Nucl - Phys.B v.711

Nuclear Physics B 711 (2005) 353
Resolving the holography in the plane-wave limit

of AdS/CFT correspondence
Suguru Dobashi, Tamiaki Yoneya
Institute of Physics, University of Tokyo, Komaba, Meguro-ku, Tokyo 153-8902, Japan
Received 15 July 2004; accepted 19 January 2005
Abstract
The issue of holographic mapping between bulk and boundary in the plane-wave limit of
AdS/SYM correspondence is reexamined from the viewpoint of correlation functions. We first study
the limit of large angular momentum for the so-called GKP-W relation in supergravity approximation, connecting directly the effective action in the bulk and the generating functional of correlation
functions on the boundary. The spacetime tunneling picture which has been proposed in our previous works naturally emerges. This gives not only a justification of our previous proposal, with some
important refinements, on the mapping between bulk effective interaction and the CFT coefficients
on the boundary in the plane-wave limit, but also implies various insights on the interpretation of
holography in the plane-wave limit. Based on this result, we construct a new holographic string
field theory. We confirm for several nontrivial examples that this gives the CFT coefficients derived
by perturbation theory on the gauge-theory side. Our results are useful for understanding how apparently different duality maps proposed from different standpoints are consistent with each other and
with our definite spacetime picture for the AdS holography in the plane-wave limit.
2005 Elsevier B.V. All rights reserved.
PACS: 11.25.-w; 04.60.-m
E-mail addresses: doba@hep1.c.u-tokyo.ac.jp (S. Dobashi), tam@hep1.c.u-tokyo.ac.jp (T. Yoneya).

0550-3213/$ see front matter 2005 Elsevier B.V. All rights reserved.
doi:10.1016/j.nuclphysb.2005.01.024
S. Dobashi, T. Yoneya / Nuclear Physics B 711 (2005) 353
1. Introduction
An impressive amount of computations have been done following the BMN conjecture [1] as to the identification of stringy operators in AdS/CFT correspondence. In spite
of all those important works of two years, however, it seems that the question of holographic correspondence of correlation functions for the BMN operators still has not been
appropriately understood. In the case of the original AdS5 /SYM4 correspondence, the relation between the bulk fields {i (z, x)} and the gauge-invariant operators {Oi (
x )} of 4D
YangMills theory has been concretely formulated as the famous GKP-W relation [2]

Z[] exp
i (
x )Oi (
x) ,
d 4x
i
which connects the boundary values limz0 zi 4 i (z, x) = i (

x ) of the bulk fields to the
source fields coupled with {Oi (
x )} at the conformal boundary of AdS spacetime. If one
naively followed the Penrose limit in the bulk of AdS spacetime in obtaining a plane-wave
approximation, one would end up in a puzzling situation that the plane-wave geometry
corresponding to the large angular momentum along a direction of S 5 cannot be related
to the conformal boundary, since the null trajectory adopted by the BMN proposal never
reaches the conformal boundary. Because of this difficulty, some different ways of comparing both sides without relying directly upon the GKP-W relation have been discussed
in the literature [3]. However, lacking for more direct links to physical observables, these
proposals seem to be still regarded as phenomenological data towards a better understanding of holography. It is very important to resolve the issue of holographic correspondence
from the viewpoint of correlation functions, since it would be a crucial basis in addressing
physically more relevant questions related to the duality between closed string theories and
gauge theories.
In previous works, we have presented basic ideas on a possible reconciliation of the
BMN proposal with the GKP-W relation. In Refs. [4,5], we proposed to interpret the GKPW relation in the plane-wave limit as a consequence of the tunneling propagation of the
BMN states of closed strings from boundary to boundary. The motivation for our proposal
was puzzles which arises in connection of holography when we adopt seemingly familiar
premises in the literature, especially, the identification of the global AdS time (or lightcone time) with the time of radial quantization on the CFT side.1 We have argued that
these puzzles are resolved, by considering the tunneling trajectory connecting AdS boundary to AdS boundary. Since the role of time parameter is played by the affine parameter
along the tunneling trajectory which is orthogonal to the conformal boundary, we cannot
identify it with radial time directly, and consequently all the puzzles are naturally resolved.
In subsequent works [7], our ideas have been extended successfully to a more general nonconformal case of Dp-brane backgrounds, by deriving the generalized correspondence [8]
obtained previously between nonconformal SYM theories and Dp-brane backgrounds.
1 In our opinion, the origin of a puzzle discussed in [6] is also related to this identification. For a list of other
approaches on the plane-wave holography, we would like to refer the reader to several review articles cited in [3].
However, our original argument in [4] has not yet been completely satisfactory, since it
involves some ambiguities with respect to normalization and short-distance cutoff when we
discuss 3-point and higher correlation functions. One of the purposes of the present work
is to reexamine our basic ideas from an equivalent but a more systematic standpoint by
studying the plane-wave limit directly on the GKP-W relation, and to strengthen our picture
by presenting further supports and extensions. In our first work [4], we have not started
from the GKP-W relation. Instead, we have treated the bulk field equation in the WKB
approximation and then proposed a natural ansatz for relating correlation functions and
the Euclidean S-matrix in the same spirit as the GKP-W relation. In the present work, we
study the limit of large R-charge (J ) for the GKP-W relation directly and confirm that our
original picture emerges automatically within the supergravity approximation. Then, on the
basis of the known relation [11] for chiral operators between supergravity and SYM gauge
theory, we establish a definite relation between 3-point correlators and the bulk effective
interactions which are consistent with our original picture, including precise normalization
and cutoff.
It turns out that the resultant supergravity effective theory in the plane-wave limit cannot
be obtained from any versions of previously known string field theories. Our result can
be adopted as a strong constraint in constructing a string field theory describing higher
stringy modes in accordance with our picture. We propose a new holographic string field
theory, which reduces to the derived effective action when restricted in the supergravity
sector, and simultaneously gives the correct 3-point correlation functions on the SYM side
derived by perturbation theory, via our holographic mapping. As a byproduct, we also give
some clarifications on the relation of our picture with other approaches which have been
discussed in some of recent works for mapping the known versions of string field theories
to gauge theory. We believe that our results not only resolve holography in the plane-wave
limit, but also lay a foundation for understanding relation among different proposals and
for investigating further extensions, on the basis of holography for correlation functions. In
particular, we give a definite prediction for impurity nonpreserving cases which have not
been treated in the literature before.
The present paper is organized as follows. In Section 2, we analyze the large J -limit
directly on the usual diagrammatic rules (Witten diagrams) for computing 3-point functions from the viewpoint of bulk theory. We demonstrate how our tunneling picture arises
in the WKB approximation and establish the validity of this picture by comparing with
exact computations. In Section 3, we derive directly the effective action in the plane-wave
limit from the bulk supergravity effective action given previously in Ref. [11]. We then
formulate the holographic relation which should be valid for (non-BPS) stringy operators.
In Section 4, we construct the holographic string field theory which is consistent with
our effective action for supergravity sector. In Section 5, we clarify the relation of our results with other approaches. In Section 6, we confirm our general discussion by explicit
examples. We conclude the paper in Section 7 by giving further remarks. Appendices A
and B are devoted to a detail of calculation and to a summary of the properties of stringinteraction vertices of string field theory, respectively.
2. The direct large J -limit of GKP-W relation

Let us start from briefly recalling the standard perturbative computation of correlation
functions from bulk supergravity theory. For simplicity, suppose that the bulk theory is
effectively described by scalar fields i (x) in the AdS5 background with action
3

3
1

m2i 2
5
2
(i ) +
+ g1 2 3 , m2i = Ji (Ji + 4)/R 2 .
S = d x g
2
2 i
i=1
i=1
(2.1)
We assume the Euclidean AdS metric (R 4 = 4gs N ( )2 ),
R2 2
(2.2)
dz + d x2
2
z
using the Poincar coordinate. According to the GKP-W dictionary, a 3-point correlation
xi ) (conformal dimension = i = Ji + ki = ki ) correspondfunction of three operators Oi (
ing to the bulk fields i is given, up to an overall normalization factor and to the lowest
order with respect to the coupling constant g, as

O1 (
x1 )O2 (
x2 )O3 (
x3 )
4
d x dz
=
(2.3)
K1 (z, x; x1 )K2 (z, x; x2 )K3 (z, x; x3 ),
z5
ds 2 =
where the bulk-to-boundary propagator

z
()
K (z, x; y) = 2
( 2) z2 + (
x y)2
satisfies

2
z3
+ z2 2 m2 K (z, x; y) = 0,
z5
z
z
x
(2.4)
for z > 0
(2.5)
and
lim z4 K (z, x; y) = (
x y).
z0
(2.6)
The PP-wave limit amounts to taking the limit where Ji , R with Ji /R 2 being
kept fixed and ki O(1). The angular momentum which comes from an SO(2) part of
the SO(6) R-symmetry must be assumed to be conserved, say, J1 = J2 + J3 . Obviously,
since i Ji , the integral in the expression (2.3) can be studied by saddle-point
methods.2 For our purpose it is useful to do a warm-up in the case of two-point functions.
In what follows until stated otherwise explicitly, it is convenient to adopt the unit such that
the AdS5 (S 5 ) radius to be one, R = 1, since R is the only length scale characterizing this
system in the supergravity limit 0 with fixed R.
2 A preliminary discussion on the approach of the present paper has been given by one of the present authors
in a talk at the Strings 2003 conference [12].
2.1. Two-point functions

Consider a 2-point function of the following form:
4
d x dz
G2 (
x1 , x2 )
K (z, x; x1 )K (z, x; x2 )z ,
z5
where 0+ is a parameter for regularization. The saddle-point equations are

z
z
+ ln 2
= 0,
ln 2
z
z + (
x x1 )2
z + (
x x2 )2

z
z
+
ln
= 0.
ln 2
x
z + (
x x1 )2
z2 + (
x x2 )2
(2.7)
(2.8)
(2.9)
The general solution is

1
1
|
x1 x2 |
z0 =
x1 + x2 ) (
x1 x2 ) tanh ,
,
x0 = (
(2.10)
2
2
2 cosh
with being an undermined integration constant. Thus the integral can be approximated
as a one-dimensional integral over the collective coordinate . In conformity with our
previous works, the solution describes a tunneling process from one boundary point x1 at
to another boundary point x2 at +. Thus we have naturally arrived at the
same picture for the PP-wave holography as we have proposed in previous works.
Following the standard method of semi-classical path-integrals, the integration measure
in (2.3) is replaced by
dz d 4 x
d d z d 3 x J ( )
z5
(2.11)
which are defined by

for the collective coordinate and the fluctuating coordinates z , x,
the following shift of the bulk coordinates,
z = z0 ( ) + z ,

x = x0 ( ) + x,
(2.12)
with the orthogonality constraint

d x0
dz0
z +
x = 0.
(2.13)
d
d
The effective metric for the fluctuations is found to be, to the lowest nontrivial order in the
fluctuations,
1
2
(dz0 + d z )2 + (d x0 + d x)
2
z
4 cosh2
cosh2 (d z )2 + (d x )2 ,
(d )2 +
2
|
x1 x2 |
(2.14)
where we have used the solution for the constraint (2.13),

x =
n sinh z + x ,
n x = 0,
(2.15)
with n being the unit vector along the direction of the vector x1 x2 , which connects two
points x1 , x2 on the boundary. Thus, the Jacobian is given as

4 cosh2 4
J ( ) =
(2.16)
cosh2 .
|
x1 x2 |2
On the other hand, the effective second-order action for the saddle-point integral is
(2)
Seff =
4
cosh4 z 2 + cosh2 (x )2 + -dependent factor.
2
|
x1 x2 |
(2.17)
The classical part of the action gives only a factor which is independent of the collective
coordinate and has the correct dependence on the distance of boundary points, as can be
checked by using

K z0 ( ), x0 ( ); x1,2 =
()
|
x1 x2 | e ,
2 ( 2)
(2.18)
where the sign on the exponentials depends on the points x1 or x2 , respectively. It is now
evident that the Jacobian factor is canceled by the integrations over the fluctuating coordinates, up to a -dependent proportionality constant. Consequently, the two-point function
in the large limit is given simply by
2
G2 (
x1 , x2 ) 2 |
x1 x2 |2()
+
d (2 cosh ) ,
(2.19)
where the -dependent contributions come from the factor z in the defining expression
(2.7). Thus, in the limit 0+, we reproduce the correct behavior for two-point correlators for conformal operators, up to the pole singularity
+
2
d (2 cosh ) .

(2.20)
We can compare this result with that of exact integration:

2 2

( 2)(/2)2
()
G2 (
|
x1 x2 |2()
x1 , x2 ) =
2
2
()()
( 2)
22 1
|
x1 x2 |2 .
2
(2.21)
Here we have used the general formula for this type of integral [9],

dz
0
d D x
za
(z2
+ (
x x1
)2 )b (z2
+ (
x x2 )2 )c
= |
x1 x2 |1+a+D2b2c I (a, b, c, D),
I (a, b, c, D)
D/2
a
2
1
2

b + c D2 a2 12 12 + a2 + D2 b 12 +
(b)(c)(1 + a + D b c)
a
2
D
2
(2.22)
In reality, for the effective theory described by (2.1), it is more appropriate to consider
4

d x dz
g (z, x) K (z, x; x1 ) K (z, x; x2 ) + m2 K (z, x; x1 )K (z, x; x2 )
5
z
(2.23)
than (2.7). It is easy to repeat the above calculation for this case. Only difference of the
final result from the case of G2 is the multiplication of (m2 2 )/m2 4/.
2.2. 3-point functions
Armed by this exercise, we now go back to the 3-point function (2.3). The saddle-point
equations are
3
i=1
3
i=1
1
2z
2
z z + (
xi x)2

= 0,
xi x
= 0.
z2 + (
xi x)2
(2.24)
(2.25)
It is easy to convince oneself that, for generic configurations of three boundary points, there
is no solution to these equations. However, if we take a limit where two of the boundary
points, say, x2 and x3 approach sufficiently to one point xc = (
x2 + x3 )/2, the same trajectory connecting x1 and xc as we have discussed in the previous subsection can be regarded
as an approximate solution. Therefore, let us try to reduce the integral to the one along the
following trajectory,
1
1
|
x1 xc |
x1 + xc ) (
x1 xc ) tanh ,
,
z0 ( ) =
x0 ( ) = (
(2.26)
2
2
2 cosh
with the fluctuations, x = x x0 and z = z z0 . For our purpose it is sufficient to evaluate
the integral to the leading order in the short-distance limit,
x3 + xc 0.
x2 xc =
In order to avoid unnecessary complications, we assume that all three points x1 , x2 , x3 are
along a single line on the boundary.
The effective action for this computation
Seff =
3

i=1
is rewritten as
i ln
xi x)2
z2 + (
z
(2.27)
10
2

z + (
z2 + (
1 + 2 + 3
x1 x)2
xc x)2
ln
+ ln
2
z
z
2

2
2
z + (
z + (
x1 x)
xc x)2
1 2 3
ln
ln
+
2
z
z

2 (
xc x) + 2
2 (
xc x) + 2
+
.
ln
1
+
+ 2 ln 1 + 2
3
z + (
xc x)2
z2 + (
xc x)2
(2.28)
The first line of (2.28) can be treated in exactly the same way as for the 2-point case, by
replacing by (1 + 2 + 3 )/2 ( J1 ). Next, since
1 2 3 O(1),
(2.29)
the second line can be approximated by its value on the classical trajectory

2
z + (
x1 x0 )2
x2 x0 )2
z2 + (
1 2 3
= (1 2 3 ).
ln 0
ln 0
2
z0
z0
(2.30)
The third line gives, to the zeroth order with respect to the fluctuations,
2
2
(2 + 3 )
2(2 3 )
+
(
+
)
e2
2
3
|
x1 xc |
|
x1 xc |2
|
x1 xc |2

+ O 3 .
(2.31)
Here we have kept the third term which is of second order in 0, since it shows that
for sufficiently large there is a natural cutoff for the range of the affine parameter for
0. The other terms which are independent of can be ignored in
arbitrary small = ||
the limit 0.
As for the contribution of fluctuating coordinates in the third line, it is easy to convince
ourselves that keeping only the first order term with respect both to the fluctuations and to
the short-distance cutoff is sufficient for our purpose. The relevant terms are arranged as

xc x0 )) 1
x
(
xc x )(z z0 x (
4(2 3 )
+
2 z02 + (
(z02 + (
xc x0 )2 )2
xc x0 )2
4(2 3 )
e cosh2 ,
= z
(2.32)
|
x1 xc |2
which involves only the z -fluctuation. Combining this with the relevant part of the Gaussian
factor coming from the first line, the integral with respect to z is

4
4(2 3 )
2 4J1 cosh
2
d z [measure] exp z
z
e cosh
|
x1 x2 |2
|
x1 xc |2

2
4
(2 3 )2 e2
2 4J1 cosh
d
z
[measure]
exp
z
= exp
(2.33)
J1 |
x1 x2 |2
|
x1 x2 |2
to the present order of approximation. The prefactor here implies that, together with the
contribution from the 3rd term of (2.31), the total factor which is responsible for the cutoff
at large region is

2
2 (2 + 3 )
2
2 (2 3 )
exp e2
+
e
J1
|
x1 x2 |2
|
x1 x2 |2

2
J2 J3 2
exp 4
e .
2
|
x1 x2 | J1
11
(2.34)
Putting all of these results together and remembering that the Jacobian and the Gaussian
integral cancel, the integral of the 3-point function now takes the form
2
|
x1 xc |(1 +2 +3 )
J12

+
J2 J3 (2)2
(1 2 3 )
2
(2.35)
d e
exp
e
J1 |
x1 xc |2
in the limit of large Ji and small . This expression shows that the precise form of the
cutoff mentioned above is c with

J2 J3
2
c
.
e
(2.36)
=
J1 |
x1 xc |
Thus, the 3-point integral can be expressed by a Gamma function and leads to

(2 +3 1 )/2 2 +3 1

+1
2
21
(2 +3 1 ) J2 J3
2
|
x1 xc |
(2)
.
J1
1 + 2 + 3
J12
(2.37)
After fixing the convention for normalization suitably, this give the correct short-distance
limit for the 3-point correlator, from which we can identify the CFT coefficient by equating
the expression with
lim
0
C123
C123
,
21
|
x1 x2 |23 |
x2 x3 |21 |
x3 x1 |22
|
x1 xc |21 |2|
(2.38)
where 1 = (2 + 3 1 )/2, etc. The precise normalization will be discussed in the

next section.
Let us compare this result with the exact computation. The same integral formula cited
before gives
i
3
z
d 4 x dz
z5
z2 + (
x xi )2
i=1
c(1 , 2 , 3 )
,
x3 |2 +3 1 |
x3 x1 |3 +1 2
|
x1 x2 |1 +2 3 |
x2
c(1 , 2 , 3 )

2 12 (1 + 2 3 ) 12 (2 + 3 1 ) 12 (3 + 1 2 )
=
2
(1 )(2 )(3 )

1
(1 + 2 + 3 4) .
2
(2.39)
(2.40)
12
By taking the large Ji limit using the Stirling formula

ln J J ,
lim (J + k) 2 exp J + k
J
2
we obtain

(2 +3 1 )/2 2 +3 1

+1
2
21
(2 +3 1 ) J2 J3
2
|
x1 xc |
(2)
,
J1
1 + 2 + 3
J12
(2.41)
which exactly matches (2.37).
Thus we have confirmed that the 3-point correlators in the short distance limit 0
can be computed effectively as a process in the bulk occurring along a single tunneling
trajectory connecting boundary to boundary, whose amplitude essentially takes the general
form,
1 2 3
g,
1 + 2 + 3
1 2 3

2
1 + 2 + 3
J2 J3
+1 ,

g g
J1
2
(2.42)
= 2,
(2.43)
with a suitable normalization convention, apart from the usual spacetime-dependent factor
|
x1 xc |21 .
The form (2.42) is consistent with what we have proposed as the general structure
for 3-point correlators which is expected from the spacetime picture for holography
in the PP-wave limit. The parameter ec is the cutoff related to the distance of
x2 ) and O3 (
x3 ) at the boundary, as (2.36). In our first original disthe operators O2 (
cussion,
the
emergence
of
this
particular form is originated from the formal integral

d exp[(1 2 3 ) ] which appears in perturbation theory along the tunneling trajectory. This is actually ill-defined as it stands. To justify the expression (2.42), we had to
invoke a wave-packet picture which is rather subtle in the Euclidean tunneling as is alluded
to in [4]. According to the present argument, the ill-defined integral must be replaced by
(2.35) which can be defined unambiguously by analytic continuation, due to the presence
of the natural cutoff for large region. Intuitively, the large cutoff corresponds to the
limitation of the picture using the single tunneling trajectory in discussing 3-point correlators for small but nonzero . Note that if we take the limit 0 naively inside the integral,
this factor would have simply disappeared. In fact, the additional factor

J2 J3 (1 2 3 )/2
1 + 2 + 3

+1
J1
2
was missing in the previous discussions, owing to this ambiguity.
It is known that the coupling constants of supergravity modes in the bulk vanish for the
so-called extremal case with 1 2 3 = 0. However, as is well known, the extremal
correlation functions themselves do not vanish [10]. This remarkable fact corresponds to
the presence of the singular denominator in (2.42). We propose that the holographic correspondence relating the bulk 3-point couplings and the CFT coefficients should obey the
13
above general relation. Of course, the computation of this section is restricted to the supergravity approximation. Below, we will argue that this should generalize to non-BPS
stringy modes too, with slight corrections. Actually, if we only consider a restricted set of
the correlators in which the numbers of impurities are conserved, as has been assumed
in the literature, the correction factor can be ignored in the large limit to the leading
order in making comparison with perturbative computation on the YangMills side, since
1 2 3 O(1/2 ) for such cases.
3. Effective action along the tunneling trajectory

Since we have established that the computation of correlation functions (or CFT coefficients) can be reduced to the processes along a single tunneling trajectory connecting AdS
boundary to AdS boundary, we can now proceed to derive an effective action along the
trajectory directly from a more general effective action of supergravity in the bulk.
3.1. Bulk effective action and the CFT coefficients
We study the bulk effective action for general chiral primary operators consisting of the
SO(6) scalar fields i (x)

O I = CiI1 i2 ...ik Tr i1 i2 ik .
(3.1)
Here, C is a totally symmetric traceless tensor whose contraction is normalized as
I I
C 1 C 2 = CiI11i2 ...ik C I2 ,i1 i2 ...ik = I1 I2 ,
(3.2)
2 N )k/2 k is the normalization factor such that the 2-point functions

and = (2)k /(gYM
takes the form (
x12 = x1 x2 ),

I1 I2
x1 )O I2 (
x2 ) =
.
O I1 (
|
x12 |2k
(3.3)
Since the BMN operators in the supergravity sector are essentially contained in this set
of local operators, it should be possible to derive the effective action along the tunneling
trajectory starting from an appropriate effective action for these operators. The BMN supergravity modes are a subset of (3.1), expressed by using the complex basis for the SO(2)
directions i = 5, 6 with Z = 1 (5 + i6 ):
2
O I = CiI1 i2 ...i Tr Z J i1 i2 ik + permutations ,

k
k = J + k,
J , (3.4)
where CiI1 i2 ...i are now completely symmetrized traceless SO(4) tensors with unit normalk
ization (similarly as (3.2)) under contraction. The normalization constant is related to that
of the original set as
1/2
k
=
(3.5)
.
J
14
Note that in the large J -limit the operators with vector excitations Di Z are regarded
essentially as the derivatives of (3.4), and hence are included in this set defined by (3.4) if
we suitably take into account the variation of the boundary coordinates x . In fact, as we
will argue later on we have to take a special care in the interpretation of vector excitations.
The bulk effective action for the SO(6) operators has been derived in Ref. [11]:

1
4N 2
1
5
d x g
(I )2 + k(k 4)I2
Sbulk =
5
2
2
(2)
I

GI I I
1
1 2 3 I1 I2 I3 ,
(3.6)
3
AI1 AI2 AI3
I ,I ,I
1
where we used Euclidean metric and assumed a particular normalization for the scalar
fields:
AI = 26k 3
k(k 1)
.
(k + 1)2
(3.7)
This is the effective action in the sense that it reproduces correlation functions through the
Witten diagrams as studied in the previous section. Therefore, the freedom of field redefinition on the bulk side and correspondingly the choice of the basis of chiral operators on the
boundary side are both fixed already. Remarkably, the set of bulk fields {I } corresponding
to the set {O I } are effectively treated as scalar fields propagating in the AdS5 background
without derivative coupling. The derivative interactions are removed by making a particular field redefinition in the derivation of this action.3 Actually, the effective action (3.6)
was obtained from the equation of motion of IIB supergravity. It was argued that the result
must be correct including the normalization of the CFT coefficients by relating them to the
R-symmetry currents which have been known to be given exactly by the free gauge theory.
We refer the reader to [11] and the references therein for more details on this effective action. Also for a summary on the nonrenormalization properties of 2- and 3-point functions
of chiral operators, see, e.g., [13] and the references therein.
The CFT coefficients C I1 I2 I3 of the above operators defined by

x1 )O I2 (
x2 )O I3 (
x3 ) =
O I1 (
C I1 I2 I3
|
x12 |23 |
x23 |21 |
x31 |22
(3.8)
with
2 + 3 1
, etc.,
2
is related to the 3-point interaction in the effective action (3.6) by
k1 k2 k3 I 1 I 2 I 3
I1 I2 I3
C C C ,
=
C
N
I I I
(k1 + 1)(k2 + 1)(k3 + 1)
GI1 I2 I3
C 1C 2C 3 =
,
7
a(k1 , k2 , k3 ) 2 ((/2)2 1)((/2)2 4)1 2 3
1 =
3 This does not mean that the interactions containing derivatives along the S 5 directions are removed.
(3.9)
(3.10)
(3.11)
= k1 + k 2 + k 3 ,
a(k1 , k2 , k3 ) =
2
15
(3.12)
k1 !k2 !k3 !
3
.
+ 2 !2(2)/2 1 !2 !3 !
(3.13)
The first expression in this list is nothing but the result of free field computation of correlators on the YangMills side, with C I1 C I2 C I3 being all the possible contractions among the
SO(6) indices of C-tensors. The second expression is its representation in terms of the bulk
quantities, in which the factor a(k1 , k2 , k3 ) arises as the coefficient relating C I1 C I2 C I3
to the overlap integral of S 5 harmonics Y I , which appears in reducing 10D theory to 5D
effective theory on AdS5 backgrounds, as

(3.14)
Y I1 Y I2 Y I3 = a(k1 , k2 , k3 ) C I1 C I2 C I3 .
S5
The definitions of the S 5 harmonics and their integrals are summarized in Appendix A, to
which we refer the reader for more details. The fact that the supergravity effective action
is exactly matched to the free-field results on the gauge-theory side is interpreted as a
consequence of nonrenormalization properties of 3-point functions of chiral operators. This
is quite remarkable since the supergravity limit 0 is nothing but a strong-coupling
2 N = R 4 /( )2 on the gauge-theory side.
limit gYM
Before going to our main task of this section, let us here check that the relation between
the above effective action and the CFT coefficients is consistent with the prediction of
previous section. Taking the limit J1 (= J2 + J3 ), J2 , J3 , we find
C I1 I2 I3 =
1 2J1 + 2 9
N
3

J1 J2 J3 J1 1 1 !
GI I I .
J2 J3
1 1 2 3
J12
(3.15)
Apart from the normalization factor which is independent of 1 = (2 + 3 1 )/2 =

(k2 + k3 k1 )/2 1 , this indeed coincides with the result obtained in the previous section,
as it should. The difference of the normalization is owing to the fact that we have not yet
fixed the precise normalization of two-point functions in the previous section.
The CFT coefficients can be expressed in terms of the SO(4) basis (3.4) of the BMN
operators, using the relation

1 k2 /2 k3 /2
k1 !k2 !k3 ! I I I
I I I
J1
J2
J3
C 1 C 2 C 3 = 1 !
(3.16)
C 1 C 2 C 3 ,
J2 J3
J1
J1
1 ! 2 ! 3 !
which is valid in the large Ji limit. For a derivation of this relation, see Appendix A. Thus,
the 3-point coupling coefficient in the above effective action is

k2 /2 k3 /2
k1 !k2 !k3 ! I I I
J2
J3
C 1 C 2 C 3
GI1 I2 I3 = 1 3 2J1 2 +9 J12
(3.17)
J1
J1
1 ! 2 ! 3 !
in terms of the SO(4) contractions.
16
3.2. Effective (0 + 1)-dimensional action

With these preparations, we are now in the position of deriving the effective action
along the tunneling trajectory starting from (3.6). Let us parametrize the coordinates near
the trajectory as in the previous section,
x = x0 ( ) + x
z = z0 ( ) + z ,
(3.18)
with
x =
nz sinh + x .
(3.19)
Since the fluctuations around the trajectory can be assumed to be of order 1/ J from the
discussion of the previous section, we derive the effective metric which is correct to the
second order in the fluctuations. The result is

ds 2 = 1 + z 2 + x 2 d 2 + d z 2 + d x 2 + higher-order term
after rescaling as
z
|
x1 xc |
2
2 cosh
z ,
|
x1 xc |
x
x ,
2 cosh
and then making the shift of the time coordinate

+
sinh 2 2
z + x .
2 cosh
To avoid notational confusions, we denote the rescaled fluctuations (x , z ) by a four

vector y = (y1 , y2 , y3 , y4 ). Thus, the original fluctuating four-vector x in the bulk is now
expressed as

|
x1 xc |
sinh
(x , n x ) =
(3.20)
y1 , y2 , y3 ,
y4 ,
2 cosh
cosh
with the time parameter being redefined as
= +
sinh 2
y .
2 cosh
The effective metric

ds 2 = 1 + y 2 d 2 + d y 2
(3.21)
(3.22)
has an SO(4) symmetry with respect to the rotation of y. We also note that, since the
order of magnitude of y is supposed to be constant in on the basis of this -independent
metric, (3.20) implies that the original fluctuating coordinates x decrease as e| | , when
we approach the boundary .
Using this metric, the quadratic terms of the bulk effective action (3.6) are given, to the
accuracy of second order in y, as

4N 2
1 2
4
I I + y I y I
y
y

1
d
2
(2)5

1
+ k 4) I I .
+ 1 + y 2 (J + k)(J
2
17
(3.23)
Note that we have changed the SO(6) index to the complex SO(2) SO(4) index.
The assumption that the magnitude of the fluctuating coordinates is of order 1/ J is
justified in the language of the effective action as follows. Redefine the fields by
(, y) = eJ 0 (
y )( ),
(3.24)
y) = e+J (J ) (
),
(,
y )(
0
(3.25)

1
0(J ) (
y ) = (J /)2 exp J y 2
2
(3.26)
(J )
with
being the ground state wave function for the kinetic operator for the fluctuating coordinates,
h = y2 + J 2 y 2 .
(3.27)
The hermiticity condition for our Euclidean field theory is [4]

y) = (, y),
(,
(3.28)
which requires the choice of signs in the definition (3.25) on the exponential. Here and in
what follows, we rewrite and z by and z without tilde again for notational brevity.
), the leading term of the free action is then of
In terms of the reduced field ( ), (
order O(J ), taking the familiar nonrelativistic form in 0 + 1 dimensions as

4N 2
+ 2k
.

(3.29)
d J
(2)5
are canceled, and the zero-point energy of the
The order O(J 2 ) terms of the form J 2
operator (3.27) is responsible for cancelling a part of the order O(J ) term in the last term in
the Lagrangian density. Note that the term | |2 is of order one and hence can be ignored
in the large J -limit. Absorbing the normalization factor for the field, the free Hamiltonian
is seen to be k which correctly reproduces the O(1) part of the energy = J + k.

By performing the above redefinition for the cubic interaction term and rescaling,
(2)5/2 1
,
2N
2J
to renormalize the quadratic terms in the standard form, the total effective action is

1
( I I I I ) + kI I I
d
2
I

1
+
(3.30)
I1 ,I2 ,I3 ( I1 I2 I3 + h.c.),
d
2
I1 ,I2 ,I3
18
1 8+J1 2/2
2
J1 J2 J3 GI1 ,I2 ,I3 ,
3N
J12
I1 ,I2 ,I3 =
(3.31)
with J1 = J2 + J3 . Comparing this result with the known relation (3.15) connecting the
CFT coefficients to the 3-point coupling constant, we confirm that the relation is again
precisely the one discussed in (2.42) and (2.43). This establishes that the effective theory
along the tunneling trajectory for the BMN operators with only scalar excitations is (3.30).
Now it is very natural to extend this action by including the higher excited states
with respect to the Schrdinger operator (3.27). Since the fluctuating coordinates near the
boundary z0 0 ( ) are essentially the four vector x of the 4D base space up to an
exponential factor |
x1 xc |e| | , taking these excited states into account seems to corre of the BPS-BMN operators. Here
spond to the inclusion of the vector excitations Di Z(Z)
we write down this extension and will give a precise discussion of this correspondence in
the next section. The decomposition (3.24) is generalized to the infinite expansion over the
complete set of excited states,
I (, y) =
(2)5/2 1 J (J )
n (
y )n ( ),
e
2N
2J
n
(3.32)
I (, y) =
(2)5/2 1 +J (J )
n (
y ) n ( ),
e
2N
2J
n
(3.33)
where
n(J ) (
y) =
4 1/4 ni /2

J
2
2
Hni J y eJ y /2
ni !
i=1
(3.34)
are the normalized eigenfunctions. The kinetic term is then extended to

1
I,n

( I,n I,n I,n I,n ) + kI +
4
ni I,n I,n
(3.35)
i=1
and similarly for the interaction term

1
2

I1 ,n(1) ,I2 ,n(2) ,I3 ,n(3)
1 8+J1 2/2
2
J1 J2 J3
3N
J12
(1) (2) (3)

GIn ,I ,n,I ,n I1 ,n(1) I2 ,n(2) I3 ,n(3) + h.c.
1
(3.36)
with
(1) (2) (3)
GIn ,I ,n,I ,n
1 2 3
J1
= GI1 ,I2 ,I3
J2 J3
(J )
(J )
(J )
y )n(2)2 (
y )n(3)3 (
y ).
d 4 y n(1)1 (
(3.37)
19
3.3. Vector excitations

The form of the effective action (3.36) and (3.37) including higher excited states with
respect to the fluctuations y indicates that the vector excitations of BMN operators in the
bulk can be treated in the same manner as the scalar excitations. On the boundary, BMN
operators with vector excitations restricted to supergravity modes can be regarded essentially as derivatives of those without vector excitations; for sufficiently large J ,

C iI1 i2 ...i KjL1 j2 ...j j1 j2 j Tr Z J i1 i2 ik + permutations

k

C iI1 i2 ...i KjL1 j2 ...j Tr Z J Dj1 ZDj2 Z Dj Z i1 i2 ik + permutations ,

k
(3.38)
where KjL1 j2 ...j is again a completely symmetric and traceless tensor4 along the base-space
directions on the 4D boundary, with the normalization condition of the same type as for
C I . The permutation of the second line means a summation over all possibilities of different orderings of operators including Dj Z. Since we consider the limit of large J with
fixed k and , the derivatives acting on -fields can be ignored. We expect that the excitations with respect to yj components correspond to the action of j on the boundary,
as suggested in the original BMN proposal. However, we encounter two puzzles with this
naive expectation.
Firstly, we have stressed in Section 3.2 that the fluctuations with respect to the original 4-coordinate x vanish exponentially e| | as we approach the boundary. We then
naively expect that the higher excited states do not affect the boundary theory. However,
this is actually as it should be in accordance with our S-matrix picture [4] for the PP-wave
holography. When an asymptotic state is an excited state, we have to supply an additional
energy-factor e| | to the wave function, by the definition of the S-matrix, which would
just cancel the decreasing exponential associated with y fluctuations. In the case of scalar
excitations discussed in Section 2, the boundary condition (2.6) of the bulkboundary propagator already takes this into account, as is visible in the boundary condition (2.6).
Secondly, the effective (0 + 1)-dimensional action clearly demands that the 2-point
functions must satisfy an orthogonality condition, since the quadratic terms are diagonalized. To the contrary, the two-point functions obtained from the standard form |x 1x |2 by
1
2
acting derivative do not satisfy orthogonality: for instance,
nj
,
1,j ln |
(3.39)
x1 x2 | =
|
x1 x2 |
1
1,j1 2,j2 ln |
(3.40)
x1 x2 | =
(j j 2nj1 nj2 ),
|
x1 x2 |2 1 2
with nj = (x1 x2 )j /|
x1 x2 |. The appearance of the tensor factor j1 j2 2nj1 nj2 Cj1 j2
is consistent with the relation (3.20) of fluctuating coordinates with the vector y, requiring that the directions of y along n are actually opposite at two asymptotic regions
4 Strictly speaking, we should also include higher derivatives appropriately in the right-hand side of (3.38). In
the present section, we do not treat the trace part for simplicity. They will be partly considered in Section 5 after
we construct the holographic string field theory.
20
, corresponding to x1 and x2 , respectively, at the boundary. The tensor Cij is usually associated with the conformal inversion x x/|x|2 . In our case, this is an automatic
consequence of the bulkboundary correspondence. However, (3.39) evidently leads to an
inconsistency with the orthogonality between scalar and vector excitations. In the literature
(see, e.g., [14]), a particular way out of this difficulty has been discussed without a definite
physical picture for holography. The suggested procedure recovers orthogonality only after taking the limit |
x1 x2 | . Since our picture for holography must be valid for any
finite |
x1 x2 |, it is not satisfactory from our standpoint.
From our viewpoint of a direct correspondence between bulk and boundary as clarified up to this point, the origin of this apparent discrepancy lies in an important difference
with respect to the SO(4) symmetry on both sides. For the bulk, we use the metric (3.22)
which characterizes the geometry close to the tunneling trajectory traversing from a fixed
boundary point to another fixed boundary point. In particular, the trajectory is asymptotically orthogonal to the boundary spacetime, and the fluctuations near the boundary have
an SO(4) symmetry under global rotations of y. The SO(4) rotations must be performed
simultaneously at all points along the trajectory, and consequently at both ends on the
boundary. It is important to recognize that this SO(4) symmetry cannot be identified with
the SO(4) isometry of the AdS metric with respect to x , as is indicated by the relation
(3.20). Since the latter isometry should correspond to the SO(4) symmetry of the boundary
theory, the rotations of y cannot be directly related to the usual SO(4) symmetry at the
boundary.
For the boundary theory, we are using the flat Euclidean metric along the boundary
space. The change of a distance caused by a small variation of points in the flat Euclidean
metric is not of second order with respect to the variations of coordinates
(
x1 + x1 x2 x2 )2

| x1 x2 | 2
x1 x2
2
+
= |
x1 x2 | 1 + 2 n
.
|
x1 x2 |
|
x1 x2 |
(3.41)
This is not SO(4) symmetric under the rotation of y. Note that the SO(4) transformations
of y must be done with fixed fiducial points, x1 and x2 , in such a way that the vicinities
around them are simultaneously rotated. Obviously, such an SO(4) symmetry cannot be
identified with the usual global SO(4) isometry of the original AdS metric. One might
wonder that this is then a self-contradiction. But that is not so. As we have emphasized, the
magnitude of fluctuations around the trajectory vanishes exponentially as we approach the
boundary, and hence there is no contradiction. This is also consistent with the fact that the
tunneling trajectory is a classical solution under the restriction that the fluctuations satisfy
the Dirichlet boundary condition at z = 0. The Euclidean S-matrix, however, requires us
to blow up the vanishing fluctuations near the boundary at the vicinities of the fiducial
points by multiplying the powers of e| | = eT in such a way that the SO(4) symmetry
with respect to y is kept. The process of blowing-up amounts to introducing a particular
UV cutoff for the boundary theory.
Therefore, the naive identification of the vector excitations in the bulk with the derivations Di Z on the boundary is not precise. We have to check from the bulk point of view
how the small variations near the boundary are treated by deriving two-point functions for
21
them following the approach of Section 2. Suppose that external lines are general vector
states corresponding to the operators (3.38) with two traceless symmetric tensors K L1 and
K L2 . This amounts to extracting the parts proportional to
KjL11j2 ...j y j1 y j2 y j1 ,
1
KjL 2j ...j y j1 y j2 y
j
2
1 2
from the bulkboundary propagators in the integrand (2.7),

J +k
J +k

z
z
,
,
z2 + (
x + x x1 x1 )2
z2 + (
x + x x2 x2 )2
respectively. They are given by
J +k+
1
j1
j1 j2
z
1 1 x x x
,
2 J
z 1
z2 + (
x x1 )2
2 2 x
2 J
j1 j2
x x
z 2
j
z
2
z + (
x x2 )2
(3.42)
J +k+
2
,
(3.43)
respectively, in the large Ji -limit, where

x i = ij 2ni nj x j .
(3.44)
Note that the use of redefined x i is required by the change of sign in the relation (3.20)
between the fluctuating coordinates with y in the limit = T (T ). Although the
4th components actually give the powers of | tanh | in converting from x/z (or x /z) to y,
they can be replaced by one in the leading singular part with respect to the regularization .
Then, the result of saddle-point integral is
L1 L2
J2
2
(2J )1 1 !|

x1 x2 |2(J +k+1 ) L1 L2 .
2

(3.45)
The Kronecker L1 L2 and the prefactor arises by the Gaussian integral

4 j j

2
j
dy y 1 y 2 y j1 y j1 y j2 y 2 eJ y
L2
1 +2 L1
(2J )
Kj1 j2 ...j Kj j ...j

2
1
2
1 2
dy 4 eJ y
= L1 L2 1 !(2J )1 .
(3.46)
In terms of derivatives with respect to (

x1 , x2 ), this result is expressed as
KjL11j2 ...j 1,j1 1,j2 1,j1 KjL 2j ...j 2,j1 2,j2 2,j
1
1 2
(2J ) 1 !
1)
|
x1 x2 |2(J +k+
L1 L2
2
2(J +k)
|
x12 |g
(3.47)
with
i = (ij 2ni nj )j
(3.48)
22
being the derivatives with respect to x i . Here the subscript g for the distance |
x12 | indicates
that the derivatives are computed by assuming that the small variations with respect to the
boundary positions are defined by

x1 x2
x1 x2 )2 1 2
|
x12 |2 (
(3.49)
|
x1 x2 |2
instead of the naive expression (3.41). Apart from the terms x12 , x22 which do not contribute owing to the traceless condition for the tensors K L1 and K L2 , this amounts to
dropping all terms which violates the SO(4) symmetry in (3.41). Strictly speaking, we
should have denoted the derivatives i , i by using different notations to indicate the situation. Effectively, our prescription is equivalent to dropping terms which violates the SO(4)
symmetry, and is consistent with the prescriptions adopted in other approaches. Practically, the following rule is valid: first we consider only the directions orthogonal to n to
obtain SO(3) symmetric answer. Then we extend the results formally to SO(4) symmetric
ones.
The variation indicated by (3.49) can be interpreted as the variation of the distances
measured using the plane-wave metric (3.22). The minimal distance defined by the distance
functional

2

2

d y
1 2
d y
2
(
y1 , y2 ; T ) = d 1 + y +
(3.50)
y +
d 1 +
d
2
d
with the boundary condition
y(T ) = y1 ,
y(T ) = y2
(3.51)
behaves as
y1 y2
1 2
y + y22 tanh 2T
2 1
sinh 2T

1
2T + y12 + y22 2
y1 y2 e2T + O e4T .
2
Thus the transition amplitude is
(
y1 , y2 ; T ) = 2T +
y1 ,
(J +k)(
y2 ;T )
2(J +k)T
x1 x2
12
|
x1 x2 |2
(3.52)
J +k
(3.53)
with
x1
eT y1
,
|
x1 x2 |
x2
eT y2
,
|
x1 x2 |
(3.54)
which are required by the relation (3.20) for large | | T . Note that we can ignore the
term 12 (
y12 + y22 ) in (3.52), which corresponds to the redefinition (3.21) and does not contribute owing to the traceless condition.
We can extend the above argument to 3-point correlators with vector excitations. By an
obvious simplification of notations, we are lead to the integral
23

i=1 (2Ji )i K Li

2
j
j
dy 4 y j1 y j2 y j1 y j1 y j2 y 2 y j1 y j2 y 3 eJ1 y

2
4
J
y
1
dy e
(scalar integral with i Ji + ki + i )
= 2(1 +2 +3 )/2 J2 3 J3 2
1 !2 !3 ! (1 + 1 1)! L1 L2 L3

K K K
1 !2 !3 ! (1 1)!
C I1 I2 I3
|
x1c
|2(J1 +k1 +1 ) |2|2(1 +1 )
(3.55)
where
1 = (2 + 3 1 )/2,
(3.56)
etc.
Taking into account the normalization factor, this leads to the CFT coefficient

J1 1 J2 2 /2 J3 3 /2
I1 I2 I3 ,L1 L2 L3
C
=
J2 J3
J1
J1
1 !2 !3 ! (1 + 1 1)! I1 I2 I3 L1 L2 L3

K K K .
C
1 !2 !3 !
(1 1)!
(3.57)
From the viewpoint of boundary, this result can again be formulated as the consequence
of the replacements

x1 x2
|
x12 |2 (
,
x1 xc )2 1 2
|
x1 xc |2

x1 x3
2
2
|
x13 | (
(3.58)
,
x1 xc ) 1 2
|
x1 xc |2

|2| (2)
2
x2 x3
12
|2|2

(3.59)
in acting the derivation

L
KjL11j2 ...j 1,j1 1,j2 1,j1 KjL 2j ...j 2,j1 2,j2 2,j Kj 3j ...j 3,j1 3,j2 3,j
1
1 2
2
1 2
3
directly to the general form (3.8) of the 3-point correlator without vector excitations. The
prescriptions (3.58) are equivalent to (3.49) to the leading order in , while (3.59) is necessary to be consistent with the SO(4) symmetry.
Combining with the CFT coefficient of pure scalars, we obtain
1 2J1 + 2 9 J1 J2 J3 J1 1 +1 (1 + 1 )!
C I1 I2 I3 ,L1 L2 L3 =
GI1 I2 I3
N
J2 J3
1 + 1
3
J12
2 /2 3 /2
J3
1 !2 !3 ! L1 L2 L3
J2
K K K
(3.60)
J1
J1
1 !2 !3 !
in terms of the pure-scalar 3-point coupling, or using the expression for the scalar vertex
(3.17)
24
I1 I2 I3 ,L1 L2 L3

J1 J2 J3 J1 1 +1
1
(1 + 1 )!
2
J
J
J1
2 3
1 + 1
(k2 +2 )/2 (k3 +3 )/2
J2
J3
J1
J1

k1 !k2 !k3 ! I I I 1 !2 !3 ! L L L

C 1 C 2 C 3
K 1K 2K 3
1 ! 2 ! 3 !
1 !2 !3 !
1
=
N
(3.61)
in terms of the SO(4) tensors. On the other hand, the effective interaction (3.37) including
the vector excitations shows that the 3-point coupling for the external lines in consideration
is given by replacing the coefficient GI1 I2 I3 for purely scalar excitations as
2 /2 3 /2
J2
J3
1 !2 !3 ! L1 L2 L3
L L L
K K K .
GI11I2 I23 3 = GI1 I2 I3
(3.62)
J1
J1
1 !2 !3 !
Comparing this with (3.60) taking into account the normalization factor in the 3-point
interaction term of the effective action, we can conclude that the general form, as derived
in Section 2 for the holographic correspondence of scalar supergravity excitations, extends
to a more general case including the vector excitations,
L L L
I11I2 I23 3
L L L
CI11I2 I23 3 =
L L L
I11I2 I23 3 =
2 + 3 1

J2 J3
J1
(1 2 3 )/2

2 + 3 1
L L L

+ 1 I11I2 I23 3 ,
2
(3.63)
(3.64)
with i = ki + i = Ji + ki + i .
Supersymmetry demands that this should extend further to fermionic excitations. Since
we treat the full supersymmetric string field theory later, we do not discuss the supersymmetrization of our arguments explicitly. We only mention that to discuss orthogonality
including fermionic excitations, the spinor equivalent to the inversion tensor j1 j2 2nj1 nJ2
transforming spinor indices between two boundary points is the SO(4) gamma matrix
i i ni . In computing correlation functions, we have to extract SO(4) invariant pieces in similar manner as for bosonic excitations by taking into account the transformation of spinor
indices between two boundary points. This is effectively equivalent with the procedures
adopted in other works.
3.4. Operator representation and the corrections
On the basis of our results, we are now prepared to proceed to the construction of a full
holographic string field theory. Before that, it is convenient to first rewrite the foregoing
results using oscillator representation. By substituting the expression for the general 3point coupling, the effective action is written as
Seff = S2 + S3 ,
(3.65)
S2 =

d
I,L
S3 =

1
1
( I,L I,L I,L I,L ) + (kI + L ) I,L I,L ,
2
R
25
(3.66)
(k2 +2 )/2 (k3 +3 )/2

J2
J3
1
1 J1 J2 J3 I1 ,L1 I2 ,L2 I3 ,L3
d
NR
J1
J1
I1 ,I2 ,I3

k1 !k2 !k3 1 !2 !3 ! I I I L L L

C 1 C 2 C 3 K 1 K 2 K 3 + h.c.,
(3.67)
1 ! 2 ! 3 ! 1 !2 !3 !
where we have recovered the length dimension. The integrals over the angular momentum
J should be understood implicitly. We introduce the braket notation for the field by using
the 8-dimensional oscillator algebra of ai , ai : ai |0 = 0, [ai , aj ] = ij (i = 1, 2, . . . , 8),
where the first 4 directions correspond to the SO(4) indices of the vector directions y,
and the remaining 4 directions (i = 5, . . . , 8) represents the SO(4) indices of the scalar
excitations

I,L ( )|I, L ,
I ( ) ( ) =
(3.68)
I,L
1
|I, L = C iI1 i2 ...i KjL1 j2 ...j
ai ai ai aj1 aj2 aj |0 .
k
k
k!! 1 2
The free term of the action is then

1
1
| |
| + |h
sv |
S2 = d |
2
2
(3.69)
(3.70)
with
1
ai ai ,
R
4
hsv = hs + hv ,
hs =
1
aj aj .
R
8
hv =
(3.71)
j =5
i=1
The parameters familiar in string field theory literature are R 4 /J12 ( )2 = = 1/(p + )2
2 N/J 2 , g = J 2 /N . In terms of the string-length parameter,
+
= gYM
2
(r) = pr , (r) =
1
1
Jr /R 2 . Note that 1/R since pr+ Jr /R. The supergravity limit corresponds to the
2 N )1/4 .
limit 0 with fixed R = (gYM
The interaction term is expressed in terms of the overlap state defined by

4

3
8

1 rs
rs
ai(r) n00 ai(s) +
aj (r) n00 aj (s) |0
|v0 = exp
(3.72)
2
r,s=1
i=1
j =5
with the supergravity part of the familiar Neumann functions in the zero-slop limit 0

J
J
Jr
r
s
rs
1r
nrs
,
nr1
, for r, s = 1, 2 and n11
00 = n00 =
00 = 0. (3.73)
00 =
2
J1
J1
26
The unusual minus sign on the exponent is owing to our phase convention for the creation
annihilation operators. As is well known, the overlap state is the ground-state solution for
the continuity conditions satisfying

3

J2
J3
p(r) |v0 = 0,
x(1) x(2) x(3) |v0 = 0,
(3.74)
J1
J1
r=1
where

Jr
R

a(r) + a(r) ,
p(r) = i
a(r) a(r)
x(r) =
(3.75)
2
2R
2Jr
for all 8 directions. The ground state means that it also satisfy the locality condition of
the form
(x(1) x(2) )|v0 = (x(1) x(3) )|v0 = 0.
(3.76)
We find that
(1) I1 , L1 |(2) I2 , L2 |(3) I3 , L3 ||v0
1 !2 !3 ! L L L
C C C
K 1K 2K 3 .
=
1 ! 2 ! 3 !
1 !2 !3 !
(3.77)
This shows that the 3-point interaction part of the effective action takes the following simple form:

1
(3)
(1)
(2) |(3) | J1 J2 J3 h(2)
S3 =
(3.78)
d (1) |
s + hs hs |v0 + h.c.
2N
The integral with respect to the angular momentum under the conservation condition
J1 = J2 + J3 should be understood implicitly as before. The most important characteristic
of this expression is that the so-called prefactor involves only the energies of scalar excitations. As we have stressed in previous works, the holographic relation of our type does
not necessarily demand that the prefactor itself is the difference of the free Hamiltonian
(2)
(3)
(1)
hsv + hvs hvs . As a consequence of this remarkable property, the 3-point vertices vanish whenever the scalar energies are preserved. This implies that, except for special cases
when the vector energies are simultaneously preserved, the 3-point correlators vanish, as
has been explicitly exhibited in the expression of the CFT coefficient (3.61). Since we have
constraints k1 k2 + k3 and 1 2 + 3 , the conservation of both scalar and vector energies is possible only when the total energies are conserved, namely, when 1 = 2 + 3 .
Note that in this extremal cases we have finite results for the 3-point correlators, owing to
the vanishing denominator in our holographic relation.
Of course, the above action for supergravity modes is valid in the limit 0 with
fixed R. The limit 0 corresponds to the imposition of the locality conditions (3.76)
for the overlap state. However, when we consider a finite , the latter conditions cannot be
preserved, as it should be for general string states which have nonzero extension. Then only
the more general continuity conditions (3.74) are satisfied. Correspondingly, the Neumann
rs
functions nrs
00 for the zero modes must be replaced by N00 , defined by

J2
J1
k2 +2
2
J3
J1
k3 +3
2
k1 !k2 !k3 !
I1
I2
I3

rs
N00
= f nrs
00
=f
rs

r1
= nr1
N00
00 =
Jr
,
J1
Jr Js
J12
27

,
11
for r, s = 1, 2 and N00
= 0,
(3.79)
where f = 1 4(1) (2) (3) K is a nontrivial function of (r) (r = 1, 2, 3). For the
definition of the quantity K, we refer the reader to Appendix B. In the supergravity limit
0, keeping R and Ji fixed, we have f 1. In the opposite large limit, it is known
that
f
(1)
R 2 J1
=
.
4(2) (3) 4 J2 J3
(3.80)
We denote the overlap state with this modification for finite by |V0 .

4

3
8

1
rs
rs
|V0 = exp
ai(r) N00 ai(s) +
aj (r) N00 aj (s) |0 .
2
r,s=1
(3.81)
j =5
i=1
The use of the -corrected 3-point vertex which is given by the same form as (3.78)
but |v0 being replaced by |V0 leads to a correction factor f 1 for the matrix elements,
multiplying the original local form.
On the other hand, the normalization of the CFT coefficients for the BPS operators used
in the foregoing subsections is believed to be exact to all orders in the gauge coupling =
2 N , because of the nonrenormalization property of 3-point functions of chiral operators.
gYM
Indeed, as we stressed before, the CFT coefficients are nothing but the results from the
free field theory, which corresponds to the large limit. This implies that for nonzero ,
the rule of mapping between the matrix elements of the bulk 3-point vertex and the CFT
coefficients must be extended as a natural ansatz to
L L L
CI11I2 I23 3 =
L L L
I11I2 I23 3
2 + 3 1
(3.82)

J J (2 +3 1 )/2
2 + 3 1
L L L
L1 L2 L3 = f 2 3
+ 1 I11I2 I23 3

I1 I2 I3
J1
2
with
L L L
I11I2 I23 3 = (1) I1 , L1 |(2) I2 , L2 |(3) I3 , L3 |
J1 J2 J3 (2)
(1)
R hs + h(3)
s hs |V0 ,
N
(3.83)
(3.84)
and the 3-point interaction term of the action is
J1 J2 J3 (2)
1
(1)
hs + h(3)
S3 =
(3.85)
d (1) |(2) |(3) |
s hs |V0 + h.c.
2
N
This is the main prediction of the present paper. By construction, this gives the same CFT
coefficients for the purely supergravity modes. But for higher stringy modes, the correction
factor would play an important role for impurity nonpreserving processes. The appearance
28
of the factor f with nontrivial dependence is not surprising if we recall that the origin of
the factor (J2 J3 /J1 )(2 +3 1 )/2 is the relation between the length and the cutoff with
respect to the integration, as exhibited in the integral (2.35). Our prediction
indicates that
the nonlocality of the vertex is essentially represented by a rescaling f . Finally, we
also mention that this interaction vertex is not SO(8) symmetric, contrary to a possibility
suggested in [4]. The origin of this phenomenon is the asymmetric roles played by scalar
and vector excitations from the viewpoint of boundary theory, as we have analyzed in the
previous subsection.
4. Holographic string field theory

In our first work [4] on the PP-wave holography, we have emphasized that string field
theory describing the plane-wave background cannot be constructed uniquely by the requirement of supersymmetry alone. Our claim was that there should exist a unique holographic string field theory which realizes our holographic relation. We now have a more
concrete constraint for the holographic string field that it should have the 3-point vertex
(3.85) when restricted to the bosonic supergravity sector.
There are basically two different proposals for 3-point vertex, conforming to the requirement of supersymmetry algebra {Qa , Qa } = 2a a H up to the first order with respect
to the string coupling. Let us first briefly review them. The most familiar is the one which
was first proposed in [15,16] as a generalization of the light-cone string field theory in flat
spacetime, as constructed by Brink, Green, and Schwarz long time ago. This proposal was
corrected and established in subsequent works [1720]. We denote this vertex by
|H3 SV = PSV |E ,
(4.1)
where |E is the overlap vertex including both bosonic and fermionic oscillators. The explicit form of the prefactor PSV is given in Appendix B, where all of the other necessary
definitions and formulas of string field theory are summarized, in order to make the present
paper reasonably self-contained. It has been shown that this vertex has a nontrivial relation with the matrix of operator mixing which appears in the perturbative computation of
anomalous dimensions on the gauge-theory side. We will come back to this point later.
Another version of the 3-point vertex, which was proposed [23] later in connection with
the duality relation of our type, takes the following form:
(2)
(3)
(1)
|H3 D = H2 + H2 H2 |E ,
(4.2)
where
H2(r) =
1
n(r) an(r) an(r) + bn(r) bn(r)
|(r) | n=
with n(r) =
n2 + ((r) )2 (4.3)
are the free Hamiltonian operators. Here the string-length parameters are chosen such that
(1) + (2) + (3) = 0 with (2) , (3) > 0 and (1) < 0 to be consistent with our conventions
used in the foregoing analyses. Note that this Hamiltonian coincides with the total free
Hamiltonian hsv , (3.71), of the previous section when it is restricted to supergravity modes
29
with the identification = 1/R. The form (4.2) is the simplest realization of the way
which was discussed in our first work [4] to obtain susy-compatible interaction vertices. In
particular, it satisfies the SO(8) symmetry suggested there. The authors of [23] has shown
that this vertex gives nontrivial matrix elements in the leading order of the 1/-expansion
for impurity preserving processes, following the duality relation of our type
J1 J2 J3 1
1, 2, 3|H3 ,
(2 + 3 1 )C123 =
(4.4)
N
when the mixing of gauge-theory operators is ignored. This duality relation itself for
impurity-preserving processes was first proposed in [26] from a viewpoint which is entirely
different from ours, and was actually abandoned later since it turned out that the relation
was not appropriate for the original purpose after correcting a sign error in some earlier
versions of works on PP-wave string field theory. This relation is in fact obtained from our
holographic relation in the limit of large for arbitrary 3-point processes when the number of impurities is preserved. We stress that in the light of our works, the more general
relation summarized in (3.82)(3.84) has a firm physical foundation from the holographic
principle. We should also recall that we are using the Euclidean picture of the tunneling
and hence this form of the interaction proportional to energy difference must not be thrown
away by a possible nonlinear unitary transformation: the energies are not conserved for
the Euclidean S-matrix. For a detailed discussion of our S-matrix picture, we invite the
reader to [4].
These two candidates for the 3-point vertex conform to different duality relations connecting bulk and boundary. The first type |H3 SV does not fit to our holographic relation.
The second type |H3 D does not either, though motivated by (4.4), since for all of our arguments from Sections 2 and 3, the gauge-theory operators have to be definite conformal
operators of which 2-point functions are diagonalized and 3-point correlators take the standard conformal form (3.8). However, it is important here to recall that the requirement of
supersymmetry algebra to the first order in the string coupling puts only a linear constraint.
Hence, any linear combination of two possible vertices with some global
de
coefficients
pending only on string length parameters and the curvature parameter /R can
be an allowed candidate for the 3-point vertex.
We can now study whether it is possible to obtain our effective action (3.85) for supergravity modes from these vertices. Let us first examine how the above two vertices behave
when they are restricted to purely supergravity modes. For purely bosonic external states,
they take the following forms:
8

4
1 i
2 i
2
2
2
XI XII
XI XII
|H3 SV
|Ea
2
i=5
=1

3
8
(r)
(r)

m

m (r)i (r)i
(r)i (r)i
=
a
am
a
a
|Ea
(r) m
(r) m m
r=1 i=5 m=0
m=1

3
4
(r)
(r)

m
m
(r)i
(r)i
(r)i
(4.5)
a (r)i am
a
a
|Ea ,
(r) m
(r) m m
r=1 i=1
m=0
m=1
30
8

4
1 i
2 i
2
2
2
XI + XII
+
XI + XII
|H3 D
|Ea
2
i=5
=1

3
8
(r)
(r)

m
m
(r)i
(r)i
(r)i
=
a (r)i am
+
a
a
|Ea .
(r) m
(r) m m
r=1 i=1
m=0
(4.6)
m=1
The state |Ea is the bosonic part of the overlap state. The expression XI and XII , defined
in Appendix B, contain the oscillators of only cos and sin modes, respectively, which are
(r)
(r)
represented by the creation-and-annihilation operator of positive (an ) and negative (an )
indices, respectively. As briefly discussed there, the factorization formula for the Neumann
matrices [17,21] allows us to derive the above simple expressions in terms of world-sheet
r .
energies m
By restricting to the zero modes, we see that neither of them separately reduces to the
effective action of the previous section. However, it is now evident that if we combine them
with equal weights we can get the desired form
1
|H3 h |H3 SV + |H3 D
2

8
4
1 i
2 i
2
XI +
XII
|Ea
2
i=5
i=1
8

3
4
(r)
(r)

m
m
(r)i
(r)i
(r)i
=
(4.7)
a (r)i am
+
a
a
|Ea ,
(r) m
(r) m m
r=1
i=5 m=0
i=1 m=1
(2)
(3)
(1)
which in the zero-mode sector reduces to the form (hs + hs hs )|V0 , involving only
the scalar oscillators in the prefactor, as derived in the previous section. Apart from the
overall normalization fixed by comparing with (3.85), this combination is unique. Therefore, we have arrived at the uniquely possible string field 3-point interaction term which is
consistent with the conclusion that we have reached by foregoing discussions.
We note that when restricted to the scalar modes the form (4.7) coincides with the one
suggested previously in [33] as the possible prefactor which is compatible with the holographic relation (4.4). Namely, they have shown that this gives the correct CFT coefficients
after taking into account the operator mixing for scalar operators with two impurities. The
result (4.7) shows that the scalar prefactor indeed consists only of the cos modes as they
have proposed, while actually the vector prefactor must consist only of the sin modes, in
order to be consistent with supersymmetry.
Thus the conclusion of this section is that the holographic string field theory is given in
the following form up to the first order in the string coupling
Sh = Sh,2 + Sh,3 ,

1
1
Sh,2 = d | | | | + |H2 | ,
2
2

1
(2) |(3) | J1 J2 J3 |H3 h + h.c.
Sh,3 =
d (1) |
2N
(4.8)
(4.9)
(4.10)
31
The integral over Ji under the condition J1 = J2 + J3 is implicit as before.

This action has no SO(8) symmetry, nor even the Z2 symmetry. We also note that the
holographic string field theory does not directly reduce to the familiar GreenSchwarz
form in the flat limit R . This was the case already in the supergravity sector. Since
the AdS/CFT correspondence between bulk gravity (closed string theory) and CFT on the
boundary is a concept which requires a global consideration of the whole AdS spacetime,
the fact that the flat limit is not direct is not surprising.
5. Relation with other possible duality maps

In the present section, we discuss the relation of our results to other proposals for duality maps, especially the one advocated in Refs. [24,29]. As we have already mentioned,
it has been shown that the 3-point vertex |H3 SV of the first type has a close connection
with the matrix of operator mixing in the perturbative computation of anomalous dimensions. It is indeed reasonable to assume that there exists a quantity in the bulk which plays
the role of the operator mixing. The operator mixing is necessarily associated with perturbative renormalization procedure in higher orders with respect both to gauge coupling
and to genus expansion in our situation where there is a large degeneracy with respect to
conformal dimensions at the lowest order.
The first indication for the necessity of taking into account the operator mixing comes
from the behavior of 3-point functions at the first order in /J 2 . Take, for instance,
the simplest case of the 3-point function of operators with two impurities,

J
yJ
(1y)J
(x3 )
O ij,n (x1 )Oij,m (x2 )Ovac

x31 x12

2
,
= g2 Cn;my 1 an;my log(x12 ) + bn;my log
(5.1)
x23
where
J

2inl
1
J
Oij,n
=
e J Tr i Z l j Z J l
J N J +2 l=0
J is just the BMN operator corresponding to the ground state, with
and Ovac
bn;my = n(n m/y),

an;my = m2 /y 2 ,

1 y sin2 (ny)
Cn;my =

.
yJ 2 n m 2
y
(5.2)
For notational brevity, in the present section, we suppress the classical part (namely, the zeroth order in ) of spacetime dependence for gauge-theory correlation functions. Note that
yJ
an;my is noting but the anomalous dimension of Om since it comes from Feynman diayJ
grams where the interaction occurs only between OnJ (x1 ) and Om (x2 ), while bn;my comes
yJ
from graphs where the interaction occurs among all three operators OnJ (x1 ), Om (x2 ) and
(1y)J
Ovac
(x3 ). This form, however, does not take the standard form of a 3-point correlation
32
function of conformal operators. In other words, to this first order in , the BMN operator
J cannot be regarded as a conformal operator characterized by the standard conformal
Oij,n
transformation property.
As is well known now, this difficulty is resolved if we take into account the operator
mixing with multi-trace operators. Consider a 2 2 matrix of two-point functions of operators,
J

2inl
1
J
=
e J Tr i Z l j Z J l ,
Oij,n
J N J +2 l=0
J,y
yJ
(1y)J
Tij,n = :Oij,n Ovac
:.
(5.3)
With respect to the genus-expansion parameter g2 J 2 /N , the diagonal elements begin

from the zeroth order, while the off-diagonal elements, being interpreted as a 2-body to
1-body (or 1-body to 2-body) process, starts from the first order. Therefore, we can express
the general structure of this matrix to the first order both in and g2 as [2528]

O A (0)OB (x) GAB + AB log(x)2 ,
(5.4)
with

GAB =
mn
0
0
mn yz

+ g2
0
Cn;my
Cn;my
0

,
(5.5)

AB =

n2 nm
0
2
m
0

y 2 nm yz

0
+ g2
(an;my + bn;my )Cn;my
(an;my + bn;my )Cn;my

0

.
(5.6)
Note hat the 2-point functions between a single-trace and a double-trace operator are
obtained by taking the limit x23 1/ 0, and also that 1/ is identified with the shortdistance cutoff parameter associated with loops in the sense of Feynman diagrams.
The specific ansatz proposed in [29] in order to relate this mixing matrix to string-field
theory vertex |H3 SV is as follows. We first transform the basis for these operators by
O A = U AB OB such that

O A (0)O B (x) = AB + O( ).
(5.7)
However, this requirement alone does not determine the basis completely. To restrict the
basis further the authors demand that the transformation should be symmetric and real.
Then, the transformation matrix U can be fixed to the present order O(g2 ) as
g2
U = 1 G(1) ,
(5.8)
2
where G(1) is the O(g2 ) part of GAB . By this change of the basis, the matrix AB is
transformed to
g2 (0) (1)
,G
= (0) + g2 (1)
(5.9)
2
to the present order of approximation. The observation of [29] is that the off-diagonal
part of this matrix coincides with the matrix elements of the interaction vertex of the first
33
type, namely, |H3 SV . This has been confirmed in several cases, for instance, for scalar
impurities in [14,29,30,34], and for other cases including vectors and fermions in [35,36].
For the supergravity chiral operators, the mixing matrix AB vanishes. In connection
with this, we note that in this case the matrix GAB should not be interpreted as representing operator mixing, as we will touch briefly below. This is as it should be, since the
3-point functions of the chiral supergravity operators are not renormalized and hence take
the standard conformally-invariant form even after including the higher-order effects.
We now show that this ansatz can be understood as a consequence of our holographic
string field theory, given the observation of [23]. In other words, all existing duality relations are actually compatible. Since our results are based on a clear spacetime picture
which is lacking, unfortunately, in other approaches, checking the consistency with more
formal correspondences seems to provide a useful guide for obtaining a unified viewpoint
on the holography in the plane-wave limit.
Let us consider a general class of three general operators O1 (x1 ), O2 (x2 ) and O3 (x3 ),
0
which have the classical conformal dimensions r satisfying the condition of degeneracy
0
0
0
1 = 2 + 3 . The number of impurities contained in O1 is thus equal to the sum of
those in O2 and O3 . Here and in what follows, we use the subscript 0 to denote the order
with respect to . The order with respect to gs will be denoted by the usual subscript (0),
(1), etc. The 3-point function of them takes the following general form

O 1 (x1 )O2 (x2 )O3 (x3 )

x31 x12
0
1
1

2
2
,
= g2 C123 1 2 log(x12 ) + 3 log(x13 ) + b123 log
x23
(5.10)
0
where C123 denotes the 3-point coefficient of the free gauge theory and r is the
anomalous dimension of the operator Or (xr ) to the first order in . As above, this expression in general does not conform to the standard form of the 3-point correlation
functions of conformal operators. However, we can easily check that this is the most
general form which would lead to the standard form ((5.16) below) after taking into account the operator mixing. In the O( ) part on the right-hand side of (5.10), the first term
1
2 log(x12 )2 comes from a class of Feynman diagrams in which the interaction occurs
0
only between O2 (x2 ) and O1 (x1 ). Similarly, the second term 3 log(x13 )2 comes from
the ones where the interaction occurs only between O3 (x3 ) and O1 (x1 ), and the third term
b123 log((x31 x12 )/x23 ) from the remaining ones, where the interaction occurs among all
three operators. Supposing that the double-trace operator in consideration is obtained from
the product of O2 (x2 ) and O3 (x3 ) by taking the limit x23 1/, the matrices GAB and
AB in the subspace of operators O1 (0) and :O2 (x)O3 (x): takes the form

0
0
C123
1 0
,
+ g2
GAB =
(5.11)
0
0 1
C123
0

1
0
1
AB =
1
1
0
+ 3
2
1
1
0
0
(2 + 3 + b123 )C123
+ g2
(5.12)
,
1
1
0
(2 + 3 + b123 )C123
0
34
where we have taken into account the fact that the O(g20 ) part of AB gives the anomalous
dimensions of O1 and :O2 O3 :. In the case of two different scalar impurities discussed
1
1
1
0
above, 2 = m2 /y 2 , 3 = 0, 1 = n2 , b123 = bn;my = n(nm/y), and C123 = Cn;my .
Now, in order to extract the correct CFT coefficient with the operator mixing being taken
into account, we introduce the transformation matrix UAB which diagonalizes the matrix
O A (0)OB (x) as a whole. Namely, contrary to the previous U , both of the matrices GAB
and AB are simultaneously diagonalized. It takes the form

0
D123
1 0
,
U=
+ g2
(5.13)
E123
0
0 1
where
1
D123 =
2 + 3 1 + b123
1
1
E123 =
1
2
1
3
b123
1
1
1
2
1
3
C123 ,
C123 .
(5.14)
Then the 3-point function of the operators OA = UAB OB in the new basis is given by

O 1 (x1 )O2 (x2 )O3 (x3 )

= O 1 (x1 )O2 (x2 )O3 (x3 ) + g2 D12 3 :O2 O3 :(x1 )O2 (x2 )O3 (x3 ) + O g22
0
= g2 C123 + D123
1 0
0
g2 22 C123 + D123 + b123 C123 log(x12 )
1 0

0
0
+ 23 C123 + D123 + b123 C123 log(x31 ) b123 C123 log(x23 ) .
(5.15)
Here, the mixing effect does not affect the third term log(x23 ), since a correlation function which contains a double trace operator at x2 or x3 gives a O(g2 ) contribution by itself.
Taking into account the definition of D123 , we then obtain the following result regardless of
the expression of b123 , which is consistent with the canonical form of the 3-point function
1
1
1
for operators with (anomalous) conformal dimensions 1 , 2 , and 3 , respectively:

O 1 (x1 )O2 (x2 )O3 (x3 )

1
1
1
= g2 C123 1 1 + 2 3 log(x12 )
1
1

1
1
1
1
+ 3 + 1 2 log(x31 ) + 2 + 3 1 log(x23 ) ,
(5.16)
0
where the true CFT coefficient C123 = C123 + D123 is expressed in terms of C123 and b123
as
0
C123 =
b123 C123
.
1 2 3
(5.17)

1
35
1
Here, we have used the relation 1 2 3 = (1 2 3 ), which is valid

for impurity-preserving processes. Though we have taken into account the operator mixing
to higher order with respect to both g2 and , the correction to the CFT coefficients thus
0
starts from the lowest order, namely the same order as C123 .
Note that this argument can be applied for any kind of impurities, except for the case
0
of pure chiral operators where a123 , b123 and i all vanish: the relation (5.17) holds
with different bn;my s depending on impurities [14,35,36]. Also, for the chiral operators of
supergravity, this procedure would lead to a nonsensical result C123 = 0, indicating that for
the supergravity BMN operators the matrix GAB cannot be regarded as the mixing matrix.
By identifying the true CFT coefficient with the 3-point vertex |H3 h in accordance with
our result for the holographic string field theory, we must have

1
1
0
1, 2, 3|
(5.18)
|H3 D +
|H3 SV = b123 C123 .
2
2
On the other hand, the result of [23] indicates that
1
0
1, 2, 3|H3 D = (2 + 3 1 )C123 .
(5.19)
The off-diagonal matrix elements of the -matrix in the particular basis which makes the
partial diagonalization are given in the form
1
0
0

off
= (2 + 3 1 )C123 + b123 C123 .
2
Using Eqs. (5.17), (5.18) and (5.19), we finally find

off
=
1
1, 2, 3|H3 SV .
2
(5.20)
(5.21)
This is nothing but the claim made in [29,30], except for the overall factor 1/2. This factor can be understood from the difference of normalization. Our convention differs just by
this factor from the one adopted in the literature discussing this subject; see, for example,
(B.14) and (B.9) in [36].
It should be remarked that the above relation of the 3-point vertex |H3 SV with the operator mixing in perturbation theory at the boundary is intrinsically restricted to the processes
where the numbers of impurities are conserved. For this particular class of processes, our
argument clarified why the correct interaction vertex of the holographic string field theory
must be the particular combination of the two types of string interaction vertices: roughly
speaking, the part |H3 D describes the bare part of the interaction of BMN operators,
while the part |H3 SV describes the mixing among them. Both are necessary for describing the processes in the bulk, corresponding to a propagation of them from boundary to
boundary along the tunneling null geodesic. Note that the observation of [23] that the
string overlap vertex, the bare part of the interaction, in the large limit precisely corresponds to the free-field contraction is also quite natural for impurity preserving processes.
Through the holographic string field theory, this natural property is related to the specific
ansatz relating |H3 SV to the matrix of operator mixing, which singles out the particular
basis of the gauge-theory operators.
36
We warn the reader, however, that the above intuitive interpretation on the different roles
of |H3 SV and |H3 D does not apply to more general processes in which the numbers of
impurities are not conserved. It is difficult to extend the above argument directly to such
cases.
6. Explicit examples
The purpose of this section is to present some concrete computations in order to confirm
our general discussions. In the present work, for simplicity we restrict ourselves to the cases
of two (conserved) impurities, for which we can utilize many results by other authors
on the gauge theory side. It is sufficient to focus on the cases of vector, mixed scalarvector, and fermionic impurities, since as we have mentioned before the case of pure scalar
impurities has been practically treated in [33] and was shown to be consistent with our
holographic relation. We are planning to study more general cases, especially the cases
where the numbers of impurities are not conserved, in a forthcoming work. In the present
section, we denote the first 4-vector indices (i = 1, . . . , 4) by Greek letters (, ) for a clear
discrimination between the vector and scalar directions.
6.1. Vector impurities
Let us begin from the BMN operators with two vector impurities. The CFT coefficients
on the gauge theory side have been already computed in Ref. [14]. Nontrivial processes are
listed bellow for (2 + 3 1 )C123 :
sin2 (ny) m/y
,
y 2 n m/y
m m + vac n n :
Cvac
m m + vac n n :
Cvac
m m + vac n n :
sin2 (ny)

Cvac
,
2
y 2
(6.3)
m m + vac n n :
sin(ny) n2 + 3m2 /y 2

Cvac
,
2
y 2
n2 m2 /y 2
(6.4)
sin2 (ny) m/y

,
y 2 n + m/y
(6.1)
(6.2)
where the left-hand side represents symbolically the processes with two vector impurities
with and supposed to denote different vector indices. We have already used the fact
that the difference of the conformal dimensions, 2 + 3 1 , is given by (m2 /y 2 n2 )
with y J2 /J1 at the leading order of small = 1/((1) )2 . The mode numbers n and
m are supposed to satisfy n > 0 and m 0, respectively. The overall numerical constant
Ji
(Ji N Ji )1/2 Tr(Z Ji ) with
Cvac is the 3-point function for vacuum BMN operators, Ovac
J2 + J 3 = J1 :
J1 J2 J3
y(1 y)
Cvac =
(6.5)
= g2
.
N
J1
37
Note that the normalization constant of the BMN operator has always been chosen such that
two-point functions take the form O J (x)O J (0) = 1/|x|2J . This overall factor coincides
with the corresponding factor for the 3-point vertex of the holographic string field theory,
as determined in the previous section on the basis of the comparison with the supergravity
analysis.
The 3-point vertex on the string-theory side which should be compared with is the one
with the prefactor with only sin modes, since the scalar part XI2 vanishes for these cases:
XII2 |E =
(r)

rs
(r) (s)
m
(r) (s)
rs
n |E
N N mn
m n + m
(r) mn
(6.6)
r,s=1 m,n=1
3
(r)
(s)

(r) (s)
m
n rs
rs
Nmn N mn
m n |E ,
+
(r)
(s)
r,s=1 m,n=1
where we have expressed the formula in terms of the exponential basis defined as
0 = a0 ,
1
n = (an ian ),
2
1
n = (an + ian ),
2
(6.7)
which directly corresponds to the momentum basis of the BMN operators on the boundary.
For the first process (6.1), assuming m and n are nonzero, we obtain the following
matrix element
(1) (1) (2) (2)
n m m
123 0|n
1 2
X |E
2 II
(1)

(2)
m
21
1 n 12
12
21
12
Nnm N nm
+
N mn N mn
N nm
2 (1)
(2)
(1)

(2)
m
21
1 n 12
12
21
12
Nnm N nm
+
N mn N mn
+
,
N nm
2 (1)
(2)
(1)
(6.8)
(2)
where the second line comes form the case where the oscillators n and m are con(1)
(2)
tracted through the prefactor XII2 and the operators n
and m
through the overlap |E ,
while the third line comes form the opposite case. The net results is
(1)
(2)
1 n
m 12
12 12
Nnm N nm
Nnm .
+
(1)
(2)
(6.9)
Using the explicit form of the Neumann coefficients in the large limit presented in (B.60),
we can confirm that this reduces in this limit to the gauge theory result (6.1), after including
the above overall factor. As for the special case m = 0, the absence of zero-mode oscillators
in the prefactor XII leads to the vanishing result, which matches the gauge theory.
The second case (6.2) is related by a change m m to the first one. On the string
rs satisfies the same relation
theory side, the large behavior of Neumann functions N mn
with respect to this sign change, as exhibited in (B.60). Thus the first case ensures the
correct matching in the second case.
38
For the third case (6.3), we obtain for n = 0 and m = 0,

(1) (1) (2) (2)
n m m
123 v|n
1 2
X |E
2 II
(2)
22
22
11
n(1) 11
m
11
22
Nnn N nn
N mm
N mm N mm
N nn ,
(1)
(1)
(6.10)
(1)
(1)
and n
where the first term in the second line comes form the contraction of n
(2)
(2)
through the prefactor XII2 and of m and m through the overlap |E , while the second
term comes from the opposite case. Due to the property of the Neumann coefficients, the
second term vanishes, and the first term reduces to the field theory result. It is useful to
notice here that the gauge theory result for vector impurities in the case (6.3) is equal to the
scalar correspondent (, i, j ) except for the overall sign, and to recall that the scalar
case matches the duality relation (4.4) by using the vertex (1/2)XI2 |E . In addition to this,
we can easily check that the relation

2
1
(1) (1) (2) (2)
2
X
|E
=
O
v|
+
X
(6.11)
123
n m
m
n
I
II
2
is satisfied in the present flavor-changing process. Thus, XII2 |E can be replaced with
XI2 |E at the leading order in large limit, confirming the validity of the duality relation.
The same is true for the case of m = 0.
The last case (6.4) is the sum of the above three cases (6.1), (6.2) and (6.3) on the
gauge theory side. We can check that the same is true for the expressions on the string side,
indeed,
2
(1) (1) (2) (2) 1

X |E ,
n m m
(6.12)
123 v|n
2 II
(1)
(1)
(2)
(2)
is given by summing all possible contractions of n , n , m and m through the

prefactor XII2 or the vertex |E , which coincide with the sum of all the three amplitudes
considered above on the string side in terms of the Neumann functions.
6.2. Mixed impurities
Next, we consider the case with one vector and one scalar impurities, namely, im m +
vac in n . The gauge theory result for (2 + 3 1 )C123 is [14]

m 2
sin2 (ny)

Cvac 2 2
n
+
(6.13)
.
2
y
y (n m2 /y 2 )
On the string theory side, both the XI and XII parts in the prefactor in |H3 h contribute as

8
4

2

2
(1)i (1) (2)i (2) 1
i
XI +
XII
|E
123 v|n n m m
2

n(1)
(2)

m
1
+
2 (1)
(2)
i=5
=1
12
12 12
N mn
Nmn
+ N mn
(1)
(2)
1 n
m 12
12 12
Nmn N mn
Nmn ,
+
2 (1)
(2)
39
(6.14)
where the first part in the second line is the contribution from (XIi )2 |E and the second
from (XII )2 |E , and the net result is

(1)
(2)
1 n
m
12 12
+
Nmn .
N mn
(1)
(2)
(6.15)
We can easily confirm that in the large limit this precisely reduces to the field theory
result (6.13).
6.3. Fermionic impurities
We first explain the convention for representing spinor impurities. We essentially follow Ref. [36]. Decompose the SU(4) R-symmetry group as SU(4) = SO(4) U(1) =
SU(2) SU(2) U(1) with U(1) being the subgroup corresponding to large orbital angular momentum J :
4 (2, 1)+ + (1, 2) ,
where the subscript represents the U(1) charge. With this decomposition of the R-charge
A
index, the correspondence between the fermionic fields A
and , with A and being
R-charge and Lorentz spinor indices, and the string theory creation operators b 1 2 and
b 1 2 with SO(8) (= SO(4) SO(4) = SU(2) SU(2) SU(2) SU(2)) indices is given
by
A
(6.16)
(r,1/2 , r ,1/2 ) b1 2 , b 1 2 ,

, r ,1/2
) b 1 2 , b 1 2 .
A
(6.17)
(r ,1/2
The original SU(4) index A of A
r , 1/2) and simi is represented by the set (r, 1/2) + (
larly of A
by
its
conjugate
(r,
1/2)
+
(
r
,
1/2).
Note
that
the
range
of
indices
are r = 3, 4,
r = 1, 2, 1 , 2 = 1, 2, and 1 , 2 = 1, 2. On the right end of the above symbolic relation,

the fermion oscillators are represented by the indices (r 1 , 2 ), etc., with being
originally the spinor indices of 4D target space on the boundary. We call the sector with the
U(1) charge J = 1/2 a BMN fermion while the one with J = 1/2 an anti-BMN fermion,
and we will focus on the former. The SU(2) indices are contracted in the standard way by
1
.
the symbol and
As for two fermionic impurities, it is sufficient to consider the four types listed bellow,
accompanied with the corresponding gauge theory results for (2 + 3 1 )C123 which
can be extracted from the work [36]:
31m 32m + vac 31n 32n : Cvac
sin2 (ny)
n
,
2
n m/y
y
(6.18)
31m 31m + vac 31n 31n : Cvac
sin2 (ny) 2nm/y

,
y 2 n2 m2 /y 2
(6.19)
40
31m 42m + vac 31n 42n :
sin2 (ny) n + m/y

Cvac
,
2
y 2 n m/y
(6.20)
sin2 (ny) m/y

.
y 2 n m/y
(6.21)
31m 41m + vac 31n 41n : Cvac
On the string theory side, the contribution from the |H3 SV part in the interaction vertex
is already calculated in [36] as
1 v|n n(1)11 11,m 12,m |H3 SV = CN
(1)12
(2)
(2)
sin2 (ny)
,
((1) )2 2 y
1 v|n n(1)11 11,m 11,m |H3 SV = 0,

(1)11
(2)
(2)
(1)22
(2)
(2)
v|n n(1)11 11,m 22,m |H3 SV
(1)21
(2)
(2)
(6.23)
= 0,
1 v|n n(1)11 11,m 21,m |H3 SV = CN
(6.22)
(6.24)
sin2 (ny)
,
((1) )2 2 y
(6.25)
where n is the exponential basis of each string defined in a similar manner as n :

1
1
n = (bn ibn ),
n = (bn + ibn ).
(6.26)
2
2
For these cases of pure fermionic external states with undotted indices, the prefactor reduces to
PSV = Y 4 ,
(6.27)
3
0 = b0 ,
where Y 4 is defined as Y 4 Y21 2 Y 21 2 = 12Y11 Y12 Y21 Y22 with Y21 1 = Y1 2 Y12 . See
Appendix B for the definitions of these quantities. Using the explicit form of Y 4 , we can
easily confirm that the amplitudes on the string side vanish in both case of (6.23) and (6.24),
and that (6.25) is equal to the minus of (6.22). With the definition of Y 1 2 in (B.44), the
2
(1)
(2)
string-matrix element (6.22) is given by (G
n G
m ) , which reduces in the large limit
(r) given in (B.65).
to the right-hand side of (6.22) using the asymptotic form for G
Next turning to the contribution from the |H3 D part, we first find
1 v|n n(1)11 11,m 11,m |H3 D
(1)

(2)
2n
2m
1 12
1 12
21 2
21 2
=
+
Qnm + Qmn + Qnm Qmn
(1) (2)
4
4
(1)11
= CN
(2)
(2)
sin2 (ny)
4nm/y
,
2
2
2
((1) ) y n m2 /y 2
(6.28)
12
21
21
where we have used the relation Q12
nm = Nnm , Qmn = Nmn in the large limit,
rs
which can be easily shown by the definition of Qnm , (B.26), and the asymptotic form of
U (r) , (B.65), given in Appendix B. For other three cases (6.18), (6.20) and (6.21), we find
that the |H3 D contribution gives one and the same expression
(1)
(2)
2
2n
sin2 (ny) n + m/y
2m 1 21
+
=
C
(6.29)
Qmn Q12
.
N
nm
(1)
(2) 4
((1) )2 y 2 n m/y
41
Combining these two types of contributions in each case, we obtain the matrix elements
which precisely coincide with the gauge theory results (6.18)(6.21).
7. Concluding remarks
Finally, we remark on some relevant problems left in the present paper and on possible
future directions. First of all, we have to emphasize again that our main predictions are not
restricted to the impurity-preserving processes which almost all of other works have been
limited to. In the supergravity sector, our holographic relation summarized in (3.82)(3.84)
is valid by its construction for general processes. The extremal cases of the supergravity
sector where conformal dimensions are preserved 1 = 2 + 3 are generalized by lifting
the degeneracy, owing to the higher-order effects in , to the impurity preserving processes
including string excitation modes. In the last two sections, we have confirmed that our relation is indeed valid in this case with nontrivial stringy effects. It is quite plausible that
the relation should then be naturally extended to impurity nonpreserving sectors. We can
note, for example, that the -correction factor f in the holographic relation is consistent
with the fact that the CFT coefficients for such cases start from 0th order in 1/ just as the
impurity-preserving processes, while the 3-point string vertices for such cases in general
start at most from the first order in 1/, because of the large behavior of the Neumann
functions. In fact, it is easy to confirm that our ansatz gives the correct results for a few simple cases. However, a systematic check of more general classes of impurity nonpreserving
processes is beyond the scope of the present paper and is left to a forthcoming work.
There are many possible directions following the present work: for instance, in connection with the question of impurity nonpreserving processes, we should investigate the
string-loop corrections. In most of the existing literature, impurity nonpreserving contributions have been ignored often without appropriate justification. It would also be interesting
to extend the discussions of Sections 2 and 3 to higher-point correlation functions, from
both standpoints of supergravity limit and full string theory, and to see to what extent the
structure suggested in our first work [4] is realized in higher orders. Another important
problem is to derive the holographic string-field theory in conjunction with our prescription of the holographic duality directly from the gauge-theory side. For such an attempt,
the collective-field approach discussed in [37] seems to be suggestive.
Acknowledgements
We would like to thank H. Shimada for discussions at an early stage of the present
work. The present work is supported in part by Grant-in-Aid for Scientific Research (No.
13135205 (Priority Areas) and No. 16340067 (B)) from the Ministry of Education, Science
and Culture.
42
Appendix A. Reduction to SO(4) basis

First we summarize the definitions of overlap integrals of SO(6) harmonics,
Y I = CiI1 ...ik x i1 x ik
(A.1)
following the convention of Ref. [11]. Using the formula for the integration over the 5dimensional unit sphere S 5 ( = 3 = the area of a unit 5-sphere),

21m
1
x i1 x i2m =
(A.2)
(all possible contractions),
5
(m + 2)!
S5
we derive

I1
I2
I1 I2
Y Y = Ci1 ...ik Cj1 ...jk x i1 x ik x j1 x jk
S5
S5
= 3
and

21k
I1 I2
k! C I1 C I2 = 3 k1
,
(k + 2)!
2 (k + 1)(k + 2)
(A.3)
2(2)/2 k1 !k2 !k3 ! I1 I2 I3

C C C
Y I1 Y I2 Y I3 = 3
! ! !
2 +2 ! 1 2 3
S5

a(k1 , k2 , k3 ) C I1 C I2 C I3 ,
(A.4)
!k2 !k3 !
since in this case m = /2 and k11!
is equal to the number of same contractions occur,
2 !3 !
due to the total symmetry of the tensors C I .
Now we convert these integrals on S 5 into a Gaussian average for the SO(4) directions
by taking the large J limit. Such a calculation has previously done in Ref. [6] for some
special cases. First by expressing the S 5 harmonics in terms the S 3 harmonics Y I1 ,

iJ
(J + k)!
e
Y I = 2J /2
(A.5)
cosJ sink1 Y I1 ,
J !k!

I
Y =2
J /2
iJ
(J + k)!
e
cosJ sink1 Y I .
J !k!
(A.6)
Let us first check this formula by confirming the normalization integral

I I
Y Y =2
S5
(J + k)!
J !k!
2
/2
d cos | sin |3 cos2J sin2k
d
/2
Y I Y I . (A.7)
S3
In the limit J , this can be evaluated by making a change of integration variables from
and the Cartesian coordinates x i on S 3 ( 3i=1 (x i )2 = 1) to a 4-vector y 0 = , y i = x i
43
as

lim
Y I Y I = lim 2J
J
S5
= 2J
(J + k)!
2
J !k!
2
2
d 4 y eJ y y k Y I
(A.8)
I I
2
(J + k)!
3
k!
C =
2
, (A.9)
C
k
2
J !k!
2J +k1
J2
J + 12
2 J + 12
which coincides with the (A.3).

Similarly, the 3-point integral is found to be

1/2
I1 I2 I3
J1 (J1 + k1 )!(J2 + k2 )!(J3 + k3 )!
Y Y Y =2
J1 !k1 !J2 !k2 !J3 !k3 !
5
S
2 d cos | sin |3 cosJ1 sink1 cosJ2 sink2 cosJ3 sink3

k2 /2 k3 /2
k1 !k2 !k3 ! I I I
J2
J3
1
3
C 1 C 2 C 3 . (A.10)
J1
1 ! 2 ! 3 !
21+ 2 J12 J1
Equating this result with the large J1 limit of (A.4) which is equal to

1 J2 J3 1 5 I1 I2 I3
C C C ,
21+ 2 J 2 1 ! J1
3
(A.11)
we find the relation between the SO(6) and SO(4) contractions ( 1 = 1 = (k2 + k3
k1 )/2),

1 k2 /2 k3 /2

k1 !k2 !k3 ! I I I
I I I
J
J
J
1
2
3
C 1 C 2 C 3 .
C 1 C 2 C 3 = 1 !
(A.12)
J2 J3
J1
J1
1 ! 2 ! 3 !
This is used in Section 3 in order to express the 3-point coupling in terms of the SO(4)
variables explicitly.
Appendix B. Explicit form of string-field vertices

In this appendix, we summarize various formulas which are necessary for our arguments
in Sections 5 and 6. We hope that this is useful to make the expositions of the present paper
self-contained.
B.1. Preliminaries
First, the Fourier mode expansions in terms of sin/cos basis of bosonic coordinate
x (r) (r ) and bosonic momentum p (r) (r ), as well as the fermionic ones (r) (r ) and
44
(r) (r ), are given by

x
(r)
(r)
(r ) = x0

nr
nr
(r)
(r)
+ 2
xn cos
+ xn sin
,
|(r) |
|(r) |
(B.1)
n=1
(r)

1
nr
nr
(r)
(r)
(r)
pn cos
+ pn sin
(r ) =
p + 2
,
2|(r) | 0
|(r) |
|(r) |
(B.2)
n=1
(r)
(r) (r ) = 0 +

nr
nr
(r)
n(r) cos
,
2
+ n sin
|(r) |
|(r) |
(B.3)
n=1

1
nr
nr
(r)
(r)
(r)
(r ) =
n cos
+ n sin
+ 2
,
2|(r) | 0
|(r) |
|(r) |
(r)
(B.4)
n=1
where5

xn(r) =
(r)
a + an(r) ,
(r) n
2n

pn(r) = i
(r)
n (r)
an an(r) ,

2
(r)
1 +

(r)
(r)

0 =
1 + e((r) ) b0 1 e((r) )b0
2
2 |(r) |

(r)
1
(r)

1 e((r) ) b0 + 1 + e((r) )b0
,
+
2
n(r)
(B.5)
(B.6)

n 1 + 1/2 (r)
1/2 (r)
U(r)n bn + e((r) n)U(r)n bn
=
(r)
2
2|(r) | n

1 1/2 (r)
1/2 (r)
+
U(r)n bn + e((r) n)U(r)n bn
2
(n = 0),
(B.7)
(r)
n =
|(r) | (r)
,
n
(B.8)
with the ordinal (anti-)commutation relations

(r)i (s)j
(r)a (s)b
= rs ij mn ,
= rs ab mn .
am , an
bm , bn
(r)
(B.9)
5 The definition of the oscillators a

n is different from the usual one in the literature by a factor i. We use
this definition since it is the appropriate one in the supergravity limit as we have discussed in Section 3.

(r)
Here e(x) x/|x|, n
45
(r)
2 ,U
1 2 3 4
n2 + 2 (r)
(r)n (n (r) )/|n|, and ,
with i being SO(8) gamma matrices. With these Fock space basis, the free string Hamiltonian for rth string
(r)
1
=
2
2|
(r) |

d 2 p (r)2 +
2
1
1
x (r) +
2 x (r)2

2
2
1
+
2
2|
(r) |

d 2 (r) (r) +
1 (r)
(r) + 2(r) (r)
2

(B.10)
reduces to
H=
(r)
n (r) (r)
an an + bn(r) bn(r) .
|
|
n= (r)
(B.11)
B.2. Overlap vertex

The overlap vertex takes the form
3

|E = |Ea |Eb
(r) ,
(B.12)
r=1
where |Ea and |Eb are the bosonic and fermionic overlap vertices which satisfy
3
(r)
( )|Ea = 0,
r=1
3
3
e((r) )x (r) ( )|Ea = 0,
(B.13)
e((r) ) (r) ( )|Eb = 0.
(B.14)
r=1
(r) ( )|Eb = 0,
r=1
3

r=1
Here, p (r) ( ) (| | |(1) |) is defined as p (r) ( ) r ( )p (r) (r ) with 2 ( ) =

((2) | |), 3 ( ) = (| | (2) ), and 1 = 1. The parameter r is defined as
2 = ,

3 =
(2) (2) ,
(2) , (2) ((2) + (3) ),

+ (2) , ((2) + (3) ) (2) ,
1 = ,
((2) + (3) ) ((2) + (3) ).
(B.15)
(B.16)
(B.17)
The definitions of x(r) ( ), (r) ( ) and (r) ( ) are given in the same way. We always as+
) satisfies the relation (2) , (3) > 0, (1) < 0.
sume (r) ( p(r)
46
The explicit form of the overlap vertex is

3
1 (r) rs (s)
|Ea = exp
am Nmn an
|va 123 ,
2
(B.18)
r,s=1

3

(r)1 2 rs (s)
(r) 1 2 rs (s)
|Eb = exp
bm
Qmn bn1 2 + bm
Qmn bn 1 2 |vb 123 .

r,s=1 m,n=0
(B.19)
This overlap vertex is based on the ground states |va 123 and |vb 123 which are defined
(r)
(r)
by an |va 123 = 0 and bn |vb 123 = 0 for n Z. Note that |Eb is constructed on the
(r)
Fock vacuum |vb [31,32], not on the SO(8) vacuum |0 , defined as an |0 = 0 (n Z),
(r)
(r)
bn |0 = 0 (n = 0) and 0 |0 = 0, on which the original fermionic interaction vertex
[1618] was constructed. As for the fermionic sector, the SO(8) spinor indices have been
decomposed as SO(8) = SO(4) SO(4) = SU(2) SU(2) SU(2) SU(2), according
to the works [19,20].
rs , N r and the fermionic Neumann coefficients
The bosonic Neumann coefficients Nmn
m
rs
r
Qmn , Qm are given by

(r ) (s )

r s
N00
(B.20)
,
= (1 4K) r s +
(1)

r1
N00
= r 1
(r )
,
(1)
(B.21)

1/2
rs
= 2(s ) s t (t ) C(r) N r m ,
Nm0
(B.22)
Nmr = C 1/2 A(r)T 1 B m ,
(B.23)
1/2
1/2
rs
= rs mn 2 C(r) C 1/2 A(r)T 1 A(s) C 1/2 C(s) mn ,
Nmn
(B.24)

rs
Nmn
= U(r) N rm U(s) mn ,
(B.25)

Qrs
mn
= e((r) )
|(s) | 1/2 1/2 rs 1/2 1/2
U C N C
U(s) mn ,
|(r) | (r)
e((r) ) 1/2 1/2 1/2 r
Qrr
U C C N m,
m0 = r t (r ) (t )
|(r) | (r) (r)

r1
Q1r
00 = Q00 =

(r )
1
,
2
(1)
(B.26)
(B.27)
(B.28)
otherwise = 0,
where n, m > 0,
r , s
Cmn = mmn ,
47
(B.29)
{2, 3}, r, s {1, 2, 3}, (1) (2) (3) , and
C(r)mn = m(r) mn ,
1
K = B T 1 B,
4
3
U(r) = C 1 (C(r) (r) ),
A(r) U(r) A(r)T ,
(B.30)
(B.31)
r=1
2 mn y(1)n+1
=
sin(my),
n2 y 2 m2
(1 y)
2 mn
A(3)
sin my,
mn =
2
n (1 y)2 m2
A(2)
mn
A(1)
mn = mn ,
Bm =
(B.32)
2
m3/2 sin(my),
y(1 y)(1)
(B.33)
with y = (2) /(1) and 1 y = (3) /(1) .

When we compare string amplitudes on the both sides of bulk and boundary in the
plane-wave limit, the appropriate Fock basis is the one spanned by the oscillators defined
with the exponential Fourier mode basis which corresponds directly to BMN operators:
0 = a0 ,
1
n = (an ian ),
2
1
n = (an + ian ).
2
(B.34)
rs , is
The Neumann coefficients in terms of the exponential oscillator basis, N mn
(r) rs (s)
am
Nmn an =
m,n=
(r) rs (s)
m
Nmn n ,
(B.35)
m,n=
where
rs
rs
N 00
= N00
,
1 rs
rs
rs
rs
rs
N 0m
= N m0
= N 0m
= N m0
= N0m
,
2
1 rs
rs
rs
rs
,
= N mn
= Nmn
Nmn
N mn
2
1 rs
rs
rs
rs
N mn
.
= N mn
= Nmn
+ Nmn
2
rs s below in Appendix B.4.
We present the large behavior of N mn
(B.36)
(B.37)
B.3. Prefactors
The prefactor which was first constructed for the overlap vertex based on the SO(8)
vacuum |0 [1518] can be reformulated for the overlap vertex |E which is based on the
48
genuine Fock vacuum [19,20]. The form of this prefactor is6

|H3 SV = PSV |E ,
PSV =
(B.38)

1 i j
K K + ij Vij K K + V
2

K 1 1 K 2 2 S1 2 (Y )S1 2 (Z) K 1 1 K 2 2 S1 2 (Y )S 1 2 (Z) , (B.39)
where K I , K I (I = i, ) and Y 1 2 , Z 1 2 are bosonic and fermionic constituents of the

prefactor defined as
K J = XIJ + XIIJ ,

K J = XIJ XIIJ ,

(1 4K)1/2
Fn(r) an(r) ,
XI = i
(B.40)
(B.41)
r=1 n=0

(r)
(1 4K)1/2
Un(r) Fn(r) an ,
XII =
(B.42)
r=1 n=1

Y
1 2

(r)1 2
(1 4K)1/2
G(r)
,
n bn
(B.43)
r=1 n=0

Z
1 2

(r) 1 2
(1 4K)1/2
G(r)
,
n bn
(B.44)
r=1 n=0
with

(2)
F0 =
2
(2) (3) ,

(3)
F0 =
2
(3) (2) ,

1 1 1/2
1
U(r) C(r) CN r n
Fn(r) =

1
4K
(r)

(2)
G0
1
(2) (3) ,
=

(3)
G0
(1)
F0 = 0,
(n > 0),
1
(3) (2) ,

e((r) ) 1/2 1/2 1/2 r

G(r)
U(r) C(r) C N n
n =
1 4K |(r) |
(B.45)
(B.46)
(1)
(B.47)
(n > 0).
(B.48)
G0 = 0,
6 This definition differs form the one in (3.28) of [19] by a factor 2/ . Note also the difference of the total
factor of K i between here and there.
The other quantities in the prefactor is defined as

1 4
1 4 4
Y + Z4 +
Y Z
Vij ij 1 +
12
144

1
1
i
Yij2 1 + Z 4 Zij2 1 + Y 4 + Y 2 Z 2 ij ,
2
12
4
V

1 4
1 4 4
4
Y +Z +
Y Z
1
12
144

1 4
1
i 2
4
2
+ Y 2 Z 2 ,
Y 1 Z Z 1 Y
2
12
4
i
S(Y ) Y + Y 3 ,
3
49
(B.49)
(B.50)
(B.51)
with
r r
K r r K i i
Y21 1 Y1 2 Y1 2 ,
Y31 2 Y21 1 Y 1 2 ,
ij
Y 2ij Y 21 1 1 1 ,

K r r K i i r r
(r = 1, 2)
Y22 2 Y1 2 Y 1 2 ,
(B.53)
Y 4 Y21 1 Y 21 1 ,
(B.52)
Z 2ij Z 2 1 1
ij
1 1
(B.54)
,
2 2
ij
Y Z
Y 2k(i Z 2j )k . (B.55)
We refer the reader to Ref. [19] for more details.

On the other hand, the interaction vertex presented in [23], which is of the form
8

3
8
(r)

m (r)I (r)I (r)a (r)a
am am +
bm bm
|H3 D =
(B.56)
|E
m= (r)
r=1
I =1
a=1
can be, using the factorization formula, written as

|H3 D = PD |E ,
PD =
(B.57)
1 2
K + K 2 Y 1 2 Y1 2 Z 1 2 Z 1 2 ,
4
(B.58)
where
Y 1 2 =
3

n (r) (r)1 2
,
G b
(r) n n
r=1
Z 1 2 =
3

n (r) (r) 1 2
.
G b
(r) n n
r=1
For the derivation of (B.57), see Appendix B.5 below.
(B.59)
50
B.4. Large behavior [22]

rs is given, for (m, n) = (0, 0), by
The large behavior of N mn
(1)m+n
22
N mn
=
,
4|(1) |y
33
=
N mn
23
=
N mn
1
,
4|(1) |(1 y)
(1)m+1
,
4|(1) | y(1 y)
(1)m+n+1 sin(my) sin(ny)

11
N mn
,
=
|(1) |
(1)m+n+1 sin(ny)
21
N mn
=
,
y(n m/y)
and, for m = n = 0, by
11
N 00
= 0,
12
N 00
= y,
(B.60)
31
=
N mn
(B.61)
(1)n sin(ny)
,
1 y(n m/(1 y))
(B.62)

13
N 00
= 1 y,
1
,
4|(1) | y(1 y)
1
33
N 00
=
.
4|(1) |(1 y)
23
=
N 00
22
=
N 00
(B.63)
1
,
4|(1) |y
(B.64)
(r)
(r)
When we compute string amplitudes for fermions, the large behavior of Fn , Gn ,

(r)
Um and 1 4K are also useful:7
2|(1) |
2|(1) |
Fn(2) = (1)n+1
Fn(3) =
|(1) |(1 y) y,
|(1) |y (1 y),

2|(1) | y(1 y)

Fn(1) = (1)n+1
n sin(ny),
|(1) |

(1)n+1
1
(3)
(2)

,
G
,
=
G
n
n =
2|(1) |y
2|(1) |(1 y)
(1)n+1 2 sin(ny)
(1)

Gn =
,
|(1) |
n
n
2|(1) |
,
Un(3) =
,
Un(1) =
,
Un(2) =
2|(1) |y
2|(1) |(1 y)
n
1
,
1 4K =
4|(1) |y(1 y)
(B.65)
where we have defined

(r)
Gn = (1 K)1/2 G(r)
n .
7 Note that the definition of the Neumann vector N r here, with which F (r) and G(r) are defined, differs by
n
n
n
1/2
C(r) Ur from that of the Ref. [22].
51
B.5. Factorization formula

We first prove the formula
3
(r)

n
r=1 n=0
1
an(r) an(r) |E = XI2 |E ,
(r)
2
(B.66)
using the factorization formula obtained in [17,21]. Operating the annihilation operator an
on the vertex |E , the left-hand side of (B.66) can be written as
3
(r)

n
a (r) a (r) |E
(r) n n
r=1 n=0
(r)
3
(r)

0
0
(r)
rs (r) (s)
N00
a0 a0 +
N rs a an(s)
=
(r)
(r) 0n 0
r,s=1
n(r)
n=1
(r)
rs (r) (s)
Nn0
an a0
n=1
n,m=1

(r)
m
rs (r) (s)
N a a
|E .
(r) mn m n
(B.67)
By the definition of Neumann matrices, the first term in the right-hand side becomes
3
(r)

0
1
(r) (s)
N rs a a0 = X02 ,
(r) 00 0
2
(B.68)
r,s=1
and the sum of the second and the third terms reduces to
3
3
(r)
(r)

0
n
(s)
rs (r) (s)
N0n
a0 an +
N rs a (r) a
= X0 X+ ,
(r)
(r) n0 n 0
r,s=1 n=1
(B.69)
r,s=1 n=1
where X0 and X+ is defined as zero-mode and positive-mode parts of XI , such as

3

3
(r) (r)

1/2
(r) (r)
F0 a 0 +
Fn an
XI = i (1 4K)
r=1
r=1 n=1
X0 + X + .
(B.70)
rs
Nnm
sr , and the factorization formula [17,

= Nmn
Using the property of the Neumann matrix,

21],
1 1/2
1 1/2
rs
Nnm
U(r) C(r) CN r n U(r)
=
C(s) CN s m ,
(s)
(r)
1 4K (r) m + (s) n
(B.71)
the fourth term can be written as
3
3
(r)
(r)
(s)

n
1 n
rs (r) (s)
rs
rs m
Nnm
an am =
Nnm
+ Nnm
a (r) a (s)
(r)
2
(r)
(s) n m
r,s=1 n,m=1
r,s=1 n,m=1
1 2
= X+
.
2
Combining all the results above, we obtain the formula (B.66).
(B.72)
52
Noticing that the Neumann coefficient with negative Fourier modes is given by

rs
= U(r) N rs U(s) mn (m, n > 0)
Nmn
(B.73)
and the definition of XII , which has the extra iU(r) factor compared with XI , we can easily
see that the similar formula for negative modes,
3
(r)

n
r=1 n=1
1
(r) (r)
an an |E = XII2 |E ,
(r)
2
(B.74)
can also hold.

With the definition of the fermionic Neumann coefficient Qrs
mn , it is not difficult to prove
the formula
3 (r)
n (r) (r)1 2
(r)
bn1 2 bn
+ bn 1 2 bn(r) 1 2 |Eb
r=1 nZ (r)

= Y1 2 Y 1 2 + Z 1 2 Z 1 2 |Eb
(B.75)
in the same manner as the bosonic case.
References
[1] D. Berenstein, J.M. Maldacena, H. Nastase, Strings in flat space and pp waves from N = 4 super-Yang
Mills, JHEP 0204 (2002) 013, hep-th/0202021.
[2] S.S. Gubser, I.R. Klebanov, A.M. Polyakov, Gauge theory correlators from noncritical string theory, Phys.
Lett. B 428 (1998) 105, hep-th/9802109;
E. Witten, Anti-de Sitter space and holography, Adv. Theor. Math. Phys. 2 (1998) 253, hep-th/9802150.
[3] For reviews, see, e.g.,
A. Pankiewicz, Strings in plane wave backgrounds, Fortschr. Phys. 51 (2003) 1139;
J.C. Plefka, Lectures on the plane-wave string/gauge theory duality, hep-th/0307101;
D. Sadri, M.M. Sheikh-Jabbari, The plane-wave/super-YangMills duality, hep-th/0310119;
A.A. Tseytlin, Spinning strings and AdS/CFT duality, hep-th/0311139;
R. Russo, A. Tanzini, The duality between IIB string theory on PP-wave and N = 4 SYM: a status report,
hep-th/0401155.
[4] S. Dobashi, H. Shimada, T. Yoneya, Holographic reformulation of string theory on AdS5 S 5 background
in the PP-wave limit, Nucl. Phys. B 665 (2003) 94, hep-th/0209251.
[5] T. Yoneya, What is holography in the plane wave limit of AdS(5)/SYM(4) correspondence?, hepth/0304183, expanded from the paper published in the Proceedings, Prog. Theor. Phys. 152 (2003) 108.
[6] N. Mann, J. Polchinski, ADS holography in the Penrose limit, hep-th/0305230.
[7] M. Asano, Y. Sekino, T. Yoneya, PP wave holography for Dp-brane backgrounds, Nucl. Phys. B 678 (2004)
197, hep-th/0308024;
M. Asano, Y. Sekino, Large N limit of SYM theories with 16 supercharges from superstrings on Dp-brane
backgrounds, hep-th/0405203.
[8] Y. Sekino, T. Yoneya, Generalized AdS/CFT correspondence for matrix theory in the large N limit, Nucl.
Phys. B 570 (2000) 174, hep-th/9907029;
Y. Sekino, Nucl. Phys. B 602 (2001) 147, hep-th/0011122.
[9] D.Z. Freedman, S.D. Mathur, A. Matusis, L. Rastelli, Correlation functions in the CFTd /AdSd+1 correspondence, hep-th/9804058.
[10] E. DHoker, D.Z. Freedman, S.D. Mathur, A. Matusis, L. Rastelli, Extremal correlators in the AdS/CFT
correspondence, hep-th/9908160.
53
[11] S.-M. Lee, S. Minwalla, M. Rangamani, N. Seiberg, Three-point functions of chiral operators in D = 4,
N = 4 SYM at large N , Adv. Theor. Math. Phys. 2 (1998) 697, hep-th/9806074.
[12] T. Yoneya, see http://www2.yukawa.kyoto-u.ac.jp/str2003/talks/yoneya.pdf.
[13] E. DHoker, J. Erdmenger, D.Z. Freedman, M. Prez-Victoria, Near-extremal correlators and vanishing supergravity couplings in AdS/CFT, hep-th/0003218.
[14] C.S. Chu, V.V. Khoze, G. Travaglini, BMN operators with vector impurities, Z(2) symmetry and pp waves,
JHEP 0306 (2003) 050, hep-th/0303107.
[15] M. Spradlin, A. Volovich, Superstring interactions in a pp wave background, Phys. Rev. D 66 (2002) 086004,
hep-th/0204146.
[16] M. Spradlin, A. Volovich, Superstring interactions in a pp wave background 2, JHEP 0301 (2003) 036,
hep-th/0206073.
[17] A. Pankiewicz, More comments on superstring interactions in the pp wave background, JHEP 0209 (2002)
056, hep-th/0208209.
[18] A. Pankiewicz, B. Stefanski Jr., PP wave light cone superstring field theory, Nucl. Phys. B 657 (2003) 79,
hep-th/0210246.
[19] A. Pankiewicz, An alternative formulation of light cone string field theory on the plane wave, JHEP 0306
(2003) 047, hep-th/0304232.
[20] A. Pankiewicz, B. Stefanski Jr., On the uniqueness of plane wave string field theory, hep-th/0308062.
[21] J.H. Schwarz, Comments on superstring interactions in a plane wave background, JHEP 0209 (2002) 058,
hep-th/0208179.
[22] Y.H. He, J.H. Schwarz, M. Spradlin, A. Volovich, Explicit formulas for Neumann coefficients in the plane
wave geometry, Phys. Rev. D 67 (2003) 086005, hep-th/0211198;
See also: J. Lucietti, S. Schafer-Nameki, A. Sinha, On the plane-wave cubic vertex, hep-th/0402185.
[23] P. Di Vecchia, J.L. Petersen, M. Petrini, R. Russo, A. Tanzini, The three string vertex and the AdS/CFT
duality in the pp wave limit, hep-th/0304025.
[24] D.J. Gross, A. Mikhailov, R. Roiban, A calculation of the plane wave string Hamiltonian from N = 4 superYangMills theory, JHEP 0305 (2003) 025, hep-th/0208231.
[25] C. Kristjansen, J. Plefka, G.W. Semenoff, M. Staudacher, A new double scaling limit of N = 4 super-Yang
Mills theory and pp wave strings, Nucl. Phys. B 643 (2002) 3, hep-th/0205033.
[26] N.R. Constable, D.Z. Freedman, M. Headrick, S. Minwalla, L. Motl, A. Postnikov, W. Skiba, PP wave string
interactions from perturbative YangMills theory, JHEP 0207 (2002) 017, hep-th/0205089.
[27] N. Beisert, C. Kristjansen, J. Plefka, G.W. Semenoff, M. Staudacher, BMN correlators and operator mixing
in N = 4 super-YangMills theory, Nucl. Phys. B 650 (2003) 125, hep-th/0208178.
[28] N.R. Constable, D.Z. Freedman, M. Headrick, S. Minwalla, Operator mixing and the BMN correspondence,
JHEP 0210 (2002) 068, hep-th/0209002.
[29] J. Gomis, S. Moriyama, J. Park, SYM description of SFT Hamiltonian in a pp wave background, Nucl. Phys.
B 659 (2003) 179, hep-th/0210153.
[30] J. Gomis, S. Moriyama, J. Park, SYM description of pp wave string interactions: singlet sector and arbitrary
impurities, Nucl. Phys. B 665 (2003) 49, hep-th/0301250.
[31] C.S. Chu, V.V. Khoze, M. Petrini, R. Russo, A. Tanzini, A note on string interaction on the PP wave background, Class. Quantum Grav. 21 (2004) 1999, hep-th/0208148.
[32] C.S. Chu, M. Petrini, R. Russo, A. Tanzini, String interactions and discrete symmetries of the PP wave
background, Class. Quantum Grav. 20 (2003) S457, hep-th/0211188.
[33] C.S. Chu, V.V. Khoze, Correspondence between the three point BMN correlators and the three string vertex
on the pp wave, JHEP 0304 (2003) 014, hep-th/0301036.
[34] G. Georgiou, V.V. Khoze, BMN operators with three scalar impurities and the vertex correlator duality in pp
wave, JHEP 0304 (2003) 015, hep-th/0302064.
[35] G. Georgiou, V.V. Khoze, G. Travaglini, New tests of the pp wave correspondence, JHEP 0310 (2003) 049,
hep-th/0306234.
[36] G. Georgiou, G. Travaglini, Fermion BMN operators, the dilatation operator of N = 4 SYM, and pp wave
string interactions, JHEP 0404 (2004) 001, hep-th/0403188.
[37] R. de Mello Koch, A. Donos, A. Jevicki, J.P. Rodrigues, Derivation of string field theory from the large N
BMN limit, Phys. Rev. D 68 (2003) 065012, hep-th/0305042.
Impurity non-preserving 3-point correlators

of BMN operators from PP-wave holography I:
bosonic excitations
Suguru Dobashi, Tamiaki Yoneya
Institute of Physics, University of Tokyo, Komaba, Meguro-ku, Tokyo 153-8902, Japan
Received 10 September 2004; received in revised form 11 November 2004; accepted 14 December 2004
Available online 27 December 2004
Abstract
As a continuation of our previous works studying the holographic principle in the plane-wave
limit, we discuss the 3-point correlation functions of BMN operators with bosonic excitations when
impurities are not conserved. We show that our proposal for a holographic mapping between the
conformal OPE coefficients of super-YangMills theory and the 3-point vertex of the holographic
string field theory is valid to the leading order in the large limit. Our results provide for the first
time a direct holographic relation for the 3-point correlators of BMN operators including impurity
non-preserving processes.
PACS: 11.25.-w; 04.60.-m
1. Introduction
What is the correct interpretation of holographic principle in the PP-wave limit of the
AdS5 /SYM 4 correspondence has been one among several open problems since the first
original discussion [1] of the BMN limit. In the previous paper [2], we proposed a simple
direct relation between the 3-point OPE coefficients of conformal BMN operators and
the bulk 3-point interaction vertex of string field theory, by developing the basic ideas
E-mail addresses: doba@hep1.c.u-tokyo.ac.jp (S. Dobashi), tam@hep1.c.u-tokyo.ac.jp (T. Yoneya).
doi:10.1016/j.nuclphysb.2004.12.013
55
presented first in Ref. [3]. In [2], we constructed the holographic string field theory which
meets requirements for the validity of the GKPWitten relation and confirmed explicitly
that it indeed reproduces the 3-point OPE coefficients of N = 4 super-YangMills theory
in the leading large- expansion for stringy BMN operators in the impurity-preserving
sector. Furthermore, it was also clarified that two entirely different proposals for relating
3-point vertices of (different versions of) string field theories to corresponding quantities
on the gauge-theory side in the impurity-preserving sector are actually compatible to each
other through our holographic map.
It is important to recall, as we have stressed there, that our holographic relation in principle should be valid for much more general impurity non-preserving processes too, since by
construction the holographic relation is satisfied for the so-called non-extremal 3-point
functions of chiral primary operators in the BMN limit. The latter functions are regarded to
be the supergravity sector of general impurity non-preserving 3-point functions. The goal
of the present paper is to confirm that our stringy holographic relation is indeed satisfied
for impurity non-preserving cases in the leading large- expansion, which is at the same
level as in the existing treatments of the 3-point functions of impurity-preserving sector on
both sides.
The significance of this result seems evident, since the usual discussions in most of
other works being focused almost exclusively on the dilatation operator are practically restricted to impurity-preserving sector and therefore are difficult to extend them to impurity
non-preserving processes.1 In contrast to this, the string-field Hamiltonian in our approach
cannot exactly be interpreted as the dilatation operator, since it generates an infinitesimal
translation along a geodesic connecting from boundary to boundary in the Euclidean AdS
spacetime. Except for regions approaching asymptotically to the conformal boundary,
such a translation cannot be identified with dilatation. In fact, this discrepancy between
bulk Hamiltonian and dilatation operator on the boundary is, in our opinion, even more so
in the usual derivation of the Penrose limit, leading to a geodesic which is disconnected
from the boundary: there would be no asymptotic region where the Hamiltonian is related
to the dilatation, corresponding to the fact that the translation with respect to the global
time coordinate of the AdS spacetime does not coincide with dilatation on the boundary
at least within the usual geometric interpretation of the AdS spacetimes. This puzzle was
our main motivation for undertaking the present series of works [35], to which we would
like to invite the reader for further discussions of this basic issue. From our viewpoint,
the usual Penrose limit is obtained formally by a Wick rotation of the affine time parameter along the geodesics connecting boundary to boundary. Note that the relative topology
of trajectory with respect to the conformal boundary is completely changed by this Wick
rotation. This however explains why the dilatation comes about even in Minkowski formulation. In any case, given the apparent correspondence for conformal dimensions {r }
on one hand, it would be very strange if there could be no clear way of mapping the OPE
coefficients which, together with conformal dimensions, constitute the crucial data of CFT.
1 On the gauge-theory side, the matrix elements of a dilatation operator in the sense of renormalized pertur-
bation theory would vanish unless the states have degenerate conformal dimensions in the lowest order. This is
borne out in computations in Refs. [911].
56
We note that, even though the matrix elements of 3-point interaction vertices of string
field theory in the impurity non-preserving sector often have additional powers in 1/,2
the OPE coefficients are not necessarily of lower order in the large- expansion for large
but fixed R-charge angular momentum J . It is also well known that a class of impurity
non-preserving processes cannot simply be ignored in summing up intermediate states in
higher-loop calculations, at least within the logic of string field theory as presently understood. Once the summation over intermediate states are involved, the limiting procedures
required in the studies of amplitudes for BMN states should be a very subtle problem, since
results would depend on the orders of various infinite sums and the limiting procedures.
We hope that our concrete results for impurity non-preserving 3-point correlators provide
a useful first step for further exploration of the AdS holography in the plane-wave limit
beyond the approximation of keeping only the impurity preserving sector.
In the next section, we start from summarizing briefly the holographic relation and the
holographic string field theory proposed in the previous work. In Section 3, we treat a
general case of impurity non-preserving 3-point functions where only non-singlet (more
precisely, traceless) scalar (and bosonic) impurities corresponding to the fluctuations along
4-directions among the S 5 of AdS5 S 5 are involved. In Section 4, we extend the results to
include non-singlet vector excitations. In Section 5, we discuss a few typical cases where
singlet impurities are involved. The concluding Section 6 contains some remarks on relevant issues, especially, the uniqueness of our holographic relation and the question of
higher-order effects. Some useful formulas which are relevant for discussions in the main
text are summarized in Appendix A.
2. The holographic relation and string field theory

The key concept behind our holographic relation is that the correspondence between
conformal boundary and bulk spacetime in the plane-wave limit of the AdS/CFT correspondence should be based on a tunneling geodesics traversing inside the AdS spacetime
with the Euclidean signature which start from the conformal boundary and end again at
conformal boundary, instead of the usual procedure of taking the Penrose limit in the AdS
spacetime with Minkowski signature. Such amplitudes can be regarded as an Euclidean
analog of the usual S-matrix. This picture, which has been previously proposed by us for
the purpose of resolving some puzzles associated with holography in the PP-wave limit in
[3], automatically emerges by studying the large-J limit of the GKPWitten relation for
single-trace chiral primary operators. Since the 3-point functions of chiral primary operators are protected against the perturbative corrections on the gauge-theory side, we can
derive the effective action describing the interaction of the BMN operators along the tunneling geodesics in the supergravity sector. The full string field theory, the holographic
string field theory, should reduce to this effective action in the supergravity sector.
2 This is due to an additional power of 1/ in the asymptotic forms of Neumann functions other than N 21
mn
31 . See Appendix A.
and N mn
57
Under this requirement, we are led to a unique 3-point vertex which should obey the
holographic relation based on the above tunneling picture. We stress that in the GKP
Witten conjecture the bulk partition function as a functional of boundary fields is fixed
uniquely with a given basis of linearized supergravity fields, independently of field redefinitions. For 3-point correlators, the effect of such a field redefinition would be proportional
to the equation of motion, and integration over the interaction point in the bulk would give
vanishing result. In this sense, even though energies being now the conformal dimensions
are not conserved, the Euclidean S-matrix is very similar [3] to the ordinary S-matrix of
flat Minkowski spacetime. We will give a remark on the issue of uniqueness of our prescription in the final section of the present paper. A subtlety in the case of the extremal case
where the conformal dimensions are strictly conserved will also be clarified.
The holographic relation is summarized as
123
,
(2 + 3 1 )

J J (2 +3 1 )/2
2 + 3 1
123 = f 2 3

+ 1 123
J1
2
C123 =
with
(2.1)
(2.2)
J1 J2 J3
(2.3)
|H3 h .
N
The conformal dimensions r are those in the planar limit. The CFT coefficients C123
is defined under the standard normalization of two point functions as
123 =
(1)1|(2)2|(3)3|

x1 )O2 (
x2 ) =
O 1 (
12
|
x12 |21
(2.4)
by

O 1 (
x1 )O2 (
x2 )O3 (
x3 ) =
C123
2
3
|
x12 | |
x23 |21 |
x31 |22
(2.5)
with (
x12 = x1 x2 , etc.)
2 + 3 1
, etc.
(2.6)
2
The symbol |H3 h is the 3-point interaction vertex of the holographic string field theory

J J J
1
(2)|(3)| 1 2 3 |H3 h + h.c.
S3 =
(2.7)
d (1)|
2
N
1 =
Here the integration over the R-charge angular momenta Jr associated with the string fields
should be implicitly understood, under the conservation condition
(r = 2, 3), (1) |
J1 = J2 + J3 . The quantity f in (2.2) depending on is responsible for the non-locality
caused by the extended nature of strings and is given by
(r) |
f = 1 4(1) (2) (3) K,
(2.8)
58
where K is the well-known expression in string field theory which is defined in terms of
various infinite matrices associated with the familiar overlap condition for string interaction. For full details of the string field theory, we refer the reader to Appendix B of the
previous paper [2] and to the references cited therein and below. The 3-point interaction
vertex takes just an equal-weight sum of the two previously constructed vertices which are
compatible with supersymmetry algebra:

1
|H3 h |H3 SV + |H3 D
(2.9)
2
where |H3 SV and |H3 D are those proposed in [6] and in [7], respectively. In particular,
the prefactor of the part |H3 D manifestly takes the form of the energy difference 21
while that of the part |H3 SV is the one obtained by a natural generalization of the familiar
flat-space vertex.
Our convention for various parameters is as follows: R 4 /J12 ( )2 = = 1/(p1+ )2 =
2
gYM N/J12 , g2 = J12 /N , (r) = pr+ , |(r) | = Jr /R 2 (r = 1, 2, 3). Do not confuse the
symbol (1) , etc. with the previously defined 1 , etc. as (2.6). Using these conventions, the
large limit of f is given as
f
J1
J2 J3
.
J1
4|(1) |
(2.10)
We assume (1) (= (2) (3) ) < 0, (2) > 0, (3) > 0. Note that the mass parameter
becomes a meaningful curvature parameter = 1/R as we identify the light-like momentum with the angular momentum by |pr+ | = Jr /R, which is the correct one for comparing
the action with flat-space form. The interaction term is of order gs forfixed R and p + as it
should be. The overall factor J1 J2 J3 /N is rewritten as 4gs ( )2 |p1+ p2+ p3+ |/R 5/2 in
terms of the usual string coupling constant gs . This form indicates that the effective action
in fact has a non-trivial curvature dependence.
The above normalization of the interaction vertex is fixed by matching the effective
action in the supergravity sector. The string field is normalized such that the free action
takes the standard Euclidean form

1
1
| |
| + |H
2 | ,
|
S2 = d
(2.11)
2
2

1
(r)
H2 =
n(r) an(r) an(r) + bn(r) bn(r) with n(r) = n2 + ((r) )2 ,
|(r) | n=
(2.12)
(r)
(r)
(r)
(r)
where (an , an )s and (bn , bn )s represent 8-component bosonic and fermionic oscillators of rth string.
One of the most characteristic features of this string field theory is that for purely
bosonic string states the (total) prefactor reduces to the special form which consists of
(r)
(r)
only cos modes an (n 0) [8] and of only sin modes an (n > 0) for scalar (i = 58)
and vector (i = 14) impurities, respectively:
8

3
4
(r)
(r)

m

m (r)i (r)i
(r)i (r)i
a
am +
a
a
|H3 h
(2.13)
|Ea ,
(r) m
(r) m m
r=1
i=5 m=0
i=1 m=1
where |Ea given as

3
1
|Ea = exp
2
59

(r) rs (s)
m
Nmn n
|0(1)(2)(3)
(2.14)
r,s=1 m,n=
is the standard bosonic overlap vertex.3 Thus, Z2 (or higher SO(8) symmetry) of the free
(r) (r)
string-field theory is completely violated by the interaction. The oscillators (n , n ,
n Z) in the exponential basis which directly corresponds to the standard convention for
the BMN operators are related to the trigonometric basis by (n > 0)
0 = a0 ,
1
n = (an ian ),
2
1
n = (an + ian ).
2
(2.15)
In our formalism, the BMN operators must have definite conformal dimensions to at
least the first order in g2 and to all orders with respect to . This means that for unprotected stringy BMN operators we have to take into account the effect of various operator
mixing including double-trace operators [9,10] to the leading approximation in 1/N expansion in extracting the CFT coefficients C123 on the gauge-theory side. In real life, we
have to be satisfied by studies of leading 1/-expansions at current stage of development,
since computation of 3-point OPE coefficients at the order g2 and beyond has not been
carried out so far. Such a computation on the gauge-theory side in general requires twoloop calculations including operator mixing effects.
In [2], we have presented a general argument that the above relation must be correct for
arbitrary impurity-preserving 3-point functions, by reinterpreting appropriately the previously known results [7,12] for comparison between string field theory vertices and gaugetheory calculations in the leading large- expansion. We explicitly confirmed the above
relation for two-impurity processes including vector and spinor excitations. The peculiar correction factor (f JJ2 J1 3 )(2 +3 1 )/2 ( 2 +23 1 + 1) appearing in the expression
(2.3) can be neglected to the leading order in the 1/-expansion for impurity-preserving
processes. Thus, in these cases, the relation reduces to the one first conjectured in Ref. [11]
from a different viewpoint. Our previous work, however, clarified that this particular relation is valid only with our holographic string field theory vertex |H3 h which takes into
account the operator mixing of gauge-theory operators by the above specific combination
of two different prefactors.
When the impurities are not conserved, the correction factor plays a crucial role, as
has already been shown by the construction in [2] for non-extremal correlators of chiral
primary operators on the basis of the GKPWitten relation. Our task is now to confirm
this for unprotected stringy BMN operators by studying the large- limit. We can divide
the impurity non-preserving 3-point interactions into two classes, class I and II, respeccl
cl
cl
tively, depending on 1cl = (cl
2 + 3 1 )/2 > 0 or < 0 where r denotes the classical
conformal dimension, counting the number of fields and (spacetime) derivatives involved
3 As in [2], the minus sign on the exponential factor is due to our phase convention in defining the world-sheet
oscillators. This is different from the standard one in the literature, but is necessary for matching between bulk
and boundary.
60
in each BMN operator. Because of the SO(4) SO(4) symmetry, class II processes are
possible only when we allow singlet representations for the external line 1.
3. Class I non-singlet scalar impurities

Let us start from considering a simple example of class I processes. We denote the directions of scalar excitations by i, j, . . . (58). The operator 2 is assumed to involve 4 scalar
excitations in all different directions i, j, k, with world-sheet momenta m, m, p, p,
respectively. The operators 3 and 1 are assumed to involve 2 scalar impurities in directions k, with momenta q, q and directions i, j with momenta n, n, respectively. The
explicit forms are given, suppressing possible mixing terms with double-trace operators,
as

1
(2)
Tr i Z a j Z b k Z c Z d
O(i,m;j,m;k,p;,p) =
(J2 + 3)3 N J2 +4 a+b+c+d=J2
2i

[am+(a+b+1)p(a+b+c+2)p]
e (J2 +3)
+ permutations , (3.1)
where the permutations indicates the summation over all non-equivalent positioning of
the impurities (a, b, c, d 0), and
(3)
O(k,q;,q) =
1
(J3
+ 1)N J3 +2
J3

2i aq
Tr k Z a Z J3 a e (J3 +1) ,
(3.2)
a=0
J1

1
2i an
(1)
O(i,n;j,n)
=
Tr i Z a j Z J1 a e (J1 +1) .
(J1 + 1)N J1 +2 a=0
(3.3)
The overall constants correspond to the normalization of scalar fields such that the
1 )Z(x2 ) =
free propagators are equal to (no summation over i) i (x1 )i (x2 ) = Z(x
2
1/|x12 | . Note that the phases associated with non-zero momenta are determined only by
relative (and oriented) distances among impurities. Here we have adopted the normalization constants and phase factors which are slightly different from those adopted in the
recent literature. In the large J limit, the difference between J and J + k 1 with k being
the number of impurities is inconsequential for our leading order computations restricted
to bosonic excitations. Therefore, in the following we will ignore this difference unless
otherwise stated explicitly and use the more familiar convention by ignoring these shift of
J , in order to save the space for mathematical expressions. However, for fermionic excitations it turns out that these shifts in the phases actually play an important role. Fermionic
excitations will not, however, be treated in the present paper and be left for a separate work.
(1)
(2)
(3)
It is obvious that the gauge-theory correlator O (i,n;j,n) O(i,m;j,m;k,p;,p) O(k,q;,q)
in the leading planar approximation must necessarily involve contractions of impurity
fields k and between 2 and 3, aside those between 1 2 and 1 3. This implies that
to the lowest non-trival order in g2 we can ignore mixing with double-trace operators in
calculating the 3-point function, since the effect of the mixing terms becomes of higher
order with respect to N1 , in contrast to the impurity preserving case where there is no
61
contraction between 2 and 3 for the mixing contribution which corresponds to the topology of a product of two cylinders (1/N ). We then find by a straightforward free-field
computation that the CFT coefficient is given by
2
1
2
sin2 (yn)
J1 J2 J3
1
2 sin (yn)
C123 =
2J1 2
=
2
2
N J J 3J
N
(n m
J12 y 2 (1 y) 2 (n m
y)
y)
1 2 3
(3.4)
with y = J2 /J1 . The first factor in the first equality comes from the overall normalization,
and the second factor is the result of free-field contractions and of summation over the
positioning of impurities. In particular, the factor 2 originates from the permutation of two
contractions between operators 2 and 3. Since the contractions between 2 and 3 can occur
in the planar limit only when the impurity fields k and are adjacent to each other, they
do not have any momentum dependence in the present large-J limit. This is in general
true for arbitrary configurations of momenta of such impurities at least for exchanges of
bosonic fields, beyond the above special case.
Now let us turn to the corresponding calculation in string field theory. The string states
are
i(1) j (1)
(1) 0|n n ,
i(2) j (2) k(2) (2)

(2) 0|m m p p ,
k(3) (3)
(3) 0|q q ,
(3.5)
respectively. The relevant part of the interaction vertex takes the form

|H3 h P123 exp N 12 N 23 |0(1),(2),(3)
where the prefactor is
(2)
P123 =
(2)

p
m
(2) (2)
(2)
(2)
(2) (2)
2 + m
+
2 + p(2) p + p p(2)
m + m m
(2)
(2)
(3)
(1)

q
n
(3)
(3)
(1)
(1)
2 + n(1) n + n n(1)
2 + q(3) q + q q(3)
(3)
|(1) |
(3.6)
and
N rs
are expressed in terms of the Neumann functions in the exponential basis as

(1) (2)
(1) (2)
(1) (2)
(2)
12
12
N 12 = N nm
n m + n m + N nm
n m + n(1) m

(1) (2)
(1) (2)
(2)
12 (1) (2)
12
(3.7)
n p + n
n p + n(1) p
,
p + N np
+ N np

(2) (3)
(2)
(3)
23 (2) (3)
23
p q + p q + N pq
p q(3) + p(2) q .
N 23 = N pq
(3.8)
Note that there is no contraction between 1 3.

(r)
In the leading large limit we can simply replace the energy factor n /|(r) | by for
class I processes, since there is no singularity in the multiplying factor (denoted by G) of
the holographic relation

cl

+
1
1cl !
J1
J2 J3 (2 +3 1 )/2 ( 2 23 1 + 1)
,
G f
J1
(2 + 3 1 ) 21cl 4|(1) |
(3.9)
62
with 1 = (2 + 3 1 )/2 1cl + O(1/2 ). In the present example, 1cl = 2. Usrs

rs
rs
rs = N
nm
ing the properties of Neumann functions such as N nm
, N nm
= N nm
for
m, n > 0, we can easily see that the contributions from two parts of the prefactor involv(2) (2)
(2) (2)
(1) (1)
(1) (1)
ing m m + m m and n n + n n , respectively, cancel against each other.
This cancellation corresponds to the fact that two terms |H3 SV and |H3 D have equal contributions in this case. The situation is in contrast to the impurity-preserving sector where
they play different roles (roughly speaking, bare interaction and mixing effect, respectively), owing to the existence of the singularity in the factor G. Furthermore, using the
23 | is independent of the momenta, we find that the
property that the large limit of |N pq
matrix element of the interaction vertex is equal to
12 23 2
4 N nm
N pq .
Using the explicit expressions of Neumann functions (see Appendix A for a summary) in
the leading large limit,
(1)m+n+1 sin(ny)
12
,
=
N nm
y(n m
y)
23
N pq
=
(1)p+1
,
4|(1) | y(1 y)
(3.10)
we find that the CFT coefficient (3.4) from gauge-theory side precisely matches the string
interaction vertex with the holographic relation (2.1)(2.3).
It is not difficult to extend the above result to a more general case of scalar impurities.
In this section, we restrict ourselves to operators without any singlet representation with
respect to O(4) group of rotations of scalar directions. More precisely, the scalar impurities
i1 , i2 , . . . are contracted to polarization tensors CiI1 i2 ... which are traceless with respect
to arbitrary pair of the O(4) indices (i1 , i2 , . . .) for each conformal BMN operator in an
appropriate irreducible representation of SO(4):
CiI1 i2
I
=
O(p
1 ,p2 ,...)
J k1 N J +k

Tr i1 Z a1 i2 Z a2 e2i(a1 p2 +a2 p3 +)/J
a1 +a2 +=J

+ permutations ,
(3.11)
where k ( 1) is the number of impurities and as above we have to sum over non-equivalent
positioning of impurities.4 The traceless condition allows us to ignore the mixing of the
pairs of operators Z and Z which would be needed [13] for defining operators with definite conformal dimensions even if the mixing with double-trace operators can be ignored. If
we consider higher orders in gYM , mixing between purely bosonic operators with antisymmetrized scalar indices (or mixed scalar and vector indices) and those involving fermion
impurities are expected to play important roles. In this paper, we ignore such complications
by restricting ourselves to the leading order effect in the large- expansion on 3-point functions. The symbol I of the polarization tensor indicates collectively the configurations of
4 Note that we have changed the notations slightly from Ref. [2]. For instance, k k. The convention on
the summation over permutations (for sugra modes in particular) is also changed. In the present paper, nonequivalent permutations exclude those which correspond to cyclic permutations, while in [2] they were not
excluded and hence the normalization constants had an additional power of 1/J .
63
both momenta and SO(4) representation. Its normalization is most conveniently expressed
(1) (1)
using oscillators ap , ap , etc. of strings, as

I I
(1)
(2)
1
C 2 exp

(3.12)
|0(1)(2) = I1 I2 ,
(1) C
p
(2)

C I1 =

(1)
p=
I1
i1 (1) i2 (1)
(1)0|Ci1 i2 ... p1 p2
etc.,
(3.13)
where the SO(4) indices on the exponential is suppressed. According to this convention,
various symmetry factors associated with the symmetry property of the SO(4) indices are
absorbed in the normalization of polarization tensors themselves. This representation naturally takes into account the summation over all allowed contractions with equal weights
corresponding to those of free-gauge theory. Of course, the momenta must satisfy the level
matching condition
k

pk = 0.
(3.14)
i=1
Suppose the rth operator has kr impurities. Then, the number of contractions between
r = 2 and r = 3 is equal to
1cl = (k2 + k3 k1 )/2.
Now, in computing the gauge-theory correlators for this general type of operators, we note
the following properties, all of which already appeared in the above example.
(1) We can ignore operator mixing with double-trace operators, since the mixing contributions are always of higher order in the 1/N -expansion because of the existence of
non-zero number of 23 contractions. In the absence of the mixing with double-trace operators, the CFT coefficients to the leading order in 1/ can be determined entirely from the
free-field contractions which give the correct leading order form of the spacetime factor
of conformal 3-point functions.
(2) In the planar limit, the contractions between 2 and 3 must belong to a single group
of adjacent products of the impurity i fields in which no Z field is contained, since all Z
fields in the operators 2 and 3 must be contracted with Z fields in the operator 1 and hence
would lead to a non-planar contribution otherwise. In other words, in the sum over ai in the
definition (3.11), such contractions are possible only when ais = ais+1 = = ais+1 1 = 0
with some s for a consecutive set of impurities (is , is+1 , . . . , is+1 1 ). See Fig. 1.
(3) By the cyclic symmetry of Tr operation and the level-matching condition, we can
always choose s = 1. This means that we can replace the momentum-dependent exponential by one for these impurities in the large-J limit. Therefore, the 2 3 contractions do
not give any momentum dependent factor.
Using these properties, the 3-point correlation function is computed as follows. First, the
overall constant factor is

cl
J1 J2 J3 cl (k1 +k2 +k3 )/2 J2 k2 /2 J3 k3 /2
N J1 +k1 +1 1
1cl !
=
1 !J1
N
J1
J1
ki 1 Ji +ki
3
N
i=1 Ji
64
Fig. 1. The dotted lines represent ZZ contractions, while real lines represent contractions of scalar fields i s.
The outer circle denotes the trace of operator product in the line 1 and two inner circles the traces of operator
products in the line 2 and 3, respectively.
which can also be written as
k2 /2 k3 /2

cl
J2
J3
J1 J2 J3
cl
1
J1k1
1cl !J1 1 y(1 y)
N
J1
J1
(3.15)
where k2 = k2 1cl and k3 = k3 1cl represent the number of impurities in the operators
2 and 3, respectively, which contract with those of the operator 1. Note that the factor 1cl !
comes from the permutations of the ordering of impurities contracting between 2 and 3.
(1)
(2)
Secondly, each single contraction with momenta pi and pi respectively between
1 and 2 after the summation over the permutation of their positioning gives the factor
(suppressing the spacetime dependent factor 1/|
x12 |2 )

(2)
(1)
(1)
(1)
pi
sin ypi
pi
J1 eiypi
exp 2ia
(2)

p
J2
J1
a=0
pi(1) yi
J2

(3.16)
in the large J limit, which is expressed in terms of Neumann function as

(1)
(2)
(1)
J1 y(1)pi +pi +1 eiypi N 12(1)
(2)
pi pi
(3.17)
Here the Kronecker delta of the corresponding SO(4) indices is suppressed. Similarly, each
(1)
(3)
single contraction with momenta qi and qi respectively between 1 and 3 gives
J3

a=0

(3)
(1)

(1)
q
q
J1 1 yei(1y)qi N 13(1) (3) .
exp 2ia i + i
qi qi
J3
J1
(2)
(3)
(3.18)
On the other hand, each contraction with momenta ri and ri respectively between 2
and 3 gives factor 1 (again apart from the spacetime factor) which is replaced by the
corresponding Neumann functions using the identity

(2)
1 = (1)ri +1 4|(1) | y(1 y)N 23
(3.19)
(2) (3) .
ri ri
65
Collecting all these factors together, we conclude that the CFT coefficient is given as
k2 /2 k3 /2
I I I
J3
k1 J2
1cl +2cl +3cl J1 J2 J3
1
2
3
C123 = C C C (1)
J1
N
J1
J1

cl

1
cl
12
13
1
1 !J1
y(1 y)
J1 y N (1) (2)
J1 1 y N (1) (3)

12

4|(1) | y(1 y)N 23
(2)

cl
cl
cl
= C I1 C I2 C I3 (1)1 +2 +3

qi qi
13
(3)
ri ri
23
pi pi
N 12(1)

(2)
pi pi
12
N 13(1)

cl
1
J1
J1 J2 J3
1 !
cl
N
4|(1) |

23
N
(3)
(2) (3) ,
qi qi
13
ri ri
23
(3.20)

where rs denotes the product over the contractions and the symbol (C I1 C I2 C I3 ) denotes
the part of SO(4) contractions of polarization tensors associated with the free-field contractions. We note that the phase factors of the expressions (3.17)(3.19) cancel using the
level-matching condition and that in terms of string notation, the following equality

I I I
cl
cl
cl
N 12(1) (2)
N 13(1) (3)
N 23
C 1 C 2 C 3 (1)1 +2 +3
(2) (3)
12

= (1) C I1 (2) C I2 (3) C I3 |Ea
pi pi
13
qi qi
23
ri ri
(3.21)
is valid.
Let us next confirm that this result matches the string field theory through our holographic relation. Since we have already expressed a main part of the CFT coefficient
obtained on the gauge-theory side in terms of string states, only non-trivial part for this
task is to examine the prefactor. In the large -limit, it takes the form

P123 = 0(2) 0(2) + 0(3) 0(3) 0(1) 0(1)
+
+

(2) (2)
(3) (3)
(1) (1)
m m + m
m m
m + [m m]
2
m=1
(2) (2)

(3) (3)
(1) (1)
m m + m
m m
m + [m m] .
(3.22)
m=1
Obviously, the contributions from the first and second lines is simply determined by counting the difference of the numbers of impurities in the process 2 + 3 1. Furthermore, in
the third line, the impurity preserving part corresponding to contractions 1 2 and 1 3
rs = N
rs , N
rs
rs = N
23
nm
nm
nm
cancels due to the equalities N nm
. Then, the property that N mn
is independent of the signs of momenta (m, n) shows that the contribution of the third term
is 1cl . Thus all together the total contribution of the prefactor is 21cl . But this is just
canceled by the denominator factor 1/(2 + 3 1 ) of the holographic relation. Using
66
(3.21), it is now clear that the CFT coefficient precisely satisfies our holographic relation
(2.1)(2.3) including the sign and numerical factors.
4. Class I non-singlet vector impurities

In this section, we extend the result of the previous section to the situation where (nonsinglet) vector excitations are involved. A vector excitation of momentum p at the position
a corresponds to the insertion of a derivative e2iap/J Di (i (1 4)) in the trace (3.11).
We assume here that only traceless part are involved with respect to the vector SO(4)
indices too. So, the vector SO(4) indices which will be mostly suppressed below should be
understood to be contracted with traceless polarization tensors similarly as in the previous
section. As we have argued in detail in [2] on the basis of the GKPWitten relation, the
derivatives must then be computed by assuming the following form for the variation of the
distance function |
xrs | under the shift of the spacetime coordinates xr x + xr ,
1
|
xrs
|2
(
xr xs
)2
1
2 xr xs
(4.1)
in order to be consistent with the SO(4) symmetry and orthogonality of vector states. The
invariants such as ( x)2 can be ignored because of the traceless condition.
This allows us to treat the pairs of vector indices almost as those of additional virtual
scalar indices. Namely, the pair of derivatives of the spacetime factor of a free propagator
can contribute in the form
2ia( p(r) q (s) ) (r) (s) 1
2
Jr
Js
e
N rs
phase factor,
i j
2
4 ij p (r) ,q (s)
|
x
|
|
x
|
rs
rs
s
where the phase factor depending on (r, s) is the same as for the case of scalars. Thus a
pair of vector derivatives yields essentially the same factor as in the case of contractions of
scalar impurities, except for the factor 2 which is absorbed in the normalization, provided
that they act on the free propagator. This implies that the derivatives acting on 1 2
and 1 3 contractions give the same factors on both sides of gauge theory and string
field theory. In other words, the holographic relation is satisfied as it stands when vector
impurities are conserved for arbitrary impurity non-preserving processes in which impurity
non-preserving contractions between 2 and 3 only occur for scalar excitations.
Therefore, in the rest of this section, it is sufficient to concentrate to the case where all
vector impurities contribute to (2 3) contractions. Suppose first that we add 1cl such
derivatives for a process with non-zero scalar exchange in the 23 channel (1cl > 0). Since
cl
the spacetime factor 1/|

x23 |21 only comes from a single group of adjacent scalar impurities, the derivatives act as
1
1cl /2
(2J2 )
1cl /2
(2J3 )
1
1 (2) 1cl (3) 1cl
cl
cl
(1 )!
|
x23 |21
where the first factor comes from the normalization constant associated with the vector excitations. For notational simplicity, we suppress the SO(4) vector indices. The denominator
67
factor 1/(1cl )! is the symmetry factor to cancel the over-counting. This is equal to
cl /2 cl /2
1
1
(1cl + 1cl )
J3
1
1cl J2
J1
cl
cl
cl
2(
J1
J1
(1 ) |
x23 | 1 +1 )
cl (1cl + 1cl + 1)
21cl
1
cl
1
= J1 1
y(1 y)
.
cl
cl
cl
cl
cl
2(
(1 + 1) 2(1 + 1 ) |
x23 | 1 +1 )
As in the scalar case, we further multiply the factor

(2)
1 = (1)ri +1 4|(1) | y(1 y)N 23
(2) (3)
(4.2)
ri ri
corresponding to each pair of derivatives

By this procedure, the power of y(1
y) is canceled and replaced by the Neumann functions for (23) channel, leaving us the
(2) (3) .
correction factor (
J1
1cl
.
cl | )
4|(1)
Furthermore, the factor
21cl
(1cl +1cl +1)
(1cl +1) 2(1cl +1cl )
is just the
one required for modifying the energy factor in the holographic relation as
(1cl + 1)
21cl
(1cl + 1cl + 1)
2(1cl + 1cl )
corresponding to the shift of 2 + 3 1 due to the addition of vector impurities in the

(23) channel.
Turning now to the string side, only difference for vector excitations from the purely
scalar case is that the contribution of vector modes to the prefactor consists only of sin
modes,

(2) (2)
(3) (3)
(1) (1)
m m + m
m m
m + [m m]
2
m=1

(2) (2)
(3) (3)
(1) (1)
m m + m
m m
m + [m m] .
2
(4.3)
m=1
Then, the independence of (23) Neumann functions on the momentum (except for the
sign factor) shows that the vector contribution in this channel is zero due to cancellation
between the first and the second line in this expression. Hence, it does not contribute at
all to the modification of the 3-point function. Also it is easy to check that the sign and
phase factors exactly match using level matching condition, in the same way as purely
scalar case of the previous section. Thus we can conclude that the holographic relation is
precisely satisfied.
The case with 1cl = 0 requires a separate consideration. In this case, there is no direct
contraction for scalar fields in the (23) channel. Also the correction factor before taking
derivatives is singular in the large -limit, and we have to take into account mixing with
double-trace operators. The spacetime dependence come from the order contributions
to the anomalous dimensions. Let us proceed inductively, starting from 1cl = 1. We have
to consider
1
1
1
(2) (3)
(J2 J3 )1/2 1
(2J2 )1/2 (2J3 )1/2
|
x23 |21
|
x23 |2
68
with 1 O(((1) )2 ). The factor 1 is precisely the necessary factor which modifies
the singular energy factor into non-singular one corresponding to non-zero 1cl ,
(1cl + 1) (1cl + 1cl + 1) 1
(1 + 1) (1cl + 1)
1
=
= ,
21
21
21
2
2(1cl + 1cl )
with 1cl = 1 and 1cl = 0, and the power (J2 J3 )1/2 plays the same role generating the
(23) Neumann functions as in the above case with non-zero 1cl . Since there does not
occur any correction factor from the vector prefactor on the string side, this proves the
validity of the holographic relation. Note that in this special case the mixing with doubletrace operator for the operator 1 is effective on the gauge-theory side, but on the string
side the (23) Neumann functions contribute the same common factor for both |H3 D
and |H3 SV . Once the spacetime factor 1/|
x23 |2 is generated correspondingly to a single
vector exchange as above, we can apply the same argument inductively at each time of
adding one vector exchange between 2 and 3 as we have discussed for the case starting from
non-zero 1cl . This completes our argument for the holographic relation for non-singlet
vector impurities.
5. Singlet operators and class II processes

All previous arguments are restricted to cases where there is no trace part in the external
lines. When we allow trace part for the SO(4) representation, the situation becomes more
cumbersome and it is not easy to make treatments in a general way because of various
possible mixings among different configurations of operator products even at the level of
single trace operators. It is also important that the inclusion of singlet opens the possibility of class II processes. In the present paper, we study several typical cases in order to
convince ourselves that our holographic relation must be valid in such cases too. In particular, we will see that various characteristic structures of the CFT coefficients depending
differently on scalar and vector excitations are nicely captured by the interplay between the
factor G in the holographic relation and the very specific prefactor of our string-interaction
vertex.
5.1. Class I with scalar singlets
The simplest (normalized) BMN operator with singlet representation for scalar excitation [13] is
J

2i
1
J +1 ,
Oss(p;p) =
(5.1)
Tr
i Z a i Z J a e J ap 4ZZ
2 J N J +2
a=0
where the scalar index i is summed over from i = 5 to 8 and p = 0. When p = 0, the
normalization constant should be replaced by 1 J +2 . The necessity of the mixing of
2 2J N
the second term can be easily understood when p = 0: in order for this operator to be BPS,
it must be symmetric and traceless with respect to SO(6) symmetry. The role of the second
69
Fig. 2. The contraction of Z and Z between 2 and 3, with 2 being the singlet operator and 3 the ground state
operator.
term is to meet this requirement by completing the vanishing SO(6) trace from the ZZ
directions.
Let us start from the case where the operator 1 and 3 are in the ground state,

1
1
O (3) =
Tr Z J1 ,
Tr Z J3 ,
O (1) =
(5.2)
J
J
J3 N 3
J1 N 1
and the operator 2 is the above singlet-state operator,
J

2

2i
1
ap
(2)
J2 +1 .
Os(p;p) =
(5.3)
Tr
i Z a i Z J2 a e J2 4ZZ
2 J2 N J2 +2
a=0
On the gauge-theory side, the first term of (5.3) has no contribution in the leading order in
the perturbation theory. The second term can contribute in the leading order if and only if
(2)
one of the Z in Os(p;p) is contracted with one of Z in O (3) as in Fig. 2. The mixing with
double-trace operator can again be ignored. Thus, apart from the spacetime factor (which
takes the correct form), the CFT coefficient is given as
J1 J2 J3 1
1
1

J3 J1 N J1 = 2
.
C123 = 4
(5.4)
J
J
J
+2
N
J2
1
3
2
J1 N J3 N
2 J2 N
On the string side, the state corresponding to the operator 2 is
(2) 0|
1 (2) (2)

.
2 p,i p,i
22 =
Using the explicit form of the (22) Neumann function N mn
(1)m+n
4|(1) |y ,
we find that the
prefactor is simply
= 2 and hence the matrix element of the interaction vertex is
given as
1
J1 J2 J3 22
J1 J2 J3
Np,p = 4
.
4
(5.5)
N
N
4|(1) |y
By multiplying the factor G

1
J2 J3 1
1 4|(1) |
2 + 3 1
(5.6)
f
+1

,
(2 + 3 1 )
J1
2
2
J1
21cl
70
Fig. 3. The two types of contractions between 2 and 3 which are both singlet operators with two scalar impurities.
we find that the holographic relation is precisely satisfied including numerical coefficient
and sign.
This simplest example shows that the mixing of the second term (5.3) is absolutely
necessary for this agreement, and also that it is responsible to the (22) Neumann functions
on the string side. To convince the universality of this role of the Z Z mixing term, it is
useful to treat the case where both of the operators 2 and 3 are singlet states. So, now O (3)
in (5.2) is replaced by
J

3

2i
1
(3)
a
J3 a J3 aq
J
+1
3
Tr
i Z i Z
e
4ZZ
Os(q;q) =
(5.7)
.
2 J3 N J3 +2
a=0
Then, the CFT coefficient consists of two contributions with different sets of free-field
contractions as in Fig. 3,
iijj
C123 = C123
ZZ
+ C123
(5.8)
corresponding to the contributions from the contractions between scalar excitations and
between Z, Z fields, respectively. By similar calculations as above, we find
J1 J2 J3 4
J1 J2 J3 8
iijj
Z Z
C123 =
(5.9)
,
C123 =
.
N
J2 J3
N
J2 J3
On the string-field side, we find the following matrix element of the interaction vertex,
by using self-explanatory notations,
self
123 = 23
123 + 123
with
(5.10)

2
23 2
J1
J1 J2 J3
J1 J2 J3
1
2 Npq = 4
2
(5.11)
= 4
,
N
N
4|(1) | J2 J3

2
J1
J1 J2 J3
J1 J2 J3
1
self
22
33
4Np,p Nq,q = 4
4
123 = 4
,
N
N
4|(1) | J2 J3
(5.12)
23
123
71
where we have again used the property that the Neumann functions appearing here are
independent of the signs of momentum. Since the factor G is

2 + 3 1
1
J2 J3 2
1 4|(1) | 2

,
f
+1 2
(2 + 3 1 )
J1
2
4
J1
the results on both sides exactly match separately for two contributions between (5.8) and
(5.10). These two exercises seem sufficient to convince the general validity of our holographic relation for class I processes involving singlet representation.
5.2. Class II with scalar singlets
Let us next consider an example of class II processes which are possible with the presence of trace part in the operator 1. Consider the simplest non-trivial case with
J

1

2i
1
(1)
a
J1 a J1 ap
J
+1
1
Os(p;p) =
(5.13)
Tr
i Z i Z
e
4ZZ
,
2 J1 N J1 +2
a=0
O (2) =
1
J 2 N J2

Tr Z J2 ,
O (3) =
1
J 3 N J3

Tr Z J3 .
(5.14)
On the gauge-theory side, there is no possibility of free contraction for either term of the
operator 1. This means that the CFT coefficient vanish in the leading order at least in the
1/-expansion and hence is at most of order g2 1/2 , if it does not vanish.
On the other hand, the corresponding matrix element of the 3-point interaction vertex is
given as, suppressing an obvious SO(4) indices,

J1 J2 J3
(1)
(1)
(1)
0|p(1) p 2 + p(1) p + p p(1)
N
2

(1) 2
1 11 (1) 2
(1)
11
11
exp N pp
p
p
N p,p
+ N p,p
p(1) p |0
2

J1 J2 J3 11
11
11
=
(5.15)
2Np,p + N p,p
.
+ N pp
N
2
The explicit form of the (11)-Neumann function is
sin(my) sin(ny)
11
11
11
11
N nm
= N n,m
= N n,m
= N n,m
= (1)m+n+1
|(1) |
(5.16)
in the leading order in the 1/ expansion. The correction terms for this expression is at
most of order 1/(|(1) |)3 . Therefore, the matrix element vanishes in the leading order
and possible corrections must be at most of order (| |)3 .
(1)
Let us next examine the multiplying factor,
2 +3 1

2
J2 J3
2 + 3 1
1
f
+1 .

G
(2 + 3 1 )
J1
2
72
One of new features of class II processes is that 1cl =

the present example,
cl
cl
cl
2 +3 1
2
is a negative integer. In
1cl = 1
which leads to the following leading behavior
G
1 J2 J3 ((1) )2
J1
((1) )2 J1 |(1) |
1
=
=
.
f
J1
4|(1) | p 2
p2
4p 2
(5.17)
Note that we have here taken into account the singularity of the -function () 1/.
Thus we conclude that the CFT coefficient must be at most of order 1/((1) )2 in
conformity with the result from the gauge-theory side. We have again seen the crucial role
played by the factor G in relating both sides. The precise calculation of the higher order
terms for 3-point correlation functions would require in general to take into account various
mixing terms involving both bosonic and fermionic impurities.
5.3. Vector singlets
The case of vector-singlet states is more subtle. The general prescription adopted in
Section 4 for dealing with traceless vector excitations is not applicable to this case, since
in the presence of trace part there are a variety of different ways of constructing SO(4)
invariants after taking spacetime derivatives comparing the case for purely non-singlet
processes. We consider only some simple examples.
The simplest singlet operator with vector impurities with non-zero momenta p and p
is
J
2

2i
1
Ovs(p;p) =
Tr (Dj Z)Z a (Dj Z)Z J 2a e J 1 ap
J
4 J N a=0
(5.18)
where the vector SO(4) index j is summed over from 1 to 4. The derivatives should now
be computed in the standard way. It is easy to check that this satisfies the orthonormality.
Note that we can ignore the second derivatives D 2 Z in the present approximation due to
the equation of motion.
Let us consider the simplest example of class I, in which the operator 2 is this singlet
state, while 1 and 3 are in the ground state,
J
2 2

1
2i ap
(2)
Tr (Dj Z)Z a (Dj Z)Z J2 2a e J2 1 .
Ovs(p;p) =
4 J2 N2J a=0
(5.19)
Then, the free-field contractions between 1 and 2 yields the sum

J
2 2
J2i
1 ap
2
= 0 (p = 0).
a=0
We conclude that the corresponding CFT coefficient must be zero in the leading order. On
the side of string-field theory, the matrix element of the 3-point vertex is
73
Fig. 4. The three types of contractions for the class II process with singlet vector operator for the line 1.

