Nucl - Phys.B v.644

Nuclear Physics B 644 (2002) 3–20
www.elsevier.com/locate/npe
Matrix models, topological strings, and

supersymmetric gauge theories
Robbert Dijkgraaf a , Cumrun Vafa b
a Institute for Theoretical Physics & Korteweg-de Vries Institute for Mathematics,
University of Amsterdam, 1018 TV Amsterdam, The Netherlands
b Jefferson Physical Laboratory, Harvard University, Cambridge, MA 02138, USA
Received 15 August 2002; accepted 23 August 2002
Abstract
We show that B-model topological strings on local Calabi–Yau threefolds are large-N duals of
matrix models, which in the planar limit naturally give rise to special geometry. These matrix
models directly compute F-terms in an associated N = 1 supersymmetric gauge theory, obtained by
deforming N = 2 theories by a superpotential term that can be directly identified with the potential
of the matrix model. Moreover by tuning some of the parameters of the geometry in a double scaling
limit we recover (p, q) conformal minimal models coupled to 2d gravity, thereby relating non-critical
string theories to type II superstrings on Calabi–Yau backgrounds.
 2002 Published by Elsevier Science B.V.
1. Introduction
Large-N limits of U (N) gauge theories have been a source of inspiration in physics,
ever since ’t Hooft introduced the idea [1]. In particular the large-N limit of gauge theories
should be equivalent to some kind of closed string theory. The first contact this idea had
with string theory was in the context of non-critical bosonic strings described by c 1
conformal field theories coupled to two-dimensional gravity. It was found that by taking a
“double scaling limit” of N × N matrix models, where one send N to infinity while at the
same time going to some critical point, one ends up with non-critical bosonic strings [2,3].
This relation between gauge systems and strings was not exactly in the sense that ’t Hooft
originally suggested for the large-N expansion, as it involved a double scaling limit. In
E-mail address: rhd@science.uva.nl (R. Dijkgraaf).
0550-3213/02/$ – see front matter  2002 Published by Elsevier Science B.V.

PII: S 0 5 5 0 - 3 2 1 3 ( 0 2 ) 0 0 7 6 6 - 6
4 R. Dijkgraaf, C. Vafa / Nuclear Physics B 644 (2002) 3–20
particular, before taking this limit the matrix model would not have a string dual, whereas
according to ’t Hooft’s general idea one would have expected it to have.
In one context this was remedied by a different kind of matrix model introduced by
Kontsevich [4], where without taking a double scaling limit one finds an equivalence
between a matrix model and non-critical string theory. In particular the amplitudes of
the topological string observables introduced in [5] are directly computed by these matrix
integrals. This duality was in the same spirit as ’t Hooft’s general idea and can be seen as a
low-dimensional example of a holographic correspondence.
More recently large-N dualities have come back in various forms. In the context of
M-theory a large-N matrix formulation was advanced [6]. Since here an unconventional
large-N limit is involved, this again was not quite in the same spirit as ’t Hooft’s idea,
much as the double scaling limit of matrix models in the context of non-critical strings is
not. However, the AdS/CFT correspondence [7] is in the same spirit as ’t Hooft’s original
proposal in that one did not have to take a particular limit to obtain an equivalence.
Another example of such a strict large-N duality is the relation between Chern–Simons
gauge theory and A-model topological strings [8] where one also does not have to take any
particular limit for the equivalence to hold.
The main aim of this paper is to develop a mirror version of this last duality [8]. We
find matrix models that are dual to B-model topological strings on Calabi–Yau threefolds.
This is again in the same spirit as ’t Hooft, as one does not take a double scaling limit. We
will show in particular that the special geometry of Calabi–Yau threefolds that solves the
B-model at tree level emerges naturally from the dynamics of the eigenvalues of the matrix
model.
However, even though it is not required, one can also consider a double scaling limit
of this setup and obtain a specific class of Calabi–Yau manifolds that are dual to double-
scaled matrix models. In this sense we are enlarging the original equivalence of double
scaling limits of matrix models with string theory to an equivalence of all matrix models
with some kind of closed string theory, without any need to take a double scaling limit.
In particular our result shows that studying strings on non-compact Calabi–Yau spaces
provides a unifying approach to all matrix model descriptions.
Furthermore, it turns out that one can embed these large-N dualities in the context of
type IIB superstrings [9], a relation that was further explored in [10–15]. In the context of
this embedding one obtains a dictionary in which the planar limit of the matrix models is
seen to compute superpotentials for certain N = 1 supersymmetric gauge theories, where
the potential of the matrix model gets mapped to the superpotential for an adjoint scalar of
an N = 1 theory.
The organization of this paper is as follows: in Section 2 we propose the large-N con-
jecture for topological strings with matrix models after reviewing the various geometrical
ingredients. In Section 3 we pass this conjecture through some highly non-trivial checks,
and in particular check it at the planar limit. In Section 4 we discuss some generalizations
of this conjecture and its connections with non-critical bosonic strings coupled to gravity
and the double scaling limit. We also discuss the meaning of the double scaling limit from
the viewpoint of type IIB superstrings.
R. Dijkgraaf, C. Vafa / Nuclear Physics B 644 (2002) 3–20 5
2. Large-N topological string conjectures
In this section we formulate a general class of large-N conjectures involving B-model

topological strings and certain two-dimensional topological gauge theories. The idea is
to consider the mirror of the large-N conjecture of [8], which relates large-N Chern–
Simons theory on S 3 with A-model topological strings on the resolved conifold. This
duality involves Calabi–Yau threefold transitions where a 3-cycle with branes wrapped
over it shrinks and a 2-cycle grows without any branes wrapped over it. A worldsheet
derivation of this duality has been recently presented in [16]. Moreover this has been
generalized to a large class of Calabi–Yau threefolds [17] which has been further studied in
[18] leading to development of powerful methods to compute all loop A-model topological
string amplitudes [19,20].
As suggested in [8] the mirror of A-model transitions dualities should also exist, namely
we can consider 2-cycles with branes wrapped over them, undergoing transitions where
they shrink and are replaced by 3-cycles without any branes. A large class of such Calabi–
Yau transitions were considered in [10–14] in the context of embedding such B-model
dualities in type IIB superstrings. However, the duality between gauge theory and the
topological B-model itself, which is implicit in these works, has not been studied. The aim
of this section is to elaborate on these topological B-model/large-N gauge theory dualities.
2.1. Geometry of the generalized conifold transition
Instead of being general we consider a special class of such transitions, studied in [10],
and discuss its topological lift. The situation considered in [10] involved a string theory
realization of N = 2 supersymmetric U (N) gauge theory, deformed to N = 1 theory by
addition of a tree-level superpotential Tr W (Φ), which we take to be a general polynomial
of degree n + 1 of the adjoint field Φ. The Calabi–Yau geometry relevant for this was
studied in [21] and corresponds to considering the blowup of the local threefold given as a
hypersurface
uv + y 2 + W (x)2 = 0. (2.1)
The blow up takes place at the critical points of W , i.e., at W (x) = 0. Such transitions lead
to a geometry that contains n blown up P1 ’s which are all in the same homology class. So
in the resolved geometry we can find n isolated rational curves.
More precisely, the resolved singularity can be obtained by starting with the bundle
O(0) ⊕ O(−2) over P1 . This is the normal bundle to rational curve in K3 × C and
corresponds to N = 2 supersymmetry. Let us denote the sections of the normal bundle
by φ0 , φ1 . These sections are respectively 0-forms and 1-forms on the P1 . The field φ0
corresponds to the adjoined valued Higgs field Φ in the N = 2 SYM theory.
Now the inclusion of the superpotential W should give n isolated P1 ’s at the critical
values W (φ0 ) = 0. This is achieved by the following transition function. If z and z are
the coordinates on the northern and southern hemispheres of the P1 , then the resolution is
given by relating the patches (z, φ0 , φ1 ) and (z , φ0 , φ1 ) by
φ0 = φ0 , z φ1 = zφ1 + W (φ0 ), z = 1/z. (2.2)

To relate it to the geometry given by (2.1) we make the identifications
x = φ0 , u = 2φ1 , v = 2φ1 ,

ω = z φ1 , y = i 2ω − W (φ0 ) . (2.3)
Here we have introduced another variable ω in terms of which the geometry would have
been given by
uv − 4ω2 + 4ωW (x) = 0. (2.4)

This makes a natural connection with objects that will be introduced in the next section.
If we now distribute N D5-branes wrapped over these S 2 ’s and filling the spacetime,
this corresponds to a choice of the vacuum in the corresponding N = 1 supersymmetric
gauge theory, where the distribution of the branes among the n critical points corresponds
to distributing the N eigenvalues of Φ among the n classical values. At large-N gaugino
condensation takes place and this leads to a geometric transition in which the S 2 ’s are all
blown down and replaced by “blown up” S 3 ’s. This results in the geometry [10]
uv + y 2 + W 2 (x) + f (x) = 0, (2.5)

where f (x) is a polynomial of degree n − 1, whose precise coefficients depend on the
distribution of the N eigenvalues among the n critical points.
In [15] it has been shown that, if one chooses W (x) to be of degree N + 1 and if all the
critical points are equally occupied, the N = 2 Seiberg–Witten geometry can be recovered
by considering the limit W → W as → 0.
Note in particular the identification of x with the eigenvalues of Φ. This is rather
natural from Eq. (2.1). In particular if W = 0 then we have an A1 geometry formed as
a hypersurface in (u, v, y) space, where we have wrapped a D5-brane around the blowup
sphere. The transverse position of the D5-brane along the x direction would correspond to
changing the vev for Φ. Having a non-trivial fibration of the A1 geometry over the x-plane
dictated by (2.1) gives rise to the superpotential W (Φ), as reviewed above.
We will be interested in the B-model topological string on the deformed geometry given
by equation (2.5). The genus zero prepotential is determined by period integrals of the
holomorphic (3, 0) form
dx ∧ du ∧ dv
Ω= .
uv + W (x)2 + f (x)
This non-compact Calabi–Yau has n compact three-cycles Ai that correspond to the n
“blown-up” S 3 and there are n canonically conjugated cycles Bi that are non-compact
and have the topology of a three-ball. The usual special geometry relations determine the
tree-level prepotential F0 (Si ) by the periods

∂F0
Si = Ω, = Ω. (2.6)
∂Si
Ai Bi
Here one should be careful in regulating the periods over the non-compact B cycles [10].
The degree of the deformation f (x), which corresponds to normalizable deformations, is
such that these periods can be sensibly defined. In particular their variations with respect
to the variation of the coefficients of f (x) are cutoff independent.
As explained in [10] the A and B cycles can be represented as S 2 fibrations over paths
in the complex x-plane. After integrating the 3-form over these S 2 fibers we are left with
the integrals over these curves of the 1-form
η = y dx.
Here y is determined by the hyperelliptic curve
y 2 + W (x)2 + f (x) = 0. (2.7)

We will refer to this curve as the spectral curve. It has n branch cuts that are the projections
of the cycles Ai onto the x-plane. The cycles Bi are represented by half-lines that start at
the cuts and run to some cut-off point x = Λ far away from the branch cuts.
2.2. Lifting to topological string dualities
We now consider lifting these dualities to topological strings. As discussed in [9] the
key point is the observation in [22] that topological strings computes superpotential terms
of the corresponding gauge theory arising in type II superstrings. In particular the leading
planar diagram computes superpotential terms involving the gaugino bilinear field on the
gauge theory side, and the N = 2 prepotential on the dual gravity side (with some vev’s
for auxiliary fields, corresponding to turning on fluxes). More generally the topological
gravity in the presence of Calabi–Yau 3-fold is the B-model theory studied in [22] and
called “Kodaira–Spencer theory of gravity”. Thus all we need to do is to specify the gauge
theory dual, which should be the gauge theory on the topological branes wrapping the P1 ’s.
This theory is the reduction of the holomorphic Chern–Simons theory studied in [23] from
complex dimension three to complex dimension one, which we now turn to.
First suppose that we have no superpotential, i.e., W (x) = 0. In this case we simply
get the A1 geometry times the x-plane. We wrap N branes around the P1 . In this case the
normal directions to the P 1 correspond to the cotangent bundle and the trivial bundle C
associated to x. Let us call the Higgs fields in these two direction respectively Φ1 (z) for the
cotangent direction and Φ0 (z) for the x-direction. The topological theory on the B-brane
we obtain in this case is given by the action

1
S= Tr(Φ1 D A Φ0 ),
gs
P1
where Φ1 (z) is a U (N) adjoint valued (1, 0) form on P1 , Φ0 (z) is an adjoint valued scalar,
A(z) is a U (N) holomorphic (0, 1) form connection on P1 , and D A = ∂¯ + [A, −]. Here gs
denotes the string coupling constant. If we turn on the Higgs fields thereby deforming the
P1 to a non-holomorphic curve C, this action computes the integral

1
S= Ω
gs
Y
with Y a 3-chain connecting P1 and C.

If we turn on the superpotential W by deforming the geometry, the above action
changes. This has been studied in [21] by studying the Beltrami differential that deforms
the complex structure in the appropriate way, leading (including a minor generalization) to

1
SW = Tr Φ1 D A Φ0 + W (Φ0 )ω , (2.8)
gs
P1
where W (Φ0 ) corresponds to the superpotential and ω is a (1, 1) form which can be taken
to correspond to have unit volume on P1 .
A consistency check for this action is to note that the equations of motion following
from the above action agree with the fact that the classical solutions correspond to
holomorphic curves. In particular integrating out A gives
[Φ0 , Φ1 ] = 0,
so that Φ0 and Φ1 commute, i.e., we can assume they are simultaneously diagonal.
Variation with respect to the eigenvalues of Φ1 leads to
¯ 0 = 0,
∂Φ
which together with compactness of P1 implies that Φ0 is a constant. Variation with respect
to Φ0 gives
¯ 1 = W (Φ0 )ω
∂Φ
which, together with the fact that the integral of ∂Φ1 over P1 has to be zero for non-singular
Φ1 , leads to
W (Φ0 ) = 0 = Φ1 .
Thus the classical vacua indeed are localized at points W (Φ0 ) = 0, which describe the
positions of the n P1 ’s.
In fact the action (2.8) and the resulting quantum theory is rather trivial. In particular Φ1
also appears linearly and can be integrated out exactly, leading to the constraint ∂Φ ¯ 0 = 0,
which leads to the statement that Φ0 is a constant N × N matrix
Φ0 (z) = Φ = const.
Thus the full action just reduces to its last potential term, which after integration of ω over
P1 leads to the matrix action
1
SW (Φ) = Tr W (Φ). (2.9)
gs
Thus we see that the partition function of the gauge system is equivalent to a simple matrix
model!
2.3. Operator formalism
We can give another derivation for the action (2.8) starting directly from the patching
functions (2.2) that determine the blown-up geometry. We split the P1 in two hemispheres
connected by a long cylinder. We denote the fields on these two patches as Φ0 , Φ1
respectively Φ0 , Φ1 . In the case W = 0 we are simply dealing with a gauged chiral CFT
given by an adjoined valued β–γ system of spin (1, 0). The partition function computes
the number of holomorphic blocks and this is one on the two-sphere. (Here we are ignoring
for a moment the factor coming from the volume of U (N).) In an operator formalism this
partition is simply given by pairing the left and right vacuum
Z = 0|0 = 1.
Now we want to implement the deformation induced by W . From the transition function
(2.2) we see that the fields are related in the following way (here we write the fields in
coordinates on the cylinder so that factors of z and z are absorbed)
Φ1 = Φ1 + W (Φ0 ).
Now there is an obvious operator U that implements this transformation on the Hilbert
space. If we define

1
U = exp Tr W Φ0 (z) dz, (2.10)
gs
(recall that the operator Φ0 (z) is an holomorphic field, so the contour does not matter as
long as it encircles the poles) then one easily verifies that
Φ1 = U Φ1 U −1 .
Here one uses the fact that the fields Φ0 and Φ1 are canonically conjugated
gs
Φ0 (z)Φ1 (w) ∼ .
z−w
Therefore in an Hamiltonian formalism the partition function should be given by inserting
the transformation U between the left and right vacua
Z = 0|U |0.
This is the familiar way to implement changes in complex structure in the operator
formalism of conformal field theory.
In Eq. (2.10) we have written
the deformation of the action in terms of the contour
integration or Wilson line Tr W (Φ0 ). Alternatively, this can be written as a surface
integral

1
U = exp Tr W (Φ0 ) ω,
gs
P1
where the volume form ω has been localized to a band along the equator of the P1 . But, as
noted above, one can take any 2-form, as long as it integrates to 1 over the sphere.
2.4. Precise conjecture
We are now ready to state our conjecture in precise terms. Let FW,f (gs ) denote the
partition function of topological B-model for the Calabi–Yau manifold given by
uv + y 2 + W (x)2 + f (x) = 0,
where W is a fixed polynomial of degree n + 1 in x and f is a polynomial of degree n − 1
in x. Let Si denote the integral of the holomorphic 3-form over the ith S 3 coming from
the ith critical point of W . The periods Si will vary as we vary the n coefficients of f .
Inverting this map, given the variables Si we can find the coefficients of the polynomial
f (x) compatible with these periods.
On the gauge theory side we now consider the matrix model given by the action
1
SW (Φ) = Tr W (Φ).
gs
We expand this matrix model near the classical vacuum given by partitioning
N = N1 + · · · + Nn ,
and by putting Ni eigenvalues of Φ in the ith critical point of W . (Here we use both
stable and unstable critical points. In fact, since we work in the holomorphic context, this
difference does not really make sense.) Let FW,Ni (gs ) denote the free energy of this matrix
model expanded near this classical vacuum. Then the claim is
FW,f (gs ) = FW,Ni (gs )

with the understanding that Ni gs = Si and, as discussed above, the periods Si fix the
coefficients in the polynomial f (x).
A few comments are in order: the matrix model integral
is over a holomorphic matrix
Φ, i.e., we have integrals of the form dΦ and not like dΦ dΦ. This is a general issue
in topological B-branes.1 The world-volume actions, such as the holomorphic Chern–
Simons action in six dimensions, is a holomorphic function of the field variables. The non-
perturbative path-integral should therefore by defined by picking some appropriate contour
in the complex field configuration space. Indeed in our two-dimensional example we ended
up with a chiral CFT and the path-integral definition of such a theory is rather subtle.
One often ends up defining it as a holomorphic square root of a non-chiral theory. For a
perturbative computation the situation is much easier. One simply performs the Feynman
diagrams as if one was dealing with a real field. Similarly in this case we can effectively
treat the matrix Φ as if it is a real, that is Hermitean, matrix.
Secondly we should note that the matrix integral, in the particular vacuum we end up,
has a prefactor of
1
.
Vol(U (N1 ) × · · · × U (Nn ))
1 We thank E. Witten for a useful discussion on this point.

This comes from the fact that for this vacuum U (N1 ) × · · · × U (Nn ) denotes the unbroken
gauge group and we have to mod out by the corresponding volume of the constant gauge
transformations. This piece gives, as discussed in [16] the partition function of c = 1 at
self-dual radius. In particular the genus 0 answer will involve
1
F0 = S 2 log Si .
2 i
i
embedding in type IIB superstring this leads to the gaugino superpotential Weff =
In
i Ni ∂F0 /∂Si + αSi . (Note that within the type IIB context the parameters Si and Ni
are independent.) This is a first check on our conjecture. We are now ready to test the
above large-N conjecture in more detail.
3. Matrix models and special geometry
We will first show how our conjecture can be proven in the planar limit using standard
manipulations in matrix model technology. A useful reference is for example [3]. We will
see how the special geometry of Calabi–Yau’s emerges naturally.
3.1. Matrix integrals in the planar limit
We consider the one-matrix integral over N × N Hermitean matrices Φ

1 1
Z= dΦ exp − Tr W (Φ)
Vol(U (N)) gs
with W (Φ) a polynomial of degree n + 1. By diagonalizing the matrix Φ such integrals
can be reduced to an integral over the eigenvalues λ1 , . . . , λN . In this way we obtain

1
Z= dλi ∆(λ)2 exp − W (λi ) ,
gs
i i
where the Vandermonde determinants
j −1
∆(λ) = (λi − λj ) = det λi
i<j
appear from the Jacobian picked up by the diagonalization process. After exponentiating
this contribution the effective action for the eigenvalues is given by
1
S(λ) = W (λj ) − 2 log(λi − λj ). (3.1)
gs
i i<j
In this way we end up with a collection of N variables λ1 , . . . , λN in a potential W (λ).

These eigenvalues interact through the second term that is a consequence of integrating
out the off-diagonal components. This term gives the famous Coulomb repulsion between
the eigenvalues. Because of the effective Pauli exclusion principle induced by the
Vandermondes it makes the eigenvalues behave as fermions. From this action we see that
the equation of motion satisfied by the eigenvalues is

1 1
W (λi ) − 2 = 0. (3.2)
gs λi − λj
j =i
We will now take the limit N → ∞ of this system while keeping fixed the ’t Hooft
coupling
µ = gs N.
In this standard large-N limit we will have a continuum of eigenvalues and their density
1
ρ(λ) = δ(λ − λi )
N
i

becomes a continuous function on the real axis normalized to ρ(λ) dλ = 1. (In the
following λ will always denote a real variable, in contrast with the variable x that can be
complex). The eigenvalues will fill a domain on the real axis. This domain might consist of
several disconnected components known as cuts. In the case of more than one components
one speaks of a multi-cut solution. In the present case they are at most n of these cuts. We
will denote the corresponding intervals in the complex x plane as Ai , i = 1, . . . , n.
3.2. The spectral curve
To further analyze the model it will be convenient (and standard practice) to introduce
the trace of the resolvent of the matrix Φ

1 1 1 1
ω(x) = Tr = , x ∈ C, x = λi .
N Φ −x N λi − x
i
This resolvent plays an crucial role in matrix model technology. It also has an interesting
physical interpretation. For example, it can be thought of as a loop operator.
By multiplying the equation of motion (3.2) by the factor 1/(λi − x) and summing over
i one obtains the important relation (loop equation) [3]
1 1 1
ω2 (x) − ω (x) + ω(x)W (x) + f (x) = 0, (3.3)
N µ 4µ2
where the polynomial f (x) is of degree n − 1 and is given by
4µ W (x) − W (λi )
f (x) = . (3.4)
N x − λi
i
In some sense the function f (x) determines through (3.3) the whole solution of the matrix
integral. Since it is polynomial of degree n − 1 we only have to determine the n unknown
coefficients.
In the large-N limit the second term in (3.3) can be ignored and the differential equation
for ω(x) becomes an algebraic equation
1 1
ω2 (x) + ω(x)W (x) + f (x) = 0. (3.5)
µ 4µ2
From this we see that the resolvent ω(x) in general has a piece that can have branch cuts.
This singular part is captured by the function y(x) that we define here as
1
y(x) = 2µω(x) + W (x) = 2gs + W (x). (3.6)
λi − x
i
In terms of the variables (x, y) the relation (3.5) associated to the matrix model now takes
the form
y 2 − W (x)2 + f (x) = 0. (3.7)

This has an interpretation as a hyperelliptic curve in the (x, y)-plane. We immediately
recognize this curve as the curve (2.7) that determined the periods and thereby the tree-
level topological string on the deformed Calabi–Yau geometry. In fact, it is even more
natural to identify directly the parametrization of the curve in terms of (x, ω) in (3.5) with
Eq. (2.4) of Section 2. From this we see that we should identify ω with the field zΦ1 .
Note that the function y(x) naturally appears from the dynamics of the eigenvalues. It
is given by the variation of the action (3.1) with respect to a particular eigenvalue λ
∂S
y(λ) = gs . (3.8)
∂λ
In fact, from this we see that the quantity y(x) has an elegant physical interpretation. Being
the derivative of the potential energy, it equals the force on an eigenvalue if it tries to go
away from its stationary position and moves into the complex x-plane.
3.3. Filling fractions and periods
We now want to evaluate the matrix integral around a particular stationary point where
particular fractions of the eigenvalues cluster around the different critical points. Around
such a multi-cut configuration it makes sense to make a perturbative expansion of the
matrix integral using large-N techniques. It is the contribution from one of these saddle
points that we are after.
Consider such a multi-cut solution. The filling fractions Ni /N , i.e., the relative number
of eigenvalues around each critical point, are given by the integrals

Ni
= ρ(λ) dλ.
N
Ai
Here the Ai are the cuts in the complex x-plane.

Now an important standard result is that the density of eigenvalues is given by the jump
of the resolvent ω(x) across the cut
1
ρ(λ) = ω(λ + i0) − ω(λ − i0) .
2πi
Since only the singular piece contributes, we can also write this relation as
µ
ρ(λ) = y(λ + i0) − y(λ − i0) . (3.9)
πi
Therefore we can compute the fraction of eigenvalues in a specific cut by doing a contour
integral around the cut. In this way we find that the quantities Si = gs Ni are given by the
period integrals around the cut, that is the periods on the Riemann surface

1
Si = y(x) dx. (3.10)
2πi
Ai
Here we make contact with the period integrals (2.6) as obtained in the topological B-
model computation. This is the first half of the derivation of the genus zero part of our
conjecture.
In order to complete the derivation, we now need to compute the change in the free
energy F0 (Si ) if we vary the filling fractions Si by adding an eigenvalue to the cuts
1
2Si = 2Ni .
gs
This change in the action is given by the work done by the force F (x) acting on an
eigenvalue if we move this eigenvalue from one branch to infinity. We have seen in (3.8)
that this force is given by
1
F (x) = y(x).
gs

So the variation of the free energy is computed in terms of the action F (x) dx, that is by
integrating the one-form y(x) dx along one of the B-cycles to the cut-off point x = Λ. We
therefore immediately find the special geometry relation

∂F0
= y(x) dx (3.11)
∂Si
Bi
expressing the B-periods in terms of the A-periods through the free energy F0 (S). Together
(3.10) and (3.11) give the precise match with the Calabi–Yau geometry. This concludes
the derivation of the planar version of our conjecture relating the matrix model to the
topological B-model on the deformed Calabi–Yau.
As we have mentioned, for a given potential W (x) the function f (x) that deforms the
singular Calabi–Yau and determines the solution of the matrix model can be expressed in
terms of the periods Si and vice versa. In fact, using definition (3.4) we can give a useful
relation valid in terms of the prepotential F0 . If one parametrizes the potential as

n+1
W (x) = uk x k ,
k=0
and the deformation as

n−1
f (x) = 4µbk x k ,
k=0
then by plugging definition (3.4) in the matrix integral we obtain

n+1
∂F0 (ui , Si )
bk = (k + 2)uk+2 + j uj .
∂uj −k−2
j =k+2

Note that if we introduce Virasoro operators Lk = j j uj ∂/∂uj +k this equation can be
written as [24]
bk = (k + 2)uk+2 + L−k−2 F0 , k = 0, . . . , n − 1.
3.4. Higher genus
Can one extend the derivation of our conjecture to higher genera? One possibility might
be to use the loop equations that are derived by taking the expectation value of expression
(3.3) for the resolvent ω(x)
1 1
ω2 (x) + ω(x) W (x) + 2 f (x) = 0,
µ 4µ
where f (x) is now given by the expectation value of expression (3.4). These loop equations
have been studied for multi-cut solutions, for example in [25], and one can try to solve them
order by order in the string coupling constant gs . In principle these equations give recursion
relations that relate the higher genus amplitudes in terms of the tree-level free energy. It
would be very interesting to see if these equations are directly related to the equations of
motion for Kodaira–Spencer string field theory [22]. This is not completely unlikely since
collective field theories for the eigenvalue densities are known to have a similar form [26],
and morally the connection between topological strings and matrix models should go along
these lines.
It might be interesting to also briefly discuss the case of the pure conifold, here given by
the quadratic superpotential W (x) = x 2 . The large-N dual is from our point of view just
the Gaussian matrix model

1
dΦ e−1/gs Tr Φ .
2
Z=
Vol(U (N))
Since the integral is trivial, the only contribution comes from the normalization factor
1
N−1
∼ k!
Vol(U (N))
k=1
which has been shown to reproduce the all genus answer for the B-model on the conifold
in the 1/N expansion [16]. In this case the spectral curve is given by
y2 − x2 + µ = 0
and the eigenvalue density

ρ(λ) = λ2 − µ
is Wigner’s famous semi-circle distribution. In the eigenvalue basis the all genus answer is
alternatively obtained by the method of orthogonal polynomials which indeed gives [27]

N−1
− g1s λ2i
dλi ∆(λ)2 e i ∼ k! .
i k=1
3.5. Domain walls and eigenvalue tunneling
There is an interesting relation that directly connects the D-branes in the type II string
theory and the behaviour of the eigenvalues in the matrix models. In the “old days” it
was pointed out by Shenker [28] that the characteristic non-perturbative effect observed in
matrix models was the tunneling of eigenvalues and this was an effect of strength 1/gs .
This remark in some sense anticipated the importance of D-branes. Here we can connect
the two effects.
In the type IIB theory on the resolved geometry we can consider D5-branes wrapped
around an S 3 that interpolates between two S 2 . Such an object manifest itself as a domain
wall in the four uncompactified spacetime dimensions. After the geometric transition such
a D5-brane will connect two three-cycles. It will describe a process where one unit of RR
flux is transported. That is, the RR flux in one S 3 is decreased by one unit and the flux in
another S 3 is increased by one. Since the space–time superpotential is given by

∂F0
Weff = Ni + αSi , (3.12)
∂Si
i
the tension of a domain wall transferring flux from the ith to the j th cycle is given by

∂F0 ∂F0
T= − = y(x) dx.
∂Si ∂Sj
Bij
We now recognize this as the instanton action in the matrix model of an eigenvalue
tunneling from the cut Ai to the cut Aj along the path Bij .
4. Double scaling limits and other further generalizations
Given that we have found a natural stringy interpretation of ordinary matrix model, one
could ask what is the meaning of the double scaling limit in the context of the old matrix
model [3]. Following that limit on the gravity side for the single matrix model leads to the
Calabi–Yau geometry
H = uv + y 2 + x 2m+1 + deformations = 0
as corresponding to the (2, 2m + 1) bosonic minimal model coupled to two-dimensional
gravity. The deformations correspond to the m observables of the (2, 2m + 1) model. In
fact for generic deformations of this geometry there are m A-cycles and m B-cycles and
we thus can choose m independent parameters to parametrize these deformations. The
infinitesimal deformations which map this geometry to the deformations of the (2, 2m + 1)
models are of the form
m
uv + y + x (x − i )2 = 0,
2
(4.1)
i=1
where i are related to the deformation of the (2, 2m + 1) model with primary fields. For
example, the case of pure 2d gravity, i.e., the (2, 3) model has only one observable and
in that case we have = µ1/2 . In the usual matrix model this is obtained by considering
a quartic superpotential W (x). Let us take that to be an even function of x. Then there
are three critical points, including one at x = 0. Upon deformation by f (x) the critical
points can split. If we put all the N eigenvalues at the well corresponding to x = 0 then
only the x = 0 double point splits and the other two do not split. The double scaling limit
corresponds to taking the limit where one of the double points reaches one pair of doubles
and the other reaches the other pair. Taking the limit where the geometry is localized near
one set of triple zeros we obtain the (2, 3) local geometry given in (4.1). Embedding this
kind of theory in type IIB strings gives exactly the kind of N → ∞ dualities proposed
recently in [19] which relate certain limits of gauge systems with type IIB strings on
Calabi–Yau geometries without fluxes.
Note that this map to topological B-model is in perfect accord with the fact that the
bosonic strings have a hidden N = 2 superconformal symmetry [29], which in this case
we can identify with the superconformal theory on the above Calabi–Yau threefold. As a
check note that, if we only turn on the cosmological constant, then the genus g answer for
topological B-model scales as the holomorphic threeform Ω to the power of 2 − 2g [22].
For the (2, 3) model since we have
du dv dy dx
Ω= ∼ µ5/4 .
dH
So we learn that the free energy of the B-model is given by an expansion of the form
2g−2 5/4(2−2g)
F= cg gs µ
g0
for some coefficients cg , in perfect accord with the expected answer.

As a further generalization one can consider the topological lift of the models considered
in [14]. These involve many U (N) gauge fields with bifundamental matter. In particular
for the case of a linear chain of p − 1 gauge fields one obtains the dual geometry (before
blow up)

p

uv + y − Wi (x) = 0,
i=1
where the Wi (x) are related to the superpotentials of the corresponding gauge fields. Again
a double scaling limit exists which lead to geometries of the form
uv + y p + x q + deformations = 0.
This would correspond to the (p, q) bosonic minimal model coupled to gravity. There are
2 (p − 1)(q − 1) A-cycles for deformations in the above equation, which correspond to the
1
observables of the (p, q) minimal model. As discussed in [13] in this case the reduction
of holomorphic 3-form leads to p − 1 1-forms ηi which naturally get identified with the
p − 1 eigenvalue densities in a (p − 1) matrix model. It would be interesting to study these
model in more detail and in particular the corresponding multi-matrix model duals.
4.1. Final remarks
There are many indications that topological strings on a general target space might be
described by some kind of integrable systems. This was originally shown for the c 1
topological string theories [30]. More recently in the context of A-model topological
strings on non-compact Calabi–Yau this integrability was shown by dualizing to certain
observables of Chern–Simons gauge theory on three-manifolds [19,20]. For A-models on
compact target spaces evidence has been accumulating in the mathematical literature on
Gromov–Witten invariants (see, for example, [31,32]). If we manage to formulate a matrix
model dual of the topological string this integrability is in some sense manifest. Both the
finite-N and the large-N matrix models are well known to give tau-functions of the KP and
Toda hierarchies [3,27]. This integrability of matrix models was the underlying reason that
non-critical strings with c 1 were exactly solvable. Our results indicate that for a large
class of non-compact Calabi–Yau manifolds this integrability is present.
Furthermore one should also like to be able to include gravitational descendents within
the topological string model. In terms of the B-model we expect that these are given by
non-normalizable deformations of the complex structure.
A related issue is the description of the c = 1 string, which is equivalent to topological
strings on the conifold geometry [33]. In a forthcoming paper [34] we will describe how
tachyon scattering processes are indeed reproduced in the conifold string theory and can
be described by a large-N dual gauge system, making contact with [35] and the recent
work [36].
Acknowledgements
We would like to thank F. Cachazo, V. Kazakov, G. Moore and H. Ooguri for valuable
discussions. R.D. would like to thank the Harvard Physics Department and the Institute for
Advanced Study, Princeton for kind hospitality during part of this work. The research of
R.D. is partly supported by FOM and the CMPA grant of the University of Amsterdam,
C.V. is partly supported by NSF grants PHY-9802709 and DMS-0074329.
References
[1] G. ’t Hooft, A planar diagram theory for strong interactions, Nucl. Phys. B 72 (1974) 461.
[2] P. Ginsparg, G.W. Moore, Lectures on 2D gravity and 2D string theory, hep-th/9304011.
[3] P. Di Francesco, P. Ginsparg, J. Zinn-Justin, 2D gravity and random matrices, Phys. Rep. 254 (1995) 1,
hep-th/9306153.
[4] M. Kontsevich, Intersection theory on the moduli space of curves and the matrix airy function, Commun.
Math. Phys. 147 (1992) 1.
[5] E. Witten, On the structure of the topological phase of two-dimensional gravity, Nucl. Phys. B 340 (1990)
281.
[6] T. Banks, W. Fischler, S.H. Shenker, L. Susskind, M-theory as a matrix model: a conjecture, Phys. Rev. D 55
(1997) 5112, hep-th/9610043.
[7] O. Aharony, S.S. Gubser, J.M. Maldacena, H. Ooguri, Y. Oz, Large N field theories, string theory and
gravity, Phys. Rep. 323 (2000) 183, hep-th/9905111.
[8] R. Gopakumar, C. Vafa, On the gauge theory/geometry correspondence, Adv. Theor. Math. Phys. 3 (1999)
1415, hep-th/9811131.
[9] C. Vafa, Superstrings and topological strings at large N , J. Math. Phys. 42 (2001) 2798, hep-th/0008142.
[10] F. Cachazo, K.A. Intriligator, C. Vafa, A large N duality via a geometric transition, Nucl. Phys. B 603 (2001)
3, hep-th/0103067.
[11] J.D. Edelstein, K. Oh, R. Tatar, Orientifold, geometric transition and large N duality for SO/Sp gauge
theories, JHEP 0105 (2001) 009, hep-th/0104037.
[12] K. Dasgupta, K. Oh, R. Tatar, Geometric transition, large N dualities and MQCD dynamics, Nucl. Phys.
B 610 (2001) 331, hep-th/0105066;
K. Dasgupta, K. Oh, R. Tatar, Open/closed string dualities and Seiberg duality from geometric transitions in
M-theory, hep-th/0106040;
K. Dasgupta, K. Oh, R. Tatar, Geometric transition versus cascading solution, JHEP 0201 (2002) 031, hep-
th/0110050.
[13] F. Cachazo, K.A. Intriligator, C. Vafa, A large N duality via a geometric transition, Nucl. Phys. B 603 (2001)
3, hep-th/0103067.
[14] F. Cachazo, B. Fiol, K.A. Intriligator, S. Katz, C. Vafa, A geometric unification of dualities, Nucl. Phys.
B 628 (2002) 3, hep-th/0110028.
[15] F. Cachazo, C. Vafa, N = 1 and N = 2 geometry from fluxes, hep-th/0206017.
[16] H. Ooguri, C. Vafa, Worldsheet derivation of a large N duality, hep-th/0205297.
[17] M. Aganagic, C. Vafa, G2 manifolds, mirror symmetry, and geometric engineering, hep-th/0110171.
[18] D.E. Diaconescu, B. Florea, A. Grassi, Geometric transitions and open string instantons, hep-th/0205234.
[19] M. Aganagic, M. Marino, C. Vafa, All loop topological string amplitudes from Chern–Simons theory, hep-
th/0206164.
[20] D.E. Diaconescu, B. Florea, A. Grassi, Geometric transitions, del Pezzo surfaces and open string instantons,
hep-th/0206163.
[21] S. Kachru, S. Katz, A.E. Lawrence, J. McGreevy, Open string instantons and superpotentials, Phys. Rev.
D 62 (2000) 026001, hep-th/9912151.
[22] M. Bershadsky, S. Cecotti, H. Ooguri, C. Vafa, Kodaira–Spencer theory of gravity and exact results for
quantum string amplitudes, Commun. Math. Phys. 165 (1994) 311, hep-th/9309140.
[23] E. Witten, Chern–Simons gauge theory as a string theory, hep-th/9207094.
[24] M. Bertola, B. Eynard, J. Harnad, Partition functions for matrix models and isomonodromic tau functions,
nlin.SI/0204054.
[25] G. Akemann, Higher genus correlators for the Hermitian matrix model with multiple cuts, Nucl. Phys. B 482
(1996) 403, hep-th/9606004.
[26] S.R. Das, A. Jevicki, String field theory and physical interpretation of D = 1 strings, Mod. Phys. Lett. A 5
(1990) 1639.
[27] A. Morozov, Matrix models as integrable systems, hep-th/9502091.
[28] S.H. Shenker, The strength of nonperturbative effects in string theory, in: Proc. Random Surfaces and
Quantum Gravity, Cargese, 1990, pp. 191–200.
[29] M. Bershadsky, W. Lerche, D. Nemeschansky, N.P. Warner, Extended N = 2 superconformal structure of
gravity and W gravity coupled to matter, Nucl. Phys. B 401 (1993) 304, hep-th/9211040.
[30] R. Dijkgraaf, Intersection theory, integrable hierarchies and topological field theory, in: Cargese Summer
School on New Symmetry Principles in Quantum Field Theory, 1991, hep-th/9201003.
[31] A.B. Givental, Gromov–Witten invariants and quantization of quadratic Hamiltonians, math.AG/0108100.
[32] A. Okounkov, R. Pandharipande, Gromov–Witten theory, Hurwitz theory, and completed cycle,
math.AG/0204305.
[33] D. Ghoshal, C. Vafa, c = 1 string as the topological theory of the conifold, Nucl. Phys. B 453 (1995) 121,
hep-th/9506122.
[34] R. Dijkgraaf, C. Vafa, in preparation.

[35] P.B. Wiegmann, A. Zabrodin, Conformal maps and integrable hierarchies, hep-th/9909147.
[36] S.Y. Alexandrov, V.A. Kazakov, I.K. Kostov, Time-dependent backgrounds of 2D string theory, hep-
th/0205079.
On geometry and matrix models

Robbert Dijkgraaf a , Cumrun Vafa b
a Institute for Theoretical Physics & Korteweg-de Vries Institute for Mathematics, University of Amsterdam,
1018 TV Amsterdam, The Netherlands
b Jefferson Physical Laboratory, Harvard University, Cambridge, MA 02138, USA
Received 15 August 2002; accepted 23 August 2002
Abstract
We point out two extensions of the relation between matrix models, topological strings and N = 1
supersymmetric gauge theories. First, we note that by considering double scaling limits of unitary
matrix models one can obtain large-N duals of the local Calabi–Yau geometries that engineer N = 2
gauge theories. In particular, a double scaling limit of the Gross–Witten one-plaquette lattice model
gives the SU(2) Seiberg–Witten solution, including its induced gravitational corrections. Secondly,
we point out that the effective superpotential terms for N = 1 ADE quiver gauge theories is similarly
computed by large-N multi-matrix models, that have been considered in the context of ADE minimal
models on random surfaces. The associated spectral curves are multiple branched covers obtained as
Virasoro and W -constraints of the partition function.
 2002 Published by Elsevier Science B.V.
1. Introduction
In [1] we have shown how the effective superpotential in N = 1 supersymmetric

gauge theories, that are obtained by breaking an N = 2 super-Yang–Mills theory by
adding a tree-level superpotential W (Φ) for the adjoint scalar Φ, can be computed by
a large-N Hermitian matrix models. More precisely, the effective superpotential of the
N = 1 theory considered as a function of the gluino condensates Si is—apart from the
universal Si log(Si /Λ) terms coming from the pure N = 1 Yang–Mills theory—given
exactly by a perturbative series that is computed by the planar diagrams of the matrix model
with potential W (Φ). Furthermore, the contributions of this gauge theory to the induced
E-mail address: rhd@science.uva.nl (R. Dijkgraaf).
0550-3213/02/$ – see front matter  2002 Published by Elsevier Science B.V.

