Nucl - Phys.B v.698

Nuclear Physics B 698 (2004) 352
Supersymmetric gauge theories with flavors

and matrix models
Changhyun Ahn a,b , Bo Feng a , Yutaka Ookouchi c ,
Masaki Shigemori d
a School of Natural Sciences, Institute for Advanced Study, Olden Lane, Princeton, NJ 08540, USA
b Department of Physics, Kyungpook National University, Taegu 702-701, South Korea
c Department of Physics, Tokyo Institute of Technology, Tokyo 152-8511, Japan
d Department of Physics and Astronomy, UCLA, Los Angeles, CA 90095-1547, USA
Received 24 May 2004; accepted 15 July 2004

Available online 7 August 2004
Abstract
We present two results concerning the relation between poles and cuts by using the example of
N = 1 U (Nc ) gauge theories with matter fields in the adjoint, fundamental and antifundamental
representations. The first result is the on-shell possibility of poles, which are associated with flavors
and on the second sheet of the Riemann surface, passing through the branch cut and getting to the
first sheet. The second result is the generalization of hep-th/0311181 (Intriligator, Kraus, Ryzhov,
Shigemori, and Vafa) to include flavors. We clarify when there are closed cuts and how to reproduce
the results of the strong coupling analysis by matrix model, by setting the glueball field to zero
from the beginning. We also make remarks on the possible stringy explanations of the results and on
generalization to SO(Nc ) and USp(2Nc ) gauge groups.
2004 Elsevier B.V. All rights reserved.
PACS: 11.15.Tk; 11.30.Pb; 11.25.Db
E-mail addresses: ahn@ias.edu (C. Ahn), fengb@ias.edu (B. Feng), ookouchi@th.phys.titech.ac.jp

(Y. Ookouchi), shige@physics.ucla.edu (M. Shigemori).
0550-3213/$ see front matter 2004 Elsevier B.V. All rights reserved.
doi:10.1016/j.nuclphysb.2004.07.015
C. Ahn et al. / Nuclear Physics B 698 (2004) 352
1. Introduction
String theory can be a powerful tool to understand four-dimensional supersymmetric
gauge theory which exhibits rich dynamics and allows an exact analysis. In [1], using the
generalized Konishi anomaly and matrix model [2], N = 1 supersymmetric U (Nc ) gauge
theory with matter fields in the adjoint, fundamental and antifundamental representations
was studied. The resolvents in the quantum theory live on the two-sheeted Riemann surface defined by the matrix model curve. Their quantum behavior is characterized by the
structure around the branch cuts and poles, which are related to the RR flux contributions
in the CalabiYau geometry and flavor fields, respectively. A pole associated with flavor
on the first sheet is related to the Higgs vacua (corresponding to classical nonzero vacuum
expectation value of the fundamental) while a pole on the second sheet is related to the
pseudo-confining vacua where the classically vanishing vacuum expectation value of the
fundamental gets nonzero values due to quantum correction.
It is known [1] that Higgs vacua and pseudo-confining vacua, which are distinct in the
classical theory, are smoothly transformed into each other in the quantum theory. This
transition is realized on the Riemann surface by moving poles located on the second sheet
to pass the branch cuts and enter the first sheet. This process was analyzed in [1] at the
off-shell level by fixing the value of glueball fields during the whole process. However,
in an on-shell process, the position of poles and the width and position of branch cuts are
correlated (when the flavor poles are moved, the glueball field is also changed). It was
conjectured in [1] that for a given branch cut, there is an upper bound for the number of
poles (the number of flavors) which can pass through the cut from the second sheet to the
first sheet.
Our first aim of this paper is to confirm this conjecture and give the corresponding
upper bound for various gauge groups (in particular, we will concentrate on the U (Nc )
gauge group). The main result is that if Nf Nc , the poles will not be able to pass through
the cut to the first sheet where Nc is the effective fluxes associated with the cut (and can be
generalized to other gauge groups).
Another important development was made in [3], which was inspired by [4]. In [3],
which we will refer to as IKRSV, it was shown that, to correctly compute the prediction of
string theory (matrix model), it is crucial to determine whether the glueball is really a good
variable or not. A prescription was given, regarding when a glueball field corresponding to
a given branch cut should be set to zero before extremizing the off-shell glueball superpotential. The discussion of IKRSV was restricted to N = 1 gauge theories with an adjoint
and no flavors, so the generalization to the case with fundamental flavors is obviously the
next task.
Our second aim of this paper is to carry out this task. The main result is the following.
Assuming Nf poles around a cut associated with gauge group U (Nc,i ), when Nf Nc,i
there are situations in which we should set Si = 0 in matrix model computations. More
concretely, situations with Si = 0 belong to either of the following two branches: the baryonic branch for Nc,i Nf < 2Nc,i , or the r = Nc,i nonbaryonic branch for Nf 2Nc,i .
Moreover, when Si = 0, the gauge group is completely broken and there should exist some
extra, charged massless field which is not incorporated in matrix model.
In Section 2, as background, we review basic materials for N = 1 supersymmetric

U (Nc ) gauge theory with an adjoint chiral superfield, and Nf flavors of quarks and antiquarks. The chiral operators and the exact effective glueball superpotential are given. We
study the vacuum structure of the gauge theory at classical and quantum levels. We review
also the main results of IKRSV. In addition to all these reviews, we present our main
motivations of this paper.
In Section 3, we apply the formula for the off-shell superpotential obtained in [1] to
the case with quadratic tree level superpotential, and solve the equation of motion derived
from it. We consider what happens if one moves Nf poles associated with flavors on the
second sheet through the cut onto the first sheet, on-shell. Also, in Section 3.4, we briefly
touch the matter of generalizing IKRSV in the one cut model.
In Section 4, we consider cubic tree level superpotential. On the gauge theory side, the
factorization of the SeibergWitten curve provides an exact superpotential. We reproduce
this superpotential by matrix model, by extremizing the effective glueball superpotential
with respect to glueball fields after setting the glueball field to zero when necessary. We
present explicit results for U (3) theory with all possible breaking patterns and different
number of flavors (Nf = 1, 2, 3, 4, and 5).
In Section 5, after giving concluding remarks, we repeat the procedure we did in previous sections for SO(Nc )/USp(2Nc ) theories, briefly.
In the appendix, we present some proofs and detailed calculations which are necessary
for the analysis in Section 4.
Since string theory results in the dual CalabiYau geometry are equivalent to the matrix
model results, we refer to them synonymously through the paper. There exist many related
works to the present paper. For a list of references, we refer the reader to [5].
2. Background
In this section, we will summarize the relevant background needed for the study of
N = 1 supersymmetric gauge theory with matter fields.
2.1. The general picture of matrix model with flavors
The generalized Konishi anomaly interpretation to the matrix model approach for N =
1 supersymmetric gauge theory with flavors was given in [1,6]. Here we make only a brief
summary on some points we will need.
Let us consider N = 1 supersymmetric U (Nc ) gauge theory, coupled to an adjoint chiral
superfield , Nf fundamentals Qf , and Nf antifundamentals Q f . The tree level superpotential is taken to be
Wtree = Tr W () +

f,f
Q f mf f ()Qf ,
(2.1)

f
where the function W (z) and the matrix mf (z) are polynomials
W (z) =
n

gk zk+1
k=0
k+1
mf f (z) =
l+1
(mk )f f zk1 .
k=1
Classically we can have the pseudo-confining vacua where the vacuum expectation values of Q, Q are zero, or the Higgs vacua where the vacuum expectation values of Q,
Q are nonzero so that the total rank of the remaining gauge groups is reduced. These two
vacua, which seem to have a big difference classically, are not fundamentally distinguishable from each other in the quantum theory and in fact can be continuously transformed
into each other, as we will review shortly, in the presence of flavors [1].
Supersymmetric vacua of gauge theory are characterized by the vacuum expectation
values of chiral operators [7]. They are nicely packaged into the following functions called
resolvents [1,6]1

1
,
T (z) = Tr
(2.2)
z

1
W W
R(z) =
(2.3)
Tr
,
32 2
z

1
Qf ,
M(z)f f = Q f
(2.4)
z
where W is (the lowest component of) the field strength superfield. Classically, R(z)
vanishes while T (z), M(z) have simple poles on the complex z-plane at infinity and at
the eigenvalues of . Each eigenvalue of is equal to one of zeros of W (z) or B(z),
where
W (z) = gn
n

(z ai ),
B(z) det m(z) = BL
i=1
L
(z zI ).
(2.5)
I =1
In the pseudo-confining vacuum, every eigenvalue of is equal to ai for some i. On the

other hand, in the Higgs vacuum, some eigenvalues of are equal to zI for some I .
In the quantum theory, the resolvents (2.2)(2.4) are determined by the generalized
Konishi anomaly equations [1,6,8]:

W (z)T (z)

+ Tr m (z)M(z) = 2R(z)T (z),

W (z)R(z) = R(z)2 ,

f
M(z)m(z) f = R(z) f f ,

f
m(z)M(z) f = R(z) f f ,
1 We set w (z) 1 Tr W to zero because in supersymmetric vacua w (z) = 0.
z
4
(2.6)
where the notation [ ] means to drop the nonnegative powers in a Laurent expansion in z.
From the second equation of (2.6), one obtains [9]

1
R(z) = W (z) W (z)2 + f (z) ,
2
where f (z) is a polynomial of degree (n 1) in z. This implies that in the quantum theory
the zeros z = ai (i = 1, 2, . . . , n) of W (z) are blown up into cuts Ai along intervals2
[ai , ai+ ] by the quantum effect represented by f (z), and the resolvents (2.2)(2.4) are
defined on a double cover of the complex z-plane branched at the roots ai of W (z)2 +
f (z). This double cover of the z-plane can be thought of as a Riemann surface described
by the matrix model curve
:
2
= W (z)2 + f (z).
ym
(2.7)
This curve is closely related to the factorization form of N = 2 curve in the strong coupling
analysis.
Every point z on the z-plane is lifted to two points on the Riemann surface which
we denote by q and q,
respectively. For example, zI is lifted to qI on the first sheet and qI
on the second sheet. We write the projection from to the z-plane as zI = z(qI ) = z(qI ),
following the notation of [1].
The classical singularities of the resolvents T (z), M(z) are modified in the quantum
theory to the singularities on , as follows. For T (z), the classical poles at zI are lifted to
poles at qI or qI , depending on which vacuum the theory is
in, while the classical poles at
1
ai with residue Nc,i are replaced by cuts with periods 2i
Ai T (z) dz = Nc,i . For M(z),
the classical poles at zI are also lifted to poles at qI or qI . More specifically, by solving
the last two equations of (2.6), one can show [1]

L

(1 rI )R(qI ) 1
dx
1
M(z) = R(z)
m(z)
(z zI ) 2i
m(x)
I =1
L

I =1
rI R(qI ) 1
(z zI ) 2i

qI
qI
dx
,
m(x)
(2.8)
where (qI , qI ) are the lift of zI to the first sheet and to the second sheet of , and rI = 0
for poles on the second sheet and rI = 1 for poles on the first sheet. Furthermore, for T (z),
by solving the first equation of (2.6)
B (z) (1 2rI )y(qI ) c(z)
+
,
2B(z)
2y(z)(z zI )
y(z)
(2.9)

L
W (z) W ()
1 W (z) W (zI )
c(z) = Tr
.
z
2
z zI
(2.10)
T (z) =
I =1
where
I =1
2 a are generally complex and in such cases we take A to be a straight line connecting a and a + . Note
i
i
i
i
that there is no physical meaning to the choice of the cut; it can be any path connecting ai and ai+ .
Practically it is hard to use (2.10) to obtain c(z) and we use the following condition instead:

1
(2.11)
T (z) dz = Nc,i .
2i
Ai
Finally, the exact, effective glueball superpotential is given by [1]

1
Weff =
Nc,i
2
n
i=1
1
rI
2
L
I =1

B ir
1
y(z) dz
(1 rI )
2
L
I =1
0
y(z) dz
qI
0
L
1
1
y(z) dz + (2Nc L)W (0 ) +
W (zI )
2
2
I =1
qI
i(2Nc L)S + 2i0 S + 2i
n1
bi Si ,
(2.12)
i=1
where

BL 2Nc Nf
2i0 = log
.
c L
2N
0
(2.13)

Here, 0 is the cut-off of the contour integrals, is the dynamical scale, S ni=1 Si ,
and bi Z. B ir is the regularized contour from 0 to 0 through the ith cut and 0 and
0 are the points on the first sheet and on the second sheet, respectively. The glueball field
is defined as

1
R(z) dz.
Si =
2i
Ai
In the above general solutions (2.8), (2.9), we have rI = 1 or rI = 0, depending on

whether the pole is on the first sheet or on the second sheet of , respectively. The relation between these choices of rI and the phase of the system is as follows. Let us start
with all rI = 0, i.e., all the poles are on the second sheet. This choice corresponds
to
n
)
U
(N
)
the pseudo-confining
vacua
where
the
gauge
group
is
broken
as
U
(N
c
c,i
i=1

with ni=1 Nc,i = Nc . Now let us move a single pole through, for example, the nth cut

to the first sheet. This will break the gauge group as ni=1 U (Nc,i ) n1
i=1
U (Nc,i )
U (Nc,n 1). Note that the rank of the last factor is now (Nc,n 1) so that ni=1 Nc,i =
Nc 1 < Nc . Namely, the gauge group is higgsed down. In this way, by passing poles
through cuts, one can go continuously from the pseudo-confining phase to the Higgs phase,
as advocated before.
However, if we consider this process of passing poles through a cut to the first sheet
on-shell, then there should be an obstacle at a certain point. For example, if initially we
have Nc,n = 1, after passing a pole we would end up with an U (0). This sudden jump of
the number of U (1)s in the low energy gauge theory is not a smooth physical process,
because the number of massless particles (photons) changes discontinuously. So we expect
some modifications to the above picture. In [1], it was suggested that in an on-shell process,
the nth cut will close up in such a situation so that the pole cannot pass through. It is one
of our motivations to show that this is indeed true. More precisely, the cut does not close
up completely and the pole can go through a little bit further to the first sheet and then will
be bounced back to the second sheet.
2.2. The vacuum structure
In the last subsection we saw that different distributions of poles over the first and the
second sheets correspond to different phases of the theory. In this subsection we will try to
understand this vacuum structure of the gauge theory at both classical and quantum levels
for a specific model (for more details, see [1015]). For simplicity we will focus on U (Nc )
theory with Nf flavors and the following tree level superpotential3
Nf

1
Q I ( + mf )QI .
Wtree = mA Tr 2
2
(2.14)
I =1
This corresponds to taking polynomials in (2.1) as

mA 2
z ,
W (z) =
mI I (z) = (z + mf )II .
2
All Nf flavors have the same mass mf , and the mass function defined in (2.5) is given by
B(z) = (1)Nf (z + mf )Nf .
Therefore, poles associated with flavors are located at
zI = mf zf ,
I = 1, 2, . . . , Nf .
(2.15)
In the quantum theory, some of these poles are lifted to qf on the first sheet and others are
lifted to qf on the second sheet.
The D- and F -flatness for the superpotential (2.14) is given by

0 = QQ Q Q,
0 = mA QQ,
0 = , ,
0 = ( + mf )Q = Q(
+ mf ).
Solutions are a little different for mf = 0 and mf = 0, because mf = 0 is the root of
W (z) = z. The case of W (mf ) = 0 was discussed in [10,13] which we will refer to as
the classically massless case.
In the mf = 0 case, the solution is given by

mf IKK
0
=
,
0
0(Nc K)(Nc K)

t

AKK
B K(Nf K)
0
AKK
t
,
Q=
Q=
0
0(Nc K)(Nf K)
0
0(Nc K)(Nf K)
(2.16)
3 We used the convention of [16] for the normalization of the second term. Different choices are related to
and Q.
each other by redefinition of Q
10
with
mf mA IKK = A t A,
AA = A A + B t B.
The gauge group is higgsed down to U (Nc K) where

Kmf =0 min(Nc , Nf ).
To understand the range of Kmf =0 , first note that the breaks the gauge group as
U (Nc ) U (K) U (Nc K). Now the U (K) factor has effectively Nf massless fla
vors and because QQ
= 0, U (K) is further higgsed down to U (0).
For mf = 0, we have = 0 and Q, Q are still of the above form (2.16) with one special
requirement: A = 0. Because of this we have

Nf
Kmf =0 min Nc ,
,
2
where [ ] means the integer part. The integer Kmf =0 precisely corresponds to the rth
branch discussed in [10,13]. The mf = 0 case is different from the mf = 0 case as follows.
First, does not break the U (Nc ) gauge group, i.e., U (Nc ) U (Nc ). Secondly, the rth
branch is the intersection of the Coulomb branch in which QQ

= 0 and the Higgs branch
must be
in which QQ = 0, whereas for mf = 0 the vacuum expectation value QQ
nonzero and the gauge group must be higgsed down. For these reasons, Kmf =0 and Kmf =0
have different ranges.
The above classical classification of rth branches is also valid in the quantum theory
(including the baryonic branch).
The quantum rth branch can also be discussed by using the SeibergWitten curve. In
the rth branch, the curve factorizes as
2
2
2Nc Nf
(x + mf )Nf
yN
=2 = PNc (x) 4

= (x + mf )2r PN2 c r (x) 42Nc Nf (x + mf )Nf 2r .
Because Nc r 0 (coming from PNc r (x)) and Nf 2r 0 (coming from the last
term), we have r Nc and r Nf /2, which leads to the range

Nf
.
r min Nc ,
(2.17)
2
The relation between this classification of the SeibergWitten curve and the above classification of r-branches, in the mf = 0 and mf = 0 cases, is as follows. In the mf = 0 case, we
have one-to-one correspondence where the r is identified with Kmf =0 . In the mf = 0 case,
for a given r of the curve, there exist two cases: either Kmf =0 = r for Kmf =0 [Nf /2],
or Kmf =0 = Nf r for Kmf =0 [Nf /2].
2.3. The work of IKRSV
Now we discuss another aspect of the model. In the above,
we saw that there is a period

condition (2.11) for T (z). So, if for the ith cut we have Ai T (z) dz = Nc,i = 0, then it
seems that, in the string theory realization of the gauge theory, there is no RR flux provided
11
by D5-branes through this cut and the cut is closed. Because of this, it seems that we
should set the corresponding glueball field Si = 0. Based on this naive expectation, Ref. [4]
calculated the effective superpotential of USp(2Nc ) theory with an antisymmetric tensor
by the matrix model, which turned out to be different from the known results obtained by
holomorphy and symmetry arguments (later Refs. [17,18] confirmed this discrepancy).
This puzzle intrigued several papers [3,15,1924]. In particular, in [20], it was found
that although Nc,i = 0, we cannot set Si = 0. The reason became clear by later studies.
Whether a cut closes or not is related to the total RR flux which comes from both D5branes and orientifolds. For USp(2Nc ) theory with antisymmetric tensor, although the RR
flux from D5-branes is zero, there exists RR flux coming from the orientifold with positive
RR charges, thus the cut does not close. That the cut does not close can also be observed
from the SeibergWitten curve [15] where for such a cut, we have two single roots in the
curve, instead of a double root. All these results were integrated in [3] for N = 1 gauge
theory with adjoint. Let us define N c = Nc for U (Nc ), Nc /2 1 for SO(Nc ) and 2Nc + 2
for USp(2Nc ). Then the conclusion of [3] can be stated as
If N c > 0, we should include Si and extremize Weff (Si ) with respect to it. On the other
hand, if N c 0, we just set Si = 0 instead.
In [3], it was argued that this prescription of setting Si = 0 can be explained in string theory
realization by considering an extra degree of freedom which corresponds to the D3-brane
wrapping the blown up S 3 and becomes massless in the S 0 limit [25]. Our second
motivation of this paper is to generalize this conclusion to the case with flavors. We will
discuss the precise condition when one should set Si = 0 in order to get agreement with
the gauge theory result, in the case with flavors.
2.4. Prospects from the strong coupling analysis
Before delving into detailed calculations, let us try to get some general pictures from
the viewpoint of factorization of the SeibergWitten curve. Since we hope to generalize
IKRSV, we are interested in the case where some Si vanish. Because Si is related to the
size of a cut in the matrix model curve, which is essentially the same as the SeibergWitten
curve, we want some cuts to be closed in the SeibergWitten curve. Namely, we want a
double root in the factorization of the curve, instead of two single roots.
For U (Nc ) theory with Nf flavors of
the same mass mf = zf , tree level superpotential
(2.1), and breaking pattern U (Nc ) ni=1 U (Nc,i ), the factorization form of the curve is
[13]
PNc (z)2 42Nc Nf (z zf )Nf = HNn (z)2 F2n (z),
F2n (z) = W (z)2 + fn1 (z),
(2.18)
where the degree 2n polynomial F2n (z) = W (z)2 + fn1 (z) generically has 2n single
roots. How can we have a double root instead of two single roots?
For a given fixed mass, for example zf = a1 , there are three cases where we have a
double root, as follows. (a) There is no U (Nc,1 ) group factor associated with the root a1 ,
12
namely Nc,1 = 0. (b) The U (Nc,1 ) factor is in the baryonic branch. This can happen for
Nc,1 Nf < 2Nc,1 . (c) The U (Nc,1 ) factor is in the rth nonbaryonic branch with r = Nc,1 .
This can happen only for Nf 2Nc,1 . Among these three cases, (b) and (c) [13] are new
for theories with flavors, and will be the focus of this paper. However, it is worth pointing
out that the factorization form in the cases (b) and (c) are not the one given in (2.18) for a
fixed mass, but the one given in (A.3).
Can we keep the factorization form (2.18) while having an extra double root? We can,
but instead of a fixed mass we must let the mass floating, which means the following.
There will be multiple solutions to the factorization form (2.18), and for any given solution
the 2n single roots of F2n (x), denoted by ai , i = 1, . . . , n, are functions of zf . Now, we
tune zf so that ai+ (zf ) = ai (zf ), i.e., so that two single roots combine into one double
root. Since for different solutions this procedure will lead to different values of zf , we call
this situation the floating mass.
Now we have two ways to obtain extra double roots: one is with a fixed mass, but to
go to the baryonic or the r = Nc,1 nonbaryonic branch, while the other is to start with a
general nonbaryonic branch but using a floating mass. In fact it can be shown that these two
methods are equivalent to each other when the double root is produced. In the calculations
in Section 4 we will use the floating mass to check our proposal.
3. One cut modelquadratic tree level superpotential

In this section we will study whether a cut closes up if one tries to pass too many poles
through it. If the poles are near the cut, the precise form of the tree level superpotential
(namely the polynomials W (z), mf f (z)) is inessential and we can simplify the problem
to the quadratic tree level superpotential given by (2.14). For this superpotential, we will
compute the effective glueball superpotential using the formalism reviewed in the previous
section. Then, by solving the equation of motion, we study the on-shell process of sending
poles through the cut, and see whether the poles can pass or not.
Also, on the way, we make an observation on the relation between the exact superpotential and the vacuum expectation value of the tree level superpotential.
3.1. The off-shell Weff , M(z) and T (z)
First, let us compute the effective glueball superpotential for the quadratic superpotential (2.14). The matrix model curve (2.7) is related to W (z) in this case as

2
= W (z)2 + f0 (z) = m2A z2 m2A z2 ,
ym
.
m2A
(3.1)
Let us consider the case with K poles on the first sheet at qI = qf , for which rI = 1, and
with (Nf K) poles on the second sheet at qI = qf , for which rI = 0 (recall that (qf , qf )
is the lift of zf defined in (2.15)). Using the curve (3.1) and various formulas summarized
13
in the previous section, one can compute

1
mA
=
S=
R(z) dz =
,
2i
4
4mA
A
0
y(z) dz = mA 20 2S 2S log
=2
20 mA
,
S
rI =0
f,I
0
0
= y(z) dz = y(z) dz
q I
rI =1
f,I
qI
mA 20
zI
2S log
2
0

mA zI2
1 1
4S
4S
1
+ 2S log
1
1
+ ,
+
+
2 2
4S
mA zI2
mA zI2 2
0
0
= y(z) dz = y(z) dz
qI
qI
mA 20
zI
2S log
2
0

mA zI2
1 1
4S
4S
1
,
+ 2S log
1
1
+
2 2
4S
mA zI2
mA zI2 2
where we dropped O(1/0 ) terms. We have traded qI , qI for zI in the square roots, so

1 4S 2 1 2S 2 and zI2 m4SA zI for |zI |
that the sign convention is such that
mA zI
very large. Substituting this into (2.12), we obtain

Weff (S)

Nc 2Nc Nf

mA
I zI
= S Nc + log
S Nc

1 1
4S
+
+
S log
1
2 2
mA zI2
I,rI =0

1 1
4S
S log
1
2 2
mA zI2
I,r =1
I
mA zI

4S
1
1
1 +
2
mA zI2

mA zI2
4S
1
1
,
1
+
4S
2
mA zI2
(3.2)
mA zI2
4S
where is the dynamical scale of the corresponding N = 2 gauge theory defined in (2.13).
14
Let us compute resolvents also. The resolvent M(z) is an Nf Nf matrix. Using (2.8),
we find that for the I th eigenvalue
MI (z) =
R(z)
(1 rI )R(qI ) rI R(qI )
+
+
.
(z qI )
(z qI )
(z qI )
Expanding MI (z) around z = we can read off the following vacuum expectation values

mA
zI + (2rI 1) zI2 ,
2

mA 2
2zI + 2(2rI 1)zI zI2 ,

QQ
I =
4

Nf mA
= Nf S.
(QQ
zI QQ)
=
4
I=
QQ
There is something worth noting here. One might naively expect that the exact superpotential is simply the vacuum expectation value of the tree level superpotential (2.14), as
is the case without
flavors. However, this naive expectation is wrong! Although we have
Wtree,fund = I (QQ
zI QQ)
= 0 if S = 0, we still have

Weff,on-shell =

mA
Tr 2 ,
2
(3.3)
as we will see shortly. The reason for (3.3) can be explained by symmetry arguments
[26,27]. Although the tree level part Wtree,fund for fundamentals is generically nonzero
and also contributes to Wlow , the contribution is precisely canceled by the dynamically
generated superpotential Wdyn , leaving only the part of Wtree .4
N
Let us calculate the resolvent T (z) also. For the present example, c(z) = mA (Nc 2f )
from (2.10). Therefore, using (2.9), we obtain the expansion of the resolvent T (z):

2rI 1 2
Nc zI
1
+
+
T (z) =
zI 2
z
2
2
z
I

z2 2rI 1
(2N
1
c Nf )
I
2
+
+ .
+
+
zI zI
4
2
2
z3
I
From this we can read off Tr n . For example, for K = 0 we have

Nf
zf zf2 ,
2

(2Nc Nf )
N
f
zf zf zf2 +
.
Tr 2 =
2
4
Tr =
4 We would like to thank K. Intriligator for explaining this point to us.
(3.4)
15
3.2. The on-shell solution

Now we can use the above off-shell expressions to find the on-shell solution. First we
rewrite the superpotential as

3Nc Nf
1
Weff = S Nc + log
S Nc

zf
1 2
4S
+
(Nf K)S log
zf
2
2
mA

mA zf
4S
1
+
zf2
zf +
4S
mA
2

zf
mA zf
4S
4S
1 2
1
2
KS log
z
zf + ,
zf
+
2
2 f mA
4S
mA
2
(3.5)
3Nc N
f
c
2Nc Nf . We have set all z to be z , i.e., all masses are the same,
where 1
mN
I
f
A
as in (2.15). From this we obtain5

z z2 4S
3Nc Nf
f
1
f
mA
+ K log
0 = log
S Nc
2

z + z2 4S
f
f
mA
+ (Nf K) log
(3.6)
,
2
or6

z z 2 4S
f
Nc
f
0 = log S
+ K log
2

z + z 2 4S
f
f
+ (Nf K) log
,
2
(3.7)
5 It is easy to check that Eq. (3.6) with parameters (N , N , K) is the same as the one with parameters
c
f
(Nc r, Nf 2r, K r). Also from the expression (3.4) it is straightforward to see we have Tr 2 Nc ,Nf ,K =
Tr 2 Nc r,Nf 2r,Kr + rzI2 . All of these facts are the result of the addition map observed in [13]. Furthermore, one can show that both (3.6) and (3.4) for K = 0 are exactly the same as the one given by the strong
coupling analysis in [16] and the weak coupling analysis in [28,29].
6 As we mentioned before, for m = 0 the allowed Higgs branch requires K N but the strong coupling
f
f
analysis gives K Nf /2. The resolution for that puzzle is that if K > Nf /2, it is given by (Nf K)th branch
of the curve. Since the same rth branch of the curve gives both r and (Nf r) Higgs branches, we expect that r
and (Nf r) Higgs branches are related. This relation is given by S = Sb2 , z f = zf b with b = mA 2 /S. It can
be shown that with the above relation, the equation of motion of S for the Kth branch is changed to the equation
of motion of S for the (Nf K)th branch.
16
where we have defined dimensionless quantities S =
42
and z f = f . Note that

2 S].
using these massless quantities the cut is from along the interval [2 S,
Using (3.6) or (3.7) it is easy to show that

mA zf2
Nf
4S
.
S+
Nf + (2K Nf ) 1
Weff,on-shell = Nc
2
4
mA zf2
S
mA 2
Also by expanding T (z) in the present case, just as we did in (3.4), we can read off

mA zf2
Nf
4S
mA
2
Tr = Nc
,
S+
Nf + (2K Nf ) 1
2
2
4
mA zf2
which gives us the relation

mA
Tr 2 = Weff,on-shell
2
as we promised in (3.3).
Eq. (3.6) is hard to solve. But if we want just to discuss whether the cut closes up when
we bring zf 0, we can set K = 0,7 for which (3.6) reduces to
r (Nf Nc )/Nf
r Nc /Nf
zf = N
+ N
.
S
S
f
f
(3.8)
Here, Nf is the Nf th root of unity, Nf = e2i/Nf , and r = 0, 1, . . . , (Nf 1) corresponds to different branches of solutions. It is also amusing to note that above solution
has the Seiberg duality [30] where electric theory with (Nc , Nf ) is mapped to a magnetic
theory with (Nf Nc , Nf ).
With these preparations, we can start to discuss the on-shell process of passing poles
through the cut from the second sheet.
3.3. Passing poles through a branch cut
Consider moving Nf poles on top of each other at infinity on the second sheet toward
the cut along a line passing through the origin and making an angle of with the real axis.
Namely, take8
z(qf ) = z(qf ) = pei ,
p, R,
(3.9)
and change p from p = to p = (see Fig. 1). This Eq. (3.9) needs some explanation.
Remember that zf = z(qf ) = z(qf ) denotes the projection from the Riemann surface
(Eq. (3.1)) to the z-plane. For each point z on the z-plane, there are two corresponding
points: q on the first sheet and q on the second sheet. Although we are starting with poles
at qf on the second sheet, we do not know in advance if the poles will pass through the cut
7 If K = 0, we will have U (N ) U (K) U (N K) and the problem reduces to of U (N K) with
c
c
c
(Nf 2K) flavors in the 0th branch.
8 Throughout this subsection, we will use the dimensionless quantities z , S,
f etc., and omit the hats on them
to avoid clutter, unless otherwise mentioned.
17
Fig. 1. A process in which Nf poles at qf on the second sheet far away from the cut approach the branch cut on
the double sheeted z-plane, along a line which goes through the origin and makes an angle with the real z axis.
The with dotted lines denotes the poles on the second sheet, moving in the direction of the arrow. The two
branch points a are connected by the branch cut, which is denoted by a zigzag.
and end up on the first sheet, or it will remain on the second sheet. Therefore we cannot
specify which sheet the poles are on, and that is why we used z(qf ), z(qf ) in (3.9), instead
of qf or qf .
Below, we study the solution to the equation of motion (3.8), changing p R from
p = to p = . By redefining zf , S by zf zf e2ir/(Nf 2Nc ) , S Se4ir/(Nf 2Nc ) ,
we can bring (3.8) to the following form:
zf = pei = S t + S 1t ,
(3.10)
where
Nc
t.
Nf
Henceforth we will use (3.10). Because zf as well as S is complex, the position of the
branch points (namely, the ends of the cut), a, where
a 4S
is also complex, which means that in general the cut makes some finite angle with the real
axis, as shown in Fig. 1.
Nf = Nc
As the simplest example, let us first consider the Nf = Nc (i.e., t = 1) case. We will see
that the poles barely pass through the cut but get soon bounced back to the second sheet.
The equation of motion (3.10) is, in this case,
zf = pei = S + 1.
(3.11)
Therefore, as we change p, the position of the branch points changes according to

a = 4S = 2 zf 1 = 2 pei 1.
(3.12)
18
Let us look closely at the process, step by step. The point is that transition between the
first and the second sheet can happen only when the cut becomes parallel to the incident
direction of the poles, or when the poles pass through the origin.
(1) p +, on the second sheet. In this case, we can approximate the right-hand side of
(3.12) as
1
1 i
1 i
1 1 i
a = 2p 2 e 2 1 p1 ei 2 2p 2 e 2 e 2 p e
= 2p 2 e 2 p
1
1 cos
ei( 2 + 2 p
1 sin )
Therefore, when the poles are far away, the angle between the cut and the real axis
is approximately 2 > 0 (we assume 0 < < 2 ). Furthermore, as the poles approach
1
(p becomes smaller), the cut shrinks (because of p 2 ) and rotates counterclockwise

i 1
(because of e 2 p sin ). This corresponds to Fig. 2(a).
(2) Because the cut is rotating counterclockwise, as the poles approach, the cut will eventually become parallel to the incident direction, at some point. This happens when

a 2 = 4 pei 1 = 4 (p cos 1) + ip sin e2i = cos 2 + i sin 2.
By simple algebra, one obtains
p = 2 cos ,
a = 2ei .
(3.13)
Note that this is the only solution; the cut becomes parallel to the incident direction
only once. Because 0 < p < |a| = 2 (we are assuming 0 < < /2), by the time
the cut becomes parallel to the incident direction, the poles have come inside of the
interval [2ei , 2ei ], along which the cut extends when it is parallel to the incident
direction. This implies that the poles cross9 the cut at this point, and enter into the
first sheet. Fig. 2(b) shows the situation when this transition is about to happen. In
Fig. 2(c), the poles are just crossing the cut. Fig. 2(d) corresponds to the situation just
after the transition happened; the poles have passed the cut and are now proceeding on
the first sheet. Here we implicitly assumed that the cut is still rotating counterclockwise
with a finite angular velocity, but this can be shown by expanding a around (3.13) as
p = 2 cos + p. A short computation shows
1
a 2e 2 p cos ei 2 p sin
which implies that the cut shrinks and rotates counterclockwise if we move the poles
to the left (p decreases).
(3) p 0. If the poles proceed on the real line further, it eventually reaches the origin
p = 0. By expanding (3.12) around p = 0, one obtains

1
1
1
a = 2 ei + pei 2 2e 2 p cos ei( 2 2 p sin ) .
9 As mentioned in footnote 2, there is no real physical meaning to the position of the cut itself.
(3.14)

Fig. 2. Six configurations of the branch cut and the poles. The poles are depicted by and moving along a line at an angle with the real axis, as the arrow on
it indicates. The in solid (dotted) lines denotes poles on the first (second) sheet. The branch cut is rotating counterclockwise (as the arrows on its sides indicate),
changing its length.
19
20
Therefore the cut has a finite size (|a| = 2) at p = 0 and along the imaginary axis,
still rotating counterclockwise, but now expanding. Because the cut goes through the
origin, the poles pass through the cut again and comes back onto the second sheet
(Fig. 2(e)).
(4) p , on the second sheet. If the poles have gone far past the cut so that p < 0,
|p| 1, we can approximate (3.12), as before, as

1
1 i
1 1
1
1
1
a = 2|p| 2 e 2 (+) 1 p1 ei 2 2|p| 2 e 2 |p| cos ei( 2 + 2 2 |p| sin ) .
Therefore, as the poles go away, the cut expands and rotates counterclockwise. The
angle between the cut and the real axis asymptotes to ( 2 + 2 ) (Fig. 2(f)).
In the above we assumed that 0 < < 2 . If 2 < < , the only difference is that
the order of steps (2) and (3) are exchanged. If < 0, the cut rotates clockwise instead of
counterclockwise.
When is the cut shortest in this whole process? From (3.12), one easily obtains

1/4
|a| = 2 (p cos )2 + sin2
2| sin |1/2.
(3.15)
Therefore, when the poles are at zf = pei = cos ei , which is between the steps (2)
and (3) above, the cut becomes shortest. In particular, in the limit 0 or , the
cut completely closes up instantaneously. These correspond to configurations with either
a horizontal cut with poles colliding sideways, or a vertical cut with poles colliding from
right above or from right below. Actually the existence of the S = 0 solution is easy to see
in (3.11): it is just zf = 1, S = 0.
Summary: for Nf = Nc , when one moves poles on the second sheet from infinity along
a line toward a cut, poles pass through the cut onto the first sheet and move away from the
cut by a short distance. Then poles are bounced back to the second sheet again. Therefore,
one can never move poles far away from the cut on the first sheet. During the process, in
certain situations, the cut completely closes up.
Nf =
Nc
Now let us consider a more general case with Nf = Nc . We again consider a situation
where poles on the second sheet approach a cut. This time we will be brief and sketchy,
because a detailed analysis such as the one we did for the Nf = Nc case would be rather
lengthy due to the existence of multiple branches, and would not be very illuminating.
First, let us ask how we can see whether poles are on the first sheet or on the second
sheet, from the behavior of S versus p. Because this is not apparent in the equation of
motion of the form (3.10), let us go back to

z
zf2 4S
f
Weff
Nc
log S

0=
+ Nf log
S
2

zf zf2 4S
St =
(3.16)
2
21
which led to Eq. (3.10). Here the (+) sign corresponds to q f on the first (second) sheet. For |zf |2 |4S|, the square root can be approximated as zf2 4S = zf (1
4S/zf2 )1/2 zf (1 2S/zf2 ) (our sign convention was discussed above (3.2)). Therefore
(3.16) is, on the first sheet,
St
zf zf (1 2S/zf2 )
2
S
zf
zf S 1t ,
(3.17)
while on the second sheet

zf + zf (1 2S/zf2 )
zf zf S t .
(3.18)
2
Now let us solve (3.10) for |zf | 1. By carefully comparing the magnitude of the two
terms in (3.10), one obtains

1
zf S t |S| |p| t , |S| 1,
Nf < Nc (1 < t)
1
zf S 1t |S| |p| t1 , |S| 1,

1
1
zf S t |S| |p| t , |S| 1,
<t <1
Nc < Nf < 2Nc
2

1
1
zf S 1t |S| |p| 1t , |S| 1.
0<t <
2Nc < Nf
2
(3.19)
S
t
It is easy to show that |zf |2 |4S| in all cases. So by using (3.17), (3.18), we conclude
that the first and the third lines in (3.19) correspond to poles on the second sheet, while the
second and the last lines correspond to poles on the first sheet. This implies that, only for
Nf < Nc , poles on the second sheet can pass through the cut all the way and go infinitely
far away on the first sheet from a cut, as we will see explicitly in the examples below. For
Nc < Nf < 2Nc , if one tries to pass poles through a cut, then either poles will be bounced
back to the second sheet, or the cut closes up before the poles reach it. For Nf > 2Nc ,
there is no solution corresponding to poles moving toward a cut from infinity on the second
sheet. This should be related to the fact that in this case glueball S is not a good IR field.
Therefore, this one cut model is not applicable for Nf > 2Nc .
To argue that these statements are true, rather than doing an analysis similar to the one
we did for the Nf = Nc case, we will present some explicit solutions for some specific
values of t = NNfc and , and argue general features. Before looking at explicit solutions,
note that there are multiple solutions to Eq. (3.10) which can be written as
N
N N

zf = S 1/Nf c + S 1/Nf f c .
(3.20)
From the degree of this equation, one sees that the number of the solutions to (3.20) is:
Nf Nc
Nc Nf 2Nc
2Nc Nf
2Nc Nf solutions,

Nc solutions,
Nf Nc solutions.
22
branch (i)
branch (ii)
branch (iii)
Fig. 3. The graph of |S| versus p for Nf = 12 Nc (t = 2), = /6. The vertical axis is |S| and the horizontal axis
is p. Although we are showing just the = /6 case, there are similar looking three branches for any value of ,
which change into one another when is changed by 2/3.
Therefore, if we solve the equation of motion (3.10), in general we expect multiple

branches of solutions. We do not have to consider the Nf branches of the root S 1/Nf ,
because it is taken care of by the phase rotation we did above (3.10).
Now let us look at explicit solutions for U (2) example. For Nf = 12 Nc (t = 2), the
solution to the equation of motion (3.10) is

27 + 729 108z3 1/3 27 + 729 108z3 1/3
f
f
1
,
S =
3
2
2
where three branches of the cubic root are implied. In Fig. 3 we plotted |S| versus p for
these branches, for a randomly chosen value of the angle of incidence, = /6. Even if
one changes , there are always three branches whose general shapes are similar to the
ones in Fig. 3. These three branches changes into one another when is changed by 2/3.
One can easily see which branch corresponds to what kind of processes, by the fact that on
the first sheet |S| 1 as |p| , while on the second sheet |S| 1 as |p| . The
three branches correspond to the process in which: (i) poles go from the second sheet to
the first sheet through the cut, without any obstruction, (ii) poles go from the first sheet to
the second sheet (this is not the process we are interested in), and (iii) poles coming from
the second sheet get reflected back to the second sheet. Note that the cut has never closed
in all cases, because |S| is always nonvanishing.
Similarly, for Nf = 32 Nc (t = 23 ), the solution to the equation of motion (3.10) is
S=

1
3zf 1 (zf + 1) 4zf + 1 .
2
branch (i)
23
branch (ii)
Fig. 4. The graph of |S| versus p for Nf = 32 Nc (t = 23 ) or Nf = 3Nc (t = 13 ), for = /2. The two values
of t give the same graph. The vertical axis is |S| and the horizontal axis is p. Although we are showing just the
= /2 case, there are similar looking two branches for any value of , which change into each other when is
changed by .
This time there are two branches, which change into each other when the is changed
by . We plotted |S| versus p for = /2 in Fig. 4. It shows two possibilities: (i) poles
coming from the second sheet get reflected back to the second sheet, for which |S| = 0 as
p 0, (ii) the cut closes up before poles passes through it, for which |S| 0 as p 0.
The Nf = 3Nc (t = 13 ) case is also described by the same Fig. 4. However, as we
discussed below (3.19), it does not correspond to a process of poles approaching the cut
from infinity on the second sheet; it corresponds to poles on the first sheet and we cannot
give any physical interpretation to it.
These demonstrate the following general features:
For Nf < Nc , one can move poles at infinity on the second sheet through a cut all the
way to infinity on the first sheet without obstruction, if one chooses the incident angle
appropriately. If the angle is not chosen appropriately, the poles will be bounced back
to the second sheet.
For Nc Nf < 2Nc , one cannot move poles at infinity on the second sheet through
a cut all the way to infinity on first sheet. If one tries to, either (i) the cut rotates and
sends the poles back to the second sheet, or (ii) the cut closes up before the poles reach
it.
For 2Nc < Nf , the one cut model does not apply directly.
Nf = 2Nc is an exceptional case, for which the equation of motion (3.10) becomes
zf = 2S 1/2 .
(3.21)
Therefore |S| 0 as zf 0, and the cut always closes before the poles reach it.
Subtlety in S = 0 solutions
If p = 0, or equivalently if zf = 0, there is a subtle, but important point we overlooked
in the above arguments. For zf = 0, the superpotential (3.5) becomes

Nc N /2
(1)Nf /2 mA f 2Nc Nf
Nf
Weff = S Nc
+ log
2
S Nc Nf /2
24

Nc N /2 2Nc Nf
(1)Nf /2 mA f 0
Nf
+ log
+ 2i0 S
Nc
2
S Nc Nf /2

3
0
Nf
= Nc
(3.22)
S 1 + log
+ 2i0 S
2
S
=S
3(Nc N /2)
Nc N /2
2Nc N
f
f
(1)Nf /2 mA f 0
. Only from here to (3.23), S means
with 0
see below (3.7)). In addition, in the second line
the dimensionful quantity (S = mA 2 S;
of (3.22), we rewrite the renormalized scale in terms of the bare scale 0 and the bare
coupling 0 using the relation (2.13). In our case, BL = (1)Nf . The equation of motion
derived from (3.22) is

3
0
Nf
Nc
log
+ 2i0 = 0
2
S
and the solution is

2 i0
3 e Nc Nf /2 , Nf = 2Nc ,
S=
0
no solution, Nf = 2Nc .
(3.23)
For Nf < Nc , (3.23) is consistent with the fact that the |S| versus p graphs in Fig. 3
all go through the point (p, |S|) = (0, 1) (now S means the dimensionless quantity). Also
for Nf = Nc , (3.23) is consistent with the result (3.14) (|a| = 2, so |S| = 1). On the other
hand, for Nf > Nc , (3.23) implies that we should exclude the origin (p, |S|) = (0, 0) from
the |S| p graphs in Fig. 4, which is the only S = 0 solution (this includes the Nf = 2Nc
case (3.21)).10
Therefore, the above analysis seems to indicate that, for Nf > Nc , the S = 0 solution at
p = 0, or equivalently zf = 0 is an exceptional case and should be excluded. On the other
hand, as can be checked easily, gauge theory analysis based on the factorization method
shows that there is an S = 0 solution in the baryonic branch. Thus we face the problem
of whether the baryonic S = 0 branch for Nf > Nc can be described in matrix model, as
alluded to in the previous discussions.
Note that, there is also an S = 0 solution for Nc = Nf in certain situations, as discussed
below (3.15). For this solution, which is in the nonbaryonic branch, there is no subtlety in
the equation of motion such as (3.23), and it appears to be a real on-shell solution. This
will be discussed further below.
3.4. Generalization of IKRSV at one cut model
In the above and in Section 2.4, we argued that for Nf Nc the S = 0 solutions are
real, on-shell solutions based on the factorization analysis. More accurately, there are two
cases with S = 0: the one in the maximal nonbaryonic branch with Nf 2Nc and the other
10 One may think that if one uses the first line of (3.22), then for N = 2N , W = S log[(1)Nf /2 ] and there
c
f
eff
are solutions for some Nf . However, the glueball superpotential that string theory predicts [31,32] is the third
line of (3.22) which is in terms of the bare quantities 0 and 0 . If Nf = 2Nc , then the log term vanishes and
one cannot define a new scale as we did in (2.13) to absorb the linear term 2 i0 S.
25
one in the baryonic branch with Nc Nf < 2Nc . The case of nonbaryonic branch cannot
be discussed in the one cut model, which is applicable only to Nf < 2Nc . On the other
hand, the baryonic one did show up in the previous subsection, but we just saw above that
those solutions should be excluded by the matrix model analysis. What is happening? Is it
impossible to describe the baryonic branch in matrix model?
Recall that the glueball field S has to do with the strongly coupled dynamics of U (Nc )
theory. That S = 0 in those solutions means that there is no strongly coupled dynamics
any more, namely the U (Nc ) group has broken down completely. The only mechanism for
that to happen is by condensation of a massless charged particle which makes the U (1)
photon of the U (Nc ) group massive. Therefore, in order to make S = 0 a solution, we
should incorporate such an extra massless degree of freedom, which is clearly missing in
the description of the system in terms only of the glueball S. This extra degree of freedom should exist even in the Nf = Nc case where S = 0 really is an on-shell solution as
discussed below (3.15); we just could not directly see the degree of freedom in this case.
The analysis of [3] hints on what this extra massless degree of freedom should be in the
matrix model/string theory context. Note that, the superpotential (3.22) is of exactly the
same form as Eq. (4.5) of [3], if we interpret Nc Nf /2 N as the amount of the net
RR 3-form fluxes. In [3] it was argued that, if the net RR flux N vanishes, one should take
into account an extra degree of freedom corresponding to D3-branes wrapping the blown
up S 3 in the CalabiYau geometry [25], and condensation of this extra degree of freedom
indeed makes S = 0 a solution to the equation of motion. The form of the superpotential
(3.22) strongly suggests that the same mechanism is at work for Nf = 2Nc in the r = Nc
nonbaryonic branch; condensation of the D3-brane makes S = 0 a solution. Furthermore,
as discussed in [3], for Nf > 2Nc the glueball S is not a good variable and should be set
to zero. A concise way of summarizing this conclusion is: if the generalized dual Coxeter number h = Nc Nf /2 is zero or negative, we should set S to zero in the r = Nc
nonbaryonic branch.
However, this is not the whole story, as we have discussed in Section 2.4. As we saw
above, we need some extra physics also for Nc Nf < 2Nc in order to explain the matrix
model result in the baryonic branch. We argue below that this extra degree of freedom at
least in the Nc < Nf < 2Nc case should also be the D3-brane wrapping S 3 which shrinks
to zero when the glueball goes to zero: S 0.
The original argument of [3] is not directly applicable for Nf < 2Nc because there are
nonzero RR fluxes penetrating such a D3-brane (N = 0). These RR fluxes induce fundamental string charge on the D3-brane. Because the D3-brane is compact, there is no place
for the flux to end on (note that this flux is not the RR one but the one associated with the
fundamental string charge). Hence it should emanate some number of fundamental strings.
If there are no flavors, there is no place for such fundamental strings to end on, so they
should extend to infinity. This fact led to the conclusion of [3] that the D3-brane wrapping
S 3 is infinitely massive and not relevant unless N = 0.
However, in our situation, there are places for the fundamental strings to end on
noncompact D5-branes which give rise to flavors [31,32]. In particular, precisely in the
zf = 0 case, where we have S = 0 solutions for Nc < Nf < 2Nc , the D3-brane wrapping
S 3 intersects the noncompact D5-branes in the S 0 limit, hence the 35 strings stretching
between them are massless. Therefore the D3-brane with these fundamental strings on it
26
is massless and should be included in the low energy description. It is well known [33]
that such a D-brane with fundamental strings ending on it can be interpreted as baryons
in gauge theory.11 Condensation of this baryon degree of freedom should make S = 0 a
solution, making the photon massive and breaking the U (Nc ) down to U (0). The precise
form of the superpotential for this extra degree of freedom must be more complicated than
the one proposed in [3] for the case without flavors.
All these analyses tell us the following prescription:
Using the floating mass condition that all Nf poles are on top of one branch
point12 on the Riemann surface, we will have an S = 0 solution for Nf
2Nc . For Nc < Nf < 2Nc there are two solutions: one with S = 0 in the
(3.24)
baryonic branch and one with S = 0 in the nonbaryonic branch. In multi-cut
cases, this applies to each cut by replacing Nc , S with the corresponding
Nc,i , Si for the cut.
In the next section we will discuss the condition we have used in above prescription.
Also by explicit examples, we will demonstrate that when the gauge theory has a solution
with closed cuts (Si = 0), one can reproduce its superpotential in matrix model by setting
the corresponding glueballs Si to zero by hand.
4. Two cut modelcubic tree level superpotential

Now, let us move on to U (Nc ) theory with cubic tree level superpotential, where we
have two cuts. We will demonstrate that for each closed cut we can set S = 0 by hand to
reproduce the correct gauge theory superpotential using matrix model.
Specifically, we take the tree level superpotential to be
N
f

Wtree = Tr W ()
Q I ( zf )QI ,
I =1
g
m
W (z) = z3 + z2 ,
3
2

m

W (z) = gz z +
g(z a1 )(z a2 ).
g
(4.1)
Here we wrote down W (z) in terms of g2 = g, g1 = m for definiteness, but mostly we

will work with the last expression in terms of g, a1,2 . The general breaking pattern in the
pseudo-confining phase is U (Nc ) U (Nc,1 ) U (Nc,2 ), Nc,1 + Nc,2 = Nc , Nc,i > 0. In
the quantum theory, the critical points at a1 and a2 blow up into cuts along the intervals
[a1 , a1+ ] and [a2 , a2+ ], respectively. Namely, we end up with the matrix model curve (2.7),
11 That the D3-brane wrapping S 3 cannot exist for N < N can probably be explained along the same line as
c
f
[33], by showing that those 35 strings are fermionic. Also, note that the gauge group here is U (Nc ), not SU (Nc )
as in [33], hence the baryon is charged under the U (1).
12 This condition will not work for the N = N
f
c,i case.
27
which in this case is

2
= W (z)2 + f1 (z) = g 2 z a1 z a1+ z a2 z a2+ .
ym
(4.2)
We will call the cuts along [a1 , a1+ ] and [a2 , a2+ ] respectively the first cut and the second cut henceforth. One important difference from the quadratic case is that, we can study
a process where Nf 2Nc,i flavor poles are near the ith cut in the cubic case.
As we have mentioned, our concern is whether the cut is closed or not. Also from the
experiences in the factorization it can be seen that for Nf > Nc,i , when closed cut is produced, the closed cut and the poles are on top of each other.13 With all these considerations
we take the following condition to constrain the position of the poles14
zf = a1 .
(4.3)
If there are S1 = 0 solutions in which the closed cut and the poles are on top of each
other, then all such solutions can be found by solving the factorization problem under
the constraint (4.3), since for such solutions z = a1 = a1+ obviously. One could impose a
further condition S1 = 0, or equivalently a1 = a1+ if one wants just closed cut solutions,
but we would like to know that there also are solutions with S1 = 0 for Nf < 2Nc,1 , so we
do not do that.
To summarize, what we are going to do below is: first we explicitly solve the factorization problem under the constraint (4.3), and confirm that the S1 = 0 solution exists when
Nf > Nc,i . Then, we reproduce the gauge theory superpotential in matrix model by setting
S1 = 0 by hand.
Before plunging into that, we must discuss one aspect of the constraint (4.3) and the
r-branches, in order to understand the result of the factorization method. If one solves the
factorization equation for a given flavor mass mf = zf (without imposing the constraint
(4.3)), then in general one will find multiple r-branches labeled by an integer K with range
N
0 K min(Nc , [ 2f ]) (see Eq. (2.17)). This is related to the fact that the factorization
method cannot distinguish between the poles on the first sheet and the ones on the second
sheet. The r-branch labeled by K corresponds to distributing Nf K poles on the second
sheet and K poles on the first sheet. We are not interested in such configurations; we want
to put Nf poles at the same point on the same sheet. However, as we discuss now, we
actually do not have to worry about the r-branches under the constraint (4.3).
The r-branches with different K are different vacua in general. However, under the
constraint (4.3) these r-branches become all identical because at the branch point z = ai
there is no distinction between the first and second sheets. This can be easily seen in the
matrix model approach. From Eq. (2.12), the effective glueball superpotential for the two
cut model with Nf K poles at qf on the second sheet and K poles at qf on the first sheet
13 We do not discuss the N = N case where closed cut and poles are not at the same point. However because
f
c,i
the S = 0 solution in this case is an on-shell solution, we can reproduce the gauge theory result in matrix model
without setting S = 0 by hand.
14 We could choose z = a + or z = a instead of (4.3), but the result should be all the same, so we take
f
f
1
2
(4.3) without loss of generality.
28
is
1
1
1
Weff = (Nc,1 1 + Nc,2 2 ) (Nf K)f(2) Kf(1)
2
2
2
1
1
+ (2Nc Nf )W (0 ) + Nf W (q) i(2Nc Nf )S
2
2
+ 2i0S + 2ib1S1 ,
where the periods are defined by
1
Si =
2i
0
i = 2 y(z) dz,

R(z) dz,
ai
Ai
f(2)
0
= y(z) dz,
q f
(1)
f
qf 0
0
(2)
= y(z) dz =
+
y(z) dz f + f
qf
qf
(1)
q f
(2)
with i = 1, 2. The periods f , f
are associated with the poles on the first sheet and

(2)
the ones on the second sheet, respectively. The contour C2 for f is totally on the second
sheet, while the contour C1 for f(1) is from qf on the first sheet, through a cut, to 0
on the second sheet. These contours are shown in Fig. 5. This r-branch with K poles on
the first sheet can be reached by first starting from the pseudo-confining phase with all Nf
poles at qf (K = 0) and then moving K poles through the cut to qf . The path along which
the poles are moved in this process is the difference in the contours, C1 C2 C.15
When we impose the constraint (4.3), then the difference C vanishes (Fig. 6). Therefore there is no distinction between C1 , C2 and hence f(1) = f(2) for any K. In other
words, all Kth branches collapse16 to the same branch under the constraint (4.3).
Now, let us explicitly solve the factorization problem under the constraint (4.3), and
check that the S = 0 solutions exist as advertised before. In solving the factorization problem, we do not have to worry about the r-branches because there is no distinction among
them under the constraint (4.3). Then, we compute the exact superpotential using the data
from the factorization and reproduce it in matrix model by setting S = 0 by hand when
S = 0 on the gauge theory side.
15 There is ambiguity in taking C; for example we can take C to go around a + in Fig. 5. However, the dif1

ference in C y(z) dz for such different choices of C is 2 inS1 , n Z, which can be absorbed in redefinition
of the theta angle and is immaterial.
16 In fact this collapse was observed in [14,15] for SO(N ) and USp(2N ) gauge groups with massless flavors.
c
c
We have seen that there are only two branches, i.e., Special branch and Chebyshev branch, which correspond to
the baryonic branch and the nonbaryonic branch in U (Nc ) case.
(1)
29
(2)
Fig. 5. Contours C1 and C2 defining f and f , respectively. The part of a contour on the first sheet is drawn
in a solid line, while the part on the second sheet is drawn in a dashed line. The Nf K poles on the second
sheet and the K poles on the first sheet are actually on top of each other (more precisely, their projections to the
z-plane are).
Fig. 6. Under the constraint (4.3), contours C1 and C2 become degenerate: C1 = C2 .
For simplicity and definiteness, we consider the case of U (3) gauge group henceforth.
We consider Nf < 2Nc flavors, namely 1 Nf 5, because Nf 2Nc cases are not
asymptotically free and cannot be treated in the framework of the SeibergWitten theory.
4.1. Gauge theory computation of superpotential
In this subsection, we solve the factorization equation under the constraint (4.3) and
compute the exact superpotential, for the system (4.1) with U (3) gauge group and with
various breaking patterns.
4.1.1. Setup
The factorization equation for U (3) theory with Nf flavors with mass mf = zf is
given by17 [13]
P3 (z)2 46Nf (z zf )Nf

= H 1 (z)2 W (z)2 + f1 (z)
2

= H 1 z z a1 z a1+ z a2 z a2+ ,
(4.4)
17 From the result of (A.3), the matrix model curve with flavors does not change even for N > N , contrary
c
f
to [7].
30
where we set g = 1 for simplicity and W (z) is given by (4.1). The breaking pattern is
assumed to be U (3) U (Nc,1 ) U (Nc,2 ) with Nc,i > 0. Here we used new notations
to clarify the shift of the coordinate below. For quantities after the shift, we use letters
without tildes. Enforcing the constraint (4.3) and shifting z as z = z + a1 , we can rewrite
this relation as follows:
P3 (z)2 46Nf zNf

= H1 (z)2 W (z)2 + f1 (z)

= H1 (z)2 z z a 1+ z a 2 z a 2+

H1 (z)2 z z3 + Bz2 + Cz + D .
(4.5)
Because of the shift, the polynomials P3 (z) and H1 (z) are different in form from P3 (z)
and H 1 (z) in (4.4). We parametrize the polynomials P3 (z) and H1 (z) as
P3 (z) = z3 + az2 + bz + c,
H1 (z) = z A.
(4.6)
The parameters B, C can be written in terms of the parameters in

W (z) = W (z) = z (z + mA ) = z + a1 z + a1 + mA (z a1 )(z a2 )
by comparing the coefficients in (4.2):
B 3 B 2 8 C
B 3B 2 8C
,
a2 =
,
a1 =
4
4
3B 2 8C
2 (a 2 a 1 )2 = (a2 a1 )2 =
.
4
(4.7)
The ambiguity in signs in front of the square roots can be fixed by assuming a1 < a2 .
Finally we undo the shift by noting that

1
mA 2 z 3
z2
1
W (z) = z 3 +
z =
(a1 + a2 ) + a1 a2 z + a13 3a12 a2 .
3
2
3
2
6
(4.8)
With all this setup, we can compute the superpotential as follows. First we factorize the
curve according to (4.5). Then we find the Casimirs U1 , U2 , U3 from P3 (z)18 and solve for
a1 , a2 using the last equation of (4.7). Finally we put all these quantities into (4.8) to get
the effective action as

Nc 3
a 3a12 a2 .
Wlow = Tr W () = U3 (a1 + a2 )U2 + a1 a2 U1 +
6 1
(4.9)
Here, for a1 , a2 , one can use the first two equations of (4.7). We also want to know whether
the first cut (the one along the interval [a1 , a1+ ]) is closed or not; we expect that the cut
18 From the coefficients of P (z), namely a, b and c, one can compute the Casimirs U = 1 Tr k using the
k
3
k
quantum modified Newton relation (A.1) as explained in Appendix C.
31
Table 1
The result of factorization of curves for U (3) with up to Nf = 5 flavors. denotes which inequality Nf and
Nc,1 satisfy
Nf
Breaking pattern
U
(Nc,1 ) U (Nc,2 )
Nf Nc,1
Nc,1 < Nf < 2Nc,1
2Nc,1 Nf
The first cut is

U
(2) U (1)

U
(1) U (2)
open
open

U
(1) U (2)

U
(2) U (1)
closed
open

U
(1) U (2)

U
(2) U (1)

U
(2) U (1)
closed
closed
open

U
(1) U (2)

U
(2) U (1)
closed
closed

U
(1) U (2)

U
(2) U (1)
closed
closed
closes if we try to bring too many poles near the cut. As is obvious from (4.5), this can be
seen from the value of D. If D = 0, the cut is closed, while if D = 0, the cut is open.
4.1.2. The result of factorization problem
We explicitly solved the factorization problem for U (3) gauge theories with Nf =
1, 2, . . . , 5 and summarized the result in Table 1. Let us explain about the table. U
(Nc,1 )
U (Nc,2 ) denotes the breaking pattern of the U (3) gauge group. The hat on the first factor
means that the pole is at one of the branch points of the first cut (Eq. (4.3)) which is associated with the first factor U (Nc,1 ). Of course this choice is arbitrary and we may as well
choose U (Nc,2 ), ending up with the same result. Finally, whether the cut is closed or not
depends on whether D = 0 or not, as explained below (4.9).
Now let us look carefully at Table 1, comparing it with the prescription (3.24) based on
the analysis of the one cut model.
First of all, for Nf 2Nc,1 , the first cut is always closed. We will see below that in these
cases with a closed cut the superpotential can be reproduced by setting S1 = 0 by hand in
the corresponding matrix model, confirming the prescription (3.24) for Nf > 2Nc,1 .

Secondly, the lines for Nf = 3 and U
(2) U (1) correspond to the Nc,1 < Nf < 2Nc,1
part of the prescription (3.24). There indeed are both an open cut solution and a closed
cut solution. We will see below that the superpotential of the closed cut solution can be
reproduced by setting S1 = 0 by hand in the corresponding matrix model. On the other
hand, the superpotential of the open cut solution can be reproduced by not setting S1 = 0,
namely by treating S1 a dynamical variable and extremizing Weff with respect to it. In fact
these two solutions are baryonic branch for a closed cut and nonbaryonic branch for an
open cut.
Finally, for Nf Nc,1 , the cut is always open, which is also consistent with the prescription (3.24). In this case, the superpotential of the open cut solution can be reproduced
32
by extremizing Weff with respect to it, as we will see below. For Nf = Nc,1 there should
be an S1 = 0 (a1 = a1+ ) solution for some zf (corresponding to U (2) theory with Nf = 2
in the r = 0 branch) in the quadratic case, but under the constraint (4.3) we cannot obtain
that solution.
Below we present resulting exact superpotentials, for all possible breaking patterns. For
simplicity, we do not take care of phase factor of which gives rise to the whole number
of vacua. For details of the calculation, see Appendix C.
Results

(1) U (2) and Wcl = 16 for U
(2) U (1). For simplicity
Definitions: Wcl = 13 for U
we set g = 1 and = a2 a1 = m/g = 1.

U
(1) U (2) with Nf = 1
Wlow = Wcl 2T
5T 2 115T 3 245T 4 30501T 5 12349T 6

+
+ ,
2
12
4
64
3
T 2 .

U
(2) U (1) with Nf = 1
Wlow = Wcl
5T 2 5T 3 11T 4
235T 6
+
+ 11T 5
+ ,
2
3
3
6
T 3 .

U
(1) U (2) with Nf = 2
Wlow = Wcl + 2T 2 6T 4
T .

U
(2) U (1) with Nf = 2
Wlow = Wcl 2T 4
32T 6
3136T 12
40T 8 192T 10
+ ,
3
3
16T 6
2240T 12
24T 8 128T 10
+ ,
3
3
T .

U
(1) U (2) with Nf = 3
Wlow = Wcl + 2T
19T 2 51T 3 157T 4 5619T 5 33T 6

+
+
+
+
+ ,
2
4
4
64
2
T 2 .

U
(2) U (1) with Nf = 3: two solutions
Wlow,baryonic = Wcl + T
T 3 ,
Wlow = Wcl + 3 .
5T 2
396591T 6
33T 3 543T 4 10019T 5
+ ,
2
2
33

U
(1) U (2) with Nf = 4
176T 3
138T 4 + 792T 6
3
9288T 8 + 137376T 10 2286144T 12 + ,
Wlow = Wcl + 2T 13T 2 +

T .

U
(2) U (1) with Nf = 4
40T 3
4928T 6
56T 4 288T 5
9984T 7
3
3
1244672T 9
2782208T 10 19009536T 11 + ,
63360T 8
3
T 2 .
Wlow = Wcl + T 6T 2

U
(1) U (2) with Nf = 5
Wlow = Wcl 2T
33T 2 1525T 3 3387T 4 314955T 5 74767T 6
+ ,
2
12
4
64
3
T 2 .

U
(2) U (1) with Nf = 5
Wlow = Wcl + T
T .
19T 2 154T 3
+
132T 4 + 828T 6 + ,
2
3
4.2. Matrix model computation of superpotential

In this subsection we compute the superpotential of the system (4.1) in the framework of [1]. If all the Nf flavors have the same mass mf = zf , the effective glueball
superpotential W
(Sj ) for the pseudo-confining phase with breaking pattern U (Nc )
eff

n
n
i=1 U (Nc,i ),
i=1 Nc,i = Nc is, from (2.12),

n
Nf
Nf
Nf
1
f + Nc
W (zf )
Weff (Sj ) =
Nc,i i
W (0 ) +
2
2
2
2
i=1

n1

Nf
bi Si ,
2i Nc
S + 2i0 S + 2i
2
i=1
where the periods associated with adjoint and fundamentals are defined by
0
i (Sj ) 2 y(z) dz,
ai
(4.10)
34
0
0
f (Sj ) y(z) dz = y(z) dz,
y(z) =
z f
zf
W (z)2 + f1 (z).
For cubic tree level superpotential (4.1), the periods 1,2 (Sj ) were computed by explicitly evaluating the period integrals by power expansion in [31], as

2

1
1
+ 2s2 log 0
W (0 ) W (a1 ) + s1 1 + log 0
=
3
3
2g
g
s1

32
91
+ 2s12 + 10s1 s2 5s22 + s13 + 91s12 s2 118s1s22 + s23
3
3

280 4 3484 3
5272 3 871 4
s +
s s2 2636s12s22 +
s1 s2
s + ,
+
3 1
3 1
3
3 2

2

0
2
1
W (0 ) W (a2 ) + s2 1 + log
=
+ 2s1 log 0
3
3
2g
g
s2

32
91
+ 2s22 10s1 s2 + 5s12 + s23 + 91s1 s22 118s12s2 + s13
3
3

5272 3
871 4
280 4 3484 3
s
s1 s2 + 2636s12s22
s s2 +
s + , (4.11)
+
3 2
3
3 1
3 1
where a2 a1 , si Si /g3 , and 0 0 /.
Under the constraint (4.3), the contours defining 1 and f coincide, so
1
1 = f =
2
0
y(z) dz.
(4.12)
zf =a1
Using this, we can rewrite (4.10) as

Weff (S1 , S2 )

Nf

= Nc,1 W (a1 ) + Nc,2 W (a2 )
W (a1 ) W (zf )
2

1
1
Nc,1 1 W (0 ) + W (a1 ) Nc,2 2 W (0 ) + W (a2 )

2
2
2i(Nc,1 + Nc,2 )S + 2i0 S + 2ib1S1 .
(4.13)
Here we rearranged the terms taking into account the fact that the periods take the form
1
2 i = W (0 ) W (ai ) + (quantum correction of order O(Si )), and also the fact that we
are considering zf = a1 a1 (thus the second term). The first line corresponds to the
classical contribution, while the second and third lines correspond to quantum correction.
Furthermore, we defined N c,1 Nc,1 Nf /2.
35
We would like to extremize this Weff (4.13) with respect to S1,2 , and compute the low
energy superpotential that can be compared with the Wlow obtained in the previous subsection using gauge theory methods. In doing that, one should be careful to the fact that one
should treat the mass zf as an external parameter which is independent of S1,2 although
we are imposing the constraint (4.3), zf = a1 = a1 (S1 , S2 ). Where is the zf dependence
in (4.13)? Firstly, zf appears explicitly in the second term in (4.13). Therefore, when we
differentiate Weff with respect to S1,2 , we should exclude this term. Secondly, there is a

more implicit dependence on zf in f = zf 0 y(z) dz, which we replaced with 1 /2
using (4.12). If we forget to treat zf as independent of Si , then we get an apparently un
zf
wanted, extra contribution as Sif = zf 0 y(z)
Si dz Si y(z)|z=zf . However, this last
term actually does not make difference because

y(z)|z=zf = g (z zf ) z a1+ z a2 z a2+
= 0.
z=zf
Therefore what one should do is: (i) plug the expression (4.11) into (4.13), (ii) solve the
equation of motion for S1,2 using (4.13) without the second term, and then (iii) substitute
back the value of S1,2 into (4.13), now with the second term included.
Solving the equation of motion can be done by first writing the VenezianoYankielowicz
term (log and linear terms) as

1
1
Nc,1 1 W (0 ) + W (a1 ) Nc,2 2 W (0 ) + W (a2 )

2
2
2i(N c,1 + Nc,2 )S + 2i0 S + 2ib1S1

= g3 N c,1 s1 1 log s1 /31 + Nc,2 s2 1 log s2 /32 + O si2 ,
where
3N c,1
2(N c,1 +2Nc,2 ) 2i(N c,1 +Nc,2 )2i0 2ib1
= 0
3N
2 c,2
2N +2Nc,2 2i(N c,1 +Nc,2 )2i0

= (1)Nc,2 0 c,1
e
,
and then solving the equation of motion perturbatively in 1,2 . In this way, one can straightforwardly reproduce the results obtained in the previous section in the case with the first
cut open. In the case with the first cut closed, in order to reproduce the results in the previous section, one should first set S1 = 0 by hand, and then extremize Weff with respect to
the remaining dynamical variable S2 .
Following the procedure above, we checked explicitly that extremizing Weff (S1 , S2 )
(open cut) or Weff (S1 = 0, S2 ) (closed cut) reproduces the Wlow up the order presented in
the previous section, for all breaking patterns for U (3) theory.
In the above, we concentrated the explicit calculations of effective superpotentials in
U (3) theory with cubic tree level superpotential. These explicit examples are useful to see
that the prescription (3.24) really works; one can first determine using factorization method
when we should set Si = 0 by hand, and then explicitly check that the superpotential obtained by gauge theory can be reproduced by matrix model.
However, if one wants only to show the equality of the two effective superpotentials
on the gauge theory and matrix model sides, one can actually prove it in general cases. In
36
Appendix B, we prove this equivalence for U (Nc ) gauge theory with an degree (k + 1)
tree level superpotential where k + 1 < Nc . There, we show the following: if there are
solutions to the factorization problem with some cuts closed, then the superpotential Wlow
of the gauge theory can be reproduced by extremizing the glueball superpotential Weff (Si )
on the matrix model side, after setting the corresponding glueball fields Si to zero by hand.
Note that, on the matrix side we do not know when we should set Si to zero a priori; we
can always set Si to zero in matrix model, but that does not necessarily correspond to a
physical solution on the gauge theory side that solves the factorization constraint.
5. Conclusion and some remarks

In this paper, taking N = 1 U (Nc ) gauge theory with an adjoint and flavors, we studied
the on-shell process of passing Nf flavor poles on top of each other on the second sheet
through a cut onto the first sheet. This corresponds to a continuous transition from the
pseudo-confining phase with U (Nc ) unbroken to the Higgs phase with U (Nc Nf ) unbroken (we are focusing on one cut). We confirmed the conjecture of [1] that for Nf < Nc
the poles can go all the way to infinity on the first sheet, while for Nf Nc there is obstruction. There are two types of obstructions: the first one is that the cut rotates, catches
poles and send them back to the first sheet, while the second one is that the cut closes
up before poles reaches it. The first obstruction occurs for Nc Nf < 2Nc whereas the
second one occurs for Nc < Nf .
If a cut closes up, the corresponding glueball S vanishes, which means that the U (Nc )
group is completely broken down. This can happen only by condensation of a charged
massless degree of freedom, which is missing in the matrix model description of the system. With a massless degree of freedom missing in the description, the S = 0 solution
should be singular in matrix model in some sense. Indeed, we found that the S = 0 solution of the gauge theory does not satisfy the equation of motion in matrix model (with
an exception of the Nf = Nc case, where the S = 0 solution does satisfy the equation of
motion). How to cure this defect of matrix model is simplethe only thing the missing
massless degree of freedom does is to make S = 0 a solution, so we just set S = 0 by hand
in matrix model. We gave a precise prescription (3.24) when we should do this, i.e., in
the baryonic branch for Nc,i Nf < 2Nc,i and in the r = Nc,i nonbaryonic branch, and
checked it with specific examples.
The string theory origin of the massless degree of freedom can be conjectured by generalizing the argument in [3]. We argued that it should be the D3-brane wrapping the blown
up S 3 , along with fundamental strings emanating from it and ending on the noncompact
D5-branes in the CalabiYau geometry.
Although we checked that the prescription works, the string theory picture of the S = 0
solution needs further refinement, which we leave for future research. For example, although we argued that some extra degree of freedom makes S = 0 a solution, we do not
have the precise form of the superpotential including that extra field. It is desirable to derive
it and show that S = 0 is indeed a solution, as was done in [3] in the case without flavors.
Furthermore, we saw that there is an on-shell S = 0 solution for Nf = Nc . Although this
solution solves the equation of motion in matrix model, there should be a massless field
37
behind the scene. It is interesting to look for the nature of this degree of freedom. It cannot
be the D3-branes with fundamental strings emanating from it, since for this solution the
noncompact D5-branes are at finite distance from the collapsed S 3 and the 3-5 strings are
massive. Finally, we found that the S = 0 solution is in the baryonic branch. It would be
interesting to ask if one can describe the baryonic branch in the matrix model framework
by adding some extra degrees of freedom.
In the following, we study some aspects of the theory, which we could not discuss so
far. We will discuss generalization to SO(Nc ) and USp(2Nc ) gauge groups by computing
the effective superpotentials with quadratic tree level superpotential.
5.1. SO(Nc ) theory with flavors
Here we consider the one cut model for SO(Nc ) gauge theory with Nf flavors. The tree
level superpotential of the theory is obtained from N = 2 SQCD by adding the mass mA
for the adjoint scalar
mA

Tr 2 + Qf Qf Jff + Qf m
ff Qf ,
Wtree =
(5.1)
2
ff
where f = 1, 2, . . . , 2Nf and the symplectic metric Jff and mass matrix for quark m
are given by

0 1
J=
INf Nf ,
1 0

0 1
m
=
diag(m1 , . . . , mNf ).
1 0
For this simple case the matrix model curve is given by

y(z)2 = m2A z2 42 .
(5.2)
2
This Riemann surface is a double cover of the complex z-plane branched at the roots of ym
(that is z = 2).
The effective superpotential receives contributions from both the sphere and the disk
amplitudes in the matrix model [6] and the explicit form was given in [1] for U (Nc ) gauge
theory with flavors. Now we apply this procedure to our SO(Nc ) gauge theory with flavors
and it turns out the following expression
1
Weff =
2
n

i=n
!
Nc,i 2
B ir
2Nf 0
1
y(z) dz
4
I =1 q
y(z) dz
I
2Nf
1
1
+ (2Nc 4 2Nf )W (0 ) +
W (zI )
2
2
I =1
i(2Nc 4 2Nf )S + 2i0 S + 2i
n

i=1
bi Si ,
38
where S = S0 + 2
n
i=1 Si
and zI is the root of
Nf

2

z zI2 .
B(z) = det m(z) =
I =1
Since the curve (5.2) is same as the one (3.1) of U (Nc ) gauge theory, we can use the
integral results given there to write down the effective superpotential as
Nc 2 Nc22 Nc 2Nf

2 2 mA
det z
(Nc 2)
+ log
Weff = S
Nc 2
2
S 2

Nf

2S
1 1
+
1
S
log
2 2
mA zI2
I =1,rI =0

mA zI2
2S
1
1
1 +
+
2S
2
mA zI2

Nf

1 1
2S
S
log
1
2 2
mA zI2
I =1,rI =1

mA zI2
2S
1
1 + .
+
1
2S
2
mA zI2

We can solve M(z) and T (z) as did for U (Nc ) gauge theory. For simplicity we take
all rI = 0, i.e., all poles at the second sheet. For the I th block diagonal matrix element of
M(z) (I = 1, . . . , Nf ) it is given by

I =mI )
0
R(z)R(q
zmI
MI (z) = R(z)R(q =m )
I
I
0
z+mI
where

R(z) = mA z z2 42 .
Expanding MI (z) in the series of z we can find Qf Qf Jff + Qf m

ff Qf = 2Nf S.
The gauge invariant operator T (z) can be constructed similarly as follows:
Nf
B (z) y(qI )zI

c(z) 2 R(z)
T (z) =
,
+
2
2
2B(z)
y(z)(z zI ) y(z) z y(z)
I =1
where

Nf

W (z) W ()
zW (z) zI W (zI )
c(z) = Tr
.
z
(z2 zI2 )
I =1
(5.3)
39
For the theory without the quarks, the Konishi anomaly was derived in [17,18,34]. The last
term in (5.3) reflects the action of orientifold. For our example we have
c(z) = mA (Nc Nf )
and
Nf

1
1
2
2
zI zI zI 42 + 2 (Nc 2 Nf )
T (z) = Nc + 3
z
z
I =1
Nf

Nf
2

1 4 2
2
4
4
2
zI
zI 4 zI zI + 2 + 6 (Nc Nf ) 12
+ 5
z
I =1
I =1

1
+O 7 ,
z

2
2
2
2
where for equal mass of flavor, we
get Tr = Nf q(q q 4 ) + 2 (Nc 2
4
4
2
2
4
Nf ) and Tr = Nf q Nf q 2 42 q(q + 2 ) + 6 (Nc 2 Nf ).
Let us assume the mass of flavors are the same and K of them (in this case, rI = 1)
locate at the first sheet while the remainder (Nf K) where rI = 0 are at the second
sheet. Then from the effective superpotential, it is ready to extremize this with respect to
the glueball field S 19

z z 2 2S
32 (Nc 2)Nf
f
f
1
mA
+ K log
0 = log
Nc 2
2
S 2

z + z 2 2S
f
f
mA
+ (Nf K) log
(5.4)
.
2
Or by rescaling the fields S = 2mS 2 , z f =
A
case

z + z 2 4S Nf
f
Nc 2
f
1 = S 2
.
2
zf
one gets the solution and consider for K = 0
This equation is the same as the one in U (Nc ) case with Nc Nc22 , so the discussion of
passing poles will go through without modification and the result is when Nf Nc22 , the
on-shell poles at the second plane cannot pass the cut to reach the first sheet far away from
the cut.
19 One can easily check that this equation with parameters (N , N , K) is equivalent to the one with parac
f
meters (Nc 2r, Nf 2r, K r). The equation of motion for glueball field is the same. Since the equation of
motion for both rth Higgs branch and (Nf r)th Higgs branch is the same, one expects that both branches have
4m2 4
some relation. By redefinition of S AS

Higgs branch and (Nf K) Higgs branch.
2
zf 2mA zf z f we get the final relation between K
S,
S
40
By using the condition (5.4), one gets the on-shell effective superpotential
1
Weff,on-shell = (Nc 2 Nf )S
2

2S
1
2
+ mA zf Nf + (2K Nf ) 1
.
2
mA zf 2
It can be checked that this is the same as 12 mA Tr 2 .
5.2. USp(2Nc ) theory with flavors
For the USp(2Nc ) gauge theory with Nf flavors we will sketch the discussion because
most of them is similar to SO(Nc ) gauge theory. The tree level superpotential is given by
(5.1) but with J the symplectic metric, and m
ff the quark mass given by

0 1
J=
INc Nc ,
1 0

0 1
m
=
diag(m1 , . . . , mNf ).
1 0
We parametrize the matrix model curve and the resolvent R(z) as

2
ym
= W (z)2 + f (z) = m2A z2 + 42 ,

R(z) = mA z z2 + 42 .
The effective superpotential is given by

3
Weff = S(Nc + 1) 1 + log

S

Nf

1 1
2S
+
log
S
1+
2 2
mA zI2
I =1,rI =0

mA zI2
2S
1
1+
1
+
2S
2
mA zI2

Nf

2S
1 1
1+
S
log
2 2
mA zI2
I =1,rI =1

mA zI2
2S
1
1 +
1+
2S
2
mA zI2
and T (z) is given by
Nf
c(z) 2 R(z)
B (z) y(qI )zI
+
.
+
T (z) =
2
2
2B(z)
y(z)
z y(z)
y(z)(z zI )
I =1
41
Note the last term (different sign) compared with the SO(Nc ) gauge theory. From the solu

ff Qf J = 2Nf S = 0,
tion of M(z), we can show that although Qf (J )ff Qf + Qf m
we still have on-shell relation 12 mA Tr 2 = Weff .
The equation of motion is given by20

z z 2 + 4S
f
f
0 = log S Nc 1 + K log
2

z + z 2 + 4S
f
f
+ (Nf K) log
,
2
where S = 2mS 2 , z f = f . From this we can read out the following result: when Nf
A
Nc + 1, the on-shell poles at the second sheet cannot pass through the cut to reach the first
sheet far away from the cut.
z
Acknowledgements
We would like to thank Freddy Cachazo, Eric DHoker, Ken Intriligator, Romuald Janik,
Per Kraus, and Hirosi Ooguri for enlightening discussions. This research of C.A. was
supported by a grant in aid from the Monell Foundation through Institute for Advanced
Study, by SBS Foundation, and by Korea Research Foundation Grant (KRF-2002-015CS0006). The work of B.F. is supported by the Institute for Advanced Study under NSF
grant PHY-0070928. The work of Y.O. was supported by JSPS Research Fellowships for
Young Scientists. The work of M.S. was supported by NSF grant 0099590.
Appendix A. On matrix model curve with Nf (> Nc ) flavors
In this appendix we prove by strong coupling analysis that matrix model curve corresponding to U (Nc ) supersymmetric gauge theory with Nf flavors is exactly the same as the
one without flavors when the degree (k + 1) of tree level superpotential Wtree is less than
Nc .21 This was first proved in [32] but the derivation was valid only for the range Nf < Nc .
Then in [13], the proof was extended to the cases with the range 2Nc > Nf Nc . However,
in [13], the characteristic function PNc (x) was defined by PNc (x) = det(x ), without
taking into account the possible quantum corrections due to flavors. In consequence, it appeared that the matrix model curve is changed by addition of flavors. In this appendix, we
use the definition of PNc (x) proposed in Eq. (C.2) of [1]:
20 One can easily check that this equation with parameters (2N , N , K) is equivalent to the one with parac
f
meters (2Nc 2r, Nf 2r, K r). In other words, the equation of motion for glueball field is the same. Since
the equation of motion for both rth Higgs branch and (K r)th Higgs branch is equivalent to each other, one
expects that both branches have some relation.
21 The generalized Konishi anomaly equation of R(z) given in (2.6) is same with or without flavors, so the
form of the solution is the same for gauge theory with or without flavor. In this appendix we use another method
to prove this result.
42
PNc (x) = x
Nc
exp

Ui
i=1
xi
!
2Nc Nf
Ui
B(x)
,
exp
x Nc
xi
(A.1)
i=1
which incorporates quantum corrections and reduces to PNc (x) = det(x ) for Nf = 0,
and see clearly that the matrix model curve is not changed, even when the number of flavors
is more than Nc . Since PNc (x) is a polynomial in x, (A.1) can be used to express Ur with
r > Nc in terms of Ur with r Nc by imposing the vanishing of the negative power terms
in x.
Assuming that the unbroken gauge group at low energy is U (1)n with n k, the factorization form of SeibergWitten curve can be written as,
PN2 c (x) 42Nc Nf B(x)

= HN2 c n (x)F2n (x).
The effective superpotential with this double root constraint can be written as follows:22

P (x) 2 Nc N2f B(x)
N
k
c n

Nc
i
Weff =
gr Ur+1 +
dx
Li
x pi
r=0
i=1

P (x) 2 Nc N2f B(x)
Nc
i
+ Bi
dx
.
(x pi )2
The equations of motion for Bi and pi are given as follows respectively:

P (x) 2 Nc N2f B(x)
Nc
i
dx,
0=
(x pi )2

P (x) 2 Nc N2f B(x)
Nc
i
0 = 2Bi
dx.
(x pi )3
Assuming that the factorization form does not have any triple or higher roots, we obtain
Bi = 0 at the level of equation of motion. Next we consider the equation of motion for Ur :

!
N
c n

PNc
Li
Ui
x Nc
0 = gr1 +
2 r exp
dx,
r
i
x
x
x
x pi
i=1
i=1
PNc
Ur
where we used Bi = 0 and (A.1) to evaluate

. Now, as in [8], we multiply this by zr1
and sum over r:

Nc n
Li
PNc
W (z) =
dx
x z
x pi
i=1
! N n

c

2x Nc
Uk
Li
exp
dx.
+
xz
xk
x pi
k=1
i=1
22 If we want to generalize this proof to more general cases in which k + 1 is greater than and equals to N , we
c
have to take care more constraints like Appendix A in [8], which should be straightforward.
43
Defining the polynomial Q(x) in terms of

N
c n
i=1
Li
Q(x)
,
=
x pi
HNc n (x)
(A.2)
and also using (A.1) and factorization form, we obtain

Q(x)
Q(x)
PNc
PNc
W (z) =
dx +
dx
x z HNc n (x)
x z HNc n (x)

Q(x) F2n (x)
dx
+
x z

Q(x) F2n (x)
=
dx.
x z
This is nothing but (2.37) in [8]. Since W (z) is a polynomial of degree k, the Q should be
a polynomial of degree (k n). Therefore, we conclude that the matrix model curve is not
changed by addition of flavors:

2
ym
(A.3)
= F2n (x)Q2kn (x) = Wk (x)2 + O x k1 .
Appendix B. Equivalence between Wlow and Weff (Si ) with flavors
In this appendix we prove the equivalence Wlow in U (Nc ) gauge theory with Weff (Si )
in corresponding dual geometry when some of the branch cuts on the Riemann surface
are closed and the degree (k + 1) of the tree level superpotential Wtree is less than Nc .
This was first proved in [32], however, the proof was only applicable in the Nf < Nc
cases. Especially, the field theory analysis in [32] did not work for Nc Nf < 2Nc cases.
Furthermore, as we saw in the main text, for some particular choices of zI (position of the
flavor poles), extra double roots appear in the factorization problem. In Section 4, we dealt
with U (3) with cubic tree level superpotential and saw the equivalence of two effective
superpotentials for such special situations. To include these cases we are interested in the
Riemann surface that has some closed branch cuts. Therefore our proof is applicable for
U (Nc ) gauge theories with Wtree of degree k + 1 (< Nc ) in which some of branch cuts are
closed and number of flavors is in the range Nc Nf < 2Nc . In addition, we restrict our
discussion to the Coulomb phase.
In the discussion below, we follow the strategy developed by Cachazo and Vafa in [36]
and use (A.1) as the definition of PNc (x). We have only to show the two relations:

Wlow (gr , zI , )|0 = Weff Si 0 ,
(B.1)
Wlow (gr , zI , ) Weff (Si )
=
,
(B.2)
the equivalence of two effective superpotentials in the classical limit and that of the derivatives of the superpotentials with respect to .
44
B.1. Field theory analysis

and n ( k) be the number of U (1) at low energy. Since we
Let k be the order of Wtree
are interested in cases with degenerate branch cuts, let us consider the following factorization form:23

2
PNc (x)2 42Nc Nf B(x)

= F2n (x) Qkn (x)H Nc k (x)

2
F2n (x) HNc n (x) ,
(B.3)
W (x)2 + fk1 (x) = F2n (x)Qkn (x)2 .

If k equals to n, all the branch cuts in F2k (x) are open. The low energy effective superpotential is given by

n+1
l
Nf

i)
Li PNc (pi ) 2i Nc 2 B(p
gr Ur +
Wlow =
r=1
i=1
Nf

+ Qi
PNc (pi ) 2i Nc 2
pi

i) ,
B(p
where l N n and PNc (x) is defined by (C.3) or (C.4) in [1],

!

Ui
2Nc Nf B(x)
exp
.
PNc (x) = det(x ) +
x Nc
xi
(B.4)
i=1
The second term is specific to the Nf Nc case, representing quantum correction. Define
!

Ui
2Nc Nf B(x)
exp
.
K(x)
x Nc
xi
+
i=1
c Nc k
The first term in (B.4) can be represented as det(x ) N
k=0 x sk . The relation
between Ui s and sk s are given by the ordinary Newton relation, ksk + kr=1 rUr sks = 0.
From the variations of Wlow with respect to pi and Qi , we conclude that Qi = 0 at the level
of the equation of motion. In addition, the variation of Wlow with respect to Ur leads to
c
K(pi )
PNc (pi )
N j
=
Li pi c sj r
Li
Ur
Ur
i=1
i=1 j =0
i=1

!
Nc
l
l

Uk
Nc j
2Nc Nf B(pi )
=
Li pi
sj r
Li
exp
.
Nc +r
k
p
p
i
i
i=1 j =0
i=1
k=1
+
gr =
Let us define
l
Li

2Nc Nf
Gr (pi )
i)
B(p
piNc +r
exp

Uk
k=1
pik
!
.
+
23 In the computation below, we will use relation (A.2) and put g

k+1 = 1.
45
By using these relations, let us compute Wcl

Wcl =
Nc
gr x r1
r=1
Nc
Nc
l
Nc j
x r1 pi
r= i=1 j =0
Nc
l
1
Li det(pi )
x
l
sj r Li
i=1
Li Gr (pi )x r1
r=1 i=1
l

i=1
l

i=1
c

det(x )
1
Li
Li det(pi )
Li Gr (pi )x r1
x pi
x
i=1
PNc (x)
Li
x pi
Nc
l
l

i=1
r=1 i=1
K(x)
1
Li
x pi
x
l
1
Li K(pi )
x
l
Li PNc (pi ) +
i=1
i=1
Li Gr (pi )x r1 ,
(B.5)
r=1 i=1
where we dropped O(x 2 ). The fifth term above can be written as
Nc
l
Li Gr (pi )x r1
r=1 i=1
Nc
l
Li Gr (pi )x
r1
r= i=1
l

1
i=1

Li K(pi ) + O x 2 .
After some manipulation with the factorization form (B.3), we obtain a relation
!

Ui
PNc (x) 22Nc Nf B(x)
Qkn (x) F2n (x) =
exp
xi
H Nc k
H Nc k x N
i=1

PNc (x) 2K(x)
=
+ O x 2 .
HNc k
HNc k
Substituting this relation into (B.5) we obtain
l
l

1
K(x)
Wcl = Qkn (x) F2n (x)
Li PNc (pi ) 2Li K(pi ) +
Li
x
x pi
i=1
Nc
l
r= i=1

Li Gr (pi )x r1 + O x 2 .
i=1
46
Finally let us use the following relation:24

Nc
l
l

K(x)
Li
Li Gr (pi )x r1 = O x 2 .
x pi
r=
i=1
(B.6)
i=1
After all, by squaring Wcl , we have

Qkn (x)2 F2n (x) = Wcl 2 (x) + 2
l

1
i=1
bk1 = 2
l

i=1

Li PNc (pi ) 2Li K(pi ) x k1 + O x k2 ,

1
Li PNc (pi ) 2Li K(pi ) .
x
On the other hand the variation of Wlow with respect to is given by

Wlow
1
bk1
=
Li K(pi )
Li PNc (pi ) =
.
2N
N
c
f
2
4
log
i=1
i=1
l
(B.7)
This is one of the main results for our proof. In the dual geometry analysis below, we will
see the similar relation.
In the classical limit, we have only to consider
the expectation value of .
In our assumption, gauge symmetry breaks as U (Nc ) ni U (Nc,i ) we have Tr = ni=1 Nc,i ai .
Therefore in the classical limit Wlow behaves as
Wcl =
n
Nc,i W (ai ).
(B.8)
i
24 For simplicity, we ignore l
i=1 Li . To prove the relation, let us consider the circle integral over C. Until
now, since we assumed pi < x, the point pi should be included in the contour C. Multiplying 1k k 0 and
x
taking the circle integral, we can pick up the coefficient ck1 of x k1 . In addition, if we denote a polynomial

Ui
j
M 2Nc Nf B(x)
exp(
i=1 x i ) j = aj x we can obtain following relation:
x Nc

k1

[M]+
M
=
+
aj x (kj )
xk
xk +

K(x)
= Gk (x) +
aj x (kj ) ,
xk
k1
j =0
j =0
where right-hand side means circle integral of left-hand side. Thus, we obtain the ck1 as
ck1 =
k1

j =0x=0
k1

aj x j k
dx +
x pi
k1

j =0x=0
j =0x=pi
aj x j k

xn
n+1
n=1 pi
aj x j k
dx
x pi
dx +
k1

j =0
j k
aj pi
= 0,
1 =
xn
where we used that around x = 0, xp
n=1 p n+1 . Therefore, the left-hand side of (B.6) can be written
i
i

2
j
j
2
as
j = cj x = j = cj x = O(x ).
47
By comparing (B.7) and (B.8) and the similar result which we will see in the dual geometry
analysis below, we will show (B.1) and (B.2). Let us move to the dual geometry analysis.
B.2. Dual geometry analysis with some closed branch cuts
As we have already seen in the main text, a solution with Si = 0 appears for some
special choice of zI , the position of flavor poles. In our present proof, however, we put
some of Si to be zero from the beginning, without specifying zI . More precisely, what we
prove in this Appendix is as follows: for a given choice of zI , if there exists a solution
to the factorization problem with some of Si vanishing, then we can construct a dual
geometry which gives the same low energy effective superpotential as the one given by the
solution to the factorization problem, by setting some of Si to zero from the beginning.
Therefore, this analysis does not tell us when a solution with Si = 0 appears. To know
that within the matrix model formalism, we have to go back to string theory and consider
an explanation such as the one given in [3].
(x)
Now let us start our proof. Again, let k be the degree of tree level superpotential Wtree
and n be the number of U (1) at low energy. To realize this situation, we need to consider
that (k n) branch cuts on the Riemann surface should be closed, which corresponds to
Si = 0. Here, there is one important thing: as we know from the expansion of Weff in
terms of (e.g., see (4.13)), we cannot obtain any solutions with Si = 0 if we assume
that Si is dynamical and solve its equation of motion. Therefore to realize the situation with
vanishing Si , we must put Si = 0 at the off-shell level by hand. With this in mind, let us
study dual geometry which corresponds to the gauge theories above. In the field theory,
we assumed that the Riemann surface had (k n) closed branch cuts. Thus, in this dual
geometry analysis we must assume that at off-shell level, (k n) Si s must be zero. For
convenience, we assume that first n Si s are nonzero and the remaining (k n) vanish,
Si = 0,
i = 1, . . . , n,
Si = 0,
i = n + 1, . . . , k.
Therefore the Riemann surface can be written as

y 2 = F2n (x)Q2kn = W (x)2 + bk1 x k1 + .
The effective superpotential in dual geometry corresponding to U (Nc ) gauge theory with
Nf flavors was given in [1] (see also (2.12)) and in the classical limit it behaves as25
Weff |cl =
n
Nc,i W (ai ).
(B.9)
i=1
As discussed in the previous section, existence of flavors does not change the Riemann
surface y(x). In other words, Riemann surface is not singular at xI (roots of B(x)),

y(x) dx = 0, y(x) = W (x)2 + bk1 x k1 + .
xI
25 Remember that in this appendix we are assuming only Coulomb branch. For the Higgs branch, see (7.11)
and (7.12) in [1].
48
Therefore as in [36], by deforming contours of all Si s and evaluating the residue at infinity
on the first sheet, we obtain the following relation,
n

i=1
Si =
k

i=1
1
Si = bk1 ,
4

where we used ki=n+1 Si = 0.
With this relation in mind, next we consider the variation of Weff with respect to Si :
Weff (Si , )
(B.10)
= 0, i = 1, . . . , n.
Si
Solving these equations, we obtain the expectation values, Si . Of course, these vacuum expectation values depend on , gr and zI . Thus when we evaluate the variation
of Weff (Si , ) with respect to , we have to pay attention to implicit dependence on .
However the implicit dependence does not contribute because of the equation of motion
(B.10):
dWeff (Si , ) Si Weff (Si , ) Weff (Si , )
=
+
d
Si
i=1
Weff (Si , )
.
On the other hand, explicit dependence on can be easily obtained by monodromy analysis. Here let us recall the fact that the presence of fundamentals does not change the
Riemann surface. In fact, looking at (2.12) we can read off the dependence from the term
=
2Nc Nf
2i0 = log( BL 2Nc L ),

0
dWeff (Si , )
bk1
.
=S =
4
d log 2Nc Nf
(B.11)
To finish our proof, we have to pay attention to fk1 (x), on-shell. Namely putting Si
into fk1 (x) what kind of property does it have? To see it, let us consider change of variables from Si s to bi s. As discussed in [9] the Jacobian of the change is nonsingular if
0 j k 2,

1
xj
Si
=
dx
.
bj
8i
W (x)2 + f (x)
Ai
In our present case, since only

n of k Si s are dynamical variable, we use bi , i = 0, . . . ,
n 1 in a function fk1 (x) = bi x i as new variables, instead of Si s. As discussed in
[1,32,36], by using Abels theorem, the equation of motions for bi s is interpreted as an
existence condition of a meromorphic function that has an Nc th order pole at infinity on
the first sheet and an (Nc Nf )th order zero at infinity on the second sheet of and a
first order zero at qI . For a theory with Nf 2Nc , such a function can be constructed as
follows [1,35]:

(x) = PNc (x) + P 2 (x) 42Nc Nf B(x).

Nc
49
For this function to be single valued on the matrix model curve y(x), the following condition must be satisfied,
PN2 c (x) 42Nc Nf B(x)

= F2n (x)HN2 c n (x),
W (x)2 + f (x) = F2n (x)Q2kn (x).
This is exactly the same as the factorization form we already see in the field theory analysis.
Therefore the value bk1 of on-shell matrix model curve in dual theory is the same one for
field theory analysis. Comparing two results, (B.7) and (B.8) with corresponding results
for the dual geometry analysis, (B.9) and (B.11) we have shown the equivalence between
these two descriptions of effective superpotentials.
Appendix C. Computation of superpotentialgauge theory side

In this appendix, we demonstrate the factorization method used in Section 4.1 to compute the low energy superpotential, taking the Nf = 4 case as an example. Therefore there
are two kinds of solutions for the factorization problem (4.5) and (4.6).

The breaking pattern U
(2) U (1)
The first kind of solution for the factorization problem is given by
A = 0,
B = 2a,
D = 0,
c = 0,
C = a 2 42 ,
b = 0.
In the classical limit 0, we can see the characteristic function goes as P3 (x) x 2 (x +

a), which means that the breaking pattern is U
(2) U (1). Note that since we are assuming
mf = 0, the notation " should be used for the gauge group that corresponds to the cut
near the critical point at x = 0. Inserting these solutions into (4.7) we obtain one constraint,
2 = a 2 + 82 .
We can easily represent a as a Taylor expansion of :
a = 1 + 4T + 8T 2 + 32T 3 + 160T 4 + 896T 5 + 5376T 5 + ,
where we put = 1 and defined T 2 .26 The coefficients of P3 (x) are related to the
Casimirs Uj = j1 Tr[ j ] as follows. For Nc = 3, Nf = 4, (A.1) reads
!
!
4

Uj
Uj
3
5x
P3 (x) = x exp
+ 3 exp
xj
x
xj
j =1
j =1
26 If we take care of a phase factor of , we will obtain the effective superpotentials corresponding to each
vacuum. However in our present calculation, we want to check whether the effective superpotentials of two
method, field theory and dual geometry, agree with each other. Therefore, we have only to pay attention to the
coefficients in Wlow , neglecting the phase factor.
50

U12
2
+ x
= x U1 x + U2 +
2

U3
+ U3 + U1 U2 1 2 U1 + .
6
3
Comparing the coefficients, we obtain

a2
a3
+ 2 ,
a2 .
U3 = c + ab
2
3
Furthermore, one can compute a1,2 from (4.7). Plugging all these into (4.9), we finally
obtain
U1 = a,
U2 = b +
Wlow = Wcl + T 6T 2
T 2 ,
40T 3
4928T 6
56T 4 288T 5
+ ,
3
3
1
Wcl = .
6

The breaking pattern U
(1) U (2)
The other kind of solution for the factorization problem is given by
1
A = (a 2),
2
D = 0,
c = 0,
B = a 2,
1
C = (a + 2)2 ,
4
1
b = (a + 2)2 ,
4

where 1. These solutions correspond to the breaking pattern U
(1) U (2) in the
classical limit. Inserting these solutions into (4.7) we obtain one constraint,

1 2
a 20a + 42 .
4
Again, let us represent a as a Taylor series of :
2 =
a = 2 + 10T 24T 2 + 144T 4 1728T 6 + ,

where we put = 1, = 1 and defined T . Doing the same way as previous breaking
pattern, we can compute the effective superpotential as
Wlow = Wcl + 2T 13T 2 +
176T 3
138T 4 + 792T 6 + ,
3
1
Wcl = .
3
The other cases with Nf = 1, 2, 3 and 5 can be done analogously.
T ,
References
[1] F. Cachazo, N. Seiberg, E. Witten, Chiral rings and phases of supersymmetric gauge theories, JHEP 0304
(2003) 018, hep-th/0303207.
51
[2] R. Dijkgraaf, C. Vafa, Matrix models, topological strings, and supersymmetric gauge theories, Nucl. Phys.
B 644 (2002) 3, hep-th/0206255;
R. Dijkgraaf, C. Vafa, On geometry and matrix models, Nucl. Phys. B 644 (2002) 21, hep-th/0207106;
R. Dijkgraaf, C. Vafa, A perturbative window into non-perturbative physics, hep-th/0208048.
[3] K. Intriligator, P. Kraus, A.V. Ryzhov, M. Shigemori, C. Vafa, On low rank classical groups in string theory,
gauge theory and matrix models, Nucl. Phys. B 682 (2004) 45, hep-th/0311181.
[4] P. Kraus, M. Shigemori, On the matter of the DijkgraafVafa conjecture, JHEP 0304 (2003) 052, hepth/0303104.
[5] R. Argurio, G. Ferretti, R. Heise, An introduction to supersymmetric gauge theories and matrix models,
hep-th/0311066.
[6] N. Seiberg, Adding fundamental matter to Chiral rings and anomalies in supersymmetric gauge theory,
JHEP 0301 (2003) 061, hep-th/0212225.
[7] P. Svrcek, Chiral rings, vacua and gaugino condensation of supersymmetric gauge theories, hep-th/0308037.
[8] F. Cachazo, N. Seiberg, E. Witten, Phases of N = 1 supersymmetric gauge theories and matrices, JHEP 0302
(2003) 042, hep-th/0301006.
[9] F. Cachazo, M.R. Douglas, N. Seiberg, E. Witten, Chiral rings and anomalies in supersymmetric gauge
theory, JHEP 0212 (2002) 071, hep-th/0211170.
[10] P.C. Argyres, M.R. Plesser, N. Seiberg, The moduli space of N = 2 SUSY QCD and duality in N = 1 SUSY
QCD, Nucl. Phys. B 471 (1996) 159, hep-th/9603042.
[11] G. Carlino, K. Konishi, H. Murayama, Dynamical symmetry breaking in supersymmetric SU(n(c)) and
USp(2n(c)) gauge theories, Nucl. Phys. B 590 (2000) 37, hep-th/0005076.
[12] G. Carlino, K. Konishi, S.P. Kumar, H. Murayama, Vacuum structure and flavor symmetry breaking in
supersymmetric SO(n(c)) gauge theories, Nucl. Phys. B 608 (2001) 51, hep-th/0104064.
[13] V. Balasubramanian, B. Feng, M.X. Huang, A. Naqvi, Phases of N = 1 supersymmetric gauge theories with
flavors, Ann. Phys. 310 (2004) 375, hep-th/0303065.
[14] C. Ahn, B. Feng, Y. Ookouchi, Phases of N = 1 SO(N (c)) gauge theories with flavors, Nucl. Phys. B 675
(2003) 3, hep-th/0306068.
[15] C. Ahn, B. Feng, Y. Ookouchi, Phases of N = 1 USp(2N (c)) gauge theories with flavors, Phys. Rev. D 69
(2004) 026006, hep-th/0307190.
[16] Y. Demasure, R.A. Janik, Explicit factorization of SeibergWitten curves with matter from random matrix
models, Nucl. Phys. B 661 (2003) 153, hep-th/0212212.
[17] P. Kraus, A.V. Ryzhov, M. Shigemori, Loop equations, matrix models, and N = 1 supersymmetric gauge
theories, JHEP 0305 (2003) 059, hep-th/0304138.
[18] L.F. Alday, M. Cirafici, Effective superpotentials via Konishi anomaly, JHEP 0305 (2003) 041, hepth/0304119.
[19] M. Aganagic, K. Intriligator, C. Vafa, N.P. Warner, The glueball superpotential, hep-th/0304271.
[20] F. Cachazo, Notes on supersymmetric Sp(N ) theories with an antisymmetric tensor, hep-th/0307063.
[21] M. Matone, The affine connection of supersymmetric SO(N )/Sp(N ) theories, JHEP 0310 (2003) 068, hepth/0307285.
[22] K. Landsteiner, C.I. Lazaroiu, On Sp(0) factors and orientifolds, Phys. Lett. B 588 (2004) 210, hepth/0310111.
[23] S.G. Naculich, H.J. Schnitzer, N. Wyllard, Matrix-model description of N = 2 gauge theories with nonhyperelliptic SeibergWitten curves, Nucl. Phys. B 674 (2003) 37, hep-th/0305263.
[24] M. Gomez-Reino, S.G. Naculich, H.J. Schnitzer, Improved matrix-model calculation of the N = 2 prepotential, JHEP 0404 (2004) 033, hep-th/0403129.
[25] A. Strominger, Massless black holes and conifolds in string theory, Nucl. Phys. B 451 (1995) 96, hepth/9504090.
[26] K.A. Intriligator, N. Seiberg, Phases of N = 1 supersymmetric gauge theories in four-dimensions, Nucl.
Phys. B 431 (1994) 551, hep-th/9408155.
[27] S. Elitzur, A. Forge, A. Giveon, K.A. Intriligator, E. Rabinovici, Massless monopoles via confining phase
superpotentials, Phys. Lett. B 379 (1996) 121, hep-th/9603051.
[28] Y. Demasure, AffleckDineSeiberg from SeibergWitten, hep-th/0307082.
[29] J. Erlich, S. Hong, M. Unsal, Matrix models, monopoles and modified moduli, hep-th/0312054.
52
[30] N. Seiberg, Electricmagnetic duality in supersymmetric non-Abelian gauge theories, Nucl. Phys. B 435
(1995) 129, hep-th/9411149.
[31] F. Cachazo, K.A. Intriligator, C. Vafa, A large N duality via a geometric transition, Nucl. Phys. B 603 (2001)
3, hep-th/0103067.
[32] Y. Ookouchi, N = 1 gauge theory with flavor from fluxes, JHEP 0401 (2004) 014, hep-th/0211287.
[33] E. Witten, Baryons and branes in anti-de Sitter space, JHEP 9807 (1998) 006, hep-th/9805112.
[34] C. Ahn, Y. Ookouchi, The matrix model curve near the singularities, hep-th/0309156.
[35] S.G. Naculich, H.J. Schnitzer, N. Wyllard, Matrix model approach to the N = 2 U (N ) gauge theory with
matter in the fundamental representation, JHEP 0301 (2003) 015, hep-th/0211254.
[36] F. Cachazo, C. Vafa, N = 1 and N = 2 geometry from fluxes, hep-th/0206017.
Gravitational corrections for supersymmetric gauge

theories with flavors via matrix models
Hiroyuki Fuji, Shunya Mizoguchi
Theory Division, Institute of Particle and Nuclear Studies,
High Energy Accelerator Research Organization (KEK), Tsukuba, Ibaraki 305-0801, Japan
Received 2 June 2004; accepted 27 July 2004
Abstract
We study the gravitational corrections to the F-term in four-dimensional N = 1 U (N) gauge
theories with flavors, using the DijkgraafVafa theory. We derive a compact formula for the annulus
contribution in terms of the prime form on the matrix model curve. Remarkably, the full Tr R R
correction can be reproduced as a special momentum sector of a single c = 1 CFT correlator, which
closely resembles that in the bosonization of fermions on Riemann surfaces. The N = 2 limit of the
torus contribution agrees with the multi-instanton calculations as well as the topological A-model
result. The planar contributions, on the other hand, have no counterpart in the topological gauge
theories, and we speculate about the origin of these terms.
PACS: 02.10.s; 11.15.-q; 12.60.J
1. Introduction
Since the discovery by Dijkgraaf and Vafa, we have recognized that non-perturbative aspects of four-dimensional supersymmetric gauge theories can be studied via matrix models.
In this framework the effective superpotential for N = 1 supersymmetric gauge theories
can be determined as the large N free energy of a matrix model [1], and by minimizing it
the non-perturbative vacua and their phase structures can be investigated [2]. This part of
E-mail addresses: hfuji@post.kek.jp (H. Fuji), mizoguch@post.kek.jp (S. Mizoguchi).
doi:10.1016/j.nuclphysb.2004.07.039
54
H. Fuji, S. Mizoguchi / Nuclear Physics B 698 (2004) 5391
the proposal has been elegantly proven by using the chiral ring relations of N = 1 supersymmetry in the generalized Konishi anomaly equations [3,4].
On the other hand, the non-planar diagrams have been shown to correspond to the gravitational corrections to the gauge theory [510], in particular, the genus one free energy of
the matrix model computes the gravitational F-term

1
d 4 x F1 (S) Tr R R.
(1.1)
16 2
Recently, there has been much progress in the analysis of N = 2 gravitational F-terms
in terms of the multi-instanton calculations [11] and geometric engineering [1214]. To
compare them with the matrix model results, it is essential to know the precise form of the
gravitational F-terms computed using the matrix model. For the pure gauge theory case,
it has been checked that the torus free energy F1 coincides with the topological partition
function [15] of the DonaldsonWitten theory, and hence the N = 2 gravitational F-terms
[5,6]. The planar free energy is also known to contribute to the gravitational corrections
[16].
This framework can be extended to N = 1 supersymmetric gauge theories with Nf
flavors in the fundamental representation [1719]. The dual matrix model consists of a
bosonic matrix and vectors which correspond to the adjoint scalar and Nf flavor fields
q I , qI (I = 1, . . . , Nf ) of the gauge theory, respectively. In the vector coupled matrix
model, there are two kinds loops with N and Nf indices; the former are summed up but
the latter remain as boundaries of Feynman diagrams. It looks as if the model has not only
the closed string but also the open string sectors [20]. In order to evaluate R 2 terms for this
gauge theory, it is necessary to consider both the torus and annulus contributions.
In this paper, we evaluate the R 2 correction terms for N = 1 U (N) gauge theories
with Nf flavors. Using the ordinary large N analysis [21], we derive a compact formula
for the annulus contribution to the correction in terms of the prime form on the matrix
model curve. Remarkably, the full Tr R R correction containing the torus as well as all
the planar contributions is reproduced as a special momentum sector of a single c = 1
CFT correlator, which closely resembles that in the bosonization of fermions on Riemann
surfaces. It is in accord with the recent observation in topological string theory that the
non-compact B-branes correspond to fermions in a chiral boson theory on a hyper-elliptic
curve [22,23].
The plan of this paper is as follows: in Section 2 we briefly review the Dijkgraaf
Vafa theory including a string theory derivation of gravitational corrections. In Section 3
we compute the planar gravitational corrections for the N = 1 U (N) gauge theories. In
particular, we derive in Section 3.4 a compact formula for the annulus contributions in
terms of the prime form on the matrix model curve. Section 4 is devoted to some examples,
in which our formula is confirmed explicitly. In Section 5 we show that the full Tr R R
gravitational correction, including both the torus and planar contributions, is reproduced
as a single chiral correlator of a c = 1 conformal field theory, which closely resembles that
in the bosonization of fermions on Riemann surfaces. In Section 6 we consider the N = 2
limit, and speculate about the origin of the planar contributions. Finally, we conclude our
paper with a summary and outlook on future work in Section 7. The Appendices AC
contain some technical details which we need in the text.
55
2. Gravitational F-terms from matrix models

2.1. The DijkgraafVafa theory with flavors
The original proposal of Dijkgraaf and Vafa [1] was summarized as follows:
1. The low energy effective superpotential of N = 1 gauge theories can be computed
by summing over the planar diagrams of the matrix model with the same tree-level
potential.
2. The non-planar diagrams compute the gravitational F-terms for these theories.
The first statement was proven in [3], and was generalized to the cases with flavors in [18].
The latter part was also supported by many arguments [8,9], and was explicitly confirmed
in the pure gauge theory cases [5,6].
Let us consider the N = 1 U (N) gauge theory coupled to matter superfields qI , qI
(I = 1, . . . , Nf ) in the (anti-)fundamental representation with the superpotential [17]
V (, qI ) = tr W () +
Nf
qI ( mI )qI ,
I =1
W () =
n+1

gp
p=1
p ,
(2.1)
is the adjoint chiral superfield. In the classical vacuum the gauge group is broken as
U (N)
n

i=1
U (Ni ),
n
Ni = N.
(2.2)
i=1
The claim of [1] is that the non-perturbative vacuum structure of this theory can be analyzed
by a vector coupled matrix model with action
Smatrix (, QI , Q I ) = TrN N W () +
Nf
I ( mI )QI ,
Q
(2.3)
I =1
where is an N N Hermitian matrix and QI , Q I are N -component vectors. The vectormatrix coupling Q I QI leads to Feynman diagrams with boundaries in various topologies
(see Fig. 1).
1
In the low energy effective theory, the glueball superfields Si = 32

2 trSU(Ni ) W W
(i = 1, . . . , n) play the role of the fundamental fields. According to the proposal, Si is
Fig. 1. Feynman diagrams with boundaries.
56
identified with the t Hooft coupling gs N i for each i, where gs is the matrix model coupling constant and N i is the number of eigenvalues distributed on the ith cut. Under this
identification, the effective superpotential for the gauge theory is given by

n

F0(0)(Si )
(1)
+ Si + F0 (Si ),
Ni
Weff (Si ) =
(2.4)
Si
i=1
and F0 (Si ) are given by the large N sphere and disk free energies of the
where
matrix model, respectively, is the bare coupling.1 By extremizing this superpotential, one
can analyze the non-perturbative vacuum structure of the gauge theory [2].
(0)
F0 (Si )
(1)
2.2. String theory derivation of gravitational F-terms

To extend the analysis to the study of the gravitational F-terms, it is useful to consider
how they arise in string theory. The gauge theory setup above can be realized in type IIB
string theory on a CalabiYau three-fold [24,25]
W (z)2 + y 2 + v 2 + w2 = 0,
(z, y, v, w) C4 ,
(2.5)
with n singular points. By blowing them up, we obtain a smooth geometry with exceptional
2-cycles Ci (i = 1, . . . , n). We then consider N D5-branes wrapped around Ci s and Nf
D5-branes around non-compact 2-cycles CI (I = 1, . . . , Nf ). The N = 1 gauge theory
above is realized on the spacetime filling world-volume of the D5-branes.
To evaluate the F-term corrections it is advantageous to utilize the hybrid formalism
[7,26]. The string Lagrangian density for the four-dimensional part is given by
1
+ p
+ p
+ p + p ,
L = X X
(2.6)
2
where
p, p,
and are the fermionic fields. Inserting two gluino vertex operators

p on the boundaries of the world-sheet, the stringy corrections lead to the F-terms
W
containing the glueball superfields.
Fig. 2. The brane configuration in the resolved geometry.
1 The definition of W (S ) (2.4) has an ambiguity in the linear terms in S depending on the choices of the
i
eff i
integration constants in the free energy (see, e.g., [18]).
57
The string world-sheet can end either on N compact or Nf non-compact branes in the
CalabiYau direction. Resorting to the fermion zero-mode arguments [27,28], we can conclude that only the planar string world-sheets with at most one boundary ending on the
non-compact branes can contribute to the effective superpotential [1]. A simple combinatorial argument then yields the effective superpotential (2.4) where the sphere and disk free
energies of string theory are identified with those of the matrix model.
The gravitational F-term corrections are given by inserting RR vertex operators on the
bulk of the string world-sheet. A candidate for such an operator is the self-dual gravitino
vertex operator [7,8]

+ p X X )
+ G p p ( ).
G (p X X
(2.7)
Due to the chiral ring relations
{W , W } 2G W ,
G4 0
(mod D),
(2.8)
at most two G insertions can contribute to the gravitational F-term corrections. Such G2
terms are given by the torus (Fig. 3) as well as the planar world-sheet corrections (Fig. 4).
In particular, there are three types of planar world-sheets depending on which (compact or
non-compact) branes the string ends on.
Fig. 3. Gravitational corrections from the torus.
Fig. 4. Three types of planar string world-sheets.
58
The gravitational F-term containing G2 takes the form

d 4 x d 2 F=0 (Si )G2 ,
(2.9)
where F=0 (Si ) (= the planar + torus gravitational corrections) is the contribution from
the = 0 string world-sheets.2 If we further consider a self-dual graviphoton background
F , then we also have gravitational F-terms from the graviphoton fields. Such F-terms
are obtained by decomposing N = 2 gravitational F-terms into the N = 1 F-terms. It
was shown in [7] that the superstring computation of these Feynman diagrams is identical
to that in the field theory limit, and the combinatorial factors can be calculated in the
associated matrix model.
In the following sections, we will derive a formula for the free energy F=0 (Si ).
3. Planar gravitational corrections

3.1. Gravitational corrections from sphere and disk diagrams
There are three kinds of planar diagrams which contribute to the Tr R R correction
(Fig. 4). The first and second types of contributions are determined by the sphere and the
disk free energies through the following combinatorial arguments: for the unbroken gauge
group case (one-cut case), the coefficient of planar gravitational corrections yields
(0)
2 h2
h C2 F0,h (trSU(N) 1) S
h=1
(0)
N 2 2 F0 (S)
=
,
2
S 2
(0)
F0 (S) =
(0)
F0,h S h ,
(3.1)
h=1
for the first type, and
(1)
(0)
2 h1
h C1 Fh (trSU(N) 1) S
h=1
F (S)
,
=N 0
S
F0(1)(S) =
(1) h
F0,h
S ,
(3.2)
h=1
(0)
(1)
and F0,h
are the symmetric factors of Feynman diagrams with
for the second type. F0,h
h boundaries for the sphere and disk topologies, respectively. Generalizing to the cases of
arbitrary breaking patterns, we have
n
2 F0(0)
1
Ni Nj
2
Si Sj
(for a sphere with 2 index loops),
(3.3)
(for a disk with 1 index loop).
(3.4)
i,j =1
n

i=1
Ni
F0(1)
Si
The third type of gravitational corrections come from annulus diagrams, which we will
compute below using the familiar large N technique in matrix models [21].
2 Here only the empty loops are understood as boundaries in the definition of the Euler number .
59
3.2. Planar diagrams with boundaries

Let be an N N Hermitian matrix, and QI (I = 1, . . . , Nf ) be N -dimensional
complex vectors. We consider the partition function
Z=
vol U (N )

Nf

dQI d Q I
I =1
Nf

1
I QI
Q I QI mI Q
exp
Tr W () +
gs
(3.5)
I =1
with a polynomial tree level potential W (). It allows a topological expansion [29]

Z = exp gs2 F ,
F=
2g+k
gs
Fg(k) ,
(3.6)
g=0 k=0
(k)
where Fg is the contributions from the connected genus g diagrams with k boundaries.
(0)
F0 also includes the non-perturbative piece coming from the gauge volume [30].
(2)
I integrations first.
To evaluate the annulus amplitude F0 , we carry out the dQI d Q
This yields

Nf

1
log( mI )
Z d exp Tr W () + gs
(3.7)
gs
I =1
up to an irrelevant multiplicative constant which is independent of , W () or mI s. These

integrations organize a resummation of diagrams. The gs factor of the log potential indi loop at each log vertex. Thus, instead of Z, we consider
cates that there is a matter (= QQ)
the following double expansion

Nf

1
log( mI )
Z(gs ,
) := d exp Tr W () +
gs
I =1

:= exp gs2 F (gs ,
) ,
(3.8)
F (gs ,
) =
2g
gs Fg (
),
(3.9)
g=0
Fg (
) =
k Fg(k)
(3.10)
k=0
(0)
(2)
and set
= gs . Then F1 and F0 are of the same gs2 order in the ordinary gs expansion
of the free energy; the former is the torus, whereas the latter is the annulus amplitude.
(2)
One of the aims of this paper is to determine the precise form of F0 for general n and
for a general polynomial potential W (z), in terms of the language of Riemann surfaces that
the matrix model defines. F0(k) (k = 1, 2, . . .) can be extracted from Z(gs ,
) as follows:
60
writing the expectation value of any function f () of with respect to the measure (3.8)
as f () gs ,
, we have

k F (gs ,
)
1 k F0 (
)
1
lim
=
F0(k) =

k!
k
=0 k! gs 0
=0
Nf

k1
1

g
Tr
lim
=
log(
m
)
(3.11)
.
s
I
k1

k! gs 0
I =1
gs ,
=0
Keeping S := gs N fixed, gs 0 implies N . In this limit the expectation value can

be written as

1
Tr f () gs ,
=
lim
gs 0 N

d (,
)f (),
(3.12)
for any f (), where (,

) is the eigenvalue density function of normalized as

d (,
) = 1.
(3.13)
Therefore, expanding (,
) as
(,
) =
k (k) (),
(3.14)
k=0
we find
(k)
F0
Nf

1 k1

=
d
(,
)
log(
m
)
S

I

k!
k1
I =1
S
k
=0
Nf
d (k1) ()
log( mI ),
(3.15)
I =1
in particular, the disk [17,18]

(1)
F0

= S
d (0) ()
Nf
log( mI )
(3.16)
I =1
and the annulus

F0(2)
(k1) ()
S
=
2

d
(1)
()
Nf
log( mI ).
(3.17)
I =1
can be computed by means of the standard large N technique for evaluating the
planar diagrams [21].
61
To be rigorous, one would need to show that the gs 0 limit and the operation of /
commute. In Appendix A we will give a different derivation of the formula (3.17) for an
arbitrary polynomial potential W () using Riemanns bilinear identity, thereby providing
an alternative proof of it.
3.3. The large N technique
We will now briefly review the large N (= N ) technique developed in [21] to fix our
notation (see also [31]). As we introduced above, (, gs ) is the density function of the
eigenvalue of normalized as (3.13). The non-zero support of (, gs ) (as a function
of ) consists of several disjoint intervals on the real axis. These intervals are the branch
cuts of the resolvent (z, gs ) defined by

(z, gs ) :=
(, gs )
z
(3.18)
on the complex z-plane. (3.18) is equivalent to

( + i0, gs ) ( i0, gs ) = 2i(, gs ),
(3.19)
in the boundary value representation. If the number of cuts is n, (z, gs ) dz is a meromorphic differential on a hyper-elliptic curve of genus n 1 with a parameter gs .
gs
Let n1
be a family of genus n 1 hyper-elliptic curves defined by the equation
y2 =
2n

z ai (gs )
(i = 1, . . . , 2n)
(3.20)
i=1

2n
with a parameter gs . We denote by
i=1 (z ai (gs )) a branch of y such that y > 0 when
z R, z . We call it the first sheet, and the other the second sheet.
The n cut solution to the large N saddle point equation

Nf
2n
gs
n
W (w) + I =1
i=1 (z ai (gs ))
wmI

dw
(z, gs ) =
(3.21)
4iS
2n
j =1A
(z
w)
(w
a
(g
))
i
s
i=1
j
g
s
. The contour surrounds all the cuts but not
defines a meromorphic differential on n1
the points w = z, mI (I = 1, . . . , Nf ). The definitions of the contours are summarized in
Fig. 5. The positions of the 2n branch points ai (gs ) (i = 1, . . . , 2n) are determined by the
2n conditions

Nf
Nf
gs

wk (W (w) + I =1
wmI )
= 4iSk,n (k = 0, . . . , n) (3.22)
dw
2n
(w
a
(g
))
I
=1
i
s
i=1
A
CI
62
Fig. 5. The contours. Those on the second sheet are shown in broken lines. A is defined to go along |z| = |0 |
counter-clockwise on the z-plane.
and

dw (w, gs ) = 2i
Sj
S
(j = 1, . . . , n).
(3.23)
Aj
The first n + 1 conditions (3.22) arise from the requirement that (z, gs ) behaves like 1/z
as z on the first sheet, whereas the latter are the conditions for a given set of filling
fractions. Only n 1 of (3.23) are independent.
The large N free energy for the potential
W (, gs ) := W () + gs
Nf

I =1
log( mI )
(3.24)
63
is given by

F0 (gs ) = S
d (, gs )W (, gs )

+S

d
d (, gs )( , gs ) log
| |
.
(3.25)
is an arbitrary integration constant, which may be regarded as the physical scale parameter of the corresponding gauge theory up to a potential-dependent constant.
3.4. Annulus diagrams
We would now like to write (3.17) in terms of the language of Riemann surfaces.
(, gs ) is expanded as
(z, gs ) =
gsk (k) (z).
(3.26)
k=0
The differential (1) (z) dz must have zero A-periods because the right-hand side of (3.23)
does not depend on gs .
Let us analyze the singularities of the differential (1) (z) dz. Since the discontinuity of
(z, gs ) is the eigenvalue density, it satisfies
n
1
dz (z, gs ) = 1.
(3.27)
2i
j =1A
(z, gs ) also satisfies

1
dz (z, gs ) = 1,
2i
(3.28)
because (z, gs ) goes to 1/z at infinity on the first sheet. Eqs. (3.27) and (3.28) imply
that (z, gs ) dz has a pole of first order at z = on the first sheet with residue 1,
but otherwise it is regular everywhere else on the first sheet. Therefore, we conclude that
(1) (z) dz is regular everywhere on the first sheet. On the other hand, (z, gs ) can be
written in the form

gs
1

(z, gs ) =
y(z, gs ) ,
W (z) +
2S
z mI
I

Nf
2n

i=1 (z ai (gs ))
y(z, gs ) =
dw
2i
W (w) +
(z w)

Nf
I =1C
gs
I =1 wmI
2n
i=1 (w ai (gs ))
(3.29)
64
We similarly expand
y(z, gs ) =
gsk y (k)(z).
(3.30)
k=0
Note that the difference of the contours of y(z, gs ) and (z, gs ); thanks to this change, the
contour integral of y(z, gs ) gives some rational function of z.3 This implies that y(z, gs )
changes its sign under the interchange of the first and second sheets.
Since the second term of (z, gs ) has a first order singularity at each z = mI (I =
g N
gs
and at z = with residue s2Sf on both sheets, the y(z, gs )
1, . . . , Nf ) with residue 2S
term must cancel them on the first sheet, and hence in turn doubles the residues on the second sheet. To summarize, denoting coordinates on the second sheet with tildes, (1) (z) dz
has first order singularities at
z=m
I with residue 1/S for all I = 1, . . . , Nf , and
with residue Nf /S,
z=
g
s
otherwise holomorphic everywhere else on n1
. Thus we have shown that (1) (z) dz is
an Abelian differential of the third kind with zero A-periods and is given by
(1)
Nf
1
(z) dz =
m I
,
S
(3.31)
I =1
where, following the notation used in [32], ab (z) denotes a zero A-period Abelian differential which has simple poles at z = a, b with residue 1, respectively.
One of the nice things for this kind of differentials is that their integrals are compactly
gs
described in terms of the prime form E(z, w) on n1
. The prime form E(z, w) is known
to be the unique bi-holomorphic half differential on a compact Riemann surface such that
E(z, w) = 0 if and only if z = w. One of the basic properties of the prime form is [32]
ab (z) = dz log
E(z, a)
.
E(z, b)
(3.32)
Using this formula, we finally obtain the third type of contributions to the gravitational
correction4
(2)
F0 =
Nf
Nf
n Nf

1
1
m J
(z)
log(z
m
)
=
m J
I
(z)
4i
2
j =1A
J =1
I =1
I,J =1mI
3 If g = 0, it is a polynomial of z. If, in particular, n saturates the degree of W (z) (the maximally broken
s
case), it is a constant up to which y (0) (z) coincides with y(z) (3.20).
(2)
4 Of course, the computation of the annulus amplitude F

0 itself can also be (and has been) done by other
means (see, e.g., [33,34]). What is new here is that we have expressed it in terms of a special form (that is, the
prime form) on the Riemann surface, and we cannot compare it with the CFT result until we do so. We thank
H. Kawai for discussions on this point.

Nf
m
J)
1
E(mI , )E(,
.
=
log
2
E(mI , m
J )E(, )
65
(3.33)
I,J =1
The full planar gravitational correction F (pl) is thus

F (pl) =
n
(0)
n
(1)

2 F0
F
1
Ni Nj
+
Ni 0 + F0(2)
2
Si Sj
Si
i,j =1
with
(3.34)
i=1
n

n

2 F0(0)
1
0
Ni Nj
= lim
Ni Nj i j (z) + log
, (3.35)
2
Si Sj |0 |
i,j =1
n

i=1
Ni
i,j =1
(1)
n

F0
=
Ni
Si
i,k=1
lim
|0 |
B i
Nf
i (z)
log(z mI )
I =1
Ak
n

i=1
Ni
Nf

0
2i i (z) log(0 )
I =1
(3.36)
mI
and F0(2) given in (3.33). In deriving the above, we have used the special geometry relation
[1,18]

F0(0)
0
1
(0)
(3.37)
= lim
dz y (z) W (0 ) + 2S log
,
|0 | 2
Si
B i
as well as the variation formula

y (0) (z)
dz = 4i i (z)
Si
(i = 1, . . . , n),
(3.38)
where i (i = 1, . . . , n) are related with the holomorphic differentials i (i = 1, . . . , n 1)

gs =0
through the relations
on n1
i = i
2i
(i = 1, . . . , n 1),
1
.
2i
They are normalized so that

j (z) = ij (i, j = 1, . . . , n 1),
n =
(3.39)
Ai
j (z) = ij
(i, j = 1, . . . , n).
Ai
See Appendix B for details.
(3.40)
66
4. Examples
4.1. One-cut solutions for quadratic potential with Nf = 1
The first example is the conifold case
W () = t 2
(4.1)
with a single matter field. The resolvent (z, gs ) is

1
gs
(z, gs ) =
2tz +
2S
zm

+ (z c(gs ))2 (gs )2 2t
gs
zm
(m c(gs ))2

(gs )2
(4.2)
where c(gs ) and (gs ) are related with the positions of the end points by
a1 (gs ) = c(gs ) + (gs ),
a2 (gs ) = c(gs ) (gs )
(4.3)
and satisfy (3.22) and (3.23)

0 = 2tc(gs ) +
gs
,
(m c(gs ))2 (gs )2

m c(gs )
2
.
2S = t(gs ) + gs 1
(m c(gs ))2 (gs )2
(4.4)
(4.5)
They are solved as a power series

m2 2 + 2m 2
c(gs ) =
gs +
gs + O gs3 ,
2
2
2
2
2t (m )
2t m2 2

m 2 2 m
(gs ) = +

gs
2t m2 2

(m2 22 m m2 2 )2 + 54 2
+
gs + O gs3
2
3
2
2
2
4t (m )
1
(4.6)
(4.7)
with := 2S/t. The integral for the genus zero free energy F0 (3.25) can be completely performed in this case and is given in Appendix C. F0(1) (the disk amplitude) and
(2)
F0 (the annulus amplitude) are obtained as coefficients of the Taylor series of F0 (gs ).
After some calculations using (4.6) and (4.7) we find

m
m + m 2 2
1
(1)

F0 = S log
(4.8)
+
,
2
m + m 2 2 2

m + m 2 2
1
2S
(2)
.
=
F0 = log
(4.9)
2
t
2 m 2 2
67
(1)
F0 (4.8) agrees with the calculations in [19].

Let us show that (4.9) can be written in terms of prime forms. Let w, w C {} = P1
be coordinates of some two points in the natural coordinate system, then the prime form is
simply given by
w w
E(w, w ) =
.
dw dw
(4.10)
P1 is also realized as a two-sheeted Riemann surface with a single branch cut. We take
two end points at z = ( R, > 0), where z is the coordinate of the two-sheeted
Riemann surface. The relation between these two coordinate systems is

z + z 2 2
2
, w=
.
z=w+
(4.11)
4w
2
This map is two to one for generic z, except at the two end points of the cut where it is
one-to-one.
The cut [, ] on the two-sheeted z-plane gets mapped to a circle |w| = /2 on
the w-plane. Defining the first and second sheets as before, the region outside this circle
corresponds to the first sheet, while inside the second sheet. It is easy to see that

2
w 4w
z 2 2

E(z, z ) =
=
,
dw(z) dw(z)
dw(z) dw(z)
z+ z2 2
w0
2
=
=
,
E(z, )
dw(z) dw()
dw(z) dw()
E(, z ) =
2
4w
,
dw() dw(z)
0
=
E(, )
,
dw() dw()
(4.12)
where, as before, we have added tildes to the z coordinates for the points on the second
sheet. Thus one may write (4.9) as
F0(2) =
E(m, )E(,
m)
1
log
.
2
E(m, m)E(,
(4.13)
This agrees with our general formula (3.33).

4.2. Two-cut solutions for a quartic potential with Nf = 2
The next example is an even quartic superpotential with two flavors with masses
m1 = m2 := m,
(4.14)
in which we find a symmetric 2-cut solution. The potential is

W () = t 2 u 4 .
(4.15)
68
Without matter, a large N symmetric solution was obtained long time ago by [35] in terms
of elementary functions. Despite the symmetric potential, asymmetric filling is also possible and the solution is written using elliptic functions in general [36,37].
After integrating out the matter fields, we have

W (, gs ) = t2 u4 + gs log 2 m2 .
With a symmetric ansatz, we find the resolvent
gs
gs
+
2S(z, gs )=2tz 4uz3 +
zm z+m

+ z z2 a 2 z2 b2 4u
2gs

,
(z2 m2 ) (m2 a 2 )(m2 b2 )

(4.16)
where two cuts are created at [b, a] and [a, b]. The density function of eigenvalues
yields

1
2 a 2 b 2 2
(, gs ) =
2S

2gs

4u +
(4.17)
.
(2 m2 ) (m2 a 2 )(m2 b2 )
Being symmetric, the condition (3.23) is automatically satisfied, while the asymptotic
conditions (3.22) yield
gs
0 = 2uc + t +
(4.18)
,
(c M)2 2

M c
,
S = u2 + gs 1
(4.19)
(M c)2 2
where parameters c, and M are defined by

a+b
ba
,
=
,
M = m2 .
2
2
From these conditions we find an iterative solution
gs
t
c=
2u
(t + 2uM)2 + 4uS

2gs2 u 2t + 4uM (t + 2uM)2 + 4uS
+
+ O gs3 ,
2
2
[(t + 2uM) + 4uS]

t + 2uM
S gs
1
1
= +
u
2
uS
(t + 2uM)2 + 4uS

2
1
u
2
t + 2uM
gs2
+

1
+
2
S
(t + 2uM)2 + 4uS 4uS
(t + 2uM)2 + 4uS

2(t + 2uM) 2t + 4uM (t + 2uM)2 + 4uS
+ O gs3 .
+
(4.20)
2
2
((t + 2uM) + 4uS)
c=
69
The multi-cut large N free energy is given in Appendix A, and is in the present case

F0 = S
d (, gs )

|2 20 |
1
S
S
W (, gs ) log
W (0 , gs ),
2
2
2
(4.21)
a
b
where b d (, gs ) = a d (, gs ) = 1/2. Plugging (4.17), (4.20) into (4.21) and
picking up the O(gs2 ) terms, we obtain the free energy of annulus diagrams as before. The
result is

t + 2uM + (t + 2uM)2 + 4uS
1
(2)
F0 = log
(4.22)

.
2
2 (t + 2uM)2 + 4uS
Let us compare this with (3.33). The prime form is defined on the curve

y 2 = z2 a 2 z2 b 2 ,

t
S
t
S
2
2
a = ,
b = + .
2u
u
2u
u
(4.23)
In the present case (3.33) reads

1
E(m, m)E(m,
m)E(m,
m)E(m,
m)
log
2
2
E(m, )
2
E(m, )

1
=
+
(m

+ (m)

).
2
(4.24)
The relevant meromorphic 1-form with zero A-periods is

m

+ (m)

y(m) 1
1
1
1 y(m) 1
z dz
+
.
=
dz
2 y(z) z m
y(z) z + m z m z + m
y(z)
(4.25)
By changing the integration variable z to x = z2 , the above integration reduces to a simple

form5
1
(4.24) =
2

M
y(M) 1
1
1
dx
x M
y(x) x M y(x)

t + 2uM + (t + 2uM)2 + 4uS
1
= log

.
2
2 (t + 2uM)2 + 4uS
This result exactly coincides with the matrix model calculation.
5 Identifying z z, we obtain the same curve and singular points as those in the conifold case.
(4.26)
70
4.3. Two-cut solutions for a cubic potential with Nf = 1

4.3.1. Perturbative computations in gauged matrix models
The final example is the perturbative computations of the two-cut free energy for a cubic
potential [38]

m)Q.
W = g Tr 3 + 2 + Q(
(4.27)
3
2
In this two-cut case we need to consider the fluctuation around the vacuum

a1 1N 1
0
11 12
=
+
,
0
a2 1N 2
21 22
(4.28)
where a1 and a2 are classical roots of W (z) = 0. Around this vacuum the original gauge
symmetry of matrix model
U U 1 ,
U U (N )
(4.29)
reduces to U (N ) U (N 1 ) U (N 2 ), and the matrix model action is given by

1 3
2
2
1 3
Wtree () = g Tr 11 + 11 + g Tr 22 22 .
3
2
3
2
(4.30)
Since the off-diagonal block of the matrix does not appear in the action, a convenient gauge
choice is
ij = 0,
i = j.
(4.31)
In this gauge the coupling to vectors Q, Q is

Wmatter = Q 1 (11 m)Q1 + Q 2 (22 m )Q2 .
(4.32)
The gauge fixing requires the introduction of the ghost matrices B and C with action
Wghost (B, C) = Tr(B21 C12 ) Tr(B12 C21 ) + Tr(B21 11 C12 + C21 11 B12 )
+ Tr(B12 22 C21 + C12 22 B21 ).
(4.33)
From the matrix model action

W (, B, C) = Wtree () + Wghost (B, C),
(4.34)
we can read off the propagators

11 11 =
1
,
g
22 22 =
B12 C21 = B21 C12 =

Q1 Q1 =
1
,
m
1
,
g
1
,
g
Q2 Q2 =
1
.
m+
(4.35)
For 3 and BC vertices we assign weight g, and for QQ

vertex we assign weight 1
(Fig. 6). For a ghost loop, we add an extra factor of 1.
71
Fig. 6. Vertices in the gauged matrix model.
Fig. 7. The dump-bell diagram.
Fig. 8. Three-loop diagrams contributing to S12 or S22 .
Having found the Feynman rules, we can calculate the annulus contributions perturbatively. The annulus diagrams are drawn in Figs. 79 up to three loops. The contributions to
annulus free energy from the dump-bell diagram Fig. 7 is
S1
S2
.
2
2gm
2g(m + )2
(4.36)
From the three-loop diagrams, we obtain the terms involving the S12 , S22 and S1 S2 . The
diagrams in Fig. 8 lead to the S12 and S22 terms. We obtain the S12 terms

1
1
1
1
1
1
2 3 3 2 3 3 + 2 2 4 + 2 2 4 + 2 4 2 + 2 4 2 S12 .
g m
g m
4g m
g m
g m
g m
(4.37)
72
Fig. 9. Three-loop diagrams contributing to S1 S2 .
The S22 terms can be obtained by replacements S1 S2 , and m m +

1
1
1
1
+ 2 3
+ 2 2
+ 2 2
2
3
3
3
4
g (m + )
g (m + )
4g (m + )
g (m + )4

1
1
+ 2 4
+ 2 4
(4.38)
S2.
2
g (m + )
g (m + )2 2
Finally, the diagrams in Fig. 9 lead to the S1 S2 terms.