J1 J2 J3
(2)
(2)
(2)
0|p(2) p 2 p(2) p p p(2)
N 2
(2) 2
1 22 (2) 2 1 22
22
(2) (2)
Np,p p Np,p p p |0

exp Npp p
2
2
(5.20)
which is equal to

J1 J2 J3
22
22
22
(5.21)
2 2N p,p
+ N pp
+ N p,p
=0
N
by using explicit forms of the Neumann functions which has already been used for the
case of scalar singlet. Note that the crucial relative sign in the prefactor in the first line in
the expression (5.20), which is opposite to the corresponding scalar case. Since there is no
singularity in the factor G in the holographic relation, this shows that the both sides match.
It is easy to check that the vanishing of the CFT coefficient continues to be valid when we
increase the number of singlet vector impurities in the lines 2 and 3 on both sides.
The situation drastically changes for the class II case. Consider the simplest such
process where the operator 1 is the vector singlet state with 2 and 3 being in the ground
state:
J
1 2
2i

1
(1)
J1 2a e J1 1 ap .
Tr (Dj Z)Z a (Dj Z)Z
Ovs(p;p) =
4 J1 N J1 a=0
(5.22)
On the gauge-theory side, the 3-point function consists of three different types of contributions
(1)

1(2,3)
12
13
O vs(p;p) (
(5.23)
x1 )O (2) (
x2 )O (3) (
x3 ) = F123
+ F123
+ F123
.
In the first and second terms, both of two derivatives in (5.22) acts only on (1 2)
contractions and on (1 3) contractions, respectively. In the third contribution, one of the
derivatives acts on (1 2) contractions and the other on (1 3) contractions. See Fig. 4.
We find
12
F123
=
sin2 yp
1
2
J2 J3 J12
,
2
2J
+2
2
(p) |
x12 |
|
x13 |2J3
N J1 J2 J3
(5.24)
13
F123
=
sin2 (1 y)p
1
2
J2 J3 J12
,
2
2J
(p)
|
x12 | 2 |
x13 |2J3 +2
N J1 J2 J3
(5.25)
74
4
J2 J3
N J1 J2 J3
sin yp sin (1 y)p
(
x12 x13 )
J12 (1)p
.
(5.26)
2
2J
(p)
|
x12 | 2 +2 |
x13 |2J3 +2
Summing up these three contributions, we have the expression
J1 J2 J3 sin2 yp
|
x23 |2
J1
2
.
2
2J
+2
N
(p) |
x12 | 2 |
x13 |2J3 +2
The spacetime factor is precisely the form we should expect for a type II process with
1cl = 1, and hence the CFT coefficient is
J1 J2 J3 sin2 yp
C123 = 2
(5.27)
J1
.
N
(p)2
Let us turn to the string side. The matrix element of the interaction vertex is given by

J1 J2 J3
(1)
(1)
(1)
0|p(1) p 2 p(1) p p p(1)

N
2
(1) 2
1 11 (1) 2
(1)
11
11
exp N pp
+ N p,p
p(1) p |0
p
p
N p,p
2

J1 J2 J3
J1 J2 J3 8
11
11
11
2Np,p + Np,p + Npp =

sin2 (p).
= 4
N
2
N
|(1) |
(5.28)
The difference from the scalar type II case treated in the previous subsection is of course
the relative sign in the prefactor apart from the trivial difference of the suppressed internal
indices. By multiplying the factor G which is the same as (5.17), we confirm that the CFT
coefficient (5.27) is precisely reproduced.
In the present paper, we have restricted our discussions only to bosonic excitations. If
we consider fermionic excitations, the correlation functions generically start from higher
orders in 1/, since impurity non-preserving exchanges of BMN fermion fields with positive U (1)-charge 1/2 between 2 and 3 in general require Yukawa-type interactions with
scalar fields because of charge conservation. Furthermore, we have to take into account
complex mixings among operators which have different numbers of bosonic and fermionic
excitations but with degenerate conformal dimensions. For a precise treatment of these phenomena, it is very important to formulate the correspondence of supersymmetries between
bulk and boundary. Supersymmetry would put various constraints on the correlation functions among both bosonic and fermionic impurities. For examples of such constraints in
the context of the pp-wave limit, see, e.g., [14,15] and references therein. We are planning
to discuss these aspects separately in the next work.
1(2,3)
F123
6. Remarks
In this final section, we give a few remarks on some relevant issues related to the main
problem of the present work. We hope that this is useful for having further perspective on
our approach.
75
6.1. Uniqueness of the holographic relation and the integrability of string field theory
The holographic relation summarized in Section 2 was derived on the basis of the GKP
Witten relation and its perturbative computation in the case of 3-point function using bulkto-boundary propagator. The original effective action for supergravity modes on which
our derivation is based is written in terms of a particular choice of fields [16] by which
derivatives with respect to the AdS5 spacetime coordinates are completely eliminated.
This is justified since the 3-point function is independent of field redefinition. To the present
order of approximation, it is sufficient to show that a would-be interaction term which could
be generated from the quadratic term by a field redefinition does not contribute to the 3point function. For notational simplicity, let us take the case of a single scalar field with
action

1
d 5 x g ()2 + m2 2 .
2
By a field redefinition + c 2 , a 3-point interaction vertex
v3 =
gc 2 (K),
1
K = i gi + m2
g
is generated. However, since the bulk-boundary propagator (z, x; y) satisfies the equation
of motion K(z, x; y) = 0 for arbitrary bulk spacetime positions (z, x), we conclude that
this vertex v3 does not contribute to the correlation function, according to the GKPWitten
prescription. Therefore, the holographic relation before the large-J limit is unique.
Actually, in the special case of the so-called extremal correlators with 2 + 3 1 =
0, there is a well-known subtlety. For generic configurations of U (1) R-charge, this is
possible only for protected supergravity modes. The on-shell matrix elements of 3-point
interaction vertex in the bulk vanish for this case, but the 3-point correlation functions do
not. In our formalism, the zero of the interaction vertex is canceled by the zero of the
denominator in relating the interaction vertex to the CFT coefficient. In this sense, the
extremal correlator has the 0/0 ambiguity. However, we can circumvent this problem in
the GKPWitten relation and in our holographic relation if we slightly shift the conformal
dimensions i i + i by analytic continuation using the generic matrix elements of
the interaction vertex such that the degeneracy 2 + 3 = 1 is lifted and, take the limit
i 0 finally after we compute the CFT coefficients.
By taking the large-J limit, we arrived at the first order action described in Section 2
which describes the dynamics along a single tunneling trajectory. A question now arises as
to the possibility of similar field redefinitions after the large-J limit. Here we have to be
very careful: In computing 3-point correlation functions using the tunneling trajectory, we
had to introduce a cutoff near the boundary, since the approximation of a single trajectory
is violated near the boundary even if we take a short distance limit by which two among
three boundary points approach to each other up to a short distance of order . Finding the
precise prescription for this cutoff was the essence of Section 3 of our previous work [2].5
5 To avoid possible confusions, we emphasize that taking the short-distance limit 0 here is different from
considering double-trace operators. In our previous work [2], this limit was used for the purpose of extracting the
76
Namely, the integral with respect to the positions of interacting point along the tunneling
trajectory in the naive perturbation theory
+
d e(1 2 3 )
must be replaced by

+
J2 J3 (2)2
2
,
d e(1 2 3 ) exp f
e
J1 |
x1 xc |2
(6.1)
where f plays the role of taking into account the corrections. This implies that we
cannot ignore total derivatives of the type
d (1 2 3 )
e
= (1 2 3 )e(1 2 3 )
d
inside the integral. It should be remarked that for this integral to be well defined we have to
assume an analytic continuation in 1 2 3 . Thus the shift i i + i of conformal dimensions which is necessary for the extremal case is actually a general premise for
our approach. Furthermore, the short distance limit 0 must be taken after evaluating
the integral with f JJ2 J1 3 2 = fy(1 y)J1 2 being kept fixed. In other words, the order of
two limits, large J and small , must be carefully chosen such that the integral is meaningful. Note that by overall scaling xr xr / the short-distance limit can equivalently be
regarded as a large-distance limit in which |
x23 | is fixed while x1 is sent to infinity.
The expression (6.1) is basically why we cannot impose conservation of energy for our
Euclidean S-matrix. Recall that in the argument for the uniqueness of 3-point functions
before the large-J limit, total spacetime derivatives are assumed to be vanishing. That
is justified in the case of the GKPWitten relation since the integral with respect to the
interaction point extends to the whole bulk spacetime. In contrast to this, in our tunneling picture emerging by taking the large-J limit, we have to be very careful when total
derivatives are involved.
In connection with this, it is appropriate to reconsider the suggestion made in our
first work [3] concerning a possible integrable structure of holographic string field theory, namely, the interaction term is actually obtained from the free theory by making a
similarity transformation expressed using the CFT coefficient. In the present first order
approximation, this can equivalently be formulated by a canonical transformation
1 2
)( ),
( ) C (
( ) c ( ) = ( ) + C
2
1
) c ( ) = (
) C(
) + C 2 ( ),
(
)(
2
(6.2)
(6.3)
3-point OPE coefficients. Its connection with the mixing of double-trace operators for the impurity-preserving
sector was clarified there. For impurity non-preserving processes, 3-point functions cannot be related to the mixing matrix of a dilatation operator, since by definition we are treating operators with definite conformal dimensions
which are eigenstates of dilatation.
where C is proportional to the CFT coefficient itself

C123
J2 J3 (2 +3 1 )/2
C 123 = f
2 +3 1
J1
(
+ 1)
2
because of the relation between the CFT coefficient and the 3-point coupling 123
(2)
(3)
(1)
H2 + H2 H2 C 123 = 123 .
77
(6.4)
(6.5)
Note that we are here using a symbolic notation for brevity. It should be a trivial task to
convert those formal expressions using the bra-ket notation for string fields. As is easily
seen, the canonical transformation, (6.2) and (6.3), generates the 3-point interaction term
123 from the free Hamiltonian. This shows that the equation of motion of our holographic
string field theory is integrable provided that the CFT coefficient is well defined.
Does this imply that the three-point functions are obtained from the free string-field theory by the field redefinition corresponding to this canonical transformation? The answer to
this question is clearly no, since the Euclidean S-matrix is invariant under field redefinition. Recall that the equivalence with respect to the equation of motion does not in general
imply the equivalence of quantum mechanical amplitudes, because the surface terms often
play crucial roles.
Let us briefly check this using the LSZ formalism. The 3-point S-matrix elements
2 )(
3 ) at the
are defined as the residue of the 3-point Green functions (1 )(
poles at 1 = i1 , 2 = i2 and 3 = i3 , respectively, in the complex-energy
(i , i = 1, 2, 3) plane corresponding to external lines. If the field redefinition could generate a 3-point S-matrix element from the free theory, that would be obtained from the Green
function

c (1 ) c (2 ) c (3 ) free ,
(6.6)
where we compute the Green function using the free field action for the original string
) (not for the transformed fields c and c ). Assuming that the
fields ( ) and (
R-charge angular momenta of external states are J1 (= J2 + J3 ), J2 and J3 respectively
corresponding to our convention throughout the present work, it is sufficient to study the
following part of this Green function,

1 2
2 )(
3)
2 )(
3)
c (1 )(
C
(1 )(
free
free
2

3) .
= C (1 )(2 ) free (1 )(
free
(6.7)
Note that in this computation we do not use any partial integration with respect to the
positions of interaction, since we do not make the change of field variables explicitly for
the action itself. This expression has poles at the correct positions in the complex energy
plane for the lines 2 and 3, but not for the line 16 except for the extremal case where the
interaction vertex itself vanishes. As above, we assume the prescription i i + i for
treating the extremal correlators to remove the degeneracy. Under this definition, the Green
6 The cutoff (6.1) does not affect the positions of the leg poles.
78
Fig. 5. The diagram which contributes to the Green function of free string field after the field redefinition.
functions in general, including the extremal case, do not have correct poles corresponding
to external line 1, and hence the S-matrix elements vanish. Therefore, the free action cannot
in general generate the S-matrix elements.
When we represent the Green function in the path-integral formalism, we would be able
to discuss the above problem by making the change of the integrated string-fields from
to the redefined ones (c , c ). Then, it would transform the free action to one with
(, )
the interaction term and with crucial surface terms. The above direct treatment without
using the change of the integration variables clearly shows that such an action generated
by the canonical transformation does give only vanishing S-matrix, due to cancellation
between local interaction terms and surface terms which are simultaneously generated by
the change of integration variables. To summarize, the conclusion of this subsection is that
our holographic relation is essentially unique, independently of possible field redefinitions,
from both viewpoints of the large-J limit of the GKPWitten relation and of the logic of
string field theory to the present order of approximations.
In passing, we note that the above ambiguity of the extremal case in the bulk S-matrix
is essentially the same thing on the gauge-theory side as the ambiguity mentioned in [17]
with respect to the mixing of sugra BMN operators with double-trace operators. As we
have argued in [2], however, we should not mix the double-trace operators for sugra modes.
But if one wishes, one could mix the double-trace operators and correspondingly make a
field-redefinition appropriately such that one obtains the same 3-point extremal correlators on both sides. The degree of arbitrariness is the same on both sides. So the existence
of this ambiguity itself may be regarded as the consistency of our Euclidean S-matrix
interpretation of holographic mapping in the PP-wave limit. We do not however adopt
this viewpoint since without analytic continuation the extremal correlators cannot be computed from string-field theory umambigously. This peculiar ambiguity is related to a well
known accidental degeneracy among multi-particle states of massless particles with strictly
collinear momenta. In any case, there is no ambiguity at all for impurity non-preserving
cases treated in the present paper.
6.2. Higher-order effects
Extension of our ideas to higher orders is an important next problem. In the present
paper, we have only treated the leading order results with respect to 1/-expansion (more
precisely, 1/(r) R 2 /J -expansion) to the first order in the genus-expansion parameter 1/N . But in principle our holographic relation should be valid to all order in 1/
within the planar approximation. Actually, it is straightforward to compute next or next-
79
to-next orders on the string side, since the Neumann functions are known [18,20] apart
from non-perturbative contributions. Corresponding computations on the gauge-theory
side, however, require us to obtain 3-point functions at least up to 2-loop orders, including all mixing effects. Once we go to such higher-orders, an interesting puzzle arises. The
factor G which played crucial roles in the foregoing discussions gives corrections of the
form 2n logm (n > 0), since the large behavior (2.10) of the function f is correct
up to non-perturbative exponential corrections, according to the above references. If we
ignore the would-be non-perturbative terms of the Neumann functions, this implies that,
when they are not protected, 3-point correlation functions are subject to these peculiar corrections which cannot be derived by the ordinary perturbation theory on the gauge-theory
side.
At present, we are not sure how this should be interpreted. A possibility may be that
the total sum of non-perturbative corrections on the Neumann functions might generate
such logarithmic terms and cancel these corrections when m = 0. For possible subtleties
of non-perturbative corrections, we refer the reader to [19] and also to [20]. Or, if such
terms with non-zero m are not canceled on the string side, non-perturbative effects (or
exact summation over the whole perturbation series) both on the gauge-theory side and on
the string side might be responsible for complete understanding of string/gauge duality. Of
course, it is also possible that our holographic relation itself may be subject to corrections.
The higher orders with respect to 1/N is also important. Since our holographic relation
is originally derived for sugra modes in the planar approximation, it is not immediately
clear whether our conjecture should be valid for non-planar cases. In principle, we have
to examine the same limiting procedure for such cases as we have discussed for planar
3-point functions. Following our philosophy, all rules of the holographic mapping should
be derived on the basis of the correspondence between correlation functions on both sides.
As for higher-genus corrections, it is not evident whether the computation of loop corrections on the string side can be closed within the PP-wave in the presence of the sum
over intermediate states. Recently, it has been reported [21] that the energy shift computed
from the |H3 SV -type string vertex disagrees with the gauge-theory prediction even at the
leading order in the first non-planar correction, if the intermediate states are restricted to
impurity-preserving sector.
Even if it could be closed within the PP-wave limit, we have to fix possible higherorder corrections for susy generators and Hamiltonians themselves on the string field side.
In our first paper [3] in this series, we have suggested that in planar (tree in the sense of
string field theory) approximation, the higher-point vertices may be derived by a similarity transformation from free theory whose lowest-order form is nothing but the canonical
transformation discussed in the previous subsection. If we include non-planar corrections,
these (similarity or perhaps unitary in the space of string fields) transformations generate
1/N corrections for vertices and even for quadratic kinetic terms. In order to confirm this
suggestion, it is important to investigate first the extension of our arguments for higherpoint amplitudes in the planar limit. It has been pointed out that the structure of 4-point
correlation functions on the gauge-theory side suffers from some ambiguity [17] in determining the large-J limit. From our point of view, we have to take into account the fact
that the approximation of a single tunneling trajectory can be justified only if we first take
the short-distance limit for operator products such that the order of the distances at each
80
set of operator products at initial and final points at the boundary satisfies 2 c/J with
some finite c and finally let c 0. It is clear that the naive limit J on fixing the
spacetime configuration of operators at the boundary, which indeed leads to discontinuous results depending on the positions of operators, is not allowed. Careful analyses are
required, however, to make more precise the general structure suggested in [3] for holographic mapping for higher-point S-matrix elements.
Note added
After completing this manuscript, we received a preliminary manuscript by H. Shimada
[22] who tried to establish the holographic relation in an approach which is slightly different from ours. His argument and result, though restricted yet to impurity-preserving sector
with scalar excitations, seem to be consistent with ours in [2] and complementary to our
argument for the uniqueness of the holographic relation.
Acknowledgements
We would like to thank H. Shimada for discussions. The present work is supported
in part by Grant-in-Aid for Scientific Research (No. 13135205 (Priority Areas) and No.
16340067 (B)) from the Ministry of Education, Science and Culture.
Appendix A. Large behavior of Neumann coefficients

rs for readers convenience folWe summarize here the leading large behavior of N mn
lowing [18,20]:
(1)m+n
22
,
=
N mn
4|(1) |y
33
=
N mn
23
N mn
=
1
,
4|(1) |(1 y)
(1)m+n+1 sin(ny)
21
N mn
,
=
y(n m/y)
for (m, n) = (0, 0), and
11
N 00
= 0,
12
N 00
= y,
1
,
4|(1) | y(1 y)
1
33
=
N 00
,
4|(1) |(1 y)
23
N 00
=
for m = 0 = n.
(1)m+1
,
4|(1) | y(1 y)
(1)m+n+1 sin(my) sin(ny)

11
N mn
,
=
|(1) |
31
N mn
=
(1)n sin(ny)
1 y(n m/(1 y))

13
N 00
= 1 y,
22
=
N 00
(A.1)
(A.2)
(A.3)
(A.4)
1
,
4|(1) |y
(A.5)
81
The asymptotic form in the large limit of f = 1 4(1) (2) (3) K is given by
f=
1
.
4|(1) |y(1 y)
References
Mills, JHEP 0204 (2002) 013, hep-th/0202021.
[2] S. Dobashi, T. Yoneya, Resolving the holography in the plane-wave limit of AdS/CFT correspondence,
hep-th/0406225.
[3] S. Dobashi, H. Shimada, T. Yoneya, Holographic reformulation of string theory on AdS5 S 5 background
in the PP-wave limit, Nucl. Phys. B 665 (2003) 94, hep-th/0209251.
[4] T. Yoneya, What is holography in the plane wave limit of AdS(5)/SYM(4) correspondence?, hep-th/0304183,
expanded from the paper published in the Proceedings, Prog. Theor. Phys. 152 (2003) 108.
[5] M. Asano, Y. Sekino, T. Yoneya, PP wave holography for Dp-brane backgrounds, Nucl. Phys. B 678 (2004)
197, hep-th/0308024;
M. Asano, Y. Sekino, Large N limit of SYM theories with 16 supercharges from superstrings on Dp-brane
backgrounds, hep-th/0405203;
M. Asano, Stringy effect of the holographic correspondence for Dp-brane backgrounds, hep-th/0408030.
[6] M. Spradlin, A. Volovich, Superstring interactions in a pp wave background, Phys. Rev. D 66 (2002) 086004,
hep-th/0204146;
M. Spradlin, A. Volovich, Superstring interactions in a pp wave background 2, JHEP 0301 (2003) 036,
hep-th/0206073;
A. Pankiewicz, More comments on superstring interactions in the pp wave background, JHEP 0209 (2002)
056, hep-th/0208209;
A. Pankiewicz, B. Stefanski Jr., PP wave light cone superstring field theory, Nucl. Phys. B 657 (2003) 79,
hep-th/0210246;
A. Pankiewicz, An alternative formulation of light cone string field theory on the plane wave, JHEP 0306
(2003) 047, hep-th/0304232;
A. Pankiewicz, B. Stefanski Jr., On the uniqueness of plane wave string field theory, hep-th/0308062.
[7] P. Di Vecchia, J.L. Petersen, M. Petrini, R. Russo, A. Tanzini, The three string vertex and the AdS/CFT
duality in the pp wave limit, hep-th/0304025.
[8] C.S. Chu, V.V. Khoze, Correspondence between the three point BMN correlators and the three string vertex
on the pp wave, JHEP 0304 (2003) 014, hep-th/0301036.
JHEP 0210 (2002) 068, hep-th/0209002.
[11] N.R. Constable, D.Z. Freedman, M. Headrick, S. Minwalla, L. Motl, A. Postnikov, W. Skiba, PP wave string
[12] J. Gomis, S. Moriyama, J. Park, SYM description of SFT Hamiltonian in a pp wave background, Nucl. Phys.
B 659 (2003) 179, hep-th/0210153;
J. Gomis, S. Moriyama, J. Park, SYM description of pp wave string interactions: singlet sector and arbitrary
impurities, Nucl. Phys. B 665 (2003) 49, hep-th/0301250.
[13] A. Parnachev, A.V. Ryzhov, Strings in the near plane wave background and AdS/CFT, JHEP 0210 (2002)
066, hep-th/0208010.
[14] N. Beisert, BMN operators and superconformal symmetry, Nucl. Phys. B 659 (2003) 79, hep-th/0211032.
[15] U. Grsoy, Vector operators in the BMN correspondence, JHEP 0307 (2003) 048.
[16] S.-M. Lee, S. Minwalla, M. Rangamani, N. Seiberg, Three point functions of chiral operators in D = 4,
N = 4 SYM at large N , Adv. Theor. Math. Phys. 2 (1998) 697, hep-th/9806074.
82
[17] C. Kristjansen, J. Plefka, G.W. Semenoff, M. Staudacher, A new double scaling limit of N = 4 super-Yang
Mills theory and pp wave strings, Nucl. Phys. B 643 (2002) 3, hep-th/0205033.
[18] Y.H. He, J.H. Schwarz, M. Spradlin, A. Volovich, Explicit formulas for Neumann coefficients in the plane
wave geometry, Phys. Rev. D 67 (2003) 086005, hep-th/0211198.
[19] I.R. Klebanov, M. Spradlin, A. Volovich, New effects in gauge theory from pp-wave superstrings, Phys.
Lett. B 548 (2002) 111, hep-th/0206221.
[20] J. Lucietti, S. Schafer-Nameki, A. Sinha, On the plane-wave cubic vertex, Phys. Rev. D 70 (2004) 026005,
hep-th/0402185.
[21] P. Gutjahr, A. Pankiewicz, New aspects of the BMN correspondence beyond the planar limit, hepth/0407098.
[22] H. Shimada, in preparation.
The neutralino sector of the next-to-minimal

supersymmetric Standard Model
S.Y. Choi a , D.J. Miller b , P.M. Zerwas c
a Physics Department, Chonbuk National University, Chonju 561-756, South Korea
b School of Physics, The University of Edinburgh, Edinburgh EH9 3JZ, Scotland, UK
c Deutsches Elektronen-Synchrotron DESY, D-22603 Hamburg, Germany
Received 24 August 2004; accepted 5 January 2005

Available online 28 January 2005
Abstract
The next-to-minimal supersymmetric Standard Model (NMSSM) includes a Higgs iso-singlet superfield in addition to the two Higgs doublet superfields of the minimal extension. If the Higgs fields
remain weakly coupled up to the GUT scale, as naturally motivated by the concept of supersymmetry, the mixing between singlet and doublet fields is small and can be treated perturbatively. The mass
spectrum and mixing matrix of the neutralino sector can be analyzed analytically and the structure
of this 5-state system is under good theoretical control. We also determine decay modes and production channels in sfermion cascade decays to these particles at the LHC and pair production in e+ e
colliders.
PACS: 12.60.Jv
1. Introduction
The minimal supersymmetric Standard Model (MSSM) [1,2] opens the path to the
analysis of supersymmetric theories. Arguments have been advanced however that suggest extensions beyond this minimal version. One well-motivated example is the nextto-minimal supersymmetric Standard Model (NMSSM) [3] in which an iso-singlet Higgs
E-mail address: zerwas@desy.de (P.M. Zerwas).
doi:10.1016/j.nuclphysb.2005.01.006
84
S.Y. Choi et al. / Nuclear Physics B 711 (2005) 83111
superfield S is introduced in addition to the two iso-doublet Higgs fields H u,d incorporated
in the MSSM to generate electroweak symmetry breaking. Such an extension offers a possible solution of the problem, generating in a natural way, a value of the order of the
electroweak breaking scale v; this is achieved by identifying , apart from the O(1) coupling, with the vacuum expectation value of the scalar component S of the new iso-singlet
field. [For a recent summary of this construct see Ref. [4]; a useful code has been made
available in Ref. [5].]
The superpotential of the NMSSM includes, besides the usual MSSM WY Yukawa components, an additional term, which couples the iso-singlet to the two iso-doublet Higgs
fields, plus the self-coupling of the iso-singlet:
H u H d ) + 1 S 3 .
W = WY + S(
3
(1)
The two parameters and are dimensionless. By demanding the Higgs fields remain
weakly interacting up to the GUT scale, the two couplings are bounded at the electroweak
scale by the inequalities , 0.7. While the scalar Higgs sector includes several soft
supersymmetry breaking parameters, the Lagrangian of the gaugino/higgsino sector is
complemented only by the familiar SU(2) and U(1) gaugino mass terms. As a result, the
parameter space of the neutralino sector is much less complex than the Higgs space.
The superpotential without the singlet self-coupling, i.e., = 0, incorporates a Peccei
and
Quinn (PQ) symmetry: {H u (1), H d (1), S(2),
Q(1),
U (0), D(0),
L(1),
E(0)}.
Q
L are the quark and lepton SU(2) doublet superfields, while U , D and E are the up- and
down-quark and lepton SU(2) singlet superfields, respectively. The integer of each parenthesis indicates the PQ charge of the corresponding superfield. The
spontaneous breaking
of this symmetry by the non-zero vacuum expectation value vs / 2 of the scalar S field
gives rise to a massless Goldstone boson. However, when = 0, the mass is lifted to a
non-zero value by the self-interaction of the S field. Still, a discrete Z3 symmetry is left
which would lead to the formation of domain walls in the early Universe. This problem
can be tamed by introducing new interactions of the inverse Planck size that, however, do
not affect the low-energy effective NMSSM theory [6].
In contrast to the Higgs sector, masses and mixings in the chargino system are not
affected by the singlet extension. [Of course new decays such as S + or +
5 H + may be possible if allowed kinematically.]
So far the supersymmetric particle spectrum of the NMSSM has received only little
attention in the NMSSM literature, Refs.[3,79]. In this report we attempt a systematic
analytical analysis of the neutralino system. In contrast to the MSSM where exact solutions
of the mass spectrum and mixing parameters can be constructed mathematically in closed
form, this is not possible any more for the NMSSM in which the eigenvalue equation is of
5th order, not allowing closed
solutions.2 However, since the coupling between singlet and
doublet fields is weak, v/
2 O(10 ) GeV, compared with the typical supersymmetry
scale M1,2 and = vs / 2 O(103 ) GeV, a perturbative expansion of the solution gives
rise to a good approximation of the mass spectrum while the magnitude of the matrix
elements in the mixing matrix is at least qualitatively well understood. The usefulness of
a perturbative expansion has also been noticed in Ref. [9]; however, here, extending the
85
Higgs analysis in Ref. [4], we work out this approach systematically for all facets of the
NMSSM.
While plays a crucial rle in the Higgs sector, it is less crucial for the neutralino
system. The size of vs , with vs 15v to maintain a link with the electroweak scale, just
determines the singlino mass before modified by mixing effects. Once masses and mixings
are determined, the couplings of the neutralinos to the electroweak gauge bosons and to
scalar/fermionic matter particles are fixed. Decay widths and production rates of the five
neutralinos can subsequently be predicted for squark cascades at the LHC [10] and e+ e
annihilation at prospective linear colliders [11].
The report is organized as follows. In Section 2 we describe the neutralino sector of
supersymmetric models in which the pair of Higgs doublet superfields is augmented by
an additional iso-singlet field. In Section 3 we show how, for a naturally expected weak
coupling, the properties of the four standard neutralinos are modified; moreover the properties of the fifth neutralino, the new singlino-dominated state, are calculated. All these
spectra and mixings are predetermined analytically before the surprisingly good quality
of the weak-coupling expansion is demonstrated by comparison with numerical solutions.
In this way we achieve a satisfactory theoretical understanding of the system. In the limit
of large gaugino mass parameters M1,2 compared with the higgsino mass parameter , or
vice versa, the MSSM part can be easily diagonalized analytically and a clear and simple picture of the entire system emerges. The section is concluded by a lovely toy model
in which we set M1 = M2 and tan = 1; this set allows us to solve the system exactly,
leading to transparent closed expressions for the neutralino mass spectrum and the mixing
parameters. A sample of decay widths and production cross sections for the neutralinos
is presented in Section 4. The results are summarized in Section 5 and technical details
of the diagonalization procedure for the 5 5 neutralino mass matrix are described in the
Appendix A.
2. The NMSSM neutralino sector

2.1. The NMSSM neutralino mass and mixing matrix
The Lagrangian of the neutralino system can be derived from the superpotential defined
in Eq. (1), complemented by the SU(2) and U(1) mass terms in the soft supersymmetry
breaking Lagrangian. After breaking the [electroweak] symmetry spontaneously by introducing non-zero vacuum expectation values of the iso-doublet and singlet Higgs fields,

1
1
0
v
,
Hd = cos
,
S = vs / 2,
Hu = sin
(2)
v
0
2
2
the Higgs-higgsino mass parameter
= vs / 2
(3)
is generated and, subsequently, the neutralino mass matrix

M X
M5 =
X T
86
with a hierarchical structure as analyzed in Appendix A, can be written in detail as:
M1
0
mZ c sW mZ s sW
0
mZ c cW mZ s cW
0
0
M2
s
M5 = mZ c sW mZ c cW
.
mZ s sW mZ s cW
0
c
0
(4)
This 5 5 mass matrix is constructed from the standard 4 4 MSSM neutralino mass
matrix M in the upper left corner, the mass term of the higgsino component S of the
singlet superfield S,
= 2vs / 2
(5)
and the mixing between doublets and singlet parameterized by
= v/ 2.
(6)
W 3 , H 0 , H u0 , S).
As usual, M1 and
The mass matrix M5 is defined in the group basis (B,
d
M2 are the soft SUSY breaking U(1) and SU(2) gaugino mass parameters, tan is the ratio
of the vacuum expectation values of the two neutral SU(2) Higgs doublet fields (as defined
in Eq. (2)), s = sin , c = cos , and sW , cW , tW are the sine, cosine and tangent of the
electroweak mixing angle W .
Since the neutralino mass matrix (4) is symmetric and real, it can be diagonalized by
an orthogonal matrix V 5 . The mass eigenvalues are real but not necessarily positive. They
can be mapped onto positive values by supplementing the rotation matrix to N 5 = 5 V 5
with the diagonal phase matrix ( 5 )kl = 1(i)kl in case of positive (negative) eigenvalues
so that N 5 M5 N 5 is positive diagonal. The physical neutralino states i0 [i = 1, . . . , 4]
are ordered according to ascending mass values while 50 is the predominantly singlino
state.1 They are mixtures

W 3 , H d , H u , S
[i = 1, . . . , 5],
i0 = Nij5 B,
(7)
j
of the U(1), SU(2) gauginos, the doublet higgsinos and the singlino.
The unitary matrix N 5 defines the couplings of the mass eigenstates i0 to other particles. For the neutralino production processes it is sufficient to consider the neutralino
neutralino-Z vertices
0 0
g 5 5
5 5
iL Z j L =
Ni3 Nj 3 Ni4
Nj 4 ,
2cW
0 0
g 5 5
5 5
Ni3 Nj 3 Ni4
Nj 4
iR Z j R = +
(8)
2cW
1 Note that the ordering of the masses according to ascending values is accomplished easily after the diagonalization process is finalized. For the intermediate steps it is however convenient to use the indices i = 1, 2, 3, 4
for the former MSSM type states and i = 5 for the additional state originating from the singlino field as suggested
by the structure of M5 in Eq. (4).
and the fermionsfermionneutralino vertices

g f 5

0
f
5
sW ,
I3 Ni2 cW + Qf I3 Ni1
iR fL |fL = 2
cW
0

fR |fR = 2gQf tW N 5 .
iL
i1
87
(9)
The coupling g is the SU(2) gauge coupling, I3 is the SU(2) isospin 3-component and Qf
is the electric charge of the fermion f . In Eq. (9) the coupling to the higgsino component,
which is proportional to the fermion mass, has been neglected for light flavors. The more
involved Higgs couplings to the neutralinos are listed in detail in Section 4.
2.2. NMSSM parameter range
In contrast to the Higgs sector only two additional parameters and are introduced
in the NMSSM neutralino sector as compared to that of the MSSM including . Assuming that the fields remain weakly interacting up to the GUT scale, the two couplings are
bounded at the electroweak scale by the inequality

2 + 2 0.7.
(10)
Moreover, the renormalization group (RG) evolution of the couplings points to as
preferential target domain if the evolution starts from a random distribution of the couplings
U , U 2 at
the GUT scale [4].
While v = vu2 + vd2 = 246 GeV is fixed by the Fermi coupling GF , the parameter vs
should be expected in the same range,
vs 15v
(11)
in compliance with the arguments for introducing the NMSSM. A RG analysis of the entire
set of parameters shows that a low value of tan is favored [4]. Current experimental analyses of tan assume MSSM relations for the couplings; they are modified in the NMSSM
and the results in this extended scenario are less restrictive.
Since the size of the doubletsinglet mixing is set by = v/ 2, the mixing interaction
is expected to be small2 compared with the standard supersymmetry scales,
= vs / 2 and/or M1,2 for which values O(1 TeV) are anticipated. As a result, transparent expressions can be found by performing a systematic expansion for small mixing
between the gauginos/doublet higgsinos and the singlino, measured by the small size of
the parameter relative to the other parameters in the mass matrix.
In summary, at tree-level the NMSSM neutralino sector described above has six
free parameters which we choose as and in addition to the MSSM parameters:
{{M1 , M2 , tan , }; , }. Sometimes it is convenient to re-express , and in
terms of , and vs . The spectrum of the NMSSM neutralino sector will now be analyzed
in detail.
2 is expected to have a lower limit from cosmological arguments; private communication with U. Ell
wanger, see also Ref. [12]. For too small a value of , i.e., very much below the typical scale v/ 2, the amount
of cold dark matter may exceed the measured value of CDM 0.25; detailed analyses are not available yet.
88
3. NMSSM small-mixing scenarios

In general the diagonalization of the 5 5 NMSSM mass matrix M5 cannot be performed analytically in closed form. However, if the doubletsinglet coupling is weak, an
approximate analytical solution can be found after the 4 4 MSSM submatrix M is analytically diagonalized following the elaborate standard procedures in Ref. [13].
The orthogonal matrix V 5 which transforms M5 to the diagonal mass matrix MD
5 =
diag[m1 , . . . , m5 ] is conveniently split into a matrix V diagonalizing the 4 4 submatrix
M and a matrix performing subsequently the block diagonalization of the 4 4 and 1 1
submatrices. After the block-diagonalization, the upper left MSSM mass matrix MD =
2, m
3, m
4 ] needs not be rediagonalized for small doubletsinglet mixing, as
diag[m
1, m
proved in Appendix A. The final result for the orthogonal matrix V 5 may be written in the
simple form:

(V )
144 12 (V )(V )T
V
0
5
.
V
(12)
0 111
(V )T
1 12 (V )T (V )
The doubletsinglet 4-component mixing vector can be expressed in terms of the gaugino/higgsino parameters as
M2 mZ sW c2

M1 mZ cW c2 2
=
(13)
det(M ) M M (c s ) M m s
1
12
m2 c
M1 M2 (s c ) M12
Z
with the abbreviations

M1 = M1 ,
M2 = M2 ,
( )
( )
( )
2
2
M12 = M1 cW
+ M2 sW
(14)
and the determinant

det(M ) = M1 M2 2 2 + M12
(s2 + )m2Z .
(15)
The mixing with the singlet alters the MSSM mass eigenvalues m
i [i = 1, . . . , 4] to
O( 2 ),3 and correspondingly the singlet mass
m
5 = .
(16)
The shifts are given as

mi = m
i +
m5 = m
5
1
(V X)2i
m
i m
5
4

i
[i = 1, . . . , 4],
1
(V X)2i
m
i m
5
(17)
with the 4-component vector X (0, 0, s , c ). [The eigenvalues are not necessarily ordered sequentially, and, if some of them are negative, the additional phase rotation
3 Note however that small mass differences |m
i m
5 | may enhance the mixing effects.
89
transforms them to positive physical masses.] Even for small mixing, the 5th eigenvalue m5
may differ significantly from the singlino mass parameter m
5 = if is small. However,
even though the relative shift may be large, the absolute shift remains small, of second
order. Trivially, the eigenvalues fulfill the spur formula
5

mi = M1 + M2 + ,
(18)
i=1
which is independent of the parameters and .

The doubletsinglet mixing generates a singlino component in the wave functions of
the original MSSM neutralinos i0 [i = 1, . . . , 4] of the size
Vi55
4

Vij j
(19)
j =1
linear in the mixing parameter to first approximation as expected for off-diagonal elements.
Reciprocally, the singlino component in the wave function of 50 is reduced to
1 2
i
2
4
5
V55
1
(20)
i=1
differing from unity only to second order in the mixing as expected for diagonal elements.
As long as the mixing parameter is significantly smaller than the other parameters,
we find that the approximation works remarkably well, as demonstrated in Fig. 1. As an example both the exact numerical solutions and the approximate solutions for the neutralino
masses are shown as a function of for a favored parameter set P of broken PQ symmetry, = 120 GeV with M1 = 250 GeV, M2 = 500 GeV, = 170 GeV and tan = 3. The
exact and approximate solutions agree rather well as long as is less than about 80 GeV,
as the mixing corrections are of second order in .
In Fig. 2 the exact numerical solution (solid) and the approximate solution (dashed)
are compared for the gaugino/higgsino and singlino components, {|(N 5 )51 |, |(N 5 )53 |,
|(N 5 )55 |}, of the lightest singlino-dominated neutralino as a function of for the same
parameter set P. Since the matrix V 5 is in general linear in the mixing term , the approximate solution differs from the exact solution already for smaller values of in the
reference point P in which m
5 = is quite close to the higgsino parameter , though the
characteristic features remain valid up to 40 GeV.
To fully exhaust the potential of our analytical method we perform the complete
NMSSM diagonalization for the two standard limits analyzed in general within the MSSM:
M1,2 || and vice versa, both complemented of course by small doubletsinglet mixing
max{M1,2 , ||}.
3.1. Small singlino mass parameter
The first special analysis should be performed for small singlino mass parameter ,
which implies a slightly broken PQ symmetry 1 as favored by the RG flow of this
90
Fig. 1. The exact numerical solution (solid) and the approximate solution (dashed) for the masses of the five
neutralino states in the NMSSM as a function of for the parameter set P = { = 120 GeV, M1 = 250 GeV,
M2 = 500 GeV, = 170 GeV, tan = 3}. The ordering of the mass spectrum is m5 , m1 , m2 , m3 , and m4 in
increasing mass, i.e., the state 50 is the lightest neutralino for the given parameter set.
Fig. 2. The exact numerical solution (solid) and the approximate solution (dashed) for the gaugino/higgsino and
5 |, |N 5 |, |N 5 |}, of the lightest singlino-dominant neutralino as a function of for
singlino components, {|N51
53
55
the same parameter set P as in Fig. 1.
91
coupling in grand unified theories. Due to the small doublet-singlet mixing the structure of
the original MSSM neutralinos i0 [i = 1, . . . , 4] is changed little while the properties of
the 5th neutralino 50 , the lightest for small , are determined jointly by both the singlino
parameter and the mixing parameter .
3.1.1. Large gaugino mass parameters
As a first example, we consider the case with large gaugino mass parameters, i.e.,
M1,2 || mZ , .
To begin, the 4 4 diagonalization matrix V defined in Eq. (12) can be parameterized
up to second order according to standard MSSM procedure, cf. Ref. [13], as

122
122
VX
0
VG 0
.
V
(21)
VXT 122
0 VH
0
R/4
The 2 2 /4 rotation R/4 = (1 iy )/ 2 shifts the [34] off-diagonal elements

[, ] onto the diagonal axis [, ]. The second matrix, VX ,

c sW mZ /M1 s sW mZ /M1
VX =
(22)
c cW mZ /M2 s cW mZ /M2
removes the mixing between the blocks of the two gaugino and the two higgsino states.
The components VG and VH diagonalize the gaugino and higgsino blocks themselves:

2 2
1 sW
mZ /M12
0
VG 122
2 m2 /M 2 ,
0
cW
2
Z
2

2 m2 /2M 2 M 2
1 (1 + s2 )M12
0
Z
1
2
VH 122
2 m2 /2M 2 M 2
0
(1 s2 )M12
2
Z
1 2
(23)
2 = M 2 c2 + M 2 s 2 , respectively. V and V relate to a diagonal form of the
with M12
G
H
1 W
2 W
gauginohiggsino mass matrix for large M1,2 and . Their off-diagonal matrix elements
are of second order and can be omitted consistently as they would effect the eigenvalues
only to fourth order.
After these steps are performed, the 4 4 mass submatrix is diagonal and the complete
symmetric mass matrix M5 takes the form
m
1
0
0
m
2
+ c
m
3
,
5
M5 M
(24)
m
4 c+
0 0 + c c+
where, in an obvious
notation, zero elements are suppressed for easier reading, and
c = (c s )/ 2 is used as abbreviation. The MSSM neutralino mass eigenvalues are

given by
m
1 = M1 +
m2Z 2
s ,
M1 W
m
2 = M2 +
m2Z 2
c ,
M2 W
92
m
3 =
M12
m2 (1 + s2 ),
2M1 M2 Z
m
4 =
M12
m2 (1 s2 ).
2M1 M2 Z
(25)
5 by choosing the proper form of V in V 5 .

It remains to diagonalize M
In the limit of large gaugino mass parameters, the doubletsinglet 4-component mixing
vector reduces to a simple expression

(0, 0, c , s )T
(26)
and the entire matrix V 5 can be written, up to second order, in the form
1
0
1

2
c
1 42 (1 s2 )
V 0
5
0 1
2
1 42 (1 + s2 ) c+
00
c
c+
1 22
(27)
with zeros suppressed in the upper 4 4 matrix, and antisymmetric in the off-diagonal
elements.
The rotations lead eventually to the diagonal mass matrix, of which the mass eigenvalues
to the desired order are given by
m1 M1 +
m3
m2Z 2
s ,
M1 W
m2 M2 +
m2Z 2
c ,
M2 W
2
M12
m2Z (1 + s2 ) + (1 s2 ),
2M1 M2
2
m4
2
M12
m2Z (1 s2 ) (1 + s2 ),
2M1 M2
2
m5 +
2
s2
(28)
2 + M s 2 ]. For the ordering of the eigenvalues and the flipping of the

[recall M12 = M1 cW
2 W
signs to positive physical masses the previous general remarks apply.
Two points should be emphasized explicitly. While the large gaugino masses m1 , m2
are not affected by the singlino, it does affect the higgsino states 3, 4 to second order. The
singlino mass is also affected to second order; however the mixing term can be leading if
the singlino mass parameter is small.
The mixing in the wave-functions is described by the components of itself [since the
4 4 matrix V deviates from unity, apart from the /4 rotation, only to second order in
the small parameters of the order of the SUSY scales]:

1
1
0, 0, (c s ), (c + s ) ,
Vi55
2
2
i
V5i5
(0, 0, c , s )i ,
2
.
22
5
V55
1
93
(29)
3.1.2. Large higgsino mass parameter

As a second example, we consider the case with large higgsino mass parameter, i.e.,
|| M1,2 mZ , . This example is complementary to the previous case.
The overall diagonalization 4 4 matrix V can be parameterized in the same form as
that in Eq. (21). The 2 2 matrix VX describing the mixing between the two ensembles of
the gaugino states and the higgsino states reads
mZ
VX
sW s
cW s
sW c
cW c

,
(30)
leading to a block-diagonal mass matrix composed of a 2 2 matrix, depending on M1 and

M2 with small corrections of the order of m2Z /, and a 2 2 mass matrix, depending only
on the higgsino parameter . The 2 2 blocks VG and VH in the gaugino and higgsino
sector may be written
VG 122
1
2
1
VH 122
2
2 m2 /2
sW
Z
0
1
0
2 m2 /2
cW
Z
2
2
2 (1 + s2 )mZ /

,
0
1
2
2
2 (1 s2 )mZ /

,
(31)
respectively, after the higgsino submatrix has been diagonalized by the standard R/4 rotation.
These transformations diagonalize the 4 4 submatrix within the block-diagonal matrix
5 of the same form as (24), of which the first four diagonal elements are given by
M
m
1 = M1
m
3 =+
m2Z 2
s s2 ,
W
m2Z
(1 + s2 ),
2
m
2 = M2
m2Z 2
c s2 ,
W
m
4 =
m2Z
(1 s2 ),
2
(32)
The mixing between the doublet-higgsino and singlino states is then described in an
analytic form by a 4-component column vector

(0, 0, c , s )T
(33)
mixing the singlino both with the gauginos and with the doublet-higgsinos. The entire
matrix V 5 can be written up to second order in the form, with antisymmetric off-diagonal
94
elements,
5
V
2 c2
2 m2Z sW
2
2M12 2
2 c2
2 m2Z cW
2
2M22 2
2
2 c
22
2
2 c+
22
c+
1
0
c+
2
1 22

V
0
1
(34)
and with the same abbreviations as before in Eq. (24).
The rotations lead eventually to a diagonal mass matrix, consisting of the mass eigenvalues:
m2Z 2
m2 2
sW s2 ,
m2 M2 Z cW
s2 ,
m2
2
m3 + Z (1 + s2 ) + (1 s2 ),
2
2
m2Z
2
(1 s2 ) (1 + s2 ),
m4
2
2
2
m5 +
s2
m1 M1
(35)
with apparent reciprocity in the MSSM subsystem between gaugino and higgsino parameters in comparison with the previous case, but universal modifications from the doublet
singlet mixing.
Correspondingly, to leading order the coefficients of V 5 involving the singlino index 5
coincide with the elements of the doubletsinglet mixing matrix in Eq. (29).
3.2. Large singlino mass parameter
In the alternative extreme, the PQ symmetry is strongly broken if is large and, equivalently, the singlino mass parameter is large, i.e., , , M1,2 . This limit is not favored
by the renormalization group flow from the GUT scale down to the electroweak scale but
cannot be ruled out a priori on general grounds. The new fifth eigenstate, predominantly
composed of the singlino, would in general be the heaviest state, mixed only weakly with
the iso-doublets and, as a result, coupling weakly to electroweak gauge bosons and matter
fields.
Applying the approximation method described in Appendix A and the general introduction to this section, the neutralino mass matrix can be transformed into the 4 4 and 1 1
block-diagonal form by inserting the mixing column vector
(0, 0, s , c )T

(36)
95
in the V 5 matrix Eq. (12). Note that the mixing column vector (36) is directly proportional
to the 4-component off-diagonal column vector of the mass matrix (4) unlike the column
vector (26) for a small singlino mass parameter.
From the general analysis it is apparent that the first four neutralino masses, of MSSM
type, are modified to the order 2 / through the higgsino part, as is the 5th neutralino
mass. The mass and the 55 wave-function are approximately given by
2
(37)
2
,
22
(38)
m5 +
and
5
V55
1
while doublet components are mixed in to first order,

V5i5
(0, 0, s , c )i
(39)
in parallel to the singlino components of the first doublet-type neutralinos.

In summary, the gaugino/doublet higgsino dominated neutralinos follow the pattern of
the MSSM quite narrowly. Increasing the value of will increase the mass of the new
singlino state (almost) linearly, causing the state to decouple and making the NMSSM very
difficult to distinguish from the MSSM.
3.3. The case with M1 = M2 in the limit of tan = 1
When the two soft-breaking SU(2) and U(1) gaugino masses are equal, M1 = M2 = M
and tan = 1, cf. Ref. [13], the electroweak gauge symmetry guarantees the existence of
a physical neutral state which does not mix with the other states and which has a mass
eigenvalue identical to the modulus M. Furthermore, the gaugino states do not mix with
the singlino state S that
couples only to the specific linear combination of the higgsino
states H b0 = (H u0 + H d0 )/ 2. As a result, one gaugino state mixes only with one higgsino
state while the other orthogonal higgsino state mixes with the singlino state, leading to a
block-diagonal matrix composed of one scalar and two 2 2 matrices.
H a0 ,
This special structure can be made apparent by switching to the mixed basis { , Z,
0
0
3
0
Hb , S} from the original group basis {B, W , Hd , Hu , S} by means of the transformation

cW
sW
W 3
Z

Ha = A5 Hd = 0

Hb
Hu
0
S
S
0
sW
cW
0
0
0
0
0
1
2
1
2
0
0
1
1
2
0 B
0 W 3
0
H d .
0 H u
S
1
(40)
96
5 takes the block-diagonal form

H a0 , H 0 , S}
basis the mass matrix M
In this new { , Z,
b
M 0 0
0
0
0 M mZ 0
0
0
M5 = A5 M5 A5 = 0 mZ 0
(41)
.
0 0 0
0 0 0
This mass matrix generates two two-state mixings between Z and H a0 , and between H b0 and
respectively. The block-diagonal matrix can be diagonalized by the orthogonal matrix
S,

1
5
V =
(42)
Rg/ h
Rh/s
consisting of two 2 2 rotation matrices Rg/ h and Rh/s ,

cos g/ h sin g/ h
cos h/s
Rg/ h =
,
Rh/s =
sin g/ h
cos g/ h
sin h/s
sin h/s
cos h/s

(43)
with the mixing angles determined by the relations

tan g/ h =
tan h/s =
2mZ

,
M (M )2 + 4m2Z
+ +
(44)
( + )2 + 42
The mass eigenvalues can be written completely in analytic form,

m1 = M,

1
m2 = M + + (M )2 + 4m2Z ,
2

1
m3 = M + (M )2 + 4m2Z ,
2

1
m4 = ( + )2 + 42 ,
2

1
m5 = + ( + )2 + 42
2
(45)
with the wave-functions of the neutralinos i0 [i = 1, . . . , 5] determined by the cos/sin of

the mixing angles g/ h and h/s .
It is instructive to study the neutralino mass spectrum in this model for a set of fixed
parameters M = 200 GeV and = 120 GeV, = 100 GeV, by varying the higgsino
mass parameter . The branch character of the eigenvalues {23} and {45} is exemplified
in Fig. 3(a). The tans of the corresponding mixing angles are displayed in Fig. 3(b). With
rising tans we move from scenarios of no mixing, to maximal gaugino/doublet higgsino
in {23} and doublet/singlet higgsino mixing in {45}, finally to gaugino/doublet higgsino
flipping {23}, and doublet/singlet flipping {45}, while the gaugino {1} remains untouched.
97
Fig. 3. (a) The neutralino masses |mi |, mapped onto positive values, and (b) the tangent values of the mixing
angles g/ h and h/s as a function of the higgsino mass parameter for the parameter set: M = 200 GeV,
= 120 GeV, = 100 GeV.
4. Neutralino production and decays

In the MSSM the neutralino sector consists of two gauginos and two higgsinos. Typically the lightest supersymmetric particle (LSP), which is stable under the assumption of
R-parity conservation, is the lightest state of the neutralino mass matrix. The LSP will
appear as one of the final states of each sparticle decay and its non-observability is responsible for the well-known missing energy/momentum signature of supersymmetric particle
production.
The neutralino production and decay properties in the NMSSM with the additional
singlino state depend crucially on the singlino mass with respect to the MSSM neutralino
masses [8]. If the singlino is much heavier than the other states, it will be very rarely produced and so practically unobservable. On the contrary, if the singlino is lighter than the
other states, a singlino-dominated state will be the LSP so that the other neutralino states
will decay, possibly through cascades, into the singlino-dominated LSP.
In this section, we present a qualitative description of the production of neutralinos,
involving at least one singlino-dominated state, such as 50 50 and 10 50 and the subsequent
decays of the neutralino 10 into leptons and light Higgs bosons.
4.1. Singlino production in e+ e annihilation
The production processes
e+ e i0 j0
[i, j = 1, . . . , 5]
(46)
98
are generated by s-channel Z exchange, and t- and u-channel eL,R exchanges. After appropriate Fierz transformations of the selectron exchange amplitudes [with the electron
mass neglected], the transition matrix element of the production process can be written as

T e+ e i0 j0 =
(47)
Q v e+ u e u i0 v j0 .
,=L,R
The transition amplitudes are built up by the sum of the products of chiral neutralino currents and chiral fermion currents. The four generalized bilinear charges Q correspond to
independent helicity amplitudes, describing the neutralino production processes for polarized electrons/positrons [13]. They are defined by the fermion and neutralino currents and
the propagators of the exchanged (s)particles as follows:
QLL = +
DZ
2
2
sW cW
2
I3 Qf sW
Zij DuL gLij ,
DZ
Qf Zij + DtR gRij ,
2
cW

DZ f
2
QLR = 2 2 I3 Qf sW
Zij + DtL gLij
,
sW cW
DZ
QRR = + 2 Qf Zij DuR gRij

cW
QRL =
(48)
with f = e in the production channel. The first term in each bilinear charge is generated
by Z exchange and the second term by selectron exchange; DZ , DtL,R and DuL,R denote
the s-channel Z propagator and the t- and u-channel left/right-type selectron propagators
DZ =
m2Z
s
,
+ imZ Z
D(t,u)L,R =
s
(t, u) m2
(49)
fL,R
with s = (pe + pe+ )2 , t = (pe p 0 )2 and u = (pe p 0 )2 representing the Mandeli
stam variables for neutralino pair production in e+ e collisions. Finally, the matrices Zij ,
gLij and gRij are products of the neutralino diagonalization matrix elements Nij5
5 5
5 5
Zij = Ni3
Nj 3 Ni4
Nj 4 /2,
f 5

2 2
f
5
f
gLij = I3 Ni2
cW + Qf I3 Ni1
sW If3 Nj52 cW + Qf I3 Nj51 sW /sW
cW ,
5 5 2
gRij = Q2f Ni1
Nj 1 /cW .
(50)
The e+ e annihilation cross sections follow from the squares of the bilinear charges,

2 1/2
e+ e i0 j0 = Sij
2s PS
1

2

1 2i 2j + PS cos2 Q1

1/2
+ 4i j Q2 + 2PS Q3 cos d cos ,
(51)
99
where Sij is a statistical factor: 1 for i = j and 1/2 for i = j ; i = m 0 / s, is the

i
polar angle of the produced neutrinos; and PS = PS (1, 2i , 2j ) denotes the familiar 2body phase space function PS (x, y, z) x 2 + y 2 + z2 2xy 2xz 2yz. The quartic
charges Qi (i = 1, 2, 3) are given by the bilinear charges as follows:

1
|QRR |2 + |QLL |2 + |QRL |2 + |QLR |2 ,
4

1
Q2 = e QRR QRL + QLL QLR ,
2

1
Q3 = |QRR |2 + |QLL |2 |QRL |2 |QLR |2 .
4
Q1 =
(52)
An example for the production of 50 in association with another singlino-type ( 50 ), or a

gaugino-type ( 10 ) or a higgsino-type ( 30 ) neutralino is presented in Fig. 4 for the parameter set P [Fig. 1] with meR = 200 GeV and meL = 250 GeV. [Of course the {55} final state
is unobservable without additional ISR emission.] The increase of the cross-sections
with increasing doubletsinglet gaugino/higgsino mixing parameterized by is obvious.
The gaugino character of 10 is responsible for the dominant size of the {51} cross-section.

With the anticipated integrated luminosity L = 1 ab1 , sufficiently large event rates
of order 103 are predicted if is not too small.
4.2. Decays to a singlino, with no Higgs bosons
(i) If kinematically allowed, two-body decays of neutralinos to the electroweak gauge
bosons Z are among the dominant channels. The widths of decays i0 j0 Z are given
by
(m2 m2 )2

1/2
0

i0
j0
g 2 PS 2
0
2
2
2

Zij
i j Z =
+ m 0 + m 0 2mZ
16m 0
i
j
m2Z
i

+ 6m 0 m 0 e Zij2 ,
i
(53)
where PS = PS (1, m2 0 /m2 0 , m2Z /m2 0 ), with Zij defined in Eq. (50). The widths of the
j
chargino 2-body decays into a neutralino and a W boson, i j0 W , read correspondingly

i j0 W

(m2 m2 )
1/2
i
j0
g 2 PS
|WLij |2 + |WRij |2
2
2
2
+ m + m 0 2mW
=
16m
2
i
j
m2W
i

6m m 0 e WLij WRij
(54)
i
100
Fig. 4. The production cross-sections of neutralino pairs, {51} (dashed), {55} (thin-solid), {53} (dotted) and {11}
(thick-solid), in e+ e collisions with the center-of-mass energy s = 500 GeV as a function of for the
parameter set P [Fig. 1] with meR = 200 GeV and meL = 250 GeV.
where PS = PS (1, m2 0 /m2 , m2W /m2 ) and the bilinear charges WL,R are defined as
j
Nj52 + ULi2
Nj53 ,
WLij = ULi1
2
WRij = URi1
Nj52 URi2
Nj54 .
2
(55)
The unitary matrices UL and UR diagonalize the chargino mass matrix as UR MC UL =

diag{m , m }, cf. Ref. [14] for details.
1
If 2-body decay channels are closed kinematically, the 3-body neutralino decays, i0
j0 f f, are generated by s-channel (virtual) Z exchange, and t- and u-channel sfermion
exchanges. Neglecting fermion masses, the transition matrix element, cf. Ref. [15], is determined by the bilinear charges Q which are related to the bilinear charges Q introduced
for the production, by crossing symmetry as
Q = Q
(56)
with the transformed Mandelstam variables, s = (pf + pf )2 , t = (p 0 + pf )2 and

j
u = (p 0 + pf )2 for the decays. [Neutralino decays to charginos and W bosons can be dej
scribed in the same way after obvious redefinitions of the bilinear charges.] Decay widths
and distributions depend on the quartic charges Q 1 , Q 2 and Q 3 defined analogously to
Eq. (52).
101
Fig. 5. The widths, lifetimes and flight distances [broken lines] of the decays 10 50 l + l and lR 50 l
as a function of for the parameter set P [Fig. 1]. The mass of the right-handed slepton is taken to be
ml = 200 GeV > m 0 for the 3-body neutralino decays (upper panels) and to be ml = 130 GeV < m 0
R
for the 2-body slepton decays (lower panels). The masses of the squarks are assumed to be mqL = 250 GeV
and mqR = 200 GeV. Right: flight distances for s = 500 GeV are shown by broken lines. The kink in the 10
lifetime and flight distance in the upper right panel is caused by accidental cancellations between sfermion and
Z exchange diagrams. [The value of the lower bound, expected from cosmological arguments on is presently
not yet known.]
(ii) At the LHC, cascade sfermion decays, f f i0 , are of great experimental interest.
The width of the sfermion 2-body decay into a fermion and a neutralino follows from
g 2 1/2

PS
|gfi |2 m2f m2 0 m2f ,
f f i0 =
16mf
i
(57)
where the 2-phase space function PS = PS (1, m2 0 /m2 , m2f /m2 ) with f = fL or fR ; the
i
couplings are expressed in terms of the neutralino mixing matrix N 5 as

f 5

f
5
5
gfL i = 2 I3 Ni2
and gfR i = 2Qf tW Ni1
+ Qf I3 Ni1
tW
(58)
102
in obvious notation.
The reverse decays, neutralino [chargino] decays to sfermions plus fermions, i0 ff
etc, are given by the corresponding partial widths,
1/2

g 2 PS
i0 ff =
|gfi |2 m2 0 + m2f m2f
32m 0
i
(59)
with the same couplings as before and PS = PS (1, m2 /m2 0 , m2f /m2 0 ). [Analogous exf
pressions hold for chargino decays.]

Examples of these partial decay widths are shown in Fig. 5 [with the parameter set P
as in Fig. 1]. For an illustrative purpose, the mass of the R-sleptons lR is assumed to be
mlR = 200 GeV > m 0 for the 3-body neutralino decays and to be mlR = 130 GeV < m 0
1
for the 2-body slepton decays, respectively.4 The masses of the squarks are assumed to be
mqL = 250 GeV and mqR = 200 GeV. For small mixing the lifetimes of the second
lightest neutralino 10 and the R-sleptons lR can be quite large, giving rise potentially to
macroscopic flight paths [9]. However, cosmological bounds on must be analyzed before any (realistic) experimental conclusions can be drawn. The kink in the 10 lifetime
and flight distance in the upper right panel of Fig. 5 is caused by accidental cancellations
between sfermion and Z exchange diagrams in the decays 10 50 q q and 50 ; these accidental cancellations do not occur [to any significant degree] in the decay 10 50 l + l .
4.3. Decays to a singlino, involving Higgs bosons
Decays involving Higgs bosons can be quite different for different Higgs boson mass
spectra. Following the procedure outlined in Ref. [4] we decompose the neutral Higgs
states into real and imaginary parts as follows:
1
Hd0 = [vd S1 s + S2 c + iP1 s ],
(60)
2
1
Hu0 = [vu + S1 c + S2 s + iP1 c ],
(61)
2
1
S = [vs + S3 + iP2 ],
(62)
2
where the Goldstone states are removed by using the unitary gauge. We then further rotate
these states onto the mass eigenstates, Hi (i = 1, . . . , 3) and Ai (i = 1, 2) labeled in order
of ascending mass, by using the orthogonal rotation matrices5 O H and O A :
Hi = Sj OjHi ,
Ai = Pj OjAi .
(63)
4 In either case 0 or l is the next-to-lightest SUSY particle NLSP with just one decay channel open to the
R
1
lightest SUSY particle LSP = 50 .
5 Note that the definitions of these mixings matrices differ slightly from those in Ref. [4], where the scalar
rotation matrix is defined via O T = O H and the pseudoscalar rotation is defined by a rotation through an angle
A .
103
Fig. 6. The Higgs boson mass spectrum as a function of for the parameter set P [Fig. 1] and maximal mixing.
For the purposes of example, the Higgs mass parameter MA is set to 2/ sin 2. Heavy scalar, pseudoscalar and
charged states are nearly mass degenerate: MH3 MA2 MH MA .
The resulting mass spectrum, composed of three scalars, two pseudocalars, and two
charged Higgs bosons, is shown in Fig. 6 as a function of . For the purposes of example,
we have chosen the mass parameter MA (defined to be the heavy pseudoscalar mass in the
MSSM limit) to be 2/ sin 2 567 GeV, setting the scale of the heavy Higgs bosons.
The lighter Higgs bosons consist of two scalars and one pseudoscalar. The lightest scalar
and pseudoscalar in our example are predominantly singlet states, with masses set by the
scale of .
Generally, the width of a 2-body neutralino or chargino i decay to a neutralino or
chargino j and a Higgs boson k (Hk or Ak ) is given by
1/2
2
2

PS 2
mi + m2j m2k CijL k + CijRk
[ i j k ] =
16mi

+ 2 mi mj CijL k CijRk + CijLk CijRk ,
(64)
L/R
where PS = PS (1, m2j /m2i , m2k /m2i ) and the left/right couplings Cij k must be specified in each individual case; = 1 for = Hk , H , and 1 for = Ak .
(i) For the decay of a neutralino i0 to a neutralino j0 and a scalar Higgs boson Hk ,
i0 j0 Hk , the couplings are given by,
104
CijRk i0 j0 Hk
H
1 5
5
5
= g Ni2
Ni1
tW Nj53 s + Nj54 c + 2 Ni3
c Ni4
s Nj55 O1k
2
H
1 5
5
5
+ g Ni2
Ni1
tW Nj53 c + Nj54 s + 2 Ni3
s + Ni4
c Nj55 O2k
2
H
1 5 5
5 5
+ Ni3
Nj 4 Ni5
Nj 5 O3k
+ (i j ),
(65)
2

CijL k i0 j0 Hk = CijRk i0 j0 Hk ,
(66)
While the first term in each of the two square brackets in Eq. (65) are reminiscent of the
MSSM couplings i0 j0 h and i0 j0 H , respectively, the other terms are genuinely new in
origin, arising from the extra interaction terms in the NMSSM superpotential.
0 H
The widths for the kinematically allowed decays 40 1,5
1,2 are shown in
0
Fig. 7(left) as a function of . For = 0 the 5 state is decoupled from the other
neutralinos; as is switched on, the coupling, and therefore the decay widths, increase.
The decay widths for 40 50 H1 and 40 50 H2 are comparable, within an order of
magnitude, due to the large 40 mass and the near mass degeneracy of H1 and H2 . With
partial widths of order GeV, these decay modes are in the observable range of branching
ratios.
(ii) Similarly, a 2-body neutralino decay to a neutralino and a pseudoscalar Higgs boson, i0 j0 Ak , follows Eq. (64) with the left/right couplings given by

CijRk i0 j0 Ak
A
1 5
5
5
= g Ni2
Ni1
tW Nj53 s + Nj54 c + 2 Ni3
c + Ni4
s Nj55 O1k
2
A
1 5 5
5 5
+ Ni3
Nj 4 Ni5
Nj 5 O2k
+ (i j ),
(67)
2

CijL k i0 j0 Ak = CijRk i0 j0 Ak .
(68)
Again, only the first term in the square brackets is similar to the MSSM coupling i0 j0 A.
0 A are shown in Fig. 7(right)
The widths for the kinematically allowed decays 40 1,5
1
as a function of for our chosen example scenario. In comparison with the scalar case,
many of the decays are kinematically disallowed, only leaving the decays of the heaviest
two neutralinos to 50 and the lightest pseudoscalar (A1 ). Note that pseudoscalar decays
are strongly suppressed compared with the scalar modes and may not be observed easily.
(iii) For completeness, we describe the decays of charginos to a neutralino and charged
Higgs boson i j0 H (i = 1, 2; j = 1, . . . , 5). These follow a similar pattern, now
with the last index of the coupling removed:

1 5
L
0
5
5
5
Cij i j H = gc Ni4
ULj 1 + Ni2
+ Ni1
tW ULj
2 s Ni5 ULj 2 ,
2
(69)

5
5
5
5
CijR i j0 H = gs Ni3
URj 1 Ni2
+ Ni1
tW URj
2 c Ni5 URj 2 .
2
(70)
105
0 H
0
0
Fig. 7. The decay widths for 40 1,5
1,2 (left) and 4 1,5 A1 (right) for the parameter set P[Fig. 1] and
maximal mixing. For the purposes of example, the Higgs mass parameter MA is set to 2/ sin 2.
However, the large mass of the charged Higgs boson means that these 2-body decays are
kinematically disallowed for our specific parameter choice.
(iv) It is also possible for Higgs bosons themselves to decay into the singlino-dominated
state, via the decays Hi 50 j0 , Ai 50 j0 and H 50 i , if kinematically allowed.
Clearly this is only possible for the heavier Higgs states; the lightest Higgs boson is never
heavy enough to decay in this way. The general form of the width for these decays i
j k (i = Hi , Ai , H ), is given by the crossing of Eq. (64):
2
2

PS 2
[i j k ] = Sj k
mi m2j m2k CijL k + CijRk
16mi

2 mj mk CijL k CijRk + CijLk CijRk ,
1/2
(71)
where PS = PS (1, m2j /m2i , m2j /m2i ) and Sj k = 1 or 1/2 is the usual statistical factor.
Again, = 1 for = Hk , H , and 1 for = Ak . The couplings Cij k are related to
their neutralino decay counterparts in the obvious way:
L/R
L/R
L/R
Cij k Hi j0 k0 = Ckij j0 k0 Hi ,
L/R
L/R
Cij k Ai j0 k0 = Ckij j0 k0 Ai ,
L/R
L/R
Cij H i0 j = Cij i j0 H .
(72)
(73)
(74)
Some of these decays widths are plotted in Fig. 8. Note that a significant fraction of the
Higgs boson H3 and A2 decays go into the invisible channel 50 50 only if the partial decay
width exceeds the range of 1/10 GeV. The upper left panel, showing the partial width
for the decay H3 i0 50 [i = 1, 5], has been allowed to extend down to widths of order
106
0 0 (upper left), A 0 0 (upper right) and H 0 (lower)

Fig. 8. The decay widths for H3 1,5
2
5
1,5 5
1 1,5
for the parameter set P [Fig. 1] and maximal mixing. For the purposes of example, the Higgs mass parameter MA
is set to 2/ sin 2.
105 GeV to show the switching off of the 10 50 H3 coupling at 58 GeV. This is caused
by destructive interference between the different constituent fields in both the Higgs and
the neutralinos, and is directly analogous to the cancellations seen in Ref. [4].
5. Summary and conclusions

In this study, we have investigated the neutralino sector of the NMSSM, suggested
by many GUT and superstring models. Moreover, this model attempts to explain the the scalar
problem of the MSSM by introducing a new iso-singlet Higgs superfield, S,
component of which acquires a non-zero vacuum expectation value.
We have given expressions for the new 5 5 neutralino mass matrices and mixing
matrices and we have presented, besides the numerical analyses, approximate analytical
solutions for the neutralino masses and mixings which provide a nice insight into the
structure of the spectrum and the mass hierarchies in case of small couplings between
the MSSM and the new iso-singlet.
107
The renormalization group flow of the parameters and from the GUT scale down to
the electroweak scale gives rise to strong upper bounds on their values at the electroweak
scale, where small is favored. The qualitative features of the neutralino masses are dependent on how strongly the PQ symmetry of the model is broken by non-zero values;
this is quite accurately described by the approximate analytical solutions.
If the PQ symmetry is slightly broken for small , the qualitative pattern for the particle
spectrum remains intact, except that the lightest singlino-dominant neutralino acquires a
mass of the order of the electroweak scale. Thus the model contains four MSSM-type
heavy gaugino/higgsino dominant states and one light singlino dominant state. Since the
couplings to the Z boson can be very much reduced, the NMSSM with a slightly broken
PQ symmetry constitutes a valid scenario.
In contrast, a strongly broken PQ symmetry, though disfavored by the flow of the
couplings from the GUT scale down to the electroweak scale, could provide an extra moderately heavy neutralino state, which is only weakly coupled to the Z and (s)fermions. Such
decoupled scenarios would be more difficult to distinguish the NMSSM from the MSSM.
The methods described for real mass matrices, can readily be generalized to complex
5 5 neutralino mass matrices. In analogy to the MSSM, cf. Ref. [13], the procedure is
mathematically based upon the Takagi theorem [16] which assures that any complex symmetric matrix can be diagonalized by an orthogonal transformation. The technical details
are outlined in an addendum to the Appendix A. The system can then be solved analytically by following the same steps of approximations which have been applied to the real
mass matrices.
Acknowledgements
The work of S.Y.C. was supported in part by the Korea Research Foundation Grant
(KRF-2002-070-C00022) and in part by KOSEF through CHEP at Kyungpook National
University.
Appendix A. The small-mixing approximation

The 5 5 neutralino mass matrix of Eq. (4) in general cannot be diagonalized analytically to derive the physical neutralino masses. However, simple analytical expressions for
masses and mixing parameters can be found by making use of approximations for small
doubletsinglet mixing which is theoretically very well motivated.
To construct this approximate solution in the neutralino sector, we treat the doublet
singlet mixing parameter Mi , , together with the Z-boson mass mZ , as small
parameters of generic size 1 in units of the typical SUSY masses. Then, as long as
these SUSY masses are not as small as , we observe a hierarchical structure in the neutralino mass matrix of the form:

A X
,
H=
(A.1)
XT B
108
where A is a 4 4 matrix incorporating elements of the order of the large SUSY scale, B
is a scalar and X is a 4-component vector of order .
Performing an auxiliary orthogonal transformation O defined by the matrix,6

144 12 T
O=
(A.2)
T
1 12 T
with the mixing column vector = [A B]1 X, the mass matrix takes block diagonal
form, accurate to order 2 :

0
A + A
T
,
OHO =
(A.3)
0
B + B
where
A =

1
[A B]1 , XX T ,
2
B = X T [A B]1 X.
(A.4)
Both s are of order 2 . If A is diagonal, only the diagonal elements of A need be kept as
re-diagonalization would change the mass matrix (A.16) and the orthogonal matrix (A.2)
only beyond the order considered in the systematic expansion. We note that the correction
terms satisfy the simple sum rule Tr A + B = 0.
If B is also as small as the elements of the low vector X, the mixing column vector
= A1 X and the correction terms A and B are further simplified to be

1 1
(A.5)
A , XX T
and B = X T A1 X.
2
On the contrary, if B is much larger than the other parameters, the mixing column vector
= X/B and the correction terms take the following simple form
A =
A = XX T /B,
B = X T X/B.
(A.6)
Both these approximations have been used in the derivation of all the mass and mixing
formulae discussed earlier in the report.
The diagonalization of the mass matrix M5 ,

M X
M5 =
(A.7)
5
XT m
with M being the 4 4 MSSM mass submatrix, makes use of the block-diagonalization
method in the following way:
(1) In the first step M is diagonalized by applying the well-elaborated MSSM procedure
MD = V MV T
(A.8)
generating the eigenvalues MD = diag[m

1, . . . , m
4 ].
6 Note that by standard notation T is a 4 4 matrix with the elements while T is a scalar with
i j
!
the value i2 .
109
(2) The ensuing 5 5 matrix can subsequently be block-diagonalized as worked out

above by applying the orthogonal transformation in Eq. (A.2) with

1
5 V X = (M m
5 )1 X.
= V : = V T MD m
(A.9)
Note that is of order quantum mechanically enhanced however if mass differences
|m
i m
5 | are, accidentally, small.
(3) The block-diagonalization affects the upper left diagonalized 4 4 submatrix MD
only beyond second order and likewise the orthogonal matrix V 5 beyond the second and
first order considered, respectively, for on- and off-diagonal elements. As a result, we obtain the final diagonal form of the mass matrix as
5
5T
MD
5 V M5 V
with

V5
(A.10)
144 12 (V )(V )T
(V )T
(V )
1
1 2 (V )T (V )
V
0
0
111

(A.11)
in obvious notation. While the right-most part solves the MSSM diagonalization, the leftmost part diagonalizes the NMSSM under the assumption of small doubletsinglet mixing.
Addendum
This procedure of diagonalizing the neutralino mass matrix can readily be generalized
to complex mass matrices when the gaugino and higgsino matrix elements are no longer
assumed to be real.
The mathematical key to this generalization is the Takagi theorem [16] which assures
that any complex symmetric matrix can be diagonalized by an orthogonal transformation
of the form
N HN = diag(mi ),
mi = real and positive
(A.12)
with the unitary matrix N acting on the neutralino Majorana fields. This is widely used for
complex extensions of the MSSM mass matrix, cf. Ref. [13].
The matrix N can be decomposed, N = MU , into the unitary matrix U blockdiagonalizing the hermitian matrix H H and the diagonal phase matrix M. Writing

A2 X2
H H =
(A.13)
X2 B2
with A2 = A A + X X T , B2 = B B + X X and X2 = A X + X B, the vector 2 in

2
144 12 2 2
U=
(A.14)
2
1 12 2 2
is defined, in analogy to (A.2), by
2 = [A2 B2 ]1 X2 .
(A.15)
110
The matrix elements of the block-diagonalized matrix can easily be derived from

0
A2 + A2
,
U H HU =
(A.16)
0
B2 + B2
where

1
[A2 B2 ]1 , X2 X2 ,
2
The matrix M is built up by the phases

e2ii = U HU ii /mi .
A2 =
B2 = X2 [A2 B2 ]1 X2 .
(A.17)
(A.18)
From this point on, the same steps can be followed as for the real mass matrix.
References
[1] P. Fayet, Phys. Lett. B 64 (1976) 159.
[2] H.P. Nilles, Phys. Rep. 110 (1984) 1;
H.E. Haber, G.L. Kane, Phys. Rep. 117 (1985) 75.
[3] P. Fayet, Nucl. Phys. B 90 (1975) 104;
M. Drees, Int. J. Mod. Phys. A 4 (1989) 3635;
J. Ellis, J.F. Gunion, H. Haber, L. Roszkowski, F. Zwirner, Phys. Rev. D 39 (1989) 844, and other references
quoted therein.
[4] D.J. Miller, R. Nevzorov, P.M. Zerwas, Nucl. Phys. B 681 (2004) 3, hep-ph/0304049.
[5] U. Ellwanger, J.F. Gunion, C. Hugonie, NMHDECAY, hep-ph/0406215.
[6] S.A. Abel, S. Sarkar, P.L. White, Nucl. Phys. B 454 (1995) 663;
S.A. Abel, Nucl. Phys. B 480 (1996) 55;
C. Panagiotakopoulos, K. Tamvakis, Phys. Lett. B 446 (1999) 224;
C. Panagiotakopoulos, K. Tamvakis, Phys. Lett. B 469 (1999) 145;
A. Dedes, C. Hugonie, S. Moretti, K. Tamvakis, Phys. Rev. D 63 (2001) 055009;
C. Panagiotakopoulos, A. Pilaftsis, Phys. Rev. D 63 (2001) 055003.
[7] M. Dine, W. Fischler, M. Srednicki, Phys. Lett. B 104 (1981) 199;
H.P. Nilles, M. Srednicki, D. Wyler, Phys. Lett. B 120 (1983) 346;
J.M. Frere, D.R. Jones, S. Raby, Nucl. Phys. B 222 (1983) 11;
J.P. Derendinger, C.A. Savoy, Nucl. Phys. B 237 (1984) 307;
A.I. Veselov, M.I. Vysotsky, K.A. Ter-Martirosian, Sov. Phys. JETP 63 (1986) 489, Zh. Eksp. Teor. Fiz. 90
(1986) 838 (in Russian);
R.B. Nevzorov, M.A. Trusov, J. Exp. Theor. Phys. 91 (2000) 1079, Zh. Eksp. Theor. Fiz. 91 (2000) 1251
(in Russian), hep-ph/0106351.
[8] F. Franke, H. Fraas, A. Bartl, Phys. Lett. B 336 (1994) 415;
U. Ellwanger, M. Rausch de Traubenberg, C.A. Savoy, Phys. Lett. B 315 (1993) 331;
U. Ellwanger, M. Rausch de Traubenberg, C.A. Savoy, Nucl. Phys. B 492 (1997) 21;
S.F. King, P.L. White, Phys. Rev. D 52 (1995) 4183;
F. Franke, H. Fraas, Z. Phys. C 72 (1996) 309;
F. Franke, H. Fraas, Int. J. Mod. Phys. A 12 (1997) 479;
B. Ananthanarayan, P.N. Pandita, Int. J. Mod. Phys. A 12 (1997) 2321;
S.P. Martin, Phys. Rev. D 62 (2000) 095008;
M. Bastero-Gil, C. Hugonie, S.F. King, D.P. Roy, S. Vempati, Phys. Lett. B 489 (2000) 359;
U. Ellwanger, C. Hugonie, Eur. Phys. J. C 25 (2002) 297;
F. Franke, S. Hesselbach, Phys. Lett. B 526 (2002) 370;
U. Ellwanger, J.F. Gunion, C. Hugonie, S. Moretti, hep-ph/0305109.
111
[9] U. Ellwanger, C. Hugonie, Eur. Phys. J. C 5 (1998) 723;

U. Ellwanger, C. Hugonie, Eur. Phys. J. C 13 (2000) 681.
[10] ATLAS Technical Proposal, CERN/LHCC/94-43, LHCC/P2 (1994);
CMS Technical Proposal, CERN/LHCC/94-38, LHCC/P1 (1994).
[11] TESLA Collaboration, in: R.D. Heuer, D.J. Miller, F. Richard, P. Zerwas (Eds.), Technical Design Report
(Part 3), DESY 01-011, 2001, hep-ph/0106315;
American LC Working Group, T. Abe, et al., SLAC-R-570, hep-ex/0106055;
ACFA LC Working Group, T. Abe, et al., KEK-REPORT-2001-11, hep-ex/0109166;
CLIC Study Team, R.W. Assman, et al., CERN-2000-008.
[12] A. Stephan, Phys. Lett. B 411 (1997) 97;
A. Stephan, Phys. Rev. D 58 (1998) 035011;
B.A. Dobrescu, K.T. Matchev, JHEP 0009 (2000) 031;
D.J. Miller, R. Nevzorov, hep-ph/0309143;
A. Menon, D.E. Morrissey, C.E.M. Wagner, hep-ph/0404184.
[13] S.Y. Choi, J. Kalinowski, G. Moortgat-Pick, P.M. Zerwas, Eur. Phys. J. C 22 (2001) 563;
S.Y. Choi, J. Kalinowski, G. Moortgat-Pick, P.M. Zerwas, Eur. Phys. J. C 23 (2002) 769.
[14] S.Y. Choi, A. Djouadi, M. Guchait, J. Kalinowski, H.S. Song, P.M. Zerwas, Eur. Phys. J. C 14 (2000) 535.
[15] S.Y. Choi, H.S. Song, W.Y. Song, Phys. Rev. D 61 (2000) 075004.
[16] T. Takagi, Jpn. J. Math. 1 (1925) 83.
Hierarchically split supersymmetry with

FayetIliopoulos D-terms in string theory
Boris Krs a , Pran Nath b
a Center for Theoretical Physics, Laboratory for Nuclear Science and Department of Physics,
Massachusetts Institute of Technology, Cambridge, MA 02139, USA

b Department of Physics, Northeastern University, Boston, MA 02115, USA
Received 29 November 2004; accepted 19 January 2005
Abstract
We show that in string theory or supergravity with supersymmetry breaking through combined
F-terms and FayetIliopoulos D-terms the masses for charged scalars and fermions can be hierarchically split. The mass scale for the gauginos and higgsinos of the MSSM is controlled by the gravitino
mass m3/2 , as usual, while the scalars get extra contributions from the D-terms of extra Abelian
U (1) factors, which canmake them much heavier. The vanishing of the vacuum energy requires that
their masses lie below m3/2 MPl , which for m3/2 = O(TeV) sets a bound of 101013 GeV. Thus,
scalars with non-vanishing U (1) charges typically become heavy, while others remain light, producing a spectrum of scalars with masses proportional to their charges, and therefore non-universal.
This is a modification of the split supersymmetry scenario, but with a light gravitino. We discuss how
FayetIliopoulos terms of this size can arise in orientifold string compactifications with D-branes.
Furthermore, within the frame work of D-term inflation, the same vacuum energy that generates the
heavy scalar masses can be responsible for driving cosmological inflation.
PACS: 12.60.Jv; 04.65.+e; 11.25.Mj
E-mail addresses: kors@lns.mit.edu (B. Krs), nath@neu.edu (P. Nath).

doi:10.1016/j.nuclphysb.2005.01.030
B. Krs, P. Nath / Nuclear Physics B 711 (2005) 112132
113
1. Introduction
The conventional approach to supersymmetry breaking in models of supergravity
(SUGRA) is to assume some form of spontaneous breaking in a hidden sector, mediated
to the visible sector with contains the MSSM via gravitational interactions [1]. The overall
mass scale is then set by the gravitino mass m3/2 , and all other masses for squarks, sleptons,
gauginos and higgsinos come out roughly proportional to it, by demanding the cancellation
of the vacuum energy. Low energy supersymmetry, as required by a natural explanation of
the Higgs potential, fixes m3/2 to the electro-weak scale. Now, recently it was argued [2]
that the fine tuning problem of the Higgs mass may be insignificant compared to the fine
tuning of the cosmological constant, and that an anthropic selection mechanism (see, e.g.,
[3]) may then involve actual fine tuning of MSSM parameters. The mass pattern that was
proposed under the name of split supersymmetry [4] has all the MSSM fermions at the
electro-weak scale, whereas all scalars, except for the one fine tuned Higgs doublet, get
ultra-heavy at a high mass scale. This scenario has attracted some attention recently [5].
More concretely, the challenge for model building is to keep the gauginos and higgsinos
light, while letting the scalars become very heavy. The motivation for this originates from
supersymmetric grand unification, even without low energy supersymmetry in the usual
sense, and the model is designed to keep the merits of gauge coupling unification as in the
MSSM. Here, we pursue the perspectives of such patterns in the mass spectrum in the context of SUGRA and string theory models, based on the paradigm of spontaneous breaking
in a hidden sector.
Given the above mentioned relations that govern gravity mediated supersymmetry
breaking, it seems very hard to achieve hierarchically split mass scales. If all masses are
proportional to m3/2 , there is no room for flexibility. There is, however, a loophole to the
argument, which has not so far been explored in the conventional approach in any depth,
probably because it quickly leads to large masses, which were thought unacceptable. It
consists of assuming not only auxiliary F-terms but also D-terms to be generated. In global
supersymmetry, this is well-known under the label of supersymmetry breaking mediated
by an anomalous U (1), where large masses can be avoided [6,7], as will be seen later. The
mechanism basically adds a FayetIliopoulos (FI) term for an extra anomalous U (1) gauge
symmetry, independent of the F-terms which may also be present. In local supergravity, the
two contribute both to the vacuum energy, and thus get tied together at roughly the same
scale. Still, the contribution to scalar masses can be very different, since the F-terms are
mediated via gravity, while the D-terms are mediated via gauge interactions, which opens
up the possibility to have hierarchically split mass scales, splitting scalars charged under
the relevant U (1), from all other fields, i.e., gauginos, higgsinos as well as scalars not
charged under the relevant U (1)s. This is not quite along the lines of high scale supersymmetry breaking, as advocated in split supersymmetry, where the gravitino mass itself was
assumed at the high scale, and the main difficulty lies in keeping the gauginos and higgsinos lighter than m3/2 .1 Instead, we propose extra contributions to the masses of charged
1 To achieve this, it is usually assumed that gravity mediation of gaugino masses can be avoided first of all.
Furthermore, one has to find ways to suppress contributions from anomaly mediation [8]. Since the latter is not
114
scalars, which make them heavier than m3/2 , while the fermions including the gravitino
remain light.
The purpose of this paper is to study the confluence of this combined approach with
F- and D-terms in supergravity and string theory models. By this we mean that we assume
that some hidden sector dynamics generates F- and D-terms at some scale of supersymmetry breaking, but we do not present a full dynamical model, how this happens. Instead,
we analyze the various scenarios that can emerge in the visible sector, and in particular
identify classes of models which generically lead to hierarchically split mass scales.
1.1. FI-terms in global supersymmetry
In many models of grand unification, compactification of higher-dimensional supergravity or string theory, the minimal gauge symmetries that can be achieved at low energies
involve various extra Abelian U (1) gauge factors beyond the Standard Model gauge group,
i.e., the total gauge symmetry is SU(3)C SU(2)L U (1)n , where among the U (1) there
is also the hypercharge. In string theory it often happens that some of the extra factors are
actually anomalous, the anomaly being canceled by a (generalized) GreenSchwarz (GS)
mechanism, in which the gauge boson develops a Stueckelberg mass and decouples (see,
e.g., [10]). In any case, it is then permissible to add FI-terms a Da to the supersymmetric
Lagrangian, one for each U (1)a . The D-term potential in global supersymmetry is
2
g2
g2
a 2
a
i 2
D =
Qa |fi | + a
VD =
(1)
2 a
2
a
a
i
and thus a > 0 leads to the formation of a condensate for at least some field fi of negative
charge Qia < 0. This breaks the gauge symmetry spontaneously, but supersymmetry can be
restored at the minimum if Da = 0.2 In the MSSM it is usually assumed that the FI-term
of the hypercharge is absent or very small, and does not play a role in the Higgs potential.
Whenever the auxiliary field obtains a non-vanishing expectation value Da = 0, supersymmetry is broken, and mass terms are generated for all the charged scalars,

m2i =
(2)
ga2 Qia Da ,
a
where it is now assumed that the charges are positive for the MSSM fields, to avoid breaking of the Standard Model gauge symmetries. This scenario can be achieved in a global
supersymmetric model with a single extra U (1)X by adding two scalars to the MSSM,
singlets under SU(3)c SU(2)L U (1)Y , but with charges 1 under U (1)X [6]. The
crucial ingredient is an interaction in the superpotential of the form
W = m + .
(3)
fully understood within string theory (see [9]) and we include the effects of gravity mediation anyway, we will
not consider anomaly mediation in the following.
2 In string theory, the FI-parameter is usually a function of the moduli, = (T , T ). Therefore, turning on
a
a I I
the FI-term can correspond to a flat direction |fi |2 = a in the total potential, for Qia = 1.
115
Minimizing the full potential

2 2 g 2
V = m2 + + + X
2
2 2
QiX |fi |2 + + + X
2
(4)
drives the fields to

+ = 0,
2
m2
= X 2 ,
gX
(5)
and
DX =
m2
,
2
gX

F + = m X + .
(6)
Gaugino masses may originate from higher-dimensional operators, and are suppressed by
powers of MPl ,3
m

X
1
F + + F + m 2 .
2
MPl
MPl
(7)
2 ) on gets masses at the electro-weak scale. DeAssuming m O(TeV) and O(MPl

pending on the precise scale and the charges of the MSSM scalars under the extra U (1)X ,
these contributions to their masses can be very important in the soft breaking Lagrangian.
A central point to notice here is the fact that the masses that follow from the FI-terms are
directly proportional to the expectation values of the auxiliary Da fields, they are mediated
by the anomalous U (1)X , whereas the masses induced via the F-terms are suppressed by
MPl through their mediation by gravity. The function of the extra fields lies in absorbing the potentially large FI parameter a , such that Da 1/2 O(TeV), consistent with the
standard scenario of superpartner masses at the electro-weak scale.
We will discuss this type of model and its modifications in the frame work of supergravity and string theory. However, before getting into the details of the extended model, we
discuss how such FI-terms arise in string theory.
1.2. A single anomalous U (1) and the heterotic string

The four-dimensional GS mechanism consists of the cancellation of the anomalies of
the usual one-loop triangle diagrams with tree-level exchange of an axionic scalar . This
refers to the mixed Abelian-gravitational and Abeliannon-Abelian anomalies at the same
time. The relevant terms in the action are usually written in terms of the Hodge-dual 2-form
B , related to by B , as
cA B 3
X
+ mX B F
,
(8)
where cA and mX are two coupling constants, 3 the ChernSimons (CS) 3-form, and
F X the gauge field strength of the relevant U (1)X . Now, is the imaginary part of some
3 The Planck mass M is defined so that M = 1 = (8 G)1/2 = 2.4 1018 GeV.
Pl
Pl
116
complex scalar in a chiral multiplet. For the heterotic string, the only such scalar that
participates in the GS mechanism is the dilaton-axion field S|= =0 = s + i (see [11]
= ln(S + S).
for an overview). Its action is described by the Khler potential K(S, S)

Since Eq. (8) implies a non-linear gauge transformation X S mX X under the U (1)X ,
ln(S +
the gauge invariance demands a redefinition of the Khler potential ln(S + S)
S mX VX ), where VX is the vector multiplet superfield. The Lagrangian then involves a

Stueckelberg mass term with mass proportional to mX for the gauge boson of this U (1)X ,
which absorbs the scalar as its longitudinal component. In addition, an FI-term X DX is
present, with 2 X mX /s. For the heterotic string, this FI-term is generated at one-loop
and the coefficient reads [12]
2 X
g 2 tr(QX )
mX
X
.
s
192 2
(9)
In the presence of an interaction (3) the scalar fields fi charged under the U (1)X acquire
masses given by
m2i = QiX m2 O(TeV).
(10)
The remarkable feature of Eq. (10) is that it is independent of the FI-parameter. Thus, the
vector boson gets a mass of the order of mX , which is close to the Planck scale, whereas
the charged sfermions and gauginos remain massless at the high scale, and get masses
of the order of the electro-weak scale. This is the standard scenario of supersymmetry
breaking via an anomalous U (1) with GS mechanism.
1.3. Multiple anomalous U (1) symmetries and D-branes
Orientifold string compactifications [13] usually involve more than one anomalous U (1)
factor. While in the heterotic string it is only the axionic partner of the dilaton that participates in the GS anomaly cancellation, now all the axionic scalars that follow from the
reduction of the RR forms from ten dimensions can do so [14]. In orientifold compactifications of type IIB strings, the relevant RR scalars originate from the twisted sectors.
The FI-parameters a are then functions of the expectation values of the real parts of these
twisted scalars, instead of the dilaton s. For a special example of this class of models,
in a toroidal orbifold T6 /Z3 , it was shown, that no FI-term was generated at one-loop,
consistent with the fact that the twisted scalars vanish at the orbifold point [15]. As another class of models, orientifolds with intersecting (type IIA) or magnetized (type IIB)
D-branes have been studied extensively in the recent past, most prominently for their very
attractive features to produce Standard Model or MSSM like gauge groups and spectra. For
these, untwisted RR scalars participate in the GS mechanism. Again the FI-term at treelevel (i.e., from a disc diagram, or the dimensional reduction of the BornInfeld action) is
proportional to the modulus that combines with the axionic scalar from the GS mechanism
into a complex scalar. The GS couplings analogous to Eq. (8) now involve many scalars

I
I
I
a
(11)
cA
B
3 +
mIa B
F
,
I,A
a,I
117
where I labels the scalars I , given by I B I , and a the anomalous U (1)a

I are labeled by A
factors with field strengths F a for the superfield Va . The constants cA
for the different anomalies, i.e., the different CS forms that can appear. We let TI |= =0 =
tI + iI be the complex scalars. Again Eq. (11) implies that the TI transform under U (1)a
whenever the coupling coefficient mIa = 0. Then the Khler coordinate TI is replaced in
the following way

mIa Va .
K(TI + TI ) K TI + TI
(12)
a
Depending on the precise form of the Khler potential, a FI-term will be generated from
this expression, that will depend on the vacuum expectation value of tI . The simplest expression would be

mIa tI .
2 a
(13)
I
It was stressed in [14] that the a given by Eq. (13) can in principle be of any size, as
opposed to the result for the heterotic string case. Another important observation is to note,
that the FI-terms are not necessarily tied to anomalous gauge symmetries, but only the
non-vanishing Stueckelberg coupling mIa = 0 has to exist.4
2. FI-terms in supergravity and string theory
We now examine the patterns of soft supersymmetry breaking that arise from an effective string theory Lagrangian with one or more FI-terms, motivated by the appearance of
multiple U (1) factors in orientifold models, that can develop FI-terms.
2.1. The vacuum energy
The degrees of freedom of the model are assumed to be given by the fields fi of the
MSSM, the extra gauge vector multiplets for the U (1)a , the moduli TI of the gravitational
sector, plus the axion-dilaton S, which includes the fields that participate in the GS mechanism, producing Stueckelberg masses for the gauge bosons and FI-terms. Furthermore, we
can add extra fields like the of the globally supersymmetric model, with charges a
under U (1)a . The effective scalar potential is given by the N = 1 supergravity formula
[16]
V = 4 eG GM N GM GN + 3 + VD ,
(14)
with

G = 2 K ln 6 W W ,
(15)
4 As mentioned earlier, we shall here not attempt to model the dynamics of the hidden sector in any detail,
and therefore also do not try to answer, how the FI-parameters are generated dynamically. Since they are modulidependent functions, a meaningful answer would have to address the moduli-stabilization at the same time.
118
where indices M, N run over all fields. We define the dilaton and moduli fractions of the
vacuum energy by
1
|S |2 = GS S GS GS ,
3
1
|I |2 = GI I GI GI ,
3
(16)
and | |2 in a similar fashion. It also turns out to be useful to introduce the following
combinations

2
y = .
m = e K/2 m,
x = + ,
m3/2 = 1 eG/2 ,
(17)
Similarly, all other fields are made dimensionless. Imposing the restriction on the model
that the vacuum energy vanishes (through fine tuning) one has
|S |2 +
|I |2 + |+ |2 + | |2 +
1
2
3m23/2 MPl
g2
a
Da2 = 1,
(18)
where Da = a | + |2 a | |2 + a . This implies an immediate bound on the expectation

values of the auxiliary fields FI = DI W = I W + 2 (I K)W and Da ,
FI m3/2 MPl ,
Da m3/2 MPl ,
(19)
where we ignore prefactors involving the Khler potential. As long as m3/2 O(TeV),
Eq. (19) implies roughly Da 1/2 101013 GeV, which is the usual intermediate supersymmetry breaking scale in SUGRA models. The masses that are generated by the F-terms
are given by FI /MPl O(TeV), whereas the D-terms would be able to produce much
1/2
larger mass terms proportional to Da . This means that the mass parameter in the superpotential (3), which had to be fine tuned
to the electro-weak scale, can now also be assumed
as large as the intermediate scale, m m3/2 MPl . Further, we note that the scenario with
a Planck scale sized FI-parameter, as is unavoidable for the heterotic string in the presence
of an anomalous U (1), is only consistent with a Planck scale sized gravitino mass. In orientifold D-brane models, as mentioned earlier, the FI-parameter can in principle be of any
value, and the problem does not occur.
In scenarios with split supersymmetry, the gravitino mass itself is not restricted to a
small value. However, gravity mediation generically leads to a contribution to gaugino and
higgsino masses which is proportional to the gravitino mass, and therefore m3/2 O(TeV)
is unavoidable in the present context. This then really puts an upper bound 101013 GeV
on the high mass scale allowed for the sleptons and squarks.
Before going into the various scenarios, let us first assemble a few general definitions
and formulas for the supergravity version of the model of supersymmetry breaking mediated by one or many anomalous U (1). For the Khler potential K we write
K = Khid (TI , TI ) + Ki (TI , TI )fi f + K+ (TI , TI ) + + + K (TI , TI ) ,
where we have now included S among the TI . The gauge kinetic functions are modulidependent,
fa = fa (TI ),
1
= (fa ),
ga2
(20)
119
but independent of . For the superpotential we assume the following factorized form

W = WMSSM (fi ) + Whid (TI ) + W + , = m + + W0 ,
(21)
where WMSSM contains the quark, lepton and Higgs fields, Whid contains the fields of
the hidden sector which break supersymmetry spontaneously by generating auxiliary field
components for FI = DI Whid , while W is still given by Eq. (3).5 With this, the total
D-term potential VD is given by
2
g2
2
2
a
MSSM
+
Qia Ki |fi |2 + a K+ + a K + a .
VD = VD
2
a
i
(22)
MSSM is the D-term arising from the SU(2) U (1) sector, which will not be
Here VD
L
Y
important. The standard expressions for the soft breaking terms that originate from the
F-terms only, are [17]
m =
1
F I I fa ,
2(fa )
(23)
for gaugino masses, and
m2gr,i = m23/2 F I F J I J ln(Ki ),
(24)
for scalar masses. For FI m3/2 MPl , both masses are of the order of m3/2 .
2.2. The simplest model
The simplest model that already displays the effects of the FI-terms is given by assuming
one or more FI-terms being generated by extra U (1)a gauge factors, and only including the
MSSM fields with arbitrary positive charges, but leaving out the extra fields .6 In that
case, supersymmetry is broken, and the D-terms are trivially given by
ga2 2 ga2 2
D = a .
2 a
2
Together with potential F-terms, they generate masses

m2i = m2gr,i +
ga2 Qia Ki a ,
(25)
(26)
a
5 This implies an assumption on the absence of any coupling among MSSM fields and in the superpotential, which may be problematic in the context of a concrete model. Furthermore, we also ignored any
cross-coupling in the Khler potential, where in principle the moduli-dependence of the various coefficients could
also involve .
6 It may sound very restrictive to allow only positive charges here, and it really would be in any reasonable
model derived from a GUT or string theory. However, there are well-known cases in string theory compactifications, where higher order corrections in the derivative expansion of the effective action (such as the BornInfeld
Lagrangian) lead to a lifting of tachyonic negative masses in the presence of FI-terms, even if some fields have
negative charge. We will come to explain this in some more detail later.
120
for all charged scalars. On the other hand, the masses of gauginos (and higgsinos) are unaffected by the D-term, at least at leading order, and would be dictated by the F-terms
to be of order m3/2 . In a scenario with m3/2 O(TeV), and Da 101013 GeV for at
least one FI-term, this provides a hierarchical split of energy scales, however, the high
scale cannot move up all the way to the Planck scale. The most simple charge assignment would give positive charges of order one to all MSSM fields, except the Higgs,
and thus the sfermion sector of the MSSM would become very heavy and undetectable
at LHC.7 There may of course also be other interesting patterns to consider, such as different charges for the three generations, different charges for different SU(5) multiplets,
which would leave the gauge coupling unification intact, or different charges for left- and
right-handed fields, which could be more easily realized in certain D-brane models from
string theory.
2.3. The full potential with a single U (1)X
The essence of the model of [6], where an anomalous U (1) with its FI-term is responsible for supersymmetry breaking, is the scalar condensate for that cancels the
contribution of the FI-parameter to the D-term, up to small electro-weak scale sized contribution, given the interaction (3) in the superpotential. Since such a condensate breaks
the gauge symmetry, one has to add extra chiral multiplets to the MSSM, neutral under the MSSM gauge symmetries. The model of [6] is within the framework of global
supersymmetry, and it is of obvious interest to study the embedding of this original model
into supergravity. So, we focus first on the case when there is just one extra U (1)X , which
develops a FI-term. The potential can be written as

2
4 V = 4 Vhid + e K 2 m2 |x|2 + |y|2 + 6 |W |2 |K+ x|2 + |K y|2 3

g2
2
) + X K+ |x|2 K |y|2 + X ,
+ 4 m(K+ + K )(xy W + x yW
2
(27)
where we also have replaced X 2 X , and defined

2
Vhid = e K KI J DI W DJ W ,
(28)
for the contribution of the hidden sector. It is essential for the fine tuning of the vacuum
energy. The two minimization conditions are x V = y V = 0. To keep things simple, we
now also set all relevant phases to zero, i.e., we treat x, y, W as real. Then one gets the
following two equations
m23/2 2
m2
2
K+ x 2 gX
DX + 4 VF + 2 x + xy 2 (K+ + K ) + 2 K+
x
MPl
MPl

m m3/2
y (K+ x)2 + (K y)2 + 2K 3 = 0,

+
2
MPl
7 This charge assignment would indeed render the U (1) anomalous.
m2
m23/2 2
2
K y 2 gX
DX + 4 VF + 2 y + x 2 y(K+ + K ) + 2 K
y
MPl
MPl

m m3/2
x (K+ x)2 + (K y)2 + 2K+ 3 = 0,

+
2
MPl
121
(29)
2 D 2 , where V and V
where VF and DX are defined so that V = VF + 12 gX
F
hid are related
X
by
2

2
2
VF = Vhid + m2 MPl
|x| + |y|2 + |m3/2 |2 MPl
|K+ x|2 + |K y|2 3

2
(K+ + K )(xy m
3/2 + x ym
3/2 ) .
+ m MPl
(30)
Note that the redefined mass parameters are field-dependent. In the following, we also
restrict to canonically normalized fields, and set K+ = K = 1. We have illustrated the
full potential in Fig. 1, using values
4 Vhid = 3,
gX = 200,
2
6 e K |W0 |2 = 1,
2
2 e K m2 = 1,
2 X = 101 .
2 m2 ,
This corresponds to setting some parameters equal, dividing the total potential by MPl
3/2
and rescaling the Planck mass by many order of magnitude, so that MPl /m3/2 = 10, just
to suppress the very steep part of the potential. It turns out, that for the relevant range of
parameters x = 0 = y at the minimum.
Since the minimization of the F-term potential basically consists of balancing terms
that scale like m2 | |2 , m23/2 | |2 or with inverse powers of MPl , with the negative term
2 , it is intuitively clear that x and y will be roughly bounded through the most
m23/2 MPl
2
dangerous term by m3/2 m1
or O(1). This is of course only valid as long as | |
2
X , otherwise | | gets tied to X . In any case one has that F /MPl m3/2 , which is
sufficient for light gauginos.
Fig. 1. Rescaled potential V(x, y).
122
2.4. Limiting cases: m3/2 = 0 or m = 0

We consider now some limiting cases for the above. To see how the model of [6]
emerges, we take the flat limit by setting
m3/2 = 0.
(31)
The two minimization conditions reduce to equations homogeneous in x, y, respectively.

One can convince oneself that x = 0 is stable. The solution then reduces to the known case
of global supersymmetry, where

m2
2m2
x = 0,
y 2 = X + 1 1 + 2 2 X 2 2 .
(32)
gX MPl
gX MPl
The auxiliary fields are also identical to (6), since DI W = I W . Thus, in this limit, all
masses are of the order of m m.
On the other hand, it is interesting to study the supergravity corrections to the case that
allowed to restore supersymmetry in global supersymmetry, when the mass term in the
superpotential is absent,
m = 0.
(33)
In this case the solution is given by

x2 + y2 = 2
Vhid
,
2
2
m3/2 MPl
x 2 y 2 = X .
(34)
Obviously, there is no supersymmetry breaking by D-terms, and DX = 0. Actually, in this

limit the vacuum energy cannot be fine tuned without extra contributions to the potential,
2 . This is however an artifact of
irrespective of Vhid , since it is always given by m23/2 MPl
the limit m = 0, and for non-vanishing m , the value at the minimum can be shifted by
varying Vhid . Finally as a special case of Eq. (34), one may consider the case
2
Vhid = (2 X )m23/2 MPl
(35)
which gives
x = 0,
y 2 = X .
(36)
The values of Eq. (36) give the minimum of the potential that is in some sense analogous
to the case Eq. (32).
2.5. Multiple U (1) gauge symmetries
With multiple FI-terms in the potential, it is clear that supersymmetry breaking can
occur more generically. We have seen above, that FI-terms that cannot be canceled by
scalar condensates lead to large scale masses, whereas a scalar condensate with a single
FI-term was able to lower the masses to the electro-weak scale m3/2 , similar to [6]. If
multiple FI-terms now come accompanied by the same number of extra charged fields to
develop condensates, then the Lagrangian would just be a sum of identical copies of the
123
one of the previous sections, and nothing new is found. An interesting case arises, when
there is a mismatch, and not all the FI-parameters can be absorbed by fields like , and
thus some large masses can be generated. We now analyze the situation, when there still is
only a single set of extra charged fields , that takes a non-vanishing expectation value,
but there are multiple U (1)a gauge symmetries, under which it is charged. The potential is
only modified by summing over D-terms,
4 VD =
g2
2
a
a K+ |x|2 a K |y|2 + a ,
2
a
again making a dimensionless by a 2 a . Regarding the vacuum energy cancellation, the new minimization conditions read

m23/2 2
m2
K+ x 2
a ga2 Da + 4 VF + 2 x + xy 2 (K+ + K ) + 2 K+
x
MPl
MPl
a

m m3/2
+
y (K+ x)2 + (K y)2 + 2K 3 = 0,
2
MPl

m23/2 2
m2
2
2
4
K y
a ga Da + VF + 2 y + x 2 y(K+ + K ) + 2 K
y
MPl
MPl
a

m m3/2
+
(37)
x (K+ x)2 + (K y)2 + 2K+ 3 = 0.
2
MPl
It is not necessary to study the general solution here, since the multiple FI-terms already
break supersymmetry without the superpotential (3), and one can therefore set m = 0.
With K+ = K = 1 one can easily solve for x, y, finding
2
Vhid
a ga a
2
2
x2 + y2 = 2 2
(38)
,
x
y
=
2.
2
2
m3/2 MPl
MPl
a ga
Thus, in a generic situation, where not all a are equal, supersymmetry will be broken,
and Da a m3/2 MPl . In this case, the vacuum energy is given by the negative F-term
contribution plus D-terms, and can be fine tuned even without Vhid . The F-terms are given
by F = 2 W0 m3/2 MPl . The contributions to the scalar masses now look like (with
dimensionful a )

ga2 Qia Da
ga2 Qia a + ,
m2i = m2gr,i +
(39)
a
where the terms in parentheses can be of the same order of magnitude, but the gravitymediated contributions are negligible. This realizes the split supersymmetry scenario, if
a 101013 GeV, for at least two FI-parameters, but with a low gravitino mass m3/2
O(TeV).
It may also be interesting to note that at the same time all other soft breaking parameters
will get modified in the presence of FI-terms and scalar field condensates. This happens
2
through the prefactor e K in the total potential. For instance, the bi-linear couplings B
124
and the tri-linear couplings A are (see, e.g., [18])

A0 ,
1
2
0
1/2 e 2 (|x|2 +|y|2 )+ ,
B
m3/2 e K/2 = m3/2 (S + S)
(40)
which may lead to extra suppression factors, depending on the model. Further, we note that
the Higgs mixing parameter that enters in the superpotential in the form H1 H2 , where
H1 and H2 are the two Higgs doublets of MSSM, remains essentially unaffected. This is
so because in string/supergravity scenarios one expects the term to arise from the Khler
potential using a Khler transformation after the spontaneous breaking of supersymmetry
has taken place. The Khler transformation is sensitive to the F-term breaking and not the
D-term. Thus, one expects the same mechanism that produces a term of electro-weak
size for supergravity models to hold in this case as well.
2.6. Scenarios with partial mass hierarchies
To summarize, there are several scenarios that are possible, which would lead to different patterns in the mass spectrum: (i) There is only one FI-term, and an extra scalar field
beyond the MSSM fields which forms a vacuum expectation value. Here all the soft
scalar masses will be of electro-weak size. (ii) In the case of two or more U (1)a with nonvanishing a , and the charges Qia non-zero and sufficiently generic, the scalar masses are
of the order of a , while the gauginos stay light. One can thus get a hierarchical splitting of
scalar and fermion masses. (iii) One may also achieve hybrid scenarios, when only some of
the charges Qia are non-vanishing, or such that the high mass terms just cancel out. Below
we consider two specific scenarios with partial splitting of scalars. The implications of this
partially split scenario will be very different from the usual SUGRA scenario and also from
the high scale supersymmetry scenario of split supersymmetry.
2.6.1. Model I: 2 + 1 generations
As the first model we consider the case with non-vanishing charges for the first and
second generation of squarks and sleptons, but vanishing charges for the third generation.
In this circumstance the former will develop super heavy masses and will not be accessible
at the LHC, as opposed to the third generation. The above implies that the dangerous
flavor changing neutral currents will be automatically suppressed. Some of the signals of
this scenario will be very unique, such as the decay of the gluino (g).
In SUGRA its decay
modes are as follows

ui u i j0 , di di j0 ,
di di
g u i u i ,
(41)
ui di k , di u i k+ ,
where i = 1, 2, 3 are the generational indices, j = 1, 2, 3, 4, and k = 1, 2 for neutralinos and charginos. However, for the case of the model under consideration the decay
through the first two generations is highly suppressed, and the gluino decay can only occur
through third generation squarks via the modes g bb j0 with admixtures of g bt k+ ,
g t b k , and g t t j0 , depending on the gluino mass. There will be no contribution
to the anomalous magnetic moment of the muon at the one-loop level. The following set
of phases can arise: , i , and A , where is the phase of the Higgs mixing parameter
125
, i (i = 1, 2, 3) are the phases of the SU(3)c , SU(2)L and U (1)Y gaugino masses, and
A is the phase of the common trilinear coupling for the third generation scalars (however,
it should be kept in mind that not all the phases are independent and only certain combinations enter in physical processes). Because of the super-heavy nature of the first two
generations, the one-loop contributions to the electric dipole moment (edm) of the electron
and of the neutron are highly suppressed. However, there can be higher loop corrections
to the edms. Specifically, the neutron edm can get a contribution from the CP violating
dimension six operator. Unification of gauge coupling constants at the one-loop level will
remain unchanged, although at the two-loop level there will be effects from the splittings.
In this model the staus can be light and thus co-annihilation of the LSP neutralinos with
the staus can occur, allowing for the possibility that the neutralino relic density could be
consistent with the current data. Finally, proton decay from dimension five operators would
not arise via dressings from the first and second generation squarks and sleptons, but can
arise from the dressings of the third generation sfermions.
2.6.2. Model II: 5 + 10 split spectrum
As another illustrative example, we consider a model where the squarks and sleptons
belonging to a 10 of SU(5) have non-vanishing charges and high scale masses, while the
squarks and sleptons belonging to a 5 have vanishing charges. In this case, the only light
scalars aside from Higgs bosons will be diC , eLi , i (i = 1, 2, 3). In this model, unlike the
case of model I, there is a one-loop supersymmetric contribution to the muon anomalous
magnetic moment g 2. The CP phases in this model consist of and i (i = 1, 2, 3).
There also is a one-loop supersymmetric contribution to the edms of the electron and of the
neutron. Further, the decays of the gluino, the chargino and the neutralino can occur only
via a smaller subset of states and thus their decay widths will be relatively smaller, though
they will still decay within the detection chamber. Finally proton decay through dimension
five operators will be highly suppressed in this model and the dominant decay will occur
through the usual vector boson interactions.
3. Hierarchical breaking in D-brane models

We turn now to the question of how the models discussed so far fit into string theory,
in particular, in the class of intersecting or magnetized D-brane models [19,20]. These
are models within orientifold string compactifications of type II strings with D-branes
that wrap parts of the internal compactification space, and either with magnetic field
backgrounds on the brane world volume (in type IIB) or with the branes intersecting nontrivially on the internal space (in type IIA) [21]. For these models a great deal about the
Lagrangian of their low energy field theory description is known, e.g., in [22,23], and the
soft supersymmetry breaking terms in the conventional setting with F-terms generated in
the hidden sector, have been determined. Furthermore, it is known how FI-terms can appear [24].8
8 These models have also been recently discussed in the context of split supersymmetry in [25].
126
The gauge group for any single stack of Na D-branes is U (Na ) = SU(Na ) U (1)a
(only sometimes an Sp(Na ) or SO(2Na ) subgroup thereof), and thus involves extra Abelian
U (1)a factors, when the Standard Model is engineered. In the first models that were constructed to reproduce the non-supersymmetric Standard Model [20], there are four extra
U (1)a , including the anomaly-free hyper charge and gauged B L quantum numbers,
but also two extra U (1) factors which are anomalous. The relevant anomaly is actually a mixed Abeliannon-Abelian anomaly, i.e., there are triangle diagrams of the type
SU(Nb )2 U (1)a non-vanishing for either SU(2) or SU(3). In particular, tr(Qa ) = 0, and
there is no gravitational anomaly. While the first models were non-supersymmetric from
the beginning, the structure of charge assignments, given in Table 1 of [20], and the number
of U (1)a can serve as a representative example. From this it is clear that these models at
least contain two candidate U (1)a , which may develop FI-terms. In generality, it is known
that, when D-branes violate supersymmetry, this is reflected by FI-terms in the effective
theory [24]. The violation of supersymmetry translates into a violation of the -symmetry
on their world volume, and is geometrically captured by a violation of the so-called calibration conditions. For intersecting D-branes models, this has a very simple geometric
interpretation. Any single brane wraps a three-dimensional internal space. In the case of an
orbifold compactification it is characterized by three angles ia , i = 1, 2, 3, one for each T2
in T6 = (T2 )3 , measuring the relative angle of the D-brane with respect to some reference
orientifold plane. The supersymmetry condition reads 1a 2a 3a = 0 mod 2 , with
some choice of signs. The FI-parameter at leading order is proportional to the deviation,

a 1a 2a 3a mod 2.
(42)
The angles are moduli-dependent quantities, and thus the question if a FI-term is generated
cannot be finally answered without solving the moduli stabilization problem for the relevant moduli. In the mirror symmetric type IIB description with magnetized branes this is
manifest, and the relation reads

a
fa1
fa2
fa3
,
T1 + T1
T2 + T2
T3 + T3
(43)
where fai /(Ti + T ) = tan(ia ), and the Ti are the three (dimensionless) moduli, whose real
parts measure the sizes of the three T2 , while the fai are rational numbers. The natural
scale for the FI-parameter is the string scale Ms = (
)1/2 , and a suppression means that
the right-hand side is small numerically.
Another very important property of the string theoretic embedding of D-terms is the fact
that tachyonic masses (negative mass squared) can be lifted to positive values. Inspecting
the charge assignments, e.g., in [20], one realizes that it is not feasible to have positive
charges under the U (1)a for all the fields of the MSSM. This would naively mean that
some m2i ga2 Qia Da are negative, which would lead to a breakdown of gauge symmetry.
However, in the particular case of orbifold models, the exact string quantization can be
performed, and the mass spectrum computed for small FI-parameters, without using effective field theory. It turns out that for small angles (iab = ia ib < /2), the mass of the
127
lowest excitation of two intersecting branes a and b is given by

m2i =
3

1 ab
i max iab ,
i
2
(44)
i=1
which, for a proper choice of signs, vanishes precisely if a = 0, consistent with the effective description. However, depending on the angles, there is a region in parameter space,
where the deviation from a = 0 only induces positive squared masses, and no tachyons
(see, e.g., [26] in this context). This comes as a surprise from the low energy point of view,
and is explained by the presence of higher order corrections in the derivative expansion of
the BornInfeld effective action, which become important for strings stretching between
intersecting branes [27]. The conclusion is, that the effective mass that follows from the
D-term potential is corrected to positive values, and we can tolerate tachyons in the field
theoretic models.
Taking these observations together, it seems first of all possible that D-brane models of
the type discussed have mass spectra with important contributions from FI-terms. Since
multiple extra U (1)a factors exist and turn anomalous, a scenario, where the Da fields
cannot be relaxed to the electro-weak scale, appears plausible. This would imply a mass
hierarchy between charged scalars and fermions. To solve the moduli stabilization problem,
and show convincingly how the D- and F-terms of the desired magnitude are generated
dynamically, of course, remains an open challenge. One may also want to turn the argument
around, to conclude that within the conventional approach with low energy supersymmetry,
the potential presence of many FI-terms is a great danger for these types of models, and
one has to find ways to suppress them dynamically.
Finally, we like to mention a caveat which makes the classes of D-branes models that
we discussed rather not so good candidates to realize split supersymmetry. It is well known
that in D-brane models in general the unification of gauge interactions really only happens
at the string scale, and not in the field theory regime. This means that the original motivation to keep the gauginos and higgsinos light, while giving up on the scalars, is upset.
The gauge kinetic functions are moduli-dependent fa = fa (TI ), and a unification of couplings requires some relations among moduli to hold [23,25,28], which so far can appear
accidentally in various models in the literature, but do not seem to have any independent
justification. This means, that an essential motivation for split supersymmetry, gauge unification, is only accidentally realized.
4. Hierarchical D-term inflation

In hierarchical supersymmetry breaking of the type discussed here it seems an intriguing
suggestion to relate the mass scale of the heavy scalars to cosmology. In our model, the potential energy is generated at the conventional supersymmetry breaking scale (m3/2 MPl )2 ,
and with standard values comes out about (101013 GeV)4 . This is the scale of the individual contributions of the F- and D-terms to the full potential, and only the fine tuning
of the cosmological constant leads to a cancellation. It is now very tempting to identify
the vacuum energy of these individual components with the vacuum energy that drives
128
inflation, by undoing the fine tuning. A possible scenario is very easily illustrated along
the lines of D-term inflation [29]. Roughly speaking, the only required modification of the
model we used so far, with the MSSM extended by one or many extra U (1)a gauge factors,
plus a pair of charged fields , is to promote the mass term (3) in the superpotential to a
dynamical field , neutral under U (1)a , which plays the role of the inflaton. We now write
W = + + W0 ,
(45)
and use a canonically normalized Khler potential,

2 2
K = + + + ||2 + K0 .
(46)
With this one finds the scalar potential of the form

2
2
2
2 2 2
2

V = Vhid + e K + + + + + 4 + + + |W |2

2
ga2
+ 2 2

+ a 2 ,
+ 3 2 + 3 2 |W0 |2 +
(47)
2
a
where the only negative contribution comes from 3 2 e K |W0 |2 . Above some threshold
value for , the two charged fields are stabilized at the origin = 0. Their masses at
zero are

2

2
m2 = 0 = 2 Vhid + e K0 1 + 6 |W0 |2 ||2 2 4 |W0 |2
ga a , (48)
2
which turns positive, when is large enough (but still well below the Planck scale). The
potential simplifies to

ga2 2

2
.
V = 0 = Vhid + 2 e K0 |W0 |2 2 ||2 3 +
2 a
a
(49)
The inflationary slow-roll condition for the second derivative of the potential is [30]
2
V
|| = 2 1,
(50)
V
where || 1 numerically means || 0.01 is acceptable. This implies that
2 2
2
MPl
a ga a
(51)
100.
2K
e
|W0 |2
The first derivative is then automatically also small (with a ), and inflation can
be successful. This means that the very same D-term vacuum energy that is responsible
for the large scalar masses can drive inflation, if either the D-terms are enhanced or the
F-term vacuum energy is sufficiently suppressed during that period. Thus, during the de
Sitter phase the relation between the Hubble parameter H and the energy density , i.e.,
2 H 2 = = 1 2 + V, shows that the Hubble expansion in this phase is
the relation 3MPl
2
dominated by the FI-term
2 2
H
3MPl
g2
a 2
a .
(52)
129
In summary, if in the phase of large the vacuum energy of the D-terms is larger than that
of the F-terms by a factor of about 100 or more, the D-term energy can drive inflation at a
scale up to 101215 GeV, slightly above the mass scale of the heavy scalars. After the end
of the slow roll period, will eventually fall below the threshold value, and will form
condensates themselves. This can then lead to a partial relaxation of the D-term vacuum
energy, but a more elaborate model of the hidden sector would be needed to describe
this
phase
transition
properly.
In
the
minimum,
one
may
expect
to
settle
down
to

m3/2 MPl on dimensional grounds, which is compatible with small gaugino masses.9
5. Conclusion
We have presented a model of supersymmetry breaking in the context of string and supergravity scenarios by inclusion of both F- and FayetIliopoulos D-terms, arising from
extra U (1) factors in the gauge group. Such extra U (1) gauge symmetries arise quite naturally in string based models. It was shown that scalars charged under the extra U (1)
gain large masses from the FI-terms, proportional to the charges of the respective scalar
fields under the extra U (1). This leads generically to non-universal masses
for the heavy
scalars. The cancellation of the vacuum energy puts an upper bound of m3/2 MPl on the
scalar masses, and thus also puts a bound on the FI-term X . The bound on X can be met
2.
in heterotic string models only for m3/2 close to MPl , since there X is scaled by MPl
Thus, heterotic string scenarios are not preferred from the vacuum energy constraint, when
m3/2 = O(TeV). However, m3/2 = O(TeV) and X m3/2 MPl , i.e., of size 101013 GeV,
could arise in orientifolds models which allow X of a variable moduli-dependent size.
The fact that the D-term contributions to the scalar masses depend on their U (1) charges
opens up the possibility of building a new class of models with some scalars (with vanishing U (1) charges) light and others (with non-vanishing U (1) charges) heavy, while the
gauginos and higgsinos gain masses only of electro-weak size. Further, the term is essentially unaffected by the FI contribution, and can be of electro-weak size. We investigated
two illustrative examples of models with light and heavy scalars (model I and model II
in Section 2) and showed that they lead to significantly different phenomenologies which
could be tested at colliders and in non-accelerator experiment. The class of models we have
discussed here are different from the high scale supersymmetry models of Ref. [2], where
all scalars and the gravitino are super heavy. However, which scalars are heavy and which
are light is now a model-dependent question. A further interesting feature is the possibility
that the vacuum energy responsible for generating heavy scalars may also drive inflation.
An analysis of how this can come about was discussed in Section 4. It would be interesting
to investigate more explicit D-brane constructions to build models of the type advocated
here.
9 As mentioned earlier, we so far ignore here the problem of moduli stabilization, which is even more severe
in the context of inflation with D-brane degrees of freedom, where already the correct choice of supersymmetric
coordinates is a very subtle question [31]. The FI-parameters depend on the moduli fields, and it will be necessary
to stabilize the relevant fields (as, for instance, along the lines of [32]) at a scale larger that the scale of inflation,
or they would have to be considered as dynamical (see also the last reference of [29]).
130
Acknowledgements
B.K. would like to thank Angel Uranga for helpful advice. The work of B.K. was supported by the German Science Foundation (DFG) and in part by funds provided by the
US Department of Energy (D.O.E.) under cooperative research agreement #DF-FC0294ER40818. The work of P.N. was supported in part by the US National Science Foundation under the grant NSF-PHY-0139967.
References
[1] A.H. Chamseddine, R. Arnowitt, P. Nath, Locally supersymmetric grand unification, Phys. Rev. Lett. 49
(1982) 970;
R. Barbieri, S. Ferrara, C.A. Savoy, Gauge models with spontaneously broken local supersymmetry, Phys.
Lett. B 119 (1982) 343;
P. Nath, R. Arnowitt, A.H. Chamseddine, Gauge hierarchy in supergravity GUTs, Nucl. Phys. B 227 (1983)
121;
L.J. Hall, J. Lykken, S. Weinberg, Supergravity as the messenger of supersymmetry breaking, Phys. Rev.
D 27 (1983) 2359.
[2] N. Arkani-Hamed, S. Dimopoulos, Supersymmetric unification without low energy supersymmetry and signatures for fine-tuning at the LHC, hep-th/0405159.
[3] L. Susskind, The anthropic landscape of string theory, hep-th/0302219;
M.R. Douglas, The statistics of string/M theory vacua, JHEP 0305 (2003) 046, hep-th/0303194;
T. Banks, M. Dine, E. Gorbatov, Is there a string theory landscape?, JHEP 0408 (2004) 058, hep-th/0309170.
[4] G.F. Giudice, A. Romanino, Split supersymmetry, Nucl. Phys. B 699 (2004) 65, hep-ph/0406088;
N. Arkani-Hamed, S. Dimopoulos, G.F. Giudice, A. Romanino, Aspects of split supersymmetry, hepph/0409232.
[5] A. Arvanitaki, C. Davis, P.W. Graham, J.G. Wacker, One-loop predictions of the finely tuned SSM, hepph/0406034;
A. Pierce, Dark matter in the finely tuned minimal supersymmetric standard model, Phys. Rev. D 70 (2004)
075006, hep-ph/0406144;
S.H. Zhu, Chargino pair production at linear collider and split supersymmetry, hep-ph/0407072;
B. Mukhopadhyaya, S. SenGupta, Sparticle spectrum and phenomenology in split supersymmetry: some
possibilities, hep-th/0407225;
W. Kilian, T. Plehn, P. Richardson, E. Schmidt, Split supersymmetry at colliders, hep-ph/0408088;
R. Mahbubani, Bounds on the Higgs mass in variations of split supersymmetry, hep-ph/0408096;
M. Binger, The Higgs boson mass at 2 loops in the finely tuned split supersymmetric Standard Model,
hep-ph/0408240;
J.L. Hewett, B. Lillie, M. Masip, T.G. Rizzo, Signatures of long-lived gluinos in split supersymmetry,
JHEP 0409 (2004) 070, hep-ph/0408248;
L. Anchordoqui, H. Goldberg, C. Nunez, Probing split supersymmetry with cosmic rays, hep-ph/0408284;
K. Cheung, W.Y. Keung, Split supersymmetry, stable gluino, and gluinonium, hep-ph/0408335;
D.A. Demir, Effects of flavor violation on split supersymmetry, hep-ph/0410056;
R. Allahverdi, A. Jokinen, A. Mazumdar, Gravitino production from reheating in split supersymmetry, hepph/0410169;
V. Barger, C.W. Chiang, J. Jiang, T. Li, Axion models with high-scale supersymmetry breaking, hepph/0410252;
J.A. Casas, J.R. Espinosa, I. Hidalgo, Implications for new physics from fine-tuning arguments. I: application
to SUSY and seesaw cases, hep-ph/0410298;
B. Bajc, G. Senjanovic, Radiative seesaw: a case for split supersymmetry, hep-ph/0411193.
[6] G.R. Dvali, A. Pomarol, Anomalous U(1) as a mediator of supersymmetry breaking, Phys. Rev. Lett. 77
(1996) 3728, hep-ph/9607383.
131
[7] P. Binetruy, E. Dudas, Gaugino condensation and the anomalous U(1), Phys. Lett. B 389 (1996) 503, hepth/9607172.
[8] L. Randall, R. Sundrum, Out of this world supersymmetry breaking, Nucl. Phys. B 557 (1999) 79, hepth/9810155;
G.F. Giudice, M.A. Luty, H. Murayama, R. Rattazzi, Gaugino mass without singlets, JHEP 9812 (1998)
027, hep-ph/9810442.
[9] I. Antoniadis, T.R. Taylor, Topological masses from broken supersymmetry, Nucl. Phys. B 695 (2004) 103,
hep-th/0403293.
[10] M. Klein, Anomaly cancellation in D = 4, N = 1 orientifolds and linear/chiral multiplet duality, Nucl. Phys.
B 569 (2000) 362, hep-th/9910143;
H. Ruegg, M. Ruiz-Altaba, The Stueckelberg field, Int. J. Mod. Phys. A 19 (2004) 3265, hep-th/0304245;
B. Krs, P. Nath, A Stueckelberg extension of the Standard Model, Phys. Lett. B 586 (2004) 366, hepph/0402047;
H. Ruegg, M. Ruiz-Altaba, A supersymmetric Stueckelberg U(1) extension of the MSSM, hep-ph/0406167;
J. Louis, W. Schulgin, Massive tensor multiplets in N = 1 supersymmetry, hep-th/0410149.
[11] Z. Lalak, S. Lavignac, H.P. Nilles, String dualities in the presence of anomalous U(1) symmetries, Nucl.
Phys. B 559 (1999) 48, hep-th/9903160;
H.P. Nilles, Remarks on anomalous U(1) symmetries in string theory, hep-ph/0003102.
[12] M. Dine, N. Seiberg, E. Witten, FayetIliopoulos terms in string theory, Nucl. Phys. B 289 (1987) 589;
J.J. Atick, L.J. Dixon, A. Sen, String calculation of FayetIliopoulos D terms in arbitrary supersymmetric
compactifications, Nucl. Phys. B 292 (1987) 109.
[13] C. Angelantonj, A. Sagnotti, Open strings, Phys. Rep. 371 (2002) 1, hep-th/0204089;
C. Angelantonj, A. Sagnotti, Phys. Rep. 376 (2003) 339, Erratum.
[14] L.E. Ibanez, R. Rabadan, A.M. Uranga, Anomalous U(1)s in type I and type IIB D = 4, N = 1 string vacua,
Nucl. Phys. B 542 (1999) 112, hep-th/9808139.
[15] E. Poppitz, On the one-loop FayetIliopoulos term in chiral four-dimensional type I orbifolds, Nucl. Phys.
B 542 (1999) 31, hep-th/9810010.
[16] A. Chamseddine, R. Arnowitt, P. Nath in [1];
E. Cremmer, S. Ferrara, L. Girardello, A. Van Proeyen, YangMills theories with local supersymmetry:
Lagrangian, transformation laws and super-Higgs effect, Nucl. Phys. B 212 (1983) 413.
[17] V.S. Kaplunovsky, J. Louis, Model independent analysis of soft terms in effective supergravity and in string
theory, Phys. Lett. B 306 (1993) 269, hep-th/9303040.
[18] P. Nath, T.R. Taylor, Modular invariance, soft breaking, and tan() in superstring models, Phys. Lett.
B 548 (2002) 77, hep-ph/0209282.
[19] R. Blumenhagen, L. Grlich, B. Krs, D. Lst, Noncommutative compactifications of type I strings on tori
with magnetic background flux, JHEP 0010 (2000) 006, hep-th/0007024;
C. Angelantonj, I. Antoniadis, E. Dudas, A. Sagnotti, Type-I strings on magnetised orbifolds and brane
transmutation, Phys. Lett. B 489 (2000) 223, hep-th/0007090;
G. Aldazabal, S. Franco, L.E. Ibanez, R. Rabadan, A.M. Uranga, D = 4 chiral string compactifications from
intersecting branes, J. Math. Phys. 42 (2001) 3103, hep-th/0011073;
M. Cvetic, G. Shiu, A.M. Uranga, Chiral four-dimensional N = 1 supersymmetric type IIA orientifolds from
intersecting D6-branes, Nucl. Phys. B 615 (2001) 3, hep-th/0107166.
[20] L.E. Ibanez, F. Marchesano, R. Rabadan, Getting just the standard model at intersecting branes, JHEP 0111
(2001) 002, hep-th/0105155.
[21] C. Bachas, A way to break supersymmetry, hep-th/9503030;
M. Berkooz, M.R. Douglas, R.G. Leigh, Branes intersecting at angles, Nucl. Phys. B 480 (1996) 265, hepth/9606139.
[22] D. Cremades, L.E. Ibanez, F. Marchesano, SUSY quivers, intersecting branes and the modest hierarchy
problem, JHEP 0207 (2002) 009, hep-th/0201205;
M. Cvetic, I. Papadimitriou, Conformal field theory couplings for intersecting D-branes on orientifolds,
Phys. Rev. D 68 (2003) 046001, hep-th/0303083;
M. Cvetic, I. Papadimitriou, Phys. Rev. D 70 (2004) 029903, Erratum;
S.A. Abel, A.W. Owen, Interactions in intersecting brane models, Nucl. Phys. B 663 (2003) 197, hepth/0303124;
132
[23]
[24]
[25]
[26]
[27]
[28]
[29]
[30]
[31]
[32]
P.G. Camara, L.E. Ibanez, A.M. Uranga, Flux-induced SUSY-breaking soft terms, Nucl. Phys. B 689 (2004)
195, hep-th/0311241;
D. Lst, P. Mayr, R. Richter, S. Stieberger, Scattering of gauge, matter, and moduli fields from intersecting
branes, Nucl. Phys. B 696 (2004) 205, hep-th/0404134.
B. Krs, P. Nath, Effective action and soft supersymmetry breaking for intersecting D-brane models, Nucl.
Phys. B 681 (2004) 77, hep-th/0309167.
M.R. Douglas, G.W. Moore, D-branes, quivers, and ALE instantons, hep-th/9603167;
D. Cremades, L.E. Ibanez, F. Marchesano, Intersecting brane models of particle physics and the Higgs
mechanism, JHEP 0207 (2002) 022, hep-th/0203160;
R. Blumenhagen, V. Braun, B. Krs, D. Lst, Orientifolds of K3 and CalabiYau manifolds with intersecting
D-branes, JHEP 0207 (2002) 026, hep-th/0206038.
I. Antoniadis, S. Dimopoulos, Splitting supersymmetry in string theory, hep-th/0411032.
R. Rabadan, Branes at angles, torons, stability and supersymmetry, Nucl. Phys. B 620 (2002) 152, hepth/0107036.
A. Hashimoto, W.I. Taylor, Fluctuation spectra of tilted and intersecting D-branes from the BornInfeld
action, Nucl. Phys. B 503 (1997) 193, hep-th/9703217;
F. Denef, A. Sevrin, J. Troost, Non-Abelian BornInfeld versus string theory, Nucl. Phys. B 581 (2000) 135,
hep-th/0002180.
R. Blumenhagen, D. Lst, S. Stieberger, Gauge unification in supersymmetric intersecting brane worlds,
JHEP 0307 (2003) 036, hep-th/0305146.
P. Binetruy, G.R. Dvali, D-term inflation, Phys. Lett. B 388 (1996) 241, hep-ph/9606342;
E. Halyo, Hybrid inflation from supergravity D-terms, Phys. Lett. B 387 (1996) 43, hep-ph/9606423;
P. Binetruy, G. Dvali, R. Kallosh, A. Van Proeyen, FayetIliopoulos terms in supergravity and cosmology,
Class. Quantum Grav. 21 (2004) 3137, hep-th/0402046.
A.R. Liddle, D.H. Lyth, Cosmological Inflation and Large Scale Structure, Cambridge Univ. Press, Cambridge, 2000.
S. Kachru, R. Kallosh, A. Linde, J. Maldacena, L. McAllister, S.P. Trivedi, Towards inflation in string theory,
JCAP 0310 (2003) 013, hep-th/0308055;
J.P. Hsu, R. Kallosh, S. Prokushkin, On brane inflation with volume stabilization, JCAP 0312 (2003) 009,
hep-th/0311077;
H. Firouzjahi, S.H.H. Tye, Closer towards inflation in string theory, Phys. Lett. B 584 (2004) 147, hepth/0312020;
M. Berg, M. Haack, B. Krs, Loop corrections to volume moduli and inflation in string theory, hepth/0404087;
M. Berg, M. Haack, B. Krs, On the moduli dependence of nonperturbative superpotentials in brane inflation, hep-th/0409282;
H. Jockers, J. Louis, The effective action of D7-branes in N = 1 CalabiYau orientifolds, hep-th/0409098.
S. Kachru, R. Kallosh, A. Linde, S.P. Trivedi, De Sitter vacua in string theory, Phys. Rev. D 68 (2003)
046005, hep-th/0301240.
Auxiliary field methods in supersymmetric

non-linear sigma models
Muneto Nitta
Department of Physics, Purdue University, West Lafayette, IN 47907-1396, USA 1
Received 10 December 2003; accepted 19 January 2005
Abstract
Auxiliary field methods in D = 2 (or 3), N = 2 supersymmetric (SUSY) non-linear sigma models
(NL Ms) are studied. For these models auxiliary fields as Lagrange multipliers belong to a vector or
a chiral superfield, which gives a Khler quotient of complexified gauge group or a holomorphic constraint on it, respectively. Using these, NL Ms on all Hermitian symmetric spaces were formulated
previously. In this paper, we formulate new SUSY NL Ms on some rank-two Khler coset spaces as
SUSY gauge theories with two FayetIliopoulos parameters.
PACS: 11.30.Pb; 11.30.Na; 11.10.Lm; 11.10.Kk; 11.30.Qc
1. Introduction
The auxiliary field method and the large-N method are very powerful tools to study
non-perturbative effects in a lot of theories, such as the GrossNeveu model, non-linear
sigma models (NL Ms), gauge theories, matrix models and so on [1]. For two-dimensional
(D = 2) NL Ms on coset spaces G/H , this method displays similarities with fourdimensional (D = 4) gauge theories very easily: dynamical mass generation, dynamical
symmetry breaking/restoration, dynamically induced gauge bosons, the asymptotic freeE-mail addresses: nitta@physics.purdue.edu, nitta@th.phys.titech.ac.jp (M. Nitta).
1 Current address: Department of Physics, Tokyo Institute of Technology, Tokyo 152-8551, Japan.
doi:10.1016/j.nuclphysb.2005.01.025
134
M. Nitta / Nuclear Physics B 711 (2005) 133162
dom and so on. Supersymmetric (SUSY) extensions are also studied for D = 2, N = 1
and N = 2 SUSY NL Ms [26]. D = 3 NL Ms are non-renormalizable in perturbative method but renormalizable in the large-N expansion [7]. SUSY extensions of D = 3
NL Ms are also studied extensively [8,9].
As an example, the auxiliary field method for the O(N ) model can be illustrated briefly
as follows. Let gij () be the metric on S N 1 = O(N )/O(N 1) with some coordinates
i (i = 1, . . . , N 1). Then the partition function for the O(N ) model is given by

1
Z = [d] exp d D x gij () i j .
(1.1)
2
Classically O(N ) symmetry is spontaneously broken down to O(N 1). Introducing an
auxiliary field as a Lagrange multiplier, this can be rewritten as

1
Z = [d d ] exp d D x + 2 r 2
(1.2)
,
2
with = { A } (A = 1, . . . , N ) being an O(N ) vector of scalar fields. The integration over
supplies a constraint 2 = r 2 and the Lagrangian in (1.1) is rederived. On the other
hand, leaving , we can first perform the integration over dynamical fields A exactly as
the Gaussian integral. Then, the integration over can be approximated by the saddle point
in the large-N limit with taking r 2 = N/g 2 . We thus obtain the gap equation. In D = 2
gets non-zero vacuum expectation value (VEV) = 0 by solving it. Thus the O(N )
symmetry is dynamically recovered and the mass generation is found as a non-perturbative
effect contrary to masslessness in the perturbative analysis. This agrees with the Colemans
theorem which prohibits massless bosons in D = 2 [10]. In D = 3 there exist two phases
of broken and unbroken O(N ) symmetry.
Therefore, the auxiliary field formulation is crucial for a study of non-perturbative ef
fects. It is very easy to find the metric gij () solving the given constraint on linear fields :
N
for instance, for the O(N ) model, putting = (,
) and eliminating the N th component
i j
N
2
2
by = r , we obtain the metric gij () = ij + r2 2 expressed in terms of the
independent fields i . It is, however, in general difficult to find proper constraints among
linear fields to give a metric gij () of a given model. What we really want to do for a
study of non-perturbative effects is in this direction and this is the most difficulty of the
auxiliary field method. In SUSY theories, the situation becomes far more complicated as
will be explained below.
For D = 2, 3, N = 1 SUSY NL Ms, the situation is the same with bosonic case because
target spaces are Riemannian. The D = 2, N = 1 SUSY O(N ) model was investigated in
[3] and dynamical chiral symmetry breaking was found. The D = 3, N = 1 SUSY O(N )
model was discussed in [8]. D = 2, 3, N = 2 SUSY NL Ms are obtained as dimensional
reduction from D = 4, N = 1 SUSY NL Ms. For these cases target spaces must be Khler
[11,12] so that this makes the auxiliary field formulation difficult. D = 2 (D = 3), N = 2
SUSY NL Ms on Khler coset spaces G/H may have similarities with D = 4 (D = 5),
N = 2 SUSY QCD [13] as the same as bosonic theories. So pursuing similarities of these
SUSY models using non-perturbative method is very interesting.
In SUSY theories, auxiliary fields as Lagrange multipliers belong to superfields.
For these N = 2 SUSY theories, there exist vector and chiral superfields in terms of
135
Table 1
Hermitian symmetric spaces (HSS)
Type
AIII1
AIII2
BDI
CI
DIII
EIII
EVII
G/H
Complex coordinates
dimC (G/H )
)
CP N 1 = SU(NSU(N
1)U (1)
U (N )
GN,M = U (N M)U
(M)
(N 1)-vector
N 1
[M (N M)]-matrix
M(N M)
)
QN 2 = SO(NSO(N
2)U (1)
(N 2)-vector
N 2
Sp(N )
U (N )
SO(2N )
U (N )
E6
SO(10)U (1)
E7
E6 U (1)
Symmetric (N N )-matrix
Asymmetric (N N )-matrix
1 N (N + 1)
2
1 N (N 1)
2
16-spinor
16
27-vector
27
Classification of Hermitian symmetric spaces (HSS) by Cartan, the standard complex coordinates belonging to
the representation of H and their complex dimensions are shown.
D = 4, N = 1 SUSY [12]. Using vector superfields as auxiliary fields, the CP N model

[4,5] (CP N = SU(N )/[SU(N 1) U (1)]) and the Grassmann model [14] (GN,M =
U (N )/[U (N M) U (M)]) were constructed very long time ago.
The CP N model is simply given in the auxiliary field formulation [4,5] as

L = d 4 eV cV ,
(1.3)
with chiral superfields of an N -vector, V an auxiliary vector superfield for U (1) gauge
symmetry, and c a real positive constant called the FayetIliopoulos (FI) parameter. In the
CP N model, V acquires kinetic term so that U (1) gauge boson is dynamically induced
in the large-N limit [4,5]. In D = 2 the scalar components in V acquire the vacuum expectation value (VEV), and there exists the mass generation without breaking of the gauge
symmetry through the Schwinger mechanism [15]. D = 3, N = 2 SUSY CP N model was
discussed in Ref. [9].
The GN,M model is given in [14] by

L = d 4 tr eV c tr V ,
(1.4)
with chiral superfields of an N by M matrix and V auxiliary vector superfields of an M
by M matrix for U (M) gauge symmetry. This model is expected to induce U (M) gauge
bosons in the large-N limit.
There was not, however, auxiliary field formulation for other D = 2, 3, N = 2 SUSY
NL Ms up to a few years ago. To overcome such situation, the auxiliary field formulation
for SUSY NL Ms on a broad class of Khler coset spaces, Hermitian symmetric spaces
(HSS) summarized in Table 1, has been given in Refs. [1618]. Besides auxiliary vector
superfields V with the Khler potential (1.3) or (1.4), we introduced auxiliary chiral superfields () as summarized in Table 2. The integration over V gives the Khler potential for
CP N or GN,M , whereas the integration over () gives holomorphic constraints which
embed the manifold into CP N or GN,M .
The large-N analysis of these models has become possible. The simplest model other
than CP N is the QN model, which is called the quadric surface and is the coset of QN 2 =
136
Table 2
Auxiliary field formulation for HSS
G/H
()
()
Superpotential
Constraints
Hosts
SO(N )
SO(N 2)U (1)
SO(2N ) Sp(N )
U (N ) , U (N )
E6
SO(10)U (1)
E7
E6 U (1)
U (1)
2 = 0
CP N 1
U (N )
tr( T J )
T J = 0
G2N,N
ij k j k = 0
CP 26
d = 0
CP 55
2N N
N N
i j k
27
27
U (1)
ij k
56
56
U (1)
Field contents, say dynamical chiral superfields () and auxiliary chiral and vector superfields () and V ,
are displayed in the first three rows. (We have given the representation of G or matrix sizes for () and ()
and gauge symmetry for V .) The superpotentials invariant under G and gauge symmetries and holomorphic
constraints obtained by the integration over () are also shown. In the last row, the host
spaces determined by
0 1
the integration over V are shown. The second rank symmetric tensor J is given by J = 1 0N with = +1
N
(or 1) for SO(2N ) (or Sp(N )), and (d) is the E6 (E7 ) symmetric third (fourth) rank tensor whose explicit
expressions can be found in [16].
SO(N )/[SO(N 2) U (1)]. It is given by [6,16]

L = d 4 eV cV +
d 2 2 + c.c.
(1.5)
with dynamical chiral superfields of an N -vector, V an auxiliary vector superfield for

U (1) gauge symmetry and an auxiliary chiral superfield. The integration over V gives
the Khler potential on CP N and the integration over gives a holomorphic constraint
2 = 0 among dynamical fields . So this model is a hybrid of the O(N ) model (1.2)
and the CP N model (1.3). On the other hand, performing the integration over dynamical fields in (1.5) exactly, the non-perturbative analysis of the D = 2 QN model has
been investigated in the large-N method [6]. It has turned out that the QN model has very
interesting features which did not exist in the previous models. It contains two kinds of
non-perturbatively stable vacua; one is the Schwinger phase in which the scalar components of V acquire VEV like the CP N model and the other is the Higgs phase in which the
scalar components of get VEV. The latter is a new vacuum for the SUSY NL Ms and is
interesting if we investigate similarities with N = 2 SUSY QCD. Both vacua are asymptotically free and the mass gap exist due to the Schwinger and Higgs mechanisms. D = 3,
N = 2 SUSY QN model is also discussed recently [19].
Since the rests of HSS are formulated using auxiliary chiral superfields besides auxiliary
vector superfield, these models are also expected to contain the Higgs phase besides the
Schwinger phase.
Hence our interest is naturally led to SUSY NL Ms on more general Khler coset
spaces other than HSS. Does dynamical mass generation occur for NL Ms on any Khler
coset G/H ? Is any model asymptotically free? Which gauge symmetry is dynamically induced for a given model? It is very important to give answers to these questions. However,
before investigating these problems, we have to ask if NL Ms on any Khler coset G/H
can be formulated by the auxiliary field method or not. This question is, however, a quite
difficult problem. In this paper we will make some progress in this problem.
137
Any Khler coset space can be written as G/H = G/[Hs.s. U (1)r ] where Hs.s. is the
semi-simple subgroup in H and r rank G rank Hs.s. is called the rank of this Khler
coset space [20]. Every HSS is a rank one Khler coset space, whose Lagrangian has a
U (1) or U (M) gauge symmetry and one FI parameter if formulated by the auxiliary field
method. Other rank one Khler coset spaces seem to be relatively easy to be constructed in
the auxiliary field formulation with U (1) or U (M) gauge group [16]. In this paper, we give
the auxiliary field formulation for some rank two Khler coset spaces, SU(N )/[SU(N
2) U (1)2 ] and SU(N )/[SU(N M L) SU(M) SU(L) U (1)2 ]. These models
have U (1)2 and U (M) U (L) gauge symmetries, respectively, and two FI-parameters. In
addition to auxiliary vector superfields for these gauge symmetries, some auxiliary chiral
superfields are also needed even though G = SU(N ). Non-perturbative studies for these
new models become possible which will remain as a future work. We expect that U (1)2 or
U (M) U (L) gauge symmetry is dynamically induced.
This paper is organized as follows. In Section 2, we give the minimum ingredient of
SUSY non-linear realizations needed for later discussions. Section 3 explains how to obtain compact Khler coset spaces using the super-Higgs mechanism eliminating unwanted
QNG bosons. How this works in the simplest CP N and GN,M is shown. In Sections 4 and 5
we generalize these discussions to the rank-two coset spaces SU(N )/[SU(N 2) U (1)2 ]
and SU(N )/[SU(N M L) SU(M) SU(L) U (1)2 ], respectively. Section 6 is
devoted to summary and discussions. In Appendix A, we give a review on the supersymmetric non-linear realization with Khler G/H focusing on the case of G = SU(N ). In
Appendix B, we discuss some geometric aspects of these models, a relation with some
hyper-Khler manifolds and an application to construction of a new CalabiYau metric.
2. Supersymmetric non-linear realizations

The most general discussion for SUSY non-linear realizations was discussed by Bando
et al. [21]. (For a review see [22].) Then they were extensively studied for both compact
[2327] (and references in [26]) and non-compact [2837] target manifolds. In this section
we briefly review the minimum of SUSY non-linear realizations needed for discussions in
the following sections.
A chiral superfield comprises of a complex scalar and a Weyl spinor in terms of the
D = 4, N = 1 superfield formalism. In SUSY non-linear realizations, we consider GC ,
the complexification of a group G, because the symmetry group G of the superpotential is
enhanced to GC by its holomorphic structure. When a global symmetry G is spontaneously
broken down into its subgroup H by vacuum expectation values (VEVs) with SUSY preserved, there appear ordinary NambuGoldstone (NG) bosons for broken G together with
additional massless bosons called the quasi-NG (QNG) bosons for broken GC and their
fermionic SUSY partners [38].2 The unbroken subgroup of GC is also a complex group H
2 If we assume that all vacuum degeneracy come from a symmetry, there are no more massless bosons other
than these NG and QNG bosons. This happens if all GC invariants composed of fundamental fields are fixed to
some values. So in this case, the vacuum manifold becomes a GC -orbit.
138
which contains H C as a subgroup. Its Lie algebra can be written as3

H = HC B,
(2.1)
with B called the Borel algebra comprising of (non-Hermitian) nilpotent generators. Therefore the vacuum manifold M parameterized by massless bosons is topologically a complex
coset space GC /H , which is a non-compact Khler manifold in general. However, we note
that the isometry of this coset space should be still G but not GC because symmetry of the
Khler potential is not complexified, with being different from the superpotential.
A group element in g GC can be uniquely divided into g = h with the coset representative and h H . Here the coset representative is given by
= exp( Z)
GC
H
(2.2)
A set of Z comwith chiral superfields and Z complex broken generators in G C H.
prises of both Hermitian and non-Hermitian generators, because H in general contains

non-Hermitian generators in the Borel algebra B as in Eq. (2.1). Corresponding to these
Hermitian and non-Hermitian broken generators, there exist two kinds of massless chiral
multiplets, called the M-type and P-type multiplets, respectively.
In every P-type multiplet, both real and imaginary parts of a complex scalar field are NG
bosons, parametrizing compact directions of the manifold M. Whereas, in every M-type
multiplet, one degree of freedom is an NG boson and the other is a QNG boson, parametrizing a non-compact direction of M. This can be understood as follows. Correspondingly
to each non-Hermitian broken generator ZP , there exists a non-Hermitian unbroken generator B in the Borel algebra B in H (2.1) with ZP = B. Then, both X i(ZP B) and
X ZP + B are Hermitian. The coset representative can be transformed to an element in
its equivalent class as

= exp( + ZP ) exp( + ZP ) exp B

= exp + i(Re )X + i(Im )X + O 2 .
(2.3)
Therefore both real and imaginary parts of every P-type multiplet , generated by a nonHermitian ZP , parameterize compact directions of M. However, for every M-type, its real
part parameterizes a non-compact direction of M because it is generated by a Hermitian
generator.
Let NNG , NQNG , NP and NM be numbers of NG and QNG bosons and P- and M-type
multiplets, respectively. Then relations
G
NQNG = NM
= 2NP + NM ,
H
hold. The complex dimension of the manifold M can be written as
NNG = dim
(2.4)
1
dimC M = NP + NM = (NNG + NQNG ).
2
(2.5)
3 We denote the Lie algebra of the group by its calligraphic font.
139
If there are no QNG bosons (NQNG = NM = 0) the manifold becomes compact, whereas
if there exists at least one QNG bosons (NQNG = NM 1) the manifold becomes noncompact. For non-compact cases, NG bosons parameterize a compact homogeneous submanifold G/H embedded into the total manifold M GC /H [34]. For both cases, the
isometry of M is G but not GC although GC acts on M transitively.
Even if the same NG bosons appear for the same G/H , NQNG depends on the underlying linear model. Two extreme or natural cases can be considered: the case of the maximal
NQNG equal to NNG (NP = 0) and the case of NQNG = 0 (NM = 0).
(1) If there exist only M-type multiplets without P-type multiplets (NP = 0, NQNG =
NNG ), it is called the maximal realization (or fully-doubling), which corresponds
to GC /H = GC /H C T (G/H ).4 Some sufficient conditions for maximal realizations are known [29]: it occurs if a symmetry G is broken down to H by VEVs of
linear fields (1) which belong to a real representation of G or (2) with G/H a symmetric space. In both cases, absence of gauge fields is assumed.
(2) On the other hand, when there exist only P-type multiplets without M-type multiplets
(NM = NQNG = 0), it is called the pure realization, which corresponds to a compact
homogeneous Khler manifold (Khler coset) G/H . In this case GC /H G/H holds.
The Khler metric on arbitrary Khler coset G/H was constructed by Borel [20]. All
Khler coset spaces G/H were classified by using the painted Dynkin diagrams [23].
Their Khler potentials were given by Itoh, Kugo and Kunitomo (IKK) [24]. For pure
realizations a no-go theorem due to Lerche and Shore is known [29,30] (see also [16,
32,34]): there must appear at least one QNG bosons and therefore pure realizations
cannot be realized if a symmetry G is broken by VEVs of linear fields and if there are
no gauge symmetries.
When we reformulate NL Ms by auxiliary fields, dynamical fields must be embedded
into fields in some linear representations of G. Therefore the LercheShore theorem is
the most difficulty for the auxiliary field formulation of Khler coset spaces G/H . As
discussed in this paper in detail, this theorem can be avoided introducing appropriate gauge
symmetry.
3. Auxiliary field formulation for compact manifolds

The CP N model and the GN,M model can be easily formulated by the auxiliary field
method as (1.3) and (1.4) without discussing the non-linear realization method [4,5,14].
The non-linear realization method and the super-Higgs mechanism play essential roles to
construct more complicated coset spaces. We can eliminate QNG bosons by gauging a
subgroup of G introducing vector multiplets V . The important thing is that vector multiplets V can become massive absorbing only M-type chiral multiplets including one NG and
4 Only the maximal realization cases have a dual description by a non-Abelian tensor gauge theory in D = 4
[37].
140
one QNG bosons, with preserving SUSY.5 Hence, one V can eliminate one non-compact
direction with one compact direction. If we can eliminate all non-compact directions of
QNG bosons, a compact manifold of the pure realization can be realized. Such consideration for the CP N model and the GN,M model was given in [16,22] and then applied to
HSS [16]. We briefly review the cases of CP N and GN,M in this section, with some new
consideration on the geometry of non-compact manifolds before gauging.
3.1. Auxiliary field formulation for CP N
First, let us discuss CP N 1 . Prepare linear fields N of SU(N ). The system has an
additional phase symmetry U (1)D , given by = ei , and the total global symmetry becomes G = U (N ) = SU(N ) U (1)D . When the fields develop the VEV it can be
transformed by G into = (0, . . . , 0, v)T with v real positive. By this VEV, G is spontaneously broken down to its subgroup H = U (N 1). There appear NNG = dim G/H =
2N 1 NG bosons.
To discuss the whole massless bosons including the QNG bosons, we consider the complexification of the groups. The complex unbroken and broken generators are
0
P
.
..
C .
.
H = U(N 1) . ,
(3.1)
G C H = 0N 1
,
0
P
B
0 0 M
B 0
where B denote generators in a Borel subalgebra B in Eq. (2.1). Here, M and P denote one
Hermitian and N 1 non-Hermitian broken generators, generating M- and P-type chiral
superfields, respectively (NM = 1, NP = N 1). So the numbers of NG and QNG bosons
are NNG = 2N 1 and NQNG = 1, respectively. The number of the QNG bosons coincides
can be
with the number of G-invariant ||2 as was discussed in [36]. The manifold M
locally written as
R
M
SU(N )
U (N )
R
U (N 1)
SU(N 1)
(3.2)
which is cohomogeneity one. (From now on we denote non-compact manifolds before

and compact manifolds by M.)
gauging by M
To gauge away unwanted QNG boson let us promote U (1)D symmetry to a gauge symmetry introducing an auxiliary vector superfield V . The gauge transformation is defined
by
= ei ,
eV eV = eV ei+i ,
(3.3)
with (x, , ) a gauge parameter of a chiral superfield. Note that this gauge symmetry is
complexified to U (1)C = GL(1, C) because the scalar component of is a complex scalar
5 When a gauge field in V absorbs an NG boson in a P-type multiplet, SUSY is spontaneously broken [39].
field. The simplest invariant Lagrangian for these matter contents is written as

L = d 4 eV cV ,
141
(3.4)
with c a real positive parameter called the FayetIliopoulos (FI) parameter.6 If we eliminate
V by its equation of motion eV c = 0, we obtain

4
L = d c log + 1 = d 4 c log ,
(3.6)
where the second term has disappeared under the superspace integral. There still exist the
gauge symmetry (3.3) for matter fields in this Lagrangian which can be fixed as T =
( T , 1) using U (1)C . Then the non-linear Lagrangian in terms of independent fields is
obtained as

L = d 4 c log 1 + .
(3.7)
We thus have obtained the Khler potential for the FubiniStudy metric on CP N 1 .
This construction of CP N is known as the Khler quotient method [51,52]: CP N 1
M/U
(1)C .

Note that can be rewritten as = v with v = (0, . . . , 0, 1)T and = exp 0N1
=
0 0

1N1
being the same as (A.7). Comparing the coset generator in and (3.1), we find
0 1
that one QNG boson is absorbed, with one NG boson, by the U (1) gauge field V and that
a pure realization occurs.
Although we have used the classical equation of motion to eliminate V here, we can
show that this holds in the quantum level using the path integral formalism [17,18].
3.2. Auxiliary field formulation for GN,M
This can be generalized to non-Abelian gauge symmetry to construct GN,M . This case
is a little bit complicated due to non-uniqueness of vacua. Let be an (N M)-matrix
chiral superfield, on which the global symmetry G = SU(N ) SU(M) U (1) as
g
= gL gR ,
(3.8)
with gL SU(N ) and gR U (M) = SU(M) U (1)D . There exist M independent

G-invariants composed of , given by

2

M
,
X2 tr , . . . , XM tr
X1 tr ,
(3.9)
6 The most general invariant Lagrangian is

L=

d 4 f eV cV ,
(3.5)
with an arbitrary function f . However, we can show that we get the same Lagrangian (3.7) below when V is
eliminated [17].
142
because the G-invariants det and tr( )n (n > M) are not independent with these
due to the CayleyHamilton equation for M by M matrices. The G-invariants (3.9) de to be M. The most general invariant
termine the cohomogeneity of the manifold M
Lagrangian is

L = d 4 f X 1 , X 2 , . . . , XM
(3.10)
with an arbitrary function f .
Using G symmetry, generic VEVs can be transformed into the form of
0(N M)M
v
0
1
Vgeneric = =
,
..
.
0
vM
(3.11)
with vi (i = 1, . . . , M) M real positive

These vi s correspond to (VEVs
constants.
j . When all v s differ, G is spontaof) xi s in Eq. (3.9) through xj = M
(v
)
i
j =1 i
neously broken down into Hg SU(N M) U (1)M with ith U (1) generated by
diag(1, . . . , 1, 0, . . . , 0, N + M, 0, . . . , 0) in SU(N ) associated with opposite phase rota

N M
i1
Mi
tion by U (1)D . Hence the number of NG bosons is NNG = dim(G/Hg ) = 2MN M. The
M constants vi in (3.11) are (VEVs of) QNG bosons parametrizing non-compact directions
and hence NQNG = M. Then NNG + NQNG = 2MN = 2 dimC = 2 dimC M

RM in M,
correctly holds. At generic points, the manifold can be locally written as
RM G
M
Hg
= RM
SU(N ) U (M)
SU(N ) SU(M)
RM
,
SU(N M) U (1)M
SU(N M) U (1)M1
(3.12)
and so it is cohomogeneity M. In the last equality, the overall phase rotation is cancelled.
When some vi s coincide, unbroken symmetry is enhanced. For instance, when two of
them coincide vi = vj , the unbroken symmetry is H = SU(N M) U (2) U (1)M2 .
So the number of NG bosons NNG = dim(G/H ) = 2MN 3 is less than the generic
points. Some NG bosons have changed into QNG bosons with total number of massless
bosons unchanged. This phenomenon was found by Shore [34] and was named the SUSY
vacuum alignment. The number of QNG bosons changes from a point to a point, but
the minimum number of QNG bosons realized at the generic points is bounded below by
the number of the G-invariants composed of fundamental fields as found in [36]. So it
determines the cohomogeneity of the manifold.
At the most symmetric point v1 = v2 = = vM v, the VEV becomes

0(N M)M
Vsymmetric = v
(3.13)
.
1M
The unbroken symmetry is H0 SU(N M) SU(M) U (1) generated by

SU(N M) 0(N M)M
, AMM U(1),
H=
0M(N M)
AMM
143
(3.14)
def
with AMM M by M generators in SU(M). Here U (1) is generated by Q1 =

diag(M, . . . , M , N + M, . . . , N + M ) in the SU(N ) generators combined with the op

N M
posite phase rotation by U (1)D with fixing the M M unit matrix in the VEV (3.13). The
number of NG bosons is NNG = dim(G/H0 ) = 2MN M 2 .
At this point complex unbroken and broken generators become

SU(N M)C 0(N M)M
H =
, AMM U(1)C ,
BM(N M)
AMM

P(N M)M
0N M
, 0M U(1)C
G C H =
(3.15)
M.
0M(N M) MMM
Here B represents the matrix of the Borel generators and M and P represent the matrices
of M- and P-type broken generators, respectively, and U (1) in G C H is generated by Q1
itself which is also M-type. So NP = M(N M) and NM = M 2 hold. The numbers of NG
and QNG bosons are NNG = 2NP + NQNG = 2MN M 2 and NQNG = M 2 , respectively.
Here NNG is the least and NQNG is the most. The manifold looks like
RM 2
M
SU(N ) U (M)
,
SU(N M) U (M)
(3.16)
where we denoted a fiber bundle over a base B with a fiber F by F B.

Now let us eliminate unwanted QNG bosons with gauging U (M) by introducing auxiliary vector superfields V taking a value in U(M). The U (M) gauge transformation is given
by
= ei ,
eV eV = ei eV ei ,
(3.17)
with gauge parameters of an M by M matrix chiral superfield. The gauge symmetry is

enhanced to its complexification U (M)C = GL(M, C). Using SU(N ) GL(M, C)local , the
VEV can be taken as

0(N M)M
.
Vgauge =
(3.18)
1M
Since this VEV takes the form of (3.13) with v = 1, breaking pattern is the same with
(3.15). The most symmetric point is realized at which the number of QNG bosons is the
maximum NQNG = M 2 with coinciding with the dimension of the U (M) gauge group.
The simplest invariant Lagrangian is

L = d 4 tr eV c tr V ,
(3.19)
144
with c real positive.7 The equation of motion for V

eV c1M = 0
(3.21)
can be used to eliminate V . Solving this equation as V = log( /c) and substituting
this back into the original Lagrangian (3.19), we obtain

L = d 4 c log det ,
(3.22)
where a constant has disappeared under the superspace integral. This still has the
GL(M, C) gauge symmetry (3.17) for matter fields. We can fix this gauge degree of freedom as

=
(3.23)
1M
with an N M by M matrix of chiral superfields. Therefore, we obtain

L = d 4 c log det 1M + .
(3.24)
This is the Khler potential for the Grassmann manifold GN,M . In terms of Khler quotient
(M)C .
we can write GN,M M/U

0
NM
Note that in (3.23) can be rewritten as = Vgauge using = exp 0M(NM)
0M =

1NM
in (A.10) and Vgauge in (3.18). Comparing these coset generators and

0M(NM) 1M
Eq. (3.15), we conclude that M 2 QNG bosons at the most symmetric point are absorbed,
with the same number of NG bosons, by the U (M) gauge fields V and that a pure realization occurs. The less symmetric points do not contribute to the resultant manifold.
The U (M) gauge symmetry can be replaced with the U (N M) gauge symmetry with
the same SU(N ) global symmetry considering an N by N M matrix, due to the duality
GN,M GN,N M .
4. Auxiliary field formulation of SU(N)/[SU(N 2) U (1)2 ]
Generalizing the discussion in CP N , we formulate the rank-two Khler coset space
SU(N )/[SU(N 2) U (1)2 ] by the auxiliary field method. In the first subsection, we
construct a non-compact Khler manifold putting a holomorphic constraint by an auxiliary
chiral superfield without gauging. Then in the second subsection, we obtain a compact
manifold by gauging U (1)2 part of the isometry of the non-compact manifold introducing
auxiliary vector superfields with two FI-parameters.
7 The most general invariant Lagrangian is

L=

2
M
d 4 f tr eV , tr eV , . . . , tr eV
c tr V
(3.20)
with an arbitrary function f . However, we think that the resulting Lagrangian after eliminating V coincides with
the simplest case although we do not have a proof.
145
4.1. Non-compact Khler manifold

Let 1 (x, , ) and 2 (x, , ) being column vectors of chiral superfields, belonging
of SU(N ), respecto the fundamental and the anti-fundamental representations, N and N,
tively. They transform under SU(N ) as
T

1 1 = g1 ,
(4.1)
2 2 = g 1 2 ,
where g SU(N ) is a matrix element of the fundamental representation.8 We put the constraint invariant under SU(N ),9
1 2 = 0.
(4.2)
There exists additional
U (1)2
symmetry,
i i = eii i
(4.3)
with i (i = 1, 2) being real parameters. So the total global symmetry is G SU(N )

U (1)2 . There exist two G-invariants |1 |2 and |2 |2 so the manifold is cohomogeneity two.
The most general invariant Lagrangian is given by

4
2
L = d f 1 1 , 2 2 +
(4.4)
d 1 2 + c.c. ,
with f an arbitrary function. Here (x, , ) is an auxiliary chiral superfield belonging to
the singlet of SU(N ), whose U (1)2 transformation is given by = ei(1 +2 ) .
Using the G symmetry, the VEVs can be taken as
v1 = 1 = ( 0, . . . , 0, v)T ,

N 1
v2 = 2 = ( 0, . . . , 0, u, 0)T ,

(4.5)
N 2
with v and u real positive. By these VEVs, G = SU(N ) U (1)2 is spontaneously broken
down to H = SU(N 2) U (1)2 . Its generators are given by
0 0
. .
SU(N 2) .. ..
U(1) U(1),
H=
(4.6)
0 0
0 0 0 0
0 0 0 0
with two U (1)s given by generators
Q1 diag( 1, . . . , 1, 2 N, 0),

N 2
Q2 diag( 1, . . . , 1, 0, 2 N ),

(4.7)
N 2
8 (g 1 )T is not equivalent to g when we consider complex extension of the group for the symmetry of the
superpotential. We should define the transformation law of the anti-fundamental representation by the former.
9 The constraint = a 2 , with a being a real constant, was discussed in [36]. In this case, one of U (1)
1
2
symmetries (4.3) is explicitly broken.
146
in SU(N ) accompanied with the 2 - and 1 -phase rotations (4.3) with opposite angles,
with fixing v2 and v1 , respectively. The number of NG bosons is NNG = dim(G/H ) =
4N 4.
To discuss QNG bosons, we consider the action of the complexification of G. The complex unbroken and broken generators are
B 0
. .
SU(N 2)C .. ..
U(1)C U(1)C ,
H =
B 0
0 0 0 0
B
B B 0
..
0N 2
.
C
C
U(1)C
G H =
M U(1)M ,
0 P
P P 0 P
0 0 0 0
0
..
.
(4.8)
where B denote generators in a Borel subalgebra, P denote non-Hermitian broken generators P-type chiral superfields and two U(1)C
M s are given by Q1 and Q2 defined in
Eq. (4.7), both of which generate M-type chiral superfields. The numbers of the P- and the
M-type superfields are NP = 2N 3 and NM = 2, respectively. Therefore, there appear
NNG = 4N 4 NG bosons, whose number coincides with the dimension of G/H , and
NQNG = 2 QNG bosons. The number of the QNG bosons coincides with the number of the
G invariants, |1 |2 and |2 |2 , as discussed in [36]. The manifold can be locally written as
R2
M
SU(N ) U (1)2
SU(N )
R2
2
SU(N 2)
SU(N 2) U (1)
(4.9)
which is cohomogeneity two.

4.2. Gauging: SU(N )/[SU(N 2) U (1)2 ]
To eliminate unwanted QNG bosons and to obtain a compact Khler manifold, we promote U (1)2 symmetry (4.3) to a gauge symmetry introducing auxiliary vector superfields
Vi (x, , ) (i = 1, 2). The gauge symmetry U (1)1 U (1)2 is defined by
i i = eii i ,
= ei1 i2 ,
eVi eVi = eVi eii +ii ,

(4.10)
with i (x, , ) being a chiral superfield as a gauge parameter of U (1)i . We thus obtain the
invariant Lagrangian for the auxiliary field formulation of SU(N )/[SU(N 2) U (1)2 ]
as

V
4
V2
2
1
L = d e 1 1 + e 2 2 c1 V1 c2 V2 +
d 1 2 + c.c. .
(4.11)
147
To confirm if this really gives a compact Khler coset space, we now eliminate auxiliary
superfields , V1 and V2 by their equations of motion. The integration over gives the
holomorphic constraint
1 2 = + + = 0,
in which we have used the notation

1 = ,
2 =
,
(4.12)
(4.13)
with and N 2 vectors and the rests singlets. Introducing a new chiral superfield
(x, , ), we rewrite this constraint as
2 = 2,
2 = 2.
(4.14)
If we eliminate from these equations, we get the constraint (4.12) again. Instead, we
solve and by other fields:

1

1
+
,
=
.
=
(4.15)
2
The integration over Vi gives eVi i i ci = 0, which can be solved as

Vi = log i i /ci .
(4.16)
Substituting these solutions and (4.15) back into the Lagrangian (4.11), we obtain

L=
d 4
2

ci log i i
i=1

1
2
4
2
2

= d c1 log || + || + +

2

2

1

,
+ c2 log ||2 + ||2 +

2
(4.17)
where additional constant has disappeared under the integration over the superspace. Since
this still has gauge invariance (4.10) of U (1)2 for matter fields, we can take a gauge of
= = 1:

2
4
2

L = d c1 log 1 + || + +
2
2

.
+ c2 log 1 + ||2 +
(4.18)
2
This is the Khler potential of SU(N )/[SU(N 2) U (1)2 ] with a suitable complex
structure [31] (see Eq. (A.17) in Appendix A).
148
To see the relation with the non-compact case, we note the coset representative is given
by
1N 2 0
= eZ = T 1 + 12
0 0
1
0N 2 0
with Z = T 0 ,
0 0 0
(4.19)
where Z represent complex broken generators, and = {, , }. Using this representative, the superfields, after solving the constraint and fixing a gauge, can be written as
T

1 = v1 ,
(4.20)
2 = 1 v2 ,
where vi are VEVs given in (4.5) with u = v = 1. By comparing the generators (4.8) and
(4.19), it is now obvious that two M-type superfields are eliminated by gauging of U (1)2 .
From the structure of generators (4.19), we find that this manifold has one of two complex
structures, called II , for SU(N )/[SU(N 2) U (1)2 ] [25].
If we forget the superpotential term in the Lagrangian (4.11), it becomes the auxiliary field formulation for CP N 1 CP N 1 . Therefore SU(N )/[SU(N 2) U (1)2 ] is
embedded into CP N 1 CP N 1 by a holomorphic constraint 1 2 = 0 so that it is
algebraic.
5. Auxiliary field formulation of

SU(N )/[SU(N M L) SU(M) SU(L) U (1)2 ]
In this section we show the results of the last section can be generalized to the auxiliary
field formulation for SU(N )/[SU(N M L) SU(M) SU(L) U (1)2 ] by promoting
gauge groups to non-Abelian groups.
5.1. Non-compact Khler manifold
Let A (x, , ) (A = 1, . . . , M) and (x, , ) ( = 1, . . . , L) be chiral superfields
belonging to the fundamental and the anti-fundamental representations of SU(N ), respectively (N L + M). The transformation law under SU(N ) is the same as (4.1). We define
matrix chiral superfields (1 , . . . , M ) and (1 , . . . , L ). Additional global symmetries of U (M) and U (L) act from the right of and , respectively, as
= g1 ,
g1 U (M),
g2 U (L).
= g2 ,
(5.1)
The total global symmetry is G = SU(N ) U (M) U (L).

We impose LM holomorphic, GC invariant constraints on these fields:
A = 0 or
T = 0LM .
There exist M + L G-invariants

2
X1 tr ,
X2 tr ,
(5.2)
...,

M
XM tr
,

2
Y2 tr ,

Y1 tr ,
...,

L
YL tr
,
149
(5.3)
so that this manifold is cohomogeneity M + L.10 The most general invariant Lagrangian is

L=
d 4 f (X1 , . . . , XM , Y1 , . . . , YL ) +

d 2 tr T + c.c. ,
(5.4)
with an M L matrix of auxiliary chiral superfields A (x, , ) of Lagrange multipliers

for the constraints (5.2).
Using G = SU(N ) U (M) U (L), the generic VEVs can be transformed into
0(N M)M
v
0
1
V1 = =
,
..
.
0
vM
V2T = T = 0L(N ML)
u1
0
..
0LM ,
(5.5)
uL
with vi (i = 1, . . . , M) and ua (a = 1, . . . , L) M + L real positive constants.

These VEVs

(v
)i and Ya =
are related with VEVs of the invariants Xi and Ya through Xi = M
j
j =1
L
a
M+L and
b=1 (ub ) . G is spontaneously broken down to Hg = SU(N M L) U (1)
NNG = dim(G/Hg ) = 2(N M + LN LM) M L. The number of the QNG bosons
is NQNG = M + L. So equations NNG + NQNG = 2(N M + LN LM) = 2(dimC +
correctly holds. At generic points, the manifold can be locally
dimC LM) = 2 dimC M
written as
RM+L G = RM+L SU(N ) U (M) U (L)
M
Hg
SU(N M L) U (1)M+L
RM+L
SU(N ) SU(M) SU(L)

,
SU(N M L) U (1)M+L2
(5.6)
so that it is cohomogeneity M + L.
At the most symmetric points, the VEVs become
0
V1 = = v (N M)M
1M

,
V2T = T = u(0L(N ML) , 1L , 0LM ).
(5.7)
10 The invariants made by traces of products M and N are probably not independent of these
invariants although we do not have a proof. It is plausible considering the relation with VEVs as below.
150
The unbroken subgroup is H0 = SU(N M L) SU(M) SU(L) U (1)2 whose

generators are given by
SU(N M L) 0(N ML)L 0(N ML)M
XLL
0LM
H = 0L(N ML)
, XLL , YMM
0M(N ML)
0ML
YMM
U(1) U(1),
(5.8)
with X and Y generators of SU(L) and SU(M), respectively. Here the two U (1)s are
given by
Q1 diag( L, . . . , L, N + M + L, . . . , N + M + L, 0, . . . , 0 ),

N ML
Q2 diag( M, . . . , M , 0, . . . , 0, N + M + L, . . . , N + M + L ),

N ML
(5.9)
in SU(N ) combined with phase rotations with opposite angles by U (1)1 and U (1)2 , respectively.
Complex unbroken and broken generators become like
SU(N M L)C B(N ML)L 0(N ML)M
H = 0L(N ML)
XLL
0LM
, XLL , YMM
BM(N ML)
BML
YMM ,
U(1)C U(1)C ,
0N ML 0(N ML)L P(N ML)M
MLL
PLM
G C H = PL(N ML)
, 0LL , 0MM
0M(N ML)
0ML
MMM
C
U(1)C
M U(1)M ,
(5.10)
respectively, where each subscript denotes the size of each block, and two U (1)s in G C
H are generated by Q1 or Q2 in Eq. (5.9), both of which are M-types. The numbers of the
M- and P-type superfields are NM = L2 + M 2 and NP = M(N M) + L(N M L),
respectively. There appear NNG = 2(N M + N L L) M 2 L2 NG bosons and NQNG =
(L2 + M 2 ) QNG bosons. The manifold looks like
RM 2 +L2
M
SU(N ) U (M) U (L)

.
SU(N M L) U (M) U (L)
(5.11)
5.2. Gauging: SU(N )/[SU(N M L) SU(M) SU(L) U (1)2 ]

To absorb M-type superfields, we promote the global symmetries of (5.1) to gauge symmetries, by introducing auxiliary vector superfields V1 and V2 of U (M) and U (L) gauge
fields, respectively. These U (M) U (L) gauge symmetries are defined by
= ei1 ,
= ei2 ,

eV1 eV1 = ei1 eV1 ei1 ,
151
eV2 eV2 = ei2 eV2 ei2 ,
= ei1 ei2 ,
T
(5.12)
where 1 (x, , ) and 2 (x, , ) are chiral superfields of gauge parameters, taking values
in U(M) and U(L), respectively. Then the Lagrangian can be written as

L = d 4 tr eV1 + tr eV2 c1 tr V1 c2 tr V2

+
d 2 tr T + c.c. .
(5.13)
We decompose the matrix chiral superfields into submatrices, like

I A
I
= A ,
=
,
AB
A
(5.14)
where a new index I , which runs from 1 to N M L, has been introduced. In terms of
these decompositions, the integration over A yields the LM holomorphic constraints
T
T
T
A = I
I A +
A + B
BA = 0,
(5.15)
in which the summation over repeated indices is implied. Introducing a new L M matrix
chiral superfield, A (x, , ), these constraints can be rewritten as
T
T
2
A I
I A = 2A ,
T
T
2B
BA I
I A = 2A .
(5.16)
Elimination of from these equations gives the original constraints (5.15). Instead, we
express and by other fields in the region such that det = 0 and det = 0 hold:

1
1
= 1T + T ,
(5.17)
= 1T T ,
2
2
where we have used the matrix notation.
The equations of motion for V1 and V2 read eV1 c1 1M = 0 and eV2 c2 1L =
0, respectively. These equations can be solved, to give

,
V2 = log det
.
V1 = log det
(5.18)
c1
c2
Substituting these solutions back into the Lagrangian (5.13), we obtain

L = d 4 c1 log det + c2 log det

= d 4 c1 log det + + + c2 log det + + .
(5.19)
152
By substituting (5.17) into this Lagrangian and taking a gauge fixing of = 1M and =
1L , we obtain

1
1
L = d 4 c1 log det 1M + + + + T
2
2

1
1 T
.
+ c2 log det 1L + +
(5.20)

2
2
We thus have obtained the Khler potential of SU(N )/[SU(N M L) U (M) U (L)]
with a suitable complex structure through a Khler quotient: M M/[U

(M)C U (L)C ].
Since the Lagrangian (5.13) becomes the auxiliary field formulation for GN,M GN,L
throwing away its superpotential, we find that SU(N )/[SU(N M L) U (M) U (L)]
is embedded into GN,M GN,L by holomorphic constraints by T = 0LM .
An interesting thing is that there exists the triality between theories with the same U (N )
flavor symmetry and following three different gauge groups: U (M) U (L), U (N M
L) U (L) and U (M) U (N M L).
6. Summary and discussions

We have given the auxiliary field formulation for D = 2, 3, N = 2 SUSY NL Ms on
the rank two Khler coset spaces SU(N )/[SU(N 2) U (1)2 ] and SU(N )/[SU(N
M L) SU(M) SU(L) U (1)2 ] as U (1)2 and U (M) U (L) gauge theories with
the Lagrangians (4.11) and (5.13), respectively. In addition to auxiliary vector superfields
V1 and V2 for these gauge groups, we have also needed auxiliary chiral superfields to
give holomorphic constraints among two irreducible representations. For both cases the
Lagrangian includes two FI parameters c1 and c2 which represent two free parameters
(decay constants) of the resultant Khler coset spaces. Non-perturbative analyses using
the large-N method for these new models in D = 2, 3 dimensions have become possible,
which remains as a future work.
Let us discuss possibility to construct more general models by the auxiliary field
method. First of all the Khler coset spaces discussed in this paper have particular complex structure II (see Appendix A). However, in general, the same rank-two coset space
allows two inequivalent complex structures I and II as discussed in Appendix A. So the
question is whether it is possible to construct a model with another complex structure I
or not. In the case of SU(N )/[SU(N 2) U (1)2 ] with I , it is natural to prepare the two
sets of fields 1 and 2 both belonging to the fundamental representation N of SU(N ), in for II . Actually, the authors in Ref. [40] constructed a bosonic NL M on
stead of N and N
SU(N )/[SU(N 2) U (1)2 ] with the complex structure I by the auxiliary field method.
They introduced an U (2) gauge symmetry explicitly broken by two constraints 1 1 = c1
and 2 2 = c2 with c1 = c2 . These constraints can be embedded into bosonic parts of the
SUSY U (1)2 gauge theory with two FI-parameters c1 and c2 in the WessZumino gauge.
They moreover imposed an additional constraint 1 2 = 0 instead of 1 2 = 0 for I . It
is, however, difficult to embed this constraint to a SUSY theory.
153
Second, let us discuss rank-two coset spaces G/H with other groups G. For rankone case it was possible to reduce G = SU(N ) to other groups G G imposing G invariant F-term (holomorphic) constraints by auxiliary chiral superfields which give an
embedding for the whole coset G /H G/H , as in the Lagrangian (1.5) for QN 2 =
SO(N )/[SO(N 2) U (1)] CP N 1 . In principle, it should work also for rank-two
cases but naive attempts failed. This remains for a future work.
Third, we would like to discuss higher-rank coset spaces. The LercheShore theorem
[29,30] implies that any Khler G/H needs a gauge group to be formulated by linear
fields. It is natural for a gauge group to include the U (1)r factor with r FI-parameters for
a rank-r coset space. Moreover we should introduce at least r irreducible representations
of G and put suitable F-term constraints among them which are needed to fix all GC invariants composed of them. However the similar problem for SU(N )/[SU(N 2)
U (1)2 ] with I occurs and we are unable to achieve this in the present time.
Perturbatively, dynamics of D = 2, N = 2 SUSY NL Ms have very different features
according to their first Chern classes c1 (M) on the target manifold M. CalabiYau manifolds M have vanishing first Chern classes c1 (M) = 0. D = 2, N = 2 SUSY NL Ms on
these manifolds are conjectured to be finite to all orders and are considered as models of
superstring theory [41]. On the other hand, all Khler coset spaces have positive first Chern
classes: c1 (G/H ) > 0. For all NL Ms on M with c1 (M) > 0, it is conjectured that these
models are asymptotically free and have the mass gap like D = 4 QCD.
Non-perturbative analyses for these features were discussed extensively using the mirror
symmetry [42]. Exact beta functions for D = 2 NL Ms on Hermitian symmetric spaces
were derived using the instanton method [43]. D = 2, 3, N = 2 SUSY NL Ms were also
discussed using the Wilsonian renormalization group (WRG) [44], in which they used
the so-called Khler normal coordinates [45] to derive the WRG equation. In particular,
D = 2, 3, N = 2 SUSY NL Ms on the KhlerEinstein manifolds including Khler coset
spaces were discussed.
Combined with these several methods, we expect that the large-N method plays an
important role to reveal non-perturbative aspects of SUSY NL Ms and their similarity
with D = 4, 5 QCD.
Acknowledgements
The author thanks Kiyoshi Higashijima and Makoto Tsuzuki for a collaboration in early
stages of this work. His work is supported by the US Department of Energy under grant
DE-FG02-91ER40681 (task B).
Appendix A. Pure realizations for G = SU(N) cases

We work out theories with the global symmetry SU(N ). For details see the original
Refs. [21,24] or the reviews
with G = SU(N ) occur if and only
[22,26]. Pure realizations

r+1
r with
if H is the form of H = r+1
SU(n
)
U
(1)
i
i=1
i=1 ni = N and r( 1) called the
154
rank of this Khler coset space G/H :
SU(n1 )
SU(n2 )
H=
..
U(1) U(1) .

(A.1)
SU(nr+1 )
All off-diagonal blocks are zero matrices and r U (1) generators Q ( = 1, . . . , r) are
given by
Q = diag( n+1 , . . . , n+1 , 0, . . . , 0, n1 , . . . , n1 0, . . . , 0 ).

n1
=2 n
n+1
(A.2)
r+1
=+2 n
A complex structure on G/H is defined as follows. Taking linear combination of U (1)

generators Q in H, we define the Y -charge by
Y=
r

c Q
(A.3)
=1

with c R. Complex linear combination I CI XI (CI C) of the real coset generators
XI G H can be divided into Bi H and Zi G C H according to positive and negative Y -charges, respectively: [Y, Bi ] +Bi and [Y, Zi ] Zi . Note that the generators
in H carry zero Y -charges and therefore all generators in H carry non-negative Y -charges.
Thus, rank one coset spaces have one complex structure. Rank (more than) two coset spaces
have (more than) two inequivalent complex structures. The complex coset representative
can be defined by = exp( Z) GC /H with i NG chiral multiplets whose scalar
components are both genuine NG bosons. There exists homomorphism GC /H G/H for
pure realizations because there are no QNG bosons.
Once a complex structure is provided we can construct a G-invariant Khler potential
on G/H . There exist r projection operators ( = 1, . . . , r) satisfying the conditions
H = H ,
2 = ,
(A.4)
in the space of the fundamental representation N of G. We can take these according to

the Y -charges of the fundamental representation space as follows: first decompose the fundamental representation of G-into H -irreducible sectors. Second take the th projection
to project out the first sectors with lower Y -charges from the lowest Y -charge sector.
Using these projections, the Khler potential on G/H is given by
K=
r

c log det ,
(A.5)
=1
with GC /H in the fundamental representation and det denoting the determinant in

the subspace projected by . Here c can be shown to coincide with the coefficients in the
Y -charge (A.3) [24].
We give the following four examples discussed in this paper: (1) CP N 1 = SU(N )/
[SU(N 1) U (1)], (2) GN,M = SU(N )/[SU(N M) SU(M) U (1)], (3) SU(N )/
[SU(N 2) U (1)2 ] and (4) SU(N )/[SU(N M L) SU(M) SU(L) U (1)2 ].
155
(1) CP N 1 = SU(N )/[SU(N 1) U (1)].

This is the simplest example called the projective space. If we define the Y -charge by Y
Q1 with Q1 diag( 1, . . . , 1, N + 1) H, complex unbroken and broken generators,

N 1
carrying positive and negative (or zero) Y -charges, are found to be
H = SU(N 1)
0
..
.
U(1)C ,
0
G C H =
0N 1
P
..
. ,
(A.6)
0 0 0
B 0
respectively, with U(1)C generated by Q1 . Then the complex coset representative is given
by

1N 1
0N 1
,
=
Z=
(A.7)
0 0
0 1
with an (N 1)-vector of SU(N 1). The projection operator is = diag( 0, . . . , 0, 1)

N 1
and the Khler potential is

K = c log det = c log 1 + ||2 .
(A.8)
(2) GN,M = SU(N )/[SU(N M) SU(M) U (1)].