PII: S 0 5 5 0 - 3 2 1 3 ( 0 2 ) 0 0 7 6 4 - 2
supergravity corrections R 2 F 2g−2 (with R the Riemann curvature and F the graviphoton
field strength) are similarly computed exactly by the genus g > 0 matrix diagrams.
This gauge theory/matrix model correspondence was a consequence of the large-N
dualities of [2–4] that relate the computation of holomorphic F-terms in the world-volume
theories of D-branes to partition functions of topological strings in local Calabi–Yau
geometries—a relation that was further explored in [5–11]. In the simplest case these local
non-compact Calabi–Yau manifolds take the form
vv + y 2 − W (x)2 + f (x) = 0.
One finds that in the B-model topological string the tree-level free energy can be computed
in terms of the periods of the meromorphic differential y dx on the associated Riemann
surface
y 2 − W (x)2 + f (x) = 0.
As we argued in [1] this curve and the associated special geometry arises naturally from
the large-N dynamics of the matrix integral with action W (Φ). But we should stress again
that the relation with matrix models goes beyond the planar limit. The higher genus string
partition functions Fg and the related gravitational couplings of the gauge theories are
exactly computed in the 1/N expansion of the matrix models.
Let us briefly summarize these connections, for more details see [1]. We start with the
Hermitian matrix integral

dΦ e−S(Φ)
with action
1
S(Φ) = Tr W (Φ),
gs
and W (x) is a polynomial of degree n + 1. The matrix integral can be reduced to an integral
over the eigenvalues x1 , . . . , xN of Φ in the potential W (x). In the classical limit gs → 0,
where one ignores the interactions among the eigenvalues, the equation of motion is given
by
∂S
y(x) = gs = W (x) = 0. (1.1)
∂x
The associated classical spectral curve is
y 2 − W (x)2 = 0, (1.2)

where x, y can be considered as complex variables. Writing W (x) =
i (x − ai ) we see
that this singular genus zero planar curve has n double points at the critical points x = ai .
Sometimes it can be helpful to think of the (x, y)-plane as a phase space, with y the
momentum conjugate to x, as given by the Hamilton–Jacobi equation (1.1). Then (1.2) has
an interpretation as the zero-energy level set of the (bosonic part of the) supersymmetric
quantum mechanics Hamiltonian associated to the superpotential W (x), and S(x) can be
thought of as the semi-classical WKB action of the associated quantum mechanical ground
state Ψ (x) ∼ e−S(x).
Classically, the N eigenvalues will cluster in groups of Ni in the critical points ai where
they will form some meta-stable state. The relative number of eigenvalues or filling fraction
of the critical point ai we will denote as
νi = Ni /N.
If gs is not zero, we have to take into account the Coulomb interaction that results from
integrating out the angular, off-diagonal components of the matrix Φ. The equation of
motion of a single eigenvalue x in the presence of the Dyson gas of eigenvalues x1 , . . . , xN
is now modified to

N
1
y = W (x) − 2gs .
x − xI
I =1
We will now take the large-N ’t Hooft limit keeping both µ = gs N and the filling
fractions νi fixed. In this case each critical point has its own ’t Hooft coupling
µi = gs Ni = µνi .
The collective dynamics of these eigenvalues in the large-N limit can be summarized
geometrically as follows. Each of the n double points x = ai gets resolved into two
branch points ai+ , ai− . The resulting branch cuts Ai = [ai− , ai+ ] are filled by a continuous
density of eigenvalues that behave as fermions and spread out due to the Pauli exclusion
principle. This process of splitting up of double points is very analogous to transition
from the classical to the quantum moduli space in the Seiberg–Witten solution of N = 2
supersymmetric gauge theories [12]—a relation that was explained in [11]. The resolution
of double points is captured by deforming the classical spectral curve (1.2) into the
quantum curve
y 2 − W (x)2 + µf (x) = 0, (1.3)

where the quantum deformation f (x) is a polynomial of degree n − 1. The filling fraction
νi is related to the size of the branch cut Ai . Roughly, the higher the proportion of
eigenvalue at the critical point ai , the larger the cut. More precisely, we have

1
µi = µνi = y(x) dx.
2πi
Ai
(This equation can be considered as analogous to the Bohr–Sommerfeld quantization

condition.) Finally the tree-level free energy F (µi ) can be computed in terms of the dual
B-periods by the special geometry relations

∂F
= y(x) dx,
∂µi
Bi
where the cycles Bi run from the branch cuts to some cut-off point at infinity.
2. Unitary matrix models and the Seiberg–Witten solution
We have reviewed how the effective superpotential in N = 1 supersymmetric gauge

theories obtained by breaking an N = 2 super-Yang–Mills theory can be computed by a
large-N Hermitian matrix model. This raises the question whether there exists a matrix
model that computes directly holomorphic F-terms in the underlying undeformed N = 2
theory as described by the rigid special geometry of the Seiberg–Witten solution [12].
As explained in [11] it is indeed possible to extract the geometry of the N = 2 solution
from the effective superpotential of the N = 1 deformation of N = 2 theory and thereby,
indirectly, from the associated matrix model. But here we will pursue a more direct
relation—we will show that indeed there is a large-N matrix model that computes the SW
solution directly. The matrix model for N = 2 supersymmetric SU(2) gauge theory turns
out to be a double scaling limit of the most simple unitary matrix model—the so-called
one-plaquette or Gross–Witten model [13].
2.1. Geometrical engineering of N = 2 theories
Our strategy will be the following. The N = 2 theory can be geometrically engineered
by taking a suitable limit of type IIB string theory on a local Calabi–Yau [14–17]. This
local CY produces directly the Seiberg–Witten curve that encodes the dynamics of the
N = 2 gauge theory. More precisely the genus zero topological B-model amplitudes on the
local CY capture the Seiberg–Witten geometry. The higher genus amplitudes compute the
contributions of the gauge theory to certain gravitational terms of the form R 2 F 2g−2 [18,
19]. We will now engineer a matrix model that is large-N dual to this local CY geometry.
In particular the planar limit gives the SW geometry and the 1/N corrections capture the
generation of the corresponding gravitational terms.
To be specific, let us discuss here the simplest case of pure SU(2) N = 2 Yang–Mills
theory. The SW solution is given in terms of the familiar elliptic curve
2
w2 = y 2 + u − Λ4 , (2.1)
where u is the coordinate on the moduli space, i.e., the vev of the adjoint 12 tr Φ 2
, and Λ
the gauge theory scale, that we will sometimes set conveniently to Λ = 1.
The local CY obtained in the geometric engineering is given by the algebraic variety
[15–17]

1
vv + Λ z +
2
+ 2 y2 + u = 0
z
with z ∈ C∗ , i.e., an invertible variable. After reducing over the (v, v )-plane the associated
Riemann surface is

1
Λ z+
2
+ 2 y 2 + u = 0. (2.2)
z
Since z = 0 we can multiply by z and substituting w = Λ2 z + y 2 + u to bring the curve in
the form (2.1).
Furthermore, the reduction of the holomorphic three-form on the CY gives directly the
SW differential y dz/z. The prepotential F (u) is then obtained by computing the periods
of this meromorphic one-form along the A-cycle √and B-cycle of the elliptic curve (2.1).
Note that there are four branch points at y = ± −u √ ± Λ . In the classical limit Λ → 0
2
they coalesce pairwise in two double points y = ± −u. The moduli space contains two
singularities at u = ±1 where monopoles respectively dyons become massless.
Since z is a C∗ variable, it makes sense to write z = eix with the variable x periodic
modulo 2π , and reexpress the original local CY and the resulting curve (2.2) as
Λ2 cos x + y 2 + u = 0,
with SW differential y dx. This way of writing the equation suggests a relation to a matrix
model with some suitable potential W (x). In fact, since x is now a periodic variable this
suggest a unitary matrix model where z = eix will get interpreted as the eigenvalue of a
unitary matrix U .
2.2. Unitary matrix models
Unitary matrix models are defined as integrals over the group manifold U (N) of the
form

1 1
Z= dU exp − Tr W (U ) , (2.3)
Vol(U (N)) gs
U (N)
where dU is the Haar measure. As in the Hermitian matrix models, one can diagonalize U
and express everything in integrals over its eigenvalues

U ∼ diag eiα1 , . . . , eiαN ,
where the αi are periodic variables, giving

2 αI − αJ 1
Z= dαI sin exp − W (αI ) .
2 gs
I I <J I
Note that such a unitary matrix model can be viewed as a special case of a Hermitian
model by writing U = eiΦ with Φ a ‘compactified’ Hermitian matrix, i.e., a matrix with a
periodic spectrum Φ ∼ Φ +2π . Such a periodicity is achieved by adding multiples of 2π to
the eigenvalues αI of Φ. This addition of multiple images is the familiar way to compactify
transverse directions for D-branes or for matrix models in M-theory [20]. For example, in
this way the Vandermonde determinants in the measure become after regularization

αI − αJ
(αI − αJ + 2πn) = sin .
2
n∈Z
The unitary matrix model describes a collection of particles in a potential W (α) on the
unit circle interacting through a Coulomb potential. The equation of motion of the unitary
model is
αI − αJ
W (αI ) − 2gs cot = 0.
2
J
The large-N solution proceeds exactly as in the uncompactified case. One introduces again
a resolvent, this time defined as

1 x − αI
ω(x) = − cot ,
N 2
I
that satisfies a quadratic loop equation that can be derived in exactly the same way as in
the Hermitian case. In the limit N → ∞ with gs N = µ fixed this loop equation takes the
following familiar form, when written in terms of the variable y = W (x) + 2µ ω(x),
y 2 − W (x)2 + 4µf (x) = 0. (2.4)

The quantum correction f (x) in this unitary case is given by the expression

1 x − αI
f (x) = W (x) − W (αI ) cot . (2.5)
N 2
I
2.3. The Gross–Witten model
Our candidate unitary matrix model will be a much studied one, namely the so-
called Gross–Witten model [13] with potential W (α) ∼ cos α. This model was originally
introduced as a lattice discretization of two-dimensional (non-supersymmetric) Yang–
Mills theory. In such a lattice model to each plaquette with holonomy U around the edge
one associates the Wilson action (in other contexts known as the Toda potential)
, ,
S(U ) = Tr U + U −1 = Tr cos(Φ).
2gs gs
Here Φ can be thought of as the lattice approximation to the gauge field strength Fµν , and
in the limit Φ → 0 this gives the quadratic Yang–Mills action Tr Φ 2 . (The parameter ,
we introduce for convenience. It can of course be absorbed by rescaling gs = gYM 2 .) In a
general lattice model one integrates over a collection of plaquettes, but in two dimensions
a single plaquette suffices to compute for instance a Wilson loop action.
The GW model has two critical points
W (α) = −, sin(α) = 0
at a1 = 0 and a2 = π . Note that for real and positive , (the case relevant for Yang–Mills
theory) the second point is an unstable critical point. But that issue is irrelevant for the
holomorphic matrix models that we are considering here. The parameter , can be complex,
and our eigenvalues eiα are allowed to move off the unit circle into the punctured complex
plane C∗ . In fact, following our general philosophy, we will consider the perturbative
expansion of the matrix integral (2.3) around the saddle point where N1 eigenvalues are at
the first critical point a1 and N2 = N − N1 are at the second point a2 . So we are dealing
with a two-cut, meta-stable solution to the matrix integral. These two cuts introduce a
second parameter, besides the overall ’t Hooft coupling µ = gs N , namely the relative
filling fraction
ν = (N1 − N2 )/N.
If we introduce the separate ’t Hooft couplings for the two critical points
µ1 = gs N1 , µ2 = gs N2 ,
then the difference of these couplings is related to the filling fraction
µ = µ1 − µ2 = gs (N1 − N2 ) = µ · ν.
We will be interested in computing the planar limit of the free energy F as a function of
the two coupling µ1 and µ2 , or equivalently as a function of the ’t Hooft coupling µ and
the filling fraction ν. Although ν takes values in the interval [−1, 1] the final result will
turn out to be a holomorphic function of ν.
The planar limit can be computed solving the loop equation (2.4). To this end we have
to compute the quantum correction f (x) defined in (2.5). For our choice of potential
W (x) = , cos x this becomes the following average over the eigenvalues

1 x − αI
f (x) = − ,(sin x − sin αI ) cot
N 2
I
1
=− ,(cos x + cos αI )
N
I
= −,(cos x + u). (2.6)
Here the constant u is defined as the average
1
u= cos αI .
N
I
In the semi-classical approximation gs → 0 we have
N1 − N2
u≈ = ν.
N
Since there are N1 eigenvalues at the critical point αI = 0 and N2 eigenvalues at αI = π ,
that contribute respectively +1 and −1 to the average of cos αI .
Inserting our expression for f (x) into (2.4) gives the spectral curve
y 2 − , 2 sin2 x + 4µ,(cos x + u) = 0. (2.7)

We see that the original double points a1 = 0 and a2 = π now split up in four branch points
a1± , a2± . The two branch cuts

A1 = a1− , a1+ , A2 = a2− , a2+
describe the condensation of eigenvalues around the two critical points. The discontinuities
in y give the eigenvalue density ρ(α). Integrating the one-form y dx around the branch cuts
Ai gives the filling fractions Ni /N . These relations allow one to express the filling fraction,
or more precisely the relative coupling µ = µν, in terms of the period

1
µ = y(x) dx
2πi
A
where A is a one-cycle homologous to A1 − A2 . This equation describes the exact relation

between the variables µ and u. Integrating the same differential over the conjugated
(1) (2)
B-cycle encircling the cut [α+ , α− ] computes the variation of the free energy F under a
change of the relative number of eigenvalues in the two cuts,

∂F
= y(x) dx.
∂µ
B
Note that the GW solution assumes that all eigenvalues center around the stable vacuum
a1 = 0, in which case N2 = 0 and ν = 1. In fact we claim that for this vacuum we have
exactly u = 1, so that the spectral curve is given by
y 2 − , 2 sin2 x + 4µ,(cos x + 1) = 0.
Indeed, since one only fills the stable critical point a1 = 0 with eigenvalues, only this
double point will get resolved into two branch points. The second critical point a2 = π
remains unresolved. Inserting u = 1 gives the GW solution for the large-N eigenvalue
density that can be written as in [13]

α µ 2 α
ρ(α) ∼ cos − sin .
2 2, 2
2.4. Double scaling limit
Now we will take a double scaling limit of the GW model to obtain the SW solution
relevant for N = 2 supersymmetric gauge theory. In this limit we will send N → ∞ and at
the same time , → 0 and ν → 0, keeping ,Ni , ν/, and gs fixed—or, equivalently, we will
send the ’t Hooft coupling µ → ∞ keeping the difference of the two couplings µ1 and µ2
µ = gs (N1 − N2 ) = µν
fixed. In this limit the absolute difference in eigenvalues N1 − N2 remains finite, but N1
and N2 become both infinite, and therefore the relative filling fraction ν = (N1 − N2 )/N
goes to zero.
After rescaling y appropriately, the spectral curve reduces in this limit exactly to the
SW curve (with Λ = 1)
y 2 + cos x + u = 0.
Note that the double scaled curve depends only on a single parameter u that at weak
coupling could be identified with the filling fraction ν = (N1 − N2 )/N of the matrix model.
Of course the limit we are taking is at a strong coupling point and the relation between u
and matrix model modulus µ is more complicated, as discussed above.
As we have already mentioned, the prepotential is now computed by the periods of the
differential y dx = y dz/z along the A and B cycles. This allows us to identify the SW
periods as

a = y(x) dx = µ ,
A

∂F
aD = y(x) dx = . (2.8)
∂µ
B
From the original four branch points a1± , a2± obtained in resolving the two double
points a1 , a2 , our double scaling limit takes the branch points a1− , a2+ to infinity while
keeping a1+ , a2− at finite distance. This leaves two homology one-cycles: the B-cycle that
runs around the cut [a1+ , a2− ] and the dual A-cycle that is homologous to A1 − A2 . This
behaviour of the branch points is exactly the behaviour in the double scaling limit one takes
in the old matrix models [21,22], as we also noted in [1]. For example, the (2, 3) critical
point of the one-matrix model was obtained starting from a curve of the form y 2 = x 6 + · · ·
and the double scaling limit got rid of the all monomials with power more than x 3 giving
an equation of the form y 2 = x 3 + · · ·. (Note that these branch points in the x-plane should
not be confused with the branch points in the y-plane that are relevant for the SW solution.)
The double scaling is very analogous to the limit that was used in the A-model
topological string in [23]. In fact, when the scaling limit is embedded in type II string
theory, the resulting CY geometry based on (2.7) will have RR flux through the compact
cycles A1 and A2 . It is crucial that the remaining compact cycle B does not carry any
Ramond flux. We are thus engineering a large-N dual of a geometry without fluxes.
The GW model has a famous third order phase transition at (in our convention) µ/, = 2.
This signals a transition of the eigenvalue distribution in which the single cut changes
topology and starts to cover the whole unit circle. After the GW phase transition the
eigenvalue distribution is given by ρ(α) ∼ cos α + µ/2. Geometrically speaking, in that
phase all four branch points are on top of each other.
This phase transition is however not relevant in our model. First of all, we are studying
a more general question by considering a stationary phase approximation around a meta-
stable state with two clusters of eigenvalues and consequently have to work with a two-
dimensional phase diagram (µ, ν). As we argued the GW solution puts ν = 1 and that is
very far away from our double scaling limit in which ν tends to zero. Indeed in our limit
the number of eigenvalues in the two cuts is roughly equal. Secondly, we are dealing with
an holomorphic object, and holomorphy excludes any phase transitions, one can just go
around the singularity. The GW phase transition is just a (very special) real slice of our
complex phase diagram.
Finally it would be interesting to connect this approach to the beautiful semi-classical
computation of the SW solution and its gravitational counterparts in [24]. That computation
was inspired by matrix integrals appearing in D-branes formulas.
2.5. Generalizations for general groups
It is not difficult to guess how the SW solution for gauge group SU(n) can be
engineered. In this case the curve associated to the local CY is of the form
cos x + Pn (y) = 0,
with Pn (y) a polynomial of degree n. More generally we can consider a chain of U (ni )
gauge theories with bifundamental matter, for which the corresponding curve has been
obtained from the M5-brane viewpoint in [25] and from the viewpoint of geometric
engineering in [17]. This will give rise to a curve of the form

F eix , y = 0,
where F is a polynomial in e±ix and y. In particular if we consider the rank of all the
gauge groups to be equal to n, then F is a polynomial in y of degree n. Moreover the
difference in power of eix between the highest and lowest powers is the number of U (n)
gauge groups plus 1. As we will discuss in greater detail in the next section in the context
of Hermitian multi-matrix models such curves are typically produced by a multi-matrix
model consisting of n − 1 matrices. The choice of the coefficients in F will be related to
the choice of the action and some suitable double scaling limit, as we studied in the context
of SU(2) gauge theory here.
2.6. Connections with A-model topological strings
It is natural to ask if there is a connection with A-model topological strings, and in

particular for A-models on local toric CY. As was demonstrated in [26,27] the B-model
mirrors are given by

vv + F eix , eiy = 0
for some F . This is analogous to an infinite matrix model version of the unitary matrix
models, as follows from our discussion above. In this case the A-model has a gauge theory
dual involving certain correlations functions of Chern–Simons theory [23,28]. It would be
interesting to connect these matrix models directly with the Chern–Simons gauge theory
computation, thus completing the circle of ideas. There are some hints that this idea indeed
works [29].
3. Quiver matrix models
We will now turn to a related generalization of [1] where we will connect superpotential
computations of quiver gauge theories to multi-matrix models.
3.1. N = 1 quiver gauge theories and topological strings
We will restrict our discussion here to the ADE quivers, in particular the Ar case,
although one can also include the affine quivers based on the extended Dynkin diagrams
D
A E.
Let r denote the rank of the quiver G and consider a partition
N = N1 + · · · + Nr .
In the associated N = 2 quiver gauge theory we assign to each of the r vertices vi of
the Dynkin diagram of G a U (Ni ) gauge field and to links connecting vertices vi and
vj we associate bifundamentals Qij transforming in the representation (Ni , N j ) with a
Hermiticity condition Q†ij = Qj i .
For such a gauge theory we can write a general tree-level superpotential

W (Φ, Q) = sij Tr Qij Φj Qj i + Tr Wi (Φi ), (3.1)
i,j i
with sij = −sj i = 1 (for some ordering i < j ), if the vertices vi and vj are linked in the
Dynkin diagram (we will write this relation also as i, j
), and sij = 0 otherwise. Here the
first term is the standard superpotential of the N = 2 theory with bifundamental matter. The
additional potentials Wi (Φi ) are introduced to break the supersymmetry down to N = 1.
Within type II string theory these quiver gauge theories are obtained by wrapping
D5-branes over a particular CY geometry that is a fibration of the corresponding ADE
singularity over the complex plane [7]. This geometry contains r intersecting P1 ’s.
According to [7,8] in the large-N limit the geometry undergoes a transition to a deformed
geometry where these P1 ’s are blown down and a number of S 3 ’s with RR flux are “blown
up”. The corresponding smooth CY geometry gives a dual description of the gauge theory
system.
In the context of B-model topological strings, the deformed CY geometry is dual
to a two-dimensional large-N gauge system, obtained from a collection of B-branes
wrapped on the intersecting P1 ’s. The world-volume theory of these branes consists of
open topological strings. So each P1 gives rise to a two-dimensional field theory with
Lagrangian [30]

1
S(Φ) = A Φ 0 + Wi Φ 0 ω ,
Tr Φi1 D (3.2)
i i
gs
P1
where ω is some volume form on P1 , and Φi0 and Φi1 are adjoint fields of respectively
spin 0 and spin 1 coupled to an U (Ni ) holomorphic gauge field. Here we included the
effect of the superpotential Wi (Φ0 ). The open topological strings connecting different P1 ’s
give as physical fields the bifundamentals Qij . Since the different P1 ’s intersect in points,
the action of these bifundamental scalar fields localizes to the intersection point x and is
given by

S(Q) = Tr Qij (x)Φj0 (x)Qj i (x) − Qj i (x)Φi0 (x)Qij (x) .
i,j
x∈P1 ∩P1
i j
(Compare the similar computation for the coupling of open topological strings connecting
Lagrangians A-branes intersection along one-dimensional curves in [31].)
As in [1] one can see that in the end this two-dimensional topological field theory can
be completely reduced to the zero modes of the fields Φi0 (x) = Φi and Qij . Thereby the
path-integral reduces to the “quiver matrix integral”

1
Z= dΦi dQij exp − Tr W (Φ, Q) , (3.3)
gs
i i,j
where Φi (with i = 1, . . . , r) is an Ni × Ni Hermitian matrix and for every linked indices

i, j
the variable Qij is a Ni × Nj rectangular matrix, satisfying Q†ij = Qj i .
The generalization of the conjecture in [1] will now identify the free energy of the
large-N quiver matrix model, for given filling fractions of the saddle points, with the closed
topological string partition function in the corresponding deformed CY geometry, and this
in turn with the effective superpotential of the quiver gauge theory.
3.2. Saddle points and dual CY geometry
The saddle points of the quiver superpotential have been discussed extensively in [7–9]
following the mathematical literature. The eigenvalues x of the adjoints Φi have to satisfy
a series of equations: one for every positive roots αk of G. If that root is expressed in the
simple roots ei as

αk = nik ei ,
i
then the associated condition reads

nik Wi (x) = 0. (3.4)
i
The saddle points can be labeled as xa,k with a = 1, . . . , d and d the maximal degree
occurring in (3.4). If such a critical point appears with multiplicity Na,k then the total
number of eigenvalues of the matrix Φi in this saddle point is Na,k · nik . A general saddle
point is therefore parametrized by the filling fractions νa,k = Na,k /N . We will consider the
matrix integral in the limit where both the Na,k and N tend to infinity keeping the filling
fractions and the ’t Hooft coupling finite.
In the case of an Ar quiver there is a more straightforward description of the saddle
points [7–9]. Introduce the r + 1 potentials

i
t0 (x) = 0, ti (x) = Wj (x), i = 1, . . . , r.
j =1
The the associated rational planar curves
y − ti (x) = 0
intersect in various double points given by

j
tj (x) − ti (x) = Wk (x) = 0.
k=i+1
The saddle points of the Ar quiver matrix potential correspond exactly to these double
points.
In this Ar case the original singular CY geometry is given by

r

uv + y − ti (x) = 0
i=0
which after reduction over u, v gives precisely this collection of nodal curves. After the
deformation the corresponding smooth Riemann surface is given by

r

y − ti (x) + f (x, y) = 0 (3.5)
i=0
for a suitable normalizable quantum deformation f (x, y). Again every double point gets
resolved into two branch points. The resulting quantum curve is now an r + 1 fold cover
of the x-plane. By moving around in the x-plane these sheets will be exchanged through
Weyl reflections acting on the parameters ti .
The analogues of the meromorphic one-form are constructed by reducing the holomor-
phic three-form of the local CY over various cycles and for Ar have the following descrip-
tion [7,8]. Write the curve (3.5) in the factorized form

r

y − ai (x) = 0.
i=0
Then we have a basis of r − 1 meromorphic one-forms η1 , . . . , ηr−1 that is in one-to-one

correspondence with a basis of positive roots for Ar−1 given by

ηi = ai+1 (x) − ai (x) dx. (3.6)
In the undeformed case, with ai (x) = ti (x), this gives ηi = Wi (x) dx = dWi . Different
choices of positive roots correspond to Weyl reflections, which are in fact the generalized
Seiberg dualities of gauge theories (which should also have some direct interpretation as
dualities of the matrix models). Note that in the case of A1 we have
(y − a)(y + a) = 0, a 2 = W (x)2 + f (x)

and this gives the usual definition η = ±y dx.
We will argue that this process of smoothing out the singular curve including the
meromorphic differentials is exactly described by the large-N dynamics of the quiver
matrix integral.
3.3. ADE matrix models
Quite remarkably it turns out that the quiver matrix integrals (3.3) (up to some minor
details) have already been studied in the context of the “old matrix models”. They have
been used to describe the coupling of ADE conformal minimal models to two-dimensional
gravity by Kostov [32], see also the reviews [33,34], and they have naturally emerged in
the study of matrix models and integrable systems in the work of the ITEP group [35]. We
will follow closely these works in presenting the main results, leaving the details to the
literature.
First of all, one can immediately integrate out the bifundamental fields Qij in the quiver
matrix integral to give an effective interaction between the adjoint fields Φi and Φj
det(Φi ⊗ 1 − 1 ⊗ Φj )−1 .
After expressing everything in terms of the eigenvalues of the remaining Hermitian

matrices Φi
Φi ∼ diag(λi,1 , . . . , λi,Ni )
the quiver matrix integral reduces to

i,I <J (λi,I − λi,J )
2
1
Z= dλi,I |sij |
exp − Wi (λi,I ) .
i,I i<j,I,J (λi,I − λj,J ) gs
i,I
Note that if we introduce the Cartan matrix of G
Cij = 2δij − |sij | = ei · ej ,
then the quiver eigenvalue measure, that generalizes the usual “fermionic” Vandermonde
determinants of the one matrix model

∆(λ)2 = (λI − λJ )2 ,
I <J
can now be written as

(λi,I − λj,J )ei ·ej /2 .
(j,J ) =(i,I )
The resemblance to a correlation function of vertex operators is not accidental—it was in

fact the main motivation to study these kind of matrix models in [35] since it allows one to
express the partition function as a particular state in a two-dimensional chiral CFT, to be
more precise the level one realization of the corresponding ADE current algebra.
In the large-N limit this many-flavor Dyson gas of eigenvalues will spread out in cuts
around the saddle points and will form a continuum of eigenvalues described by a series of
densities functions
1
ρi (x) = δ(x − λi,I ).
N
I
The solution of the model proceeds again through the resolvents or loop operators of the
matrices Φi

1 1 1 1
ωi (x) = Tr = .
N x − Φi N x − λi,I
I
The jump of ωi (x) across a branch cut measure the eigenvalue density ρi (x).
In fact, it is natural to work with a closely related object—the derivative of the matrix
model action S with respect to an eigenvalue of type i evaluated at a general position x in
the complex plane (away from the cuts) as it interacts with all the other eigenvalues,
yi (x) = gs ∂i S = Wi (x) − 2µC ij ωj (x). (3.7)
As we already mentioned in [1] for a multi-matrix model we want to identify the one-forms
ηi (3.6) coming from the local CY geometry with the expressions yi (x) dx, up to a possible
change of basis. The most powerful and general technique that can be used to relate the
CY geometry to the matrix model are the loop equations.
3.4. Loop equations and collective fields
Before we discuss the loop equations of the multi-matrix models, let us first rewrite the
solution of the one-matrix model as used in [1] in a more suggestive form, that is actually a
standard technique in matrix model technology. Here we found among others the reviews
[33,34,36] very helpful.
The resolvent ω(x) has a natural interpretation as a loop operator. More precisely, the
inverse Laplace transform

dx ix:
e ω(x) = Tr e:Φ
2π
is the zero-dimensional analogue of the Wilson loop. The non-linear all-genus loop
equation is usually written in terms of ω(x) as

dz W (z)
ω(z) = µ ω(x)2 , (3.8)
2πi x − z
C
where · · ·
indicates an expectation value within the matrix integral. The contour C
encircles all the cuts but not the point x. This equation is supplemented with the boundary
condition ω(x)
∼ 1/x at infinity. The loop equation acts as a Schwinger–Dyson equation
of the matrix model. It gives a recursive relation to solve for the loop operator and the free
energy. In the planar limit we have large-N factorization ω(x)2
= ω(x)
2 and the loop
equation becomes algebraic.
Loop operators are closely connected to collective fields. By integrating out the angular
variables the individual eigenvalues start to behave as fermions, and the collective field
is essentially constructed by bosonization of these fermion fields. In [1] we have already
speculated that this collective field should be identified with the Kodaira–Spencer field [18]
describing the closed strings moving on the local CY geometry.
For a single matrix model the collective field is defined as the chiral two-dimensional
scalar field

ϕ(x) = W (x) − 2gs log(x − λI ).
I
It clearly satisfies ∂ϕ(x) = y(x) with

1
y = W (x) − 2gs = W (x) − 2µω(x).
x − λI
I
So in view of (3.7) we can identify the function ϕ(x) with the action S(x) of a single
eigenvalue as a function of its position x in the complex plane in the presence of the gas
of other eigenvalues λ1 , . . . , λN . The function ϕ(x) is multi-valued in the x-plane. It has
branch cuts around which it changes sign. It is therefore only properly defined on the
double cover
y 2 − W (x)2 + f (x) = 0. (3.9)

On this Riemann surface ϕ(x) has quantized periods around the A-cycles, given by the
filling numbers µi = gs Ni . Since it is a chiral field the periods around the dual B-cycles
are not independent and expressed by the special geometry relations as ∂F /∂µi .
Note that if we work with a general, not necessarily polynomial, superpotential

W (x) = tn x n ,
n0
then the expectation value of the field ∂ϕ(x) inserted in the matrix integral can be
represented by a linear differential operator in the couplings tn acting on the partition
function. For example,

−n−1 ∂
∂ϕ(x) = ntn x n−1
− 2gs2
x Z,
∂tn
n>0 n0
and similarly for multi-point functions. (Here we used that the derivative ∂/∂tn brings
down a factor g1s Tr Φ n .)
With this notation there is an elegant way to write the loop equations. Introduce the
holomorphic stress-tensor

T (x) = (∂ϕ)2 = Ln x −n−2 .
n
Then the all-genus loop equation (3.8) of the one-matrix model can be rewritten in the
suggestive form

dz 1
T (z) = 0. (3.10)
2πi x − z
C
That is, the expectation value T (x)
has no singular terms if x → 0. Therefore we can
also express (3.10) equivalently as the Virasoro constraints [37,38]
Ln Z = 0, n −1.
The derivation of the constraints in the matrix model is completely standard—it
simply expresses the Ward identities following from the invariance under infinitesimal
reparametrization of Φ → Φ + ,Φ n+1 of the matrix variable Φ.
In the planar limit we can substitute the classical values for ∂ϕ(x) = y in T (x) and then
Eq. (3.10) is a consequence of (3.9) that can now be written as
T (x) = W (x)2 − f (x),
which shows that T (x) is indeed regular (even polynomial) at x = 0.
3.5. Quiver theories and W -constraints
The large-N solution of the quiver matrix integral (3.3) now proceeds along similar
lines [32,35]. One introduces r scalar fields ϕi (x) through the one-forms (3.7) as
yi (x) dx = ∂ϕi (x).
One can then show that the multi-valued fields ϕi (x) are actually the values of one single-
valued field ϕ(x) (essentially the full matrix model action) on a r + 1 branched cover of the
complex x-plane. This branched cover is the spectral curve associated to the quiver matrix
integral, and turns out to be given by (3.5) in the Ar case.
The general derivation of the curve proceeds through generalized loop equations.
For these multi-matrix model we do not only have the Virasoro constraints, expressing
reparametrization invariance in the matrix variables Φi . There are also higher order
relations [37,38]. The full set of loop equations are obtained by showing that the partition
function Z satisfies a set of W -constraints, labeled by the Casimirs of the corresponding
ADE Lie algebra, which contains the Virasoro constraints. These constraints take the form

dz 1 (s)
W (z) = 0,
2πi x − z
C
where W (s) (x) is a spin s current in the W -algebra. When expressed in modes these
equations take the form
Wn(s) · Z = 0, n 1 − s.
In the case of Ar there is leading spin r + 1 current that with a suitable basis of vectors
ϕ0 , . . . , vr can be written as

r
W (r+1) (x) ∼ (vi · ∂ϕ) + · · · .
i=0
We claim that in the planar limit this loop equation translates directly into the curve (3.5).
To be completely explicit let us give some more detail for the simplest case of A2 . Here
we have two matrices Φ1 , Φ2 with potentials Wi (Φ1 ) and W2 (Φ2 ). The classical singular
curve is after a shift in y given by

y − t1 (x) y − t2 (x) y − t3 (x) = 0
with
t1 = −(2W1 + W2 )/3, t2 = (W1 − W2 )/3, t3 = (W1 + 2W2 )/3,
all polynomials in x. To find the quantum curve we introduce the resolvents
1 1
w1 (x) = , w2 (x) = ,
x − λ1,I x − λ2,I
I I
and the one-forms yi (x) dx
y1 = W1 − µ(2ω1 − ω2 ), y2 = W2 − µ(2ω2 − ω1 ).
We now claim that the quantum curve is given by

y − a1 (x) y − a2 (x) y − a3 (x) = 0,
where the functions ai (x) are no longer polynomials, but instead are defined as
a1 = t1 + µω1 , a2 = t2 − µ(ω1 − ω2 ), a3 = t3 − µω2 .
With this choice we have, as claimed before,
a2 − a1 = y1 , a3 − a2 = y2 .
Now after some algebra, expanding out terms like ωi (x)3 , one verifies that indeed

y − a1 (x) y − a2 (x) y − a3 (x)

= y − t1 (x) y − t2 (x) y − t3 (x) + f (x)y + g(x) = 0
with f (x) and g(x) polynomials.
Acknowledgements
We would like to thank J. de Boer, M. Mariño, and E. Verlinde for discussions. The
research of R.D. is partly supported by FOM and the CMPA grant of the University of
Amsterdam, C.V. is partly supported by NSF grants PHY-9802709 and DMS-0074329.
References
[1] R. Dijkgraaf, C. Vafa, Matrix models, topological strings, and supersymmetric gauge theories, hep-
th/0206255.
[2] R. Gopakumar, C. Vafa, On the gauge theory/geometry correspondence, Adv. Theor. Math. Phys. 3 (1999)
1415, hep-th/9811131.
[3] C. Vafa, Superstrings and topological strings at large N , J. Math. Phys. 42 (2001) 2798, hep-th/0008142.
[4] F. Cachazo, K.A. Intriligator, C. Vafa, A large N duality via a geometric transition, Nucl. Phys. B 603
(2001) 3, hep-th/0103067.
[5] J.D. Edelstein, K. Oh, R. Tatar, Orientifold, geometric transition and large N duality for SO/Sp gauge
theories, JHEP 0105 (2001) 009, hep-th/0104037.
[6] K. Dasgupta, K. Oh, R. Tatar, Geometric transition, large N dualities and MQCD dynamics, Nucl. Phys.
B 610 (2001) 331, hep-th/0105066;
K. Dasgupta, K. Oh, R. Tatar, Open/closed string dualities and Seiberg duality from geometric transitions in
M-theory, hep-th/0106040;
K. Dasgupta, K. Oh, R. Tatar, Geometric transition versus cascading solution, JHEP 0201 (2002) 031, hep-
th/0110050.
[7] F. Cachazo, S. Katz, C. Vafa, Geometric transitions and N = 1 quiver theories, hep-th/0108120.
[8] F. Cachazo, B. Fiol, K.A. Intriligator, S. Katz, C. Vafa, A geometric unification of dualities, Nucl. Phys.
B 628 (2002) 3, hep-th/0110028.
[9] K.h. Oh, R. Tatar, Duality and confinement in N = 1 supersymmetric theories from geometric transitions,
hep-th/0112040.
[10] H. Fuji, Y. Ookouchi, Confining phase superpotentials for SO/Sp gauge theories via geometric transition,
hep-th/0205301.
[11] F. Cachazo, C. Vafa, N = 1 and N = 2 geometry from fluxes, hep-th/0206017.
[12] N. Seiberg, E. Witten, Electric–magnetic duality, monopole condensation, and confinement in N = 2
supersymmetric Yang–Mills theory, Nucl. Phys. B 426 (1994) 19;
N. Seiberg, E. Witten, Nucl. Phys. B 430 (1994) 485, hep-th/9407087, Erratum.
[13] D.J. Gross, E. Witten, Possible third order phase transition in the large N lattice gauge theory, Phys. Rev.
D 21 (1980) 446.
[14] S. Kachru, A. Klemm, W. Lerche, P. Mayr, C. Vafa, Nonperturbative results on the point particle limit of
N = 2 heterotic string compactifications, Nucl. Phys. B 459 (1996) 537, hep-th/9508155.
[15] A. Klemm, W. Lerche, P. Mayr, C. Vafa, N. Warner, Self-dual strings and N = 2 supersymmetric field
theory, Nucl. Phys. B 477 (1996) 746, hep-th/9604034.
[16] S. Katz, A. Klemm, C. Vafa, Geometric engineering of quantum field theories, Nucl. Phys. B 497 (1997)
173, hep-th/9609239.
[17] S. Katz, P. Mayr, C. Vafa, Mirror symmetry and exact solution of 4D N = 2 gauge theories I, Adv. Theor.
Math. Phys. 1 (1998) 53, hep-th/9706110.
[18] M. Bershadsky, S. Cecotti, H. Ooguri, C. Vafa, Kodaira–Spencer theory of gravity and exact results for
quantum string amplitudes, Commun. Math. Phys. 165 (1994) 311, hep-th/9309140].
[19] I. Antoniadis, E. Gava, K.S. Narain, T.R. Taylor, Topological amplitudes in string theory, Nucl. Phys. B 413
(1994) 162, hep-th/9307158.
[20] W.I. Taylor, D-brane field theory on compact spaces, Phys. Lett. B 394 (1997) 283, hep-th/9611042.
[21] P. Ginsparg, G.W. Moore, Lectures on 2D gravity and 2D string theory, hep-th/9304011.
[22] P. Di Francesco, P. Ginsparg, J. Zinn-Justin, 2D gravity and random matrices, Phys. Rep. 254 (1995) 1,
hep-th/9306153.
[23] M. Aganagic, M. Marino, C. Vafa, All loop topological string amplitudes from Chern–Simons theory, hep-
th/0206164.
[24] N.A. Nekrasov, Seiberg–Witten prepotential from instanton counting, hep-th/0206161.
[25] E. Witten, Solutions of four-dimensional field theories via M-theory, Nucl. Phys. B 500 (1997) 3, hep-
th/9703166.
[26] K. Hori, C. Vafa, Mirror symmetry, hep-th/0002222.
[27] K. Hori, A. Iqbal, C. Vafa, D-branes and mirror symmetry, hep-th/0005247.
[28] D.E. Diaconescu, B. Florea, A. Grassi, Geometric transitions, del Pezzo surfaces and open string instantons,
hep-th/0206163.
[29] M. Marino, Chern–Simons theory, matrix integrals, and perturbative three-manifold invariants, hep-
th/0207096.
[30] S. Kachru, S. Katz, A.E. Lawrence, J. McGreevy, Open string instantons and superpotentials, Phys. Rev.
D 62 (2000) 026001, hep-th/9912151.
[31] H. Ooguri, C. Vafa, Knot invariants and topological strings, Nucl. Phys. B 577 (2000) 419, hep-th/9912123.
[32] I.K. Kostov, Gauge invariant matrix model for the A-D-E closed strings, Phys. Lett. B 297 (1992) 74, hep-
th/9208053.
[33] I.K. Kostov, Bilinear functional equations in 2D quantum gravity, in: Razlog 1995, New trends in quantum
field theory 77–90, hep-th/9602117.
[34] I.K. Kostov, Conformal field theory techniques in random matrix models, hep-th/9907060.
[35] S. Kharchev, A. Marshakov, A. Mironov, A. Morozov, S. Pakuliak, Conformal matrix models as an
alternative to conventional multimatrix models, Nucl. Phys. B 404 (1993) 717, hep-th/9208044.
[36] A. Morozov, Integrability and matrix models, Phys. Usp. 37 (1994) 1, hep-th/9303139.
[37] R. Dijkgraaf, H. Verlinde, E. Verlinde, Loop equations and Virasoro constraints in nonperturbative 2D
quantum gravity, Nucl. Phys. B 348 (1991) 435.
[38] M. Fukuma, H. Kawai, R. Nakayama, Continuum Schwinger–Dyson equations and universal structures in
two-dimensional quantum gravity, Int. J. Mod. Phys. A 6 (1991) 1385.
Gauge-invariant action for superstring in

Ramond–Ramond plane-wave background
Machiko Hatsuda a , Kiyoshi Kamimura b , Makoto Sakaguchi a
a Theory Division, High Energy Accelerator Research Organization (KEK), Tsukuba, Ibaraki, 305-0801, Japan
b Department of Physics, Toho University, Funabashi, 274-8510, Japan
Received 18 July 2002; received in revised form 7 August 2002; accepted 29 August 2002
Abstract
We present an alternative form of gauge-invariant action for the superstring in the plane wave
background with Ramond–Ramond (RR) five-form flux. The Wess–Zumino term is given explicitly
in a bilinear form of the left-invariant currents by introducing a fermionic center to define the
nondegenerate group metric. The reparametrization invariance generators, whose combinations are
conformal generators, and fermionic constraints, half of which generate κ-symmetry, are obtained.
Equations of motion are obtained in conformal-invariant and background-covariant manners.
 2002 Elsevier Science B.V. All rights reserved.
PACS: 11.30.Pb; 11.17.+y; 11.25.-w
Keywords: Wess–Zumino term; Superalgebra; Plane wave background
1. Introduction
Recently the plane wave solution with the Ramond–Ramond (RR) 5-form flux was
found as a maximally supersymmetric type IIB supergravity solution [1] in addition to
the Minkowski flat and the AdS5 × S5 spaces, based on studies of the plane wave
solutions with the 4-form flux in the 11-dimensional supergravity [2]. The Penrose’s
limiting procedure [3] was applied to the AdS spaces to obtain these plane wave solutions
with fluxes (pp-wave) [4–6]. It is recognized as an approximation of AdS spaces and leads
to interesting approaches to the AdS/CFT correspondence [7].
E-mail addresses: mhatsuda@post.kek.jp (M. Hatsuda), kamimura@ph.sci.toho-u.ac.jp (K. Kamimura),

Makoto.Sakaguchi@kek.jp (M. Sakaguchi).
0550-3213/02/$ – see front matter  2002 Elsevier Science B.V. All rights reserved.
PII: S 0 5 5 0 - 3 2 1 3 ( 0 2 ) 0 0 7 9 0 - 3
M. Hatsuda et al. / Nuclear Physics B 644 (2002) 40–64 41
The gauge-invariant action for a superstring in the RR plane wave background was
presented [8] and it was also shown that the action in the light-cone gauge becomes simply
an action for 8 bosons and 8 fermions which are free and massive in 2 dimensions. Brane
actions in the RR pp-wave background have been widely studied [9] mostly in the light-
cone gauge. In the light-cone gauge the conformal symmetry is broken by a 2-dimensional
mass term at the gauge fixed level though it should be recovered in whole string theory
even in the RR pp-wave background [10]. The light-cone Hamiltonian does not commute
with other global space–time charges thus states in a supergravity multiplet have different
light-cone energy values [11] and the light-cone energy is not minimized for BPS states.
From a point of view of the symmetry, the light-cone approach is not always suitable
to understand systems. Manifest conformal-invariant approaches have been providing us
elegant formulations and practical computation methods in developments of string theories.
However, the conformal-invariant treatment in the RR pp-wave background has not been
explored except in an alternative hybrid approach [12].
In this paper we will study the superstring in the RR pp-wave background in a covariant
and manifestly conformal-invariant way. The RR backgrounds are usually described by the
Green–Schwarz type actions which contain Wess–Zumino (WZ) terms. In Ref. [8] the WZ
term of the superstring in the pp-wave background is given in a one parameter integral of
a closed three-form. Such integrals of the WZ terms are hardly performed explicitly for
general curved background cases, so it is difficult to discuss local symmetry constraints
and covariant equations of motion which will be needed in covariant string field theories.
In the flat background the integration in the WZ term is performed explicitly giving the
Green–Schwarz action [13]. In the pp-wave background it is performed only in the light-
cone gauge leading to the solvable action [8].
On the other hand, it was shown that an alternative form of the WZ term for the covariant
superstring in AdS spaces can be constructed in a bilinear form of the left-invariant (LI)
currents [14–17]. In contrast to the integral representation of the WZ term [18] the bilinear
form WZ term allows concrete computations of local symmetry constraints [19] and global
charges [17]. The WZ term can be constructed in a bilinear form of the LI currents due to
the existence of a nondegenerate group metric depending on the scale parameter. For the
super-pp-wave algebra the fermionic component of group metric is degenerate, so it cannot
be constructed in an analogous way. In this paper we find the bilinear form WZ term in the
super-pp wave background using the Penrose limit [5] of that in the super-AdS background
[17]. The limiting procedure must be taken carefully, since the bilinear form WZ term
contains a divergent coefficient when the Penrose limit parameter, Ω, is brought to zero.
The divergent term is a closed form which is a bilinear product of a leading term of the LI
current in Ω power series expansion, and it is subtracted. The finite contribution is given
by the next to leading term. The next to leading term of the LI currents can be obtained
not from the super-pp-wave algebra but from the nondegenerate super-pp-wave algebra.
42 M. Hatsuda et al. / Nuclear Physics B 644 (2002) 40–64
It corresponds to introducing a fermionic “center”1 to make the group metric nondegene-

rate [20].
The organization of this paper is as follows. The Penrose limit of the bosonic string is
explained as a simpler case in Section 2. In Section 3, the correct limiting procedure of
the bilinear form WZ term is explained. The resulting WZ term is shown to produce the
closed three-form H[3] . It is used to construct the κ-invariant action in the RR pp-wave
background. It is also shown that this action reduces to the light-cone action obtained
by Metsaev [8] and the Green–Schwarz action [13] in the flat limit. In Section 4, we
examine a particle, a superparticle, a bosonic string and a superstring cases. For each
systems generators of reparametrizations and fermionic constraints are calculated. We
obtain Hamiltonians and equations of motion in the conformal gauge.
2. Penrose limit of bosonic string action
The superstring in the AdS5 × S5 space is given as a sigma model action on

SU(2,2|4)
SO(4,1)×SO(5) [18], and we begin with a bosonic part of it. We use the same notation
of the superalgebra as [17,18]. Bosonic part of the LI Cartan 1-forms are given by
GAdS (y)−1 dGAdS (y) = eAdS a P + e a P + 1 ωab M + 1 ωa b M , where we use the
a AdS a 2 AdS ab 2 AdS a b
subscript “AdS” to indicate objects on the AdS5 × S5 space. Parametrizing an element of
the coset as GAdS (y) = exp(y â Pâ ), â = (a, a ), it follows the Cartan 1-forms as

sin y
a
eAdS = dy a + − 1 dy b ϒ b a ,
y

a sin y
eAdS = dy a + − 1 dy b ϒ b a , (2.1)
y
with

y= −y 2 = −y a y a, y = y 2 = y a ya ,

ya y b ya y b
ϒ ba = δa b − , ϒ ba = δa b − . (2.2)
y2 y 2
A scale parameter R, which is the radius of AdS5 and S5 , is introduced in the normalization
of F5 = (1/2R)(dvol(AdS5 ) + dvol(S5 )), by rescaling Pâ → RPâ . The bosonic part of the
sigma model action is written as
L0,AdS = eAdS
â
eAdS,â

y 2 2 y 2 2
= −(dy)2 + sin dΩ42 + dy + sin d Ω4 (2.3)
R R
1 For super-AdS and super-pp-wave algebra, this fermionic generator is not center of the superalgebra,
because it does not commute with Lorentz generators. In the flat case, the fermionic generator is center of the
supertranslation algebra (not the super-Poincaré algebra). We use the word “center” throughout this paper but will
not cause confusion.
where dΩ42 and dΩ42 are 4-sphere metrics. The Penrose limit is obtained as Ω → 0 after
rescaling
y + → Ω 2y +, y − → y −, y î → Ωy î , (2.4)
±
√ −2 −1
where y = (y ± y )/ 2, corresponding to P+ → Ω P+ , P− → P− , Pî → Ω Pî .
9 0
Rewriting (2.1) in terms of (2.4) and taking Ω infinitesimally small, (2.1) turns out to be
a power series of Ω. Cartan 1-forms should be also rescaled as e+ → Ω 2 e+ , e− → e− ,
eî → Ωeî for consistency. Taking Ω → 0 limit, leading terms of the expansion in Ω are
identified to the LI Cartan 1-forms in the plane wave background
−
sin √y y î −
2 + + 2R î dy
Ω eAdS = Ω dy +2
− − 1 − y − − dy î
+ o(Ω 4)
√y y y
2R
+
≡ Ω 2 epp + o(Ω 4),
−
eAdS = dy − + o(Ω 4) ≡ epp
−
+ o(Ω 4),
−
sin √y −
2R î dy
ΩeAdS = Ω dy −
î î
− − 1 y − − dy î
+ o(Ω 3 )
√y y
2R
≡ Ωepp
î
+ o(Ω ).3
(2.5)
Bosonic part of the sigma model action in the pp-wave background is written, using the
leading terms, as
L0,pp = epp
â
epp,â = 2dy + dy −
√ 2 î
sin(y − / 2R) y î 2 − 2 y −
+ √ −1 (dy ) − 2 − dy dy
î
y − / 2R y− y
√ 2
sin(y − / 2R)
+ √ dy î dy î
−
y / 2R