1
2
1
1
1
1
+ 2 4
2 4
+
+
g m2 (m + )2
g m2 (m + )2

2
1
1
2
+ 2 3 3+
+ 2 4
S1 S2 .
g
m
(m + )3
g m(m + )
(4.39)
4.3.2. Comparison with the annulus formula

To compare these diagrammatic computation with the annulus formula (3.33), let us
expand the latter in powers of S1 and S2 . For the two-cut solution the prime form is defined
on the curve

2
1
2
W (z) + f1 (z) = (z z1 )(z z2 )(z z3 )(z z4 ),
y =
g

1 3 2
W (z) = g z + z .
(4.40)
3
2
There are two 1-cycles: A1 around [z1 , z2 ] and A2 around [z3 , z4 ]. We assume that the two
cuts to be small so that we shall find a solution in a power series of the periods.
Using the addition theorem for the prime form, we write the annulus formula (3.33) on
the curve (4.40) as
) 1
E(m, m)E(
,
1
log
=
2
E(m, )E(
m,
) 2

m

.
(4.41)
The meromorphic 1-form m

on this curve is given by

y(m)
(1 z + 2 ) dz
1 dz
m
1
+
.
=

2 zm
y(z)
y(z)
(4.42)
73
The coefficients 1 , 2 for the holomorphic forms are determined by the zero A-period
conditions
2

m

= 0,
i=1 A
m

= 0.
A2
The integration in the first condition can be performed by deforming the contour to A
(Cm ). As a result, we obtain 1 = 1. The integration in the second condition is rather
difficult. To perform this integration, we introduce the parameterization [24]
QI
QI
21 ,
+ 21 ,
z2 =
2
2
Q+I
Q+I
43 ,
+ 43 ,
z4 =
z3 =
2
2
z3 + z4 z1 + z2
z1 + z2 + z3 + z4
,
I :=
.
Q :=
2
2
2
z
By expanding the condition z34 m

= 0 in terms of 21 , we determine the coefficient
4
2 perturbatively up to O(21 ) as
z1 =
2 (z + z )(m + (m z )(m z ) )

3
4
3
4
21
2 = (m z3 )(m z4 ) +
4z3 z4 m

421
2
m
m + (m z3 )(m z4 ) (z3 + z4 )
+
3 3 3
128z3 z4 m
2

9z3 10z3 z4 + 9z42

+ (m z3 )(m z4 ) 8z32 z42 (z3 + z4 ) + 2z3 z4 m 3z32 + 2z3 z4 + 3z42

+ O 621 ,
where m := m + I Q
2 .
The periods of the curve (4.40)
z4
S1 = g
z3
dz
y(z),
2
z2
S2 = g
dz
y(z),
2
(4.43)
z1
are evaluated in terms of the variables (Q, I, 21 , 43 ). We can iteratively solve the inverse
relations 21 (S1 , S2 , g, ), 43 (S1 , S2 , g, ) as [6]
43 (S1 , S2 , g, )2

4
8
8
S1 + 2 4 2S12 3S1 S2 + 3 7 S1 (5S1 13S2 )(4S1 3S2 ) + O S 4 ,
=
g
g
g
21 (S1 , S2 , g, )2 = 43 (S1 , S2 , g, ).
(4.44)
74
The integration of meromorphic 1-form (4.41) can be performed by expanding in terms

of 21 and 43 . By plugging the inverse relations (4.44), we obtain the perturbative expansion
(2)
F0 =
S1
S2
2gm2 2g(m + )2
52 8m + 8m2 2 52 + 8m + 8m2 2
S1 +
S
4g 2 4 m4
4g 2 4 (m + )4 2

24 + 33 m 52 m2 16m3 8m4
+
S1 S2 + O S 3 .
2
4
3
3
g m (m + )
(4.45)
These terms precisely coincide with (4.37), (4.38) and (4.39). The absence of the contribution from (2) (z) can also be checked to all orders by numerical calculations.
5. CFT techniques
So far we have considered the planar contributions to the Tr R R corrections in supersymmetric gauge theories. On the other hand, there are also corrections coming from the
torus diagrams [31,3941], which are known to be elegantly computed using CFT techniques [5,42,43]. In this section we will see how the full Tr R R correction F=0 (= the
(0)
planar gravitational corrections F (pl) (3.34) + the torus contribution F1 ) is reproduced in
this framework for the gauge theories with matter. We will first recall how it works in the
case without matter [33,4446], and then examine how the matter fields fit in the story.
5.1. CFT techniques without matter
Consider the matrix model partition function

1
Z = d exp Tr W () ,
gs
(5.1)
with an arbitrary tree level potential

W () =
tk k .
(5.2)
k=0
CFT techniques are based on the equivalence of the loop equation and the Virasoro constraints [33,4446]. That is, defining the loop operator
(z) :=
1
1
Tr
(5.3)
and the corrective field

(z) := W (z) + 2gs Tr log(z ),
(z) = 2S(z) W (z),
(5.4)
75
the loop equation

N
1 W (i )
(z) =
gs
z i
2
(5.5)
i=1
can be written in the equivalent form

1
((w))2
dw
= 0,
2i
wz
(5.6)
where denotes the expectation value. Here the contour C encircles all w = i , the
eigenvalues of , but not w = z. Since
(z) =
ntn zn1 +
1
2gs2 zn1
n=
n=0
tn
inside a correlator, (z) may be regarded as a free chiral boson of conformal field theory
[47], and Eq. (5.5) can be written as the Virasoro constraints

Tem (w)
dw
(5.7)
= 0,
wz
C
where
Tem (z) = +
2
1
(z)
2
4gs
(5.8)
is the energymomentum tensor. Note that the prefactor 4g1 2 has been chosen so that their
s
moments generate the Virasoro algebra in the standard normalization.
To leading order in gs , Z is computed by the saddle point approximation in which (z)
is replaced with its large N expectation value y (0)(z) [1]. We need to compute the nextto-leading order, which may be obtained by the Gaussian approximation around a classical
solution [43], and hence is described by a free conformal field theory.
For the case of pure gauge theories without matter, CFT techniques were utilized [5]
to compute the genus one contribution to the Tr R R correction. This goes as follows:
consider the correlation function of 2n twist operators (ai , a i ) (i = 1, . . . , 2n) on a sphere
[4852]
2n
CFT

(ak , a k )
k=1
= | det A|1
2n

i<j
|ai aj |1/4

({pi },{p i })

1
,
exp i(p p p p)
2
(5.9)
where is the period matrix of the double cover with genus n (not n 1!; the extra handle
see Section 5.3) and
arises by the plumbing fixture connecting z = and ,

j
1
z
.
Aij := dz
(5.10)
y
Ai
76
The loop momenta {pi }i=1,...,n1 run over a certain momentum winding lattice.
It was argued in [43], and was recently confirmed in [40], that the genus one free energy
can be built from the chiral determinant of a c = 1 free boson CFT, or the chiral piece of
(5.9) with particular loop momenta
2n
CFT

1/2
(ak )
= eiN N det 0
k=1
pi = 2 Ni
=e
iN N
det A
1/2
2n
(ai aj )1/8 .
(5.11)
i<j
The important point is that the twist operator (z) does not satisfy the Virasoro constraint
(because, for example, it has non-zero conformal weight 1/16) and hence must be replaced
with an appropriately modified star operator S(z) [53] given by
S(z) = 3/2 (z)1/24 (z),

(z) =
r (ai )(z ai )r1
(5.12)
(5.13)
r1/2
to this order of the approximation. If the number of cuts n is equal to the degree of W (),
we find6
2n
log 3/2 (ai ) = log
(ai aj ).
(5.14)
i<j
i=1
The replacement (5.13) then correctly yields the genus one free energy
2n
CFT

F (pl) +F1(0)
=
S(ak )
e
k=1
pi = 2 Ni
= eiN N det A1/2

(ai aj )1/6
(5.15)
i<j
as was confirmed in [31,40].7 Note that the CFT result (5.15) also automatically includes
the planar gravitational correction from sphere diagrams [16] with the momentum identification8
pi = 2 Ni (i = 1, . . . , n).
(5.16)
6 If the number of cuts n is smaller than the degree of W (), (5.14) has additional contributions correspond-
ing to the moment factors in Eq. (4.9) of [40], which reduce to constants in the maximally broken case.
7 det A reduces to that of genus n 1 up to an irrelevant multiplicative constant in the | | limit.
0
8 There are reasons why the assignment (5.16) is natural. First, T (z) := Tr 1 has a total residue N from
f
z
matter poles [18] associated with the total Nf / 2 momentum of vertex operators (see Section 5.2), whereas the
difference of the residues at z = and
is 2N . Another related reason is that if Nf = 2N , the total momentum
vanishes at infinity with this assignment, naturally reflecting the conformal bound of the gauge theory.
77
5.2. The full Tr R R correction and fermion correlators

We next consider the CFT interpretation of the gravitational corrections for gauge theories with matter. The partition function is now
! Nf
"

1
Z = d exp Tr W () exp
(5.17)
Tr log( mI ) .
gs
I =1
Treating the second factor perturbatively, it is an expectation value of e

with respect to the measure without matter. Since
Tr log( mI ) =

1
(mI ) + W (mI ) + Tr log(1),
2gs
Nf
I =1 Tr log(mI )
(5.18)
it should be obtained as a correlator of vertex operators of charge 2g1 s . If we take the

convention that

CFT
log(z w),
(z)(w)
(5.19)
the prefactor of (5.8) must be 1/2 so that gs2 = 1/2. This motivates us to consider the
CFT correlator
CFT
2n
Nf
N

1 i(mI )
f i()
(ak , a k )
e 2
e 2
.
(5.20)
k=1
I =1
This is equivalent to a correlation function of conjugate pairs of the vertex operators on a

hyper-elliptic Riemann surface without twist operators, except that their zero-mode contributions differ by a factor of two [52,54]. In fact, to correctly reproduce the annulus
amplitude we also need to take the normal ordering of the vertex operators on the same
sheets:
CFT
Nf
Nf
Nf
1 i(mI ) Nf i()
i()
1 i(m
I)
e 2
e 2
::
e 2
e 2
:
:
I =1
= (det 0 )
I =1
{pi } {pi }
A0 A0 ,
(5.21)
({pi },{p i })
{p }
where A0 i is the holomorphic block with loop momenta {pi } (i = 1, . . . , n). Again, let
us focus on a particular block of (5.21) with pi = 2 Ni (i = 1, . . . , n). Taking the

logarithm of this block, we have [52]
Nf
n
n

{ 2 Ni }
log A0
= 2 i
ij Ni Nj + 2i
Ni
i
i,j =1
i=1
I =1 mI
Nf
m
J)
1
E(mI , )E(,
,
log
2
E(mI , m
J )E(, )
I,J =1
(5.22)
78
ij is the period matrix of the Riemann surface on which the CFT is defined (see Section 5.3). The first and second terms are twice of the planar gravitational corrections (3.35)
and (3.36). As we mentioned, the zero-mode part of the twist correlator is half of that on
its double cover (Eqs. (4.3) and (5.21) in [52]). The last term is equal to the annulus amplitude F0(2) (3.33); by definition of normal ordering, it does not contain the correlations
between pairs of points on the same sheet. The starization of the matter vertices is not
necessary, since the contour C in (5.7) does not encircle z = mI . Thus we have shown that
the chiral block (5.22) together with the corrected chiral determinant precisely reproduces
all the contributions to the full Tr R R correction for the gauge theories with matter.
Remarkably, the correlator (5.20) closely resembles that in the bosonization of fermions
on Riemann surfaces [54]. This is in accord with the fact that the correlators of det( mI )
can be described as free fermion insertions in the soliton theory [55], and in some special
case they are realized as CFT correlators [56]. However, a significant difference here is
that they are all for the full matrix model correlators, whereas our result concerns only the
contributions of O(1/N 2 ). Another important point is that the normal ordering is required
in (5.21).
5.3. Non-compact CalabiYau as a pinched Riemann surface
The N = 1 supersymmetric gauge theories are realized in type IIB superstring theory on
non-compact CalabiYau three-folds. If one computes the effective superpotential by using
the geometric transition, one needs to introduce a cutoff 0 to regularize some periods
of a reduced non-compact Riemann surface. Although the quantities computed in matrix
models are finite (since the eigenvalues are distributed on finite intervals), a cutoff is needed
again if the free energy (or the disk amplitude) is written as a contour integral around
infinity. Thus a matrix model curve with n cuts is a genus n1 punctured Riemann surface.
On the other hand, the CFT correlator formula obtained in [52] and used above assumes
that the Riemann surface on which the CFT is defined is compact with no punctures. How
can we understand the cutoff 0 here? In fact, the Riemann surface for the CFT can be
thought of as a pinched Riemann surface of genus n, where |0 /|2 is identified as the
pinching parameter t of the plumbing fixture. Indeed, suppose that we identify two annular
regions near ,
2

gs =0
W1 = p n1
on the 1st sheet || < |z(p)| < 0 ,

2

gs =0
W2 = p n1
(5.23)
on the 2nd sheet || < |z(p)| < 0 ,
by gluing them with a cylinder

$
#
S = (X, Y ) C2 XY = t
through the identification

|| tz
,
(X, Y ) =
,
z ||
for W1 ,
(5.24)

(X, Y ) =

tz ||
,
,
|| z
for W2
79
(5.25)
with t = |/0 |. Then B i s (i = 1, . . . , n) become closed contours,and we can verify

by a change of the canonical homology basis that the period matrix { B i dz j }i,j =1,...,n
coincides with Fays period matrix formula for the pinched Riemann surface [32] in the
|0 | limit.
6. On-shell condition
6.1. The N = 2 limit and the torus contribution
The N = 2 supersymmetry can be restored for U (N) gauge theories if the superpotential is in the form

N

gp p
N+1
+
W () = gN+1
(6.1)
p
p=1
so that the gauge breaking pattern is U (N) U (1)N ; an N = 2 theory can be realized by
taking the gN+1 0 limit [25]. The SeibergWitten curve is recovered from the matrix
model curve in this limit if the moduli Si satisfy the on-shell condition Weff (Si )/Si = 0
[17,18]. (See also [57].) Turning off the superpotential term in this way, the N = 1 Weyl superfield G ( ) and the graviphoton superfield F ( ) are combined into a single N = 2
Weyl superfield W (, ) = F ( ) + G . Consequently, the N = 1 gravitational
F-terms yield to an N = 2 F-term

d 4 x d 4 F=0 (Si )W W .
(6.2)
One can thus recover the N = 2 results by simply replacing the matrix model curve with
the SeibergWitten curve.
The W 2 factor contains the Tr R R component; it can be calculated by topologically
twisted N = 2 YangMills theories [15], since the effect of the topological twist becomes
invisible on a hyper-Khler four-manifold M4 and the coefficient of Tr R R term of
physical N = 2 theories agrees with that of topological theories. The latter yields up to a
constant factor
2
F=0 = b(Si ) c(Si ),
3

1
zj 1
b(Si ) = log det Aij , Aij = dz
,
2
y
Ai
c(Si ) =
1
log ,
12
2N

i<j
(zi zj )2 ,
(6.3)
80
where b(Si ) and c(Si ) are defined on the SeibergWitten curve

y 2 = PN (z)2 2NNf
Nf
(z mI ) =
I =1
2N

(z zi ).
(6.4)
i=1
On the SeibergWitten curve, the torus free energy which is obtained by conformal
field theory techniques exactly coincides with (6.3). Recently, it was confirmed that the
topological partition function coincides with the multi-instanton counting formula [11] for
g = 1 [13]. Thus we find consistency of our formula with the N = 2 SYM results.
6.2. The origin of the planar contributions?
We have, however, also the planar gravitational corrections, which have no counterpart
in the instanton calculations of N = 2 theories. Where do these terms come from? A suggestive observation is that all the planar terms can be combined into a simple integral
Nf
n
n

1
Ni Nj ij +
Ni
i
F (pl) = 2i
2
i,j =1
i=1
I =1mI
Nf
m
J)
1
E(mI , )E(,
log
2
E(mI , m
J )E(, )
I,J =1

n
Nf

Ni
dz T (z),
= i
+
i=1
B i
(6.5)
I =1mI
where
T (z) dz =
n

i=1
Ni i +
Nf
1
mI
,
2i
(6.6)
I =1
1
and solves the Konishi anomis nothing but the gauge theory expectation value of Tr z
aly equations [18]! Since the B-period integrals of T (z) on shell may naturally be regarded
as NSNS fluxes on a deformed CalabiYau three-fold [24]
W (z)2 + fn1 (z) + y 2 + v 2 + w2 = 0,
(z, y, v, w) C4 ,
(6.7)
F (pl)
would correspond to some interaction term on the D5-branes couple

it suggests that
to 2-form field B2 = BR + BNS before the geometric transition. This motivates us to
consider the induced ChernSimons term on Dp-branes [58]

I=
(6.8)
C ch(F ) A(R),
Mp+1
where C is a form field on the Dp-branes. ch(F ) and A(R)

are defined for the ChanPaton
gauge and tangent bundles on Mp+1 , respectively. In our case D5-branes are wrapped
81
around 2-cycles in a CalabiYau three-fold and filling the four-dimensional spacetime.

From this ChernSimons term we obtain the Tr R R term

(6.9)
C (tr 1) Tr R R,
M6
where tr 1 comes from ch(F ) and C is restricted to a 2-form in the CalabiYau direction.
In this brane setup we will have the Tr R R term
n

i=1

+
Ni
B i
Nf

BNS
I =1m
Tr R R.
(6.10)
R4
Note that the weight factors Ni naturally arise from tr 1. Thus, we conjecture that the onshell planar gravitational corrections may have their origin in the ChernSimons coupling
induced on the D5-branes.
Although these arguments are not conclusive, the relation (6.5) is certainly suggestive
and would be worth being studied.
7. Summary and discussion

In this paper we have studied the gravitational corrections to the effective superpotential
in four-dimensional N = 1 U (N) gauge theories with flavors in the fundamental representation, using the matrix model approach of Dijkgraaf and Vafa. We derived a compact
formula for the annulus contribution to the corrections in terms of the prime form on the
matrix model curve. We also showed that the full Tr R R correction containing the torus
as well as all the planar contributions can be reproduced as a special momentum sector of
a single c = 1 CFT correlator. The N = 2 limit of the torus contribution agrees with the
answer of the multi-instanton calculations and also with the geometric engineering argument from the topological A-model. The planar contributions, on the other hand, have no
counterpart in the instanton calculations of N = 2 gauge theories, and we speculated that
the latter might correspond to the ChernSimons term induced on the D5-branes.
The CFT correlator we found is very close to the fermion correlator on the Riemann surface. In [22] the correspondence between the non-compact B-branes and fermions has been
discussed. Since the matter contributions come from the Nf D5-branes wrapped around
non-compact 2-cycles, our result is consistent with the picture [22,23]. In the topological
B-model, the genus one partition function is given by a generalized holomorphic Ray
Singer torsion on the CalabiYau geometry [28]
F1 =
3
pq(1)p+q log det p,q ,
(7.1)
p,q=1
where p,q is the Laplacian acting on (p, q)-forms. By using Quillens anomaly, we also
find the one-loop open topological string partition function. Our formula may lead to the
explicit expression for it on a non-compact CalabiYau three-fold.
82
There are various interesting directions to extend our analysis carried out in this paper.
A simple extension of our analysis will be to investigate the gravitational corrections to
the SO/Sp gauge theories. If the N = 2 theory is deformed by an adjoint chiral superfield, the ChernSimons coupling term is also induced by the orientifold plane [59]. It
would be interesting to see whether such terms arise from the Klein bottle amplitudes in
orthogonal matrix models. On the other hand, real symmetric/symplectic matrix models
describe non-perturbative aspects of the SO/Sp gauge theories with flavors in the symmetric/antisymmetric tensor representations, respectively. The CFT description of these matrix
models can be given by a c = 2 CFT [60]. We will report on this elsewhere.
Another interesting issue to consider concerns higher-dimensional gauge theories. It
was proposed in [61,62] that some five-dimensional gauge theories can be described by
unitary matrix models. The associated matrix model curve is represented by a pair of
cylinders, where z = + and cannot be distinguished. As a result there exist two
chiral boson fields in the CFT description. It is interesting to investigate whether the fivedimensional multi-instanton calculations can be reproduced in this CFT analysis. We also
expect that the meaning of the five-dimensional ChernSimons term [14,62] will be clarified.
Acknowledgements
We would like to thank M. Naka for his collaboration at an early stage of this work,
and N. Ishibashi and H. Kanno for useful discussions. We also thank H. Eguchi, T. Eguchi,
S. Hosono, S. Iso, K. Ito, M. Jinzenji, A. Kato, H. Kawai, T. Kawai, Y. Kitazawa, Y. Matsuo, Y. Nakayama, M. Natsuume, Y. Ookouchi, H. Suzuki, Y. Sugawara, Y. Tachikawa,
Y. Yamada and A. Yamaguchi for helpful comments. The research of S.M. was supported
in part by Grant-in-Aid for Scientific Research (C)(2) #14540286 from Ministry of Education, Culture, Sports, Science and Technology.
Appendix A
In this appendix we give an alternative proof of (3.17). See Section 3.3 for the definitions
of (2)(z), y (2) (z), etc.
A.1. Singularities of (2)(z) dz
We will first determine (2) (z) dz from its singularities. In fact, to prove the assertion,
the precise form is not needed. We will nevertheless derive it for future convenience for it
(3)
is relevant to the computation of F0 .
(2)
First of all, (z) dz (as well as other higher (k) (z) dzs for all k 1) has zero
A-periods, for the same reason as beforeSi s are independent of gs . In fact, (2) (z) dz
has only singularities at the 2n branch points z = ai (0) (i = 1, . . . , 2n), which are of second order. They arise since the matter insertions change the locations of the cuts. We will
show that (2) (z) dz is an Abelian differential of the second kind with zero A-periods and
83
given by
(2)
(z) =
2n
2n 2 (1)

ai Mi
l=1 (z al )
i=1
2n
k=1,k =i (ai
ak )
16S(z ai )
n2
j =0 pj z
+
2n
l=1 (z al )
(A.1)
ai
:= ai (0)
for some pj (j = 0, . . . , n 2), where ai := ai (0),

and

1
W (z)
(1)

Mi =
(i = 1, . . . , 2n).
dz
2i
2n
(z
a
)
(z
a
)
i
l
l=1
A
(A.2)
(1)
Mi s are the moments introduced in [31]. pj s (j = 0, . . . , n 2) are determined so that

(2) (z) has zero A-periods.
Instead of studying the singularities of (z, gs ) directly, it is easier to investigate those
of y(z, gs ) (3.29). y (2)(z) and 2S(2)(z) have the same singularities since the difference
between the two is only linear in gs .
gs
It is easy to see that y (2)(z) dz is regular at z = mI , m
I (I = 1, . . . , Nf ) since I zm
I
has no singularity of O(gs2 ). The integral representation (3.29) shows that y (2)(z) dz has
and the branch points z = aj (0) (j = 1, . . . , 2n). Howpossible singularities at z = ,
2
ever, using the conditions (3.22) as well as those obtained by acting g s |gs =0 and g
2 |gs =0 ,
s
precisely cancel. Therefore, we have
it can be shown that the singularities at z = ,
only to examine the singular behavior of y (2)(z) at the branch points.
Since

1 2 y(z, gs )
(2)
y (z) =
(A.3)
,
2
gs2 gs =0
2
we see that the singularities at the branch points arise when g

2 |gs =0 acts on the first factor
s

2n
(2)
i=1 (z ai (gs )) of y(z, gs ), and therefore y (z) is proportional to
1
2i

+
1
=
2i
Cz
Nf

gs
W (w) + I =1

wmI

dw

2n
gs =0
(z w)
i=1 (w ai (gs ))
dw
A
zaj
W (w)
(1)

Mj
2n
(z w)
i=1 (w ai (0))
(j = 1, . . . , 2n).
(A.4)
The contour C surrounds all the cuts, and Cz surrounds w = z. Matching the Laurent
coefficients at the branch points, we obtain
y (2)(z) =
2n
2n

ai 2 Mi(1)
(ai ak )
8(z ai ) k=1
2n
l=1 (z al ) i=1
+ holomorphic differentials.
k =i
(A.5)
84
This implies Eq. (A.1).

Unlike (1)(z), (2) (z) depends on the potential W () not only through the positions
(1)
(3)
of the cut but also through the moments Mi , reflecting the non-universality of F0 .
A.2. An alternative proof of (3.17)
We will now prove (3.17). Using the saddle point equation
W (, gs )
= 2S
d
( , gs )
,

(A.6)
the large N free energy (3.25) is written as
F0 (gs ) =
n

Sj
j =1

| 0j |
1
d (, gs ) W (, gs ) + S log
2

1
W (0j , gs ) .
2
(A.7)
Note that (A.6) is true only if belongs to the non-zero support of (, gs ), that is, only
if is such that (, gs ) = 0. 0j is an arbitrary point on the j th cut, whose location is
independent of gs .
We can extract the annulus contribution
F0(2) = F0(2) (I) + F0(2) (II),
S
(2)
F0 (I) :=
(2)
F0 (II) :=

d (1) ()
Nf
log( mI ),
I =1

n

Sj
j =1
(A.8)

d (2) () W () 2S log | 0j | .
(A.9)
(2)
(2)
d () = 0.) Since F0 (I) is already the right-hand side
(2)
of (3.17), we must show that F0 (II) = 0. For this purpose let us first rewrite (A.8) in a
contour integral on the first sheet using (2) (z). We take the branch cut of log(z 0j )
(log drops out since
so that it runs from z = + i0 (z = on the upper side of the cut) to z = . Then

depending on i < j , = j or > j we have
85
a2i
d (2) () log | 0j |
a2i1

1
dz (2) (z)(log(z 0j ) + i)
(i < j ),
2i Ai
0j i0

1
(2)
(2)
= 2i Aj (0j ) dz (z) log(z 0j ) + i 0j +i0 dz (z) (i = j ),
1 dz (2) (z) log(z )

(i > j ),
0j
2i Ai
(A.10)
where the contour Aj (0j ) starts from z = 0j + i0, surrounds the j th cut once anticlockwise and goes back to z = 0j + i0 again but on the other side of the branch cut of
log(z 0j ). From (A.10) we find

d (2) () log | 0j | =
1
4S

lim
|0 |
B j
dz y (2)(z)
(j = 1, . . . , n).
(A.11)
Therefore, to prove F0(2) (II) = 0 it suffices to show that

n
n

1
dz y (2) (z)W (z) +
Sj dz y (2) (z) = 0.
2i
j =1A
j =1
(A.12)
Bj
This can be proved by applying Riemanns bilinear identity, as we will show below.
Cut out along
1
1 1
An1 Bn1 A1
n1 Bn1 A1 B1 A1 B1 ,
(A.13)
s
n1
is represented by a 4(n 1)-sided polygon with identification in a standard manner.
Since y (2)(z) dz is an Abelian differential of the second kind, its integral
z
Y (z) :=
dx y (2)(x)
(A.14)
z0
defines a single-valued meromorphic function Y (z) inside the polygon for an arbitrary
reference point z0 . Let us evaluate the integral of
:= Y (z)y (0)(z) dz
(A.15)
along the 4(n 1) sides of the polygon. Applying Riemanns bilinear identity, we obtain

2i
Resp
g
s
pn1
n1

j =1
Aj

dz y
(2)
(z)
Bj

dx y
(0)
(x)
dz y
Bj

(2)
(z)
Aj
dx y
(0)
(x) .
(A.16)
86
The right-hand side is

right-hand side of (A.16) = 4i
n1

dz y (2)(z).
Sj
j =1
(A.17)
Bj
Let us compute the left-hand side. Since

2S
+ W (z) + O z2 ,
z

2S
z
W (z) + O z2 ,
+
z
z
y (0)(z)
we find
(A.18)
2i Res =
dz y (2) (z)W (z) + 4iSY (),
dz y (2)(z)W (z) 4iSY ()
2i Res
=
A
dz y (2)(z)W (z) 4iSY (),
=+
(A.19)
where A
is the same as A but on the second sheet. We have shown in Appendix A.1
that y (2)(z) dz has second order poles at every branch point z = ai (i = 1, . . . , 2n), and
therefore Y (z) has first order poles there, while we also see from (3.29) that y (0) dz has
first order zeroes at those branch points. Therefore, is regular there and hence has no
non-zero residues. Thus, in all, we have
left-hand side of (A.16)

.
= 2 dz y (2)(z)W (z) + 4iS Y () Y ()
(A.20)
By definition

=
Y () Y ()
dz y (2)(z),
(A.21)
where the contour cannot cross any side of the polygonal region since Y (z) is single-valued
only inside. Therefore, the contour can pass only through the nth cut:

= lim
Y () Y ()
(A.22)
dz y (2)(z).
|0 |
B n
Plugging (A.22) into (A.20) and equating it with (A.17), we obtain the desired Eq. (A.12).
87
Appendix B. The special geometry relation

(0)
It is now well known that the sphere amplitude F0 satisfies the special geometry relation

F0(0)
0
1
= lim
dz y (0) (z) W (0 ) + 2S log
(B.1)
.
|0 | 2
Si
B i
This fact was pointed out and proven using the energy cost argument in [1]. A proof using
a Legendre transformation was given in [18]. In this appendix we will give a sketch of
an alternative proof neither using Legendre transformations nor resorting to any physical
argument.
Before we consider (B.1), we first recall the relation [36]
y (0) (z)
dz = 4i i (z)
Si
(i = 1, . . . , n),
(B.2)
where, as we defined in the text, y (0)(z) is related to the large N resolvent (0) (z) by
y (0)(z) = W (z) 2S(0) (z).
(B.3)
W (z)
Note that the relation (B.2) is also true when the degree of
exceeds the number of
the cuts n and y (0)(z) has extra zero factors (see Footnote 3). To see this we first use the
integral representation of y (0)(z) (obtained by setting gs = 0 in (3.29)) to verify that the
Next we write
possible singularities of y (0) (z)/Si are only located at z = and .

y (0)(z) = W (z)2 + f z, {Sj } ,
(B.4)
for some f (z, {Sj }) to find that

f (z, {Sj }) z
1
O
W (z)
z
since
2S(0) (z) = W (z)
=
(B.5)

W (z)2 + f z, {Sj }
f (z, {Sj })

,
W (z) + W (z)2 + f (z, {Sj })
(B.6)
must behave like 2S/z + O(1/z2 ) as z . Therefore, we find that
y (0)
Si f (z, {Sj })
dz,
dz =
Si
2 W (z)2 + f (z, {Sj })
(B.7)
has at most a first order pole at z = , and hence can be written as a linear combination
of i s. Then (B.2) follows from the fact that the Ai -period of y (0)(z) dz is 4iSi (i =
1, . . . , n).
88

(0)
We will now prove the special geometry relation (B.1). F0 is given by

F0(0) =
S
2

d (0) ()W ()
+S
n

Sj
j =1
(B.8)
|
1
0j
W (0j ) .
d (0) () log
2S
(B.9)
We express it in terms of contour integrals of the complex plane. Note that the integral of
dz y (0)(z) log(z 0j ) along A does not vanish as |0 | , and hence must be taken
into account:

n

Sj
1
S
(0)
(0)
F0 = lim
dz y (z)W (z) +
dz y (0) (z) W (0 )
|0 | 8i
4
2
j =1
+ S 2 log
B j
0
.
(B.10)
Differentiating it by Si , we find

n
(0)

F0
1
0
= lim
i (z)W (z) i
Sj i (z) + S log
|0 |
Si
2
j =1
1
+
4

dz y
(0)
B i
B j

1
0
(z) W (0 ) + S log
,
2
(B.11)
where the S log 0 terms are separated so that each line gives a finite result. Again, we
can simplify it by using Riemanns bilinear relation. In this case we cut out the Riemann
surface along
1
1 1
1
An1 Bn1 A1
Bn A (Bn )
n1 Bn1 A1 B1 A1 B1 A
and evaluate

dz y (0)(z)i (z),
(B.12)
z
i (z) :=
i (x)
(i = 1, . . . , n),
z0
n (z) := n (z),
(B.13)
along the boundary of the resulting polygon. The result we obtain is

n

1
lim
i (z)W (z) + 2i
Sj i (z) +
dz y (0)(z) W (0 ) = 0,
|0 |
2
A
j =1
B j
B i
(B.14)
for i = 1, . . . , n. Using (B.14) in (B.11), we obtain the special geometry relation (B.1).
89
Appendix C. The large N free energy for a quadratic potential

S t 2 (gs )2 c(gs )2 (gs )2
+
F0 =
2
S
2
8

gs t
m2
(gs )(m + c(gs ))
S
(gs ) (gs )1
2
(gs )

gs t(gs )2
(gs )(gs )
1
log
+
2S
2
2(gs )2

(g )
g 2 (gs )1 (log( 2 s ) 1) + (gs ) log (gs )
s
S
(gs ) (gs )1

c(gs )2
(gs )
1
+
log (gs ) (gs )1 + t(gs )2 log
2
(gs )2 2

mc(gs ) log (gs ) + c(gs )
m
1
(gs )
2
(gs )
+ 2gs
(C.1)
+ gs log(m) ,
log
2
(gs )
(gs ) (gs )1
where
(gs )
m c(gs )
(m c(gs ))2 (gs )2

.
(gs )
References
[1] R. Dijkgraaf, C. Vafa, Nucl. Phys. B 644 (2002) 3, hep-th/0206255;
R. Dijkgraaf, C. Vafa, Nucl. Phys. B 644 (2002) 21, hep-th/0207106;
R. Dijkgraaf, C. Vafa, hep-th/0208048.
[2] F. Cachazo, N. Seiberg, E. Witten, JHEP 0302 (2003) 042, hep-th/0301006.
[3] F. Cachazo, M.R. Douglas, N. Seiberg, E. Witten, JHEP 0212 (2002) 071, hep-th/0211170;
N. Seiberg, JHEP 0301 (2003) 061, hep-th/0212225.
[4] H. Kawai, T. Kuroki, T. Morita, Nucl. Phys. B 664 (2003) 185, hep-th/0303210;
H. Kawai, T. Kuroki, T. Morita, Nucl. Phys. B 683 (2004) 27, hep-th/0312026;
T. Morita, hep-th/0403259.
[5] R. Dijkgraaf, A. Sinkovics, M. Temurhan, hep-th/0211241.
[6] A. Klemm, M. Marino, S. Theisen, JHEP 0303 (2003) 051, hep-th/0211216.
[7] H. Ooguri, C. Vafa, Adv. Theor. Math. Phys. 7 (2003) 53, hep-th/0302109;
H. Ooguri, C. Vafa, Adv. Theor. Math. Phys. 7 (2004) 405, hep-th/0303063.
[8] J.R. David, E. Gava, K.S. Narain, JHEP 0309 (2003) 043, hep-th/0304227;
L.F. Alday, M. Cirafici, J.R. David, E. Gava, K.S. Narain, JHEP 0401 (2004) 001, hep-th/0305217;
L.F. Alday, M. Cirafici, JHEP 0309 (2003) 031, hep-th/0306299;
J.R. David, E. Gava, K.S. Narain, JHEP 0309 (2003) 043, hep-th/0311086.
[9] H. Ita, H. Nieder, Y. Oz, JHEP 0312 (2003) 046, hep-th/0309041;
B.M. Gripaios, hep-th/0311025.
[10] S. Chiantese, A. Klemm, I. Runkel, JHEP 0403 (2004) 033, hep-th/0311258.
[11] N. Nekrasov, hep-th/0206161;
A.S. Losev, A. Marshakov, N. Nekrasov, hep-th/0302191;
N. Nekrasov, A. Okounkov, hep-th/0306238;
H. Nakajima, K. Yoshioka, math.AG/0306198.
(C.2)
90
[12] A. Iqbal, A.-K. Kashani-Poor, Adv. Theor. Math. Phys. 7 (2004) 457, hep-th/0212279;
A. Iqbal, A.-K. Kashani-Poor, hep-th/0306032;
T.J. Hollowood, A. Iqbal, C. Vafa, hep-th/0310272.
[13] T. Eguchi, H. Kanno, JHEP 0312 (2003) 006, hep-th/0310235;
T. Eguchi, H. Kanno, Phys. Lett. B 585 (2004) 163, hep-th/0312234.
[14] Y. Tachikawa, JHEP 0402 (2004) 050, hep-th/0401184.
[15] E. Witten, Selecta Math. 1 (1995) 383, hep-th/9505186;
G. Moore, E. Witten, Adv. Theor. Math. Phys. 1 (1998) 298, hep-th/9709193;
M. Marino, G. Moore, Commun. Math. Phys. 199 (1998) 25, hep-th/9802185.
[16] R. Dijkgraaf, M.T. Grisaru, H. Ooguri, C. Vafa, D. Zanon, JHEP 0404 (2004) 028, hep-th/0310061.
[17] S. Naculich, H. Schnitzer, N. Wyllard, JHEP 0301 (2003) 015, hep-th/0211254.
[18] F. Cachazo, N. Seiberg, E. Witten, JHEP 0304 (2003) 018, hep-th/0303207.
[19] R. Argurio, G. Ferretti, R. Heise, Phys. Rev. D 67 (2003) 065005, hep-th/0210291;
R. Argurio, G. Ferretti, R. Heise, hep-th/0311066;
J. McGreevy, JHEP 0301 (2003) 047, hep-th/0211009;
H. Suzuki, JHEP 0303 (2003) 005, hep-th/0211052;
H. Suzuki, JHEP 0303 (2003) 036, hep-th/0212121;
I. Bena, R. Roiban, Phys. Lett. B 555 (2003) 117, hep-th/0211075;
Y. Demasure, R.A. Janik, Phys. Lett. B 553 (2003) 105, hep-th/0211082;
Y. Demasure, R.A. Janik, Nucl. Phys. B 661 (2003) 153, hep-th/0212212;
I. Bena, R. Roiban, Radu Tatar, Nucl. Phys. B 679 (2004) 168, hep-th/0211271;
K. Ohta, JHEP 0302 (2003) 057, hep-th/0212025;
C. Hofman, JHEP 0310 (2003) 022, hep-th/0212095;
T.J. Hollowood, K. Ohta, hep-th/0405051;
C.H. Ahn, B. Feng, Y. Ookouchi, M. Shigemori, hep-th/0405101.
[20] V.A. Kazakov, Phys. Lett. B 237 (1990) 212.
[21] E. Brezin, C. Itzykson, G. Parisi, J.B. Zuber, Commun. Math. Phys. 59 (1978) 35.
[22] M. Aganagic, A. Klemm, M. Marino, C. Vafa, hep-th/0305132;
M. Aganagic, R. Dijkgraaf, A. Klemm, M. Marino, C. Vafa, hep-th/0312085.
[23] H. Ita, H. Nieder, Y. Oz, T. Sakai, hep-th/0403256.
[24] C. Vafa, J. Math. Phys. 42 (2001) 2798, hep-th/0008142;
F. Cachazo, K.A. Intriligator, C. Vafa, Nucl. Phys. B 603 (2001) 3, hep-th/0103067.
[25] F. Cachazo, C. Vafa, hep-th/0206017;
Y. Ookouchi, JHEP 0401 (2004) 014, hep-th/0211287.
[26] N. Berkovits, Class. Quantum Grav. 17 (2000) 971, hep-th/9910251;
N. Berkovits, Nucl. Phys. B 431 (1994) 258, hep-th/9404162.
[27] I. Antoniadis, E. Gava, K.S. Narain, T.R. Taylor, Nucl. Phys. B 413 (1994) 162, hep-th/9307158.
[28] M. Bershadsky, S. Cecotti, H. Ooguri, C. Vafa, Commun. Math. Phys. 165 (1994) 311, hep-th/9309140.
[29] G. t Hooft, Nucl. Phys. B 72 (1974) 461.
[30] H. Ooguri, C. Vafa, Nucl. Phys. B 641 (2002) 3, hep-th/0205297.
[31] J. Ambjorn, L. Chekhov, C.F. Kristjansen, Yu. Makeenko, Nucl. Phys. B 404 (1993) 127;
J. Ambjorn, L. Chekhov, C.F. Kristjansen, Yu. Makeenko, Nucl. Phys. B 449 (1995) 681, Erratum, hepth/9302014;
G. Akemann, Nucl. Phys. B 482 (1996) 403, hep-th/9606004.
[32] D. Mumford, Tata Lectures on Theta, vol. II, Birkhuser, Boston, MA, 1983;
J. Fay, Theta Functions on Riemann Surfaces, in: Lecture Notes in Mathematics, vol. 352, Springer, Berlin,
1973.
[33] J. Ambjorn, J. Jurkiewicz, Y.M. Makeenko, Phys. Lett. B 251 (1990) 517.
[34] M. Hanada, M. Hayakawa, N. Ishibashi, H. Kawai, T. Kuroki, Y. Matsuo, T. Tada, hep-th/0405076.
[35] Y. Shimamune, Phys. Lett. B 108 (1982) 407.
[36] F. Ferarri, Phys. Rev. D 67 (2003) 085013, hep-th/0211069.
[37] D. Shih, JHEP 0311 (2003) 025, hep-th/0308001;
H. Fuji, S. Mizoguchi, Phys. Lett. B 578 (2004) 432, hep-th/0309049;
C. Ahn, Y. Ookouchi, Phys. Rev. D 69 (2004) 086006, hep-th/0309156;
91
C. Ahn, Y. Ookouchi, JHEP 0402 (2004) 009, hep-th/0312162.

[38] R. Dijkgraaf, S. Gukov, V.A. Kazakov, C. Vafa, Phys. Rev. D 68 (2003) 045007, hep-th/0210238.
[39] D. Bessis, Commun. Math. Phys. 69 (1979) 147;
D. Bessis, C. Itzykson, J.B. Zuber, Adv. Appl. Math. 1 (1980) 109.
[40] L. Chekhov, hep-th/0401089.
[41] B. Eynard, A. Kokotov, D. Korotkin, hep-th/0401166;
B. Eynard, A. Kokotov, D. Korotkin, hep-th/0403072.
[42] M.R. Douglas, hep-th/9311130;
R. Dijkgraaf, Nucl. Phys. B 493 (1997) 588, hep-th/9609022.
[43] I.K. Kostov, hep-th/9907060.
[44] R. Dijkgraaf, E. Verlinde, H. Verlinde, Nucl. Phys. B 348 (1991) 435;
M. Fukuma, H. Kawai, R. Nakayama, Int. J. Mod. Phys. A 6 (1991) 1385.
[45] H. Itoyama, Y. Matsuo, Phys. Lett. B 255 (1991) 202.
[46] A. Mironov, A. Morozov, Phys. Lett. B 252 (1990) 47.
[47] E. Witten, Commun. Math. Phys. 144 (1992) 189.
[48] L.J. Dixon, D. Friedan, E.J. Martinec, S.H. Shenker, Nucl. Phys. B 282 (1987) 13;
S. Hamidi, C. Vafa, Nucl. Phys. B 279 (1987) 465.
[49] A.B. Zamolodchikov, Nucl. Phys. B 285 (1987) 481.
[50] M. Bershadsky, A. Radul, Int. J. Mod. Phys. A 2 (1987) 165;
K. Miki, Phys. Lett. B 191 (1987) 127.
[51] E. Verlinde, H. Verlinde, Nucl. Phys. B 288 (1987) 357.
[52] R. Dijkgraaf, E. Verlinde, H. Verlinde, Commun. Math. Phys. 115 (1988) 649.
[53] T. Miwa, Publ. Res. Inst. Math. Sci. 17 (1987) 665;
G.W. Moore, Commun. Math. Phys. 133 (1990) 261.
[54] D. Bernard, Nucl. Phys. B 302 (1988) 251.
[55] E. Date, M. Kashiwara, M. Jimbo, T. Miwa, in: M. Jimbo, T. Miwa (Eds.), Proceedings of RIMS Symposium
on Non-Linear Integrable SystemClassical Theory and Quantum Theory, Kyoto, 1981, World Scientific,
Singapore, 1983.
[56] N. Ishibashi, Y. Matsuo, H. Ooguri, Mod. Phys. Lett. A 2 (1987) 119.
[57] M. Matone, L. Mazzucato, JHEP 0307 (2003) 015, hep-th/0305225;
R. Flume, F. Fucito, J.F. Morales, R. Poghossian, JHEP 0404 (2004) 008, hep-th/0403057.
[58] M. Bershadsky, C. Vafa, V. Sadov, Nucl. Phys. B 463 (1996) 420, hep-th/9511222;
M.B. Green, J.A. Harvey, G.W. Moore, Class. Quantum Grav. 14 (1997) 47, hep-th/9605033.
[59] K. Dasgupta, D.P. Jatkar, S. Mukhi, Nucl. Phys. B 523 (1998) 465, hep-th/9707224;
K. Dasgupta, S. Mukhi, JHEP 9803 (1998) 004, hep-th/9709219;
B. Craps, F. Roose, Phys. Lett. B 450 (1999) 358, hep-th/9812149;
J.F. Morales, C.A. Scrucca, M. Serone, Nucl. Phys. B 552 (1999) 291, hep-th/9812071;
B. Stefanski, Jr., Nucl. Phys. B 548 (1999) 275, hep-th/9812088;
M. Gomez-Reino, S.G. Naculich, H.J. Schnitzer, JHEP 0404 (2004) 033, hep-th/0403129.
[60] G.R. Harris, Nucl. Phys. B 356 (1991) 685;
H. Awata, Y. Matsuo, S. Odake, J. Shiraishi, hep-th/9503028;
P. Svrcek, hep-th/0311238.
[61] R. Dijkgraaf, C. Vafa, hep-th/0302011.
[62] M. Wijnholt, hep-th/0401025.

www.elsevier.com/locate/npe
Matter unification in warped supersymmetric SO(10)

Yasunori Nomura a,b , David Tucker-Smith c
a Department of Physics, University of California, Berkeley, CA 94720, USA
b Theoretical Physics Group, Lawrence Berkeley National Laboratory, Berkeley, CA 94720, USA
c Department of Physics, Williams College, Williamstown, MA 01267, USA
Received 22 March 2004; received in revised form 28 May 2004; accepted 23 July 2004
Abstract
We construct models of warped unification with a bulk SO(10) gauge symmetry and boundary
conditions that preserve the SU(4)C SU(2)L SU(2)R PatiSalam gauge group (422). In the
dual 4D description, these models are 422 gauge theories in which the apparent unification of gauge
couplings in the minimal supersymmetric standard model is explained as a consequence of strong
coupling in the ultraviolet. The weakness of the gauge couplings at low energies is ensured in this 4D
picture by asymptotically non-free contributions from the conformal sector, which are universal due
to an approximate SO(10) global symmetry. The 422 gauge symmetry is broken to the standard model
group by a simple set of Higgs fields. An advantage of this setup relative to SU(5) models of warped
unification is that matter is automatically required to fill out representations of 422, providing an
elegant understanding of the quantum numbers of the standard-model quarks and leptons. The models
also naturally incorporate the see-saw mechanism for neutrino masses and bottom-tau unification.
Finally, they predict a rich spectrum of exotic particles near the TeV scale, including states with
different quantum numbers than those that appear in SU(5) models.
PACS: 12.60.-i; 12.10.-g
E-mail address: yasunori@thsrv.lbl.gov (Y. Nomura).

doi:10.1016/j.nuclphysb.2004.07.031
Y. Nomura, D. Tucker-Smith / Nuclear Physics B 698 (2004) 92110
93
1. Introduction
Two successful aspects of grand unification are the unification of gauge couplings and
the unification of matter into a smaller number of representations. Both features explain
something about nature. The first explains why the running gauge couplings appear to meet
at a very high energy in the context of weak scale supersymmetry. The second helps to explain the quark and lepton gauge quantum numbers in the standard model. The apparent
unification of couplings can be easily addressed if the unified group is a simple group containing all the standard model gauge interactions. The smallest successful group is SU(5)
[1], and the next smallest is SO(10) [2]. Of course, in these theories coupling unification
and matter unification are closely related, as the enlarged gauge symmetry requires quarks
and leptons to appear in representations of either SU(5) or SO(10).
Based on matter unification alone, however, the simplest approach is arguably SU(4)C
SU(2)L SU(2)R (422) la Pati and Salam [3]. This group provides a very elegant explanation of the quantum numbers of the standard model fermions, with a full generation
of quarks and leptons (including a right-handed neutrino) filling out the representations
(4, 2, 1) + (4 , 1, 2). The breaking of the 422 group to its standard model subgroup is
quite straightforward: it is attained by a simple set of Higgs fields, and this same set of
fields is sufficient to break unwanted quark and lepton mass relations (while preserving
bottom-tau unification) through non-renormalizable operators. This breaking also almost
automatically leads to small neutrino masses through the see-saw mechanism (which is not
the case in SU(5)). The simplicity of the gauge breaking makes 422 attractive compared
with larger unified groups such as SO(10), which require a more complicated structure for
realistic gauge breaking.
In this paper we explore the interesting role that the 422 gauge group can play in models
of warped supersymmetric unification. In the SU(5) model introduced in [4], the unified
gauge symmetry realized in the bulk is explicitly broken by boundary conditions on the
Planck brane. This implies that in the 4D dual description the theory does not have an
SU(5) gauge symmetry. Rather, SU(5) appears as an approximate global symmetry, possessed by the strongly interacting conformal sector that arises as the dual description of
the bulk physics. The successful supersymmetric prediction relating the low-energy gauge
couplings then follows from the assumption that the theory becomes strongly coupled in
the ultraviolet. The role of the SU(5) global symmetry in this context is to ensure that the
contributions from the conformal sector to the gauge coupling evolution are SU(5) symmetric, so that the non-universal contributions are given purely by the elementary states,
which are identical to the states of the minimal supersymmetric standard model (MSSM).
Thus, a simple group still plays an important role in the unification prediction, but in the
infrared, rather than the ultraviolet, and as a global symmetry, rather than a gauge symmetry.
This extra global symmetry does not require a complicated breaking mechanismit is
explicitly broken from the beginningbut nor does it require that the matter fields form
unified representations. Suppose we introduce quarks and leptons on the Planck brane in
the SU(5) model of [4]. Then there is no reason why they must fill out SU(5) representations, and no reason why their hypercharges must satisfy the appropriate quantization
condition. This point becomes especially important in the class of models introduced in [5],
94
where the unified symmetry is broken both on the Planck and TeV branes, in which case
the quarks and leptons cannot arise from bulk matter alone. We thus clearly need some
other ingredient for understanding matter quantum numbers in these setups. We propose
to use the 422 gauge group for this purpose. We promote the gauge group in the 4D dual
picture to 422 in these models. This in turn requires that we promote the global group of
the conformal sector, which is the bulk gauge group in the 5D picture, to SO(10), at least.
The correct prediction relating the gauge couplings at low energies then arises through the
global SO(10) group, with the assumption of strong coupling. Enlarging the global group
to SO(10) has direct consequences for physics at low energies, because it results in light
exotics with different quantum numbers than those in the SU(5) models. The resulting
phenomenology is, naturally, quite rich.
The organization of the paper is as follows. In Section 2 we construct a model based
on the structure of the model of [5], so that the bulk SO(10) group is broken both on the
Planck and TeV branes. In Section 3 we present a model in which the bulk SO(10) is broken only on the Planck brane, similarly to the model of [4]. In both theories the matter
fields are located on the Planck brane, but without the problems such a setup produces in
the SU(5) models. Moreover, we gain several desirable features coming from 422, including quarklepton unification (and thus hypercharge quantization), bottom-tau unification,
and a natural see-saw mechanism. Some of the group theory in our models shares certain
features with SO(10) models in flat space [6], but the physics involved is quite different.
Conclusions are given in Section 4.
2. Model with TeV-brane symmetry breaking

In this section we construct a model having the properties described in the introduction,
and in which the bulk gauge symmetry is reduced at the TeV brane. This model can be
viewed as an extension of the 321321 model of Ref. [5]. The main new ingredient of the
model presented here is the unification of matter fields localized on the Planck brane. This
leads to an understanding of the quark and lepton quantum numbers, as well as the ratio
of the bottom-quark and tau-lepton masses through Yukawa unification. Small neutrino
masses also arise quite naturally through the see-saw mechanism. All these features can
be accommodated without spoiling the interesting features of the 321321 model: automatic doublettriplet splitting, suppression of proton decay, and a rich phenomenology of
superparticles and grand-unified-theoretic (GUT) particles.
2.1. Basic setup
The model is formulated in a 5D warped spacetime with the extra dimension y compactified on an S 1 /Z2 orbifold: 0 y R. The metric is given by
ds 2 = e2k|y| dx dx + dy 2,
(1)
where k is the AdS curvature, which is taken to be somewhat (typically a factor of a few)
2 M 3 /k
smaller than the 5D Planck scale M5 . The 4D Planck scale, MPl , is given by MPl
5
95
and we take k M5 MPl . We choose kR 10 so that the TeV scale is naturally generated
by the AdS warp factor: k kekR TeV [7].
We choose the bulk gauge group to be SO(10). This bulk SO(10) symmetry is then
broken by boundary conditions imposed at the boundaries both at y = 0 (Planck brane) and
R (TeV brane). Using the 4D N = 1 superfield notation, in which the 5D gauge multiplet
is described by a vector superfield V (A , ) and a chiral superfield ( + iA5 , ), the
boundary conditions are given by

V
P V P 1
x ,y ,
x , y =
P P 1

P V P 1
V
(2)
x ,y ,
x , y =
P P 1
where y = y R. The matrix P is chosen such that it leaves the PatiSalam SU(4)C
SU(2)L SU(2)R (422) subgroup of SO(10) invariant. Specifically, in the basis where
the generators of SO(10), which are imaginary and antisymmetric 10 10 matrices, are
given by 0 A5 , 1 A5 , 2 S5 and 3 A5 , the matrix P can be chosen as P =
0 diag(1, 1, 1, 1, 1). Here, 0 is the 2 2 unit matrix and 1,2,3 are the Pauli spin
matrices; S5 and A5 are 5 5 matrices that are real and symmetric, and imaginary and
antisymmetric, respectively. This reduces the gauge group at the Planck and the TeV branes
to 422, and leaves the (15, 1, 1) + (1, 3, 1) + (1, 1, 3) component of V and the (6, 2, 2)
component of as zero modes, where the numbers in parentheses represent quantum
numbers under 422. All components of the SO(10) gauge multiplet have KaluzaKlein
(KK) towers with the typical mass scale of k TeV.
The Higgs fields are introduced in the bulk as a hypermultiplet transforming as 10 of
SO(10), which is described by two chiral superfields as {H, H c } in the 4D N = 1 superfield
notation. They obey the boundary conditions

H
P H
x
x ,y ,
,
y
=
Hc
PHc

P H
H

=
,
y
(3)
x
x ,y ,
PHc
Hc
which leave the (1, 2, 2) component of H and the (6, 1, 1) component of H c as zero
modes.1 The Higgs multiplet can have a mass parameter in the bulk, which we parameterize as cH k. The parameter cH then controls the wavefunction profiles for the zero modes
arising from {H, H c }. As in the 321321 model of [5], the unwanted zero modes from
and H c obtain masses when supersymmetry is broken.
Matter fields are introduced on the Planck brane as chiral superfields in the (4, 2, 1) +
(4 , 1, 2) representation of 422 for each generation, which contain our quarks and leptons (and a right-handed neutrino) as = {Q, L} and = {U, D, E, N}. Since the gauge
1 Alternatively, we could introduce the Higgs field H on the Planck brane in the (1, 2, 2) representation of
422. Another possibility is to put the Higgs fields in the bulk and impose the boundary conditions of Eq. (3)
with an extra minus sign in the right-hand side of the second equation (i.e., flipping the TeV-brane boundary
conditions) [5].
96
group on the Planck brane is non-Abelian, the charges of the matter fields are quantized.
This setup also requires the existence of right-handed neutrinos, which is an important
ingredient for the see-saw mechanism.
To reproduce successful phenomenology at low energies, the low-energy gauge group
must be reduced to SU(3)C SU(2)L U (1)Y (321). To this end, we break the 422 group
by the Higgs mechanism on the Planck brane. The simplest possibility is to introduce chiral
superfields on the Planck brane in the (4, 1, 2) + (4
, 1, 2) representation of 422, and
give appropriate vacuum expectation values for them.2 The expectation
values are easily
induced, for example, by introducing the superpotential interaction d 2 S( v2 ) +
h.c. on the Planck brane, where S is a singlet chiral superfield. Here we take the expectation
values =
= v to be of order k, which is a natural scale on the Planck brane. This
then breaks the gauge group on the Planck brane to 321 at the scale k, and gives masses to
the 422/321 component of V , which would otherwise be massless.
Quark and lepton masses are generated on the Planck brane through the following operators:

S=

R
2
2
d x dy 2(y)
d y HD +
( ) + h.c. ,
M
4
(4)
where HD represents the (1, 2, 2) component of H under the 422 decomposition, and
we have omitted generation indices. M is the cutoff scale of order M5 . The first term
gives Yukawa couplings for quarks and leptons while the second term gives Majorana
masses for right-handed neutrinos of order 2 /M , which generates small masses
for the observed neutrinos through the see-saw mechanism. The unwanted mass relations arising from Eq. (4) for the first-two generation quarks and leptons are broken
by higher-dimensional operators involving and ,
allowing for realistic quark and
lepton masses and mixings.3 For relatively suppressed 422-breaking expectation values
= k M , the Yukawa couplings for the third generation fermions are approximately unified at the scale k, which gives a successful mb /m prediction at low energies.
It also leads the theory to the large tan region, tan Hu /Hd 50, where Hu and
Hd ( HD ) are the Higgs fields giving masses to the up-type and down-type quarks, respectively. The couplings in Eq. (4) respect a U (1)R symmetry with the charges given by
This symmetry,
when imposed on the
V (0), (0), H (0), H c (2), (1), (1), (0), (0).

theory, forbids potentially dangerous operators, such as d 2 HD2 + h.c., on the Planck
brane. As in the SU(5) models, proton decay is not a problem for the Planck-brane localized matter because all potentially dangerous gauge bosons (and their KK towers) have
wavefunctions strongly peaked towards the TeV brane.
2 We could also introduce (4, 2, 1) + (4 , 2, 1) fields on the brane and make the breaking of leftright
symmetry entirely spontaneous.
3 The higher dimensional operators that allow for realistic fermion masses have the same form as those employed in 4D 422 theories, see, e.g., [8].
97
2.2. Prediction for gauge couplings

The SO(10) generators in the present model are naturally divided into three classes:
(i) the 321 generators, (ii) the generators belonging to 422/321, which we call PS, and
(iii) the generators belonging to SO(10)/422, which we call XY. The spectrum of the gauge
sector is then given as follows. The 321 gauge multiplet has a zero mode V 321 and a KK
tower, which consists of V 321 and 321 at each KK level, with the masses mn given by the
solutions of
g2

J0 mkn + 2B mn J1 mkn
J0 mk n
g 0,a
(5)
g2
= Y mn ,
0 k
Y0 mkn + 2B mn Y1 mkn
g 0,a
where Jn (x) and Yn (x) are the Bessel functions of order n, and a = 1, 2, 3 represents
2
U (1)Y , SU(2)L and SU(3)C , respectively; gB is the bulk SO(10) gauge coupling and g 0,a

are the Planck-brane gauge couplings appropriately renormalized at the scale k (for more
precise definitions and our assumptions regarding the ultraviolet values of these parameters, see [5]). The PS gauge multiplet does not have a zero-mass mode, and its KK tower
consists of V PS and PS at each KK level with the masses approximately given by4

J1 mkn
J0 mk n
= mn .
(6)
Y1 mkn
Y0 k
Finally, the XY gauge multiplet has a zero mode XY , and its KK tower {V XY , XY } has
masses given by

J1 mk n
J1 mkn
= mn .
(7)
Y1 mkn
Y1 k
This spectrum can be summarized, for k k, as
321
XY
V , : m0 = 0,

321
V , 321 + V PS , PS : mn n 14 k ,

XY XY
: mn n + 14 k ,
V ,
(8)
where n = 1, 2, . . . . The transformation properties of these fields under the 321 gauge
group are given by (8, 1)0 + (1, 3)0 + (1, 1)0 for V 321 and 321 , (3, 1)2/3 + (3 , 1)2/3 +
(1, 1)1 + (1, 1)1 + (1, 1)0 for V PS and PS , and (3, 2)5/6 + (3 , 2)5/6 + (3, 2)1/6 +
(3 , 2)1/6 for V XY and XY . A schematic depiction of the spectrum is given in Fig. 1(a).
4 More
precisely,
the
left-hand
side
of
Eq.
(6)
is
given
by
{J0 (mn /k)
2 v 2 /m )J (m /k)}/{Y (m /k) (Cg 2 v 2 /m )Y (m /k)}, where C is an O(1) coefficient de(CgB
n 1 n
n 1 n
0 n
B
pending on the gauge component. For v k , however, the denominator is well approximated by
2 v 2 /m )Y (m /k), which is enough to guarantee that the mass eigenvalues are almost given by
(CgB
n 1 n
the zeros of the right-hand side of Eq. (6). The expression for the left-hand side further simplifies for v k and
2 k 1 to J (m /k)/Y (m /k), which we have used in Eq. (6). However, this is not essential because the
gB
1 n
1 n
solutions are quite insensitive to the detailed expression of the left-hand side for v k . The same comment
applies also to the expressions involving the PS gauginos, e.g., Eq. (14).
98
Fig. 1. Schematic depiction for the lowest-lying masses for the gauge multiplet. The three figures represent the
spectrum (a) in the supersymmetric limit, (b) for small supersymmetry breaking and (c) for large supersymmetry
breaking. Each bullet for 321 (PS and XY ) represents a Majorana (Dirac) degree of freedom.
With the gauge spectrum of Eq. (8) and the Higgs fields of Eq. (3), the MSSM prediction for the low-energy gauge couplings is preserved. Specifically, the low-energy 321
gauge couplings are given by setting (T1 , T2 , T3 )(V++ ) = (0, 2, 3), (T1 , T2 , T3 )(V+ ) =
(0, 0, 0), (T1 , T2 , T3 )(V+ ) = (14/5, 0, 1) and (T1 , T2 , T3 )(V ) = (26/5, 6, 4) in Eq. (9)
of Ref. [4] and adding the Higgs contribution, which is identical to that of the 321321
model (the spectrum of the PS gauge multiplet is effectively reproduced by imposing the
(, +) and (+, ) boundary conditions on V PS and PS , respectively, and the Higgs spectrum is identical to the SU(5) Higgs spectrum of [5] with cH = cH (5) = cH (5 ) ). We then
99
find that for cH 1/2 the prediction for low-energy 321 gauge couplings ga is given by
1
ga2 (k )
where

1
2
3
(universal) +

1 a
,
8 2

33/5
k
ln ,
1
k
3
(9)
(10)
hence reproducing the successful MSSM prediction.5 In a suitable renormalization scheme,

the logarithmic contribution of Eq. (10) arises entirely from the Planck brane cou2 a /8 2 in Eq. (5). The result of Eqs. (9),
plings g0,a , implying that we should use 1/g0,a
(10) can also be understood in the 4D dual picture as follows. In the 4D picture the theory
between v k and the TeV scale is described by an N = 1 supersymmetric gauge theory with the gauge group SU(3)C SU(2)L U (1)Y G with the G sector possessing
a global SO(10) symmetry, where G represents some gauge interaction whose coupling
evolves very slowly over this energy interval. The quark, lepton and Higgs doublets are
interpreted as elementary fields, while various GUT states are regarded as composite states
arising from the non-trivial infrared (of order TeV) dynamics of G. Since the contribution
from the G sector is universal due to the global SO(10), the differences among the lowenergy 321 gauge couplings arise entirely from the contribution of the elementary fields
(assuming strong coupling at ultraviolet, see [4]). This gives the desired MSSM prediction
because the elementary sector is identical to the MSSM.
Here we comment on calculability in this model. In a theory of warped supersymmetric
unification, the size of the bulk gauge coupling gB is related to the 4D gauge coupling
as 1/g42 = R/gB2 , where g4 represents the unified gauge coupling in conventional 4D
supersymmetric unification, g4 0.7. Defining M to be the scale where the 5D theory
becomes strongly coupled, we obtain 1/gB2 CM /L, where C and L are the grouptheoretic and 5D-loop factors, respectively [11]. For the SO(10) theory, C 8. Using the
5D-loop factor of L 24 3 [12] and kR 10, we obtain M /k 2. This implies that
the infrared cutoff of the theory, M M ekR , is close to the scale of the first KK
excitation, k : M /k 2. This strongly restricts calculabilitywe generically expect
errors of order (k /M )n in various predictions, where n depends on the quantity (errors
for the masses of the lightest 321 gauginos could be suppressed further by 1/ ln(k/k )).6
The equations that follow should thus be interpreted with care: their precision is not very
high and the results for higher KK towers are not meaningful. In general, this is the case for
any warped theory with a large bulk gauge symmetry, so the same comment also applies to
the model presented in the next section. We stress, however, that the main features of the
model, such as its unified understanding of matter quantum numbers and the qualitative
aspects of its spectrum, are not affected by these numerical limitations.
5 Successful gauge coupling unification in warped unified theories was anticipated in [9] based on a heuristic
argument, and was shown explicitly in [4]. Techniques for calculating gauge coupling evolution in warped space
were developed in Refs. [10].
6 The situation is somewhat better in theories with the bulk SU(5) symmetry because of the smaller value of
C: C 5.
100
2.3. Supersymmetry breaking

We now consider the effects of supersymmetry breaking in the present model. Here we
follow the notation of Ref. [5]. Supersymmetry breaking is introduced on the TeV brane
through the following potential [13]:

S=
R
d x dy 2(y R)
4
e2kR

d 4 Z Z + e3kR d 2 2 Z + h.c. ,
(11)
where Z is a singlet chiral superfield and is a mass parameter of order M M5 . This

potential gives the vacuum expectation value Z = ekR 2 2 , breaking supersymmetry and the U (1)R symmetry (to the Z2,R subgroup). This breaking does not destroy
the successful prediction relating the low-energy gauge couplings, although it causes a
distortion of the spectrum.
The masses for the 321 and PS gauginos, 321 and PS , are generated through the operators on the TeV brane of the form d 2 Z Tr[W W ] + h.c. Since the gauge symmetry on
the TeV brane is 422, we have three independent coefficients C , L and R for these operators, corresponding to the SU(4)C , SU(2)L and SU(2)R factors, respectively. Specifically,
the operators are given by

S=
R
d x dy 2(y R)
4

A
Z Tr WA WA + h.c. , (12)
d 2
2M
A=C,L,R
where A = C, L, R denotes SU(4)C , SU(2)L and SU(2)R . This implies that the masses for
PS PS
PS
the 321 gauginos 321
a (a = 1, 2, 3) and the PS gauginos U , E and S are all determined

in terms of four parameters C M , L M , R M and M /k , where M ekR 2 /M .
PS
PS
Here PS
U , E , and S transform under 321 as (3, 1)2/3 + (3 , 1)2/3 , (1, 1)1 + (1, 1)1 ,
and (1, 1)0 , respectively. Moreover, if leftright symmetry is unbroken on the TeV brane,
which is natural if leftright symmetry is broken spontaneously only on the Planck brane,
we have an additional relation L = R . In this case the masses for the above gauginos are
all determined by the three parameters C M , L M and M /k .
Solving the equations of motion in 5D, we find that the masses for the SU(3)C and
SU(2)L gauginos (a = 3 and 2, respectively) are given by
J0
Y0
mn
k
mn
k
+
+
mn
gB2
2 mn J1 k
g 0,a
mn
gB2
2 mn Y1 k
g 0,a
mn
2
k + gB M,a J1 k

,
Y0 mk n + gB2 M,a Y1 mk n
J0
mn
(13)
where M,3 = C 2 /M and M,2 = L 2 /M . The masses for the SU(3)C and SU(2)L
gaugino towers are given as the solutions to this equation, which can be mn < 0 as well as
mn > 0 (the physical masses are given by |mn |). The masses for the PS gauginos PS
U and
101
PS
E are similarly given by
J1
Y1
mn
mkn =
k
mn
2
k gB M,A J1 k

,
Y0 mk n gB2 M,A Y1 mk n
J0
mn
(14)
PS
2
where A takes U and E for PS
U and E respectively, and M,U = C /M and M,E =
2
R /M . An interesting point is that for large supersymmetry breaking (i.e., large )
the lowest 321
and 321
gauginos both become pseudo-Dirac states with the masses
3
2
PS
1/2

(2/kR) k k /4, while the lowest PS
U and E modes become very light with the
2
2

masses given by (2/gB M,U )k and (2/gB M,E )k , respectively. This can be easily
understood by noticing that the form of Eq. (13) and Eq. (14) are identical, respectively,
to the equations determining the masses of the 321 and SU(5)/321 gauginos in the model
321
of [4,14]. For small supersymmetry breaking, the lowest modes of 321
3 and 2 (i.e., the
PS
PS

MSSM gauginos) have masses much smaller than k , while U and E do not have a
mode lighter than k .
PS
The masses for the U (1)Y gaugino, 321
1 , and the PS gaugino S obey a somewhat more
complicated equation, since they generically mix with each other. The gauginos 321
1 and
are
associated
with
two
U
(1)
factors
arising
from
SU(4)
SU(2)
:
U
(1)
321
PS
C
R
Y
S
and U (1) , respectively. Taking the two U (1)s to be orthogonal, U (1) is the fiveness
charge arising as U (1) SO(10)/SU(5) in the standard GUT embedding. The generators
for U (1)
, Y and QS , normalized
are given
Y and U (1)
in theCSO(10)
covariant manner
C
C
by Y = 2/5 T15
+ 3/5 T3R and QS = 3/5 T15
+ 2/5 T3R , where T15
(T3R ) is a
generator of SU(4)C (SU(2)R ) that commutes with 321. The masses for these gauginos
are then given by the equation

J0 mkn
J1 mkn
mn
mn
mn
mn
JS
JY
m YY
mn YS

n
k
k
k
k
Y1 k
Y0 k

J0 mkn
J1 mkn
mn
mn
mn
mn
JM
m YM
JM
mn YM
= 0,

n
k
k
k
k
Y1 k
Y0 k
(15)
where J0 (mn /k), JY (mn /k ), JS (mn /k ) and JM (mn /k ) are defined by

gB2
mn
mn
mn
J0
J0
+ 2 mn J1
,
k
k
k
g 0,1

mn
mn
mn
2
J0
+ gB M,1 J1
,
JY

k
k
k

mn
mn
mn
2
J
+
g
,
M
J
JS
0
B ,S 1

k
k
k

mn
mn
2
JM
gB M,M J1
,
k
k
(16)
102
and similarly for Y0 (mn /k), YY (mn /k ), YS (mn /k ) and YM (mn /k ). Here M s are given
by
2
2

2
3
3
2
M,1 =
C + R
C + R
,
M,S =
,
5
5
M
5
5
M
2

6
6
C +
R
.
M,M =
5
5
M
In the supersymmetric limit, Eq. (15) gives two decoupled KK towers for each of 321
1
and PS
S , reproducing the spectrum given in Eq. (8). When supersymmetry is broken the
two towers mix, but for small supersymmetry breaking the resulting tower can still be
and PS
effectively described by the sum of the two independent towers for 321
1
S . The
321
lightest state is almost purely 1 with the mass given as the lowest solution of Eq. (13)
with a = 1, and all the other states are heavier than k (the mixings are not negligible for
the excited states). With an increased strength for supersymmetry breaking, the mixing
among the states becomes more important, giving, for example, a non-negligible effect on
the mass of the lightest state. For very large supersymmetry breaking, the lowest state is a
Majorana fermion with the mass given by

M,1
M
2k
= 2 3 1 + 2 1
k;
2

2
5 C
5 R gB2 2
gB M,S M,1 M,M
the next state is a pseudo-Dirac fermion with the mass g1 k (2/gB2 k)1/2 0.2 k .
The effects of supersymmetry breaking on the XY states are similar to those on the
SU(5)/321 states in the model of [5]. Before supersymmetry breaking, the massless XY
states consist of two Dirac fermions XXY and QXY , and four sets of real scalars XXY ,
XY , AXY and AXY , where the subscripts X and Q represent the (3, 2)
Q
5/6 + (3 , 2)5/6
5,X
5,Q
and (3, 2)1/6 + (3 , 2)1/6 components under the 321 decomposition. The masses for XXY
and QXY are generated through the operator

S=

R

d x dy 2(y R) e2kR d 4
4

Z Tr P[A]P[A] + h.c. , (17)

2M
where

A eV y eV + y eV eV 2 eV eV 2 eV eV .
(18)
Here, the trace is taken over SO(10) space and P[X ] is a projection operator: with X an
adjoint of SO(10), P[X ] extracts the (6, 2, 2) component of X under the decomposition
to 422. The coefficient is a dimensionless parameter. Since the X and Q components are
embedded in a single 422 multiplet (6, 2, 2), the masses of their fermionic components are
determined by this single coefficient. The equation determining the XXY and QXY masses
is given by

J1 mkn
J1 mk n gB2 M,X J0 mk n
= m
,
(19)
Y1 mkn
Y1 k n gB2 M,X Y0 mk n
103
where M,X 2 /M . Note that the masses for the XXY and QXY towers are degenerate
at tree level, although they split at loop level through the 321 gauge interactions (these
splittings are finite and calculable as the XY towers are localized to the TeV brane while the
XY are generated
breaking of 422 resides at the Planck brane). The masses for XXY and Q
through the operator

S=

R

d x dy 2(y R) e2kR d 4
4

Z
Z
Tr
P[A]P[A]
,
4M2
(20)
where the consistency of the effective theory requires to take the form = 8gB2 ||2 (0)
+ , where is a dimensionless parameter [5]. The equation determining the XXY and
XY masses is then given by
Q
J1
Y1
mn
mkn =
k
J1
Y1
mn
k
k
mn
2 k
mn
gB2 M,X
mn k J0 k
2 k
mn ,
gB2 M,X
Y
0
mn k
k
(21)
2 ||4 /M 2 . Again the masses for the XY and XY towers are degenerate
where M,X
X
Q
XY
at tree level, although they split at loop level. The masses of AXY
5,X and A5,Q are not generated by the operators in Eqs. (17), (20). In fact, the 5D gauge invariance forbids any local
XY
operator giving these masses (in the 4D dual picture the zero modes of AXY
5,X and A5,Q are
pseudo-Goldstone bosons associated with the SO(10) 422 breaking at the TeV scale,
which encodes the TeV-brane gauge breaking in 5D). These masses, however, are generated at loop level, picking up the effects of both Planck-brane and TeV-brane breakings.
The resulting masses are finite and approximately given by
m2AXY
5
g2 C 2
k
m
ln
,

XY
4
m XY
(22)
where g represents a 4D gauge coupling and C the group theoretical factor. The masses
XY
for AXY
5,X and A5,Q are different, because they have the different 321 quantum numbers
and thus different values of Cs. This difference, however, will be small, since the quantum
XY
numbers for AXY
5,X and A5,Q are the same under SU(3)C and SU(2)L so that the mass
difference only comes from the U (1)Y part, which is expected to give a mass splitting of
O(10%).
The Higgs spectrum is identical to the 321321 model, except that the two SU(5) multiplets, 5 and 5 , of the 321321 model are now unified into a single multiplet, 10 of SO(10).
As in [5], the unwanted zero modes in H c acquire TeV-scale masses through their tree-level
couplings to the supersymmetry breaking on the TeV brane. The Higgs spectrum of our
model can be obtained from the expressions given in [5] by setting cH = cH , HD = HD
and HT = H T . Note, however, that the Planck-brane kinetic terms can still be different
for the up-type and down-type Higgs fields, i.e., zH = zH , due to the gauge breaking on
the Planck brane. This could be important for obtaining the correct electroweak symmetry
breaking vacuum, depending on the details of the Higgs sector.
It is useful here to consider a limit of small supersymmetry breaking M and to see
what the spectrum looks like. In this limit we can expand Eqs. (13), (14), (15), (19), (21)
104
in powers of /M . We then find that the PS gauginos do not have any light mode with
the mass smaller than k , while the 321 gauginos, 321
a (a = 1, 2, 3), and the XY gauginos
and scalars, ZXY and ZXY (Z = X, Q), do with the masses approximately given by

m321
= ga2 M,a
,
a
(23)

m XY = 2gB2 kM,X
,
(24)
2
= 2gB2 kM,X
,
(25)
m2 XY
Z

where M,a
M,a ekR , M,X
M,X ekR and M,X
M,X ekR are parame2
2
1/2
are the 4D gauge couplings. Conters of order TeV, and ga (R/gB + 1/g0,a )
2
sidering ga = O(1) and gB k = O(kR), we expect that the XY states are generically
heavier than the 321 gauginos. In fact, in the case of M,a M,X M,X /4 , as
suggested by naive dimensional analysis, the ratios of the masses are roughly given by
: m XY : m XY 1 : kR : kR. The masses for AXY
m321
5,Z are given by Eq. (22) so they
a
Z
could be somewhat lighter than ZXY and ZXY . Similarly in the Higgs sector, the coloredtriplet states are generically heavier than the doublet states. These little mass hierarchies
among the TeV states arise because the wavefunctions for the exotic states are localized
to the TeV brane, where supersymmetry breaking occurs, while those of the MSSM states
are not. This effect is therefore related to the models successful prediction relating the
low-energy gauge couplings, which crucially relies on the fact that all the exotic states are
strongly localized to the TeV brane [5]. The spectrum for small supersymmetry breaking
is depicted in Fig. 1(b) for the gauge sector.
In the small supersymmetry breaking limit, the masses of the 321 gauginos, 321
a , are
given by

3
2
M2 = g22 L M ,
M3 = g32 C M ,
M1 = g12 C + R M ,
(26)
5
5
where M1 m321 , M2 m321 , and M3 m321 are the bino, wino, and gluino masses,
1
and M = ekR 2 /M is a parameter of order TeV. A particularly interesting case is

where leftright symmetry is unbroken on the TeV brane. In this case L = R , so that we
have a non-trivial relation among the 321 gaugino masses, which can be written as
M1 2 M3 3 M2
=
+
.
5 g32
5 g22
g12
(27)
Note that, in contrast to high-scale supersymmetry breaking scenarios, this relation arises
from the physics at an energy scale of order TeV as a threshold effect.7
We finally discuss the squark and slepton masses. They are generated at one-loop level
through the standard model gauge interactions. Because of the geometrical separation between supersymmetry breaking and the place where squarks and sleptons are located, the
7 The relation (27) is essentially determined by the symmetry of the theory 422 is preserved in the G sector
and its breaking comes only through the gauge kinetic terms of the gaugino fields. Thus, Eq. (27) holds accurately
even for relatively small values of M / k , i.e., it is not subject to errors of O( k /M ) coming from unknown
TeV-brane operators (note that Ma here are the running masses and not the physical pole masses).
105
generated squark and slepton masses are finite and calculable in the effective field theory.
Although the remaining gauge symmetry on the Planck brane after the orbifolding is 422,
the squarks and sleptons interact effectively only with the 321 gauge multiplet at the scale
where their masses are generated (due to the spontaneous breaking of 422 at a high scale).
This means that the squark and slepton masses in the present model are given by
m2f =
1
f
F a1 Ca Ia ,
2 2
(28)
a=1,2,3
l,
e represents the MSSM squarks and sleptons, and the Caf are
where f = q,
u,
d,
f
the group theoretical factors given by (C1 , C2 , C3 ) = (1/60, 3/4, 4/3), (4/15, 0, 4/3),
l and e,
(1/15, 0, 4/3), (3/20, 3/4, 0) and (3/5, 0, 0) for f = q,
u,
d,
respectively. The
functions Ia are defined in Eq. (21) of [14], where we have to use three different gaugino mass parameters M,a because of the non-universal gaugino masses.8 The quantity F ,
which represents a mixing effect between the U (1)Y and U (1) gaugino towers, is a function of M,1 , M,S and M,M and takes a value of order 1. Since the mixing effect vanishes
for small supersymmetry breaking, F approaches to 1 for small M s. Note that because
the squark and slepton masses are generated through the gauge interactions, they are flavor
universal and the supersymmetric flavor problem is absent. Small mass splittings among
different generations arise through the Yukawa couplings at two loop orders, making the
masses for the third generation squarks and sleptons slightly lower than those for the fist
two generation ones. However, they do not generate flavor changing neutral currents at a
dangerous level.
The spectrum described above provides a rich phenomenology. For example, it gives a
variety of possibilities for the next-to-lightest supersymmetric particle (NLSP) (the lightest supersymmetric particle is the gravitino with the mass k 2 /MPl 0.010.1 eV).
Possible patterns for the superparticle spectrum are similar to those discussed in [5].
XY
The model also predicts relatively light GUT particles AXY
5,X and A5,Q , transforming as
(3, 2)5/6 + (3 , 2)5/6 and (3, 2)1/6 + (3 , 2)1/6 under 321, respectively. The masses for
XY
AXY
5,X and A5,Q are close but expected to have a relative splitting of O(10%). The lighter
one will presumably be AXY
5,Q , which is stable for collider purposes due to the conservation
of SU(3)C charges and the location of fields [5]. Once AXY
5,Q is produced, it hadronizes to
either of four fermionic mesons T 0 , T 0 , T + , T + (and their anti-particles), depending
on whether it picks up an up or down quark or anti-quark. All these states are sufficiently
long-lived, so that the charged ones would be detectable through highly ionizing tracks. For
strong supersymmetry breaking (large ), the gauginos of the PS multiplet UPS , EPS and
SPS , transforming as (3, 1)2/3 + (3 , 1)2/3 , (1, 1)1 + (1, 1)1 and (1, 1)0 , also become
light (see Fig. 1(c) for the overall spectrum of the gauge sector for large supersymmetry
breaking). Some of these states could also be stable and seen at colliders.
8 The expressions for squark and slepton masses, Eq. (28), are subject to errors of O( k /M ) arising from
the TeV-brane localized 321 gauge kinetic terms (only errors of O(( k /M )2 ) are mentioned in [14]).
106
3. Model without TeV-brane symmetry breaking

In this section we construct a model in which the bulk gauge symmetry is not reduced
at the TeV brane. The construction closely follows that of the previous section. The model
presented here can be viewed as an extension of the SU(5) model of [4] to a larger unified
gauge group.
The model is again formulated in the 5D warped spacetime compactified on S 1 /Z2
(0 y R), with the metric given by Eq. (1). The parameters M5 , M , k and R take
similar values. The gauge group in the bulk is taken to be SO(10), with the boundary
conditions given by

V
P V P 1
x ,y ,
x , y =
P P 1

V
V
x , y =
x ,y ,
(29)
where y = y R. The matrix P is the same as before: P = 0 diag(1, 1, 1, 1, 1).

This reduces the gauge group at the Planck brane to 422, but leaves the gauge group at
the TeV brane to be SO(10). The resulting zero modes are only the (15, 1, 1) + (1, 3, 1) +
(1, 1, 3) component of V (the 422 N = 1 gauge supermultiplet). The excited KK states all
have masses of order k or larger.
The Higgs fields are introduced in the bulk as a hypermultiplet transforming as 10 of
SO(10), with the boundary conditions given by

P H
H
x , y =
x ,y ,
PHc
Hc

H
H
x , y =
x ,y .
(30)
H c
Hc
This leaves the (1, 2, 2) component of H as zero modes. All the other modes have masses
of order k or larger.9
The matter and the 422 gauge breaking sectors are identical to those in the model of
Section 2. The quark and lepton chiral superfields are introduced on the Planck brane in
the (4, 2, 1) + (4 , 1, 2) representation of 422 for each generation.10 The 422 gauge
breaking is introduced on the Planck brane. The simplest possibility is the Higgs breaking
9 We can alternatively put the Higgs field on the Planck brane in the (1, 2, 2) representation of 422, or in
the bulk but with the boundary conditions of Eq. (30) with an extra minus sign in the right-hand side of the
second equation (cf. footnote 1). The latter case gives two triplet zero modes from H c and four relatively light
doublet modes from H and H c (two for each) in the supersymmetric limit (the doublet states are even exponenlighter than k for cH > 1/2). A realistic model is then obtained by introducing a mass term of the form
tially
d 2 H c H c + h.c. on the TeV brane. The MSSM prediction for gauge coupling unification is preserved in these
cases, too (for cH 1/2 in the case of the bulk Higgs).
10 In the present model, (some of) matter fields could be introduced in the bulk as hypermultiplets transforming
as 16 of SO(10) (a hypermultiplet transforming as 16 of SO(10) yields either or as a zero mode, depending
on the boundary conditions). The prediction for the low-energy gauge couplings is the same as that in the brane
matter case if the bulk mass parameters, cM , for matter fields take the values cM 1/2.
107
= = O(k), where and transform as (4, 1, 2) and (4 , 1, 2) under 422, respectively. We introduce the Planck brane couplings Eq. (4), which give the Yukawa couplings
for the quarks and leptons as well as Majorana masses for the right-handed neutrinos.
With suitable non-renormalizable operators realistic quark and lepton masses and mixings
are reproduced. For , k M , the Yukawa couplings for the third generation
fermions are approximately unified at the scale k, giving a successful low-energy prediction for mb /m . It also leads to tan 50. Small masses for the observed neutrinos are
naturally obtained through the see-saw mechanism. The model possesses a U (1)R symme
(broken to Z2,R at the TeV brane
try: V (0), (0), H (0), H c (2), (1), (1), (0), (0)
throughsupersymmetry breaking), which prevents potentially dangerous operators such
as (y) d 2 HD2 + h.c. There is not a proton decay problem for Planck-brane localized
matter.
The 422 breaking at the Planck brane gives masses for the 422/321 component of V .
The resulting spectrum for the gauge sector is then given by Eq. (5) for the 321 component
and by Eq. (6) for the SO(10)/321 component (with the comment in footnote 4 applying
to 422/321). The spectrum, therefore, can be summarized as
321
V : m0 = 0,
321 321
GUT GUT

(31)
V ,
+ V
: mn n 14 k ,
,
where n = 1, 2, . . . . Here, the superscript GUT represents the SO(10)/321 component,
whose transformation properties are (3, 2)5/6 + (3 , 2)5/6 + (3, 2)1/6 + (3 , 2)1/6 +
(3, 1)2/3 + (3 , 1)2/3 + (1, 1)1 + (1, 1)1 + (1, 1)0 under the 321 decomposition. Note
that the spectrum for the excited KK towers are approximately SO(10) symmetric. This
is because in the 4D dual picture the global SO(10) symmetry of G is not dynamically
broken so that the composite states of G, which are identified as KK towers in 5D, are
approximately SO(10) symmetric.
As in the model of Section 2, the MSSM prediction for low-energy gauge couplings is preserved. Specifically, low-energy 321 gauge couplings are given by setting (T1 , T2 , T3 )(V++ ) = (0, 2, 3), (T1 , T2 , T3 )(V+ ) = (8, 6, 5) and (T1 , T2 , T3 )(V+ ) =
(T1 , T2 , T3 )(V ) = (0, 0, 0) in Eq. (9) of Ref. [4] and adding the Higgs contribution,
which is identical to that of the SU(5) model of [4] with cH = cH (5) = cH (5 ) . Here, cH
is the dimensionless bulk mass parameter for the Higgs multiplet, {H, H c }. Therefore, for
cH 1/2 we find that the prediction for low-energy 321 gauge couplings ga is given by
Eq. (9) with Eq. (10), reproducing the successful MSSM prediction.
Supersymmetry breaking is introduced on the TeV brane through the potential Eq. (11).
This gives the vacuum expectation value Z = ekR 2 2 , breaking supersymmetry
and the U (1)R symmetry. The breakings are transmitted to various fields through the operator

S=

R
d x dy 2(y R) d 2
4

Z Tr W W + h.c. ,
2M
(32)
where the trace is taken over SO(10) space. Since the TeV brane respects full SO(10), the
coefficient is universal for the entire gauge components. The above operator gives masses
for the 321 gaugino zero modes and modifies the spectrum for the gaugino towers of both
108
321 and SO(10)/321 components. The masses for the 321 and SO(10)/321 gaugino towers
are given, respectively, by Eq. (13) and Eq. (14), but now the gaugino mass parameters
M s take a universal value: M,a = M,A = 2 /M M . The values of the Planckbrane gauge couplings renormalized at the TeV scale are the same as in the previous model:
2 a /8 2 should be used in Eq. (13).
1/g0,a
For the Higgs sector, the only light states are those of the two MSSM Higgs doublets. They obtain supersymmetry-breaking as well as supersymmetry-preserving masses
through the couplings to the Z field on the TeV brane. The resulting masses are identical
to those in the model of Section 2, whose explicit expressions are given in Ref. [5].
Squark and slepton masses are generated at one-loop level through gauge interactions.
Despite the 422 symmetry after the orbifolding, the squark and slepton masses come almost entirely from the 321 gauge loops because 422 is broken at a high scale. The resulting
masses are finite and calculable, due to the spatial separation between supersymmetry
breaking and the matter location, and are given by
m2f =
1 f
Ca Ia ,
2 2
(33)
a=1,2,3
l,
e,
where f = q,
u,
d,
and Ca are the group theoretical factors given below Eq. (28). The
functions Ia are defined in [14], where we have to use the gaugino mass parameter M
for all a = 1, 2, 3. Because of the universality of the operator Eq. (32), the effect of the
U (1)Y -U (1) mixing is negligible. The supersymmetric flavor problem is absent because
the squark and slepton masses are flavor universal (up to small higher-order corrections
from the Yukawa couplings).
The phenomenology of the present model is similar to that of the model in [4]. In fact,
the masses for all the superparticles as well as the KK towers are all determined in terms
of only two parameters M /k and k , up to parameters associated with the Higgs sector
[14]. The main difference is that the SU(5)/321 gauge multiplet of [4] is now replaced by
the SO(10)/321 multiplet. Therefore, the light exotics now consist of five components
transforming as (3, 2)5/6 + (3 , 2)5/6 , (3, 2)1/6 + (3 , 2)1/6 , (3, 1)2/3 + (3 , 1)2/3 ,
(1, 1)1 + (1, 1)1 , and (1, 1)0 under 321. In particular, in the case of large supersymmetry
breaking the gauginos for all these components become light, thus providing the possibility
of testing the underlying enlarged group structure of the model at the LHC.11
4. Conclusions
In this paper we have seen that the PatiSalam (422) gauge group fits extremely naturally into the framework of warped unification. The two models we have constructed both
11 In the present model, the Z GUT parity of [4] is extended to two Z parities: one which acts non-trivially
2
2
on the SO(10)/422 component of the gauge multiplet and the colored Higgs states and the other which acts nontrivially on the SO(10)/(SU(5) U (1) ) component of the gauge multiplets, i.e., the U and E of PS and Q of
XY. These symmetries ensure the quasi-stability for the lightest of XXY and QXY and of QXY , UPS and EPS ,
respectively.
109
have a bulk SO(10) gauge symmetry broken by boundary conditions to 422 on the Planck
brane, corresponding in the 4D picture to a 422 gauge theory with an approximate global
SO(10) symmetry. They differ in whether the full SO(10) is realized on the TeV brane, or
equivalently, in whether or not the global SO(10) is spontaneously broken in the infrared
by the strong dynamics of the conformal sector.
These models incorporate matter unification in a very economical way, with a full generation of quarks and leptons transforming as (4, 2, 1) + (4 , 1, 2) under 422. They also
require only a very simple Higgs sector for breaking of the 422 gauge symmetry and easily
accommodate the see-saw mechanism for neutrino masses, realistic fermion masses, and
bottom-tau unification. At the same time they explain the successful unification of gauge
couplings in the MSSM, something that PatiSalam unification by itself does not do. The
same prediction relating the low energy couplings given in the MSSM applies here to good
approximation under the assumption that the gauge interactions grow strong in the ultraviolet. The SO(10) symmetry plays an important role in this prediction; in the 4D description,
it ensures that the contribution to the evolution of the gauge couplings from the conformal
sector is universal.12
The phenomenology of these models is quite different from conventional supersymmetric unification, and different as well from models of warped unification built on SU(5)
symmetry. In both models a rich array of exotic particles appears near the TeV scale:
these include supermultiplets with 321 quantum numbers (3, 2)5/6 + (3 , 2)5/6 as in
warped SU(5), but also states with quantum numbers (3, 2)1/6 + (3 , 2)1/6 , (3, 1)2/3 +
(3 , 1)2/3 , and color-neutral states transforming as (1, 1)1 + (1, 1)1 + (1, 1)0 . In the
model with SO(10) broken to 422 on the TeV brane, some of these states are massless
in the supersymmetric limit. The prospects for producing these particles at future colliders such as the LHC depend on the scale k , the strength of supersymmetry breaking on
the TeV brane, and in the model with TeV-brane symmetry breaking, on the free parameters that determine how strongly the pseudo-Goldstone multiplet feels the supersymmetry
breaking ( and of Eqs. (17), (20)).
The spectrum of MSSM superparticles differs between the two models. In the model
with TeV-brane symmetry breaking, the gaugino mass terms on the TeV brane are nonuniversal, although there is the possibility of one relation among the 321 gaugino masses
if leftright symmetry is unbroken on the TeV brane. One interesting point is that in the
TeV-brane symmetry breaking model, there is generally a mixing between the bino and
the gaugino associated with the other U (1) factor contained in SU(4)C SU(2)R . The
effect of this mixing is small for weak supersymmetry breaking, but becomes significant as
the supersymmetry breaking is increased. Because of the non-universality in the gaugino
masses, there is a broad range of possibilities for what the gaugino and scalar spectrum
will look like, and in particular, there are many possibilities for what the NLSP will be, as
discussed in [5]. The model without TeV-brane symmetry breaking, on the other hand, has
a more constrained spectrum due to the universal gaugino mass terms on the TeV brane.
12 It is worth noting that the SO(10)-symmetric conformal sector can be replaced by extra vector-like states,
such as (4, 2, 1) + (4 , 2, 1) and (4, 1, 2) + (4 , 1, 2) with TeV-scale masses, without spoiling many of the desired features of the model. This provides a class of purely 4D 422 theories with the successful gauge coupling
prediction arising from strong coupling in the ultraviolet [15].
110
The spectrum of MSSM particles has similar features to that of [4,14] in this case, but it
should be stressed again that the spectrum of exotic particles is different.
Acknowledgement
The work of Y.N. was supported in part by the Director, Office of Science, Office of
High Energy and Nuclear Physics, of the US Department of Energy under Contract DEAC03-76SF00098.
References
[1] H. Georgi, S.L. Glashow, Phys. Rev. Lett. 32 (1974) 438.
[2] H. Georgi, in: C.E. Carlson (Ed.), Particles and Fields, AIP, New York, 1975, p. 575;
H. Fritzsch, P. Minkowski, Ann. Phys. 93 (1975) 193.
[3] J.C. Pati, A. Salam, Phys. Rev. D 10 (1974) 275.
[4] W.D. Goldberger, Y. Nomura, D.R. Smith, Phys. Rev. D 67 (2003) 075021, hep-ph/0209158.
[5] Y. Nomura, D. Tucker-Smith, B. Tweedie, hep-ph/0403170.
[6] T. Asaka, W. Buchmuller, L. Covi, Phys. Lett. B 523 (2001) 199, hep-ph/0108021;
L.J. Hall, Y. Nomura, T. Okui, D.R. Smith, Phys. Rev. D 65 (2002) 035008, hep-ph/0108071;
R. Dermisek, A. Mafi, Phys. Rev. D 65 (2002) 055002, hep-ph/0108139.
[7] L. Randall, R. Sundrum, Phys. Rev. Lett. 83 (1999) 3370, hep-ph/9905221.
[8] S.F. King, Phys. Lett. B 325 (1994) 129;
S.F. King, Phys. Lett. B 325 (1994) 538, Erratum.
[9] A. Pomarol, Phys. Rev. Lett. 85 (2000) 4004, hep-ph/0005293.
[10] L. Randall, M.D. Schwartz, Phys. Rev. Lett. 88 (2002) 081801, hep-th/0108115;
L. Randall, M.D. Schwartz, JHEP 0111 (2001) 003, hep-th/0108114;
W.D. Goldberger, I.Z. Rothstein, Phys. Rev. Lett. 89 (2002) 131601, hep-th/0204160;
W.D. Goldberger, I.Z. Rothstein, Phys. Rev. D 68 (2003) 125011, hep-th/0208060;
K.W. Choi, H.D. Kim, I.W. Kim, JHEP 0211 (2002) 033, hep-ph/0202257;
K.W. Choi, H.D. Kim, I.W. Kim, JHEP 0303 (2003) 034, hep-ph/0207013;
K. Agashe, A. Delgado, R. Sundrum, Nucl. Phys. B 643 (2002) 172, hep-ph/0206099;
R. Contino, P. Creminelli, E. Trincherini, JHEP 0210 (2002) 029, hep-th/0208002;
K.W. Choi, I.W. Kim, Phys. Rev. D 67 (2003) 045005, hep-th/0208071.
[11] See, for example, L.J. Hall, Y. Nomura, Phys. Rev. D 65 (2002) 125012, hep-ph/0111068.
[12] Z. Chacko, M.A. Luty, E. Ponton, JHEP 0007 (2000) 036, hep-ph/9909248.
[13] T. Gherghetta, A. Pomarol, Nucl. Phys. B 586 (2000) 141, hep-ph/0003129.
[14] Y. Nomura, D.R. Smith, Phys. Rev. D 68 (2003) 075003, hep-ph/0305214.
[15] M. A. Luty, Y. Nomura, D. Tucker-Smith, in preparation.

On-shell gauge invariants and field strengths

in open superstring field theory
Yoji Michishita
Theory Division, CERN, CH-1211 Geneva 23, Switzerland
Received 8 July 2004; accepted 28 July 2004
Abstract
We study gauge invariant quantities in the open superstring field theory proposed by Berkovits,
extending the precedent discussion in bosonic string field theory. Our gauge invariants are on-shell.
As its applications, we define quantities which are expected to be related to the U(1) field strength
a RR coupling and a component of the string field equation of motion, and consider their naive
extensions to off-shell. Order by order calculations show that the field strength extracted from the
RR coupling is not gauge invariant, while from the component of the equation of motion we obtain
an off-shell field strength which is gauge invariant under full gauge transformation if on-shell, and
under linearized gauge transformation even off-shell.
PACS: 11.25.-w; 11.25.Sq
1. Introduction
Open string field theory has been developed in the last several years, mainly in the study
of tachyon condensation. However, it is not yet a tool as useful as first quantized formalism
when we consider other phenomena. Further development is necessary to make it as useful
as first quantized formalism. Especially, in this theory it is very difficult to extract physical
E-mail address: yoji.michishita@cern.ch (Y. Michishita).

doi:10.1016/j.nuclphysb.2004.07.043
112
Y. Michishita / Nuclear Physics B 698 (2004) 111131
quantities, because of its computational complexity and nonlocality. It is helpful to know

how to construct physically meaningful quantities.
String field theory is a kind of gauge theory, and gauge invariant quantities in gauge
theories have some physical meanings. Therefore, we want to construct gauge invariant
quantities in string field theory. Studies along this direction in bosonic string field theory
have been advanced in [13]. These works have shown that, roughly speaking, couplings
of one closed string mode and one open string field are gauge invariant. In this paper we
study these gauge invariants in open superstring field theory proposed by Berkovits [4].
This gauge invariant is on-shell because of the on-shell property of the closed string
mode.
The simplest gauge invariant quantity in U(1) gauge theory is the field strength of the
gauge field. Since in lowest order approximation U(1) open string field theory is equivalent
to U(1) gauge theory, it is natural to consider string field theory analog of the field strength.
The success of the analysis of tachyon condensation [5] in boundary string field theory
(BSFT) [6] also suggests that we can define such an analog, because gauge field in BSFT
transforms in the same way as ordinary gauge field, and the BSFT action is expected to be
related to cubic string field theory and its superstring extension by some field redefinition.
As applications of our gauge invariant quantities, we make two attempts to extract gauge
invariant field strength. Firstly, we consider string field theory counterpart of the coupling
of RamondRamond (p 1)-form and one field strength of Dp-brane in the effective
action. From it we extract quantities analogous to gauge invariant field strength and gauge
field, and extend it to off-shell in the most naive and straightforward manner. It is not clear
from the general expression that the analog of gauge field transforms in the same way as
ordinary one. To make it clear and to investigate how far this quantity can be extended to
off-shell, we give the component expression of this quantity up to level 2. We find that
the field strength is not gauge invariant at level 2, even on-shell, and the gauge field does
not transform in the same way as ordinary gauge field.
Secondly, we extract a gauge invariant component of the string field equation of motion. The string field equation of motion contains a gauge invariant extension of that of
ordinary gauge field, i.e., F = 0. This string field theory counterpart has correction
terms from massive fields. We compute those terms up to level 1. From this gauge invariant component we can extract an analog of field strength, and at the linearized level this
component is gauge invariant even off-shell.
For definiteness, in this paper we consider one single D9-brane in type IIB theory. Extension of our discussion to lower-dimensional D-branes is straightforward, but is restricted
to one single D-brane. In Section 2, we introduce couplings of one closed string mode and
one open string field which are gauge invariant in Berkovits open string field theory. Our
discussion is entirely in terms of conformal field theory. In Section 3, we consider coupling of RR 8-form and one open string field, and give explicit expression up to level 2. We
show that field strength defined from it is not gauge invariant even on-shell. In Section 4,
we define another type of on-shell gauge invariant which reduces to the previous ones,
and define gauge invariant components of the equation of motion. From one of them we
extract field strength and gauge field, and show that the field strength is gauge invariant
under linearized gauge transformation even off-shell. Section 5 contains discussions. In
the Appendix A bases for expanding string fields are tabulated.
113
2. On-shell gauge invariants

In this section we introduce gauge invariant quantities linear in string field , extending
the argument in [13] for bosonic open string field theory. Our argument is based on the
conformal field theory viewpoint. If necessary, we can rewrite the following argument in
terms of oscillator expression as in [2], in the case of a flat background.
The action of the NS part of the open string field theory proposed by Berkovits [4] is

1
S = 2 e QB e e 0 e
2g

1

t t t
e
,
t e
QB et , et 0 et
dt e
(1)
0
where the string field is Grassmann even, GSO(+), ghost number 0 and picture number 0 operator. CFT correlators are defined in large Hilbert space. For details of the
definition see, for instance, [7].
This action is invariant under the following gauge transformation:
e = (QB )e + e (0 ),
(2)
where the gauge transformation parameters and are Grassmann odd, GSO(+) operators. Their ghost number and picture number are (1, 0) and (1, 1), respectively. From
this expression, we can compute the transformation of order by order:
1

1
(QB ), + , (0 )
2
2
1
1 2
1
2
+ (QB ) (QB ) + (QB )
12
6
12
1 2
1
1
+ (0 ) (0 ) + (0 ) 2 + .
12
6
12
This transformation contains infinitely many terms with arbitrarily high power in .
The gauge invariant quantity (V ; ) we consider in this paper is defined as:

(V ; ) V (0, 0) f1(1) (0) disk,

= (QB ) + (0 ) +
(3)
(4)
where disk is the CFT correlation function evaluated on a unit disk with appropriate
(n)
boundary condition on the boundary. Conformal mappings fk (z), which are used to define star products, are
2/n

(n)
2i(k1)/n 1 + iz
.
fk (z) = e
(5)
1 iz
V (z, z ) is a closed string vertex operator that satisfies the following conditions:
[QB , V (z, z )] = 0, i.e., V (z, z ) is BRST invariant;
[0 , V (z, z )] = 0, i.e., V (z, z ) is in small Hilbert space;
V (z, z ) is a dimension (0, 0) primary field.
114
Ordinary on-shell closed string vertex operators in first quantized formalism satisfy all of
these conditions.
(1)
There is a subtlety in the definition of (V ; ). The mapping f1 is singular at the center
of the unit disk where V is inserted. Geometrically, this mapping glues the left half and the
(1)
right half of the open string and therefore the midpoint is singular. However, in (4), f1
acts only on at the point where this mapping is not singular. Hence we can forget that
the unit disk is formed by the gluing procedure and can evaluate (4) regarding it as merely
a CFT correlation function on the unit disk. Then (4) is not singular and well-defined.
We may have to take conformal transformations of (V ; ), which is singular at the
point where V is inserted. For such cases, we can define (V ; ) by taking a limit:

(1)
f1 (0) disk .
(V ; ) lim V (, )
(6)
0
Since the gauge transformation of is given implicitly by (2) and is quite different
from the bosonic string field theory counterpart = QB + , at first
glance it is not clear that (V ; ) is gauge invariant. However, for proving gauge invariance
it is sufficient to show the following relation:
(V ; A B) = (V ; B A).
(7)
Since no explicit explanation for this relation in CFT viewpoint has not been given in the
literature, we will give one shortly. In fact this is valid for any dimension (0, 0) primary
operator V , not necessarily BRST invariant or in small Hilbert space. In [2] this has been
proved for the case where V is a tachyon vertex operator in terms of oscillator expression
in flat background.
Now let us proceed assuming this relation. Then the invariance of (V ; ) can be
proved as follows. In general is not equal to e e , but thanks to (7), (V ; ) =
(V ; e e ). Then plugging (2) into this,
(V ; ) = (V ; QB + 0 ).
(8)
So far we did not use the fact that V is BRST invariant and in small Hilbert space, and by
using it, we obtain

(1)
(1)
(V ; ) = QB V (0) f1 (0) + 0 V (0) f1 (0) = 0.
(9)
We now make some comments on the properties of (V ; ).
(V ; ) is on-shell in the following sense. For example, in flat background momentum q of V satisfies the on-shell condition q 2 = const, because of the BRST invariance
or the condition of (0, 0) conformal dimension. Therefore, by momentum conservation
q + k = 0, the momentum k of also satisfies k 2 = const;
(V ; ) is linear in . In terms of component expression of the string field, it is surprising that all the non-linear terms in the gauge transformation of cancel;
Since V does not have 0 , has to have 0 to obtain a non-zero contribution. Therefore
we can concentrate on only those operators that survive the 0 = 0 gauge condition;
An obvious physical interpretation of (V ; ) is that these quantities represent couplings of one on-shell closed string mode and one open string mode [13].
115
Now we give an explanation for the relation (7) in terms of CFT. We assume that V is a
dimension (0, 0) primary operator. Comparing the following two relations,

(3)
(3)
(3)
A B C = f1 A(0) f2 B(0) f3 C(0) disk

1 (3)
= A(0) f1(3)
f2 B(0) f3(3) C(0)
(2)

(2) (3) 1 (3)
(3)
= f1 A(0) f1 f1
f2 B(0) f3 C(0)
1 (3)

f2 B(0) f3(3) C(0) , (10)

= f1(2) A(0) f2(2) I f1(3)

A (B C) = f1(2) A(0) f2(2) (B C)(0) disk ,

(11)
we obtain the following CFT expression of (V ; A B)
(3) 1 (3)

(1)
(3)
f1 I f1
(V ; A B) = lim V (, )
f2 A(0) f3 B(0) S
0
1 (3)

h f1(1) I f1(3)
= lim V (, )
f2 A(0) f3(3) B(0) disk,
0
(12)
where I (z) = z1 and h(z) = z1/2 . Let us trace the successive actions of conformal
(3)
(3)
mappings in the above expression. By f2 on A and f3 on B, worldsheets of two open
strings are glued together to form a sector with the central angle of 4/3 (Fig. 1). Then
by I (f1(3) )1 it is mapped to two copies of half-disk with the left half arc of one copy
(1)
and the right half arc of the other identified (Fig. 2). Next by f1 they are mapped to two
copies of unit disk. These two disks have cuts along the negative part of the real axis, and
when we go across one of them, we jump to the other. Thus we get the Riemann surface S
in Fig. 3. Finally by h, (V ; A B) becomes a correlation function on a unit disk. Note
Fig. 1. Worldsheets of two open strings are glued together.
116
Fig. 2. The sector is mapped to two copies of half-disk. Identified parts are shown by bold lines.
Fig. 3. Two half-disks are mapped to one double-valued unit disk S. Identified parts are shown by bold and
dashed lines.
that V is invariant under h. At this stage we can take the limit 0 safely and obtain the
following non-singular result:

(V ; A B) = V (0, 0) f1(2) A(0) f2(2) B(0) disk .
(13)
We can easily see that a rotation exchanges the positions of A and B, and leaves
V (0, 0) invariant. (If A and B are Grassmann odd we have an extra sign factor.) Thus we
have established relation (7).
By extending the above argument, we obtain
(V ; 1 2 n )

= V (0, 0) f1(n) 1 (0) f2(n) 2 (0) fn(n) n (0) disk.
(14)
117
From this expression we see that (V ; 1 2 n ) is invariant under a cyclic exchange

of 1 , 2 , . . . , n .
Let us give a comment on the case where V is a dimension (hL , hR ) primary operator. In
this case V is not invariant under h and obtains the factor (/2)hL /2 ( /2)hR /2 . In general
this factor is divergent or vanishing, and makes the second expression of (12) ill-defined.
3. Gauge invariant RR coupling and field strength

In this section we investigate the coupling of a string field and RR field as an application
of the gauge invariant quantities defined in the previous section. From it we extract a string
field theory analog of gauge invariant field strength of ordinary U(1) gauge theory up to
level 2. We will see that this field strength is not gauge invariant, even on-shell.
We consider one single type IIB D9-brane in a flat background, and take RR 8-form
and
vertex operator VRR8 as V . Then (VRR8 ; ) is interpreted as the coupling of RR 8-form

one open string field. We can expect that this corresponds to ChernSimons term C (8)

= F (9) A in the effective action, where A is the ordinary gauge field on the DF
is its field strength. Therefore we can extract a string field theory analog of
brane, and F
by computing (VRR8 ; ).
F
d 10 k
e ikX +B(k)
The lowest level component of is equal to (2)
10 [A (k) c e
cce2 eikX ]. In lowest order calculation, A corresponds to A [8], and the lowest order
term of (VRR8 ; ) should be proportional to F , the field strength of A . It is easy to see
d 10 k
e ikX (1)
that this is indeed the case: (VRR8 ; ) VRR8 (0, 0) (2)
disk +
10 A (k) c e
and noting that c e eikX is the vertex operator corresponding to the gauge field in
first quantized formalism, this CFT correlator represents the coupling of RR 8-form and
gauge field on D-brane in first quantized formalism, and gives C (8) F .
Higher order terms distort the equality of A and A , as can be seen from the gauge
while the transformation
transformation law. A is expected to transform as A = ,
of A contains terms of arbitrarily high order in infinitely many modes of . Therefore the
string theory counterpart of U(1) field strength has correction terms to F . (VRR8 ; ) gives
a gauge invariant extension of C (8) F , and can be expected to give some information on
the correction terms.
In the following we compute (VRR8 ; ) order by order, up to level 2. The RR p-form
vertex operator is
(q)(C 1 ...p+1 )AB VRR (z, z )AB ,
VRRp (z, z ) = F(p+1)
1 ...p+1
VRR (z, z )AB = c(z)e 2 (z) S A (z)c(

S B (z)eiqX (z, z ),
z)e 2 (z)
1
(15)
(p+1)
where S A (z) and
S B (z) are spin operators and, as is well known, F1 ...p+1 (q) corresponds
to field strength of the RR p-form. BRST invariance of VRRp requires F (p+1)(q)1 ...p+1 =
(p + 1)q[1 C (p) (q)2 ...p+1 ] , q 1 C (p) (q)1 2 ...p = 0 and q 2 = 0. VRRp also satisfies
[0 , VRRp ] = 0, and is a dimension (0, 0) primary field.

In order to perform an order by order calculation, let us decompose and into sums
of contributions of each level. The level is defined as L0 k 2 , i.e., the eigenvalue of L0
118
with the contribution of eikX subtracted.

= 0 + 1 + 2 + ,
= 0 + 1 + 2 + ,
(16)
where n and n are level n parts of and , respectively.

By the general argument given in the previous section, (VRR8 ; ) is gauge invariant. The
linear part (QB ) + (0 ) and the higher order part in (VRR8 ; ) cancel separately.
Henceforth we consider only the linear part 0 of the gauge transformation, i.e., 0 =
QB . 0 consists of only those without 0 , and does not give non-zero contribution to
(VRR8 ; ). Therefore we do not consider the parameter .
Since QB does not change the level, QB n gives the gauge transformation of n , and
we can investigate the contribution of each level separately.
First we consider the level 0 contribution. As we have already seen,

d 10 k
A (k) c e eikX + B(k) cce2 eikX
0 =
10
(2)
gives

AB

1
1
2
2
AB
VRR
; 0 = 22 q (2i) q /2 1 + 10 C 1
A (q),
2
2
(17)
where we did not use q 2 = 0.

To extract a quantity analogous to the gauge field, we define a (k) as follows:
AB
1
2
2
; off-shell,
a (q) = 2 22 q (2i) q /2 (C )AB VRR
16
(18)
AB ; )
AB
where (VRR
off-shell is equal to (VRR ; ) computed without using the on-shell condi2
tion q = 0. In other words, a (k) is the gauge field obtained by naive and straightforward
off-shell extension of (VRR8 ; ). At level 0, a (k) = A (k). Then the field strength of
a (k) is defined by f (k) = ik a (k) ik a (k).
Then (VRR8 ; ) is given by
16
2
2
(q)a10 (q)
(VRR8 ; ) = 22 q (2i) q /2 1 ...10 F(9)
1 ...9
2
16i
2
2
= 22 q (2i) q /2 1 ...10 C(8)
(19)
(q)f9 10 (q).
1 ...8
2
d 10 k i
(k) c
Let us see the linearized gauge transformation of a (k). 0 = (2)
10

e2 eikX gives the following gauge transformation:
0 a (k) = ik (k).
(20)
At this level, a transforms in the same way as ordinary gauge field. Note that the gauge
transformation is defined off-shell. Therefore at this level f (k) is an off-shell gauge
invariant.
119
Next we consider the level-1 contribution. At this level 1 and 1 are expanded by
AB has two c ghosts and charge
the basis given in the tables in Appendix A. Since VRR
1, and does not have and , only those with bc ghost number 1, charge 1 and
AB ; )
ghost number 1 and with 0 give non-zero contribution to (VRR
1 off-shell and a .
Then, the relevant basis consists of c e eikX , c e eikX , c e eikX ,
c e eikX and c e :X eikX :. By computing the CFT correlators of these
operators, we find that none of them contribute to (VRR8 ; 1 )off-shell and a .
Finally we consider level-2 contribution, and see whether a (k) transforms in the same
way as an ordinary gauge field. 2 and 2 are expanded by the basis given in the tables in
Appendix A. Among them only the following components give non-zero contributions to
(VRR0 ; 2 )off-shell and a (k):

d 10 k (1)
2 =
B (k) c 2 e eikX + B(2) (k) c : 2 e :eikX
10
(2)

(4)
+ B(3) (k) 2 c e eikX + B
(k) c e : 2 X eikX :

(6)
+ B(5) (k) c 2 e eikX + B
(k) c: :e eikX

(8)
+ B(7) (k) : :c e eikX + B (k) c e :X X eikX :

+ B(9) (k) :bcc: e eikX ,
(21)
(6)
where the coefficients B (n) have appropriate symmetry, i.e., B
is antisymmetric under
(8)
the exchange of and , and B is symmetric under the exchange of and .

2 is expanded, and coefficients of components are defined as follows:

d 10 k (1) 2
2 =
(k) ce2 eikX + (2)(k) c 2 e2 eikX
10
(2)

+ (3) (k) c: 2 e2 :eikX + (4)(k) 3 ce2 eikX

+ (5) (k) 2 ce2 :X eikX : + (6)(k) ce2 :X eikX :

(7)
(8)
+
(k) 2 c: :e2 eikX +
(k) c: :e2 eikX

+ (9) (k) 2 c 2 c e3 eikX + (10)(k) ce2 : 2 X eikX :

(11)
+
(k) ce2 :X X eikX :

(12)
(13)
+
(k) c: :e2 :X eikX : +
(k) c: :e2 eikX

(14)
+ (k) c: :e2 eikX + (15)(k) e eikX

+ (16)(k) 2 ce2 eikX + (17) (k) :bc: e eikX

(19)
+ (18)(k) e eikX +
(k) e :X eikX :

(20)
+ (k) : :e eikX + (21)(k) 2 cc e3 eikX

+ (22)(k) 3 cc e3 eikX

+ (23)(k) 2 3 cc 2 ce4 eikX + (24) 2 cc e3 eikX
120

(25)
+ (k) 2 cc: :e3 eikX

(26)
+
(k) 2 cc e3 :X eikX :

+ (27)(k) 2 ce2 eikX + (28) (k) ce2 eikX

+ (29)(k) :bcc:e2 eikX + (30)(k) ce2 :X eikX :

(31)
(k) c: :e2 eikX + (32) (k) beikX .
+
(22)
Again (n) have the appropriate symmetry. Then we can compute a and the gauge transformation of B (n) :
(4)
a (k) = A (k) B(1) (k) + 2B(3) (k) + 2i k B
(k) B(5) (k)
1
(6)
(8)
+ 3B
(23)
(k) B(7) (k) B
(k),
2

k (3) (k)
0 B(1) (k) = 6 2 k (1) (k) + 13 2 k (2)(k) +
2

(5)
(6)
(4)

(k) + 3i
(k)
+ 3 2 k (k) i
2
2

(10)
(7)
(8)
(9)

(k)
2 2 k (k) + 6 2 k (k) + 8 (k) i
2

1
(12)
(13)
(k) +
k (k) + (15)(k) + (17)(k)
+i
2
2
2
+ 36(21)(k) 24(22)(k),
(24)
0 B(2) (k) = 8 2 k (1) (k) + 12 2 k (2)(k) + 2 2 k (3)(k)
+ 6 2 k (4) (k) i 2 (5)(k) + 2i 2 (6) (k)
(7)
(8)
(k) + 8 2 k
(k) + 12(9)(k) i 2 (10) (k)
4 2 k
3
(12)
(13)
+ i 2
(k) + 2 k
(k) + (17)(k) + 48(21)(k)
2
36(22)(k),
(25)
1
0 B(3) (k) = 2(9)(k) + (15)(k) 2 k (16)
2
3 (17)
1

(19)
(k) (18)(k) i k
(k),
(26)
2
2
2

2
2
2
(4)
(1)
(2)
(k) + 2i
(k) + i
(3) (k)
0 B (k) = 4i

2
2 (7)
2 (8)
(4)
(k) 4i
(k) + 4i
(k)
+ 6i

2 (13)
(19)
+ 2 k (10)(k) + i
(27)
(k) 2ik (17)(k) +
(k),

121

(1)
(2)

k (3) (k)
2 k (k) + 2 k (k) +
2

(5)
(6)
(10)
(4)

(k) + i
(k) i
(k)
+ 3 2 k (k) i
2
2
2
3
(17)(k) (18)(k),
(28)
4

(12)
(6)
(7)
(8)
(k)
0 B
(k) = 2 2 k
(k) 2 2 k
(k) i
2
1
(13)
(17)
(20)
+ 2 [ (k)k] [ ] (k) + 3 (k),
(29)
2
0 B(7) (k) = 2 2 k (2) (k) 2 k (3)(k) i 2 (6) (k)
(8)
(12)
(k) 12(9)(k) + i 2 (10)(k) i 2
(k)
4 2 k
(13)
(k) (17)(k) 36(21)(k) + 24(22)(k),
2 k
(30)

2
2
(8)
(5)
(6)
(11)
( ) (k) 2i
( ) (k) + 2 k (k)
0 B (k) = 2i

2 (12)
1
(19)
2i
(31)
() (k) + (17) (k) + 2i( (k)k),
(9)
2
(17)
(21)
(22)
(29)
0 B (k) = ( k + 2) (k) 12 (k) + 12 (k) + 2 k (k). (32)
(5)
0 B
(k) = 2
By using the above expression of a and 0 B (n) , we can calculate 0 a

0 a (k) = ik (k) 16i 2 (1) (k) + 16i 2 (2)(k) + 2i 2 (3) (k)

( )3 (11)
3

(k)
+ 18i 2 (4) (k) + 2 k (10)(k) + i
2

(13)
(k) + 2i 2 (16)(k)
+ 3i
2
(7)
(8)
(k) + 16 2 k
(k) + 4 k 2 (17)(k).
16 2 k
(33)
Unfortunately 0 a is not in the form of an ordinary gauge transformation. Terms contain(7) (8)
(17)
ing , and are not proportional to k .
The gauge transformation of f is given by

(7)
(7)
0 f (k) = 16i 2 k k (k) k k (k)

(8)
(8)
+ 16i 2 k k
(k) k k
(k)

+ 4i k 2 k (17)(k) k (17) (k) .
(34)
Even on-shell f is not invariant. Of course (VRR8 ; ) is invariant, thanks to
k 1 C(8)
1 2 ...8 = 0.
Thus we have seen that the field strength extracted from the RR coupling is not gauge
invariant. Therefore it has some physical meaning only when it is coupled with the on-shell
RR-field. It is not immediately clear whether we can modify the definition of a so that it
transforms in the same way as ordinary gauge field.
122
4. Gauge invariant component of the equation of motion and field strength

As a second application of our gauge invariant quantities, we consider in this section we
gauge invariant components of the equation of motion, and extract a string field theory
analog of field strength from it, up to level 1. We will find that this quantity is invariant
under the linearized gauge transformation, even off-shell.
Let us begin with a more general consideration. If a function f () of transforms under the gauge transformation covariantly, i.e., if f () = [QB , f ()] or
[f (), 0 ], V does not have to be BRST invariant or in small Hilbert space for gauge
invariance of (V ; f ()). What we need is only (7) and therefore the only necessary condition on V is that V is a dimension (0, 0) primary operator.
As an example of f (), we take QB (e (0 e )). Since QB (e (0 e )) = 0 is the
equation of motion derived from the gauge invariant action, it is obvious that this transforms covariantly, and indeed [QB (e (0 e ))] = [QB , QB (e (0 e ))]. We may
also take 0 (e (QB e )), but (V ; 0 (e (QB e ))) is equal to (V ; QB (e (0 e )))
because of 0 (e (QB e )) = e QB (e (0 e ))e .
In fact (V ; QB (e (0 e ))) is a kind of gauge invariant we have considered in the
previous sections. This can be shown as follows:

= QB V ; e 0 e .
V ; QB e 0 e
(35)
QB V is a dimension (0, 0) primary operator. Therefore we can apply (7) to the right-hand
side, and we obtain

= (QB V ; 0 ) = (0 QB V ; ).
V ; QB e 0 e
(36)
0 QB V is BRST invariant, in small Hilbert space, and is a dimension (0, 0) primary operator.
From this calculation we see that only the linear part of the equation of motion contributes to these gauge invariants. In general each coefficient of component operators of
the basis in the linear part of the equation of motion is not gauge invariant. Therefore
our gauge invariants give gauge invariant linear combinations of components of the linearized equation of motion. Since the linearized equation of motion is QB 0 = 0, at the
linearized level these quantities are gauge invariant even off-shell, where off-shell means
evaluating this quantity without imposing q 2 = 0.
= 0, and the left-hand side is
In free U(1) gauge theory equation of motion is F

gauge invariant. Let us consider a string field theory counterpart that reduces to F

in lowest order, and extract a string field theory counterpart of F from it. We choose
V (z, z ) = (z)c(z) (z)e (z)e2iqX (z) as V , where the factor e2iqX (z) has only
left moving part. This is not an ordinary closed string vertex operator, but is a dimension
for the level
(0, 0) primary operator if q 2 = 0. We can see that this choice gives F
zero component 0 . The linear part of the equation of motion for 0 is

d 10 k
QB 0 0 =
cc e eikX k 2 A (k) 2 k B(k)
10
(2)

+ ceikX 2 k A (k) + 2B(k) .
(37)
123
The coefficient of the second component in the right-hand side gives an algebraic equation
of motion for the auxiliary field B(k). By using it, we obtain k (ik A (k) ik A (k)) =
k F (k) = 0 from the first component. This is the ordinary equation of motion of free
U(1) gauge theory. Then

V ; QB 0 0 off-shell

2
2 q 2 q 2
q 2
2
= (i)
2
(2i) q A (q) +
q
B(q)

2
2
2
= (i)2 q 2 q (2i) q iq iq A (q) iq A (q) ,
(38)
where we keep V off-shell.
Next we calculate the contribution of level-1 part 1 . Only operators with 0 contribute
to the linearized equation of motion, and their coefficients are defined as follows:

d 10 k (1)
D (k) c e eikX + D (2) (k) c 2 ce2 eikX
1 =
(2)10

+ D (3) (k) eikX + D(4) (k) c e eikX

(5)
(6)
(k) c: :e eikX + D
(k) c e :X eikX :
+ D

+ D (7) (k) 2 cce2 eikX + D (8) (k) cce2 eikX

(9)
+ D
(k) cc: :e2 eikX + D(10) (k) cce2 :X eikX :

+ D(11) (k) c e eikX + (no 0 part) .
(39)
The linearized equation of motion is given as follows:
QB 0 1

2
(1)
d 10 k
ikX
cc
=
e
e
k
+
1
D
(k)
+
2
2 k D (7) (k)
(2)10

(10)
(8)
(9)
(11)

4 2 k D (k) 2 2 k D (k) i
D (k) + D (k)
2

+ cc 2ce2 eikX k 2 + 1 D (2) (k) D (7) (k) + i k D(10) (k)
2

+ ceikX k 2 + 1 D (3) (k) 6D (7) (k) + 8D (8) (k) + 2 k D(11) (k)

+ cc e eikX k 2 + 1 D(4) (k) 2 2 k D (7) (k)

(10)
D (k) + D(11) (k)
2

(5)

(9)
+ cc: :e eikX k 2 + 1 D (k) 2 k[ D] (k)
+ 2 2 k D (8) (k) + i
124

2
(6)
2
+ cc e :X e
: k + 1 D (k) 2i
D (7) (k)

2
2 (9)
(8)
(10)
(11)

+ 2i
D (k) + 2i
D (k) 2 k D (k) + 2ik D (k)

2 ikX 1 (1)
1
+ c c e e
D (k) 2 k D (2) (k) D(4) (k)
2
2

(6)
(k) + D(11) (k)
+ i k D
2

+ ceikX 2 k D(1) (k) + 6D (2) (k) D (3) (k) + 2 k D(4) (k)
ikX

(6)
D (k) + 8D (7) (k) 12D (8) (k)
2

ikX
+ ce
2 2 k D(1) (k) + 8D (2) (k) + 2 k D(4) (k)
i

(6)
D (k) + 12D (7)(k) 20D (8)(k)
2

(1)
(5)
ikX
+ c: :e
2 k[ D] (k) + 3 2 k D (k)
i

(6)
(9)
D[] (k) + 2D
(k)
2

2 (1)
2 (4)
ikX
(3)
+ c:X e
: i
D (k) 2ik D (k) + i
D (k)

(6)
+ 2 k D
(k) + 2D(10) (k) .
+i
(40)
The first six components of the above equation have KleinGordon operator ( k 2 + 1),
and the last five components do not. Therefore the first six are dynamical equations of
(9)
(10)
motion and the rest is algebraic equations for auxiliary fields D (7) , D (8) , D , D , and
(11)
D . This can also be seen from the fact that operators corresponding to these auxiliary
fields contain c0 , and can be gauged away by taking the gauge b0 = 0.
Computation of (V ; QB 0 1 ) is straightforward
(V ; QB 0 1 )
2
2
2
= i(i)2 q 2 q (2i) q 2 q 2 D(1) (q) + 4 2 q D (2) (q)
(6)
(q) 4 2 q D (7) (q)
2 q 2 D(4) (q) + 2i( )2 q 2 q D

+ 2i 2 q q D(10) (q) + 4 q 2 D(11) (q) .
(41)

(10)
(11)
Eliminating auxiliary fields D (7) , D , and D

obtain
125
by algebraic equations of motion, we
(V ; QB 0 1 )

2
2
2
= i(i)2 q 2 q (2i) q q 2 8 2 q 2 + 1 q D (2) (q)

4 2 q 2 + 1 q D (3) (q) + 4 2q q D(4) (q) + q 2 D(4) (q)
(6)

(6)
(6)
+ 2i q D
(q) + D
(q) 2i q D(6) (q) 4i( )2 q q q D (q) .
(42)
Thus, up to this order, we get the following quantity:

iq iq A (q) iq A (q)

+ q 2 8 2 q 2 + 1 q D (2) (q) 4 2 q 2 + 1 q D (3) (q)

(6)
(6)
(q) + D
(q)
+ 4 2q q D(4) (q) + q 2 D(4) (q) + 2i q D

(6)
2i q D(6) (q) 4i( )2 q q q D (q) + .
(43)
This is gauge invariant if q 2 = 0, and as we have already noticed, this is invariant under
a linearized gauge transformation even off-shell. This is interpreted as an extension of
q F (q). However the contribution from the level-1 part is not in the form of q (. . .)[] .
It is thus not possible to extract the counterpart of F (q) from this quantity.
Therefore we want to make a better choice for V . To give a gauge invariant in the
form of q (. . .)[] , (V ; QB 0 ) has to satisfy q (V ; QB 0 ) = 0. Note the following
equation:

QB ce2 e2iqX = q V e2iqX ,
(44)

q cce2 e2iqX .
V = V +
(45)
2

This shows that if we add 2 q ( cce2 e2iqX ; QB 0 ) to (V ; QB 0 ), we
obtain a gauge invariant in the form of q (. . .)[] . This is because the second term in the
right-hand side of (44) does not contribute since there is no zero mode of in the correlator,
and when V is contracted with q the whole thing in the correlator becomes BRST exact.
The component expression of (V ; QB 0 ) up to level 1 is given by

2
2
2
(V ; QB 0 ) = (i)2 q 2 q (2i) q q (q A q A )

2iq q D(1) q D(1) 2iq q D(4) q D(4)

2 q q q D(6) q D(6) + 4iq q D(11) q D(11)
2
2
2
= (i)2 q 2 q (2i) q q

q A 2iD(1) 2iD(4) 2 q D(6) + 4iD(11)

q A 2iD(1) 2iD(4) 2 q D(6) + 4iD(11) .
(46)
126
This is in the form of q (. . .)[] . We have thus succeeded in extracting string field theory
and A , up to this level. Note that we did not use algebraic
counterparts f and a of F
equations of motion for auxiliary fields to extract these quantities
f (q) = iq a (q) iq a (q),
a (q) = A (q) 2iD(1) (q) 2iD(4) (q) 2 q D(6) (q) + 4iD(11)(q).
(47)
The linearized gauge transformation of a (q) is

0 a (q) = iq (q) 2i 2 (1)(q) 2 q (4)(q) + 2i 2 (7)(q) , (48)
where (n) are defined as:

d 10 k (1)
1 =
(k) 2 ce2 eikX + (2)(k) ce2 eikX
(2)10
(3)
+
(k) c: :e2 eikX + (4) (k) ce2 :X eikX :
+ (5)(k) e eikX + (6)(k) 2 cc e3 eikX

+ (7)(k) ce2 eikX ,
and the linearized gauge transformation for each component is
(3)
0 D(1) (k) = 2 2 k (1)(k) + 4 2 k (2) (k) + 2 2 k
(k)

(4)
(k) (5)(k) + 4(6)(k),
+i
2

0 D (2) (k) = (1)(k) + i k (4)(k) + (7)(k),
2
0 D (3) (k) = 6 (1)(k) 8 (2)(k) 2 k (5)(k) + 2 (7)(k),

(4)
(4)
(1)
(2)

(k) (5)(k),
0 D (k) = 2 2 k (k) 2 2 k (k) i
2
(5)
(3)
0 D (k) = 2 k[ ] (k),

2
2
2 (3)
(6)
(1)
(2)
(k) 2i
(k) 2i
(k)
0 D (k) = 2i

+ 2 k (4) (k) 2ik (5)(k),

0 D (7) (k) = k 2 + 1 (1)(k) 2 k (6) (k) + (7)(k),

0 D (8) (k) = k 2 + 1 (2)(k) 2 k (6) (k) + (7)(k),

(3)
(6)
(9)
(k) = k 2 + 1
(k) + 2 2 k[ ]
(k),
0 D

2 (6)
(k) + 2ik (7)(k),
0 D(10) (k) = k 2 + 1 (4) (k) + 2i

0 D(11) (k) = k 2 + 1 (5) (k) + 2(6)(k) + 2 k (7) (k).
(49)
(50)
(51)
(52)
(53)
(54)
(55)
(56)
(57)
(58)
(59)
(60)
127
V is one of the simplest choices for V , but of course we can take different ones. For
example,
V (z, z ) = (z)c(z) (z)e (z)eiqX (z, z )
+ (z)(z)(z)c(z)e (z)e2 (z) (z)eiqX (z, z )

1
q (z)(z)c(z) c(z) + c(z) e2 (z)eiqX (z, z )
+
2 2

2
(z)(z)c(z)c(z)e2 (z):X (z)eiqX (z, z ):,
i
(61)
where eiqX (z, z ) is an ordinary momentum factor for closed strings, and the normal ordering of :X (z)eiqX (z, z ): acts on left mover
and right mover separately. This satisfies
QB [(z)(z)c(z)e2 (z)eiqX (z, z )] = 2 q V (z, z ) eiqX (z, z ). This equation shows that V (z, z ) is one of the right choices for V .
We have considered only the linearized part of the gauge transformation. Some choices
for V might give gauge invariant quantities under full gauge transformation even off-shell,
and others might not.
5. Discussion
We have considered on-shell gauge invariants in Berkovits open superstring field
theory and, as an application of these we have computed a string field theory analog of
up to level 2, and a gauge invariant component of the equation
RR coupling C (8) F
of motion up to level 1. We have extracted analogs of the field strength from them. The
former is not gauge invariant even on-shell, and the corresponding gauge field a does
not transform as an ordinary gauge field. The latter is gauge invariant under full gauge
transformation if on-shell, and invariant under linearized gauge transformation even offshell.
It seems that the latter is more promising, but obviously there are many open problems.
First of all, it is not clear whether it is fully invariant off-shell for some choices of V and,
if not, whether there is a way of extending it to off-shell. By momentum conservation
we always have some kind of on-shell condition if we use on-shell closed string vertex
operators, or dimension (0, 0) operators.
Another unclear point is what our gauge invariant gives after integrating out massive
modes, and how it is related to the gauge field given in [9,10] and marginal deformation
in [11,12].
The construction of field strength in the non-Abelian case, i.e., multiple D-brane case,
is an interesting direction of extension, although our method is not directly applicable to
this case. For RR coupling, the non-Abelian part of the field strength does not couple
to the RR 8-form. More technically, relation (7) does not hold for matrix valued string
fields.
128
Acknowledgements
This work is supported by the Japan Society for the Promotion of Science and the Swiss
National Science Foundation.
Table 1
Basis for . Only those with 0 and without c0 are shown. For the level-2 basis, the operator expressions are
given only for those relevant to our calculation
Level
Oscillator expression
Operator expression

0 1/2 |
c e e2ikX

0 1/2 1/2 1/2 |

0 1/2 c1 |
c e e2ikX

0 1/2 b1 |

0
|
: :e2ikX
c e e2ikX
3/2

0 1/2 1/2
1/2
|

0
|
1/2 1
c 2 ce2 e2ikX
c: :e e2ikX
c e :X e2ikX :
c 2 e eikX

0 5/2 |

0 3/2 1/2 1/2 |

0 3/2 1/2 1/2 |

0 3/2 1/2 1/2 |

0 3/2 1/2 1/2 |
( c 2 e 2: :ce ) eikX
( c( + 2 )e 2: :ce ) eikX
c: :e eikX

0 3/2 c1 |

0 3/2 b1 |
|

0 3/2 1

0 1/2 1/2 1/2 1/2 1/2 |
|
0 1/2 1/2
1/2 1/2 1/2

1/2
1/2 1/2
|
0 1/2 1/2

0 1/2 1/2 1/2 c1 |
c: :e eikX

0 1/2 1/2 1/2 b1 |

0 1/2 1/2
|
1/2 1

c1 |
0 1/2 1/2 1/2

b1 |
0 1/2
1/2 1/2
1/2
1 |
0 1/2 1/2

0 1/2 c2 |

0 1/2 b2 |
|

0 1/2 2

0 1/2 c1 1 |

0 1/2 b1 1 |
|

0 1/2 1 1

0 1/2 b1 c1 |
c e : 2 X eikX :
c e :X X eikX :
2 c e eikX
129
Appendix A
In this appendix we tabulate the bases for expanding the string field and the gauge
transformation parameter . The following two vacua are used in Tables 14:
| = c1 eikX(0)|0,
= e (0)c1 eikX(0)|0,
|
(A.1)
Table 2
Basis for . Only those with 0 and c0 are shown. For the level-2 basis, the operator expressions are given only
for those relevant to our calculation
Level
Operator expression

0 c0 1/2 |
cce2 e2ikX

0 c0 3/2 |
( 2 cce2 + 12 cce2 )e2ikX

0 c0 1/2 1/2 1/2 |

0 c0 1/2
|
1/2 1/2

0 c0 1/2 1 |

0 c0 1/2 b1 |
( 2 cce2 + cce2 )e2ikX

cc: :e2 e2ikX
cce2 :X e2ikX :
c e e2ikX

0 c0 5/2 |

0 c0 3/2 1/2 1/2
|

0 c0 3/2 1/2 1/2 |

0 c0 3/2 1/2 1/2 |

0 c0 3/2 1/2 1/2
|

0 c0 3/2 1 |

0 c0
b1 |
3/2

0 c0 1/2 1/2 1/2 1/2 1/2 |

0 c0 1/2 1/2 1/2
|
1/2 1/2

0 c0 1/2 1/2 1/2
1/2
1/2 |

0 c0 1/2 1/2 1/2 1 |

0 c0 1/2 1/2
c1 |
1/2

0 c0 1/2 1/2 1/2 b1 |
|

0 c0 1/2 1/2 1/2
1

0 c0 1/2 1/2
1/2
b1 |

0 c0 1/2 2 |

0 c0
b2 |
1/2

0 c0 1/2 1 1
|

0 c0 1/2 b1 c1 |

0 c0
b1 |
1/2
:bcc: e eikX
130
Table 3
Basis for . Only those without c0 are shown
Level
Operator expression

0 1/2 |
ce2 eikX

0 1/2 1/2 1/2
|
c: :e2 eikX

0 1/2 1 |

0 1/2 b1 |

0 3/2 |

0 1/2 1/2 1/2 |
2
cX e2 eikX
e eikX
( 2 ce2 + 12 ce2 )eikX
( ce2 + 2 ce2 )eikX

0 5/2 |
( 3 e2 2 2 :e2 :
+ :( 2 )e2 :)ceikX

0 3/2 1/2 1/2 |
( 2 e2 + 12 3 e2
+ 12 :( 2 )e2 :)ceikX

0 3/2 1 |
( 2 e2 + 12 e2 )c:X eikX :

0 3/2 1/2 1/2
|

0 1/2 1/2 1/2 1/2 1/2 |
( 2 e2 + 12 e2 )c: :eikX
(6 :e2 : + 3 2 e2 + 3 e2 )ceikX

0 1/2 1/2 3/2 |
( :( + 2 )e2 : 2 :e2 :)ceikX

0 1/2 1/2 1/2 1 |

0 1/2 1/2 1/2 1/2 1/2
|

0 1/2 1/2 c1 1/2 |

0 1/2 2 |

0 1/2 1 1
|

0 1/2 1 1/2 1/2
|

0 1/2 3/2 1/2 |

0 1/2 1/2 1/2 1/2
1/2 |

0 1/2 b1 1/2 1/2 |
( e2 + 2 e2 )c:X eikX :
( e2 + 2 e2 )c: :eikX
2 c 2 ce3 eikX
ce2 : 2 X eikX :
ce2 :X X eikX :
ce2 : ::X eikX :
ce2 : :eikX
ce2 : :eikX
e eikX

0 1/2 b1 c1 |
2 ce2 eikX

0 b2 1/2 |
:bc:e eikX
e eikX

0 b1 3/2 |
|

0 b1 1/2 1

0 b1 1/2 1/2
1/2
|
e : :eikX
b2 b1 |
beikX
e :X eikX :
where |0 is Fock vacuum defined by n1 |0 = n1/2 |0 = bn1 |0 = cn2 |0 =
n1/2 |0 = n3/2 |0 = 0. We show both oscillator expressions and operator expressions. Oscillator expression |osc. and the corresponding operator expression v is related
by |osc. = (numerical factor)v(0)|0.
131
Table 4
Basis for . Only those with c0 are shown
Level
none

0 c0 1/2 b1 |
Operator expression
ce2 eikX

0 c0 1/2 1/2 1/2 |
2 cc e3 eikX

0 c0 3/2 1/2 1/2 |

0 c0 3/2 b1 |
( 3 e3 2 2 :e3 :)cc eikX

0 c0 1/2 1/2 1/2 c1 |

0 c0 1/2 1/2
|
2 3 e4 cc 2 ceikX

0 c0 1/2 1/2 1/2 1/2 1/2 |
3/2

0 c0 1/2 1/2 1/2 1/2
1/2
|

0 c0 1/2 1/2 1/2 1 |

0 c0 1/2 1/2 b1 1/2 |

0 c0 1/2 b2 |

0 c0 1/2 b1 1 |

0 c0 1/2 b1
|
1/2 1/2
( 2 e2 + 12 e2 )ceikX
( 2 e3 + 3 e3 )cc eikX
2 e3 cc eikX
2 e3 cc: :eikX
2 e3 cc :X eikX :
( e2 + 2 e2 )ceikX
e2 :bcc:eikX
e2 c:X eikX :
e2 c: :eikX
References
[1] B. Zwiebach, Interpolating string field theories, Mod. Phys. Lett. A 7 (1992) 1079, hep-th/9202015.
[2] A. Hashimoto, N. Izhaki, Observables of string field theory, JHEP 0201 (2002) 028, hep-th/0111092.
[3] D. Gaiotto, L. Rastelli, A. Sen, B. Zwiebach, Ghost structure and closed strings in vacuum string field theory,
Adv. Theor. Math. Phys. 6 (2002) 403, hep-th/0111129.
[4] N. Berkovits, Super-Poincar invariant superstring field theory, Nucl. Phys. B 450 (1995) 90, hepth/9503099.
[5] D. Kutasov, M. Marino, G. Moore, Some exact results on tachyon condensation in string field theory,
JHEP 0010 (2000) 045, hep-th/0009148;
D. Kutasov, M. Marino, G. Moore, Remarks on tachyon condensation in superstring field theory, hepth/0010108.
[6] E. Witten, On background independent open-string field theory, Phys. Rev. D 46 (1992) 5467, hepth/9208027.
[7] N. Berkovits, A. Sen, B. Zwiebach, Tachyon condensation in superstring field theory, Nucl. Phys. B 587
(2000) 147, hep-th/0002211.
[8] N. Berkovits, M. Schnabl, YangMills action from open superstring field theory, JHEP 0309 (2003) 022,
hep-th/0307019.
[9] J.R. David, U(1) gauge invariance from open string field theory, JHEP 0010 (2000) 017, hep-th/0005085.
[10] E. Coletti, I. Sigalov, W. Taylor, Abelian and non-Abelian vector field effective actions from string field
theory, JHEP 0309 (2003) 050, hep-th/0306041.
[11] A. Sen, B. Zwiebach, Large marginal deformations in string field theory, JHEP 0010 (2000) 009, hepth/0007153.
[12] A. Iqbal, A. Naqvi, On marginal deformations in superstring field theory, JHEP 0101 (2001) 040, hepth/0008127.

Towards the exact dilatation operator of N = 4

super-YangMills theory
A.V. Ryzhov a , A.A. Tseytlin b,1
a Department of Physics, Brandeis University, Waltham, MA 02454-9110, USA
b Department of Physics, The Ohio State University, Columbus, OH 43210-1106, USA

Abstract
We investigate the structure of the dilatation operator D of planar N = 4 SYM in the sector of
single trace operators built out of two chiral combinations of the 6 scalars. Previous results at low
orders in t Hooft coupling suggest that D has a form of an SU(2) spin chain Hamiltonian with
long range multiple spin interactions. Instead of the usual perturbative expansion in powers of , we
split D into parts D (n) according to the number n of independent pairwise interactions between spins
at different sites. We determine the coefficients of spinspin interaction terms in D (1) by imposing
the condition of regularity of the BMN-type scaling limit. For long spin chains, these coefficients
turn out to be expressible in terms of hypergeometric functions of , which have regular expansions
at both small and large values of . This suggest
that anomalous dimensions of long operators in
the two-scalar sector should generically scale as at large , i.e., in the same way as energies of
semiclassical states in dual AdS5 S 5 string theory.
PACS: 11.15.Kc; 11.15.Pg; 11.25.-w; 11.25.Hf; 11.25.Tq
Keywords: Classical and semiclassical techniques; Expansions for large numbers of components; Strings and
branes; Conformal field theory; Algebraic structures; Gauge/string duality
E-mail address: ryzhovav@brandeis.edu (A.V. Ryzhov).

1 Also at Imperial College London and Lebedev Institute, Moscow.
doi:10.1016/j.nuclphysb.2004.07.037
A.V. Ryzhov, A.A. Tseytlin / Nuclear Physics B 698 (2004) 132148
133
1. Introduction
The N = 4 supersymmetric SU(N) YangMills theory [1] is the basic example of a 4-d
conformal theory [2]. It is actually a family of CFTs parametrized by N and the t Hooft
2 N . To solve a CFT amounts, at least, to being able to compute dimencoupling = gYM
sions of local gauge invariant conformal operators as functions of the parameters. This
problem should simplify in the planar limit of N , fixed. In this limit the AdS/CFT
duality conjecture [3] suggests that conformal dimensions should be smooth functions of ,
and have regular expansions at both large and small .
Important progress towards this non-trivial goal of understanding how anomalous dimensions depend on was recently made by concentrating on states with large quantum
numbers (see, in particular, [421]). For some non-BPS states there are new expansion limits (like large J , fixed J2 for S 5 -rotating pointlike [4] and multispin [12] string states)
where one can directly compare perturbative SYM anomalous dimensions to semiclassical
string results.
One would obviously like to go beyond the restriction to long and/or scalar-only operators and compute, e.g., the exact dimension of the Konishi scalar tr(i i ), or the
coefficient f () of the ln S term in the anomalous dimension of the minimal twist operators such as tr(i D S i ) [5,8,22]. The main obstacle is our lack of tools for obtaining
bn

n ) side.
exact all-order results on either gauge theory ( cn n ) or string theory (
( )
One potentially fruitful idea of how to go beyond the first few orders in SYM perturbation
theory is to try to determine the exact structure of the dilatation operator D by imposing
additional conditions (like superconformal symmetry, BMN limit, integrability, etc., as in
[2325]) implied by the expected correspondence with AdS5 S 5 string theory. Having
found the resulting anomalous dimensions as functions of , one may then be able to see
if they admit a regular expansion not only at small but also at large .
This is the approach we would like to explore below using as an input the condition of regularity of BMN-type scaling limit in the form suggested in [18] and further
clarified in [21]. We shall concentrate on the planar SU(2) sector of single trace SYM operators built out of chiral combinations X and Z of two the 6 SYM adjoint scalars, i.e.,
tr(X . . . XZ . . . ZX . . .) with canonical dimension L. This sector is closed under renormalization [23]. The eigen-operators of D with J1 Zs and J2 Xs (so L = J1 + J2 ) should be
dual to string states with two components of the SO(6) spin [4,12]. On general grounds,
the SYM dilatation operator computed in the planar limit should be a series in
D=

r=0
r
D2r .
(4)2r
(1.1)
Let us review what is known already about the structure of D2r .