This is called the (complex) Grassmann manifold. We take the Y -charge as Y Q1
with Q1 = diag( M, . . . , M , M N, . . . , M N ) H. Then complex unbroken and bro

N M
ken generators are given by

SU(N M)C 0(N M)M
U(1)C ,
BM(N M) SU(M)C

P(N M)M
0N M
,
G C H =
0M(N M)
0M
H =
(A.9)
respectively, with U(1)C generated by Q1 . The complex coset representative is obtained as

1N M
0N M
,
=
Z=
(A.10)
0M(N M) 0M
0M(N M) 1M
with an N M by M matrix. The projection operator is = diag( 0, . . . , 0, 1, . . . , 1 )

N M
and the Khler potential is

K = c log det = c log detMM 1M + .
(A.11)
156
(3) SU(N )/[SU(N 2) U (1)2 ].

This coset space allows two inequivalent complex structures [25,31]. The U (1) generators
in H are
Q1 diag( 1, . . . , 1, N + 2, 0),

N 2
Q2 diag( 1, . . . , 1, 0, N + 2).

(A.12)
N 2
Two complex structures I and II denoted in [25] are represented, for instance, by YI
Q2 and YII Q1 Q2 , respectively. According to these Y -charges, complex unbroken
and broken generators are given by
P P
0 0
.. ..
. .
0N 2
. .
SU(N 2)C .. ..

C
C
P P
2U(1) ,
HI =
G H I =
0 0
,
B B 0 0
0
0
0
P
0 0 0 0
B B 0
for I and
0
..
SU(N 2)C
.
H II =
2U(1)C ,
B
0
0 0 0 0
B B B 0
B
..
.
P
..
0N 2
.
C

G H II =
,
0
P
P P 0 P
0 0 0 0
0
..
.
for II . For both cases two U (1) generators are given by Eq. (A.12). The coset representative for I is calculated as
1N 2 + 12
0N 2
( Z)I = 0 0 ,
(A.13)
I = 0 1
,
0 0 0
0 0
1
with and belonging to N 2 of SU(N 2). The coset representative for II is
0N 2 0
( Z)II = T 0 ,
0 0 0
1N 2 0
II = T 1 + 12
0 0
1
(A.14)
with ( ) belonging to N 2 (N 2). The projection operators are given by

(1 )I = diag( 0, . . . , 0, 0, 1),

(2 )I = diag( 0, . . . , 0, 1, 1),

(A.15)
(1 )II = diag( 0, . . . , 0, 0, 1),

(2 )II = diag( 1, . . . , 1, 0, 1),

(A.16)
N 2
N 2
N 2
N 2
157
and the Khler potentials can be calculated as [31]

1 2
2

KI = c1 log 1 + || + +
2

2
1 2
+ c2 log 1 + ||2 + + ||2 ||2 ,
2
2
2

1
1
KII = c1 log 1 + ||2 + + + c2 log 1 + ||2 + ,
2
2
(A.17)
for I and II , respectively. The one formulated by the auxiliary field method is the second
one.
(4) SU(N )/[SU(N M L) SU(M) SU(L) U (1)2 ].
Two U (1) generators in H are given by11
Q1 diag( L, . . . , L, N + M + L, . . . , N + M + L, 0, . . . , 0 ),

N ML
Q2 diag( M, . . . , M , 0, . . . , 0, N + M + L, . . . , N + M + L ).

N ML
(A.18)
As the same with the last example, the Y -charge can be taken, for instance, as YI Q2 or
YII MQ1 LQ2 for the complex structure I or II , respectively. Complex unbroken
and broken generators are given by
SU(N M L)C 0(N ML)L 0(N ML)M
C
SU(L)C
0LM
H I = BL(N ML)
2U(1) ,
BM(N ML)
BML
SU(M)C
0N ML P(N ML)L P(N ML)M

C
0L
PLM
G H I = 0L(N ML)
(A.19)
,
0M(N ML)
0ML
0M
for I and
SU(N M L)C B(N ML)L 0(N ML)M
C
SU(L)C
0LM
H II = 0L(N ML)
2U(1) ,
BM(N ML)
BML
SU(M)C
0N ML 0(N ML)L P(N ML)M

C
0L
PLM
G H II = PL(N ML)
,
0M(N ML)
0ML
0M
11 We have chosen these charges in a different way from those in Refs. [22,48].
(A.20)
158
for II . The complex coset representatives can be calculated as
0N ML
1N ML + 12
,
0
0L ,
I =
( Z)I =
(A.21)
0
1L
0
0 0M
0
0
1M
1N ML 0
0N ML 0
( Z)II = T
II = T
0L ,
1L + 12 T , (A.22)
0
0 0M
0
0
1M
for I and II , respectively. The projection operators are given by
(1 )I = diag( 0, . . . , 0, 0, . . . , 0, 1, . . . , 1 ),

N ML
(2 )I = diag( 0, . . . , 0, 1, . . . , 1, 1, . . . , 1 ).

N ML
(A.23)
for I and
(1 )II = diag( 0, . . . , 0, 0, . . . , 0, 1, . . . , 1 ),

N ML
(2 )II = diag( 1, . . . , 1, 0, . . . , 0, 1, . . . , 1 ),

N ML
(A.24)
for II . Using these projection operators, the Khler potentials are calculated as

1
1
KI = c1 log det 1M + + + +
2
2
1L +
+ ( + 12 )
+ c2 log det
1
1
1
+ ( + 2 ) 1M + + ( + 2 )( + 2 )

1
1 T
KII = c1 log det 1M + + +

+
2
2
T
1N ML +
+ ( + 12 T )
+ c2 log det
1 T
1
+ ( + 2 )

,
1M + + ( + 2 )( + 12 T )

,
(A.25)
for I and II , respectively.
The second one should coincide with the Khler potential (5.20) up to a holomorphic
coordinate transformation and a Khler transformation, but we are unable to show their
equivalence.
Appendix B. Some geometric structures

In this appendix, we discuss more geometric structures of the manifolds presented in
this paper; their bundle structures and their relation with hyper-Khler manifolds and the
CalabiYau manifolds of cohomogeneity one [4649].
159
First we discuss the bundle structures using gauging/ungauging technique [47]. If we

introducing vector superfields V and then integrate
gauge a part of isometry I G on M
C . On the other hand, if we unV out, we obtain a Khler quotient manifold M = M/I
In the case
gauge I by freezing V in a Khler quotient formulation of M, we obtain M.
M
of I = U (1) [I = U (M)], M can be regarded as a complex line (C -)bundle over M. By

applying this technique to Hermitian symmetric spaces (HSS) formulated as gauge theories
[16], canonical complex line bundles over HSS are constructed in [47].
We now discuss relations between manifolds in this paper with other manifolds. First let
us consider SU(N )/[SU(N 2) U (1)2 ]. If we put V V1 = V2 (or freezing V1 + V2 )
in the Lagrangian (4.11) we get

L=

d 4 eV 1 1 + eV 2 2 cV +

d 2 1 2 + c.c. ,
(B.1)
with c c2 c1 . SUSY is enhanced to D = 2, N = 4 (D = 4, N = 2) SUSY. For this

SUSY, target space must be hyper-Khler (HK) [50]. Actually, Eq. (B.1) is the HK quotient
construction [51,52] for the HK Calabi metric [53] on the cotangent bundle over CP N 1 ,
T CP N 1 [54,55]. Therefore SU(N )/[SU(N 2) U (1)2 ] is a U (1) Khler quotient
of T CP N 1 , and the latter is a complex line bundle over the former. T CP N 1 is the
only one HK manifold of cohomogeneity one [56]. In our method, the cohomogeneity of
T CP N 1 is easily found to be one, because we construct it from a compact manifold by
freezing one gauge degree of freedom.
Next let us consider non-Abelian case of M = L: SU(N )/[SU(N 2M) SU(M)2
U (1)2 ]. If we put V V1 = V2T in the Lagrangian (5.13) with M = L, we get

L=

d tr eV + tr T eV c tr V +
4

T
d tr + c.c. ,
2
(B.2)
with c c2 c1 . As the same as the above discussions, this is the HK quotient construction
for the LindstrmRocek metric on T GN,M [51,57]. (Putting = 0 in the Lagrangian
(B.2) we obtain the Lagrangian (1.4) for GN,M , and therefore we find this bundle structure.) Thus SU(N )/[SU(N 2M) U (M)2 ] is a U (M) Khler quotient of T GN,M , and
the latter is a CM -bundle over the former.
Before closing this appendix, we discuss possibility to construct a new CalabiYau (CY)
metric of cohomogeneity one [4649]. If we freeze out V1 + V2 in SU(N )/[SU(N 2)
U (1)2 ] starting from the most general Lagrangian, we obtain

L=

d 4 eV 1 1 + f eV 2 2 cV +

d 2 1 2 + c.c. ,
(B.3)
with f an arbitrary function.12 This is a deformation of HK metric (B.1) preserving only

12 We can take this Khler potential as the most general one instead of f (eV , eV ) because it can
1 1
2 2
be shown that one variable can be linearized when V is integrated [17].
160
the Khler structure. If we freeze out aV1 + bV2 (with a, b R) in SU(N )/[SU(N 2)
U (1)2 ], we get

V
qV

4
2
L = d e 1 1 + f e 2 2 cV +
(B.4)
d 1 2 + c.c. ,
with q a/b being the relative charge of the remained U (1) gauge for 1 and 2 , and
c c1 + qc2 . This is the Lagrangian suggested in [49] ((A.2) in Appendix A) as a generalization of the construction of a CY metric using matter coupling in the CP N model.
Integrating V and we obtain

q

L = d 4 log 1 1 + h 1 1
(B.5)
2 2
with the constraint 1 2 = 0 and some function h related with f . In the case of q = 0, the
CY metric was obtained in [49] solving the Ricci-flat condition for h. The case of q = 1
corresponds to the HK Calabi metric. If we solve it for general q, we will be able to obtain
the most general CY metric on the line bundle over SU(N )/[SU(N 2) U (1)2 ], if it
exists, which is cohomogeneity one and can be locally written as
R
SU(N )
,
SU(N 2) U (1)
(B.6)
where freedom to embed U (1) H into SU(N ) corresponds to q. The holonomy structure
for the case of N = 3, R SU(3)/U (1), was discussed in detail in [58] including Spin(7)
holonomy.
References
[1] A.M. Polyakov, Gauge Fields and Strings, Harwood Academic, Reading, UK, 1987;
S. Coleman, Aspects of Symmetry, Cambridge Univ. Press, Cambridge, 1985;
E. Abdalla, M.C.B. Abdalla, K.D. Rothe, Non-Perturbative Methods in 2-Dimensional Quantum Field Theory, World Scientific, Singapore, 1991;
K. Higashijima, Prog. Theor. Phys. Suppl. 104 (1991) 1.
[2] A. DAdda, M. Lscher, P. Di Vecchia, Nucl. Phys. B 146 (1978) 63.
[3] E. Witten, Phys. Rev. D 16 (1977) 2991;
O. Alvarez, Phys. Rev. D 17 (1978) 1123.
[4] E. Witten, Nucl. Phys. B 149 (1979) 285.
[5] A. DAdda, P. Di Vecchia, M. Lscher, Nucl. Phys. B 152 (1979) 125.
[6] K. Higashijima, T. Kimura, M. Nitta, M. Tsuzuki, Prog. Theor. Phys. 105 (2001) 261, hep-th/0010272.
[7] I.Ya. Arefeva, Ann. Phys. 117 (1979) 393;
I.Ya. Arefeva, S.I. Azakov, Nucl. Phys. B 162 (1980) 298;
B. Rosenstein, B.J. Warr, S.H. Park, Nucl. Phys. B 336 (1990) 435.
[8] V.G. Koures, K.T. Mahanthappa, Phys. Rev. D 43 (1991) 3428;
V.G. Koures, K.T. Mahanthappa, Phys. Rev. D 45 (1992) 580.
[9] M. Ciuchini, J.A. Gracey, Nucl. Phys. B 454 (1995) 103, hep-th/9508176;
T. Inami, Y. Saito, M. Yamamoto, Prog. Theor. Phys. 103 (2000) 1283, hep-th/0003013;
T. Inami, Y. Saito, M. Yamamoto, Phys. Lett. B 495 (2000) 245, hep-th/0008195;
T. Inami, Y. Saito, M. Yamamoto, Mod. Phys. Lett. A 16 (2001) 1643, hep-th/0106204.
[10] S. Coleman, Commun. Math. Phys. 31 (1973) 259.
[11] B. Zumino, Phys. Lett. 87B (1979) 203.
161
[12] J. Wess, J. Bagger, Supersymmetry and Supergravity, second ed., Princeton Univ. Press, Princeton, 1992.
[13] N. Seiberg, E. Witten, Nucl. Phys. B 426 (1994) 19, hep-th/9407087;
N. Seiberg, E. Witten, Nucl. Phys. B 431 (1994) 484, hep-th/9408099;
P.C. Argyres, M.R. Plesser, N. Seiberg, Nucl. Phys. B 471 (1996) 159, hep-th/9603042.
[14] S. Aoyama, Nuovo Cimento A 57 (1980) 176.
[15] J.S. Schwinger, Phys. Rev. 125 (1962) 397.
[16] K. Higashijima, M. Nitta, Prog. Theor. Phys. 103 (2000) 635, hep-th/9911139.
[17] K. Higashijima, M. Nitta, Prog. Theor. Phys. 103 (2000) 833, hep-th/9911225.
[18] K. Higashijima, M. Nitta, in: H. Suganuma, et al. (Eds.), Proceedings of Confinement 2000, World Scientific,
Singapore, 2001, pp. 279286, hep-th/0006025;
K. Higashijima, M. Nitta, in: C.S. Lim, et al. (Eds.), Proceedings of ICHEP2000, World Scientific, Singapore, 2001, pp. 13681370, hep-th/0008240.
[19] K. Higashijima, E. Itou, M. Tsuzuki, in preparation.
[20] A. Borel, Proc. Natl. Acad. Sci. 40 (1954) 1147.
[21] M. Bando, T. Kuramoto, T. Maskawa, S. Uehara, Phys. Lett. B 138 (1984) 94;
M. Bando, T. Kuramoto, T. Maskawa, S. Uehara, Prog. Theor. Phys. 72 (1984) 313;
M. Bando, T. Kuramoto, T. Maskawa, S. Uehara, Prog. Theor. Phys. 72 (1984) 1207.
[22] T. Kugo, Soryushiron Kenkyu (Kyoto) 95 (1997) C56;
T. Kugo, in: J. Nishimura, K. Yamawaki (Eds.), Proceedings of 1996 International Workshop on
Perspectives of Strong Coupling Gauge Theories, SCGT 96, World Scientific, Singapore, 1996,
http://www.phys.nagoya-u.ac.jp/Scgt/proc/.
[23] M. Bordemann, M. Forger, H. Rmer, Commun. Math. Phys. 102 (1986) 605.
[24] K. Itoh, T. Kugo, H. Kunitomo, Nucl. Phys. B 263 (1986) 295;
K. Itoh, T. Kugo, H. Kunitomo, Prog. Theor. Phys. 75 (1986) 386.
[25] W. Buchmller, O. Napoly, Phys. Lett. B 163 (1985) 161.
[26] M. Bando, T. Kugo, K. Yamawaki, Phys. Rep. 164 (1988) 217.
[27] S. Aoyama, Nucl. Phys. B 578 (2000) 449, hep-th/0001160.
[28] A.J. Buras, W. Slominski, Nucl. Phys. B 223 (1983) 157.
[29] W. Lerche, Nucl. Phys. B 238 (1984) 582.
[30] G.M. Shore, Nucl. Phys. B 248 (1984) 123.
[31] W. Buchmller, U. Ellwanger, Phys. Lett. B 166 (1985) 325.
[32] U. Ellwanger, Nucl. Phys. B 281 (1987) 489;
U. Ellwanger, Fortschr. Phys. 36 (1988) 881.
[33] W. Buchmller, W. Lerche, Ann. Phys. 175 (1987) 159.
[34] A.C. Kotcheff, G.M. Shore, Int. J. Mod. Phys. A 4 (1989) 4391;
G.M. Shore, Nucl. Phys. B 320 (1989) 202;
G.M. Shore, Nucl. Phys. B 334 (1990) 172.
[35] K. Higashijima, M. Nitta, K. Ohta, N. Ohta, Prog. Theor. Phys. 98 (1997) 1165, hep-th/9706219.
[36] M. Nitta, Int. J. Mod. Phys. A 14 (1999) 2397, hep-th/9805038.
[37] K. Furuta, T. Inami, H. Nakajima, M. Nitta, Prog. Theor. Phys. 106 (2001) 851, hep-th/0106183.
[38] T. Kugo, I. Ojima, T. Yanagida, Phys. Lett. B 135 (1984) 402;
W. Lerche, Nucl. Phys. B 246 (1984) 475.
[39] J. Bagger, E. Witten, Phys. Lett. B 118 (1982) 103.
[40] T. Itoh, P. Oh, C. Ryou, Phys. Rev. D 64 (2001) 045005, hep-th/0101041.
[41] L. Alvarez-Gaum, Nucl. Phys. B 184 (1981) 180;
C.M. Hull, Nucl. Phys. B 260 (1985) 182;
L. Alvarez-Gaum, P. Ginsparg, Commun. Math. Phys. 102 (1985) 311.
[42] K. Hori, C. Vafa, Mirror symmetry, hep-th/0002222.
[43] V.A. Novikov, M.A. Shifman, A.I. Vainshtein, V.I. Zakharov, Phys. Lett. B 139 (1984) 389;
A.Y. Morozov, A.M. Perelomov, M.A. Shifman, Nucl. Phys. B 248 (1984) 279.
[44] K. Higashijima, E. Itou, Prog. Theor. Phys. 108 (2002) 737, hep-th/0205036;
K. Higashijima, E. Itou, Prog. Theor. Phys. 109 (2003) 751, hep-th/0302090;
K. Higashijima, E. Itou, Prog. Theor. Phys. 110 (2003) 563, hep-th/0304194.
162
[45] K. Higashijima, M. Nitta, Prog. Theor. Phys. 105 (2001) 243, hep-th/0006027;
K. Higashijima, E. Itou, M. Nitta, Prog. Theor. Phys. 108 (2002) 185, hep-th/0203081.
[46] K. Higashijima, T. Kimura, M. Nitta, Phys. Lett. B 515 (2001) 421, hep-th/0104184.
[47] K. Higashijima, T. Kimura, M. Nitta, Phys. Lett. B 518 (2001) 301, hep-th/0107100;
K. Higashijima, T. Kimura, M. Nitta, Nucl. Phys. B 623 (2002) 133, hep-th/0108084;
K. Higashijima, T. Kimura, M. Nitta, Ann. Phys. 296 (2002) 347, hep-th/0110216.
[48] K. Higashijima, T. Kimura, M. Nitta, Nucl. Phys. B 645 (2002) 438, hep-th/0202064;
K. Higashijima, T. Kimura, M. Nitta, in: S. Bentvelsen, et al. (Eds.), Proceedings of the 31st International Conference on High Energy Physics, ICHEP2002, Elsevier, Amsterdam, 2002, pp. 867869, hepth/0210034.
[49] M. Nitta, Non-compact CalabiYau metrics from nonlinear realizations, hep-th/0309004.
[50] L. Alvarez-Gaum, D.Z. Freedman, Commun. Math. Phys. 80 (1981) 443.
[51] U. Lindstrm, M. Rocek, Nucl. Phys. B 222 (1983) 285.
[52] N.J. Hitchin, A. Karlhede, U. Lindstrm, M. Rocek, Commun. Math. Phys. 108 (1987) 535.
[53] E. Calabi, Ann. Sci. cole Norm. Sup. 12 (1979) 269.
[54] T.L. Curtright, D.Z. Freedman, Phys. Lett. B 90 (1980) 71;
L. Alvarez-Gaum, D.Z. Freedman, Phys. Lett. B 94 (1980) 171;
M. Rocek, P.K. Townsend, Phys. Lett. B 96 (1980) 72.
[55] M. Arai, M. Naganuma, M. Nitta, N. Sakai, Nucl. Phys. B 652 (2003) 35, hep-th/0211103;
M. Arai, M. Naganuma, M. Nitta, N. Sakai, BPS wall in N = 2 SUSY nonlinear sigma model with Eguchi
Hanson manifold, in: Garden of Quantain Honor of Hiroshi Ezawa, World Scientific, Singapore, 2003,
pp. 299325, hep-th/0302028.
[56] A. Dancer, A. Swann, J. Geom. Phys. 21 (1997) 218.
[57] M. Arai, M. Nitta, N. Sakai, Vacua of massive hyper-Khler sigma models of non-Abelian quotient, hepth/0307274, Prog. Theor. Phys., in press.
[58] H. Kanno, Y. Yasui, J. Geom. Phys. 43 (2002) 293, hep-th/0108226;
H. Kanno, Y. Yasui, J. Geom. Phys. 43 (2002) 310, hep-th/0111198.
Gauged linear sigma models for noncompact

CalabiYau varieties
Tetsuji Kimura
Theory Division, Institute of Particle and Nuclear Studies,
High Energy Accelerator Research Organization (KEK), Tsukuba, Ibaraki 305-0801, Japan
Abstract
We study gauged linear sigma models for noncompact CalabiYau manifolds described as a line
bundle on a hypersurface in a projective space. This gauge theory has a unique phase if the Fayet
Iliopoulos parameter is positive, while there exist two distinct phases if the parameter is negative.
We find four massless effective theories in the infrared limit, which are related to each other under
the CalabiYau/LandauGinzburg correspondence and the topology change. In the T-dual theory, on
the other hand, we obtain two types of exact massless effective theories: one is the sigma model on
a newly obtained CalabiYau geometry as a mirror dual, while the other is given by a Landau
Ginzburg theory with a negative power term, indicating N = 2 superconformal field theory on
SL(2, R)/U (1). We argue that the effective theories in the original gauged linear sigma model are
exactly realized as N = 2 Liouville theories coupled to well-defined LandauGinzburg minimal
models.
PACS: 11.15.Ex; 11.25.Hf; 11.25.Pm
1. Introduction
Field theory in two-dimensional spacetime is one of the most powerful tools for analyzing dynamical phenomena in particle physics. It has been studied as a toy model of
E-mail address: tetsuji@post.kek.jp (T. Kimura).
doi:10.1016/j.nuclphysb.2005.01.029
164
T. Kimura / Nuclear Physics B 711 (2005) 163198
low energy effective theory including symmetry breaking mechanism. Nonlinear sigma
model (NLSM) on the projective space is a typical example to investigate chiral symmetry
breaking [1,2]. String worldsheet theory is also described as a two-dimensional conformal
field theory (CFT). Conformal invariance in the worldsheet theory gives a set of equations
of motion of spacetime physics [3]. When we discuss string theory on curved spacetime,
supersymmetric NLSMs play significant roles.
Coupled to a gauge field, two-dimensional field theory has been applied to more complicated physics. The gauge field plays a key role in the compactification of the target
space via the Higgs mechanism. This theory is also useful to study mathematical problems such as Morse theory [4] and mirror symmetry [5]. Many people have constructed
two-dimensional supersymmetric gauge theories in order to understand various kinds of
physical and mathematical structures.
Higashijima and Nitta formulated supersymmetric NLSMs on hermitian symmetric
spaces (HSSs), which are specific Khler manifolds, starting from supersymmetric gauge
theories with four supercharges [6]. By using this, we constructed Khler metrics on complex line bundles over compact EinsteinKhler manifolds [7,8]. These noncompact Khler
manifolds have vanishing Ricci tensors, and hence are CalabiYau (CY) manifolds [9,10].
In attempt to investigate the gauge/gravity duality in string theory [1113] on these CY
manifolds, however, it is indispensable to understand global aspects such as cohomology
classes.
On the other hand, it is a well-known fact that the gauged linear sigma model (GLSM)
is useful to investigate worldsheet string theories on toric varieties [14]. In this framework,
we can understand some global properties of the CY manifolds. GLSM includes at least
two kinds of SCFTs in the infrared (IR) limit. One is a supersymmetric NLSM on CY
manifold and the other is an N = 2 LandauGinzburg (LG) theory. We can read the cohomology ring of the CY manifold from the chiral ring derived from the LG superpotential
[15]. Furthermore, its T-dual theory provides us with mirror descriptions of the original
geometry [16,17].
Since each HSS is constructed as a submanifold of a complex projective space or of
a Grassmannian, we can, in general, construct the GLSMs for the line bundles on HSSs.
Investigating LG theories in the IR limit of the GLSMs, we will be able to understand
cohomology rings of the noncompact CY manifolds. Unfortunately, however, it is difficult
to embed the sigma models on HSSs into the GLSMs. Thus we study the GLSM for a line
bundle of homogeneous hypersurface in the projective space whose Ricci tensor vanishes
[18]. This noncompact manifold is represented as O(N + ) bundle on CPN 1 [], which
can be seen as a toy model of the line bundles on HSSs.
In this paper we will find the following theories in the IR limit: NLSMs on CY manifolds, orbifolded LG theories, gauged WessZuminoWitten (WZW) models on coset
SL(2, R)/U (1) and Liouville theories. They appear as N = 2 SCFTs. The former two
theories give unitary conformal field theory. Under the conformal invariance, the sigma
model and the LG theory become appropriate SCFTs from differential geometric and
algebro-geometric points of view, respectively. They often emerge when we analyze superstring theory on compact manifolds [19,20]. The latter two theories are slightly different.
These theories appear in string theory on noncompact curved spacetime such as a twodimensional black hole [21,22]. They are also utilized in non-critical string theory and
165
matrix model [23,24]. It is quite important to study all four SCFTs simultaneously when
we consider string theories on noncompact CY manifolds. Thus we study the GLSM for
noncompact CY manifolds including all the above four theories in the low energy limit.
This paper is organized as follows. In Section 2 we study the GLSM for line bundles
and discuss how massless effective theories appear in some specific limits. In this analysis
we find that there exist two distinct phases in the negative FI parameter region. This phenomenon newly appears, while other well-known GLSMs do not include this. In Section 3
we discuss the T-dual of the GLSM. We obtain two types of exact effective theories, the
sigma models on newly constructed mirror CY geometries and the LG theories with negative power. There we discuss the exact effective theories in the original GLSM. We devote
Section 4 to the summary and discussions. In Appendix A we introduce conventions of
N = 2 supersymmetry in two-dimensional spacetime. In Appendix B we review a definition of weighted projective space. Finally, we briefly introduce the linear dilaton CFT and
discuss an interpretation of LG superpotential with a negative power term in Appendix C.
This argument is useful to understand the LG theories in Section 3.
2. Gauged linear sigma model

2.1. Lagrangian: review
First of all, let us briefly review of a general formulation of the GLSM [14]. In this model
there appear various superfields such as a chiral superfield a , a vector superfield V and
a twisted chiral superfield , whose definitions are in Appendix A. We also incorporate a
complexified abelian gauge transformation
a a = e2iQa a ,
V V = V + i( ),
a a = e+2iQa a ,
= ,
where Qa is a U (1) charge of the chiral superfield a . For convenience, we restrict these
charges to integers: Qa Z. The complexified gauge parameters are described by a chiral
, ), respectively: D = 0, D = 0.
and an anti-chiral superfields (x, , ) and (x,
By using superfields, we construct a supersymmetric gauge invariant Lagrangian:

1
+
LGLSM = d4 2
a e2Qa V a
e
a

1
2
2
+
d W () + c.c. +
d WGLSM (a ) + c.c. ,
2
where we assume that all chiral superfields have non-zero U (1) charges Qa = 0 because
a neutral chiral superfield is completely free from the system. The abelian gauge coupling
constant e, which appears in front of the kinetic term of , has mass dimension one. There
exist two types of superpotentials. One is a superpotential written as WGLSM (a ). This is a
holomorphic function of chiral superfields a . The other is called a twisted superpotential
W () described as
W () = t,
t = r i,
166
where t is a complex parameter defined by the FayetIliopoulos (FI) parameter r and the
theta-angle . We also refer t to the (complexified) FI parameter.
We are interested in supersymmetric low energy effective theories. Thus we need to
study the potential energy density U() described by the scalar component fields of superfields:

e2
|Fa |2 + U (),
U() = D2 +
(2.1a)
2
a

1
D=r
Qa |a |2 ,
2
e
a

2
Q2a |a |2 ,
U () := 2| |
D :=
Fa =
WGLSM (),
a
(2.1b)
where the scalar components of a and are expressed by a and . We sometimes

abbreviate scalar component fields of all superfields to a . The functions D and Fa are
auxiliary fields of and a , respectively. We need not include fermionic components into
the above functions (2.1) if we simply investigate supersymmetric vacua. The supersymmetric vacuum manifold M is defined by the vanishing potential energy density U() = 0:

M := (a ) Cn D = Fa = U = 0 U (1),
where n is the number of scalar component fields in the GLSM. The dividing U (1) group
indicates the abelian gauge symmetry. Since we consider N = (2, 2) supersymmetric theories, the manifold M becomes a Khler manifold [25] where the FI parameter denotes
the scale of M.
Under a generic configuration for chiral superfields a of charges Qa , the FI parameter
r receives a renormalization via wave-function renormalizations of a . Thus the bare FI
parameter r0 is related to the renormalized one rR under the following equation:

UV
,
r0 = rR +
(2.2)
Qa log
a
where UV and are the ultraviolet cut-off and the scale parameter, respectively. Thus we
observe that the scale of M changes under the renormalization group (RG) flow. Studying
the -function of the FI parameter derived from (2.2), we find whether the effective theories
expanded on M are asymptotically free or not. In particular, if we impose

Qa = 0,
(2.3)
a
the FI parameter does not receive the renormalization. Thus there appears a non-trivial

conformal field theory in the IR limit. From the geometric point of view, the sum a Qa
is equal to the first Chern class c1 (M) of the vacuum manifold M. If the condition (2.3)
is satisfied on M, this manifold becomes a CY manifold. Thus we refer (2.3) to the CY
condition. In attempt to study CY manifolds, we impose this on the GLSM.
We usually study how massless effective theories are realized on the supersymmetric
vacuum in M. Recall that in two-dimensional field theory the massless modes are not
167
well-defined because of the IR divergence in their two-point functions. The Colemans theorem on non-existence of NambuGoldstone modes [26] is closely related to this difficulty.
In order to avoid this problem, we assume that there exists an IR cut-off parameter. Furthermore we take the large volume limit r when we consider a NLSM whose target
space is the vacuum manifold M. In this limit the FI parameter r of GLSM can be related
to the coupling constant g of the NLSM:
r=
1
.
g2
This means that the large volume limit r is the weak coupling limit g 0.
Next we consider fluctuation fields around the vacuum. It is so complicated to analyze
massless/massive fluctuation modes that we perform here a general calculation. Let us
decompose scalar component fields a into three kinds of variables:

a = a + a + a ,
(2.4)
d2 x (a + a ) = 0,
where a , a and a mean the vacuum expectation values (VEVs), the fluctuation modes
tangent and non-tangent to the vacuum manifold M, respectively. They satisfy the following relations:

F ()VEV := F 0,
(2.5a)

F ()

()
:=
a
0,
F
(2.5b)

VEV
a

a

F ()

()
F
(2.5c)
:=
a
= 0.

VEV
a

a
Note that F () are the set of functions given by (2.1b): F = {D, Fa , U }. The symbols
and denote holomorphic variations with respect to the complex variables a . Of course
the VEVs a satisfy Eq. (2.5a). Eq. (2.5b) provides that the first order variations of F ()
with respect to the fluctuation modes a vanish. This is nothing but the definition that a
move only tangent to the vacuum manifold M. The third equation (2.5c) means that the
other fluctuation modes a do not propagate tangent to M. Substituting (2.4) and (2.5) into
the potential energy density U() described by (2.1), we investigate the behaviors of the
low energy effective theories. If Eqs. (2.5a) and (2.5b) furnish non-trivial relations among
the fluctuation modes a , these modes constitute a supersymmetric NLSM whose target
space is M. However, if these equations are trivially satisfied, a are free from constraints
and propagate on a flat space with potential energy. Then we find that a field theory appears
described by a superpotential of fluctuation fields such as a LG superpotential WLG .
2.2. Field configuration and supersymmetric vacuum manifold
Now we are ready to analyze massless low energy effective theories in the GLSM for
O(N + ) bundle on CPN 1 []. We consider a U (1) gauge theory with N + 2 chiral
168
superfields a of charges Qa . We set the field configuration to

chiral superfield a
U (1) charge Qa
S1
1
...
...
SN
1
P1 P2
N +
(2.6)
In addition, we introduce a superpotential WGLSM () = P1 G (S), where G (S) is a

function of chiral superfields Si . This is a holomorphic homogeneous polynomial of degree
. Owing to the homogeneity, this polynomial has a following property:
if G (s) = 1 G (s) = = N G (s) = 0 then si = 0.
(2.7)
By definition, the numbers N and are positive integers: , N Z>0 . We assume that these
two integers satisfy 1 N 1 and 2 N . The sum of all charges Qa vanishes (2.3)
in order to obtain non-trivial SCFTs on the CY manifold.
Now we consider the potential energy density and look for supersymmetric vacua. Imposing the WessZumino gauge, we write down the bosonic part of the potential energy
density U():
U() =
N
2

e2 2
p1 i G (s)2 + U (),
D + G (s) +
2
(2.8a)
i=1
D=r
N
|si |2 + |p1 |2 + (N )|p2 |2 ,
i=1
U () = 2| |
N
(2.8b)

|si | + |p1 | + (N )|p2 |

2
(2.8c)
i=1
Imposing zero on them, we obtain the supersymmetric vacuum manifold M. Since the Lagrangian has N = (2, 2) supersymmetry and the single U (1) gauge symmetry, the vacuum
manifold becomes a Khler quotient space:

M = (a ) CN +3 D = G = p1 i G = U = 0 U (1).
(2.9)
In attempt to study effective theories, we choose a point on M as a vacuum and give VEVs
of scalar component fields: a a . Then we expand the fluctuation modes around the
vacuum. In general, the structure of M is different for r > 0, r = 0 and r < 0 and there
appear various phases in the GLSM. The phase living in the r > 0 region is referred to the
CY phase, and the phase in r < 0 is called to the orbifold phase. A singularity of the
model emerges in the phase at r = 0. Thus we sometimes call this the singularity phase.
We will treat these three cases separately. We comment that in each phase the vacuum
manifold is reduced from the original M. We often refer the reduced vacuum manifold to
Mr M. The VEVs of the respective phases can be set only in Mr .
2.3. CalabiYau phase
In this subsection we analyze the CY phase r > 0. In this phase, D = 0 requires
some si cannot be zero and therefore must vanish. If we assume p1 = 0, the equations
169
G (s) = i G (s) = 0 with the condition (2.7) imply that all si must vanish. However, this
is inconsistent with D = 0. Thus p1 must be zero. The variable p2 is free as long as the
condition D = 0 is satisfied. Owing to these, the vacuum manifold M is reduced to MCY
defined by

MCY = (si ; p2 ) CN +1 D = G (s) = 0, r > 0 U (1).

(2.10)
Here we explain this manifold in detail. This is an (N 1)-dimensional noncompact Khler
manifold. The components si denote the homogeneous coordinates of the complex projective space CPN 1 . The constraint G (s) = 0 reduces CPN 1 to a degree hypersurface
expressed to CPN 1 []. We find that p2 is a fiber coordinate of the O(N + ) bundle
on CPN 1 []. Furthermore, the vanishing sum of U (1) charges indicates that the FI parameter r is not renormalized. This is equivalent to c1 (MCY ) = 0. Thus we conclude that
the reduced vacuum manifold MCY is nothing but a noncompact CY manifold on which a
non-trivial superconformal field theory is realized.
Let us consider a low energy effective theory. We choose a vacuum and take a set of
VEVs of the scalar component fields. Because si = 0, the U (1) gauge symmetry is
spontaneously broken down completely. Next, we expand all fields in terms of fluctuation
modes such as a = a + a + a . We set p1 , and to be zero. Substituting them
into the potential energy density (2.8), we obtain

N

e2
U=
2si si + (N )p 2 p 2
2 Re
2
i=1
2
N

2
2
2
|si + si | + |p 1 | + (N )|p 2 + p 2 |

i=1
2
N

1

+
si i G s +
(s + s )i1 (s + s )ik i1 ik G s

k!
i=1
k=2
i1 ,,ik

2
1

1

+ |p 1 |2
(s + s )j1 (s + s )jk i j1 jk G s
i G s +

k!
j1 ,...,jk
i=1
k=2
N

2

2
2
2
2
2

si + si + si + |p 1 | + (N ) p2 + p 2 + p 2 .
+ 2| |
N
i=1
Fluctuation modes si and p 2 remain massless and move only tangent to MCY because they
VEV = G
|VEV = 0. The variation (p
1 i G )|VEV = 0 indicates p 1 = 0.
are subject to D|
The modes , p 1 , si and p 2 have mass m2 = O(e2 r). The gauge field vm also acquires mass
of order O(e2 r) by the Higgs mechanism. The fermionic superpartners behave in the same
way as the scalar component fields because of preserving supersymmetry. In the IR limit
e and the large volume limit r , the massive modes decouple from the system.
Thus we obtain
N = (2, 2) supersymmetric NLSM on MCY
(2.11)
170
Table 1
Classification of O(N + ) bundle on CPN 1 []
Degree
Vacuum manifold MCY
=1
2N 1
O(N + 1) bundle on CPN 2

O(N + ) bundle on CPN 1 []
as a massless effective theory. Notice that this description is only applicable in the large
volume limit because the NLSM is well-defined in the weak coupling limit from the viewpoint of perturbation theory. This effective theory becomes singular if we take the limit
r +0 because the decoupled massive modes becomes massless. This phenomenon also
appears in the SeibergWitten theory [27,28], the black hole condensation [29,30], and
so on.
Let us make a comment on the target space MCY . By definition, the number means the
degree of the vanishing polynomial G (s) = 0, which gives a hypersurface in the projective
space CPN 1 . We can see that if = 1, G=1 (s) = 0 gives a linear constraint with respect
to the homogeneous coordinates si and the hypersurface CPN 1 [ = 1] is reduced to (N
2)-dimensional projective space CPN 2 . This reduction does not occur if 2 N 1.
Here we summarize the shape of the target space MCY in Table 1.
Although the = 1 case has been already analyzed in the original paper [14], the other
cases 2 N 1 are the new ones which have not been analyzed.
2.4. Orbifold phase
Here we consider the negative FI parameter region r < 0. In this region the total vacuum
manifold (2.9) is restricted to a subspace defined by

Morbifold = (p1 , p2 ; si ) CN +2 D = G = p1 i G = 0, r < 0 U (1).

(2.12)
Since D = 0 does not permit p1 and p2 to vanish simultaneously, must be zero. This
subspace is quite different from MCY in the CY phase. In addition, the shape of Morbifold
is sensitive to the change of the degree because of the existence of the constraints
G = p1 i G = 0 and the property (2.7). Thus let us analyze Morbifold and study massless
effective theories on it in the case of 3 N 1, = 2 and = 1, separately.
2.4.1. Effective theories of 3 N 1
Here we analyze the vacuum manifold Morbifold and massless effective theories of
3 N 1. Owing to the constraints G = p1 i G = 0 and their property (2.7), the
manifold Morbifold is decomposed into the following two subspaces:

Morbifold 3N 1 = M1r<0 M2r<0 ,
(2.13a)

1
2
Mr<0 := (p1 , p2 ) C D = 0, r < 0 U (1),
(2.13b)

2
N +1
M
(2.13c)
D = G = 0, r < 0 U (1).
:= (p2 ; si ) C
r<0
In the former subspace the condition (2.7) is trivially satisfied whereas in the latter subspace
it is satisfied non-trivially. Both of the two subspace include a specific region p1 = si = 0.
171
The subspace M1r<0 is defined as a one-dimensional weighted projective space WCP1,N

represented by two complex fields p1 and p2 of U (1) charges and (N ), respectively. The precise definition of the weighted projective space is in Appendix B.
Let us choose a supersymmetric vacuum and set VEVs of all scalar fields. Then we
expand all the fields around the VEVs. Expanding the potential energy density (2.8) in
terms of VEVs and fluctuation modes, we obtain the following form:

e2
2 Re p 1 p 1 + (N )p 2 p 2
U=
2
2
|si + si |2 + |p 1 + p 1 |2 + (N )|p 2 + p 2 |2

i

2

2
i G (s + s )2
+ G (s + s ) + p1 + p 1 + p 1
+ 2| |2

2

2
|si + si |2 + 2 p1 + p 1 + p 1 + (N )2 p2 + p 2 + p 2 ,
where p1 and p2 are VEVs of scalar components of P1 and P2 , respectively. They live
in the weighted projective space (2.13b). Because the VEVs of si are all zero, the U (1)
gauge symmetry is spontaneously broken to Z , where is the great common number
between and N : = GCM{, N }. This potential energy density provides that all
fluctuation modes si and si appear as linearly combined forms such as si + si , which do
not acquire any mass terms. The modes p 1 and p 2 remain massless and move tangent to
the subspace (2.13b). The other fluctuation modes acquire mass of order m2 = O(e2 |r|).
Thus, in the IR limit e , all the massive modes are decoupled from the system. Thus
we obtain the following massless effective theory:
N = (2, 2) supersymmetric NLSM on WCP1,N

coupled to LG theory with WLG = p1 + P1 G (S) Z ,
(2.14)
where P1 and Si are massless chiral superfields. Note that the sigma model sector also contains the Z orbifold symmetry coming from the property of WCP1,N . As is well known
that the term p1 G (S) forms an ordinary LG superpotential. Thus in the IR limit we can
interpret that this term is marginal and flows to the N = (2, 2) minimal model. The second
term P1 G (S) is somewhat mysterious. Since this term has not any isolated singularities
we might not obtain well-defined unitary CFT. This difficulty causes the noncompactness
of the manifold MCY which appears in the CY phase.
There are two specific points in the subspace WCP1,N . One is the point p2 = 0 and
the other is p1 = 0. In the former point the gauge symmetry is enhanced to Z . Furthermore, the mode p 1 disappears and p 2 becomes massless, which combines with a massless
fluctuation modes p 2 linearly. This combined mode is free from any constraints. The
other massless modes si + si in (2.14) remain massless and are also free from constraints.
Thus in the IR limit and the large volume limit, the massless effective theory becomes an
N = (2, 2) supersymmetric theory as

CFT on C1 LG theory with WLG = p1 G (S) Z .
(2.15)
172
This effective theory consists of N + 1 massless chiral superfields such as P2 and Si ,

which live in the free and the LG sectors, respectively. Since we take the IR limit, this
effective theory becomes an SCFT. The LG sector flows to a well-known LG minimal
model [14]. Thus the sigma model sector is also a superconformal field theory. Here we
notice that we did not integrate out but just decomposed all massive modes in the above
discussion because it is generally impossible to calculate the integration of them. Thus the
above effective theory is merely an approximate one. If we will be able to integrate out all
massive modes exactly, the obtaining effective theory will be different from the above one.
In later section we will discuss the exact form of the effective theory.
Next, let us consider the latter point p1 = 0 in the space WCP1,N . On this point the
broken gauge symmetry is partially restored to ZN . The massless fluctuation mode p 2
becomes zero whereas the massive mode p 1 becomes massless, which combines with p 1
being free from any constraints. Thus P1 appears as a massless chiral superfield. In the IR
limit we obtain the supersymmetric massless effective theory such as

LG theory with WLG = P1 G (S) on CN +1 ZN ,
(2.16)
which consists of N + 1 massless chiral superfields such as P1 and Si . This theory is not a
well-defined LG theory because the superpotential WLG has no isolated singularities. We
interpret the defect of isolated singularities as a noncompactness of the manifold MCY in
the CY phase via CY/LG correspondence (if this correspondence is satisfied in the case
of sigma models on noncompact CY manifolds). This property prevents from calculating
a chiral ring of this model in the same way as unitary LG minimal models describing
compact CY manifolds [15].
Here we study massless effective theories on the subspace M2r<0 defined in (2.13c).
As mentioned before, there are non-trivial constraints in M2r<0 . Thus, as we shall see, the
effective theories are also under these constraints. In the same way as discussed before,
we choose one point in the subspace M2r<0 and make all the scalar fields fluctuate around
it. Then we write down the expanded potential energy density (2.8) in terms of VEVs and
fluctuation modes a , a and a :

e2
2 Re
U=
si si + (N )p 2 p 2
2
i
2

|si + si |2 + |p 1 |2 + (N )|p 2 + p 2 |2
2

1

+
si i G s +
(s + s )i1 (s + s )ik i1 ik G s

k!
i
k=2
i1 ,...,ik

2
1

1
2
+ |p 1 |
(s + s )j1 (s + s )jk i j1 jk G s
i G s +

k!
i
j1 ,...,jk
k=1

si + si + si 2 + 2 |p 1 |2 + (N )2 p2 + p 2 + p 2 2 .
+ 2| |2
i
173
This potential energy density indicates the following: the fluctuation modes si , p 1 and p 2
are massive; si and p 2 move tangent to M2r<0 . Thus, taking e and |r| , we
obtain
N = (2, 2) supersymmetric NLSM on M2r<0 .
(2.17)
In this theory there exist P2 and Si as massless chiral superfields, which move tangent to
M2r<0 . Notice that in general points in M2r<0 the U (1) gauge symmetry is completely
broken because of the existence of non-zero VEVs si . However, taking si = p1 = 0
and p2 = 0 in the subspace M2r<0 , we find that the gauge symmetry is partially restored
to ZN .
Even though the vacuum manifolds M1r<0 and M2r<0 are connected on p1 = si = 0,
the effective theories given by (2.16) and (2.17) are quite different from each other. The
reason is that while the subspace M1r<0 is free from constraints G = p1 i G = 0, in the
subspace M2r<0 these constraints are still valid on the region p1 = si = 0. On account of
the existence of these constraints, a phase transition occurs when the theory moves from
one to the other. Thus we conclude that a new phase appears on the subspace M2r<0 , which
has not been discovered in well-known GLSMs such as the models for O(N ) bundle on
CPN 1 , for CPN 1 [N], for resolved conifold, and so on. We refer this phase to the 3rd
phase. Here we refer the phase on M1r<0 to the orbifold phase, as usual.
2.4.2. Effective theories of = 2
Let us consider the orbifold phase of = 2. In the same way as the previous analysis,
the constraints G = p1 i G = 0 and the property (2.7) decompose the manifold Morbifold
into two subspaces:

Morbifold =2 = M1r<0 M2r<0 ,

M1r<0 := (p1 , p2 ) C2 D = 0, r < 0 U (1) WCP12,N 2 ,

M2r<0 := (p2 ; si ) CN +1 D = G2 = 0, r < 0 U (1).
(2.18a)
(2.18b)
(2.18c)
These two subspaces are glued in the region given by p1 = si = 0. Although this situation
is same as to the case of 3 N 1, the appearing massless effective theories are quite
different.
Here let us analyze the effective theories on the subspace M1r<0 = WCP12,N 2 . We
choose a point in this subspace as a supersymmetric vacuum and take VEVs of all scalar
fields. Then we make all scalar fields fluctuate around the VEVs. Fluctuation modes p 1
and p 2 are subject to the constraints such that they move only tangent to WCP12,N 2 . The
fluctuation modes si have no degrees of freedom because of the variation of the constraint
p1 i G2 = 0. (In the case of (2.13b), the equations G = 0 and p1 i G = 0 are trivially
satisfied in WCP1,N . These variations are also trivial. However, the case of = 2 is
quite different. By definition, some i j G2 must have non-zero values. Thus even though
the above equations are trivially satisfied in the subspace WCP12,N 2 , their variations give
non-trivial constraints on the fluctuation modes.) Under these conditions we write down
the potential energy density (2.8) in terms of VEVs a and fluctuation modes a and a :
174
U=

e2
2 Re 2p 1 p 1 + (N 2)p 2 p 2
2
2

2
2
2
|si | + 2|p 1 + p 1 | + (N 2)|p 2 + p 2 |

i

2

2
i G2 (s )2
+ G2 (s ) + p1 + p 1 + p 1
+ 2| |2

2

2
|si |2 + 4p1 + p 1 + p 1 + (N 2)2 p2 + p 2 + p 2 .
This function denotes the following: the fluctuation modes si , p 1 and p 2 acquire masses
m2 = O(e2 |r|); the modes p 1 and p 2 remain massless and move tangent to WCP12,N 2 .
Thus taking e and |r| , we obtain the massless effective theory described by
N = (2, 2) supersymmetric NLSM on WCP12,N 2 .
(2.19)
This sigma model has Z orbifold symmetry coming from the property of
where = GCM{2, N 2}. This effective theory does not include massless LG theory. The
reason is that the degree two polynomial G2 generates mass terms such as
|p1 |2 i |i G2 |2 . (See, for example, [31].)
Now we consider the effective theory on two specific points in WCP12,N 2 like (2.15)
and (2.16). Expanding the theory on the one point (p1 , p2 ) = (p1 , 0), the gauge symmetry
is partially restored to Z2 . Thus we obtain the effective theory on this specific point as
WCP12,N 2 ,
N = (2, 2) SCFT on C1 /Z2 .
(2.20)
Note that this theory can possess the LG theory with a quadratic superpotential WLG =
p1 G2 (S), which gives massive modes of Si .
The effective theory drastically changes if we expand the theory on another point
(p1 , p2 ) = (0, p2 ) in WCP12,N 2 . On this point, the broken gauge symmetry is enhanced to
ZN 2 and the fluctuation modes si become massless with being free from any constraints.
Both p 1 and p 1 are massless and linearly combined in the potential energy density. The remaining field p 2 becomes zero because there exists a non-trivial variation of the constraint
D = 0. Summarizing these results, we find that the following massless effective theory
appears in the limit e, |r| :

N = (2, 2) LG theory with WLG = P1 G2 (S) on CN +1 ZN 2 .
(2.21)
Although this superpotential also has no isolated singularities, this theory should describe
a non-trivial SCFT. We shall return here in later discussions.
We next study the massless effective theories on the subspace M2r<0 defined in (2.18c).
The potential energy density is obtained as

e2
2 Re
U=
si si + (N 2)p 2 p 2
2
i
2

|si + si |2 + 2|p 1 |2 + (N 2)|p 2 + p 2 |2
175

1
2

+
si i G2 s +
(s + s )i (s + s )j i j G2 s
2!
i
i,j

2
2

+ |p 1 |
(s + s )j i j G2 s
i G2 s +
i

si + si + si 2 + 4|p 1 |2 + (N 2)2 p2 + p 2 + p 2 2
+ 2| |2
i
under the following constraints on fluctuation modes: the fluctuations si and p 2 move tangent to M2r<0 ; the other tangent mode p 1 is zero; the fluctuations a are all massive of
m2 = O(e2 |r|). Thus the effective theory expanded around generic points in M2r<0 becomes
N = (2, 2) supersymmetric NLSM on M2r<0
(2.22)
in the IR and large volume limit: e, |r| . The U (1) gauge symmetry is completely
broken if some si = 0 exist. On the other hand, if we expand the theory on a specific
point p1 = si = 0, the gauge symmetry is partially restored to ZN 2 .
So far we have studied the effective theories on all regions of the vacuum manifold
Morbifold of = 2. From the same reason discussed in the case of 3 N 1, there
exists a phase transition between the theories (2.21) and (2.22) because of the non-trivial
constraint coming from the variation of the equation G2 = 0. Thus we find that the GLSM
for the O(N + 2) bundle on CPN 1 [2] also includes two phases in the negative FI parameter region. The phase on (2.19) is called the orbifold phase, and we refer the phase on
(2.22) to the 3rd phase.
Here we illustrate the relation among the phases in the GLSM schematically in Fig. 1.
In the large volume limit |r| , these three effective theories (2.11), (2.14) and
(2.17) become well-defined. In later discussions we shall consider how these effective theories deform in the small FI parameter limit |r| 0. There we must consider the singular
phase [14].
Fig. 1. Various phases in GLSM for O(N + ) bundle on CPN 1 [] with 2 N 1. The axes with
thin/thick lines represent the vacuum space coordinates in the positive/negative FI parameter regions, respectively.
176
2.4.3. Effective theory of = 1

Finally we investigate the = 1 case. Since the polynomial G=1 (S) is of degree one,
there exist some non-zero values of i G1 (S). Thus, combining this condition with the other
constraints which define Morbifold , we find that p1 must be zero and obtain the following
reduced vacuum manifold:

Morbifold =1 = (p2 ; si ) CN +1 D = G1 = 0, r < 0 U (1) =: M2r<0 .

(2.23)
Since this space is defined in the same way as (2.13c) and (2.18c), we also referred it to
M2r<0 .
After taking VEVs of scalar fields which live in (2.23), we make scalar fields fluctuate
around the VEVs. These fluctuation modes are subject to constraints: si and p 2 move only
tangent to M2r<0 ; p 1 is zero. Substituting these into (2.8), we obtain the expanded potential
energy density

e2
2 Re
si si + (N 1)p 2 p 2
U=
2
i
2

2
2
2
|si + si | + |p 1 | + (N 1)|p 2 + p 2 |

2

i G1 s 2
+ G1 (s ) + |p 1 |2
i

2
2
2
2

si + si + si + |p 1 | + (N 1) p2 + p 2 + p 2 .
+ 2| |
i
This indicates that the modes si and p 2 remain massless whereas the modes si , p 1 and p 2
become massive. Thus the following massless effective theory appears in the limit e, |r|
:
N = (2, 2) supersymmetric NLSM on M2r<0 ,
(2.24)
where the U (1) gauge symmetry is completely broken because si = 0. While if we set
the VEVs to si = 0, the broken U (1) gauge symmetry is enhanced to ZN 1 . In addition
the modes si become massless and are combined with the tangent modes si , which are still
under constraint G1 = 0. The mode p 2 becomes zero, which is derived from the variation
of D = 0. In this specific point we can see that the space M2r<0 is deformed to CN 1 /ZN 1
and the effective theory are
N = (2, 2) SCFT on CN 1 /ZN 1 .
(2.25)
The two effective theories (2.24) and (2.25) are smoothly connected without any phase
transitions coming from the variations of constraints. Thus we find that in the = 1 case
there exists only one phase in the negative FI parameter region. We refer this to the orbifold
phase, as usual. Here we illustrate the schematic relation between the CY phase and the
orbifold phase in Fig. 2.
Note that the GLSM for O(N + 1) bundle on CPN 1 [1] is completely the same as
the one for O(N + 1) bundle on CPN 2 . This is because the hypersurface CPN 1 [1] is
nothing but CPN 2 . Thus the vacuum structure and phases are also equal to each other.
177
Fig. 2. Various phases in GLSM for O(N + 1) bundle on CPN 1 [1].
2.5. Singularity phase

In this subsection let us analyze the singularity phase. As mentioned before, the effective
theory (2.11) becomes singular if r +0. The effective theories in the orbifold and the 3rd
phases also become singular if r 0. Thus we will study the singularity phase r = 0 in
order to avoid the singularities in effective theories. In this analysis we will find that there
appears a new branch. Then we will discuss how to avoid the singularity.
Here we study how the vacuum
manifold (2.9) is reduced in the r = 0 phase. If we
assume p1 = 0, then we obtain i |si |2 = 0 from D = 0. However, the equations G (s) =
0 and p1 i G (s) = 0 insist that all si vanish. This is a contradiction. Thus p1 must be
zero. Under this condition, we obtain two solutions. One is obtained by D = 0 and = 0.
In general this solution has non-zero a , where a are scalar component fields of chiral
superfields. The other solution is given by a = 0 and is free. We refer the former
and latter solutions to the Higgs and Coulomb branches, respectively. These branches are
similar to the ones of N = 2 SQCD in four dimensions [28]. (The CY, orbifold and 3rd
phases are all in the Higgs branch.) They are connected if all scalar component fields
vanish: a = 0. Now we analyze the effective theories on these two branches.
2.5.1. Higgs and Coulomb branches
Let us consider the Higgs and Coulomb branches in detail. In the Higgs branch, there
exist two supersymmetric vacuum solutions. One is

a = O |r| 0,
(2.26a)
= 0.
This solution is smoothly connected with the supersymmetric vacuum solutions in the
phases of non-vanishing FI parameter. The other is
p1 = 0,
si , p2 : arbitrary order,
= 0,
(2.26b)
which is satisfied only on r = 0. Although the first solution (2.26a) appears in each GLSM,
the second solution (2.26b) does not satisfy the supersymmetric vacuum condition U() =
0 in some GLSMs, for example, the GLSM for CPN 1 [N ].
178
In the Coulomb branch, we can set that a = 0 and is free. This solution appears
only when the FI parameter vanishes. If we choose to be zero, i.e., all the VEVs of
scalar fields vanish a = 0, the Coulomb branch connects with the Higgs branch.
Let us consider a massless effective theory in the Coulomb branch. Since the scalar field
has mass dimension one, we take the VEV to be very large. Owing to this, all chiral
superfields Si , P1 and P2 acquire very large masses via U () in (2.8). Taking
and integrating out all massive fields1 , we obtain the following effective Lagrangian:

1
+
d2 W eff () + c.c. ,
Leff = d4 Keff (, )
2

Qa
1
Qa log
W eff () = t
a

= t log() (N ) log(N + ) .
Note that the twisted superpotential was deformed by the quantum effects coming from the
integration of massive fields. This effective Lagrangian presents the asymptotic form of the
potential energy density
Ueff ( ) =
2

eeff
e2
W eff ( )2 = eff t log() (N ) log(N + )2 .
2
2
(2.27)
In order that the effective theory remains supersymmetric, the potential energy density
must be zero in a specific value of . Notice that if the complexified FI parameter is given
by
t = log() + (N ) log(N + ),
(2.28)
the potential energy density becomes always zero. If it happens, the effective theory does
not have any mass gap and becomes singular as a two-dimensional field theory. Thus (2.28)
is the quantum singular point of the GLSM. In the classical point of view, the value t = 0
looks like a singular point in the theory. Integrating out the massive fields, we find that
the singular point moves to (2.28). The massless effective theories in Coulomb and Higgs
branches are connected with each other avoiding this singular point.
2.5.2. CY/LG correspondence and topology change
As mentioned before, the massless effective theories are only valid if we take the FI
parameter to be infinitely large |r| . In this limit the effective theories are (partly)
described by the NLSMs. However, if we change the FI parameter to be small, the NLSM
representations are no longer well-defined and must be deformed. This phenomenon has
been already studied in [14] as following: if the FI parameter goes to zero r 0, the effective theory on the CY phase moves to the theory on the Coulomb branch in the singularity
phase avoiding the singular point. Furthermore, the effective theory connects to the LG
theory in the orbifold phase when r .
1 Here we can integrate out because the superpotential W
a
GLSM does not contribute to the deformation of
the effective twisted superpotential W eff (). For the precise derivation, see Chapter 15 in [17].
179
The above phenomenon suggests that, rather than the LG theory being equivalent to the
sigma model on the CY manifold, they are two different phases of the same system, i.e., the
system of the single GLSM. Thus the CY/LG correspondence [3238] can be read from
the phase transition. In fact, it has been proved that the sigma model on CP4 [5] and the Z5 orbifolded LG theory with WLG = G5 (S), which are equivalent to each other, appear as the
distinct phases in the single GLSM. Furthermore, the topology change is also understood
in the framework of the phase transition of the GLSM. The flop of the resolved conifold
O(1) O(1) CP1 is a typical example [14].
Now let us apply the above discussions to the GLSM for O(N + ) bundle on
CPN 1 []. For example, we consider the relations among the various phases of 3
N 1, where we have found three phases: the CY phase on MCY , the orbifold phase on
M1r<0 and the 3rd phase on M2r<0 . Furthermore, we have found four effective theories.
Let us discuss the relations among them:
The effective theory on the CY phase (2.11) and the theory on the 3rd phase (2.17)
are related to each other via a topology change, because the defining equations of the
target spaces are equal except for the sign of the FI parameter. Furthermore, these two
effective theories are both sigma models, which do not include the potential theory
sectors such as a LG theory.
(2.16) and (2.17) are connected at the point p1 = si = 0 in Morbifold . Since the former is the sigma model and the latter is the LG theory, there exists a phase transition
between these two theories, which are equivalent to each other by the CY/LG correspondence.
Both the theories (2.15) and (2.16) are on the weighted projective space and are included in the theory (2.14).
The LG theory (2.15) is equivalent to the sigma model (2.11) by the CY/LG correspondence.
Notice that the sigma model (2.11) is not related to (2.16) directly, because the theory
(2.16) has already been connected to (2.17) while the CY/LG correspondence connects
between two theories by one-to-one. These connections are realized through the singularity
phase. Even though we wrote down the connections only from the qualitative point of view,
we can acquire non-trivial relations among the effective theories as illustrated in Fig. 3.
In the case of compact CY manifolds, we have already understood that the local rings
in the LG theory are identified with the chiral rings of the SCFT and these chiral rings are
related to the harmonic forms on such manifolds [15]. However, we have no proof that this
relation is also satisfied in the case of noncompact CY manifolds. Thus we must investigate
the spectra of the above effective theories as a future problem.
As discussed before, we have obtained various massless effective theories by decomposing all massive modes. Thus they are just approximate descriptions which must be
deformed if we can exactly integrate out massive modes. In the next section we will study
the T-dual theory of the GLSM [16]. This formulation is so powerful to obtain the exact effective theories. Analyzing them exact theories we will reinvestigate the massless effective
theories in the original GLSM.
180
Fig. 3. The relation among various phases around the singularity: a conjecture.
3. T-dual theory
In this section we consider T-dual of GLSMs. It is quite significant to study it because
we can obtain exact descriptions of the low energy effective theories. Furthermore, they
will also indicate how the exact effective theories are realized in the original GLSM. In
fact, in the original GLSM, we obtained just approximate effective theories. There we did
not perform integrating-out but just decomposed all massive modes because it is generally
impossible to integrate them out. In the model proposed in [16], we calculate a function
which is directly related to the partition function. Thus we will obtain exact effective theories as quantum field theory.
3.1. General construction
Here we briefly review the T-duality of a generic GLSM without any superpotentials
[16]. We start from the following Lagrangian in two-dimensional worldsheet:

1
1
2Qa V +Ba
L = d 2 +
(Ya + Ya )Ba
e
2
e
a

1
d2 (t) + c.c. ,
+
2
(3.1)
where Ya are twisted chiral superfields whose imaginary parts are periodic of period 2 .
We incorporate real superfields Ba as auxiliary fields.
Integrating out twisted chiral superfields Ya , we obtain D + D Ba = D+ D Ba = 0,
whose solutions are written in terms of chiral superfields a and a such as Ba = a + a .
When we substitute them into the Lagrangian (3.1), a GLSM Lagrangian appears:

1
+
a e2Qa V a
d4 2
e
a

1
2
d (t) + c.c.
+
2
LGLSM ,
181

L B
a
a =a +
(3.2)
where we rewrote a := ea . On the other hand, when we first integrate out Ba in the
original Lagrangian (3.1), we obtain

Ya + Ya
.
Ba = 2Qa V + log
(3.3)
2
Let us insert these solutions into (3.1). By using a deformation

1
1
d4 Qa V Ya = Qa d2 D + D V Ya = Qa d2 Ya ,
2
2
we find that a Lagrangian of twisted chiral superfields appears:

1
1

(Ya + Ya ) log(Ya + Ya )
LT = d 4 2
2
e
a

1
d2 W + c.c. ,
+
2

W =
Qa Ya t +
eYa .
a
(3.4a)
(3.4b)
This Lagrangian is T-dual of the gauge theory (3.2). Notice that the twisted superpotential
W is corrected by instanton effects where the instantons
are the vortices of the gauge

theory. In attempt to analyze a model satisfying a Qa = 0, the scale parameter is
omitted by field re-definitions. Relations between chiral superfields a in (3.2) and twisted
chiral superfields Ya in (3.4) are
2 a e2Qa V a = Ya + Ya .
(3.5)
We can see that the shift symmetry Ya Ya + 2i comes from the U (1) rotation symmetry
on a . In the IR limit e , becomes non-dynamical and generates a following
constraint from W :

Qa Ya = t,
a
which corresponds to the condition D = 0 in the original GLSM.

In this formulation it is convenient to incorporate a function defined by

dYa exp(W ),
:= d
(3.6)
where W is defined in (3.4b). When we consider low energy effective theories of the theory
(3.4), we take the gauge coupling constant to be infinity e . In this limit is no
182
longer dynamical and becomes just an auxiliary field. Thus the function (3.6) is rewritten
by integrating-out of :

=
dYa
Qa Ya t exp
eYa .
a
Via suitable field redefinitions, we can read a LG theory of twisted chiral superfields. Moreover, we also obtain a period integral of a mirror pair of the manifold which appeared in
the effective theories in the original GLSM. Thus we often refer the function (3.6) to the
period integral.
Suppose the theory is topologically A-twisted [39]. In the topologically A-twisted theory, twisted chiral superfields are only valid while the other fields such as chiral superfields
and real superfields are all BRST-exact. Due to this the Lagrangian is reduced only to the
twisted superpotential and the partition function is obtained as the integral of weight eW .
This is nothing but the period integral defined in (3.6). Thus as far as considering the Atwisted sector, the effective theories derived from this function are exact.
Unfortunately, no one knows an exact formulation of a T-dual Lagrangian LT of a
GLSM with a generic superpotential WGLSM . This is partly because the above formulation
is only powerful when we consider T-duality of topologically A-twisted GLSMs. As mentioned above, any deformations of WGLSM are BRST-exact in the topological A-twisted
theory. However, even though in the A-theories, we can analyze T-dual theories of specific
GLSMs with superpotentials of type WGLSM = P G (S), where G (S) is a homogeneous
polynomial of degree with respect to chiral superfields S. In order to do this, let us deform
the above period integral (3.6) to

dYa () exp(W ).
= d
(3.7)
a
This function can be derived through the discussions of Cecotti and Vafa [40], and Morrison and Ronen Plesser [41]. Here we omit a precise derivation. Please see it in [16].
In this formulation we also take the IR limit e and integrate out the superfield ,
because we want to obtain the mirror dual descriptions of the effective theories of the
original GLSM with superpotential. However, the factor in (3.7) prevents from exact
integrating-out of . Thus we need to replace this factor to other variable which does not
disturb the integration. If we wish to obtain the LG description we replace the factor to the
differential with respect to the FI parameter such as t . On the other hand when
we derive the mirror geometry we replace this to the differential operator of an appropriate
twisted chiral superfield derived from the negatively charged chiral superfield, for example,
YP , where YP is the twisted chiral superfield of the chiral superfield P of charge
. The resulting geometry has a Z -type orbifold symmetry. For example, we start from
the GLSM for quintic hypersurface CP4 [5]. Performing T-duality and taking the IR limit,
we obtain not only the mirror dual geometry CP4 [5]/(Z5 )3 but also its LG description [16].
This procedure is so powerful that we develop it in order to obtain the mirror descriptions
of the noncompact CY manifolds. In Sections 3.3 and 3.4 we will study how to obtain LG
theories defined by the twisted superpotential and mirror dual geometries, respectively. We
can also obtain another geometry with a different orbifold symmetry if we replace to
the differential of other suitable twisted chiral superfield.
183
3.2. Field configuration

Let us analyze the T-dual theory of the GLSM for the O(N +) bundles on CPN 1 [].
The field configuration is assigned as follows:
Chiral superfield a
U (1) charge Qa
Twisted chiral superfield Ya
S1
1
Y1
. . . SN
... 1
. . . YN
P1

YP1
P2
N +
YP2
(3.8)
The twisted chiral superfields Ya are periodic variables Ya Ya + 2i. They are defined
from the chiral superfields a via (3.5). As we have already discussed, the twisted superpotential W and the period integral , given by the followings, play key roles:

W =
N

Yi YP1 (N )YP2 t +
i=1

d
N

N
eYi + eYP1 + eYP2 ,
(3.9a)
i=1
dYi dYP1 dYP2 () exp(W ).
(3.9b)
i=1
Let us take the IR limit e in order to consider the low energy effective theories.
It is clear that the dynamics of is frozen and this superfield becomes just an auxiliary
superfield. Thus we must replace the factor in the period integral (3.9b) to appropriate
variables.
3.3. Mirror LandauGinzburg descriptions
In this subsection we will derive LG theories with orbifold symmetries. In order to do
this, we change the variable in the period integral (3.9b) to

.
t

This replacing can be easily performed because of the existence of the term ( a Qa Ya
t) in . Then we integrate out the superfield and obtain
=

N
dYi dYP1 dYP2

Yi YP1 (N )YP2 t
i=1

YP1
YP2
Yi
.
e
e
e
exp
(3.10)
Next let us solve the -function in this function. We note that there are two ways to solve
it. One is to write the variable YP1 in terms of Yi and YP2 . The other is to solve YP2 by Yi
and YP1 . Both two solutions give consistent LG theories with orbifold symmetries.
184
3.3.1. Solution one: Z orbifolded LG theory

Let us solve the variable YP1 via the -function in (3.10):

N

1
YP1 =
Yi + (N )YP2 .
t

i=1
Performing the t-derivative in (3.10) after the substitution of this solution, we obtain
= et/

N
1Y
N

e i dYi e YP2 dYP2
i=1

1 N
YP2
YP2
Yi
t/
Yi

exp
e
e
e
e
e
.
i
It is clear that the integral measure is not canonical. Transforming the variables into
Xi := eYi and XP 2 := e(N )YP2 , we obtain the following period integral with a canonical
measure up to an overall constant:2

N

=
dXi dXP2 exp
Xi XP2N et/ X1 XN XP2 .
(3.11)
i
i=1
Since Ya are periodic with respect to the shifts of their imaginary parts Ya Ya + 2i, the
new variables Xi and XP2 are symmetric under the following phase shifts:
Xi i Xi ,
XP2 P2 XP2 ,
i = P2N = 1 2 N P2 = 1. (3.12)
We can read from (3.11) and (3.12) that the following orbifolded LG theory appears:

N

Xi + XP2N + et/ X1 XN XP2 (Z )N .
(3.13)
W =
i=1
This theory is still ill-defined from the minimal model point of view. Even though the terms
of positive powers such as Xi are well-defined and they consist of N = 2 LG minimal
model, there exists a term XP2N , which does not generate any critical points at finite XP2 .
However, there is an interpretation to avoid this difficulty. Recall a discussion on the linear
dilaton CFT and the Liouville theory [42,43]. (We prepare a brief review in Appendix C.)
Based on this argument, we can interpret the negative power term corresponds to Z0k in
(C.4), which gives an N = 2 SCFT on the coset SL(2, R)k /U (1) at level k assigned by

.
N
This assignment is correct because the conformal
weights ra in Appendix C are all
ra = 1/, where n + 1 = N . Thus we obtain r = a ra 1 = N/ 1 1/k, which
gives the above equation. This theory is given as an N = 2 KazamaSuzuki model on
k=
2 It is not serious to ignore an overall constant.
185
the coset SL(2, R)k /U (1) at level k [44], which is the gauged WZW model on the twodimensional Euclidean black hole [21]. Furthermore, this theory is exactly equivalent to
N = 2 Liouville theory of background charge Q2 = 2/k via T-duality [22,43]. We will
continue to argue in later discussions.
3.3.2. Solution two: ZN orbifolded LG theory
In the same analogy of the previous discussion,3 we study the theory of the second
solution

N

1
Yi + YP1 ,
YP2 =
t
N
i=1
which comes from the -function in the period integral (3.10). Substituting this into (3.10),
we find

N
1 Yi

=
e N dYi e N YP1 dYP1
i=1

t
1

N
Yi N
YP1
YP1
Yi
N
.
exp
e
e
e
e
e
i
XiN
Performing the redefinitions

:= eYi and XPN1 := eYP1 , we find that the period
integral has a canonical measure and the ill-defined LG theory with orbifold symmetry
appears:

N
t
N
N

X
+X
+ e N X1 XN XP1 (ZN )N .
W N =
(3.14)
i
P1
i=1
Applying the discussions in Appendix C to the negative power term in the superpotential
W N , we find that the theory is also described by the well-defined LG theory with an
orbifold symmetry coupled to N = 2 KazamaSuzuki model on the coset SL(2, R)k /U (1)
at level k, which is given by
N
2
=k = 2,

Q
where Q is the charge of equivalent N = 2 Liouville theory.
3.4. Mirror geometry descriptions
In the previous subsection we found two orbifolded LG theories as exact effective theories. They are obtained by solving the twisted chiral superfields YP1 and YP2 , respectively.
Next, we will read geometric informations from the same period integral (3.9b). Here we
will also obtain two solutions which are related to the LG theories. The derivation procedure is so complicated that we try to imitate the method discussed in Section 7.3 of [16]
3 From now on we omit overall constant factors which appear in the period integral.
186
and we develop detailed calculations, explicitly. In order to obtain the geometric informations in the IR limit, we integrate out the superfield in the period integral (3.9b) after the
replacement of in (3.9b) to other variables, as we performed before.
3.4.1. Z orbifolded geometry
Let us study how to obtain the geometry with Z -type orbifold symmetry. Replacing
in the period integral (3.9b) to

,
YP1
we can perform the integration of and obtain

=

N

dYi eYP1 dYP1 dYP2
i=1

YP1
YP2
Yi
.
Yi YP1 (N )YP2 t exp
e
e
e

(3.15)
We perform the redefinitions of the variables Yi , YP1 and YP2 :

eYP1 =: P1 ,
eYP2 =: P2 ,
eYa =: P1 Ua
eYb =: P2 Ub
for a = 1, . . . , ,
for b = + 1, . . . , N.
Substituting these redefined variables into (3.15), we continue the calculation:

N
dUi
d
P
2
dP1
log
Ui + t
=
Ui
P2
i
i=1

N

exp P1
Ua + 1 P2
Ub + 1
a=1
b=+1

dUi
dP2 du dv log
=
Ui + t
Ua + 1
Ui
a
i
i

Ub + 1 uv
exp P2
b

dUi
du dv log
=
Ui + t
Ua + 1
Ub + 1 uv ,
Ui
a
i
i
b
(3.16)
where we introduced new variables u and v taking values in C and used a following equation

1
= du dv exp(P2 uv).
P2
187
It is obvious that the resulting function (3.16) still includes a non-canonical integral measure. Thus we perform further redefinitions such as
Ua =: et/
Za
,
Z1 ZN
Ub =: Zb .
Note that the period integral (3.16) is invariant under the following transformations acting
on the new variables Zi :
Za a Za ,
Zb b Zb ,
a = b = 1 N = 1,
where is an arbitrary number taking in C . The i come from the shift symmetry of the
original variables Yi Yi + 2i. Combining these transformations we find that the period
integral has C (Z )N 2 symmetries. Substituting Zi into (3.16), we obtain

N

N

1

t/

dZi du dv
Za + e Z1 ZN
Zb + 1 uv ,
=
vol(C )
i=1
a=1
b=+1
which indicates that the resulting mirror geometry is described by

= (Zi ; u, v) CN +2 {F(Zi ) = 0}/C , G(Zb ; u, v) = 0 (Z )N 2 ,

M
F(Zi ) :=

Za + Z1 Z ,
G(Zb ; u, v) :=
a=1
N
Zb + 1 uv,
(3.17a)
(3.17b)
b=+1
:= et/ Z+1 ZN .
(3.17c)
is a CY maniThis is an (N 1)-dimensional complex manifold. It is guaranteed that M

fold because of the following reason: we have already
seen
that
the
FI
parameter
t in (3.9b)

does not renormalized owing to the CY condition a Qa = 0, which is also valid in the
T-dual theory. In addition, we took the IR limit e and obtained the above non-trivial
result. This means that the sigma model on the above geometry is a superconformal sigma
model.
defined in (3.17) more in detail. The equation F(Zi ) = 0
Let us study the manifold M
denotes that the complex variables Za consist of the degree hypersurface in the projective
space: CP1 []. This subspace itself is a compact CY manifold, which is parametrized by
a parameter which is subject to the equation G(Zb ; u, v) = 0. Moreover, we can also
interpret that the total space is a noncompact CY manifold whose compact directions are
described by Zi , while the variables u and v run in the noncompact directions under the
equations (3.17b).
and the LG twisted suHere let us comment on a relation between the manifold M
has (Z )N 2 orbifold symmetry,
perpotential (3.13). As we have described in (3.17), M
while the LG theory (3.13) also holds this type of orbifold symmetry, i.e., the (Z )N orbifold symmetry. When we combine the two equations in (3.17b) as follows:
F (Zi , u, v) F(Zi ) + G(Zb , u, v) =
N

i=1
Zi + et/ Z1 ZN + (1 uv) = 0.
188
This function F = 0 is quite similar to the LG twisted superpotential W including negative

power term (3.13). Recall that a LG theory written by a superpotential W is identical with
a CY space defined by W = 0 in a (weighted) projective space. (See, for example, [32,35].)
If we can apply this argument to the above result, the LG theory (3.13) is identical with
the sigma model on (3.17) and there also exists the CY/LG correspondence in the T-dual
theory.
3.4.2. ZN orbifolded geometry
We have constructed the two LG theories: (Z )N orbifolded LG theory and (ZN )N
. It is natural to conorbifolded LG theory. The former is related to the CY geometry M
sider there also exists a dual geometry related to the latter LG theory. In the previous
calculation, we replaced the in the period integral (3.9b) to YP and we obtained the
1
(Z )N 2 orbifolded geometry. Here we replace to the differential with respect to YP2 ,
which is dual of the chiral superfield P2 of charge (N ):

.
N YP2
Substituting this into (3.9b), we obtain the following expression:

=

N

dYi dYP1 eYP2 dYP2
i=1

Yi YP1 (N )YP2 t

exp
eYi eYP1 eYP2 .
(3.18)
Let us perform the following redefinitions of the variables Yi , YP1 and YP2 :
eYP1 =: P1 ,
eYP2 =: P2 ,
eYa =: P1 Ua
eYb =: P2 Ub
for a = 1, . . . , ,
for b = + 1, . . . , N.
Substituting the re-defined variables into (3.18) and introducing auxiliary variables u and
v in order to integrate out P1 completely, we obtain
N

N

dUi
du dv log
Ui + t
=
Ui
i=1
i=1
N

(3.19)
Ub + 1
Ua + 1 uv .
b=+1
a=1
The integral measure still remains non-canonical. We next introduce further redefinitions
of Ui :
Ua =: ZaN ,
Ub =: et/(N)
ZbN
.
Z1 ZN
189
We can see that the map from Zi to Ui is one-to-one modulo the C (ZN )N 2 action
given by
Za a Za ,
Zb b Zb ,
aN = bN = 1 N = 1,
where takes value in C . On account of the above redefinitions and symmetries we find
that the period integral is rewritten as

N

1
N

dZi du dv
Za + 1 uv
=
vol(C )
i=1
a=1
N

N
t/(N)
Zb
+e
Z1 ZN ,
b=+1
from which we can read the geometric information described by

N = (Zi ; u, v) CN +2 F(Za ; u, v) = 0, {G(Zi ) = 0}/C (ZN )N 2 ,
M
(3.20a)

N

ZaN + 1 uv,
G(Zi ) :=
ZbN + Z+1 ZN ,
F(Za ; u, v) :=
a=1
:= et/(N) Z1 Z .
b=+1
(3.20b)
(3.20c)
This is also a noncompact CY manifold including a compact CY hypersurface

CPN 1 [N ], which is defined by G(Zi ) = 0 and parametrized by with being
subject to F(Za ; u, v) = 0. Since the variables are the twisted chiral superfields, we ob N as a low energy effective theory of the
tained the N = 2 supersymmetric NLSM on M
T-dual theory. We can see that the sigma model on this manifold is identical with the LG
theory described by (3.14).
3.5. Return to the gauged linear sigma model
As discussed before, it has been proved that the N = 2 SCFT on coset SL(2, R)k /U (1)
at level k is exactly T-dual with the N = 2 Liouville theory of background charge Q under
the relation Q2 = 2/k. Let us apply this argument to the GLSM and its T-dual. Notice
that the massless effective theories in the T-dual theory are exact, whereas the ones in the
original GLSM are approximately realized.
Now let us recall that if a CFT C has an abelian discrete symmetry group , the orbifold
CFT C = C/ has a symmetry group which is isomorphic to and a new orbifold
CFT C / is identical to the original CFT C. Including this argument into the effective
theories of the GLSM and its T-dual theory of 2 N 1, we find that the theories
(2.15) and (2.16) are equivalent to (3.13) and (3.14), respectively. Furthermore, we can
interpret that the theories (2.15) and (2.16) are described by N = 2 Liouville theories
coupled to the well-defined LG minimal models as exact effective theories. As a result we
obtain the non-trivial relations among the various effective theories in the GLSM. Here
we refer one typical result. The CY sigma model on (2.10) corresponds to (2.15), which
190
is deformed to the LG theory coupled to the Liouville theory as an exact quantum theory.
This is equivalent to (3.13) via T-duality. On account of the CY/LG correspondence, (3.13)
and sigma model on (3.17) are identical with each other. Finally the original CY manifold
(2.10) and (3.17) are mirror dual with each other. Notice that the CY manifold MCY is
also deformed because the Liouville theory indicates that the dilaton field propagates on
the target space [45,46]. Of course, we find that there are the same relations among effective
theories (2.16), (2.17), (3.14) and (3.20).
Let us consider the case = 1. As discussed before, the GLSM has only two massless effective theories (2.11) and (2.25). In addition, the subspace CP1 [] in (3.17) is
ill-defined if = 1 and then the LG description (3.13) is also ill-defined. Thus the T-dual
theory has only two descriptions (3.14) and (3.20) in the IR limit. This situation is consistent with the result in [16], where the GLSM for O(N ) bundle on CPN 1 and its T-dual
was discussed.
4. Summary and discussions

We have studied the GLSM for noncompact CY manifolds realized as a line bundle on
a hypersurface in a projective space. This gauge theory has three non-trivial phases and
includes two types of four massless effective theories in the IR limit. Two theories are
NLSMs on two distinct manifolds, whereas the other two are LG theories coupled to complex one-dimensional SCFTs. Following the conventional arguments, we have interpreted
that these four theories are related to each other under phase transitions such as CY/LG
correspondences and a topology change. Performing the T-duality, we have also obtained
two types of four exact massless effective theories; the two theories are the sigma models
on newly appeared mirror CY manifolds, while the other two are the LG theories including the terms of negative power k, which may be regarded as indicating N = 2 SCFTs
on coset SL(2, R)k /U (1) at level k. Since the SCFT on this coset is exactly equivalent to
the N = 2 Liouville theory via T-duality, we have argued that the LG effective theories
derived from the original GLSM are exactly realized by the Liouville theories coupled to
the well-defined LG minimal models. The relations among the theories are illustrated in
Fig. 4.
Utilizing the above relations, we will obtain the topological charges of a CY manifold
from the exact effective theories in the T-dual theory, even though we cannot directly calculate them in the original sigma model. Furthermore, we will understand noncompact CY
manifolds in detail from the mathematical point of view. In addition, we can interpret the
holographic duality in type II string theory on noncompact (singular) CY manifolds [43]
as the phase transition and T-duality of the two-dimensional worldsheet theory and will
be able to understand this duality more closely in the framework of the worldsheet sigma
model description [4750].
As mentioned in the introduction, we have constructed the noncompact CY manifolds
as line bundles on HSSs [7]. The base spaces HSSs can be seen as the submanifolds in
the projective spaces obtained by polynomials with additional symmetries [6]: the quadric
surface SO(N )/[SO(N 2) U (1)] is given by a polynomial of degree two with SO(N )
symmetry, and E6 /[SO(10) U (1)] has a set of differential equations including E6 isom-
191
Fig. 4. Relations among IR effective theories of GLSM and its T-dual.
etry group. These symmetries give the information of the complex structures of not only
the base spaces HSSs but also the noncompact CY manifolds. However, the T-dual theory
[16] is only valid when we consider the GLSM without a superpotential or with a superpotential given simply by a homogeneous polynomial such as WGLSM = P G (S). Even
though the polynomial G (S) has an additional symmetry, the period integral (3.6) or (3.7)
cannot recognize the existence of this additional symmetry. Thus the T-dual theory does
not map all structures of the CY M to the mirror geometry completely. For example, we
can argue the sigma model on the resolved conifold and its mirror dual in the framework
of GLSM and its T-dual, however, we have not even understood any correct descriptions
for the deformed conifold represented by the GLSM. Therefore, if we wish to obtain the
correct T-dual theories of the sigma models on such noncompact CY manifolds, we must
improve the formulation so that it may recognize the complex structure of the manifold. It
is quite significant to solve this problem in order to understand mirror symmetry for more
general noncompact CY manifolds.
Acknowledgements
The author would like to thank Hiroyuki Fuji, Kentaro Hori, Takahiro Masuda, Shunya
Mizoguchi, Toshio Nakatsu, Kazutoshi Ohta, Takuya Okuda, Makoto Sakaguchi, Hitoshi
Sato, Dan Tomino and Atsushi Yamaguchi for valuable comments. The author also thank
Debashis Ghoshal for correspondence. The author would also like to express the gratitude
to Yukawa Institute for Theoretical Physics (YITP) for the hospitality during the authors
stay. This work was supported in part by the JSPS Research Fellowships for Young Scientists (#15-03926).
Appendix A. Conventions
In this appendix we will write down the notation and convention which are modified from the ones defined in WessBaggers book [51]. In [51] supersymmetric field
192
theory is defined in four-dimensional spacetime. However, in this paper we discuss supersymmetric field theory in two dimensions. Thus let us first perform the dimensional
reduction. The coordinates in two dimensions (x 0 , x 1 ) are related to the four-dimensional
ones (y 0 , y 1 , y 2 , y 3 ):
0 1 0 3
x ,x y ,y .
Note that we perform the dimensional reduction for y 1 - and y 2 -directions. Next we redefine
the irreducible representation for spinors. Weyl spinors in four dimensions becomes
Dirac spinors. For convenience, we define the Dirac spinor indices in two dimensions as
[14]:
1 2 +
, = , ,
(1 , 2 ) = ( , + ),
= + ,
+ = ,
12 = 21 = 1 + = + = 1,
+ +
,
= , = ( + , ).
Under the above convention, the super differential operators D are also changed as follows:
D =
i (0 1 ),

D = + i (0 1 ).

Notice that the ordinary coordinate differentials are defined as 0 /x 0 and 1 /x 1 .

So far we wrote down the convention with respect to the two-dimensional Minkowski
spacetime. When we consider a theory in two-dimensional Euclidean worldsheet, we modify the coordinate x 0 to x 0 = ix 2 .
Under the above convention, let us briefly introduce the following irreducible superfields, i.e., chiral superfields, vector (real) superfields and twisted chiral superfields.
Chiral superfield. As in the case of four-dimensional theory, a chiral superfield is defined by D = 0. We can expand a chiral superfield in terms of the fermionic coordinates
{ , } in the superspace:
(x, , ) = (x) + 2 + + (x) + 2 (x) + 2 + F (x) + ,

where (x) is a complex scalar field, (x) are Dirac spinors and F (x) is a complex
auxiliary field, whose mass dimensions are 0, 1/2 and 1, respectively. The part written by
+ involves only the derivatives of these component fields and .
Vector superfield. A vector superfield is defined by V = V . This setup is also same
in four-dimensional spacetime. We expand a vector superfield under the WessZumino
gauge:
V (x, , )
= + + (v0 + v1 ) + (v0 v1 ) 2 + 2 +

2i + + + + + 2i + + + + 2 + + D.
193
Note that we consider only U (1) gauge theories in this paper, where v0 and v1 are components of a U (1) gauge potential, are gaugino fields as Dirac spinors and D is a real
auxiliary field. The complex fields and are coming from the dimensionally reduced
components of four-dimensional U (1) gauge potential. In general we set this superfield to
be dimensionless.
Twisted chiral superfield. A twisted chiral superfields is also an irreducible superfield
in two dimensions. The definition is D + Y = D Y = 0. Expanding a twisted chiral superfield Y in terms of { , }, we obtain
Y (x, , ) = y(x) + 2 + + (x) + 2 (x) + 2 + G(x) + .

We denote a complex scalar field, Dirac spinors and an auxiliary field as y(x), { (x),
+ (x)} and G(x), respectively. The part + means derivatives of component fields
y(x), (x) and + (x).
We can construct a superfield for the field strength Fmn m vn n vm in the following way:
1
:= D + D V
2
= i 2 + + i 2 + 2 + (D iF01 )
i (0 1 ) i + + (0 + 1 ) + 2 + (0 1 ) +

+ 2 + + (0 + 1 ) + + 02 12 .
This is also a twisted chiral superfield D + = D = 0. This superfield is gaugeinvariant under the U (1) gauge transformation.
Here let us define integral measures of the fermionic coordinates and in the
superspace:
1
d2 := d + d ,
2
1
d2 := d + d ,
2
1
d4 := d + d d + d .
4
Thus the integral over and are obtained as follows:

1
d2 + = .
d2 = 1,
2
These definitions are slightly different from the ones in other papers, for example, [16,17].
We notice that we use the above convention in this paper.
Appendix B. Weighted projective space

In this appendix we discuss a definition of one-dimensional weighted projective space.
The weighted projective space is slightly different from the (ordinary) projective space. As
we shall see, the most significant difference is that there exists an orbifold symmetry in the
weighted projective space, while the projective space does not have this symmetry.
194
First let us review one-dimensional (ordinary) projective space CP1 . We prepare a twodimensional complex plane without the origin such as W = C2 {0}. The coordinates on
this plane are described as (z1 , z2 ). The projective space CP1 is defined as a space whose
coordinate is given by the ratio of the complex variables z1 and z2 . (The complex variables
are called homogeneous coordinates in the projective space.) Under this definition, the two
points (z1 , z2 ) and (z1 , z2 ) in W -plane are identified with each other:
(z1 , z2 ) (z1 , z2 ),
(B.1)
= C {0}. In other words, all the points on the straight line

where is a variable of
through the origin of C2 are identified via the above projection. Note that the projective
space CP1 is diffeomorphic to the two-sphere: CP1 S 2 .
Next we define a weighted projective space WCP1,N . Here we also prepare a twodimensional complex plane W , whose coordinates are expressed by (z1 , z2 ). The weighted
projective space is given as a space of complex coordinate defined by the ratio of z1 and z2
with appropriate weights. The identification in the W -plane is the following:

(z1 , z2 ) z1 , N z2 ,
(B.2)
where both and N are positive integers: , N Z>0 . This identification has a
residual symmetry with respect to the phases such as

z1 , N z2 = (z1 , z2 ),
(B.3)
where = exp(2i) is the phase of and is a great common number of and N
described by = GCM{, N }. This does not exist in the definition of CP1 . In the
case of CP1 , the identification (B.1) fixes the phase of homogeneous coordinates z1 and
z2 completely. On the other hand, the identification (B.2) does not fix the phases of z1
and z2 , and the residual symmetry (B.3) exists. Due to this, roughly speaking, we can
see that the weighted projective space is a projective space with Z orbifold symmetry.
There are two specific points. On the point (z1 , z2 ) = (z1 , 0), the orbifold symmetry is
enhanced to Z , the other point (z1 , z2 ) = (0, z2 ) generates ZN . In addition, if we choose
= N , the weighted projective space can be reduced to the ordinary projective space
WCP1,N = = CP1 , which has no longer an orbifold symmetry.
Appendix C. Linear dilaton CFT and Liouville theory
In this appendix we demonstrate the linear dilaton CFT and the Liouville theory discussed in [42,43]. Let us consider the superstring propagating on the following the tendimensional spacetime:
Rd1,1 X 2n ,
2n = 10 d,
where X 2n is a 2n-dimensional singular CY manifold. Sending the zero string coupling

limit gs 0 at fixed string length ls gives rise to a d-dimensional theory without gravity describing the dynamics of modes living near the singularity on X 2n . This theory is
holographic dual to string theory on a following background which approaches at weak
195
coupling region:
Rd1,1 R M = Rd1,1 R S 1 M/U (1),
where M is a compact and non-singular manifold. The real line R is parameterized by .
We can define CFT on each subspace. On the flat space Rd1,1 we can define N = 1
SCFT whose central charge is
3
cd = d.
2
(C.1)
We describe the theory on R in terms of a linear dilaton given by = Q

2 . The linear
dilaton CFT has a central charge
c = 1 + 3Q2 .
(C.2)
From the consistency of superstring propagation, the worldsheet theory on M should be

an N = 1 SCFT with central charge cM = 3(n 1/2 Q2 ). Moreover, if the manifold has
a U (1) symmetry, the theory on the coset manifold M/U (1) must be an extended N = 2
SCFT with central charge

cM/U (1) = 3 n 1 Q2 .
(C.3)
Let us specialize the N = 2 SCFT on M/U (1) to the N = 2 LG minimal model whose
superpotential is defined in terms of n + 1 chiral superfields Za :

F ra Za = F (Za ), a = 1, 2, . . . , n + 1,
WLG = F (Za ),
where ra are the conformal weights of the chiral superfields Za , respectively. Note that we
have already understood properties of this minimal model [15,31]. The worldsheet central
charge should correspond to (C.3) such as
n+1

(1 2ra ) cM/U (1) .
cLG = 3
a=1
If
weights ra such as r
we introduce a new variable r with respect to the conformal
2 = 2r .
r
1,
we
can
express
the
background
charge
Q
to
Q
a a
Here let us combine the above discussion with the conjectures proposed by Muhki and
Vafa [52], Ghoshal and Vafa [53], and Ooguri and Vafa [54], where they insisted that
an N = 2 SCFT on the noncompact space R M can be given formally by the LG
superpotential
W = Z0k + F (Za ),
(C.4)
where Z0 is an additional chiral superfield and

k=
1
2
= 2.
r
Q
(C.5)
This formulation is useful to describe the sigma model on deformed conifold [53]. The first
term in the superpotential appears to be ill-defined from the LG minimal model point of
view. The corresponding potential does not have a minimum at the finite value of Z0 . The
196
topological LG model with such a superpotential has already been studied by Ghoshal
and Mukhi [55], and Hanany, Oz and Ronen Plesser [56] in order to investigate twodimensional string theory. Moreover, in general, k is not an integer, which makes (C.4)
non-single valued. Thus, it was proposed that this first term can be interpreted as an N = 2
SCFT on the coset SL(2, R)/U (1) at level k. From the geometric point of view, this coset
space corresponds to a semi-infinite cigar, and in the IR limit this geometry deforms to
the two-dimensional Euclidean black hole [21]. This SCFT on the coset had been believed
to be isomorphic to the Liouville theory in the sense of SCFT. They are related by strongweak coupling duality on the worldsheet: the theory (C.4) can be valid as a Liouville theory
in the large Q limit, while this can be seen as a coset SCFT in the large k limit (k = 2/Q2 ).
Finally, it has been proved that the N = 2 SCFT on the coset SL(2, R)k /U (1) is exactly
equivalent (or T-dual) to the N = 2 Liouville theory to each other in any values of k > 0
[22]. This equivalence was also proved by Tong in the framework of two-dimensional domain wall physics in three-dimensional theory [57].
To summarize, we find that the string theory on a singular CY manifold X 2n can be
holographic dual to string theory as a product theory of the N = 2 SCFT on the coset
SL(2, R)k /U (1) and the N = 2 LG minimal model on M/U (1). The coset SCFT sector
is also equivalent to the N = 2 Liouville theory on R S 1 .
References
[1] A. DAdda, P. Di Vecchia, M. Lscher, Confinement and chiral symmetry breaking in CPN 1 model with
quarks, Nucl. Phys. B 152 (1979) 125.
[2] D.J. Gross, A. Neveu, Dynamical symmetry breaking in asymptotically free field theories, Phys. Rev. D 10
(1974) 3235.
[3] D. Friedan, E.J. Martinec, S.H. Shenker, Conformal invariance, supersymmetry and string theory, Nucl.
Phys. B 271 (1986) 93.
[4] E. Witten, Topological sigma models, Commun. Math. Phys. 118 (1988) 411.
[5] B.R. Greene, M. Ronen Plesser, Duality in CalabiYau moduli space, Nucl. Phys. B 338 (1990) 15.
[6] K. Higashijima, M. Nitta, Supersymmetric nonlinear sigma models as gauge theories, Prog. Theor. Phys. 103
(2000) 635, hep-th/9911139.
[7] K. Higashijima, T. Kimura, M. Nitta, Gauge-theoretical construction of non-compact CalabiYau manifolds,
Ann. Phys. 296 (2002) 347, hep-th/0110216.
[8] K. Higashijima, T. Kimura, M. Nitta, CalabiYau manifolds of cohomogeneity one as complex line bundles,
Nucl. Phys. B 645 (2002) 438, hep-th/0202064.
[9] E. Calabi, Mtriques khlriennes et fibrs holomorphes, Ann. Scient. col. Norm. Sup. 12 (1979) 269.
[10] S.-T. Yau, Calabis conjecture and some new results in algebraic geometry, Proc. Nat. Acad. Sci. 74 (1977)
1798.
[11] I.R. Klebanov, E. Witten, Superconformal field theory on threebranes at a CalabiYau singularity, Nucl.
Phys. B 536 (1998) 199, hep-th/9807080.
[12] I.R. Klebanov, M.J. Strassler, Supergravity and a confining gauge theory: duality cascades and SBresolution of naked singularities, JHEP 0008 (2000) 052, hep-th/0007191.
[13] M. Cvetic, G.W. Gibbons, H. L, C.N. Pope, Ricci-flat metrics, harmonic forms and brane resolutions,
Commun. Math. Phys. 232 (2003) 457, hep-th/0012011.
[14] E. Witten, Phases of N = 2 theories in two dimensions, Nucl. Phys. B 403 (1993) 159, hep-th/9301042.
[15] W. Lerche, C. Vafa, N.P. Warner, Chiral rings in N = 2 superconformal theories, Nucl. Phys. B 324 (1989)
427.
[16] K. Hori, C. Vafa, Mirror symmetry, hep-th/0002222.
197
[17] K. Hori, S. Katz, A. Klemm, R. Pandharipande, R. Thomas, C. Vafa, R. Vakil, E. Zaslow, Mirror Symmetry,
Clay Mathematics Monographs, vol. 1, American Mathematical Society, Providence, 2003.
[18] T. Kimura, Towards mirror symmetry on non-compact CalabiYau manifolds, contribution to the Proceedings of the 12th International Conference on Supersymmetry and Unification of Fundamental Interactions
(SUSY 2004), hep-th/0409003.
[19] D. Gepner, Exactly solvable string compactifications on manifolds of SU(N ) holonomy, Phys. Lett. B 199
(1987) 380.
[20] D. Gepner, Spacetime supersymmetry in compactified string theory and superconformal models, Nucl.
Phys. B 296 (1988) 757.
[21] E. Witten, String theory and black holes, Phys. Rev. D 44 (1991) 314.
[22] K. Hori, A. Kapustin, Duality of the fermionic 2d black hole and N = 2 Liouville theory as mirror symmetry, JHEP 0108 (2001) 045, hep-th/0104202.
[23] P.H. Ginsparg, G.W. Moore, Lectures on 2D gravity and 2D string theory, Lectures given at TASI Summer
School (1992), hep-th/9304011.
[24] Y. Nakayama, Liouville field theory: a decade after the revolution, Int. J. Mod. Phys. A 19 (2004) 2771,
hep-th/0402009.
[25] B. Zumino, Supersymmetry and Khler manifolds, Phys. Lett. B 87 (1979) 203.
[26] S.R. Coleman, There are no Goldstone bosons in two dimensions, Commun. Math. Phys. 31 (1973) 259.
[27] N. Seiberg, E. Witten, Electricmagnetic duality, monopole condensation, and confinement in N = 2 supersymmetric YangMills theory, Nucl. Phys. B 426 (1994) 19, hep-th/9407087;
N. Seiberg, E. Witten, Nucl. Phys. B 430 (1994) 485, Erratum.
[28] N. Seiberg, E. Witten, Monopoles, duality and chiral symmetry breaking in N = 2 supersymmetric QCD,
Nucl. Phys. B 431 (1994) 484, hep-th/9408099.
[29] A. Strominger, Massless black holes and conifolds in string theory, Nucl. Phys. B 451 (1995) 96, hepth/9504090.
[30] B.R. Greene, D.R. Morrison, A. Strominger, Black hole condensation and the unification of string vacua,
Nucl. Phys. B 451 (1995) 109, hep-th/9504145.
[31] N.P. Warner, Lectures on N = 2 superconformal theories and singularity theory, in: M. Green, et al. (Eds.),
Superstrings 89, World Scientific, Singapore, 1990.
[32] E.J. Martinec, Algebraic geometry and effective Lagrangians, Phys. Lett. B 217 (1989) 431.
[33] C. Vafa, String vacua and orbifoldized LG models, Mod. Phys. Lett. A 4 (1989) 1169.
[34] C. Vafa, N.P. Warner, Catastrophes and the classification of conformal theories, Phys. Lett. B 218 (1989) 51.
[35] B.R. Greene, C. Vafa, N.P. Warner, CalabiYau manifolds and renormalization group flows, Nucl. Phys.
B 324 (1989) 371.
[36] E.J. Martinec, Criticality, catastrophes and compactifications, in: L. Brink, et al. (Eds.), Physics and Mathematics of Strings, V.G. Knizhnik Memorial Volume, EFI, Chicago, 1989.
[37] K. Intriligator, C. Vafa, LandauGinzburg orbifolds, Nucl. Phys. B 339 (1990) 95.
[38] S. Cecotti, N = 2 LandauGinzburg vs. CalabiYau -models: non-perturbative aspects, Int. J. Mod. Phys.
A 6 (1991) 1749.
[39] E. Witten, Mirror manifolds and topological field theory, hep-th/9112056.
[40] S. Cecotti, C. Vafa, Topological and antitopological fusion, Nucl. Phys. B 367 (1991) 359.
[41] D.R. Morrison, M. Ronen Plesser, Summing the instantons: quantum cohomology and mirror symmetry in
toric varieties, Nucl. Phys. B 440 (1995) 279, hep-th/9412236.
[42] A. Giveon, D. Kutasov, O. Pelc, Holography for non-critical superstrings, JHEP 9910 (1999) 035, hepth/9907178.
[43] A. Giveon, D. Kutasov, Little string theory in a double scaling limit, JHEP 9910 (1999) 034, hep-th/9909110.
[44] Y. Kazama, H. Suzuki, New N = 2 superconformal field theories and superstring compactification, Nucl.
Phys. B 321 (1989) 232.
[45] N. Nakayama, K. Sugiyama, Construction of supergravity backgrounds with a dilaton field, hep-th/0411143.
[46] M. Nitta, Conformal sigma models with anomalous dimensions and Ricci solitons, hep-th/0411149.
[47] T. Eguchi, Y. Sugawara, Modular invariance in superstring on CalabiYau n-fold with A-D-E singularity,
Nucl. Phys. B 577 (2000) 3, hep-th/0002100.
[48] S. Mizoguchi, Modular invariant critical superstrings on four-dimensional Minkowski space twodimensional black hole, JHEP 0004 (2000) 014, hep-th/0003053.
198
[49] J.M. Maldacena, G.W. Moore, N. Seiberg, Geometrical interpretation of D-branes in gauged WZW models,
JHEP 0107 (2001) 046, hep-th/0105038.
[50] K. Hori, A. Kapustin, Worldsheet descriptions of wrapped NS five-branes, JHEP 0211 (2002) 038, hepth/0203147.
[51] J. Wess, J. Bagger, Supersymmetry and Supergravity, second ed., Princeton Univ. Press, Princeton, NJ, 1992.
[52] S. Mukhi, C. Vafa, Two-dimensional black hole as a topological coset model of c = 1 string theory, Nucl.
Phys. B 407 (1993) 667, hep-th/9301083.
[53] D. Ghoshal, C. Vafa, c = 1 string as the topological theory of the conifold, Nucl. Phys. B 453 (1995) 121,
hep-th/9506122.
[54] H. Ooguri, C. Vafa, Two-dimensional black hole and singularities of CY manifolds, Nucl. Phys. B 463
(1996) 55, hep-th/9511164.
[55] D. Ghoshal, S. Mukhi, Topological LandauGinzburg model of two-dimensional string theory, Nucl. Phys.
B 425 (1994) 173, hep-th/9312189.
[56] A. Hanany, Y. Oz, M. Ronen Plesser, Topological LandauGinzburg formulation and integrable structure of
2d string theory, Nucl. Phys. B 425 (1994) 150, hep-th/9401030.
[57] D. Tong, Mirror mirror on the wall: on two-dimensional black holes and Liouville theory, JHEP 0304 (2003)
031, hep-th/0303151.
The one-loop partition function of N = 4

super-YangMills theory on R S 3
Marcus Spradlin, Anastasia Volovich
Kavli Institute for Theoretical Physics, University of California, Santa Barbara, CA 93106, USA
Received 21 September 2004; accepted 5 January 2005
Abstract
We study weakly coupled SU(N ) N = 4 super-YangMills theory on R S 3 at infinite N , which
has interesting thermodynamics, including a Hagedorn transition, even at zero YangMills coupling.
We calculate the exact one-loop partition function below the Hagedorn temperature. Our calculation
employs the representation of the one-loop dilatation operator as a spin chain Hamiltonian acting on
neighboring sites and a generalization of Plyas counting of necklaces (gauge-invariant operators)
to include necklaces with a pendant (an operator which acts on neighboring beads). We find that
the one-loop correction to the Hagedorn temperature is ln TH = +/8 2 .
PACS: 11.15.-q
1. Introduction
The past several years have witnessed a tremendous amount of effort invested into the
careful study of N = 4 super-YangMills (SYM) theory at large N from a number of complementary approaches. One motivation for much of this work is the fact that this theory
is believed to provide, via the AdS/CFT correspondence, the simplest context in which we
might hope to understand how to solve large N gauge theories in four dimensions. Optimistically, the apparent integrability [1] of the string theory dual to N = 4 SYM theory
suggests that it might be possible to calculate all physical quantities (at least at infinite N )
E-mail address: spradlin@kitp.ucsb.edu (M. Spradlin).
doi:10.1016/j.nuclphysb.2005.01.007
200
M. Spradlin, A. Volovich / Nuclear Physics B 711 (2005) 199230
2 N . The successful calculation of the

as exact functions of the t Hooft parameter = gYM
circular BPS Wilson loop for all [2,3] provided an early realization of this hope.
More recently, significant progress has been made on the problem of calculating anomalous dimensions of gauge theory operators. Two related approaches relevant to this problem include the study of semiclassical string solutions following [4], and the study of the
dilatation operator directly as the Hamiltonian of an integrable spin chain following [5,6].
Comprehensive reviews of the most recent progress on these approaches can be found
in [7,8] and [9]. BMN operators provide a prime example of a class of operators whose
anomalous dimensions can be calculated for all [10,11]. Very recent work with partial
results on the problem of summing up all orders in includes [12,13].
In this paper we continue in the fine tradition of chipping away at N = 4 SYM theory
from a variety of angles. Our present interest lies in the partition function of the theory on
R S 3 at infinite N , which displays interesting thermodynamic behavior [14], including
a Hagedorn transition [15], even at zero t Hooft parameter. The partition function can be
calculated exactly at = 0 by simply counting gauge-invariant operators using Plya theory (see, for example, [1518]). Here we calculate the one-loop correction to the partition
function below the Hagedorn temperature and find the result

D2 (xk )
k
P D2 (xk , xm ) ,
+
ln Tr eH 1-loop = 2
(1)
1 z(xk )
4
k=1
m=1
where xn = n+1 en , = e2i (so that m/2 = 1 depending on whether m is even or

odd), and the function z(x) is given below in (6). The remaining quantities D2 (x) and
P D2 (w, y) are traces of the one-loop dilatation operator D2 acting on two neighboring
fields inside an operator. We evaluate these traces explicitly in (87) and (90) below. We
also obtain from (1) the one-loop correction to the value of the Hagedorn temperature,

TH () = TH (0) 1 +
(2)
+ , TH (0) 0.380
8 2
(measured in units where the radius of the S 3 is one), which is presumably another example
of an interesting physical quantity which we might hope to one day calculate as an exact
function of .
We begin in Section 2 by reviewing how the free partition function may be calculated by
first assembling the partition function for the elementary fields into a partition function for
single-trace operators, and then into the full partition function for multi-trace operators.
We discuss the general structure of the one-loop correction to the partition function and
comment on the role of operator mixing and 1/N effects. In Section 3 we rephrase the
problem of counting gauge theory operators into the language of spin chains and introduce
a generalization of Plyas necklace problem to what we call necklaces with a pendant.
such as a spin chain Hamiltonian, inserted
These are necklaces with some local operator O,
at neighboring beads on the necklace. We derive a general formula expressing a partition
function for such necklaces in terms of some basic quantities O(x) and P O(w, y)
In Section 4 we apply this machinery to various familiar
which are easily obtained from O.
subsectors (SU(2), SO(6), and SL(2)) of the N = 4 SYM theory. This section is mostly
a warm-up for Section 5, where we calculate D2 (x) and P D2 (w, y) for the one-loop
201
dilatation operator D2 of the full N = 4 SYM theory. Finally in Section 6 we present our
final results for the one-loop corrections to the single- and multi-trace partition functions
of N = 4 SYM.
2. The N = 4 SYM partition function
We begin in this section with a discussion of weakly coupled N = 4 SYM theory on
R S 3 following [15,17]. Since there is more than one way to calculate the free partition
function, we review here that derivation which best sets the stage for our calculation of the
one-loop correction in the following sections.
2.1. Initial considerations
In general, the partition function is given by

Z() = Tr eH ,
where is the inverse temperature and H is the Hamiltonian of the theory on
convenient to introduce the bookkeeping parameter
x = e = e1/T ,
(3)
R S3.
It is
(4)
which ranges from x = 0 (zero temperature) to x = 1 (infinite temperature). According to

the state-operator map there is a one-to-one correspondence between states of the theory
on R S 3 and gauge-invariant operators on R4 . The Hamiltonian H on R S 3 is identified
with the dilatation operator D on the plane, so counting states weighted by x to the power
of their energy is equivalent to counting gauge-invariant operators weighted by x to the
power of their dimension on the plane. Therefore the partition function can be written as

Z(x) = Tr x D ,
(5)
where we implicitly set the radius R of the S 3 to one. The dimensions of quantities such
as energy and temperature can be restored by affixing the appropriate power of R.
We start by calculating Z(x) at tree level, so the dilatation operator reduces to D =
D0 , which just counts the engineering dimension of an operator. To calculate (5) all we
need to do is write down a complete basis of operators and count them. The most general
gauge-invariant operator can be written as a linear combination of operators with a definite
number of traces, and the most general k-trace operator can be expressed as a product of
k single-trace operators (keeping in mind that separate traces behave as identical particles
and are subject to the appropriate Bose or Fermi statistics). Therefore, a complete basis for
arbitrary gauge-invariant operators follows naturally after we specify a complete basis of
single-trace operators.
2.2. The N = 4 alphabet
A single-trace operator is a product of the fields ( I , a , F ) of N = 4 SYM theory
and their covariant derivatives. Covariant derivatives must always be symmetrized with
202
traces removed, since antisymmetric derivatives can be replaced by field strengths and
traces of derivatives give terms which are zero by the equations of motion. Following
Polyakov [19] we refer to such objects as the letters of the N = 4 alphabet. We will
use A to denote the collection of letters.
We define d(A) of a letter A to be its engineering dimension, so d( I ) = 1, d(a ) =
3/2, d(F ) = 2, and each covariant derivative adds one to the dimension. The enumeration of letters weighted by dimension gives rise to the following elementary partition
function for the N = 4 alphabet [15,17],

2x(3 x )
d(A)
x
=
.
z(x) =
(6)
(1 x )3
AA
This function has a power series which converges for all temperatures 0 x < 1 and can
be understood as follows:
3/2
z(x) =
6x + 16x
+
6( I )
16(a )
+ .
2
30x

5/2
+ 48x
+
70x 3

24(D I )+6(F )
48(D a )
54(D D I )+16(D F )
(7)
There are only 48 D a instead of 64 because 16 components are set to zero by the Dirac
equation. Similarly there are only 54 components of D D I because D 2 I is zero. Finally, 4 of the 24 components of D F are zero by the equations of motion and another 4
are equal to zero by the Bianchi identity.
In what follows we will frequently need to know various partition functions with the
fermion number operator (1)F inserted. These factors can be easily dealt with by making
use of the fact that bosonic and fermionic operators respectively have integer or half-integer
dimensions at tree level. To this end we introduce the quantity
= e2i ,
(8)
which is equal to +1 if it is raised to an integer power and 1 if it is raised to a half-integer

power. To see in action consider the formula

2x(3 + x )
2id(A) d(A)
z(x) =
e
x
=
= 6x 16x 3/2 +
(1 + x )3
AA

(1)F (A) x d(A) .
=
(9)
AA
2.3. Single-trace operators

The next step is to string individual letters of the alphabet A together to form singletrace operators. The quantity we would like to calculate is the single-trace partition function

Z(x) =
(10)
x D0 (O) .
single-trace O
(Note here the notational distinction between the single-trace partition function Z(x)
and the full partition function in (5), denoted Z(x).) The only constraint on O =
203
Tr[A1 A2 AL ] is the overall cyclic invariance of the trace. The enumeration of such
single-trace operators is identical to the combinatorial problem of counting the number
of distinct necklaces composed of a collection of different types of beads. The solution to
this problem, which we review in the next section, may be expressed in terms of z(x) as
Z(x) = z(x)

(n)
n=1

ln 1 z n+1 x n ,
(11)
where (n) is Eulers totient function which counts the number of integers less than n
which are relatively prime to n. The first term z(x) in (11) is present simply to subtract
away traces of a single letter Tr[A], since these vanish automatically in the SU(N ) theory.
Plugging (6) into (11) gives an expansion which goes as follows:
Z(x) =
2
21x

21(Tr[ I J ])
5/2
+ 96x

96(Tr[ I a ])
+.
3
376x

76(Tr[ I J K ])+144(Tr[ I D
J ])+36(Tr[ I F
(12)
])+120(Tr[a b ])
Two comments are in order. The first is that the result (11) is only valid in the N = limit
of the SU(N ) gauge theory, since we allow arbitrarily high powers of the individual letters.
If N were finite, then trace identities would allow a single trace of more than N letters to
be reexpressed in terms of higher-trace operators, indicating that the basis of operators we
are using would be overcomplete.
The second observation is that the power series expansion of (11) converges only for
0 x < xH , where
xH = 7 4 3 0.072.
(13)
The divergence of the partition function at this value, which corresponds to the temperature
TH ( = 0) =
1
1
=
0.380
ln xH ln(7 + 4 3 )
(14)
(measured in units of R 1 ) has been argued to be the gauge theory dual of the Hagedorn
transition in string theory [15,17]. Hagedorn-like behavior in other free large N systems
was also observed in [20].
2.4. From single-trace to multi-trace operators
Having determined the partition function Z(x) for single-trace operators, it is an easy
combinatoric problem to calculate from this the partition function for an arbitrary number
of traces, since the only constraint is that traces should be treated as indistinguishable
bosons or fermions. The result is

Z(n+1 x n )
Z(x) = exp
(15)
.
n
n=1
204
The partition function Zk (x) for k-trace operators can be extracted by inserting y n into the
sum and then reading off the coefficient of y k in the expansion of the exponential.
From the results reviewed above we can see that the complete partition function of
N = 4 SU(N ) gauge theory at infinite N and zero YangMills coupling is given by the
formula

z(n+1 x n )
1
Z(x) = exp
(16)
.
n
1 z(n+1 x n )
n=1
n=1
The exponential term in (16) is (one over) the partition function for gauge group U (1) and
the infinite product is the partition function for gauge group U (N ), so (16) expresses the
expected fact that
ZSU(N ) =
ZU (N )
.
ZU (1)
(17)
In fact, it is interesting to note that although this analysis has been done at infinite N , the
result (16) remains correct to all orders in 1/N (though certainly not at finite N ). This
is true because the tree-level dilatation operator D0 obviously does not receive any 1/N
corrections, and trace relations are non-perturbative in 1/N .
2.5. Turning on the YangMills coupling
In (16) we have reviewed the complete partition function for SU(N ) N = 4 SYM theory
on R S 3 at infinite N and zero YangMills coupling. The goal of this paper is to calculate
the first-order correction to this result when we turn on the YangMills coupling gYM . The
2 N as
dilatation operator D can be expanded in powers of the t Hooft parameter = gYM
D2 + ,
4 2
so to first order in the partition function (5) is given by
D = D0 +

ln x D

D + D +
= Tr x D0 +
Tr x 0 D2 + .
Z(x, ) = Tr x 0 4 2 2
4 2
Therefore we need to calculate the trace of the one-loop dilatation operator,

Z (1) (x) = Tr x D0 D2 .
(18)
(19)
(20)
Our calculation will proceed by first calculating the one-loop partition function in the
single-trace sector and then assembling together the result for multi-trace operators as in
the previous subsection.
2.6. Operator mixing and 1/N
The tree-level dilatation operator D0 is diagonal in the trace basis, since the engineering
dimension of a k-trace operator is obviously just the sum of the engineering dimensions of
the k individual operators. At one loop this is no longer true. Instead we have a formula of
205
the form
D2 |k =
k

i=1

1
1
Tr[O1 ] D2 Tr[Oi ] Tr[Ok ] + |k 1 + |k + 1,
N
N
(21)
for a generic k-trace operator |k = Tr[O1 ] Tr[Ok ]. At first glance one might be tempted
to disregard the O(1/N) terms since we are working at infinite N . However it is well
known that there are classes of k-trace operators whose one-loop anomalous dimensions
receive non-zero contributions from mixing with k 1-trace
operators [2123]. This occurs
generically for BMN operators, which consist of L N letters. For such operators the
1/N suppression is overwhelmed by the growth of the number of non-planar diagrams [22].
Therefore we will not use N = as a justification to omit the last two terms in (21).
Fortunately we have an even better excuse, which is simply that these terms are nondiagonal in the trace basis and therefore do not contribute to the quantity (20) that we
are computing. The situation becomes more complicated starting at two loops, where the
correction to the partition function involves Tr[x D0 D4 ] and Tr[x D0 D22 ], both of which
have some diagonal terms of order 1/N 2 which cannot necessarily be dropped. (In some
sense this like an order of limits problem, like the one discussed in [24].) Of course, if one
focuses on calculating the planar partition function (as opposed to the large N partition
function), then none of these issues arise and one can ignore the 1/N terms in (21) from
the beginning. Subtleties in the 1/N expansion have been noted in [25].
Note that we have implicitly been using a scalar product on the space of operators which
is diagonal in the trace basis,
k|l kl .
(22)
The N = 4 SYM theory provides, via the state-operator correspondence, a natural inner
product Skl = k|l on the space of operators which is diagonal in the trace basis at infinite
N but receives non-diagonal, trace-mixing corrections beginning at O(1/N ). This operator
mixing does not concern us here since the trace

1
k|x D0 D2 |lSkl
,
Tr x D0 D2 =
(23)
k,l
is completely independent of the scalar product S. Therefore we are free to choose the
most convenient inner product Skl = kl . A non-trivial inner product S appeared in studies
of 1/N corrections to the BMN correspondence, such as [26,27] where matrix elements
of the Hamiltonian were compared between gauge theory and string theory. The utility of
ignoring the gauge theory inner product for the purpose of calculating basis-independent
quantities was emphasized in [28] in the context of calculating eigenvalues of D2 in the
BMN sector.
3. Plya necklaces
In this section we develop the machinery which will reduce the calculation of the oneloop partition function

Z (1) (x) = Tr x D0 D2
(24)
206
in the single-trace sector to the calculation of some elementary traces involving O = D2

acting on two neighboring letters. Roughly speaking, we consider here the problem of
tracing out all but two letters of any gauge-invariant operator, and express the result in
terms of traces of O over the remaining two letters. This is accomplished by translating
the calculation into the language of spin chains, and then using a generalization of Plyas
counting theory.
3.1. Free necklaces
A necklace N of length L is a collection of L objects (A1 AL ) chosen from some
fixed set A A of beads, such that two necklaces are identified if they differ from each
other by a cyclic rotation. It is useful to introduce a counting function d on the beads, and
define d to act additively on the beads of a necklace,
d(A1 AL ) =
L
d(Ai ).
(25)
i=1
The analysis of this section will be general, but of course for the desired application to
N = 4 SYM theory we remember in the back of our mind that we will take A to be the
N = 4 alphabet, and d(A) will be the engineering dimension of the letter A.
A central result of Plyas counting theory is that the necklace partition function

Z(x) =
(26)
x d(N ) ,
N
where we sum over necklaces N of arbitrary length L 1, is given by

Z(x) =

(n)
n=1

ln 1 z x n ,
where z(x) is the generating function for the beads,

z(x) =
x d(A) .
(27)
(28)
AA
The formula (27) is valid when the beads are all bosons. The generalization to include
fermions is straightforward and will be presented below.
3.2. Necklaces with a pendant
Instead of reviewing the elementary derivation of the result (27), we will consider a
useful generalization from which we will recover (27) as a special case. To describe the
generalization that we are interested in, it is useful to think of a necklace of length L as a
spin chain of length L, where on each site of the spin chain the spin vector takes values in
the set A. Of course we have the constraint that only cyclically invariant spin configurations
correspond to necklaces. Therefore, we can recast the calculation of (26) into spin chain
207
language by writing the partition function as

Z(x) =
x d(N ) =

TrL Px d ,
(29)
L=1
where TrL denotes the trace over spin chains of length L and P denotes the projection
operator onto the subspace of cyclically invariant spin configurations. The projector can be
written explicitly as

1
1 + T + T 2 + + T L1 ,
(30)
L
where T is the translation operator which sends site i on the chain to site i + 1 and satisfies
T L = 1.
The generalization of TrL [Px d ] that we would like to consider is to the case

TrL Px d O ,
(31)
P=
where O is any homogeneous operator which commutes with d and which acts only on
two neighboring sites at a time, so that it may be decomposed into the form
O =
L
O i,i+1 ,
O i,i+1 = 11 1i1 O 1i+2 1L .
(32)
i=1
Spin chain Hamiltonians are of course prime examples of such operators, and eventually
we will apply the present machinery to the case O = D2 , but our analysis will continue
to be as general as possible. Since the operator O only connects two neighboring sites at
a time, but can slide all the way around the necklace, we can think of O as a pendant
hanging from two adjacent beads on the necklace.
Given any such operator O, an obvious quantity of interest is the expectation value

x d(A1 )+d(A2 ) A1 A2 |O|A1 A2 .
O(x) = TrAA x d O =
(33)
A1 ,A2 A
We will see that knowing O(x) is almost, but not quite enough information to allow for
the calculation of (31). The other quantity that we will need to know is the permuted trace

w d(A1 ) y d(A2 ) A1 A2 |O|A2 A1 , (34)
P O(w, y) = TrAA P w d1 y d2 O =
A1 ,A2 A
where P is the permutation operator on A A and the two sites are counted with different
variables w and y. Let us now see how to reduce the calculation of (31) to the calculation
of these two quantities.
Since P projects onto cyclically invariant states anyway, the sum over i in (32) is actually redundant. We can affix the pendant to sites (i, i + 1) = (1, 2), and the sum in (32) just
gives a factor of L which cancels the 1/L in (30) giving

d L1

TrL T k x d O 12 .
TrL Px O =
k=0
(35)
208
This trace may be expressed as

L1
TrL Px d O =
A1 AL |T k x d O 12 |A1 AL
k=0 A1 ,...,AL
L1
x d(A1 )++d(AL ) A1 A2 |O|A1+k A2+k
k=0 A1 ,...,AL
L

Ai ,Ai+k . (36)
i=3
Upon contemplating the formula (36), it is clear that after we sum over A3 , . . . , AL ,
only two possible structures can emerge,
(i)
A1 A2 |O|A1 A2
or (ii) A1 A2 |O|A2 A1 ,
(37)
depending on whether k and L are such that the Kronecker delta functions in (36) end up
connecting site 1 + k to site 1 or to site 2. For example, k = 0 clearly gives the former
structure while k = 1 clearly gives the latter. But for general k and L, what is the criterion
which tells us which of the two possibilities (37) we get?
Let us start with site 1 + k and follow it around the necklace using n of the Kronecker
delta functions,
1 + k 1 + 2k 1 + 3k 1 + (n + 1)k.
(38)
Since there are only L 2 delta functions in total, we have 0 n L 2. Now let m =
(L, k) be the greatest common divisor of L and k. If m > 1, then by step n = L/m 1 at
the latest, site 1 + k will have been connected to site
1 + k 1 + (n + 1)k = 1 + (k/m)L = 1 mod L.
(39)
On the other hand, if k and L are relatively prime (m = 1), then there is no n < L such
that 1 + (n + 1)k = 1 mod L, so 1 + k cannot get connected to site 1 and hence must be
connected to site 2. We conclude that we get the trace structure of type (ii) if and only if k
and L are relatively prime, and we get structure (i) otherwise. Let us therefore study these
cases separately.
3.3. Case (i): k and L have a common factor
As a warm-up exercise let us consider the case L = 15, k = 6, so that m = (15, 6) = 3.
Then as we sum over sites 3, 4, . . . , L, the delta functions in (36) sew together the beads of
the necklace as follows:
7 13 4 10 1,
8 14 5 11 2,
9 15 6 12 3 9.
(40)
Each line in this formula can be though of as a strand of the necklace. So the necklace
with L = 15 and k = 6 is composed of m = (15, 6) = 3 strands, and the Kronecker delta
functions in (36) force all of the beads on any given strand to be the same.
209
The strands starting with 1 + k and 2 + k end up respectively at sites 1 and 2, confirming
our earlier analysis. The third line in (40) denotes the following contribution to (36):

x d(A3 )+d(A9 )+d(A15 )+d(A6 )+d(A12 ) A3 ,A9 A9 ,A15 A15 ,A6 A6 ,A12 A12 ,A3
A3 ,A9 ,A15 ,A6 ,A12

x 5d(A3 ) = z x 5 .
(41)
A3
The first two strands in (40) involve the sites 1 and 2 where the pendant is attached, and it
is not hard to see that they end up contributing a factor of

(42)
x 5d(A1 )+5d(A2 ) A1 A2 |O|A1 A2 = O x 5 ,
A1 ,A2
using the definition (33). Combining these results, we find that the total contribution to the
sum (36) for the case L = 15 and k = 6 is

z x5 O x5 .
(43)
The generalization of this analysis is straightforward. A necklace with general L and k
will have m = (L, k) different strands, each with L/m beads. Two of those strands (like
the first two in (40) will involve the sites 1 and 2 and give rise to
L/m
.
O x
(44)
The remaining m 2 strands (note that we are assuming here that m 2), like the last line
in (40), give a factor of
L/m m2
.
z x
(45)
Combining (44) and (45), we conclude that the total contribution to (36) from all k such
that m = (k, L) > 1 is
L1
L/m m2 L/m

z x
O x
.
(46)
k=0
m=(k,L)>1
3.4. Case (ii): k and L are relatively prime

As a warm-up exercise let us consider the case L = 14, k = 5. The Kronecker delta
functions in (36) now sew together the beads
6 11 2,
7 12 3 8 13 4 9 14 5 10 1,
(47)
confirming our general analysis that site 1 + k gets connected to 2, and site 2 + k gets
connected to 1, giving trace structure (ii). The first line indicates a contribution of

A6 ,A11 A11 ,A2 x d(A6 )+d(A11 )+d(A2 ) = x 3d(A2 ) ,
(48)
A6 ,A11
210
while the second line similarly denotes a contribution of x 11d(A1 ) . Putting everything together, we find that for L = 14 and k = 5, (36) reduces to

x 11d(A1 )+3d(A2 ) A1 A2 |O|A2 A1 = P O x 11 , x 3 ,
(49)
A1 ,A2
using the definition (34).

The generalization to arbitrary k and L is straightforward. After n 1 steps in the first
line of (47), site 1 + k will connect to site 1 + nk. Therefore, the length of the first strand
is the smallest non-negative n such that 1 + nk = 2 mod L, or equivalently nk = 1 mod L.
The total contribution to (36) from all k which are relatively prime to L is therefore
L1

P O x Ln(L,k) , x n(L,k) ,
n(L, k) = min{n 0: nk = 1 mod L}.
(50)
k=0
(k,L)=1
Solving nk = 1 mod L for fixed k and L is equivalent to finding n, m such that nk

Lm = 1. Given that k is relatively prime to L, it is clear that a solution exists only if n
and L are also relatively prime. Moreover, it is clear that n L (otherwise subtracting L
from n would give a smaller solution). Finally, the set of {k: (k, L) = 1} is in one-to-one
correspondence with the set of n(L, k), simply because the condition nk = 1 mod L is
symmetric in n and k. Therefore, although n(L, k) is not generically equal to k, the sum in
(50) is equivalent to
L1

P O x Lk , x k .
(51)
k=0
(k,L)=1
3.5. Summary and main result

We now combine the contributions (46) and (51), writing the result as

(k,L)2 L/(k,L)
L1
TrL Px d O =
z x L/(k,L)
O x
k=0
L1

1 L
P O x Lk , x k z x L
O x
,
(52)
k=0
(k,L)=1
where in the first term we omitted the constraint m > 1 from (46) at the expense of subtracting off the extra terms on the second line. Now we can trade the sum over k in the first
term of (52) for a sum over divisors a of L, to write

L/a2 a

O x
(a) z x a
TrL Px d O =
a|L
L1

k=0
(k,L)=1

1 L
P O x Lk , x k z x L
O x
.
(53)
211
At this step let us pause for a moment to explain how to recover the Plya formula
(27) as promised. To this end we need to consider the special case where the operator
O is proportional to the identity matrix, and specifically O = 1/L. This is the correct
normalization which gives rise to O = 1 when plugged into the sum over sites in (32). For
O = 1/L we easily find

1

1
O(x) = z(x)2 ,
P O(w, y) = z(wy).
L
L
The second term in (53) therefore drops out, leaving just

1
L/a
TrL Px d =
(a) z x a
.
L
(54)
(55)
a|L
The sum over L is performed in the usual way, and we obtain

L=1

(n)
ln 1 z x n ,
TrL Px d =
n
(56)
n=1
which is the desired result (27).

Having confirmed that the formula (53) reduces to the known answer for the special case
O = 1, let us now consider operators O which do not depend explicitly on L. In particular
this implies, through (32), that the eigenvalues of O scale linearly with L. Then we can
perform the sum over L > 1 to arrive at

L=2

1
O(x)
O(x n )
+
TrL Px d O =
(n)
n
z(x)
z(x ) 1 z(x n )
n=1
L1
L=2
k=0
(k,L)=1

O(x L )
P O x Lk , x k
.
z(x L )
(57)
The first term is present to subtract off the part of the second term which would correspond
to L = 1, which we omit since it cannot support a pendant (and moreover is irrelevant
in the SU(N ) gauge theory). A final step is to simplify (57) by changing the summation
variable from n to L and combining everything into the main result

TrL Px d O
L=2

O(L+1 x L )

Lk+1 Lk k+1 k
+ L=1 P O
.
x
, x
1 z(L+1 x L )
(58)
L=1 k:(k,L)=1
It is a straightforward exercise to generalize the analysis of the previous subsections to

allow for fermionic beads, and we have included here the appropriate factors of which
keep track of the minus signs appearing when such beads are permuted. (The permutation
operator P is understood to be graded, i.e., P |A1 A2 = (1)F1 F2 |A2 A1 .)
212
The first term in (58) for L = 1 is precisely what one would obtain by making the crude
estimate that the only effect of the projection P onto cyclically invariant states is to insert
a factor of 1/L. The rest of (58) is the detailed correction to this approximation.
In all the cases relevant to N = 4 SYM theory that we study below, we will see that
the second term in (58) is a very small correction in the sense that its contribution to
the coefficient of x n is negligible for large n. In particular, we will find that O(x) and
P O(w, y) converge for all temperatures so that the large temperature behavior of (58) is
dominated by the pole 1/(1 z(x)) in the first term.
4. Examples
We can gain some insight into the formula (58) by applying it to some subsectors of the
gauge theory. The implication of the results presented in this section for the thermodynamics of N = 4 SYM theory is unclear since there is no sense in which the sectors decouple
from each other at finite temperature (we do not consider here the addition of chemical
potentials for various charges). However, we believe this section is a useful warm-up exercise for the more complicated analysis which follows. Moreover, the results given here
for the traces of the SU(2), SO(6) and SL(2) spin chain Hamiltonians may be of interest
from the point of view of integrability in those sectors. An additional subsector which is
of independent interest is the SU(2|4) subsector, which at one loop is isomorphic to the
t Hooft large N limit of the plane-wave matrix model [10]. This subsector is considered
in [29].
4.1. The SU(2) subsector
This subsector consists of all operators of the form
Tr[XXZZZXZZ XXZ],
(59)
where X and Z are two holomorphic scalar fields. The alphabet for this sector is
A = {X, Z}, the dimension formula is d(A) = 1, and the elementary partition function
is z(x) = 2x. In the natural basis {XX, XZ, ZX, ZZ} for A A, the matrix elements of
the permutation operator P and the one-loop Hamiltonian D2 are simply
1 0 0 0
1
0 0 1 0
P =
D2 = (1 P ).
(60)
,
0 1 0 0
2
0 0 0 1
We immediately find

D2 (x) = x 2 ,
(61)
P D2 (w, y) = wy.
Plugging these into the main formula (58) gives, after some simplification, the following
formula for the trace of the Hamiltonian in the SU(2) sector:

1 3x n
.
(n)x n
Tr x D0 D2 = x
(62)
1 2x n
n=1
213
As a check, we used a computer to calculate the trace of the SU(2) spin chain Hamiltonian
for all chains of length L 26 and successfully matched the coefficients in the expansion
of (62) up to order x 26 .
4.2. The SO(6) subsector
This subsector consists of all operators built only out of scalar fields with no derivatives,

Tr I1 IL ,
(63)
Ii = {1, . . . , 6}.
The alphabet is A = { 1 , . . . , 6 }, the dimension formula is d(A) = 1, and the elementary
partition function is z(x) = 6x. In the natural basis |I1 I2 = I1 I2 for A A, the matrix
elements of D2 and P are [5]
1
1
1
I1 I2 |D2 |J1 J2 = I1 I2 J1 J2 + I1 J1 I2 J2 I1 J2 I2 J1 ,
4
2
2
I1 I2 |P |J1 J2 = I1 J2 I2 J1 .
(64)
A simple calculation yields

33
D2 (x) = x 2 ,
2
which leads to the result

27
P D2 (w, y) = wy,
2
D
27
3
n 9 65x
0
Tr x D2 = x
(n)x
2
2
1 6x n
(65)
(66)
n=1
for the trace of the SO(6) Hamiltonian. As a check, we used a computer to calculate the
trace of the SO(6) spin chain Hamiltonian for all chains of length L 11 and successfully
matched the coefficients in the expansion of (66) up to order x 11 .
4.3. The SL(2) subsector
This subsector consists of all operators of the form

Tr D i1 Z D iL Z , in {0, 1, . . .},
(67)
where Z is a single holomorphic scalar field and D is a single holomorphic covariant

derivative. The alphabet is A = {Z, DZ, D 2 Z, . . .}, the dimension formula is d(D i Z) =
i + 1, and the elementary partition function is
z(x) =

AA
x d(A) =

i=0
x i+1 =
x
.
1x
In the basis |i1 i2 = D i1Z D i2Z for A A the matrix elements of D2 are [30]

i1 =j1
1
,
i1 i2 |D2 |j1 j2 = i1 +i2 ,j1 +j2 i1 ,j1 h(j1 ) + h(j2 )
2
|i1 j1 |
(68)
(69)
214
where h(j ) are the harmonic numbers

h(j ) =
j

1
n=1
h(0) = 0.
(70)
The matrix elements of P are obviously

i1 i2 |P |j1 j2 = i1 j2 i2 j1 .
(71)
A simple calculation yields

D2 (x) =
x2
ln(1 x),
(1 x)2

1 wy
(1 w)(1 y)
P D2 (w, y) =
ln
.
2 1 wy
(1 wy)2
(72)
After some simplification, we find for the trace of the SL(2) Hamiltonian the result

Tr x D0 D2 =
(n)
n=1

L=1

xn
ln 1 x n
n
1 2x
xL
1 xL
L

ln 1 x k .
(73)
k=1
(k,L)=1
As a check, we used a computer to calculate the trace of the SL(2) spin chain Hamiltonian
for all chains with total dimension D0 19 and successfully matched the coefficients in
the expansion of (73) up to order x 19 .
5. Traces of the PSL(4|4) spin chain Hamiltonian

We now turn to our next step, which is to apply the result (58) to the calculation of
the one-loop correction to the partition function of N = 4 SYM theory on R S 3 in the
single-trace sector:

Z (1) (x) = Tr x D0 D2 .
(74)
To this end, we compute in this section the quantities

D2 (x) = TrAA x D0 D2 ,

P D2 (w, y) = TrAA P w D0(1) y D0(2) D2
(75)
needed to invoke (58) for the full PSL(4|4) spin chain Hamiltonian D2 .
The calculation of D2 (x) is greatly facilitated by making use of the PSL(4|4) symmetry of N = 4 SYM theory. Since the dilatation operator D2 commutes with this symmetry,
the action of D2 on an arbitrary state can be decomposed into its action on irreducible
representations of PSL(4|4). We therefore begin with a discussion of the relevant representations. Unfortunately, the operator w D0(1) y D0(2) does not commute with the two-letter
PSL(4|4) Casimir, so the calculation of P D2 (w, y) will prove more difficult.
215
5.1. Digraphs in the N = 4 language

The elementary fields and their covariant derivatives which make up the alphabet A of
N = 4 SYM theory constitute the so-called singleton representation of PSL(4|4). The
superconformal primary state is the scalar field I , with quantum numbers
[0, 1, 0](0,0)
(76)
under SL(4) flavor rotations and the SL(2) SL(2) Lorentz algebra. The singleton representation is frequently denoted VF , although we shall continue to refer to it as A for
consistency.
Since the one-loop dilatation operator D2 only acts on two letters at a time and com2
mutes with the two-letter Casimir J(12)
of PSL(4|4), it is sufficient to consider the decomposition of the product of two copies of the singleton representation into irreducible
representations of PSL(4|4). The decomposition is
AA=
Vj ,
(77)
j =0
where Vj is the module whose superconformal primary is an eigenstate of the PSL(4|4)

Casimir with eigenvalue j (j + 1) and quantum numbers
[0, 2, 0](0,0) ,
[1, 0, 1](0,0)
and [0, 0, 0]( j 1, j 1) ,

2
for j 2.
(78)
In the notation of [31], we have

1 1
2 2
A = B[0,1,0](0,0)
,
Vj = C 1,1
1 1
2 2
V0 = B[0,2,0](0,0)
,
[0,0,0]( 12 j 1, 12 j 1)
1 1
4 4
V1 = B[1,0,1](0,0)
,
and
for j 2.
(79)
In linguistics, a group of two successive letters whose phonetic value is a single sound,
such as ng in Yang or th in theory, is called a digraph, so we can think of the Vj as the
digraphs of N = 4 YangMills theory.
It is straightforward to count the primary states in Vj , weighted in the usual way by
x D0 . For j = 0 and j 2 the results can be read off respectively from (6.13) and (5.45)
of [31], or from Tables 7 and 8 of [18]. We did not immediately find the primary content of
V1 in the literature, but the derivation thereof is straightforward and the result is presented
in Appendix A. The results for all Vj can be summarized in the expressions

4
V0 (x) = 4x 2 1 + x (5 x),

7

Vj (x) = x j 1 + x j 1 + (j + 2) x j 1 + 5 x (j + 2)x , j 1.
(80)
Setting x = 1 counts the total number of primaries in Vj , which is
+ 1) for any
j 0.
Note that for j = 0, 1 some powers of x in Vj (x) have negative coefficients. This may
be thought of as a bookkeeping device (explained in [31]) which allows for easily keeping
28 (2j
216
track of fields which are eliminated by the requirement of imposing equations of motion
or conservation laws. One consequence of this is that the full partition function for the
module Vj (including descendants) may be calculated naively with derivatives treated as
if they acted freely, without worrying about equations of motion or conservation laws. The
partition function for the module Vj is therefore simply

Vj (x)
,
TrVj x D0 = TrAA Pj x D0 =
(1 x)4
(81)
where we have defined Pj to be the projection operator Pj : A A Vj .

The Pj form a complete set of orthogonal projection operators, so
Pj = 1.
(82)
j =0
This implies the identity

TrAA Pj x D0 = TrAA x D0 = z(x)2 ,
(83)
j =0
which is indeed satisfied by (80) and (81).

5.2. Simple trace D2 (x)
Here we calculate the expectation value (33) for the one-loop dilatation operator D2 .
The calculation only takes one line since we have all of the machinery in place. In [30] it
was shown that the eigenvalue of D2 in the module Vj is simply the harmonic number
h(j ) =
j

1
n=1
h(0) 0,
(84)
and therefore that D2 may be written as

D2 =
h(j )Pj .
(85)
j =0
From (81) and (85) we immediately have

Vj (x)
h(j )
.
D2 (x) = TrAA x D0 D2 =
(1 x)4
(86)
j =0
Plugging in (80) and performing the sum over j gives

(1 + x )2
2

D2 (x) =
1 4 x + x ln(1 x) x 1 8 x + 2x .
(1 x )6
(87)
217
5.3. Permuted trace P D2 (w, y)

As mentioned above, the calculation of P D2 (w, y) is complicated by the fact the two2 commutes only with the sum D
letter PSL(4|4) Casimir operator J(12)
0(1) + D0(2) , but not
with D0(1) and D0(2) separately. In particular, it does not commute with w D0(1) y D0(2) , so the
beautiful decomposition into the modules Vj that we employed in the previous subsection
is of no use here. This apparent breaking of PSL(4|4) is merely an artifact of our choice
to simplify the calculation of TrL [Px D0 D2 ] by tracing out L 2 sites on the chain and
expressing (58) in terms of the two remaining sites.
A manifestly PSL(4|4)-invariant calculation of TrL [Px D0 D2 ] would proceed as follows. First, one would need to know the collection CL of PSL(4|4) modules which appear
in higher powers of the singleton representation A,

Tr(A A) = Tr AL =
(88)
VI
I CL
(where Tr denotes the projection onto singlets of the cyclic group ZL ). Then PSL(4|4)
invariance guarantees that in each resulting module VI the dilatation operator D2 is proportional to the identity operator, with some calculable eigenvalue hI . The desired trace
would then be given by

hI Tr PI x D0 ,
Tr x D0 D2 =
(89)
L=2 I CL
where PI is the projection operator from Tr(AL ) onto VI .

The decomposition of A A A appears in [32] (see also [9]), where it was used
to determine the one-loop anomalous dimensions of some operators consisting of three
elementary fields. However, it seems quite challenging to implement the procedure outlined
in the previous paragraph for arbitrary L, although of course it would be very interesting
to do so.
Instead, we will proceed by using the matrix elements of the operator D2 , written down
in [30] in a GL(4|4) oscillator basis, and then calculating the desired trace by hand. This
is quite a lengthy calculation, so we begin by presenting the result. The interested reader
can find more details below. We find:

P D2 (w, y)
1
(1 + w )2 (1 + y )2
=
2
(1 w ) (1 y )2 (1 + wy )3 ( w + y )2

(1 w)(1 y) (1 wy )( w + y )2
1
p2 (w, y)
wy p1 (w, y) + ln
2
(1 w)(1 y)
(1 wy)2

1 w (1 + wy )3
1( w y)
p3 (w, y) ,

(90)
ln
2( w+ y)
1 y (1 w)(1 y)
in terms of the three polynomials
218
p1 = 4 16w 1/2 + 7w 16y 1/2 + 22w 1/2 y 1/2 16wy 1/2

+ w 3/2 y 1/2 + 7y 16w 1/2 y + 6wy + w 3/2 y 3/2 ,
p2 = 1 4w 1/2 + w 4y 1/2 + 20w 1/2 y 1/2 20wy 1/2 + 4w 3/2 y 1/2 + y
20w 1/2 y + 42wy 20w 1/2 y + w 2 y + 4w 1/2 y 3/2 20wy 3/2
+ 20w 3/2 y 3/2 4w 2 y 3/2 + wy 2 4w 3/2 y 2 + w 2 y 2 ,
(91)
and
p3 (w, y) = y 2 p2 (1/y, w) = w 2 p2 (y, 1/w).
(92)
It is useful to subject this complicated result to a simple check. When we set w = y = x,

then the calculation of P D2 (x, x) can be done using the group theoretic techniques of
the previous subsection. In particular, we have

Vj (x)
P D2 (x, x) = TrAA P x D0 D2 =
TrVj P x D0 D2 =
(1)j h(j )
,
(1 x)4
j =0
j =0
(93)
where we used the fact that the permutation operator P acts as
in Vj [33,34]. If we
now substitute the expressions for Vj (x) from (80) and perform the sum over j , we obtain

P D2 (x, x)

1
(1 + x )3
=
x 1 7x 1/2 + x + x 3/2 6x 2 + 2x 5/2
4
3
(1 x ) (1 + x)

1 7x 1/2 + 15x 25x 3/2 + 25x 2 15x 3/2 + 7x 3 x 7/2 ln(1 + x) .
(94)
Encouragingly, this expression agrees precisely with that obtained by setting w = y = x in
(90).
Now we begin in earnest the calculation of (90). The first step is to use the matrix
elements of D2 in a GL(4|4) oscillator basis, as presented in [30], to write down a sum
which gives (90):
(1)j

P D2 (w, y) =
4
s1 ,s2 ,p1 ,p2 ,k=0 F1 ,F2 ,j =0

4j
p1 !p2 !
4
4j
(1)
j F1 j F2 j k!(k + 1)!
j
c(n1 + n2 , n1 k j, n2 k j )w 1+s1 /2+p1 /2 y 1+s2 /2+p2 /2

F(1 k, k, s1 , s2 ; 2, 1 k + p1 , 1 k + p2 ; 1)
2

1
1
1
1 si + pi Fi (1 + si )(1 + pi ),
2
2
2
(95)
i=1
where F is the regularized hypergeometric function,

ni = si + pi + Fi ,
(96)
219
and the coefficients c are the matrix elements of D2 given in [30]:

( 1 (n12 + n21 )) (1 + 12 (n n12 n21 ))
1
c(n, n12 , n21 ) = (1)1+n12 n21 2
,
2
(1 + 12 n)

1 n
.
c(n, 0, 0) = h
2 2
(97)
The detailed derivation of (95), which is not entirely straightforward, is presented in Appendix B. The next several steps of the calculation will be shown schematically. The
skeptical reader is free to verify that the power series expansion of (95) agrees with that of
(90) to any desired order.
After several manipulations similar to the ones in Appendix B, Eq. (95) can be cast into
the form

P D2 (w, y) =
(1)n (n + 1)(n + 3)w 1+n/2 y n/2
n=0

(1 + y )2
n+1 1+n/2 2 ln(1 y)
,
Pn (y) + 1 + (1) y
(98)
y
(1 y )2
where Pn (y) is a polynomial in y whose highest term is O(y n+1 ). We did not obtain
an explicit formula for Pn (y), though presumably it could be reverse engineered from the
final answer (90). Instead, we use a trick by investigating the quantity

(1 y ) (1 w ) 2

.
Q(w, y) = P D2 (w, y)
(99)
(1 + y ) (1 + w )

Now we use (98) and the power series expansion

1 w 2
=
dm w m/2 , dm = m,0 + 4m(1)m
1+ w
(100)
m=1
to write
Q(w, y) =

dm (1)n (n + 1)(n + 3)w 1+n/2+m/2 y n/2 Pn (y) + Wn (y) ,
m,n=0
where
Wn (y) = y
n/2
(101)

2 ln(1 y)
1 y 2
.
1 + (1)n+1 y 1+n/2
1+ y
y
(102)
The trick is now to break the power series expansion of Q(w, y) into the upper diagonal
terms (where the power of y is greater than the power of w), the diagonal terms, and the
lower diagonal terms. Since Q(w, y) is a symmetric function, the lower diagonal terms
will be known once the upper diagonal terms are known. Furthermore, the diagonal terms
can be extracted from (94), so all we have left to calculate are the upper diagonal terms.
But since Pn is a polynomial whose highest-order term is y n+1 , we see from (101) that it
220
never contributes to the upper diagonal. Therefore, for purposes of computing the upper
diagonal we are free to omit the Pn (y) term in (101), and make the replacement
Wn (y) Wn (y)|y k : k>1+n/2+m/2 ,
(103)
where the notation means that we write out Wn (y) as a series in y and throw away all
powers of y less than or equal to 1 + n/2 + n/2. This finally gives a formula for Q(w, y)
which is amenable to a calculation in M ATHEMATICA. In this manner, we obtain after a
tedious but straightforward calculation the result (90).
6. One-loop N = 4 SYM partition function

6.1. Single-trace
The complete one-loop correction to the partition function of N = 4 SYM theory on
R S 3 in the single-trace sector is given by substituting (87) and (90) into (58). We will
not rewrite the formulas because no significant simplification seems to occur. Instead, let
us note that the result has the expansion

ln x D0 ln x 2
3x + 48x 5/2 + 384x 3 + 2064x 7/2 + .
(104)
Tr x D2 =
2
2
4
4
(Each coefficient receives contributions from both kinds of traces, D2 and P D2 .) The
first term in (104) encodes the one-loop anomalous dimension of the Konishi operator,
3/4 2 . The second term is 3 16, coming from 16 descendants of the Konishi operator.
The third term is 384 = 6 2 + 20 3 + 104 3, which come respectively from the 6
primary states Tr[ I I J ], which have anomalous dimension 1/2 2 according to Table 3
in [30], the 20 Konishi descendants of the form Tr[ I [ J , K ]], and finally 104 Konishi
descendants which are traces of two elementary fields.
6.2. Multi-trace
Now let us extend the result of the previous subsection to the complete one-loop correction to the partition function, including multi-trace operators. As discussed in Section 2.6,
the diagonal matrix elements of the one-loop dilatation operator act additively on k-trace
operators. Therefore, to go from the single-trace partition function to the multi-trace partition function we can still use the formula (15). Substituting
ln x (1)
(105)
Z (x)
4 2
and expanding to first order in , we find that the first order correction to the multi-trace
partition function is
Z(x) = Z (0) (x) +
(1)
ln x (0) (1) n+1 n

(x) =
Z (x)
Z x
4 2
n=1
(106)
221
(a factor of 1/n is canceled by ln x n = n ln x) where we recall that the tree level result
Z (0) (x) is written in (16).
Now let us plug the result from (58) into (106). The term proportional to P D2 gives
the sum

P D2 n(Lk)+1 x n(Lk) , nk+1 x nk
n=1 L=2 k:(k,L)=1

P D2 b+1 x b , a+1 x a .
(107)
a,b=1
To see why this equation is true, pick any positive a and b, and look at the left-hand side to
see how many times (if any) the term P D2 (a+1 x a , b+1 x b ) appears. This is equivalent
to asking how many solutions, for given a and b, there are to the equations
nk = a,
nL = a + b,
(108)
for L 2 and k: (k, L) = 1. The answer is that there is always precisely one solution:
n = (a, b), L = (a + b)/n and k = a/n. Certainly if n were not the greatest common
divisor of a and b but some smaller common divisor, then (108) would still give solutions
for k and L, but these would not satisfy the constraint (k, L) = 1.
Next we plug the D2 (x) terms from (58) into (106), which gives the sum

D2 (nL+1 x nL )
D2 (k+1 x k )
=
(L)
(L)
1 z(nL+1 x nL )
1 z(k+1 x k )
k=1 L|k
n=1 L=1

D2 (k+1 x k )
.
k
=
1 z(k+1 x k )
(109)
k=1
Combining (107) and (109) into (106) gives the final result

D2 (k+1 x k )

k+1 k m+1 m
ln x (0)
(1)
+
P D2 x ,
,
Z (x)
k
x
Z (x) =
4 2
1 z(k+1 x k )
k=1
k,m=1
(110)
as advertised already in (1), for the one-loop partition function of N = 4 SYM theory on
R S 3 , expressed in terms of the free partition function Z (0) , the elementary partition
function (6), and the traces (87) and (90).
6.3. One-loop Hagedorn temperature
The partition function Z(x) has a simple pole at the Hagedorn temperature
c
Z(x)
(111)
,
xH x
where c is an irrelevant overall numerical coefficient. To compute the one-loop correction
xH to the Hagedorn temperature, we simply expand

c
c
xH
(112)
=
1
+ .
xH + xH x xH x
xH x
222
When we compare this to (110) and recall that xH is such that z(xH ) = 1, we note that
only the k = 1 term in the sum of D2 (x k ) contributes to the double pole at the Hagedorn
temperature.1 Reading off the residue of this pole, we find

2 ln xH
ln x D2 (x)
=
D
(x
)
.
xH = lim (xH x)
(113)
x
H
2
H
xxH
3
4 2 1 z(x)
4 2
Remarkably, we find from (87) that D2 (xH ) = 3/4, which gives
ln xH
xH
=
,
xH
8 2
and hence
TH
1 xH
=
=
.
TH
ln xH xH
8 2
The one-loop Hagedorn temperature is therefore

1
+ , TH (0) =
TH = TH (0) 1 +
.
2
8
ln(7 + 4 3 )
(114)
(115)
(116)
It is encouraging that the sign is positive, consistent with the simple guess that the Hagedorn temperature is a monotonically increasing, smooth function of from TH = T0 at zero
coupling to the AdS/CFT prediction that TH 1/4 at strong coupling.
7. Discussion
In this paper we have presented, in (1), the one-loop correction to the partition function
of SU(N ), N = 4 SYM theory on R S 3 at infinite N and below the Hagedorn temperature.
Several recent papers including [17,3537] address the thermodynamics and phase transition structure of weakly coupled gauge theories with the goal of smoothly connecting
onto the strong coupling predictions implied by the AdS/CFT correspondence. String theory in AdS5 S 5 has both a Hagedorn transition [14] at TH 1/4 and a HawkingPage
transition [38] involving the nucleation of AdS5 black holes at THP = 3/2 . The authors
of [17] lay out the ways in which the phase diagram of the weakly coupled theory can
be matched onto these strong coupling predictions. Various qualitatively different phase
diagrams can in part be distinguished by the sign of a particular coefficient in the effective action for the Polyakov loop U , which is the order parameter for the phase transition.
The computation of this coefficient requires a three-loop calculation in thermal YangMills
theory on S 3 which is in progress [39].
One of the motivations for the present work was the desire to provide an independent
check of some pieces of the calculation of [39] from a completely orthogonal starting
1 It is possible that the second term in (110) develops a pole at the Hagedorn temperature after evaluating the
sum, in which case it would add a finite correction to the one-loop Hagedorn temperature. However, this does
not happen in any of the subsectors that we studied, and numerical evidence suggests that it does not occur here
either.
223
point. The one-loop calculation in this paper is equivalent to a two-loop calculation in

thermal YangMills theory and clearly has some, but not complete, overlap with the work
of [39]. In one sense our calculation contains less information than the effective action
for the Polyakov loop because we integrate out all degrees of freedom in the theory, including U . Moreover, our method is only applicable for temperatures below the Hagedorn
transition. On the other hand, our calculation contains more information since we have
also separately calculated, in (104), the one-loop correction to the partition function in the
single-trace sector. It requires extra work because the trace basis it not particularly natural
at finite temperature, and the result is slightly messy because of the appearance of number
theoretic quantities, such as the Euler function (n) or the condition k: (k, L) = 1 in (58).
These always disappear at the end of the day in any formula (such as (16) or (110)) which
includes arbitrary-trace operators.
The finer information present in the single-trace result (104) is deeply related to recent
studies of integrability in N = 4 SYM theory because the one-loop correction to the partition function is essentially just the trace of the one-loop dilatation operator, and therefore
encodes the sum of all anomalous dimensions in the theory (sorted according to bare dimension by the powers of x D0 ). The one essential subtlety is that only cyclically invariant
spin configurations correspond to gauge theory operators, because traces of elementary
YangMills fields are automatically cyclically invariant. Although the spin chain Hamiltonian only acts on two neighboring sites at a time, the projection onto cyclically invariant
states induces an effective long-range interaction on the spin chain which is irrelevant at
temperatures near the Hagedorn transition but dominates the partition function, and significantly complicates the calculation thereof, at low temperatures. Interestingly, the fact that
the PSL(4|4) Hamiltonian is integrable [6] played absolutely no apparent role in our calculation. It would be very interesting to understand our result in the context of integrability.
It would also be interesting to consider higher-loop corrections to our results. Although
the dilatation operator of the full N = 4 theory is only known to one loop, in the SU(2)
sector its precise form is known up to three loops (at the planar level). Furthermore, depending on the assumptions (such as integrability) that one is willing to make, one can go
all the way to five loops [9].
Acknowledgements
It is a pleasure to thank O. Aharony, S. Minwalla, H. Reall and R. Roiban for useful
discussions, correspondence, and comments on the manuscript. We are grateful to J. Plefka
for pointing out a flaw in our original proof of Eq. (51). This research was supported in
part by the National Science Foundation under Grant No. PHY99-07949.
Appendix A. The module V1

We tabulate here the SL(4) SL(2) SL(2) decomposition of the primary states in the
1 1
4 4
module V1 , using the notation of [31] (where it is referred to as B[1,0,1](0,0)
).
224
(A.1)
With the help of the SL(4) SL(2) SL(2) dimension formula
dim[k, p, q](j1 ,j2 ) =
1
(k + p + q + 3)(k + p + 2)(p + q + 2)(k + 1)
12
(p + 1)(q + 1)(2j1 + 1)(2j2 + 1)
(A.2)
we can immediately read off the partition function for primary states in V1 ,
V1 (x) = 15x 2 + 96x 5/2 + 252x 3 + 336x 7/2 + 210x 4 + 0x 9/2 84x 5
48x 11/2 9x 6 ,
(A.3)
in agreement with the result written in (80).
Appendix B. Some details

In this appendix we show how to obtain the formula (95) from the matrix elements of
D2 given in [30].
225
B.1. The GL(4|4) oscillator basis

The GL(4|4) oscillator basis for A is realized by a set of four bosonic oscillators a , b
(, {1, 2}) and four fermionic oscillators ca (a {1, 2, 3, 4}) with the usual relations

a
a , a = ,
(B.1)
b , b = ,
c , cb = ba ,
and a vacuum |0 annihilated by all of the lowering operators. The only constraint on
physical states is that they should be annihilated by the central charge
1
1
1
C = 1 a a + b b ca ca .
2
2
2
The tree-level dilatation operator corresponds to
(B.2)
1
1
D0 = 1 + a a + b b .
(B.3)
2
2
To consider two letters A A we simply have two copies the above algebra, indexed
by a subscript (i) {(1), (2)}. A general state in A(i) will be labeled by its oscillator
1 , a 2 , b1 , b2 , c1 , c2 , c3 , c4 ) N . In this basis we have
occupation numbers (a(i)
(i)
(i) (i) (i) (i) (i) (i) (i)
1 1
1 1

1 1
2
2
2
3
4
a(i) + a(i)
+ b(i) + b(i)
c(i) + c(i)
,
+ c(i)
+ c(i)
2
2
2
1 1

1 1
2
2
+ b(i) + b(i)
.
+ a(i)
D0(i) = 1 + a(i)
(B.4)
2
2
Then to calculate a trace TrA(i) literally means that we perform a sum of the form
C(i) = 1

A(i)
(C(i) ) =
N(i)
1
1 ,a 2 ,b1 ,b2 =0
a(i)
(i) (i) (i)
1 ,c2 ,c3 ,c4 =0

c(i)
(i) (i) (i)
(C(i) )
(B.5)
over all possible oscillator numbers, subject to the physical state constraint. As a check, it
is straightforward to confirm using this formula and (B.4) that

D
2x(3 x )
TrA x D0 =
(B.6)
x 0=
3 ,
(1 x )
A
in agreement with (6).

Now we consider the action of the one-loop dilatation operator D2(12) on two sites,
following [30]. If we let AI = (a , b , ca ) schematically denote all of the raising operators,
then a general state in A A can be written as

s1 , . . . , sn ; {Ii } = A
(B.7)
I1 (s1 ) AIn (sn ) |0,
where si {1, 2} indicates on which site the oscillator acts. The dilatation operator does
not change the type or number of elementary GL(4|4) oscillators, but can only cause them
to hop from site 1 to 2 or vice versa according to the rule given in [30]:
226

D2(12) s1 , . . . , sn ; {Ii }

c(n, n12 , n21 )(C(1) )(C(2) )s1 , . . . , sn ; {Ii } ,
=
(B.8)
s1 ,...,sn =1,2
where n12 , n21 count the number of oscillators hopping from site 1 to 2 or vice versa and the
coefficients c(n, n12 , n21 ) are given in (97). In what follows we will consider a generalized
operator of the form (B.8),

n12 n21
q n q12
q21 s1 , . . . , sn ; {Ii } ,
Qs1 , . . . , sn ; {Ii } =
(B.9)
s1 ,...,sn =1,2
n12 n21
with matrix elements q n q12
q21 instead of c(n, n12 , n21 ).
B.2. The combinatorics of hopping

We are therefore interested in the studying the combinatorics of oscillators hopping
between two sites. Consider first a toy system with just a single bosonic oscillator a. The
most general state in A(1) A(2) would then be
n

a a
|a(1) , a(2) = a(1) (1) a(2) (2) |0 =

a(s
|0,
i)
(B.10)
i=1
with
n = a(1) + a(2) ,
si = {1, . . . , 1, 2, . . . , 2}.

a(1)
(B.11)
a(2)
A quantity of interest is the matrix element

ha(1) ,a(2) (q, q12 , q21 ) = a(1) , a(2) |P Q|a(1) , a(2)
=
2

s1 ,...,sn =1
n12 n21
q n q12
q21 a(1) , a(2) |P
n

i=1
a(s
) |0,
(B.12)
where Q is defined in (B.9). The function h counts the number of ways that the initial state
|a(1) , a(2) is mapped to itself, up to a permutation P , under the action of an operator of the
form (B.8), weighted according to the powers of q12 and q21 which tell us the number of
oscillators that have hopped from 1 to 2 or vice versa. An elementary combinatoric analysis
reveals that

a(1) a(2) a(1) +a(2) a(1) a a(2) a
q
q12
q21
ha(1) ,a(2) (q, q12 , q21 ) =
a
a
a=0

= (qq12 )a(1) (qq21 )a(2) F a(1) , a(2) , 1, (q12 q21 )1 , (B.13)
where F is the hypergeometric function.
A similar analysis can be done for the case of a single fermionic oscillator c. For a
general state |c(1) , c(2) we find

c(1) c(2)
.
gc(1) ,c(2) (q, q(12) , q(21 ) = (qq12 )c(1) (qq21 )c(2) 1
(B.14)
q12 q21
227
Since fermionic oscillators only have occupation numbers which are 0 or 1, this formula
encodes the four cases
g0,0 = 1,
g0,1 = qq21 ,
g1,0 = qq12 ,
g1,1 = q 2 (q12 q21 1).
(B.15)
For g0,0 there are no oscillators, so there is no hopping possible. For g0,1 we start with one
oscillator on site 2, which moves to site 1 giving a factor of q21 . In the final case, g1,1 we
have one fermionic oscillator on each site. They can either stay where they are, or they can
flip, accounting for the two terms in g1,1 .
For a system with multiple oscillators we simply multiply together the appropriate partition functions (B.13) and (B.14) for the individual oscillators. For GL(4|4) this gives
PN(1) ,N(2) (q, q12 , q21 ) = N(1) , N(2) |P Q|N(1) , N(2)
2
2
4

a
a
.
=
h a(1) , a(2)
h b(1) , b(2)
g c(1)
, c(2)
=1
=1
(B.16)
a=1
The quantity

R(w, y; q, q12 , q21 ) = TrAA P w D0(1) y D0(2) Q

(C(1) )(C(2) )w D0(1) y D0(2) PN(1) ,N(2) (q, q12 , q21 )
=
N(1) ,N(2)
(B.17)
then represents a trace over A A which counts all of the possible hoppings between
a state |N(1) , N(2) and its permutation |N(2) , N(1) , weighted appropriately by w to the
power of the dimension of site 1, times y to the power of the dimension of site 2, times q
to the power of the total number of oscillators on both sites, times q12 to the power of the
number of oscillators hopping from site 1 to 2, times q21 to the power of the number of
oscillators hopping from site 2 to 1. Concretely, we obtain from (B.5), (B.13) and (B.14)
the formula
R(w, y; q, q12 , q21 )
1
1 ,a 2 ,b1 ,b2 ,a 1 ,a 2 ,b1 ,b2 =0

a(1)
(1) (1) (1) (2) (2) (2) (2)
1 ,c2 ,c3 ,c4 ,c1 ,c2 ,c3 ,c4 =0

c(1)
(1) (1) (1) (2) (2) (2) (2)
(C(1) )(C(2) )
(qq12 )n(1) (qq21 )n(2) w D0(1) y D0(2)

4
2
a ca

c(1)

(2)
F a(1)
, a(2)
, 1, (q12 q21 )1
1
q12 q21
a=1
=1
2

F b(1)
, b(2)
, 1, (q12 q21 )1
(B.18)
=1
where
n(i) =
2

=1
a(i)
+
2

=1
b(i)
+
4

a=1
a
c(i)
(B.19)
228
denotes the total number of oscillators on site i.