8
2 2
8
= 2dx + dx − − 2µx î dx − + dx î dx î , (2.6)
î=1 î=1
where the last expression is obtained by the following field redefinition
sin 2µy − î
x î = y,
2µy −

sin 4µy −
8
(y î )2
x − = y −, x+ = y+ + 1 − . (2.7)
2y − 4µy −
î=1
A scale parameter µ is introduced in the normalization F5 = µdx − ∧ (dx 1dx 2 dx 3 dx 4 +

dx 5 dx 6 dx 7 dx 8 ). This coordinate system is obtained if one begins with the coset
+P −P î P
parameterization Gpp (x) = ex + ex − ex î .
3. Penrose limit of superstring action
3.1. Wess–Zumino term and nondegenerate super-pp-wave algebra
For an Inönü–Wigner group contraction [21] the Cartan 1-forms are expanded with
respect to a parameter s which is brought to zero as
∞

TA → s −NA TA −→ LA (z) → s NA LA s NB zB = s N LA
(N) (z), (3.1)
N=NA
as was seen in (2.5) for bosonic part. The Maurer–Cartan equations are also expanded as
∞
∞ ∞
1 A
(N) + fBC
s N dLA (M) L(K) = 0,
s M+K LB C
(3.2)
2
N=NA M=NB K=NC
A is the structure constant of the original Lie algebra. Usually only leading terms
where fBC
of Cartan 1-forms are kept in the limiting procedure as was done in the previous section.
The Maurer–Cartan equations for the leading terms
1 A(NA )
(NA ) + fB(NB ) C(NC ) L(NB ) L(NC ) = 0,
dLA B C
A(N ) 2
fB(NBA) C(NC ) = fBC
A , for N = N + N ,
A B C
A(NA ) (3.3)
fB(NB ) C(NC )
= 0, for NA = NB + NC ,
describe the resultant group structure in the s → 0 limit.
The Penrose limit from the super-AdS group to the super-pp-wave group makes the
super-pp-wave group metric to be degenerate. This is the similar situation as in the flat limit
from the super-AdS group to the super-translation group where the metric is degenerate.
However, if the next to leading term in the fermionic Cartan 1-forms (3.1) is maintained in
the limiting procedure, the nondegenerate group metric of the central extended super-pp-
wave group can be constructed as we will show below.
The nondegenerate group metric can be defined in the super-AdS space and the bilinear
form WZ term can be constructed as follows. In terms of light-cone indices â = (+, −, i, i )
î = (i, i ) and the light cone projection operators for spinors #±
θ ± = #± θ, ζ± = ζ #± , Q± = Q#± ,
1 1
#± ≡ Γ± Γ∓ , Γ± = √ (Γ9 ± Γ0 ), (3.4)
2 2
the Cartan 1-forms for the super-AdS5 × S5 space are written as
G−1 + − + −
AdS dGAdS = LAdS P+ + LAdS P− + LAdS Pî + LAdS Q+ + LAdS Q−
î
1 î jˆ
+ Lî∗AdS P ∗ + LAdS Mî jˆ , (3.5)
î 2
where
Pi∗ = M0i , Pi∗ = M9i . (3.6)
The MC equation for Q− is given by

√
1 î jˆ 2 î
dL− + L Γ ˆ L−
− L ΠΓî ΠΓ− L+
AdS 4 AdS î j AdS 4 ∗AdS AdS
1 +
+ √ * −2L+ −
AdS ΠLAdS − LAdS ΠΓî Γ− LAdS = 0
î
(3.7)
2 2R
with Π = Γ1234 and * = iτ2 , and the MC equation for Q+ has a similar form to this. From
Q Q Q
these MC equations it can be seen that structure constants fP+−Q− , fP −Q+ , fP−+Q+ and
î
Q
fP +Q− are present and the nondegenerate group metric can be defined consistently. The
î
bilinear form WZ term can be constructed for the super-AdS space [17]
I ββ J αα I
[2],AdS = R Lαα
B
AdS Cαβ Cα β τ1,I J LAdS − dθ Cαβ Cα β τ1,I J dθ ββ J . (3.8)
Here the nondegenerate metric for spinor index is ρQαα I Qββ J = Cαβ Cα β (τ1 )I J ≡ ραβ ,
and this is used for the WZ term as Lα Lβ ραβ . It gives d(Lα Lβ ραβ ) = Lα Lc Lβ fαcβ ∼ H[3]
γ
with totally antisymmetric structure constants defined as fcβ ργ α . In [17] it was shown that
(3.8) gives correct exterior derivative d B[2] = H[3] , κ-invariance of the total action and
the correct flat limit. It was also shown that the second term in (3.8) is required for the
pseudo-supersymmetry invariance giving the correct string charge in the superalgebra.
Let us discuss the Penrose limit of this bilinear form WZ term, (3.8). The Penrose limit
is taken as the following rescaling [5]
θ + → Ωθ + , θ− → θ− (3.9)
in addition to (2.4). This corresponds to s = Ω in (3.1) and the scaling dimensions NA in
(3.2) to be
NP+ = 2, NPî = 1, NP− = 0, NP ∗ = 1,

î
NMî jˆ = 0, NQ+ = 1, NQ− = 0. (3.10)
Under this rescaling the AdS–WZ term (3.8) becomes

iR − −
[2],AdS = √
Ω 2B −L(0) Γ τ1 ΠL− − −
(0) + d θ̄ Γ τ1 Π dθ
−
2
+ Ω 2L+ Γ + τ1 ΠL+ − Ω 2 2L − Γ − τ1 ΠL−
(1) (1) (0) (2)
+ + +

− Ω d θ̄ Γ τ1 Π dθ + o(Ω )
2 3
(3.11)
= LC and C is the charge conjugation matrix in 10 dimensions. The scaling factor
where L
2
Ω of B[2] is determined by requiring the same scaling weight as the string kinetic term,
the Nambu–Goto action.
Under the Penrose limit the 0th-order equation of the MC equation for Q− in (3.3)
becomes a trivial one according to (3.10)
î jˆ
dL− −
− Q
(0) + fM ˆ Q− L(0) L(0) = 0.
î j
(3.12)
This implies that the group metric is degenerate in the ‘−’ spinor direction. In order to
Q Q
make it nondegenerate fP+−Q− and fP −Q+ must be included, so additional form is required
î
whose scaling dimension is NP+ + NQ− = NPî + NQ+ = 2. It is L− −
(2) ≡ Lpp and must be
maintained in the limiting procedure, and explicit computation confirms that L − −
pp = L(2)
is the next to leading term. Under the Penrose limit the Cartan 1-forms of the super-AdS
group reduce to those of the super-pp-wave background as
NA A NA +2 ) for A = Q ,
Ω LAdS = Ω NA LA pp + o(Ω −
(3.13)
L−
AdS = L − + Ω 2
pp L − + o(Ω 4 ),
pp
and they satisfy the following MC equations for bosonic and fermionic Cartan 1-forms,
respectively,
1 î î
dL+ +
pp − √ Lpp L∗pp − i Lpp Γ L = 0,
2
dL− −
pp − i Lpp Γ Lpp = 0,
1 jˆ jî̂
dLîpp + √ L−
pp L∗pp + Lpp Lpp − i Lpp Γ Lpp = 0,
î î
2
√ √
dLi∗pp − 4 2 µ2 L− i j ji
pp Lpp + L∗pp Lpp + 2 2 iµLpp ΠΓ *Lpp = 0,
i
√ i j j i √ i
dLi∗pp − 4 2 µ2 L−
pp Lpp + L∗pp Lpp + 2 2 iµLpp Π Γ *Lpp = 0,
ij
dLpp + Lki
jk
pp Lpp + 2iµLpp ΠΓ
−ij
*Lpp = 0,
j k
ij pp Π Γ −ij *Lpp = 0,
dLpp + Lkppi Lpp + 2iµL (3.14)
1 î jˆ √ î
dL+
pp + Lpp Γî jˆ L+ −
pp + 2 L∗pp Γ+î Lpp
4
−
+ µ* 2Lpp ΠL+ − ΠLîpp Γî+ L− pp = 0,
1 î jˆ
dL− −
pp + Lpp Γî jˆ Lpp = 0,
4
− 1 î jˆ − √ î
dLpp + Lpp Γî jˆ Lpp − 2 L∗pp ΠΓî ΠΓ− L+ pp
4

+ µ* −2L+ −
pp ΠLpp − Lpp ΠΓî Γ− pp
î +
= 0, (3.15)
where Π = Γ5678 . These MC equations coincide with the ones obtained by the direct
computation [8] using
G−1 + − + −
pp dGpp = Lpp P+ + Lpp P− + Lpp Pî + Lpp Q+ + Lpp Q−
î
1 î jˆ
+ Lî∗pp P ∗ + Lpp Mî jˆ , (3.16)
î 2
except for the last equation for L−
pp . It would be obtained if one uses an extended super-pp-
wave algebra with a fermionic center. The last equation of (3.15) has the same form as the
MC equation of the AdS (3.7) then it becomes nondegenerate.
Following to above limiting procedure from the super-AdS WZ term (3.8) we propose
a superstring action in the super-pp-wave background as

S = d σ L = d 2 σ (L0 + LWZ ),
2
√
L0 = −T −h huv Lu â Lv b̂ ηâb̂ , [2],pp ,
LWZ = T B (3.17)
where the WZ term is
+I
[2],pp = i L
B Γ− (τ1 )I J ΠL+J −I −J
pp − 2Lpp Γ+ (τ1 )I J Π Lpp
4µ pp

− d θ̄ +I Γ− (τ1 )I J Π dθ +J , (3.18)
and u, v = τ, σ = 0, 1 are the world volume indices and Cartan 1-forms are LA =
dzM LM A = dσ u Lu A with zM = (x m , θ µ ).
It satisfies the following criteria:
(i) giving the correct three-form, d B[2] = H[3] for dH[3] = 0,

(ii) κ-invariance of the total action.
(i) Three-form H[3]

From now on the currents LA A
pp is denoted just as L . The first condition (i) is checked
by using the MC equations (3.14) and (3.15),
[2] = i +I −I Γ+ (τ1 )I J Π L

−J
dB 2L Γ− (τ1 )I J Π dL+J − 2 d L
4µ

− 2L−I Γ+ (τ1 )I J Π d L
−J
+ −
= −i L + Lî Γ τ3 L− + i √1
L Γ− τ3 L+ + L
L+ −Li∗ Γi + Li∗ Γi Πτ1 L−
î
2 2µ
− +
−i L − Lî Γ τ3 L+ + i √1 L
L Γ− τ3 L− + L + Li∗ Γi − Li∗ Γi Πτ1 L−
î
2 2µ
Γâ τ3 L
= −i LL â
= H[3] . (3.19)
(ii) κ-invariance
The second condition (ii) κ-invariance of the action is confirmed as follows. Let us
denote an arbitrary variation of the coset element δG as the following combination
G−1 δG = ;LA TA = δzM LM A TA , (3.20)

and the variation of the Cartan 1-form is given by

δLA TA = δ G−1 dG = d(;LA ) − ;LB LC fBC A
TA . (3.21)
Then an arbitrary variation of a Cartan 1-form is obtained by referring to the corresponding
MC equation.
The κ-variations (δθ µ = κ µ ) of Cartan 1-forms are given by
δκ Lâ = −fBC
â
;κ LB LC ,

δκ Lα = d ;κ Lα − fBC α
;κ LB LC , (3.22)
where ;κ Lâ = 0 is imposed for the κ-variations. Concrete expression is
1
+ ;κ L = 0,
δκ L+ − √ ;κ Lî Lî∗ + Lî ;κ Lî∗ − 2i LΓ
2
− ;κ L = 0,
δκ L− − 2i LΓ
1 ˆ ˆ ˆ ˆ î ;κ L = 0,
δκ Lî + √ ;κ L− Lî∗ + L− ;κ Lî∗ + ;κ Lj Lj î + Lj ;κ Lj î − 2i LΓ
2
(3.23)

δκ L+ + µ* 2 ;κ L− ΠL+ + L− Π;κ L+ − Π ;κ Lî Γî Γ+ L− + Lî Γî Γ+ ;κ L−
1 ˆ ˆ 1
+ ;κ Lî j Γî jˆ L+ + Lî j Γî jˆ ;κ L+ − √ ;κ Lî∗ Γî+ L− + Lî∗ Γî+ ;κ L− = 0,
4 2 2
1 ˆ ˆ
δκ L− + ;κ Lî j Γî jˆ L− + Lî j Γî jˆ ;κ L− = 0,
4

δκ
L− − µ* 2 ;κ L+ ΠL− + L+ Π;κ L− − Π ;κ Lî Γî Γ− L+ + Lî Γî Γ− ;κ L+
1 ˆ ˆ
+ ;κ Lî j Γî jˆ
L− + Lî j Γî jˆ ;κ
L−
4
1
− √ Π ;κ Lî∗ Γî− ΠL+ + Lî∗ Γî− Π;κ L+ = 0. (3.24)
2 2
Using these relations, the κ-variation of the total action is calculated as
√
δκ (L0 + bLWZ ) = −T δκ −g g uv Lâu Lâ,v + bT * uv B[2]uv

= −2iT ;κ L − det Guv GuvL / u 1 + b* uvL
/ u τ3 Lv

= −2iT ;κ L (−Γ(1) + b)* uvL / u τ3 Lv (3.25)
with
1
Γ(1) = √ /,
τ3L 2
Γ(1) = 1, tr Γ(1) = 0,
2 − det Guv

− − det Guv GuvL
/ v = Γ(1)τ3 * uvL
/v. (3.26)
The action has the κ-invariance whose parameter κ is satisfying
(−Γ(1) + b)κ = 0, b = ±1. (3.27)

3.2. Gauge-invariant action for a super-pp-string
We present an explicit expression of the gauge-invariant action for a superstring in the

RR pp-wave background, with b = 1 of (3.27),

√
S = d 2 σ L = d 2 σ (L0 + LWZ ), L0 = −T −h huv Lu â Lv b̂ ηâb̂ ,
i + u − Γ+ τ1 Π L
v − + ∂u θ̄ + Γ− τ1 Π∂v θ +

LWZ = T * uv −Lu Γ− τ1 ΠLv + + 2L
4µ
(3.28)
where the Cartan 1-forms in the pp-wave background are obtained from the ones in the
super-AdS5 × S5 given in Appendix A. The results are

sin(Ψ+ /2) 2 sin(Ψ/2) 2
L+ = e+ + i θ̄ + Γ + (Dθ )+ + i θ̄ + Γ + (Dθ )− Ω 0 ,
Ψ+ /2 Ψ/2
2
sin(Ψ− /2)
L− = e− + i θ̄ − Γ − (Dθ )− Ω 0 ,
Ψ− /2
2 2
+ î sin(Ψ− /2) −
− î sin(Ψ+ /2)
L = e + i θ̄ Γ
î î
(Dθ ) Ω 0 + i θ̄ Γ (Dθ )+
Ψ− /2 Ψ+ /2

sin(Ψ/2) 2
+ i θ̄ − Γ î (Dθ )− Ω 0 ,
Ψ/2 Ω

sin Ψ+ sin Ψ
+
L = +
(Dθ ) + #+ #− (Dθ )− Ω 0 ,
Ψ+ Ψ Ω
− sin Ψ− −

L = (Dθ ) Ω 0 ,
Ψ−

sin Ψ− −
sin Ψ− −
sin Ψ
−
L = (Dθ ) Ω 2 + (Dθ ) Ω 0 + #− #+ (Dθ )+ ,
Ψ− Ψ− Ω2 Ψ Ω
(3.29)
with the covariant derivatives
1 ˆ
(Dθ )− Ω 0 = dθ − + ωî j Γî jˆ θ − ,
4
√
2 î
(Dθ )− Ω 2 = −µ* 2e+ Πθ − − Πeî Γî Γ− θ + − e ΠΓî ΠΓ− θ + ,
4 ∗
√
1 ˆ 2 î
(Dθ )+ = dθ + + ωî j Γî jˆ θ + + µ* 2e− Πθ + − Πeî Γî Γ+ θ − − e Γ Γ+ θ − ,
4 4 ∗ î
(3.30)
and arguments of the sin’s are

Ψ 2 = 2µi *Π 2θ + θ̄ − Γ+ − 2θ − θ̄ + Γ− + Γ+ Γ î θ − θ̄ Γî − Γ− Γ î θ + θ̄ Γî

Γ+ Γ î θ − θ̄ − Γî Π − Γ î θ − θ̄ + ΠΓî + Γ− Γ î θ + θ̄ − ΠΓî − Γ î θ + θ̄ + Γî Π *
1 ij −
− Γ θ θ̄ Γij − Γ i j θ θ̄ − Γi j Γ+
2
ij +
− Γ θ θ̄ Γij − Γ i j θ θ̄ + Γi j Γ− Π* , (3.31)

#+ Ψ 2 #+ ≡ Ψ+2 = 2µi *ΠΓ+ Γ î θ − θ̄ − Γî + Γ+ Γ î θ − θ̄ − Γî Π* ,

#− Ψ 2 #− ≡ Ψ−2 = −µi Γ ij θ − θ̄ − Γij − Γ i j θ − θ̄ − Γi j Γ+ Π*,

#+ Ψ 2 #− Ω = 2µi *Π 2θ + θ̄ − Γ+ + Γ+ Γ î θ − θ̄ + Γî − Γ+ Γ î θ − θ̄ + ΠΓî *
1
− Γ ij θ + θ̄ − Γij − Γ i j θ + θ̄ − Γi j Γ+ Π* ,
2

#− Ψ 2 #+ Ω = 2µi *Π −2θ − θ̄ + Γ− − Γ− Γ î θ + θ̄ − Γî + Γ− Γ î θ + θ̄ − ΠΓî *
1
+ Γ ij θ − θ̄ + Γij − Γ i j θ − θ̄ + Γi j Γ− Π* ,
2

#− Ψ #− Ω 2 = 2µi *ΠΓ− Γ î θ + θ̄ + Γî − Γ− Γ î θ + θ̄ + Γî Π* .
2
The Cartan 1-forms are same as the one in [8] except for L − . From now on we use the
“x-coordinates” in (2.7) for simplicity in which the bosonic Cartan 1-forms are given by
2
e+ = dx + − 2µ2 x î dx − , e− = dx − , eî = dx î ,
√ ˆ
e∗î = −4 2 µ2 x î dx − , ωî j = 0. (3.32)
We confirm that this action reduces to following known ones:
(i) light-cone action in the light cone gauge,

(ii) flat action in the flat limit µ → 0.
(i) Light-cone gauge action

In the light-cone gauge with the conformal metric
x+ = p+ τ, θ − = 0, huv = ηuv (3.33)

the Cartan 1-forms reduce into

L+ = e+ + i θ̄ + Γ + dθ + + 2µ*e+ Πθ + , L− = e+ , Lî = eî ,
L+ = dθ + + 2µ*e+ Πθ + , L− = 0,
√
− = µ*Πeî Γ θ + − 2 e∗î ΠΓ ΠΓ− θ + .
L (3.34)
î− 4 î
The light-cone action of a superstring in the super-pp-wave background is given by

1 u î 2
S = −T d 2 σ ∂ x ∂u x î + m2 x î
2

+ +
− ip+ θ̄ Γ (∂0 + τ3 ∂1 )θ + + mθ̄ + Γ + *Πθ + , m ≡ 2µp+ , (3.35)
which coincides with the one obtained by Metsaev [8].

(ii) Flat action
The flat limit is realized by taking µ → 0 limit. The kinetic term L0 in the flat limit is
straightforwardly obtained by taking µ → 0 in Cartan 1-forms Lâ as
Lâ = dx â + i θ̄Γ â dθ. (3.36)
The WZ term LWZ in the flat limit must be evaluated with care. Term proportional to 1/µ
is absent in (3.17). Terms proportional to µ0 are obtained from terms proportional to µ1 in
L’s

+ +
− + −
1 2 + −

L = dθ + µ *Π 2e θ − e Γî Γ+ θ +î
Ψ dθ + #+ Ψ #− dθ2
3! +
+ o(µ2 ),

L = µ −*Π 2e+ θ − − eî Γî Γ− θ +
−

1 2 −
+ Ψ− dθ + #− Ψ 2 #− Ω 2 dθ − + #− Ψ 2 #+ dθ + + o(µ2 ),
3!
L− = dθ − + o(µ2 ). (3.37)
The WZ term becomes
i
+ + − − 3
LWZ = T µ 2d θ̄ Γ− τ1 Π(L ) µ1 − 2d θ̄ Γ+ τ1 Π(L ) µ1 + o(µ )
4µ µ→0

1 1
= T i dx â θ̄ Γâ dθ + θ̄ Γ â τ3 dθ θ̄ Γ â dθ + θ̄ΠΓ î τ1 dθ θ̄ ΠΓ î * dθ
3 3

1 −
− θ̄ ΠΓ −ij τ1 dθ θ̄ ΠΓ −ij * dθ − θ̄ − ΠΓ −i j τ1 dθ θ̄ ΠΓ −i j * dθ
12

1
= T i dx â θ̄ Γâ dθ + θ̄ Γ â τ3 dθ θ̄ Γ â dθ , (3.38)
2
where the last equality is derived by using the relation in Section 2.3 of [17]. This action
is the Green–Schwarz superstring action in a flat space [13].
4. Hamiltonian and equations of motion
In this section we calculate the Hamiltonian, constraints and equations of motion

of the system. Before examining the superstring in the pp-wave background, particle,
superparticle and bosonic string cases are examined.
4.1. Bosonic particle
The action for a bosonic particle system in a curved background is

1
SPA = dτ ẋ m gmn ẋ n .
2e
The canonical momentum of x m is pm = δSPA /δ ẋ m = (1/e)gmn ẋ n and the gauge-invariant

Hamiltonian is
1
HPA = eHPA , HPA = pm g mn pn = 0. (4.1)
2
It leads to the following equations of motion in e = 1 gauge
1
ẋ m = g mn pn , ṗm = − (∂m g nl )pn pl , (4.2)
2
then
ẍ m = −Γ m nl ẋ n ẋ l (4.3)
with the Affine connection coefficients defined by Γ l nk = 12 g lm (−∂m gnk + ∂(n gk)m ).
For the pp-wave background the metric is given by
   
g++ g+− g+jˆ 0 1 0
gmn =  g−+ g−− g−jˆ  =  1 −4µ2 x 2 0  ,
gî+ gî− gî jˆ 0 0 δî jˆ
 2 2 
4µ x 1 0
g mn =  1 0 0 , (4.4)
0 0 δî jˆ
the Hamiltonian in the e = 1 gauge is given by
1
HPA = 2p+ p− + p 2 + (2µp+ )2 x 2 (4.5)
2
and equations of motion are
ẋ + = p− + (2µ)2 x 2 p+ , ẋ − = p+ , ẋ = p,
ṗ− = 0, ṗ+ = 0, ṗ = −(2µp+ ) x, 2
(4.6)
with x = x î and p = pî . The second-order equations of motion can be written as
ẍ + = (2µ)2 p+ (x˙2 ), ẍ − = 0, ẍ = −ω2 x, ω = 2µp+ , (4.7)

and (4.6) and (4.7) are solved as
1
x + = p− + 2µ2 p+ α 2 + α̃ 2 τ − µ α 2 − α̃ 2 sin(4µx − )
2
− µα · α̃ cos(4µx − ) − 1 + x0+ ,
x − = p+ τ + x0− ,
x = α sin(2µx − ) + α̃ cos(2µx − ), (4.8)
where constants α and α̃ are amplitudes of harmonic motions in transverse directions. The
Hamiltonian is also written as
1
HPA = 2p+ p− + ω2 α 2 + α̃ 2 . (4.9)
2
The constants of motion are the global translation and boost charges Pâ and P ∗ , instead
î
of the canonical momenta pî . Their forms in terms of the canonical variables are calculated
from
λâ Pâ = pm δλ x m = pm ;λ LA (L−1 )A m ,

G−1 −1 â
pp δλ Gpp = Gpp (λ Pâ )Gpp = ;λ L TA ,
A
(4.10)
etc., and are

 P− = p− ,
Pâ = P+ = p+ ,

Pî = pî cos(2µx − ) + 2µp+ xî sin(2µx − ) = ωαî ,

∗ 1 − −
1
P = − √ pî sin(2µx ) + 2µp+ xî cos(2µx ) = − √ ωα̃î . (4.11)
î 2 2µ 2 2µ
In terms of global charges the Hamiltonian can be written as the quadratic Casimir operator
of the pp-algebra
1 î jˆ î jˆ ˆ √ 2
HPA = Pâ ηâ b̂ Pb̂ + Pî∗ η∗ Pj∗ˆ , η∗ ≡ δ î j 2 2 µ . (4.12)
2
This Hamiltonian is justified algebraically as it commutes with all global space–time
symmetry charges. It is also obtained by the Penrose limit of the Hamiltonian of the AdS
particle which is also the quadratic Casimir operator of the AdS algebra:
 $ $
 cAdS =
 Pa Pb ηab + (Mab )2
a=0,1,...,4 ab=0,...,4
$ $

 cS = Pa Pb ηa b + (Ma b )2
a =5,...,9 a b =5,...,9
 $ $
 cpp(−2) = î jˆ
Penrose limit Pâ Pb̂ ηâb̂ + P ∗ P ∗ˆ η∗ ⇒ 2HPA ,
−→ â=0,1,...,9 î=1,...,8
î j
(4.13)

cpp(−4) = P+2 .
There are two quadratic Casimir operators each for the AdS5 and S5 . They turn to become
two quadratic Casimir operators in the pp algebra. The one, which contains the Lorentz-
invariant mass term in a flat limit µ → 0, is the Hamiltonian of this system.
4.2. Superparticle
The action for a superparticle system in the pp-wave background is obtained by

eliminating σ dependence in (3.17), (3.29), (3.30), (3.31),

1
SSUPA = dτ L0 â L0 b̂ ηâb̂ , (4.14)
2e
where L0 â is the coefficient of dτ in the pullback of the MC form Lâ,
L0 â = ẋ m Lm â + θ̇ µ Lµ â , Lm â ≡ em â + Θm µ Lµ â . (4.15)
Canonical momenta for x m and θ µ are

δSSUPA T
pm ≡ = Lm â L0 b̂ ηâ b̂ ,
δ ẋ m e
δ r SSUPA T
ζµ ≡ = Lµ â L0 b̂ ηâb̂ . (4.16)
δ θ̇ µ e
The gauge-invariant Hamiltonian of the pp-superparticle is obtained as
HSUPA = eHSUPA + FSUPA,µ Λµ , (4.17)
where primary constraints are
1
HSUPA = πâ πb̂ ηâb̂ = 0, (4.18)
2
FSUPA,µ = ζµ − πâ Lµ â = 0, (4.19)
with super-invariant (up to local Lorentz) combination
T
πâ = (e−1 )â m pm + ζµ Θm µ = L0 b̂ ηâ b̂ ≡ (e−1 )â m πm . (4.20)
e
Half of the fermionic constraints generate the κ-symmetry transformations whose
existence has been proven in Subsection 3.2. Other half is the set of second class constraints
and the multipliers, Λµ ’s in (4.17), associating with the second class constraints vanish by
the consistency condition.
Equations of motion in e = 1 and Λ = 0 gauge are obtained in a background-covariant
way as
1
ẋ m = π m , ṗm = − ∂m g nl πn πl − ζ(∂m Ξn )θ π n , (4.21)
2
θ̇ µ = π â (Ξâ θ )µ , ζ̇µ = −π â (ζ Ξâ )µ , (4.22)
with Θm = (Ξm
µ )µ ν θν
and second-order equation is given by

ẍ m = −Γ m nl ẋ n ẋ l − g mn ẋ l ζ ∂[n Ξl] − [Ξn , Ξl ] θ. (4.23)
where indices â and m are raised and lowered by ηâ b̂ and g mn , respectively.
2
The world line element is calculated as dsSUPA = eHSUPA = 0, where (4.20) is used.
The superparticle, which is the zero mode of the superstring, in the pp-wave background
moves along the null geodesics and it satisfies the “massless” constraint HSUPA = 0.
In order to solve equations of motion the concrete expression of the RR pp-wave
background metric (4.4) and

 Ξ+ = 0,
Ξm = Ξ− = #+ 2µ*Π − 2µ2 x î Γî Γ+ , (4.24)

Ξî = −µ*ΠΓî Γ+ ,
are inserted. The Hamiltonian is given by
1
HSUPA = 2p+ (p− + ζ Ξ− θ ) + 4µ2 p+ x 2 + (pî + ζ+ Ξî θ )2 (4.25)
2
1
= 2p+ π− + ω2 x 2 + π 2 ,
2
with pm = (p+ , p− , p), πm = (π+ , π− , π) and x m = (x + , x − , x). The equations of

motion become
ẋ + = π− + (2µ)2 x 2 p+ , ẋ − = p+ , ẋ î = πî ,
ṗ− = 0, ṗ+ = 0, ṗî = −ω2 xî + 2µ2 p+ ζ Γî Γ+ ,
θ̇ µ = p+ (Ξ− θ )µ + πî (Ξî θ )µ , ζ̇µ = −p+ (ζ Ξ− )µ − πî (ζ Ξî )µ . (4.26)
Using facts π̇− = 0, π̇ = −ω2 x, the second-order equations for x m take the same form as
the bosonic particle (4.7). Furthermore, θ µ and ζµ satisfy second-order harmonic equation
with frequency ω. It is shown by using ∂î Ξ− = −Ξ− Ξî , (Ξ− )2 = −4µ2 + 4µ2 x î Ξî and
Ξî Ξjˆ = 0. Solutions are found in the same form as (4.8) and (4.12) with replacing p− , pî
by π− , πî . The Hamiltonian is expressed as
HSUPA = EB + EF ,
1
EB = 2p+ π− + ω2 α 2 + α̃ 2 = const,
2
2
+ + sin(Ψ+ /2)
EF = 2p+ µi θ̄ Γ
2
* Πθ +
Ψ+ /2

2 2 − sin(Ψ+ 2) 2 − −
− 2p+ µ i θ̄ π/ x/ Γ θ = −EB , (4.27)
Ψ+ /2
where ‘+’ projected part of fermionic constraints (4.19) are used.
In terms of global charges the Hamiltonian can be written as the quadratic Casimir
operator of the super-pp-wave algebra which is obtained by the Penrose limit from the
quadratic Casimir operator of the super-AdS algebra:
1
csAdS = Pâ Pb̂ ηâb̂ + Mâ2b̂ − Qαα A * AB C −1αβ C −1α β Qββ B
4
â=0,...,9 â b̂=0,...,4
or 5,...,9
Penrose limit î jˆ iη∗
−→ cspp(−2) = Pâ Pb̂ ηâb̂ + Pî∗ Pj∗ˆ η∗ − Q+ C −1 Γ+ Π* Q+
4
â=0,...,9 î=1,...,8
⇒ 2HSUPA (4.28)
î jˆ √ ˆ √
with η∗ = (2 2µ)2 δ î j and η∗ = 2 2µ. The supercharge is obtained by the Penrose limit
from the one of the super-AdS result [17]

i − −
+ + − î
sin(Ψ+ /2) 2
Q+ = ζ+ 1 − √ Γ+î θ θ̄ Γî Π* − p+ i θ̄ Γ + pî i θ̄ Γ
2 2 Ψ+ /2

Ψ+
× cos(2µx − ) − *Π sin(2µx − ) (4.29)
sin Ψ+
which is consistent with that the Hamiltonian (4.26) and the quadratic Casimir operator in
(4.28).
4.3. Bosonic string
The action for a bosonic string system is obtained by eliminating θ dependence in

(3.28),

√
SST = −T d 2 σ −h huv ∂u x m gmn ∂v x n .
√
Canonical momentum of x m (σ ) is pm (σ ) = δSST /δ ẋ m (σ ) = −T −h hu0 ∂u x n gmn . The
gauge-invariant Hamiltonian of the pp-string is obtained as
√
−h h01
HST = dσ H⊥ + H , (4.30)
h11 h11
1 n
pm g mn pn + T 2 x gmn x = 0,
m
H⊥ (σ ) = (4.31)
2T
H (σ ) = pm x = 0.
m
(4.32)
The equation of motion in the conformal gauge, huv = ηuv , are obtained as
1 mn
ẋ m (σ ) = g pn ,
T

n T 1 n l n l
ṗm (σ ) = T gmn x + (∂m gnl ) p p −x x , (4.33)
2 T2
and
n l
✷x m (σ ) = −Γ m nl ẋ n ẋ l − x x (4.34)
with ✷ = ∂τ2 − ∂σ2 = ∂02 − ∂12 .
For the RR pp-wave background case, by using (4.4), the Hamiltonian becomes

1
HST = dσ 2p+ p− + p 2 + (2µp+ )2 x 2 + T 2 (x + ) (x − )
2T

+ x 2 − (2µ)2 x 2 (x − ) 2 . (4.35)
The equations of motion are
1 1 1
ẋ + (σ ) = p− + (2µ)2 x 2 p+ , ẋ − (σ ) = p+ , ẋ(σ ) = p,
T T T

ṗ− (σ ) = T (x + ) − (2µ)2 x 2 (x − ) , ṗ+ (σ ) = T (x − ) ,
2 %
p+ − 2
ṗ(σ ) = −T x + (2µ) x 2
− (x ) (4.36)
T
and ones of the second-order form become
✷x + (σ ) = (2µ)2 (x˙2 )ẋ − , ✷x − (σ ) = 0,

✷x(σ ) = (2µ)2 x (x − ) − (ẋ − )2 .
2
(4.37)
By solving equation of motion for x − in (4.37) as

p+,0 1 − 2in(σ −τ )
x − (σ ) = τ+√ αn e + α̃n− e−2in(σ +τ ) , (4.38)
T 2π n=0
the last term of the last equation in (4.37) becomes

− 2
(x ) − (ẋ − )2

p+,0 − 2in(σ −τ ) p+,0 − −2in(σ +τ )
=− − 4i nαn e − 4i nα̃n e . (4.39)
T T
n=0 n=0
It is difficult to solve the equation for x except in a case where only the zero-mode term
(p+,0 /T )2 is present. It is the same as the light-cone result [18]
1
x = α 0 sin(ω0 τ ) + α̃ 0 cos(ω0 τ ) + √ α n ei(ωn τ +2nσ ) + α̃ n ei(ωn τ −2nσ ) ,
2π n=0

ωn = sgn(n) (2µp+,0 )2 + (2n)2 , (4.40)
with α †n = α −n and α̃ †n = α̃ −n and the Hamiltonian becomes

1 2
HST = ω02 α 20 + α̃ 20 + ωn α n α −n + · · · . (4.41)
2T
n=0
4.4. Superstring
Now let us examine a superstring in the RR pp-wave background which is described by

the covariant action (3.28). The conjugate momenta are
δL δL0 δLWZ
pm (σ ) ≡ m
= Lm â + Lm α , (4.42)
δ ẋ δL0 â δL0 α
δr L δL0 δ r LWZ α
ζµ (σ ) ≡ µ = L µ
â
+ Lµ . (4.43)
δ θ̇ δL0 â δL0 α
The gauge-invariant Hamiltonian of the pp-superstring is obtained as
√
−h h01
H= H⊥ + H ,
h11 h11
H⊥ (σ ) = H⊥ − F τ3 θ , H (σ ) = H + F θ , (4.44)
1
H⊥ (σ ) = π̃â π̃b̂ ηâb̂ + T 2 L1 â L1 b̂ ηâ b̂ = 0, (4.45)
2T
H (σ ) = π̃â L1 â = 0, (4.46)
δLWZ β
Fν (σ ) = ζν − π̃â Lν â − β
Lν = 0. (4.47)
δL0
Half of the above fermionic constraints generate κ-symmetry as shown in Section 3.2 for
the action (3.28).
Super-invariant (up to local Lorentz) combinations are Lâ1 and

π̃â = (e−1 )â m p̃m + ζ̃µ Ξm θ µ
T
= −√ −L0 b̂ G11 + L1 b̂ G01 ηb̂â ≡ (e−1 )â m π̃m
−G
δLWZ δLWZ β
p̃m = pm − β
Lm β , ζ̃µ = ζµ − β
Lµ .
δL0 δL0
The equations of motion for a superstring in the pp-wave background in the conformal
gauge, huv = ηuv , are obtained as
1 m
ẋ m = π̃ + (e−1 )â m Lµ â (τ3 θ )µ ,
T

ṗm = T Lm â L1â

1 n l T â −1 n b̂ −1 l
+ (∂m gnl ) π̃ π̃ − L1 (e )â L1 (e )b̂ − π̃ n Lµ â (e−1 )â l (τ3 θ )µ
2T 2
1 µ
− ζ (∂m Ξn )θ π̃ n − T x n (∂m Ξn )θ Lµ â L1â
T

δLWZ
− ∂m Lµ α (τ3 θ )µ , (4.48)
δL0 α

µ 1 m −1 m ν
θ̇ = −(τ3 θ ) + (Ξm θ )
µ µ
π̃ + (e )â Lν (τ3 θ ) ,
â
T

ζ̇µ = −(ζ τ3 )µ − T Lµ â L1â − π̃â Lµ â τ3

1 â ν
− (ζ Ξâ )µ π̃ + Lν (τ3 θ )
â
T

m ν ∂
left
+ T x (Ξm )ν µ Lν â − x (Ξm θ )ν + θ
m â
Lν L1â
∂θ µ
left
∂ left δLWZ α ν ∂
+ µ Lν (τ3 θ ) + π̃â Lν (τ3 θ )ν ,
â
(4.49)
∂θ δL0 α ∂θ µ
which are background-covariant.
In the components by using (4.4) and (4.24) the Hamiltonian in the conformal gauge is
rewritten as
H = H1 + H2 + H3 ,

1
H1 = dσ 2p+ π̃− + 4µ2 x 2 p+
2
+ π̃ 2 + p+ Lµ + (τ3 θ )µ
2T

ν
+ π̃− + 2µ2 x 2 p+ Lν − τ3 (θ − ) + π̃î Lµ î (τ3 θ )µ ,
&
T
H2 = dσ 2 (x + ) − 2µ2 x 2 (x − )
2
µ ν
+ θ + (x − ) Ξ− θ + x î Ξî θ − Lµ + (x − ) + (θ − ) Lν −

µ 2 ' δLWZ α
+ x î + θ + (x − ) Ξ− θ + x î Ξî θ − Lµ î + Lµ (τ 3 θ µ
) ,
δL0 α

H3 = dσ [−ζ τ3 θ ]. (4.50)
( (
For a flat case H1 and H2 reduce simply into (1/2T )p2 and (T /2)x 2 , respectively.
Equations of motion are written as
1 ν
ẋ + =π̃− + 4µ2 x 2 p+ + Lµ + (τ3 θ )µ + 2µ2 x 2 Lν − τ3 (θ − ) ,
T
1 ν 1
ẋ = p+ + Lν − τ3 (θ − ) ,
−
ẋ î = π̃î + Lµ î (τ3 θ )µ , (4.51)
T T
& '
ṗ− = T −2µ2 x 2 L1 − + (Ξ− θ )µ Lµ + L1 − + Lµ î L1 î

∂ δLWZ &
−
µ µ
'
+ dσ 2Lµ
α
# + Ξ − θ ẋ + Ξ θ ẋ î
− L µ
α
(τ 3 θ )
∂x − (σ ) δL0 α î
T
ν sin Ψ− sin Ψ−
ṗ+ = T (x − ) + (θ − ) Lν − + iT *Πθ − CΓ+ τ1 Π τ3 (θ − )
Ψ− Ψ−

ṗî = +T L1 î + Ξî θ − Lµ + L1 − + Lµ î L1 î
µ
%
1 −

− ν − −
− 4µ x p+
2 î
p+ + Lν τ3 (θ ) − (x ) L1
T

∂ δLWZ
− T x − (Ξ− Ξi θ − )µ Lµ + L1 − + Lµ î L1 î − dσ α
Lµ α (τ3 θ )µ
∂x î (σ ) δL0
T %
iT sin Ψ+ sin Ψ+ ˆ
− Ξî θ − CΓ− τ1 Π Ξ− θ ẋ − + Ξjˆ θ − ẋ j
µ Ψ+ Ψ+
T %
iT sin Ψ+ − sin Ψ+ ˆ
− (x ) Ξ− Ξî θ − CΓ− τ1 Π Ξ− θ ẋ − + Ξjˆ θ − ẋ j ,
µ Ψ+ Ψ+
(4.52)

+ λ

+ λ λ 1 −

− ν
(θ̇ ) = − τ3 (θ ) + (Ξ− θ ) p+ + Lν τ3 (θ )
T

1
+ (Ξî θ − )λ π̃î + Lµ î (τ3 θ )µ ,
T
θ̇ − = −τ3 (θ − ) ,

(ζ̇+ )λ = − (ζ+ ) τ3 λ

iT + sin Ψ+ 1 −

− µ
− ζ+ − L1 Γ− τ1 Π ν
(Ξ− #+ ) λ p+ + Lµ τ3 (θ )
µ Ψ+ ν T

µ
+ T − Lµ + L1 − + Lµ î L1 î (#+ )µ λ + (x − ) Ξ− + Lµ+ L1− + Lµ î L1 î
µ
− θ + (x − ) Ξ− θ + x î Ξî θ −
left left %
∂ + − ∂ jˆ jˆ
× L ν L 1 + Lν L 1
∂(θ + )λ ∂(θ + )λ

+ p+ Lν + (#+ )ν λ τ3 + π̃î Lν î (#+ )ν λ τ3
left left
∂ + ν ∂ jˆ
+ Lν (τ3 θ ) p+ + Lν (τ3 θ )ν π̃î
∂(θ + )λ ∂(θ + )λ

iT ∂ left +
+ dσ 1 Γ− τ1 Π Lµ + (τ3 θ )µ (σ )
L
2µ ∂(θ + )λ (σ ) +
%
iT sin Ψ+ T sin Ψ+ − − jˆ

+ CΓ− τ1 Π Ξ− θ ẋ + Ξjˆ θ ẋ
µ Ψ+ Ψ+
T %
iT sin Ψ+ − sin Ψ+ − − jˆ

− (x ) Ξ− CΓ− τ1 Π Ξ− θ ẋ + Ξjˆ θ ẋ , (4.53)
µ Ψ+ Ψ+
and second-order form equations become
✷x + (σ ) = (2µ)2 x˙2 ẋ − + θ ± -dependent terms,

ν ν
✷x − (σ ) = ∂τ Lν − τ3 (θ − ) − ∂σ Lν − (θ − )
T %
sin Ψ− − sin Ψ− −
+i *Πθ CΓ+ τ1 Π τ3 (θ ) ,
Ψ− Ψ−
2
✷x(σ ) = (2µ)2 x (x − ) − (ẋ − )2 + θ − -dependent terms. (4.54)
The right-hand sides of the first and third equations are complicated functions of θ − or θ −
and θ + .
5. Conclusions and discussions
In this paper an alternative form of the gauge invariant action for the superstring in
the RR pp-wave background is proposed. It is explicit since the Wess–Zumino term is
bilinear with respect to the LI currents of the centrally extended super-pp-wave algebra. It
is obtained by the Penrose limit from the superstring action in the AdS background with the
bilinear WZ term [17]. The Penrose limit of the WZ term is given essentially as follows:
1. Rescale the coordinates in the LI 1-forms with a parameter Ω with suitable weights,
for example,
m
L−
AdS z → Ω z = L−
Nm m m 2 −
(0) (z ) + Ω L(2) (z ) + o(Ω ).
m 4
(5.1)
2. Rescale the WZ term with the same weight of the Nambu–Goto term as
LWZ,AdS → Ω −2 LWZ,AdS . (5.2)

The term which would have negative power of Ω is − Γ − Πτ1 L−
L and becomes
AdS AdS
Ω −2
L− − −
AdS Γ Πτ1 LAdS

= Ω −2 L − Γ − Πτ1 L− + Ω 2 2L
− Γ − Πτ1 L− + o(Ω 4 ) .
(0) (0) (0) (2)
3. Subtract the divergent term in Ω → 0 limit which is proportional to 1/Ω 2 and closed.
4. Take the Ω → 0 limit in the bilinear WZ term
− Γ − Πτ1 L− + L+ -dependent term.
LWZ,AdS → LWZ,pp = 2L(0) (2)
This procedure corresponds to make a nondegenerate super-pp-wave group by introduc-

ing the fermionic center associating to L− −
(2) = L . This is contrasted with the conventional
WZ term case:
1. Rescale coordinates in the LI 1-forms with a parameter Ω with suitable weights and
take the leading terms in the limit Ω → 0
m
L−AdS z → Ω z → L−
Nm m m
(0) (z ),