Restricting D to planar graphs suggests that D2r should be given by local sums over
sites a = 1, . . . , L with Z and X interpreted as a spin up and spin down state of a
periodic (a + L a) spin chain [11,23] for which D is the Hamiltonian,
D2r =
L

a=1
D2r (a),
D0 = 1.
(1.2)
134
The one-loop term D2 turns out to be equivalent to the Hamiltonian of the ferromagnetic
XXX 1 Heisenberg spin chain [11],
2
D2 = 2Qa,a+1 ,
(1.3)
1
(1.4)
Qa,b = (1 a b ),
2
where Pa,b is the permutation operator and a are the Pauli matrices acting on the spin
state at site a.2 The two-loop term D4 was found to be [23]
Qa,b 1 Pa,b ,
i.e.,
D4 = 2(4Qa,a+1 + Qa,a+2 ),
(1.5)
while the expression for the 3-loop term D6 conjectured in [23] on the basis of integrability
D6 = 4(15Qa,a+1 6Qa,a+2 + Qa,a+3 )
+ 4(Qa,a+2 Qa+1,a+3 Qa,a+3 Qa+1,a+2),
(1.6)
was shown in [25] to be uniquely fixed by the superconformal symmetry algebra, constraints coming from the structure of Feynman graphs and the correct BMN limit.3 Finally,
there is also a proposal [24] for the 4-loop term D8 based on assuming integrability [23]
and the BMN scaling. Written in terms of factorized permutations as in [19] it reads
D8 = 10(56Qa,a+1 + 28Qa,a+2 8Qa,a+3 + Qa,a+4 )
2
+ (421Qa,a+1Qa+2,a+3 + 986Qa,a+3Qa+1,a+2 183Qa,a+2Qa+1,a+3)
3
+ 8(Qa,a+3Qa+2,a+4 + Qa,a+2 Qa+1,a+4 Qa,a+4 Qa+2,a+3

Qa,a+4 Qa+1,a+2 Qa,a+4 Qa+1,a+3 + Qa,a+3 Qa+1,a+4 ) .
(1.7)
Generalizing the above expressions for r = 2, 3, 4 it is then natural to expect that generic
r-loop term in (1.2) will contain a term linear in Qa,b , a term quadratic in Qa,b , and so on:

(1)
(2)
(n)
D2r = D2r
(1.8)
+ D2r
+ ,
D2r
Qn ,
(1)
D2r = 2
r
ar,c Qa,a+c .
(1.9)
c=1
(1)
(n)
At order r there can be at most r spinspin interactions in D2r
[6,23]. Qn in D2r
stands
for products of independent projectors, i.e., with all indices corresponding to different sites
as in (1.6) and (1.7). The above explicit expressions (1.3), (1.5), (1.6), (1.7) for D2 , . . . , D8
imply that for r 4 the coefficients ar,c are (c = 1, 2, . . . , r)
a1,1 = 1,
a2,c = (4, 1),
a3,c = (30, 12, 2),
a4,c = (280, 140, 40, 5).
(1.10)
1
2 For a = b one should set Q
2
a,a = 0; note that P = 1 and 2 Q is a projector.
3 The same expression was found in a closely related context of SYM matrix model [26]. Also, the 3-loop
anomalous dimension of (a descendant) of the Konishi operator found in [23] from the above form of D6 received
a remarkable indirect confirmation in a recent computation of anomalous dimension in twist 2 sector [22] which
also contains a descendant of the Konishi operator (N. Beisert and M. Staudacher, private communication).
135
Then
D = D0 + D (1) + D (2) + ,
D (1) = 2

r=1
L
r
r
ar,c Qa,a+c . (1.11)
(4)2r
a=1 c=1
Using the periodicity of the chain (Qa,b+L = Qa,b , etc.) D (1) can be rewritten as
D (1) =
L L1

hc (L, )Qa,a+c .
(1.12)
a=1 c=1
Our aim below will be to determine the general expression for the coefficients ar,c and thus
the functions hc (L, ), i.e., to find the spinspin (linear in Q) part of the exact dilatation
operator D.
To find the coefficients ar,c in (1.9), we will demand that the BMN-type scaling limit
L ,
= fixed
L2
(1.13)
of the coherent-state expectation value of D(1) [18,21] is well defined. This turns out to be
(nearly) equivalent to the consistency with the BMN expression [4,6,7] for the anomalous
dimensions of the 2-impurity operators. Imposing the condition of agreement with the
BMN square root formula fixes one remaining free coefficient at each order in r.
Our approach is thus similar to the previous important investigations of the constraints
on the dilatation operator imposed by the BMN limit [6,2325]. The new elements of the
present discussion are that (i) we follow [19,21] and classify the structuresin D interms of
independent interactions between sites as in (1.8), (1.11), i.e., D = D0 + Q + QQ +
, and (ii) we resum the loop expansion to find the coefficients in D (1) = Q as explicit
functions of and study their strong-coupling limit.
The resulting D (1) (1.12) may be interpreted as a Hamiltonian of a periodic spin
chain with long-range interactions. One could conjecture that, like in some known examples [27,28], this spin chain may be integrable; and, furthermore, the higher-order terms
D (2) , D (3) , . . . in (1.11) may be effectively determined by D (1) , e.g., expressed in terms
of higher conserved charges of the chain. This would then determine the full D. Remarkably, this is indeed true up to order 3 [19]: the sum of one, two, and three-loop dilatation
operators can be viewed as a part of the Inozemtsev integrable spin chain [28], with the
QQ-terms in D6 in (1.6) being proportional to a leading term in the -expansion of a
higher conserved charge of the Inozemtsev chain. At order 4 the Hamiltonian of the Inozemtsev chain does not, however, agree with the BMN perturbative scaling [19]. Here,
instead of starting with the Inozemtsev chain, we reverse the logic and determine which
spin chain Hamiltonian is actually consistent with the BMN limit, leaving the issue of
integrability open.4
4 Footnote 5 in [19] points out, following [28], that the only integrable spin chain with spinspin interactions
is the Inozemtsev chain. This implies that higher-order terms D (n) may not be directly determined by D (1) , and
only the whole D may represent a Hamiltonian of an integrable spin chain.
136
For comparison, let us recall that the Inozemtsev Hamiltonian that interpolates between
the Heisenberg and HaldaneShastry spin chain Hamiltonians is given by [28]
H=
L L1

pc (L, q)Qa,a+c ,
(1.14)
a=1 c=1
where pc (L, q) is a double-periodic Weierstrass function with periods L and q

1
1
1
.
pc = 2 +
c
(c mL inq)2 (mL + inq)2
(1.15)
(m,n)=(0,0)
Note that pc = pLc , so the sum in (1.14) may be restricted to c [L/2]. The limiting cases are limq pc = ( L )2 ( 21 c 13 ) (the HaldaneShastry chain limit), and
sin
limL pc = ( q )2 (
1
sinh2
c
q
+ 13 ) (q 0 corresponds to the infinite Heisenberg chain).
As was suggested in [19], to

relate H in (1.14) to the 3-loop SYM dilatation operator one
2 n 1
is to relate q to by (4)
2 =
n=1 (4 sinh q ) .
The rest of this paper is organized as follows. In Section 2 we determine the coefficients
in (1.9) by first imposing the regularity of the scaling limit (1.13) of the coherent-state expectation value of D (1) following [18,21] (Section 2.1) and then checking consistency with
the 2-impurity BMN spectrum (Section 2.2). In principle, we could fix all the coefficients
just by imposing the second (BMN) condition but we believe the approach of Section 2.1
is more straightforward and has its own conceptual merit, having close connection to string
theory [18,21].
In Section 3 we discuss how to sum up the t Hooft coupling expansion of the coefficients in D (1) . We first show that in the large L limit the coefficients of Qa,a+c can be
expressed in terms of hypergeometricfunctions which smoothly interpolate between perturbative power series at small and growth at large . We then comment on the finite
L case, in particular on the resulting contribution of D (1) to the exact anomalous dimension
of the Konishi operator.
Section 4 contains some conclusions and a discussion of open problems. Some useful
relations and definitions are summarized in Appendix A.
2. Determining the coefficients in D (1)

Our first task will be to obtain the general expression for the coefficients ar,c in the
linear in Qa,a+c part of the dilatation operator in (1.2), (1.9). To do this we shall follow
[18,21] and consider the spin coherent state path integral representation for the quantum
mechanics of D (1.1) as a generalized spin chain Hamiltonian. The corresponding action
for a collection of unit 3-vectors na (t) at each of L sites of the chain ( n|ai |n
= nia ,
(nia )2 = 1, i = 1, 2, 3) will then be given by
L

S = dt
(2.1)
LWZ (na ) D
, D
= n|D|n
,
a=1
137
where LWZ (
na ) = Ci (
na )n ia ensures the proper SU(2) commutation relations if one considers n a back as spin operators (see, e.g., [29]). Motivated by the explicit results (1.3)
(1.7) where the leading (at small ) coefficient of Qa,a+c is always positive, it is natural
to assume that the spin chain in question is ferromagnetic.5 Then in the long spin chain
limit L one may expect [18,21] that the low energy excitations of the spin chain will
be captured by the semiclassical dynamics of (2.1). The correspondence with string theory
then suggests [18,21] that S should have a regular scaling limit (1.13), or, more explicitly,
that the low energy effective action for the system governed by (2.1) should have a well
defined continuum limit. To take the continuum limit one may introduce a field n (, t),
0 < 2, with na (t) = n( 2a
L , t), so that (2.1) becomes

S L
2
dt

d
Ci (
n)n ia H n , 2 n , . . . ; .
2
(2.2)
H which originated from D

should be a regular function of the effective coupling and
-derivatives of n (t, ) in the limit L , fixed (with subleading L1n terms omitted).
Quantum corrections are then suppressed because of the large prefactor L in front of the
action.
Writing D in terms of factorized permutation operators as in (1.5)(1.7) one observes
that since the Qa,b in (1.4) satisfy
1
1
n|Qa,b |n
= (1 na nb ) = (
na nb )2 ,
2
4
(2.3)
n|D(1) |n
in (1.8) contains terms quadratic in n (but all orders in derivatives); n|D(2) |n
,
terms quartic in n, etc. The approximation that distinguishes D(1) from all higher D(k)
in (1.8) is the one in which one keeps only small fluctuations of n (t, ) near its (all where |n|
1, so that
spins-up) ground-state value n0 = (0, 0, 1). Then n = n0 + n,

higher powers of the fluctuating field n are suppressed, regardless the number of spatial
derivatives acting on them. Such configurations correspond to semiclassical spinning string
states with J1 J2 , and are close to a single-spin BPS state. They should indeed represent
semiclassical or coherent-state analogs of few-impurity BMN states, having the same BMN
energyspin relation which is indeed reproduced in the limit J1 J2 [8,12,14,15] by the
classical two-spin string solutions.
Let us therefore consider this BMN-type approximation, concentrating on the part of
n|D|n
which is quadratic in n, i.e., on n|D (1) |n
, and demand that the continuum version
of n|D (1) |n
have a regular scaling limit (1.13). As was shown in [18,21], this condition
is indeed satisfied at r = 1, 2, 3 loop orders, i.e., for (1.3), (1.5) and (1.6) (in general, this
should be a consequence of the supersymmetry of the underlying SYM theory which restricts the structure of D). One finds that the coefficients in the order Q-terms in (1.5)(1.7)
are such that all lower than r-derivative terms in the continuum limit of n|Dr(1) |n
cancel
out. If they did not, the limit L would be singular, as there would be a disbalance
between the powers of and the powers of L. Explicitly, one gets for r = 1, 2, 3, 4 using
5 Then the state with all spins up (represented by the operator tr Z L ) is a true vacuum.
138
(1.10) (after integrating by parts)

2r+2

r
r 2
r (1)
r
2
r
D
=
a
(
n
n

)
d
n

+
O
,
r,c a
a+c
r
(4)2r 2r
2(4)2r
L2
c=1
(2.4)
where
1
1
1
5
d1 = ,
d2 = ,
d3 = ,
d4 =
.
8
32
64
512
The sum over a in (1.2) becomes an integral over as in (2.2), and we find
2

L

d
1
D2r ( ) + O
.
D2r (a ) L
2
L
a=1
(2.5)
2.1. Regularity of the continuum limit

Let us now demand that the same pattern of cancellations (2.4) should persist to all
orders in -expansion.
Taking the continuum limit in (2.4) one may use the Taylor expansion to show that, up
to a total derivative,

(1)m1 (2c)2m
m 2
n + ( ).
(
na na+c ) = 2
(2m)!L2m
(2.6)

drm
m 2
r (1)
r
D
n ,
2r
(4)2r
L2m
(2.7)
(1)m1 (2)2m 2m
drm =
c ar,c .
(4)2r (2m)!
(2.8)
m=1
Then
m=1
c=1
(1)
should scale as L12r so that

To make sure that the limit (1.13) of D (1) is well defined, D2r
r
(1)
r D2r
r = L2r , up to subleading O( L1 ) terms. This implies that dr1 , . . . , dr,r1 must
vanish, i.e., that the coefficients ar,c must satisfy
r
c2m ar,c = 0,
for 0 < m < r.
(2.9)
c=1
This gives (r 1) equations for r unknowns. The coefficients ar,c with c = 2, . . . , r 1

are then uniquely determined in terms of a single multiplicative constant ar,r from
a
...
(r 1)2
1
22
r2
r,1
24
...
(r 1)4 ar,2
r4
1
. = ar,r . .
.
(2.10)
.
.
.
.
..
.
..
..
..
.
.
r 2(r1)
1 22(r1) . . . (r 1)2(r1)
ar,r1
139
The matrix in (2.10) is invertible for any r, and one finds6

(1)rc (2r)!
(2.11)
ar,r , c = 1, . . . , r 1.
(r c)!(r + c)!
This generalizes (1.10) to all values of r. Given (2.11), the first non-vanishing coefficient
vr in (2.8) becomes
ar,c =
(1)r1 2r
(1)r1
c
a
=
ar,r ,
r,c
22r (2r)!
22r+1
r
drr dr =
(2.12)
c=1
and the r-loop contribution to the expectation value of D (1) takes the form (cf. (2.4))

2
d
r 2
r (1)
1
r
.
(2.13)
D
L dr
n + O
2
L
(4)2r 2r
0
What remains is to find the values of ar,r generalizing (1.10). This can be done by analyzing
the spectrum of BMN operators.
2.2. Constraints from the BMN limit
Let us consider the two-impurity BMN operators of the form

J

1
n(2p + 1)
BMN
On
=
cos
tr XZ p XZ J p .
J
+
1
J + 1 p=0
(2.14)
These operators are multiplicatively renormalized for any J , and so are eigenstates of D
(see [30] and references there). Here the total number of fields, i.e., the length of the spin
chain, is L = J + 2, with J1 J fields Z and J2 = 2 fields X. Anomalous dimensions of
the BMN operators can be computed in both string theory and gauge theory [4,7] in the
large J , fixed J2 limit, and one finds

2
1
.
BMN = J + 2 1 + 2 n + O
(2.15)
J
J
Note that since L = J + 2, this BMN limit is essentially the same as the scaling limit (1.13)
discussed above, with being the same, up to subleading J1 terms, as J2 (which is usually
denoted as ). To reproduce (2.15) (i.e., the coefficients in the expansion of the square root
, we should require
in powers of ) by acting with the dilatation operator (1.1) on OBMN
n
that

(1)r1(2r 1) 2n 2r BMN
1
.
=
4
O
+
O
D2r OBMN
(2.16)
n
n
(r)(r + 1)
J
J 2r+1
6 The same relation appeared in BMN limit in the first reference in [6]. There the authors computed contributions to the scaling dimensions of the BMN operators by analyzing the relevant Feynman diagrams. Here
instead we determine the part of the full dilatation operator which, in particular, computes the dimensions of
BMN operators, but which can also be applied systematically to general operators made out of scalars X and Z.
140
To evaluate D2r OBMN

explicitly let us note since we are interested in the large J limit, we
n
can ignore what happens near the ends of the spin chain. Then for generic values of p in
the sum (2.14) we find that the permutation operators Pa,b = 1 Qa,b act as follows (c can
be positive or negative)

P1,1+c Op = Pp+2,p+2c Op = Opc , Op tr XZ p XZ J p ,

(2.17)
while for all other labels a, b we get Pa,b Op = Op . Then
J
+2
Qa,a+c Op = 2(Op Opc ) + 2(Op Op+c ),
(2.18)
a=1
where there are four non-zero combinations, from Q1,1|c| and from Qp+2,p+2|c| . Other
(n)
terms D2r in D2r (1.8) containing two and more products of Qs (which appear starting
with 3-loop term (1.6)), i.e., Qa,b . . . Qc,d with all labels a, b . . . c, d distinct, annihilate Op
unless we are near the ends of the chain, i.e., they do not contribute to (2.16) in the large J
limit. Ignoring these higher order terms is equivalent to the usual dilute gas approximation
which applies when the number of impurities is small and the spin chain is very long. Thus
it is the D (1) part of the full dilatation operator (1.11) that is responsible for the anomalous
dimensions of the operators OBMN
in the BMN limit,
n
(D D0 )OBMN
= D (1) OBMN
+ ,
n
n
where dots stand for subleading
in J1 ,
(1) BMN
D2r
On
r
1
J
(2.19)
terms. Using (2.18) we find that to the leading order

J

n(2p + 1)
(2Op Opc Op+c )
cos
J +1
ar,c
J + 1 p=0

r

(1)m1 2n 2m 2m
=8
c ar,c OBMN
.
n
(2m)!
J
c=1
m=1
(2.20)
c=1

To arrive at this expression we expanded the cosines at large J . The sum rc=1 c2m ar,c in
(2.20) is precisely the same as in (2.7), (2.8). Using the results (2.9) and (2.12) for ar,c ,

2n 2r BMN
1
r1
D2r OBMN
=
4(1)
a
O
+
O
r,r
n
n
J
J 2r+1

1
= (1)r1 ar,r (D2 )r OBMN
(2.21)
+
O
,
n
J 2r+1
so to match the BMN expression (2.16) we must have
ar,r =
22r3 (r 12 )
(2r 1)
=
.
(r)(r + 1)
(r + 1)
(2.22)
For r = 1, 2, 3, 4 this gives ar,r = 1, 1, 2, 5 as in (1.10) (note that the limit of r 1 of

the second expression in (2.22) is well defined). Combining (2.11) with (2.22) we finally
141
conclude that
ar,c =
(1)rc (2r + 1)(2r 1)

.
(r c + 1)(r + c + 1)(r)(r + 1)
(2.23)
Note that, as required (cf. (1.9)), ar,c = 0 for r < c (for integer r and c).
3. Summing up: structure of D (1) to all orders in
Let us now try to draw some conclusions about the exact structure of the spinspin
part of the dilatation operator D (1) in (1.11) to all orders in . First, let us observe that the
value of ar,r found in (2.22) implies that summing (2.13) over r with dr in (2.12) gives
a very simple formula for the quadratic in n (small fluctuation or BMN) part of the
coherent state effective Hamiltonian in (2.2):

D (1) =
r=1
r (1)
D
(4)2r 2r
2
L

d 1
1
2
.
n 1 1 n + O
2 4
L
(3.1)
Remarkably, this is the same expression that follows from the classical AdS5 S 5 string
sigma model action expanded in the limit (1.13) (Eq. (2.90) in [21]). There is a closely
related square root formula (see Eq. (3.25) in [21]) that expresses D (1) as a function of D2
in the dilute gas approximation:

(1)
(1)
1+2
D2 1
.
D =
(3.2)
(4)2
The superscript (1) in the right-hand side means that one should drop all terms with
higher than first power of independent Qa,b s (written in the factorized form) in the products of D2 . This relation should be understood in the sense of equality of the n2 terms in
the coherent-state expectation values of the two sides.7
Next, let us substitute the values (2.23) for the coefficients ar,c we have found above
into D (1) in (1.11) and try to formally perform the summation over r first, independently
for each Qa,a+c term. We get
D (1) = 2
L

a=1 c=1
fc ()Qa,a+c ,
fc () =

r=c
r
ar,c .
(4)2r
(3.3)
7 If we wanted to apply D (1) to a single trace operator, the compact expression in the right-hand side of (3.2)
would be of limited advantage. Multiplying D2 s and taking the single Qa,b part do not commute, and we would
have to explicitly compute all powers of D2 and then apply the (1)-operation to them (like it was done in
Appendix C of [21] where connected expectation values of products of some operators were computed).
142
Remarkably, the series representation for the coefficients fc () can then be summed up in
terms of the standard hypergeometric functions (see Appendix A):
r
(1)rc (2r 1)(2r + 1)
2r
(4) (r + 1)(r)(r c + 1)(r + c + 1)
r=c

c (c 12 )
1
1
2 F1 c 2 , c + 2 ; 2c + 1; 2 .
4 2 4 (c + 1)
fc () =
(3.4)
2
The coefficient in front of 2 F1 is equal to (4)
2c ac,c (cf. (2.22)). The fc go to 0 rapidly at
large c, so we effectively have a spin chain with short range interactions.
In general, the hypergeometric functions 2 F1 (a1 , a2 ; b1; z) have a cut in the z plane
running from 1 to . Note also that y(z) = 2 F1 (c 12 , c + 12 ; 2c +1; z) solves the following
differential equation z(1 z)y + (2c + 1)(1 z)y (c2 14 )y = 0.
The resulting coefficients fc () are smooth positive functions of having regular expansion at both small (see (3.4)) and large

2
1
1
fc () = 2
(3.5)
+
1
2H
+
O
,
ln
c 21
4c2 1 4
2
2
where the harmonic numbers Hp are defined in Appendix A. The square root fc
asymptotics of (3.5) is related to the cut structure of 2 F1 . The ln subleading terms are
likely to be an artifact of our resummation procedure (they will be absent in the explicit
L = 4 example discussed below).
One may be tempted to interpret the behavior of the coefficients fc as an indication
of how anomalous dimensions of particular operators should scale with (one should remember of course that D (1) is only a part of the full dilatation operator in (1.11)). Their
asymptotics may then seem to be in contradiction with the usual expectation that dimensions of generic operators corresponding to string modes should have slower
growth
with they should scale as square root of the effective string tension, i.e., as 4 [5].
A possible resolution of this paradox is that the above resummation procedure leading
to (3.3), (3.4) is useful only in the infinite chain L limit, i.e., it corresponds to the
case when D acts on long operators. The latter should be dual to semiclassical
string
modes for which dimensions are expected to grow as string tension [5]. Indeed, we
have treated all Qa,a+c terms as independent but for finite L the terms with c and c + mL
are the same because of the periodicity of the chain (implied by cyclicity of the trace in the
operators). Also, under the sum over a one has Qa,a+c = Qa,a+Lc , i.e.,
c = c+mL = mLc ,
L
Qa,a+c ,
(3.6)
a=1
where m is any positive integer number. Therefore, for finite L the sum over c should, in
fact, be restricted to run from c = 1 to c = L 1,
D
(1)
=2
L

a=1 c=1
fc () Qa,a+c =
L L1

a=1 c=1
hc (L, )Qa,a+c ,
(3.7)
143
or, equivalently,
D (1) = 2
[L/2]
hc (L, )c .
(3.8)
c=1
The new coefficients hc depend on both the t Hooft coupling and the length of the
chain L,
hc (L, )

fc+mL () + fLc+mL () ,
m=0
hL/2 (L, )
fL/2+mL (),
c = 1, . . . , L 1,
if L is even.
c = L/2,
(3.9)
m=0
They satisfy the periodicity condition hc (L, ) = hLc (L, ), reflecting the fact that for
pairwise interactions it matters only which sites participate in the interaction. Explicitly,
the coefficients appearing in (3.9) can be written using (3.4) as

c+mL (c + mL 12 )
hc (L, ) =
4 2
4 (c + mL + 1)
m=0

2 F1 c + mL 12 , c + mL + 12 ; 2c + 2mL + 1; 2
+ (c L c).
(3.10)
When L and 0 < c L, the only contribution to the sum (3.10) comes from m = 0 in
the first term, and (3.10) reduces to (3.4). For finite L, we may expand the hypergeometric
functions in (3.10) at large as in (3.5) and then do the sum over m. Ignoring the issue of
convergence of the resulting strong-coupling expansion, that leads to the following simple
result for the leading-order term

1
sin L
hc (L, ) =
(3.11)
+O
.
2L cos L cos 2c
L
In the c L limit (3.11) reduces back to the large asymptotics (3.5) of the m = 0 term
of the sum in (3.10). Using the integral representation (A.5) one can write the sum over m
in (3.10) as
1

hc (L, ) =
dt
2 2

0
dy
y 2cL + y L2c
.
y L y L
(3.12)
( 2 +t)
t (1t)
Representation (3.12) may be useful for studying various limits of hc (L, ).

It is possible that for finite L the contributions of higher order
Qn interaction terms
4
in D (1.11) may transform this asymptotics into the expected behavior. At the same
(1)
time, one may wonder if our basic assumption about the structure of D2r
in (1.9) actually
144
applies for finite values of L and all values of r. After all, to fix the coefficients ar,c we
used the condition of regularity of the scaling limit which assumes that L .8
Ignoring this cautionary note let us go ahead and apply the above relations to the first
non-trivial small L caseL = 4 (the operators with lengths L = 2, 3 are BPS: the antisymmetric combinations vanish because of trace cyclicity). The non-BPS operator with L = 4
is the level four descendant K of the Konishi scalar operator
K = tr[X, Z]2 = 2 tr(XZXZ XXZZ).
(3.13)
The action of D (1) (3.7) on K is determined by noting that (see (3.6))

1 K = 3 K = 6K,
2 K = 0,
D (1) K = (1) K.
(3.14)
Using that the periodicity implies that 1 = 3 = 5 = 7 = one finds then directly
from (1.11)

2k
2k1

k
(1)
a2k,2p1 +
a2k1,2p1 .
=
(3.15)
16 2
16 2
k=1 p=1
Using (2.23) the sums over p can be found explicitly

k
a2k,2p1 =
p=1
24k2(4k 1)
,
(2k)(2k + 1)
k

p=1
a2k1,2p1 =
24k4(4k 3)
,
(2k)(2k 1)
(3.16)
and finally we obtain the following surprisingly simple result (cf. (A.2))

3
(1)
=
1+ 2 1 .
2
Then for small we reproduce the previously known results [23,31]

2
3
4
(1)
=3 2 3
+6
15
+ O 5 ,
2
2
2
4
4
4
4
(3.17)
(3.18)
while in the large limit one finds
(1)

1
3
3
3
=
+ + O 3/2 ,
2
2 2
(3.19)
where the leading term agrees with (3.11). Note also that in the 3 and 4 terms in (3.18)
(1)
(1)
we included only the contributions of D6 and D8 , i.e., the linear in Q terms in (1.6)
and (1.7). The contributions of the QQ terms in (1.6) and (1.7) change the 3- and 4-loop
705
coefficients in (3.18) from 6 to 21
4 [23] and from 15 to 64 [24] (but see footnote 8).
Again, one may hope that a systematic account of contributions of all higher D(n) terms in
(1.8) will
the strong-coupling asymptotics of the dimension of the Konishi operator
change
from to 4 .
8 Possible subtleties in applying the general expression for the dilatation operator to operators with small
length L were mentioned in [23] (footnote 18), [24] (footnote 4) and [25] (Section 4.3).
145
4. Concluding remarks
Inspired by recent work in [23,24] and especially [19], in this paper we suggested to
organize the dilatation operator as an expansion (1.11), (1.8) in powers of independent
projection operators Qa,b [19,21] at L sites of spin chain
D = D0 +
D (n)
n=1
with D0 = L, D (1) =
L L1

hc (L, )Qa,a+c , . . .
(4.1)
a=1 c=1
where D (n) are given by sums of products of n Qs at independent sites of the spin chain.
We determined the coefficients in D (1) by demanding that its BMN-type scaling limit
[18,21] be regular, and found that it admits a very simple representation (3.3), applicable at least in the large L limit. This representation includes all
orders in and suggests
that the corresponding anomalous dimensions should grow as for large .
A natural extension of this work would be to try to find the next term in the expansion
(4.1), namely
D (2) =
L
L1
hc1 ;c2 ,c3 (L, )Qa,a+c1 Qa+c2 ,a+c2 +c3 .
(4.2)
a=1 c1 ,c2 ,c3 =1
The prime on the sum here means that certain terms should be omitted: since Qa,b Qb,c +
Qb,c Qa,b = Qa,b + Qb,c Qa,c the terms with c2 = c1 and c2 = c3 c1 have already
been included in D (1) . The contributions of higher D (n) terms should be crucial
at finite L,
resolving, in particular, the above-mentioned
contradiction
between
the
asymptotics
of the coefficients in D (1) and the expected 4 scaling of dimensions of operators corresponding to string modes.
The AdS/CFT duality suggests that D should correspond to an integrable spin chain.
The simplest possibility could be that, by analogy with the Inozemtsev chain [19], the
operator D (1) (with interaction coefficients given by (3.10)) represents a Hamiltonian of an
integrable spin 1/2 chain, while all higher order terms D (n) are effectively determined by
D (1) through integrability. This, however, is unlikely in view of the low loop order results of
[2325] and the very recent paper [32] suggesting that the full dilatation operator satisfying
the requirements of integrability, BMN scaling and consistency with gauge theory should
be essentially unique.
Acknowledgements
We are grateful to G. Arutyunov, N. Beisert, S. Frolov, M. Kruczenski, A. Parnachev,
A. Onishchenko, M. Staudacher and K. Zarembo for useful discussions, e-mail correspondence on related issues and comments on an earlier version of this paper. The work of
A.R. was supported in part by NSF grants PHY99-73935 and PHY04-01667. The work of
146
A.T. was supported by DOE grant DE-FG02-91ER40690, the INTAS contract 03-51-6346
and Wolfson award. We also thank the Michigan Center for Theoretical Physics for the
hospitality during the beginning of this work.
Appendix A. Some useful relations and definitions
Here we summarize some useful formulae used in the paper. The usual binomial expansion is given by
(1 + x)n =

k=0
(n + 1)
xk
(n k + 1)(k + 1)
(A.1)
(for integer n the series terminates at k = n). For n = 12 (A.1) can be written as

(1)k1 (2k 1) x k
(1 + x)1/2 = 1 + 2
.
(k)(k + 1)
4
(A.2)
k=1
To transform the arguments of -functions one uses

22z1
.
(z)(1 z) =
(2z) = (z) z + 12 ,
(A.3)
sin z
The hypergeometric functions are given, within the radius of convergence, by the series
p Fq (a1 , . . . , ap ; b1 , . . . , bq ; z) =

(a1 )k . . . (ap )k zk
,
(b1 )k . . . (bq )k k!
(a)k
k=0
(a + k)
.
(a)
(A.4)
They reduce to simpler functions in particular cases; for example, one finds 2 F1 (n, a;
a, z) = (1 + z)n as one can see by comparing (A.1) and (A.4). If p = q + 1,
p Fq (a1 , . . . , ap ; b1 , . . . , bq ; z) have a branch cut in the z-plane running from z = 1 to
. p Fq satisfy second order differential equations, and by a change of variables one can
relate the values of p Fq at z and at 1/z, although for different arguments ai and bj . There
is an integral representation
(c)
2 F1 (a, b; c; z) =
(b)(c b)
1
dt t b1 (1 t)cb1 (1 tz)a .
(A.5)
One can show that for large z

2 F1 (b 1, b; 2b; z)

(z)1b (2b)
(b)(b + 1)
1+

b(b 1)
1 + ln(z) 2Hb 1 + O z2 ,
2
(z)
(A.6)
where the Harmonic numbers are given by

Hp Hp(1),
Hp(s) = (s, 1) (s, p + 1),
(s, p)

k=0
(k + p)s , (A.7)

(s)
and for integer p > 1 one has Hp =

(b) =
(b)
n
k=1 k
s . One finds that
147
Hb 1 = E + (b), where
2
(b) , and E = (1) 0.577216 is the Eulers constant.
References
[1] F. Gliozzi, J. Scherk, D.I. Olive, Supersymmetry, supergravity theories and the dual spinor model, Nucl.
Phys. B 122 (1977) 253;
L. Brink, J.H. Schwarz, J. Scherk, Supersymmetric YangMills theories, Nucl. Phys. B 121 (1977) 77.
[2] M.F. Sohnius, P.C. West, Conformal invariance in N = 4 supersymmetric YangMills theory, Phys. Lett. B
100 (1981) 245;
L. Brink, O. Lindgren, B.E.W. Nilsson, The ultraviolet finiteness of the N = 4 YangMills theory, Phys.
Lett. B 123 (1983) 323;
P.S. Howe, K.S. Stelle, P.K. Townsend, Miraculous ultraviolet cancellations in supersymmetry made manifest, Nucl. Phys. B 236 (1984) 125.
[3] J.M. Maldacena, The large N limit of superconformal field theories and supergravity, Adv. Theor. Math.
Phys. 2 (1998) 231, hep-th/9711200;
S.S. Gubser, I.R. Klebanov, A.M. Polyakov, Gauge theory correlators from non-critical string theory, Phys.
Lett. B 428 (1998) 105, hep-th/9802109;
E. Witten, Anti-de Sitter space and holography, Adv. Theor. Math. Phys. 2 (1998) 253, hep-th/9802150.
[4] D. Berenstein, J.M. Maldacena, H. Nastase, Strings in flat space and pp waves from N = 4 super Yang
Mills, JHEP 0204 (2002) 013, hep-th/0202021.
[5] 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.
[6] D.J. Gross, A. Mikhailov, R. Roiban, Operators with large R charge in N = 4 YangMills theory, Ann.
Phys. 301 (2002) 31, hep-th/0205066;
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.
[7] 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.
[8] S. Frolov, A.A. Tseytlin, Semiclassical quantization of rotating superstring in AdS5 S 5 , JHEP 0206 (2002)
007, hep-th/0204226.
[9] J.G. Russo, Anomalous dimensions in gauge theories from rotating strings in AdS(5) S(5), JHEP 0206
(2002) 038, hep-th/0205244.
[10] J.A. Minahan, Circular semiclassical string solutions on AdS5 S 5 , Nucl. Phys. B 648 (2003) 203, hepth/0209047.
[11] J.A. Minahan, K. Zarembo, The Bethe-ansatz for N = 4 super YangMills, JHEP 0303 (2003) 013, hepth/0212208.
[12] S. Frolov, A.A. Tseytlin, Multi-spin string solutions in AdS5 S 5 , Nucl. Phys. B 668 (2003) 77, hepth/0304255.
[13] N. Beisert, J.A. Minahan, M. Staudacher, K. Zarembo, Stringing spins and spinning strings, JHEP 0309
(2003) 010, hep-th/0306139.
[14] S. Frolov, A.A. Tseytlin, Rotating string solutions: AdS/CFT duality in non-supersymmetric sectors, Phys.
Lett. B 570 (2003) 96, hep-th/0306143.
[15] N. Beisert, S. Frolov, M. Staudacher, A.A. Tseytlin, Precision spectroscopy of AdS/CFT, JHEP 0310 (2003)
037, hep-th/0308117.
[16] G. Arutyunov, M. Staudacher, Matching higher conserved charges for strings and spins, JHEP 0403 (2004)
004, hep-th/0310182;
G. Arutyunov, M. Staudacher, Two-loop commuting charges and the string/gauge duality, hep-th/0403077.
[17] J. Engquist, J.A. Minahan, K. Zarembo, YangMills duals for semiclassical strings on AdS5 S 5 , JHEP 0311
(2003) 063, hep-th/0310188.
[18] M. Kruczenski, Spin chains and string theory, hep-th/0311203.
148
[19] D. Serban, M. Staudacher, Planar N = 4 gauge theory and the Inozemtsev long range spin chain, hepth/0401057.
[20] V.A. Kazakov, A. Marshakov, J.A. Minahan, K. Zarembo, Classical/quantum integrability in AdS/CFT,
hep-th/0402207.
[21] M. Kruczenski, A.V. Ryzhov, A.A. Tseytlin, Large spin limit of AdS5 S 5 string theory and low energy
expansion of ferromagnetic spin chains, hep-th/0403120.
[22] A.V. Kotikov, L.N. Lipatov, V.N. Velizhanin, Anomalous dimensions of Wilson operators in N = 4 SYM
theory, Phys. Lett. B 557 (2003) 114, hep-ph/0301021;
A.V. Kotikov, L.N. Lipatov, A.I. Onishchenko, V.N. Velizhanin, Three-loop universal anomalous dimension
of the Wilson operators in N = 4 SUSY YangMills model, hep-th/0404092.
[23] N. Beisert, C. Kristjansen, M. Staudacher, The dilatation operator of N = 4 super-YangMills theory, Nucl.
Phys. B 664 (2003) 131, hep-th/0303060.
[24] N. Beisert, Higher loops, integrability and the near BMN limit, JHEP 0309 (2003) 062, hep-th/0308074.
[25] N. Beisert, The su(2|3) dynamic spin chain, hep-th/0310252.
[26] T. Klose, J. Plefka, On the integrability of large N plane-wave matrix theory, Nucl. Phys. B 679 (2004) 127,
hep-th/0310232.
[27] F.D.M. Haldane, Exact JastrowGutzwiller resonating valence bond ground state of the spin 1/2 antiferromagnetic Heisenberg chain with 1/R 2 exchange, Phys. Rev. Lett. 60 (1988) 635;
S. Shastry, Exact solution of an S = 1/2 Heisenberg antiferromagnetic chain with long ranged interactions,
Phys. Rev. Lett. 60 (1988) 639.
[28] V.I. Inozemtsev, On the connection between the one-dimensional S = 1/2 Heisenberg chain and Haldane
Shastry model, J. Stat. Phys. 50 (1990) 1143;
V.I. Inozemtsev, Integrable Heisenbergvan Vleck chains with variable range exchange, Phys. Part. Nucl. 34
(2003) 166, Fiz. Elem. Chast. Atom. Yadra 34 (2003) 332 (in Russian), hep-th/0201001.
[29] E.H. Fradkin, Field Theories of Condensed Matter Systems, Frontiers in Physics, vol. 82, AddisonWesley,
Redwood City, CA, 1991, 350 p.
[30] N. Beisert, BMN operators and superconformal symmetry, Nucl. Phys. B 659 (2003) 79, hep-th/0211032.
[31] D. Anselmi, The N = 4 quantum conformal algebra, Nucl. Phys. B 541 (1999) 369, hep-th/9809192.
[32] N. Beisert, V. Dippel, M. Staudacher, A novel long range spin chain and planar N = 4 super-YangMills,
hep-th/0405001.
Brane solutions with/without rotation

in pp-wave spacetime
Rashmi R. Nayak a , Kamal L. Panigrahi b,1 , Sanjay Siwach c
a Institute of Physics, Bhubaneswar 751 005, India
b Dipartimento di Fisica, Universit di Roma Tor Vergata, INFN, Sezione di Roma Tor Vergata,
Via della Ricerca Scientifica 1, 00133 Roma, Italy

c Institute of Mathematical Sciences, CIT Campus, Taramani, Chennai 600 113, India

Abstract
We present two classes of brane solutions in pp-wave spacetime. The first class of branes with a
rotation parameter are constructed in an exact string background with NSNS and RR flux. The
spacetime supersymmetry is analyzed by solving the standard Killing spinor equations and is shown
to preserve the same amount of supersymmetry as the case without the rotation. This class of branes
do not admit regular horizon. The second class of brane solutions are constructed by applying a
null Melvin twist to the brane solutions of flat spacetime supergravity. These solutions admit regular
horizon. We also comment on some thermodynamic properties of this class of solutions.
PACS: 11.25.-w; 11.25.Sq
1. Introduction
Study of string theory in plane wave (or pp-wave) background has drawn lots of attention in the last couple of years, in search of establishing AdS/CFT like dualities. These
E-mail addresses: rashmi@iopb.res.in (R.R. Nayak), kamal.panigrahi@roma2.infn.it (K.L. Panigrahi),
sanjay@imsc.res.in (S. Siwach).
1 INFN fellow.
doi:10.1016/j.nuclphysb.2004.07.042
150
R.R. Nayak et al. / Nuclear Physics B 698 (2004) 149162
backgrounds can be seen as a small deformation of ten-dimensional Minkowski spacetime [1]. Plane wave (or pp-wave) spacetime qualifies, the most, for analyzing certain
issues of quantum gravity and give a consistent background for studying string theory
in light-cone gauge. These backgrounds are obtained by applying PenroseGuven limit
on AdSp S q type of geometries and also from the near horizon geometries of various
supergravity solutions in diverse dimensions. Of particular interest, is the maximally supersymmetric pp-wave background which is obtained from the near horizon geometry of
coincident D3-branes in ten-dimensional spacetime in the Penrose limit. String theory in
this background is exactly solvable in light-cone gauge and is shown to be dual to N = 4
super-YangMills theory in large R-charge sector [2]. The pp-wave/CFT dualities have
been analyzed (see [3] for the updated references on this subject), and some speculations
have been made regarding the holography in plane wave backgrounds. Whereas the above
issues are slightly more clear in backgrounds with NSNS 3-form flux [49] (e.g., the
NappiWitten backgrounds and AdS3 S 3 spacetime), they are not very profound in the
case of maximally supersymmetric plane wave background with RR flux.
Plane wave spacetime in the presence of non-constant flux is also an interesting background to study string theory, as it provides examples of integrable models on the worldsheet [10,11]. These backgrounds can also be interpreted as the deformation of the flat
spacetime and are supported by null matter fields. The corresponding worldsheet theory is
described by the non-linear sigma model [1013] and represent the non-trivial examples
of interacting theories in light-cone gauge. The pp-wave backgrounds with non-constant
3-form NSNS (H3 ) and RR (F3 ) flux do not admit, in contrast to their 5-form RR flux
(F5 ) counterpart, the linearly realized supernumerary killing spinors. Moreover these theories are closely related to the closed strings in a constant magnetic field, in the presence of
antisymmetric tensor fields and a non-trivial metric [14], which also provides an example
of exact string theory background. These backgrounds are known to be homogeneous
plane wave backgrounds and string theory in these type of spacetime has been analyzed in
great detail [15,16].
In recent years, several important aspects of string dualities have been revealed and
D-branes have played important role in these developments. While D-branes can be treated
as the black branes in supergravity theories, the effective field theory on the brane are
of super-YangMills type theories. So the study of D-branes in various non-trivial and
non-generic backgrounds with/without flux give ideas about the structures of string theory as well as the related gauge theories. D-branes in various curved background have
generated renewed interest in the context of plane wave background for various reasons.
First, these non-perturbative objects are easily tractable in pp-wave background and second, the supergravity solutions can be constructed with not much efforts as compared to
its AdS counterpart. Various D-brane supergravity solutions in maximally and less than
maximally supersymmetric plane wave backgrounds have been analyzed in the past couple
of years [1728]. In this context various attempts have also been made in finding out black
hole/brane solutions with regular horizon in plane wave spacetime. The analysis of [25,29]
shows strong evidence in favour of the non-existence of horizon in the spacetime with covariantly constant and null Killing vectors, proposed in [30]. Moreover, the brane solutions
seem to be singular. In [31], however, a consistent method for obtaining a black string
solution with regular horizon has been discussed, which relies on a solution generating
151
technique [32] known as null Melvin twist (NMT). This particular mechanism, transforms
a flat spacetime to a plane wave spacetime. So the natural guess would be to start with a
black brane solution in flat spacetime and apply NMT to obtain a solution in plane wave
spacetime which preserves horizon. As proved in [31], this particular transformation indeed preserve the horizon and that the area of the horizon remains the same even after the
NMT.
Motivated by the recent interest in finding out exact string backgrounds and the classical
solutions of branes and their bound states, in various non-trivial backgrounds with/without
flux, in this paper, we present some Dp and DpDp branes in homogeneous pp-wave
background with non-constant flux and with a rotation parameter. This class of solutions
are seen not to admit a regular horizon. We also examine another class of D-branes which
are obtained by applying a solution generating technique known as null Melvin twist. These
class of solutions do keep the asymptotic of the spacetime as that of the plane waves but do
not give rise to the null matter content of the theory. The rest of the paper, is organized as
follows. In Section 2, after a small digression for the discussion of the homogeneous plane
wave spacetime, we present the classical solutions of some Dp as well as DpDp branes
in this background with an explicit inclusion of the rotation parameter. We keep the fluxes
completely general and show that they solve supergravity field equations. In Section 3, we
analyze the supersymmetry of the background and the branes in this background by solving
the Killing spinor equations. The solutions of the Killing spinor equations are shown to
constrain the structure of the 3-form fluxes. We also make some remarks on the properties
of horizon of these brane solutions. In Section 4, we present classical solutions of black
branes in asymptotically plane wave spacetime by applying the null Melvin twist on the
non-extremal brane solutions of flat spacetime supergravity. We also compute the horizon
area and temperature of these black branes. In Section 5, we conclude with some remarks
and discussions.
2. Branes with rotation in pp-wave background

2.1. The background
As a warm up exercise, below, we recapitulate few basic facts about the homogeneous
plane wave background discussed in [16], which will be helpful in fixing the notations,
etc. The most general null Brinkmann metric in d-dimensions with flat transverse space is
given by
ds 2 = 2 du dv + H(u, x) du2 + 2Ai (u, x) dx i du + dx i dxi .
(2.1)
Exact string backgrounds, with this metric, can be constructed by switching on the appropriate field strengths and the dilaton
NS
Biu
= Bi (u, x),
= (u).
(2.2)
152
For the above ansatz, the one-loop conformal invariance, or in other words, the type II
supergravity field equations give the following conditions
1
1
1
2H + u i Ai + Fij F ij Hij H ij + 2u2 = 0,
2
4
4
i Hij = 0,
i Fij = 0,
(2.3)
where Fij = i Aj j Ai and Hij = i Bj j Bi . In principle, the general solutions to

these equations do not define an exact background. Some of the special cases where it
does, has been discussed in [16].
In the present paper, we shall be interested in the supergravity background with the
metric, NSNS 2-form (B) and RR two form (B ) (and a constant dilaton)
ds 2 = 2 du dv + H(xi ) du2 + 2J Ai (xi ) du dx i +
4
8

i 2
a 2
dx +
dx ,
i=1
B = bi (xi ) du dx ,
a=5
H3 = Hij (xi ) du dx dx ,
B = J Ai (xi ) du dx ,
i
F3 = J Fij (xi ) du dx i dx j ,
(2.4)
B,
respectively: Fij =
where H3 and F3 are the field strengths associated with the B and
i Aj j Ai and Hij = i bj j bi . J is an arbitrary constant parameter. Note that we
are considering the case in which the above metric functions are independent of the lightcone time u. We keep the most general form of the NSNS and RR field strengths, but
the restrictions on them would be imposed by the requirement of supersymmetry, as we
will see in the subsequent analysis. Few remarks regarding the structure of A(x i ) are in
order now. If we restrict: J Ai dx i du = J ij x i dx j du, where 12 = 34 = 1, then this can
be interpreted as the rotation in the x i space of solutions, parametrized by J .2 That would
further restrict the background fields turned on, which in turn play an important role in the
analysis of Killing spinor equations. We will come back to this issue later on.
To be a consistent solution of supergravity, the above ansatz should be supplimented by
the constraints
2(i) H(xi ) = (i bj )2 ,
k Fki = 0,
k Hki = 0.
(2.5)
We have checked that the solution (2.4) supplimented by the conditions (2.5) satisfy
all the type IIB field equations and the Bianchi identities. We shall be interested in this
background for the subsequent analysis of the present paper. As can be seen from the
metric (2.4) that switching off the gauge field (A) we get back to the pp-wave metric of
[11] with non-constant NS flux.
2.2. Supergravity solutions
In this section, we present classical solutions of branes in the above background supported by NSNS (H3 ) and RR (F3 ) flux in the transverse direction of the branes. We
2 Similar analysis have also been performed in the context of closed strings in the presence of magnetic field,
and proved to be conformally invariant background [14].
153
start by writing down the supergravity solution of N D-strings lying on top of each other
in this background. The metric, the dilaton, and the field strengths of such a configuration
is given by

1/2
ds 2 = f1
2 du dv + H(xi ) du2 + 2J Ai (xi ) du dx i
4

8
2
a 2
1/2
i
+ f1
dx +
dx
,
i=1
a=5
H3 = Hij (xi ) du dx dx ,
e2 = f1 ,

J
B = f11 1 du dv + Ai (xi ) du dx i ,
f1
i
(2.6)
where f1 = 1 + Q1 /r 6 is the harmonic function in the transverse space of the D-string

and B is the RamondRamond potential. We have checked that the above solution
solves type IIB field equations provided the constraints (2.5) are also imposed. Similarly,
a D5-brane solution is given by

8

a 2
1/2
2
2
i
ds = f5
dx
2 du dv + H(xi ) du + 2J Ai (xi ) du dx +
a=5
4

1/2
+ f5
2
dx i ,
i=1
H3 = Hij (xi ) du dx i dx j ,
F3 = J Fij (xi ) du dx i dx j ,
e2 = f51 ,
Fij k = ijl k l f5 ,
(2.7)
where f5 = 1 + Q5 /r 2 is the harmonic function in the transverse 4-space. F3 and Fij k are
the RamondRamond field strengths. We have once again checked that the above ansatz
solves type IIB field equations provided the constraints (2.5) are also imposed. The above
solutions reduce to those presented in [24] for J = 0. Other Dp-brane (for p 2) solutions
can also be constructed first by smearing a = x 5 , . . . , x 8 directions and then by applying
T-dualities along those.
Now we present the classical solution of D1D5 system, as an example of DpD(p + 4)
brane bound state in the background (2.4). The metric, dilaton, NSNS and RR fields of
such a configurations are given by
1/2
8

a 2
f1
ds 2 = (f1 f5 )1/2 2 du dv + H(xi ) du2 + 2J Ai (xi ) du dx i +
dx
f5
a=5
+ (f1 f5 )1/2
4
2
dx i ,
i=1
H3 = Hij (xi ) du dx i dx j ,
e2 =
f1
,
f5
154

J
B = f11 1 du dv + Ai (xi ) du dx i ,
f1
Fij k = ijl k l f5 ,
(2.8)
where f1,5 = 1 + Q1,5 /r 2 are the harmonic functions of D1 and D5 brane in the common
transverse 4-space. We have once again checked that the above ansatz solves type IIB
field equations provided the identities (2.5) are also imposed. This solution reduces to the
D1D5 brane solution of [24] for J = 0. We would like to point out that a similar solution
has already been presented in [26]. However, the choice of the background flux turned on
to compensate the effect of H is different here. The corresponding differential equations
for H is given by (2.5) and is independent of the parameter J for all the solutions. The
explicit solution can be read off from Ref. [25] for the branes presented here.
3. Supersymmetry
The supersymmetry variation of dilatino and gravitini fields of type IIB supergravity in
ten dimensions, in string frame, is given by [33,34]

1
1
1
1
(1)
(3)
H + e M FM
,
=
+ F
2
12
2
12
(3.1)

1
1
wa b Ha b a b
= +
4
2

1
1
1 (5)
(3)
+ e F(1) F
F ,
(3.2)
8
3!
2.5!
where we have used (, , ) to describe the ten-dimensional spacetime indices, and hats
represent the corresponding tangent space indices.
3.1. Background supersymmetry
Before analyzing the supersymmetry of the rotating Dp and DpDp brane solutions in
homogeneous plane wave background, let us first discuss the supersymmetry of the background itself. The dilatino (3.1) and gravitino (3.2) variations impose non-trivial conditions
on the spinor . First the dilatino variation gives

H + J u
F = 0.
u
(3.3)
Similarly, from the gravitini variations, we get the following conditions on the spinors to
have non-trivial solutions

J
1
J
v
= 0,
u u + F H F u
4
4
8
v v = 0,
a a = 0,
i i = 0.
(3.4)
155
In writing down the above supersymmetry variations we have made use of the standard
supersymmetry condition3
u = 0.
(3.5)
After imposing this condition the dilatino variation, above, is satisfied. One notices that for
the variations of the remaining terms in the gravitino variation u , for a constant spinor
0 , we need to restrict the structure of the background flux Fij and also Hij . One such
(3)
(3)
(3)
(3)
possibility has been discussed in [24]. For the case: Fu12
= Fu34
and Hu12
= Hu34
, with
all other components of Fij and Hij set to zero, we have to impose the condition

1 1234 = 0.
(3.6)
So the amount of supersymmetry preserved, after imposing the above two conditions, (3.5)
and (3.6) is 1/4 of the original one. This fact has also been shown in [35]. We would
like to point out that the addition of the rotation J , does not destroy more supersymmetry
compared to the case without J . So the natural guess would be that the fate of the remaining
supersymmetry in the presence of D-branes will be the same as the case without J , that
has been explained in [24,27]. We will examine this fact by giving examples of D-branes
that we have considered in the previous section.
3.2. D-brane supersymmetry
In this section, we analyze the supersymmetry conditions for the D-string (2.6) and
the D1D5 brane bound state (2.8) solutions presented in the previous section. First, the
dilatino variation equation for the D-string solution presented in (2.6) gives
5/4

f1
J 7/4

u
f1
H + f1 u
F = 0,

4
4
= 1, . . . , 8.
u v
(3.7)
On the other hand, the gravitini variations gives the following conditions on the spinors to
have non-trivial solutions

J 1/2
1
J 1/2

v
u u + f1
F H f1 F u
= 0,
4
4
8
v u = 0,

1 i f1
i i +
= 0,
8 f1

1 a f1
a a +
= 0.
8 f1
(3.8)
In writing down the above gravitini variations, we have made use of the D-string supersymmetry condition
u v = 0,
3 This condition does not depend on the details of the pp-wave background that we are considering.
(3.9)
156
in addition to the necessary condition (3.5). By imposing (3.5), the dilatino variation is
satisfied. The gravitini variations v , i and a , solve for the spinor

1
0
= exp ln f1
,
8
0 being a spinor which can depend on the coordinate u, leaving the following equawith
tions to have a non-trivial solution4

J 1/2
1
J 1/2
0
v 0
f1
F H
f1 F u
= 0,
(3.10)
4
4
8
and
0
u
= 0.
(3.11)
One can see that the existence of the solution to the above equations can be obtained by
restricting the form of the functions Fij and Hij . By making the choice: Fu12 = Fu34 and
0
Hu12 = Hu34 , we get an additional condition on the spinors
0
1 1234
= 0.
(3.12)
Coming back to the counting of the surviving supersymmetry (which are of standard type
only), it is easy to see that the D-string solution (2.6), after imposing the conditions (3.9),
(3.12) along with (3.5), preserves 1/8 of the original supersymmetry. Therefore, even with
the addition of the rotation J , the D-string supersymmetry remains the same as in the nonrotating case, discussed in [24]. One can also show in a similar way that the D5-brane
solution presented in (2.7) preserves 1/8 of the supersymmetries.
Now let us analyze the supersymmetry of the D1D5 brane bound state solution presented in (2.8). First the dilatino variation gives

f1,

f5,
1

u v

k l kl
(f1 f5 )1/4 u
H
f1
3!
f5
3 1/4
f

+J 1
u
F = 0.
(3.13)
f5
The gravitino variations are very similar to those presented for the D-string case. Therefore,
we skip the detailed expressions for those. After imposing, respectively, the D-string and
the D5-brane supersymmetry conditions
u v
= 0,
(3.14)
and

1 k l
= 0,
3! k l
4 Since 0 is a function of u only and both F and H are functions of x i only.

ij
ij
(3.15)
157
along with the standard supersymmetry condition (3.5), the dilatino variation is fully satisfied. The gravitini variation equations, however, would require, an additional condition
0
1 1234
= 0,
(3.16)
0 ) solution similar to that presented for the D-string

for the existence of a constant spinor (
case. Let us now count the amount of supersymmetry preserved after imposing all these
conditions. First of all, u = 0 breaks half of the supersymmetries. The fate of the
remaining supersymmetries can be found out by examining the conditions (3.14)(3.16). It
is easy to see however that they indeed are only two independent conditions on the spinor .
So the D1D5 solution presented in (2.8) preserves 1/8 of the supersymmetries.
Few remarks are in order now. As can be seen from the supersymmetry analysis of the
D-branes in the pp-wave background, there is always a decoupling between the standard D-brane supersymmetry conditions and the supersymmetry condition imposed by the
pp-wave (which in turn comes from the light-cone gauge fixing). We would like to note
that the branes that we are considering here are longitudinal branes [19] (all the light-cone
directions fall into the worldvolume directions of the branes and other pp-wave directions
are transverse to the brane).
3.3. Horizon/no horizon

The plane wave spacetime with covariantly constant and null Killing isometries, as such
do not admit horizon [30]. Relaxing the covariant constancy condition raised some hope
that there might exist horizon in spacetime admitting null Killing isometry only (e.g.,
p-brane solutions in pp-wave spacetime) [29]. However, the analysis of [25,29] shows
strong evidence in favour of the non-existence of regular horizon for p-branes in pp-wave
spacetime.
To examine the issue that the solutions presented in previous section admit regular horizon or not, one would like to see how the curvature tensors behave in the near horizon limit.
Without addition of the rotation term, it has been noticed earlier that the potential divergent quantities are the components of the Riemann tensor [25,36]. An invariant measure of
the divergence are the Riemann tensor components as measured in an orthonormal frame.
A natural choice for it is the parallel transported frame as emphasized in [36]. One can
show that in the parallel transported frame some of the Riemann tensors diverge in the near
horizon geometry, thereby showing the appearance of pp-curvature singularities [25,36].
Hence their does not exist the regular event horizon. Generalization to the non-extremal
solutions along the lines of [25,29] does not improve the situation. It is not difficult to
see that the addition of rotation term does not change the situation in both extremal and
non-extremal cases.
However, it has been argued that the pp-singularities close-off of spacetime near the
horizon [36], thereby acting as the boundary of the spacetime and if one accepts this interpretation, the issue that the horizon is a singular surface becomes less important and the
solutions can be thought of well behaved.
158
4. pp-wave branes from flat spacetime brane solutions

We start by writing down the most general non-extremal p-branes in ten dimensions. The metric, dilaton and the field strengths of such non-extremal p-branes in tendimensional spacetime is given by

p

2

1/2
1/2
2
2
2
f (r) dt +
+ Hp f 1 (r) dr 2 + r 2 dd+1
dy
,
ds = Hp
=1
f (r)
E (p+1) = Q
dt dy 1 dy 2 dy p ,
Hp
Qp
Hp = 1 + ,
(4.1)
f (r) = 1 ,
d
r
rd
where Hp is the harmonic function for the p-branes which satisfies the Green function
equation in the transverse d + 2 = 9 p space and f (r) is the non-extremal parameter.
Now we apply the solution generating technique, NMT as described in [31] on (4.1) to
generate black p-branes in the plane wave spacetime. The first step involves a boost along
y1 (which is one of the isometry directions along the brane). The resulting metric and field
strength of the boosted p-brane becomes

p
2 2
1/2
2
1
2
K (r)f (r) d t + K(r)

ds = Hp
dy
d y1 + A(r) d t +
e = H
3p
4
1/2
+ Hp f 1 (r) dr 2
e = H
3p
4
+r
2
dd+1
f
E (p+1) = Q
d t d y 1 dy 2 dy p ,
Hp
(4.2)
where
dy 1 = sinh d t + cosh d y 1 ,
dt = cosh d t sinh d y 1 ,
Q
Q
A(r) = K 1 ,
K(r)
=1+ ,
rd
rd
2
Q = sinh cosh .
Q = sinh ,
y 1 .
(4.3)
The second step involves a T-duality along

The resulting solution is a boosted (p
1)-brane in ten-dimensional spacetime with the following form of the metric, dilaton and
the other fields

p

2
1/2
2
1
2
dy
K (r)f (r) d t +
ds = H
p1
2

1/2
+ Hp1 K 1 (r) d y 1
=2
+f

2
,
(r) dr 2 + r 2 dd+1
3(p1)
Hp14
e =
,
K(r)
B = A(r) dt d y 1 ,
f
E (p) = Q
d t dy 2 dy (p1).
Hp1
159
(4.4)
The next and the vital step of the NMT is to apply a twist + 2 d y1 , where
the one form is defined such that d 2 = 14 2 + d 2 . The spacetime, after this twist
d+1
d
becomes

ds
1/2
= Hp1
(r)f (r) d t +
p

dy

2
=2

1/2
+ Hp1
(r) d y

1 2
+f

2
1
(r) dr 2 + r 2 + 2 d y 1 + r 2 dd2 ,
4
3(p1)
Hp14
e =
,
K(r)
B = A(r) dt d y 1 ,
E (p) = Q
f
Hp1
d t dy 2 dy (p1).
(4.5)
Next one T-dualises back along y 1 to get back a Dp-brane solution and apply the inverse
boost along y1 . The purpose of inverse boost is to cancel the boost charge as in step one.
One gets the following configuration of metric and other fields
ds
1/2
= Hp

(A(r) cosh + sinh )2
K 1 (r)f (r) cosh2 +
dt 2
(K 1 (r) + r 2 2 )

+ 2 K 1 (r)f (r) sinh cosh +
2
1
A (r) sinh cosh
1
2
2
K (r) + r

+ A(r) cosh2 + sinh2 + sinh cosh dt dy1

p

(A sinh b + cosh )2
1
2
2
2
dy1 +
+ K(r) f (r) sinh +
dy
(K 1 (r) + r 2 2 )
=2

r 2 K 1 (r)
1
1/2 1
+ Hp
dr 2 +
2 + r 2 dd2 ,
f
4 (K 1 (r) + r 2 2 )

1
r 2
sinh + A(r) cosh dt
1
2
2
2 K (r) + r

+ cosh + A(r) sinh dy1 d,
B =
(3p)
4
e =
Hp
K(r)(
K 1 (r) + r 2 2 )
f
dt dy 1 dy p .
E (p+1) = Q
Hp
(4.6)
160
The final step is to take the limit 0 and while keeping = 12 e fixed.
The net effect of this step is to make the twist null hence the name null twist

f (r)(1 + 2 r 2 ) 2 2 2 r 2 f (r)
dt
dt dy 1
k(r)
k(r)

p
2 r 2 1 2 2
dy
+ 1
+
dy
k(r)
=2

1 2 r 4 (1 f (r)) 2
1/2 1
2
dr 2
+ r 2 dd+1
+ Hp
,
f
4
k(r)
1/2
ds 2 = Hp
(3p)
Hp 4
,
e =
k(r)
E (p+1) = Q
B =

1 r 2
f (r) dt + dy 1 ,
2 k(r)
f
dt dy 1 dy p ,
Hp
k(r) = 1 +
r d2
(4.7)
The above solution reduces to the black p-brane in flat spacetime (4.1) in 0 limit
and goes to the plane wave metric in ten dimensions asymptotically. For = 0, it is well
known that the above metric admits a regular event horizon at
r+ = 1/d .
It was noticed by the authors of [31] in the context of black string solution in asymptotically plane wave spacetime that the solution obtained by NMT also admits regular horizon
at r+ . It also remains true for the black brane solutions constructed above. Moreover, the
area of event horizon remains invariant under NMT, i.e., it is independent of . The area
of event horizon (as measured in Einstein frame) is given by
A=

d+1
k(r) 2 r 2 H r d+1 d+1

= H (r+ ) d .
(4.8)
One quarter of the area of event horizon measured in Planck units furnish the statistical
entropy of the black branes according to the laws of black hole thermodynamics. Defining
1 r d = 2 , and Euclidean time i = t, the metric can be put in the following form

2
4r+
d2 1 2 2
d/8
2
d + 2 H d + .
ds =
H
4r+
d2
2
(4.9)
The temperature of the black brane solutions is given by the inverse periodicity of the
Euclidean time
T=
d
H 1/2 (r+ ),
4r+
which is also independent of .
(4.10)
161
5. Discussion
In this paper we have presented two classes of brane solutions. The first one represent
a class of rotating Dp-branes in homogeneous plane wave spacetime with both NSNS
and RR flux. We discussed the supersymmetry of these branes and their bound states by
solving the type II Killing spinor equations explicitly. We briefly reviewed the possibility
of having a regular horizon for this class of branes. The worldsheet analysis the background and the rotating branes presented in Section 2 is rather straightforward. Following
[11,14], one can write down the bosonic sigma model action in the presence of the nonconstant RR and NSNS flux. By using the D-brane boundary conditions it is easy to
write down the mode expansions and thereby the classical Hamiltonian of the system. So
we skip the details here. Regarding the event horizon, it has been shown in [26], in the context of Godel background, that one indeed find out a regular horizon by applying suitable
T-duality/dimensional reduction as that removes the pp-singularities. One could possibly
try to find out the Godel type solutions from the branes solutions presented here and analyze the properties of horizon, the holographic screens and the closed timelike curves. The
gauge theory duals of these branes could also be found out by identifying the supergravity
modes with the corresponding operators in the boundary theory. In the line of the supergravity theories, an interesting exercise would be to find out the most general structure of
the metric and the other field strengths and try to solve the field equations. The later would
impose certain constraints of the structure of the flux compatible with the supersymmetry
of the background.
The second class of branes were constructed from the non-extremal branes in the flat
spacetime by applying a NMT along the translational isometry of the brane solutions. This
class of branes admit regular horizon and the corresponding thermodynamical quantities
like the entropy and temperature are computed. They are found to be independent of the parameter that defines the plane wave spacetime. In summary, the generic solution obtained
by null Melvin twist inherit their thermodynamic properties from the parent solution in flat
spacetime. It would be interesting to apply the procedure to construct general rotating and
charged black brane solutions using this procedure.
Acknowledgements
We would like to thank N. Ohta for useful discussions at the early stage of the work. We
also thank O. Zapata for some comments on the draft. It is a pleasure to thank M. Bianchi
and A. Sagnotti for encouragement and for useful discussions. R.R.N. would like to thank
string theory group at the Universit di Roma, Tor Vergata for their kind hospitality where
part of the work was done. The research of K.P. was supported in part by INFN, by the E.C.
RTN programs HPRN-CT-2000-00122 and HPRN-CT-2000-00148, by the INTAS contract 99-1-590, by the MURST-COFIN contract 2001-025492 and by the NATO contract
PST.CLG.978785.
162
References
[1] M. Blau, J. Figuero-OFarrill, C. Hull, G. Papadopoulos, JHEP 0001 (2000) 047, hep-th/0110242;
M. Blau, J. Figuero-OFarrill, G. Papadopoulos, Class. Quantum Grav. 19 (2000) L87, hep-th/0201081;
M. Blau, J. Figuero-OFarrill, G. Papadopoulos, hep-th/0202111.
[2] D. Berenstein, J. Maldacena, H. Nastase, JHEP 0204 (2002) 013, hep-th/0202021.
[3] D. Sadri, M.M. Sheikh-Jabbari, hep-th/0310119.
[4] E. Kiritsis, B. Pioline, JHEP 0208 (2002) 048, hep-th/0204004.
[5] O. Lunin, S.D. Mathur, Nucl. Phys. B 642 (2002) 91, hep-th/0206107.
[6] J. Gomis, L. Motl, A. Strominger, JHEP 0211 (2002) 016, hep-th/0206166.
[7] E. Gava, K.S. Narain, JHEP 0212 (2002) 023, hep-th/0208081.
[8] G. DAppollonio, E. Kiritsis, Nucl. Phys. B 674 (2003) 80, hep-th/0305081.
[9] M. Bianchi, G. DAppollonio, E. Kiritsis, O. Zapata, JHEP 0404 (2004) 074, hep-th/0402004.
[10] J. Maldacena, L. Maoz, JHEP 0212 (2002) 046, hep-th/0207284.
[11] J.G. Russo, A.A. Tseytlin, JHEP 0209 (2002) 035, hep-th/0208114.
[12] N. Kim, Phys. Rev. D 67 (2003) 046005, hep-th/0212017.
[13] G. Bonelli, JHEP 0301 (2003) 065, hep-th/0301089.
[14] J.G. Russo, A.A. Tseytlin, Nucl. Phys. B 448 (1995) 293, hep-th/9411099.
[15] D. Sadri, M.M. Sheikh-Jabbari, JHEP 0306 (2003) 005, hep-th/0304169.
[16] M. Blau, M. OLoughlin, G. Papadopoulos, A.A. Tseytlin, Nucl. Phys. B 673 (2003) 57, hep-th/0304198.
[17] A. Kumar, R.R. Nayak, S. Siwach, Phys. Lett. B 541 (2002) 183, hep-th/0204025.
[18] K. Skenderis, M. Taylor, JHEP 0206 (2002) 025, hep-th/0204054;
K. Skenderis, M. Taylor, Nucl. Phys. B 665 (2003) 3, hep-th/0211011;
K. Skenderis, M. Taylor, JHEP 0307 (2003) 006, hep-th/0212184.
[19] P. Bain, P. Meessen, M. Zamaklar, Class. Quantum Grav. 20 (2003) 913, hep-th/0205106.
[20] M. Alishahiha, A. Kumar, Phys. Lett. B 542 (2002) 130, hep-th/0205134.
[21] A. Biswas, A. Kumar, K.L. Panigrahi, Phys. Rev. D 66 (2002) 126002, hep-th/0208042.
[22] R.R. Nayak, Phys. Rev. D 67 (2003) 086006, hep-th/0210230.
[23] L.F. Alday, M. Cirafici, JHEP 0305 (2003) 006, hep-th/0301253.
[24] K.L. Panigrahi, S. Siwach, Phys. Lett. B 561 (2003) 284, hep-th/0303182.
[25] N. Ohta, K.L. Panigrahi, S. Siwach, Nucl. Phys. B 674 (2003) 306, hep-th/0306186.
[26] D. Brecher, U.H. Danielsson, J.P. Gregory, M.E. Olsson, JHEP 0311 (2003) 033, hep-th/0309058.
[27] R.R. Nayak, K.L. Panigrahi, Phys. Lett. B 575 (2003) 325, hep-th/0310219.
[28] S.F. Hassan, R.R. Nayak, K.L. Panigrahi, hep-th/0312224.
[29] J.T. Liu, L.A. Pando Zayas, D. Vaman, Class. Quantum Grav. 20 (2003) 4343, hep-th/0301187.
[30] V.E. Hubeny, M. Rangamani, JHEP 0211 (2002) 021, hep-th/0210234.
[31] E.G. Gimon, A. Hashimoto, V.E. Hubeny, O. Lunin, M. Rangamani, JHEP 0308 (2003) 035, hepth/0306131.
[32] M. Alishahiha, O.J. Ganor, JHEP 0303 (2003) 006, hep-th/0301080.
[33] J.H. Schwarz, Nucl. Phys. B 226 (1983) 269.
[34] S.F. Hassan, Nucl. Phys. B 568 (2000) 145, hep-th/9907152.
[35] C.A.R. Herdeiro, Nucl. Phys. B 665 (2003) 189, hep-th/0212002.
[36] D. Brecher, A. Chamblin, H.S. Reall, Nucl. Phys. B 607 (2001) 155, hep-th/0012076.

Supersymmetric PatiSalam models from

intersecting D6-branes: a road to the Standard Model
Mirjam Cvetic a , Tianjun Li b , Tao Liu a
a Department of Physics and Astronomy, University of Pennsylvania, Philadelphia, PA 19104, USA
b School of Natural Science, Institute for Advanced Study, Einstein Drive, Princeton, NJ 08540, USA
Received 22 March 2004; received in revised form 4 June 2004; accepted 23 July 2004
Abstract
We provide a systematic construction of three-family N = 1 supersymmetric PatiSalam models
from Type IIA orientifolds on T6 /(Z2 Z2 ) with intersecting D6-branes. All the gauge symmetry factors SU(4)C SU(2)L SU(2)R arise from the stacks of D6-branes with U (n) gauge
symmetries, while the hidden sector is specified by USp(n) branes, parallel with the orientifold planes or their Z2 images. The PatiSalam gauge symmetry can be broken down to the
SU(3)C SU(2)L U (1)BL U (1)I3R via D6-brane splittings, and further down to the Standard Model via D- and F-flatness preserving Higgs mechanism from massless open string states in a
N = 2 subsector. The models also possess at least two confining hidden gauge sectors, where gaugino condensation can in turn trigger supersymmetry breaking and (some) moduli stabilization. The
systematic search yields 11 inequivalent models: 8 models with less than 9 Standard Model Higgs
doublet-pairs and 1 model with only 2 Standard Model Higgs doublet-pairs, 2 models possess at
the string scale the gauge coupling unification of SU(2)L and SU(2)R , and all the models possess
additional exotic matters. We also make preliminary comments on phenomenological implications
of these models.
PACS: 11.10.Kk; 11.25.Mj; 11.25.Uv
E-mail address: tli@sns.ias.edu (T. Li).

doi:10.1016/j.nuclphysb.2004.07.036
164
M. Cvetic et al. / Nuclear Physics B 698 (2004) 163201
1. Introduction
Prior to the second string revolution the efforts in string phenomenology focused on
constructions of four-dimensional solutions in the weakly coupled heterotic string theory:
the goal was to construct N = 1 supersymmetric models with features of the Standard
Model (SM). On the other hand, the M-theory unification possesses in addition to its perturbative heterotic string theory corner, also other corners such as perturbative Type I,
Type IIA and Type IIB superstring theory, which should provide new potentially phenomenologically interesting four-dimensional string solutions, related to the heterotic ones via
a web of string dualities. In particular, the advent of D-branes [1], as boundaries of open
strings, plays an important role in constructions of phenomenologically interesting models
in Type I, Type IIA and Type IIB string theories. Conformal field theory techniques in the
open string sectors, which end on D-branes, allow for exact constructions of consistent
4-dimensional supersymmetric N = 1 chiral models with non-Abelian gauge symmetry on
Type II orientifolds. Within this framework chiral matters can appear (i) due to D-branes
located at orbifold singularities with chiral fermions appearing on the worldvolume of such
D-branes [28] and/or (ii) at the intersections of D-branes in the internal space [9]. (These
latter models also have a T-dual description in terms of magnetized D-branes [10,11].)
Within the models with intersecting D6-brane on Type IIA orientifolds [1214], a
large number of non-supersymmetric three-family Standard-like models and Grand Unified
models were constructed [1225]. These models satisfy the RamondRamond (RR) tadpole cancellation conditions, however, since the models are non-supersymmetric, there are
uncancelled NeveuSchwarzNeveuSchwarz (NSNS) tadpoles. In addition, the string
scale is close to the Planck scale because the intersecting D6-branes typically have no
common transverse direction in the internal space. Therefore, these models typically suffer from the large Planck scale corrections at the loop level, i.e., there exists the gauge
hierarchy problem.
On the other hand, the supersymmetric models [26,27] with quasi-realistic features of
the supersymmetric Standard-like models have been constructed in Type IIA theory on
T 6 /(Z2 Z2 ) orientifold with intersecting D6-branes. Subsequently, a larger set of supersymmetric Standard-like models and a PatiSalam model [28], as well as a systematic
construction of supersymmetric SU(5) Grand Unified models [29] have been constructed.
Their phenomenological consequences, such as renormalization group running for the
gauge couplings, supersymmetry breaking via gaugino condensations, moduli stabilization, and the complete Yukawa couplings that include classical and quantum contributions,
have been studied [3033]. Furthermore, the supersymmetric PatiSalam models based
on Z4 and Z4 Z2 orientifolds with intersecting D6-branes were also constructed [34,35].
In these models, the leftright gauge symmetry was obtained via brane recombinations, so
the final models do not have an explicit toroidal orientifold construction, where the conformal field theory can be applied for the calculation of the full spectrum and couplings.
We shall concentrate on constructions of supersymmetric three-family PatiSalam models based on T 6 /(Z2 Z2 ) orientifold. Although previous constructions provided a number
of supersymmetric three family examples with Standard-like gauge group (or Grand Unified group), these models have a number of phenomenological problems. Previous models
had at least 8 pairs of the SM Higgs doublets and a number of exotic particles, some of
165
them fractionally charged. In addition, in the previous supersymmetric SM constructions

[2628], there are at least two extra anomaly free U (1) gauge symmetries. They could be
in principle spontaneously broken via Higgs mechanism by the scalar components of the
chiral superfields with the quantum numbers of the right-handed neutrinos, however, they
break D-flatness conditions, and thus supersymmetry, so the scale of symmetry breaking
should be near the electroweak scale. On the other hand, there were typically no candidates, preserving D-flatness and F-flatness conditions, which could break these gauge
symmetries at an intermediate scale. In addition, there is no gauge coupling unification.
(The gauge coupling unification can be realized in the quasi-supersymmetric U (n)3 models [39]. But the filler branes, which are on top of orientifold planes in these models, are
anti-branes and break the supersymmetry at the string scale.) Furthermore, there exist three
multiplets in the adjoint representation for each U (n) group, which is a generic property
for the supersymmetric and non-supersymmetric toroidal orientifold constructions because
the typical three cycles wrapped by D6-branes are not rigid. One possible solution is that
we consider the CalabiYau compactifications with rigid supersymmetric cycles, but, the
calculational techniques of conformal field theory may not be applicable there. (For studies
of N = 1 supersymmetric solutions of D-branes on CalabiYau manifolds see [3638] and
references therein.)
Well-motivated by the Standard Model constructions, we shall study systematically the
three family N = 1 supersymmetric PatiSalam models from Type IIA orientifolds on
T6 /(Z2 Z2 ) with intersecting D6-branes where all the gauge symmetries come from
U (n) branes. On the one hand, PatiSalam model provides a natural origin of U (1)BL
and U (1)I3R both of which are generically required due to the quantum numbers of the SM
fermions and the hypercharge interaction in the SM building from intersecting D6-brane
scenarios. On the other hand, it also provides one road to the SM without any additional
anomaly-free U (1)s near the electroweak scale, which was a generic feature of the previous supersymmetric SM constructions [28].
The paper is organized in the following way. In Section 2 we briefly review the rules
for supersymmetric model building with intersecting D6-branes on Type IIA orientifolds,
the conditions for the tadpole cancellations and conditions on D6-brane configurations for
N = 1 supersymmetry in four-dimension. We specifically focus on Type IIA theory on
T 6 /(Z2 Z2 ) orientifold.
In Section 3, we discuss in detail the T-duality symmetry and its variations within the
supersymmetric intersecting D6-brane model building. The first set of symmetries are general and can be applied to any concrete particle physics model building (type I), and the
second set is special and only valid for the specific PatiSalam model building (type II).
We also find that any two models T-dual to each other have the same gauge couplings at
string scale.
In Section 4, we study the phenomenological constraints in the construction of the
supersymmetric PatiSalam models. In particular, we highlight why the models where
the PatiSalam gauge symmetry comes from U (n) branes are phenomenologically interesting. We show that the PatiSalam gauge symmetry can be broken down to the
SU(3)C SU(2)L U (1)BL U (1)I3R via D6-brane splittings, and further down to
the Standard Model gauge symmetry via Higgs mechanism, i.e., brane recombination in
geometric interpretation [22], where the Higgs fields come from the massless open string
166
states in a N = 2 subsector. In order to stabilize the moduli and provide a way to break
supersymmetry, we require that at least two hidden sector gauge factors are confining, thus
allowing for the race-track gaugino condensation mechanism.
In Section 5, by employing the T-duality and its variations, we systematic search for
the inequivalent models, and discuss the possible classes of solutions in detail. As a result,
we obtain total of 11 inequivalent models, all of them arising from the orbifold with only
one of the three two-tori tilted. Compared to the previous SM constructions [2628], eight
of our models have fewer pairs of the SM Higgs doublets ( 8). Interestingly, the gauge
coupling unification for SU(2)L and SU(2)R can be achieved at string scale in two models.
As explicit examples, we present the chiral spectra in the open string sector for the models
I-NZ-1a, I-Z-6 and I-Z-10.
In Section 6, we briefly comment on the other potentially interesting setups. And the discussions and conclusions are given in Section 7. In Appendix A, we present the D6-brane
configurations and intersection numbers for supersymmetric PatiSalam models.
2. Conditions for supersymmetric models from T 6 /(Z2 Z2 ) orientifolds with
intersecting D6-branes
The rules to construct supersymmetric models from Type IIA orientifolds on T 6 /(Z2
Z2 ) with D6-branes intersecting at generic angles, and to obtain the spectrum of massless
open string states have been discussed in [27]. Following the convention in Ref. [29], we
briefly review the essential points in the construction of such models.
The starting point is Type IIA string theory compactified on a T 6 /(Z2 Z2 ) orientifold.
We consider T 6 to be a six-torus factorized as T 6 = T 2 T 2 T 2 whose complex coordinates are zi , i = 1, 2, 3, for the ith two-torus, respectively. The and generators for the
orbifold group Z2 Z2 , which are associated with their twist vectors (1/2, 1/2, 0) and
(0, 1/2, 1/2) respectively, act on the complex coordinates of T 6 as
: (z1 , z2 , z3 ) (z1 , z2 , z3 ),
: (z1 , z2 , z3 ) (z1 , z2 , z3 ).
(1)
The orientifold projection is implemented by gauging the symmetry R, where is

world-sheet parity, and R acts as
R : (z1 , z2 , z3 ) (z1 , z2 , z3 ).
(2)
So, there are four kinds of orientifold 6-planes (O6-planes) for the actions of R, R ,
R, and R , respectively. To cancel the RR charges of O6-planes, we introduce
stacks of Na D6-branes, which wrap on the factorized three-cycles. Meanwhile, we have
two kinds of complex structures consistent with orientifold projection for a two-torus
rectangular and tilted [13,27,29]. If we denote the homology classes of the three cycles
wrapped by the D6-brane stacks as nia [ai ] + mia [bi ] and nia [ai ] + mia [bi ] with [ai ] = [ai ] +
1
2 [bi ] for the rectangular and tilted tori respectively, we can label a generic one cycle by
(nia , lai ) in either case, where in terms of the wrapping numbers lai mia for a rectangular
ia = 2mia + nia for a tilted two-torus. Note that for a tilted two-torus,
two-torus and lai 2m
167
lai nia must be even. For a stack of Na D6-branes along the cycle (nia , lai ), we also need
to include their R images Na with wrapping numbers (nia , lai ). For D6-branes on top
of O6-planes, we count the D6-branes and their images independently. So, the homology
three-cycles for stack a of Na D6-branes and its orientifold image a take the form
[a ] =
3

i
na [ai ] + 2i lai [bi ] ,
[a ] =
i=1
3

i
na [ai ] 2i lai [bi ] ,
(3)
i=1
where i = 0 if the ith two-torus is rectangular and i = 1 if it is tilted. And the homology
three-cycles wrapped by the four O6-planes are
R:
R:
[R ] = 23 [a1 ] [a2 ] [a3 ],

[R ] = 2
32 3
R : [R ] = 2
R :
[R ] = 2
[a1 ] [b2 ] [b3 ],
31 3
31 2
[b1 ] [a2 ] [b3 ],
[b1] [b2 ] [a3 ].
(4)
(5)
(6)
(7)
Therefore, the intersection numbers are

Iab = [a ][b ] = 2
3

nia lbi nib lai ,
(8)
i=1
Iab = [a ][b ] = 2k
3

nia lbi + nib lai ,
(9)
i=1
Iaa = [a ][a ] = 23k
3

nia lai ,
(10)
i=1

IaO6 = [a ][O6 ] = 23k la1 la2 la3 + la1 n2a n3a + n1a la2 n3a + n1a n2a la3 ,
(11)
where [O6 ] = [R ] + [R ] + [R ] + [R ] is the sum of O6-plane homology

three-cycles wrapped by the four O6-planes, and k = 1 + 2 + 3 is the total number of
tilted two-tori. For future convenience, we shall define nia lbi nib lai as the intersection factor
from the ith two-torus.
The general spectrum of D6-branes intersecting at generic angles, which is valid for
both rectangular and tilted two-tori, is given in Table 1. And the 4-dimensional N = 1 supersymmetric models from Type IIA orientifolds with intersecting D6-branes are mainly
constrained in two aspects: RR tadpole cancellation (Section 2.1) and N = 1 supersymmetry in four dimensions (Section 2.2).
2.1. Tadpole cancellation conditions
As sources of RR fields, D6-branes and orientifold O6-planes are required to satisfy the
Gauss law in a compact space, i.e., the total RR charges of D6-branes and O6-planes must
vanish since the RR field flux lines are conserved. The RR tadpole cancellation conditions
168
Table 1
General spectrum on intersecting D6-branes at generic angles which is valid for both rectangular and tilted twotori. The representations in the table refer to U (Na /2), the resulting gauge symmetry [27] due to Z2 Z2 orbifold
projection. For supersymmetric constructions, scalars combine with fermions to form chiral supermultiplets. In
our convention, positive intersection numbers imply left-hand chiral supermultiplets
Sector
Representation
aa
U (Na /2) vector multiplet

3 adjoint chiral multiplets
ab + ba
Iab (a , b ) fermions
ab + b a
Iab (a , b ) fermions

1 I
1
2 aa 2 Ia,O6 fermions

1 I
1
2 aa + 2 Ia,O6 fermions
aa + a a
are
Na [a ] +
Na [a ] 4[O6 ] = 0,
(12)
where the last contribution comes from the O6-planes which have 4 RR charges in the
D6-brane charge unit.
To simplify the notation, let us define the products of wrapping numbers
Aa n1a n2a n3a ,
A a la1 la2 la3 ,
Ba n1a la2 la3 ,
Ca la1 n2a la3 ,

C a n1a la2 n3a ,
B a la1 n2a n3a ,
Da la1 la2 n3a ,

D a n1a n2a la3 .
(13)
To cancel the RR tadpoles, we can also introduce an arbitrary number of D6-branes wrapping cycles along the orientifold planes, the so-called filler branes, which contribute to
the tadpole conditions but trivially satisfy the supersymmetry conditions. Thus, the tadpole
conditions are

Na Aa = 2k N (2) +
Na Ba = 2k N (3) +
Na Ca
2k N (1) +
a
= 2 N
k
(4)
Na Da = 16,
(14)
where 2N (i) are the number of filler branes wrapping along the ith O6-plane which is
defined in Table 2.
The tadpole cancellation conditions directly lead to the SU(Na )3 cubic non-Abelian
anomaly cancellation [15,16,27]. And the cancellation of U (1) mixed gauge and gravitational anomaly or [SU (Na )]2 U (1) gauge anomaly can be achieved by GreenSchwarz
mechanism mediated by untwisted RR fields [15,16,27].
2.2. Conditions for 4-dimensional N = 1 supersymmetric D6-brane
The four-dimensional N = 1 supersymmetric models require that 1/4 supercharges
from ten-dimensional Type I T-dual be preserved, i.e., these 1/4 supercharges should survive two projections: the orientation projection of the intersecting D6-branes, and the
169
Table 2
Wrapping numbers of the four O6-planes
Orientifold action
O6-plane
(n1 , l 1 ) (n2 , l 2 ) (n3 , l 3 )
R
R
R
R
1
2
3
4
(21 , 0) (22 , 0) (23 , 0)

(21 , 0) (0, 22 ) (0, 23 )
(0, 21 ) (22 , 0) (0, 23 )
(0, 21 ) (0, 22 ) (23 , 0)
Z2 Z2 orbifold projection on the background manifold. Analysis shows that, the 4dimensional N = 1 supersymmetry (SUSY) can be preserved by the orientation projection
if and only if the rotation angle of any D6-brane with respect to the orientifold-plane is
an element of SU(3) [9], or in other words, 1 + 2 + 3 = 0 mod 2 , where i is the
angle between the D6-brane and the orientifold-plane in the ith two-torus. Meanwhile,
this 4-dimensional N = 1 supersymmetry will automatically survive the Z2 Z2 orbifold
projection. The SUSY conditions can therefore be written as [29]
xA A a + xB B a + xC C a + xD D a = 0,
Aa Ba Ca Da
+
+
+
< 0,
xA
xB
xC
xD
(15)
where xA = , xB = 22 +3 /2 3 , xC = 21 +3 /1 3 , xD = 21 +2 /1 2 , and i =
Ri2 /Ri1 are the complex structure moduli. The positive parameter has been introduced to
put all the variables A, B, C, D on an equal footing. Based on these SUSY conditions, all
possible D6-brane configurations preserving 4-dimensional N = 1 supersymmetry can be
classified into three types:
(1) Filler brane with the same wrapping numbers as one of the O6-planes in Table 2. It
corresponds to the USp group. And among coefficients A, B, C and D, one and only
one of them is non-zero and negative. If the filler brane has non-zero A, B, C or D,
we refer to the USp group as the A-, B-, C- or D-type USp group, respectively.
(2) Z-type D6-brane which contains one zero wrapping number. Among A, B, C and D,
two are negative and two are zero.
(3) NZ-type D6-brane which contains no zero wrapping number. Among A, B, C and D,
three are negative and the other one is positive. Based on which one is positive, these
NZ-type branes are defined as the A-, B-, C- and D-type NZ brane. Each type can
have two forms of the wrapping numbers which are defined as
A1: (, ) (+, +) (+, +),
A2: (, +) (, +) (, +),
(16)
B1: (+, ) (+, +) (+, +),
B2: (+, +) (, +) (, +),
(17)
C1: (+, +) (+, ) (+, +),
C2: (, +) (+, +) (, +),
(18)
D1: (+, +) (+, +) (+, ),
D2: (, +) (, +) (+, +).
(19)
In the following, we will call the Z-type and NZ-type D6-branes as U -branes since they
carry U (n) gauge symmetry.
170
3. T-duality symmetry and its variations

T-duality relates equivalent models, and thus employing this symmetry can simplify
significantly the search for the inequivalent models. In this section, we shall study the
action of the T-dualities (and its variants). These symmetries correspond to two types: one
is general and can be applied to any D6-brane model building (type I) and the other one is
special and only effective in the PatiSalam model building (type II). Our philosophy is:
consider only one model for each equivalent class characterized by these T-dualities.
Before discussing the T-duality, we point out that: (1) Two models are equivalent if their
three two-tori and the corresponding wrapping numbers for all the D6-branes are related
by an element of permutation group S3 which acts on three two-tori. This is a trivial fact.
(2) Two D6-brane configurations are equivalent if their wrapping numbers on two arbitrary
two-tori have the same magnitude but opposite sign, and their wrapping numbers on the
third two-torus are the same. We call it as the D6-brane Sign Equivalent Principle (DSEP).
In the following, we discuss type I and type II T-dualities separately.
3.1. Type I T-duality
T-duality transformation happens on two two-tori simultaneously, for example, the j th
and kth two-tori
j j
j j
k k

nx , lx lx , nx ,
(20)
nx , lx lxk , nkx ,
where x runs over all stacks of D6-branes in the model. It does not change anything about
the D-brane model except that it makes an interchange among the four pairs of the products
(B, B),
(C, C)
and (D, D),
which indicates that the particle
of wrapping numbers (A, A),
spectra are invariant under this transformation while the complex structure moduli may not
be.
Without loss of generality, we assume that j = 2 and k = 3. In this case, the interchange
will take place between A and B pairs, and C and D pairs
(B, B),
(A, A)
(D, D).
(C, C)
(21)
And the corresponding transformations of moduli parameters are

xA = xB ,
xB = xA ,
xC = xD ,

xD
= xC .
(22)
After these transformations, we obtain the new complex structure moduli and radii of T 6

1 xA xB
1 = 2
= 1 ,
(23)
xC xD

xD xB
2 = 22
(24)
= 222 (2 )1 ,
xA xC

xC xB
3 = 23
(25)
= 223 (3 )1 ,
xA xD
2
1
R1 = R12 ,
R1 = R11 ,
(26)
171
22 Ms2
R21 =
,
R21
2 22 Ms2
R2 =
,
R22
(27)
23 Ms2
R31 =
,
R31
2 23 Ms2
R3 =
,
R32
(28)
where Ms is the string scale.

Sometimes, it will be more convenient if we combine this T-duality with the trivial two
two-tori interchange Tj Tk where the transformations of the wrapping numbers and j,k
are
j
nx nkx ,
lx lxk ,
(29)
j k .
(30)
Thus, under this extended T-duality, the wrapping numbers and j,k transform as
j j

k k
j j
nx , lx lxk , nkx ,
nx , lx lx , nx ,
j k .
(31)
(32)
Still for j = 2 and k = 3, only A and B pairs are interchanged under this T-duality
(B, B),
(A, A)
(33)
and the corresponding transformations of moduli parameters are

xA = xB ,
xB = xA ,
xC = xC ,

xD
= xD .
In this case, the new complex structure moduli and radii are

1 xA xB
1 = 2
= 1 ,
xC xD

xB xC
2 = 23
= 223 31 ,
xA xD

2 xB xD
3 = 2
= 222 21 ,
xA xC
1
2
R1 = R11 ,
R1 = R12 ,

R21 =

R31 =
23 Ms2
,
R31
22 Ms2
,
R21
2
R2 =
2
R3 =
23 Ms2
,
R32
22 Ms2
.
R22
(34)
(35)
(36)
(37)
(38)
(39)
(40)
One pair of the models related by this extended type I T-duality have been shown in Table 7
and Table 8 in Appendix A up to DSEP on the first two two-tori of b stack of D6-branes.
Since this extended T-duality only interchanges two pairs of the products of wrapping numbers, if all two-tori are rectangular or tilted, all models characterized by the permutations
of these four parameter pairs will be T-dual to each other. By the way, for the case where
172
two two-tori are rectangular or tilted, this conclusion is not valid if the rectangular and
tilted two-tori have been fixed.
x for
As a remark, in this kind of D6-brane models, the YangMills gauge coupling gYM
x-stack of D6-branes at string scale is [33,39]
x 2
8 Ms
1

,
gYM =
(41)

MPl
3
(ni )2 1 + (2i l i )2
x
i=1
where MPl is the 4-dimensional Planck scale. So the gauge coupling is invariant under the
T-duality and its variation. This is a typical property of T-duality (e.g., see [40]).
3.2. Type II T-duality
Under it, the transformations of the wrapping numbers for any stacks of D6-branes in
the model are
nix nix ,
lxi lxi ,
nx lx ,
nkx lxk ,
(42)
where i = j = k, and x runs over all D6-branes in the model. Comparing with the general
(B, B),
(C, C)
and (D, D),
the
type I T-duality, besides the interchanges among (A, A),
signs of A, B, C and D are also changed, which lead to the sign changes of all intersection
numbers. On the other hand, for the models in our model construction, we require that
Iab + Iab = 3,
Iac = 3,
Iac = 0,
(43)
which will be discussed in detail in the next section. Obviously, there is only one sign
difference between the intersection numbers of Iab and Iac if Iab = 3. By combining with
b c,
(44)
therefore, we may get one equivalent model satisfying our requirements. For this type of
T-duality, the moduli and radii will obey the same transformation rules as those in the
type I T-duality. But, unlike the type I, the quantum numbers for SU(2)L and SU(2)R in
the particle spectrum and two gauge couplings at string scale will be interchanged due
to b c. Models in Table 7 and Table 9 are such a pair of examples related by type II
T-duality up to DSEP on b and c stacks of D6-branes.
If Iab = 1 or 2, which can be achieved only when nia lbi = 0 or nib lai = 0 are satisfied on
two two-tori, these intersection numbers become
Iab = 3,
Iac = 1 or 2,
(45)
under the type II T-duality transformation. Therefore, if we relax the intersection number
requirement to
Iab + Iab = 3,
Iac = 3,
Iac = 0,
(46)
Iac + Iac = 3,
(47)
and
Iab = 3,
Iab = 0,
173
we only have to consider one case if the D6-brane wrapping numbers in two setups can be
related by Eq. (42): the derivations and conclusions in the first setup can be applied to the
second one as well.
By combining with type I T-duality and DSEP, we obtain an variation of type II
T-duality. Under it, the transformations of the wrapping numbers for any stacks of
D6-branes in the model are
lx1 lx1 ,
b c,
lx2 lx2 ,
lx3 lx3 ,
(48)
where x runs over all D6-branes in the model. Since the transformations in the first line
B,
C and D,
the moduli and radii are
of the above equations only change the signs of A,
invariant.
4. Supersymmetric PatiSalam models and gauge symmetry breaking via D6-brane

splittings
To build the SM or SM-like models in the intersecting D6-brane scenarios, besides
U (3)C and U (2)L stacks of branes, we need at least two extra U (1) gauge symmetries
for both SUSY and non-SUSY versions due to the quantum number of the right-handed
electron [16,2729]. One (U (1)L ) is lepton number symmetry, and the other one (U (1)I3R )
is like the third component of right-handed weak isospin. Then, the hypercharge is obtained
via
QB QL
,
QY = QI3R +
(49)
2
where U (1)B arises from the overall U (1) in U (3)C . Meanwhile, to forbid the gauge field
of U (1)I3R to obtain a mass via B F couplings, the U (1)I3R can only come from the nonAbelian part of U (2)R or USp gauge symmetry. In this case, the U (1) gauge symmetry,
which comes from a non-Abelian symmetry, is generically anomaly free and its gauge field
is massless. Similarly, to generate the anomaly-free U (1)BL symmetry, the U (1)L should
come from non-Abelian group. Considering that the U (1)L stack should be parallel to the
U (3)C stack on at least one two-tori, we generate it by splitting branes from one U (4)
stack, resulting in the U (3)C stack at the same time.
In the previous supersymmetric SM constructions [27,28], U (1)I3R arises from the stack
of D6-branes on the top of orientifold, i.e., from the USp group. These models have at least
8 pairs of the SM Higgs doublets, and generically there exist two additional anomaly free
U (1) gauge symmetries. They could be in principle spontaneously broken via Higgs mechanism by the scalar components of the chiral superfields with the quantum numbers of the
right-handed neutrinos, however, they break D-flatness conditions, and thus supersymmetry, so the scale of symmetry breaking should be near the electroweak scale. On the other
hand, there were typically no candidates, preserving the D-flatness and F-flatness conditions, and that could in turn break these gauge symmetries at an intermediate scale.
Therefore, we focus on PatiSalam model where U (1)I3R arises from the U (2)R symmetry. Failing to find interesting models with SU(2)L from the D6-branes on the top of
174
O6-plane, we would like to construct the supersymmetric SU(4)C SU(2)L SU(2)R

models from three stacks of D6-branes that are not on the top of orientifold planes. We
will show that the PatiSalam gauge symmetry can be broken down to SU(3)C SU(2)L
U (1)BL U (1)I3R via D6-brane splittings, and then down to the SM gauge symmetry via
Higgs mechanism where the Higgs particles come from a N = 2 subsector. In particular, in
our models, we do not have any extra anomaly free U (1) gauge symmetry at electroweak
scale which was a generic problem in previous constructions [27,28].
Suppose we have three stacks of D6-branes, a, b, and c with number of D6-branes 8,
4, and 4. So, a, b, and c stacks give us the gauge symmetry U (4)C , U (2)L and U (2)R ,
respectively. The anomalies from three U (1)s are cancelled by the GreenSchwarz mechanism, and the gauge fields of these U (1)s obtain masses via the linear B F couplings.
So, the effective gauge symmetry is SU(4)C SU(2)L SU(2)R . In addition, we require
that the intersection numbers satisfy
Iab + Iab = 3,
Iac = 3,
(50)
Iac = 0.
(51)
The conditions Iab + Iab = 3 and Iac = 3 give us three families of the SM fermions
1, 2) under SU(4)C SU(2)L SU(2)R gauge
with quantum numbers (4, 2, 1) and (4,
symmetry. Iac = 0 implies that a stack of D6-branes is parallel to the orientifold (R)
image c of the c stack of D6-branes along at least one tow-torus, for example, the third
two-torus. Then, there are open strings which stretch between the a and c stacks of D62
branes. If the minimal distance squared Z(ac
) (in 1/Ms units) between these two stacks of
D6-branes on the third two-torus is small, i.e., the minimal length squared of the stretched
2
2
string is small, we have the light scalars with squared-masses Z(ab
) /(4 ) from the
NS sector, and the light fermions with the same masses from the R sector [15,16,39].
These scalars and fermions form the 4-dimensional N = 2 hypermultiplets, so, we obtain
(2)

the Iac
(the intersection numbers for a and c stacks on the first two two-tori) vector 1, 2) and (4, 1, 2). These particles
pairs of the chiral multiplets with quantum numbers (4,
are the Higgs fields needed to break the PatiSalam gauge symmetry down to the SM
2
gauge symmetry. In particular, these particles are massless if Z(ac
) = 0. By the way, the

intersection numbers Iac = 0 and Iac = 3 are equivalent to Iac = 3 and Iac = 0 due to
the symmetry transformation c c .
In order to break the gauge symmetry, we split the a stack of D6-branes into a1 and a2
stacks with 6 and 2 D6-branes, respectively. The U (4)C gauge symmetry is broken down
to the U (3) U (1). Let us assume that the numbers of symmetric and anti-symmetric
representations for SU(4)C are, respectively, na and na , similar convention for SU(2)L
and SU(2)R . After splitting, the gauge fields and three multiplets in adjoint representation
for SU(4)C are broken down to the gauge fields and three multiplets in adjoint representations for SU(3)C U (1)BL , respectively. The na and na multiplets in symmetric
and anti-symmetric representations for SU(4)C are broken down to the na and na mul
tiplets in symmetric and anti-symmetric representations for SU(3)C , and na multiplets in
symmetric representation for U (1)BL . However, there are Ia1 a new fields with quantum
2
175
number (3, 1) under SU(3)C U (1)BL from the open strings at the intersections of
D6-brane stacks a1 and a2 . The rest of the particle spectrum is the same. Moreover, we
can show that the anomaly free gauge symmetry from a1 and a2 stacks of D6-branes is
SU(3)C U (1)BL , which is the subgroup of SU(4)C .
Furthermore, we split the c stack of D6-branes into c1 and c2 stacks with 2 D6-branes
for each one. Similarly, the gauge fields and three multiplets in adjoint representation for
SU(2)R are broken down to, respectively, the gauge fields and three multiplets in adjoint
representation for U (1)I3R , The nc multiplets in symmetric representation for SU(2)R are
broken down to the nc multiplets in symmetric representation for U (1)I3R , while the nc
multiplets in anti-symmetric representation for SU(2)R are gone. In addition, there are Ic1 c
2
new fields which are neutral under U (1)I3R from the open strings at the intersections of
D6-brane stacks c1 and c2 . And the rest of the particle spectrum is the same. Moreover, the
anomaly free gauge symmetry from c1 and c2 stacks of D6-branes is U (1)I3R , which is the
subgroup of SU(2)R .
After D6-brane splittings, we obtain that the gauge symmetry is SU(3)C SU(2)L
U (1)BL U (1)I3R . To break this gauge symmetry down to the SM gauge symmetry,
2
between the a2 and c1 stacks of
we assume that the minimal distance squared Z(a
c )
2 1
(2)

2 c1
D6-branes on the third two-torus is very small, then, we obtain Ia

c1
(the intersection
numbers for a2 and stacks on the first two two-tori) pairs of chiral multiplets with quantum numbers (1, 1, 1, 1/2) and (1, 1, 1, 1/2) under SU(3)C SU(2)L U (1)BL
U (1)I3R from the light open string states which stretch between the a2 and c1 stacks of
D6-branes. These particles can break the SU(3)C SU(2)L U (1)BL U (1)I3R down
to the SM gauge symmetry and keep the D- and F-flatness because their quantum numbers
are the same as those for the right-handed neutrino and its complex conjugate. Especially,
2
these particles are massless if Z(a
= 0. In summary, the complete symmetry breaking
2 c1 )
chains are
aa1 +a2
SU(4) SU(2)L SU(2)R SU(3)C SU(2)L SU(2)R U (1)BL

cc1 +c2
SU(3)C SU(2)L U (1)I3R U (1)BL

Higgs mechanism
SU(3)C SU(2)L U (1)Y .
(52)
The dynamical supersymmetry breaking in D6-brane models from Type IIA orientifolds
has been addressed in [33]. In the D6-brane models, there are some filler branes carrying
USp groups which are confining, and thus could allow for gaugino condensation, supersymmetry breaking and moduli stabilization.
The gauge kinetic function for a generic stack x of D6-branes is of the form (see, e.g.,
[33]):

3

1 1 2 3
j
fx =
n n n S
2j k nix lx lxk U i ,
4 x x x
i=1
(53)
176
where the real parts of dilaton S and moduli U i are

Ms3 R11 R12 R13
,
2gs
(54)
= Re(S)j k ,
(55)
Re(S) =
Re U

i
where i = j = k, and gs is the string coupling. Also, the Khler potential is

K = ln(S + S)
3

ln U I + U I .
(56)
I =1
In our models, three stacks of D6-branes with U (4)C U (2)L U (2)R gauge symmetry
generically fix the complex structure moduli 1 , 2 and 3 due to supersymmetry conditions. So, there is only one independent modulus field. To stabilize the modulus, we need
at least two USp groups which are confining, i.e., their functions are negative, and thus
allow for gaugino condensations [4143]. Suppose there are 2N (i) filler branes which are
on top of ith O6-plane and carry USp(N (i) ) group. Its beta function is

(i)
(i)
N
N
g
1k
k
k
+ 1 + 2 |Iai | + 2 |Ibi | + 2 |Ici | + 3
1
i = 3
2
2
= 6 + 21k |Iai | + 2k |Ibi | + 2k |Ici |,
(57)
where 3(N (i) /2 1) is a contribution from three multiplets in the anti-symmetric representation of the USp branes. The negative functions of USp groups give strong constraints
on the intersection numbers of the associated filler branes and the observable branes, and
thus constrain the number of allowed models.
If supersymmetry turns out to be broken due to the gaugino condensations, the supersymmetry breaking will be mediated via gauge interactions because the gravity mediated
supersymmetry breaking is much smaller. So, the supersymmetry CP problem can be
solved in our models. In this paper, we will neither study the stabilization of this complex structure modulus (or for that reason also Khler moduli, which could enter the gauge
coupling corrections due to the one-loop threshold corrections) due to gaugino condensation, nor the issue of supersymmetry breaking and postpone this for further study. However,
since we do eventually want to address these issues, we confine our search only to models
with at least two USp gauge group factors with negative functions.
5. Systematic search for supersymmetric PatiSalam models

The basic properties for the models that we want to construct are given in Section 3.
Let us summarize them here. There are three stacks of D6-branes, a, b, and c with number
of D6-branes 8, 4, and 4, which give us the gauge symmetry U (4)C , U (2)L and U (2)R ,
respectively. We require that their intersection numbers satisfy Eqs. (50), (51). In addition,
to stabilize the modulus and possibly break the supersymmetry, we require that at least two
USp groups in the hidden sector have negative functions.
177
Our searching strategy is the following: first, analytically exclude most of the parameter space for the D6-brane wrapping numbers which cannot give the models with above
properties, and then scan the rest parameter space by employing a computer program. If no
two-torus is tilted, we cannot have the particle spectra with odd families of the SM fermions. So, there are three possibilities: one tilted two-torus, two tilted two-tori, and three
tilted two-tori. The complete searching shows no-go for the last two possibilities. As for
the first one, all solutions are tabulated in Appendix A. The detailed discussions are given
in the following three subsections. People only interested in phenomenology may safely
skip these parts.
5.1. One tilted two-torus
Without loss of generality, let us suppose that the third two-torus is tilted. Then, we
may consider the two cases where the a stack of D6-branes is of NZ- and Z-type, which
are characterized by no and one zero wrapping number, respectively.
5.1.1. Case I: NZ-type a stack of D6-branes with one negative wrapping number
For NZ-type a stack of D6-branes, among Aa , Ba , Ca and Da , there is at least one
equal to 1 in order to avoid the tadpole cancellation condition (TCC) violations. Due
to T-duality, we may assume that Da = 1. Obviously, the setup Iab = 1 or 2 cannot be
realized in this case. As for |la3 | there are only two possible absolute values (ABS): 1 and 3,
because the third wrapping number pairs (WNP) should be responsible for the even factors
of both 2k Iac = 6 and 2k Iac = 0 (k = 1). We shall discuss this case according to the
number of negative wrapping numbers for a-brane because this number cannot be larger
than 3 due to the D6-brane Sign Equivalent Principle (DSEP). Next, let us take a look at
the case with one negative wrapping number first.
Since Da = 1, the minus sign can only come from la1 , la2 and n3a . Noticing that SUSY
conditions cannot be satisfied if n3a < 0, we set la1 < 0 without loss of generality. If la3 = 3,
the a-brane will be
(+, 1) (+, 1) (1, 3).
(58)
= 0, in order to generate the

For the general intersection numbers 2k Iax = 3 2k and Iax
associated even factor requires at least one set of WNPs from the tilted two-torus which satisfy |nia | = |nix |, |lai | = |lxi |, and nia lai = nix lxi = 3 or 1 while the co-prime conditions
are also implemented. As a result, the 3rd WNP of the C-type brane should be (1, 3),
and thus one C2-brane is needed for CTCC (Tadpole Cancellation Condition related to C)
requirement
(, +) (+, +) (1, 3).
(59)
As for the third brane, if it is of A2-, B2- or D1-type,

(, +) (, +) (1, 3),
(60)
(+, +) (, +) (1, 3),
(61)
(+, +) (+, +) (1, 3),
(62)
178
there will be a problematic intersection factor generated from the second or the first WNP.
This factor has a ABS larger than 1, which will lead to 2k Iax > 6. If the third brane is of
C2-type also ATCC and DTCC require all wrapping numbers of a-, b- and c-branes have
unit ABS, which obviously is forbidden in our model building. Therefore, the third brane is
of Z-type and cannot provide extra positive Tadpole Charges. A direct result for this is that
|n1a n2a | = 1 and |n1C n2C | 2, |lC1 lC2 | 2 due to ATCC and DTCC. But, any one of |n1C n2C |
and |lC1 lC2 | cannot be equal to 1 to avoid vanishing IaC , which may lead to n33 = 0 due to
ATCC and DTCC again. This is impossible. Therefore, there is no solution while la3 = 3.
If la3 = 1, the most general form for the a-brane is
(L1 , 1) (L2 , 1) (1, 1),
(63)
where L1 and L2 are positive. If L2 > 2 for a-brane, no matter what value L1 has, both
A2- and C2-type branes are needed
(, +1) (, +1) (1, +1),
(64)
(, +1) (+, +1) (1, +1).
(65)
It is easy to see that the 2nd WNP of the A2-brane will contribute an intersection factor
larger than L2 or 3. Since we have another even factor from the third WNP, this may lead
to 2k IaA > 6. Therefore, there is no solution in this case.
For L2 = 2, both A2- and C2-branes are still required. If both of the 2nd and 3rd branes
are of Z-type, to avoid ATCC and CTCC violation, we must have n22 = n23 = 0. Then the
second WNPs of these branes will yield an even intersection factor because L2 is even, thus
yielding even SM families. So, the 2nd and 3rd branes cannot be Z-type at the same time.
If one of them is of NZ-type, the other one must be NZ-type also because the combination
of a-brane and the second one will necessarily violate ATCC and/or CTCC. As a result, band c-branes will be

(L1 1), +1 (1, +1) (1, +1),
(66)
(, +1) (+, +1) (1, +1).
(67)
It is obvious that the factor 3 for 2k IaA should be generated from the second WNP, thus
2 | = 1 here. ATCC cannot be satisfied while L 2; as for L = 1, ATCC
we have |n2A lA
1
1
1
requires |nC n2C | 2, which will forbid that IaC be a multiple of 3. Therefore, the only
possible solution for a-brane in this case is
(L, 1) (1, 1) (1, 1).
(68)
Let us consider the case where L > 2 first. In this case, one A2-brane is required, and
C- and D-type USp groups cannot appear in the hidden sector since they are not asymptotically free according to Eq. (57). Noticing the intersection factor 3 of 2k IaA should be
generated by the second WNP, we have two kinds of possible solutions
(L, 1) (1, 1) (1, 1),

(L 1), 1 (2, 1) (1, 1),
(69)
(70)
179
and
(L, 1) (1, 1) (1, 1),

(L 1), 1 (1, 2) (1, 1),
(71)
(+, +) (1, 0) (1, 1).
(73)
(72)
For the former, no matter what the third brane is, the combination of a- and A-branes
yield no solution for moduli consistent with supersymmetry; for the latter, obviously, a
problematic intersection factor larger than 3 comes out from the first WNPs of a- and the
third branes.
Finally, let us prove that the ABSs of b- and c-branes wrapping number cannot be larger
than 8 for L 2. If they are NZ-type branes, this conclusion is obvious since for each WNP
only one components ABS can be larger than 2 to avoid TCC violation. Thus the possible
largest ABS of wrapping numbers should be less than 2L + 3 = 7. For Z-type brane, the
possible ABS larger than 8 should come from the WNP without zero wrapping number.
Meanwhile, the corresponding two-torus is untilted. Let us focus on such a WNP. Suppose
this brane contributes the non-vanishing X- and Y-type tadpole charges. If X and Y do not
match with the types of a- and the 3rd brane, due to the same reason applied to NZ-type
brane, the wrapping numbers ABS of this brane still cannot be larger than 7. If X = B
and the 3rd brane is Y-type brane, we shall have lZ1 n2Z = 0. For lZ1 = 0, in order to avoid
the problematic factor generated from the 2nd WNPs, we must have |n2Z lZ2 | = 2 since n2Z
and lZ2 have different signs due to SUSY conditions. For n2Z = 0, given |BZ | 2L + 4 8
or |n1Z | 2L + 4 8, lZ1 cannot be larger than 7 due to the co-prime conditions for both
L = 1 and 2.
5.1.2. Case II: NZ-type a stack of D6-branes with two and three negative wrapping
numbers
For the case with two minus signs, if they are from different two-tori, the only possible
setup for a-brane due to SUSY conditions is
(L1 , 1) (L2 , 1) (1, ),
(74)
which corresponds to case I by the type II T-duality inference. Therefore, we need not
consider this case any more.
If the two minus signs come from the same two-torus, based on DSEP, let us suppose
that they are from the first two-torus. While la3 = 3, the a-brane will be of the form
(L1 , 1) (L2 , 1) (1, 3),
(75)
and both B2- and C2-branes

(+, +) (, +) (1, 3),
(76)
(, +) (+, +) (1, 3),
(77)
are required. Obviously, a problematic intersection factor larger than 1 will be generated
for 2k IaC . Therefore, there is no solution at this time.
180
If la3 = 1, the a-brane will be of the form

(L1 , 1) (L2 , 1) (1, 1).
(78)
For L1 2 and L2 > 2 or L1 > 2 and L2 2, B2- and C2-branes are needed
(+, +1) (, +1) (1, +1),
(79)
(, +1) (+, +1) (1, +1),
(80)
where at least one problematic intersection factor will be contributed by the 2nd WNP of
B2-brane or the 1st WNP of C2-brane. Therefore, the only possible solution for a-brane
will be
(2, 1) (2, 1) (1, 1),
(81)
(L, 1) (1, 1) (1, 1).
(82)
and
Noticing that the second setup can be transformed to the setup in Case I through type II
T-duality, we only need to consider the first one.
For the first setup, B2- and C2-branes are required and the only possible solutions are
(2, 1) (2, 1) (1, 1),
(83)
(1, 1) (2 1, 1) (1, 1),
(84)
(2 1, 1) (1, 1) (1, 1).
(85)
Regretfully, these solutions are excluded because there are no moduli solutions for supersymmetric D6-branes configuration. And thus, the case with NZ-type a stack of D6-branes
with two negative wrapping numbers should be ruled out. For the case with three negative
wrapping numbers, due to SUSY conditions the only possible setup for a-brane is
(L1 , 1) (L2 , 1) (1, 1),
(86)
which obviously corresponds to the case with two minus signs according to the type II
T-duality variation. Thus, this case can be excluded, too.
5.1.3. Case III: Z-type a stack of D6-branes
Now let us consider the Z-type a-brane. According to the type I T-duality and its extended version, there are two possible setups for a-brane: Aa = Ba = 0 and Aa = Da = 0.
The latter cannot give the required intersection numbers with b- and c-branes. If |n3a la3 | = 3,
according to SUSY conditions, the a-brane has four kinds of possible setups
(0, 1) (L1 , L2 ) (1, 3),
(87)
(0, 1) (L1 , L2 ) (3, 1),
(88)
(0, 1) (L1 , L2 ) (1, 3),
(89)
(0, 1) (L1 , L2 ) (3, 1).
(90)
181
The 1st and 2nd setups correspond to the 4th and 3rd ones due to type I T-duality, respectively, and the 1st one also corresponds to the 3rd one due to type II T-duality. Therefore,
we only need to consider the 1st one. For the 1st one, one C2-brane is required and it has
the form
(1, +) (+, 1) (1, 3).
(91)
In order to avoid BTCC and DTCC violation (since the third brane can provide at most
one kind of positive tadpole charge), we must have lC2 = 1. In addition, since n13 = 0 due
to Ia3 = 0, in order to avoid the BTCC violation we have to require that the third brane
satisfy l32 = 0
(, ) (+, 0) (1, +).
(92)
If l32 = 0, we must have l33 = 3 and thus the third brane is of B-type
(+, +) (, +) (1, 3).
(93)
A problematic intersection factor will arise from the second WNP. Eq. (92) implies that the
third brane is the b-brane and thus c-brane is of C-type. Note that here Iac + Iac = 3,
which is T-dual (type II) to the 3rd possible setup of a-brane with Iab + Iab = 3, cannot be
achieved here. Since DTCC requires la2 = 1 and |lc1 | 2, Iac = 3 will yield n2a = n2c + 1 >
0, and thus CTCC cannot be satisfied. So, there is no solution while la3 = 3.
For the case with |n3a la3 | = 1, based on the type I T-duality and SUSY conditions, we
can have only a-brane of the form:
(0, 1) (L1 > 0, L2 > 0) (1, 1).
(94)
Meanwhile, due to type II T-duality, we only need consider the case L1 > L2 . Let us prove
first that there is no solution while L1 > 2. If L1 > 2, one C2-brane is required, but the
third brane must be of Z-type. The reason is that, if the third brane is of A2- or B2-type
(, +) (, +) (1, 1),
(95)
(+, +) (, +) (1, 1),
(96)
there is a problematic intersection factor larger than L1 generated from the second WNPs.
If it is C2- or D1-brane,
(, +) (+, +) (1, 1),
(97)
(, ) (+, +) (1, 1),
(98)
n23 = l32 L1 + 1 or l32 L1 + 3 is forbidden in order to avoid the ATCC violation. Therefore, Iab
and Iac have the same signs if both of them are of C2- or D1-type, which is not allowed.
Thus L2 = 1 is required.
Let us return to the C2-brane. If the intersection factor 3 for IaC comes from the first
two-torus, one additional A2-brane is needed because n1C = 3 and n2C 2 has led to
ATCC violation. Therefore, there is no solution in this case. If the factor 3 comes from the
second two-torus, lC2 cannot be larger than 2, otherwise, DTCC cannot be satisfied. The
182
only two possible solution for a- and C-type branes will be

(0, 1) (3, 1) (1, 1),
(99)
(1, +) (3, 2) (1, 1),
(100)
(0, 1) (L, 1) (1, 1),
(101)
(1, +) (L 3, 1) (1, 1).
(102)
and
For the first possible solution, because DTC (ATC, BTC, DTC and DTC denote the A-, B-,
C- and D-type tadpole charges, respectively) is already filled, the 3rd brane should satisfy
l31 l32 = 0 and have the form

1, l31 1, l32 (1, 1),
(103)
where A3 = 1 is required by ATCC. Obviously, the intersection factor 3 cannot be generated for Ia3 . For the second possible solution, ATCC requires 4 L 7. To avoid the
CTCC violation, lC1 > 2 is necessary, which, however, will lead to the DTCC violation. As
a result, there is no solution if L1 > 2.
For L = 1, 2, a solution is possible. Here we shall prove that the wrapping numbers of
b- and c-branes cannot be larger than 8. For a NZ-type brane, the only possible wrapping
number with its ABS larger than 3 is l 1 . |n2 | and |l 2 | must be smaller than 2L + 3 since the
smaller one still cannot be larger than 2. |l 1 | can be larger than 3 only when the NZ-brane
is of C- or D-type. Correspondingly, the third brane should be of D- or C-type. Since the
components of the 3rd WNPs for both branes have different signs, the two branes should
be of C2- and D1-types
(0, 1) (L, 1) (1, 1),

+, lC1 < 0 (+, +) (1, 1),
(104)
(+, +) (+, +) (1, 1).
(106)
(105)
Due to ATCC and BTCC, one of IaC and IaD cannot be a multiple of 3. For a Z-type brane,
as we discussed before, the possible wrapping number with its ABS larger than 3 can only
come from an untilted two-torus and the corresponding WNP contains no zero wrapping
number. If this WNP is from the second torus, SUSY conditions require that its two components have different signs and thus their ABS cannot be larger than 2 to avoid a problematic
intersection factor. If this WNP is from the first two-torus, the only possible large wrapping
number is lZ1 . If lZ1 > 2, the third brane is of C-type or D-type. For the C-type case, given
that ATCC and DTCC require |A3 | 3 and |D3 | 2, respectively, CTCC will yield |CZ |
or |lZ1 | |A3 | |D3 | + 4 2 = 8. As a result, the largest ABS of these wrapping numbers
is less than 9. We shall have the same situation for D-type 3rd brane.
5.2. Two tilted two-tori
Similar to the previous subsection, we consider the NZ-type a stack of D6-branes first
and then turn to the Z-type one. Here we assume that the second and the third two-tori are
tilted.
183
5.2.1. Case I: NZ-type a stack of D6-branes with one negative wrapping number
For the a-brane of NZ type, let us still suppose D = 1 in accordance with T-duality.
Then this brane has two forms
(L1 , 1) (L2 , 1) (1, L3 ),
(107)
(L1 , 1) (L2 , 1) (1, L3 ),
(108)
and
where L1 , L2 , L3 are positive, and L2 , L3 are odd. According to the requirement of moduli
stabilization for gauge symmetries in the hidden sector, there is at most one with its ABS
larger than 2 among all wrapping numbers of a-brane, which implies three possibilities:
(1) L3 = 1, L1 2 and L2 3; (2) L2 = 1, L1 2 and L3 3; (3) L2 = L3 = 1 and
L1 2.
Now let us consider the first form of a-brane and set L2 3:
(L1 , 1) (L2 , 1) (1, 1),
(109)
then the other two branes are of A- and C-types, respectively, and have DA = DC = 1 to
avoid DTCC violation. Meanwhile, to avoid a problematic intersection factor, the C-brane
should be of C2-type:
(L1 1 < 0, +1) (L2 6 > 0, +1) (1, 1),
(110)
where the intersection factor 3 of IaC cannot come from the first WNP. If it does, we shall
have n1C = (L1 + 3) because of 0 < L1 2. The C-type USp group is no longer asymptotically free, and then we do not have enough USp groups in the hidden sector to stabilize
the modulus. Now let us consider the A-type brane. For A1-brane, the 0 intersection factor
for IaA cannot be generated. The A-brane, therefore, should be of A2-type. The factor 3
of IaA comes from the second WNP, and thus the brane will be
(L1 1 < 0, +1) (L2 6 < 0, +1) (1, 1).
(111)
Under this setup, we always have n1A = n1C no matter L1 is equal to 1 or 2. This is obvious
for L1 = 1. If L1 = 2, both n1A and n1C cannot be equal to 3. Otherwise, C-type USp group
is not asymptotically free again. So we shall have n1A = n1C = 1. On the other hand,
because of n2C > 0 and n2A < 0, we must have n2C = L2 + 6 and n2A = L2 6, and thus
|n2C | |n2A | = 2L2 > 0. Due to L2 3, the ATCC cannot be satisfied for any value of L2 .
As a result, we have L2 = 1 for the first setup of a-brane.
If L3 3, the a-brane has the form
(L1 , 1) (1, 1) (1, L3 ),
(112)
and then one C-type brane is required due to CTCC. At this time A- and D-type USp
groups are not asymptotically free due to the large value of L3 . If the C-brane is of C1-type,
the intersection factor 3 for IaC should come from the 1st WNP and the only possible
solutions for the a- and C1-branes are
(L1 , 1) (1, 1) (1, L3 ),
1

nC > 0, lC1 > 0 (1, 1) (1, L3 + 2),
(113)
(114)
184
where L1 n1C lC1 = 2 to provide the factor 3 for IaC , n3C cannot be equal to 2 to keep AT C =
DT C = 0, and lC3 cannot be equal to L3 2 to avoid CTCC violation. Now it is easy
to check that there is no moduli solution while L1 = 2 or lC1 = 2, and B-type USp group
cannot be generated in the hidden sector while n1C = 2. If this is a C2-brane, the intersection
factor 3 will come from the third WNP, which means L3 = 3, thus we shall have n2C lC2 3.
Noticing that the third brane is of NZ-type no matter which one is larger, it is easy to check
that there are no enough asymptotically free USp groups available in the hidden sector.
Finally, let us consider the last kind of setup of a-brane
(L1 2, 1) (1, 1) (1, 1).
(115)
For L1 3, one brane of A-type is required. This is also true for L1 = 2. Otherwise, to
avoid one A-type brane, the fact that ATC is full will lead to n1 n2 n3 = 0 for the other
two branes. n1 cannot be equal to zero to avoid an even problematic intersection factor. If
n2 = 0, the intersection factor 3 comes from the first WNP and this brane has the form

2l 1 3 < 0, l 1 (0, 2) (1, 1).
(116)
To avoid the DTCC violation, the third brane must have n33 = 0 and |l33 | = 2. This indicates
that no enough USp groups are available in the hidden sector now. As a result, one A-type
brane is definitely necessary while L1 2.
This A-type brane is a A2-brane to get IaA = 0. If the intersection factor 3 of IaA comes
from the first WNP, the possible solution for this brane is

1
1
(1, 1) (1, 1),
lA L1 3 < 0, lA
(117)
which gives out no moduli solutions by combining with a-brane setup. If the factor 3 is
given out by the second WNPs, we must have L1 = 2, in order to preserve at least two
asymptotically free USp groups in the hidden sector. Thus the possible solution for a- and
A-type brane will be
(2, 1) (1, 1) (1, 1),

2
2
< 0, lA
(1, 1),
(1, 1) 6 lA
(118)
(119)
2 < 3. If l 2 = 4 or 5, DTCC requires that the

which yield no moduli solutions while lA
A
third brane be of D-type, which necessarily leads to less than two asymptotically free USp
groups available in the hidden sector.
As for the second form of a-brane in this case,
(L1 , 1) (L2 , 1) (1, L3 ),
(120)
where L1 2, L2 3 and L3 3, the derivation is similar. For the case L1 2, both Aand B-type branes are required. This is obvious for L1 3. For L1 = 2, if there is no Aand B-type branes, the fact that ATC and BTC are filled requires n12 = n13 = 0, in order
to avoid the ATCC and BTCC violation, which in turn will lead to a problematic even
factor from the first WNPs since L1 is even. These two necessary NZ-type branes have
DA = DB = 1 due to DTCC. If the B-brane is of B1-type, the factor 3 for IaB comes
from the first WNP and thus we have n1a = 2 and L2 = 1. Then, the possible solutions for
185
a-brane and the B-brane will be

(2, 1) (1, 1) (1, 1),
(121)
(1, 1) (1, 1) (1, 3),
(122)
with L3 = 1 required by the BTCC because of BB L3 + 2. It is easy to check that under

this setup, A1-brane cannot obtain the intersection factor 3 to avoid the BTCC and CTCC
violation, and A2-brane cannot yield the moduli solution by combining with the B-type
brane.
Now let us consider the combination of B2- and A1-branes. L2 cannot be larger than 3,
otherwise, one problematic factor will be generated for IaA from the second WNPs. If
L2 = 3, B- and D-type USp groups are no longer asymptotically free. To ensure C-type
USp group is asymptotically free, we must have L1 = 2. Then the only possible solutions
are
(2, 1) (3, 1) (1, 1),
(123)
(+, 1) (, 1) (1, 1),
(124)
(1, 1) (3, 1) (1, 3).
(125)
n1B
But, due to BTCC,

cannot be smaller than 3, i.e., C-type USp group is still not asymptotically free. So we must have L2 = 1. As for L3 , we shall have the same situation if it is
larger than 1. So, the last possible solutions for a-, B- and A-branes are
(L1 , 1) (1, 1) (1, 1),
(126)
(L1 3 > 0, 1) (3, 1) (1, 1),

(L1 3) < 0, 1 (1, 1) (1, 3),
(127)
(128)
3 = n2 = 3 is because the intersection factor 3s of I

where lA
aB and IaA cannot come from
B
the tilted two-tori. This is straightforward to see. For example, if the intersection factor 3
of IaB comes from the 2nd WNP, we must have n2B = 7 and then the ATCC will require
n1A 7. This is not allowed since B-, C- and D-type USp groups are not asymptotically
free in this case. For this set of possible solutions, ATCC requires L1 3, then we shall
have n1B = n1A = L1 + 3; BTCC is violated. As for the B2A2 combination, the factor 3
for IaA comes from the first WNP, and thus we have L3 = 1. The possible solutions are
(2, 1) (1, 1) (1, 1),

1
nB > 0, 1 (3, 1) (1, 1),
(129)
(1, 1) (3, 1) (1, 1),
(131)
(130)
where L2 = 1 is required by ATCC because of AA L2 + 2, and n2B 3 is required by

CTCC. Now it is easy to see that ATCC will forbid IaB obtain the intersection factor 3. As
a result, for L1 2, there is no solution.
We now turn to L1 = 1 case. If L2 3, B- and D-type USp groups cannot appear in
the hidden sector and a-brane will have the form
(1, 1) (L2 , 1) (1, 1).
(132)
186
One A-type brane is required. For A1-brane, the factor 3 for IaA will be contributed by the
3 = 3. If n3 = 3, the third brane is of D1-type since
second WNPs and thus we have n3A lA
A
D2-type brane cannot generate the zero factor for IaD . Because no extra positive BTC is
available, we must have Ba = BA = BD = 1. Then the only possible solutions are
(1, 1) (3, 1) (1, 1),
(133)
(1, 2) (3, 1) (3, 1),
(134)
(1, 2) (3, 1) (1, 1),
(135)
3 = 3, the third brane is of B-type and it should

which, however, is excluded by CTCC. If lA
be B2-brane since the problematic intersection factor cannot be avoided for B1-brane. Due
to BTC = DTC = 0, the possible solutions are
(1, 1) (3, 1) (1, 1),
(136)
(2, 1) (3, 1) (1, 3),
(137)
(4, 1) (1, 1) (1, 1),
(138)
which means that the C-type USp group is not asymptotically free, either.
If the A-brane is of A2-type, the factor 3 of IaA will be provided by the first WNPs,
1 | = 2. Meanwhile, l 2 cannot be larger than 1 to satisfy BTC = DTC = 0
indicating |n1A lA
A
while the 3rd brane is also counted in. Then the possible solution for a- and A2-brane will
be
(1, 1) (3, 1) (1, 1),

(1, 2) n2A < 0, 1 (1, 1),
(139)
(140)
1 = 2 is because there is no moduli solutions for l 1 = 1. But, no proper value is

where lA
A
available for n2A : |n2A | = 1 will lead to ATCC or DTCC violation when the third brane is
included, and |n2A | = 5 will make C-type USp group unavailable since the third brane must
be of Z-type to satisfy DTC = 0. As a result, there is no solution for L2 3. As for the
case where L3 3, due to the extended type I T-duality, it is dual to the case where L2 3
according to Eq. (32), and thus all above conclusion can be applied directly to this case.
At last, let us recall that the ABS of all branes wrapping numbers are less than 8 under
the survived two setups of a-brane
(1, 1) (1, 1) (1, 1),
(141)
(1, 1) (1, 1) (1, 1).
(142)
For the former, the proof is exactly the same as in the case of one-tilted two-torus. For
the latter, the only difference happens where n12 = 0 and the third brane is of D-type. In
that case, ATCC, BTCC and CTCC will impose on the 3rd brane the constraint D3 5.
Combining this constraint and CTCC, it implies that the wrapping numbers of the 2nd
brane cannot be larger than 8 in this case, either.
187
5.2.2. Case II: NZ-type a stack of D6-branes with two and three negative wrapping
numbers
In this case, the forms of a-brane allowed by the SUSY conditions include
(L1 , 1) (L2 , 1) (1, L3 ), (L1 , 1) (L2 , 1) (1, L3 ),
(143)
(L1 , 1) (L2 , 1) (1, L3 ), (L1 , 1) (L2 , 1) (1, L3 ).
(144)
The first case corresponds to

(L1 , 1) (L3 , 1) (1, L2 ),
(145)
by a T-duality combination of the Eqs. (32) and (48). As for the last three cases, they
correspond to
(L1 , 1) (L2 , 1) (1, L3 ),
(146)
(L1 , 1) (L2 , 1) (1, L3 ),
(147)
(L1 , 1) (L2 , 1) (1, L3 ),
(148)
respectively, according to the extended type II T-duality (48). Therefore, these setups are
also excluded due to T-duality.
5.2.3. Case III: Z-type a stack of D6-branes
Let us consider the case where the a-brane is of Z-type. Suppose that the zero wrapping
number comes from the untilted two-torus, then due to T-duality, we can assume Aa =
Ba = 0 or n1a = 0 and this a-brane has the form
(0, 1) (+, +) (+, +).
(149)
For the four unknown positive wrapping numbers, at least one of them is larger than 2;
if not, all of them are equal to 1. Without loss of generality, we may suppose n2a is such
a wrapping number, then it must be true that la2 = n3a = la3 = 1. Before verifying this, we
point out first that the other two branes should be of C- and D-type only if one of la2 , n3a and
la3 is not equal to 1. This is obvious if la3 is not the largest one among these three wrapping
numbers. If la3 is the largest one, it should be larger than 2, and then A- and B-type USp
groups are no longer asymptotically free. C-type brane is definitely required due to n2a 3
here. So, to preserve the D-type USp group in the hidden sector, we must have one D-type
brane to provide the extra positive DTC. Next, we shall discuss these cases separately.
For n3a > 1, the intersection factor 3s of IaC and IaD will come from the tilted two-tori,
which means, among n2C , n3C , n2D and n3D , there are two whose ABS is equal to 3. This
will lead to ATCC violation. Thus we must have n3a = 1. For la3 , it cannot be larger than 3,
otherwise, we shall have n2C = n2D = 3 to generate the intersection factor 3 and ATCC will
be violated again. If la3 = 3, the only possible solutions are
(0, 1) (3, 1) (1, 3),
1
1, l2 (3, 1) (1, 1),
1
1, l3 (1, 1) (1, 3).
(150)
(151)
(152)
188
Obviously, no matter what values l21 and l31 take, CTCC and DTCC cannot be satisfied at
the same time. So la3 must be equal to 1 also. For la2 > 1, suppose that n2a > la2 due to Tduality, the intersection factor 3, no matter for IaC or IaD , cannot come from the first WNP.
For example, if IaC s does, ATCC and BTCC will require AD = BD = 1 since now we
have AC = BC = 3. Then the D-type brane cannot obtain the intersection factor 3. As a
result, the only possible solutions will be

(0, 1) n2a , la2 (1, 1),
(153)
2

1
2
1, lC > 0 nC > 0, lC > 0 (1, 1),
(154)
2

1
2
1, lD > 0 nD > 0, lD > 0 (1, 1).
(155)
2 | 3, it is not hard to check one by one that the intersection
Since |n2C lC2 | 3 and |n2D lD
factor 3 of IaC and IaD cannot be generated from the second WNPs at the same time. So,
the a-brane cannot have two wrapping numbers with their ABSes larger than 1 at the same
time. The last possibility is then

(0, 1) n2a > 0, 1 (1, 1).
(156)
For the last possible setup of a-brane, if n2a 3, one C-type brane is required. For a
C1-brane, the intersection factor 3 for IaC comes from the second WNP, and the possible
solutions of a- and C1-branes are
(0, 1) (3, 1) (1, 1),

1
1, lC > 0 (3, 1) (1, 3),
(157)
(158)
where n3C must be equal to 1 in order to avoid ATCC and/or DTCC violation after the third
brane is included. Meanwhile, to generate the intersection factor 3 of Ia3 , |n13 n23 | 3 or
|A3 | 3 cannot be avoided, so the last brane should be of A-type. Noticing that |n13 | = 3 is
not allowed by BTCC, we have |n23 | = 3, which necessarily leads to no enough asymptotically free USp groups in the hidden sector. For a C2-brane, the intersection factor 3 of IaC
cannot come from the first WNPs. If it does, one extra D- or A-type brane is required. For
a D-type brane, AD = BD = 1 will forbid IaD to obtain the intersection factor 3. For an
3 | = 1 which will lead to I
A1-brane, BTCC and DTCC require |n3A lA
aA = 0. For an A2brane, we have BA = DA = DC = 1 and thus the intersection factor 3 of IaA can only
come from the second WNPs. The only possible solutions are
(0, 1) (3, 1) (1, 1),

(3, 1) n2C , 1 (1, 1),
(159)
(1, 1) (3, 1) (1, 1).
(161)
n2C
(160)
Obviously,
has no solution to satisfy CTCC and ATCC at the same time. As a result,
the intersection factor 3 of IaC has to come from the 2nd WNPs and the only possible
solutions for a- and C2-brane will be

(0, 1) n2a 3, 1 (1, 1),
(162)
2

1
1, lC > 0 na 6 > 0, 1 (1, 1),
(163)
189
where lC2 cannot be larger than 3 to avoid the requirement of additional B- and D-type
branes. If it is equal to 3, there is no solution for n2a satisfying the co-prime condition. It
cannot be equal to 2 or 3 also to avoid the requirement of additional A- and D-type branes
are required at the same time. If it is equal to 2, we have n2C 4 and two additional ATC
and DTC are required together. As for the n2C , if it is equal to n2a + 6, the third brane is of
2 = 1 and |n1 l 3 | = 3. If l 3 = 3, this is
A-type, and we have BA = 3 and DA = 1, or lA
A
A A
a A1-brane and n2A = n2a 6 > 0. It is easy to check that in either case, ATCC and CTCC
3
= 1, this is a A2-brane and we have n1A = 3. At
cannot be satisfied at the same time. If lA
this time, a problematic intersection factor for IaA will be generated from the 2nd WNPs.
If n2C = n2a 6, CTCC requires lC1 > 2 and thus the 3rd brane is of D-type. Meanwhile, we
2 l 3 = 3. For a D2-type brane, one problematic intersection factor for I
have BD = n1D lD
aD
D
3
2 = 3,
will be generated by the 2nd WNPs due to n2a 7. For a D1-type brane, lD
= 1; if lD
n1D = 1 and there is no solution for n2D satisfying the co-prime condition. Therefore, the
possible solutions are

(0, 1) n2a 7, 1 (1, 1),
(164)

1, lC1 > 0 (1, 1) (1, 1),
(165)

1
3, lD > 0 (1, 1) (1, 1),
(166)
1 satisfying CTCC and DTCC at the same time.
regretfully, there is no solution for lC1 and lD
Based on the above analysis, we can declare the no-go theorem for n2a > 1. As for the case
where n2a = 1, following the same logic as applied in the case of one-tilted two-torus, it is
easy to see that no wrapping number can have ABS larger than 8.
If the zero wrapping number comes from one of the tilted two-tori, due to type I
T-duality and its extended version again, we may take n2a = 0 and thus |la2 | = 2. The other
two branes will be B2- and D1-branes. If they are of Z-type, their second WNPs have
to satisfy |n2 | = 2 and l 2 = 0, which will yield a problematic intersection factor 4 since
the third WNPs have yielded another even factor. Furthermore, they must be of B2- and
D1-types because B1- and D2-types branes cannot provide the zero factor to IaB and IaD .
The possible solutions then are

1
na > 0, la1 > 0 (0, 2) (1, 1),
(167)
(+, +) (1, ) (1, 1),
(168)
(+, +) (1, +) (1, 1),
(169)
up to the inferential Type II T-duality. Here the intersection factor 3 for IaB and IaD cannot
come from the 2nd WNPs. If one of them does obtain the intersection factor 3 from the
2nd WNPs, the other one will be forbidden to obtain the intersection factor 3 due to ATCC
and CTCC constraints. The factor 3 cannot come from the 3rd WNPs also since one of
ATCC and CTCC will be violated only if n3a or la3 is equal to 3. Now let us consider the
last casethey are from the 1st WNPs. To generate these intersection factor 3, n1a = la1 = 1
1 3 due to ATCC and CTCC. Without loss of
is not allowed, since n1B lB1 3 and n1D lD
1
1
generality, let us suppose na < la . Then according to IaB = IaD , we have
1 1

1
1
na = nB + n1D la1 .
lB + lD
(170)
190
1 4, n1 + n1 4 and the co-prime conditions, (n1 , l 1 ) have three possible

Given lB1 + lD
a a
B
D
solutions (3, 4), (2, 3) and (1, 2). Now it is not hard to figure out one by one that the
intersection factor 3s cannot be generated from the 1st WNPs for IaB and IaD at the same
time. As a result, there is no solution in this case.
5.3. Three tilted two-tori

If all two-tori are tilted, for one set of definite values arranged to {A, B, C, D}, models
characterized by any permutations of them actually are T-dual to each other. If a-brane is of
NZ-type, considering that there is only one wrapping number L with its ABS larger than 1
to preserve enough asymptotically free USp groups, we only need to consider two kinds
of possible setups for a-brane: {A = B = L, C = D = 1} and {A = B = L, C =
D = 1}. And the C- and D-type USp groups are not asymptotically free.
For the former, without loss of generality, we may suppose that a-brane is of A1-type
up to the inferential type II T-duality,
(L, 1) (1, 1) (1, 1).
(171)
Then one of the other two branes must be of B-type. In the case of a B1-brane, the intersection factor 3 will come from the 1st WNP and L = n1B = n2B lB2 = n3B lB3 = 3. So, no
enough asymptotically free USp groups can be preserved. In the case of a B2-brane, still
for the sake of enough asymptotically free USp groups in the hidden sector, the intersection
factor 3 must come from the first WNPs. Meanwhile, due to CTCC and DTCC, we have
lB1 = 1 or lB1 = lB2 = lB3 = n2B = n3B = 1 where there are no moduli solutions for a- and
B2-branes.
For the latter, we may suppose that a-brane is a C1-brane
(L, 1) (1, 1) (1, 1).
(172)
Then the other two branes must be of A- and B-types and DA = DB = 1 since a-brane
is of C-type. In order to keep enough asymptotically free USp groups, the intersection
factor 3 must come from the first WNPs and only the combinations of A1- and B2-branes,
and A2- and B1-branes are allowed
(L, 1) (1, 1) (1, 1),

(L 6) < 0, 1 (1, 1) (1, 3),
(173)
(L 6 > 0, 1) (3, 1) (1, 1),
(175)
(L, 1) (1, 1) (1, 1),

(L 6) < 0, 1 (3, 1) (1, 1),
(176)
(L 6 > 0, 1) (1, 1) (1, 3).
(178)
(174)
and
(177)
For the first one, ATCC and BTCC cannot be satisfied only if L > 1; for the second one,
the combination of A2- and B1-branes gives no moduli solutions.
191
As for the case where L = 1, it is easy to see that the ABS of all branes wrapping numbers are less than 8. Meanwhile, if a-brane has one vanishing wrapping number, supposing
(0, 2) for the second WNP of a-brane, we know that the intersection factor 3 of IaB
and IaD cannot come from the 2nd WNPs. Meanwhile, they cannot come from the same
WNPs also due to ATCC and CTCC. So, the only possible solutions are
(1, 3) (0, 2) (3, 1),

(1, 3) 1, l22 (1, 1),

(1, 1) 1, l32 (3, 1).
(179)
(180)
(181)
However, in this case the BTCC or DTCC is violated. Thus, no solution exists!
5.4. Preliminary phenomenological features of the models
After analytically excluding most of parameter space for wrapping numbers, we wrote
a computer program to scan the rest parameter space. The results indicate that no model
is available for the cases with two and three tilted two-tori. For the case with one-tilted
two-torus, we obtain 11 inequivalent models which can be specified by two classes.
The first class include Tables 717, which correspond to the models that do not possess
the gauge coupling unification for SU(2)L and SU(2)R at the string scale and whose Higgs
doublets are generated by the intersections of b- and c-stacks of branes.
For the second class, which includes Tables 18 and 19, they have the SU(2)L and
SU(2)R gauge coupling unification at the string scale. Because the b-stack branes for these
models are parallel to both c-stack branes and their images on one of the three two-tori,
their Higgs doublet pairs come from the massless open string states in a N = 2 subsector
and form vector-like pairs.
For all of these models, except I-Z-7, I-Z-8, and I-Z-9, the number of the pairs of Higgs
doublets is less than 9, which could make them phenomenologically interesting. In particular, there are two pairs of Higgs doublets in model I-NZ-1a. As an example, the chiral
particle spectra for models I-NZ-1a, I-Z-5 and I-Z-10 are presented in Tables 35, respectively. For model I-NZ-1a, it has 2 pairs of Higgs doublets and two confining hidden sector
gauge group factors. Model I-Z-5 has 3 pairs of Higgs doublets and three confining gauge
group factors. Model I-Z-10 has the gauge coupling unification of SU(2)L and SU(2)R at
the string scale and 4 confining gauge group factors. These models thus possess potentially
phenomenologically attractive features.
Hidden sector gauge groups with negative beta functions have a potential to be confining. In these cases, the gaugino condensations generate the non-perturbative effective
superpotential. The minimization of this supergravity potential determines the ground state
which can stabilize the dilaton and complex structure toroidal moduli, and in some cases
breaks supersymmetry. For the models with two confining USp(N) gauge groups, a general analysis of the non-perturbative superpotential with tree-level gauge couplings shows
[33] that there can be extrema with the dilaton and complex structure moduli stabilized,
however, the extrema are saddle points in general and they do not break supersymmetry
in general. On the other hand, for the models with three or four confining USp(N) gauge
groups, the non-perturbative superpotential in general allows for the stabilization of moduli
192
Table 3
The chiral spectrum in the open string sector of model I-NZ-1a
ab
SU(4) SU(2)L SU(2)R

USp(2)USp(4)USp(2)
1, 1, 1, 1)
3 (4, 2,
ac
bc
a1
1, 2, 1, 1, 1)
3 (4,
2 (1, 2, 2, 1, 1, 1)
1 (4, 1, 1, 2, 1, 1)
0
1
1
0
1
0
a2
1, 1, 1, 4, 1)
1 (4,
1, 0, 0, 1
1, 1
6
2
a3
1 (4, 1, 1, 1, 1, 2)
16 , 12
b1
1, 2, 1, 1)
2 (1, 2,
b2
1 (1, 2, 1, 1, 4, 1)
I-NZ-1a
Qem
BL
Field
13 , 23 , 1, 0
1 , 1
3
31 , 1
QL , LL
1 , 2 , 1, 0
3
3
Q4
Q2L
Q2R
1
3 , 1
31 , 1
1
3 , 1
c3
4 (1, 1, 2, 1, 1, 2)
a
4 (6, 1, 1, 1, 1, 1)
1, 1
6
2
12
12
12
13 , 1
b
b
1 (1, 3, 1, 1, 1, 1)
1, 1, 1, 1)
1 (1, 1,
0
0
2
2
0
0
0, 1
0
0
0
c
c
1, 1, 1)
3 (1, 1, 3,
3 (1, 1, 1, 1, 1, 1)
0
0
0
0
2
2
0, 1
0
0
0

QR , LR
0
0
0
32 , 2
Table 4
The chiral spectrum in the open string sector of model I-Z-5
I-Z-5
SU(4) SU(2)L SU(2)R USp(2)3
ab
1, 1, 1, 1)
3 (4, 2,
ac
bc
1, 2, 1, 1, 1)
3 (4,
2,
1, 1, 1)
3 (1, 2,
a1
1, 1)
1 (4, 1, 1, 2,
a2
1, 1, 1, 2, 1)
1 (4,
b2
1)
1 (1, 2, 1, 1, 2,
c1
2, 1, 1)
2 (1, 1, 2,
c3
b
b
3 (1, 1, 2, 1, 1, 2)
2 (1, 3, 1, 1, 1, 1)
1, 1, 1, 1)
2 (1, 1,
0
0
0
0
2
2
1
0
0
c
c
1, 1, 1)
1 (1, 1, 3,
1 (1, 1, 1, 1, 1, 1)
0
0
0
0
2
2

Q4
Q2L
Q2R
Qem
BL
Field
13 , 23 , 1, 0
1 , 2 , 1, 0
3
3
1 , 1
3
31 , 1
QL , LL
1, 0, 0, 1
1, 1
6
2
16 , 12
12
12
12
1 , 1
3
31 , 1
0, 1
0
0
0
0
0, 1
0
0
0
QR , LR
H
0
0
and breaking of supersymmetry at the stable extremum. (For an explicit analysis of three
confining USp gauge group factors see Ref. [33].) In our case, six models (I-NZ-1a, I-Z-2,
I-Z-3, I-Z-7, I-Z-8 and I-Z-9) have two USp(N) gauge groups with negative beta functions, three models (I-Z-1, I-Z-4 and I-Z-5) have three confining USp(N) gauge groups,
and two models (I-Z-6 and I-Z-10) have four confining USp(N) gauge groups. Therefore,
193
Table 5
The chiral spectrum in the open string sector of model I-Z-10
I-Z-10
ab
1, 1, 1, 1, 1)
3 (4, 2,
ac
1, 2, 1, 1, 1, 1)
3 (4,
a1
1 (4, 1, 1, 2, 1, 1, 1)
a2
1, 1, 1, 2, 1, 1)
1 (4,
b2
1 (1, 2, 1, 1, 2, 1, 1)
b4
1, 1, 1, 1, 2)
3 (1, 2,
c1
2, 1, 1, 1)
1 (1, 1, 2,
c3
b
b
3 (1, 1, 2, 1, 1, 2, 1)
2 (1, 3, 1, 1, 1, 1, 1)
1, 1, 1, 1, 1)
2 (1, 1,
0
0
0
0
2
2
1
0
0
c
c
1, 1, 1, 1)
2 (1, 1, 3,
2 (1, 1, 1, 1, 1, 1, 1)
0
0
0
0
2
2

Q4
Q2L
Q2R
Qem
B L
Field
13 , 23 , 1, 0
QL , LL
1 , 2 , 1, 0
3
3
1, 1
6
2
16 , 12
12
12
12
12
1 , 1
3
13 , 1
1 , 1
3
13 , 1
QR , LR
0
0
0
0, 1
0
0
0
0
0, 1
0
0
0
for the latter five models, due to gaugino condensations there may be stable extrema with
the stabilized moduli and broken supersymmetry. These five models are also very interesting from other phenomenological points of view. Note also that similar to the case studied
in Ref. [33], at the extremum the cosmological constant is likely to be negative and close to
the string scale, and thus in these models the gaugino condensation is not likely to address
the cosmological constant problem.
We should emphasize that all the models possess exotic particles charged under the hidden gauge group factors. There is a possibility that the strong coupling dynamics of the
hidden sector at some intermediate scale would provide a mechanism for all of these particles to form bound states or composite particles (compatible with anomaly cancellation
constraints); these particles would in turn be charged only under the SM gauge symmetry
[30], which is similar to the quark condensation in QCD. Generally speaking, USp groups
have two kinds of neutral bound states. The first one is the pseudo inner product of two
fundamental representations, which is generated by reducing the rank 2 anti-symmetric
representation and is generic for USp groups. This is somehow the reminiscent of a meson,
which is the inner product of one pair of SU(3) fundamental and anti-fundamental representations in QCD. The second one is rank 2N anti-symmetric representation for USp(2N)
group with N 2, which is also a singlet under USp(2N) transformation and is similar to
a baryon, a rank 3 anti-symmetric representation in QCD. A definite example containing
such a singlet is model I-Z-2 whose groups in the hidden sector are USp(4) USp(4).
For N = 1, this singlet is identified with the first one. In order to explicitly show how to
form the composite states, one concrete example from Model I-Z-10 is given out, with
the confined particle spectra tabulated in Table 6. Model I-Z-10 has four confining gauge
groups. Thereinto, both USp(2)1 and USp(2)2 have two charged intersections. So for them,
besides self-confinement, the mixed-confinement between different intersections is also
1, 1, 1, 1) and (4,
1, 2, 1, 1, 1, 1).
possible, which yields the chiral supermultiplets (4, 1, 2,
194
Table 6
The composite particle spectrum of Model I-Z-10, which is formed due to the strong forces from hidden sector
Model I-Z-10
Confining force
Intersection
Exotic particle spectrum
Confined particle spectrum
USp(2)1
a1
c1
a2
b2
c3
b4
1 (4, 1, 1, 2, 1, 1, 1)
2, 1, 1, 1)
1 (1, 1, 2,
1, 1, 1, 2, 1, 1)
1 (4,
1 (1, 2, 1, 1, 2, 1, 1)
3 (1, 1, 2, 1, 1, 2, 1)
1, 1, 1, 1, 2)
3 (1, 2,
1, 1, 1, 1)
1 (42 , 1, 1, 1, 1, 1, 1), 1 (4, 1, 2,
1 (1, 1, 2 2 , 1, 1, 1, 1)
1, 2, 1, 1, 1, 1)
1 (4 2 , 1, 1, 1, 1, 1, 1), 1 (4,
1 (1, 22 , 1, 1, 1, 1, 1)
3! (1, 1, 22 , 1, 1, 1, 1)
3! (1, 2 2 , 1, 1, 1, 1, 1)
USp(2)2
USp(2)3
USp(2)4
As for USp(2)3 and USp(2)4 , they have only one charged intersection. Thus, there is no
mixed-confinement, and self-confinement leads to 3! tensor representations for both ones.
In addition, it is not hard to check from the spectra that no new anomaly is introduced to
the remaining gauge symmetry, i.e., this model is still anomaly-free. This set of analysis
works as well for the other models except Model I-NZ-1(ac) and Model I-Z-8 where a
non-asymptotical free gauge symmetry appears in the hidden sector, and the states charged
under this symmetry have no chance to be confined. By the way, the anomaly-cancellation
for the confined particle spectra is not automatically guaranteed. For our models, all gauge
factors in the hidden sectors are pseudoreal, and thus there is no anomaly constraint on the
number of chiral supermultiplets charged under these gauge symmetries. Sometimes one
additional field associated with composite states is also required to satisfy t Hooft anomaly matching condition. Here, we only consider one relatively simple example in order to
avoid the unnecessary complications.
6. Some other potentially interesting setups

In addition to the models that we have discussed, there are some other potentially interesting constructions that could lead to the SM constructions. For example, if we assume
Iac = (3 + h),
Iac = h,
(182)
with h a positive integer, we can have massless vector-like Higgs fields which can break
the PatiSalam gauge symmetry down to the SM gauge symmetry or break the U (1)BL
U (1)I3R down to U (1)Y . But, due to the large wrapping numbers required by the increased
ABSes of Iac and Iac , it is very difficult to find such models. For instance, let us focus on
the h = 1 case with one tilted two-torus. Considering that all factors of 2k Iac = 8 and
2k Iac = 2 are even, they should come from the 3rd tilted two-torus, i.e., we should have
|n3a lc3 | = 5 or 3 and |n3c la3 | = 3 or 5. This implies that the 1st and 2nd WNPs will contribute
a unit factor to both 2k Iac = 8 and 2k Iac = 2, and thus we must have n1a la1 = 0 and
n2c lc2 = 0. Therefore, both a- and c-branes are of Z type. If |n3a la3 | = 15, obviously there
is no solution due to the TCC violation. If |n3c lc3 | = 15, given |n3b lb3 | = 1 to generate the
even factors of 2k Iab = 6 and 2k Iab = 0 and the constraints from BTCC and CTCC, the
ATCC cannot be satisfied. If |n3a la3 | = 5, to avoid the TCC violation, the even factors of
195
2k Iab = 6 and 2k Iab = 0 cannot be generated from the third WNPs at the same time. As
for |n3a la3 | = 3, there are less than two asymptotically free USp groups available in order to
satisfy ATCC or BTCC for |n3a n3c | = 15 or |la3 lc3 | = 15, respectively. In short, it is hard to
find such models.
Another interesting possibility would correspond to constructions where the SU(2)L
and/or SU(2)R gauge symmetries come from filler branes, i.e., the SU(2)L,R = USp(2)L,R .
In this case the number of the SM Higgs doublet pairs may be decreased. However, we do
not want to obtain SU(2)L,R from the splittings of higher rank USp(N) (N 4) branes
which would generically lead to even number of families. In this case, one wrapping number with its ABS larger than 2 cannot be avoided for the U (4) brane. This may make the
model building very hard due to tadpole cancellation constraints. However, further exploration of these models is needed.
7. Discussions and conclusions

In this paper we reviewed the rules for supersymmetric model building, and the conditions for the tadpole cancellations and 4-dimensional N = 1 supersymmetric D6-brane
configurations in the Type IIA theory on T 6 /(Z2 Z2 ) orientifold with D6-brane intersections. Subsequently, we highlighted the interesting features of the three-family supersymmetric SU(4)C SU(2)L SU(2)R models where all the gauge symmetries arise from the
stacks of D6-branes with U (n) gauge symmetries. In particular, we demonstrated that the
PatiSalam gauge symmetry can be broken down to the SU(3)C SU(2)L U (1)BL
U (1)I3R via D6-brane splittings, and further down to the Standard Model gauge symmetry
via the D- and F-flatness preserving Higgs mechanism where Higgs fields arise from the
massless open string states in a specific N = 2 subsector of the theory. In order to stabilize
the (complex structure) modulus and provide a possibility to break supersymmetry via a
race-track scenario, we required that there be at least two confining USp groups in the
hidden sector.
In order to facilitate the systematic search, we discussed the T-duality and its variations that are in effect for the intersecting D6-brane constructions on Type IIA orientifolds.
Employing these symmetries allowed us to search for all inequivalent models with above
specified properties. We found no models in the case when zero, two, or three two-tori
are tilted. For the case with only one tilted two-torus, we obtain 11 inequivalent models.
Eight of them have 8 or fewer pairs of the SM Higgs doublets. Especially, one model has
only 2 pairs of the SM Higgs doublets. Furthermore, two models have the gauge coupling
unification for SU(2)L and SU(2)R gauge group factors at the string scale and the Higgs
pairs for them arise from the massless open string states in a N = 2 subsector. The explicit
brane configurations, their intersections, the gauge group structure (and the hidden sector
beta functions) are tabulated for all these models in Appendix A. As explicit examples,
we also present the chiral spectra in the open string sector for the models I-NZ-1a, I-Z-6
and I-Z-10. We also briefly comment on the other potentially interesting configurations
of intersecting D6-branes which could lead to the three family supersymmetric Standard
Model. In particular, the setup when the origin of SU(2)L and/or SU(2)R comes from the
196
USp brane configurations is extremely constraining; it seems to be very difficult to find the
supersymmetric three-family models of this type.
Models presented in this paper provide a promising stepping stone toward the realistic
SM models from string theory. In particular, the symmetry breaking chain of the original PatiSalam models (via brane splitting and subsequent Higgs mechanism) allows for
obtaining only the Standard Model gauge group structure at electroweak scale. While we
made some preliminary comments on the phenomenological features of these models constructed, there are a number of avenues opened for further study.
One should study the renormalization group equations for the running of the gauge
couplings, both in the observable and hidden sectors. We expect that due to the exotic
matters and the additional adjoint chiral superfields, the low energy predictions for the SM
gauge couplings may not be consistent with these from experiments. Note, however, that
the leftright gauge coupling unification at the string scale for some models may provide
interesting consequences for the low energy couplings.
The issue of supersymmetry breaking and modulus stabilization via gaugino condensation in the hidden sector (with at least two confining USp gauge group factors) should be
addressed in detail. If supersymmetry breaking does take place in these models, this breaking can be mediated via gauge interactions, thus providing a possibility to address the CP
problem in this framework. In addition, the nature of the soft supersymmetry breaking parameters and their model dependence, deserve further detailed study. Another important
role of the strong dynamics in the hidden sector could play, is to bind fractionally charged
exotic matter into the composite objects with only SM quantum numbers [30].
An important topic is further study of the Higgs mechanism (that preserves D- and
F-flatness condition) for the breaking of U (1)BL U (1)I3R down to U (1)Y . In order
to be consistent with the see-saw mechanism to explain the tiny neutrino masses, and with
the leptogenesis to produce the baryon asymmetry, we need the symmetry breaking scale at
about 1015 GeV which hopefully can be realized in our models (with radiative corrections
fixing this scale as one possibility).
The study of the SM fermion masses and mixings is also important. Yukawa couplings
can be calculated exactly in conformal field theory [32] and they have a beautiful geometric
interpretation in terms of the angles and areas of the triangles, specified by the location of
brane intersections in the internal space. In particular, the models with few SM Higgs pairs
should provide an important framework to address the textures of the SM fermion mass
matrices in detail [44].
Acknowledgements
We would like to thank Paul Langacker for useful discussions. The research was supported in part by the National Science Foundation under Grant No. INT02-03585 (M.C.)
and PHY-0070928 (T. Li), DOE-EY-76-02-3071 (M.C., T. Liu) and Fay R. and Eugene L.
Langberg Chair (M.C.).
197
Appendix A. Tables for supersymmetric PatiSalam models

In appendix we tabulate all 11 inequivalent models, found by the systematic search.
Thereinto, we present only its equivalence class, specified by T-dualities. In the first column
of each table, a, b and c denote the U (4), U (2)L , and U (2)R stacks of branes, respectively.
1, 2, 3, and 4 represent the filler branes along respective R, R, R and R orientifold planes, resulting in USp(N) gauge groups. N in the second column is the number
Table 7
D6-brane configurations and intersection numbers for the three-family leftright symmetric model I-NZ-1a
Model I-NZ-1a
U (4) U (2)L U (2)R USp(2) USp(4) USp(2)
Stack
(n1 , l 1 ) (n2 , l 2 ) (n3 , l 3 )
a
b
c
8
4
4
(1, 1) (1, 1) (1, 1)

(0, 1) (1, 2) (1, 1)
(1, 0) (1, 4) (1, 1)
(1, 0) (1, 0) (2, 0)
(1, 0) (0, 1) (0, 2)
XA = 12 XB = 5XC = 54 XD
(1 = 5/ 2, 2 = 1/2 2, 3 = 2 2 )
(0, 1) (1, 0) (2, 0)
1 = 2, 2 = 3, 3 = 0
n
n
b
c
0
1
3
4
1
3
3
0
0
2
1
2
0
1
1
0
1
0
4
Table 8
D6-brane configurations and intersection numbers for the three-family leftright symmetric model I-NZ-1b
Model I-NZ-1b
U (4) U (2)L U (4)R USp(2) USp(2) USp(2)
Stack
(n1 , l 1 ) (n2 , l 2 ) (n3 , l 3 )
a
b
c
8
4
4
(1, 1) (1, 1) (1, 1)

(2, 1) (1, 0) (1, 1)
(4, 1) (0, 1) (1, 1)
(1, 0) (0, 1) (0, 2)
(0, 1) (1, 0) (2, 0)
XA = 25 XB = 4XC = 45 XD
(1 = 2 2, 2 = 2/5, 3 = 2 2 )
(0, 1) (0, 1) (2, 0)
2 = 3, 3 = 0, 4 = 2
n
n
b
c
0
1
3
4
1
3
3
0
0
2
1
1
0
1
0
4
1
2
0
Table 9
D6-brane configurations and intersection numbers for the three-family leftright symmetric model I-NZ-1c
Model I-NZ-1c
U (4) U (2)L U (2)R USp(4) USp(2) USp(2)
Stack
(n1 , l 1 ) (n2 , l 2 ) (n3 , l 3 )
a
b
c
8
4
4
(1, 1) (1, 1) (1, 1)

(1, 0) (4, 1) (1, 1)
(0, 1) (2, 1) (1, 1)
(1, 0) (1, 0) (2, 0)
(1, 0) (0, 1) (0, 2)
XA = 2XB = 52 XC = 10XD
(1 = 5/ 2, 2 = 2 2, 3 = 2 )
(0, 1) (0, 1) (2, 0)
1 = 3, 2 = 2, 4 = 0
n
n
b
c
0
3
1
4
3
1
3
0
0
2
1
0
1
1
0
2
1
4
0
198
Table 10
D6-brane configurations and intersection numbers for the three-family leftright symmetric model I-Z-1
Model I-Z-1
U (4) U (2)L U (2)R USp(2) USp(2) USp(2)
Stack
(n1 , l 1 ) (n2 , l 2 ) (n3 , l 3 )
n
n
b
c
a
b
c
8
4
4
(0, 1) (1, 1) (1, 1)

(1, 0) (4, 1) (1, 1)
(1, 1) (1, 2) (1, 1)
0
3
2
0
3
6
3
0
0
7
1
0
2
1
0
1
0
1
2
1
2
3
2
2
2
(1, 0) (1, 0) (2, 0)

(1, 0) (0, 1) (0, 2)
(0, 1) (1, 0) (2, 0)
XA = XB = 94 XC = 9XD
g
g
g
1 = 2, 2 = 3, 3 = 3
Table 11
Model I-Z-2
U (4) U (2)L U (2)R USp(4) USp(4)
Stack
(n1 , l 1 ) (n2 , l 2 ) (n3 , l 3 )
n
n
b
c
a
b
c
8
4
4
(0, 1) (1, 1) (1, 1)

(3, 1) (1, 0) (1, 1)
(1, 1) (1, 2) (1, 1)
0
2
2
0
2
6
3
0
0
4
1
0
2
1
1
1
1
2
4
4
(1, 0) (1, 0) (2, 0)

(1, 0) (0, 1) (0, 2)
XA = XB = 3XC = 3XD
g
g
1 = 2, 2 = 2
Table 12
Model I-Z-3
U (4) U (2)L U (2)R USp(2) USp(4)
Stack
(n1 , l 1 ) (n2 , l 2 ) (n3 , l 3 )
a
b
c
8
4
4
(0, 1) (1, 1) (1, 1)

(3, 1) (1, 0) (1, 1)
(1, 0) (1, 4) (1, 1)
3
4
2
4
(0, 1) (1, 0) (0, 2)

(0, 1) (0, 1) (2, 0)
n
n
b
c
0
2
3
0
2
3
3
0
0
4
0
0
4
0
3
1
XA = XB = 12XC = 3XD
g
g
3 = 2, 4 = 2
Table 13
Model I-Z-4
U (4) U (2)L U (2)R USp(2) USp(2) USp(2)
Stack
(n1 , l 1 ) (n2 , l 2 ) (n3 , l 3 )
a
b
c
8
4
4
(0, 1) (1, 1) (1, 1)

(1, 1) (1, 0) (3, 1)
(3, 2) (0, 1) (1, 1)
(1, 0) (1, 0) (2, 0)
(1, 0) (0, 1) (0, 2)
(0, 1) (1, 0) (2, 0)
n
0
2
1
n
b
c
0
2
1
3
5
0
2
1
0
2
1
3
0
0
0
3
XA = XB = 32 XC = 13 XD
g
g
g
1 = 2, 2 = 1, 3 = 3
199
Table 14
Model I-Z-5
U (4) U (2)L U (2)R USp(2) USp(2) USp(2)
Stack
(n1 , l 1 ) (n2 , l 2 ) (n3 , l 3 )
a
b
c
8
4
4
(0, 1) (1, 1) (1, 1)

(3, 1) (1, 0) (1, 1)
(3, 2) (0, 1) (1, 1)
(1, 0) (1, 0) (2, 0)
2
3
2
2
(1, 0) (0, 1) (0, 2)

(0, 1) (1, 0) (2, 0)
n
n
b
c
0
2
1
0
2
1
3
0
0
3
1
0
2
1
1
0
0
0
3
XA = XB = 32 XC = 3XD
g
g
g
1 = 2, 2 = 3, 3 = 3
Table 15
Model I-Z-6
U (4) U (2)L U (2)R USp(2) USp(2) USp(2) USp(2)
Stack
(n1 , l 1 ) (n2 , l 2 ) (n3 , l 3 )
a
b
c
8
4
4
(0, 1) (1, 1) (1, 1)

(1, 1) (1, 0) (3, 1)
(3, 1) (0, 1) (1, 1)
(1, 0) (1, 0) (2, 0)
XA = XB = 3XC = 13 XD
2
3
4
2
2
2
(1, 0) (0, 1) (0, 2)

(0, 1) (1, 0) (0, 2)
(0, 1) (0, 1) (2, 0)
1 = 3, 2 = 1, 3 = 3, 4 = 5
n
0
2
2
n
b
c
0
2
2
3
4
0
4
1
0
1
1
3
0
0
0
3
0
1
0
Table 16
Model I-Z-7
U (4) U (2)L U (2)R USp(2) USp(4)
Stack
(n1 , l 1 ) (n2 , l 2 ) (n3 , l 3 )
a
b
c
8
4
4
(0, 1) (1, 1) (1, 1)

(1, 1) (1, 0) (3, 1)
(1, 0) (1, 4) (1, 1)
(0, 1) (1, 0) (0, 2)
(0, 1) (0, 1) (2, 0)
n
0
2
3
n
b
c
0
2
3
3
4
0
8
0
0
4
0
1
1
XA = XB = 43 XC = 13 XD
g
g
3 = 2, 4 = 4
of D6-branes in each stack. The third column shows the wrapping numbers of the various
branes, and we have specified the third set of wrapping number for the tilted two-torus.
(Recall, only one two-torus is tilted.) The intersection numbers between the various stacks
are given in the remaining right columns where b and c are, respectively, the R images
of b and c. For convenience, we also tabulate the relation among the moduli parameters
g
imposed by the supersymmetry conditions, as well as the functions (i ) of the gauge
groups in the hidden sector. Again, we required at least two asymptotically free USp groups
with negative functions in the hidden sector. For example, model I-NZ-1a, which forms
200
Table 17
Model I-Z-8
U (4) U (2)L U (2)R USp(2) USp(2) USp(2)
Stack
(n1 , l 1 ) (n2 , l 2 ) (n3 , l 3 )
a
b
c
8
4
4
(0, 1) (1, 1) (1, 1)

(1, 2) (1, 0) (3, 1)
(3, 1) (0, 1) (1, 1)
(1, 0) (1, 0) (2, 0)
2
4
2
2
(1, 0) (0, 1) (0, 2)

(0, 1) (0, 1) (2, 0)
n
n
b
c
0
5
2
3
7
0
10
1
0
1
1
6
0
0
1
0
0
5
2
XA = XB = 3XC = 16 XD
g
g
g
1 = 3, 2 = 2, 4 = 5
Table 18
Model I-Z-9
U (4) U (2)L U (2)R USp(2) USp(2)
Stack
(n1 , l 1 ) (n2 , l 2 ) (n3 , l 3 )
a
b
c
8
4
4
(0, 1) (1, 1) (1, 1)

(3, 2) (1, 0) (1, 1)
(3, 2) (0, 1) (1, 1)
(1, 0) (1, 0) (2, 0)
(1, 0) (0, 1) (0, 2)
n
n
b
c
0
1
1
0
1
1
3
0
0
0
1
0
2
1
2
0
XA = XB = 32 XC = 32 XD
g
g
1 = 2, 2 = 2,
Table 19
Model I-Z-10
U (4) U (2)L U (2)R USp(2) USp(2) USp(2) USp(2)
Stack
(n1 , l 1 ) (n2 , l 2 ) (n3 , l 3 )
n
n
b
c
a
b
c
8
4
4
(0, 1) (1, 1) (1, 1)

(3, 1) (1, 0) (1, 1)
(3, 1) (0, 1) (1, 1)
0
2
-2
0
2
2
3
0
0
0
1
0
1
1
1
0
0
0
3
0
3
0
(1, 0) (1, 0) (2, 0)
XA = XB = 3XC = 3XD
2
3
4
2
2
2
(1, 0) (0, 1) (0, 2)

(0, 1) (1, 0) (0, 2)
(0, 1) (0, 1) (2, 0)
1 = 3, 2 = 3, 3 = 3, 4 = 3
one equivalent class together with I-NZ-1b and I-NZ-1c, has two asymptotically free USp
groups among its three USp groups in the hidden sector. Moreover, I-, -Z- and -NZ- imply
that only one two-torus is tilted, a-brane is Z-type, and a-brane is NZ-type, respectively.
References
[1] J. Polchinski, E. Witten, Nucl. Phys. B 460 (1996) 525.
[2] C. Angelantonj, M. Bianchi, G. Pradisi, A. Sagnotti, Y.S. Stanev, Phys. Lett. B 385 (1996) 96.
[3]
[4]
[5]
[6]
[7]
[8]
[9]
[10]
[11]
[12]
[13]
[14]
[15]
[16]
[17]
[18]
[19]
[20]
[21]
[22]
[23]
[24]
[25]
[26]
[27]
[28]
[29]
[30]
[31]
[32]
[33]
[34]
[35]
[36]
[37]
[38]
[39]
[40]
[41]
[42]
[43]
[44]
M. Berkooz, R.G. Leigh, Nucl. Phys. B 483 (1997) 187.