The first step in simplifying (B.18) is to rearrange the bosonic sums according to
s(i)

1 2
f a(i) , a(i) =
f (t(i) , s(i) t(i) ).
(B.20)
s(i) =0 t(i) =0
1 ,a 2 =0
a(i)
(i)
(For the b oscillators we will use s and t as the new summation variables.) After this
substitution, the t variables only appear in the arguments of the hypergeometric functions.
The sum over t(1) and t(2) can be done with the help of the identity
s(1) s(2)

F (t(1) , t(2) , 1, z)F (s(1) + t(1) , s(2) + t(2) , 1, z)
t(1) =0 t(2) =0
= (1 + s(1) )(1 + s(2) )F (s(1) , s(2) , 2, z).
(B.21)
The sum over fermionic occupation numbers can be similarly simplified with the formula
4
4
1
4

a
a
a a
1 c(1)
f
c(1) ,
c(2)
c(2) z
1 ,c2 ,c3 ,c4 ,c1 ,c2 ,c3 ,c4 =0
c(1)
(1) (1) (1) (2) (2) (2) (2)
4
4
F(1) =0 F(2) =0
a=1
a=1
a=1
4

4j
4j
j j 4
.
f (F(1) , F(2) )
(1) z
j F(1) j F(2) j
(B.22)
j =0
The results (B.20) and (B.22) allow (B.18) to be written as

R(w, y; q, q12 , q21 )
=
4
(1)j zj
s(1) ,s(2) ,s(1) ,s(2) =0 F(1) ,F(2) ,j =0
4
4j
4j
j F(1) j F(2) j
w 1+s(1) /2+s(1) /2 y 1+s(2) /2+s(2) /2 (qq12 )n(1) (qq21 )n(2)

F (s(1) , s(2) , 2, z)F (s(1) , s(2) , 2, z)
2

1
1
1
1 s(i) + s(i) F(i) (1 + s(i) )(1 + s(i) ),

2
2
2
(B.23)
i=1
where
z=
1
,
q12 q21
n(i) = s(i) + s(i) + F(i) .
(B.24)
Of course, we are not particularly interested in the quantity R(w, y; q, q12 , q21 ). It is
simply an auxiliary quantity which counts the possible ways for the operators to hop between sites. We are interested in the trace of (B.8) instead of the trace of (B.9), which we
can calculate via the replacement

P D2 (w, y) = R(w, y; q, q12 , q21 )|q n q n12 q n21 c(n,n12 ,n21 ) ,
(B.25)
12
21
229
where the notation means simply that we expand (B.23) in powers of q, q12 and q21 and
then make the substitution indicated for the various powers.
In order to proceed, we expose the powers of q12 and q21 in (B.23) by using the identity
F (s(1) , s(2) , 2, z)F (s(1) , s(2) , 2, z)

s(1) !s(2) !
F(1 k, k, s(1) , s(2) ; 2, 1 k + s(1) , 1 k + s(2) ; 1)
=
zk
k!(k + 1)!
k=0
(B.26)
where F is the regularized hypergeometric function. Combining (B.25), (B.26) and (B.23)
finally leads to the formula (95) for the desired trace, after some simplification of the
notation.
References
[1] I. Bena, J. Polchinski, R. Roiban, Hidden symmetries of the AdS5 S 5 superstring, Phys. Rev. D 69 (2004)
046002, hep-th/0305116.
[2] J.K. Erickson, G.W. Semenoff, K. Zarembo, Wilson loops in N = 4 supersymmetric YangMills theory,
Nucl. Phys. B 582 (2000) 155, hep-th/0003055.
[3] N. Drukker, D.J. Gross, An exact prediction of N = 4 SUSYM theory for string theory, J. Math. Phys. 42
(2001) 2896, hep-th/0010274.
[4] S.S. Gubser, I.R. Klebanov, A.M. Polyakov, A semi-classical limit of the gauge/string correspondence, Nucl.
Phys. B 636 (2002) 99, hep-th/0204051.
[5] J.A. Minahan, K. Zarembo, The Bethe-ansatz for N = 4 super-YangMills, JHEP 0303 (2003) 013, hepth/0212208.
[6] N. Beisert, M. Staudacher, The N = 4 SYM integrable super-spin chain, Nucl. Phys. B 670 (2003) 439,
hep-th/0307042.
[7] A.A. Tseytlin, Spinning strings and AdS/CFT duality, hep-th/0311139.
[8] A.A. Tseytlin, Semiclassical strings in AdS5 S 5 and scalar operators in N = 4 SYM theory, hepth/0407218.
[9] N. Beisert, The dilatation operator of N = 4 super-YangMills theory and integrability, hep-th/0407277.
Mills, JHEP 0204 (2002) 013, hep-th/0202021.
[11] A. Santambrogio, D. Zanon, Exact anomalous dimensions of N = 4 YangMills operators with large R
charge, Phys. Lett. B 545 (2002) 425, hep-th/0206079.
[12] A.V. Ryzhov, A.A. Tseytlin, Towards the exact dilatation operator of N = 4 super-YangMills theory, Nucl.
Phys. B 698 (2004) 132, hep-th/0404215.
[13] G. Arutyunov, S. Frolov, M. Staudacher, Bethe ansatz for quantum strings, JHEP 0410 (2004) 016, hepth/0406256.
[14] E. Witten, Anti-de Sitter space, thermal phase transition, and confinement in gauge theories, Adv. Theor.
Math. Phys. 2 (1998) 505, hep-th/9803131.
[15] B. Sundborg, The Hagedorn transition, deconfinement and N = 4 SYM theory, Nucl. Phys. B 573 (2000)
349, hep-th/9908001.
[16] B. Sundborg, Stringy gravity, interacting tensionless strings and massless higher spins, Nucl. Phys. B (Proc.
Suppl.) 102 (2001) 113, hep-th/0103247.
[17] O. Aharony, J. Marsano, S. Minwalla, K. Papadodimas, M. Van Raamsdonk, The Hagedorn/deconfinement
phase transition in weakly coupled large N gauge theories, hep-th/0310285.
[18] M. Bianchi, J.F. Morales, H. Samtleben, On stringy AdS5 S 5 and higher spin holography, JHEP 0307
(2003) 062, hep-th/0305052.
[19] A.M. Polyakov, Gauge fields and spacetime, Int. J. Mod. Phys. A 17S1 (2002) 119, hep-th/0110196.
[20] M.B. Halpern, On the large N limit of conformal field theory, Ann. Phys. 303 (2003) 321, hep-th/0208150.
230
JHEP 0210 (2002) 068, hep-th/0209002.
[23] N.R. Constable, D.Z. Freedman, M. Headrick, S. Minwalla, L. Motl, A. Postnikov, W. Skiba, PP-wave string
[24] I.R. Klebanov, M. Spradlin, A. Volovich, New effects in gauge theory from pp-wave superstrings, Phys.
Lett. B 548 (2002) 111, hep-th/0206221.
[25] N. Beisert, C. Kristjansen, M. Staudacher, The dilatation operator of N = 4 super-YangMills theory, Nucl.
Phys. B 664 (2003) 131, hep-th/0303060.
[26] D.J. Gross, A. Mikhailov, R. Roiban, A calculation of the plane wave string Hamiltonian from N = 4
super-YangMills theory, JHEP 0305 (2003) 025, hep-th/0208231.
[27] J. Pearson, M. Spradlin, D. Vaman, H. Verlinde, A. Volovich, Tracing the string: BMN correspondence at
finite J 2 /N , JHEP 0305 (2003) 022, hep-th/0210102.
[28] N. Beisert, C. Kristjansen, J. Plefka, M. Staudacher, BMN gauge theory as a quantum mechanical system,
Phys. Lett. B 558 (2003) 229, hep-th/0212269.
[29] M. Spradlin, M. Van Raamsdonk, A. Volovich, Two-loop partition function in the planar plane-wave matrix
model, Phys. Lett. B 603 (2004) 239, hep-th/0409178.
[30] N. Beisert, The complete one-loop dilatation operator of N = 4 super-YangMills theory, Nucl. Phys. B 676
(2004) 3, hep-th/0307015.
[31] F.A. Dolan, H. Osborn, On short and semi-short representations for four-dimensional superconformal symmetry, Ann. Phys. 307 (2003) 41, hep-th/0209056.
[32] N. Beisert, M. Bianchi, J.F. Morales, H. Samtleben, Higher spin symmetry and N = 4 SYM, JHEP 0407
(2004) 058, hep-th/0405057.
[33] L. Dolan, C.R. Nappi, E. Witten, A relation between approaches to integrability in superconformal Yang
Mills theory, JHEP 0310 (2003) 017, hep-th/0308089.
[34] L. Dolan, C.R. Nappi, E. Witten, Yangian symmetry in D = 4 superconformal YangMills theory, hepth/0401243.
[35] L. Fidkowski, S. Shenker, D-brane instability as a large N phase transition, hep-th/0406086.
[36] H. Liu, Fine structure of Hagedorn transitions, hep-th/0408001.
[37] O. Aharony, J. Marsano, S. Minwalla, T. Wiseman, Black hole-black string phase transitions in thermal
(1 + 1)-dimensional supersymmetric YangMills theory on a circle, Class. Quantum Grav. 21 (2004) 5169,
hep-th/0406210.
[38] S.W. Hawking, D.N. Page, Thermodynamics of black holes in anti-de Sitter space, Commun. Math. Phys. 87
(1983) 577.
[39] O.Aharony, J. Marsano, S. Minwalla, K. Papadodimas, M. Van Raamsdonk, in preparation.
Spacetime-filling branes in ten and nine dimensions

Fabio Riccioni
DAMTP, Centre for Mathematical Sciences, University of Cambridge, Wilberforce Road,
Cambridge CB3 0WA, UK
Abstract
Type-IIB supergravity in ten dimensions admits two consistent Z2 truncations. After the insertion
of D9-branes, one of them leads to the low-energy action of type-I string theory, and it can be performed in two different ways, in correspondence with the fact that there are two different consistent
ten-dimensional type-I string theories, namely, the SO(32) superstring and the USp(32) model, in
which supersymmetry is broken on the D9-branes. We derive here the same results for type-IIA theory compactified on a circle in the presence of D8-branes. We also analyze the -symmetric action
for a brane charged with respect to the S-dual of the RR 10-form of type-IIB, and we find that the
tension of such an object has to scale like gS2 in the string frame. We give an argument to explain
why this result is in disagreement with the one obtained using Weyl rescaling of the brane action, and
we argue that this brane can only be consistently introduced if the other Z2 truncation of type-IIB
is performed. Moreover, we find that one can include a 10-form in type-IIA supersymmetry algebra,
and also in this case the corresponding -symmetric brane has a tension scaling like gS2 in the string
frame.
PACS: 11.25.-w; 11.30.Pb
1. Introduction
Type-II string theories in the non-perturbative regime contain in their spectrum BPS
D-branes, that are charged states with respect to RR fields [1], and are defined as hyE-mail address: f.riccioni@damtp.cam.ac.uk (F. Riccioni).
doi:10.1016/j.nuclphysb.2005.01.034
232
F. Riccioni / Nuclear Physics B 711 (2005) 231252
persurfaces on which open strings end [2]. In the low-energy effective action, these states
appear as 1/2-supersymmetric solitonic solutions carrying electric or magnetic charge with
respect to the RR fields of type-IIA and type-IIB supergravities. The effective action describing the massless modes of a D-brane is characterized by a DiracBornInfeld (DBI)
term and a WessZumino (WZ) term. In [3,4] it was shown that the effective action describing the massless open string states at string tree level is the DBI action in the approximation
in which one neglects derivatives of the field strength, while the coupling to the RR fields
is contained in the WZ term. The relative coefficient of the DBI and WZ terms is fixed,
since the tension and the RR-charge of the brane are related by the BPS condition.
The method for constructing actions for supersymmetric D-branes is known in the literature [59]. These actions are obtained embedding the world-volume of the brane in
superspace. The fermionic superspace coordinate becomes consequently a fermion on the
brane. Apparently, this seems to imply that the brane breaks all the supersymmetries, since
the fermion plays the role of the goldstino field. The solution of this apparent paradox is
the fact that the brane action possesses an additional local fermionic symmetry, known as
-symmetry [10,11], whose role is to decouple half of the fermions in the brane action.
After -gauge fixing half of the supersymmetries become linearly realized, while the other
half are still non-linearly realized, la VolkovAkulov [12]. Therefore, -symmetry is a
basic ingredient in the construction of brane actions, and it is the world-volume remnant of
the BPS-condition.
Spacetime-filling D-branes characterize the vacua of type-I models. Type-I string theory is obtained from type-IIB through an orientifold projection [13] that removes the states
that are odd under orientation reversal of the string. From a target space point of view, this
can be pictured in terms of orientifold planes, and the appearance of tadpoles corresponds
in this picture to non-vanishing tension and charge of the O-plane. Tadpole cancellation
typically requires the introduction of an open sector, and this corresponds to D-branes. In
ten dimensions, one can thus introduce an O -plane (with negative tension and negative
charge) and 32 D9-branes, with a resulting gauge group SO(32) [14]. The cancellation
of the overall tension and charge of the configuration corresponds to the cancellation of
dilaton and RR tadpoles [15,16]. The resulting theory is N = 1 supersymmetric, and the
massless spectrum contains the gravity multiplet from the closed sector and an SO(32)
YangMills multiplet from the open sector. There is actually a second possibility, corresponding to a change of sign of the tension and the charge of the orientifold plane, so that
RR tadpole cancellation requires the addition of 32 anti-D9 branes, with a resulting gauge
group USp(32) [17]. The overall tension of the configuration does not vanish, so that the
resulting theory has a dilaton tadpole. Nevertheless, the theory is anomaly-free, as a consequence of the vanishing of the RR tadpole [16,18]. The spectrum is not supersymmetric,
and more precisely the closed sector is not modified, still describing at the massless level
the N = 1 gravity multiplet, while the massless fermions in the open sector are not in
the adjoint but in the antisymmetric representation of USp(32), so that supersymmetry is
broken on the brane [19]. The gravitino couplings can then only be consistent if supersymmetry is non-linearly realized in the open sector. Since the antisymmetric representation of
symplectic groups is reducible, the massless spectrum contains a spinor that is an USp(32)
singlet, and this spinor is the goldstino of the non-linearly realized supersymmetry [20].
The presence of the NS tadpole is a manifestation of the fact that the theory has been ex-
233
panded around the wrong vacuum, and an analysis of this problem, addressed long time
ago in [21], has been recently performed in [22].
From the point of view of the low-energy effective action, the closed sector of type-I
strings is obtained performing a consistent Z2 truncation of the type-IIB theory, while the
open sector corresponds to the first order in the low-energy expansion of the D9-brane
action in a type-I background. It is then natural to ask what is the fate of -symmetry in
this background. The result is that there are two possibilities of performing this truncation
[23], and in a flat background, with all bulk fields put to zero, the D9-brane action reduces
in one case to the VolkovAkulov (VA) action [12], and in the other case to a constant. In
[24] these results were extended to a generic background, showing that also in the curved
case there are two possibilities of performing the truncation. In one case one gets a dilaton
tadpole and a RR tadpole plus goldstino couplings, while in the other case the goldstino
couplings vanish and one is left with a dilaton and a RR tadpole. This result is equivalent
to the string result: the two different truncations correspond to the two different choices of
the relative sign of tension and charge of orientifold plane and D9-branes. The first case
corresponds to the non-supersymmetric one, in which the orientifold plane and the D-brane
have both positive tension, and in the case of 32 coincident D9-branes it gives rise to the
low-energy action [20,25] of the USp(32) model. The second case, in which the goldstino
disappears, corresponds to an orientifold plane with negative tension, and in the case of 32
coincident D9-branes it gives rise to the low-energy action of the supersymmetric SO(32)
superstring. In other words, the supersymmetric truncation projects out the spinor that
would not be projected out fixing the -symmetry gauge. As a result, in general one expects
that only linearly realized supersymmetry survives. The non-supersymmetric truncation
does the opposite, namely, it projects out the spinor that would have been projected out
by -symmetry, so that no -symmetry is left in the truncated theory, and in the resulting
action supersymmetry is only non-linearly realized, i.e., completely broken.
If one wants to generalize these results to lower-dimensional cases, the first possibility is to consider the T-dual of this configuration, that is a type-I orientifold of type-IIA
compactified on a circle [2]. In this case the orientifold projection has fixed points on the
circle, corresponding to the positions of the O-planes. In [26] the low-energy action for
a D8-brane located at one of the fixed points was constructed in the type-I background,
without including the fermionic fields. In this paper we want to apply the techniques used
in [24] to this case, in order to obtain the low-energy brane + bulk action up to four Fermi
terms, for a generic 9-dimensional background. We will see that the results are in complete
agreement with T-duality, since also in this case one has two possible consistent truncations, leading in one case to a supersymmetric model, and in the other case to a model in
which supersymmetry is non-linearly realized on the brane.
S-duality is a symmetry of type-IIB string theory mapping weak coupling to strong
coupling [27]. On the other hand, type-I string theory is related in ten dimensions by a
strong-weak coupling S-duality to the heterotic SO(32) theory [28]. In this respect, it is
interesting to study the behavior of the O9D9 system of [23,24] under S-duality, and
whether the result can be related to the low-energy action of the heterotic theory. Type-IIB
supersymmetry algebra includes two 10-forms [23,29]. One of them is the RR 10-form
that couples to D9-branes, while the other couples to other spacetime-filling branes, called
NS9-branes in [29]. Type-IIB supergravity also admits an additional Z2 truncation, re-
234
moving all the RR fields. This truncation was conjectured in [29] to be the S-dual of the
orientifold projection, and consequently the introduction of 32 NS9-branes was conjectured to give origin to the SO(32) heterotic string after performing this projection [29,30].
Under S-duality, the DBI part of the D9-brane action acquires a dilaton factor e4 in the
string frame, and this led to conjecture that the tension of the NS9-branes is proportional to
gS4 [29,31]. We will argue in this paper that this is actually not the case. We will prove that
-symmetry requires a dilation factor e2 in front of the DBI term of an NS9-brane action, and thus a tension proportional to gS2 . This seems to be inconsistent with the analysis
of [31], since starting from a D9-brane action and performing an S-duality transformation
one should end up with a -symmetric action. The solution of this paradox is that in the
presence of NS and RR 10-forms S-duality is no longer a symmetry of the type-IIB algebra.
This does not mean that S-duality symmetry of type-IIB is actually broken, since introducing spacetime-filling branes is only consistent after performing a truncation. We will also
analyze the type-IIA case, since type-IIA supersymmetry algebra can be extended including an NS 10-form, and the resulting supersymmetric NS9-brane action is -symmetric if
the tension scales like e2 in the string frame.
The paper is organized as follows. In Section 2 we review some known results about
super-D-branes. In Section 3 we discuss the type-I truncation of type-IIA compactified on
a circle. We make use of the democratic formulation of the theory [26], in which both
the RR forms and their magnetic duals appear as independent fields in the supersymmetry
algebra, and duality relations between electric and magnetic field strengths are imposed as
constraints (the same formulation was introduced in [23] for the type-IIB case). We show
that the two possible truncations lead in one case to a supersymmetric model, and in the
other case to a model in which supersymmetry is spontaneously broken on the D8-brane.
In Section 4 we discuss S-duality of type-IIB in the presence of NS and RR 10-forms.
First of all, one realizes that these two fields, besides transforming as a doublet under Sduality, acquire a dilaton dependence e2 . Moreover, an additional constraint has to be
imposed for S-duality to be a symmetry. In other words, S-duality is broken in the presence
of spacetime-filling branes. Section 5 is devoted to the study of the supersymmetric NS9brane in both IIA and IIB. We show that -symmetry implies that the tension of the NS9brane in the string frame scales like e2 . In analogy to the D9 [24] and the D8 cases, one
can perform a truncation to show that half of the fermions decouple from the spectrum. In
this case the spectrum is projected by means of a heterotic truncation. Finally, Section 6
contains the conclusions.
2. Generalities about D-brane actions

In this section we review the basic ingredients for the construction of supersymmetric
D-branes, and in particular spacetime-filling D-branes. We will concentrate here on the IIB
case, while the straightforward generalization to the IIA case will be outlined in the next
section.
In order to construct supersymmetric actions for D-branes, one has to embed the
D-brane in IIB (or IIA) superspace. A basic ingredient is therefore the supersymmetry
algebra of type IIB in 10 dimensions. Since we want a formulation that is suitable for all
235
the D-branes of type-IIB, we write down the IIB algebra in the democratic formulation, in
which all the forms and their magnetic duals appear in the algebra. Following the notations
of [23], the supersymmetry transformations of the IIB bulk fields are
e a = a ,

1
1
1
= D H 3 + e
G(2n+1) 1 ...2n+1 Pn ,
8
16
(2n + 1)! 1 ...2n+1
5
n=0
(2)
B
= 2 3 [ ] ,
(10)
B1 ...10 = e2 3 (10[1 ...9 10 ]
C(2n)
1 ...2n
1 ...10 ),

1
2n ]
= (2n)e Pn [1 ...2n1 2n ]
2(2n)
(2n2)
+ n(2n 1)C[1 ...2n2 B2n1 2n ] ,

1
1 n2
H 3 + e
G(2n+1) Pn 1 ...2n+1 ,
12
4
(2n + 1)! 1 ...2n+1
5
=
n=0
1
= ,
(2.1)
2
where Pn is 1 for n odd and i2 for n even. We are neglecting terms cubic in the fermions
in the case of the transformations of the spinors. We have introduced the field strengths for
the RR fields and their duals, related by duality according to the relations
G(7) = G(3) ,
G(9) = G(1) ,
G(5) = G(5) .
(2.2)
An advantage of this formulation is that all the ChernSimons terms in the supergravity
Lagrangian are hidden in the definitions of the field strengths and their magnetic duals.
The matching between bosonic and fermionic degrees of freedom is of course restored
only once these duality relations are imposed.1 The field strengths are defined through the
relations
H = dB,
G(2n+1) = dC (2n) H C (2n2) ,
(2.3)
and the gauge transformations of the fields are

B = dN S ,
(2n1)
C (2n) = dRR
(10)
B (10) = dN S ,
(2n3)
RR
H,
(2.4)
so that the field strengths are gauge invariant. The dilaton dependence in the variations of
the forms shows that the algebra of Eq. (2.1) is expressed in the string frame. Moreover,
it is important to observe that two 10-forms are present in the algebra. We stress again
that, even though these forms do not have any dynamics since they do not have any field
1 This is a generalization of what is typically done for the self-dual 5-form field strength, when one writes a
Lagrangian for an ordinary 5-form, and imposes self-duality as a constraint on the equations of motion.
236
strength, they are associated to spacetime-filling branes, whose presence is consistent only
after one performs a suitable projection of the spectrum.
The general idea is to describe supersymmetric D-branes through the embedding of
a bosonic brane in superspace [5,7,8]. We thus introduce the world-volume fields as the
supercoordinates

Z M i = x i , I i
(2.5)
defining the position of the brane in superspace. Here i are the world-volume coordinates (i = 0, . . . , 9 for a 9-brane), while = 0, . . . , 9 is a spacetime vector index and
= 1, . . . , 32 a spinor index, and I = 1, 2. The Majorana spinors I are both left-handed.2
We denote with V i ( ) the Abelian world-volume vector. The bulk superfields are denoted
with

(2n)
, EM A , BMN , BM1 ...M10 , CM
(2.6)
, n = 0, . . . , 5,
1 ...M2n
and the brane action is
S = SDBI + SWZ =

d 10 e det(g + F) +
CeF ,
M10
(2.7)
M10
where
Fij = Fij + Bij ,
(2.8)
and one defines the pull-back of the bulk fields on the world-volume according to
gij = Ei a Ej b ab ,
Bij = i Z M j Z N BMN
(2.9)
and
C=
5

(1)n C (2n) ,
C (2n) =
n=0
1
(2n)
dZ M1 dZ M2n CM1 ...M2n .
(2n)!
(2.10)
In a flat space background [6,9] these expressions have a simpler form, since from the
(global) supersymmetry transformations of the supercoordinates,
1
,
2
one derives a supersymmetry invariant object
= ,
i = i x +
x =
(2.11)
1
i ,
2
(2.12)
that is the flat space analogous of i Z M EM a . Consequently, the pull-back of the metric
becomes
gij = i j = i x j x + (i x j ) + ,
2 In the IIA case the two chiral spinors I are substituted with a single non-chiral Majorana spinor.
(2.13)
237
where we neglect higher terms in the fermions. Analogously, the pull-back of the NS
2-form is
3 j ] + ,
Bij = [i x
(2.14)
while the pull-back of the RR forms is

Ci(2n)
= ne [i1 x 1 i2n1 x 2n1 Pn 1 ...2n1 i2n ] + .
1 ...i2n
(2.15)
The brane action (2.7) is then supersymmetric, provided that one chooses the supersymmetry transformation for the world-volume vector Vi to be
1
1
Vi = i 3 j Fj i
(2.16)
2
2
up to a gauge transformation.
The action (2.7) is invariant under world-volume general coordinate transformations,
and one can then choose a static (or Monge) gauge, in which the coordinates i are
identified with x i , i = 0, . . . , p, where p + 1 is the spacetime dimension of the brane.
A supersymmetry variation then induces a compensating general coordinate transformation, and the resulting variation for is
1 i
i .
(2.17)
2
The other xs in this gauge become world-volume scalars, whose supersymmetry transformations is

1
1
a = a i i a , a = p + 1, . . . , 10.
(2.18)
2
2
Focusing again on the flat space limit, one can recognize in Eq. (2.17) the Volkov
Akulov (VA) transformations [12]. We will concentrate in the following on space-filling
9-branes, so that the target spacetime -matrices can be identified with the world-volume
-matrices, and the spacetime index is the same as the world-volume index i. The commutator of two transformations (2.17) is a translation,

[1 , 2 ] = 2 1 ,
(2.19)
=
and thus Eq. (2.17) provides a realization of supersymmetry. The 1-form

1 a
,
2
transforms under supersymmetry as
e a = a +
(2.20)
ea = L ea ,
with L the Lie derivative with respect
(2.21)
to3
1
= ( ).
2
3 The parameter should not be confused with the world-volume coordinates.
(2.22)
238
The action of supersymmetry on e is thus a general coordinate transformation, with a parameter depending on , and therefore
L = det e
(2.23)
is clearly an invariant Lagrangian. Using the same technique, for a generic field A that
transforms under supersymmetry as
A = L A,
(2.24)
defining the induced metric as g = e

is determined by the substitution
me
m ,
a supersymmetric Lagrangian in flat space
L(, A) eL(g, A).
(2.25)
This is what happens in the brane action (2.7) in the Monge gauge in a flat space background, since the pull-back of the metric of Eq. (2.13) in the Monge gauge equals the VA
metric g . Moreover, the second term in the variation of V i is a general coordinate transformation with the same parameter plus an additional gauge transformation, while the
first term combines with the variation of the pull-back of the NS form of Eq. (2.14) in such
a way that F transforms covariantly. Finally, the pull-backs of the RR forms in Eq. (2.15)
are such that the WZ term in the brane action transforms as a total derivative.
It is then natural to generalize this VA construction to D9-branes in a generic background. One must construct from the bulk fields quantities whose supersymmetry variations are general coordinate transformations with the parameter plus additional gauge
transformations [20,25]. Supersymmetry guarantees that this way of constructing the D9brane action coincides with the superspace construction of [7,8].4 For instance, from the
supersymmetry variation of one defines
1 n3
1
1
(2n1) i1 ...i2n1
ij k 3 + e
G
Pn ,
= + Hij k
2
48
16
(2n 1)! i1 ...i2n1
n=1
(2.26)
whose supersymmetry transformation is a general coordinate transformation with the correct parameter i given in (2.22), up to higher order Fermi terms. With the same technique,
one can construct all the other hatted fields [20,25], so that the resulting D9-brane action
in a generic type-IIB background is

F,
S = SDBI + SWZ =
(2.27)
d 10 e det(g + F) +
Ce
6
M10
M10
where
Fij = Fij + B ij .
(2.28)
We come now to a brief discussion of the degrees of freedom that the action (2.7) propagates. If all the fermions were dynamical this would lead to a complete spontaneous
4 See [32] for a similar construction in the case of a generic p-brane.
239
supersymmetry breaking, since the s transform non-linearly under supersymmetry. It is

well known that this is actually not the case because of -symmetry gauge invariance,
whose fixing leads to a cancellation between the DBI and the WZ term that makes only
half of the fermions propagate. To leading order in the fermions, and in the Monge gauge,
the -symmetry transformation for and Vi reads

1
i
ij
1 1 2 F ij + ,
=
2
2
1
Vi = i 3 ,
(2.29)
2
with an SL(2, R) doublet of spinors, and neglecting higher order terms in and F in the
variation of . This gauge invariance can be used to put 1 2 = 0. After a supersymmetry
transformation, this gauge choice is maintained through a compensating -transformation
of parameter 1 2 = 1 2 , and this results in the linear supersymmetry transformations
1
(1 + 2 ) = F ij ij (1 2 ),
4
1
Vi = (1 2 )i (1 + 2 ),
(2.30)
2
and expanding the DBI action with this gauge choice one obtains that these are the correct linear supersymmetry transformations [9,33].5 In other words, symmetry is the
brane effective action equivalent of the statement that a brane solution of supergravity
is a BPS solution preserving half of the supersymmetries. In the case of spacetime-filling
branes, that do not correspond to any solution of supergravity, we assume in this paper that
-symmetry is the only requirement that these branes have to satisfy.
At the end of this section, we now want to review the results of [23] and [24]. One can
perform a type-I truncation of IIB supersymmetry algebra, imposing
C (2n2) = 0,
B = 0,
n = 1, 3, 5,
(10)
= 0,
(1 1 )f = 0,
(2.31)
where we have denoted with f the gravitino and the dilatino. The surviving bosonic fields
are thus the dilaton, the metric, the RR 2-form and its dual, and the RR 10-form, while the
two different signs in the projection of the fermions indicate that there are two possible
type-I truncations.
The truncation on the D9-brane action was performed in [23] in flat space, and generalized in [24] to an arbitrary background. We review here the results. The brane fields are
projected according to
Vi = 0,
(1 1 ) = 0.
(2.32)
The lower sign choice leads to no surviving -symmetry, since it projects out the spinor
components that would have been put to zero using -symmetry before the truncation,
while the upper sign choice leads to no leftover components of . This last choice, then,
5 See [34] for a similar analysis.
240
corresponding to the vanishing of all the terms containing the goldstino, results in a supersymmetric type-I spectrum. Actually, in the case of a single D9-brane, there are no
remaining world-volume degrees of freedom after the truncation, but the generalization to
a stuck of branes would result in a spectrum in which supersymmetry is linearly realized,
and the goldstino is projected out. The resulting action contains a dilaton tadpole and a RR
tadpole, that in the SO(32) string are both canceled against the orientifold plane contribution. The other choice, instead, corresponds to the curved generalization of the VA action.
The resulting spectrum breaks supersymmetry in the brane sector [19], or more precisely
N = 1 supersymmetry is non-linearly realized on the brane. The brane action again contains a dilaton tadpole and a RR tadpole, but in this case, a suitable orientifold projection
only cancels the brane RR charge, and a dilaton tadpole remains [17].
The type-IIB supersymmetry algebra in D = 10 also admits an alternative Z2 truncation, projecting out all the RR fields and acting as (1 3 )f = 0 on the fermions, and for
this reason called heterotic truncation [23]. We will show in Section 5 how this truncation
can be consistently implemented on spacetime-filling branes electrically charged with respect to B (10) , after a discussion about S-duality of type-IIB carried out in Section 4. First,
in the next section, we are going to describe the T-duals of these results, i.e., the type-I
truncation of IIA in the presence of D8-branes.
3. Type-I truncation of IIA
After reduction to D = 9, T-duality relates the system described in the previous section
to type-IIA theory compactified on a dual circle, and the corresponding truncation is in this
case the low-energy manifestation of the type-I orientifold projection. In this section we
want to discuss this truncation in the presence of D8-branes. Since D8-branes are charged
with respect to the RR 9-form, whose field strength is dual to a cosmological constant,
the massive Romans IIA supergravity [35] is the bulk low-energy theory describing this
system [28]. In has been shown in [26] that both massless and massive 10-dimensional IIA
supergravities can be described in terms of the same supersymmetry algebra, once the Romans cosmological constant is treated as a dynamical 0-form dual to the RR 10-form field
strength. Again, it is convenient to work in the democratic formulation [23,26], treating all
the RR-forms and their magnetic duals as independent, and imposing the duality relations
as constraints. The resulting supersymmetry algebra is
e a = a ,
1
1
1
G(2n) 1 ...2n (11 )n ,
= D + H 11 + e
8
16
(2n)! 1 ...2n
5
n=0
(2)
B
= 2 11 [ ] ,
C(2n1)
1 ...2n1
= (2n 1)e

(11 ) [1 ...2n2
n
1
2n1 ]
]
2(2n 1) 2n1
(2n3)
B2n2 2n1 ] ,
+ (n 1)(2n 1)C[
1 ...2n3
241
1
1 5 2n (2n)
H 11 + e
G
(11 )n 1 ...2n ,
12
8
(2n)! 1 ...2n
5
=
n=0
1
= .
2
The RR field strengths are defined as
(3.1)
(2)
G(2n) = dC (2n1) dB (2) C (2n3) + G(0) eB ,

where it is understood that one has to extract the 2n-form out of
by the duality relations
(3.2)
(2)
eB ,
and they are related
G(2n) = ()n G(102n) .
(3.3)
The field equations of type-IIA supergravity obtained in this formulation are supersymmetric only after these duality relations are imposed. This algebra can be naturally extended to
include a 10-form, whose supersymmetry transformation is
= e2 (10 [1 ...9 10 ] + 1 ...m10 ).
B(10)
1 ...10
(3.4)
Actually, there is also another consistent 10-form, whose transformation is

= e2 (10 [1 ...9 11 10 ] 1 ...m10 11 ),
B (10)
1 ...10
(3.5)
but we will show that it does not correspond to any spacetime-filling -symmetric brane.
Consequently, we will only consider (3.4) as a natural extension of the type-IIA supersymmetry algebra.
We now continue reviewing the results of [26] concerning the possible consistent Z2
truncations of the algebra of Eqs. (3.1) and (3.4). In 10 dimensions, only a single truncation
is available, projecting out all the RR fields, and acting on the fermions as
{ , , } 11 { , , }.
(3.6)
We will construct in Section 5 the resulting -symmetric spacetime-filling brane, for which
this truncation is consistent in the way described in the previous section. It will turn out
that this brane is electrically charged with respect to the 10-form B (10) .
If we compactify the theory on a circle S 1 , the resulting theory admits another Z2 truncation, acting on the compactified coordinate as
x 9 x 9 ,
(3.7)
thus acting as an orbifold projection, being the low-energy manifestation of the orientifold
projection generated by introducing two orientifold 8-planes at the fixed points. If we only
consider spacetime indices in the uncompactified directions, the projection acts on the
fields according to

(2)
(2)
g , , B
g , , B
,
C(2n1)
()n+1 C(2n1)
,
1 ...2n1
1 ...2n1
{ , , } 9 { , , }.
(3.8)
242
Any index in the 9-direction corresponds to an additional minus sign with respect to these
projection rules, and consequently the 10-form B (10) of Eq. (3.4) (having an index in the
9-direction) is consistently projected out.6 In order to make the analogy with the IIB case
in 10 dimensions manifest, we define the 10-dimensional -matrices as
= 2 ,
9 = 1 1 ,
11 = 1 3
(3.9)
in terms of the 9-dimensional -matrices. Consequently, denoting the 10-dimensional IIA

spinors as doublets of 9-dimensional spinors, the truncation acts as
(1 1 ) = 0,
(1 1 ) = 0.
(3.10)
We now want to consider the introduction of D8-branes, and we will only take into
account the case in which a single D8-brane is located at one of the two fixed points of the
orientifold projection. Consistency requires that the truncation acts on the world-volume
fields as7
V i = 0,
(1 1 ) = 0.
(3.11)
The brane action contains, in a massive background, an additional ChernSimons term

[36,37] that we will not take into account because it vanishes after the truncation. The
relevant terms in the brane action are thus

det g + C (9) .
e
(3.12)
M9
M9
The supersymmetrization of this action is obtained in the Monge gauge following the same
arguments of the previous section. Taking into account only the terms that are relevant after
the truncation, and neglecting higher order Fermi fields, we thus define the hatted fields
1
= + + ,
2
g = g + 2 ( ) + ( D) + ,
1
C (9)1 ...9 = C(9)1 ...9 9e [1 ...8 11 9 ] e 1 ...9 11
2
9
e [1 ...8 11 D9 ] + ,
(3.13)
2
whose supersymmetry transformation has the form of a -dependent general coordinate
transformation (plus an additional gauge transformation in the case of the 9-form). Expressing then the brane action in terms of these hatted fields, it turns out that if one chooses
the upper sign in the projection of the fermions, all the terms containing the goldstino
disappear in the action, while the lower sign choice leads to an action of the VA type for .
We interpret this result in the same way as we did for the IIB case in 10 dimensions. The
6 In the case of B (10) defined in Eq. (3.5), consistency would require that this form survive the projection.
7 The other world-volume field, the scalar x 9 , is of course projected out because of Eq. (3.7).
243
upper sign choice corresponds to a supersymmetric spectrum. Again, just as in the case of
a single D9-brane, for a single D8-brane there are no remaining world-volume degrees of
freedom after the truncation, but the generalization to a stuck of branes would result in a
spectrum in which supersymmetry is linearly realized, and the goldstino is projected out.
The resulting action contains a dilaton tadpole and a RR tadpole, that in consistent supersymmetric orientifold models are both canceled against the orientifold plane contribution.
The other choice, instead, corresponds to the case in which the brane and the orientifold
plane have both positive tension. Consequently, N = 1 supersymmetry is non-linearly realized on the brane, and a suitable orientifold projection only cancels the brane RR charge,
while a dilaton tadpole remains. The fact that this result is in agreement with the IIB result
of Refs. [23,24] is a manifestation of T-duality [36,38].
4. S-duality of type-IIB
Type-IIB superstring theory is conjectured to be invariant under SL(2, Z) transformations [27], a discrete subgroup of the isometry group SL(2, R) of type-IIB supergravity
[39]. This group acts on the complex scalar
= C0 + ie
(4.1)
as
where
a
c
a + b
,
c + d
b
d
(4.2)

SL(2, R),
while the 2-forms B (2) and C (2) transform as a doublet. The matrix

0 1
S=
1 0
(4.3)
(4.4)
generates the S-duality transformation 1/ , that for a vanishing axion background

corresponds to , and in type-IIB string theory this results in mapping weak coupling to strong coupling. SL(2, Z) symmetry thus implies a strong-weak coupling selfduality of type-IIB string theory. Since S maps B (2) to C (2) and vice versa, the duality
interchanges the fundamental string and the NS5-brane with the D1-string and the D5brane. We want to study here how spacetime-filling branes transform under S-duality, and
since the 10-forms B (10) and C (10) cannot appear consistently in the low-energy effective action, the only way of deducing their behavior under an S-transformation is to make
use of the supersymmetry algebra. In the remaining of this section, we thus study how a
transformation S acts on the supersymmetry algebra (2.1).
In the string frame, S-duality acts on the metric as
g e g .
(4.5)
244
Because of the explicit dilaton dependence of this transformation, it is easier to consider a

configuration with vanishing axion background. This is what we will do in the following,
and it is understood that our results do not depend on this assumption. For completeness, we
write again the IIB supersymmetry transformations in the string frame in this background:
e a = a ,
1
1
1 2 3
= D H 3 + e G(3)
1 ,
1 2 3
8
48
(2)
B
= 2 3 [ ] ,
B(10)
= e2 3 (10[1 ...9 10 ] 1 ...10 ),
1 ...10

1
(2)
C
= 2e 1 [m ] ] ,
4

1
=
10e
,
C(10)
1 [1 ...9
10 ]
]
1 ...10
20 10
1
1
= H 3 e G(3)
,
1
12
12
1
= ,
(4.6)
2
again neglecting higher order Fermi terms in the transformations of the fermions.
Our strategy will be to derive the transformations of the fields under S-duality requiring
that the supersymmetry algebra is preserved. We already know that
,
e a e/2 e a .
(4.7)
We now obtain the transformations of , and e requiring that they are consistent with
Eq. (4.7), i.e., imposing that the supersymmetry variation of the S-transformed fields is
still Eq. (4.6), up to other local symmetry transformations of the type-IIB theory. We know
from the supersymmetry transformation of that and must acquire the same dilaton
dependence. Moreover, we expect that all the fermions undergo an overall SL(2, R) rotation determined by a 2 2 unitary matrix . Finally, the transformation of the gravitino
can contain a term proportional to . Hence, imposing that the transformed vielbein has
the correct supersymmetry variation, one gets
1
(4.8)
e/4 e/4 .
4
It turns out that the supersymmetry transformation of the vielbein is mapped to itself plus
an additional local Lorentz transformation of parameter
e/4 ,

1
ab = e/2 ab .
(4.9)
4
We neglect this term when we study the S-transformations of the supersymmetry variations
of the fermions, since they would lead to cubic Fermi terms. The S-duality transformation
of is straightforwardly obtained imposing that the transformed dilaton varies according
245
to (4.6) under supersymmetry, and the result is

e/4 .
(4.10)
Proceeding this way, one realizes that the two 2-forms

doublet, transforming as
B (2) C (2) ,
C (2) B (2)
B (2)
and
C (2)
form an SL(2, R)
(4.11)
if
1 3 = 1 ,
whose solution is8
= ei2 /4 =
1 1 = 3 ,

1/2 1/ 2
.
1/ 2 1/ 2
(4.12)
(4.13)
Implementing these transformations on the supersymmetry variation of C (4) , one then obtains
(2)
(2)
C(4)1 ...4 C(4)1 ...4 6B[1 2 C3 4 ] ,
(4.14)
leaving invariant the field strength

G(5) = dC (4) H (3) C (2) .
(4.15)
The correctness of the transformations (4.8) and (4.10) is finally proven by showing that
the supersymmetry variation of the transformed Fermi fields is consistent with Eq. (4.6).
Following the same arguments, one can now determine the S-duality transformations
of the two 10-forms B (10) and C (10) . One expects these fields to transform as a doublet,
but the surprising result is that the first requirement one has to make to impose S-duality
is that the transformation of this doublet must have a non-trivial dilaton dependence. More
precisely, the only possibly consistent transformation is
B (10) e2 C (10) ,
C (10) e2 B (10) .
(4.16)
This still does not guarantee that S-duality is preserved, and in fact the additional variation
of in (4.16) is canceled only once one imposes the additional constraints9
C(10)
( ) = e ( 1 1 ...10 ),
1 ...10
B(10)
( ) = e2 ( 3 1 ...10 ).
1 ...10
(4.17)
After performing the type-I truncation of Eq. (2.31), in which B (10) is projected out, the
first constraint becomes

1 1 ...10
C1 ...10 = e det g.
(4.18)

10!
8 The inverse choice for corresponds to a sign change in the transformations of C (2) and B (2) .
9 As a consistency check, one can show that these constraints are related by an S-duality transformation.
246
A similar result holds for the second constraint, after performing the heterotic truncation,
in which all the RR fields are projected out and the spinors are projected according to
(1 3 )f = 0.
(4.19)
As we will see in the next section, a possible interpretation of this result is that S-duality
is actually broken in the presence of spacetime-filling branes. The constraint of Eq. (4.17)
can be justified only in the truncated theory, and this is in agreement with the fact that only
in the truncated theory the presence of spacetime-filling branes can be consistent.
We will see in the next section that if one tries to construct the S-dual of a supersymmetric D9-brane using Eqs. (4.7), (4.8), (4.10), (4.11) and (4.16), the action one gets is
no longer -symmetric, and the constraints of Eq. (4.17) have to be imposed to restore
-symmetry. This means that the breakdown of -symmetry is consistent with the breakdown of S-duality. The brane action obtained in this way has a DBI term proportional to
e4 and a WZ term proportional to e2 . As we are going to prove, it turns out that, in
the untruncated theory, a -symmetric action for a spacetime-filling brane charged with
respect to B (10) has an e2 dilaton dependence in the DBI term, and no dilaton factor in
the WZ term.
5. Spacetime-filling branes and S-duality

In this section we want to describe the -symmetric spacetime-filling branes that are
charged with respect to the NS 10-forms of IIB and IIA supergravities. We start from
the type-IIB case, performing an S-duality transformation on the D9-brane action. In the
following of this section, we will always have in mind to perform a Z2 truncation that
leaves the NS 10-form invariant. In the case of IIB, this transformation projects out all
the RR fields, leaving the NS fields invariant. This projection can actually be worked out
using the results of the previous section, performing an S-duality transformation of the
truncations of Eq. (2.31). In the fermionic sector, performing the transformations (4.8) and
(4.10) and using Eq. (4.12), the projection becomes
(1 3 )f = 0.
(5.1)
The same projection applies to the spinor in the brane sector, since transforms under
S-duality like , while the world-volume vector V i is projected out, again in agreement
with Eqs. (4.8) and (4.12). We will assume that the 10-forms B (10) and C (10) transform
according to Eq. (4.16), keeping in mind that this is consistent only if the constraints (4.17)
are satisfied. We will see that the action we end up with is consistently -symmetric only
if these constraints are satisfied.
After the projection, the S-dual of the action (2.7) becomes

10
4
S=
(5.2)
d e
det g +
e2 B (10) .
M10
M10
247
The supersymmetrization of this action is then worked out in the same way as the D9 and
D8 cases. One first constructs the hatted fields
1
= + + ,
2
g = g + 2 ( ) + ( D) + ,
B (10)
= B(10)
+ 10e2 [1 ...9 3 10 ] e2 1 ...10 3
1 ...10
1 ...10
+ 5e2 [1 ...9 3 D10 ] + ,
and then writes the supersymmetric action

S=
d 10 e4 det g +
e2 B (10) .
M10
(5.3)
(5.4)
M10
If this procedure preserved -symmetry, it would be expected that one of the two truncations (the one with the upper sign choice in Eq. (5.1), as one would get using Eq. (4.12))
leaded to a brane action in which all the goldstino terms disappear. This is actually not the
case, since for the upper sign choice a term proportional to
B(10)
( ) + e2 ( 1 ...10 )
1 ...10
(5.5)
survives. This term vanishes if the second constraint of Eq. (4.17) is imposed. This is
not surprising, since only if this constraint is valid the S-duality transformations can be
performed. Thus, the picture that emerges is that the breakdown of S-duality is in agreement with the breakdown of -symmetry, and the constraint of Eq. (4.17) provides a
restoration of both. On the other hand, the constraint (4.17) leads to a vanishing action
for the NS9-brane, and this could simply mean that such an object does not exist. We
will comment about this in the conclusions. The lower sign choice in (5.1), again analogously to the D-brane case, corresponds to a VA-type action for , after Eq. (4.17) is
imposed.
Let us consider now the action

S=
(5.6)
d 10 e2 det g +
B (10) .
M10
M10
In this case, using Eqs. (5.3), one obtains that the upper sign choice in Eq. (5.1) leads to
an action with no goldstino, while the lower sign choice leads to a VA action, and in both
cases no constraint is required. This means that the untruncated action is -symmetric,
and thus we argue that this is the correct action for an NS9-brane. More precisely, the
complete action would result from rescaling the S-dual of the Lagrangian of Eq. (2.7), and
in order to compute this, one should know how C (8) and C (6) transform under S-duality.
This analysis is currently under investigation, and we expect that it would shed some light
on the problem of studying the S-dual of a D7-brane as well. Anyway, we do not expect
the results of this section to be altered by the inclusion of additional terms, since we do not
see how the constraint of Eq. (4.17) can be removed modifying B (10) by the inclusion of
other bulk fields.
248
One could also discuss the S-dual of this picture, starting from the action of Eq. (5.6),
and then performing an S-duality transformation. The result is that one ends up with the
action

S=
(5.7)
d 10 e3 det g +
e2 C (10) .
M10
M10
Again, -symmetry corresponds to the existence of a Z2 truncation removing the goldstino

completely, and one can show that this happens only after imposing the constraint for C (10)
in Eq. (4.17). The picture is thus completely symmetric, since from this low-energy point
of view assuming that the D9-brane action is (2.7) instead of (5.7) is S-dual to assuming
that the action for an NS9-brane is (5.6) instead of (5.2).
At the end of this section, we want to determine the supersymmetric action for an
NS9-brane in type-IIA, where again -symmetry corresponds to the vanishing of all the
goldstino terms in the suitably Z2 -truncated action. The 10-dimensional Z2 -truncation
projects out all the RR fields, acting on the fermions as
= 11 ,
= 11 .
(5.8)
From the supersymmetry transformation of Eq. (3.4) we obtain

B (10)
= B(10)
10e2 [1 ...9 10 ] + e2 1 ...10
1 ...10
1 ...10
5e2 [1 ...9 D10 ] + .
(5.9)
Vi
The truncation acts on the world-volume fields as usual: the vector

is projected out,
while transforms in the same way as . The final result is that the truncated action

d 10 e2 det g +
S=
(5.10)
B (10)
M10
M10
is supersymmetric choosing the upper sign in Eq. (5.8), while supersymmetry is spontaneously broken if one chooses the lower sign. It can be shown that, after an S 1 reduction,
T-duality relates this action with the one of Eq. (5.6).
6. Conclusions
The starting point of this paper was a continuation of [24], where the results of [23]
were generalized to a curved background, showing that the possible type-I truncations of
type-IIB are in correspondence with the possible consistent type-I strings in D = 10. We
showed here that the same results apply to the D = 9 truncations of type-IIA, in accordance
with T-duality. We then proceeded constructing the -symmetric spacetime-filling branes
that are charged with respect to the NS 10-forms of type-IIB and type-IIA.
In [29] it was argued that S-duality of type-IIB implies the existence of NS9-branes, that
together with the D9-branes form an SL(2, Z) doublet. From the standard Weyl-rescaling
argument, it turns out that the tension of these branes appears to scale like 1/gS4 in the
string frame. Here we have argued that the actual tension of these branes scales like 1/gS2 ,
249
like the other solitonic NS objects, namely NS5-branes. The solution of the paradox is
that the doublet of NS and RR 10-form potentials does not transform covariantly under
SL(2, Z). It is therefore not possible to derive the action for an NS9-brane performing a
Weyl rescaling. In [29,30] it was also conjectured that S-duality implies the existence of a
dual of the orientifold projection, and the SO(32) heterotic theory should result from this
projection, after the introduction of 32 NS9-branes. This projection would naturally act
like 3 on the fermion doublets, since in the heterotic theory the fermions come only from
the left sector. We do not expect this truncation to be an ordinary Z2 orbifold, since it is
well know that a Z2 orbifold of type-IIB gives rise to type-IIA. In any case, if there is a
way of deriving the heterotic SO(32) theory from type-IIB, we expect that the twisted
sector of the projection would correspond to inserting -symmetric branes, that would
therefore have the structure of Eq. (5.6). In any case, should a brane interpretation of the
heterotic theory be possible, a natural question would arise, namely what is the heterotic
string equivalent of brane supersymmetry breaking.10
Similar arguments hold for the IIA case. Since the type-IIA superstring and the E8 E8
heterotic theory have both an M-theory origin [40,41], it has also been conjectured [29] that
the E8 E8 heterotic theory can arise in ten dimensions from a projection of type-IIA,
that would result in the low-energy action in a Z2 truncation removing the RR fields. We
emphasize again that if this projection exists, it cannot act as a Z2 orbifold of type-IIA,
since such an orbifold gives rise to type-IIB. The twisted sector of the heterotic theory
would result in this case form the insertion of NS9-branes. Starting from type-IIB and using
Weyl-rescaling arguments, it has been argued that T-duality would imply that the tension of
this branes, apart from having an e4 dilaton dependance, is proportional to R 3 , where R
is the radius of the isometry direction [31], and this would mean that they are not defined
in 10 uncompactified dimensions. The NS9-brane, as well as the D8-brane, would then
result from a 9-brane in M-theory whose effective action and target space solution [42]
can be written only if the 11-dimensional supergravity has an isometry, and thus cannot be
covariant in 11 dimensions. Again, the -symmetric spacetime-filling brane we obtained
in this paper has instead an e2 dilaton dependance, and it is related by T-duality to the
type-IIB NS-brane of Eq. (5.6). This different scaling with respect to the one of [31] implies
that this NS9-brane and the D8-brane cannot have a common M-theory origin. Since the
field-strength of a 10-form in 11 dimensions would be dual to a cosmological constant,
this result is basically rephrasing the fact that no cosmological constant can be included
in 11-dimensional supergravity [43], and Romans IIA supergravity cannot be obtained
by dimensional reduction from 11 dimensions. After compactification on a 2-torus Mtheory is related to type-IIB by T-duality, and thus this picture is the T-dual analogous
of the type-IIB picture, where S-duality is broken by the presence of spacetime-filling
branes.
Finally, it would be interesting to see if the S-duality rules of this paper can be used
to understand the strong coupling behavior of the D7-branes. In [44] it was shown that
D7-branes of type-IIB belong to a triplet of 7-branes. One could then determine a supersymmetric effective action for these branes requiring -symmetry, and relate them to the
10 I am grateful to E. Dudas for discussions about this point.
250
half-BPS 7-brane solutions of type-IIB supergravity [4547]. This analysis is currently

under investigation.
Acknowledgements
I am grateful to E. Bergshoeff, M. Bianchi, E. Dudas, M. Green and G. Pradisi for
discussions. This work is supported by a European Commission Marie Curie Postdoctoral
Fellowship, Contract MEIF-CT-2003-500308.
References
[1] J. Polchinski, Dirichlet-branes and RamondRamond charges, Phys. Rev. Lett. 75 (1995) 4724, hepth/9510017.
[2] J. Dai, R.G. Leigh, J. Polchinski, New connections between string theories, Mod. Phys. Lett. A 4 (1989)
2073.
[3] E.S. Fradkin, A.A. Tseytlin, Nonlinear electrodynamics from quantized strings, Phys. Lett. B 163 (1985)
123.
[4] A. Abouelsaood, C.G. Callan, C.R. Nappi, S.A. Yost, Open strings in background gauge fields, Nucl. Phys.
B 280 (1987) 599.
[5] M. Cederwall, A. von Gussich, B.E. Nilsson, A. Westerberg, The Dirichlet super-three-brane in tendimensional type IIB supergravity, Nucl. Phys. B 490 (1997) 163, hep-th/9610148.
[6] M. Aganagic, C. Popescu, J.H. Schwarz, D-brane actions with local kappa symmetry, Phys. Lett. B 393
(1997) 311, hep-th/9610249.
[7] M. Cederwall, A. von Gussich, B.E. Nilsson, P. Sundell, A. Westerberg, The Dirichlet super-p-branes in
ten-dimensional type IIA and IIB supergravity, Nucl. Phys. B 490 (1997) 179, hep-th/9611159.
[8] E. Bergshoeff, P.K. Townsend, Super-D-branes, Nucl. Phys. B 49 (1997) 145, hep-th/9611173.
[9] M. Aganagic, C. Popescu, J.H. Schwarz, Gauge-invariant and gauge-fixed D-brane actions, Nucl. Phys.
B 495 (1997) 99, hep-th/9612080.
[10] W. Siegel, Hidden local supersymmetry in the supersymmetric particle action, Phys. Lett. B 128 (1983) 397.
[11] M.B. Green, J.H. Schwarz, Covariant description of superstrings, Phys. Lett. B 136 (1984) 367.
[12] D.V. Volkov, V.P. Akulov, Is the neutrino a goldstone particle?, Phys. Lett. B 46 (1973) 109;
U. Lindstrom, M. Rocek, Constrained local superfields, Phys. Rev. D 19 (1979) 2300;
S. Samuel, J. Wess, A superfield formulation of the nonlinear realization of supersymmetry and its coupling
to supergravity, Nucl. Phys. B 221 (1983) 153;
S. Samuel, J. Wess, Realistic model building with the AkulovVolkov superfield and supergravity, Nucl.
Phys. B 226 (1983) 289;
S. Samuel, J. Wess, Secret supersymmetry, Nucl. Phys. B 233 (1984) 488;
J. Bagger, A. Galperin, Matter couplings in partially broken extended supersymmetry, Phys. Lett. B 336
(1994) 25, hep-th/9406217.
[13] A. Sagnotti, in: G. Mack, et al. (Eds.), Non-Perturbative Quantum Field Theory, in: Cargese 87, Pergamon,
Elmsford, 1988, p. 521;
A. Sagnotti, Open strings and their symmetry groups, hep-th/0208020;
G. Pradisi, A. Sagnotti, Open string orbifolds, Phys. Lett. B 216 (1989) 59;
M. Bianchi, A. Sagnotti, On the systematics of open string theories, Phys. Lett. B 247 (1990) 517;
M. Bianchi, A. Sagnotti, Twist symmetry and open string Wilson lines, Nucl. Phys. B 361 (1991) 519;
M. Bianchi, G. Pradisi, A. Sagnotti, Toroidal compactification and symmetry breaking in open string theories, Nucl. Phys. B 376 (1992) 365.
[14] M.B. Green, J.H. Schwarz, Anomaly cancellation in supersymmetric D = 10 gauge theory and superstring
theory, Phys. Lett. B 149 (1984) 117;
M.B. Green, J.H. Schwarz, Infinity cancellations in SO(32) superstring theory, Phys. Lett. B 151 (1985) 21.
251
[15] N. Ohta, Cancellation of dilaton tadpoles and two loop finiteness in SO(32) type I superstring, Phys. Rev.
Lett. 59 (1987) 176.
[16] J. Polchinski, Y. Cai, Consistency of open superstring theories, Nucl. Phys. B 296 (1988) 91.
system and the USp(32) string theory, Prog. Theor.
[17] S. Sugimoto, Anomaly cancellations in type I D9D9
Phys. 102 (1999) 685, hep-th/9905159.
[18] M. Bianchi, J.F. Morales, Anomalies and tadpoles, JHEP 0003 (2000) 030, hep-th/0002149] .
[19] I. Antoniadis, E. Dudas, A. Sagnotti, Brane supersymmetry breaking, Phys. Lett. B 464 (1999) 38, hepth/9908023.
[20] E. Dudas, J. Mourad, Consistent gravitino couplings in non-supersymmetric strings, Phys. Lett. B 514 (2001)
173, hep-th/0012071.
[21] W. Fischler, L. Susskind, Dilaton tadpoles, string condensates and scale invariance, Phys. Lett. B 171 (1986)
383;
W. Fischler, L. Susskind, Dilaton tadpoles, string condensates and scale invariance. 2, Phys. Lett. B 173
(1986) 262.
[22] E. Dudas, G. Pradisi, M. Nicolosi, A. Sagnotti, On tadpoles and vacuum redefinitions in string theory, hepth/0410101.
[23] E. Bergshoeff, M. de Roo, B. Janssen, T. Ortin, The super-D9-brane and its truncations, Nucl. Phys. B 550
(1999) 289, hep-th/9901055.
[24] F. Riccioni, Truncations of the D9-brane action and type-I strings, Phys. Lett. B 560 (2003) 223, hepth/0301021.
[25] G. Pradisi, F. Riccioni, Geometric couplings and brane supersymmetry breaking, Nucl. Phys. B 615 (2001)
33, hep-th/0107090.
[26] E. Bergshoeff, R. Kallosh, T. Ortin, D. Roest, A. Van Proeyen, New formulations of D = 10 supersymmetry
and D8O8 domain walls, Class. Quantum Grav. 18 (2001) 3359, hep-th/0103233.
[27] C.M. Hull, P.K. Townsend, Unity of superstring dualities, Nucl. Phys. B 438 (1995) 109, hep-th/9410167.
[28] J. Polchinski, E. Witten, Evidence for heterotictype I string duality, Nucl. Phys. B 460 (1996) 525, hepth/9510169.
[29] C.M. Hull, Gravitational duality, branes and charges, Nucl. Phys. B 509 (1998) 216, hep-th/9705162.
[30] C.M. Hull, The non-perturbative SO(32) heterotic string, Phys. Lett. B 462 (1999) 271, hep-th/9812210.
[31] E. Bergshoeff, E. Eyras, R. Halbersma, J.P. van der Schaar, C.M. Hull, Y. Lozano, Spacetime-filling branes
and strings with sixteen supercharges, Nucl. Phys. B 564 (2000) 29, hep-th/9812224.
[32] I.A. Bandos, J.A. de Azcarraga, J.M. Izquierdo, J. Lukierski, An action for supergravity interacting with
super-p-brane sources, Phys. Rev. D 65 (2002) 021901, hep-th/0104209;
I.A. Bandos, J.A. de Azcarraga, J.M. Izquierdo, J. Lukierski, D = 4 supergravity dynamically coupled to a
massless superparticle in a superfield Lagrangian approach, hep-th/0207139;
I.A. Bandos, J.A. de Azcarraga, J.M. Izquierdo, J. Lukierski, On dynamical supergravity interacting with
super-p-brane sources, hep-th/0211065.
[33] E.A. Bergshoeff, M. de Roo, A. Sevrin, Non-Abelian BornInfeld and kappa-symmetry, J. Math. Phys. 42
(2001) 2872, hep-th/0011018.
[34] J. Gomis, K. Kamimura, P.K. Townsend, Non-relativistic superbranes, hep-th/0409219.
[35] L.J. Romans, Massive N = 2a supergravity in ten dimensions, Phys. Lett. B 169 (1986) 374.
[36] E. Bergshoeff, M. De Roo, D-branes and T-duality, Phys. Lett. B 380 (1996) 265, hep-th/9603123.
[37] M.B. Green, C.M. Hull, P.K. Townsend, D-brane WessZumino actions, T-duality and the cosmological
constant, Phys. Lett. B 382 (1996) 65, hep-th/9604119.
[38] E. Bergshoeff, C.M. Hull, T. Ortin, Duality in the type II superstring effective action, Nucl. Phys. B 451
(1995) 547, hep-th/9504081.
[39] J.H. Schwarz, Covariant field equations of chiral N = 2 D = 10 supergravity, Nucl. Phys. B 226 (1983) 269.
[40] E. Witten, String theory dynamics in various dimensions, Nucl. Phys. B 443 (1995) 85, hep-th/9503124.
[41] P. Horava, E. Witten, Heterotic and type I string dynamics from eleven dimensions, Nucl. Phys. B 460 (1996)
506, hep-th/9510209.
[42] E. Bergshoeff, J.P. van der Schaar, On M9-branes, Class. Quantum Grav. 16 (1999) 23, hep-th/9806069.
[43] K. Bautier, S. Deser, M. Henneaux, D. Seminara, No cosmological D = 11 supergravity, Phys. Lett. B 406
(1997) 49, hep-th/9704131.
252
[44] P. Meessen, T. Ortin, An Sl(2, Z) multiplet of nine-dimensional type II supergravity theories, Nucl. Phys.
B 541 (1999) 195, hep-th/9806120.
[45] G.W. Gibbons, M.B. Green, M.J. Perry, Instantons and seven-branes in type IIB superstring theory, Phys.
Lett. B 370 (1996) 37, hep-th/9511080.
[46] M.B. Einhorn, L.A. Pando Zayas, On seven-brane and instanton solutions of type IIB, Nucl. Phys. B 582
(2000) 216, hep-th/0003072.
[47] E. Bergshoeff, U. Gran, D. Roest, Type IIB seven-brane solutions from nine-dimensional domain walls,
Class. Quantum Grav. 19 (2002) 4207, hep-th/0203202.
Universality of nonperturbative effect

in type 0 string theory
Hikaru Kawai a,b , Tsunehide Kuroki b , Yoshinori Matsuo a
a Department of Physics, Kyoto University, Kyoto 606-8502, Japan
b Theoretical Physics Laboratory, RIKEN, 2-1 Wako, Saitama 351-0198, Japan

Abstract
We derive the nonperturbative effect in type 0B string theory, which is defined by taking the
double scaling limit of a one-matrix model with a two-cut eigenvalue distribution. However, the
string equation thus derived cannot determine the nonperturbative effect completely, at least without
specifying unknown boundary conditions. The nonperturbative contribution to the free energy comes
from instantons in such models. We determine by direct computation in the matrix model an overall
factor of the instanton contribution, which cannot be determined by the string equation itself. We
prove that it is universal in the sense that it is independent of the detailed structure of potentials in
the matrix model. It turns out to be a purely imaginary number and therefore can be interpreted as a
quantity related to instability of the D-brane in type 0 string theory. We also comment on a relation
between our result and boundary conditions for the string equation.
1. Introduction
The nonperturbative effect in string theory can be studied using matrix models. In particular, the noncritical string theory, which is a simplified model of string theory, is exactly
solvable via matrix models [1]. The string equation, which can be derived from them,
contains the nonperturbative effect of the noncritical string theory [2]. On the other hand,
E-mail address: ymatsuo@gauge.scphys.kyoto-u.ac.jp (Y. Matsuo).
doi:10.1016/j.nuclphysb.2005.01.002
254
H. Kawai et al. / Nuclear Physics B 711 (2005) 253274
study of the Liouville theory [3] enables us to find the effect of the D-brane, which is the
nonperturbative effect of string theory and can be identified with the effect that appears in
the string equation. In [4], we have shown that the string equation does not describe the
nonperturbative effect completely, at least in c = 0 noncritical string theory. To obtain the
whole nonperturbative effect, it is necessary to study the matrix model directly.
Recently, a matrix model that corresponds to the noncritical string with worldsheet supersymmetry has been proposed [5,6]. For the c = 0 noncritical string, which is described
as two-dimensional pure supergravity on the worldsheet [7], we consider the double scaling limit around the GrossWitten transition [8]. This critical point can be found in the
one-matrix model with a two-cut eigenvalue distribution. In string theoretical interpretation, the one-matrix model with two cuts corresponds to the NSR string theory of type 0B.
This matrix model can be solved with the string equation. However, as in the case of c = 0
string theory, it does not contain the nonperturbative effect completely.
In this paper we study the nonperturbative effect of type 0B string theory by analyzing
the matrix model with a two-cut eigenvalue distribution. We compute the effect of instantons directly in the matrix model, which corresponds to the D-brane in the string theory.
The result is summarized as follows.
From the string equation, the nonperturbative effect in the free energy is obtained as

2 3/2
C
,
= 3/4 exp t
(1.1)
3
t
where C cannot be determined from the string equation by itself without specifying unknown boundary conditions. From the direct computation using the matrix model, we can
determine the constant C as
i
C= .
(1.2)
4
Moreover, it is shown that this value is universal, namely, it does not depend on the detailed structure in the potential of the matrix model. Because it is purely imaginary, this
nonperturbative effect is related to the instability of the D-brane.
The paper is organized as follows. In Section 2, we identify the instantons in the matrix
model and the contribution from instantons to the free energy. In Section 3, we compute
the effect of the instantons using the method of orthogonal polynomials. In Section 4, we
take the double scaling limit and consider the universal behavior of the effect of instantons.
In Section 5, we present the conclusions. Appendices AC show the details of calculations.
2. Instanton in one-matrix model

In this section, we consider a one-matrix model. We discuss how an instanton contributes to the partition function and the free energy. The one-matrix model with a one-cut
eigenvalue distribution corresponds to c = 0 noncritical string theory, while that with a
two-cut distribution corresponds to c = 0 type 0B string theory [5]. In both cases, an instanton can be interpreted as the ZZ-brane [9]. In the one-cut case, the instanton can be
described as a configuration in which all eigenvalues are at the minimum of the effective potential except that a single eigenvalue is at its maximum. This description can be
255
extended to the case of two-cut distribution. In this case, the effective potential has two
separated minima. Half of the eigenvalues are in one of these minima and the other half
are in the other minimum except that a single eigenvalue is at the maximum.
In the one-matrix model, the partition function is given by

Z = d eN tr V () .
(2.1)
Here, is an N N Hermitian matrix. Hereafter we consider the case where the potential
V (x) is invariant under x x and thus the eigenvalue distribution of has this Z2
symmetry.
Diagonalizing the matrix , the partition function can be expressed as

Z=
(2.2)
di 2 () eN i V (i ) .
i
Here, () = i<j (i j ) is the Vandermonde determinant. We concentrate on the N th

eigenvalue N , and represent it as x. The other N 1 eigenvalues can be regarded as those
of a (N 1) (N 1) matrix. The partition function of an N N matrix model can be
expressed using an (N 1) (N 1) matrix model as

N
1
N
1
N1

Z = dx
(2.3a)
di 2N 1 ()
(x i )2 eN i=1 V (i )N V (x)

= ZN 1
= ZN 1
i=1
i=1

dx det(x )2 N 1 eN V (x)
(2.3b)
dx eN Veff (x) .
(2.3c)
Here, the subscript N 1 indicates that the quantities concerned are those in the (N
1) (N 1) matrix model, and the expectation value O is defined as

1
O =
(2.4)
d O eN tr V () .
Z
In the large-N limit, the system of an (N 1) (N 1) matrix is the same as the system
of an N N matrix. Hence, we can use the standard N N matrix model to calculate
these expectation values.
The effective potential Veff (x) defined above can be expressed in terms of connected
diagrams. After some algebra, we obtain
2

1
N Veff (x) = N V (x) 2 tr log(x ) c
(2.5)
2 tr log(x ) c .
2
Here, the subscript c indicates the connected part. In the large-N limit, the first and
second terms are of order N and third term is of order N 0 . If we restrict ourselves to the
leading order of N , the terms other than the first two can be neglected. Using the resolvent1

1
1
1
= V (x) V 2 (x) + p(x) ,
R(x) =
(2.6)
tr
N x
2
1 Here the branch of the square root is chosen so that R(x) 1/x as x .
256
the equation can be expressed as follows:

(0)
(1)
N Veff (x) = N Veff + Veff +

(0)
1 (2)
V + ,
N eff

1
2 tr log(x )
N
x
x

= V (x) 2 Re dx R(x ) = Re dx V 2 (x ) + p(x).
Veff = V (x)
(2.7a)
(2.7b)
(2.7c)
The resolvent R(x) has the cut on the real axis. If x is on the cut, the effective potential
becomes constant and the eigenvalue density (x) takes a nonzero value. Interpreting this
in physical terms, we are considering the N th eigenvalue x and the other N 1 eigenvalues are those of whose distribution is expressed by (x). In the cut, where the N 1
eigenvalues are distributed, the forces from them acting on the N th eigenvalue cancel each
other. The effective potential Veff (x) has is at the minimum over the entire cut. From the
standpoint of the original system of the N N matrix including the N th eigenvalue, it
is natural that the eigenvalue density (x) should not change due to the N th eigenvalue
at the leading order of N ; that is, the back reaction is at the subleading order. Hence, in
the integration with respect to x in (2.3), most of the contribution comes from the region
where x is inside the cut. In the case of one cut, the integration over this region gives the
partition function of the N N matrix system. To extend this to type 0B string theory, we
should consider the case of two cuts. In the case of two cuts, there are two minima of the
effective potential. Because the potential under consideration is Z2 symmetric, these two
minima should be symmetric under x x and give the same contribution to the partition
function. Hence, we can deal with these two regions together as inside the cut. There is
another nonzero contribution from the region where x lies outside the cut. Because there
is a maximum of the effective potential, we should take this into account. If the N th eigenvalue x lies outside the cut, the eigenvalue density of the N N matrix system differs
from that of the (N 1) (N 1) matrix system. It can be identified as the instanton
characterized by the configuration where x is located at the local maximum of the effective potential. The resolvent is related to the disk amplitude of the Liouville field theory
with the boundary condition corresponding to the FZZT-brane [10]. The effect of the instanton comes from the (integrated) resolvent with some fixed value of the cosmological
constant corresponding to the local maximum of the effective potential. It is related to the
ZZ-brane [9].
Now, we divide the integration region into two parts, namely, inside the cut and outside
the cut, as

dx det(x )2 eN V (x)
Z = ZN 1
inside the cut
+ ZN 1

dx det(x )2 eN V (x) .
(2.8)
outside the cut
The second term comes from the integration outside the cut, and is identified as the contribution of the instanton. So far, we have restricted ourselves to the N th eigenvalue x and
257
identified the instanton as the configuration where x lies outside the cut. However, there
are N eigenvalues and all other N 1 eigenvalues can be possibly outside the cut as well.
Hence, there is an n-instanton sector; that is, n eigenvalues lie outside the cut. The partition
function can be expressed the sum of those in the n-instanton sector as
Z = Z (0-inst) + Z (1-inst) + Z (1-inst) + ,

Z (0-inst) =
di 2 () eN i V (i )
i
inside the cut
(0-inst)
= ZN 1
(0-inst) N V (x)

dx det(x )2
e
,
(2.9a)
(2.9b)
inside the cut
(1-inst)
(0-inst)
= N ZN 1

(0-inst) N V (x)
dx det(x )2
e
.
(2.9c)
outside the cut
Here, the superscript (n-inst) indicates the n-instanton sector. In the partition function
Z (0-inst) and the expectation value O(0-inst) in the 0-instanton sector, all eigenvalues are
inside the cut. Hereafter we will omit the superscript (0-inst) in the expectation value.
The factor N in front of the 1-instanton sector partition function (2.9c) reflects the number
of ways of specifying an eigenvalue that lies outside the cut. If we consider all n-instanton
sectors, neglecting interaction between instantons, which is valid in the large-N limit, the
partition function can be expressed in terms of those of the 0-instanton and the 1-instanton
sector

Z (1-inst)
(0-inst) +
F
(0-inst)
1 + (0-inst) + = eF
,
e =Z=Z
(2.10a)
Z

Z (1-inst)
= (0-inst) = N outside the cut
.
2 N V (x)
Z
inside the cut dx det(x ) e
(2.10b)
The additional term is the contribution from instantons. We regard this term as the
chemical potential of the instanton. This effect corresponds to the ZZ-brane in the noncritical string theory.
3. Effective potential and orthogonal polynomials

In the previous section we have seen that the partition function has contributions which
come from the multi-instanton sectors, that are characterized as configurations where some
of the eigenvalues are located at the maximum of the effective potential. On the other
hand, in order to compute the chemical potential of the instanton, which is nothing but the
instanton effect in the free energy, it is sufficient to consider the effect of only one instanton.
This amounts to considering det(x )2 in the 0-instanton sector and performing the
integration with respect to x as shown in (2.10b). However, the integration of det(x )2
inside the cut generally contains divergent contributions in the subleading order of N [4].
258
Our expectation here is that these divergences cancel the overall N in (2.10b) to make
the chemical potential finite. In order to confirm this, we have to retain the N -dependence
in the computation. For this purpose, it is appropriate to use the method of orthogonal
polynomials, because it is available for any N .
We begin with definitions and properties of the orthogonal polynomials. The partition
function of the one-matrix model can be expressed in terms of orthogonal polynomial
Pn (x) as

Z=
(3.1)
di det Pn (n ) det Pm (m ) eN i V (i ) .
i
nn
mm
Here, the orthogonal polynomial Pn (x) = x n + O(x n1 ) satisfies the orthogonality condition

Pn (x), Pm (x) = dx Pn (x)Pm (x) eN V (x) = hn nm .

(3.2)
Using this inner product of the orthogonal polynomials, the partition function can be expressed as
Z = N! det(Pn , Pm ) = N ! hN
0
nm
N
1
fnN n .
(3.3)
n=1
Here, fn = hn / hn1 . The orthogonal polynomials can be determined by recursion relations

xPn (x) = Xnm Pm (x) = Pn+1 (x) + sn Pn (x) + rn Pn1 (x),
(3.4a)

Pn (x) = Pnm Pm (x) = N V (X) nm Pm (x).
(3.4b)
It can be easily seen that rn = fn . Eliminating rn and sn , we will obtain a differential

equation that determines Pn (x). However, we use these relations in a slightly different
way. Because the free energy is expressed in terms of rn s, we should determine them.
This can be done using (3.4b). Indeed, rn and sn can be determined by (3.4b), then using
(3.4a), Pn (x) can be expressed in terms of rn and sn .
In the large-N limit, it is natural that any quantity fn with index n can be approximated
by a continuous function f ( ) with = n/N . In fact, fn becomes continuous in the case
of one cut. In order to do this, however, the values of fn and fn+1 should become closer
in the large-N limit. In the case of two cuts, fn cannot be approximated by a continuous
function, but if we consider fn with even n or odd n separately, they can be approximated
by different continuous functions [11]. We use fn to indicate that the index n is even, and
fn for odd n. In the large-N limit, they are approximated by different functions as

fn = fn = f ( ) (n: even),
(3.5)
fn = f( ) (n: odd).
In this limit, a summation over the index n can be approximated by an integration. If we
want to calculate up to the next-to-leading order in 1/N , we should use the EulerMclaurin
259
summation formula. For example, for even N

N

n=0
N
fn =
2
1+ N1
N
d f( ) +
2
0 N1
1
d f( ).
(3.6)
In this manner we treat corrections of the next-to-leading order in a systematic way.

Now, we compute the chemical potential of the instanton

= N outside the cut
(2.10b)
.
2 N V (x)
inside the cut dx det(x ) e
Here, it is easy to show that the expectation value det(x ) can be identified with Pn

Pn (x) = det(x ) n ,
(3.7)
where the subscript n again indicates a quantity for an n n matrix system. Because
the quantity under consideration is not the expectation value of the trace, but of the determinant, the large-N factorization does not hold in this case. In fact, if we define Dn (x)
as

Dn (x) = det(x )2 n ,
(3.8)
it satisfies a recursion relation
Dn = Pn2 (x) + rn Dn1 .
(3.9)
Using this relation recursively, we obtain

DN (x) = PN2 (x) + rN PN2 1 (x) + + rN r1 P02 (x).
(3.10)
This formula enables us to evaluate the chemical potential. Substituting this formula into
(2.10b), the chemical potential can be expressed in terms of the orthogonal polynomials,
which can be determined by (3.4), and we will obtain the definite value of the chemical
potential.
When we evaluate det(x )2 , the result will be different depending on whether
x is inside the cut or outside the cut. This difference can be described as follows. The
orthogonal polynomial Pn can be expressed as the expectation value in the n n matrix
system as in (3.7), and thus if x is inside the cut of this system, Pn (x) is oscillating rapidly,
otherwise Pn (x) is monotonic as x n . As n increases, the oscillatory region becomes wider.
When x is inside the cut, we define n (x) as the minimum value of n such that x is
inside this
oscillatory region of Pn (x). The normalized orthogonal polynomials n (x) =
Pn (x)/ hn take values of the same order for n n (x), while those for n < n (x) n (x)
become small so that the contributions from the last n (x) terms containing Pn (x) (n <
n (x)) in (3.10) are damped exponentially. Hence, we can neglect the term of Pn with
n < n (x) in (3.10). On the other hand, if x is outside the cut, Pn (x) for all n are not
oscillating but monotonic with respect to x as x n and we cannot neglect the latter terms
with Pn (x) for n < n (x). Thus, in the computation of (3.10), we should consider these
two cases, namely, inside the cut and outside the cut, separately.
260
First, we consider the case in which x is outside the cut. In this case, the largest contribution to det(x )2 comes from the first term in (3.10) and contributions from latter terms
become smaller like a geometric series. Using the ratio of the orthogonal polynomials
ekn
Pn (x)
,
Pn1 (x)
(3.11)
DN = det(x )2 can be expressed as

DN = PN2 (x) 1 + rN e2kN + rN rN 1 e2kN 2kN1 + .
In the case of one cut, we can approximate this by a geometric series

n

1
1+O
.
PN2 (x) rN e2kN
DN =
N
(3.12)
(3.13)
n=0
In the case of two cuts, however, we should distinguish a quantity with even n and odd n.
Hence, we approximate this up to O(1/N ) by a sum of two geometric series as
DN
n
PN2 1 + rN e2kN rN rN e2(kN +kN )
(3.14a)
n=0
PN2 (x)(1 + rN e2kN )

,
1 rN rN exp[2(kN + kN )]
(3.14b)
for even N , and

DN
n
PN2 1 + rN e2kN rN rN e2(kN +kN )
(3.15a)
n=0
PN2 (x)(1 + rN e2kN )

,
1 rN rN exp[2(kN + kN )]
(3.15b)
for odd N . In the case of two cuts, we obtain (see Appendix A)

kn = kn(0) +
1 (1)
k + ,
N n
(3.16a)
(0)
1 2
x + rn rn
2x
(0)
1 2
x rn + rn
2x
ekn =
ekn =
2
x 2 rn rn 4rn rn ,
(3.16b)
2
x 2 rn rn 4rn rn ,
(3.16c)
2

1
kn(1) + kn(1) = log x 2 rn rn 4rn rn ,
2
(3.16d)

(0)
where the double sign is understood to be + for x 2 > rn + rn + 2 rn rn so that ekn x as

(0)
|x| , and for x 2 < rn + rn 2 rn rn so that ekn does not diverge at x = 0. Using
these relations, we obtain

2

(0)
(0)
(x rN rN ) exp[kN
+ kN ] N W (x)
1

e
1+O
,
DN =
N
(x 2 rN rN )2 4rN rN
1
W (x) =
d k (0) ( ) + k (0) ( ) = 2
x
dx R(x).
261
(3.17a)
(3.17b)
Here, the sign distinguishes the cases of even N and odd N . The relation between
W (x) and resolvent R(x) can be checked with a concrete potential V (x), or in the double
scaling limit. We will examine this point later.
Next, we consider the case where x is inside the cut. In this case, the contribution from
Pn (x) for n < n (x) is exponentially suppressed. Hence, DN = det(x )2 can be
approximated as
DN =
N

Pn2 (x)
N n1
rN m .
(3.18)
m=0
n=n (x)
In the large-N limit, the summation over n can be replaced by an integration. By definition,
x is in the oscillatory region of Pn (x) with n n (x). Because its frequency is of the order
N as shown in Appendix A, in order to calculate Pn (x)2 in the large-N limit it is sufficient
to take the average of this oscillation, which amounts to dividing the amplitude of Pn (x)2
by 2. In the case of two cuts, generally, we should distinguish a quantity for even
n and a
quantity for odd n. However, after we average this oscillation, n (x) = Pn (x)/ hn can be
approximated by a continuous function of = n/N whether n is even or odd. Finally we
should take the extra factor 2 into account. This factor comes from a relative normalization
between the Pn (x) inside the cut and Pn (x) outside the cut as in the continuity formula in
the usual WKB approximation [4]. Using the asymptotic formula for n (x) inside the cut
given in Appendix A, (3.18) takes a value up to O(1/N ) in the exponent
DN =
N

hN n2 (x)
n=n
= 2e
N Re W (x)
N n1
rN m
(3.19a)
m=0
1
d
rN N
x
4r ( )r ( ) (x 2 r ( ) r ( ))2
= 2N(x) rN eN (V (x)+W0 ) ,
(3.19b)
(3.19c)
where = n /N and W (x) is again related to the resolvent as in (3.17b). Because the
real part of the resolvent becomes V (x) inside the cut, Re W (x) = V (x) + W0 , where
(0)
W0 is a constant that determines the origin of Veff (x). Note that because Re W (x) does
not depend on = n/N for n n , as shown in Appendix A, we can put it outside the
integration with respect to . We have also used the fact that the integration in the second
line can be identified with the eigenvalue density (x), which is shown in Appendix B.
262
Now, we are ready to take the ratio of the partition functions inside and outside the cut,
and obtain the chemical potential of the instanton. This can be expressed as

dxdet(x )2 eN V (x)
.
= N outside the cut
(2.10b)
2 N V (x)
inside the cut dxdet(x ) e
Most of the contribution from outside the cut comes from the saddle point of the effective
potential in the large-N limit. In our case, there is a maximum of the effective potential at
x = 0. This maximum always exists when we consider the case where there are two cuts
that are symmetric under x x. We consider only the instanton that is located at this
maximum. Using the method of steepest descent, we obtain the chemical potential with an
overall factor of O(N 0 ) as

|rN + rN + |rN rN ||

=i
1

N (R (0) 2 V (0)) 4 rN |rN rN |

0

dx 2R(x) V (x) ,
exp N
(3.20)
a
where a corresponds to W0 and can be chosen as any point in the left one of the two cuts. As
(0)
mentioned above, it indeed determines the origin of Veff (x). If there is another maximum
or minimum of the effective potential, we should take it into account in the contribution
from outside the cut. Under the double scaling limit, however, there can be no contribution
from such a saddle point. Hence, it is sufficient to consider the instanton at the maximum
of x = 0. We will elucidate this point further in the next section.
4. Universality of chemical potential

In this section, we consider the double scaling limit in the case of two cuts. This is
obtained by taking the limits N and g gc with a certain combination of them
fixed. Here, g is a parameter of the potential V (x) and gc is its critical value. The critical
point around which the type 0B noncritical string is described is one of the GrossWitten
phase transitions. It can be found as the critical point of the one-matrix model where two
cuts, which are the supports of the eigenvalue distribution, become closer and are merged.
If this critical point is exceeded, we have a one-cut eigenvalue distribution. Hence, this
critical point distinguishes two phases, namely, the one-cut phase and the two-cut phase.
Before taking up the double scaling limit, we consider the behavior of the matrix model
near the critical point in the two-cut case. To compute quantities with the method of orthogonal polynomials, it is necessary to evaluate the value of rn . This can be done by using
(3.4b). Because Pn (x) is monic, Pn,n1 = n. Picking up the coefficient of Pn1 in (3.4b),
we obtain at the leading order of 1/N
g = F (r , r ) = F (r , r ).
(4.1)
[gV (X)]n,n1 .
Here, F (x, y) originates from

In fact, there are two relations derived from
(3.4b), namely, one for even n and one for odd n. If we define F (r , r ) for even n, the same
263
variable for odd n can be obtained by interchanging r and r , which is nothing but F (r , r ).
At the critical point, the two cuts merge and r and r take the same value rc . This means
that r = r = rc for n = N at the critical point. Hence, we obtain gc = F (rc , rc ). Expanding
(4.1) around rc , it can be expressed as
c r ) A(r
c r )
F (r , r ) = gc A(r
c r )2 B(r
c r )2 C(rc r )(rc r ).
B(r
(4.2)
It is necessary for at least one of following conditions to be satisfied so that F (r , r ) =

F (r , r ) holds:
Case 1: r = r .
Case 2: A = A and (rc r ) + (rc r ) = 0.
Case 3: A = A and B = B.
The condition that is relevant to the two-cut case is case 2. In this case, if g gc , a solution
of (4.1) with real rs exists that can be rewritten as

r = rc gc g ,
(4.3a)

r = rc + gc g .
(4.3b)
For g > gc , we cannot take this condition. In this case, the condition r = r will be
satisfied, which is the case of one cut. Using the definition in (4.3), a solution of (4.1) in
this case can be expressed as (see Appendix C)
r = rc
2
(gc g ).
4rc
(4.4)
First, we evaluate the free energy without the contribution from the instanton. In the
GrossWitten phase transition comprising the third order, the third derivative of the free
energy in general has a discontinuity, which is universal in the sense that it is not affected by
details of the potential. Thus, in order to extract the universal part of the free energy, we will
compare the free energy of each phase|one cut and two cuts|and pick up a discontinuity of
(the third derivative of) the free energy, and which is the only contribution to the universal
part. For the free energy in the two-cut phase, from (3.3) we obtain
N2
F=
2
1

d (1 ) log r ( ) + log r ( )
(4.5a)
2 2
d (1 )
(gc g )
rc
(4.5b)
N2
1
0

N2
12
grc
2
(gc g)3 ,
(4.5c)
264
and for the one-cut phase,

1
d (1 ) log r( )
F = N2
(4.6a)
2 2
d (1 )
(gc g )
4rc
1
2
0
N2
0
2 2
d (1 )
(gc g )
rc

N2
24
grc
(4.6b)
2
(gc g)3 .
(4.6c)
Here, we have omitted the terms that do not contribute to the discontinuity of the free energy. In computation of the free energy in the one-cut phase, we have noticed the following
fact: in the one-cut phase, in general there is only one cut constructed from N eigenvalues.
However, if there were n < n0 = Ngc /g eigenvalues, the cut would split into two. Hence,
for < 0 n0 /N , we should take the two-cut solution (4.3a), even in the one-cut phase.
Comparing (4.5c) and (4.6c), we can see that the third derivative of the free energy indeed
has a discontinuity.
Second, we consider the contribution from the instanton in a background with a fixed
number of instantons. In the one-instanton background, an instanton is located at the top
of the effective potential. The contribution to the free energy from this instanton can be
obtained by the height of the potential barrier. The effective potential can be obtained from
the resolvent that is expressed near the critical point as
1
R(x) = x
2
rc

C = 2 ,
g
1

d k (0) ( ) + k (0) ( )
C x a 2 x 2 ,
(4.7a)
a2 =
2
(gc g),
rc
(4.7b)
where we have dropped the contribution from the potential V (x), which is nonuniversal.
Then, the height of the potential barrier is
0
2N
a
2
dx R(x) = N
(gc g)3/2 .
3 grc
(4.8)
This is the contribution to the free energy from the instanton in the one-instanton background. In the n-instanton background, the contribution of the instantons at the leading
order is multiplied by n.
265
Third, we evaluate the chemical potential of the instanton. From (2.10b), (4.3), and
(4.8), we obtain

grc
2
i
3/2
.
N
exp
(g
g)
=
(4.9)
c
3 grc
4 N (gc g)3/4
Now, we are ready to consider the double scaling limit. Two limits, N and g
gc , are related as

2/3
3
N =a ,
(4.10)
(gc g) = a 2 t, a 0.
grc
There is an ambiguity in the definition of t. Here, we define t so that the discontinuity of
the free energy is described as F (t) = |t|/8. Using this definition, the free energy in our
calculation and that in the string equation coincide. The string equation in this definition
takes the form of the Painlev II equation,
th = h3 2t2 h.
(4.11)
Here, h(t) is related to the free energy as

1
t2 F = h2 + f,
t = h2 4f.
2
The solution of (4.11) in the two-cut phase is given by
h(t) = t +
(4.12)
(4.13)
in perturbative expansion. The solution in the one-cut phase is h(t) = 0. Thus, the discontinuity of the free energy is consistent with ours.
In this double scaling limit, the chemical potential of the instanton becomes

2 3/2
i
.
= 3/4 exp t
(4.14)
3
4 t
As shown above, the value does not depend on the details of the potential and hence it is
a universal quantity. Moreover, it agrees with the nonperturbative effect derived from the
string equation. This can be seen as follows. If there are two solutions of the string equation
with the same asymptotic expansion for large t
1
h(t) = t t 5/2 + ,
(4.15)
4
then the difference between them due to the nonperturbative effect is given by
hinst (t) = Const t 1/4 e2/3t
3/2
(4.16)
From this, the nonperturbative effect of the free energy can be obtained as
Finst = Const t 3/4 e2/3t
3/2
(4.17)
which is in accord with our chemical potential of the instanton. This justifies our identification of the instanton in the matrix model as the nonperturbative effect of string theory.
However, the overall normalization factor cannot be determined from the string equation
itself. We have determined it from direct computations in the matrix model and have shown
that it is universal.
266
5. Conclusions
In this paper, we have investigated in full detail the nonperturbative effect in type 0B
string theory, which is defined by taking the double scaling limit of the one-matrix model
with a two-cut eigenvalue distribution. We have computed the contribution from the instanton to the free energy directly in the matrix model. In the double scaling limit, the
chemical potential of the instanton does not depend on the details of the potential, and is
a universal quantity. It takes exactly the same form as the nonperturbative effect derived
from the string equation. Moreover, by computation via the matrix model keeping N finite,
we have fixed the overall factor of the chemical potential, which cannot be determined by
the string equation itself. In [4], it is shown that in (bosonic) c = 0 noncritical string theory,
only the closed string, or the string equation, does not describe the nonperturbative effect
completely, and that the matrix model is more fundamental, capturing the nonperturbative
effect correctly. Here, we have found that this is also the case with type 0B, or c = 0,
noncritical string theory.
It is worth noting a crucial difference between c = 0 and c = 0 string theory. In the
case of c = 0 string theory, the potential of the matrix model is unbounded from below.
Therefore, the vacuum with one cut is unstable and the imaginary part of the free energy
obtained in [4] reflects this instability. On the other hand, in the case of c = 0 string theory
defined by a matrix model with a two-cut eigenvalue distribution, the potential is a doublewell type and bounded from below. Thus, the matrix integration in the definition of the
partition function or the free energy as in (2.1) is well defined and gives a real number. In
fact, in the double scaling limit, the integration outside the cut in (2.10b) becomes
0
rc
t 2/3(t 2 )3/2
d
e
.
2
t 2
(5.1)
Therefore, the contribution from the instanton to the free energy is given by an integration
of a real function, but it diverges. This divergence can be attributed
to the large-N limit,
in the double
where some eigenvalues are at the edge of the cut = t. Moreover,
scaling limit, some of the eigenvalues are pushed out of the cut t, so that
the edge
of the eigenvalue distribution is smeared. In this sense, the boundary at = t between
the outside and inside of the cut becomes subtle, and it becomes somewhat artificial in the
double scaling limit to divide the integration region into these two regions. Because this
divergence originates from only a part of the eigenvalues that spread into the outside of
the cut, it is less divergent compared to the integration over the inside of the cut given as
the denominator in (2.10b), which is of order N as shown in (3.19). Therefore, it may be
possible to change the definition of inside the cut slightly and then renormalize the above
divergence into the contribution in the denominator in (3.19). In that case, it is important
to confirm that the chemical potential is still universal irrespective of a slight change in the
definition of the inside of the cut. However, at least as long as the string coupling constant
gs is sufficiently small,2 we can take advantage of the saddle point method to evaluate the
2 g can be restored on dimensional grounds as t 3/2 t 3/2 /g .
s
s
267
above integration. Thus, the above integration is dominated by the saddle point 0 and
if we can choose the contour of as the whole imaginary axis, the integration becomes
finite and gives an imaginary number as in (4.14). The chemical potential obtained in this
way is meaningful at least for small gs . In fact, the instanton in the matrix model we have
considered corresponds to the D-brane known as a ZZ-brane [9]. This can be checked
by noting that it gives an open boundary to the worldsheet, or more quantitatively, by
comparing the disk amplitude in the fixed instanton background to that in the ZZ-brane
background computed in the super Liouville theory as done in [4]. In type 0B string theory,
the ZZ-brane is unstable. The chemical potential that we have computed as above is a
purely imaginary number. It can be considered to reflect the fact that the D-brane under
consideration is unstable. Although c = 0 string theory does not have the time direction
and the instanton does not decay, if we consider the additional time direction associated
with the energy of the system, the instanton will decay due to its instability. The chemical
potential of the imaginary number can be considered to show this instability. In this sense,
the chemical potential is analogous with the energy of a statistical system, where there
is no time direction, but the energy gives the probability that the system is realized in the
ensemble. Likewise, because the instanton corresponds to the ZZ-brane in c = 0 noncritical
string theory, the chemical potential is related to the decay rate of the ZZ-brane. Note that
it is not necessary for our result to agree with the computations in [5] of the decay rate of
the ZZ-brane (D-particle) in c = 1 noncritical string theory, because the time in the target
space is a priori not the same as the extra time direction associated with the free energy
mentioned above.
The most important problem remaining unsolved is to identify the boundary conditions
of the string equation. Note that the matrix model from which the string equation is derived has no ambiguity. Therefore, it is natural to expect that the matrix model specifies
some boundary conditions, by which we can determine the nonperturbative effect precisely. Because the string equation is a differential equation of the second order, there are
two boundary conditions to be specified. It can be readily seen that the asymptotic behavior
(4.15) needs fine-tuning of one parameter, and therefore there is one boundary condition
remaining. It is not clear in general whether the contribution from the instanton to the
free energy, or the nonperturbative effect in the free energy, has something to do with this
boundary condition. In particular, if gs is sufficiently small, the nonperturbative effect is
exponentially small and it is impossible in general to determine it as an ambiguity to be
added to a perturbative series of the free energy, which is an asymptotic expansion. However, if we concentrate on the process where each term of a perturbative series is completely
zero, we can identify the nonperturbative effect unambiguously like the tunneling effect.
The chemical potential we have computed can be regarded as such an example. That is, it
is purely imaginary and the free energy cannot have an imaginary part perturbatively. We
therefore conclude that by defining the contour of the integration with respect to the eigenvalue as the imaginary axis, we can compute the chemical potential of the instanton, and
this will give a boundary condition for the imaginary part of a complexified string equation.
However, we have not yet fixed a boundary condition for the real part of the string equation. Note here that in order to identify the nonperturbative ambiguity, we should specify a
choice of the contour in general as mentioned in [12].
268
Finally, it would be interesting to apply the computations employed in this paper to

other matrix models; for example, the two-matrix model. This is left for future studies.
Acknowledgements
The authors would like to thank M. Hanada, N. Ishibashi, and T. Matsuo for their fruitful discussions. This work is supported in part by the Grant-in-Aid for Scientific Research
(14540254) and the Grant-in-Aid for the 21st Century COE Center for Diversity and
Universality in Physics from the Ministry of Education, Culture, Sports, Science and
Technology (MEXT) of Japan. The work of T.K. is supported in part by the Special Postdoctoral Researchers Program.
Appendix A. Orthogonal polynomials

In this section, we derive the asymptotic behavior of the orthogonal polynomials both
inside and outside the cut. This can be deduced using (3.4). From (3.4a), we write the
orthogonal polynomials in terms of rn . Using the ratio of the orthogonal polynomials ekn =
Pn (x)/Pn1 (x), (3.4a) can be expressed as
x = ekn+1 + rn ekn .
(A.1)
Here we have used the fact that sn is identically zero due to the symmetry with respect to
x x. In the case of two cuts, we should distinguish quantities for even n and odd n as
follows:

r (for even n),

rn = n
kn = kn (for even n),
(A.2)
rn (for odd n).
kn (for odd n),
From (A.1) we obtain two relations, one for even n and one for odd n:
x = ekn+1 + rn ekn
x=e
kn+1
+ rn e
kn
(for even n),
(A.3a)
(for odd n).
(A.3b)
In the large-N limit, kn , kn , rn , and rn can be expanded as

1
n
n
+ f
+ ,
fn+1 = f
(A.4a)
N
N
N
1
fn = fn(0) + fn(1) + .
(A.4b)
N
Here, fn represents kn , kn , rn , or rn and f ( ) is a continuous function of = n/N corre(1)
(1)
sponding to fn in the large-N limit. We can see that rn and rn are identically zero from
(3.4b). From now on, we will omit the superscript (0) for rn and rn . Substituting (A.4)
into (A.3), we obtain
(0)
(0)
x = ekn + rn ekn
(for even n),
(A.5a)

(0)
(0)
x = ekn + rn ekn
(for odd n),
269
(A.5b)
at the leading order, and

(0)
(0)
0 = kn(1) + kn(0) ekn rn kn(1) ekn

(0)
(0)
0 = kn(1) + kn(0) ekn rn kn(1) ekn
(for even n),
(A.6a)
(for odd n),
(A.6b)
at the next-to-leading order. From (A.5), we obtain

2
(0)
1 2
kn
e =
(A.7a)
x + rn rn x 2 rn rn 4rn rn ,
2x

2
1 2
(0)
ekn =
(A.7b)
x rn + rn x 2 rn rn 4rn rn ,
2x

(0)
where the sign of
is + for x 2 > rn + rn + 2 rn rn so that ekn x as |x| , and
for x 2 < rn + rn 2 rn rn not to diverge at x = 0. Moreover, from (A.6),
2

1
kn(1) + kn(1) = log x 2 rn rn 4rn rn .
2
By using these ks, the orthogonal polynomials can thus be expressed as
n

Pn (x) = exp
kn .
(A.8)
(A.9)
m=1
Because we need to perform computations up to the next-to-leading order of 1/N , we use

the EulerMclaurin summation formula to obtain

N
Pn (x) = exp
2
1+ N1
)+ N
d k(
2
N
Pn (x) = exp
2
1

)
d k(
(for even n),
)+ N
d k(
2
1+ N1

)
d k(
(for odd n).
Substituting (A.7) and (A.8) into (A.10), we find

1/4

n
1 (0)
x2
N
Pn (x) =
W0 x,
+ kn ,
exp
2
N
2
(x 2 rn rn )2 4rn rn
W0 (x, ) =
(A.10b)
0+ N1
(A.10a)
0+ N1
1
d k (0) ( ) + k (0) ( ) .
(A.11a)
(A.11b)
Here,
is
for even n and kn for odd n.
If x is in the oscillatory region of Pn (x), where (x 2 rn rn )2 4rn rn < 0, kn is no
longer real, but Pn (x) should be real. We note here that Pn (x) is a solution of a set of
linear differential equations (3.4) and we have obtained its solution in a classically allowed
(0)
kn
(0)
kn
(0)
270
region, (A.11b) via the WKB approximation. Therefore, in order to find Pn in the oscillatory region which is an analog of a forbidden region, we invoke the continuity condition
[4]. That is, it is sufficient to take a linear combination of two complex solutions of (3.4a)
to obtain the real solution, which is nothing but the real part of the naive expression in
(A.11b) with an appropriate phase factor. We thus obtain Pn (x) in the oscillatory region as

1/4
x 2 rn
Pn (x) = 2
4rn rn (x 2 rn rn )2

n
n
N
1 (0)
N
sin Im
+ kn + . (A.12)
exp
Re W0 x,
W0 x,
2
N
2
N
2
Here, we have used
1
1
Re kn(0) = log rn ,
(A.13)
Re kn(0) = log rn ,
2
2
in the oscillatory region. The additional factor 2 comes from a relative normalization to the
solution for the allowed region (x 2 rn rn )2 4rn rn > 0 in the continuity condition, and
is a constant phase factor. The quartic root is defined as real and positive, and the overall
sign is included in .
If we consider the normalized orthogonal polynomials n (x) = Pn (x)/ hn , This can

be expressed as
1/4

x 2 rN
n (x) = 2

N
1 (0)
n
n
N
sin Im
+ kn + ,
exp
Re Wr x,
W0 x,
2
N
2
N
2
(A.14a)
n
N

n
n
1
Wr x,
= W0 x,
d log r ( ) + log r ( ) .
N
N
2
(A.14b)
For n n (x), where n (x) is defined by (x 2 rn rn )2 4rn rn = 0, n satisfies

n n (x) if x is in the oscillatory region of Pn (x). From (A.13), for n n ,

Re Wr x, =
n
N
= nN

1
1
d k (0) ( ) + k (0) ( ) log rn log rn ,
2
2
(A.15)
hence Re Wr (x, ) does not depend on . Thus, we can set Wr (x, ) = Wr (x, 1) for
n /N . As seen in (A.12), the frequency of the oscillation of Pn (x) is of order N . Therefore,
if we take the average of this oscillation for Pn2 (x), it simply gives 1/2. We thus obtain

|x| r0
2
exp N Re Kr (x, 1) .
n (x)
2
(A.16)
Here, the branch of the square root is chosen to be real and positive. This expression is
convenient for calculating DN because of the independence of at order N .
271
Appendix B. Effective potential and resolvent

In this section, we clarify the relation between the effective potential and the resolvent.
We show that the first derivative of the effective potential can be identified with the resolvent. The coefficient of the resolvent in the double scaling limit can be determined by this
relation.
The relation between the effective potential and resolvent can be derived as follows.
Because PN (x) = det(x ), we have
1/4

n
1 (0)
x2
N
x,
, (B.1)
+
det(x ) =
exp
W
k
0
2
N
2 n
for x outside the cut. Differentiating both sides with respect to x, we obtain

1
1
1
1
(B.2)
tr
det(x ) = det(x )
x K(x) + O
.
N x
2
N
In the large-N limit, the expectation value on the left-hand side factorizes. Hence, we have

1
1
tr
R(x) =
(B.3a)
N x
1

1
1
= x K(x) = x d k (0) ( ) + k (0) ( ) .
(B.3b)
2
2
0
The resolvent inside the cut can be obtained by analytic continuation, and it is consistent
with the choice of the double sign in the definition of k (0) ( ) and k (0) ( ) given in (3.16b)
and (3.16c). We can evaluate the integration in (3.19) by using this relation. In the case of
two cuts, we have

2x
,
x k (0) ( ) + k (0) ( ) =
(B.4)
which is again understood as an analytic function. Thus, the eigenvalue density can be
obtained as

1
i 1
1
i 1
tr
tr
(x) =
(B.5a)
2 N x + i
2 N x i
1
1
x
= Re d
(B.5b)
.
4r ( )r ( ) (x 2 r ( ) r ( ))2
Here, the square root is again defined as an analytic function. The sign of the square root
is positive for positive x. Note that (x) is positive even for negative x because the square
root also becomes negative there. Thus, the integration in (3.19) gives the eigenvalue density.
The relation between the effective potential and the resolvent (B.3) can be checked by
a direct computation near the critical point, where it can be expressed as

x

,
x k (0) ( ) + k (0) ( )

(B.6a)
2
b( ) a ( ) x 2
272
a( )2
2
(gc g ),
rc
b2 ( )
4rc .
(B.6b)
Here, a = a(1) and b = b(1) are identified with the endpoints of the cut. Integrating this
equation with respect to x, the effective potential near the critical point can be obtained
as
1
1 (0)
Veff (x) = x
2
2
rc
C = 2 ,
g
1

) + k(
)
C x a 2 x 2 ,
d k(
(B.7a)
a2 =
2
(gc g),
rc
(B.7b)
and can be identified with the resolvent.
Appendix C. Relation between r in one-cut phase and in two-cut phase

There is a relation between the behavior of r in the one-cut phase and in the two-cut
phase. Here, we use the notations in (4.2)
c r ) A(r
c r )
g = F (r , r ) = gc A(r
2
c r )2 C(rc r )(rc r ),
c r ) B(r
B(r
(C.1)
In the two-cut phase, r and r satisfy

and from (4.1) we consider the case where A = A = A.
the condition (rc r ) + (rc r ) = 0. From these conditions, r near the critical point in the
two-cut phase is expressed as

1/2
r = rc B + B C
(C.2)
gc g .
The double sign indicates + for r and for r , because r vanishes for = 0. In the one-cut
phase, r = r = r , and then r near the critical point is
r = rc
1
(gc g ).
2A
B,
and C can be expressed in terms of F (r , r ) as
A, B,

F (r , r )
F (r , r )
A=
=
,
r r =r =rc
r r =r =rc

1 2 F (r , r )
B =
,
2 r 2 r =r =rc

1 2 F (r , r )
B =
,
2 r 2 r =r =rc

2 F (r , r )
C=
.
r r r =r =rc
(C.3)
(C.4a)
(C.4b)
(C.4c)
(C.4d)
273
Here, we use the fact that F (r , r ) originates from [gV (X)]n,n1 , and that it can be expressed in terms of V (x). Because the potential we consider is even, V (x) has only the
odd power of x and can be expressed as gV (x) = xU (x 2 ). Using this, F (r , r ) is expressed
as

r
r
r
U z+
z+
,
F (r , r ) = dz z +
(C.5)
z
z
z
for even n, and whereas odd n we should use F (r , r ). The condition A = A is not trivial,
but comes from (4.1). Using (C.4), (4.1) expressed as

r
dz
r
z+
.
0 = (r r )
(C.6)
U z+
z
z
z
To satisfy this condition with r = r , we need

r
r
dz
U z+
z+
.
0 = G(r , r ) =
z
z
z
(C.7)
From (C.3), (C.4), and (C.6), we can see

A
B + B C =
.
2rc
(C.8)
This is the relation between r in the one-cut phase and in the two-cut phase.
References
[1] E. Brezin, V.A. Kazakov, Exactly solvable field theories of closed strings, Phys. Lett. B 236 (1990) 144;
M.R. Douglas, S.H. Shenker, Strings in less than one-dimension, Nucl. Phys. B 335 (1990) 635;
D.J. Gross, A.A. Migdal, Nonperturbative two-dimensional quantum gravity, Phys. Rev. Lett. 64 (1990)
127;
D.J. Gross, A.A. Migdal, A nonperturbative treatment of two-dimensional quantum gravity, Nucl. Phys.
B 340 (1990) 333.
[2] F. David, Phases of the large N matrix model mnd nonperturbative effects in 2-D gravity, Nucl. Phys. B 348
(1991) 507;
F. David, Nonperturbative effects in matrix models and vacua of two-dimensional gravity, Phys. Lett. B 302
(1993) 403, hep-th/9212106;
B. Eynard, J. Zinn-Justin, Large order behavior of 2-D gravity coupled to d < 1 matter, Phys. Lett. B 302
(1993) 396, hep-th/9301004.
[3] V.G. Knizhnik, A.M. Polyakov, A.B. Zamolodchikov, Fractal structure of 2d-quantum gravity, Mod. Phys.
Lett. A 3 (1988) 819;
F. David, Conformal field theories coupled to 2-D gravity in the conformal gauge, Mod. Phys. Lett. A 3
(1988) 1651;
J. Distler, H. Kawai, Conformal field theory and 2-D quantum gravity or whos afraid of Joseph Liouville?,
Nucl. Phys. B 321 (1989) 509.
[4] M. Hanada, M. Hayakawa, N. Ishibashi, H. Kawai, T. Kuroki, Y. Matsuo, T. Tada, Loops versus matrices:
the nonperturbative aspects of noncritical string, Prog. Theor. Phys. 112 (2004) 131, hep-th/0405076.
[5] M.R. Douglas, I.R. Klebanov, D. Kutasov, J. Maldacena, E. Martinec, N. Seiberg, A new hat for the c = 1
matrix model, hep-th/0307195.
[6] I.R. Klebanov, J. Maldacena, N. Seiberg, Unitary and complex matrix models as 1-d type 0 strings, hepth/0309168.
274
[7] J. Distler, Z. Hlousek, H. Kawai, Superliouville theory as a two-dimensional, superconformal supergravity

theory, Int. J. Mod. Phys. A 5 (1990) 391.
[8] D.J. Gross, E. Witten, Possible third order phase transition in the large N lattice gauge theory, Phys. Rev.
D 21 (1980) 446.
[9] A.B. Zamolodchikov, Al. Zamolodchikov, Liouville field theory on a pseudosphere, hep-th/0101152.
[10] V. Fateev, A.B. Zamolodchikov, Al. Zamolodchikov, Boundary Liouville field theory. I: boundary state and
boundary two-point function, hep-th/0001012;
J. Teschner, Remarks on Liouville theory with boundary, hep-th/0009138;
B. Ponsot, J. Teschner, Boundary Liouville field theory: boundary three point function, Nucl. Phys. B 622
(2002) 309, hep-th/0110244.
[11] K. Demeterfi, N. Deo, S. Jain, C.I. Tan, Multiband structure and critical behavior of matrix models, Phys.
Rev. D 42 (1990) 4105.
[12] P. Di Francesco, P.H. Ginsparg, J. Zinn-Justin, 2-D gravity and random matrices, hep-th/9306153.
SO(10) MSGUT: spectra, couplings and threshold

effects
Charanjit S. Aulakh, Aarti Girdhar
Department of Physics, Panjab University, Chandigarh 160014, India
Abstract
We compute the complete gauge and chiral superheavy mass spectrum and couplings of the minimal SUSY GUT (based on the 21012612610 irreps as the Higgs system) by decomposing SO(10)
labels in terms of PatiSalam subgroup labels. The spectra are sensitive functions of the single
complex parameter that controls MSGUT symmetry breaking. We scan for the dependence of the
threshold corrections to the Weinberg angle and unification scale as functions of this parameter. We
find that for generic values of the GUT scale parameters the modifications are within 10% of the one
loop values and can be much smaller for significant regions of the parameter space. This shows that
contrary to longstanding conjectures, high precision calculations are not futile but rather necessary
and feasible in the MSGUT. The couplings of the matter supermultiplets are made explicit and used
to identify the channels for exotic (B = 0) processes and to write down the associated bare d = 5
operators (some of both are novel). The mass formulae for all matter fermions are derived. This sets
the stage for a comprehensive RG based phenomenological analysis of the MSGUT.
PACS: 12.10.Dm; 12.10.Kt; 12.60.-i; 12.60.Jv
1. Introduction
The supersymmetric SO(10) GUT based on the 126, 126, 210 Higgs multiplets [14]
has, of late, enjoyed a much delayed bloom of interest motivated by its economy and preE-mail address: aulakh@pu.ac.in (C.S. Aulakh).
doi:10.1016/j.nuclphysb.2005.01.008
276
C.S. Aulakh, A. Girdhar / Nuclear Physics B 711 (2005) 275313
dictivity. Besides the traditional virtues of SO(10) this is the minimal renormalizable model
which has shown itself capable of matching the observed fermion spectra, including the
prima facie GUT repellent feature of maximal mixing in the neutrino sector [57]. Beyond
the traditional scenario of perturbative unification of couplings due to the RG flow between
MS and MX it also offers strong indications that the gauge coupling becomes strong above
the GUT scale. We have argued [8,9] that this necessarily leads to dynamical symmetry
breaking of the GUT symmetry at a scale U (just above the perturbative unification scale
MU 1016 GeV). Utilizing the quasi-exact supersymmetry at the GUT scale we made
plausible [9] a scenario in which U is calculably determined by only the low energy data
and structural features of the theory (such as the gauge symmetry group, supersymmetry
and the very restricted Higgs multiplets available to generate fermion massesparticularly
neutrino massesin a renormalizable theory). This scenario offers interesting possibilities
of a novel picture of elementarity and dual unification characterized by a new fundamental
length scale 1
U characterizing the hearts of quarks [8,9].
The MSGUT is thus the focus of multi-faceted interest and a detailed phenomenological
analysis of the theory in terms of the structure dictated by its GUT scale minimality is thus
called for. However such an analysis has been delayed by the computational difficulty of
obtaining the GUT scale spectra and couplings and the effective Lagrangian describing the
normal and exotic features (baryon and lepton violation, etc.) of the GUT derived MSSM
(i.e., extended by the leading (d = 5) exotic operators of the theory). The spectra and
couplings are necessary to analyse threshold corrections to the gauge couplings near the
GUT scale and are also a crucial input into deriving the Lagrangian for exotic processes
and parameters mediated by GUT scale massive fermions. In [11] we presented techniques
for computing the decomposition of SO(10) invariants in terms of the unitary labels of its
maximal (PatiSalam) subgroup SU(4) SU(2)L SU(2)R . Once this decomposition is
performed the computation of the complete spectrum and couplings is quite easy and the
long standing vagueness regarding the Clebsches that arise can finally be banished. This
allowed us to present, by way of illustration of the power of our method, the two most
important mass matrices (4 4 and 5 5 respectively) affecting electroweak symmetry
breaking, fermion masses and nucleon decay: namely those for the MSSM type Higgs
SU(2)L doublets and baryon number violation mediating SU(3)c colour triplets that mix
with the doublets and triplets of the fermion mass (FM) Higgs (10, 126). Moreover since
our methods allow computation of the actual couplings of Higgs to spinors we could also
obtain the d = 5 operators for baryon violation generated via exchange of triplet higgsinos
contained not only in the traditional (6, 1, 1) submultiplets (of the 10 or those in the 126
[12]) which had been noticed to provide a connection between neutrino masses and proton
decay, but also in other channels arising from the exchange of colour triplets contained
in (10, 1, 3)126 submultiplets involved in neutrino mass generation [11]. A more complete
calculation of these spectra and effective Lagrangians and an initial estimate of their effects
is the subject of this paper.
While the calculations presented were in their final stages we were collaborating and
cross checking with another parallel calculation [13] of mass spectra using a different [10]
method which has since been published. Moreover another group [14,15] has also recently
published a calculation (using the same methods as [13]) of spectra and baryon decay effective potentials recently. As far as computation of chiral superfield spectra are concerned
277
our results coincide (upto normalization and phase conventions) with those of [13]. However both our results diverge [11,13] in certain details from the chiral spectra given in [14].
Moreover as already noted by us in (an update to) [11] we also disagreed with the results
of [14] regarding the higgsino channels available for baryon decay in this model. We found
[11] that [14] obtained couplings between the 126 multiplet and matter in the spinorial 16
representation which were in contradiction with ours [11] not only as regards the numerical
coefficients but also in the heavy Higgs channels to which matter fields couple in a baryon
number violating way. In the revised version of [14], i.e., [15] this defect has apparently
been corrected at least modulo disagreement on values of Clebsches. We shall try to settle
these questions by tracing the reasons for the continuing discrepancy in explicit detail and
confirm our previous assertions. We have also analyzed the gauge Dirac multiplet structure
arising from the super-Higgs effect and the masses and vevs responsible for the type I and
type II mechanisms [16,17] of neutrino mass generation.
We emphasize that our method allows computation, not only of spectra but also of the
couplings of all the multiplets in the theory (whether they are renormalizable or heavyexchange induced effective couplings) without any ambiguity. Moreover our results are
obtained by an analytic tensorial reprocessing of labels of fields in the Lagrangian. This
approach might thus find preferment with field theorists in comparison with the more
restricted capabilities of the approach of [10], which, so far, has not proved capable of
generating all the Clebsches of the SO(10) theory and which relies on an explicit multiplet representative and computer based approach which is tedious to connect to the unitary
group tensor methods so familiar to particle theorists.
It has long been held by some that SO(10) GUTs specially SUSY SO(10) GUTs,
are essentially self-contradictory [18] due to the apparently enormous threshold effects
that might arise due to the large number of superheavy Higgs residuals in these theories.
Thus the authors of [18] speak of the futility of high-precision calculations in SO(10).
However these assertions have never been tested against any actual computations of mass
spectra of a SUSY SO(10) GUT and are only worst case estimates assuming that no cancellations occur. However we expect that cancellations will generically occur since the
lepto-quark mass has no reason to lie at the edge of the mass spectrum nor are the coefficients all of the same sign. With the computed spectra now available, the threshold effects
on observable quantities such as Sin2 w , MX , etc., become computable in terms of the
relatively small number [4] of GUT scale parameters of the MSGUT. In fact [4] a single
complex parameter controls the solutions of the cubic equation in terms of which all the
superheavy vevs are defined. We have performed a preliminary scan of the parameter space
of the MSGUT to see what is the typical size of these corrections. We find the striking result that such threshold corrections are generically 10% of the 1-loop results [2326]
that underpin the GUT scenarios viability. Moreover, for significant and possibly interestingly restricted regions of parameter space these corrections can be much smaller, i.e., as
small as 0.5% of the one loop results. Thus far from indicating futility our results indicate
that a thoroughgoing investigation of the compatibility of low energy precision data with
the threshold corrections may significantly constrain the parameter space of the MSGUT.
In any case we show that inclusion of threshold corrections is necessary and not futile.
Such an analysis is now in progress [27].
278
In Section 2 we present a brief review of the principal features of the minimal SUSY
SO(10) theory [14] and compute the gauge supermultiplet masses. In Section 3 we provide the PS reduction of the SO(10) Higgs superpotential. From this we computed the
chiral fermion mass terms and thus the supermultiplet spectra which we discuss here and
list in Appendix A. In Section 4 we compute the threshold corrections to the 1-loop values
of the unification scale MX and sin2 W (MS ). In Section 5 we present the couplings of
the matter fields to FM Higgs fields in the superpotential as well as their couplings to the
gauginos of the SO(10) model. This permits us to identify the possible channels for baryon
violation in the low energy theory via exchange of higgsinos or gauginos and compute
the relevant effective Lagrangians. Using the associated mass matrices we write down the
d = 5 effective Lagrangians for baryon and lepton number violation which arise via exchange of superheavy fermions. In Section 6 we discuss the mass formulae for the matter
fermions in this model. The Majorana mass terms of the left and right handed neutrinos
and the SU(2)L triplet micro-vev responsible the type II mechanism for neutrino mass is
calculated along with the charged fermion mass matrices. In a final section we discuss our
conclusions and results and plans for further investigations using the results derived here.
2. The minimal SUSY GUT

In accordance with our basic rationale we shall deal with a renormalizable globally
supersymmetric SO(10) GUT whose chiral supermultiplets consist of adjoint multiplet type (or AM) totally anti-symmetric tensors: 210(ij kl ), 126( ij klm ), 126( ij klm )
(i, j = 1, . . . , 10) which serve to break the GUT symmetry to the SM, together with
fermion mass (FM) Higgs 10-plet (Hi ). The 126 plays a dual or AMFM role since besides
enabling SUSY preserving GUT symmetry breaking, it also enables the generation of realistic charged fermion masses and neutrino masses and mixings (via the type I and/or type II
mechanisms); three spinorial 16-plets A (A = 1, 2, 3) contain the matter supermultiplets
126() pair is required to
together with the three conjugate neutrinos ( LA ). The 126(),
be present together to preserve SUSY while breaking U (1)R U (1)BL U (1)Y and is
capable [57] of generating realistic neutrino masses and mixings via the type I or type II
seesaw mechanisms [16,17]. The complete superpotential in this theory is the sum of
1
m
M
WAM = MH Hi2 + ij kl ij kl + ij kl klmn mnij + ij klm ij klm
2
4!
4!
5!
1
+ ij kl ij mno klmno + Hi j klm ( ij klm + ij klm )
4!
4!
(1)
and
1
(5)
f T C i i5 B i1 ,...,i5 .
(2)
5! AB A 2 1
Our notations and conventions for spinors can be found in [11]. The Yukawa couplings

are complex symmetric 3 3 matrices and one of them, say f can thus be
hAB , fAB
diagonalized (by an orthogonal transform Uf U T using a unitary matrix U which leaves the
= F
matter kinetic terms invariant) to a real positive diagonal form fAB
A AB , thus leaving
15 residual real parameters in WFM . In addition the 7 complex parameters in WAM can be
WFM = hAB AT C2 i B Hi +
(5)
279
reduced to 10 real ones by absorbing 4 phases by Higgs field redefinitions. Then together
with the gauge coupling one has in all exactly 26 non-soft parameters [4]. Coincidentally,
MSSM also has 26 non-soft couplings consisting of 9 quark and charged lepton masses,
3 Majorana neutrino masses 3 quark (CKM) and 3 lepton (PMNS) mixing angles and
1 quark but 3 lepton CP phases together with 3 gauge couplings and a parameter. Thus
we see that the 15 parameters of WFM must be essentially responsible for the 22 parameters
describing fermion masses and mixings in the MSSM.
The kinetic terms are given by covariantizing in the standard way the global SO(10)
invariant D-terms

1
1
ij klm + ij klm ij klm + ij kl ij kl + Hi Hi + A A .

(3)
2 5! ij klm
4!
D
Note that the extra factor of (1/2) achieves canonical normalization for the 126 independent component fields of the self-dual (anti-self-dual) 126 (126) representations which
would be otherwise be overcounted. We thus, unfortunately, differ from the normalization
used in the parallel computations of [13] with which our results are nevertheless in agreement after appropriate rescalings and rephasings of parameters and fields. We emphasize
that all our redefinitions of labels are unitary and thus maintain unit norm relative to the
above kinetic terms.1
The economy of the above superpotential Eqs. (1), (2) is remarkable [4]. Its few couplings together with the functional flexibility of the chosen Higgs multiplet set and its
ability (in common with other renormalizable models using just 10126 FM Higgs) to fit
the all the fermion mass data [6,7], justify its claim to being the MSGUT. The small
number of non-soft parameters (26 as in the MSSM) implies that after fitting [57] the
known quark, charged lepton masses and quark mixing angles together with the neutrino
mass splittings very little play is left in the model and it becomes predictive and thus falsifiable. The nearest related model (NMSGUT?) (in some ways more logically complete
since all the FM channels allowed by renormalizability would then be utilized) might be
considered to be the one obtained by adding a 120-plet SO(10) FM Higgs. Alternatively
one may consider SU(5) supplemented with right handed neutrinos or non-renormalizable
terms [4]. Both models are far less economical and not so predictive. Therefore, as advocated in detail in [4], the first priority should be to pin down the predictions of this model.
We began the development of a detailed framework for handling the group theoretic complexity of SUSY SO(10) models generally in [11] and this paper presents the results of
calculations using the techniques developed there for computing couplings and spectra for
MSSM fields from the MSGUT tree action by decomposing the fields according to the
SU(4) SU(2)L SU(2)R or PatiSalam (PS) maximal subgroup.
We now specify how the symmetry is broken down to the MSSM gauge group [1,2,4] by
superlarge vevs contained in the 210-, 126-, 126-plet scalar vevs. Before doing so we introduce our submultiplet naming and indexing conventions. A host of further details related
to the PatiSalam decomposition of SO(10) can be found in our earlier paper [11] where
the foundation for the current program of computation of states, masses and couplings of
1 The relations between the quantities of [13] (denoted by primes) and ours are = / 2,
= /

(also for vevs) = 2 , = 2 , = 2, M = 2M.
280
this theory was laid and the spectrum of MSSM like SU(2)L doublets and SU(3)c baryon
decay triplets first computed.
We denote quantum numbers w.r.t. the SM gauge group by enclosing them in square
brackets while those with respect to the PS group are denoted by round brackets. We have
adopted the rule that any PS submultiplet of an SO(10) field is always denoted by the
same symbol as its parent field, its identity being established by the indices it carries or
by additional sub/superscripts ((a), (s), , L, R) denoting (anti-)symmetry or (anti)-self
duality, if necessary. On the other hand, since one often encounters several chiral MSSM
multiplets of the same type arising from different SO(10) Higgs multiplets we will also
introduce a naming convention using roman letters for these multiplets. If we need to denote the scalar component of a chiral superfield we use a tilde over the superfield symbol
and sometimes use a superscript F to denote fermionic components of chiral superfields
while gauginos are denoted by . Our notation for indices is as follows: the real indices
of the vector representation of SO(10) are denoted by i, j = 1, . . . , 10. The real vector
index of the upper left block embedding (i.e., the embedding specified by the breakup of
the vector multiplet 10 = 6 + 4) of SO(6) in SO(10) are denoted a, b = 1, 2, . . . , 6 and
of the lower right block embedding of SO(4) in SO(10) by ,
= 7, 8, 9, 10. These in 2,
3,
4,
5,
6
dices are complexified via a unitary transformation and denoted by a,
b = 1,
1 , 2,
2 , 3,
3 where 1 1,
2 1 , etc. Similarly we denote the complexified
,
= 1,
9,
10
Using this complexification we showed [11] how
0.
versions of ,
by ,
= 7 , 8,
all SO(6) SO(4) subinvariants of SO(10) tensor invariants could be systematically converted to SU(4) SU(2)L SU(2)R invariants whose indices are as follows: the indices of
2).
Finally the index of
= 1,
the doublet of SU(2)L (SU(2)R ) are denoted , = 1, 2 (,
the fundamental 4-plet of SU(4) is denoted by a (lower) , = 1, 2, 3, 4 and its upper-left
block SU(3) subgroup indices are ,
= 1, 2, 3. The corresponding indices on the 4 are
carried as superscripts. These doublets and quartets correspond to the chiral spinor representations of the SO(4) and SO(6) subgroups of SO(10). Details of the spinorial invariant
decomposition techniques may be found in [11]. The component of the SU(4) adjoint in
(15)
the direction of the Gell-Mann generator i is labeled with a superscript (15) or (B L).
2
Thus the PS decomposition of our SO(10) multiplets is
= 210 = (15, 1, 1) + (1, 1, 1) + (L)

(15, 3, 1) + (R)
(15, 1, 3)
2, 2),
+ , (a)
(6, 2, 2) + , (s)
(10, 2, 2) + (s)
(10,
(4)
= + = 126
(s)
L
(10, 1, 3) + (s)
(10, 3, 1) + (6, 1, 1) + , (15, 2, 2),
=
(5)
= = 126
(s) (10, 3, 1) + (6, 1, 1) + (15, 2, 2),
= R(s)
(10, 1, 3) +
(6)
H = 10 = H (1, 2, 2) + H (6, 1, 1),
(7)
(a)
(a)
1, 2) + F (4, 2, 1),
A = 16 = 16+ = (4+ , 2+ ) + (4 , 2 ) = F (4,
(8)
F (4, 2, 1) = (Q
, L ),
with

Q=

U
,
D
L=

1, 2) = Q
, L
F (4,

,
e
=
Q

d
,
u
L =
281
(9)

e
.
(10)
The GUT scale vevs that break the gauge symmetry down to the SM symmetry are [1,2]

a
(15, 1, 1) 210 : abcd
= abcdef ef ,
(11)
2
where [ef ] = Diag(2 , 2 , 2 ), 2 = i2 . One dualizes
ab
1
abcdef cdef .
4!
(12)
Then in SU(4) notation [ ] this vev is
ia
ia
Diag(I3 , 3)
,
=
2
2

(15, 1, 3) 210 :
ab
= ab ,
(13)
(14)
where [ ] = Diag(2 , 2 ) which translates to

(R)
1 2 = i (R) 0 ,
2

(1, 1, 1) 210 :

= p ,
(15)
(16)

(10, 1, 3) 126 :
(R) 441 1
= ,
1 3 5 8 0
= = i 44(+)
2
(17)

(10, 1, 3) 126 :
(R)44 44
2 4 6 7 9
= = i () = 22 .
2
(18)
Substituting these vevs into the superpotential one obtains

W = m p 2 + 3a 2 + 62 + 2 a 3 + 3p2 + 6a2

+ M + (p + 3a 6)
(19)
the non-trivial F-term conditions are thus

2mp + 62 + = 0,

2ma + 2 a 2 + 22 + = 0,
2m + 2(p + 2a) = 0,

M + (p + 3a 6) = 0.
(20)
(21)
(22)
(23)
282
The vanishing of the D-terms of the SO(10) gauge sector potential imposes only the
condition
| | = | |.
(24)
Except for degenerate cases corresponding to enhanced unbroken symmetry (SU(5)

U (1), SU(5), G3,2,2,BL , G3,2,R,BL , etc.) [4,13] this system of equations is essentially
cubic and can be reduced to the single cubic equation [4] for a variable x = /m:
8x 3 15x 2 + 14x 3 = (1 x)2 ,
(25)
where =
and the other vevs can be expressed in terms of values of the variable x
which solve Eq. (25). This parametrization of the MSGUT ssb problem [4] is of great help
computationally and clearly exhibits the crucial importance of the , i.e., of the ratio M/m.
The important role played by a similar ratio in the other renormalizable SO(10) GUT based
on the 45, 54, 126, 126 representations has already been noted [20].
When we measure vevs or masses in units of m
= m/ we will put a tilde over the sym
bol. We also define the additional dimensionless parameters = / and M H = MH /m.
Then the dimensionless vevs are = x [4] and
M
m
a =
(x 2 + 2x 1)
,
(1 x)
p =
x(5x 2 1)
,
(1 x)2
2 x(1 3x)(1 + x 2 )
.
(1 x)2
(26)
The solutions of the cubic equation (25) are generically complex. We will therefore
nowhere assume hermiticity for our mass matrices, preferring to leave them undiagonalized
for eventual numerical diagonalization so that all our results are applicable in the general
case. We will not generate arrays of expressions in terms of the variable x, although it is
easy to do so since, practically speaking, the substitutions are now handled via a computer
anyway.
We conclude this section with a description of the super-Higgs effect for the breaking
SO(10) SU(3) SU(2) U (1)Y which is achieved by the above superheavy vevs. As
is well known, as a consequence of gauge symmetry breaking, each massive gauge boson forms a massive supermultiplet together with its longitudinal Goldstone pseudo scalar
(and its real scalar partner) as the 4 bosonic degrees of freedom. Its gaugino and the chiral
fermion superpartner of the Goldstone scalar pair make up one Dirac fermion superpartner also with 4 degrees of freedom. This is the so-called Dirac or massive vector gauge
supermultiplet. These gauge boson/gaugino masses are the most fundamental thresholds
of the GUT and it is appropriate to begin with a discussion of their values for this model.
In Section 4 we follow [21,22] to compute the threshold corrections using the spectra we
compute. Then a Dirac gauge coset multiplet in representation R of the MSSM gives rise to
a RG mass threshold above which the gauge and chiral components of the Dirac multiplets
separately contribute 3Ti (R) and Ti (R), respectively, to the beta function coefficients of
the individual MSSM couplings (including the U (1)Y coupling!).
It is easiest to keep track of the gaugino masses and mixings. The combination of chiral
fermions that forms a Dirac fermion together with a gaugino must, for consistency, be a
zero mode of the mass matrix arising from the superpotential and this makes it easy to
disentangle the gauge spectrum even in the case of complex vevs and parameters. For the
symmetry breaking to the MSSM the gauginos of the coset SO(10)/G321 lying in the PS
283
representations (6, 2, 2) (1, 1, 3) plus the triplets and anti-triplets in (15, 1, 1) (i.e., 33
Dirac multiplets in all) obtain a mass by pairing with chiral AM Higgs fermions. One need
only substitute the vevs given above into the PS decomposition of the gaugino Yukawa
terms which have the form

1

1
F
F
F
g 2
+ H.c.
+
ij kl im mj
im
im
kl
ij kln
mj kln
ij kln
mj kln
3!
2 4!
(27)
One finds the following gaugino masses.
(i) G[1, 1, 0]: mG = 10g| |.

The mass term is

g (R0) (15) 44
R + 44R+ + H.c.
2
3
2

G6
mG ei G4 + ei G5 + H.c.,
2

2 (R0)
3 (15)
G6
,
5
5
44
G4 R ,
2
R+
G5 = 44 .
2
(28)
The naming conventions for the chiral states are given in Section 4 and Appendix A. Here
, are the phases of , . Since the representation is real, the mass matrix G in this
sector is symmetric. The complete G[1, 1, 0] sector mass matrix (including gauginos) G is
6 6 while its pure chiral part G (which arises only from the superpotential) is 5 5 and
symmetric and the 5-tuple (0, 0, 0, , ) is both a left and right null eigenvector of Gas
will be obvious when it is presented further
on (Section 2 and Appendix A).
1, 4 ] J [3, 1, 4 ]: mJ = g 8|a|2 + 16||2 + 2| |2 .
(ii) J[3,
3
3
In this case (J4 ) = 4 , (J4 ) = 4 pair up with thecombinations corresponding
to the
T
=
N
(,
2a,
2
2) of
left and right null eigenvectors v0J L = NJ ( , 2a, 2 2), v0J
J
R
the complex, non-symmetric, upper left 3 3 submatrix J of the 4 4 mass matrix J in
the J sector. The gaugino mass terms are

ig J4 2 2a J2 + 4 J3 2 J1

ig 2 2a J2 + 4 J3 + 2 J1 J4 + H.c.
(29)

(iii) F [1, 1, 2] F [1, 1, 2]: mF = g 24||2 + 2| |2 .
The chiral partners of the gauginos

R correspond to the right and left
F3 R+ , F3
null eigenvectors v0F R = (, 12i)T ; v0F L = ( , 12i) of the 2 2 F F chiral
fermion mass matrix. The mass terms are

g F3 i 24F2 + 2 F1 + g i 24F2 2 F1 F3 .
(30)
284

2, 1 ] E[3, 2, 1 ]: mE = g (4|a |2 + 2|w p|2 + 2| |2 ).
3,
(iv) E[
3
3
5 4
The chiral partners of the gauginos E5 4
,E
1
correspond to the null eigenvec2
tors v0ER = (i, 2(a ), p)T ; v0EL = (i , 2(a ), p) of the upper left
3 3 corner E of the E sector 4 4 chiral fermion mass matrix E. E1 , E 1 do not mix with
other E-sector multiplets. The mass terms are

g E 5 i 2 E2 + 2 a E3 + 2 p E4

+ g i 2 E 2 + 2 a E 3 + 2 p E 4 E5 .
(31)

2, 5 ] X[3, 2, 5 ]: mX = g 4|a + |2 + 2|p + |2 .
(v) X[3,
3
3
3 4
The chiral partners of the gauginos X3 4
,X
2
correspond to the null eigenvec1
tors v0XR = ( 2(a + ), + p)T = v0XL of the upper left 2 2 corner X of the 3 3
X-sector chiral fermion mass matrix X . The X-gaugino mass terms are

g X 3 2 a + X1 + 2 p + X2

+ g 2 a + X 1 + 2 p + X 2 X3 .
(32)
3. AM chiral masses via PS

Our approach to opening up the maze of MSSM interactions coded in the deceptive
simplicity of the SO(10) form of the GUT action is to rewrite SO(10) invariants as combinations of PS invariants using the translation techniques developed by us [11]. Although
tedious, our approach allows one to keep track of all phases and normalizations without any
ambiguity. Once this is done making contact with the MSSM phenomenology becomes
trivial since the embedding SU(3) SU(2)L U (1)Y SU(4) SU(2)L SU(2)R is
trivial and transparent if one keeps in mind that
Y = 2T3R + B L.
We obtain for the PS form of the different terms in WAM

m 2
(s)
ij kl = m + (s)
+ (R)
(R) + (L)
(L)
4!

1 (a)
+ (a) 2 ,
2
(a)
M
+ 2
ij klm ij klm = M (a)
5!

(s)
+ (s)
+ (s)
(R)
(R)
(L)

1 3
2
(s)
= i 2i (s)

4!
3

+ (L)

2i (R)
(R)
(L)
(s)(L)

,
(33)
(34)

(s) (L)
(a) (s) (R)

(s)
+ (a)
(R) +
(L)

+ 2 (s)
(s) (R) + (s) (L)

(R)
(R)
(L)
(L)

1
(a)
(a) (R) + (a) (L)
2

2

(R) (R) (R) + (L) (L) (L) ,

+
3
285
(35)

1
= iH(a) (a) + H
H
4!

1
(s)
(s)
H (s) (s) (R)
+ (L)
2
1
(s)
H (a) (s) + H(a)
2

i

H
(R) (L)
2

(s)
+ H (a) (R)
(s)(R) + H(a)
(L)
(L)
1

(a) (a)
H
2
iH(a)
(a)

1 (a)
+ H
(a) ,
2
(36)

1
H = iH(a) (a) + H
4!
(s) (s)

1
H
+ (s)(L)
(R)
2
1
(s)
H (a) (s) + H(a)
2

i

+ H
(R) (L)
2

(s)(L)
(s)(R)
(a)
+ H (a) (L)

+ H

(R)

1 (a)
1 (a)
(a) H + iH(a) (a) + H(a) ,

2
2

= 2i +
4!
+ 2i (s)(R)
+ (s)(L)
(s)(R)
(s)(L)
(37)
286

+ (s)(R) (s)(R) (s)(L) (s)(L)

(s)
+ i 2 (a) (s) + (a)

(s)
+ i 2 (a) (s) (a)

(s)
(s)
2i (s) (L) + (R)
(s)

(s)
(s)
(s)
2i (s) (R)
+ (L)

(R) + (a) (L)

(L)
2 (a) (R)

+ 2 (a) (L)
(L) + (a) (R)
(R)

2 2 (R) + (L)
(s)(R) (s)
(s)(L) (s)
(R)
(R)
(L)
(L)
(s)(L)
(s)

+ i 2 (a) + (a)
(R)
(s)(R)

(s)
+ i 2
(a)
(a)
(L)
(38)
The purely chiral superheavy supermultiplet masses can be determined from these expressions simply by substituting in the AM Higgs vevs and breaking up the contributions
according to MSSM labels.
It is again easiest to keep track of chiral fermion masses since all others follow using
supersymmetry and the organization provided by the gauge super-Higgs effect.
There are three types of mass terms involving fermions from chiral supermultiplets in
such models: (A) unmixed chiral, (B) mixed pure chiral, (C) mixed chiral and gaugino.
3.1. Unmixed chiral
A pair of chiral fermions transforming as SU(3) SU(2)L U (1)Y conjugates pairs
up to form a massive Dirac fermion. For example for the properly normalized fields
44
(R+)
1, 4] = 44(R)
,
A[1, 1, 4] =
A[1,
2
2
one obtains the mass term

= mA AA.
2 M + (p + 3a + 6) AA
(39)
The physical Dirac fermion mass is then |mA | since the phase can be absorbed by a field
redefinition. By supersymmetry this mass is shared by a pair of complex scalar fields with
the same quantum numbers. If the representation is real rather than complex one obtains
an extra factor of 2 in the masses. There are in fact 19 types of such multiplets and their
(roman letter) labels are given along with their masses and SO(10) origins in Table 1 in Appendix A. The case of the sectors C[8, 2, 1] and D[3, 2, 1] bears special mention. The
mass terms for these multiplets arise only between pairs drawn one each from (15, 2, 2),
(15,
2, 2) and there is no mixing between a C, C or D, D drawn from the same SO(10)
287
multiplet simply because the superpotential does not contain any term containing 2 or
2
. This was the reason for the discrepancy in this sector between the results of [11,13]
and [14]: there simply is no such mixing.
3.2. Mixed pure chiral
In this case there are no contributions from the gaugino Yukawas or the D-terms to
the supermultiplet masses, but there is a mixing among several multiplets of the same SM
quantum numbers. There are only three such multiplet types:
).
(a) [8, 1, 0](R1 , R2 ) ( , (R0)
These mix with mass matrix

(m a)
2
R=2
2 m + (p a)
(40)
with both rows and columns labeled by (R1 , R2 ). The masses are the magnitudes of the
eigenvalues of the matrix R:

2

p
p
2

|R | = 2m 1 +
(41)
a
+ 2 = mR .
2
4
While the corresponding eigenvectors can be found by diagonalizing the matrix RR .
2, 1] and colour triplets
The mass matrices of the electroweak doublets h[1, 2, 1], h[1,
1, 2 ] which mix with the multiplets contained in the 10 plet FM Higgs
t[3, 1, 23 ], t[3,
3
are the most crucial ones for determining the phenomenology of the effective MSSM that
arises from this GUT. These matrices were first calculated in [11] (v2) and later, stimulated
by a contradiction with a recent paper [14], the d = 5 baryon violating operators induced
by the exchange of heavy higgsinos were computed and added to a revised version [11]
(v4) by using the Clebsches for the 16 16 126 and 16 16 10 invariants calculated earlier
by us. Thus one has.
(b) [1, 2, 1](h 1 , h 2 , h 3 , h 4 ) [1, 2, 1](h1 , h2 , h3 , h4 ) (H2 , 2
(15)
(15) 441
2
(15) 44
, )
2
, 2
(H 1 , 1 , 1 , ).
2
These multiplets label the 4 rows and columns of the 4 4 mass matrix H [11] which is
given in the collection of mixing matrices in Appendix A. We note that we have redefined
our mass parameters m, M by a factor of 2 relative to those we used in [11]. To achieve
the MSSM spectrum of one pair of light doublets, it is necessary to fine tune one of the
parameters of the superpotential (e.g., MH ) so that Det H = 0. By extracting the null eigenvectors of H H and HH one can compute the coefficients of the various bi-doublets in
the light doublet pair, and, in particular, we can find those for the doublets coming from the
10, 126 multiplets which couple to the matter sector (see Section 6). In this way the SO(10)
can be brought into
constraints on the fitting of the Yukawa coupling matrices hAB , fAB
focus and the invalid assumption that the squares of these coefficients [5,7] add up to 1 can
be dispensed with.
(15)
288
,
1, 2 ](t1 , t2 , t3 , t4 , t5 ) [3, 1, 2 ](t1 , t2 , t3 , t4 , t5 ) (H 4
4
(c) [3,
(a) , (a) , R0 ,
3
3
(a) , (a) , (R0) , 4

).
4(R+) ) (H4
,
4
(R)
For generic values of the couplings all these particles are superheavy. These triplets and
anti-triplets participate in baryon violating process since the exchange of (t1 , t2 , t4 )
The strength of
(t1 , t2 ) higgsinos generates d = 5 operators of type QQQL and lu u d.
the operator is controlled by the inverse of the t t mass matrix T which we computed in
[11] and is given in Appendix A. We shall examine how d = 5 baryon and lepton number
violating operators are generated in Section 5.
3.3. Mixed chiral-gauge

Finally we come to the mixing matrices for the chiral modes that mix with the gauge particles as well as among themselves. Apart from threshold effects, these are of some interest
since one might ask whether the new types of coset gauginos present in SO(10) but not in
2, 1 ] F [1, 1, 2] F [1, 1, 2]
3,
SU(5) namely SO(10)/SU(5) E[3, 2, 13 ] E[
3
4
4
1, ] (the SU(5)/G321 leptoquarks are X[3, 2, 5 ]
G[1, 1, 0] J [3, 1 3 ] J[3,
3
3
2, 5 ]) might not mediate interesting exotic processes by inducing d = 5 operators
X[3,
3
via mixed gaugino-chiral exchange. We have examined this question in some detail in Section 5.
These multiplet sets are:
(15)
44 / 2,
44(R+) / 2,
(a) [1, 1, 0](G1 , G2 , G3 , G4 , G5 , G6 ) (, (15) , (R0) , (R)
( 2(R0) 315 )/ 5)
which mix via a 6 6 mass matrix G given in Appendix A. The complex conjugates of
the 6th row and column form left and right null eigenvectors v0GL , v0GR of the upper left
5 5 block G of G. The determinant of G is generically non-zero although the determinant
of the submatrix G vanishes. It will clearly not affect the evolution of the MSSM gauge
couplings at one loop due to the singlet quantum numbers.
(a)4
2, 1 ](E 2 , E 3 , E 4 , E 5 ) [3, 2, 1 ](E2 , E3 , E4 , E5 ) ( , 4(s)

(b) [3,
, 2 ,
3
3
4 1
2
4
(s)
(a)
4 ,
, 4
, 1 ).
2 ) (
2
4
1
1
4 , ) do not mix with the others) has the

The 4 4 mass matrix E ((E1 , E 1 ) (
2
4 1
usual super-Higgs structure: complex conjugates of the 4th row and column are left and
right null eigenvectors of the upper left 3 3 submatrix E. E has non zero determinant
although the determinant of E vanishes. As for the case of C[8, 2, 1] and D[3, 2, 7/3]
type multiplets one finds that the conjugate types of E type multiplets drawn from the
same SO(10) representation cannot mix. Furthermore explicit computation using the decomposition of the superpotential given in Section 3 shows that E1 [3, 2, 13 ] = 42 and
2, 1 ] = in fact decouple from the other members of the E sector so that the
E 1 [3,
3
4 1
E sector mixing matrix is 4 4 (including gauginos) and 3 3 excluding gauginos. Note
that our assertion is not that these couplings cancel but simply that they do not appear. To
(s)
see why, for instance, there is no term mixing say E1 = 42 with E3 = 4
coming
1
from the (10, 2, 2) we observe that the terms mixing (15, 2, 2) and (decuplet, 2, 2)
289
via a right-handed vev could only come from the following two terms in Eq. (38):
(s)

(s)
= 2i (R)
.
(s) 2i (s) (R)
4!
We see that the pairs (10,

1, 3) and (10, 2, 2) and (10, 1, 3) and (10, 2, 2) simply
do not mix. Now it is obvious that a right handed vev will mix only E 2 coming from
(15,
2, 2) with E3 coming from (10, 2, 2) but not E 1 coming from (15, 2, 2) with
E3 coming from (10, 2, 2). Similar considerations account for the other decouplings
between E1 , E 1 and the rest of the E sector. Ultimately this correlation is accounted for by
the correlation between the duality properties of SO(6) decuplets and SO(4) triplets within
the SO(10) self-dual and anti-self-dual multiplets , .
(15)
44 ,
(c) [1, 1, 2](F1 , F2 , F3 ) [1, 1, 2](F1 , F2 , F3 ) ( 44(R0) , (R) , (R) ) ((R0)
(15)
(R+) , (R+) ).
The mixing matrix F has the usual structure. The residual massive eigenstates after separating off the two Dirac fermions of mass
1/2

mF = g 24||2 + 2| |2
(42)
is a Dirac fermion of mass

mF = | |2 + 12||2
and the form of its chiral parts is
F = NF (i 12F1 + F2 )ei( ) ,
F = NF (i 12F1 + F2 ),

NF1 = 12||2 + | |2 .
(43)
(44)
(R0)
1, 4 ](J1 , J2 , J3 , J4 )[3, 1, 4 ](J1 , J2 , J3 , J4 ) ( 4

(d) [3,
, 4 )
(R) , 4 , 4
3
3
4 , 4 ).
( 4(R+)
, 4 , (R0)
The 4 4 mass matrix J has the usual super-Higgs structure: complex conjugates of the
4th row and column are left and right null eigenvectors of the upper left 3 3 submatrix J.
J has non-zero determinant although the determinant of J vanishes.
4(s)
4(a)
(s)
(e) [3, 2, 53 ](X 1 , X 2 , X 3 ) [3, 2, 53 ](X1 , X2 , X3 ) ( 1 , 1 , 1 ) (4
,
2
(a)
, 2 ).
4
2 4
Mix via a 3 3 symmetric matrix X so the left and right null eigenvectors of the upper left
2 2 submatrix X, formed by the complex conjugates of the third row and column of X ,
are the same. Separating off the two Dirac [3, 2, 53 ] gauge fermions of mass

mX = g 4|a + |2 + 2|p + |2
one is left with a Dirac fermion of mass

2|m|(2|x|2 + |1 x|2 )

mX = 2 2m + (a + ) + |m + | =
|1 x|
290
whose chiral parts are also neatly expressed in terms of x

=
(X, X)
+ |1 x|2

i( )
e m 1x
2xX1 + (1 x)X2 , 2x X 1 + (1 x)X 2 .
2|x|2
(45)
This concludes our description of the superheavy mass spectrum of the minimal SUSY
GUT. As mentioned earlier our results were calculated in collaboration with the authors
of [13] and are in agreement with the chiral spectra calculated in [13]: whose results also
confirm our earlier results [11] on the phenomenologically important matrices H, T . Moreover we have evaluated the mixing of the gauginos with the chiral fermions explicitly and
calculated the gauge spectra and eigenstates besides furnishing all the couplings in the
superpotential sector explicitly. The gauge couplings and the gauge Yukawa couplings to
matter will be given in Section 5. The gaugino mixing with the chiral fields will be useful
to us when we examine B + L violation mediated by gaugino exchange as well by higgsino triplet exchange and when one wishes to examine the flow of gauge couplings past
the gauge thresholds.
4. RG analysis
The first phenomenological success of GUTs was the 1-loop calculation of the numerical value of Weinberg angle [23]. This was followed by the prediction [24] and then the
verification [26] of an amazingly exact compatibility between UV gauge coupling convergence in the MSSM and the precision LEP data. The large mass at which the top quark was
eventually discovered and the associated large value Sin2 W 0.23 verified the originally
somewhat far fetched conjecture of [24]: a historical fact that is still not always appreciated. The proposal of Weinberg [21] for calculating threshold effects within an effective
field theory picture using mass independent renormalization schemes such as the standard
MS renormalization scheme was taken up and developed in detail in [22]. Thereafter, using
these results, it was argued [18] that high-precision calculations in SO(10), and particularly in supersymmetric SO(10) models which used large representations such as 210, 54,
126, etc., were futile. This was due to the huge corrections to the one loop predictions that
they expected in view of the large number of superheavy fields and the expected span in
their masses. It should be remarked however that without an explicit calculation cancellations that might naturally occur would be overlooked. Such calculations were never done.
These negative expectations were a motivation for the development [29] of a whole genre
of SO(10) models that eschewed large representations (and thus parameter counting minimality) in favor of models with a plethora of small representations and non-renormalizable
interactions.
The other approach [14,19,20] approach has all along been to retain renormalizability
of the fundamental theory. We regard retention of Higgs multiplets just adequate to account
for the gauge and fermion spectrum via renormalizable couplings as a sine qua non for
even being clear as to what is testable about a given model. The inverse approach where
representations (hypotheses) are multiplied without necessity seems regressive to us.
291
Thus the proposal of the SUSY SO(10) GUT based on the 21012612610 [1,2] Higgs
system as being the Minimal SUSY GUT [3,4] must live or die by the criterion: Are
the one loop values of Sin2 W and MX generically stable against superheavy threshold
calculations?. By generically we mean: for a non-singular subset of the parameter space.
So far this question could not be answered definitively since no complete mass spectrum
was available in any SUSY SO(10) model to settle the issue. Partly this was due to the lack
of accessible techniques to calculate mass spectra and couplings in these models due to the
difficulty in obtaining the relevant SO(10) Clebsches. Over the last few years we have
developed [11] a complete technology for translating SO(10) tensor and spinor labels into
those of the unitary labels of the PatiSalam maximal subgroup SU(4) SU(2)L SU(2)R
of SO(10). This allowed us to compute first the mass matrices of SM type doublets and
proton decay mediating triplets and then the complete spectrum and couplings reported
in this paper. The partial technology of [3,10] has also been used to compute [13,14] this
spectrum (but not the couplings). With the correct spectrum in hand we can apply the
standard formulae of Hall [22] to compute the changes in the 1-loop GUT predictions as
functions of the few MSGUT parameters ( = M/m, , , , , m, g, MH ) which are
relevant at the GUT scale. Thereafter we can scan the parameter space to see how the
corrections vary with these parameters.
A few remarks on the role of the parameters are in order. The parameter = M/m
is the only numerical parameter that enters into the cubic equation (25) that determines
the parameter x in terms of which all the superheavy vevs are given. It is thus the most
crucial determinant of the mass spectrum. The dependence of the threshold corrections on
the parameters , , , seems quite mild (logarithmic) (this is especially obvious for the
unmixed chiral multiplets) and is also suggested by our preliminary scans of the parameter
space. Thus changing , by a factor of 100 each yields plots vs that seem indistinguishable from the ones presented below. From Eq. (26) we see that m/ can be extracted as the
overall scale parameter of the vevs. Since the threshold corrections we calculate are dependent only on (logarithms of) ratios of masses the parameter m does not play any crucial
role in our scan of the parameter space: it is simply fixed in terms of the (threshold and
two loop corrected) mass MV = MX of the lightest superheavy vector particles mediating
proton decay: which mass is chosen, in the approach of Hall, as the common physical
matching point in the equations relating the running MSSM couplings to the SO(10) coupling [22]. Inasmuch as we take the parameters , as given, and the parameter m is set
by the overall mass scale, the freedom in the parameter is essentially that of choosing the
126126 mass parameter M, i.e., the freedom in choosing the dimensionless parameter ,
is essentially that of the ratio M/m: which ratio is already known to be a crucial control
parameter of symmetry breaking in renormalizable models that utilize the 126, 126 to complete and enforce the symmetry breaking down to the SM symmetry [20]. As for MH it is
fine tuned to keep a pair of doublets light. The relation between the MSSM couplings at
the SUSY breaking scale MS 1 TeV and the GUT coupling at the scale MX is given by
[21,22]
bij
1
1
MX
=
+ 8bi ln
+ 4
ln Xj 4i (MX )
i (MS ) G (MX )
MS
bj
j
(46)
292
here

M0
Xj = 1 + 8bj G MX0 ln X
MS
(47)
is understood to be evaluated at the values of MX0 , G (MX0 ) determined from the one loop
calculations. In this equation the contribution of the Yukawa couplings has not been taken
into account and this should also be done in a full investigation [27]. Here we will confine
ourselves to estimating the corrections using the equations as given above, since these were
already conjectured [18] to lead to a breakdown of the unification scenario. The coefficients

33
1
{b1 , b2 , b3 } =
, 1, 3 ,
16 2 5
1
[bij ] =
(16 2 )2
199/25 27/5 88/5

9/5
25
24
11/5
9
14
(48)

(49)
are the standard one loop and two loop gauge evolution coefficients for the MSSM [28].
The term containing i represents the leading contribution of the superheavy thresholds:
i () =
2
MV
MV
MF
(biV + biGB ) + 2(biV + biGB ) ln
+ 2biS ln
+ 2biF ln
,
21
(50)
where V , GB, S, F refer to vectors, Goldstone bosons, scalars and fermions respectively
and a sum over heavy mass eigenstates is implicit. The formulae for the threshold corrections are
(th) (ln MX ) =
51 (MX0 ) + 32 (MX0 ) 83 (MX0 )

,
10b1 + 6b2 16b3
(51)
M
(th) (Log10 MX ) = 0.0217 + 0.0167(5b1 + 3b2 8b3 ) Log10 0 ,
MX

10(MS )
ij k (bi bj )k MX0
(th) sin2 W (MS ) =
(5b1 + 3b2 8b3 )
(52)
ij k
= 0.00004 0.00024(4b1 9.6b2 + 5.6b3 ) Log10
M
MX0
(53)
where bi = 16 2 bi are the 1-loop beta function coefficients for multiplets with mass M .
To evaluate these formulae it is convenient to group the gaugino contributions along with
the chiral fermions they mix with. The values of the indices S1 , S2 , S3 combined as in
Eqs. (52), (53), i.e., SW = 4S1 9.6S2 + 5.6S3 ; SX = 5S1 + 3S2 8S3 are given in Table 2
in Appendix A.
293
The two loop contributions

(2-loop)

5b1j + 3b2j 8b3j
1
(ln MX ) =
ln Xj ,
10b1 + 6b2 16b3
bj
j

(2-loop) sin2 w (MS ) =

10(MS )
bkl
ij k (bi bj )
ln Xl .
(5b1 + 3b2 8b3 )
bl
(54)
ij kl
Using the values

0
G
(MX )1 = 25.6,
11 (MS ) = 57.45,
MX0 = 1016.25 GeV,

21 (MS ) = 30.8,
MS = 1 TeV,
1
3 (MS ) = 11.04
extrapolated from the global averages of current data, the two loops effects give

MX
2-loop
log10
= 0.08,
2-loop sin2 W (MS ) = 0.0026.
MS
(55)
(56)
The values of the 1-loop coefficients bi = 16 2 bi corresponding to vector, complex

scalar and Weyl fermion fields are 11S(R)/3, S(R)/3, 2S(R)/3 where S(R) is the index
of the relevant representation. Note in particular that this implies that the non-zero hypercharge superheavy vector multiplets which are present in SO(10) models will contribute
with negative coefficients to the evolution of even the U (1)Y coupling.
We have computed the threshold corrections for a range of values of keeping the other
insensitive parameters fixed at randomly chosen representative values
= 0.12,
= 0.21,
= 0.23,
= 0.35.
(57)
The results for different values of these parameters (but with the same ) are very similar. We will therefore keep them fixed at these values throughout since here we only wish
to illustrate the feasibility of precision RG calculations in the SO(10) MSGUT.
For real values of the superpotential parameters the cubic equation (25) that determines
the vevs has one real and two complex (conjugate) solutions. The latter give essentially
identical corrections. So for real we need to present plots for two solutions only. These
are given as Figs. 16.
From Figs. 1, 3 see that for most real values of the threshold effects on sin2 W (MS )
are less than 10 % of the 1-loop values. There are three exceptional values of very close
to which this limit is breached but even then the change is only about 25%. For large magnitudes of an asymptotic regime of around 10% change seems to supervene. Similarly
Figs. 2, 4 show that the change in MX is also not drastic (though possibly phenomenologically interesting since the gauge contribution to the nucleon lifetime goes as MX4 )
except at certain special points among which one recognizes certain known points of enhanced symmetry [4,13] such as = 5, 10 (SU(5)), = 3 (GLR ), = 2/3 (flipped
SU(5) U (1)). It is natural to expect that something similar accounts for the other sharp
peaks and dips in these plots. Moreover their narrowness emphasizes that for generic values of the parameters one may expect the threshold corrections to be small for the real
real x cases. There are also regions in which the threshold corrections to Log10 MX are
as large as 5 and these need special examination with regard to their phenomenological
294
Fig. 1. Plot of the threshold corrections to Sin2 w vs for real : real solution for x.
0 vs for real : real solution for x.

Fig. 2. Plot of the threshold corrected Log10 MX /MX
viability and consistency with the one scale breaking picture. It is interesting that in this
way one can scan the parameter space of the MSGUT and obtain a global tomograph of
the variation in its character with the ratio M/m.
Figs. 3, 4 give a magnified view of the region | | < 2. A comparison of the graphs shows
clearly that the peaks in the threshold corrections coincide by either measure, obviously
because some particles are becoming very light and enhancing the mass ratios that enter
the formulae. It will be amusing to use these plots to identify and unravel the special regions
of the MSGUT parameter space.
295
Fig. 3. Magnified plot of the threshold corrections to Sin2 w vs for real : real solution for x.
0 vs for real : real solution for x.

Fig. 4. Magnified plot of the threshold corrected Log10 MX /MX
Let us turn next to the complex solutions of the cubic equation for x but still with real
values of . We obtain the typical plots Fig. 5, 6. The corrections to sin2 W (MS ) are very
small for small | | < 2, with a minimum close to = 1. From Fig. 6 we see that apart
from the two peaks near = 5 the corrections to the unification scale are quite small for
small .
When we consider complex values of as shown in Figs. 712 we see that the behavior
is quite regular (like the case of the complex solutions for real ) and once again there are
large regions of parameter space where the corrections are less than 10% for Sin2 w while
296
Fig. 5. Plot of the threshold corrections to Sin2 w vs for real : complex solution for x.
0 vs for real : complex solution for x.

MX changes by a factor of 10 or less. Thus even this cursory scan of the MSGUT parameter space shows that, quite contrary to expectations in the literature [18], precision RG
analysis of the SO(10) MSGUT is far from being futile, since the hierarchy of magnitudes
between MSSM one loop gauge coupling convergence values (O( 1 ) effects) and the one
loop threshold and two loop gauge coupling corrections (O(1) effects [22]) is generically
maintained at the level of 10% or less. Furthermore the RG analysis and parameter scan
in terms of the single parameter can teach us much about the structure of the parameter
297
Fig. 7. Plot of the threshold corrections to Sin2 w vs Re( ) for complex : Im = 1.2, first solution for x.
0 vs Re( ) for complex : Im = 1.2, first solution for x.

space since it shows a sharp sensitivity to points of enhanced symmetry. We will return to
these questions at length in the sequel [27].
5. d = 5 operators for B, L violation
5.1. Higgsino exchange mediated violation
In SO(10) B L is a gauge symmetry. Thus the vertices preserve this symmetry and the
vevs are just
leading effects of the spontaneous violation of B L by the superheavy ,
298
Fig. 9. Plot of the threshold corrections to Sin2 w vs Re( ) for complex : Im = 1.2, second solution for x.
0 vs Re( ) for complex : Im = 1.2, second solution

for x.
the neutrino mass phenomena. To examine the generation of effective (non-renormalizable)

operators that violate B + L in the low energy theory via higgsino and gaugino exchange
we need the MSSM break-up of the SO(10) invariants 16 16 10 and 16 16 126. Since
we had already presented the PS decomposition of these invariants in [11] it is a trivial
exercise to use that result to obtain the MSSM wise decompositions:
299
Fig. 11. Plot of the threshold corrections to Sin2 w vs Re( ) for complex : Im = 1.2, third solution for x.
0 vs Re( ) for complex : Im = 1.2, third solution for x.

H
WFM
= hAB AT C2 i B Hi
= 2hAB H A B + H A
B H A B + A B

= 2 2hAB t1 ( u A dB + QA LB ) + t1 QA QB + u A eB dA B
2

2 2hAB h 1 [dA QB + eA LB ] + 2 2hAB h1 [u A QB + A LB ],
(58)
(5) (5)
300
1 T (5)
C2 i1 i5 i1 ,...,i5
5!

(a)
+ 4 2 +
= 2 2 (a)

+
+ 4
WFM
=
(59)
whence

t2 QA QB u A eB + A dB + t2 (QA LB u A dB )
WFM
= 4 2fAB
2

i
+ 4 2fAB
h 2 (dA QB 3eA LB ) h2 (u A QB 3A LB )
3
+ 2(C1 dA QB C2 u A QB ) + 2(E1 dA LB D2 u A LB )
+ 2(D2 eA QB E2 A QB )

A Q
+ 2i(A eA eB G5 A B ) 2 2i F1 eA B
+ 4fAB
Q
B
A LB )
+ (W QA QB + 2P QA LB + 2OL

2it4 (dA B + u A eB ) + 2i 2(K dA eB J1 u A B ) .
(60)
We have suppressed G321 indices and used a sub multiplet naming convention specified
in Section 2 and Table 1 in Appendix A.
In order that the exchange of a higgsino that couples to matter with a given B + L
lead to a B + L violating d = 5 operator in the effective theory at sub GUT energies it
is necessary that it have a non-zero contraction with a conjugate (MSSM) representation
higgsino that couples to a matter chiral bilinear with a B + L different from the conjugate
of the first B + L value. Inspecting the above superpotentials one finds that only {t(1) , t(2) }
and {t(1) , t(2) , t(4) } satisfy this requirement. Terms containing the right handed neutrinos A
must be further processed to integrate out the heavy field A in favor of the light neutrinos
A . This will introduce an extra factor of mDirac /MMajorana

and effectively lead to amplitudes suppressed like those of d = 6 operators. Thus on integrating out the heavy triplet
Higgs supermultiplets one obtains the effective d = 4 superpotential for baryon number
violating processes:

1
B=0
Weff
(61)
= LABCD QA QB QC LD + RABCD ( eA u B u C dD ),
2
where the coefficients are
LABCD = S1 1 hAB hCD + S1 2 hAB fCD + S2 1 fAB hCD + S2 2 fAB fCD
(62)
RABCD = S1 1 hAB hCD S1 2 hAB fCD S2 1 fAB hCD + S2 2 fAB fCD
i 2S1 4 fAB hCD + i 2S2 4 fAB fCD ,
(63)
and
301
where S = T 1 and T is the mass matrix for [3, 1, 2/3]-sector triplets: W = tT t + ,

while

fAB = 4 2fAB
.
hAB = 2 2hAB ,
(64)
This expression and the Clebsches contained in it, as well as the new baryon decay
126 (10, 1, 3) (the same PS multiplet
channel mediated by the triplets (t(4) ) contained in
that contains the Higgs field responsible for the right handed neutrino Majorana mass),
126 mediated decay focused on the multiplets
were given in [11]. Previous work [12] on
(2)
(2)
t , t and found that there was no contribution of t (4) , t(4) in their models. This new
channel nominally strengthens the emergent link between neutrino mass and baryon decay. Note however that t(4) couples only to the RR combinations d + u e and as such its
exchange will contribute only to the RRRR channel which, at least in SO(10), seems [12]
generically suppressed except at very large tan . However the mixing in the triplet mass
matrix could also strengthen the effects of this channel.
5.2. Novel d = 5, (B + L) = 0 operators via superheavy gaugino exchange?
A novel situation apparently arises in this GUT due to exchange of superheavy gaugino
Dirac multiplets that couple to matter both via the gauge Yukawa couplings of their gaugino part and the superpotential couplings of their (126) chiral components to the matter
sector. Such gauginos are not present in the case of SU(5) As is evident from Eq. (60)
2, 1/3], F1 [1, 1, 2], J1 [3, 1, 4/3] (which mix with the
the 126 submultiplet fields E 2 [3,
superheavy SO(10)/SU(5) coset gauginos: see Sections 2 and 3) couple only to terms
containing at least one superheavy neutrino field A . Thus, to leading order in MU1 , the
exchange of such gaugino Dirac multiplets will not lead to any d = 5 operator with 4 light
external fields. However a puzzle remains.
The superheavy neutrinos mix with the usual light neutrinos via Dirac masses. So in the
effective theory one trades them for the light neutrinos by using their equations of motion
Majorana
to leading order in their (inverse) masses (effectively = 2(mDirac
/M
) + ).
The chiral parts E2 , F1 , J1 of the gauge Dirac E, F , J multiplets therefore couple to light
neutrinos and another light matter field with a small coupling O(mDirac /MMajorana ). Exchange of the gaugino Dirac fermion between a gauge Yukawa vertex and a 126 16 16
vertex can lead to effective operators involving 4 light matter fields of which at least one
is a light neutrino and one is anti-chiral. This appears to violate the usual argument that in
the effective MSSM arising from a SUSY GUT, supersymmetric D terms involving 4 light
(mixed chiral and anti-chiral) fields must be d 6 or equivalently that the d = 5, B, L vi F . Exchange of SO(10)/SU(5) coset
olating terms are either of form [QQQL]F or [eu u d]
gauginos peculiar to SO(10) however appears to lead to (admittedly suppressed) d = 5
chiralanti-chiral operators with 4 light fields. These operators arise once the electroweak
scale vev that gives rise to neutrino Dirac masses is turned on. This vev is smaller than MS
and arises after soft SUSY breaking terms are included. In this theory B L is spontaneously broken giving rise to the Majorana mass for conjugate neutrinos (which was used
to eliminate them in favor of the SM neutrinos). Thus perhaps the contradiction is not as
violent as it seems at first. We emphasize that there are no analogous processes in SU(5)
302
SUSY GUTs since there the 12 coset gauginos acquire partners from the purely AM type
24 plets which do not couple to the matter sector.
The couplings of the gauginos of SO(10) to the matter fields are easily computed by
adapting the PS reduction of the SO(10) covariant derivative for the spinor 16 [11]:
A

tA
A t
A

+
LgY = 2ig
2
2

L
+
+
2
2
+
(65)
+ + H.c.
2
The terms carrying B, L are
L(B+L)=0

= 2g L J4 Q + Q J4 L 2g d J4 e + u J4 + e J4 d + J4 u

+ (g/ 2) d X 3 L u E (5) L + e X 3 Q + E 5 Q + d E5 Q + u X3 Q

(X3 e E5 ) + Q (E 5 d X 3 u)
+ (g/ 2) L (X3 d E5 u)
Q
+ .
(66)
There are no X[3, 1, 5/3] sector submultiplets in the 126. Thus we can focus on just
the E[3, 2, 1/3] and the J [3, 1, 4/3] sectors here. As discussed in Section 2, the super 2, 1/3].
heavy gauginos J4 , E5 mix with 126 derived fermions J1 [3, 1, 4/3] and E 2 [3,
Examining Eq. (60) we see that J1 , E2 couple only to operators involving at least one
superheavy field (E1 126 does not mix with E-gauginos):

WFM
= 8 2fAB
[E 2 QB + iJ1 u B ]A

= [E 2 QA + iJ1 u A ] f M 1 mD
B + .
(67)
AB
Since J1 couples to a B = 1/3, L = 1 operator while J4 couples only to B = 1/3,

L = 1, J exchange does not lead to B + L violation. The E sector gaugino, i.e., E5
couples as

g
(68)
d E5 Q + Q E5 L E5 u .
2
Only the first terms in (67), (68) are relevant and thus we find the following effective
Lagrangian due to superheavy gaugino exchange:
1 (D)

A B ) ,
m
E 15 2 d C QC (QA B + Q
L(B+L)=2 = 4g f M
(69)
AB
where E 15 2 is essentially the mass of the exchanged gaugino times mixing factors written
compactly in terms of the inverse of the relevant fermion mass matrix (in the E[3, 2, 13 ]
sector). By dressing this with MSSM gauginos we obtain B = L = 1 violating 4-Fermi
vertices responsible for processes like
uL dL dL + L .
(70)
303
This is a vertex quite distinct from the higgsino mediated vertices since it involves exchange of massive gauginos between a chiral and an anti-chiral vertex. It requires non
zero external momenta for the fermions and vanishes in the limit of zero external momenta. Thus the coefficient of the corresponding 4-Fermi operators for B violation in the
effective Lagrangian is M D mNucl /MX2 MS2 where MS is the SUSY breaking scale. This
magnitude seems hopelessly suppressed (relative even to gauge boson exchange) to be observable. Nevertheless the contrast of its structure with that of the standard QQQL and
u u d e operators perhaps warrants a more thorough investigation of the conditions for the
possibility of its appearance in the effective theory.
6. Fermion mass formulae

A vital issue for any SO(10) GUT is the type of predictions it makes for the relations
among the parameters of the (type I and type II) seesaw mechanisms [16] by which neutrino
masses and mixings arise. From the coupling of neutrinos to the 126 we find that the
Majorana mass matrix of the superheavy neutrinos A is (Eq. (60))

(R+)
MAB
(71)
44
= 4 2fAB
= 4i 2fAB
.
Similarly the Majorana mass matrix for the left neutrinos A is (Eq. (60)).
11

MAB
O = 8ifAB
= 4 2fAB
O
,
(72)
where O
is the small vev of the SU(2)L triplet in the (10, 3, 1) induced by a tadpole
that arises as a consequence of SU(2)L breaking (see below).
In addition to this there is the Dirac mass which mixes the left and right neutrinos:
(2)

D
mAB
(73)
= 2 2hAB h(1)
2 + 4i 6fAB h2 .
We must make the fine tuning Det H = 0 necessary to keep a pair of Higgs doublets H(1) ,
H (1) (which is to develop the EW scale vev) light. Then these doublets will be the left and
right null eigenstates of the mass matrix H. If the bi-unitary transformations responsible
for diagonalizing H H and HH are U , U , i.e.,
(1) (2)

Diag mH , mH , . . . = U T HU
(74)
then writing
h(i) = Uij H (j ) ,
h (i) = U ij H (j ) ,
(75)
where H (j ) , H(j ) are the mass eigenstate doublets, the contributions of any coupling in
which h(i) , h (i) enter can be accounted for in the effective MSSM below the heavy thresholds of the GUT just by replacing h(i) i H (1) , h (i) i H (1) where i = Ui1 , i = U i1
and we have numbered the massless doublet pair 1. These components are easily obtained from the normalized null eigenvectors V , V of H H and HH to be Ui1 = Vi , U i1 =
Vi . Thus the neutrino Dirac mass matrix becomes

D
MAB
(76)
= (2 2hAB 1 + 4i 6fAB
2 )vu .
304
To obtain the final formula for the neutrino masses and mixings we must eliminate the
fields which are superheavy and evaluate the tadpole that gives rise to the type II seesaw.
3, 1)126 vevs, inspection of the

The first step is standard. As for the O(10, 3, 1)126 , O(10,
mass spectrum (Table 1 in Appendix A) and Eqs. (36)(38) yields the relevant terms in the
superpotential as
44
= MO O O H 44 O H O
WFM
2
2

44
2 2i 4 4 44 O + 4 4 O
(77)
(4)
44
(4)
since 4 4 1 = 2 3i h (3) , 4 4 2 = 23i h(2) , 441
= 2h
, 2
= 2h , one gets for
the relevant terms

i
i
2
MO O O+ + O 1 + 2i 3 3 4 vd O+ 1 + i2 32 4 vu2 .
2
2
(78)
Thus the vev we need, i.e., O

is immediately determined to leading order in MW /MU
by the equation for O+ as
2

vu
i
O
= 1 + 2i 32 4
(79)
MO
2
and MO can be read off from Table 1 to be MO = 2(M + (3a p)).
The quark and charged lepton mass matrices are

2
d
if 2 vd ,
M = 2 2h 1 4
3

2
u

M = 2 2h 1 4
if 2 vu ,
3
M l = (2 2h 1 + 4 6if 2 )vd .
(80)
(81)
(82)
These formulae are now in a form ready to use for fitting the fermion mass and mixing data
after lifting it via the RG equations of the MSSM to the GUT scale.
7. Discussion and outlook
In this paper we have calculated the complete superheavy spectrum of the minimal supersymmetric GUT along with the gauge and chiral couplings of all MSSM multiplets in
a readily accessible form. Partial calculations of these spectra and couplings [3,11,13,14]
have been published earlier but our method is different from the computer based method
of [3,13,14] and is more complete, especially regarding couplings. Being analytic and explicit it also allows us to trace and resolve discrepancies arising within the computer based
approach. We used the calculated spectra to perform a preliminary scan of the parameter
space of the MSGUT as regards the magnitude of the threshold corrections to two crucial
phenomenological parameters of the MSGUT: the Weinberg angle at low energy and the
mass of the X lepto-quark gauge supermultiplet. We obtained a result that is in sharp contrast with expectations in the literature [18] that precision RG calculations in SO(10) are
305
futile. On the contrary we find that the 1-loop GUT threshold and gauge two loop contributions are modest but significant. Thus, on the one hand, the basic GUT picture suggested
by the convergence of gauge couplings in the MSSM is in fact not destroyed by the contributions of the large number of superheavy fields. On the other hand extant precision
calculations that ignore threshold effects in SO(10) GUTs seem to be of dubious validity. In
particular the proper RG analysis of the MSGUT taking into account EWRSB, all fermion
masses and GUT threshold effects still remains to be done. This calculation is now being
performed [27]. In view of the other phenomenological successes of renormalizable SUSY
SO(10) [57] and the unforeseen correlations between disparate phenomena like neutrino
oscillations and nucleon decay that have emerged [11,12] the mildness and calculability
of threshold effects in the MSGUT is a most welcome and promising development. Our
preliminary scans of the MSGUT parameter space (whose very feasibilitybased on there
being just one sensitive control parameter ( )is a matter of some astonishment) show
that the threshold effects can potentially narrow down the allowed regions of the MSGUT
parameter space and indicate correlations between the GUT scale and the B L violating scale which can be of crucial significance when cross checking the particle physics
phenomenology against cosmology. We have also argued that above the perturbative unification scale realistic renormalizable SO(10) GUTs are necessarily strongly coupled [8,9].
We have recently reported the results of 2-loop calculations of MSGUT RG equations
above the SO(10) restoration scale which we used to show that the strong growth of the
SO(10) coupling above MX cannot be evaded by taking shelter in a weakly coupled fixed
point [30]. On the other hand our work [8,9] has shown that a scenario of a calculable
dynamical symmetry breaking of the GUT symmetry which utilizes the nearly exact supersymmetry at the GUT scale offers rich possibilities for the significance of the new length
scale associated with the condensation of SO(10)/G321 coset gauginos implied by both
holomorphic analysis and by the Konishi anomaly. The present calculations show the way
for crossing the threshold and entering into the SO(10) regime in a controlled way. The
emerging coherence of the low energy phenomenology, B and L violation, perturbative
GUT structures (such as the natural R-parity preservation [20,31], successful seesaw scenarios, leptogenesis, etc.) and exciting hints of deeper mysteries, perhaps unavailable [8,9],
carry to our hopeful nostrils the spoor of a grail perhaps within reach.
Note added
After this paper was posted on the arXive as hep-ph/0405074 the authors of hepph/0405300 claimed that the mass spectra listed in Appendix A were not internally consistent with the requirements of SU(5) or SU(5) U (1) symmetry (at the special vevs where
p = a = ). However this is incorrect. The mass spectra we derived via a PS decomposition of SO(10) organize straightforwardly and termwise into appropriate SU(5) invariants
for SU(5) invariant vevs as given in Appendix B (added in hep-ph/0205074 v2). This termwise reorganization of several hundred G123 invariant mass terms into SU(5) invariant
mass terms is a more stringent consistency test than the tests of hep-ph/0405300 which are
based on traces and determinants and valid only for their conventions. The phase conventions and field normalizations of hep-ph/0405300 are quite different from our work. Thus
the blind application of their trace and determinant consistency tests to our results can-
306
not but fail. We maintain unit field normalizations throughout by using only unitary field
redefinitions of fields with canonical kinetic terms. Finally our results for chiral spectra
also coincide, upto minor convention related adjustments, with those obtained in a parallel
computation reported in [13].
Finally we stress that our method [11] yields all coupling coefficients between both
spinor and tensor irreps and not just the tensor irrep ones relevant for masses and symmetry
breaking which were obtained using the method of [10]. Moreover we note that the most
complex of the mass matrices given here namely those of the Higgs doublets and triplets
relevant for proton decay were already derived by us in hep-ph/0204097 v2 (2003) [11].
Further note added
After version 2 of this paper (including the SU(5) reorganization given in Appendix B)
was accepted for publication the authors of hep-ph/0405300 issued yet another preprint
(hep-ph/0412348 v1) this time claiming that although our results pass the SU(5) reorganization test for = = 0 they failed to do so for = = 0 and that the counting
of Goldstone modes and distinct mass eigenvalues was, in their opinion incorrect. Further
they claimed that our results were inconsistent since they failed to pass certain trace and
hermiticity tests that they had applied successfully to their own results. All these claims
are incorrect. Our results in fact pass all three tests. We have issued a preprint showing this
explcitly [33]. Here we only remark that once the super-Higgs effect for SO(10) MSSM
has been verified it is scarcely feasible that the GoldstoneHiggs counting could fail for
SO(10) SU(5) since the latter is a special case of the same spectra! However the reader
can easily check that the SU(5) singlet and 10-plet mass matrices have zero determinant confirming that the required Goldstone supermultiplets 1 + 10 + 10 are present. The
demonstration that the trace constraints and hermiticity tests of hep-ph/0205300 are also
satisfied is also straightforward once proper account is taken of the difference in the phase
conventions of the two calculations. Details may be found in [33]. Finally as this paper goes
to press the authors of hep-ph/0412348 v1 have reissued the preprint hep-ph/0412348 v2
in which all the claims of the inconsistency of our results are totally retracted.
Acknowledgements
It is a pleasure for C.S.A. to acknowledge much correspondence and collaboration with
B. Bajc, A. Melfo, G. Senjanovic and F. Vissani while calculating the chiral spectra reported here. C.S.A. is also grateful to B. Bajc and Prof. D.R.T Jones for correspondence
related to RG analysis, to Sumit Kumar for technical help with Appendix B and to the
High Energy Group of the Abdus Salam ICTP, Trieste for hospitality while completing
Appendix B. A.G. thanks H.R.I., Allahabad for its facilities and hospitality. This work was
done under Project SP/S2/K-07/99 of the Department of Science and Technology of the
Government of India.
Appendix A. Tables of masses and mixings
In this appendix we collect our results for the chiral fermion/gaugino states, masses
and mixing matrices for the readers convenience. Apart from the discussion of gauge
307
multiplet masses our results have been obtained in parallel with and are compatible with
those of [13], which, however, are computed with a different normalization for the ,
fields resulting in a difference between the mass and Yukawa coupling parameters M, of
these multiplets in the two starting actions. Moreover certain minor phase differences also
exist between the definitions of representative states used by them and our definitions for
the same states (which follow directly from our consistent definitions of PS tensors from
SO(10) submultiplets). Mixing matrix rows are labeled by barred irreps and columns by
unbarred. Unmixed cases (i) are given as Table 1.
Table 1
(i) Masses of the unmixed states in terms of the superheavy vevs. The SU(2)L contraction order is always F F .
The primed fields defined for SU(3)c sextets maintain unit norm. The absolute value of the expressions in the
column Mass is understood
Field [SU(3), SU(2), Y ]
1, 4]
A[1, 1, 4], A[1,
C1 [8, 2, 1], C 1 [8, 2, 1]
C2 [8, 2, 1], C 2 [8, 2, 1]
2, 7 ]
D1 [3, 2, 73 ], D 1 [3,
3
2, 7 ]
D2 [3, 2, 73 ], D 2 [3,
3
2, 1 ]
E1 [3, 2, 13 ], E 1 [3,
3
1, 8 ]
3,
K[3, 1, 83 ], K[
3
1, 2 ]
6,
L[6, 1, 23 ], L[
3
1, 8 ]
6,
M[6, 1, 83 ], M[
3
1, 4 ]
N [6, 1, 43 ], N [6,
3
O[1, 3, 2], O[(1,

3, +2]
3, 2 ]
P [3, 3, 23 ], P [3,
3
3, 2 ]
W [6, 3, 23 ], W [6,
3
10

I [3, 1, 10
3 ], I [3, 1, 3 ]
S[1, 3, 0]
Q[8, 3, 0]
3, 4 ]
U [3, 3, 43 ], U [3,
3
V [1, 2, 3], V [1, 2, 3]
2, 5 ]
6,
B[6, 2, 53 ], B[
3
2, 1 ]
Y [6, 2, 13 ], Y [6,
3
1, 2]
Z[8, 1, 2], Z[8,
PS fields
Mass
44
(R+)
, 44(R)
2
2
,
1
2
,
1
2
4 ,
4 2
1
4 ,
4 2
1
4 ,
2
4 1
4
4(R) , (R+)
(R0)
( , (R0) )

= , =
=
2

(R+)
( (R+) , (R) )

(R)
( , (R+) )

44

44(L) (L)
,
2
2
4

4(L)
, (L)

(L) ,
(L)
4
(R+) , 4(R)
(15)
(L)
(L)
4
(L)
, 4(L)
2(M + (p + 3a + 6))
2(M + (a + ))
2(M + (a ))
2(M + (a + ))
2(M + (a + 3))
2(M + (a ))
2(M + (a + p + 2))
2(M + (p a))
2(M + (p a + 2))
2(M + (p a 2))
2(M + (3a p))
2(M + (a p))
2(M (a + p))
2(m + (p + a + 4))
2(m + (2a p))
2(m (a + p))
2(m (p a))
44
44 2 1
,
2
2

( , )
1 2

( , )
2 1
(R+)
(R)
2(m + 3(a + ))
2(m + ( a))
2(m (a + ))
2(m + (p a))
308
Table 2
Index values for the 26 different chiral multiplet types (used in the threshold corrections). Except for Q, R, S all
other reps come in complex pairs. SW = 4S1 9.6S2 + 5.6S3 , SX = 5S1 + 3S2 8S3 are the combinations that
enter the threshold corrections to Sin2 W and to Log10 MX
Field [SU(3), SU(2), Y ]
{S3 , S2 , S1 }
SW
SX
A[1, 1, 4]
{0, 0, 12/5}
9.6
12
B[6, 2, 5/3]
C[8, 2, 1]
D[3, 2, 7/3]
E[3, 2, 1/3]
F [1, 1, 2]
G[1, 1, 0]
{5, 3, 5}
{6, 4, 12/5}
{1, 3/2, 49/10}
{1, 3/2, 1/10}
{0, 0, 3/5}
{0, 0, 0}
19.2
4.8
10.8
8.4
2.4
0
6
24
21
3
3
0
h[1, 2, 1]
I [3, 1, 10/3]
J [3, 1, 4/3]
K[3, 1, 8/3]
L[6, 1, 2/3]
M[6, 1, 8/3]
{0, 1/2, 3/10}

{1/2, 0, 5}
{1/2, 0, 4/5}
{1/2, 0, 16/5}
{5/2, 0, 2/5}
{5/2, 0, 32/5}
3.6
22.8
6
15.6
15.6
39.6
3
21
0
12
18
12
N [6, 1, 4/3]
O[1, 3, 2]
P [3, 3, 2/3]
Q[8, 3, 0]
R[8, 1, 0]
S[1, 3, 0]
{5/2, 0, 8/5}
{0, 2, 9/5}
{3/2, 6, 3/5}
{9, 16, 0}
{3, 0, 0}
{0, 2, 0}
20.4
12
46.8
103.2
16.8
19.2
12
15
9
24
24
6
t[3, 1, 2/3]
U [3, 3, 4/3]
V [1, 2, 3]
W [6, 3, 2/3]
X[3, 2, 5/3]
Y [6, 2, 1/3]
{1/2, 0, 1/5}
{3/2, 6, 12/5}
{0, 1/2, 27/10}
{15/2, 12, 6/5}
{1, 3/2, 5/2}
{5, 3, 1/5}
3.6
39.6
6
68.4
1.2
0
3
18
15
18
9
30
Z[8, 1, 2]
{3, 0, 24/5}
36
(ii) Chiral mixed states
(a) [8, 1, 0](R1 , R2 ) ( , (R0)

)

(m a)
2
.
R=2
2 m + (p a)
mR

2

p
p
a
= |R | = 2m 1 +
+ 2 2 .
2
2
The corresponding eigenvectors can be found by diagonalizing the matrix RR .
(A.1)
(A.2)
309
(15)
(15) 44
(b) [1, 2, 1](h 1 , h 2 , h 3 , h 4 ) [1, 2, 1](h1 , h2 , h3 , h4 ) (H2 , 2
, 2
, )
2
441
, (15)
, )
(H 1 , (15)
1
1
2
MH
+ 3( a)
3( + a)

3( + a)
0
(2M + 4(a + ))
0
.
H=
3( a) (2M + 4(a ))
0
2 3
0
2m + 6( a)

2 3
The above matrix is to be diagonalized after imposing the fine tuning condition Det H = 0
to keep one pair of doublets light.
,
1, 2 ](t1 , t2 , t3 , t4 , t5 ) [3, 1, 2 ](t1 , t2 , t3 , t4 , t5 ) (H 4
4
(c) [3,
(a) , (a) , R0 ,
3
3
(a)4
) (H4
, 4(R0)
, 4
)
,
, 4(a)
(R)
4(R+)
(a + p)
(p a)
2M
2M
4 2i
2i
MH
(p a)
T = (p + a)

2 2i
i
2 2i
0
4 2i
2M + 2p + 2a
2 2
0
2i
2 2
2m 2(a + p 4)
(iii) Mixed gauge chiral
(15)
44 / 2,
44((R+) / 2,
(a) [1, 1, 0](G1 , G2 , G3 , G4 , G5 , G6 ) (, (15) , (R0) , (R)
(R0) (15)
( 2
3 )/ 5)
i
i
m
0
6
0
2

0
m + 2a
2 2
i 32
i 32
6 22 m + (p + 2a)
i 3
i 3
G = 2
i
i 32
i 3
0
M + (p + 3a 6)
2
i
i 3
M + (p + 3a 6)
0
2 i 32
5g
2
5g
2
0
5g
2
5g
2
0
4
(a)4
2, 1 ](E 2 , E 3 , E 4 , E 5 ) [3, 2, 1 ](E2 , E3 , E4 , E5 ) ( ,

(b) [3,
, 2
3
3
4 1
(s) 2
4 , (s) , (a) ,
2 ) (
1 )
2
4
1
4
1
2(M + (a 3))
2 2i
2i
2i 2
2(m + (a ))
2 2
E =
2i
2 2
2(m )

ig 2
2g(a )
g 2( p )
ig 2
2g(a )
.
2g( p )
0
(A.3)
310

(15)
44 ,
(c) [1, 1, 2](F1 , F2 , F3 ) [1, 1, 2](F1 , F2 , F3 ) ( 44(R0) , (R) , (R) ) ((R0)
(15)
(R+) , (R+) )
2(M + (p + 3a))
F =
2i 3
g 2
2i 3
2(m + (p + 2a))
24ig
g 2
24ig ) .
0
(A.4)
(R0)
1, 4 ](J1 , J2 , J3 , J4 ) [3, 1, 4 ](J1 , J2 , J3 , J4 ) ( 4

(d) [3,
, 4 )
(R) , 4 , 4
3
3
4
( 4(R+)
, 4 , (R0)
, 4 )
ig 2
2(M + (a + p 2))
2
2 2
2
2(m + a)
2 2
2ig 2a
J =
.
2 2
2 2
2(m + (a + p))
4ig
ig 2
2 2iga
4ig
0
(A.5)
(s)4
(e) [3, 2, 53 ](X 1 , X 2 , X 3 ) [3, 2, 53 ](X1 , X2 , X3 ) ( 1

(a)
, 2 )
4
2 4
2(m + (a + ))
X=
2 2
2g(a + )
2 2
2(m + )
2g( + p )
2g(a + )
2g( + p ) .
0
(a)4
, 1
(s)
, 1 ) (4
,
2
(A.6)
Appendix B. SU(5) U (1) reassembly crosscheck

Given the complexity of the spectra and couplings derived here it would be useful to
have a method of cross checking the internal consistency of our results. A stringent check
is provided by verifying that at special values of the vevs, i.e.,
p = a = ,
(B.1)
where the unbroken symmetry includes SU(5) the MSSM labeled mass spectra and couplings given in Appendix A do indeed reassemble into SU(5) invariant form. For the mass
spectra this is fairly straightforward to check and is reported explicitly below. A similar
calculation [32] for the superpotential couplings is much more tedious but furnishes an
SO(10) SU(5) U (1) analog of the SO(10)-PS Clebsches reported here.
The decomposition of the chiral multiplets of the MSGUT into SU(5) U (1) multiplets and of those into MSSM multiplets (named as per the alphabetic convention of
Appendix A) is given below. The only complication is that certain MSSM multiplet types
occur in several copies and (orthogonal) mixtures of these are present in the different SU(5)
mutiplets. Thus, for instance, the 210 contains a 24 and a 75 of SU(5) both of which contain mixtures of the G123 multiplets R1 (8, 1, 0) and R2 (8, 1, 0). These mixtures must be
orthogonal and must be precisely the eigenstates of the mass matrices in this G123 sector
which have the same masses as the rest of the G123 submultiplet sets within the 24-plet and
311
75-plets as two wholes. The fact that this follows in every case from our results appears to
confirm their reliability. The decompositions we need are
H = 10 = 51 + 5 1 ,
= 126 = 15 (G4 ) + 5 1 + 103 + 153 + 451 + 501 ,

1, 2 ,
51 = h 3 (1, 2, 1) + t3,4 3,
3

4
1
103 = F1 (1, 1, 2) + J1 3, 1,
+ E2 3, 2,
,
3
3

2, 1 + N 6,
1, 4 ,
153 = O(1, 3, 2) + E 1 3,
3
3

2
2
2, 7
1, 8 + D 1 3,
451 = h3 (1, 2, 1) + t3 3, 1,
+ P 3, 3,
+ K 3,
3
3
3
3

2
1,
+ C1 (8, 2, 1),
+ L 6,
3

1, 2 + D2 3, 2, 7 + W 6, 3, 2
501 = A(1, 1, 4) + t3,4 3,
3
3
3

8
1,
+ M 6,
+ C 2 (8, 2, 1),
3
= 126 = 15 (G5 ) + 51 + 103 + 153 + 451 + 501 ,

2
51 = h2 (1, 2, 1) + t2,4 3, 1, ,
3

4
2, 1 ,
103 = F1 (1, 1, 2) + J1 3, 1,
+ E 2 3,
3
3

1
4
+ N 6, 1, ,
153 = O(1,
3, 2) + E1 3, 2,
3
3

2
3, 2 + K 3, 1, 8 + D1 3, 2, 7
451 = h 2 (1, 2, 1) + t2 3, 1,
+ P 3,
3
3
3
3

2
+ C 1 (8, 2, 1),
+ L 6, 1,
3

2, 7 + W 6,
3, 2
1, 4) + t2,4 3, 1, 2 + D 2 3,
501 = A(1,
3
3
3

8
+ C2 (8, 2, 1),
+ M 6, 1,
3
= 210 = 10 + 54 + 5 4 + 102 + 102 + 240 + 402 + 402 + 750 ,
10 = G1,2,3 ,

2
54 = h4 (1, 2, 1) + t5 3, 1, ,
3

2
1,
5 4 = h 4 (1, 2, 1) + t5 3,
,
3
312

1, 4 + E3,4 3, 2, 1 ,
102 = F2 (1, 1, 2) + J2,3 3,
3
3

4
2, 1 ,
+ E 3,4 3,
102 = F2 (1, 1, 2) + J2,3 3, 1,
3
3

5
2, 5
240 = (1, 1, 0)G1,2,3 + S(1, 3, 0) + X1,2 3, 2,
+ X 1,2 3,
3
3
+ R1,2 (8, 1, 0),

1
1, 4 + U 3,
3, 4
+ J2,3 3,
402 = V (1, 2, 3) + E3,4 3, 2,
3
3
3

1
2,
+ Z(8, 1, 2) + Y 6,
,
3

1
4
4
+ J2,3 3, 1,
+ U 3, 3,
402 = V (1, 2, 3) + E 3,4 3, 2,
3
3
3

1
1, 2) + Y 6, 2, ,
+ Z(8,
3

10
1, 10 + X1,2 3, 2, 5
75 = (1, 1, 0)G1,2,3 + I 3, 1,
+ I 3,
3
3
3

5
5
5
2,
2,
+ B 6, 2,
+ B 6,
+ R1,2 (8, 1, 0)
+ X 1,2 3,
3
3
3
+ Q(8, 3, 0).
(B.2)
If we insert a = = p in the mass matrices of Appendix A we find that, after diagonalizing the mass matrices of the submultiplets that mix, the resultant spectra group
precisely as indicated by the decompositions above with all the subreps of a given SU(5)
irrep obtaining the same mass. One obtains the SU(5) invariant mass terms
2(M + 10p)1 1 + 2(M + 4p)5 5 + 2(M 2p)50 50
+ 2(M + 4p)10 10 + 2(M + 2p)15 15 + 2M45 45
+ MH 5 H 5H + (m + 6p)(1 )2 + 2(m + 6p)5 5 + 2(m + 3p)10 10
+ (m + p)(24 )2 + 2m40 40 + (m 2p)(75 )2

+ 2 3 (5 5 + 10 10 ) + (5 5 + 10 10 )
+ 2 3p( 5 5H + 5 H 5 ) + 2i 5( 1 1 1 1 )
+ 5 5H + 5 H 5 ,
(B.3)
where every SU(5) invariant has been normalized so that the individual G123 subrep
masses can be read off directly from the coefficient of the invariant for complex SU(5)
representations which pair into Dirac supermultiplets and is 2 times the coefficient for
the real representations which remain unpaired Majorana/chiral supermultiplets. For a =
= p, = = 0 the 20 Goldstone supermultiplets G, J , J, F , F , E, E of the coset
SO(10)/SU(5) U (1) remain heavy as they should since they are eaten in the spontaneous breaking SO(10) SU(5) U (1) while the 12 fields in the {X3 , X 3 } multiplets
313
lose their mass terms with {X1,2 , X 1,2 } since they form part of the unbroken SU(5) gauge
supermultiplet. When a = = p, i.e., for flipped SU(5), the roles of the {X3 , X 3 } and
{E5 , E 5 } gauge multiplets are interchanged, with the Es remaining massless and the Xs
becoming heavy, so that one obtains the SU(5) invariant groupings corresponding to the
flipped SU(5) U (1) SO(10) embedding. Note that this successful SU(5) reassembly
is a much more fine-grained consistency test than any overall trace or determinant test.
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]
C.S. Aulakh, R.N. Mohapatra, Phys. Rev. D 28 (1983) 217.