L+AdS z → Ω
m
z → Ω 2 L+
Nm m m
(2) (z ). (5.3)
2. Construct the WZ term in the conventional form given by an integral of the three-
form [8]

d σ LWZ,AdS → d 2 σ LWZ,pp
2

− +
= d 3σ L L − ±
(0) / (2) τ3 L(0) + L -dependent terms .
In this case the next to leading term L−(2) is not necessary to be kept in taking the Penrose
limit of the WZ term. However, the resultant WZ term has the integral expression and does
not allow us simple treatment of the Hamiltonian and equations of motion in covariant
gauges.
The Hamiltonians for the bosonic particle, superparticle, string and superstring in
the RR pp-wave background are obtained in the conformal gauge. The particle and the
superparticle Hamiltonians are identified with the quadratic Casimir operators of the pp-
wave and the super-pp-wave algebras respectively. Once the superparticle Hamiltonian
(4.18) is recognized as the super-pp-invariant “mass operator” of the superstring theory, all
states in a supergravity multiplet are “massless” and the supersymmetry is manifest.
The world-sheet reparametrization generators and the local fermionic constraints are
also obtained. The combinations of the reparametrization constraints (4.45) and (4.46) as
H⊥ ± H and the first class part of FνI , where I = 1, 2 correspond to right/left (±) modes,
will make a closed set of constraint algebra, namely ABCD constraint system [22]. It
was shown that the local constraints of the AdS superstring satisfy the ABCD algebra
[19] as well as of the flat superstring. Since the super-pp-wave algebra is obtained by the
Penrose limit from the super-AdS algebra [5] as well as the flat algebra by the flat limit,
the local symmetry algebra is also expected to be obtained by the same limiting procedure
preserving the same structure of the ABCD algebra. The background independence of the
local symmetries is plausible.
Equations of motion in the conformal gauge are obtained and are background-covariant.
The equations of motion for x ± are obtained and will be important for taking into
account interactions. Quantization of the superstring theory in the conformal gauge may
require a suitable change of variables such as to the GL(4|4) matrix variable as was
done for the AdS5 × S5 case [16]. The covariant approach will be useful to examine
symmetry structures toward the covariant superstring field theory, S-T-U dualities which
are deeply related to the background symmetry and non-perturbative properties such as
BPS conditions.
Appendix A. Cartan 1-forms for AdS5 × S5
The Cartan 1-forms in the AdS5 × S5 space are presented [17,18]. The left-invariant
Cartan one-forms of a coset SU(2, 2|4)/[SO(4, 1) × SO(5)] G = G(x, θ ) = exP eθQ are
defined by
1 1
G−1 dG = La Pa + La Pa + Lab Jab + La b Ja b + Lαα I Qαα I .
2 2
They are given by

sin(Ψ/2) 2
La = ea + iθ CC γ a Dθ,
Ψ/2
2
a a a sin(Ψ/2)
L = e − θ CC γ Dθ,
Ψ/2

sin(Ψ/2) 2
Lab = ωab − θ CC γ ab * Dθ,
Ψ/2

a b a b a b sin(Ψ/2) 2
L =ω + θ CC γ * Dθ,
Ψ/2
sin Ψ
Lα = Dθ,
Ψ

1 sinh(x/2) 2 [a b]
ω =
ab
dx x ,
2 x/2

1 sin(x /2) 2 [a b ]
ωa b = − dx x , (A.1)
2 x /2
where [ab] = ab − ba and charge conjugation matrix for AdS5 space and S5 space are C
and C , respectively, and

i a a
1 ab a b

Dθ = d − * γ ea + iγ ea + γ ωab + γ ωa b θ,
2 4
I αα I
θ CC γa ββ J − *γ a θ θ CC γa ββ J
αα
(Ψ 2 )αα I ββ J = *γ a θ
1 ab αα I 1 αα I
− γ θ θ CC γab * ββ J + γ a b θ θ CC γa b * ββ J .
2 2
(A.2)
After the Penrose limit using (2.4) and (3.9) they are written as (3.29). The relation between
the AdS variables and the Penrose variables are described in [5].
References
[1] M. Blau, J. Figueroa-O’Farrill, C. Hull, G. Papadopoulos, JHEP 0201 (2002) 047, hep-th/0110242.
[2] J. Kowalski-Glikman, Phys. Lett. B 134 (1984) 194;
C.M. Hull, Phys. Lett. B 139 (1984) 39;
P.T. Chrusciel, J. Kowalski-Glikman, Phys. Lett. B 149 (1984) 107;
J. Figueroa-O’Farrill, G. Papadopoulos, JHEP 0108 (2001) 036, hep-th/0105308.
[3] R. Penrose, Any space-time has a plane wave as a limit, in: M. Cahen, M. Flato (Eds.), Differential Geometry
and Relativity, Reidel, Dordrecht, 1976, p. 271.
[4] M. Blau, J. Figueroa-O’Farrill, C. Hull, G. Papadopoulos, Class. Quantum Grav. 19 (2002) L87, hep-
th/0201081;
R. Güven, Phys. Lett. B 482 (2000) 255;
R. Güven, Phys. Lett. B 637 (1–3) (2002) 168, hep-th/0005061;
M. Blau, J. Figueroa-O’Farrill, G. Papadopoulos, Penrose limit, supergravity and brane dynamics, hep-
th/0202111.
[5] M. Hatsuda, K. Kamimura, M. Sakaguchi, Nucl. Phys. B 632 (2002) 114, hep-th/0202190.
[6] M. Hatsuda, K. Kamimura, M. Sakaguchi, Nucl. Phys. B 637 (2002) 168, hep-th/0204002.
[7] D. Berenstein, J. Maldacena, H. Nastase, JHEP 0204 (2002) 013, hep-th/0202021.
[8] R.R. Metsaev, Nucl. Phys. B 625 (2002) 70, hep-th/0112044.
[9] A. Dabholkar, S. Parvizi, Dp branes in pp-wave background, Nucl. Phys. B 641 (2002) 223, hep-th/0203231;
A. Kumar, R.R. Nayak, Sanjay, D-brane solutions in pp-wave background, Phys. Lett. B 541 (2002) 183,
hep-th/0204025;
M. Alishahiha, A. Kumar, D-brane solutions from new isometries of pp-waves, Phys. Lett. B 542 (2002)
130, hep-th/0205134;
S.S. Pal, Solution to worldvolume action of D3 brane in pp-wave background, Mod. Phys. Lett. A 17 (2002)
1735, hep-th/0205303;
S. Hyun, H. Shin, Branes from matrix theory in pp-wave background, Phys. Lett. B 543 (2002) 115, hep-
th/0206090;
D. Bak, Supersymmetric branes in pp wave background, hep-th/0204033;
K. Skenderis, M. Taklor, JHEP 0206 (2002) 025, hep-th/0204054;
H. Singh, M5-branes with 3/8 supersymmetry in pp-wave background, hep-th/0205020;
P. Bain, P. Meessen, M. Zamaklar, Supergravity solutions for D-branes in Hpp-wave backgrounds, hep-
th/0205106;
M. Alishahiha, A. Kumar, D-brane solutions from new isometries of pp-waves, hep-th/0205134;
O. Bergman, M.R. Gaberdiel, M.B. Green, D-brane interactions in type IIB plane-wave background, hep-
th/0205183;
Y. Hikida, Y. Sugawara, JHEP 0206 (2002) 037, hep-th/0205200;
S.S. Pal, Solution to worldvolume action of D3 brane in pp-wave background, hep-th/0205303;

K. Sugiyama, K. Yoshida, Supermembrane on the pp-wave background, hep-th/0206070;
K. Sugiyama, K. Yoshida, BPS conditions of supermembrane on the pp wave, hep-th/0206132;
S. Hyun, H. Shin, Branes from matrix theory in pp-wave background, hep-th/0206090;
Y. Michishita, D-branes in NSNS and RR pp-wave backgrounds and S-duality, hep-th/0206131;
N. Kim, J. Plefka, On the spectrum of pp wave matrix theory, hep-th/0207034;
K. Dasgupta, M.M. Sheikh-Jabbari, M. Van Raamsdonk, Protected multiplets of M-theory on a plane wave,
hep-th/0207050;
N. Kim, J.-H. Park, Superalgebra for M-theory on a pp-wave, hep-th/0207061.
[10] T. Takayanagi, Modular invariance of strings on pp-waves with RR-flux, hep-th/0206010.
[11] R.R. Metsaev, A.A. Tseytlin, Phys. Rev. D 65 (2002) 126004, hep-th/0202109;
J.G. Russo, A.A. Tseytlin, JHEP 0204 (2002) 021, hep-th/0202179.
[12] N. Berkovits, JHEP 0204 (2002) 037, hep-th/0203248.
[13] M.B. Green, J.H. Schwarz, Phys. Lett. B 136 (1984) 367;
M.B. Green, J.H. Schwarz, Nucl. Phys. B 243 (1994) 285.
[14] N. Berkovits, M. Bershadsky, T. Hauer, S. Zhukov, B. Zwiebach, Nucl. Phys. B 567 (2000) 61, hep-
th/9907200.
[15] M. Hatsuda, K. Kamimura, M. Sakaguchi, Phys. Rev. D 62 (2000) 105024, hep-th/0007009.
[16] R. Roiban, W. Siegel, JHEP 0011 (2000) 024, hep-th/0010104.
[17] M. Hatsuda, M. Sakaguchi, WZ term for AdS superstring, Phys. Rev. D 66 (2002) 0450220, hep-th/0205092;
M. Hatsuda, M. Sakaguchi, WZ term for the AdS superstring and generalized Inönü–Wigner contraction,
hep-th/0106114.
[18] R.R. Metsaev, A.A. Tseytlin, Nucl. Phys. B 533 (1998) 109, hep-th/9805028.
[19] M. Hatsuda, K. Kamimura, Nucl. Phys. B 611 (2001) 77, hep-th/0106202.
[20] M. Green, Phys. Lett. B 223 (1989) 157;
W. Siegel, Phys. Rev. D 50 (1994) 2799, hep-th/9403144.
[21] E. Inönü, E.P. Wigner, Proc. Nat. Acad. Sci. USA 39 (1953) 510;
E. Inönü, E.P. Wigner, in: F. Gursey (Ed.), Group Theoretical Concepts and Methods in Elementary Particle
Physics, Gordon and Breach, New York, 1964.
[22] W. Siegel, Nucl. Phys. B 263 (1985) 93;
W. Siegel, Introduction to String Field Theory, World Scientific, Singapore, 1988;
F. Essler, E. Laenen, W. Siegel, J.P. Yamron, Phys. Lett. B 254 (1991) 411;
F. Essler, M. Hatsuda, E. Laenen, W. Siegel, J.P. Yamron, Nucl. Phys. B 364 (1991) 67;
W. Siegel, Nucl. Phys. B 263 (1985) 93.
M-theory pp-waves, Penrose limits and

supernumerary supersymmetries
M. Cvetič a,1 , H. Lü b,2 , C.N. Pope c,3
a Department of Physics and Astronomy, University of Pennsylvania, Philadelphia, PA 19104, USA
b Michigan Center for Theoretical Physics, University of Michigan, Ann Arbor, MI 48109, USA
c Center for Theoretical Physics, Texas A&M University, College Station, TX 77843, USA
Received 1 July 2002; received in revised form 26 August 2002; accepted 29 August 2002
Abstract
We study supersymmetric pp-waves in M-theory, their dimensional reduction to D0-branes or
pp-waves in type IIA, and their T-dualisation to solutions in the type IIB theory. The general class of
pp-waves that we consider encompass the Penrose limits of AdSp × S q with (p, q) = (4, 7), (7,4),
(3,3), (3,2), (2,3), (2,2), but includes also many other examples that can again lead to exactly-solvable
massive strings, but which do not arise from Penrose limits. All the pp-waves in D = 11 have 16
“standard” Killing spinors, but in certain cases one finds additional, or “supernumerary,” Killing
spinors too. These give rise to linearly-realised supersymmetries in the string or matrix models.
A focus of our investigation is on the circumstances when the Killing spinors are independent of
particular coordinates (x + or transverse-space coordinates), since these will survive at the field-
theory level in dimensional reduction or T-dualisation.
1. Introduction
The Penrose limit [1] of the AdS5 × S 5 solution of type IIB theory is a pp-wave with
maximal supersymmetry [2,3]. This result is of considerable interest within the framework
of the AdS/CFT correspondence, since the pp-wave provides a background for which
the string theory action in light-cone gauge describes a massive free string, which is
E-mail address: honglu@umich.edu (H. Lü).

1 Research supported in part by DOE grant DE-FG02-95ER40893 and NATO grant 976951.
2 Research supported in full by DOE grant DE-FG02-95ER40899.
3 Research supported in part by DOE grant DE-FG03-95ER40917.
PII: S 0 5 5 0 - 3 2 1 3 ( 0 2 ) 0 0 7 9 2 - 7
66 M. Cvetič et al. / Nuclear Physics B 644 (2002) 65–84
exactly solvable [4,5], thus allowing explicit comparisons with results in the dual gauge
theory [5]. The Penrose limit of the AdS4 × S 7 and AdS7 × S 4 solutions of M-theory also
gives rise to a maximally-supersymmetric pp-wave, obtained in [6], and this provides a
simple background for the DLCQ description of M-theory, and the corresponding matrix-
model in this regime [5]. Subsequent papers have explored a variety of consequences and
generalisations of these observations [7–21].
In a previous paper [17], we studied a wider class of supersymmetric pp-waves in the
type IIB theory, generalising the maximally-supersymmetric one that arises as the Penrose
limit of AdS5 × S 5 . In particular, we allowed for a more general structure of the constant
self-dual five-form field strength; these structures were motivated by the flat (orbifold)
limit of a special holonomy transverse space for the pp waves. In fact any pp-wave within
the general class automatically has 16 Killing spinors, which we, therefore, denoted as
“standard” Killing spinors. In special cases one finds that there can be additional Killing
spinors, which we denoted as “supernumerary” Killing spinors. The maximum number,
16, of these is achieved for the Penrose limit of AdS5 × S 5 . (In this class we also found
another example of the Penrose limit of AdS3 × S 3 arising form an D3/D3-intersection.)
The focus of our study in [17] was to determine the circumstances under which one
obtains supernumerary Killing spinors in the type IIB pp-waves. These are important when
one considers the exactly-solvable string models in the pp-wave backgrounds; we found
that it is the supernumerary Killing spinors that are in one to one correspondence with
the associated linearly-realised worldsheet supersymmetries of the corresponding string
action. In fact the string theory is solved by going to the light-cone gauge, with the x +
coordinate in the pp-wave being set equal to the world-sheet time coordinate. In order that
the linearly-realised world-sheet supersymmetries be unbroken, it is necessary therefore
that the associated supernumerary Killing spinors be independent of the coordinate x + ,
which is indeed the case in all the type IIB pp-waves. For instance, all 16 supernumerary
Killing spinors in the Penrose limit of AdS5 × S 5 have this property [5,17].
A further significance of having Killing spinors in the type IIB pp-waves that are
independent of x + is that after performing a T-duality transformation on the x + coordinate
(which is always a Killing direction), the resulting type IIA solution will also be
supersymmetric. It can be lifted to M-theory, where it acquires an interpretation as a
supersymmetric deformed M2-brane, i.e., an M2-brane in which an additional 4-form
flux is turned on in the transverse space [22–28]. An intriguing feature of the deformed
M2-branes obtained by this T-dualisation procedure is that if any of the Killing spinors
originate from supernumerary Killing spinors (which are x + -independent), then in the
M-theory picture they solve the Killing-spinor equations despite violating the criterion that
is usually applied [22,25,29] for testing whether a supersymmetry survives when the extra
4-form flux is turned on [17].
In this paper, we study supersymmetric pp-waves in M-theory. In particular, we allow
for rather general structures for the constant 4-form field strength of M-theory, motivated
from the flat (orbifold) limit of special holonomy transverse space for the pp-waves.
These possible structures fall into two classes. Focusing on the nature of supersymmetry,
we again find that there are always 16 “standard” Killing spinors, and that additional
“supernumerary” Killing spinors can arise in special cases. Unlike the case of pp-waves
in type IIB, however, it is no longer automatic that supernumerary Killing spinors are
M. Cvetič et al. / Nuclear Physics B 644 (2002) 65–84 67
independent of x + . The Penrose limit of AdS4 × S 7 or AdS7 × S 4 provides the unique

example where the number of supernumerary Killing spinors attains its maximum, namely
16 [6]. Unlike the maximally-supersymmetric pp-wave in type IIB, however, here in the
M-theory maximally-supersymmetric pp-wave the 16 supernumerary Killing spinors all
depend on x + .
Again our focus is on the occurrence of supernumerary Killing spinors, and also on
determining the dependence of all the Killing spinors on the x + coordinate and the
9 transverse coordinates zi . These dependences are of importance when one considers
reductions to type IIA, and subsequent T-dualisation to type IIB, since they determine
whether there will be supersymmetries in the type IIA and IIB supergravity solutions.
Depending upon whether one reduces from D = 11 on x + or on one of the transverse
coordinates zi , one either obtains a D0-brane or a pp-wave in the type IIA theory. In the
case of a pp-wave, the string theory in this background is again exactly-solvable by going
to the light-cone gauge, giving rise to a free massive theory. In the case of a reduction
instead on x + , the D0-brane world-particle action leads to a DLCQ description of the
M-theory matrix model in this sector. Thus these backgrounds of M-theory provide for
dual descriptions, either in terms of a solvable type IIA string action or in a matrix theory
model. In particular, the supernumerary supersymmetry plays a key role in determining the
supersymmetry of the string action as well as supersymmetry of the matrix model.
We begin in Section 2 by setting up our formalism for the pp-waves in M-theory, and
obtaining the criterion for the existence of Killing spinors. We then study the coordinate
dependences of the Killing spinors, for the various choices of constant 4-form fluxes in the
flat nine-dimensional transverse space that we are considering. This allows us to discuss the
supersymmetry of the type IIA D0-branes and pp-waves that we can obtain by dimensional
reduction. We also show how some of our pp-wave solutions are related to Penrose limits
of M2-brane and M5-brane intersections.
In Section 3 we derive the light-cone action for type IIA strings in arbitrary bosonic
backgrounds, making use of earlier covariant results for the Green–Schwarz action up
to quadratic order in fermions that were obtained in [30]. We also obtain an analogous
result for the light-cone action for the type IIB string, again valid for arbitrary bosonic
backgrounds, and based on covariant results obtained in [30]. Using these light-cone
actions, we study the properties of some of the pp-wave solutions that we obtain by
dimensional reduction and T-duality.
In certain cases the M-theory pp-waves have no isometry direction within the nine-
dimensional transverse space, and so they cannot be reduced to give pp-waves in ten
dimensions. Under these circumstances, where the pp-waves are intrinsically eleven-
dimensional, one is led to considering a DLCQ description in this background, leading to a
matrix model. In particular, we derive the matrix-model action, whose supersymmetry is in
one to one correspondence with the supernumerary supersymmetries of the corresponding
pp-wave background.
2. Supersymmetry of pp-waves in M-theory
2.1. General formalism
We shall consider pp-wave solutions of D = 11 supergravity, where the metric and

4-form are given by
= −4 dx + dx − + H dx + + dzi2 ,
2 2
ds11 (1)
+
F(4) = µ dx ∧ Φ(3) , (2)
where Φ(3) is a harmonic 3-form in the flat nine-dimensional transverse space whose
metric is dzi2 , µ is a constant, and we are taking H here to depend only on zi . In the
vielbein basis e+ = dx + , e− = −2 dx − + 12 H dx + , ei = dzi , for which the metric is
2 = 2e + e − + e i e i , the spin connection is given by
ds11
1
ω+i = ∂i H e+ , ω−i = ω+− = ωij = 0, (3)
2
and the only non-vanishing Riemann tensor components, in the vielbein basis, are
1
R+i+j = − ∂i ∂j H. (4)
2
This implies that the only non-vanishing Ricci-tensor component is R++ = − 12 H . The
D = 11 supergravity equations are therefore satisfied if H obeys the equation
1
H = − µ2 |Φ(3) |2 . (5)
6
In this paper we shall focus on the cases where Φ(3) is a covariantly-constant 3-form. It is
sufficient for our purposes to take the solution for H to be
Q 2 2
H = c0 + − µi zi , (6)
r7
i
where c0 , Q and µi are constants, and r 2 ≡ zi zi . It follows from (5) that the µi are subject
to the condition
1
µ2i = µ2 |Φ(3) |2 . (7)
12
i
When µ = 0 (and, hence, µi = 0), the solution becomes a standard pp-wave in D = 11,
whose dimensional reduction gives a D0-brane in the type IIA theory, and the pp-wave
charge Q becomes the charge of the D0-brane.
The supercovariant derivative appearing in the supersymmetry transformation rule
δψM = DM is given by
1 N1 ···N4
DM = ∇M − ΓM FN1 ···N4 − 8FMN1 ···N3 Γ N1 ···N3 . (8)
288
Defining DM = ∇M + ΩM , we therefore find

1
∇+ = ∂+ + ∂i H Γ− Γi , ∇− = ∂− , ∇i = ∂i ,
4
i i
Ω+ = − µ(1 + Γ− Γ+ )W, Ω− = 0, Ωi = µΓ− (Γi W + 3W Γi ),
12 24
(9)
where we have defined
i
W ≡ Φij k Γij k . (10)
6
It follows immediately from (9) that Killing spinors , satisfying DM = 0, are
independent of x − . Since Ωi Ωj = 0 we have ∂i ∂j = 0 and hence it follows that

= 1 − zi Ωi χ, (11)
where χ depends only on x + . Finally, from D+ = 0 one deduces that
i
∂+ χ − µ(Γ− Γ+ + 1)W χ = 0 (12)
12
and

µ2 zi Γi W 2 + 9W 2 Γi + 6W Γi W + 72∂i H Γi Γ− χ = 0. (13)
Thus (13) determines the number of Killing spinors, while (11) and (12) determine their zi
and x + dependence.
Since Γ+ Γ− + Γ− Γ+ = 2 and Γ+ 2 = Γ− 2 = 0, we have a unique decomposition
χ = χ+ + χ− for any χ , where χ+ ≡ 12 Γ+ Γ− χ and χ− ≡ 12 Γ− Γ+ χ have the defining
properties Γ+ χ+ = 0 and Γ− χ− = 0.
It is evident from (13) that there will always be 16 Killing spinors corresponding to
χ = χ− . Accordingly, we refer to these as “standard Killing spinors”, since they exist for
any choice of the function H that satisfies the field equation (5). In particular, the pp-wave
charge Q can be non-zero. In certain cases there can also be additional Killing spinors
corresponding to χ = χ+ . We refer to these as “supernumerary Killing spinors”. We shall
construct the two categories of Killing spinors, and discuss their coordinate dependences,
in the following two subsections.
Before discussing the two categories of Killing spinors, let us be a little more specific in
our choice of constant 3-forms Φ(3) . It turns out to be natural to restrict attention to cases
where the associated matrix W , defined by (10), is a sum of individual terms Wα that all
commute with each other. This can be seen to lead to two inequivalent maximal sets of
terms, which we shall refer to as Case 1 and Case 2. For the two cases we have
Case 1:
Φ(3) = m1 dz129 + m2 dz349 + m3 dz569 + m4 dz789. (14)
Case 2:
Φ(3) = m1 dz123 + m2 dz145 + m3 dz167 + m4 dz246 + m5 dz257 + m6 dz347
+ m7 dz356 , (15)
where we have defined dzij k ≡ dzi ∧ dzj ∧ dzk . It should be noted that unless all four of
the mα coefficients are non-zero in Case 1, it is in fact encompassed (after a relabelling of
coordinates) within Case 2.
It is straightforward to verify that if we construct W as in (10), and write it as

W= mα Wα , (16)
α
where Wα denotes the individual Γij k structures (for example, W1 = iΓ129 is one of the
four structures in Case 1), then we shall have [Wα , Wβ ] = 0. In consequence, for either
Case 1 or Case 2 we can choose a basis for the gamma matrices in which the Wα are all
diagonal. It is useful to have in mind such a diagonal choice of basis in the subsequent
discussion.
For our canonical choices, one can see that if the mα are taken equal then Φ(3) in Case 1
can be expressed asmdz9 ∧ J , where J is the Kähler form for the eight-dimensional flat
space with metric 8i=1 dzi2 . Likewise, if the mα are set equal in Case 2, Φ(3) can be
expressed as mΨ(3) , where Ψ(3) is a G2 -invariant associative 3-form in the flat seven-

dimensional space with metric 7i=1 dzi2 .
2.2. The 16 standard Killing spinors
The 16 standard Killing spinors correspond to taking χ = χ− , i.e., they are defined
by Γ− χ = 0. It is evident from (9) and (11) that they are all independent of all of the zi
coordinates. It is also evident from (12) that they will have x + dependence given by
i +W
χ = e 4 µx χ0 , (17)
where χ0 is any constant spinor satisfying Γ− χ0 = 0. If W annihilates any of these spinors,
then the associated “standard” Killing spinor will be independent of x + (and so, in fact, it
will be independent of all the coordinates). The discussion now divides into two, according
to whether we take Φ(3) to be given by (14) or (15):
Case 1:
For Φ(3) given by (14), the eigenvalues of W are
λi = ±m1 ± m2 ± m3 ± m4 , (18)
where the ± choices are all independent. Each eigenvalue occurs twice, making the 32 in
total. In the subspace of eigenspinors annihilated by Γ− one gets each eigenvalue once,
and likewise in the subspace annihilated by Γ+ . For the standard Killing spinors arising in
Case 1, it therefore follows that the possible numbers that are independent of x + can be
Nstan = 0, 2, 4 or 8.
The number Nstan = 0 is achieved for generic choices of the mα ; Nstan = 2 is achieved
for choices where m1 + m2 + m3 + m4 = 0; Nstan = 4 is achieved for choices with m4 = 0
and m1 + m2 + m3 = 0 (or permutations); and Nstan = 8 is achieved for choices with
m3 = m4 = 0 and m1 + m2 = 0 (or permutations).
Case 2:
When Φ(3) is given by (15), we find that again W generically has sixteen different
eigenvalues, each occurring twice. One copy of the sixteen again occurs in each of the Γ−
and Γ+ subspaces. The eigenvalues are given by ±λi , where
λ8 = m1 + m2 + m3 − m4 + m5 + m6 + m7 , (19)
and λi for 1 i 7 is given by reversing the sign of each mα that occurs as a coefficient of
any term containing the gamma matrix Γi . The numbers of standard Killing spinors that are
independent of x + that can be achieved for these Case 2 examples are therefore Nstan = 2n,
where 0 n 6 is the number of λi = 0 that are arranged to vanish by choosing the mα
appropriately. Thus for Case 2 we can have Nstan = 0, 2, 4, 6, 8, 10 or 12 standard Killing
spinors that are independent of x + .
2.3. Supernumerary Killing spinors
We now turn to the discussion of supernumerary Killing spinors, for which Γ+ χ = 0.

In a generic pp-wave solution there will be none of these, but they can arise in special
cases when H given in (6) is quadratic in zi (i.e., the pp-wave charge Q = 0), and the
distribution of µi coefficients (which must in any case satisfy (7)) is chosen appropriately.
The numbers of supernumerary Killing spinors that can be achieved depends also on the
choice of Φ(3) .
Eq. (13) that determines the number of Killing spinors admits solutions for supernumer-
ary Killing spinors (Γ+ χ = 0) if H in (6) depends on zi only quadratically, and
2
µ Γi W 2 + 9W 2 Γi + 6W Γi W − 144µ2i Γi χ = 0, (20)
where Γ+ χ = 0.
As we discussed when we made our choices (14) or (15) for Φ(3) , the resulting terms
Wα defined by (10) and (16) can all be simultaneously diagonalised (for each of Case 1
or Case 2 separately), by means of an appropriate similarity transformation of the gamma
matrices. It is convenient to assume that such a basis for the gamma matrices has been
chosen.
With respect to a diagonal basis, it is evident that
Xi ≡ Γi W Γi + 3W (21)
is also diagonal, and therefore that (20) can be rewritten as
2 2
µ Xi − 144µ2i Γi χ = 0, for each i. (22)
From (11), it now follows that the solutions of (22) will give the supernumerary Killing
spinors, with zi dependence given by

i
= 1 − µi zi Γ− Γi χ. (23)
2
In particular, this means that a supernumerary Killing spinor is independent of a given
coordinate zi if and only if the associated coefficient µi in (6) is zero.
The discussion of the supernumerary Killing spinors now divides into the two
possibilities for Φ(3) , given by (14) or (15).
Case 1:
In this case Φ(3) is given by (14). In the direction i = 9 we have X9 = 4W , and so
µ29 = 19 µ2 λ2 , where λ is one of the eigenvalues of W given in (18). Without loss of
generality, since the other eigenvalues differ only in sign permutations of the mα , we can
take
1
µ29 = µ2 (m1 + m2 + m3 + m4 )2 . (24)
9
The remaining µi for 1 i 8 are then given by
1 2
µ21 = µ22 = µ (−2m1 + m2 + m3 + m4 )2 ,
36
1
µ23 = µ24 = µ2 (m1 − 2m2 + m3 + m4 )2 ,
36
1
µ25 = µ26 = µ2 (m1 + m2 − 2m3 + m4 )2 ,
36
1
µ27 = µ28 = µ2 (m1 + m2 + m3 − 2m4 )2 . (25)
36
To see this, we note that if (X9 − κ9 )Γ9 χ = 0 then 4W χ = κ9 χ . We are taking κ9 =
4(m1 + m2 + m3 + m4 ). Substituting into (X1 − κ1 )Γ1 χ = 0 we therefore find 14 κ9 χ +
3W χ − κ1 χ = 0, where W = Γ1 W Γ1 . From (14) and (10) it follows that the diagonal
matrix W has eigenvalues that are just those of W but with m2 , m3 and m4 reversed in sign,
and so W χ = (m1 − m2 − m3 − m4 )χ . Thus we deduce that κ1 = 2(2m1 − m2 − m3 − m4 ).
Applying an analogous argument for each direction i, we arrive at (25).
For a generic choice of the constants mα , there are precisely two supernumerary Killing
spinors. This is because a given bosonic solution has fixed values for the coefficients µα ,
and so there are two solutions to (20) since there is a twofold degeneracy in (25). In special
cases, where the mα are chosen so that two or more of the expressions in (25) are equal,
there can therefore be more solutions of (20). It is an elementary exercise to enumerate
all the possible numbers of supernumerary supersymmetries that can be achieved for
specific choices of mα . As in the case of type IIB pp-wave solutions, the supernumerary
supersymmetries can lead to a variety of “non-standard” fractions of total supersymmetry
that exceed 1/2 [17].
It is worth remarking that for Case 1, as a consequence of the equation X9 = 4W , it
follows that supernumerary Killing spinors are independent of x + if and only if they are
independent of z9 .
Case 2:
When the 3-form Φ(3) is given by (15), it follows that we shall have X8 = X9 = 2W ,
and so from (22) we shall have µ28 = µ29 = 36 1 2 2
µ λ , where λ is one of the eigenvalues
of W . These are now given by λ8 in (19), together with λi for 1 i 7 as described below
(19). Without loss of generality, since the µα have not yet been specified, we may choose
1 2 2 1
µ28 = µ29 = µ λ8 = µ2 (m1 + m2 + m3 − m4 + m5 + m6 + m7 )2 . (26)
36 36
For the other constants µi , we shall, therefore, have

1 2
µ2i = µ (λ8 − 3λi )2 , 1 i 7, (27)
144
where λ8 and λi are given by (19) and below. For example, we shall have µ21 =
36 µ (2m1 + 2m2 + 2m3 + m4 − m5 − m6 − m7 ) .
1 2 2
2.4. Supersymmetry of type IIA pp-waves and D0-branes
Having obtained the M-theory pp-waves, we can dimensionally reduce the solutions
to D = 10, giving rise to D0-branes if we reduce on the x + coordinate, or to type IIA
pp-waves if we reduce instead on any of the zi coordinates. Of course a reduction on a
particular zi coordinate is possible only if it is a Killing direction, which means that the
associated coefficient µi in the metric function H must vanish.4
First let us consider a reduction on x + . This is a Killing direction for all pp-wave
solutions. However, some or all of the Killing spinors in a given solution may be dependent
on the x + coordinate, in which case they will not survive in the reduction of the solution
to type IIA supergravity. As we saw from (12), the criterion for a Killing spinor to be
independent of x + is that it should be annihilated by W . For the standard Killing spinors,
the fraction of the 16 Killing spinors that will survive the reduction on x + depends on
the detailed structure of W . The 16 standard Killing spinors exist for any solution for H ,
subject to (5). In particular, they exist when the D0-brane charge Q is turned on. The
supernumerary Killing spinors, on the other hand, are all eigenvectors of W with the
same eigenvalue, and hence they are all x + -independent if W χ = 0, but x + -dependent if
W χ = 0. The supernumerary Killing spinors exist only if Q = 0 and the µi are distributed
appropriately.
For a reduction on one of the transverse coordinates zi , the corresponding constant µi in
the expression for H in the metric (1) must vanish, in order that ∂/∂zi be a Killing vector.
As we saw in (23), the Killing spinors are then also independent of the coordinate zi , and,
hence, they will all survive in the reduction. In Section 2.2, it was observed that the 16
standard Killing spinors are all independent of zi .
When Φ(3) is contained within the Case 1 in (14), the direction z9 is singled out. It
was observed in the discussion of Case 1 in Section 2.3 that if µ9 = 0 then we have also
W χ = 0, and so this implies that if z9 is a Killing direction then the supernumerary Killing
spinors will not only be independent of z9 , but also of x + .
In Case 2, where Φ(3) is given by (15), the directions z8 and z9 are singled out. If we
arrange for µ8 = µ9 = 0 (they are always equal), then as shown in Section 2.3 we also
have W χ = 0, and so the supernumerary Killing spinors will be independent of x + as well
as the reduction coordinate z8 or z9 .
4 There are, of course, many other Killing vectors in the pp-wave metric, which could be used for Kaluza–
Klein reduction, but we are not considering these here (see, however, Section 2.5). Some examples are discussed
in [20]; these typically give rise to an extra (constant) flux in the lower dimension, coming from a non-vanishing
Kaluza–Klein vector potential.
In fact these various reductions to type IIA can be related to the general type IIB
pp-waves obtained in [17], by means of T-duality. If we dimensionally reduce on z9 in
Case 1, or on z8 or z9 in Case 2, which is possible if parameters are chosen so that the
corresponding µi coefficient vanishes, the resulting supernumerary Killing spinors are
all independent x + . This implies that the type IIA string action then has linearly-realised
supersymmetries.
We can also obtain large classes of type IIA pp-wave solutions in which other µi
parameters are instead zero. (That is to say, µi other than µ9 in Case 1, or µ8 = µ9 in
Case 2.) In these circumstances, there can exist supernumerary Killing spinors that are
dependent on x + . One would then obtain a type IIA solution where some, or all, of the
world-sheet supersymmetries were non-linearly realised.
2.5. An explicit example
Here we consider an explicit example where only m1 and m2 are non-vanishing. (This
can equally well be for either Case 1 in (14) or Case 2 in (15), since the two are then
equivalent after coordinate relabellings.) Taking m1 and m2 as given, we can then arrange
for two choices for the µi coefficients that will give rise to supernumerary Killing spinors.
These are summarised in the following table.
The second choice is nothing but the first, with one of the two mi reversed in sign.
However, if the mα are given, fixed parameters, then these two choices for the µα
correspond to two independent solutions.5
For generic but fixed values of mα , each choice in Table 1 gives rise to 8 supernumerary
Killing spinors. When m1 = m2 in the second choice, the 8 supernumerary Killing spinors
are all independent of x + . This then corresponds to the Penrose limit of the M2/M5-brane
system (AdS3 × S 3 × T 4 ). By contrast, in the first choice in Table 1 the 8 supernumerary
Killing spinors depend on the x + coordinate. For both choices, 8 of the 16 standard Killing
spinors are independent of x + . Another special case arises when either m1 = ±2m2 or
m2 = ±2m1 , which implies that µ1 = µ2 = 0 or µ3 = µ4 = 0. Interestingly enough, this
case is T-dual to the maximally supersymmetric pp-wave arising from the Penrose limit
of AdS5 × S 5 . Suppose, for example, we have the choice giving µ1 = µ2 = 0. We can
Table 1
Choices for µi with fixed m1 and m2 that give supernumerary Killing spinors
µ21 = µ22 µ23 = µ24 µ25 = µ26 µ27 = µ28 µ29
1 1 1 1 1
36 (−2m1 + m2 ) 36 (m1 − 2m2 ) 36 (m1 + m2 ) 36 (m1 + m2 ) 9 (m1 + m2 )
2 2 2 2 2
1 1 1 1 1
36 (2m1 + m2 ) 36 (m1 + 2m2 ) 36 (m1 − m2 ) 36 (m1 − m2 ) 9 (m1 − m2 )
2 2 2 2 2
5 In our previous discussion, we adopted an “active” viewpoint when discussing the possible occurrences of
supernumerary Killing spinors, rather than the “passive” viewpoint we are adopting here. Namely, we previously
took a fixed choice for how the eigenvalues were to be expressed in terms of the mα , and then covered the
spectrum of possibilities by allowing the mα to be chosen freely. The two viewpoints are clearly equivalent, if
appropriate care is taken.
then reduce on z1 and T-dualise on z2 . The resulting type IIB solution is the maximally-
supersymmetric pp-wave, in a slightly non-standard coordinate system that was introduced
in [20] to make certain Killing directions in the transverse space manifest. The reverse
procedure of T-dualisation and lifting was performed in [20], to give the pp-wave in
D = 11.
2.6. AdS Penrose limits
Various AdS spacetimes can arise in M-theory as near-horizon limits of M-brane

intersections. The M2-brane and M5-brane themselves have near-horizon limits AdS4 × S 7
and AdS7 × S 4 , respectively; these both have the same Penrose limit. The resulting pp-wave
is maximally supersymmetric, and is the one obtained in [6], with
1 1
H = c0 − µ2 z12 + z22 + z32 − µ2 z42 + · · · + z92 ,
9 36
Φ(3) = µ dz123. (28)
(The constant c0 was not included in the Penrose limit taken in [6], but it can easily be
included, as was shown in [17] for any Penrose limit.) There is no isometry along any of
the zi directions. All the Killing spinors depend on x + , and so a dimensional reduction
on the x + coordinate will result in a type IIA supergravity solution that breaks all the
supersymmetry.
The Penrose limits of intersecting M-branes that give rise to AdS structure were
discussed in [8]. An M2/M5 brane intersection gives rise to AdS3 × S 3 × T 4 in the near-
horizon limit. The Penrose limit is then given by
1
H = c0 − µ2 z12 + z22 + z32 + z42 ,
4

Φ(3) = µ dz129 + dz349 . (29)
There are in total 24 Killing spinors, of which 16 are the standard Killing spinors, together
with 8 supernumerary Killing spinors. Out of the 16 standard Killing spinors, 8 are
independent of x + . All 8 supernumerary Killing spinors are independent of x + , giving
a total of 16 Killing spinors independent of x + . Thus the D0-brane after reduction to type
IIA will have 16 Killing spinors, but reduced to 8 if the D0-brane charge Q is turned on
(since then the supernumerary Killing spinors will already be lost in D = 11). The solution
has also isometries along the five zi coordinates with i = 5, 6, 7, 8, 9. All 24 Killing spinors
are independent of these coordinate, and so they will all survive if the M-theory solution
is instead reduced on any of these five zi coordinates, giving a pp-wave in type IIA. In
particular, if we reduce the solution on the z9 coordinate the pp-wave will have a constant
(NS–NS) 3-form F(3) , and in fact this pp-wave is itself the Penrose limit of the NS1/NS5
system in type IIA. If instead we reduce the D = 11 solution on z5 , z6 , z7 or z8 , the ten-
dimensional pp-wave is supported by a constant RR 4-form F(4) .
M2/M2/M2 and M5/M5/M5 brane intersections give rise to AdS2 × S 3 and AdS3 × S 2
respectively. Both have the same Penrose limit, which can be written as
1
H = c0 − µ2 z72 + z82 − µ2 z92 ,
4

Φ(3) = µ dz129 + dz349 + dz569 . (30)
In this case there are the 16 standard Killing spinors plus 4 supernumerary Killing spinors,
giving a total of 20 in all. All the Killing spinors depend on x + , and, hence, after reduction
on x + to type IIA the resulting D0-brane will have no supersymmetry. There are also
isometries in the zi coordinates with i = 1, . . . , 6. Since none of the Killing spinors
depends on any of these coordinates, the type IIA pp-wave that results from reducing
instead on one of these will have all 20 Killing spinors.
The M2/M2/M5/M5 brane intersection system gives an AdS2 × S 2 in its near-horizon
limit. The Penrose limit is given by

H = c0 − µ2 z12 + z62 ,

Φ(3) = µ dz123 + dz145 + dz246 + dz356 . (31)
In this case, 4 out of the 16 standard Killing spinors are independent of x + . Additionally,
there are 4 supernumerary Killing spinors, which are all x + -independent.
3. Matrix model and type II string actions
3.1. Type IIA string action
Whenever there is an isometry in any of the zi directions, the D = 11 pp-wave can be

dimensionally reduced on such a coordinate, to give a pp-wave in the type IIA theory. In
general, the resulting pp-wave can have non-vanishing constant backgrounds both for the
NS–NS 3-form and RR 4-form. The precise details depend in the usual way on the structure
of Φ(3) in D = 11.
The Green–Schwarz action for the type IIA string in an arbitrary bosonic background
was derived, in component form up to and including second-order in the fermionic
coordinates θ , in [30] (the form of the RR couplings had previously been obtained
schematically in [31]):
1√ 1
L2 = − −h hij ∂i Xµ ∂j Xν gµν + ij ∂i Xµ ∂j Xν Aµν
2 2
i
− iθ̄ β Γµ Dj θ ∂i X + ∂i X ∂j Xν θ̄ β ij Γ11 Γµ ρσ θ Fνρσ
ij µ µ
8

i 1
− ∂i Xµ ∂j Xν eφ θ̄ β ij Γ11 Γµ Γ ρσ Γν Fρσ + Γµ Γ ρσ λτ Γν Fρσ λτ θ, (32)
16 12
where
√ 1
β ij ≡ −h hij − ij Γ11 , Di θ ≡ ∂i θ + ∂i Xµ ωµ mn Γmn θ. (33)
4
The field strengths are given by
F(4) = dA(3) − A(1) ∧ dA(2), F(3) = dA(2) , F(2) = dA(1). (34)

1
The fermionic coordinates θ are non-chiral, and can be written as θ = θ 2 , where
1 θ
Γ11 θ = θ 2 . In a notation adapted to the passage to light-cone gauge, we can introduce
−θ
world-sheet Dirac matrices 9i , with 90 = −iτ2 , 91 = τ1 and 92 = τ3 , where τi are the
Pauli matrices. The 9i act on the upper and lower 16 components θ 1 and θ 2 of the column
1
vector Ψ = θ 2 , and 92 is the chirality operator. The conjugate spinor in this notation
θ
is then Ψ̄ = Ψ † Γ0 90 , and we therefore have the “dictionary” θ̄ Oθ → −Ψ̄ 90 OΨ and
θ̄ Γ11 Oθ → Ψ̄ 91 OΨ , where O is any matrix or operator√constructed from the Γi matrices.
In the light-cone gauge, where X+ = τ , Γ− θ = 0 and −h hij = ηij , the fermionic part
of the type IIA Green–Schwarz action (32), therefore, becomes
i i
LF = iΨ̄ Γ+DΨ
/ − Ψ̄ 91 Γ+F/ (3) Ψ − eφ Ψ̄ Γ+ (91F/ (2) − 90F/ (4) )Ψ, (35)
4 4
1 1
/ (2) ≡ Γ i F+i ,
F / (3) ≡ Γ ij F+ij ,
F / (4) ≡ Γ ij k F+ij k .
F (36)
2 6
In the pp-wave backgrounds we are considering here, the world-sheet Dirac operator D / just
reduces to ∂/ in the light-cone gauge.
If the dimensional reduction from the D = 11 pp-wave to D = 10 is performed on a
Killing direction zi whose differential dzi does not appear in the expression for Φ(3) in
D = 11, then the solution in D = 10 is a pp-wave with only the RR 4-form F(4) as a
source. This situation can be achieved for any Φ(3) contained within Case 2, provided that
one reduces on the z8 or z9 coordinate. Since Case 1 is encompassed by Case 2 (after
appropriate coordinate relabellings) in all situations except where all four mα are non-
vanishing, it is only in this last circumstance that one is forced into a dimensional reduction
in which the differential dzi of the reduction coordinate is present in Φ(3) in D = 11. Of
course in other cases too, one may choose to perform the dimensional reduction on such a
coordinate zi , provided that the associated coefficient µi in the quadratic metric function
H vanishes, implying that ∂/∂zi is a Killing vector.
Let us first consider the case where the differential dzi of the reduction coordinate zi
does not appear in Φ(3) in D = 11. It then follows from (32) and (35) that after choosing
the light-cone gauge, the associated type IIA string action will be6
8

1 1 1 1
L= żi2 − zi 2 − µ2i zi2 + Ψ̄ i/
∂ + µ90 W Γ+ Ψ. (37)
2 2 2 4
i=1
Thus the masses of the fermions will be given by the eigenvalues of W .