G. Shiu, S.H. Tye, Phys. Rev. D 58 (1998) 106007.
J. Lykken, E. Poppitz, S.P. Trivedi, Nucl. Phys. B 543 (1999) 105.
M. Cvetic, M. Plmacher, J. Wang, JHEP 0004 (2000) 004;
M. Cvetic, A.M. Uranga, J. Wang, Nucl. Phys. B 595 (2001) 63.
G. Aldazabal, A. Font, L.E. Ibez, G. Violero, Nucl. Phys. B 536 (1998) 29;
G. Aldazabal, L.E. Ibez, F. Quevedo, A.M. Uranga, JHEP 0008 (2000) 002.
M. Klein, R. Rabadan, JHEP 0010 (2000) 049.
M. Berkooz, M.R. Douglas, R.G. Leigh, Nucl. Phys. B 480 (1996) 265.
C. Bachas, hep-th/9503030.
J.F.G. Cascales, A.M. Uranga, hep-th/0311250.
R. Blumenhagen, L. Grlich, B. Krs, D. Lst, JHEP 0010 (2000) 006.
R. Blumenhagen, B. Krs, D. Lst, JHEP 0102 (2001) 030.
G. Aldazabal, S. Franco, L.E. Ibez, R. Rabadn, A.M. Uranga, JHEP 0102 (2001) 047.
G. Aldazabal, S. Franco, L.E. Ibez, R. Rabadan, A.M. Uranga, J. Math. Phys. 42 (2001) 3103.
L.E. Ibez, F. Marchesano, R. Rabadn, JHEP 0111 (2001) 002.
C. Angelantonj, I. Antoniadis, E. Dudas, A. Sagnotti, Phys. Lett. B 489 (2000) 223.
S. Frste, G. Honecker, R. Schreyer, Nucl. Phys. B 593 (2001) 127;
S. Frste, G. Honecker, R. Schreyer, JHEP 0106 (2001) 004.
R. Blumenhagen, B. Krs, D. Lst, T. Ott, Nucl. Phys. B 616 (2001) 3.
D. Cremades, L.E. Ibez, F. Marchesano, Nucl. Phys. B 643 (2002) 93.
D. Cremades, L.E. Ibez, F. Marchesano, JHEP 0207 (2002) 009.
D. Cremades, L.E. Ibez, F. Marchesano, JHEP 0207 (2002) 022.
D. Bailin, G.V. Kraniotis, A. Love, Phys. Lett. B 530 (2002) 202;
D. Bailin, G.V. Kraniotis, A. Love, JHEP 0302 (2003) 052.
J.R. Ellis, P. Kanti, D.V. Nanopoulos, Nucl. Phys. B 647 (2002) 235.
C. Kokorelis, JHEP 0209 (2002) 029;
C. Kokorelis, JHEP 0208 (2002) 036, hep-th/0207234;
C. Kokorelis, JHEP 0211 (2002) 027, hep-th/0210200.
M. Cvetic, G. Shiu, A.M. Uranga, Phys. Rev. Lett. 87 (2001) 201801.
M. Cvetic, G. Shiu, A.M. Uranga, Nucl. Phys. B 615 (2001) 3.
M. Cvetic, I. Papadimitriou, Phys. Rev. D 67 (2003) 126006.
M. Cvetic, I. Papadimitriou, G. Shiu, hep-th/0212177.
M. Cvetic, P. Langacker, G. Shiu, Phys. Rev. D 66 (2002) 066004.
M. Cvetic, P. Langacker, G. Shiu, Nucl. Phys. B 642 (2002) 139.
M. Cvetic, I. Papadimitriou, Phys. Rev. D 68 (2003) 046001.
M. Cvetic, P. Langacker, J. Wang, Phys. Rev. D 68 (2003) 046002.
R. Blumenhagen, L. Grlich, T. Ott, hep-th/0211059.
G. Honecker, hep-th/0303015.
R. Blumenhagen, JHEP 0311 (2003) 055.
R. Blumenhagen, T. Weigand, JHEP 0402 (2004) 041.
I. Brunner, K. Hori, K. Hosomichi, J. Walcher, hep-th/0401137.
T. Li, T. Liu, Phys. Lett. B 573 (2003) 193, hep-th/0304258.
A. Sen, hep-th/9802051.
T.R. Taylor, Phys. Lett. B 252 (1990) 59.
R. Brustein, P.J. Steinhardt, Phys. Lett. B 302 (1993) 196.
B. de Carlos, J.A. Casas, C. Munoz, Nucl. Phys. B 399 (1993) 623.
M. Cvetic, P. Langacker, T. Li, T. Liu, in preparation.
201

Meanfield approximation for field theories

on the worldsheet revisited
Korkut Bardakci a,b
a Department of Physics, University of California at Berkeley, Berkeley, CA 94720, USA
b Theoretical Physics Group Lawrence Berkeley National Laboratory, University of California,
Berkeley, CA 94720, USA

Received 28 April 2004; accepted 5 August 2004
Abstract
This work is the continuation of the earlier efforts to apply the mean field approximation to the
continuum worldsheet formulation of planar 3 theory. The previous attempts were either simple but
without solid foundation or better founded but excessively complicated. In this article, we present
an approach both simple, and also systematic and well founded. We are able to carry through the
leading order mean field calculation analytically, and with a suitable tuning of the coupling constant,
we find string formation.
PACS: 12.38.Cy; 03.70.+k
1. Introduction
Over the last couple of years, the present author and Charles Thorn have been pursuing
a program of summation of planar graphs in field theory [14]. Because of its simplicity, the field theory that has been most intensively investigated so far is the 3 theory,
This work was supported in part by the Director, Office of Science, Office of High Energy and Nuclear
Physics, of the US Department of Energy under Contract DE-AC03-76SF00098, and in part by the National
Science Foundation Grant PHY-0098840.
E-mail address: kbardakci@lbl.gov (K. Bardakci).
doi:10.1016/j.nuclphysb.2004.08.010
K. Bardakci / Nuclear Physics B 698 (2004) 202232
203
although Thorn and collaborators have made considerable progress in extending the program to more physical theories [57]. The basic idea, which goes back to t Hooft [8],
is to represent Feynman graphs on the worldsheet by a suitable choice of the light cone
coordinates. t Hoofts representation, which was non-local, was later reformulated as a local field theory on the worldsheet by introducing additional non-dynamical fields [1]. This
reformulation has no new physical content; it merely reproduces perturbation theory. However, it provides a setup well suited for the study of string formation in field theory. This
is an old problem that has attracted recent renewed interest [912] following the discovery
of AdS/CFT correspondence [13,14].
Our approach to the problem of string formation starts with the worldsheet description
of the 3 field theory mentioned above, and we look for the phenomenon of condensation
of Feynman graphs. This phenomenon will be defined more precisely later in the paper,
but roughly it means that the lines that form Feynman graphs on the worldsheet become
dense, and graphs of asymptotically infinite order dominate the perturbative expansion.
Furthermore, the original non-dynamical worldsheet fields become dynamical and string
formation takes place. Whether the scenario described above really happens is of course a
question of dynamics. So far, the only tool used to investigate this problem in the present
context has been the mean field or the self consistent field approximation [24]. The accuracy of the mean field approximation is questionable; however, one can hope that at least a
qualitative understanding of the relevant dynamical issues would emerge.
The main virtue of the mean field method is its simplicity. There is, however, no unique
way to do the mean field calculation, and it all depends on the choice of the order parameter. In Ref. [2], the simplest choice was made for the order parameter by taking it
to be the expectation value of a local field. This makes the subsequent calculation quite
tractable. However, this early attempt, at least in its continuum version, relied on a number of questionable steps and approximations that are hard to justify. In order to overcome
these difficulties, in Refs. [3,4] the order parameter was taken to be the expectation value
of two fields at different points (two point function). This also has the advantage of providing a better probe of the problem; but the disadvantage is that the calculation becomes too
complicated to carry out analytically.
In this article, we revisit the earlier calculation [2] based on a simple order parameter to
see whether the difficulties associated with it can be overcome without sacrificing its basic
simplicity. We will mainly focus on the treatment based on the continuum worldsheet given
in this reference. Reexamining this treatment, we identify the following problems:
(a) The boundary conditions on the worldsheet were imposed only approximately
through the so-called trick. The exact boundary conditions could only be recovered in
the problematic limit;
(b) If we try to impose the boundary conditions exactly by means of Lagrange multipliers, as was done in [1] and as we shall do here, we avoid the problem discussed in (a), but
instead we encounter another difficulty. The action has then an unfixed gauge invariance,
which can lead to ill defined results;
(c) There are two kinds of fields in the problem: the matter and the ghost fields. The
contribution of each sector is quadratically divergent, but there is a subtle partial cancella-
204
tion between them. Unless great care is exercised, the result can depend on the regulation
scheme used;
(d) The use of light cone coordinates obscures the covariance of the theory. An approximation scheme, such as the mean field approximation, could easily violate Lorentz
invariance.
In this article, we propose the following improvements to overcome the problems listed
above:
(a) The boundary conditions will be imposed exactly by means of Lagrange multipliers;
(b) To avoid the resulting gauge invariance problem, we introduce a gauge fixing term.
This is probably the most important new idea introduced in the present work. As is well
known, working with a gauge unfixed action can lead to lots of confusion;
(c) To keep track of the cancellation between the matter and ghost fields, we impose
supersymmetry on the worldsheet. This idea was already in the air in [1], what we have
done here is to formulate it explicitly. Also, in treating divergent quantities, we first combine the contributions of the matter and ghost sectors, and then we regulate the sum in a
way that does not spoil the cancellation between them. The answer obtained in this fashion
is unambiguous;
(d) There is a particular subgroup of the Lorentz group under which the light cone variables transform linearly. In particular, under boosts along the 1 direction, the variables x
and p scale (see Section 2 for the definition of these variables). All of this is familiar from
the light cone treatment of the bosonic string, where, among other things, the importance
of invariance under this special boost was recognized [15]. In the present context, this was
discussed in Ref. [4]. In this paper, following the treatment given in [4], we will try to
preserve invariance under this special subgroup at each step. We will not, however, try to
investigate invariance under the full Lorentz group. In string theory, full Lorentz invariance
in the light cone framework is realized only for a fixed critical value of D, where D is the
dimension of the transverse space [15]. If the same is true here, this problem is clearly
beyond the scope of a leading large D approximation.
When these improvements are incorporated in the mean field approximation of Ref. [2],
the result is a systematic approximation scheme without any of the ambiguities encountered
in [2]. It is also simple enough so that we are able to carry out the leading order calculation
analytically. With a suitable tuning of the coupling constant, we find string formation, and
in addition, we discover that a new string mode has been dynamically generated.
In organizing this paper, the goal was to present a complete and self-contained treatment
which should be intelligible even to a reader unfamiliar with the previous work on the
subject. As a result, there a good deal of overlap with Ref. [2], and some overlap with
Refs. [3,4]. When we preview the rest of the paper below, we will try to make clear what
is new and what is a review or elaboration of the material covered in the earlier work cited
above.
In Section 2, we briefly review both the rules for Feynman graphs in light cone variables
[8] and the local field theory on the worldsheet from which these graphs follow [1]. We also
discuss the transformation properties of the fields under the special boost mentioned earlier,
205
which manifests itself as a scale transformation on the worldsheet. This is an abridged

version of a more complete discussion given in [4].
In Section 3, worldsheet supersymmetry is introduced and a supersymmetric free action
S0 is constructed. This is a new idea in the present context. S0 differs from the corresponding action introduced in [1] and used in all the subsequent work in the structure of the terms
involving ghosts, and also in the presence of a relatively insignificant bosonic field r required by SUSY. The cancellation of quantum corrections between matter and ghost fields,
which was the reason for the introduction of ghosts in the first place, is now guaranteed by
supersymmetry.
The boundary conditions accompanying S0 are enforced by a term in the action, S1 ,
given in Section 4. Both the boundary conditions and S1 are substantially the same as
those in [1]. To express the new term in the action fully in the language of field theory, one
needs worldsheet fermions which were introduced earlier [1,4]. We have found it slightly
more convenient to work with a somewhat different set of fermions, although it is easy to
show by means of a Bogoliubov transformation that the two are completely equivalent.
In Section 5, we show that the action constructed so far is invariant under a simple gauge
transformation. It is therefore important to fix this gauge, and we show how to do it. We
also note that it is possible to introduce some arbitrary constants in the boundary and gauge
fixing terms in the action. Although the exact theory is not sensitive to these constants, we
keep them around to see how the affect an approximate calculation. The material covered
in this section is completely new.
The mean field approximation which is at the basis of the present work is discussed
in Section 6 from the point of view of the large D limit. All of this is standard material,
well known from the solution of the vector models in in the large N (in this case, large
D) limit [16]. The only thing new here compared to [2] and [3] is the manner in which
the singular determinants resulting from integration over the matter and ghost fields are
regulated: the two determinants are combined into an expression less singular than each
one separately, and regulating the combined expression, we get an unambiguous and scale
invariant answer.
In Section 7, an effective action is constructed and evaluated by the saddle point method
in the large D limit. This effective action is pretty close to but still different in detail from
the one derived in [2]. A standard bosonic string action with positive slope emerges from
this calculation. The important question is then whether this slope is finite. We find that, by
suitably tuning the coupling constant of the 3 theory, the slope can be made finite. This
tuning can be regarded as renormalization: a cutoff dependent coupling constant is traded
for the finite slope parameter.
It seems somewhat surprising that starting with an unstable field theory, no sign of
instability has so far appeared in the string picture. One possibility is that we have not gone
far enough. The calculation of the intercept, which we have not undertaken here, may show
that, as in the bosonic string, some lowest lying states are tachyonic. Or, it may be that the
instability is not visible in the leading order of the large D limit. These possibilities are
discussed in at the end of Section 7.
So far, all the calculations were carried out in the leading order of the large D limit. In
Section 8, we compute a non-leading correction in this limit by expanding the composite
field (see Section 4 for its definition) around its mean value 0 . In addition, is assumed
206
to be slowly varying, and an expansion up to second order in powers of derivatives of this

field is carried out. This is essentially a repetition of the computation done in Appendix B
of [2] from the standpoint of the present approach. We find that, from the worldsheet perspective, becomes a dynamical, propagating field, and from the string perspective, the
string acquires an additional mode, with the same slope as all the other modes. Finally,
in Appendix A, we show that, the mean field computation presented here is completely
equivalent to the standard large N (D) treatment of vector models.
2. A brief review
The Feynman graphs of massless 3 have a particularly simple form in the mixed light
cone representation of t Hooft [8]. In this representation, the evolution parameter is x +
also denoted by , and the conjugate Hamiltonian is p , and the Minkowski evolution
operator is given by

T = exp ix + p .
(1)
In this notation, aMinkowski vector v has the light cone components (v + , v , v), where
v = (v 0 v 1 )/ 2, and the boldface letters label the transverse directions. A propagator
that carries momentum p is pictured as a horizontal strip of width p+ and length = x +
on the worldsheet, bounded by two solid lines (Fig. 1). The lines forming the boundary
carry transverse momenta q1 and q2 , where
p = q1 q2 ,
and the corresponding propagator is given by

( )
(p) = + exp i + p2 .
2p
2p
(2)
More complicated graphs consist of several horizontal solid line segments (Fig. 2). The
beginning and the end of each segment is where the 3 interaction takes place, and a factor
Fig. 1. The propagator.
207
Fig. 2. A typical graph.
of g is associated with each such point, where g is the coupling constant. Finally, one has
to integrate over the position of the interaction vertices, as well as the momenta carried
by the solid lines. We note that momentum conservation is automatic in this formulation.
A typical light cone graph is pictured in Fig. 2.
It was shown in [1] that the light cone Feynman rules described above can be reproduced
by a local field theory on the worldsheet. The worldsheet is parametrized by coordinates
along the p+ direction and along the x + direction, and the transverse momentum q
is promoted to a bosonic field q(, ) defined everywhere on the worldsheet. In addition,
two fermionic fields (ghosts) b(, ) and c(, ) are needed. In contrast to q, which has
D components, b and c each have D/2 components, where D is the dimension of the
transverse space, assumed to be even. The action on the worldsheet, with the Minkowski
signature (+, ), is given by
p
S0 =
f
d
0

1 2

d b c q ,
2
(3)
where the prime denotes derivative with respect to . This action is supplemented by the
Dirichlet conditions
q = 0,
b = c = 0,
(4)
on the solid lines (boundaries), where the dot denotes derivative with respect to . Imposing
Dirichlet conditions on q is equivalent to fixing them to be independent on the solid lines
and then integrating over them. If we fix q to be q1 and q2 on two adjacent solid lines and
solve the equations of motion
q = 0
subject to these boundary conditions, we find that the resulting classical action reproduces
the exponential factor in Eq. (2). However, there is also an unwanted quantum contribution,
given by

1
D det 2
2
208
which is exactly cancelled by the corresponding determinant resulting from integrating

over b and c.
The action formulation described above, which was extensively used in the previous
work [24]. Although it is basically correct, it has some unsatisfactory features. For one
thing, although we have nothing new to say about this problem in this paper, the factor of
1/(2p+ ) in front of the exponential in Eq. (2) is missing. Also, the splitting of the ghost
fields into two components b and c is somewhat unnatural; it leads to the artificial condition
that D is even, and it does not look rotationally invariant, although there is of course no real
violation of rotation symmetry in the transverse space. In the next section, we will show
that the introduction of supersymmetry on the worldsheet leads to a more symmetric ghost
sector and the condition that D be even is no longer needed. We also think that it results in
a more natural and elegant approach.
Finally, we would like to discuss briefly the question of Lorentz invariance. This is a
non-trivial problem, since the use of the light cone coordinates obscures the Lorentz transformation properties of the dynamical variables. There is, however, a special subgroup of
the Lorentz group under which the light cone coordinates have simple linear transformation properties.1 If Li,j are the angular momenta and Ki are the boosts, the generators of
this subgroup are
Li,j ,
M+, = K1 ,
M+,i = Ki + L1,i ,
(5)
where indices i and j run from 2 to D + 2. It turns out that invariance under all the generators except for one is rather trivially satisfied by the propagator (2) or the field theories
(3). The non-trivial transformation, generated by K1 , corresponds to scaling of x + and p+
by a constant u:
x + x + /u,
p+ p+ /u,
(6)
and the transverse momenta q are unchanged. The reason this transformation is critical is
that although it is easy to construct classically scale invariant theories, this symmetry is
in general broken by quantum corrections. This is familiar from the study of conformal
invariance in field theory.
Now, let us examine the scale invariance of the action (3) [4]. Since the coordinates
and must transform like p+ and x + , the transformation laws of the fields are
q(, ) q(u, u ),
b, c(, ) b, c(u, u ),
(7)
and the classical action is invariant under this transformation. On the other hand, the quantum corrections are singular and need a cutoff for their definition. This cutoff would break
scale invariance, were it not for the cancellation between the ghost and matter fields. So
what we have here is a potential violation of scale invariance by an anomaly, which is
eliminated by the cancellation between the matter and ghost sectors. This is nothing but
1 In this article, we will not investigate invariance under the remaining Lorentz generators. If the string analogy
is to be trusted, this is where the critical dimension becomes important [15], and a large D approximation is clearly
inadequate for treating this problem.
209
the cancellation of the determinants discussed earlier. We would like to stress that the quantum contribution in question is a worldsheet effect; it potentially present even in the case
of a free propagator and it has nothing to do with the target space ultraviolet divergences.
In addition to the scaling of the bulk, we have to consider the scaling behaviour of
the boundary conditions given by Eq. (4). These are scale invariant as they stand, but the
integration over the position of the boundaries is not invariant, since the coordinate
scales. The factor of 1/(2p+ ) provides the measure needed to make this integration scale
invariant. Although we will not present here a general recipe for the inclusion of this factor,
we will be careful to preserve the scale invariance of various integrals that occur in the
course of the mean field calculation. It can be shown that, in any case, this factor does not
contribute to the leading order of the mean field calculation [2,3].
3. SUSY on the worldsheet

We generalize the momentum q to form a SUSY multiplet:
Q = q + 1 b + 2 c + 1 2 r,
(8)
where 1,2 are the usual anticommuting coordinates and all the fields represented by boldface letters are vectors in the D-dimensional transverse space. The two SUSY generators
are (i = 1, 2)
Qi =
+ i ,
i
(9)
and they satisfy

Q21 = Q22 = ,
[Q1 , Q2 ]+ = 0.
The infinitesimal transformations are given by

QQ+
i Qi , Q ,
(10)
(11)
where 1,2 are anticommuting infinitesimal parameters. We also define two covariant derivatives
Di =
i ,
i
(12)
which anticommute with the SUSY generators and satisfy the same equations as (10), apart
from the change of the sign of .
Having introduced the framework of supersymmetry, we write down the supersymmetric analogue of the action (3):
p
S0
d
0

d

1
d 2 D1 Q D2 Q
2
210
p

1
d q 2 + b b + c c + r2 ,
2
(13)
supplemented by the boundary conditions

q = 0,
b = 0,
c = 0,
r = 0,
(14)
on the solid lines. We now have twice as many ghost fields as before, but they appear
with only first order derivative in , and as a result, there is again complete cancellation
between the determinants (quantum corrections) coming from the integration of the matter
and ghost fields. It is clear that this cancellation is a consequence of supersymmetry. In fact,
the main motivation for introducing SUSY was to obtain this cancellation as a systematic
consequence of a symmetry, and not as some accident. In the rest of this article, we will
exclusively use this supersymmetric form of the action.
Although there is supersymmetry in the bulk of the worldsheet, it is broken by the
boundary conditions, since the condition on q differs from those on b and c. This breakdown of SUSY is essential, since supersymmetric boundary conditions would lead to
a complete cancellation between the matter and the ghost sectors, resulting in a trivial
propagator. This has no effect on the cancellation of the quantum contributions, since the
cancellation occurs in the bulk and it is insensitive to the boundary conditions.
Since we now have a new expression for S0 , we have to reinvestigate the invariance
under the scaling transformations (6), (7). The action given by Eq. (13) is invariant if the
fields transform as
q(, ) q(u, u ),
b(, ) u b(u, u ),
r(, ) ur(u, u ).
c(, ) uc(u, u ),
(15)
Again, a potential quantum anomaly that could violate scale invariance is cancelled as a
consequence of supersymmetry.
4. The worldsheet action

In this section, we review the construction of an action that incorporates the boundary
conditions (14), which will be implemented by introducing a bosonic Lagrange multipliers
y(, ) and z(, ), and the fermionic Lagrange multipliers b(,

) and c (, ) [1]. The
corresponding term in the action is
p
S1 =
d (y q + b b + c c + z r),
(16)
where (, ) is equal to 1 on the solid lines and it is equal to 0 elsewhere. This definition
is singular on the continuum worldsheet; we will give a more precise definition on the discretized worldsheet below. The factor ensures that the boundary conditions are imposed
only on the solid lines; elsewhere, vanishes and there are no constraints on the fields.
211
Fig. 3. Solid and dashed (dotted) lines.
In the earlier work [1,2,4], it was shown how to construct in terms of a fermionic
field. We will present here a slightly different version of this construction. In order to avoid
singular expressions, it is best to start by discretizing the coordinate into segments of
length = a, with p+ = Na. The parameter a plays the role of a cutoff, which will in
any case be needed later on. The specific form of this cutoff is not important; for example,
a cutoff in the Fourier modes conjugate to would serve just as well. Fig. 3 shows N
equally spaced lines parallel to the direction. For the convenience of exposition, the
absence of a solid line in a given position is pictured by the presence of a dotted line in the
same position. In other words, the solid lines mark the boundaries and the dotted lines fill
the bulk of the worldsheet. On each line, we introduce two fermionic variables i (n , )
and their conjugate i (n , ), where i = 1, 2, n = na, and n is an integer in the range
0 n N . The free fermionic action is given by

n , ),
n , )(
d i (
S2 =
(17)
n
and the fermions satisfy the usual anticommutation relations:

i (m , ), j (n , ) + = i,j m,n .
The function of the fermions is to keep track of the solid and dotted lines. The vacuum
state, defined by
i |0 = 0,
corresponds to the trivial situation where all the lines in the graph are dotted. The independent state corresponding to a single solid line at = n is
|n = 1 (n , ) 2 (n , )|0.
(18)
This is an eternal solid line, extending indefinitely for both positive and negative , since
and hence the above state, are independent of by the equation of motion
the operators ,
following from (17). Several solid lines are represented by the state

1 (n ) 2 (n )|0.
n
212
We note that, a state with a double solid line, which has no graphical interpretation, vanishes by Fermi statistics. Having set up our fermionic system, we can express in Eq. (16)
in terms of fermions:
1
i (n , )i (n , ),
(n , ) =
(19)
2
i=1,2
and it is easy to check that is 1 if there is a solid line located at = n , and it is zero
if the line is dotted. The ground state expectation value of this composite field, 0 , will
play an important role in what follows. In any finite order of perturbation theory, where
the density of solid lines is essentially zero, 0 is zero. On the other hand, a non-zero 0
means that a portion of the worldsheet with finite area is occupied by the solid lines, which
we interpret as condensation of solid lines. It is also clear that this can only happen if large
(infinite) order graphs dominate the perturbation series. It will be shown later that, at least
in the mean field approximation, the condensation of solid lines leads to string formation.
One can think of 0 as an order parameter that distinguishes between the stringy phase and
the perturbative phase of the same theory.
So far, all the lines, whether solid or dotted, are eternal. We need an interaction term
in the action which will convert dotted lines into solid lines and vice versa. Remembering
that the transition between solid and dotted lines is accompanied by a factor of the coupling
constant g, we set

d 1 (n , ) 2 (n , ) + 2 (n , )1 (n , ) ,
S3 = g
(20)
n
and it is easy to show that this term in the action does the required job.
As we mentioned earlier, the fermions introduced in this paper are somewhat different
from those used in the earlier work [1,2,4]. However, it is not difficult to show that the two
are connected by the Bogoliubov transformation
1 1 ,
and, as a result, they are physically equivalent.
Finally, we will consider the continuum limit, with a 0. The dictionary for taking
this limit is

1
1
d,
i (n , ) i (, ),
a
a
n
1
1
(21)
(n , ) (, ) =
i (, )i (, ).
a
2
i
For the continuum fermions, we use the same notation as the discrete ones, but with n
replaced by . They satisfy anticommutation relations similar to the discretized version,
with the Kroenecker delta replaced by the Dirac delta function. The expressions for S2 and
S3 (Eqs. (17) and (20)) remain unchanged if one replaces the sum over n by integration
over . In particular, there is no explicit dependence on a. The state corresponding to a
solid line at = is now represented by
| = 1 ( , ) 2 ( , )|0,
213
and
(, )| = ( )| .
(22)
In what follows, although we will mostly work with the continuum fermions, from time to
time we will also need the worldsheet with discretized to have well defined expressions.
The total action is the sum of various pieces given by Eqs. (13), (17) and (20):
S = S0 + S1 + S2 + S3 .
(23)
As it stands, this action suffers from a serious problem: it is not well defined. We will
discuss this problem and present its solution in the next section.
We close with a discussion of the scaling properties of the fermions. Invariance of the
canonical algebra, or of the free action for the continuum fermions requires the transformation law
(,
) u(u,
u ),
(, ) u(u, u ),
(24)
and as result
(, ) u(u, u ),
(,
) u(u,
u ).
(25)
We also note that, from its definition Na = p+ , it follows that the cutoff parameter a must
scale as p+ :
a a/u.
(26)
It is now easy to check that, with the possible exception of the interaction term S3 , all the
terms of the fermionic action are scaling invariant. The interaction term violates scaling,
unless the coupling constant g is allowed to transform. Of course, the original field theory
coupling constant is a Lorentz scalar and cannot transform. However, the coupling constant
on the worldsheet is closely related to but not identical to the field theory coupling constant,
and it need not be a Lorentz scalar. In fact, Lorentz invariance requires that
g ug
under scaling. The simplest way to secure this is to use p+ , the only other physical parameter at our disposal and make the replacement
g
g
.
p+
(27)
p+ takes care of the scaling (see Eq. (6)), and the newly defined g is a true scalar.
5. The gauge fixed worldsheet action

Consider Eq. (16) for S1 : since vanishes everywhere except on the solid lines (the
c and z in the
boundary), the integrand is independent of the Lagrange multipliers y, b,
bulk of the worldsheet. As a result, the functional integration over the bosonic variables
214
is divergent2 and the corresponding integration over the fermionic variables vanishes. The
result is an ill defined expression of the type infinity times zero. This is similar to what
happens in gauge theories before gauge fixing. In fact, S1 is invariant under the gauge
transformation
y y + y
0,
b b + b 0 ,
z z + z
0,
c c + c 0 ,
(28)
where y0 , z0 , b 0 and c 0 are arbitrary functions and is defined by

= 1 ,
for discretized and
=
1
1
i i
=
a
2
(29)
for continuous , where is given by (19) in the first case and by (21) in the second case.
In either case, since
= 0,
(30)
invariance of S1 under the transformations (28) follows. It is this gauge invariance that is
responsible for the singular behaviour of the functional integrals over the Lagrange multipliers mentioned above.
The cure for this problem is to gauge fix the action, but only in the bulk of the worldsheet
(on the dotted lines), where = 0. We will also demand the gauge fixing term to be supersymmetric, since our philosophy is to keep intact the SUSY in the bulk and violate it only
on the boundaries. We therefore promote the Lagrange multipliers into a supersymmetric
multiplet
Y = z + 1 c + 2 b + 1 2 y,
(31)
and write the gauge fixing part of the action as

p
S4 =

d
1
Q
d 2 Y
2
p
=
d (y
z + b c ),
(32)
where is a gauge fixing parameter. This is the simplest gauge fixing term which is supersymmetric and which vanishes on the boundaries (solid lines) where = 0 and no gauge
fixing is needed. The previously singular functional integral over the bulk of the worldsheet
2 I thank Charles Thorn for stressing this point, although at the time I did not think it was important.
215
where = 0 is now equal to unity:

Dy Dz Db Dc exp(iS4 ) = 1.
(33)
=0
We note that the -dependent integration measures of the bosonic and fermionic functional
integrals cancel. We believe that the absence of gauge fixing on the continuum worldsheet
caused some of the problems encountered in the earlier work.
There is one more technical issue which we have to address here. There is some arbitrariness in the equation for (16) for S1 ; it could be replaced by the following more general
expression:
p
S1

d 1 y q + 2 b b + 3 c c + 4 z r .
(34)
Here, 1,2,3,4 are arbitrary constants. They can be eliminated by absorbing them into the
definition of the Lagrange multipliers. So long as one is dealing with the exact expression
for the action, the introduction of the s changes nothing, and one could just as well
set them all equal to one, as in Eq. (16). However, if an approximation scheme is used,
the results may well depend on these constants. These remarks also apply to the gauge
fixing parameter : the exact theory is independent of this constant but an approximate
calculation may introduce some dependence. In the next section, when we carry out a
mean field calculation, we will be able to see to what extend our results are sensitive to the
choice of these constants.
Finally, we have to make sure that the gauge fixing term S4 and also S1 are scale invariant. This is indeed the case for S4 if under scaling
y(, ) y(u, u ),
c (, ) u c (u, u ),
b(,
)
u b(u,
u ),
z(, ) uz(u, u ),
(35)
and Eq. (25) is taken into consideration. Now that we know the scaling properties of all the
fields, we can check S1 (Eq. (16)). The first three terms are indeed invariant as they stand,
but in the last term, we have to let
4 p + 4 .
6. The mean field approximation

In the last section, the final form of the action which is supersymmetric, gauge fixed and
scale invariant was worked out. Since the mean field approximation will be based on it, we
216
start by writing it down in full:

S=
n=4

Sn
n=0

1
d q 2 + b b + c c + r2 + 1 y q + 2 b b + 3 c c
2
0

g
+
z + b c ) + i + + (1 2 + 2 1 ) ,
+ 4 p z r + (y
(36)
p
p
where and are given by (21) and (29).

The mean field approximation was developed and applied to the worldsheet action in
the earlier work [24]. Here, we need the simplest version of it used in [2]. We notice that
Eq. (36) represents a vector model, which can be solved in the large D limit [16]. The
standard approach is to replace scalar products of various vector fields, such as the bilinear
terms in the expression for S0 (Eq. (13)), by their vacuum expectation values, which are
then treated as classical c-number fields. The functional integral over the remaining fields
is carried out exactly, and the resulting effective action is minimized with respect to the
classical fields. This method is known to be equivalent to a large N (in this case, D replaces
N ) saddle point calculation [16]. Instead of the approach sketched above, we find it much
simpler to replace the bilinears in the fermions, and by their classical expectation
values. In Appendix A, we show that this is completely equivalent to replacing the bilinears
in vector fields by their expectation values.
The problem is considerably simplified by starting with the setup where the total transverse momentum p carried by the whole graph is zero:
p
p=
d q = 0.
(37)
This configuration can always be reached by a suitable Lorentz transformation. It allows

us to impose the periodic boundary conditions

q( = 0, ) = q = p+ , .
(38)
This setup is translationally invariant in both the and the directions, and we shall
see that translation invariance will play an important role in simplifying the mean field
calculation.
We start by explicitly introducing the composite field by adding a new term S to the
action:
S S + S,
p
S =
d
0

1
i i ,
d
2
i
(39)
217
where is a Lagrange multiplier, and is given in terms of through Eq. (29). In the
large D limit, we can treat and as classical fields (see Appendix A). In other words,
we make the replacement
0 = ,
0 = ,
0 = ,
(40)
where denotes the expectation value in the ground state of the field in question. Translation invariance on the worldsheet means that both 0 and 0 are constants independent of
the coordinates and . With this simplification, it is possible to carry out explicitly all the
functional integration over the fields. We first consider the integration over the fermions;
that part of the action involving the fermions is given by
p
Sf =

g
0
d i + + (1 2 + 2 1 ) +
i i .
p
2
(41)
Instead of working with the action, we find it more convenient to diagonalize the corresponding Hamiltonian. In order to avoid singular expressions, we first discretize the
coordinate as in Eqs. (17) and (20). There is a complete decoupling of the dynamics in
the direction; as a result, the total Hamiltonian can be written as a sum of N mutually
commuting Hamiltonians:

H=
(42)
Hn ,
n
where,

0
Hn = g ( 1 2 + 2 1 ) +
,
i i
2
=n
(43)
and
g = g/p+ .
We observe that Hn acts on the two states
|0
corresponding to a dotted line at = n and
|n = 1 (n ) 2 (n )|0
corresponding to a solid line at the same position as a two by two matrix:
Hn |0 = g |n,
Hn |n = g |0 0 |n.
(44)
The eigenvalues are

1
1
Ef = 0
2 + 4g 2 ,
2
2 0
(45)
218
where the square root is defined to be positive. Since we are interested in the ground state,
we have to pick the lower energy Ef at each = n and add up to get the total fermionic
energy:
p+
E .
(46)
a f
Next, we will carry out the functional integrations over the vector fields in Eq. (36),
setting
Ef = NEf =
1
0 .
a
We first split the action into Sm , the matter action, and Sg , the ghost action:
0 ,
p
0 =
Sm =

1 2
2
+
d q + r + 0 1 y q + 4 p z r + 0 y z ,
2

1

d (b b + c c ) + 0 (2 b b + 3 c c) + 0 b c .
2
d
0
p
Sg =
(47)
The integrations over y, r, b and c are Gaussian and they can be done trivially. The (singular) Jacobians coming from integrations over the matter and the ghost fields cancel, with
the result,
p
Sm

1
1 1 4 p+ 02 2 2
q ,
d q 2 +
2
2
0

2 3 02
1
1
d b b c c +
cb .
2
2
0
d
0
+
p
Sg
d
0
(48)
We note that Sm is the (Minkowski) worldsheet action for a string, with the slope parameter , where

1 4 p+ 02 2
.
4 2 =
(49)
0
It may seem like string formation is almost automatic; however, the string picture breaks
down if the slope is zero, which happens for 0 = 0. The parameter 0 is therefore the
order parameter that distinguishes between the stringy and the perturbative phases of the
same field theory. Roughly speaking, since 0 measures the fraction of the worldsheet area
occupied by the solid lines, each graph in perturbation theory corresponds to 0 = 0. This
is because any finite number of solid lines, being one-dimensional, have vanishing area,
and as to be expected, perturbative field theory is then the zero slope limit of the string
theory. A non-zero slope requires 0 = 0, which means that the solid lines condense to
occupy a finite fraction of the area of the worldsheet. Therefore, 0 serves as an order
219
parameter which distinguishes between two phases: 0 = 0 in the perturbative field theory
phase and 0 = 0 in the stringy phase. In the next section, we will find that 0 = 0 in the
mean field approximation.
Since Sm,g have quadratic dependence on the fields, the functional integrals can be
carried out. Defining

e
,
exp(iSm,g ) = exp iSm,g
we have, after Euclidean rotation in ,

1
e
Sm
= D Tr ln 2 A2 2 ,
2

e
Sg = D Tr ln 2 B 2 2 ,
(50)
where
A=
1 4 p+ 02
,
0
B = i
2 3 02
.
0
Rewriting the Tr lns in terms of momenta k0 and k1 conjugate to and gives

(k n )2 + B 2
f i
e
1
S e = Sm
+ Sge = D
ln
,
dk0
n 2
2 2
4
(k
1 ) + A k0
n
(51)
with
k1n =
2n
.
p+
This is only a formal result, since the integration over k0 and summation over n lead to a
quadratic divergence, and before one can make sense of it, it must be regulated. In fact,
we are precisely interested in the coefficient of this quadratic divergence, which, after it
is regulated by a cutoff, will be the dominant term in the answer. Since the answer is
somewhat sensitive to the cutoff procedure, we will try to explain the motivation for the
regulator we use. First, we make use of a simplification: only the leading cutoff dependent
part of the answer is of interest and this dependence is sensitive only to the large k1n and
large k0 behaviour of the integrand in the above equation. Therefore, we can safely replace
the summation over n by the integration over the continuous variable k1 . By the same
token, we can set B = 0 in the integrand:

k12
Dp+
e
dk0 dk1 ln 2
S (f i )
(52)
.
8 2
k1 + A2 k02
In Eq. (51), the contributions of the matter and ghost sectors were combined into a
single log. This was intentional: we are going to regulate the combined contributions of
the two sectors, rather then regulate them separately. To see why, we observe that
(a) The combined term is less singular then each term treated separately. In fact, for a
fixed k0 , the integral over k1 converges, so that we need to regulate only the integral over k0 .
This is no accident: it can be traced back to the cancellation of the singular determinants
220
between the matter and the ghost sectors. Regulating each term separately could spoil this
cancellation;
(b) We saw in Section 2 that scale invariance on the worldsheet is necessary for Lorentz
invariance in the target space. Therefore, the regulated expression for S e should be scale
invariant. Under scaling, k0 and k1 transform as
k0 uk0 ,
k1 uk1 ,
and from Eq. (50), it is easy to check that A is scale invariant. It then follows that the integrand in Eq. (52) is scale invariant. This is of course related to the cancellation discussed
above, if we recall from Section 3 that the cancellation of determinants and scale invariance are intimately connected. Again, it was essential to combine the matter and the ghost
terms to arrive at a scale invariant integrand.
The scale invariance of the integrand in Eq. (52) is necessary but not sufficient for the
scale invariance of S e ; one also needs a regulator that respects scaling. We have seen that
only the k0 integration has to be regulated, which we regulate by introducing a second
cutoff (in addition to a) in the direction. Again the precise form of the regulator is not
important, so long as it respects scaling. As a simple example, we will consider a sharp
cutoff in k0 :
Dp+
S (f i )
8 2
+
/p

dk0
/p +

dk1 ln
k12

,
2
k12 + A2 k0
(53)
where is the cutoff parameter. The factor of p+ is inserted so that the limits of k0 integration are invariant under scaling. By a change of variables, we can evaluate this integral
as
D f i
Se = 2
(54)
A,
8
p+
where,

=

dk0
dk1 ln
k12 + k02
k12

.
(55)
We note that
(a) is a positive cutoff dependent constant. Since this is the only cutoff dependent
parameter in the result, we may as well replace the original cutoff parameter by ;
(b) The simple linear dependence of S e on A is going to be important in the following development. This dependence is a fairly robust result: it follows from the change of
variable
k1 Ak1 ,
independent of the details of how the integral over k0 is regulated;
221
(c) The dependence on p+ is required by scale invariance, again independent of the

form of the regulator.
The preceding discussion leads to the conclusion that S e has the unique form given by
Eq. (54), provided that we combine the matter and ghost determinants before regularizing
and we use a regulator that respects scale invariance.
Finally, we would like to comment on the factor of i in the definition of B in Eq. (50).
If the product 2 3 is real, B 2 will be negative, and from Eq. (51), S e will be complex.
This is, of course, a signal for instability. This does not concern us here: since we are
interested only in the leading cutoff dependence, we use Eq. (52), where the B 2 term is
absent. However, even for the non-leading part of Se , it is possible to avoid this problem
by taking the product 2 3 to be pure imaginary. We recall that the constants 2 and 3
appeared in front of the Lagrange multipliers that set the fields b and c equal to zero. If
these fields were bosonic, complex values for these constants would not be permissible, but
for fermionic fields, complex coefficients are allowed. Nevertheless, this may still show up
as an instability in some non-leading order in the large D limit.
7. String formation
We can now put together various terms derived in the last section and write down the
full effective action S eff :
f i
S,
S eff = S e (f i ) Ef + p+ 0 0 =
p+

D 1 4 p+ 02
(p+ )2
2
S = 2
(56)
0 + 02 + 4g 2 p+ 0 0 .
+
0
2a
8
Now let us go back to the question posed in Section 5: how does the effective action depend
on the arbitrary constants 1,2,3,4 that appear in Eq. (36) and the gauge fixing parameter
? The answer is that, at least within the present approximation, it does not depend on 2
and 3 at all. Also, the dependence on and 1 and 4 is rather trivial. Since the cutoff
parameter is arbitrary to begin with, by redefining it through
1 4
,
one can eliminate all the reference to these parameters.

Next we will search for the saddle point of the effective action in the variables 0 and 0 ,
or equivalently in the variables x and y, and see whether this corresponds to the minimum
value of the ground state energy. To start with, the expression for S can be simplified
considerably by a series of redefinitions:
1 4
,
8 2
1x
,
0 =
a
=
0 =
0 = 2
x
,
a
D
p+
y,
g =
g
D
= + g.
+
p
p
(57)
222
It is also convenient to define a scale invariant cutoff parameter a by

a = a/p+ .
In terms of these new variables, S is given by
S = D F (x, y),
a
where
F (x, y) =
x2
2xy + y +
1x
(58)

y 2 + g 2 ,
(59)
and y ranges from to +, whereas x is limited to the interval 0 x 1.

A number of features of this expression are worth noting:
(a) There is a factor of D multiplying the whole expression. Therefore, in the large D
limit, the dominant contribution comes from the saddle point;
(b) Every variable in this expression is scale invariant;
(c) The factor of D appearing in the definition of g is the standard large N factor [8]
needed to have the correct limit;
(d) Although we have so far introduced two independent cutoff parameters and a, or
appears in the expression for S.

equivalently, and a , only the combination /a
The integral to be evaluated in the large D limit is
+ 1

dy dx exp iS eff ,
Z=
eff
= F (x, y),
(60)
and the constant is defined by

=D
f i
.
a p+
First, let us consider the integration over y for a fixed value of x in the interval (0,1). This
integral can be evaluated exactly in terms of a Bessel function, but since we need only the
large D result, we will instead use the saddle point approximation. The saddle point ys is
at
2x 1
y F (x, y) = 0 ys =
g.
2 x x2
(61)
In this equation, both g and the square root are defined to be positive. The contour of
integration in the complex y plane can be distorted into the curve whose equation is

Re 2xy + y + y 2 + g 2 = 2g x x 2 .
(62)
223
Fig. 4. The integration contour in the complex y plane for g = 1.
This curve passes through the saddle point, and with the above choice of the branch of the
square root, as
Im(y) ,
on the curve,

Im F (x, y) +,
and therefore the integral

dy exp i F (x, y)
y= curve
is exponentially convergent. Using this contour of integration justifies the evaluation of the
integral over y by setting y = ys in the integrand. In Fig. 4, the curve defined by Eq. (62) in
the complex y plane is pictured for g = 1 and x = 1/2, when the saddle point is at ys = 0.
The branch cuts of the square root run from i g to +i and from i g to i and the
contour asymptotes the vertical lines Re(y) = g.
After the saddle point evaluation of the integral over y, we are left with the integral
over x:
1
Z

dx exp if (x) ,
(63)

x2
2g x x 2 .
1x
(64)
where
f (x) =
224
Fig. 5. The function f (x) for g = 20.
The function f (x) is pictured in Fig. 5 for g = 20. It has a single minimum in the interval
0 x 1 at x = xm , which satisfies
f (xm ) =
2
2xm xm
1 2xm
g
= 0.
2
2
(1 xm )
xm xm
(65)
At the minimum, f (x) is negative, corresponding to a negative minimum energy

E0 = D
f (xm ),
a p+
which is the energy of the ground state in this approximation. The situation is similar for
other positive values of g:
there is only a single minimum in the interval (0, 1), corresponding to a negative ground state energy.
We have just seen that the ground state corresponds to a value of x between 0 and 1. Let
us remember that
x = 0 0 = 0
corresponds to the trivial case of a worldsheet with all dotted lines. The opposite limit of
x = 1 0 = 0
corresponds to a worldsheet with only solid lines. A value of x in between these two
extremes implies an intermediate worldsheet texture: the solid and dotted lines each occupy
a finite fraction of the area of the worldsheet. Recalling our earlier discussion following
Eq. (49), we see that indeed a condensate of the solid lines has formed.
Let us now see whether a sensible string picture emerges. In particular, the slope parameter (Eq. (49))
=
2
1 4 p+ 02
1 4 x m
=
,
2 0
2(1 xm )a
(66)
225
is the only physical parameter to emerge from the mean field calculation. The theory should
be renormalized by requiring it to be a finite number independent of any cutoff. To see how
this happens, we first get rid of the irrelevant constants 1,4 and by suitably redefining
the cutoff parameter a :

2
xm
.
2(1 xm )a
(67)
In order to have a finite slope in the limit a 0, xm , and therefore, g should also go to
zero in the same limit. Solving (65) in the small g limit, we have
2/3
g
xm
(68)
,
2
and
g 4/3
.
(69)
27/3 a
Therefore, in the limit of the cutoff a tending to zero, the coupling constant g should be
fine tuned so that the ratio

g 4/3
a
stays constant. The theory is renormalized by trading the cutoff dependent coupling constant g for the cutoff independent physical parameter through Eq. (69).
There is one more comparison one can make between the parameters of the 3 theory
and the string theory: one could try to relate the string intercept to the field theory mass and
the coupling constant. If we identify the mass square of the lowest lying state on the string
trajectory with the ground state energy corresponding to S (Eq. (58)), the meanfield approximation gives a negative cutoff dependent answer. In this article, we will not consider
the question of renormalization of the ground state energy, which is the same as the renormalization of the intercept. There is, however, the following simple possibility. Instead of
starting with a 3 theory with zero bare mass, we could have started with a non-zero cutoff
dependent bare mass. Fine tuning this bare mass, it may be possible to end up with a finite
renormalized ground state energy. To carry out this program, however, our formalism has
to be extended to include a non-zero bare mass. This we leave to future research.
It may seem surprising that, starting with an unstable field theory, so far we have not
encountered any sign of instability on the string side. Of course, as mentioned above, we
have not calculated the renormalized intercept. In the end, upon calculating this intercept,
just as in the case of the bosonic string, some of the lowest lying states may turn out to be
tachyonic. Another possibility is that, in the leading order of the large D approximation,
the instability may not be visible.3 For example, it is easy to construct a simple quantum
mechanics problem with D degrees of freedom, where there is a metastable state which
decays by tunneling.4 It is usually the case that tunneling is suppressed in the leading large
3 I would like to thank Jeff Greensite and Charles Thorn for this suggestion.
4 I would like to acknowledge a helpful conversation with Eliezer Rabinovici on this subject.
226
D limit, and the metastable state becomes stable. One has to go beyond the leading order to
see signs of instability. It is possible that this is what happens in the model we are studying
here.
8. An additional string mode

In this section, we are going to compute a particular correction to the leading mean
field or large D result.5 We recall that, in this limit, the composite field can be replaced
by its ground state expection value 0 , but to go beyond the leading term, one has to
expand in powers of the fluctuations around the mean. Of course, an exact computation
of the full expansion is impractical; however, as in Section 6, we look for the dominant
contribution in the limit of large cutoff. In this case, this is a logarithmically divergent term
which dominates the rest of the terms, which are finite. To isolate this contribution, we
split the composite field into the constant mean value 0 (see Eq. (70)), and a fluctuating
part . Along with a power series expansion in , we treat as a slowly varying function,
and we expand it around a fixed point in powers of the coordinates and . This latter
expansion, which is sometimes called the derivative expansion, is very useful in isolating
divergent terms in the perturbation series and it is widely used in literature. The point is
that increasing order in this expansion goes with increasing number of derivatives on ,
and by dimensional arguments, this results in greater convergence. The divergent terms
therefore appear only in the lowest orders of the derivative expansion and they are easy to
identify. We shall see that the leading divergence is quadratic, but this is already included
in the calculation done in Section 6 with a constant 0 . The next leading divergence is
logarithmic, which is the contribution we are going to calculate. The rest of the terms in the
expansion are finite. The logarithmic term has a special physical significance: it provides a
kinetic energy term for in the action and so it promotes into a new propagating degree
of freedom. This phenomenon should be familiar from other two-dimensional models, such
as the CP(N) model [17] or the GrossNeveu model [18]. In contrast, the finite terms in
the expansion are non-local and they do not seem to have any special physical significance.
Our starting point is Eq. (48) for Sm , but with 0 replaced by , since we are considering
fluctuations of around the mean value 0 . We define

1 4 p + 2 2
(70)
= A2 (1 + ),

where A is defined by Eq. (50) and,

0 2 2
1,
=
(71)
02
is the fluctuating field. Doing the functional integral over q gives the following contribution
to the action:

i
S = D Tr ln 2 A2 (1 + ) .
(72)
2
5 See also Appendix B of Ref. [2].
227
Fig. 6. First order tadpole graph.
Fig. 7. Higher order tadpole graph.
We are going to examine in detail only terms up to second order in ; it will then be easy to
figure out the contribution of the higher order terms. We therefore expand S up to second
order:
S = S0 + S1 + S2 + ,

i
S0 = D Tr ln 2 A2 2 ,
2

1

i
S1 = DA2 Tr 2 A2 2 ,
2

1

1

i
S2 = DA4 Tr 2 A2 2 2 A2 2 .
4
(73)
e in Section 6 and it was already calculated there

The zeroth order term S0 was called Sm

(Eq. (50)). The first order term S1 is represented by the graph in Fig. 6, where the external
line in this graph carries zero momentum. Its contribution is given by

S1 = c1 d d (, ),
where c1 is a quadratically divergent constant. In addition, there other contributions from

higher order terms, represented by graphs of the form given in Fig. 7, which generate the
following series in S :

n
cn (, ) .
S d d
(74)
n=1
This can be thought of as a potential for , which, when minimized, determines the expectation value of . However, all this does is to shift the value of 0 , which we have already
228
Fig. 8. One loop contribution to the propagator.
calculated in the previous section. This ambiguity is due to the intrinsic arbitrariness in the
split made in Eq. (70): only the particular combination of 0 and that appears on the
right-hand side of (70) is well defined: shifting 0 and the expectation value of while
keeping the right-hand side of (70) constant will not change anything. We can resolve this
ambiguity by setting the expectation value of equal to zero. In that case, there is no shift
in the value of 0 , and we can drop the terms given in Eq. (74). We shall do so in what
follows.
Let us now focus on S2 , the term quadratic in , which is represented by the graph in
Fig. 8. In momentum space, we can set
+
i

4 p

0 , k1 )(k
d 2 k I (k0 , k1 )(k

S2 = DA
(75)
0 , k1 ),
4
2 3
where is the Fourier transform of , and

(4k02 (k0 )2 )2
I (k0 , k1 ) = d 2 k
.
((2k1 + k1 )2 A2 (2k0 + k0 )2 )((2k1 k1 )2 A2 (2k0 k0 )2 )
(76)
So far, this expression is exact, but now, we are going to carry out the derivative expansion
explained earlier, which coincides with the expansion in powers of the momentum k . The
zeroth order term in k contributes to the potential in , and we have explained above that
if one starts with the correct value of the expectation value of , this term is already taken
care of. The first order term in k vanishes, and the second order term has a logarithmic
divergence. Setting
I = I0 + I1 + I2 + ,
I1 is zero and I2 has the form
I2 = I2,0 (k0 )2 + I2,1 (k1 )2 ,
and, after Euclidean rotation in k0 ,

k 2 (3A2 k 2 k 2 + k 4 )
i
I2,0 =
d 2k 0 2 0 1 2 1 ,
2
(k1 + A2 k0 )4

k 4 (k 2 A2 k02 )
i
.
I2,1 =
d 2k 0 2 1
2
(k1 + A2 k02 )4
(77)
(78)
229
The integrals are elementary. After the change of variables by

k1 = r sin( ),
Ak0 = r cos( ),
the integrals are easily done, leaving a logarithmically divergent r integral:

i
I2,0 =
2A3
dr
,
r
i
I2,1 = 5
2A
dr
.
r
(79)
We have introduced both an infrared cutoff and an ultraviolet cutoff to regulate the
divergent r integral. Putting this back into the equation for S2 (Eq. (75)), and rewriting it
in the position space, we have

DR
S2 =
(80)
d d A2 ( )2 ( )2 ,
32A
where we have defined

R=
dr
.
r
We close this section with a couple of comments on this result:

(a) The contribution of the mode to the action is exactly the same as the contribution
of q to Sm (Eq. (48)), apart from an overall cutoff dependent multiplicative constant. This
constant can be eliminated by rescaling (wave function renormalization). The important
point is that it is positive; otherwise, this term would have the wrong (ghostlike) signature;
(b) There is therefore an additional string mode represented by , with the same slope
= A/2
as the other modes. This changes the dimension of the target space from D to D + 1, and
therefore it can regarded as a non-leading contribution in the large D limit;
(c) If we continued the derivative expansion of S2 beyond second order, we would get
terms with higher derivatives of with respect to and . By simple power counting,
these terms would be finite and therefore they would not be of interest to us;
(d) Let us now consider terms higher then second order in in the expansion of S .
The logarithmically divergent contribution still comes from second order in the derivative
expansion. Consider the graph of Fig. 2, where two external lines carry finite momenta and
the rest of the external carry zero momentum. A simple generalization of the calculation
given above in the case of the second order term in shows that this term must be of the
form

S = d d h() ( )2 A2 ( )2 ,
(81)
and h() is a function whose power series expansion in starts with the positive constant
that appears in Eq. (80). Redefining the field by
g(),
230
and choosing the function g so that it satisfies

2
h f () g () = 1,
one can get replace h by one. This shows that a simple field redefinition gets us back to the
second order result given by Eq. (80).
To complete our discussion, we should also consider the contribution coming from Sg in
Eq. (50). This contribution has a quite different structure compared to the one coming from
the matter sector. It has a linear cutoff dependence, and it depends only on derivatives of
and not derivatives. We remind the reader of the motivation behind integrating over the
matter fields in Sm : the integration produced a kinetic energy term for the originally nonpropagating field . Since integrating over the ghost fields produces nothing of comparable
interest, it is probably best to leave Sg as it is.
9. Conclusions
This article is a direct follow up of Ref. [2]. The goal is to give a systematic treatment which resolves various problems encountered in [2], while still staying within the
framework of the simple meanfield approximation developed in that reference. The crucial
improvement over the treatment given in [2] is the fixing of an accidental gauge invariance.
In addition, there are various other technical improvements: supersymmetry is introduced
on the worldsheet to keep better track of matter-ghost cancellation, and a better treatment
of the singular determinants is presented. Also, following Ref. [4], we show how to impose
partial Lorentz invariance.
The meanfield method used here and in [2] is sufficiently simple to allow the carrying
out of a fairly complete treatment. In the leading order of this approximation, a condensation of Feynman graphs takes place, which means that large order Feynman graphs
dominate the perturbation series. As a consequence, with a suitable tuning of the coupling
constant, a string with finite slope is formed. We also show that a new dynamical degree
of freedom emerges, extending the transverse dimension of the string from D to D + 1.
Although the string appears to be stable in the leading order, we argue that the fundamental
instability of the underlying field theory probably shows up in the non-leading orders of
the approximation.
The question whether string formation takes place in a given field theory is fundamentally important but difficult to answer in practice. The meanfield approximation used in
this article, free from its earlier shortcomings, is simple yet powerful enough to answer
this question in the case of the 3 theory. The possibility of applying this method to more
realistic theories appears more appears within reach. Another valuable line of future research would be to try to improve over the meanfield approximation.
231
Acknowledgements
This work was supported in part by the Director, Office of Science, Office of High
Energy and Nuclear Physics, of the US Department of Energy under Contract DE-AC0376SF00098, and in part by the National Science Foundation under Grant PHY-0098840.
Appendix A
In this appendix, we will discuss an alternative, and more conventional way of doing the
mean field approximation, and we will show that it is completely equivalent to the approach
used in this article. Instead of (Eq. (39)), we introduce a different set of composite fields:

S = d d 1 (y q 1 ) + 2 (b b 2 ) + 3 (c c 3 )

+ 4 (z r 4 ) + 5 (y z 5 ) + 6 (b c 6 ) .
(A.1)
In the large D limit, the fields i and i become classical and they can be replaced by their
ground state expectation values:
i i ,
i i .
The justification for this is well known [16]: the composite fields i are each sum of D
terms, and consequently, they grow like D as D becomes large. On the
other hand, compared to this, the quantum fluctuations are suppressed by a factor of 1/ D. As a result,
in leading order large D these fields become classical. In contrast, there is no comparable direct argument for why should become classical in this limit. Therefore, from the
perspective of large N (large D) physics, the approach sketched in this appendix is better
motivated than the approach developed in the main body of the article. The disadvantage
is that, at least initially, many more auxilliary fields have to be introduced. We will show
presently that the expectation values of these extra fields can easily be expressed in terms
0 and 0 introduced earlier. To do this, we first write Sf , the fermionic part of the action,
in terms of the fields introduced in Eq. (A.1):

g
Sf = d d i + + ( 1 2 + 2 1 )
p

1
+ 1 + 2 + 3 + p+ 4 5 6
(A.2)
i i + (5 + 6 ) ,
2
a
i
where, to keep things simple, we have set all the s equal to one. Comparing this with the
Sf given by Eq. (41) leads to the identification
0 = 1 + 2 + 3 + p+ 4 5 6 ,
(A.3)
so that the two expressions for Sf , apart from an additive term, agree. We now consider all
the -dependent terms in
S + Sf
232
and write down the equations of motion by varying the s, subject, however, to the constraint that 0 (Eq. (A.3)) is held fixed. These equations give

4
1
1 .
5 = 6 =
1 = 2 = 3 = + ,
(A.4)
p
a
Consequently, there is only one independent , say 1 . With the further identification
1
1 = 0 ,
(A.5)
a
we recover the action of Eq. (36) with s set equal to one, establishing the equivalence of
the two approaches.
1 = 0 ,
References
[1]
[2]
[3]
[4]
[5]
[6]
[7]
[8]
[9]
[10]
[11]
[12]
[13]
[14]
[15]
[16]
K. Bardakci, C.B. Thorn, Nucl. Phys. B 626 (2002) 287, hep-th/0110301.

K. Bardakci, Nucl. Phys. B 667 (2004) 354, hep-th/0308197.
C.B. Thorn, Nucl. Phys. B 637 (2002) 272, hep-th/0203167.
S. Gudmundsson, C.B. Thorn, T.A. Tran, Nucl. Phys. B 649 (2003) 3, hep-th/0209102.
C.B. Thorn, T.A. Tran, Nucl. Phys. B 677 (2004) 289, hep-th/0307203.
G. t Hooft, Nucl. Phys. B 72 (1974) 461.
A. Clark, A. Karsch, P. Kovtun, D. Yamada, Phys. Rev. D 68 (2003) 066011.
R.M. Koch, A. Jevicki, J.P. Rodrigues, Phys. Rev. D 68 (2003) 065012, hep-th/0305042.
O. Aharony, J. Marsano, S. Minwalla, K. Papadodimas, M.V. Van Raamsdonk, hep-th/0310285.
R. Gopakumar, hep-th/0402063.
J.M. Maldacena, Adv. Theor. Math. Phys. 2 (1998) 281, hep-th/9711200.
O. Aharony, S.S. Gubser, J. Maldacena, H. Ooguri, Y. Oz, Phys. Rep. 323 (2000) 183, hep-th/9905111.
P. Goddard, J. Goldstone, C. Rebbi, C.B. Thorn, Nucl. Phys. B 56 (1973) 109.
For a recent review of the large N method, see M. Moshe, J. Zinn-Justin, Phys. Rep. 385 (2003) 69, hepth/0306133.
[17] A. DAdda, M. Luscher, P. Di Vecchia, Nucl. Phys. B 146 (1978) 63.
[18] D. Gross, A. Neveu, Phys. Rev. D 10 (1974) 3235.

Higher charge calorons with non-trivial holonomy

Falk Bruckmann, Dniel Ngrdi, Pierre van Baal
Instituut-Lorentz for Theoretical Physics, University of Leiden,
P.O. Box 9506, NL-2300 RA Leiden, The Netherlands
Received 28 April 2004; accepted 27 July 2004
Abstract
The full ADHM-Nahm formalism is employed to find exact higher charge caloron solutions with
non-trivial holonomy, extended beyond the axially symmetric solutions found earlier. Particularly
interesting is the case where the constituent monopoles, that make up these solutions, are not necessarily well-separated. This is worked out in detail for charge 2. We resolve the structure of the
extended core, which was previously localized only through the singularity structure of the zeromode density in the far field limit. We also show that this singularity structure agrees exactly with
the Abelian charge distribution as seen through the Abelian component of the gauge field. As a
by-product zero-mode densities for charge 2 magnetic monopoles are found.
PACS: 11.10.Wx; 12.38.Lg; 14.80.Hv
1. Introduction
Calorons are instantons at finite temperature. For a long time the influence of a background Polyakov loop on the properties of these topological excitations has been neglected.
Solutions were constructed long ago [1] and were studied in detail in the semi-classical approximation [2]. In all these studies the Polyakov loop at spatial infinity (also called the
holonomy) was trivial, i.e., an element of the center of the gauge group. That the influence
of the background Polyakov loop on the topological excitations can be dramatic is particE-mail address: vanbaal@lorentz.leidenuniv.nl (P. van Baal).
doi:10.1016/j.nuclphysb.2004.07.038
234
F. Bruckmann et al. / Nuclear Physics B 698 (2004) 233254
ularly clear in the confined phase, where on average its trace vanishes. Caloron solutions
in such backgrounds were constructed only relatively recently [3,4] and can be seen as
composed of massive monopole constituents with their magnetic charges adding to zero.
It was observed that the one-loop correction to the action for configurations with a nontrivial asymptotic value of the Polyakov loop gives rise to an infinite action barrier, which
were therefore considered irrelevant [2]. However, the infinity simply arises due to the
integration over the finite energy density induced by the perturbative fluctuations in the
background of a non-trivial Polyakov loop [5]. The proper setting would therefore rather
be to calculate the non-perturbative contribution of calorons (with a given asymptotic value
of the Polyakov loop) to this energy density, as was first successfully implemented in supersymmetric theories [6], where the perturbative contribution vanishes. It has a minimum
where the trace of the Polyakov loop vanishes, i.e., at maximal non-trivial holonomy.
In a recent study at high temperatures, where one presumably can trust the semiclassical approximation, the non-perturbative contribution of these monopole constituents
(also called dyons) was computed [7]. When added to the perturbative contribution [5] with
its minima at center elements, a local minimum develops where the trace of the Polyakov
loop vanishes, deepening further for decreasing temperature. This gives support for a phase
in which the center symmetry, broken in the high temperature phase, is restored and provides an indication that the monopole constituents are the relevant degrees of freedom for
the confined phase.
Also lattice studies, both using cooling [8] and chiral fermion zero-modes [9] as filters, have now confirmed that monopole constituents do dynamically occur in the confined
phase. A charge 1 caloron is seen for SU(n) to consist of n constituent monopoles. In the
deconfined phase, due to the fact that the average Polyakov loop becomes a center element,
the caloron returns to the form known as the HarringtonShepard solution [1]. The latter
can also be interpreted as consisting of constituent monopoles, however, with n 1 of them
being massless.
To be precise, for self-dual configurations in the background of non-trivial holonomy
the masses of constituent monopoles are given by 8 2 j /, with j j +1 j . The i
are related to the eigenvalues of the Polyakov loop at spatial infinity,

P = lim Pexp
(1)
A0 (t, x) dt = g exp 2i diag(1 , 2 , . . . , n ) g
x
(this expression assumes the periodic gauge A (t, x ) = A (t + , x)) where g is the gauge
rotation used to diagonalize P and is the period in the imaginary time direction, related
to be ordered on the circle such
to the inverse temperature. The eigenvalues exp(2ij ) are
that 1 2 n n+1 , with n+j 1 + j and ni=1 i = 0, which guarantees
that the masses add up to 8 2 /, the instanton action per unit (imaginary) time. At higher
topological charge k, the parameter space of widely separated constituent monopoles is
described by kn monopole constituents, k of each of the n types of Abelian charges (with
overall charge neutrality).
We established in an earlier paper [10] that well-separated constituents act as point
sources for the so-called far field (that is far removed from any of the cores). When constituents of opposite charge (n constituents of different type) come together, the action
235
Fig. 1. In the middle is shown the action density in the plane of the constituents at t = 0 for an SU(2) charge 2
caloron with tr P = 0, in the regime where constituents are not well separated. On a scale enhanced by a factor
10 2 are shown the densitities for the two zero-modes, using either periodic (left) or anti-periodic (right) boundary conditions in the time direction. This solution is for the so-called crossed configuration with k = 0.997 and
D = 8.753, see Section 4 for more details.
density no longer deviates significantly from that of a standard instanton. Its scale parameter is related to the constituent separation d through 2 / d. Yet, the gauge field
is vastly different, as is seen from the fact that within the confines of the peak there are n
locations where two of the eigenvalues of the Polyakov loop coincide [11,12], thus in some
sense varying over the maximal range available (e.g., for SU(2) from 12 to 12 ), whereas
for trivial holonomy only one such location is found.
On the other hand, when constituents of equal charge come together typically an extended core structure is found. This was deduced, in particular for the case of charge 2
calorons, from our ability to analytically determine the zero-mode density (summed over
the two zero-modes implied by the index theorem) in the far field limit, neglecting exponential contributions.1 In this limit it forms a singular distribution on a disc bounded by
an ellipse, but approaches two delta functions for well-separated like-charge constituents.
This zero-mode density only sees constituents of a given charge, depending on the boundary condition for the fermions in the imaginary time direction, which can be chosen to be a
U(1) phase (containing the physically relevant choice of anti-periodic boundary conditions
for thermal field theory). We will show for SU(2) that their difference for periodic and
anti-periodic boundary conditions coincides exactly with the (Abelian) charge distribution
extracted from the gauge field in the same limit, making contact with an old result due to
Hurtubise [13] for the asymptotic behaviour of the monopole Higgs field.
1 This is in some sense equivalent to the high temperature limit, with constituent masses given by 8 2 /,
m
such that the range of the exponential contributions shrinks inversely proportional with the temperature.
236
We found two particular parametrizations within the SU(2) charge 2 moduli space that
exhibited these extended charge distributions. The first includes as a limit arbitrary charge
2 monopoles. The second of these parametrizations contains as a limiting case the axially
symmetric configurations constructed for arbitrary charge in Ref. [14]. Deforming away
from the axial configuration the two discs overlap. Describing the intricate behaviour for
the non-Abelian core of these configurations in this region of the parameter space requires
one to find exact solutions, which are presented here. We rely on early work of Nahm
[15] and Panagopoulos [16] for charge 2 magnetic monopoles, which is simplified to some
extent by our formalism that uses the relation between the Fourier transformation of the
ADHM construction (as relevant for the finite temperature case) and the Nahm transformation, a crucial ingredient for our success to find explicit solutions [3]. Fig. 1 gives a
particular example for the action and zero-mode densities of a charge 2 caloron solution.
The two-dimensional zero-mode basis is chosen such that each zero-mode only localizes
on one of the constituents of a given charge, showing both the zero-modes with periodic
and anti-periodic boundary conditions in the imaginary time direction.
This paper is organized as follows. In Section 2 we will outline the construction, introduce the Greens function that is computed through the analogy of an impurity scattering
problem, and summarize the various limits that can be formulated before explicitly solving
for the Greens function. In Section 3 we present the method that allows one to find the
exact solution for the Greens function, first for the general case and then applied in more
detail to that of topological charge 2 calorons. Readers only interested in the results could
skip Sections 2.2 and 3. In Section 4 we discuss the two classes of configurations in the
moduli space of the charge 2 calorons and provide plots of the various quantities to illustrate the properties of the exact results. In Section 5 we discuss the relation between the
algebraic tail of the gauge field and the zero-mode density. We end with some discussions.
An Appendix A presents a new result for the limiting behaviour of the action density.
2. Outline of the construction

There are two steps in the construction of charge k caloron solutions. The first step
involves finding a U (k) gauge field A (z), which satisfies the self-duality equation, i.e.,
the Nahm equation [15], on a circle parametrized by z, with z introduced through replacing
the original SU(n) gauge field A (x) by A (x) 2iz0 1n . Although not affecting the
field strength, this changes the holonomy to exp(2iz)P , revealing that z has period
1 . The index theorem guarantees the existence of k zero-modes (x; z) which satisfy
the Dirac equation, or in the two-component Weyl form

D z (x; z) Dz (x; z) + A (x) 2iz0 1n (x; z) = 0
(2)
with = = (12 , i ) (i are the usual Pauli matrices). We may remove z from
the gauge field A (x) by transforming the zero-mode to z (x) exp(2itz) (x; z),
which is at the expense of making the zero-mode only periodic up to a phase factor,
z (t + , x) = exp(2iz) z (t, x ). In a similar way we could introduce z through
(x; z, z) exp(2iz x) (x; z), which replaces in Eq. (2) A (x) by A (x) 2iz1n ,
237
where z0 z. Assuming the k zero-modes (a)(x; z, z) to be orthonormal one has

A ab
(z) =

(a)(x; z, z)
(b) (x; z, z) d 4x,

z
(3)
or equivalently (demonstrating as well that A does not depend on z)

A ab
0 (z) =

(a)(x; z)
A ab
k (z) = 2i
(b)
(x; z) d 4x,
z
(a) (x; z)xk (b) (x; z) d 4x.
(4)
We have shown how A (z) can alternatively be related to a Fourier transformation of the
ADHM construction of instantons [17], that periodically repeat (up to a gauge rotation with
P ) in the imaginary time direction, so as to turn an infinite charge instanton in R 4 to one
of finite charge and finite temperature. The derivation of this relation will not be repeated
here, see Refs. [3,14] for the details.
The connection to the ADHM construction has been useful to simplify the second step
in the construction of the caloron solutions, namely how to reconstruct the original gauge
field when given a solution to the Nahm equation [15] (which is equivalent to the quadratic
ADHM constraint),

1

d
j
Aj (z) + A 0 (z), A j (z) + j k A k (z), A (z) = 2i
(z m )m .
dz
2
m
(5)
For convenience of notation we will henceforth use the classical scale invariance to set
= 1. The singularities in the Nahm equation appear precisely for those values where
e2iz P has one of its eigenvalues equal to 1, at z = i . This is where some fermion
field components become massless, i.e., the zero-mode becomes delocalized, whereas for
generic values of z it is exponentially localized, which has turned out to be a useful tool to
pinpoint the constituent monopoles.
One could apply the Nahm transformation again, introducing A (z) 2ix 1k (with
x0 t) and find the n chiral fermion zero-modes in the background of this gauge field.
The construction is somewhat complicated due to the presence of the singularities, whose
structure is determined from the matrices m which appear in the Nahm equation. Not all
m are independent, e.g., integrating and tracing the Nahm equation yields the conditions,

n
Further constraints are implied [14] by the fact that one may introduce
m = 0.
m=1 Tr
(as is most easily seen in relation to the ADHM construction) k two-component spinors a
in the n representation of SU(n), such that
ab
ab
m
,
2a Pm b = 0 Sm
where Pm are projections defined through P =

Eq. (8)).
(6)
n
m=1 e
2im P
m
(Sm will appear in
238
2.1. The Greens function

However, with reference to the ADHM construction, there is great benefit in first finding
the solution for the Greens function, fx (z, z ) g (z)fx (z, z )g(z
), where

d2
2 + V (z; x) fx (z, z ) = 4 2 1k (z z ),
(7)
dz
with
V (z; x) 4 2 R 2 (z; x ) + 2
(z m )Sm ,
Sm g(
m )Sm g (m ),
Rj (z; x) xj (2i)1 g(z)

A j (z)g (z),
(8)
and Sm playing the role of impurities. This is formulated in the gauge where first we
transform A 0 (z) to a constant (diagonal) matrix, 2i0 , as is always possible in one dimension, and then use

g(z)
exp 2i(0 x0 1k )z ,
(9)
Tr 0 = 0
(when Tr 0 = 0 it is absorbed in a shift of x0 ) to transform A 0 2ix01k to zero. This is
at the expense of introducing periodicity up to a gauge transformation; although fx (z, z )
is periodic in z and z with period 1 (for = 1), fx (z, z ) no longer is.2
Given a solution for the Greens function, there are straightforward expressions for the
gauge field [14] (only involving the Greens function evaluated at the impurity locations)
and the fermion zero-modes [10,18]. For the zero-mode density this gives
z(a) (x) z(b) (x) = (2)2 2 fxab (z, z).
In this paper we will only have need for the Greens function at
be expressed as

fx (z, z) = 4 2 (12k Fz )1 12 ,
z+1
Fz g (1) Pexp
0
V (w; x )
1k
0
(10)
z
= z, which formally can

dw,
(11)
where the (1, 2) component on the right-hand side of the first identity is with respect to the
2 2 block matrix structure. In particular this leads to a compact expression for the action
density [3,14]

1
1
2
S(x) tr F
(x) = 2 2 log det ieix0 (12k Fz ) ,
2
2
which can be shown to be independent of the choice of z.
(12)
2 It is in this respect interesting to note that g(1)
plays the role of the holonomy associated to the dual Nahm
gauge field A (z).
239
The formal expression for Fz can be made explicit by a decomposition into the impurity contributions Tm at z = m and the propagation Hm Wm (m+1 , m ) between
m and m+1 . For z (m , m+1 ) this gives
Fz = Wm (z, m )Tm Hm1 T2 H1 T1 g (1)Hn Tn Hn1 Tm+1 Hm Wm (m , z),
(13)
with

0
1k
,
2Sm 1k
+
f (z)

Wm (z, z ) d m +
dz fm (z)
Tm
fm (z)
d
dz fm (z)
fm+ (z )
d +
dz fm (z )
fm (z )
d
dz fm (z )
(14)
The columns of the two k k matrices fm (z), defined for z (m , m+1 ), form the 2k
solutions of the homogeneous Greens function equation,

2
d
2 2
R
4
(z;
x

)
v(z)
= 0,
(15)
dz2
of which those in fm+ (z) are exponentially rising and those in fm (z) are exponentially
for |
falling. This implies [14] fm (z) exp(2|
x |(z m )1k )Cm
x | , in which
f ( ) can be arbitrary non-singular (to ensure a complete set of solutions) maCm
m
m
= 1 , but here we find it convenient to leave this choice
trices. In Ref. [10] we put Cm
k
open. With the impurity scattering problem solved, constructing the exact solutions of
the homogeneous Greens function equation is the last step in finding analytic expressions
for the higher charge calorons.
2.2. Limiting cases
Nevertheless, approximate solutions can be derived, either assuming x is far removed
from any core such that the gauge field has become Abelian (which we called the far field
limit, denoted by a subscript ff), or assuming x and the constituents of type m (belonging
to the mth interval) are well separated from all others, but not necessarily from each other
(which was called the zero-mode limit, denoted by a subscript zm). This is because we
found [10] that for z (m , m+1 ) the k zero-modes z(a) (x) only see the constituents
of type m. For m z m+1 we have

fxzm (z, z) = fm (z)fm (m+1 )1 fm+ (z)fm+ (m+1 )1 Zm+1

1
+ +
fm (m )fm (m+1 )1 Zm
fm (m )fm+ (m+1 )1 Zm+1

1
+ +
fm (m )fm (z)1 Zm
fm (m )fm+ (z)1 Rm
(z),
(16)
up to exponential corrections in the distance to the constituents of type m = m, with
1
Zm
1k 2m
Rm1 (m ),

1 +
Rm (z) Rm (z) + Rm
(z) ,
2
+
1
Zm
1k 2m
Rm (m ),
+
m Rm
(m ) + Rm1
(m ) + Sm
(17)
240
and
(z)
2Rm

d
fm (z) fm (z)1 .
dz
(18)
+ and Z
In the zero-mode limit Zm
m+1 approach 1k up to algebraic corrections. Neglecting
these contributions as well, e.g., by sending the constituents of type m = m to infinity,
leads to the so-called monopole limit (denoted by a subscript mon), further simplifying
the expression for the Greens function to
1
(z),
fxmon (z, z) = U (z, m+1 )Um1 (m , m+1 )Um (m , z)Rm
(19)
Um (z, z ) fm+ (z)fm+ (z )1 fm (z)fm (z )1 .
(20)
with3
In turn, from Eq. (16) one may derive the far field limit, giving up to exponential corrections in the distance to all constituents
1
,
fxff (m , m ) = 2m
(21)
as is relevant for the expression of the gauge field in this limit [14]. For the zero-mode
density (Eq. (10)) we may use for m < z < m+1 (z strictly different from m , m+1 )
1
(z).
fxff (z, z) = Rm
(22)
1 = R
The discontinuity in this limit, Rm (m ) = 2m
m1 (m ), arises due to the
zero-mode developing a massless component when z approaches m . It might seem that
Eq. (22), combined with Eq. (10), is inconsistent with an exponential decay. However, it
turns out that [10]
1
x ) (4)1 Tr Rm
(z)
Vm (
(23)
is independent of z in the mth interval and harmonic everywhere (hence giving vanishing
zero-mode density) except for some singularities in the cores of the constituents of type
m. It is this feature, and our ability to compute Vm (
x ) exactly for SU(2) charge 2 calorons,
that allowed us to make statements about the localization of the cores, without solving the
Greens function exactly.
In Appendix A we derive the following new result for the monopole limit of the action
density (Eq. (12)). In the limit where x and the constituent locations of type m are well
separated from all other constituents, for which both the action and zero-mode densities
are static, we find (with Um defined in Eq. (20))

1
1
x ) = i2 j2 log det Um (m+1 , m )Rm
(m ) ,
S mon (
(24)
2
up to algebraic corrections. This is a direct generalization for the action density of a single
BPS monopole, S(
x ) = 12 i2 j2 log[sinh(2m |
x |)/|
x |] (located at the origin and with
3 Note that U (z, z ) satisfies the Greens function equation with boundary conditions U (z, z) = 0 and
m
m
d U (z, z ) = d U (z, z ) = 2 R (z), for z z.
m
m
m

dz
dz
241
mass 8 2 m ). We emphasize that this result can be used irrespective of the distance between the constituents of the same type m. Eq. (24) therefore gives in terms of fm (z)
a closed form expression for the multi-monopole energy density. The same holds, when
using Eq. (19), for the zero-mode density (Eq. (10)). For the caloron it involves taking
the limit where all constituents of type m = m are sent to infinity, which is why this is
called the monopole limit. We recall, that in the far field one should use [14] (see also
Appendix A)
n

1
1
x ) = i2 j2
log det fm+ (m+1 )fm+ (m )1 Rm
(m )m .
S ff (
2
(25)
m=1
3. Exact results
Finding the exact homogeneous solutions of the Greens function equation, Eq. (15),
closely follows Nahms method [15] to construct the dual chiral zero-modes. The main
advantage of our approach is that we need not worry about boundary conditions, as this is
solved by the impurity scattering formalism [14]. In the following we restrict ourselves
to a given interval z (m , m+1 ) and work in the gauge where A 0 (z) 2ix0 1k = 0.
Using that A (z) is self-dual, a consequence of the Nahm equation, one easily shows
d2
2 2
that D x D x = dz
), such that it is natural to consider the equation
2 + 4 R (z; x

d
x) (z)
2 R(z;
= 0.
D x (z)
(26)
= D x (z)
=
dz
It follows that (z)

would be a homogeneous solution of the Greens function equation,
albeit that (z)

is a spinor (with a chirality opposite to that for the zero-modes involved
in the Nahm transformation, cf. Eq. (2)). We follow Nahm [15] and use the ansatz (z)
=
x ) )|s v(z),
where v(z)
is a k-dimensional complex vector, u (

x ) is a unit
(12 + u(
vector that does not depend on z and |s is an arbitrary normalized constant spinor (as long
x ) ). It then follows that v(z)

= s|(12 u (
x ) )(z)
as it is not annihilated by 12 + u(
satisfies Eq. (15).
The unit vector u(
x ) is found from a complex vector y(x) which squares to 0,
y(
x ) y(
x ) = 0, implying its real and imaginary parts are perpendicular and of equal length
such that (for ease of notation the x dependence of y and u will
(= 0 as long as y(
x ) = 0),
henceforth be left implicit)

u = i y y y y
(27)
is well defined and u y = i y, i.e., Re(
y ), Im(
y ) and u form an orthogonal set of vectors.
Using the ansatz for (z)

and introducing
x ),
Y (z)
y R(z;
x),
U (z) 2 u R(z;
(28)
leads to the equations

Y (z)v(z)
= 0,
d
v(z)
= U (z)v(z)
dz
for which the first one can only have a solution provided det Y (z) = 0.
(29)
242
It is the great beauty of Nahms formalism that det Y (z) is a conserved quantity. That is,
= 12 j k [A k (z), A (z)] (the Nahm equation (5) restricted to an interval and in
d
det Y (z) = 0 for any choice of x and y on a null-cone
the gauge where A 0 = 0) implies dz
2
3
in C (or rather in CP since we may rescale y with a non-zero complex factor without
changing the equations). The conserved quantities are generated by the symmetric traceless
monomials, Mi1 i2 ...i , built from Tr(A i1 (z)A i2 (z) A i (z)) with arbitrary, as one readily
verifies. For example, Tr A i (z) is constant and defines the center of mass. A natural way
to project on the traceless symmetric monomials is precisely through introducing y CP 2
on a null-cone, forming yi1 yi2 yi Mi1 i2 ...i = yi1 yi2 yi Tr(A i1 (z)A i2 (z) A i (z)). It
is interesting to note that xi1 xi2 xi Mi1 i2 ...i |
x | is always a spherical harmonic of or1 (z)) that it is
der , used in Ref. [10] to show through the multipole expansion of Tr(Rm
conserved and harmonic, except for singularities in the core of the constituents.
Once it is established that det Y (z) is independent of z for any choice of x , we can look
for its zeros. Using the null-cone parametrization y = ( 12 (1 2 ), 2i (1 + 2 ), ), and the
fact that the matrix Y (z) is k-dimensional, we obtain a polynomial equation in of order
2k and hence there are for generic x exactly 2k solutions. It is useful to note that these
solutions come in complex conjugate pairs, where the symmetry y y implies u
u
and 1/ (giving y up to a multiple, equivalent to y in CP 2 ).
Given a particular zero y , we may conveniently write a vector in the kernel of Y (z)
as [19] va (z) = (z)(adj

Y (z))ac for a fixed choice of c, where adj Y (z) is the matrix
d
Y (z) =
formed by the minors of Y (z). The Nahm equation is easily seen to imply dz
d
4
[U (z), Y (z)] for any choice of y on the null-cone. From this one derives that dz adj Y (z) =
[U (z), adj Y (z)]. Substituting va (z) = (z)(adj

Y (z))ac into Eq. (29) gives
d
dz Aj (z)

d (z)
adj Y (z) ac = (z)

adj Y (z)U (z) ac .
dz
(30)
A j (z)g (z) 2uj xj , we get

Using the fact that U (z) = iuj g(z)
(z)
2z
u x )
= exp(
,
(z)
(adj Y (z))ac
i{adj Y (z), uj g(z)
A j (z)g (z)}ac
d (z)
=
,
dz
2(adj Y (z))ac
(31)
where the equation for (z)
is the same for any value of a (it may depend on the value
of c). This is useful for studying the asymptotic behaviour of the solution. For large |
x |,
det Y (z) = 0 implies that y x 0, such that (cf. Eq. (27)) u
x /|
x |. We may use the
symmetry u
u to guarantee that there are k zeros y(b) with the sign of u(b) x negative,
leading to solutions that rise as exp(2z|
x |). It then follows that the k zeros y(b+k) = y(b)
4 Assume first that y is such that det Y = 0, in which case adj Y = Y 1 det Y and therefore d adj Y =
dz
d det Y = [U , adj Y ] + Y 1 Tr(Y 1 d Y ) = [U , adj Y ]. Observe that adj Y is anY 1 [U , Y ]Y 1 det Y + Y 1 dz
dz
alytic in y , such that the result is valid also when y leads to det Y = 0.
243
give u(b+k) =
u(b) , leading to the solutions that decay as exp(2z|
x |) for large |
x |.
+
Hence we may put fab

(z) = va(b) (z) and fab
(z) = va(b+k) (z). Defining

x) ,
x ) ,
(b) R(z;
m
(b) R(z;
m
+
(32)
ab (z) = adj y
ab (z) = adj y
ac
ac
which are algebraic in A j (z) and xj , we find
+
(b)
(z) = m
+
fab
ab (z) (z),
(b+k) (z),
fab
(z) = m
ab (z)
(33)
where (z)
contains the exponential dependencies, and thus seemingly all the information
about the cores of the constituents. To make this more precise we compute
R (z), see
d
(b+k)
(b)
=
u ) dz fab (z) = 2 kd=1 u(b)
Eq. (18), using that Eq. (29) implies (with u
(z), finding that the factors (b) (z) drop out

Rad (z; x)fdb
(z) =
Rad
1
u(b) Rae (z; x )m
(z) bd .
eb (z) m
(34)
b,e=1
(z) is purely algebraic in A
j (z) and xj , as are m and Rm (z), which
This proves that Rm
determine the far field limit for the zero-mode density and the gauge field.
One might wonder how, given that the (b) (z) are of the wrong chirality in the context of the Nahm transformation, one could use these results to reconstruct the gauge field
for magnetic monopoles where the relation to the ADHM construction is not readily available. For this one observes that the columns of (b) (z) can be used to form a 2k 2k
matrix w(z). Using that D x w(z) = 0, one finds D x (w (z)1 ) = 0. Thus the columns of
w (z)1 give 2k independent solutions for each interval, from which n normalizable solutions (p) (z; x) should remain after imposing the appropriate boundary (cq. matching)
conditions. These are then used in Nahms original construction to compute the gauge field
(cf. Eq. (3))
(q) (z; x0, x ) dz,

(x)
=
(p) (z; x0, x)
Apq
(35)
where (p) (z; x0, x) g (z) (p) (z; x ).

There seems to be considerable advantage in using the Greens function (Fourier transformed ADHM) method, since it can solve the matching conditions without relying on the
availability of exact solutions for the normalizable dual zero-modes. To go beyond the approximations discussed in Section 2.2 and resolve the constituent cores we need to solve
for (z).
This cannot always be done in closed form, but it is given by an explicit integral
which can be performed numerically when required. For charge 2 monopoles Panagopoulos [16] was, however, able to find the exact integral. We can use the same ingredients
for the caloron case and explicitly solve for the Greens function in the case of charge 2
calorons.
3.1. Analytic expressions for charge 2
For charge 2 the number of invariants associated to the conserved quantities of the
Nahm equation is 8, of which Tr A j (z) 4iaj are related to the center of mass for the
244
constituents of given magnetic charge, coming from the interval under consideration. Assuming now that A j (z) is traceless, 5 invariants remain, given in terms of the symmetric
traceless tensor

1

1
Mij Tr A i (z)A j (z) ij Tr A 2k (z) .
(36)
2
3
Three of its parameters are associated to the rotation R which diagonalizes the 33 matrix,
M = R diag(c1 , c2 , c3 )Rt , where R is fixed by requiring c2 c1 c3 . The ci add to zero
and can be expressed in terms of the so-called scale (D) and shape (k) parameters,
1 2k2
k2 2
1 + k2
,
c2 = D 2
,
c3 = D 2
.
(37)
12
12
12
The Nahm equation for the case of charge 2 can be solved completely in terms of Jacobi
elliptic functions [20,21], which was summarized in Ref. [10]
c1 = D 2

1
g(z)
A j (z; a , R, h, D, k)g (z) 2iaj 12 + iDRj b fb D(z z0 ) h b h,
(38)
2
where h is a global gauge parameter and

k
k snk (z)
dnk (z)
f1 (z)
,
f2 (z)
,
f3 (z)
, k 1 k2
cnk (z)
cnk (z)
cnk (z)
(39)

with5 snk (z) = sin((z)), cnk (z) = cos((z)) and dnk (z) = 1 k2 sn2k (z) the standard
Jacobi elliptic functions. This does not yet address the matching of A j (z) on the different
intervals, where some difference between the monopole and caloron application appears.
For the caloron, apart from the axially symmetric solutions constructed in Ref. [14], we
found two sets of non-trivial solutions that interpolate between overlapping and wellseparated constituents. It is for these classes of solutions that we will resolve the cores,
when constituents overlap and the non-linearity plays an important role.
The next step in the construction is finding the zeros y of det Y (z). One has

1 2
2
det Y (z) = yi yj xi xj x ij (2) Mij ,

(40)
3
where we used y2 = 0. Substituting a parametrization for this null-cone in CP 2 , e.g.,
y = ( 12 (1 2 ), 2i (1 + 2 ), ), gives a 4th order polynomial. However, for finding
the 4 solutions we find it in this case more convenient to first diagonalize the matrix
xi xj 13 x 2 ij (2)2 Mij = Oik k Oj k . Introducing y = Ot y , the equation for the zeros
reduces to (y1 )2 1 + (y2 )2 2 + (y3 )2 3 = 0, which in addition to the null-cone condition,
(y1 )2 + (y2 )2 + (y3 )2 = 0, is now easily solved by

y(a) = 2 3 , (1)a+1 3 1 , i 2 1 ,

y(a+2) = y(a) , a = 1, 2,
(41)

5 (z) implicitly defined by the elliptic integral of the first kind z = (z) 1/ 1 k2 sin2 d .
0
245
where we fixed (for generic x ) the diagonalizing rotation O by ordering 1 3 2 .

Using Eq. (27) we find for u Ot u

3 1
2 3
u(a) = (1)a
,
,0 ,
u(a+2) =
u(a), a = 1, 2. (42)
2 1
2 1
One easily checks that u(1,2) and u(3,4) give rise to, respectively, the exponentially rising
and falling solutions.
It is also instructive to give the explicit expressions for the matrices m
(z) in Eq. (32)
(choosing c = 2). We note that for a 2 2 matrix adj Y = (Tr Y )12 Y , and without loss
R = 13 , D = 1 and h = 12 in Eq. (38), such that
in generality6 we take z0 = 0, a = 0,

(1)
(1)
(2)
(2)
1 y1 f1 (z) iy2 f2 (z) y1 f1 (z) iy2 f2 (z)
+
,
m
(z) =
4 4 x y(1) y (1)f3 (z) 4 x y(2) y (2)f3 (z)
3
3

(3)
(3)
(4)
(4)
1 y1 f1 (z) iy2 f2 (z) y1 f1 (z) iy2 f2 (z)
m
(z) =
(43)
.
4 4 x y(3) y (3)f3 (z) 4 x y(4) y (4)f3 (z)
3
3
We checked that Eq. (23), V(
x ) = (2)1 Tr(R + (z) + R (z))1 , evaluated using Eq. (34)
is independent of z and agrees with the result derived in Ref. [10].
We next address solving Eq. (31), which for charge 2 can be written as7
Mij ui yj ac 2i(
x y)ui A ac
d (z)
i (z)
=
.
ac
ac
dz
2(
x y) iyi Ai (z)
(44)
R = 13 , D = 1 and
For the same choice of parameters in Eq. (38) as above, z0 = 0, a = 0,
h = 12 , this gives (with a = 1 and c = 2)
d (z)
u1 f1 (z) iu2 f2 (z)

= 2(
x y)
.
dz
y1 f1 (z) iy2 f2 (z)
(45)
Although the dependence on x is complicated, the integral over z turns out to be manageable (as was observed before in the context of charge 2 monopoles, although we here
choose not to express the solution in terms of theta functions [16]). To solve the equation
we first rewrite the right-hand side of Eq. (45) using the fact that u y = i y (cf. Eq. (27)),
u1 f1 (z) iu2 f2 (z) (u3 y1 + iy2 )f1 (z) i(u3 y2 iy1 )f2 (z)
=
y1 f1 (z) iy2 f2 (z)
y3 (y1 f1 (z) iy2f2 (z))
=
f1 (z)f2 (z)y32 + 4iy1y2 (k )2

y3
(16 2 (
x y)2
y32 f32 (z))
u3
.
y3
In the last identity we used that y2 = 0, f12 (z) f22 (z) = 1 k2 = (k )2 and the fact that
x y)2 y32 f32 (z).
det Y (z) = 0 implies (y1 f1 (z) iy2 f2 (z))(y1 f1 (z) + iy2 f2 (z)) = 16 2 (
6 We may change z , a
0 , h, R and D by, respectively, translations, (gauge) rotations, and suitable rescalings.
7 Using that 1 {A , A }y u = 1 1 Tr(A A )y u = M y u 1 , since u
y = 0.
i
j i j
i j i j
ij i j 2
2
2 2
246
With

d
dz f3 (z) = f1 (z)f2 (z)
we can now integrate Eq. (45),

4 x y + f3 (z)y3
u3 1
sign(z)(k )2 y1 y2
(z)
= 2z(
x y) + log
+i
I (z),
y3 4
4 x y f3 (z)y3
2(
x y) 4y3

(4 x y)2
,
I (z) k f31 (z), n k (1, n) + |z|, n
(46)
y32
up to an irrelevant constant, where k (s, n) is the elliptic integral of the third kind8
s
k (s, n)
0
(1 nt 2 )
dt
(1 k2 t 2 )(1 t 2 )
(47)
We now combine these ingredients to give in terms of y the exact form for the homogeneous solution of the Greens function equation,

(adj Y (z))a2
adj Y (z) a2 = exp (z)
2z
u x
va (z) = (z)
(adj Y (z))12
(48)
or putting in all the relevant expressions

1

z
y1 y2 (k )2
|z| + k f3 (z), n k (1, n)
2(
y x)3 +
va (z) = exp i
y3
8(
x y)|z|

1/2 4 x y (1)a y3 f3 (z) 1/4
.
(4)1/2 y1 f1 (z) (1)a iy2 f2 (z)
4 x y + (1)a y3 f3 (z)
(49)
+
Substituting y = y(b) = Ot y(b) , with y(b) as defined in Eq. (41) gives fab
(z) = va (z)
(b+2)
and fab (z) = va

(z), and thereby the Greens function, once we specify the parameters
involved in the solutions to the Nahm equation.
(b)
4. Action and zero-mode density plots

The discontinuities in A j (z) at z = 1 and z = 2 implied by the Nahm equation,
Eq. (5), impose constraints which are (like the quadratic ADHM constraint) in general
difficult to solve. Work is in progress to describe the full parameter space for SU(2) and
charge 2, but in Ref. [10] we did find two non-trivial parametrizations for which we illustrate here in a number of figures how the full caloron solutions look like, using the exact
Greens function as constructed in the previous section. Taking advantage of the possibility
8 More commonly the elliptic integral of the third kind is defined as (n; , k) = (sin , n). Note that
k
f (z)
I (z) can alternatively be written as 1 3 (1 t 2 /n)1 (t 2 k2 )1/2 (t 2 1)1/2 dt, from which it follows that
d
2
1 d f (z) = sign(z)(1 nf 2 (z))1 , using f 2 (z) 1 = f 2 (z), f 2 (z)
dz I (z) = ((1 nf3 (z))|f1 (z)f2 (z)|)
dz 3
3
3
2
3
d
2
2
k = f1 (z) and dz f3 (z) = f1 (z)f2 (z).
247
Fig. 2. An example to illustrate the location of the disc singularities (light and dark shaded according to magnetic
charge) for a rectangular (left) and crossed (right) configuration. The latter is shown at = /2, for d = /32
and = /4 (k = 0.962, D = 3.894 and = /4). The curves indicate the would-be constituent locations at
fixed d and , varying from (to which the arrows point) to 0. For = and 0 the discs collapse to lines
(k = 1) with no singularities remaining, except at the endpoints.
to work with arbitrary arithmetic precision the programme Maple has been used for these
calculations. The configurations are formulated in terms of the two intervals z [1 , 2 ]
and z [2 , 1 + 1 ], each associated with two constituent monopoles of equal magnetic
charge, but opposite in sign from one interval to the next. Apart from a shift and (gauge)
orientation, the configuration is described by a shape (k) and scale (D) parameter (see
Section 3.1), for simplicity assumed to be the same on both intervals. We also take all
constituents to be of equal mass, 1,2 = 12 (2 = 1 = 14 ), most relevant for the confined
phase with tr P = 0.
The two periodic and two anti-periodic chiral fermion zero-modes each have support
on oppositely charged constituents (see Fig. 1) and in the far field limit it was found that
the zero-mode density (summed over the two zero-modes implied by the index theorem)
is described by a disc singularity, bounded
by an ellipse with semi-major axis D/4 and
eccentricity (1 k )/(1 + k ) (where k = 1 k2 ). This revealed that the core of a cluster
of like-charged constituents is in general extended, unless the individual constituents are
well separated. The far field only describes the (algebraic) Abelian component of the gauge
field and to resolve the structure of the core we need to determine the full non-Abelian
structure.
For the first parametrization (called rectangular) the two discs are parallel and separated in height by a distance d. The configuration is charaterized by the two extremal points
along the major axis of the disc, also called would-be constituent locations9

1
1
(j )
m
j
1
ym = 0, (1) d, (1) (4) D ,
2d = Df2 D ,
(50)
2
4
up to an overall shift and orientation (the definition of f2 (z), which involves k, can be found
in Eq. (39)). A typical example is shown in Fig. 2(left). Apart from k and D, the parameters
9 In their immediate neighbourhood the action density is maximal; they are the constituent locations in the
point-like limit, k 1.
248
Fig. 3. Shown is the energy density (middle) for 2 = 0.25, k = 0.570 and D = 6.915, in the monopole limit
d and in the xz plane through one of the discs for the rectangular configuration, see Fig. 2(left). On a
scale enhanced by a factor 4 2 are shown the densities for the two monopole zero-modes (left and right).
that enter Eq. (38) for the mth interval are a = (0, 12 (1)m d, 0), R = 13 , h = 12 and
z0 = 14 (1 + (1)m ). In all cases discussed here we have also put 0 = 0. One verifies that
j
the discontinuities of A j (z) at z = m are given by 2im with appropriately chosen a
(cf. Eq. (6)), as discussed in detail in Ref. [10].
For the second parametrization (called crossed) the two discs are coplanar and intersect, see Fig. 2(right) for a typical example. Their relative orientations can vary between
perpendicular and coinciding (for which k is forced to 1). Here we choose for the parameters in Eq. (38) h and R to be non-trivial (isospin) rotations around the y-axis with angles
(1)m , respectively, (1)m , and a = (0, 0, 12 (1)m d cos ), whereas z0 and 0 are as
in the rectangular case. The would-be constituent locations are now given (up to an overall
shift and orientation) by

1
(j )
ym = (1)j (4)1 D sin , 0, (1)m+j (4)1 D cos (1)m d cos , (51)
2
where conveniently gives the orientation of each of the two discs with respect to the
z-axis. The angle originates from the definition of a , which through Eq. (6) determines
the discontinuity of A j (z). To ensure the proper matching, the following three equations
need to be satisfied [10]

1
1
D sin( ) f3 D f1 D = 8d(1 sin ),
4
4

1
Df2 D + 8d sin = 0,
(52)
4
which determine , k and D for given , and d. Fig. 2 illustrates that for these crossed
configurations the two discs always overlap, unlike for the rectangular case. This formed
an important motivation for the present study, so as to determine in how far the overlapping
discs would affect the behaviour in the core.
249
Fig. 4. In the middle is shown the action density in the plane of the constituents at t = 0 for an SU(2) charge
2 caloron with tr P = 0 in the crossed configuration of Fig. 2, hence with k = 0.962 and D = 3.894. On a
scale enhanced by a factor 16 2 are shown the densities for the two zero-modes, using either periodic (left) or
anti-periodic (right) boundary conditions in the time direction.
We will illustrate, using the exact formalism, how the action and zero-mode densities for
these solutions behave. For the rectangular case we are mostly interested in the monopole
limit, d . For finite d, the density separates in two contributions, where each in the
limit d is exactly the density for a charge 2 monopole with the same values of k and
D. Note that in the limit d the matching conditions for A j (z) decouple the intervals, turning into pole conditions at z = 1,2 , as is appropriate for multi-monopoles [20].
We do recover from this the known results for the charge 2 monopoles [22], like the doughnut structure for k = 0, corresponding to two superimposed monopoles. The interest in the
monopole limit comes from the fact that the zero-mode densities for multi-monopoles had
not been studied in detail before [23]. In Fig. 3 we give the densities for k = 0.570 and
D = 6.915, which is intermediate between the doughnut and well-separated monopole
configurations. On the other hand, for d small (compared to = 1) the configuration will
look like two non-dissociated calorons, and in particular is no longer static. When D remains much bigger than d, forcing k 1, these behave as two well-separated charge 1
instantons. Otherwise, when D is comparable to d, one finds overlapping instantons [24].
An example for the crossed configuration with k = 0.997 and D = 8.753 was already
shown in Fig. 1. In this case D is large enough for the like-charge constituents to be
separated, as is particularly clear from the zero-modes, which are essentially no longer
overlapping. But two nearest neighbour (oppositely) charged constituents
still show appreciable overlap. The distance between these nearest neighbours is 14 D 2/ = 0.985. As
this is comparable to = 1 we would expect the configuration to depend on time. Indeed,
at the maxima of the action density its value of 1.18 16 2 at t = 0 is reduced by almost
50% at t = 0.5. At the center of mass, where the action dencity is much lower, there is still
250
a time dependence. However, far from all cores the field becomes static.10 Increasing D
further (which will push k closer to 1) the configuration quickly turns into well separated
spherically symmetric static BPS monopoles.
More interesting is to consider the case with smaller D, like D = 3.894 and k = 0.962,
for which the disc singularities were illustrated in Fig. 2. The corresponding densities are
shown in Fig. 4. We see that the constituents are now so close that they form a doughnut, but
we stress this is different from the static monopole doughnut, which has k = 0. Since the
oppositely charged constituents now are as close as 0.438, which is considerably smaller
than the time extent, the solution will have a strong time dependence. When D is decreased
even further, it will turn into a charge 2 instanton localized in both space and time.
5. Higgs field asymptotics

In this section we make some comments on the far field limit of the gauge field. As we
discussed before, the gauge field far removed from any core becomes Abelian (as well as
static). The Abelian subgroup is the one that leaves the holonomy invariant, in the periodic
gauge equivalent to leaving the constant asymptotic value of the adjoint Higgs field A0
invariant. For definiteness, let us consider the case of SU(2) with k = 2, = 1 and P =
exp(2i
) (i.e., 2 = 1 = ||).
Up to exponential corrections we have
1
x ),
Aff0 (
(53)
x ) = 2i
i (
2
where we can express (
x ) in terms of the far field limit of the Greens function at the
impurities [14]

1

(
x ) = 1 1 1 Tr fxff (2 , 2 )S2
(54)
i Tr fxff (2 , 2 )2i .
Using the twistor description of magnetic monopoles Hurtubise [13] was able to explicitly compute the asymptotic Higgs field for the SU(2) magnetic monopole long ago. The
x ), Eq. (23),
function he found for this algebraic tail amazingly agrees exactly with Vm (
which was introduced to describe the caloron zero-mode density (m denotes the interval
and hence the type of constituents to which the corresponding zero-modes would localize
[10]). As mentioned before, from our multi-caloron results one can recover the multimonopole results by sending the constituent monopoles with the unwanted magnetic
charge to infinity, cf. Eq. (24).
Although this tends to be cumbersome to show, (
x ) in the far field can be written
x ) 2 (
x ), where m (
x ) is the contribution coming from the type m constituent
as 1 (
monopoles and the difference in sign is due to the sign change in the magnetic charge. This
is simply because the field is Abelian in the far field and linear superposition preserves the
x ) = 2Vm (
x ), such that for the SU(2) caloron
self-duality. Hence, m (
(
x ) = 2V1 (
x ) 2V2 (
x ).
(55)
10 For any SU(n) and topological charge k the far field limit is static, provided all constituent monopoles have
a non-vanishing mass.
251
We checked that this result indeed holds for the solutions discussed in Section 4, even
when the two types of monopole structures are not well separated. This relation trivially
holds for the axially symmetric solutions that were introduced in Ref. [14], where (
x)
was explicitly shown to factorize in a sum of point charge contributions, compatible with
what was found in Ref. [10] for the zero-mode densities. Therefore, in the far field limit
(i.e., for the algebraic part) the singularity structure in the zero-mode density agrees exactly
x ). Such a relation is at the heart
with the Abelian charge distribution, as given by i2 (
of using chiral fermion zero-modes as a filter to isolate the underlying topological lumps
from rough lattice Monte Carlo configurations [9].
6. Discussions
In this paper we have analyzed the higher charge caloron solutions and showed how
to obtain exact results by suitably combining techniques developed in the context of the
Nahm transformation and the ADHM formalism. The aim of these studies has been to establish that SU(n) caloron solutions of charge k can be described in terms of kn monopole
constituents, and that these can be viewed as independent constituents. A natural way to
get an ensemble would be to consider approximate superpositions of k charge 1 calorons,
but this would lead to an unwanted memory effect, with constituents remembering from
which caloron they originated [14]. Our studies, within the context of self-dual configurations, have shown nevertheless that the constituents have an independent identity, with the
only requirement that the net magnetic and electric charge of the configuration vanishes
(each of the n types of constituents should occur with the same number). A recent lattice
study [25], using the technique of over-improvement [26], fully confirms this picture.
It is therefore reasonable to consider the constituents as the independent building blocks
for constructing an ensemble of monopole constituents, something that was not questioned in Ref. [7], but like for the instanton liquid [27] forms an essential assumption
in a semi-classical study. Clearly the expectation is that semi-classical methods no longer
work in the confined regime, at least for the part of the parameter space that corresponds
to well-separated constituents, that is typically associated to instantons with a large scale
parameter. It is not unlikely that the density of these constituents at low temperatures is so
high that they form a coherent background and as such will no longer easily be recognized
as lumps. With high quark densities leading to deconfinement, it may perhaps be that a high
constituent monopole density will lead to confinement [28], although for now we have to
leave this as a speculation.
Instantons that overlap get deformed and depending on the relative gauge orientation
tend to repel, i.e., inspecting the action density distribution they do not get closer than
a certain distance [24]. When deconstructing instantons in monopole constituents, interestingly only like-charge constituents will show this effect, manifesting itself through the
extended core structure. For unlike charges, from the point of view of the Abelian field, the
configuration behaves as with linear superposition. If as a consequence of this all Abelian
charge is annihilated, it disappears through forming a small instanton (localized in space
and time), which in the limit of zero size describes the boundary of the moduli-space. The
interaction between constituents of opposite duality is more complicated [8,29].
252
In conclusion, calorons with non-trivial holonomy have revealed a rich structure, incorporating traditional instanton physics, but allowing for gauge fields that inherit some
essential features associated to a confining background not present in the traditional formulations. The fact that the underlying constituents are monopoles opens the way to describe
the confining aspects of the theory in terms of these degrees of freedom. Much work remains to be done when it comes to understanding the dynamics, but we hope to have
convinced the reader that a consistent picture is developing that holds considerable promise
for the future.
Acknowledgements
We thank Conor Houghton for pointing us to Hurtubises paper, and Chris Ford for
discussions on the analytical aspects. We are grateful to Christof Gattringer, Michael Ilgenfritz, Boris Martemyanov and Michael Mller-Preussker for stimulating discussions
and correspondence on lattice implementations. The research of F.B. is supported by FOM.
He is grateful to Chris Ford and Conor Houghton for their hospitality while visiting Trinity
college in Dublin, and thanks them, as well as Werner Nahm and Samson Shatashvili, for
their interest.
Appendix A
In this appendix we derive the zero-mode limit for the action density, which assumes
that the distance of x and the constituents of type m to all constituents of type m = m is
large, but where x and the constituent locations of type m may otherwise be arbitrary.
As in Ref. [10] we take z = m + 0 for computing Fz and use that we can write
det(12k Fm ) = det(12k LK), where K Fm1 m1 1 g (1)Fn n m+2 Fm+1
and L m+1 Fm m , with

1

1k
1k
1k
1k
,
Tm
m
+
+ ( ) 2R ( )
2Rm
2Rm1
(m ) 2Rm1
(m )
m
m
m
+

fm (m+1 )fm+ (m )1
0
Fm
(A.1)
.
0
fm (m+1 )fm (m )1
We note the K has no remaining dependence on the constituent locations of type m. Writ with
ing LK L K + L K,

K++ K+
0
0
K
,
K
,
K+ K
0
1k

0 L+
L++ 0
,
L
,
L
(A.2)
0 L
L+ 0
det(K 1 L L K K 1 ).
we find det(12k LK) = det(K)
We next use

1

1
K+
K++ K++
0
,
K K 1 =
K 1 =
1
0
1k
K+ K++
(K 1 )

(A.3)
253
1
1
1
and note that in the zero-mode limit K++
, K++
K+ , K+ K++
and (K 1 ) are exponentially small (cf. Ref. [10, Appendix A]), such that

det ieix0 (12k LK) = det e2ix0 K++ det(L++ ).
(A.4)
With the definition of L we now find

+

1 1
L++ = Rm+1
(m+1 ) Rm+1
(m+1 ) + Sm+1 U m Rm1
(m ) + Sm ,
4
where
(A.5)
+
1
+
fm+ (m+1 )fm+ (m )1 Rm
(m )Zm
U m = Zm+1
Zm+1
fm (m+1 )fm (m )1 Rm
(m )Zm
,

1
= 1k Rm
(m ) + Sm Rm1
(m ),
Zm
1

+
= 1k Rm
(m ) Rm1
(m ) + Sm ,
Zm
(A.6)
and U m contains all contributions due to the constituent locations of type m, up to exponential corrections in the distance of these, and of x , to the other constituents. Hence
log det(ieix0 (12k LK)) splits into the sum of two contributions, log det(U m ) and
+
1
log det[ 14 (Rm1

(m ) + Sm )e2ix0 K++ Rm+1
(m+1 )(Rm+1
(m+1 ) + Sm+1 )], where the
last term only depends on the constituent locations of type m = m whose contribution
will decay inversely proportional to the fourth power of their distance. Allowing for algebraic decay (or in the monopole limit, sending all constituents of type m = m to infinity)
= 1 , one thus finds Eq. (24).

= Zm
such that in addition Zm+1
k
A simple way to derive the result for the far field limit in Eq. (25) is by noting that
in this case all fm (m+1 )fm (m )1 are exponentially small and Fm can be approximated by diag(Fm++ , 0), with Fm++ = fm+ (m+1 )fm+ (m )1 . This therefore acts as a
projection on the ++ component and is thus seen to lead to det(ieix0 (12k LK)) =
++ = 1 R 1 ( )
det(e2ix0 g (1)Fn++ n++ F1++ 1++ ). Using the fact that [10] m
m
m
2 m
gives the required result.
References
[1] B.J. Harrington, H.K. Shepard, Phys. Rev. D 17 (1978) 2122;
B.J. Harrington, H.K. Shepard, Phys. Rev. D 18 (1978) 2990.
[2] D.J. Gross, R.D. Pisarski, L.G. Yaffe, Rev. Mod. Phys. 53 (1981) 43.
[3] T.C. Kraan, P. van Baal, Phys. Lett. B 428 (1998) 268, hep-th/9802049;
T.C. Kraan, P. van Baal, Nucl. Phys. B 533 (1998) 627, hep-th/9805168;
T.C. Kraan, P. van Baal, Phys. Lett. B 435 (1998) 389, hep-th/9806034.
[4] K. Lee, P. Yi, Phys. Rev. D 56 (1997) 3711, hep-th/9702107;
K. Lee, Phys. Lett. B 426 (1998) 323, hep-th/9802012;
K. Lee, C. Lu, Phys. Rev. D 58 (1998) 025011, hep-th/9802108.
[5] N. Weiss, Phys. Rev. D 24 (1981) 475.
[6] N.M. Davies, T.J. Hollowood, V.V. Khoze, M.P. Mattis, Nucl. Phys. B 559 (1999) 123, hep-th/9905015.
[7] D. Diakonov, N. Gromov, V. Petrov, S. Slizovskiy, Quantum weights of dyons and of instantons with nontrivial holonomy, hep-th/0404042.
254
[8] E.-M. Ilgenfritz, M. Mller-Preussker, A.I. Veselov, in: V. Mitrjushkin, G. Schierholz (Eds.), Lattice Fermions and Structure of the Vacuum, Kluwer, Dordrecht, 2000, p. 345, hep-lat/0003025;
E.-M. Ilgenfritz, B.V. Martemyanov, M. Mller-Preussker, S. Shcheredin, A.I. Veselov, Phys. Rev. D 66
(2002) 074503, hep-lat/0206004.
[9] C. Gattringer, Phys. Rev. D 67 (2003) 034507, hep-lat/0210001;
C. Gattringer, S. Schaefer, Nucl. Phys. B 654 (2003) 30, hep-lat/0212029;
C. Gattringer, R. Pullirsch, hep-lat/0402008.
[10] F. Bruckmann, D. Ngrdi, P. van Baal, Nucl. Phys. B 666 (2003) 197, hep-th/0305063.
[11] M. Garca Prez, A. Gonzlez-Arroyo, A. Montero, P. van Baal, J. High Energ. Phys. 9906 (1999) 001,
hep-lat/9903022;
P. van Baal, in: V. Mitrjushkin, G. Schierholz (Eds.), Lattice Fermions and Structure of the Vacuum, Kluwer,
Dordrecht, 2000, p. 269, hep-th/9912035.
[12] E.M. Ilgenfritz, B.V. Martemyanov, M. Mller-Preussker, A.I. Veselov, Phys. Rev. D 69 (2004) 114505,
hep-lat/0402010.
[13] J. Hurtubise, Commun. Math. Phys. 97 (1985) 381.
[14] F. Bruckmann, P. van Baal, Nucl. Phys. B 645 (2002) 105, hep-th/0209010.
[15] W. Nahm, Self-dual monopoles and calorons, in: G. Denardo, et al. (Eds.), Lecture Notes in Physics,
vol. 201, 1984, p. 189.
[16] H. Panagopoulos, Phys. Rev. D 28 (1983) 380.
[17] M.F. Atiyah, N.J. Hitchin, V.G. Drinfeld, Yu.I. Manin, Phys. Lett. A 65 (1978) 185;
M.F. Atiyah, Geometry of YangMills fields, in: Fermi Lectures, Scuola Normale Superiore, Pisa, 1979.
[18] M. Garca Prez, A. Gonzlez-Arroyo, C. Pena, P. van Baal, Phys. Rev. D 60 (1999) 031901, hepth/9905016;
M.N. Chernodub, T.C. Kraan, P. van Baal, Nucl. Phys. B (Proc. Suppl.) 8384 (2000) 556, hep-lat/9907001.
[19] W. Nahm, The algebraic geometry of multi-monopoles, in: M. Serdaroglu, et al. (Eds.), Group Theoretical
Methods in Physics, Istanbul, 1982, p. 456.
[20] W. Nahm, Multi-monopoles in the ADHM construction, in: Z. Horvath, et al. (Eds.), Gauge Theories and
Lepton Hadron Interactions, CRIP, Budapest, 1982.
[21] S.A. Brown, H. Panagopoulos, M.K. Prasad, Phys. Rev. D 26 (1982) 854;
A.S. Dancer, Commun. Math. Phys. 158 (1993) 545.
[22] P. Forgcs, Z. Horvth, L. Palla, Nucl. Phys. B 192 (1981) 141;
M.F. Atiyah, N.J. Hitchin, The Geometry and Dynamics of Magnetic Monopoles, Princeton Univ. Press,
Princeton, NJ, 1988.
[23] E.F. Corrigan, P. Goddard, Ann. Phys. (N.Y.) 154 (1984) 253.
[24] M. Garca Prez, T.G. Kovcs, P. van Baal, Phys. Lett. B 472 (2000) 295, hep-ph/9911485.
[25] F. Bruckmann, M. Ilgenfritz, B. Martemyanov, P. van Baal, hep-lat/0408004.
[26] M. Garca Prez, A. Gonzlez-Arroyo, J. Snippe, P. van Baal, Nucl. Phys. B 413 (1994) 535, heplat/9309009.
[27] T. Schfer, E.V. Shuryak, Rev. Mod. Phys. 70 (1998) 323, hep-ph/9610451;
D. Diakonov, Prog. Part. Nucl. Phys. 51 (2003) 173, hep-ph/0212026.
[28] F. Bruckmann, D. Ngrdi, P. van Baal, Acta Phys. Pol. B 34 (2003) 5717, hep-th/0309008.
[29] E.M. Ilgenfritz, B.V. Martemyanov, M. Mller-Preussker, A.I. Veselov, Eur. Phys. J. C 34 (2004) 439, heplat/0310030;
E.M. Ilgenfritz, M. Mller-Preussker, B.V. Martemyanov, P. van Baal, Phys. Rev. D 69 (2004) 097901,
hep-lat/0402020.

Solution of the fan diagram equation

in (2 + 1)-dimensional QCD
J. Bartels a , V.S. Fadin b,1 , L.N. Lipatov c,1
a II Institut fr Theoretische Physik, Universitt Hamburg, Luruper Chaussee 149, D-22761 Hamburg, Germany
b Institute for Nuclear Physics and Novosibirsk State University, 630090 Novosibirsk, Russia
c Petersburg Nuclear Physics Institute, Gatchina, 188 300 St. Petersburg, Russia

Abstract
We investigate the BalitskyKovchegov (BK) equation for D = 3 spacetime dimensions, corresponding to one transverse coordinate, and we show that it can be solved analytically. The explicit
solutions are found in the linear approximation and for the particular cases when they do not depend
on the impact parameter of the dipole or on its rapidity. It is shown, that in a general case different
solutions are related by an infinite parameter group of transformations. The key observation is that
the equation has the KowalewskayaPainlev property, which gives the possibility of reducing the
non-linear problem to the solution of a linear integral equation.
2004 Published by Elsevier B.V.
PACS: 12.38.Bx
1. Introduction
In the framework of perturbative QCD the most general basis for the theoretical description of small-x processes is given by the BFKL approach [1], based on the gluon
reggeization. Taking into account also analyticity and unitarity constraints for the S-matrix
Work supported in part by INTAS and by the Russian Fund of Basic Researches.
E-mail address: jochen.bartels@desy.de (J. Bartels).
1 Humboldt Preistrger.
0550-3213/$ see front matter 2004 Published by Elsevier B.V.

doi:10.1016/j.nuclphysb.2004.07.035
256
J. Bartels et al. / Nuclear Physics B 698 (2004) 255276
one can calculate the high energy asymptotics of scattering amplitudes with a good accuracy [2]. In particular, using this approach the kernel of the BFKL equation was obtained in
the next-to-leading order (NLO) [3]. To avoid a strong renormalization scale dependence
of its solution in the MS-scheme one can chose the BrodskyLepageMackenzie (BLM)
optimal scale setting within physical renormalization schemes [4]. It is important that the
BLM procedure allows to preserve, approximately, some good properties of the leading
order approximation, in particular, its conformal invariance [5]. Incorporation of the corrections related to the renormalization group symmetry [6] should make the predictions of
the BFKL equation for the small-x physics more reliable.
However, at asymptotically large energies the BFKL approach should be generalized,
since it leads to a power-like growth of cross sections. The problem of the Froissart bound
violation cannot be solved by the calculation of radiative corrections at any fixed order
and requires other methods. One of the possibilities is to take into account the multireggeon contributions in the framework of the BKP equations [7]. A general framework
for considering unitarization effects is the reggeon field theory, describing the interactions of reggeized gluons. Possible ways of deriving such a field theory are based on the
high-energy effective QCD action [8], generalizing the BFKL approach for multi-particle
production and multi-Reggeon interaction or, alternatively, on the BFKL analysis of multiparticle production amplitudes [9].
In a first step, the unitarization program can be realized by non-linear generalizations of
the BFKL equation, related to the idea of saturation of parton densities [10]. The most important ingredient is the transition vertex for 2 4 reggeized gluons which, in the BFKL
approach, has been calculated in [11] and further investigated in [12,13]. Using this transition vertex, as a first step beyond the BFKL equation, one can formulate a nonlinear
evolution equation which sums the QCD fan diagrams consisting of BFKL Greens functions [14]. In a future step towards finding solutions to the reggeon field theory, closed
loops of BFKL Pomerons should be included and calculated.
At present it is fashionable to describe the saturation in DIS within the color dipole
approach [15], which in the target rest frame has an attractive and clear physical interpretation: the incoming splits into a q q colour dipole long before the dipole interacts with
the target. In this approach the cross section is presented as

(x, Q ) =
2
1
2
d r

2
dz r, z, Q2 dp (r, x),
(1)
where (r, z, Q2 ) is the photon wave function, z the quark momentum fraction, and
r = |r | the transverse size of the colour dipole, r = r1 r2 ; the transverse vectors r1 and
r2 are the quark and antiquark coordinates, and dp (r, x) is the dipole cross section,

dp (r, x) = 2

d 2 b 1 S(r1 , r2 ; Y ) .
(2)
Here b = (r1 + r2 )/2 is the impact parameter of the quarkantiquark system, Y = log(1/x),
and S(r1 , r2 ; Y ) = tr(U + (r1 )U (r2 ))/Nc Y is the S-matrix for the elastic scattering, which
257
can be presented as an average of the product of two Wilson lines

+
+
a
U (x ) = P exp ig dx Aa (x , x )t .
There exist two different approaches for calculating the Y dependence of S(r1 , r2 ; Y ).
The first one starts from an infinite hierarchy of coupled equations [16] for products of any
given number of Wilson lines. Taking the limit of large Nc and considering the case of the
target being a large nucleus, the equation for the two-line correlator decouples and takes a
simple form [17]:

(r1 r2 )2
S(r1 , r2 , Y ) s Nc
2
=
d
r

Y
2 2
(r1 r)2 (r2 r)2

S(r1 , r, Y )S(r , r2 , Y ) S(r1 , r2 , Y ) .
(3)
At small N = 1 S, one can neglect terms nonlinear in N , and this equation turns into
the colour-dipole version of the BFKL equation for the dipole cross section N . In [14] it
has been shown that Eq. (3) can be considered as a special limit of the QCD fan diagram
equation mentioned before: apart from taking the large-Nc limit, the dipole cross section
has to be in the Mbius class of functions. An alternative approach for describing saturation
is the Color Glass Condensate [18]. In this framework, the Y -evolution is based upon
functional equations. It is affirmed [19] that both approaches give identical results for the
calculation of observables.
Eq. (3) has attracted much interest. First, numerical solutions have been found for a bindependent approximation [20], and later on also for the full b-dependent equation [21].
Several attempts have recently been made to find analytical solutions, at least in some
approximation [22]. However, so far no exact analytical solution is known. Therefore, it
will be interesting to find a model in which the BK equation could be solved.
In this paper we generalize the BK-equation (3) to an arbitrary dimension D of the
spacetime. For the case D = 2 + 1, which corresponds to one transverse dimension, we
shall present analytical solutions (the BFKL dynamics was investigated at D = 2 + 1 in
the papers [23]). The key observation in finding these solutions is that the equation has the
KowalewskayaPainlev property. We proceed in several steps of approximations. After
writing down the equation in one transverse dimension (Section 2) we first discuss the linear approximation (Section 3). Then we turn to the nonlinear equation, beginning with the
b-independent case (Section 4), where we find explicit solutions for arbitrary initial conditions. Passing on to the b-dependence we first find all stationary (rapidity independent)
solutions (Section 5), generalizing then to the rapidity dependent case (Section 6). We conclude with the investigation of the infinite parameter symmetry of the equation (Section 7)
and its general solution (Section 8). A few details are put into Appendix A.
2. The fan equation for 2 + 1 QCD

Let us begin by writing down the fan equation for the S-matrix describing the dipole
evolution in the (2 + 1)-dimensional QCD. The kernel of the BK equation (3) is equal to
258
the probability density in the space of transverse coordinates and rapidity for the soft gluon
emission by the quarkantiquark pair:

2

s Nc
(r1 r2 )2
d D2 k k i k(
r1 r )
r2 r )
i k(

e
(4)
=
2
N
e
s c

2 2 (r1 r)2 (r2 r)2
(2)D2 k2
at D = 4. It is clear that if the spacetime dimension takes the value D = 4 + 2, the BK
equation preserves its form, but since

r )k
d 2+2 k exp(i k
i (1 + )
r
(5)
=
,
(2)2+2
2

(r 2 )1+
k2
the kernel changes according to
s Nc
(r1 r2 )2
2 2 (r1 r)2 (r2 r)2

2
(r1 r)
s Nc (1 + ) 2
(r2 r)
.

2 2
(r1 r)2(1+) (r2 r)2(1+)
(6)
Putting = 1/2 we obtain:

S2 1 (y)
=
y
2

d0 S2 0 (y)S0 1 (y) S2 1 (y) ,
y=
s
Y,
2
(7)
where the inequality 2 > 1 is imposed without any restriction of generality. Note that the
scattering amplitude N2 1 (y) is related to the S-matrix S2 1 (y) by the identity
N2 1 (y) = 1 S2 1 (y).
(8)
We introduce the dipole impact parameter b and its size

2 + 1
,
= 2 1 ,
= 0 1 .
2
Then the fan equation can be presented as follows:
b=
S(b, , y)
=
y

d S b +

2 ,

, y S b
(9)

2 , , y S(b, , y)
(10)
or, in a simpler way,

+ S(b, , y) = d S b
y

2 ,

, y S b +

2 , ,y .
(11)
Now we go to the mixed representation (b, p)

i
S(b, , y) () =
i
dp p
e s(b, p, y)
2i
(12)
259
with the Laplace transform

s(b, p, y) =
d ep S(b, , y).
(13)
In this representation is the differential operator
s(b, p, y) s(b, p, y).

p
The convolution of two functions in the -representation
s(b, p, y) =

F12 () =
d F1 ( )F2 ( )
(14)
(15)
is reduced to the product of their Mellin transforms

f12 (p) = f1 (p)f2 (p),
f (p) =
d ep F ().
(16)
Moreover, a more general relation holds

kl
f12
(p) = l f1 (p) k f2 (p)
(17)
for the Mellin transform of the function

kl
F12 () = d ( )k ( + )l F1 ( )F2 ( ).
(18)
Let us introduce the anti-normal ordering N of operators depending on the two momenta
pr (r = 1, 2) and the corresponding derivatives r = /(pr ), according to the rules
N (pr s ) = N (s pr ) s pr = pr s + rs .
(19)
Then one can verify that the fan equation in the mixed representation (b, p) can be written
in the form

s(b, p, y) = lim N s b + 12 2 , p1 , y s b 12 1 , p2 , y 1,
(20)
pr p
y p
where it is implied that, after performing the transformation of the differential operator to
the anti-normal form, one omits all terms with the differential operators due to the relations
1l 2k 1 = 0,
k, l 1
and, at the end, puts p1 = p2 = p.