T.E. Clark, T.K. Kuo, N. Nakagawa, Phys. Lett. B 115 (1982) 26.
D.G. Lee, Phys. Rev. D 49 (1995) 1417.
C.S. Aulakh, B. Bajc, A. Melfo, G. Senjanovic, F. Vissani, Phys. Lett. B 588 (2004) 196, hep-ph/0306242.
K.S. Babu, R.N. Mohapatra, Phys. Rev. Lett. 70 (1993) 2845.
B. Bajc, G. Senjanovic, F. Vissani, Phys. Rev. Lett. 90 (2003) 051802, hep-ph/0210207.
H.S. Goh, R.N. Mohapatra, S.P. Ng, Phys. Lett. B 570 (2003) 215;
H.S. Goh, R.N. Mohapatra, S.P. Ng, Phys. Rev. D 68 (2003) 115008.
C.S. Aulakh, Truly minimal unification: asymptotically strong panacea?, hep-ph/0207150.
C.S. Aulakh, Taming asymptotic strength, hep-ph/0210337.
X.G. He, S. Meljanac, Phys. Rev. D 41 (1990) 1620.
C.S. Aulakh, A. Girdhar, SO(10) a la PatiSalam, hep-ph/0204097, Int. J. Mod. Phys. A, in press.
K.S. Babu, J.C. Pati, F. Wilczek, Phys. Lett. B 423 (1998) 337, hep-ph/9712307.
B. Bajc, A. Melfo, G. Senjanovic, F. Vissani, Phys. Rev. D 70 (2004) 035007, hep-ph/0402122.
T. Fukuyama, A. Ilakovac, T. Kikuchi, S. Mejanac, N. Okada, hep-ph/0401213.
T. Fukuyama, A. Ilakovac, T. Kikuchi, S. Mejanac, N. Okada, hep-ph/0401213, version 2, April 2004.
M. Gell-Mann, P. Ramond, R. Slansky, in: P. van Niewenhuizen, D.Z. Freedman (Eds.), Supergravity, NorthHolland, Amsterdam, 1979;
T. Yanagida, in: O. Sawada, A. Sugamoto (Eds.), Proceedings of Workshop on Unified Theory and Baryon
number in the Universe, KEK, 1979;
R.N. Mohapatra, G. Senjanovic, Phys. Rev. Lett. 44 (1980) 912.
R.N. Mohapatra, G. Senjanovic, Phys. Rev. D 23 (1981) 165;
G. Lazarides, Q. Shafi, C. Wetterich, Nucl. Phys. B 181 (1981) 287.
V.V. Dixit, M. Sher, Futility of high precision SO(10) calculations, Phys. Rev. D 40 (1989) 3765.
D.-G. Lee, R.N. Mohapatra, Phys. Rev. D 51 (1995) 1353, hep-ph/9406328.
C.S. Aulakh, B. Bajc, A. Melfo, A. Rain, G. Senjanovic, Nucl. Phys. B 597 (2001) 89, hep-ph/0004031.
S. Weinberg, Phys. Lett. B 91 (1980) 51.
L.J. Hall, Nucl. Phys. B 178 (1981) 75.
H. Georgi, H.R. Quinn, S. Weinberg, Phys. Rev. Lett. 33 (1974) 451.
W. Marciano, G. Senjanovic, Phys. Rev. D 25 (1982) 3092.
M.B. Einhorn, D.R.T. Jones, Nucl. Phys. B 196 (1982) 475.
U. Amaldi, W. de Boer, H. Furstenau, Phys. Lett. B 260 (1991) 447.
C.S. Aulakh, B. Bajc, A. Melfo, G. Senjanovic, F. Vissani, in preparation.
See, e.g., S.P. Martin, M.T. Vaughn, Phys. Rev. D 50 (1994) 2282, hep-ph/9311340.
See, e.g., K.S. Babu, S.M. Barr, Phys. Rev. D 51 (1995) 2463, hep-ph/9409285;
S.M. Barr, S. Raby, Phys. Rev. Lett. 79 (1997) 4748, hep-ph/9705366.
C.S. Aulakh, ICTP seminar, November 2003.
C.S. Aulakh, K. Benakli, G. Senjanovic, Phys. Rev. Lett. 79 (1997) 2188, hep-ph/9703434;
C.S. Aulakh, B. Bajc, A. Melfo, A. Rain, G. Senjanovic, Phys. Lett. B 460 (1999) 325, hep-ph/9904352;
C.S. Aulakh, A. Melfo, A. Rain, G. Senjanovic, Phys. Lett. B 459 (1999) 557, hep-ph/9902409;
C.S. Aulakh, A. Melfo, G. Senjanovic, Phys. Rev. D 57 (1998) 41744178, hep-ph/9707256.
C.S. Aulakh, S. Kumar, in preparation.
C.S. Aulakh, hep-ph/0501025.
Five-brane dynamics and inflation in heterotic

M-theory
Evgeny I. Buchbinder
School of Natural Sciences, Institute for Advanced Study, Einstein Drive, Princeton, NJ 08540, USA
Abstract
Generic heterotic M-theory compactifications contain five-branes wrapping non-isolated genus
zero or higher genus curves in a CalabiYau threefold. Non-perturbative superpotentials do not depend on moduli of such five-branes. We show that fluxes and non-perturbative effects can stabilize
them in a non-supersymmetric AdS vacuum. We also show that these five-branes can be stabilized in
a dS vacuum, if we modify the supergravity potential energy by FayetIliopoulos terms. This allows
us to stabilize all heterotic M-theory moduli in a dS vacuum in the most general compactification
scenarios. In addition, we demonstrate that, by this modification, one can create an inflationary potential. The inflationary phase is represented by a five-brane approaching the visible brane. We give
a qualitative argument how extra states becoming light, when the five-brane comes too close, can
terminate inflation. Eventually, the five-brane hits the visible brane and disappears through a small
instanton transition. The post-inflationary system of moduli has simpler stability properties. It can be
stabilized in a dS vacuum with a small cosmological constant.
PACS: 11.25.Mj
1. Introduction
Recently, there has been a considerable interest in cosmological aspects of string theory
(see [1] for a recent review). In [26], several methods of producing de Sitter (dS) vacua
E-mail address: evgeny@sns.ias.edu (E.I. Buchbinder).
doi:10.1016/j.nuclphysb.2005.01.015
E.I. Buchbinder / Nuclear Physics B 711 (2005) 314344
315
in string compactifications were presented. In [2,3,5], it was suggested that various corrections to the supergravity potential energy can raise a supersymmetric anti-de Sitter (AdS)
vacuum to a metastable dS vacuum. In [4], based on the earlier work [7], a dS vacuum was
created by balancing various exponential superpotentials and in [6], it was argued that a
dS vacuum can be created by taking into account higher order corrections to the moduli
Kahler potential. In addition, in [8,9], it was studied how effects of gravity and quantum
particle production could trap moduli at enhanced symmetry points. Furthermore, a substantial progress has been achieved towards inflation in string theory [1030]. In these
models, inflation was studied within the context of D-branes. Under certain conditions the
D-brane modulus can be treated as an inflaton. However, all these models usually have two
common problem. Inflation is often realized in compactification or brane world scenarios
which do not correspond to realistic four-dimensional physics. The other problem is that,
in addition to the inflaton, there are, usually, other moduli whose stabilization could be a
problem and whose presence can violate the slow roll conditions.
In this paper, we would like to explore the possibility of creating an inflationary potential within the framework of strongly coupled heterotic string theory, or heterotic M-theory
[3133]. Such compactifications have a lot of attractive phenomenological features (see
[34] for a review on phenomenological aspects of M-theory). Various GUT- and Standard
Model-like theories were obtained from heterotic compactifications on CalabiYau threefold [3538]. For example, in [38], vector bundles on CalabiYau manifolds with Z3 Z3
homotopy group were constructed. A compactification on such a manifold can lead to the
Standard Model with suppressed nucleon decay. The actual particle spectrum in such theories was studied in [3941]. One more attractive feature of such compactifications is that
it is possible to stabilize moduli in a phenomenologically acceptable range [5,42]. The
set of moduli considered in [5,42] was very general. Nevertheless, it was not complete.
In [5,42], it was assumed that the CalabiYau threefold had enough isolated genus zero
curves to stabilize all h1,1 moduli. It was also assumed that the five-branes in the bulk
wrapped only isolated genus zero curves. Even though such compactifications can certainly exist, a generic compactification with h1,1 greater than one contains various, not
necessarily isolated genus zero, cycles as well five-branes wrapped on them. In this case, it
is quite possible that not all h1,1 moduli can be stabilized by methods presented in [5,42].
The moduli of a five-brane wrapped on a non-isolated genus zero curve or a higher genus
curve cannot be stabilized by methods of [5,42] either. In this paper, we add these new
moduli. We show that this new additional h1,1 moduli can be stabilized in a supersymmetric AdS minimum by the slight modification of ideas of [5,42]. The five-brane moduli
cannot be stabilized this way. Surprisingly, we find that they can be stabilized in a nonsupersymmetric AdS minimum. Of course, what we really mean by this is that the system
of moduli containing the moduli of this new five-brane admits a non-supersymmetric AdS
vacuum. However, the potential energy has one more minimum when the five-brane coincides with the visible brane. A heterotic M-theory vacuum can contain several five-branes
wrapped on non-isolated genus zero or higher genus curves. Those which are located relatively far away from the visible brane will be stabilized. On the other hand, those which
are located close enough to the visible brane will roll towards it and, eventually, collide
with it. We show that these five-branes can be stabilized as well by balancing the supergravity potential energy against the FayetIliopoulos terms [43] induced by an anomalous
316
U (1) gauge group in the hidden sector. This shows that the most general set of heterotic
M-theory moduli can be stabilized. Furthermore, the cosmological constant can be positive and fine tuned to be very small. By balancing the supergravity potential energy against
FayetIliopoulos terms, it is also possible to create a positive potential satisfying the slow
roll conditions and treat the five-brane translational modulus as an inflaton. Inflation takes
place when the five-brane approaches the visible brane. However, this potential has one
negative feature. It has a vanishing first derivative when the five-brane coincides with the
visible brane. This means that it takes infinite time for the branes to collide. This also means
that the primordial fluctuations will become infinite. On the other hand, at very short distances, one cannot trust the low-energy field theory because new states are expected to
become massless. At the present time, physics at short distances in heterotic M-theory is
not known. Nevertheless, we present an argument how these new state can terminate inflation before the fluctuations became too big. Once the five-brane hits the visible brane, it
gets dissolved into it and turns into new moduli of the vector bundle, so-called transition
moduli, studied in [44,45]. This process is called small instanton transition [4649]. Thus,
the post-inflationary phase does not have the inflaton but has extra moduli of the vector
bundle. These moduli are easier to stabilize [42]. Therefore, the new system of moduli can
be stabilized, whereas during inflation this was not the case. In addition, we argue that, after
a small instanton transition, generically, the cosmological constant changes. It is possible
to decrease the cosmological constant and, by fine tuning, make it consistent with observations. Let us point out that, though, besides the inflaton, there are various other moduli
during inflation, they are all taken into account. The potential energy has a minimum in
all the other directions. Therefore, dynamically, one expects that all these moduli will roll
very fast in their minimum leaving the five-brane to roll slowly towards the visible brane.
This paper is organized as follows. In Section 2, we discuss the system of moduli in
compactifications with h1,1 greater than one. The reason is twofold. First, we would like to
obtain more general results on moduli stabilization. In particular, we would like to stabilize
the h1,1 moduli that do not have a non-perturbative superpotentials and, hence, cannot
be stabilized by methods of [5,42]. Second, before we begin to study potentials for the
five-brane, whose stability properties are more complicated, it is important to understand
how the remaining moduli of the system are stabilized. The system of (complex) moduli
considered in this section includes the complex structure moduli, the volume modulus, two
h1,1 moduli and the moduli of the five-brane wrapped on an isolated genus zero cycle.
One of the two h1,1 moduli is assumed to be associated with an isolated genus zero curve
and, hence, has a non-perturbative superpotential. The other one is associated with a nonisolated genus zero curve or a higher genus curve. The non-perturbative superpotential does
not depend on this modulus. By the slight modification of ideas of [5,42], we show that this
system can be stabilized in a supersymmetric AdS vacuum. The crucial moment is that, if
h1,1 is greater that one, it is possible to choose one of the contributions to the tension of
the hidden brane to be positive without having the gauge coupling constant stronger in the
visible sector. The system of moduli can be supplemented by vector bundle moduli [45].
They can be stabilized as well [42]. For simplicity, we will ignore them. In Section 3, we
add one more five-brane to the system. This five-brane is wrapped on a non-isolated genus
zero or higher genus curve. Approximately, we can treat the rest of the moduli fixed and
consider an effective potential for the remaining five-brane modulus. This potential is very
317
difficult to analyze analytically. A graphical analysis shows that, generically, it has a nonsupersymmetric AdS minimum. Nevertheless, if a five-brane was originally located close
to the visible brane, it will roll towards it. In the rest of the paper, we concentrate only on
dynamics of rolling five-branes. We modify this effective potential with FayetIliopoulos
terms and show that addition of a FayetIliopoulos term in the hidden sector can stabilize
a rolling five-brane. We also show that the cosmological constant in such a vacuum can
be positive and small. The results from Sections 2 and 3 provide stabilization of heterotic
moduli in the most general set-up in a vacuum with a positive cosmological constant. They
also indicate that it is conceivable to obtain two distinct dS vacua. One of them is the
lift of the non-supersymmetric AdS minimum. The other one is created by addition of a
FayetIliopoulos term in the hidden sector. In Section 4, we argue that, by balancing the
supergravity potential energy and FayetIliopoulos terms, it is also possible to construct
a potential with inflationary properties. One of the slow roll parameters turns out to be
naturally much less than one. The other one can be (not necessarily fine) tuned to be much
less than one. We also discuss the amount of inflation and primordial fluctuations. In the
last subsection of Section 4, we discuss how the system can escape from inflation at very
short distances. We give a qualitative argument how the appearance of new light states in
the field theory can provide such an escape. In Section 5, we discuss the post-inflationary
phase. After inflation, the five-brane hits the visible brane and disappears through a small
instanton transition. The new system of moduli does not contain the five-brane but has
extra vector bundle moduli. Unlike the five-brane modulus, these moduli can be stabilized.
We show that the new system of moduli can be stabilized in a vacuum with a positive
cosmological constant which can be fine tuned to be very small.
2. Supersymmetric AdS vacua in models with h1,1 > 1
2.1. The system of moduli
In this paper, we work in the context of strongly coupled heterotic string theory [31,32]
compactified on a CalabiYau threefold [33,50]. To one of the orbifold fixed planes we will
refer as to the visible brane (or the visible sector). To the other one we will refer as to the
hidden brane (or the hidden sector). Such compactifications also allow five-branes wrapped
on holomorphic cycles in the CalabiYau manifold and parallel to the orbifold fixed planes.
Moduli stabilization in this theory was performed in a relatively general setting in [5,42].
Nevertheless, it was assumed in [5,42] that the CalabiYau manifold has enough isolated
genus zero curves to stabilize all the h1,1 and five-brane moduli. In this paper, we would
like to consider a more complicated set-up when the CalabiYau threefold has two-cycles,
one represented by isolated genus zero curve and the other one by curves of a different type.
They could be either non-isolated genus zero curves or curves of a higher genus. In both of
these two cases, no non-perturbative superpotential for the corresponding h1,1 modulus can
318
be generated by string or, more precisely, open membrane instantons [51,52].1 This also
means that one cannot generate a non-perturbative superpotential for moduli of a five-brane
wrapped on such a cycle. However, compactifications on a CalabiYau manifold with h1,1
represented only by isolated genus zero cycles are, of course, very restrictive. A generic
compactification scenario involves a CalabiYau manifold with non-isolated genus zero or
higher genus cycles and five-branes wrapped on such cycles. As an example, consider a
CalabiYau manifold elliptically fibered over the Hirzebruch surface Fr , r = 0, 1, . . . . The
Hirzebruch surface Fr is a P1 bundle over P1 . We denote the class of the base of this bundle
by S and the class of the fiber by E. These CalabiYau threefolds are simply connected
and admit a (generically unique) global holomorphic section which we denote by . For
such a manifold, generically, we have
h1,1 = 3
(2.1)
and the basis of curves can be chosen to be

S,
E,
F.
(2.2)
Here is the projection map from the threefold onto the base Fr and F is the class of the
elliptic fiber. The curves S and E have genus zero. The curve F has genus
one. The curve S has a self-intersection r and, thus, is an isolated genus zero
curve for r > 0. It is a non-isolated genus zero curve for r = 0. The curve S has a
self-intersection zero for any r and, thus, is non-isolated genus zero curve. Therefore, it is
important to stabilize the h1,1 moduli corresponding to non-isolated genus zero or higher
genus curves. It is also important to understand whether or not it is possible to stabilize fivebranes wrapped on such cycles. In this paper, for simplicity, we consider the case h1,1 = 2.
We will assume that there is one isolated genus zero curve and one curve of a different
type. The generalization to the case involving many curves isolated curves of various types
is conceptually straightforward but technically more difficult.
At this point, we would like to make a remark. It may happen that the pullback of
more than one harmonic form I onto a given isolated curve is non-zero. As a result,
the non-perturbative superpotential associated with this isolated curve may depend on the
linear combination of more than one h1,1 modulus. In particular, it may depend on all
h1,1 moduli. In this case, all h1,1 moduli can be stabilized by methods presented in [42].
However, one might expect that, generically, there can be h1,1 moduli of two sorts, those
that appear in the non-perturbative superpotential and those which do not. In this paper, we
simply assume that our compactification has one modulus of each sort. In this section, we
will not consider five-branes wrapped on non-isolated genus zero or higher genus curves.
We will add such a five-brane in the next section.
The system of moduli that we would like to consider in this section includes the following complex moduli
S,
T 1,
T 2,
Y,
Z .
(2.3)
1 The statement that strings on non-isolated genus zero curves do not contribute to the non-perturbative su-
perpotential was conjectured by Witten [51]. The author is very grateful to Edward Witten for discussions on this
issue.
319
The modulus S is related to the volume of the CalabiYau manifold

S = V + i,
(2.4)
where is the axion. The moduli T 1 and T 2 are the h1,1 moduli. They are defined as
follows [50,53]
T I = RbI + ip I ,
I = 1, 2,
(2.5)
bI
where R is the size of the eleventh dimension, are the Kahler moduli of the CalabiYau
threefold and p I s come from the components of the M-theory three-form C along the
interval and the CalabiYau manifold. The moduli bI are not all independent. They satisfy
the constraint
2
dI J K bI bJ bK = 6,
(2.6)
I,J,K=1
where coefficients dI J K are the CalabiYau intersection numbers

1
dI J K =
I J K .
V
(2.7)
CY
The constraint (2.6) reduces the number of independent b-moduli by one. We will take
T 1 to correspond to the area of the isolated genus zero curve and T 2 to the area of the
remaining curve. Y is the modulus of the five-brane wrapped on the isolated genus zero
curve. In this case, there is only one five-brane modulus [54], whose real is the position of
the five-brane in the bulk

p1
,
Y=y +i a+
(2.8)
Rb1
where a is the axion arising from dualizing the three-form field strength propagating on
the five-brane world-volume. At last, by Z we denote the complex structure moduli. The
actual number of them is not relevant for us. A generic heterotic compactification contains
also instanton moduli [45]. Their stabilization was considered in [42]. In this section, for
simplicity, we will ignore them. They can be added and treated as in [42]. However, we
will come back to them in the last section. The moduli V , T 1 , T 2 and y are assumed to be
dimensionless normalized with respect to the following reference scales
1/6
vCY 1016 GeV,
()1 1014 1015 GeV.
(2.9)
In order to obtain the four-dimensional coupling constants in the correct phenomenological

range [33,55], the corresponding moduli should be stabilized at (or be slowly rolling near)
the values
V 1,
R 1.
(2.10)
The Kahler potential for this system is as follows [50,56,57]

K
= KZ + KS,T 1 ,T 2 ,Y ,
2
MPl
(2.11)
320
where

KZ = ln i ,
(2.12)
and

ln dI J K T I + T I T J + T J T K + T K
KS,T ,Y = ln(S + S)
2
(Y + Y)
+ 25
.
1 + T 1 )
(S + S)(T
(2.13)
Here MPl is the four-dimensional Planck scale and 5 is given by

5 =
T5 v5 ()2
,
2
MPl
(2.14)
where v5 is the area of the cycle on which the five-brane is wrapped and T5 is

1 2/3
T5 = (2)1/3
,
2
211
(2.15)
with 11 being the eleven-dimensional gravitational coupling constant. It is related to the

four-dimensional Planck mass as
vCY
2
=
.
11
(2.16)
2
MPl
Evaluating 5 by using (2.16) and (2.9) gives
v5
5 1/3 .
vCY
(2.17)
Generically this coefficient is of order one.

The superpotential for this system consists of three different contributions
W = Wf Wg Wnp .
Wf is the flux-induced superpotential [5860]

2
MPl
Wf =
dx 11 G ,
vCY
(2.18)
(2.19)
CY
where G is the M-theory four-form flux. The order of magnitude of Wf was estimated in
3 . In fact, this is flexible. The super[42] and was found to be, generically, of order 108 MPl
potential Wf may receive certain higher order corrections from ChernSimons invariants.
In [61] it was argued that these ChernSimons invariants can reduce the order of magnitude
of Wf .
By Wg we denote the superpotential induced by a gaugino condensate in the hidden
sector [6265]. A non-vanishing gaugino condensate has important phenomenological consequences. Among other things, it is responsible for supersymmetry breaking in the hidden
sector. When that symmetry breaking is transported to the observable brane, it leads to
321
soft supersymmetry breaking terms for the gravitino, gaugino and matter fields [6669].
See [70] for a good review on gaugino condensation in string theory. This superpotential
has the following structure

Y2
(2) 1
(2) 2
S + 1 T + 2 T 1 .
T
3
Wg = hMPl
exp
(2.20)
The order of magnitude of h is approximately 106 [64]. The coefficient is related to the
coefficient b0 of the one-loop beta-function and is given by
=
6
.
b0 GUT
(2.21)
For example, for the E8 gauge group 5. The coefficients I(2) represents the tension
(up to the minus sign) of the hidden brane measured with respect to the Kahler form I
(2)
I =

11 2/3
1
I tr F (2) F (2) tr R R ,
16vCY 4
2
(2.22)
CY
where F (2) is the curvature of the gauge bundle on the hidden brane. Similarly, the coefficient is the tension of the five-brane. It is given by [71]

2 2 11 2/3
= 2/3
1 W,
4
v
CY
(2.23)
CY
where W is the four-form Poincar dual to the holomorphic curve on which the five-brane
(2)
is wrapped. Generically both I and are of order one. In fact, from Eqs. (2.14), (2.15)
and (2.23) it follows that
5 .
(2.24)
Let us note the following fact which will be crucial for stabilization of T 2 . If h1,1 = 1,
apparently, it is important to have (2) positive (and, correspondingly, the tension negative)
in the hidden sector. This, in particular, happens when the bundle on the hidden brane is
trivial. The reason is that the quantity

(2)
Y2
Re S
I T I + 1
T
(2.25)
represents the inverse square of the gauge coupling constant in the hidden sector
Furthermore, the quantity

Re S

I

(1)
I T I
+T
Y2
1 1
T
1
.
2
ghidden

(2.26)
322
represents the inverse square of the gauge coupling constant in the visible sector
Here (1) is the tension (up to the minus sign) of the visible brane

11 2/3
1
(1)
(1)
(1)
=
tr F F tr R R ,
16vCY 4
2
1
.
2
gvisible
(2.27)
CY
where F (1) is the curvature of the gauge bundle on the visible brane. If, for example,
h1,1 = 1 and there are no five-branes, we have

1
= Re S (2) T
(2.28)
2
ghidden
and
1
2
gvisible

= Re S (1) T .
(2.29)
The anomaly cancellation condition in the absence of five-branes,

c2 (Vvisible ) + c2 (Vhidden ) = c2 (T X),
(2.30)
(2) = (1) .
(2.31)
sets
Now it is clear that if (2) < 0, the gauge coupling constant in the hidden sector is weaker
that the gauge coupling constant in the visible sector and the whole assumption about the
gaugino condensation in the hidden sector breaks down. It is unlikely that this statement
changes when five-branes are included. However, when h(1,1) is greater than zero, there
(2)
is nothing wrong with having some I s negative. It is still possible to keep the gauge
coupling constant stronger in the hidden sector. We will assume that
(2)
(2.32)
(2)
(2.33)
1 > 0
and
2 < 0.
It is important to note that the quantity given by Eq. (2.25) must be positive. This means
that the superpotential (2.20) cannot be trusted for large values of the interval size R. One
should expect that higher order corrections to the combination (2.25) will make the gauge
coupling constant 2 1 well defined for large values of R. Partial support for this comes
ghidden
from [72,73].
The last contribution to the superpotential that we have to discuss is the non-perturbative
superpotential Wnp [52,7482]. Such a superpotential is induced by open membranes
wrapped on an isolated genus zero curve. Therefore, it depends on the h1,1 modulus T 1
and on the five-brane modulus Y. However, it does not depend on the h1,1 modulus T 2 .
The non-perturbative superpotential has three parts
Wnp = Wvh + Wv5 + W5h .
(2.34)
323
Wvh is induced by a membrane stretched between the visible and the hidden branes. It
behaves as
Wvh e T .
1
(2.35)
Wv5 is induced by a membrane stretched between the visible brane and the five-brane. It
behaves as
Wv5 e Y .
(2.36)
At last, W5h is induced by a membrane stretched between the five-brane and the hidden
brane. It behaves as
W5h e (T
1 Y)
(2.37)
The coefficient is given by [78,79]

1/3
1
= ()vi
,
2
211
(2.38)
where vi is the reference area of the isolated curve. Generically, is much bigger than
one. As in [5,42], we will assume that the five-brane is close to the hidden sector. It was
argued in [5,42] that only in this case it is possible to stabilize the size of the interval in
a phenomenologically acceptable range. Therefore, the contributions Wvh and Wv5 decay
very fast and we have
3
ae (T
Wnp = W5h = MPl
1 Y)
(2.39)
For concreteness we assume that the coefficient a 1.

2.2. Supersymmetric AdS vacua
In this subsection, we will argue that this system of moduli has an AdS minimum. The
consideration is, somewhat, similar to [5,42] and we will be relatively brief. Let us first
discuss the imaginary parts of the moduli. A consideration analogous to [42] shows that
the imaginary parts of T 1 and Y are stabilized at values
Im T 1
1
0,
Im Y 0.
(2.40)
(2)
The imaginary part of the linear combination S 2 T 2 is stabilized in such a way that the
superpotentials Wf and Wg are out of phase. Similarly, Wf and Wnp are also out phase.
We already took this into account in Eq. (2.18) by putting the minus sign in appropriate
places. Unfortunately, the superpotential of the form (2.18)(2.20) and (2.39) does not
allow us to stabilize the remaining linear combination of S and T 2 . It can be shown to be
a flat direction. This problem cannot be resolved even by considering a multiple gaugino
condensation in the hidden sector. Nevertheless, it is easy to realize that this problematic
linear combination can be stabilized by taking into account the higher order T -corrections
to the gauge coupling in the hidden sector. We will make a more detailed comment on it
later in this subsection. Therefore, stabilization of the remaining imaginary part does not
324
represent a conceptual problem. In the rest of the paper, we will concentrate only on the
real parts of the moduli ignoring their imaginary parts. Now let us consider the real parts of
the moduli and show that the system under study indeed has an AdS minimum satisfying
Dall fields W = 0,
(2.41)
where D is the Kahler covariant derivative. We will not distinguish between the superpotentials and their absolute values. First, we consider equations
DZ W = 0.
(2.42)
Assuming that
Wf Wg , Wnp
(2.43)
in the interesting regime, Eq. (2.43) can be written as

KZ
(2.44)
Wf = 0.
Z
In [42], it was shown that inequality (2.43) is indeed satisfied. In Eq. (2.44), all quantities
depend on the complex structure moduli only. We will assume that this equation fixes all
the complex structure moduli. Partial evidence that equations of the type (2.44) fix all the
complex structure moduli comes, for example, from [83]. The next equation to consider is
Z Wf +
DS W = 0.
(2.45)
By using Eqs. (2.13), (2.20) and (2.43), it can be written as

Wg = F1 Wf ,
(2.46)
where

y2
1
.
1 + 5
F1 =
2V
(Rb1 )2
(2.47)
By using Eqs. (2.13), (2.20), (2.39), (2.43) and (2.46),

DT 1 W = 0
can be rewritten as

(2)
Wnp =
1 +
(2.48)

y2
F
+
F
1
2 Wf ,
(Rb1 )2
(2.49)
where

3 I J d1I J bI bJ
5 y 2
F2 =
+
.
R I J K dI J K b I b J b K
V (Rb1 )2
(2.50)
Now let us consider

DT 2 W = 0.
(2.51)
T 2,
T2
Note that the non-perturbative superpotential (2.39) does not depend on

thus,
cannot be stabilized by the same mechanism as T 1 . By using Eqs. (2.13), (2.20), (2.43), we
325
obtain
Wg = F3 Wf ,
(2.52)
where

3 I J d2I J bI bJ
.
F3 = (2)
2 R I J K dI J K bI bJ bK
(2.53)
Eqs. (2.46) and (2.52) are consistent only if F1 and F3 are equal to each other. In particular,
(2)
they must have the same sign. This is possible only if 2 is negative. As we argued before,
this does not lead to any contradictions. Note, that F1 and F3 are both real. This is the
reason why only one linear combination of the imaginary parts of S and T 2 moduli can
be stabilized. On the other hand, if higher order T -corrections to the quantity (2.25) are
present, F3 is really complex and, hence, the imaginary parts of both S and T 2 moduli can
be stabilized. The last equation to consider is
DY W = 0.
(2.54)
By using Eqs. (2.13), (2.20), (2.39), (2.47) and (2.50), we obtain

y
y2
y
(2)
1 +
= 0.
F1 + F3 + 2 1 F1 + 25
1
2
(Rb )
Rb
V Rb1
(2.55)
Eqs. (2.46), (2.49), (2.52) and (2.55) are the four equations with four independent variables
V , R, y and one of two bI s. Equations of this type were analyzed in detail in [5,42] in the
case of only one h1,1 modulus. It was shown that they admit a solution with the following
properties
V is of order one;
R is of order one;
2
does not become imaginary;
The gauge coupling constant ghidden
The five-brane is close to the hidden brane (R y 0.1).
In this paper, we will not perform a detailed analysis. Let us just point out that in Eqs. (2.46)
and (2.49), the run-away moduli are stabilized by fluxes. Eqs. (2.46) and (2.52) lead to
equation
F1 = F 3 ,
(2.56)
(2)
2
which is well defined if

is negative. Eq. (2.55) is a purely algebraic equation. It is
possible to show that it admits a numeric solution with the right properties as in [5,42].
We will not give a numeric result in this paper. See [5,42] for a detailed analysis of similar
equations.
In this subsection, we have provided stabilization of moduli listed in (2.3). This list
includes the modulus T 2 , corresponding to the area of a non-isolated genus zero curve or
a curve of a higher genus. Stabilization of such a modulus differs from stabilization of the
modulus T 1 , corresponding to the area of an isolated genus zero curve. The crucial point
in stabilization of T 2 is that, in the case when h1,1 > 1, it possible to choose the coefficient
326
(2)
2 to be negative. It is not possible to do in the case when h1,1 = 1, because it would

follow that the gauge coupling in the hidden sector became weaker than in the visible
sector. This would not be consistent with the assumption about gaugino condensation in
the hidden sector.
The AdS vacuum constructed in this section can be raised to a metastable dS vacuum
along the lines of [5]. This can be achieved by either adding FayetIliopoulos terms to the
supergravity potential energy or by working within the context of E8 E 8 theory.
3. Addition of a five-brane and dS vacua

3.1. Effective potential for a five-brane modulus and non-supersymmetric AdS vacua
Now we would like to see what happens if we add a five-brane wrapped on a nonisolated genus zero curve or on a higher genus curve to the system of moduli considered
above. We will denote the complex five-brane modulus by X and its real part by x. The
Kahler potential (2.13) receives the contribution
K = 25
2
(X + X)
.
2 + T 2 )
(S + S)(T
(3.1)
The gaugino condensate superpotential gets modified and becomes

2
Wg Wg e
X2
T
(3.2)
The coefficients 5 and are given by expressions similar to Eqs. (2.14) and (2.23). Unfortunately, if a five-brane wraps a non-isolated cycle, one should expect other five-brane
moduli in addition to X. Such moduli have never been considered in the literature in detail.
Nevertheless, one should expect that the gaugino condensate superpotential (3.2) depends
on them. This might provide their stabilization. Thus, we will assume that these additional
moduli are fixed and not consider them in this paper. In principle, one can avoid this issue
by taking a five-brane wrapping an isolated higher genus curve. Let us first see if we can
stabilize X in an AdS vacuum. By using Eqs. (3.1) and (3.2), the equation
DX W = 0
can be written as
X
x
2 Wg + 5
Wf = 0.
T
V Rb2
It is easy to realize that the only solution for X is
X = 0.
(3.3)
(3.4)
(3.5)
The point x = 0 corresponds to the five-brane coinciding with the visible brane. Such a
vacuum is unstable in the sense that the five-brane will disappear through a small instanton
transition [44,4749] and turn into new vector bundle moduli.
As an approximation, we will assume that the presence of this extra five-brane will not
modify much the vacuum constructed in the previous section. As a result, we can talk about
327
the effective potential U (x) describing dynamics of the five-brane. In fact, it is possible
to show that the vacuum value of the moduli S, T I , Y receive corrections of order x 2 .
Therefore, for very small values of x, x 1, their vacuum values will not shift much. This
suggest that the effective potential U (x) is a decent approximation. Of course, in order to
describe the system exactly, one has to solve all the equations for moduli including the
equations for the imaginary parts. This is not possible to do analytically. However, it is
natural to argue that the qualitative behavior of this system will be captured assuming that
there is the effective potential U (x) with the rest of the moduli fixed along the lines of
the discussion in the previous section. Thus, we consider dynamics of one field X with the
Kahler potential
1
K(X)
2
= K0 + K1 (X + X)
2
4
MPl
(3.6)
and the superpotential

W (X) = W0 W1 e X ,
2
(3.7)
where K0 is a constant independent of X, K1 is given by

25
,
V Rb2
W0 is a constant of order fluxes,
K1 =
(3.8)
W0 Wf ,
(3.9)
W1 is approximately given by (see Eq. (2.46))

F1
F1
Wf =
W0

and the coefficient is given by
W1 =

.
T2
(3.10)
(3.11)
Without loss of generality, we can set Im T 2 = 0. Then

.
Rb2
The effective potential for the X modulus is given by
=
K(X)

2
3W (X)W (X)
,
(X)
U (X) = e MPl G1
W
DX W (X)DX
XX
(3.12)
(3.13)
where the Kahler covariant derivative is defined as usual

DX W (X) = X W (X) +
1
K(X)W (X).
2 X
MPl
(3.14)
As was argued before, the imaginary part of X can be stabilized by this potential. Therefore,
the potential U (X) can be treated as an effective potential for one real field x. We will
328
denote it U (x). From Eqs. (3.6) and (3.7), we obtain

2
2 2
U (x) = U0 eK1 x 3 + 2K1 x 2 1 + e x
,
(3.15)
2 , K and are given by Eqs. (3.8) and (3.12),

where U0 is a constant of order Wf2 /MPl
1
respectively, and is given by
2 F1
.
K1
(3.16)
Eq. (3.15) gives an effective potential U (x). Unfortunately, it is very difficult to analyze
this potential analytically. A graphical analysis shows that, generically, this potential has a
non-supersymmetric AdS vacuum for a non-zero value of x. The form of U (x) for various
choices of parameters is shown on Figs. 1 and 2. It is possible to adjust parameters in
such a way that the vacuum becomes dS. However, in this case, the parameter has to
Fig. 1. The graph of UU(x) for K1 = 3, = 3, = 1, = 10. There exists a non-supersymmetric AdS minimum.
0
Fig. 2. The graph of UU(x) for K1 = 5, = 2.5, = 2, = 10. There exists a non-supersymmetric AdS minimum.
0
329
be taken to be sufficiently greater than one, whereas Eqs. (3.16), (3.8) and (3.12) require
that be of order one. Therefore, for reasonable values of the parameters, the minimum is
always AdS. It is possible to adjust parameters so that x is less than the size of the interval,
which, as discussed in the previous section, can be stabilized at a value of order one. This
AdS vacuum can be raised to a metastable dS vacuum by methods discussed in [5]. This
demonstrates that the most general system of heterotic M-theory moduli can be stabilized
in a dS vacuum.
In the rest of the paper, we will be interested in dynamics of a five-brane in the regime
x 1. Heterotic M-theory vacua can contain several five-branes wrapped on non-isolated
genus zero or higher genus curves. We have just argued that those five-branes which are
located sufficiently far away from the visible brane can be stabilized. Now we would like to
understand the fate of the five-branes which are close to the visible sector. Such five-brane
will roll towards x = 0. The potential U (x) in this regime does not lead to any interesting
physics. It does not provide stabilization of x. It is also hard to imagine how to use it in
any cosmological framework. On the one hand, it is negative and, hence, cannot be used
for inflation. On the other hand, it does not satisfy conditions necessary for Ekpyrotic
cosmology [8486]. To make use of this potential, we will modify it by FayetIliopoulos
terms. Depending on relations among various coefficients, FayetIliopoulos terms can lead
to either stabilization of x or a potential with certain inflationary properties.
3.2. FayetIliopoulos terms and dS vacua
In both weakly and strongly coupled heterotic string models, there can be anomalous
U (1) gauge groups. They can arise in both the visible and the hidden sectors. The anomaly
is canceled by a four-dimensional version of the GreenSchwarz mechanism. This anomalous U (1) gives rise to the FayetIliopoulos term [43], which, in turn, gives rise to the
moduli effective potential of the form
b
,
(3.17)
V2
where b is a constant and g is the gauge coupling constant. In the context of the strongly
coupled heterotic string theory, the coefficient b was estimated in [5] and was found to be,
generically, of order
4 2
g
UD = MPl
b 1018 .
(3.18)
The potential UD depends on in what sector there appears an anomalous U (1). The reason
is that the coupling constants in the visible and the hidden sectors are different. They are
given by [65]
2
=
gvisible
g02
Re(S + 1(1) T 1 + 2(1) T 2 + (T 1
Y2
) + (T 2
T1
X2
))
T2
(3.19)
and
2
ghidden
=
g02
(2)
(2)
Re(S 1 T 1 2 T 2 + TY1 + TX2 )
(3.20)
330
where g0 is a moduli independent constant of order GUT . In [3,5], FayetIliopoulos potentials UD were used to raise AdS vacua to dS vacua. In this paper, we will be interested in
the x dependence of UD . If the anomalous U (1) appears in the hidden sector, the potential
UD (x) takes the form
UDvisible (x) =
B1
,
B2 x 2
(3.21)
whereas, if the anomalous U (1) appears in the hidden sector, the potential UD is
UDhidden (x) =
C1
,
C2 + x 2
(3.22)
where, B1 , B2 , C1 and C2 can be read off from Eqs. (3.17), (3.19) and (3.20) and is given
by Eq. (3.12).
We would like to modify our potential U (x) by UD (x). In this section, we take UD (x)
to be UDhidden (x). We will now show that the potential energy
U (x) = U (x) + UDhidden (x)
(3.23)
can provide stabilization of x in the regime x 1 in a dS vacuum. We should point out

that, if we modify U (x) by some other moduli dependent correction, it is not very obvious
that this correction will not destabilize other, additional to x, moduli. However, in [5], it
was shown that if the order of magnitude of UD is the same as (or less than) the order
of magnitude of U , it is possible to find a solution to equation d(U + UD ) = 0 fixing all
the moduli considered in the previous section. Therefore, it is still a decent approximation
to consider the effective potential U (x) assuming that all the remaining moduli are fixed.
Now note the following simple facts. Since x = 0 is the minimum of the function U , for
small x we have
U (x)
> 0.
(3.24)
x
On the other hand, from Eq. (3.22), it follows that
UDhidden (x)
< 0.
x
This means that it should be possible to find a solution to the equation
(3.25)
U (x)
(3.26)
=0
x
under mild assumptions. For x 1, the potential U (x) is governed by the quadratic function
U (x) = 3U0 + a2 U0 x 2 ,
where a2 is given by

a2 = K1 2(1 + )2 3 .
(3.27)
(3.28)
Using Eqs. (3.8), (3.12), (3.16) and (2.47), one can show that a2 is greater than zero for
any choice of the parameters. It is straightforward to solve Eq. (3.26) in this regime. The
approximate solution is
1/2

C1
1
C2
.
xmin
a 2 U0
331
(3.29)
It is possible to adjust the parameters so that xmin is real and much less than one. It is also
straightforward to show that, if the solution for xmin exists, it is always a minimum. The
simplest way to do it is prove that, if the solution (3.29) exists, then x = 0 is always a
maximum. Since xmin 1, the value of the cosmological constant is approximately given
by
3U0 +
C1
.
C2
(3.30)
It is obvious that can be of both signs. By fine-tuning it is possible to set

4
10120 MPl
(3.31)
which is consistent with observations. The form of the potential U (x) in the regime x 1
is shown on Fig. 3.
In Sections 2 and 3, we showed that the most general system of heterotic M-theory
moduli can be stabilized in a dS vacuum. In addition to moduli considered in [5], we
also provided stabilization for extra h1,1 moduli and an extra five-brane associated with
a non-isolated genus zero curve or with a higher genus curve. In the presence of such a
five-brane, the system of moduli can be stabilized by fluxes and non-perturbative effects in
a non-supersymmetric AdS vacuum which then can be lifted to a dS vacuum as in [5]. This
five-brane can also be stabilized by balancing the supergravity potential energy against a
FayetIliopoulos term induced by an anomalous U (1) gauge group in the hidden sector.
Thus, the potential energy U (x) might admit two dS vacua. One of the them is the lift
of the non-supersymmetric AdS vacuum. The other one can additionally arise for x 1,
though it did not existed in the absence of the FayetIliopoulos term.
Fig. 3. The graph of UU(x) in the regime x 1 for a2 = 2.95, = 1, U1 = 3.01, C2 = 1. There exists a dS
0
0
minimum.
332
4. The five-brane modulus as an inflaton

4.1. Constructing an inflationary potential
We begin this section with modifying the potential energy U (x) by the FayetIliopoulos
term UDvisible (x) given by (3.21). The first derivative of UDvisible (x) is positive, hence, the
potential
U (x) = U (x) + UDvisible (x)
(4.1)
does not have a non-trivial minimum for x 1 and x rolls towards x = 0. We will assume
that the potential U (x) is positive. The potential (4.1) has the following form

2
2 2
+
U (x) = U0 eK1 x 3 + 2K1 x 2 1 + e x
B1
.
B2 x 2
(4.2)
Let us recall that the coefficients , K1 , and are given by Eqs. (2.21), (3.8), (3.12) and
2 and B and B can be read off from Eqs. (3.17)
(3.16), U0 is a constant of order Wf2 /MPl
1
2
and (3.19). We assume that this potential is positive, that is,
B1
> 3U0 .
B2
(4.3)
Our goal is to examine whether this potential can satisfy the slow roll conditions required
by inflation. As in the previous section, we are interested in the regime x 1. Then we
can expand U (x) in powers of x. For our purposes, it is enough to keep only two leading
terms. We obtain
U (x) A0 + A2 x 2 ,
(4.4)
where
A0 = 3U0 +
B1
B2
(4.5)
and

B1
B1
A2 = K1 2(1 + )2 3 U0 + 2 = a2 + 2 .
B2
B2
(4.6)
In order to study the standard slow roll parameters (x) and (x), we have to canonically
normalize the kinetic energy. From the Kahler potential (3.6), it follows that we have to
redefine x as

2 x
x
(4.7)
.
2
K1 MPl
This new x is canonically normalized and has dimension one. The potential energy now
looks as follows
2A2 2
x .
U (x) = A0 +
2
K1 MPl
(4.8)
To have inflation as x rolls towards x = 0, the two parameters

M 2 U (x) 2
(x) = Pl
2
U (x)
333
(4.9)
and
2
(x) = MPl
U (x)
U (x)
(4.10)
have to be much less than one. From Eq. (4.8) we obtain

(x) =
2A22
2
A20 K12 MPl
x2.
(4.11)
Clearly, for x MPl , (x) is naturally much less than one. For the parameter (x) we have
(x) =
4A2
.
K1 A0
(4.12)
Therefore, we need to impose

4
A2 < A0 .
K1
(4.13)
Using Eqs. (4.5) and (4.6), this condition can be rewritten as

4(2(1 + )2 3)U0 +
3U0 +
4 B1
K1 B22
B1
B2
1.
(4.14)
The only way this can be fulfilled is when

B1
U0
B2
(4.15)
4
1.
K1 B2
(4.16)
and
Condition (4.15) is a relatively mild constraint. Using Eqs. (3.8), (3.11), (3.17) and (3.19),
condition (4.16) can be rewritten as
2V
Re(S
(1)
+ 1 T 1
(1)
+ 2 T 2
+ (T 1
Y2
) + T 2)
T1
1.
(4.17)
Unfortunately, it does not seem to be possible to fulfill this condition, at least in the context
of low-energy field theory. Inequality (4.17) requires that some of the tensions I(1) , or
be much greater than one. In this case, one cannot trust, even approximately, expressions
(3.19) and (3.20) for the coupling constants because they can be substantially modified by
higher order corrections [72,73]. On the other hand, Eq. (4.17) may make perfect sense
in the context of M-theory. However, we would like to stay within the context of lowenergy field theory. All we have to do to make the parameter (x) small is to decrease
334
the parameter A2 in Eq. (4.4). This, in fact, can easily be done. We just have to replace a
FayetIliopoulos term in the visible sector by a FayetIliopoulos term in the hidden sector.
Equivalently, we could just add the FayetIliopoulos term in the hidden sector to Eq. (4.1).
In both cases, addition of such a FayetIliopoulos term increases A0 and decreases A2 .
It is possible to (not necessarily fine) tune the parameter to be much less than one. Let
us consider it in slightly more detail. Assuming, for simplicity, that only a hidden sector
FayetIliopoulos term is present and using (3.22), we have
(x) =
4(2(1 + )2 3)U0
3U0 +
4 C1
K1 C22
C1
C2
1.
(4.18)
We can rewrite (4.18) as

(x) =
4(2(1 + )2 3
3
K1 C2 )U0
A0
4
K1 C2 A0
1.
(4.19)
Since the quantities U0 and A0 are, generically, of the same order of magnitude [5], inequality (4.19) is a relatively mild constraint. In the next subsection, we will show that the
parameter does not have to be fine tuned to be very small. As discussed in the previous
section, addition of a FayetIliopoulos term in the hidden sector can stabilize x. This happens if the numerator in (4.19) becomes negative. In this case, the point x = 0 becomes
a maximum and the potential U (x) acquires a minimum at a non-zero value of x. This
was studied in the previous section. In this section, we assume that the effect of such an
addition is to make the potential flat, rather than to produce a non-trivial minimum.
In this subsection, we showed that the five-brane effective potential, with various Fayet
Iliopoulos terms included, can satisfy the slow roll conditions
(x) 1,
(x) 1
(4.20)
necessary for inflation. Let us recall that the system of fields contains various other moduli,
in addition to x. The potential energy, by construction, has a minimum in these directions.
Therefore, dynamically, one expects that they will roll fast in the minimum, leaving the
modulus x to roll slowly. Since inequalities (4.20) are satisfied for x 1, the five-brane
modulus x can be viewed as an inflaton.
4.2. The amount of inflation and primordial fluctuations
In this subsection, we will consider the amount of inflation and primordial fluctuations.
The amount of inflation is defined by
af
N = ln ,
(4.21)
ai
where ai and af are the initial and final values of the expansion parameter. The evolution
of a and x can be found from the Friedmann equation

1 2
1
(x)
x
H2 =
(4.22)
+
U
2 2
3MPl
335
and the x-equation of motion

x + 3H x + U (x) = 0,
(4.23)
where
a
(4.24)
a
is the Hubble constant. Since during the period of inflation the kinetic energy is much less
than the potential energy, it follows from Eq. (4.25) that

1
A0
H
(4.25)
.
MPl 3
This gives
H=
a(t) ai e
A0
2 t
3MPl
(4.26)
Similarly, integrating Eq. (4.23) we find that

x(t) xi e
4A2
t
K1 MPl 3A0
(4.27)
From Eqs. (4.26) and (4.27) we obtain

N = ln
af
A0 K1 xi
1 xi
=
ln
= ln ,
ai
4A2
xf
xf
(4.28)
where Eq. (4.12) has been used. By xi and xf we denoted the initial and final positions of
the five-brane during inflation. Taking, as an example,
xi
104 ,
0.1,
(4.29)
xf
we get
N 80
(4.30)
which is consistent with observations. Primordial fluctuations are determined by the following quantity

1
1
A0
4 H 2 H
U (x)
2
H =
(4.31)
=
,
4 (x)
4 (x)
25 x
2
150 2 MPl
150 2 MPl
where, in our case, (x) is given by Eq. (4.11). Note that as x goes to zero, (x) goes to
2 goes to infinity. Therefore, it is important to terminate inflation before the
zero and H
fluctuations became too big. This will be discussed in the next subsection. Taking the order
of magnitude of A0 set by the fluxes (see Eq. (2.19) and discussion below it),
4
1018 ,
A0 MPl
(4.32)
(x) to be
(x) 1012 ,
(4.33)
336
corresponding, for example, to

xf
0.1,
105 ,
MPl
(4.34)
we obtain
2
1010 ,
H
(4.35)
which is consistent with measurements of the cmb anisotropy.

Thus, this model of inflation gives appropriate values for the amount of inflation and
primordial fluctuations. However, these results really make sense only if it is possible to
escape from inflation before the fluctuations became too big.
4.3. Escape from inflation
At very small values of x, we cannot really trust the potential U (x) because one should
expect extra light states to become light as we approach the singularity x = 0. At the
present time, the new physics at distances much less than the eleven-dimensional Planck
scale is not known. It may happen that these new states are string-like, rather than particles
[87]. In this subsection, we would like to give a qualitative argument how such new states
can terminate inflation. Let us emphasize that we cannot prove that this is the actual mechanism. We just would like to point out that the appearance of new physics at short distances
can help to terminate inflation. We will assume that the new states are particles and, in the
absence of fluxes and non-perturbative effects, the moduli space of heterotic M-theory is
describable by the superpotential
W = W(, X),
(4.36)
where the fields come from a membrane stretching between the visible brane and the
five-brane. These fields are expected to be charged under E8 . The moduli space, that is the
space of solutions of
dW = 0,
(4.37)
must consist of two branches. The first branch, the five-brane branch, is characterized by
a non-zero expectation value of the five-brane translational modulus x. In this branch,
the five-brane multiplet is massless, while the fields are massive and integrated out
form the low-energy field theory. The mass of the fields is proportional to x. The second branch, the instanton branch, is characterized by the vanishing expectation value of
x and coincides with the moduli space of transition moduli [44] of an instanton on our
CalabiYau threefold. This five-brane-instanton transition is called small instanton transition [44,4749]. The interpretation of the transition is the following. As the five-brane hits
the visible brane, it changes the vector bundle on the CalabiYau manifold. The second
Chern class of the new vector bundle changes by the amount of the curve on which the
five-brane was wrapped. This new bundle has more moduli. The new moduli are precisely
the transition moduli parameterizing the instanton branch of the superpotential W. In the
instanton branch, only those components of , which correspond to the transition moduli,
take non-zero expectation values and are massless, the remaining ones become massive
337
and get integrated out. The five-brane is also massive and integrated out in the instanton
branch. The origin represents a singularity, where all the multiplets become massless. From
the bundle viewpoint, the singularity at the origin corresponds to a vector bundle becoming singular and turning into a torsion free sheaf [49]. From the five-brane view-point,
the singularity at the origin corresponds to a five-brane coinciding with the visible brane.
An analogous, but simpler, transition takes place in string theory in the Dp D(p + 4)
system [88]. The Dp D(p + 4) system is describable by a supersymmetric field theory
with eight supercharges. The moduli space of this system consists of two branches, the
Coulomb branch and the Higgs branch. The Coulomb branch describes positions of the
Dp-brane away from the D(p + 4)-brane. The Higgs branch describes how the Dp-brane
can get dissolved into the D(p + 4)-brane. Geometrically, the Higgs branch is isomorphic
to the one-instanton moduli space on a 4-manifold which is just the ADHM moduli space.
In the heterotic M-theory case, the analog of the Coulomb branch is the moduli space of
five-branes. The analog of the Higgs branch is the space of transition moduli. Thus, transition moduli can be understood as a CalabiYau threefold generalization of the ADHM
one-instanton moduli space.
Now let us see how this picture changes in the presence of fluxes and non-perturbative
effects. For x very small, we have to include into the Lagrangian. The Kahler potential
receives an extra contribution
K trE8 ( ).
(4.38)
Then it is not hard to show that the potential energy will contain, among others, a term
U0
tr ( ).
2 E8
MPl
(4.39)
This is a consequence of the following very simple statement. If we take the theory with the
constant superpotential and quadratic Kahler potential, the potential energy has a maximum
at zero and for small fields is dominated by a negative quadratic term. Therefore, the mass
of the fields schematically behaves as
W(, X)
U0 .
(4.40)
2
2
MPl
=0
This means that as the five-brane gets very close to the visible sector, the fields become
tachyonic and begin to roll downhill. Since no slow roll conditions on are satisfied, this
terminates inflation. Eventually, the five-brane hits the wall and disappears (gets massive
and integrated out from the low-energy field theory) through a small instanton transition.
In addition, those components of , which do not correspond to the transition instanton
moduli, get a mass, according to the superpotential (4.36), and get integrated out. Now their
mass is set by the vacuum expectation values of the transition moduli. The new system of
moduli after the small instanton transition involves the moduli discussed in Section 2 plus
the new transition moduli. The physics of them will be considered in the next section.
This escape from inflation is, somewhat, analogous to one in D3D7 inflation studied
in [17]. In [17], the negative mass for the charged hypermultiplets (analogs of our fields )
was created by FayetIliopoulos terms. The same FayetIliopoulos terms were responsible
for stabilization at a non-zero value.
338
5. The post-inflationary phase

After inflation, the five-brane disappears through a small instanton transition and turns
into new vector bundle moduli, which we denote by i . The new system of moduli contains
now the following fields
S,
T 1,
T 2,
Y,
Z ,
i .
(5.1)
This system of moduli can be stabilized. The moduli S, T 1 , T 2 , Y and Z can be stabilized
by the same mechanism as in Section 2. In this section, we will concentrate on the moduli
i . Without loss of generality, we can assume that there is only one such modulus . Vector
bundle moduli have a non-perturbative superpotential. It appears as a factor in Eqs. (2.35)
(2.37). Since after the small instanton transition only the bundle on the visible brane has
changed, only Wv5 and Wvh will depend on after the transition. In fact,
Wv5 Wvh ,
(5.2)
since the coefficient in Eqs. (2.35) and (2.36) is much greater than one. Therefore, the
potential for is
W () = P()e Y ,
(5.3)
where we have denoted the factor depending on by P().

Let us make a remark. One can worry that, after the transition, the superpotential (5.3)
will not depend on . Indeed, the superpotential is induced by a string wrapping isolated
genus zero curves. On the other hand, the five-brane that turned into the modulus was
wrapped on a non-isolated genus zero curve or a higher genus curve. It seems possible that
the bundle over the isolated curves did not really change and the bundle moduli contribution to the superpotential will remain unchanged. However, this logic is not quite correct
because the curves might intersect. If the curve on which the five-brane was wrapped intersects at least one isolated genus zero curve, the non-perturbative superpotential will change
and it will depend on .
Even though the five-brane modulus x could not be stabilized, the new vector bundle
modulus can. The equation
D W = 0
(5.4)
has a solution. The analysis of this equation will be analogous to the one in [42]. Let us
recall that the superpotential W is a sum of various contributions
W = Wf + Wg + Wnp ,
(5.5)
where Wnp is now the sum of W5h and W (). Then Eq. (5.4) can be written as
P()e Y = K()Wf ,
(5.6)
where we have assumed that

Wf W .
(5.7)
339
In [42], it was shown that inequality (5.7) is indeed satisfied. Vector bundle moduli superpotentials were studied in detail in [81,82]. They were found to be high degree polynomials.
Therefore, we will take
P() = d .
(5.8)
The Kahler potential K() represents a more complicated problem. It was evaluated
explicitly only for bundles which can be written as the pullback of a bundle on a fourmanifold [89]. A generic bundle on a threefold is, obviously, not of this form. In this paper,
we will not need the actual form of the Kahler potential. Using Eq. (5.8) and ignoring the
imaginary part of Y, Eq. (5.6) can be written as
d ey = K()Wf .
(5.9)
Clearly, such an equation has a solution for a generic function K(). To present a numeric
solution, we need to know the order of magnitude of K. It was estimated in [42] and found
to be 105 in Planck units. Then, if d is sufficiently large, we can approximately write
Eq. (5.9) as follows
d 105 ey Wf .
(5.10)
Taking, as an example,
Wf 1010 ,
200,
y 0.8,
d 40,
(5.11)
we obtain
20.
(5.12)
The relatively large value of means that the gauge connection is spread out over the
CalabiYau manifold, rather than sharply peaked over some curve. As long as we stay
away from singularities in the moduli space of vector bundles, any value of is acceptable.
To be slightly more precise, Eq. (5.12) is a solution for the absolute value of . The phase
of can also be found form Eq. (5.6) [42].
Thus, we have shown that the new system of moduli has much simple stability properties. That is, it is possible to fix all moduli after inflation. Now we are going to argue that
the cosmological constant in the new vacuum can be positive and fine tuned to coincide
with observations. Recall that the cosmological constant during inflation is given by three
contributions
+ visible
.
inflation = SUGRA + hidden
D
D
(5.13)
The first term in Eq. (5.13) is the contribution from the supergravity potential energy. It is
2 . The second term comes from the Fayet
negative and approximately given by Wf2 /MPl
Iliopoulos term in the hidden sector. The last term is the contribution from the Fayet
Iliopoulos term in the visible sector. They second and the third terms are positive. In the
previous section, we argued that the existence of FayetIliopoulos terms in the visible
sector is not relevant for inflation. Nevertheless, we will assume that they are present.
The cosmological constant inflation has to be positive and large. After the small instanton
transition, some of the contributions to the cosmological constant might change, because
340
they depend on the properties of the vector bundle. The contribution SUGRA might change
because, as was argued in [61], the flux-induced superpotential may receive higher order
corrections from ChernSimons invariants which depend on the choice of the bundle. The
will change because it depends on the gauge connection. Moreover,
contribution visible
D
since after a small instanton transition there is a possibility of changing the number of
families of quarks and leptons [48,49], the corresponding U (1) gauge group might not be
will be zero. However, hidden
will not change
anomalous anymore. In this case visible
D
D
because the bundle on the hidden brane remains unchanged. All these arguments show
that, in principle, the cosmological constant changes after inflation. Since it receives both
negative and positive contributions, it is possible, by fine tuning, to make it consistent with
observations.
6. Conclusion
In this paper, we considered dynamics of the five-brane wrapped on a non-isolated genus
zero or higher genus curve. Non-perturbative superpotentials do not depend on moduli of
such five-branes. We showed that fluxes and non-perturbative effects can stabilize such a
five-brane in a non-supersymmetric AdS vacuum. We also showed that addition of Fayet
Iliopoulos terms not only can raise this vacuum to a dS vacuum but also can create one
more dS vacuum. We also stabilized h1,1 moduli which do not have non-perturbative superpotentials. The cosmological constant of the vacuum can be positive and fine tuned
to be consistent with observations. This provides a generalization of results of [5,42] and
shows that the most complete system of heterotic string moduli can be stabilized in a vacuum with a positive cosmological constant. In addition, we showed that, by modifying the
supergravity potential energy with FayetIliopoulos terms, one can create an inflationary
potential and treat the five-brane translational modulus as an inflaton. However, the potential cannot be trusted at very small distances because one should expect extra light states
to appear. We give a qualitative argument how such new states can terminate inflation. The
idea is that, in the presence of fluxes and non-perturbative effects, these extra states can become tachyonic when the five-brane comes very close to the visible brane. Eventually, the
five-brane hits the visible brane. The system undergoes a small instanton transition which
changes the vector bundle on the visible brane. The five-brane disappears from the lowenergy spectrum while the vector bundle moduli are created. They have simpler stability
properties and, as a result, the new system of moduli can be stabilized. We also argue that
the cosmological constant changes after the transition. The cosmological constant after the
transition can be fine tuned to be consistent with observations.
Acknowledgements
The author is indebted to Juan Maldacena for lots of interesting and helpful discussions.
The work is supported by NSF grant PHY-0070928.
341
References
[1] F. Quevedo, Lectures on string/brane cosmology, Class. Quantum Grav. 19 (2002) 57215779, hepth/0210292.
[2] S. Kachru, R. Kallosh, A. Linde, S.P. Trivedi, De Sitter vacua in string theory, Phys. Rev. D 68 (2003)
046005, hep-th/0301240.
[3] C.P. Burgess, R. Kallosh, F. Quevedo, De Sitter string vacua from supersymmetric D-terms, JHEP 0310
(2003) 056, hep-th/0309187.
[4] M. Becker, G. Curio, A. Krause, De Sitter vacua from heterotic M-theory, Nucl. Phys. B 693 (2004) 223
260, hep-th/0403027.
[5] E.I. Buchbinder, Raising anti-de Sitter vacua to de Sitter vacua in heterotic M-theory, Phys. Rev. D 70 (2004)
066008, hep-th/0406101.
[6] V. Balasubramanian, P. Berglund, Stringy corrections to Kahler potentials, SUSY breaking, and the cosmological constant problem, hep-th/0408054.
[7] G. Curio, A. Krause, G-fluxes and non-perturbative stabilization of heterotic M-theory, Nucl. Phys. B 643
(2002) 131156, hep-th/0108220.
[8] L. Kofman, A. Linde, X. Liu, A. Maloney, L. McAllister, E. Silverstein, Beauty is attractive: moduli trapping
at enhanced symmetry points, JHEP 0405 (2004) 030, hep-th/0403001.
[9] T. Mohaupt, F. Saueressig, Dynamical conifold transitions and moduli trapping in M-theory cosmology,
hep-th/0410273.
[10] G. Dvali, S.-H.H. Tye, Brane inflation, Phys. Lett. B 450 (1999) 7282, hep-ph/9812483.
brane annihilation, Phys. Rev. D 65 (2002) 023507, hep-th/0105032.
[11] S.H.S. Alexander, Inflation from DD
[12] C.P. Burgess, M. Majumdar, D. Nolte, F. Quevedo, G. Rajesh, R.-J. Zhang, The inflationary braneantibrane
universe, JHEP 0107 (2001) 047, hep-th/0105204.
[13] G. Shiu, S.-H.H. Tye, Some aspects of brane inflation, Phys. Lett. B 516 (2001) 421430, hep-th/0106274.
[14] D. Choudhury, D. Ghoshal, D.P. Jatkar, S. Panda, Hybrid inflation and braneantibrane system, JCAP 0307
(2003) 009, hep-th/0305104.
[15] C. Herdeiro, S. Hirano, R. Kallosh, String theory and hybrid inflation/acceleration, JHEP 0112 (2001) 027,
hep-th/0110271.
[16] J. Garcia-Bellido, R. Rabadan, F. Zamora, Inflationary scenarios from branes at angles, JHEP 0201 (2002)
036, hep-th/0112147.
[17] K. Dasgupta, C. Herdeiro, S. Hirano, R. Kallosh, D3/D7 inflationary model and M-theory, Phys. Rev. D 65
(2002) 126002, hep-th/0203019.
[18] S. Kachru, R. Kallosh, A. Linde, J. Maldacena, L. McAllister, S.P. Trivedi, Towards inflation in string theory,
JCAP 0310 (2003) 013, hep-th/0308055.
[19] J.P. Hsu, R. Kallosh, S. Prokushkin, On brane inflation with volume stabilization, JCAP 0312 (2003) 009,
hep-th/0311077.
[20] L. Pilo, A. Riotto, A. Zaffaroni, Old inflation in string theory, JHEP 0407 (2004) 052, hep-th/0401004.
[21] C.P. Burgess, J.M. Cline, H. Stoica, F. Quevedo, Inflation in realistic D-brane models, JHEP 0409 (2004)
033, hep-th/0403119.
[22] O. De Wolfe, S. Kachru, H. Verlinde, The giant inflaton, JHEP 0405 (2004) 017, hep-th/0403123.
[23] N. Iizuka, S.P. Trivedi, An inflationary model in string theory, Phys. Rev. D 70 (2004) 043519, hepth/0403203.
[24] R. Kallosh, A. Linde, P-term, D-term and F-term inflation, JCAP 0310 (2003) 008, hep-th/0306058.
[25] J.J. Blanco-Pillado, C.P. Burgess, J.M. Cline, C. Escoda, M. Gomez-Reino, R. Kallosh, A. Linde, F.
Quevedo, Racetrack inflation, hep-th/0406230.
[26] A. Buchel, A. Ghodsi, Braneworld inflation, hep-th/0404151.
[27] M. Berg, M. Haack, B. Kors, Loop corrections to volume moduli and inflation in string theory, hepth/0404087.
[28] E. Gravanis, N.E. Mavromatos, Vacuum energy and cosmological supersymmetry breaking in brane worlds,
Phys. Lett. B 547 (2002) 117127, hep-th/0205298.
[29] J. Ellis, N.E. Mavromatos, D.V. Nanopoulos, A. Sakharov, Brany Liouville inflation, gr-qc/0407089.
342
[30] M. Gomez-Reino, I. Zavala, Recombination of intersecting D-branes and cosmological inflation, JHEP 0209
(2002) 020, hep-th/0207278.
[31] P. Horava, E. Witten, Heterotic and type I string dynamics from eleven dimensions, Nucl. Phys. B 460 (1996)
506524, hep-th/9510209.
[32] P. Horava, E. Witten, Eleven-dimensional supergravity on a manifold with boundary, Nucl. Phys. B 475
(1996) 94114, hep-th/9603142.
[33] E. Witten, Strong coupling expansion of CalabiYau compactification, Nucl. Phys. B 471 (1996) 135158,
hep-th/9602070.
[34] A.E. Faraggi, Phenomenological survey of M-theory, hep-th/0307037, in: Proceedings of SUGRA 20 Conference, Boston, 1721 March 2003, in press.
[35] R. Donagi, A. Lukas, B.A. Ovrut, D. Waldram, Holomorphic vector bundles and non-perturbative vacua in
M-theory, JHEP 9906 (1999) 034, hep-th/9901009.
[36] R. Donagi, B.A. Ovrut, T. Pantev, D. Waldram, Standard Model bundles on non-simply connected Calabi
Yau threefolds, JHEP 0108 (2001) 053, hep-th/0008008.
[37] R. Donagi, B.A. Ovrut, T. Pantev, R. Reinbacher, SU(4) instantons on CalabiYau threefolds with Z2 Z2
fundamental group, hep-th/0307273.
[38] V. Braun, B.A. Ovrut, T. Pantev, R. Reinbacher, Elliptic CalabiYau threefolds with Z3 Z3 Wilson lines,
hep-th/0410055.
[39] R. Donagi, Y.-H. He, B.A. Ovrut, R. Reinbacher, Moduli dependent spectra of heterotic compactifications,
Phys. Lett. B 598 (2004) 279284, hep-th/0403291.
[40] R. Donagi, Y.-H. He, B.A. Ovrut, R. Reinbacher, The particle spectrum of heterotic compactifications, hepth/0405014.
[41] R. Donagi, Y.-H. He, B.A. Ovrut, R. Reinbacher, Higgs doublets, split multiplets and heterotic SU(3)C
SU(2)L U (1)Y spectra, hep-th/0409291.
[42] E.I. Buchbinder, B.A. Ovrut, Vacuum stability in heterotic M-theory, Phys. Rev. D 69 (2004) 086010, hepth/0310112.
[43] M. Dine, N. Seiberg, E. Witten, FayetIliopoulos terms in string theory, Nucl. Phys. B 289 (1987) 589.
[44] E.I. Buchbinder, R. Donagi, B.A. Ovrut, Vector bundle moduli and small instanton transitions, JHEP 0206
(2002) 054, hep-th/0202084.
[45] E.I. Buchbinder, B.A. Ovrut, R. Reinbacher, Instanton moduli in string theory, hep-th/0410200.
[46] E. Witten, Small instantons in string theory, Nucl. Phys. B 460 (1996) 541559, hep-th/9511030.
[47] N. Seiberg, E. Witten, Comments on string dynamics in six dimensions, Nucl. Phys. B 471 (1996) 121134,
hep-th/9603003.
[48] S. Kachru, E. Silverstein, Chirality changing phase transitions in 4d string vacua, Nucl. Phys. B 504 (1997)
272284, hep-th/9704185.
[49] B.A. Ovrut, T. Pantev, J. Park, Small instanton transitions in heterotic M-theory, JHEP 0005 (2000) 045,
hep-th/0001133.
[50] A. Lukas, B.A. Ovrut, D. Waldram, On the four-dimensional effective action of strongly coupled heterotic
string theory, Nucl. Phys. B 532 (1998) 4382, hep-th/9710208.
[51] E. Witten, private communications.
[52] M. Dine, N. Seiberg, X.G. Wen, E. Witten, Nonperturbative effects on the string world sheet. 2, Nucl. Phys.
B 289 (1987) 319.
[53] A. Lukas, B.A. Ovrut, K.S. Stelle, D. Waldram, Heterotic M-theory in five dimensions, Nucl. Phys. B 552
(1999) 246290, hep-th/9806051.
[54] R. Donagi, B.A. Ovrut, D. Waldram, Moduli spaces of fivebranes on elliptic CalabiYau threefolds,
JHEP 9911 (1999) 030, hep-th/9904054.
[55] T. Banks, M. Dine, Couplings and scales in strongly coupled heterotic string theory, Nucl. Phys. B 479
(1996) 173196, hep-th/9605136.
[56] P. Candelas, X. de la Ossa, Moduli space of CalabiYau manifolds, Nucl. Phys. B 355 (1991) 455.
[57] J.-P. Derendinger, R. Sauser, A five-brane modulus in the effective N = 1 supergravity of M-theory, Nucl.
Phys. B 598 (2001) 87114, hep-th/0009054.
[58] S. Gukov, C. Vafa, E. Witten, CFTs from CalabiYau four-folds, Nucl. Phys. B 584 (2000) 69108, hepth/9906070;
343
S. Gukov, C. Vafa, E. Witten, Nucl. Phys. B 608 (2001) 477478, Erratum.

[59] K. Behrndt, S. Gukov, Domain walls and superpotentials from M-theory on CalabiYau three-folds, Nucl.
Phys. B 580 (2000) 225242, hep-th/0001082.
[60] M. Becker, D. Constantin, A note on flux induced superpotentials in string theory, JHEP 0308 (2003) 015,
hep-th/0210131.
[61] S. Gukov, S. Kachru, X. Liu, L. McAllister, Heterotic moduli stabilization with fractional ChernSimons
invariants, Phys. Rev. D 69 (2004) 086008, hep-th/0310159.
[62] M. Dine, R. Rohm, N. Seiberg, E. Witten, Gluino condensation in superstring models, Phys. Lett. B 156
(1985) 55.
[63] P. Horava, Gluino condensation in strongly coupled heterotic string theory, Phys. Rev. D 54 (1996) 7561
7569, hep-th/9608019.
[64] A. Lukas, B.A. Ovrut, D. Waldram, Gaugino condensation in M-theory on S 1 /Z2 , Phys. Rev. D 57 (1998)
75297538, hep-th/9711197.
[65] A. Lukas, B.A. Ovrut, D. Waldram, Non-standard embedding and five-branes in heterotic M-theory, Phys.
Rev. D 59 (1999) 106005, hep-th/9808101.
[66] V. Kaplunovsky, J. Louis, Model-independent analysis of soft terms in effective supergravity and in string
theory, Phys. Lett. B 306 (1993) 269275, hep-th/9303040.
[67] A. Brignole, L.E. Ibez, C. Muoz, Towards a theory of soft terms for the supersymmetric standard model,
Nucl. Phys. B 422 (1994) 125171, hep-ph/9308271;
A. Brignole, L.E. Ibez, C. Muoz, Nucl. Phys. B 436 (1995) 747748, Erratum.
[68] H.P. Nilles, M. Olechowski, M. Yamaguchi, Supersymmetry breaking and soft terms in M-theory, Phys.
Lett. B 415 (1997) 2430, hep-th/9707143.
[69] Z. Lalak, S. Thomas, Gaugino condensation, moduli potentials and supersymmetry breaking in M-theory
models, Nucl. Phys. B 515 (1998) 5572, hep-th/9707223.
[70] H.P. Nilles, Gaugino condensation and SUSY breakdown, in: Lectures at Cargese School of Physics and
Cosmology, Cargese, France, August 2003, hep-th/0402022.
[71] R. Donagi, J. Khoury, B.A. Ovrut, P.J. Steinhardt, N. Turok, Visible branes with negative tension in heterotic
[72] G. Curio, A. Krause, Four-flux and warped heterotic M-theory compactifications, Nucl. Phys. B 602 (2001)
172200, hep-th/0012152.
[73] G. Curio, A. Krause, Enlarging the parameter space of heterotic M-theory flux compactifications to phenomenological viability, Nucl. Phys. B 693 (2004) 195222, hep-th/0308202.
[74] M. Dine, N. Seiberg, X.G. Wen, E. Witten, Nonperturbative effects on the string world sheet. 2, Nucl. Phys.
B 278 (1986) 769.
[75] E. Witten, Non-perturbative superpotentials in string theory, Nucl. Phys. B 474 (1996) 343360, hepth/9604030.
[76] K. Becker, M. Becker, A. Strominger, Fivebranes, membranes and non-perturbative string theory, Nucl.
Phys. B 456 (1995) 130152, hep-th/9507158.
[77] E. Witten, World-sheet corrections via D-instantons, JHEP 0002 (2000) 030, hep-th/9907041.
[78] E. Lima, B.A. Ovrut, J. Park, R. Reinbacher, Non-perturbative superpotentials from membrane instantons in
heterotic M-theory, Nucl. Phys. B 614 (2001) 117170, hep-th/0101049.
[79] E. Lima, B.A. Ovrut, J. Park, Five-brane superpotentials in heterotic M-theory, Nucl. Phys. B 626 (2002)
113164, hep-th/0102046.
[80] G. Moore, G. Peradze, N. Saulina, Instabilities in heterotic M-theory induced by open membrane instantons,
Nucl. Phys. B 670 (2003) 2789, hep-th/0206092.
[81] E.I. Buchbinder, R. Donagi, B.A. Ovrut, Superpotentials for vector bundle moduli, Nucl. Phys. B 653 (2003)
400420, hep-th/0205190.
[82] E.I. Buchbinder, R. Donagi, B.A. Ovrut, Vector bundle moduli superpotentials in heterotic superstrings and
[83] S. Kachru, M.B. Schulz, S. Trivedi, Moduli stabilization from fluxes in a simple IIB orientifold, hepth/0201028.
[84] J. Khoury, B.A. Ovrut, P.J. Steinhardt, N. Turok, The Ekpyrotic universe: colliding branes and the origin of
the hot big bang, Phys. Rev. D 64 (2001) 123522, hep-th/0103239.
344
[85] J. Khoury, B.A. Ovrut, N. Seiberg, P.J. Steinhardt, N. Turok, From big crunch to big bang, Phys. Rev. D 65
(2002) 086007, hep-th/0108187.
[86] J. Khoury, B.A. Ovrut, P.J. Steinhardt, N. Turok, Density perturbations in the Ekpyrotic scenario, Phys. Rev.
D 66 (2002) 046005, hep-th/0109050.
[87] O.J. Ganor, A. Hanany, Small E8 instantons and tensionless non-critical strings, Nucl. Phys. B 474 (1996)
122140, hep-th/9602120.
[88] M.R. Douglas, Branes within branes, in: Cargese 1997, Strings, Branes and Dualities, pp. 267275, hepth/9512077.
[89] J. Gray, A. Lukas, Gauge five-brane moduli in four-dimensional heterotic models, hep-th/0309096.
Inclusive production of single hadrons with finite

transverse momenta in deep-inelastic scattering
at next-to-leading order
B.A. Kniehl, G. Kramer, M. Maniatis
II. Institut fr Theoretische Physik, Universitt Hamburg, Luruper Chaussee 149, 22761 Hamburg, Germany
Abstract
We calculate the cross section for the inclusive production of single hadrons with finite transverse momenta in deep-inelastic scattering at next-to-leading order (NLO), i.e., through O(s2 ), in
the parton model of QCD endowed with non-perturbative parton distribution functions (PDFs) and
fragmentation functions (FFs). The NLO correction is found to produce a sizeable enhancement in
cross section, of up to one order of magnitude, bringing the theoretical prediction to good agreement
with recent measurements for neutral pions and charged hadrons at DESY HERA. This provides a
useful test for the universality and the scaling violations of the FFs predicted by the factorization
theorem.
PACS: 12.38.Bx; 12.39.St; 13.87.Fh; 14.40.Aq
1. Introduction
The predictive power of the parton model of quantum chromodynamics (QCD) lies in
the factorization theorem. In deep-inelastic scattering (DIS), factorization in short- and
long-distance parts allows us to describe the observed cross sections of inclusive hadron
production as a convolution of the partonic cross sections with non-perturbative parton
E-mail address: bernd.kniehl@desy.de (B.A. Kniehl).
doi:10.1016/j.nuclphysb.2005.01.031
346
B.A. Kniehl et al. / Nuclear Physics B 711 (2005) 345366
density functions (PDFs) and fragmentation functions (FFs) [1]. Single-hadron inclusive
production in electronproton DIS,
e (k) + p(P ) e (k ) + h(p) + X,
(1)
occurs partonically already in the absence of strong interactions, at O(s0 ), where s is the
strong-coupling constant, when one parton of the proton (a quark) interacts with the lepton
current and fragments into the hadron h (naive parton model). If the virtuality Q2 = q 2
of the four-momentum transfer q = k k satisfies Q2 m2Z , where mZ denotes the Zboson mass, then process (1) is essentially mediated by a virtual photon ( ), while the
contribution from Z-boson exchange is negligible. In the following, this is the situation we
are interested in.
Since we are interested in perturbative QCD effects, we require the hadron to carry
non-vanishing transverse momentum (pT ) in the centre-of-mass (c.m.) frame of the virtual photon and the incoming proton. At leading order (LO), the corresponding partonic
subprocesses thus contain two partons in the final state, one of which fragments into the
hadron, while the other one balances the transverse momentum. At next-to-leading order
(NLO), three-parton final states contribute to the real correction, while the virtual correction arises from one-loop diagrams with two final-state partons.
The investigation of single-hadron production is interesting for several reasons. First of
all, it provides a test of perturbative QCD and of factorization. Apart from the partonic
cross sections obtainable from perturbative QCD, the theoretical predictions essentially
depend on universal PDFs and FFs. In particular, the FFs, which are fitted to electron
positron-annihilation data, may be tested with regard to their universality. Furthermore,
the theoretical predictions allow for a direct comparison with experimental data, without
resorting to any kind of Monte Carlo model to simulate the hadronization of the outgoing
partons. Thus, we may expect very meaningful results. Moreover, the theoretical predictions are directly sensitive to the gluon PDF of the proton with the potential to constrain
the latter.
On the experimental side, precise data were collected by the H1 [24] and ZEUS [5,6]
Collaborations at the ep collider HERA at DESY. They refer to 0 mesons in the forward
region [3,4], with small angles with respect to the proton remnant, and to charged hadrons
[2,5,6].
More than 25 years ago, the cross section of process (1) with finite transverse momentum of the hadron h was calculated by Mndez [7] at LO, to O(s ). Since QCD corrections
are typically large and we are confronted with precise experimental data, it is desirable
to compare these data with predictions of at least NLO accuracy, including the terms of
O(s2 ). For this purpose, a first NLO QCD prediction was computed in Ref. [8], neglecting
the longitudinal degrees of freedom of the virtual photon.
The theoretical description can be rendered more reliable by resumming the leading
logarithmic contributions of the perturbation expansion. In this sense, the BalitskyFadin
KuraevLipatov (BFKL) [9] and DokshitserGribovLipatovAltarelliParisi (DGLAP)
[10] equations resum at LO the (s ln(1/xB ))n and (s ln(Q2 /Q20 ))n contributions, respectively, where xB is the Bjorken variable and Q0 is the cut-off scale for the perturbative
evolution. Disregarding the fact that these resummations are just approximations, they evidently fail in the kinematic regions of large xB values and small Q2 values, respectively.
347
In this paper, we perform a full NLO QCD calculation, also taking into account the
longitudinal degrees of freedom of the virtual photon. We encounter ultraviolet (UV) and
infrared (IR) singularities, which we all regularize using dimensional regularization. In order to overcome the difficulties in connection with the IR singularities emerging in different
parts of the NLO correction, we employ the dipole subtraction formalism [11]. In contrast
to the more conventional phase-space slicing method, there is no need to introduce any unphysical parameter to cut the phase space into soft, collinear, and hard regions. Moreover,
all cancellations of IR singularities occur before any numerical phase-space integration is
performed. We thus conveniently obtain numerically stable predictions.
An independent calculation was recently presented in Ref. [12], where the matrix elements of the hard-scattering processes were adopted from the DISENT program package
[13]. In Ref. [12], the phase-space slicing method was applied to handle the IR singularities. Another related work, focusing on fracture functions of the proton, was published in
two parts, related to incoming gluons [14] and quarks [15].
This paper is organized as follows. In Section 2, we describe our analytical analysis.
The LO result and a specific part of the real NLO correction are relegated to Appendices A
and B, respectively. In Section 3, we present our numerical results. Our conclusions are
summarized in Section 4.
2. Analytical analysis
According to the factorization theorem [1], the differential cross section for process (1)
p
is given as a convolution of the hard-scattering cross sections d ab with the PDFs Fa of
h
the proton and the FFs Db of hadron h, as

d 4 h
=
d x dy d z d
1
ab x
dx
x
1
z
dz p
d 4 ab
Fa (x/x,
D h (z/z, f ),
i )
z
dx dy dz d b
(2)
where i and f are the factorization scales related to the initial and final states and the
sum runs over all tagged initial- and final-state partons, a and b, respectively. As usual,
the dimensionless variables x, y, and z are defined as x = Q2 /(2pa q), y = (pa q)/(pa k),
and z = (pa pb )/(pa q) with respect to the partonic four-momenta pa and pb , and their
barred counterparts x = xB = Q2 /(2P q), y = y = (P q)/(P k), and z = (Pp)/(P q) with
respect to the hadronic four-momenta. We have Q2 = xB yS, where S = (P + k)2 is the
square of the ep c.m. energy. It is convenient to describe the kinematics in the c.m. frame
of the virtual photon and the incoming parton a as is done in Fig. 1, where we take the
three-momentum of the virtual photon to point along the z axis and the three-momenta of
the incoming and scattered electrons to lie in the xz plane. Then, the azimuthal angle of
the hadron h is enclosed between the plane spanned by the three-momenta of the incoming
and scattered electrons and the one spanned by those of the virtual photon and the outgoing
parton b.
348
Fig. 1. C.m. frame of the virtual photon and the initial-state parton a, where the three-momenta of the leptons are
rotated into the xz plane.
The hard-scattering cross sections may be written as contractions of a lepton tensor l

ab , as
with hadron tensors H
2 y ab
d 4 ab
=
l H ,
dx dy dz d 16 2 Q4
(3)
where is Sommerfelds fine-structure constant. If the virtual photon and the initial-state
parton are both unpolarized, then there cannot be any dependence on the azimuthal angle
. Integrating over the latter, we find the decomposition

2 y 2 2y + 2 ab
y 2 6y + 6 ab
d 3 ab
=
H
+
2
H
(4)
T
L ,
dx dy dz 8
2yQ2
y3s2
ab , H ab = p p H ab , and s = (p + q)2 .
where HTab = g H
a
a a
L
The partonic subprocesses contributing at LO are
+ q q + g,
(5)
(6)
(7)
+ q g + q,
+ g q + q,
where it is understood that the first of the final-state partons is the one that fragments
into the hadron h. Here, q = q1 , q1 , . . . , qnf , qnf , where nf is the number of active quark
flavours, which are ordered according to their masses, i.e., q1 = u, q2 = d, q3 = s, q4 = c,
and q5 = b, and we identify q = q. There are two Feynman diagrams for each of the
processes (5)(7). The matrix elements of processes (5)(7) are interrelated through crossing symmetry. Owing to charge-conjugation (C) invariance, the counterparts of processes
(5)(7) with quarks and antiquarks interchanged yield equal cross sections and do not have
to be calculated separately. However, this is not generally true for the PDFs and FFs. Therefore, we have to explicitly sum over all possible pairings of partons a and b in Eq. (2). For
the readers convenience, the LO expressions for the Lorentz invariants HTab and HLab in
349
Eq. (4) are listed in Appendix A. They are of O(s ) and proportional to eq2 , where eq
denotes the electric charge of quark q in units of the positron charge.
In order to determine the NLO correction to the cross section of process (1), we have to
compute the virtual and real corrections of O(s2 ) to the hadron tensors. We then encounter
a rather involved pattern of singularities. All these singularities are regularized using dimensional regularization with D = 4 2 spacetime dimensions yielding poles in in
the physical limit D 4. The integrations over the loop four-momenta in the virtual correction lead to UV and IR singularities, where the IR ones comprise both soft and collinear
singularities. All UV singularities are removed through the renormalizations of the wave
functions and the strong-coupling constant. The remaining soft and collinear singularities
cancel partly against counterparts originating from the phase-space integration of the real
correction. The remaining collinear poles have to be factorized into the bare PDFs and FFs
so as to render them finite.
The virtual correction is obtained as the interference of the Born and one-loop matrix
elements. The latter receive contributions from self-energy, triangle, and box diagrams.
These involve two-, three-, and four-point tensor integrals, which are reduced to scalar integrals via tensor reduction [16]. The scalar integrals contain both UV and IR singularities.
They are computed analytically in dimensional regularization. Our analytic expressions for
the contractions of the resulting hadron tensors with g agree with the literature [17]. The
virtual correction is renormalized in the modified minimal-subtraction (MS) scheme and
thus UV finite.
The partonic subprocesses contributing to the real correction read
+ q q + g + g,
(8)
(9)
(10)
(11)
(12)
+ q g + q + g,
+ g q + q + g,
+ g g + q + q,
+ q q + q + q,
+ q q + q + q,
+ q q + q + q ,
+ q q + q + q,
(13)
(14)
(15)
where q, q = q1 , q1 , . . . , qnf , qnf with q = q . As in processes (5)(7), the first partons

in the final states of processes (8)(15) are taken to fragment into the hadron h. The order in which the residual final-state partons appear is irrelevant. There are eight Feynman
diagrams for each of the processes (8)(13) and four ones for each of the processes (14)
and (15). Crossing symmetry interrelates the matrix elements of processes (8)(11), those
of processes (12), (13), and those of processes (14), (15). The cross sections of processes
(8)(15) are of O(s2 ). Those of processes (8)(13) are proportional to eq2 , while, at first
sight, those of processes (14), (15) contain pieces proportional to eq2 , eq eq , and eq2 . However, in the case of process (14), the piece proportional to eq eq vanishes by Furrys
theorem, as is explained below. The squared matrix elements of processes (8)(11) involve one quark trace, those of processes (12) and (13) contain pieces with one or two
350
Fig. 2. Cut-diagram involving two fermion traces each coupled to three gauge bosons, as illustrated in Ref. [18].
The cut proceeds along the numbered ticks representing the four on-shell quarks. If the charges of both quark
loops are tagged, there is no Furry cancellation with an analogous diagram with one fermion-number flow flipped.
quark traces, and those of processes (14) and (15) involve two quark traces. Due to C invariance, the counterparts of processes (8)(15) with quarks and antiquarks interchanged
yield equal cross sections. Notice that, in the case of process (15), we have to distinguish
between the case where the tagged quarks q and q are both particles or anti-particles and
the case where one is a particle and the other one is an anti-particle. Processes (8) and (13)
each contain two identical untagged partons in the final state, so that their cross sections
receive a statistical factor of 1/2 to avoid double counting in the phase-space integration.
We derived the matrix elements of processes (8)(15) in two steps. First, we calculated the
ones of the corresponding processes of e+ e annihilation via a virtual photon, which may
also be found in Ref. [18]. Then, we employed crossing symmetry. The squared matrix
elements of processes (8)(15), excluding the Furry terms discussed below, are also implemented in the DISENT program package [13]. Performing a numerical comparison with
the latter, we find agreement.
In Ref. [18], the squared matrix elements, which may be visualized as cut diagrams, are
classified with respect to colour factors. One specific class, called F terms, contains all cut
diagrams with two fermion loops, which are both coupled to three vector bosons, namely
to one photon and two gluons. This class constitutes a gauge-parameter-independent subset
of the NLO correction. The cut diagram of one specific member of this class is shown in
Fig. 2, where the on-shell quarks are indicated by numbers. As was noticed in Ref. [18],
by Furrys theorem, each cut diagram within this class exactly cancels against one counterpart in which one fermion-number flow is reversed if the on-shell quarks associated with
the loop whose fermion-number flow is reversed are not tagged in the experiment, i.e., if
the three-momenta of these quarks are integrated over. This argument is also true in the
case where only one fermion charge is identified, for instance in single-hadron production
by e+ e annihilation, since there is still a counterpart diagram where the other fermionnumber flow is reversed. In our case, there are two tagged partons, one coming from the
proton and one fragmenting into the hadron h. Suppose the two tagged partons are the
quarks 1 and 2 in the cut diagram of Fig. 2. This situation can occur for processes (12) and
(13), which involve only one quark flavour, and for process (15), which involves two different quark flavours. Then, the Furry cancellation is impeded because there is no counterpart
diagram. Thus, we are not allowed to omit this class of cut diagrams in our calculation.
The corresponding squared matrix elements are listed in Appendix B.
The differential cross sections of processes (8)(15) have to be integrated over the threemomenta of the second and third final-state partons keeping the three-momentum of the
first one fixed. Performing the phase-space integrations, we encounter IR singularities of
351
the soft and/or collinear types, which, for consistency with the virtual correction, must be
extracted using dimensional regularization. It is convenient to do this by means of the dipole subtraction formalism [11]. The general idea of this formalism is to subtract from the
contribution to the real correction due to a given partonic subprocess some artificial counterterm which has the same point-wise IR-singular behaviour in D spacetime dimensions
as the considered part of the real correction itself. Thus, the limit 0 can be performed,
and the phase-space integration can be evaluated numerically in four dimensions. The artificial counterterm is constructed in such a way that it can be integrated over the one-parton
subspace analytically leading to poles in . Adding the terms thus constructed to the virtual correction, the IR singularities of the latter are cancelled analytically. In the present
case, where the three-momenta of two tagged partons need to be kept fixed, additional,
more complicated artificial counterterms appear than in situations where only one parton
is tagged, such as inclusive jet production. A technical advantage of the dipole subtraction
method compared to the phase-space slicing method is that all IR singularities cancel before any numerical integration is performed. Furthermore, there is no need to introduce a
slicing parameter to separate soft and/or collinear phase-space regions from the remaining
hard region, which needs to be tuned in order to obtain a numerically stable result. For the
factorization of the collinear singularities associated with the tagged partons, we choose
the MS scheme. In turn, we have to employ PDFs and FFs which are defined in the same
scheme.
Finally, we end up with two contributions, the real correction with the artificial counterterms subtracted and the virtual correction with the integrated artificial counterterms
included, which are both finite in the physical limit 0 and can be integrated over their
three- and two-particle phase spaces, respectively, in three spacial dimensions. These integrations are performed numerically using a custom-made C++ routine. On the other hand,
all algebraic calculations are executed with the help of the symbolic-manipulation package
FORM [19].
3. Numerical results
We are now in a position to present our numerical results for the cross section of
single-hadron inclusive production in ep DIS. We start by specifying our input. We
work in the MS renormalization and factorization scheme with nf = 5 massless quark
flavours. At NLO (LO), we employ set CTEQ6M (CTEQ6L1) of proton PDFs by the
coordinated theoreticalexperimental project on QCD (CTEQ) [20], the NLO (LO) set
of FFs for light charged hadrons ( , K , and p/p)
by Kniehl, Kramer, and Ptter
(n )
(KKP) [21], and the two-loop (one-loop) formula for s f (r ) with asymptotic scale
(5)
parameter QCD = 226 MeV (165 MeV) [20]. This value is compatible with the result
(5)
QCD = (213 80) MeV ((88 41) MeV) determined in Ref. [22]. We approximate the
0 FFs as
1
0
Da (x, f ) = Da (x, f ),
2
(16)
352
where Da refers to the sum of the + and mesons, which is supported by LEP1
data of hadronic Z 0 -boson decays [23]. Furthermore, we assume the charged hadrons to
be exhausted by the charged pions, charged kaons, protons, and antiprotons, viz
p/p
Dah (x, f ) = Da (x, f ) + DaK (x, f ) + Da
(x, f ).
(17)
For simplicity, we identify the renormalization scale r and the initial- and final-state
factorization scales, i and f , respectively, and relate them to the characteristic dimensionful variables Q2 and pT by setting 2r = 2i = 2f = [Q2 + (pT )2 ]/2, where is a
dimensionless parameter of order unity introduced to estimate the theoretical uncertainty
due to unphysical-scale variations. As usual, we consider variations of between 1/2 and
2 about the default value 1.
We now compare our theoretical predictions with HERA data on 0 mesons in the
forward region from the H1 Collaboration [3,4] and on charged hadrons in the currentjet region from the ZEUS Collaboration [5]. We start by discussing the H1 data [3,4],
which were taken in DIS of positrons with energyEe = 27.6
GeV on protons with energy
Ep = 820 GeV in the laboratory frame, so that S = 2 Ee Ep = 301 GeV, during the
running periods 1996 and 1996/1997, and correspond to integrated luminosities of 5.8 and
21.2 pb1 , respectively. In Refs. [3,4], the 0 mesons were described by their transverse
momentum pT in the p c.m. frame and by their angle with respect to the proton flight
direction, their pseudorapidity = ln[tan(/2)], and their energy E = xE Ep in the laboratory frame. They were detected within the acceptance cuts pT > 2.5 GeV or 3.5 GeV,
5 < < 25 , and xE > 0.01. The DIS phase space was restricted to the kinematic regime
defined by 0.1 < y < 0.6 and 2 < Q2 < 70 GeV2 . The cross section was measured differentially in pT [3,4], [3], xE [4], and xB [3,4] for various Q2 intervals, differentially in
xE for various xB intervals [4], and differentially in Q2 [3]. The differential cross sections
0
0
0
0
0
d /dpT , d /d, d /dxE , d /dxB , and d /dQ2 presented in Refs. [3] (open
circles) and [4] (solid circles) are compared with our LO (dashed histograms) and NLO
(solid histograms) predictions in Figs. 37, respectively. In Figs. 3, 4, 5(a), and 6(a), the
upper three frames refer to the Q2 intervals 2 < Q2 < 4.5 GeV2 , 4.5 < Q2 < 15 GeV2 ,
and 15 < Q2 < 70 GeV2 . In Fig. 5(b), the upper three frames refer to the xB intervals
0.000042 < xB < 0.0002, 0.0002 < xB < 0.001, and 0.001 < xB < 0.0063. In Fig. 6(b),
the upper three frames refer to the Q2 intervals 2 < Q2 < 8 GeV2 , 8 < Q2 < 20 GeV2 ,
and 20 < Q2 < 70 GeV2 . In all figures, the minimum-pT cut is pT > 2.5 GeV, expect
for Fig. 6(b), where it is pT > 3.5 GeV. In Figs. 37, the shaded bands indicate the theoretical uncertainties of the NLO predictions due to the variation described above. The
K factors, defined as the NLO to LO ratios of our default predictions, are shown in the
downmost frames of Figs. 37.
We observe from Figs. 37 that the H1 data generally agree with our NLO predictions
within errors, while they significantly overshoot our default LO predictions. Indeed, the K
factors always exceed unity and even reach one order of magnitude at low values of pT , Q2 ,
or xB . Not only do the LO predictions disagree with the H1 data in their normalizations,
but they also exhibit deviating shapes. On the other hand, under the effect of asymptotic
freedom, the K factors approach unity for increasing values of r , i.e., for increasing
values of pT and/or Q2 .
353
Fig. 3. Differential cross section d /dpT (in pb/GeV) of e+ p e+ 0 + X in DIS with 0.1 < y < 0.6 and
2 < Q2 < 4.5 GeV2 (first frame), 4.5 < Q2 < 15 GeV2 (second frame), or 15 < Q2 < 70 GeV2 (third frame)
at HERA with Ee = 27.6 GeV and Ep = 820 GeV for 0 mesons with 5 < < 25 and xE > 0.01. H1 data
from Ref. [3] (open circles) and [4] (solid circles) are compared with our default LO (dashed histograms) and
NLO (solid histograms) predictions including theoretical uncertainties due to variation (shaded bands). The K
factors (fourth frame) are also shown.
There is an obvious explanation for the sizeable K factors at low values of r in terms
of the different kinematic constraints at LO and NLO. The LO processes (5)(7) are 2 2,
and their cross sections are sensitive to collinear singularities only as pT 0. By contrast,
processes (8)(15) contributing to the real NLO correction are 2 3, so that collinear
configurations can also arise for finite values of pT . After mass factorization of the corresponding collinear singularities, the finite remainders can be sizeable, leading to large
NLO corrections. A similar line of reasoning was presented in Ref. [12].
Unfortunately, the theoretical uncertainties in our NLO predictions due to variation are
rather sizeable, especially at low values of pT , Q2 , or xB , where the K factors themselves
are abnormally large. This is partly related to the opening of new partonic production
channels at NLO, which are still absent at LO, namely those of Eqs. (11), (13) and (15).
354
Fig. 4. Same as in Fig. 3, but for d /d (in pb) with pT > 2.5 GeV.
Obviously, a reduction in dependence can only be expected to happen at next-to-next-toleading order (NNLO), which is beyond the scope of this work.
Besides the freedom in the choice of the renormalization and factorization scales, there
are other sources of theoretical uncertainty, including the variations of the PDF and FF sets.
However, in view of the considerable spread in cross section induced by the moderate
variations described above, we conclude that the residual sources of theoretical uncertainty
are of minor importance. Furthermore, we must bear in mind that the factorization theorem
itself is only valid up to terms of O(2QCD /(pT )2 ), which may become large in the low-pT
range.
We now turn to the ZEUS data on charged hadrons [5], which were produced in DIS
of electrons with energy
Ee = 26.7 GeV on protons with energy Ep = 820 GeV in the
laboratory frame, giving S = 296 GeV, during the 1993 running period and correspond
to an integrated luminosity of 0.55 pb1 . They refer to the DIS phase space defined by
10 < Q2 < 160 GeV2 and 75 < W < 175 GeV, where W is the p invariant mass, with
355
(a)
0
Fig. 5. (a) Same as in Fig. 3, but for d /dxE (in nb) with pT > 2.5 GeV. (b) Same as in (a), but
for 2 < Q2 < 70 GeV2 and 0.00042 < xB < 0.0002 (first frame), 0.0002 < xB < 0.001 (second frame), or
0.001 < xB < 0.0063 (third frame).
W 2 = (P + q)2 = (1 xB )yS, and come as multiplicities differential in pT or Feynmans x variable xF = 2pL /W , where pL = pT sinh is the projection of the hadron
three-momentum onto the flight direction of the virtual photon in the p c.m. frame, and
normalized to the total number of DIS events. Unfortunately, the xF distribution of the multiplicity includes charged hadrons with pT values down to zero, while our NLO analysis
is only valid for finite values of pT , so that a comparison is impossible. However, a comparison is feasible for the pT distribution (1/Nevt ) dNhad /dpT [5], which includes charged
hadrons with xF > 0.05. The differential cross section d h /dpT may be obtained using
the conversion formula [24]
1 d h
1 dNhad
=
,
DIS
Nevt dpT
tot dpT
(18)
356
(b)
Fig. 5. (continued)
DIS is the total cross section in the DIS regime specified above,
where tot
2
Q
max
DIS
tot
W
max
2
dQ
Q2min
dW
d 2 DIS
.
dQ2 dW
(19)
Wmin
We have [25]

d 2 DIS
W
= 4 2 6 xB 1 + (1 y)2 F2 xB , Q2 ,
dQ2 dW
Q
(20)
where xB = Q2 /(Q2 + W 2 ) and y = (Q2 + W 2 )/S. Using the parameterization [24]

F2 xB , Q2 = c1
1
xB
c2 +c3 ln(1+Q2 /Q2 )

0
(21)
357
(a)
0
Fig. 6. (a) Same as in Fig. 3, but for d /dxB (in nb) with pT > 2.5 GeV. (b) Same as in (a), but for
pT > 3.5 GeV and 2 < Q2 < 8 GeV2 (first frame), 8 < Q2 < 20 GeV2 (second frame), or 20 < Q2 < 70 GeV2
(third frame).
where Q20 = 0.4 GeV2 , c1 = 0.2030 0.0086, c2 = 0.0727 0.0046, and c3 = 0.0448
DIS = (35.4 2.1) nb assuming
0.0012, obtained from a fit to ZEUS data, we thus find tot
the errors on c1 , c2 , and c3 to be statistically independent. This nicely agrees with the result
DIS = 33.9 nb obtained in the parton model of QCD at LO, where [25]
tot

F2 xB , Q2 = xB
nf
p

p
eq2i Fqi xB , Q2 + Fqi xB , Q2 ,
(22)
i=1
using set CTEQ6L1 [20] of proton PDFs with nf = 5. For consistency, we use the ZEUS
DIS to convert the ZEUS data for (1/N ) dN

result for tot
evt
had /dpT [5]. The result for
h
d /dpT thus obtained (solid circles) is compared with our LO (dashed histogram) and
NLO (solid histogram) predictions in Fig. 8 (upper frame). As in Figs. 37, the shaded band
358
(b)
Fig. 6. (continued)
indicates the theoretical uncertainty in the NLO prediction due to the variation described
above, and the K factor is also shown (lower frame). Again, our NLO prediction leads to a
better description of data than our LO one. Here, the K factor takes more moderate values
than under H1 kinematic conditions, being of order 1.5 or below. As explained above, our
LO and NLO predictions break down in the limit pT 0. This drawback can be fixed by
the resummation of multiple parton radiation, as demonstrated in Ref. [26] on the basis of
the LO result.
In Section 2, we explained why the Furry terms do not vanish in our case, in contrast
to inclusive jet production in DIS [13]. It is interesting to investigate their importance
0
quantitatively. To this end, we reconsider the differential cross sections d /dxB for
0.1 < y < 0.6, 4.5 < Q2 < 15 GeV2 , pT > 2.5 GeV, 5 < < 25 , and xE > 0.01 and
d h /dpT for 10 < Q2 < 160 GeV2 , 75 < W < 175 GeV, and xF > 0.05, which we already studied in the second frame of Fig. 6(a) and the first frame of Fig. 8, respectively,
and turn off the Furry terms in our default NLO prediction. The results are shown together
359
Fig. 7. Same as in Fig. 3, but for d /dQ2 (in pb/GeV2 ) with pT > 2.5 GeV.
with our default LO and NLO predictions and the H1 [4] and ZEUS [5] data in Figs. 9(a)
and (b), respectively. We observe that the Furry terms are very important. In Fig. 9(a),
they account for roughly 20% of the NLO correction, while, in Fig. 9(b), they practically
exhaust the latter.
We expect our fixed-order predictions to break down in three extreme kinematic regimes
corresponding to the limits (i) Q2 0; (ii) 0 or, equivalently, or xF 1;
and (iii) xB 0. Case (i) corresponds to the photoproduction limit, in which the resolvedphoton contribution gains importance, especially at small values of pT and/or . Case (ii)
is related to the possibility that the observed hadron h originates from the proton remnant,
so that the notion of fracture functions is invoked. Case (iii) is expected to correspond to the
realm of BFKL [9] dynamics, although it is unclear precisely where the onset of the latter
is supposed to be located. Our analysis is puristic in the sense that resolved virtual photons,
fracture functions, and BFKL dynamics are disregarded, so as to test their actual relevance
in the confrontation of the QCD-improved parton model with the experimental situation
of Refs. [35]. Let us now scrutinize these issues. Doing this, however, we have to bear
in mind that the theoretical uncertainty in our NLO predictions due to the arbitrariness in
the choice of the unphysical scales is particularly large in these corners of phase space, so
that any conclusions are likely to be premature prior to the advent of a full NNLO analysis.
360
Fig. 8. Differential cross section d h /dpT (in nb/GeV) of e p e h + X in DIS with

10 < Q2 < 160 GeV2 and 75 < W < 175 GeV at HERA with Ee = 26.7 GeV and Ep = 820 GeV for charged
hadrons with xF > 0.05. ZEUS data [5] (solid circles) are compared with our default LO (dashed histogram) and
NLO (solid histogram) predictions including theoretical uncertainties due to variation (shaded band). The K
factor (lower frame) is also shown.
0
From Fig. 7, we observe that our NLO prediction for d /dQ2 tends to undershoot the
H1 data [3] in the low-Q2 range, so that there is indeed some room for a resolved-photon
contribution. Similar conclusions were reached in Ref. [27]. On the other hand, we see from
0
Fig. 4 that the H1 data for d /d [3] significantly exceed our NLO prediction in the very
forward region, i.e., in the rightmost bin, for low values of Q2 . In fact, for 2 < Q2 <
4.5 GeV2 , the measured distribution exhibits a plateau in the upper range, whereas
the NLO prediction is rapidly suppressed by the shrinkage of the available phase space
for increasing value of . This plateau might be partly caused by 0 mesons originating
from the remnant jet, which contaminate the proper data sample. Such events cannot be
described within our puristic NLO QCD framework. Finally, thanks to the support from the
Furry terms, we find in Fig. 6(a) and (b) satisfactory overall agreement between our NLO
0
prediction for d /dxB and the H1 data [4] down to the lowest xB values considered.
0
A similar conclusion can be drawn from Fig. 5(b) for d /dxE in the low-xB bin 4.2
105 < xB < 2 104 . This suggests that, in the case of light-hadron inclusive production
361
(a)
(b)
Fig. 9. Same as in (a) the second frame of Fig. 6(a) and (b) the first frame of Fig. 8, but also including our default
NLO predictions with the Furry terms turned off (dotted histograms). For clarity, the theoretical uncertainties due
to variation are omitted.
362
in DIS at HERA, the influence of the BFKL dynamics is likely to be still feeble for xB
4.2 105 .
4. Conclusion
We analytically calculated the cross section for the inclusive electroproduction of single
hadrons with finite transverse momenta via virtual-photon exchange at NLO in the QCDimproved parton model, with nf massless quark flavours, on the basis of the collinearfactorization theorem. We worked in the MS renormalization and factorization scheme and
handled the IR singularities using the dipole subtraction formalism [11]. As for the virtual
correction, we reproduced the result of Ref. [17]. As for the real correction, we established
agreement with Ref. [13], up to the Furry terms, which vanish upon phase-space integration
in the case of single-jet inclusive electroproduction considered in Ref. [13], but yield a
finite contribution in the case under consideration here.
Using nonperturbative FFs recently extracted from data of e+ e annihilation [21], we
provided theoretical predictions for the production of 0 mesons in the forward region
and of charged hadrons in the current-jet region, and compared them in all possible ways
with H1 [3,4] and ZEUS [5] data, respectively. Specifically, we considered cross section
distributions in pT , , xE , xB , and Q2 .
We found that our LO predictions always significantly fell short of the HERA data
and often exhibited deviating shapes. However, the situation dramatically improved as we
proceeded to NLO, where our default predictions, endowed with theoretical uncertainties estimated by moderate unphysical-scale variations, led to a satisfactory description
of the HERA data in the preponderant part of the accessed phase space. In other words,
we encountered K factors much in excess of unity, except towards the regime of asymptotic freedom characterized by large values of pT and/or Q2 . This was unavoidably
accompanied by considerable theoretical uncertainties. Both features suggest that a reliable
interpretation of the HERA data [35] within the QCD-improved parton model ultimately
necessitates a full NNLO analysis, which is presently out of reach, however. For the time
being, we conclude that the successful comparison of the HERA data with our NLO predictions provides a useful test of the universality and the scaling violations of the FFs,
which are guaranteed by the factorization theorem and are ruled by the DGLAP evolution
equations, respectively.
Significant deviations between the HERA data and our NLO predictions only occurred
in certain corners of phase space, namely in the photoproduction limit Q2 0, where
resolved virtual photons are expected to contribute, and in the limit , where fracture functions are supposed to enter the stage. Both refinements were not included in our
analysis. Interestingly, distinctive deviations could not be observed towards the lowest xB
values probed, which indicates that the realm of BFKL [9] dynamics has not actually been
accessed yet.
Note added
After finalizing this manuscript, a paper has appeared which also reports on a NLO
analysis of the inclusive electroproduction of single hadrons with finite transverse momenta
363
[28] reaching conclusions similar to ours. A dedicated comparison with results presented
in Ref. [28] using identical input led to agreement within the numerical accuracy.
Acknowledgements
We thank Gnter Wolf for a clarifying communication [24] regarding the extraction of
d h /dpT from Ref. [5] and Elisabetta Gallo for drawing Ref. [6] to our attention. We
are grateful to Michael Klasen for his collaboration at the initial stage of this work, to
Michael Spira for a beneficial communication regarding the application of the dipole subtraction formalism to the case of two tagged partons, and to Ingo Schienbein and Dominik
Stckinger for helpful discussions. This work was supported in part by the Bundesministerium fr Bildung und Forschung through Grant No. 05 HT4GUA/4 and by the Deutsche
Forschungsgemeinschaft through Grant No. KN 365/3-1.
Appendix A. LO results
In this appendix, we list the LO expressions for HTab and HLab in Eq. (4) pertaining to
processes (5)(7), with ab = qq, qg, gq, respectively. We have
1 + (1 x z)2
,
(1 x)(1 z)
z
qq
HL = 8s CF eq2 Q2 ,
x
1 + (x z)2
qg
HT = 16s CF eq2
,
(1 x)z
1z
qg
,
HL = 8s CF eq2 Q2
x
16s Nc CF eq2 1 2x(1 x) 2z(1 z)
gq
HT =
,
z(1 z)
Nc2 1
qq
HT = 16s CF eq2
gq
HL =
16s Nc CF eq2 Q2 1 x
Nc2 1
(A.1)
where Nc = 3 is the number of quark colours and CF = (Nc2 1)/(2Nc ) = 4/3 is the
eigenvalue of the Casimir operator in the fundamental representation of the QCD gauge
group SU(Nc ).
Appendix B. Real correction: Furry terms

In this appendix, we list the NLO expressions for HTab and HLab in Eq. (4) that originate
from hindered Furry cancellations in the squared matrix elements of processes (12) and
(13), with ab = qq, and of process (15), with ab = qq . We denote the four-momenta
364
of the second and third final-state quarks by pc and pd , respectively, and introduce the
invariants sij = pi pj , where i, j = a, b, c, d with i = j . For given q, pa and pb , we
need to integrate over pc , while pd is fixed through four-momentum conservation to be
pd = q + pa pb pc . We work in the coordinate frame defined in Fig. 1 and parameterize
pc as
Q2
x
c
pc =
(B.1)
(1, cos sin , sin sin , cos ),
2 x(1 x)
where and are the azimuthal and polar angles, respectively. Then, we have
HTF,ab
,L
1x
2
= s2 CF ea eb Q2
z
2
dxc
1z
1
d
d cos hF,ab
T ,L ,
(B.2)
where
F,qq
hT

1
1
sab sac sbd sab sac scd + sab sad sbc 2sab sbc sbd
2
ssab scd (pc q)
2
2
2
2
sab sbd scd + sab scd
sab
scd + sac
sbd sac sad sbc sac sbd

sac sbd scd 3sad sbc sbd sad sbc scd 2sad sbd scd
(pa pb )
+ (pa pd )
(pc pd )
+ (pb pc )
+ (pb pc , pa pd )
(pa pc , pb pa , pc pb )
F,qq
hL
(pb pc , pc pd , pd pb ),

2

1
1
=
sac sbd sac
sab sad sac sad
2
2
sab sac sbd scd (pb q) (pd q)
(sad sbc + sac sbd sab scd )
2

1
1
+
sab scd sab
sab sad sac sad
(pc q)2 (pd q)2
(sab scd sad sbc sac sbd )
1
1
(sac sbd sad sbc + sab scd )

2
(pb q) (pc q)2
2
2
2
2
2
sac
sad sbd sac sad
sbd + sab
sac scd + sab
sad scd + sab sac
sbd

2
sab sad
scd ,
(B.3)
(B.4)

F,qq
hT
365

1
1
sab sac sbd sab sac scd + sab sad sbc 2sab sbc sbd
2
ssab scd (pc q)
2
2
2
2
sab sbd scd + sab scd
sab
scd + sac
sbd sac sad sbc sac sbd

sac sbd scd 3sad sbc sbd sad sbc scd 2sad sbd scd
(pa pb )
(pc pd )
F,qq
hL
+ (pb pc , pa pd ),

2

1
1
1
sac sab sad sac sad
=
2
2
sab scd (pb q) (pd q)
(sac sbd sad sbc sab scd )

1
2
sab sac + sad sac sad (sac sbd sad sbc + sab scd ) .
(pc q)2
(B.5)
(B.6)
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]
J.C. Collins, D.E. Soper, G. Sterman, Adv. Ser. Dir. High Energy Phys. 5 (1988) 1.
H1 Collaboration, C. Adloff, et al., Nucl. Phys. B 485 (1997) 3.
H1 Collaboration, C. Adloff, et al., Phys. Lett. B 462 (1999) 440.
H1 Collaboration, A. Aktas, et al., Eur. Phys. J. C 36 (2004) 441.
ZEUS Collaboration, M. Derrick, et al., Z. Phys. C 70 (1996) 1.
ZEUS Collaboration, J. Breitweg, et al., Eur. Phys. J. C 11 (1999) 251.
A. Mendez, Nucl. Phys. B 145 (1978) 199.
P. Bttner, PhD thesis, University of Hamburg, 1999, Report No. DESY-THESIS 1999-004.
E.A. Kuraev, L.N. Lipatov, V.S. Fadin, Zh. Eksp. Teor. Fiz. 72 (1977) 377, Sov. Phys. JETP 45 (1977) 199;
I.I. Balitsky, L.N. Lipatov, Yad. Fiz. 28 (1978) 1597, Sov. J. Nucl. Phys. 28 (1978) 822.
V.N. Gribov, L.N. Lipatov, Yad. Fiz. 15 (1972) 781, Sov. J. Nucl. Phys. 15 (1972) 438;
G. Altarelli, G. Parisi, Nucl. Phys. B 126 (1977) 298;
Yu.L. Dokshitser, Zh. Eksp. Teor. Fiz. 73 (1977) 1216, Sov. Phys. JETP 46 (1977) 641.
S. Catani, M.H. Seymour, Nucl. Phys. B 485 (1997) 291;
S. Catani, M.H. Seymour, Nucl. Phys. B 510 (1997) 503, Erratum.
P. Aurenche, R. Basu, M. Fontannaz, R.M. Godbole, Eur. Phys. J. C 34 (2004) 277.
S. Catani, M.H. Seymour, in: G. Ingelman, A. De Roeck, R. Klanner (Eds.), Future Physics at HERA,
Proceedings of the Workshop, vol. 1, 19951996, p. 519, hep-ph/9609521.
A. Daleo, C.A. Garcia Canal, R. Sassot, Nucl. Phys. B 662 (2003) 334.
A. Daleo, R. Sassot, Nucl. Phys. B 673 (2003) 357.
G. Passarino, M.J.G. Veltman, Nucl. Phys. B 160 (1979) 151.
D. Graudenz, Phys. Rev. D 49 (1994) 3291.
R.K. Ellis, D.A. Ross, A.E. Terrano, Nucl. Phys. B 178 (1981) 421.
J.A.M. Vermaseren, Symbolic Manipulation with FORM, Computer Algebra Netherlands, Amsterdam,
1991.
J. Pumplin, D.R. Stump, J. Huston, H.-L. Lai, P. Nadolsky, W.-K. Tung, JHEP 0207 (2002) 012.
B.A. Kniehl, G. Kramer, B. Ptter, Nucl. Phys. B 582 (2000) 514.
B.A. Kniehl, G. Kramer, B. Ptter, Phys. Rev. Lett. 85 (2000) 5288.
DELPHI Collaboration, W. Adam, et al., Z. Phys. C 69 (1996) 561;
ALEPH Collaboration, R. Barate, et al., Phys. Rep. 294 (1998) 1.
G. Wolf, private communication.
366
[25]
[26]
[27]
[28]
Particle Data Group, S. Eidelman, et al., Phys. Lett. B 592 (2004) 1.

P.M. Nadolsky, D.R. Stump, C.P. Yuan, Phys. Rev. D 64 (2001) 114011.
M. Fontannaz, Nucl. Phys. B (Proc. Suppl.) 135 (2004) 173.
A. Daleo, D. de Florian, R. Sassot, hep-ph/0411212.
BRST approach to Lagrangian construction

for fermionic massless higher spin fields
I.L. Buchbinder a , V.A. Krykhtin b , A. Pashnev c,
a Department of Theoretical Physics, Tomsk State Pedagogical University, Tomsk 634041, Russia
b Laboratory of Mathematical Physics and Department of Theoretical and Experimental Physics,
Tomsk Polytechnic University, Tomsk 634050, Russia

c Bogoliubov Laboratory of Theoretical Physics, JINR, Dubna 141980, Russia
Abstract
We develop the BRST approach to Lagrangian formulation for all massless half-integer higher
spin fields on an arbitrary dimensional flat space. General procedure of Lagrangian construction
describing the dynamics of fermionic field with any spin is given. It is shown that in fermionic case
the higher spin field model is a reducible gauge theory and the order of reducibility grows with the
value of spin. No off-shell constraints on the fields and the gauge parameters are used. We prove
that in four dimensions after partial gauge fixing the Lagrangian obtained can be transformed to
FangFronsdal form however, in general case, it includes the auxiliary fields and possesses the more
gauge symmetries in compare with FangFronsdal Lagrangian. As an example of general procedure,
we derive the new Lagrangian for spin 5/2 field containing all set of auxiliary fields and gauge
symmetries of free fermionic higher spin field theory.
PACS: 11.10.-z; 11.10.Ef; 11.10.Kk; 11.15.-q
E-mail addresses: joseph@tspu.edu.ru (I.L. Buchbinder), krykhtin@mph.phtd.tpu.edu.ru (V.A. Krykhtin).

Deceased.
doi:10.1016/j.nuclphysb.2005.01.017
368
I.L. Buchbinder et al. / Nuclear Physics B 711 (2005) 367391
1. Introduction
Construction of the self-consistent Lagrangian theory of interacting higher spin fields
is one of the longstanding problems of the theoretical physics. First success in the theory of massless higher spin fields was the formulation of Lagrangians for free bosonic [1]
and fermionic [2] fields in four dimensions. Since then the various approaches to higher
spin fields problem were developed (see, e.g., [3] for reviews and [4,5] for recent developments) however the general problem is still open. We would like to point out two modern
approaches.
An approach, called the unfolded formalism, was developed by Vasiliev et al. (see, e.g.,
[6] and references therein). It allows to construct both the theory of free higher spin fields
and the theory of higher spin fields coupled to AdSD background (see [7] for the bosonic
case and [8] for the fermionic case and references therein). Also this formalism turned out
to be fruitful for constructing the consistent equations of motion for interacting higher spin
fields.
Another approach to higher spin field problem, called BRST1 approach [10], was initiated by development of string field theory where the interacting model of open strings was
constructed (see, e.g., [11]) on the base of BRST techniques.2 Higher spin BRST approach
is analogous to string field theory however it contains two essential differences related with
structure of constraints, which are used in construction of BRST charge, and with presence
only massless fields in the spectrum of higher spin field model. If ones try to consider
the tensionless limit of the string field theory (see, e.g., for free string theory [13,14]) we
expect to get the theory of interacting massless fields. Since the string field theory contains lesser number of constraints on the fields than we need to construct an irreducible
representation of Poincar group, fields in string spectrum do not belong to irreducible
representations with fixed spin and their equations of motion describe rather propagation
of Regge trajectories, instead of one spin mode. For the equations of motion to describe
propagation of one spin mode the additional, in compare with string, off-shell constraints
on the fields must be imposed. In order to get Lagrangian which contains the additional
constraints as equations of motion we have to include these additional constraints into the
set of constraints which is used in constructing the BRST charge and then try to get the Lagrangian of the higher spin field theory. Using this approach one can hope to construct the
theory of interacting higher spin fields analogously to the string field theory. (An attempt
to do that was undertaken in [15].)
The first natural step of constructing the massless higher spin interacting model in the
BRST approach is a formulation of corresponding free model. This problem was studied in
[10] and finally solved for the bosonic massless higher spin fields both on the flat [16,17]
and the AdS [18,19] backgrounds. However, the BRST approach to fermionic fields has
not been developed at all so far.
The present paper is devoted to formulation of BRST approach to derivation the Lagrangian for free fermionic massless higher spin fields on the flat Minkowski space of
1 BRST construction was discovered at first in context of YangMills theories [9].
2 Also we point out the approach [12] to finding the gauge invariant actions for arbitrary representations of the
Poincar group.
369
arbitrary dimension. The method which we use here slightly differs from the one in the
bosonic case, but application of our method in the bosonic case leads to the same final
result for the Lagrangian describing propagation of all spin fields. The difference of the
methods is rather technical and consists in that we do not use the similarity transformation
like in [17] and therefore one can construct Lagrangian for the field of one fixed value of
spin while the approach in [17] demands to use fields of all spins together. As a future purpose we hope, using our method, to construct the free theory of massive higher fermionic
fields (see, e.g., [20] and references therein), to get the Lagrangian describing propagation
of fermionic higher spin fields through AdS background, and to consider an application
of BRST approach to supersymmetric higher superspin models (for recent development in
these directions see, e.g., [21] for massive higher spin field in AdS background and [22]
for supersymmetric higher spin field models).
The paper is organized as follows. In Section 2 we investigate the superalgebra generated by the constraints which are necessary to define a irreducible half-integer spin
representation of Poincar group. It is shown that an naive use of the BRST charge, constructed on the base of these constraints, leads us to the equations of motion only for spin
1/2 fields. We argue, to overcome this difficulty we should reformulate the constraint algebra and find a new representation for the constraints.
Section 3 is devoted to actual formulation of new representation for constraints.
In Section 4 we construct Lagrangian describing a propagation of field with any fixed
half-integer spin. We find that in the case of arbitrary spin fermionic field, the theory has
reducible gauge symmetry with the finite order of reducibility which increases with the
spin value. Next, in Section 5 we construct Lagrangian describing propagation of all halfinteger spin fields simultaneously.
Then in Section 6 we show that in four dimensions found Lagrangian may be transformed, after partial gauge fixing, to the Fang and Fronsdal Lagrangian [2].
In Section 7 we illustrate the general procedure of Lagrangian construction by finding the Lagrangian and gauge transformations for the spin 5/2 field model without gauge
fixing, keeping all auxiliary fields and higher spin gauge symmetries.
This paper is devoted to the memory of our friend and collaborator, a remarkable human
being and scientist A.I. Pashnev who tragically passed away on 30 March 2004.
2. Algebra of the constraints

It is well known that the totally symmetrical tensor-spinor field 1 ...n (the Dirac index
is suppressed), describing the irreducible spin s = n + 1/2 representation must satisfy the
following constraints (see, e.g., [23])
1 ...n = 0,
(1)
2 ...n = 0.
(2)
Here are the Dirac matrices { , } = 2 , = (+, , . . . , ).

In order to describe all higher tensor-spinor fields together it is convenient to introduce
Fock space generated by creation and annihilation operators a+ , a with vector Lorentz
370
index = 0, 1, 2, . . . , D 1 satisfying the commutation relations

a , a+ = .
(3)
These operators act on states in the Fock space

| =
1 ...n (x)a +1 a +n |0
(4)
n=0
which describe all half-integer spins simultaneously if the following constraints are taken
into account
T0 | = 0,
T1 | = 0,
(5)
where T0 = p , T1 = a . If constraints (5) are fulfilled for the general state (4)
then constraints (1), (2) are fulfilled for each component 1 ...n (x) in (4) and hence the
relations (5) describe all free higher spin fermionic fields together. In order to construct
hermitian BRST charge we have to take into account the constraints which are hermitian
conjugate to T0 and T1 . Since T0+ = T0 we have to add only one constraint T1+ = a+ to
the initial constraints T0 and T1 .
Algebra of the constraints T0 , T1 , T1+ is not closed and in order to construct the BRST
charge we must include in the algebra of constraints all the constraints generated by T0 ,
T1 , T1+ . The resulting constraints and their algebra are written in Table 1.
+
The constraints T0 , T1 , T1+ are fermionic and the constraints L0 , L1 , L+
1 , L2 , L2 , G0
are bosonic. All the commutators are graded, i.e., graded commutators between the fermionic constraints are anticommutators and graded commutators which include any bosonic
constraint are commutators. In Table 1 the first arguments of the graded commutators and
explicit expressions for all the constraints are listed in the left column and the second argument of graded commutators are listed in the upper row. It is worth pointing out that this
+
algebra involves all the bosonic constraints L0 , L1 , L+
1 , L2 , L2 , G0 which were used to
describe the irreducible bosonic representation in [16] as a subalgebra.
Table 1
Algebra of the constraints
T0
T1
T1+
L0
L1
L+
1
L2
L+
2
G0
T0
T1+
T1
T0
T1
T0 = p
2L0
2L1
2L+
1
T1 = a
+
T1+ = a
2
L0 = p
L1 = p a
+
L+
1 = p a
1
L2 = 2 a a
1 + +
L+
2 = 2 a a
+ a + D
G0 = a
2
2L1
4L2
2G0
2L+
1
2G0
4L+
2
T0
L0
L+
1
L1
T0
L0
L1
L+
1
T1+
T1
0
L+
1
2L2
0
0
T1
T1+
L1
L1
G0
G0
L+
1
2L2
2L+
2
T1+
2L+
2
371
Let us introduce the BRST charge for the enlarged system of constraints
+
Q = q0 T0 + q1+ T1 + q1 T1+ + 0 L0 + 1+ L1 + 1 L+
1 + 2 L2

+
+
+ +
+ 2 L+
2 + G G0 + i 1 q1 1 q1 p0 i G q1 + 2 q1 p1

+ i G q1+ + 2+ q1 p1 + q02 1+ 1 P0 + 2q1 q1+ 2+ 2 PG

+ G 1+ + 2+ 1 2q0 q1+ P1 + 1 G + 1+ 2 2q0 q1 P1+

+ 2 G 2+ q1+2 P2 + 2 2 G q12 P2+ .
(6)
Here q0 , q1+ , q1 are the bosonic ghosts corresponding to the fermionic constraints T0 , T1 ,
T1+ respectively and 0 , 1+ , 1 , 2+ , 2 , G are fermionic ghosts corresponding to the
bosonic constraints. The momenta for these ghosts are p0 , p1 , p1+ for bosonic and P0 , P1 ,
P1+ , P2 , P2+ , PG for fermionic ones. They satisfy the usual commutation relations

{0 , P0 } = {G , PG } = 1 , P1+ = 1+ , P1 = 2 , P2+ = 2+ , P2 = 1,
(7)
+

+
[q0 , p0 ] = q1 , p1 = q1 , p1 = i
(8)
and act on the vacuum state as follows
p0 |0 = q1 |0 = p1 |0 = P0 |0 = PG |0 = 1 |0 = P1 |0 = 2 |0 = P2 |0 = 0. (9)
The BRST charge (6) acts in enlarged space of state vectors depending both on a + and
on the ghost operators q0 , q1+ , p1+ , 0 , G , 1+ , P1+ , 2+ , P2+ and having the structure

k k
k k k k
| =
(q0 )k1 q1+ 2 p1+ 3 (0 )k4 (G )k5 1+ 6 P1+ 7 2+ 8 P2+ 9
ki
...k9
a +1 a +k0 k11 ...
(x)|0.
k
0
(10)
The corresponding ghost number is 0. The sum in (10) is assumed over k0 , k1 , k2 , k3

running from 0 to infinity and over k4 , k5 , k6 , k7 , k8 , k9 running from 0 to 1. It is evident
that the state vectors (4) are the partial cases of the above vectors.
The physical states are defined in the BRST approach by the equation
Q | = 0
(11)
which is treated as an equation of motion. Besides, if | is a physical state, then due to

nilpotency of the BRST operator, the state | + Q | will also be physical for any |.
It means we have the gauge transformations
| = Q |.
(12)
Let us decompose the BRST charge Q (6), the state vector | and the parameter of
the gauge transformations | as follows

0 , Q0 ] = 0,
0 + 2q + q1 + 2 PG ,
Q = Q0 + G G
(13)
[G
1
2
+
+
+
+
+
+
G = G0 iq1 p1 + iq1 p1 + 1 P1 1 P1 + 22 P2 22 P2 ,
(14)
| = |0 + G |G ,
(15)
372
| = |0 + G |G .
(16)
Here the state vectors |0 , |G and the gauge parameters |0 , |G are independent of
G . Then the equations of motion and the gauge transformations take the form

Q0 |0 + 2q1+ q1 2+ 2 |G = 0,
(17)
0 |0 Q0 |G = 0,
G
(18)
+
+
|0 = Q0 |0 + q1 q1 2 2 |G ,
(19)
0 |0 Q0 |G .
|G = G
(20)
Now ones try to simplify these equations. First, we decompose the
| and
state vector
0 : | = |n , | = |n ,
the gauge parameter | in the eigenvectors of operator G
0 |n = (n + D4 )|n , n = 0, 1, 2, . . . . Then using the
0 |n = (n + D4 )|n , G
with G
2
2
1
gauge transformation we can make all |Gn = 0 choosing |n = n+(D4)/2
|G except
the case n + D4
2 = 0. When D = 4 we have n = 0 and the field |G after this gauge
transformation is reduced to
|G |G0 = G (x)|0,
(21)
i.e., it contains only spin 1/2 field. Substituting (21) in the equations of motion (17), (18)
ones get
Q0 |0n = 0,
n=0

n=0
(22)
D4
n+
|0n = Q0 |G0 .
2
0 on both sides of Eq. (23) ones obtain

Acting by the operator G

D4 2
|0n = 0.
n+
2
(23)
(24)
n=0
Since all the states vectors |0n are linear independent, Eq. (24) means that all |0n = 0,
except n = D4
2 = 0. Thus analogously to (21) one can write
|0 |00 = (x)|0.
(25)
Ultimately we have two independent equations of motion

T0 |00 = 0,
T0 |G0 = 0
(26)
both for |00 and |G0 . These equations in component form read
p (x) = 0,
p G (x) = 0.
(27)
So, we see that the above construction leads to double number of equations and only for
spin 1/2 fields. Hence, such a procedure is unsatisfactory.
To clarify a situation we pay attention to two points.
373
First, if we suppose that the state vectors and the gauge parameters do not depend on the
ghost field G then we have only one Dirac equation and avoid the doubling the physical
component states.
Second, the above construction has led us to the equations only for spin 1/2 fields. This
0 has the structure
happens because of G
0 = N 0 + D 4 ,
G
2
(28)
where N 0 is proportional to the particle number operators associated with the operators
a + , q1+ , p1+ , 1+ , P1+ , 2+ , P2+ and therefore if we want G0 to be considered as a constraint, we get that there are no particles (in D = 4 case) in the physical states and hence
only equations of motion for the field with spin 1/2 arise.
another constraint G
0 + h with h being an
We note that if we had instead of constraint G
arbitrary constant, we could get equations of motion for fields of any spin by choosing the
arbitrary parameter h in the proper way for each spin. As a result, we could put n to any inD4
teger number since instead of condition n + D4
2 = 0 we had condition n + 2 + h = 0.
However, if we simply change the constraint G0 G0 + h we break the algebra of the
constraints which is given in Table 1. Thus, introducing of this arbitrary parameter h must
be carry out in such a way that the algebra of the constraints will not be broken. This discussion shows that the representation for the constraints we used is too naive and improper
and we have to find another representation.
Such a new representation may be realized as follows. We enlarge the number of creation and annihilation operators and extend the expressions for the constraints using the
prescription: the new expressions for the constraints should have the general structure
new constraint = old constraint + additional part,
(29)
with some additional parts which will be found in Section 3 in explicit form. This new
representation must be constructed in such a way that the arbitrary parameter h appears in
constraint G0new as follows
G0new = G0 + Nadd + C + h,
(30)
where Nadd is proportional to the number operators of additional particles, associated

with extra annihilation and creation operators, C is some fixed constant which may arise.
It is evident that we must construct these additional parts first of all for those constraints
whose commutators give G0 . The corresponding operators are T1 , T1+ , L2 , L+
2 . Then we
must go on and construct additional parts for those constraints whose commutators give T1 ,
+
+
T1+ , L2 , L+
2 . Fortunately, they are G0 , T1 , T1 , L2 , L2 and we may construct no additional
parts for the other operators. Note that these operators form a subalgebra. Thus we have to
construct a representation for the subalgebra of the constraints G0 , T1 , T1+ , L2 , L+
2 only.
In the next section we describe construction of a representation for such an subalgebra.
Certainly, we can construct additional parts for all the operators of the algebra given in
Table 1, but in this case we must use some massive parameter which is absent in the true
massless theory. Of course, one can try to introduce such a massive parameter to constraints
by hand. However in this case we expect to get a massive higher spin field theory [24].
374
3. New representation of the constraints

In this section we construct a new representation for the algebra of the constraints so
that the new expressions for the constraints have the structure (29) and the parameter h
appears in the constraint G0 in the proper way (30). Algebra of the new constraints still
has the form given by Table 1. As was explained at the end of the previous section, for this
purpose it is enough to construct the additional parts only for G0 , T1 , T1+ , L2 , L+
2 and the
new expressions for the constraints should be
+
+
L+
2new = L2 + L2add ,
+
T1new
= T1+
+
+ T1add
,
L2new = L2 + L2add ,
(31)
T1new = T1 + T1add ,
(32)
G0new = G0 + G0add
(33)
+
(all the other constraints do not change). Since the constraints G0new , T1new , T1new
, L2new ,
+
L2new form a subalgebra and since the old and additional expressions of the constraints
+
commute, the G0add , T1add , T1add
, L2add , L+
2add form a subalgebra too with the same commutation relation among them as for the old expressions for the constraints G0 , T1 , T1+ , L2 ,
L+
2 . Thus it is enough to find a representation of the subalgebra in terms of new creation
and annihilation operators which will be introduced later.
Let us turn to the construction of the subalgebra representation.
Note that the commutation relations between G0 and the other operators of the subalgebra resemble the commutation relations between a number operator and creation and
annihilation operators. Therefore let us consider the representation of the subalgebra of the
constraints with the state vector |0V annihilated by the operators T1 and L2
T1 |0V = L2 |0V = 0
(34)
and being the eigenvector of the operator G0

G0 |0V = h|0V ,
(35)
where h is an arbitrary constant.3 It is the relation (35) where the vacuum state |0V is an
eigenvector of the number operator G0 with eigenvalue h gives us the desired structure
of the operator G0new (30). Since (T1+ )2 = 2L+
2 we can choose the basis vectors in this
representation as follows
m
m
|0, mV = L+
(36)
|0V ,
|1, mV = T1+ L+
|0V .
2
2
The next step is to find the action of the operators T1 , T1+ , L2 , L+
2 , G0 on the basis
vectors (36). The result is
L+
2 |0, nV = |0, n + 1V ,

L2 |0, nV = n2 n + nh |0, n 1V ,
(37)
L+
2 |1, nV = |1, n + 1V ,
2

L2 |1, nV = n + nh |1, n 1V ,
(38)
3 The representation which is given by (34) and (35) is called in the mathematical literature the Verma module.
It explains the subscript V at the state vectors.
T1+ |0, nV = |1, nV ,
375
T1+ |1, nV = 2|0, n + 1V ,
(39)
T1 |0, nV = n|1, n 1V ,
T1 |1, nV = 2(n + h)|0, nV ,
(40)
G0 |0, nV = (2n + h)|0, nV ,
G0 |1, nV = (2n + 1 + h)|1, nV . (41)
Now, in order to construct the new representation for the subalgebra ones introduce the
additional creation and annihilation operators. The number of pairs of these operators is
equal to the number of the mutually conjugate pairs of the constraints. So we introduce one
pair of fermionic d + , d (corresponding to the constraints T1+ , T1 ) and one pair of bosonic
b+ , b (corresponding to the constraints L+
2 , L2 ) creation and annihilation operators with
the standard commutation relations

b, b+ = 1.
d, d + = 1,
(42)
Making use of the map of the basis vectors (36) and the basis vectors of the Fock space of
the operators d + , b+
n
|0, nV b+ |0 |n,
|1, nV d + |n
(43)
we can construct a representation of the subalgebra. From (37)(41) and (43) ones find

+
L2add = b+ b + d + d + h b,
L+
(44)
2add = b ,
+

+
+
+
+
T1add = 2 b b + h d d b,
T1add = 2b d + d ,
(45)
G0add = 2b+ b + d + d + h.
(46)
It is easy to see, the operators (44), (45) are not hermitian conjugate to each other
+
(T1add )+ = T1add
,
(L2add )+ = L+
2add
(47)
if we use the usual rules for hermitian conjugation of the additional creation and annihilation operators
(d)+ = d + ,
(b)+ = b+ .
(48)
The reason is that the map (43) does not preserve the scalar product. If we have two
vectors |1 V and |2 V and corresponding them two vectors in the Fock space |1
|1 V , |2 |2 V then in general
V 1 |2 V
= 1 |2 ,
(49)
where we assumed that V 0|0V = 1. In order to improve the situation we change the scalar
product in the Fock space so that
V 1 |2 V
= 1 |2 new = 1 |K|2 ,
(50)
with some operator K.

This operator may be found as follows. If we have a map between two bases
|nV |n,
then
|V =
cn |nV
(51)
cn |n = |.
(52)
376
Therefore if we preserve the scalar product for the basis vectors V m|nV = m|K|n then
it will be preserved for all vectors.
In the case of orthogonal basis in the Fock space, m|n = Cn mn with Cn being some
constants (as we have in our case) it may be proved by direct substitution to (50) that the
operator K is

V m|nV
n|.
K=
|m
(53)
Cm Cn
Hence, in the case under consideration we get
K=

1
|n n|C(n, h) 2d + |n n|dC(n + 1, h) ,
n!
(54)
n=0
C(n, h) =
n1

(k + h),
C(0, h) = 1.
(55)
k=0
It is a simple exercise to check that operators (44) and (45) are now mutually conjugate in
the following sense
+ +
KT1add = T1add
(56)
K,
+ +
KL2add = L2add K.
(57)
Thus we have found the new representation of the algebra of constraints which is given
by (31)(33) with (44)(46). (Remind that all the other constraints of the algebra do not
change.) Since the algebra of the constraints have not been changed, a new BRST charge
is constructed substituting in (6) the new constraints instead of old. As a result ones get

= q0 T0 + q + T1 2 b+ b + h d bd + + q1 T + + 2b+ d + d +
Q
1
1
+

+

+
+
+
+ 0 L0 + 1+ L1 + 1 L+
1 + 2 L2 + b b + h + d d b + 2 L2 + b

+ G G0 + 2b+ b + d + d + h + i 1+ q1 1 q1+ p0 i G q1 + 2 q1+ p1+

+ i G q1+ + 2+ q1 p1 + q02 1+ 1 P0 + 2q1 q1+ 2+ 2 PG

+ G 1+ + 2+ 1 2q0 q1+ P1 + 1 G + 1+ 2 2q0 q1 P1+

+ 2 G 2+ q1+2 P2 + 2 2 G q12 P2+ .
(58)
Let us notice that the new BRST charge (58) is selfconjugate in the following sense
+ K = K Q,
Q
(59)
with operator K (54). Now we turn to the construction of the Lagrangians for free fermionic higher spin fields.
4. Lagrangians for the free fermionic fields of single spin

In this section we construct the Lagrangians for free higher spin fermionic fields using the BRST charge (58). Unlike the bosonic case [17,18] we use here slightly another
procedure.
377
First, let us extract the dependence of the new BRST charge (58) on the ghosts G , PG

= Q + G ( + h) + 2q + q1 + 2 PG ,
Q
(60)
1
2
+
+
2
[Q, ] = 0
Q = 2 2 2q1 q1 ( + h),
(61)
with
= G0 + 2b+ b + d + d iq1 p1+ + iq1+ p1 + 1+ P1 1 P1+
(62)
+ 22+ P2 22 P2+ ,

+
2
+
+
+
+
Q = q0 T0 2q1 P1 2q1 P1 + i 1 q1 1 q1 p0 + 0 L0 + q0 1 1 P0

+ q1+ T1 2 b+ b + h d bd + + q1 T1+ + 2b+ d + d + + 1+ L1 + 1 L+
1

+ +
+
+
i
+ 2+ L2 + b+ b + h + d + d b + 2 L+
+
b
q
p
+
i
q
p
2 1 1
2
2 1 1
+ 2+ 1 P1 + 1+ 2 P1+ 2q1+2 P2 2q12 P2+ .
(63)
Second, in order to avoid the doubling the physical component states as it was in Section 2, Eq. (26) we suppose that the state vectors are independent of G , i.e., PG | = 0.
Now the general structure of the states looks like

k k
k k k k
(q0 )k1 q1+ 2 p1+ 3 (0 )k4 d + 5 1+ 6 P1+ 7 2+ 8
| =
ki
k k
10
P2+ 9 b+ 10 a +1 a +k0 k11...k
...k (x)|0.
0
(64)
The corresponding ghost number is 0, as usual. The sum in (64) is assumed over k0 , k1 , k2 ,
k3 , k10 running from 0 to infinity and over k4 , k5 , k6 , k7 , k8 , k9 running from 0 to 1.
After this assumption the equation on the physical states in the BRST approach
Q|
= 0 yields two equations
Q| = 0,
(65)
( + h)| = 0.
(66)
Eq. (66) is the eigenvalue equation for the operator (62) with the corresponding eigenvalues h
D4
, n = 0, 1, 2, . . . .
(67)
2
The numbers n are related with the spin s of the corresponding eigenvectors as s = n + 1/2.
Let us denote the eigenvectors of the operator corresponding to the eigenvalues n + D4
2
as |n

D4
|n .
|n = n +
(68)
2
h = n +
Then solutions to the system of Eqs. (65), (66) are enumerated by n = 0, 1, 2, . . . and
satisfy the equations
Qn |n = 0,
(69)
378
where in the BRST charge (63) we substitute n + D4

2 instead of h for each given equation on spin s = n + 1/2 field. Thus we get that the BRST charge depends on n

Qn = q0 T0 2q1+ P1 2q1 P1+ + i 1+ q1 1 q1+ p0 + 0 L0 + q02 1+ 1 P0

+ q1+ T1 2b+ bd bd + + q1 T1+ + 2b+ d + d + + 1+ L1 + 1 L+
1
+

+

+
+ +
+
+
+
+ 2 L2 + b b + d d b + 2 L2 + b i2 q1 p1 + i2 q1 p1
+ 2+ 1 P1 + 1+ 2 P1+ 2q1+2 P2 2q12 P2+

D4
.
+ 2q1+ d 2+ b n +
2
(70)
Let us rewrite the operators Qn (70) in the form independent of n. This may be done by
replacing n + D4
2 in (70) by the operator (62). Then we obtain

Q = q0 T0 2q1+ P1 2q1 P1+ + i 1+ q1 1 q1+ p0 + 0 L0 + q02 1+ 1 P0

+ q1+ T1 2b+ bd bd + + q1 T1+ + 2b+ d + d + + 1+ L1 + 1 L+
1

+ +
+
+
i
+ 2+ L2 + b+ b + d + d b + 2 L+
+
b
q
p
+
i
q
p
2 1 1
2
2 1 1

+ 2+ 1 P1 + 1+ 2 P1+ 2q1+2 P2 2q12 P2+ + 2q1+ d 2+ b ,
(71)
where Q = Qn |n+ D4 . Operator (71) analogous to the BRST operator which obtained
2
in the bosonic case [17,18] after the dependence on the ghost fields G , PG was removed.
Now we can rewrite the set of Eqs. (69) in the equivalent form as one equation for all
half-integer spins. Since the operators Q and commute then vectors Q |n belong to
different eigenvalues of the operator and consequently are linear independent. Therefore
we may write the set of Eqs. (69) as one equation summing them
Q |n = Q
n=0
|n = Q | = 0,
(72)
n=0
where we denote
| =
|n .
(73)
n=0
Thus we obtain that the equation

Q | = 0
(74)
with | defined by (73) describes propagation of all half-integer spin fields simultaneously.
Let us turn to the gauge transformations. Analogously we suppose that the parameters
of the gauge transformations are also independent of G . Due to Eq. (61) we have the
following tower of the gauge transformations and the corresponding eigenvalue equations
for the gauge parameters
| = Q|,

| = Q(1) ,
( + h)| = 0,

( + h)(1) = 0,
(75)
(76)

(i) = Q(i+1) ,

( + h)(i+1) = 0.
379
(77)
where h has already been determined for each spin. Since the ghost number of the gauge
parameters is reduced with the stage of reducibility gh(|(i) ) = (i + 1) we get that for
each n (and for the spin s = n + 1/2, respectively) the tower of the gauge transformations
must be finite. Thus in case of fermionic higher spin fields we have gauge symmetry with
reducible generators.
Doing the same procedure as above for the equations of motion we may write the gauge
transformations (75)(77) for each given spin

D4
|n ,
|n = n +
|n = Qn |n ,
(78)
2

D 4 (i)
n,
(i) n = n +
|n = Qn (1) n ,
(79)
2

(i) n = Qn (i+1) n
(80)
and for all half-integer spins simultaneously
| = Q |,
| =
|n ,
(81)
n=0

| = Q (1) ,

(i) = Q (i+1) ,
(82)
(i)
(i)
=

.
n
(83)
n=0
Next step is to extract the zero ghost mode from the operator Q (71). This operator
has the structure

Q = 0 L0 + q02 1+ 1 P0 + q0 T0 2q1+ P1 2q1 P1+

+ i 1+ q1 1 q1+ p0 + Q ,
(84)
where Q is independent of 0 , P0 , q0 , p0 . Also we may decompose the state vector and
the gauge parameters as
| =

q0k 0k + 0 1k ,
k=0
(i)
=
(i)k
(i)k
q0k 0 + 0 1 ,

gh mk = (m + k),
(85)

gh (i)km = (i + k + m + 1).
(86)
k=0
Following the procedure described in [14] we get rid of all the fields except two |00 , |01
and Eq. (74) is reduced to
1

Q 00 + T0 , 1+ 1 01 = 0,
2

T0 00 + Q 01 = 0,
(87)
(88)
380
where T0 = T0 2q1+ P1 2q1 P1+ , and {A, B} = AB + BA.

To be complete we show how Eqs. (87) and (88) can be derived from the (74). First we
extract the zero ghost modes from the BRST charge Q (84), the state vector | (85) and
the gauge parameter | (86). After this the gauge transformation for the fields |0k , k 2
are

0k = Q k0 1+ 1 k1 + (k + 1) 1+ q1 1 q1+ k+1
0

k2
+ 1 .
+ T0 k1
(89)
0
We see that we can make all fields |0k , k 2, to be zero using the gauge parameters |k1 .
Second step is to consider the equations of motion at coefficients (q0 )k , k 3. Taking
into account that all fields |0k = 0, k 2, these equations are reduced to
k2

= + 1 k ,
1
k 3,
(90)
and we find that all |1k = 0, k 1.

Finally we consider the equation at coefficient (q0 )2 and express |10 field from |01
0

= T0 1 .
1
(91)
So it remained only two fields |00 and |01 and the independent equations of motion for
them are (87) and (88).
Since the operators Q , T0 , {T0 , 1+ 1 } commute with the operator , then from
Eqs. (87), (88) we may get equations of motion for fixed spin fields

1
Q 00 n + T0 , 1+ 1 01 n = 0,
2
0

T0 0 n + Q 01 n = 0.
(92)
(93)
These field equations can be deduced from the following Lagrangian4

1
Ln = n 00 Kn T0 00 n + n 01 Kn T0 , 1+ 1 01 n
2

+ n 00 Kn Q 01 n + n 01 Kn Q 00 n ,
(94)
where the standard scalar product for the creation and annihilation operators is assumed.
This Lagrangian is now written for fields with given spin which are defined by n chosen
according to (66), (67)

00 n = n + (D 4)/2 00 n ,
(95)
01 n = n + (D 4)/2 01 n
and the operator Kn is the operator K (54) where the following substitution is assumed be
done h (n + (D 4)/2)
4 The Lagrangian is defined up as usual to an overall factor.

D4
1
|k k|C k, n
Kn =
k!
2
k=0

D4
+
.
2d |k k|dC k + 1, n
2
381
(96)
Thus the operators Kn depend on the spin of the fields. Note also that we can write
Qn instead of Q in the equations of motion (92), (93), in the Lagrangian (94) and in
the gauge transformations (97)(100) for fixed spin fields below.
The equations of motion (92), (93) and the Lagrangian (94) are invariant under the
gauge transformations

1
00 n = Q 00 n + T0 , 1+ 1 10 n ,
2

01 n = T0 00 n + Q 10 n ,
which are reducible
(i)0
(i+1)0
1 + (i+1)1
0 n = Q
0 n + 2 T0 , 1 1
0 n,

(i)10 n = T0 (i+1)00 n + Q (i+1)10 n ,
(97)
(98)
(0)0
0

0 n = 0 n , (99)
(0)1
1

0 n = 0 n , (100)
with finite number of reducibility stages imax = n 1 for spin s = n + 1/2. In Section 6 we
show that the Lagrangian (94) is transformed to the Fang and Fronsdal Lagrangian [2] in
four dimensions after eliminating the auxiliary fields. So, we construct the Lagrangian for
arbitrary fixed spin fermionic fields using the BRST approach. Now we turn to construction
of Lagrangian describing propagation of all half-integer spin fields simultaneously.
5. Lagrangian for all half-integer spin fields

In this section we construct the Lagrangian which describes all half-integer spin fields
simultaneously, i.e., we construct the Lagrangian in terms of the fields containing all halfinteger spins
i
i
=
,
0
0 n
i = 0, 1.
(101)
n=0
As we mentioned above the operator commutes with the operators Q , T0 , {T0 , 1+ 1 }

and moreover it commutes with each term of these operators. Therefore we can write all
the operators Kn in the Lagrangians (94) in the same form for any spin. This is done
analogously to that when we transformed Qn into Q . Namely, we stand all h to the right
(or to the left position) in the expression for K (54) and substitute instead of h. As a
result we have

1
|n n|C(n, ) 2d + |n n|dC(n + 1, ) .
K =
n!
n=0
(102)
382
Thus we can substitute K (102) instead of Kn (96) in the expression for the Lagrangian
corresponding to one fixed spin field (94).
Evidently that Lagrangian describing all half-integer spin fields simultaneously should
be a sum of all the Lagrangians for each spin (94)

0
0
1 1
+ 1

Ln =
L=
n 0 K T0 0 n + n 0 K T0 , 1 1 0 n
2
n=0
n=0

+ n 00 K Q 01 n + n 01 K Q 00 n .
(103)
Now we rewrite this Lagrangian in terms of two concise fields |00 and |01 (101)

containing all half-integer spin fields. Using that n 0i |0i n ii nn , we transform each
term in Lagrangian (103) as

0

0

0
0

K T0

=
= 0 K T0 0
(104)
n K T0
n
0
0 n
n=0
n=0
0 n
n=0
and find
1

L = 00 K T0 00 + 01 K T0 , 1+ 1 01
2

+ 00 K Q 01 + 01 K Q 00 ,
(105)
where we have used (101). Thus we have constructed the Lagrangian describing propagation of all half-integer spin fields simultaneously (105), the equations of motion which are
derived from it are (87), (88).
Let us turn to the gauge transformations for the fields (101). Summing up the gauge
transformation for the fields of fixed spins (97), (98) over all n we get the gauge transformations for the fields containing all half-integer spins (101)
1

00 = Q 00 + T0 , 1+ 1 10 ,
2

01 = T0 00 + Q 10 ,
which are also reducible

1

(i)00 = Q (i+1)00 + T0 , 1+ 1 (i+1)10 ,
2
(i)1
(i+1)0
(i+1)1

0 = T0
0 + Q
0 ,
(106)
(107)
(0)0 0

= ,
0
(108)
(109)
(0)1 1

= ,
where we introduced
(j )i

=
n=i+j +1
(j )i

0 n,
i = 0, 1, j = 0, 1, . . . .
(110)
Since the fields (101) contain infinite number of spins and since the order of reducibility
grows with the spin value, then the order of reducibility of the gauge symmetry for fields
(101) will be infinite.
383
It should be noted that the procedure developed here for constructing the Lagrangians
both for fixed spin fields (94) and for all half-integer spin fields (105) may be used for constructing Lagrangians for bosonic fields as well. Also this procedure may be generalized
for constructing Lagrangians with mixed symmetry tensor-spinor fields as it was done in
the bosonic case [17].
Now we turn to the reduction of (94) to the Fang and Fronsdal Lagrangian [2].
6. Reduction to Fang and Fronsdal Lagrangian

In this section we show that Lagrangian (94) is transformed to the Fang and Fronsdal
Lagrangian [2] in four dimensions after elimination of the auxiliary fields.
Let us consider Lagrangian (94) with some fixed n. In this case the gauge symmetry
(99), (100) is reducible with imax = n 1. We can write down the dependence of the fields
and the gauge parameters on the ghost fields explicitly. For the lowest gauge parameters
we have
(n1)0

+ n1 +

(111)
P1 |0 + p1+ |1 0 ,
0 n = p1
(n1)1

(112)
0 n = 0.
Here we have taken into account that the gauge parameters are the eigenvectors of the
operator with the eigenvalue (n + D4
2 ) and that they have the ghost numbers n and
(n + 1), respectively. Let us recall that the subscripts at the state vectors are associated
with the eigenvalues of the corresponding state vectors (68).
With the help of these parameters we can get rid of the dependence on the ghost P2+
(n2)0
(n2)1
in the parameters |
0 n . (The parameter |
0 n has no dependence on the ghost
+
P2 .) We may go on and get rid of any dependence on the ghost P2+ in all the fields and
the parameters. The restriction which appears on the fields and the parameters is that they
can depend on the ghost p1+ maximum in the first power. (They must be annihilated by the
operator q12 .) Due to this restriction the remain gauge symmetry is reducible with the first
stage reducibility.
Then we can get rid of the dependence on the ghost 2+ in all the remain fields and the
gauge parameters. The restriction appeared is that the fields and the gauge parameters must
be annihilated by the operator L2 L2 + (b+ b + h)b + d + db + 1 P1 + iq1 p1 .
After this we get rid of the gauge transformation parameter |10 n with the help of
(1)0
the parameter | 0 n . Now we write down the remain fields and the gauge parameter
explicitly
0
= | n + + P + |1 n2 + q + p + |2 n2 + p + + |3 n2
0 n
1 1
1 1
1 1
+ q1+ P1+ |4 n2 + q1+ p1+ 1+ P1+ |5 n4 + q1+2 p1+ P1+ |6 n4 ,
1
= P + |n1 + p + |1 n1 + p + + P + |2 n3 + q + p + P + |3 n3 ,
0 n
1
1
1 1 1
1 1 1
0
= P + | n1 + p + |1 n1 + q + p + P + |2 n3 + p + + P + |3 n3
0 n
1
1
1 1 1
1 1 1
(113)
(114)
(115)
384
with the restriction L2 |00 n = L2 |01 n = L2 |00 n = 0. Here |i k , |i k , |i k do not
depend on the ghost fields. Using the gauge transformations we first get rid of the fields
|2 n2 , |4 n2 , |6 n4 after which the gauge parameters are restricted by T1 |1 n1 =
T1 | n1 = T1 |2 n3 = 0, with T1 T1 2(b+ b + h)d d + b. Now we can see that
|3 n3 = 0 and then |5 n4 = 0 as the equation of motion. Then we eliminate one after
another the fields |2 n3 , |3 n2 and |1 n1 . The new restrictions on the gauge parameters are T0 |3 n3 = L1 |1 n1 = T0 |1 n1 = 0. The remain gauge freedom is enough
to get rid of the dependence on b+ and d + in | n , |1 n2 and |n1 . After this the
remain equations of motion and the gauge transformation are
T0 |0 n + L+
1 |0 n1 = 0,
T1 |0 n = |0 n1 ,
(116)
T1 |0 n1 = 2|10 n2 ,
(117)
T0 |10 n2 + L1 |0 n1 = 0,
T1 |10 n2 = 0,
(118)
|0 n1 = T0 |0 n1 ,
|10 n2 = L1 |0 n1 ,
(119)
|0 n1 = T0 |0 n1 ,
|10 n2 = L1 |0 n1 .
(120)
T0 |0 n1 L1 |0 n + L+
1 |10 n2
= 0,
Here subscript 0 means that the corresponding fields and the gauge parameter do not depend on b+ and d + . Besides, the gauge parameter is restricted as in the Fang and Fronsdal
theory T1 |0 n1 = 0. The equations which stand in the left column of (116)(118) can be
derived from the Lagrangian

L = n 0 | T0 |0 n + L+
1 |0 n1

n1 0 | T0 |0 n1 L1 |0 n + L+
1 |10 n2

n2 10 | T0 |10 n2 + L1 |0 n1 .
(121)
Using the restrictions on the fields which stand in the right column of (116), (117) we
can express the fields |10 n2 and |0 n1 through |0 n and substitute them in the Lagrangian (121). As a result ones get the Lagrangian which is generalization of the Fang
and Fronsdal Lagrangian [2] for arbitrary dimensional spacetime

+
+
+
L = n 0 | T0 T1+ T0 T1 L+
2 T 0 L 2 + T 1 L 1 + L 1 T 1 + L2 L 1 T 1

+ T1+ L+
(122)
1 L2 |0 n
with the vanishing triple -trace (T1 )3 |0 n = 0 and gauge transformation (119) with the
constrained gauge parameter T1 |0 n1 = 0.
To see that the Lagrangian (122) indeed coincides with the Lagrangian of Fang and
Fronsdal [2] we calculate it (122) explicitly for an arbitrary spin field s = n + 1/2
1
|0 n = a +1 a +n h1 ...n (x)|0.
n!
(123)
First we find that

n 0 |T0 |0 n
p h,
= (1)n h/
+
n 0 |T1 T0 T1 |0 n
= (1)n nh p
/ h ,
(124)
n(n 1)
/h ,
hp
4
+
n
h,
n 0 |T1 L1 |0 n = (1) nhp
+
n 0 |L2 T0 L2 |0 n
= (1)n
+
n 0 |L1 T1 |0 n
= n 0 |T1+ L1 |0 n
+
n n(n 1)
h p h ,
n 0 |L2 L1 T1 |0 n = (1)
2
+ +
+
n 0 |T1 L1 L2 |0 n = n 0 |L2 L1 T1 |0 n ,
385
(125)
(126)
(127)
where we have used the notation of [2]. Substituting the found relations in (122) ones get

1
ph + nh p
/ h n(n 1)h p
/ h n(h p h + H.c.)
L = (1)n h/
4
1
+ n(n 1)(h p h + H.c.) .
(128)
2
The Lagrangian (128) coincides in D = 4 with the Lagrangian of Fang and Fronsdal up
to overall factor (1)n . It can be treated as FangFronsdal Lagrangian for arbitrary Ddimensional space.
Thus we showed that Lagrangian (94) is reduced to the Fang and Fronsdal Lagrangian
(122) with all necessary conditions on the field and the gauge parameter.
We point out that from the Lagrangian (122) we may get one Lagrangian describing
propagation of all half-integer spin fields simultaneously. Summing up the Lagrangians
(122) over all half-integer spins and noticing that n 0 |0 n nn we get this Lagrangian
in the form

+
+
L = 0 | T0 T1+ T0 T1 L+
2 T0 L2 + T 1 L1 + L1 T1

+ +
+ L+
(129)
2 L1 T1 + T1 L1 L2 |0 ,
where we used the notation
|0 =
|0 n .
(130)
n=0
In the next section, we apply our procedure for derivation a new Lagrangian for spin
5/2 field model, where, unlike FangFronsdal Lagrangian, all auxiliary fields stipulated
by general Lagrangian construction, are taken into account.
7. Construction of the Lagrangian for field with spin 5/2

In this section we show how the generic Lagrangian construction (94), given in terms of
abstract state vectors, is transformed to standard spacetime Lagrangian form. We explicitly
derive a Lagrangian for the field with spin-5/2 which contains the auxiliary fields and
more gauge symmetries in compare with FangFronsdal Lagrangian. Of course, it can
be reduced to FangFronsdal Lagrangian after partial gauge-fixing and putting D = 4.
However, this new Lagrangian possesses the interesting properties, in particular it has a
reducible gauge symmetry.
386
Let us start. Since s = n + 1/2 = 5/2 we have n = 2 and h = D2 (67). Then we first
extract the ghost fields dependence of the fields and the gauge parameters
0
= | 2 + + P + |1 0 + q + p + |2 0 + p + + |3 0 + q + P + |4 0 ,
(131)
0 2
1 1
1 1
1 1
1 1
1
+
+
+
= P |1 + p |1 1 + P |4 0 ,
(132)
0 2
1
1
2
0
= P + | 1 + p + |1 1 + P + |4 0 ,
(133)
0 2
1
1
2
1

2
= p + P + |0 + p + |1 0 ,
(134)
0 2
1 1
1
(1)0

2
+ +
+

(135)
0 2 = p1 P1 |0 + p1 |1 0 .
Here the ghost numbers of the fields and the gauge parameters are also taken into account.
In the following we omit the subscripts at the state vectors associated with the eigenvalues
of the operator (68).
Substituting the fields in this concise form in the Lagrangian (94) ones find

+
+
L2 = |K2 T0 | + L+
1 | + iT1 |1 + L2 |4

+ 1 |K2 T0 |1 2i|3 L1 | + |4

+ 2 |K2 T0 |2 2|3 + T1 |1 i|4

+ 3 |K2 2i|1 2|2 + iT0 |4 + iT1 |

+ 4 |K2 iT0 |3 + iL1 |1

+
+ |K2 T0 | i|1 + L1 | L+
1 |1 iT1 |3

+ 1 |K2 i| iT1 | + T1+ |2 iL+
1 |4

+ 4 |K2 L2 | + |1 + i|2 ,
(136)
where we have used that the ghost fields commute with the operator Kn (96). Next we find
the gauge transformations (97), (98)
+
+
| = L+
1 | + iT1 |1 + L2 |4 ,
|1 = L1 | |4 + i|,
|2 = T1 |1 i|4 |,

|4 = T1 | ,
|3 = L1 |1 T0 | 2i|1 , (138)
(137)
| = T0 | 2i|1 iT1+ |,
+
|1 = T0 |1 + L+
1 | + 2iT1 |1 ,
|4 = T0 |4 + 4|1 ,
(139)
(140)
and the gauge for gauge transformations (99), (100)

| = iT1+ |,
+
|1 = L+
1 | + 2iT1 |1 ,
| = T0 | 4i|1 ,
|1 = T0 |1
|4 = 4|1 ,
(141)
(142)
in the concise form.