Suppose now that we instead perform a reduction on a coordinate zi whose differential
dz does appear in Φ(4) in D = 11. We now find that F(4) in D = 11 reduces as
i
F(4) −→ F4 + dz ∧ F(3) (38)
6 See also [32] for a related discussion of the type IIA Green–Schwarz action in a gravitational wave
background.
in D = 10, where z is the reduction coordinate. We now have the non-vanishing NS–NS
field F(3) in the type IIA background, together, possibly, with a non-vanishing RR 4-form
F(4) . If the reduction of the 3-form Φ(3) is written as Φ(3) → Φ3 + dz ∧ Φ2 , and if we
define
i
Y ≡ Φij Γ ij , (39)
2
then it follows from (32) and (35) that after choosing the light-cone gauge, we shall obtain
the string action
8
1 2 1 2 1 1 1
L= żi − zi + µzi Bi − µ2i zi2 + Ψ̄ i/∂ + µ90 W + µ90 Y Γ+ Ψ.
2 2 2 4 4
i=1
(40)
Here, Bi denotes the components of the 1-form B(1) whose exterior derivative gives
Φ(2) = dB(1) . Since Φ(2) = 12 Φij dzi ∧ dzj where Φij are constants, we may, therefore,
take Bi to be given by
1
Bi = Φj i zj . (41)
2
Thus the string action in this case is given by
8 2
1 2 1 1 1 2 2 1
L= ż − z − µΦij zj − µi zi + µ2 Φik Φj k zj zk
2 i 2 i 2 2 8
i=1

1 1
+ Ψ̄ i/ ∂ + µ90 W + µ90 Y Γ+ Ψ. (42)
4 4
Thus the boson masses, as well as the fermion masses, are modified by the presence of the
NS–NS 3-form field. This is a generalisation of a result obtained in [5].
As an example, let us consider the pp-wave in D = 11 resulting from taking Φ(3) to be
given by Case 1, as in (14). After dimensional reduction on the coordinate z9 , which will
be a Killing direction provided that
m1 + m2 + m3 + m4 = 0 (43)
(see (24)), the type IIA light-cone action will be given by
8 2
1 2 1 1 1 2 2
L= ż − z − µΦij zj − µi zi
2 i 2 i 2 2
i=1

1 2 2 2 1
+ µ m1 z1 + z2 + · · · + m4 z7 + z8 + Ψ̄ i/
2 2 2 2
∂ + µ90 Y Γ+ Ψ, (44)
8 4
with Y given by
Y = im1 Γ12 + im2 Γ34 + im3 Γ56 + im4 Γ78 . (45)
(The matrix W is absent here, since all terms in Φ(3) in D = 11 involved a factor dz9 . Thus
the D = 10 background is purely NS–NS in this example.)
Note that the choice of gauge for writing B(1) ≡ Bi dzi is not unique. In this example
we could, for instance, choose, instead of writing it in the “symmetrical” gauge
1 1 1
B(1) = Φij zi dzj = m1 (z1 dz2 − z2 dz1 ) + · · · + m4 (z7 dz8 − z8 dz7 ), (46)
2 2 2
to write it in the “asymmetrical” form
B(1) = m1 z1 dz2 + m2 z3 dz4 + m3 z5 dz6 + m4 z7 dz8 . (47)
In this choice of gauge we would instead obtain the string action
8 2
1 2 1 1 1
L= żi − zi − µΦij zj − µ2i zi2
2 2 2 2
i=1
1 1
+ µ2 m21 z12 + m22 z32 + m23 z52 + m24 z72 + Ψ̄ i/
∂ + µ90 Γ+ Ψ. (48)
2 4
Of course the different gauge choices just change the action by a total derivative, and so
they are equivalent in the closed string sector.
3.2. Type IIB string action
Many of our examples can be T-dualised to pp-waves in type IIB theory when there are
two µi that vanish. In some cases, when the type IIA pp-wave is supported only by the
F(3) , the solution is also valid in type IIB theory supported by the NS–NS F(3) or the RR
F(3) , or both using S-duality rotation. Thus in this section, we consider the light-cone type
IIB string action in such a background.
In [30], the Green–Schwarz action for the type IIB string in an arbitrary bosonic
background was derived, giving all terms up to and including quadratic order in the
fermionic coordinates. In the notation of [30], the two Majorana–Weyl fermions were
1
denoted by θ 1 and θ 2 . If we put these in a column vector θ ≡ θ 2 , and define world-sheet
θ
Dirac matrices 9i by 90 = −iτ2 , 91 = τ1 and 92 = τ3 , where τi are the Pauli matrices, then
we find the following “dictionary” for converting the notation in [30] to the one we wish to
use here. For any matrix or operator O constructed from the target-space Dirac matrices,
we shall have
θ̄ 90 Oθ = −θ̄ 1 Oθ 1 − θ̄ 2 Oθ 2 , θ̄ 91 Oθ = θ̄ 1 Oθ 1 − θ̄ 2 Oθ 2 ,
θ̄ Oθ = 2θ̄ [1 Oθ 2] , θ̄92 Oθ = −2θ̄ (1Oθ 2) , (49)
where the conjugate of θ is defined by θ̄ = θ † Γ0 90 = (−θ̄ 2 , θ̄ 1 ). Substituting into
Eq. (3.29) of [30] (in the updated v2. where a minor typographical error has been corrected
and conventions adjusted), the type IIB Green–Schwarz action up to O(θ 2 ) is given by
1√ 1
L=− −h hij ∂i Xµ ∂j Xν gµν + ij ∂i Xµ ∂j Xν Bµν
2 2
i
+ i∂i Xµ θ̄ γ ij 90 Γµ Dj θ − ∂i Xµ ∂j Xν θ̄ γ ij 91 Γµ ρσ θ Gνρσ
8
i
+ eφ ∂i Xµ ∂j Xν θ̄ γ ij Γµ
8

1 1 ρ1 ···ρ5
× Γ ρ ∂ρ χ + 92 Γ ρ1 ρ2 ρ3 Fρ1 ρ2 ρ3 + Γ Fρ1 ···ρ5 Γν θ, (50)
6 240
where G(3) = dB(2) is the NS–NS 3-form, φ and χ are the dilaton and axion, and F(3) and
F(5) are the RR 3-form and self-dual 5-form, and we have defined
√
γ ij ≡ −h hij − ij 92 . (51)
+
√
In the light-cone gauge, X = τ , θ = Ψ with Γ− Ψ = 0, and −h h = η , we, ij ij
therefore, find that the fermionic part of the action becomes

i i 1
LF = iΨ̄ Γ+DΨ / + Ψ̄ 91 Γ+G / 3 Ψ − eφ Ψ̄ Γ+ ∂+ χ + 92F/ (3) + F/ (5) Ψ, (52)
4 4 2
1 1 1
/ (3) ≡ Γ ij G+ij ,
G / (3) ≡ Γ ij F+ij ,
F / (5) ≡ Γ ij k? F+ij k? .
F (53)
2 2 24
/ is just ∂/ .
Note that for all the pp-waves, it follows from (3) that D
To apply this action to our examples, let us first consider Case 2, with µ8 = µ9 = 0.
This example can be T-dualised to type IIB, where it gives rise to pp-waves of the sort
described in [17]. In these cases the pp-wave is supported only by the self-dual RR 5-form.
The action was obtained in [17].
For another example consider Case 1, with all four of the mα non-vanishing, but chosen
so that µ9 = 0, thus permitting a reduction in the z9 direction to give a type IIA solution.
The solution is then supported purely by the NS–NS 3-form. There is in general no further
isometry direction among the remaining zi coordinates that could allow us to perform a
T-duality transformation. However, since only the NS–NS 3-form field is involved, we can
clearly take the identical configuration and view it as a solution instead of the type IIB
theory, again supported by the NS–NS 3-form. Having done so, we can then choose
to perform an S-duality transformation, thereby introducing a non-vanishing RR 3-form
as well (or instead). It is then straightforward to see from (52) and (53), together with
imposing the light-cone gauge in the bosonic part of (50), that these type IIB pp-wave
backgrounds will have a light-cone string action given by
8 2
1 2 1 1 1
L= żi − zi − Bij zj − µ2i zi2
2 2 2 2
i=1

i i
+ Ψ̄ i/ ∂ + 91G / (3) − 92F/ (3) Γ+ Ψ, (54)
4 4
/ (3) and F/ (3) are the NS–NS and RR 3-form contributions, respectively.
where G
3.3. Matrix model action
There are many examples in our general discussion where all the coefficients µi in the
metric function H are non-vanishing, implying that the pp-wave is intrinsically eleven-
dimensional. In these cases, the system is best described by a D0-brane action. Namely,
one can perform a DLCQ compactification [33–36] along the light-cone coordinate
x − ≡ x − + 2πR, and consider the sector with momentum 2p+ = −p− = N/R. The
dynamics of this sector is then described by a U (N) matrix model with the strength of
interactions governed by g ∼ 2R. The procedure as it applies to the case of the Penrose
limit of AdS4 × S 7 or AdS7 × S 4 was given in [5]. The form of the action for the general,
constant W , as studied in this paper, can be derived along the same lines, and is structurally
of the same form. This is due to the fact that the 4-form field strength enters the D0-
brane particle action in the light-cone gauge (i.e., the U (N) matrix model) only through
W = 6i Φij k Γ ij k . The form of the action is thus given by
9
i 2 2 i 2 9

L= Ẋ − µ2i Xi + Ψ T Ψ̇ + µΨ T W Ψ − µg Tr Xi Xj Xk Φij k
4 3
i=1 i,j,k=1
2
+ 2g 2 Tr Xi , Xj + 2ig Tr Ψ T Γ i Ψ, Xi . (55)
Note that in addition to the standard matrix-model interactions there are also the fermionic
and bosonic mass terms, and additionally the term tri-linear in Xi that is related to the
Myers effect [37].
The supersymmetry of this quantum mechanical matrix model fixes the coefficients
in front of the fermionic mass terms and the interaction terms in the same way as
it was derived for the special case of W = Φ123 Γ 123 in [5]. Indeed, the existence of
supersymmetry is dictated by the existence of the supernumerary Killing spinors. In fact,
the supersymmetry transformation parameter is exactly the supernumerary Killing spinor:
δXi = Ψ Γ i ,

1 1 i j
δΨ = Ẋ Γi + µX − W Γi + Γi W + ig X , X Γij ,
i i
4 12
= eµW t 0 . (56)
The case where W = Γ123 was given in [5]. In that case, the system is fully supersym-
metric, and hence 0 is an arbitrary constant spinor. Furthermore, since W has no zero
eigenvalues in that example, all the supersymmetry parameters are time-dependent.
For the more general W ’s that we have considered in this paper, 0 is subject to further
projection constraints, in accordance with the supernumerary Killing spinors. In our more
general cases W can annihilate 0 , implying that is then time-independent. In such an
example, the pp-wave can also be reduced to give rise to a pp-wave in type IIA, thus giving
an exactly-solvable string action. The existence of two routes, one corresponding to the
matrix model of the D0-particle action, and the other corresponding to the free massive
Type IIA string action, therefore suggests that these are dual descriptions of the theory
when the background is of this particular type.
4. Conclusions
In this paper, we studied a general class of supersymmetric pp-waves in M-theory, by

turning on constant 3-forms, motivated by the orbifold (flat) limit of a special holonomy
transverse space for the pp-wave. These 3-forms fall into two classes, one motivated by
the Kähler form of the eight-dimensional special holonomy transverse space, and the
other motivated by the associative 3-form of a seven-dimensional transverse space of G2
holonomy.
This general class of pp-waves encompass the Penrose limits of AdSp × S q with
(p, q) = (4, 7), (7, 4), (3, 3), (3, 2), (2, 3), (2, 2) which are associated with the near hori-
zon limits of the M2-brane, M5-brane, and M2/M5, M5/M5/M5, M2/M2/M2 and
M2/M2/M5/M5 intersections, respectively. In addition this general class contains many
additional examples of pp-waves that are do not correspond to any known Penrose limit.
We focused on the study of the target space supersymmetry. In addition to 16 “standard”
Killing spinors that always arise, we determined the conditions under which additional
“supernumerary” Killing spinors appear. We also analysed the conditions under which the
Killing spinors are independent of the light cone x + coordinate, or of one or more of the
nine transverse coordinates. These conditions determine whether the reduction of the M-
theory pp-waves to type IIA supergravity, and subsequent T-dualisation to type IIB, remain
supersymmetric.
Since x + is always a Killing direction the M-theory pp-wave can always be reduced
on this coordinate, leading to a D0-brane configuration of the type IIA theory. Its
world-particle action corresponds to a DLCQ description of a matrix-theory action for
M-theory, with unbroken supersymmetry governed by x + -independent supernumerary
supersymmetries of the M-theory pp-wave background.
On the other hand the independence of a Killing spinor on a transverse coordinate allows
for a reduction on this coordinate down to a supersymmetric type IIA pp-wave. The light
cone string actions in these backgrounds correspond to exactly-solvable free massive string
theories, and again the supernumerary supersymmetries play a key role in determining the
supersymmetry of the string action.
Note added
We have updated the discussion of the type IIA and type IIB string actions in this version
of the paper, taking into account the corrections and improvements in v2. of [30]. There was
only one minor typographical in the type IIB action in the earlier version of [30], but there
were various inelegant notations and conventions, all of which have now been changed, and
these changes are incorporated in this version of the present paper. A detailed discussion
of the changes is given in the Addendum section in v2. of [30]. We are grateful to Kelly
Stelle for discussions leading to these improvements.
Acknowledgements
We are grateful to Gary Gibbons, Jim Liu and Justin Vázquez-Poritz for conversations.
H.L. and C.N.P. are grateful to University of Pennsylvania for hospitality and financial
support during the course of this work.
References
[1] R. Penrose, Any spacetime has a plane wave as a limit, in: Differential Geometry and Relativity, Reidel,
Dordrecht, 1976.
[2] M. Blau, J. Figueroa-O’Farrill, C. Hull, G. Papadopoulos, A new maximally supersymmetric background of
IIB superstring theory, JHEP 0201 (2002) 047, hep-th/0110242.
[3] M. Blau, J. Figueroa-O’Farrill, C. Hull, G. Papadopoulos, Penrose limits and maximal supersymmetry, hep-
th/0201081.
[4] R.R. Metsaev, Type IIB Green–Schwarz superstring in plane wave Ramond–Ramond background, Nucl.
Phys. B 625 (2002) 70, hep-th/0112044.
[5] D. Berenstein, J. Maldacena, H. Nastase, Strings in flat space and pp waves from N = 4 super-Yang–Mills,
hep-th/0202021.
[6] J. Kowalski-Glikman, Vacuum states in supersymmetric Kaluza–Klein theory, Phys. Lett. B 134 (1984) 194.
[7] R.R. Metsaev, A.A. Tseytlin, Exactly solvable model of superstring in plane wave Ramond–Ramond
background, hep-th/0202109.
[8] M. Blau, J. Figueroa-O’Farrill, G. Papadopoulos, Penrose limits, supergravity and brane dynamics, hep-
th/0202111.
[9] N. Itzhaki, I.R. Klebanov, S. Mukhi, PP-wave limit and enhanced supersymmetry in gauge theories, hep-
th/0202153.
[10] J. Gomis, H. Ooguri, Penrose limit of N = 1 gauge theories, hep-th/0202157.
[11] J.G. Russo, A.A. Tseytlin, On solvable models of type IIB superstring in NS–NS and RR plane wave
backgrounds, hep-th/0202179.
[12] L.A. Pando-Zayas, J. Sonnenschein, On Penrose limits and gauge theories, hep-th/0202186.
[13] M. Alishahiha, M.M. Sheikh-Jabbari, The pp-wave limits of orbifolded AdS5 × S 5 , hep-th/0203018.
[14] M. Billo’, I. Pesando, Boundary states for GS superstrings in an Hpp wave background, hep-th/0203028.
[15] N. Kim, A. Pankiewicz, S.-J. Rey, S. Theisen, Superstring on pp-wave orbifold from large-N quiver gauge
theory, hep-th/0203080.
[16] T. Takayanagi, S. Terashima, Strings on orbifolded pp-waves, hep-th/0203093.
[17] M. Cvetič, H. Lü, C.N. Pope, Penrose limits, pp-waves and deformed M2-branes, hep-th/0203082.
[18] U. Gursoy, C. Nunez, M. Schvellinger, RG flows from Spin(7), CY 4-fold and HK manifolds to AdS,
Penrose limits and pp waves, hep-th/0203124.
[19] E. Floratos, A. Kehagias, Penrose limits of orbifolds and orientifolds, hep-th/0203134.
[20] J. Michelson, (Twisted) toroidal compactification of pp-waves, hep-th/0203140.
[21] C.S. Chu, P.M. Ho, Noncommutative D-brane and open string in pp-wave background with B-field, hep-
th/0203186.
[22] S.W. Hawking, M.M. Taylor-Robinson, Bulk charges in eleven dimensions, Phys. Rev. D 58 (1998) 025006,
hep-th/9711042.
[23] M.J. Duff, J.M. Evans, R.R. Khuri, J.X. Lu, R. Minasian, The octonionic membrane, Phys. Lett. B 412
(1997) 281, hep-th/9706124.
[24] M. Cvetič, H. Lü, C.N. Pope, Brane resolution through transgression, Nucl. Phys. B 600 (2001) 103, hep-
th/0011023.
[25] K. Becker, A note on compactifications on Spin(7)-holonomy manifolds, JHEP 0105 (2001) 003, hep-
th/0011114.
[26] M. Cvetič, G.W. Gibbons, H. Lü, C.N. Pope, Ricci-flat metrics, harmonic forms and brane resolutions, hep-
th/0012011.
[27] M. Cvetič, G.W. Gibbons, H. Lü, C.N. Pope, Hyper-Kähler Calabi metrics, L2 harmonic forms, resolved
M2-branes, and AdS4 /CFT 3 correspondence, Nucl. Phys. B 617 (2001) 151, hep-th/0102185.
[28] M. Cvetič, G.W. Gibbons, H. Lü, C.N. Pope, New complete non-compact Spin(7) manifolds, Nucl. Phys.
B 620 (2002) 29, hep-th/0103155.
[29] K. Becker, M. Becker, M-theory on eight-manifolds, Nucl. Phys. B 477 (1996) 155, hep-th/9605053.
[30] M. Cvetič, H. Lü, C.N. Pope, K.S. Stelle, T-duality in the Green–Schwarz formalism, and the mass-
less/massive IIA duality map, Nucl. Phys. B 573 (2000) 149, hep-th/9907202.
[31] A.A. Tseytlin, On dilaton dependence of type II superstring action, Class. Quantum Grav. 13 (1996) L 81,
hep-th/9601109.
[32] S. Hyun, H. Shin, Supersymmetry of Green–Schwarz superstring and matrix string theory, Phys. Rev. D 64
(2001) 046008, hep-th/0012247.
[33] T. Banks, W. Fischler, S.H. Shenker, L. Susskind, M theory as a matrix model: a conjecture, Phys. Rev. D 55
(1997) 5112, hep-th/9610043.
[34] L. Susskind, Another conjecture about M(atrix) theory, hep-th/9704080.
[35] A. Sen, D0 branes on T n and matrix theory, Adv. Theor. Math. Phys. 2 (1998) 51, hep-th/9709220.
[36] N. Seiberg, Why is the matrix model correct?, Phys. Rev. Lett. 79 (1997) 3577, hep-th/9710009.
[37] R.C. Myers, Dielectric-branes, JHEP 9912 (1999) 022, hep-th/9910053.
Spectra of supersymmetric Yang–Mills quantum

mechanics
J. Wosiek
M. Smoluchowski Institute of Physics, Jagellonian University Reymonta 4, 30-059 Kraków, Poland
Received 3 April 2002; accepted 5 September 2002
Abstract
The new method of solving quantum mechanical problems is proposed. The finite, i.e., cut-off,
Hilbert space is algebraically implemented in the computer code with states represented by lists of
variable length. Complete numerical solution of a given system is then automatically obtained. The
technique is applied to Wess–Zumino quantum mechanics and D = 2 and D = 4 supersymmetric
Yang–Mills quantum mechanics with SU(2) gauge group. Convergence with increasing cut-off was
observed in many cases well within the reach of present machines. Many old results were confirmed
and some new ones, especially for the D = 4 system, are derived. Extension to D = 10 is possible
but computationally demanding for higher gauge groups.
PACS: 11.10.Kk; 04.60.Kz
Keywords: M-theory; Matrix model; Quantum mechanics; Non-Abelian
1. Introduction
Since the original conjecture of Banks, Fischler, Shenker and Susskind of the
equivalence between M-theory and the D = 10 supersymmetric Yang–Mills quantum
mechanics (SYMQM) [1], a lot of effort has been put to understand, and ultimately solve,
the latter [4–17].
The general, not necessarily gauge, supersymmetric quantum mechanical systems have
much longer history [2,3]. Claudson and Halpern considered for the first time the gauge
systems also well before the BFSS hypothesis [4]. In particular a complete solution of the
D = 2, N = 2 SYMQM was given there (see also Ref. [5]). Later the specific gauge models
E-mail address: wosiek@th.if.uj.edu.pl (J. Wosiek).
PII: S 0 5 5 0 - 3 2 1 3 ( 0 2 ) 0 0 8 1 0 - 6
86 J. Wosiek / Nuclear Physics B 644 (2002) 85–112
were studied in more detail and some candidates for the ground state were constructed in
the, gauge invariant, Born–Oppenheimer approximation [6,7].
Another important development was achieved by de Wit, Lüscher and Nicolai who have
shown that the spectrum of supersymmetric Yang–Mills quantum mechanics is continuous
[8,9] due to the cancellations between fermions and bosons. This is different from the
pure Yang–Mills case where the transverse fluctuations across the vacuum valleys do not
cancel and effectively block the valley, resulting in the discrete spectrum of the 0-volume
glueballs [10]. The continuous spectrum was first regarded as a setback, however it has
turned out into a virtue with the advent of the BFSS hypothesis and new interpretation
in terms of the scattering states. Still this new connection requires existence of the
localized state at the threshold of the spectrum—the supergraviton [11]. This question
triggered intense studies of the Witten index of SYMQM for various D and different
gauge groups [12–15]. Powerful techniques were developed to calculate analytically
non-Abelian integrals [16–18] related to the Witten index. They were accompanied by
complementary and original numerical methods [19,20]. The emerging picture is rather
satisfactory, indeed: for D < 10 there is no threshold bound state while for D = 10 there
is one, exactly as required by BFSS. This has been proven for N = 2, but the evidence is
being accumulated that it holds for higher N , as well as for other gauge groups.
The large N limit of SYMQM, which is relevant to M-theory, was studied in the
framework of the mean field approximation in Refs. [21,22]. This provided an interesting
realization of the black hole thermodynamics predicted by the M-theory [11,23].
In spite of all above results, SYMQM remains unsolved for D > 2. Therefore, it seems
natural to study this model with lattice methods which proved so successful in treating
more complex field theoretical systems. Such a programme has been proposed in Ref. [24],
beginning with the yet simpler (quenched, D = 4, N = 2) case which was later extended
to higher gauge groups 2 < N < 9 [25]. The asymptotic behavior in N was observed,
and indications of an onset of the interesting phase structure was found in agreement with
Refs. [21,22]. Including dynamical fermions is possible for D = 4, and may be feasible at
D = 10 for the first few N . However, an arbitrary N case is plagued by the sign problem
caused by the generally complex fermionic determinant.
Another interesting approach studied by many authors, follows the Eguchi–Kawai
trick to trade entirely the D-dimensional configuration space into a group space [26,27].
Proceeding in this way one is led to consider the fully reduced (to a single point in the
Euclidean space) model which nevertheless possesses a version of the supersymmetry [28].
Many analytical and numerical results were obtained in this way (for a review see, e.g.,
[29]). Numerical simulations were pushed to quite high N in simplified models [30].
However the sign problem which is also present there limits the Monte Carlo approach.
This may be alleviated by the new method proposed recently in Ref. [31].
It is important to remember that the subject has a lot of overlap with the small volume
study of gauge theories where the valuable expertise has been accumulating for a long time
[10,32–34]. Although the final goal there is to increase the space volume and to match
ultimately the standard large volume physics, the starting point is the pure Yang–Mills
quantum mechanics identical to the bosonic part of our supersymmetric systems. In fact
the classical results of Lüscher and Münster are the special case of the solution of the
D = 4 SYMQM in the gluino-free sector of the latter. This will be shown in Section 5.
J. Wosiek / Nuclear Physics B 644 (2002) 85–112 87
Recently van Baal has used the full machinery of the small volume approach to study
the supersymmetric vacuum state of the more complicated, compact version of D = 4
supersymmetric YM quantum mechanics [35].
Summarizing, even though above quantum mechanics is much simpler that the M-
theory, it remains unsolved and thereby still poses an interesting challenge.
In this paper we propose a new approach to this problem and present a series of
quantitative results for simpler systems not always solved up to now. To this end we
use the standard Hamiltonian formulation of quantum mechanics in the continuum,1
construct explicitly the (finite) basis of physical states and calculate algebraically matrix
representations of all relevant observables. This done, we proceed to calculate numerically
the complete spectrum, the energy eigenstates, and identify (super)symmetry multiplets.
This approach is entirely insensitive to the sign problem and is equally well applicable to
systems with and without fermions. Similarly to the lattice approach, the method has an
intrinsic cut-off: any quantitative results can be obtained only within the finite-dimensional
subspace of the whole Hilbert space. It turns out, however, that in all cases studied below
many important characteristics (in particular the low energy spectrum) can be reliably
obtained before the number of basis vectors grows out of control.
General Hamiltonian methods have been applied before to complete, space extended
field theories [38]. Recently Matsumura and collaborators [39] and Pinsky et al. have
applied this technique to study a variety of partly reduced, supersymmetric theories in
lower dimensions (see Refs. [40,41] and references therein).
In the next section we describe present approach and its algebraic computer implemen-
tation. In Sections 3 and 4 the spectra of Wess–Zumino quantum mechanics and D = 2
supersymmetric Yang–Mills SU(2) quantum mechanics are derived. Section 5 contains
main results of this paper. It is devoted to D = 4 supersymmetric Yang–Mills quantum
mechanics with the SU(2) gauge group. The global picture of the whole system, with the
quantitative spectra in all fermionic channels and their supersymmetric interrelations, will
be presented for the first time. Summary and discussion of the future applications follow
in the last two sections.
2. Quantum mechanics in a PC
We begin with the simple observation that the action of any quantum mechanical
observable can be efficiently implemented (e.g., in an algebraic program) if we use the
discrete eigen basis
1 n
{|n}, |n = √ a † |0, (1)
n!
1 This work was inspired by the discussion with C.M. Bender and the methods he developed in studying
various quantum mechanical systems [36,37].
of the occupation number operator a † a. For example, the bosonic coordinate and
momentum operators can be written as
1 1
x = √ a + a† , p = √ a − a† , (2)
2 i 2
where all dimensionfull parameters are set to 1. Since typical quantum observables are
relatively simple functions of x and p, they can be represented as the multiple actions
of the basic creation and annihilation operators.2 Including fermionic observables is
straightforward and will be done in subsequent sections. Generalization to more degrees of
freedom is also evident and will be carried out separately for individual systems.
The first step to quantify above considerations is to implement the Hilbert space in
an algebraic program. Any quantum state is a superposition of arbitrary number, ns , of
elementary states |n

ns

|st = aI n(I ) , (3)
I
and will be represented as a Mathematica list3

st = ns , {a1 , . . . , ans }, n(1) , n(2) , . . . , n(ns ) , (4)
with ns + 2 elements. The first element specifies the number of elementary states entering
the linear combination, Eq. (3), the second is the sublist supplying all (in general complex)
coefficients aI , I = 1, . . . , ns , and the remaining ns sublists specify the occupation
numbers of elementary, basis states. In particular, convention (4) implies for an elementary
state

|n ↔ 1, {1}, {n} . (5)
In the case of more degrees of freedom the single occupation numbers n(I ) in Eq. (4) will
become lists themselves containing all occupation numbers corresponding to the individual
degrees of freedom.
In the next step we implement basic operations defined in the Hilbert space: addition
of two states, multiplication by a number and the scalar product. All these can be simply
programmed as definite operations on Mathematica lists. Table 1 displays corresponding
routines which transform lists in accord with the principles of quantum mechanics.
This done, it is a simple matter to define the creation and annihilation operators
which act as a list-valued functions on above lists. This also defines the action of the
position and momentum operators according to Eq. (2). Then we proceed to define any
quantum observables of interest: Hamiltonian, angular momentum, generators of gauge
transformations and supersymmetry generators, to name only the most important examples.
Now our strategy should be obvious: given a particular system, we define the list
corresponding to the empty state, then we generate a finite basis of Ncut vectors and
2 The method can be also extended to non-polynomial potentials.

3 Of course any other algebraic language can be used. In more advanced and time consuming applications we
shall be using much faster, compiler based, languages anyway.
Table 1
Quantum states in the Hilbert space and their computer implementation
Operation Quantum mechanics PC Mapping
Any state |st list
Sum |st1 + | st2 add[list 1 , list 2 ] (list1 , list 2 ) → list3
Number multiply α|st1 mult[α, list 1 ] list1 → list 2
Scalar product st1 |st2 sc[list1 , list2 ] (list1 , list 2 ) → number
Empty state |0 {1,{1},{0}}
Null state 0 {0,{}}
calculate matrix representations of the Hamiltonian and other quantum operators using
above rules. Thereby the problem is reduced to a simple question in linear algebra, namely
the spectrum of the system is given by the eigenvalues and eigenvectors of the Hamiltonian
matrix, their behaviour under rotations by the angular momentum matrix, etc.
Of course the most important question is how much the ultimate physics at Ncut = ∞
is distorted by the finite cut-off. This can be answered quantitatively by inspecting the
dependence of our results on Ncut for practically available sizes of the bases. In all systems
studied so far (and discussed below) the answer is positive: one can extract the meaningful
(i.e., Ncut = ∞) results before the size of the basis becomes unmanageable. However, this
can be answered only a posteriori and individually for every system.
The method outlined above turns out to be a rather powerful tool capable to solve
quantitatively various quantum mechanical problems with finite but large number of
degrees of freedom.
3. Wess–Zumino quantum mechanics
This system is obtained by dimensional reduction of the four-dimensional Wess–

Zumino model leaving only the time dependence of the dynamical degrees of freedom.
Resulting quantum mechanics (WZQM) has one complex bosonic variable φ(t) = x(t) +
iy(t) ≡ x1 (t) + ix2 (t) and two complex Grassmann-valued fermions ψα (t), α = 1, 2 [42].
The Hamiltonian reads
1
H = px2 + py2 + |F |2 + (m + 2gφ)ψ1 ψ2 + h.c. , (6)
2
where the auxiliary field F is determined in terms of φ by the classical equation of motion,

F = − mφ + gφ 2 , (7)
since it enters only quadratically into the action. Independent variables satisfy the standard
commutation relations

[xi , pk ] = iδik , ψα , ψβ† = δαβ , (8)
which allow us to parameterize these coordinates in terms of the creation and annihilation
operators
1 1
x = √ ax + ax† , px = √ ax − ax† , (9)
2 i 2
1 1
y = √ ay + ay† , py = √ ay − ay† , (10)
2 i 2
ψα = fα , ψα = fα† ,
†
(11)
which in turn satisfy

ai , ak† = δik , fα , fβ† = δαβ . (12)
Now we implement this quantum mechanical system in the computer as explained in the
previous section. The empty state will be represented by a list

(0, 0), (0, 0) ↔ 1, {1}, {0, 0}, {0, 0} , (13)
where the first brackets specify occupation numbers of the bosonic, and the second
fermionic, states. To define properly the fermionic creation/annihilation operators we
follow the original construction of Jordan and Wigner [43,44]
fα = Πβ<α (−1)Fβ σα− , fα† = Πβ<α (−1)Fβ σα+ , (14)
where the fermionic number operator Fα = fα† fα (no sum) and σα± are rising and lowering
operators, commuting for different α, with (σα± )2 = 0. The Jordan–Wigner phases ensure
uniform anticommutation rules for fermionic operators.
We proceed to construct the finite eigen basis of the fermionic, Fα , and bosonic, Bi ,
number operators. For more than one degree of freedom the organization of a basis and
definition of the cut-off, Ncut , is not unique. In this case we have chosen the cut-off to
be the maximal number of bosonic quanta. That is, the basis subject to the cut-off Ncut
consists of all orthonormal elementary states with B = B1 + B2 Ncut and all allowed
fermionic quanta, i.e., F = F1 + F2 2. In practice the basis is created by action on
the empty state with all elementary independent monomials of bosonic creation operators
up to Ncut ’th order, followed by the three independent monomials of fermionic creation
operators. Hence, the Hilbert space, subject to the cut-off Ncut , has 2(Ncut + 1)(Ncut + 2)
dimensions.
Given the basis, the matrix representation of the Hamiltonian (and any other observable
of interest) can be readily calculated with our computer based “quantum algebra”. The
spectrum and energy eigenstates are then obtained by numerical diagonalization of the
Hamiltonian matrix.
Fig. 1 shows the spectrum of low energy states as a function of the cut-off up to
Ncut = 15 for m = g = 1. Energies of the fourth and fifth level are shifted by 1 to avoid
confusion of levels for low Ncut . Clear convergence with Ncut is seen, well within a capacity
of a medium class PC. The convergence is faster for lower states. This general feature of
our method can be easily understood. The basis generated by creation operators, Eq. (12),
is nothing but the eigen basis of some normalized harmonic oscillator. It is obvious that
the ground state can be easier approximated by a series of harmonic oscillator wave
functions that the higher states with all their detailed structures, zeroes, etc. Quantitatively,
at Ncut = 10 the exact supersymmetric ground state energy is reproduced with the 3%
accuracy which further improves to below 1% at Ncut = 15.4 Supersymmetric pattern of
4 Since the exact value is zero, we take the first excited energy as a reference scale.
Fig. 1. Spectrum of WZ quantum mechanics as a function of cut-off, Ncut .
Table 2
Splittings within the first three supersymmetric multiplets for different cut-offs
Ncut M1 M2 M3
9 0.1321 0.1574 0.1835
10 0.0924 0.1213 0.1245
11 0.0595 0.0941 0.0886
12 0.0420 0.0657 0.0609
13 0.0282 0.0440 0.0407
14 0.0212 0.0332 0.0287
15 0.0149 0.0207 0.0192
the spectrum is also evident. First, the lowest bosonic state has no fermionic counterpart, its
energy tends to zero, hence we identify it with the supersymmetric vacuum of the model.5
Second, all higher states group into bosonic–fermionic multiplets with the same energy.
Splittings inside the multiplets decrease with Ncut as shown in Table 2. We conclude that
supersymmetry at low energies is restored at the level of few percent in the cut Hilbert
space with states containing up to 15 bosonic quanta.
Let us see how the above convergence to the supersymmetric spectrum shows up in the
Witten index, which in this approach can be calculated directly from the definition

IW (T ) = (−1)Fi exp(−T Ei ). (15)
i
5 The model exhibits additional two-fold degeneracy in bosonic and fermionic sectors which is not related to
supersymmetry [42]
Fig. 2. Witten index for Wess–Zumino quantum mechanics.
Fig. 2 shows Witten index for 4 < Ncut < 15. Nice convergence to the exact, time
independent, result IW (T ) = 2 is observed. Two things are happening which ensure this
behaviour. First, there are two (due to the above degeneracy within bosonic and fermionic
sectors) supersymmetric ground states. At the largest cut-off, Ncut = 15, their energies are
not exactly zero but tiny E0,0 = 0.0046, E0,1 = 0.0071, hence giving rise to the exponential
fall-off at large T with a very small slope. Second, cancellations between non-zero energy
states within supersymmetric multiplets are not exact and still leave exponential terms,
albeit with smaller and smaller (with increasing Ncut ) coefficients. This gives the residual
time dependence which is however much weaker than the exponential one.
Another, rather general feature of IW (T ), can be also observed here. Exactly at T = 0
Witten index vanishes for every Ncut just because the bases generated with our procedure
automatically satisfy a “global” supersymmetry requirement. Namely, the number of
bosonic and fermionic states in a basis is the same for every Ncut . Therefore, we conclude
that
lim IW (0, Ncut ) = 0. (16)

Ncut →∞
On the other hand exact supersymmetry requires that this complete correspondence
between fermionic and bosonic states is violated since the ground state at E = 0 does not
have its fermionic counterpart while each non-zero energy state has. This apparent paradox
is reconciled by noting that exact supersymmetry emerges only in the limit of the infinite
Ncut . In another words the unbalanced fermionic state is pushed (with increasing Ncut )
to higher and higher energies and eventually “vanishes from the spectrum”. However, it
leaves a visible effect—Witten index has a discontinuity at T = 0 which can be expressed
as the non-commutativity of the T → 0 and Ncut → ∞ limits
lim lim IW (T , Ncut ) = lim lim IW (T , Ncut ). (17)

Ncut →∞ T →0 T →0 Ncut →∞
The right-hand side of the above equation, known as the bulk contribution, may not be
integer for the continuous spectrum [13].
Above results were obtained for m = g = 1, but the method works equally well for
any other choices of parameters. The case m = 0 has a rotational symmetry and can
be solved by other methods [45]. To our knowledge no quantitative calculation of the
spectra and wave functions for m = 0 exists. As mentioned earlier this approach provides
a complete quantum mechanics of a system. In particular, wave functions in the coordinate
representation are simply given by the well defined linear combinations of the harmonic
oscillator wave functions. More detailed discussion of the model, and comparison with
other authors will be presented elsewhere.
4. D = 2 supersymmetric SU(2) Yang–Mills quantum mechanics
This system, although solved analytically [4], has two new features: gauge invariance
and continuous spectrum. Therefore, we have chosen it as a next exercise. Reducing from
D = 2 to one time dimension one is left with the three real, colored, bosonic variables xa (t)
and three complex, fermionic degrees of freedom ψa (t) also in the adjoint representation
of SU(2), a = 1, 2, 3.
The Hamiltonian reads [4]
1
H = pa pa + ig$abc ψa† xb ψc , (18)
2
where the quantum operators x, p, ψ, ψ † satisfy

[xa , pb ] = iδab , ψa , ψb† = δab . (19)
Hence they can be written in terms of the creation and annihilation operators as before
1 1
xa = √ aa + aa† , pa = √ aa − aa† , (20)
2 i 2
ψa = fa , ψa = fa .
† †
(21)
To implement fermionic creation and annihilation operators we again used the Jordan–
Wigner construction.
The system has a local gauge invariance with the generators

Ga = $abc xb pc − iψb† ψc . (22)
Therefore, the physical Hilbert space consists only of the gauge invariant states. This can
be easily accommodated in our scheme noting that the gauge generators of the SU(2) are
just the angular momentum operators acting in color space. In fact the fermionic part of
Eq. (22) can be also interpreted in this way since the momentum canonically conjugate to
ψ: πψ = iψ † . Therefore, we construct all possible invariant under SU(2) combinations of
the creation operators (referred for short as creators) and use them to generate a complete
gauge invariant basis of states. There are four lower order creators:
aa† aa† ≡ (aa), aa† fa† ≡ (af ), (23)

$abc aa† fb† fc† ≡ (aff ), $abc fa† fb† fc† ≡ (fff ). (24)
Pauli principle implies that (ff ) = (af )2 = (aff )2 = (fff )2 = 0, therefore, the whole
basis can be conveniently organized into the four towers of states, each tower beginning
with one of the following states
|0F = |0, |1F = (af )|0, |2F = (aff )|0, |3F = (fff )|0, (25)
where we have labeled the states by the gauge invariant fermionic number F = fa† fa . The
empty state (and its Mathematica representation) reads

|(0, 0, 0), (0, 0, 0) ↔ 1, {1}, {0, 0, 0}, {0, 0, 0} , (26)
with the obvious assignment of bosonic and fermionic occupation numbers. To obtain the
whole basis it is now sufficient to repeatedly act on the four vectors, Eq. (25), with the
bosonic creator (aa), since action with other creators either gives zero, due to the Pauli
blocking, or produces linearly dependent state from another tower. Basis with the cut-off
Ncut is then obtained by applying (aa) up to Ncut times to each of the four “ground” states.
Of course our cut-off is gauge invariant, since it is defined in terms of the gauge invariant
creators. Moreover, since the gauge generators can be implemented into the algebraic
program as any other observables, we can directly verify gauge invariance of the basis
and any other state in question.
Once the above basis is constructed, we proceed to calculate the matrix representation
of the Hamiltonian, the spectrum and the eigenstates according to Section 2.6 Again
all computation can be done on a small PC and the longest run, corresponding to 20
bosonic quanta, takes approximately 500 sec. It is worth to mention that the most time
consuming are the algebraic operations on the states (cf. Table 1) in the PC-based abstract
Hilbert space. Consequently the program spends most of the time calculating the matrix
representations of various operators 7 while the numerical diagonalization is rather fast.
The Hamiltonian, Eq. (18) can be rewritten
1
H = pa pa + gxa Ga , (27)
2
hence, it reduces to that of a free particle in the physical, gauge invariant basis. It
follows that it preserves the gauge invariant fermionic number and, consequently, can be
diagonalized independently in each sector spanned by the four towers in Eq. (25).
The spectrum is doubly degenerate because of the particle-hole symmetry which relates
empty and filled fermionic states (|0F ↔ |3F ) and their 1-particle 1-hole counterparts
(|1F ↔ |2F ). This symmetry is not violated by the cut-off and indeed we see it exactly
in the spectrum. On the other hand, supersymmetry connects sectors which differ by 1 in
the fermionic number, e.g. (|0F ↔ |1F ) and in general is restored only at infinite Ncut .
A good measure of the SUSY violation is provided by the energy of the ground state which
should be 0. We see that the energy of the lowest (doubly degenerate) state converges to
0 but rather slowly, cf. Fig. 3. This is interpreted as the indication that the spectrum is
continuous at infinite cut-off. Indeed, it is hard to approximate the non-localized state by
6 All numerical results are for g = 1.