Note that one can consider a modified equation for the S-matrix in the case of a non-zero
temperature T in the t-channel (see [24]). In this case S(, y) is assumed to be a periodic
function of k

S2 1 (y) = S2 1 (y) +1/T .
(21)
k
260
3. The BFKL equation in (2 + 1)-dimensional QCD

To begin with, we consider the linearized case corresponding to the BFKL equation
in (2 + 1)-dimensional spacetime. Using the relation S(b, , y) = 1 N(b, , y) and
neglecting nonlinear contributions in (11), one can obtain

+ N(b, , y) = d N b
y

2 ,

, y + N b +

2 ,
, y
.
(22)
This equation is simplified significantly in the case when N does not depend on b

+ N(, y) = 2 d N( , y).
y
(23)
For the Mellin transform

n(p, y) =
d e
i
N(, y) () =
N(, y),
dp p
e n(p, y)
2i
we obtain the differential equation

n(p, y) = n(p, y).

y p
p
(24)
(25)
Its solution with the initial condition n(p, 0) = n0 (p) is

n(p, y) =
(y + p)2 n0 (y + p)
.
p2
(26)
The second order pole at p = 0 provides a linear dependence of N(, y) at large

N(, y)
y 2 n0 (y),
(27)
which leads to a violation of the S-matrix unitarity at y = 0 for any initial condition. Note
that Eq. (23) has a stationary (y-independent) solution N(, y) = c, with arbitrary c. But
this solution is not acceptable from the physical point of view.
Let us turn now to Eq. (22) for the amplitude depending on b. Performing the Mellin
and Fourier transformations in the variables and b, respectively,
i
N(b, , y) () =
i

n(q, p, y) =
0
dp p
e
2i
d ep
dq iqb
e n(q, p, y),
2
db eibq N(b, , y),
(28)
we obtain

8p
n(q, p, y).
n(q, p, y) = 2
y p
4p + q 2
261
(29)
Its general solution is

f (p + y; q)
,
4p2 + q 2
n(q, p, y) =
(30)
with the arbitrary function f (p + y; q). For the initial condition n(q, p, 0) = n0 (q, p) the
result can be written as follows:
(4(y + p)2 + q 2 )n0 (q, y + p)
,
4p2 + q 2
n(q, p, y) =
(31)
where

n0 (q, p) =
d e
db eibq N(b, , 0).
(32)
Therefore
i
N(b, , y) =
dp p
e
2i
dq iqb 4(y + p)2 + q 2

e
2
4p2 + q 2
(p+y)
d e
db eib q N(b , , 0).
(33)
By introducing the Green function Gy (b, ; b, , 0) according to the definition

N(b, , y) =
y
d e
db Gy (b, ; b, )N(b , , 0),
(34)
we obtain for it
Gy (b, ; b, )
= (b b )( )

q( )
q( ) 2y 2
dq iq(bb )

e
+
sin
,
2y cos
+ ( )
2
2
q
2
(35)
so that the Green function for the BFKL Pomeron is a polynomial in y

Gy (b, ; b, )
= (b b )( )

2

y b b +
+ ( )
+
y
.
b
+
2
2
=
(36)
262
Therefore the y-evolution of the initial dipole distribution N2 1 (0) in the (2 , 1 )representation is simple:
N2 1 (y) = e
y21
2
N2 1 (0) + y

d0 ey20 N2 0 (0) + ey01 N0 1 (0)
2
+y
2
d2
d1 e(2 1 )y N2 1 (0).
(37)
For large y 1/, where in (30) one can neglect p in comparison with y, the solution
in the coordinate representation is

N(b, , y) =

dq iqb sin(q/2) 2
e
4y + q 2 n0 (q, y).
2
2q
(38)
In particular, for large b we obtain the above discussed expression N(b, , y)
y 2 n0 (y). On the other hand, at b one has

N(b, , y)
dq sin(q/2) 2
2y n0 (q, y).
2
q
(39)
Note, that Eq. (29) admits stationary solutions

n(q, p, y) =
f (q)
4p2 + q 2
with an arbitrary function f (q). For N(b, , y) we obtain correspondingly

N(b, , y) = a b +
a b
2
2
(40)
(41)
with an arbitrary amplitude a. Actually, the existence of such solutions can be seen from
Eq. (22). They are given by a difference of the amplitudes depending on transverse coordinates of quark 1 and antiquark 2 , respectively. Splitting of the dipole does not lead
to any evolution of such a solution: new quarks and antiquarks on the edges of the dipole
have the same transverse coordinates, and therefore their contributions cancel in the linear
approximation.
4. Solutions depending only on the dipole size

Next let us investigate the simple case when the solution does not depend on b, which,
formally, corresponds to the asymptotics of s(b, p, y) at b
lim s(b, p, y) = s(p, y).
(42)
263
This function describes the S-matrix for the scattering of the dipole with the size on the
hadron system with a large radius R , provided that the impact parameter b is restricted
to the interval
R b .
In this case the fan equation is significantly simplified. In terms of variables
1
= (y p),
2
it takes the form:
1
= (y + p)
2
(43)
s(, ) = s 2 (, ).
(44)
In accordance with the fact that this is a degenerated Riccati equation, this equation has the
KowalewskayaPainlev property [25], and its solution is a simple meromorphic function
1
.
()
s(, ) =
(45)
Returning to the -representation, we obtain the following result:

i
S(, y) =
i
dp p
1
e
,
2i
(s(p + y))1 y
(46)
where s(p) is the Laplace transform of the S-matrix at the initial moment y = 0

s(p) =
d ep S(, 0).
(47)
Assuming the colour transparency property in the form

lim S(, 0) = 1 c + ,
> 0,
c > 0,
(48)
one can calculate the function s(p, y) at large y or p

s(p, y)
1
.
p + c(1 + )(p + y)1
(49)
Therefore, if 0 < < 1, we obtain in the region p y 1 y the scaling behaviour

s(p, y)
1
,
p + c(1 + )y 1
(50)
which leads to the S-matrix

S(, y)
ec(1+ )y
(51)
in the region
1. It means, that at large y the distribution in is proportional to a
smeared -function ().
y 1
264
If, on the other hand, > 1 (for example, when lim0 S(, 0) = 1 c 2 ) the Smatrix tends to unity in the large region y 1 , which corresponds to the vanishing of
the dipole interaction. In the particular case when = 1 and
lim S(, 0) = 1 c + ,
c > 0,
(52)
the asymptotic behaviour of the S-matrix at large y is especially simple:

lim S(, y) = ec .
(53)
It corresponds to the fact that in the (2 + 1)-dimensional QCD the fan equation has the
stationary solutions
S(, y) = ec ,
(54)
corresponding to the conservation of the total length of dipoles. This expression is similar
to the Maxwell distribution in energies for molecules in a thermal equilibrium. In our case
the energy is proportional to the total length of dipoles. Thus, there is an analogy between
the Boltzman and BK equations.
5. Stationary solutions of the fan equation

The stationary fan equation
2
d0 (S2 0 S0 1 S2 1 )
0=
(55)
has the following solution

S(2 )
S2 1 =
,
S(1 )
(56)
where S() is an arbitrary function. In a linear approximation it corresponds to the amplitude N(b, , y) (41) for the BFKL equation, being the sum of two functions depending
on 1 and 2 . In order to show that S2 1 is a general solution, we expand the equation in
terms of small fluctuations S2 1 around the solution
S2 1 = S2 1 S2 1 ,
(57)
and we obtain, in the linear approximation,

2
0=

d0 S2 0 S0 1 + S2 0 S0 1 S2 1 .
(58)
2 1
By introducing the relative correction S
2 1 ,
S2 1 = S2 1 S
(59)
265
we simplify the above equation

2
0=

0 1 + S
2 0 S
2 1 .
d0 S
(60)
There is an obvious solution of this equation

2 1 = s(2 ) s(1 ),
S
(61)
where s() is an arbitrary function. This correction, however, represents only a redefinition
of the function S() in the expression for S2 1 . In fact, by introducing the new variables
= 2 1 , b = (2 + 1 )/2 we can reduce the above equation for S2 1 to the stationary
2 1 , as it was shown in Section 3.
equation for N(b, , y), which has only the solutions S
The stationary fan equation has the KowalewskayaPainlev property. Namely, if we
the solution in the series

S2 1 =
(62)
an (b) n , a0 = 1,
n=0
the first coefficient a1 (b) is an arbitrary function of with the following asymptotic behaviour at b

cr
.
a1 (b) =
(63)
br
r=1
Indeed, the above solution of the stationary equation can be written as follows:

rk
1+

2(b+d
)
k
S2 1 =
,
1 2(b+d
k)
(64)
k=1
where dk and rk are arbitrary complex parameters. Therefore, the above function a1 (b) is
a1 (b) =

k=1
rk
.
b + dk
(65)
In contrast to this, as a consequence of the KowalewskayaPainlev property, for a given

a1 (b) the solution S2 1 is obtained in an unique way. For example, if we start from the
scaling ansatz
a
a1 (b) = ,
(66)
b
the stationary solution is
a

b + 2
a3 3
a2 2
a
+
S2 1 =
(67)
1+a +
+
+ O 4 .
2
3
b 2
b
2 b
12
6 b
We can also write solutions of the stationary equation which do not have any singularities
at real 1,2 . For example,
S2 1 =
1 + 12
1 + 22
(68)
266
6. Scaling solution of the non-stationary equation

Let us now consider the general case when s(b, p, y) depends on b, and let us introduce
the new variables
1
1
= (y + p),
(69)
= () (y p),
2
2
where () is an arbitrary function which later on will be derived from the requirement
that s(b, p, y) satisfies initial conditions at y = 0. In terms of the new variables the fan
equation looks as follows:

S(b, , ) = lim N S b + 14 2 + 1 + (2 ) 2 , 1 , 1
r ,

r

(70)
S b 14 1 + 1 + (1 ) 1 , 2 , 2 1.
Its solution can be searched in the form of an expansion in powers of 1/:

, ),
, ) = 1 +
r S(b,
S(b,

r=1
, ) =
r S(b,
cr (b, )
.
r+1
(71)
Here c1 (b, ) turns out to be an arbitrary function, fixed by initial conditions for s(b, p, y)
at y = 0. According to the fan equation the residues cr (b, ), near b = , are analytic
functions with the following asymptotic behaviour:
cr (b, ) =
1 cr,s ()
.
br
bs
(72)
s=0
In particular, for the first correction 1 S(b, , ) to the pole 1/ we obtain the differential
equation

(73)
1 S(b, , ) = 0.

Its solution is
c1 (b, )
,
2
(74)
1 c1,s ()
b
bs
(75)
1 S(b, , ) =
where
c1 (b, ) =
s=0
is an arbitrary function which is analytic and vanishing at b . It should be chosen to

satisfy the initial conditions for s(b, , y) at y = 0.
Therefore, in the expansion of S(b, , ) in powers of 1/, there is a phenomenon
similar to the KowalewskayaPainlev property for the integrable differential equations.
Namely, one can construct the general solution in the class of meromorphic functions by
fixing the functions () and c1 (b, ) from the initial conditions for s(b, p, y). The next
267
correction 2 S(b, , ) satisfies the equation

2
1
2 S(b, , ) = 4 N c1 b + 14 2 , 1 c1 b 14 1 , 2 = =

2
1

1
c1 b + 14 2 , 1 + (2 ) = , (76)
2
2 b
where, in the last term, it is implied that after the differentiation of c1 the removed operator
b + 14 2 is substituted by the expression 1 + (2 ) in the corresponding place. Therefore,
we obtain
2 S(b, , ) =
c2 (b, )
,
3
(77)
where

c2 (b, ) = N c1 b +

= =
2 1

1
c1 b + 14 2 , 1 + (2 )
.
2 b
2 =
1
4 2 , 1

c1 b
1
4 1 , 2
(78)
In an analogous way one can calculate the residues cr (b, ) in terms of () and ck (b, )
with k < r.
Let us now consider the solution of the fan equation in the scaling regime
1,
b 1,
b 1,
(79)
for fixed . In this case the equation (70) is significantly simplified

S(b, ) = lim N S b + u 2 , 1 S b u 1 , 2 1,
r

(80)
where the parameter u = u() is defined by the expression

1
1 + () ,
4
and the solution has the following expansion at large b
u=
(81)
v
1
+ 2 + .
(82)
b
Here the function v = v() is fixed by the initial conditions for s(b, p, y). The first two
terms on the rhs of (82) determine all the other coefficients of the expansion
S(b, ) =
S(b, ) =
1 an
,

(b)n
a0 = 1,
a1 = v.
(83)
n=0
The recurrence relations for the coefficients an are given in Appendix A, and one obtains
S(b, ) =
1

v
v2
1
+ 2 + 3 2 + 2u2 v + v 3 4 3 + .
b b
b
(84)
268
It is easy to see that the substitution

1
s b, 2u , 2up
(85)
2u
turns (80) into (20) for the stationary (y-independent) case. Therefore, we have the remarkable result: the general solution of the non-stationary equation at b 1 can be expressed
in terms of the stationary solution depending on /b. Using the stationary solution obtained
in the previous section (see (67)) one can derive

v

b + 2 2u

1
2u
d e
.
S(b, ) =
(86)
2u
b 2
S(b, )
7. Infinite parameter symmetry of the fan equation

In this section we consider another ansatz:
R(2 )
,
S2 1 (y) = S2 1 (y) = e((2 )(1 ))y
T (1 )
(87)
where R() and T () are arbitrary functions. Putting this ansatz into Eq. (7) it is easy to
verify that it is indeed a non-stationary solution, provided that

() =
d0
R(0 )
1 .
T (0 )
(88)
For small fluctuations N2 1 (y) around this solution

S2 1 (y)
S2 1 (y) 1 + N2 1 (y)
(89)
we obtain, inserting (89) into (7) and linearizing the result in N2 1 (y),
N2 1 (y)
=
y
2
d0
1

R(0 )
N2 0 (y) + N0 1 (y) N2 1 (y) .
T (0 )
We have zero mode solutions of this equation

2 1 (y) = R(2 ) T (1 ) + y (2 ) (1 ) ,
N
(90)
(91)
where R() and T () are arbitrary functions related to () as follows:

() =
d0

R(0 )
R(0 ) T (0 ) .
T (0 )
(92)
Introducing the new integration variable ()

() =
d0
R(0 )
,
T (0 )
(93)
269
we obtain the following linear equation for N2 1 (y) N2 1 (y):

N2 1 (y)
=
y
2

d0 N2 0 (y) + N0 1 (y) N2 1 (y) ,
(94)
which coincides with the BFKL equation having the general solution given by (33) with
the substitution
2 + 1
= 21 2 1 ,
b
,
r (r ).
2
Let us factorize S2 1 (y) as follows:
S2 1 (y)
S2 1 (y)s2 1 (y).
(95)
Then we obtain the following equation for s2 1 (y):

s2 1 (y)
=
y
2
d0
1

R(0 )
s2 0 (y)s0 1 (y) s2 1 (y) .
T (0 )
(96)
Therefore, the fan equation has a renormalization group property. Namely, s2 1 (y) satisfies the same equation as S2 1 (y), but in the new variables , defined by (93). As a
result, we obtain the following Backlund transformation between different solutions of
the fan equation
(y) =
e(2 1 )y S(1)
2 1
R12 (2 ) (2 1 )y (2)
e
S2 1 (y)
T12(1 )
(97)
for arbitrary R() and T (), provided that the new variable satisfies

() =
d0
R12 (0 )
.
T12 (0 )
(98)
The function () is chosen to grow with and to have the additional constraint
() = .
(99)
We can once more perform the above transformation:

e(2 1 )y S(2)
(y) =
2 1
R23 (2 ) (2 1 )y (3)
e
S2 1 (y),
T23 (1 )
r = (r ),
(100)
where

( ) =
d0
R23 (0 )
,
T23 (0 )
(101)
and verify its group property

e(2 1 )y S(1)
(y) =
2 1
R13 (2 ) (2 1 )y (3)
e
S2 1 (y),
T13(1 )

r = (r ) .
(102)
270
Here
R13 (2 ) R12 (2 ) R23 (2 )
=
,
T13 (1 )
T12(1 ) T23 (1 )
()

R23 (0 )
R12 (0 ) R23 ( (0 ))
= d0
.
() =
d0
T23 (0 )
T12 (0 ) T23 ( (0 ))
In particular, for

() =
(103)
(104)
(105)
one obtains
R12 () R23 ( ())
= 1.
T12 () T23( ())
(106)
Generally,
R13 (2 ) R12 (2 ) R23 (2 )
=
= 1,
T13 (1 )
T12(1 ) T23 (1 )
(107)
and we thus obtain a family of solutions

S(f2 )1 (y) =
f (2 )
S (y)
f (1 ) 2 1
(108)
for arbitrary f (). The symmetry group of the fan equation is a product of this abelian
transformation of the dipole S-matrix S2 1 (y) and of the reparametrization group (98).
Without any loss of generality it is convenient to write the action of the second group
on S2 1 (y) in the form
(y) = a(2 )a(1 )e(2 1 )y S2 1 (y),
e(2 1 )y S(a)
2 1
r
r = (r ) =
d a 2 (),
(109)
where a() = R12 ()/T12 (). We can express this equation directly in terms of the
transformation = ()

d (2 ) d (1 ) (2 1 )y (2)
(y) =
e
S2 1 (y).
e(2 1 )y S(1)
(110)
2 1
d2
d1
It means that under the general covariance transformation () the quantity
T2 1 (y) = e(2 1 )y S2 1 (y)
(111)
transforms as a tensor with the spinor indices 2 , 1 . The functional expressing the Smatrix S2 1 (y) through initial conditions

S2 1 (y) = S2 1 (0)
(112)
should be invariant under the full symmetry group (together with the above multiplication
of S(1)
2 1 by the ratio f (2 )/f (1 )). In particular, we can use the above transformations to
271
change in a convenient way the initial conditions

S(1)
(0) =
2 1
R(2 ) (2)
(0).
S
T (1 ) 2 1
(113)
8. General solution of the fan equation

The fan equation for the function
G2 1 (y) = e(2 1 )y S2 1 (y)
(114)
can be written as follows

G2 1 (y)
=
y
2
d0 G2 0 (y)G0 1 (y).
(115)
The function G2 1 (y) can be considered as a matrix element of the infinite-dimensional

matrix G(u) having a triangular form, in accordance with the fact that it is nonzero only
for 2 1
G2 1 (y) = 2 |G|1 .
(116)
We can define the product of two matrices Ga (u) and Gb (u) as the matrix Gc (u) with the
matrix element
2
Gc2 1 (y) =
d0 Ga2 0 (y)Gb0 1 (y).
(117)
Correspondingly, the inverse matrix G1 (u) is defined as the matrix satisfying the relations
2
(2 1 ) =

d0 G2 0 (y) G1
0 1
2
(y) =

d0 G1
2 0
(y)G0 1 (y). (118)
Its matrix elements (G1 )2 1 (y) are non-zero only for 2 1 . Thus, the operators G(y)
belong to the (solvable) group of triangular matrices. The matrix element of the unit element of this group is (2 1 ).
In the operator form the fan equation looks as follows:
G(y)
= G2 (y),
y
(119)
and its general solution has the meromorphic property of KowalewskayaPainlev [25]
G(y) =
G(0)
.
1 yG(0)
(120)
272
Note that in accordance with the Riccati theory the operator G1 satisfies the free Newton
equation
d 2 1
G (y) = 0.
dy 2
(121)
The corresponding solution for the S-matrix S2 1 (y) written in the operator form is
S(y) = ey
G(0)
ey ,
1 yG(0)
(122)
where is the operator with the eigenvalues 1 for the eigenfunctions |1 . This solution is
similar to the general solution of a linear equation expressed in terms of its Green function.
In a matrix form it can be expanded in a series as follows:
e
(2 1 )y
2
S2 1 (y) = S2 1 (0) + y
d0 S2 0 (0)S0 1 (0)
1
2
+y
0
d0 S2 0 (0)S0 0 (0)S0 1 (0) + .
d0
1
(123)
In particular, according to above discussions one can simplify this general result as follows:
S2 1 (y) = e
((2 )(1 ))y R(2 )
T (1 )

() =
d0
R(0 )
1 ,
T (0 )
(124)
providing that S2 1 (0) has the factorized form

R(2 )
.
S2 1 (0) =
T (1 )
(125)
It is also obvious that the solutions S2 1 (y) for different S2 1 (0) are related each to other
by the above established group of transformations depending on two arbitrary functions
f () and ()

f (2 ) d (2 ) d (1 ) (2 1 )y
(2 1 )y (f, )
e
S2 1 (y) =
e
S2 1 (y),
f (1 )
d2
d1
r = (r ).
(126)
In particular, we can consider the case where the initial condition S2 1 (0) is only a
function of 2 1 , which gives us a possibility to construct the solution depending on
three arbitrary functions
(y) =
e(2 1 )y S(f,,s)
2 1
+i
(2)
(1)
dp c2 (p)c1 (p)
,
2i (s(p))1 y
(127)
273
where the parameter is chosen from the condition that all singularities of the integrand
are situated to the left of the integration contour in p and

s(p) =
0
c(2)
(p) =
2
d ep S/2,/2 (0),

d (2 ) 2 p
e f (2 ),
d2
c(1)
(p) =
1
1
d (1 ) 1 p
e f (1 ) .
d1
(128)
The solution S (f,,s) can be considered as a particular case of the general solution transformed to a diagonal form by an appropriate similarity transformation. The function s(p)
has the interpretation of the eigenvalue of S2 1 (0), and c(2)
1 (p) is its eigenfunction. The
similarity transformation is unitary (c(2) = c(1)+ = 1/c(1) ) provided that f () is a phase:
f = f 1 .
In general, the functions c(r)(p) have a more complicated (non-exponential) dependence on p which enumerates the eigenvalues. This dependence can be found from the
solution of the equation

d1 S2 1 (0)c1 (p) = s(p)c2 (p).
(129)
For the triangular matrix S2 1 (0) the eigenfunction c (p) can be calculated easily,
provided that we substitute the continuous coordinate by a finite number of points k . In
general, the behaviour of S2 1 (y) at large y depends on the initial conditions encoded in
the asymptotics of the functions c (p) at large p. Due to the colour transparency property
we have
S2 1 (0) < 1,
lim S2 1 (0) = 1 c21 ,
2 1
> 0,
(130)
where the coefficient c > 0 is a function of b. At large y and fixed r it is natural to expect
an universal dependence of S2 1 (y) from y. In particular, for large c the S-matrix should
have the form of a smeared (21 )-function. It is related to the fact that in this case, as a
result of the y-evolution, the average size of the dipole tends to zero, and one can neglect
the b-dependence of S (up to a common factor). In the pre-asymptotics a scaling behaviour
of S2 1 (y) is possible (cf. Section 4).
It is important that the solution of the BalitskyKovchegov equation satisfies also the
linear integral equation
2
G2 1 (y) = G2 1 (0) + y
d G2 0 (0)G0 1 (y).
(131)
In particular, the theory of the Fredholm equations can be applied to this equation. In this
case, the perturbation theory in y should work rather well. For large y one can use a quasiclassical approximation. We hope to return to the investigation of the this linear equation
in future publications.
274
9. Conclusions
In this paper we have obtained analytic solutions of a simplified version of the nonlinear
BK equation, namely the BK equation in one transverse dimension. Our explicit formulae
allow to study, for any initial condition, the rapidity evolution, both as a function of the
dipole size and of the dipole position in impact parameter.
We also have found several remarkable properties of the nonlinear equation: the
general solution has the meromorphic KowalewskayaPainlev property, and the equation has symmetries which are reminiscent of general covariance. Some of these results
might be helpful in addressing also the more realistic case of two transverse dimensions.
Acknowledgements
This work was started while two of us (V.S. Fadin and L.N. Lipatov) were visiting
the II Institut fr Theoretische Physik, Universitt Hamburg, and DESY. We gratefully
acknowledge their hospitality.
Appendix A
Here we derive the recurrence relations for the coefficients an in (83). Inserting the
expansion (83) in (80) one can obtain the relation

1 an
u d k 1
al
1 ak
(n + 1) 2
1+
=

(b)n 2
(b)k (b)l
b dx
x l+1
n=0
k=0 l=0
x=1

l

1
u d
1
.
k+1

b dx
x
x=1
Using the expansion

d
1 + z dx

k l+1
x
1

r=0
x=1

r=0
where
r
Ck,l
=
(k + r)(l + r)
,
(r + 1)(k)(l)
(k + r)
(r + 1)(k)
r
Ck,l+1
zr ,

d r 1
z

dx x l+1
x=1
275
one can easily verify that a0 = 1, and a1 is arbitrary. For the coefficients an with n > 1 we
obtain the recurrence relation
(n 1)an =
n1

nk nk
u
ak Ck,1
+ (u)nk
k=1
n m1

m=2 k=1
ak amk unm
nm

nml
l
(1)l Ck,mk+1
Cmk,k+1
.
l=0
Evidently, the terms with odd powers of u are cancelled.
References
[1] L.N. Lipatov, Sov. J. Nucl. Phys. 23 (1976) 338;
V.S. Fadin, E.A. Kuraev, L.N. Lipatov, Phys. Lett. B 60 (1975) 50;
E.A. Kuraev, L.N. Lipatov, V.S. Fadin, Zh. Eksp. Teor. Fiz. 71 (1976) 840 (in Russian), Sov. Phys. JETP 44
(1976) 443;
E.A. Kuraev, L.N. Lipatov, V.S. Fadin, Zh. Eksp. Teor. Fiz. 72 (1977) 377 (in Russian), Sov. Phys. JETP 45
(1977) 199;
Ya.Ya. Balitskii, L.N. Lipatov, Sov. J. Nucl. Phys. 28 (1978) 822.
[2] L.N. Lipatov, V.S. Fadin, Zh. Eksp. Teor. Fiz. Pisma 49 (1989) 311 (in Russian), Sov. Phys. JETP Lett. 49
(1989) 352;
L.N. Lipatov, V.S. Fadin, Yad. Fiz. 50 (1989) 1141 (in Russian), Sov. J. Nucl. Phys. 50 (1989) 712.
[3] V.S. Fadin, L.N. Lipatov, Phys. Lett. B 429 (1998) 127;
M. Ciafaloni, G. Camici, Phys. Lett. B 430 (1998) 349.
[4] S.J. Brodsky, V.S. Fadin, V.T. Kim, L.N. Lipatov, G.B. Pivovarov, Zh. Eksp. Teor. Fiz. Pisma 70 (1999)
161 (in Russian).
[5] L.N. Lipatov, Sov. Phys. JETP 63 (1986) 904.
[6] G.P. Salam, JHEP 9807 (1998) 19;
M. Ciafaloni, D. Colferai, Phys. Lett. B 452 (1999) 372;
M. Ciafaloni, D. Colferai, G.P. Salam, Phys. Rev. D 60 (1999) 114036;
G. Altarelli, R.D. Ball, S. Forte, Nucl. Phys. B 575 (2000) 313;
G. Altarelli, R.D. Ball, S. Forte, Nucl. Phys. B 599 (2001) 383;
G. Altarelli, R.D. Ball, S. Forte, Nucl. Phys. B 621 (2002) 359.
[7] J. Bartels, Nucl. Phys. B 175 (1980) 365;
J. Kwiecinski, M. Praszalovicz, Phys. Lett. B 94 (1980) 413.
[8] L.N. Lipatov, Nucl. Phys. B 452 (1995) 369.
[9] J. Bartels, C. Ewerz, JHEP 9909 (1999) 026, hep-ph/9908454.
[10] L.V. Gribov, E.M. Levin, M.G. Ryskin, Phys. Rep. 100 (1983) 1.
[11] J. Bartels, M. Wsthoff, Z. Phys. C 66 (1995) 157.
[12] J. Bartels, L.N. Lipatov, M. Wsthoff, Nucl. Phys. B 464 (1996) 298.
[13] M. Braun, G.P. Vacca, Eur. Phys. J. C 6 (1999) 147.
[14] J. Bartels, L.N. Lipatov, G.P. Vacca, hep-ph/0404110.
[15] N.N. Nikolaev, B.G. Zakharov, Z. Phys. C 49 (1991) 607;
N.N. Nikolaev, B.G. Zakharov, Z. Phys C 53 (1992) 331;
N.N. Nikolaev, B.G. Zakharov, Z. Phys. C 64 (1994) 651;
N.N. Nikolaev, B.G. Zakharov, JETP 78 (1994) 598;
A.H. Mueller, Nucl. Phys. B 415 (1994) 373;
A.H. Mueller, B. Patel, Nucl. Phys. B 425 (1994) 471;
A.H. Mueller, Nucl. Phys. B 437 (1995) 107.
[16] Ia. Balitsky, Nucl. Phys. B 463 (1996) 99.
276
[17] Yu. Kovchegov, Phys. Rev. D 60 (1999) 034008.

[18] L. McLerran, R. Venugopalan, Phys. Rev. D 49 (1994) 2233;
L. McLerran, R. Venugopalan, Phys. Rev. D 49 (1994) 3352;
L. McLerran, R. Venugopalan, Phys. Rev. D 50 (1994) 2225;
E. Iancu, A. Leonidov, L. McLerran, Phys. Lett. B 510 (2001) 133;
E. Iancu, A. Leonidov, L. McLerran, Nucl. Phys. A 692 (2001) 583;
E. Ferreiro, E. Iancu, A. Leonidov, L. McLerran, Nucl. Phys. A 701 (2002) 489;
J. Jalilian-Marian, A. Kovner, A. Leonidov, H. Weigert, Nucl. Phys. B 504 (1997) 415;
J. Jalilian-Marian, A. Kovner, A. Leonidov, H. Weigert, Phys. Rev. D 59 (1999) 014014;
For a recent review, see: E. Iancu, R. Venugopalan, hep-ph/0303204.
[19] J.P. Blaizot, E. Iancu, H. Weigert, Nucl. Phys. A 713 (2003) 441.
[20] M. Braun, Eur. Phys. J. C 16 (2000) 337;
K. Golec-Biernat, L. Motyka, A.M. Stasto, Phys. Rev. D 65 (2002) 074037;
J.L. Albacete, N. Armesto, A. Kovner, C.A. Salgado, U.A. Wiedemann, Phys. Rev. Lett. 92 (2004) 082001,
hep-ph/0307179.
[21] K. Golec-Biernat, A.M. Stasto, Nucl. Phys. B 668 (2003) 345, hep-ph/0306279;
A.M. Stasto, Acta Phys. Pol. B 35 (2004) 261.
[22] E. Levin, K. Tuchin, Nucl. Phys. A 691 (2001) 779, hep-ph/0012167;
S. Munier, R. Peschanski, Phys. Rev. Lett. 91 (2003) 232001;
S. Munier, R. Peschanski, Phys. Rev. D 69 (2004) 034008, hep-ph/0401215.
[23] G.M. Cicuta, G. Marchesini, E. Montaldi, Phys. Lett. B 96 (1980) 141;
L. N. Lipatov, L. Szymanovski, Preprint of the Warsaw University IBJ 11 (7) (1980) (unpublished);
M. Li, Ch.-I Tan, Phys. Rev. D 50 (1994) 1140;
M. Li, Ch.-I Tan, Phys. Rev. D 51 (1995) 3287;
D.Y. Ivanov, R. Kirschner, E.M. Levin, L.N. Lipatov, L. Szymanowski, M. Wusthoff, Phys. Rev. D 58 (1998)
074010, hep-ph/9804443.
[24] H. de Vega, L.N. Lipatov, Phys. Lett. B 578 (2004).
[25] S.V. Kowalewskaya, Acta Math. 12 (1889) 177;
S.V. Kowalewskaya, Acta Math. 14 (1890) 81;
P. Painlev, Acta Math. 25 (1902) 1.

Master integrals with 2 and 3 massive propagators

for the 2-loop electroweak form factorplanar case
U. Aglietti a , R. Bonciani b,1
a Dipartimento di Fisica Universit di Roma La Sapienza and INFN, Sezione di Roma, I-00185 Roma, Italy
b Physikalisches Institut, Albert-Ludwigs-Universitt Freiburg, D-79104 Freiburg, Germany
Received 26 January 2004; received in revised form 30 June 2004; accepted 20 July 2004
Abstract
We compute the master integrals containing 2 and 3 massive propagators entering the planar
amplitudes of the 2-loop electroweak form factor. By this me mean the process f f X, where
f f is an on-shell massless fermion pair and X is a singlet under the electroweak gauge group
SU(2)L U (1)Y . This work is a continuation of our previous evaluation of master integrals containing at most 1 massive propagator. The masses of the W , Z and Higgs bosons are assumed to
be degenerate. The 1/ poles and the finite parts are computed exactly in terms of a new class of
1-dimensional harmonic polylogarithms of the variable x = s/m2 , with = 2 D/2, D the space
time dimension and s the center-of-mass energy squared. Since thresholds and pseudothresholds in
s = 4m2 do appear in addition to the old ones in s = 0, m2 , an extension of the basis function set
involving complex constants and radicals is introduced, together with a set of recursion equations to
reduce integrals with semi-integer powers. It is shown that the basic properties of the harmonic polylogarithms are maintained by the generalization. We derive small-momentum expansions |s| m2
of all the 6-denominator amplitudes. Comparison with previous results in the literature is performed
finding complete agreement.
PACS: 11.15.Bt; 12.15.Lk
E-mail addresses: ugo.aglietti@roma1.infn.it (U. Aglietti), roberto.bonciani@physik.uni-freiburg.de

(R. Bonciani).
1 This work was supported by the European Union under contract HPRN-CT-2000-00149.
doi:10.1016/j.nuclphysb.2004.07.018
278
U. Aglietti, R. Bonciani / Nuclear Physics B 698 (2004) 277318
Keywords: Feynman diagrams; Multi-loop calculations; Vertex diagrams; Harmonic polylogarithms
1. Introduction
We present in this paper the computation of the master integrals containing 2 and 3
massive propagators entering the planar amplitudes of the 2-loop electroweak form factor.
The model process we consider is
f (p1 ) + f(p2 ) X(q),
(1)
where f f is an on-shell massless fermion pair, p12 = p22 = 0, and X is a singlet under the
electroweak gauge group SU(2)L U (1)Y . This work is the natural continuation of our
previous evaluation of master integrals containing at most 1 massive propagator. At the
2-loop level, the annihilation in Eq. (1) involves the emission of 2 virtual bosons among
, W , Z and H s. The cases of emission of (i) 2 photons and (ii) 1 photon and 1 massive
boson, have already been treated in our previous work [1]. Since the above amplitudes have
only thresholds in s = 0, m2 and pseudothreshold in s = m2 , we succeeded to express
them in terms of 1-dimensional harmonic polylogarithms [2,3] with maximum weight up
to 4 included.
In this paper we consider the amplitudes with 2 or 3 bosons of mass m exchanged: the
masses of W , Z and Higgs bosons are assumed to be degenerate, mW mZ mH m.2
In general, these amplitudes have thresholds in s = 0, m2 , 4m2 and pseudothresholds in
s = m2 , 4m2 . As a consequence of this more complicated structure, the basis function
set of the harmonic polylogarithms considered in [2] is no more sufficient
and a gen
eralization is presented. New basis functions such as 1/(4 x), 1/ x(4 x), etc., are
introduced and recursion relations to reduce integrals with semi-integer powers coming
from the evaluation of the master integrals are derived. The basic properties of the harmonic polylogarithms, such as the uniqueness of representation as repeated integration,
the algebra structure, the closure under the inverse transformation x 1/x, etc., are all
maintained.
We use dimensional regularization [4] to regulate both ultraviolet and infrared divergences, which then appear as poles in 1/, with = 2 D/2 and D the spacetime
dimension.
The paper is organized as follows.
In Section 2 we briefly review the method of the differential equations for the analytical
evaluation of the master integrals. Specific properties of the present computations are also
2 Small corrections of order g 4 (m2
2 n
Z,H mW ) can be included by means of expansions of the denominators
of the form
1
k 2 + m2Z,H
m2Z,H m2W
+ .
k 2 + m2W
(k 2 + m2W )2
1
(2)
The amplitudes with powers of the W denominators can be reduced to the master integrals by means of the
integration-by-parts identities.
279
discussed. The whole strategy for the analytical evaluation of the 2-loop electro-weak form
factor has been discussed in our previous work [1], to which we refer for details.
In Section 3 we discuss the properties of the harmonic polylogarithms. An overview of
the ordinary harmonic polylogarithms is presented, in order to introduce later the generalization necessary to represent the master integrals. As already anticipated, the basic tool
is a set of recursion equations to reduce the integrals coming from the evaluation of the
master integrals to a unique form, specified by the choice of the basis functions.
Section 4 contains the main results of our work. We present exact analytical results for
the 21 non-trivial master integrals containing 2 and 3 massive denominators. All the master
integrals are represented in terms of generalized harmonic polylogarithms.
In Section 5 the small momentum expansion |s| m2 of all the 6-denominator amplitudes is presented. We compare our results with those present in the literature finding
complete agreement.
In Section 6 we draw our conclusions and we discuss the open problems in the computation of the master integrals.
Finally, in Appendix A, we report on the 1-loop master integrals with 2 massive propagators which occur in our computation. They are written in terms of generalized harmonic
polylogarithms, like the 2-loop amplitudes.
2. The computation of the master integrals

The planar diagrams containing 2 and 3 massive propagators entering the 2-loop electroweak form factor are plotted in Figs. 1 and 2. We have omitted 1-particle-reducible
contributions, related to field and mass renormalization of the external particles f , f
and X. The graphical conventions in use are the following. A massless propagator is represented by a wavy internal line, while a massive propagator is denoted by a straight segment.
An external line carrying the light-cone momentum p1 or p2 is represented by an external wavy line, while the external particle X carrying the general momentum q is depicted
as a straight external line.3 The amplitudes can be obtained, for example, by means of a
DysonSchwinger expansion of the ff X vertex function [5].
Any Feynman diagram of Figs. 1 and 2 can be decomposed, on the basis of his Lorentz
structure, in form factors containing only scalar integrals. The calculation of the form factors, therefore, reduces to the calculation of these scalar integrals. By means of the so-called
Laporta algorithm [6], using integration-by-parts identities [7] and eventual symmetry relations [8,9], one can reduce the calculation of all the scalar integrals to a small set of
independent ones called the master integrals (MIs) of the problem (see [1]). All the possible resulting diagrams of the above-mentioned reduction are shown in Figs. 36, while
the MIs are shown in Fig. 7. Let us note that the amplitudes computed in our previous
work [1], containing 0 or 1 massive denominators, appear in the decomposition of the
present amplitudes as well, but they are not explicitly shown in the present paper.
3 This set of conventions is clearly a generalization of the well-known QED ones.
280
Fig. 1. Vertex-correction diagrams with 2 and 3 massive propagators. The topologies related to these diagrams
are real 6-denominator topologies (see text).
Once the MIs are found, the problem of their computation arises. Let us sketch in this
Section how to explicitly evaluate the MIs with the differential equation method [1014].
We may identify 2 basic steps:
(1) Generation of the differential equations.
This step is rather automatic in the sense that it does not offer specific difficulties. It
involves, in turn, the following steps:
(a) We take the derivatives of the MIs with respect to x at fixed a by differentiating
the integrand with respect to the external momenta p1 or p2 ; the derivatives of the
MIs are then expressed as a superposition of linearly-dependent scalar amplitudes;
(b) We reduce the independent amplitudes to MIs according to the previous discussion. This way the system closes on the MIs themselves. Linear systems of first-
281
Fig. 2. Self-energy correction diagrams with 2 and 3 massive propagators. The topologies related to these diagrams are 5 and 4 denominator topologies (see text).
order differential equations with variable coefficients are then generated, whose
unknowns are the MIs themselves.
(2) Solution of the differential equations.
In order to have a unique solutionas it shouldinitial conditions can often be obtained by studying the behaviour of the MIs close to thresholds or pseudothresholds.
The system of differential equations is then solved by recursion in . Let us remark
that this step is not automatic, in the sense that obtaining a solution is in some cases
a matter of intuition and experience.
Let us make a few comments about the specific characteristics of the present computation.
The most complicated cases involve topologies with more than 1 MI; we label with t the
282
Fig. 3. The set of 7 independent 6-denominator diagrams, with 2 and 3 massive propagators.
number of different denominators appearing in the scalar integrals (t is called the topology
number of the integral):
t = 4: we encounter 2 vertex topologies having 2 MIs, which we take as (q) and (r) in
Fig. 7 for the first topology and (s) and (t) for the second one;
t = 5: we encounter 2 vertex topologies having 2 MIs, which we take as (g) and (h)
for the first one and (i) and (j ) for the second one. There is also 1 topology having 3
MIs,4 ,5 which we take as (l), (m) and (n).
A convenient basis for topologies with 2 MIs consists of the following amplitudes:
(1) the basic amplitude, i.e., the amplitude with unitary numerator and with all the denominators having unitary indices;
(2) the amplitude with unitary numerator and with one of the denominators squared, or
with an independent scalar product left on the numerator.
The resulting system of 2 differential equations is often triangular in D = 4, allowing for
a simple solution. For the only case of 3 MIs, we found that a convenient basis comprises
the following amplitudes:
(1) the basic amplitude defined above;
(2) the amplitude with unitary numerator and one denominator squared, as in the previous
case;
4 The t = 6 crossed ladder topology also has 3 MIs.
5 We have also used the symbolic method to reduce the topology with 3 MIs, with similar results.
283
Fig. 4. The set of 17 independent 5-denominator diagrams, with 2 and 3 massive propagators.
(3) the amplitude with an irreducible scalar product in the numerator and the denominators
all with unitary indices.
With this choice, the system of 3 ordinary differential equations becomes triangular in
D = 4, allowing for a simple recursive solution in .
3. Harmonic polylogarithms of one variable

As discussed in the introduction, amplitudes with 2 and 3 massive propagators may have
thresholds in s = 4m2 and pseudothresholds in s = 4m2 . As a consequence, a generalization of the usual harmonic polylogarithms of one variable is needed. In the next section
we summarize the salient features of the ordinary harmonic polylogarithms (HPLs) that
we want to preserve with the generalization, while in Section 3.2 we discuss such a generalization.
284
Fig. 5. The set of 11 independent 4-denominator diagrams, with 2 and 3 massive denominators.
Fig. 6. The set of 4 independent 3-denominator diagrams together with the single 2-denominator diagram, with 2
and 3 massive propagators.
3.1. Ordinary harmonic polylogarithms

The basic idea of the HPLs is that of representing a given integral introducing a minimal set of transcendental functions by making only linear transformations, such as partial
fractioning and integration by parts. Furthermore, the transcendental functions are defined
once and for all. Let us begin with a simple example. The integration of the power function
285
Fig. 7. The set of 25 MIs with 2 and 3 massive denominators. Out of them, 4 are the product of 1-loop MIs. The
dot on a line indicates a square of the corresponding denominator and the scalar product in parenthesis denotes
an irreducible numerator.
286
with an integer exponent n,

n+1
x
x
n

x dx = n+1
log x
1
n+1
if n = 1,
(3)
if n = 1,
is again a power function, except for the case n = 1 for which we have the logarithm.
Integration then introduces 1 transcendental function, plus the original set of elementary
functions. Larger sets of transcendental functions are needed to represent double integrals,
such as for instance
x
0
dx
log x =
1 x
x
dx
1 x
x
dx
= log x log(1 x) Li2 (x),
x
(4)
which involves a new transcendental functions, the well-known dilogarithm

x
Li2 (x)
dx
x
x
0
xn
dx
=
1 x
n2
for |x| 1.
(5)
n=1
However, many double integrals do not involve new transcendental functions with respect
to the ones needed to represent single integrals such as (3). Let us indeed consider
x
0
dx
log x =
(1 x )2
x
0
dx
(1 x )2
x
1

dx
1
1 log x.
= log(1 x) +
x
1x
(6)
The reduction above is done with an integration by parts: we integrate 1/(1 x)2 , whose
integral is elementary, and differentiate log x, obtaining a simpler, elementary function. We
then make a partial fractioning of
1
1
1
= +
,
x(1 x) x 1 x
(7)
i.e., another linear transformation. In general, whenever we have to integrate the product of
a transcendental function times an elementary function whose integral is also elementary,
we do a by-parts integration in order to derive, i.e., to simplify, the transcendental function.
This applies in particular to the integral of the transcendental function itself:
x
dx log x = x log x x
(8)
and in general to all the integrals of the form

x
0
dx
log x
(1 x )n
(9)
287
with n = 1. The examples given above generalize to multiple repeated integrations, of the
form
x
x3
x2
dxl
dx2
dx1
,
(10)
(xl + al )nl
(x2 + a2 )n2
(x1 + a1 )n1
with a1 al some constants. One has to introduce transcendental functions given by the
repeated integrations of the inverse linear functions 1/(x + ai ):
x
x3
x2
dxl
dx2
dx1
.
(11)
xl + al
x2 + a2
x1 + a1
For many applications, such as our previous computation of 2-loop electroweak amplitudes with 1 massive propagator [1], it is sufficient to introduce the following set of basis
functions:
1
g(0; x) = ,
(12)
x
1
,
g(1; x) =
(13)
1x
1
.
g(1; x) =
(14)
1+x
Note that g(0; x) has a non-integrable singularity for x 0, while the other functions are
finite in the same limit. The HPLs of weight 1 are defined as integrals of the basis functions:
x
H (0; x) =
dt
= log x,
t
(15)
dt
= log (1 x),
1t
(16)
x
H (1; x) =
0
x
H (1; x) =
0
dt
= log (1 + x).
1+t
(17)
Note the slight asymmetry related to the lower bound of integration for H (0; x).
Let us now define the general HPL H (w;
x), where w
is a vector with w components,
consisting of a sequence of 1, 0, and 1. The weight of a HPL is the number of its
indices, w, coinciding with the number of the repeated integrations. The HPLs of weight
w + 1 have the following integral recursive definition
x
H (a, w;
x) =
f (a; x )H (w,

x ),
(18)
except for the case with all the indices zero

1
H (0 w ; x) =
logw x,
w!
(19)
288
or, if a recursive definition is preferred

H (0, 0 w ; x) =
x
f (0; x )H (0 w , x ) =
1
logw+1 x,
(w + 1)!
(20)
where 0 w = (0, 0, . . . , 0) is a vector containing w zeroes.

3.2. Generalized harmonic polylogarithms
The basis functions for the GHPLs involve various kinds of extensions of the basis
functions defining the HPLs.
A trivial extension concerns functions involving different real constants
1
,
(21)
4+x
1
.
g(4; x) =
(22)
4x
The above functions are related to amplitudes with threshold/pseudothreshold at s =
4m2 , respectively, just as the old functions g(1; x) represent amplitudes with threshold/pseudothreshold in s = m2 . We will see in a moment however that the above extension in not sufficient.
To represent 3-point functions with 3 massive propagators one also needs to introduce
basis functions involving a complex constant6
g(4; x) =
g(c; x) =
g(c;
x) =
1
x
1
2
i
1
1
2
+i
3
2
(23)
3
2
(24)
Note that 12 i 23 = exp(i/3) are the non-trivial cubic roots of 1 and are the inverse
of each other.
The non-trivial extension, however, involves:
radicals of the form
1
,
x(4 + x)
1
g(r; x) =
.
x(4 x)
g(r; x) =
(25)
(26)
These functions also describe amplitudes with thresholds and pseudothresholds in s =

4m2 , respectively. It is indeed well known that amplitudes involving 2 particles with
6 An alternative representation uses only real functions at the price of introducing squares of the integration
variable: 1/(x 2 x + 1) and x/(x 2 x + 1).
289
the same mass m = 0 contain square roots of similar form as a phase-space reduction
effect;
products of radical with inverse linear functions7
1
,
g(1 r; x) =
x(4 + x)(1 + x)
1
g(1 r; x) =
,
x(4 + x)(1 x)
1
,
g(1 + r; x) =
x(4 x)(1 + x)
1
g(1 + r; x) =
.
x(4 x)(1 x)
(27)
(28)
(29)
(30)
In total, we have added 10 functions to the old basis. All the new
basis functions above are
finite in x = 0 or have at most an integrable singularity 1/ x for x 0.
The GHPLs of weight 1 are given by integrals of the basis functions and can be written,
in general, in terms of logarithms of complex functions of x:
x
H (4; x) =
0
x
H (4; x) =
0
dt
= log (4 + x) 2 log 2,
4+t
(31)
dt
= log (4 x) + 2 log 2,
4t
(32)
x
H (c; x) =
dt
t
1
2
3
2
= log(x 12 i 32) log(1/2 i 3/2),

x
dt
H (c;
x) =
1
t 2 + i 23
0
= log(x 1/2 + i 3/2) log(1/2 + i 3/2),

x
dt
x
= 2 arcsinh
H (r; x) =
2
t (4 + t)
(33)
(34)
= 2 log ( x + 4 + x ) 2 log 2,
(35)
7 Alternative basis functions having the same asymptotic limits 1/x for x of the previous ones can be
taken for instance as (4 x)/x/(1 x).
290

dt
x
H (r; x) =
= 2 arcsin
2
t (4 t)
0

4x +i x
= i log
,
4x i x

x
2
dt
3x
= arctan
H (1 r; x) =
4+x
t (4 + t)(1 + t)
3
0

i
4 + x i 3x
= log
,
3
4 + x + i 3x

x
1
dt
4 + x + 5x
H (1 r; x) =
,
= log
t (4 + t)(1 t)
5
4 + x 5x
x
(36)
(37)
(38)

2
dt
5x
H (1 + r; x) =
= arctan
4x
t (4 t)(1 + t)
5
0

4 x i 5x
i
= log
,
5
4 x + i 5x

x
dt
4 x + 3x
1
H (1 + r; x) =
= log
.
t (4 t)(1 t)
3
4 x 3x
x
(39)
(40)
Note that all the above integrals have zero as the lower limit of integration.
The GHPLs of general weight w > 1 are defined exactly in the same way as the
HPLs, according to Eq. (18), where now the components of w
can also take the values
4, c, c, r, 1 r.
Let us now discuss how the integrals appearing in the evaluation of the MIs can be
reduced to the GHPLs plus, of course, elementary functions. As far as the indices 4, c
and c are concerned, there is basically nothing new with respect to the HPLs: the only
difference is that individual terms in the expressions for the MIs are in general complex
and only their sum is, at it should, real.
When radicals, i.e., semi-integer powers are involved, the situation is more complicated.
The integral of a semi-integer power,

(x + a)n+1/2
(x + a)n1/2 dx =
(41)
(n is an integer),
n + 1/2
is always a semi-integer power, i.e., an elementary function. There is no need, then, to
introduce basis functions of this kind. On the other hand, integrals of products of radicals,
such as

I (, ) = (x + a) (x + b) dx,
(42)
291
with and half integers, involve in general new transcendental functions.8 By taking
a = 0 and b = 4 (or a = 4 and b = 0), we cover with I (, ) the cases related to the
basis functions with indices r. The idea is to shift the indices to reference values, such as
= 1/2 and = 1/2 in our (arbitrary) choice of the basis functions, by using recursive
relations.
Let us consider the general case in which both indices differ from their target values:
= 1/2 and = 1/2. The integrand of (42) is conveniently rewritten as
1
(x + a)n (x + b)l ,
(x + a) (x + b) =
(x + a)(x + b)
(44)
with n = + 1/2 and l = + 1/2 general integers.

We first make an algebraic reduction on the integrand to take one of the two indices n
and l to its target value zero. There are two possibilities
If 1 of the 2 indices is positive, let us say n > 0, we expand (x + a)n in powers of x + b
with the binomial formula
n

n
(a b)ns (x + b)s ,
(x + a)n =
(45)
s
s=0
so that the integrand takes the form

n

n
(x + a) (x + b) =
(a b)ns (x + a)1/2(x + b)s+l1/2
s
s=0
(if n > 0),
(46)
in which the index reached its target value.

If both indices are negative, n < 0 and l < 0, we do ordinary partial fractioning:
1
1
(x + a)|n| (x + b)|l|
A|n|
A|n|1
A1
=
+
+ +
|n|
|n|1
(x + a)
x +a
(x + a)
B|l|
B|l|1
B1
+
,
+
+ +
|l|
|l|1
(x + b)
(x + b)
x+b
(47)
so that
(x + a) (x + b)
= A|n| (x + a)1/2|n| (x + b)1/2 + A|n|1 (x + a)1/2|n| (x + b)1/2
8 On the contrary, products of integer powers do not require the introduction of new basis functions because
partial fractioning completely disentangles the factors:
An
A1
B1
Bk
1
1
=
+ +
+ +
+
.
(x + a)n (x + b)k
(x + a)n
x+a
x+b
(x + b)k
(43)
292
+ + A1 (x + a)3/2(x + b)1/2 + B|l| (x + a)1/2(x + b)1/2|l|

+ B|l|1 (x + a)1/2(x + b)1/2|l| + + B1 (x + a)1/2(x + b)3/2
(if n < 0 and l < 0).
(48)
In any of the above terms on the l.h.s. one of the two indices reached its target value.
To take also the second index to its target value, we use the integral identity
I (, ) =
(a b)
1
(x + a)+1(x + b)
I (, 1).
+ +1
+ +1
(49)
The above relation is obtained by doing an integration by parts on I (, ): we integrate

(x + a) and differentiate (x + b) with respect to x; we then simplify the ratio (x + a)/
(x + b) as 1 + (a b)/(x + b).
The first term on the r.h.s. is a finite one and is therefore known. The above equation can
then be used to lower the index by one unit. For example I (, 3/2) can be transformed
to a linear combination of I (, 1/2) and I (, 1/2), while I (, 1/2) can be reduced to
I (, 1/2). In general, by recursively using Eq. (49), one can reduce any < 1/2 to
= 1/2.9
By solving Eq. (49) with respect to the last term on the r.h.s. I (, 1), and shifting
+ 1, one obtains an identity to raise the index by one unit10
I (, ) =
+ +2
(x + a)+1(x + b)+1
I (, + 1).
( + 1)(a b)
( + 1)(a b)
(50)
By using Eqs. (49) and (50) we can then take the index to any desired reference value,
such as the value 1/2 of our basis. Then, we succeeded in reducing an integral of the
form (42) to H (r; x) plus terms containing elementary functions only.
In general, we encounter integrals containing both radicals and GHPLs, of the form

J (, ) = (x + a) (x + b) H (v, w;
(51)
x) dx,
where H (v, w;
x) is a GHPL whose first index v has been separated out for a reason that
will become clear soon. Since w
has w components, H (v, w;
x) has weight w + 1. As with
the previous integral, we can transform one of the two indices and , let us say , to its
target value = 1/2. The identity to reduce the second index is
J (, ) =
1
(x + a)+1(x + b) H (v, w;
x)
+ +1

1
x) dx
(x + a)+1 (x + b) g(v; x)H (w;
+ +1
(a b)
J (, 1).
+ +1
(52)
9 Since we can take = 1/2 and the equation is used for > 1/2, it holds that + > 1, so the
singularity in + = 1 is always avoided.
10 Note the singularity for = 1 is never reached since takes half-integer values only.
293
The second term on the r.h.s. involves the integration of a GHPL of smaller weight. We
then consider the last term as the only relevant one in the recursion. As before, the above
equation can be directly used to lower the index by one unit. Note that the indices inside
the GHPL are not touched by the reduction. Analogously to the previous case, by solving
Eq. (52) with respect to J (, 1) and sending + 1, one obtains an identity to
raise by one unit.
In some cases, it is also necessary to integrate expressions involving 3 factors, of the
form

L(, ; k) = dx (x + a) (x + b) (x + c)k ,
(53)
with and half integers and k a general integer.11 For our computation, the relevant cases
are a = 0, b = 4 and c = 1 (or a = 4, b = 0 and c = 1). Let us consider the various
possibilities for the k index. For k < 0 one can reduce the integral (53) to the simpler form
of I (, ) in (42) by means of the binomial expansion of (x + c)|k| in powers of x + a or
of x + b:
|k|
|k|

|k|
|k|
(x + c)|k| =
(c a)|k|s (x + a)s =
(c b)|k|s (x + b)s . (54)
s
s
s=0
s=0
Therefore we just need to consider the case k > 0 from now on.
By means of the algebraic relations in (46) and (48), we can shift one of the half-integer
indices and to the target value 1/2; let us assume for instance that = 1/2. We
can then assume the integral of the form

dx

L (; k) = L(1/2, ; k) =
(55)
(x + b)l (x + c)k ,
(x + a)(x + b)
with l = + 1/2 a general integer.
With an algebraic reduction analogous to the one in Eqs. (46) and (48), we can reduce
the product (x + b)l (x + c)k to a linear combination of terms involving powers of either
(x + b) or (x + c), but not the product. The terms not containing (x + c) belong to the
simpler class of the integrals I (, ) defined in Eq. (42), whose reduction has already
been discussed. The terms not containing (x + b) have the indices and both equal to
their target value 1/2. Our task is then the evaluation of integrals of the form

dx
(x + c)k ,
L(k)
= L(1/2, 1/2; k) =
(56)
(x + a)(x + b)
with k integer and strictly positive, k N . The required identity is
1
(x + a)1/2(x + b)1/2(x + c)k+1

L(k)
=
k1

1
1
dx (x + a)3/2(x + b)1/2
2(k 1) (c a)k1
11 We have put a minus sign in front of k just for practical convenience, in order to simplify the forthcoming
results a little bit.
294

1
1
dx (x + a)1/2(x + b)3/2
2(k 1) (c b)k1

k1

1
1
1
+
+
L(l).
2(k 1)
(c a)kl (c b)kl
(57)
l=1
The second and the third terms on the r.h.s. do not contain any power of (x + c), can be
reduced with the previous identities and are therefore known terms. The last term contains
integrals L(l)
of the same form as the one on the l.h.s.: L(k).
Eq. (57) is then to be used
to relate L integrals, treating the other terms as known functions. By using Eq. (57) for
instance for k = 3 one can reduce L(3)

to a superposition of L(2)
and L(1),
while using it
for k = 2 one can reduce L(2) to L(1). In general, by recursively using Eq. (57), one can
reduce any k > 1 to k = 1, i.e., to the basic integral12

1
,
L(1)
= dx
(58)
(x + a)(x + b)(x + c)
which defines the H (r 1; x)s.
The derivation of Eq. (57) is analogous to the ones of the previous identities: one has
to integrate (x + c)k and to differentiate (x + a)1/2(x + b)1/2 with respect to x. The
resulting expressions have to be simplified using (x + c)/(x + a) = 1 + (c a)/(x + a)
and an analogous equation with b replacing a. One also needs the following result:

1
1
1
1
1
1
=
+
k
l+1
kl
k
(x + c) x + a
(c a)
(x + c)
(c a) x + a
k1
(59)
l=0
and an analogous equation with b replacing a.

Finally, the last class of integrals we need to consider is

F (, ; k) = dx (x + a) (x + b) (x + c)k H (v, w;
x).
(60)
As with the previous case, we can reduce ourselves to the case = 1/2, = 1/2 and
k > 0, i.e., to

dx
F (k) = F (1/2, 1/2; k) =

(61)
H (v, w;
x).
(x + a)(x + b)(x + c)k
The identity reads
1
(x + a)1/2(x + b)1/2(x + c)k+1 H (v, w;
x)
F (k) =
k1

1
1
x)
dx (x + a)3/2(x + b)1/2H (v, w;
2(k 1) (c a)k1

1
1
x)
dx (x + a)1/2(x + b)3/2H (v, w;
2(k 1) (c b)k1
12 Note that the singularity of the coefficients in Eq. (57) for k = 1 forbids any further reduction.
1
k1
295
x)
dx (x + a)1/2(x + b)1/2(x + c)k+1 g(v; x)H (w;

k1

1
1
1
+
+
F (l).
2(k 1)
(c a)kl (c b)kl
(62)
l=1
The fourth term on the r.h.s. involves the integration of an elementary function times a
GHPL of weight w, while the l.h.s. involves the integration of a GHPL of weight w + 1.
This term then has a smaller recursive weight than F and has to be considered as a known
integral. For the rest, analogous considerations to the one of the previous identity hold. By
recursively using Eq. (62), one can reduce any k > 1 to k = 1, i.e., to the basic integral

1
H (v, w;
x),
F (1) = dx
(63)
(x + a)(x + b)(x + c)
which defines the GHPLs of weight w + 2.
3.3. Numerical evaluation
The numerical evaluation of 1- and 2-dimensional HPLs has been discussed in [3]
and [16], respectively. Fortran routines are also available for a fast numerical evaluation
of the HPLs of weight up to 4 included for a real argument: the running time is of order
1 millisecond on a present-day PC for an accuracy O(1015). These routines are based on
power expansions of the HPLs and on a set of transformation rules to bring the argument
x to a positive value as close to 0 as possible.
In the case of GHPLs, the extension of the above strategy is not straightforward, because
the transformation rules are far more complicated when square roots are involved. In [17]
GHPLs have been computed for a large set of arguments simply by numerically evaluating
the multiple integrals. The only subtlety is related to the i prescription. Since x = s/m2
and s s + i, we have that x x i, with a positive infinitesimal number.
As an example let us consider:
1
(x i)(4 + x i)
1
i
= (x)
+ (x) (4 + x)
x(4 + x)
(x)(4 + x)
1
(4 x)
,
x(4 + x)
g(r; x) =
where the square root on the r.h.s. is intended as the elementary (arithmetic) one:
a > 0. The weight-1 GHPL is correspondingly given by

x
x
H (r; x) = (x)2 log 1 + +
4
4

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

x
x
i .
+ (4 x) 2 log 1 +
4
4
(64)
a 2 = a,
(65)
296
Note that

1 + x i x = 1,

4
4
(66)
and therefore, for 4 < x < 0, the harmonic polylogarithm is purely imaginary

x
x
x
log 1 + i
= i arctan
.
4
4
4+x
(67)
4. Results for the master integrals

In this section we present the results of our computation of the MIs involving up to 6
denominators included, which constitute a necessary input for the calculation of the 2-loop
vertex diagrams in Figs. 1 and 2. They are expanded in a Laurent series in
= 2 D/2,
(68)
up to the required order in . The coefficients of the series are expressed in terms of GHPLs
(see Section 3) of the variable x, defined as
x=
s
,
a
(69)
where s = q 2 is the c.m. energy squared,13 q = p1 + p2 and a = m2 .

Note that the GHPLs can be expressed in many casesbut not alwaysin terms of
usual Nielsens polylogarithms of more complicated arguments. For instance, in the case
of H (4, r; x), we have
x
H (4, r; x) =
0
dt
4+t
t
0
du
u(4 + u)
= H (0, 0; y) 2H (1, 0; y) (2)

1
= log2 y 2 log y log (1 + y) 2 Li2 (y) (2),
2
(70)
where
x +4 x
y=
.
x +4+ x
(71)
Therefore, a possible presentation of our results could be to write the MIs, whenever possible, in terms of Nielsens polylogarithms, keeping the GHPLs notation for the remaining
cases. For simplicitys sake, we prefer however to use the GHPLs notation in all the cases.
Furthermore, the latter has the following advantages:
13 We define the scalar product of two 4-vectors as: a b = a b + a

b.
0 0
297
the GHPLs representation of a MI is unique: multiple representations or hidden zeroes are absent by construction, while this is not at all guaranteed for Nielsens
polylogarithms of fancy arguments. In essence, the GHPLs offer a linear basis for
the MIs, while Nielsens polylogarithms of rational functions, radicals, etc., constitute
a non-linear basis, with much more complicated transformation rules;14
the analytic properties of the MIssuch as analytic continuation, location of singular
points, etc.are manifestly shown;
if the MIs enter as known terms in a differential equation for a more complicated
diagram, the representation linear in the H s is by far more convenient.
We denote by the mass scale of the dimensional regularization (DR)the so-called
unit of mass. We work in Minkowski space and we normalize the loop measures as
DD k =
dDk
D
i 2 (3
D
2)
d 42 k
.
i 2 (1 + )
(72)
This definition makes the expression of the 1-loop tadpolethe simplest of all loop
diagramsparticularly simple [1]. The denominators appearing in the master integrals are
listed below:
D1 = k12 ,
(73)
D2 = k22 ,
(74)
D3 = (k1 + k2 ) ,
(75)
D4 = (p1 k1 ) ,
(76)
D5 = (p2 + k1 ) ,
(77)
2
2
D6 = (p2 k2 ) ,
2
(78)
D7 = (p1 k1 + k2 ) ,
(79)
D8 = (p2 + k1 k2 ) ,
(80)
D9 = (p1 + p2 k1 ) ,
(81)
2
2
D10 = (p1 + p2 k2 ) ,
2
(82)
D11 = (p1 + p2 k1 k2 ) ,
(83)
D12 = k12
D13 = k22
(84)
+ a,
+ a,
(85)
D14 = (k1 + k2 ) + a,
(86)
D15 = (p1 k1 ) + a,
(87)
D16 = (p2 + k1 ) + a,
(88)
D17 = (p2 k2 ) + a,
(89)
2
2
2
14 As well known, L. Lewin could fill a thick book with such transformation rules, which correspond, as shown
in the previous section, to a few formulas in the GHPLs basis [18].
298
D18 = (p1 k1 + k2 )2 + a,
(90)
D19 = (p2 + k1 k2 ) + a,
(91)
D20 = (p1 + p2 k1 ) + a,
(92)
D21 = (p1 + p2 k2 )2 + a,
(93)
D22 = (p1 + p2 k1 k2 )2 + a.
(94)
All the results for the MIs of the present paper as well as of [1] can be downloaded as a
FORM input file in the personal web-site of one of us (R.B.) [19].
4.1. Topology t = 3

= 2(4D)

=
2
a
DD k1 DD k2
2
2
(1)
i Fi
1
D12 D13 D14

+ O 3 ,
(95)
(96)
i=2
where
(1)
F2
3
= ,
a
2
(1)
F1
9
= ,
a
2
(1)
F0
21
= 3H (r, 0; 1),
a
2
(1)

F1
45
= 3 3H (r, 0; 1) + H (r, 0, 0; 1) + H (4, r, 0; 1) ,
a
2
F2(2)
93
= 3 7H (r, 0; 1) + 3H (r, 0, 0; 1) + 3H (4, r, 0; 1)
a
2

+ H (r, 0, 0, 0; 1) + H (4, r, 0, 0; 1) + H (4, 4, r, 0; 1) .
(97)
(98)
(99)
(100)
(101)
This diagram has been originally evaluated in Ref. [20].

Even though the above MI is a vacuum amplitude, its computation is non-trivial because
of the presence of 3 massive propagators. We have evaluated it with the following method.
We consider a vacuum sunrise with 2 internal lines with equal mass m and the third internal
line with the different mass squared m 2 = zm2 . We then differentiate the vacuum sunrise
with respect to z and rewrite the result in terms of MIs by using the ibps identities. The
resulting differential equation represents the evolution in one of the masses and is solved
as in usual cases. We set at the end z = 1 to obtain the equal-mass case.
Let us make a few remarks about the above result:
The finite part of the MI, i.e., the O( 0 ), involves 1 transcendental constant: H (r, 0; 1),
related to the Clausen function;
299
the O() part involves 2 new independent transcendental constants: H (r, 0, 0; 1) and
H (4, r, 0; 1);
the O( 2 ) part involves 3 new transcendental constants: H (r, 0, 0, 0; 1), H (4, r, 0, 0; 1)
and H (4, 4, r, 0; 1).
To simplify the above expressions, we have used the following identities to move the 0
indices to the right
H (0, r; 1) = H (r, 0; 1),
H (0, r, 0; 1) = 2H (r, 0, 0; 1),
H (0, r, 0, 0; 1) = 3H (r, 0, 0, 0; 1),
H (0, 0, r, 0; 1) = 3H (r, 0, 0, 0; 1),
H (0, 4, r, 0; 1) = H (4, 0, r, 0; 1) 2H (4, r, 0, 0; 1).
(102)
The above relations are obtained by transforming products of H s into linear combinations
of H s, as for example in
0 = H (0; 1)H (r; 1) = H (0, r; 1) + H (r, 0; 1).
(103)
4.2. Topology t = 4

=

=
2(4D)
2
a
DD k1 DD k2
2
2
(2)
i Fi
1
D2 D11 D12 D20

+ O 3 ,
(104)
(105)
i=2
where
1
(2)
F2
= ,
2
x +4
5
(2)
F1 =
H (r; x),
2
x(x + 4)

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

1
3
+ 1 H (1; x),
H (r, 1; x) H (4, r; x)
2
x
65
(2)
F1 =
+ 5 (2) (3) 3H (0, 1; x) + 2H (0, 1, 1; x)
2

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

x+4
H (0, 1; x) +
2 6 + (2) H (r; x)
x(x + 4)
(106)
(107)
(108)
300
+ 6H (r, 1; x) 6H (r, 1, 1; x) + 3H (r, 0, 1; x)

3
+ 4H (4, r; x) H (4, r, 1; x) H (4, 4, r; x) ,
(109)
2

211
9
(2)
+ 19 (2) + 2 (2) 5(3) 13 + (2) H (0, 1; x)
F2 =
2
5
8H (0, 1, 1, 1; x) + 3H (0, 1, 0, 1; x) + 12H (0, 1, 1; x)
1
3H (0, 0, 1; x) + 2H (0, 0, 1, 1; x) H (0, 0, 0, 1; x)
2

x+4
+
2 (3) 4(2) 16 H (r; x)
x(x + 4)

+ 6 + (2) 3H (r, 1; x) + 2H (4, r; x)
24H (r, 1, 1; x) + 24H (r, 1, 1, 1; x)
9H (r, 1, 0, 1; x) + 12H (r, 0, 1; x) 12H (r, 0, 1, 1; x)
+ 3H (r, 0, 0, 1; x) 6H (4, r, 1; x) + 6H (4, r, 1, 1; x)
3
3H (4, r, 0, 1; x) 4H (4, 4, r; x) + H (4, 4, r, 1; x)
2

1
+ H (4, 4, 4, r; x)
+ 1 33 + 2(2) H (1; x)
x
28H (1, 1; x) + 16H (1, 1, 1; x) 6H (1, 0, 1; x)

+ 7H (0, 1; x) 4H (0, 1, 1; x) + H (0, 0, 1; x) .
(110)
The above 2-point function is the simplest amplitude having thresholds both in s = m2
and s = 4m2 . It does not have any pseudo-threshold. The indices appearing in the GHPLs are indeed only 0, 1, 4 and r. Note that the index r appears eventually only
once in the H s. The related terms have coefficients always containing radicals, in order to
reproduce the right causality structure.

=

=
2(4D)
2
a
DD k1 DD k2
2
1
1
D5 D7 D12 D13

i Fi(3) + O 3 ,
(111)
(112)
i=2
where
1
(3)
F2 = ,
2
3
(3)
F1 = ,
2
5
(3)
F0 = + H (1; x) H (0, 1; x)
2

1
+ H (1; x) (2)H (1; x) + 2H (1, 0, 1; x) ,
x
(113)
(114)
(115)
301
1
(3)
F1 = (2) + 7H (1; x) (2)H (1; x) 4H (1, 1; x) 2H (0, 1; x)
2
1
+ 4H (0, 1, 1; x) + H (0, 0, 1; x) + 2H (1, 0, 1; x) + 7H (1; x)
x

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

+ 2H (0, 1, 0, 1; x) + 2H (1, 0, 1; x) 8H (1, 0, 1, 1; x) .
(116)

(p2 k2 ) =

=
2(4D)
2
a
DD k1 DD k2
2
2
p2 k2
D5 D7 D12 D13

i Fi(4) + O 3 ,
(117)
(118)
i=2
where
(4)
F2
a
(4)
F1
1
= x,
8
5
= x,
a
16
F0(4) 1 1
1
= H (1; x) H (1; x)
a
8 2
8x

3
1
7
+ H (1; x) H (0, 1; x) ,
x
32 8
4
(119)
(120)
(121)
F1(4)
9
1
=
+ (2) 3H (1; x) + 2H (1, 1; x) H (0, 1; x)
a
16 4
1 7
1
1
H (1; x) + (2)H (1; x) H (1, 1; x)

x 16
4
2

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

41
1
3
123 3
+ (2) H (1; x) + (2)H (1; x) + H (1, 1; x)
+x
64
8
16
4
2
1
1
+ H (0, 1; x) H (0, 1, 1; x) H (0, 0, 1; x)
4
4

1
H (1, 0, 1; x) ,
(122)
2
(4)
F2
39
1
25
1
=
+ (2) (3) H (1; x) (2)H (1; x) (2)H (1; x)
a
32
4
2
2
21
+ 12H (1, 1; x) H (0, 1; x) 8H (1, 1, 1; x)
4
1
+ 3H (1, 0, 1; x) + 4H (0, 1, 1; x) H (0, 0, 1; x)
2
302

1
1 21
H (1; x) + (2)H (1; x)
x 32
4
1
7
1
+ (2)H (1; x) (3)H (1; x) H (1, 1; x)
2
4
4
7
1
+ H (0, 1; x) + (2)H (0, 1; x) + 2H (1, 1, 1; x)
16
4
1
1
3
H (1, 0, 1; x) H (0, 1, 1; x) + H (0, 0, 1; x)
4
2
8

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

1681 41
3
379
3
+ (2) (3)
H (1; x) (2)H (1; x)
+x
128
16
8
32
4
41
1
+ (2)H (1; x) (3)H (1; x) + H (1, 1; x)
4
4
17
1
1
H (0, 1; x) + (2)H (0, 1; x) (2)H (0, 1; x)
8
2
4
9
6H (1, 1, 1; x) + H (1, 0, 1; x) H (0, 1, 1; x)
4
7
H (0, 0, 1; x) 2H (1, 0, 1; x) + 4H (0, 1, 1, 1; x)
4
3
3
H (0, 1, 0, 1; x) + H (0, 0, 1, 1; x) + H (0, 0, 0, 1; x)
2
4

1
+ H (0, 1, 0, 1; x) + 2H (1, 0, 1, 1; x) .
(123)
2
+ H (1, 0, 1; x)
The above 2 MIs, which contain 2 massive denominators, can be expressed in terms of
ordinary HPLs. The reason is that the 2 massive lines, roughly speaking, are in different
channels: one is in the s channel while the other is in the t channel. The amplitudes do not
have thresholds/pseudothresholds in s = 4m2 , but only in s = m2 . Both the indices 1
and 1 do indeed appear inside the HPLs. The presence of 2 massive denominators is
then a necessary but not a sufficient condition in order to have thresholds or pseudothresholds in s = 4m2 .

1
= 2(4D) DD k1 DD k2
(124)
D7 D8 D12 D13

=
2
a
2
1
(5)
i Fi

+ O 3 ,
(125)
i=2
where
1
(5)
= ,
F2
2
5
(5)
F1 = H (0; x),
2
(126)
(127)
303
4x
19
(2) 5H (0; x) + H (0, 0; x) +
H (r, 0; x)
2
x(4 x)
2
+ H (r, r, 0; x),
(128)
x
65
(5)
5 (2) 2 (3) 19H (0; x) + (2)H (0; x) H (r, r, 0; x)
F1 =
2
4x
+ 5H (0, 0; x) H (0, 0, 0; x) +
(2)H (r; x) + 5H (r, 0; x)
x(4 x)
2
H (r, 0, 0; x) H (0, r, 0; x) + 2H (4, r, 0; x) + (2)H (r, r; x)
x
+ 3H (r, r, 0; x) H (r, r, 0, 0; x) H (r, 0, r, 0; x) + 2H (r, 4, r, 0; x)

+ H (0, r, r, 0; x) .
(129)
(5)
F0 =
The double and the simple poles in the MI above have ultraviolet origin. The amplitude
has indeed an ultraviolet sub-divergence related to the integration of the bubble together
with an over-all UV divergence.

=

=
2(4D)
2
a
DD k1 DD k2
2
2
1
D72 D8 D12 D13

i Fi(6) + O 3 ,
(130)
(131)
i=2
where
(6)
F2
a
1
= ,
x
(132)
(6)
F1
1
H (0; x),
a
x

F0(6)
1
4x
= 4 (2) 2H (0; x) + H (0, 0; x)
H (r, 0; x),
a
x
x x(4 x)
=
(133)
(134)
F1(6)
1
= 2 (2) + 2 (3) + 4H (0; x) (2)H (0; x) 2H (0, 0; x) + H (r, r, 0; x)
a
x

4x
+ H (0, 0, 0; x)
(2)H (r; x) + 2H (r, 0; x) H (r, 0, 0; x)
x x(4 x)

H (0, r, 0; x) + 2H (4, r, 0; x) ,
(135)

(6)

F2
1
9
= 16 4 (2) 2 (2) 4(3) 2 4 (2) (3) H (0; x)
a
x
10

+ 4 (2) H (0, 0; x) (2)H (r, r; x) 2H (r, r, 0; x) + H (r, r, 0, 0; x)
+ H (r, 0, r, 0; x) 2H (r, 4, r, 0; x) H (0, r, r, 0; x) 2H (0, 0, 0; x)
304

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

4x

2 (2)H (r; x) + 2(3)H (r; x) + 4H (r, 0; x)
x x(4 x)
(2)H (r, 0; x) (2)H (0, r; x) + 2(2)H (4, r; x) + 3H (r, r, r, 0; x)
2H (r, 0, 0; x) + 4H (4, r, 0; x) 2H (0, r, 0; x) + H (r, 0, 0, 0; x)
+ H (0, r, 0, 0; x) + H (0, 0, r, 0; x) 2H (0, 4, r, 0; x) 2H (4, r, 0, 0; x)

2H (4, 0, r, 0; x) + 4H (4, 4, r, 0; x) .
(136)
The double pole in in the MI above is the product of a simple UV pole coming from the
nested bubble and of an IR pole coming from the massless line squared.
The above 2 MIs have a pseudothreshold in s = 4m2 related to the exchange of 2
massive particles in the t channel and consequently only GHPLs with indices 0, 4, and r
do appear in the expansion. These MIs have also been computed in [21] by means of a
transformation well-known in QED eliminating the square roots:
x=
(1 + z)2
.
z
(137)
In general, this change of variable is very convenient for amplitudes not having the
pseudothreshold in s = m2 .
4.3. Topology t = 5

=

=
2(4D)
2
a
DD k1 DD k2
2
1
(7)
i Fi
1
D2 D3 D11 D12 D20

+ O 3 ,
(138)
(139)
i=0
where
1
H (0, 0, 1; x) + 3H (r, r, 1; x) 2H (r, r, 0; x)
x

2H (r, 0, r; x) ,
1
(7)
aF1 = 2H (0, 0, 1; x) 6(2)H (r, r; x) + 6H (r, r, 1; x)
x
4H (r, r, 0; x) 4H (r, 0, r; x) 4H (0, 0, 1, 1; x)
(7)
aF0 =
(140)
+ H (0, 0, 0, 1; x) 12H (r, r, 1, 1; x) + 6H (r, r, 0, 1; x)

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

+ 2H (r, 0, 0, r; x) .
(141)
305
The above 2-point function has thresholds in s = 0, m2 as well as in s = 4m2 and no

pseudothresholds. The index r always appears twice in the H s and the related coefficients contain no radicals. The finite part O( 0 ) of F (7) has been computed by the authors
of [22] by means of a resummed small momentum expansion. With the help of the Mathematica system [23] we have computed the first 20 terms of the small momentum expansion
of our result and compared with their result, finding complete agreement.

=

=
DD k1 DD k2
2(4D)
2
a
2
1
D4 D6 D12 D13 D14

F0(8) + O 3 ,
(142)
(143)
where

1
(2) H (0, c; x) + 2H (0, 1; x) + H (0, c; x) 3H (0, c, 0, 1; x)
x
i

3H (0, c, 0, 1; x) 4H (0, 1, 0, 1; x) + H (r, 0; 1) H (0, c; x)
x

H (0, c; x) .
(144)
aF0(8) =

(p2 k1 ) = 2(4D)

=
2
a
DD k1 DD k2
2
1
(9)
i Fi
p2 k1
D4 D6 D12 D13 D14

+ O 3 ,
(145)
(146)
i=0
where

1
x) + H (c; x)
F0(9) = 1 + H (1; x) (2)H (1; x) (2) H (c;
2

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

3
1 1
0, 1; x)
3H (r, 0; 1) + (2) H (c; x) + H (c;
x) H (c,
+
x 4
4

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

i
1
3
+
H (r, 0; 1) 2
H (c; x) H (c;
x) +
(2) H (c; x)
4
x
x

H (c;
x) 3H (c, 0, 1; x) + 3H (c,
0, 1; x) ,
(147)
(9)
F1 = 7 (2) + 6H (1; x) 2(2)H (1; x) + (3)H (1; x)

1
x) 4H (1, 1; x)
(2) (3) H (c; x) + H (c;
2
306

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

1

+ (2) H (c, c; x) + H (c,
c;
x) + (2) H (c,
c; x) + H (c, c;
x)
2

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

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

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

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

1
0, 0, 1; x) + H (c, 0, 0, 1; x) 4 H (c,
1, 0, 1; x)
H (c,
2
9

+ H (c, 1, 0, 1; x) + H (0, c, 0, 1; x) + H (0, c, 0, 1; x)
2
1
3 H (r, 0; 1)
+ 6H (0, 1, 0, 1; x) 8H (1, 0, 1, 1; x) +
4x

+ H (r, 0, 0; 1) H (c; x) + H (c;
x) + H (r, 0; 1) H (c, c;
x)

+ H (c,
c; x) 2H (c, c; x) 2H (c,
c;
x) + H (0, c;
x) + H (0, c; x)

+ H (4, r, 0; 1) H (c; x) + H (c;
x)

i
+
H (r, 0; 1) + H (r, 0, 0; 1) + H (4, r, 0; 1) H (c; x) H (c;
x)
2

+ H (r, 0; 1) H (c, c;
x) H (c,
c; x) 2H (c, c; x) + 2H (c,
c;
x)

1
H (r, 0; 1) + H (r, 0, 0; 1)
+ 3H (0, c; x) 3H (0, c; x)
2x

+ H (4, r, 0; 1) H (c; x) H (c;

x) + H (r, 0; 1) H (c, c;
x)

H (c,
c; x) 2H (c, c; x) + 2H (c,
c;
x) + H (0, c; x) H (0, c;
x)
(2) (3) H (c; x) H (c;

x) + (2)H (0, c; x)
+
2x
(2)H (0, c; x) 4(2)H (c, 1; x) + 4(2)H (c,
1; x) 2(2)H (c, c; x)
+ 2 (2)H (c,
c;
x) (2)H (c, c; x) + (2)H (c,
c; x) 3H (c, 0, 1; x)
+ 3H (c,
0, 1; x) + 8H (c, 1, 0, 1; x) 8H (c,
1, 0, 1; x)
3H (c,
c, 0, 1; x) + 3H (c, c, 0, 1; x) + 6H (c, c, 0, 1; x)
6H (c,
c,
0, 1; x) + 12H (c, 0, 1, 1; x) 12H (c,
0, 1, 1; x)
+ H (c, 0, 0, 1; x) H (c,
0, 0, 1; x) 3H (0, c, 0, 1; x)

+ 3H (0, c, 0, 1; x) .
(148)

=

=
DD k1 DD k2
2(4D)
2
a
2
1
1
2
D4 D6 D12 D13 D14

i Fi(10) + O 3 ,
307
(149)
(150)
i=0
where
a
F0(10)
1
=
x
+i

3
H (r, 0; 1) H (c; x) + H (c;
x)
3

3
(2) H (c; x) H (c;
x)
3

3H (c, 0, 1; x) + 3H (c,
0, 1; x) ,
a
F1(10)
(151)

3
H (r, 0, 0; 1) + H (4, r, 0; 1) H (c; x) + H (c;
x)
3

+ H (r, 0; 1) H (c, c; x) + H (c,
c; x) 2H (c, c; x) 2H (c,
c;
x)

3
+ H (0, c; x) + H (0, c; x) i
(3) H (c; x) H (c;
x)
3
(2) H (0, c; x) H (0, c; x) 4H (c, 1; x) + 4H (c,

1; x)

2H (c, c; x) + 2H (c,
c;
x) H (c, c; x) + H (c,
c; x)
1
=
x
6H (c, c, 0, 1; x) + 6H (c,
c,
0, 1; x) 3H (c, c, 0, 1; x)
+ 3H (c,
c, 0, 1; x) 12H (c, 0, 1, 1; x) + 12H (c,
0, 1, 1; x)
H (c, 0, 0, 1; x) + H (c,
0, 0, 1; x) 8H (c, 1, 0, 1; x)

+ 8H (c,
1, 0, 1; x) + 3H (0, c, 0, 1; x) 3H (0, c, 0, 1; x) .
(152)
The above 3 MIs are real as complex H ( c ; x)s always appear in the combinations:
H ( c ; x) + H ( c ; x) or i[H ( c ; x) H ( c ; x)]. As discussed in the
previous section, the above topology is the only one having 3 MIs. The amplitudes have a
threshold in s = m2 in agreement with Cutkowsky rule as well as a pseudo thresholds in
s = m2 .

=

=
2(4D)
2
a
DD k1 DD k2
2
1
i=1
(11)
i Fi
1
D4 D5 D12 D13 D14

+ O 3 ,
(153)
(154)
308
where
1
(11)
aF1 = H (1, 0; x),
(155)
x
1
aF0(11) = (2)H (1; x) + 2H (1, 0; x) H (r, r, 0; x) + H (0, 1, 0; x)
x
H (1, 0, 0; x) + H (1, 1, 0; x) 3H (1 + r, r, 0; x)

+ 3H (r, 0; 1)H (1; x) ,
(156)

1
aF1(11) = 2 (2)H (1; x) 2(3)H (1; x) (2)H (0, 1; x) 4H (1, 0; x)
x
+ (2)H (1, 0; x) (2)H (1, 1; x) + (2)H (r, r; x)
+ 3 (2)H (1 + r, r; x) 2H (0, 1, 0; x) + 2H (1, 0, 0; x) 2H (1, 1, 0; x)
+ 2H (r, r, 0; x) + 6H (1 + r, r, 0; x) H (0, 0, 1, 0; x) + H (0, 1, 0, 0; x)
H (0, 1, 1, 0; x) H (1, 0, 0, 0; x) H (1, 0, 1, 0; x) + H (1, 1, 0, 0; x)
H (1, 1, 1, 0; x) H (r, r, 0, 0; x) H (r, 0, r, 0; x) + 2H (r, 4, r, 0; x)
+ H (0, r, r, 0; x) 3H (1 + r, r, 0, 0; x) 3H (1 + r, 0, r, 0; x)
+ 6H (1 + r, 4, r, 0; x) + 3H (0, 1 + r, r, 0; x) + 3H (1, 1 + r, r, 0; x)

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

+ H (r, 0; 1) H (0, 1; x) + H (1, 1; x) .
(157)
The above amplitude has pseudothresholds in s = m2 and s = 4m2 and is the only one
containing GHPLs with the index 1 + r. The latter is then related to the (virtual) transition
of a particle with mass m = 0 into a pair of particles with the same mass, i.e., a bubble with
2 equal mass lines. The index r appears in the GHPLs only 0 or 2 times.

= 2(4D)

=
2
a
DD k1 DD k2
2
(12)
F0
1
D3 D4 D6 D12 D13
+ O(),
(158)
(159)
where
aF0(12) =

2
(2)H (0, 1; x) + H (0, 1, 1, 0; x) .
x
(p2 k1 ) = 2(4D)

=
2
a
DD k1 DD k2
2
1
i=0
p2 k1
D3 D4 D6 D12 D13

i Fi(13) + O 2 ,
(160)
(161)
(162)
309
where
(13)
F0
a
= 1 + (2) + H (0; x) + (2)H (1; x) + H (1, 0; x) + H (1, 1, 0; x)

+

1
(2)H (1; x) H (1, 0; x) H (1, 1, 0; x) ,
x
(163)
(13)
F1
a
= 7 + 4 (2) (3) + 6H (0; x) + 3(2)H (1; x) (3)H (1; x)

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

5H (1, 1, 1, 0; x) .
(164)
The above 2 MIs contain HPLs with indices 0 and 1 only. They represent the emission of a photon by a charged vector boson in the t channel, and therefore have only a
pseudothreshold in s = m2 . They are IR (as well as UV) finite because the photon is
emitted internally to the basic 1-loop triangle.

= 2(4D)

=
2
a
DD k1 DD k2
2
(14)
F0
1
D1 D3 D6 D13 D15
+ O(),
(165)
(166)
where
(14)
aF0
1
(2)H (0, 1; x) + (2)H (0, 1; x) + H (0, 1, 0, 1; x)
x

2H (0, 1, 0, 1; x) .
(p1 k2 ) =

=
2(4D)
2
a
DD k1 DD k2
2
1
p1 k2
D1 D3 D6 D13 D15

i Fi(15) + O 2 ,
i=0
where
(15)
F0

1
3
3 1
= + 3 + (2) H (1; x) + (2)H (1; x) H (0, 1; x)
2 2
2
2
(167)
(168)
(169)
310

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

1
1
(2)H (1; x) + H (1, 0, 1; x) + H (1, 0, 1; x) ,
(170)
2
2

1
1
(15)
F1 = 9 (2) + 15 (3) H (1; x) (2) + (3) H (1; x)
2
2

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

1
1 1
15 (3) H (1; x) + (2) + (3) H (1; x)
+
x 2
2

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

= 2(4D)

=
2
a
DD k1 DD k2
2
1
(16)
i Fi
(171)
1
D1 D2 D8 D15 D16

+ O 2 ,
(172)
(173)
i=1
where
2
(174)
H (r, r; x),
x
1
(16)
aF0 = 4H (r, r; x) 3H (r, r, 1; x) 2H (r, 4, r; x)
x

+ 2H (0, r, r; x) ,
(175)

1
(16)
aF1 = 4 2 + (2) H (r, r; x) 6H (r, r, 1; x) + 4H (0, r, r; x)
x
4H (r, 4, r; x) + 12H (r, r, 1, 1; x)
(16)
aF1
=
6H (r, r, 0, 1; x) + 3H (r, 4, r, 1; x)
+ 2H (r, 4, 4, r; x) 3H (0, r, r, 1; x)

2H (0, r, 4, r; x) + 2H (0, 0, r, r; x) .
311
(176)
The above MI has a simple UV pole coming from the sub-divergence in the bubble.

=
DD k1 DD k2
2(4D)

=
2
a
2
1
D5 D7 D8 D12 D13
F0(17) + O () ,
(177)
(178)
where
(17)
aF0

1
(2)H (0, 1; x) 2H (0, 1, 0, 1; x) + 2H (0, r, r, 0; x) .
x
(179)
4.4. Topology t = 6

=

=
DD k1 DD k2
2(4D)
2
a
2
(18)
F0
1
D4 D5 D6 D12 D13 D14

+ O 2 ,
(180)
(181)
where
a 2 F0(18) =
1
(2) H (0, 1; x) + H (1, 1; x) + H (1, c; x) + H (1, c; x)
x

+ H (0, c; x) + H (0, c; x) 2H (1, 1, 0, 1; x) 2H (0, 1, 0, 1; x)
3H (0, c, 0, 1; x) 3H (0, c, 0, 1; x) 2H (1, r, r, 0; x)

2H (0, r, r, 0; x) 3H (1, c, 0, 1; x) 3H (1, c, 0, 1; x)

1
i H (r, 0; 1) H (0, c; x) H (0, c; x) + H (1, c; x)
x

H (1, c; x) .
(182)
The index r appears in the GHPLs only 0 or 2 times, so the coefficients of the related terms
do not contain radicals.

1
= 2(4D) DD k1 DD k2
(183)
D2 D3 D4 D5 D12 D17
2 2
=
F0(19) + O(),
(184)
a
where

1
(19)
a 2 F0 = (2) H (0, 1; x) H (0, 1; x) H (1, 1; x) + H (1, 1; x)
x
2H (0, 1, 0, 1; x) + H (0, 1, 0, 0; x) + 2H (0, 1, 0, 1; x)

2H (1, 1, 0, 1; x) + H (1, 1, 0, 0; x) + 2H (1, 1, 0, 1; x) . (185)
312
Because of analogous considerations to the previous ones, the above MI is expressed in

terms of ordinary HPLs.

=

=
DD k1 DD k2
2(4D)
2
a
2
1
D1 D2 D7 D8 D15 D16
F0(20) + O(),
(186)
(187)
where
(20)
a 2 F0

1
12(2)H (r, 1; x) 6H (r, 1, 0, 1; x)
=
x x(x + 4)
+ 6H (r, 1, 0, 0; x) 12H (r, r, r, 1; x)
+ 8H (r, r, r, 0; x) + 8H (r, r, 0, r; x)

+ 4H (r, 0, r, r; x) 2H (r, 0, 0, 1; x) .

=

=
2(4D)
2
a
2
DD k1 DD k2
1
D4 D5 D7 D8 D12 D13
F0(21) + O(),
(188)
(189)
(190)
where
(21)
a 2 F0
1
6 (2)H (1, 1; x) + 6H (1, r, r, 0; x) 2H (1, 0, 0, 1; x)
x2

+ H (1, 0, 1, 0; x) 12H (1, 1, 0, 1; x) + 4H (1, 1, 0, 0; x) .
(191)
The masses of the vector bosons exchanged in the t channel completely cut-off the infrared singularities, so the above MI is IR (as well as UV) finite. As noted in our previous
work [1], a non-zero mass on the outer boson line is already sufficient to completely screen
the IR singularities.
The above MI has been computed in [22] by fitting a small momentum expansion to
an assumed form for the exact expression. We have compared the first 15 terms of the
small momentum expansion in [22] with an analogous expansion of our expression, finding
complete agreement. The first few terms of the small momentum expansion of F (21) are
given in the next section. In [24] a leading-twist large-momentum expansion of the above
MI has been presented, based on the separation of the loop space in leading IR regions
and a related approximation on the integrand. The above result is in agreement with a
preliminary large momentum expansion of our expression [15].
5. Small momentum expansion of six denominator amplitudes

In this section we present the small momentum expansions |s| m2 of all the
6-denominator diagrams, i.e., expansions in powers of x and L = log x up to first order
313
in x included. The expansion of the GHPLs for a small value of the argument |x| 1 is
obtained in the following way:
we explicitly write the GHPLs as a repeated integration over the basis functions, as for
example:
x
H (1, r, r, 0; x) =
0
dx1
1 x1
x1
0
dx2
x2 (4 x2 )
x2
0
dx3
log x3 ;
x3 (4 x3 )
(192)
we expand the basis functions in powers of x up to the required order in x,15 such as
for instance:
1
1
x
+ ;
(193)
= +
16
x(4 x) 2 x
we integrate term by term the expanded functions. This involves in general the integration of functions of the form x q log x k , with q integer or half integer and k integer.
The expansions are given below

=
2
a
2
2

x j A0j + O x 3 ,
(194)
j =0
where
a 2 A00 = 4 +
3H (r, 0; 1) L,
3 3
13
19 1
+ (2) +
H (r, 0; 1) L,
a 2 A01 =
9
2
4
24
197
3
12509 2
+ (2) +
H (r, 0; 1)
L.
a 2 A02 =
16200 3
2
540

=
2
a
2
2

x j Bj0 + O x 3 ,
(195)
(196)
(197)
(198)
j =0
where
1
a 2 B00 = 3 2L + L2 ,
2
1
1
11 1
(2) L + L2 ,
a 2 B10 =
16 2
2
8
7 1
41
5
2 0
a B2 = (2)
L + L2 .
8 3
108
36
15 The factors 1/x and 1/x are clearly not expanded.
(199)
(200)
(201)
314

=
2
a
2
2
9
x j Cj0 + O x 2 ,
(202)
j =0
where
1
a 2 C00 = 1 + 2 (2) L + L2 ,
2
5
1
17
2 0
a C1 = (2) + L L2 ,
24
12
4
1
61
1
827
2 0
+ (2)
L + L2 .
a C2 =
2160 2
360
8

=
2
a
2
2

x j Ej0 + O x 3 ,
(203)
(204)
(205)
(206)
j =0
where
a 2 E00 = 4 + 3 (2) L + L2 ,
5
341
+ 3 (2) L + L2 ,
a 2 E10 =
72
6
79
11
2617 11
2 0
+ (2)
L + L2 .
a E2 =
600
4
120
12

=
2
a
2
2

x j G0j + O x 3 ,
(207)
(208)
(209)
(210)
j =0
where
1
a 2 G00 = 3 (2) 2L + L2 ,
2
5
3
25 3
(2) L + L2 ,
a 2 G01 =
16 4
4
8
107
11
251 11
2 0
(2)
L + L2 .
a G2 =
216 18
108
36

=
2
a
2
2
j =0

x
0

i Iji

+ O x3 ,
(211)
(212)
(213)
(214)
i=1
where
a 2 I01 = L,
1
a 2 I00 = 6 + 4 (2) + 3L L2 ,
2
1 3
a 2 I11 = + L,
4 4
(215)
(216)
(217)
21
3
11
+ 3 (2) + L L2 ,
2
8
8
11
1
2 1
a I2 = + L,
12 18
22
11
859 22
2 0
a I2 =
+ (2) + L L2 .
216
9
9
36
a 2 I10 =

=
2
a
2
2
j =1

x
1

i Jji

+ O x3 ,
315
(218)
(219)
(220)
(221)
i=2
where
2
a 2 J1
= 1,
(222)
1
a 2 J1
= 1 L,
(223)
1
0
a 2 J1
= 1 (2) L + L2 ,
2
(224)
1
1
1
a 2 J1
= 1 (2) 2 (3) + (2)L L + L2 L3 ,
2
6
1
a 2 J01 = ,
4
1
35
a 2 J00 =
+ L,
72 12
7
1
359 13
2 1
+ (2) + L L2 ,
a J0 =
432 12
24
24
1
2 1
a J1 = ,
18
209
1
2 0
a J1 =
+
L,
675 180
1
23
1 2
141091
+
(2) +
L
L ,
a 2 J11 =
162000 180
1080
360
1
a 2 J21 = ,
48
1
11063
2 0
+
L,
a J2 =
78400 1680
47
17
1
105455939
+
(2) +
L
L2 .
a 2 J21 =
296352000 560
6720
3360
(225)
(226)
(227)
(228)
(229)
(230)
(231)
(232)
(233)
(234)
6. Conclusions
We have presented the exact analytic evaluation of the 25 master integrals containing
2 and 3 massive propagators entering the planar amplitudes of the 2-loop electroweak
form factor. While the reduction to master integrals does not present any new element
with respect to our previous computation [1] and is done with the same algorithm, the
316
analytic evaluation of the master integrals requires a non-trivial extension of the harmonic
polylogarithm theory. The presence of 2 massive particles in the s or in the t channel opens
indeed thresholds and pseudothresholds in s = 4m2 , respectively, in addition to the old
ones in s = 0, m2 . The generalization of the 1-dimensional harmonic polylogarithms has
basically required:
the introduction of new basis functions, in addition to the usual one, involving complex
constants and radicals;
a set of recursion relations to take the integrals with semi-integer powers coming from
the evaluation of the master integrals to a unique form fixed by the basis function
choice.
The basic properties of the ordinary harmonic polylogarithms are maintained by the generalization. The small momentum expansion of all the 6-denominator amplitudes has been
obtained by means of a series expansion of the basis functions. We compared our results
with those present in the literature usually in the form of resummed small momentum expansions or truncated large momentum expansions, finding complete agreement.
In order to complete the evaluation of the master integrals, 2 steps are still to be taken:
the large-momentum expansion of all the MIs, which requires the expansion for a large
value of the argument x of all the H (w;
x)s [15];
the evaluation of the master integrals related to the crossed ladder topology. The reduction, using both the Laporta method and the symbolic method, shows that this topology
has 3 master integrals. The resulting system of 3 differential equations cannot be completely triangularized by means of the techniques discussed in this work, but can be
split into a second-order and a first-order differential equations [15].
Acknowledgements
We are grateful to J. Vermaseren for his kind assistance in the use of the algebra manipulating program FORM [25], by which all our calculations were carried out.
We wish to thank E. Remiddi for discussions and for use of the C program SOLVE [26]
to solve the linear systems generated by the ibp identities.
U.A. wishes to thank A. Pellissetto and M. Testa for discussions.
R.B. wishes to thank the Universit di Roma La Sapienza for hospitality during the
final part of this work.
Appendix A. One-loop master integrals

In this appendix we present the results for the 1-loop master integrals containing 2 massive propagators. We have recomputed them with the method of the differential equations
described in the main body of the paper in terms of GHPLs. In the case of the bubble,
we found that it was necessary to push the expansion up to third order included. In our
317
previous work [1] we gave the expressions of the 1-loop master integrals containing at
most 1 massive propagator. The above amplitudes are necessary for the computation of the
factorized 2-loop master integrals, which are presented in the next Section.
A.1. Bubble

=

=
(4D)
2
a
DD k

3
(k 2
1
+ a)[(p k)2 + a]

i Bi + O 4 ,
(A.1)
i=1
where
B1 = 1,
(A.2)
x+4
B0 = 2
H (r; x),
x(x + 4)

x+4
2H (r; x) H (4, r; x) ,
B1 = 4
x(x + 4)

x+4
B2 = 8
4H (r; x) 2H (4, r; x) + H (4, 4, r; x) ,
x(x + 4)
x +4
8H (r; x) 4H (4, r; x) + 2H (4, 4, r; x)
B3 = 16
x(x + 4)

H (4, 4, 4, r; x) .
(A.3)
(A.4)
(A.5)
(A.6)
A.2. Vertex

=

=
(4D)
2
a
DD k

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

i Vi + O 3 ,
(A.7)
i=0
where
2
H (r, r; x),
x

2
aK1 = H (r, 4, r; x) H (0, r, r; x) ,
x
2
aK2 = H (r, 4, 4, r; x) H (0, r, 4, r; x)
x

+ H (0, 0, r, r; x) .
aK0 =
(A.8)
(A.9)
(A.10)
318
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]
U. Aglietti, R. Bonciani, Nucl. Phys. B 668 (2003) 3, hep-ph/0304028.

E. Remiddi, J.A.M. Vermaseren, Int. J. Mod. Phys. A 15 (2000) 725, hep-ph/9905237.
T. Gehrmann, E. Remiddi, Comput. Phys. Commun. 141 (2001) 296, hep-ph/0107173.
G. t Hooft, M. Veltman, Nucl. Phys. B 44 (1972) 189;
C.G. Bollini, J.J. Giambiagi, Phys. Lett. B 40 (1972) 566;
C.G. Bollini, J.J. Giambiagi, Nuovo Cimento B 12 (1972) 20;
J. Ashmore, Lett. Nuovo Cimento 4 (1972) 289;
G.M. Cicuta, E. Montaldi, Lett. Nuovo Cimento 4 (1972) 289;
R. Gastmans, R. Meuldermans, Nucl. Phys. B 63 (1973) 277.
See, for example, C. Itzykson, J. Zuber, Quantum Field Theory, McGrawHill, 1980.
S. Laporta, E. Remiddi, Phys. Lett. B 379 (1996) 283, hep-ph/9602417;
S. Laporta, Int. J. Mod. Phys. A 15 (2000) 5087, hep-ph/0102033.
F.V. Tkachov, Phys. Lett. B 100 (1981) 65;
K.G. Chetyrkin, F.V. Tkachov, Nucl. Phys. B 192 (1981) 159.
T. Gehrmann, E. Remiddi, Nucl. Phys. B 580 (2000) 485, hep-ph/9912329.
R. Bonciani, P. Mastrolia, E. Remiddi, Nucl. Phys. B 661 (2003) 289, hep-ph/0301170.
A.V. Kotikov, Phys. Lett. B 254 (1991) 158.
E. Remiddi, Nuovo Cimento A 110 (1997) 1435, hep-th/9711188.
M. Caffo, H. Czyz, S. Laporta, E. Remiddi, Acta Phys. Pol. B 29 (1998) 2627, hep-ph/9807119;
M. Caffo, H. Czyz, S. Laporta, E. Remiddi, Nuovo Cimento A 111 (1998) 365, hep-ph/9805118.
U. Aglietti, R. Bonciani, in preparation.
T. Gehrmann, E. Remiddi, Comput. Phys. Commun. 144 (2002) 200, hep-ph/0111255.
U. Aglietti, R. Bonciani, G. Degrassi, A. Vicini, hep-ph/0404071.
L. Lewin, Polylogarithms and Associated Functions, North-Holland, 1981.
The results can be downloaded from http://pheno.physik.uni-freiburg.de/~bonciani/ as an input file for
FORM.
A.I. Davydychev, B. Tausk, Phys. Rev. D 53 (1996) 7381, hep-ph/9504431;
A.I. Davydychev, Phys. Rev. D 61 (2000) 087701, hep-ph/9910224.
R. Bonciani, P. Mastrolia, E. Remiddi, Nucl. Phys. B 690 (2004) 138, hep-ph/0311145.
R. Scharf, J.B. Tausk, Nucl. Phys. B 412 (1994) 523;
F.A. Berends, A.I. Davydychev, V.A. Smirnov, J.B. Tausk, Nucl. Phys. B 439 (1995) 536, hep-ph/9410232;
J. Fleischer, M.Y. Kalmykov, A.V. Kotikov, Phys. Lett. B 462 (1999) 169, hep-ph/9905249;
J. Fleischer, A.V. Kotikov, O.L. Veretin, Nucl. Phys. B 547 (1999) 343, hep-ph/9808242.
Mathematica 4.2, Copyright 19882002 Wolfram Research, Inc.
V.A. Smirnov, Phys. Lett. B 404 (1997) 101, hep-ph/9703357.
J.A.M. Vermaseren, Symbolic Manipulation with FORM, Version 2, CAN, Amsterdam, 1991;
J.A.M. Vermaseren, New features of FORM, math-ph/0010025.
Solve, by E. Remiddi.

Sectional curvature bounds in gravity: regularisation

of the Schwarzschild singularity
Frederic P. Schuller a , Mattias N.R. Wohlfarth b
a Perimeter Institute for Theoretical Physics, Waterloo N2J 2W9, Ontario, Canada
b Department of Applied Mathematics and Theoretical Physics, Centre for Mathematical Sciences,
University of Cambridge, Wilberforce Road, Cambridge CB3 0WA, United Kingdom

Received 12 March 2004; accepted 28 July 2004
Abstract
A general geometrical scheme is presented for the construction of novel classical gravity theories
whose solutions obey two-sided bounds on the sectional curvatures along certain subvarieties of the
Grassmannian of two-planes. The motivation to study sectional curvature bounds comes from their
equivalence to bounds on the acceleration between nearby geodesics. A universal minimal length
scale is a necessary ingredient of the construction, and an application of the kinematical framework
to static, spherically symmetric spacetimes shows drastic differences to the Schwarzschild solution
of general relativity by the exclusion of spacelike singularities.
PACS: 04.20.Cv; 04.50.+h; 02.40.-k
1. Introduction
The history of modifications of Einsteins theory of gravity is a long and winding one, and its beginnings almost date back to the publication of general relativity itself [14]. A main motivation for all proposed modifications of general relativity comes from the fact that the spacetime solutions of this theory, even in simple
E-mail addresses: fschuller@perimeterinstitute.ca (F.P. Schuller), m.n.r.wohlfarth@damtp.cam.ac.uk
(M.N.R. Wohlfarth).
doi:10.1016/j.nuclphysb.2004.07.040
320
F.P. Schuller, M.N.R. Wohlfarth / Nuclear Physics B 698 (2004) 319334
cases, may contain curvature singularities. The existence of such solutions creates problems, both mathematical and physical; a famous example is the black hole information paradox which seems to make it impossible to reconcile general relativity with
quantum mechanics [5]. The solution of these problems is usually postponed to an unforeseeable future with the claim that a quantum theory of gravity would smooth out
the singularities. It is indeed sensible to expect a radically altered short distance behaviour in quantum gravity and string theory, and this becomes apparent, respectively,
in the quantisation of the area or volume operators in loop quantum gravity, or in
the appearance of higher order curvature corrections to the EinsteinHilbert action in
string theory, see, e.g., [6,7]. But postponing the solution of existing classical problems to a future quantum theory means accepting to quantise the already problematical classical theory. It might well be the case that this causes unnecessary additional difficulties. For this reason alone, it is worthwhile to change the perspective and
to reverse the procedure: try to improve on classical general relativity before quantising it.
Modified theories of gravity should conform to a number of criteria, of which the most
important ones are that they should reduce to general relativity in a limit that covers the
range of observations and they should not contain ghost poles. In the construction of such
theories, one can essentially embark on two different routes. The first is to change the stage,
i.e., to enrich or alter the geometrical structure of the spacetime manifold; examples falling
into this category are theories with non-vanishing torsion, like EinsteinCartan theory, e.g.,
[8], or those based on a non-symmetric metric, e.g., [9]. The second possibility is to provide
a new screenplay, i.e., to change the dynamics of the gravitational field. This happens, for
example, in string theory which provides higher order corrections to the EinsteinHilbert
action while retaining the Lorentzian structure of the spacetime manifold. The standard
way of proceeding, in either case, is to formulate a consistent new action principle and
to hope that the solutions of the corresponding equations of motion have nicer properties
than the solutions to Einsteins theory. Such constructions, however, can be at best educated guesses, because underlying geometrical concepts, which are of utmost importance
in any theory of gravity, are very hard, or even impossible, to identify. In this article, the
programme is a different one. Our starting point is the physical notion of accelerations
between nearby geodesics, and we will investigate the possibility to place bounds on these
accelerations. The kinematical structure of these bounds will be analysed and the results
will be used for the construction of novel gravity theories. In consequence, all solutions of
these theories will obey the acceleration bounds. It is clear that this might not be enough
to get rid of all curvature singularities, but, in any case, this is an important first step of
improvement on classical general relativity.
Throughout this article, we use the geometrical setting of a d-dimensional torsion-free
manifold (M, g) with a symmetric metric g and the compatible metric connection which
satisfies g = 0. Consider a congruence X of affinely parameterised geodesics such that
X X = 0 and a connecting vector field Y with [X, Y ] = 0. Then the acceleration, in the
direction Y , between nearby geodesics of the congruence X is measured by the geodesic
deviation equation, or Jacobi equation,
X X Y = R(X, Y )X,
(1)
321
where R(X, Y )Z = X Y Z Y X Z [X,Y ] Z defines the Riemann tensor for all vector fields X, Y, Z. The projection of the above acceleration onto the connecting vector Y is
given by the RiemannChristoffel tensor, which carries all its indices downstairs, since

g(X X Y, Y ) = g R(X, Y )X, Y = R(X, Y, X, Y ).
(2)
In order to place a bound on this physically meaningful quantity, we have to normalise the
expression. There are essentially two ways of doing so. The first is simply to divide out the
squared lengths of the vectors X and Y , i.e., to normalise by g(X, X)g(Y, Y ). The second
possibility is to divide by the square of the area of the parallelogram spanned by X and Y ,
i.e., by g(X, X)g(Y, Y ) g(X, Y )2 . It is the latter possibility that we will choose because
it relates the projection of the geodesic deviation acceleration to the concept of sectional
curvatures of the manifold.
We will consider the kinematics of bounded sectional curvatures, in Riemannian and
Lorentzian geometries in turn, in Sections 2.1 and 2.2. In Section 3, we will devise gravitational dynamics that automatically enforce these bounds for all solutions of the gravitational field equations. The crucial requirement that the theory should be free of ghosts
narrows down the admissible gravitational actions to those with a DiracBornInfeld invariance group, as will be explained in Section 4. Highly interesting applications of the new
framework, especially to the case of static, spherically symmetric spacetimes, will be derived in Section 5. There, we will find that the kinematics of bounded sectional curvatures
imply drastic changes in the spacetime structure in comparison to the Schwarzschild solution. In particular, spacelike singularities will be excluded. We conclude with a discussion
in Section 6.
2. Kinematics of bounded sectional curvatures

2.1. Riemannian geometry
Consider a d-dimensional Riemannian spacetime (M, g) with a symmetric positive definite metric g. The sectional curvature Sp (X, Y ) at a point p M with respect to a plane
spanned by two vectors X, Y Tp M is the total curvature of the two-surface of geodesics
through p and tangent to that plane. It is induced from the RiemannChristoffel tensor by
Sp (X, Y )
R(X, Y, X, Y )
,
G(X, Y, X, Y )
(3)
where the tensor G : (Tp M)4 R is defined by

G(X, Y, Z, W ) g(X, Z)g(Y, W ) g(X, W )g(Z, Y ).
(4)
Note that G(X, Y, X, Y ) is the squared area of the parallelogram formed by X and Y ,
so that the sectional curvature Sp (X, Y ) determines exactly the normalised value of the
projection of the geodesic deviation acceleration on the connecting vector, as has been
discussed in the introduction. The tensor G has the same symmetry properties as the
RiemannChristoffel tensor and the squared parallelogram area G(X, Y, X, Y ) is always
322
positive as long as X and Y are linearly independent. For later use, we define the set

Wp (X, Y ) (Tp M)2 | X, Y linearly independent .
(5)
The sectional curvature is a function on the Grassmannian of two-planes at the point p,

i.e., Sp : G2 (Tp M) R. This follows from the fact that, under a non-singular change of
basis from X, Y to X , Y , according to
X = aX + bY,

Y = cX + dY,
(6a)
(6b)
which is a GL(2, R)-transformation with non-vanishing determinant (ad bc) = 0, both

R(X, Y, X, Y ) and G(X, Y, X, Y ) are multiplied by the same positive factor (ad bc)2 .
Thus, the value of the ratio Sp (X , Y ) = Sp (X, Y ) stays unchanged. The Grassmannian
G2 (Tp M) can be understood as the set of pairs of linearly independent vectors Wp with its
elements identified under the equivalence relation of a GL(2, R) basis change, G2 (Tp M)
=
Wp /GL(2, R). Note that the Grassmannian is not a vector space, but a variety. We will
explain in Section 2.2 why this is a point of crucial importance when studying Lorentzian
manifolds. A further important fact is that all sectional curvatures at a point p determine
the Riemann tensor there uniquely [10].
Now we establish a bound on the absolute value of all sectional curvatures and require

S(X, Y ) 1
(7)
for all linearly independent (X, Y ) Wp and for all p M (dropping the subscript p). The
constant has the dimension length squared. Therefore, any theory of gravity realising
such a bound would involve a fundamental, very short length scale at which considerable
differences to known gravity solutions could be expected. The requirement of bounded
sectional curvature can be rewritten in the form
H [ab] [cd] X[a Yb] X[c Y d] 0.
(8)
This condition has to be satisfied for both signs, which arise from the removal of the absolute value, and the H are defined by
H [ab] [cd] = [ab] [cd] R [ab] [cd] ,
(9)
where arises from G by raising two indices. The H are linear maps H : (Tp M)2
2
Tp M. Due to the symmetries of G and R, they map any two-tensor into an antisymmetric two-tensor. For the discussion of the condition on the sectional curvatures,
and consider
however, wecan restrict H to endomorphisms of antisymmetric two-tensors
H End( 2 Tp M). The endomorphism simply becomes the identity on 2 Tp M.

We need another definition. An element 2 Tp M is called simple, if it can be
decomposed as = X Y for X, Y Tp M. We write

2
Tp M | simple .
Qp
(10)

Any element of 2 Tp M can be written as a sum of simple elements because the basis
elements of this space can be chosen to be simple. This allows us to consider the tensor
323

G, defined in (4), as a map G : ( 2 Tp M)2 R by using the identification G(X Y, Z
W ) = G(X, Y, Z, W ).
are symmetric with respect to G, i.e., G(, H ) =
The endomorphisms H
G(H , ) for all , 2 Tp M. Therefore, the H are diagonalisable with real

eigenvalues by an orthogonal transformation S O(dim( 2 Tp M)) which satisfies
G(S, S) = G(,
). This implies that there exists an orthonormal basis {I |
G(I , J ) = I J } in 2 Tp M of common eigenvectors of H + and H (or, more pre
cisely, of R) with H I =
I I . In terms of this basis, we may write any plane as a

linear combination, X Y = I aI I . From (8), the bound on the sectional curvatures
then is seen to be equivalent to

2
G X Y, H (X Y ) =
(11)
a I I 0
I
for all (X, Y ) Wp and all points of the spacetime manifold. This is satisfied if, and only
if, the eigenvalues

I are non-negative, I 0. Note that this is a fully covariant requirement because the characteristic polynomials of H , and hence their eigenvalue equations

det H = 0,
(12)
are diffeomorphism-invariant. Hence, the following holds:
An upper bound (7) on the absolute value of all sectional curvatures of a Riemannian
manifold is equivalent to the requirements that

(i) the endomorphisms H End( 2 Tp M) defined in (9) should be diagonalisable

by an element S of the local orthogonal group O(dim( 2 Tp M)), where dim( 2 Tp M) =
d(d 1)/2, such that S 1 H S = diag(
I );
(ii) H should have non-negative eigenvalues
I 0.
The first requirement is, of course, redundant in the Riemannian setting. But the formulation used above lends itself much better to a generalisation to the Lorentzian case than the
simpler statement that H should be positive semi-definite.
2.2. Lorentzian geometry
In the Lorentzian case, the spacetime metric is indefinite with signature (1, d 1)
(, +, , +). Consequently, the squared area G(X, Y, X, Y ) in the definition of sectional
curvature in (3) may have either sign or even vanish. Accordingly, a plane spanned by
the basis vectors X, Y Tp M is called timelike when G(X, Y, X, Y ) < 0, spacelike for
G(X, Y, X, Y ) > 0 and null for G(X, Y, X, Y ) = 0. Although the sectional curvature thus
is only defined on non-null planes, the knowledge of it on all these planes still determines
the Riemann tensor uniquely.
Bounding the sectional curvatures on all non-null planes by condition (7) is equivalent
to requiring that

G X Y, H (X Y ) 0 for G(X Y, X Y ) > 0,
(13a)

G X Y, H (X Y ) 0 for G(X Y, X Y ) < 0

(13b)
324
for all (X, Y ) Wp and for all p M. Again these conditions have to be satisfied for
both signs. Now comes an important point! While in the Riemannian case a large class of
manifolds possesses bounded sectional curvatures, it turns out that imposing such bounds
in the Lorentzian case is extremely restrictive: on null planes, the Riemann curvature has
to vanish, R(X, Y, X, Y ) = 0. Otherwise the sectional curvature could not be bounded for
nearby planes [11]. But then it follows that the sectional curvature at any given point of
the manifold has the same value for every non-null plane [12]; hence, the whole manifold
is one of constant curvature [13]. Clearly, this is not desirable for a theory of gravity as it
could never cover a large enough range of solutions! These arguments show that a theory of
gravity on a Lorentzian spacetime manifold cannot have bounds on all sectional curvatures.
But the same conclusion, that the manifold has constant curvature, already follows from
certain weaker assumptions, namely from bounds on either all timelike or all spacelike
planes, or from either an upper or a lower bound on all non-null planes [1315].
Thus, in order to avoid these rigidity theorems, a less restrictive condition has to be
devised. Such a condition amounts to a selection of planes for which the sectional curvatures should be bounded. Observe that the above restriction of the sectional curvature
map S : G2 (Tp M) R to all non-null (timelike, spacelike) planes in the Grassmannian
variety G2 (Tp M) is not a natural one from the point of view of algebraic geometry, as
these sets are subsets but not subvarieties of G2 (Tp M). A morphism between varieties,
however, can only be sensibly restricted to subvarieties. In the following, we will therefore identify the maximal subvariety contained in the set of timelike (spacelike) planes and
only constrain
the sectional curvature for these planes. More precisely, consider the vector

space 2 Tp M which has already played an important rle in the Riemannian construction. Not all elements of that space describe planes, but only the simple ones, Qp .
Each plane corresponds to a pair of linearly independent basis vectors, i.e., to an element
of Wp . But Wp /SL(2, R) is isomorphic to Qp \ {0} via (X, Y )
X Y . So Qp additionally contains a degenerate plane for which X becomes
parallel to Y . The set of planes

Qp forms a variety embedded in the vector space 2 Tp M, for its elements satisfy the
polynomial equation = 0, compare, e.g., [16]. (If the lengthof the basis vectors
is projected out, one gets the Plcker embedding G2 (Tp M) P( 2 Tp M).) A mathematically well-motivated subset of planes, on which one might hope to achieve bounded
sectional curvatures in Lorentzian manifolds, therefore, should be a subvariety of the set
of all planes Qp and not, for instance, the subset of all timelike planes which is not a
subvariety since G(X, Y, X, Y ) < 0 is not a polynomial equation.
We return to the conditions (13) for bounded sectional curvature on all non-null planes.
The set of timelike planes for which the conditions hold can be written as the intersection

2

Qp
Tp M | G(, ) < 0, G , H 0 ,
(14)
and in an analogous form with reversed relations for the spacelike planes. This set is not a
subvariety of Qp , as it is not defined by additional polynomial equations. But into it, we
can embed a maximal subvariety by constructing the vector space of maximal dimension
which can be embedded into the set in curly brackets. The intersection of such a linear
subspace with Qp is then the desired subvariety. We show now that this vector space can
325
be found, in the present Lorentzian case, by requiring the conditions (i) and (ii) on the
endomorphisms H which we have established in the Riemannian case.
By (i), the endomorphisms H should be diagonalisable by the local orthogonal group
which is now O(d 1, (d 1)(d 2)/2). This requirement is non-trivial in the Lorentzian
case: the H are still symmetric, but now with respect to an indefinite metric, which does
not guarantee diagonalisability. Requiring diagonalisability, now, implies the existence of
an orthonormal basis

I , I | I = 1 (d 1), I = 1 (d 1)(d 2)/2
(15)
2
Tp M with G(I , J ) = I J , G(I , J ) = IJ and G(I , J ) = 0 which conof
sists of common eigenvectors of H with corresponding eigenvalues {

I , }. And by
I
(ii), these eigenvalues

I and I should be non-negative.
Both these requirements together imply that the linear span of timelike eigenvectors

I | I = 1 (d 1)
(16)

is the sought-for highest-dimensional linear subspace of 2 Tp M which can be embedded
into the set in curly
brackets appearing
in Eq. (14). This is most easily seen from the
decomposition = I aI I + I aI I , which gives

G , H =
(17)
aI2
aI2 .
I +
I
For elements of the linear span (16) of the timelike eigenvectors, the above expression
is clearly negative. Another basis vector that could be added to enlarge the dimension of
this vector space would necessarily be spacelike, so that the augmented space could not
be properly embedded anymore. An analogous argument where the spacelike eigenvectors
I replace the timelike eigenvectors I holds for the case of the spacelike planes. Thus
we have shown the following.
On a Lorentzian manifold, the requirements (i) and (ii) on the endomorphisms H , as
formulated at the end of the preceding subsection, imply bounds
on the sectional curvatures

of certain subvarieties of the variety Q of all planes in 2 T M. In terms of the common
eigenvector basis {I , I } of H , these subvarieties are given by the timelike planes in
I Q and by the spacelike planes in I Q.
Note that this mechanism of choosing the planes of bounded sectional curvature is a dynamical one, as it is the Riemann tensor (of particular spacetime solutions) that determines
the eigenvectors of the endomorphisms H . This construction, by bounding the sectional
curvatures on certain subvarieties of the variety of all planes, hence allows to circumvent
the rigidity theorems obtained by Nomizu and others.
3. Gravitational dynamics
We have now established algebraic conditions that allow placing sensible bounds on the
sectional curvatures along certain planes. But how does one implement that solutions of
326
gravitational equations, yet to be devised, automatically give rise to non-negative eigenvalues for the endomorphisms H ? A fairly general answer to this question is to use, in the
equations of motion, absolutely converging power series in the Riemann tensor that converge exactly on a domain with non-negative eigenvalues. For definiteness, consider the
square roots
A
A
H B = R B ,
(18)
2
Tp M. The index A [ab] may take
where we use the Petrov notation for indices of
d(d 1)/2 values and corresponds to a pair of antisymmetrised indices of Tp M. If the
above expressions appear in the equations of motion, then any diagonalisable solution will
force the eigenvalues of H to be real and non-negative. Otherwise the solution will become complex.
It is very convenient, for a number of reasons, to derive the gravitational dynamics from
an action principle. Not only does this allow to check easily whether the modified theory
of gravity is consistent with Einsteins theory in a certain limit, which is sufficient for its
compatibility with experiment, it also simplifies the discussion and exclusion of ghosts.
Gravitational ghosts are modes of negative kinetic energy in a perturbative expansion of
the metric around the Minkowski vacuum. Ghosts appear whenever there are quadratic
terms in the curvature expansion of the action; the only exception from this rule is the
quadratic GaussBonnet combination

2
2
2
,
g R 4Rab
+ Rabcd
(19)
which is a total derivative in four dimensions, and for which, also in higher dimensions,
the kinetic terms of the perturbative expansion cancel.
As a consequence of the kinematical construction above, the Lagrangian for an action
implying
bounded sectional curvatures should be constructed from a power se
partially nA
a
(R)
ries
B with the desired finite convergence radius to enforce non-negative
n
n=0
eigenvalues for H . The requirement of exact convergence on the domain of non-negative
eigenvalues is
satisfied when the power series is obtained from functions with a branch
cut, such as 1 z in the example (18). Furthermore, the theory should be ghost-free. It
is natural to use the same power series for H + and H , i.e., for , in constructing the
dynamics. Hence, the Lagrangian should be built from the expression

n=0
an (R)nA B
an (R)nA B ,
(20)
n=0
where the minus sign between the sums is dictated by the requirement of ghost-freedom.
This is the only possible way to achieve ghost-freedom in the proposed setup: contractions
of the Riemann tensor in Petrov notation can never yield the Ricci tensor which appears
in the GaussBonnet term. Note that the cancellation of the ghosts automatically kills an
otherwise possible cosmological constant term which, however, could again be included
by hand without inconsistency. Because of their absolute convergence, the two series may
be rearranged, and they collapse to

n=0

a2n+1 (R)(2n+1)A B = a1 R A B + O 3 .
(21)
327
The simplest way to build a scalar action from the above expression is to act on it with the
linear integral-trace functional, to obtain

(22)
g Tr
an (R)nA B .
n odd
This is then seen to be a direct generalisation of Einsteins theory, where a2n+1 = 0

for n > 0. So one particular action whose solutions by construction have bounded sectional curvature (on all planes in the Riemannian case, and on certain subvarieties in the
Lorentzian case) could be taken as

A
A
1
Sexample =
(23)
g Tr H + B H B .
The equations of motion are derived in the usual manner by variation with respect to the
spacetime metric gab : they present a system of fourth-order partial differential equations.
In the limit 0, this action reproduces the EinsteinHilbert action up to a constant that
can be divided out. The same limit simplifies the equations of motion to yield Einsteins
vacuum equations Rab 12 Rgab = 0.
Although the action (22) appears to be the mathematically most straightforward one, we
could have made several alternative choices in the construction. Another obvious choice
would have been to take a determinant instead of a trace. It is also possible first to make
scalars out of the power series for and then to subtract. This would lead, for instance,
to the action

+
g det H det H .
(24)
The first part of this expression is almost of the form as has been considered in [17].
There are, however, two important differences here. Firstly, the inclusion of the second
term bounds the value of the sectional curvatures (on the planes that can be linearly combined solely from spacelike or timelike eigenvectors of H ) additionally from below, and,
notably, this removes the ghosts as well. Secondly, it is important to take the square root of
the matrix first and then the determinant, because this is what constrains the eigenvalues of
H when only real solutions are considered.
4. DiracBornInfeld invariance
Our construction of gravity theories with partially bounded sectional curvatures features
a fundamental length scale, appearing in the constant which has the dimension length
squared. Hence, one expects that gravitational dynamics respecting that scale should lead
to solutions very different
from those of general relativity, especially on scales which are
small compared to . Such a crucial change in the short distance behaviour should also
be mirrored in structural differences. To show that this is indeed the case, we need to briefly
review the kinematics associated to DiracBornInfeld theory.
The singularities of Maxwell electrodynamics, most prominently the divergence of the
electric field energy density of a point charge, prompted Born and Infeld in [18] to devise a
328
non-linear theory of electrodynamics, which regulates this divergence by the introduction

of a parameter of dimension length squared. In string theory, the corresponding action
emerges [19] as the low energy effective action of a U (1) gauge field A, with corresponding
field strength F = dA, on a Dp-brane ,

g det( + F ).
(25)
This theory is manifestly invariant under worldvolume diffeomorphisms of the (p + 1)dimensional Lorentzian manifold , and hence possesses a local O(1, p) Lorentz symmetry. The feature of a finite length scale in (25), however, has profound structural consequences. Based on the observation that this action only depends on even powers of the
electromagnetic field strength F , it has been shown in [20] that there is a hidden dynamical
invariance, extending the local invariance group to a product group,
O(1, p) O(1, p) O(1, p),
(26)
into which the Lorentz symmetry group is diagonally embedded. An investigation of the
kinematical meaning of the product group, acting on curves in the frame bundle, shows that
it contains, apart from the standard Lorentz transformations, transformations to arbitrarily
rotating and accelerated frames, whose covariant
acceleration for non-rotating observers is
bounded by the inverse of the length scale, 1/ . As much as the extension of the rotation
group O(3) to O(1, 3) captures the dynamical symmetries of Maxwell theory due to the
existence of a fundamental speed, the extension (26) is implied by the existence of a fundamental squared length scale in DiracBornInfeld theory. That the extended group also
captures the regulation of BornInfeld theory is further illustrated by the following fact: if
one defines particles in quantum field theory as the irreducible representations of a corresponding doubling of the Poincar group, then every particle automatically is accompanied
by a PauliVillars regulating Weyl ghost [21].
These insights are intimately linked to the findings of the present paper: in the previous
chapter, we have seen that the criterion of ghost-freedom for gravity theories with partially
bounded sectional curvatures leads to actions of the type
(22), to which solely odd powers
of the Riemann tensor, viewed as an endomorphism on 2 Tp M, contribute. Hence one can
show in close technical analogy to [20] that gravitational ghost-freedom can be understood
as singling out those theories that possess a hidden invariance (26). In turn, the extended
structure group O(1, p) O(1, p) appears not only to capture the regulation of Born
Infeld theory, but also the absence of ghosts there, and transfers this property to curvature
regulated gravity theories in our sense.
In summary, we arrive at the remarkable fact that ghost-free gravity theories with tidal
acceleration bounds, as constructed in the preceding sections,
automatically imply kine
matics that put an upper bound of the same value 1/ on covariant accelerations as
well.
These structural insights justify the classification of the gravity theories presented in
this paper as of BornInfeld type.
329
5. Simple applications
The appeal of the geometrical construction scheme presented here is the characterisation
of a spacetime with partially bounded sectional curvatures, and hence, partially bounded
geodesic deviation accelerations, by a simple algebraic property of the Riemann tensor:
the endomorphisms H should be diagonalisable and have real non-negative eigenvalues. Even without considering specific gravitational dynamics this has interesting consequences, valid for the whole class of theories constructed here. Two applications will be
considered: for asymptotically flat, static, spherically symmetric spacetimes we will show
the non-existence of spacelike singularities, and a discussion of pp-wave backgrounds will
illustrate some limitations of the proposed scheme.
5.1. pp-wave spacetimes
The first application considers pp-wave spacetimes (plane-fronted waves with parallel
rays) which are defined by the existence of a covariantly constant null Killing vector field
/v. The line element can be written in the form

ds 2 = f u, x i du2 2 du dv + dx dx,
(27)
and it can be shown that in these spacetimes all curvature invariants vanish [22]. This holds,
in particular, for the invariants

Tr R nA B ,
for n N,
(28)
which determine the eigenvalues of H since they appear as the coefficients of the characteristic polynomial in Eq. (12). For the pp-wave spacetimes, the eigenvalue equation
becomes ( 1)d(d1)/2 = 0, and so all eigenvalues equal = 1, as is the case for
Minkowski space. This also means that the Riemann tensor, although it is non-trivial, is
nilpotent. More explicitly, in four dimensions and in Petrov notation using the labelling
([uv], [ux 1 ], [ux 2], [vx 1 ], [vx 2 ], [x 1x 2 ]) (1, . . . , 6), one finds the non-zero components

R24
R25
R34
R35

=
1
2
x21 F
x 1 x 2 F
x 1 x 2 F
x22 F

.
(29)
This Riemann tensor is not diagonalisable

by a local O(3, 3) transformation, since it does

not admit an eigenvector basis for 2 Tp M.
But because the Riemann tensor is nilpotent with R 2A B = 0, the pp-waves are solutions
of any theory constructed within our framework, as they solve Einsteins theory on the remaining first order level of the curvature expansion, compare (22). The polarisation of the
waves may diverge and lead to null-singularities. These singularities appear in the tidal
forces given by the geodesic deviation equation and have been argued to be unavoidable
in any theory of gravity whose Lagrangian depends only on the metric, covariant derivatives and the Riemann tensor [23]. In particular, our construction does not escape that
conclusion.
330
5.2. Static spherical symmetry

The second, particularly nice example considers static, spherically symmetric solutions
in four dimensions with a spacetime ansatz of the form
ds 2 = A(r) dt 2 + B(r) dr 2 + r 2 d22 ,
(30)
where A and B are arbitrary functions of the radial coordinate r and d22 gives the standard
line element on the unit two-sphere.
Now we establish sectional curvature bounds in the sense of Section 2.2, by assuming
that the endomorphisms H , which are already diagonal in the adjoint basis induced by
the Schwarzschild coordinates, have non-negative eigenvalues. One of these eigenvalues is
algebraic in B. The corresponding condition is
1
1 B(r)
0
r 2 B(r)
(31)
and implies that B is bounded

from below for all r by the function Blow (r) = /( + r 2 )
and from above, for r < , by Bup (r) = /( r 2 ). For asymptotically flat spacetimes,
this implies that B is strictly positive everywhere and that B(0) = 1 with slope B (0) = 0.
An argument given in [24] shows that such behaviour of the function B is necessary in
any theory of gravity on a Lorentzian spacetime manifold that regulates the Schwarzschild
singularity. The framework presented here obviously passes this test.
When the gravitational equations of motion are constructed from an action
principle,
there is another implicit requirement to satisfy. The density g = r 2 sin AB in the

integrand is only well-defined when
A(r)B(r) 0
(32)
everywhere. But as B(r) > 0 this shows A(r) 0.