Now in order to derive Lagrangian (94) (or (136)) in component form we write the
fields |i and |i entering into (131) and (132) explicitly (taking into account the fields
eigenvalues associated with the operator )

1 + +
+ +
+
| =
a a (x) + d a (x) + b (x) |0,
2
|2 = 2 (x)|0,
|1 = 1 (x)|0,
|3 = 3 (x)|0,
|4 = 4 (x)|0,
+

+
| = a (x) + d (x) |0,

|1 = a + 1 (x) + d + 1 (x) |0,
|4 = 4 (x)|0
387
(143)
(144)
(145)
(146)
(147)
and substitute them into (136). As a result we get the Lagrangian (94) for the field with
spin 5/2 in the explicit form

1
i
L = i + 1 + 4
2
2

i

iD + 1 1 + D 21 + i4
2

+ i 1 1 23 i4

2 i 2 + 23 + 1 D1 + i4

+ 3 2i1 22 + 4 i + iD 4 3 + 1

i 1 + + 1 3

iD + 1 + 3

i 1 + D i 2 + i 4

1
D
+ iD 1 + i2 + i 4 + 1 + i2 . (148)
2
2
Here D is dimension of the spacetime, is basic spin 5/2 field and all other fields are
auxiliary. In order to write the gauge transformations in the explicit form we write the
gauge fields |i and |i entering into (133) and (134) as follows (also taking into account
the gauge parameters eigenvalues associated with the operator )

|1 = a + 1 (x) + d + 1 (x) |0,
| = a + (x) + d + (x) |0,
(149)
|4 = 4 (x)|0,
(150)
| = (x)|0,
|1 = 1 (x)|0,
(151)
and substitute them into (137)(140). As a result we get the gauge transformations for the
field associated with spin 5/2 in the explicit form
= i( + ) + i( 1 + 1 ) + 4 ,
(152)
= i + i1 i 1 ,
= 2i1 + 4 ,
(153)
1 = i 4 + i,
2 = 1 + D1 i4 , (154)
3 = i 1 + i 2i1 ,
4 = D,
(155)
= i 2i1 i ,
= i 2i1 i,
(156)
388
1 = i 1 i + 2i 1 ,
1 = i 1 + 2i1 ,
4 = i 4 + 41 .
(157)
(158)
Finally we get in the explicit form the gauge for gauge transformations (141), (142).
Writing the gauge for gauge parameters as
| = (x)|0,
|1 = 1 (x)|0
(159)
ones find (141), (142) in the explicit form

= i ,
= i,
1 = i + 2i 1 ,
1 = 2i1 ,
= i 4i1 ,
1 = i 1 .
(160)
4 = 41 ,
(161)
(162)
Thus, following the general procedure described in Section 4 we have constructed the
Lagrangian (94), the gauge transformations (97), (98) and the gauge for gauge transformations (99), (100) for the field model of spin 5/2 in the explicit form (148), (152)(158),
(160)(162), respectively. Unlike FangFronsdal construction, we obtained the Lagrangian
containing all proper set of auxiliary fields.
8. Summary
We have developed the new BRST approach to derivation of Lagrangians for fermionic
massless higher spin models in arbitrary dimensional Minkowski space. We investigated
the superalgebra generated by the constraints which are necessary to define an irreducible
massless half-integer spin representation of Poincar group and constructed the corresponding BRST charge. We found that the model is reducible gauge theory and the order of
reducibility linearly grows with the value of spin. It is shown that this BRST charge generates the correct Lagrangian dynamics for fermionic fields of any value of spin. We construct
Lagrangians in the concise form for the fields of any fixed spin in arbitrary spacetime dimension and show that our Lagrangians are reduced to the FangFronsdal Lagrangians
after partial gauge-fixing. As an example of general scheme we obtained the Lagrangian
and the gauge transformations for the field of spin 5/2 in the explicit form without any
gauge fixing.
The main results of the paper are given by the relations (94), where Lagrangian for
the field with arbitrary half-integer spin is constructed, and (97)(100) where the gauge
transformations for the fields and the gauge parameters are written down. In the case when
ones consider all half-integer spin fields together, the analogous relations are (105) for
the Lagrangian and (106)(109) for the gauge transformations. Our formulation does not
impose any off-shell constraints on the fields and the gauge parameters5 (see the discussion
of this point in [14]).
5 The possibility to formulate a higher spin field theory without restrictions on traces of the fields and the
gauge parameters was considered in [4].
389
The procedure for Lagrangian construction developed here for higher spin massless
fermionic field can be also applied to bosonic higher spin massless theories and leads to
the same results as in [16]. There are several possibilities for extending our approach. This
approach may be applied to Lagrangian construction for mixed symmetry tensor-spinor
fields (see [17] for corresponding bosonic case), for Lagrangian construction for fermionic
fields in AdS background, for massive higher spin fields using the dimensional reduction
and for supersymmetric higher spin models.
Acknowledgements
I.L.B. is very grateful to X. Berkaert, M. Grigoriev, M. Tsulaia, M.A. Vasiliev for fruitful discussions. We are thankful to A. Sagnotti and W. Siegel for useful comments. This
work was supported in part by the INTAS grants, projects INTAS-03-51-6346 and INTAS00-00254, The work of I.L.B. and V.A.K. was also supported by the RFBR grant, project
No. 03-02-16193, the joint RFBR-DFG grant, project No. 02-02-04002, the DFG grant,
project No. 436 RUS 113/669, the grant for LRSS, project No. 1252.2003.2 and the grant
PD02-1.2-94 of Russian Ministry of Education and Science. I.L.B. and V.A.K. are thankful the Humboldt-Universitt zu Berlin, where part of this work was done and D. Lst for
warm hospitality.
References
[1] C. Fronsdal, Massless fields with integer spin, Phys. Rev. D 18 (1978) 36243629.
[2] J. Fang, C. Fronsdal, Massless fields with half-integer spin, Phys. Rev. D 18 (1978) 36303633.
[3] M. Vasiliev, Progress in higher spin gauge theories, in: Proceedings of the International Conference Quantization, Gauge Theory and Strings, vol. 1, Moscow, 510 June 2000, World Scientific, Singapore, 2001,
pp. 452471, hep-th/0104246;
M. Vasiliev, Higher spin gauge theories in various dimensions, Fortschr. Phys. 52 (2004) 702717, hepth/0401177;
D. Sorokin, Introduction to the classical theory of higher spins, hep-th/0405069;
N. Bouatta, G. Compre, A. Sagnotti, An introduction to free higher-spin fields, hep-th/0409068.
[4] D. Francia, A. Sagnotti, Free geometric equations for higher spins, Phys. Lett. B 543 (2002) 303310, hepth/0207002;
D. Francia, A. Sagnotti, On the geometry of higher-spin gauge fields, Class. Quantum Grav. 20 (2003)
S473S486, hep-th/0212185.
[5] I.L. Buchbinder, V.A. Krykhtin, V.D. Pershin, On consistent equations for massive spin 2 field coupled to
gravity in string theory, Phys. Lett. B 466 (1999) 216226, hep-th/9908028;
I.L. Buchbinder, D.M. Gitman, V.A. Krykhtin, V.D. Pershin, Equations of motion for massive spin 2 field
coupled to gravity, Nucl. Phys. B 584 (2000) 615640, hep-th/9910188;
I.L. Buchbinder, D.M. Gitman, V.D. Pershin, Causality of massive spin 2 field in external gravity, Phys.
Lett. B 492 (2000) 161170, hep-th/0006144;
E. Sezgin, P. Sundell, Analysis of higher spin field equations in four dimensions, JHEP 0207 (2002) 055,
hep-th/0205132;
K.B. Alkalaev, M.A. Vasiliev, N = 1 supersymmetric theory of higher spin gauge fields in AdS(5) at the
cubic level, Nucl. Phys. B 655 (2003) 5792, hep-th/0206068;
J. Engquist, E. Sezgin, P. Sundell, On N = 1, N = 2, N = 4 higher spin gauge theories in four dimensions,
Class. Quantum Grav. 19 (2002) 61756196, hep-th/0207101;
390
[6]
[7]
[8]
[9]
[10]
[11]
[12]
[13]
P. de Medeiros, C. Hull, Exotic tensor gauge theory and duality, Commun. Math. Phys. 235 (2003) 255273,
hep-th/0208155;
J. Engquist, E. Sezgin, P. Sundell, Superspace formulation of 4D higher spin gauge theory, Nucl. Phys.
B 664 (2003) 439456, hep-th/0211113;
X. Bekaert, N. Boulanger, On geometric equations and duality for free higher spins, Phys. Lett. B 561 (2003)
183190, hep-th/0301243;
M. Plyushchay, D. Sorokin, M. Tsulaia, GL flatness of OSp(1|2n) and higher spin field theory from dynamics in tensorial space, hep-th/0310297;
K.B. Alkalaev, O.V. Shaynkman, M.A. Vasiliev, On the frame-like formulation of mixed-symmetry massless
fields in (A)dS(d), Nucl. Phys. B 692 (2004) 363393, hep-th/0311164;
K.B. Alkalaev, Two-column higher spin massless fields in AdS(d), hep-th/0311212;
O.V. Shaynkman, I.Yu. Tipunin, M.A. Vasiliev, Unfolded form of conformal equations in M dimensions and
o(M + 2)-modules, hep-th/0401086;
N. Boulanger, S. Cnockaert, Consistent deformations of [p, p]-type gauge field theories, JHEP 0403 (2004)
031, hep-th/0402180;
C.C. Ciobirca, E.M. Cioroianu, S.O. Saliu, Cohomological BRST aspects of the massless tensor field with
the mixed symmetry (k, k), hep-th/0403017;
A.K.H. Bengtsson, An abstract interface to higher spin gauge field theory, hep-th/0403267;
G. Barnich, M. Grigoriev, A. Semikhatov, I. Tipunin, Parent field theory and unfolding in BRST firstquantized terms, hep-th/0406192;
I. Bandos, P. Pasti, D. Sorokin, M. Tonin, Superfields theories in tensorial superspace and the dynamics of
higher spin fields, hep-th/0407180;
S. Deser, A. Waldron, Arbitrary spin representations in de Sitter from dS/CFT with applications to dS supergravity, Nucl. Phys. B 662 (2003) 379392, hep-th/0301068;
M. Bianchi, Higher spin symmetry (breaking) in N = 4 SYM and holography, hep-th/0409292;
M. Bianchi, Higher spins and stringy AdS5 S 5 , hep-th/0409304;
B. Sathiapalan, Loop variables and the (free) open string in a curved background, hep-th/0412033.
M. Vasiliev, Higher-spin gauge theories in four, three and two dimensions, Int. J. Mod. Phys. D 5 (1996)
763797, hep-th/9611024;
M. Vasiliev, Higher spin gauge theories: star-product and AdS space, Contributed article to Golfands Memorial Volume, M. Shifman (Ed.), World Scientific, hep-th/9910096;
M. Vasiliev, Higher spin symmetries, star-product and relativistic equations in AdS space, hep-th/0002183;
M. Vasiliev, Higher spin superalgebras in any dimension and their representations, hep-th/0404124.
V.E. Lopatin, M.A. Vasiliev, Free massless bosonic fields of arbitrary spin in D-dimensional de Sitter space,
Mod. Phys. Lett. A 3 (1998) 257.
M.A. Vasiliev, Free massless fermionic fields of arbitrary spin in D-dimensional anti-de Sitter space, Nucl.
Phys. B 301 (1988) 26.
C. Becchi, A. Rouet, R. Stora, Renormalization of the Abelian HiggsKibble model, Commun. Math.
Phys. 42 (1975) 127;
C. Becchi, A. Rouet, R. Stora, Renormalization of gauge theories, Ann. Phys. 98 (1976) 287;
I.V. Tyutin, Gauge invariance in field theory and statistics in operator formulation, preprint FIAN, N39
(1975).
S. Ouvry, J. Stern, Gauge fields of any spin and symmetry, Phys. Lett. B 177 (1986) 335340;
A.K.H. Bengtsson, A unified action for higher spin gauge bosons from covariant string theory, Phys. Lett.
B 182 (1986) 321325.
A. Witten, Noncommutative geometry and string field theory, Nucl. Phys. B 268 (1986) 253;
C.B. Thorn, String field theory, Phys. Rep. 175 (1989) 1101;
W. Taylor, B. Zwiebach, D-branes, tachyons, and string field theory, hep-th/0311017.
W. Siegel, B. Zwiebach, Gauge string fields from the light cone, Nucl. Phys. B 282 (1987) 125;
W. Siegel, Gauging Ramond string fields via OSp(1, 1/2), Nucl. Phys. B 284 (1987) 632;
W. Siegel, Introduction to String Field Theory, World Scientific, Singapore, 1988, hep-th/0107094;
W. Siegel, Fields, hep-th/9912205.
J. Isberg, U. Lindstrom, B. Sundborg, Spacetime symmetries of quantized tensionless strings, Phys. Lett.
B 293 (1992) 321326, hep-th/9207005;
[14]
[15]
[16]
[17]
[18]
[19]
[20]
[21]
[22]
[23]
[24]
391
U. Lindstrom, M. Zabzine, Tensionless strings, WZW models at critical level and massless higher spin fields,
Phys. Lett. B 584 (2004) 178185, hep-th/0305098;
G. Bonelli, On the tensionless limit of bosonic strings, infinite symmetries and higher spins, Nucl. Phys.
B 669 (2003) 159172, hep-th/0305155.
A. Sagnotti, M. Tsulaia, On higher spins and the tensionless limit of string theory, Nucl. Phys. B 682 (2004)
83116, hep-th/0311257.
F. Fougre, M. Knecht, J. Stern, Algebraic construction of higher spin interaction vertices, preprint LAPPTH-338/91.
A. Pashnev, M. Tsulaia, Description of the higher massless irreducible integer spins in the BRST approach,
Mod. Phys. Lett. A 13 (1998) 18531864, hep-th/9803207.
C. Burdik, A. Pashnev, M. Tsulaia, On the mixed symmetry irreducible representations of the Poincar
group in the BRST approach, Mod. Phys. Lett. A 16 (2001) 731746, hep-th/0101201.
I.L. Buchbinder, A. Pashnev, M. Tsulaia, Lagrangian formulation of the massless higher integer spin fields
in the AdS background, Phys. Lett. B 523 (2001) 338346, hep-th/0109067.
I.L. Buchbinder, A. Pashnev, M. Tsulaia, Massless higher spin fields in the AdS background and BRST
constructions for nonlinear algebras, in: Proceedings of XVI Max Born Symposium Supersymmetries and
Quantum Symmetries (SQS01), Karpacz, Poland, 2125 September 2001, Dubna, 2002, pp. 310, hepth/0206026.
X. Bekaert, I.L. Buchbinder, A. Pashnev, M. Tsulaia, On higher spin theory: strings, BRST, dimensional
reduction, Class. Quantum Grav. 21 (2004) S1457S1464, hep-th/0312252.
P. de Medeiros, Massive gauge-invariant field theories on space of constant curvature, Class. Quantum
Grav. 21 (2004) 25712593, hep-th/0311254;
R.R. Metsaev, Totally symmetric fields in AdS(d), Phys. Lett B 590 (2004) 95104;
Yu.M. Zinoviev, First order formalism for massive mixed symmetry tensor fields in Minkowski and (A)dS
spaces, hep-th/0306292.
S.M. Kuzenko, A.G. Sibiryakov, Phys. At. Nucl. 57 (1994) 1257;
S.J. Gates, S.M. Kuzenko, A.G. Sibiryakov, Phys. Lett. B 394 (1997) 343;
S.J. Gates, S.M. Kuzenko, A.G. Sibiryakov, Phys. Lett. B 412 (1997) 59;
I.L. Buchbinder, S. James Gates Jr., W.D. Linch Jr., J. Phillips, New 4D, N = 1 superfield theory: model of
free massive superspin- 32 multiplet, Phys. Lett B 535 (2002) 280288, hep-th/0201096;
I.L. Buchbinder, S. James Gates Jr., W.D. Linch Jr., J. Phillips, Dynamical superfield theory of free massive
superspin-1 multiplet, Phys. Lett. B 549 (2002) 229236, hep-th/0207243.
I.L. Buchbinder, S.M. Kuzenko, Ideas and Methods of Supersymmetry and Supergravity, Institute of Physics,
Bristol, 1988.
I.L. Buchbinder, V.A. Krykhtin, in preparation.
Giant gravitons in AdS3 S 3 T 4 as fuzzy cylinders

B. Janssen a , Y. Lozano b , D. Rodrguez-Gmez b
a Instituut voor Theoretische Fysica, KU Leuven Celestijnenlaan 200D, B-3001 Leuven, Belgium
b Departamento de Fsica, Universidad de Oviedo, Avda. Calvo Sotelo 18, 33007 Oviedo, Spain
Received 6 July 2004; received in revised form 22 November 2004; accepted 14 January 2005
Abstract
Using the non-Abelian action for coincident type IIB gravitational waves proposed in hepth/0303183 we show that giant gravitons in the AdS3 S 3 T 4 background can be described in
terms of coincident waves expanding into a fuzzy cylinder, spanned by two embedding scalars and
one worldvolume scalar. This fuzzy cylinder has dipole and magnetic moments with respect to the
2-form and 6-form potentials of the background, and can be interpreted as a bound state of D1-branes
and D5-branes (wrapped on the 4-torus) wrapped around the basis of the cylinder. We show the exact
agreement between this description and the Abelian, macroscopical description given in the literature.
PACS: 11.25.-w
1. Introduction
There is strong evidence by now that giant gravitons are described microscopically
in terms of dielectric gravitational waves, expanding into massless higher-dimensional
p-branes. From this point of view, the expansion of the waves happens because the transverse coordinates to the coincident waves are matrix valued, which allows the waves to
E-mail addresses: bert.janssen@fys.kuleuven.ac.be (B. Janssen), yolanda@string1.ciencias.uniovi.es

(Y. Lozano), diego@fisi35.ciencias.uniovi.es (D. Rodrguez-Gmez).
doi:10.1016/j.nuclphysb.2005.01.022
B. Janssen et al. / Nuclear Physics B 711 (2005) 392406
393
couple non-trivially to the RR potentials of the background. The effect is analogous to

Myers dielectric (or magnetic moment) effect for D-branes [1].
The non-Abelian worldvolume effective action for coincident Dp-branes contains nonAbelian couplings to RR potentials of order higher than (p + 1). Under the influence of a
RR (p + 4)-field strength the stable configuration corresponds to an expansion of the Dp2
branes into a D(p + 2)-brane with topology R 1,p Sfuzzy
. This brane is stable because it
carries a dipole (or magnetic) moment with respect to the RR potential, that cancels the
contraction due to its tension.
Stable expanded brane configurations had been previously found in the literature [2] as
single spherical D(p + 2)-branes with non-vanishing dipole moment and Dp-brane charge
dissolved in their worldvolumes. In fact, this is the large N limit of the N Dp-branes expanding into a fuzzy D(p + 2)-brane [1]. The description of expanded brane configurations
in terms of the expanding non-Abelian Dp-branes is commonly referred in the literature
as the microscopical description, whereas the description in terms of the spherical Abelian
D(p + 2)-brane is referred as the macroscopical description. A non-trivial check of the
validity of a given microscopical configuration of branes is its agreement with the corresponding macroscopical description when the number of branes is large.
Giant gravitons in AdSm S n spacetimes have been studied macroscopically in [37] as
stable brane configurations with non-zero angular momentum (or graviton charge) wrapped
around (m 2)- or (n 2)-spheres in the spacetime background and with a non-vanishing
dipole or magnetic moment with respect to the background gauge potential. The dynamical
equilibrium of these configurations is reached through the cancellation between the tension
of the brane and the coupling of the angular momentum to the background flux field. At
this equilibrium point, the configurations can be shown to have a zero mass, and behave
essentially as massless, finite size objects, which explains the name of giant gravitons.
Giant gravitons expanded in the spherical part of AdSm S n were first considered [3] as
a possible way to realise the stringly exclusion principle [8]. The radius of these expanded
gravitons is proportional to their angular momentum, and since this radius is bounded by
the radius of the S n , the configuration has associated a maximum angular momentum, in
agreement with the CFT predictions. Giant gravitons expanding in the AdS part of the
geometry do not satisfy however the stringy exclusion principle [4,5], given that AdS is
non-compact. They are referred in the literature as dual giant gravitons (whereas we will
refer to the giant gravitons in S n as genuine giant gravitons). Some discussion about how
the stringy exclusion principle can still be satisfied having these two types of giant graviton
configurations can be found in [4,6] and in [9,10].
In the microscopic picture, the giant gravitons are generated by gravitational waves
expanding into the spherical brane configurations due to Myers dielectric effect [1113].
Therefore, in order to describe giant gravitons microscopically we need a non-Abelian action for coincident gravitational waves, and then to identify the dielectric couplings that
will be responsible for their expansion. The non-Abelian effective action describing coincident gravitational waves in arbitrary backgrounds was constructed in [14,15].1 Using
this action we provided a microscopical description for the giant gravitons in AdSm S n
1 Giant gravitons in plane wave backgrounds have also been studied microscopically in [12,1620].
394
spacetimes expanding into 2-spheres: the genuine giant graviton in AdS7 S 4 , the dual
giant graviton in AdS4 S 7 and the genuine and dual giant gravitons in AdS5 S 5 .2 Other
macroscopical giant graviton solutions are known in these spacetimes, though the microscopic discussion is much harder, mainly due to the technical difficulties involved in the
construction of fuzzy n-spheres with n > 2 [21]. For example, the genuine giant graviton
of AdS4 S 7 and the dual one of AdS7 S 4 involve the construction of a fuzzy 5-sphere.
The microscopical description of these gravitons has not been given yet, though we hope
to report on this in a forthcoming paper [22].
In this paper we center on the microscopical description of giant gravitons in another
background, with its own peculiarities, namely AdS3 S 3 T 4 . It is known that macroscopic giant gravitons in AdS3 S 3 T 4 have features which are different from the ones
in AdSm S n spacetimes with m, n > 3. First of all, giant gravitons in AdS3 S 3 T 4
only exist when their angular momentum has a very specific value, namely a multiple of
the number of branes that create the geometry. Moreover, for this value of the momentum the graviton can have arbitrary size. The fact that the potential governing the size of
the giant graviton is flat in this background was already noted in [3,5]. Therefore, these
configurations seem to be completely unrelated to the stringy exclusion principle in this
background [8]. In fact, it has been suggested in [23,24] that giant gravitons are not the
correct supergravity description of the chiral primary states of the D1D5 system, which
are in turn described by the more general family of metrics given in [23,25,26].
Secondly, it is also known that various types of macroscopic giant graviton configurations can exist in this background. Besides the usual distinction between genuine and dual
giant gravitons, living respectively in the spherical and the AdS part of the spacetime, it is
possible to construct so-called mixed giant gravitons, which are basically a linear combination of the previous two. The one-cycle these mixed giant gravitons are wrapped on is
the sum of the one-cycles in the AdS and in the S parts. Also these configurations can have
arbitrary size in either part of the geometry, for the specific value of the angular momentum mentioned above. Furthermore, all types of giant gravitons (genuine, dual or mixed)
can be built by using D1-branes, D5-branes or both. The D1s wrap the one-cycle in the
adequate part of the background, while the D5s wrap the same one-cycle and the T 4 part.3
In this paper we will directly deal with the most general case: combined D1D5, mixed
giant gravitons, with the understanding that each separate case (or combinations thereof)
can be obtained by putting the appropriate parameters to zero.
A specific problem of the microscopic giant gravitons (of all types) in AdS3 S 3 T 4
seems to be the fact that the gravitational waves should expand into a fuzzy S 1 .4 The question then arises how such a fuzzy S 1 can be realised. The solution proposed here is that the
waves expand into a fuzzy cylinder, spanned by the S 1 in the background geometry and a
2 The giant and dual giant gravitons in AdS S 5 involve a fuzzy 3-sphere which is however described in
5
terms of an Abelian S 1 bundle over a fuzzy 2-sphere. See [15] for the details of this construction.
3 It should be clear that the D5 giant gravitons are in fact the T-duals of the D1s, after dualisation over the
directions of the T 4 .
4 For those blowing up into a D1 giant graviton, or a fuzzy S 1 times an Abelian T 4 for the D5 giant gravitons.
395
worldvolume scalar field .5 This scalar field arises naturally from T-duality in the derivation of the type IIB wave action [13]. Since the worldvolume field has no geometrical
meaning in the background, one effectively sees the cylinder for a given value of this scalar
field, which results into an S 1 .
This paper is organised as follows. In Section 2 we discuss briefly the AdS3 S 3 T 4
background and how it arises as the near horizon limit of a D1D5 intersection. Its main
purpose is to set our notation. In Section 3 we discuss macroscopically the most general
case of combined D1D5, mixed giant gravitons living both in AdS3 and S 3 . In Section 4
we construct the microscopic picture, deriving first the non-Abelian action for gravitational
waves in type IIB, involving the scalar field needed for the cylinder algebra. We also discuss
some properties of the fuzzy cylinder and its algebra and we come to the actual construction
of the microscopic picture in Section 4.3. We summarize our conclusions in Section 5.
2. The background
The AdS3 S 3 T 4 background arises as the near horizon geometry of the intersecting
D1D5 system [8,30]

1/2 1/2
1/2 1/2
1/2 1/2
2
dt 2 + dz2 + H1 H5 d 2 + 2 d32 + H1 H5
H5
dym
,
ds 2 = H1
1/2 1/2
1
3
,
Ftz = H1 ,
F1 2 3 = H5 gS 3 ,
e = H1 H5
(2.1)
where the i are the angles and gS 3 the determinant of the metric on the S 3 in the overall
transverse space. The coordinates ym (m = 1, . . . , 4), describing the relative transverse
space, can in principle have an unlimited range, though here we will choose them periodic
with period 2 . This can be thought of as wrapping the D5 on a four-torus.6 The harmonic
functions H1 and H5 are given by
Q1
Q5
(2.2)
,
H5 = 1 + 2 ,
2
with Q1 and Q5 the total D1- and D5-brane charge. In the near horizon limit 0, this
solution goes to AdS3 S 3 T 4
H1 = 1 +
L2
2
2
dt 2 + dz2 + 2 d 2 + L2 d32 + R2 dym
,
2
L
e = R2 ,
Ftz =
,
F1 2 3 = 2Q5 gS 3 ,
Q1
ds 2 =
(2.3)
where the radius R of the 4-torus as well as the radii L of curvature of AdS3 and S 3 , which
coincide in this background, are functions of the numbers of D1-branes and D5-branes:

L2 = Q1 Q5 .
R2 = Q1 /Q5 ,
(2.4)
5 The fuzzy cylinder has also been shown to play a role [27,28] in the microscopical description of the super-
tube [29].
6 In general the D5 can be wrapped on any Ricci-flat surface M 4 , giving rise in the near horizon limit to
spacetimes of the form AdS3 S 3 M 4 .
396
In this paper, we will work in global coordinates for AdS3 :

r2
r 2 1 2
dr + r 2 d 2
ds 2 = 1 + 2 dt 2 + 1 + 2
L
L

2
+ L2 d 2 + cos2 d 2 + sin2 d 2 + R2 dym
,
e = R2 ,
(2)
Ct =
Q5 2
r ,
L3
(2)
C = Q5 sin2 ,
(2.5)
(2.6)
where , , , ym [0, 2] and [0, ]. The RR background field C (2) can also be
expressed in terms of its Hodge-dual 6-form potential as:
(6)
C1234 = Q1 sin2 ,
(6)
Ct1234 =
Q1 2
r .
L3
(2.7)
Notice that the same expressions for the 6-form potential arise from performing four
T-duality transformations along the T 4 directions, after also renaming Q1 and Q5 . It is
clear that an AdS3 S 3 T 4 solution, due to the number of isometry directions, can be
embedded both in type IIA as type IIB supergravity. The above form, however, with all
four radii of the T 4 being equal, is only a solution of type IIB. We will limit ourselves for
the rest of this paper to this specific case.
3. Macroscopic giant gravitons in AdS3 S 3 T 4

As we have mentioned, the most general giant graviton solution in the AdS3 S 3 T 4
background is in terms of a test brane consisting on a bound state of D1-branes and D5branes wrapped on the 4-torus, both with angular momentum in S 3 and expanding at the
same time in the AdS3 and in the S 3 parts of the geometry, while maintaining constant radii
in these spaces [24]. Therefore we look at a trial solution with = const, r = const and
= ( ), where t = in static gauge, that is, the giant graviton runs around the sphere
along the coordinate . We wrap our combination of D1-branes and wrapped D5-branes
around the circle parametrised by7
=
+
.
2
(3.1)
Then the pullbacked metric on the D-branes is given by

r2
2
ds = 1 + 2 dt 2 + r 2 + L2 sin2 d 2 + L2 cos2 d 2 + R2 dym
.
L
(3.2)
We have as well non-vanishing, constant RR potentials given by (2.6) and (2.7).

7 One could either choose wrapping the branes around the circle parametrised by ( )/2, in which case
the graviton moves along in the opposite direction. This is a consequence of the symmetry of the background
under the simultaneous interchange , .
397
Considering a giant graviton made of n D1-branes and m D5-branes8 we obtain the

following action, after integrating over :

Q1
S = 2 nT1 + (2)4 mT5
Q5

Q5 2
r2
2
2
2
2
2
d
r + L sin 1 + 2 L cos
Q1
L
2

r
+ sin2
Q5
L3
(3.3)
with T1 (T5 ) the tension of a D1-brane (D5-brane). For the Hamiltonian we get

1
r2
2T1 (nQ5 + mQ1 )
1+ 2
H=
L
cos
L

2
P
r2
r2
sin2 2 ,
cos2 sin2 + 2 +
2T1 (nQ5 + mQ1 )
L
L
(3.4)
where we have taken into account that T1 = (2)4 T5 . P is the angular momentum carried
by the combination of branes, which is constant given that is a cyclic coordinate in (3.3).
It is clear that the minimum energy solution corresponds to
P = 2T1 (nQ5 + mQ1 ),
(3.5)
for which
H=
2T1 (nQ5 + mQ1 ) P

=
,
L
L
(3.6)
and this happens independently of the size of the giant graviton. It is easy to see that the
genuine and dual giant graviton solutions given in [5] arise as the special limits r = 0 or
= 0 of this solution, since in these limits the giant graviton expands either on the S 3
or on the AdS3 part of the geometry. That the potential governing the size of the giant
graviton is flat in the AdS3 S 3 background was already noted in [3,5]. This fact poses a
puzzle with the realisation of the stringy exclusion principle. However, as we mentioned in
the introduction, giant graviton configurations do not seem to be the correct supergravity
description of the chiral primary states of the dual CFT [23,24]. Indeed in the CFT there are
no chiral primary states beyond JL = JR = Q1 Q5 [8], whereas giant gravitons only exist
with angular momentum P = 2T1 (nQ5 + mQ1 ), a result which seems to be completely
unrelated to the CFT predictions.
8 Alternatively one can consider m D5-branes with n D1-brane charge dissolved in their worldvolumes, and
arrive at the same results.
398
4. The microscopical description

We expect to describe microscopically the giant graviton configuration of the previous
section in terms of coincident gravitons expanding into a 1-brane with the topology of a
fuzzy circle, consisting on a bound state of D1-branes and D5-branes wrapped on the 4torus. In this description the expansion of the gravitons takes place due to their interaction
with the RR 2-form and 6-form potentials of the background, which in this case have
both electric and magnetic components. At the level of the graviton worldvolume effective
action this interaction occurs in the form of non-Abelian dielectric and magnetic moment
couplings.
4.1. The action
The effective action describing a system of coincident gravitons in type IIB was constructed in [13] for weakly curved backgrounds, and later extended to more general ones
in [15]. The extended action presented in [15] can be used to study the AdSm S n background, which is not a linear perturbation to Minkowski. This action was truncated for
simplicity to certain worldvolume fields equal to zero. One of these fields will however be
non-vanishing in the AdS3 S 3 T 4 background, so our first task will be to extend the
computation in [15] to this case.
The worldvolume dynamics of type IIB gravitons is determined by D-strings and
F-strings ending on them [13], in a way which is manifestly S-duality invariant. These
strings are wrapped around a transverse direction that appears automatically as an isometric direction in the T-duality derivation of the action. This direction is in fact the direction
along which the T-duality is performed. Let us call it the z-direction, and l the Killing
vector pointing along z, i.e., l = z in the adapted coordinate system. Each type of string
ending on the gravitons has associated a worldvolume scalar forming an invariant field
strength either with il C (2) or with il B (2) , where il denotes the interior product with l .
We call the worldvolume scalar associated to F-strings. This worldvolume scalar plays
the role of the T-dual of the z-direction. Since we are dealing with a gauged sigma model
in which the translations along this direction are gauged, the embedding scalar Z disappears as a transverse scalar but a new worldvolume scalar is generated in the process
which accounts for the corresponding degree of freedom. This situation is analogous to the
one that is found in the relation between the NS5-brane and the KaluzaKlein monopole
via T-duality. In this case the transverse direction in which the NS5-brane is dualised becomes the Taub-NUT direction of the monopole, which is isometric in the action, and a
new worldvolume scalar is generated which is associated to wrapped F-strings ending on
the monopole [31].
The effective action for type IIB gravitons constructed in [15] is modified for nonvanishing, but constant in the following way (we restrict for simplicity to vanishing RR
4-form potential):

i

BI
SWB = T0 d STr k 1 P E + Ei Q1 k E kj Ej det Qij , (4.1)
399
where now
Eij = Gij ,

Eiz = e k 1 l 1 ik C (2) i ,
Ezz = l 2 ,

Qij = ji 1 + ie kl X i , X k Gkj i X i , ik C (2) j ,

Qiz = ie kl 1 X i , + i X i , X k ik C (2) k ,

Qzi = ie kl , X k Gki ,

Qzz = 1 + i , X k ik C (2) k .
Ezi = Eiz ,
(4.2)
Here i, j, k exclude the z-direction and G denotes the reduced metric, typical for gauged
sigma models
G = g k 2 k k l 2 l l ,
(4.3)
which projects out the embedding scalars corresponding to the two isometry directions l
and k , where the latter is pointing along the direction of propagation of the gravitons
(see below). The scalar k and vector k are defined as k 2 = g k k and k = g k .
The same notation applies to l . We have also taken in the above action that g k l = 0,
a condition that is satisfied for the background that we consider in this paper.
The ChernSimons part of the action contains the term [13]

CS
(2)
=
iT
=
iT
)C
SW
d
STr
P
(i
d STr X i , Cij(2) DX j
0
[X,]
0
B
(4.4)
which will be playing an important role in the description of the giant graviton, as we will
see.
The fact that the direction of propagation appears as an isometric direction is common
to all gravitons in type II and M-theories (see [14,15]). It is in fact easy to see that, in the
Abelian limit, a Legendre transformation restoring the dependence along the time derivative of this direction yields the usual action for massless particles in terms of an auxiliary
-metric (see the previous references for the details). The second isometric direction, z, is
however special to the type IIB case. The reason why the dependence along the T-duality
direction cannot be restored in this case seems to be a technical one. It is remarkable
however that only due to the presence of this isometric direction we can obtain the right
dielectric couplings to higher order type IIB RR potentials relevant in the background we
are considering.
4.2. The fuzzy cylinder
We expect to describe microscopically the giant graviton configuration of Section 3 in
terms of gravitons expanding into a 1-brane with the topology of a fuzzy circle with
radius (see (3.2))

R = r 2 + L2 sin2 .
(4.5)
400
It is clear however that a circle cannot be made non-commutative unless we embed it in

a higher-dimensional non-commutative manifold. The simplest thing is to embed it in a
non-commutative cylinder.
A non-commutative version of the circle condition x12 + x22 = R 2 can be obtained by
making the non-commutative ansatz
1 2
1 3
2 3
X , X = 0,
(4.6)
X , X = if X 2 ,
X , X = if X 1 ,
i.e., taking the coordinates X 1 and X 2 , defining the circle, together with a third generator X 3 , to satisfy the algebra of the two-dimensional Euclidean group, which is the algebra
defining the fuzzy cylinder [27,3234], with X 1 , X 2 parametrising the base and X 3 the
axis of the cylinder. The length scale f is the non-commutative parameter.
The quadratic Casimir associated to the algebra (4.6) is (X 1 )2 + (X 2 )2 , so the base
of the cylinder is indeed a non-commutative circle, since we can realise the condition
x12 + x22 = R 2 as
1 2 2 2
X + X
(4.7)
= R 2 1.
But, which generator in our geometry is associated to X 3 ? It turns out that the worldvolume
scalar , that couples in the effective action describing the system of coincident gravitons
(4.1) and (4.4), must play the role of the coordinate X 3 along the axis of the cylinder.
Therefore the fuzzy cylinder is not a geometrical object in which the gravitons expand,
given that it is not defined entirely in terms of embedding scalars with the interpretation
of transverse coordinates. This will become clearer below when we construct the giant
graviton solution explicitly.
The representations of the fuzzy cylinder algebra are infinite dimensional (see for instance [27]). This means that we can only provide a microscopical description of the giant
graviton solution for an infinite number of gravitons. Still, even though the dimension of
the matrices is infinite, the algebra is non-trivial, since the non-commutative parameter
f is independent of the dimension of the representation. This situation is different from
the fuzzy S 2 case, where the limit of infinite number of gravitons is at the same time the
commutative limit.
One explicit realisation of (4.6) is [33,34]
1
1
X 1 = c T 1 ,
X 2 = c T 2 ,
= f T 3 ,
2
2
where we already take as the direction along the axis, with
1
T mn = m+1,n + m1,n ,
2
T mn = im+1,n im1,n ,

3
1
m,n
T mn = m
2
(4.8)
(4.9)
and m and n running from to +. The quadratic Casimir depends on the parameter
c as
1 2 2 2
= c2 1,
X + X
(4.10)
401
so we can have a fuzzy version of the circle defined by x12 + x22 = R 2 if we choose c = R
and we embed the circle in a cylinder whose axis is taken along the -direction and is
therefore infinite, since the eigenvalues of range from to +.
The length of the cylinder is in fact related to the non-commutative parameter f . The
matrix algebra (4.6) has a discrete translational symmetry along , with shift unit f [34].
The unitary matrix U defined by Umn = m+1,n acts as
U 1 X i U = X i ,
i = 1, 2,
U 1 U = + f,
(4.11)
so the fuzzy cylinder is invariant under translations along with no deformation in the
(X 1 , X 2 )-plane, with shift unit fixed by the non-commutative parameter f . This parameter
can then be regarded as a minimal distance in the -direction, and the size of this direction
can be estimated as
l = f Tr 1,
(4.12)
which is indeed infinite for f = 0.9

In the commutative limit f 0 the system is invariant under continuous translations
along . This reflects a symmetry under overall translations in the direction of the axis of
the cylinder, which is analogous to that of the supertube [29,35]. We must stress however
that in our case the worldvolume scalar does not have an interpretation as a transverse
coordinate,10 so the circle that we are making non-commutative by embedding it in the
cylinder is not physically located in a cylinder in the transverse space. On the contrary, in
our construction one effectively sees the cylinder for a given value of , and this results into
an S 1 . From this point of view it is natural to find that the fuzzy cylinder has an invariance
along this direction.
4.3. The microscopic giant graviton
Looking at the background metric
given by (3.2) we expect the gravitons to expand into

a fuzzy circle with radius R = r 2 + L2 sin2 . In the most general case this fuzzy circle
corresponds to a bound state of n D1-branes and m D5-branes wrapped on the 4-torus.
Taking Cartesian coordinates

x1 = r 2 + L2 sin2 cos ,
(4.13)
x2 = r 2 + L2 sin2 sin
the line effective element (3.2) as seen by the waves reduces to (i = 1, 2)

r2
2
ds 2 = 1 + 2 dt 2 + dxi2 + L2 cos2 d 2 + R2 dym
,
L
(4.14)
9 It is however possible to take the commutative limit such that the resulting cylinder has finite length by
taking f going to zero exactly to compensate the divergence of Tr 1.
10 Given that = 0 we can neither interpret it as inducing F-string charge in the configuration (recall that
forms an invariant field strength with il B (2) , F = + il B (2) [13]). If that were the case we would be describing
a configuration different from the giant graviton that we want to study.
402
and the electric and magnetic 2-form and 6-form potentials to

sin2
(2)
Ci = Q5
r 2 + L2 sin2
(6)
Ci1234 = Q1
(2)
Cti =
ij xj ,
sin2
r 2 + L2 sin2
r2
Q5
ij xj ,
L3 r 2 + L2 sin2
(6)
Cti1234 =
ij xj ,
(4.15)
r2
Q1
ij xj . (4.16)
L3 r 2 + L2 sin2
We now make the fuzzy cylinder ansatz for the 2-sphere parametrised by x 1 , x 2 and the
worldvolume scalar appearing in the wave action:
1 2
X , X = 0,
1
X , = if X 2 ,
2
X , = if X 1 .
Rewriting the 6-form potentials in terms of their hodge dual, the ansatz for the RR potentials becomes
(2)
Ci = (nQ5 + mQ1 )
(2)
Cti =
sin2
r 2 + L2 sin2
ij Xj ,
r2
(nQ5 + mQ1 )
ij Xj .
3
2
L
r + L2 sin2
(4.17)
(2)
Here, Ci
couples in the BI part of the non-Abelian wave action (see (4.2)), and Cti(2) in
the CS part, given by (4.4).
Taking into account that the gravitons propagate along the -direction, so that k = ,
and substituting the background and the non-commutative ansatz above in the action for
coincident gravitons (4.1), (4.4), we find:

SWB = T0

d STr

1
L cos
1+

1 f (nQ5 + mQ1 )
r2
L2
sin2
r2
+ L2 sin2
2
Xi
2

i 2
r2
1
.
f (nQ5 + mQ1 ) 3
X
L r 2 + L2 sin2
2
f2
(nQ5 + mQ1 )2 cos2 X i
L2
(4.18)
In this description, since the direction of propagation is isometric, we are effectively dealing with a static configuration, and we can compute the potential as minus
the Lagrangian. Note that the embedding scalars X i only appear via their quadratic
Casimir (4.10). Also, it is easy to check that the corrections due to the contributions of
(X i )2n in the symmetrised trace prescription vanish, so that we can write the potential exactly in terms of an ordinary trace. We stress that this is an exact expression, in contrast
to the giant gravitons expanded in fuzzy 2-spheres [14,15], where the symmetrised trace
induces corrections of order 1/N 2 . We find a potential:
VWB

Tr 1T0
1
r2
f (nQ5 + mQ1 )
=
1+ 2
L
cos
L

2
1
r2
r2
sin2 2 .
cos2 sin2 + 2 +
f (nQ5 + mQ1 )
L
L
403
(4.19)
Here Tr 1 is infinite, since the irreducible representations of the fuzzy cylinder algebra are
infinite dimensional. The potential per unit length of the cylinder is however finite, and it
is given by

T0
1
r2
VWB = (nQ5 + mQ1 )
1+ 2
L
cos
L

2

1
r2
r2
cos2 sin2 + 2 +
sin2 2 , (4.20)
f (nQ5 + mQ1 )
L
L
where we have used (4.12).
The minimum energy is reached when the non-commutative parameter
f = (nQ5 + mQ1 )1 ,
(4.21)
for which the radius of the cylinder remains however arbitrary. For this value of f the
energy per unit length is
P
T0
(4.22)
(nQ5 + mQ1 ) =
,
L
L
with P the momentum (3.5). Therefore, the configuration describes a massless brane with
momentum P , and we find perfect agreement with the macroscopical description.
We can compare the microscopical potential (4.20) and the minimum energy condition
(4.21) with the corresponding quantities in the macroscopical calculation. Indeed, the momentum and energy that should be compared to those of the macroscopical calculation are
the ones per unit length of the cylinder, since we need to project onto the (X 1 , X 2 )-plane
to make connection with the description in terms of the string wrapped around the circle
x12 + x22 = R 2 . The corresponding quantities in the macroscopical calculation are given
by (3.4) and (3.5), respectively. We then find that there is exact agreement, given that the
macroscopical momentum P is given, in microscopical quantities, by
E=
P =
Tr 1T0 T0
= ,
l
f
(4.23)
where we have taken into account that microscopically the momentum is given by the tension of a wave times the number of them, and we have used (4.12). We must stress that, as
shown in [1], the agreement between the microscopical or non-Abelian calculation and the
macroscopical or Abelian one is found when the number of expanding branes goes to infinity. For the giant graviton configurations that we have studied microscopically in [14,15],
which involved fuzzy 2-spheres, we indeed found this agreement for infinite number of
gravitons. In the case of the AdS3 S 3 T 4 background the irreducible representations of
the fuzzy cylinder are infinite dimensional, so our calculation is only valid for an infinite
404
number of gravitons. From this point of view it is no surprise that we find exact agreement
with the macroscopical description.
The crucial difference between the fuzzy cylinder and the fuzzy S 2 is that for the
fuzzy S 2 the limit of infinite number of gravitons is at the same time the commutative
limit, so the microscopical description, in terms of a fuzzy S 2 , tends in this limit to the
macroscopical, or commutative, description, formulated in terms of a classical spherical
test brane. In the case of the fuzzy cylinder the non-commutative parameter f is independent of the dimension of the representation and can therefore be independently sent to
zero. This justifies why the agreement between the microscopical and macroscopical descriptions occurs for f = 0, and therefore for a non-commutative cylinder. Physically this
is due to the fact that the gravitons do not expand onto the whole cylinder, but only onto its
projection on the (X 1 , X 2 )-plane. Therefore, the value of the shift unit along the axis of the
cylinder, which is given by f , as discussed around (4.11), should be physically irrelevant.
In the classical limit, however, when the integer charges of the D1-branes and D5-branes
that create the background geometry, Q1 and Q5 , are very large, so that the supergravity
solution is valid, f , as given by (4.21), must be very small, so in this limit the cylinder
indeed becomes commutative.
5. Conclusions
We have shown that the action proposed in [15] to describe multiple type IIB gravitational waves is suitable for the microscopical study of giant gravitons in the AdS3 S 3 T 4
background in terms of dielectric gravitational waves. This action was used in [15] to describe the giant gravitons in the AdS5 S 5 background as expanded gravitational waves.
The genuine (dual) giant graviton was shown to be described in terms of gravitational
waves expanding into a fuzzy 3-sphere contained inside S 5 (AdS5 ), with a non-vanishing
magnetic (dipole) moment with respect to the RR 4-form potential of the background. In
both cases for large number of gravitons we found perfect agreement with the macroscopical description of [35], which provided a strong support for the validity of our action.
Remarkably, the fuzzy S 3 solution to the equations of motion derived from our action
could simply be described as an S 1 bundle over a fuzzy S 2 base manifold.
In this paper we have used the same action to describe microscopically the giant graviton configurations of another type IIB background, AdS3 S 3 T 4 . This background has
the special feature that both giant and dual giant gravitons can be studied in a unified way
in terms of strings winding at the same time around a circle in AdS3 and in S 3 [24]. This
can happen because the background has both electric and magnetic RR 2-form potentials
switched on. That genuine giant gravitons (satisfying stringy exclusion principle for general AdSm S n backgrounds) and dual giant gravitons (not satisfying stringy exclusion
principle) can be described in a unified way in this background is in consonance with the
fact that giant graviton configurations in the AdS3 S 3 T 4 background do not seem to be
the supergravity duals of the chiral primary states of the two-dimensional CFT of the D1
D5 system. Instead, they seem to be associated to dissasociated D1D5 systems [24].
Therefore in this spacetime giant graviton configurations would be completely unrelated
to the stringy exclusion principle.
405
It would perhaps shed some light on this issue to study dual descriptions of IIB waves
expanded onto circular D1- and (wrapped) D5-branes along the lines of [9], where the
relations between giant and dual giant graviton configurations in M-theory and Polchinski
and Strasslers [36] N D3 D5 and N D3 NS5 dielectric brane configurations are
used to argue that each type of giant graviton can only exist in a given regime of the space
of parameters.11
We should stress that one limitation of our microscopical description is that we are
constrained to work with infinite coincident gravitons, given that the representations of the
algebra of the fuzzy cylinder are infinite-dimensional. A possible way to consider finite N
coincident gravitons would be by taking the fuzzy cylinder algebra as the limit of a fuzzy
ellipsoid, for which the algebra is a deformed SU (2) algebra (see for instance [34]). In
this case however a circular section as the one involved in the giant graviton configurations
can only be recovered in the limit in which the ellipsoid becomes a cylinder, for which we
would be stuck again with a infinite number of gravitons.
Giant gravitons in AdSm S n spacetimes preserve the same fraction of the supersymmetries than the point-like gravitons in these spacetimes. The condition over the spinors
reads [4,5]
t

+ 1 = 0,
(5.1)
which coincides with the supersymmetry preserving condition of a gravitational wave with
momentum P propagating in flat space [37]. One would expect that the same holds true
in the AdS3 S 3 T 4 background, although this has not been checked out explicitly. On
the other hand, the supersymmetry properties of the microscopical configurations have not
been examined so far, though the agreement with the macroscopical description suggests
that one should be able to arrive at the same condition. It would be interesting to check
whether this is indeed the case.
Acknowledgements
We wish to thank Jos Gheerardyn and Jan Rosseel for useful discussions. The work of
B.J. has been done as a Post-doctoral Fellow of the FWO-Vlaanderen. B.J. is also partially supported by the European Commission RTN-program HPRN-CT-2000-00131, by
the FWO-Vlaanderen project G0193.00N and by the Belgian Federal Office for Scientific, Technical and Cultural Affairs through the Interuniversity Attraction Pole P5/27. The
work of Y.L. and D.R.-G. has been partially supported by CICYT grant BFM2003-00313
(Spain). D.R.-G. was supported in part by a FPU Fellowship from MEC (Spain). Y.L. and
D.R.-G. would like to thank the Institute for Theoretical Physics at the University of Leuven for its hospitality while part of this work was done.
11 We would like to thank a suggestion of the referee along these lines.
406
References
[1]
[2]
[3]
[4]
[5]
[6]
[7]
[8]
[9]
[10]
[11]
[12]
[13]
[14]
[15]
[16]
[17]
[18]
[19]
[20]
[21]
[22]
[23]
[24]
[25]
[26]
[27]
[28]
[29]
[30]
[31]
[32]
[33]
[34]
[35]
[36]
[37]
R.C. Myers, JHEP 9912 (1999) 022, hep-th/9910053.

R. Emparan, Phys. Lett. B 423 (1998) 71, hep-th/9711106.
J. McGreevy, L. Susskind, N. Toumbas, JHEP 0006 (2000) 008, hep-th/0003075.
M.T. Grisaru, R.C. Myers, O. Tafjord, JHEP 0008 (2000) 040, hep-th/0008015.
A. Hashimoto, S. Hirano, N. Itzhaki, JHEP 0008 (2000) 051, hep-th/0008016.
S.R. Das, A. Jevicki, S.D. Mathur, Phys. Rev. D 63 (2001) 044001, hep-th/0008088.
S.R. Das, A. Jevicki, S.D. Mathur, Phys. Rev. D 63 (2001) 024013, hep-th/0009019.
J.M. Maldacena, A. Strominger, JHEP 9812 (1998) 005, hep-th/9804085.
I. Bena, D. Smith, Towards the solution to the giant graviton puzzle, hep-th/0401173.
D. Berenstein, A toy model for the AdS/CFT correspondence, hep-th/0403110.
S.R. Das, S.P. Trivedi, S. Vaidya, JHEP 0010 (2000) 037, hep-th/0008203.
D. Berenstein, J. Maldacena, H. Nastase, JHEP 0204 (2002) 013, hep-th/0202021.
B. Janssen, Y. Lozano, Nucl. Phys. B 643 (2002) 399, hep-th/0205254.
B. Janssen, Y. Lozano, Nucl. Phys. B 658 (2003) 281, hep-th/0207199.
B. Janssen, Y. Lozano, D. Rodrguez-Gmez, Nucl. Phys. B 669 (2003) 363, hep-th/0303183.
A. Mikhailov, Nonspherical giant gravitons and matrix theory, hep-th/0208077.
S.R. Das, J. Michelson, A.D. Shapere, Fuzzy spheres in pp-wave matrix string theory, hep-th/0306270.
K. Sugiyama, K. Yoshida, Phys. Rev. D 66 (2002) 085022, hep-th/0207190.
H. Takayanagi, T. Takayanagi, JHEP 0212 (2002) 018, hep-th/0209160.
Y.X. Chen, J. Shao, Phys. Rev. D 69 (2004) 106010, hep-th/0310062.
S. Ramgoolam, Nucl. Phys. B 610 (2001) 032, hep-th/0101001;
S. Ramgoolam, JHEP 0210 (2002) 064, hep-th/0207111.
B. Janssen, Y. Lozano, D. Rodrguez-Gmez, in preparation.
J.M. Maldacena, L. Maoz, Desingularization by rotation, hep-th/0012025.
O. Lunin, S.D. Mathur, A. Saxena, Nucl. Phys. B 655 (2003) 185, hep-th/0211292.
V. Balasubramanian, J. de Boer, E. Keski-Vakkuri, S.F. Ross, Phys. Rev. D 64 (2001) 064011, hepth/0011217.
O. Lunin, S.D. Mathur, Nucl. Phys. B 623 (2002) 342, hep-th/0109154.
D. Bak, K. Lee, Phys. Lett. B 509 (2001) 168, hep-th/0103148.
I. Bena, Phys. Rev. D 67 (2003) 026004, hep-th/0111156.
D. Mateos, P.K. Townsend, Phys. Rev. Lett. 87 (2001) 011602, hep-th/0103030.
J.M. Maldacena, Adv. Theor. Math. Phys. 2 (1998) 231, hep-th/9711200.
E. Eyras, B. Janssen, Y. Lozano, Nucl. Phys. B 531 (1998) 275, hep-th/9806169.
M. Chaichian, A. Demichev, P. Presnajder, Nucl. Phys. B 567 (2000) 360, hep-th/9812180.
Y. Hyakutake, Nucl. Phys. B 675 (2003) 241, hep-th/0302190.
K. Hashimoto, JHEP 0404 (2004) 004, hep-th/0401043.
D. Bak, Y. Hyakutake, N. Ohta, Phase moduli space of supertubes, hep-th/0404104.
J. Polchinski, M.J. Strassler, The string dual of a confining four-dimensional gauge theory, hep-th/0003136.
E.A. Bergshoeff, R. Kallosh, T. Ortn, Phys. Rev. D 47 (1993) 5444, hep-th/9212030.
Nuclear Physics B 711 [FS] (2005) 409479
Superconformal Ward identities and their solution

M. Nirschl, H. Osborn 1
Department of Applied Mathematics and Theoretical Physics,
Wilberforce Road, Cambridge CB3 0WA, England, UK
Received 6 August 2004; accepted 11 January 2005
Available online 1 February 2005
Abstract
Superconformal Ward identities are derived for the four point functions of chiral primary BPS
operators for N = 2, 4 superconformal symmetry in four dimensions. Manipulations of arbitrary
tensorial fields are simplified by introducing a null vector so that the four point functions depend
on two internal R-symmetry invariants as well as two conformal invariants. The solutions of these
identities are interpreted in terms of the operator product expansion and are shown to accommodate
long supermultiplets with free scale dimensions and also short and semi-short multiplets with protected dimensions. The decomposition into R-symmetry representations is achieved by an expansion
in terms of two variable harmonic polynomials which can be expressed also in terms of Legendre
polynomials. Crossing symmetry conditions on the four point functions are also discussed.
PACS: 11.25.Hf; 11.30.Pb
Keywords: Superconformal symmetry; Chiral primary operators; Correlation functions; Operator product
expansion
1. Introduction
Since the discovery of the AdS/CFT correspondence there has been a huge resurgence of
interest in superconformal theories in four dimensions, for a review see [1]. In particular for
the N = 4 superconformal SU(N ) gauge theory, to which the AdS/CFT correspondence
E-mail addresses: mn244@damtp.cam.ac.uk (M. Nirschl), ho@damtp.cam.ac.uk (H. Osborn).
1 Address for correspondence: Trinity College, Cambridge, CB2 1TQ, England.
doi:10.1016/j.nuclphysb.2005.01.013
410
M. Nirschl, H. Osborn / Nuclear Physics B 711 [FS] (2005) 409479
is most directly applicable, many new and exciting results have been obtained. Much has
been discovered concerning the spectrum of operators and their scale dimensions both in
the large N limit, through the supergravity approximation to the AdS/CFT correspondence,
and also perturbatively as an expansion in the coupling g.
Of the operators present in the theory the simplest are the chiral primary operators belonging to SU(4)R R-symmetry representations with Dynkin labels [0, p, 0]. These are
represented by symmetric traceless rank p tensors formed by gauge invariant traces of the
elementary scalar fields and satisfy BPS like constraints so that they belong to short supermultiplets of the superconformal group PSU(2, 2|4). They are therefore protected against
renormalisation effects and have scale dimension = p. Their three point functions have
been fully analysed in [2] and perturbative corrections shown to vanish in [3,4] and, using
harmonic superspace, in [5]. For the case of p = 2, when the supermultiplet contains the
energymomentum tensor, the four point functions have been found both perturbatively [6]
and in the large N limit [7]. Such results have also been extended more recently to chiral
primary operators with p = 3, 4 [810]. The explicit results for the four point correlation
functions has then allowed an analysis of those operators which contribute to the operator
product expansion for two chiral primary operators [1117].
To take the analysis of the operator product expansion of correlation functions beyond
the lowest scale dimension operators for each SU(4)R representation it is necessary to have
an explicit form for the conformal partial waves which give the contribution of a quasiprimary operator of arbitrary scale dimension and spin and all its conformal descendants
to conformally covariant four point functions. In four dimensions a simple expression was
found in [18]. In addition since all operators in a superconformal multiplet must have the
same anomalous dimensions it is desirable to have a procedure for analysing the operator
product expansion for each supermultiplet as a single contribution. This depends on a solution of all superconformal Ward identities since this should allow all possible operator
product expansion contributions to be found in a form compatible with the superconformal symmetry. For the simplest case of the four point function for [0, 2, 0] chiral primary
operators this was undertaken in [19] and applied to determine the one-loop anomalous
dimensions for all operators with lowest order twist two.
The procedure adopted in [19] is somewhat involved and does not simply generalise
to correlation functions of more general chiral primary operators. As was shown in [19]
the superconformal Ward identities are simplified if they are expressed in terms of new
variables x, x rather than the usual conformal invariants. In terms of the standard correspondence for the spacetime coordinates x a x = x a a , where x is a 2 2 spinorial
matrix such that det x = x 2 , then, for four points x1 , x2 , x3 , x4 and xij = xi xj , x, x
may be defined, as shown in [20], as the eigenvalues of x12 x42 1 x43 x13 1 . By conformal

transformations we may choose a frame such that x2 = 0, x3 = , x4 = 1 and x1 = x0 x0 .
The two conformal invariants are then given in terms of x, x by,2

u = det x12 x42 1 x43 x13 1 = x x,

2 Since 1 + u v = x + x and 1 + u2 + v 2 2uv 2u 2v = (x x)
2 it is easy to invert these results

to obtain x, x in terms of u, v up to the arbitrary sign of the square root (x x)
2 . For any f (u, v) there is a
corresponding symmetric function f(x, x)

= f(x,
x) such that f(x, x)
= f (u, v).

v = det 1 x12 x42 1 x43 x13 1 = det x14 x24 1 x23 x13 1 = (1 x)(1 x).
411
(1.1)
For a Euclidean metric on spacetime x, x are complex conjugates. In our analysis we find
that the linear equations which follow from superconformal invariance naturally separate
into ones involving just x and conjugate equations with x x.
A technical complication in dealing with arbitrary chiral primary operators represented

by symmetric traceless tensorial fields r1 ...rp is that the four point function for four chiral
primary fields, for arbitrary p1 , p2 , p3 , p4 , involves in general a proliferation of independent tensorial invariants as the pi are increased. The construction of projection operators
corresponding to different R-symmetry representations also becomes a non-trivial exercise. Such tensorial complications in the analysis of superconformal Ward identities and
also in applying the operator product expansion are avoided here by taking
r1 ...rp (x) (p) (x, t) = r1 ...rp (x)tr1 trp ,
(1.2)
where t is an arbitrary complex vector satisfying3

t 2 = 0.
(1.3)
(For a more mathematical discussion of using such vectors for the treatment of representations of SO(n) see [21], see also Appendix A in [22]). Clearly r1 ...rp can be recovered from
(p) . The four point function then becomes a homogeneous polynomial in t1 , t2 , t3 , t4 , of
respective degree p1 , p2 , p3 , p4 , invariant under simultaneous rotations on all ti s. Due to
the condition (1.3) for each ti the conformally covariant four point function is reducible to
an invariant function F(u, v; , ) with , the two independent invariants, homogeneous
of degree zero in each ti , which are analogous to the conformal invariants u, v,
=
t1 t3 t2 t4
,
t1 t2 t3 t4
t1 t4 t2 t3
.
t1 t2 t3 t4
(1.4)
In general F(u, v; , ) is a polynomial in , , with degree determined by the pi , where

the number of independent terms match exactly the number of tensorial invariants necessary for the general decomposition of the four point function for the corresponding
symmetric traceless tensorial fields, for pi = p there are 12 (p + 1)(p + 2) terms.
Just as the invariants u, v are expressed in terms of x, x it is convenient to write , in
a similar form involving new variables , ,
= ,
= (1 )(1 ).
(1.5)
For the N = 2 case further restrictions impose = .

The superconformal Ward identities
are then simply expressed in terms of F(x,

x;
, )
= F(u, v; , ), with F symmetric
in x, x and also a symmetric polynomial in , .
The superconformal identities constrain
F(x,
x;
, )
for = 1/x to be expressible only in terms of a function involving x and
3 Such null vectors may also be motivated by considering the harmonic superspace approach and were used
similarly, for instance, in [8,9]. Our application is independent of the harmonic superspace formalism and is
essentially motivated just by the requirement of simplifying the treatment of arbitrary rank symmetric traceless
tensors, we do not anywhere consider the conjugate of t .
412
(for N = 2 just a single variable function of x appears). Taking into account the symmetry
conditions the unconstrained or dynamical part of F is therefore of the form
F(u, v; , )dynamical
= (x 1)( x 1)(x
1)( x 1)H(u, v; , ),
(1.6)
with H(u, v; , ) = H(x,

x;
, )
a polynomial in , , or , ,
of reduced degree. In the
N = 2 the corresponding result is
F(u, v; , )dynamical = (x 1)( x 1)H(u, v; , ),
H(u, v; , ) = H(x,
x;
).
(1.7)
Similar results were previously obtained by Heslop and Howe [20] based on expansions
in terms of Schur polynomials for SU(2, 2|2) and PSU(2, 2|4).4 Using the formalism of
harmonic superspace [23] also provides a method for deriving superconformal identities
which are equivalent to those obtained here.
These results have a natural interpretation in terms of the operator product expansion
when the four point function is expanded in terms of conformal partial waves corresponding to operators with various scale dimensions and spins belonging to the various
possible representations of the R-symmetry group. The conformal partial waves are explicit functions of u, v, more simply given in terms of x, x [18]. To disentangle the different
R-symmetry representations the correlation functions are also expanded in terms of two
variable harmonic polynomials corresponding to the possible R-symmetry representations
which may be formed. Explicit simple expressions are found here for these harmonic polynomials in the N = 4 case using the variables , (for N = 2 they reduce to a single
variable Legendre polynomial). If H in (1.6) is simultaneously expanded in such harmonic polynomials and conformal partial waves then the factors multiplying H, for each
term in the expansion, through various recurrence relations generate contributions corresponding to all operators belonging to a single superconformal long multiplet. For such
a long multiplet the scale dimension has only a lower bound due to unitarity and, in a
perturbative expansion, H therefore includes dynamical renormalisation effects leading to
anomalous scale dimensions. The remaining parts in the solution of the superconformal
identities which involve functions of x or x are also analysed here. For N = 2 there is a
single variable function f (x) whereas for N = 4 the superconformal identities allow for
f (x, ), which is polynomial in and satisfies the constraint that it is a constant k when
= 1/x. These functions correspond to semi-short multiplets with protected scale dimensions, determined by and the R-symmetry representation, and also in special cases to
short multiplets. The full set of possible semi-short supermultiplets are obtained by decomposing long multiplets at the unitarity threshold and the short multiplet contributions
are realised by extending the semi-short results to = 1, 2.
4 In Eq. (49) of [20] S
020 (Z) = (X1 Y1 )(X1 Y2 )(X2 Y1 )(X2 Y2 ) which appears as an overall factor
in the Schur polynomial for long representations.
413
The superconformal Ward identities are most powerful for four point functions which
are extremal, so that there is only one possible SU(4)R invariant coupling, or next-toextremal when there are just three invariant couplings. Various calculations [24] have
shown that the correlation functions are identical with the results obtained in free field
theory. The superconformal Ward identities for the extremal case show that the correlation function depends only on the constant k whereas the next-to-extremal case is given
just by the function f (x, ), without any dynamical contribution of the form exhibited in
(1.7). These results are naturally interpreted in terms of the operator product expansion
since these correlation functions require only operators belonging to short or semi-short
multiplets, with no contributions from operators in long multiplets which have anomalous
dimensions depending on the coupling.
For four point functions of identical primary operators there are further constraints arising from crossing symmetry which corresponds to the permutation group S3 . In such four
point functions S3 acts on u, v and also , so that F(u, v; , ) is invariant up to an
explicit overall factor. All invariant contributions to F may be formed by combining S3
irreducible representations constructed from functions of u, v and also , . Crossing symmetry further constrains the single variable functions f (x) or f (x, ) that arise in solving
the superconformal Ward identities. We argue here, based on superconformal representation theory and analyticity requirements, that these can be extended to a fully crossing
symmetric contribution to F(u, v; , ) in terms of two variable crossing symmetric polynomials which are constructed here. These essentially correspond to generalised free field
contributions to the correlation function. A similar discussion is also applicable to the nextto-extremal correlation function in the case of thee identical operators and there is still a
S3 symmetry.
The discussion in this paper applies only to four point functions of 12 -BPS operators
obeying the strongest shortening conditions. For N = 4 there are also 14 -BPS operators
whose correlation functions should satisfy superconformal Ward identities but such cases
are not pursued here.
In detail the structure of this paper is then as follows. In Section 2 we derive the superconformal Ward identities for N = 2 superconformal symmetry and these are applied in
Section 3 by analysing the contributions of different supermultiplets in the operator product
expansion. The discussion is extended to the N = 4 case in Sections 4 and 5. For the operator product expansion it is shown how there are potential contributions from non-unitary
semi-short supermultiplets although they may be cancelled so that only unitary multiplets
remain. In Section 6 we take into account the restrictions imposed by crossing symmetry making use of S3 representations. In Section 7 we summarise some results obtained
previously for large N in the framework of this paper and a few comments are made in a
conclusion. Various technical issues are addressed in four appendices. In Appendix A we
discuss how derivatives involving the null vector tr are compatible with t 2 = 0. In Appendix B we consider two variable harmonic polynomials, depending on , given in (1.4),
which are used in the expansion of general four point correlation functions. Appendix C
describes some differential operators which play an essential role in our analysis whereas
in Appendix D we consider non-unitary semi-short representations for PSU(2, 2|4) which
are important in our operator product analysis.
414
2. Superconformal Ward identities, N = 2

The algebraic complications involved in the analysis of Ward identities are much simpler for N = 2 superconformal symmetry. In this case the R-symmetry group is just U (2)
and discussion of the representations is much easier. In order to facilitate the comparison
with the N = 4 case later we consider BPS chiral primary operators which belong to representations of SU(2)R symmetry for R = n, an integer. The BPS condition requires that
the scale dimension = 2n. Such fields form superconformal primary states for a short
supermultiplet with necessarily unrenormalised scale dimensions. The fields in this case
are represented by symmetric traceless tensors r1 ...rn with ri = 1, 2, 3. To derive the Ward
identities we need to consider just the superconformal transformations at the lowest levels
of the multiplet. First
r1 ...rn = (rn r1 ...rn1 ) + (r1 ...rn1 rn ) ,
(2.1)
where ir1 ...rn1 , ir1 ...rn1 are spinor fields, traceless and symmetric on the indices
r1 . . . rn1 , satisfying, with i = 1, 2 and r the usual Pauli matrices,
r rr1 ...rn2 = 0,
r1 ...rn2 r r = 0.
(2.2)
Thus both and belong to SU(2)R representations with R = n 12 . In (2.1) we have5
i (x) = i i i x
,
i (x) = i + i x
i ,
(2.3)
where i , i , i ,
i are the R = anticommuting parameters for an N = 2 superconformal transformation. In addition to (2.1) we use
1
2
r1 ...rn1 = i r1 ...rn1 s s + 4nr1 ...rn1 s s

n1
(r Jr ...r )s s ,
+ Jr1 ...rn1
(2.4)
2n 1 1 2 n1
where Jr1 ...rn1 , a symmetric traceless rank n 1 tensor, is a R = n 1 current. Using
(2.4) together with its conjugate we may verify closure of the superconformal algebra
acting on r1 ...rn ,
[2 , 1 ]r1 ...rn = vr1 ...rn n( + )r1 ...rn + nt(rn |s r1 ...rn1 )s ,
(2.5)
where
which is quadratic in x, and , , trs = tsr , which are linear in x, are constructed from 1 , 1 , 2 , 2 .
For the general analysis here we define (n) (x, t) as in (1.2), where tr is here a 3-vector,
and in a similar fashion also (n1) (x, t) = r1 ...rn1 (x)tr1 trn1 and (n1)
(x, t) =
(n1)
r1 ...rn1 (x)tr1 trn1 while J (x, t) = Jr1 ...rn1 (x)tr1 trn1 . With this notation (2.1) may be rewritten as
va ,
(n) (t) = t (n1) (t) + (n1) (t) t ,
(2.6)
5 Thus 4-vectors are identified with 2 2 matrices using the Hermitian -matrices , , = 1,
a a (a b)
ab
a
= x a ( )
= x , with inverse x a = 1 tr( a x ). We have xy = x a y =
x x = x a (a ) , x
a
a
2
12 tr(xy), det x = x 2 , x1 = x/x 2 .
415
and (2.4) becomes

1
(n1) (t) = i (n) (t) + 4 (n) (t)

n t
t

1
(n1)
J
+ 1
(t) .
t
2n 1
t
(2.7)
A precise form for differentiation with respect to tr satisfying (1.3) is given in Appendix A.
The conditions (2.2) are now
(n1)
(t) = 0,
(n1) (t) = 0.
t
t
The four point correlation functions of interest here then have the form
(n )

1 (x1 , t1 ) (n2 ) (x2 , t2 ) (n3 ) (x3 , t3 ) (n4 ) (x4 , t4 )
r23 2n2 2n3 r34 2n3 2n4

F (u, v; t),
r13 2n1 r24 2n3
(2.8)
= n1 + n2 + n3 + n4 ,
(2.9)
where
rij = (xi xj )2 ,
(2.10)
and u, v are the two independent conformal invariants,

u=
r12 r34
,
r13 r24
v=
r14 r23
,
r13 r24
(2.11)
which is equivalent to (1.1). F (u, v; t) is also a SU(2)R scalar which is specified further
later, clearly there is a freedom to modify it by suitable powers of u or v at the expense of
changing the terms involving rij in (2.9). The choice made on (2.9) has some convenience
in the later discussion.
The fundamental superconformal Ward identities arise from expanding

(n1 1) (x1 , t1 ) (n2 ) (x2 , t2 ) (n3 ) (x3 , t3 ) (n4 ) (x4 , t4 ) = 0,
(2.12)
using (2.6) and (2.7). This gives, suppressing the arguments ti for the time being,

(n1 )
1
(x1 ) (n2 ) (x2 ) (n3 ) (x3 ) (n4 ) (x4 ) (x1 )
i1
n1
t1

(n1 )
(x1 ) (n2 ) (x2 ) (n3 ) (x3 ) (n4 ) (x4 )
+ 4
t1

1
(n1 1)
J
t1
+ 1
(x1 ) (n2 ) (x2 ) (n3 ) (x3 ) (n4 ) (x4 ) (x1 )
2n1 1
t1
(n 1)

(n2 1)
1
+
(x1 )
(x2 ) (n3 ) (x3 ) (n4 ) (x4 ) t2 (x2 )

(n 1)
+ (n1 1) (x1 ) (n2 ) (x2 ) 3 (x3 ) (n4 ) (x4 ) t3 (x3 )

(n 1)
+ (n1 1) (x1 ) (n2 ) (x2 ) (n3 ) (x3 ) 4 (x4 ) t4 (x4 )
= 0.
(2.13)
416
To apply this we make use of general expressions compatible with conformal invariance
for each four point function which appears. Thus
(n 1)

(n 1)
1 (x1 ) 2 (x2 ) (n3 ) (x3 ) (n4 ) (x4 )

1
r23 2n2 2n3 r34 2n3 2n4 1
,
x
R
+
(x
x
x
)
S
= 2i
12
2
13
34
42
2
r12
r13 r24
r13 2n1 r24 2n3
(n 1)

(n 1)
1 (x1 ) (n2 ) (x2 ) 3 (x3 ) (n4 ) (x4 )

1
r23 2n2 2n3 r34 2n3 2n4 1
,
x
R
+
(x
x
x
)
S
= 2i
13
3
14
42
23
3
r13
r14 r23
r13 2n1 r24 2n3
(n 1)

(n 1)
1 (x1 ) (n2 ) (x2 ) (n3 ) (x3 ) 4 (x4 )

1
r23 2n2 2n3 r34 2n3 2n4 1
x
R
+
(x
x
x
)
S
= 2i
14 4
13 32 24 4 ,
r14
r13 r24
r13 2n1 r24 2n3
(2.14)
where Rn , Sn are functions of u, v and also scalars formed from ti (to verify completeness
of the basis chosen in (2.14) we use relations such as x13 x 34 x42 + x14 x 43 x32 = r34 x12 ). In
addition we have
(n1 1)

J
(x1 ) (n2 ) (x2 ) (n3 ) (x3 ) (n4 ) (x4 )
= 2i
r23 2n2 2n3 r34 2n3 2n4

(X1[23] I + X1[43] J ),
r13 2n1 r24 2n3
(2.15)
for
Xi[j k] =
xij x j k xki
1
1
=
xij
xik ,
rij rik
rij
rik
(2.16)
which transforms under conformal transformations as a vector at xi and is antisymmetric

in j k.
Using (2.14) and (2.15) in (2.13), noting that
i1
1
r13
2n1
(x1 ) + 4n1
1
r13
2n1
= 4n1 i
1
r13
2n1 +1
(x3 ),
(2.17)
and 1 u = 2uX1[23] , 1 v = 2vX1[43] , we may decompose (2.13) into independent contributions involving (x1 ) and (x3 ) (note that x12 (x2 )/r12 = x13 (x2 )/r13 +
X1[23] (x1 ) and also for x2 x4 ) giving two linear relations,
1
(X1[23] uu + X1[43] vv )F
n1 t1

1
(X1[23] I + X1[43] J )
t1
+ 1
2n1 1
t1
+ X1[23] R2 t2 + X1[43] R4 t4 + (uX1[23] vX1[43] )(S2 t2 S4 t4 )
= 0,
(2.18a)
417

1
F + R2 + (u v)S2 t2 + R3 t3 + R4 + (1 u + v)S4 t4
x13
t1
r13
1

+ vS2 t2 + S3 t3 vS4 t4
x14 x 42 x23
r14 r23
= 0.
(2.18b)
It is easy to decompose (2.18a), (2.18b) into independent equations but crucial simplifications are obtained essentially by diagonalising each 2 2 spinorial equation in terms
of new variables x, x which, as mentioned in the introduction, are the eigenvalues of
x12 x42 1 x43 x13 1 . These are related to the conformal invariants u, v defined in (2.11) by
(1.1). In (2.18b) the spinorial matrix x 41 1 x 42 x 32 1 x 31 may be replaced by 1/(1 x) and
in (2.18a) we may effectively replace X1[23] 1/x and X1[43] 1/(1 x) and in each
case also for x x.
Using
= x
(1 x)
,
x
u
v
and the definitions
T2 = R2 + xS2 ,
(2.19)
T 3 = R3 +
1
S3 ,
1x
T4 = R4 + (1 x)S4 ,
1
1
I
J,
x
1x
we may then obtain from (2.18a), (2.18b)
K=
(2.20)
F
n1 t1 x

1
1
1
K T2 t2 +
= 1
(2.21a)
t1
T4 t4 ,
2n1 1
t1
x
1x
2 F = T2 t2 + T3 t3 + T4 t4 .
(2.21b)
t1
Together with the corresponding equations obtained by x x in (2.21a), (2.21b) with also
which are defined just as in (2.20) for x x.
Ti Ti , K K,
The equations in (2.21a), (2.21b) are equations for F /t1 . The integrability conditions,
which are required by virtue of ( t1 )2 = 0, are satisfied since we have, for i = 2, 3, 4,
Ti = 0,
t1
Ti
= 0,
ti
(2.22)
as a consequence of (2.8). To reduce (2.21a), (2.21b) into equations which ultimately allow
Ti and K to be eliminated we first write, since Ti is a 2 2 matrix,
Ti ti = Vi + Wi 1,
(2.23)
where Wi and Vi,r are respectively a scalar and a vector. From the results of Appendix A
we further decompose Vi uniquely in the form
Vi =
1
Ui + Vi ,
n1 t1
t1 Vi = Ui ,
t1 Vi = 0.
(2.24)
418
The first equation in (2.22) then separates into SU(2) scalar and vector equations,
Vi = 0,
t1
Vi +
Wi = 0,
t1
t1
(2.25)
where we may let Vi Vi without change since both 12 Ui and 1 1 Ui are zero. From
(2.25) we may then find
L1 Wi = in1 Vi ,
(2.26)
where we define the SU(2)R generators by
.
ti
Substituting (2.23) into (2.21a) gives
Li = t i
(2.27)
1
F = U2 +
U4 ,
x
x
1x
(2.28)
and
1
1
n1
K = W2 +
W4 ,
2n1 1
x
1x
which is just an equation giving K, and also
(2.29)
1
1
1
(2.30)
iL1 K = V2 +
V4 .
2n1 1
x
1x
It is easy to see that this follows from (2.29) as a consequence of (2.26). Similarly substituting (2.23) into (2.21b) gives three equations
2n1 F =
4

Ui ,
(2.31)
i=2
and
4

Vi = 0,
(2.32)
Wi = 0.
(2.33)
i=2
as well as
4

i=2
Clearly (2.32) follows from (2.33).

An essential constraint may also be obtained from the second equation in (2.22) which
gives
(ni + 1)Ti ti = (Ti ti )i L i .
(2.34)
With the decomposition (2.23) this leads to

(ni + 1)Wi = iLi Vi ,
(2.35a)
(ni + 1)Vi = Li Vi iLi Wi .

Contracting (2.35b) with Li , and using Li Li = Li ,
(2.35a). In addition we have from Ti ( ti )2 = 0
ti Vi = 0,
419
(2.35b)
L2i Wi
= ni (ni + 1)Wi , gives
iti Vi = ti Wi .
(2.36)
With the aid of the results in Appendix A we may obtain (2n1 + 1)i (ti Wi iti Vi ) =
(2ni + 3)((ni + 1)Wi + iLi Vi ) so that (2.36) implies (2.35a). Similarly, since i (ti
Vi ) = (i ti ) Vi ) + i (ti Vi ) i (ti Vi ), we have from Appendix A (2ni + 1)i (ti
Vi ) = (2ni + 3)(Li Vi + (ni + 4)Vi ) and (2ni + 1)i (ti Wi ) = (2ni + 3)Li Wi .
Hence it is clear that (2.36) also implies (2.35b).6
Using (2.24) and (2.26) for Vi in (2.35a) we obtain

1
L1 Li + n1 (ni + 1) Wi = (L1 + Li )2 + (n1 + ni )(n1 + ni + 1) Wi
2
= i
Li Ui .
t1
(2.37)
Ui (u, v; t), which is defined by (2.23), is a homogeneous polynomial in t1 , ti of O(t1n1 , tini )

such that the SU(2)R representation with R(1i) = n1 + ni is absent. In consequence the
operator (L1 + Li )2 + (n1 + ni )(n1 + ni + 1), which commutes with 1 Li , in (2.37) may
be inverted to give Wi in terms of Ui . Alternatively we may obtain from (2.36)
iti 1 Ui = t1 L1 Wi + n1 ti Wi .
(2.38)
To analyse these equations further we now consider the decomposition of F and also
Ui in terms of SU(2)R scalars. We first assume the ni are ordered so that
n1 n2 n3 n4 ,
(2.39)
and further assume

n4 = n1 + n2 + n3 2E,
(2.40)
for integer E = 0, 1, 2, . . . , where E is a measure of how close the correlation function is

n
to the extremal case. With (2.39) and (2.40) F , which is O(t1n1 , t2n2 , t3 3 , t4n4 ), can in general
be written in the form
F (u, v; t) = (t1 t 4 )n1 E (t2 t 4 )n2 E (t1 t 2 )E (t3 t 4 )n3 F(u, v; , ),
(2.41)
where F is a polynomial in , , defined in (1.4), with all terms p q satisfying p +q E.

If E > n1 then all terms in F must contain a factor En1 to cancel negative powers of
t1 t4 in (2.41). Since ti are three-dimensional vectors t1[r t2s t3t t4u] = 0 so that , are not
independent but obey the relation
2 + 2 + 1 2 2 2 = 0.
6 The converse follows using t L = 0, t L = t t and t (L V ) = t V .
i i
i
i
i i i
i
i
i
i
i
(2.42)
420
This may be solved in terms of a single variable by7

= 2,
= (1 )2 ,
(2.43)
F(u, v; , ) = F(x,
x;
).
(2.44)
so that
F(x,
x;
) is symmetric in x, x and, for E n1 , is a polynomial in of degree 2E, so
that there are 2E + 1 independent coefficients, while if E > n1 then it must be of the form
(1 )2(En1 ) p() with p a polynomial of degree n1 , so that the number of coefficients
is 2n1 + 1. These results correspond exactly of course to the number of SU(2)R invariants
which can be formed in the four point function, subject to (2.40), together with (2.39), that
can be found using standard SU(2) representation multiplication rules.
A similar expansion to (2.41) can be given for each Ui
Ui (x, x;
t) = (t1 t 4 )n1 E (t2 t 4 )n2 E (t1 t 2 )E (t3 t 4 )n3 Ui (x, x;
, ),
, ) = Ui (x, x;
).
Ui (x, x;
(2.45)
The analysis of (2.29) and (2.33) depends on using (2.37), or (2.38), as shown in
Appendix C, to relate Wi and Ui . Defining
Wi = it2 (t3 t4 )(t1 t4 )n1 E (t2 t4 )n2 E (t1 t2 )E1 (t3 t4 )n3 1 Wi ,
(2.46)
then we obtain
2(2n1 1)Wi = D i Ui ,
(2.47)
where D i are linear operators given by

2(E n1 )
2n1 2(E n1 )
d
d
+
,
D 3 =
+
,
D 2 =
d
1
d
1
2E
d
D 4 =
+
.
d 1
The superconformal identities (2.28), (2.31) and (2.33) then become
(2.48)
1

1
F = U2 +
U4 ,
x
x
1x
2n1 F = U2 + U3 + U4 ,
(2.49b)
D 2 U2 + D 3 U3 + D 4 U4 = 0.
(2.49c)
(2.49a)
By acting on (2.49b) with D 3 and using (2.49c) we may obtain

1
1
U4 ,
D 3 F = U2
(1 )
(2.50)
7 In three dimensions a null vector t may be represented [21] in terms of two component spinors u = (u1 , u2 )
by ta = ua u for u = uT for the 2 2 antisymmetric matrix. Then t1 t2 = 2(u2 u 1 )2 . In this case
u u u u
u u u u
= u3 u 1 u2 u 4 = 1 u2 u 3 u1 u 4 . Not that also it1 t2 t3 = 4u1 u 2 u2 u 3 u3 u 1 .
2 1 3 4
2 1 3 4
421
and substituting in (2.49a) gives

x
1
x
D3 F =
+
U4 .
x
1x 1
(2.51)
The right-hand side of (2.51) vanishes when = 1/x leaving an equation for F alone.
With the explicit form for D 3 in (2.48) we have

1 2(n1 E)
1
x 2n1 1
= 0.
F x, x;
x
x
x
(2.52)
Together with its partner or conjugate equation involving / x (2.52) provides the final
result for the constraints due to superconformal identities for the four point function when
N = 2.
For the N = 2 case we may also require instead of (2.40)
n4 = n1 + n2 + n3 2E 1,
(2.53)
since F can then be written as
F (u, v; t) = (t1 t4 )n1 E (t2 t4 )n2 E1 (t1 t2 )E (t3 t4 )n3 1 t2 t3 t4 F(x,

x;
).
(2.54)
There is an essentially unique expression in (2.54), with a single function F as a consequence of identities for the various possible vector cross products for null vectors which
take the form
1
t1 t2 t3 t2 t4 = ( + 1)t2 t3 t4 t1 t2 ,
2
1
t1 t2 t4 t2 t3 = ( 1)t2 t3 t4 t1 t2 ,
2
1
t1 t3 t4 t2 t4 = ( + 1)t2 t3 t4 t1 t4 .
2
(2.55)
Since, as shown in Appendix B, effectively t1 t4 t2 t3 t4 = O(1 ) we can take in (2.54),
if n1 E 1, (1 )F(x,
x;
) to be a polynomial of degree 2E + 1. If n1 E < 1 then
F(x, x;
) must contain a factor (1 )2(En1 )1 . It is easy to see that the number of
independent coefficients matches with the number of independent terms in the four point
function obtained by counting possible representations in each case.
There is a similar expansion as (2.54) for Ui . Instead of (2.46) and (2.47) we now have
Wi =
i
(t1 t4 )n1 E1 (t2 t4 )n2 E (t1 t2 )E (t3 t4 )n3 D i Ui ,
2n1 1
(2.56)
with D i exactly as in (2.48). In consequence the superconformal identities reduce to

(2.49a)(2.49c) and we may derive the final result (2.52), albeit with E given by (2.53).
422
3. Solution of identities, N = 2
Although in the N = 2 case the identities can be solved rather trivially we show here
how they may be put in a form which makes the connection with the operator product
expansion, and the possible supermultiplets which may contribute to it, rather obvious. For
the purposes of analysing the operator product expansion for x1 x2 an alternative form
to (2.9) is more convenient so we write
(n )

1 (x1 , t1 ) (n2 ) (x2 , t2 ) (n3 ) (x3 , t3 ) (n4 ) (x4 , t4 )
n1 n2 n3 n4
r24
r14
1
=
(3.1)
G(u, v; t),
r12 n1 +n2 r34 n3 +n4 r14
r13
where
G(u, v; t) = un1 +n2 v n1 +n4 n2 n3 F (u, v; t).
(3.2)
For application of the superconformal Ward identities here it is convenient here to replace the variable by y where
y = 2 1,
(3.3)
and x, x by z, z given by
z=
2
1,
x
z =
2
1.
x
(3.4)
Assuming now
G(u, v; t) = (t1 t4 )n1 E (t2 t4 )n2 E (t1 t2 )E (t3 t4 )n3 G(u, v; y),
(3.5)
the solution of (2.52) and its conjugate equation, maintaining the symmetry under z z ,
becomes
G(u, v; z) = un1 +n2 2E f (z),
G(u, v; z ) = un1 +n2 2E f (z),
(3.6)
where f is an unknown single variable function. Since G(u, v; y) is just a polynomial in y

(3.6) requires
(y z )f (z) (y z)f (z)
z z
+ (y z)(y z )K(u, v; y),
G(u, v; y) = un1 +n2 2E
(3.7)
where K(u, v; y) is undetermined, if G(u, v; y) is a polynomial of degree 2E in y then

clearly K is a polynomial of degree 2E 2.
The operator product expansion applied to this correlation function is realised by ex()
panding it in terms of conformal partial waves G (u, v; 21 , 43 ), ij = i j , which
represent the contribution to a four point function for four scalar fields, with scale dimensions i , from an operator of scale dimension and spin , and all its conformal
423
descendants. Explicit expressions, in four dimensions, were found in [18] which are simple in terms of the variables x, x defined in (1.1),8 which satisfy

()
21 43 () u 1
, ; 21 , 43
G
G u, v; 21 , 43 = (1) v
v v

1
()
= v 2 (43 21 ) G u, v; 21 , 43 .
(3.8)
For this case the expansion is also over the contributions for differing SU(2)R R-representations and has the form, if n1 E,
G(u, v; y) =
n
1 +n2
R=n4 n3 ,
(2n 2E,2n2 2E)
aR,, PR+n1 3 n4
(y)

()
G u, v; 2(n2 n1 ), 2(n4 n3 ) ,
(3.9)
with Pn
a Jacobi polynomial. For a a negative integer Pn (y) (1 y)a and
n + a 0. Hence when n1 < E we require a similar expansion to (3.9) but with R =
n2 n1 , . . . , n1 + n2 and then G(u, v; y) En1 as required in (3.5) to avoid negative
powers of t1 t4 . The different terms appearing in the sum in (3.9) then determine the necessary spectrum of operators required by this correlation function. The symmetry properties
(a,b)
(b,a)
of this operator product expansion follow from (3.9) and Pn (y) = (1)n Pn (y).
We first consider the case when n1 = n2 = n3 = n4 = n, so that E = n. To apply (3.6)
we first consider the expansion in terms of Legendre polynomials (to which the Jacobi
polynomial reduce in this case),
(a,b)
(a,b)
G(u, v; y) =
2n

aR (u, v)PR (y),
R=0
K(u, v; y) =
2n2

AR (u, v)PR (y).
(3.10)
R=0
The PR (y) in (3.10) correspond to the 2n + 1 possible SU(2)R invariants for the four point
function (3.1) and, as a consequence of results in Appendix B, the coefficients aR represent
the contribution to the correlation function from operators belonging just to the SU(2)R Rrepresentation in the operator product expansion for (n) (x1 , t1 ) (n) (x2 , t2 ). From (3.7) it is
easy to see that the single variable function f involves terms linear in y and so contributes
8
()
G (u, v; 21 , 43 )

u 2 () 1 1
2 x xF 2 ( + 21 + ), 12 ( 43 + ); + ; x
x x

F 12 ( + 21 2), 12 ( 43 2); 2; x x x .
1
424
only for R = 0, 1 giving

f
a0 =
zf (z) z f (z)
,
z z
a1 =
f (z) f (z)
.
z z
(3.11)
Using the expansion in (3.10) for K in (3.7) and standard recurrence relations for Legendre
polynomials gives corresponding expressions for aR . For the terms involving AR we have
(R + 1)(R + 2)
(R 1)R
AR
AR ,
AR ,
aR2
=
(2R + 1)(2R + 3)
(2R 1)(2R + 1)
2(R + 1) 1 v
2R 1 v
AR
AR
aR+1
AR ,
AR ,
=
aR1
=
2R + 1 u
2R + 1 u

1
1+v 1
aRAR = 2
+
AR .
u
2 2(2R 1)(2R + 3)
AR
aR+2
=
(3.12)
For R 2 aR is therefore given in terms AR2 , AR1 , AR while for R = 0, 1, with (3.11),
we have

1+v 2
21v
2
f
A0
A1 + A2 ,
a0 = a 0 + 2
u
3
3 u
15

2
1v
1
v
1
+
v
4
6
f
A0 + 2
A1
A2 + A3 .
a1 = a 1 2
(3.13)
u
u
5
5 u
35
In (3.12) and (3.13) any contributions involving AR for R > 2n 2 should be dropped.
The significance of the results given by (3.12) and (3.13) is that they correspond exactly to the N = 2 supermultiplet structure of operators appearing in the operator product
()
()
expansion. Each aR (u, v) may then be expanded in terms of G
(u, v) G
(u, v; 0, 0)

()
bR,, G
(u, v).
aR (u, v) =
(3.14)
,
()
The conformal partial waves G (u, v) satisfy crucial recurrence relations [19],
2
1 v ()
G (u, v)
u
1
(+1)
(1)
(+1)
(+1)
= 4G1 (u, v) + G1 (u, v) + as G+1 (u, v) + at1 G1 (u, v),
4

1+v
()
1 G (u, v)
2
u
1
1
()
(+2)
(2)
()
= 4G2 (u, v) + 4as G (u, v) + at1 G (u, v) + as at1 G+2 (u, v),
4
4
(3.15)
where
1
s = ( + ),
2
1
t = ( ),
2
In (3.15) at1 > 0 if > + 3.
as =
s2
.
(2s 1)(2s + 1)
(3.16)
425
If AR is restricted to a single partial wave so that

()
AR G+2 ,
(3.17)
then, using (3.15) with (3.12),

( )
b( ; ) G ,
aRAR aR A
R, =
( , )

R R = 1,
R R = 2,
; = ( + 2; ),

; = ( + 3, + 1; 1),
; = ( + 4, ; ),
R R = 0,

; = ( + 2; 2, ).
(3.18)
This gives exactly the expected contributions corresponding to those operators present in a
long N = 2 supermultiplet, which we may denote A
R, , whose lowest dimension operator
has dimension , spin belonging to the SU(2)R R-representation. From (3.12) and the
positivity constraints for (3.15) we may then easily see that in (3.14) b( , ) > 0 for >
+ 1. For a unitary representation, so that all states in A
R, have positive norm, (we
consider here multiplets whose U (1)R charge is zero) the requirement is
2R + + 2.
()
G (u, v)
(3.19)
1
2 ()
=u
F (u, v) with F (u, v) expressible as a power series in u, 1 v
Since
we must have from (3.17) for u 0,
AR (u, v) uR+2+ ,
0.
(3.20)
The contribution of the single variable function f (3.11) represents operators just with
twist = 2. From the results in [18] we have
()
G+2 (u, v) = u
g+1 (x) g+1 (x)
g+1 (x) g+1 (x)

= 2
,
x x
z z
for

1
2
g (x) = 12 x
xF (, ; 2; x) = F
z
1
1
1
1 1
2 , 2 + 2 ; + 2 ; z2
(3.21)

,
(3.22)
where F is just an ordinary hypergeometric function.9 As shown in [19] g satisfies

zg (x) = g1 (x) a g+1 (x).
(3.23)
In general we therefore expand the single variable function f in (3.11) in the form
f (z) =
b g (x).
=0
9 g (x) Q

1 (z) with Q an associated Legendre function.
(3.24)
426
For this to be possible f (z) must be analytic in 1/z, or equivalently in x. If we consider

f
just f 2g+2 and use (3.23) in (3.11) then aR aR (C0, ) where
(+1)
a1 (C0, ) = G+3 ,
()
(+2)
a0 (C0, ) = G+2 + a+2 G+4 .
(3.25)
These results for a0 , a1 then correspond to the contributions of operators belonging to a

semi-short N = 2 supermultiplet C0, whose lowest dimension operator is a SU(2)R singlet
with spin and = + 2, i.e., at the unitarity threshold (3.19).
In general we denote by CR, the semi-short multiplet whose lowest dimension operator
has spin , belongs to the representation R, and has = 2R + , so that the bound (3.19) is
saturated. At the unitarity threshold given by (3.19) a long multiplet A
R, may be decomposed into two semi-short supermultiplets CR, and CR+1,1 , [25]. This is reflected in the
contributions to the four point function since, with aR (A
R, ) defined by (3.17) and (3.18),
2R++2
R+1
aR AR,
(3.26)
= 4aR (CR, ) +
aR (CR+1,1 ),
2R + 1
where we take
1
()
()
(+2)
aR (CR, ) = G2R++2 + aR G2R++4 + aR++2 G2R++4 ,
4

R
1 (1)
1
(+1)
(+1)
,
G2R++3
+ G2R++3
+ aR++2 G2R++5
aR1 (CR, ) =
2R + 1
4
4
(R 1)R
()
aR2 (CR, ) =
G
,
4(2R 1)(2R + 1) 2R++4
R + 1 (+1)
G
.
aR+1 (CR, ) =
(3.27)
2R + 1 2R++3
For R = 0 (3.27) coincides with (3.25). Thus the contribution of any semi-short supermultiplet CR, , R = 0, 1, . . . , 2n 1, to the four point function may be obtained by combining
the results for long supermultiplets at unitarity threshold with (3.25). There is no reason
why any particular CR, , except C0,0 which contains the energymomentum tensor and the
conserved SU(2)R current, should be present but if f (z) is non-zero it is necessary for there
to be at least one semi-short contribution involving operators with protected dimensions.
A special case arises if we set = 1. Formally, as shown in [25], CR,1 BR+1
where BR denotes the short supermultiplet whose lowest dimension operator belongs to
the R-representation with = 2R, = 0, obeying the full N = 2 shortening conditions.
The conformal partial waves as shown in [19] satisfy
1 1 ()
(2)
(1)
(3.28)
G = G ,
G = 0,
4
and hence from (3.27) we have
aR (CR,1 ) =
R+1
aR (BR+1 ),
2R + 1
(3.29)
where
(0)
aR (BR ) = G2R ,
aR1 (BR ) =
R
(1)
G
,
2R + 1 2R+1
aR2 (BR ) =
(R 1)R
(0)
G
.
4(2R 1)(2R + 1) 2R+2
427
(3.30)
The operators whose contributions appear in (3.30) are just those expected for the short
supermultiplet BR and there are possible contributions to the four point function for R =
1, 2, . . . , 2n. Since
(0)
G0 (u, v) = 1,
(3.31)
then it is easy to see from (3.30) that

aR (B0 ) = aR (I) = R0 ,
(3.32)
where aR (I) denotes the contribution of the identity in the operator product expansion.
Besides (3.29) we may also note that
aR (CR,2 ) = 4aR (BR ).
(3.33)
G0(2)
= 4.
For R = 0 this is in accord with (3.32) since
Apart from the case of the correlation function for four identical operators as considered above there are other solutions of the superconformal Ward identities which are of
interest corresponding to extremal and next-to-extremal correlation functions [24]. The
extremal case corresponds to taking E = 0 in (2.40). There is then a unique SU(2)R invariant coupling which also follows from the requirement that F in (2.41), or G where in
(3.2) G(u, v; t) = (t1 t4 )n1 (t2 t4 )n2 (t3 t4 )n3 G(u, v), must be independent of , and hence
equivalently also of . In this case the result (2.52) for x and its conjugate for x simply
imply
G(u, v) = Cun1 +n2 ,
(3.34)
where C is independent of both x, x and thus a constant. To interpret this in terms of the
operator product expansion for (n1 ) (x1 , t1 ) (n2 ) (x2 , t2 ) we may use the result from [18],
G1 +2 (u, v; 21 , 1 + 2 ) = u 2 (1 +2 ) .
(0)
(3.35)
The result (3.34) then shows that the only operators contributing to the operator product
expansion in the extremal case have = 2(n1 + n2 ), = 0 and necessarily R = n1 + n2 .
Such operators can only be found as the lowest dimension operator in the short supermultiplet Bn1 +n2 .
For the next-to-extremal case we set E = 0 in (2.53). The solution of (2.52) can be
conveniently expressed as
(1 z)G(u, v; z) = un1 +n2 1 f (z),
(3.36)
)n2 1 (t
)n3 1 t
where in (3.2) we have G(u, v; t) = (t1 t4

2 t4
3 t4
2 t3 t4 G(u, v; y). For
E = 0, (1 y)G(u, v; y) is linear in y and from (3.36) and its conjugate we may find
)n1 (t
(1 y)G(u, v; y) = un1 +n2 1
(y z )f (z) (y z)f (z)

,
z z
so that G is determined just by the single variable function g in this case.
(3.37)
428
For the next-to-extremal correlation function there are just two independent SU(2)R
invariant couplings. In a similar fashion to (3.10), we have an expansion, from Appendix B,
in terms of two Jacobi polynomials
(1 y)G(u, v; y)
(2n1 1,2n2 1)
= an1 +n2 1 (u, v)P0
(2n1 1,2n2 1)
(y) + an1 +n2 (u, v)P1
(y),
(3.38)
where aR , R = n1 + n2 1, n1 + n2 represent the contribution of the two possible

R-representations of SU(2)R in this case. From (3.37) we obtain
f (z) f (z)
1
,
un1 +n2
n1 + n2
z z

zf (z) z f (z) n1 n2 f (z) f (z)
.
= un1 +n2
+
z z
n1 + n2
z z
an1 +n2 1 =
an1 +n2
(3.39)
To interpret this in terms of the operator product expansion we may use, extending (3.21),
()
G1 +2 + (u, v; 21 , 1 + 2 + 2)
g+1 (x; 1 , 2 ) g+1 (x;
1 , 2 )
,
x x

1
xF ( + 2 1, ; 2 + 1 + 2 2; x).
g (x; 1 , 2 ) = 12 x
= u 2 (1 +2 )
1
(3.40)
In consequence only operators with twist 1 + 2 can contribute for the solution for aR
given by (3.39). If in (3.39) let we f (z) 2g+2 (x; 2n1 , 2n2 ) then aR aR (Cn1 +n2 1, )
where
an1 +n2 (Cn1 +n2 1, ) =
1
(+1)
G
,
n1 + n2 2n1 +2n2 ++1
an1 +n2 1 (Cn1 +n2 1, )

()
= G2n1 +2n2 +
( + 1)(2n1 + 2n2 + )
(+1)
G
(n1 + n2 + )(n1 + n2 + + 1)(n1 + n2 ) 2n1 +2n2 ++1
( + 2)(2n1 + + 1)(2n2 + + 1)(2n1 + 2n2 + )
+
(n1 + n2 + + 1)2 (2n1 + 2n2 + 2 + 1)(2n1 + 2n2 + 2 + 3)
+ (n2 n1 )
(+2)
G2n1 +2n2 ++2 ,
(3.41)
( )
where G2n1 +2n2 + are as in (3.40) with i 2ni . The contributions appearing in
(3.41) correspond to those expected from the semi-short supermultiplet Cn1 +n2 1, . Using CR,1 BR+1 again we may obtain the contribution for the short multiplet Bn1 +n2 ,
aR (Cn1 +n2 1,1 ) =
1
aR (Bn1 +n2 ),
n1 + n2
giving
(0)
an1 +n2 (Bn1 +n2 ) = G2n1 +2n2 ,
(3.42)
an1 +n2 1 (Bn1 +n2 ) =
4n1 n2
(1)
G
.
(n1 + n2 )(2n1 + 2n2 + 1) 2n1 +2n2 +1
429
(3.43)
For the next-to-extremal correlation function therefore only the protected short and semishort supermultiplets BR and CR1, can contribute to the operator product expansion.
By analysis [26,27] of three point functions the possible N = 2 supermultiplets which
may appear in the operator product expansion of two N = 2 short supermultiplets is determined by the decomposition, for n2 n1 ,
Bn1 Bn2
n
2 +n1
Bn
n=n2 n1

1 1
n2 +n

0
Cn,
n=n2 n1
n2 +n
1 2

A
n,
(3.44)
n=n2 n1
where for A
n, all > 2n + + 2 is allowed. By considering also the corresponding result
for Bn3 Bn4 in all cases discussed above the general solution of the N = 2 superconformal identities accommodates all possible N = 2 supermultiplets which may contribute to
the four point function in the operator product expansion according to (3.44). In the extremal case it is clear that only Bn1 +n2 contributes while for the next-to-extremal case long
multiplets which undergo renormalisation are also excluded.
4. Superconformal Ward identities, N = 4

We here describe an analysis of the superconformal Ward identities for the four point
function of N = 4 chiral primary operators belonging to the SU(4)R [0, p, 0] representation with = p represented by symmetric traceless fields r1 ,...,rp (x), ri = 1, . . . , 6. As
in (1.2) we define (p) (x, t), homogeneous of degree p in t, in terms of a six-dimensional
null vector tr . The superconformal transformation of (p) (x, t) is then expressible in the
form
(p) (x, t) = (x) t (p1) (x, t) + (p1) (x, t) t (x),
(4.1)
where the conformal Killing spinors i (x), i (x) are as in (2.3), with i = 1, 2, 3, 4 and
r ij = r j i , rij = 12 ij kl r kl are SU(4) gamma matrices, r s + s r = 2rs 1, r =
r . In (4.1) (p1) i (x, t), (p1)i (x, t) are homogeneous spinor fields of degree p 1
in t and satisfy constraints similar to (2.8)
(p1)
(x, t) = 0,
t
(p1) (x, t)
= 0,
t
(4.2)
which are necessary for them to belong to SU(4)R representations [0, p 1, 1],
[1, p 1, 0]. At the next level the superconformal transformations involve a current belonging to the [1, p 1, 1] representation which corresponds to a homogeneous field of
degree p 1 with one SU(4)R vector index J (p1) r (x, t) satisfying
tr J (p1) r (x, t) = 0,
(p1)
J
r (x, t) = 0.
tr
(4.3)
430
The superconformal transformation of (p1) (x, t), neglecting terms, is then

(p1) (x, t) =
i (p) (x, t) (x) + 2 (p) (x, t)

p t
t

1
t
+ 1+
J (p1) r (x, t)r (x).
2p + 2
t
(4.4)
Superconformal transformations which generate the full BPS multiplet listed in [28] can
be obtained similarly to [19] but the superconformal Ward identities depend only on (4.1)
and (2.6).
The general four point function of chiral primary operators can be written in an identical
form to (2.9),

(p )
1 (x1 , t1 ) (p2 ) (x2 , t2 ) (p3 ) (x3 , t3 ) (p4 ) (x4 , t4 )
=
r23 p2 p3 r34 p3 p4
F (u, v; t),
r13 p1 r24 p3
= 12 (p1 + p2 + p3 + p4 ).
(4.5)
The derivation of superconformal Ward identities initially follows an almost identical path
as that in Section 2 leading to (2.21a), (2.21b). With similar definitions to (2.14), (2.15),
taking 2ni pi , and (2.20) we find
1

F
p1 t1 x

1
1
1
K T2 t2 +
t1
T4 t4 ,
= 1+
2p1 + 2
t1
x
1x
F = T2 t2 + T3 t3 + T4 t4 .
t1
(4.6a)
(4.6b)
Instead of (2.22) we have the constraints, which follow from (4.2) and (4.3),
Ti = 0,
t1
Ti
= 0,
ti
t1 K =
K = 0.
t1
(4.7)
As with (2.23) we exhibit the dependence on SU(4) gamma matrices by writing

1
Ti ti = Vi + [r s u] Wi,rsu .
6
(4.8)
Since we take10 [r s u v w z] = irsuvwz ,

1
[r s u] = irsuvwz v w z ,
6
so that we must require the self-duality condition
1
Wi,rsu = irsuvwz Wi,vwz .
6
10 Note that ( ) = .
1 2 3 4 5 6
1 2 3 4 5 6
(4.9)
(4.10)
431
Imposing the first equation in (4.7) we have
Vi = 0,
t1
1[r Vi,s] = 1u Wi,rsu .
(4.11)
Just as in (2.24) we write,

Vi =
1
Ui + Vi ,
p1 t1
t1 Vi = Ui ,
t1 Vi = 0,
(4.12)
so that in (4.11) we may let Vi Vi .

Using (4.8) with (4.12), and t1 1 K = p1 K + 12 [r s u] L1,rs Ku , (4.6a),
(4.6b) may be decomposed into three pairs of equations,
1
1
F = U2 +
U4 ,
x
x
1x
p1 F =
4

Ui ,
(4.13)
i=2
and
p1 + 2
1
1
Kr = V2,r +
V4,r ,
2p1 + 2
x
1x
4

Vi,r = 0,
(4.14)
i=2
and
3
1
1
(L1,[rs Ku] )sd = W2,rsu +
W4,rsu ,
2p1 + 2
x
1x
4

Wi,rsu = 0,
(4.15)
i=2
where we define for i = 1, 2, 3, 4

Li,rs = tir is tis ir ,
(4.16)
and for any Xrsu = X[rsu] the self dual part, satisfying (4.10), is given by
(Xrsu )sd = 12 Xrsu +
1
12 irsuvwz Xvwz .
(4.17)
Since 2t1s 1[r Vi,s] = p1 Vi,r we may obtain from (4.11)

p1 Vi,r = L1,su Wi,rsu ,
(4.18)
which gives Vi,r in terms of Wi,rsu . Furthermore from (4.7) L1,rs Ks = Kr and using also,
as a consequence of the commutation relations for L1 , [L1,rs , L1,ru ] = 4L1,su we have
L1,rs L1,ru Ks = 3Ku . With, in addition, 12 L1,rs L1,rs Ku = (p1 1)(p1 + 3)Ku we may
then obtain
3L1,rs L1,[rs Ku] = 2p1 (p1 + 2)Ku .
(4.19)
Since also rsuvwz L1,su L1,vw Kz = 0 it is clear from (4.18) and (4.19) that Eqs. (4.15)
imply (4.14). However, if we define
W i,rsu = 3(L1,[rs Vi,u] )sd (p1 + 2)Wi,rsu ,
(4.20)
432
with Vi,u determined by (4.18), then as a consequence of (4.14) and (4.15) we must also
require
1
1
W2,rsu =
W4,rsu .
x
1x
(4.21)
From the second equation in (4.7) we may obtain 12 Li,rs (Ti ti )r s = (pi + 4)Ti ti
which leads to the relations
Li,rs Vi,s Li,su Wi,rsu = (pi + 4)Vi,r ,
(4.22a)
3(Li,[rs Vi,u] )sd + 3Li,[r|v Wi,su]v = (pi + 4)Wi,rsu ,
(4.22b)
where Li,[u|v Wi,rs]v is self dual as a consequence of (4.10). We also have from
Ti ti ti = 0
ti Vi = 0,
ti[r Vi,s] + Wi,rsu tiu = 0.
(4.23)
For consistency we note that is (ti[r Vi,s] + Wi,rsu tiu ) = 0 is identical with (4.22a). Furthermore using (4.10) we have (i[r Wi,su]v tiv )sd = 12 (i[r Wi,su]v tiv iv ti[r Wi,su]v ) +
1
3 iv (Wi,rsu tiv ) and, from Appendix A, (pi + 2)iv (Wi,rsu tiv ) = (pi + 3)(pi + 4)Wi,rsu
while acting on Wi,suv similarly (pi +2)i[r tiv] = (pi +3) 12 Li,rs . Hence we have demonstrated that (i[r (tis Vi,u] + Wi,su]v tiv )sd = 0 is identical to (4.22b) so that this equation is
also implied by (4.23).
Combining (4.23) with (4.12) gives the essential equation
ti[r 1s] Ui + p1 ti[r Vi,s] + p1 Wi,rsu tiu = 0,
(4.24)
where p1 Vi,s is determined by (4.18).

As in (2.45) we may expand the correlation function F , as defined in (4.5), in terms of
SU(4) invariants
F (u, v; t) = (t1 t4 )p1 E (t2 t4 )p2 E (t1 t2 )E (t3 t4 )p3 F(u, v; , ),
(4.25)
where we assume
p1 p2 p3 p4 ,
2E = p1 + p2 + p3 p4 .
(4.26)
p
p
p
F (u, v; t) = O(t1 1 , t2 2 , t3 3 ,
In (4.25) F(u, v; , ) is a polynomial in , consistent with

p
t4 4 ) and hence E 0 is a integer. For p1 E then F is expressible as a polynomial of
degree E in , , i.e., a linear expansion in the 12 (E + 1)(E + 2) independent monomials p q with p + q E. For p1 < E it is necessary also that q E p1 giving only
1
2 (p1 + 1)(p1 + 2) independent terms. It is easy to see that this matches the number
of invariants that may be constructed by finding common representations in [0, p1 , 0]
[0, p2 , 0] and [0, p3 , 0] [0, p4 , 0] using the tensor product result
[0, p1 , 0] [0, p2 , 0]
p1 p
1 r

[r, p2 p1 + 2s, r].
(4.27)
r=0 s=0
Hence representations [r, p2 p1 + 2s, r] may contribute for s = 0, . . . , n r, r = 0, . . . , n

with n = E if p1 E, otherwise n = p1 .
433
In an exactly similar fashion to (4.25) we may express Ui (x, x;

t) in terms of
, ) so that (4.13) becomes
Ui (x, x;
1
1
p1 F = U2 + U3 + U4 .
(4.28)
F = U2 +
U4 ,
x
x
1x
Furthermore we may also decompose Wi,rsu (x, x;
t) for i = 2, 3, 4 in terms of four
independent self dual tensors,
Wi,rsu = (t1 t4 )p1 E (t2 t4 )p2 E (t1 t2 )E2 (t3 t4 )p3 1

(t1[r t2s t3u] )sd t2 t4 Ai + (t1[r t4s t2u] )sd t2 t3 Bi
+ (t1[r t3s t4u] )sd t2 t3 t2 t4

1
Ci + (t2[r t3s t4u] )sd t1 t2 Wi ,
t3 t4
(4.29)
with Ai , Bi , Ci and Wi polynomials in , of degree E 2 and E 1, if p1 E. From

its definition in (4.8) we must have
C2 = B3 = A4 = 0.
(4.30)
The result (4.15) then requires

A2 + A3 = 0,
B2 + B4 = 0,
C3 + C4 = 0,
W2 + W3 + W4 = 0.
(4.31)
We may similarly decompose Vi,r in the form

Vi,r = (t1 t4 )p1 E1 (t2 t4 )p2 E (t1 t2 )E1 (t3 t4 )p3 1

(t2r t1 t4 t4r t1 t2 )t3 t4 Ii + (t3r t1 t4 t4r t1 t3 )t2 t4 Ji + t1r t2 t4 t3 t4 Vi ,
(4.32)
where we impose t1 Vi = 0. The coefficient of t1r is determined by the requirement

1 Vi = 0,
(pi + 2)Vi = O Ii + Ii ( O O )Ji + Ji ,
(4.33)
with differential operators
+ 2
+ p1 2E + 1,

p1 E
+ ( + 1)
+
p1 + 1.
O = 2
O = ( + 1)
(4.34)
Using (4.18) we get

6p1 Ii = (p1 + 2)( Ai Bi ) ( O O )Wi ,
6p1 Ji = (p1 + 2)(Ai Ci ) + O Wi .
From (4.33) we then obtain

6p1 Vi = (O + 1)Bi (O + 1)Ai (O + 1) (O + 1) Ci .
(4.35)
(4.36)
434
As a consequence of (4.24) the coefficients in (4.25) are not independent but we have
relations which determine Ai , Bi , Ci for each i,
1
U2 + (O p1 )W2 ,
2

p1 E
1
+
U2 + (O p1 )W2 ,
(p1 + 1)B2 = 3
2

1
+
E U3 + ( O O p1 )W3 ,
(p1 + 1) A3 = 3
2

p1 E
1
+
U3 (O + p1 )W3 ,
(p1 + 1) C3 = 3
2

(p1 + 1) B4 = 3
+
E U4 ( O O + p1 )W4 ,
(p1 + 1)A2 = 3
(p1 + 1) C4 = 3
1
U4 (O + p1 )W4 .
(4.37)
Combining this with (4.31) and also the result in (4.28) for p1 F leads to
1
+ p1 E F = U4 + (
6

+
E F = U2 +
1
1)W4 W2 ,
3
1
1
( 1)W2 W4 .
6
3
(4.38)
i as in (4.29) then it is easy

If W i,rsu given by (4.20) is defined in terms of A i , Bi , Ci and W
i = (p1 + 2)Wi and, as a consequence of (4.35) and (4.37),
to see that W
2(p1 + 1)A i

( O O + p1 + 2) +
+
E + 1 (O p1 2) Wi ,
=
Ci A i = O Wi ,
Bi A i = ( O O )Wi .
(4.39)
Hence (4.21) reduces to just

1
1
W2 =
W4 .
x
1x
(4.40)
In terms of the variables , ,

defined in (1.5), (4.38) becomes
F + EF p1 F

1
= (1 )U2 U4 + ( )
(1 )W2 + W4 ,
6
(1 )
(4.41)
435
together with the conjugate equation obtained for .

If this is used together with (4.28)
for x F we may eliminate U2 to obtain

E
+ p1
F
x
x
1
1

1
x
1
1
+
U4 ( )
W2 +
W4
=
1x 1
6
1

1 x
1
U4 ( )W
4 ,
=
(4.42)
(1 )(1 x)
6
where we have used (4.40). Writing F(u, v; , ) = F(x,

x;
, )
evidently

+ E + (p1 E)
F(x,
x;
, )
=1/x = 0,
x
x
1
which is solved by writing

1
).
, = f (x,
uE (1 v)p1 E F x, x;
x
(4.43)
(4.44)
Together with the conjugate equation in which (4.44) is the basic solution of the
superconformal Ward identities in this context.
5. Solution of identities, N = 4
We here extend the results of Section 3 to the N = 4 case. As previously it is more convenient for consideration of the operator product expansion to change from F (u, v; t) to
G(u, v; t), defined in a similar fashion to (3.1) with 2ni pi . Writing G(u, v; t) in a similar fashion to (4.25) then the corresponding function G is given in terms of F(u, v; , )
by
G(u, v; , ) = u 2 (p1 +p2 ) v p1 E F(u, v; , ).
1
(5.1)
For the applications in the section it is convenient to write

v; y, y)
v; y,
G(u, v; , ) = G(u,
= G(u,
y),
(5.2)
where G depends on the variables

y = 2 1,
y = 2 1.
(5.3)
The solution (4.44) then gives, with z, z defined in (3.4),

1
v; z, y)
G(u,
= u 2 (p1 +p2 )E f (z, y),
1
v; z , y)
G(u,
= u 2 (p1 +p2 )E f (z, y).
(5.4)
For consistency, since f (z, z) = f (z, z ), we must have

f (z, z) = k.
(5.5)
436
In general the conformal partial wave expansion and the decomposition into contributions for differing SU(4)R representations further into conformal partial waves is realised
by writing for p1 E.

(p E,p2 E)
v; y, y)
G(u,
=
anm (u, v)Pnm1
(y, y)
0mnE
(p E,p2 E)
anm,, Pnm1
(y, y)
0mnE ,
()
G (u, v; p2 p1 , p4 p3 ),
()
(5.6)
(a,b)
where G are described in Section 3 and Pnm (y, y)

are symmetric polynomials of degree n (i.e., for an expansion in terms of the form (y y)
s (y t + y t ), s + t n) which are
discussed in Appendix B and which are given in terms of Jacobi polynomials
(a,b)
(a,b)
Pnm
(y, y)
=
(a,b)
Pn+1 (y)Pm
(a,b)
(y)
Pm
(a,b)
(y, y).
= Pm1n+1
(a,b)
(y)Pn+1 (y)
y y
(5.7)
In (5.6) anm,, then corresponds to the presence of an operator in the operator product
expansion for (p1 ) and (p2 ) belonging to the SU(4)R representation with Dynkin labels
[nm, p1 +p2 2E +2m, nm] and with scale dimension , spin . The expansion (5.6)
(a,b)
Ep1 .
also extends to p1 < E save that then m E p1 and Pnm (y, y)
is a
We consider initially in detail the case with pi = p = E for all i and G(u, v; y, y)
symmetric polynomial in y, y with degree p. Since it must also be symmetric in z, z (5.4)
implies
v; y, y)
G(u,
(y z)(y z )(f (z, y)

+ f (z, y)) (y z )(y z)(f (z, y) + f (z, y))
= k +
(z z )(y y)
+ (y z)(y z )(y z)(y z )K(u, v; , ),

(5.8)
with K(u, v; , ) = K(u,

v; y, y)
defining an undetermined symmetric polynomial in y, y
of degree p 2. This term corresponds to the result (1.6) described in the introduction. To
take account of the constraint (5.5) we write
f (z, y) = k + (y z)f(z, y),
(5.9)
with f(z, y) a free function, polynomial in y of degree p 1.

v; y, y)
The decomposition of G(u,
into the contributions for different possible SU(4)R
representations is given by (5.6) where anm are for this case the coefficients corresponding to the representation with Dynkin labels [n m, 2m, n m]. For this case in (5.7)
(0,0)
Pn (y) = Pn (y), conventional Legendre polynomials.
We first consider the contribution resulting from the constant k in (5.8) and (5.9). It is
easy to see that this gives only
k
= k.
a00
(5.10)
437
To analyse the contributions arising from the function f(z, y) this may be expanded as
f(z, y) =
p1

fn (z)Pn (y).
(5.11)
n=0
Using this in (5.9) and (5.8) then fn gives rise to the following contributions to anm just
for m = 0, 1,
(n + 1)(n + 2)
(n 1)n
fn
an3
=
Fnm (z, z ),
Fnm (z, z ),
m
(2n + 1)(2n + 3)
(2n 1)(2n + 1)
n+1
n
f
fn
an nm =
an2
(z + z )Fnm (z, z ),
m = 2n + 1 (z + z )Fnm (z, z ),
2n + 1

1
1
fn
+
Fnm (z, z ),
=
z
z
+
an1
(5.12)
m
2 2(2n 1)(2n + 3)
f
n
an+1
m=
where
fn (z) fn (z)
zfn (z) z fn (z)
,
Fn0 (z, x)
.
=
(5.13)
z z
z z
For low n the results need to be modified but these can be obtained from (5.12) by taking
f
f
f
into account the symmetry relation in (5.7). For n = 0, a110 , a100 are as in (5.12) but for a000
we need to take
Fn1 (z, z ) =
0
a000 a11
a000 =
(z2 1/3)f0 (z) (z2 1/3)f0 (z)

,
z z
(5.14)
while for n = 1, a211 , a201 , a111 , a101 are given by (5.12) but
z (z2 13 )f1 (z) z(z2 13 )f1 (z)
4
(5.15)
+ F11 .
z z
15
It remains to consider the contribution of the two variable function K in (5.8) which is
(0,0)
Pnm (y, y),
as
expanded, for Pnm (y, y)

Anm (u, v)Pnm (y, y),
K(u,
v; y, y)
=
(5.16)
f
a001 =
0mnp2
with 12 (p 1)p terms. In this case the Legendre recurrence relations give
(m 1)mn(n + 1)
Anm ,
(2m 1)(2m + 1)(2n + 1)(2n + 3)
(m + 1)(m + 2)n(n + 1)
An m
an2
m+2 = (2m + 1)(2m + 3)(2n + 1)(2n + 3) Anm ,
(m 1)m(n + 2)(n + 3)
An m
an+2
m2 = (2m 1)(2m + 1)(2n + 3)(2n + 5) Anm ,
(m + 1)(m + 2)(n + 2)(n + 3)
An m
an+2
m+2 = (2m + 1)(2m + 3)(2n + 3)(2n + 5) Anm ,
1v
2mn(n + 1)
An m
an2
m1 = (2m + 1)(2n + 1)(2n + 3) u Anm ,
An m
an2
m2 =
438
1v
2(m + 1)n(n + 1)
Anm ,
(2m + 1)(2n + 1)(2n + 3) u
1v
2(m 1)m(n + 1)
An m
an1
m2 = (2m 1)(2m + 1)(2n + 3) u Anm ,
2(m + 1)(m + 2)(n + 1) 1 v
An m
an1
m+2 = (2m + 1)(2m + 3)(2n + 3) u Anm ,
1v
2m(n + 2)(n + 3)
An m
an+2
m1 = (2m + 1)(2n + 3)(2n + 5) u Anm ,
2(m + 1)(n + 2)(n + 3) 1 v
An m
an+2
m+1 = (2m + 1)(2n + 3)(2n + 5) u Anm ,
1v
2(m 1)m(n + 2)
An m
an+1
m2 = (2m 1)(2m + 1)(2n + 3) u Anm ,
2(m + 1)(m + 2)(n + 2) 1 v
An m
an+1
m+2 = (2m + 1)(2m + 3)(2n + 3) u Anm ,
An m
an2
m+1 =
An m
an1
m1 =
(1 v)2
4m(n + 1)
Anm ,
(2m + 1)(2n + 3) u2
An m
an1
m+1 =
4(m + 1)(n + 1) (1 v)2

Anm ,
(2m + 1)(2n + 3) u2
An m
an+1
m1 =
(1 v)2
4m(n + 2)
Anm ,
(2m + 1)(2n + 3) u2
4(m + 1)(n + 2) (1 v)2

Anm ,
(2m + 1)(2n + 3) u2
2n(n + 1)
An m
an2
m = (2n + 1)(2n + 3) Bm Anm ,
2(n + 2)(n + 3)
An m
an+2
m = (2n + 3)(2n + 5) Bm Anm ,
2(m 1)m
nm
Bn+1 Anm ,
anAm2
=
(2m 1)(2m + 1)
2(m + 1)(m + 2)
nm
Bn+1 Anm ,
anAm+2
=
(2m + 1)(2m + 3)
1v
4(n + 1)
An m
an1
m = 2n + 3 Bm u Anm ,
1v
4(n + 2)
An m
an+1
m = 2n + 3 Bm u Anm ,
1v
4m
nm
anAm1
Bn+1
Anm ,
=
2m + 1
u
1v
4(m + 1)
nm
=
anAm+1
Bn+1
Anm ,
2m + 1
u
n m = 4B B
anAm
m n+1 Anm ,
An m
an+1
m+1 =
(5.17)
439
where
Bm =
m2 + m 1
1+v
.
u
(2m 1)(2m + 3)
(5.18)
For m = n, n 1, n 2, n 3, and also if n = 0, 1, 2, (5.7) may be used to combine terms

to ensure that we only have anAnm
m for 0 m n . For m = n = 0 this prescription gives

1+v
4
41v
4
a22 = A00 ,
A00 ,
3
1 A00 ,
a21 =
a20 =
15
5 u
15
u

2
(1 v)
4
1+v
a11 =
10
+ 1 A00 ,
5
2
15
u
u

1v
4 1+v
1
A00 ,
a10 = 2
3
u
u

(1 + v)2
4
(1 v)2
1+v
15
+
1
A00 ,
8
a00 =
(5.19)
15
u
u2
u2
which is equivalent to the results in [19]. Similarly for n = 1, m = 0, 1 the resulting anAnm
m
correspond to those in [9].
The solution of the superconformal identities given by (5.10), (5.12) and (5.17) may
now be naturally interpreted in terms of the operator product expansion. If in (5.17) we
consider a single conformal partial wave for Anm by letting
()
Anm G+4 ,
(5.20)
then, if A
[q,p,q], denotes a long superconformal multiplet whose lowest state has spin
, scale dimension and which belongs to a SU(4)R representation with Dynkin labels
[q, p, q], we obtain

anAnm
(5.21)
A
nm, A[nm,2m,nm], .
m an m Anm, ,
The non-zero results obtained from (5.17) with (5.20) may be conveniently expressed in
the form

nm
an+i m+j A
(5.22)
nm, = Nn+1,i Nm,j A|i||j | , i, j = 2, 1, 0,
for
(m + 1)(m + 2)
m+1
,
Nm,1 =
,
(2m + 1)(2m + 3)
2m + 1
m
(m 1)m
,
Nm,2 =
Nm,1 =
2m + 1
(2m 1)(2m + 1)
Nm,2 =
and using (3.15) we have

( )
an+i m+j A
b( ; ) G ,
nm, =
( ; )
|i| = |j | = 2,

; = ( + 4; ),
Nm,0 = 1,
(5.23)
440

|j | = 1,
|i| = 1,
|j | = 2,
; = ( + 5, + 3; 1),

|i| = |j | = 1,
; = ( + 6, + 4, + 2; 2, ),
|i| = 2,
|i| = 1,
j = 0,
i = 0,
|j | = 1,

; = ( + 7, + 1; 1), ( + 5, + 3; 3, 1),
i = j = 0,

; = ( + 8, ; ), ( + 6, + 2; 2, ), ( + 4; 4, 2, ).
(5.24)
an m (A
nm, )
corresponds to the contribution in the operator product exIn consequence

pansion applied to the correlation function for all expected operators belonging to A
nm, . In
(5.24) b( ; ) > 0 if > + 1. If m n m + 3 the results are modified since we then obtain from (5.17) contributions with m > n . In this case, for n m and am 1 n +1 (A
nm, )
non-zero, we should take

an m A
(5.25)
nm, am 1 n +1 Anm, an m Anm, .
Furthermore any contribution with m = n + 1 should be dropped. Using this result and
(5.25) we may then easily show that

an m A
(5.26)
nn+1, = 0,
and for later reference we also note the symmetry relation

an m A
nm, = am 1 n +1 Am1 n+1, .
(5.27)
The unitarity condition for a long multiplet A

nm, requires
2n + + 2,
(5.28)
and so, as in (3.20), using (5.20) we must have for u 0,

Anm (u, v) un+3+ ,
0.
(5.29)
We now consider the operator product expansion interpretation of the remaining terms
in the solution of the superconformal identities given by (5.8) and (5.9). The constant k,
whose contribution is just given by (5.10), clearly corresponds to the identity operator,
anm (I) = n0 m0 .
(5.30)
To analyse the contribution of the single variable functions fn in (5.11) we use the result
(3.21) for the conformal partial wave for twist two operators as well as

()
G (u, v)
(5.31)
fn+1 (z) 12 g+2 (x),
n = 0, 1, 2, . . . ,
(5.32)
1 z g (x) zg (x)

,
2
z z
with g as in (3.22), for twist zero. Taking
1 =2 =3 =4
in (5.13) and (5.11) then leads to results corresponding to only twist zero and twist
two operators. These operators can be interpreted as belonging to a multiplet Dn0, ,
441
where in general we denote by Dn m, D[nm,2m,nm], the semi-short supermultiplet

in which the lowest dimension operator has = 2m + , or twist 2m, and belongs to the
[n m, 2m, n m] SU(4)R representation. These non-unitary super multiplets are discussed in Appendix D. For Dnm, the conformal partial waves may be expressed in general
in the form
nm
,
an+i m+j (Dnm, ) = Nn+1,i Nm,j D|i|j
nm
D|i|2
= 0.
(5.33)
Corresponding to (5.32) we then have

1 (+1)
n0
D21
= G+3 ,
4
()
1
(+2)
n0
D20
= G+2 + a+2 G+4 ,
4
1 ()
(+2)
(+2)
n0
D11 = G+2 + G+2 + a+2 G+4 ,
4
1 (1)
(+1)
(+3)
(+1)
(+3)
n0
D10 = G+1 + a+2 G+3 + G+1 + b G+3 + a+2 a+3 G+5 ,
4
1
(+1)
(+3)
(+1)
n0
D01 = G+1 + a+2 G+3 + bn G+3
,
4
1 ()
()
(+2)
(+4)
(+2)
n0
D00
= G + b G+2 + a+2 a+3 G+4 + bn G+2 + a+2 G+4 ,
4
whereas a is as in (3.16) and
(5.34)
22 + 6 + 3
.
(5.35)
(2 + 1)(2 + 5)
A list of relevant representations for differing dimensions contained in Dn0, D[n,0,n],
is listed in Appendix D, the twist zero and twist two representations correspond with those
necessary for (5.34). For f0 these results are modified. From (5.14) only twist two contributions are required since, taking now f0 (z) 2g+3 (x),
b = a+2 + a+1 =
2( + 2)( + 3) (+2)

(+4)
G
+ a+3 a+4 G+6 ,
3(2 + 3)(2 + 7) +4
2 (+1)
(+3)
a10 (C00, ) = G+3 + a+3 G+5 ,
3
2 (+2)
a11 (C00, ) = G+4 .
(5.36)
3
Here we denote by Cnm, C[nm,2m,nm], the semi-short supermultiplet in which the
lowest dimension operator has = 2n + + 2 and belongs to the [n m, 2m, n m]
SU(4)R representation.
The multiplets D[q,p,q], fail to satisfy the unitarity condition (5.28) on and so their
contributions as in (5.34) must be cancelled in a unitary theory. This may be achieved by a
corresponding long multiplet contribution. When = 2m + or = 2n + + 2 the long
multiplet A
nm, can be decomposed into semi-short multiplets resulting in
()
a00 (C00, ) = G+2 +
2m+
4(m + 1)
an m (Dnm+1,1 ),
an m Anm,
= 16an m (Dnm, ) +
2m + 1
(5.37)
442
and, at the unitarity threshold (5.28),

2n++2
4(n + 2)
an m (Cn+1m,1 ).
= 16an m (Cnm, ) +
an m Anm,
2n + 3
(5.38)
When n = m we have the special case

(n + 1)(n + 2)
an m (Cn+1n+1,2 ).
an m A2n+
nn, = 16an m (Dnn, ) +
(2n + 1)(2n + 3)
(5.39)
The results (5.37), (5.38) and (5.39) reflect a decomposition of long multiplets at particular values of as described in Appendix D. From (5.37) we may obtain an m (Dnm, )
iteratively starting from (5.34). With the notation in (5.33) the results are
1 ()
G
,
16 2m++4
1 (+1)
D2nm
1 = G2m++3 ,
4
1 (1)
(+1)
(+1)
G
D2nm
+ 4G2m++3 + am++2 G2m++5 ,
1 =
16 2m++3
1 ()
()
(+2)
D2nm
4G2m++2 + am G2m++4 + 4am++2 G2m++4 ,
0 =
16

1
1
(1)
(+1)
(1)
(+1)
G
a
D1nm
=
+
4G
+
G
+
a
G
m+1
m++2
2
2m++3
2m++5
2m++5 ,
16 2m++3
4

1 ()
1
(+2)
()
(+2)
G
a
D1nm
=
+
4G
+
G
+
a
G
m 2m++4
m++2 2m++4 ,
1
2m++2
4 2m++2
4

1
1
(2)
(2)
nm
G2m++2 + am+1 G2m++4
D1 1 =
16
4
1
()
()
()
+ 8G2m++2
+ (bm+ + am )G2m++4
+ am+1 am++2 G2m++6
4

(+2)
(+2)
(+2)
+ 16G2m++2 + 8am++2 G2m++4 + am++2 am++3 G2m++6 ,
D2nm
2 =
D1nm
0 =

1
1
(1)
(1)
(1)
4G2m++1 + 2am G2m++3 + am am+1 G2m++5
16
4
(+1)
(+1)
(+1)
+ 16G2m++1 + 4(bm+ + am )G2m++3 + 2am am++2 G2m++5

(+3)
(+3)
+ 16am++2 G2m++3 + 4am++2 am++3 G2m++5 ,
D0nm
2

1
1
()
(2)
()
4G2m++2 + am+1 G2m++4 + bn G2m++4
=
16
4

1
(+2)
()
+ 4am++2 G2m++4 + am+1 am++2 G2m++6 ,
4
443

1
1
(+1)
(1)
(+1)
4G2m++1
+ am G2m++3
+ bn G2m++3
4
4

1
(+3)
(+1)
+ 4am++2 G2m++3 + am am++2 G2m++5 ,
4

1 1
(3)
(1)
(1)
am+1 G2m++3 + 4G2m++1 + (am + bn )G2m++3
D0nm
1 =
16 4
1
(1)
(+1)
(+1)
+ am+1 bm+ G2m++5 + 16G2m++1 + 4(bm+ + bn )G2m++3
4
1
(+1)
(+1)
+ am++2 (am + bn )G2m++5
+ am+1 am++2 am++3 G2m++7
4

D0nm
1 =
(+3)
(+3)
+ 16am++2 G2m++3 + 4am++2 am++3 G2m++5 ,

1
1
(2)
(2)
()
()
D0nm
am G2m++2
=
+ am am+1 G2m++4
+ 16G2m+
+ 4(am + bn )G2m++2
0
16
4
1
()
()
+ am (bm+ + bn )G2m++4 + am am+1 am++2 G2m++6
4
(+2)
(+2)
+ 16bm+ G2m++2 + 4am++2 (am + bn )G2m++4

(+2)
(+4)
+ am am++2 am++3 G2m++6 + 16am++2 am++3 G2m++4 .
(5.40)
The corresponding results for the semi-short multiplet Cnm, may be obtained from those
for Dnm, given above by taking
an m (Cnm, ) = am 1 n +1 (Dm1 n+1, ).
(5.41)
Using (5.27) then (5.38) easily follows from (5.37). We may also verify that (5.39) is
satisfied. Combining (5.37) for m = n + 1 with (5.26) we may then obtain

2n+
an m A2n+
n n, am 1n +1 An n,

= 16 an m (Dn n, ) am 1 n +1 (Dn n, )

(n + 1)(n + 2)
am 1 n +1 (Cn+1 n+1,2 ) + an m (Cn+1 n+1,2 ) ,
+
(2n + 1)(2n + 3)
(5.42)
which for n m , and noting the requirement (5.25), gives exactly (5.39).
In general the results from (5.41) can be expressed as
an+i m+j (Cnm, ) = Nn+1,i Nm,j Cijnm ,
C2nm
j = 0.
(5.43)
For general n, m the necessary operators are just those given in Table 4 of [25]. For m = n
the relation (5.41) combined with (5.40) in this case and applying the corresponding results
to (5.25) gives
(+2)
,
C1nn1 = G2n++4
444
1
(+1)
(+1)
(+3)
C1nn0 = G2n++3 + an G2n++5 + an++3 G2n++5 ,
4
1
()
()
(+2)
C0nn0 = G2n++2 + an G2n++4 + (bn++1 an+1 )G2n++4
4
1
1
()
(+2)
(+4)
+ an an+1 G2n++6 + an an++3 G2n++6 + an++3 an++4 G2n++6 ,
16
4
1 ()
1
(+2)
(+2)
C1nn1 = G2n++4 + G2n++4 + an++3 G2n++6 ,
4
4
1 (1)
1
1
(+1)
(1)
(+1)
C0nn1 = G2n++3 + G2n++3 + an+1 G2n++5 + bn++1 G2n++5
4
16
4
1
1
(+3)
(+1)
(+3)
+ an++3 G2n++5 + an+1 an++3 G2n++7 + an++3 an++4 G2n++7 ,
16
4
1 (2)
1 ()
1
(+2)
()
nn
G
C1
+ G
+ G2n++4 + (bn++1 an+2 )G2n++6
1 =
16 2n++4 4 2n++4
16
1
1
(+2)
(+2)
+ an++3 G2n++6 + an++3 an++4 G2n++8 ,
4
16
1 (+1)
,
C1nn2 = G2n++5
4
1 ()
1
1
()
(+2)
C0nn2 = G2n++4
+ an+1 G2n++6
+ an++3 G2n++6
,
4
16
4
1 (1)
1 (+1)
1
(+1)
nn
G
+ G
+ an++3 G2n++7 ,
C1
2 =
16 2n++5 4 2n++5 16
1 ()
nn
G
.
C2
(5.44)
2 =
16 2n++6
The necessary operators correspond exactly to those listed in [25] (see Table 3) as present
in the semi-short supermultiplet for this case. For n = 0 (5.44) reproduces (3.21). We may
also note that, since for m 1, 1/4 < am 1/3 and bn > 1/2, all coefficients in (5.44) are
positive as required by unitarity.
As in the N = 2 case the semi-short results also include the contributions for short
BPS multiplets when extended to negative . Formally as shown in [25] C[q,p,q],1
B[q+1,p,q+1] where B[q,p,q] denotes the BPS supermultiplet whose lowest state has spin
zero, = 2q + p, and belongs to the SU(4)R [q, p, q] representation. For q > 0 the low supercharges whereas when q = 0 we
est state is annihilated by 1/4 of the Q and also Q
1
have a 2 -BPS multiplet with 1/2 the Q and Q supercharges annihilating the lowest state.
As earlier we identify, for n m, Bnm B[nm,2m,nm] and we then have
an m (Cn m,1 ) =
n+1
an m (Bn+1 m ),
2n + 1
(5.45)
where
an+i m+j (Bnm ) = Nn+1,i Nm,j Binm
j ,
nm
B2nm
j = B1 j = 0.
(5.46)
For general n, m we have

nm
Binm
j = Bi |j | ,
(5.47)
445
and
1 (0)
B0nm
2 = G2n+2 ,
4
1 (0)
nm
B2 2 = G2n+4 ,
16
1 (1)
nm
B1
2 = G2n+3 ,
4
1
(1)
(1)
nm
B0 1 = G2n+1 + an+1 G2n+3 ,
4
1 (1)
1
(1)
nn
B2
an+2 G2n+5
,
1 = G2n+3 +
4
16
1 (0)
1
1
(2)
(0)
(2)
nm
an G2n+4 + an+2 G2n+4 ,
B1
1 = G2n+2 + G2n+2 +
4
16
4
1
1
(0)
(0)
(2)
(0)
B0nn0 = G2n + (bm1 an )G2n+2 + an+1 G2n+2 + an an+1 G2n+4 ,
4
16
1 (0)
1
1
1
(0)
(2)
(0)
nm
(bm1 an+1 )G2n+4 + an+2 G2n+4 + an+1 an+2 G2n+6 ,
B2
0 = G2n+2 +
4
16
4
64
1
1
(1)
(1)
(3)
(1)
nm
B1
(5.48)
an an+2 G2n+5 .
0 = G2n+1 + bm1 G2n+3 + an+2 G2n+3 +
4
16
Again all coefficients are positive and the necessary operators are exactly as expected for
this supermultiplet (see Table 2 in [25]). For n = m + 1 the multiplet is truncated with, in
(5.46), the following non-zero,
(1)
m
B0m+1
= G2m+3 ,
1
1
(0)
(0)
(2)
m
= G2m+2 + am G2m+4 + am+2 G2m+4 ,
B0m+1
0
4
1
(1)
(1)
(3)
m+1 m
B1
0 = G2m+3 + 4 am G2m+5 + am+3 G2m+5 ,
1
(1)
(1)
m
B0m+1
1 = G2m+3 + 4 am+2 G2m+5 ,
1 (0)
1
1
(2)
(0)
(2)
m+1 m
B1
1 = 4 G2m+4 + G2m+4 + 16 am+1 G2m+6 + 4 am+3 G2m+4 ,
1 (1)
1
(1)
m+1 m
B2
1 = 4 G2m+5 + 16 am+3 G2m+7 ,
1 (0)
m
B0m+1
2 = 4 G2m+4 ,
1 (1)
m+1 m
B1
2 = 4 G2m+5 ,
1 (0)
m+1 m
B2
2 = 16 G2m+6 .
(5.49)
The necessary operators correlate again with those expected for this 14 -BPS multiplet (see
Table 5 in [25]).
446
If we consider the semi-short multiplet for = 2 we get

an m (Cn m,2 ) = 4an m (Bn m ),
(5.50)
which allows results for an m (Bn m ) to be derived for m = n in addition to m < n as given
by (5.45). However, in this case there is a further decomposition into contributions corresponding to 12 -BPS multiplets. Such 12 -BPS contributions are obtained in (5.46) by letting
Bnm Bnn and Bijnm B ijnn where
(0)
(1)
(2)
nn
B 0nn0 = G2n ,
B 0nn1 = G2n+1 ,
B 1
1 = G2n+2 ,
1 (0)
1 (1)
1 (0)
nn
nn
G
B 0nn2 = G2n+2 ,
(5.51)
B 1
B 2
2 = G2n+3 ,
2 =
4
4
16 2n+4
(the relevant operators here correspond to Table 1 in [25]). With the result given in (5.51)
we can then write in (5.50)
an m (Bn n ) = an m (Bn n )
(n + 1)(n + 2)
an m (Bn+1 n+1 ).
4(2n + 1)(2n + 3)
(5.52)
From (3.31) and (5.30) it is also easy to see that

anm (B0 0 ) = anm (I).
(5.53)
Any 12 -BPS contribution an m (Bn n ) may then be isolated by considering appropriate linear
combinations of an m (Cn n,2 ) together with an m (I).
We also consider the extremal and next-to-extremal cases. When E = 0 G is independent of y, y and so must also be the function f in (5.4). From (5.5) and (3.35) we then get
the solution
1
G(u, v) = u 2 p+ k,
(5.54)
where we define
p = p 2 p 1 .
(5.55)
Noting that
(p ,p2 )
P00 1
(y, y)
= 12 (p+ + 2),
Gp(0)
(u, v; p , p+ ) = u 2 p+ ,
+
(5.56)
it is clear that the only operator which is necessary in the operator product expansion has
= p+ and is spinless belonging to the [0, p+ , 0] representation. This is of course may be
identified with the contribution of just the 12 -BPS operator belonging to the short B[0,p+ ,0]
supermultiplet so that for the extremal case, up to a constant factor,
.
anm (B[0,p+ ,0] ) = n0 m0 Gp(0)
+
(5.57)
The correlation function in this case has the very simple form
(p )

1 (x1 , t1 ) (p2 ) (x2 , t2 ) (p3 ) (x3 , t3 ) (p4 ) (x4 , t4 ) p =p
4
(t1 t4 )p1 (t2 t4 )p2 (t1 t3 )p3

=
k.
r14 p1 r24 p2 r34 p3
1 +p2 +p3
(5.58)
447
For the next-to-extremal case, E = 1, we have a similar solution to that given by (3.7)
and (5.9), but with no arbitrary K term and f a single variable function of z,
v; y, y)
G(u,
=u
=
1
2 p+ 1

k

1
(y z)(y z)f (z) (y z )(y z )f (z)

z z
(p 1,p2 1)
anm (u, v)Pnm1
(y, y),
(5.59)
0mn1
where we have expanded in terms of the different possible SU(4)R representations. From
this we obtain
1
p+ (p+ + 1)(p+ + 2)a11 = a 11 = F0 ,
16
1
p
(p+ + 1)(p+ + 2)a10 = a 10 = F1 +
F0 ,
8
p+
1
1
2p
p 2 (p+ + 2)
p+ a00 = a 00 = ku 2 p+ 1 + F2 +
F1 +
F0 ,
2
p+ + 2
(p+ + 1)(p+ + 2)
(5.60)
for
Fn (z, z ) = (1)n u 2 p+ 1
1
zn f(z) z n f(z)
.
z z
(5.61)
Keeping only the term in (5.60) involving k we may easily from (5.56) see that this
represents the contribution of just the 12 -BPS chiral primary operator belonging to the
B[0,p+ 2,0] supermultiplet so that in the next-to-extremal case we have
(0)
a nm (B[0,p+ 2,0] ) = n0 m0 Gp+ 2 .
(5.62)
If in (5.60) and (5.61) we let f(z) 2g+3 (x; p1 , p2 ) and use the definitions in (3.40) we
obtain the contributions for the semi-short supermultiplet C[0,p+ 2,0], ,
(+2)
a 11 (C[0,p+ 2,0], ) = Gp+ ++2 ,

a 10 (C[0,p+ 2,0], )
(+1)
(+3)
= Gp+ ++1 + b+2 Gp+ ++3 +
4( + 2)p (p+ + + 1)

(+2)
G
,
p+ (p+ + 2 + 2)(p+ + 2 + 4) p+ ++2
448

()
(+4)
(+2)
a 00 (C[0,p+ 2,0], ) = Gp+ + + b+2 b+3 Gp+ ++4 + c+2 Gp+ ++2
8( + 1)p (p+ + + 1)

(+1)
G
(p+ + 2)(p+ + 2)(p+ + 2 + 4) p+ ++1
8( + 2)p (p+ + + 2)b+2
(+3)
G
+
,
(p+ + 2)(p+ + 2 + 2)(p+ + 2 + 6) p+ ++3
(5.63)
for
4( + 1)(p1 + )(p2 + )(p+ + 1)
,
(p+ + 2 1)(p+ + 2)2 (p+ + 2 + 1)
2(p+ + 1)
c =
(p+ + 1)(p+ + 2 3)(p+ + 2 + 1)

p 2 (8( 1)(p+ + ) p+ (p+ 1))
.
p+ 1 +
(p+ + 2)(p+ + 2 2)(p+ + 2)
b =
(5.64)
The necessary operators required for (5.63) correspond exactly with those in this semishort supermultiplet (see Table 3 in [25]).
Just as previously we may extend these (5.63) to = 1, 2 to obtain results for short
multiplets. Thus
a nm (C[0,p+ 2,0],1 ) = a nm (B[1,p+ 2,1] ),
a nm (C[0,p+ 2,0],2 ) = a nm (B[0,p+ ,0] ) 4a nm (B[0,p+ 2,0] ),
(5.65)
where, together with (5.62),

a 11 (B[1,p+ 2,1] ) = Gp(1)
,
+ +1
4p
(1)
(2)
G
a 10 (B[1,p+ 2,1] ) = Gp(0)
+
+ b1 Gp+ +2 ,
+
p+ (p+ + 2) p+ +1

8p (p+ + 2)
(1)
(2)
(3)
Gp+ +2 + b2 Gp+ +3 , (5.66)
a 00 (B[1,p+ 2,1] ) = b1 Gp+ +1 +
p+ (p+ + 2)(p+ + 4)
and
,
a 11 (B[0,p+ ,0] ) = Gp(0)
+
(1)
a 10 (B[0,p+ ,0] ) = b0 Gp+ +1 ,
.
a 00 (B[0,p+ ,0] ) = b0 b1 Gp(2)
+ +2
(5.67)
The necessary operators here correspond to Tables 5 and 1 in [25].

The results obtained above show that the operator product expansion for 12 -BPS operators can be decomposed into short, semi-short and long supermultiplets. For p =
p2 p1 0,
B[0,p1 ,0] B[0,p2 ,0]

B[nm,p +2m,nm]
0mnp1
0 0mnp1 2
C[nm,p +2m,nm],
0 0mnp1 1
A
[nm,p +2m,nm], ,
(5.68)
449
in accordance with the results of Eden and Sokatchev [27]. In (5.68) we identify B[0,0,0]
I, corresponding to the unit operator in the operator product expansion. It immediately follows from (5.68) that long supermultiplets, with non-zero anomalous dimensions, cannot
contribute to extremal and next-to-extremal correlation functions.
6. Crossing symmetry
The operator product expansion provides the strongest constraints when combined with
crossing symmetry. For a correlation function for four identical chiral primary operators
the correlation function is invariant under permutations of all xi , ti for all i = 1, 2, 3, 4.
Permutations of the form (ij )(kl) act trivially so we may restrict to permutations leaving
x4 , t4 invariant so that crossing symmetry transformations correspond to the permutation
group S3 , which is of order 6. The action of each permutation on the essential conformal
invariants u, v or x, x or y, z and also on the R-symmetry invariants , or , or y, y is
given in Table 1, where the transformations of x are identical to those of x, and similarly
for z , ,
y.
For the N = 4 case with pi = p the crossing symmetry conditions on the correlation
function G(u, v; , ) are generated by considering just (12) and (13) which give

p

u 1
u
1
G(u, v; , ) = G
(6.1)
, ; , =
G v, u; ,
.
v v
v

The general construction of such invariant correlation functions follows by determining
polynomials in , which transform according to the irreducible representations of S3 .
We first consider symmetric polynomials satisfying

1
p
Sp (, ) = Sp (, ) = Sp
(6.2)
,
.

As described by Heslop and Howe [20], for any given p, S3 acts on the 12 (p + 1)(p + 2)
monomials r s , r + s p, giving chains of length 6 or 3 or 1 which may be added to
give minimal polynomial solutions of (6.2). If the chain contains a monomial ( )r , for
0 r [ 12 p], where [x] denotes the integer part of x, then this term is invariant under
the action of the permutation (12) and the chain is of length 3, except if p is divisible by
3 then ( )p/3 satisfies (6.2) by itself and so forms a chain of length 1. All other chains
are of length 6. With this counting the number of independent such minimal symmetric
Table 1
Symmetry transformations of variables under crossing
e
(12)
(13)
(23)
(123)
(132)
u
v
1
v
x
x1
1x
z+3
z1
v
u
1
u
x1
x
3+z
1z
1
v
u
v
1
u
v
u
1
x
3z
1+z
1
1x
z3
z+1
(12)
(13)
(23)
(123)
(132)
3y
1+y
y3
y+1
1
y+3
y1
1
1
y+3
y1
450
Table 2
Symmetric polynomials
p
Polynomial
(i, j )
+ +1
(0, 0)
2 + 2 + 1
+ +
3 + 3 + 1
2 + 2 + 2 + 2 + +
(0, 0), (1, 0)
4 + 4 + 1
3 + 3 + 3 + 3 + +
2 2 + 2 + 2
2 + 2 +
5 + 5 + 1
4 + 4 + 4 + 4 + +
3 2 + 2 3 + 3 + 3 + 2 + 2
3 + 3 +
2 2 + 2 + 2
6 + 6 + 1
5 + 5 + 5 + 5 + +
4 2 + 2 4 + 4 + 4 + 2 + 2
3 3 + 3 + 3
4 + 4 +
3 2 + 2 3 + 3 + 3 + 2 + 2
2 2
7 + 7 + 1
6 + 6 + 6 + 6 + +
5 2 + 2 5 + 5 + 5 + 2 + 2
4 3 + 3 4 + 4 + 4 + 3 + 3
5 + 5 +
4 2 + 2 4 + 4 + 4 + 2 + 2
3 3 + 3 + 3
3 2 + 2 3 + 2 2
polynomials is,

(n + 1)3n + 1,
Np =
(n + 1)(3n + q),
p = 6n,
p = 6n + q,
(0, 0), (1, 0)

(0, 1)
(0, 0), (1, 0), (2, 0)

(0,
1)
(0, 0), (1, 0), (2, 0)

(0, 1), (1, 1)
(0, 0), (1, 0), (2, 0), (3, 0)

(0, 1), (1, 1)
(0,
2)
(0, 0), (1, 0), (2, 0), (3, 0)

(0, 1), (1, 1), (2, 1)
(0, 2)
q = 1, 2, 3, 4, 5.
(6.3)
We list the first few non-trivial cases in Table 2, of course S0 (, ) = 1.

An alternative basis for Sp , valid for general p, may be obtained by constructing from
, two invariants I1 , I2 under S3 and then introducing for any p a factor to ensure that
(6.2) holds. With suitable restrictions the result becomes a polynomial expressible in the
form
Sp,(i,j ) (, ) = ( + + 1)p I1 (, )i I2 (, )j ,
+ +
I1 (, ) =
,
I2 (, ) =
,
2
( + + 1)
( + + 1)3
i = 0, 1, . . . ,
1
2p ,
j = 0, 1, . . . ,
1
3 (p 2i)
451
(6.4)
Lists of possible (i, j ) for p up to 7 are given in Table 2. This result may also be easily
expressed as symmetric polynomial in y, y by using

1
1 y 2 1 y 2 ,
16
1
2
= ( + + 1) 4( + + ) = (y y)
(6.5)
2,
4
where is defined in (2.42). Completeness of the basis provided by (6.4) is straightforwardly demonstrated by showing that it gives the same number of independent polynomials
Np as given in (6.3).
For the antisymmetric representation of S3 we require,
1
+ + 1 = (y y + 3),
2
a a,
(12)
a a,
(6.6)
(123)
while the two-dimensional mixed symmetry representation of S3 is defined on a basis (b, c)

where

12 23
b
1 0
b
b
b
.
,

(6.7)
3
1
c
0 1
c
c (123)
c (12)
2
2
It is easy to see that the tensor products formed by aa and bb + cc are symmetric while
(bc + cb , bb + cc ) is a basis for a mixed symmetry representation and bc cb is
antisymmetric.
For functions of , (6.6) is satisfied by
a(, ) =
( )( 1)( 1)
.
( + + 1)3
(6.8)
For p 3, a(, )Sp,(i,j ) (, ) is a polynomial if we allow i = 0, 1, . . . , [ 12 (p 3)] and

j = 0, 1, . . . , [ 13 (p 2i 3)] giving Np3 antisymmetric polynomials. For the mixed symmetry transformations in (6.7) there essentially two independent possibilities
b1 (, ) =

,
+ +1
b2 (, ) =

,
( + + 1)2
+ 2
c1 (, ) =
3( + + 1)
(6.9)
and
+ 2
c2 (, ) =
.
3( + + 1)2
(6.10)
By considering (br (, ), cr (, ))Sp,(i,j ) (, ) for p r, r = 1, 2, for appropriate i, j we

obtain Npr polynomial mixed symmetry representations of S3 . Together with the symmetric polynomials Sp,(i,j ) and aSp,(i,j ) these provide a complete basis for two variable
polynomials in , of order p since Np + 2(Np1 + Np2 ) + Np3 = 12 (p + 1)(p + 2).
We may also note that these polynomials form a closed set under multiplication since

4
b1 2 + c1 2 = 4I1 ,
3 2b1 c1 , b1 2 c1 2 = 2(b1 3b2 , c1 3c2 ),
3
452
2
b1 b2 + c1 c2 = I1 6I2 ,
3
3(b1 c2 + c1 b2 , b1 b2 c1 c2 ) = 2(I1 b1 b2 , I1 c1 c2 ),
4
b2 2 + c2 2 = I1 2 4I2 ,
3

3 2b2 c2 , b2 2 c2 2 = 2(3I2 b1 I1 b2 , 3I2 c1 I1 c2 ),
3(b1 c2 c1 b2 ) = 2a,
a 2 = I1 2 4I1 3 + 18I1 I2 4I2 27I2 2 ,
(6.11)
where I1 , I2 are the invariants defined in (6.4).

For N = 2 there are further restrictions as a consequence of (2.42). Taking p 2n we
construct, instead of (6.2) since , are expressible in terms of just by (2.43), the single
variable polynomials fn of degree 2n, satisfying under the action of S3

1
= 2n fn
fn () = fn (1 ) = ( 1)2n fn
1

1
1
= ( 1)2n fn
.
= ( 1)2n fn
(6.12)
1
As shown by Heslop and Howe, [20], the sum of terms produced by the action of S3 as
given by (6.12) starting from r generates a linearly independent set of polynomials for
r = 0, 1, . . . , [ 13 n], giving [ 13 n] + 1 solutions for fn . Alternatively an equivalent basis is
provided by

n

Sn,j () = 2 + 1 s()j , j = 0, 1, . . . , 13 n ,
(6.13)
where s() is the S3 invariant
s() =
2 (1 )2
(1 y 2 )2
=
4
.
( 2 + 1)3
(y 2 + 3)3
(6.14)
The solutions given by (6.13) correspond to (6.4) for i = 0 since = 0 in this case. A general polynomial solution of (6.12) is then given by

n

fn () = 2 + 1 P s() ,
(6.15)
with P (s) a polynomial of degree [ 13 n].
We may also consider other representations of S3 . For the antisymmetric representation,
as in (6.8), we may define
a() = (2 1)
( 2)( 1)( + 1)
y(y 2 1)(y 2 9)
=
4
,
( 2 + 1)3
(y 2 + 3)3
(6.16)
so that a()Sn,j () is then a polynomial for n 3 and j = 0, 1, . . . , [ 13 n] 1. For the

mixed symmetry representation there are two essential solutions which can be written in
the form
b1 () =
y
2 1
=4 2
,
+1
y +3
2 y2 3
1 2 2 2 1
=
,
c1 () =
3 2 + 1
3 y2 + 3
453
(6.17)
and
( 1)
y(y 2 1)
=
4
,
( 2 + 1)2
(y 2 + 3)2
y2 1
( 1)
=
4
3 2
.
c2 () = 3 2
( + 1)2
(y + 3)2
b2 () = (2 1)
(6.18)
It is easy to see that (br (), cr ())Sn,j () are polynomials for j = 0, 1, . . . , [ 13 (n r)]
if n r for r = 1, 2. The basis provided by Sn,j (), (br (), cr ())Sn,j (), r = 1, 2
and a()Sn,j () is then complete in that it gives 2n + 1 linearly independent polynomials, allowing for the expansion of any arbitrary polynomial of degree 2n, 2([ 13 n] +
[ 13 (n 1)] + [ 13 (n 2)]) + 5 = 2n + 1.
For N = 4 the superconformal Ward identities require
G(u, v; , )|=1/
x = f (x, ),
(6.19)
so that (6.1) gives

x
f (x, ) = f
,1
x 1

1 1
x( 1) p
= (x)p f
,
.
f 1 x,
=
1x
1
x
(6.20)
To obtain an extension to a fully crossing symmetric correlation function we may consider

for any Sp satisfying (6.2)

u
G(u, v; , ) = Sp u, ,
(6.21)
v
which obeys (6.1) as a consequence of (6.2). From (6.19) we obtain

x(1 )
,
f (x, ) = Sp x,
x 1
which automatically satisfies (6.20).
For the N = 2 case instead of (6.1) we have
2 n

u
1
u 1
, ; , =
,
G v, u; ,
G(u, v; , ) = G
v v

v2
where, with , constrained as in (2.43), the superconformal Ward identities are

2n

x
x

=
G(u, v; , ) =1/x = f (x) = f
f (1 x)
x 1
x 1

1
,
= x 2n f
x
(6.22)
(6.23)
(6.24)
454
where we also exhibit the crossing symmetry relations for the single variable function f .
The corresponding solution to (6.21) is given by

u2
2
G(u, v; , ) = Sn u , 2 ,
(6.25)
v
which implies

f (x) = Sn x 2 ,

x2
.
(1 x)2
(6.26)
In this case if we consider the contribution of individual factors in the basis given by (6.4)
to f (x) as expected from (6.25) and (6.26) we have

u2
u4
2
2

=p ,
Q = 2
= q 2,
P = u + 2 +1
v
v
=1/x
=1/x
4

u
u2
2
R=
(6.27)
+
u
= 2pq,
v2
v 2 =1/x
where
p(x) =
x2 x + 1
,
1x
q(x) =
x2
,
1x
(6.28)
so that we have the relation R 2 = 4P Q. In consequence we may restrict in (6.4) to those

polynomials with i = 0, 1.
Conversely we may argue that for the N = 2 case all single variable functions f (x)
may be expressible in terms of Sn as in (6.26) and therefore may be extended to a fully
crossing symmetric form for G(u, v; , ) as exhibited in (6.25). To demonstrate this we
suppose all solutions of the crossing symmetry relations in (6.24) for f are solvable by
writing

f (x) = p(x)2n g s(x) ,
s(x) =
q(x)
,
p(x)3
(6.29)
for some function g of the crossing invariant s given by (6.14). Note that for x 0, s x 2 ,
x 1, s (1 x)2 and for x , s 1/x 2 . From the superconformal representation
theory for the corresponding contributions to the operator product expansion f (x) should
be analytic in the neighbourhood of x = 0 with singularities only at x = 1, . In consequence g(s) must be a polynomial which is then restricted to have maximal degree [ 23 n]
to avoid singularities when x 2 x + 1 = 0. It is then easy to see that f can be written as
a polynomial in P , Q with terms also linear in R, as defined in (6.27), which is consistent
with (6.26) where Sn has an expansion in terms of Sn,(i,j ) with i = 0, 1 and j restricted as
in (6.4).
A similar discussion is possible for N = 4. The function f (x, ) is required to be
a general solution of the crossing symmetry conditions given by (6.20) which is also a
polynomial of degree p in . It is also analytic in x in the neighbourhood of x = 0 with
455
singularities only at x = 1, . If we write

f (x, ) = P (x, )p g(x, ),
P (x, ) =
x 2 2x + 2x 1
,
x 1
(6.30)
then g is an invariant under the action of S3 , as displayed in Table 1. Determining a general

form for g is then reducible to finding a basis for all possible independent invariants which
may be formed from x and . Since the action of S3 on any polynomial in may be
decomposed, up to functions of the invariant s(), into contributions linear in 1, a()
and (br (), cr ()), r = 1, 2, as given in (6.16) and (6.17), (6.18), then a basis for such
invariants is obtained, in addition to the separate invariants s(x), s(), by combining these
non-trivial irreducible representations with corresponding representations involving x to
give

A(x, ) = a x 1 a(),
(6.31)
where a(x 1 ) = a(x), and also

S1 (x, ) = b1 x 1 b1 () + c1 x 1 c1 ()
2(x 1)2
4
2
,
3 ( + 1)(x 2 x + 1)

S2 (x, ) = b2 x 1 b2 () + c2 x 1 c2 ()
=
2(1 )x(1 x)
(x 2 2x + 1),
( 2 + 1)2 (x 2 x + 1)2

S3 (x, ) = b2 x 1 b1 () + c2 x 1 c1 ()
2

2x(1 x)
= 2
x 2x + 2 1 ,
2
2
( + 1)(x x + 1)

S4 (x, ) = b1 x 1 b2 () + c1 x 1 c2 ()
2

2(1 )
x 2x + 2x 1 .
= 2
2
2
( + 1) (x x + 1)
=
(6.32)
These are not independent since

3
S1 (x, )S2 (x, ) S3 (x, )S4 (x, ) ,
4
1
1
S2 (x, ) S1 (x, )S3 (x, ) S3 (x, ) = 2s(x),
2
3
1
1
S2 (x, ) S1 (x, )S4 (x, ) S4 (x, ) = 2s(),
2
3

2
8
2 S3 (x, ) + S4 (x, ) 6S2 (x, ) + S1 (x, )2 S1 (x, ) = .
3
9
A(x, ) =
(6.33)
A crucial further constraint arises from (5.5) which here requires that f (x, x 1 ) is a constant. Since P (x, x 1 ) = 3 we also require that g depends on invariants sr (x, ) such that
456
sr (x, x 1 ) are constants. Taking account of the relations in (6.33) there are then two independent solutions which we take as
s1 (x, ) = 2
S3 (x, )s()
R(x, )
=
,
S4 (x, )2
P (x, )2
s2 (x, ) = 8
s(x)s()2
Q(x, )
=
,
3
S4 (x, )
P (x, )3
R(x, ) =
x(x 2 2x + 2 1)
,
1x
Q(x, ) =
x 2 (1 )
,
x 1
(6.34)
where R(x, x 1 ) = 3, Q(x, x 1 ) = 1. It is then evident that g in (6.30) must be of the form

g=
cij s1i s2 .
(6.35)
i,j 0
2i+3j p
Noting that

u
u2

P (x, ) = u + + 1
,
Q(x, ) =
,
v
v
=1/
x
=1/
x

2
u
u
+ u +
R(x, ) =
,
v
v
=1/
(6.36)
it is easy to see, as a consequence of (6.4), that Ir (u, u/v)|=1/
x = sr (x, ) and hence

the expression given by (6.30) and (6.35) for the function f may always be extended to
a fully crossing symmetric result
for the full correlation function G of the form (6.21)
with Sp (, ) = ( + + 1)p i,j cij I1 (, )i I2 (, )j and where f satisfies (6.22). With
appropriate coefficients for the independent terms in Sp (6.21) corresponds to the results of
free field theory. In general, using the formalism of harmonic superspace, the Intriligator
insertion technique [29] demonstrates that only H as in (1.6) or (1.7), or K as in (3.7) or
(5.8), can depend on the coupling g, and so are dynamical. The functions f (x) or f (x, )
are then identical with the free theory, or g = 0, results.
The remaining part of the correlation function may also be expressed in terms of S3
representations. It is convenient to define from (6.9) and (6.10) (br (u, v), cr (u, v)) =
(br (1/u, v/u), cr (1/u, v/u)). For the N = 2 case we may then write for the factor which
appears in the solution of the superconformal identities in (1.7)
(x 1)( x 1)

2

1 1

= + 1 (u + v + 1)
b (u, v)b1 () + c1 (u, v)c1 () .
3 2 1
For N = 4 we may also note
(6.37)
457
(x 1)( x 1)(x
1)( x 1)
1 2
= u (y z)(y z )(y z)(y z )
16
= v + 2 uv + 2 u + v(v 1 u) + (u + v 1) + u(u 1 v)
= ( + + 1)2 (u + v + 1)2

1
1
1
I1 (u, v) + I1 (, ) 2I1 (u, v)I1 (, ) b1 (u, v)b2 (, )
3
3
2
+ c1 (u, v)c2 (, ) + b2 (u, v)b1 (, ) + c2 (u, v)c1 (, ) 3b2 (u, v)b2 (, )

3c2 (u, v)c2 (, ) .
(6.38)
The function K in (5.8) must then satisfy the crossing symmetry relations
p+2

u
u 1
1
.
K(u, v; , ) = K , ; , =
p2 K v, u; ,
v v
v

(6.39)
It is also of interest to extend the considerations of crossing symmetry to the next-toextremal case when p1 = p2 = p3 = p, p4 = 3p 2. In this case G, defined by (5.1), must
satisfy for the permutations (12) and (23)

u 1
p1
G(u, v; , ) = v
, ; , ,
G
v v

1 v 1
2p1
G(u, v; , ) = u
(6.40)
, ; ,
.
G
u u
The solution (5.59) can be rewritten as
G(u, v; , )

x x
1
1
1
1
= up1 k +

f (x)

f (x)
,
x x
x
x
x
x
(6.41)
and then (6.40) requires

x
1
2
f (x) = f
(6.42)
,
f (x) = x f
+ kx.
x 1
x
A particular solution of (6.42) is given by

k
x
f (x) =
x
.
3
x1
(6.43)
To obtain a general solution of (6.42) it is then sufficient to seek the general solution f0 (x)
of (6.42) with k = 0. Using results obtained above this is
f0 (x) =

(x 2)x(x + 1)(2x 1)
h s(x) ,
2
(x 1)(x x + 1)
(6.44)
458
where s is the invariant defined by (6.29) and (6.28). This introduces unphysical singularities for x 2 x + 1 = 0 unless cancelled by h. However, for compatibility with semi-short
representations, h(s) must be analytic in s for s 0 (if h(s) = 1/s, which cancels the singularity at x 2 x + 1 = 0, then f0 (x) 1/x for x 0). Hence we conclude that there is
no possible solution of the form (6.44) and hence we only have (6.43). In this case

1 p1
u
G(u, v; , ) = ku
1 + u +
.
3
v
(6.45)
7. Large N results
In this paper we have endeavoured to work out the consequences of superconformal
symmetry for the four point correlation functions of BPS operators. As a result our considerations are lacking in dynamical input since we do not consider any details of N = 2
or N = 4 superconformal theories. The results of our analysis demonstrate that the details
of the dynamics resides in the function H, which appears in (1.6) and (1.7), or K as in
(5.8). In particular cases results have been obtained using perturbation theory [6] or with
the AdS/CFT correspondence [7,9,10]. We here summarise some of the results obtained in
[7,9,10] in the context of this paper.
For pi = p in the large N limit the leading result for single trace operators may be, with
a suitable normalisation, simply obtained from disconnected graphs in free field theory
p
u
G0 (u, v; , ) = 1 + ( u)p +
.
v
(7.1)
The definitions (5.4) and (5.5) then give

f0 (z, y) = 1 +
1+y
1+z
p
+
1y
1z
p
,
k = 3.
Using (5.8) we can then determine, assuming K(u, v; , ) =

H0 (u, v; , ) = 1 +
1
,
v2
(7.2)
1 2
16 u H(u, v; , ), for
p=2
(7.3)
and for p = 3
H0 (u, v; , )

1

1 1
= 3
( + )u 1 + v 3 + ( ) 3u 1 v 3 + 2(1 v) 1 + v 3
2
v 2

3
3
4
+ u 1 + v 1 + 2v + 2v v ,
(7.4)
while for p = 4
H0 (u, v; , )

1

1
= 4 u2 1 + v 4 + ( )2 2(1 v)2 1 + v 4 5u 1 + v 5
2
v

+ 3uv 1 + v 3 + 4u2 1 + v 4

1
+ 2 2 u(1 v) 1 + v 4 2u2 1 v 4
2

1
+ ( + ) (1 v)2 + u(1 + v) 1 + v 4
2

1
+ ( ) (1 v) 3(1 + v) + 7u 1 + v 4 + 8v(1 v) 1 + v 3
2

4u2 1 v 4 + 1 + v 6 3v(1 v) 1 v 3

4
2
4
2u(1 v) 1 v + u 1 + v
.
459
(7.5)
In each case the crossing symmetry relation H0 (u, v; , ) = H0 (u/v, 1/v; , )/v 2 is satisfied but the corresponding one for u v is not since it is necessary to take account of
the function f0 (z, y) then as well.
The large N results obtained through the AdS/CFT correspondence are expressible in
terms of functions D n1 n2 n3 n4 (u, v) which satisfy various identities listed in [19] and [9].
When p = 2
H(u, v; , ) =
4 2
u D2422 (u, v),
N2
(7.6)
and for p = 3,

9 3
u (1 + + )D 3533 + D 3522 + D 2523 + D 2532 .
2
N
For p = 4 the results obtained in [10] can be rewritten as
H(u, v; , ) =
(7.7)
H(u, v; , )

4
= 2 u4 1 + 2 + 2 + 4 + 4 + 4 D 4644 + 2(D 4633 + D 4622 )
N
+ 2 2 (D 3634 + D 2624 ) + 2 2 (D 3643 + D 2642 ) 4 (D 4624 2D 3623 )

4 (D 4642 2D 3632 ) 4 (D 2644 2D 2633 ) .
(7.8)
1 2
u H(u, v; , ) it is easy to verify both the crossing symmetry
Since K(u, v; , ) = 16
conditions (6.39) using D identities. Furthermore the results given by (7.6)(7.8), in which
overall factors of up are present, are manifestly compatible with the unitarity conditions
flowing from (5.16) and (5.29) since the leading log. term D n1 n2 n3 n4 (u, v) is log u itself.
()
When expressed in terms of conformal partial waves G+4 it is easy to see in each case that
only contributions with minimum twist = 2p are required. Hence (7.6)(7.8) require
the presence of operators belonging to long multiplets which have anomalous dimensions
with twist, at zeroth order in 1/N , = 2(p + t), t = 0, 1, 2, . . . for the lowest scale
dimension operators in each multiplet. The condition = 2p is stronger than that
460
required by unitarity (5.29), with n p 2, which shows that for any representation some
low twist multiplets decouple (thus for the singlet case twist 2 is absent as it disappears
in the large N limit but twist 4 multiplets, which are necessary in the p = 2 correlation
function, decouple from the correlation functions for p = 3, 4).
To obtain the anomalous scale dimensions in detail it is necessary to decompose both
(7.6)(7.8) and (7.3)(7.5) in terms of different representations, as in (5.16), and then to
expand each term in conformal partial waves. The expressions (7.3)(7.5) require contributions with twist zero and above but the corresponding low twist operators in long
supermultiplets, for which there are no anomalous dimensions, are cancelled by semi-short
multiplets which are required by the expansion of f0 (z, y). For p = 2 and p = 3 a detailed
discussion is contained in [19] and [9] (although some details are different the analysis
is equivalent to the results that would be obtained by expanding H as given by (7.3) and
(7.4)).
8. Conclusion
In this paper we have derived requirements arising from N = 2 and N = 4 superconformal symmetry for the four point correlation functions of BPS operators. The conditions in
essence are obtained from the absence of representations at the first level, obtained by the
action of a single supercharge on the superconformal primary state. The derived conditions
are clearly necessary but not manifestly sufficient in that it is possible to imagine that there
are further constraints arising from superconformal transformations involving higher levels
which are obtained by the action of more than one supercharge, although we have no reason
to suppose that there any such additional conditions. A related question is whether the four
point correlations functions of all descendant operators are determined uniquely from the
basic correlation function for the superconformal primary BPS operators. In the simplest
three point function case the correlation function for various descendants, including the
energymomentum tensor three point function, was calculated by hand in [19]. In a superspace formalism these questions are straightforward to address, the question of uniqueness
depends on whether there are any nilpotent superconformal invariants which are formed
from the anti-commuting coordinates. Nevertheless it would be nice to show directly
that the correlation function of all descendant operators could in principle be obtained by
the action of differential operators acting on the basic correlation function, subject to the
conditions for superconformal invariance derived here.
Another area of possible future investigation is whether the requirements of crossing
symmetry and superconformal invariance might be extended further using constraints arising from factorisation and the operator product expansion, as in the classic bootstrap
framework. In the above we showed how there were conditions on the single variable
functions that arise in the solution of superconformal identities which dictate that they are
essentially of free field form. Our arguments are restricted to the case where all but one of
the operators in the correlation function are identical.
Finally we may mention that the use of null vectors t to conveniently express arbitrary
rank traceless symmetric tensor fields in the form (p) (x, t), homogeneous of degree p
in t, may for conformal fields be also written in terms of homogeneous coordinates A on
461
the null cone 2 = 0 [30] such that (p)(, t) = (p) (, t). For N = 4 both and t
are 6-vectors. The expansion in terms of harmonic polynomials as discussed in Appendix B
has a direct analogue to the conformal partial wave expansion which was explored in [31].
Acknowledgements
We are particularly grateful to Francis Dolan for much help and collaboration during
various stages of this investigation. H.O. would also like to thank Gleb Arutyunov and
Emeri Sokatchev for many very useful conversations and emails and for showing us details
of their work at an early stage. M.N. would like to thank PPARC, the Cambridge European
Trust and the Isaac Newton Trust for financial assistance.
Appendix A. Results for null vectors

We discuss here some results for null vectors tr which are useful in the text. For generality we allow t to be d-dimensional. As a consequence of (1.3) differentiation requires
some care but for any null vector a we may define as usual
(at)n = n(at)n1 a.
t
More generally for a set of null vectors a1 , a2 , . . . , ap we have
(A.1)
p

(ai t)ni
t
i=1
p

ni (a1 t)n1 (ai t)ni 1 (ap t)np ai
i=1
ni nj ai aj (a1 t)n1 (ai t)ni 1 (aj t)nj 1 (ap t)np t,
1i<j p
(A.2)
where
2
,
R=
2N + d 4
N=
p

ni .
(A.3)
i=1
The right-hand side of (A.2), with (A.3), may be represented in the form

p
1
2
(ai t)ni ,
t
t
2t + d
(A.4)
i=1
where the action of the derivatives is as usual, without regard to the constraint t 2 = 0.
The resulting operator is equivalent to a definition given in [21] for an interior differential
operator on the complex null cone.
462
From (A.2) we may readily find

p

,
(ai t)ni = 0,
tr ts
i=1
p

p

i=1
i=1
(ai t)ni = N
p

(ai t)ni = 0,
t t
i=1
(ai t)ni .
(A.5)
We also have

p

p

2
, ts
(ai t)ni = rs
(ai t)ni ,
tr
tr
2N + d 2 ts
i=1
(A.6)
i=1
which implies
p

p

(2N + d)(N + d 2)
ni
(ai t)
(ai t)ni .
tr
=
tr
2N + d 2
i=1
(A.7)
i=1
Defining the generators of SO(d) by

Lrs = tr s ts r ,
(A.8)
then the above results give

Lru Lsu
p
p

(ai t)ni = (N + d 3)tr s + (N 1)ts r + N rs
(ai t)ni ,
i=1
i=1
(A.9)
and

1
Lrs Lrs
(ai t)ni = N (N + d 2) (ai t)ni ,
2
p
i=1
i=1
(A.10)
which reproduces the appropriate eigenvalue of the Casimir operator for the representation
formed by traceless rank N tensors.
If Vr (t) is homogeneous of degree N then in general
1
(ts Vs ),
tr Vr = 0,
N + 1 tr
as used in (2.24) and (4.12). If Vr also satisfies
Vs
Vr = 0,
Vr = 0,
tr
ts
tr
Vr = Vr +
(A.11)
(A.12)
then, by contracting with ts and using (A.5), (A.6), we easily see that Vr = 0. As a further
corollary if Vr = s Urs , Urs = Usr , [r Usu] = 0 then (N + 1)Vr = r ( 12 Lsu Usu ) with
Lsu as in (A.8). In general we have the decomposition
Vr =
2N + d 4
tr V
(2N + d 2)(N + d 3)

1
(2N + d 2)s (tr Vs ts Vr ) + 2r (tV ) .