7 In fact this part can be (and often is) done analytically in the integer arithmetic.
Fig. 3. D = 2 supersymmetric Yang–Mills quantum mechanics. The energy of the lowest state as a function of he
cut-off and the 1/Ncut fit (dotted line).
the harmonic oscillator states (this in fact is our basis) and consequently the convergence
of such an approximation must be slow.8 This deficiency can be sometimes turned into an
advantage as will be seen in the D = 4 case.
Surprisingly, however, there exists a particular scheme of increasing the basis which
renders exact supersymmetry at every finite cut-off Ncut in this model. To see that, let us
first discuss tests of the SUSY on the operator level. SUSY generators read
Q = ψa pa , a = ψ̄a pa ,
Q (28)
with ψ̄ = ψ † , and satisfy the algebra
= 2H − gxa Ga ,
{Q, Q} 2 = 0.
Q2 = Q (29)
Since the matrix elements of the SUSY generators can be easily calculated in our PC-
based Hilbert space approach, one can readily check the matrix element version of
Eqs. (29). Indeed, we find that these relations are satisfied to better and better precision
with increasing Ncut . If so, it is natural to ask what is the spectrum of the Hamiltonian

matrix H QQ defined by Eq. (29) at finite cut-off. To this end we diagonalize the matrix

QQ 1 + N|Q|MM|Q|N

HN,N = N|Q|MM|Q|N , (30)
2
where N, N and M label vectors of our finite basis.9 Of course SUSY generators mix
the fermionic number, therefore, one should combine all four towers of Eq. (25) into one
big basis. One more trick is required to achieve exact SUSY at finite cut-off: we have to
increase the basis allowing from the beginning for the disparity between the fermionic and
bosonic sectors. That is, the size of the basis grows as: 2 + 4 + 4 + · · ·. Hence, dimension
8 This argument can be turned into a proof that the free spectrum converges like 1/N
cut [46] which is also
confirmed here (cf. Fig. 3).
9 The second term in Eq. (29) does not contribute in the physical basis.
Table 3

The spectrum and the degeneracy, d, of the Hamiltonian H QQ defined at finite Ncut as {Q, Q}/2
2 + 4Ncut 10 14 18 22 26 d
E0 0 0 0 0 0 2
E1 0.815 0.610 0.489 0.409 0.351 4
E2 2.685 1.904 1.495 1.236 1.056 4
E3 4.235 3.160 2.558 2.160 4
E4 0 5.856 4.522 3.743 4
E5 0 7.525 5.960 4
E6 0 0 9.230 4
of the cut Hilbert space is 2 + 4Ncut in this scheme. With this choice the spectrum of the

H QQ has exact supersymmetry as shown in Table 3.
As required, the ground state has zero energy and does not have the supersymmetric
image while higher states form supersymmetric doublets with the same energy. Additional
degeneracy is caused by the particle-hole symmetry as explained earlier. All non-zero
eigenvalues (with finite “principal” quantum number) tend to zero with increasing Ncut and
form, at Ncut = ∞, the continuum spectrum of a free Hamiltonian, Eq. (27). To reach the
solutions with non-zero energy, one should appropriately scale the index of an eigenvalue
with Ncut . The four towers of eigenstates are in the direct correspondence with the four
gauge invariant plane waves constructed in Ref. [4].
Even though existence of the supersymmetry preserving cut-off is quite interesting, we
belive that it is related to the simple structure of the D = 2 SYM quantum mechanics and
its complete solubility. Nevertheless, it is important to keep in mind that at finite Ncut one
is free to define the Hamiltonian within the “O(1/Ncut )” tolerance, i.e., as long as various

definitions converge to the same limit at infinite Ncut [39–41]. We decided to use H QQ
because it is the anticommutator of charges which is used to derive basic properties of the
SUSY spectrum. Further study of this exact realization of supersymmetry will be continued
elsewhere.
Finally let us shortly discuss the Witten index for this model. Because the particle-
hole symmetry interchanges odd and even fermionic numbers, the Witten index vanishes
identically. This is also true for any finite cut-off since Ncut preserves particle-hole
symmetry. Nevertheless one can obtain a non-trivial and interesting information by
defining the index restricted to the one pair of fermion–boson sectors. For example,

IW (T )(0,1) = (−1)Fi exp (−T Ei ), (31)
i,Fi =0,1
where the sum is now restricted only to the F = 0 and F = 1 sectors. Since supersymmetry
balances fermionic and bosonic states between these sectors (with the usual exception
of the vacuum), the restricted index is a good and non-trivial measure of the amount
of the violation/restoration of SUSY, even when the total Witten index vanishes.
(2,3) (0,1) (0,3) (1,2)
Obviously IW = −IW , and IW = IW = 0. Studying restricted index is particularly
interesting in this model since, due to the continuum spectrum, it does not have to be
integer and provides some information about the density of states. We have calculated the
index from our spectrum of the original Hamiltonian, Eq. (18), for a range of cut-offs:
Fig. 4. Restricted Witten index for D = 2 SYMQM.
5 < Ncut < 20. Fig. 4 shows the result for the 2 + 4 + 4 + · · · scheme, i.e., when the basis
is increased by every four vectors allowing from the beginning for the two unbalanced
states from the empty and filled sectors each. We see a slow approach towards the time
independent constant which seems to be 1/2.
Witten index, as a sum over all states, provides an average measure of the break-
ing/restoration of SUSY. Indeed, even though individual supersymmetric multiplets have
not yet clearly formed (at cut-offs used for this calculation), we see definite flattening of
the T dependence of IW (T ) which must then be a result of the average cancellation be-
tween many levels. Therefore, it is easier to see the onset of the supersymmetric behavior
in the Witten index than in the location of the individual levels. As before Fig. 4 shows
that the limiting value of the index is discontinuous at T = 0 with the same interpretation
as in Section 3. However, contrary to the discrete spectrum of the Wess–Zumino model,
the value of the index is not integer.10 The value of the index at T = 0 is the direct conse-
quence of our scheme (2 + 4 + 4 + · · ·) of increasing the basis and, of course, is scheme
dependent. As a final remark we note an intriguing existence of a “universal” point in T at
which the asymptotic value seems to be attained at all Ncut .
5. D = 4 supersymmetric SU(2) Yang–Mills quantum mechanics
5.1. Preliminaries
Dimensionally reduced (in space) system is described by nine bosonic coordinates

xai (t), i = 1, 2, 3, a = 1, 2, 3, and six independent fermionic coordinates contained in the
Majorana spinor ψaα (t), α = 1, . . . , 4. Equivalently (in D = 4) one could have chosen to
impose the Weyl condition and worked with Weyl spinors. Hamiltonian reads [7]
H = HB + HF , (32)
10 It would be an interesting exercise to calculate the restricted index from the continuous spectrum of Ref. [4].
1 g2 j j
HB = pai pai + $abc $ade xbi xc xdi xe ,
2 4
ig
HF = $abc ψaT Γ k ψb xck , (33)
2
where ψ T is the transpose of the real Majorana spinor, and Γ in D = 4 are just the
standard Dirac α matrices. In all explicit calculations we are using Majorana representation
of Ref. [47].
Even though the three-dimensional space was reduced to a single point, the system
still has the internal spin(3) rotational symmetry, inherited from the original theory, and
generated by the angular momentum

j 1
J i = $ ij k xa pak − ψaT Σ j k ψa , (34)
4
with
i j k
Σjk = − Γ ,Γ . (35)
4
Further, the system has gauge invariance with the generators

i T
Ga = $abc xb pc − ψb ψc ,
k k
(36)
2
and is invariant under the supersymmetry transformations generated by
j
Qα = Γ k ψa α pak + ig$abc Σ j k ψa α xb xck . (37)
The bosonic potential (written now in the vector notation in the color space)
g2 2
V= Σj k x j × x k , (38)
4
exhibits the famous flat directions responsible for a rich structure of the spectrum.
At first sight one might expect that the spectrum of the purely bosonic (hence, non-
supersymmetric) system does not have localized states because of these flat valleys (cf.
Introduction). However, the flat directions are blocked by the energy of the transverse
quantum fluctuations since valleys narrow as we move from the origin. As a consequence
the spectrum of the model is discrete. Energies of the first lower states, known as zero
volume glueballs, were first calculated by Lüscher and Münster [32]. On the other hand,
in the supersymmetric system, transverse fluctuations cancel among bosons and fermions,
valleys are not blocked, and the spectrum of the model is expected to be continuous [8].
The story has its interesting continuation in D = 10 model where the evidence of the
threshold bound state was accumulated [12–17]. Existence of such a state is necessary
for the M-theory interpretation of the model where it is considered as a prototype of the
graviton multiplet. Therefore, detailed study of the low energy spectra of the whole family
of Yang–Mills quantum mechanical systems, including identification of the localized and
non-localized states is an important and fascinating subject. We shall face some of these
questions also in the D = 4 system.
5.2. Creation and annihilation operators
There are many ways to write quantum coordinates in terms of creation and annihilation
operators
i k† ρ σ † ρσ
aa , ab = δab
ik
, fa , fb = δik , ρ, σ = 1, 2. (39)
The only constraint comes from the canonical (anti)commutation relations
i k α β αβ
xa , pb = iδ ik δab , ψa , ψb = δab . (40)
For bosonic variables we just use the straightforward extension of Eq. (20) to more degrees
of freedom
1 1
xai = √ aai + aai† , pai = √ aai − aai† . (41)
2 i 2
For fermionic variables we begin with the classical Majorana fermion in the Weyl
representation [48]

ψW = ζ24 , −ζ14 , ζ1 , ζ2 , (42)
replace the classical Grassmann variables ζ, ζ 4 by fermionic creation and annihilation
operators f, f † , and transform to the Majorana representation. Final result is a quantum
hermitian Majorana spinor
 
−fa1 − ifa2 + ifa1† + fa2†
1 + i  +ifa1 − fa2 − fa1† + ifa2† 
ψa = √  
2† , (43)
2 2  −f + if + ifa − fa 
1 2 1†
a a
−ifa1 − fa2 + fa1† + ifa2†
which satisfies Eq. (40) due to (39). Other choices of fermionic creation and annihilation
operators are also possible [6,7,15].
The next step is to define an empty state and its, computer based, algebraic
representation
|(0, 0, 0), (0, 0, 0), (0, 0, 0), (0, 0, 0), (0, 0, 0) (44)

↔ 1, {1}, {{0, 0, 0}, {0, 0, 0}, {0, 0, 0}, {0, 0, 0}, {0, 0, 0}} , (45)
where the first three vectors (in color) specify bosonic, and the last two fermionic,
occupation numbers.
The Jordan–Wigner transformation requires additional specification in this case. As
before, the action of fai and fai† on elementary states is defined by the spin-like raising
and lowering operators corrected by the non-local Jordan–Wigner phases
fA = ΠB<A (−1)FB σB− , fA† = ΠB<A (−1)FB σA+ . (46)
However, since individual states are now labeled by a double index A, one must define the
ordering of the two-dimensional indices. We choose the lexicographic order. If A = (a, α)
and B = (b, β) than
ΠB<A = Πβ<α,b3 Πβ=α,b<a . (47)
Any other unambiguous definition of the ordering is admissible.
5.3. Constructing the basis
There are many ways to construct bases in higher-dimensional systems. Apart from
theoretical requirements one must also take into account practical limitations of computer
implementation. Of course the number of states at finite cut-off is generally bigger for
higher D. However, the theory also has more symmetries and consequently the number of
states in a particular channel can be kept manageable.
First important symmetry is the conservation of the number of Majorana fermions
[F, H ] = 0, F = fai† fai . (48)
That is, the vector interaction HF cannot produce Majorana pairs in this model. As a
consequence the whole Hilbert space splits into seven sectors of fixed F = 0, 1, . . . , 6.
Second, the system has the particle-hole symmetry
F ↔ 6 − F, (49)
therefore the first four sectors F = 0, 1, 2, 3 contain all information about the spectrum.
Further, the gauge invariance requires that our basis is built only from the gauge
invariant creators. All this is not different from the earlier D = 2 case.
The new element is the SO(3) invariance, with the angular momentum Eq. (34), which
can be used to split further the problem into the different channels of fixed J . However,
more we specify the basis more complicated its vectors become, and consequently more
computer time goes into the generation of such a basis and subsequent calculation of
matrix elements. Similarly one might contemplate using the reduced version of Lorentz
invariant composite operators as possible creators of a basis. Again this would generate
more complex basis since fields contain both creation and annihilation operators. Also,
Lorentz covariance is not that relevant in our fixed frame, Hamiltonian formulation.
Taking all above into account we have decided to produce the simplest basis of gauge
invariant vectors using elementary creation operators. To this end we proceed as follows.
Consider each fermionic sector separately, i.e., fix the fermionic number 0 F 3. At
given F create all independent vectors with fixed number of bosonic quanta B = aai† aai ,
and define a cut-off as the maximal number of bosonic quanta B Ncut , independently for
each F . Hence, Ncut can depend on F . To create all independent, gauge invariant states at
fixed F and B consider all possible contractions of color indices in a creator of (F, B) order
aai11† · · · aaiBB† fbσ11 † · · · fbσFF † , (50)
for all values of the spatial indices i and σ . All color contractions fall naturally into dif-
ferent gauge invariant classes. Creators from different classes differ by color contractions
between bosonic and fermionic operators. For example,
j†
aai† aa abk† abl† fcσ † fcρ† (51)
and
j† ρ†
aai† aa abk† acl† fcσ † fb (52)
belong to different gauge invariant classes according to our definition. Another example
involves odd number of operators where one “contraction” consists of one triple of color
Table 4
Sizes of the bases generated in each fermionic sector, F . Ns is the number of basis vectors with given number of
bosonic quanta, B, while Σ gives the cumulative size up to B. The last column gives the difference between the
total number of the bosonic and fermionic states in all seven sectors
F 0 1 2 3
B Ns Σ Ns Σ Ns Σ Ns Σ B −F
0 1 1 – – 1 1 4 4 0
1 – 1 6 6 9 10 6 10 0
2 6 7 6 12 21 31 42 52 0
3 1 8 36 48 63 94 56 108 0
4 21 29 36 84 111 205 192 300 0
5 6 35 126 210 240 445 240 540 0
6 56 91 126 336
7 21 112
8 126 238
jmax 8 11/2 6 11/2
indices coupled with the Levi-Civita symbol.11 Namely,

j†
$cde aai† aa abk†abl† acm† fdσ † feρ† (53)
and
j† ρ†
$cde aai† aa abk†acl† adm† feσ † fb (54)
also belong to different gauge invariant classes. Different contractions within bosonic and
fermionic family of operators do not lead to independent states hence are considered to be
in the same class. Still there will be many linearly dependent states generated by above
creators with all values of spatial indices i and σ . It is however a simple matter to recog-
nize linearly independent ones given our rules of “quantum algebra”. We therefore leave
it to the computer for the time being. At fixed F and B the final procedure is as follows:
(1) identify all gauge invariant classes of creators, (2) loop over all values of spatial in-
dices and for each i1 , . . . , iB , σ1 , . . . , σF create corresponding state from the empty state,
Eq. (45), (3) identify and reject linearly dependent vectors, (4) orthonormalize the remain-
ing set of states. In principle this procedure depends exponentially on Ncut , hence, can be
further shortened and improved, however it is sufficiently fast for our present purposes. It
also has an advantage of generating all channels of angular momentum (available within
the cut-off Ncut ). This will prove useful in studying supersymmetry.
Table 4 shows the sizes of bases we have reached so far in each of the fermionic
sectors.12 Due to the particle-hole symmetry the structure in the F = 4, 5, 6 sectors is
identical with that in F = 2, 1, 0, respectively.
Calculation of the spectrum of H proceeds exactly as earlier. Naturally we can also
readily obtain the spectrum of other observables, e.g., of the angular momentum, Eq. (34),
11 For the SU(2) gauge group, this step can be generalized to higher SU(N ) groups.
12 By the time of printing the whole Table 4 has been filled out [51].
as well as the angular momentum content of the energy eigenstates. For example, the last
row of Table 4 gives the highest angular momentum which can be constructed in each
sector with the bases generated so far.
5.4. Results
The spectrum of the theory is shown in Fig. 5. A sample of states is labeled with
their angular momenta. It is important that our cut-off preserves the SO(3) symmetry.
Consequently, the basis described in previous section contains complete representations
of the rotation group for any Ncut . Hence, the spectrum displayed in Fig. 5 has appropriate
degeneracies for each value of the angular momentum quantum number j .
To maintain some clarity of the figure we have cut arbitrarily the upper part of the
spectrum, which extends to about Emax ∼ 35 with current Ncut . One expects that the
individual higher states have a considerable dependence on Ncut . However, they also carry
some relevant information, e.g., for quantum averages.
Apart from relating the bases in corresponding fermionic sectors, particle-hole symme-
try also implies equality of all corresponding energy levels. Since our cut-off respects this
symmetry we indeed observe, in sectors with F = 6, 5, 4, exact repetition of the eigenval-
ues from F = 0, 1, 2 subspaces. Therefore, only first four sectors are displayed in Fig. 5.
Evidently the spectrum of D = 4 theory is very rich and raises many interesting questions.
We shall discuss some of them, beginning with the relation to the already known results.
Fig. 5. Spectrum of D = 4 SYMQM.

Fig. 6. First three energy levels in the F = 0 sector of D = 4 SYMQM for different cut-offs. Solid lines show the
0-volume results of Ref. [32].
5.4.1. Correspondence with pure Yang–Mills quantum mechanics

It turns out that the fermionic part of the Hamiltonian, Eq. (33), annihilates any state in
the F = 0 sector
HF |0F = 0. (55)
Therefore, the supersymmetric Hamiltonian, Eq. (33) reduces in that sector to the
Hamiltonian of the pure Yang–Mills quantum mechanics, well known from the small
volume approach [10,32–35]. This offers us a chance to compare one special case of the
present spectrum with the classical results of Ref. [32]. Fig. 6 shows such a comparison for
the first three states as a function of Ncut . Nice convergence of our results towards those
of Lüscher and Münster is seen. They used 120 states in each angular momentum channel
and variationally optimized frequencies of the basis states. Our F = 0 basis contains 238
states which span all angular momenta up to j = 8, and has only 12 states with j = 0. No
variational adjustment of the frequencies was attempted. In view of this we consider above
agreement as a rather satisfactory confirmation of the present approach. The assignment of
the angular momentum, which is done automatically within our method, is also the same.
In particular, the characteristic to the 0-volume, ordering of the tensor glueball and the first
scalar excitation is reproduced.
5.4.2. The cut-off dependence

The dependence of the spectrum on Ncut is of course crucial, as it provides the
quantitative criterion if the size of the basis is adequate. It is displayed in Fig. 7 for the
first few lower levels in all fermionic sectors. It is clear that the cut-off dependence is
different in various fermionic sectors. For F = 0, 1 (and F = 6, 5) the dependence is rather
weak, suggesting that the energies of lower levels have already converged (within few
percent) to their asymptotic values. On the other hand, in F = 2, 3, 4 sectors convergence
is considerably slower. As explained earlier, one can use the rate of convergence with Ncut
to distinguish between the continuous and discrete spectrum. It was proved that in the
case of the free spectrum the convergence is O(1/Ncut ) i.e., slow [46]. For the localized
bound states we conjecture that the rate of convergence with Ncut is sensitive to the large x
Fig. 7. Cut-off dependence of the lower levels of the D = 4 SYMQM in four independent fermionic channels.
asymptotic of the wave function. We have observed this in a simple anharmonic oscillator
and other models. This regularity is also clearly confirmed in the Wess–Zumino and D = 2
SYMQM models discussed in previous sections.
Taking this into account we claim that the low energy spectrum of D = 4 SYMQM is
discrete in F = 0, 1, 5, 6 and continuous in the F = 2, 3, 4 sectors. This is an interesting
quantification of the result of Ref. [8] which was mentioned earlier. Since the fermionic
modes are crucial to provide continuous spectrum, it is natural that it does not show up in
the sectors where they do not exist at all, or are largely freezed out by Pauli blocking.
Second result, which is evident from Fig. 7, concerns the supersymmetric vacuum in
this model. Assuming that indeed the eigen energies have approximately converged in
F = 0, 1 sectors (none of them to zero) it follows that the SUSY vacuum cannot be in
empty and filled sectors (F = 1, 5 is also ruled out by the angular momentum).13 The
obvious candidates are the lowest states in F = 2, 4 sectors with their energy consistent
with zero at infinite Ncut , cf. Fig. 7. It follows, together with the conjectured correlation
between localizability and Ncut dependence, that the SUSY vacuum in this model is non-
normalizable. Further evidence is based on the structure of the supersymmetric multiplets,
and will be presented in the next section. Very recently van Baal studied F = 2 sector in
a more complex, non-compact, supersymmetric model of the same family [35]. Although
our results cannot be directly compared, due to the specific boundary conditions required in
[35], he finds that the energy of the lowest state in F = 2 sector is indeed very close to zero.
The deviations from exact zero are caused by the SUSY violating boundary conditions.14
It is important to note that not all states in F = 2, 3, 4 sectors have to belong to the
continuum. Supersymmetry together with the discrete spectrum in F = 0, 1 channels
implies existence of the normalizable states among the continuum of F = 2, 3, 4 states.
13 In early attempts, empty and filled sectors were considered as possible candidates for SUSY vacuum.
14 See Ref. [52] for more recent comparison.
This will be discussed later, here we only give one explicit example. Indeed the energy of
the F = 2, j = 1 state, shown in Fig. 7 (flatter curve beginning at E = 6), has definitely
weaker dependence on Ncut than the others. We interpret this as a signature that the
lowest |2F , 1j state is localized. This situation may be a precursor of the more complex
phenomenon expected in the D = 10 theory. There, the zero energy, localized bound state
of D0 branes should exist at the threshold of the continuous spectrum. Present example
suggests that one way to distinguish such a state from the continuum may be by the
different Ncut dependence.
5.4.3. Supersymmetry
Operator level To check the supersymmetry algebra
{Qα , Qβ } = 2δαβ H + 2gΓαβ

k k
x a Ga , (56)
at finite cut-off, we sandwich both sides of this operator relation between our basis,
introduce the intermediate complete set of states to saturate the product of Q’s and test
the equivalent relations between matrix elements
N|{Qα , Qβ }|N = 2δαβ N|H |N . (57)

The gauge term in Eq. (56) does not contribute since the basis is gauge invariant at any Ncut .
There is, however, one limitation. As is evident from Eq. (37)), supersymmetry generators
change fermionic number by one, and can change number of bosonic quanta, B, at most
by two
N|Qα |M = 0 if FN = FM ± 1, |BM − BN | 2. (58)

It follows that, if a cut-off representation of unity |MM|, BM < Ncut is used to saturate
any matrix element of a product N|Q|MM|Q|N , it will be biased for high states in
N, N sectors. Quantitatively, the supersymmetric relations, Eq. (57) are valid if
BN , BN Ncut − 2. (59)
This limits the number of matrix elements available for the test. Still many interesting
predictions for lower states can be verified and should be satisfied exactly at finite Ncut .
One more property of supersymmetry generators should be kept in mind. Since they
change F by 1, they move between different fermionic sectors. Hence, the generic matrix
equation (57), when restricted to a particular sector, reads (α = β, no sum)
F |H |F = F |Qα |F − 1F − 1|Qα |F + F |Qα |F + 1F + 1|Qα |F . (60)

That is, the Hamiltonian matrix receives contribution from a single fermionic sector, only
if it is computed in empty or filled sectors. Of course above SUSY relations have many
other consequences. Some of them will be used later, when discussing emergence of the
supersymmetric multiplets.
Within the above limitations we have checked some consequences of SUSY algebra at
finite Ncut . The spectrum of the Q2 matrix agrees exactly with that shown in Fig. 5 and
derived directly form the Hamiltonian. Moreover, the off-diagonal commutators indeed
vanish in our basis. Many other interesting questions emerge and will be studied elsewhere.
SUSY on the level of states The next goal is to identify supersymmetric multiplets in the
spectrum and to asses the effect of the cut-off on their splittings and mixings. It is clear
from Fig. 5 that any direct search for approximately degenerate states is difficult and may
not be conclusive. However our construction provides another simple way to move within
the SUSY multiplets. Namely, it is sufficient to use the explicit matrix representation of
supersymmetry generators in bases summarized in Table 4. If supersymmetry was exact,
acting with Q on an eigenstate of H , would generate states within the same multiplet. At
finite Ncut this is no longer true, however we expect that the resulting state, Q|Ψ say, will
be spread around the appropriate multiplet member (or members) in another fermionic
sector. This information will then be correlated with the spectrum of Fig. 5. Important
simplification comes from the rotational symmetry, which is exact at every Ncut in this
approach. Supersymmetric generators carry the angular momentum 1/2 which clearly
limits the range of possible targets Q|Ψ . Indeed, we confirm that
1
F, j |Q|F ± 1, j = 0 |j − j | = .
if (61)
2
To proceed, we denote the kth energy eigenvector in the mth fermionic sector and the
nth angular momentum channel by |mF , nj ; k. The subscript j will be omitted where
evident. A summary of various transitions we have analyzed is shown in Fig. 8. For
example, the action of Q1 on the first state in F = 0 sector gives
(|1F , 1/2; 1|Q1|0F , 0j ; 1|2 + |1F , 1/2; 2|Q1|0F , 0j ; 1|2)

= 95%, (62)
Q1 |0F , 0j ; 12
which provides rather satisfactory confirmation of the first-glance guess from Fig. 5. For
higher states situation is more complex. The second scalar state in this sector goes under
Q1 into Q1 |0F , 0j ; 2 which decomposes in 40% into the |1F , 1/2; 3, 4 doublet and 43%
of yet higher, but also close in energy, |1F , 1/2; 5, 6 doublet. This is not surprising, since
higher states have not yet converged to their infinite Ncut limit.
Even more interesting is to analyze SUSY relations between F = 1 and F = 2 sectors.
Again the lowest |1F , 1/2; 1 state transforms predominantly (84%) into the third scalar
|2F , 0j ; 3 which is closest in the energy. Another consequence of the SUSY algebra can
be seen here. Naively one would expect that combining above transitions |0F , 0j ; 1 →
|1F , 1/2; 1 and |1F , 1/2; 1 → |2F , 0j ; 3 would lead to the connection of F = 0 and
F = 2 sectors and extension of the multiplet. This contradicts Eq. (57) which implies
that the Q21 is diagonal in the eigen basis of the Hamiltonian. Solution of this apparent
paradox is simple. There are two states with j = 1/2 and while the multiplet beginning
in F = 0 sector ends on one linear combination, the second multiplet must start on the
orthogonal one. In other words the supersymmetric images of |0F , 0j ; 1 and |2F , 0j ; 3
in F = 1 sector must be orthogonal. We confirm this exactly (e.g., to the full precision of
Mathematica numerics)15
2F , 0j |Q†1 Q1 |0F , 0j ∼ 10−16 . (63)
15 This is because we use the limited basis Eq. (59), which guarantees Eq. (57) exactly.
Fig. 8. Supersymmeric images of a sample of eigenstates.
The next (in the energy) |1F , 3/2; 1 state goes under Q1 in 30% to |2F , 1j ; 1, 2, 3 triplet,
with the rest shared among higher F = 2 states. In fact, the |2F , 1j ; 1, 2, 3 states are those
with the fast Ncut convergence, which we have already pointed earlier as good candidates
for localized states. Very little of |1F , 3/2; 1 goes back to the F = 0 sector (cf. fat head
arrows in Fig. 8)
Q1 |1F , 3/2; 12F =0

= 6.3%, (64)
Q1 |1F , 3/2; 12
since there is no state with similar energy there.16 This means that the j = 3/2 states start
a new SUSY multiplet which extends into F = 2 sector and not into F = 0 one.
We are using only one of the four SUSY generators just as an example. Similarly we
do not consider systematically all members of j = 0 multiplets. The full analysis, which in
fact would be quite straightforward in the present setup, would involve complete discussion
of the extended, N = 4, SUSY algebra. However, it is out of the scope of this article and
will be done elsewhere.
All above examples have a common feature. Clear SUSY regularities show up if the
states have already converged, i.e., if their energy depends weakly on Ncut . This is not the
case for the lowest states in F = 2, 3, 4 sectors which, we believe, form a continuum at
infinite cut-off. Accordingly, and similarly to the D = 2 SYM QM case, identification of
multiplets and the SUSY vacuum in this sector is not clear at the moment. As one possible
signature of the vacuum one might take the average norm of the supersymmetric image of
16 All surrounding states have already converged within few percent which suggests that the above small rate
is also a O(1/Ncut ) effect.
Fig. 9. Witten index of the D = 4, supersymmetric Yang–Mills quantum mechanics for the number of bosonic
quanta bounded by Ncut = 2, . . . , 5. The bulk value 1/4 is also shown.
the lowest state Q1 |2F , 0j ; 12 / dim[3F ]. Indeed, it is smaller than that for other states
and falls with Ncut . However, it is too early to draw definite conclusions.
Clearly we have only begun. The aim here is just to introduce the new approach and its
potential,17 leaving specific applications for more focused articles.
Witten index The total number of bosonic and fermionic states is the same for each B,
cf. last column of Table 4. This is not a direct consequence of supersymmetry, but rather
of a combinatorics of binary fermionic systems, which is not spoiled by gauge invariance.
Total number of states gives the Witten index at T = 0. Hence, we expect
lim IW (0, Ncut ) = 0. (65)
Ncut →∞
As was already discussed Witten index is discontinuous at T = 0. In particular IW (0)
depends on the regularization, therefore, this number is tied to our B − F symmetric
scheme of increasing the basis.
Fig. 9 shows the (Euclidian) time dependence of the Witten index calculated from our
spectrum with up to five bosonic quanta. Clearly we are far from the convergence, but some
interesting signatures can be already observed. First, one can see how the discontinuity
at T = 0 emerges even at this early stage. Second, the exponential fall-off common to
all curves at large (T > 5) times is evident. This is where the lowest state dominates.
Since at this Ncut its energy is not zero, the fall-off is exponential. Finally, and most
interestingly, we observe the flattening shoulder appearing at T ∼ 2–3. This is the signal
of the supersymmetric cancellations which occur on the average, even though the exact
SUSY correspondence between individual states does not yet appear. The behavior with
Ncut is not inconsistent with the exact bulk value 1/4 [13] obtained also from the non-
Abelian integrals [14,16,17]. Clearly higher Ncut are needed, and, what is important, they
are perfectly within the range of present computers.
17 Only Hamiltonian methods are capable to provide such a detailed and quantitative characteristics as
discussed in this section.
6. Summary
The new approach to quantum mechanical problems is proposed. The Hilbert space of
quantum states is algebraically implemented in the computer code. In particular realization,
used here, states are represented as Mathematica lists. All basic operations on quantum
states are mirrored as definite operations on above lists. Any quantum observable is
represented as a well defined function on these lists. This allows for automatic calculation
of matrix representation of a Hamiltonian and other quantum operators of interest. To this
end we use the discrete eigen basis of the operators of the numbers of quanta. The length
of above lists is not fixed. Similarly to the length of an arbitrary quantum combination of
basic states, it can vary dynamically allowing for any number of quanta.
Of course in any finite computer the maximal length of a combination must be limited.
Therefore, we impose a cut-off Ncut which bounds from above the number of allowed
quanta. The stringent and quantitative test of this approach is provided by checking the
dependence of any physical observable, e.g., the spectrum or the wave functions, on
the cut-off. This is similar to studying the infinite volume statistical systems via the
finite size scaling. We have applied above technique to the three progressively more
complicated systems: Wess–Zumino quantum mechanics, supersymmetric Yang–Mills
quantum mechanics in D = 1 + 1 dimensions, and D = 4 SYM quantum mechanics, both
for the SU(2) gauge group. In distinction from many other approaches (for example, lattice
simulations) the method is completely insensitive to the sign problem and works equally
well for bosonic and fermionic systems.
For Wess–Zumino quantum mechanics we can calculate the discrete spectrum for any
values of parameters. Clear restoration of the supersymmetry was observed for the cut-offs
well within the capabilities of present computers. Witten index was also obtained and its
convergence to the known result IW (T ) = 2 is clearly seen.
The next system, D = 2 SYM QM, possesses the gauge invariance which was readily
incorporated in our approach. The physical subspace of gauge invariant states was
explicitly constructed. Known structure of the solutions in terms of four fermionic sectors
was reproduced. Convergence of the method, and emergence of the supersymmetry, was
also studied in this more difficult case of the continuum spectrum. Witten index, restricted
to one supersymmetric branch of the model, was defined and computed for the first time.
Clear, albeit slow, convergence to the time independent fractional value IW R (T ) = 1/2 was
observed. Moreover, we have found a special scheme of increasing the basis such that the
supersymmetry is exact at any finite cut-off. This however, may be related to the exact
solubility of the model.
Finally, the method was applied to the unsolved up to now D = 4 SYMQM. This
much richer system has the SO(3) rotational symmetry inherited from its space extended
predecessor. Our approach preserves this symmetry exactly at any finite cut-off. The
Hilbert space splits again into seven sectors with fixed fermionic number. We have obtained
the complete spectra in all these sectors and studied their cut-off dependence. The spectrum
in F = 0 sector agrees with the classical 0-volume calculation for pure Yang–Mills
quantum mechanics [32]. An efficient method to distinguish between the discrete and
continuous spectrum was proposed. It turns out that the asymptotics of the wave function at
large distances determines the convergence of our calculations with the number of allowed
quanta Ncut . The continuous spectrum with non-localized wave functions converges slowly
(∼ O(1/Ncut ), while discrete, localized bound states lead to faster (sometimes even
exponential) convergence. Accordingly, we have found an evidence that the spectrum of
D = 4 SYM QM is discrete in F = 0, 1, 5, 6 sectors while it is continuous in the F = 2, 3, 4
sectors. This provides an explicit realization of the claim of Ref. [8] in particular fermionic
sectors. Interestingly, localized states exist also in the sectors with continuous spectrum.
This is a simple consequence of supersymmetry whose transformations move between
adjacent fermionic sectors. Our method allows to monitor directly the action of SUSY
generators and analyse supersymmetric images of any state. In this way we have identified
some candidates for lowest supersymmetric multiplets. They do not have the same energy
at current values of the cut-off, however the splittings are small and consistent with
vanishing at infinite Ncut . In particular, the 0-volume glueballs found in F = 0 sector
are relevant to fully supersymmetric theory in that there exist gluino–gluon bound states
with the same masses. We have also found a candidate for the supersymmetric vacuum
which seems to belong to the continuum. However, identification of SUSY multiplets in
the continuum part of the spectrum requires higher Ncut . Supersymmetry was also tested
on the operator level. In particular we confirmed that the spectrum of the Hamiltonian
coincides with that of Q21 . As a final application we have calculated the Witten index for
this theory. It still depends strongly on Ncut . Nevertheless an early evidence for some of its
asymptotic properties can be already seen.
7. Future prospects
The main goal of this work is to asses the feasibility of attacking with the new approach
the BFSS model of M-theory. Certainly the method is more efficient for smaller number
of degrees of freedom. Application to the Wess–Zumino quantum mechanics and D = 2
Yang–Mills quantum mechanics give quite satisfactory and quantitative results including
the new intriguing scheme which preserves exact supersymmetry for finite cut-off. For
more complex, and correspondingly richer, D = 4 SYMQM the method is able to provide
new results including detailed information about the supersymmetric structure of the
spectrum and the observables. Obviously we also see a room for improvement which is
especially needed in the continuum sector. However, and that we find most important,
further increase of the size of the Fock space is possible within the available technology.
The whole programme can be (and is being) implemented in the standard, compiler based
languages which usually improves performance by a factor 10–100. Further, the action
of the quantum operators can be optimized taking into account symmetries of the states.
Finally, one can go for more powerful computers.
Taking all above into account, one can reasonably expect substantial improvement in
the quality of the present D = 4, N = 2 results. At the same time one should be able to
study D = 4 systems with higher N . To answer the main question: yes, we think that the
quantitative study of the D = 10 theory is feasible, and reaching current quality results
for the D = 10 is realistic, beginning with the SU(2) gauge group. D = 10 Hamiltonian
and spin(9) generators do not conserve fermionic number [7,15]. The reason for this
complication should be better understood in the first place.
Apart from increasing the number of degrees of freedom, higher gauge groups pose
an interesting problem of constructing gauge invariant states. This has already been done
in the small volume approach for SU(3) [49], and should not present any fundamental
problem for higher N as well. At the same time the possibility of some large N
simplification, specifically within the present approach, should be investigated.
Last, but not least, we would like to mention a host of applications of this method to
other quantum mechanical systems. For example, one may simply extend the 0-volume
calculations to the full non-supersymmetric QCD with dynamical quarks in fundamental
representation. Apart from known glueballs, this would give us a spectrum of quark-made-
hadrons in the “femtouniverse”.
As another rather different application, we mention that this program is already
being used as a routine in solving, to a high precision, quantum mechanics of a simple
two-dimensional building blocks of a prototype quantum computer [50]. In particular,
a complete quantum evolution in time with the time dependent Hamiltonian, was
straightforward to simulate.
The D = 4, N = 2 SYMQM studied here has 15 degrees of freedom. There are many
unsolved quantum mechanical systems with this or smaller complexity. Present approach
should be well applicable in some of these cases.
Obviously there are many routes which can be followed from this point and we are
looking forward to explore some of them.
Note added in proof
By the time this article was in print, a faster recursive method of calculating matrix
elements has been developed [53].
Acknowledgements
I would like to thank C.M. Bender for an instructive discussion which inspired this
approach. I also thank P. van Baal, P. Breitenlohner, L. Hadasz, M. Rostworowski, H. Saller
for intensive discussions. This work is supported by the Polish Committee for Scientific
Research under the grant no. PB 2P03B01917.
References
[1] T. Banks, W. Fishler, S. Shenker, L. Susskind, Phys. Rev. D 55 (1997) 6189, hep-th/9610043.
[2] E. Witten, Nucl. Phys. B 185/188 (1981) 513.
[3] F. Cooper, A. Khare, U. Sukhatme, Phys. Rep. 251 (1995) 267, hep-th/9405029.
[4] M. Claudson, M.B. Halpern, Nucl. Phys. B 250 (1985) 689.
[5] S. Samuel, Phys. Lett. B 411 (1997) 268, hep-th/9705167.
[6] U.H. Danielsson, G. Ferretti, B. Sundborg, Int. J. Mod. Phys. A 11 (1996) 5463, hep-th/9603081.
[7] M.B. Halpern, C. Schwartz, Int. J. Mod. Phys. A 13 (1998) 4367, hep-th/9712133.
[8] B. de Wit, M. Lüscher, H. Nicolai, Nucl. Phys. B 320 (1989) 135.
[9] H. Nicolai, R. Helling, hep-th/9809103, in: Trieste 1998, Non-perturbative aspects of strings, branes and
supersymmetry, pp. 29–74.
[10] M. Lüscher, Nucl. Phys. B 219 (1983) 233.
[11] J. Polchinski, String Theory, Cambridge Univ. Press, Cambridge, 1998.
[12] P. Yi, Nucl. Phys. B 505 (1997) 307, hep-th/9704098.
[13] S. Sethi, M. Stern, Commun. Math. Phys. 194 (1998) 675, hep-th/9705046.
[14] A.V. Smilga, Nucl. Phys. B 266 (1986) 45.
[15] V.G. Kac, A.V. Smilga, Nucl. Phys. B 571 (2000) 515, hep-th/9908096.
[16] G. Moore, N. Nekrasov, S. Shatashvili, Commun. Math. Phys. 209 (2000) 77, hep-th/9803265.
[17] M.B. Green, M. Gutperle, JHEP 01 (1998) 005, hep-th/9711107.
[18] F. Sugino, Int. J. Mod. Phys. A 14 (1999) 3979, hep-th/9904122.
[19] W. Krauth, H. Nicolai, M. Staudacher, Phys. Lett. B 431 (1998) 31, hep-th/9803117.
[20] W. Krauth, M. Staudacher, Nucl. Phys. B 584 (2000) 641, hep-th/0004076.
[21] D. Kabat, G. Lifschytz, Nucl. Phys. B 571 (2000) 419, hep-th/9910001.
[22] D. Kabat, G. Lifschytz, D.A. Lowe, Phys. Rev. D 64 (2001) 124015, hep-th/0105171.
[23] E. Martinec, hep-th/9909049, in: Cargesse 1999, Progress in string theory and brane theory, pp. 117–145.
[24] R.A. Janik, J. Wosiek, Acta Phys. Pol. B 32 (2001) 2143, hep-th/9903121.
[25] P. Bialas, J. Wosiek, Nucl. Phys. B (Proc. Suppl.) 106 (2002) 968, hep-lat/0111034.
[26] T. Eguchi, H. Kawai, Phys. Rev. Lett. 48 (1982) 1063.
[27] A. Gonzalez-Arroyo, J. Jurkiewicz, C.P. Khortals, Altes, in: J. Honercamp, et al. (Eds.), Proceedings of 11th
NATO Summer Institute, Plenum, New York, 1982.
[28] N. Ishibashi, H. Kawai, Y. Kitazawa, A. Tsushijya, Nucl. Phys. B 498 (1997) 467.
[29] H. Aoki, et al., Prog. Theor. Phys. Suppl. 134 (1999) 47, hep-th/9908038.
[30] J. Ambjorn, et al., JHEP 0007 (2000) 011, hep-th/0005147.
[31] J. Ambjorn, K.N. Anagnostopoulos, A. Krasnitz, JHEP 0106 (2001) 069, hep-ph/0101309.
[32] M. Lüscher, G. Münster, Nucl. Phys. B 232 (1984) 445.
[33] P. van Baal, Acta Phys. Pol. B 20 (1989) 295.
[34] P. van Baal, in: M. Shifman (Ed.), in: At the Frontiers of Particle Physics—Handbook of QCD, Boris Ioffe
Festschrift, Vol. 2, World Scientific, Singapore, 2001, p. 683, hep-ph/0008206.
[35] P. van Baal, hep-th/0112072, in: M. Olshanetsky, A. Vainstein (Eds.), The Michael Marinov Memorial
Volume Multiple Facets of Quantization and Supersymmetry, World Scientific, in press.
[36] C.M. Bender, et al., Phys. Rev. D 32 (1985) 1476.
[37] C.M. Bender, K.A. Milton, Phys. Rev. D 34 (1986) 3149.
[38] J. Kogut, L. Susskind, Phys. Rev. D 11 (1975) 395.
[39] Y. Matsumura, N. Sakai, T. Sakai, Phys. Rev. D 52 (1995) 2446.
[40] J.R. Hiller, S. Pinsky, U. Trittmann, hep-th/0112151.
[41] J.R. Hiller, S.S. Pinsky, U. Trittmann, hep-th/0106193.
[42] M.A. Shifman, ITEP Lectures on Particle Physics and Field Theory, World Scientific, Singapore, 1999.
[43] P. Jordan, E.P. Wigner, Z. Phys. 47 (1928) 631.
[44] J.D. Bjorken, S.D. Drell, Relativistic Quantum Fields, McGraw–Hill, New York, 1965.
[45] A. Nakamura, F. Palumbo, Phys. Lett. B 135 (1984) 96.
[46] J. Trzetrzelewski, J. Wosiek, in preparation.
[47] C. Itzykson, J.-B. Zuber, Quantum Field Theory, McGraw–Hill, New York, 1980.
[48] S. Weinberg, The Quantum Theory of Fields III—Supersymmetry, Cambridge Univ. Press, Cambridge,
2000.
[49] P. Weisz, V. Ziemann, Nucl. Phys. B 284 (1987) 157.
[50] V. Corato, et al., cond-mat/0205514.
[51] J. Kotanski, J. Wosiek, hep-lat/0208067, in: Proceedings of the XX Symposium on Lattice Field Theory,
June 2002, MIT, Cambridge, MA, in press.
[52] J. Wosiek, hep-th/0204243, in: Proceedings of the NATO Workshop QCD, Stara Lesna, Slovakia, January
2002, in press.
[53] M. Campostrini, J. Wosiek, hep-th/0209140.
Supermembrane on the pp-wave background

Katsuyuki Sugiyama a , Kentaroh Yoshida b
a Department of Fundamental Sciences, Faculty of Integrated Human Studies, Kyoto University,
Kyoto 606-8501, Japan
b Graduate School of Human and Environmental Studies, Kyoto University, Kyoto 606-8501, Japan
Received 13 June 2002; received in revised form 8 August 2002; accepted 2 September 2002
Abstract
We study the closed and open supermembranes on the maximally supersymmetric pp-wave
background. In the framework of the membrane theory, the superalgebra is calculated by using the
Dirac bracket and we obtain its central extension by surface terms. The result supports the existence
of the extended objects in the membrane theory in the pp-wave limit. When the central terms are
discarded, the associated algebra completely agrees with that of Berenstein–Maldacena–Nastase
matrix model. We also discuss the open supermembranes on the pp-wave and elaborate the possible
boundary conditions.
Keywords: Supermembranes; Matrix theory; M-theory; pp-waves
1. Introduction
It is considered that the supermembrane theory in eleven dimensions should describe

the M-theory [1–3] (for the review, see [4–10]). For the past years, many works toward
the investigation of the M-theory have been done, and in particular the matrix model
approach, which comes from the supermembrane theory via the matrix regularization
procedure [3], seems greatly successful [11]. These are attempts to describe the scattering
in the eleven-dimensional supergravity theory. For many years, the eleven-dimensional
supergravity backgrounds have been studied, and the Minkowski-space, AdS4 × S 7 ,
AdS7 × S 4 and Kowalski-Glikman pp-wave solution [12] are only known cases as the
maximally supersymmetric backgrounds. They are possible candidates for the M-theory
E-mail addresses: sugiyama@phys.h.kyoto-u.ac.jp (K. Sugiyama), yoshida@phys.h.kyoto-u.ac.jp

(K. Yoshida).
PII: S 0 5 5 0 - 3 2 1 3 ( 0 2 ) 0 0 7 9 4 - 0
114 K. Sugiyama, K. Yoshida / Nuclear Physics B 644 (2002) 113–127
backgrounds. In particular, by taking a certain limit called Penrose limit [13,14], the
maximally supersymmetric pp-wave solution can be obtained from the AdS4 × S 7 or
AdS7 × S 4 backgrounds [15]. Also, the maximally supersymmetric IIB supergravity
background has been lately found [16] and it has been shown that the Green–Schwarz
type IIB superstring theory on the pp-wave is exactly solvable [17–19]. The pp-wave
background used in the works [17–19] can be also obtained by taking Penrose limit in
the AdS5 × S 5 [16].
The fact that the maximally supersymmetric pp-wave backgrounds are obtained by
taking the Penrose limit in the AdS background leads to the work [20], where the IIB
string theory on the pp-wave is used for investigating the AdS/CFT correspondence [21,
22] in the string theoretic analysis. That is the exactly solvable model with nontrivial
string background and it provides an interesting area to study properties of strings with
background fluxes. Moreover, the matrix model on the maximally supersymmetric pp-wave
has been proposed from the considerations for the superparticles [20]. The action of the
matrix model on the pp-wave has been also derived directly from the membrane theory on
the maximally supersymmetric pp-wave [23].
In this paper we consider the closed and open supermembranes on the eleven-
dimensional maximally supersymmetric pp-wave background. We calculate the super-
charges and associated algebra. In the case of the flat space, the correspondence of the
superalgebra has been shown [24]. We will show this correspondence on the pp-wave. In
contrast with the algebra in the matrix model on the pp-wave, surface terms are included in
our membrane case. The resulting algebra is the central extension of the superalgebra on
the pp-wave and we can discuss the extended objects contained in the membrane theory on
the pp-wave.
Next we discuss the boundary conditions for the open supermembrane on the pp-wave
by calculating surface terms under the variations of the supersymmetry transformations.
In the case of flat background, the open supermembrane can end on the p-dimensional
hypersurface only for the values p = 1, 5 and 9. However, we show that some additional
surface terms arise in the pp-wave case and only the value p = 1 is allowed for the open
supermembrane on the pp-wave.
This paper is organized as follows. In Section 2, as a short review we provide an
explanation of the action of the supermembrane and supersymmetries on the maximally
supersymmetric pp-wave background. In Section 3 we will calculate the supercharges
and associated algebra by the use of the Dirac bracket procedure. In order to discuss
the extended objects, we carefully analyze the surface terms. In Section 4, the boundary
conditions for the open supermembranes on the pp-wave will be considered. Section 5 is
devoted to considerations and discussions. In appendix, our notation is summarized.
2. Supermembrane on maximally supersymmetric pp-wave
We consider the (closed and open) supermembranes [1–3] on the eleven-dimensional

maximally supersymmetric pp-waves (Kowalski-Glikman solution) [12]. Its metric is given
K. Sugiyama, K. Yoshida / Nuclear Physics B 644 (2002) 113–127 115
by
2
9
µ 2
ds 2 = −2 dx + dx − + G++ dx + + dx ,
µ=1
2 2
µ 2 µ 2
G++ ≡ − x1 + x2 + x3 +
2 2
x4 + · · · + x9 ,
2
(2.1)
3 6
where the constant 4-form flux for +, 1, 2, 3 directions,
F+123 = µ, (µ = 0) (2.2)
is equipped.
The Lagrangian of supermembrane on the maximally supersymmetric pp-wave is given
as a sum of L0 and Wess–Zumino term LWZ 1

L = L0 + LWZ , L0 = − −g(X, θ ), (2.3)
where the induced metric gij is given by
gij = Πir̂ Πjŝ ηr̂ ŝ , g = det gij (2.4)
and the supervielbein Π A and covariant derivative Di θ for θ are defined by using vielbein
eµ̂r̂ and spin connection ωr̂ ŝ
Π r̂ = dXµ̂ eµ̂r̂ − i θ̄ Γ r̂ Dθ, (2.5)

ᾱ 1
Π ᾱ = (Dθ )ᾱ ≡ dθ ᾱ + er̂ Tr̂ ŝ tû̂v̂ θ Fŝ tû̂v̂ − ωr̂ ŝ (Γr̂ ŝ θ )ᾱ , (2.6)
4
1 ŝ tû̂v̂
Tr̂ ŝ tû̂v̂ = Γr̂ − 8δr̂ [ŝ Γ tû̂v̂] . (2.7)
288
The maximally supersymmetric pp-wave background is achieved by taking Penrose
limit [15] in the AdS4 × S 7 or AdS7 × S 4 where the supervielbeins are given by [25,26]
ᾱ
sinh M
Π = ᾱ
Dθ , (2.8)
M

2 M 2
Π r̂ = dx µ̂ eµ̂r̂ − i θ̄ Γ r̂ sinh Dθ, (2.9)
M 2
1
iM2 = 2 Tr̂ ŝ tû̂v̂ θ Fŝ tû̂v̂ θ̄ Γ r̂ − (Γr̂ ŝ θ ) θ̄ Γ r̂ ŝ tû̂v̂ ŵ Ftû̂v̂ŵ + 24Γtû̂ F r̂ ŝ tû̂ .
288
(2.10)
When we take the light-cone gauge in the Penrose limit, M2 = 0 is satisfied. In addition
the Dθ becomes a simple formula
Dθ = dθ + e+ T+ +123 θ F+123 , (2.11)
1 Our notation and convention are summarized in Appendix A.

and supervielbeins are dramatically simplified, though the action in the AdS background
has nontrivial interaction terms. In this gauge we write down the Wess–Zumino term LWZ
1
LWZ = ! ij k Cµ̂ν̂ ρ̂ ∂i Xµ̂ ∂j Xν̂ ∂k Xρ̂
6

i ij k µ̂ ν̂ µ̂ 1 µ̂
+ ! θ̄ Γµ̂ν̂ Di θ Πj Πk + iΠj θ̄ Γ Dk θ − θ̄ Γ Dj θ θ̄ Γ Dk θ .
ν̂ ν̂
2 3
(2.12)
Here the supervielbeins on the maximally supersymmetric pp-wave are given by Eqs. (2.5)
and (2.11). The Cµ̂ν̂ ρ̂ is the 3-form potential and its field strength is described by Eq. (2.2).
The above supermembrane action is difficult to analyze directly, and so we shall rewrite
Lagrangian (2.3) following the work [3] in the light-cone gauge in terms of SO(9) spinor ψ
3 9
−1 1 1 r s 2 1 µ 2 2 1 µ 2 2
w L = Dτ X Dτ X − X , X
r r
− XI − XI
2 4 2 3 2 6 I =1 I =4
µ
3

− !I J K XK XI , XJ + iψ T γ r Xr , ψ
6
I,J,K=1
µ
+ iψ T Dτ ψ + i ψ T γ123ψ. (2.13)
4
We used a convention P0+ = 1. Here “τ ” is the time coordinate on the worldvolume
and { , } is Lie bracket given by using an arbitrary function w(σ ) of worldvolume spatial
coordinates σ a (a = 1, 2)
1 ab
{A, B} ≡ ! ∂a A∂b B (a, b = 1, 2)
w
with ∂a = ∂σ∂ a . Also this theory has large residual gauge symmetry called the area-
preserving diffeomorphism (APD) and the covariant derivative for this gauge symmetry
is defined by a gauge connection ω