Already at this point, we see that theories with partially bounded sectional curvatures
show a completely different behaviour than does general relativity, where, by Birkhoffs
theorem, the Schwarzschild solution is the unique static, spherically symmetric and asymptotically flat vacuum solution, with A = B 1 = 1 2m/r for a mass parameter m.
For positive mass m > 0, the Schwarzschild spacetime has a horizon at r = 2m, in whose
interior A and B become negative. All timelike and null geodesics are trapped behind this
horizon, and there is a spacelike singularity at r = 0. In our theories with partially bounded
sectional curvature, however, a change of signs in the functions A and B is forbidden.
Hence, hypersurfaces of constant r can never be spacelike, such that spacelike singularities
are excluded from all possible solutions. (Singularities can only occur at constant radius
due to spherical symmetry.) This mechanism does not prevent the occurrence of timelike
singularities or, as seen previously, null singularities.
Horizons may still appear in solutions. To see this more explicitly consider the spherically symmetric spacetimes in which the functions A and B are finite for all r 0. When
A > 0, i.e., when A is strictly positive everywhere, nothing interesting happens. There are
no horizons and the spacetime is conformally equivalent to Minkowski space. Now assume that A vanishes at an isolated point, A(r0 ) = 0. In the usual manner, it is possible to
331
Fig. 1. The CarterPenrose diagram of static, spherically symmetric, asymptotically flat spacetimes with A, B
finite and strictly positive everywhere on r 0 except A(r0 ) = 0 demonstrates the possibility of horizons in
theories with partially bounded sectional curvatures. Several asymptotic regions with their own future and past
infinities occur. A few surfaces of constant radius are indicated.
introduce null coordinates by setting

r
r (r) =
dr
B(r )
A(r )
1/2
(33)
and defining u = t r and v = t + r , where u and v are constant on outgoing or ingoing

radial null geodesics, respectively. These coordinates allow the continuation of the radial
null geodesics across r = r0 because it takes a finite affine parameter distance 10 to reach
this hypersurface from a point r1 > r0 ,
r1
10

1/2
dr A(r)B(r)
.
(34)
r0
This argument extends and shows the existence of a horizon at r = r0 which, however,
is permeable from both sides. The CarterPenrose diagram of the resulting spacetime is
shown in Fig. 1.
Finally, let us take a closer look at the planes on which the sectional curvatures are
bounded in the static spherically symmetric case. We use the Petrov notation and the labelling ([tr], [t ], [t], [ ], [r], [r ]) = (1, . . . , 6). The six-vectors Z A = (Z1 , Z2 ) that
describe planes then are those which satisfy Z1 Z2 = 0. This follows from the choice of
basis X = (X0 , X) and Y = (Y0 , Y) which implies
Z A = (X0 Y Y0 X, X Y).
(35)
332
The endomorphisms
2 H 2 are diagonal in the static spherically symmetric example, as is
the tensor G : ( Tp M) R, for which

GAB = diag AB, Ar 2 , Ar 2 sin2 , r 4 sin2 , Br 2 sin2 , Br 2 .
(36)
The timelike and spacelike eigenvectors of H are I = (eI , 0) and I = (0, eI ), respectively. The sectional curvature is thus bounded on the subvarieties of all timelike
planes described by Z A = (Z, 0) with BZ12 + r 2 (Z22 + sin2 Z32 ) = 1/A (where the Z A
are normalised to avoid overcounting). More explicitly, these planes can be described by
an orthonormal basis: given any normalised timelike vector X with g(X, X) = 1, we
can calculate
the second, spacelike basis vector Y with g(Y, Y ) = 1 from (35). When
X=
(1/ A, 0), we find Y = (0, Y) for any normalised three-vector Y. For X0 = 1/ A,

we find Y = (Y0 , X0 X/Y0 ) where

AX02 1
.
Y0 =
(37)
A
All these planes are easily visualised:
they contain the local time axis /t. (This is obvious
in the first case with X0 = 1/ A, while in the second case it follows from the fact that
X is parallel to Y = X0 X/Y0 .) A similar calculation for the spacelike planes shows that
among those with bounded sectional curvature are the planes orthogonal to the local time
axis, defined by dt = 0.
6. Discussion
In contrast to past attempts to modify Einsteins theory of gravity, where the consequences of new actions, differing from the EinsteinHilbert action and proposed in a more
or less ad hoc fashion, are analysed, this work has been based on clear-cut geometrical assumptions. Motivated by bounds on the deviation accelerations between nearby geodesics,
we have investigated possible bounds on the sectional curvatures of a spacetime, with the
aim of potentially improving on the singularity problems of classical general relativity.
In the Riemannian case, the sectional curvatures may be bounded for all planes, whereas
on the physically relevant Lorentzian manifolds, the situation is more involved. Due to the
existence of strong rigidity theorems, certain subsets of planes have to be chosen to avoid
trivial theories allowing for constant curvature solutions only. A consistent choice of such
subsets can be made by demanding that they should form subvarieties of the variety of all
possible planes through a given point.
We have shown that the bounds on the sectional curvatures along maximal subvarieties
of the set of all planes can be reformulated in a nice algebraic way. The conditions for these
bounds translate into the requirement that certain endomorphisms, which are constructed
from the Riemann tensor and act on the space of antisymmetric two-tensors, should be
diagonalisable by the local orthogonal group and should have non-negative eigenvalues. In
addition to these kinematical results, it has been shown that it is possible to construct ghostfree gravity actions such that the dynamical solutions of their corresponding equations of
motion automatically satisfy the sectional curvature bounds. This is strong evidence that
333
there exist classical gravity theories on Lorentzian manifolds that are better-behaved than
Einsteins theory.
The strengths and limitations of the kinematical framework constructed here have been
illustrated by two instructive examples. The expectation that a regularisation, at least for
some types of singularities, might be possible has been found justified. This expectation
emerged from the DiracBornInfeld structure of our theories and from earlier results in
[17], where actions similar to those obtained here had been investigated. In spherically
symmetric spacetimes in four dimensions, without even considering specific dynamics, it
has been shown that there cannot be spacelike singularities, which is a very strong result,
as it applies to the whole class of curvature regulated theories in our sense. Another application to pp-wave spacetimes explains why null singularities cannot be removed in our
theories.
A new fundamental length scale appears as an important ingredient in the construction.
We have shown that the appearance of the length scale here is intimately linked to its rle in
DiracBornInfeld theory [25]. The covariant acceleration of non-geodesic observers turns
out to be bounded by the same value that one chooses to impose on tidal accelerations. It
would be interesting to understand the thereby established relations to other theories with
fundamental length scales, such as, for example, string theory or quantum gravity, which
all predict modifications of general relativity. A study of exact, or numerical, solutions of
specific gravitational dynamics constructed within our framework will allow to constrain
the possible range for the length scale parameter from comparison with observations.
Our investigations here indicate the emergence of at least one further generic feature
besides the finite, fundamental length scale: the existence of curvature corrections to all
orders seems to be an essential ingredient in order to regulate gravity.
Acknowledgements
The authors are grateful to Paul K. Townsend and Achim Kempf for useful discussions.
F.P.S. thanks the Mathematical Institute at Oxford University for the opportunity to work
there. M.N.R.W. thanks the Perimeter Institute where this research was begun for its generous hospitality. He gratefully acknowledges financial support from the Gates Cambridge
Trust.
References
[1] H. Weyl, Gravitation und Elektrizitt, Sitzungsber. Preu. Akad. Wiss. Berlin 26 (1918) 465;
H. Weyl, Eine neue Erweiterung der Relativittstheorie, Ann. Phys. 59 (1919) 101.
[2] A.S. Eddington, The Mathematical Theory of Relativity, Cambridge Univ. Press, Cambridge, 1923.
[3] A. Einstein, The Meaning of Relativity, 4th ed., Methuen, 1950.
[4] E. Schrdinger, SpaceTime Structure, Cambridge Univ. Press, Cambridge, 1950.
[5] S.W. Hawking, Breakdown of predictability in gravitational collapse, Phys. Rev. D 14 (1976) 2460.
[6] T. Thiemann, Lectures on Loop Quantum Gravity, Lecture Notes in Physics, vol. 631, Springer, Berlin,
2003, p. 631.
[7] J. Polchinski, String Theory, vols. I, II, Cambridge Univ. Press, Cambridge, 1998.
[8] M. Blagojevic, Gravitation and Gauge Symmetries, Institute of Physics Publishing, Bristol, 2002.
334
[9]
[10]
[11]
[12]
[13]
[14]
[15]
[16]
[17]
[18]
[19]
[20]
[21]
[22]
[23]
[24]
[25]
J.W. Moffat, Regularity theorems in the nonsymmetric gravitational theory, J. Math. Phys. 36 (1995) 5897.
S.S. Chern, W.H. Chen, K.S. Lam, Lectures on Differential Geometry, World Scientific, Singapore, 1998.
B. ONeill, Semi-Riemannian Geometry with Applications to Relativity, Academic Press, New York, 1983.
M. Dajczer, K. Nomizu, On sectional curvatures in indefinite metrics. II, Math. Ann. 247 (1980) 279.
E. Garca-Ro, D.N. Kupeli, R. Vzquez-Lorenzo, Osserman Manifolds in Semi-Riemannian Geometry,
Springer, Berlin, 2002.
L. Graves, K. Nomizu, On sectional curvature of indefinite metrics, Math. Ann. 232 (1978) 267.
R.S. Kulkarni, The values of sectional curvature in indefinite metrics, Comment. Math. Helv. 54 (1979) 173.
J. Harris, Algebraic Geometry, Springer, New York, 1992.
M.N.R. Wohlfarth, Gravity la BornInfeld, Class. Quantum Grav. 21 (2004) 1927.
M. Born, L. Infeld, Foundations of the new field theory, Proc. R. Soc. A 144 (1934) 425.
E.S. Fradkin, A.A. Tseytlin, Non-linear electrodynamics from quantized strings, Phys. Lett. B 163 (1985)
123.
F.P. Schuller, H. Pfeiffer, Invariant length scale in relativistic kinematics: lessons from Dirichlet branes,
Phys. Lett. B 578 (2003) 402.
F.P. Schuller, M.N.R. Wohlfarth, T.W. Grimm, PauliVillars regularization and BornInfeld kinematics,
Class. Quantum Grav. 20 (2003) 1.
G.T. Horowitz, A.R. Steif, Spacetime singularities in string theory, Phys. Rev. Lett. 64 (1990) 260.
G.T. Horowitz, R. Myers, The value of singularities, Gen. Relativ. Gravit. 27 (1995) 915.
B. Holdom, On the fate of singularities and horizons in higher derivative gravity, Phys. Rev. D 66 (2002)
084010.
F.P. Schuller, Almost product manifolds as the low energy geometry of Dirichlet branes, Eur. Phys. J. C
(2004) 006.
Evolution of cosmological perturbations

in the universe dominated by resonant scalar fields
Takashi Hamazaki
Kamiyugi 3-3-4-606 Hachioji-city, Tokyo 192-0373, Japan
Received 9 January 2004; accepted 5 August 2004
Abstract
Recently a Hamiltonian formulation for the evolution of the universe dominated by multiple oscillatory scalar fields was developed by the present author and was applied to the investigation of
the evolution of cosmological perturbations on superhorizon scales in the case that scalar fields have
incommensurable masses.
In the present paper, the analysis is extended to the case in which the masses of scalar fields
satisfy resonance conditions approximately. In this case, the action-angle variables for the system
can be classified into fast changing variables and slowly changing variables. We show that after an
appropriate canonical transformation, the part of the Hamiltonian that depends on the fast changing
angle variables can be made negligibly small, so that the dynamics of the system can be effectively
determined by a truncated Hamiltonian that describes a closed dynamics of the slowly changing
variables. Utilizing this formulation, we show that the system is unstable if this truncated Hamiltonian
system has hyperbolic fixed point and as a consequence, the Bardeen parameter for a perturbation of
the system grows.
PACS: 98.80.Cq
E-mail address: yj4t-hmzk@asahi-net.or.jp (T. Hamazaki).

doi:10.1016/j.nuclphysb.2004.08.008
336
T. Hamazaki / Nuclear Physics B 698 (2004) 335385
1. Introduction and summary

The inflationary universe model is the most successful model in explaining the origin of the present cosmological structures such as galaxies and clusters of galaxies. In
this model, quantum fluctuations of an inflaton field, a scalar field driving the inflationary expansion, provide seed perturbations which grow and form the present cosmological
structures by gravitational instability. During the slow rolling phase of inflation, these seed
perturbations are stretched beyond the Hubble horizon and their wavelengths stay larger
than the horizon scale until the perturbations come back inside the Hubble horizon during the Friedmann stage after the inflation. The amplitudes of perturbations at this second
horizon crossing, which have a direct relevance to the CMB anisotropy observations and
provide the initial condition for detailed astrophysical models of galaxy formation, are determined by the so-called Bardeen parameter. Hence, in order to obtain information on the
inflationary stage of the universe from observations of the present universe, we have to
determine the behavior of the Bardeen parameter of perturbations during the superhorizon
stage.
If the cosmic matter has a regular equation of state and is dominated by a single component, this Bardeen parameter is conserved with a good accuracy on superhorizon scales
[2]. However, in a realistic model, during the period between the first horizon crossing in
the inflationary regime and the socond horizon crossing in the Friedmann regime, the inflaton field oscillates coherently around a local minimum of a potential and their energy
is gradually transformed into matter and radiation which constitute the present universe.
During this reheating phase, the equation of state becomes singular periodically [6], and
entropy modes can be produced [7,11]. Further, in an inflationary model with a multiplecomponent inflaton field, isocurvature modes can appear [11,12]. Therefore, the Bardeen
parameter may not be converved even approximately in realistic models, and a detailed
analysis of its behavior is mandatory.
On superhorizon scales, this problem can be reduced to the analysis of a spatially
homogeneous model for the following reasons. First, the evolution of cosmological perturbations on superhorizon scales is well described by that in the long wavelength limit
with a good accuracy [6]. Further, the long wavelength limit of a solution to the perturbation equation can be easily constructed from a homogeneous perturbation of the
background universe model [35]. Hence, we only have to analyse a spatially homogeneous system, which is much simpler than the analysis of the exact perturbation equation
with a finite wave number. If the background system is exactly solvable, this reduction
solves the problem completely. However, in the case of multiple scalar fields in an expanding universe, it is not the case. Further, the system exhibits quite complicated oscillatory
behavior.
There is, however, one powerful method to treat such an oscillatory system. It is the
Hamiltonian formulation in terms of the action-angle variables [15]. In fact, in a previous
paper by the present author [8], we have developed a Hamiltonian formulation for a universe model dominated by multiple oscillatory scalar fields and have shown that it works
well at least in the case in which the scalar fields have incommensurable masses. In this
formulation, we introduce an expansion parameter that represents the ratio of the cosmic
expansion rate to the masses of the scalar fields, and decompose the Hamiltonian is into an
337
unperturbed part that depends only on the action variables and a perturbative part that is
of the order and bounded by a constant multiple of 1/t, where t is a time parameter of
the system given by a 3/2 in terms of the cosmic scale factor a. Next, we look for a canonical transformation that transforms the perturbative part to a quantity of higher order with
respect to and 1/t. As was shown in Ref. [8], we can construct such a canonical transformation iteratively, and by repeated applications of such transformations, we can transform
the perturbative part to a quantity of the order of an arbitrary power of and 1/t. Furthermore, the new Hamiltonian system obtained by these transformations becomes solvable if
the small perturbative part is neglected. By this method, we have proved that the Bardeen
parameter is conserved with a good accuracy under the assumption that the masses of the
scalar fields do not satisfy resonant relations.
In the reheating phase, a dynamical instability caused by the parametric resonance plays
a crucial role in the energy transfer from a macroscopic homogeneous mode to finite
wavelength modes [9]. This instability can have a significant effect on the evolution of cosmological perturbations [10,13]. In fact, a numerical example showing a non-conservation
of the Bardeen parameter was presented in [11]. In order to treat this problem, it is necessary to extend our Hamiltonian formulation to the resonant case. From this point of view,
in this paper, we undertake this extension and with the help of it, we investigate the dynamical behavior of the universe dominated by multiple oscillatory scalar fields whose masses
satisfy a resonance condition at least approximately.
The present paper is organized as follows. First, in the next section, on the basis of the
paper [4], we explain how to construct a solution to the perturbation equation in the long
wavelength limit from an exactly spatially homogeneous perturbation of the background
universe model. Then, in Section 3, we put the spatially homogeneous system of multiple
scalar fields in an expanding universe into the Hamiltonian form and introduce the actionangle variables.
In Section 4, for the case in which the masses of scalar fields satisfy resonance conditions approximately, we decompose the action-angle variables to fast changing variables
and slowly changing variables, and reduce the dynamics of the system to that of slowly
changing variables by a canonical transformation. Then, with the help of this formulation,
we estimate the growth rate of perturbations of the system in the case in which the time
parameter t of the system is smaller than 1/. On the basis of this estimate and the analysis
of some soluble examples, we argue that the Bardeen parameter of the system can grow if
the system has a hyperbolic fixed point. Next, in Section 5, we analyze the evolution of a
perturbation of the system in the time range t > 1/ and show that the Bardeen parameter
is conserved with a good accuracy in this time range. Section 6 is devoted to discussions.
In order to make the presentation clear, the proofs of most of the mathematical statements
and the calculation of the growth rates of perturbations in concrete models are given the
Appendices AE.
Throughout the paper, the natural units c = h = 1 are adopted, and 8G is denoted
as 2 . Further, the notation adopted in the article [1] is used for perturbation variables, and
their definitions are sometimes omitted except for those newly defined in this paper.
338
2. Evolution of cosmological perturbations in the long-wavelength limit

As mentioned in the introduction, we can determine the dynamical behavior of cosmological perturbations on superhorizon scales by studying an exactly spatially homogeneous
system. Since the perturbative analysis of the latter system can be used to determine its
Lyapunov exponent, which is an index for a dynamical instability including chaos of a
Hamiltonian system, we can also analyse the dynamical instability and integrability of the
inflaton dynamics by such a study.
In this section, we summarize the main results of the paper [4] in the case in which the
universe is dominated by multi-component scalar fields, and explain how to construct a
solution to the perturbation equation in the long-wavelength limit from an exactly homogeneous perturbation of the model. We assume that the universe is spatially flat (K = 0)
throughout the paper. Hence, the background metric is given by
ds 2 = d 2 + a( )2 dx 2 .
(2.1)
We consider the universe dominated by multi-component scalar fields whose energy

momentum tensor is given by

1
T = + 2U .
(2.2)
2
In the long-wavelength limit, the gauge-invariant variable Yi representing the fluctuation
of the scalar field i in the flat time slice can be expressed as

i
H2
W
, ,
Yi = i +
d
H
2U
H
W (X1 , X2 ) := X1 X 2 X 1 X2 ,
(2.3)
where i is the combination of the exactly homogeneous perturbation of i , i , and the
perturbation of the cosmic scale factor a, a, given by
i a
.
(2.4)
H a
From the equations of motion, it follows that this quantity always satisfies the equation

a 3H 2
(2.5)
W
, = const.
U
H
i = i
In general, the general solution to the homogenous dynamical system can be expressed
in terms of the scale factor a and a set of integration constants as
i = i (a, C).
(2.6)
For this expression, i can be simplify written as

i =
i (a, C)
.
C
(2.7)
339
As explained in the introduction, one of the most important quantities describing the
evolution behavior of perturbations is the Bardeen parameter defined by
=R
aH
g ,
k
(2.8)
where R and g are perturbations of the three curvature and the shear of each constant time
slice, respectively. In the present case, the Bardeen parameter can be written in terms of
the gauge-invariant variable Yi as
= H
Y
.
2
()
(2.9)
Hence, from (2.3), is represented in the long-wavelength limit as

H2

W
, .
d
= H
2
2U
H
()
(2.10)
From
(2.5), the second term on the right-hand side of this equation is proportional to

d/a 3 , which rapidly approaches a constant as the cosmic time increases. Hence, in
order to see whether the Bardeen parameter is conserved or not in the superhorizon stage,
we can concentrate on the first term. For this reason, from this point, we assume that is
expressed as
= H

.
2
()
(2.11)
3. Evolution equations of corresponding exactly homogeneous universe

If we use the cosmic scale factor a as the time variable, the action for the homogeneous dynamical system introduced in the previous section can be put into the following
Hamiltonian form

S=
(3.1)
pi di ha da,
i
1/2

2 3
1 1 2
a 2
p
+
U
()
,
ha =
i
2 2a 6
(3.2)
U () =
1 2 2
mi i + Uint (),
2
(3.3)
where Uint is assumed to be a sum of monomials in of degrees not less than 3. After
changing the time variable to
3/2
a
t=
(3.4)
,
a0
340
this action can be expressed in terms of the non-dimensional canonical variables

=
,
0
P =
p
,
3
a 0 m 0 0
(3.5)
and the dimensionless parameters
3
mi
0 ,
,
=
i =
m0
2
as
S =
=
(3.6)
S
a03m0 02

Pi di
1/2

2 11 2
1 2 2 2
2 1
Pi + t
i i + t 2 2 Uint(0 )
dt.
2 t2
2
m 0 0
i
(3.7)
Now, let us introduce the action-angle variables (Ji , i ) by

2 1
i =
Ji cos i ,
i
t

Pi = 2i Ji t sin i ,
(3.8)
(3.9)
which corresponds to the canonical transformation generated by the generating function

1
W (, , t) =
(3.10)
i t 2 i2 tan i .
2
i
Then, the Hamiltonian of the system is transformed to

1/2
Ji
2
t2
sin 2i .
H=
i Ji + 2 2 Uint

t
m 0 0
i
i
Further, in terms of these new variable, the Bardeen parameter is expressed as
1/2

2
1
t2
=
J + 2 2 Uint
i
i
3t
m 0 0
i i J (1 cos 2 )

1

J i sin 2 i J i 1 cos 2 i i .
(3.11)
(3.12)
4. Evolution for t 1/

In this section, we evaluate the growth rate of a perturbation of the system defined by
the Hamiltonian (3.11) during the period 1 t 1/, assuming that the masses of scalar
341
fields satisfy a resonant condition approximately. Before going to general arguments, we

first explain the basic ideas by two simple models consisting of two scalar fields.
The first model is defined by
Uint () = 12 2 ,
21 2 .
(4.1)
If we change the canonical variables from (i , Ji ) (i = 0, 1) to (qi , pi ) (i = 0, 1) by the

linear symplectic transformation
1 = q0 ,
2 = 2q0 + q1 ,
J1 = p0 2p1 ,
J2 = p1 ,
(4.2)
we obtain
1 J1 + 2 J2 = 0 p0 + 1 p1 ,
(4.3)
where
0 = 1 ,
1 = 21 + 2 .
(4.4)
Uint can be written

t2
1/2
Uint = J1 J2 cos (21 2 ) + cos (21 + 2 ) + 2 cos 2
2
2
t
m 0 0

1/2
= (p0 2p1 )p1 cos q1 + cos (4q0 + q1 ) + 2 cos (2q0 + q1 ) ,
t
(4.5)
where
=
0
1
.
2
m 0 1 2 2
(4.6)
Hence, the contribution of this interaction term to the Hamiltonian is of order 1/t, and
when averaged over q0 and q1 , it becomes of order 1/t 2 . It generally holds on the interaction terms of the third degree of . This feature will play an important role in the argument
in Section 5.
The second model is defined by
Uint () = 12 22 ,
1 2 .
(4.7)
By the linear symplectic transformation

1 = q0 ,
2 = q0 + q1 ,
J1 = p0 p1 ,
J2 = p1 ,
(4.8)
we obtain
1 J1 + 2 J2 = 0 p0 + 1 p1 ,
(4.9)
where
0 = 1 ,
1 = 1 + 2 .
(4.10)
342
The interaction term is now written

t2
1
1
U
=
J
J
int
1 2 1 + cos (21 22 ) + cos (21 + 22 ) + cos 21
2
2
t2
m20 02

+ cos 22

1
1
= 2 (p0 p1 )p1 1 + cos 2q1 + cos (4q0 + 2q1 ) + cos 2q0
t
2
2

+ cos (2q0 + 2q1 ) ,
(4.11)
where
=
02
1
.
m20 1 2
(4.12)
Hence, in this model, the contribution of this interaction term to the Hamiltonian is of order
1/t 2 , and its dominant part does not vanish even after it is averaged with respect to q0 and
q1 , in contrast to the first model.
In both of these models, 1 represents the deviation from the exact resonance, which is
given by 21 = 2 for the first model and by 1 = 2 for the second model, respectively.
Hence, 1 is much smaller than 0 , and as a consequence, the motion of q0 is much faster
than that of q1 in general. For this reason, we call q0 and q1 the fast angle variable and the
slow angle variable, respectively.
This behavior of the variables qi suggests that the dynamics of the slow variable is well
described by a Hamiltonian H obtained from H by averaging it with respect to the fast
angle variable q0
1
H =
2
2
dq0 H.
(4.13)
For the first model, this averaged Hamiltonian can be expressed as
2
1
2
1
1
1/2
,
H = ( p)1/2 +
(p
2p
)p
cos
q
+
O
0
1
1
1
1/2

t ( p)
t2
and for the second model, as
(4.14)

2
1
1
1
1
2
1/2
,
(p0 p1 )p1 1 + cos 2q1 + O
H = ( p) + 2
1/2

t ( p)
2
t4
(4.15)
where p = 0 p0 + 1 p1 .
This simple procedure, however, does not give a useful approximation, since the higherorder terms with respect to 1/t still contains the large parameter 1/, as is seen from the
above expressions.
Now, we will show for a generic system consisting of n ( 2) oscillatory scalar fields
that this difficulty can be resolved by taking the average after applying an appropriate
343
canonical transformation to the Hamiltonian. For that purpose, we transform the original action-angle variables to the fast canonical variables (q 0 , p 0 ) and the slow variables
(q 1 , p1 ), as in the above examples.
For the time being, suppose that the masses of the system satisfy n1 resonance relations
of the form k = 0, where k is a vector with irreducible integer coefficients. Let R =
(Rij ) be a unimodular integral matrix such that its last n1 rows are given by the n1 vectors
k defining the resonant relations, and consider the canonical transformation generated by
W = pi Rij j ,
W
= Rj k k ,
pj
W
= pi Rij .
Jj =
j
qj =
(4.16)
(4.17)
Here, note that due to the unimodularity of R, we can assume that each of the new angle
variables qi also has the period 2 .
Now, let us decompose these new variables (q, p) into two sets of variables as
(q 0 )j = qj ,
(p 0 )j = pj
(q 1 )i = qn0 +i ,
(j = 1, . . . , n0 ),
(p 1 )i = pn0 +i
(i = 1, . . . , n1 ),
(4.18)
(4.19)
where n0 + n1 = n. Then, the set (q 0 , p 0 ) becomes the fast variable and the set (q 1 , p1 )
becomes the slow variable, because J can be written as
J = p0 0 ,
(4.20)
where
(0 )i := Rij j
(i = 1, . . . , n0 ).
(4.21)
In the above, we have imposed exact resonant conditions on the mass parameters. However, such strong conditions are rarely satisfied. In fact, it is known that for d > n 1, the
set of points in the space satisfying the condition

inf n |k|d (k ) = 0,
(4.22)
k=0,kZ
has measure zero, where

|k| := |k1 | + |k2 | + + |kn |.
(4.23)
Therefore, from this point, we only require that the mass parameters satisfy resonance
conditions approximately. To be precise, we assume that
inf
k 0 =0,kZn
inf
k=0,kZn
|k|d |k 0 0 + k 1 1 | = C > 0,
|k|d |k 0 0 + k 1 1 | C,
(4.24)
(4.25)
where C is a constant of order unity. When satisfies (4.24) for some positive constants
d, C, we say that is of the class D0 (d, C).
344
Next, we show that there exists a canonical transformation such that in a system obtained
by that transformation, the dynamics of the slow variables can be determined independent
of the fast angle variables with a good accuracy. For that purpose, first note that the Hamiltonian of the system can be written in terms of the variables q 0 , p0 , q 1 and p1 as
2
H = (0 p 0 + 1 p1 )1/2 + A(q 1 , p, t) + B(q, p, t),

by Taylor expanding H with respect to 1/t, where
(1 )i := Rn0 +i j j
(i = 1, . . . , n0 ).
(4.26)
(4.27)
We say that this Hamiltonian is of the type Cm (, M1 , M2 , ), if the following conditions

are satisfied for some positive constants , M1 , M2 , and :
(i) p = 0 p0 + 1 p 1 is bounded as
| p| .
(4.28)
(ii) tA can be extended to an analytic function in the domain D1 () in Cn+n1 +1 defined

by

1
D() :=
p, q,
; Re p D + , | Im p| , | Im q| ,
t

1
1

1 + , Im ,
|t|
t
D1 () := D()|q 0 =0 ,
(4.29)
(4.30)
where D + denotes the -neighbourhood of an interval D, and in this domain, satisfies

the inequality
|tA| M1 .
(4.31)
Further, tA is periodic with respect to q 1 and real if (q 1 , p, 1/t) are real.

(iii) t m B can be extended to an analytic function in the domain D() C2n+1 defined
above and satisfies the inequality
m
t B m1 M2 .
(4.32)
Further, t m B is periodic with respect to q, real if (q, p, 1/t) is real, and satisfies
1
(2)n0
2
2
d n0 q 0 t m B = 0.
(4.33)
Under this notation, the following proposition holds.

Proposition 4.1. Let m be some positive interger, and consider the Hamiltonian H (m)
(m)
written in terms of fast canonical variables (q (m)
0 , p 0 ) and slow canonical variables

(m)
345
(m)
(q 1 , p1 ) as
H (m) =
(m)

2
(m)
(m) 1/2
0 p 0 + 1 p 1
+ Am q 1 , p (m) , t

+ Bm q (m) , p(m) , t .
(4.34)
Suppose that is of the class D0 (d, C) and that this Hamiltonian is of the type
Cm (m , M1(m) , M2(m) , m ). Then, for any > 0, there exists 0 > 0 such that, for an arbitrary satisfying
0 < < 0 ,
(4.35)
there exists a function Sm (q, p, t) satisfying the following conditions:

(i) Sm is periodic with respect to q and real if (q, p, t) is real.
(ii) t m Sm can be extended to an analytic function in the multi-dimensional complex domain D(m+1 ), where m+1 = m , and in this domain, satisfies the inequality
m
t Sm m L(m) ,
(4.36)
1
for some positive constant L(m)
1 .
(iii) Let H (m+1)(q (m+1) , p(m+1) , t) be a Hamiltonian obtained from H (m) by the canonical transformation generated by Sm (q (m) , p (m+1) , t):
Sm
,
q (m)
Sm
,
q (m+1) = q (m) +
p(m+1)
Sm
H (m+1) = H (m) +
t

2
(m+1) 1/2
= p
+ Am+1 q (m+1)
, p (m+1) , t
1

+ Bm+1 q (m+1) , p(m+1) , t .
p (m) = p(m+1) +
(4.37)
(4.38)
(4.39)
(4.40)
Then, H (m+1) is of the type Cm+1 (m+1 , M1(m+1) , M2(m+1) , m+1 ) for some positive
constants m+1 , M1(m+1) , and M2(m+1) , and the change of the A-term in the Hamiltonians satisfies the inequality
m+1
m (m+1)
t
M
,
Am+1 (q 1 , p, t) Am (q 1 , p, t)
2 2
for (q 1 , p, 1/t) D1 (m+1 ).
(4.41)
(For the proof see Appendix A.)

Since the generating function Sm in this proposition is of the order m /t m , we have
Bm m1 /t m and Bm+1 m /t m+1 . This implies that we can make the part of the
Hamiltonian that depends on the fast variables q 0 arbitrarily small by taking the original
Hamiltonian as the starting point (m = 1) and applying canonical transformations given
346
in the proposition repeatedly. Therefore, we can expect that the evolution of the slow variables (q 1 , p1 ) can be determined with a quite good accuracy from an effective Hamiltonian
(m)
H (m) (q 1 , p(m) , t) obtained by discarding Bm term from H (m) , if we take m sufficiently
(m)
(m)
large. In this truncated system, p0 becomes constant, and the behavior of q 0 can be
obtained by a simple time integration of a function that has a definite t-dependence when
(m)
(m)
p(m)
0 and a corresponding solution for (q 1 , p 1 ) are given. The evolution of the original
variables can be calculated from this solution by applying the known canonical transformation connecting these two sets of variables.
Next, we show that this expectation is correct by estimating the errors produced by
the trunction. Let us use the symbol to represent the difference of a quantity for the
exact system and the corresponding quantity for the truncated system. Further, let us denote a quantity for the truncated system by the upper case letter of the lower case symbol
representing the corresponding quantity for the exact system. For example, the errors of
canonical variables are denoted as
Q = q Q,
(4.42)
P = p P ,
(4.43)
and the errors of solutions to the perturbation equations for both systems are denoted as
Q = q Q,
(4.44)
P = p P .
(4.45)
Further, let us denote the sets of slow variables, (q 1 , p1 ) and (Q1 , P 1 ), by z and Z, respectively. Finally, for a function f (t), let us define f (t) by

f (t) = sup f (s).
(4.46)
1st
Before evaluating Z, Z and Z, we give the general technique to evaluate an upper
bound on the norm of the solution X to the first order differential equation
d
(4.47)
X = X + S,
dt
where X, S are N column vectors and is N N matrix. For the detail, see the appendix
of the paper [7]. If we define the norm of the solution X by
|X|2 := X X,
(4.48)
we decompose into a sum of 1 , 2 : 1 from the perturbed part is of order 1/t and 2
from the unperturbed part is of order and mi is the largest eigenvalue of the Hermitian
matrix H i := i + i , |X| is bounded as
d
|X| |X| + |S|,
dt
2
(4.49)
:= m1 + m2 .
(4.50)
where
347
If is an eigenvalue of H i , is also an eigenvalue of H i , because i can be written

in the form of I Si where Si is the symmetric matrix and I is defined by (4.79). So mi are
non-negative. Therefore, the norm of the solution is bounded as

dt X(1) + dt|S| .
|X| exp
(4.51)
2
1
In evaluation of the contribution from the source term, we use

k

k

S
|S i |,

i

i=1
(4.52)
i=1
which is obtained from the CauchySchwarz inequality

A B |A||B|.
Since the different definition of the norm

|X|m := max (X)i
1iN
(4.53)
(4.54)
satisfies
1
|X| |X|m |X|,
N
(4.55)
we can identify |X| and |X|m assuming that the number of degrees of freedom N is of
order unity, so we simply omit the subscripts m.
Then, the truncation error for the fourth-order system can be estimated as follows. From
now on we omit constant coefficients of order unity except .
Proposition 4.2A. Let be an upper bound of the eigenvalues of the Hermitian matrix
tH 1 /2 defined by
H 1 := 1 + 1 ,
2A
Q1 P 1
1 :=
2A
Q1 Q1
2A
P 1 P 1
,
2A
P 1 Q1
(4.56)
where A is the A-term of the fourth-order Hamiltonian H = H (4) , and all the elements of
1 are bounded by M1(4)/ 2 t in the domain D1 (4 ). This quantity gives an upper
bound on the growth rates of the errors Q1 , P 1 , Q1 , and P 1 , and the growth of
the perturbations Q1 , P 1 .
Let us define by
=
1
.
+1
(4.57)
348
Then, for an arbitrary positive , in the time interval

1
,

the truncation errors of canonical variables are given by

Q0
3
2
2

|P 0 | ,
|Z| ,
t ,
1t
and the errors of perturbation variables are

Q0

2

t Q0 (1) + P 0 + Z ,

P 0 3 Q0 (1) + 2 P 0 + 3 Z ,

Z 2 Q0 (1) + P 0 + Z ,
(4.58)
(4.59)
(4.60)
under the initial conditions:

P 0 (1) = Z(1) = Q0 (1) = 0,
(4.61)
Q0 (1) = P 0 (1) = Z(1) = 0.
(4.62)
(For the proof see Appendix B.1.)

This proposition shows that the truncation errors can be made small in the fourth-order
system (m = 4). In order to obtain the information on the behavior of the original variables,
we also have to estimate how this truncation error affects the original variables through the
canonical transformation which connects the fourth-order system and the original system.
The next proposition gives that estimate.
Proposition 4.2B. The difference between (q (1) , p(1) ) obtained from (q (4) , p (4)) by the
canonical transformation and (Q(1) , P (1) ) obtained from (Q(4) , P (4) ) by the same canonical transformation has the upper bound

Q(1)

0
2
P (1) 3 ,
Z (1) 2 ,

(4.63)
0
t ,
and the corresponding difference in the perturbation variables is estimated as

Q(1)

(4)
(4)

0

2 Q0 (1) + P 0 + Z (4) ,

t

P (1) 2 Q(4)(1) + P (4) + 2 Z (4),
0
0
0

Z (1) 2 Q(4) (1) + P (4) + Z (4) ,
0
0
(4.64)
in the same time interval as in the previous proposition.

(For the proof see Appendix B.2.)
These two propositions show that the truncation errors can be made small even with
respect to the original variables, if we truncate the system at the fourth-order. This order
349
is minimal in the sense that the trunction at a lower-order system produces errors of order
unity in the perturbation variables. Conversely, if we go to higher-order systems, we can
obtain a better approximation. Further, we can prolong the time interval in which the approximation is good. In fact, we can show that in the mth order system with m 4, the
same estimates for the truncation errors as in the above propositions hold in the interval
1t
1
(m3)/( +1)
(4.65)
In particular, for m larger than + 4, the approximation is good in the interval 1 < t < 1/.
Finally, let us evaluate the growth of perturbations of a truncated higher-order system.
The Hamiltonian of such a system can be in general expressed as
2
H = (0 P 0 + 1 P 1 )1/2 + A(Q1 , P , t),

where
|tA| M,
1 = O().
(4.66)
(4.67)
As mentioned before, the fast action variables P 0 for this system become constants of
motion, and the equations of motion for the slow variables (P 1 , Q1 ) do not contain small
parameter essentially. Therefore, in contrast to P 0 , the slow action variables P 1 are not
conserved in general, and as a consequence, the Bardeen parameter can grow considerably
in the time interval 1 t 1/. (For estimate of the growth rates of the Bardeen parameter in the concrete examples, see Appendix E.) This can be confirmed by the following
estimate of the upper bound of the perturbation variables Q and P .
Proposition 4.3. In the time interval 1 t 1/, the perturbation variables Q and P
are bounded as

|Z| t Z(1) + P 0 (1)(t 1) ,
(4.68)

|P 0 | = P 0 (1),
(4.69)

1

t +1 1
|Q0 | Q0 (1) + P 0 (1)(t 1) + Z(1)

+1

t +2 1 t +1 1

,
+ P 0 (1)
+2
+1
(4.70)
where is the quantity defined in Proposition 4.2B. In particular, the growth rate of the
perturbation variables is not exponential and at most a power of t.
Proof. The evolution of the slow variables Q1 and P 1 is determined by the equation
2A
1 1
2A

+

Q1 P 1
d Q1
2( P )3/2 P 1 P 1
Q1
=
P 1
dt P 1
2A
2A
Q1 Q1
P 1 Q1
350
1 0
2A
+
2( P )3/2 P 0 P 1
.
+ P 0
2A
P 0 Q1
By evaluating the coefficients in the right-hand side, we obtain

d
|Z|
+ |Z| + P 0 (1),
dt
t
(4.71)
(4.72)
where we have used P 0 = P 0 (1). By integrating this inequality, we obtain

|Z| t exp (t 1) Z(1) + P 0 (1)(t 1) .
(4.73)
Since we are considering the time interval 1 t 1/, we have

|Z| t Z(1) + P 0 (1)(t 1) .
(4.74)
Next, Q0 obeys the equation

1 0
1
d
(0 P 0 + 1 P 1 )
Q0 =
dt
2 ( P )3/2

A
+ P 0
+ Q1
+ P 1
.
P 0
Q1
P 1 P 0
(4.75)
Integrating the right-hand side of this equation with respect to t yields

1

1
|Q0 | Q0 (1) +
dt P 0 (1) + dt|P 1 | + dt P 0 (1) + |Z|

t
1
1
1

1

Q0 (1) +
(4.76)
dt P 0 (1) + dt|Z|.

1
From (4.74) and this equation, we obtain (4.70). 2

Next, we show that the above general estimate of the upper bound on the growth of perturbations is rather good, by presenting an example in which the upper bound is nearly
saturated. Let us consider a system in which the Hamiltonian flow in the phase space
(Q1 , P 1 ) has a equilibrium fixed point. For simplicity, we assume the exact resonance
condition 1 = 0. Then, in terms of the phase space variable Z defined by
Z i = Qi1
(1 i n1 ),
Z n1 +i = P i1
(1 i n1 ),
(4.77)
the Hamiltonian equations of motion can be written

dZ
H
(4.78)
=I
,
dt
Z
where I is the matrix of degree 2n1 expressed in terms of the unit matrix E of degree n1
as

0
E
I=
.
(4.79)
E 0
351
We consider the Hamiltonian of the form

2
1
H = (0 P 0 )1/2 + a(Q1 , P ),
(4.80)

t
where = 1 or = 2. When the resonant interactions come from interaction terms of
the third order in the scalar fields , = 1, and when they come from interaction terms
of the fourth order in , = 2. We focus on the dynamical behavior near the fixed point
(P 0 (1), Z(1)) and expand a(P 0 , Z) as
1
1
a = a0 + bT P 0 + P T0 CP 0 + Z T DP 0 + Z T F Z,
(4.81)
2
2
dropping the terms of degree not less than 3 with respect to deviations from the equilibrium;
P 0 = P 0 P 0 (1), Z = Z Z(1), where T implies to take the transposition of a matrix,
a0 is a constant, b and P 0 are n0 -dimensional vectors, C is an n0 n0 symmetric matrix,
D is a 2n1 n0 matrix, and F is a 2n1 2n1 symmetric matrix. Since Z(1) is the fixed
points, a does not contain g T Z. Then, the evolution equations of the slow variables Z can
be written in the matrix form as
1
dZ
= [I F Z + I DP 0 ].
dt
t
Taking the variation of this equation and using the relations
P 0 = P 0 (1),
(4.82)
(4.83)
we obtain

1
d
Z = I F Z + I DP 0 (1) .
dt
t
Here, note that the 2n1 2n1 matrix X := I F satisfies the equation
XT = I XI.
(4.84)
(4.85)
From this and I 2 = 1, it follows that the characteristic polynomial

() = det(X E),
(4.86)
is an even function of :
() = ().
In addition, since () is real polynomial, we have

() = .
(4.87)
(4.88)
Hence, if a + bi, (a, b real) is an eigenvalue of X, all of a bi are also eigenvalues of X.

From this, it follows that if X has an eigenvalue with a non-vanishing real part, the fixed
point is unstable. We say that such a fixed point is hyperbolic. In contrast, if all eigenvalues
of X are pure imaginary, the fixed point is stable. We say that such a fixed point is elliptic.
In the case in which the fixed point is neither hyperbolic nor elliptic, i.e., some of the
eigenvalues of X vanish and the other eigenvalues are pure imaginary, the flow around the
352
fixed point is stable in the linear analysis but may become unstable if higher-order terms
are taken into account.
To be precise, this stability argument applies to this autonomous systems. In the present
case, due to the existence of the time-dependent factor 1/t , the real stability depends on
the value of . Since the system under consideration is linear, we can check it directly by
solving the equation. First, by diagonalizing the matrix X as
S 1 XS = ,
(4.89)
where is a diagonal matrix, the general solution for Z can be written

dt
Z = S exp
S 1 K X1 I DP 0 (1),
t
(4.90)
where
K = Z(1) + X1 I DP 0 (1).
(4.91)
Next, the equation for Q0 is given by

1 0 0 P 0 (1)
1
d
Q0 =
+ CP 0 (1) + Z T D .
1/2
dt
2 (0 P 0 (1))
t
(4.92)
Hence, Q0 can be expressed as

Q0 = Q0 (1)
where
R = CP 0 (1)
0 0 P 0 (1)
(t 1) + R,
2 (0 P 0 (1))1/2
dt
+
t
dt T
Z D.
t
(4.93)
(4.94)
When = 1, inserting the above expression for Z into this equation, we obtain
t 1
+ ln t,
for > 0, and

|R|
(4.95)
|R| ln t,
(4.96)
for = 0, where is the maximum of the real parts of the eigenvalues of X. Thus, Q0 is
always unstable. In contrast, when = 2, |R| is bounded from above as
|R| e + 1,
(4.97)
and the stability of Q0 depends on the value of P 0 .

Finally, we give an exactly soluble example such that there appear both hyperbolic fixed
points and elliptic fixed points and the phase flow pattern is similar to that of the pendulum
in a conservative field. It is given by the Hamiltonian

c
1 1 2
2
aP1 + b cos Q1 + 2 ,
H = (0 P0 )1/2 + 2
(4.98)

t 2
t
353
where a, b, c are positive functions of P0 .

Since there exists a conserved quantity,
1
C = aP12 + b cos Q1 = const,
(4.99)
2
in this model, we can determine the phase flow pattern in the (Q1 , P1 ) plane easily. In
particular, we find that the phase plain is divided into a region of oscillatory motions and a
region of rotationary motions by the separatrix defined by

Q1
b
sin
.
P1 = 2
(4.100)
a
2
Further, we find two types of fixed points. One is elliptic points, P1 = 0, Q1 = (2k + 1) ,
where k is an integer, each of which is surrounded by flows corresponding to oscillatory
motions. The other is hyperbolic fixed points, P1 = 0, Q1 = 2k, where k is an integer. These fixed points are connected by the heteroclinic orbits (4.100). This implies that
near the heteroclinic orbits, the perturbations Q and P , therefore the Bardeen parameter
grows.
5. Evolution for t 1/
We have treated the case in which is of the class D0 (d, C) and satisfies the resonance
relations at least approximately (4.25) for t 1/. When there exists deviation from the
exact resonance, as the time proceeds, the instability characteristic of resonance disappears
and the evolution of the system is reduced to the superposition of oscillations with amplitudes and frequencies of different orders. We demonstrate this fact by investigating the case
in which is of the classes D0 (d, C) and D(d, C) for t 1/: we say that is of the
class D(d, C) if
inf
k=0,kZn
|k|d |k 0 0 + k 1 1 | = C,
(5.1)
is satisfied for some positive constants d, C.

By redefining t, H by t, H /, respectively, we investigate the evolutionary behavior
for t 1 of the system defined by the Hamiltonian of the type C1,1 (, M11 , M12 , M2 , ):
we say that this Hamiltonian
2
( p)1/2 + A(q 1 , p, t) + B(q, p, t),
2
A(q 1 , p, t) = D(p, t) + E(q 1 , p, t),
H=
(5.2)
(5.3)
is of the type Cm,l (, M11 , M12 , M2 , ), if the following conditions are satisfied for some
positive constants , M11 , M12 , M2 , and :
(i) p = 0 p0 + 1 p 1 is bounded as
| p| .
(5.4)
354
(ii) t 2 D can be extended to an analytic function in the domain D2 () := D()|q=0 in

and satisfies the inequality
2
t D M11 .
(5.5)
Cn+1
Further, t 2 D is real if (p, 1/t) are real.

(iii) t l E can be extended to an analytic function in the domain D1 () Cn+n1 +1 and
satisfies the inequality
l
t E l1 M12 .
(5.6)
Further, t l E is periodic with respect to q 1 , real if (q 1 , p, 1/t) is real, and satisfies
1
(2)n1
2
2
d n1 q 1 t l E = 0.
(5.7)
(iv) t m B can be extended to an analytic function in the domain D() C2n+1 and
m
t B 2(m1)M2 .
(5.8)
Further, t m B is periodic with respect to q, real if (q, p, 1/t) is real, and satisfies
1
(2)n0
2
2
d n0 q 0 t m B = 0.
(5.9)
Since the interaction terms of the third degree of do not contribute D, D begins with
terms of order /t 2 .
First we show that we can make q 0 depenent part B small by the canonical transforma2 (, M , M , ) if the conditions with
tion S. We say that the Hamiltonian is of the type Cm
1
2
2
replaced with in the definition of the type Cm (, M1 , M2 , ) are satisfied.
Proposition 5.1. Let m be some positive interger, and consider the Hamiltonian H (m)
(m)
written in terms of fast canonical variables (q (m)
0 , p 0 ) and slow canonical variables
(m)
(m)
(q 1 , p1 ) as
H (m) =
(m)

2
(m)
(m) 1/2
0 p 0 + 1 p 1
+ Am q 1 , p (m) , t
2

+ Bm q (m) , p(m) , t .
(5.10)
Suppose that is of the class D0 (d, C) and that this Hamiltonian is of the type
Then, for any > 0, there exists 0 > 0 such that, for an arbitrary satisfying
2 ( , M (m) , M (m) , ).
Cm
m
m
1
2
0 < < 0 ,
(5.11)
355
there exists a function Sm (q, p, t) satisfying the following conditions:

(i) Sm is periodic with respect to q and real if (q, p, t) is real.
(ii) t m Sm can be extended to an analytic function in the multi-dimensional complex
domain D(m+1 ), where m+1 = m , and in this domain, satisfies the inequality
m
t Sm 2m L(m) ,
(5.12)
1
for some positive constant L(m)
1 .
(iii) Let H (m+1) (q (m+1) , p (m+1) , t) be a Hamiltonian obtained from H (m) by the canonical transformation generated by Sm (q (m) , p (m+1) , t):
Sm
,
q (m)
Sm
,
q (m+1) = q (m) +
p(m+1)
Sm
H (m+1) = H (m) +
t

(m+1) (m+1)
2
(m+1) 1/2
= 2 p
+ Am+1 q 1
,p
,t

(m+1) (m+1)
+ Bm+1 q
,p
,t .
p (m) = p (m+1) +
(m+1)
2
Then, H (m+1) is of the type Cm+1
(m+1 , M1
(m+1)
,
m+1 , M1
and
stants
(m+1)
M2
,
(m+1)
, M2
(5.13)
(5.14)
(5.15)
(5.16)
, m+1 ) for some positive con-
and the change of the A-term in the Hamiltonians
m+1
2m (m+1)
t
M2
Am+1 (q 1 , p, t) Am (q 1 , p, t)
,
2
for (q 1 , p, 1/t) D1 (m+1 ).
(5.17)
The proof is obtained by replacing with 2 in the proof of Proposition 4.1.

We show that the fast action variables p0 are perpetually stable and oscillate around the
initial values with amplitudes of order 2 and with frequencies of order 1/ 2 .
Proposition 5.2. There exists some constant C0 such that

(1)
p p (1) (1) 2 C0 .
0
0
(5.18)
Proof. We consider the transformed Hamiltonian given by

H (2) =
1/2
2
p(2)
+ A2 + B2 ,
2
(5.19)
where
(2)
|tA2 | M1 ,
2
t B2 2 M (2) ,
2
(5.20)
(5.21)
356
which gives
dp(2)
B2
0
= (2) .
dt
q
(5.22)
We obtain
(2)

p p (2) (1)
0
(2)
dt
2 M2
t2
2
M2(2)
.
(5.23)
By using the transformation law

2 (1)
(1)

p p (2) S1 L1 ,
q (1)
t
(5.24)
we obtain

(1)
p p (1) (1) p (1) p (2) + p(2) p(2) (1) + p(2) (1) p (1) (1)
0
0
0
0
0
0
0
0
(1)
(2)
L1
M
+ 2 2
which completes the proof. 2

2
(1)
+ 2
L1
,
(5.25)
Next we show that we can make q 1 dependent part E small by the canonical transformation T .
Proposition 5.3. Let m, l be some positive intergers, and consider the Hamiltonian H (m,l)
written in terms of fast canonical variables (q (m,l)
, p (m,l)
) and slow canonical variables
0
0
(m,l)
(m,l)
(q 1 , p 1 ) as
H (m,l) =
1/2

2
0 p(m,l)
+ 1 p(m,l)
+ Dm,l p (m,l) , t + Em,l q (m,l)
, p(m,l) , t
0
1
1
2

+ Bm,l q (m,l) , p (m,l) , t .
(5.26)
Suppose that is of the class D(d, C) and that this Hamiltonian is of the type
(m,l)
(m,l)
(m,l)
Cm,l (m,l , M11 , M12 , M2 , m,l ). Then, for any > 0, there exists 0 > 0 such that,
for an arbitrary satisfying
0 < < 0 ,
(5.27)
there exists a function Tm,l (q 1 , p, t) satisfying the following conditions:

(i) Tm,l is periodic with respect to q 1 and real if (q 1 , p, t) is real.
(ii) t l Tm,l can be extended to an analytic function in the multi-dimensional complex
domain D1 (m,l+1 ), where m,l+1 = m,l , and in this domain, satisfies the inequality
l

t Tm,l l L(m,l)
(5.28)
2
for some positive constant L(m,l)
.
2
357
(iii) Let H (m,l+1)(q (m,l+1) , p(m,l+1) , t) be a Hamiltonian obtained from H (m,l) by the
, p(m,l+1) , t):
canonical transformation generated by Tm,l (q (m,l)
1
(m,l)
p0
(m,l+1)
= p0
= p 1(m,l+1) +
p (m,l)
1
(5.29)
Tm,l
,
(m,l)
q 1
Tm,l
q (m,l+1) = q (m,l) +
,
p(m,l+1)
Tm,l
H (m,l+1) = H (m,l) +
t
1/2

2
= 2 p(m,l+1)
+ Dm,l+1 p (m,l+1) , t

(m,l+1) (m,l+1)

,p
, t + Bm,l+1 q (m,l+1) , p(m,l+1) , t .
+ Em,l+1 q 1
(m,l+1)
(m,l+1)
(5.30)
(5.31)
(5.32)
(5.33)
(m,l+1)
, M12
, M2
, m,l+1 ) for
Then, H (m,l+1) is of the type Cm,l+1 (m,l+1 , M11
(m,l+1)
(m,l+1)
(m,l+1)
some positive constants m,l+1 , M11
, M12
, and M2
, and the change of the
D-term in the Hamiltonians satisfies the inequality
l+1
l (m,l+1)
t
Dm,l+1 (p, t) Dm,l (p, t) M12
,
2
for (p, 1/t) D2 (m,l+1 ).
(5.34)
(For the proof see Appendix C.)

We show that the slow action variables p 1 are perpetually stable and oscillate around
the initial values with amplitudes of order and with frequencies of order 1/. Notice that
T is independent of q 0 .
Proposition 5.4. There exists some constant C1 such that
(1)

p p (1) (1) C1 .
1
1
(5.35)
Proof. We consider the transformed Hamiltonian H (m,l) (m = 2, l = 2) given by

H (2,2) =
1/2
2
p (2,2)
+ D22 + E22 + B22 ,
2
where
2

t D22 M (2,2),
11
2

t E22 M (2,2),
12
2

t B22 2 M (2,2).
2
(5.36)
(5.37)
(5.38)
(5.39)
The evolution equations for the slow action variables p1 are given by
(2,2)
dp1
dt
E22
(2,2)
q 1
B22
(2,2)
q 1

,
t2
(5.40)
358
which yields
(2,2)

(2,2)
p
p1 (1) C,
1
(5.41)
for some positive constant C.

(1,1)

(1,1)
p
(t) p 1 (1)
1
(2,2)
(1,1)
(2,2)

(2,2)
(2,2)
(1,1)
p 1 (t) p1 (t) + p1 (t) p1 (1) + p1 (1) p 1 (1)
C1 .
(5.42)
We investigate the dynamical behavior of the cosmological perturbations. We obtain the

(3,3)
(3,3)
(3,3)
Hamiltonian H (3,3) of the type C3,3 (3,3 , M11 , M12 , M2 , 3,3 ) by carrying out the
canonical transformations in the order S1 , S2 , T3,1 , T3,2 .
Later we omit the superscript (3, 3) and constant coefficients of order 1.
From now on we assume that

q 0 (1) p0 (1) q 1 (1) p1 (1) 1,
(5.43)
since in the linear perturbation, the scale of the perturbation variables is arbitrary. Then we
obtain the proposition below.
Proposition 5.5. The transformed perturbation variables are estimated as

p 0 p 0 (1) 2 ,

p 1 p 1 (1) ,

1 0
1
q 0 = 2 (t 1)
0 p 0 (1) + 1 p1 (1) + R 0 ,
3/2
2
( p(1))

1 1
1
0 p 0 (1) + 1 p1 (1) + R 1 ,
q 1 = 2 (t 1)
2
( p(1))3/2
(5.44)
(5.45)
(5.46)
(5.47)
where the residual part R 0 , R 1 is bounded as

|R 0 | C0 (t 1),
(5.48)
|R 1 | C1 (t 1),
(5.49)
for some positive constants C0 , C1 .

(For the proof see Appendix D.1.)
We have evaluated the canonically transformed variables q (3,3) , p(3,3) . Then we evaluate the difference between such transformed variables and the original variables q (1,1) ,
p(1,1) .
Proposition 5.6.

(1,1)
q
q (3,3) 1,
(1,1)

p
p(3,3) 1.
(5.50)
(5.51)
359
(For the proof see Appendix D.2.)

According to Propositions 5.5, 5.6, for t 1/, the Bardeen parameter stays constant
in a good accuracy.
6. Discussion
In this paper we have constructed the method for analyzing the resonance phenomena
of the Hamiltonian system obtained from the multiple oscillatory scalar fields in the expanding universe. We have shown that the perturbations including the Bardeen parameter
can grow when the resonant interactions between the homogeneous modes exist. If the
truncated Hamiltonian system has hyperbolic fixed points in the phase space of the slow
variables, the perturbations grow at the speed of a power of t in the three leg interaction
systems, while the growth of perturbations is bounded from above in no less than four leg
interaction systems. In the models where 1 = 0, we have found the network constructed
by the heteroclinic orbits around which perturbations are unstable. On this network, the
orbits become irregular, unperiodic, complicated and probabilistic. In order to evaluate the
amplitude of the Bardeen parameter which is directly related the cosmic structure formation, we must calculate the measure occupied by the stochastic network in the whole phase
space of the homogeneous modes.
Recently the relaxation in reheating is investigated numerically and is interpreted in
terms of the turbulence phenomena of the dynamical system of large number of degrees of
freedom [14]. In order to understand the turbulence phenomena, our fixed point analysis of
the Hamiltonian system may be useful.
Acknowledgements
The author would like to thank Professor H. Kodama for reading his manuscript carefully, helping him improve his English and giving him useful comments. The author would
also like to thank Professor V.I. Arnold for his writing the excellent textbooks and/or reviews [15], from which the author learned quite a lot about the Hamiltonian dynamical
system.
Appendix A. Proof of Proposition 4.1
We consider the canonical transformation induced by the generating function given by
(m+1) ikq (m)

Sk p
,t e
,
Sm p(m+1) , q (m) , t =
(A.1)
k 0 =0
(m)
(m)
where k q (m) = k 0 q 0 + k 1 q 1 and the sum is taken over k such that k 0 = 0. Bm is

decomposed as

(m+1) ikq (m)
bk p
,t e
,
Bm p(m+1) , q (m) , t =
(A.2)
k 0 =0
360
where the sum is taken over k satisfying k 0 = 0.

The transformed Hamiltonian is
H (m+1) = H (m) +
=
Sm
t
1/2 1
1
Sm (q (m) , p(m+1) , t)
2
p (m+1)
+
+ R1

( p (m+1) )1/2
q (m)
(m+1) (m+1)
+ Am q 1
,p
, t + R2

Sm (q (m) , p (m+1) , t)
,
+ Bm q (m) , p (m+1) , t + R3 +
t
where R1 , . . . , R3 are defined as
1/2 2
1/2
2
p(m)
p (m+1)

p(m) p(m+1) ,
(m+1)
1/2
( p
)
(m) (m)
(m+1) (m+1)
R2 = Am q 1 , p , t Am q 1
,p
,t ,
(m) (m)
(m) (m+1)
,t .
R3 = Bm q , p , t Bm q , p
(A.3)
R1 =
(A.4)
(A.5)
(A.6)
We determine the generating function Sm so that the leading term depending on the fast
angle variables q (m)
0 can be eliminated

1
1
Sm (q (m) , p (m+1) , t)
= Bm q (m) , p(m+1) , t .
(m+1)
1/2
(m)
( p
)
q
(A.7)
By comparing the Fourier components in the both hand sides, we obtain

1/2

Sk p(m+1) , t = i p (m+1)

1
bk p(m+1) , t .
( k)
(A.8)
Before we prove the analyticity and evaluate the upper bound of Sm (q (m) , p (m+1) , t), we
use the following lemma.
Lemma. As for the Fourier series

F (q) =
Fk eikq ,
(A.9)
where
kq =
n

ki qi .
(A.10)
(1) If F (q) is analytic and satisfies

F (q) C,
(A.11)
i=1
361
on the domain | Im q| , then

|Fk | Ce|k| ,
(A.12)
where
|k| = |k1 | + + |kn |.
(A.13)
(2) If on the domain | Im q| , Fk satisfies

|Fk | Ce|k| ,
(A.14)
then F (q) is analytic on the domain | Im q| , and for an arbitrary (0 < < ), on the
domain | Im q|
n

F (q) 4 C ,
(A.15)
n
where we assume 3.
Proof of Lemma.
(1) The Fourier coefficients Fk are given by
Fk =
1
(2)n
2
2
d n qeikq F (q).
(A.16)
By Cauchys theorem, the path of integration in the above integral can be shifted to qi =
xi i, 0 xi 2 where we choose the sign equal to ki , we get
|Fk | Ce|k| .
(A.17)
(2) For an arbitrary positive (0 < < ), on the domain | Im q| ,

ikq
e e|k|() .
(A.18)
On the domain | Im q|

|k| |k|()

|F |
|Fk |eikq
Ce
e
k

n

1 + e n
=
Ce|k| = C 1 + 2
ek = C
1 e
k>0
k
n
4
<C
,
where we use the inequality

4
1 + e
,
<
1 e
(A.19)
(A.20)
for 0 < 3. Namely, on the domain | Im q| , F (q) converges uniformly and

absolutely, therefore, F (q) is analytic. Since is arbitrary, F (q) is analytic on the domain
| Im q| . 2
362
We evaluate the right-hand side of (A.8). For arbitrary p(m+1) D(m ), we put m as
the maximum value of p (m+1)
m p(m+1) m .
(A.21)
Since t m Bm is analytic and bounded by m1 M2(m) on the domain D(m ),

m (m+1)
t b k p
, t m1 M2(m) e|k|m .
(A.22)
As |k| increases, while bk decay exponentially, the contribution from the denominator
grows like power
|k|d
1

.
( k)
C
(A.23)
The exponential decay defeats the power grow. We use the inequality
d
d
1
|k|d e|k|
,
e d
(A.24)
for an arbitrary positive . The proof is as follows. f (x) = x d ln x has a minimum at

x = d. Therefore,
ex
ed
(A.25)

.
xd
dd
For x = |k|, we obtain this inequality. We evaluate the Fourier components Sk as
d
m (m+1)

1/2 |k| m1 (m)
t Sk p

, t m
M2 exp |k|m
C
(m) d

1
d
1/2 M2
m m
exp |k|(m ) .
d
C
e
Therefore, t m Sm is analytic on the domain D(m ) and bounded as

m
t Sm m L(m) ,
1
(A.26)
(A.27)
where
(m)
L1
(m) d
d
1/2 M2
= m
4n
n+d
(A.28)
on the domain D(m 2) for an arbitrary positive 3. We put m+1 = m 3 and on

the domain D(m+1 )

(m)
(m)

p p (m+1) Sm m L1 (1 + m+1 )m ,
(A.29)
q (m)

(m)
Sm
(m+1)
L
(m)

q
q
(m+1) m 1 (1 + m+1 )m .
(A.30)
363
Therefore, if satisfies
m
L(m)
1
(1 + m+1 )m 3,
(A.31)
when (p (m+1) , q (m+1) ) D(m+1 ), (p(m) , q (m) ) D(m ).

Next on the domain D(m+1 ), we evaluate the residual part R defined as
Sm (q (m) , p (m+1) , t)
.
t
For evaluation, we use the Taylor expansion

n1

1 (k) 1 (n)

F (0)
F
,
F (1) F (0)

n!
k!
R = R1 + R2 + R3 +
(A.32)
(A.33)
k=1
where [A] means the maximum value of |A|. As the function F ( ), we take

F ( ) = f x 0 + (x x 0 ) ,
where x is the multi-dimensional vector.

f
F (1) (0) = i (x x 0 )i ,
x 0

2 f
(2)
F (0) = i j (x x 0 )i (x x 0 )j .
x x
(A.34)
(A.35)
(A.36)
The residual part is

nF
1 (n) 1
i1
in
(x x 0 ) (x x 0 ) .
=
F
n!
n! x i1 x in
(A.37)
As for R1 ,

(m)

1 i j (m)
(m+1) i
(m+1) j
.
p p
p p
|R1 |
4 ( p)3/2
(A.38)
Then we evaluate

m (m) 2
2m
t R1 1 1 n L1
.
4 m3/2
As for R2 ,
|R2 |

Am (m)
Am (m)
(m+1)
p p(m+1) .
+
q1 q1
q 1
p
(A.39)
(A.40)
Then we evaluate
(m)
(m)
m+1
M
L
t
R2 2n 1 m 1 .
(A.41)
364
As for R3 ,

|R3 |

Bm (q (m) , p, t) (m)
p p (m+1) .
p
(A.42)
Then we evaluate
(m)
(m)
2m
t R3 n m1 M2 m L1 .
In addition,

m+1 Sm
t
m L(m) m + 1 + m+1 .
1

t
Therefore, we obtain
m+1
t
R m M,
(A.43)
(A.44)
(A.45)
where
2

L(m)
M1(m) L(m)
1
1
n
(1 + m+1 )
+
2n
M =
3/2
2
4m

(m) (m)
M L
1 + m+1
.
m
+
+ m1 n(1 + m+1 )m1 2 2 1 + L(m)
1
m1
m1
(A.46)
We decompose R into the slowly varying part Rs and the fast varying part Rf as
R = Rs + Rf ,
(A.47)
where
Rs =

R0k1 p(m+1) , t exp ik 1 q (m+1)
,
1
(A.48)

Rk 0 k 1 p (m+1) , t exp ik 0 q (m+1)
+ ik 1 q (m+1)
.
0
1
(A.49)
k 0 =0
Rf =
k 0 =0
Since Rs is written as
1
Rs =
(2)n0
2
2
d n0 q (m+1)
R,
0
(A.50)
Rs is bounded as
m+1 m+1
t
Rs t
R m M.
(A.51)
Then

m+1 m+1
t
Rf t
(R Rs ) 2 m M.
(A.52)
365
We define
(m+1) (m+1)
(m+1) (m+1)
(m+1) (m+1)
Am+1 q 1
,p
, t = Am q 1
,p
, t + Rs q 1
,p
,t ,
(A.53)
and

Bm+1 q (m+1) , p(m+1) , t = Rf q (m+1) , p(m+1) , t .
(A.54)
We evaluate
|tAm+1 | M1(m+1) ,
(A.55)
m+1
(m+1) (m+1)
(m+1) (m+1)
t
Am+1 q
,p
, t Am q
,p
,t
1
m+1

(m+1)
t
Bm+1 m M
,
(m+1)
m M2
, (A.56)
(A.57)
where
M1(m+1) = M1(m) + (1 + m+1 )m m M,
(A.58)
(m+1)
M2
(A.59)
= 2M,
on the domain D(m+1 ). We complete the proof.

Appendix B. Proof of Propositions 4.2A, 4.2B
B.1. Proof of the Proposition 4.2A
We consider the system obtained by three times of canonical transformations
S1
S2
S3
H (1) H (2) H (3) H (4).
(B.1)
(4)
We evaluate in how good accuracy the truncated system Htr approximates H (4)
2
(4)
Htr = ( P )1/2 + A4 (Q1 , P , t),

2
(4)
H = ( p)1/2 + A4 (q 1 , p, t) + B4 (q, p, t),

where
4
t B4 3 M (4) .
|tA4 | M1(4),
2
(B.2)
(B.3)
(B.4)
By taking the difference between

B4
dp0
=
,
dt
q 0
(B.5)
dP 0
= 0,
dt
(B.6)
and
366
we obtain
B4
d
P 0 =
.
dt
q 0
(B.7)
By integrating the above inequality, we obtain

3

P 0 (t) P 0 (1) + dt P 0 (1) + 3 .
4
t
(B.8)
If we fix

P 0 (1) 3 ,
(B.9)
we obtain
|P 0 | 3 .
(B.10)
Next we evaluate Q1 = q 1 Q1 . By taking the difference between

1
A4 B4
dq 1 1
=
+
+
,
1/2
dt
( p)
p1 p 1
(B.11)
and
A4
1
dQ1 1
+
,
=
1/2
dt
( P )
P 1
we obtain

1
1
d
A4 A4
B4
1
Q1 =
.
+
+
1/2
1/2
dt
( p)
( P )
p1 P 1
p 1
(B.12)
(B.13)
By using the mean value theorem,

1
1
1
1
=
(0 P 0 + 1 P 1 ),
2 ( P )3/2
( p)1/2 ( P )1/2

A4 A4
A4
= P 0
+ Q1
+ P 1
,
p1 P 1
P 0
Q1
P 1 P 1
(B.14)
(B.15)
are obtained where the differentiations in the right-hand side are taken at appropriate values
between (p0 , z) and (P 0 , Z). Since we are only interested in the upper bounds of the
coefficients, we understand that the differentiations are taken at appropriate values between
the original variables (p 0 , z) and the truncated variables (P 0 , Z) without notice from now
on. Then we get
1
1 1
d
Q1 =
(0 P 0 + 1 P 1 )
dt
2 ( P )3/2

A4 B4
+ P 0
+ Q1
+ P 1
+
.
P 0
Q1
P 1 P 1 p1
In the same way, as for P 1 = p1 P 1 , we obtain

d
B4
A4
P 1 = P 0
+ Q1
+ P 1
.
dt
P 0
Q1
P 1 Q1 q 1
(B.16)
(B.17)
By using the notation z = (q 1 , p1 ), Z = (Q1 , P 1 ), Z = z Z, we obtain

2

1
d
1 1
3
|Z|
+
|Z| +
+
|P 0 | + 4 .
dt

t

t
t
Under the assumption 1 , we get

d
3
|Z|
+ |Z| + |P 0 | + 4 .
dt
t
t
(B.18)
(B.19)

3

+
Z(1) + dt|P 0 | + dt 4 .
dt
|Z| exp
t
t
1
367
(B.20)
By using (B.10), and assuming

Z(1) 3 ,
(B.21)
we obtain

|Z| t exp (t 1) 3 t + 3 .
We consider the time interval as
1
1t ,
t
where
1
< 1.
=
+1
In this interval, we obtain the evaluation as
|Z| 2 .
(B.22)
(B.23)
(B.24)
(B.25)
The equation for Q0 = q 0 Q0 is given by

1 0
d
1
Q0 =
(0 P 0 + 1 P 1 )
dt
2 ( P )3/2

A4 B4
+ P 0
+ Q1
+ P 1
+
,
P 0
Q1
P 1 P 0 p0
which yields

3
d

Q0 1 + 1 |P 0 | + 1 |Q1 | + 1 + 1 |P 1 | + .
dt

4

t
t
t
t
(B.26)
(B.27)

1
dt
3

|Q1 | + dt|P 1 | + dt 4 ,
|Q0 | Q0 (1) +
dt|P 0 | +

t
t
1
1
1
1
(B.28)
368
which yields

Q0
1
3

t Q0 (1) + |P 0 | + |Z| +
2,
where we used (B.10), (B.25) and assumed that

Q0 (1) 2 .
(B.29)
(B.30)
By obtaining (B.10), (B.25), (B.29), we complete the proof of (4.59).

(4)
Next we consider in how good accuracy the perturbations of the truncated system Htr
approximate the perturbations of the transformed system H (4) . We try to obtain the equation for Q0 = q 0 Q0 . We take the difference between
1 0
1
d
q =
(0 p 0 + 1 p1 )
dt 0
2 ( p)3/2

A4
+ p0
+ q 1
+ p 1
p0
q 1
p1 p0

B4
+ q 0
+ p0
+ q 1
+ p 1
,
q 0
p 0
q 1
p1 p0
(B.31)
and
1 0
1
d
Q0 =
(0 P 0 + 1 P 1 )
dt
2 ( P )3/2

A4
+ P 0
+ Q1
+ P 1
.
P 0
Q1
P 1 P 0
(B.32)
By using the mean value theorem, we obtain

1
1
1 0
1 0
(0 p0 + 1 p 1 ) +
(0 P 0 + 1 P 1 )
3/2
2 ( p)
2 ( P )3/2
1
1 0
(0 P 0 + 1 P 1 )
=
2 ( p)3/2
1
3 0
(0 P 0 + 1 P 1 )(0 P 0 + 1 P 1 ),
+
(B.33)
4 ( P )5/2

A4
p 0
+ q 1
+ p1
p
q 1
p1 p0
0

A4
P 0
+ Q1
+ P 1
P 0
Q1
P 1 P 0

A4
= P 0
+ Q1
+ P 1
p0
q 1
p1 p0

+ P 0
+ Q1
+ P 1
P 0
Q1
P 1

A4
P 0
(B.34)
+ Q1
+ P 1
,
P 0
Q1
P 1 P 0
and
369

B4
+ p 0
+ q 1
+ p1
q 0
q
p0
q 1
p1 p 0

0
B4
= Q0
+ P 0
+ Q1
+ P 1
q 0
p0
q 1
p1 p 0

B4
+ Q0
+ P 0
+ Q1
+ P 1
,
q 0
p0
q 1
p1 p0
(B.35)
where the differentiations in the right-hand side are taken at the appropriate values between
(Q, P ) and (q, p), and it will not be noticed from now on. By evaluating the coefficients,
we obtain

d

Q0 1 |P 0 | + 1 |P 1 | + 1 |P 0 | + 1 |P 1 | |P 0 | + 1 |P 1 |

dt

1

+ |P 0 | + |Q1 | + |P 1 |
t

1
+ |P 0 | + |Q1 | + |P 1 | |P 0 | + |Q1 | + |P 1 |
t
3

+ |Q0 | + |P 0 | + |Q1 | + |P 1 | 4
t

3
+ |Q0 | + |P 0 | + |Q1 | + |P 1 | 4 .
t
By using the inequalities as
(B.36)
|P 0 | + |Q1 | + |P 1 | 2 ,
(B.37)
|P 0 | + 1 |P 1 | ,
(B.38)
we obtain

3

d
Q0 1 + 1 |P 0 | + |P 1 | + 1 |Q1 | + |Q0 |

dt

t
t
t4

1
2
3
+ 2 |P 0 | + 2 +
|P 1 | + |Q1 | + 4 |Q0 |.
t
t
t
In the same way as Q0 , we obtain

d
3
P 0 |Q0 | + |P 0 | + |Q1 | + |P 1 |

dt
t4
3

+ |Q0 | + |P 0 | + |Q1 | + |P 1 | 4 ,
t
and

d
3
|Z|
+ |Z| + |P 0 | + 4 |Q0 |
dt
t
t
(B.39)
(B.40)
370

1
2
1
3
|P 0 | + |Q1 | + 2 2 +
|P 1 | + 4 |Q0 |.
+ 2 +
t
t
t
t
(B.41)
By using
|Q1 |,
|P 1 | |Z|,
|Q1 |,
|P 1 | |Z|,
(B.42)
we obtain

3
d
|Z| +
|Z| + |P 0 | + 4 |Q0 |
dt
t
t
3
(B.43)
|Q0 |,
t4

3
3

d
Q0 1 |P 0 | + |Z| + |Q0 | + 2 |P 0 | + 2 |Z| + |Q0 |,

dt
4
4
t
t
(B.44)
+ 2 |P 0 | + 2 |Z| +
and

d

3
3
P 0 |Q | + |P 0 | + |Z| + |Q | + |P 0 | + |Z| .
0
0
dt

t4
t4
(B.45)
The inequality (B.43) yields

3

Z(1) + dt |P 0 | + dt |Q0 |

|Z| exp
dt +
t
t4
1

3
+ dt 2 |P 0 | + dt 2 |Z| + dt 4 |Q0 |
t
1
1
1

3 Q0

t exp (t 1) Z(1) + t P 0 + 2 + 2 t P 0
t

Q
0
+ 2 t Z + 3
(B.46)
t2 .
The inequality (B.44) yields

3
dt 4
|Q0 | exp
t
1

t

Q0 (1) + P 0 + t Z

Q0

+ 2 t P 0 + 2 t Z + 3
t2

t

Q0 (1) + P 0 + t Z

Q0

+ 2 t P 0 + 2 t Z + 3
(B.47)
t2 ,
and the inequality (B.45) yields

|P 0 | P 0 (1)

Q0 Z Q0 P 0 Z
+
+
+

+ 3
t2 t2 t2 t2 + t2 .
371
(B.48)
If we consider the time interval

1
,
t
(B.49)
1
1,
+1
(B.50)
1t
where
=
the growth factor of |Z| is bounded as

1
t exp (t 1) t .
(B.51)

Then in this interval (B.49), we obtain

Q0

Q0 (1) + 1 P 0 + Z + 2 P 0 + 2 Z + 3 Q0 ,

t2
t

(B.52)

|P 0 | P 0 (1)

Z Q0 P 0 Z
3 Q0

+ ,
+ 2 + 2 + 2 + 2
(B.53)
t
t
t
t t2

1
Q0

1
+ P 0 + Z + 2 Q0 .
|Z| Z(1) + P 0 + 2

2
2

t
t
(B.54)
By using (B.53) to (B.52) (B.54), we obtain

Q0

Q0 (1) + 1 P 0 (1) + Z + 2 Q0
t

2

t
+ 2 Z + 2 P 0 ,
and
(B.55)

1
Q0

1
+ 2 Q0
Z Z(1) + P 0 (1) + 2

2
2

t
t
+ P 0 + Z .
By using (B.56) to (B.55), we get

Q0
1

t Q0 (1) + P 0 (1) +
+ P 0 + Z .
(B.56)

Q0
1

Z(1) + 2
t2

(B.57)
372
By using (B.57) to (B.56), we obtain

1

Z Q0 (1) + P 0 (1) +

+ P 0 + Z .

2

1
2 Q0

Z(1) + 2

t
(B.58)
By using (B.57), (B.58) to (B.53), we obtain

2
3 Q0

P 0 Q0 (1) + P 0 (1) + Z(1) + 2
t

3
+ 3 P 0 + 3 Z .
(B.59)
By substituting
Q0 (1) = P 0 (1) = Z(1) = 0
to (B.57)(B.59), we obtain

Q0

2 Q0 + P 0 + Z ,

t
t2

Q0
3
3

P 0 3
t 2 + P 0 + Z ,

2 Q0
Z 2 + P 0 + Z .
t
(B.60)
(B.61)
(B.62)
(B.63)
Next we evaluate Q0 /t 2 by
d
1
1 0
Q0 =
(0 P 0 + 1 P 1 )
dt
2 ( P )3/2

A4
+ P 0
+ Q1
+ P 1
,
P 0
Q1
P 1 P 0
which yields

1
Q0

t Q0 (1) + P 0 + Z .
By using (B.65) to (B.61)(B.63), we get

Q0

2

t Q0 (1) + P 0 + Z ,

P 0 3 Q0 (1) + 2 P 0 + 3 Z ,

Z 2 Q0 (1) + P 0 + Z .
These complete the proof of (4.60).
(B.64)
(B.65)
(B.66)
(B.67)
(B.68)
373
B.2. Proof of Proposition 4.2B

We evaluate how the errors Q(4) , P (4) , Q(4) , P (4) are transmitted to Q(1) ,
P (1) , Q(1) , P (1) by the canonical transformations
Sm
,
q (m)
Sm
q (m+1) = q (m) +
p(m+1)
p (m) = p (m+1) +
(B.69)
(m 1),
where
m
t Sm m L(m) .
(B.70)
(B.71)
Lemma.

P (1) P (m) Q(m) + P (m) ,
t
t

Q(1) Q(m) Q(m) + P (m) ,
t
t
(1)
(m) (m)
(m)
P P Q + P ,
t
t
(1)

Q Q(m) Q(m) + P (m) .
t
t
(B.72)
(B.73)
Proof of Lemma. We can prove (B.72), (B.73) in the almost same way. So we prove
(B.72) as the representative. We consider the difference P = p P . Taking the difference between
Sm
p (m) = p (m+1) + (m) ,
(B.74)
q
and
P (m) = P (m+1) +
Sm
Q(m)
(B.75)
yields
Sm
Sm
(m)
q
Q(m)

Sm
(m+1)
= P (m+1) + Q(m)
+
P
,
(m)
(m+1)
P
Q
Q(m)
P (m) = P (m+1) +
(B.76)
where differentiations in the right-hand side are taken at the appropriate values between
(q, p) and (Q, P ) according the mean value theorem. So we obtain
m

P (m) P (m+1) Q(m) + P (m+1) ,
m
t

m

(m)
(m+1)
Q(m) + P (m+1) .
Q Q

m
t
(B.77)
374
As for the first term in the right-hand side of the above inequalities, we obtain

Q(m) Q(m) Q(m+1) + Q(m+1)

m
Q(m) + P (m+1) + Q(m+1) ,
m
t
(B.78)
which yields
m

Q(m) Q(m+1) + P (m+1) .
m
t
By using (B.79) to (B.77), we obtain
m

P (m) P (m+1) X(m+1) ,
m
t

m

(m)
(m+1)
Q Q

X(m+1) ,
m
t
where

X(m) = Q(m) + P (m) .
(B.79)
(B.80)
(B.81)
In the same way as (B.79), we obtain

m

P (m) = P (m+1) + Q(m+1) .
tm
From (B.79), (B.82), we obtain

X(m) X(m+1) .
(B.82)
(B.83)
We evaluate

P (1) P (m)

P (1) P (2) + P (2) P (3) + + P (m1) P (m)
2

m1
X(2) + 2 X(3) + + m1 X(m)
t
t
t

(m)
X
,
t
where we used (B.80), (B.83). In the same way, we get

Q(1) Q(m) X(m) .
t
These complete the proof. 2
(B.84)
(B.85)
Proof of Lemma. From the inequalities (4.59), we obtain
So
1

1
Q(4) + P (4) 2 .
t
t
(B.86)

P (1) P (4) 3 ,
(B.87)

Q(1) Q(4) 3 ,
375
are obtained. Therefore, we obtain

P (1) P (4) + P (1) P (4) 3 ,
0
0
0
0

Z (1) Z (4) + Z (1) Z (4) 2 ,

(4)
Q(4)
Q(1)
Q(1)
Q0
0
0
0

2.

t t + t
t
(B.88)
(B.89)
(B.90)
We have proved (4.63).

Lemma.

P (1) P (m) X(m) X(m) + X (m) ,
t

Q(1) Q(m) X (m) X(m) + X(m) ,
t
where

X(m) = Q(m) + P (m) .
(B.91)
(B.92)
Proof of Lemma. By using (B.79), we obtain

Q(m) + P (m+1) X(m+1) .
(B.93)
In the same way,

(m) (m+1) (m+1)
Q + P
X

(B.94)
is obtained. We take the difference between

Sm
p (m) = p (m+1) + q (m) (m) + p(m+1)
,
(m+1)
q
p
q (m)
and
P
(m)
= P
(m+1)

+ Q(m)
(m)
+ P
(m+1)
P (m+1)
By the mean value theorem, we get

Sm
(m)
(m+1)
q (m) + p
q
p(m+1) q (m)

Sm
(m)
(m+1)
Q
+ P
(m)
(m+1)
P
Q
Q(m)

(m)
(m+1)
= Q
+ P
P (m+1)
Q(m)

Sm
(m)
(m+1)
Q
+ P
(m)
(m+1)
P
Q
Q(m)

Sm
+ Q(m) (m) + P (m+1)

,
(m+1)
q
p
q (m)
Sm
Q(m)
(B.95)
(B.96)
(B.97)
376
where the differentiations in the right-hand side are taken at the appropriate values between
(q, p) and (Q, P ). So we obtain

P (m) P (m+1)

m

Q(m) + P (m+1) Q(m) + P (m+1) m
t
m

+ Q(m) + P (m+1) m
t
m

m (m+1)
(m+1)
X
+ m Q(m) + P (m+1) ,
m X
t
t
where (B.93), (B.94) are used. In the same way, we get

Q(m) Q(m+1)

m

m
m X(m+1) X(m+1) + m Q(m) + P (m+1) .
t
t
From the above inequality,

Q(m) Q(m) Q(m+1) + Q(m)

m

m
m X(m+1) X(m+1) + m Q(m) + P (m+1)
t
t

+ Q(m+1) ,
which yields
m
m

Q(m) X(m+1) X(m+1) + Q(m+1) + P (m+1) .
m
m
t
t
Then we get
m

Q(m) + P (m+1) X (m+1) X(m+1) + X(m+1) .
tm
By using (B.102) to (B.98), (B.99), we obtain
m

P (m) P (m+1) X(m+1) X (m+1) + X(m+1) ,
m
t

m (m+1)

(m)
(m+1)
Q Q
X

X(m+1) + X(m+1) .
m
t
By using the above inequalities,

X(m) X(m+1)

m
+ m X(m+1) X(m+1) + X(m+1) ,
t
is obtained. Then we get

P (1) P (m)

P (1) P (2) + + P (m1) P (m)

X(m) X (m) + X(m) .
t
(B.98)
(B.99)
(B.100)
(B.101)
(B.102)
(B.103)
(B.104)
(B.105)
In the same way, we obtain

Q(1) Q(m) X (m) X(m) + X(m) .
t
These complete the proof. 2
From the inequalities (4.59), we obtain

X(4)
2

t .
377
(B.106)
(B.107)
From the inequalities (B.65), we obtain

(4)
X Q0 (1)t + 1 P 0 t + Z t.

From the inequalities (4.60), we obtain
(4)

(4)

1
X(4) 2 Q0 (1) + P 0 + Z (4) .
t
By using the above three inequalities, we obtain

P (1) P (4) 2 Q(4) (1) + P (4) + 2 Z (4) ,
0
0

Q(1) Q(4) 2 Q(4)(1) + P (4) + 2 Z (4),
0
0
for 1 t 1/. By combining (4.60), (B.110), (B.111), we obtain

(4)

Q(1)
(4)

0

2 Q0 (1) + P 0 + Z (4) ,

t

P (1) 2 Q(4)(1) + P (4) + 2 Z (4),
0
0
0

Z (1) 2 Q(4) (1) + P (4) + Z (4) .
0
0
This completes the proof of Proposition 4.2B.
(B.108)
(B.109)
(B.110)
(B.111)
(B.112)
Appendix C. Proof of Proposition 5.3

The proof is essentially the same as that of the preceding Proposition 4.1. So in this
proof we estimate the generating function of the canonical transformation, and the residual
parts which cannot be eliminated by the canonical transformation only roughly. Later we
omit m. We consider the canonical transformation induced by the generating function given
by

(l)
(l+1)
Tl q (l)
(C.1)
,
p
,
t
=
Tk1 p(l+1) , t eik1 q 1 ,
1
k 1 =0
where Tl does not depend on q (l)

0 .
378
The transformed Hamiltonian is

Tl
H (l+1) = H (l) +
t
(l)
1/2
Tl (q 1 , p(l+1) , t)
2
1
1
= 2 p(l+1)
+ 2
+ R1
1
(l)

( p (l+1))1/2
q 1

(l)

+ Dl p (l+1), t + R2 + El q 1 , p(l+1) , t + R3
(l)

Tl (q 1 , p(l+1) , t)
,
+ Bl q (l+1), p (l+1), t + R4 +
(C.2)
t
where R1 , . . . , R4 are defined as
1/2
1/2
(l)
1
2
2
1
(l+1)
,
2 p(l+1)
2
p1 p 1
R1 = 2 p(l)
(l+1)
1/2

( p
)
(C.3)
(l)
(l+1)
R2 = Dl p , t Dl p
(C.4)
,t ,
(l) (l)
(l) (l+1)
R3 = El q 1 , p , t El q 1 , p
(C.5)
,t ,
(l) (l)
(l+1) (l+1)
R4 = Bl q , p , t Bl q
(C.6)
,p
,t .
We determine the generating function Tl so that the leading term depending on the slow
angle variables q (l)
1 can be eliminated
(l+1) , t)
(l)

Tl (q (l)
1
1
1 ,p
= El q 1 , p(l+1) , t .
1
2
(l+1)
1/2
(l)
( p
)
q
(C.7)
When El is decomposed as

(l)

ek1 p (l+1), t eik1 q 1 ,
El =
(C.8)
k 1 =0
Tl is written as

1/2
Tl =
i 2 p(l+1)
k 1 =0
(l)
1
ek1 eik1 q 1 .
( 1 k 1 )
(C.9)
Since Tl is roughly estimated as

l
(C.10)
,
tl
the differences between the original variables and the transformed variables are estimated
as
Tl
(l)
(l+1)
p 0 p0
= 0,
(l+1)
p (l)
1 p1
l
,
tl
l
q (l+1) q (l) l .
t
(C.11)
(C.12)
(C.13)
By using these estimates, we evaluate the residual parts Ri and Tl /t

j
1 1 i1 1 (l)
2l
(l+1) i (l)
(l+1) j
|R1 | 2
p
p
p
,
1
1
1
4 ( p)3/2 1
t 2l
where we use 1 ,

Dl (p, t) (l)
l+1
(l+1)
|R2 |
p1 p1
l+2 ,
p 1
t

2l1

El (l)
|R3 |
2l ,
p1 p(l+1)
1
p1
t

Bl (l)
2(m1) l
Bl (l)
(l+1)
(l+1)
q q
+
p1 p 1
,
|R4 |
q
p 1
tm tl
Tl
l
l+1 .
t
t
(C.14)
(C.15)
(C.16)
(C.17)
(C.18)
(l)
(l+1)
Then q 0 independent (so q 0

R1 + R2 + R3 +
379
independent) part is estimated as
l
Tl
l+1 ,
t
t
(l)
(l+1)
while q 0 dependent (so q 0
(C.19)
dependent) part is estimated as
2(m1) l
.
tm tl
The residual part
R4
(C.20)
Tl
(C.21)
t
is decomposed into non-oscillating part R0 , slowly oscillating part Rs and fast oscillating
part Rf
R = R1 + R2 + R3 + R4 +
R = R0 + Rs + Rf ,
(C.22)
where

l
R0 = R0 p(l+1) , t l+1 ,
t

(l+1)

l
(l+1)
l+1 ,
Rk 1 p
, t exp ik 1 q 1
Rs =
t
(C.23)
(C.24)
k 1 =0
Rf =

k 0 =0

2(m1) l
(l+1)
Rk 0 k 1 p (l+1), t exp ik 0 q (l+1)
+
ik
q
.
1
0
1
tm tl
(C.25)
We define

Dl+1 p(l+1) , t = Dl p(l+1) , t + R0 ,
(l+1)

El+1 q 1 , p(l+1) , t = Rs ,

Bl+1 q (l+1) , p(l+1) , t = Bl q (l+1) , p(l+1) , t + Rf .
(C.26)
(C.27)
(C.28)
380
By transforming canonically, we succeeded in lowering the residual part El l1 /t l to

El+1 l /t l+1 .
Appendix D. Proof of Propositions 5.5, 5.6
D.1. Proof of Proposition 5.5
Lemma.

q 0

1
1

t q 0 (1) + q 1 (1) + 2 p 0 (1) + p 1 (1) ,

p 0 4 q 0 (1) + 4 q 1 (1) + p 0 (1) + 3 p 1 (1),

q 1

1

3

t q 0 (1) + q 1 (1) + p0 (1) + p 1 (1) ,

p 1 4 q 0 (1) + 2 q 1 (1) + p 0 (1) + p1 (1).
Proof. From the variational equations of H (3,3), we obtain

q 0

q 0 (1) + 1 p 0 + 1 p1 + 4 q 0 + 2 q 1 ,

t

2

t
t

4 q 0
4 p 0
4 q 1
4 p 1

+
+
+
,
|p0 | p0 (1) +
t
t
t
t

q 1

1
2 q 1
4 q 0

t q 1 (1) + p 0 + p 1 + t + t ,

2 p 0
2 q 1
2 p 1
4 q 0

+
+
+
.
|p1 | p1 (1) +
t
t
t
t
(D.1)
(D.2)
(D.3)
(D.4)
(D.5)
(D.6)
(D.7)
(D.8)
From these inequalities, we can deduce the inequalities of the Lemma by tedious manipulations. 2
We assume that

q 0 (1) p0 (1) q 1 (1) p1 (1) 1,
(D.9)
since in the linear perturbation, the scale of the perturbation variables is arbitrary. Then we
obtain the proposition below.
Lemma.

q 0
1

t 2 ,
p 1,
0
q 1 1

t ,
p 1 1.
(D.10)
(D.11)
(D.12)
(D.13)
381
By using the above Lemma to the evolution equations of perturbation variables, we

obtain the evaluation of Proposition 5.5.
D.2. Proof of Proposition 5.6
Since Sm 2m /t m , we obtain
2

(1,1)
q
q (m,1) x (m,1),
t

(1,1)
2
p
p(m,1) x (m,1) ,
t
where
|x| = |q| + |p|.
On the other hand, since Tl l /t l and Tl does not depend on q 0 , we obtain
(m,1)

q
q (m,l) y (m,l) ,
t
(m,1)
(m,l)
p
p0
= 0,
0
(m,1)

(m,l)
p
p1 y (m,l) ,
1
t
where
|y| = |q 1 | + |p|.
(D.14)
(D.15)
(D.16)
(D.17)
(D.18)
(D.19)
(D.20)
By using the above inequalities, we obtain the Lemma below.

Lemma.
(1,1)

2 (m,l)
q
q (m,l) y (m,l) + q 0 ,
t
t
2
(1,1)

(m,l)
p
p0 = x (m,l) ,
0
t
2
(1,1)

p
y (m,l) + q (m,l) .
p(m,l)
1
1
0
t
t
Since
1 (3,3) 1
y
,
t

1
1 (3,3)
q 0
2,
t

1
1 (3,3)
2,
x
t

we obtain Proposition 5.6.
(D.21)
(D.22)
(D.23)
(D.24)
(D.25)
(D.26)
382
Appendix E. Growth index of perturbations

In this appendix, we calculate the growth rates of perturbations in the first model
(12 2 , 21 2 ) and the second model (12 22 , 1 2 ) which are presented in the
beginning of Section 4, assuming that / is of order unity.
The fourth order Hamiltonians H (4) of the first model and the second model are written
by
2
H = ( P )1/2 + A + R,

1
A = ( P )1/2 (M + N cos kQ1 ),
t

|R| 2 ,
t
(E.1)
(E.2)
(E.3)
where
= 1,
M = 0,
k = 1,
(E.4)
1/2
N = (P0 2P1 )P1 ,
(E.5)
k = 2,
(E.6)
and
= 2,
M = 2N = (P0 2P1 )P1 ,

respectively. The coefficients of the perturbation equations
1 12
2A
1
2A
+

2 ( P )3/2 P12
d Q1
Q1
P1 Q1
=
P1
dt P1
2A
2A
2
Q1 P1
Q1
1 0 1
1
2A
+
2 ( P )3/2 P1 P0
P0 + R,
+
2A
Q1 P0

2
1
d
1 0
2A
2A
Q0 =
P
+
+
Q1
0
dt
2 ( P )3/2 P02
P0 Q1

1
1 0 1
2A
P1 + R,
+
+
2 ( P )3/2 P0 P1
P0 = P0 (1),

|R| 2 |P0 | + |Q1 | + |P1 | ,
t
(E.7)
(E.8)
(E.9)
(E.10)
(E.11)
383
are given by
1
t
1
t
1
1
2A
N
1
1
1
= k
sin kQ1 + k
N sin kQ1 ,
1/2
P1 Q1
( P ) P1
2 ( P )3/2
(E.12)
2A
P12

12
M
3
1
N
(M
+
N
cos
kQ
)
+
cos
kQ
1
1
4 ( P )5/2
( P )3/2 P1 P1
2

1
M 2N
+
cos kQ1 ,
+
( P )1/2 P12
P12

1 1 2 A
t
P1 P0

3 0 1
1
N
M
0
=
(M
+
N
cos
kQ
)
+
cos
kQ
1
1
4 ( P )5/2
2 ( P )3/2 P1 P1

1
N
1
M
+
cos kQ1
2 ( P )3/2 P0 P0
2

M
2N
1
,
+
cos
kQ
+
1
( P )1/2 P1 P0 P1 P0

1
1 1 2 A
= k 2
N cos kQ1 ,
2
t
( P )1/2
Q1

1 1 2 A
0
N
1
1
= k
N sin kQ1 k
sin kQ1 ,
3/2
1/2
t
Q1 P0 2 ( P )
( P ) P0
=
(E.13)
(E.14)
(E.15)
(E.16)
and
1
t
=
1
2A
P02

02
3
0
N
M
(M
+
N
cos
kQ
)
+
cos
kQ
1
1
4 ( P )5/2
( P )3/2 P0 P0
2

M 2N
1
+
cos kQ1 .
+
( P )1/2 P02
P02
(E.17)
Since the correction to the growth rate of perturbations by R is of order as long as we

consider the finite small time range, we consider the fixed points with dropping R.
For simplicity, we consider the case 1 = 0.
In the first model, around the fixed point
Q1 =
+ k,
2
2P1 = P0 = c,
(E.18)
384
where k is integer, the perturbations are given by

Q1 = t ()
P1 = t
k
1
Q1 (1),

1
1
P1 (1) + P0 (1) P0 (1),
2
2
()k+1 1
1/2

1 0
1
k
Q0 = Q0 (1)
P0 (1)(t 1) t () 1 1 Q1 (1),
2 c3/2
2
P0 = P0 (1),
where 1 is given by
1
1 = 2
.
1/2
0
(E.19)
(E.20)
(E.21)
(E.22)
(E.23)
The lines Q1 = /2 + k and P1 = c/2 are the heteroclinic orbits. Since the definition of
the Bardeen parameter contains the prefactor 1/t, when 1 is larger than 1, grows in
proportion to t 1 1 . Around the elliptic fixed point
Q1 = k,
6P1 = P0 = c,
(E.24)
the evolution of perturbations is oscillatory.

In the second model, the fixed points are given by
k
2P1 = P0 = c,
Q1 = ,
(E.25)
2
where for odd k hyperbolic, for even k elliptic. Around this hyperbolic fixed point the
perturbations are given by

1
1
Q1
= c1 exp 2 1
P1
t
22 c

1
0
1
+
P
+ c2 exp 2 1
(E.26)
(1)
0
2
1 ,
t
2 c
2
where
21
1
c1 = Q1 (1)
P1 (1) +
2
2 c
1
21
P1 (1)
c2 = Q1 (1) +
2
2 c
21
P0 (1),
4 c
21
P0 (1),
4 c
(E.27)
(E.28)
and
Q0 = Q0 (1)
P0 = P0 (1),

1/2
1
1 1 0
5
1
+ R,
P
(1)(t
1)
P
(1)
1
0
0
2 c3/2
32 (0 c)1/2
t
(E.29)
(E.30)
where
|R| P1 .
(E.31)
The growth rate 2 is given by
2 c1/2
2 =
.
2 1/2
385
(E.32)
The contribution of the dependent part to the Bardeen parameter is proportional to

f (t):

1
1
f (t) = exp 2 1
(E.33)
t
t
which increases for t 2 and
1
f (2 )
exp(2 1).
=
f (1)
2
(E.34)
In case 1 = 0, as the time proceeds the term originating from the unperturbed part
becomes dominant while the perturbation parts decay as 1/t , therefore the fixed points
disappear.
References
[1]
[2]
[3]
[4]
[5]
[6]
[7]
[8]
[9]
[10]
[11]
[12]
[13]
[14]
[15]
H. Kodama, M. Sasaki, Prog. Theor. Phys. Suppl. 78 (1984) 1166.

J.M. Bardeen, Phys. Rev. D 22 (1980) 1882.
A. Taruya, Y. Nambu, Phys. Lett. B 428 (1998) 3743.
H. Kodama, T. Hamazaki, Phys. Rev. D 57 (1998) 71777185.
M. Sasaki, T. Tanaka, Prog. Theor. Phys. 99 (1998) 763782.
H. Kodama, T. Hamazaki, Prog. Theor. Phys. 96 (1996) 949970.
T. Hamazaki, H. Kodama, Prog. Theor. Phys. 96 (1996) 11231146.
T. Hamazaki, Phys. Rev. D 66 (2002) 023529.
J. Traschen, R.H. Brandenberger, Phys. Rev. D 42 (1990) 2491;
Y. Shtanov, J. Traschen, R.H. Brandenberger, Phys. Rev. D 51 (1995) 5438;
K.A. Kofman, A.D. Linde, A.A. Starobinsky, Phys. Rev. Lett. 73 (1994) 3195;
K.A. Kofman, A.D. Linde, A.A. Starobinsky, Phys. Rev. D 56 (1997) 3258.
S.Yu. Khlebnikov, I.I. Tkachev, Phys. Rev. Lett. 77 (1996) 219.
C. Gordon, D. Wands, B.A. Bassett, R. Maartens, Phys. Rev. D 63 (2001) 123506.
A.D. Linde, Phys. Lett. B 158 (1985) 375;
D. Polarski, A.A. Starobinsky, Nucl. Phys. B 385 (1992) 623;
A.A. Starobinski, J. Yokoyama, gr-qc/9502002;
A.D. Linde, K.A. Kofman, Nucl. Phys. B 282 (1987) 555;
A.R. Liddle, A. Mazumdar, Phys. Rev. D 61 (2000) 123507.
J.P. Zibin, R. Brandenberger, D. Scott, Phys. Rev. D 63 (2001) 043511.
G. Felder, L. Kofman, Phys. Rev. D 63 (2001) 103503.
V.I. Arnold, Mathematical Methods of Classical Mechanics, Springer, New York, 1978;
V.I. Arnold, A. Avez, Problmes Ergodiques de la Mcanique Classique, GauthierVillars, Paris, 1967.

Neutrino masses and mixing:

singular mass matrices and quarklepton symmetry
I. Dorsner a , A.Yu. Smirnov a,b
a International Centre for Theoretical Physics, Strada Costiera 11, 31014 Trieste, Italy
b Institute for Nuclear Research, Russian Academy of Sciences, Moscow, Russia
Received 19 May 2004; received in revised form 6 July 2004; accepted 21 July 2004
Abstract
We suggest an approach to explain the observed pattern of the neutrino masses and mixing which
employs the weakly violated quarklepton equality and does not require introduction of an ad hoc
symmetry of the neutrino sector. The mass matrices are nearly equal for all quarks and leptons.
They
have very small determinant and hierarchical form with expansion parameter sin c m /m .
The latter can be realized, e.g., in the model with U (1) family symmetry. The equality is violated at
the 2 level. Large lepton mixing appears as a result of summation of the neutrino and charged
lepton rotations which diagonalize corresponding mass matrices in contrast with the quark sector
where the up quark and down quark rotations cancel each other. We show that the flip of the sign of
rotation in the neutrino sector is a result of the seesaw mechanism which also enhances the neutrino
mixing. In this approach one expects, in general, deviation of the 2-3 mixing from maximal, s13
(13)2 , hierarchical neutrino mass spectrum, and mee < 102 eV. The scenario is consistent with
the thermal leptogenesis and (in SUSY context) bounds on lepton number violating processes, like
e .
PACS: 12.10.Kt; 12.15.Ff; 14.60.P
E-mail addresses: idorsner@ictp.trieste.it (I. Dorsner), smirnov@ictp.trieste.it (A.Yu. Smirnov).

doi:10.1016/j.nuclphysb.2004.07.028
I. Dorsner, A.Yu. Smirnov / Nuclear Physics B 698 (2004) 386406
387
1. Introduction
One of the main results in the neutrino physics is a surprising pattern of the lepton
mixing which differs substantially from the quark mixing pattern. The 2-3 leptonic mixing
is maximal or nearly maximal, the 1-2 mixing is large but not maximal and the 1-3 mixing
is small or very small (see [1,2] for recent reviews). No apparent regularities or relations
between mixing parameters as well between mass ratios of different fermions have been
found, except for probably accidental relation 12 + C = 23 45 . Furthermore, the data
on masses and mixings show some degree of chaoticity.
In this connection, there are two essential issues on the way to the underlying physics:
Quarklepton symmetry: is it still realized at some level?, and
New symmetry of Nature behind neutrino masses and mixings: does it exist?
As is well known, the exact quarklepton symmetry is violated by difference of masses
of quarks and charged leptons of the first and second generations. It seems that neutrino
mixing further deepens this difference.
On the other hand, several features of the neutrino data indicate certain symmetry which
is not realized in the quark and charged lepton sectors (we will call it the neutrino symmetry):
Maximal (or near maximal) 2-3 mixing;
Small 1-3 mixing: the fact that
sin 13 sin 12 sin 23
(1)
indicates some special structure of the mass matrix;

Possible quasi-degenerate neutrino mass spectrum. This is hinted by (i) a general consideration in physics according to which large mixing is associated with degeneracy,
(ii) the neutrinoless beta decay result [37], (iii) the cosmological analysis which uses
particular set of observations [8] (see, however, [911]).
These features can be related. The same symmetry can lead to the maximal 2-3 mixing
and zero 1-3 mixing. So, breaking of the symmetry will generate simultaneously the nonzero 1-3 mixing and deviation of the 2-3 mixing from maximal value. (See, however, [12].)
Maximal mixing can be related to the quasi-degenerate mass spectrum, etc.
There is a number of studies which explore various neutrino symmetries like Z2 ,
A4 [13] or SO(3) [1416] (see [17] for review). Apparently these symmetries being exact
or approximate cannot be extended to the charged lepton sector where the hierarchy of
masses, and in particular, inequality m m , exists. Even more difficult is to include in
the same scheme quarks which show small mixings. Realization of neutrino symmetries
usually requires introduction of (i) new leptons and quarks, (ii) complicated Higgs sector
to break the symmetry, (iii) additional symmetries to forbid unwanted couplings associated
to new fermions and scalars, etc. Thus, in the neutrino symmetry scenario the observed
pattern of mixing has profound implications and requires substantial extensions of known
structures.
388
It is not excluded, however, that the neutrino symmetry is just misleading interpretation. In fact, till now the only solid indication of the new symmetry is the maximal 2-3
mixing. Notice, however, that sin2 223 = 1 is obtained as the best fit point in the 2 analysis of the atmospheric neutrino data [18]. At 90% C.L. sin2 223 > 0.9 [18]. The K2K
experiment gives even weaker bound on 2-3 mixing [19]. Furthermore, sin2 223 is a bad
quantity to describe the deviation of mixing from maximal. From theoretical point of view
the relevant parameter would be
1
sin2 23 .
2
Then the present experimental bound on the deviation is
D23
|D23 | < 0.15 (90% C.L.).
(2)
(3)
That is, |D23 | sin 23 is still possible and at the moment we cannot say that the 2-3
mixing is really near maximal one. Moreover, the latest analysis, of the atmospheric neutrino data (without renormalization of the original fluxes) shows some excess of the e-like
events at low energies (the sub-GeV events) and the absence of excess in the multi-GeV
sample. This gives a hint of non-zero D23 [18]. The deviation can show up in the generic
3 analysis of the data with the solar oscillation parameters taken into account.
In this connection we will explore an opposite no-neutrino symmetry approach which
does not rely on a special symmetry for the neutrino sector. In contrast, we will employ the
quarklepton symmetry as much as possible.
Some elements of our approach have already been considered before.
We use the mass matrix structure which leads to mixing angles of the order

mi
,
tan ij
(4)
mj
2
where mi are the eigenvalues [2022].

The enhancement of lepton mixing is a result of summation of rotations which diagonalize the neutrino and charged lepton mass matrices [23]. In contrast, the rotations cancel
each other in the quark sector thus leading to small quark mixing. In this case the atmospheric mixing angle equals

m2
m
+
.
23
(5)
m3
m
The ratio of neutrino masses is bounded from below by mass squared differences measured
in the solar (m212 ) and the atmospheric (m223) neutrino experiments:

m212
m2

= 0.18+0.22
(6)
0.08 .
m3
m223
The corresponding mass ratio for the charged leptons is smaller: m /m 0.06. Even
taking equality in (6) (which would correspond to the hierarchical mass spectrum) we find
23 38 which is well within the allowed region.
We employ the seesaw mechanism [24] and partial seesaw enhancement of the neutrino
mixing [25].
389
We also posit a symmetric form for the mass matrix structure. It is in the case of symmetric matrices that the strong mass hierarchy and large mixing can be reconciled provided
that the determinant of matrix is very small.
Finally, in the democratic approach the idea that to leading approximation all the mass
matrices in the lepton sector are proportional to each other has been pursued in [26]. It has
been further extended to the quark sector as well in [27].
The paper is organized as follows. In Section 2 we formulate our no-neutrino symmetry approach. In Section 3 we describe main features of the mass matrices and find
masses and mixing angles. In Section 4 we obtain generic predictions of the approach.
In Section 5 we consider the theoretical implications. Conclusions follow in Section 6.
Numerical results are presented in Appendix A.
2. No-neutrino symmetry approach

In what follows we assume the following.
(1) The weakly broken quarklepton symmetry is realized in terms of the mass matrices
and not in terms of observables (masses and mixing angles). The Yukawa couplings for all
quarks and leptons are nearly equal, so that the matrices of the couplings can be written as
YK Y0 + YK ,
K = u, d, l, D.
(7)
Here index D refers to the Dirac type matrix of neutrinos. The dominant structure is given
by Y0 which is common for all fermions, whereas the matrices of small corrections, YK ,
are different for different fermions. The smallness of YK can be specified in two different
ways which have different theoretical implications:
( YK )ij (Y0 )ij ,
(8)
that is, the relative corrections are small to all matrix elements, or
( YK )ij 1,
(9)
if the largest element, (Y0 )33 , is normalized to 1. In what follows for definiteness we will
elaborate on the first possibility.
(2) We assume that the matrix Y0 is singular: whole matrix Y0 as well as the sub-matrices
2-3 and 1-3 have zero (very small) determinants. As a consequence, Y0 is unstable in
a sense that small perturbations, YK , lead to significant difference in the eigenvalues
(masses) and eigenstates (mixings). This allows us to explain (see Section 3) substantial
deviation from the quarklepton symmetry at the level of observables.
In what follows we consider the following symmetric singular structure:
4
3 2
Y0 = 3 2 ,
(10)
2 1
where the expansion parameter
sin c 0.20.3.
(11)
390
We will comment on other possibilities in Section 5.

(3) The smallness of neutrino mass is explained by the seesaw mechanism [24]:
D M R1 m
TD ,
m
= m
(12)
where m
D YD v1 is the Dirac mass matrix, and v1 is the electroweak vacuum expectation
value (VEV) which generates masses of the upper fermions. The seesaw type II contribution, if exists, is small and can contribute to the correction matrix YK .
For simplicity we assume that mass matrix of the right-handed (RH) neutrinos, M R ,
has the same structure as given in Eqs. (7) and (10). This could correspond to a situation
when all fermionic components are in the same multiplet and the flavor information is in
fermions, whereas Higgs multiplets are flavorless. In general, this is not necessary, since
the RH neutrino mass matrix has different gauge properties and is generated by different
Higgs multiplet VEV. Also it may have different expansion parameter.
The seesaw mechanism plays the triple role here. (i) It explains smallness of the neutrino
mass. (ii) It flips the sign of rotation which diagonalizes the light neutrino mass matrix, so
that in the lepton sector the up and down rotations sum up (in contrast to the quark sector)
thus leading to large lepton mixing. (iii) It enhances moderately (by factor of 2) the
mixing angles which come from the neutrino mass matrix. The last two factsflipping of
the relative sign of rotations and the moderate seesaw enhancement of the neutrino mixing
anglelead to large lepton mixings.
The situation is different in the quark sector. The same dominant form for the mass
matrices of the up and down quarks leads due to cancellation of rotations to zero mixing
equal to the identity matrix. The CKM matrix originates from the mismatch between correction matrices Yu and Yd which appear small in our approach. This in turn guarantees
the smallness of the CKM angles.
We parametrize the complete matrix of Yukawa coupling (7) as
K )4 (1 + K )3 (1 + K )2
(1 + 11
12
13
K
3
K
2
(1 + K ) yK ,
YK = (1 + ) (1 + )
(13)
12
K )2
(1 + 13
22
K )
(1 + 23
23
1
where the range for the corrections, ij , is restricted by :

, K = u, d, l, D, M,
ij
(14)
for all i, j in the first case (8). The overall multipliers, yK 1, describe the amount of
non-unification of the third generation of quarks and leptons. They can also be introduced
K ).
as the corrections to 33 elements: 1 (1 + 33
The mass matrices (without renormalization group effects) equal:
m
K = YK v1 ,
m
K = YK v2 ,
K = u, D,
M R = YK M0 ,
K = M.
K = d, l,
(15)
Here v1 and v2 are the VEVs of the two Higgs doublets and M0 is the overall scale of RH
neutrino masses.
In what follows we will consider for simplicity ij to be real.
391
3. Singular mass matrices, masses and mixings

3.1. Expansion parameter
The value of expansion parameter is determined essentially by the condition (14) which
encodes degree of violation of the quarklepton symmetry in our approach and by the ratio
of muon to tau lepton masses which shows the weakest mass hierarchy. According to (13)
we obtain
l

m
l
2 3 ,
2 22
223
(16)
m
where, in general, by we will denote combinations of ij of the order ij .
There are two solutions of Eq. (16) depending on the sign of the mass ratio. If m /m
l and l = , so that the ratio
< 0, the smallest value of would correspond to 22
23
3
equals 3 , and consequently,

m 1/3

(17)
.
3m
Using value of the mass ratio at the GUT scale, m /m = 0.045, we obtain 0.26. In
this case the corrections enhance the mixing:
l
=
tan 223
l
2(1 + 23
)
l )2
1 (1 + 22
2(1 + )
.
1 2
(18)
l
l
For m /m > 0 the smallest corresponds to 22
and 23
= . The required value
of is approximately the same but the mixing is smaller.
l
l
223
= 12, that is, large corrections.
Notice that = sin c = 0.22 would require 22
Stronger mass hierarchy of quarks can be obtained taking values of 22 and 23 closer
to zero. In Fig. 1 we show the lines of constant mass ratios m2 /m3 in the 22 23 plane
for the quarks and charged leptons. The figure indicates certain hierarchy of the 22 and
23 corrections: u d l . However, this hierarchy cannot be established for all matrix
elements due to the need to reproduce observed mixing angles. In particular, value of the
2-3 CKM mixing still prevents the deviations in the 2-3 sector of the up and down quarks
from being extremely small simultaneously.
Explanation of other observables, especially in the 1-2 sectors, requires that some
u , d (see the Table 4 in Appendix A).
3.2. Masses and mixing from YK

The matrix Y0 can be diagonalized by U0 = U23 U13 , where the corresponding rotation
angles equal tan 23 = + O(3 ) and tan 13 = 2 + O(4 ). After these rotations the 1-2
matrix becomes zero and therefore masses and mixing of the first and second generations
are determined completely by the corrections to Y0 . In fact, only 33 elements of matrices
are nonzero and we will call this basis the 33 basis.
Formally one could work immediately in the 33 basis. In this basis, however, there is
no guideline (apart from the experimental data) how corrections should be introduced. One
392
Fig. 1. The lines of constant ratio m /m = 0.045 (thick solid line), ms /mb = 0.011 (thin solid line), and
mc /mt = 0.0022 (dashed line) in 22 23 plane at the GUT scale.
can consider the matrix (13) as an ansatz for introduction of the corrections. It by itself
leads to certain qualitative pattern of masses and mixing though quantitative predictions
depend substantially on particular values of ||s within interval (0). Furthermore the
ansatz (13) has certain theoretical implications which we will outline in Section 5. For a
different ansatz where the dominant structure of the Yukawa matrices has a democratic
form and related phenomenological considerations see [26,27].
In what follows we find the parametric expressions for the observables in terms of and
ijK . We discuss then restrictions on ijK and relations between them. The detailed study of
ijK and their possible origins will be given elsewhere [28].
The complete mass matrix YK can be diagonalized with high accuracy by three successive rotations: U = U23 U13 U12 . The 2-3 rotation is determined by
sin 23 (1 + 23 ),
(19)
and the 1-3 rotation by

sin 13 2 (1 + 13 ).
(20)
Here we omit the superscript for ij and ij , since these results apply to all fermions.
As a result of these two rotations we find the mass of the heaviest eigenstate

m3 = 1 + 2 + O 2 ,
(21)
393
Table 1
The values of combinations of in (26) that yield correct values of the mass ratios at the GUT scale. We take
= 0.26
m2 /m3
m1 /m3
u
d
l
0.032
0.0010
0.16
0.14
0.66
0.047
and the matrix for the first and second generations:

2

(12 13 23 )
(11 213 )
m
12 = 2
,
(12 13 23 )
22 223
(22)
where each matrix element is given in the lowest order in . Diagonalization of (22) gives
tan 212 = 2
(12 13 23 )

= r(),
22 223 + O(2 )
(23)
and masses of the lightest fermions

m2 = 2 (22 223 ) = 2 3 ,

(12 13 23 )2
m1 = 4 11 213
= 4 .
22 223
(24a)
(24b)
Here
r()
1
,
2
(25)
where 1 , 2 are functions of the order and parametrically r() = O(1). However, strong
cancellation can occur in i . Also in some cases different terms in i can sum up producing
an enhancement. As a result, the ratio can be in rather wide range r() 101 10.
Notice that the lightest mass is of the order 4 O() 5 103 which gives correct
order of magnitude for the down quarks and charged leptons.
The scenario predicts the following hierarchy of masses:
m2
= 2 ,
m3
m1
= 2 r(),
m2
m1
= 4 .
m3
(26)
K
K
K
The experimental values of mass ratios, mK
2 /m3 and m1 /m3 , can be obtained provided that the combinations of in (26) take on the values given in the Table 1. So,
cancellation or enhancement in the combinations of is needed which testifies that certain relations or/and hierarchy between ijK exist. Random selection of parameters |ijK | in
the intervals (0) will not produce correct values of masses in most of the cases. The
observables are very sensitive to choice of . It is this high sensitivity to that produces
substantially different masses of up and down quarks and leptons.
Notice that according to (22) and (24b) both m1 and 1-2 mixing will be enhanced if
22 223 .
The physical mixing matrix is a mismatch of the left rotations which diagonalize
the mass matrices of the up and the down components of the weak doublets: U =
394
U (up) U (down). Since the mass matrices of the up and down fermions are very similar, especially in 2-3 sector, they are diagonalized by rather similar rotations. In particular,
the angles of up and down rotation have the same sign thus cancelling each other in the
physical mixing matrix, so that U I . This explains the smallness of the quark mixing
angles. In contrast, due to the seesaw the neutrino rotation may flip the sign, so that the
rotations in lepton sector will sum up leading to large mixing angles.
3.3. Quark mixing
The CKM matrix is given by
u u u d d d
VCKM = U12
U13 U23 U23 U13 U12 .
(27)
Using Eqs. (19), (20), and (23) we obtain the elements of the CKM matrix in the leading
order in and :

d
u
= ,
23
Vcb
(28a)
= 23

d
u
u
u
u
( 23 )(12 13 23 )
d
u
Vub
(28b)
= 2 ,
13
23
= 2 13
u 2 u
22
23
d
d d
u u u
12 13
12
23
13
23
Vus =
(28c)
= r().
u
u
d 2 d

2
22
22
23
23
These elements have correct order of magnitude without any need for some special correlation between ijK . Indeed, for = 0.26, Vcb requires = 0.12 0.46, Vub : = 0.042
0.16, and Vus : r() 0.86.
The hierarchy of the quark mixings is naturally reproduced:
Vus ,
Vcb
,
Vus
Vub
.
Vcb
(29)
In Eqs. (A.1) and (A.2) of Appendix A we present two examples of corrections which
reproduce all parameters of the quark sector. Notice that indeed, the inequalities iju , ijd <
are satisfied for all i, j . Both up and down matrices contain some elements of the order .
Some corrections are much smaller than . Furthermore, two examples have different dominant structures (sets of matrix elements of the order ). The detailed study of properties
of ij will be given elsewhere [28].
3.4. Lepton mixing: flipping the sign of rotation
In our approach an enhancement of the lepton mixing is a consequence of the seesaw
mechanism. The seesaw produces two effects:
1. It flips the sign of rotation which diagonalizes the mass matrix of light neutrinos m
with respect to the sign of the rotations which diagonalize the Dirac neutrino matrix
m
D and charged lepton mass matrix. As a result, the rotations of the neutrinos and
charged leptons sum up in the lepton mixing matrix;
395
2. It enhances moderately the mixing produced by the neutrino mass matrix.

Let us consider these effects for the 2-3 mixing explicitly. Diagonalizing the 2-3 submatrix of m
we find

D

M D 2 D D 2

23 23
22
23
23
D
.
tan 223 = 2 1 + 23 +
(30)

M 2 M 2 D M + D 2 + 2 O()
22
23
23 23
23
The first term in square brackets corresponds to diagonalization of the Dirac mass matrix;
the second one is the effect of seesaw. An explanation of the magnitude of the 2-3 mixing
requires the second term to be 3. So that in combination with the first term it gives
4.
tan 223
Notice that the seesaw contribution is proportional to the difference of the off-diagonal
(2-3) corrections and, approximately, the ratio of determinants of the Dirac and Majorana
neutrino mass matrices. Since the determinants equal the corresponding mass hierarchies,
the enhancement of mixing requires much stronger hierarchy of the RN neutrino masses
than hierarchy of the eigenvalues of the Dirac matrix.
This can be seen explicitly by considering the mass matrix of light neutrinos:

A22 2 A23
1
m
M
(31)
,
M ( M )2
A23 A33
22 223
23
D , M ). We find explicitly that
where Aij Aij (kl
kl

M
M
Aij = 22
223
+ O ij2 .
(32)
That is, the coefficients Aij are all equal to each other in the lowest (first) order in ij .
Therefore, to enhance the mixing and to flip the sign of rotation the terms of the order 2
in (32) should be important. Consequently,

M
M
= 223
+ O ij2
22
(33)
and Aij = O(ij2 ). The equality (33) means that the determinant of the Majorana matrix of
the RH neutrino components is of the order 2 O(ij2 ) or smaller, and consequently, the RH
neutrino masses have strong hierarchy:
M2
2 2 4 ,
M3
(34)
D
2
whereas mD
2 /m3 . It is this difference of hierarchies which leads to the seesaw enhancement of the 2-3 mixing.
There are three different possibilities to realize the flip of the sign of the neutrino rotation:
1. Change the sign of the off-diagonal mass terms (m

)23 .
2. Change the sign of the diagonal mass term (m
)33 (provided that |(m
)33 | > |(m
)22 |).
3. Enhance the 22 element, so that (m
)22 > (m
)33 .
396
In terms of Eq. (30) the sign of the second (seesaw) term can be changed in four different
ways by appropriately changing the sign of the factors in its numerator and/or denominator.
M,
Numerically, we find this to happen in 5% of cases for randomly generated coefficients 22
M
D
D
23 , 22 and 23 in the allowed range given in (14).
Summarizing, generically, the mass matrix of the left-handed (LH) neutrinos has the
form (31) with moderately enhanced off-diagonal term: |A23 /A33 | 23. The relative sign
of A23 and A33 is negative. In large region of parameter space A22 can be comparable with
two other elements. That corresponds to summing up different (order 2 ) contributions,
thus producing not too strong mass hierarchy.
3.5. Seesaw and the 1-2 neutrino sector
The mass matrix of the light neutrinos can be written as
diag diag 1 T diag T
m
= UL m
D V M R
V m
D UL ,
(35)
where
V = UR UM = UR12
UR13
UR23
UM23 UM13 UM12 ,
(36)
D
and UR and UM are the rotations of the RH neutrino components which diagonalize m
and M R correspondingly.
We find in the lowest order in and :
R sin sin R sin

cos 12
sin 12 cos 12
13
23
12
sin 12
cos 12
sin 23
V
,
M sin + sin M sin
cos 12
13
23 sin 23
12
1
(37)
ij ijM
ijR , and the angles ij are determined in Eqs. (19), (20) and (23).
M 2 M , according to (23) the angle M can be near /4, so that
equality 22
23
12
R = O().
1, cos 12 1, sin 23 = O(), sin 13 = O(2 ) and sin 12
where
Due to
sin 12
Using these estimations we find
1
13 23
,
V 1
23
23
23
(38)
M D ) and = ( M D )2 . Taking the hierarchy of the mass eigenwhere 23 = (23

13
23
13
13
diag
diag
D (4 , 2 , 1) we find from (38) an estimate of
values as M R (4 , 2 2 , 1) and m
the light neutrinos mass matrix (before the LH rotations):
3
6 / 2
diag diag 1 T diag

V m
D
m
D V MR
3

23 1
23 /2

23 1
23 /2 m3 .
1
(39)
397
Table 2
Majorana masses of the RH neutrinos, mee , m1 and predicted value of |Ue3 |. To extract the value of the RH
neutrino masses we take m3 = 0.045 eV. Mi are given in GeVs, whereas the masses mee and m1 are in eVs
Example I(l)
Example II(l)
M1
M2
M3
mee
m1
|Ue3 |2
1.3 1010
3.0 1010
8.6 1014
2.5 108
2.2 1011
3.8 1014
0.0006
0.0007
0.002
0.004
0.008
0.001
Note that 13 does not contribute in the leading order in . If we set 23 2 the light
neutrino mass matrix in the flavor basis with the LH rotating included takes the form
2

m
1 m3 ,
(40)
1 1
where we show only the leading terms in both and . Notice that in the 12 element of (40)
the combination should be enhanced to generate large 1-2 mixing. This will lead, after
the 2-3 rotation, to the 1-3 term of the order 2 according to our general considerations.
4. Phenomenological consequences
The corrections ijK have been introduced in a certain way (13) and they are restricted
to be small enough (14). This allows us to draw some qualitative consequences. Though
exact predictions would require determination of ijK .
For illustration, in the Table 4 we present two examples of the matrices of corrections.
They correspond to two different realizations of the sign flip: the Example I(l) implements
the inequality (m
)22 > (m
)33 and in the Example II(l) the element (m
)23 changes the
sign. For simplicity we take 13 = 11 = 0. With these corrections the mass matrices reproduce precisely the lepton mixings, charged lepton masses and the neutrino mass squared
differences.
The predictions from these two sets of matrices are given in the Table 2 where we
present values of the lightest neutrino mass, the effective Majorana mass of the electron
neutrino, the value of Ue3 and masses of the RH neutrinos. Since the neutrino mass spectrum is hierarchical the radiative corrections are very small [29,30].
4.1. 1-3 mixing
K = 0 (K = l, D, M),
Generically for the 1-3 mixing we expect Ue3 2 0.07. If 13
we find

me
l
sin 23 + O 4 ,
Ue3 = sin 13 sin 13 cos 23
(41)
m
and l are the angles of rotations which diagonalize the neutrino and charged
where 13
13
l . In (41) the last term is induced by simultaneous
lepton mass matrices, and 23 23
23
12 and 23 rotations. Notice that in the sum each contribution is of the order 2 and the next
398
order correction is very small. So, depending on sign of the angle and phase one may get
substantial cancellation of the terms, and even Ue3 = 0 can be achieved. If, however, the
terms sum up we can get Ue3 0.2 which corresponds to the present upper experimental
bound.
We also refer to the results of the numerical analysis summarized in the Table 2. (For details on numerical procedure see Appendix A.) In the examples considered in Appendix A,
|Ue3 |2 is indeed of the order 4 .
4.2. The absolute scale of neutrino mass and mee
According to our general consideration in Section 3.5, the spectrum of light neutrinos is
hierarchical, so that numerically m3 and m2 are determined by the mass
squared differences
measured in the atmospheric and solar neutrino experiments: m3 m2atm
= 0.045 eV

and m2 m2sol . Parametrically m2 /m3 = 2 which implies that = 2.7. The lightest
mass can be found evaluating the determinants of the matrices in (12). Indeed, parametri2 , where represents linear combinations of K coefficients, so that
cally Det(YK ) = 6 K
K
ij
4 / 2 . Then, we have:
the determinant of the seesaw matrix Det(m
) = 6 D
M
m1 =
4
Det(m
)
= 4 D2 m3 ,
m2 m3
M
(42)
where D = D (ijD ), M = M (ijM ) and = 2.7. In the examples presented in the Table 4
the hierarchy of light masses is rather weak: m1 /m2 = 0.20.5 which is partly related to
strong hierarchy of the RH neutrino masses. For the lightest mass we get (see the Table 2)
typically
m1 (0.15) 103 eV.
(43)
4 / 2 (1102 ) which can be used to estimate how

Taking this into account we obtain D
M
singular YD and YM are with respect to each other.
The effective Majorana mass of the electron neutrino can be calculated immediately
as the ee-element of the neutrino mass matrix in the flavor basis (40). Parametrically this
gives mee = 2 m3 , where stands for the linear combination of ij s.
Alternatively, we can use the neutrino masses and known neutrino mixing and present
mee as the sum of contributions of mass eigenstates:

|Uei |2 mi eii = mee (1) + mee (2) + mee (3).
mee =
(44)
i
In general,
due to smallness of the 1-3 mixing the contribution of 3 is very small:
mee (3) = m2atm 4 2 104 eV. The contribution of 2 is phenomenologically de
termined: mee (2) = m2sol sin2 sol (23) 103 eV, and usually dominates. The
contribution of 1 can be comparable with the previous one due to weak mass hierarchy
and larger admixtures of e . Furthermore, typically the masses and therefore contributions
399
of 1 and 2 have an opposite sign cancelling each other in mee . For this reason the predictions for mee in the examples of the Table 4 are small: mee 103 eV.
If the HeidelbergMoscow positive result [37] is confirmed, either our approach, at
least in its present form, is not correct or another mechanism, different from the Majorana
mass of the light neutrinos gives main contribution to the decay rate.
4.3. Leptogenesis
The corrections ijK are in general complex numbers and this is the source of CP violation in our approach.
Since M3 M2 , M1 , only two lighter RH neutrinos are relevant for leptogenesis and
the lepton number asymmetry can be written as [3136]
L =
1 M1 (h h)212
.
8 M2 (h h)11
(45)
Here h is the matrix of the Yukawa couplings in the basis where M R is diagonal. Apparently,
diag2
D
T m
V,
h h=V
(46)
v12
where V is determined in (38). Using estimations of the matrix elements of V (and assuming that the imaginary parts of these elements can be as large as the real ones) we find the
asymmetry
5
5 105 .
8
Then the baryon to photon ratio is given by
L =
(47)
B 0.01Lk1 ,
(48)
where k describes the washout of the produced lepton asymmetry due to weak deviation
from the thermal equilibrium. The factor k depends on the effective mass parameter
m
1 =
v12 (h h)11
(0.11) eV.
M1
(49)
For this value of the effective mass we get k1 103 , and therefore B 5 1010 in
agreement with the observed value.
Notice that the key difference of our scenario from the analysis in [37] is that the lightest
eigenvalue of the neutrino Dirac matrix is much larger than the up quark mass mu ; in the
Example I(l): m1D 300 MeV. Also, the left rotations are not negligible here.
4.4. Lepton number violating effects
In the SUSY context one expects observable flavor violating decays, like e , due
to slepton mixing related to the neutrino mixing. The approximate formula for the e
400
branching ratio, which has the most stringent experimental limit, reads
2
2
3
mL 21
tan2 ,
BR( e ) 2
m8s
GF
(50)
where ms stands for the effective mass of the superparticles, and m2 represents the offL
diagonal corrections to the slepton mass matrix. They appear due to the renormalization
group running between the scale where universality conditions on SUSY breaking parameters are imposed, which we take to be the GUT scale MGUT , and scale where the RH
neutrinos decouple from the theory.
We find

m2L

ij
diag
3m20 + A20
MGUT
(VL )i3 YD 33 VL 3j ln
,
M3
8 2
(51)
where the matrix VL = Ul UL represents the mismatch in the LH rotations that diagonalize
Yl and YD . The relevant coefficients (VL )13 and (VL )23 for the e process are proportional to 2 O() ( 3 ) and O() ( 2 ), respectively. Though, the exact values depend
on combinations of , we expect the product (VL )13 (VL )23 to be close to Vub Vcb 5 .
(See also the form of V in Eq. (37) and the discussion on the mixing in the quark sector.)
Rather precise approximation for the effective mass ms is given by [38]
2

2
2 2
m0 + 0.6M1/2
,
m8s 0.5m20M1/2
(52)
where m0 is the typical slepton mass and M1/2 is the gaugino mass. Taking for simplicity,
m0 = M1/2 m and value (VL )13 (VL )23 = 5 we obtain
4

m
BR( e ) 1.8 109
(53)
,
100 GeV
diag
where = 0.26, tan = 55.9, (YD )33 0.7, MGUT /M3 = 100 and A0 = 0 were used.
In the case of an exact quarklepton symmetry: (VL )13 (VL )23 = Vub Vcb we find
4

m
BR( e ) 1.1 1011
(54)
.
100 GeV
According to Eqs. (53) and (54) for m (300400) GeV we expect BR( e )
10131011 . This interval is close to the current experimental limit of BR( e ) <
1.2 1011 [39] and clearly within reach of the MEG experiment at PSI [40] which will
have a sensitivity down to BR( e ) 5 1014 .
The results of an exact numerical running1 of slepton mass matrix are also in an excellent agreement with the approximations presented in this section.
Finally, we note that the majority of the SUSY GUT models yields significantly larger
value for the product (VL )13 (VL )23 than what is generated in our approach. This puts them
1 Due to the highly non-degenerate spectrum of the RH neutrinos the care has been taken to integrate them
out at the appropriate energy scales as suggested in [41].
401
in precarious position with respect to the experimental constrains on lepton flavor violating
processes. Namely, they typically yield (VL )13 (VL )23 102 101 (see, for example, [42]
and references therein) which makes them violate the experimental bounds even for low
values of tan ( 5). On the other hand generically we obtain (VL )13 (VL )23 104 103 ,
which can be traced back to the ansatz (10). This rather large suppression more than compensates the enhancement of e branching ratio that originates from the large value
of tan . The suppression brings our prediction for e branching ratio close to but
below the current experimental limit.
5. Discussion and implications

(1) There are two different approaches to the theory of fermion masses. One possibility
(widely explored in the literature) is to build up the theory immediately on the basis of
observablesthe mixing and mass ratiosconsidering them as fundamental parameters.
In this case the quarklepton symmetry is strongly broken at least by masses of the first and
second generations. In a number of models this is described by introduction of different
charges for the leptons and quarks. Another approach is when the quarklepton symmetry
is weakly broken. In this case, the observables appear as diagonalization of the nearly equal
mass matrices. They are determined by small corrections to the dominant structure equal
for quarks and leptons.
(2) The main feature of our approach is the nearly singular matrices YK . This allows us,
using small perturbations, to generate strong difference of the mass hierarchies of quarks
and leptons and simultaneously enhance the lepton mixing. The lepton mixing (due to the
seesaw mechanism) is unstable with respect to perturbation of the RH mass matrix which
appears in the denominator of the expression for the light neutrino masses. The perturbations of MR influence strongly the mass hierarchy of the RH neutrinos and therefore (via
the seesaw enhancement) the mixing of light neutrinos.
(3) The matrix Y0 (10) can be obtained in the model with U(1) family symmetry in the
context of the FroggattNielsen [43] (FN) mechanism. According to this mechanism the
Yukawa couplings are generated by the operators

qi +qj
c
fj L
Hk ,
aij fiL
(55)
MF
where fi are fermion components, aij are dimensionless constants of order 1, and is
the scalar fieldsinglet of the SM gauge symmetry group with a U (1)F charge of 1.
Hk (k = 1, 2) are the Higgs doublets of the MSSM, MF corresponds to mass scale at
which the non-renormalizable operators describing interactions of with fermion fields
are generated and qi (i = 1, 2, 3) are the U (1)F charges of the family i.
After develops the VEV
the following Yukawa couplings are generated:
(Y )ij = aij qi +qj ,
.
MF
(56)
If aij = a0 = 1, the matrix Y = Y0 is singular. Furthermore, prescribing the U (1)F charges

0, 1, and 2 for the third, second, and first family we reproduce the required structure of
402
matrix Y0 (10). Corrections to Y0 can be generated by deviations of aij from universality:

aij = a0 (1 + ij ).
(57)
In general, the required singular matrix can be represented as the product:

Y0 = W W T ,
W T (a1 , a2 , a3 ).
(58)
In turn, such a structure can appear as a result of interaction of the light fermion fields
(f1 , f2 , f3 ) with a single heavy field F . Let us consider the following mass terms:
FL
3

i fiR + FR
i=1
3

i fiL + h.c.
(59)
i=1
with i < M and the Dirac mass terms formed by fiL and fiR are forbidden by some
symmetry. Then after decoupling of F we get for the light masses
i j
mij =
(60)
M
with required properties. Notice, however, that this mechanism cannot be applied immediately to top quark since mt i |max .
(4) To reproduce observables we still need small deviations of coefficients aij from 1.
This may come from the F-N mechanism itself as it is indicated in (57) or from new physics
at some higher scale as an additional contribution to mass matrices. The correction matrix
is of the order 2 3 (13) 102 . So, if the flavor symmetry is realized at the GUT
scale the correction matrix can be related to some physics at the string scale.
(5) The case of unstable matrices reproduces to some extent a situation of anarchy:
small perturbations of the otherwise symmetric pattern lead to significant difference in the
observables.
Selecting ai in (58) one can further optimize the structure of the dominant singular
matrix to reduce spread of the corrections ijK , to diminish their absolute values or to get
certain relations [28].
(6) To explain the observed masses and mixing certain relations between the correction
parameters should be satisfied and some of them should be in narrow ranges. These
relations should be used to construct the theory of . Random selection of values of ||s
in the intervals 0 produces typically incorrect values of the observables.
6. Conclusions
We have elaborated an approach in which no ad hoc symmetry for the neutrino sector
is introduced. The difference of parameters in the quark and the lepton sectors arises essentially from the seesaw mechanism as well as from instability of the mass matrices.
The difference of the mass hierarchies follows from small perturbations of the singular
matrices. Singularity can be a consequence of certain family symmetry.
The explanation of features of observables is reduced to large extent to explanation of
perturbations (corrections). Particular values of s are needed. Still our proposal opens
alternative approach to explain the data. Furthermore, in this approach one gets
1.
2.
3.
4.
5.
6.
403
correct hierarchy of the quark mixings;

hierarchical mass spectrum of light neutrinos;
1-3 mixing of the order 2 ;
small effective Majorana mass of the electron neutrino: mee 102 eV;
in general, deviation of the 2-3 mixing from maximal;
generic prediction is a strong mass hierarchy of the second and third RH neutrinos
which is of the order 4 .
Perturbations of the singular matrix introduced in the form (13) with |ij | = 0.26
allow us to reproduce all available experimental results. Even in parametric form the
approach leads to correct qualitative pattern of masses and mixings though quantitative
description of the data requires precise determination of |ij | within the interval 0. Let
us summarize the information on we have obtained:
We have shown that the data can be well described for all 0.26.
There is rather strict relation (33) required by the enhancement of the 2-3 leptonic
mixing. The same relation also gives enhancement of the 1-2 mixing.
23 for the Dirac mass matrices are determined by the mass ratios for the second and
third generations.
22 elements correlate with 23 and they are restricted by the 2-3 CKM mixing in the
quark sector.
Values of 11 are practically irrelevant.
There are rather complicated relations between other parameters (they also include
parameters of the 2-3 sector) which follow from masses of first generation, the 1-2
leptonic mixing and CKM mixing. These relations do not restrict a given parameter
once other parameters are allowed to change in the intervals |0|.
The description of all available data leaves substantial freedom of variations of these
parameters (12 , 13 ). So one can impose on them additional conditions motivated by
theoretical context (zeros, equalities, etc.) [28].
Acknowledgements
We would like to thank Z. Berezhiani, R. Dermisek, W. Liao and Y. Takanishi for fruitful
discussion.
Appendix A
We present the numerical results for ijK corrections. Input parameters, the masses and
mixings of the matter fields, at the GUT scale used for the numerical fit, are given in the
Table 3. We assume the MSSM particle content below the GUT scale and determine tan
requiring the unification of b and t Yukawa couplings.
In Eqs. (A.1) and (A.2) we present two examples of correction matrices of quarks which
yield exact agreement with the experimental input in the Table 3.
404
Table 3
Experimental values of the quark and charged lepton masses and relevant CKM angles extrapolated to the GUT
scale. Three-loop QCD and one-loop QED renormalization group equations are used in running up to mt . Further
extrapolation from mt to MGUT = 2.3 1016 GeV is done using the two-loop MSSM beta functions taking all
SUSY particles to be degenerate at mt and assuming tan = 55.9. Masses are given in GeV
mu
mc
mt
md
ms
mb
me
|Vus |
|Vub |
|Vcb |
0.000558
0.264
121
0.00137
0.0239
2.16
0.000530
0.110
2.45
0.222
0.00284
0.0320
Table 4
K in the lepton sector (K = l, D, M) for two different cases which realize different scenarios for
Corrections ij
the flip of the sign of rotations. The fit is performed assuming m2 /m3 = 0.187 at the GUT scale. We also require
tan2 sol = 0.4 and sin2 2atm = 0.95
Example I(l) ((m
)33 < (m
)22 )
0
0.171 0.036
l
= 0.171 0.254 0.268
0.036 0.268
0
0
0.233
0
D
= 0.233 0.104 0.042

0
0.042
0
0
0.0065
0
M
= 0.0065 0.0098 0.005

0
0.005
0
Example I(q):

0
u
= 0.0683
0.0103

0
d
= 0.0387
0.163

0.0683 0.0103
0.144
0.0526 ,
0.0527
0

0.0387
0.163
0.00386 0.0821 .
0.0821
0
Example II(q):

0
0.00811
0.0100
0.0200
0.00782 ,
u
= 0.00811
0.0100 0.00782
0

0
0.0112 0.160
d
= 0.0112 0.110 0.141 .
0.160 0.141
0
Example II(l) ((m

)23 < 0, (m
)33 > 0)
0
0.093 0.006
l
= 0.093 0.262 0.262
0.006 0.262
0
0
0.213
0
D
= 0.213 0.200 0.098

0
0.098
0
0
0.130
0
M
= 0.130 0.264 0.129

0
0.129
0
(A.1)
(A.2)
The coefficients (yu , yd ) in the Examples I(q) and II(q) are (0.645, 0.655) and (0.650,
0.659), respectively. This difference can also be accounted as 2 correction to (33) elements. The parameter is always 0.26.
We next specify two examples in the lepton sector in the Table 4. In the spirit of the
simplest SO(10) model we set yD = yu
= 0.645 for both cases. Our fit yields yl = 0.753
in the Example I(l) and yl = 0.754 in the Example II(l).
405
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]
C. Giunti, hep-ph/0309024.
A.Y. Smirnov, hep-ph/0311259.
H.V. Klapdor-Kleingrothaus, et al., Eur. Phys. J. A 12 (2001) 147.
H.V. Klapdor-Kleingrothaus, et al., Mod. Phys. Lett. A 16 (2001) 2409.
F. Feruglio, A. Strumia, F. Vissani, Nucl. Phys. B 637 (2002) 345;
F. Feruglio, A. Strumia, F. Vissani, Nucl. Phys. B 659 (2003) 359, Addendum.
A.M. Bakalyarov, A.Y. Balysh, S.T. Belyaev, V.I. Lebedev, S.V. Zhukov, C03-06-23.1 Collaboration, hepex/0309016.
H.V. Klapdor-Kleingrothaus, A. Dietz, I.V. Krivosheina, O. Chkvorets, hep-ph/0403018.
S.W. Allen, R.W. Schmidt, S.L. Bridle, astro-ph/0306386.
D.N. Spergel, et al., Astrophys. J. Suppl. 148 (2003) 175, astro-ph/0302209.
O. Elgaroy, O. Lahav, JCAP 0304 (2003) 004.
S. Hannestad, JCAP 0305 (2003) 004.
C.I. Low, hep-ph/0404017.
E. Ma, G. Rajasekaran, Phys. Rev. D 64 (2001) 113012;
K.S. Babu, E. Ma, J.W.F. Valle, Phys. Lett. B 552 (2003) 207.
R. Barbieri, L.J. Hall, G.L. Kane, G.G. Ross, hep-ph/9901228.
S. Antusch, S.F. King, hep-ph/0402121.
S. Antusch, S.F. King, hep-ph/0403053.
G. Altarelli, F. Feruglio, hep-ph/0306265.
Super-Kamiokande Collaboration, Y. Hayato, talk given at the HEP2003 International Europhysics Conference, Aachen, Germany, 2003, http://eps2003.physik.rwth-aachen.de.
M.H. Ahn, et al., Phys. Rev. Lett. 90 (2003) 041801.
S. Weinberg, Trans. New York Acad. Sci. 38 (1977) 185.
F. Wilczek, A. Zee, Phys. Lett. B 70 (1977) 418;
F. Wilczek, A. Zee, Phys. Lett. B 72 (1978) 504, Erratum.
H. Fritzsch, Phys. Lett. B 70 (1977) 436.
S. Barshay, G. Kreyerhoff, Europhys. Lett. 63 (2003) 519;
S. Barshay, P. Heiliger, Astropart. Phys. 6 (1997) 323.
M. Gell-Mann, P. Ramond, R. Slansky, in: P. van Niewenhuizen, D.Z. Freedman (Eds.), Supergravity, NorthHolland, Amsterdam, 1980;
P. Ramond, Sanibel talk, retroprinted as hep-ph/9809459;
T. Yanagida, in: O. Sawada, A. Sugamoto (Eds.), Proc. of Workshop on Unified Theory and Baryon Number
in the Universe, KEK, Tsukuba, 1979;
S.L. Glashow, in: M. Lvy (Ed.), Quarks and Leptons, Cargse Lectures, Plenum, New York, 1980, p. 707;
R.N. Mohapatra, G. Senjanovic, Phys. Rev. Lett. 44 (1980) 912.
A.Yu. Smirnov, Phys. Rev. D 48 (1993) 3264;
M. Tanimoto, Phys. Lett. B 345 (1995) 477;
T.K. Kuo, G.-H. Wu, S.W. Mansour, Phys. Rev. D 61 (2000) 111301;
G. Altarelli, F. Feruglio, I. Masina, Phys. Lett. B 472 (2000) 382;
S. Lavignac, I. Masina, C.A. Savoy, Nucl. Phys. B 633 (2002) 139;
A. Datta, F.S. Ling, P. Ramond, hep-ph/0306002;
M. Bando, et al., hep-ph/0309310.
E.K. Akhmedov, G.C. Branco, F.R. Joaquim, J.I. Silva-Marcos, Phys. Lett. B 498 (2001) 237, hepph/0008010.
R. Dermisek, hep-ph/0312206.
I. Dorsner, A.Yu. Smirnov, in preparation.
K.S. Babu, C.N. Leung, J. Pantaleone, Phys. Lett. B 319 (1993) 191, hep-ph/9309223.
M. Frigerio, A.Y. Smirnov, JHEP 0302 (2003) 004, hep-ph/0212263.
M.A. Luty, Phys. Rev. D 45 (1992) 455.
M. Flanz, E.A. Paschos, U. Sarkar, Phys. Lett. B 345 (1995) 248;
M. Flanz, E.A. Paschos, U. Sarkar, Phys. Lett. B 382 (1996) 447, Erratum, hep-ph/9411366.
406
[33]
[34]
[35]
[36]
[37]
[38]
[39]
[40]
[41]
[42]
[43]
M. Plumacher, Z. Phys. C 74 (1997) 549, hep-ph/9604229.

L. Covi, E. Roulet, F. Vissani, Phys. Lett. B 384 (1996) 169, hep-ph/9605319.
W. Buchmuller, M. Plumacher, Phys. Lett. B 431 (1998) 354, hep-ph/9710460.
G.F. Giudice, A. Notari, M. Raidal, A. Riotto, A. Strumia, hep-ph/0310123.
E.K. Akhmedov, M. Frigerio, A.Y. Smirnov, JHEP 0309 (2003) 021, hep-ph/0305322.
S.T. Petcov, S. Profumo, Y. Takanishi, C.E. Yaguna, Nucl. Phys. B 676 (2004) 453, hep-ph/0306195.
K. Hagiwara, et al., Particle Data Group Collaboration, Phys. Rev. D 66 (2002) 010001.
T. Mori, Nucl. Phys. B (Proc. Suppl.) 111 (2002) 194.
S. Antusch, J. Kersten, M. Lindner, M. Ratz, Phys. Lett. B 538 (2002) 87, hep-ph/0203233.
S.M. Barr, Phys. Lett. B 578 (2004) 394, hep-ph/0307372.
C.D. Froggatt, H.B. Nielsen, Nucl. Phys. B 147 (1979) 277.
Nuclear Physics B 698 [PM] (2004) 409449
Wn(2) algebras
B.L. Feigin a , A.M. Semikhatov b
a Landau Institute for Theoretical Physics, Russian Academy of Sciences, Russia
b Theory Division, Lebedev Physics Institute, Russian Academy of Sciences, Russia
Received 6 February 2004; received in revised form 21 June 2004; accepted 25 June 2004
Available online 20 July 2004
Abstract
(2)

We construct W -algebra generalizations of the s(2)
algebraW algebras Wn generated by two
currents E and F with the highest pole of order n in their OPE. The n = 3 term in this series is
(2)
the BershadskyPolyakov W3 algebra. We define these algebras as a centralizer (commutant) of
the Uq s(n|1) quantum supergroup and explicitly find the generators in a factored, Miura-like
(2)

s(n).
The relation
form. Another construction of the Wn algebras is in terms of the coset s(n|1)/

between the two constructions involves the duality (k + n 1)(k + n 1) = 1 between levels k

algebras.
and k of two s(n)
PACS: 11.25.Hf
1. Introduction

The affine Lie algebra s(2)
(probably the second popular in conformal field theory

after the incontestable Virasoro algebra) is not only the first term in the sequence {s(n)}
(2)
of affine Lie algebras, but also the first term in a sequence {Wn } of W -algebras generated
by two dimension- 2n currents En (z) and Fn (z) whose operator product expansion starts
with a central term over the nth-order pole. These Wn(2) algebras are defined and studied in
this paper.
E-mail addresses: feigin@mccme.ru (B.L. Feigin), ams@sci.lebedev.ru (A.M. Semikhatov).
doi:10.1016/j.nuclphysb.2004.06.056
410
B.L. Feigin, A.M. Semikhatov / Nuclear Physics B 698 [PM] (2004) 409449
(2)

Their structural similarity with s(2)
allows studying the Wn minimal models of

conformal field theory by essentially the same tools that are successfully used in s(2)
(2)
minimal models. For n = 3, W3 is the BershadskyPolyakov (BP) algebra [1,2]. Its study
in [3] reveals the mechanism underlying the construction of unifying W -algebras [4] and
shows how representations of W3(2) (and presumably of the higher algebras) can be built
along the lines of semi-infinite constructions [5,6], from currents subject to constraints

generalizing the famous s(2)
relation e(z)k+1 = 0; this also naturally evokes Landau
Ginzburg models (see [7] and references therein). Another motivation for the study of Wn(2)
algebras is that they can be defined in terms of fermionic screening operators, the interest in
which is in particular related to toric manifolds [8] and, again, LandauGinzburg models.
(2)

The Wn algebras depend on a parameter k C. For n = 2, k is the s(2)
level; for
(2)

n = 3, k is the level of the s(3)
algebra from which the BP W -algebra W3 is obtained via
(2)
the defining Hamiltonian reduction [1,2,9]. Each Wn (k) contains a Virasoro subalgebra
generated by the energymomentum tensor Tn (z) with central charge
cn (k) =
((k + n)(n 1) n)((k + n)(n 2)n n2 + 1)

k+n
(1.1)
and a Heisenberg subalgebra generated by the modes of the dimension-1 current Hn (z),
with the OPEs1
n (k)
n1
k + n 2,
, n (k) =
2
(z w)
n
En
Fn
Hn (z)En (w) =
,
Hn (z)Fn (w) =
.
zw
zw
Hn (z)Hn (w) =
(1.2)
The operator product expansion of En and Fn starts as

En (z)Fn (w) =
n1 (n, k) nn2 (n, k)Hn (w)

+
+ ,
(z w)n
(z w)n1
(1.3)
where i (n, k) are numerical coefficients defined in (A.2) and the dots denote lower-order
poles involving operators of dimensions 2. The energymomentum tensor is extracted
from the pole of order n 2 such that are En and Fn are dimension- 2n primary fields (and
Hn is a dimension-1 primary), see (A.1).
(2)
(2)
The notation Wn extends the notation W3 used for the BP algebra, which was origi
nally derived as a second Hamiltonian reduction of s(3)
and therefore had to be denoted
similarly to but distinctly from W3 . We tend to interpret the superscript (2) differently, as a
(2)
(2)

reminder that the Wn algebras are relatives of s(2).
The lower Wn algebrasthe BP
(2)
(2)
algebra W3 and the otherwise not celebrated algebra W4 are described in more detail
in Appendix A.
Each algebra Wn(2) can be derived in at least two different ways: from the centralizer of

the quantum group Uq s(n|1) and from the coset theory s(n|1)/
s(n).
1 Normal-ordered products of composite operators are understood and regular terms are omitted in operator
products.
411
(2)
Wn as a centralizer of Uq s(n|1) The standard definition of the W -algebra associated

with a given root system is as the centralizer of the corresponding quantum group [10]
(cf. [11]); the W -algebra can be thought of as invariants of the quantum group acting on a Heisenberg algebra. We recall that a general principle in the theory of vertexoperator algebras consists in a certain duality between vertex-operator algebras and quantum groups. In free-field realizations of a vertex-operator algebra A, the corresponding
quantum group Uq g is represented by screening operators; more precisely, screenings are
elements of Uq n, the nilpotent subalgebra in Uq g. They can be thought of as symmetries
of the conformal field theory associated with A: the two algebras, A and Uq g, are each
others centralizers (commutants) in the algebra of free fields. More precisely, the vertexoperator algebra A is the centralizer of Uq g in the algebra of local operators; in turn, the
centralizer of A in fact contains two quantum groups, which commute with each other and
one of which typically suffices to single out A as its centralizer.2
In the standard construction of W -algebras, the centralizer of Uq g is sought in the algebra V of operators that are descendants of the identity (in other words, correspond to states
in the vacuum representation). We generalize this by first adding one extra Heisenberg algebra (scalar current) and then extending V by a one-dimensional lattice vertex-operator
algebra. We thus seek the quantum group centralizer in V = {P ()em(,) | m Z},
where is a lightlike lattice vector, is a collection of scalar fields, ( , ) is the Euclidean
scalar product, and P are differential polynomials in all components of (and = z

);
this involves one scalar field (Heisenberg algebra) more than in the standard construction.
For chosen specially (and for generic values of k), such a centralizer of Uq s(n|1) is
the Wn(2) algebra; the required choice of the vector is determined by quantum group
considerations, as we discuss in detail in the paper.
We find the Wn(2) generators (En (z), Hn (z), Fn (z)) in the centralizer of Uq s(n|1)
explicitly. Before describing our construction, we note that because s(n|1) admits inequiv(2)
alent simple root systems, we have to consider different realizations Wn[m]
, 0 m n,
which are isomorphic W -algebras, but whose generators are constructed through free fields
differently.3 For m = 0 and m = n, the respective Dynkin diagrams are (with n nodes in
each case)
and
which are the same. For 1 m n 1, the corresponding Dynkin diagrams are (n nodes)
2 For example, for Virasoro theories, both quantum groups are the quantum s(2); for s(2)

theories, the two

s(2)/
s(2),
these are Uq D(2; 1|) and
quantum groups are Uq s(2|1) and Uq s(2); for the coset s(2)
Uq1 s(2) Uq2 s(2) Uq3 s(2) [12].
3 In general, there seems to be no theorem that the W -algebras obtained as centralizers of screenings corresponding to different root systems of the same superalgebra are isomorphic, but in all known examples, the
different root systems lead to the same W -algebra.
412
with n m 1 black dots (even roots) to the left of the white dots (odd roots). To describe the generators in the centralizer of the corresponding quantum group, we let Ri+ ,
i = 0, . . . , n m 1, and Ri , i = 0, . . . , m 1, be free-field currents with the OPEs
1
,
(z w)2
k n + 1
Ri+ (z)Rj+ (w) = Ri (z)Rj (w) =
,
(z w)2
k+n1
,
Ri+ (z)Rj (w) =
(z w)2
and let Y be the current with the OPEs
1
Ri (z)Y (w) =
,
Y (z)Y (w) = 0
(z w)2
Ri+ (z)Ri+ (w) = Ri (z)Ri (w) =
i = j,
and the corresponding scalar field, such that e (z) = :Y e :(z) (with =
(1.4)
(1.5)
z ).
(2)
(k) are given by
Theorem 1.1. The two currents generating Wn[m]

En[m] (z) = : (k + n 1) + Rm1

(z) (k + n 1) + R1 (z) R0 (z) e (z):,

+
(z)
Fn[m] (z) = : (k + n 1) + Rnm1

(k + n 1) + R1+ (z) R0+ (z) e (z) :.
Here, = z
and the action of the derivatives is delimited by the outer brackets (i.e., the
derivatives do not act on the exponentials). For m = 1, only the R0 factor is involved in
En[1] , and for m = 0, En[0] (z) = e (z) . For m = n 1, only the R0+ factor is involved in
Fn[n1] , and for m = n, Fn[n] (z) = e (z) .
These formulas can be obtained by noting that free-field realizations of the generators of
(2)
(2)
Wn[m] (k) are related to free-field realizations of the generators of W(n1)[m] (k + 1). They
(2)
generalize the following well-known situation: the bosonic system (which is W1 ) is

bosonized through two scalar fields; this bosonization is involved in the symmetric real(2)

), which can be obtained by rebosonizing the Wakimoto
ization of s(2)
(which is W2[1]
representation. This has an analogue for all n; in addition, relations between the maximally asymmetric realizations of two subsequent algebras lead to the factored form of the
generators in Theorem 1.1. We do not prove that the En[m] (z) and Fn[m] (z) currents in
Theorem 1.1 generate the entire centralizer, but we believe that this is true for generic k.
(2)
where Ws
It follows from the definition that Wn contains the subalgebra Ws Uh,
is the standard W -algebra associated with the s(n|1) root system and h is a Heisenberg algebra commuting with it. The Ws(n|1) Uh subalgebra is merely the zero momentum
(2)
sector of Wn , i.e., consists of elements in the centralizer of Uq s(n|1) in V, descendants
of the identity operator. Introducing the generalized parafermions W n(2) as the quotient over
the Heisenberg subalgebra, we thus have
W n(2) = Wn(2)/h,
W n(2) Ws(n|1).
413
(2)
The algebra W n is nonlocal, but we use it because it naturally appears in some constructions and its locality can easily be restored by tensoring with an additional free scalar
field, thus recovering Wn(2) .
Although this is not in the focus of our attention in this paper, we note that Ws(n|1) are
the unifying W -algebras [4] that interpolate the rank of Ws() algebras. The underlying
numerology is as follows [4]. We take the W -algebra Ws(m), m N, and consider its
(p, p ) minimal model. Its central charge is given by
cp,p (m) = 2m3 m 1 (m 1)m(m + 1)

(2)
The central charge of Wn (k) with k = 1 n +
p
p
(m 1)m(m + 1) .
p
p
m+1
n1
satisfies the relation
cn (k) 1 = cm+1,m+n (m).

This suggests a ranklevel duality of the corresponding minimal models,
Ws(n|1)1n+ m+1 = Ws(m)m+n,m+1 .
n1
(1.6)
The algebra of this minimal model closes on normal-ordered differential polynomials in

the currents of conformal dimensions 2, 3, . . . , 2n + 1 [4], which is often expressed in
the notation W(2, 3, . . . , 2n + 1) for the model. The mechanism underlying (1.6) was also
discussed in [3]. The Ws(m)m+n,m+1 minimal model can be extended by taking the tensor
product with a certain lattice vertex-operator algebra generated by a free field . A certain
primary field of Ws(m)m+n,m+1 can then be dressed with e such that the resulting
operator becomes local and can be identified with the E current of Wn(2) ; another (dual)
primary is dressed into the F current by e . This gives integrable representations of
(2)

Wn , generalizing the integrable representations of s(2).
(2)
(2)

Alternatively to the quantum-group description, the Wn (k)
Wn algebras from s(n|1)
algebras can be given another characterization, in terms of coset conformal field theories.
The algebras involved in the coset construction are with the dual level k related to k by
(k + n 1)(k + n 1) = 1.
(1.7)
For k = n, Eq. (1.7) is equivalent to

1
1
+
= 1.
k+n k +n
(1.8)

Theorem 1.2. Let k be related to k by (1.7). Then the coset of s(n|1)
k with respect to its

even subalgebra g(n)k is given by

k = W (2) (k)/h
s(n|1)
k /g(n)
n[m]
(1.9)
(where the h subalgebra is generated by (the modes of ) the current Hn (z)) for any m
(2)
algebras with m [0, n] are isomorphic.
[0, . . . , n]. In particular, all the Wn[m]
414

For n = 2, we thus recover the well-known identification s(2)/
h = s(2|1)/
g(2).