(2N + d)(N + d 3)
(A.13)
463
Appendix B. Two variable harmonic polynomials

For the expansion of four point functions in terms of R-symmetry representations we
consider here the eigenfunctions of the SO(d) Casimir operator
1
L2 = Lrs Lrs ,
2
where the generators are
(B.1)
Lrs = t1r 1s t1s 1r + t2r 2s t2s 2r ,
(B.2)
formed by homogeneous functions of the null vectors t1 , t2 , t3 , t4 . Obviously Lrs t1 t2 = 0

and hence L2 (t1 t2 )k (t3 t4 )l f (, ) = (t1 t2 )k (t3 t4 )l L2 f (, ), where , are given by
(1.4). We therefore first consider eigenfunctions which are polynomials in ,
Y (, ) =
t

ct,q tq q ,
(B.3)
t0 q=0
satisfying
L2 Y (, ) = 2CY (, ).
With the aid of the given in Appendix A we may easily calculate the action of
monomial formed from , ,

L2 p q = 2 (d 2)(p + q) + 4pq p q

+ 2(1 ) p 2 p1 q + q 2 p q1 ,
or

1 2
L Dd = (1 )
4
2

+
.
(d 2)
(B.4)
L2
on a
(B.5)
(B.6)
Alternatively Dd may be written in the form

1 T
(1 )
2
Dd = wG , G =
,
2
(1 )
w

= ,
(B.7)
where, with = ( + + 1)( + 1)( + 1)( 1) as in

(2.42),
1
w = 2 (d5) .
(B.8)
In general, for a polynomial as in (B.3) with tmax = n, we must have that cn,q forms an
eigenvector for an (n+1) (n+1) matrix Mn ,
Mn,pq cn,q = Ccn,p ,
464

Mn,pq = pq n(n + d 2) + 2p(n p) + pq1 q 2 + pq+1 (n q)2 .
(B.9)
The coefficients ct,q with t < n may then be obtained by solving recurrence relations. For
given n there are n + 1 eigenvectors solving (B.9) and the corresponding eigenfunctions
are
Ynm (, ),
Cnm = n(n + d 3) + m(m + 1),
n = 0, 1, 2, . . . ,
m = 0, . . . , n.
(B.10)
As a consequence of (B.7) and

(B.8)
the polynomials are orthogonal for d > 5 with respest
to integration over , 0, + 1 with weight w (for a general discussion of such
two variable orthogonal polynomials see [32,33]).
The polynomials Ynm are also eigenfunctions for higher order Casimir invariants. Letting
1
1
1
Lrs Lst Ltu Lur L2 L2 + (d 2)(d 3)L2 Q
(B.11)
4
2
4
then acting on any Y (, ) we may express Q in a form similar to (B.7),
2
2
1
1

Q= 1
( 2 2 ) 2 (d3)
2
2
(d5)
2

1
1
2
+ 1
(d5)
2
+ (d 3) 1
( )
1
2
(d5)
2

1
1
1
+ (d 2) 1
(B.12)
2 (d3) + 2 (d3) .
(d5)
2
The harmonic polynomials then satisfy
QYnm = (n m)(n + m + 1)(n + m + d 3)(n m + d 4)Ynm .
(B.13)
Using (B.5) it is straightforward to construct the first few eigenfunctions satisfying (B.4)
by hand. With an arbitrary normalisation, we find for n = 0, 1, 2, 3,
Y00 (, ) = 1,
Y10 (, ) = ,
Y11 (, ) = +
2
,
d
Y20 (, ) = 2 + 2 2
2
2
( + ) +
,
d 2
(d 2)(d 1)
4
( ),
d +2
8
8
Y22 (, ) = 2 + 2 + 4
( + ) +
,
d +4
(d + 2)(d + 4)

12
6
Y30 (, ) = 3 3 2 + 3 2 3 2 2 +
( ),
d
d(d + 1)
Y21 (, ) = 2 2
465

8(d 6)
8(d 1) 2
+ 2 +
(d + 4)(d 2)
(d + 4)(d 2)
4(3d + 2)
8
+
( + )
,
(d + 1)(d + 4)(d 2)
(d + 1)(d + 4)(d 2)

24
12 2
2 +
( ),
Y32 (, ) = 3 + 3 2 3 2 3
d +6
(d + 4)(d + 6)

72
18 2
+ 2
Y33 (, ) = 3 + 9 2 + 9 2 + 3
d +8
d +8
72
48
+
(B.14)
( + )
.
(d + 6)(d + 8)
(d + 4)(d + 6)(d + 8)
Y31 (, ) = 3 2 2 + 3
Up to an overall normalisation for d = 6 each term may be identified with terms in the
projection operators constructed in [9] where Ynm corresponds to the SU(4) SO(6) repre 2
sentation with Dynkin labels [n m, 2m, n m]. For m = n in (B.10) we have cn,q = qn
and the recurrence relations may be easily solved giving

Ynn (, ) = An F4 n, n + 12 d 1; 1, 1; , ,
(B.15)
where F4 is one of Appells generalised hypergeometric functions11 and An is some overall
constant.
To obtain more general forms (see [34]) we used the variables , defined in (1.5).
Acting on Y (, ) = P(, )
= P(,
)
1 2
= D d P(, ),
2 L P(, )
(B.16)
where, using (B.5) or (B.6), we now have
D d =
(1 )
+
(1
)

(1 )
.
+ (d 4)
(1
)

Corresponding to (B.4) and (B.10) we have

D d Pnm (, )
= n(n + d 3) + m(m + 1) Pnm (, ),
(B.17)
(B.18)
are generalised symmetric Jacobi polynomials. For particular d simpliwhere Pnm (, )

fied formulae may be found in terms of well know single variable Legendre polynomials
Pn . When d = 4 it is clear from (B.17) that D 4 is just the sum of two independent Legendre
differential operators so that

Pnm (, )
(B.19)
= 12 Pn (y)Pm (y)
+ Pm (y)Pn (y)
, n m,
11
(a)m+n (b)m+n m n

x y .
F4 a, b; c, c ; x, y =
(c)m (c )n m!n!
m,n
466
with y, y defined in (5.3). For d = 6 we may use the result

D 6
1
1
=
(D 4 + 2),

(B.20)
to see that we can take the eigenfunctions to be of the form

Pnm (, )
= pn+1 m (y, y),
n m,
(B.21)
where
= pmn (y, y)
=
pnm (y, y)
Pm (y)Pn (y)
Pn (y)Pm (y)
.
y y
(B.22)
It is also of interest to consider d = 8 when we take

P(, )
=
F (, )
,
( )
2
and the eigenvalue equation becomes

2

D6 F (, )
( )F
(, )

(1 )
2
( )

( )F
(, )
(1
)

= (C + 4)F (, ).
(B.23)
(B.24)
If we assume
F (, )
=
anm pnm (y, y),
(B.25)
n,m
and use, from standard identities for Legendre polynomials,

1
1 y2
1 y 2
(y y)p
nm (y, y)
y y
y
y

m(m + 1)
pn m+1 (y, y)
pn m1 (y, y)
=
2m + 1

n(n + 1)
pn+1 m (y, y)
pn1 m (y, y)
,
2n + 1

1
(n + 1)pn+1m (y, y)
(y y)p
nm (y, y)
=
+ npn1 m (y, y)
2n + 1

1
(m + 1)pn m+1 (y, y)

+ mpn m1 (y, y)
,
2m + 1
(B.26)
then we may set up recurrence relations for anm which for the appropriate value of C have
just four terms. For C = n(n + 1) + m(m + 1) 6 (B.25) gives a solution
qnm (y, y)
=

n+1
1
(n + m)(n m 1)pn+1 m (y, y)
2
2n + 1
(y y)
n
(n + m + 2)(n m + 1)pn1 m (y, y)
+
2n + 1
m+1
(n + m)(n m + 1)pn m+1 (y, y)
2m + 1

m
,
(n + m + 2)(n m 1)pn m1 (y, y)
2m + 1
467
(B.27)
where qnm (y, y)

= qmn (y, y),
qnn (y, y)
= qn+1 n (y, y)
= 0. Hence we can take
= qn+2 m (y, y).
Pnm (, )
(B.28)
The above results for harmonic polynomials in , are relevant for discussing four point
functions when each field belongs to the same SO(d) representation. For the more general
case we also consider instead of (B.4),

L2 (t1 t4 )a (t2 t4 )b Y (a,b) (, ) = 2C (t1 t4 )a (t2 t4 )b Y (a,b) (, ) ,
(B.29)
where now the action of L2 is determined by

L2 (t1 t4 )a (t2 t4 )b p q

= (t1 t4 )a (t2 t4 )b 2Dd (a + b)(a + b + d 2) 4ap 4bq p q

+ 2(1 ) bp p1 q + aq p q1 ,
(B.30)
or

1 2
L (t1 t4 )a (t2 t4 )b f (, )
2

(a,b)
= (t1 t4 )a (t2 t4 )b Dd
where
(a,b)
Dd

1
(a + b)(a + b + d 2) f (, ),
2

+b
2a
2b .
= Dd + (1 ) a
(B.31)
(B.32)
This may also be written in the form (B.7) with w = b a 2 (d5) . The possible eigenvalues for polynomial eigenfunctions with maximum power p + q = n are then determined
by the matrix

1
Mn,pq = p q n(n + d 2 + a + b) + (a + b)(a + b + d 2)
2

+ 2p(n p) + a(n p) + bp
+ p q1 q(q + a) + p q+1 (n q)(n q + b).
(a,b)
The eigenfunctions Ynm (, ) for m = 0, 1, . . . , n then have eigenvalues
(B.33)
468

1
1
Cnm = n + (a + b) n + (a + b) + d 3
2
2

1
1
+ m + (a + b) m + (a + b) + 1 .
2
2
(B.34)
(a,b)
For d = 6 Ynm corresponds to the representation [nm, a +b +2m, nm]. The simplest
non-trivial examples are
ab
,
a+b+d 2
1
1
1
(a,b)
Y11
(, ) =
.
+
b+1
a+1
a + b + 12 d
(a,b)
Y10
(, ) = +
(B.35)
Corresponding to (B.15) we have in general

(a,b)
Ynn
(, ) = An F4 n, n + a + b + 12 d 1; b + 1, a + 1; , .
(B.36)
Again more explicit results can be obtained by using the variables , .

In (B.32) the
differential operator now becomes

(a,b)
= D d a b(1 )
D d
a b(1 )
(B.37)
by Pnm (, )
then previwith D d given in (B.17). Denoting the eigenfunctions of D d
ous results for d = 4, 6 for the eigenfunctions can be extended by using Jacobi polynomials
(a,b)
(a,b)
(a,b)
(a,b)
Pn . For d = 4 D 4
= D + D
where D (a,b) is the ordinary differential operator defined by
(a,b)
D(a,b) =
d
d
d
d
(1 )
a
+ b(1 ) .
d
d
d
d
(a,b)
(a,b)
(B.38)
(a,b)
The eigenfunctions of D
are just Pn (y), where y = 2 1 and the eigenvalues are
n(n + a + b + 1). For d = 6 the generalisation of (B.21) and (B.22) is then
(a,b)
(a,b)
(, )
=
Pnm
(a,b)
Pn+1 (y)Pm
(a,b)
(y)
Pm
y y
(a,b)
(y)Pn+1 (y)
(B.39)
When d = 3 the above results need to be considered separately since , are not in(a,b)
dependent and satisfy the constraint (2.42).12 The eigenfunctions Ynm (, ) are also
(a,b)
(, ) = 0 for m < n 1 as a consequence of (2.42). To obtain eigenrestricted since Ynm
2
functions of L in general we make use of the solution (2.43) which amounts to setting
= in the above, so that we are restricted just to single variable functions. Instead of
12 In terms of the two component spinors described in footnote 6 we may also define O(3) generators by
. Since
La = 12 ua u

1
(a,b)
+ (a + b)(a + b + 2) f (),
(L1 + L2 )2 (u1 u 4 )a (u2 u 4 )b f () = D
4
the relevant O(3) eigenfunctions for each representation may be directly obtained.
(B.31) we have

L2 (t1 t4 )a (t2 t4 )b f ()

= (t1 t4 )a (t2 t4 )b D(2a,2b) (a + b)(a + b + 1) f (),
469
(B.40)
using the definition (B.38). In consequence

L2 (t1 t4 )a (t2 t4 )b Pn(2a,2b) (y)
= (n + a + b)(n + a + b + 1)(t1 t4 )a (t2 t4 )b Pn(2a,2b) (y),
(B.41)
corresponding to the (n + a + b)-representation for SU(2) SO(3). Hence for d = 3 we

may then take
(2a,2b)
(a,b)
Pnn
(, ) = P2n
(y),
(a,b)
(2a,2b)
Pn n1 (, ) = P2 n1 (y),
(B.42)
(a,b)
with Pnm (, ) = 0 for m < n 1.

For d = 3 there are also eigenfunctions involving cross products. To consider these we
first define
T1 = t1 t3 t4 (t1 t4 )a1 (t2 t4 )b ,
T2 = t2 t3 t4 (t1 t4 )a (t2 t4 )b1 ,
(B.43)
and consider eigenfunctions of the form T1 f1 () + T2 f2 (). The action of L2 on such

functions is given by

2
L + (a + b 1)(a + b) (T1 f1 + T2 f2 )
(2a1,2b+1)

2a
2a
D
f1
.
= ( T1 T2 )
(B.44)
f2
2b
D (2a+1,2b1) 2b
However for ti three-dimensional null vectors the basis given by (B.43) is not independent
since we have from (2.55)
T1 (1 ) + T2 = 0,
(B.45)
so that f1 , f2 are not unique. If we use this freedom to set f2 = 0 the eigenvalue equation
for L2 reduces to

1
(2a1,2b+1)
D
(a + b 1)(a + b) f1
2a + 2b

1
= D (2a1,2b1) (a + b 1)(a + b) (f1 ) = Cf1 ,
(B.46)
which has solutions proportional to Jacobi polynomials,

1 (2a1,2b1)
(B.47)
P
(y),
C = (n + a + b 1)(n + a + b).
n
For n 1 the apparent singularity for 0 may be removed by using (B.45) to give an
appropriate non-zero f2 . The eigenfunctions for the solution in (B.47) correspond to the
SU(2) (n + a + b 1)-representation. Alternatively we may set f1 = 0 and obtain the
f1 () =
470
corresponding equation

1 (2a1,2b1)
D
(a + b 1)(a + b) (1 )f2 = Cf2 .
1
(B.48)
Appendix C. Calculation of differential operators

A non-trivial aspect in the derivation of the superconformal identities is the determination of the differential operators (2.48) which appear in (2.47). To sketch how these were
obtained we first obtain, for any dimension d and arbitrary f (, ),

1
1
k + a + d 2 L2[rs 1u] (t1 t2 )k (t3 t4 )l (t1 t4 )a (t2 t4 )b f
2
2

= (t1 t2 )k2 (t3 t4 )l1 (t1 t4 )a (t2 t4 )b t1[r t2s t3u] t2 t4 D1 + t1[r t4s t2u] t2 t3 D2

1
+ k + a + d 2 t2[r t3s t4u] t1 t2 (D D ) f,
(C.1)
2
where

D +
+
+ 1 k (D D ),
D1 =

a
+
D +
+
+ 1 k (D D ),
D2 =
(C.2)
for
2
2
2
+ (b + 1)
D = (1 ) 2 2 2 2

1
1
+
+k k+a+b+ d 1 ,
a+b+ d
2
2
2
2
2
D = (1 ) 2 2 2 2
+ (a + 1)

1
1
+
+k k+a+b+ d 1 .
a+b+ d
2
(C.3)
In terms of (B.32) and (B.6) we have

(a,b)
D + D + ( )(D D )

1
= Dd(a,b) + 2k k + a + b + d 1 .
2
(C.4)
The operators in (C.2) satisfy the identity

2( + 1)D1 + 2( + 1)D2

(a,b)
+
+
(D D ) (2k + 2a 1)(D D ),
= D2 d
(C.5)
with as in (2.42) and, as well as (C.4), defining

a
D2 = ( + 1)
+ ( 1)
+
.
471
(C.6)
When d = 3, and = 0, this result with (2.55) leads to the simplified form for (C.1)

1
L2 (t1 t2 )k (t3 t4 )l (t1 t4 )a (t2 t4 )b f
4 k+a
2 t1
= t2 t3 t4 (t1 t2 )k1 (t3 t4 )l1 (t1 t4 )a (t2 t4 )b

D 2 D (2a,2b) + 2k(2k + 2a + 2b + 1) f,
(C.7)
letting f (, ) = f() and D2 D 2 for

2a
d
.
D 2 =
d 1
(C.8)
From (B.40) the operator D (2a,2b) + 2k(2k + 2a + 2b + 1) acting on f() corresponds to

L2 + (2k + a + b)(2k + a + b + 1). We may also note that
1
D (2a1,2b+1) (1 )D 2 ,
1
and in (C.7) from (B.44)
D 2 D (2a,2b) =
(C.9)
t2 t3 t4 (t1 t2 )k1 (t3 t4 )l1 (t1 t4 )a (t2 t4 )b

1 (2a1,2b+1)
D
+ 2k(2k + 2a + 2b + 1) (1 )f
1

= L2 + (2k + a + b)(2k + a + b + 1) t2 t3 t4
(t1 t2 )k1 (t3 t4 )l1 (t1 t4 )a (t2 t4 )b f.
(C.10)
The equivalent results to (C.7) for L2 L3 and L2 L4 can be found by using

the permutations 2 3 4 2, along with a a = k l, b a, k a + l,
l a + b + l and = (1 )/, and also 2 4 3 2, along with in this
case a a = b, b l k, k a + b + l, l b + k and = 1/(1 ). From
(C.7) we then find
D 3 = 2(a+l1) (1 )2a D 2 2(a+l) (1 )2a
2a
2(k + a)
d
+
,
=
d 1
d
2k
D 4 = 2b (1 )2(b+k1) D 2 2b (1 )2(b+k) =
(C.11)
+
.
d 1
Together with (C.8), (C.11) is equivalent to (C.4).
For the analysis of the N = 4 superconformal identities a particular solution of the
constraints (4.7) is obtained by expressing Ti in terms of scalar functions Yi (u, v; t)
Ti = Yi .
t1
ti
(C.12)
472
(4.8) and (4.12) then give

Ui = (L1,rs Li,rs + p1 pi )Yi ,
Wi,rsu = 3(1,[r Li,su] Yi )sd ,

1
1
L1,su Li,su Yi .
Vi,r = 1s Li,rs Yi 1r
p1
2
(C.13)
Writing
Yi (u, v; t) = (t1 t4 )p1 E (t2 t4 )p2 E (t1 t2 )E (t3 t4 )p3 Yi (u, v; , ),
(C.14)
then for i = 2, using (C.1) with k = E, k + a = p1 , k + b = p2 , we find

(p1 E,p2 E)
U2 = 6
Y2 ,
W2 = 6(D D )Y2 .
(C.15)
A2 and B2 are then given by (C.1) and (C.2) in terms of U2 , W2 in accord with (4.37).
The other results may be obtained by cyclic permutations. For 2 3 4 2, when
/ , 1/ and E E p4 p2 , then U2 p1 E p2 p4 E U3 , W2
p1 E p2 p4 E+1 W3 , A2 p1 E p2 p4 E+2 C3 and B2 p1 E p2 p4 E+2 A3 .
For 2 4 3 2, so that 1/ , / and E E + p1 + p2 , in this
case U2 p2 p2 E U4 , W2 1p2 p2 E W4 , A2 2p2 p2 E B4 and B2
2p2 p2 E C4 .
However, the representation (C.12) is not valid in general since it excludes contributions
involving the -tensor. Nevertheless equivalent results may be obtained by use of (4.24).
With the expansion
t2[r 1s] U2 + p1 t2[r V2,s] + p1 W2,rsu t2u
= (t1 t4 )p1 E1 (t2 t4 )p2 E (t1 t2 )E1 (t3 t4 )p3 1

1
t2[r t3s] t1 t4 t2 t4 U2 + p1 J2 p1 (A2 + W2 )
6

1
1

+ t2[r t4s] t1 t4 t2 t3 U2 p1 (I2 + J2 ) + p1 (B2 + W2 )
6

+ (E ) + U2
+ t1[r t2s] t2 t4 t3 t4
p1 + 1

1
p1 V2 p1 (A2 B2 ) ,
6
(C.16)
where = + (p1 E)/ , then (4.24) requires

6 U2 = 6p1 J2 + p1 (A2 + W2 ) = 2(p1 + 1)A2 (O p1 )W2 ,
1
6 U2 = 6p1 (I2 + J2 ) p1 (B2 + W2 ) = 2(p1 + 1)B2 + (O p1 )W2 ,
(C.17)
using (4.35) for i = 2, which gives the first two equations in (4.37). The remaining results
in (4.37) can be obtained by using permutations. In addition with (4.36) we also obtain

(p1 + 1) (O p1 + 1)B2 (O p1 + 1)A2

= 6 (E 1 )( + ) + U2

= 3 (O p1 + 1) + (O p1 + 1) U2 .
473
(C.18)
It is then straightforward to see that (C.18) follows from (C.17) using [O , O ] = O

O .
Appendix D. Non-unitary semi-short representations

In Section 5 the analysis of the operator product expansion in general potentially required contributions below the unitarity threshold on the scale dimension . We show here
how such truncations of the full representation space arise for the superconformal algebra
PSU(2, 2|4), following the approach in [25].13
The essential results are found by considering the chiral subalgebra SU(2|4) (although no hermeticity conditions are imposed) which has generators Qi , S i , = 1, 2,
i = 1, . . . , 4, where

i
1
Q , Sj = 4 i j M + D R i j ,
(D.1)
2
as well as {Qi , Qj } = {Si , Sj } = 0. In (D.1) M are generators of SU(2) and R i j are

generators of SU(4), i R i i = 0, with standard commutation relations. D is the dilation
operator, with eigenvalues the scale dimension. The commutators with Qi and Si are
then

i
M , Si = Si + 12 Si ,
M , Q = Qi 12 Qi ,

i
R i j , Sk = i k Sj + 14 i j Sk ,
R j , Qk = k j Qi 14 i j Qk ,

Si = 1 Si .
Qi = 1 Qi ,
D,
D,
(D.2)
2
2
In terms of the usual J3 , J

J3 J+
M =
,
J J3
(D.3)
and it clear then that (Qi 1 , Qi 2 ) and (Si 2 , Si 1 ) form j = 12 doublets. In terms of a
standard Chevalley basis Er , Hr , r = 1, 2, 3, where Hr = Hr , Er + = Er with commutators [Hr , Hs ] = 0, [Er + , Es ] = rs Hs , [Hr , Es ] = Ksr Es , for [Krs ] the SU(4)
Cartan matrix,

2 1 0
[Krs ] = 1 2 1 ,
(D.4)
0 1 2
13 This reproduces an analysis in [35].
474
then we may take R 1 2 = E1 + , R 2 3 = E2 + , R 3 4 = E3 + and R i i = 14 (3H1 + 2H2 + H3 )

i1
r=1 Hr .
For SU(4) SU(2) highest weight states |p1 , p2 , p3 ; j hw |p; j hw we have
Hr |p; j hw = pr |p; j hw ,
J3 |p; j hw = j |p; j hw ,
J+ |p; j hw = Er + |p; j hw = 0,
(D.5)
from which states defining a representation with Dynkin labels [p1 , p2 , p3 ]j are constructed by the action of Er , J . The representations of SU(2|4) may then be formed
from a highest weight state which is also superconformal primary,
D|p;
j hw = |p; j hw ,
Si |p; j hw = 0.
(D.6)
The states of a generic supermultiplet, labelled by a[p

, are obtained by the ac1 ,p2 ,p3 ]j

i
n
hw
i
tion of the supercharges, giving i, (Q ) |p; j with ni = 0, 1, together with the
lowering operators Er . The possible SU(4) SU(2) representations [p1 , p2 , p3 ]j , with
scale dimension , forming the supermultiplet a[p

are obtained by adding the
1 ,p2 ,p3 ]j
SU(4), SU(2) weights with ni = 0, 1 so that
pr = pr +

(nr nr+1 ),
= +
1
2
j = j +
1
(ni1 ni2 ),
2
i
ni .
(D.7)
i,
= 28 d(p1 , p2 , p3 )(2j + 1), where d(p1 , p2 , p3 ) is

It is easy to see that dim a[p
1 ,p2 ,p3 ]j
the dimension of the SU(4) representation with Dynkin labels [p1 , p2 , p3 ]. If in (D.7) any
pr or j are negative the RacahSpeiser algorithm, described in [25], provides a precise
prescription for removing such [p1 , p2 , p3 ]j .
Shortening conditions arise for suitable when descendant representations satisfy the
conditions (D.6) to be superconformal primary. Since all Si are obtained by commutators
of S1 1 with Er + and J+ it is sufficient to impose only that S1 1 annihilates the highest
weight state of the representation. In such cases we may impose that the appropriate combinations of Qi annihilate |p; j hw . For application here it is convenient to define, acting
on states | such that J3 | = j | ,
i = Qi 2 1 Qi 1 J .
Q
2j
(D.8)
i | . From (D.1) we have

i | = 0 and J3 Q
i | = (j 1/2)Q
If J+ | = 0 then J+ Q

1 1 1
1
1
j S1 , Q = 2j J3 D + (3H1 + 2H2 + H3 ) J ,
2
2
4

1
2 = E 1 J .
j S1 1 , Q
(D.9)
2
475
1 |p; j hw |p1 + 1, p2 , p3 ; j 1/2 . The shortening

It is straightforward to show that Q
conditions considered in [25] and previously are obtained by imposing
i = 1,
if p1 = 0,
i
hw
|p; j = 0 i = 1, 2,
Q
(D.10)
if p1 = p2 = 0,
i = 1, 2, 3,
i = 1, 2, 3, 4, if p1 = p2 = p3 = 0.
In each case there is a consistency condition on which can be found by using (D.9) and
(D.6),
= 2 + 2j + 12 (3p1 + 2p2 + p3 ).
(D.11)
The corresponding supermultiplet is here denoted by c[p1 ,p2 ,p3 ]j . Detailed results were
given in [25], the SU(4) SU(2) representations present may be calculated as in (D.7)
with the restriction ni2 = 0 for those i listed in (D.10) for each case.
There are also additional shortening conditions of the form

2 1 Q
1Q
1 E1 |p; j hw = 0, p1 > 0,
2 |0, p2 , p3 ; j hw = 0,
Q
Q
p1
(D.12)
where the left-hand sides correspond to highest weight states |p1 1, p2 + 1, p3 ; j 1/2
and |0, p2 + 1, p3 ; j 1 respectively. Using (D.9) these conditions then require
= 2j + 12 (p1 + 2p2 + p3 ).
(D.13)
1 [E2 , E1 ])|p; j hw = 0. The

For p2 = 0 the condition (D.12) extends also to (Q 3 p11 Q
supermultiplet in each case is denoted by d[p1 ,p2 ,p3 ]j . The representations are obtained as
in (D.7) with n22 = 0, or if p2 = 0 then n22 = n32 = 0. For p1 = 0 it is sufficient to exclude
n12 = n22 = 1.
These semi-short representations lead to decompositions of the generic multiplet,
2+2j + 1 (3p1 +2p2 +p3 )
a[p1 ,p2 ,p2 3 ]j

a
a
c[p1 ,p2 ,p3 ]j c[p1 +1,p2 ,p3 ]j 1/2 ,
2j + 12 (p1 +2p2 +p3 )

[p1 ,p2 ,p3 ]j
[p1 1,p2 +1,p3 ]j 1/2
[p1 ,p2 ,p3 ]j
2j + 12 (2p2 +p3 )
[0,p2 ,p3 ]j
[0,p2 +1,p3 ]j 1
[0,p2 ,p3 ]j
(D.14)
Formally, as discussed in [25], we have

c[p1 ,p2 ,p3 ]1/2 b[p1 +1,p2 ,p3 ] ,
c[p1 ,p2 ,p3 ]1 b[p1 ,p2 ,p3 ] ,
(D.15)
where b[p1 ,p2 ,p3 ] is the short supermultiplet formed by imposing

= 0 where
1
we require = 2 (3p1 + 2p2 + p3 ). Just as in (D.15) d[p1 ,p2 ,p3 ]1/2 may be identified
with a multiplet obtained from the highest weight state |p1 1, p2 + 1, p3 ; 0 hw , with
= 12 (p1 + 2p2 + p3 + 1), annihilated by Q2 p11 Q1 E1 . Formally we have
Q1
d[p1 ,p2 ,p3 ]j c[p1 2,p1 +p2 +p3 +2,p3 2]j ,
|p; 0 hw
(D.16)
where, in accord with the RacahSpeiser algorithm described in [25], we may identify
SU(4) representations [p1 , p2 , p3 ] [p1 2, p1 + p2 + p3 + 2, p3 2] which are
476
related by an even element of the Weyl group. This allows the detailed representation
content and dimension for d[p1 ,p2 ,p3 ]j to be determined from the results given in [25].
The generators of the superconformal group PSU(2, 2|4) are obtained by extend i = Qi , S i = Si ,
ing those of SU(2|4) to include the Hermitian conjugates, Q
M = (M ) , with an algebra obtained by conjugation of that for SU(2|4), assuming

j } =
D = D and (R i j ) = R j i . In addition {Si , Q j } = {Qi , S j } = 0 and {Qi , Q
i
a
i
i
a
2 j ( ) Pa , {S , Sj } = 2 j ( ) Ka where Pa is the momentum operator and Ka

the generator of special conformal transformations. The supermultiplets are generated from
highest weight superconformal primary states |p1 , p2 , p3 ; j, hw , where is the SU(2)
quantum number for J , J3 obtained from M , which is annihilated by Si , S i and

Ka . The representations are of course infinite-dimensional, since they are generated by
arbitrary powers of Pa , but they are formed by a finite set of conformal primary representations, annihilated by Ka , which are straightforwardly constructed from SU(2|4)
supermultiplet representations, as described above, combined with their conjugates formed
i . For these j and p1 p3 . The generic supermultiplet is deby the action of Q
noted A
[p1 ,p2 ,p3 ](j,) where [p1 , p2 , p3 ](j, ) are the labels for the representation with
lowest scale dimension . The conformal primary states form representations labelled by
[p1 , p2 , p3 ](j , ), with scale dimension , which are given by
pr = pr +

(nr nr+1 ) +
(n r+1 n r ),
1
(ni1 ni2 ),
j = j +
2
= +
= +
1
2
i,
ni +
1
2
1
(n i2 n i1 ),
2
i
n i ,
ni , n i = 0, 1.
(D.17)
i,
The total dimension is 216 d(p1 , p2 , p3 )(2j + 1)(2 + 1).
We here consider the case when shortening conditions are imposed for both the Q and Q
charges. Requiring (D.10) with (D.11) together with its conjugate we have the semi-short
multiplets,
C[p1 ,p2 ,p3 ](j,) ,
p1 p3 = 2( j ),
= 2 + j + + p1 + p2 + p3 ,
(D.18)
where we impose in (D.17) n12 = n 41 = 0, with further restrictions if p1 or p3 are zero.

Requiring (D.12) and (D.13) in both cases gives
D[p1 ,p2 ,p3 ](j,) ,
p1 p3 = 2(j ),
= j + + p2 ,
(D.19)
and we require in (D.17) n22 = n 31 = 0. If p2 = 1 then we exclude n32 = n 21 = 1 while if

p2 = 0 then we require n32 = n 21 = 0 as well. Corresponding to (D.16) we have
D[p1 ,p2 ,p3 ](j,) C[p1 2,p1 +p2 +p3 +2,p3 2](j,) .
(D.20)
477
Table 3
Diagonal representations for each in D[q,0,q]( 1 , 1 )
2
[q, 0, q]
+ 1 [q 1, 0, q 1]+1 , [q 1, 2, q 1]+1 , 3[q, 0, q]+1 , [q + 1, 0, q + 1]+1

[q 1, 0, q 1]1 , 2[q, 0, q]1 , [q + 1, 0, q + 1]1
+2
[q 2, 2, q 2]+2 , [q 1, 0, q 1]+2 , 2[q 1, 2, q 1]+2 , 4[q, 0, q]+2 , [q, 2, q]+2 , [q + 1, 0, q + 1]+2

[q 2, 0, q 2] , [q 2, 2, q 2] , 5[q 1, 0, q 1] , 2[q 1, 2, q 1] , 8[q, 0, q] , [q, 2, q] , 5[q + 1, 0, q + 1] , [q + 2, 0, q + 2]
[q, 0, q]2
+3
[q 1, 0, q 1]+3 , [q 1, 2, q 1]+3 , 3[q, 0, q]+3 , [q + 1, 0, q + 1]+3

[q 3, 2, q 3]+1 , [q 2, 0, q 2]+1 , 4[q 2, 2, q 2]+1 , 6[q 1, 0, q 1]+1 , 6[q 1, 2, q 1]+1
10[q, 0, q]+1 , 4[q, 2, q]+1 , 6[q + 1, 0, q + 1]+1 , [q + 1, 2, q + 1]+1 , [q + 2, 0, q + 2]+1
[q 1, 0, q 1]1 , [q 1, 2, q 1]1 , 3[q, 0, q]1 , [q + 1, 0, q + 1]1
+4
[q, 0, q]+4
[q 2, 0, q 2]+2 , [q 2, 2, q 2]+2 , 5[q 1, 0, q 1]+2 , 2[q 1, 2, q 1]+2 , 8[q, 0, q]+2 , [q, 2, q]+2 ,
5[q + 1, 0, q + 1]+2 , [q + 2, 0, q + 2]+2
[q 2, 2, q 2] , [q 1, 0, q 1] , 2[q 1, 2, q 1] , 4[q, 0, q] , [q, 2, q] , [q + 1, 0, q + 1]
+ 5 [q 1, 0, q 1]+3 , 2[q, 0, q]+3 , [q + 1, 0, q + 1]+3

[q 1, 0, q 1]+1 , [q 1, 2, q 1]+1 , 3[q, 0, q]+1 , [q + 1, 0, q + 1]+1
+ 6 [q, 0, q]+2
This result has essentially been used in (5.41). We may also impose (D.10) with (D.11) for
the Q charges and (D.12) and (D.13) for Q giving
E[p1 ,p2 ,p3 ](j,) ,
p1 + p3 = 2( j 1),
= 2 + j + + p1 + p2 ,
(D.21)
and we may also obtain a conjugate E[p1 ,p2 ,p3 ](j,) . Only for (D.18), where is at the
unitarity threshold, is there a unitary representation.
For relevance in Section 5 we list the self-conjugate representations, when p1 = p2 , j =
super, arising in D[q,0,q](j,j ) , obtained by the action of equal powers of the Q and Q
charges for each
For application in the text we have the decompositions of self conjugate multiplets
2+2j +p+2q
A[q,p,q](j,j ) C[q,p,q](j,j ) C[q+1,p,q](j 1/2,j ) C[q,p,q+1](j,j 1/2)

C[q+1,p,q+1](j 1/2,j 1/2) ,
2j +p
A[q,p,q](j,j ) D[q,p,q](j,j ) D[q1,p+1,q](j 1/2,j ) D[q,p+1,q1](j,j 1/2)

D[q1,p+2,q1](j 1/2,j 1/2) ,
2j +p
A[0,p,0](j,j )
D[0,p,0](j,j ) E[0,p+1,0](j 1,j ) E[0,p+1,0](j,j 1)

C[0,p+2,0](j 1,j 1) .
(D.22)
The first case represents the decomposition of a long multiplet into semi-short multiplets at
the unitarity threshold, the second plays a crucial role in Section 5 in relating the solution
of the superconformal Ward identities to the operator product expansion.
478
References
[1] E. DHoker, D.Z. Freedman, Supersymmetric gauge theories and the AdS/CFT correspondence, Lectures
given at Theoretical Advanced Study Institute in Elementary Particle Physics, TASI 2001, hep-th/0201253.
[2] S. Lee, S. Minwalla, M. Rangamani, N. Seiberg, Three-point functions of chiral operators in D = 4, N = 4
SYM at large N , Adv. Theor. Math. Phys. 2 (1998) 697, hep-th/9806074.
[3] E. DHoker, D.Z. Freedman, W. Skiba, Field theory tests for correlators in the AdS/CFT correspondence,
Phys. Rev. D 59 (1999) 045008, hep-th/9807098.
[4] K. Okuyama, L.-S. Tseng, Three-point functions in N = 4 SYM theory at one-loop, hep-th/0404190.
[5] P.S. Howe, E. Sokatchev, P.C. West, 3-point functions in n = 4 YangMills, Phys. Lett. B 444 (1998) 341,
hep-th/9808162.
[6] F. Gonzalez-Rey, I. Park, K. Schalm, A note on four-point functions of conformal operators in N = 4 superYangMills, Phys. Lett. B 448 (1999) 37, hep-th/9811155;
B. Eden, P.S. Howe, C. Schubert, E. Sokatchev, P.C. West, Four-point functions in N = 4 supersymmetric
YangMills theory at two loops, Nucl. Phys. B 557 (1999) 355, hep-th/9811172;
B. Eden, P.S. Howe, C. Schubert, E. Sokatchev, P.C. West, Simplifications of four-point functions in N = 4
supersymmetric YangMills theory at two loops, Phys. Lett. B 466 (1999) 20, hep-th/9906051;
M. Bianchi, S. Kovacs, G. Rossi, Y.S. Stanev, On the logarithmic behaviour in N = 4 SYM theory,
JHEP 9908 (1999) 020, hep-th/9906188;
M. Bianchi, S. Kovacs, G. Rossi, Y.S. Stanev, Anomalous dimensions in N = 4 SYM theory at order g 4 ,
Nucl. Phys. B 584 (2000) 216, hep-th/0003203;
B. Eden, C. Schubert, E. Sokatchev, Three loop four point correlator in N = 4 SYM, Phys. Lett. B 482
(2000) 309, hep-th/0003096.
[7] G. Arutyunov, S. Frolov, Four-point functions of lowest weight CPOs in N = 4 SYM 4 in supergravity
approximation, Phys. Rev. D 62 (2000) 064016, hep-th/0002170.
[8] G. Arutyunov, S. Penati, A. Santambrogio, E. Sokatchev, Four-point correlators of BPS operators in N = 4
SYM at order g 4 , Nucl. Phys. B 670 (2003) 103, hep-th/0305060.
[9] G. Arutyunov, F.A. Dolan, H. Osborn, E. Sokatchev, Correlation functions and massive KaluzaKlein modes
in the AdS/CFT correspondence, Nucl. Phys. B 665 (2003) 273, hep-th/0212116.
[10] G. Arutyunov, E. Sokatchev, On a large N degeneracy in N = 4 SYM and the AdS/CFT correspondence,
Nucl. Phys. B 663 (2003) 163, hep-th/0301058.
[11] E. DHoker, S.D. Mathur, A. Matusis, L. Rastelli, The operator product expansion of N = 4 SYM and the
4-point functions of supergravity, Nucl. Phys. B 589 (2000) 38, hep-th/9911222.
[12] G. Arutyunov, S. Frolov, A.C. Petkou, Operator product expansion of the lowest weight CPOs in N = 4
SYM 4 at strong coupling, Nucl. Phys. B 586 (2000) 547, hep-th/0005182;
G. Arutyunov, S. Frolov, A.C. Petkou, Nucl. Phys. B 609 (2001) 539, Erratum.
[13] G. Arutyunov, S. Frolov, A.C. Petkou, Perturbative and instanton corrections to the OPE of CPOs in N = 4
SYM 4 , Nucl. Phys. B 602 (2001) 238, hep-th/0010137;
G. Arutyunov, S. Frolov, A.C. Petkou, Nucl. Phys. B 609 (2001) 540, Erratum.
[14] B. Eden, A.C. Petkou, C. Schubert, E. Sokatchev, Partial non-renormalisation of the stress-tensor four-point
function in N = 4 SYM and AdS/CFT, Nucl. Phys. B 607 (2001) 191, hep-th/0009106.
[15] G. Arutyunov, B. Eden, A.C. Petkou, E. Sokatchev, Exceptional non-renormalization properties and OPE
analysis of chiral four-point functions in N = 4 SYM 4 , Nucl. Phys. B 620 (2002) 380, hep-th/0103230.
[16] M. Bianchi, S. Kovacs, G. Rossi, Y.S. Stanev, Properties of the Konishi multiplet in N = 4 SYM theory,
JHEP 0105 (2001) 042, hep-th/0104016.
[17] L. Hoffmann, L. Mesref, W. Rhl, Conformal partial wave analysis of AdS amplitudes for dilaton-axion
four-point functions, Nucl. Phys. B 608 (2001) 177, hep-th/0012153;
T. Leonhardt, A. Meziane, W. Rhl, Fractional BPS multi-trace fields of N = 4 SYM 4 from AdS/CFT, Phys.
Lett. B 552 (2003) 87, hep-th/0209184.
[18] F.A. Dolan, H. Osborn, Conformal four point functions and the operator product expansion, Nucl. Phys.
B 599 (2001) 459, hep-th/0011040.
[19] F.A. Dolan, H. Osborn, Superconformal symmetry, correlation functions and the operator product expansion,
Nucl. Phys. B 629 (2002) 3, hep-th/0112251.
[20] P.J. Heslop, P.S. Howe, Four-point functions in N = 4 SYM, JHEP 0301 (2003) 043, hep-th/0211252.
479
[21] V. Bargmann, I.T. Todorov, Spaces of analytic functions on a complex cone as carriers for the symmetric
tensor representations of SO(n), J. Math. Phys. 18 (1977) 1141.
[22] V.K. Dobrev, G. Mack, V.B. Petkova, S.G. Petrova, I.T. Todorov, Harmonic analysis on the n-dimensional
Lorentz group and its application to conformal quantum field theory, Springer Lecture Notes in Physics,
vol. 63, Springer-Verlag, Berlin, 1977.
[23] F.A. Dolan, L. Gallot, E. Sokatchev, On four-point functions of 12 -BPS operators in general dimensions,
hep-th/0405180.
[24] E. DHoker, D.Z. Freedman, S.D. Mathur, A. Matusis, L. Rastelli, Extremal correlators in the AdS/CFT
correspondence, in: M.A. Shifman (Ed.) The Many Faces of the Superworld, hep-th/9908160;
B. Eden, P.S. Howe, C. Schubert, E. Sokatchev, P.C. West, Extremal correlators in four-dimensional SCFT,
Phys. Lett. B 472 (2000) 323, hep-th/9910150;
B. Eden, P.S. Howe, E. Sokatchev, P.C. West, Extremal and next-to-extremal N point correlators in fourdimensional SCFT, Phys. Lett. B 494 (2000), hep-th/0004102;
M. Bianchi, S. Kovacs, Nonrenormalization of extremal correlators in N = 4 SYM theory, Phys. Lett. B 468
(1999) 102, hep-th/9910016;
J. Erdmenger, M. Prez-Victoria, Nonrenormalization of next-to-extremal correlators in N = 4 SYM and
the AdS/CFT correspondence, Phys. Rev. D 62 (2000) 045008, hep-th/9912250;
E. DHoker, J. Erdmenger, D.Z. Freedman, M. Prez-Victoria, Near extremal correlators and vanishing
supergravity couplings in AdS/CFT, Nucl. Phys. B 589 (2000) 3, hep-th/0003218.
[25] F.A. Dolan, H. Osborn, On short and semi-short representations for four-dimensional superconformal symmetry, Ann. Phys. 307 (2003) 41, hep-th/0209056.
[26] G. Arutyunov, B. Eden, E. Sokatchev, On non-renormalization and OPE in superconformal field theories,
Nucl. Phys. B 619 (2001) 359, hep-th/0105254.
[27] B. Eden, E. Sokatchev, On the OPE of 1/2 BPS short operators in N = 4 SCFT 4 , Nucl. Phys. B 618 (2001)
259, hep-th/0106249.
[28] M. Gnaydin, N. Marcus, The spectrum of the S 5 compactification of the chiral N = 2, D = 10 supergravity
and the unitary supermultiplets of U (2, 2/4), Class. Quantum Grav. 2 (1985) L11.
[29] K. Intriligator, Bonus symmetries of N = 4 super-YangMills correlation functions via AdS duality, Nucl.
Phys. B 551 (1999) 575, hep-th/9811047.
[30] P.A.M. Dirac, Wave equations in conformal space, Ann. Math. 37 (1936) 823;
P.A.M. Dirac, in: R.H. Dalitz (Ed.), The Collected Works of P.A.M. Dirac 19241948, Cambridge Univ.
Press, Cambridge, 1995.
[31] F.A. Dolan, H. Osborn, Conformal partial waves and the operator product expansion, Nucl. Phys. B 678
(2004) 491, hep-th/0309180.
[32] T. Koornwinder, Two-variable analogues of the classical orthogonal polynomials, in: R. Askey (Ed.), Theory
and Applications of Special Functions, Academic Press, New York, 1975.
[33] C.F. Dunkl, Y. Xu, Orthogonal Polynomials of Several Variables, Cambridge Univ. Press, Cambridge, 2001.
[34] L. Vretare, Formulas for elementary spherical functions and generalized Jacobi polynomials, Siam J. Math.
Anal. 15 (1984) 805.
[35] F.A. Dolan, Aspects of superconformal quantum field theory, University of Cambridge, Ph.D. Thesis, 2003.
Effective boundary field theory for a Josephson

junction chain with a weak link
Domenico Giuliano a , Pasquale Sodano b
a Dipartimento di Fisica, Universit della Calabria and I.N.F.N., Gruppo collegato di Cosenza,
Arcavacata di Rende I-87036, Cosenza, Italy

b Dipartimento di Fisica e Sezione I.N.F.N., Universit di Perugia, Via A. Pascoli I-06123, Perugia, Italy
Abstract
We show that a finite Josephson junction (JJ) chain, ending with two bulk superconductors, and
with a weak link at its center, may be regarded as a condensed matter realization of a two-boundary
sine-Gordon model. Computing the partition function yields a remarkable analytic expression for
the DC Josephson current as a function of the phase difference across the chain. We show that, in
a suitable range of the chain parameters, there is a crossover of the DC Josephson current from a
sinusoidal to a sawtooth behavior, which signals a transition from a regime where the boundary term
is an irrelevant operator to a regime where it becomes relevant.
PACS: 03.70.+k; 11.25.Hf; 74.50.+r; 74.81.Fa
Keywords: Boundary conformal field theories; Josephson junction arrays; Wire networks
1. Introduction
In this paper, we analyze a superconducting (1 + 1)-dimensional system, defined on a
finite interval of length L. If the bulk is described by a massless theory, and conformal
boundary conditions are chosen, one could understand the properties of the model, using
E-mail addresses: giuliano@fis.unical.it (D. Giuliano), pasquale.sodano@pg.infn.it (P. Sodano).
doi:10.1016/j.nuclphysb.2005.01.037
D. Giuliano, P. Sodano / Nuclear Physics B 711 [FS] (2005) 480504
481
the formalism of boundary conformal field theories [1]. If one deviates from this situation
by either adding an interaction in the bulk, or at the boundary, or both, the behavior of the
system becomes much more interesting, since it involves crossovers depending on the bulk
and boundary energy scales, as well as on the finite size L.
In the sequel, we shall show that a JJ-chain with a weak link at its center and ending with
two bulk superconductors at fixed phase difference , is the prototype of a condensed matter realization of a two-boundary sine-Gordon model [2], whose Hamiltonian is given by
1
H=
4
L
dx

g
1 2
2
L cos
(0)
+v
v t
x
2

g(L)
.
R cos
2
(1)
Boundary field theories appear to be relevant in several different contexts. In condensed

matter physics, they are mostly generalizations of quantum impurity models, which may be
described using the TomonagaLuttinger liquid (TLL)-paradigm [3]; for instance, boundary interactions appear in the analysis of the Kondo problem [4], in the study of a onedimensional conductor in presence of an impurity [5], and in the derivation of the tunneling
between edge states of a Hall bar [6]. The TLL paradigm shows that many interactions are
simply diagonalizable in the basis of appropriate collective bosonic modes, and that nondiagonalizable interactions usually correspond to exactly solvable Hamiltonians, such as
sine-Gordon models [7,8]. Recently, boundary field theories have been investigated in the
context of string theories. For instance, in studying tachyon instabilities [9], one is faced
with the fact that the space of interacting string theory [10] may be mapped onto the space
of boundary perturbations of conformal theories [11], and that the renormalization group
flow determined by boundary perturbations may be identified with tachyon condensation
[12]. Affleck and Ludwig [13] showed that the boundary entropy g is decreasing along the
renormalization group trajectories, triggered by the boundary interaction.
In an inspiring paper [14], Glazman and Larkin analyzed the quantum phase diagram of
a JJ-chain in the Vg J -plane, where Vg is an external gate voltage applied to each junction,
while J is the Josephson coupling between neighboring grains. They found evidence that
this system undergoes a phase transition between an attractive TLL phase, with g < 1, and
a repulsive TLL phase, with g > 1. While the former phase is the one-dimensional analog of the superconducting phase [15], the repulsive TomonagaLuttinger phase is peculiar
of a one-dimensional system [3]. To be self-contained, here we shall provide a detailed
field-theoretical description of the one-dimensional infinite chain analyzed in Ref. [14]:
our rather pedagogical derivation evidences how the one-dimensional JJ-chain may be described in terms of interacting (1 + 1)-dimensional chiral fermions and how, using the TLL
paradigm, the interaction is exactly diagonalized in a pertinent basis. The TLL-g parameter
[16] is crucial for the analysis of the phase diagram. Indeed, while for g < 1 the system
supports an attractive TLL phase (superconducting), for g > 1 the JJ-chain is described by
a repulsive TLL phase [14]. The g = 1-line corresponds to a noninteracting TLL model.
482
All these features may be quantitatively derived within the framework of the bosonized
(1 + 1)-dimensional TLL-model.1
Using the TLL paradigm, we show that a finite JJ-chain with a weak link at its center
is mapped onto a two-boundary sine-Gordon model, with fixed Dirichlet boundary conditions at the outer boundary, and with dynamical boundary conditions at the inner boundary.
To study the effects of the interaction at the inner boundary, we perform a renormalization
group (RG) analysis, to derive how the effective parameters of the system scale as a function of the size of the chain L. We find that in the repulsive TLL-phase (g > 1) the boundary
term is perturbative for any size L. At variance, in the attractive TLL-phase (g < 1), we
find that there is an RG-invariant length, L , such that, for L < L the boundary term is
perturbative, while for L L it becomes nonperturbative. As for the models analyzed
in [2], the crossover from the perturbative to the nonperturbative regime is evidenced by
a change of the DC Josephson current (as a function of the phase difference at the bulk
superconductors ) from a sinusoidal to a sawtooth behavior.
The paper is organized as follows:
In Section 2 we analyze the infinite one-dimensional JJ-array described in Ref. [14]
and provide a detailed derivation of the mapping of this chain onto the anisotropic
(XXZ) spin-1/2 model.
In Section 3 we construct the effective field theory for the equivalent XXZ chain. We
bosonize the theory and identify the various parameters of the continuum model in
terms of the microscopic parameters of the lattice Hamiltonian.
In Section 4, using the TLL-paradigm, we derive the phase diagram of the JJ-chain.
In Section 5 we show that the effective field theory for the JJ-chain with a weak link
and ending with two bulk superconductors is indeed the two-boundary sine-Gordon
model.
In Section 6, using the Coulomb Gas Renormalization Group approach, we provide
a careful estimate of the partition function of the two-boundary sine-Gordon model
for any value of g. We then derive the DC Josephson current across the chain, as a
function of , at both fixed points and explicitly show the existence of a crossover
from a sinusoidal to a sawtooth behavior.
Section 7 is devoted to a discussion of our results.
2. Mapping of the one-dimensional JJ-chain onto the XXZ spin-1/2 model
The simplest model Hamiltonian describing a one-dimensional JJ-chain is given by

L/a
L/a

EC
N 2
H = HC + HJ
(2)
i
J
cos(j j +1 ).
2
j
2
j =1
j =1
i j
In Eq. (2)
is the operator representing the number of Cooper pairs at site j in the
phase representation and, thus, it takes only integer eigenvalues, nj , EC is the charging
1 Notice that here the TLL-g parameter is the inverse of the parameter used in Ref. [14].
483
energy of a grain, J is the Josephson coupling energy and N accounts for the influence of
a gate voltage, since eN Vg . The sum over j ranges over the (L/a) sites, with L being
the length of the chain, and a being the intergrain distance; imposing periodic boundary
conditions amounts to fix L/a+j = j . For J /EC 0, the chain is an insulator for almost
any value of N , since it costs an energy EC , to change the number of pairs at any grain:
EC measures, then, the insulating gap.
When N = n + 1/2 + h, with integer n and h 1, the two states nj = n and nj = n + 1
become almost degenerate in energy, even for large EC ; in this limit one may restrict the
set of physical states to the Fock space F spanned by the 2L/a states

L/a
{n} =
|nj ,
nj = n, n + 1.
j =1
The Josephson coupling lifts the degeneracy between n and n + 1, since HJ may be represented as

J
HJ = eij +1 eij + eij eij +1 ,

(3)
2
with the operator eij (eij ) raising (lowering) the charge nj by 1.
Resorting to the well-known procedure [17], one may easily construct the effective
Hamiltonian Heff , describing the JJ-chain on the reduced space F [14]. Let P be the projector onto F and P be the projector onto the subspace F , to O(J 2 /EC ), Heff takes the
form:

HJ P HJ
P.
Heff = P (HJ + HC )P + P
(4)
9
16
EC
When restricted to F , the operators eij and i j may be represented with the spin-1/2
operators Sj and Sjz , as
P eij P = Sj ,

1
P = Sjz .
P i
n
j
2
(5)
From Eq. (5), one immediately sees that a charge n (n + 1)-state corresponds to a spin(1/2) (1/2)-eigenstate of Sjz , and that, to O(J 2 /EC ), Heff is given by
Heff =
L/a
L/a
L/a

J
+
3 J2 z z
Sj Sj +1 + Sj++1 Sj EC h
Sjz
Sj Sj +1 .
2
16 EC
j =1
j =1
(6)
j =1
To account for the contributions coming from intergrain capacitances, it is sufficient

to retain only the nearest-neighbor terms [14], since next-to-nearest neighbor hopping
terms would give rise to irrelevant operators, and to add to the Hamiltonian (6) the term
L/a
EZ j =1 Sjz Sjz+1 (EZ > 0) [14]. Thus, the system is usefully described [14] by
L/a
L/a
L/a

J
+
+

z
Heff =
Sj Sj +1 + Sj +1 Sj H
Sj +
Sjz Sjz+1 ,
2
j =1
j =1
j =1
(7)
484

2
3 J
with H = EC h and = EZ 16
EC .
Eq. (7) is the Hamiltonian for a spin-1/2 XXZ-chain in an external magnetic field H .
The anisotropy parameter may take positive, as well as negative values, depending on
the constructive parameters of the JJ-chain. As elucidated in the following sections, the
sign of is crucial for the emergence of a repulsive TomonagaLuttinger (RTL) phase in
a JJ-chain.
3. Continuum field theory and bosonization of the XXZ-chain

Using the standard bosonization technique [16], we derive in this section the effective
low-energy long-wavelength field theory associated to the Hamiltonian (7). For this purpose, one starts to write the spin operators Sja in terms of JordanWigner (JW) spinless
lattice fermions aj [18], obeying standard anticommutation relations

aj , ak = j k .
(8)
The JW transformation amounts to define
j 1

1
Sj+ aj exp i
al al ,
Sjz aj aj ,
2
l=1

j
1
Sj exp i
al al aj ,
(9)
l=1
which, in turn, implies
a b
Sj , Si = ij i abc Sjc .
(10)
From Eqs. (9), the fermionic effective Hamiltonian may be written as

f
Heff HK + HP + H
L/a

aj aj
j =1

L/a
L/a

J

1
1
aj +1 aj +1
aj aj
aj aj +1 + aj +1 aj +
=
2
2
2
j =1
+H
L/a

j =1
aj aj .
(11)
j =1
The hopping term HK is readily diagonalized by resorting to the Fourier components of

aj , ak ,
aj =
1
ak eik(j a) ,
L/a k
(12)
leading to
HK =
(k)ak ak

(k) = J cos(ka) H .
485
(13)
If |H | < J , the Fermi surface is disconnected and consists of two isolated points at kF ,
with kF = arccos(H /J ). Keeping only the excitations about the Fermi points with momenta k such that |k kF | , one obtains

HK
(k)ak ak +
(k)ak ak
|kkF |
J sin(akF )
|k+kF |
sin(pa)aL (p)aL (p)
|p|
J sin(akF )
sin(pa)aR (p)aR (p),
(14)
|p|
with
aL (p) ap+kF ,
aR (p) apkF

|p| .
For kF , one may define the continuum chiral fields L/R (x) as
aj
eikF xj L (xj ) + eikF xj R (xj ),
2a
with xj = j a; one gets, then

p aL (p)aL (p) aR (p)aR (p)
HK J sin(kF a)
(15)
|p|
L
= ivF

dL (x)
dR (x)
dx L (x)
R (x)
dx
dx
(16)
where the Fermi velocity is given by vF = 2J sin(akF ).

Eq. (16) is, of course, the effective low-energy theory of the hopping Hamiltonian HK ;
the cutoff will be specified later.
The dynamics of the fermionic fields L and R in the Heisenberg representation, is
described by
1 ip(xvF t)
e
L (p),
L (x, t) = L (x vF t) =
L p
1 ip(x+vF t)
R (x, t) = R (x + vF t) =
(17)
e
R (p),
L p
and the equal time anticommutation relations are given by

L (p), L (p ) = p,p ,
R (p), R (p ) = p,p ,

R (p), L (p ) = 0.
(18)
486
Since L (p) with p > 0 creates positive energy left-handed states, while R (p) creates
positive energy right-handed states if p < 0, the Fermi sea fermionic ground state is
defined as

|FS =
(19)
L (p)R (p) |0 L (p)|0 = R (p)|0 = 0 .
p<0
Thus, by choosing = 1/(4a), one gets

FS|2a e2ikF xj L (xj )R (xj ) + e2ikF xj R (xj )L (xj ) |FS = 0
(20)
and

1
FS|a L (xj )L (xj ) + R (xj )R (xj ) |FS = ,
2
(21)
thus, Sjz is normal ordered respect to |FS, i.e.,

Sjz = 2a :L (xj )L (xj ): + :R (xj )R (xj ):

+ 2a :L (xj )R (xj ):e2ikF xj + :R (xj )L (xj ):e2ikF xj ,
(22)
where :: denotes normal ordering.

Using fermionic coordinates, one should now evaluate the IsingNel interaction HP as
HP
N

j =1
Sjz Sjz+1

4 2 a
L/a

j =1
L
dxj
aj aj

aj+1 aj +1

:L (xj )L (xj ): + :R (xj )R (xj ):

L (xj )R (xj ) + e2ikF xj R (xj )L (xj )
+e
:L (xj +1 )L (xj +1 ): + :R (xj +1 )R (xj +1 ):

2ikF xj

+ e2ikF xj +1 L (xj +1 )R (xj +1 ) + e2ikF xj +1 R (xj +1 )L (xj +1 )
(1)
(2)
= HP + HP ,
(23)
where
(1)
HP

= 4 2 a
L
dxj :L (xj )L (xj )::L (xj +1 )L (xj +1 ):
+ :R (xj )R (xj )::R (xj +1 )R (xj +1 ):

+ :L (xj )L (xj )::R (xj +1 )R (xj +1 ):

+ :R (xj )R (xj )::L (xj +1 )L (xj +1 ): ,
(24)
487
and
(2)
HP

= 4 2 a
L

dxj L (xj )R (xj )e2ikF xj + R (xj )L (xj )e2ikF xj
L (xj +1 )R (xj +1 )e2ikF xj +1 + R (xj +1 )L (xj +1 )e2ikF xj +1 .

(25)
(1)
While evaluating HP is rather straightforward, since it contains only normal-ordered
fermionic left- and right-densities, evaluating HP(2) is a little bit more involved, due to
crossed LR-interaction. In fact, at any kF , momentum conservation selects the perti-
nent contribution to Eq. (25), given by

(2)
HP

= 4 2 a
L
dxj e2ikF a L (xj )R (xj )R (xj +1 )L (xj +1 )
0

2ikF a
R (xj )L (xj )L (xj +1 )R (xj +1 ) ,
+e
(26)
where a possible Umklapp contribution, arising when kF a /2, has been neglected.2
(2)
To normal order HP , one may rewrite it as

(2)
HP = 4 2 a e2ikF a
L

dx :L (x)L (x + a): +
i
2a

i
2a
L

4 2 a e2ikF a dx :L (x + a)L (x):
:R (x + a)R (x): +
i
2a
:R (x)R (x

i
,
+ a):
2a
(27)
which, for a 0, becomes

(2)
HP

= 2 4 2 a cos(2kF a)
L
dx :L (x)L (x)::R (x)R (x):
0
L
+ 4 sin(2kF a)

dx :L (x)L (x): + :R (x)R (x):
L
i4a cos(kF a)

dL (x)
dR (x)
R (x)
.
dx L (x)
dx
dx
2 This is the case, for instance, of the half-filled fermionic sea in zero chemical potential.
(28)
488
The various terms in Eq. (28) may be interpreted as follows:

A shift in the chemical potential
L
4 sin(2kF a)

dx :L (x)L (x): + :R (x)R (x): ,
which is accounted for by simply redefining kF through the equation

(aJ ) cos(kF a) + 2 sin(2kF a) = H.
(29)
An LR-interaction term

2 4 2 a cos(2kF a)
L
dx :L (x)L (x)::R (x)R (x):,
0
that adds up to a similar term coming from HP(1) , giving

2 4 2 a 1 cos(2kF a)
L
dx :L (x)L (x)::R (x)R (x):.
(30)
A renormalization of the Fermi velocity given by

L
i4a cos(kF a)

dL (x)
dR (x)
R (x)
.
dx L (x)
dx
dx
(31)
0
f
Using the well-known bosonization rules (A.9), the fermionic Hamiltonian Heff may be
written in bosonic coordinates as
v F + g2
H =
4
L
dx
L
x
2
R
+
x
2
g4
+2
4
L

L R
,
dx
x x
(32)
where g2 = g4 = 4(a)[1 cos(2kF a)].

One may readily see that H b corresponds to the Hamiltonian for a free, massless, real
bosonic field in 1 + 1 dimensions, which is described by the Hamiltonian
v
H [, ] =
4
L
dx

4 2 2
2
+g
g
x
where the momentum conjugate to is = (2/g)

t .
g
g
Upon defining two independent chiral fields, L and R , as

g
L (x vt)
1 2

= + g
,
x
g
x
2
(33)

g
R (x + vt)
1
2

= + g
,
x
g
x
2
489
(34)
one immediately sees that
g g
v
H [, ] H L , R =
4
L
dx
L
x
2
R
+
x
2
,
(35)
0
g=1
which, when expressed in terms of L

v=

(vF + g2
)2
g42 ,
g=
g=1
and R , yields Eq. (32), provided that

v F + g2 + g4
.
v F + g2 g4
(36)
Thus, the correlation functions of all the operators depending on L and R may be evaluated by the replacements

1
g

g
g
g
L + R ,
L R = g L R ,
(37)
L + R =
g
g
with L , R free, chiral bosonic fields.
4. Phase diagram of the JJ-chain

In Ref. [14], it has been evidenced that the phases allowed to a JJ-chain are:
(1)
(2)
(3)
(4)
A Mott insulating (MI) phase;

A band insulating (BI) phase;
A repulsive TomonagaLuttinger phase (RTL);
A superconducting (S), attractive TomonagaLuttinger phase.
Here, we shall determine the range of the JJ-chain parameters associated to each allowed
phase in the Vg J -plane and, using the TLL approach, we shall provide a careful derivation
of the phase boundaries; of course, our results crucially depend on the approximations
made in Section 2. Our subsequent analysis is based on the bosonic, low-energy effective
Hamiltonian given in Eq. (32).
To analyze the onset of the MI phase, one has to include also the Umklapp term HPu in
Eq. (25), given by
L
HPu
(a)
dxj L (xj )L (xj +1 )R (xj )R (xj +1 )

+ R (xj )R (xj +1 )L (xj )L (xj +1 ) ,
(38)
490
whose bosonized version yields

HPu
L

gU
= 2 2
dx :cos 2 2(x) :,
L
4

2a g
gU = a
.
L
(39)
Eqs. (39) and (33) yield the Hamiltonian for a (1 + 1)-dimensional sine-Gordon model,
whose phase structure, as a function of the parameters g and gU has been extensively studied [19]. There are two distinct regimes: if g < 2, the interaction is irrelevant and the theory
is perturbative in gU , while, if g > 2, the interaction is relevant. In the thermodynamic limit
(L ) the system flows towards a strongly-coupled regime, where the Umklapp interaction is responsible for the creation of a gap in the excitation spectrum and for the onset
of long range IsingNel order [7]. In the language of the JJ-chain, this corresponds to a
checkboard charge ordered state with the charge at each grain either equal to n, or to n + 1:
this is the MI phase.
The MI charge-ordered region in the Vg J -plane may be identified with the condition
g > 2, which reads

3
3 J2
> J.
4 sin(akF ) EZ
(40)
16 EC
2
As J = 0, Eq. (40) is satisfied by any value of kF (provided that there is a real solution
to Eq. (29)). When there is no real solutions to Eq. (29), that is, for large enough |H |, the
chain undergoes a phase transition towards the BI phase. This shows that, for J = 0, the
only possible phases are either the BI phase, or the MI charge-ordered phase.
To see how the transition towards the BI phase extends for J > 0, one may start again
from Eq. (29), describing the boundary of the BI phase. If H > 0, there are no real solutions
of Eq. (29) for
H 2 > J
H 2EZ J +
3 J2
> 0.
8 EC
(41)
As H < 0, on the other hand, there are no real solutions if

H + 2 < J
H + 2EZ + J
3 J2
< 0.
8 EC
(42)
Eqs. (41), (42) define two regions in the phase diagram corresponding to BI phases, since
the density of charge-carrying states at the Fermi surface is 0.
Furthermore, as J > 0, Eq. (40) admits real solutions only if

3
16
8J
< 1 J < EC2 + EZ EC EC ,
(43)
2
3
(EZ 3 J )
16 EC

which implies that the MI phase closes when J = EC2 + 16
3 EZ EC EC . Since changes

sign for J = J = 16
3 EZ EC , one finds that, as the MI phase closes, the Tomonaga
Luttinger liquid interaction is still repulsive (that is, g > 1).
491
Fig. 1. Sketch of the phase diagram of the JJ-chain in the Vg J -plane derived within TLL-approach, as discussed
in Section 4.
The phase where, instead, the TomonagaLuttinger liquid is attractive (which is a necessary condition, to achieve superconducting correlations in the 1-d system) takes place for
J > J , that is, for g < 1.
In Fig. 1 we plot the phase diagram obtained using the TLL-approach. We observe
that, due to the renormalization of the Fermi velocity, the line corresponding to g = 1 is
a straight horizontal line: thus, as long as the TLL-description of the JJ-chain holds, one
cannot push the system across this line by acting on the gate voltage Vg . We expect that
this behavior is a byproduct of the approximations introduced in Section 2; higher order
contributions to Heff should strongly modify the line corresponding to g = 1.
5. Two-boundary sine-Gordon-model description of a finite JJ-chain

In the following, we shall consider a one-dimensional JJ-chain with a weak link (i.e.,
a junction with a different nominal value of the Josephson coupling, EW ) at its center,
whose position is set at x = 0, and ending with two bulk superconductors, whose phase
difference is held fixed at (i.e., R = L = /2). Using the bosonization method, in
this section we show that this finite JJ-chain is pertinently described by a two-boundary
sine-Gordon model [2].
Upon introducing JW fermions on both sides of the weak link, one gets
L/a1
+
SL/a,>
= ei l=1 al al aL/a
= ei/2 ,
L/a1
+
SL/a,<
= ei l=1 al al aL/a
= ei/2 ,
(44)
where the labels > (< ) refer to observables at the right (left)-hand side of the weak link.
492
Using the long wavelength approximation, the fermionic string in the exponential of
Eqs. (44), is easily evaluated as
L/a1

al al =
l=1
L
+
2a
La
dxl :L,>
(xl )L,> (xl ): + :R,>
(xl )R,> (xl ):

L 1
+ L,> (L a) + R,> (L a) ,
=
2a
2
1

2a 2 3i L,> (L) i R,> (L)
+
SL/a,>
:e 2
::e 2
:
L
1

L 2 i L,> (L) i R,> (L)
+
:e 2
::e 2
:,
2a
(45)
(46)
and, by keeping only the leading contributions to Eq. (46) in the cutoff a, as a 0, one
gets, for x > 0,

1
i
L 2 i L,> (L) i R,> (L)
+
:e 2
::e 2
: = e 2 [L,> (L)R,> (L)] .
SL/a,>
(47)
2a
Similarly, for x < 0, one obtains
+
= e 2 [L,< (L)R,< (L)] .
SL/a,<
i
(48)
The boundary condition at x = 0 is, instead, dynamical, since it depends on the strength
of the weak link, EW . In terms of the spin variables, the weak link interaction may be
represented as a pointwise contact Hamiltonian given by
HW =
EW
+
+

S0,< S0,> + S0,>
.
S0,<
2
(49)
Taking into account that S0z = 12 [a0 a0 + a0 a0 ] = La [:L (0)L (0) + R (0)R (0):] and the
requirement that S0+ S0 S0 S0+ = 2S0z , for any value of g, the operators S0+ and S0 are
realized as

g
a 2a 2 i [L (0)R (0)]
+
S0 =
:e 2
:,
L L

g
a 2a 2 i [L (0)R (0)]
S0 =
(50)
:e 2
:.
L L
From Eqs. (50), the dependence of HW on the bosonic coordinates is given by

i
aEW 2a g
:exp
L,< (0) + R,< (0) L,> (0) + R,> (0) :
HW =
2L
L
2

+ h.c. .
(51)
493
Using Eq. (37), one immediately sees that the boundary interaction Hamiltonian at x = 0
takes the form

g

aEW 2a g
i g [L,+ (0)R,+ (0)]
:e 2
: + :ei 2 [L,+ (0)R,+ (0)] : ,
HW =
(52)
2L
L
where

1 2
L,+ (x vt)
+
= + + g
,
x
g
x
2

R,+ (x + vt)
2
1
+
,
= + + g
x
g
x
2
with (x, t) = 1 [> (x, t) < (x, t)].

2
The boundary conditions at the bulk superconductors may be written as

g
g
g L,+ (L, t) R,+ (L, t) = (mod 2k).
(53)
(54)
By inspection of Eq. (52), one sees that the field (x, t) fully decouples from the weak
link dynamics. Furthermore, its boundary condition is (L, t) = 0 t, thus implying that
is insensitive to variations in the phase difference between the bulk superconductors.
As a result, the field does not contribute to the dynamics of the JJ-network. Using
Eq. (52), one gets that the pertinent effective Hamiltonian HJJ is given by
v
HJJ =
4
L
dx
0
aEW
L
L
x
2a
L
2
g

+
R
x
2

g
L (0) R (0) :,
:cos
2
with (L,+ , R,+ ) (L , R ), while the pertinent boundary condition is given by

g L (L, t) R (L, t) = (mod 2k).
(55)
(56)
The model described by HJJ , supplemented with the boundary condition in Eq. (56), coincides with the two-boundary sine-Gordon Hamiltonian introduced in Eq. (1), provided
in Eq. (1) is identified with (L R )/ 2, R is sent to , and L is identified

g
with aELW ( 2a
L ) . Intuitively speaking, while the boundary condition at x = L is always
Dirichlet-like, at x = 0, since EW is finite, the boundary condition is dynamical, and is
given by

g
aEW 2a g
L (0, t) R (0, t)
g sin
L
L
2

L (0, t) R (0, t) = 0.
+v
(57)
x
In particular, for small EW , Eq. (57) provides Neumann-like boundary conditions for L
R at x = 0, while it provides Dirichlet-like boundary conditions for large values of EW .
494
For g < 1, the boundary term is a relevant operator and one may use the renormalization
group to describe how the renormalization of E W affects the ground-state energy: as we
shall evidence in Section 6, there is a renormalization group invariant length scale L
such that, for L L the JJ-chain behaves nonperturbatively in E W while, for L < L , it
behaves perturbatively. At variance, for g > 1, the JJ-chain is always perturbative in E W ,
since the boundary term is now an irrelevant operator. For g = 1, the boundary term is
marginal and the bulk system is fully described by a pair of noninteracting chiral fermions
and the partition function may be computed exactly [2].
To summarize, for g > 1 E W always flows to 0, while, for g < 1, there is a characteristic
1
healing length L = L(J /(E W ( = 1))) 1g , separating a small-E W perturbative regime
from the nonperturbative one E W 1. Similar features are exhibited by a superconducting
loop closed by a Josephson junction of strength EJ , when EJ is regarded as an effective
coupling strength [20].
In the next section, we shall prove that the behavior of the DC Josephson current as a
function of depends crucially on whether E W flows to zero, or to large values. Namely,
we shall find that, when E W flows to zero, the DC Josephson current has a sinusoidal
behavior, while, when E W 1, one gets the sawtooth behavior.
6. Josephson current across the JJ-network with a weak link

In this section, we shall compute the functional dependence of the DC Josephson current on the phase difference between the bulk superconductors, by evaluating the zero
temperature canonical partition function Z[EW ], from which the Josephson current may
be evaluated as
2e EJJ []
I () =
(58)
c
with

Z[EW ]
1
ln
.

Z[0]
EJJ [] = lim
(59)
The partition function of the two-boundary sine-Gordon model has been exactly computed for particular values of g [2]; due to our interest in providing an estimate of the
Josephson current for any value of g, we resort to an approximate computation based on
the Coulomb Gas Renormalization Group scheme [21], described in detail in Appendix B.
Our analysis shows that, while for Neumann boundary conditions (EW 0), one gets
I () sin(), for Dirichlet boundary conditions ((EW /J ) 1) one gets a sawtooth dependence of I () on .
Case A: EW 0
As EW 0, Eq. (57) provides Neumann-like boundary condition at x = 0,

L (0 vt) R (0 + vt) = 0,
x
(60)
495
together with the Dirichlet-like boundary condition at x = L
L (L vt) R (L + vt) = .
g
Using the mode expansion given in Eq. (A.5), one sees that Eq. (60) is satisfied if
q L qR = ,
g
L (n)e
ikn L
pL = pR p,
= R (n)e
ikn L

(n)

1
2
n+
, nZ
kn =
L
4
(61)
4 eikn vt
L (0 vt) R (0 + vt) = +
(n) (t).
g
L
kn
(62)
n=0
The relevant vertex operators are given by

:ei
g(t)
g+ (t) i g (t)
: = ei ei
(63)
with
+ (t) =
4 eikn vt
(n),
L
kn
(t) =
n<0
4 eikn vt
(n).
L
kn
(64)
n>0
The partition function is given by

Z[EW ] = Z[EW = 0] T
aEW
exp
L
2a
L
g

g
( ) :
,
d :cos
2
(65)
where = it, T is the (imaginary) time-ordered product, and the brackets . . . mean that
expectation values should be computed with respect to the ground state of the Hamiltonian
in Eq. (35).
Using the expansion of Eq. (65) in a power series of EW as
Z[EW ] = Z[EW = 0]

j

(E W )j
j =0
with E W =
aEW 2a g
L ( L ) ,
2j j !

dk T
j =1 0 k=1

:e
ik (k )
(66)
k=1
and using the identity
2v(1 2 )
[1 2 ] (1 ), + (2 ) = 4 ln 1 e L
,
one gets
(67)
496

1a/v

Z[EW ]
(E W )j
=
d1
d2
Z[EW = 0]
2j
j =0
j =1 0
j 1
a/v
dj e
i g
k=1 k
2j

2gu r
2v
1 e L |u r |
,
(68)
u<r=1
since Wicks theorem implies3

T
2j

:e

2 k (k )
=e
i g
k=1 k
k=1
2j

2gu r
2v
1 e L |u r |
.
(69)
u<r=1
To analyze the short-distance divergences in Eq. (68), one has to rescale the cut2j
2v
off a a/, > 1, in order to approximate u<r=1 [1 e L |u r | ]2gu r with
2j
2v
2gu r . As a result, one obtains
u<r=1 | L (u r )|
j 1a/v
1 a/v
d2
d1
0
i g
2gu r
2j

2v

(
)
r
L u
k=1 k
u<r=1
u r
u=r
2j

1a/v
j 1
a/ v
d2
d1
0
dj e

2gu r
2v

.
L (u r )
dj e
i g
k=1 k
(70)
u<r=1
At a given order 2j , and for , the most diverging contributions come from integrals
containing an equal number of positive and negative s. Thus, -scaling of the integrals
appearing in Eq. (70) is taken into account by means of a multiplicative renormalization of
the effective coupling strength
E W E W () = 1g E W ( = 1).
(71)
Eq. (71) implies that the boundary interaction at x = 0 is irrelevant and the Neumann fixed
point is always stable for g > 1 (i.e., in the repulsive TomonagaLuttinger phase). The
RTL phase is always associated to a stable Neumann fixed point.
To evaluate I (), one may retain only the first order terms in the EW -expansion in
Eq. (66), getting
3 In Eq. (68), the cutoff a has been introduced to regularize possible short-distance divergences in the argument
of the integral. It should be identified with the lattice step introduced in Eq. (2).
497

g
g

E W
Z[EW ]
1
d ei :ei 2 ( ) : + ei :ei 2 ( ) :
Z[0]
2
0
g
(aEW )( 2a
L ) cos()
(72)
from which the network energy is derived as

1 Z[EW ]
2a g
EJJ [] = lim
cos().
ln
= (aEW )

Z[0]
L
(73)
Using Eq. (58) one gets I () sin().

It is comforting to see that Eq. (73) reproduces the pertinent renormalization of the
effective coupling constant given in [14].
Case B: (EW /J ) 1
When analyzing the case in which the effective coupling grows, as the size of the system
goes large, we must consider that our analysis started from expanding fermionic fields
whose band energy is equal to J . Accordingly, the scaling should stop as E W J . The
scale at which this happens, , is found by the condition
E W ( = 1)( )1g = J.
This implies that scaling stops as the size of the system becomes of order of
1

1g
J
L = L
.
E W ( = 1)
(74)
L ,
given by
(75)
For L < L , the theory is still perturbative. As L L , instead, the system enters the
nonperturbative region. In this limit, the field L R has to satisfy Dirichlet-like boundary
conditions both at x = L and at x = 0. From Eq. (57), one gets

1
L (0 vt) R (0 + vt) =
L (0 vt) R (0 + vt) ,
sin
2
x
g EW v
(76)
thus, the Dirichlet-like boundary condition at x = 0 is
L (0 vt) R (0 + vt) = 0.
(77)
Using the mode expansion given in Eq. (A.5), Eq. (77) may be easily satisfied by setting
qL = qR q,
pL = pR p,
+ R (n)eikn L = 0,
(78)
2n
, n Z, g4p = 2k + ,
L
L (n) = R (n) = (n).
(79)
L (n)e
ikn L
provided that
kn =
498
The partition function is now given by

2v
v 2
2
p + pR
L (n)L (n) + R (n)R (n) ,
Z = Tr exp
L L
L
n>0
(80)
with the pertinent boundary conditions given by Eq. (79).
The trace in Eq. (80) may be factorized into a contribution from the oscillatory modes,
and a contribution from the zero modes, so that
Z = Zosc Z0-modes ,

4v
L )n ]]1 , and
with Zosc = [
n=1 [1 (e
Z0-modes =

k=

2
v
.
k
exp
2gL
2
(81)
(82)
From Eqs. (81), (82), one gets I () , for ; this yields the well-known
sawtooth behavior [20].
The switch to this behavior from the sinusoidal behavior obtained for EW 0 signals
the crossover from a perturbative to a nonperturbative regime of the chain. It should be
observed that, for g > 1, the DC Josephson current has always a sinusoidal dependence on
the phase difference between the bulk superconductors, since, in this region, the boundary
term is an irrelevant operator and, thus, the chains behavior is always bulk-dominated.
7. Concluding remarks
Our analysis shows how a two-boundary sine-Gordon model emerges as a pertinent
effective description of a finite JJ-chain. Since our approach heavily relies on the bosonization method, we expect that boundary sine-Gordon models may be useful where the
TomonagaLuttinger liquid paradigm is relevant. For instance, a magnetic spin system
with a pertinent impurity at its center and with the spins at its extrema held fixed, may
support a spin current across the chain, with different behaviors, depending on the boundary conditions around the impurity. Similarly, one may envisage other applications of our
results to quantum wires [22] and carbon nanotubes [23].
According to the g-theorem [13], the boundary entropy of the chain should always
decrease, as one gets towards the thermodynamic limit. Thus, for L L , one has

SD SN = lim lim [ln ZD ln ZN ] ,
(83)
L
where ZD/N is the partition function computed with Dirichlet/Neumann boundary conditions, respectively. Eq. (83) yields a nonvanishing result only because of the contribution
of the zero modes; namely, from Eq. (82), one gets that

SD SN = ln g .
(84)
499
Remarkably, the entropy variation depends on the sign of ln g. Thus, as g > 1 (i.e., within
the RTLL phase) the Dirichlet boundary entropy is higher than the Neumann boundary entropy and then, in the thermodynamic limit the system flows from the Dirichlet to the
Neumann fixed point. Conversely, as g < 1 (i.e., within the superconducting region) the
Neumann fixed point carries an entropy that is higher than the one associated to the Dirichlet fixed point. Accordingly, the flow now goes from the Neumann to the Dirichlet fixed
point.
The evaluation of the Josephson current from the partition function of the two-boundary
sine-Gordon model explicitly shows a crossover from a perturbative regime (E W 0), in
which the current is a sinusoidal function of the phase difference at the boundary, , to a
nonperturbative regime (E W /J 1), where it exhibits a sawtooth functional dependence
on .
There is a striking, and yet intuitive, similarity between the finite JJ-chain investigated
in this paper, and an rf-SQUID in an external magnetic field [20]. For the latter system,
a variation of the flux threaded by a superconducting loop operates the crossover in the
behavior of the Josephson current, as a function of the applied flux. This might suggest that
other very interesting condensed matter realization of boundary sine-Gordon models, may
be provided by superconducting loops, interrupted by two, or more, Josephson junctions
[24]. Remarkably, using the results in Section 6, one immediately sees that the effective
potential of the finite JJ-chain, as a function of the phase difference at the boundary of the
weak link, i.e., = L (0) R (0), exhibits only one minimum within the RTLL phase
(i.e., = ), since E W 0, while it is a two-level quantum system in the superconducting
region (E W /J 1), provided that .
Acknowledgements
We thank G. Zemba for actively participating to our research efforts at the very early
stages of this work. We acknowledge useful discussions with G. Delfino, A. de Martino,
G. Grignani, H. Saleur, G.W. Semenoff and insightful correspondence with L. Glazman.
The work has been partly supported by the M.I.U.R. national project Josephson Networks
for Quantum Coherence and Information (grant No. 2004027555).
Appendix A. Bosonization rules

In this appendix the bosonization rules, used in this paper, are reviewed.
It is a peculiar property of (1 + 1)-dimensional theories that it is possible to realize
chiral fermionic fields in terms of chiral bosonic fields, and vice versa. If one starts, for
instance, from the chiral components of a free, massless, KleinGordon field in 1 + 1
dimensions, the equation of motion for is
2
2
2
(A.1)
(x, t) = 0,
v
t 2
x 2
500
where periodic boundary conditions are assumed. may be written, as the sum of two
chiral fields as

1
(x, t) = L (x vt) + R (x + vt)

(A.2)
2
with L and R chiral FubiniVeneziano fields [25], whose mode-expansion is given by
L (x, t) = L (x vt) = qL
2
2i eikn (xvt)
pL (x vF t) +
L (n)
L
L
kn
(A.3)
2
2i eikn (x+vt)
pR (x + vt) +
R (n).
L
L
kn
(A.4)
kn
and
R (x, t) = R (x + vt) = qR +
kn
The basic commutation rules are

[qL , pL ] = [qR , pR ] = i,
L (n), L (m) = nn+m,0 ,

R (n), R (m) = nn+m,0 ,
(A.5)
and {kn } is a (discrete) set of nonzero modes depending on the boundary conditions
imposed on the bosonic fields (for instance, for periodic boundary conditions, one gets
kn = 2n
L , n Z).
Due to the commutation rules in Eq. (A.5), the bosonic vacuum |Bos is defined by
pL |Bos = pR |Bos = 0,
L (n)|Bos = R (n)|Bos = 0 (n > 0),
(A.6)
and one may then define a creation and an annihilation part for each field, i.e.,
L+ (x) = qL +
2i eikn x
L (n),
L
kn
kn <0
L (x) =
2
2i eikn x
xpL +
L (n)
L
L
kn
(A.7)
kn >0
and
R+ (x) = qR +
2i eikn x
R (n),
L
kn
kn >0
R (x) =
2
2i eikn x
R (n).
xpR +
L
L
kn
(A.8)
kn <0
1 L
From the commutators given in Eqs. (A.5), one sees that the modes of the operator 2
x
obey the same algebra as the modes of the fermionic density operator, normal ordered with
respect to |FS. Thus, the fermionic bilinear density operator :L L : may be identified
1 L
with the bosonic density operator, 2
x , provided that |Bos is identified with |FS. The
501
same identification may be carried for the R-modes. Therefore, one gets a first bosonization
rule
1 L (x vt)
,
2
x
1 R (x + vt)
.
:R (x + vt)R (x + vt):
(A.9)
2
x
The second rule is obtained if one identifies the chiral fermionic fields with normal ordered
vertex operators of bosonic fields, defined by
:L (x vt)L (x vt):
:eiL/R (xvt) : = ei[L/R (xvt)+L/R (xvt)] e

=e
2 +
2 [L/R (xvt),L/R (xvt)]
i[L/R
(xvt)+L/R
(xvt)] iL (xvt)
2
L
2a
2
2
(A.10)
The correspondence rules are now given by

1
L (x vt) :eiL (xvt) :,
L
1
R (x + vt) :eiR (x+vt) :.
L
(A.11)
To check the consistency of Eq. (A.11), one has to consider the braiding rule
:eiL/R (x) ::eiL/R (y) : = ei :eiL/R (y) ::eiL/R (x) :,
(A.12)
and the vertexvertex correlators

Bos|:eiL/R (xvt) ::eiL/R (x vt ) :|Bos

2
i
1
=
.

2 sin[ 2
L (x x v(t t ))]
(A.13)
From Eqs. (A.12), (A.13), one derives the basic anticommutators

L (x vt), L (x vt ) = x x v(t t )
and

R (x + vt), R (x + vt ) = x x + v(t t ) .
(A.14)
The other correlators used in the paper are derived from Eq. (A.13) and from Wicks theorem applied to normal ordered vertices [25].
Appendix B. Renormalization group equations for the JJ-chain with a weak link
In this appendix, the flow of E W is derived within the Coulomb Gas Renormalization
Group approach.
To analyze the renormalization group flow for the JJ-chain with a weak link, one needs
to observe that, at order 2j , the short-distance most diverging contribution to the partition
502
function is given by
1a/v

2jg
2j 2a
(EW )
d1
d2
L
0
2j 1
a/v
d2j
2j

2gu r
2v
1 e L |u r |
.
(B.1)
1 ++2j =0 u<r=1
Rescaling the short-distance cutoff as a a/ and sending implies the following

renormalization group scaling equation for E W :
d E W ()
= (1 g)E W ().
d ln
(B.2)
1
From Eq. (B.2), one may easily identify the cutoff scale = (J /(EW ( = 1))) 1g , at
which E W becomes J . It means that a system of size L, with a weak link of nominal
strength EW = EW ( = 1), crosses over towards the strongly coupled regime, as its size
is increased to L = L [14].
In addition, it has to be noticed that, to remove the cutoff, one needs a further renormalization, due to one-dimensional charge annihilation processes. This may be evidenced,
for instance, by applying AndersonYuvalHamann analysis of a one-dimensional instanton gas [21].
As the cutoff is rescaled from a to a/, two charges, of opposite sign, may annihilate
with each other, if they were originally separated by a distance between a/(v) and a/v.
As a result, the integral at order 2j + 2 should provide an extra contribution to the integral
at order 2j , which we are now going to calculate.
Upon defining
T=
+ +
,
2
= + ,
(B.3)
where + being the coordinate of the +1-charge, and the coordinate of the 1-charge,
the extra contribution arising to order 2j , is given by

a/v
(E W )
2j +2
a/(v)
2j
dT +

exp 2g
k=1
d2j
0

dT +
2j 2 a/v
2j

d2
d1
2j 1
a/v
2j 1
a/v
a/L
1
1
2g
[ 2v
L ]
2v
|T
|
k
k
ln 1 e L
T
2j

2gu r
2v
1 e L |u r |
u<r=1

a/v
+ (E W )
2j +2
a/(v)
dT +
d2j
0

dT +
2j 3 a/v
2j

d2
0
2j 2
a/v
2j a/v
2j 1
a/v
2j 1
a/v
a/L
1
d1
503
2gu r
2v
1 e L |u r |
2g
[ 2v
L ] u<r=1
exp 2g
2j

k=1

2v

ln 1 e L |T k | .
k
T
(B.4)
If one expands the exponentials as

2j

2v

exp 2g
ln 1 e L |T k |
k
T
k=1
1 2g
2j

k=1

2v

ln 1 e L |T k | ,
T
one may derive the renormalization group equation for g.4 The result is [21]
g

L
g g + dg = g
g[E W ]2 d ln .
2va
(B.5)
(B.6)
As expected [19], the wavefunction renormalization is needed only for g 1.

This completes the renormalization scheme derived within the perturbative approach
for the system in the presence of a weak link.
References
[1] J. Cardy, Conformal invariance and surface critical behavior, in: C. Itzykson, H. Saleur, J.B. Zuber (Eds.),
Conformal Invariance and Applications to Statistical Mechanics, World Scientific, 1988;
A. Cappelli, G. DAppollonio, M. Zabzine, J. High Energy Phys. 0404 (2004) 010.
[2] A. LeClair, G. Mussardo, H. Saleur, S. Shorik, Nucl. Phys. B 453 (1995) 581;
J.-S. Caux, H. Saleur, F. Siano, Nucl. Phys. B 672 (2003) 411;
J.-S. Caux, H. Saleur, F. Siano, Phys. Rev. Lett. 88 (2002) 106402.
[3] J.M. Luttinger, J. Math. Phys. 4 (1963) 1154;
S. Tomonaga, Prog. Theor. Phys. 5 (1950) 544.
[4] I. Affleck, Nucl. Phys. B 336 (1990) 517;
I. Affleck, A.W.W. Ludwig, Nucl. Phys. B 352 (1991) 849;
I. Affleck, A.W.W. Ludwig, Nucl. Phys. B 360 (1991) 341;
I. Affleck, A.W.W. Ludwig, Phys. Rev. Lett. 67 (1991) 161;
4 In field theory language, such an extra renormalization is equivalent to a wavefunction renormalization.
504
I. Affleck, A.W.W. Ludwig, Phys. Rev. B 48 (1993) 7292.

[5] C.L. Kane, M.P.A. Fisher, Phys. Rev. B 46 (1992) 1220.
[6] P. Fendley, A.W.W. Ludwig, H. Saleur, Phys. Rev. Lett. 74 (1995) 3005;
P. Fendley, A.W.W. Ludwig, H. Saleur, Phys. Rev. B 52 (1995) 8934;
C.L. Kane, M.P.A. Fisher, Phys. Rev. B 46 (1992) 15233.
[7] F.D.M. Haldane, Phys. Rev. Lett. 45 (1980) 1358;
J.L. Black, V.J. Emery, Phys. Rev. B 23 (1981) 429;
M.P.M. den Nijs, Phys. Rev. B 23 (1981) 6111.
[8] P. Gueret, N. Blanc, R. Germann, H. Rothuizen, Phys. Rev. Lett. 68 (1992) 1896.
[9] K. Bardakci, A. Konechny, hep-th/0009214;
S.A. Harvey, D. Kutasov, E.J. Martinec, hep-th/0003101.
[10] A. Sen, J. High Energy Phys. 9927 (1999) 9912.
[11] A. LeClair, M.E. Peskin, C.R. Preitschopf, Nucl. Phys. B 317 (1989) 411;
J.A. Harvey, D. Kutasov, E.G. Martinec, G. Moore, hep-th/0111154.
[12] E. Gava, K.S. Narain, N.H. Sarmadi, Nucl. Phys. B 504 (1997) 214;
A. Sen, J. High Energy Phys. 9809 (1998) 013;
A. Sen, Int. J. Mod. Phys. A 14 (1999) 4061.
[13] I. Affleck, A.A. Ludwig, Phys. Rev. Lett. 67 (1991) 161.
[14] L.I. Glazman, A.I. Larkin, Phys. Rev. Lett. 79 (1997) 3736.
[15] R.M. Bradley, S. Doniach, Phys. Rev. B 30 (1984) 1138.
[16] H.J. Shultz, G. Cuniberti, P. Pieri, Fermi liquids and Luttinger liquids, in: G. Morandi, P. Sodano, V. Tognetti,
A. Tagliacozzo (Eds.), Field Theory for Low-Dimensional Condensed Matter Systems, Springer-Verlag,
2000.
[17] V.J. Emery, in: J.T. Devreese, R.P. Evrard, V.E. van Doren (Eds.), Highly Conducting One-Dimensional
Solids, Plenum, 1979.
[18] P. Jordan, E. Wigner, Z. Phys. 47 (1928) 631.
[19] S. Coleman, Phys. Rev. D 11 (1975) 2088;
D.J. Amit, Y.Y. Goldshmidt, G. Grinstein, J. Phys. A 13 (1980) 585.
[20] F.W.J. Hekking, L.I. Glazman, Phys. Rev. B 55 (1997) 6551.
[21] P.W. Anderson, G. Yuval, D.R. Hamann, Phys. Rev. B 1 (1970) 4464.
[22] C. Nayak, M.P.A. Fisher, A.W.W. Ludwig, H.H. Lin, Phys. Rev. B 59 (1999) 15694;
C. Chamon, M. Oshikawa, I. Affleck, Phys. Rev. Lett. 91 (2003) 206403;
I. Affleck, J.-S. Caux, A.M. Zagoskin, Phys. Rev. B 62 (2000) 1433.
[23] See, for example, J.-S. Caux, A. Lpez, D. Suppa, Nucl. Phys. B 651 (2003) 413, and references therein.
[24] I. Chiorescu, Y. Nakamura, C.J.P.M. Harmans, J.E. Mooij, Science 299 (2003) 1869.
[25] See, for example, P. Ginsparg, Applied conformal field theory, in: E. Brzin, P. Zinn-Justin (Eds.), Field,
Strings and Critical Phenomena, 1988, Les Houches, Section XLIX.
KacMoody theories for colored phase space

(quantum Hall) droplets
Alexios P. Polychronakos
Physics Department, City College of the CUNY, Convent Avenue and 138th Street, New York, NY 10031, USA
Abstract
We derive the canonical structure and Hamiltonian for arbitrary deformations of a higherdimensional (quantum Hall) droplet of fermions with spin or color on a general phase space manifold.
Gauge fields are introduced via a KaluzaKlein construction on the phase space. The emerging theory is a nonlinear higher-dimensional generalization of the gauged KacMoody algebra. To leading
order in h this reproduces the edge state chiral WessZuminoWitten action of the droplets.
PACS: 05.30.Fk; 11.10.Kk; 71.10.-w; 73.40.Hm
1. Introduction
Describing fermions in terms of bosonic variables has been the source of much of our
progress in understanding their many-body dynamics. Such descriptions are collectively
termed bosonization [15].
An intuitive approach to such a description is to consider a dense collection of fermions
forming a droplet on their phase space and study the dynamics of the Fermi surface of
the droplet. In two dimensions this leads to a chiral theory in 1 + 1 dimensions [5].
In a previous paper [6] we used a phase space canonical approach to derive the Poisson structure and Hamiltonian for arbitrary deformations of constant-density droplets for
E-mail address: alexios@sci.ccny.cuny.edu (A.P. Polychronakos).
doi:10.1016/j.nuclphysb.2005.01.016
506
A.P. Polychronakos / Nuclear Physics B 711 [FS] (2005) 505529
spinless (Abelian) fermions on a general higher-dimensional phase space. This provided a

nonlinear generalization of the results of Karabali and Nair [7] who derived a chiral boundary action for the higher-dimensional quantum Hall effect proposed by Zhang and Hu [8].
The nonlinear terms derived in [6] captured higher order quantum corrections in the 1/N
approximation. Such nonlinear deformations for the fluid dynamics of the two-dimensional
quantum Hall effect have been studied in [9].
The analysis in [7] uses the quantum density matrix formulation and large-N approximations [10] and also applies to a non-Abelian situation [11], where it leads to a generalization
of the chiral sigma model known as the WessZuminoWitten model [3]. This model describes boundary perturbations of the quantum Hall (droplet) state of fermions with color
degrees of freedom.
The inclusion of spin, color or other internal degrees of freedom for the fermions in
the phase space formulation presents new challenges, since the usual semiclassical droplet
picture has to be modified and extended to accommodate the new degrees of freedom.
Nevertheless, such a description is possible and leads to an interesting gauge generalization
of the results for scalar particles. This will be derived in the present paper.
The organization of the paper is as follows. In Section 2 we give a general analysis
of phase space density dynamics and review the relevant results of [6]. In Section 3 we
introduce internal degrees of freedom as classical phase space variables and derive the corresponding droplet dynamics; we argue that the internal space needs to be quantized for
an accurate description of the system and present the corresponding dynamics in terms of
matrix generalizations of the boundary field obeying a generalized KacMoody algebra.
We further demonstrate that the dynamics to leading order in h reproduce the Wess
ZuminoWitten model. In Section 4 we introduce gauge degrees of freedom and define
gauge transformations in the phase space structure, through a mechanism analogous to
a quantum KaluzaKlein reduction; we generalize the results of the previous section for
the gauged case, reproducing a gauge KacMoody algebra and a gauged WessZumino
Witten model. Finally, in Section 5 we discuss some outstanding issues.
Note on h : it is customary to put h = 1 and eliminate it from all expressions. In our
case, however, we have to keep track of various orders of h in our calculations in order to
reproduce the correct leading-order model. We could have kept the convention h = 1 and
reintroduce it where appropriate and needed. For the sake of clarity, however, we preferred
to keep h explicit as a book-keeping device, in order to indicate the appropriate scale of
each term in the formulae, accepting the price of some h -litter in the expressions.
2. Review of phase space droplet dynamics
2.1. General formulation
We shall start by considering noninteracting (spinless) particles on a general Ddimensional phase space manifold with coordinates , = 1, . . . , D and Poisson structure

, sp =
(1)
507
where the subscript sp stands for single-particle. For a nondegenerate Poisson structure the
dimension D should be even. The volume element in this phase space is
d
dD = ,
where d =
D
d , = det .
(2)
=1
The particles have Hamiltonian V () and perform classical motion:

= , V sp = V .
(3)
A dense collection of particles on this phase space is described in terms of its density
(, t). (This is essentially a phase space fluid dynamical description; for a recent review
see [12].) The motion of the underlying particles induces a time dependence for the density . Its time evolution is given by a canonical transformation generated by V :
= {, V }sp = V .
(4)
We can obtain the same dynamics without referring to the underlying particles by assuming
a Hamiltonian and canonical structure for the field
= {, H }.
(5)
(These brackets should not be confused with the single-particle brackets (1).) Choosing as
Hamiltonian the total particle energy

d
H = V
(6)
the appropriate Poisson brackets are

(1 ), (2 ) = (+ ) (+ ) (+ ) ( ),
(7)
2
where we defined relative and mid-point coordinates = 1 2 and + = 1 +
2 .
The above brackets (7) are the standard infinite-dimensional Poisson algebra of functions on the phase space manifold. In terms of test functions their form becomes more
obvious: defining

d
[F ] = F ()()
(8)
for some function on the phase space F , then the brackets of two such integrals are

[F ], [G] = {F, G}sp .
In deriving the above we used the identity

=0

(9)
(10)
which is a corollary of the Bianchi identity for . The equation of motion (4) arises as
the canonical evolution with Hamiltonian (6).
508
The above algebra has Casimirs. For any function of a single variable f (x), the integral

d
C[f ] = f ()
(11)
has vanishing Poisson brackets with and constitutes a Casimir. There are, thus, an infinite
tower of Casimirs spanned by Cn C[x n+1 ] for n = 0, 1, 2, . . . .
2.2. Droplet dynamics
We now specialize to the case where the particles underlying the density are (temporarily spinless) fermions. A dense collection of fermions in phase space will form a
Fermi liquid. Semiclassically, the fermions will fill densely a region of the phase space,
with one particle per volume (2 h)
D/2 , forming a constant density droplet of arbitrary
shape. Under time evolution, the density remains constant inside the droplet (by Liouvilles theorem) while each point on its boundary moves according to the single-particle
equation of motion, thus deforming the shape of the droplet.
To describe the droplet it suffices to determine the shape of its (D 1)-dimensional
boundary. This can be done by expressing one of the phase space coordinates on the boundary, say D R, as a function of the remaining phase space coordinates . (In the sequel
we use early Greek letters for the set of indices = 1, 2, . . . , D 1 while middle Greek
letters will take values in the full D-dimensional space.) So the dynamical variable is the
function R(, t). (For a finite droplet, it is convenient to assume that the origin of coordinates is inside the droplet and to think of R as a radial coordinate and i as angular
coordinates.)
The canonical structure of the droplet variable R arises from a Hamiltonian reduction
of the full density canonical structure: a constant-density droplet of arbitrary shape constitutes a particular class of density functions and thus a submanifold of the full manifold of
configurations for , of the form

= o R D ,
(12)
where (x) = 12 [1 + sgn(x)] is the step function. To find the droplet Poisson brackets we
need to project the canonical two-form of on this submanifold. This can be done with
the help of the so-called cartographic transformation of the density . (We refer the reader
to [6] for further details and a full derivation.) Introducing the shorthand fb for quantities
defined on the boundary of the droplet

fb f D = R,
(13)
the induced Poisson brackets for R are expressed as

b+ D
b+ ( ) b+ R+ ( )
R(1 ), R(2 ) =
o
and the Hamiltonian becomes

d
H = o V () R D .
(14)
(15)
509
As before, + and stand for mid-point and relative coordinates.

The above Poisson structure and Hamiltonian encode the full dynamics of the droplet
and imply the canonical evolution for the boundary
R = bD Vb b R Vb .
(16)
This equation refers only to the boundary, although the Hamiltonian is defined in the bulk
of the droplet. The same equation can be obtained by following the single-particle evolution
of the particles on the boundary of the droplet [6].
It is worth noting that if the coordinate D is chosen to parametrize the potential (that
is, surfaces D = const are equipotential), V = V ( D ), the second term above drops and
we get
R = bD Vb .
(17)
In the special case when bD is nonzero only for a single value of the index (there is
a global variable conjugate to D , call it 1 ), and is a function only of D , the above
equation becomes
R = b01 V b 1 R
(18)
which can easily be solved by a hodographic transformation, interchanging R and 1 .

We should warn that the droplet may have more than one boundaries, depending on its
topology. In such cases we need to introduce several commuting boundary fields Rn , one
for each boundary. Similarly, the boundary could intersect = const lines at more than
one D , in which case we need again to introduce several boundary fields, one for each
branch, with appropriate matching conditions tying them into a unique boundary.
We conclude this section with the following remarks:
(1) The Poisson brackets (14) of R contain an affine chiral part (the first term in the
bracket) as well as an ordinary Poisson density structure over the gauge manifold { }/ D .
The quotient arises because is degenerate, being odd-dimensional, and effectively the
variable conjugate to D drops out.
(2) (14) satisfies the Bianchi identity, as a corollary of the Bianchi identity of the Poisson brackets for , although its direct check is highly nontrivial. In the special case when
is independent of D the affine and linear terms decouple and individually satisfy the
Bianchi identity. In the generic case, however, both terms are needed to satisfy the identity.
This will be relevant to the case of particles with internal degrees of freedom.
(3) The Casimirs of the original density for the droplet become Cn = C0 . So they are
all neutralized, the only remaining Casimir being the total particle number C0 = N .
(4) The constant o appears both in the Hamiltonian and the Poisson brackets and is
irrelevant for classical dynamics. The semiclassical interpretation of the droplet, however,
fixes the value o = 1/(2 h)
D/2 , which will be important for quantization and for the case
of spinning particles.
510
3. Particles with internal degrees of freedom

The generalization of the above semiclassical construction for fermions with internal
degrees of freedom (spin, color, flavor, etc.) presents some conceptual problems, due to the
fact that internal quantum numbers are never really classical.
One approach would be to view the internal degrees of freedom simply as different
species of fermions and apply the above procedure separately for each species. We would
obtain a set of partially overlapping droplets with mutually commuting boundary fields R.
This, however, has several drawbacks. One is that such a description would not allow the
particle number of each species to fluctuate (remember that the total particle number for
each droplet is a Casimir), thus excluding the possibility of having transitions of fermions
from one species to the other. Another related problem is that the Hamiltonian may not
be diagonal in the particular chosen basis of flavors. This would happen, e.g., for a spindependent Hamiltonian or in the case where we include gauge fields that act on the spins
or colors.
We need, therefore, to start from a proper semiclassical description of the full set of
degrees of freedom without the above limitations. This will be done in this section.
3.1. Introducing spins as classical phase space variables
We shall start from a description where the classical phase space encodes also the internal degrees of freedom of the particles. This will be done by considering the internal
quantum numbers as arising from the quantization of an internal, compact phase space for
the particles. For shortness, we shall refer to this space as representing spin variables, understanding that it can also represent color, flavor or any other internal degrees of freedom.
Consider the direct product of the original phase space and an additional compact
phase space with coordinates i and Poisson structure ij (middle Latin letters will stand
for indices of the components of this compact phase space). Clearly i = 0. The dimensionality DI of the space will be left arbitrary. The total volume of this space, however,
will be chosen as n(2 h )DI /2 . We see that the size of this space is microscopic, involving
Plancks constant.
Semiclassically, we will have one quantum state per volume (2 h)
DI /2 in the internal
phase space. The above choice of volume implies that we shall have n quantum states
associated with this space, thus endowing the particles with n internal states. The classical
variables i represent spin operators for the particle. The droplet procedure of the previous
section can, then, be applied to the total phase space (, ).
We shall choose conventions in which the canonical two-form of the spin space ij is
of order h while the range of the coordinates i is of order h 0 . This makes the Poisson
structure ij of order 1/h . We shall also take the internal phase space to be homogeneous
and the determinant of ij equal to (2 h)
DI , therefore making the determinant of the
i
total Poisson structure independent of

DI
det ij = det det ij = (2 h)
.
(19)
There are many potential realizations of the spin space i . A specific example of such a
space is S 2 with canonical two-form proportional to the area form. Choosing 1 = /2 ,
511
2 = (n/2) cos , with (, ) polar and azimuthal angles on the sphere, the canonical twoform and Poisson structure will be
=
hn
sin d d = 2 h d 1 d 2
2
2 1
1
.
, =
2 h
(20)
The range of (1 , 2 ) is (1, n) and the total area of this space is 2 h n. Semiclassically it
can support n quantum states. The quantization of this phase space reproduces the lowest
Landau level of a particle on the sphere with a magnetic monopole of strength n at the
center. It is known that these states form a spin- n2 multiplet of the group of rotations, the
Cartesian coordinates of the particle becoming spin- n2 SU(2) matrices. The number of
states is 2j + 1 = n + 1, the shift due to the nonzero curvature of the space.
Another realization of the internal phase space would be the Grassmanian manifold
G = U (M)/U (M1 ) U (Mk ) (M1 + + Mk = M) with an appropriate canonical
form. This can be visualized as the lowest Landau level of a particle moving on the group
manifold U (M) with an appropriate magnetic field. The canonical structure = dA is
determined by the Kirillov one-form

A = i h tr KU 1 dU ,
(21)
where U is a U (M) matrix and K is a Hermitian M M matrix that can be chosen diagonal. The above is invariant under right-multiplication of U by a unitary matrix commuting
with K, so the corresponding U (N ) coordinates have to be eliminated. The phase space
manifold, then, is the Grassmanian G where M1 , . . . , Mk are the degeneracies of the eigenvalues of K.
Quantization of the above phase space requires that the eigenvalues of K be integers
(a condition akin to the monopole quantization on the sphere). The quantum states will
form irreducible representations of U (M) with lengths of Young tableau rows given by the
eigenvalues of K and U (1) charge equal to the total number of boxes. So this phase space
will reproduce internal color quantum numbers in a given representation of SU(M) and
the classical coordinates i represent color matrices in this representation. The previous S 2
monopole construction can be realized as the Grassmanian manifold U (2)/U (1) U (1)
with the two eigenvalues of K differing by n.
The exact realization of the spin space is unimportant at this point. The only thing that
matters is the fact that we will have n internal states for each fermion. Specific realizations, however, will be more convenient depending on the dynamics and symmetries of the
problem, as will be apparent later.
3.2. Realization of droplets with internal degrees of freedom
We are now set to apply the droplet formalism to the problem. The total phase space has
dimension D + DI and can accommodate one fermion per volume (2 h)
(D+DI )/2 . Fermi(D+D
I )/2 , reserving the
ons on this space will form a droplet with density o = 1/(2 h)
D/2
notation o = 1/(2 h)
for the density in coordinate phase space. The droplet boundary
variable R will be a function of both and i .
512
The Poisson brackets of R are given by the general formula (14) applied to the present
phase space structure:

R(1 , 1 ), R(2 , 2 )

b+ D
ij
b+ ( )( ) b+ R+ ( )( ) + i R+ j ( ) , (22)
=
o
where we used (19) and (2 h)
DI /2 o = o . Assuming a particle Hamiltonian V (, ) that
depends also on the spin variables, the Hamiltonian for R will be

d d
H = o
(23)
V (, ) R(, ) D .
The above would be an exact classical description of the droplet. The fact, however, that
the internal space is of Planck size and supports a few quantum states renders it essentially
quantum mechanical and makes the classical description inadequate. We need, therefore,
to quantize the internal degrees of freedom and incorporate them in the Poisson brackets
for R. This can be done by quantizing the spin coordinates .
This quantization goes along standard lines. The i become noncommutative and are
represented by n n matrices. Functions of the i become matrices on the n-dimensional
internal Hilbert space; real functions such as R( ) become Hermitian matrices R ab , where
early Latin letters a, b, . . . , = 1, . . . , n will stand for spin indices. Integration over the phase
space amounts to summing over Hilbert space states with a volume of (2 h)
DI /2 per
state, that is, a trace over the Hilbert space

d
D /2

(24)
d tr
(2 h)
I tr or
det( ij )
while the -Poisson brackets become matrix commutators over i h:
1
{A, B} ij i Aj B [A, B].
(25)
i h
We also need the Dirac delta-function (1 2 ). Since this relates two different points in
the space it should carry two sets of matrix indices (a1 , b1 ; a2 , b2 ) and implement the
defining property

d1 F (1 )(1 2 ) = F (2 ).
(26)
This implies for matrix quantities
tr1 (F1 12 ) = F2 ,
(27)
where the subscripts 1, 2 refer to the first and second set of matrix indices. This means that
12 is proportional to the matrix that exchanges the matrix spaces 1 and 2, which is written
in terms of Kronecker deltas as
(12 )a1 b1 ;a2 b2 = a1 b2 a2 b1
(28)
and satisfies
F1 12 = 12 F2 ,
F2 12 = 12 F1 .
Now we may determine the dynamics of the matrix field variable R.
(29)
513
3.2.1. Poisson structure

In order to obtain Poisson brackets for the matrix variable R in a form that is not too
unwieldy we shall assume that is independent of D . This can always be achieved
with an appropriate choice of coordinates; full generality can be restored after obtaining
the Poisson bracketsof R by performing the inverse change of variables. This choice guar
antees that b and b become independent of R and are scalar functions of .
To translate the expression in (22) to matrix spin variables, observe that the last term in
the Poisson brackets of R, involving -space derivatives, can be written as

ij
+ i R+ j ( ) = R(1 ), (1 2 ) = R(2 ), (1 2 ) .
(30)
1
This translates to the matrix expression

1
1
[R1 , 12 ] = [R2 , 12 ]
(31)
i h
i h
the equality of the two expressions being ensured by (29). We also have to express functions
F+ defined on mid-point coordinates + in terms of matrix indices. For this, we remark
that, when F+ multiplies delta functions or their derivatives, it can also be expressed as

F (1 ) + F (2 )
1 + 2
=
F+ = F
(32)
2
2
and this translates to 12 (F1 + F2 ) in matrix notation.
We now have all the ingredients. Each classical expression in (22) becomes a corresponding matrix in the spin space with a residual dependence on ; subscripts + and
refer to mid-point and relative coordinates. Altogether we obtain

ab
R (1 ), R cd (2 )

+ D
1
ad cb
cb ad
=
+ ( ) ad cb + R+
+ R+
( )
o
2

1 ad cb
cb ad
R+ R+
(33)
( ) .
i h
Note that the last term involves an explicit h .
These brackets are a generalization of the KacMoody algebra. To make this explicit,
define the generators T A , A = 1, 2, . . . , n2 1 of the fundamental representation of SU(n)
(capital Latin letters will stand for U (N ) generator
indices) and append the U (1) generator
(proportional to the n n unit matrix) T 0 = I / n for the full set of generators of U (n).
We choose their normalization so that they satisfy the orthogonality and completeness
conditions

tr T A T B = AB ,
2 1
n
A ab
A cd
T
T
= ad cb .
(34)
A=0
Their commutators define the U (N) structure constants

A B
T , T = if ABC T C
(35)
514
(with f 0AB = 0). We further define the symmetrized trace

d ABC =

1
A B C
tr T T T + T B T A T C .
2
(36)
The symbol d ABC is also known as the anomaly of the group U (n), since it appears in the
evaluation of the triangle anomaly graphs in (3 + 1)-dimensional gauge theories.
The matrices T A can be used as a basis to express the Hermitian matrix R:

ab

R A T A , R A = tr RT A .
R ab =
(37)
A
In terms of the dynamical variables R A the Poisson brackets (33) become

A

R (1 ), R B (2 )

+ D
1
C
C
+ ( ) AB d ABC + R+
=
( ) + f ABC R+
( ) . (38)
o
h
We recognize this as a KacMoody type algebra generalized to higher dimension with an
additional term proportional to the anomaly of U (n).
This algebra inherits the Casimir C0 of the classical algebra. The integral

d
d 0
N = o tr R = o
(39)
nR
commutes with R. This is the total particle number.

3.2.2. Hamiltonian
The Hamiltonian of the droplet can similarly be expressed in matrix form. The only
question is matrix ordering. A -dependent single-particle Hamiltonian V (, ) and its
product with the step function (R D ) of a matrix R introduce ordering ambiguities.
Ordering will not be an issue as long as V is at most linear in the i . This is, in fact,
related to the question of the specific realization of the internal spin space, as commented
at the end of the previous section and as we shall analyze presently.
Remember that the spin space is microscopic (of order h ). The relevant variables are
really S i = h i , representing quantum spin operators, and their products are higher order
in h.
Such higher-order terms will be equally relevant only if they come with large coefficients (of order 1/h for quadratic terms etc.), which is unnatural. This, however, can be
avoided if the internal phase space is chosen wisely so that all spin terms in the Hamiltonian can be expressed linearly in the coordinates i . This can best be illustrated with the
following example.
Consider the case that there are n internal states. As exposed in the previous section,
2
we may realize them either as a spin- n1
2 representation of SU(2), in terms of an S phase
space, or as the fundamental representation of SU(n), in terms of the Grassmanian space
U (n)/U (n 1) U (1), that is, the Kirillov action (21) with K = diag(1, 0, . . . , 0).
In either case the Hilbert space is n-dimensional, so there are n2 linearly independent
Hermitian operators on this space (including the identity operator) that can be used to expand any operator. In the SU(2) case these are the three spin matrices S 1,2,3 = h 1,2,3 (the
515
i are Pauli or higher-spin SU(2) matrices) as well as their ordered products up to degree n 1. A general spin-dependent term in the single-particle Hamiltonian, then, can be
expressed as a sum of such monomials. In the SU(n) realization, however, the n2 1 fundamental generators (represented by coordinates i ) are a complete set and therefore any
Hamiltonian can be expressed as a linear expression in the i . Clearly, physics originating
from SU(2) spin will only involve linear expressions in the S i , while physics originating
from color or flavor degrees of freedom will give rise to a linear expression in the full set
of SU(n) generators.
In conclusion, an appropriate choice of realization of the internal states will lead to an
expression for V (, ) linear in and we may write
V (, ) = V () + Vi () i V0 + h V ,
(40)
where V is the spinless part and h V = Vi i is the spin part of the single-particle energy
(where we explicitly indicate the fact that it is of order h ). The Vi are magnetic field terms.
The matrix representation of V is unambiguous. Further, there are no ordering problems in
the definition of (R D ), since it is defined pointwise in and does not couple matrices
R at different points that could be noncommuting. We may define it, for instance, via the
expression

1
dk ik(R D )
D
R =
(41)
e
.
2i
k + i
The multiplication of V and (R D ) inside the integral is also free of ordering issues: V
is linear in and the trace representing -integration makes its exact placement immaterial.
Altogether the droplet Hamiltonian will be given by (23) with the integral expressed as
a trace

d
H = o tr V (, ) R(, ) D .
(42)
The matrix expression inside the trace can also be obtained by integrating the classical
expression in terms of D and then promoting R to a matrix, which will give the same
result as using the matrix definition of (R D ).
The above Poisson structure and Hamiltonian imply an equation of motion for R: R =
{R, H }. In evaluating this Poisson bracket it is useful to use the relations
F+ =
F (1 ) + F (2 )
,
2
F+ G+ =
F (1 )G(2 ) + F (2 )G(1 )
2
(43)
which hold true for any two functions F, G when multiplying the delta-function ( ) or
its derivatives. We obtain
1
1
R = bD Vb b ( R Vb + Vb R) [R, Vb ].
2
i h
(44)
This is the matrix version of the equation of motion (16) with a symmetric ordering of the
middle term.
516
3.3. The leading-order action

Assuming that the droplet deviates slightly from an equilibrium configuration, we can
analyze its motion as a perturbation of its equilibrium shape. This will be useful for large
droplets (large number of particles) where we can recover a boundary action to leading
order in h .
Consider a reference droplet configuration filling the phase space up to an energy level
Eo for the scalar part of the single-particle Hamiltonian V0 . That is, the boundary field
of the reference configuration R = ro ( ) is -independent (proportional to the identity
matrix) and satisfies

V0 ro ( ), = Eo = const.
(45)
(From now on, a subscript o in any quantity will signify the value of the quantity at D =
ro .)
Note that ro ( ) is not a time-independent solution of the equations of motion, since it
does not minimize the full Hamiltonian. The true static solution Ro includes a -dependent
part h
o and satisfies

V Ro ( ), , = Eo .
(46)
Substituting Ro = ro + h o and V = V0 + h V in the above and expanding to first order in
h we obtain

V
V0
.
o = , where uo =
(47)
uo
D o
Nevertheless, we may expand our droplet around the reference configuration ro . Such a
perturbation can be written as
R = ro + h ,
(48)
where ( ) is a matrix and we have explicitly indicated that this is an order h perturbation. Correspondingly, the Poisson brackets (33) or (38) and Hamiltonian (42) have to be
expanded to that order.
3.3.1. Leading-order Poisson brackets and Hamiltonian
For the Poisson brackets, the leading part of the first two terms in (38) (affine and
proportional to d ABC ) is of order h 0 and involves the scalar equilibrium term ro alone.
In the last term the scalar part ro (proportional to the identity matrix A0 ) drops out; due to
the explicit 1/h in its coefficient, however, the part survives. Overall we have
A

o+
D
B
(1 ), (2 ) = 2
o+ o+ ro ( ) AB + f ABC C ( ) .
h o
(49)
(Note that we reinstated the dependence of on R; the dependence on the matrix part
that would complicate the Poisson brackets is down by h and can be neglected.)
517
The single-particle Hamiltonian perturbed to order h around ro is

V = Eo + h uo ( )( ) + h Vo
and this gives for the droplet Hamiltonian

1
d
d
H = Ho + h o Eo tr + h 2 o tr uo 2 + Vo .
2
o
o
(50)
(51)
Ho is the energy of the unperturbed droplet; it is a constant and can be discarded. The
next term, of order h , is proportional to Eo (N No ), where N is the Casimir (39) and
No the value of the Casimir for the unperturbed droplet. It is therefore itself a Casimir and
does not contribute to the equations of motion. It can be set to zero as an initial condition,
corresponding to a constantvolume perturbation of the droplet (total number of particles
N constant). We end up with a Hamiltonian of order h 2 with a linear and a quadratic term
in :

1
d
H = h 2 o tr uo 2 + Vo .
(52)
2
o
Finally, the equation of motion as obtained by the above Hamiltonian and Poisson brackets, or from the full equation of motion (44) to first order in h , becomes

= oD o ro (uo + Vo ) + i[, Vo ].
(53)
Note that the factors h 2 o have canceled.
3.3.2. WessZuminoWitten action
The form of the above equation of motion is suggestive. Let us define the differential
operator

= oD o ro
(54)
representing a vector field along the classical trajectory of a particle on the boundary of the
anti-self-adjoint on the boundary of the droplet under the
droplet. Note that is properly
integration measure d/ o due to the relation

= 0.

(55)
o
In terms of the equation of motion becomes
+ i[Vo , ] (uo ) = Vo .
Similarly, the Poisson brackets of become
A

o+
+ ( ) AB + f ABC C ( ) .
(1 ), B (2 ) = 2
h o
(56)
(57)
We recognize the above Poisson structure as the KacMoody algebra of a chiral current
J+ = . This algebra is realized by one of the light-cone components of the current in
a WessZuminoWitten (WZW) model, with the corresponding light-cone coordinate x +
518
identified as the trajectory along which acts and the other light-cone coordinate x
identified as time. The remaining directions of appear simply as parameters.
This immediately provides a Lagrangian realization of the above Hamiltonian structure,
in terms of the WZW action of a unitary nn matrix field U . The chiral field is identified
as the current
= iU 1 U.
(58)
The action that reproduces the Poisson brackets of is the WZW action at critical
coupling. The WZW model gives equation of motion J+ = 0, corresponding to = 0,
which implies a vanishing Hamiltonian. The full action, then, is the WZW action minus
the Hamiltonian

d
S = SWZ + h 2 o dt
o

2
1 1
1
1
1
1
tr U t U U U + uo U U + i Vo U U .
(59)
2
2
The WessZumino action SWZ can be written as an integral over a D-dimensional manifold
whose boundary is the boundary of the droplet. Introducing a variable s [0, 1] we can take
U (s, , t) to be any extension of U away from the boundary such that
U (1, , t) = U (, t),
U (0, , t) = 1.
The WZ action with the proper normalization to reproduce (57) is

h 2 o
d
SWZ =
ds dt tr U 1 s U U 1 U, U 1 t U .
6
o
(60)
(61)
The above WZ action can be written in a more suggestive way by choosing the Ddimensional manifold of integration to be the
bulk of the droplet itself, identifying s with
D and allowing the integration measure 1/ to vary accordingly in the bulk of the
droplet. In doing that, we have to take into account the following:
= (oD o ro ) is defined on the boundary and must be appropriately extended in the bulk;
unlike s, D is not constant on the boundary, but rather D = ro ( );
the integrand should be a closed form so that its variation reproduce the same boundary
term as (61).
These actually combine to give the WZ bulk action a simple geometrical form as the
integral of a (D + 1)-form, in terms of the canonical two-form = 12 d d and the
exterior derivative d = dt t + d . In anticommuting form notation

1 k1
1 3
h 2 o
D
tr U dU , k = .
SWZ =
(62)
(k 1)!
3
2
D
This has the form of a KhlerWessZumino term on the bulk of the droplet with playing the role of the Khler structure [13]. It is obviously a closed form, since both and
519
tr(U 1 dU )3 are closed (d = 0 is equivalent to the Jacobi identity for ). Its variation
will give the boundary integral

2

h 2 o
SWZ =
(63)
ok1 tr U 1 dU U 1 U .
(k 1)!
D1
To see that this is what we want, note that k is a top form on the phase space and thus
k!
k = k! det d 1 d D =
det 1 ...D d 1 d D ,
(2k)!
(k 1)!
k1 =
(64)
det 1 ...D 1 2 d 3 d D
2(2k 2)!
( det is really the Pfaffian of the antisymmetric

). Restriction of the form
matrix
k1 on the boundary will produce a factor of det o = 1/ o and will induce the substitutions
d D = d ro ,
(U 1 dU )2
dU = dt t U + d U.
(65)
= dt d [U 1 t U, U 1 U ].
So
Taking into account the combinatorics, the
terms in (64) with 1 = D or 2 = D reproduce the term oD in , while the terms with
any of the remaining s equal to D reproduce the term o in .

Finally, we may recast the full action (59) above in a more familiar form by renaming
Vo = A0 and defining the gauged time derivative
D0 U = t U i[A0 , U ] = t U + i[Vo , U ].
(66)
The action becomes

h 2 o
dt d
S = SWZ
2
o
1

tr U (D0 U uo U )U 1 U + i A0 U 1 + U 1 A0 U .
(67)
This has the form of a gauged WZW model. The last term is the extra term needed in
the action to absorb the anomaly of the WessZumino term under gauge transformations
U W 1 U W . We have recovered the action of Karabali and Nair for a temporal gauge
field, generalized to an arbitrary phase space droplet and with a -dependent potential
gradient uo .
3.3.3. Comments on the dynamics of the model
The expression for = iU 1 U implies that the deformation of the droplet is generated by the unitary field U , in analogy with the spinless case. Specifically, assume a small
canonical transformation of the coordinates , induced by an order-h generating function
h (, ):
ro ro + h oD ,
+ h o ,
i i + h ij j .
(68)
The deformations of
= ro and
are of order h . The deformation of
however, is
of order h 0 (since ij is of order h 1 ) and of the same order as i ; it cannot be written
D
i,
520
in the above infinitesimal form. Instead, we must write the analog of a finite canonical
transformation on the spin space, which is a unitary transformation.
The dynamical meaning of the above WZW structure is as follows: the operator
L = uo generates classical motion on the manifold D = ro under the spinless part of the
single-particle Hamiltonian V0 . (The variable represents time of flight along the classical
path.) In the absence of the spin-dependent (matrix) part V , particles would simply move
along the flow of L, and so would the surface of the droplet. Such motion also rescales
distances away from the droplet. The rescaled deformation of the surface uo , then, would
evolve as a co-moving matrix: uo = L(uo ). The trace (scalar) part of represents total
particle number (charge) fluctuations, while its traceless part represents spin fluctuations
of the droplet. The scalar and traceless parts actually decouple, signaling spin-charge separation in this limit.
In the absence of V the motion of the various matrix components of would decouple
and could be described as a collection of independent Abelian chiral models. The presence
of the spin part V , however, causes a Pauli rotation of the coordinates i and thus couples
the matrix components inducing an extra unitary transformation of the matrix that can
be understood as gauge (spin) rotation. The gauged WZW action is the proper dynamical
setting for describing such motion.
The appearance of gauge structure in the problem is somewhat surprising, since we
have not introduced gauge fields or considered gauge transformations. In the next section
we shall complete the picture by doing that.
4. Introducing gauge fields

In the analysis so far we have described spin in terms of an internal compact phase
space of the particles. Its canonical structure decoupled from the one of the kinematical
phase space ( i = 0) and any nontrivial spin dynamics arose out of the Hamiltonian.
We may further couple spin and kinematical degrees of freedom by introducing nonzero
phase space structure constants between the two spaces. As we shall demonstrate, this
amounts to introducing non-Abelian gauge fields and endows the dynamics with a nonAbelian gauge symmetry. For other examples of introducing gauge degrees of freedom in
the canonical description of particles or fluids see [12,14].
4.1. Coupling the phase spaces
The most convenient setting for analyzing the situation is in terms of the canonical
one-form formulation of the phase space. We give below the relevant facts for our purpose.
We consider a phase space x endowed with a canonical one-form A = A dx and
a Hamiltonian V . (In our case, x will comprise both and i .) A and V could be
time-dependent. The phase space action and Lagrangian are
L = A x V ,
dS = L dt = A dx V dt
(69)
which leads to the canonical two-form = dA inverse to the Poisson structure :

= A A ,
= .
(70)
521
The above action has the standard phase space invariances. The first is generated by
adding to the Lagrangian the total time derivative of an infinitesimal phase space function
(x; t)
L =
(71)
which amounts to the Abelian gauge transformation

A = ,
V = t
(72)
leaving the canonical two-form and Poisson structure invariant. The other is general
coordinate invariance, generated by arbitrary infinitesimal coordinate redefinitions
x = (x; t)
(73)
which is compensated by the transformation

A = A + A ,
V = V A t .
(74)
(The minus sign in (73) is put to stress the fact that this is a passive transformation of
coordinates.) The above can be rewritten as

V = (t A + V ) t A
A = + A ,
(75)
involving canonically invariant quantities and an Abelian gauge transformation generated
by = A .
Canonical transformations are a special case of coordinate transformations leaving the
Poisson structure invariant. Choosing as coordinate deformation parameters

c = x , =
(76)
we get for the change in A and V

c A = + A ,
c V = {V , } + t A .
(77)
invariThis corresponds to an Abelian gauge transformation on the A that leaves

ant. For time-independent A , V transforms by a canonical transformation, while a time
dependence in A contributes an extra correction.
We specialize now to the phase space of interest ( , i ). The original Lagrangian is

L = A () + Ai ( ) i V .
(78)
The Poisson structure decouples the spin and kinematical phase spaces, which reflects to
the fact that A () and Ai ( ) depend only on their corresponding phase space variables,
ensuring i A = Ai = 0.
We shall couple and by relaxing the above condition. In doing so we do not want
to distort the structure of the internal phase space. Its volume, as well as the area of all
noncontractible two-submanifolds, must remain fixed to appropriate integers for a consistent quantization (cf. to monopole quantization for S 2 and K-eigenvalue quantization for
the Grassmanian case of Section 3.1). It should also stay a homogeneous space to allow
for a linear Poisson algebra in terms of appropriate spin generators. We shall, therefore,
522
keep its one-form Ai ( ) the same as above and independent of i and shall write it h A i to
explicitly indicate the fact that it is of order h .
We will however allow A to depend on . The new one-form will consist of the old
one, denoted by A , plus a -dependent order-h perturbation h Ai . We also write the Hamiltonian in the form V + h V = V h A0 , taking a hint from the previous section in renaming
V to A0 . Further, we will allow A and A0 to be time-dependent. Altogether the Lagrangian becomes
L = A () + h A i ( ) i V () + h A (, , t) + h A0 (, , t).
(79)
In terms of scales, ij = h (i A j j A i ) is of order h and ij = (1 )ij is of order 1/h

as it should.
4.2. Gauge transformations
We come now to the issue of gauge transformations in the spin space of our phase space
structure (79). Interpreting as spin variables, we understand that gauge transformations
should amount to local rotations of the -coordinates in their phase space; that is, canonical
transformations in the space that depend on the kinematical phase space coordinates.
These will be generated by an order-h function h (, , t)

i = h i , = h ij j ,
(80)
= 0.
Note that the above is not a canonical transformation on the full space, since the kinematical coordinates are not transformed and the Poisson bracket in the transformation of
is restricted. The canonical one-form and Hamiltonian are transformed according to (75),
which implies for A , A0 and A i :
ij

ij
A = h
i A j + h Ai j ,
ij

A0 = h i A0 ij j + h
0 Ai j ,
ij

A i = i + h
(81)
i A i j .
We observe that A and A0 transform as the space and time components of a one-form.
Calling x 0 = t and using middle-Greek letters for spacetime indices, , = 0, 1, . . . , D
(not to be confused with , = 1, 2, . . . , D used in early sections), we can write the above
as

A = h {A , } + h ij A i j ,

ij
A i = i + h
(82)
A i j .
Under the above transformation, the spin one-form transforms away from its reference
value A i by a total derivative. We may restore it to its original form by adding to the
Lagrangian the total derivative of h
ij A i j . This further transforms A to
g A = + h{A
, } .
(83)
This has exactly the form of a non-Abelian gauge transformation. Indeed, upon quantization of the spin space, -dependent quantities such as and A become matrices and
523
Poisson brackets become (1/i h)

commutators. So the above transform becomes
g A = i[A , ],

i = i i ,
(84)
which corresponds to the transformation of a covariant derivative D = iA and an

anticovariant spin matrix i under an infinitesimal unitary rotation U = 1 + i:
D U D U 1 ,
i U 1 i U.
(85)
The A , thus, can be rightfully considered as non-Abelian gauge fields on the phase space.
Gauge transformations are local rotations of the spin coordinates, which is a passive transformation, reflected in the anticovariant nature of i . We recover the dynamics of a single
spinning (in fact, colored) particle interacting with a non-Abelian gauge field in phase
space.
The group of the non-Abelian gauge transformations is determined by the realization
of the spin phase space. Gauge invariance derives from canonical spin transformations
and therefore inherits the full SU(n) symmetry group of the spin Hilbert space. Its realization, however, may be restricted by the physics of the problem (cf. the discussion of
Section 3.2.2).
As an example, in the realization in terms of a spin- n1
2 representation of SU(2), as
argued in Section 3.2.2, only operators linear in the spin variables S i = h i are natural and
thus the expression for A will be restricted to
A (, , t) = Ai (, t)S i ,
i = 1, 2, 3.
(86)
A general SU(n) transformation will take the above A away from this form. Only unitary transformations in the SU(2) subgroup transforming linearly the i will be proper
gauge transformations, the field A being in the spin- n1
2 representation of the group. The
Grassmanian representation with a Kirillov form K = diag(n1 , . . . , nM ) and a corresponding linear restriction for the form of A would produce an SU(M) gauge group in the
representation corresponding to Young tableau with ni blocks in row i.
4.3. Equations of motion
The above interpretation can be further justified by looking at the single-particle equations of motion in the presence of the extra coupling between and due to the (timedependent) A and A0 . These can be obtained by varying the action with an arbitrary
and setting the variation to zero. Using (75) this implies
x = 0,
(87)
where x stands for {x 0 , , i }. (In principle, since we do not vary x 0 = t, we only obtain
the above equations for = 0. The = 0 equation, however, holds true as a corollary of
the remaining equations, due to the identity x x = 0, so we obtain the full covariant
524
set of equations.) These can also be written as

x = ( V + 0 A ) = x , V sp + 0 A
(, = 0)
(88)
which, for time-independent A , reduce to the usual canonical equations of motion.

Applying the above Eq. (88) to the case of the Lagrangian (79) for , i we obtain

i
+ h ( A A ) = V + h 0 A A0 + h
i A ,

ij j + h i A0 + i A = 0,
(89)
where = d A is the reference (uncoupled) canonical two-form. The above equations can
be combined and rewritten as
( + h F ) = V + h F0 ,

i h A0 + A , i = 0,
(90)
where
F = A A + h {A , A }
(91)
is the non-Abelian field strength of the gauge field A . These equations have the structure of the equations of motion of a particle with non-Abelian degrees of freedom and
only involve gauge covariant quantities: the first is the standard minimal coupling of the
particles coordinates to a field strength coupled to its (non-Abelian) charge (with a scalar
potential V and an electric field F0 ), while the second is the covariant parallel transport of
the spin over the particles phase-space-time trajectory. Due to the anticovariant (passive)
nature of , its equation of motion involves covariant derivatives with the opposite sign for
A . Observables of the form

Q = d Qi i tr Qi i
(92)
are gauge invariant, while Qi transform covariantly.
4.4. Gauged droplet dynamics
The generalization of the droplet construction for the gauged phase space considered
above is straightforward. The Poisson structure is, now, time-dependent, involving the
gauge fields A , but the counting of states and fermion exclusion principle that led to
constant-density droplets remain the same. The construction of the boundary field Poisson
brackets, Hamiltonian and equation of motion for the classical theory is as in Section 3.2
with the generalized form of b , involving gauge fields, appearing in the formulae.
The quantization of the spin space in this case presents some new ordering ambiguities,
since we cannot any more assume that is independent of D and R. The proper ordering of the full nonlinear matrix Poisson brackets for R ab will be partly determined by the
requirement that they satisfy the Jacobi identity.
To leading-order in h,
however, there are no ambiguities. All nonlinear terms in the
Poisson structure that would require ordering are of higher order and can be ignored. The
leading terms reproduce a fully gauged KacMoody algebra, as we shall demonstrate.
525
The canonical two-form as derived from (79), denoted , consists of a leading part
and an order-h part:
= + h(
A A ),
i = h
i A ,
ij = ij = h (i A j j A i ).
(93)
The expansion of = 1 in h is complicated by the fact that, to leading order, ij vanishes and so the h 0 part of is singular. To overcome this,
we temporarily change the
scale of the spin coordinates by incorporating a factor of h in each i , which has the
effect
i h 2 i ,
1
ij h 1 ij .
(94)
The rescaled becomes

= + h(
A A ),
1
i = h 2 i A ,
ij = ij = i A j j A i .
(95)
1
2
This is an order h perturbation over a nonsingular form . We can calculate the

inverse in the standard expansion
+
= +
(96)
( = 1 ). The result to order h is

= h F ,
i = h 2 j i j A ,
ij = ij + h ik j l k A l A .
1
(97)
We see that we now have a nonvanishing i . Finally, we may restore the original scale of
1
the spin coordinates, which amounts to i h 2 i , ij h 1 ij . We also revert to
ij
the original spin Poisson structure, = , = h 1 ij . The final result is
= h F ,
i = h j i j A ,
ij = ij + h 2 ik j l k A l A .
Similarly, the determinant det = (det )1 will receive corrections according to

2
1
1
det = det 1 + tr( ) tr( )2 + tr( ) +
2
2
(98)
(99)
and these will be of higher order in h.
We may now use the new expressions (98) in the Poisson brackets for the boundary field
(22). The new terms i appear with derivatives acting on R or . Such terms, acting on a
526
function g (=R or ) create new terms in the Poisson brackets of the form
{A , g} .
i i g = h j i j A i g = h
(100)
Upon quantization of the spin space, h {A , g} i[A , g] and the above terms become
commutators. Combined with the corresponding term they give

g + i i g = g i[A , g] = D g.
(101)
Their net effect is to gauge all the derivatives appearing in the Poisson brackets. This is the
leading change in h . Other terms will produce higher order effects. For instance, the new
term in ij will produce the term
[R, A ][A , ].
(102)
Although this does not involve explicit factors of h , upon putting R = ro + h

the contribution of the scalar leading term Ro vanishes and the above term is of order h.
Altogether and obtain

A
o+
0
B
o+ o+ ro DAB ( ) + f ABC C ( ) , (103)
(1 ), (2 ) = 2
h o
where DAB is the adjoint expression of the covariant derivative D
DAB = AB f ABC AC
and D ro = ro since ro is a scalar. Similarly, the Hamiltonian obtains as

1
d
2
2
H = h o tr uo A0 .
2
o
(104)
(105)
We may define as before a derivative along the direction of classical motion on the boundary and the corresponding covariant version

D = oD o ro D .
(106)
In terms of D the equation of motion becomes
D0 D (uo ) = F0
(107)
where
F0 = [D0 , D ].
Similarly, the Poisson brackets of become

A
o+ AB
B
D ( )( ) + f ABC C ( ) .
(1 ), (2 ) = 2
h o
(108)
(109)
We obtain a gauged KacMoody algebra and corresponding equation of motion for the
chiral current = J+ . As before, this structure derives from the action of a fully gauged
WessZuminoWitten model in the space defined by classical motion trajectories and time,
527
integrated over the remaining phase space variables. In terms of a unitary field U we have
= iU 1 D U
(110)
and the action is

1 1
dt d
2
1
S = h o
tr U (D0 uo D )U U D U
2
o

2

h o
+
ok1 tr AU 1 dU + A dU U 1 + AU 1 AU
(k 1)!
D1
h2
o
+
(k 1)!
3
1 k1
1
tr U dU
(111)
with D = 2k. The first term is the gauged kinetic term on the boundary. The last term is the
standard WessZumino term; it does not involve gauge fields and is obtained by integrating
the WessZumino form over the bulk of the droplet with an appropriately extended unitary
field U as in Section 3.3. The middle term is defined on the boundary of the droplet and
involves the gauge fields A0 and A ; it is needed to absorb the gauge noninvariance of the
WessZumino term and contributes the term F0 in the equation of motion for (107).
Overall, we have recovered the action of [11] for a fully general gauge field, generalized
to an arbitrary phase space droplet and with a -dependent potential gradient uo .
5. Conclusions and discussion

We have presented an analysis of the phase space dynamics of droplets representing
fermions with internal degrees of freedom in an arbitrary phase space and derived their
Hamiltonian and canonical structure.
To leading order in h we recovered the WZW chiral action of edge excitations. In the
nonlinear theory we do not have an explicit form for the action. This is not crucial, since
we have derived the complete Hamiltonian dynamics, but it remains an issue for further
investigation, especially if we are interested in applying path-integral or effective field
theory techniques.
The nature of the obtained theories is halfway between classical and quantum: spin is
quantized and gives rise to a matrix structure, while phase space coordinates are still treated
classically. As such, it is reminiscent of the matrix formulation of the quantum Hall effect
[15]. The exact correspondence between the two formulations, if any, should be further
examined.
Finally, the theories derived in this paper represent a non-Abelian phase space bosonization of the fermionic systems they describe. Just as in the Abelian case, however, this
bosonization fails quantum mechanically in dimensions higher than D = 2. The main problems are, first, that this theory overestimates the degrees of freedom of the system, due to
the infinity of excitations normal to the direction of propagation and, second, that the theory is essentially local in phase space and thus does not take into account processes where
528
fermions would undergo transitions to faraway phase space states. This is an issue that will
be treated in an upcoming publication.
Acknowledgements
I would like to thank D. Karabali and V.P. Nair for useful comments on the manuscript.
This research was supported in part by the National Science Foundation under grant PHY0353301 and by the CUNY Research Foundation under grant PSC-CUNY-66565-0035.
References
[1] S.R. Coleman, Quantum sine-Gordon equation as the massive Thirring model, Phys. Rev. D 11 (1975) 2088;
S. Mandelstam, Soliton operators for the quantized sine-Gordon equation, Phys. Rep. 23 (1976) 307.
[2] A. Jevicki, B. Sakita, The quantum collective field method and its application to the planar limit, Nucl. Phys.
B 165 (1980) 511;
S.R. Das, A. Jevicki, String field theory and physical interpretation of D = 1 strings, Mod. Phys. Lett. A 5
(1990) 1639.
[3] E. Witten, Non-Abelian bosonization in two dimensions, Commun. Math. Phys. 92 (1984) 455.
[4] G.W. Semenoff, Canonical quantum field theory with exotic statistics, Phys. Rev. Lett. 61 (1988) 517;
G.W. Semenoff, P. Sodano, Exotic spin and statistics in (2 + 1)-dimensional canonical quantum field theory,
Nucl. Phys. B 328 (1989) 753;
T. Matsuyama, Canonical structures and BoseFermi transmutations in (2 + 1)-dimensional u(1) gauge
theories, Phys. Lett. B 228 (1989) 99.
[5] J. Polchinski, Classical limit of (1 + 1)-dimensional string theory, Nucl. Phys. B 362 (1991) 125.
[6] A.P. Polychronakos, Chiral actions from phase space (quantum Hall) droplets, Nucl. Phys. B 705 (2005)
457, hep-th/0408194.
[7] D. Karabali, V.P. Nair, Quantum Hall effect in higher dimensions, Nucl. Phys. B 641 (2002) 533, hepth/0203264;
D. Karabali, V.P. Nair, The effective action for edge states in higher-dimensional quantum Hall systems,
Nucl. Phys. B 679 (2004) 427, hep-th/0307281.
[8] S.C. Zhang, J.P. Hu, A four-dimensional generalization of the quantum Hall effect, Science 294 (2001) 823,
cond-mat/0110572;
S.C. Zhang, J.P. Hu, Phys. Rev. B 41 (1990) 12838, cond-mat/0112432;
B.A. Bernevig, C.H. Chern, J.P. Hu, N. Toumbas, S.C. Zhang, Effective field theory description of the
higher-dimensional quantum Hall liquid, Ann. Phys. 300 (2002) 185, cond-mat/0206164.
[9] O. Agam, E. Bettelheim, P. Wiegmann, A. Zabrodin, Viscous fingering and a shape of an electronic droplet
in the quantum Hall regime, Phys. Rev. Lett. 88 (2002) 236802, cond-mat/0111333, and references therein.
[10] B. Sakita, Collective variables of fermions and bosonization, Phys. Lett. B 387 (1996) 118, hep-th/9607047;
R. Ray, B. Sakita, Bulk and edge excitations of a = 1 Hall ferromagnet, cond-mat/0105626.
[11] D. Karabali, V.P. Nair, Edge states for quantum Hall droplets in higher dimensions and a generalized WZW
model, hep-th/0403111.
[12] R. Jackiw, V.P. Nair, S.Y. Pi, A.P. Polychronakos, Perfect fluid theory and its extensions, J. Phys. A 37
(2004) R327, hep-ph/0407101.
[13] V.P. Nair, J. Schiff, A KhlerChernSimons theory and quantization of instanton moduli spaces, Phys. Lett.
B 246 (1990) 423;
V.P. Nair, J. Schiff, KhlerChernSimons theory and symmetries of antiselfdual gauge fields, Nucl. Phys.
B 371 (1992) 329.
[14] A.P. Balachandran, G. Marmo, B.-S. Skagerstam, A. Stern, Gauge Symmetries and Fibre Bundles, SpringerVerlag, Berlin, 1983;
529
C.H. Chou, V.P. Nair, A.P. Polychronakos, On the electromagnetic interactions of anyons, Phys. Lett. B 304
(1993) 105, hep-th/9301037;
B. Bistrovic, R. Jackiw, H. Li, V.P. Nair, S.Y. Pi, Non-Abelian fluid dynamics in Lagrangian formulation,
Phys. Rev. D 67 (2003) 025013, hep-th/0210143.
[15] L. Susskind, The quantum Hall fluid and non-commutative ChernSimons theory, hep-th/0101029;
A.P. Polychronakos, Quantum Hall states as matrix ChernSimons theory, JHEP 0104 (2001) 011, hepth/0103013;
A.P. Polychronakos, Quantum Hall states on the cylinder as unitary matrix ChernSimons theory, JHEP 0106
(2001) 070, hep-th/0106011.
Scaling behavior of tethered crumpled manifolds

with inner dimension close to D = 2:
Resumming the perturbation theory
H.A. Pinnow a , K.J. Wiese a,b
a Fachbereich Physik, Universitt Essen, 45117 Essen, Germany
b Laboratoire de Physique Thorique, Ecole Normale Suprieure, 24 rue Lhomond, 75005 Paris, France
Received 5 April 2004; received in revised form 8 December 2004; accepted 6 January 2005
Abstract
The field theory of self-avoiding tethered membranes still poses major challenges. In this article,
we report progress on the toy-model of a manifold repelled by a single point. Our approach allows the
summation of the perturbation expansion in the strength g0 of the interaction exactly in the limit of
internal dimension D 2, yielding an analytic solution for the strong-coupling limit. This analytic
solution is the starting point for an expansion in 2 D, which aims to interpolate to the well studied
case of polymers (D = 1). We give results to fourth order in 2 D, where the dependence on g0
is again summed exactly. As an application, we discuss plaquette density functions, and propose a
Monte Carlo experiment to test our results. These methods shed light on the more complex problem
of self-avoiding manifolds.
PACS: 68.35.Md; 05.40.-a; 11.10.-z
Keywords: Polymer; Polymerized membrane; Renormalization group; Exact resummation
E-mail addresses: hpinnow@sinits.com (H.A. Pinnow), wiese@lpt.ens.fr (K.J. Wiese).

doi:10.1016/j.nuclphysb.2005.01.010
H.A. Pinnow, K.J. Wiese / Nuclear Physics B 711 [FS] (2005) 530564
531
1. Introduction
One major problem in statistical physics is the effect of interactions on the thermodynamical properties of extended fluctuating geometric objects. In general, multi-particle
attractive or repulsive interactions are involved. One may divide these into two classes:
(i) The interaction of a single fluctuating object with itself: a well known example is the
excluded volume interaction between any two monomers within a long polymer chain in a
good solvent, which results in the anomalous scaling of the mean squared end-to-end distance. (ii) The interaction between different manifolds or between a single manifold and a
fixed non-fluctuating object. Thermal fluctuations then affect the depinning of the manifold
from an attractive substrate as well as the steric repulsions from a wall. Finally, cases (i)
and (ii) can appear together.
Whatever the situation is, it is usually well understood as long as the fluctuating objects
are one-dimensional [14]. Referring to the example above, the long-distance properties
of self-avoiding polymers can be analyzed with renormalization group techniques [57],
either in the continuous Edwards Hamiltonian [8],

2 b0

1
H[r ] =
(1.1)
r (x) +
d r(x) r(y) ,
2
2
xM
xM yM
or by mapping this model on a local O(N ) symmetric 4 -theory in the limit of N = 0 components [1,3,9]. The critical exponents describing the long-distance properties are related
to the critical exponents of the corresponding N -vector model at the critical point. What
makes (1.1) a non-standard theory is that the interaction is non-local, and not a polynomial
of the field.
Obtaining the corresponding results for membranes poses considerable challenges. The
generalization of polymers to 2D-surfaces are crystalline fixed-connectivity membranes as
they appear, for instance, in the spectrin network of cell membranes. Considering phantom membranes which can freely fold into themselves, the existence of a bending rigidity
induced phase transition separating a high rigidity, low temperature flat phase from a low
rigidity, high temperature crumpled phase is well established [1015]. This in contrast to
polymers, which are always crumpled on large scales. The scaling properties of the crumpled phase of phantom membranes are described by the 2D generalization of the free field
part in (1.1). Taking self-avoidance into account, which is modeled in (1.1) through the
short-range two-body interaction, we expect more swollen manifolds than those predicted
by the free theory. This is expressed in a non-trivial radius of gyration exponent :
Rg L ,
0 1,
(1.2)
where L denotes the linear internal size of the membrane, and the radius of gyration Rg
is obtained from the effective extend of the membrane in external space. In the case of
polymers Rg scales like the end-to-end distance. Much effort has been spent on calculating
corrections to the radius of gyration exponent within an expansion in the deviation from
the critical space dimension [16,17]. These calculations cannot be performed directly for
the membrane-dimension D = 2, since the naive scaling dimension of the coupling in (1.1)
532
equals
(D, d) := [b0 ] = 2D
2D
d,
2
(1.3)
where d denotes the dimension of the embedding space, such that is always non-zero as
D 2, for any embedding dimension d. Equivalently, the critical embedding dimension
defined through (D, dc (D)) = 0 becomes infinite in this limit. The reason is that the nonself-avoiding membrane densely fills out the embedding space, such that it always sees
the interaction. A way to circumvent this problem is to set up the expansion about any
point (D < 2, dc (D)) and to extrapolate along an appropriate path in the (D, d)-plane to
the physically interesting point (D, d) = (2, 3) [1824]. To second order in the radius of
gyration exponent is then found to be 0.86 [16,17]. This is a strong correction with
respect to the only logarithmic dependence in the non-interacting theory, and indicates the
existence of a crumpled phase, for which 2/3 follows from the fact that a membrane
has a finite volume.
However, there is no evidence for a crumpled phase in experiments [2528]. Latest
Monte Carlo simulations on plaquette-models [31,32] starting from a discretization of the
2D generalized Hamiltonian (1.1) with system sizes of up to 17000 plaquettes show
considerable evidence for a vanishing of the crumpling transition as soon as self-avoidance
is switched on, in contrast to the earlier references [29,30]. Even on large scales fixedconnectivity membranes seem to stay flat with a radius of gyration exponent of 1. It
is however not clear whether any of the existing simulations is large enough to settle the
problem.
The final goal is to develop techniques, which allow to go beyond the two-loop result.
So far, we developed such techniques for a simplified model, which reduces the non-local
self-avoiding interaction in (1.1) to self-avoidance with only a single point, e.g., the origin o
in the membrane:

2

1
H[r ] =
(1.4)
r (x) + g0
d r(x) r(o) .
2
xM
xM
This is a special case of a phantom tethered manifold interacting with a single point in
embedding space and which is related to case (ii). The corresponding physical situation is
the binding and unbinding of a long chain as, e.g., a polymer or a membrane from a wall
or the wetting of an interface. More precisely, we study the interaction of a single freely
fluctuating manifold with another non-fluctuating, fixed object. Depending on whether the
interaction is attractive or repulsive, one may observe two different scenarios: either the
manifold delocalizes from an attractive substrate as in wetting phenomena or it is sterically
repelled from a fixed object (a wall). Both cases have in common that excluded volume
effects become important. We already discussed these scenarios in [33].
In [33] we performed a complete resummation of the perturbation series for the effective
coupling in the case of 2D-membranes. The long-distance behavior of the resummed theory then turned out to be non-trivial in the sense that it emerged from the limiting behavior
of a scale invariant theory. As a result of this the effective coupling grows logarithmically
instead of approaching some finite fixed-point value as one would expect. This and the
533
extremely slow convergence of the perturbation series makes the analysis of the fully resummed theory indispensable: all finite loop calculations fail to extract the correct large
distance properties. The importance of this result becomes evident as soon as one compares it with extrapolations obtained from the -expansion at the 2-loop level [33]. The
latter not only required the numerical calculation of diagrams, with considerably raising
effort as the loop order becomes higher, but also turned out to be unable to make reliable
predictions for D 2. This problem persisted, though we exploited the freedom to set up
2D
the expansion about any point (D < 2, dc (D)), dc (D) = 2D
being the critical embedding
dimension for given internal dimension D and to expand both in D and d along any appropriate extrapolation path to some physically interesting point (D = 2, d). As soon as D
was approaching 2, the result became strongly dependent on the selected expansion point.
The aim of this paper is two-fold: first, we reconsider the techniques to perform loop
calculations within a massive scheme, that is on a manifold of finite size and with fixed
space dimension 0 < D < 2 and d. We show that the perturbation series of the effective
coupling can be completely summed up in D = 2, and analyze the long-distance properties
in this limit. In addition to [33], instead of analytically continuing loop integrals to D = 2
from below we perform calculations directly in 2D, which need an explicit short-distance
(UV)-cutoff. It turns out that results in D = 2 are independent of the procedure, i.e., they
are universal.
Second, we construct a systematic expansion of the effective coupling in powers of
2 D. It is based on our techniques to resum the perturbation series at each order in 2 D.
A first attempt to go beyond D = 2 has already been made in [33]. However, there, the
effect of the boundaries of the finite manifolds was not taken properly into account, a problem that has now been circumvented by considering closed manifolds. We specialize to a
toroidal internal topology, which corresponds to imposing periodic boundary conditions.
Of course, the propagator of the perturbation series needs to be modified, and diagrams
become more difficult to calculate. Slightly below D = 2 we expect power-law behavior
of the effective coupling. We present a possible ansatz for the exact effective coupling as a
function of the internal dimension D 2, which is consistent with the expansion in 2 D.
However, it remains an open problem to extract more information about the power-law
behavior in order to make this expansion unique.
A short account of this work has already appeared in [34].
2. Model and physical observables

2.1. The model
The problem of a membrane avoiding only a single point (1.4) may appear artificial.
However, it not only provides a toy-model for the analysis of the full self-avoidance problem, but also specializes case (ii). We consider a phantom tethered membrane interacting
with some -potential located at the origin of the configuration space: the Hamiltonian is
534
given by (0 D 2)

2
1
r (x) + g0
H[r ] =
2
xM

d r(x) ,
(2.1)
xM
where any point in the membrane is labeled by some D-component vector x, and its position in external space is given by the d-component field r(x),
r : x RD r(x) Rd .
(2.2)
The partition function is defined as

Z = D[r ] exp H[r ] .
(2.3)
To remove the translational 0-mode, we will consider

Z = D[r ] r(y) exp H[r ] .
(2.4)
Let us discuss (2.1) in more detail: the first term is the elastic energy of the manifold which
is entropic in origin. Elasticity and temperature have been scaled to unity. The second term
models the interaction of the manifold with a single point at the origin in the d-dimensional
configurational space. The physical interpretation [33,35] depends on the dimensionality:
in the case that Rd is identical to the embedding space, (2.1) describes a phantom crumpled
manifold interacting with a single defect as sketched in Fig. 1. However, setting d = 1 (2.1)
may as well describe a solid-on-solid like fluctuating interface parameterized by some
displacement field and interacting with a parallel plane (D = 2) as shown in Fig. 1.
The coupling constant g0 may either be positive (repulsive interaction) or negative (attractive interaction). We now give the dimensional analysis: in internal space units, the
engineering dimensions are
2D
,
dim[x] = 1,
:= dim[r ] =
2

:= dim
dD x d r(x) = D d.
(2.5)
(a)
(b)
Fig. 1. (a) A D-dimensional manifold (D = 2) interacting with a point in the origin of the configurational
space Rd . (b) A directed membrane (interface) interacting with a parallel subspace of same dimension D.
535
The interaction is naively relevant for > 0, i.e., d < dc with (see Fig. 2)
2D
,
(2.6)
2D
irrelevant for < 0 and marginal for = 0. It has been shown [35,36] that the model is
renormalizable for 0 < D < 2 and 0. Results for negative are obtained via analytical
continuation. One can define the renormalized coupling g as
dc =

N
(2.7)
Z(0) Z(g0 ) L ,
VM
where VM denotes the internal volume of the manifold. The normalization N depends on
the definition of the path-integral (but not on L) and is chosen such that

g = g0 L + O g02 .
(2.8)
g :=
Universal quantities emerge at fixed-points of the -function, which is defined as

g
(g) := L .
L g0
(2.9)
The -function describes, how the effective coupling g changes under scale transformations, while keeping the bare coupling g0 fixed. Let us state the 1-loop result, see, e.g., [33,
35,36]: it reads

1
(g) = g + g 2 + O g 3 ,
(2.10)
2
where g is the dimensionless renormalized coupling. Apart from the trivial solution, g = 0,
the flow equation given by (2.9) and (2.10) has a non-trivial fixed point at the zero of the
-function

g = 2 + O 2 .
(2.11)
We will show below that the scaling behavior is encoded in the slope of the RG-function at
the fixed point, which is universal as a consequence of renormalizabilty. The long-distance
2D . The interaction is relevant for points that lie above

Fig. 2. Critical line defined through = 0 dc (D) = 2D
that line.
536
(a)
(b)
(c)
Fig. 3. RG-function and flow for increasing manifold size L for the dimensionless renormalized coupling g: (a) in
the case > 0; (b) in the case < 0; (c) in the case = 0.
behavior is then governed by the -interaction as considered in our model (2.1), which is
the most relevant operator at large scales. Let us now discuss possible physical situations
(see Fig. 3):
(a) > 0: The RG-flow has an infrared stable fixed point at g > 0 and an IR-unstable
fixed point at g = 0. The latter corresponds to an unbinding transition whose critical
properties are given by the non-interacting system, while the non-trivial IR stable fixed
point determines the long-distance properties of the delocalized state, the long-range
repulsive force exerted by the fluctuating manifold on the originwhich we remind
may be a point, a line or a plane.
(b) < 0: Now, the long-distance behavior is Gaussian, while the unbinding transition
occurs at some finite value of the attractive potential, g < 0, which corresponds to
an infrared unstable fixed point of the -function. Below g the RG-flow is to strong
coupling and the manifold is always attracted.
(c) = 0: This is the marginal situation, where the transition takes place at g = 0; we
expect logarithmic corrections to scaling.
We discussed these scenarios and possible observables already in [33]. Here we specialize
to membranes avoiding a single point. It turns out that this situation allows to calculate
observables staying non-singular even for 2D membranes and which can be measured in a
Monte Carlo experiment.
2.2. Plaquettes-density correlation functions
Interesting physical observables for a membrane avoiding a single point are the
plaquettes-density functions at the repelling point. Generally, these are defined as follows:

d
r(xi ) ,
n :=
(2.12)
i=1 x M
i
where the expectation value

is taken within the pinned ensemble as defined in (2.4).
The quantity, which is accessible to perturbation theory, is the effective coupling as defined
in (2.7). It is a generating function for observables like (2.12). Let us first show how to obtain the constrained partition function (2.4) from (2.7): since we consider closed manifolds,
internal translational invariance implies

1
1 (gL )

Z Z (g0 ) =
Z(g0 ) =
,
VM g0 L
N g0 L
537
(2.13)
where g is the renormalized or effective coupling defined in (2.7) and VM denotes the
internal volume of the membrane. Introducing the dimensionless bare coupling,
z := g0 L ,
(2.14)
(2.13) can be written in terms of dimensionless quantities as

g
Z g 0 L =
(2.15)
,
z
where N has been set to unity. In the same way, all observables of the type (2.12) can be
easily derived from g according to:

L g
.
n =
(2.16)
g/z z z
Observables, which are to be measured in a Monte Carlo simulations, should be universal:
the -function written in terms of the bare coupling reads
g
(2.17)
.
z
(Note that in a slight abuse of notation, we write (z) = (g(z)).) The universal slope at
the fixed point, which is defined as

(g)
,
:=
(2.18)
g g
(z) = z
is obtained from
z (z)
(z) =
(z) z
(2.19)
in the limit z . We furthermore need the second derivative of the RG-flow function
with respect to the effective coupling, which is defined as

2 (g) z z (z)

.
=
:=
(2.20)
(z) z
g 2 g
Let us now show that the universal slope (2.18) can be obtained from the measurement
of appropriate combinations of observables of type (2.12). For this purpose we need the
plaquettes-density ( = 1) and the densitydensity function ( = 2), which are obtained
after some straightforward, but tedious algebra from the above definitions:

(z) z 1
1
1+
1+
,
n =
g0
g0

2
3(z) 2 (z) (z)(z) z 1

3 2
1
n = 2 2+
+ 2 +
+ 2 . (2.21)
2 2 +
g0
g0
538
These quantities depend on the bare coupling g0 , which is not accessible. Instead, consider
the following ratio:

n z 1 + /

,
(2.22)
=
2 + /
n2
which obviously is universal.
2.3. Delocalization transition
For completeness let us shortly discuss the physical situation at the UV-stable fixed point
in Fig. 3. The fixed point corresponds to a delocalization transition of the manifold, which
is at vanishing coupling g = 0 for > 0 and at some finite attractive coupling g < 0 for
< 0.
In the localized phase g < g , correlation functions such as
[r (x) r(y)]2 and the
associated correlation length (in the D-dimensional internal space) should be finite,
as well as the radius of gyration . Approaching the transition these quantities diverge
as [37]

g g .
g g ,
(2.23)
Since , the exponents and are related through
= ,
(2.24)
being the dimension of the field (2.5).

Furthermore, they are related to the correction-to-scaling exponent :
=
1
,
(g )
.
(g )
(2.25)
Note that (g ) < 0 at the transition. Specializing to (D, d) = (1, 1), we find
= 1,
= 2.
(2.26)
These exponents are also valid for the delocalization transition of a 1-dimensional interface from an attractive hard wall in 2-dimensional bulk space [33,3739].
3. Complete summation of the perturbation series

3.1. Perturbation theory
In (2.2) we saw that physical observables can be derived from the renormalized coupling
g (2.7). To obtain g we need the perturbation series of the partition function Z (2.3):
Z=

(g0 )N +1
ZN ,
(N + 1)!
N =1
(3.1)
where
ZN =
N +1

d r(xi ) ,
i=1 xi
N 0,
539
(3.2)
and the normalization of the -distribution has been chosen to be

d r(x) := (4)d/2 r(x) = eikr(x)
(3.3)
with

:=
d/2

dd k.
(3.4)
The advantage of these normalizations is that

2
ek = 1.
(3.5)
Accordingly, the perturbation expansion of the effective coupling (2.7) reads

N +1

g0 L (g0 )N
d
r(xi ) .
g(z) =
VM
(N + 1)!
N =0
i=1 xi
(3.6)
Performing the averages within the Gaussian theory with normalization

d
1
r(x) 0 = 1,
VM
(3.7)
one arrives at

N +1
N +1 N +1

1
g0 L (g0 )N
d
ki exp
ki kj C(xi xj ) ,
g(z) =
VM
(N + 1)!
2
N =0
i=1 k xi
i
i=1
i,j =1
(3.8)
where
2
1
r(xi ) r(xj ) 0
(3.9)
2d

denotes the correlator, and the d ( i ki ) stems from the integration over the global translation. Shifting
C(xi xj ) :=
kN +1 kN +1
N

ki ,
i=1
the quadratic form in (3.8) transforms to
(3.10)
540
N +1
1
ki kj C(xi xj )
2
i,j =1
N

kN +1 kj C(xN +1 xj )
j =1
N

C(xN +1 xi ) + C(xN +1 xj ) C(xi xj )

.
2
ki kj
i,j =1
Integrating out the momenta k1 , . . . , kN +1 in (3.8), one obtains

N

(z)N
g(z) = z
(det D)d/2 ,
(N + 1)!
N =0
(3.11)
(3.12)
=1 x
from the loop integration (such that the integrals now run
where we have factored out
over a torus of size 1), and the matrix elements Dij are
Dij =

1
C(xN +1 xi ) + C(xN +1 xj ) C(xi xj ) .
2
(3.13)
3.2. Complete summation in fixed internal space dimension D = 2

Let us compute the N -loop order of (3.12): the behavior of the propagator C(x) for
arguments x large compared to a is of the form
x
1
(3.14)
ln ,
2 a
where c0 denotes some positive constant (note C(x) 0), and the logarithmic growth (for
large x) is universal (see Appendix A). In D = 2 we need an additional short-distance
cutoff a, which we want to take to 0. We can (somehow arbitrary) decompose
N

det D =
(3.15)
Dii det D.
C(x) = c0 +
i=1
1
In the limit of a 0 each C(x) = 2
ln(L/a) + O(a 0 ), such that

a0 1
ij = 1 1 + C(xN +1 xj ) C(xi xj )
, i = j,
D
2
C(xN +1 xi )
2
ii = 1.
D
N

=1 x

d/2
(det D)

L
N L
(0) d/2 .
= I1
det D
=: IN
a
a
(3.16)
(3.17)
(0) denotes the limit a 0 of (3.16). It can be written as D

(0) = 1 (I +
The matrix D
2
NP), where I denotes the identity and P the projector onto (1, 1, . . . , 1), whose image
541
a0
(0) = 1+N
has dimension 1, such that det D
[33]. Furthermore, to one loop I1 (L/a) =
2N
c1 (ln La )d/2 , where c1 denotes some (finite) constant. One then arrives at
g(z) = z

(z(ln La )d/2 )N
.
N !(1 + N )d/2+1
(3.18)
N =0
A factor c1 2d/2 has been absorbed into a rescaling of both z and g.