Dτ Xr ≡ ∂τ Xr − ω, Xr . (2.14)
In this model, if we replace the variables in the Lagrangian (2.13) according to the rule as
follows:

X ξ i → X(τ ),

ψ ξ i → ψ(τ ),

d 2 σ w(σ ) → Tr,
{ , } → −i[ , ],
we can obtain the matrix model on the pp-wave [20], starting from the Lagrangian for
supermembrane on the maximally supersymmetric pp-wave.
We have taken the light-cone gauge and so original symmetries are not seen manifestly
but the Lagrangian (2.13) still has residual supersymmetries,
δ! Xr = 2ψ T γ r !(τ ), δ! ω = 2ψ T !(τ ),
i
δ! ψ = −iDτ Xr γr !(τ ) + Xr , Xs γrs !(τ )
2
µ 3
µ I
9
+ i XI γI γ123 !(τ ) − i X γI γ123!(τ ),
3 6
I =1 I =4

µ
!(τ ) = exp γ123τ !0 (!0 : constant spinor). (2.15)
12
These transformation rules are 16 linearly-realized supersymmetries on the maximally
supersymmetric pp-wave. In taking the limit µ → 0, we recover the supersymmetry
transformations on the flat space. In the context of the eleven-dimensional supersymmetry,
this corresponds to the dynamical supersymmetry. The Lagrangian (2.13) has other 16
nonlinearly realized supersymmetries,
δη Xr = 0, δη ω = 0,
δη ψ = η(τ ),

µ
η(τ ) = exp − γ123τ η0 (η0 : constant spinor). (2.16)
4
It corresponds to the kinematical supersymmetry in the eleven-dimensional theory.
3. Supercharge algebra from the supermembrane
To begin, we derive supercharges for the supersymmetries (2.15) and (2.16), and then
study associated superalgebra by the use of the Dirac bracket. We discuss the extended
objects on the pp-wave from the viewpoint of the central charges of the superalgebra.
Supercharges Q+ and Q− of the linearly and nonlinearly realized supersymmetries,
respectively, are obtained as Noether charges

+ µ
− 12 1
Q = d σ w −2e
2 γ123 τ
DXr γr ψ + Xr , Xs γrs ψ
2

µ I µ I
3 9
+ X γI γ123ψ + X γI γ123ψ , (3.1)
3 6
I =1 I =4

−
µ
µ
Q = d 2 σ w −2ie 4 γ123 τ ψ = −2ie 4 γ123 τ ψ0 , (3.2)

where ψ0 is the zero-mode of ψ and we have used the normalization with d 2 σ w(σ ) = 1.
Next we shall calculate the superalgebra satisfied by (3.1) and (3.2). The supermem-
brane theory contains the fermionic field ψ α and this leads to the second class constraint
Ξα ≈ 0 for the theory

∂L
Ξα = Sα − iwψαT ≈0 Sα ≡ = iwψα .
T
(3.3)
∂R (∂0 ψ α )
The fermionic field is the SO(9) spinor with 16 components. We have to deal properly with
this second class constraint by the use of the Dirac bracket. The calculation of the Dirac
bracket needs only a constraint matrix Cαβ

Cαβ ≡ Ξα (σ ), Ξβ (σ ) PB = −2iwδαβ δ (2) (σ − σ ), (3.4)
and its inverse matrix (C −1 )αβ is given by
i
C −1 αβ
= δαβ δ (2) (σ − σ ). (3.5)
2w
By the use of the matrix (C −1 )αβ , we can introduce the Dirac bracket { , }DB in terms of
the Poisson bracket { , }PB
αβ
{F, G}DB ≡ {F, G}PB − {F, Ξα }PB C −1 {Ξβ , G}PB
i
= {F, G}PB − {F, Ξα }PB Ξ α , G PB . (3.6)
2w
Thus, we can define the commutation relations on the bosonic and fermionic fields with
their canonical momenta as follows:
r
X (σ ), Ps (σ ) DB = δsr δ (2) (σ − σ ), (3.7)
1
ψα (σ ), Sβ (σ ) DB = δαβ δ (2)(σ − σ ), (3.8)
2
Pr = wDτ Xr .
The commutation relations (3.7) and (3.8) are rewritten in terms of ψ T and Dτ Xs as
1 r (2)
Xr , Dτ Xs δ δ (σ − σ ),
= (3.9)
DB w s
i
ψα (σ ), ψβT (σ ) DB = − δαβ δ (2) (σ − σ ). (3.10)
2w
The superalgebra is calculated by the use of the Dirac bracket (3.9) and (3.10) and we
obtain the following results

1 − 1 − T
i √ Qα , √ Q β = −δαβ , (3.11)
2 2
DB
1 1 − T
i √ Q+ α,√ Q β
2 2 DB
3
µ µ
=i P0I + X0I γ123 γI e− 3 γ123 τ
3 αβ
I =1
9
µ µ
+i P0I − X0I γ123 γI e− 6 γ123 τ

6 αβ
I =4
3
µ
−i d 2 σ ∂a SIaJ γ I J e− 3 γ123 τ αβ
I,J =1

9
µ
−i d 2 σ ∂a SIa J γ I J e− 3 γ123 τ αβ
I ,J =4
9
3
µ
− 2i d 2 σ ∂a SIaI γ I I e− 6 γ123 τ αβ , (3.12)
I =1 I =4

1 1 + T
i √ Q+ α , √ Q β
2 2 DB
µ IJ µ
3 9

= 2H δαβ + M0 (γI J γ123 )αβ − M0I J (γI J γ123)αβ
3 6
I,J =1 I ,J =4
3
9

µ
−2 d 2 σ ϕXI γ I αβ − 2 d 2 σ ϕXI γ I e 6 γ123 τ αβ
I =1 I =4
3
9

µ
+2 2
d σ ∂a SIa γ I
αβ
+2 d 2 σ ∂a SIa γ I e 6 γ123 τ αβ
I =1 I =4

3
9

+2 d 2 σ ∂a SIaJ I J γ I J I J αβ
I,J =1 I ,J =4

9

+2 d 2 σ ∂a SIa J K L γ I J K L αβ
I ,J ,K ,L =4

3 9
µ
+2 d 2 σ ∂a SIaJ KI γ I J KI e 6 γ123 τ αβ
I,J,K=1 I =4

3
9
µ
+2 d 2 σ ∂a SIaI J K γ I I J K e 6 γ123 τ αβ
I =1 I ,J ,K =4

3
9

+ 2µ d 2 σ ∂a UJaKI J γ J K γ I J αβ
J,K=1 I ,J =4
9
µ
+ 2µ d 2 σ ∂a UIa γ I γ123e 6 γ123 τ αβ . (3.13)
I =4

Here M I J and M I J are defined by
1
M I J ≡ XI P J − P I XJ − Sγ I J ψ, (3.14)
2
1
M I J ≡ XI P J − P I XJ − Sγ I J ψ, (3.15)
2

and the SO(3) × SO(6) Lorentz generators M0I J and M0I J are given as

M0I J ≡ d 2 σ M I J , (3.16)

I J
M0 ≡ d 2 σ M I J . (3.17)
They satisfy the SO(3) × SO(6) Lorentz algebra,

IJ
M0 , M0KL DB = δ I K M0J L − δ I L M0J K − δ J K M0I L + δ J L M0I K , (3.18)
I J K L
I K J L I L J K J K I L J L I K
M0 , M0 DB
=δ M0 − δ M0 −δ M0 + δ M0 . (3.19)
The zero-modes of P r (= wDτ Xr ) and Xr are written by

P0 ≡ d σ wDτ X ,
r 2 r
X0 ≡ d 2 σ wXr .
r
(3.20)
Also, the Hamiltonian H is expressed as

3
1 P r 2 1 r s 2 1 µ 2 I 2
H = d σw2
+ X ,X + X
2 w 4 2 3
I =1
9
1 µ 2 I 2 µ
3

+ X + !I J K XK XI , XJ
2 6
6
I =4 I,J,K=1

µ
− w−1 Sγ123 ψ − w−1 Sγr Xr , ψ . (3.21)
4
Other quantities in the above algebra are defined by
1
a
Srs ≡ − ! ab X[r ∂b Xs] , (3.22)
2

ϕ ≡ w w−1 P r , Xr + iw ψ T , ψ , (3.23)

−1 3 s T
Sr ≡ !
a ab
w Xr Ps ∂b X + Xr iψ ∂b ψ + iX ∂b ψ γrs ψ ,
s T
8
i
u≡
a
Srst ! ab X[r ∂b ψ T γst u] ψ , (3.24)
48
1
3

UJaKI J ≡ − !I J K ! ab XI ∂b XI XJ , (3.25)
12
I =1
3
1 I 2 1 J 2
9
1 ab I
UI ≡ − ! X ∂b
a
X − X .
2 3 6
I =1 J =4
The above superalgebra (other than the central charges) completely agrees with that of the
matrix model on the pp-wave [20].2 Also, in the limit µ → 0, the above algebra realizes
the superalgebra of the supermembrane in the flat space given in [3].
√
2 We can absorb the factor 1/ 2 in front of the supercharges in the definition of the fermion ψ .
Also, the above superalgebra includes some central charges. These charges indicate
the existence of extended objects in the supermembrane theory on the maximally
supersymmetric pp-wave. First, the charges Srs a and S a correspond to the transverse M2-
r
brane (D2-brane in type IIA string theory) and longitudinal M2-brane (fundamental string
a
in type IIA string theory), respectively. Next, Srst u corresponds to the longitudinal M5-
brane charge (D4-brane in type IIA string theory). As is well-known, these charges appear
in the supermembrane theory on the flat eleven-dimensional Minkowski space. In addition,
in our supermembrane theory the superalgebra includes the additional central charges,
UJaKI J and UIa . We do not properly confirm the physical interpretation of these extra
extended objects only living on the pp-wave. These might be related to the fuzzy membrane
and giant graviton discussed in [20], or another new extended object due to a certain kind
of the Myers effects on the pp-wave [27,28].
4. Open supermembrane on pp-wave
In the case of the open membrane, which has the boundary on the worldvolume
toward the spatial directions σ 1 and σ 2 , the surface terms do not vanish automatically.
Thus we must properly treat the total derivative terms under the variation of the above
supersymmetry transformations, and consider the boundary conditions in order for the
surface terms to vanish. Let us recall that the membrane p-branes are allowed for p = 1, 5
and 9 in the flat background due to the boundary conditions [29,30]. The M5-brane
corresponds to p = 5. The case of p = 9 is related to “the end of the world” in Hor̆ava–
Witten’s works [31].
In our pp-wave case, we obtain the total derivative terms for the linear supersymmetry
(2.15) explicitly

1 s t T
w X , Dτ X ψ γs γr !(τ ) + X , X ψ γst γr !(τ )
r s T
2
µ I J T
3

− w X , X ψ γI γJ γ123 !(τ )
3
I,J =1
µ I J T
9

− w X , X ψ γI γJ γ123!(τ )
6
I ,J =4
µ I I T
3 9

+ w X , X ψ γI I γ123!(τ )
3
I =1 I =4
µ I I T
3 9

− w X , X ψ γI I γ123!(τ ) , (4.1)
6
I =1 I =4
and for the nonlinear supersymmetry (2.16), we can calculate the corresponding term

w Xr , iηγr ψ . (4.2)
This surface term for the nonlinear supersymmetry has the same form as in the flat space.
However, some additional terms proportional to µ appear for the linear supersymmetry in
addition to the surface terms in the flat background. The variations of the action under the
linear and nonlinear supersymmetry transformations can be written as
δS = δ! S + δ!(µ) S + δη S,

1
δ! S = − dτ dξ ∂t Xr · Dτ Xs ψ T γs γr !(τ ) + Xs , Xt ψ T γst γr !(τ ) ,
2
∂Σ
(4.3)

µ
3
δ!(µ) S = − dτ dξ − ∂t XI · XJ ψ T γI γJ γ123!(τ )
3
∂Σ I,J =1
µ
9

− ∂t XI · XJ ψ T γI γJ γ123!(τ )
6
I ,J =4
µ
3 9

+ ∂t XI · XI ψ T γI I γ123!(τ )
3 I =1 I =4

µ
3 9
I T
− ∂t X · X ψ γI I γ123!(τ ) ,
I
(4.4)
6
I =1 I =4

δη S = −i dτ dξ ∂t Xr · η(τ )γr ψ, (4.5)
∂Σ
where ∂Σ is the boundary of the open supermembrane worldvolume and ξ is the
coordinate for the tangent direction of the boundary. Note that the tangential derivative
∂t and normal derivative ∂n on the boundary are defined by
∂t Xr ≡ ! ab na ∂b Xr , (4.6)
∂n X ≡ n ∂a X .
r a r
(4.7)
Here na is the unit vector toward the normal direction on the boundary. We would like to
consider the p-dimensional hypersurface (membrane p-brane) on which supermembranes
can end, and investigate the condition that such a surface can exist. First, by following the
discussion of the p-brane in string theory, the boundary conditions for our membrane are
classified
Neumann: ∂n X m = 0 (m = 0, 10 and some p − 1 coordinates), (4.8)
Dirichlet: ∂t X = 0
m
( m = other 10 − p coordinates). (4.9)
By applying these boundary conditions to (4.3) and (4.5), the constraints
η0T γm ψ = !0T γm γn ψ = !0T γm γn γn ψ = 0, (4.10)
can be obtained. These are the same conditions as in the flat case and (4.10) leads us to
the well-known results p = 1, 5 and 9. However, in the pp-wave case we also need to take
account of the constraints coming from the additional surface terms (4.4).
Here, let us define the following operators

1
P± ≡ 1 ± γ m1 γ m2 . . . γ m10−p . (4.11)
2
These are the projection operators if and only if 12 p(p + 1) is odd. Thus, the value of p is
limited to p = 1, 2, 5, 6 and 9. The requirement that boundary term should vanish leads to
the constraints Eq. (4.10), and so it provides a further restriction for the value of p. If we
assume that 1/2 BPS boundary hypersurface, then the condition
P− ψ = 0, (4.12)
is in our hand. Then we can write ψ as ψ = P+ ψ. To begin, from the second equation in
(4.10), P+ !0 = 0 is followed. Next, we can see from the third equation in (4.10) that 9 − p
should be even. As a result, p = 1, 5 and 9 are allowed in the flat case for the boundary
hypersurface. However, the story does not end because the additional boundary terms exist
in the case of the pp-wave. We can easily check whether the additional surface terms (4.4)
vanish or not in each p = 1, 5 and 9 case. In the p = 1 case we can immediately see
that the additional terms (4.4) vanish. Here, it can be seen from the constraints (4.10) that
only the even number of gamma matrices with Neumann indices m’s and arbitrary number
of gamma matrices with Dirichlet indices m’s are allowed to appear between !0T and ψ.
Equivalently, odd number of gamma matrices with Neumann indices m’s cannot appear
between !0T and ψ. However, it is found from the expression (4.4) that such a condition
cannot be satisfied in the cases p = 5 and 9 because there are inevitably several terms
including odd number of Neumann components. In conclusion, only p = 1 is allowed for
membrane p-brane on the pp-wave, and for p = 5 and 9 membrane p-brane cannot exist.
This result would be also plausible from the viewpoint of the chirality matrix [32]. It is
because the flux is turned on the 1, 2, 3 directions on the pp-wave, and so the SO(4) and
SO(8) chirality, which is important for p = 5 and p = 9 cases, cannot be respected. The
reason that p = 1 case is allowed is unknown, since what p = 1 means physically has not
been well understood.
In the above discussion, we have assumed that the 1/2 BPS boundary hypersurface,
that is, the flat boundary hypersurface. However, it might be clear that such flat boundaries
cannot exist, because the pp-wave background is curved. Possibly, the curved hypersurface
as discussed in [33] might become the boundary of the supermembrane. But, we do not
know how to treat such curved boundaries, and do not discuss such a case here.
In this paper, we have studied the supercharges and its associated algebra. In particular,
by treating the surface terms carefully, the central extension of the superalgebra has been
derived. The superalgebra apart from the central charges completely agrees with that of
the matrix model on the pp-wave. The central charges obtained in our derivation realize
the flat space results in the limit µ → 0, and also include some additional ones. We do
not confirm the physical interpretation of the additional central charges. These seem to
indicate the extra extended objects coming from a kind of the Myers effect on the pp-wave
background.
Moreover, we have discussed the boundary conditions of the open supermembrane on
the maximally supersymmetric pp-wave background. It is well-known that the membrane
p-branes in the flat space are allowed to exist only for p = 1, 5 and 9. In the pp-wave
case, more strict constraints for such hypersurfaces arise, and so only the value p = 1
is allowed. In our discussion, we have not included the 2-form which can couple to the
boundary hypersurface. It might be possible by turning on the 2-form on the boundary that
5- and 9-dimensional hypersurfaces exist as the boundaries of the open supermembranes
on the pp-wave.
In this paper, we have used the SO(9) formulation for simplicity, but it is also
interesting to work in the SO(10, 1) covariant formulation, where the nature of longitudinal
components are clear and more definite considerations would be possible. This is an
interesting future work.
Acknowledgements
The work of K.S. is supported in part by the Grant-in-Aid from the Ministry of
Education, Science, Sports and Culture of Japan (? 14740115).
Appendix A
In this appendix, we summarize several notations used in the paper.
A.1. Notation
We consider the supermembrane in the eleven-dimensional curved spacetime and use a

notation of supercoordinates (D = 11):

XM = Xµ̂ , θ α , µ̂ = (+, −, µ), µ = 1, . . . , D − 2.
The background metric is expressed by GMN .
In the Lorentz frame, we also use the coordinates (D = 11):

XA = Xr̂ , θ ᾱ , r̂ = (+, −, r), r = 1, . . . , D − 2.
The metric is flat and is described by ηAB . In these notations, we introduced a set of light-
cone coordinates X± ≡ √1 (X0 ± XD−1 ).
2
The membrane has three-dimensional worldvolume and its coordinates are parameter-
ized by ξ i = (τ, σ a ), a = 1, 2, and its metric is given by gij .
Next we shall summarize the SO(10, 1) gamma matrices (D = 11):
µ̂ ν̂ r̂ ŝ
Γ , Γ = 2Gµ̂ν̂ , Γ , Γ = 2ηr̂ ŝ ,
µ̂
Γ µ̂ ≡ er̂ Γ r̂ , Γ r̂ ≡ eµ̂r̂ Γ µ̂ ,

0 γµ
Γ µ = γ µ ⊗ σ3 = (real symmetric),
−γ µ0

0 −I16
Γ 0 = 1 ⊗ (−iσ2 ) = (real skew symmetric),
I16 0

0 I16
Γ 10 = 1 ⊗ σ1 = (real symmetric),
I16 0
1 + −
Γ ± ≡ √ Γ 0 ± Γ 10 , Γ ,Γ = −2I32 ,
2

+
√ 0 0 −
√ 0 −I16
Γ = 2 , Γ = 2 .
I16 0 0 0
We take the light-cone gauge and decompose the 32 component SO(10, 1) spinor θ in
terms of SO(9) spinor ψ with 16 components as follows:
X+ = τ, Γ +θ = 0 (θ̄ Γ + = 0),

1 0
⇒ θ = 1/4 ,
2 w ψ
1
θ̄ = θ T −Γ 0 = − 1/4 ψ T , 0 .
2 w
In the light-cone gauge, there are several useful identities
θ̄ Γ r̂ ∂i θ = 0 (for r̂ = −),
θ̄ Γrs ∂i θ = 0,
θ̄ Γ +r ∂i θ = 0,
θ̄ Γ +− ∂i θ = 0.
In the pp-wave background, the vielbein is calculated as
+ − − + 1
eµ̂r̂ : e+ = e− = 1, e+ = 0,
= − G++ , eµr = δµr ,
e−
4
µ̂ + − + − 1 µ µ
er̂ : e+ = e− = 1, e− = 0, e+ = G++ , er = δr ,
4
eµ̂r̂ : e++ = e+r = e−r = eµ+ = eµ− = 0,
1
e+− = e−+ = −1, e−− = − G++ , eµr = δ µr ,
4
eµ̂r̂ : eµ+ = eµ− = e−− = e+r = e−r = 0,
1
e+− = e−+ = −1, e++ = + G++ , eµr̂ = δµr ,
4
and the spin connection is evaluated as
1
ωr̂ ŝ ≡ ωµ̂r̂ ŝ dx µ̂ ⇒ ωr− = ∂ r G++ dx + , otherwise = 0,
4
1
⇒ ω = ∂ µ G++ dx + , otherwise = 0.
µ̂
ωµ̂ν̂ ≡ er̂ eŝν̂ ωr̂ ŝ µ−
4
References
[1] E. Bergshoeff, E. Sezgin, P. Townsend, Supermembranes and eleven-dimensional supergravity, Phys.

Lett. 189 (1987) 75.
[2] E. Bergshoeff, E. Sezgin, P. Townsend, Properties of the eleven-dimensional supermembrane theory, Ann.
Phys. 185 (1988) 330.
[3] B. de Wit, J. Hoppe, H. Nicolai, On the quantum mechanics of supermembranes, Nucl. Phys. B 305 (1988)
545.
[4] P.K. Townsend, Three Lectures on Supermembranes, in: Superstrings’ 88.
[5] M. Duff, Supermembranes, hep-th/9611203.
[6] H. Nicolai, R. Helling, Supermembranes and M(atrix) theory, hep-th/9809103.
[7] B. de Wit, Supermembranes and super matrix models, hep-th/9902051.
[8] B. de Wit, Supermembranes in curved superspace and near-horizon geometries, hep-th/9902149.
[9] W. Taylor, M(atrix) theory: matrix quantum mechanics as a fundamental theory, Rev. Mod. Phys. 73 (2001)
419, hep-th/0101126.
[10] A. Dasgupta, H. Nicolai, J. Plefka, An introduction to the quantum supermembrane, Grav. Cosmol. 8 (2002)
1, hep-th/0201182.
[11] T. Banks, W. Fischler, S.H. Shenker, L. Susskind, M-theory as a matrix model: conjecture, Phys. Rev. D 55
(1997) 5112, hep-th/9610043.
Dordrecht, 1976, pp. 271–275.
[14] R. Güven, Plane wave limits and T-duality, Phys. Lett. B 482 (2000) 255, hep-th/0005061.
[15] J. Figueroa-O’Farrill, G. Papadopoulos, Homogeneous fluxes, branes and a maximally supersymmetric
solution of M-theory, JHEP 0108 (2001) 036, hep-th/0105308.
[16] M. Blau, J. Figueroa-O’Farrill, C. Hall, G. Papadopoulos, A new maximally supersymmetric background of
IIB superstring theory, JHEP 0201 (2001) 047, hep-th/0110242;
M. Blau, J. Figueroa-O’Farrill, C. Hall, G. Papadopoulos, Penrose limits and maximal supersymmetry, hep-
th/0201081.
Phys. B 625 (2002) 70, hep-th/0112044.
background, Phys. Rev. D 65 (2002) 126004, hep-th/0202109.
[19] J.G. Russo, A.A. Tseytlin, On solvable models of type IIB superstring in NS–NS and R–R plane wave
backgrounds, JHEP 0204 (2002) 021, hep-th/0202179.
[20] D. Berenstein, J. Maldacena, H. Nastase, Strings in flat space and pp-waves from N = 4 super-Yang–Mills,
JHEP 0204 (2002) 013, hep-th/0202021.
[21] J. Maldacena, The large N limit of superconformal field theories and supergravity, Adv. Theor. Math. Phys. 2
(1998) 231;
J. Maldacena, The large N limit of superconformal field theories and supergravity, Int. J. Theor. Phys. 38
(1999) 1113, hep-th/9711200.
[22] O. Aharony, S. Gubser, J. Maldacena, H. Ooguri, Y. Oz, Large N field theories, string theory and gravity,
Phys. Rep. 323 (2000) 183, hep-th/9905111, and references therein.
[23] K. Dasgupta, M.M. Sheikh-Jabbari, M. Van Raamsdonk, Matrix perturbation theory for M-theory on a
pp-wave, JHEP 0205 (2002) 056, hep-th/0205185.
[24] T. Banks, N. Seiberg, S. Shenker, Branes from matrices, Nucl. Phys. B 490 (1997) 91, hep-th/9612157.
[25] P. Claus, Super-M-brane actions in AdS4 × S 7 and AdS7 × S 4 , Phys. Rev. D 59 (1999) 066003, hep-
th/9809045.
[26] B. de Wit, K. Peeters, J. Plefka, A. Sevrin, The M-theory two-brane in AdS4 × S 7 and AdS7 × S 4 , Phys.
Lett. B 443 (1998) 153, hep-th/9808052.
[28] B. Janssen, Y. Lozano, On the dielectric effect for gravitational waves, hep-th/0205254.
[29] K. Ezawa, Y. Matsuo, K. Murakami, Matrix regularization of an open supermembrane—towards M-theory
five-branes via open supermembranes, Phys. Rev. D 57 (1998) 5118, hep-th/9707200.
[30] B. de Wit, K. Peeters, J. Plefka, Open and closed supermembranes with winding, Nucl. Phys. (Proc.
Suppl.) 68 (1998) 206, hep-th/9710215.
[31] P. Hor̆ava, E. Witten, Heterotic and type I string dynamics from eleven dimensions, Nucl. Phys. B 460 (1996)
506, hep-th/9510209;
P. Hor̆ava, E. Witten, Eleven-dimensional supergravity on a manifold with boundary, Nucl. Phys. B 475
(1996) 94, hep-th/9603142.
[32] K. Becker, M. Becker, Boundaries in M-theory, Nucl. Phys. B 472 (1996) 221, hep-th/9602071.
[33] D. Bak, Supersymmetric branes in pp-wave background, hep-th/0204033.
Type IIA string and matrix string on pp-wave

Katsuyuki Sugiyama a , Kentaroh Yoshida b
a Department of Fundamental Sciences, Faculty of Integrated Human Studies, Kyoto University,
Kyoto 606-8501, Japan
b Graduate School of Human and Environmental Studies, Kyoto University, Kyoto 606-8501, Japan
Received 9 August 2002; received in revised form 30 August 2002; accepted 9 September 2002
Abstract
We study type IIA string theories on the pp-waves with 24 supercharges. The type IIA pp-
wave backgrounds are derived from the maximally supersymmetric pp-wave solution in eleven
dimensions through the toroidal compactification on the spatial isometry directions. The associated
actions of type IIA strings are obtained by using these metrics and other background fields of the
type IIA supergravities on the one hand. On the other hand, we derive these theories from D = 11
supermembrane on the pp-wave via double-dimensional reduction for the spatial isometry directions.
The resulting actions agree with those of type IIA strings obtained in the study of the supergravities.
Also, the action of the matrix string is written down. Moreover, the quantization of closed and open
strings is discussed. In particular, we study Dp-branes allowed in one of the type IIA theories.
PACS: 11.25.-w; 04.65.+e; 12.60.Jv; 11.25.Sq
Keywords: Supermembrane; Matrix theory; M-theory; pp-wave; Double-dimensional reduction; Matrix string
1. Introduction
The maximally supersymmetric pp-wave background in eleven dimensions is a

classical solution (Kowalski–Glikman (KG) solution) [1] of the eleven-dimensional
supergravity and is considered as one of the candidates for supersymmetric background
of M-theory [2]. This pp-wave background is obtained from AdS7 × S 4 or AdS4 × S 7
via Penrose limit [3]. Recently, the maximally supersymmetric type IIB pp-wave [4] has
been found and it has been shown that the type IIB string on this pp-wave is exactly
E-mail addresses: sugiyama@phys.h.kyoto-u.ac.jp (K. Sugiyama), yoshida@phys.h.kyoto-u.ac.jp

(K. Yoshida).
PII: S 0 5 5 0 - 3 2 1 3 ( 0 2 ) 0 0 8 2 0 - 9
Table 1
Maximal and less supersymmetric pp-waves: the less supersymmetric pp-waves are obtained by compactifica-
tions of the maximal pp-waves and the T-duality. The circles indicate the known solutions, ×’s that no such
solutions exist, and blank that it is not yet known whether such solutions exist. The superscript “∗” denotes there
are no supersymmetric D-branes
SUGRA 16 18 20 22 24 26 28 30 32
11 dim ◦ ◦ ◦ ◦ ◦ ◦ × × unique
Type IIA ◦ ◦ ◦ ◦ ◦ ◦∗ × ×
Type IIB ◦ ◦ ◦ ◦ ◦ unique
solvable in the Green–Schwarz (GS) formulation [5–7] with a light-cone gauge. This
pp-wave background [4] is also obtained from the AdS5 × S 5 via Penrose limit [3].
With this progress, the intensive studies of strings on the pp-waves were initiated. In
particular, this type IIB string has been combined with the AdS/CFT correspondence
and the almost BPS sector of a large N gauge theory has been studied [8]. Moreover,
the matrix model on the KG background has been proposed [8]. This model is often
referred as the Berenstein–Maldacena–Nastase (BMN) matrix model. As the de Wit–
Hoppe–Nicolai (dWHN) supermembrane [9–11] is closely related to the Banks–Fischler–
Shenker–Susskind (BFSS) matrix model [12] in the flat space, the BMN matrix model is
also intimately related to a supermembrane on the pp-wave [13–15]. In our previous works
[14,15], we have shown that the algebra of supercharges in the supermembrane theory on
the pp-wave agrees with that of the BMN matrix model in the same manner as the flat
space [16]. We have also discussed BPS conditions in the supermembrane on the pp-wave.
BPS multiplets in the BMN matrix model are also widely studied [13,17]. Moreover,
the classical solutions of the BMN matrix and the supermembrane are intensively
researched [18,19]. In particular, we have lately investigated the quantum stability of giant
gravitons [19], which are classical solutions of the BMN matrix model and exist due to the
presence of the constant 4-form flux [20].
With recent progress, less supersymmetric type IIB and IIA pp-wave backgrounds,
or strings on these pp-waves are greatly focused [21–28]. The maximal and less
supersymmetric pp-wave backgrounds of the eleven-dimensional supergravity, type IIA
and IIB theories are listed in Table 1 as far as we know. Motivated by these attempts,
we consider the type IIA strings on the pp-waves from two viewpoints in this paper.
On the one hand, we study the type IIA pp-waves and strings from the supergravity side
through the toroidal compactification. On the other hand, we use the double-dimensional
reduction (DDR) [29] for the supermembrane action on the maximally supersymmetric
pp-wave. Both results are equivalent as expected. We show that both compactifications
are done for a spatial isometry direction, which can be found in the same way as in the
type IIB case [21]. When we compactify this spatial direction, 8 supercharges are inevitably
broken. Therefore, the resulting type IIA theory has 24 supercharges, and is not maximally
supersymmetric. The type IIA string on this pp-wave is also exactly-solvable but it is
different from the one obtained from a type IIB string theory via the T-duality [28]. This
comes from the fact that the type IIA pp-wave with 24 supercharges is not unique and
the type IIA pp-wave considered in this paper is different from the one in [28]. Moreover,
the matrix string theory is considered. We also discuss the quantization of closed and open
strings in our type IIA theory. There we study the allowed Dp-branes in the theory. The
values p = 2, 4, 6 and 8 are allowed but the directions of D-branes are restricted as in the
case of type IIB string on the pp-wave [30–32].
This paper is organized as follows. In Section 2 we consider the type IIA pp-wave
backgrounds and actions of strings from two viewpoints. One is based on the analysis in
the supergravity and the other is based on the double-dimensional reduction. We will show
both results are equivalent. In Section 3 we consider the matrix string on the pp-wave and
formally write down the action of the matrix string from the supermembrane action on
the pp-wave in eleven dimensions. In Section 4 we will discuss the mode-expansions and
quantization of closed and open strings in the type IIA theory. We also discuss Dp-branes
and investigate the allowed value p and the direction of D-branes. Section 5 is devoted to
conclusions and discussions. In Appendix A we will briefly explain the compactification
on an SO(3)-direction. The different points from the SO(6)-case considered in the text are
summarized.
2. Type IIA strings on pp-wave
2.1. Type IIA pp-wave solution from KG solution
We can consider the toroidal compactification of a spatial isometry direction in the

eleven-dimensional maximally supersymmetric pp-wave (KG solution) given by
2
9
r 2
ds 2 = −2 dX+ dX− + G++ XI , XI dX+ + dX ,
r=1
3 2
µ 2 I 2
9
I I µ I 2
G++ X , X ≡ − X + X , (2.1)
3 6
I =1 I =4
where the constant 4-form flux for +, 1, 2, 3 directions,
F+123 = µ (µ = 0) (2.2)
is equipped. It is a unique pp-wave solution with 32 supercharges in eleven dimensions.
The Killing vectors of the KG solution are constructed as follows [2]:
ξe+ = −∂+ , ξe− = ∂− ,

µ + µ µ +
ξeI = − cos X ∂I + XI sin X ∂− (I = 1, 2, 3),
3 3 3
2
µ µ + µ µ +
ξeI = − sin
∗ X ∂I − I
X cos X ∂− ,
3 3 3 3

µ + µ µ +
ξeI = − cos X ∂I + XI sin X ∂− (I = 4, . . . , 9),
6 6 6
2
µ µ + µ µ +
ξe∗ = − sin X ∂I − XI cos X ∂− ,
I 6 6 6 6
ξMI J = XI ∂J − XJ ∂I (I, J = 1, 2, 3),

I J
ξMI J = X ∂J − X ∂I (I , J = 4, . . . , 9).
We utilize the procedure used in deriving the type IIA pp-wave from the maximally
supersymmetric type IIB pp-wave through the T-duality [21]. The spatial isometries are
given by
3 6
ξeI ± ξe∗ and ξeI ± ξe∗ .
µ J µ J
It is sufficient to consider only two cases; ξe1 + (3/µ)ξe2∗ (SO(3)-direction) and ξe4 +
(6/µ)ξe5∗ (SO(6)-direction) due to the SO(3) × SO(6) symmetry of the KG background.
The resulting type IIA pp-wave background has 24 supercharges since 8 supercharges are
inevitably broken in the toroidal compactification on the spatial isometry direction in the
same way as the construction of type IIA pp-wave from type IIB via T-duality. In below,
we will discuss mainly the SO(6)-direction case. The case of SO(3) is discussed in Appen-
dix A, since the story is very similar to the SO(6) case though there are a few differences,
such as the field contents.
Let us discuss the Killing spinor in the above type IIA pp-wave. The Killing spinor in
the KG solution is given by [2]

µ µ
(ψ+ , ψ− ) = exp − X+ I ψ− + exp − X+ I ψ+
4 12
3
µ I 1 I
9
µ
+ X ΓI − X ΓI I exp + X+ I Γ− ψ+ , (2.3)
6 2 12
I =1 I =4
where Γµ ’s are 32 × 32 gamma matrices and I ≡ Γ123 obeys I 2 = −1. The spinors ψ+
and ψ− with 32 components satisfy the conditions
Γ+ ψ+ = 0, Γ− ψ− = 0, (2.4)
hence they have 16 non-vanishing components. If X is a Killing vector, we can define an
associated Lie derivative LX on any spinor ψ by
1
LX ψ = XM ∇M ψ + ∇[M XN] Γ MN ψ, (2.5)
4
where ∇ is defined by
1 ab
∇M ≡ ∂M + ωM Γab .
4
This has the following properties:
1. If X is a Killing vector field, f is any smooth function and ψ is any spinor, then
LX (f ψ) = (Xf )ψ + f LX ψ.
2. When the symbol “·” (dot) denotes the Clifford action of vector fields on spinors, then
LX (Y · ψ) = [X, Y ] · ψ + Y · LX ψ.
3. If X, Y are Killing vector fields and ψ is any spinor, then

LX LY ψ − LY LX ψ = L[X,Y ] ψ.
The Lie derivatives for the Killing vector fields are given by [2]

µ µ
Lξe− (ψ+ , ψ− ) = 0, Lξe+ (ψ+ , ψ− ) = − I ψ+ , − I ψ− ,
12 4

µ
LξeI (ψ+ , ψ− ) = − I ΓI Γ− ψ+ , 0 ,
6

µ
Lξe (ψ+ , ψ− ) = − I ΓI Γ− ψ+ , 0 ,
I 12

µ2
Lξe∗ (ψ+ , ψ− ) = − ΓI Γ− ψ+ , 0 ,
I 18

µ2
Lξe∗ (ψ+ , ψ− ) = − ΓI Γ− ψ+ , 0 ,
I 72

1
LξMI J (ψ+ , ψ− ) = − ΓI J ψ+ , 0 ,
2

1
LξM (ψ+ , ψ− ) = − ΓI J ψ+ , 0 .
I J 2
By the use of the above results, we can count the remaining unbroken supersymmetries.
For example, in the case of ξeI + (3/µ)ξeJ∗ , we obtain the following expression

µ
LξeI +(3/µ)ξe∗ = − QΓJ Γ− ψ+ , 0 ,
J 3
1
QI J ≡ (1 + I ΓI ΓJ ). (2.6)
2
Clearly, 16 spinors are annihilated by Γ− . Furthermore, the constant matrix Q plays the
role of the projection operator and so annihilates additional 8 spinors in the same manner
as in the type IIB string [21]. For another example, in the case of ξeI + (6/µ)ξe∗ , the Lie
J
derivative is given by

µ 1
Lξe +(6/µ)ξe∗ = − QΓJ Γ− ψ+ , 0 , QI J ≡ (1 + I ΓI ΓJ ). (2.7)
I J 6 2
In the same way as in the case of ξeI + (3/µ)ξeJ∗ , 24 supersymmetries are preserved.
In conclusion, the above two cases of the type IIA pp-wave backgrounds preserve 24
supersymmetries.
2.2. Type IIA string from pp-wave solution in D = 11 supergravity
Here, we shall consider the toroidal compactification on an SO(6) direction. Let us

transform the variables Xr ’s (r = +, −, 1, . . . , 9) into x’s
µ
X+ = x + , X− = x − + x 4 x 5 ,
6
XI = x I (I = 1, 2, 3), Xa = x a (a = 6, 7, 8, 9),

µ + µ +
X = x cos
4 4
x − x sin
5
x ,
6 6

µ + µ +
X = x sin
5 4
x + x cos
5
x , (2.8)
6 6
then the metric is rewritten as
2 9
r 2 2 5 + 4
ds 2 = −2 dx + dx − + G++ x I , x a dx + + dx − µx dx dx ,
3
r=1
3 2
µ 2 I 2
9
µ a 2
G++ x I , x a ≡ − x + x , (2.9)
3 6
I =1 a=6
but the constant 4-form flux is still expressed in Eq. (2.2). We can easily see from the above
metric (2.9) that the x 4 -direction is a manifest spatial isometry direction [21] and obtain
the metric of the type IIA by the standard technique of the dimensional reduction from the
eleven-dimensional supergravity to the type IIA supergravity in ten dimensions,
4 2
= e− 3 φ gµν dx µ dx ν + e 3 φ dy + dx µ Aµ ,
2
2
ds11 (2.10)
where gµν is a ten-dimensional metric, Aµ is a Kaluza–Klein gauge field (RR 1-form) and
φ is a dilaton. The ten-dimensional metric gµν is given by
2
4
a 2 8
b 2
gµν dx µ dx ν = −2 dx + dx − + g++ x a , x b dx + + dx + dx ,
a=1 b=5
4 2
µ 2 a 2
8
a b µ b 2
g++ x , x ≡ − x + x , (2.11)
3 6
a=1 b=5
where we have made rearrangement of 8 coordinates x 1 , x 2 , x 3 , x 5 , . . . , x 9 into x 1 , . . . , x 8 .