The cosets s(n|1)/
s(n),
which are almost the left-hand side of (1.9), can be con
structed in rather explicit terms as follows. In s(n|1),
the 2n fermions are organized into

n
two s(n) n-plets Cn (z) and C n (z). Then
Cn (z) and n C n (z) are in the centralizer of

the s(n)
subalgebra. The coset s(n|1)/
s(n)
is isomorphic to the centralizer of s(n)
in

Us(n|1).
This is an essential simplification compared with the general case, where a coset
is usually defined as the cohomology of some BRST operator, and only the Virasoro alge

bra can be realized by operators commuting with the subalgebra. The coset s(n|1)/
s(n)
is quite close to the Wn(2) algebra: after a correction by e n(z) , where is an auxiliary
scalar with the OPE
1
(z)(w) =
(1.10)
,
(z w)2
n n

C (z) and n C n (z) become the two currents generating Wn(2) (k).
The contents of this paper can be outlined as follows. In Section 2, we study the Wn(2)
algebra defined as the centralizer of Uq s(n|1) in a certain lattice vertex-operator algebra.
(2)
In Section 2.1, we define the different realizations of Wn(2) , denoted by Wn[m]
. In Section 2.2, we describe the motivation of our approach leading to the construction of the
(2)
Wn[m] generators in the centralizer of the corresponding screenings. Actual calculations,
eventually leading to Theorem 1.1, are given in Section 2.3. The OPEs following from
(2)
Theorem 1.1 are considered in Section 2.4. In Section 3, we alternatively define Wn in

terms of the coset s(n|1)/
s(n).
The actual statement to be proved is in Theorem 3.1 and
the proof is outlined in Section 3.1. It involves finding another set of screenings representing the Uq s(n) quantum group. The quantum-group structure is then used to construct
(2)

a vertex-operator extension of Wn by the denominator g(n)
into the numerator
(2)

s(n|1),
thus inverting the coset and hence showing that Wn is indeed given by the coset
construction. Some speculations on vertex-operator extensions are given in Section 3.2.
(2)
2. Wn algebras from Uq s(n|1)

In this section, we define the Wn(2) algebra as the centralizer of Uq s(n|1) in a certain
lattice vertex-operator algebra. We then find the algebra generators in the centralizer; we
do not prove that the centralizer is thus exhausted, but we believe that it is for generic k.
2.1. Realizations
(2)
To define the Wn algebra as a centralizer of the screenings representing (the nilpotent

subalgebra of) Uq s(n|1), we proceed as follows. We represent nq s(n|1), the nilpotent subalgebra of Uq s(n|1), by operators expressed through n + 1 scalar fields i ,
i = 1, . . . , n + 1. These operators are called screenings in what follows. We next extend the space of differential polynomials in the currents 1 , . . . , n+1 by adding a
one-dimensional lattice vertex-operator algebra, i.e., the operators em(,) , m Z, with
415
(2)
a specially chosen vector . The Wn algebra is then defined as the centralizer of the
nq s(n|1)-screenings in the space V = {P ()em(,) | m Z}, where P are differential
polynomials.
To specify this in more detail, we note that the s(n|1) Lie superalgebra admits inequivalent simple root systems, and we must therefore distinguish between centralizers of the
screenings associated with each of the inequivalent simple root systems. To the algebras
(2)
constructed for each of the root systems, we refer as realizations of Wn , to be denoted by
(2)
(2)
Wn[m] . We now consider the definitions of Wn[m] .
2.1.1. n[0]
(2)
The maximally asymmetric realization, denoted by Wn[0] , corresponds to the simple root
system of s(n|1) represented by the Dynkin diagram
where filled (open) dots denote even (odd) roots. The corresponding Cartan matrix of
s(n|1) is given by
2 1 0 0
1 2 1 0 0
0 0 1 2 1
0 0 1 0
From this Cartan matrix, we construct the screenings representing nq s(n|1). For this, we
first introduce vectors a n1 , . . . , a 1 , in Cn whose Gram matrix (pairwise scalar products)
is given by the dressed Cartan matrix

0
a n1 2(k + n) k n

0
a n2 k n 2(k + n) k n 0
0 k n 2(k + n) k n
a1

0
0
k n
1
(2.1)
where the leftmost column shows how the vectors are associated with the rows. The determinant of this matrix is (k + n)n1 nn (k), and the vectors are therefore determined
uniquely modulo a common rotation for k C \ {n, n(n2)
n1 }.
Let be the n-plet of scalar fields with the OPEs
i (z)j (w) =
i,j
.
(z w)2
With the above vectors a n1 , . . . , a 1 , , we define the operators

a i .
Ei = e , i = 1, . . . , n 1,
= e. ,
416
where the dot denotes the Euclidean scalar product in Cn . These screening operators represent nq s(n|1). The screenings Ei are said to be bosonic and fermionic.
We next define the vector Cn by its scalar products with a 1 , . . . , a n1 , ,
. a i = 0,
i = 1, . . . , n 1,
. = 1,
(2.2)
(2)
and set V = {P ()em . | m Z}. By definition, the n[0] realization of W n , denoted
(2)
by W n[0] , is the centralizer of (Ei )i=1,...,n1 and in V . Restoring localityi.e., recon(2)
structing Wn[0]
requires one scalar field more.
(2)
To construct Wn , we embed Cn in Cn+1 as a coordinate hyperplane, let ,

a1 , . . . , an1 denote the respective images of , a 1 , . . . , a n1 , and extend the set of free
scalar fields accordingly, to = {, n+1 }. In addition, we define = { , n+1 } such
that is isotropic with respect to the Euclidean scalar product ( , ) in Cn+1 :
2
=0
(, ) = . + n+1
(the other scalar products remain the same as among the n-dimensional vectors above).
With a slight abuse of notation, we then let and Ei denote the screening operators

(ai ,)
Ei = e
(2.3)
, i = 1, . . . , n 1,
= e(,) .
(2)
, is the centralizer of (Ei )i=1,...,n1
The n[0] realization of Wn(2) , denoted by Wn[0]
m(,)
and in V = {P ()e
| m Z}.
In what follows, the differential polynomials in are expressed through the currents
Ai = (ai , ),
Q = (, ),
Y = (, ),
which have the nonzero operator products (read off from the matrix (2.1))
k n
,
(z w)2
k n
,
A1 (z)Q(w) =
(z w)2
Ai (z)Ai+1 (w) =
2(k + n)
,
(z w)2
1
1
Q(z)Q(w) =
,
Q(z)Y (w) =
.
(z w)2
(z w)2
Ai (z)Ai (w) =
Remark 2.1.2. It follows from (2.2) that the n-dimensional part of is given by
=

1
a n1 + 2a n2 + + (n 1)a 1 + n
nn (k)
and
. =
1
.
n (k)
(2.4)
417
We also note that the determinant of the (n + 1) (n + 1) Gram matrix of the vectors
an1 , . . . , a1 , , ,
2(k + n) k n
0
0
0
0
0
k n 2(k + n) k n 0
n (k) =
0 k n 2(k + n) k n 0
0
k n
1
1
0
0
1
0
(2.5)
is given by n(k + n)n1 , and hence these n + 1 vectors form a basis in Cn+1 and are
determined uniquely modulo a common rotation for k C \ {n}. For future use, we note
that
n (k)1
(n1)1
k+n
(n2)1
k+n
(n3)1
k+n
(n4)1
k+n
1
=
11
k+n
0
1
(n2)1
k+n
(n2)2
k+n
(n3)2
k+n
(n4)2
k+n
(n3)1
k+n
(n3)2
k+n
(n3)3
k+n
(n4)3
k+n
(n4)2
k+n
(n4)3
k+n
(n4)4
k+n
(n5)3
k+n
(n5)4
k+n
(n6)4
k+n
12
k+n
13
k+n
14
k+n
0
2
11
k+n
12
k+n
13
k+n
14
k+n
1(n1)
k+n
0
n1
0
0
0
0
0
0
n
4
.

n1
2
3
n
nn (k)
2.1.3. n[1]
(2)
The realization Wn[1]
corresponds to the Dynkin diagram
with two odd roots. Similarly to the n[0] case, we introduce vectors a n2 , . . . , a 1 , + ,
in Cn whose Gram matrix is given by the Cartan matrix dressed into

a n2 2(k + n) k n
0
0

a n3 k n 2(k + n) k n
0
a1
0
0
k n 2(k + n) k n
0
+
0
0
k n
1
k+n1

0
k+n1
1
We also introduce an n-tuple of scalar fields and define the screenings

Ei = eai . , i = 1, . . . , n 2,

+ .
+ = e
,
= e . ,
418
where the dot denotes the Euclidean scalar product in Cn . These screenings represent
nq s(n|1). We next define the vector by its scalar products with a n2 , . . . , a 1 , + , ,
. a i = 0,
i = 1, . . . , n 2,
. + = 1,
. = 1.
(2.6)
(2)
(2)
The n[1] realization of W n , denoted by W n[1] , is the centralizer of (Ei )i=1,...,n2 , + ,
and in V = {P ()em . | m Z}.
(2)
To construct Wn[1] , we embed Cn in Cn+1 as the coordinate hyperplane, let an2 , . . . , a1
+ , denote the respective images of a n2 , . . . , a 1 , + , , and extend the free scalar
fields as = {, n+1 }. In addition, we define = { , n+1 } such that is isotropic
with respect to the Euclidean scalar product ( , ) in Cn+1 (the other scalar products remain
the same as among the n-dimensional vectors above). We again use , + , and Ei to
denote the screenings

Ei = e(ai ,) , i = 1, . . . , n 2,

+ = e(+ ,) ,
= e( ,) .
The screenings + and are said to be fermionic and the other screenings bosonic.
(2)
(2)
The n[1] realization Wn[1] of Wn is the centralizer of (Ei )i=1,...,n2 , + , in V =
{P ()em(,) | m Z}. As in the maximally asymmetric case, we express differential
polynomials in through the currents
Ai = (ai , ),
Q = ( , ),
Y = (, ),
whose OPEs are determined by the matrix above and scalar products (2.6),
k n
,
(z w)2
k n
A1 (z)Q+ (w) =
,
(z w)2
1
Q (z)Q (w) =
,
(z w)2
Ai (z)Ai+1 (w) =
2(k + n)
,
(z w)2
k+n1
Q+ (z)Q (w) =
,
(z w)2
1
Q (z)Y (w) =
.
(z w)2
Ai (z)Ai (w) =
Remark 2.1.4. It follows that

1
=
a n2 + 2a n3 + + (n 2)a 1 + (n 1) +
nn (k)
and
1
. =
.
n (k)
The determinant of the (n + 1) (n + 1) Gram matrix of the vectors an2 , . . . , a1 , + ,
, is given by n(k + n)n1 , and these n + 1 vectors therefore form a basis in Cn+1
and are determined uniquely modulo a common rotation for all k C \ {n}.
419
2.1.5. n[m]
The subsequent n[m] realizations correspond to the Dynkin diagrams
with m 1 even roots to the right of the odd roots. We can restrict m to 0 m [ n2 ],
because taking m > [ n2 ] amounts to applying an automorphism (inducing E F on the
algebra) to the realization where m is replaced by n m.
(2)
We do not repeat the definition of W n[m]
in terms of n scalar fields and proceed to
(2)
defining Wn[m] in terms of n + 1 scalar fields. For this, we introduce (n + 1)-dimensional

vectors anm1 , . . . , a1 , + , , a1 , . . . , am+1 , whose Gram matrix (with the determinant n(k + n)n1 ) is given by

0
0
anm1 2K K

0
0
0
anm2 K 2K K

..
.
0 0 K 2K K
0
0
a1
0 0 K
1
K 1
0
0
1
+
0
K 1
1
K
0
0
1
0
,
a1 0
0
0
0
0
K
2K K

.

..
a2
0
K 2K K 0
0
am+1
0
0
0
K 2K

where K = k + n, the leftmost column indicates labeling of rows, and diagonal elements
are underlined to guide the eye. The screenings that represent nq s(n|1) are given by

Ei = e(ai ,) , i = 1, . . . , n m 1,

(+ ,)
+ = e
,
= e( ,) ,

Ei = e(ai ,) , i = 1, . . . , m + 1.
(2)
The n[m] realization of Wn(2) , denoted by Wn[m]
, is the centralizer of these operators in
V = {P ()em(,) | m Z}. As before, Ei are said to be bosonic and fermionic
screenings.
The information contained in the Gram matrix above can be conveniently reexpressed
as a rigged Dynkin diagram for s(n|1),
(2.7)
420
where K = k + n. The vertices of the diagram are assigned the dimension-1 currents
Anm1 (z), . . . , A1 (z), Q+ (z), Q (z), A1 (z), . . . , Am+1 (z) as indicated. The labels
2K, K, and 1 at the links mean that these currents have the OPEs
2K
K
,
Ai (z)Ai+1 (w) =
,
Ai (z)Ai (w) =
2
(z w)
(z w)2
1
K 1
Q (z)Q (w) =
,
Q+ (z)Q (w) =
,
(z w)2
(z w)2
K
K
A1 (z)Q+ (w) =
,
A1 (z)Q (w) =
.
2
(z w)
(z w)2
In addition, for the current Y = (, ), which is not associated with a vertex in the diagram, we have the nonzero OPEs (as indicated by the + and signs),
Q (z)Y (w) =
1
.
(z w)2
2.2. Centralizer of the screenings: the step-back strategy

(2)
We now describe the strategy that leads to the construction of the Wn[m] generators in
the centralizer of the corresponding screenings. The contents of this subsection is not a
proof, but because it motivates our construction, we hope that it can be useful in finding
centralizers of some other quantum supergroups (a necessary condition is the existence of
fermionic screening(s)). The crucial point is the conditions on the vector (see (2.2), (2.6),
and similar conditions read off from the matrix in Section 2.1.5), chosen from quantum
group considerations.
Quite generally, we recall that the action of screening operators S1 , . . . , SN Uq n on
an operator X = P V (p , z) in a lattice vertex-operator algebra (where V (p , z) is a vertex
with momentum p and P is a differential polynomial) generically gives nonlocal expressions which represent elements of a module K over the quantum group Uq g. This module
can be either a Verma module or (which is most often the case) some of its quotients. But
whenever a singular vector, e.g., v = S1 Sr X (where we write a monomial expression
for simplicity) occurs in K, the corresponding field is locala descendant of the shifted
vertex V (p, z), where the momentum p differs from p by the sum of the momenta of the
relevant screenings, p = p + a1 + + ar . We are interested in the case where a singular
vector in K generates a one-dimensional submodule. Then the corresponding local field is
necessarily in the centralizer of the quantum group. To construct a local field of the form
P V (p, z), we must then start with an appropriate p = p a1 ar .
That is, we seek a quantum group module K with a singular vector v that generates
a one-dimensional submodule in K (the singular vector must therefore have weight 0).
Such singular vectors (and hence the corresponding local fields) can indeed be found for
the s(n|1) root system. We now use this method to construct fields in the centralizer of
Uq s(n|1).
2.2.1. Maximally asymmetric realization
We first consider the maximally asymmetric realization, corresponding to the s(n|1)
simple root system described in Section 2.1.1. We must find a Verma module quotient
421
Fig. 1. Mapping between two s(3|1) modules whose highest-weight vectors are annihilated by the s(3) subalgebra. Filled dots denote elements of submodules. Those in the right module (M3 (0)) vanish under the mapping
into the left module (M3 (2)).
with a singular vector generating a one-dimensional submodule. For this, we take the
Uq s(n|1)-module induced from the one-dimensional representation of the parabolic subalgebra generated byUq g(n) and the fermionic simple root generator. The module is
then isomorphic to Cn as a vector space. Its weight diagram is shown in Fig. 1 for
n = 3 (the upper-left part, with the highest-weight vector at the top). For generic q, such a
Uq s(n|1)-module is a deformation of an s(n|1)-module and is isomorphic to the latter
as a vector space, and we proceed with the corresponding s(n|1)-module for simplicity.
Such s(n|1)-modules depend on a single parameter, the eigenvalue of the g(n) Cartan
generator that is not in s(n). We normalize this generator as
h0 =
..
.
1
n
1
and let |
be the highest-weight vector of the above module with h0 |
= |
; the module is denoted by Mn (). The vector at the bottom of the weight diagram of Mn () is then
singular if and only if = (n 1).
In particular, the vector at the bottom of the weight diagram of Mn (1 n) is annihilated
by h0 , and the highest-weight vector of the Mn (0) module is then mapped onto this singular vector, see Fig. 1. This picture is preserved under deformation to Uq s(n|1). By the
correspondence between modules over the quantum group and the vertex-operator algebra,
the singular vector in (the deformation of) Mn (1 n) determines an intertwining oper(2)
ator between the corresponding Wn[0] representations, realized in a sum of Fock spaces;
this intertwining operator is morally the product of the n fermionic screenings , , . . . ,
422
(n1) obtained by the action of the bosonic screenings on ,

(n1) : Fp Fp ,
where Fp is the module generated from ep. . The image of the highest-weight vector in Fp
is a descendant of the highest-weight vector in Fp ,

(n1) ep . = Pn ep. ,
(2.8)
where Pn is a degree-n differential polynomial in = (1 , . . . , n ). The difference
p p must be equal to the sum of the momenta of , , . . . , (n1) , which are for
the fermionic screening in (2.3) and + a 1 + + a i for each of the other fermionic
screenings obtained by dressing with the bosonic screenings. Therefore,
p = p + n + (n 1)a 1 + (n 2)a 2 + + a n1 = p nn (k) ,
(2.9)
with defined in (2.4).

Next, to ensure that the module is induced from the one-dimensional representation of
the parabolic subalgebra, we require that
p . a i = 0,
i = 1, . . . , n 1
(with the scalar products among a i and given in Section 2.1.1, this is equivalent to

p . a i = 0). Therefore, [Ei , ep. ] = 0 and [Ei , ep . ] = 0. Further, the vector represented
by the bottom dot in the upper-left module in Fig. 1 is singular, and the module is therefore
(the quantum deformation of) Mn (1 n), if
p . = 1.
With these conditions satisfied, applying ( (n1) ) to ep .(z) gives a level-n

descendant Pn ep.(z) of ep.(z) that necessarily commutes with the screenings.4 It follows
that p = , see (2.4). The currents
En[0] (z) = e .(z) ,
Fn[0] (z) = Pn e .(z)
(2)
(2)
then generate W n[0]
. To recover Wn[0]
, it remains to introduce an additional free scalar
field and embed the above n-dimensional vectors in Cn+1 as ai = {a i , 0}, = {, 0}, and
= { , n+1 } with isotropic , as in Section 2.1.1. We introduce the currents
Ai = (ai , ),
Q = (, )
(2.10)
(where the brackets denote Euclidean scalar product) and the scalar field
= (, )
4 More generally, we expect that for generic k, the entire centralizer of U s(n|1) in the sector with
q

momentum p is given by ( (n1) )(P ep (z) ), where P is a differential polynomial such that

[Ei , P ep ] = 0.
423
such that
Y = = (, ).
(2.11)
From the scalar product (, ) = 1, we then have the OPEs

Q(z)e (w) =
1 (w)
e
.
zw
(2)
-currents have the form
In the maximally asymmetric realization, the Wn[0]
En[0] (z) = e (z),

Fn[0] (z) = Pn (An1 , . . . , A1 , Q)e (z) ,
where Pn (An1 , . . . , A1 , Q) is the polynomial in (2.8) expressed through An1 , . . . ,
A1 , Q.
2.2.2. Other realizations
In other realizations, both the E and F currents are given by a normal-ordered product of
an exponential and a differential polynomial. We recall that in the well-known symmetric

bosonization of s(2)
(see Appendix A.2), the E and F currents follow by the action of the
corresponding fermionic screening and therefore involve order-1 polynomials in front of
the exponentials, see (A.3). In the general case, the simplest singular vectors given by the
action of a single Uq s(2|1) generator are replaced with singular vectors in the appropriate
Uq s(n|1) modules.
A step-back strategy similar to the one used in the maximally asymmetric case also
involves 2n -dimensional Uq s(n|1)-modules; the classical s(n|1)-analogue of every such
module can be viewed as an Mn (1 n) module turned on its side. This is illustrated in
Fig. 2 for n = 4. If the module is viewed as generated from the top vector, it is M4 (3).
Fig. 2. A matrioshka arrangement of special s(n|1)-modules (left). Considered as generated from the top
vector by the simple root generators corresponding to the Dynkin diagram (a), the 24 states represent the
s(4|1)-module M4 (3). The diamonds show the s(3|1)-module M3 (2), as in the left part of Fig. 1. The
filled dot is a singular vector in both M4 (3) and M3 (2). The bigger diamonds (together with , which is again
a singular vector) form the weight diagram of the s(2|1)-module M2 (1). With the simple root generators
corresponding to the Dynkin diagram (b), the 24 -dimensional s(4|1) module is generated from the state s.
424
In this case, the simple root generators correspond to the Dynkin diagram (a). But with
the simple root generators corresponding to the Dynkin diagram (b), the 24 -dimensional
s(4|1) module is generated from the vector s east-south-east of the top one. The vector
at the bottom in the M4 (3) coordinates is then again singular and generates a onedimensional submodule. Moreover, restricting to the s(3|1) subalgebra (corresponding
to the first three nodes in diagram (b)), we see that the s(3|1)-module generated from s
(shown with diamonds in Fig. 2) is M3 (2), as in the left-hand part of Fig. 1. The filled
dot is also a singular vector in M3 (2), with a one-dimensional submodule (it is also clear
how the s(2|1)-module M2 (1) fits this picture (bigger diamonds)). Construction of the
singular vector from the s state, actually via the s(3|1) generators, then translates into a
formula for the F current in the centralizer of the screenings.
Moreover, the E current in the centralizer of the screenings then follows by dressing the
corresponding exponential with a single simple root generator; in Fig. 2, the corresponding
module should be viewed as generated from the state s , at distance 1 from the singular
vector with a 1-dimensional submodule.
This hierarchy of the special s(n|1)-modules under consideration (in fact, of their
Uq s(n|1)-deformations) shows that singular vectors generating one-dimensional submodules can be constructed via the same step-back mechanism as in the maximally asymmetric case described above, but applied to a subset of simple root generators. The singular
vector is in a sense the same for the different root systems in the s(n|1) algebras with
different n, only constructed from differently chosen highest-weight vectors by the appropriate set of simple root generators. Accordingly, the E and F currents in the centralizer
of the screenings are then obtained by dressing the corresponding exponential with complementary sets of the simple root generators, Ei , i = 1, . . . , n m 1, and + for Fn[m] ,
and Ei , i = 1, . . . , m + 1, and for En[m] , and therefore,
En[m] (z) = Pm (A1 , . . . , Am+1 , Q )e (z),
Fn[m] (z) = ()m+1 Pnm (Anm1 , . . . , A1 , Q+ )e (z)
with differential polynomials Pm and Pnm of the respective order m and n m (normal
ordering in the right-hand sides is understood).
2.3. Centralizer of Uq s(n|1): recursion relations
Parallel to the matrioshka arrangement of the s(n|1) modules described in the previous subsection, there exist recursion relations between the differential polynomials entering
the different realizations of the Wn(2) algebras with different n. In considering these, we
must be very precise about notation.
2.3.1. Notational chores
(k)
(2)
We write Xn[m] for generators Xn[m] = (En[m] , Hn[m] , Fn[m] ) of Wn[m] (k) whenever we
need to indicate the level k of the algebra. The differential polynomials Pn[m]
and Pn[m] in
Ai and Q entering these generators depend on k explicitly, which we indicate by writing
(k)
(k)
them as Pn[m] and Pn[m] . Moreover, the free fields introduced in Section 2.1.5 also bear an
implicit dependence on k and n, in fact on k + n involved in their OPEs (where we recall
425
that K = k + n). When we need to be very precise, we write these free fields as Ai[k+n] and
[k+n]
, and similarly use the notation Ei[k+n] and [k+n] for the screenings. In the n[m]
Q
(2)
realization of Wn (k), we can thus write the generators, most generally, as (see (2.10) and
(2.11) for the definition of the currents)
(k)
(k) [k+n]
[k+n]
[k+n] (z)
(z) = Pn[m]
, Q+
, Q
,
En[m]
A
e

(z)
(k)
(k)
[k+n]
[k+n]
,
e
Fn[m] (z) = (1)m+1 Pn[m] A[k+n] , Q+ , Q
(k)
(2.12)
(k)
where Pn[m] and Pn[m] are differential polynomials of the respective degrees m and n m
and A stands for the appropriate collection of the Ai currents (see Section 2.1.5). The
conventional sign factor is chosen for future convenience. By definition, Pn[0]

= Pn[n] = 1.
For m = 0 (the maximally asymmetric realization), Q does not enter and Q+ is identified
with Q in Section 2.1.1.

Example 2.3.2. From the well-known three-boson realizations of s(2),
see Appendix A.2,
we have that the lowest P polynomials are given by
(k)
(k)
P2[1] (Q) = P2[1] (Q) = Q,

(k)
P2[0] (A, Q) = AQ + QQ + (k + 1)Q.
(2.13)
2.3.3. Field identifications

(2)
It follows that the free fields entering a realization of the Wn1 (k + 1) algebra can be
identified with a subset of the fields involved in a realization of Wn(2)(k). We can therefore
consider the universal sets of the currents
[]
[]
[]
[]
. . . , A2[] , A1[] , Q+
, Q
, A1
, A2
,...
and the screenings

[]
[]
, E2
,...,
. . . , E2[] , E1[] , +[] , [] , E1
(2)
. Any such subset is a length-n
and use their appropriate finite subsets to define the Wn[m]
5
segment including + or (or both).
Next, the scalar products of with the other vectors (see Section 2.1.5), and hence the
OPEs of Y = (, ) with the other currents are independent of k or n; this allows us to
identify the and Y fields in all the realizations of all Wn(2) .
(2)
We now find the dimension-1 current in each Wn[m] .
5 The case where only , but not is included is the maximally asymmetric realization with m = 0. The
+
case where only is included, but + is not, is the equally asymmetric, opposite realization with m = n.
We do not consider it specially because it can be obtained from the m = 0 realization by the automorphism
exchanging E and F .
426
(2)
Lemma 2.3.4. For 0 m n, n 2, the diagonal current of Wn[m] is given by

(k)
(z) = n (k)Y (z) +
Hn[m]
nm1

i=1
i=m+1
n i m [k+n]
n m [k+n]
Ai
Q+ (z)
(z) +
n
n
m [k+n]
m + i [k+n]
(z) Q
(z).
Ai
n
n
Proof. This is shown by directly finding the centralizer of the relevant screening operators
in the space of dimension-1 operators with zero momentumi.e., among descendants of
the identity operator, and hence necessarily linear combinations of the currents; the coefficients are then determined by a straightforward calculation, uniquely up to normalization,
and the overall normalization is fixed by (1.2). 2
Proceeding similarly, we establish the existence of a Virasoro algebra in the centralizer
of the screenings:
Lemma 2.3.5. For generic k, the centralizer of the screenings Ei , i = 1, . . . , n 1, and
(see (2.3)) in the space of dimension-2 operators that are descendants of the identity
is three-dimensional. In addition to HH(z) and H(z), it is generated by the energy
momentum tensor with central charge (1.1), explicitly given by

1
n (k)1 i,j Ai[k+n] Aj[k+n] (z)
Tn(k) (z) =
2
i,j {n1,n2,...,1,+,}
n1

(i 1)(k + n 1) 1 [k+n]
n [k+n]
(n i)
(z) A+
(z),
Ai
2(k + n)
2
i=1
where we write A+ Q+ Q and A Y for notational uniformity and where n (k) is

the dressed Cartan matrix (2.5).
(2)
(k) and
Lemma 2.3.4 implies a relation between the H currents in Wn[0]
(2)
W(n1)[0] (k + 1):
(k)
(k+1)
(n 1)H(n1)[0]
= (k + n 1)Y +
nHn[0]
n1

Ai[k+n] + Q[k+n] .
(2.14)
i=1
Somewhat less obviously, we have similar recursion relations for the E and F currents
(2)
.
generating Wn[0]
(k+1)
(k+1)
(k+1)
(k+1)
Lemma 2.3.6. Let E(n1)[0]
= e , H(n1)[0]
, and F(n1)[0]
= Pn1
e
be the genera(2)
tors of W(n1)[0] (k + 1), n 2. Then

(k)
(k)
(k+1) (k+1)
Fn[0]
= (k + n 1) + nHn[0]
(n 1)Hn1
F(n1)[0] ,
n 2,
427
(2)
is in the centralizer of the screenings (Ei )i=1,...,n1 and , i.e., in Wn[0] (k). In other words,
(k)
= Pn(k) e can be constructed recursively, via
Fn[0]

n1

(k+1)
[k+n]
(k)
[k+n]
Pn1
Pn = (k + n 1) + Q
+
Ai
i=1
with the initial condition
P1(k) (Q) = Q.
In accordance with a remark in Section 2.3.1, this recursion relies on identification

(2)
(k + 1) with the
of the fields An2 , . . . , A1 , Q involved in the construction of W(n1)[0]
corresponding fields among the An1 , An2 , . . . , A1 , Q involved in the construction of
(2)
Wn[0] (k). As a corollary, we note that the differential polynomial Pn depends only on
An1 , . . . , A1 , Q.
Proof of Lemma 2.3.6. The second formula in Lemma 2.3.6 is equivalent to the first one
(k)
constructed via the recursion is in the centralizer
in view of (2.14). We now show that Fn[0]
of the screenings (2.3). We assume that this is so for the preceding generators, in particular,
[k+n] (k+1)
, F(n1)[0] = 0, i = 1, . . . , n 2,
Ei
[k+n] (k+1)
, F(n1)[0] = 0,
or, in terms of OPEs for i = 1, . . . , n 2,

(k+1) [k+n]
e(ai ,(z)) Pn1 An2
, . . . , A1[k+n] , Q[k+n] (w) =
0
+ ,
(2.15)
zw
where the dots denote possible other poles (only the vanishing of the indicated pole is
essential), and

(k+1) [k+n]
e(,(z)) Pn1
(2.16)
An2 , . . . , A1[k+n] , Q[k+n] (w) = 0 (z w)0 +
(that is, the normal-ordered product vanishes).6 To be precise, we should have written
[k+n] in the left-hand sides.
We must show that relations (2.15) and (2.16) imply the OPEs
[k+n]

, . . . , A1[k+n] , Q[k+n] (w) =
e(ai ,(z)) Pn(k) An1
i = 1, . . . , n 1,
and
0
+ ,
zw
[k+n]

e(,(z)) Pn(k) An1
, . . . , A1[k+n] , Q[k+n] (w) = 0 (z w)0 + .
6 For n 1 = 1, with P (Q) = Q, the vanishing in (2.16) occurs as follows: the OPE
1
e(,(z)) Q(w) =
e(,(z))
+ :e(,) Q:(w),
zw
gives zero in the normal-ordered term after the expansion e(,(z)) = e(,(w)) + (z w)Q(w).
(2.17)
(2.18)
428
But we readily establish the OPE

n1

Ai (z) Q(w) +
Aj (w) = 0,
i = 1, . . . , n 2,
j =1
and therefore (2.15) and the recursion for P imply (2.17) for i = 1, . . . , n 2. For i =
n 1, we no longer have a similar vanishing OPE, but the only nonzero OPEs involving An1 are with itself and with An2 , see (2.7). The dependence of Pn(k) on An1 is
only through the explicit occurrences of An1 in the recursion relation, and the dependence on An2 is only through the explicit occurrences of An2 in the preceding recursion
relation. That is,

n3

(k+2)
(k+2)
(k)
Aj + Q Pn2
Pn = (k + n 1) (k + n 1)Pn2 + An2 +

+ An1 + An2 +
Aj + Q
j =1

(k
n3

j =1
(k+2)
+ n 1)Pn2 +

n3

(k+2)
An2 +
Aj + Q Pn2 ,
j =1
(k+2)
where Pn2
depends only on the fields that have zero OPEs with An1 . This shows that
(k)
the first-order pole in the OPE e(an1 ,(z)) Pn (w) vanishes; the derivation is elementary.
This proves all Eqs. (2.17).
It remains to verify (2.18), i.e., :e(,) Pn(k) : = 0. For this, we note that Q (and hence
(,)
) has nonvanishing OPEs only with Q and A1 ; in particular, it follows from Sece
tion 2.1.5 that

n1

e(,(w))
(,(z))
Q(w) +
e
.
Aj (w) = (k + n 1)
zw
j =1
(k+1)
The recursion now readily implies that :e(,) Pn(k) : = (k + n 1):e(,) Pn1
: = 0 by
assumption. 2
Example 2.3.7. For n = 3, it follows from Lemma 2.3.6 and (2.13) that
P3(k) (A2 , A1 , Q) = A1 A1 Q + A1 A2 Q + 2A1 QQ + A2 QQ + QQQ
+ (k + 2)(2A1Q + A2 Q + A1Q + 3QQ) + (k + 2)2 2 Q.
The n = 4 example is given in Appendix A.4.
(2)
(2)
Generalizing Lemma 2.3.6 from Wn[0] to Wn[m] , we have

(k+1)
(k+1)
(k+1)
(k+1)
Lemma 2.3.8. Let the currents E(n1)[m]
= P(n1)[m]
e , H(n1)[m]
, and F(n1)[m]
=
(k+1)
(2)
(k)
(1)m+1 P(n1)[m]
e be the generators of W(n1)[m]
(k + 1), n m + 2. Then Fn[m]
=
(1)m+1 Pn[m] e with

429
(k)
(k)
Pn[m]
= (k + n 1) + Q+ +
nm1

(k+1)
Ai P(n1)[m] ,
n m + 2,
i=1
(k)
and with the initial condition P(m+1)[m]
(Q) = Q is in the centralizer of the screenings
(2)
(Ei )i=1,...,nm1 , + , , (Ei )i=1,...,m+1 , i.e., in Wn[m] (k).

(k)
We note that this implies that Pn[m] depends only on Anm1 , . . . , A1 , Q+ ; anticipat
ing a similar statement for P (that the differential polynomial Pn[m]
depends only on
A1 , . . . , Am+1 , Q for m 2 (and on Q for m = 1)), we can express (2.12) much
more precisely, as
(k)
(k) [k+n]
[k+n]
[k+n] (z)
e
En[m] (z) = Pn[m] A1
, . . . , Am+1
, Q
,

(z)
(k)
(k)
[k+n]
[k+n]
[k+n]
e
, Q+
.
Fn[m] (z) = (1)m+1 Pn[m] Anm1 , . . . , A1
(2.19)
Proof of Lemma 2.3.8. The only new element compared to the proof of Lemma 2.3.6
consists in verifying the vanishing of the second-order pole in the operator product involving ,
(k)
e( ,(z)) Pn[m] (Anm1 , . . . , A1 , Q+ )(w) =
0
+ ,
(z w)2
(k+1)
assuming that this vanishing occurs for P(n1)[m] . We use the notation [A, B]n for the
coefficient at the nth-order pole in the OPE A(z)B(w) and recall the standard relations
(see, e.g., [13])
[V , P ]2 = [V , P ]1 + [V , P ]2 ,

V , [A, P ]0 2 = A, [V , P ]2 0 +
[V , A] , P 2 ,
>0
where we take V = e( ,) , A = Q+ +
nm1
i=1
(k+1)
Ai , and P = P(n1)[m]
. The assumption
is therefore that [V , P ]2 = 0. It now follows from Section 2.1.5 that V (z)A(w) =

with [V , A]1 = (k + n 1)V , and therefore

( ,) (k)
, Pn[m] 2 = V , (k + n 1)P + [A, P ]0 2
e

= (k + n 1)[V , P ]1 + [V , A]1 , P 1 = 0,
[V ,A]1
zw
(k)
(k)
e , with Pn[m]
given by the recursion
which is equivalent to the statement that Pn[m]
formula, is in the centralizer of . 2
As an immediate consequence of Lemmas 2.3.6 and 2.3.8, we have

Lemma 2.3.9. For 1 m n 1,
(k)
(k+1)
(Anm1 , . . . , A1 , Q+ ) = P(n1)[m1]
(Anm1 , . . . , A1 , Q+ ),
Pn[m]
430
and therefore
(k)
(k+m)
Pn[m] = Pnm .
This gives all the P polynomials involved in the free-field construction of the F generators. Next, the P polynomials entering the E generators are also expressed through
the Pm . This follows by reading the Dynkin diagram describing the n[m] realization from
right to left, which corresponds to the algebra automorphism interchanging E and F (and
also replacing m with n m). In the free-field realizations, this exchanges the currents
Ai with positive and negative i between each other and also exchanges the Q+ and Q
currents. As a result, Pn[m]

depends on A1 , . . . , Am+1 , Q and its actual form of a
differential polynomial in m variables is given in the following Lemma.
Lemma 2.3.10. For m = 1, . . . , n,
(k)
Pn[m] = Pm(k+nm) .
(2)
2.4. Wn[m]
OPEs
We thus see that all realizations of all the W(2) algebras are determined by a series of
degree-n differential polynomials Pn in n variables, n 1, given by the normal-ordered
expressions
Pn(k) (An1 , . . . , A1 , Q)(z)

n1

= (k + n 1) + Q(z) +

Ai (k + n 1) + Q(z) +
i=1

(k + n 1) + Q(z) + A1 (z) Q(z).
All the
currents as
n2

Ai (z)
i=1
operators are applied to the currents on the right. Linearly combining the
R0+ = Q+ ,
Ri+ = Q+ + A1 + + Ai ,
R0
Ri
i 1,
= Q ,
= Q + A1 + + Ai ,
i 1,
we readily see that their OPEs are indeed those in (1.4) and (1.5), and we obtain Theorem 1.1.
(k)
(k)
The other currents in the algebra are to be found from the OPE En[m] (z)Fn[m] (w). Calculating it, we first obtain
(k)
(k)
En[m] (z)Fn[m] (w)

= (1)m+1 e (z) (w)
431

(k + n 1)z + Rm1
(z) +

1
zw

1
1
R0 (z) +
(k + n 1)z + R1 (z) +
zw
zw

1
+
(k + n 1)w + Rnm1
(w)
zw

1
1
+
+
(k + n 1)w + R1 (w)
R0 (w)
,
zw
zw
where the OPEs with the exponentials have been taken into account and it remains to
evaluate the OPEs Ri (z)Rj+ (w) between the currents in the two (z and w) brackets. The action of the operators is delimited by the square brackets. Because the OPEs
Ri (z)Rj+ (w) are independent of i and j (see (1.4)), we obtain
(k)
(k)
En[m] (z)Fn[m] (w)

= (1)m+1 e (z) (w)

k+n1
1
(k + n 1)z + Rm1
(z) +
+
(z w)2
zw

k+n1
1
+ +
(k + n 1)z + R1 (z) +
(z w)2
zw

k
+
n
1
1
+ +
R0 (z) +
zw
(z w)2

1
+
(k + n 1)w + Rnm1
(w)
zw

1
1
+
+

(k + n 1)w + R1 (w)
R0 (w)
zw
zw
, (2.20)
+ (1)=0
where the right-hand side is normal-ordered and + is the derivation of the ring of differential polynomials such that

+ Ri+ = 1.
After all + s are evaluated in accordance with this rule, we must set + = 0, which is in
dicated by the prescription + (1) = 0. As noted above, the action of z = z
is delimited
by the first of the two groups of factors in brackets (the z-factors). But each z is involved
+
1
1
only in the combination z + (zw)
2 + , where + is then applied to Ri (w) zw . Because

+

1
1
+ Ri (w) = z
,
(z w)2
zw
(k)
(k)
(z)Fn[m]
(w) can be rewritten with all
it follows that the right-hand side of the OPE En[m]
the + dropped and with each z acting on all factors, either z or w, to the right of a
432
given one. Thus,

(k)
(k)
En[m] (z)Fn[m] (w)
= (1)m+1 e (z) (w) (k + n 1)z + Rm1

(z) +

(k + n 1)z + R1 (z) +
1
zw

1
zw

(k + n 1)z + R0 (z) +

1
zw

1
1
(k + n 1)w + R1+ (w)
R0+ (w)
.
zw
zw

+
(w)
(k + n 1)w + Rnm1
1
zw
(2.21)
In this normal-ordered expression, we expand e (z) (w) up to the terms O((z w)n1 )
and then rewrite it as
(k)
(k)
En[m]
(z)Fn[m]
(w) =
n U (k)

n[m],nj (w)
j =1
(z w)j
(k)
(k)
= n1 (n, k), the dimension-1 current Un[m],1
(w) =
This gives the central term Un[m],0
(k)
(k)
nn2 (n, k)Hn[m] (w), the energymomentum tensor related to Un[m],2 (w) as in (A.1), and
(k)
the other currents Un[m],i
(w), 3 i n 1.
Evaluating the integrals
n
(k)

(k)
dz f (z) En[m] (z)Fn[m] (w) =
j =1
1
(k)
(w)wj 1 f (w)
U
(j 1)! n[m],nj
(k)
with suitable test functions, we arrange the currents Un[m],i
(w) into the order-(n 1) differential operator
(k)
U n[m] =
n

j =1
1
(k)
U
j 1 ,
(j 1)! n[m],nj
similarly to standard cases of the quantum DrinfeldSokolov reduction. Because of the

presence of both derivatives and Cauchy kernels in the right-hand side of (2.21), the corresponding analogue of the Miura mapping is more complicated than in the classic cases.
We only note that

(k)
U (k)
V n[m] e ,
n[m] = e
where V (k)
n[m] is the order-(n 1) differential operator given by

(k)
V n[m] f (w)

(z)
= dz f (z) (k + n 1) z Rm1

(k + n 1) z R1 (z)
1
zw
433

1
zw

(k + n 1) z R0 (z)

1
zw

1
1
+
+
R0 (w)
.
(k + n 1)w + R1 (w)
zw
zw

+
(w)
(k + n 1)w + Rnm1
1
zw
Derivatives acting on f are written on the right to simplify comparison with (2.21).
Example 2.4.1. Taking n = 2 and m = 1 brings us back to the symmetric realization of

s(2).
Eq. (2.21) then becomes
(k)
(k)
(z)F2[1]
(w)
E2[1]

= e (z) (w) (k + 1)z + Q (z) +

1
1
Q+ (w)
zw
zw

Q+ Q
k
+
= 1 + (z w)Y
zw
(z w)2
k
Q+ Q + kY
=
,
+
zw
(z w)2
(k)
with Q+ Q + kY = 2H2[1]
.
Similarly, for the BP algebra (n = 3) in the maximally asymmetric realization (m = 0),
we evaluate (2.21) as
E3[0] (z)F3[0] (w)

= e (z) (w) (k + 2)w + R2+ (w)

(k + 2)w + R1+ (w)
1
zw
1
zw

R0+ (w)
1
zw

1
1
2
2
= 1 + (z w)Y (w) + (z w) Y (w) + (z w) Y Y (w)
2
2

R2+ (w) + R1+ (w) + R0+ (w)
(2k + 3)(k + 1)
+
(k
+
1)
(z w)3
(z w)2

+
+
+ +
+ +
(k + 2)(R1 + 2R0 ) + R2 R1 + R2 R0 + R1+ R0+
+
,
zw
with the rest of the calculation totally straightforward.
434

3. Wn(2) algebras from s(n|1)
(2)

to the coset s(n|1)/
s(n),
The second construction of the Wn algebras is related
(2)
n(z)
which actually becomes Wn after a correction by e
, where is an auxiliary
scalar field with the OPE (1.10). This generalizes the construction in [14] which explicitly

shows that after the appropriate correction, s(2|1)/
s(2)
is s(2).
Theorem 1.2 is formulated more specifically as follows.
Theorem 3.1. Let the level k be related to k by (1.7). Let e1 (z), . . . , en (z) and

f1 (z), . . . , fn (z) be the two s(n) n-plets of the fermionic generators of s(n|1)
k . Then
the operators
1
e1 (z)e2 (z) en (z)e n(z) ,
n/2
+ n 1)
(1)n+1
F (z) =
f1 (z)f2 (z) fn (z)e n(z)
n/2
(k + n 1)
E(z) =
(k
(2)
generate the Wn (k) algebra.

3.1. Outline of the proof
3.1.1. The second quantum group
We begin with identifying the second quantum group that commutes with Wn(2) and
with Uq s(n|1) constructed in Section 2.1.5 (see footnote 2 and the preceding text). For
this, we define

1
Sn,i = e k+n (ai ,) , i = 0,

1
Sn,0 = (aQ+ + bQ )e k+n (+ + ,) a = b, |a|2 + |b|2 = 0 .
A direct calculation shows the following lemma.
(2)
Lemma 3.1.2. The operators Sn,i , n m 1 i m + 1, commute with the Wn[m]
algebra and furnish a representation of the nilpotent subalgebra of the Uq s(n) quantum
group with q = ei/(k+n) .
For m = 0 (the maximally asymmetric realization), the operators in the lemma are Sn,i ,
i = n 1, . . . , 1. In each of the more symmetric realizations, which involve Sn,0 , the
arbitrariness in a and b is due to the possibility of adding a total derivative to the integrand and of choosing the overall normalization; a certain pair (a, b) is to be fixed in what
follows.
Remark 3.1.3. In the notation Uq s(n|1) used in the previous sections, q was part of the
general notation for quantum groups. We now have quantum groups with different values
of the deformation parameter, and we keep the notation q for the deformation parameter
435
(2)
of s(n|1); its value is q = ei(k+n) , expressed through the k parameter in Wn[m] (k) (for
i
all m). The quantum deformation of s(n) in Lemma 3.1.2 is with the parameter q = e k+n ,
and we therefore use the notation Uq s(n) for the corresponding quantum group.
For each fixed n and m, we next construct vertices that are highest-weight representations of the Uq s(n) quantum group generated by the Sn,i . Let
Vn,m (z) = e(vn,m (k),(z)),
where
1
vn,m (k) =
n(k + n)
nm1

j =m+1
j =0
(3.1)

1
(m + j )aj + m+ + m + .
n
The behavior of Vn,m (z) under Uq s(n) is described as follows. We let S . V (z) denote the
dressing of a vertex V (z) by a screening operator S = du s(u). It is given by the adjoint
action of S and can be represented as the integral du s(u)V (z) taken along the contour
running below the real axis from minus infinity to the vicinity of z, encompassing z, and
returning to minus infinity above the real axis.
Lemma 3.1.4. The vertex operator Vn,m (z) satisfies the relations
Sn,i . Vn,m (z) = 0,
n m 2 i m + 1
and
Sn,nm1 . Sn,nm1 . Vn,m (z) = 0,
but
Sn,nm1 . Vn,m (z) = 0,
where Sn,i are the screenings in Lemma 3.1.2. Therefore, Vn,m (z) generates the vector
representation of Uq s(n) (the deformation of the vector representation of s(n)).
Let Cnq denote the n-dimensional Uq s(n)-representation generated from Vn,m (z).
(2)
We next consider the properties of Vn,m (z) with respect to the Wn[m]
algebra. First,
because (, vn,m (k)) = 0, Vn,m (z) is local with respect to En[m] and Fn[m] . Moreover, it is
actually primary with respect to En[m] and Fn[m] . To formulate this, we consider the state
|Vn,m
corresponding to Vn,m and introduce the modes of En[m] and Fn[m] as

n
En[m], z 2 ,
En[m] (z) =
Z+ 21 n
Fn[m] (z) =
Z+ 21 n
Fn[m], z 2 ,
436
where n = n mod 2, and the modes of H and T in the standard way, as

Hn[m] (z) =
Hn[m], z1 ,
Tn[m] (z) =
Z
Ln[m], z2 .
Z
Lemma 3.1.5. We have

n
+ 1,
2
n
Fn[m], |Vn,m
= 0, + 2,
2
En[m], |Vn,m
= 0,
(3.2)
and
1
|Vn,m
,
n

n2 1
n
|Vn,m
.
Ln[m],0 |Vn,m
= 1 +
2 2n(k + n)
Hn[m],0 |Vn,m
=
(3.3)

For n = 2, for example, this gives the standard highest-weight conditions for s(2)
highestweight vectors, E0 |
= 0 and F1 |
= 0.
Proof. The annihilation conditions in the lemma are shown by directly calculating the
corresponding OPEs. We recall that En[m] (z) and Fn[m] (z) involve differential polynomials
of the respective degrees m and n m; a priori, such polynomials develop poles of the
respective orders m and n m in the OPE with a vertex operator. But the explicit (factored)
form of the differential polynomials above readily implies the OPEs
(k)
En[m]
(z)Vn,m (w) = 0,
(k)
Fn[m] (z)Vn,m (w) =
Fn,m
,
zw
which are equivalent to the annihilation conditions in the lemma.7 The other formulas are
established immediately. 2
For future use, we also note that the length squared of the momentum of Vn,m is given
by

n1
vn,m (k), vn,m (k) =
.
n(k + n)
(3.4)
(k+1)
7 Although we do not need this in what follows, we note that F
n,m (z) = :F(n1)[m] Vn,m :(z), the normal(2)
ordered product involving the F generator of the preceding, W(n1)[m] algebra, which is realized in the same
free-field space as discussed in Section 2.
437
3.1.6. A dual vertex

Similarly to (3.1), we define the vertex operator that carries the dual vector representa(2)
tion of Uq s(n) and is at the same time a twisted Wn[m]
-primary. The results corresponding
to the previous two lemmas are as follows.
For the vertex operator
Vn,m
(z) = e(vn,m (k),(z))
(3.5)
with
vn,m
(k) =
1
n(k + n)
n1

j =1
j =nm
1
j anmj + (n m)+ + (n m) ,
n
it follows that
Sn,i . Vn,m
(z) = 0,
n m 1 i m + 2
and
Sn,m+1 . Sn,m+1 . Vn,m

(z) = 0,
but
(z) = 0,
Sn,m+1 . Vn,m
(z) generates the dual vector representation of the quantum group
showing that Vn,m
(z).
Uq s(n). We write C nq for the n-dimensional Uq s(n)-module generated from Vn,m
is a twisted W (2) primary, namely
Further, Vn,m
n[m]

En[m], Vn,m
= 0,
n
+ 2,
2

n

Fn[m], Vn,m = 0, + 1
2
(3.6)
for the corresponding state. The last formulas are a reformulation of the OPEs
(w) =
En[m] (z)Vn,m
(k)
E n,m
,
zw
(k)
Fn[m]
(z)Vn,m
(w) = 0.
k vertices and quantum-group duality

3.1.7. s(n)

We next introduce the s(n)
algebra of the level k related to k by (1.8) and the corresponding quantum group Uq s(n). The vector representation of Uq s(n) can then be
coupled with the vector representation of Uq s(n), and similarly for the dual vector representation. Because we consider generic k, we also have generic k determined from (1.8),

and hence generic q = ei/(k+n) and q = ei/(k +n) such that qq
= 1.
438

For the s(n)
algebra of level k , let V[1 ,k ] (z) denote the vertex operator corresponding
to the vector representation weight 1 . We have
1
1 1 = 1 ,
n
where is half the sum of positive roots, and hence the conformal dimension
2 1 = n 1,
[1 ,k ] =
1 (1 + 2)
2(k + n)
of V[1 ,k ] (z) is given by

[1 ,k ] =
n2 1
.
2n(k + n)
(3.7)
The vertex operator V[1 ,k ] (z) has n components, which make a basis in the vector repre k ). We use the notation Cn (z) for
sentation of s(n) (the horizontal subalgebra in s(n)
1

this space, which is the evaluation representation of s(n)
(with the subscript 1 intended
to distinguish it from other copies of Cn ). In the Wakimoto bosonization [15] (also see [16]
and references therein), the highest-weight vector in Cn1 (z) is given by
V[(0)1 ,k ] (z) = e
k +n
1 (z)
The length squared of its momentum is

n1
1 1
=
.

k +n
n(k + n)
(3.8)
Next, V[1 ,k ] (z) carries a representation of the Uq s(n) quantum group, which is the
n-dimensional Uq s(n)-module Cnq given by the quotient of the Verma module over n 1
singular vectors. The vertex V[1 ,k ] (z) can therefore be represented as
Cn1 (z) Cnq .
(3.9)
We also introduce the vertex operator V[n1 ,k ] (z) = V[

(z) associated with the dual
1 ,k ]
vector representation of s(n), given by
,
C 1 (z) C
q
n
(3.10)
with the factors dual to the respective factors in (3.9). The dimension and the momentum
length squared of the lowest-weight component coincide with those in (3.7) and (3.8).
It is useful to consider the nonlocal algebra A[s(n)k ] of vertex operators generated
n in (3.9) and (3.10). It contains Us(n),

by Cn1 (z) Cnq and C n1 (z) C
and therefore
q

carries the adjoint action of Us(n)
(in particular, the center acts trivially); in addition,

it carries an action of Uq s(n), and for generic k , Us(n)
is the space of Uq s(n)invariants in A[s(n)k ].
(2)
(2)
For Wn[m] , a similar nonlocal algebra A[Wn[m] (k)] is generated by the vertex operators
Vn,m (z) = C(z) Cnq ,
n
Vn,m
(z) = C (z) C
q
439
(2)
constructed in Section 3.1.1. It also contains Wn[m] (the En[m] (z) and Fn[m] (z) currents

are identified in the quantum deformations of the respective products n Vn,m (z) and
n
(2)
(2)
Vn,m (z)). In addition, A[Wn[m]
(k)] carries the action of Uq s(n), and Wn[m]
is the
centralizer of Uq s(n) for generic k.
3.1.8. An almost local subalgebra
(2)
(k)] A[s(n)k ], with k and k related by (1.8), we now identify an alIn A[Wn[m]
most local subalgebra L n[m],k by coupling the special subspaces in A[W (2) (k)] and
n[m]
(z) with the

A[s(n)k ]. That is, we couple the Wn[m] vertex operators Vn,m (z) and Vn,m
k vertex operators V[ ,k ] (z) and V (z) as follows.
s(n)
1
[1 ,k ]
Let Rq and Rq be the R-matrices
(2)
Rq : Cnq Cnq Cnq Cqn ,

Rq : Cnq Cnq Cnq Cnq
for the vector representations of Uq s(n) and Uq s(n), respectively. We consider the
R-matrix

R = (Rq )13 (Rq )24 : Cnq Cnq Cqn Cnq Cqn Cnq Cnq Cnq
and recall that (1.8) implies that qq
= 1. Then the tensor product Cnq Cnq contains a 1dimensional subspace In that is invariant under R2 = R12 R21 (and in fact, also under R).
2i
The eigenvalue of the thus understood R2 operator on this subspace is e n , which we
write as
R2 : In In e
2i
n
In In .
(3.11)
The dual space also contains an invariant 1-dimensional subspace,
R2 : I n I n e
2i
n
In In
(3.12)
(with R2 understood appropriately), and moreover,
R2 : In I n e
2i
n
In I n .
(3.13)
Cnq
We next use the embedding In

the tensor product of the vertex operators
Cnq to identify an n-dimensional

(z),
Vn,m (z) and Vn,m
E(z)
Cn (z)
= Cn (z) In C(z) Cnq Cn1 (z) Cnq

Vn,m (z) V[1 ,k ] (z)
subspace in
(3.14)
and similarly with the dual spaces,
n C n (z) C
n .
F (z) C n (z)
= C n (z) In C (z) C
q
q

1

(z) V
Vn,m
[1 ,k ] (z)
(3.15)
440
In view of the eigenvalues in (3.11)(3.13), which become the monodromies of the vertex
operators constituting Cn (z) and C n (z), these operators are almost local: the mon2i
odromies of the components of E(z)

and F (z) with respect to each other are e n ; the
operator products between these components are
(z)E
(w) = (z w) n L++ (z, w),
E
F (z)F (w) = (z w) n L
(z, w),
1
(z)F (w) = (z w) n L+ (z, w),

E
where L
(z, w) are Laurent series in (z w). The almost local subalgebra Ln[m],k is
and F .
the algebra generated by E
3.1.9. A scalar-field correction
We now modify the vertex operators E(z)

and F (z) defined in (3.14) and (3.15) to
obtain fermionic currents.
We introduce an auxiliary scalar current f with the operator product
f (z)f (w) =
1
(z w)2
(3.16)
and consider the 1-dimensional lattice algebra F 1 generated by e
1n f (z)
. It then fol-
(2)
lows that A[Wn[m] (k)] A[s(n)k ] F 1 contains the local subalgebra generated by the
n
operators
1
f (z)
n
E(z) = E(z)e
,
F (z) = F (z)e
1n f (z)
We note that the dimension of the operators in each of the n-plets given by E(z)
and
F (z) is the sum of [1 ,k ] in (3.7) and the dimension n (k) of Vn,m read off from (3.3).
Because of (1.8), we have
[1 ,k ] + n (k) = 1
1
.
2n
(3.17)
1
Therefore, the vertex operators E(z) and F (z) have dimension 1 (given by 1 2n
in (3.17)
1 f
1
for e n ). Similarly, in a bosonization where (the highest-weight component of)
plus 2n
the vertex operator Cn1 (z) in the right-hand side of (3.14) is a pure exponential, the length
squared of its momentum is given by adding (3.4) and (3.8),
n1
n1
1
+
=1 .

n(k + n) n(k + n)
n
(3.18)
Therefore, the momentum length squared of E(z) and F (z) in a bosonized representation
is 1 (given by 1
1
n
in (3.18) plus
1
n
for e
1n f
).
441
To summarize, E(z) and F (z) are given by n fermionic dimension-1 currents each,
which are local with respect to each other. E(z) is in the vector representation of s(n) and
F (z) is in the dual representation. These 2n dimension-1 fermionic currents are the two

n-plets of the s(n|1)
fermionic currents, and we thus obtain a vertex-operator extension
(2)

of Wn[m] (k) Us(n)k ,
L n[m],k F 1
(2)
k
Wn[m]
(k) Us(n)

Us(n|1)
k

The diagram means that by coupling vertex operators of Wn(2) and s(n)
(and the auxil 1n f

iary e
), we obtain an algebra that contains s(n|1)k .
This vertex-operator extension inverts the coset (1.9). This means, in particular, that
k in the left-hand part of the above diagram is the subalgebra of s(n|1)

s(n)
k in the right,
and we actually have the diagram (dropping the redundant [m])
L n,k F 1
k F 1
Wn (k) Us(n)

Us(n|1)
k h
(2)
k h +
Wn (k) Us(n)

Us(n|1)
k G n
(2)
The mappings and algebras are here as follows. The Heisenberg algebra h + is generated
1 f (z)
by the current f (z), see (3.16). By e n

, this Heisenberg algebra is extended to the
(nonlocal) lattice algebra F 1 . The Heisenberg algebra h is generated by the current
n
(z), see (1.10) ((z) can be represented as a unique linear combination of the
Cartan
n(z) ,

k , the Hn current, and f that commutes with s(n|1)
currents of s(n)
k ). By e
k alit is extended to a one-dimensional lattice vertex-operator algebra Gn . The s(n)

gebra in the left-hand part is a subalgebra in s(n|1)
k . The algebra Us(n|1)k G n
contains the elements e1 (z) en (z) e n(z) and f1 (z) fn (z) e n(z) whose images in L n,k F 1 coincide with the respective images of the E(z) and F (z) currents
n
(2)
of Wn ; this describes the horizontal arrow.

3.2. Lifting the extension
This vertex-operator extension can be viewed as resulting from a Hamiltonian reduction

of another vertex-operator extension, namely of two s(n)
algebras with the dual levels
k and s(n)
k vertex operators corresponding
related by (1.8). There, we take the s(n)
to the vector representations, V[1 ,k] and V[1 ,k ] , and similarly with the dual operators.
k , we view V[1 ,k] (z) as Cn (z) Cn ; its properties with
Repeating Section 3.1.7 for s(n)
q
442
(2)
respect to the Uq s(n) quantum group are the same as for the Vn,m operators of Wn[m] .
To distinguish between different Cn spaces, we now use the notation Cn1 (z) for V[1 ,k] (z)
and Cn (z) for V[1 ,k ] (z) (which had no prime in the previous subsection, where we had
1
fewer Cn spaces). From (3.11), we again have a 1-dimensional subspace in Cnq Cnq that is
invariant under the appropriately squared R-matrix (and similarly for the dual), and hence
there are n2 -dimensional subspaces embedded as
Cn (z) C n (z) Cn1 (z) Cnq C n (z) Cnq ,

1
C (z) C (z)
n
n
n
C 1 (z) C
q
n .
Cn (z) C
q
1
k generate a nonlocal algebra

k and s(n)
The thus coupled vertex operators of s(n)
k Us(n)
k . Similarly to L n,k , N is almost local; after a free-field
N Us(n)
correction as above, we obtain a vertex-operator algebra Xn (k) with 2n2 fermionic cur
rents. Its Hamiltonian reduction then gives s(n|1);
under this reduction, the Cn1 (z) and

are reduced, respectively, to the highest-weight vector
C n1 (z) vertex operators of s(n)
v+ (z) C(z) and the lowest-weight vector v (z) C n (z), which are then identified with
the respective Wn(2) vertex operators (3.1) and (3.5). We thus have
N F 1
k Us(n)
k
Us(n)
Ham. red.
k
Wn (k) Us(n)
(2)
Xn (k)
L n,k F 1
Ham. red.

s(n|1)
k
The mappings into the algebra at the top are actually into its local subalgebras. For n = 2,
) [14], and the somewhat mysteXn (k) is the exceptional affine Lie superalgebra D(2|1;
rious algebras Xn (k) must therefore give its higher (W -)analogues.
4. Conclusions
For W -algebras whose defining set of screenings involves fermionic screenings, local
fields can be explicitly constructed by the method used in this paper, based on the properties
of quantum supergroups. Here and in [12], the simplest applications of this technique have
been given; we hope that the method can be applied in much greater generality.
(2)
For the Wn algebras constructed in this paper, it is interesting to consider integrable
representations, which have many features in common with the integrable representations

of s(2).
The E(z) and F (z) currents then satisfy a number of relations generalizing

representations. The corE(z)k+1 = 0 and F (z)k+1 = 0 satisfied in the integrable s(2)
responding higher relations are discussed in [3]. Resolutions of the butterfly type [17] are
443
expected to give further insight into the structure of these representations. It would also be
interesting to study semi-infinite realizations of such representations, as in [5,6], see [3,18,
19].
The method of vertex-operator extensions is a powerful tool in constructing infinitedimensional algebras and establishing relations between different algebras; a number of
known examples are based on duality relations of the type of Eq. (1.8) (possibly with
another integer in the right-hand side). We note that the construction of unifying W algebras [4], briefly recalled in the Introduction, involves an even more general relation
1
1
+ k +m
= 1, and is worth being reconsidered from the standpoint
between the levels, k+n
of vertex-operator extensions.
Acknowledgements
We thank I. Cherednik and D. (M.) Lebedev for a discussion. This work was supported
in part by the INTAS (Grant 00-00262), the RFBR Grants 02-02-16946 and 02-01-00930
and by the Grant LSS-1578.2003.2; AMS was also supported by the Foundation for Support of Russian Science.
(2)
Appendix A. The lower Wn algebras

The basic operator product for Wn(2) is given by
En (z)Fn (w)
n1 (n, k) nn2 (n, k)Hn (w)
=
+
(z w)n
(z w)n1
+ n3 (n, k)
n(n1)
n((n2)(k+n1)1)
Hn (w) (k
2 Hn Hn (w) +
2
n2
(z w)

Wn,3 (w) (k + n)( 12 Tn (w) +
+ n3 (n, k)
(z w)n3
+ n2 (n, k)
+ ,
+ n)Tn (w)
1
n (k) Hn Tn (w))
n
H H H (w) + 2nn(k) Hn Hn (w) + n6 2 Hn (w)
6n (k)2 n n n
(z w)n3
(A.1)
where
m (n, k) =
m

i(k + n 1) 1 ,
(A.2)
i=1
n (k) is defined in (1.2), Tn = Tn 2n1(k) Hn Hn , Wn,3 is a Virasoro primary dimension3 operator with regular OPE with Hn , and the dots denote lower-order poles involving
operators of dimensions 4. In what follows, we write the other operator products for the
(2)
lower Wn algebras and recall the realizations of these algebras obtained in accordance
with Sections 2.3 and 2.4.
444
(2)
A.1. The W1 algebra

For completeness, we note that W1(2) is merely a systemtwo bosons with the firstorder pole in their OPE. The system is well known to be singled out as the centralizer
of one fermionic screening in the two-boson system and to have two realizations
bosonizations with either or given by an exponential times a current. In accordance
with the recursion established in Section 2.3, this (current) (exponential) construct is to
(2)
be encountered in the symmetric realization of the next algebra, W2 . This is indeed the

case: the symmetric realization of s(2)
can be obtained by rebosonizing the Wakimoto
representation.

A.2. The s(2)
algebra
(2)
k algebra is singled out from a three-boson system as the
For n = 2, the W2 (k) = s(2)
centralizer of either one bosonic and one fermionic or two fermionic screenings. This gives

the familiar asymmetric and symmetric three-boson realizations of s(2).
In the asymmet(2)
ric realization (which is W2[0] in the nomenclature in this paper), the two currents that
generate the algebra are given by
E2[0] = e ,

F2[0] = A1 Q + QQ + (k + 1)Q e
(see the OPEs at the end of Section 2.1.3, where we now set n = 2). In the symmetric

realization, the s(2)
currents are given by
E2[1] = Q e ,
F2[1] = Q+ e
(A.3)
(only the two open dots remain in the rigged Dynkin diagram (2.7)).
These well-known formulas can be easily obtained by directly finding the centralizer
of the corresponding screenings (one fermionic and one bosonic in the first case and two
fermionic in the second case).
A.3. The BP algebra
(2)
By definition [1,2], the BP W -algebra W3 (k) is a partial Hamiltonian reduction

k affine algebra, also see [9]. The BP algebra is generated by the energy
of the s(3)
momentum tensor T , two bosonic spin- 23 generators E and F , and a scalar current H. In
the operator-product form, the algebra relations are written as (with X + = E and X = F
for compactness)
E(z)F (w) =
(k + 1)(2k + 3) 3(k + 1)H(w)

+
(z w)3
(z w)2
3HH (k + 3)T + 32 (k + 1)H
,
+
zw
H(z)X (w) =
445
X
,
zw
3
2 X (w)
(z w)2
X
,
zw
H(w)
H
T (z)H(w) =
,
+
(z w)2 z w
T (z)X (w) =
T (z)T (w) =
H(z)H(w) =
1
2 c3 (k)
(z w)4
2T (w)
T
,
+
(z w)2 z w
1
3 (2k + 3)
(z w)2
(with regular terms omitted in operator products and with composite operators (HH) understood to be given by normal-ordered products).
(2)
The realizations of W3 are the maximally asymmetric and the more symmetric ones
(and two more realizations obtained from these two via the automorphism exchanging E
and F ). In the maximally asymmetric realization,
(k)
F3[0]
= P3(k) (A2 , A1 , Q)e ,
(k)
E3[0]
= e ,
(k)
with P3 given in Example 2.3.7. In the more symmetric realization, the generators are
given by
(k)
F3[1]
= P2(k+1) (A1 , Q+ )e ,
(k)
E3[1]
= Q e ,
(k)
where P2 is given in (2.13). The OPEs between the currents involved here are specified
in Section 2.1.5.
(2)
A.4. The W4 algebra

A.4.1. Realizations
(2)
The realizations of W4 , described in Sections 2.3 and 2.4, are as follows. In the totally
symmetric, 4[2], realization, the generators are given by
(k)
(k+2)
E4[2] = P2
(A1 , Q )e ,
(k)
(k+2)
F4[2] = P2
(A1 , Q+ )e
(k)
with P2 (A, Q) = AQ + QQ + (k + 1)Q (see Example 2.3.2). In the 4[1] realization,

the generators are
(k)
= Q e ,
E4[1]
(k)
F4[1]
= P3(k+1) (A2 , A1 , Q+ )e ,
where P3 is given in Section (2.3.7). Finally, in the maximally asymmetric realization, the
generators are
(k)
= e ,
E4[0]
(k)
F4[0]
= P4(k) (A3 , A2 , A1 , Q)e ,
446
where we find from Lemma 2.3.6 that

P4(k) (A3 , A2 , A1 , Q)
= A1 A1 A1 Q + 2A1 A1 A2 Q + A1 A1 A3 Q + 3A1 A1 QQ + A1 A2 A2 Q
+ A1 A2 A3 Q + 4A1A2 QQ + 2A1A3 QQ + 3A1QQQ + A2 A2 QQ
+ A2 A3 QQ + 2A2 QQQ+A3 QQQ + QQQQ + (k + 3)(3A1A1 Q
+ 4A1 A2 Q + 2A1 A3 Q + A1 A2 Q + 9A1 QQ + A2 A2 Q + A2 A3 Q
+ 6A2 QQ + 3A3QQ + 3A1A1 Q + 2A1 A2 Q + A1 A3 Q + 3A1 QQ

+ A2 QQ + 6QQQ) + (k + 3)2 3QQ + A2 Q + 2 A1 Q + 3A1Q

+ A3 2 Q + 2A2 2 Q + 4 2 QQ + 3A1 2 Q + (k + 3)3 3 Q.
The remaining 4[3] and 4[4] realizations follow from 4[1] and 4[0] by the automorphism
exchanging E and F .
A.4.2. Operator product expansions
We now write the W4(2) operator products explicitly. First, Eq. (A.1) has just 4 pole
terms,
E(z)F (w)
(k + 2)(2k + 5)(3k + 8) 4(k + 2)(2k + 5)H(w)
+
=
(z w)4
(z w)3
(k + 4)T (w) + 6HH(w) + 2(2k + 5)H
+ (k + 2)
(z w)2

4(k + 4)
1
T H(w)
+ (k + 2) W (k + 4)T (w)
2
3k + 8
8(11k + 32)
+
HHH(w) + 6HH(w)
3(3k + 8)2

4(26 + 17k + 3k 2 ) 2
1
H(w)
,
+
3(3k + 8)
zw
where T is an energymomentum tensor with the central charge c4 (k) and W is a
dimension-3 Virasoro primary whose operator product with H is regular. In addition to
this OPE and those in (1.2), we have, writing X + = E and X = F for compactness, the
operator products with W given by
W(z)X (w)
2(k + 4)(3k + 7)(5k + 16) X (w)
=
(3k + 8)2
(z w)3
+
(k+4)(5k+16)
3 (k+4)(5k+16)
2(3k+8) X (w) 6 (3k+8)2 HX (w)
(z w)2

k + 4 8(k + 3)
4(3k 2 + 15k + 16)
HX +
HX
k + 2 3k + 8
(3k + 8)2
(k + 3) 2 X

2(k + 4)
4(5k + 16)
1
T X
,
HHX
2
3k + 8
zw
(3k + 8)
and
W(z)W(w)
=
2(k + 4)(2k + 5)(3k + 7)(5k + 16)

1
3k + 8
(z w)6
(k + 4)2 (5k + 16) 3T (w)
(k + 4)2 (5k + 16) 3T (w)
3k + 8
(z w)4
2(3k + 8)
(z w)3

3(k + 4)2 (5k + 16)(12k 2 + 59k + 74) 2
T (w)
+
4(3k + 8)(20k 2 + 93k + 102)

1
8(k + 4)3 (5k + 16)
T T (w) + 4(k + 4)(w)
+
2
(3k + 8)(20k + 93k + 102)
(z w)2

(k + 4)2 (5k + 16)(12k 2 + 59k + 74) 3
+
T
6(3k + 8)(20k 2 + 93k + 102)

8(k + 4)3 (5k + 16)
1
T T + 2(k + 4)
,
+
(3k + 8)(20k 2 + 93k + 102)
zw
where T = T
2
3k+8 HH
and
(k + 2)2
= X +X
4(k + 2)
k+2
W
WH
2
3k + 8
3(k + 2)2 (k + 4)(6k 2 + 33k + 46) 2

T
2(3k + 8)(20k 2 + 93k + 102)
(k + 2)(k + 4)2 (11k + 26)

2(k + 2)(k + 4)
T T +
(T H)
2
2(3k + 8)(20k + 93k + 102)
3k + 8
8(k + 2)(k + 4)
(3k + 8)2
T HH
8(k + 2)(2k + 5) 2
HH
3(3k + 8)
16(k + 2)(2k + 5)
2(k + 2)(2k + 5)
HH
HHH
3k + 8
(3k + 8)2
32(k + 2)(2k + 5)
3(3k + 8)
1
HHHH (k + 2)(2k + 5) 3 H
6
is a dimension-4 Virasoro primary field.
447
448
Next, the operator product of W and gives rise to a single dimension-5 Virasoro
2
primary (k + 2)3 Z = 52 X + X 52 X + X + 3k+8
HX + X + ,
W(z)(w)
12(k + 4)(3k + 5)(3k + 10)(4k + 11) W(w)
=
(k + 2)(3k + 8)(20k 2 + 93k + 102) (z w)4
4(k + 4)(36k 3 + 279k 2 + 695k + 550) W(w)
(k + 2)(3k + 8)(20k 2 + 93k + 102) (z w)3

+ (k + 4) (k + 4)Z(w)
312(k + 4)(3k + 5)(3k + 10)(4k + 11)
WT (w)
(k + 2)(3k + 8)(20k 2 + 93k + 102)(84k 2 + 349k + 262)

18(3k + 5)(3k + 10)(4k + 11)(4k 2 + 29k + 62)
2
W(w)
(k + 2)(3k + 8)(20k 2 + 93k + 102)(84k 2 + 349k + 262)

2
1
(k + 4)Z
+
(k
+
4)
(z w)2
5
30(k + 4)(3k + 5)(3k + 10)
+
WT
(k + 2)2 (3k + 8)(84k 2 + 349k + 262)
+
(3k + 5)(3k + 10)(96k 4 + 1652k 3 + 9647k 2 + 22746k + 18384) 3

W
2(k + 2)2 (3k + 8)(20k 2 + 93k + 102)(84k 2 + 349k + 262)

1
4(k + 4)(3k + 5)(3k + 10)(108k 2 + 523k + 634)
.
+
WT
2
2
2
zw
(k + 2) (3k + 8)(20k + 93k + 102)(84k + 349k + 262)
References
[1] A.M. Polyakov, Gauge transformations and diffeomorphisms, Int. J. Mod. Phys. A 5 (1990) 833.
[2] M. Bershadsky, Conformal field theories via Hamiltonian reduction, Commun. Math. Phys. 139 (1991) 71.
[3] B. Feigin, M. Jimbo, T. Miwa, Vertex operator algebra arising from the minimal series M(3, p) and monomial basis, math.QA/0012193.
[4] R. Blumenhagen, W. Eholzer, A. Honecker, K. Hornfeck, R. Huebel, Coset realization of unifying W algebras, Int. J. Mod. Phys. A 10 (1995) 23672430;
R. Blumenhagen, W. Eholzer, A. Honecker, K. Hornfeck, R. Huebel, Unifying W -algebras, Phys. Lett.
B 332 (1994) 5160.
[5] B.L. Feigin, A.V. Stoianovsky, Functional models of representations of current algebras and semi-infinite
Schubert cells, Funkcional. Anal. Prilozh. 28 (1) (1994) 68.
[6] A.M. Semikhatov, I. Yu Tipunin, B.L. Feigin, Semi-infinite realization of unitary representations of the
N = 2 algebra and related constructions, Teor. Mat. Fiz. 126 (2001) 362, Theor. Math. Phys. 126 (2001)
147.
[7] V. Gorbounov, F. Malikov, The chiral de Rham complex and positivity of the equivariant signature of the
loop space, math.AT/0205132.
[8] L.A. Borisov, Vertex algebras and mirror symmetry, Commun. Math. Phys. 215 (2001) 517557,
math.AG/9809094.
[9] J. de Boer, T. Tjin, The relation between quantum W -algebras and Lie algebras, Commun. Math. Phys. 160
(1994) 317332.
449
[10] B.L. Feigin, E. Frenkel, The family of representations of affine Lie algebras, Russ. Math. Surv. 43 (1989)
221;
B.L. Feigin, E. Frenkel, Quantization of the DrinfeldSokolov reduction, Phys. Lett. B 246 (1990) 75.
[11] P. Bouwknegt, K. Schoutens, W -symmetry in conformal field theory, Phys. Rep. 223 (1993) 183276.

[12] B.L. Feigin, A.M. Semikhatov, The s(2)
s(2)/
s(2)
coset theory as a Hamiltonian reduction
of D(2|1;
), Nucl. Phys. B 610 (2001) 489530.
[13] K. Thielemans, An algorithmic approach to operator product expansions, W-algebras and W-strings, KUL,
Leuven, hep-th/9506159.
[14] P. Bowcock, B.L. Feigin, A.M. Semikhatov, A. Taormina, s(2|1)

and D(2|1;
) as vertex operator extensions of dual affine s(2) algebras, hep-th/9907171, Commun. Math. Phys. 214 (2000) 495545.
[15] B.L. Feigin, E.V. Frenkel, Usp. Mat. Nauk 43 (1988) 227, Russ. Math. Surv. 43 (1989) 221;
B.L. Feigin, E.V. Frenkel, Commun. Math. Phys. 128 (1990) 161;
B.L. Feigin, E.V. Frenkel, Lett. Math. Phys. 19 (1990) 307.
[16] J.L. Petersen, J. Rasmussen, M. Yu, Free field realizations of 2D current algebras, screening currents and
primary fields, Nucl. Phys. B 502 (1997) 649670.
[17] B.L. Feigin, A.M. Semikhatov, Free-field resolutions of the unitary representations of the N = 2 Virasoro
algebra, II: the butterfly resolution, Teor. Mat. Fiz. 121 (1999) 244257, Theor. Math. Phys. 121 (1999)
14621472.
[18] B. Feigin, M. Jimbo, T. Miwa, E. Mukhin, A differential ideal of symmetric polynomials spanned by Jack
polynomials at = (r 1)/(k + 1), math.QA/0112127.
[19] B. Feigin, M. Jimbo, T. Miwa, E. Mukhin, Symmetric polynomials vanishing on the shifted diagonals and
Macdonald polynomials, math.QA/0209042.
On orientifolds of c = 1 orbifolds
T.P.T. Dijkstra, B. Gato-Rivera 1 , F. Riccioni 2 , A.N. Schellekens
NIKHEF, PO Box 41882, 1009 DB Amsterdam, The Netherlands
Received 17 May 2004; accepted 10 August 2004
Abstract
The aim of this paper is to study orientifolds of c = 1 conformal field theories. A systematic
analysis of the allowed orientifold projections for c = 1 orbifold conformal field theories is given.
We compare the Klein bottle amplitudes obtained at rational points with the orientifold projections
that we claim to be consistent for any value of the orbifold radius. We show that the recently obtained
Klein bottle amplitudes corresponding to exceptional modular invariants, describing bosonic string
theories at fractional square radius, are also in agreement with those orientifold projections.
PACS: 11.25.-w; 11.25.Hf
1. Introduction
The c = 1 orbifold CFTs on closed, oriented Riemann surfaces have been studied extensively [14] because they provide simple but non-trivial examples of various features
of conformal field theory. It has long been believed that all c = 1 bosonic theories in the
closed oriented case were known: they either belonged to the famous continuous moduli
space of the circle and its Z2 orbifold, or one of three discrete points discovered in [3].
The completeness proof [4] was based on certain assumptions, and more recently counE-mail address: f.riccioni@damtp.cam.ac.uk (F. Riccioni).
1 Permanent address: Instituto de Matemticas y Fsica Fundamental, CSIC, Serrano 123, Madrid 28006,
Spain.
2 New address: DAMTP, Wilberforce Road, Cambridge CB3 0WA, UK.
doi:10.1016/j.nuclphysb.2004.08.017
T.P.T. Dijkstra et al. / Nuclear Physics B 698 [PM] (2004) 450472
451
terexamples have been conjectured to exist [5]. The open, unoriented case has received
considerably less attention. Early results can be found in [6], reviewed recently in [7], and
there are some remarks on such theories in the appendix of [8]. Our purpose here is a
systematic study of the set of theories that is obtained if one allows open and unoriented
surfaces. We will study this problem for continuous values of the radius R and at rational
points, and match the results.
For the circle theories the solution to this problem is known. On oriented, closed surfaces the moduli space is a line parametrized by the radius R of the circle, with a T-duality
identification of R and /R. On unoriented surfaces this line splits into two lines. In one
of the two T-dual pictures the two lines are characterized by having O0-planes of either
the same tension or opposite tension on opposite points of the circle, in the other the circle
is covered by O1-planes. This matches precisely and unambiguously with the results from
rational CFT, as we will see in slightly more detail below.
Studying this problem on the orbifold line is interesting for a variety of reasons: the set
of modular invariant partition functions (MIPFs) is richer, there is an interesting one-to-one
map between the rational c = 1 orbifolds at radius R 2 = N and the DN WZW-models
at level 2 [9] and the orbifolds have a twist field degeneracy [2] which is resolved only in
the rational points. Finally we are interested in understanding the puzzling results of [10],
showing that in certain rational points the orbifold CFT does not admit a canonical Klein
bottle projection (with all coefficients equal to 1). There are non-canonical solutions, but it
is not immediately clear how they would fit into a continuum description.
Two types of approaches to boundary RCFT are used in this paper. In [11] a general
formula was given for all reflection coefficients and crosscap coefficients for all simple
current modular invariants [12] and a large class of orientifold projections. This formula is
consistent (at least in the sense of yielding integral coefficients, see [11,13]; furthermore
consistency for all resulting oriented amplitudes was demonstrated recently in [14]), but
there is no proof that it is complete. In [15] a set of polynomial equations was written
down which can be solved for annulus, Mbius and Klein bottle coefficients. This method
is complete, but it is not known if all solutions are physical. Our goal is to do a systematic
analysis of all known orbifold RCFTs and compare them with the continuum, to check if
the method of [11] misses anything, and if the method of [15] produces anything manifestly
unphysical. The orbifolds are an ideal laboratory for this because the continuum can be
obtained explicitly, and RCFTs are known in all rational points. The circle has all these
features as well, but lacks some non-trivial structure related to the fixed points, as well as
exceptional modular invariants.
A prerequisite for this work is a complete description of orientifolds in the continuous
case. Although there is a large body of work of orientifolds of tori and orbifolds, as far
as we know the complete and explicit answer for the c = 1 orbifolds is not available yet,
although a partial result can be found in [7]. We find a total of four different allowed Klein
bottle amplitudes and corresponding orientifold projections.
The structure of the paper is the following. In Section 2 we review the results concerning the bosonic string on a circle and the orientifolds thereof. In Section 3 we consider
the orbifold case. First, in Section 3.1, we study orientifolds for rational radius. A parametrization of the allowed Klein bottles in terms of RCFT characters leads to twelve different
choices. Then we enumerate the modular invariant partition functions (MIPFs), by system-
452
atically combining all known types and checking closure under multiplication. For each
MIPF we perform a systematic construction of U-NIMreps [16]. (U-NIMreps are sets of
annulus, Mbius and Klein bottle amplitudes that satisfy the aforementioned polynomial
equations; they are NIMreps [1719] extended with data concerning unoriented surfaces.)
This method is limited in practice to a finite number of primaries, but we can extend it far
enough to uncover the complete picture, and furthermore we supplement it with the formula of [11], which has no such limitation. We conclude that for integer values of R 2 / six
different cases can be distinguished, corresponding to four distinct continuum Klein bottle
amplitudes. As a result of these computations we obtain a set of boundary and crosscap
coefficients which we use in Section 3.1.4 to study the localization of branes and O-planes,
as well as the ChanPaton groups in the various cases. In Section 3.2 we study the case of
orientifolds at continuous radius from a geometric point of view, and confirm that the four
Klein bottle amplitudes are indeed the only possible ones. In Section 3.3 we consider the
case of certain non-integer rational radii (exceptional MIPFs) and show that, despite their
unexpected features, the results of [10] are precisely in agreement with the four orientifold
projections. In Section 4 we summarize our conclusions. In Appendix A we discuss a variety of orbifold maps needed to obtain distinct orbifold theories that exist in the rational
case.
2. Summary of circle results

In this section we want to review know results about orientifolds of the bosonic string
on a circle, in order to set up the discussion for the orbifold case and to introduce some notations. Orientifolds of circle compactifications for irrational values of the radius appeared
for the first time in [20], while additional references are [2126] (see [7] for a review). The
partition function for a bosonic closed oriented string compactified on a circle is
Z(R) =
1 (m/R+nR/ )2 (m/RnR/ )2
q4
q 4
,
m,n
(2.1)
where we denote with n the winding and m/R the KaluzaKlein (KK) momentum along
the circle. A well-known feature of this partition function is the fact that it is invariant
under the exchange R /R. From the point of view of the extended CFT that describes
the above theory, the partition function Z(R) hides chiral information. For each value of
R there are actually two theories with
which are each others
the same partition function,

/2(m/R nR/ ), and
T-dual, one with
momentum states | /2(m/R
+
nR/
),

one with states | /2(m/R + nR/ ), /2(m/R + nR/ ). The first of these is
obtained in a genuine compactification on a circle of radius R.
The allowed orientifold projections must respect the operator product expansion of the
CFT (we will only consider orientifold projections of order 2). This implies in particular
must be transformed into itself, and that the vertex operators corthat the operator XX
responding to the momentum and winding states must transform with a factor (1 )m (2 )n ,
with 1 and 2 equal to 1. Hence one can associate four consistent Klein bottles with this
453
partition function, namely3

1
2
q 2 (m/R) 2K+00 (R),
(2i2 ) m

1
2
(1)m q 2 (m/R) 2K00 (R),
(2i2 ) m
1 (nR)2
1
q 2
2K0+0 (R),
(2i2 ) n

1
2
1
(1)n q 2 (nR) 2K00 (R).
(2i2 ) n
(2.2a)
(2.2b)
(2.2c)
(2.2d)
The functions K will be introduced later. The Klein bottle amplitude is subject to the constraint that 12 (Z(R)+K(R)) expands into non-negative integers. The first two Klein bottles
can be combined with the diagonal theory, the other two with its T-dual. The positions of
the orientifold planes can be derived from the transverse channel Klein bottle amplitudes
by dimensional analysis. In the first two cases the transverse amplitude describes a closed
string propagating between two O1-planes. There are two different O1-planes. In the first
case, the resonance term permits only to even winding states to propagate in the transverse
channel, while in the second case only odd winding states contribute, and thus the configuration has vanishing tension, since the graviton does not propagate in the transverse
channel. We call these two configurations O1+ O1+ and O1+ O1 , respectively. In
the first case, the orientifold projection maps X to itself, while in the second case it maps
X to X + R. Both these maps square to the identity because of the periodicity of the
circle, and have no fixed points. The other two cases are the T-duals of the former. After
T-duality, the orientifold projection acts as a Z2 orbifold on the circle coordinate, so that
the model lives on a segment, with O0-planes at the endpoints. In the third case, the orientifold projection maps X to X, while in the fourth case it maps X to X + R. Both
these projections have fixed points, where the O0-planes are located. More precisely, the
third case corresponds to the configuration O0+ O0+ , in which the two O-planes have
the same tension, while the fourth case corresponds to O0+ O0 , in which the O-planes
have opposite tension.
The corresponding rational CFTs are obtained by setting R 2 = N . All primaries of
this CFT are simple currents, forming a Z2N discrete group. We denote the generator
of Z2N as J . All MIPFs are simple current invariants, and they are in one-to-one correspondence with the subgroups of Z2N generated by even powers of J , which in their
turn are in one-to-one correspondence with the divisors of N . If m is a divisor of N , the
MIPF belonging to the subgroup generated by J 2m corresponds to a circle with radius
R 2 = N/m2 . We will refer to this partition functions as Zcircle (m, N), with the convention that Zcircle(N, N) is the charge conjugation invariant and Zcircle (1, N) the diagonal
3 We assume here that the relation between the orientifold projection and the Klein bottle is straightforward.
This implies in particular that the Klein bottle coefficients are preserved under fusion, i.e., that the Klein bottle
constraint is satisfied. There are examples where this constraint is not satisfied (see, e.g., [27]) and which require
further thought, but this problem does not occur for any of the c = 1 U-NIMreps.
454
one. For every rational radius R 2 = p/q (with p and q relative prime) there is an infinite
number of rational CFTs describing it, namely Z(pr, pqr 2 ), for any r Z. For r > 1 the
corresponding partition functions involve extensions of the chiral algebra.
For all these MIPFs the allowed crosscap and boundary coefficients follow from the
general formula presented in [11] (summarizing and extending earlier work in [28]), which
in the special case ZN,N reduces to the well-known boundary state of Cardy [29] combined
with the crosscap state due to the Rome group [30].
By Fourier analyzing the closed string scattering amplitudes from the boundary and
crosscap states (a procedure that was pioneered in [31], and applied in [32] to boundary
states of WZW models and in [33] to crosscap states (see also [34,35])) one can localize the D-branes and O-planes on the circle. To do this one multiplies the boundary and
crosscap amplitude with a factor eikx/R and sums over all values of k in the primary range
N k < N . The resulting function of x has peaks that get more pronounced with increasing N , and are interpreted as the brane and plane positions. From the continuous
point of view, we expect D1-branes for the diagonal invariant (corresponding to a genuine
circle compactification, with the space-filling brane wrapped around the circle), which turn
into D0-branes for the charge conjugation invariant, the T-dual of the former.
In the rational CFT one finds the following. The MIPF Z(m, N) admits 2m boundaries,
each of which is localized at n = N/m distinct points on the circle. To make sense of this
we introduce the dual radius R = /R, which is the relevant quantity because D0-branes
live on the dual circle. The n-fold multiplicity is an indication of the fact that the original
circle of dual radius R is to be interpreted as an n-fold cover of a circle of radius R/n.

We
may label the boundaries by integers a = 0, . . . , m1, such that they are localized at points
( a+2m
2N )2 R, = 0, . . . , n 1. For m = N , n = 1 (the charge conjugation invariant) this
yields 2N D0-branes equally distributed over the circle; for m = 1, n = N (the diagonal
invariant) this gives two branes localized simultaneously on N equally distributed points,
one brane on the odd points and one on the even points. This is the RCFT realization of
a D1-brane.
In the continuum the D0 can be localized anywhere on the circle, whereas in the rational
CFT description their positions are quantized. One can approach the continuum results
either by using deformations of the boundary CFT [36], or by allowing boundaries that
break some of the extended symmetries that characterize the rational CFT. This can be done
by extending
by obtaining the circle at some dual radius R from a circle at dual radius Rr
the CFT of the latter. The RCFT notion of completeness of boundaries [37], when applied
to MIPFs of extension type, automatically implies the presence of boundaries that break the
extended symmetries. Indeed, if we consider Z(pr, pqr 2 ) we find 2pr distinct boundaries,
This circle is an r-fold cover of the
each localized in qr points on the circle of radius Rr.
so that on the latter circle we now have 2pr distinct branes each localized
circle of radius R,
in q points. Of these, 2p coincide with the ones found for r = 1, and the remaining ones
occupy intermediate positions. In the limit r , R fixed we approach the continuum
result.
Similar results hold for crosscaps. The formalism of [11] allows two ways of modifying
the crosscap coefficient for a given MIPF. The formula for the crosscap coefficient is, up
455
to normalization
i
(K, L)PKL,i ,
(2.3)
LG
where i is the Ishibashi label, P the P -matrix (P = T ST 2 S T ). Here K is a simple

current (subject to a condition given below), (K, L) a set of signs satisfying the constraint
(k, L) = ei(hK hKL ) K (L),
(2.4)
where K (L) is a set of phases solving the relation

K (LJ ) = K (L)K (J )e2iX(L,J ),
(2.5)
where X is the rational bihomorphism that specifies the MIPF, as defined in [12]. This
relation does not fix the phases completely: for every independent even cyclic factor of
the simple current group G, there is a free sign. These free signs are called the crosscap
signs. The current K (for historical reasons called the Klein bottle current) must be
local w.r.t. the currents of order two in G, and currents that differ by elements of G or by
squares of simple currents yield equivalent crosscaps. These Klein bottle currents form,
together with the crosscap signs, the set of allowed crosscap modifications.
In the present case, it is not hard to see that for each choice of G there are just two
solutions. If G has even order, there is a single crosscap sign choice, but there are no Klein
bottle currents local w.r.t. G. If G has odd order n there is no crosscap sign choice, but then
the current J n is local w.r.t. G and is a non-trivial Klein bottle current. These two choices
correspond precisely to the two orientifold choices in the continuous case. In all cases one
of the orientifold choices leads to Klein bottle coefficients that are equal to +1 for all fields
appearing diagonally.
The crosscap positions can be worked out in the same way as for D-branes. For the
MIPF Z(m, N) we find that a crosscap state occupies 2n positions, twice as many as a
boundary state. These positions are n-fold identified. In the simplest case, m = N , n = 1
there are two positions, diametrically opposite on the circle. In this case, the crosscaps are
characterized by the choice of Klein bottle current K = J k . Each k corresponds to two O0k
2R and r = ( k+2N
planes localized at r = 4N
4N )2R. If k + N is even these two O0-planes
have the same tension, if k + N is odd they have opposite tension. If k is even, the O-plane
locations coincide with a brane position; if k is odd the O-planes lie between two brane
positions. Configurations where k differs by an even integer can be obtained from each
other by rotating the circle, in agreement with the fact that the corresponding Klein bottle
currents are equivalent. The T-dual configuration corresponds to the diagonal invariant,
obtained by using the simple current J 2 . This MIPF admits just two Ishibashi states, and
hence only two non-vanishing crosscap coefficients. This is not sufficient to contain any
information about localization, in agreement with the fact that we expect the O-planes to
be O1-planes wrapping the circle (and analogously for boundaries).
456
3. Orbifolds
We are considering the c = 1 case, that is the real line modded out by the group G of
reflections and translations, resulting in the segment R/G = S 1 /Z2 . Following [38] we
will denote the action of elements of this group on the string coordinate as
(, n) G,
(, n)X = X + 2nR,
n Z, = .
(3.1)
Strings on an orbifold are closed if they are periodically identified up to an element of this
group:
X( + 2) = (, n)X( ).
(3.2)
X is then twisted by the element (, n); this defines various sectors with different periodicity conditions on X. Not all elements of G give rise to a different sector, a sector twisted
by g being the same as the one twisted by h, g, h1 . Thus we get a sector for each conjugacy class. One has the following conjugacy classes: (+, |m|), (, even) and (, odd).
The first gives the circle periodicity conditions with winding number m, where now the
winding direction is no longer significant. The last two cases give the twisted sectors. Note
that in these sectors the notion of winding is limited to being even or odd.
In the untwisted sector, X has the same mode expansion as in the circle, with the difference that now only the states that are invariant under the map X X are present.
Denoting with r the operator that performs this map on the Hilbert space, the resulting
spectrum is obtained by applying the projector 1+r
2 on the circle states.
In the twisted sectors the mode expansion for X is
X = x0 +

n
1
n
1
2
1z
n 2
1
n+ 2
+ a
1 z
n 2
1
n+ 2
(3.3)
where x0 = 0 for the (, even) sector, x0 = R for the (, odd) sector. The twisted sectors
correspond thus to states localized at the fixed points of the orbifold. Also the states created
by the modes in (3.3) must be projected by the operator 1+r
2 .
The resulting partition function is

1
Zorb (R) = Zcircle (R) + + + .
(3.4)
2
2
3
4
The first two are contributions of the untwisted sectors, while the last two are contributions
of the twisted sectors.
3.1. Orientifolds for rational radius
3.1.1. Parametrizations of the Klein bottle
In the CFT description, the allowed orientifold projections are limited by the requirement of preservation of the OPE. Of most interest are the projection signs of states appear must
ing diagonally, since those signs affect the Klein bottle. Again we find that XX
transform into itself. This implies that 2 cannot appear in the Klein bottle expression,
since this function represents the difference of contributions of the identity operator and
457
The OPE of XX
with a lowest weight twist field (z, z ) yields the
the operator XX.
has projection sign +, the twist fields and
excited twist field (z, z ). Since XX
must be projected in the same way. This implies the absence of 3 , which corresponds to
the difference of the two twist field labels.
An important issue is twist field degeneracy. The c = 1 orbifold has two twist fields
(stemming from the fact we have two different twisted sectors), denoted 1 and 2 (with
1 1
9 9
weight ( 16
, 16 )) and two excited twist fields, 1 and 2 , with weight ( 16
, 16 ). The labels 1
and 2 are not distinguished by the Virasoro algebra. On any point on the orbifold line the
Virasoro algebra is extended by operators that are even polynomial in X and its derivatives, the first one at weight 4 [2]. But these operators do not distinguish the labels either,
since X itself does not. Only in the rational
points there are operators that distinguish the

twist fields, namely the operators cos( 2/ mRX) that extend the chiral algebra to make
the CFT rational. Hence we should regard these as states with multiplicity 2. The allowed
Klein bottle coefficients in this sector are the 2, 0 or 2. The value 0 is allowed if the
twist fields appear off-diagonally in the partition function, or if they appear diagonally, but
have opposite Klein bottle projections. Based on this information we arrive at the following
twelve choices for the Klein bottle

1
2
k 2
1
1 1
1
K1 2 3 =
(1 )k q 2 ( R ) +
(2 )m q 2 (mR)
2 (2i2 )
2 (2i2 )
kZ
mZ

+ 23
,
(3.5)
4
with 1 = 1; 2 = 1 and 3 = 0, 1. The same parametrization can be used for the
circle theory, provided one allows the value 0 for 1 and 2 .
As was the case for the circle, the allowable Klein bottles depend on the interpretation
of the partition function. But in contrast to the circle theory, this interpretation is discontinuous in R. This is due to the fact that we can distinguish the twist fields only for rational R.
In addition, for rational R two orbifold fields appear that do not exist for irrational values
of R, namely the fields we denote as 1 and 2 and that have conformal weight 14 N .
3.1.2. Enumeration of modular invariants
We will now study the orbifold at rational points in order to reduce the set of orientifold
projections. In the rational points and for a sufficiently small set of primaries we have an
additional tool at our disposal, namely the systematic search for NIMreps and U-NIMreps.
The orbifold of the circle of radius R 2 = N (or its T-dual) is the well-known orbifold
rational CFT with N + 7 primaries. It has four simple currents, 1, 1 , 2 and the spin-1
current X, forming a discrete group Z4 (for N odd) or Z2 Z2 (for N even). The remaining fields will be denoted k , k = 1, . . . , N1 , i , i (i = 1, 2), following [2]. A lot is known
about the MIPFs of these CFTs, but the result are scattered throughout the literature, and
for that reason we will give here an enumeration of what is known.
In [2] it was observed that the theory at radius R 2 = p/q or R 2 = q/p has the same
chiral algebra as the one at R 2 = pq. Hence it is described by a non-trivial MIPF of the
theory at R 2 = pq. This MIPF is of exceptional type, except when q and/or p is equal
to 2, in which case the invariant is of simple current type. Just as in the circle case, one can
458
generalize this by allowing p and q to have common factors. In this way one can obtain an
infinite number of rational CFT realizations at any rational point on the orbifold line. Any
of these MIPFs can be extended by the simple current X to reobtain the circle partition
function.
But there are still more rational partition functions for every rational point, a fact that
can most easily be appreciated by using the fact that the modular group representation of
the orbifold CFT for R 2 = N is in one-to-one correspondence with the DN WZW-model
at level 2 [9]. In particular there is a one-to-one relation for partition functions, NIMreps
and U-NIMreps. While the aforementioned MIPFs describing orbifolds at non-integer radii
do not seem to have a raison detre for the WZW-models, they do exist for these models
as well. The ones of automorphism type were discovered using Galois symmetry in [39]
(subsequently the WZW automorphisms were fully classified in [40]); the extensions were
described in [9] using the aforementioned correspondence. On the other hand, the DN
WZW-models (at any level) have MIPFs related to Dynkin diagram automorphisms that
imply the existence of related invariants for the orbifolds. These are first of all the conjugation invariants, which have an off-diagonal pairing of 1 , 2 , 1 , 2 and 1 , 2 . For odd
N this is the charge conjugation invariant, which can also be described as a simple current
invariant generated by 1 (or, equivalently, 2 ). For even N this is an exceptional invariant (for even N , the simple current invariant generated by 1 gives an orbifold at reduced
radius N/4). If the MIPF involves an extension by 1 or 2 (which happens if p and q
are both even), there are still more possibilities, because one can then conjugate the left
and right chiral algebras independently. As a result one obtains two symmetric and two
asymmetric (heterotic) MIPFs. Finally, for N = 4 there are even more MIPFs related to
triality of D4 ; there are 16 MIPFs in total.
For given N = 4 the number of known modular invariants, obtained by combining
all
the above, can be described as follows. Let p be a divisor of N in the
range 1 p N ,
and define q = N/p. The number of known invariants is equal to p M(p) where the
sum is over all divisors p in this range, and M(p) = 3 if either p or q is odd, M(p) = 5
otherwise. The multiplicity three corresponds to the diagonal invariant, the conjugation
invariant and the circle extension, which we denote, respectively, as ZD (p, N), ZC (p, N)
and ZX (p, N). The multiplicity five corresponds to the four cases described above plus the
circle extension, denoted, respectively, as Z11 (p, N), Z22(p, N), Z12 (p, N), Z21(p, N)
and ZX (p, N) (the circle extension includes both 1 and 2 ).
Since the standard orbifold map X X yields just one orbifold theory for each T-dual
pair of circle theories, one has to consider more general orbifold maps to get the various
types of orbifold partition functions. These maps are discussed in the appendix.
3.1.3. U-NIMreps
Consider first ZC (1, N) and ZD (1, N). In all but one case these MIPFs are C-diagonal
or of simple current type, and a set of Klein bottle coefficients can be obtained from various
previous papers. To deal with the remaining exceptional MIPF (ZC (1, N), N even), and as
a check on all the others, we solved the U-NIMrep polynomial equations (for N 16) to
get the complete answer. This results in the following six cases:
459
(1) The diagonal invariant, with standard Klein bottle (Ki = 1 for all i). For N even,
this is the standard CardyRome case. For N odd, it is a simple current automorphism,
as treated in [11], with suitable choice of the crosscap sign. The resulting Klein bottle
amplitude is K+++ . This result was first obtained in [6] (see also [7]).
(2) The diagonal invariant, with Klein bottle coefficients 1 in the twisted sector. For
N even, this is the same as the previous case but with Klein bottle current 2. For N odd, it
is the same as the previous case, but with the opposite crosscap sign. The resulting Klein
bottle is K++ .
(3) The diagonal invariant, with Klein bottle currents 1 or 2 . This case exists only for
N even. For N odd, the diagonal invariant is generated as a simple current automorphism
of current 1 , and the only allowed sign changes are the crosscap signs, which we already
saw above. For even N one finds that all odd charged fields k get a negative Klein bottle,
i.e., K k = (1)k . Furthermore either 1 and 1 or 2 and 2 change sign, so that the total
twisted sector contribution cancels. The result is K+0 (for N even only).
(4) The charge conjugation invariant with standard Klein bottle (i.e., all Ki = +1 if i
appears diagonally). The charge conjugation invariant only differs from the diagonal one
for N odd. The effect is that 1 , 2 appear off-diagonally, and the same for the twisted
sector. The latter implies 3 = 0. The absence of 1 , 2 in the Klein bottle amplitude
implies 2 = 1, so that the contribution of i cancels between the first two terms. Hence
we get K+0 (for N odd only).
(5) The charge conjugation invariant with non-standard Klein bottle (case 4 with simple
current Klein bottle current 1 , which is equivalent to 2 ). This gives a sign flip for all odd
charge fields, implying 1 = 1 This can be taken into account by inserting a (1)k into
the first sum. Since N is odd, the i contribution cancels between the first two terms if and
only if 2 = +1. Hence we get K+0 (for N odd only).
(6) The conjugation invariant for even N . This is an exceptional invariant that pairs 1
with 2 , 1 with 2 and 1 with 2 . Here [11] does not apply, but by solving the NIMrep
conditions explicitly we find only one NIMrep with one U-NIMrep. The Klein bottle has
all allowed coefficients equal to 1. This yields K+0 (for N even only).
We summarize these results in Table 1. In the first column D denotes the diagonal
invariant, C the charge conjugation invariant, and T the twist field conjugation invariant, in which i , i and i are off-diagonal. In the fifth column we indicate the ChanPaton
group for the dominant branes (i.e., the ones that are most numerous for large N ). This will
be explained in the next subsection. The last column refers to the six cases listed above.
Note that all allowed continuous Klein bottle amplitudes make their appearance for
both odd and even N , but in rather different ways. Note also that for even N the diagonal
invariant (D) allows four different orientifold projections (the case K+0 actually consists
of two subcases with opposite signs for all Klein bottle coefficients in the twisted sector),
whereas the twist conjugation invariant (T) allows only one. This is strange because we
expect these theories to be dual to each other (in the sense of the existence of a one-to-one
map between their operators, respecting all correlators). This duality is of course not the
T-duality of the circle (which was modded out in the orbifold). We do not know if such
a duality has been proved in the literature, but it certainly seems to hold in the simplest
case, N = 2, the tensor product of two Ising models. Note that T-dual circles admit the
460
Table 1
Invariant
Boundary/Crosscap formula
Klein bottle
CP-group
odd
[11]
+ + +
+ +
SO
SO
1
2
C=T
odd
Cardy/Rome
+ 0
+ 0
SO
U
4
5
D=C
even
Cardy/Rome
+ + +
+ +
+ 0
SO
SO
U
1
2
3
even
exceptional
+ 0
SO
Case
same number (namely two) of O-plane/D-brane configurations, and the only aspect that
differs is the number of allowed D-brane positions. In the orbifold case two probably dual
theories have a different number of orientifold projections, corresponding to physically
different configurations, with different CP-groups. Although this is counterintuitive, on the
other hand it does not seem to contradict the duality in an obvious way. Note that also the
number of boundary conditions differs for the two mutually dual cases, but this merely
corresponds to a different choice of rationally allowed positions for the same D-branes.
Note furthermore that for T-dual rational circle theories the number of orientifold choices
is the same.
After this enumeration (which is exhaustive for small N ) only four of the twelve potential Klein bottles are realized. Most absences can be explained by a combination of the
following facts:
The twist fields i , i and the fields i must be simultaneously (off)-diagonal in any
MIPF. This follows from modular invariance.
The Klein bottle coefficients of 1 and 2 must be identical, because these fields are ei which must have projection sign 1.
ther each others conjugates, or they fuse to XX,
We call these coefficients K .
K can be expressed as 12 (2 + 1N ).
The fusion of 1 and 2 produces fields k , with k + N odd.
The fusion of 1 and 1 produces fields k , with k + N even.
The last two points are relevant only if the twist fields appear diagonally. These points
imply the following. For conjugation invariants we must have 3 = 0 and 2 = 1N . For
diagonal invariants we must have 2 = 1N . Furthermore from point 5 we find that 1 = 1
for N odd. Hence 2 = 1 for all diagonal invariants. If 3 = 0 the projections of 1 and 2
are opposite. This is impossible for N odd, and implies 1 = 1 for N even. On the other
hand, if 3 = 0, point 4 implies 1 = 2 = 1 for N even.
This only leaves one case that was not found, and is also not yet ruled out, namely 1 =
2 = 1 for conjugation invariants with even N . This case can be ruled out by computing
the transverse channel amplitude. It turns out that 1 and 2 propagate in the transverse
channel, although they are not Ishibashi states. Hence this case must be rejected.
461
As an additional check on the result one can now solve the U-NIMrep equations for all
other accessible MIPFs as well. The results are in complete agreement with the foregoing:
ZD (p, N) has 4 U-NIMreps for N even, 2 for N odd; ZC (p, N) has one U-NIMrep for N
even, and 2 for N odd; Zii (p, N) always has 4 U-NIMreps, whereas Zij (p, n), i = j , has
none. Finally ZX (p, N) always has 4 NIMreps, except when N = p2 , in which case it has
two. The four U-NIMreps of ZX correspond precisely to two distinct Klein bottle choices
for the circle and its dual. For N = p2 one ends up in the self-dual point, which explains
why one gets only half the number of solutions. In all cases these distinct U-NIMreps
correspond to choosing different boundary conjugations for a single NIMrep, although
ZC (3, 9) and ZD (3, 9) have, respectively, one and two additional NIMreps that do not
admit any U-NIMreps, and are presumably spurious. Finally, Zij (p, n), i = j was found
to have a single NIMrep which does not admit a U-NIMrep, in agreement with the fact that
these MIPFs are asymmetric.
The U-NIMreps for ZC (p, pq) and ZD (p, pq), with p and q prime have been given
explicitly in [10], and U-NIMreps for simple current MIPFs are described in [11]. In all
other cases these conclusions, as well as the completeness of the entire picture, rely on
extrapolation to arbitrary N .
3.1.4. Localization and ChanPaton groups
In the rational CFT description one can attempt to get information about the boundary
and crosscap states by analyzing the Fourier transformation of their coupling to closed
string states. This amounts to probing the brane/plane positions by scattering of gravitons
[31] (or, equivalently, dilatons or tachyons). This method has a clear physical interpretation in flat space, but becomes less intuitive when applied to compact spaces, although
sensible results are obtained for the circle (as explained above) and WZW-models [32,33].
Apart from the proper physical interpretation, a second caveat is that this method requires
a precise knowledge of the boundary and crosscap coefficients. In many cases the latter
are obtained by imposing integrality conditions on annulus, Mbius and Klein bottle amplitudes. These amplitudes are not sensitive to sign changes in the coupling to Ishibashi
states provided one makes the same sign change in the boundary and the crosscap coefficients. Such sign changes do not affect tadpole cancellation either. In principle the true sign
can be determined by solving the sewing constraints, but that has been done only in very
few cases. However, in the Cardy case the results of [41] imply that the signs are correct.
This should then also apply to all possible choices of orientifold projections, since this is
expected to add O-planes in different positions while keeping the branes fixed.
Keeping these caveats in mind, we can compute the positions as follows. The coupling
to the fields k provides a natural set of Fourier components for the couplings. Inspired by
the circle results we define a shape function
F (x) =
N1

k=1

eikx/R + eikx/R C(k),
(3.6)
462
where C(k) is a boundary or crosscap coefficient4 for the coupling to k . This function is
periodic with period 2R and symmetric in x x as well as R + x R x, and
therefore it is natural to identify the line segment [0, R] with the orbifold. Note that in the
diagonal and conjugation modular invariants of the orbifolds with R 2 = N all Ishibashi
labels k occur, so that there are no other points with reflection symmetry: the identification
of the two orbifold points is unambiguous. By defining C(k) = C(k) the second term can
be used to extend the sum to negative k. The coefficients C(0) C(X) and C(1 ) + C(2 )
turn out to have precisely the right value to complete the sum to the range N k < N .
The resulting function F (x) typically has one or two positive or negative peaks along the
orbifold line, which approach -functions for large N . We interpret the extrema as O0plane positions, and the sign as an O0 charge. For the six cases in the table, the coefficients
C(k) for the crosscaps either
vanish for all even k, or for all odd k. The non-vanishing
values are all equal to 1/ N , up to a sign. If this sign is positive, this leads, respectively,
to opposite charge or same-charge planes at the two endpoints of the orbifold line, x = 0
and x = R. The other possibility we encounter is a sign (1)k/2 for even k. This shifts
the plane positions by 12 R, so that they are on top of each other.
Because the orbifold incorporates the circle T-duality, which interchanges D0 (O0) with
D1 (O1) branes (planes) we expect boundary and crosscap states to describe a combination of branes and planes of dimension 0 and 1. While the D0/O0 positions and charges
can be extracted very easily from F (x), this is not the best way to determine the D1/O1
charges. The information is in fact hidden in the linear combinations C(0) + C(X) and
C(1 ) C(2 ) which are not used in the computation of F (x). A Fourier transform
of these two quantities yields identical values on all allowed brane positions in the first
case, and alternating values on even and odd positions in the second case. Furthermore
in all cases either C(0) + C(X) or C(1 ) C(2 ) is zero. Remembering how D1branes emerged for the rational circle, we are led to the conclusion that a non-vanishing
C(0) + C(X) implies the presence of two equal-charge D1/O1 branes/planes, whereas a
non-vanishing value of C(1 ) C(2 ) implies two opposite-charge D1/O1 branes/planes.
The charge referred to above always refers back to the corresponding quantity for
the circle, namely the dilaton coupling strength. This allows us to interpret any orbifold
brane/plane configuration in terms of a collection of circle configurations of different dimension and charges. Not surprisingly, this interpretation breaks down for branes labelled
by twist fields, that have no circle analog. Furthermore the values C(i ) and C(i ) (which
vanish for crosscaps and most boundaries) are also not used, and we do not have a geometric interpretation for these values.
Some more information about brane and plane positions can be gathered from boundary conjugation, which geometrically corresponds to a reflection of a brane through an
O-plane. This property affects the ChanPaton groups of the brane, which is orthogonal
or symplectic for self-conjugate (real) branes and unitary for pairs of conjugate branes.
Boundary conjugation and the allowed CP-groups are not affected by the aforementioned
sign ambiguities, but the distinction between symplectic and orthogonal for real boundaries
4 We use here the coefficients specified in [11], but without the denominator factors S . It is more natKi
ural to absorb these factors in the Ishibashi metric for the unoriented annulus, so that the boundary coefficients
themselves are independent of the choice of orientifold. See also [10,32,33].
463
does depend on the overall sign of all crosscap coefficients relative to all boundary coefficients. This sign determines the O-plane tension, and whenever we specify a CP-group
below we have fixed the tension to a negative value, so that the dilaton tadpole can be cancelled (in principle) between D-branes and the O-plane. Orbifold O-planes (unlike circle
O-planes) always have non-zero tension.
For the cases discussed listed in the table we find the following positions:
N even, diagonal invariant. This is the Cardy case, so boundary labels correspond to
primary labels, and the localization analysis should be reliable. The branes with labels 0 and X are at x = 0, the ones with label i are at x = R. The branes with
labels k are localized at points equally spread over the interval. All these branes have
in addition a D1 component. The circle-inspired Fourier analysis cannot be trusted
for the twisted sector branes and indeed gives contradictory results. The orientifold
choices correspond to the four distinct choices of the Klein bottle current, K = 0,
X, 1 and 2 . For K = 0 we get K+++ , and we find two O0+ -planes at the orbifold points plus two O1+ -planes; for K = X (K++ ) we get two O0 -planes at the
orbifold points, and again two O1+ -planes. For K = 1 or 2 (K+0 ) we get two coincident O0+ -planes at x = 12 R, plus an (O1+ + O1 ) configuration. For K+++ all
CP-groups are SO. For K++ the CP-groups of boundaries 0, X, i , i and i become unitary, while all others remain SO. For K+0 all CP-groups are unitary, except
for k , k = N/2 and the twist fields with either label 1 or 2, for which we find orthogonal groups. The group type for the k branes is easy to understand: if the O-planes
are in the middle of the orbifold line segment, they conjugate the branes mutually, except the brane in the center, which is self-conjugate. If the planes are on the endpoints,
they conjugate all -branes to their orbifold image, i.e., to themselves, so that they are
self-conjugate. A clear geometric picture suggests itself. Given a choice for the orbifold plane, there are two choices for the orientifold plane: on top of it, or orthogonal
to it. The first choice leads to K+++ and K++ and mostly self-conjugate branes, the
second to K+0 and mostly conjugate brane pairs. The proper geometric interpretation
of the CP-groups of the eight special branes (those not labelled by k ) is somewhat
less intuitive.
N even, twist automorphism. This is an exceptional invariant, and we obtained the
boundary and crosscap coefficients numerically for small values of N , up to the sign
ambiguity described above. Given the Klein bottle amplitude (K+0 ) one can easily
compute the crosscap coefficients for all N : C(0) = C() = 12 , C( 2k+1 ) = 1 ,
N
C( 2k ) = 0. This implies an O0+ -plane at x = 0 and O0 -plane at x = R. In addition there are two O1+ -planes. All CP-groups are orthogonal. We have no explicit
formula for the reflection coefficient for arbitrary N , although it could be obtained in
principle using the methods of [42] applied to the twist orbifold of the c = 1 orbifold
(which is the c = 1 orbifold with four times the value of N ). In the absence of such a
formula it is difficult to discuss brane positions with these methods. There is also no
canonical labelling of the boundary states.
N odd, charge conjugation invariant. The discussion of brane positions is identical to
the one for even N , except that there is no brane in the middle. There are four possible
choices for the Klein bottle current, but K = X and K = 2 are known to be identical
464
to K = 0, K = 1 , respectively, up to interchange of branes [11]. For K = 0 (K+0 )

the O-plane configuration is as for even N , and all CP-groups are orthogonal except
those of the i , i and i branes, which are unitary. For K = 1 (K+0 ) the O-plane
configuration is also the same as for even N , and all CP-groups are unitary except the
ones labelled by i , i (i = 1 or 2), which are orthogonal. Apart from the usual eight
special branes, these results are analogous to those for even N .
N odd, diagonal invariant. Here the formulas of [11] apply. The boundary states are
labelled by orbits of the simple current 1 . There are N + 1 branes, two for each
label k with k even, one labelled by 0, and one labelled by a twist field. Boundary
0 is localized at the orbifold endpoints, the twist field boundary is not localizable,
and all other boundaries occupy two symmetric positions on both sides of the center.
There are two orientifold projections, distinguished by opposite crosscap signs. One of
them yields K+++ , and all CP-groups are SO. The other yields K++ , and all groups
are SO except the one of the twist field boundary, which is symplectic. The O-plane
configuration consists of two O1+ -planes, plus two O0+ - (for K+++ ) and two O0 (for K++ ) planes located at the center. The fact that the boundaries are self-conjugate
is understood as a consequence of the fact that each is symmetrically located on each
side of the O0-plane. Note however that the picture seems rather different than for
even N , where the O0-planes are at the orbifold points. Note also that in this case the
caveat regarding signs of the coefficient applies. If we modify all coefficients C(k) by
a factor ()k/2 the O-plane positions are as for even N (however, the brane positions,
which also change, are still different than they are for even N ).
Finally we can extract from [10] the crosscap and boundary coefficients for R 2 =
pq odd, but only up to signs, as explained above. For the crosscaps the Fourier
transformations splits naturally into two sums, one proportional to 1/ q and the other
to 1/ p. The first gives O0-planes at multiples of 1/q of the full radius, the second
at multiples of 1/p, with signs depending on the case considered. These sums are completed by including C(0) C(X), C(1 ) + C(2 ) in one of the sums and C(0) + C(X),
C(1 ) C(2 ) in the other, in agreement with the foregoing discussion. The result can
be interpreted either in terms of a circle of radius R 2 = p/q or in terms of a circle of
radius R 2 = q/p. The first possibility corresponds a q-fold identification of the orbifold
line, the second to a p-fold identification. In the first case the planes originating from the
first Fourier sum are at the endpoints, whereas those from the second one are distributed
equally on p points of the reduced line segment. It is natural to regard the latter as rational
CFT realizations of D1-branes.
For ZC (p, q) two orientifold projections were found in [10], that differ by interchanging
p and q. The O-plane charges are alternating for one of the Fourier sums, and identical for
the other. On the reduced orbifold line segment this can be interpreted as a configuration
with two O0+ -planes at the end, plus one O1+ - and one O1 -plane, and a configuration
with one O0+ -plane and one O0 -plane at the end, plus two O1+ -planes (two, because
odd and even points are to be identified with different O1-planes, as in the previous case).
For ZD (p, q) there are also two orientifold projections, this time differing by signs in the
crosscap coefficients, that flip the two Fourier sums with respect to each other. Using the
same interpretation we now get two O0+ -planes at the end, combined with either two O1+ p/q,
465
or two O1 -planes. All this is identical to the results for odd, integer radius, except for the
positions of the two O0+ -planes. But precisely these positions are affected by the unknown
signs. This particular kind of simple current MIPF (generated by a Z4 -current with fixed
points) does not appear in the circle theory and hence the correctness of these signs cannot
be tested using brane localization on the circle.
3.2. Orientifolds for arbitrary radius
In this subsection we argue that the four Klein bottle amplitudes that we obtained in
the previous subsection are the only possible ones for arbitrary radius. Thus, we find all
possible orientifold maps, that is maps that project out states that are not invariant under
the exchange z z . Since the orientifold transformation of XL and XR must square to the
identity, the oscillators transform like an a n . The only freedom left is in the operators
coming from the z independent parts in the expansion of XL and XR .
The standard orientifold projection [6] corresponds to the map XL XR , giving rise
to the amplitude (see (3.5))

k 2 1
1 (mR)2
1
1
1
K+++ =
(3.7)
q 2 (R) +
q 2
+2
.
2 (2i2 )
2 (2i2 )
4
kZ
mZ
The first two terms arise from a trace over the states in the untwisted sectors. In the first of
these two the orientifold map is inserted, and only the states with zero winding contribute.
The second is the one with both the orientifold and orbifold map inserted, and since the
KK momentum is not invariant under reflections, only states with no KK momentum contribute. The last term comes from the two twisted sectors; they both contribute in the same
amount.
The first variation is to let the operators that create the ground states in the twisted
sectors acquire a minus sign under the orientifold transformation. This will result in a
minus sign in the last term of (3.5), giving

k 2 1
1 (mR)2
1
1
1
(R)

2
q
+
q 2
2
.
K++ =
(3.8)
2 (2i2 )
2 (2i2 )
4
kZ
mZ
In order to understand which other possible maps are allowed, one has to consider the
way the various sectors interact. The untwisted sectors combine according to
(+, n)(+, m) = (+, n + m).
(3.9)
Apart from the standard projection, this equation allows for the map

,
2 R

XR XL
(3.10)
,
2 R
that changes the sign of the states in the sectors with odd winding. Because twisted sectors
combine like
XL XR +
(, odd)(, even) = (+, odd),
(3.11)
466
consistency requires that the two twisted sectors contribute with opposite sign after the
projection. The resulting amplitude is
k 2 1 1

1
2
1 1
q 2 (R) +
(1)m q 2 (mR) ,
K+0 =
(3.12)
2 (2i2 )
2 (2i2 )
kZ
mZ
where the contribution from the twisted sectors cancels.

From T-duality one can then obtain the last consistent map,
XL XR + R,
2
XR XL + R,
(3.13)
2
that changes the sign of states with odd KK momentum. Since X X + R, the twisted
ground state localized in X = 0 is swapped for the one localized in X = R. This means
that 0 and R are no longer fixed points of the orientifold map, whose eigenstates are now
localized in X = 2 R. Moreover, since the trace over the twisted states vanishes after the
projection, there is no contribution from the twisted sectors to the Klein bottle amplitude,
whose form is

1 (mR)2
k 2
1
1 1
1
(1)k q 2 ( R ) +
q 2
.
K+0 =
(3.14)
2 (2i2 )
2 (2i2)
kZ
mZ
In order to understand why the choice 1 = 2 = 1 in Eq. (3.5) is not allowed, we

have to analyze the transverse channel. We only need to consider the untwisted sector, so
we concentrate on the case K0 . The Klein bottle in the direct channel depends on 2i2 ,
the modulus of the doubly-covering torus. The Klein bottle in the transverse channel is
obtained performing an S modular transformation on the modulus of the doubly-covering
torus, that is writing the amplitude in terms of = 1/22 , the proper time in the transverse
channel, describing a closed string propagating between two orientifold planes. The endresult of this transformation (see for instance [7] for a review) is

2 1 (kR)2 1 1 1 ( m )2
1 1
R
K++ =
q
+
2
q R 2
,

2 (i)
2 (i) R
2
kZ
mZ

2 1 (kR)2 1 1 1 ( m+1/2 )2
1 1
R
R
K+0 =
q
+
2
q
,
2 (i)

2 (i) R
kZ
mZ

1
2 1 (k+1/2)2R 2 1 1 1 ( m )2
1
R
K +0 =
q
+
2
q R ,
2 (i)

2 (i) R
kZ
mZ

2 1 (k+1/2)2R 2 1 1 1 ( m+1/2 )2
1 1
R
R
K0 =
q
+
2
q
, (3.15)
2 (i)

2 (i) R
kZ
= e2 .
mZ
The states that contribute to the Klein bottle amplitude in the transwhere here q
verse channel are closed-string states propagating between two orientifold planes. In K ++
only states with even KK momentum or even winding contribute in the untwisted sector. In
K +0 only states with odd KK momentum or even winding contribute. This is consistent
467
with the direct channel, since K+0 projects out only states with odd winding. The same is
valid for K +0 , where only states with even KK momentum and odd winding contribute,
while in the direct channel the states with odd KK momentum are projected out. Finally, in
the case of K 0 states with odd KK momentum and states with odd winding contribute
to the transverse amplitude, but these states are both projected out by K0 , so that this
Klein bottle projection is not consistent.
From the transverse channel analysis one can also derive the position of the orientifold
planes for the various Klein bottles. In all cases, namely the standard orientifold projection
XL XR , and the ones given in (3.10) and (3.13), the map has no fixed points, and this
corresponds to introducing O1-planes. Once the orbifold map X X is implemented,
all these maps develop fixed points, where O0-planes are located. This means that all these
amplitudes describe a configuration of two O1-planes, and two O0-planes located at the
fixed points of the orientifold + orbifold (r) map. The twisted sector corresponds in
the transverse channel to closed string states propagating between an O1-plane and an
O0-plane.
In the case of K++ , the and r maps are, respectively, X X and X X,
and the two O1-planes have the same tension, as well as the two O0-planes. The two sign
options for the twisted sector correspond to the fact that the tension of the O0-planes can
be positive or negative with respect to the tension of the O1-planes.5 Since the r map
has fixed points in 0 and R, the position of the two O0-planes coincides with the two
fixed points of the orbifold. Applying the T-duality transformation X = XL + XR X =
XL XR , and R R = /R, one can see that these two Klein bottles are both selfT-dual. In the case of K+0 , the projection is given in (3.10), and it maps X to itself,
meaning that the two O1-planes have the same tension. The r projection maps X to X,
so that the O0-planes are located at the fixed points of the orbifold. In order to determine
the tension of the O0-planes, one has to consider how the r transformation acts on the
T-dual coordinate. In this case one has X X + R , and the shift in the dual coordinate
is a manifestation of the fact that the two O0-planes have opposite tension. The Klein
bottle amplitude corresponds thus to the configuration O1+ O1+ O0+ O0 , and
the twisted sector cancels because of the opposite contribution from the two orbifold fixed
points, where the O0-planes are located. Finally, in the case of K+0 , the projection
is given in (3.13), and it maps X to X + R, so that the two O1-planes have opposite
tension, while r maps X to X = R, so that the O0-planes are located in the middle
of the orbifold segment. In this case the twisted sector cancels separately in any of the two
(coincident) orientifold fixed points. T-duality maps this configuration to the previous one.
In all these cases the locations and charges of the O0-planes and the charges of the O1planes agree with the results obtained from the CFT analysis in Section 3.1.4, except those
for the K++ Klein bottle of the D-invariants for odd N . In that case the charges are the
same, but the two O0+ -planes were found in the center rather than at the edges. But this
was precisely a non-Cardy case, where the signs of the crosscap coefficients (crucial for
the precise location) are not determined.
5 The K
++ case is analogous to the six-dimensional brane supersymmetry breaking model of [43], in which
the O5-planes and the O9-planes have opposite tension.
468
The foregoing discussion was for arbitrary radius, and seemed to rely in all cases on the
standard orbifold map X X. At rational radii the various orientifolds occur in combinations with specific partition functions, which require different orbifold maps, discussed
in the appendix. This changes the map r in the foregoing discussion, and hence its fixed
points. However, it also changes r, whose fixed points determine the O-planes. It is easy
to see that both modifications cancel, so that the relative position of orbifold fixed points
and O0-plane positions remains unchanged. Note that nothing in the analysis in this section
imposed any relation between the orientifold map and the orbifold map in rational points.
This relation was found in Section 3.1.3 and makes use of OPEs involving distinct twist
fields. We did not consider twist fields in this section, and furthermore for non-rational
radius they cannot be distinguished, hence there is no reason to expect such a relation to
emerge.
3.3. Orientifolds of exceptional MIPFs
As mentioned in Section 3.1, the orbifold has exceptional MIPFs, constructed using an
automorphism, , which leaves the fusion coefficients invariant: Nij k = N(i)(j ) (k) . The
exceptional torus partition functions obtained from the chiral algebra of the CFT at square
radius R 2 = pq, with p and q odd prime numbers, are

T=
i Ci(j ) j .
T=
i i(j ) j ,
(3.16)
In [10], these two invariants were called diagonal+automorphism (D+A) and Cardy+
automorphism (C + A), respectively. Geometrically, these two tori describe a free boson
compactified on an orbifold of radius R 2 = p/q and its T-dual.
We first review the results of [10] about orientifold projections. In the D + A case, the
trivial Klein bottle, that is Ki = 1 for all the fields that couple diagonally on the torus,
is allowed. A second Klein bottle is also allowed, with Ki = 1 for the twist fields and
Ki = 1 for the other diagonal fields. In the C + A case, surprisingly the trivial choice
Ki = 1 for all the fields coupling diagonally on the torus is not allowed. There are two
Klein bottles. One has Kk = 1 when k is an odd multiple of p and Ki = 1 otherwise,
and the other is obtained exchanging p with q.
Looking at the resulting amplitudes as functions of the orbifold radius, one realized
that these Klein bottles are precisely the ones obtained in Section 3.2, for a bosonic string
compactified on a circle of square radius R 2 = p/q. In particular, the Klein bottles of
the D + A modular invariant are K+++ and K++ , while the Klein bottles of the C + A
modular invariant are K+0 and K+0 . Since the twisted sector is not diagonal for the
C modular invariant, this is the only possibility that is allowed in light of the results of
the previous subsection, and thus the results of [10] are completely consistent with the
orientifold projections that are allowed for arbitrary radius.
4. Conclusions
We have identified four distinct orientifold projections for the c = 1 orbifolds. Geometrically, they can be described most easily starting from the O0-plane configurations of the
469
T-dual circle. The (O0+ , O0 ) configuration has only one axis of symmetry, namely the
line through the O0-planes. Hence the orbifold and O-plane directions must line up, and
the orbifold O0-planes are at its endpoints. The configuration (O0+ , O0+ ) has two axes of
symmetry, and the orbifold reflection line is either on top of or orthogonal to the orientifold
line. Then the O0-planes are, respectively, at the endpoints or on top of each other in the
center.
In the circle theory on can distinguish two T-dual orientifold maps, one of the form
XL +XR + const and one of the form XL XR + const. The former has fixed
points in XL XR , but not in XL + XR , whereas for the latter it is just the other way
around. Therefore the former gives rise to O1-planes on the circle and the latter to O0planes on the T-dual circle. The orbifold map (which has fixed points both in XL XR and
XL +XR ) transforms the two types of orientifold maps into each other, so that both O1- and
O0-planes are present. Inspection of the transverse channel show that the charges of the O1planes are identical if the orbifold fixed plane and the orientifold plane coincide, whereas
they are opposite if these fixed planes are orthogonal. Allowing for an additional relative
sign between the O1- and O0-planes then gives a total of four configurations (since the
overall sign is irrelevant): (O1+ O1+ ) (O0+ O0+ ), (O1+ O1+ ) (O0 O0 ),
(O1+ O1+ ) (O0+ O0 ) and (O1+ O1 ) (O0+ O0+ ). This argument also
shows why a fifth logical possibility, (O1+ O1 ) (O0+ O0 ), cannot occur.
This intuitive argument was worked out in detail in Section 3.2, and is backed up by the
complete solution for U-NIMreps for rational CFT. The latter classification can be done
exhaustively, but this is necessarily limited to a few rational points. We have shown that
all four orientifolds are realized in all rational points, although in rather different ways.
We have also shown how a known, but initially surprising solution at R 2 = p/q fits in
perfectly with the continuum.
At arbitrary R we cannot rigorously rule out additional solutions, but in view of the
agreement between the continuous R and the rational CFT descriptions, any deviations
would be quite surprising.
A few open problems remain. While all methods agree on the O-plane charges, there is
a discrepancy on their precise positions in one case, interestingly precisely the case were
the CFT results are least reliable. Secondly, the nature of the duality between diagonal
and conjugation invariants of the rational orbifolds needs to be clarified. Finally, in the
geometric description, applied to rational radii, the link between the choice among those
two invariants and the orientifold map is not manifest.
Acknowledgements
We would like to thank Christoph Schweigert and Yassen Stanev for discussions and
Herman Verlinde for clarifying some points in [2]. A.N.S. wishes to thank IMAFF-CSIC,
Madrid, where part of this work was done, for hospitality. The work of F.R. and A.N.S. has
been performed as part of the program FP 52 of the Foundation for Fundamental Research
of Matter (FOM), and the work of T.P.T.D. and A.N.S. has been performed as part of the
program FP 57 of FOM. The work of B.G.-R. and A.N.S. has been partially supported by
funding of the Spanish Ministerio de Ciencia y Tecnologa, Project BFM2002-03610.
470
Appendix A. Orbifold maps

Here we will discuss how the various partition functions enumerated in Section 3.1.2
can be obtained from the circle theory. The standard description of orbifolds starts with a
circle theory, from which the Z2 -symmetry X X is modded out. It is not hard tosee
that applying this map to the circle theories Z(p, N) or Z(N/p, N) (with 1 p N
one obtains in both cases the orbifold theory Z... (p, N). The problem is that in the rational
case the orbifold partition function has an additional label D, C or ij . Since the distinction
is not made by the T-duality of the circle, there must be more than one way to do the
orbifold map. Obviously one can generalize it to X a X, i.e., rotating the plane of
reflection, but this does not have the desired effect.
It turns out that one must consider the chiral orbifold map
XL aL XL ,
XR aR XR .
On the vertex operators V (k) corresponding to the fields k the only effect is a phase
between the two terms of which they consist; but the effect is more important for the generators of the chiral algebra and the fields 1 and 2 , which make the difference between
the rational and the non-rational case (we will ignore the twist fields here, since the difference between the various partition functions is already clear in the untwisted sector). Note
that the circle theory operators from which 1 and 2 originate, which have k = N and
chiral ground state multiplicity 2, appear in an identical way for a circle and its T-dual.
The dependence of these vertex operators on aL and aR is as follows for the chiral
algebra generators
2R
WL = V (R, 0) + eiaL V (R, 0)

and
2R
WR = V (0, R) + eiaR V (0, R),

where
2r
2s
V (r, s) = ei XL ei XR .
For the other four circle operators with |r| = |s| = 12 R we get two invariant combinations
R
V A = V (+, +) + ei (aL +aR ) V (, )

and
R
V B = V (+, ) + ei (aL aR ) V (, +),

with the arguments + and denoting +R/2 and R/2, respectively.
The operators V A and V B are Virasoro-degenerate, but are distinguished by the chiral
algebra operators WL and WR . In order to relate these operators to a partition function
interpretation we need to combine V A and V B into chiral algebra eigenstates. For this
we need the OPE of these operators, and here an important role is played by the cocycle
factors that should be added to these operators [44]. In this case these cocycles can be
conveniently represented by Pauli matrices (3 )m (1 )n where m and n are the winding
471
and momentum quantum numbers of the operator. It is easy to see that WL and WR acquire a factor 3 (1 )N , V (+, +) and V (, ) a factor (1 )N and V (+, ) and V (, +)
a factor 3 . Hence for even N the cocycles do not change anything in comparison with the
naive OPE. For arbitrary N the chiral algebra eigenstates are found to be

R
R
R
V (+, +) + ei (aL +aR ) V (, ) i N ei aR V (+, ) + i N ei aL V (, +) .
For even N this can be factorized as

R
R
VL (+) ei aL VL () VR (+) ei aR VR () ,
with correlated signs in the two factors; for odd N it cannot be factorized. For these operators to have sensible reality properties, aL and aR must be quantized as multiples of /R,
the allowed positions in the rational CFT description (these are precisely the allowed brane
positions on the circle; any other value would not allow a rational CFT interpretation).
Then one finds that for N even the operators are real, and for N odd they are each others
conjugate, in agreement with the modular matrix S [2].
For the standard case aL = aR = 0, and for N even, the operators have the expected cos
cos and sin sin form that is indicative of the diagonal invariant. By choosing aL = 0,
aR = /R one can change this to a cos sin and sin cos form, corresponding to the
conjugation invariant. For odd N the results are similar. The operators for aL = aR = 0
can be written in the form cos cos i sin sin and they change to cos sin i sin cos for
aL = 0, aR = /R. These two cases should correspond, respectively, to the diagonal and
charge conjugation invariant of the odd N orbifold. To get the heterotic orbifold invariants
we may choose aL = 0, aR = /2R. Note that this value for aR does not belong to the
set of allowed positions, but it is an allowed position for the orbifold with twice the value
of R. The heterotic theory is obtained as a chiral algebra extension of the latter CFT. The
term in the partition function corresponding to V A and V B has multiplicity 2, and it is a
simple current fixed point, which cannot be resolved using the orbifold data alone. Hence
the reality properties of these operators are not determined.
The cocycle factors are also needed in the operators that implement the various orientifold maps on the vertex operators. In some cases one has to include a factor 3 in these
operators, which affects the result only for odd N and only when acting the operators V A
and V B .
References
[1]
[2]
[3]
[4]
[5]
[6]
[7]
R. Dijkgraaf, E. Verlinde, H. Verlinde, Commun. Math. Phys. 115 (1988) 649.

R. Dijkgraaf, C. Vafa, E. Verlinde, H. Verlinde, Commun. Math. Phys. 123 (1989) 485.
P. Ginsparg, Nucl. Phys. B 295 (1988) 153.
E.B. Kiritsis, Phys. Lett. B 217 (1989) 427.
I. Runkel, G.M. Watts, Fortschr. Phys. 50 (2002) 959, hep-th/0201231.
G. Pradisi, A. Sagnotti, Phys. Lett. B 216 (1989) 59.
C. Angelantonj, A. Sagnotti, Phys. Rep. 371 (2002) 1, hep-th/0204089;
C. Angelantonj, A. Sagnotti, Phys. Rep. 376 (2003) 339, Erratum.
[8] A. Cappelli, G. DAppollonio, JHEP 0202 (2002) 039, hep-th/0201173.
[9] A.N. Schellekens, Int. J. Mod. Phys. A 14 (1999) 1283, math.QA/9806162.
472
[10] A.N. Schellekens, N. Sousa, Int. J. Mod. Phys. A 16 (2001) 3659, hep-th/0009100.
[11] J. Fuchs, L.R. Huiszoon, A.N. Schellekens, C. Schweigert, J. Walcher, Boundaries, crosscaps and simple
currents, Phys. Lett. B 495 (2000) 427, hep-th/0007174.
[12] B. Gato-Rivera, A.N. Schellekens, Complete classification of simple current modular invariants for (Z(p))k ,
Commun. Math. Phys. 145 (1992) 85;
M. Kreuzer, A.N. Schellekens, Simple currents versus orbifolds with discrete torsion: a complete classification, Nucl. Phys B 411 (1994) 97, hep-th/9306145.
[13] L.R. Huiszoon, D-branes and O-planes in string theory: an algebraic approach, PhD thesis.
[14] J. Fuchs, I. Runkel, C. Schweigert, TFT construction of RCFT correlators, III: simple currents, hepth/0403157.
[15] A.N. Schellekens, Y.S. Stanev, JHEP 0112 (2001) 012, hep-th/0108035.
[16] N. Sousa, A.N. Schellekens, Nucl. Phys. B 653 (2003) 339, hep-th/0210014.
[17] P. Di Francesco, J.-B. Zuber, Nucl. Phys. B 338 (1990) 602.
[18] G. Pradisi, A. Sagnotti, Ya.S. Stanev, Phys. Lett B 354 (1995) 279;
G. Pradisi, A. Sagnotti, Ya.S. Stanev, Phys. Lett. B 381 (1996) 97, hep-th/9603097.
[19] R. Behrend, P. Pearce, V. Petkova, J.-B. Zuber, Phys. Lett. B 444 (1998) 163, hep-th/9809097;
R. Behrend, P. Pearce, V. Petkova, J.-B. Zuber, Nucl. Phys. B 579 (2000) 707, hep-th/9908036.
[20] M. Bianchi, G. Pradisi, A. Sagnotti, Nucl. Phys. B 376 (1992) 365.
[21] A. Dabholkar, J. Park, Nucl. Phys. B 477 (1996) 701, hep-th/9604178.
[22] C. Angelantonj, M. Bianchi, G. Pradisi, A. Sagnotti, Y.S. Stanev, Phys. Lett. B 387 (1996) 743, hepth/9607229.
[23] E. Witten, JHEP 9802 (1998) 006, hep-th/9712028.
[24] J. Fuchs, C. Schweigert, Nucl. Phys. B 530 (1998) 99, hep-th/9712257.
[25] O. Bergman, E.G. Gimon, S. Sugimoto, JHEP 0105 (2001) 047, hep-th/0103183.
[26] M. Davidse, Compactifications, boundary states and tadpole equations, Masters thesis (unpublished).
[27] L.R. Huiszoon, A.N. Schellekens, N. Sousa, Nucl. Phys. B 575 (2000) 401, hep-th/9911229.
[28] J. Fuchs, C. Schweigert, Phys. Lett. B 414 (1997) 251, hep-th/9708141;
J. Fuchs, C. Schweigert, Nucl. Phys. B 558 (1999) 419, hep-th/9902132;
J. Fuchs, C. Schweigert, Nucl. Phys. B 568 (2000) 543, hep-th/9908025;
L.R. Huiszoon, A.N. Schellekens, N. Sousa, Phys. Lett. B 470 (1999) 95, hep-th/9909114;
L.R. Huiszoon, A.N. Schellekens, N. Sousa, Nucl. Phys. B 575 (2000) 401, hep-th/9911229;
L.R. Huiszoon, A.N. Schellekens, Nucl. Phys. B 584 (2000) 705, hep-th/0004100.
[29] J.L. Cardy, Boundary conditions, fusion rules and the Verlinde formula, Nucl. Phys. B 324 (1989) 581.
[30] G. Pradisi, A. Sagnotti, Y.S. Stanev, Phys. Lett. B 354 (1995) 279, hep-th/9503207.
[31] P. Di Vecchia, M. Frau, I. Pesando, S. Sciuto, A. Lerda, R. Russo, Nucl. Phys. B 507 (1997) 259, hepth/9707068.
[32] G. Felder, J. Frohlich, J. Fuchs, C. Schweigert, J. Geom. Phys. 34 (2000) 162, hep-th/9909030.
[33] L.R. Huiszoon, K. Schalm, A.N. Schellekens, Nucl. Phys. B 624 (2002) 219, hep-th/0110267.
[34] I. Brunner, On orientifolds of WZW models and their relation to geometry, hep-th/0110219.
[35] C. Bachas, N. Couchoud, P. Windey, JHEP 0112 (2001) 003, hep-th/0111002.
[36] A. Recknagel, V. Schomerus, Nucl. Phys. B 545 (1999) 233, hep-th/9811237.
[37] G. Pradisi, A. Sagnotti, Y.S. Stanev, Phys. Lett. B 381 (1996) 97, hep-th/9603097.
[38] S. Hamidi, C. Vafa, Nucl. Phys. B 279 (1987) 465.
[39] J. Fuchs, A.N. Schellekens, C. Schweigert, Nucl. Phys. B 437 (1995) 667, hep-th/9410010.
[40] T. Gannon, P. Ruelle, M.A. Walton, hep-th/9511089.
[41] G. Felder, J. Frohlich, J. Fuchs, C. Schweigert, Phys. Rev. Lett. 84 (2000) 1659, hep-th/9909140.
[42] L. Birke, J. Fuchs, C. Schweigert, Adv. Teor. Math. Phys. 3 (1999) 671, hep-th/9905038.
[43] I. Antoniadis, E. Dudas, A. Sagnotti, Phys. Lett. B 464 (1999) 38, hep-th/9908023.
[44] K. Itoh, M. Kato, H. Kunitomo, M. Sakamoto, Nucl. Phys. B 306 (1988) 362.
Construction of gauge theories on curved

noncommutative spacetime
Wolfgang Behr, Andreas Sykora
Max-Planck-Institut fr Physik, Fhringer Ring 6, 80805 Mnchen, Germany
Received 20 October 2003; received in revised form 24 May 2004; accepted 20 July 2004
Abstract
We present a method where derivations of -product algebras are used to build covariant derivatives for noncommutative gauge theory. We write down a noncommutative action by linking these
derivations to a frame field induced by a nonconstant metric. An example is given where the action
reduces in the classical limit to scalar electrodynamics on a curved background. We further use the
SeibergWitten map to extend the formalism to arbitrary gauge groups. A proof of the existence
of the SeibergWitten map for an Abelian gauge potential is given for the formality -product. We
also give explicit formulas for the Weyl-ordered -product and its SeibergWitten maps up to second
order.
PACS: 11.10.Nx; 11.15.-q; 02.40.Gh
Keywords: Noncommutative field theory; Gauge field theory; SeibergWitten map; Noncommutative geometry
1. Introduction
One hope associated with the application of noncommutative geometry in physics is a
better description of quantized gravity. At least it should be possible to construct effective
actions where traces of this unknown theory remain. If one believes that quantum gravity
E-mail addresses: behr@theorie.physik.uni-muenchen.de (W. Behr),
andreas@theorie.physik.uni-muenchen.de (A. Sykora).
doi:10.1016/j.nuclphysb.2004.07.024
474
W. Behr, A. Sykora / Nuclear Physics B 698 [PM] (2004) 473502
is in a sense a quantum field theory, then its observables are operators on a Hilbert space
and therefore elements of an algebra. Some properties of this algebra should be reflected
in the noncommutative geometry the effective actions are constructed on. As the noncommutativity should be induced by background gravitational fields, the classical limit of the
effective actions should reduce to actions on curved spacetimes [1,2].
In this paper we will investigate noncommutative geometry formulated in the -product
formalism, where gauge theory can be constructed in a particularly convenient way on
noncommutative spacetime.
The case of an algebra with constant commutator has been extensively studied. This
theory reduces in the classical limit to a theory on a flat spacetime. Therefore it is necessary
to develop concepts working with more general algebras,1 since one would expect that
curved backgrounds are related to algebras with nonconstant commutation relations. We
present here a method using derivations of -product algebras to build covariant derivatives
for noncommutative gauge theory. We are able to write down a noncommutative action by
linking these derivations to a frame field induced by a nonconstant metric. An example is
given where the action reduces in the classical limit to scalar electrodynamics on a curved
background.
Nonexpanded theories can only deal with U (n)-gauge groups, but using SeibergWitten
maps relating noncommutative quantities with their commutative counterparts makes it
possible to consider arbitrary non-Abelian gauge groups [46]. We therefore extend our
formalism to arbitrary gauge groups by introducing SeibergWitten maps. A proof of the
existence of the SeibergWitten map for an Abelian gauge potential is given for the formality -product. We also give explicit formulas for the Weyl-ordered -product and its
SeibergWitten maps up to second order.
2. The general formalism

2.1. Classical gauge theory
First let us recall some properties of a general classical gauge theory. A non-Abelian
gauge theory is based on a Lie group with Lie algebra
a b
T , T = if ab c T c .
(1)
Matter fields transform under a Lie algebra valued infinitesimal parameter
= i,
= a T a
(2)
in the fundamental representation. It follows that

( ) = i[,] .
(3)
The commutator of two consecutive infinitesimal gauge transformation closes into an infinitesimal gauge transformation. Further a Lie algebra valued gauge potential is introduced
1 As an example for the treatment of a special algebra, see also the recent paper [3], where gauge theory on
the Eq (2)-covariant plane is studied.
475
with the transformation property

ai = aia T a ,
ai = i + i[, ai ].
(4)
With this the covariant derivative of a field is

Di = i iai .
(5)
The field strength of the gauge potential is defined to be the commutator of two covariant
derivatives
iFij = [Di , Dj ] = i aj j ai i[ai , aj ].
(6)
The last equations can all be stated in the language of forms. For this a connection one
form is introduced
a = aia T a dx i .
(7)
The covariant derivative now acts as

D = d ia.
(8)
The field strength becomes a two form

F = da ia a.
(9)
2.2. Commutative actions with the frame formalism

In this section we want to recall some aspects of classical differential geometry. Suppose
we are working on a n-dimensional manifold M. Then there are locally n derivations
which form a basis of the tangent space T M of the manifold. The derivations all fulfill the
Leibniz rule on two functions. If we make a local basis transformation on T M then this
frame can always be written locally as
ea = ea (x) ,
(10)
).
has to be invertible
=
Since forms are dual to vector fields,
where ea
they may be evaluated on a frame. For the covariant derivate we get
(x)
(e a ea
(D)(ea ) = ea iaa
(11)
where
aa = a(ea ).
(12)
The field strength becomes

f (ea , eb ) = fab = ea ab eb aa a [ea , eb ] i[aa , ab ].
(13)
It is well known that in Riemannian geometry it is always possible to find a frame where
the metric is constant
ab = ea eb g .
(14)
476
Since in scalar electrodynamics we do not need a spin connection, it is simple to write

down its action on an curved manifold with the frame formalism

1
b + m2
.
S = d n x e ab cd fac fbd + ab Da D
(15)
4
Here

1
e = det ea
= det(g )
(16)
is the measure function for the curved manifold.