3.3. Asymptotic scaling behavior
In the following we will analyze the limit of large z (strong repulsion), which also is the
scaling behavior of infinitely large membranes. We need an analytical expression for sums
like (3.18) in the limit of large z. Later, it will turn out that allowing for small deviations
2 D > 0 only slightly more general sums will arise.
We claim that for all k, d > 0
(z)N
1
= d
d/2
N!(k + N )
2
N =0

dr r d/21 exp zer kr .
(3.19)
This can be proven as follows:
1
d
2

dr r d/21 exp zer kr
1 (z)N
= d
dr r d/21 e(N +k)r
N!
2
N =0

d2
1 (z)N
= d
.
N! (N + k)d/2
2
N =0
This integral-representation is not the most practical for our purpose. It is better to set
r s := er which yields
(z)N
1
=
N!(k + N )d/2 d2
N =0
1
ds s k1 ( ln s)d/21 esz .
(3.20)
This formula is already very useful for some purposes. It is still advantageous to make a
second variable-transformation s y := sz, yielding
(z)N
(ln z)d/21
=

N!(k + N )d/2
d2 zk
N =0
z
dy y
0
k1
ln y
1
ln z
d/21
ey .
(3.21)
542
Finally we remark that we usually have the following combination

ln y d/21 y
dy y k1 1
e .
ln z
0
(3.22)
It satisfies the following simple recursion relation, which is helpful to calculate the
-function:
fkd (z) := zk
(z)N
(ln z)d/21
=

N !(k + N )d/2
d2
N =0
z
d d
f (z) = fkd2 (z).
dz k
The derivative above can be rewritten as
z
(3.23)
d d
d
(3.24)
(z),
f (z) = kfkd (z) fk+1
dz k
such that one obtains a useful formula in order to isolate the dominant behavior for large z:
z
d
fk+1
(z) = kfkd (z) fkd2 (z).
(3.25)
From (3.19) fkd (z) > 0 for all k, d > 0 and the behavior for large z is obtained by expanding (1 ln y/ ln z)d/21 for small 1/ ln z

(ln z)d/21
1 d
d
k1 y
1
dy y e
dy y k1 ln yey
fk (z) =

ln z 2
d2
0
0

1
+O
(3.26)
+ O ez .
(ln z)2
The result is
fkd (z) =

1 d 2 (k)
(ln z)d/21 (k)
+ .
1
d
ln z 2 (k)
2
With the above notations, the sum (3.18) expressing g as a function of z becomes

L
L d/2
L d/2 d+2
g z,
z ln
= ln
f1
a
a
a
(3.27)
(3.28)
in the limit D = 2.
It is now easy to analyze the long-distance behavior in this limit. First, we observe
that according to (3.27) the effective coupling diverges logarithmically for all external
dimensions d > 0:
d/2

L d/2 d/2
L z ln La
d+2 ln z ln
.
g z,
(3.29)
a
a
2
This is in contrast to the one-loop result as stated in (2.11), which is exact for polymers
(D = 1) and which stays qualitatively valid as long as D < 2. This follows from the renormalizability of the theory [35] for sufficiently small > 0. A finite limit g(z ) = g
543
signals a scale invariant theory. In (3.29) we have found the limiting behavior of the latter.
Consequently, we expect the correction-to-scaling exponent to be always zero in D = 2.
In order to check that let us first compute the renormalization -function in terms of the
bare coupling as in (2.17), which can be immediately derived with the help of relation
(3.23):1
(z ) = z
d/21
g
1
z
= f1d (z ) d ln(z )
,

z
2
(3.30)
where we have introduced rescaled couplings z := g0 L (ln La )d/2 and g = g(ln La )d/2 .
Its derivative with respect to the renormalized coupling is found as a function of the bare
coupling (2.19) to be
(z ) =
z dzd f1d (z ) z 2 d z

z (z )
=

0.
(z ) z
2 ln(z )
f1d (z )
(3.31)
Note that the qualitative behavior of the -function changes depending on the external
dimension d, approaching asymptotically zero below d = 2 and being divergent above.
In the limit of large bare couplings one may as well give the RG-function in terms of the
effective (renormalized) coupling simply by inverting the asymptotic expression in (3.29)
and inserting it into (3.30), with the result:
d+2 12/d
2
(g)
g 12/d .

d2
z
(3.32)
It is interesting to compare the true asymptotic behavior of the completely resummed perturbation series as found above with predictions taking only finite loop orders into account:
if one tries to invert (3.18) and truncates it at some finite order, it is at least possible to reach
the asymptotic regime (3.32)however, for large g the truncated -function does not converge to the true -function and thus strongly deviates from the true behavior. In Fig. 4
the Pad-resummed truncated -function up to order g 160 in d = 1 is compared with the
asymptotic flow-function. One notices that the truncated -function even though improved
through a Pad-resummation hardly gets into touch with the asymptotic regime. The same
applies to the slope-function (g), which is not shown in Fig. 4. Let us finally state the
expected behavior of the plaquettes-density functions in the limit of large membranes. For
the plaquettes-density at the repelling fixed-point we find in this limit:

1
2 d z 1
n =
(3.33)
.
1+
g0
2 ln z
g0

1 Note that our definition (z ) = z g is strictly speaking equivalent to defining the -function as
z
d a d )| g, instead of (2.9). (Note that the derivative w.r.t. a disappears for D < 2.) The
(g) := (L dL
da g0
d/2 instead of z = g L , and normalizations such that
natural combination in D = 2 is z = g0 L (ln L
0
a)

2
g (z ) = z + O(z ) does not explicitly depend on L or a. The chosen definitions avoid unnecessary techni-
cal complications, but do not change the physics of the problem.
544
Fig. 4. -function in terms of the renormalized coupling g truncated at order 160, Pad-resummed, and plotting
only that part for which the truncated series converges. (This can, e.g., be tested by taking away the last few terms
of the series.) This is compared to the asymptotic behavior (3.32) (proportional to 1/g for large g). d is set to
1, and we used the diagonal (80, 80)-Pad approximant, which was find to converge best. (The non-resummed
expression starts to diverge already at g 1.8 at this order.)
Note that in the absence of the repelling interaction this quantity would diverge in this
limit. This follows from dimensional grounds, since then
n L .
(3.34)
In (3.33) we found the largest possible depopulation of monomers at the defect potential in
the case of a relevant interaction ( > 0). As we discussed in (2.2) a measurable quantity
should be the following ratio (2.22), which in the case of 2D-membranes becomes in the
limit z :

n z 1

=
,
(3.35)
2
n2
which can be compared with the 1-loop prediction (which is exact for polymers):

n z 2

, (1-loop).
=
3
n2
(3.36)
4. Crossover to polymers
Let us now analyze the theory below D = 2. Due to the renormalizability in 0 < D < 2
and the existence of an -expansion we expect the renormalized coupling to reach a finite
fixed point in the strong coupling limit as soon as D < 2. This approach is characterized
by a power-law decay of the form

g(z) = g + S(ln z)z/ + O z1 / ,
(4.1)
where S is some scaling-function growing at most sub-exponentially and 1 > > 0, with
defined in (2.18).
545
Our ultimate aim is to extract information from an expansion in powers of 2 D of

the effective coupling about the correction-to-scaling exponent in (4.1) for D 2. The
scale invariant behavior below D = 2 results in a finite fixed point of the renormalization -function as a function of the effective coupling. The qualitative behavior of the
-function is sketched in Fig. 5.
4.1. (2 D)-expansion on the torus
In order to gain information about g below D = 2 one has to expand the loop integrand
(det D)d/2 (3.12) in powers of 2 D. For convenience, we take a 0. The propagator
takes in infinite D-space the form C(x) = |x|2D /(SD (2D)), where SD = 2 D/2 / ( D2 )
denotes the volume of the D-dimensional unit-sphere. The factor (SD (2 D))1 replaces
ln( La ) and is absorbed into a rescaling of the field and the coupling according to r
r(SD (2 D))1/2 and g0 g0 (SD (2 D))d/2 , such that the factors of (ln La )d/2 in (3.18)
and (3.29) disappear. The propagator in the rescaled variable can then be written as
C(x) = 1 + (2 D)C(x),
(4.2)
where for convenience of notation we allow C(x) to depend itself on D.

Of course, on a closed manifold of finite size, C(x) is modified, but the form (4.2) is
independent of the shape of the manifold. Accordingly, one may expand the matrix D as
(0) + (2 D)D,
D=D
(4.3)
(0)
where D is defined as before and coincides with the limit D 2 when inserting the
above C(x) into D. Moreover, D is of the same form as D, but each C(x) has been replaced
with C(x):

1
Dij = C(xN +1 xi ) + C(xN +1 xj ) + C(xi xj ) .
(4.4)
2
Then,

(0) 1
(0) exp Tr ln 1 + (2 D) D
det D = det D
(4.5)
D ,
(0) ]1 = 2(I
where [D
Denoting
1

M := D(0) D
N
N +1 P)
(0) .
denotes the inverse matrix of D
Fig. 5. Qualitative behavior for the -function in D = 1, D = 2 and result anticipated for D 1.5.
(4.6)
546
we expand the determinant in (4.5) up to fourth order in 2 D:

d/2
det(D)

(0) d/2
d
(2 D)2
1
(2 D) Tr M
Tr M2
= det D
2
2

(2 D)3
(2 D)4
Tr M3
Tr M4
+
2
4

2
d
+
(2 D)2 Tr2 M (2 D)3 Tr M Tr M2
8

2
4 1
2
2
3
Tr M + Tr M Tr M
+ (2 D)
4
3

3
d4
3
d
(2 D)3 Tr3 M (2 D)4 Tr2 M Tr M2 +
(2 D)4 Tr4 M
48
2
384

5
+ O (2 D) .
(4.7)
The first step in the analysis will be to obtain the resummed perturbation series of the
effective coupling up to fourth order in 2 D. That is, we have to insert (4.7) into (4.5),
calculate the corresponding loop integrals at each order of perturbation theory, insert the
result into (3.12) and sum the appearing series to all orders.
Let us start with the first-order term in 2 D from (4.7). We only need M = [D(0) ]1 D,
which reads

N
(0) 1
2
(Mij ) = D
(4.8)
D ij = 2Dij
Dik L2 .
1+N
k=1
The trace of (4.8) can easily be performed, with the result

N
N N
2N
2
Tr M =
Dii
(1 ik )Dik L2 .
1+N
1+N
i=1
(4.9)
i=1 k=1
In each order of perturbation theory we have to integrate the expression (4.7) over internal
distances. These integrals have to be regularized in the infrared through an appropriate IR
cut-off. We are considering a finite manifold of toroidal topology (Fig. 6). The precise form
of the correlator on the torus will only later enter into the calculation.
To simplify the calculations, we further introduce the following notation:

f (xi1 , . . . , xik ) := f (xi1 , . . . , xik )
(4.10)
x1
xN
with the internal integrations defined as

=L
,
:= integral over the torus with L = 1,
xM
(4.11)
547
Fig. 6. Regularization scheme for the N -loop diagrams on manifolds with toroidal topology (periodic boundary
conditions). Here: D = 2.
such that the overbar in (4.10) can be thought of as an averaging procedure, and especially
1 = 1.
(4.12)
Thanks to our regularization prescription the integral of (4.9) over internal points can be
replaced by LN D (for the integration measure) times

2N 2
2N (N 1)
1
Tr M =
C(xN +1 xi )
C(xN +1 xi ) C(xi xj )
1+N
1+N
2
N (N 1)
2N
C(xN +1 xi ) +
C(xi xj ).
=
(4.13)
1+N
1+N
Due to the internal symmetry of the closed manifolds which we consider the expression
above can be further simplified, since
C(xN +1 xi ) = C(xi xj ) C(x).
(4.14)
Introducing a diagrammatic notation

:= C(x),
the N -loop integral reads up to first order in 2 D

d/2

d/2
N 1 + N
(det D)
=
2N

d
2
+ O (2 D) .
1 (2 D) N
2
(4.15)
(4.16)
For the further analysis we will not only need (4.13), but also the terms appearing to higher
order in 2 D in (4.7). We derived expressions like (4.13) for Tr 2 M and Tr M 2 and all
548
terms up to fourth order in 2 D with a M ATHEMATICA -program. It is based on the fact

that all terms to appear in the expansion (4.7) are of the form Tr n M m or products of the
latter and therefore can be written as P(N )/(N + 1)k , where n, m, k N and P(N ) is some
polynomial in N . It will turn out soon that it is convenient to expand the polynomial P(N )
in terms of the following base:

k

1, N, N (N 1), N (N 1)(N 2), . . . ,
(4.17)
(N j ), . . . .
j =0
We obtain:
Tr M2 =
2N (N 1) N (N 1)(N 2)
C(x)2
1+N
2N + 3N (N 1) + N (N 1)(N 2) 2
+
C (x),
1+N
(4.18)
and
4N (N 1) + N (N 1)(N 2)
2N
C(x)2 +
C2 (x).
1+N
1+N
Diagrammatically, the averages can be rewritten as
Tr2 M =
(4.19)
:= C(x)2 ,
(4.20)
:= C2 (x).
(4.21)
and
Like in the case of the first order diagram (4.18) and (4.19) are highly simplified as compared to an open manifold, see our treatment in [33].
Let us shortly discuss the reason for (4.17): inserting (4.18) and (4.19) into the perturbation series and summing all loop orders, the following series types will appear:
k1
N

(z)N
i=0 (N i)(z)
k k
=
(1)
z
z
N!(N + 1)d/2+j +1
N !(N + k)d/2+j +1
N =0
(1)
N =0
d+2(j +1)
fk
(z).
(4.22)
We may therefore identify the resummed series with a function that we know already fairly
well, in particular we know its strong coupling behavior. It is furthermore convenient to
d+2(j +1)
d+2(j +1)
(z) to sums of functions f1
(z) exploiting the formula
reduce all functions fk>1
(3.25).
4.2. Resummed contributions to the expansion in 2 D up to fourth order
We are now almost in the position to state all resummed contributions up to fourth order
in 2 D. Let us first state all necessary diagrams:
= C(xi xj ),
549
(4.23)
which contributes to first order in 2 D. To second order one needs in addition

= C2 (xi xj ).
(4.24)
To third order diagrams with new topology are

= C3 (xi xj ),
= C(xi xj )C(xj xk )C(xi xk ).
(4.25)
Finally, to fourth order arise:

= C4 (xi xj ),
= C2 (xi xj )C(xk xj )C(xk xi ),
= C(xi xj )C(xk xj )C(xl xk )C(xi xl ).
(4.26)
If one calculates diagrams, it will turn out that it is to some extend more convenient to
express the above averages in terms of averages over a connected correlation function,
which is defined as
Cc (x) := C(x) C,
(4.27)
such that, for instance,

C2c = C2 C2 .
(4.28)
Furthermore, we will need:

C3c = C3 3CC2 + 2C3
(4.29)
and

Cc = Cc (xi xj )Cc (xj xk )Cc (xk xi )

= C(xi xj )C(xj xk )C(xk xi ) + 3C2 C(xi xj )
3CC(xi xj )C(xj xk ) C3
= C C3 ,
(4.30)
where xi , xj , xk are distinct points, and the average is over their positions. In (4.30) we
exploited the symmetry of the closed manifold, and the definition of C is self-evident.
550
Furthermore, we will need to fourth order in 2 D:

C4c = C4 + 12C2 C2 4C3 C 3C4 ,
(4.31)
Cc = Cc (xi xj )Cc (xj xk )Cc (xk xl )Cc (xl xi )

= C + 5C4
(4.32)
and
Cc = C2c (xi xj )Cc (xi xk )Cc (xk xj )
=C
2C C C2 C2 + 2C4 .
(4.33)
Let us now state all terms which appear in the expansion of the renormalized coupling
g(z) up to fourth order in 2 D according to (4.7). We have to calculate at order N of
perturbation theory:
Tr M = N C.
(4.34)
Inserting this into the perturbation series and summing up the resulting terms to all orders
in N generates the following contributions in the (2 D)-expansion of the renormalized
coupling:

d/2 Tr M(z)N +1
= Cf1d+2 (z) Cf1d (z),
det D(0)
(N + 1)!
(4.35)
N =1
which contributes to first order in 2 D.

To second order in 2 D, we have (4.18) providing

d/2 Tr M2 (z)N +1
det D(0)
(N + 1)!
N =1

= 2C2c f1d+4 (z) + 4C2c + C2 f1d+2 (z) + C2 + 3C2c f1d (z) C2c f1d2 (z),
(4.36)
and (4.19) providing

d/2 Tr2 M(z)N +1
det D(0)
(N + 1)!
N =1

= 2C2c f1d+4 (z) 2C2c + C2 f1d+2 (z) + 2C2 f1d (z) C2 f1d2 (z).
(4.37)
Let us now state the terms at third order in 2 D, which we derived with the help of a
M ATHEMATICA - program (N is the loop order):

d/2 Tr M3 (z)N +1
det D(0)
(N + 1)!
N =1

= 4 C3c 4Cc f1d+4 (z) + 10C3c + 36Cc + 6CC2c f1d+2 (z)

+ 9C3c 32Cc 12CC2c + C3 f1d (z)

+ 3C3c 17Cc 9CC2c + C3 f1d2 (z)

3 2Cc + CC2c f1d4 (z) + Cc f1d6 (z),
551
(4.38)

d/2 Tr M Tr M2 (z)N +1
det D(0)
(N + 1)!
N =1

= 4 C3c 4Cc f1d+6 (z) + 8C3c + 32Cc + 2CC2c f1d+4 (z)

+ 6C3c 20Cc + 2CC2c 6C3 f1d+2 (z)

+ 2C3c + 4Cc 7CC2c + 2C3 f1d (z)

C3 4CC2c f1d2 (z) CC2c f1d4 (z),
(4.39)

d/2 Tr3 M(z)N +1
det D(0)
(N + 1)!
N =1

= 4 C3c 4Cc f1d+6 (z) + 4C3c + 24Cc 6CC2c f1d+4 (z)

+ 8Cc + 12CC2c + C3 f1d+2 (z) 3 C3 + 2CC2c f1d (z)
+ 3C3 f1d2 (z) C3 f1d4 (z).
(4.40)
To fourth order in 2 D we obtain:

d/2 Tr4 M(z)N +1
det D(0)
(N + 1)!
N =1

= 8 222C4 + 6C2 C2c + 3C2c 2 C4c + 24Cc 36Cc f1d+8 (z)

+ 4 804C4 + 12C2 C2c + 3C2c 2 2C4c + 72Cc 132Cc 4CC3c

+ 16CCc f1d+6 (z)

4 432C4 3C2 C2c 6C2c 2 + 24Cc 72Cc 8CC3c + 40CCc f1d+4 (z)

+ 287C4 36C2 C2c 12C2c 2 48Cc 16CC3c + 128CCc f1d+2 (z)

+ 4C4 + 36C2 C2c 32CCc f1d (z)

+ 6C4 2C2 C2c f1d2 (z) + 4C4 f1d4 (z) C4 f1d6 (z),
(4.41)

d/2 Tr M2 Tr2 M(z)N +1
det D(0)
(N + 1)!
N =1


+ 4 960C4 + 24C2 C2c + C2c 2 4C4c + 100Cc 156Cc f1d+6 (z)
552

4 714C4 + 20C2 C2c + 8C2c 2 3C4c + 70Cc 116Cc 4CC3c

+ 12CCc f1d+4 (z)

+ 889C4 + 36C2 C2c 2C2c 2 28CC3c 4C4c + 80Cc 144Cc

+ 88CCc f1d+2 (z)

+ 99C4 + 3C2 C2c + 8C2c 2 + 16CC3c 8Cc + 16Cc 48CCc f1d (z)

+ 3C4 11C2 C2c 2C2c 2 4CC3c + 8CCc f1d2 (z)

+ C4 5C2 C2c f1d4 (z) C2 C2c f1d6 (z),
(4.42)

d/2 Tr2 M2 (z)N +1
det D(0)
(N + 1)!
N =1


+ 4 1116C4 + 36C2 C2c + 23C2c 2 6C4c + 128Cc 180Cc + 4CC3c

16CCc f1d+6 (z)

4 1080C4 + 47C2 C2c + 28C2c 2 8C4c + 132Cc 172Cc + 8CC3c

32CCc f1d+4 (z)

+ 2111C4 + 148C2 C2c + 44C2c 2 + 24CC3c 24C4c + 272Cc 328Cc

80CCc f1d+2 (z)

+ 538C4 74C2 C2c + 10C2c 2 8CC3c + 10C4c 72Cc + 80Cc

+ 8CCc f1d (z)

+ 59C4 + 20C2 C2c 15C2c 2 2C4c + 8Cc 8Cc f1d2 (z)

+ 2C2 C2c + 6C2c 2 f1d4 (z) C2c 2 f1d6 (z),
(4.43)

d/2 Tr M Tr M3 (z)N +1
det D(0)
(N + 1)!
N =1



8CCc f1d+6 (z)

2 1818C4 + 54C2 C2c + 39C2c 2 + CC3c 9C4c + 204Cc 294Cc

22CCc f1d+4 (z)

+ 1583C4 + 48C2 C2c + 36C2c 2 CC3c 6C4c + 186Cc 258Cc
553

+ 8CCc f1d+2 (z)

+ 358C4 21C2 C2c 6C2c 2 + 6CC3c 48Cc + 60Cc 37CCc f1d (z)

+ 35C4 + 12C2 C2c 3CC3c + 6Cc 6Cc + 23CCc f1d2 (z)

+ 3C2 7CCc f1d4 (z) + CCc f1d6 (z),
(4.44)

d/2 Tr M4 (z)N +1
det D(0)
(N + 1)!
N =1



16CCc f1d+6 (z)

4 1110C4 + 39C2 C2c + 31C2c 2 + 10CC3c 7C4c + 136Cc 178Cc

36CCc f1d+4 (z)

+ 2473C4 + 72C2 C2c + 82C2c 2 + 36CC3c 16C4c + 304Cc 396Cc

128CCc f1d+2 (z)

+ 955C4 12C2 C2c 36C2c 2 12CC3c + 5C4c 92Cc + 154Cc

+ 68CCc f1d (z)

+ 288C4 + 12C2c 2 C4c + 12Cc 47Cc 24CCc f1d2 (z)

+ 60C4 2C2c 2 + 10Cc + 4CCc f1d4 (z) + (6C4 Cc )f1d6 (z). (4.45)
4.3. Renormalized coupling

Combining (3.12), (4.7) and the results (4.35)(4.45) from the preceding subsection we
may now give the exact renormalized coupling to fourth order in 2 D. For the sake of
compactness, we introduce a new notation: since all series contributions are of the form as
stated in (4.22), we introduce vectors M such that

d/2
det D(0)
l
N
ni mi
i=1 (Tr M ) (z)
N =1
max

j =min
M m1
m2 ml
n1 n2 nl
(N + 1)!
f d+2j (z) Mj
m1
1
m2 ml max
n1 n2 nl min
f d+2j (z),
1
(4.46)
where max and min are some integers, and summation over the index j is implicit. Inserting
the results for the resummed series contributions into (4.7) we find for the renormalized
554
coupling to fourth order in 2 D:

d j
d+2j
g(z) = f1d+2 (z) (2 D) M 1 1 f1
(z)
2
1 0

d2 j
d+2j
d+2j
2 d j

M 1 2 f1
+ (2 D)
(z) + M 2 2 f1
(z)
4
8
2 1
1 1

d j
d2 j
d+2j
d+2j
(z) + M 1 1 3 f1
(z)
M 1 2 f1
(2 D)3
4
8
3 3
1 2 2

d3 j
d+2j
+ M 3 3 f1
(z)
48
1 2

d j
d+2j
+ (2 D)4
(z)
M 4 4 f1
8
1 3

d2 1 j
2 j
d+2j
d+2j
M 2 4 f1
+
(z) + M 1 1 4 f1
(z)
8 4
3
2 3
1 3 3

d3 j
d4 j
d+2j
d+2j
M 4 4 f1
+ M 2 1 4 f1
(z) +
(z) + O(2 D)5 .
32
384
1 2 3
1 3
(4.47)
Mj
The vector entries

are to be taken from Section 4.2.
It is more convenient to discuss instead of g(z) an integral transform. From the expansion of fkd (z), namely
d+2j
(z) =
f1
z
d
2

dr r d/2+j 1 exp zer r ,
(4.48)
and the structure of the expansion of g(z) in powers of 2D and the integral representation
d+2j
it follows that the exact renormalized coupling can be written as
of the f1

g(z) g(D, z) = z

dr g(r)
exp zer r ,
(4.49)
where g(r)
is of the form

n

1
d/2
j
n
g(r)
=r
pnj r (2 D) .
+ (2 D)

d+2
2
n=0 j =n
(4.50)
max
4.4. Guessing the exact g(r)
Let us try to gain more information about the power-law behavior in (4.1), that is about
the expansion in 2 D of the correction-to-scaling exponent . Power-law behavior forces
the series (4.50) to turn into some exponentially decaying function g(r)
as can be seen from
555
the asymptotic form of g(z):

g(z) A + Bz

=z
dr eze
r r

A+

Ber/
+ O ez .
(1 + /)
(4.51)
In order to check the latter equation note that

2
(z) = z1+/
f1+/

+ O ez
dr exp zer (1 + /)r = 1 +
z/ =
z
(1 + /)

dr exp zer (1 + /)r + O ez

z
=
(1 + /)

dr
n=0 0

(/r)n
exp zer r + O ez
n!
(/)n 2(n+1)

1
f1
(z) + O ez ,
(1 + /)
n!
(4.52)
n=0
where it is understood that is expanded in powers of 2 D.

Let us now test a possible form of the exact g(r).
It should satisfy the following properties:

(i) In the limit of D = 2 the exact result r d/2 / ( d+2
2 ) emerges.
(ii) For D < 2 the corresponding g(z) has a finite fixed-point value together with a strong
coupling expansion. Especially, the ansatz should interpolate to the limit D = 1,
which corresponds to a Gaussian polymer closed to form a ring. The strong coupling expansion of the renormalized coupling of a closed chain interacting with a
-potential is easily obtained from the factorizability of loop integrals in D = 1 (see,
for instance, [33]). The result is:

n

1
1
.
g(z) = 1 +
(4.53)
()z (1 n)
n=1
(iii) It is consistent with the expansion (4.47).

The (non-unique) ansatz is

1 S(D, r)e r d/2
,
g(r)
=C
/
(4.54)
where S(D, r) is analytic in D = 2 of the form

Sn (r)(2 D)n ,
S(D, r) = 1 + r

n=1
(4.55)
556
and each Sn (r) has a Laurent expansion

Sn (r) =
n
max
sn,j r j .
(4.56)
j =nmin
Note, that in the limit of D 2, the expression (4.54) gives Cr d/2 , while for D < 2 it
yields upon integration the form (4.1), ensuring both properties (i) and (ii). Let us finally
check consistency with the expansion (4.47) up to the second order in 2 D: inserting
= 2 (2 D)2 + O(2 D)3

(4.57)
(the linear term in (2 D) has to vanish2 ) into the ansatz (4.54) and expanding to second
order in 2 D provides

d
d/2
1
S1 (r)(2 D)
g(r)
= Cr
2

d 2
2
2
2
r
S1 (r) + S2 (r) (2 D) + .
+
(4.58)
2
4
Explicitly, (4.47) becomes to second order in 2 D

d
g(z) = f1d+2 (z) (2 D) Cf1d+2 (z) Cf1d (z)
2

2d
2C2c f1d+4 (z) + C2 4C2c f1d+2 (z) + 3C2c C2 f1d (z)
+ (2 D)
4

d2
2
Cc f1 (z)

d 2 2 d+4
2Cc f1 (z) 2C2c + C2 f1d+2 (z) + 2C2 f1d (z)
8

2 d2
C f1 (z) + O(2 D)3 .
+ (2 D)2
(4.59)
From this, the first coefficients of the (2 D)-expansion of g(r)
are obtained. They read

d/2
d
r
d
g(r)
= d+2 1 + (2 D) C 1
2
2r
2

d2

d
d
2C2c + C2
(2 D)2 C2c r + C2 4C2c
2
4
8

!
2

3
2 d

d
d
d
d
C2 + 3C2c + C2 r 1
+
1 C2c + C2 r 2 .
8
8
8 2
2
(4.60)
d1
Comparing (4.58) and (4.60), one identifies C = 1/ ( d+2
2 ), S1 = C(1 2 r ) and 2 =
2C2c , where Cc (x) := C(x) C. Note that the terms proportional to C2 in S2 (r) mostly
cancel with S1 (r)2 , a sign that the ansatz catches some structure.
2 This is due to the fact that the order (2 D) term in g(z) scales identically in z as the leading term. Only
the order (2 D)2 diverges more strongly.
557
The diagrams to be calculated at this order are C and C2c (see Appendix B). On a manifold of toroidal shape, which is equivalent to periodic boundary conditions, two discrete
sums have to be evaluated:

1
SD
2
C=
4 2
k2 (2 D)
D
kZ , k=0
= 0.44956 + 0.3583(2 D) + O(2 D)2 ,

C2c =
2
SD
16 4

kZD, k=0
1
= 0.152661 + O(2 D).

k4
(4.61)
(4.62)
With the results given above, this leads to

= 2C2c (2 D)2 + O(2 D)3 = 0.305322(2 D)2 + O(2 D)3 ,
(4.63)
which can be compared to the exact result for D = 1 (polymers): = . As a caveat,

note that the above scheme is not unambiguous in the sense that the second order term
proportional to r in (4.61) could in principle either be attributed to 2 or S2 . However, any
ansatz in (4.54) will provide an , whose expansion starts at least quadratically in 2 D.
Though (4.54) is the best ansatz that could yet be found ensuring properties (i)(iii), the
precise form of constraints on the scaling function S remains to be discussed in order to
settle this question.
5. Conclusion
In this work we refined the analysis of a D-dimensional elastic manifold interacting
by some -potential with a fixed point in embedding space. Starting from the perturbation
expansion of the effective coupling of the problem, in a first step, we performed a new
calculation using a modified regularization prescription: evaluating loop integrals in fixed
space dimension on a manifold of finite size enforced the introduction of a microscopic cutoff as soon as D = 2. This way, we recovered the complete summability of the perturbation
theory in this limit and confirmed the strong coupling behavior as found previously in an
analytic continuation from below D = 2. In the strong coupling limit, corresponding to
strong repulsion or equivalently to large membrane sizes, the effective coupling diverges
logarithmically as a function of the bare coupling z yielding a vanishing correction-toscaling exponent . Analyzing the RG -function we found that it tends to zero at infinite
bare coupling z as 0 d < 2. The renormalization group flow then tends to a fixed point,
and the theory becomes scale invariant in this limit. Due to the logarithmic divergence of
the effective coupling, however, the corresponding zero of the -function in terms of the
latter is, too, shifted to infinity. This is a quite remarkable result showing that the scaling
behavior of the system is accessible only to an all order treatment and deviates qualitatively
from any finite loop expansion, be it within a minimal subtraction scheme or at finite .
Especially, the logarithmic growth of the effective coupling signals the limiting behavior
of a scale-invariant theory.
558
The result in D = 2 is completely independent of the regularization procedure. This

does no longer hold true beyond the leading order, which should be accessible to an expansion in 2 D. We constructed its first order in a specific regularization scheme in [33].
While this reproduces qualitatively correctly the known result in D = 1, it suffers from
a renormalization scheme, which neglects the boundaries of a finite manifold. We used a
hard cutoff in position space, while working with the infinite D-space correlator. It seems
that only in an -expansion this procedure is systematic.
Now, in a second step of the analysis we overcame this problem by constructing the
(2 D)-expansion on a manifold of toroidal shape of finite size, thus imposing periodic
boundary conditions on the field. There is no further infrared cutoff necessary. We have
carried out the expansion of the renormalized coupling up to fourth order in 2 D, revealing the general structure of the expansion. It is important to point out that in considering g
as a function of the bare coupling, the limits D 2 and strong coupling (z ) cannot
be interchanged. While g tends to infinity as z does in D = 2, we expect finiteness of this
limit as soon as 2 D > 0 and the existence of a strong coupling expansion as found for
polymers (D = 1). We were able to guess an exact g(D, z) as a function of z and the internal dimension D, which satisfies these properties and which can be reconciled with the
available expansion in 2 D by an appropriate matching of its free parameters. Though it
turned out that due to an ambiguity in the matching of parameters the precise power-law
behavior of the effective coupling below D = 2 cannot yet be isolated, we found that for
closed manifolds the expansion of in powers of 2 D starts at least quadratically as
D < 2.
The exponent is closely related to observables, which can be measured in Monte Carlo
experiments. These are, for instance, plaquettes-density functions at the repelling potential
on a membrane avoiding a single point.
While results for the pinning problem are interesting on their own, the main motivation is certainly to obtain a better understanding of self-avoiding polymerized membranes.
Preliminary studies [40] indicate that this problem can also be attacked by the methods
developed here. This would be welcome to settle the discrepancies between field theoretic
results on one hand [16,17,41] and numerical results (e.g., [32]) on the other.
Acknowledgements
It is a pleasure to thank R. Blossey, F. David, H.W. Diehl, M. Kardar, and L. Schfer
for useful discussions. We are grateful to Andreas Ludwig for persisting questions, and his
never tiring efforts to understand the limit of D 2. This work has been supported by the
DFG through the Leibniz program Di 378/2-1, under Heisenberg grant Wi 1932/1-1.
Appendix A. The propagator

The regularized difference correlator is defined as
Ca (x) = Ga (0) Ga (x)
(A.1)
559
where Ga (x) denotes the usual two-point correlator, which is obtained from:3

1
exp[i kx a 2 k 2 ]
Ga (x) =
.
dD k
D
(2)
k2
(A.2)
Here, short-wavelength modes are suppressed through a soft cutoff procedure. Introducing
a Schwinger parameterization for the evaluation of the integral in (A.2),

Ga (x) =
dt
dD k (t+a 2 )k 2 ikx
1
e
e = D
D
(2)
(2 )
where s = 1/(t
Ca (x) =
2
1/a
ds s D/21 es
x2
4
(A.3)
+ a 2 ),
we obtain for (A.1):
(2 )D
2
1/a

x2
ds s D/21 1 es 4 .
(A.4)
Further evaluation leads to:

(i) D = 2:

x 2 x 1
x2
x
1
+ 0, 2 + ln 2
ln .
Ca (x) =
4
2 a
4a
4a
(A.5)
(ii) D < 2:
2
|x|2D ( D2 )
a 2D
a 2D
x2
4a
+
e
(2 D)2 D/2 (2 D)2D1 D/2
(2 D)2D1 D/2
Ca (x) =
x
|x|2D ( D2 , 4a
2)
(2 D)2 D/2
a 2D
x
|x|2D
D2 1
ln .
D1
D/2
SD (2 D) (2 D)2
2 a
(A.6)
(z, ) denotes the incomplete -function:

(z, ) =
dt t z1 et .
(A.7)
|x|2D
,
SD (2 D)
(A.8)
Especially:
lim Ca (x) =
a0
as long as D < 2.
3 Strictly speaking, we have to consider the propagator on the torus, as is done in Appendix B. However, this
does not make any difference for the purpose of our argument.
560
Appendix B. Calculation of the diagrams in the (2 D)-expansion

In this section we calculate the diagrams which appear in the 2 D expansion on the
torus of size L = 1. It turns out that to obtain C and C2c we need to evaluate two sums
over discrete wave-vectors due to periodic boundary conditions on the torus. Let us first
derive the latter before turning to the explicit evaluation. Starting from the definition of the
difference correlator C(x),
C(x) := G(x) G(0),
(B.1)
where G(x) is the usual two-point correlator, we obtain C(x) through an inverse discrete
Fourier-transformation from G(k) = 1/k2 , which reads:
1

C(x) =
(B.2)
1 ei k x , k = 2 n, n Z Z\{0}.
2

k

k=0
Performing the averaging procedure

C(x) = C(x),
where
"
x
(B.3)
x

ei k x
= D
is to be taken into account, the calculation of C(x) reduces to
k
C(x) = I1 :=
1
,
k2
k = 2 n,
(B.4)
k=0
where k is D-dimensional, and the indices ni are integer and running from to ,
n = 0 being excluded from the summation. Of course, in the expansion in powers of 2 D
we need an analytic continuation to real values of D. Finally, to obtain C(x) we have to
subtract C(0) (x) from C(x). Due to our normalizations:

C(0) (x)
,
C(x) = SD C(x)
(B.5)
2(2 D)
where SD denotes the volume of the unit sphere and C(0) (x) = 1.
Turning to C2c (x), we first note that within our normalizations we have
2 2
C (x) = (C(x) C(0) (x)/(2(2 D)))2
SD
= C(x)2 2
1
C(x)
+
2(2 D) (2(2 D))2
(B.6)
2
SD
C(x)2 = C(x)2 2
1
C(x)
+
2(2 D) (2(2 D))2
(B.7)
and
according to (B.5), such that

2
C2c (x) C2 (x) C(x)2 = SD
C 2 (x) C(x)2 .
(B.8)
561
Knowing already the sum to be evaluated to obtain C, (B.4), what is left is:

1 1 i kx
2
2
e 1 ei px 1
C (x) = C (x) =
2
2

k p
x k=0 p=0
2
1 1
1
1
D
D
D
+
1
=
+
.

p
k
k2 p2 k+p
k4
k2
k=0 p=0
k=0
(B.9)
k=0
Therefore,
2 2
SD
Cc (x) = I2 :=
1
,
k4
ki = 2ni .
(B.10)
k=0
Let us first calculate I1 : introducing a Schwinger parameterization we have:

1
I1 =
(2)2
1
=
(2)2

ni =
n=0
1
1
=
2
n
(2)2
ds es n
ni =
0
n=0

D

2
ds
esn
1 ,
(B.11)
n=
where it is to be noted that the sum in the last line is only one-dimensional. Furthermore,
from now on it is clear, how I1 is analytically continued to real values of D.
In order to evaluate this sum, we will make use of a Poisson-transformation, which
reads:

2 l 2 +ilz
A(nz/2)2
(B.12)
e
=
e A
.
A
n=
l=
The contribution from l = 0 is the approximation of the l.h.s. through a Gaussian integral.
Our aim is to calculate the coefficients of the 2 D expansion of I1 numerically using
some algebraic manipulation program. Then, the integration interval in (B.11) has to be
made finite. This is done as follows: for any s0 > 0 we have
1
1
I1 =
(2)2
s0

ds
1
s0
ds
0
D
e
sn2
n=
1
=
(2)2

n=
D
e
sn2
1
1 +
(2)2

1
1 +
(2)2

ds
D
e
sn2
n=
s01
s0
ds
s2

D
e
n2 /s

1 .
n=
(B.13)
For any finite s0 > 0, the sum in the r.h.s. integral can be truncated at some finite nmax for
all s [0, s0 ]. For the first integral (corresponding to small values of s) we make use of the
562
Poissonian formula (B.12) with z = 0:
sn2

=
n=
2 l 2 /s
e
.
s
(B.14)
l=
Inserting this into (B.13), the sum in the first integral can be truncated at some finite l as
well, such that one may approximately write:
1
1
I1
(2)2
s0

ds
1
+
(2)2
s0
0
ds
s2

lmax

D
e
2 l 2 /s
l=lmax
n
max
D
e
n2 /s

1 .
(B.15)
n=nmax
Choosing s0 in a way that lmax can be set equal to zero the l.h.s. integral can be evaluated
analytically:

D
D/2

s0 n
max
2
1
ds
1
D/21
1
n2 /s
I1
+
s0
e
1 .
s
(2)2 2 D 0
(2)2
s2
n=nmax
0
(B.16)
There is a pole in 2D, which can be easily subtracted expanding the expression in powers
of 2 D. The pole is
I1 =

1
+ O (2 D)0 .
2(2 D)
(B.17)
The precision of the machine that we used to evaluate (B.16) was sufficient in a way that
we could select s0 from an interval, such that the sum appearing in the integrand could
be truncated at some finite nmax and the result was independent from the precise value of
s0 within the desired order of accuracy, therefore, justifying the approximation in (B.15).
Setting, for instance, s0 = 1.9 and nmax = 20 we obtain with M ATHEMATICA :
I1 =

1
0.715497(1) 0.00457046(1)(2 D) + O (2 D)2 . (B.18)
2(2 D)
On the torus we scaled the square root of the volume of the D-dimensional unitsphere into
the field. Accordingly, comparing with (B.4) and (B.5) we then find:

C = 0.44956(1) + 0.3583(1)(2 D) + O (2 D)2 .
(B.19)
Let us turn to the evaluation of I2 following the same strategy as above. Again, setting
L = 1 and introducing a Schwinger parameterization leads to:
1
I2 =
(2)4

ds s
sn2
n=

(2)4
D
563
s0
ds s
0
1
+
(2)4
s0
0
ds
s3

D
lmax

2 l 2 /s
l=lmax
n
max
D
e
n2 /s

1 ,
(B.20)
n=nmax
where we have once again applied the Poisson-transformation (B.12) with z = 0 on one
part of the integration interval and truncated both series at some finite values nmax and lmax .
There is no pole in 2 D. Since I2 appears at second order in 2 D we only need its
value at D = 2. s0 has to be chosen from an appropriate interval. Setting nmax = nmax = 10
and s0 = 1.1 we obtain with M ATHEMATICA :

I2 = 0.00386695(1) + O (2 D) ,
(B.21)
2,
or, due to the rescaling by SD

C2c = 0.152661(1) + O (2 D) .
(B.22)
References
[1] L. Schfer, Excluded Volume Effects in Polymer Solutions, Springer-Verlag, Berlin, 1999.
[2] J. des Cloizeaux, G. Jannink, Polymers in Solution, Their Modelling and Structure, Clarendon, Oxford,
1990.
[3] P.-G. de Gennes, Scaling Concepts in Polymer Physics, Cornell Univ. Press, Ithaca, NY, 1979.
[4] E. Eisenriegler, Polymers Near Surfaces, World Scientific, Singapore, 1993.
[5] M. Fixman, Excluded volume in polymer chains, J. Chem. Phys. 23 (1955) 16561659.
[6] L. Schafer, T.A. Witten, Renormalized field theory of polymer solutions, J. Chem. Phys. 66 (1977) 2121.
[7] J. des Cloizeaux, Polymers in solutions: principles and applications of a direct renormalization method, J.
Phys. 42 (1981) 635652.
[8] S.F. Edwards, The statistical mechanics of polymers with excluded volume, Proc. Phys. Soc. London 85
(1965) 613.
[9] P.-G. de Gennes, Exponents for the excluded volume problem as derived by the Wilson method, Phys. Lett.
A 38 (1972) 339340.
[10] Y. Kantor, D.R. Nelson, Crumpling transition in polymerized membranes, Phys. Rev. Lett. 58 (1987) 2774
2777.
[11] Y. Kantor, D.R. Nelson, Phase transitions in flexible polymeric surfaces, Phys. Rev. A 36 (1987) 40204032.
[12] Y. Kantor, M. Kardar, D.R. Nelson, Statistical mechanics of tethered surfaces, Phys. Rev. Lett. 57 (1986)
791795.
[13] Y. Kantor, M. Kardar, D.R. Nelson, Tethered surfaces: statics and dynamics, Phys. Rev. A 35 (1987) 3056
3071.
[14] M. Paczuski, M. Kardar, D.R. Nelson, Landau theory of the crumpling transition, Phys. Rev. Lett. 60 (1988)
2638.
[15] M. Paczuski, M. Kardar, Renormalization-group analysis of the crumpling transition in large d, Phys. Rev.
A 39 (1989) 60866089.
564
[16] F. David, K.J. Wiese, Scaling of self-avoiding tethered membranes: 2-loop renormalization group results,
Phys. Rev. Lett. 76 (1996) 4564.
[17] K.J. Wiese, F. David, New renormalization group results for scaling of self-avoiding tethered membranes,
Nucl. Phys. B 487 (1997) 529632.
[18] K.J. Wiese, Polymerized membranes, a review, in: Phase Transitions and Critical Phenomena, vol. 19, Academic Press, London, 1999.
[19] M. Kardar, D.R. Nelson, expansions for crumpled manifolds, Phys. Rev. Lett. 58 (1987) 1289;
M. Kardar, D.R. Nelson, expansions for crumpled manifolds, Phys. Rev. Lett. 58 (1987) 2280, Erratum.
[20] J.A. Aronovitz, T.C. Lubensky, Fluctuations of solid membranes, Phys. Rev. Lett. 60 (1988) 26342637.
[21] F. David, B. Duplantier, E. Guitter, Renormalization and hyperscaling for self-avoiding manifold models,
Phys. Rev. Lett. 72 (1994) 311.
[22] F. David, B. Duplantier, E. Guitter, Renormalization theory for the self-avoiding polymerized membranes,
cond-mat/9702136.
[23] T. Hwa, Generalized expansion for self-avoiding tethered manifolds, Phys. Rev. A 41 (1990) 17511756.
[24] K.J. Wiese, F. David, Self-avoiding tethered membranes at the tricritical point, Nucl. Phys. B 450 (1995)
495557.
[25] R.R. Chianelli, E.B. Prestridge, T.A. Pecorado, J.P. de Neufville, Molybdenum disulfide in the poorly crystalline rag structure, Science 203 (1979) 1105.
[26] T. Hwa, E. Kokufuta, T. Tanaka, Conformation of graphite oxide membranes in solution, Phys. Rev. A 44
(1991) 2235.
[27] X. Wen, C.W. Garland, T. Hwa, M. Kardar, E. Kokufuta, Y. Li, M. Orkisz, T. Tanaka, Crumpled and collapsed conformations in graphite oxide membranes, Nature 355 (1992) 426.
[28] M.S. Spector, E. Naranjo, S. Chiruvolu, J.A. Zasadzinski, Conformations of a tethered membrane: crumpling
in graphitic oxide?, Phys. Rev. Lett. 73 (1994) 28672870.
[29] A. Baumgrtner, Does a polymerized membrane crumple?, J. Phys. I France 1 (1991) 15491556.
[30] A. Baumgrtner, W. Renz, Crumpled self-avoiding tethered surfaces, Europhys. Lett. 17 (1992) 381386.
[31] D.M. Kroll, G. Gompper, Floppy tethered networks, J. Phys. I France 3 (1993) 1131.
[32] G. Thorleifsson, M. Bowick, A. Cacciuto, A. Travesset, Universality classes of self-avoiding fixedconnectivity membranes, Eur. Phys. J. E 5 (2001) 149.
[33] H.A. Pinnow, K.J. Wiese, Interacting crumpled manifolds, J. Phys. A 35 (2002) 11951229.
[34] H.A. Pinnow, K.J. Wiese, Interacting crumpled manifolds: exact results to all orders of perturbation theory,
Europhys. Lett. 64 (2003) 371377.
[35] F. David, B. Duplantier, E. Guitter, Renormalization of crumpled manifolds, Phys. Rev. Lett. 70 (1993)
2205.
[36] F. David, B. Duplantier, E. Guitter, Renormalization theory for interacting crumpled manifolds, Nucl. Phys.
B 394 (1993) 555664.
[37] G. Forgas, R. Lipowsky, T.M. Nieuwenhuizen, The behaviour of interfaces in ordered and disordered systems, in: Phase Transitions and Critical Phenomena, vol. 14, Academic Press, London, 1991, pp. 136376.
[38] E. Brzin, B.I. Halperin, S. Leibler, Critical wetting in three dimensions, Phys. Rev. Lett. 50 (1983) 1387.
[39] P.J. Upton, Exact interface model for wetting in the planar Ising model, Phys. Rev. E 60 (1999) 34753478.
[40] H.A. Pinnow, K.J. Wiese, in preparation.
[41] F. David, K.J. Wiese, Large orders for self-avoiding membranes, Nucl. Phys. B 535 (1998) 555595.
Bethe ansatz for the XXX-S chain with

non-diagonal open boundaries
C.S. Melo, G.A.P. Ribeiro, M.J. Martins
Universidade Federal de So Carlos, Departamento de Fsica, C.P. 676, 13565-905 So Carlos (SP), Brasil
Received 15 November 2004; accepted 10 December 2004
Available online 29 December 2004
Abstract
We consider the algebraic Bethe ansatz solution of the integrable and isotropic XXX-S Heisenberg
chain with non-diagonal open boundaries. We show that the corresponding K-matrices are similar
to diagonal matrices with the help of suitable transformations independent of the spectral parameter.
When the boundary parameters satisfy certain constraints we are able to formulate the diagonalization of the associated double-row transfer matrix by means of the quantum inverse scattering method.
This allows us to derive explicit expressions for the eigenvalues and the corresponding Bethe ansatz
equations. We also present evidences that the eigenvectors can be build up in terms of multiparticle
states for arbitrary S.
PACS: 05.50.+q; 02.30.IK
Keywords: Algebraic Bethe ansatz; Open boundary
1. Introduction
The possibility of constructing SU(2) invariant Heisenberg chain with arbitrary spinS solvable by Bethe ansatz methods was a remarkable achievement of the representation
theory underlying the associative algebra describing the dynamical symmetry of quantum
integrable systems [1]. It turns out that the Hamiltonian of such spin-S XXX Heisenberg
E-mail address: martins@df.ufscar.br (M.J. Martins).
doi:10.1016/j.nuclphysb.2004.12.008
566
C.S. Melo et al. / Nuclear Physics B 711 [FS] (2005) 565603
magnet [2] commutes with the transfer matrix TS () of a 2S + 1 state vertex on the square
L L lattice [1,3]. This connection is based on well-known relationships between onedimensional quantum spin chains and two-dimensional statistical mechanics models whose
Boltzmann weights satisfy the YangBaxter equation [4,5].
The row-to-row transfer matrix TS () of such 2S + 1 state vertex model can be conveniently written as the trace, over an auxiliary space A C 2S+1 , of an ordered product of
Boltzmann weights. More specifically,
(S)
(S)
(S)
(S)
(S)
TS () = TrA TA () , TA () = LAL ()LAL1 () LA1 (),
(1)
where is the spectral parameter and A represents the horizontal degrees of freedom of
the vertex model.
(S)
The Boltzmann weight Lab () is solution of the YangBaxter equation
(S)
(S)
(S)
(S)
Lab ( )Ta(S) ()Tb () = Tb ()Ta(S) ()Lab ( ),
(2)
invariant relative to the SU(2) Lie algebra. It can be viewed as (2S + 1) (2S + 1) matrix
on the auxiliary space whose
are operators acting non-trivially only on the bth
elements
2S+1
factor of the Hilbert space L
C
. Its explicit expression in terms of the spin-S
b=1
b
y
x
z

SU(2) generators Sa = (Sa , Sa , Sa ) is [13]
(S)
Lab () = ( + 2S)
2S
2S
2S

k Sa Sb xn
,
+ k
xl xn
l=0 k=l+1
(3)
n=0
n=l
where xl = 12 l(l + 1) S(S + 1) and is the so-called quasi-classical parameter.

(S)
Besides the YangBaxter equation the operator L12 () satisfies other relevant properties such as
(S)
(S)
Unitarity:
L12 ()L21 () = S () Id Id;
(4)
Parity invariance:
(S)
(S)
P12 L12 ()P12 = L12 ();
(S)
(S)
L12 ()t1 t2 = L12 ();
(5)
Temporal invariance:
(6)
1
1 (S)
S ()
(7)
V L12 ( )t2 V 1 ;
S ( )
2S1
( + k). Here Id is the
where functions S () = (2S)2 2 and S () = k=1
(2S + 1) (2S + 1) identity matrix, P12 is the permutation operator, t denotes trans(S)
L12 () = (1)2S
Crossing symmetry:
position on the th space, V = V Id and V = Id V . The matrix V is anti-diagonal

whose non-null elements are Vi,j = (1)i i,2S+2j .
This notion of integrability has been extended to include integrable open boundary con(S)
ditions [6,7]. In addition to the YangBaxter solution Lab () determining the dynamics
of the bulk one has to introduce (2S + 1) (2S + 1) K-matrices KS () whose elements
represent the interactions at the left and right ends of the open spin chain. Compatibility
with bulk integrability demands that these matrices should satisfy the reflection equation
567
given by [7]
(S)
(S)
(S)
(S)
L12 ( )KS ()L21 ( + )KS () = KS ()L12 ( + )KS ()L21 ( ),

(8)
1
where KS () = KS () Id and KS () = Id KS ().

In the case of open boundaries the analogue of the transfer matrix is the following
double-row operator [7]
(S)
1
(+)
(S)
()
tS () = TrA KS ()TA ()KS () TA ()
(9)
,
()
where KS () can be chosen as one of the solutions of the reflection equation (8). The
(+)
()
other matrix KS () can be directly obtained from KS () thanks to the extra relations
(S)
(4)(7) satisfied by the operator Lab (). Following a scheme devised in Ref. [8] this isomorphism becomes
()
t
(+)
KS () = KS ( ) .
(10)
The understanding of the physical properties of the XXX-S open chain includes necessarily the exact diagonalization of the double-row operator (9). If the K-matrices are
diagonal this problem can be tackled, for example, by an extension of the quantum inverse
scattering method [7] and the use of fusion hierarchy procedures [9,10]. The same does not
occur when the K-matrices are non-diagonal due to an apparent lack of simple reference
states to start Bethe ansatz analysis. In spite of this difficulty, progresses have recently been
made for the anisotropic version of the S = 12 Heisenberg magnet usually denominated the
XXZ spin chain. These achievements have been made either by a functional Bethe ansatz
analysis [11] or by means of the algebraic Bethe ansatz method [12]. The latter approach
has been based on earlier ideas developed in the context of the eight vertex model [13].
In particular, it was argued that the spectrum of the open XXZ chain can be parameterized by Bethe ansatz equation provided certain constraint between the parameters of the
Hamiltonian is satisfied. Part of the conclusions were achieved with the help of a numerical study of the spectrum for finite values of L [14]. More recently, new results have been
obtained in Ref. [15] by exploring the description of the open XXZ spin chain in terms of
the TemperleyLieb algebra. The extension of all such analysis for integrable Heisenberg
chains with arbitrary spin-S appears to be highly non-trivial and it is indeed an interesting
open problem in the field of integrable models.
In this paper we would like to take some steps towards the direction of solving the
isotropic higher spin Heisenberg model (3) with non-diagonal open boundaries. We show
that the double-row transfer matrix operator associated to the integrable XXX-S Heisenberg chain can be diagonalized by Bethe ansatz at least when the respective K-matrices
parameters satisfy one out of two possible types of constraints. We find that the roots of
the Bethe ansatz equations are fixed by integers n 2SL that play the role of standard
particle number sectors. This feature shows that the Hilbert space has a multiparticle structure which should be useful to determine the nature of the ground state and excitations
unambiguously.
568
The outline of this paper is as follows. In Section 2 we argue that the non-diagonal
K-matrices of the XXX-S Heisenberg model are diagonalizable by spectral independent
similarity transformations. In Section 3 suitable quantum space transformations are used
to show that the diagonalization of tS () is similar to an eigenvalue problem with diagonal
and triangular K-matrices provided that certain constraints are satisfied. In Section 4 we
discuss the quantum inverse scattering method for the latter system, presenting the corresponding eigenvalues and Bethe ansatz equations. Explicit expressions for the eigenvectors
in terms of similarity transformation acting on creation fields can be written for spin 12
and 1. In Section 5 our conclusions and further perspectives are discussed. In Appendix A
we summarize certain properties of the K-matrices. In Appendices B and C we discuss
the one and two particle analysis of the eigenspectrum as well as auxiliary expressions for
S = 32 , respectively. Finally, in Appendix D we exhibit general relations concerning the
one-particle unwanted terms and the two-particle state construction for arbitrary S.
2. The K-matrices properties

The most general reflection K-matrix associated to the open XXX-S Heisenberg chain
possesses three free parameters. For S = 12 it is given by [16]

+
c
()
K 1 () =
(11)
,
d

2
while the isomorphism (10) implies that

+ 1 c+ ( + 1)
(+)
K 1 () =
,
d+ ( + 1) + + 1 +
2
(12)
where , c and d are six free parameters.

A remarkable characteristic of these K-matrices is that they can be diagonalized by similarity transformations which are independent of the spectral parameter . More precisely,
it is possible to rewrite Eqs. (11), (12) as

() 1
0
()
() () +
K 1 () = 1 G 1
(13)
G1
,
2
2
2
2
and
(+)
(+) (+)
K 1 () = 1 G 1
2

+ 1
0
0
+ + 1 +
(+) 1
G1
,
(14)
()
where GS refer to appropriate (2S + 1) (2S + 1) matrices. In what follows we will

represent them in terms of the standard Weyl basis eij by the expression
()
GS
2S+1

i,j =1
()
gi,j eij .
(15)
In the specific case of S =

()
elements of G1/2 are
()
g2,1
()
g1,1
1
2
569
the expressions relating the off-diagonal and the diagonal
1 + 1 + c d
,
c
()
g1,2
()
g2,2
1 + 1 + c d
,
d
(16)
()
where + = = 1. The other variables and 1/2
entering in the formulae (13), (14)
are given by

=
,
1 + c d
()
1
2

= 1 + c d .
(17)
The K-matrices for S > 12 can be computed either by brute force analysis of the reflection equation [17] or constructed by the so-called fusion procedure [18]. Their matrix
elements expressions become very cumbersome as one increases the value of the spin and
this fact has been exemplified in Appendix A for spin 1 and 32 cases. It turns out, however,
that we have found out that such K-matrices can be rewritten in a rather compact and illuminating form with the help of appropriate spectral independent similarity transformations,
namely
() 1
()
() () ()
KS () = S GS DS () GS
(18)
,
where the overall normalizing constant is

1 + c d 2S
()
2S
.
S = ( )
2S
The diagonal matrix DS() () is defined by
DS() () =
2S+1
fj() (S; , )ejj ,
(19)
j =1
where the corresponding diagonal entries are

f() (S; , ) =
2S

+ S +

1
1
,
sign
2
2
(20a)
+ + S +

1
1
+ sign
+1 .
2
2
(20b)
=1
f(+) (S; , + ) =
2S

=1
Interesting enough, the novel parameters encode both the dependence on the spin
value and on the variables describing the off-diagonal K-matrices elements. Specifically,
we have found that
2S
=
.
1 + c d
(21)
570

()
()
()
()
()
Finally, the four elements g1,1 , g1,2 , g2,1 and g2,2 of GS are related by the expressions
()
g2,1
(1)2S 2S(1 + 1 + c d )
=
,
(22)
()
c
g1,1

()
g2,2
(1)2S 2 (S 1)c d + (2S 1)(1 + 1 + c d )
,
(23)
=
()
1 + 1 + c d
Sc
g
1,2
and the remaining elements are obtained by the following recurrence relations
() ()
() ()
2S(m 1)(2S + 2 m)g2,1 gm1,l 2S(l 1)(2S + 2 l)g1,2 gm,l1
()
gm,l =
,
()
2S(m l)g1,1
(24)
for l = m = 1, . . . , 2S + 1 while for m = l we have
()
gl,l
() ()
g2,2
gl1,l1
()
g1,1
()
()
() ()
2(2S 1)(l 2)(2S + 3 l)g3,1 (S)gl2,l + 2(l 2)g2,1 gl1,l
. (25)
()
2S(2S + 2 l)(l 1)g1,1
()
An important feature of our construction is that the matrices GS are itself representations, without spectral parameter, of the monodromy matrix associated to the YangBaxter
(S)
algebra (2) generated by the operators Lab (). In fact, the matrix (15), (24), (25) with four
free parameters are the widest possible class of non-diagonal twisted boundary conditions
compatible with integrability for the XXX-S spin chain [19]. An immediate consequence
of this symmetry is the commutation relation

(S)
L12 (), GS() GS() = 0,
(26)
which will be of great use in next section.
3. The eigenvalue problem

The purpose of this section is to show that the eigenvalue problem for the double-row
transfer matrix operator tS (),
tS ()| = S ()|,
(27)
associated to the XXX-S chain with two general non-diagonal open boundaries can be
transformed into a similar problem with only one genuine non-diagonal K-matrix.
(+)
In order to demonstrate that we use the decomposition property for the KS () matrix
described in Section 2 and the operator tS () becomes
(+) (+) (+) 1 (S)
(S)
1
tS ()
()
.
= TrA GS DS () GS
TA ()KS () TA ()
(28)
(+)
S
571
We now proceed by inserting identity terms of type [GS ]1 GS in between all the
fundamental operators that appear in the trace (28). By using the invariance of the trace
under cyclic permutation one can rewrite Eq. (28) as
(+)
tS ()
(+)
(+)
(+)
(S)
1
(S)
()
,
= TrA DS ()TA ()K S () TA ()
(29)
(S)
(S)
(S)
(S)
(S)
where TA () = L AL ()L AL1 () L A1 (). The new operator L Aj () and K-matrix
K S() () are given in terms of unitary transformations acting on the auxiliary space by the
expressions,
(+) 1 (S)
(S)
(+)
LAj ()GS ,
L Aj () = GS
(30)
and
(+) 1 ()
()
(+)
K S () = GS
KS ()GS .
(31)
(S)
LAj ()
It turns out, however, that the gauge transformation (30) on the

operators can
be reversed with the help of a second transformation on the quantum space [19]. In fact,
one can use property (26) to define quantum space matrices Vj acting non-trivially only at
the j th site
Vj = Id Id GS(+) Id Id,

(32)
j th
such that they are able to undo the transformation (30), namely
Vj1 L Aj ()Vj = LAj ().
(33)
We now can use this property in order to define a new double-row transfer matrix operator tS ()
(S)
tS () =
(S)
L

Vj1 tS ()
j =1
L

Vj ,
(34)
j =1
having only one non-diagonal K-matrix

tS ()
(+)
S
(+)
(S)
1
(S)
()
.
= TrA DS ()TA ()K S () TA ()
(35)
Clearly, the operators tS () and tS () have the same eigenvalues while their eigenstates
are related by a similarity transformation,
| and |
| =
L

Vj |.
(36)
j =1
Though this framework clearly brings a considerable simplification in the original

eigenvalue problem, it is not enough to make the diagonalization of the double-row operator tS () with six free boundary parameters amenable to a standard Bethe ansatz analysis.
572
Table 1
()
The triangular property dependence of K S () on the ratio = + /
Manifold
= + /
()
K S ()
I
II
Upper
Triangular
I
II
Lower
Triangular
()
This is because the K-matrix K S is generally non-diagonal which still imposes us the
difficulty of finding suitable reference states needed to begin the Bethe ansatz computations. However, a great advantage of this formulation is that one can easily identify the
existence of at least three cases of physical interest in which the standard SU(2) highest
weight states could be used as pseudovacuums to build up the whole Hilbert space. The
simplest occurs when one of the boundaries is free, say KS() () = Id while the other is
()
still arbitrary with three free parameters. The next one is when the K-matrices KS () are
(+)
()
diagonalizable in the same basis, i.e., GS = GS which implies that we have altogether
four distinct couplings say c , d and . This includes the important symmetric situation
where the left and right K-matrices are the same but arbitrarily non-diagonal. As far as the
Bethe ansatz technicalities are concerned the most general case in which SU(2) highest
()
weight vectors can be used as a reference state is when the effective K S () K-matrix becomes either upper or lower triangular. This leads us to an open integrable system with five
free couplings since such condition imposes certain constraint between the parameters c
and d . Substituting the representation (18) in Eq. (31) and after some algebra we find that
there are two possible classes of restrictions satisfying the above mentioned triangularity
property. It turns out that these constraints depend only upon the variables c , d and their
expressions are,
1 + 1 + c d 1 + + 1 + c + d +
(I)
(37)
=
,
c
c+
1 + + 1 + c + d +
d
=
.
(II)
(38)
c+
1 + 1 + c d
Depending on the ratio = + / the zeros entries of K S () are either bellow or

above the principal diagonal. This feature has been summarized in Table 1 for each man()
ifold. Note that the diagonal elements of the triangular matrix K S () will necessarily
()
be the eigenvalues of KS (). By considering decomposition (18) we conclude that such
() ()
eigenvalues are exactly the entries of the diagonal matrix S DS ().
Considering the above discussions, we find that the formulation of a Bethe ansatz solution for the eigenspectrum of tS () on the parameters manifolds (I) and (II) is certainly
worthwhile to pursue. It will leads us to benefit from the knowledge of the exact spectrum
with five out of six possible boundary couplings, a considerable number of free parameters
at our disposal. A fundamental ingredient in the algebraic Bethe ansatz is the quadratic
relations satisfied by the matrix elements of the double-row monodromy matrix defined
()
573
by [7]
(S)
1
(S)
(S)
()
TA () = TA ()K S () TA () ,
(39)
and consequently the double-row operator tS () can be written in the form

tS ()
(+)
S
(+)

(S)
= TrA DS ()TA () .
(40)
Taking into account the property (26) we see that the effective K S () matrix satisfies
()
the same reflection equation (8) as the original K-matrix KS (). As a consequence of
()
(S)
that and the fact the entries of K S () are c-numbers it follows that TA () is also a
solution of the reflection equation, namely
()
(S)
(S)
(S)
(S)
L12 ( )TA ()L21 ( + )TA ()
2
(S)
1
(S)
= TA ()L12 ( + )TA ()L21 ( ).

(S)
(S)
(41)
In the next section we will explore such quadratic algebra together with the existence of
(S)
a pseudovacuum on which TA () acts triangularly to present the expressions for eigenvalues of tS () as well as the corresponding Bethe ansatz equations.
4. Algebraic Bethe ansatz

In the next subsections we will consider the diagonalization of the operator tS () in the
most general restrictive condition (I) or (II) by an algebraic formulation of the Bethe ansatz.
The other two situations mentioned in Section 3 are special cases and the corresponding
eigenvalues and Bethe ansatz results can be derived from the results, for example, obtained
for manifold (I). This is obvious when GS(+) = GS() and in the case KS() () = Id one needs
()
to consider the limit with fixed S = 1 in the results to be given bellow.
4.1. The spin- 12 solution
Here we shall consider the diagonalization of the double-row transfer matrix t1/2 () by
means of the quantum inverse scattering method [5,7]. The corresponding bulk Boltzmann
weights (3) are those of the isotropic six-vertex model,
+ 0 0
0

0
0
(1/2)
L12 () =
(42)
.
0

0
0
0 0 +
Following the remarks of Section 3 we are assuming that the boundary couplings c and
d satisfy one of the two possible constraints described by Eqs. (37), (38). In this situation
()
the effective K S () K-matrix is triangular and its diagonal entries are proportional to the
574
()
eigenvalues fj ( 12 ; , ). Without losing generality one can clearly consider the case in
()
which K () is upper triangular, and after some simplifications in Eq. (31) we find that
1/2

()
()
K 1 () = 1
2
(+)
f1 ( 12 ; , )
()
12
g22
(+)
g11
f2 ( 12 ; , )
()

.
(43)
The off-diagonal term in Eq. (43) is not expected to affect the eigenvalues of t1/2 () but
it will certainly be relevant in the structure of the eigenvectors. The explicit expression for
12 has been presented in Appendix A. As discussed in Section 3 a direct consequence of
the upper triangular property of K S () is that the following SU(2) highest state vector
|0 S =
L

|S, Sj ,
j =1

1
0
|S, S =
...
0
(44)
2S+1
is an exact eigenvector of the double transfer matrix tS ().

This means that the state |0 1/2 can be used as pseudovacuum to build up the other
eigenvectors of t1/2 () following the strategy of the algebraic Bethe ansatz approach [5,7].
(1/2)
A main step in this method involves writing the double-row monodromy matrix TA ()
in the 2 2 form
(1/2)
=
TA
A()
C()

B()
.
D()
(45)
By using the intertwining relation (41) and following the procedure devised first by
Sklyanin [7] one can derive the commutation rules
A()B() =
( + )( + )
B()A()
B()A()
( + + )( )
(2 + ) ( )
B()D(),
( + + )
(46)
2( + )
( + + 2)( + )
B()D()
B()D()
D()B()
=
( )( + + )
(2 + )( )
+
4( + )
B()A(),
( + + )(2 + )(2 + )
(47)
where the new field D()

is introduced in order to simplify the commutation relations. It
is given by the following combination between the operators A() and D()
D()
= D()
A().
2 +
(48)
575
In terms of the operators A() and D()

the double-row transfer matrix eigenvalue
problem can now be written as

(+) 1
(+) 1
(+) 1
; , + + f2
; , +
A()| + f2
; , + D()|
f1

2
2
2 +
2
1/2 ()
(49)
|,
=
(+)
1/2
while the action of the fields A(), D()

and C() on the reference state |0 1/2 are given
by

( + )2 L
() () 1
01 ,
A()0 1 = 1 f1
(50a)
; ,
2
2
2
1/2 ()
2

()
() 1
() 1

D() 0 1 = 1 f2
; , f1
; ,
2
2
2
2 +
2
L

2

0 1 ,
(50b)
2
1/2 ()

C()0 1 = 0.
(50c)
2
The fields B() are interpreted as a kind of creation operators over the pseudovacuum
|0 1/2 and the multiparticle Bethe states | n (1 , . . . , n ) are supposed to be given by

n (1 , . . . , n ) = B(1 ) B(n )0 1 .
(51)
2
The rapidities j will be determined by solving the eigenvalue problem with the above
ansatz for the eigenvectors. This is done with the help of the commutation relations (46),
(47) to move A() and D()

in Eq. (49) over the creation fields until they reach the reference state |0 1/2 . The terms proportional to the eigenvectors (51) are easily collected by
keeping only the first pieces of the commutation rules. After using expressions (50a), (50b)
and some simplifications we find that the final result for the eigenvalues are

1/2 ()
( + )2 L 2( + )( + )( + + )
=
(+) ()
1/2 ()
(2 + )2

1/2 1/2
n

(j + 2 ) (j + 2 )
(j 2 ) (j + + 2 )
j =1
L

2( + )( + + + )
2
+
1/2 ()
(2 + )2
n
3

(j 3
2 ) (j + + 2 )
,
(j 2 ) (j + + 2 )
(52)
j =1
()
where we have used the values of fj ( 12 ; , ) taken from Eqs. (20a), (20b) and performed the displacements i i /2 on the rapidities.
576
1 , . . . , n ) can be canceled out

The remaining terms that are not proportional to |(
by imposing further restrictions on the rapidities j . These are known as the Bethe ansatz
equations which in our case are given by

j + + + 2
j + 2
j + 2 2L
=
j 2
j + 2
j + 2
n

(j i + ) (j + i + )
.
(j i ) (j + i )
(53)
i=1
i=j
We now can derive similar results for the open spin- 12 chain that commutes with the
double-row transfer matrix t1/2 (). The corresponding Hamiltonian is proportional to the
first-order expansion of t1/2 () in the spectral parameter [7,16]
H1 =
2
L1

1 z
1 x x
y y
z
i i+1 + i i+1 + iz i+1
+
1 + c 1+ + d 1
i=1

1 z
L + c+ L+ + d+ L ,
+
(54)
where x , and z are the Pauli matrices with = 12 (x i ). Its eigenvalues En ()

are obtained in terms of the rapidities j that satisfy the Bethe ansatz equation (53) by the
following expression
y
En () = 2
n
k=1
2k
2
4

(+)
()
1/2
1/2
L 1
.
1+
(55)
We would like to close this section with the following remark. The ferromagnetic < 0
Hamiltonian (54) is known to describe the stochastic dynamics of symmetric hopping of
particles in one dimension provided that certain relations are satisfied by the boundary parameters [20]. More specifically, letting ( ) be the rate of injection (ejection) of particles
at the left boundary and () the corresponding rate at the right boundary we have [20,21]
d
,
2
1
,
1
,
+
c
,
2
(56)
and
c+
,
2+
d+
.
2+
(57)
The above particular parameterization of the boundary parameters c and d satisfies

the constraints (I) or (II) for arbitrary values of the particle injection and ejection rates.
Though the spectrum at this special case have been determined before [20,21] not much is
known about the behavior of the wave functions. This information can now be in principle
extracted by combining the unitary transformation (32), (36) with the multiparticle state
structure (51). This knowledge of the eigenvectors can be used to calculate correlation
functions, thanks to recent developments made in the quantum inverse scattering method
577
[22,23] which allows us to reconstruct local spin operators in terms of the monodromy
matrix fields. We hope to return to this problem since this could provide us with new
insights on the physics of stochastic dynamics of interacting particle systems.
4.2. The spin-1 solution
The statistical system associated to the integrable XXX-Heisenberg model with spin-1
is a three-state vertex model with nineteen non-null Boltzmann weights given by
a()
0
0
0
0
0
0
0
0
b()
0
h()
0
0
0
0
0
0
0
e()
0
d()
0
c()
0
0
0
h()
0
b()
0
0
0
0
0
0
(1)
L12 () = 0
0
d()
0
g()
0
d()
0
0 ,
0
0
0
0
b()
0
h()
0
0
0
c()
0
d()
0
e()
0
0
0
0
0
0
0
0
h()
0
b()
0
0
0
0
0
0
0
0
0
a()
(58)
with
b()
= ,
a()
= + 2,
e()
( )
,
+
c()
22
,
+
g()
= b()
+ c(),
d()
=
,
+
h()
= 2.
(59)
(60)
As before we can consider the situation when the effective K 1 () matrix is upper triangular. In this case, carrying on few algebraic simplifications in Eq. (31) we find that
(+)

g (+) 2
g
()
K 1 () = 1
f1 (1; , )
22 ( 1 )
12 2
23
(+)
g11
f2 (1; , )
213
(1 )
2
,
g22
1
(+) 12 ( 2 ) + 23
g11
(+)
22
(+)
g11
f3 (1; , )
(61)
where the off-diagonal coefficients 12 , 13 and 23 have been collected in Appendix A.
At this point we need to start introducing suitable notation for the double monodromy
(S)
operator TA (). Here we shall use a representation which can be easily extended to accommodate arbitrary spin-S case,

A1 () B12 () B13 ()
(1)
TA () = C21 () A2 () B23 () .
(62)
C31 () C32 () A3 ()
The next step is to rewrite the eigenvalue problem in terms of the double monodromy
matrix elements. To perform this task is convenient to define new diagonal operators A i ()
in terms of appropriate linear combinations of the fields Ai () [7,24]. This is done in such
way that the action of the new fields on the state |0 1 will be proportional to a single bulk
578
term. Keeping in mind possible extension to general values of the spin we define,
A1 () = A 1 (),
h(2)
A2 () = A 2 () +
A 1 (),
a(2)
c(2)
h 1 (2)
A3 () = A 3 () +
A2 (),
A 1 () +
a(2)
h 2 (2)
where the functions h 1 () and h 2 () are the following determinants

a()
a()
c()
h()

.

,
h 2 () =
h 1 () =

h() h()
h() g()
(63)
(64)
(65)
(66)
Taking into account the representation (62) and the above redefinitions of the diagonal
fields, the diagonalization of the doubled transfer matrix t1 () becomes equivalent to the
problem
3

1 ()

(+)
i ()A i () n (1 , . . . , n ) = (+) n (1 , . . . , n ) ,
1
i=1
(67)
with
h(2)
c(2)
(+)
(+)
f2 (1; , + ) +
f3 (1; , + ),
a(2)
a(2)
h 1 (2) (+)
2(+) () = f2(+) (1; , + ) +
f (1; , + ),
h 2 (2) 3
(+)
(+)
() = f (1; , + ).
(+)
(+)
1 () = f1
(1; , + ) +
(68)
(69)
(70)
Another important ingredient is to determine the action of the double monodromy matrix elements on the pseudovacuum |0 1 . This can be done with the help of the YangBaxter
(1)
()
algebra [7,24] and the triangularity properties of both Lab () and K 1 () operators upon
|0 1 . Following Ref. [24] and taking into account Eq. (61) we have

a()
2 L
() ()
A 1 ()|0 1 = 1 1 ()
|01 ,
1 ()
2
L
b()
() ()
|0 1 ,
A2 ()|01 = 1 2 ()
1 ()

e()
2 L
|01 ,
A 3 ()|0 1 = 1() 3() ()
1 ()
C21 ()|0 1 = C31 ()|0 1 = C32 ()|0 1 = 0,
(71)
(72)
(73)
with
()
()
1 () = f1
(1; , ),
(74)
579
h(2)
()
f1 (1; , ),
(75)
a(2)
h 1 (2) ()
h 3 (2) ()
()
()
3 () = f3 (1; , )
f (1; , )
f (1; , ),
h2 (2) 2
h 2 (2) 1
(76)
c()

h()
.
where the new function h 3 () =
()
()
2 () = f2
(1; , )
h()
g()
Also one expects that the operators B12 (), B13 () and B23 () play the role of creation
operators over the reference state |0 1 . Therefore it is natural to seek for other eigenvectors
of t1 () as linear combinations of products of these creation fields acting on |0 1 . This
is done by exploring the commutation rules between the diagonal A i () and the creation
fields which can be derived from the boundary YangBaxter algebra (41). A careful analysis of these relations reveals us that the construction of the eigenvectors can be based on
either B12 () and B13 () or B23 () and B13 () pair of fields rather than on arbitrary combination of the three possible creation operators. We remark that this redundancy is not
particular of this system, but it is a general feature of the algebraic Bethe ansatz framework
developed in Refs. [25,26] for a large family of integrable vertex models with periodic
boundary. This formalism has been generalized by Li et al. [27] to include vertex models
with open boundaries based on ideas first envisaged by Fan [24] and later extended for
systems solvable by nested Bethe ansatz [28]. We also recall that in the context of three
state vertex models this approach was recently reviewed in Ref. [29]. Considering that such
algebraic framework has already been described in these references, we shall not repeat the
details here, and in what follows we will present only the main results for the eigenvectors
and the eigenvalues. Here we consider that the eigenvectors will be constructed in terms
of a linear combination of products of the creation fields B12 () and B13 () acting on the
vector |0 1 . It turns out that the eigenstates | n (1 , . . . , n ) form a multiparticle structure
and they can be constructed as

n (1 , . . . , n ) = n (1 , . . . , n )|0 1
(77)
such that the vector n (1 , . . . , n ) satisfy a second order recursion relation of the form
n (1 , . . . , n ) = B12 (1 )n1 (2 , . . . , n )
n

+ B13 (1 )
n2 (2 , . . . , i1 , i+1 , . . . , n )
i=2
(i)

(i)
1 (1 , . . . , n )A 1 (i ) + 2 (1 , . . . , n )A 2 (i ) .
Here we are assuming the identification | 0 |0 1 . The functions
and 2(i) (1 , . . . , n ) are given by
(i)
1 (1 , . . . , n ) =
(78)
(i)
1 (1 , . . . , n )
i1

1 i )
h 4 (j i )
d(
p(
1 , i )
1 + i )
a(
j i )e(
j i )
b(
j =2
n

k=2
k=i
k + i )
a(
k i )b(
,
k i )a(
k + i )
b(
(79)
580
and
(i)
2 (1 , . . . , n ) =
i1
1 + i )
h 4 (j i )
d(
1 + i )
a(
j i )e(
j i )
b(
j =2
n

k=2
k=i
h 4 (i k )
h 2 (k + i )
,
k + i )
i k )e(
i k ) a(
k + i )b(
b(
(80)
with
+ y)
d(x
e(x
+ y) h(2y)
p(x,
y) =
y)
e(x
y) a(2y)
d(x

g()
and h 4 () =
d()

d()
.

e()
1 , . . . , n ) is performed
The action of the doubled transfer matrix t1 () on the state |(
relying on similar data for the (n 1) and (n 2) particle states and with help of mathematical induction. Adapting the discussion of Refs. [27,29] to our case we can infer that
the eigenvalue expression is
1 ()
(+) ()
1 1
n

j + )
j )b(
a()
2 L a(
(+)
()
= 1 ()1 ()
j )a(
1 ()
j + )
b(
j =1
2
L
n
h 4 ( j )h 2 ( + j )
b()
(+)
()
+ 2 ()2 ()
j )a(
+ j )
1 ()
e(
j )b(
+ j )b(
j =1

(+)
()
+ 3 ()3 ()
e()
2
1 ()
L
n
j =1
j )h 5 ( + j )
b(
,
+ j )
e(
j )e(
+ j )b(
(81)
d()

and provided that the rapidities j satisfy the following Bethe
where h 5 () = b()
d() b()
ansatz equations

a(
j)
j)
b(
2L
n
(+)
()
j )]2
j i )h 2 (j + i )
1 (j )1 (j ) [b(2
b(
.
=
(+)
()
j + i )]2
e(
j i )[b(
2 (j )2 (j ) h2 (2j )
i=1
(82)
i=j
Now we are almost ready to get standard expressions for the eigenvalues and Bethe
ansatz equations. By introducing the new set of variables i = i and performing many
simplifications in the functions entering Eqs. (81), (82) we conclude that the eigenvalues
are
581
1 ()
(+) ()
1 1

(2 + 3)( + 2 )( + + 2 )( + 2 )( + + 2 )
=
(2 + )4

n
( + 2)2 L (j + ) (j + )
1 ()
(j ) (j + + )

+
j =1
( + 2 )( +
1 ()
L
n
j =1
2 )( +
4
2 )( + + +
2 )
(j + ) ( + j ) ( j + 2) ( + j + 2)
(j ) ( + j + ) ( j )
( + j )
(2 )( + 2 )( + 3
2 )( + + + 2 )( + + + 2 )
(2 + )4
() 2 L n
( j + 2) ( + j + 2)
( + )
(83)
,
1 ()
( j )
( + j )
j =1
while the Bethe ansatz equations for the new rapidities i becomes

j +
j
2L

=

j + + + 2
j + 2
j + 2
j + 2
n

(j i + ) (j + i + )
,
(j i ) (j + i )
(84)
i=1
i=j
where are taken from Eq. (21) with S = 1.

We finally remark that the results of this subsection offers us in principle the basis to
solve the O(3) non-linear sigma model with non-diagonal open boundaries. Due to the
isomorphism O(3) SU(2)2 the elements of the operator (58) can indeed be interpreted
as the scattering amplitudes of the S-matrix associated to the O(3) field theory [30]. One
expects that similar relation is also valid for the boundary scattering matrices [31]. In this
case, we need to adapt our results to include the solution of the eigenspectrum of an open
transfer matrix in the presence of inhomogeneities, following, for example, the lines of
Ref. [32]. It would be interesting to exploit this possibility to determine the effects of the
boundaries in the physics of the O(3) model.
4.3. The spin-S solution
The classical analogue of the solvable spin-S XXX model is the 2S + 1-state vertex
model (3) having the total number of 13 (2S + 1)[2(2S + 1)2 + 1] non-null Boltzmann
()
weights. The transformed upper triangular K S () matrix corresponding to the left bound-
582
ary in tS () is
f () (S; , )
()
f2 (S; , )
()
f3 (S; , )
.
.
.
0
.
.
.
0
()
2S ()
KS () = () S
.
.
.
..
.
0
()
f2S+1 (S; , )
(85)
where denotes non-vanishing values that can be directly determined from Eq. (31).
To implement the quantum inverse scattering framework we will represent the doubled
monodromy matrix by the following structure
A1 ()
B12 ()
B1(2S+1) ()
C21 ()
A2 ()
B2(2S+1) ()
..
(S)
TA () =
.
.
.
.
.
..
..
..
B2S(2S+1) ()
A2S+1 ()
C(2S+1)1 () C(2S+1)2 () C(2S+1)2S ()
(86)
The next step in the algebraic formulation consists in determining the action of the
(S)
TA () elements on the reference state |0 S which helps us to distinguish creation and annihilation fields as well as to reformulate the eigenvalues problem in terms of appropriate
linear combinations of diagonal fields. To perform that we need to know certain commu(S)
(S)
tation relations between the operators TA () and [TA ()]1 . This can be obtained by
using Eq. (2) with = [24] to get the following general matrix relation
(S)
1 (S)
(S)
1
(S)
(S)
(S)
T2 () L12 (2)T1 () = T1 ()L12 (2) T2 () .
(87)
By applying both sides of relation (87) on the pseudovacuum |0 S and by taking into
(S)
()
account the upper triangular property of both L12 (2) and K S () when acting on the
state |0 S we conclude that all the fields C () are annihilators,
C ()|0 S = 0
(88)
while B () acts as creation fields upon |0 S .

We also see that Ai ()|0 S for i = 2, . . . , 2S + 1, turns out to be proportional to many
distinct bulk terms of the form [ti ()]2L since it involves the action of upper elements of
the operator [TA(S) ()]1 on |0 S . In the specific case of the XXX-S model the functions
ti () are
ti () = ( + 2S)
S

k=Si+2
+ k S
.
+ k + S
(89)
As remarked in previous sections this difficulty can be circumvented by writing the

fields Ai () as linear combinations of new operators A i () such that their action on |0 S
is proportional only to [ti ()]2L term. The solution of this problem involves a considerable
583
amount of algebraic work but the final answer can fortunately be given in terms of the
(+)
determinants of certain j j matrices that shall be denoted by Mj,i (). Its elements
(S)
are determined in terms of the entries of the L12 () operator. More precisely, by writing
2S+1
(S)
c,d
L12 () = abcd=1
Ra,b
()ecb ead we find that such linear combination is
Ai () =
i |M (+) (2)|

j,i
(+)
j =1 |Mj,j (2)|
A j (),
(90)
(+)
where the j j matrix Mj,i () is given by
1,1
R1,1
()
1,1
R2,2 ()
(+)
Mj,i () =
..
.
1,1
()
Rj,j
j 1,j 1
2,2
R1,1
()
R1,1
2,2
R2,2
()
..
.
R2,2
2,2
Rj,j
()
Rj,j
j 1,j 1
i,i
R1,1
()
()
i,i
R2,2
()
..
()
..
.
j 1,j 1
i,i
Rj,j
()
()
(91)
j j
By using the relation (90) and the action of all Ai () on the reference state we find that
2
ti () L
() ()
|0 S ,
Ai ()|0S = S i ()
(92)
S ()
where
()
i () = ()2S
()
fi (S; , )
()
i1

|Mi1,k
(2)|
(+)
k=1
|Mi1,i1 (2)|

()
fk (S; , )
(93)
()
while the entries of a second j j auxiliary matrix Mj,i () are given by
1,1
R1,1
()
1,1
R2,2
()
()
Mj,i () = .
..
1,1
()
Rj,j
j +1,j +1
2,2
R1,1
()
i1,i1
R1,1
()
R1,1
2,2
R2,2
()
i1,i1
R2,2
()
R2,2
.
.
.
j +1,j +1
.
.
.
2,2
Rj,j
()
i1,i1
Rj,j
()
()
i+1,i+1
R1,1
()
()
i+1,i+1
R2,2
()
.
.
.
j +1,j +1
Rj,j
.
.
.
()
i+1,i+1
Rj,j
()
j,j
R1,1 ()
R2,2 ()
j,j
.
.
.
j,j
Rj,j ()
j j
(94)
Equipped with Eq. (90) one can now write the eigenvalue problem (27) in terms of the
new diagonal fields A i (), namely
2S+1
=
k ()A k ()|
(+)
k=1
S ()
|,
(95)
fi (S; , + ).
(96)
(+)
where
k(+) () =
2S+1
|Mk,i (2)|
i=k
|Mk,k (2)|
(+)
(+)
584
At this stage we would like to emphasize that the role construction presented above
is applicable to any multistate vertex model whose Boltzmann weights are invariant by
one U (1) charge conservation symmetry. In order to get manageable expressions for the
eigenvalues, however, one still needs to carry on cumbersome simplifications on the general
formulae given in Eqs. (93), (96). In the case of the XXX-S model we are able to show that
all the contributions to i() () miraculously factorized in the following product forms
()
()
i () = i
()
()i
(, ),
(97)
where
(+)
() =
i

2 + [2S + 3 i k]
,
2 + [2 k]
(98)
k=1
()
() = ()2S
2S+1

k=i
2 + [2 2i + k]
,
2 + [1 + k i]
(99)
and
(+)
i (, + ) =
2S+1i

+ + S +
j =1
()
i (, ) =
i1
1
3
+ + S + j +
j
,
2
2S

1
+ S + j +
2
j =i
2S

+ S
j =2S+2i
(100)
j =1
1
j
.
2
(101)
Before proceeding with further results we stress that the above explicit expressions for
()
i () with arbitrary S are novel in the literature since they were unknown even in the
case of diagonal boundaries [10]. Now we reached a point in which we have gathered the
basic ingredients to start an algebraic Bethe ansatz analysis of the eigenspectrum of tS ().
In particular the vector |0 S is itself an eigenstate of tS () with the eigenvalue
(0)
S ()
(+) ()
S S
2S+1

i=1

(+)
()
i ()i ()
ti2 ()
S ()
L
.
(102)
The other eigenvectors of tS () are looked as states created by the action of the fields
B () on the reference state |0 S . A single particle excitation is made by lowering the
value of the azimuthal spin component by an unity on the ferromagnetic pseudovacuum
|0 S . From the point of view of the algebraic Bethe ansatz framework this excitation can
be represented by Bjj +1 (1 )|0 S for any j = 1, . . . , 2S. As far as commutation relations
are concerned we find that it is simpler to choose the one-particle state as

1 (1 ) = B12 (1 )|0 S .
(103)
585
The action of the double transfer matrix tS () on this state can be computed with the
aid of the commutation relations between the fields A i () and B12 () that can be obtained
from the boundary YangBaxter algebra (41). In Appendix B we present details of our
analysis of the one-particle eigenvalue problem for S = 32 . This study together with the
previous results for S = 1 [27,29] and the help of mathematical induction lead us to the
following general expression

tS ()
1 (1 )
(+) ()
S S
=
2S+1
t 2 ()
L (+)

()
i
i ()i ()Qi (, 1 ) 1 (1 )
S ()
i=1
2S
(1)

(2)
Bii+1 () qi (, 1 )A 1 (1 ) + qi (, 1 )A 2 (1 ) |0 S ,
(104)
i=1
where function Qi (, j ) is given by
Qi (, j ) =
1,1
2,1
R1,1 (j )R1,2
(j +)
R 2,1 ( )R 1,1 ( +) , i = 1,
1,2 j
1,1 j
i+1,1
i+1,2
i,2

i1,1

R1,i+1 (j ) R1,i (j ) R1,i1 (+j ) R1,i1 (+j )

i,1
i1,1

i,2
i,2
R
(j ) R
(j )
(+j ) R
(+j )
2,i+1
2,i
2,i
2,i
, i = 2, . . . , 2S,
i,1
i+1,1
i1,1
i,1
R1,i
(j )R1,i+1
(j )
R1,i1
(+j )R1,i
(+j )
R 2S,1 (+ ) R 2S+1,2 (+ )
1,2S
j
j
1,2S

2S,1

2S+1,2
2S+1,2
R
R
(
)
(+
)
R
(+
j
j
j)
2,2S+1
2,2S+1
2,2S+1
, i = 2S + 1.
2S+1,1
2S,1
2S+1,1
R1,2S+1 (j )
R1,2S (+j )R1,2S+1 (+j )
(105)
From (104) we see that the unwanted terms proportional to Bii+1 () can be eliminated
by imposing that the rapidity 1 satisfies the following one-particle Bethe ansatz equation,

t1 (1 )
t2 (1 )
2L
(2)
()
1 (1 )
()
2 (1 )
qi (, 1 )
(1)
qi (, 1 )
i = 1, . . . , 2S.
(1)
(106)
(2)
We note that though the expressions for qi (, 1 ) and qi (, 1 ) have in general a

very involved dependence on the ith index, see, for instance, Appendix B, we have found
(2)
out that the ratio
qi (,1 )
(1)
qi (,1 )
does not depend of such index. Its expression for arbitrary S,
coming directly from the commutation rules, involves many complicated terms and it has
been collected in Appendix D. It turns out, however, that it is possible to carry out further
simplifications in Eqs. (D.1), (D.2) thanks to several identities between the Boltzmann
c,d
weights Ra,b
(). This also leads us to conclude that the ratio
(2)
qi (,1 )
(1)
qi (,1 )
does not depend
on the spectral parameter . This is consistent to what one would expect from a standard
586
Bethe ansatz analysis and the simplified expression for such ratio reads
(2)
qi (, 1 )
(1)
qi (, 1 )
(+)
2 (1 )
(+)
1 (1 )
(1 ),
(107)
where for later convenience we define function () separately, namely,

() =
1,1
2,2
2,2
1,1
R1,1
()R2,2
() R1,1
()R2,2
()
2,1
[R1,2
()]2
(108)
Putting all these results together we find that | 1 (1 ) is an eigenvector of tS () with

eigenvalue 1 (, 1 ) given by
2S+1
(1)
t 2 ()
L (+)
S (, 1 )
()
i
(109)
=
i ()i ()Qi (, 1 ),
(+) ()
()
S
S S
i=1
provided that the variable 1 satisfies the restriction

(+)
()
t1 (1 ) 2L 1 (1 )1 (1 )
= (1 ).
(+)
()
t2 (1 )
(1 ) (1 )
2
(110)
Here we remark that Eq. (110) is equivalent to the condition of analyticity of 1 (, 1 )

as a function of the rapidity 1 . This fact is indeed an extra verification of the validity of
our Bethe ansatz analysis.
We now turn to the analysis of the two-particle state. In this case one expects that this
state should be given in terms of two linearly independent vectors B12 (1 )B12 (2 )|0 S and
B13 (1 )|0 S . Previous experience in determining two-particle states [2426] suggests us
to look first for the commutation rule between the fields B12 (1 ) and B12 (2 ). To avoid
overcrowding this section with more heavier formulae we have exhibited this relation for
S 1 in Appendix D. From Eq. (90) and the observations made in Appendix D we clearly
see that the state

2 (1 , 2 ) = B12 (1 )B12 (2 )
3,2
3,2
(+)
R1,2 (1 + 2 )
R1,2 (1 + 2 ) |M1,2 (22 )|
A2 (2 ) +
+ B13 (1 ) 2,1
2,1
(+)
R1,2 (1 + 2 )
R1,2
(1 + 2 ) |M1,1
(22 )|

3,2
3,1
R1,2 (1 2 ) R1,3 (1 + 2 )
3,1
(111)
A 1 (2 ) |0 S ,
2,1
R1,3 (1 2 ) R1,2
(1 + 2 )
is symmetric under the exchange of the rapidities 1 and 2 . In other words we have that

2 (1 , 2 ) = ZS (1 , 2 ) 2 (2 , 1 ) ,
(112)
where ZS (1 , 2 ) is the following function,
R 3,2 ( ) R 3,1 ( )
1
1
1,2
1,3

1,2
2,2
2,1
R2,1 ( + 1 ) R2,2
(1 ) R2,3
(1 )
.
ZS (1 , 2 ) = 2,1
1,1
3,1
R1,2 ( + 1 ) R1,1
( 1 )R1,3
( 1 )
(113)
587
This state is therefore an educated ansatz for the two-particle vector for general S 1.
Note that it reproduces the previous state for S = 1 [24,27,29] and in Appendix B we have
presented all the needed evidences that it is indeed a suitable eigenvector for S = 32 . The
corresponding eigenvalue can be calculated by keeping only the terms proportional to the
vector B12 (1 )B12 (2 ) coming from the first part of the commutation relations between
the fields A i () and B12 (i ). Taking into account our previous experience with the oneparticle state and the structure of the commutation rules discussed in Appendices B and D
we find that
(2)
S (, {1 , 2 })
(+) ()
S S
2S+1
n=2
t 2 ()
L (+)

()
i
=
i ()i ()
Qi (, j ).
S ()
i=1
j =1
(114)
The associated Bethe ansatz equations are expected to be the condition on the rapidities
such that the residues at the simple poles = 1 , 2 present in functions Qi (, j ) vanish.
This condition is equivalent to the following system of equations

t1 (j )
t2 (j )
2L
= (j )
(+)
()
(+)
()
1 (j )1 (j )
2 (j )2 (j )
n=2

i=1
i=j
Q2 (j , i )
,
Q1 (j , i )
j = 1, . . . , n = 2.
(115)
By the some token, one expects that general multiparticle states can in principle be
constructed in terms of a recurrence relation of order 2S that involves the creation fields
B1j (), j = 2, . . . , 2S + 1. The precise structure of such relation for arbitrary S has however eluded us so far. This by no means prevents us to propose general expressions for the
corresponding eigenvalues and Bethe ansatz equations. In any factorizable theory, it is believed that the two-particle results already contain the main flavour about the content of the
spectrum. This means that the expressions (114) and (115) are expected to be valid for general values of n 2LS. Considering these observations and after working out the explicit
(n)
expressions for functions Qi (, j ) we find that the n-particle eigenvalue S (, {i }) is
given by
(n)
S (, {i })
(+) ()
S S
2S+1
t 2 ()
L (+)
()
i
i ()i ()
S ()
i=1
n

j =1
[ j + (S + 1)][ j S]
[ j + (S + 2 i)][ j + (S + 1 i)]
[ + j + (S + 1)][ + j S]
[ + j + (S + 2 i)][ + j + (S + 1 i)]
while the Bethe ansatz equations are given by
(116)
588
j + S
j S
2L

=

j + + + 2
j + 2
j + 2
j + 2
n

(j i + ) (j + i + )
,
(j i ) (j + i )
(117)
i=1
i=j
where we have performed the displacement i i S in order to bring the Bethe ansatz
equations in a more symmetrical form.
At this point it should be emphasized that the right-hand side of the Bethe ansatz equations (117) depend on both the spin S and the off-diagonal elements c , d through the
renormalized variable defined in Eq. (21). We also mention that we have verified numerically for several values of L and S that Eqs. (116), (117) indeed reproduces the ground
state and few low-lying excitations of the double-row transfer matrix tS (). In particular,
we have been able to check the completeness of the Bethe ansatz solution for L = 2 up to
S = 32 . Finally, we remark that the final results for the eigenvalues (114) and Bethe ansatz
equations (115) are expected to be valid for any integrable vertex model whose underlying
R-matrix possesses an unique U (1) charge symmetry and the invariance (5), (6).
5. Conclusions
The purpose of this paper was to solve the integrable XXX-S Heisenberg model with
open boundary conditions by means of the quantum inverse scattering approach. We first
argued that the corresponding K-matrices are diagonalizable by special similarity transformations without a dependence on the spectral parameter. This fact together with the
property of reversing gauge transformed Boltzmann weights leads us to an eigenvalue problem with only one non-diagonal effective K-matrix. In the cases when such K-matrix are
either upper or lower triangular we managed to present explicit expressions for the eigenvalues of the doubled transfer matrix operator tS () as well as the associated Bethe ansatz
equations for arbitrary values of the spin-S. This condition was shown to be equivalent
to two possible constraints between the four off-diagonal boundary parameters, leading us
with five free parameters out of six possible ones.
We hope that the ideas developed in this paper will be also suitable to solve a broad class
of isotropic integrable systems with non-diagonal open boundaries. In fact, the method devised here has been first applied to the fundamental SU(N ) isotropic vertex model under
more restrictive open boundary conditions [33]. We expect that the nested Bethe ansatz
approach could be further generalized to tackle effective triangular K-matrices which will
provide us the solution of the associated doubled transfer matrix operator with fewer constrained boundary parameters as compared to that presented in Ref. [33]. We also hope
that other vertex models based on higher rank symmetries such as O(N ) and sp(2N ) Lie
algebras could be dealt by the framework discussed in this work. This assumes that certain
classes of non-diagonal K-matrices of these vertex models can be classified in terms of
similarity transformations that are itself symmetries of the corresponding L-operator, acting on spectral dependent diagonal solutions for the reflection equation. This would means
589
that our observation of Section 2 for SU(2) could be generalized to other Lie algebras as
well. We plan to investigate such rather interesting possibility in a future work.
Acknowledgements
The authors C.S. Melo and G.A.P. Ribeiro thank FAPESP (Funda ao de Amparo
Pesquisa do Estado de S ao Paulo) for financial support. The work of M.J. Martins has
been supported by the Brazilian Research Council-CNPq and FAPESP.
Appendix A. The K-matrix properties

In this appendix we briefly summarize the explicit expressions of the K-matrix elements
satisfying the reflection equation (8) for S = 1 and 32 . For S = 1 [17] the corresponding
matrix is given by

K1 () =
k11 ()
k21 ()
k31 ()
k12 ()
k22 ()
k32 ()

k13 ()
k23 () ,
k33 ()
where the elements kij () are given by

1
1
2 +
k11 () = 2 +
4
2

c
1
k12 () = 2 +
,
2
2 2

c2 1
k13 () =
,
4 2

1
d
k21 () = 2 +
,
2
2 2

1
1
k22 () = 2 +
2
4
2

1
c
k23 () = 2 +
,
2
2 2

d2 1
k31 () =
,
4 2

1
d
k32 () = 2 +
,
2
2 2

1
1
k33 () = 2
2 +
4
2
(A.1)

1
cd 1
+
+
,
2
8 2
(A.2)
(A.3)
(A.4)
(A.5)
1
cd 1
+
+
2
4 2

1
+
,
2
(A.6)
(A.7)
(A.8)
(A.9)

1
cd 1
.
2
8 2
(A.10)
590
On the other hand for S =
k11 ()
k21 ()
K 3 () =
k31 ()
2
k41 ()
3
2
we have
k12 ()
k22 ()
k32 ()
k42 ()
k13 ()
k23 ()
k33 ()
k43 ()
k14 ()
k24 ()
,
k34 ()
k44 ()
where the elements kij () are given by

k11 () = cd 3 + 1
18

1
3 +
3 1 +
3 + 1 +
,
+
27

c
k12 () = cd 1
2 3 1 +
3 +
,

18 3

2
k13 () = c 1 3 1 + ,

9 3 2

c3 1
1
k14 () =
,
27 2

d
2 3 1 +
3 +
,
k21 () = cd 1

18 3
3
2

2cd
5
1 3 3
k22 () =
+
+ 3
+
27
4
4
2

1
3 + 1
3 +
3 1 +
,
+
27

2
k23 () = c cd 1 1 + 2 + 4 1 (3 )2 ,
54

2
1

c
k24 () =
3 + 1
,

9 3 2

2
k31 () = d 1 3 1 + ,

9 3 2

2
k32 () = d cd 1 1 + 2 + 4 1 (3 )2 ,
54
3
2

2cd
5
1 3 3
+
k33 () =
3 +
+
27
4
4
2

1
3
3 + 1
3 1 +
,
+
27

k34 () = c cd 1 2 3 3 + 1 ,

18 3
(A.11)
(A.12)
(A.13)
(A.14)
(A.15)
(A.16)
(A.17)
(A.18)
(A.19)
(A.20)
(A.21)
(A.22)
(A.23)
(A.24)
(A.25)

d3 1
1
,
k41 () =
27 2

d2 1
k42 () =
3 + 1
,

9 3 2

k43 () = d cd 1 2 3 3 + 1 ,

18 3

cd
3
1
k44 () =
18

1
3
3 1
3 + 1
.
+
27
591
(A.26)
(A.27)
(A.28)
(A.29)
Next we list the dependence of the off-diagonal coefficients of the transformed

()
K-matrix K S () on the parameters c and d . For S = 12 we find
+ [c+ d + c d+ 2c+ d+ ] + (c d+ d c+ ) 1 + c+ d+
12 =
,
2d+ 1 + c+ d+
(A.30)
while for S = 1 we have

12 =

(2 + c d+ + c+ d )
+ 2(c+ c ) 1 + c+ d+ ,
32 2(1 + c+ d+ )
4 (d d )2 +
+ 0 + 2c+
+
4
c+
2 1
4 [d+
2 3]
2d+ d 2 + d
,
32(1 + c+ d+ )(2 + c+ d+ + 2 + 1 + c+ d+ )2

23 =
2(c+ c ) 1 + c+ d+ + + ,
8 2 1 + c+ d+
13 =
(A.31)
(A.32)
(A.33)
where and i are given by

= 2c c+ (2 c d+ + c+ d ),

4
0 = c+
(d+ d ) 1 + c+ d+ d (2 + c+ d+ ) d+ (2 + c d+ ) ,
(A.34)
1 = 4c+ d+ + c d+ (4 + c d+ ),
(A.36)
= 2c d+ + c+ d+ (6 + c d+ ),
(A.37)
3 = c+ d+ (8 + c+ d+ ).
(A.38)
Appendix B. One and two particle states for S =
(A.35)
3
2
The purpose of this appendix is to present some of the technical details entering the
analysis of the one and two particle states for S = 32 . In order to do that it is convenient to
592
(S)
(S)
work with a new matrix R ab () = Pab Lab () where Pab is the permutator. This matrix
plays a direct role in the quantum inverse scattering method and Eq. (41) is rewritten in
(S)
terms of R ab () as
1
(S)
(S)
(S)
(S)
R 12 (u v)TA (u)R 12 (u + v)TA (v)
1
(S)
(S)
(S)
(S)
= TA (v)R 12 (u + v)TA (u)R 12 (u v).
(B.1)
In order to solve the one-particle problem one first needs to obtain the appropriate
commutation rules between the fields Ai (u) and B12 (v) coming from the boundary Yang
Baxter equation (B.1). Using the symbol [i, j ] to represent the ith row and the j th column
of Eq. (B.1) we conclude that such suitable commutation rules are derivate from the entries [1, 2], [2, 3], [3, 4], [2, 6], [3, 7] and [4, 8]. Further progress are made replacing the
fields Ai (u) by A i (u) in these equations with the help of the relations (90). After several
algebraic manipulations we obtain the following structure
A 1 (u)B12 (v)
= a11 (u, v)B12 (v)A 1 (u) + a21 (u, v)B12 (u)A 1 (v) + a31 (u, v)B12 (u)A 2 (v)
+ a41 (u, v)B13 (v)C21 (u) + a51 (u, v)B13 (u)C21 (v) + a61 (u, v)B13 (u)C32 (v)
+ a71 (u, v)B14 (v)C31 (u) + a81 (u, v)B14 (u)C31 (v)
+ a91 (u, v)B14 (u)C42 (v),
(B.2)
A 2 (u)B12 (v)
+ a42 (u, v)B23 (u)A 1 (v) + a52 (u, v)B23 (u)A 2 (v) + a62 (u, v)B13 (v)C21 (u)
+ a72 (u, v)B13 (v)C32 (u) + a82 (u, v)B13 (u)C21 (v) + a92 (u, v)B13 (u)C32 (v)
2
2
2
+ a10
(u, v)B24 (u)C21 (v) + a11
(u, v)B24 (u)C32 (v) + a12
(u, v)B14 (v)C31 (u)
2
2
(u, v)B14 (v)C42 (u) + a14
(u, v)B14 (u)C31 (v)
+ a13
2
+ a15
(u, v)B14 (u)C42 (v),
(B.3)
A 3 (u)B12 (v)
+ a43 (u, v)B23 (u)A 1 (v) + a53 (u, v)B23 (u)A 2 (v) + a63 (u, v)B34 (u)A 1 (v)
+ a73 (u, v)B34 (u)A 2 (v) + a83 (u, v)B13 (v)C21 (u) + a93 (u, v)B13 (v)C32 (u)
3
3
3
+ a10
(u, v)B13 (v)C43 (u) + a11
(u, v)B13 (u)C21 (v) + a12
(u, v)B13 (u)C32 (v)
3
3
3
+ a13
(u, v)B24 (u)C21 (v) + a14
(u, v)B24 (u)C32 (v) + a15
(u, v)B14 (v)C31 (u)
3
3
+ a16
(u, v)B14 (v)C42 (u) + a17
(u, v)B14 (u)C31 (v)
3
+ a18
(u, v)B14 (u)C42 (v),
(B.4)
593
A 4 (u)B12 (v)
+ a44 (u, v)B23 (u)A 1 (v) + a54 (u, v)B23 (u)A 2 (v) + a64 (u, v)B34 (u)A 1 (v)
+ a74 (u, v)B34 (u)A 2 (v) + a84 (u, v)B13 (v)C21 (u) + a94 (u, v)B13 (v)C32 (u)
4
4
4
+ a10
(u, v)B13 (v)C43 (u) + a11
(u, v)B13 (u)C21 (v) + a12
(u, v)B13 (u)C32 (v)
4
4
4
(u, v)B24 (u)C21 (v) + a14
(u, v)B24 (u)C32 (v) + a15
(u, v)B14 (v)C31 (u)
+ a13
4
4
(u, v)B14 (v)C42 (u) + a17
(u, v)B14 (u)C31 (v)
+ a16
4
(u, v)B14 (u)C42 (v).
+ a18
(B.5)
Before proceeding we would like to remark that several identities between the Boltzmann weights have been used in order to obtain relations (B.2)(B.5). We also note that
j
many of the coefficients ai (u, v) are proportional to annihilation operators and not all of
them are relevant in the calculations. In Appendix C we have listed only those that indeed
play an important role in our analysis since in general they are sufficiently cumbersome.
By applying Eqs. (B.2)(B.5) on the pseudovacuum |0 3/2 we see that one can rearrange
the action of the double transfer matrix t3/2 () on the one-particle state B12 (1 )|0 3/2 as in
(1)
(2)
Eq. (104). Furthermore, it turns out that the functions Qi (, 1 ), qi (, 1 ) and qi (, 1 )
can therefore be explicitly read off, namely
Qi (, 1 ) = a1i (, 1 ),
qi(1) (, 1 ) =
4
(B.6)
j(+) ()a2i (, 1 ),
(B.7)
(B.8)
j =i
(2)
qi (, 1 ) =
4
(+)
j ()a2i+1 (, 1 ),
j =i
(+)
where i = 1, . . . , 4 and function i () has been defined in Eq. (96).

(2)
As mentioned in the main text the ratio
qi (,1 )
(1)
qi (,1 )
is independent of the ith index and of
the spectral parameter . In our case this ratio is given by

(2)
qi (, 1 )
qi(1) (, 1 )
(+)
2 (1 ) (4 + 1 )
.
(+) (1 ) (2 + 1 )
(B.9)
Next we turn to the two-particle state. In order to obtain an ansatz to this vector we have
considered the commutation rules [1, 3] and [1, 6] coming from Eq. (B.1). Acting these
relations on |0 3/2 leads us to the following expression

B12 (u)B12 (v) + B13 (u) 2 (u, v)A 2 (v) + 1 (u, v)A 1 (v) 0 3
2

= Z 3 (u, v) B12 (v)B12 (u) + B13 (v) 2 (v, u)A2 (u) + 1 (v, u)A 1 (u) 0 3 ,
2
(B.10)
594
where functions 1 (u, v), 2 (u, v) and Z3/2 (u, v) are given by
1 (u, v) =
4 3v
,
( u + v)(3 + 2v)
Z 3 (u, v) =
32 + 2(u v) (u v)2
.
(3 + u v)( u + v)
2 (u, v) =
2 3
,
2 + u + v
(B.11)
(B.12)
From Eq. (B.10) it follows that an appropriate two-particle state should be

2 (1 , 2 )

= B12 (1 )B12 (2 ) + B13 (1 ) 2 (1 , 2 )A 2 (2 ) + 1 (1 , 2 )A 1 (2 ) 0 3 ,
2
(B.13)
since it is symmetric | 2 (1 , 2 ) = Z3/2 (1 , 2 )| 1 (2 , 1 ) under the exchange of rapidities.
The next step is to solve the eigenvalue problem for the two-particle state (B.13). In
order to do that we need extra commutations rules between the fields A i (u) and B13 (v),
B12 (u) and Bjj +1 (v), Cj +1j (u) and B12 (v). In the case of the fields Cj +1j (u) and B12 (v)
the rules comes from the entries [2, 5], [3, 6] and [4, 7] of Eq. (B.1) and the ones for
the other operators are obtained from [1, 3], [2, 4]; [1, 6], [2, 7], [3, 8]; [2, 10], [3, 11]
and [4, 12] entries. After long algebraic manipulations we are able to obtain the following
expressions

A i (u)B13 (v)0 3
2
= b1i (u, v)B13 (v)A i (u)0 3

2
+ unwanted terms, i = 1, . . . , 4,
(B.14)

C21 (u)B12 (v)0 3
2
1
1
1
(u, v)A 2 (v)A 1 (u) + c12

(u, v)A 1 (v)A 2 (u)
= c22 (u, v)A2 (v)A 2 (u) + c21

1
(u, v)A 1 (v)A 1 (u) 0 3 ,
+ c11
(B.15)
2

C32 (u)B12 (v)0 3
2
2
2
2
(u, v)A 2 (v)A 3 (u) + c22
(u, v)A 2 (v)A 2 (u) + c21
(u, v)A 2 (v)A 1 (u)
= c23
2
2
+ c13
(u, v)A 1 (v)A 3 (u) + c12
(u, v)A 1 (v)A 2 (u)

2
+ c11
(B.16)
(u, v)A 1 (v)A 1 (u) 0 3 ,
2

C43 (u)B12 (v)0 3
2
3
3
3
= c24
(u, v)A 2 (v)A 4 (u) + c23
(u, v)A 2 (v)A 3 (u) + c22
(u, v)A 2 (v)A 2 (u)
3
3
3
(u, v)A 2 (v)A 1 (u) + c14
(u, v)A 1 (v)A 4 (u) + c13
(u, v)A 1 (v)A 3 (u)
+ c21

3
3
+ c12
(B.17)
(u, v)A 1 (v)A 2 (u) + c11
(u, v)A 1 (v)A 1 (u) 0 3 ,
2
595

B12 (v)B12 (u)0 3
2
1

= d2 (u, v)B13 (v)A 2 (u) + d11 (u, v)B13 (v)A 1 (u) + unwanted terms 0 3 ,
2
(B.18)

B12 (v)B23 (u)0 3
2
2
= d3 (u, v)B13 (v)A 3 (u) + d22 (u, v)B13 (v)A 2 (u) + d12 (u, v)B13 (v)A 1 (u)

+ unwanted terms 0 3 ,
(B.19)
2

B12 (v)B34 (u)0 3
2

= d43 (u, v)B13 (v)A 4 (u) + d33 (u, v)B13 (v)A 3 (u) + d23 (u, v)B13 (v)A 2 (u)

+ d13 (u, v)B13 (v)A 1 (u) + unwanted terms 0 3 ,
(B.20)
2
where by unwanted terms we mean those that do not give contributions proportional to
k (u, v) and d j (u, v) are once again very involved
| 2 (1 , 2 ). The functions b1i (u, v), cij
i
and have been presented in Appendix C.
We have now the main ingredients to study the action of the operators A i () on the twoparticle state | 2 (1 , 2 ). Taking into account the commutation rules Eqs. (B.2)(B.5) and
(B.14)(B.20) and after some algebra we conclude that the two-particle wanted terms have
the following structure

t3/2 ()
2 (1 , 2 )
(+)
3/2
= B12 (1 )B12 (2 )
4

i=1
n=2

(+)
i ()
j =1

a1i (, j )

A i ()0 3
2

41

+ B13 (1 )A 4 () 42
2 , {i } A2 (2 ) + 2 , {i } A1 (2 ) 0 32

31

+ B13 (1 )A 3 () 32
2 , {i } A2 (2 ) + 2 , {i } A1 (2 ) 0 32

21

+ B13 (1 )A 2 () 22
2 , {i } A2 (2 ) + 2 , {i } A1 (2 ) 0 32

11

+ B13 (1 )A 1 () 12
2 , {i } A2 (2 ) + 2 , {i } A1 (2 ) 0 3
2
+ unwanted terms,
(B.21)
where functions lk
2 (, {i }) are given by

4k
2 , {i }

(+)
4
3
= 4 () b14 (, 1 )k (1 , 2 ) + a10
(, 1 )ck4
(, 2 )

3
(+)
4
4
3
3
+ a1 (, 1 )a5+k (, 2 )d4 (, 1 ) + 3 () a10 (, 1 )ck4
(, 2 )

3
(, 2 )d43 (, 1 ) ,
+ a13 (, 1 )a5+k
(B.22)
596

3k
2 , {i }
4
(+)
3
2
(, 1 )ck3
(, 2 ) + a94 (, 1 )ck3
(, 2 )
= 4 () a10

4
4
+ a14 (, 1 )a5+k
(, 2 )d33 (, 1 ) + a14 (, 1 )a3+k
(, 2 )d32 (, 1 )

(+)
3
3
2
(, 1 )ck3
(, 2 ) + a93 (, 1 )ck3
(, 2 )
+ 3 () b13 (, 1 )k (1 , 2 ) + a10

3
3
3
3
3
2
+ a1 (, 1 )a5+k (, 2 )d3 (, 1 ) + a1 (, 1 )a3+k (, 2 )d3 (, 1 )

(+)
2
2
+ 2 () a72 (, 1 )ck3
(B.23)
(, 2 ) + a12 (, 1 )a3+k
(, 2 )d32 (, 1 ) ,

2k
2 , {i }
4
(+)
3
2
1
(, 1 )ck2
(, 2 ) + a94 (, 1 )ck2
(, 2 ) + a84 (, 1 )ck2
(, 2 )
= 4 () a10
4
4
+ a14 (, 1 )a5+k
(, 2 )d23 (, 1 ) + a14 (, 1 )a3+k
(, 2 )d22 (, 1 )

3
(+)
4
3
(, 2 )d21 (, 1 ) + 3 () a10
(, 1 )ck2
(, 2 )
+ a14 (, 1 )a1+k
2
1
3
+ a93 (, 1 )ck2
(, 2 ) + a83 (, 1 )ck2
(, 2 ) + a13 (, 1 )a5+k
(, 2 )d23 (, 1 )

3
3
(, 2 )d22 (, 1 ) + a13 (, 1 )a1+k
(, 2 )d21 (, 1 )
+ a13 (, 1 )a3+k

(+)
2
1
+ 2 () b12 (, 1 )k (1 , 2 ) + a72 (, 1 )ck2
(, 2 ) + a62 (, 1 )ck2
(, 2 )

2
2
2
2
2
1
+ a1 (, 1 )a3+k (, 2 )d2 (, 1 ) + a1 (, 1 )a1+k (, 2 )d2 (, 1 )

(+)
1
1
(, 2 ) + a11 (, 1 )a1+k
(, 2 )d21 (, 1 ) ,
+ 1 () a41 (, 1 )ck2
(B.24)

1k
2 , {i }
4
(+)
3
2
1
(, 1 )ck1
(, 2 ) + a94 (, 1 )ck1
(, 2 ) + a84 (, 1 )ck1
(, 2 )
= 4 () a10
4
4
+ a14 (, 1 )a5+k
(, 2 )d13 (, 1 ) + a14 (, 1 )a3+k
(, 2 )d12 (, 1 )

3
(+)
4
3
(, 2 )d11 (, 1 ) + 3 () a10
(, 1 )ck1
(, 2 )
+ a14 (, 1 )a1+k
2
1
3
+ a93 (, 1 )ck1
(, 2 ) + a83 (, 1 )ck1
(, 2 ) + a13 (, 1 )a5+k
(, 2 )d13 (, 1 )

3
3
(, 2 )d12 (, 1 ) + a13 (, 1 )a1+k
(, 2 )d11 (, 1 )
+ a13 (, 1 )a3+k

(+)
2
1
+ 2 () a72 (, 1 )ck1
(, 2 ) + a62 (, 1 )ck1
(, 2 )

2
2
2
2
2
+ a1 (, 1 )a3+k (, 2 )d1 (, 1 ) + a1 (, 1 )a1+k
(, 2 )d11 (, 1 )

(+)
1
(, 2 )
+ 1 () b11 (, 1 )k (1 , 2 ) + a41 (, 1 )ck1

1
1
1
+ a1 (, 1 )a1+k (, 2 )d1 (, 1 ) .
(B.25)
It turns out that many identities between the Boltzmann weights can be used in order to
show the following remarkable property
lk
2
n=2

(+)
, {i } = k (1 , 2 )l ()
a1l (, i ).
i=1
(B.26)
597
Considering Eqs. (B.21), (B.26) and (92) together it is not difficult to derive the expression

t3/2 ()
2 (1 , 2 )
(+) ()
3/2 3/2
n=2

4 2

ti () L (+)
()
i
i ()i ()
a1 (, j ) 2 (1 , 2 )
=
3/2 ()
j =1
i=1
+ unwanted terms.
(B.27)
As a final comment we would like to stress that we have also performed extensive checks
verifying that in fact the unwanted terms are canceled out provided the rapidities i satisfy
the restriction (115).
Appendix C. Auxiliary functions for S =
3
2
j
The purpose of this appendix is to list the expressions of the functions ai (u, v), b1 (u, v),
j
lk
ci (u, v) and di (u, v) used in the previous appendix:
(3 u + v)(u + v)
,
(u + v)(3 + u + v)
6v
(C.1)
a21 (u, v) =
,
(u v)(3 + 2v)
3
2 3(3 u + v)(u + v)
a31 (u, v) =
,
a41 (u, v) =
,
3 + u + v
(u + v)(2 + u + v)(3 + u + v)
(C.2)
(
+
u
v)(3
u
+
v)(u
+
v)(4
+
u
+
v)
a12 (u, v) =
(C.3)
,
(u v)( u + v)(2 + u + v)(3 + u + v)
a11 (u, v) =
12v(92 + u(u + v) + (2u + 3v))

(C.4)
,
(3 + 2u)( u + v)(3 + u + v)(3 + 2v)
6(u(u + v) + (u + 3v))
,
a32 (u, v) =
(C.5)
(3 + 2u)(u v)(2 + u + v)
4 3v
2 3
a42 (u, v) =
(C.6)
,
a52 (u, v) =
,
( u + v)(3 + 2v)
2 + u + v
4 3(3 u + v)(u + v)(4 + u + v)(32 3v u(u + v))

2
a6 (u, v) =
,
(3 + 2u)(u v)( u + v)( + u + v)(2 + u + v)(3 + u + v)
(C.7)
4( + u v)(3 u + v)(u + v)(4 + u + v)
2
a7 (u, v) =
(C.8)
,
(u v)( u + v)( + u + v)(2 + u + v)(3 + u + v)

( + u v)(3 u + v)(u + v)(4 + u + v)
,
a13 (u, v) =
(C.9)
( u + v)(2 u + v)( + u + v)(2 + u + v)
a22 (u, v) =
a23 (u, v) =
62 v(202 + 4u(u + v) + 7(u + v))

,
( + u)( + 2u)(2 u + v)(2 + u + v)(3 + 2v)
(C.10)
598
32 (2 + 4u(u + v) + (5u + 7v))

,
( + u)( + 2u)( u + v)( + u + v)
4 3v(132 + 2u(u + v) + 5(u + v))

3
,
a4 (u, v) =
( + 2u)(2 u + v)(2 + u + v)(3 + 2v)
2 3(22 + 2u(u + v) + (u + 5v))

a53 (u, v) =
,
( + 2u)( u + v)( + u + v)
6v
3
a63 (u, v) =
,
a73 (u, v) =
,
(2 u + v)(3 + 2v)
+u+v
a33 (u, v) =
a83 (u, v) =
(C.11)
(C.12)
(C.13)
(C.14)
2 32 ( + u v)(3 u + v)(4 + u + v)(62 4u(u + v) (u + 7v))

( + u)( + 2u)(u v)( u + v)(2 u + v)( + u + v)(2 + u + v)
(C.15)
2 + (u 5v) 2u(u + v))
4(
+
u
v)(3
u
+
v)(4
+
u
+
v)(3
,
a93 (u, v) =
( + 2u)(u v)( u + v)(2 u + v)( + u + v)(2 + u + v)
(C.16)
3(
+
u
v)(3
u
+
v)(4
+
u
+
v)
2
3
,
(u, v) =
a10
(C.17)
( u + v)(2 u + v)( + u + v)(2 + u + v)

( + u v)(4 + u + v)
a14 (u, v) =
(C.18)
,
(2 u + v)( + u + v)
363 ( + u)v
,
( 2u)u( + 2u)( + u + v)(3 + 2v)
183 ( + u)
,
a34 (u, v) =
( 2u)u( + 2u)(2 u + v)
a24 (u, v) =
12 32 ( + u)v
,
( 2u)u( + u + v)(3 + 2v)
2
6 3 ( + u)
,
a54 (u, v) =
( 2u)u(2 u + v)
(C.19)
(C.20)
a44 (u, v) =
12( + u)v
,
( 2u)( + u + v)(3 + 2v)
6( + u)
a74 (u, v) =
,
( 2u)(2 u + v)
(C.21)
a64 (u, v) =
12 33 ( + u)( + u v)(4 + u + v)
,
( 2u)u( + 2u)( u + v)(2 u + v)( + u + v)
122 ( + u)( + u v)(4 + u + v)
,
a94 (u, v) =
( 2u)u( u + v)(2 u + v)( + u + v)
4 3( + u)( + u v)(4 + u + v)
4
,
a10 (u, v) =
( 2u)( u + v)(2 u + v)( + u + v)
a84 (u, v) =
(C.22)
(C.23)
(C.24)
(C.25)
599
(3 + u v)(2 + u v)(u + v)( + u + v)

,
(C.26)
(u v)( + u v)(2 + u + v)(3 + u + v)
( u v)(2 + u v)(2 u + v)(3 u + v)(u + v)(4 + u + v)
b12 (u, v) =
,
(u v)2 ( u + v)( + u + v)(2 + u + v)2
(C.27)
b11 (u, v) =
b13 (u, v) =
( u v)( + u v)(2 + u v)(3 u + v)(3 + u + v)(4 + u + v)

(u v)( u + v)2 ( + u + v)2 (2 + u + v)
( + u v)(2 + u v)(3 + u + v)(4 + u + v)

b14 (u, v) =
,
( u + v)(2 u + v)(u + v)( + u + v)
3
6u
1
1
,
c21
,
(u, v) =
(u, v) =
c22
3 + u + v
(3 + 2u)(u v)
6v
,
(u v)(3 + 2v)
12uv
1
,
(u, v) =
c11
(3 + 2u)(3 + u + v)(3 + 2v)
(C.28)
(C.29)
(C.30)
1
c12
(u, v) =
2 3
,
2 + u + v
2 3(62 + (9u 3v) + 2u(u + v))

2
,
c22 (u, v) =
( + 2u)( u + v)(3 + u + v)
(C.31)
2
c23
(u, v) =
6 32 u
,
( + u)(3 + 2u)(u v)
4 3v
2
,
c13 (u, v) =
( u + v)(3 + 2v)
(C.32)
2
c21
(u, v) =
(C.33)
4 3v(32 + 2u(u v) + 3(u + v))

,
( + 2u)(u v)(2 + u + v)(3 + 2v)
12 32 uv
2
,
c11
(u, v) =
( + u)(3 + 2u)(3 + u + v)(3 + 2v)
2
c12
(u, v) =
(C.34)
(C.35)
6(62 u(u + v) + (4u + 2v))

,
( 2u)(2 u + v)(2 + u + v)
(C.36)
3
c24
(u, v) =
3
,
+u+v
3
c22
(u, v) =
32 (92 + 3(5u + v) 4u(u + v))

,
u( + 2u)( u + v)(3 + u + v)
(C.37)
3
c21
(u, v) =
183 u
,
( + u)( + 2u)(3 + 2u)(u v)
(C.38)
3
c23
(u, v) =
600

3
c14
(u, v) =
6v
,
(2 u + v)(3 + 2v)
(C.39)
3
c13
(u, v) =
12v(u(u v) + (u + 2v))
,
( 2u)( u + v)( + u + v)(3 + 2v)
(C.40)
3
c12
(u, v) =
62 v(4u(u v) + 3(u + v))

,
u( + 2u)(u v)(2 + u + v)(3 + 2v)
(C.41)
363 uv
,
( + u)( + 2u)(3 + 2u)(3 + u + v)(3 + 2v)
2 3
4 3u
1
1
,
d1 (u, v) =
,
d2 (u, v) =
2 + u + v
(3 + 2u)( + u v)
3
c11
(u, v) =
d32 (u, v) =
4
,
+u+v
d22 (u, v) =
(C.42)
(C.43)
4(32 2u(u + v) + (7u + 3v))

,
( + 2u)(u v)(2 + u + v)
(C.44)
122 u
(C.45)
,
( + u)(3 + 2u)( + u v)
2 3
4 3(42 u(u + v) + (3u + 2v))
3
3
,
d3 (u, v) =
,
d4 (u, v) =
u+v
( 2u)( u + v)( + u + v)
(C.46)
2 2
2 3 (6 4u(u + v) + (11u + 3v))
d23 (u, v) =
(C.47)
,
u( + 2u)(u v)(2 + u + v)
12 33 u
3
.
d1 (u, v) =
(C.48)
( + u)( + 2u)(3 + 2u)( + u v)
d12 (u, v) =
Appendix D. Relations for arbitrary S

In this appendix we present certain expressions concerning the unwanted terms of
the one-particle problem as well as the construction of the two-particle vector for arbitrary S.
The commutation rules used in the solution of one-particle eigenvalue problem come
from the entries [1, 2], [2, 3], . . . , [2S, 2S + 1], [2, 2 + 2S + 1], [3, 3 + 2S + 1], . . . , [2S +
1, 2(2S + 1)] of the boundary YangBaxter equation (B.1). To cancel the unwanted terms
(2)
we need to know how to compute the ratio
qi (,1 )
(1)
qi (,1 )
which is not expected to have a depen-
dence on the ith index. This means that this ratio can be calculated collecting the simplest
unwanted contributions which turns out to be those coming from the commutation rules
between the fields A 2S (), A 2S+1 () and B12 (1 ). Considering the help of mathematical
601
(2)
induction we find that the function q2S (, 1 ) is

(2)
(, 1 )
q2S
2S+1,2
R1,2S
( + 1 )
(+)
= 2S () 2S,1
R1,2S ( + 1 )
2S+1,2
( 1 )
R1,2S
2S+1,1
R1,2S+1 ( 1 )

(+)
+ 2S+1 ()
2S,1
R1,2S
(+1 )
(+)
2S+1,2
R1,2S
( + 1 ) |M2S,2S+1 (2)|
2S,1
R1,2S
( + 1 )

2S+1,2
R1,2S
(+1 )

2S+1,2
(+)
|M2S,2S (2)|
2S,1
R2,2S+1
(+1 ) R2,2S+1 (+1 )
(D.1)
2S,1
2S+1,1
R1,2S
( + 1 )R1,2S+1
( + 1 )
(1)
while q2S (, 1 ) is given by

(1)
q2S (, 1 )
(+)
= 2S ()
(+)
2S+1,2
2S+1,1
R1,2S
( 1 )R1,2S+1
( + 1 )
2S+1,2
R1,2S
( + 1 ) |M1,2 (21 )|
2S+1,1
2S,1
2S,1
(+)
R1,2S+1
( 1 )R1,2S
( + 1 )
R1,2S
( + 1 ) |M1,1
(21 )|

(+)
2S+1,2
2S+1,1
|M2S,2S+1 (2)| R1,2S
( 1 )R1,2S+1
( + 1 )
(+)
+ 2S+1 ()
2S+1,1
2S,1
(+)
R1,2S+1 ( 1 )R1,2S ( + 1 )
|M2S,2S (2)|

(+)
2S+1,2
R1,2S ( + 1 ) |M1,2 (21 )|
2S,1
(+)
R1,2S
( + 1 ) |M1,1
(21 )|
R 2S,1 (+ ) R 2S+1,2 (+ )
1
1
1,2S
1,2S
(+)

2S+1,2
2S,1
2S+1,2
|M1,2 (21 )| R1,2S ( 1 ) R2,2S+1
(+1 ) R2,2S+1
(+1 )
2S+1,1
2S,1
2S+1,1
(+)
|M1,1
(21 )| R1,2S+1 ( 1 ) R1,2S ( + 1 )R1,2S+1 ( + 1 )

+
2S,1
2S+1,2
2S+1,2
2S+1,2
2S,1
2S+1,2
R1,2S
(1 )R1,2S
(+1 )R2,2S+1
(1 )R1,2S
(1 )R2,2S+1
(+1 )R1,2S
(1 )
2S+1,1
2S,1
2S+1,1
R1,2S+1
(1 )R1,2S
(+1 )R1,2S+1
(1 )
(D.2)
Taking into account the explicit expressions for the Boltzmann weights and for the
functions j () one can verify that the ratio
(2)
q2S (,1 )
(1)
q2S (,1 )
satisfy Eq. (107).
We close this appendix discussing the construction of the two-particle state. The appropriate commutation relation is derived by combining the entries [1, 3] and [1, 3 + 2S]
of Eq. (B.1). After some algebra we find that commutation relation between the creation
operators B12 (u) and B12 (v) is
B12 (u)B12 (v) +
3,2
R1,2
(u+ )
2,1
R1,2
(u+ )
B13 (u)A2 (v) +
2S+1

j =4
3,1
3,2
2S+1

R1,2
(u ) R1,3
(u+ )
B
(u)A
(v)
+
13
1
3,1
2,1
R1,3
(u ) R1,2
(u+ )
j =4
j,j 1
R1,2
(u+ )
B1j (u)Cj 12 (v)
2,1
R1,2 (u+ )
j,j 2
R1,3
(u+ )
2,1
R1,2
(u+ )
B1j (u)Cj 21 (v)
602

= ZS (u, v) B12 (v)B12 (u) +
1,2
R2,1
(u+ )
B13 (v)A2 (u)

R 3,2 (u ) R 3,1 (u )
1,2 1,3

1,2
1,1
2S+1
1,2
Rj,j
1 (u+ )
B1j (v)Cj 12 (u) +
1,2
j =4 R2,1 (u+ )
"
1,2
R3,2
(u+ )
1,3
2S+1

R3,1
(u+ )
B
(v)A
(u)
+
13
1
1,2
R2,1 (u+ )
j =4
R3,2 (u ) R3,3 (u )
R 3,2 (u ) R 3,1 (u )
1,2 1,3
2,2

2,1
R2,2 (u ) R2,3 (u )
#
1,3
Rj,j
2 (u+ )
B1j (v)Cj 21 (u) ,
1,2
R2,1
(u+ )
(D.3)
where function ZS (u, v) has been defined in Eq. (113).

The above relation allows us to define the following vector
(u, v) = B12 (u)B12 (v) +
+
2S+1

j =4
3,2
R1,2
(u+ )
2,1
R1,2
(u+ )
j,j 1
R1,2
(u+ )
B1j (u)Cj 12 (v)
2,1
R1,2 (u+ )
3,1
3,2
R1,2
(u ) R1,3
(u+ )
3,1
2,1
R1,3
(u ) R1,2
(u+ )
2S+1

j =4
B13 (u)A2 (v)
j,j 2
R1,3
(u+ )
2,1
R1,2
(u+ )
B13 (u)A1 (v)

B1j (u)Cj 21 (v)
(D.4)
which is symmetric under the exchange of the variables u and v, thanks to certain identities
between the Boltzmann weights. More precisely, we have
(u, v) = ZS (u, v)(v, u).
(D.5)
The two-particle state is now obtained by acting the vector (D.4) on the pseudovacuum
|0 S leading us to

R 3,2 ( + 2 )

2 (1 , 2 ) = B12 (1 )B12 (2 ) + 1,2 1
B13 (1 )A2 (2 )
2,1
R1,2
(1 + 2 )

3,2
3,1
R1,2
(1 2 ) R1,3
(1 + 2 )
B13 (1 )A1 (2 ) |0 S .
3,1
(D.6)
2,1
R1,3 (1 2 ) R1,2
(1 + 2 )
Finally, taking into account Eq. (90) we then recover the expression (111) exhibited in
Section 4.3.
References
[1] P.P. Kulish, N.Y. Reshetikhin, E.K. Sklyanin, Lett. Math. Phys. 5 (1981) 393.
[2] H.M. Babujian, Nucl. Phys. B 215 (1983) 317;
L.A. Takhtajan, Phys. Lett. A 87 (1982) 479.
603
[3] K. Sogo, Y. Akutsu, T. Abe, Prog. Theor. Phys. 70 (1983) 730.

[4] R.J. Baxter, Exactly Solved Models in Statistical Mechanics, Academic Press, New York, 1982.
[5] V.E. Korepin, G. Izergin, N.M. Bogoliubov, Quantum Inverse Scattering Method and Correlation Functions,
Cambridge Univ. Press, Cambridge, 1993.
[6] I. Cherednik, Theor. Math. Phys. 61 (1984) 977.
[7] E.K. Sklyanin, J. Phys. A: Math. Gen. 21 (1988) 2375.
[8] L. Mezincescu, R.I. Nepomechie, J. Phys. A: Math. Gen. 24 (1991) 217.
[9] L. Mezincescu, R.I. Nepomechie, V. Rittenberg, Phys. Lett. A 147 (1990) 70;
C.M. Yung, M.T. Bachelor, Nucl. Phys. B 435 (1995) 430.
[10] A. Doikou, Nucl. Phys. B 634 (2002) 591.
[11] R.I. Nepomechie, J. Phys. A: Math. Gen. 37 (2004) 433;
R.I. Nepomechie, J. Stat. Phys. 111 (2003) 1363.
[12] J. Cao, H.Q. Lin, K.J. Shi, Y. Wang, Nucl. Phys. B 663 (2003) 487.
[13] H. Fan, B.Y. Hou, K.J. Shi, Z.X. Yang, Nucl. Phys. B 478 (1996) 723.
[14] R.I. Nepomechie, F. Ravanini, J. Phys. A: Math. Gen. 36 (2003) 11391.
[15] J. de Gier, P. Pyatov, J. Stat. 03 (2004) P002.
[16] H.J. de Vega, A. Gonzalez-Ruiz, Mod. Phys. Lett. A 9 (1994) 2207;
H.J. de Vega, A. Gonzalez-Ruiz, J. Phys. A: Math. Gen. 26 (1993) L519.
[17] T. Inami, S. Odabe, Y.Z. Zhang, Nucl. Phys. B 470 (1996) 419;
A. Lima-Santos, Nucl. Phys. B 644 (2002) 568.
[18] L. Mezincescu, R.I. Nepomechie, J. Phys. A: Math. Gen. 25 (1992) 2533.
[19] G.A.P. Ribeiro, M.J. Martins, Nucl. Phys. B 705 (2005) 521.
[20] R.B. Stinchcombe, G.M. Schutz, Phys. Rev. Lett. 75 (1995) 140.
[21] M.T. Batchelor, in: S.P. Corney, et al. (Eds.), Proceedings of the 22nd International Colloquium on Group
Theoretical Methods in Physics, International Press, Boston, 1999, p. 261.
[22] N. Kitanine, J.M. Maillet, V. Terras, Nucl. Phys. B 554 (1999) 647.
[23] Y.S. Wang, Nucl. Phys. B 622 (2002) 633.
[24] H. Fan, Nucl. Phys. B 488 (1997) 409.
[25] V.O. Tarasov, Theor. Math. Phys. 76 (1988) 793.
[26] M.J. Martins, Nucl. Phys. B 450 (1995) 768;
M.J. Martins, P.B. Ramos, Nucl. Phys. B 500 (1997) 579.
[27] G.L. Li, K.J. Shi, R.H. Shi, Nucl. Phys. B 670 (2003) 401;
G.L. Li, K.J. Shi, R.H. Yue, Nucl. Phys. B 687 (2004) 220.
[28] X.W. Guan, J. Phys. A: Math. Gen. 33 (2000) 5391;
A. Foerster, X.W. Guan, J. Links, I. Roditi, H.Q. Zhou, Nucl. Phys. B 596 (2001) 525.
[29] V. Kurak, A. Lima-Santos, Nucl. Phys. B 699 (2004) 595, nlin.SI/0407006.
[30] A.B. Zamolodchikov, Al.B. Zamolodchikov, Ann. Phys. 120 (1979) 253.
[31] S. Ghoshal, Phys. Lett. B 334 (1994) 363;
E. Corrigan, Z. Sheng, Int. J. Mod. Phys. A 12 (1997) 2825;
M. Moriconi, Nucl. Phys. B 619 (2001) 396.
[32] P. Fendley, H. Saleur, Nucl. Phys. B 428 (1994) 681;
C. Ahn, R.I. Nepomechie, Nucl. Phys. B 586 (2000) 611.
[33] W. Galleas, M.J. Martins, Phys. Lett. A 335 (2005) 167.
The cluster expansion for the self-gravitating gas

and the thermodynamic limit
H.J. de Vega a,b , N.G. Snchez b
a Laboratoire de Physique Thorique et Hautes Energies, Universit Paris VI et VII,
Tour 16, 1er tage, 4 Place Jussieu, 75252 Paris cedex 05, France 1
b Observatoire de Paris, LERMA, 61, Avenue de lObservatoire, 75014 Paris, France 2
Received 5 October 2004; accepted 16 December 2004

Abstract
We develop the cluster expansion and the Mayer expansion for the self-gravitating thermal gas and
prove the existence and stability of the thermodynamic limit N, V with N/V 1/3 fixed. The
essential (dimensionless) variable is here Gm2 N/(V 1/3 T ) (which is kept fixed in the thermodynamic limit). We succeed in this way to obtain the expansion of the grand canonical partition function
in powers of the fugacity. The corresponding cluster coefficients behave in the thermodynamic limit
as (/N )j 1 cj , where cj are pure numbers. They are expressed as integrals associated to tree cluster diagrams. A bilinear recurrence relation for the coefficients cj is obtained from the mean field
equations in the Abels form. In this way the large j behaviour of the cj is calculated. This large
j behaviour provides the position of the nearest singularity which corresponds to the critical point
(collapse) of the self-gravitating gas in the grand canonical ensemble. Finally, we discuss why other
attempts to define a thermodynamic limit for the self-gravitating gas fail.
PACS: 05.20.-y; 04.40.-b; 64.60.-i; 95.30.sf
E-mail address: devega@lpthe.jussieu.fr (H.J. de Vega).

1 Laboratoire Associ au CNRS UMR 7589.
2 Laboratoire Associ au CNRS UMR 8112.
doi:10.1016/j.nuclphysb.2004.12.022
H.J. de Vega, N.G. Snchez / Nuclear Physics B 711 [FS] (2005) 604620
605
1. Introduction
The self-gravitating gas has been the subject of attention since many years [15]. In
Refs. [2,3] we recently investigated the self-gravitating thermal gas using Monte Carlo
simulations, mean field methods and low density expansions [2,3]. We have shown that
the system possess a well defined infinite volume limit in the grand canonical (GC), the
canonical (C) and microcanonical (MC) ensembles when N, V keeping N/V 1/3
fixed. A relevant variable here is the dimensionless ratio
G m2 N
,
(1.1)
V 1/3 T
which is kept fixed in the N, V limit. All physical quantities per particle turn out to
be functions of the single variable and are well defined and finite in the thermodynamic
limit N, V with fixed. has a simple physical meaning: it is a estimation of the
ratio between the potential and the kinetic energy per particle. The thermodynamic limit
considered here and in Refs. [2,3,5] is indeed dilute since the volume density of particles
tends to zero as N/V V 2/3 0.
In this paper we develop the cluster expansion and Mayers approach to the selfgravitating thermal gas and provide a rigorous demonstration of the existence of the
thermodynamic limit N, V with fixed.
The cluster expansion is a powerful tool allowing to express the partition function in
a power series of the density for short range interactions [6]. We apply and adapt this
method (the Mayer expansion) which is purely combinatorial to the self-gravitating gas.
We succeed in this way to obtain the coordinate partition function QN as a power series in
the thermodynamic limit. This is derived by generalizing the saddle point method used in
Ref. [6]. The expansion is obtained in terms of the coefficients cn which are pure numbers.
More explicitly,

1
1
N 1 1
log QN () = g(t ) log t 1 + O
,
(1.2)
N
where
g(x)
+
cj x j ,
(1.3)
j =1
and t is the solution of the equation

t g (t ) = 1,
i.e.,
+
j cj (t )j = .
j =1
N1
Moreover,
log QN () is the free energy of the self-gravitating gas minus the free energy of an ideal gas divided by N T . The series for g(t ) in Eqs. (1.2) and (1.3) is therefore
a high temperature or low density expansion (see Eq. (1.1)).
The coefficients cn can be expressed in the thermodynamic limit as a sum of 3n-uple
integrals associated to tree cluster diagrams. Loop cluster diagrams are subdominant for
N . The coefficients cn only depend on the geometry of the box. The first cn are
606
obtained by explicit evaluation of the cluster integrals. We have for the sphere

3 4 1/3
sphere
c1 = 1,
c2
=
,
5 3

51 4 2/3
4 373
sphere
sphere
.
c3
=
,
c4
=
70 3
3 315
Moreover, we use the connection with the mean field approach to obtain a nonlinear recursphere
rence relation for the coefficients cn
:

1
R
csR cns
s 2 (2n 2s + 1)
(2n + 1)(n 1)
n1
cnR =
for n 2,
s=1
where

sphere
cn
4
3
n1
3
cnR
(the label R stands for the spherical geometry). This allows to compute systematically all
cnR and to find their large n behaviour as

1
n5/2
n1
1
+
O
.
cnR = (0.309360 . . .)
(uGC )n
n
Here, uGC = 0.30034 . . . is the radius of convergence of the series Eq. (1.3). Therefore,
u = uGC is the nearest singularity of gR (u) in the u-plane.
We show below that the function g is related to the partition function ZGC (, z) in the
grand canonical ensemble by
log ZGC (, z) =
N
g(t ),
where

N
+

QN () mT 3/2
Vz .
ZGC (, z) =
N!
2
N =0
In addition, the fugacity (z = e/T ) turns out to be given by z = t e0 /T , where and 0

are the chemical potentials for the self-gravitating gas and for the ideal gas, respectively.
The point uGC = 0.30034 . . . where g(u) is singular corresponds to the critical point
(collapse) in the grand canonical ensemble. Recall that the collapse point depends on the
ensemble considered [2].
In conclusion, our investigation here on the cluster expansion and the Mayer expansion
for the self-gravitating gas shows that:
The self-gravitating gas admits a consistent thermodynamic limit N, V with
N/V 1/3 fixed. In this limit, extensive thermodynamic quantities like energy, free energy, entropy are proportional to N . That is, in this limit the energy, free energy, and
entropy per particle are well defined and finite.
607
The cluster expansion and the mean field approach provide the same results in the
thermodynamic limit N, V with N/V 1/3 fixed.
The partition function needs a small short-distance cutoff a in order to be well defined,
the thermodynamic limit has a finite limit for a 0. In other words, the contributions
to the partition function which diverge for a 0 are subdominant for N (see
Section 2). For large N and fixed cutoff a the potentially divergent contributions for
a 0 are suppressed by a factor at least 2 /N 2 compared with the dominant contribution for N . That is, the N, V = limit of the cutoff model (with N/V 1/3
fixed) has a finite limit for a 0. Therefore, subdominant corrections in /N can
always be neglected. Realistic models of the self-gravitating gas (interstellar medium,
galaxy distribution) require a small nonzero short distance cutoff since molecular and
atomic forces dominate over gravitational forces for short distances.
2. The cluster expansion for the self-gravitating gas

We investigate in this section the self-gravitating gas in thermal equilibrium at temperature T 1 . That is, we work in the canonical ensemble where the system of N particles
is in contact with a thermal bath at temperature T . We assume the gas being on a cubic box
of side L.
The partition function of the system can be written as
ZC (N, T ) =
1
N!

N
d 3 pl d 3 q l
l=1
(2)3
eHN ,
(2.1)
where
HN =
N

pl2
Gm2
2m
l=1

1l<j N
1
,
|
ql qj |A
(2.2)
G is Newtons gravitational constant.

At short distances, the particle interaction for the self-gravitating gas in physical situations is not gravitational. Its exact nature depends on the problem under consideration
(opacity limit, van der Waals forces for molecules, etc.). We shall just assume a repulsive
short distance potential, that is,
1

|ql
ql qj | A,
1
qj | for |
ql qj | =
vA |
(2.3)
=
1
|
ql qj |A
ql qj | A,
+ A for |
where A L is the short distance cut-off.
The integrals over the momenta pl (1 l N ) in Eq. (2.1) can be computed immediately.
It is convenient to introduce the dimensionless variables rl , 1 l N making explicit
the volume dependence as
ql = Lrl ,
rl = (xl , yl , zl ),
608
0 xl , yl , zl 1.
(2.4)
That is, in the new coordinates the gas is inside a cube of unit volume.
The partition function takes now the form
ZC (N, T ) =

3N
1 mT L2 2
QN (),
N!
2
(2.5)
where
1
QN ()
1
N
d 3 rl eu(r1 ,...,rN ) ,
(2.6)
0 l=1
is the dimensionless variable [2]

Gm2 N
,
LT
and u(r1 , . . . , rN ) is defined by
u(r1 , . . . , rN )
1
N
(2.7)

1l<j N
1
,
|rl rj |a
a A/L 1.
(2.8)
In this way all dependence on the volume V = L3 is buried in the variable .

The coordinate partition function QN () can be written as
1
QN () =
1
N
0 l=1
d 3 rl
e N|rl rj |a .
(2.9)
1l<j N
We are now interested to expand QN () in powers of /N . In order to do this is convenient to define:
flj e N|rl rj |a 1.

For small /N and fixed a > 0 we have
2
.
flj =
+O
N|rl rj |a
N
(2.10)
All the integrals over rl in Eq. (2.6) are finite provided we keep a > 0.
We can now multiply out the products of f s in the coordinate partition function QN (),

(1 + flj ) = 1 +
fij +
fij fkl + .
1l<j N
Thus, by introducing the fij functions the effects of interparticle forces are better exhibited.
A systematic treatment of such sum of products can be found in Ref. [6]. The outcome
is that a general term in the partition function QN () Eq. (2.9), can be factorized as the
product of several integrals over the coordinates rj . Each integral corresponds to a cluster
609
of particles and is called bj (, N ), where

1
bj (, N ) =
j!
1
1
j
d 3 rl S1,2,...,j ,
(2.11)
0 l=1
with [6]
Sj = 1,
S12 = f12 ,
S123 = f12 f23 + f12 f13 + f13 f23 + f12 f13 f23 ,
S1234 = f12 f23 f34 + nineteen permutations
+ f12 f13 f24 f34 + fourteen permutations
+ f12 f13 f14 f24 f34 + five permutations
+ f12 f13 f14 f23 f24 f34 .
(2.12)
The main result is that QN () can be expressed as the infinite sum, the so-called cluster
expansion as
QN () = N!
N

[bj (, N )]mj
,
mj !

j =1
m
, N
j =1 (j mj )=N
(2.13)
where
m
(m1 , . . . , mN ).
It must be stressed that Eqs. (2.11)(2.13) are purely combinatorial and they apply both
for the short range interactions considered in Ref. [6] as well as the long range Newton
potential.
Since flj = O(/N) for large N , and S1,2,...,j contains at least a product of j factors
fil , we see that
j 1
, for N j.
S1,2,...,j = O
N
Therefore, the cluster integrals Eq. (2.11) take the form

j 1
, for N j.
bj (, N ) =
cj 1 + O
N
N
(2.14)
Here the coefficients cj are positive numbers which only depend on the geometry of the
box. As shown in Fig. 1 the dominant terms for large N are cluster diagrams with a tree
structure. In the large N limit, cluster diagrams with a loop structure as the last term in
Eq. (2.12) are subdominant.
From Eqs. (2.10)(2.12) and (2.14) we find
c1 = 1,
610
1
c2 =
2
1 1
0 0
c3 =
1
2
d 3 r1 d 3 r2
,
|r1 r2 |
1 1 1
0 0 0
d 3 r1 d 3 r2 d 3 r3
.
|r1 r2 ||r2 r3 |
(2.15)
For a sphere of unit volume we find [2]

3 4 1/3
sphere
=
= 0.967195171 . . . ,
c2
5 3

51 4 2/3
sphere
c3
=
= 1.893206013 . . . .
70 3
(2.16)
For the cubic geometry chosen, it takes the value [2]

1
c2cube
=4
1
(1 x) dx
1

(1 y) dy
0
(1 z) dz
x 2 + y 2 + z2
= 0.94116 . . . .
Furthermore, we find for the coefficient c4 ,

1
c4 =
2
1 1 1 1
0 0 0 0
1
+
6
d 3 r1 d 3 r2 d 3 r3 d 3 r4
|r1 r2 ||r2 r3 ||r3 r4 |
1 1 1 1
0 0 0 0
d 3 r1 d 3 r2 d 3 r3 d 3 r4
.
|r1 r2 ||r2 r3 ||r2 r4 |
(2.17)
The combinatorial factors 1/2 and 1/6 take into account the symmetries of the respective
cluster diagrams.
Inserting here the expansion in spherical harmonics,
l m=+l

r<
1
1
Ylm (r )Ylm (r ),
=
4

l+1
|r r |
2l + 1 r>
l=0
m=l
max(r, r )
where r>
and r< min(r, r ), the angular integrals and then the radial integrals can be performed with the result
1 1 1 1
0 0 0 0
d 3 r1 d 3 r2 d 3 r3 d 3 r4
= (4)4
|r1 r2 ||r2 r3 ||r3 r4 |
b 4
b
2
l=1 rl drl
> r> r>
r1,2
2,3 3,4
b
b 4
4 62
3 35
4 188
.
3 105
and
1 1 1 1
0 0 0 0
d 3 r1 d 3 r2 d 3 r3 d 3 r4
= (4)4
|r1 r2 ||r2 r3 ||r2 r4 |
2
l=1 rl drl
>
>
>
r1,2 r2,3 r2,4
611
Therefore,
4 373
(2.18)
.
3 315
Notice that all these tree cluster diagrams Eqs. (2.15) and (2.17) have a finite limit for
zero cutoff a.
Divergent pieces for a 0 are subdominant as 2 /N 2 for large N . The leading divergent contribution for a 0 to QN () to the nth order in takes the form [2]
n
3
3n for n > 3,
d r1 d 3 r2 a0 n!N n2 a
n 1
N
(N
1)
(2.19)
3 log a for n = 3.
n!N n 2
|r1 r2 |na
N
sphere
c4
This gives for the physical quantities (see next section) contributions of the order
n
3
3n
a
for
n
>
3
and
log a for n = 3.
n!N n1
N2
These contributions are clearly negligible in the N limit with fixed short-distance
cutoff.
3. The Mayer expansion for the self-gravitating gas

It is convenient to consider the generating function [6]
+
X(, z)
bj (, N )zj .
(3.1)
j =1
It can be shown that [6]

+

QN () N
z = eX(,z) ,
N!
(3.2)
N =0
where z is an auxiliary variable whose physical meaning (the fugacity) will appear in Section 4.
Hence, we can compute the coefficients QN () from Eq. (3.2) by contour integration,

dz eX(,z)
QN ()
=
.
N!
2i zN +1
We choose as contour a circle of radius r,
z = rei ,
0 2.
Integrating over yields

QN ()
=
N!
+

d
exp X , rei N log r iN .
2
612
For large N , we can use Eq. (2.14) to express the bj (, N ) in X(, z) (see Eq. (3.1)) and
we find

+
z j N
z
N 1 N
,
X(, z) =
(3.3)
cj
= g
N
j =1
where
g(x)
+
cj x j .
(3.4)
j =1
Thus,
QN () N 1
=
N!
+
i
+
1 re
d
d N (r,)
exp N g
log r i =
e
.
2
N
2
(3.5)
We can now apply the steepest descent method to this integral for large N since the integrand has the structure eN (r,) , where

1 rei
log r i,
(r, ) = g
N

rei rei
(r, ) = i
g
i,
N
N

ei rei
1
(3.6)
(r, ) =
g
.
r
N
N
r
The saddle point, solution of
(r, ) = 0 =
(r, ) is thus found at
(r, )saddle = (N t, 0),

where t is N -independent and is a solution of the equation
tg (t) = 1.
(3.7)
That is, t is a function of defined by the constraint Eq. (3.7), or more explicitly,
+
j cj (t)j = .
j =1
Choosing r = N t as integration path in Eq. (3.5) and expanding the integrand around
= 0 yields
QN () N 1 N (N t,0)
= e
N!
+

1
d 1 N 2 ()
eN (N t,0)
2
1+O
e
,
=
2
N
2N()
where
1
(N t, 0) = g(t ) log[N t ],
613
()

2
(N t, 0) =
j 2 cj t j > 0,
2
(3.8)
j =1
where t is a function of defined by Eq. (3.7).

Using now Stirlings formula for the N ! factorial we find

1
1
1
log QN () = g(t ) log t 1 + O
.
N
(3.9)
Therefore, the free energy can be written as

F F0
1
= g(t ) + log t + 1,
NT
where F0 stands for the free energy of the ideal gas

eV mT 3/2
.
F0 = N T log
N 2
(3.10)
The pressure (at the surface) follows from the thermodynamic relation

F
p=
,
V T
we find
pV
2 g(t )
f ()
= +
NT
3
3
and
1
log QN () = 3
N

dx
1 f (x)
.
x
(3.11)
0
pV
NT
The dimensionless ratio

was called f () in Refs. [2,3]. We can express all physical
quantities in terms of the function f (). We find from Eqs. (3.9) and (3.11)
g(t )
3f () = 2 +
and
1
g(t ) + log t + 1 = 3

dx
1 f (x)
. (3.12)
x
That is,
1
log t = f () 1
3

dx
1 f (x)
.
x
(3.13)
We can solve Eq. (3.7) for t in powers of with the result

g(t ) = t + c2 (t )2 + c3 (t )3 + O 4 ,

t = 1 2c2 + 8(c2 )2 3c3 2 + O 3 ,
and from Eq. (3.12) we find for f () and g(t ),
f () = 1

2
c2
+ 2(c2 )2 c3 2 + O 3 ,
3
3
(3.14)
614

1 f (x)
= c2 + 2(c2 )2 c3 2 + O 3 ,
x
0

g(t ) = 1 c2 + 2 2(c2 )2 c3 2 + O 4 .
dx
(3.15)
It must be noticed that the function f () (and hence all physical quantities) have the same
expression whether we compute it from the mean field approach [2,3] or from the saddle
point Eqs. (3.5), (3.9) and (3.11).
4. The grand canonical ensemble for the self-gravitating gas

The definition of the variable z through Eq. (3.2) suggests that z is related to the fugacity. To be more precise, the grand partition function is defined in terms of the canonical
partition function as

N
+
+

QN () mT 3/2
ZC (N, T )zN =
Vz ,
ZGC (, z) =
(4.1)
N!
2
N =0
N =0
where we used Eq. (2.5).

3/2 V .
Therefore, z must be multiplied by the ideal gas factor ( mT
2 )
At the saddle point we have after this renormalization

mT 3/2
mT 3/2
V z = N t and X ,
V z = log ZGC (, z).
2
2
(4.2)
Now, we know from Ref. [2] that the chemical potential takes the form
0
= 3
T

dx

1 f (x)
3 1 f () ,
x
(4.3)
where 0 stands for the chemical potential of the ideal gas

V mT 3/2
0 = T log
.
N 2
(4.4)
From Eqs. (3.13) and (4.3) we have

0
= log t .
T
Therefore, from Eqs. (4.2)(4.5),
t = e0 /T z,
e/T = z,
(4.5)
(4.6)
that is, we can identify z with the fugacity.

We have found in Ref. [2] that f () obeys in the spherical case the first order differential
equation of Abels type,

R 3f R 1 f R + 3f R 3 + R f R = 0,
(4.7)
615
where the variable R is defined as

1/3
4
R

= (1.61199 . . .).
3
Using Eqs. (4.7) and (3.13) yields

R
log t = log f R R , i.e., t = e f R .
(4.8)
Eq. (4.1) provides the grand partition function. The above results showing that ZGC (, z)
is dominated by the canonical ensemble, together with Eq. (4.6) prove that the canonical
and grand canonical ensembles are equivalent in their common region of validity as stated
in Refs. [2,3].
In the thermodynamic limit, ZGC (, z) is given by Eqs. (3.3) and (4.2) as

z mT 3/2
N
N
log ZGC (, z) = g
(4.9)
V = g(t ).
N 2
Using here Eq. (3.12) gives

log ZGC (, z) = N 3f () 2 ,
which exactly coincides with the expression found in Ref. [2] for the grand partition function in the mean field approach.
5. Calculation of the cluster coefficients cn

We compute in this section the cluster coefficients cn for the sphere by using a nonlinear differential equation for the function g(u).
We first find from Eq. (3.7)
1/3
4
g(u).
ugR (u) = R with u t R and gR (u)
(5.1)
3
Then, combining Eqs. (4.7) and (5.1) yields the second order differential equation for
gR (u),

2
u gR (u) + ugR (u) 2ugR (u) + gR (u) 2u2 gR (u) ugR (u) + gR (u) = 0. (5.2)
Alternatively, if we choose R = ugR (u) as variable (see Eq. (5.1)), we find from Eq. (4.7)
the first order nonlinear differential equation
dgR

R

+ gR
+ R 2R + gR 2 = 0.
dR
That is, the invariance of Eq. (5.2) under the rescaling of the variable u allows to reduce
by one the order of the differential equation.
Eq. (5.1) has as regular solution around u = 0,
n1

4 3 R
sphere
R n
cn u with cn
=
cn ,
gR (u) =
(5.3)
3
n=1
616
and we can choose c1R = 1.

By inserting Eq. (5.3) into Eq. (5.2) the following nonlinear recurrence relation for the
cnR coefficients is obtained:

1
R
csR cns
s 2 (2n 2s + 1)
(2n + 1)(n 1)
n1
cnR =
for n 2.
(5.4)
s=1
We find
c1R = 1,
3
c2R = ,
5
c3R =
51
,
70
c4R =
373
,
315
14911
2047
,
c6R =
, ...,
6600
429
in agreement with Eqs. (2.15), (2.16) and (2.18).
By evaluating numerically the cnR from the recurrence relation Eq. (5.4), we find for
large n:

1
n5/2
n1
1
+
O
.
cnR = (0.309360 . . .)
(5.5)
(uGC )n
n
c5R =
Here, uGC = 0.30034 . . . is the radius of convergence of the series Eq. (5.3). Therefore,
u = uGC is the nearest singularity of gR (u) in the u-plane.
As we see from Eq. (4.9), gR (u) gives the grand canonical partition function as a function of u = t R . Since u = t R is proportional to the fugacity (see Eq. (4.6)), uGC must
be related to the critical point of the selfgravitating gas in the grand canonical ensemble.
Indeed, using the critical value for R in the grand canonical ensemble [2],
R
= 0.79735 . . .
GC
R
=
and f GC
2
,
R
3GC
from Eq. (4.8) we obtain

R 2 R
R
R
R GC
= e GC = 0.30034 . . . ,
uGC = GC
tR = GC
e
f GC
GC
3
in perfect agreement with the value for uGC in Eq. (5.5). The large order behaviour of the
expansion coefficients Eq. (5.5) corresponds to a (uGC u)3/2 behaviour of the function
gR (u). More precisely, for u uGC from Eq. (5.2) we find

u
uu
R
R
gR (u) =GC 2 1 GC
+ GC
1
uGC

u 3/2
2
R
+
2 GC 1
+ O (u uGC )2 .
3
uGC
Expanding this asymptotic behaviour in powers of u yields

R

u
1 2 GC k 32 k
uuGC
R
R
1 +
u .
gR (u) = 2 1 GC + GC
uGC
2
k!ukGC
k=0
617
This implies for the coefficients cnR the following large order behaviour:

R
5/2
1
n1 1 2 GC n
cnR =
1
+
O
.
2
(uGC )n
n
which exactly coincides with Eq. (5.5) since

R
1 2 GC
= 0.309360 . . . .
2
In summary, the investigation here on the cluster expansion for the selfgravitating gas
shows:
The selfgravitating gas admits a consistent thermodynamic limit N, V with
N/V 1/3 fixed. In this limit, extensive thermodynamic quantities like energy, free energy, entropy are proportional to N . That is, in this limit the energy, free energy, and
entropy per particle are well defined and finite.
The cluster expansion and the mean field approach provide the same results in the
thermodynamic limit N, V with N/V 1/3 fixed.
The partition function needs a small short-distance cutoff a in order to be well defined,
the thermodynamic limit has a finite limit for a 0. In other words, the contributions
to the partition function that diverge for a 0 are subdominant for N (see
Eq. (2.19)). That is, the N, V = limit of the cutoff model (with N/V 1/3 fixed) has
a finite limit for a 0.
6. The stability of the thermodynamic limit N, V = L3 with N/L fixed
(fixed )
As shown in Ref. [2], two phases exist for the selfgravitating gas: a gaseous phase if
< 0 and a collapsed phase if > 0 where 0 = 1.51024 . . . for spherical geometry and
0
1.515 for cubic geometry. At = 0 the isotherm compressibility diverges [2,3] as
0
0.24911 . . .
.
0
For > 0 the selfgravitating gas collapses into a extremely dense phase with large and
negative pressure. For < 0 the selfgravitating gas is stable. The mean field applies in this
gaseous phase and coincides with the expansion in powers of discussed in the previous
section. Such expansion converges within the gaseous phase. The variable is related to
the Jeans length of the system as
2
L
, L = V 1/3 ,
=3
dJ
where
dJ =
1
3T
,
m Gm
N
.
V
618
The gas collapses in the canonical ensemble for > 0 which corresponds to L dJ . This
corresponds to the Jeans instability. In addition, the speed of sound at the origin becomes
imaginary for > 0 implying an exponential growth in time for small disturbances in
the gas [3]. This is the physical mechanism leading to collapse. Also the specific heat at
constant volume cv is positive and finite for < C = 1.561764 . . . . cv diverges at = C ,
a value of slightly larger than 0 [2]. Therefore, the signal for collapse is in the singularity
of the compressibility and not in the specific heat singularity.
2
The relevance of the ratio VGm
1/3 T has been noticed on dimensional grounds [9]. However,
the dimensionality argument alone cannot single out the crucial factor N in the variable .
Notice that contains the ratio N/V 1/3 and not N/V . Therefore, in the thermodynamic
limit
1
N
2 0.
V
N
As N, V , is kept fixed in the same way as the temperature T . The energy, the
product P V , the free energy and entropy are expressed as a factor N times functions of .
The chemical potential, the specific heats, the compressibilities are just functions of .
For finite and large N the gas actually has a finite lifetime eN () , where () is a
function of of order one with () > 0 for < 0 in the gaseous phase (see, for example,
Ref. [10]). Therefore, for realistic values of N 1, this lifetime is a huge number much
larger than the age of the universe. Hence, it is natural to call these states stable. Monte
Carlo simulations confirm such stability features [2].
In a recent e-print [7] it was stated that the thermodynamic functions for the selfgravitating gas diverge in the thermodynamic limit N, V with N/V 1/3 fixed.
We show here below that the statements made in Ref. [7] have crucial failures which
invalidate the conclusions given in Ref. [7].
Such statements in Ref. [7] are based in the inequality (Eq. (30) in Ref. [7])

N

V0N
1 3N/2 3
exp
d ri exp
ij
ij V0 ,
ZC
N!
N!
V N3
and =
V0N
i=1
i<j
i<j
where
ij V0 =
1
V0N

N
V0N
k=1
d 3 rk ij
and ij
G
,
|ri rj |
and the last inequality follows from the property of the exponential function: exp(y)
exp(y).
To write such inequality the author of Ref. [7] considers N selfgravitating particles in a
volume V = R 3 with N R, that is, in the conditions of the above thermodynamic limit
(i.e., N/V 1/3 fixed).
Then, a portion of such volume of linear size R0 < R is considered: at this precise point
the author assumes that N R03 , that is, he assumes a extremely dense distribution of
particles within the volume of size R0 .
619
Fig. 1. Diagrams contributing to the cluster expansion. Diagram (I) gives c2 . Diagram (II) gives c3 . Diagrams
(III), (IV) and (V) contribute to c4 but (III) is subdominant for large N .
But for large N such distribution necessarily collapses since the gravitational gas avoids
collapse only when N R0 . In terms of the parameter , the assumption N R03 implies
0 N/R0 (R0 )2 1,
which is deep in the collapsed phase [2]. That is, the assumption N R03 necessarily
implies that the gas collapses. The subsequent statements made in Ref. [7] are direct consequences of this assumption and are not valid.
It is clearly true that the partition function sums over all configurations including collapsed situations. However, these collapsed configurations have a negligible weight for
< 0 N/R0 . The argument of Ref. [7] applies only for > 0 and only for such values
of collapsed states dominate the partition function. It must be noticed that in Nature, if
collapsed configurations would always dominate selfgravitating systems, then, stars, galaxies and the interstellar medium would have collapsed since longtime. Actually, the lifetime
of all such metastable objects is extremely long (as discussed above) and one can consider
them as stable.
As stressed in Refs. [2,3] there are two regimes for the selfgravitating gas in the canonical ensemble: < 0 and > 0 , with 0 = 1.51024 . . . for spherical geometry and
C
1.515 for cubic geometry. For > 0 the selfgravitating gas do collapse into a extremely dense phase. It is to this collapsed phase and only to it that the Eqs. (30)(33) of
Ref. [7] apply.
The same comments applies to the convergence of the series expansions of the thermodynamic quantities discussed in Ref. [7] (Eq. (34) of Ref. [7] above). The series in powers
of only converge for < C . Actually, the calculation of the radius of convergence of
such series in Refs. [2,3] provided us an independent check of the numerical value of C .
620
In the same token, the thermodynamic limit proposed in Ref. [8] leads to a cataclysmic
collapse for the selfgravitating gas. It is proposed in Ref. [8] to take N with V
1/N 0.
It is obvious that in such limit the gas collapses since the volume vanishes.
More precisely, that means N 4/3 + deeply in the extremely dense phase.
Furthermore, it is wrongly stated in Section 4 of Ref. [8] that the entropy is not proportional
to N as in Refs. [2,3].
References
[1] R. Emden, Gaskugeln, Teubner, Leipzig und Berlin, 1907;
S. Chandrasekhar, An Introduction to the Study of Stellar Structure, Chicago Univ. Press, Chicago, 1939;
W.B. Bonnor, Mon. Not. R. Soc. 116 (1956) 351;
R. Ebert, Z. Astrophys. 37 (1955) 217;
V.A. Antonov, Vestnik Leningrad. Univ. 7 (1962) 135;
D. Lynden-Bell, R. Wood, Mon. Not. R. Astron. Soc. 138 (1968) 495;
G. Horwitz, J. Katz, Astrophys. J. 211 (1977) 226;
G. Horwitz, J. Katz, Astrophys. J. 222 (1978) 941;
T. Padmanabhan, Phys. Rep. 188 (1990) 285;
W.C. Saslaw, Gravitational Physics of Stellar and Galactic Systems, Cambridge Univ. Press, Cambridge,
1987
[2] H.J. de Vega, N. Snchez, Nucl. Phys. B 625 (2002) 409.
[3] H.J. de Vega, N. Snchez, Nucl. Phys. B 625 (2002) 460.
[4] E.V. Votyakov, A. De Martino, D.H.E. Gross, Eur. Phys. J. B 29 (2002) 593;
E.V. Votyakov, A. De Martino, D.H.E. Gross, Nucl. Phys. B 654 (2003) 427;
P.H. Chavanis, I. Ispolatov, Phys. Rev. E 66 (2002) 036109;
B. Leong, W. Saslaw, astro-ph/0308415.
[5] H.J. de Vega, J.A. Siebert, Phys. Rev. E 66 (2002) 016112.
[6] T.L. Hill, Statistical Mechanics, McGrawHill, New York, 1956.
[7] V. Laliena, astro-ph/0303301.
[8] L. Velazquez, F. Guzmn, cond-mat/0205085;
L. Velazquez, F. Guzmn, cond-mat/0303444.
[9] See the last reference in Ref. [1].
[10] P.H. Chavanis, astro-ph/0404251;
I. Ispolatov, M. Karttunen, Phys. Rev. E 68 (2003) 036117.
AUTHOR INDEX B711
Aulakh, C.S.
B711 (2005) 275
Buchbinder, E.I.
Buchbinder, I.L.
B711 (2005) 314

B711 (2005) 367
Choi, S.Y.
B711 (2005) 83
de Vega, H.J.
Dobashi, S.
Dobashi, S.
B711 (2005) 604

B711 (2005) 3
B711 (2005) 54
Girdhar, A.
Giuliano, D.
B711 (2005) 275

B711 (2005) 480
Janssen, B.
B711 (2005) 392
Kawai, H.
Kimura, T.
Kniehl, B.A.
Krs, B.
Kramer, G.
Krykhtin, V.A.
Kuroki, T.
B711 (2005) 253

B711 (2005) 163
B711 (2005) 345
B711 (2005) 112
B711 (2005) 345
B711 (2005) 367
B711 (2005) 253
Lozano, Y.
B711 (2005) 392
Maniatis, M.
Martins, M.J.
Matsuo, Y.
B711 (2005) 345

B711 (2005) 565
B711 (2005) 253
0550-3213/2005 Published by Elsevier B.V.

doi:10.1016/S0550-3213(05)00166-5
Melo, C.S.
Miller, D.J.
B711 (2005) 565

B711 (2005) 83
Nath, P.
Nirschl, M.
Nitta, M.
B711 (2005) 112

B711 (2005) 409
B711 (2005) 133
Osborn, H.
B711 (2005) 409
Pashnev, A.
Pinnow, H.A.
Polychronakos, A.P.
B711 (2005) 367

B711 (2005) 530
B711 (2005) 505
Ribeiro, G.A.P.
Riccioni, F.
Rodrguez-Gmez, D.
B711 (2005) 565

B711 (2005) 231
B711 (2005) 392
Snchez, N.G.
Sodano, P.
Spradlin, M.
B711 (2005) 604

B711 (2005) 480
B711 (2005) 199
Volovich, A.
B711 (2005) 199
Wiese, K.J.
B711 (2005) 530
Yoneya, T.
Yoneya, T.
B711 (2005) 3
B711 (2005) 54
Zerwas, P.M.
B711 (2005) 83

Nucl - Phys.B v.711

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.711

Uploaded by

Copyright:

Available Formats

Nuclear Physics B 711 (2005) 353

Resolving the holography in the plane-wave limit

E-mail addresses: doba@hep1.c.u-tokyo.ac.jp (S. Dobashi), tam@hep1.c.u-tokyo.ac.jp (T. Yoneya).

S. Dobashi, T. Yoneya / Nuclear Physics B 711 (2005) 353

which connects the boundary values limz0 zi 4 i (z, x) = i (

S. Dobashi, T. Yoneya / Nuclear Physics B 711 (2005) 353

S. Dobashi, T. Yoneya / Nuclear Physics B 711 (2005) 353

2. The direct large J -limit of GKP-W relation

where the bulk-to-boundary propagator

S. Dobashi, T. Yoneya / Nuclear Physics B 711 (2005) 353

2.1. Two-point functions

The general solution is

which are defined by

with the orthogonality constraint

where we have used the solution for the constraint (2.13),

S. Dobashi, T. Yoneya / Nuclear Physics B 711 (2005) 353

We can compare this result with that of exact integration:

S. Dobashi, T. Yoneya / Nuclear Physics B 711 (2005) 353

S. Dobashi, T. Yoneya / Nuclear Physics B 711 (2005) 353

S. Dobashi, T. Yoneya / Nuclear Physics B 711 (2005) 353

where 1 = (2 + 3 1 )/2, etc. The precise normalization will be discussed in the

S. Dobashi, T. Yoneya / Nuclear Physics B 711 (2005) 353

By taking the large Ji limit using the Stirling formula

S. Dobashi, T. Yoneya / Nuclear Physics B 711 (2005) 353

3. Effective action along the tunneling trajectory

2 N )k/2 k is the normalization factor such that the 2-point functions

O I = CiI1 i2 ...i Tr Z J i1 i2 ik + permutations ,

S. Dobashi, T. Yoneya / Nuclear Physics B 711 (2005) 353

S. Dobashi, T. Yoneya / Nuclear Physics B 711 (2005) 353

Apart from the normalization factor which is independent of 1 = (2 + 3 1 )/2 =

S. Dobashi, T. Yoneya / Nuclear Physics B 711 (2005) 353

3.2. Effective (0 + 1)-dimensional action

and then making the shift of the time coordinate

To avoid notational confusions, we denote the rescaled fluctuations (x , z ) by a four

The effective metric

S. Dobashi, T. Yoneya / Nuclear Physics B 711 (2005) 353

The hermiticity condition for our Euclidean field theory is [4]

is seen to be k which correctly reproduces the O(1) part of the energy = J + k.

S. Dobashi, T. Yoneya / Nuclear Physics B 711 (2005) 353

are the normalized eigenfunctions. The kinetic term is then extended to

and similarly for the interaction term

S. Dobashi, T. Yoneya / Nuclear Physics B 711 (2005) 353

3.3. Vector excitations

C iI1 i2 ...i KjL1 j2 ...j j1 j2 j Tr Z J i1 i2 ik + permutations

C iI1 i2 ...i KjL1 j2 ...j Tr Z J  Dj1 ZDj2 Z Dj Z i1 i2 ik + permutations ,

S. Dobashi, T. Yoneya / Nuclear Physics B 711 (2005) 353

S. Dobashi, T. Yoneya / Nuclear Physics B 711 (2005) 353

from the bulkboundary propagators in the integrand (2.7),

respectively, in the large Ji -limit, where

(2J )1 1 !|

The Kronecker L1 L2 and the prefactor arises by the Gaussian integral

In terms of derivatives with respect to (

S. Dobashi, T. Yoneya / Nuclear Physics B 711 (2005) 353

S. Dobashi, T. Yoneya / Nuclear Physics B 711 (2005) 353

1 !2 !3 ! (1 + 1 1)! L1 L2 L3

|2(J1 +k1 +1 ) |2|2(1 +1 )

1 !2 !3 ! (1 + 1 1)! I1 I2 I3 L1 L2 L3

in acting the derivation

S. Dobashi, T. Yoneya / Nuclear Physics B 711 (2005) 353

k1 !k2 !k3 ! I I I 1 !2 !3 ! L L L

S. Dobashi, T. Yoneya / Nuclear Physics B 711 (2005) 353

(k2 +2 )/2 (k3 +3 )/2

k1 !k2 !k3 1 !2 !3 ! I I I L L L

C iI1 i2 ...i KjL1 j2 ...j j1 j2 j Tr Z J i1 i2 ik + permutations

C iI1 i2 ...i KjL1 j2 ...j Tr Z J Dj1 ZDj2 Z Dj Z i1 i2 ik + permutations ,

(2J )1 1 !|

1 !2 !3 ! (1 + 1 1)! L1 L2 L3

|2(J1 +k1 +1 ) |2|2(1 +1 )

1 !2 !3 ! (1 + 1 1)! I1 I2 I3 L1 L2 L3

k1 !k2 !k3 ! I I I 1 !2 !3 ! L L L

(k2 +2 )/2 (k3 +3 )/2

k1 !k2 !k3 1 !2 !3 ! I I I L L L

Here, we have used the relation 1 2 3 = (1 2 3 ), which is valid

For the third case (6.3), we obtain for n = 0 and m = 0,

31m 31m + vac 31n 31n : Cvac

31m 41m + vac 31n 41n : Cvac