2.3. Noncommutative gauge theory
In our approach to studying physics in the noncommutative realm, one deforms the commutative algebra of functions on a space to a noncommutative one. This noncommutative
algebra of functions we call noncommutative space, noncommutative objects are written
with a hat. We want this deformation to be controlled by some parameter so that in some
limit we can get back a commutative space. The same we expect from theories built on a
noncommutative space: in the commutative limit they should reduce to a meaningful commutative theory. For a noncommutative space where the commutator of the coordinates is
a constant, the commutative limit is the usual gauge theory on flat spacetime. But as the
noncommutativity should be related to gravity, gauge theory on a curved spacetime seems
to be a more natural limit for theories on noncommutative spaces with more complicated,
nonconstant commutators.
In a gauge theory on a noncommutative space, fields should again transform like (2)
= i .
(17)
Again the commutator of two gauge transformations should be a gauge transformation
( ) = i[,
] .
(18)
This is only the case for U (n) gauge groups, but general gauge groups can be implemented by using SeibergWitten maps (see Section 2.5). Since multiplication of a function
with a field is not again a covariant operation, we are forced to introduce a covariantizer
with the transformation property

D(f) .
D(f) = i ,
(19)
From this it follows that

f) .
D(f) = i D(
(20)
If we covariantize the coordinate functions x i we get covariant coordinates

X i = D x i = x i + A i ,
(21)
where the gauge field now transforms according to

A i .
A i = i x i , + i ,
(22)
477
Unluckily, this does not have a meaningful commutative limit, a problem that can only
be fixed for the canonical case (i.e., [x i , x j ] = i ij with a constant) and invertible .
For noncommutative algebras where we already have derivatives with a commutative
limit, it therefore seems natural to gauge these. But due to their nontrivial coproduct the
resulting gauge field would have to be derivative-valued to match the rather awkward
behaviour under gauge transformations [7]. The physical reason for this might be the following: the noncommutative derivatives are in general built to reduce to derivatives on flat
spacetime, which might not be the correct commutative limit.
We therefore advocate a solution using derivations that will later on (see Section 2.6) be
linked to derivatives on curved spacetime:
If we have a derivation of the algebra, i.e.,
fg)
(
= ( f)g + f( g),
(23)
we can introduce a noncommutative gauge parameter A and demand that the covariant
derivative (or covariant derivation) of a field
D = ( i A )
(24)
again transforms like a field

D = i D .
(25)
From this it follows that A has to transform like

A ].
A = A + i[,
(26)
This is the transformation property we would expect a noncommutative gauge potential to

have. If we have an involution on the algebra, we can demand that the gauge potential is
real A = A and the field transforms on the right-hand side. In this case expressions of
the form

and D D
(27)
become gauge invariant quantities.

2.4. Derivations and forms
We have seen that in order to construct noncommutative gauge theory in our approach,
we need derivations on the algebra. As we want to use -products to represent the algebra,
we have to investigate the derivations of such a -product algebra. We will be able to identify derivations of -products with Poisson vector fields of the Poisson structure associated
with the -product. To be more explicit, let us assume that X is a Poisson vector field

Xi i {f, g} = Xi i f, g + f, Xi i g .
(28)
Then there exists a polydifferential operator X with the following property
X (f g) = X f g + f X g.
(29)
478
Such a map from the vector fields to the differential operators, which maps the derivations of the Poisson manifold T M = {X T M|[X, ]S = 0} to the derivations of the
-product D M = { Dpoly | [, ]G = 0}, can be constructed both for the formality product (see Section 4.1) and the Weyl-ordered -product (see Section 3.2). Here we want
to investigate the general properties of such a map . For this we expand it on a local patch
in terms of partial derivatives
ij
i
X = X
i + X i j + .
(30)
i
Due to its property to be a derivation, X is completely determined by the first term X
i .
This means that if the first term is zero, the other terms have to vanish, too. If further e is
an arbitrary derivation of the -product, there must exist a vector field Xe such that
Xe = e.
(31)
If X, Y T M, then [X , Y ] is again a derivation of the -product and we can conclude

that
[X , Y ] = [X,Y ] ,
(32)
where [X, Y ] is a deformation of the ordinary Lie bracket of vector fields. Obviously it is
linear, skew-symmetric and fulfills the Jacobi identity.
We will now introduce noncommutative forms. If we have a map we have seen that
there is a natural Lie-algebra structure (32) over the space of derivations of the -product.
On this we can easily construct the Chevalley cohomology. Further, again with the map ,
we can lift derivations of the Poisson structure to derivations of the -product. Therefore it
should be possible to pull back the Chevalley cohomology from the space of derivations to
vector fields. This will be done in the following.
A deformed k-form is defined to map k Poisson vector fields to a function and has to be
skew-symmetric and linear over C. This is a generalization of the undeformed case, where
a form has to be linear over the algebra of functions. Functions are defined to be 0-forms.
The space of forms M is now a -bimodule via
(f g)(X1 , . . . , Xk ) = f (X1 , . . . , Xk ) g.
(33)
As expected, the exterior differential is defined with the help of the map .
(X0 , . . . , Xk )
k

(1)i Xi (X0 , . . . , X i , . . . , Xk )
=
i=0

(1)i+j [Xi , Xj ] , X0 , . . . , X i , . . . , X j , . . . , Xk .
(34)
0i<j k
With the properties of and [, ] it follows that

2 = 0.
(35)
479
To be more explicit we give formulas for a function f , a one form A and a two form F
f (X) = X f,
(36)
A(X, Y ) = X AY Y AX A[X,Y ] ,
(37)
F (X, Y, Z) = X FY,Z Y FX,Z + Z FX,Y

F[X,Y ] ,Z + F[X,Z] ,Y F[Y,Z] ,X .
(38)
A wedge product may be defined

1
(I, J ) 1 (Xi1 , . . . , Xip ) 2 (Xj1 , . . . , Xjq ),
p!q!
I,J
(39)
where (I, J ) is a partition of (1, . . . , p + q) and (I, J ) is the sign of the corresponding
permutation. The wedge product is linear and associative and generalizes the bimodule
structure (33). We note that it is no more graded commutative. We again give some formulas.
1 2 (X1 , . . . , Xp+q ) =
(f a)X = f aX ,
(40)
(a f )X = aX f,
(41)
(a b)X,Y = aX bY aY bX .
(42)
The differential (34) fulfills the graded Leibniz rule

(1 2 ) = 1 2 + (1)k2 1 2 .
(43)
Now we are able to translate noncommutative gauge theory into the language of these
forms. AX is the connection one form evaluated on the vector field X. It transforms like
A = + i A A .
(44)
The covariant derivative of a field is now

D = iA ,
(45)
and the field strength becomes

F = DF = A iA A.
(46)
One easily can show that the field strength is a covariant constant
DF = F iA F = 0.
(47)
2.5. SeibergWitten gauge theory

Up to now, we could only do noncommutative gauge theory for gauge groups U (n) (see
(18)). We will now show how to implement general gauge groups by using SeibergWitten
maps [4,8].
For general gauge groups, the commutator of two noncommutative gauge transformations no longer closes into the Lie algebra. The noncommutative gauge parameter and the
480
noncommutative gauge potential will therefore have to be enveloping algebra valued, but
they will only depend on their commutative counterparts, therefore preserving the right
number of degrees of freedom. These SeibergWitten maps , and D are functionals of
their classical counterparts and additionally of the gauge potential ai . Their transformation
properties (19) and (20) should be induced by the classical ones (2) and (4) like
[a] + [a] = [a + a],
[a] + [a] = + [a + a],
(48)
+ A[a]
= A[a
+ a].
A[a]
(50)
(49)
The SeibergWitten maps can be found order by order using a -product to represent
the algebra on a space of functions. Translated into this language we get for the fields [6]
[a] = i [a] [a].
(51)
From (51) we can derive a consistency condition for the noncommutative gauge parameter [8]. Insertion into (18) and the use of (3) yields
i i + [ , ] = ii[,] .
The transformation law for the covariantizer is now

D[a](f ) = i [a] , D[a](f ) .
(52)
(53)
The SeibergWitten map can be easily extended to the derivations X of the -product. The
noncommutative covariant derivative DX [a] can be written with the help of a noncommutative gauge potential AX [a] now depending both on the commutative gauge potential a
and the vector field X
DX [a] [a] = X [a] iAX [a] [a].
It follows that the gauge potential has to transform like

AX [a] = X [a] + i [a] , AX [a] .
(54)
(55)
We will give explicit formulas for the SeibergWitten maps in Sections 3.3 and 4.2.
2.6. Gauge theory on curved noncommutative spacetime
We are now ready to formulate gauge theory on a curved noncommutative space, i.e.,
a noncommutative space with a Poisson structure that is compatible with a frame ea . Later
on we will propose a method how to find such frames commuting with the Poisson structure
in the context of quantum spaces.
With the derivation X , the covariant derivative of a field and the gauge potential now
read
DX = X iAX .
(56)
With this, a field strength may be defined as

iFX,Y = [DX , DY ] D[X,Y ] .
(57)
481
The properties of and [ , ] ensure that this is really a function and not a polydifferential
operator.
For a curved noncommutative space, we can now evaluate the noncommutative covariant derivative (56) and field strength (57) on the frame ea
Da = Dea = ea iAea ,
(58)
Fab = F (ea , eb ).
(59)
To write down a gauge invariant action we further need a trace, i.e., a functional from
the algebra to the complex numbers. Again the -product will be a useful tool. For a large
class of -products there exist measure functions so that

f g = d n x (f g)
(60)
and

f g=
g f.
(61)
Obviously up to first order has to fulfill [9]

= 0.
(62)
It is known [10] that there is always a -product for which this equation holds up to all
orders.
Using the measure function and our noncommutative versions of field strength and covariant derivative we end up with the following action

1
S = d n x ab cd Fac Fbd + ab Da Db m2 .
(63)
4
By construction this action is invariant under noncommutative gauge transformations
S = 0.
Its classical limit is

1
m2
,
S d n x g g f f + g D D
4
(64)
(65)
with g the metric induced by the frame. In our example (see Sections 2.7 and 3.4), we
will further have = g and the interpretation of the classical limit is obvious.
We will now propose a method how to find Poisson structures and compatible frames.
On several quantum spaces deformed derivations have been constructed [1113]. In most
cases the deformed Leibniz rule may be written in the following form
(fg)
= fg + T (f) g,
(66)
where T is an algebra morphism from the quantum space to its matrix ring
= T (f)T (g).
T (fg)
(67)
482
Again in some cases it is possible to implement this morphism with some kind of inner
morphism
T (f) = e a fea ,
(68)
where ea is an invertible matrix with entries from the quantum space. If we define
ea = ea ,
(69)
the ea are derivations

ea (fg)
= ea (f)g + fea (g).
(70)
The dual formulation of this with covariant differential calculi on quantum spaces is the
formalism with commuting frames investigated for example in [1417]. There one can
additionally find how our formalism fits into the language of Connes spectral triples.
We can now represent the quantum space with the help of a -product. For example
we can use the Weyl-ordered -product we will construct in Section 3.1. Further we can
calculate the action of the operators ea on functions. Since these are now derivations of a
-product, there necessarily exist Poisson vector fields ea with
ea = ea .
(71)
2.7. Example: SOa (n)

In this section will examine a quantum space introduced in [18]. Since we are using a
n-dimensional generalization we will simply call it SOa (n) covariant quantum space. The
relations of this quantum space are
0 i
x , x = ia x i for i = 0,
(72)
with a a real number. In the following of the example Greek indices will run from 0 to
n 1, whereas Latin indices will run from 1 to n 1. The deformed derivations commute
and act like
o x 0 = 1 + x 0 o ,
o x i = x i o ,
j
i x j = i + x j i ,

i x 0 = x 0 + ia i .

i 2
If we define =
i (x ) and assume that it is invertible then
eo = 0 ,
ei = i
(73)
(74)
is a frame on the quantum space. The classical limit of this is obviously

eo = 0 ,
ei = i
(75)
483
and the classical metric becomes

2

2
2

g = dx 0 + 2 dx 1 + + dx n1 .
(76)
We know that we can write

2

1 2
2
+ + dx n1 = d 2 + 2 dn2
,
dx
(77)
2
where dn2
is the metric of the (n 2)-dimensional sphere. Therefore in this new coordinate system

2
2
g = dx 0 + (d ln )2 + dn2
(78)
and we see that the classical space is a cross product of two dimensional Euclidean space
and a (n 2)-sphere. Therefore it is a space of constant nonvanishing curvature. Further
the measure function is

det g = (n1) .
(79)
We will continue this example at the end of Section 3, where we will have explicit
formulas for the -product.
3. Weyl-ordered -product
To pursue our investigation further, we will have to use a specific -product and construct explicit expressions for the terms that enter into the action (63).
For the case of constant Poisson structure, one usually uses the MoyalWeyl -product,
corresponding to symmetric ordering of the generators of the noncommutative algebra.
This procedure of generating a -product by an ordering prescription can be applied to
more general algebras, too.
In this section we will therefore use a -product generated by symmetric ordering of
the generators of a noncommutative algebra, the Weyl-ordered -product. The algebra of
functions equipped with the Weyl-ordered -product is isomorphic by construction to the
noncommutative algebra it is based on.
We will present a general formula for the Weyl-ordered -product up to second order.
We will then calculate the derivations to this -product and the SeibergWitten maps for
all relevant quantities.
3.1. Construction
We start with an algebra generated by N elements x i and relations
i j
x , x = cij (x).
(80)
For such an algebra we will calculate a -product up to second order. Let

i
f (p) = d n x f (x)eipi x
(81)
484
be the Fourier transform of f . Then the Weyl-ordered operator associated to f is defined

by

d np
i
f (p)eipi x
W (f ) =
(82)
n
(2)
(see, e.g., [5]). Every monomial of coordinate functions is mapped to the corresponding
Weyl-ordered monomial of the algebra. We note that

i
i
W eiqi x = eiqi x .
(83)
The Weyl-ordered -product is defined by the equation
W (f g) = W (f )W (g).
If we insert the Fourier transforms of f and g we get

d nk
d np
i
i
f g =
f (k)g(p)W 1 eiki x eipi x .
n
n
(2)
(2)
(84)
(85)
We are therefore able to write down the -product of the two functions if we know the form
of the last expression. For this we expand it in terms of commutators. We use

B)
eA eB = eA+B R(A,
(86)

B]
1 A + 2B,
B][
A,
B]
+ O(3).
[A,
B]
+ 1 [A,
B)
= 1 + 1 [A,
R(A,
2
6
8
(87)
with
If we set A = iki x i and B = ipi x i the above-mentioned expression becomes

i
i
W 1 eiki x eipi x

1
i
i
= ei(ki +pi )x + (iki )(ipj )W 1 ei(ki +pi )x x i , x j
2

1
i
(i)(km + 2pm )(iki )(ipj )W 1 ei(ki +pi )x x m , x i , x j
6

1
i
+ (ikm )(ipn )(iki )(ipj )W 1 ei(ki +pi )x x m , x n x i , x j
8
+ O(3).
(88)
If we assume that the commutators of the generators are written in Weyl-ordered form

cij = W cij ,
(89)
we see that

m i j
= W cml l cij + O(3),
x , x , x
m n i j

x , x x , x = W cmn cij + O(3).
(90)
(91)
Further we can derive

i
W 1 eiqi x W (f )

d np
1
i(qi +pi )x i
i
i
=W
f (p)e
R iqi x , ipi x
(2)n

1
i
= eiqi x f + (iqi )cij j f + O(2).
2
Putting all this together yields

i
i
W 1 eiki x eipi x

1
i
= ei(ki +pi )x 1 + cij (iki )(ipj )
2
1 mn ij
+ c c (ikm )(ipn )(iki )(ipj )
8

1 ml ij
+ c l c (i)(km pm )(iki )(ipj )
12
+ O(3),
485
(92)
(93)
and we can write down the Weyl-ordered -product up to second order for an arbitrary
algebra
1
1
f g = fg + cij i f j g + cmn cij m i f n j g
2
8
1 ml ij
+ c l c (m i f j g i f m j g) + O(3).
12
Let us collect some properties of the just calculated -product. First
i j
x , x = cij
(94)
(95)
is the Weyl-ordered commutator of the algebra. Further, if there is a conjugation on the

algebra and if we assume that the noncommutative coordinates are real x i = x i , then the
Weyl-ordered monomials are real, too. This is also true for the monomials of the commutative coordinate functions. Therefore this -product respects the ordinary complex
conjugation
f g = g f.
(96)
On the level of the Poisson tensor this means

cij = cij .
If we have a measure function with i (cij ) = 0, then

n
d x f g = d n x g f + O(3).
(97)
(98)
486
3.2. Derivations
We now want to calculate the derivations X of the Weyl-ordered -product (94) from
the derivations X of the Poisson structure cij up to second order. We assume that X can
be expanded in the following way
ij
ij k
X = Xi i + X i j + X i j k + .
(99)
Expanding the equation

X (f g) = X (f ) g + f X (g)
(100)
order by order and using [X, c]S = 0 we find that

1 lk im
1
c k c l m Xj i j + clk cim l i Xj k m j + O(3).
12
24
For [ , ] we simply calculate [X , Y ] and get
X = Xi i
(101)

1 lk im
c k c l m Xj i j Y n clk k cim l m Y j i j Xn n
12

1 lk im
+
c c l i Xj k m j Y n clk cim l i Y j k m j Xn n
24
+ O(3).
(102)
[X, Y ] = [X, Y ]L
3.3. Explicit formulas for the SeibergWitten map

We will now present a consistent solution for the SeibergWitten maps up to second
order for the Weyl-ordered -product and non-Abelian classical gauge transformations.
The solutions have been chosen in such a way that they reproduce the ones obtained in [6]
for the constant case.
The solution for the gauge transformations is obtained by solving the consistency condition (52) order by order
i
[a] = cij {i , aj }
4
1 ij kl
+ c c 4 i , {ak , l aj } 2i[i k , j al ]
32

+ 2 j al , [i , ak ] 2i [aj , al ], [i , ak ]

+ i i , ak , [aj , al ] + aj , al , [i , ak ]

1
+ ckl l cij i , {ak , aj } 2i[i k , aj ] + O(3).
24
In the same way a solution for the field is obtained by solving Eq. (51)
1
[a] = + cij (2iai j + ai aj )
4
1
+ cij ckl (4ii ak j l 4ai ak j l 8ai j ak l
32
(103)
487
+ 4ai k aj l + 4iai aj ak l 4iak aj ai l

+ 4iaj ak ai l 4j ak ai l + 2i ak j al
4iai al k aj 4iai k aj al + 4iai j ak al
3ai aj al ak 4ai ak aj al 2ai al ak aj )
1
+ ckl l cij (2iaj k i + 2ik ai j + 2k ai aj
24
ak ai j 3ai ak j 2iaj ak ai ) + O(3).
(104)
From (53) the covariantizer becomes

D[a](f ) = f + icij ai j f

1
+ cij ckl 2{ai , j ak }l f + {ai , k aj }l f
4

+ i ai , [aj , ak ] l f {ai , ak }j l f
1
+ cil l cj k {ai , ak }j f + O(3).
4
(105)
With (55) the NC gauge potential is

i
i
AX = Xn an + ckl Xn {ak , l an + fln } + ckl l Xn {ak , an }
4
4

1 kl ij n
+ c c X 4i[k i an , l aj ] + 2i[k n ai , l aj ] 4 ak , {ai , j fln }
32

2 [k ai , an ], l aj + 4 l an , {i ak , aj } 4 ak , {fli , fj n }

+ i n aj , al , [ai , ak ] + i ai , ak , [n aj , al ] 4i [ai , al ], [ak , j an ]

+ 2i [ai , al ], [ak , n aj ] + ai , ak , al , [aj , an ]

ak , [al , ai ], [aj , an ] [ai , al ], ak , [aj , an ]

1
+ ckl cij j Xn 2i[k ai , l an ] + 2i[i ak , l an ] + 2i[i ak , l an n al ]
32

+ 4 an , {al , k ai } + 4 ak , {ai , n al l an } 2i ak , ai , [an , al ]

+ i ai , al , [an , ak ] + i an , al , [ai , ak ]

1
+ ckl cij l j Xn i k an 2i[ai , k an ] an , {ak , ai }
24

1
+ ckl l cij Xn 2i[aj , k i an ] + 2i[k ai , fj n ]
24

j an , {ak , ai } + 2 ai , {ak , fnj }

1
+ ckl l cij j Xn 4i[ai , k an ] + 2i[ak , i an ] an , {ak , ai }
24
1
ckl l cij j k Xn i an + O(3).
(106)
12
488
The resulting field strength is

Fab = F (Xa , Xb )

i
i
= Xak Xbl fkl + cij ai , j Xak Xbl fkl + cij Xak Xbl {fj l , fki }
2
2
1 ij k l
+ c Xa Xb ai , [aj , fkl ] + O(2).
4
The covariant derivative is
(107)
Da = DXa = Xa iAXa
i
= Xak Dk + Xak fki cij Dj
2

1
i ij
+ c ai j Xak Dk + cij ai aj Xak Dk + O(2).
(108)
2
4
Using partial integration and the trace property of the integral, i.e., (c ) = 0, we can
calculate

Sgauge = d n x ab cd Fac Fbd

= d n x ab cd Xa Xc Xb Xd f f

i ij
c ai , j Xa Xc f , Xb Xd f
+ d 4 x ab cd
4
1
i ij
+ c Xa Xc Xb Xd f , {fij , f } cij Xa Xc Xb Xd [f ai , f aj ]
8
4
1
cij Xa Xc Xb Xd [ai f , aj f ]
4

+ cij Xa Xc Xb Xd f , {fj , fi } + O(2).

(109)
2
Therefore the action for the gauge particles is
1
Sgauge = tr(Sgauge )

4
1
n
ab cd
= d x Xa Xc Xb Xd tr(f f )
4

i ij
i ij
c tr(fij f f ) c tr(f fj fi ) + O(2).
8
2
With cij = cij we get

Sscalar = d n x ab Da Db

i
= d n x ab X a Xb D D + cij X a Xb D fi Dj
2

i ij
i ij
+ c Xa Xb Dj fi D + c Xa Xb D fij D + O(2)
2
2
for the scalar fields.
(110)
(111)
489
3.4. Example: SOa (n)

Now we continue our example from Section 2.7. It is easy to see that the Poisson tensor
corresponding to the algebra is
c = ia0 i x i ia0 i x i .
(112)
Since we are dealing here with the case of a Lie algebra we surely have W (c ) = [x , x ].
The x i commute with each other, therefore we get with =
xi x i
W () = .
The components of the frame
X0
Xi
(113)
(e = X u )
are
= 0 ,
= i .
These we can plug into our solution of the SeibergWitten map and get

a
a
[a] = + x i {0 , ai } x i {i , a0 } + O a 2 ,
4
4

a
a
ia
[a] = x i a0 i + x i ai 0 + x i [a0 , ai ] + O a 2 ,
2
2
4

a
a
AX0 = a0 x i {a0 , i a0 + fi0 } + x i {ai , 0 a0 } + O a 2 ,
4
4
a
a i
AXj = aj {aj , a0 } x {a0 , i aj + fij }
4
4

a i
+ x {ai , 0 aj + f0j } + O a 2 ,
4

X = X + O a 2 .
(114)
(115)
A measure function induced by the trace property of the integral up to second order is
= (n1) .
We note that
= g,
(116)
(117)
where g is the classical metric induced by the noncommutative frame (see Section 2).
With this measure function the actions become

1
d n x 3n 00 ij Tr(f0i f0j )
Sgauge =
2

1
d 4 x 5n kl ij Tr(fki flj )
4

a
d n x 3n 00 ij x p Tr(f0p f0i f0j )

2

a
+
d 4 x 5n kl ij x p Tr(f0p fki flj )
4

a
(118)
d n x 5n kl ij x p Tr fjp {fki , fl0 } + O a 2
2
490
and

Sscalar =
d x
D0 D0 + d n x 3n kl Dk Dl

a
d n x 3n kl x i Dk fl0 Di
2

a
+
d n x 3n kl x i Dk fli D0
2

a
a
d n x 3n kl x i Di fl0 Dk +
d n x 3n kl x i D0 fli Dk
2
2

a d n x 3n kl x i Dk f0i Dl + O a 2 .
(119)
n
1n 00
In the classical limit a 0 the action reduces to scalar electrodynamics on a curved background or its non-Abelian generalization, respectively.
4. Formality
The Weyl-ordered -product of the last section is very useful for explicit calculations,
but these can only be done in a perturbative way order by order. Also, it is only known
in general up to the second order we calculated in Section 3.1. For closed expressions
and questions of existence, Kontsevichs formality -product [19] is the better choice. It
is known to all orders and comes with a strong mathematical framework, in which the
derivations are obtained in a very natural way.
4.1. The formality map
Kontsevichs formality map [19] is a very useful tool for studying the relations between
Poisson tensors and -products. To make use of the formality map we first want to recall
some definitions. A polyvector field is a skew-symmetric tensor in the sense of differential
geometry. Every n-polyvector field may locally be written as
= i1 ...in i1 in .
(120)
We see that the space of polyvector fields can be endowed with a grading n. For polyvector
fields there is a grading respecting bracket that in a natural way generalizes the Lie bracket
[ , ]L of two vector fields, the SchoutenNijenhuis bracket. For an exact definition see
Appendix A.1. If is a Poisson tensor, the Hamiltonian vector field Hf for a function f
is
Hf = [, f ]S = ij i f j .
(121)
Note that [, ]S = 0 is the Jakobi identity of a Poisson tensor.

On the other hand a n-polydifferential operator is a multilinear map that maps n functions to a function. For example, we may write a 1-polydifferential operator D as
ij
D(f ) = D0 f + D1i i f + D2 i j f + .
(122)
491
The ordinary multiplication is a 2-polydifferential operator. It maps two functions to one

function. Again the number n is a grading on the space of polydifferential operators. Now
the Gerstenhaber bracket is natural and respects the grading. For a exact definition see
Appendix A.2.
The formality map is a collection of skew-symmetric multilinear maps Un , n =
0, 1, . . . , that maps n polyvector fields to a m-differential operator. To be more specific
let 1 , . . . , n be polyvector fields of grade k1 , . . . , kn . Then Un (1 , . . . , n ) is a polydifferential operator of grade

ki .
m = 2 2n +
(123)
i
In particular the map U1 is a map from a k-vector field to a k-differential operator. It is

defined by

U1 i1 ...in i1 in (f1 , . . . , fn ) = i1 ...in i1 f1 in fn .
(124)
The formality maps Un fulfill the formality condition [19,20]

1
(I, J )Q2 U|I | (I ), U|J | (J )
Q1 Un (1 , . . . , n ) +
2
(125)
I J ={1,...,n}
I,J =
1
. . . , j, . . . , n)
(i, j, . . . , i,
2
i=j

Un1 Q2 (i , j ), 1 , . . . , i , . . . , j , . . . , n .
(126)
hats stand for omitted symbols, Q1 ( ) = [, ] with being

Q2 (1 , 2 ) = (1)(|1|1)|2 | [1 , 2 ]G with |s | being the
The
ordinary multiplication
and
degree of the polydifferential operator s , i.e., the number of functions it is acting on. For polyvector fields
i ...i
s1 ks i1 iks of degree ks we have Q2 (1 , 2 ) = (1)(k1 1)k2 [2 , 1 ]S .
For a bivector field we can now define a bidifferential operator
=

1
Un (, . . . , ),
n!
(127)
n=0
i.e.,
f g =

1
Un (, . . . , )(f, g).
n!
(128)
n=0
We further define the special polydifferential operators

() =

n=1
1
Un (, , . . . , ),
(n 1)!
(1 , 2 ) =

n=2
1
Un (1 , 2 , , . . . , ).
(n 2)!
(129)
(130)
492
For g a function, X and Y vector fields and a bivector field we see that
X = (X)
is a 1-differential operator and that both (g) and (X, Y ) are functions.
We now use the formality condition (125) to calculate

[, ]G = [, ]S ,

(f ), G = [f, ]S ,

[X , ]G = [X, ]S ,

[X , Y ]G + (X, Y ), G = [X,Y ]S + [, Y ]S , X [, X]S , Y ,

(), (g) G + (, g), G

= [,g]S [, g]S , [, ]S , g ,

X , (g) G = [X, g]S [, g]S , X [, X]S , g .
(131)
(132)
(133)
(134)
(135)
(136)
(137)
If is Poisson, i.e., [, ]S = 0 and if X and Y are Poisson vector fields, i.e., [X, ]S =
[Y, ]S = 0, the relations (132) to (135) become
f (g h) = (f g) h,

Hf (g) = (f ) , g ,
X (f g) = X (f ) g + f X (g),

[X , Y ] [X,Y ]L (g) = (X, Y ) , g ,
(138)
when evaluated on functions. [ , ] are now ordinary brackets. defines an associative

product, the Hamiltonian vector fields are mapped to inner derivations and Poisson vector
fields are mapped to outer derivations of the -product. Note that in [21] nonassociative
-products arised when curved D-branes were considered on curved backgrounds.
Additionally the map preserves the bracket up to an inner derivation. The last equation
can be cast into a form we used extensively in the definition of our deformed forms:
[X , Y ] = [X,Y ]
(139)
[X, Y ] = [X, Y ]L + H 1 (X,Y ) .
(140)
with
4.2. Construction of the SeibergWitten map

With the formality -product and the derivations on it we have all the key ingredients
to do NC gauge theory on any Poisson manifold. To relate the NC theory to commutative
gauge theory, we need the SeibergWitten maps for the formality -product. In [22] and
[23] the SW maps for the NC gauge parameter and the covariantizer were already constructed to all orders in . We will extend the method developed there to the SW map for
covariant derivations.
493
4.2.1. Semi-classical construction

We will first do the construction in the semi-classical limit, where the star commutator
is replaced by the Poisson bracket. As in [22,23], we define, with the help of the Poisson
tensor = 12 kl k l
d = [ , ]
(141)
and (locally)
a = ij aj i .
(142)
Note that the bracket used in the definition of d is not the SchoutenNijenhuis bracket
(Appendix A.1). For polyvector fields 1 and 2 it is
[1 , 2 ] = [2 , 1 ]S ,
(143)
giving an extra minus sign for 1 and 2 both even (see Appendix B.3.2). Especially, we
get for d acting on a function g
d g = [g, ] = [g, ]S = kl l gk .
(144)
Now a parameter t and t-dependent t = 12 tkl k l and Xt = Xtk k are introduced, fulfilling
t t = f = t f t
and t Xt = Xt f t ,
(145)
where the multiplication is ordinary matrix multiplication. Given the Poisson tensor and
the Poisson vector field X, the formal solutions are
t =

1
(tf )n = kl t ki fij j l + k l
2
(146)
n=0
and
Xt = X
(tf )n = Xk k tXi fij j k k + .
(147)
n=0
t is still a Poisson tensor and Xt is still a Poisson vector field, i.e.,

[t , t ] = 0 and [Xt , t ] = 0.
(148)
For the proof see Appendix B.1.

With this we calculate
f = t t = t f t = [a , ] = d a .
We now get the following commutation relations

at + t , dt (g) = dt (at + t )(g) ,

[at + t , Xt ] = dt Xtk ak ,
where g is some function which might also depend on t (see Appendix B.3.1).
(149)
(150)
(151)
494
To construct the SeibergWitten map for the gauge potential AX , we first define
Kt =

n=0
1
(a + t )n .
(n + 1)! t
(152)
With this, the semi-classical gauge parameter reads [22,23]

[a] = Kt ()|t =0 .
(153)
To see that this has indeed the right transformation properties under gauge transformations,
we first note that the transformation properties of at and Xtk ak are
at = tkl l k = dt
(154)

Xtk ak = Xtk k = [Xt , ].
(155)
and
Using (154), (155) and the commutation relations (150), (151), a rather tedious calculation
(see Appendix B.2) shows that

Kt Xtk ak = Xtk k Kt () + dt Kt () Kt Xtk ak .
(156)
Therefore, the semi-classical gauge potential is

AX [a] = Kt Xtk ak t =0 .
(157)
4.2.2. Quantum construction

We can now use the Kontsevich formality map to quantize the semi-classical construction. All the semi-classical expressions can be mapped to their counterparts in the -product
formalism without loosing the properties necessary for the construction. One higher order
term will appear, fixing the transformation properties for the quantum objects.
The star-product we will use is
=

1
Un (t , . . . , t ).
n!
(158)
n=0
We define
d = [ , ]G,
(159)
which for functions f and g reads

d (g)f = [f , g].
(160)
The bracket used in the definition of d is the Gerstenhaber bracket (Appendix A.2). We
now calculate the commutators (150) and (151) in the new setting (see Appendix B.3.2).
We get

(at ) + t , d (g) = d (at ) + t (f ) ,
(161)

k
(at ) + t , (Xt ) = d Xt ak (at , Xt ) .
(162)
495
The higher order term (at , Xt ) has appeared, but looking at the gauge transformation
properties of the quantum objects we see that it is actually necessary. We get
(at ) = (dt ) = d ()
(163)
with (138) and (154) and

Xtk ak (a , Xt )

= [Xt , ] (d , Xt )

= (Xt ), () [t , ], Xt + [t , Xt ], (d , Xt )

= (Xt ), () = Xt (),
(164)
where the addition of the new term preserves the correct transformation property. With
Kt =

n=0
n
1
(at ) + t ,
(n + 1)!
(165)
a calculation analogous to the semi-classical case gives

Kt Xtk ak (at , Xt )

= Xt Kt () + d Kt () Kt Xtk ak (at , Xt ) .
(166)
As in [22,23], the NC gauge parameter is

[a] = Kt () t =0 ,
(167)
and we therefore get for the NC gauge potential

AX [a] = Kt Xtk ak (at , Xt ) t =0 ,
(168)
transforming with
AX = X [ , AX ].
(169)
Acknowledgements
The authors want to thank B. Jurco and J. Wess for many interesting discussions. We
also want to thank the MPI and the LMU for their support.
Appendix A. Definitions
A.1. The SchoutenNijenhuis bracket
i ...iks
The SchoutenNijenhuis bracket for multivector fields s1

written as ([20], IV.2.1)
[1 , 2 ]S = (1)k1 1 1 2 (1)k1 (k2 1) 2 1 ,
i1 iks can be
(A.1)
496
1 2 =
k1

i1 ...ik1
(1)l1 1
j1 ...jk2
l 2
i1 il ik1
l=1
j1 jk2 ,
(A.2)
where the hat marks an omitted derivative.

For a function g, vector fields X = Xk k and Y = Y k k and a bivector field =
1 kl
2 k l we get
[X, g]S = Xk k g,
(A.3)
[, g]S = k gl ,

1
[X, ]S = Xk k ij ik k Xj + j k k Xi i j ,
2

1 kl
[, ]S = l ij + il l j k + j l l ki k i j .
3
(A.4)
kl
(A.5)
(A.6)
A.2. The Gerstenhaber bracket

The Gerstenhaber bracket for polydifferential operators As can be written as ([20], IV.3)
[A1 , A2 ]G = A1 A2 (1)(|A1 |1)(|A2 |1) A2 A1 ,
(A.7)
(A1 A2 )(f1 , . . . , fm1 +m2 1 )

=
m1

(1)(m2 1)(j 1)A1 f1 , . . . , fj 1 , A2 (fj , . . . , fj +m2 1 ),

j =1
fj +m2 , . . . , fm1 +m2 1 ,
(A.8)
where |As | is the degree of the polydifferential operator As , i.e., the number of functions
it is acting on.
For functions g and f , differential operators D1 and D2 of degree one and P of degree
two we get
[D, g]G = D(g),
[P , g]G (f ) = P (g, f ) P (f, g),

[D1 , D2 ]G (g) = D1 D2 (g) D2 D1 (g) ,

[P , D]G (f, g) = P D(f ), g + P f, D(g) D P (f, g) .
(A.9)
Appendix B. Calculations
B.1. Calculation of [t , t ] and [t , Xt ]
We want to show that t is still a Poisson tensor and that Xt still commutes with t .
For this we first define (n)kl = (f )n = ki fij rs fsl = fli ij frs sk = (f )n and
(n)kl = (f )n = ki fij frs sl . In the calculations to follow we will sometimes drop
497
the derivatives of the polyvector fields and associate k1 ...kn with k1 ...kn n1 k1 kn for
simplicity. All the calculations are done locally.
We evaluate
ij
[t , t ]S = tkl l t + c.p. in (kij )

m

j
=
(t)n+m (n)kr (o)is (m o)p rl l sp + c.p. in (kij )
n,m=0 o=0

m

(t)n+m+1 (n)kl (o)is (m o)pj l fsp + c.p. in (kij )
n,m=0 o=0
(t)n+m+o (n)kr (o)is (m)p rl l sp + c.p. in (kij )
n,m,o=0
(t)n+m+o+1 (n)kl (o)is (m)jp l fsp + c.p. in (kij ).
(B.1)
n,m,o=0
The first part vanishes because t is a Poisson tensor, i.e.,

[, ]S = kl l ij + c.p. in (kij ) = 0,
(B.2)
the second part because of

k fij + c.p. in (kij ) = 0.
(B.3)
To prove that Xt still commutes with t , we first note that

Xt = X
(tf ) = X(1 tf t ).
(B.4)
n=0
With this we can write

[Xt , t ] = [X, t ] t[Xf t , t ]
= Xn n tkl tkn n Xl + tln n Xk tXm fmi tin n tkl

+ ttkn n Xm fmi til ttln n Xm fmi tik
= Xn n tkl tkn n Xl + tln n Xk
+ ttkn n Xm fmi til ttln n Xm fmi tik
+ ttkn Xm n fmi til ttln Xm n fmi tik .
(B.5)
In the last step we used (B.2). To go on we note that

ttkn Xm n fmi til ttln Xm n fmi tik = tXn tkm n fmi til ,
(B.6)
where we used (B.3). Making use of the power series expansion and the fact that X commutes with , i.e.,
[X, ] = Xn n kl kn n Xl + ln n Xk = 0,
(B.7)
498
we further get
Xn n tkl + tXn tkm n fmi til

=
(t)r+s (r)ki Xn n ij (s)lj
=
r,s=0
r,s=0
r,s=0
(t)r+s (r)ki in n Xj (s)lj
(t)r+s (r)ki j n n Xi (s)lj .
(B.8)
Therefore (B.5) reads

[Xt , t ] =
(t)r+s (r)ki (s)lj in n Xj
r,s=0
tkn n Xl
+ tln n Xk
(t)r+s (r)ki (s)lj j n n Xi
r,s=0
kn
m
+ tt n X fmi til
ttln n Xm fmi tik
= 0.
(B.9)
B.2. The transformation properties of Kt

To calculate the transformation properties of Kt (Xtk ak ), we first evaluate

(a + t )n Xk ak
=
n1

(a + t )i d ()(a + t )n1i Xk ak
i=0
i

i
d (a + t )l () (a + t )n1l Xk ak
l
n1

i=0 l=0
and

(a + t )n Xk ak
= (a + t )n Xk k
= X k (a + t )
k
n1

(a + t )i d Xk ak (a + t )n1i
i=0
= Xk k (a + t )n
n1 n1i

n 1 i
(1)n1ij (a + t )i+j
j
i=0 j =0

d (a + t )n1ij Xk ak ()
= Xk k (a + t )n
n1 n1i

n 1 i
+
(1)n1ij (a + t )i+j
j
i=0
j =0
(B.10)
499

d () (a + t )n1ij Xk ak
= Xk k (a + t )n

i+j
n1 n1i

n1i i+j
(1)n1ij
j
l
i=0 j =0 l=0

d (a + t )l () (a + t )n1l Xk ak .
We go on by simplifying these expressions. Using

i
i 1
i 1
=
+
for i > l
l
l
l1
(B.11)
(B.12)
we get

n1
n1
m

n1i m
n m
n1m
(1)
(1)n1m .
=
mi
l
m
l
m=l i=0
(B.13)
m=l
Using (B.12) again two times and then using induction we go on to

n1
l

n m
n1i
n1m
=
,
(1)
m
n1l
l
m=l
(B.14)
i=0
giving, after using (B.12) again

l

n1i
i=0
n1l

n
.
l
(B.15)
Together with
n1

i
i=l
n
=
l
l +1

(B.16)
these formulas add up to give

m
n1
n1

n1i m
i
n+1
(1)n1m +
=
mi
l
l
l+1
m=l i=0
(B.17)
i=l
and therefore

Kt Xk ak = Xk k Kt () + d Kt () Kt Xk ak .
(B.18)
B.3. Calculation of the commutators

B.3.1. Semi-classical construction
We calculate the commutator (150) (see also [23]), dropping the t-subscripts on t for
simplicity and using local expressions:
500

a , d (g) = ij aj i kl k gl ij aj kl i k gl + kl k gl ij aj i
+ kl k g ij l aj i
= kl k ij aj i gl kl ij aj k i gl kl ij j ak i gl
= + ij fj k kl i gl kl k ( ij aj i g)l

= df g + d a (g)

= t (d )g + d a (g) .
(B.19)
For (151) we get

[a , Xt ] = ij aj i Xk k Xk k ij aj i Xk ij k aj i
= ij Xk k aj i ik k Xj aj i

= Xk fki ij j + ij i Xk ak j

= t X d Xk ak .
(B.20)
B.3.2. Quantum construction

In [24], (133), (134), (137) have already been calculated, unluckily (and implicitly)
using a different sign convention for the brackets of polyvector fields. In [23], again a
different sign convention is used, coinciding with the one in [24] in the relevant cases. In
order to keep our formulas consistent with the ones used in [23,24], we define our bracket
on polyvector fields 1 and 2 as in [24] to be
[1 , 2 ] = [2 , 1 ]S ,
(B.21)
giving an extra minus sign for 1 and 2 both even. The bracket on polydifferential operators is always the Gerstenhaber bracket.
With these conventions and
d = [ , ]
(B.22)
we rewrite the formulas (137), (135), (133), (134) so we can use them in the following

(X), (g) G = [X, g] + [, g], X [, X], g ,
(B.23)

(X), (Y ) G = d (X, Y ) + [X, Y ] + [, Y ], X

(B.24)
[, X], Y ,

d (g) = d (g) ,
(B.25)

d (X) = d (X) .
(B.26)
For the calculation of the commutators of the quantum objects we first define
a = (at )
(B.27)
f = (ft ).
(B.28)
and
501
With (B.26) we get the quantum version of (149)

f = d a .
(B.29)
For functions f and g we get

t (f g) =
=

1
t Un (t , . . . , t )(f, g)
n!
n=0
1
Un (f , . . . , t )(f, g) = f (f, g).
(n 1)!
n=1
(B.30)
With these two formulas we can now calculate the quantum version of (150) as in [23]. On
two functions f and g we have
t (f g) = f (f, g)
= d a (f, g)
= [a , ](f, g)
= a (f g) + a (f ) g + f a (g),
where we used (A.9) in the last step. Therefore

a , d (g) (f ) = a d (g)(f ) d (g) a (f )

= a [f , g] a (f ) , g

= t f , g a (g) , f

= t d (g)(f ) + d a (g) (f ).
(B.31)
(B.32)
For a function g which might also depend on t the quantum version of (150) now reads

a + t , d (g) = d a (g) .
(B.33)
We go on to calculate the quantum version of (151). We first note that
t (Xt ) =

n=1
1
t Un (Xt , t , . . . , t ) = (t Xt ) + (f , Xt ).
(n 1)!
(B.34)
With this we get

(a ), (Xt )

= d (a , Xt ) + [a , Xt ] [t a ] + [t , Xt ], a

= d (a , Xt ) + d Xtk ak + (t Xt ) (f , Xt )

= d Xtk ak (a , Xt ) t (Xt ),
where we have used (B.24).
(B.35)
502
References
[1] J. Madore, Gravity on fuzzy spacetime, gr-qc/9709002.
[2] A. Connes, Gravity coupled with matter and the foundation of non-commutative geometry, Commun. Math.
Phys. 182 (1996) 155, hep-th/9603053.
[3] F. Meyer, H. Steinacker, Gauge field theory on the Eq (2)-covariant plane, hep-th/0309053.
[4] N. Seiberg, E. Witten, String theory and noncommutative geometry, JHEP 9909 (1999) 032, hep-th/9908142.
[5] J. Madore, S. Schraml, P. Schupp, J. Wess, Gauge theory on noncommutative spaces, Eur. Phys. J. C 16
(2000) 161, hep-th/0001203.
[6] B. Jurco, L. Moller, S. Schraml, P. Schupp, J. Wess, Construction of non-Abelian gauge theories on noncommutative spaces, Eur. Phys. J. C 21 (2001) 383, hep-th/0104153.
[7] M. Dimitrijevic, F. Meyer, L. Moller, J. Wess, Gauge theories on the kappa-Minkowski spacetime, hepth/0310116.
[8] B. Jurco, S. Schraml, P. Schupp, J. Wess, Enveloping algebra valued gauge transformations for non-Abelian
gauge groups on non-commutative spaces, Eur. Phys. J. C 17 (2000) 521, hep-th/0006246.
[9] X. Calmet, M. Wohlgenannt, Effective field theories on non-commutative spacetime, Phys. Rev. D 68
(2003) 025016, hep-ph/0305027.
[10] G. Felder, B. Shoikhet, Deformation quantization with traces, math.QA/0002057.
[11] J. Wess, B. Zumino, Covariant differential calculus on the quantum hyperplane, Nucl. Phys. B (Proc.
Suppl.) 18 (1991) 302.
[12] A. Lorek, W. Weich, J. Wess, Non-commutative Euclidean and Minkowski structures, Z. Phys. C 76 (1997)
375.
[13] B.L. Cerchiai, R. Hinterding, J. Madore, J. Wess, The geometry of a q-deformed phase space, Eur. J. Phys.
C 8 (1999) 533, math.QA/9807123.
[14] A. Dimakis, J. Madore, Differential calculi and linear connections, J. Math. Phys. 37 (1996) 4647.
[15] J. Madore, Noncommutative geometry for pedestrians, gr-qc/9906059.
[16] B.L. Cerchiai, G. Fiore, J. Madore, Frame formalism for the N -dimensional quantum Euclidean spaces,
math.QA/0007044.
[17] J. Madore, An Introduction to Noncommutative Differential Geometry and Physical Applications, Cambridge Univ. Press, Cambridge, 2000.
[18] S. Majid, H. Ruegg, Bicrossproduct structure of kappa Poincare group and noncommutative geometry, Phys.
Lett. B 334 (1994) 348, hep-th/9405107.
[19] M. Kontsevich, Deformation quantization of Poisson manifolds, I, q-alg/9709040.
[20] D. Arnal, D. Manchon, M. Masmoudi, Choix des signes pour la formalite de M. Kontsevich,
math.QA/0003003.
[21] L. Cornalba, R. Schiappa, Nonassociative star product deformations for D-brane worldvolumes in curved
backgrounds, Commun. Math. Phys. 225 (2002) 33, hep-th/0101219.
[22] B. Jurco, P. Schupp, J. Wess, Noncommutative gauge theory for Poisson manifolds, Nucl. Phys. B 584
(2000) 784, hep-th/0005005.
[23] B. Jurco, P. Schupp, J. Wess, Non-Abelian noncommutative gauge theory via noncommutative extra dimensions, Nucl. Phys. B 604 (2001) 148, hep-th/0102129.
[24] D. Manchon, Poisson bracket, deformed bracket and gauge group actions in Kontsevich deformation quantization, math.QA/0003004.
Exact solution of the XXZ Gaudin model with

generic open boundaries
Wen-Li Yang a,b , Yao-Zhong Zhang b , Mark D. Gould b
a Institute of Modern Physics, Northwest University, Xian 710069, PR China
b Department of Mathematics, University of Queensland, Brisbane, QLD 4072, Australia

Abstract
The XXZ Gaudin model with generic integrable boundaries specified by generic non-diagonal Kmatrices is studied. The commuting families of Gaudin operators are diagonalized by the algebraic
Bethe ansatz method. The eigenvalues and the corresponding Bethe ansatz equations are obtained.
PACS: 03.65.Fd; 04.20.Jb; 05.30.-d; 75.10.Jm
Keywords: Gaudin model; Reflection equation; Algebraic Bethe ansatz
1. Introduction
Gaudin type models [1] have played an essential role in the study of quantum systems
appeared in many physical fields such as the BCS theory [2] of small metallic grains [35],
nuclear physics theory [6,7] and QCD theory [8,9]. They also provide a testing ground for
ideas such as the functional Bethe ansatz and general procedure of separation of variables
[1012] and the construction of integral representations of the solutions to the Knizhnik
Zamolodchikov (KZ) equation [1317].
The original Gaudins magnet Hamiltonians (or Gaudin operators) can be rewritten in
terms of the quasi-classical expansion of the transfer matrices (row-to-row transfer maE-mail address: wenli@maths.uq.edu.au (W.-L. Yang).
doi:10.1016/j.nuclphysb.2004.07.029
504
W.-L. Yang et al. / Nuclear Physics B 698 [PM] (2004) 503516
trices) of inhomogeneous spin chains with periodic boundary condition [18]. Since the
elegant work of Sklyanin [19], the powerful Quantum Inverse Scattering Method (QISM)
has been applied to various integrable models with non-trivial boundary conditions, which
are specified by K-matrices satisfying the reflection equation and its dual [20]. The quasiclassical expansion of the corresponding transfer matrices (double-row transfer matrices)
produces generalized Gaudin Hamiltonians with boundaries [16,21]. In particular, twisted
boundary conditions and open boundary conditions associated with diagonal K-matrices
give rise to Gaudin magnets in non-uniform local magnetic fields [16] and interacting electron pairs with certain non-uniform long-range coupling strengths [4,2124].
In this paper, we study the XXZ type Gaudin magnets with most generic boundary
conditions specified by the generic non-diagonal K-matrices given in [25,26]. In Section 3, we construct the generalized Gaudin operators associated with the generic boundary
K-matrices. The commutativity of these operators follows from the standard procedure
[16,18,21] specializing to the inhomogeneous spin- 21 XXZ open chain, thus ensuring the
integrability of the Gaudin magnets. In Section 4, we diagonalize the Gaudin operators
simultaneously by means of the algebraic Bethe ansatz method. This constitutes the main
new result in this paper. The diagonalization is achieved by means of the technique of
the vertex-face transformation. In Section 5, we conclude this paper by offering some
discussions.
2. Preliminaries: the inhomogeneous spin- 12 XXZ open chain

Throughout, V denotes a two-dimensional linear space and , z are the usual Pauli
matrices which realize the spin- 21 representation of the Lie algebra sl(2) on V . The spin- 21
XXZ chain can be constructed from the well-known six-vertex model R-matrix R(u)
End(V V ) [27] given by
a(u)
b(u) c(u)
R(u)
=
(2.1)
.
c(u) b(u)
a(u)
The coefficient functions read
11
22
R 11
(u) = R 22
(u) = a(u) = 1,
sin u
,
sin(u + )
sin
12
21
.
R 21
(u) = R 12
(u) = c(u) =
sin(u + )
12
21
(u) = R 21
(u) = b(u) =
R 12
(2.2)
Here u is the spectrum parameter and is the so-called crossing parameter. The R-matrix
satisfies the quantum YangBaxter equation (QYBE),
R12 (u1 u2 )R13 (u1 u3 )R23 (u2 u3 )
= R23 (u2 u3 )R13 (u1 u3 )R12 (u1 u2 ),
(2.3)
505
and the properties,

Unitarity: R 12 (u)R 21 (u) = id,
(2.4)
t2 (u 2)(R)
t2 (u) = sin u sin(u + 2) id,
Crossing-unitarity: (R)
21
12
sin(u + ) (u + )

Quasi-classical property: R12 (u) 0 = id.
(2.5)
(2.6)
Here R 21 (u) = P12 R 12 (u)P12 with P12 being the usual permutation operator and ti denotes
the transposition in the ith space. Here and below we adopt the standard notations: for any
matrix A End(V ), Aj is an embedding operator in the tensor space V V , which
acts as A on the j th space and as identity on the other factor spaces; Rij (u) is an embedding
operator of R-matrix in the tensor space, which acts as identity on the factor spaces except
for the ith and j th ones.
One introduces the row-to-row monodromy matrix T (u), which is an 2 2 matrix
with elements being operators acting on V N , where N = 2M (M being a positive integer),
T0 (u) = R 01 (u + z1 )R 02 (u + z2 ) R 0N (u + zN ).
(2.7)
Here {zj | j = 1, . . . , N} are arbitrary free complex parameters which are usually called
inhomogeneous parameters. With the help of the QYBE (2.3), one can show that T (u)
satisfies the so-called RLL relation
R 12 (u v)T1 (u)T2 (v) = T2 (v)T1 (u)R 12 (u v).
(2.8)
Integrable open chain can be constructed as follows [19]. Let us introduce a pair of
K-matrices K (u) and K + (u). The former satisfies the reflection equation (RE) [20]
R 12 (u1 u2 )K1 (u1 )R 21 (u1 + u2 )K2 (u2 )
= K (u2 )R 12 (u1 + u2 )K (u1 )R 21 (u1 u2 ),
2
(2.9)
and the latter satisfies the dual RE

R 12 (u2 u1 )K1+ (u1 )R 21 (u1 u2 2)K2+ (u2 )
= K + (u2 )R 12 (u1 u2 2)K + (u1 )R 21 (u2 u1 ).
2
(2.10)
For open spin-chains, instead of the standard row-to-row monodromy matrix T (u) (2.7),
one needs to introduce the double-row monodromy matrix T(u)
T(u) = T (u)K (u)T 1 (u).
(2.11)
Using (2.8) and (2.9), one can prove that T(u) satisfies
R 12 (u1 u2 )T1 (u1 )R 21 (u1 + u2 )T2 (u2 )
= T2 (u2 )R 12 (u1 + u2 )T1 (u1 )R 21 (u1 u2 ).
Then the double-row transfer matrix of the
boundary is given by

(u) = tr K + (u)T(u) .
inhomogeneous spin- 21
(2.12)
XXZ chain with open
(2.13)
506
With the help of (2.3)(2.5) and (2.9)(2.10), one can prove that the transfer matrices with
different spectral parameters commute with each other [19]:

(u), (v) = 0.
(2.14)
This ensures the integrability of the inhomogeneous spin- 21 XXZ chain with open boundary.
3. XXZ Gaudin models with generic boundaries

In this paper, we will consider a generic K-matrix K (u) which is a generic solution to
the RE (2.9) associated the six-vertex model R-matrix [25,26]

1
k1 (u) k21 (u)
K(u).
K (u) =
(3.1)
k12 (u) k22 (u)
The coefficient functions are
2 cos(1 2 ) cos(1 + 2 + 2 )e2iu
,
4 sin(1 + + u) sin(2 + + u)
i sin(2u)ei(1+2 ) eiu
,
k21 (u) =
2 sin(1 + + u) sin(2 + + u)
i sin(2u)ei(1+2 ) eiu
k12 (u) =
,
2 sin(1 + + u) sin(2 + + u)
2 cos(1 2 )e2iu cos(1 + 2 + 2 )
k22 (u) =
.
4 sin(1 + + u) sin(2 + + u)
k11 (u) =
(3.2)
At the same time, we introduce the corresponding dual K-matrix K + (u) which is a generic
solution to the dual reflection equation (2.10) with a particular choice of the free boundary
parameters with respect to K (u):
+1

1
k (u) k + 2 (u)
K + (u) = 1+2
(3.3)
.
2
k1 (u) k + 2 (u)
The matrix elements are
2 cos(1 2 )ei cos(1 + 2 + 2 )e2iu+i
,
4 sin(1 + u ) sin(2 + u )
1
i sin(2u + 2)ei(1+2 ) eiui
k2+ (u) =
,
2 sin(1 + u ) sin(2 + u )
2
i sin(2u + 2)ei(1+2 ) eiu+i
k1+ (u) =
,
2 sin(1 + u ) sin(2 + u )
2
2 cos(1 2 )e2iu+i cos(1 + 2 + 2 )ei
k2+ (u) =
.
4 sin(1 + u ) sin(2 + u )
k1+ (u) =
1
(3.4)
The K-matrices depend on three free boundary parameters {1 , 2 , } which specify

inte
grable boundary conditions [26]. It is very convenient to introduce a vector = 2k=1 k k
507
associated with the boundary parameters {i }, where {i , i = 1, 2} form the orthonormal
basis of V such that i , j = ij . We remark that K (u) does not depend on the crossing
parameter but K + (u) does. They satisfy the following relation:

lim K + (u)K (u) = lim K + (u) K(u) = id.
(3.5)
Let us introduce the generalized XXZ Gaudin operators [1] {Hj | j = 1, 2, . . . , N} associated with the spin- 21 XXZ model with generic boundaries specified by the boundary
K-matrices in (3.1) and (3.3):
Hj = j (zj ) +
2M

k=j

kz jz 1
1
+
+
+ k j + cos(zj zk )
sin(zj zk ) k j
2
2M K 1 (z )

j
j
k=j
where j (u) =
sin(zj + zk )
j+ k

{Kj (u)}|=0 Kj (u), j
+ j k+
+ cos(zj + zk )
jz kz 1
2

Kj (zj ), (3.6)
= 1, . . . , N , with K j (u) = tr0 {K0+ (u)R 0j (2u)P0j }
and {zj } correspond to the inhomogeneous parameters of the inhomogeneous spin- 21

XXZ chain with generic open boundary. For a generic choice of the boundary parameters {1 , 2 , }, j (u) is an non-diagonal matrix, in contrast to that of [21].
The XXZ Gaudin operators (3.6) are obtained by expanding the double-row transfer
matrix (u) (2.13) at the point u = zj around = 0:

(zj ) = (zj )=0 + Hj + O 2 ,

Hj =
(zj ) .
=0
j = 1, . . . , N,
(3.7)
(3.8)
The relations (2.6) and (3.5) imply that the first term (zj )|=0 in the expansion (3.7) is
equal to an identity, namely,

(zj )=0 = id.
(3.9)
Then the commutativity of the transfer matrices { (zj )} (2.14) for a generic implies
[Hj , Hk ] = 0,
i, j = 1, . . . , N.
(3.10)
Thus the Gaudin system defined by (3.6) is integrable. Moreover, the fact that the Gaudin
operators {Hj } (3.6) can be expressed in terms of the transfer matrix of the inhomogeneous
spin- 21 XXZ open chain enables us to exactly diagonalize the operators by the algebraic
Bethe ansatz method with the help of the vertex-face correspondence technique, as can
be seen in the next section. The aim of this paper is to find the common eigenvectors and
eigenvalues of the operators (3.6).
508
4. Eigenvalues and Bethe ansatz equations

4.1. Six-vertex SOS R-matrix and face-vertex correspondence
The simple root and fundamental weight 1 of sl(2) are given in terms of the orthonormal basis {i } as: = 1 2 , 1 = /2. Set
1
k ,
2
2
i = i ,
=
i = 1, 2,
then
k=1
2

i = 0.
(4.1)
i=1
For each dominant weight = a1 , a Z+ (the set of non-negative integer), there exists
an irreducible highest weight finite-dimensional representation V of A1 with the highest
vector |. For example, the fundamental vector representation is V1 .
Let h be the Cartan subalgebra of A1 and h be its dual. A finite-dimensional diagonalisable h-module
is a complex finite-dimensional vector space W with a weight decomposition W = h W [], so that h acts on W [] by x v = (x)v (x h, v W []).
For example, the fundamental vector representation V1 = V , the non-zero weight spaces
= Ci , i = 1, 2.
W [i]
For a generic m V , define
mi = m, i ,
mij = mi mj = m, i j ,
i, j = 1, 2.
(4.2)
Let R(u, m) End(V V ) be the R-matrix of the six-vertex SOS model, which is trigonometric limit of the eight-vertex SOS model [28] given by
R(u, m) =
2

Riiii (u, m)Eii Eii
i=1
2

ij
ji
Rij (u, m)Eii Ejj + Rij (u, m)Ej i Eij ,
(4.3)
i=j
where Eij is the matrix with elements (Eij )lk = j k il . The coefficient functions are
Riiii (u, ) = 1,
ji
Rij (u, m) =
ij
Rij (u, ) =
sin u sin(mij )
,
sin(u + ) sin(mij )
sin sin(u + mij )

,
sin(u + ) sin(mij )
i = j,
i = j,
(4.4)
(4.5)
and mij is defined in (4.2). The R-matrix satisfies the dynamical (modified) quantum Yang
Baxter equation

R12 u1 u2 , m h(3) R13 (u1 u3 , m)R23 u2 u3 , m h(1)

= R23 (u2 u3 , m)R13 u1 u3 , m h(2) R12 (u1 u2 , m).
(4.6)
We adopt the notation: R12 (u, mh(3) ) acts on a tensor v1 v2 v3 as R(u, m)id
if v3 W [].
509
Define the following functions:

(1) (u) = eiu ,
(2) (u) = 1.
(4.7)
Let us introduce an intertwiner, i.e., an 2-component column vector m,mj (u) whose kth
element is
(k)
m,mj
(u) = (k) (u + 2mj ).
(4.8)
Using the intertwiner, one can derive the following face-vertex correspondence relation
[29]
R 12 (u1 u2 )m,m (u1 ) m ,m( +) (u2 )

R(u1 u2 , m)kl
=
k)
(u1 ) m,ml(u2 ).
ij ml,(
l+
(4.9)
k,l
Then the QYBE (2.3) of for the vertex-type R-matrix R(u)

is equivalent to the dynamical YangBaxter equation (4.6) of the SOS R-matrix R(u, m). For a generic m, we can
satisfying the conditions,
introduce other types of intertwiners ,
2

(k)
(k)
m,m (u)m,m (u) = ,
(4.10)
(k)
(k)
(u)m+,
(u) = ,
m+,m
(4.11)
k=1
2

k=1
from which one can derive the following relations:

2

(j )
(i)
m,m (u)m,m (u) = ij ,
(4.12)
(j )
(i)
(u)m+,m
(u) = ij .
m+,m
(4.13)
=1
2

=1
Through straightforward calculations, we find the K-matrices K (u) given by (3.1)

and (3.3) can be expressed in terms of the intertwiners and diagonal matrices K(|u) and
K(|u)
as follows
(s)
j (t )
( ), (u)K(|u)i , (u),
K (u)st =
(4.14)
i,j
K + (u)st =
j (t )
(s)
,
(u)K(|u)
(u).
i (
),
i,j
Here the two diagonal matrices K(|u) and K(|u)

are given by
(4.15)
510

K(|u) Diag k(|u)1 , k(|u)2

sin(1 + u) sin(2 + u)
= Diag
,
,
sin(1 + + u) sin(2 + + u)

K(|u)
Diag k(|u)
1 , k(|u)
2

sin(12 ) sin(1 + + u + )
,
= Diag
sin 12 sin(1 + u )

sin(12 + ) sin(2 + + u + )
.
sin 12 sin(2 + u )
(4.16)
(4.17)
Moreover, one can check that the matrices K(|u) and K(|u)
satisfy the SOS type reflection equation and its dual, respectively [30]. Although the K-matrices K (u) given
by (3.1) and (3.3) are generally non-diagonal (in the vertex form), after the face-vertex
transformations (4.14) and (4.15), the face type counterparts K(|u) and K(|u)
simultaneously become diagonal. This fact enables us to apply the generalized algebraic Bethe
ansatz method developed in [31] for SOS type integrable models to diagonalize the transfer
matrices (u) (2.13).
4.2. Algebraic Bethe ansatz
Using the relations (4.12) and (4.13), the decomposition of K + (u) (4.15) and the diagonal property (4.17), the transfer matrix (u) (2.13) can be recasted into the following face
type form

(u) = tr K + (u)T(u)

=
tr K + (u)(
),
(u)(
), (u)
,
T(u), (u) , (u)
, (u)K + (u)(
),
(u)(
),
(u)
T(u), (u)

K(|u)
k(|u)
T (|u) .
T (|u) =
(4.18)
Here we have introduced the face-type double-row monodromy matrix T (|u)

T (|u) = (
), (u)T(u), (u)
(j )
j (i)
(u)T(u)i , (u).
(
),
(4.19)
i,j
This face-type double-row monodromy matrix can be expressed in terms of the face type
R-matrix R(|u) (4.3) and K-matrix K(|u) (4.16) (for the details, see Eq. (4.19) of [31]).
Moreover, from (2.12), (4.9) and (4.13) one can derive the following exchange relations
511
among T (|u) :

i

i ,j
R(u1 u2 , )i01 ,j01 T + (1 + 2 )u1 i1
2
i1 ,i2 j1 ,j2
j

j ,i
R(u1 + u2 , )j12 ,i23 T + (3 + 3 )u2 j2
3

j
i ,j
=
T + (1 + 0 )u2 j0 R(u1 + u2 , )i01 ,j12
1
i1 ,i2 j1 ,j2
i

j ,i
T + (2 + 2 )u1 i1 R(u1 u2 , )j23 ,i23 .
2
(4.20)
As in [31], let us introduce a set of standard notions for convenience:1

A(|u) = T (|u)11 ,
B(|u) =
T (|u)12
,
(12 )
(4.21)

(12 + )
2 1 A(|u) .
(4.22)
T (|u)22 R(2u, + 1)
12
(12 )
Hereafter, we adopt the convention: ij = i j , introduced in (4.2). After tedious calculations, we find the commutation relations among A(|u), D(|u) and B(|u). Here we
give the relevant ones for our purpose,
D(|u) =
A(|u)B( 21|v)
sin(u + v) sin(u v )
B( 21|v)A(
21|u)
=
sin(u + v + ) sin(u v)
sin(u v 12 + )
sin sin 2v
B( 21|u)A(
21|v)
sin(u v) sin(2v + )
sin(12 )
sin
sin(u + v + 21 + 2)
B( 21|u)D(
21|v),
(4.23)
sin(u + v + )
sin(21 + )
D(|u)B( 21|v)
sin(u v + ) sin(u + v + 2)
B( 21|v)D(
21|u)
sin(u v) sin(u + v + )
sin sin(2u + 2) sin(u v + 12 )
B( 21|u)D(
21|v)
sin(u v) sin(2u + ) sin(12 )
sin sin 2v sin(2u + 2) sin(u v + 12 )
+
sin(u + v + ) sin(2v + ) sin(2u + ) sin(12 )
B( 21|u)A(
21|v),
(4.24)
B( 21|u)B( 41|v) = B( 21|v)B( 41|u).

(4.25)
=
Here we have used the identity 2 = 1 which can be derived from (4.1).
In order to apply the algebraic Bethe ansatz method, in addition to the relevant commutation relations (4.23)(4.25), one needs to construct a pseudo-vacuum state (also called
1 The scalar factors in the definitions of the operators B(|u) and D(|u) are to make the relevant commutation
relations as concise as (4.23)(4.25).
512
reference state) which is the common eigenstate of the operators A, D and is annihilated
by the operator C. In contrast to the case of the spin- 21 XXZ open chain with diagonal
K (u) [19], for the open chain with generic non-diagonal K-matrices (3.1) and (3.3), the
usually highest-weight state

1
1

,
0
0
is no longer the pseudo-vacuum state. However, after the face-vertex transformations (4.14)
and (4.15), the face type counterparts K-matrices K(|u) and K(|u)
simultaneously be1
come diagonal. This suggests that one can transfer the spin- 2 XXZ open chain with generic
non-diagonal K-matrices into the corresponding SOS model with diagonal K-matrices
K(|u) and K(|u)

given by (4.16) and (4.17) (possibly after some local gauge transformations [29]). Then it is easy to construct the pseudo-vacuum in the face language and
use the generalized algebraic Bethe ansatz method [31] to diagonalize the transfer matrix.
We shall develop this game in the following.
Let us introduce the corresponding pseudo-vacuum state |
| = (N1)1,N
1 (z1 ) (N2)1,(N1)
1 (z2 )
,1 (zN ).
(4.26)
The state does only depend on the boundary parameters {1 , 2 } and the inhomogeneous
parameters {zj }. Using the technique developed in [31], after tedious calculations, we find
that the pseudo-vacuum state given by (4.26) satisfies the following equations, as required,
sin(1 + u)
|,
sin(1 + + u)
sin 2u sin(1 + + u + ) sin(2 + u )
D( N1|u)|
=
sin(2u + ) sin(1 + + u) sin(2 + + u)

N

sin(u + zk ) sin(u zk )
|,
sin(u + zk + ) sin(u zk + )
A( N1|u)|
=
(4.27)
(4.28)
k=1
C( N1|u)|
= 0,
B( N1|u)| = 0.
(4.29)
(4.30)
Then the so-called Bethe states can be constructed by applying the creation operator B on
the pseudo-vacuum state
1 )B( 41|v
2 ) B( 2M1|v
M )|.
|v1 , . . . , vM = B( 21|v
(4.31)
From (4.18), (4.21) and (4.22) we can rewrite the transfer matrices (u) (2.13) in terms
of the operators A and D
(u) =
sin(2 + u) sin(1 + + u) sin(2u + 2)

A(|u)
sin(2 + u ) sin(1 + u ) sin(2u + )
sin(2 + + u + )
+
D(|u).
sin(2 + u )
(4.32)
513
Acting the above expression of the transfer matrices on the Bethe states |v1 , . . . , vM (4.31)
and repeatedly using the relevant commutation relations (4.23)(4.25), we obtain
(u)|v1 , . . . , vM
= t (u)|v1 , . . . , vM
sin(2u + 2) sin
+
sin(2 + u ) sin(1 + u )

M

F |v1 , . . . , v1 , u, v+1 , . . . , vM .
(4.33)
=1
Here, the term associated with function t (u) is the so-called wanted term which gives
rise to the eigenvalues and the term associated with {F | = 1, . . . , M} is the so-called
unwanted term. They are given, respectively, by
t (u) =
sin(2 + u) sin(1 + u) sin(2u + 2)

sin(2 + u ) sin(1 + u ) sin(2u + )
M

sin(u + vk ) sin(u vk )
sin(u + vk + ) sin(u vk )
k=1
sin(2 + + u + ) sin(1 + + u + ) sin 2u

sin(2 + + u) sin(1 + + u) sin(2u + )
M

sin(u + vk + 2) sin(u vk + )
sin(u + vk + ) sin(u vk )
k=1
2M

k=1
sin(u + zk ) sin(u zk )
,
sin(u + zk + ) sin(u zk + )
(4.34)
and
F =
sin 2v sin(2 + v ) sin(1 + v )

sin(2v + ) sin(u + v + ) sin(u v )
M
sin(v + vk ) sin(v vk )
sin(v + vk + ) sin(v vk )
k=
sin(2 + + v + ) sin(2 + v )
sin(2 + + v ) sin(2 + v )
sin(1 + + v + ) sin(1 + v )
sin(1 + + v ) sin(1 + v )
M

sin(v + vk + 2) sin(v vk + )
sin(v + vk + ) sin(v vk )
k=

2M

sin(v + zk ) sin(v zk )
,
sin(v + zk + ) sin(v zk + )
k=1
= 1, . . . , M.
(4.35)
514
The relation (4.33) tells us that the Bethe states |v1 , . . . , vM are eigenstates of the transfer
matrices (u) if the unwanted terms vanish: F = 0, = 1, . . . , M. This leads to the Bethe
ansatz equations,
sin(2 + + v ) sin(2 + v )
sin(2 + + v + ) sin(2 + v )
sin(1 + + v ) sin(1 + v )
sin(1 + + v + ) sin(1 + v )
M

sin(v + vk + 2) sin(v vk + )
=
k=
2M

k=1
,
sin(v + zk + ) sin(v zk + )
= 1, . . . , M.
(4.36)
We have checked that the Bethe ansatz equations indeed ensure that the eigenvalues (4.34)
of transfer matrices (u) are entire functions. Our result recovers that of [29] for the very
special case zk = 0, k = 1, . . . , 2M.
4.3. Eigenstates and the corresponding eigenvalues
The relation (3.8) between {Hj } and { (zj )} and the fact that the first term of (3.7) is a cnumber enable us to extract the eigenstates of the Gaudin operators and the corresponding
eigenvalues from the results obtained in last subsection.
Let us introduce the state | (0),
i(zN 21 )
i(z1 21 )
e
e
(0)
| =
(4.37)

.
1
1
This state can be obtained from the pseudo-vacuum state | (4.26) by taking the limit:
| (0) = lim0 |. Let us introduce the matrix C(u) End(V ) associated the intertwiner vector

i(u+2 )
1
ei(u+22)
e
,
C(u) =
(4.38)
1
1
and the associated gauged Pauli operator (u) End(V ),
(u) = C(u) C 1 (u).
(4.39)
Then we define the states (v1 , . . . , vM ) and :

2M
M

sin(1 + v ) sin(v zk + 12 )
(v1 , . . . , vM ) =
sin(1 + + v ) sin(v zk )
=1 k=1

sin(2 + v ) sin(v + zk 12 )
k (zk ) (0) , (4.40)
sin(2 + + v ) sin(v + zk )
515
2M
M

sin(1 + v ) sin(v zk + 12 )
=
sin(1 + + v ) sin(v zk )
= k=1

sin(2 + v ) sin(v + zk 12 )
k (zk ) (0) .
sin(2 + + v ) sin(v + zk )
(4.41)
Noting the relations (3.8), (3.7) and (4.33)(4.35), and using the same method as in
[16], we derive the following relations
Hj (v1 , . . . , vM ) = Ej (v1 , . . . , vM )
M

sin 2zj sin 2v
+
sin(zj + v ) sin(zj v )
=1

sin(2 + v ) sin(1 + v )
f j (zj ) . (4.42)
sin(2 + zj ) sin(1 + + zj )
The functions Ej and f are

Ej = cot 2zj +
2

cot(j + zj ) +
j =1
f =
2

j =1
M

k=1
sin 2zj
,
sin(vk zj ) sin(vk + zj )
(4.43)

1
1
+
sin(j + v ) sin(j + + v )
2M
k=1
M

k=
1
.
(4.44)
The equation (4.42) suggests that the state (v1 , . . . , vM ) is an eigenstate of the Gaudin
operators {Hj } if {vj |j = 1, . . . , M} is set to satisfy f = 0 for = 1, . . . , M. This leads
to the corresponding Bethe ansatz equations
2

j =1

1
1
+
sin(j + v ) sin(j + + v )
=2
2M
k=1
M

k=
1
,
= 1, . . . , M.
(4.45)
5. Conclusions
We have studied the XXZ Gaudin model with generic boundaries specified by the nondiagonal K-matrices K (u), (3.1) and (3.3). In addition to the inhomogeneous parameters
{zj }, the associated Gaudin operators {Hj }, (3.6), have three free parameters {1 , 2 , },
which give rise to three-parameter (1 , 2 , ) generalizations of those in [4] and twoparameter (1 , 2 ) generalizations of those in [16,21]. As seen from Section 4, although
516
the vertex type K-matrices K (u) (3.1) and (3.3) are non-diagonal, the compositions,
(4.14) and (4.15), lead to the diagonal face-type K-matrices, (4.16) and (4.17), after
the face-vertex transformation. This enables us to successfully construct the corresponding
pseudo-vacuum state | (4.26) and apply the algebraic Bethe ansatz method developed in
[31] for the SOS type models to diagonalize the transfer matrix (u) of the inhomogeneous
spin- 21 XXZ open chain with the generic non-diagonal K-matrices. Furthermore, we have
exactly diagonalized the Gaudin operators {Hj }, and derived their common eigenstates
(4.40) and eigenvalues (4.43) as well as the associated Bethe ansatz equations (4.45).
Acknowledgements
The financial support from Australian Research Council is gratefully acknowledged.
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]
M. Gaudin, J. Phys. (Paris) 37 (1976) 1087.

J. Bardeen, L.N. Cooper, J.R. Schrieffer, Phys. Rev. 108 (1957) 1175.
M.C. Cambiaggio, A.M.F. Rivas, M. Saraceno, Nucl. Phys. A 624 (1997) 157.
L. Amico, A. Di Lorenzo, A. Osterloh, Phys. Rev. Lett. 86 (2001) 5759;
L. Amico, A. Di Lorenzo, A. Osterloh, Nucl. Phys. B 614 (2001) 449.
J. von Delft, D.C. Ralph, Phys. Rep. 345 (2001) 61.
D.J. Dean, M. Hjorth-Jeansen, Rev. Mod. Phys. 75 (2003) 607.
J. Dukelsky, S. Pittel, G. Sierra, Exactly solvable RichardsonGaudin models for many-body quantum systems, nucl-th/0405011.
F. Iachelo, Nucl. Phys. A 570 (1994) 145.
D.H. Rischke, R.D. Pisarski, Color superconductivity in cold, dense quark matter, nucl-th/0004016.
E.K. Sklyanin, J. Sov. Math. 47 (1989) 2473;
E.K. Sklyanin, Lett. Math. Phys. 47 (1999) 275.
E.K. Sklyanin, T. Takebe, Phys. Lett. A 219 (1996) 217.
T. Brzezinski, A.J. Macfarlane, J. Math. Phys. 35 (1994) 3261.
H.M. Babujian, J. Phys. A 26 (1993) 6981.
T. Hasegawa, K. Hikami, M. Wadati, J. Phys. Soc. Jpn. 63 (1994) 2895.
B. Feign, E. Frenkel, N. Reshetikhin, Commun. Math. Phys. 166 (1994) 27.
K. Hikami, J. Phys. A 28 (1995) 4997.
M. Gould, Y.-Z. Zhang, S.-Y. Zhao, Nucl. Phys. B 630 (2002) 492;
M. Gould, Y.-Z. Zhang, S.-Y. Zhao, J. Phys. A 35 (2002) 9381.
K. Hikami, P.P. Kulish, M. Wadati, J. Phys. Soc. Jpn. 61 (1992) 3071.
E.K. Sklyanin, J. Phys. A 21 (1988) 2375.
I.V. Cherednik, Teor. Mat. Fiz. 61 (1984) 35 (in Russian).
A. Di Lorenzo, L. Amico, K. Hikami, A. Osterloh, G. Giaquinta, Nucl. Phys. B 644 (2002) 409.
J. Dukelsky, C. Esebbag, P. Schuck, Phys. Rev. Lett. 87 (2001) 66403.
H.Q. Zhou, J.R. Links, R.H. McKenzie, M.D. Gould, Phys. Rev. B 65 (2002) 060502.
J. von Delft, R. Poghossian, Phys. Rev. B 66 (2002) 134502.
H.J. de Vega, A. Gonzalez-Ruiz, J. Phys. A 26 (1993) L519.
S. Ghoshal, A.B. Zamolodchikov, Int. J. Mod. Phys. A 9 (1994) 3841.
V.E. Korepin, N.M. Bogoliubov, A.G. Izergin, Quantum Inverse Scattering Method and Correlation Function, Cambridge Univ. Press, Cambridge, 1993.
R.J. Baxter, Exactly Solved Models in Statistical Mechanics, Academic Press, New York, 1982.
J. Cao, H.-Q Lin, K.-J. Shi, Y. Wang, Nucl. Phys. B 663 (2003) 487.
W.-L. Yang, R. Sasaki, Solution of the dual reflection equation for An1 SOS model, hep-th/0308118,
J. Math. Phys., in press.
W.-L. Yang, R. Sasaki, Nucl. Phys. B 679 (2004) 495.
Killing spinor equations from nonlinear realisations

Andr Miemiec a , Igor Schnakenburg b
a Institut fr Physik, Humboldt Universitt, Newtonstr. 15, D-12489 Berlin, Germany
b Racah Institute of Physics, The Hebrew University, Jerusalem 91904, Israel

Available online 29 July 2004
Abstract
Starting from a nonlinear realisation of eleven-dimensional supergravity based on the group G11 ,
whose generators appear as low level generators of E11 , we present a super extended algebra, which
leads to a covariant derivative of spinors identical to the Killing spinor equation of this theory. A similar construction leads to the Killing spinor equation of N = 1 pure supergravity in ten dimensions.
PACS: 04.65.+e; 11.30.Ly
1. Introduction
The search for hidden symmetries of eleven-dimensional supergravity (M-theory) [1]1
has a long history [2]. In the context of supergravities relation to string theories the existence of hidden symmetries realising U-duality was first conjectured in [3]. Identifying
the (hidden) symmetry group of a given supergravity theory is of vast interest for understanding its properties; it is essential for finding solution generating techniques, but also
for pinpointing the whole issue of dualities that apparently relate different supergravities
to each other.
E-mail addresses: miemiec@physik.hu-berlin.de (A. Miemiec), igorsc@phys.huji.ac.il (I. Schnakenburg).
1 Our conventions differ from those in [1] by rescaling all gauge fields by a factor of 1/2, working with a
mostly plus signature and replacing a by i a .
doi:10.1016/j.nuclphysb.2004.07.012
518
A. Miemiec, I. Schnakenburg / Nuclear Physics B 698 [PM] (2004) 517530
Several approaches to manifest these extra symmetries have been developed in the past.
In [4,5] it was shown that the nongravitational degrees of freedom of almost all supergravity theories can be described as a nonlinear realisation which made part of the hidden
symmetries manifest. The result was obtained by using the doubled field method. The additional degrees of freedom introduced by doubling of the fields are projected out by the
equations of motion of the gauge fields which take the form of twisted selfduality conditions. The generators of this coset construction were inert under Lorentz transformations,
and as such it is difficult to extend this method straightforwardly to include gravity or
fermions. In [6] it was shown that the entire bosonic sector of eleven- and ten-dimensional
IIA supergravity could be formulated as a nonlinear realisation. In this way of proceeding,
gravity is treated on an equal footing with the gauge fields and thus is naturally built in. The
method of nonlinear realisations has consequently been shown to extend to other gravity
theories [79]. However, it did not include the fermionic degrees of freedom.
The nonlinear realisation of M-theory of [6] is based on the nonsimple algebra G11 with
group element

1
1
x P a b K a b
c1 ...c3
c1 ...c6
Ac ...c R
e
exp
+ Ac1 ...c6 R
gB = e
(1.1)
3! 1 3
6!
that was shown to generate the covariant structure of the bosonic fields (subscript B) and
their equations of motion. The group G11 is defined by the algebra (only nontrivial commutators displayed)

a
a
K b , Pc = ca Pb ,
K b , K c d = bc K a d da K c b ,
(1.2)

a

K b , R c1 ...c3 = 3b[c1 R c2 c3 ]a ,
(1.3)
K a b , R c1 ...c6 = 6b[c1 R c2 ...c6 ]a ,

c ...c
c4 ...c6
c1 ...c6
1
3
,R
= c3,3 R
R
(1.4)
and the Cartan form of the coset (hab = (ab) , [ab] = 0) reads

1
gB1 dgB = dx P + eh eh a b K a b + (4D [ Ac1 c2 c3 ] )R c1 c2 c3 +
4!

1
+ (7D [ Ac1 ...c6 ] )R c1 ...c6
7!
with
(1.5)

D Ac1 c2 c3 = Ac1 c2 c3 + eh eh c b Abc2 c3 + eh eh c b Ac1 bc3
1
2
b

h
h
+ e e c Ac1 c2 b ,
3

D Ac1 ...c6 = Ac1 ...c6 + eh eh c b Abc2 ...c6 + eh eh c b Ac1 bc3 ...c6 +
1
2

+ eh eh c b Ac1 ...c5 b 10A[c1 ...c3 D || Ac4 ...c6 ] c3,3 .
6
The antisymmetry of the indices indicated in (1.5) is not obtained automatically by the
procedure outlined so far but it is the result of a second step, i.e., making the Cartan form
of G11 simultaneously covariant with respect to the conformal group [6,10]. Identifying the
resulting objects with the two field strengths Ga1 ...a4 and Fa1 ...a7 the equation of motion of
519
the gauge field of M-theory reads

G(4) = F(7) .
(1.6)
In this way the bosonic gauge sector of eleven-dimensional supergravity is completely

described by the covariant field strengths together with a geometric equation of motion.
The second term in Eq. (1.5) requires special attention. From the transformation properties
of the Cartan form according to the transformation gB g gB h1 and the structure of
the algebra (1.2)(1.4) it follows that the shift term of the Cartan form that appears after
performing the transformation is due to the object in front of the antisymmetric part of
the generators K a b . The same shift term also appears in the transformation of the spin
connection. So we may assume

h
e eh [ab] = ab + [ab]
with [ab] transforming as a tensor. We made the antisymmetry of ab in the latter two
indices explicit since we want to extend the definition of ab to denote the tensors in front
of the symmetric part of K a b , too. Due to the inverse Higgs effect one can put any Cartan form with a homogeneous transformation law to zero without affecting physics [11].
This allows one to neglect ab and to find relations between ab and (eh eh )ab at
the same time (see [6]). In the case of M-theory, nonlinear realisations led to the proposal
of a hidden E11 symmetry [12]. E11 appeared as the simplest KacMoody algebra which
contained the nonsimple algebra of G11 but without the momentum generator. So far the
discussion of the hidden symmetries was limited to the bosonic sector of the supergravity only. One open problem of the E11 conjecture is firstly, how to incorporate fermionic
degrees of freedom and secondly, what restrictions this places on the corresponding extension of the algebra G11 . The best possible answer would be to treat the bosons and the
11 the
fermions (gravitino) on the same footing, i.e., to generate from an extended group G
covariant derivative of the gravitino accompanied by the fermionic shifts in the bosonic
field strengths and the spin connection. Avoiding the construction of a group extension it
is, of course, possible to introduce the fermionic shifts in the bosonic field strengths just by
hand [13]. But then it remains unclear how the extended bosonic symmetries couple to the
fermionic symmetries, i.e., what the extended hidden symmetry group actually looks like.
Alternatively, one can try to generalise the fields of the theory to superfields in superspace
aiming to find the fermionic field equations by twisted superdualities. This was performed
for a two-dimensional model in [14].
Historically, the identification of the KacMoody algebra that describes the hidden symmetries of the theory was performed by an algebra, which did not take the role of the
momentum operator as a central charge of the supersymmetry algebra into account. Later
the semi-direct product of E11 and representations of the momentum generator in eleven
dimensions were considered. The semi-direct product includes nontrivial commutators of
the momentum generator with the gauge field generators, which close in the central charges
of the supersymmetry algebra in eleven dimensions [15]. Of course, the momentum generator itself appears as a central charge of this algebra.
In the following we will rather take the semi-direct product of some low-level generators
of E11 (i.e., G11 ) when split into representation of SL(11) with a spinor representation of
SO(1, 10). The Lorentz group can be obtained from the gravity line by using the Cartan
520
involution (or the temporal involution [16]). In this way, we are still free to define the anticommutation relations of this fermionic generator with itself and we will naturally choose
the supersymmetry algebra of the relevant supergravity theory. This algebra contains the
momentum generator apart from the central charges, and so the momentum generator can
effectively be added via a semi-direct product. The occurrence of a fermionic generator
parametrising the coset of a nonlinear realisation will result in a covariant expression for a
fermionic parameter, and we suggest to identify this covariant expression with the Killing
spinor equation of the theory under consideration. This fermionic parameter is, however,
not part of the fields of the theory and thus we still keep a purely bosonic background but
yet include supersymmetry into the ansatz of (1.1).
Since the spin representation we multiply is connected with the SO(1, 10) subgroup
generated by the antisymmetric combination of the generators K a b of SL(11), we just
keep these and throw the symmetric combination away (avoiding topological difficulties
[18]). We will partly answer the question as to whether there exists an extension of the
algebra of G11 by a fermionic generator Q generating a parameter , so that the Cartan
form finally produces a covariant derivative of ,

1
1

D = ab ab
G , (1.7)
8
4
2 144
identical to the Killing spinor equation. Algebraically this problem is closely related to the
full program of consistently including fermions into a nonlinear realisation and find them
to be Goldstonian. The difference is merely that we do not have to consider the fermionic
shifts induced on the bosonic fields. One should note the conceptual difference to [2],
whose Cartan form contains the potentials Ac1 ...c3 and Ac1 ...c6 and not the field strengths
G(4) and F(7) .
2. Supersymmetrisation
The ansatz for the superalgebra is mainly fixed by the structure of the covariant derivative we want to generate. Nevertheless one has to make some choices. There are several
hints contained in the literature as to how the supersymmetric extension might look like
[2,6,17]. For different reasons all of these three papers had to take a second, unphysical spinorial generator Q into account. We call this generator unphysical since we
do not identify the operator which arises in front of Q with the covariant derivative of a
physical quantity. In fact, in our calculation we observed the need for a second fermionic
generator, too. We will discuss the technical reason below. The most convincing heuristic
argument for the second fermionic generator in the approach via nonlinear realisations is
derived from the closure of G11 with the conformal group. The conformal group in d = 11
is isomorphic to SO(2, 11), whose lowest irreducible spin representation is of dimension
26 = 2 32, i.e., twice the amount of the spinor representation in d = 11. This superalgebra
was explicitly constructed in [17].
It was laid out in the introduction that the covariant field strengths, and in particular
their antisymmetrisation, are only found after taking the closure with the conformal group.
It therefore appears to be natural to include two fermionic generators Q and Q and
521
multiply the group element in (1.1) from the right by

g = e
(Q
)
+Q
(2.1)
The new Cartan form becomes:

A = g1 dg + g1 gB1 dgB g .
We work out the first term on the right-hand side ending up with an expansion of the form

) 1 (Q + Q
), d (Q + Q ) + ,
g1 dg = d (Q + Q
2
where we have only expanded up to second order in the fermions since we will soon find
that due to the Jacobi identities commutators with more than two fermionic generators
vanish (see Remark 1). For
the second bit in the expression A we introduce a shorthand
notation. We set gB1 dgB = 4i=1 ( )Gi with Gi {Pa , K a b , R c1 c2 c3 , R c1 ...c6 } where the
dots in brackets refer to the prefactors in Eq. (1.5) determined in the last paragraph. Then
it reads
4

1 1
1
( )Gi g
g gB dgB g = g
i=1
and finally we obtain for each of the four individual contributions an expansion of the type

g1 Gi g = Gi (Q + Q ), Gi

1
(Q + Q ), (Q + Q ), Gi + .
+
2!
It is feasible to organise the expansions by the power in the fermionic parameter they
contain, i.e.,
A=
A(k) ,

A(0) = gB1 dgB .
k=0
At zeroth order we just recover the purely bosonic elements of the Cartan form. Formally, the expansions goes all the way up to infinity depending on our choice of
(anti)commutation relations of the fermionic generators. For the super algebra we are going
to use it will terminate at k = 3.
2.1. Linearised analysis, i.e., O(2 )
To first order in the Cartan form looks like
A(1) = d (Q + Q )
4

(Q + Q ), ( )Gi .
i=1
Now we evaluate the terms linear in step by step. Because of the Z2 -grading of a superalgebra, the commutators between fermionic and bosonic generators can only yield
522
fermionic generators. They additionally have to fulfill the super Jacobi identities. We set2

Q , K [bc] = k [bc] Q ,
Q , K [bc] = k [bc] Q ,

Q , R c1 c2 c3 = c1 c2 c3 Q ,
Q , R c1 c2 c3 = c1 c2 c3 Q ,
2 c ...c
1 6 Q ,
Q , R c1 ...c6 =
c3,3

2 c ...c
Q , R c1 ...c6 =
1 6 Q ,
c3,3
(2.2)
where and are free parameters. Appendix A shows that this choice is consistent with the
super-Jacobis. We note that it is essential to observe that in the second line the commutator
with R a1 a2 a3 exchanges the two different Q generators! However, since one of them is
nonphysical, we only display terms proportional to Q which look

A(1) = dx Q eb [Q , Pb ] bc Q , K [bc]

0!

1
(4D Ac1 c2 c3 ) Q , R c1 c2 c3
4!

1
c1 ...c6
(7D Ac1 ...c6 ) Q , R
7!
(2.3)
and the commutator [Q, P ] vanishes due to the Jacobi identity No. (14) in Appendix A.
The further simplifications of Eq. (2.3) are straightforward. The only trick one has to keep
in mind is connected with rewriting the term containing the generator R c1 ...c6 . We have to
use the equations of motion, i.e., the condition that the two gauge field strengths G(4) and
F(7) are related by Hodge duality Eq. (1.6), to draw the following conclusion:3
1
1
F2 ...7 2 ...7 = 1 ...4 G1 ...4 .
6!
4!
(2.4)
Using this identity we may rewrite the contribution delivered by the 6-form potential into

2

1
1
c1 ...c6
1 ...4
(7D Ac1 ...c6 ) Q , R
G1 ...4 Q .
7!
c3,3
7 4!
Inserting this into Eq. (2.3) one obtains

(1)
A = dx bc k [bc]

c c c
1
2 c0 ...c3
c0
1
2
3
Gc0 c1 c2 c3 Q .
+ []
4! 7c3,3
2 Lorentz generator J ab = 2 K [ab] k [ab] = 1 ab .
4
(1)(11j)(11j1)/2
aj+1 ...a11
3
and 0...10 = sgn{0 . . . 10}.

a1 ...aj =
a
...a
1
11
(11j )!
(2.5)
523
Up to this point we have not fixed any of the free parameters appearing in our predictions
for the equations of motion of 11d supergravity. Now we want to fix the three free parameters c3,3 , and in a way, which finally produce the correct equation of motion for the
gauge field strength G(4) and the Killing spinor equation. We have to choose
c3,3 = 1,
2
1
= ,
7c3,3 12
8
.
12
(2.6)
7
These constraints lead to = 16
. The complete Cartan form for the fermionic generator
becomes

A(1) = dx D (0) Q + dx (0)

Q
with (D ) the Killing spinor equation of Eq. (1.7) and ( ) the operator in front
of the unphysical generator Q which is of no importance for the physical quantities.
The higher order corrections can be computed similarly. At next order O(3 ) we obtain
(0)
(0)

1
1

A(2) = D (0) {Q , Q } dx (0)
{Q , Q } dx
2
2

1
, Q } + (0)
D (0) {Q
{Q , Q } dx
(2.7)
which is an expression in the various central charges (cf. (A.6)). All terms A(k>2) vanish
due to the Jacobi identities Nos. (14)(17) and Remark 1. Putting all the results for A
together one obtains

1

1
(i)
(0)
A=
A = gB dgB + dx D
Q {Q , Q }
2
i=1

1
+ dx (0)
Q {Q , Q }

2

1
D (0) {Q , Q } + (0)
(2.8)
{Q , Q } dx .
2
It is important to notice that there is a correction term to the bosonic vielbein in front of
the momentum generator coming from the anticommutator {Q, Q}, which is proportional
to D (0) . If one imposes the Killing spinor equation the bosonic vielbein is left unchanged.
This is an a posteriori argument for the identification of the Killing spinor equation and our
notion of physical and unphysical fermionic generators.
3. N = 1 pure supergravity
A similar construction as the one before can also be used to find the Killing spinor
equation of N = 1 pure supergravity in ten dimensions [19].4 The group that was used to
construct the covariant objects of the bosonic sector of the theory was spelled out in [9],
4 The differences in the conventions to [19] are the same as described in footnote 1.
524
and was called GI . The group element is taken to be

g = ex
eha
bKa
e 8! Aa1 ...a8 R
a1 ...a8
e 6! Aa1 ...a6 R
a1 ...a6
e 2! Aa1 a2 R
a1 a2
eAR
(3.1)
and the commutators of the generators satisfy relations analogous to Eqs. (1.2), (1.3) but
with a new set of gauge field commutators, whose algebra is given by

aa
R, R a1 ...ap = cp R a1 ...ap ,
R 1 2 , R a3 ...a8 = c2,6 R a1 ...a8 ,
1
c2 = c6 = c2,6 = .
(3.2)
2
The corresponding field strengths (closure with the conformal group yields antisymmetric
tensors as described before) are
Fa1 = a1 A,
Fa1 a2 a3 = e
A2
(3.3)
(3[a1 Aa2 a3 ] ),
(3.4)
Fa1 ...a7 = e (7[a1 Aa2 ...a7 ] ),
(3.5)
Fa1 ...a9 = 9([a1 Aa2 ...a9 ] 7 2A[a1a2 a3 Aa4 ...a9 ] )
(3.6)
A
2
with the two first order equations of motion

F (3) = F (7),
F (1) = F (9) .
(3.7)
In our notation the two Killing spinor equations read

1

F ,
= D +
9
72

1
1
F F .
=
8
12
(3.8)
(3.9)
As in the case of eleven-dimensional supergravity treated previously we enhance the group

element by a fermionic generator exp( Q ) from the right. The extension of the algebra
is defined by commutation relations similarly to (2.2) and reads

Q , R a1 a2 = s2 a1 a2 Q ,

Q , R a1 ...a6 = s6 a1 ...a6 Q + q6 a1 ...a6 Q ,

.
Q , R a1 ...a8 = q8 a1 ...a8 Q
(3.10)
Since previous experience has taught us to introduce a second fermionic generator we do

it here again and discuss the reason later.
In Eq. (3.10) we have not written down commutators of the dilaton generator with the
supercharges. Actually, by checking the Jacobi identities it is found that the super extension
is inconsistent with the interpretation of the dilaton generator as an element of the Cartan
subalgebra. We are not surprised. The origin of this problem is connected to the difficulties
with the symmetric part of K a b . In footnote 2 we have used the antisymmetric -matrices
to parameterise the antisymmetric part of the K ab generators. The dilaton generator R can
be understood from an eleven-dimensional perspective as a generator built from the trace
525
Table 1
(p, q)
[Q , [R p , R q ]] constraint
(2, 2)
(2, 6)
(2, 8)
(6, 6)
(6, 8)
(8, 8)
none
c2,6 q8 = q6 (s2 q2 )
0 = q8 (s2 q2 )
none
0 = q8 (s6 q6 ) and 0 = q6 s6 (s2 q2 )
none
parts of the eleven-dimensional K a b . Since we have also realised the gauge generators
inside the Clifford algebra there is no algebraic possibility to realise the R generator inside
the Clifford algebra at the same time. Perhaps there is another mathematical technique to
get rid of the dilaton but it is unknown to us. Since we do not need the contributions from
the dilaton generator R, we have dropped it by hand; we have to keep in mind though that
finding a closing algebra including R needs further consideration.
As usually, we are mainly interested in those elements that close in the untilded Q .
The various constants si and qi are not linearly independent but have to be chosen such
that the Jacobi identities close. We have to formally define also commutators

,
Q , R a1 a2 = q2 a1 a2 Q

a ...a

a1 ...a6
1
6
Q , R
Q + q6 a1 ...a6 Q ,
= s6

(3.11)
Q , R a1 ...a8 = s8 a1 ...a8 Q .
The Jacobi identities put the constraint below on the free coefficients above:

Q , R p , R q = Q , R p , R q + R p , Q , R q
(3.12)
and the one with Q and Q exchanged. We note that in this case the commutation relations are actually carried by the gamma-matrices if we totally antisymmetrise the indices.
Evaluated on the totally antisymmetric part of the Clifford algebra it gives constraints in
Table 1.
This must be solved by setting
s2 = q2 ,
q8 = 0.
(3.13)
, [R p , R q ]] Jacobi identity are completely analogous. The

The constraints from the [Q
remaining two constants s2 , s6 are unconstrained from this point of view and can be chosen
as to generate the Killing spinor equation of pure supergravity in ten dimensions. The
Cartan form becomes

s2
g 1 dg = gB1 dgB + dx a a Faa2 a3 a2 a3
3!

a a

s6
q6
Faa2 a7 2 7 Q + dx a Faa2 a7 a2 a7
7!
7!

a a
q8
Faa2 a9 2 9 Q + .
(3.14)
9!
526
Using the equations of motion of Eq. (3.7) as in the last section to get rid of the dual field
strengths,5
(10p)(9p)
2
1
(1)p (1)
Fc0 c1 ...cp c1 ...cp =
p!
(10 p 1)!
11Gap+1 ...a9 c0 ap+1 ...a9 ,
the expression in front of Q in (3.14) simplifies to

aa
1 12s6 a1 a2 a3
a1
2 3
Fa1 a2 a3 .
a +
12s

2 a
72
7
Comparison with the Killing spinor equation (3.8) fixes
s6 =
7
,
12
3
s2 = .
4
(3.15)
Finally, let us take a look at the expression that builds up in front of Q in Eq. (3.14). After
using the equations of motion (3.7) we found indications that this could be connected with
the algebraic Killing spinor equation (3.9). If one contracts this expression with a one
obtains

q6 b1 b2 b3
Fb1 b2 b3 + q8 Fb b
in Q:
(3.16)
3!
which is exactly the expected structure. Due to the gauge algebra of GI in (3.2) which
requires c2,8 0 we have to place the stopper q8 = 0 (see Table 1) which deletes the
second term of the above equation. In D8+++ , however, c2,8 0 is not required anymore,
and thus q8 = 0 is possible [20]. Obviously the solution to the dilaton puzzle holds the key
to the completion of the picture.
The relation of this super extension to the one defined in the case of eleven-dimensional
supergravity is not understood.
4. Conclusions
We have shown that part of the original G11 group used in [6] to define M-theory as a
nonlinear realisation possesses an extension whose Cartan form produces a covariant derivative of spinors identical to the Killing spinor equation of eleven-dimensional supergravity.
It is appealing that the structure of the Killing spinor equation is inevitably generated by
the group structure. On the other hand this method is not yet expected to give the full super
covariant objects like the supercovariant field strengths including the fermionic shifts.
A similar construction was used for N = 1 pure supergravity but runs into difficulties
because of the dilaton generator that arises by dimensional reduction from eleven dimensions on a torus.
It would be interesting to see how dimensional reduction can be made consistent with
a super algebra, since the problems with the dilaton generator are generic. We expect that
b
...b
1
5
a1 ...aj = (10j )! a1 ...aj b1 ...b10j 10j 1 .
527
the approach presented here can be straightforwardly applied to other supergravity theories
and other low-level expansions of very-extended Lie algebras (see [20]) and might hold
some clues about the above stated problem. In this case it could be helpful to classify
all maximal supersymmetric version of supergravities [21,22] but also other solutions that
preserve different amounts of supersymmetry.
A crucial point turned out to be the need to introduce at least two fermionic generators.
Working with just one generator it is not possible to fix the free parameters of Eq. (2.6)
consistently. The doubling of the fermionic generators is interesting of itself and related to
the identification of positive and negative roots of a superalgebra.
Finally, it would be useful to see the relation to other approaches assuming infinite
dimensional KacMoody algebras as symmetry algebras of (super)gravity theories [23,
24]. Our approach of taking the semi-direct product with a spinor representation should be
applicable to these models, too.
Acknowledgements
A.M. and I.S. would like to thank P. Fr and P. West for discussion at an earlier stage
of the work. A.M. thanks I. Kirsch for fruitful discussions. Furthermore would I.S. like to
thank B. Julia, A. Keurentjes, A. Kleinschmidt, and H. Nicolai for interesting discussions,
explanations and generous hospitality. The work of A.M. is supported by the Deutsche
Forschungsgemeinschaft (DFG). The work of I.S. is partly supported by BSFAmerican
Israel Bi-National Science Foundation, the Israel Academy of SciencesCenters of Excellence Program, the GermanIsrael Bi-National Science Foundation, and the European
RTN-network HPRN-CT-2000-00122.
Appendix A. Jacobi identities of M-theory

In this appendix we list for completeness the set of Jacobi identities in the case of
eleven-dimensional supergravity. The Lie-bracket must be understood according to the parity (| |) of the generators as commutators and anticommutators, respectively, i.e.,
[X, Y ] = (1)|X||Y | [Y, X].
The corresponding Jacobi identity reads

X, [Y, Z] = [X, Y ], Z + (1)|X||Y | Y, [X, Z] .
(A.1)
The consistency of the purely bosonic generators was established in [6]. So we concentrate
on the Jacobis containing fermionic generators. In Table 2 we list all Jacobi identities containing at most one fermionic generator by displaying the left-hand side of Eq. (A.1). Here
is totally symmetric.
we restrict our attention to Jacobis containing Q but the case of Q
In contrast to Eq. (2.2) we define for the purpose of shortness

Q , K c d = k c d Q ,
(A.2)

Q , R c1 c2 c3 = c1 c2 c3 Q .
(A.3)
528
Table 2
One fermionic generator
No.
l.h.s. of Eq. (A.1)
Satisfied?
(1)
[Q , [Pb , Pc ]]
trivial
(2)
[Q , [K b c , K d e ]]
by constr.
(3)
[Q , [R b1 b2 b3 , R c1 c2 c3 ]]
cf. proof
(4)
[Q , [R b1 ...b6 , R c1 ...c6 ]]
satisf.
(5)
(6)
[Q , [Pb , K c d ]]
[Q , [Pb , R c1 c2 c3 ]]
[Q , [Pb , R c1 ...c6 ]]
[Q , [K b c , R c1 c2 c3 ]]
[Q , [K b c , R c1 ...c6 ]]
[Q , [R b1 b2 b3 , R c1 ...c6 ]]
trivial
trivial
(7)
(8)
(9)
(10)
trivial
cf. proof
satisf.
satisf.
If k c d forms a representation of K c d the Jacobi identity (2) is satisfied. We included

the symmetric part of K a b in order to simplify the algebra. To cut down to the algebra
of interest one simply has to restrict the Jacobi identities we are going to obtain to the
corresponding antisymmetric subalgebra, whose generators are realised in terms of the
generators of spin(1, 10).
Proof of No. (3). In the commutators of two R c1 ...c3 given in Eq. (1.4) we make the implicit
requirement on the symmetry explicit by introducing the anti symmetrisation symbol on
the left- and the right-hand side:

[c ...c
R 1 3 , R d1 ...d3 ] = c3,3 R [c1 ...c3 d1 ...d3 ] .
(A.4)
The additional projection onto the totally antisymmetric part makes a fermionic extension
possible, i.e., the algebra (A.4) can be realised on the total antisymmetric part in the Clifford multiplication.
The corresponding Jacobi identity defines the action of Q on R c1 ...c6 :
2 c c c c c c

1 2 3 4 5 6 Q .
2
Q , R c1 ...c6 =
(A.5)
c3,3
Proof of No. (8). The Jacobi identity boils down to [k c d , c1 c2 c3 ] = 3d[c1 c2 c3 ]c which
gives the Gamma-matrix identity

1 cd
[d c]
, c1 c2 c3 = 3 [c
c2 c3 ]
1
4
after projecting onto the antisymmetric part k [cd] = 14 cd .
The Jacobi identities with two fermionic generators in Table 3 define the action of the
bosonic generators G = {K a b , R c1 c2 c3 , R c1 ...c6 } on the central charges of the supersymmetry algebras, i.e., on Z , Z and A defined by
{Q , Q } = Z ,
{Q , Q } = Z ,
, Q } = A .
{Q
(A.6)
529
Table 3
Two & three fermionic generator
No.
l.h.s. of Eq. (A.1)
Satisfied?
(11)
[G, [Q , Q ]]
[G, [Q , Q ]]
by constr.
(12)
by constr.
(13)
]]
[G, [Q , Q
by constr.
(14)
[Q , [Q , Q ]]
, [Q , Q
]]
[Q
cf. proof
(15)
, [Q , Q ]]
[Q
]]
[Q , [Q , Q
(16)
(17)
cf. proof
cf. proof
cf. proof
Now we focus on the Jacobis with three fermionic generators in Table 3. The proofs of
Nos. (14) and (15) are just the statements, that Q (Q ) act trivial on the generators ap , Q }) anticommutator. The proofs of
pearing on the right hand side of the {Q , Q } ({Q
Nos. (16) and (17) just end up with

[Q , Z ] = 2[Q( , A||
) ],
, A ) ].
[Q , Z ] = 2[Q
(
Remark 1. The action of Q and Q on A fixes the action of Q and Q on Z and
Z . It is consistent to set this action to zero. Then all the three algebras Z , Z and
A are Abelian and do not mix with each other.
References
[1] E. Cremmer, B. Julia, J. Scherk, Supergravity theory in 11 dimensions, Phys. Lett. B 76 (1978) 409412.
[2] R. DAuria, P. Fr, Geometric supergravity in D = 11 and its hidden supergroup, Nucl. Phys. B 201 (1982)
101.
[3] C.M. Hull, P.K. Townsend, Unity of superstring dualities, Nucl. Phys. B 438 (1995) 109, hep-th/9410167.
[4] E. Cremmer, B. Julia, H. Lu, C.N. Pope, Dualisation of dualities, I, Nucl. Phys. B 523 (1998) 73, hepth/9710119.
[5] E. Cremmer, B. Julia, H. Lu, C.N. Pope, Dualisation of dualities, II: twisted self-duality of doubled fields
and superdualities, Nucl. Phys. B 535 (1998) 242, hep-th/9806106.
[6] P. West, Hidden superconformal symmetry in M-theory, JHEP 0008 (2000) 007, hep-th/0005270.
[7] N. Lambert, P. West, Coset symmetries in dimensionally reduced bosonic string theory, Nucl. Phys. B 615
(2001) 117, hep-th/0107209.
[8] I. Schnakenburg, P.C. West, KacMoody symmetries of IIB supergravity, Phys. Lett. B 517 (2001), hepth/0107181;
I. Schnakenburg, P.C. West, Massive IIA supergravity as a non-linear realisation, Phys. Lett. B 540 (2002)
137, hep-th/0204207.
[9] I. Schnakenburg, P. West, KacMoody symmetries of ten-dimensional non-maximal supergravity theories,
hep-th/0401196.
[10] A.B. Borisov, V.I. Ogievetsky, Theory of dynamical affine and conformal symmetries as gravity theory of
the gravitational field, Theor. Math. Phys. 21 (1975) 1179.
[11] A.B. Borisov, V.I. Ogievetsky, Inverse Higgs effect in nonlinear realisations, Teor. Mat. Fiz. 25 (2) (1975)
164177.
530
[12] P. West, E11 and M-theory, Class. Quantum Grav. 18 (2001) 31433158, hep-th/0104081.
[13] A.J. Nurmagambetov, The sigma-model representation for the duality-symmetric D = 11 supergravity, hepth/0312157.
[14] L. Paulot, Superconformal selfdual sigma-models, hep-th/0404027.
[15] P. West, E11 , SL(32) and central charges, Phys. Lett. B 575 (2003) 333342, hep-th/0307098.
[16] F. Englert, L. Houart, G+++ invariant formulation of gravity and M-theories: exact BPS solutions,
JHEP 0401 (2004) 002, hep-th/0311255.
[17] J. van Holten, A. van Proeyen, N = 1 supersymmetry algebras in D = 2, D = 3, D = 4 mod 8, J. Phys.
A 15 (1982) 3763.
[18] A. Keurentjes, The topology of U-duality (sub)groups, Class. Quantum Grav. 21 (2004) 1695, hepth/0309106;
A. Keurentjes, U duality (sub)groups and their topology, hep-th/0312134.
[19] Ali H. Chamseddine, N = 4 supergravity coupled to N = 4 matter and hidden symmetries, Nucl. Phys.
B 185 (1981) 403415.
[20] A. Kleinschmidt, I. Schnakenburg, P. West, Very-extended KacMoody algebras and their interpretation at
low levels, Class. Quantum Grav. 21 (2004) 2493, hep-th/0309198.
[21] J. Figueroa-OFarrill, G. Papadopoulos, Maximally supersymmetric solutions of ten- and elevendimensional supergravities, JHEP 0303 (2003) 048, hep-th/0211089.
[22] I. Schnakenburg, A. Miemiec, E(11) and spheric vacuum solutions of eleven- and ten-dimensional supergravity theories, hep-th/0312096.
[23] T. Damour, M. Henneaux, H. Nicolai, E(10) and a small tension expansion of M-theory, Phys. Rev.
Lett. 89 (2002), hep-th/0207267.
[24] J. Brown, O. Ganor, C. Helfgott, M-theory and E(10): billiards, branes, and imaginary roots, hepth/0401053.
CUMULATIVE AUTHOR INDEX B691B698
Abel, S.A.
Abramowicz, H.
Abt, I.
Adamczyk, L.
Adamus, M.
Adler, V.
Aghuzumtsyan, G.
Aglietti, U.
Aguilar-Saavedra, J.A.
Aharony, O.
Ahn, C.
Akemann, G.
Alimohammadi, M.
Alishahiha, M.
Alkalaev, K.B.
Antoniadis, I.
Antoniadis, I.
Antonioli, P.
Antonov, A.
Aoki, K.
Aoki, S.
Arakawa, G.
Arneodo, M.
B696 (2004) 141

B695 (2004) 3
B695 (2004) 3
B695 (2004) 3
B695 (2004) 3
B695 (2004) 3
B695 (2004) 3
B698 (2004) 277
B697 (2004) 207
B691 (2004) 3
B698 (2004) 3
B694 (2004) 59
B696 (2004) 55
B691 (2004) 111
B692 (2004) 363
B695 (2004) 103
B697 (2004) 3
B695 (2004) 3
B695 (2004) 3
B695 (2004) 132
B697 (2004) 271
B697 (2004) 462
B695 (2004) 3
Babichenko, A.
Bailey, D.S.
Bak, D.
Bamberger, A.
Barakbaev, A.N.
Barbagli, G.
Barbi, M.
Bardakci, K.
Bari, G.
Barreiro, F.
Bartels, J.
Bartsch, D.
B697 (2004) 481

B695 (2004) 3
B696 (2004) 251
B695 (2004) 3
B695 (2004) 3
B695 (2004) 3
B695 (2004) 3
B698 (2004) 202
B695 (2004) 3
B695 (2004) 3
B698 (2004) 255
B695 (2004) 3
0550-3213/2004 Published by Elsevier B.V.

doi:10.1016/S0550-3213(04)00691-1
Basile, M.
Beane, S.R.
Becker, M.
Behr, W.
Behrens, U.
Behrndt, K.
Bell, M.
Bellagamba, L.
Bellucci, S.
Beneke, M.
Benen, A.
Bertolin, A.
Bhadra, S.
Bigazzi, F.
Bijnens, J.
Binoth, T.
Bloch, I.
Bod, T.
Bonanno, A.
Bonciani, R.
Boos, E.G.
Borg, F.
Borras, K.
Boscherini, D.
Brihaye, Y.
Brock, I.
Brook, N.H.
Brower, R.
Bruckmann, F.
Brugnera, R.
Brmmer, N.
Bruni, A.
Bruni, G.
Buchbinder, I.L.
Bunk, B.
Buras, A.J.
Burguet-Castell, J.
B695 (2004) 3
B695 (2004) 192
B693 (2004) 223
B698 (2004) 473
B695 (2004) 3
B694 (2004) 99
B695 (2004) 3
B695 (2004) 3
B693 (2004) 51
B692 (2004) 232
B695 (2004) 3
B695 (2004) 3
B695 (2004) 3
B694 (2004) 3
B697 (2004) 319
B693 (2004) 134
B695 (2004) 3
B695 (2004) 3
B693 (2004) 36
B698 (2004) 277
B695 (2004) 3
B697 (2004) 319
B695 (2004) 3
B695 (2004) 3
B691 (2004) 79
B695 (2004) 3
B695 (2004) 3
B693 (2004) 149
B698 (2004) 233
B695 (2004) 3
B695 (2004) 3
B695 (2004) 3
B695 (2004) 3
B693 (2004) 51
B697 (2004) 343
B697 (2004) 133
B695 (2004) 217
532
Bscher, R.
Bussey, P.J.
Butterworth, J.M.
Bttner, C.
Bylsma, B.
B696 (2004) 468

B695 (2004) 3
B695 (2004) 3
B695 (2004) 3
B695 (2004) 3
Caldwell, A.
Callan Jr., C.G.
Capua, M.
Cara Romeo, G.
Carli, T.
Carlin, R.
Casper, D.
Catterall, C.D.
Chandrasekharan, S.
Chekanov, S.
Chiochia, V.
Chwastowski, J.
Ciborowski, J.
Ciesielski, R.
Cifarelli, L.
Cindolo, F.
Cloth, P.
Cole, J.E.
Collins-Tooth, C.
Contin, A.
Cooper-Sarkar, A.M.
Coppola, N.
Cormack, C.
Corradi, M.
Corriveau, F.
Costa, M.
Cotrone, A.L.
Cottrell, A.
Curio, G.
Curio, G.
Cvetic, M.
Cvetic, M.
B695 (2004) 3
B694 (2004) 115
B695 (2004) 3
B695 (2004) 3
B695 (2004) 3
B695 (2004) 3
B695 (2004) 217
B695 (2004) 3
B693 (2004) 149
B695 (2004) 3
B695 (2004) 3
B695 (2004) 3
B695 (2004) 3
B695 (2004) 3
B695 (2004) 3
B695 (2004) 3
B695 (2004) 3
B695 (2004) 3
B695 (2004) 3
B695 (2004) 3
B695 (2004) 3
B695 (2004) 3
B695 (2004) 3
B695 (2004) 3
B695 (2004) 3
B695 (2004) 3
B694 (2004) 3
B695 (2004) 3
B693 (2004) 195
B693 (2004) 223
B696 (2004) 298
B698 (2004) 163
DAgostini, G.
Dal Corso, F.
DallAgata, G.
Danilov, P.
Dannheim, D.
DAuria, R.
Davidse, M.
de Boer, J.
de Haro, S.
Della Morte, M.
del Peso, J.
Dementiev, R.K.
De Pasquale, S.
Dereli, T.
Derendinger, J.-P.
B695 (2004) 3
B695 (2004) 3
B695 (2004) 243
B695 (2004) 3
B695 (2004) 3
B693 (2004) 261
B697 (2004) 48
B696 (2004) 174
B696 (2004) 174
B697 (2004) 343
B695 (2004) 3
B695 (2004) 3
B695 (2004) 3
B691 (2004) 223
B691 (2004) 233
Derrick, M.
Descotes-Genon, S.
Devenish, R.C.E.
de Wolf, E.
Dhawan, S.
DHoker, E.
Diener, K.-P.O.
Dijkstra, T.P.T.
Dobur, D.
Dolgoshein, B.A.
Donagi, R.
Dorey, P.
Dorey, P.
Dorsner, I.
Doyle, A.T.
Drews, G.
Dubovsky, S.L.
Durkin, L.S.
Dusini, S.
B695 (2004) 3
B693 (2004) 103
B695 (2004) 3
B695 (2004) 3
B695 (2004) 3
B695 (2004) 132
B697 (2004) 387
B698 (2004) 450
B695 (2004) 3
B695 (2004) 3
B694 (2004) 187
B696 (2004) 445
B697 (2004) 405
B698 (2004) 386
B695 (2004) 3
B695 (2004) 3
B691 (2004) 91
B695 (2004) 3
B695 (2004) 3
Eisenberg, Y.
Emig, T.
Enger, H.
Ermolov, P.F.
Eskreys, A.
Eto, M.
Everett, A.
Eynard, B.
B695 (2004) 3
B696 (2004) 468
B695 (2004) 73
B695 (2004) 3
B695 (2004) 3
B696 (2004) 3
B695 (2004) 3
B694 (2004) 443
Fadin, V.S.
Faraggi, A.E.
Faraggi, A.E.
Fateev, V.A.
Feigin, B.L.
Felder, G.
Feng, B.
Ferrando, J.
Ferrara, S.
Ferrero, M.I.
Figiel, J.
Filges, D.
Fioravanti, D.
Fischbacher, T.
Fleischer, R.
Foster, B.
Foudas, C.
Fourletov, S.
Fourletova, J.
Fr, P.
Fricke, U.
Fuchs, J.
Fuji, H.
Fusayasu, T.
Fyodorov, Y.V.
B698 (2004) 255

B694 (2004) 187
B695 (2004) 41
B696 (2004) 301
B698 (2004) 409
B691 (2004) 251
B698 (2004) 3
B695 (2004) 3
B693 (2004) 261
B695 (2004) 3
B695 (2004) 3
B695 (2004) 3
B696 (2004) 445
B694 (2004) 525
B697 (2004) 133
B695 (2004) 3
B695 (2004) 3
B695 (2004) 3
B695 (2004) 3
B694 (2004) 239
B695 (2004) 3
B694 (2004) 277
B698 (2004) 53
B695 (2004) 3
B694 (2004) 59
Gabareen, A.
Gaberdiel, M.R.
Gade, R.M.
Galas, A.
Gallo, E.
Garca-Compen, H.
Garfagnini, A.
Gasperini, M.
Gato-Rivera, B.
Gattringer, C.
Gehrmann, T.
Gehrmann-De Ridder, A.
Geiser, A.
Genta, C.
Ghodsi, A.
Gialas, I.
Gibbons, G.W.
Giovannini, M.
Giusti, P.
Giveon, A.
Gladilin, L.K.
Gladkov, D.
Glasman, C.
Gliga, S.
Glover, E.W.N.
Gckeler, M.
Goers, S.
Goh, K.-I.
Gmez-Cadenas, J.J.
Gomis, J.
Gonalo, R.
Gonzlez, O.
Gonzlez-Daz, P.F.
Gosau, T.
Gttlicher, P.
Gould, M.D.
Grabowska-Bod, I.
Gracey, J.A.
Grijpink, S.
Grzelak, G.
Gnaydin, M.
Gutjahr, P.
Gutsche, O.
Gwenlan, C.
B695 (2004) 3
B693 (2004) 281
B694 (2004) 354
B695 (2004) 3
B695 (2004) 3
B694 (2004) 405
B695 (2004) 3
B694 (2004) 206
B698 (2004) 450
B694 (2004) 170
B691 (2004) 195
B691 (2004) 195
B695 (2004) 3
B695 (2004) 3
B691 (2004) 111
B695 (2004) 3
B697 (2004) 225
B694 (2004) 206
B695 (2004) 3
B691 (2004) 3
B695 (2004) 3
B695 (2004) 3
B695 (2004) 3
B695 (2004) 3
B691 (2004) 195
B694 (2004) 170
B695 (2004) 3
B696 (2004) 351
B695 (2004) 217
B696 (2004) 263
B695 (2004) 3
B695 (2004) 3
B697 (2004) 363
B695 (2004) 3
B695 (2004) 3
B698 (2004) 503
B695 (2004) 3
B696 (2004) 295
B695 (2004) 3
B695 (2004) 3
B697 (2004) 399
B692 (2004) 110
B695 (2004) 3
B695 (2004) 3
Haas, T.
Hadasz, L.
Hain, W.
Hall-Wilton, R.
Hamatsu, R.
Hamazaki, T.
Hambye, T.
Hamilton, J.
Hanlon, S.
B695 (2004) 3
B694 (2004) 493
B695 (2004) 3
B695 (2004) 3
B695 (2004) 3
B698 (2004) 335
B695 (2004) 169
B695 (2004) 3
B695 (2004) 3
533
Hart, J.C.
Hartmann, B.
Hartmann, H.
Hartner, G.
Heaphy, E.A.
Heath, G.P.
Heinrich, G.
Helbich, M.
Hernndez, P.
Heusch, C.
Hilger, E.
Hillert, S.
Hirose, T.
Hochman, D.
Holland, K.
Holm, U.
Horn, C.
Horsley, R.
Huber, P.
Hung, P.Q.
Hwang, D.S.
Hyakutake, Y.
B695 (2004) 3
B691 (2004) 79
B695 (2004) 3
B695 (2004) 3
B695 (2004) 3
B695 (2004) 3
B693 (2004) 134
B695 (2004) 3
B695 (2004) 217
B695 (2004) 3
B695 (2004) 3
B695 (2004) 3
B695 (2004) 3
B695 (2004) 3
B694 (2004) 35
B695 (2004) 3
B695 (2004) 3
B693 (2004) 3
B694 (2004) 170
B692 (2004) 83
B696 (2004) 413
B696 (2004) 251
Iacobucci, G.
Ichinose, I.
Iga, Y.
Inuzuka, M.
Irrgang, P.
Ivanov, E.A.
Ivanov, R.
B695 (2004) 3
B697 (2004) 462
B695 (2004) 3
B695 (2004) 3
B695 (2004) 3
B694 (2004) 473
B694 (2004) 509
Jakob, H.-P.
Jansen, K.
Jasklski, Z.
Jeschek, C.
Jones, T.W.
B695 (2004) 3
B697 (2004) 343
B694 (2004) 493
B694 (2004) 99
B695 (2004) 3
Kagawa, S.
Kahle, B.
Kahng, B.
Kaji, H.
Kamimura, K.
Kananov, S.
Karabali, D.
Karmanov, V.A.
Karshon, U.
Karstens, F.
Kataoka, M.
Katkov, I.I.
Kayaba, Y.
Kira, D.
Kenna, R.
Keurentjes, A.
Khalil, S.
B695 (2004) 3
B695 (2004) 3
B696 (2004) 351
B695 (2004) 3
B696 (2004) 263
B695 (2004) 3
B697 (2004) 513
B696 (2004) 413
B695 (2004) 3
B695 (2004) 3
B695 (2004) 3
B695 (2004) 3
B697 (2004) 271
B695 (2004) 3
B691 (2004) 292
B697 (2004) 302
B695 (2004) 313
534
Khein, L.A.
Khorrami, M.
Kim, D.
Kim, J.Y.
Kimura, Y.
Kind, O.
Kisielewska, D.
Kitamura, S.
Kiyo, Y.
Klemm, H.
Knechtli, F.
Kniehl, B.A.
Koffeman, E.
Kohlprath, E.
Kohno, T.
Kokotov, A.
Koma, M.
Koma, Y.
Kooijman, P.
Koop, T.
Korotkin, D.
Korzhavina, I.A.
Kotanski, A.
Ktz, U.
Kounnas, C.
Kounnas, C.
Kowal, A.M.
Kowal, M.
Kowalski, H.
Kramberger, G.
Krause, A.
Krause, A.
Kreisel, A.
Kruczenski, M.
Krumnack, N.
Krykhtin, V.A.
Kuramashi, Y.
Kutasov, D.
Kuze, M.
Kuzenko, S.M.
Kuzmin, V.A.
B695 (2004) 3
B696 (2004) 55
B696 (2004) 351
B695 (2004) 3
B692 (2004) 394
B695 (2004) 3
B695 (2004) 3
B695 (2004) 3
B692 (2004) 232
B693 (2004) 281
B697 (2004) 343
B695 (2004) 199
B695 (2004) 3
B697 (2004) 243
B695 (2004) 3
B694 (2004) 443
B692 (2004) 209
B692 (2004) 209
B695 (2004) 3
B695 (2004) 3
B694 (2004) 443
B695 (2004) 3
B695 (2004) 3
B695 (2004) 3
B691 (2004) 233
B695 (2004) 41
B695 (2004) 3
B695 (2004) 3
B695 (2004) 3
B695 (2004) 3
B693 (2004) 195
B693 (2004) 223
B695 (2004) 3
B692 (2004) 3
B695 (2004) 3
B693 (2004) 51
B697 (2004) 271
B691 (2004) 3
B695 (2004) 3
B697 (2004) 89
B695 (2004) 3
Labarga, L.
Labes, H.
Lacagnina, G.
Lainesse, J.
Lammers, S.
Lang, C.B.
Lashkevich, M.
Lebedev, O.
Lee, D.-S.
Lelas, D.
Levchenko, B.B.
Levy, A.
B695 (2004) 3
B695 (2004) 3
B693 (2004) 36
B695 (2004) 3
B695 (2004) 3
B694 (2004) 170
B696 (2004) 301
B696 (2004) 141
B696 (2004) 351
B695 (2004) 3
B695 (2004) 3
B695 (2004) 3
Li, G.-L.
Li, L.
Li, T.
Lightwood, M.S.
Lim, H.
Lim, I.T.
Limentani, S.
Lin, Y.
Ling, T.Y.
Lipatov, L.N.
Liu, T.
Liu, X.
Loaiza-Brito, O.
Lhr, B.
Lohrmann, E.
Loizides, J.H.
Long, K.R.
Longhin, A.
ukasik, J.
Lukina, O.Yu.
Lst, D.
Ltken, C.A.
uzniak, P.
B696 (2004) 381

B695 (2004) 3
B698 (2004) 163
B695 (2004) 3
B695 (2004) 3
B695 (2004) 3
B695 (2004) 3
B695 (2004) 169
B695 (2004) 3
B698 (2004) 255
B698 (2004) 163
B695 (2004) 3
B694 (2004) 405
B695 (2004) 3
B695 (2004) 3
B695 (2004) 3
B695 (2004) 3
B695 (2004) 3
B695 (2004) 3
B695 (2004) 3
B696 (2004) 205
B695 (2004) 73
B695 (2004) 3
Ma, K.J.
Maddox, E.
Magill, S.
Majumdar, P.
Mankel, R.
Margotti, A.
Marini, G.
Martin, J.F.
Martucci, L.
Maru, N.
Mastroberardino, A.
Matsui, T.
Matsuzawa, K.
Mattingly, M.C.K.
Mayr, P.
McArthur, I.N.
McCubbin, N.A.
McInnes, B.
McLoughlin, T.
Melzer-Pellmann, I.-A.
Menary, S.
Metlica, F.
Meyer, U.
Michishita, Y.
Miemiec, A.
Miglioranzi, S.
Milite, M.
Miramontes, J.L.
Mirea, A.
Mizoguchi, S.
B695 (2004) 3
B695 (2004) 3
B695 (2004) 3
B692 (2004) 209
B695 (2004) 3
B695 (2004) 3
B695 (2004) 3
B695 (2004) 3
B694 (2004) 3
B696 (2004) 3
B695 (2004) 3
B697 (2004) 462
B695 (2004) 3
B695 (2004) 3
B696 (2004) 205
B697 (2004) 89
B695 (2004) 3
B692 (2004) 270
B694 (2004) 115
B695 (2004) 3
B695 (2004) 3
B695 (2004) 3
B695 (2004) 3
B698 (2004) 111
B698 (2004) 517
B695 (2004) 3
B695 (2004) 3
B697 (2004) 405
B695 (2004) 3
B698 (2004) 53
Moch, S.
Moghimi-Araghi, S.
Mohapatra, R.
Monaco, V.
Montanari, A.
Mueller, A.H.
Musgrave, B.
B691 (2004) 129

B696 (2004) 492
B695 (2004) 313
B695 (2004) 3
B695 (2004) 3
B692 (2004) 175
B695 (2004) 3
Nagano, K.
Nair, V.P.
Nair, V.P.
Nakayama, R.
Namsoo, T.
Nania, R.
Narayanan, R.
Nayak, R.R.
Neuberger, H.
Nguyen, C.N.
Nigro, A.
Ning, Y.
Ngrdi, D.
Nomura, Y.
Nooij, S.E.M.
Notari, A.
Notz, D.
Nowak, R.J.
Nuncio-Quiroz, A.E.
B695 (2004) 3
B691 (2004) 182
B697 (2004) 513
B693 (2004) 176
B695 (2004) 3
B695 (2004) 3
B696 (2004) 107
B698 (2004) 149
B696 (2004) 107
B695 (2004) 3
B695 (2004) 3
B695 (2004) 3
B698 (2004) 233
B698 (2004) 92
B695 (2004) 41
B695 (2004) 169
B695 (2004) 3
B695 (2004) 3
B695 (2004) 3
Oh, B.Y.
Ohno, T.
Ohta, N.
Olkiewicz, K.
Ookouchi, Y.
Ossola, G.
B695 (2004) 3
B697 (2004) 462
B696 (2004) 251
B695 (2004) 3
B698 (2004) 3
B691 (2004) 259
Pac, M.Y.
Padhi, S.
Palmonari, F.
Panigrahi, K.L.
Papadimitriou, I.
Papucci, M.
Parenti, A.
Park, I.H.
Patel, S.
Paul, E.
Pavel, N.
Pawlak, J.M.
Pelfer, P.G.
Pellegrino, A.
Pepe, M.
Perlt, H.
Pesci, A.
Phong, D.H.
Piclum, J.H.
B695 (2004) 3
B695 (2004) 3
B695 (2004) 3
B698 (2004) 149
B696 (2004) 298
B695 (2004) 169
B695 (2004) 3
B695 (2004) 3
B695 (2004) 3
B695 (2004) 3
B695 (2004) 3
B695 (2004) 3
B695 (2004) 3
B695 (2004) 3
B694 (2004) 35
B693 (2004) 3
B695 (2004) 3
B695 (2004) 132
B695 (2004) 199
535
Pilaftsis, A.
Piotrzkowski, K.
Plefka, J.
Plucinski, P.
Pokrovskiy, N.S.
Polini, A.
Pope, C.N.
Posocco, M.
Pozzorini, S.
Proskuryakov, A.S.
Przybycien, M.
B692 (2004) 303

B695 (2004) 3
B692 (2004) 110
B695 (2004) 3
B695 (2004) 3
B695 (2004) 3
B697 (2004) 225
B695 (2004) 3
B692 (2004) 135
B695 (2004) 3
B695 (2004) 3
QCDSF Collaboration
B693 (2004)
Rakow, P.E.L.
Ramrez, T.
Randjbar-Daemi, S.
Rautenberg, J.
Raval, A.
Recksiegel, S.
Reeder, D.D.
Ren, Z.
Renner, R.
Repond, J.
Riccioni, F.
Richter, R.
Riederer, S.
Rim, C.
Rinaldi, L.
Riser, R.
Riveline, M.
Rizos, J.
Robins, S.
Rosin, M.
Ross, G.G.
Rouhani, S.
Ruan, J.
Rulik, K.
Runkel, I.
Rurua, L.
Ruspa, M.
Ryan, P.
Ryzhov, A.V.
Ryzhov, A.V.
B693 (2004) 3
B696 (2004) 263
B692 (2004) 346
B695 (2004) 3
B695 (2004) 3
B697 (2004) 133
B695 (2004) 3
B695 (2004) 3
B695 (2004) 3
B695 (2004) 3
B698 (2004) 450
B696 (2004) 205
B693 (2004) 149
B696 (2004) 445
B695 (2004) 3
B691 (2004) 251
B695 (2004) 3
B695 (2004) 41
B695 (2004) 3
B695 (2004) 3
B692 (2004) 50
B696 (2004) 492
B692 (2004) 417
B694 (2004) 239
B694 (2004) 277
B695 (2004) 3
B695 (2004) 3
B695 (2004) 3
B692 (2004) 3
B698 (2004) 132
Saadat, M.
Sacchi, R.
Sachrajda, C.T.
Sadooghi, N.
Sakai, N.
Salehi, H.
Snchez, F.
Santiago, J.
Sarkissian, G.
B696 (2004) 492

B695 (2004) 3
B693 (2004) 103
B691 (2004) 111
B696 (2004) 3
B695 (2004) 3
B695 (2004) 217
B696 (2004) 141
B696 (2004) 66
536
Sartorelli, G.
Savin, A.A.
Saxon, D.H.
Schagen, S.
Schellekens, A.N.
Schierholz, G.
Schiller, A.
Schioppa, M.
Schlenstedt, S.
Schleper, P.
Schmidke, W.B.
Schnakenburg, I.
Schneekloth, U.
Schnitzer, H.J.
Schnitzer, H.J.
Schrner-Sadenius, T.
Schuller, F.P.
Schwab, F.
Schweigert, C.
Sciulli, F.
Seco, M.
Semikhatov, A.M.
Sezgin, E.
Shaynkman, O.V.
Shcheglova, L.M.
Shen, Z.
Shi, K.-J.
Shigemori, M.
Shimono, Y.
Shiu, G.
Shoshi, A.I.
Sibiryakov, S.M.
Sigenza, C.L.
Sirvanl,
B.B.
Siwach, S.
Skillicorn, I.O.
Sominski, W.
Smilga, A.V.
Smirnov, A.Yu.
Smith, W.H.
Smolyakov, M.N.
Soares, M.
Soddu, A.
Sokal, A.D.
Solano, A.
Son, D.
Sosnovtsev, V.
Splittorff, K.
Stairs, D.G.
Stanco, L.
Standage, J.
Steinhauser, M.
Stieberger, S.
Stifutkin, A.
B695 (2004) 3
B695 (2004) 3
B695 (2004) 3
B695 (2004) 3
B698 (2004) 450
B693 (2004) 3
B693 (2004) 3
B695 (2004) 3
B695 (2004) 3
B695 (2004) 3
B695 (2004) 3
B698 (2004) 517
B695 (2004) 3
B695 (2004) 267
B695 (2004) 283
B695 (2004) 3
B698 (2004) 319
B697 (2004) 133
B694 (2004) 277
B695 (2004) 3
B692 (2004) 83
B698 (2004) 409
B692 (2004) 346
B692 (2004) 363
B695 (2004) 3
B692 (2004) 417
B696 (2004) 381
B698 (2004) 3
B693 (2004) 176
B696 (2004) 298
B692 (2004) 175
B691 (2004) 91
B697 (2004) 363
B692 (2004) 249
B698 (2004) 149
B695 (2004) 3
B695 (2004) 3
B694 (2004) 473
B698 (2004) 386
B695 (2004) 3
B695 (2004) 301
B695 (2004) 3
B692 (2004) 83
B691 (2004) 259
B695 (2004) 3
B695 (2004) 3
B695 (2004) 3
B695 (2004) 84
B695 (2004) 3
B695 (2004) 3
B695 (2004) 3
B695 (2004) 199
B696 (2004) 205
B695 (2004) 3
Stonjek, S.
Stopa, P.
Stsslein, U.
Straub, P.B.
Strumia, A.
Suchkov, S.
Susinno, G.
Suszycki, L.
Sutiak, J.
Sutton, M.R.
Swanson, I.
Sykora, A.
Sztuk, J.
Szuba, D.
Szuba, J.
B695 (2004) 3
B695 (2004) 3
B695 (2004) 3
B695 (2004) 3
B695 (2004) 169
B695 (2004) 3
B695 (2004) 3
B695 (2004) 3
B695 (2004) 3
B695 (2004) 3
B694 (2004) 115
B698 (2004) 473
B695 (2004) 3
B695 (2004) 3
B695 (2004) 3
Takemae, S.
Tandler, J.
Tapper, A.D.
Targett-Adams, C.
Tassi, E.
Tateo, R.
Tawara, T.
Taylor, T.R.
Terrn, J.
Theis, U.
Tiecke, H.
Tokushuku, K.
Trigiante, M.
Trigiante, M.
Tseytlin, A.A.
Tseytlin, A.A.
Tsurugai, T.
Tucker-Smith, D.
Tuckmantel, M.
Turan, G.
Turcato, M.
Tymieniecka, T.
B697 (2004) 399

B695 (2004) 3
B695 (2004) 3
B695 (2004) 3
B695 (2004) 3
B696 (2004) 445
B695 (2004) 3
B695 (2004) 103
B695 (2004) 3
B697 (2004) 48
B695 (2004) 3
B695 (2004) 3
B693 (2004) 261
B694 (2004) 239
B692 (2004) 3
B698 (2004) 132
B695 (2004) 3
B698 (2004) 92
B697 (2004) 3
B692 (2004) 249
B695 (2004) 3
B695 (2004) 3
Uehara, S.
Ukleja, A.
Ukleja, J.
Underwood, T.E.J.
B696 (2004) 36
B695 (2004) 3
B695 (2004) 3
B692 (2004) 303
van Baal, P.
Vandoren, S.
Vasiliev, M.A.
Vzquez, M.
Velasco-Sevilla, L.
Veneziano, G.
Verbaarschot, J.J.M.
Vermaseren, J.A.M.
Vernizzi, G.
Vives, O.
B698 (2004) 233

B697 (2004) 48
B692 (2004) 363
B695 (2004) 3
B692 (2004) 50
B694 (2004) 206
B695 (2004) 84
B691 (2004) 129
B694 (2004) 59
B692 (2004) 50
Vlasov, N.N.
Vogt, A.
Voss, K.C.
B695 (2004) 3
B691 (2004) 129
B695 (2004) 3
Walczak, R.
Walsh, R.
Wang, M.
West, P.
Whitmore, J.J.
Wichmann, K.
Wick, K.
Wiese, U.-J.
Wiese, U.-J.
Wiggers, L.
Wing, M.
Wohlfarth, M.N.R.
Wolf, G.
B695 (2004) 3
B695 (2004) 3
B695 (2004) 3
B693 (2004) 76
B695 (2004) 3
B695 (2004) 3
B695 (2004) 3
B693 (2004) 149
B694 (2004) 35
B695 (2004) 3
B695 (2004) 3
B698 (2004) 319
B695 (2004) 3
Yamada, S.
Yamada, S.
Yamashita, T.
Yamazaki, Y.
Yang, D.s.
B695 (2004) 3
B696 (2004) 36
B695 (2004) 3
B695 (2004) 3
B692 (2004) 232
537
Yang, W.-L.
Yelnikov, A.
Ylmaz, N.T.
Ylmaz, U.O.
Yoshida, R.
Youngman, C.
Yue, R.-H.
B698 (2004) 503

B691 (2004) 182
B691 (2004) 223
B692 (2004) 249
B695 (2004) 3
B695 (2004) 3
B696 (2004) 381
Zamaklar, M.
Zambrana, M.
Zarnecki,
A.F.
Zawiejski, L.
Zeuner, W.
ZEUS Collaboration
Zhang, Y.-Z.
Zhautykov, B.O.
Zhu, W.
Zichichi, A.
Ziegler, A.
Ziegler, Ar.
Zotkin, S.A.
Zwirner, F.
B696 (2004) 66
B695 (2004) 3
B695 (2004) 3
B695 (2004) 3
B695 (2004) 3
B695 (2004) 3
B698 (2004) 503
B695 (2004) 3
B692 (2004) 417
B695 (2004) 3
B695 (2004) 3
B695 (2004) 3
B695 (2004) 3
B691 (2004) 233

Nucl - Phys.B v.698

Încărcat de

Informații document

Titlu original

Drepturi de autor

Formate disponibile

Partajați acest document

Partajați sau inserați document

Opțiuni de partajare

Vi se pare util acest document?

Este necorespunzător acest conținut?

Drepturi de autor:

Formate disponibile

Nucl - Phys.B v.698

Încărcat de

Drepturi de autor:

Formate disponibile

Nuclear Physics B 698 (2004) 352

Supersymmetric gauge theories with flavors

Received 24 May 2004; accepted 15 July 2004

E-mail addresses: ahn@ias.edu (C. Ahn), fengb@ias.edu (B. Feng), ookouchi@th.phys.titech.ac.jp

C. Ahn et al. / Nuclear Physics B 698 (2004) 352

C. Ahn et al. / Nuclear Physics B 698 (2004) 352

In Section 2, as background, we review basic materials for N = 1 supersymmetric

C. Ahn et al. / Nuclear Physics B 698 (2004) 352

B(z) det m(z) = BL

In the pseudo-confining vacuum, every eigenvalue of is equal to ai for some i. On the

1 We set w (z) 1 Tr W to zero because in supersymmetric vacua w (z) = 0.

C. Ahn et al. / Nuclear Physics B 698 (2004) 352

C. Ahn et al. / Nuclear Physics B 698 (2004) 352

Finally, the exact, effective glueball superpotential is given by [1]

i(2Nc L)S + 2i0 S + 2i

In the above general solutions (2.8), (2.9), we have rI = 1 or rI = 0, depending on

C. Ahn et al. / Nuclear Physics B 698 (2004) 352

This corresponds to taking polynomials in (2.1) as

C. Ahn et al. / Nuclear Physics B 698 (2004) 352

The gauge group is higgsed down to U (Nc K) where

branch is the intersection of the Coulomb branch in which QQ

we saw that there is a period

C. Ahn et al. / Nuclear Physics B 698 (2004) 352

C. Ahn et al. / Nuclear Physics B 698 (2004) 352

3. One cut modelquadratic tree level superpotential

C. Ahn et al. / Nuclear Physics B 698 (2004) 352

in the previous section, one can compute

that the sign convention is such that

very large. Substituting this into (2.12), we obtain

C. Ahn et al. / Nuclear Physics B 698 (2004) 352

2zI + 2(2rI 1)zI zI2 ,

4 We would like to thank K. Intriligator for explaining this point to us.

C. Ahn et al. / Nuclear Physics B 698 (2004) 352

3.2. The on-shell solution

C. Ahn et al. / Nuclear Physics B 698 (2004) 352

where we have defined dimensionless quantities S =

and z f = f . Note that

C. Ahn et al. / Nuclear Physics B 698 (2004) 352

Therefore, as we change p, the position of the branch points changes according to

C. Ahn et al. / Nuclear Physics B 698 (2004) 352

(p becomes smaller), the cut shrinks (because of p 2 ) and rotates counterclockwise

C. Ahn et al. / Nuclear Physics B 698 (2004) 352

C. Ahn et al. / Nuclear Physics B 698 (2004) 352

C. Ahn et al. / Nuclear Physics B 698 (2004) 352

while on the second sheet

C. Ahn et al. / Nuclear Physics B 698 (2004) 352

Therefore, if we solve the equation of motion (3.10), in general we expect multiple

C. Ahn et al. / Nuclear Physics B 698 (2004) 352

C. Ahn et al. / Nuclear Physics B 698 (2004) 352

and the solution is

C. Ahn et al. / Nuclear Physics B 698 (2004) 352

C. Ahn et al. / Nuclear Physics B 698 (2004) 352

4. Two cut modelcubic tree level superpotential

Here we wrote down W (z) in terms of g2 = g, g1 = m for definiteness, but mostly we

C. Ahn et al. / Nuclear Physics B 698 (2004) 352

which in this case is

C. Ahn et al. / Nuclear Physics B 698 (2004) 352

with i = 1, 2. The periods f , f

are associated with the poles on the first sheet and

C. Ahn et al. / Nuclear Physics B 698 (2004) 352

Fig. 6. Under the constraint (4.3), contours C1 and C2 become degenerate: C1 = C2 .

C. Ahn et al. / Nuclear Physics B 698 (2004) 352

The parameters B, C can be written in terms of the parameters in

In our present case, since only