The Kaluza–Klein gauge field Aµ is expressed by
µ
A+ = − x 4 , Ai = 0 (i = 1, . . . , 8), (2.12)
3
and the RR 3-form Cµνρ is given by
µ
C+I J = − I J K x K (I, J, K = 1, 2, 3). (2.13)
3
The dilaton φ and NS–NS 2-form Bµν vanish.
Next, we discuss the action of the type IIA string theory on the above pp-wave. In
below, we shall construct the action of bosonic and fermionic sectors, respectively.
2.2.1. Bosonic sector

In general, the bosonic world-sheet action with non-zero NS–NS B-field is written as

2πL
1
SB = − dτ dσ gµν ∂a x µ ∂b x ν ηab + Bµν ∂a x µ ∂b x ν ab , (2.14)

4πα
0
where the subscript a denotes the coordinates of the string world-sheet σ a = (τ, σ ) and
η = diag(−1, 1) is the world-sheet metric. The L is the arbitrary length parameter. The
convention of the antisymmetric tensor is taken as τ σ = 1. The ten-dimensional metric
obtained previously and the light-cone gauge condition x + = τ lead to the bosonic action
of the type IIA string theory written as
2π 8
1 2 2
SB = dτ dσ ∂τ x i − ∂σ x i
4πα
0 i=1
2
4 2
8

µ
a 2 µ b 2
− x − x , (2.15)
3 6
a=1 b=5
where we have rescaled the parameters τ , σ and µ as
µ
τ → Lτ, σ → Lσ, µ→ .
L
2.2.2. Fermionic sector

In order to construct the fermionic action, we need explicit expressions of the covariant
derivatives. In the celebrated work [5], the covariant derivatives are obtained by the coset
construction in AdS5 × S 5 . However, the resulting covariant derivatives become those
in the type IIB supergravity with non-trivial background fluxes when we take the light-
cone gauge conditions. Hence, we could have a short-cut in our hand [6]. It would be
sufficient to use the covariant derivatives in the supergravity with background fluxes while
the expressions of the covariant derivatives are already known. Following the work [28],
we can derive the fermionic part of the action using the generalized covariant derivative D
defined by1

1 1
(Da )pq = ∂a δpq + ∂a x ρ ωµνρ δpq − Hµνρ (σ3 )pq γ µν
4 2

1 1
+ Fµν γ (iσ2 )pq +
µν
Fµνλδ γ µνλδ
(σ1 )pq γρ , (2.16)
2 · 2! 2 · 4!
where σi ’s (i = 1, 2, 3) are Pauli matrices. The Hµνρ is the 3-form field strength of the NS–
NS B-field. The Fµν and Fµνλδ are the 2- and 4-form field strengths of the RR 1-form Aµ
and 3-form Cµνλ , respectively. Here, we have ignored the contribution of higher Kaluza–
Klein modes which have the spectrum tower with the energy difference 1/R. Now, the
energy difference is so large that we can ignore the n = 0 sectors. After this truncation, we
are restricted to the n = 0 sector and the gauge coupling constant n/R is effectively zero.
1 It has been reported that the numerical coefficients in the covariant derivatives in the type IIA and the type IIB
include some issues [24,28]. But it should be remarked that these are based on the difference of the convention in
Ref. [33], and not on the incorrectness. We thank C.N. Pope for the valuable comment on this point.
Using this covariant derivative, we can obtain the quadratic fermionic action of the
type IIA described by

2πL

2
i ab
SF = dτ dσ η δpq − ab (σ3 )pq ∂a x µ θ̄ p γµ (Db )qr θ r , (2.17)
2π
0 p,q,r=1
where (p = 1, 2) are two 16-component spinors with different chiralities in ten

θ p ’s
dimensions. When we set the light-cone gauge conditions,
x + = τ, γ + θ p = 0,
then in the same way as in the type IIB case [6] the above action can be rewritten as

2πL

2
i
SF = − dτ dσ θ̄ p γ+ δpq (Dτ )qr + (σ3 )pq (Dσ )qr θ r , (2.18)
2π
0 p,q,r=1
where the length parameter should be fixed as L = α |p+ | now. The covariant derivatives
are also rewritten as

1 1
(Dτ )pq = ∂τ δpq + ωµν+ δpq − Hµν+ (σ3 )pq γ µν
4 2

1 1
+ Fµν γ µν + Fµνλδ γ µνλδ (σ1 )pq γ+ ,
2 · 2! 2 · 4!
(Dσ )pq = ∂σ δpq . (2.19)
µ
When we use the constant 2- and 4-form field strengths F+4 = 3 and F+123 = µ, the
fermionic action can be rewritten as
2π
i µ 1
SF = dτ dσ ψ T ∂τ ψ + ψ T γ9 ∂σ ψ + ψ T γ123 + γ49 ψ , (2.20)
2π 4 3
0
where we have introduced a 16 component spinor ψ defined by

1 √
ψ + ψ2 / 2 Ψ1
ψ= √ ≡ .
ψ1 − ψ2 / 2 Ψ2
The 8 component spinors ψ 1 and ψ 2 are given by

1 0
θ ≡ 1/4
i
(i = 1, 2)
2 ψi
due to the light-cone conditions. Also, in the same way as in the bosonic sector, we have
rescaled the parameters τ , σ , µ and fermion ψ as
µ ψ
τ → Lτ, σ → Lσ, µ→ , ψ→ . (2.21)
L L
2.3. Type IIA string via double-dimensional reduction
By following the work [11] in the light-cone gauge in terms of an SO(9) spinor ψ, we
can write down the action of the supermembrane on the pp-wave [13,14] as

2πL
2πL
1
S= dτ dσ dρ L,
;3M
0 0
2 3 2 9
−1 1 1 r s 2 µ µ
w L= Dτ X Dτ X − X , X
r r
− XI −
2
XI2
2 2 3 6
I =1 I =4

µ
3

− I J K XK XI , XJ
3
I,J,K=1
µ
+ iψ T γ r Xr , ψ + iψ T Dτ ψ + i ψ T γ123ψ, (2.22)
4
where (σ 0 , σ 1 , σ 2 ) = (τ, σ, ρ) is the set of world-volume coordinates on the membrane
and the { , } is a Lie bracket given by using an arbitrary function w(σ, ρ) of world-volume
spatial coordinates σ a (a = 1, 2) as follows:
ab
{A, B} ≡ ∂a A∂b B (a, b = 1, 2),
w
with ∂a = ∂/∂σ a . This theory has the τ -independent gauge symmetry called the
area-preserving diffeomorphism (APD). It is a residual symmetry belonging to the
reparametrization invariance of the membrane world-volume. When we use the gauge
connection ω, the covariant derivative for this gauge symmetry is defined by

Dτ Xr ≡ ∂τ Xr − ω, Xr (r = 1, 2, . . . , 9).
We have also introduced a parameter ;M , which is the M-theory scale related to the
membrane tension TM = 1/;3M . It is associated to the string coupling gs and the string
scale ;s in ten-dimensional string theory (up to some numerical constant) with a relation
1/3
;M = gs ;s . We use a normalization

0 σ 2πL, 0 ρ 2πL, dσ dρ w(σ, ρ) = L2 ,
with L being an arbitrary length parameter. In our light-cone gauge, the time coordinate
“τ ” is associated to the X+ as X+ = (;3M /(2πL)2 )P0+ τ and the longitudinal momentum
P + (σ, ρ) satisfies P + (σ, ρ) = (P0+ /L2 )w(σ, ρ). Hereafter we shall use a convention
P0+ = 1.
Here, we shall consider the double-dimensional reduction (DDR) in the SO(6)-
direction. It is considered that eleven-dimensional supermembrane theory in the flat space
should reduce to the type IIA string theory, at least classically. Based on this fact, we shall
carry out the DDR of the supermembrane on the pp-wave. We will show that the type IIA
string action on the pp-wave obtained in the previous subsection can be derived from the
supermembrane action on the pp-wave (2.22) through the double-dimensional reduction.
To begin, we rotate the variables X1 , . . . , X9 into x’s

XI = x I (I = 1, 2, 3), Xa = x a (a = 6, 7, 8, 9),

µ µ
X4 = x 4 cos τ − x 5 sin τ ,
6 6

µ µ
X5 = x 4 sin τ + x 5 cos τ , (2.23)
6 6
then an associated action is written down as follows:
2π
2πL
1
S= 3 dτ dσ dρ L,
;M
0 0
2 3 2 9
1 2 1 2 µ I 2 µ I 2
w−1 L = Dτ x r − x r , x s − x − x
2 2 3 6
I =1 I =6

µ
3
I J 2 5
− I J K x x , x − µx Dτ x
K 4
3 3
I,J,K=1

r µ T 1
+ iψ γ x , ψ + iψ Dτ ψ + i ψ γ123 + γ54 ψ,
T r T
(2.24)
4 3
where it should be noted that the fermion mass term is modified compared with flat case.
This contribution appears since we have moved to the rotated coordinate. In this time, the
additional spin connection ω+ 45
= µ/6 appears.2 Now let us consider the DDR in the x 4 -
direction. That is, we take x = ρ. We choose the density function w(σ, ρ) to be a constant
4
so that w = (2π)−2 and fix the parameter L as L = gs ;s . The resulting action is given by
2π
1
Sst = dτ dσ Lst ,
2π
0
8 2 2
1 i 2
i 2 µ
4
a 2 µ
8
b 2
Lst = ∂τ x − ∂σ x − x − x
2α 3 6
i=1 a=1 b=5

µ 1
+ iψ T 116 · ∂τ − γ9 · ∂σ + γ123 + γ49 ψ, (2.25)
4 3
where we have renamed the coordinates x 4 , x 5 , . . . , x 9 as x 9 , x 4 , x 5 , . . . , x 8 . It should
be understood that the mass term of x 4 arises from the second term in Eq. (2.10). This
term should describe the effect of the Kaluza–Klein 1-form. Also, the fermionic field and
parameters have been appropriately rescaled as
3/2
L ;M (2π)2
σ → Lσ, τ→ τ, ψ→ ψ, µ→ µ.
(2π)2 L L
2 We have modified the contribution of the spin connection in the revised version. This contribution is initially
pointed out in Ref. [34] where the correct type IIA action is obtained.
The parameters of the resulting theory are related with those of M-theory
1 (2π)L 1/3 √

= 3
, ;M = gs ;s , L = gs ;s , ;s = 2π α .
2πα ;M
It should be noted that the above action is identical with the type IIA action derived in
the previous subsection up to the sign of σ . Hereafter, we can use the following expression
of γ µ = (γ i , γ 8 , γ 9 ),

0 −i γ̃ i
γ i = γ̃ i ⊗ σ2 = (i = 1, . . . , 7), (2.26)
i γ̃ i 0

0 18 18 0
γ 8 = 18 ⊗ σ1 = , γ 9 = 18 ⊗ σ3 = , (2.27)
18 0 0 −18
where γ̃ i ’s (i = 1, . . . , 7) are SO(7) gamma matrices that obey commutation relations
γ̃ i γ̃ j + γ̃ j γ̃ i = 2δ ij . (2.28)
The 16 component fermion ψ is decomposed into two 8 component fermions Ψ 1 and Ψ 2
as
1
Ψ
ψ= .
Ψ2
Moreover, we can decompose the 8 component fermion into two eigen-spinors of the
matrix R ≡ γ̃1234 as follows:

1+R 1−R
Ψa = Ψa + Ψ a ≡ Ψ a+ + Ψ a− (a = 1, 2). (2.29)
2 2
By definition, the spinor Ψ a± satisfy
RΨ a± = ±Ψ a± . (2.30)
That is, Ψ a±are the eigen-spinors with eigen-value ±1, respectively. By the use of Ψ a± ,
the fermionic Lagrangian can be rewritten as
L = iΨ 1+T ∂− Ψ 1+ + iΨ 1−T ∂− Ψ 1− + iΨ 2+T ∂+ Ψ 2+ + iΨ 2−T ∂+ Ψ 2−
µ µ µ µ
− i Ψ 1−T Π T Ψ 2+ − i Ψ 1+T Π T Ψ 2− + i Ψ 2−T ΠΨ
1+ + i Ψ 2+T ΠΨ
1− ,
3 6 6 3
(2.31)
T T T
where Π ≡ γ̃123 , Π ≡ γ̃321 and satisfies Π Π = Π Π = 1.
3. Matrix string theory on pp-wave
We can also consider the matrix string theories [35] on the pp-wave3 from the
supermembrane by the use of the method in the work [36].
3 Matrix strings are also discussed in Refs. [37,38] from different viewpoints from ours.
Let us start with the supermembrane action (2.22), and rotate the variables into x’s as
given by (2.23). In this time, the gamma matrices are also transformed by this rotation. The
resulting supermembrane action is given by
2π
2πL
1
S= dτ dσ dρ L,
;3M
0 0
2 3 2 9
1
r 2 1 r s 2 µ I 2 µ I 2
L= Dτ x − x , x − x − x
2 2 3 6
I =1 I =6

µ
3
I J 2 5
− I J K x x , x − µx Dτ x
K 4
3 3
I,J,K=1

µ 1
+ iψ T γ r x r , ψ + iψ T Dτ ψ + i ψ T γ123 + γ54 ψ, (3.1)
4 3
where we have set w = (2π)−2 and rescaled σ as σ → (2π)2 σ . Now, the Lie bracket { , }
is simply defined by
{A, B} ≡ ∂σ A∂ρ B − ∂ρ A∂σ B.

Then, we rewrite x 4 as x 4 ≡ Y and shift Y as
Y −→ ρ + Y.
The Y is regarded as the compactified direction. As the result, the action is rewritten as
2π 2π
L
S= dτ dσ dρ L,
;3M
0 0
2
3
1 2 2 1 i j 2 µ I 2
L= 2
F0,σ + Dτ x i − DσY x i − 2
x , x − x
2 2L 3
I =1
2
µ
8
I 2 µ
3
2 4
− x − I J K x x , x − µx F0,σ
K I J
6
3L 3
I =5 I,J,K=1

1 T i i µ T 1
+ i ψ γ x , ψ + iψ Dτ ψ − iψ γ Dσ ψ + i ψ γ123 + γ49 ψ,
T T 9 Y
L 4 3
(3.2)
where we have reassigned the variables x 4 , x 5 , . . . , x 9 as x 9 , x 4 , x 5 , . . . , x 8 and rescaled
ρ → Lρ. We have also introduced the following quantities,
1 1
F0,σ ≡ ∂τ Y − ∂σ ω − {ω, Y }, Dτ x i ≡ ∂τ x i − ω, x i ,
L L
1
DσY x i ≡ ∂σ x i − Y, x i ,
L
where A0 ≡ ω and Aσ ≡ Y . The inverse compactification radius 1/L plays a role of the
gauge coupling constant. It seems that the action (3.2) is not explicitly invariant under
the area-preserving diffeomorphism. But the action indeed has this symmetry under the
transformation with an infinitesimal gauge parameter Λ
δω = L∂τ Λ + {Λ, ω}, δY = L∂σ Λ + {Λ, Y },

δx = Λ, x ,
i i
δψ = {Λ, ψ}.
The action (3.2) is very close to that of the matrix string theory. In fact, by the use
of the corresponding law of Ref. [36] in the large N limit, it is straightforward to map
the supermembrane action into matrix representations. Thus, we can formally obtain the
matrix string action on the pp-wave, up to O(1/N 2 ) described by
2π
L 2π
S= dτ dθ L,
;3M N
0

1 2 2 1 N 2 i j
2
L = Tr F0,θ2
+ Dτ x i − N 2 Dθ x i + x ,x
2 2 2πL
2 3 2 8
µ I 2 µ I 2
− x − x
3 6
I =1 I =5
3

µ N
2 4
+i I J K x x , x − µx F0,θ
K I J
3 2πL 3
I,J,K=1

N
+ Tr ψ T γ i x i , ψ + iψ T Dτ ψ − Niψ T γ 9 Dθ ψ
2πL

µ T 1
+ i ψ γ123 + γ49 ψ , (3.3)
4 3
where the quantities in the action are replaced with
N N
F0,θ = ∂τ Y − N∂θ ω + i [ω, Y ], Dτ x i = ∂τ x i + i ω, x i ,

2πL 2πL
1
Dθ x i = ∂θ x i + i Y, x i .
2πL
If we rescale some constants as
τ √ L
τ→ , ψ→ N ψ, L→ , µ → Nµ,
N 2π
then we can rewrite Eq. (3.3) as
2π
1
S= dτ dθ L,
;3M
0

1 µ 4 2 2 2 1
2
L = Tr F0,θ − x + Dτ x i − Dθ x i + x i , x j
2 3 2
2
µ 2 I 2
4 8 3
µ I 2 2
− x − x +i µ I J K x I x J x K
3 6
3
I =1 I =5 I,J,K=1

µ 1
+ Tr ψ T γ i x i , ψ + iψ T Dτ ψ − iψ T γ 9 Dθ ψ + i ψ T γ123 + γ49 ψ ,
4 3
(3.4)
where the field strength and covariant derivatives are given by

F0,θ = ∂τ Y − ∂θ ω + i[ω, Y ], Dτ x i = ∂τ x i + i ω, x i ,

Dθ x i = ∂θ x i + i Y, x i .
The action (3.4) includes the 3-point interaction and several mass terms and also the field
strength of the gauge connection is shifted by x 4 . Thus it seems that the action (3.4) is
not invariant under the area-preserving diffeomorphism. However, this action of the matrix
string is actually invariant under the gauge transformation with an matrix parameter Λ
δω = ∂τ Λ − i[Λ, ω], δY = ∂θ Λ − i[Λ, Y ],

δx = −i Λ, x ,
i i
δψ = −i[Λ, ψ].
The τ -scaling leads to the N -dependence of the physical light-cone time X+ as
;3M P0+ τ
X+ = .
N(2πL)2
We also should rescale P0+ as P0+ → NP0+ so that X+ should be independent of N . The
diagonal elements of the matrix x i describe a fundamental string bit in the large N limit
and hence the total longitudinal momentum is proportional to the number N of string bits.
It is easily observed that the above action (3.4) becomes the usual matrix string action
in the flat limit µ → 0. Moreover, let us consider the IR region. At the time, the matrix
variables are restricted to the Cartan subalgebra. That is, the matrix becomes diagonal and
so the term including commutator should vanish. Finally, integrating out the field strength
F0,θ as the auxially field, one can find that the above action should reduce to the free
type IIA string theory obtained in the previous section, as is expected.
Also, we should remark that the above action of the matrix string is included in the
family of the work [37] where the action of the matrix string and supersymmetry have been
more generally investigated from the viewpoint of the mass deformation of the Yang–Mills
theory.
Finally, we comment on the classical solution. As in the BMN matrix model [8], for
example, this matrix string theory has the static fuzzy sphere solution described by
µ I
xI = J (I = 1, 2, 3),
3
x 4 = · · · = x 8 = Y = ω = 0, (3.5)
where the J I ’s are generators of an SU(2) algebra. The existence of the fuzzy sphere
solution might be physically expected from the presence of the constant flux of RR
3-form [20]. It would be possible to consider other classical solutions.
4. Quantization of type IIA string on pp-wave
In this section we will consider the mode-expansions and quantization of closed and
open strings in the type IIA on the pp-wave. In particular, we investigate D-branes living
in the theory.
4.1. Closed strings in type IIA on pp-wave
In this subsection we will discuss the mode-expansions of bosonic and fermionic

degrees of freedom and the quantization in the type IIA closed string.
First the variation of the action previously obtained leads to equations of motion given
by
µ2 a
∂+ ∂− x a + x = 0 (a = 1, 2, 3, 4), (4.1)
9
µ2 b
∂+ ∂− x b + x = 0 (b = 5, 6, 7, 8), (4.2)
36
µ
∂+ Ψ 2+ + Π Ψ 1− = 0, (4.3)
3
µ T 2+
∂− Ψ 1− − Π Ψ = 0, (4.4)
3
µ
∂+ Ψ 2− + Π Ψ 1+ = 0, (4.5)
6
µ T 2−
∂− Ψ 1+ − Π Ψ = 0. (4.6)
6
The mode-expansions of bosonic variables are described by

µ 3 a µ
x a
(τ, σ ) = x0a cos τ + α p0 sin τ
3 µ 3

α 1 a B
+i B
αn φn + ᾱna φ̃nB (a = 1, 2, 3, 4),
2 ω
n =0 n

µ 6 b µ
x (τ, σ ) = x0 cos
b b
τ + α p0 sin τ
6 µ 6

α 1 b B b B

+i B αn φn + ᾱn φ̃n (b = 5, 6, 7, 8),

2 ωn
n =0
and those of fermionic variables are

µ µ
Ψ (τ, σ ) = Ψ0 cos
1−
τ + Ψ0 sin τ
3 3

3 F T

+ cn i ωn − n Π Ψn φn + Ψn φ̃n ,
F F
(4.7)
µ
n =0

µ µ
Ψ0 sin
Ψ 2+ (τ, σ ) = −Π τ +Π Ψ0 cos τ
3 3
3

+ cn Ψn φnF − i ωnF − n Π Ψ n φ̃nF , (4.8)
µ
n =0

µ µ
Ψ (τ, σ ) = Ψ0 cos
1+
τ + Ψ0 sin τ
6 6

6
F T F F
+ cn
i ωn − n Π Ψn φn + Ψn φ̃n , (4.9)
µ
n =0

µ µ
Ψ0 sin
Ψ 2− (τ, σ ) = −Π τ +Π Ψ0 cos τ
6 6

F 6 F F
+
cn Ψn φn − i ωn − n Π Ψn φ̃n , (4.10)
µ
n =0
where we introduced several notations

2 2
µ µ
ωn = sgn(n) n +
B 2 , ωn = sgn(n) n +
B 2 ,
3 6

φnB = exp −i ωnB τ − nσ , φ̃nB = exp −i ωnB τ + nσ ,

φnB = exp −i ωnB τ − nσ , φ̃nB = exp −i ωnB τ + nσ ,

2 2
µ µ
ωn = sgn(n) n +
F 2 , ωn = sgn(n) n +
F 2 ,
3 6

φnF = exp −i ωnF τ − nσ , φ̃nF = exp −i ωnF τ + nσ ,

φnF = exp −i ωnF τ − nσ , φ̃nF = exp −i ωnF τ + nσ ,
2 2
3 F 2 −1/2 6 F 2 −1/2
cn = 1 + ωn − n , cn = 1 + ωn − n . (4.11)
µ µ
Following the usual canonical quantization procedure, we can quantize the theory and
obtain the commutation relations. The canonical momenta are given by
1
pa = ∂τ x a (a = 1, 2, 3, 4),
2πα
1
pb = ∂τ x b (b = 5, 6, 7, 8),
2πα
Sα = iψαT ,
and the canonical (anti-)commutation relations are represented as

i

x (τ, σ ), pj τ, σ = iδ ij δ σ − σ ,
i
ψα (τ, σ ), Sβ τ, σ = δαβ δ σ − σ ,
2
1
ψα (τ, σ ), ψβ τ, σ = δαβ δ σ − σ ,
T
2
where the delta function is defined by
1 in(σ −σ )
δ σ − σ = e .
2π n
From the above results, we can obtain the commutation relations of bosonic modes as
i j
i j
ᾱm , αn = αm , ᾱn = 0 (i, j = 1, . . . , 8),

a a

αm , αn = ωm B
δm+n,0 δ aa (a, a = 1, 2, 3, 4),
a a

ᾱm , ᾱn = ωmB
δm+n,0 δ aa ,
b b
B
αm , αn = ωm δm+n,0 δ bb (b, b = 5, 6, 7, 8),
b b
B
ᾱm , ᾱn = ωm δm+n,0 δ bb ,
i j
x0 , p0 = iδ ij , otherwise is zero. (4.12)

and those of fermionic ones as
T
n
(Ψm )α , Ψ = Ψ m , (Ψn )Tβ = 0,
β α
T 1
(Ψm )α , (Ψn )Tβ = Ψ n
m , Ψ = δm+n,0 δαβ ,
α β 2
T T

Ψm α , Ψn β = Ψm α , Ψn β = 0,
T T 1
Ψm α , Ψn β = Ψ m , Ψ n = δm+n,0 δαβ . (4.13)
α β 2
Though further considerations will not be done here, we can obtain the quantum
Hamiltonian or spectrum exactly with the standard procedure.
4.2. Open strings and Dp-branes in type IIA on pp-wave
In this subsection we shall discuss the mode-expansions of open strings in the type IIA
string by imposing boundary conditions. In particular, we would like to consider D-branes,
following Ref. [30]. (For more detailed studies, see Refs. [31,32].) It has been shown in
Ref. [30] that Dp-brane is not allowed for p = 1, 9 and there are some restrictions on
directions of allowed D-branes. First we consider the open string action described by
π
1
Sst = dτ dσ Lst , (4.14)
2π
0
8 2 2
1
4 8
µ a 2 µ b 2
Lst = ∂+ x ∂− x −
i i
x − x
2α 3 6
i=1 a=1 b=5
+ iΨ ∂− Ψ + iΨ
1+T 1+ 1−T
+ iΨ
∂− Ψ 1− ∂+ Ψ + iΨ 2−T ∂+ Ψ 2−
2+T 2+
µ µ µ
− i Ψ 1−T Π T Ψ 2+ − i Ψ 1+T Π T Ψ 2− + i Ψ 2−T Π Ψ 1+
3 6 6
µ
+ i Ψ 2+T Π Ψ 1− . (4.15)
3
Similarly, we obtain equations of motion (4.1)–(4.6) from the above action (4.14). In
order to solve the above equations of motion we have to impose the following boundary
conditions on bosonic coordinates x i ’s (i = 1, 2, . . . , 8),
Neumann: ∂σ x m = 0 (m = +, −, and some p − 1 coordinates),

Dirichlet: ∂τ x m = 0 ( m = other 9 − p coordinates),
where 8 transverse indices i = 1, . . . , 8 are decomposed into a = 1, 2, 3, 4 (flux directions)
and b = 5, . . . , 8. For fermionic coordinates, boundary conditions are imposed as

Ψ 1− σ =0,π = ΩΨ 2+
σ =0,π
, Ψ 2+ σ =0,π = Ω
T Ψ 1−
σ =0,π
,

Ψ 1+ σ =0,π = ΩΨ 2−
σ =0,π
, Ψ 2− σ =0,π = Ω
T Ψ 1+
σ =0,π
,
is defined by
where Ω
= γ̃m γ̃m . . . γ̃m .
Ω 1 2 9−p
includes odd number of gamma matrices since the SO(8) chiralities of Ψ 1 and Ψ 2
The Ω
must be opposite in the type IIA theory and hence p is restricted to even.
Under these boundary conditions we can obtain classical solutions for equations of
motion, and mode-expansions of bosonic variables are given by

µ 3 µ
x a (τ, σ ) = x0a cos τ + 2α p0a sin τ (a = 1, 2, 3, 4)
3 µ 3
√ 1
α a e−iωn τ cos(nσ ) (Neumann),
B
+ i 2α (4.16)
ωnB n
n =0
√ 1
αna e−iωn τ sin(nσ ) (Dirichlet),
B
x a (τ, σ ) = 2α B
(4.17)
ω
n =0 n

µ 6 µ
x b (τ, σ ) = x0b cos τ + 2α p0b sin τ (b = 5, 6, 7, 8)
6 µ 6
√ 1
b −iωnB τ
+ i 2α αn e cos(nσ ) (Neumann), (4.18)
ωnB
n =0
√ 1
b −iωnB τ
x b (τ, σ ) = 2α B αn e sin(nσ ) (Dirichlet), (4.19)
ωn
n =0

where ωnB and ωnB have been defined by (4.11). The mode-expansions of fermionic
variables are the same as in the closed string case. The quantization can be done in the
same way as closed strings. The commutation relations of bosonic and fermionic modes
are the same as (4.12) and (4.13). The quantum Hamiltonian and spectrum can be also
studied with the standard procedure but we will not investigate them furthermore here.
Next we will study D-branes. Though the mode-expansions of fermionic variables are
the same as in the closed string case, in the open string case fermionic boundary conditions
lead to further constraints
Π
Ψ0 = Ω Ψ
0 , Ω 0 = −ΠΨ
T Ψ 0, n = ΩΨ
Ψ n (n = 0),
Ψ0 Π
=Ω Ψ
0 , ΩT 0
Ψ 0 ,
= −ΠΨ n
Ψ n
= ΩΨ (n = 0).
The self-consistency of these conditions gives us a condition
Π
Ω Ω
Π = −1. (4.20)
This condition is peculiar to the pp-wave, and gives an additional constraint for the Dp-
branes in the theory. In fact, in the massive type IIB theory, D1- and D9-branes are
forbidden and D3-, D5- and D7-branes can exist but those directions are limited. In the
massive type IIA theory similar restrictions are imposed.
We shall list the possible Dp-branes below:
• p = 8: one of I = 1, 2, 3 is Dirichlet type,

= γ̃ I .
Ω
• p = 6: (1) two of I = 1, 2, 3 and one of I = 4, . . . , 8 are Dirichlet types,

= γ̃ I γ̃ J γ̃ I ;
Ω
(2) three of I = 4, . . . , 8 are Dirichlet types,
= γ̃ I γ̃ J γ̃ K .
Ω
• p = 4: (1) all of I = 1, 2, 3 and two of I = 4, . . . , 8 are Dirichlet types,
= γ̃ I γ̃ J γ̃ K γ̃ I γ̃ J ;
Ω
(2) one of I = 1, 2, 3 and four of I = 4, . . . , 8 are Dirichlet types,
= γ̃ I γ̃ I γ̃ J γ̃ K γ̃ L .
Ω
• p = 2: two of I = 1, 2, 3 and all of I = 4, . . . , 8 are Dirichlet types,

= γ̃ I γ̃ J γ̃ I γ̃ J γ̃ K γ̃ L γ̃ M .
Ω
In conclusion, all Dp-branes can exist for p = even, but those directions are constrained
as the case of the type IIB [30]. We note that zero-point energy varies for each direction of
D-branes.
We have considered the type IIA string theory on the pp-wave background from the
eleven-dimensional viewpoint. To begin, we have discussed the type IIA pp-wave solution
through the toroidal compactification of the maximally supersymmetric pp-wave solution
in eleven dimensions on a spatial isometry direction. Next, we have derived the action of the
type IIA string theory from the type IIA pp-wave solution of the supergravity. Moreover,
we have derived the type IIA string action from the eleven-dimensional supermembrane
theory on the maximally supersymmetric pp-wave background by applying the double-
dimensional reduction for a spatial isometry direction. The resulting action agrees with
the one obtained from the supergravity side. In particular, the Kaluza–Klein gauge field
induces a mass term of a bosonic coordinate in the type IIA theory. Furthermore, we have
written down the action of the matrix string on the pp-wave. This action contains the
3-point interaction and mass terms. Also, the field strength of the gauge connection is
shifted. However, this action is still gauge invariant, though this theory is not maximally
supersymmetric. In particular, this theory is reduced to the matrix string theory in the flat
space by taking the limit µ → 0. We have also discussed the quantization of closed and
open strings in the type IIA string. In particular, the allowed Dp-branes in this theory have
been investigated. The values p = 2, 4, 6 and 8 are allowed but the directions of D-branes
are constrained.
We can also consider compactifications along other isometry directions. In such cases
the number of the remaining supercharges is less than 24. Furthermore, it is nice to
study the type IIA pp-wave background preserving 26 supercharges [27] or type IIA
string theory on such a background from the eleven-dimensional supermembrane. It is
an interesting work to discuss less supersymmetric type IIA string theories from the
supermembrane. Moreover, the supersymmetric D-branes in such type IIA string theories
are very interesting subject to study.
It is nice to study the matrix string theory written down here from several aspects. In
particular, it would be interesting to study the relation between the matrix string theory on
the pp-wave and “string bit” [39].
Acknowledgement
The work of K.S. is supported in part by the Grant-in-Aid from the Ministry of
Education, Science, Sports and Culture of Japan (F 14740115).
Appendix A. Compactification on an SO(3)-direction
In the text we have considered the compactification on an SO(6)-direction. We can also

compactify one of the directions x I ’s (I = 1, 2, 3). In this case, there are some different
points from the compactification on an SO(6)-direction and we shall summarize these
differences.
In this compactification on an SO(3)-direction the change of variables is given by

µ
X+ = x + , X− = x − + x 1 x 2 , Xa = x a (a = 3, . . . , 9),
3

µ + µ +
X = x cos
1 1
x − x sin
2
x ,
3 3

µ + µ +
X = x sin
2 1
x + x cos
2
x , (A.1)
3 3
then the metric is rewritten as
2 9
r 2 4 2 + 1
ds 2 = −2 dx + dx − + G++ x 3 , x I dx + + dx − µx dx dx ,
3
r=1

µ 2 I 2
9
µ 2 3 2
G++ x 3 , x I ≡ − x + x , (A.2)
3 6 I =4
and the constant 4-form flux is still written in Eq. (2.2). In this case the x 1 -direction is a
manifest spatial isometry direction. In the same way, we can obtain the type IIA solution
from the above expression. The ten-dimensional metric gµν is given by
2
8
i 2
gµν dx µ dx ν = −2 dx + dx − + g++ x i dx + + dx ,
i=1
2 2 2
8

2 1 2 µ 2 2 µ a 2
g++ x i ≡ − µ x + x + x , (A.3)
3 3 6
a=3
where we have relabelled the indices x2, . . . , x9 as x 1, . . . , x 8. The Kaluza–Klein gauge

field Aµ is represented by
2
A+ = − µx 1 , Ai = 0 (i = 1, . . . , 8), (A.4)
3
and non-zero NS–NS 2-form is given by
µ 2 µ
B+1 = x , B+2 = − x 1 . (A.5)
3 3
In this case, the RR 3-form Cµνρ and dilaton φ vanish.
To begin, let us discuss the bosonic sector. We can easily obtain the bosonic action from
Eq. (2.14) using the type IIA metric. The resulting action is expressed by
2π 8
1 2 2 2 2 2
SB = dτ dσ ∂τ x i − ∂σ x i + µx ∂σ x 1 − µx 1 ∂σ x 2
4πα 3 3
0 i=1
2 2 2
8

2 1 2 µ 2 2 µ a 2
− µ x − x − x , (A.6)
3 3 6
a=3
where the mass term for x 1 is induced from the Kaluza–Klein gauge field Aµ as the case of
the compactification on an SO(6)-direction. It is also an easy exercise to derive the above
action (A.6) by using the double-dimensional reduction.
Next, we shall consider the fermionic sector. Now, in the study of the supergravity, the
field strength of RR 3-form is zero, but NS–NS 2-form is non-zero and it has the constant
field strength proportional to µ. Thus, this contribution induces the fermion mass term.
However, there might be possibly an issue for the numerical constant and the fermion mass
term obtained in the supergravity analysis is not identical with the one derived via double-
dimensional reduction if we naively use the expression of the covariant derivative in the
text.
References
[2] J. Figueroa-O’Farrill, G. Papadopoulos, Homogeneous fluxes, branes and a maximally supersymmetric
solution of M-theory, JHEP 0108 (2001) 036, hep-th/0105308.
Dordrecht, 1976, pp. 271–275;
R. Güven, Plane wave limits and T-duality, Phys. Lett. B 482 (2000) 255, hep-th/0005061.
[4] M. Blau, J. Figueroa-O’Farrill, C. Hall, G. Papadopoulos, A new maximally supersymmetric background
of IIB superstring theory, JHEP 0201 (2001) 047, hep-th/0110242;
M. Blau, J. Figueroa-O’Farrill, C. Hall, G. Papadopoulos, Penrose limits and maximal supersymmetry, hep-
th/0201081.
Phys. B 625 (2002) 70, hep-th/0112044.
background, Phys. Rev. D 65 (2002) 126004, hep-th/0202109.
[7] J.G. Russo, A.A. Tseytlin, On solvable models of type IIB superstring in NS–NS and RR plane wave
backgrounds, JHEP 0204 (2002) 021, hep-th/0202179.
JHEP 0204 (2002) 013, hep-th/0202021.
[9] E. Bergshoeff, E. Sezgin, P. Townsend, Supermembranes and eleven-dimensional supergravity, Phys.
Lett. 189 (1987) 75.
[10] E. Bergshoeff, E. Sezgin, P. Townsend, Properties of the eleven-dimensional supermembrane theory, Ann.
Phys. 185 (1988) 330.
[11] B. de Wit, J. Hoppe, H. Nicolai, On the quantum mechanics of supermembranes, Nucl. Phys. B 305 (1988)
545.
[12] T. Banks, W. Fischler, S.H. Shenker, L. Susskind, M-theory as a matrix model: conjecture, Phys. Rev. D 55
(1997) 5112, hep-th/9610043.
[13] K. Dasgupta, M.M. Sheikh-Jabbari, M. Van Raamsdonk, Matrix perturbation theory for M-theory on a pp-
wave, JHEP 0205 (2002) 056, hep-th/0205185.
[14] K. Sugiyama, K. Yoshida, Supermembrane on the pp-wave background, hep-th/0206070, Nucl. Phys. B,
in press.
[15] K. Sugiyama, K. Yoshida, BPS conditions of supermembrane on the pp-wave, hep-th/0206132, Phys.
Lett. B, in press.
[16] T. Banks, N. Seiberg, S. Shenker, Branes from matrices, Nucl. Phys. B 490 (1997) 91, hep-th/9612157.
[17] N. Kim, J. Plefka, On the spectrum of pp-wave matrix theory, hep-th/0207034;
K. Dasgupta, M.M. Sheikh-Jabbari, M. Van Raamsdonk, Protected multiplets of M-theory on a plane wave,
hep-th/0207050;
N. Kim, J.-H. Park, Superalgebra for M-theory on a pp-wave, hep-th/0207061.
[18] D. Bak, Supersymmetric branes in pp-wave background, hep-th/0204033;

S. Hyun, H. Shin, Branes from matrix theory in pp-wave background, hep-th/0206090, Phys. Lett. B,
in press;
M. Alishahiha, M. Ghasemkhani, Orbiting membranes in M-theory on AdS7 × S 4 background, hep-
th/0206237.
[19] K. Sugiyama, K. Yoshida, Giant graviton and quantum stability in matrix model on pp-wave background,
hep-th/0207190, Phys. Rev. D, in press.
[21] J. Michelson, (Twisted) toroidal compactification of pp-waves, hep-th/0203140.
[22] M. Cvetič, H. Lu, C.N. Pope, M-theory pp-waves, Penrose limits and supernumerary supersymmetries,
hep-th/0203229.
[23] J.P. Gauntlett, C.M. Hull, pp-waves in 11 dimensions with extra supersymmetry, JHEP 0206 (2002) 013,
hep-th/0203255.
[24] R. Corrado, N. Halmagyi, K.D. Kennaway, N.P. Warner, Penrose limits of RG fixed points and pp-waves
with background fluxes, hep-th/0205314.
[25] E.G. Gimon, L.A. Pando Zayas, J. Sonnenschein, Penrose limits and RG flows, hep-th/0206033;
D. Brecher, C.V. Johnson, K.J. Lovis, R.C. Myers, Penrose limits, deformed pp-waves and the string duals
of N = 1 large n gauge theory, hep-th/0206045.
[26] I. Bena, R. Roiban, Supergravity pp-wave solutions with 28 and 24 supercharges, hep-th/0206195.
[27] J. Michelson, A pp-wave with 26 supercharge, hep-th/0206204.
[28] M. Alishahiha, M.A. Ganjali, A. Ghodsi, S. Parvizi, On type IIA string theory on the pp-wave background,
hep-th/0207037.
[29] M.J. Duff, P.S. Howe, T. Inami, K.S. Stelle, Superstrings in D = 10 from supermembranes in D = 11, Phys.
Lett. B 191 (1987) 70.
[30] A. Dabholkar, S. Parvizi, Dp-branes in pp-wave background, hep-th/0203231.
[31] M. Billó, I. Pesando, Boundary states for GS superstrings in an Hpp-wave background, Phys. Lett. B 536
(2002) 121, hep-th/0203028.
[32] O. Bergman, M.R. Gaberdiel, M.B. Green, D-brane interactions in type IIB plane-wave background, hep-
th/0205183.
[33] M. Cvetič, H. Lü, C.N. Pope, K.S. Stelle, T-duality in the Green–Schwarz formalism, and the mass-
less/massive IIA duality map, Nucl. Phys. B 573 (2000) 149, hep-th/9907202.
[34] S. Hyun, H. Shin, N = (4, 4) type IIA string theory on pp-wave background, hep-th/0208074.
[35] R. Dijkgraaf, E. Verlinde, H. Verlinde, Matrix string theory, Nucl. Phys. B 500 (1997) 43, hep-th/9703030.
[36] Y. Sekino, T. Yoneya, From supermembrane to matrix string, Nucl. Phys. B 619 (2001) 22, hep-th/0108176.
[37] G. Bonelli, Matrix strings in pp-wave backgrounds from deformed super-Yang–Mills theory, JHEP 08
(2002) 022, hep-th/0205213.
[38] R. Gopakumar, String interactions in pp-waves, hep-th/0205174.
[39] H. Verlinde, Bits, matrices and 1/N , hep-th/0206059.
Supersymmetric action of multiple D0-branes

from matrix theory
Masako Asano a , Yasuhiro Sekino b
a Theory Division, Institute of Particle and Nuclear Studies KEK, High Energy Accelerator Research
Organization Tsukuba, Ibaraki 305-0801, Japan
b Department of Physics, Tokyo Institute of Technology, Ookayama, Meguro-ku, Tokyo 152-8550, Japan
Received 6 August 2002; received in revised form 6 September 2002; accepted 10 September 2002
Abstract
We study one-loop effective action of Berkooz–Douglas matrix theory and obtain non-Abelian
action of D0-branes in the longitudinal 5-brane background. In this paper, we extend the analysis
of hep-th/0201248 and calculate the part of the effective action containing fermions. We show that
the effective action is manifestly invariant under the loop-corrected SUSY transformation, and give
the explicit transformation laws. The effective action consists of blocks which are closed under the
SUSY, and it includes the supersymmetric completion of the couplings to the longitudinal 5-branes
proposed by Taylor and Van Raamsdonk as a subset.
1. Introduction
The fact that multiple D-branes are described by the matrix valued coordinates led
to many surprising effects. Especially, Dp-branes are allowed to couple to Ramond–
Ramond forms of higher degree than p + 1 [1,2]. Due to those couplings, non-commutative
configurations are stabilized in the presence of certain background fields, which are
interpreted as the D-branes having finite extent [2]. To study such intrinsically stringy
effects, precise understanding of the effective action of multiple D-branes in background
fields is needed.
It is well known that a single D-brane in the low acceleration limit is described by the
Born–Infeld action which includes all the α corrections. However, it is not clear how to
generalize it to the case of multiple D-branes which have non-Abelian gauge symmetries.
E-mail addresses: asano@post.kek.jp (M. Asano), sekino@th.phys.titech.ac.jp (Y. Sekino).
PII: S 0 5 5 0 - 3 2 1 3 ( 0 2 ) 0 0 8 2 4 - 6
152 M. Asano, Y. Sekino / Nuclear Physics B 644 (2002) 151–169
In the flat background, the effective action of N coincident D-branes is given by the
maximally supersymmetric U(N) Yang–Mills theory in the small α limit, but the terms of
higher orders in α are not well-understood.
Supersymmetrization of multiple D-brane action including these higher order terms is
also a hard problem. In the Abelian case, there exists κ-symmetric formulation [3,4] which
gives a supersymmetric action after gauge fixing the world-volume diffeomorphism and
the κ-symmetry [4], but an attempt to define κ-symmetry with non-Abelian parameter [5]
does not seem successful [6]. Studies for obtaining α corrections to the SYM at the first
few orders including the fermionic part is being done from various standpoints (see the
references cited below for current status): (i) from the calculation of open string disk
amplitudes [6]; (ii) by constructing the invariants with respect to the α -corrected SUSY
transformation [7,8]; (iii) by requiring the existence of certain BPS states which are present
in string theory [9].
When we consider multiple D-branes in non-trivial backgrounds, determining the action
is further difficult. Background fields should be regarded as a function of matrix fields
somehow, but the principles for doing so is not clear. As for the supersymmetrization,
there is practically no knowledge up to now.
Effective action of matrix theory [10] gives insight for the multiple D-brane action.
Taylor and Van Raamsdonk studied the effective action of BFSS matrix theory in detail
and found the terms which can be interpreted as the supergravity interactions. From those
terms, they read off the coupling of D-branes to weak background fields, and further
proposed a form of the couplings to general weak background fields [1,11–13]. Those
couplings were applied to various contexts, including the gauge-theory calculation of
the absorption cross section of dilaton higher partial waves by D3-branes, which exactly
reproduces the semiclassical supergravity results [14].
The subject of our paper is matrix theory proposed by Berkooz and Douglas [15], which
is the matrix model for M-theory in the presence of longitudinal (L) 5-branes. Berkooz–
Douglas (BD) matrix theory is defined by the 0–4 string and 0–0 string sectors of the SYM
describing D0–D4 system. In a previous paper [16], we performed one-loop integration
of 0–4 fields and obtained the bosonic part of the non-Abelian action of D0-branes. We
found that the action consists of the terms given from the general proposal of Taylor and
Van Raamsdonk by substituting the L5-brane backgrounds, plus the corrections involving
extra commutators. Since L5-branes are degrees of freedom which are not present in BFSS
matrix theory, the fact that the proposed couplings were exactly reproduced is regarded as
a non-trivial evidence for the consistency of the BFSS and BD matrix models.
In this paper, extending the analysis of Ref. [16], we calculate the one-loop effective
action of BD matrix theory and obtain non-Abelian effective action of D0-branes
including the part containing fermionic fields. Especially, we reveal the consequence of
the supersymmetry of the classical action. We point out that the one-loop effective action
is manifestly invariant under effective SUSY transformation. The transformation law is
given simply by the one-loop expectation value of the classical SUSY transformation law.
By examining the transformation rules, we decompose the terms into the blocks which
close within themselves under the transformation. Among these blocks, we identify the
supersymmetric completion of the bosonic terms given by the Taylor and Van Raamsdonk’s
proposal applied to the longitudinal 5-brane background.
M. Asano, Y. Sekino / Nuclear Physics B 644 (2002) 151–169 153
This paper is organized as follows. In Section 2, we review the action and SUSY
transformation of BD matrix theory. In Section 3, after explaining the method of the
loop calculation, we present the results for the fermionic terms of the effective action.
In Section 4, we discuss the SUSY of the effective action. We conclude with remarks on
the directions for future studies in Section 5. We summarize the representation of spinors
and gamma matrices adopted in this paper in Appendix A, and derive the invariance of
one-loop effective action under the loop-corrected SUSY transformation in Appendix B.
2. Berkooz–Douglas matrix theory
The action of Berkooz–Douglas matrix theory is given by the 0–0 and 0–4 string sectors
of the SYM which describe the D0–D4 bound state. In the case of N D0-branes and N4
D4-branes, the theory has U(N) gauge symmetry and U(N4 ) global symmetry (which is
not gauged, for the gauge fields on D4-branes are discarded). The action is based on the
D = 6 N = 1 SYM and reads as follows.
S = S0 + S5 , (2.1)

1 1 1
S0 = dt Tr D0 Xi D0 Xi + 2 [Xi , Xj ]2 − iθ−† D0 θ− − iθ+† D0 θ+
gs s 2 4λ
1 † 0 a 1 † 0 a
+ θ− γ γ [θ− , Xa ] + θ+ γ γ [θ+ , Xa ]
λ λ
1 † 0 1 † 0
+ θ+ γ [θ− , φ2 ] − θ− γ θ+ , φ̄2
λ λ
1 1 † ∗ †
+ θ+ C[θ− , φ1 ] + θ+ C θ− , φ̄1 , (2.2)
λ λ

1 1 1
S5 = dt (D0 vI )† D0 vI − 2 vI† (Xa )2 vI − iχ † D0 χ − χ † γ 0 γ a Xa χ
gs s λ λ
1
− 2 v1† φ1 , φ̄1 + φ2 , φ̄2 v1 − v2† φ1 , φ̄1
2λ

+ φ2 , φ̄2 v2 − 2v2† φ̄1 , φ̄2 v1
√
2 † 0
+ 2v1† φ1 , φ2 v2 − χ γ θ− v1 − χ † C ∗ θ−† v2 − v1† θ−† γ 0 χ + v2† θ− Cχ
λ
1 † † † † † † † †
− 2 v1 v1 )(v1 v1 + v2 v2 v2 v2 − 2 v1 v2 v2 v1 + 4 v1 v1 v2 v2
2λ
(2.3)
where we use the indices a, b = 5, . . . , 9 for the space transverse to the D4-branes,
m, n = 1, . . . , 4 for the space along the D4-branes, and i, j = 1, . . . , 9 to denote both of
them. Dimensionful parameter λ is defined by λ = 2π2s .
The part S0 is the terms containing only the 0–0 sector fields (Xa , φ1 , φ2 , θ− , θ+ ),
which is nothing but the BFSS action, i.e., the dimensional reduction of D = 10, N = 1
SYM. The 0–0 sector fields are in the adjoint rep. of U(N) and are singlets of U(N4 ).
Covariant derivatives for those fields are defined as D0 Xi = ∂0 Xi + i[A, Xi ]. Note that we
have defined the complex combination of Xm by (φ1 , φ2 ) = (X1 + iX2 , X3 + iX4 ), and
that the fermions are expressed as 6D Weyl spinors, which have 4 complex components.
As in the previous paper [16], we do not use the SU(2) Majorana convention, for we
prefer unconstrained fermions for the loop calculations. The subscripts on the spinors
θ+ , θ− denote the positive and negative 6D chiralities, respectively (γ̄ θ± = ±θ± , where
γ̄ = γ 0 γ 1 · · · γ 5 ). The matrix C is the charge conjugation matrix. We also use the ‘complex
conjugation matrix’ B. See Appendix A for our conventions for spinors and gamma
matrices, the relation between 10D and 6D notations, and the definitions of C and B.
We note the reader that in this paper, the transpose of the spinor indices is not indicated
explicitly. For example, θ+ C[θ− , φ1 ] in Eq. (2.2) means θ+t C[θ− , φ1 ] where t denotes the
transpose of spinor indices (but not of gauge indices).
The additional part S5 contains the 0–4 sector fields (vI , χ). These fields are given by
the hypermultiplets of the 6D theory which consist of 2 complex bosons vI (I = 1, 2)
and a complex spinor χ with positive 6D chirality (γ̄ χ = +χ ). Both of them are in
the bi-fundamental rep. of U(N) × U(N4 ). Covariant derivatives are defined as D0 vI =
∂0 vI + iAvI . We remark here that only half of the 0–0 sector fermions (θ− ) couple to the
0–4 sector fields.
This model has half the amount of supersymmetry as the BFSS model. The SUSY
parameter η is a complex 6D spinor with negative chirality, thus the number of (real)
supercharges is 8. The SUSY transformation law is given as follows. The 0–0 sector fields
transform as
i
δA = δ (0) A = η† θ− − θ−† η , δXa = δ (0) Xa = i η† γ 0 γ a θ− − θ−† γ 0 γ a η ,
λ
δφ1 = δ (0) φ1 = −2iη†C ∗ θ+† , δφ2 = δ (0) φ2 = −2iη† γ 0 θ+ ,
i i
δθ+ = δ (0) θ+ = D0 φ̄1 C ∗ η† + D0 φ2 γ 0 η + Xa , φ̄1 B ∗ γ a∗ η† + [Xa , φ2 ]γ a η,
λ λ
δθ− = δ (0) θ− + δ θ− (2.4)
where
i
δ (0) θ− = D0 Xa γ 0 γ a η + [Xa , Xb ]γ ab η
2λ
i i
− φ1 , φ̄1 + φ2 , φ̄2 η + φ̄1 , φ̄2 B ∗ η† , (2.5)
2λ λ
i 2i † ∗ †
δ θ− = −v1 v1† + v2 v2† η − v2 v1 B η . (2.6)
λ λ
We have denoted δ (0) the part of the transformation laws which does not contain 0–4 sector
fields in the RHS. It is the same as the transformation law for the BFSS theory, except for
the fact that the parameter is restricted to γ̄ η = −η now. Only δθ− has extra contribution
δ θ− containing 0–4 sector fields. The transformation law for the 0–4 sector fields is as
follows
√ √
δv1 = i 2 η† γ 0 χ, δv2 = −i 2 ηCχ,
√
√ 0 ∗ †
2i
δχ = − 2 D0 v2 γ B η + D0 v1 γ η − 0
Xa γ a v2 B ∗ η† + v1 η . (2.7)
λ
3. One-loop effective action
3.1. Method of the perturbative calculation
We obtain the effective action as the one-loop determinant (Berezinian) of the

fluctuations. It is obtained from the part of the action quadratic in the fluctuations
†
S (quad) = v † Kbos v + v † Kmix χ + χ † Kmix v + χ † Kfermi χ
† −1 −1 †
= v + χ † Kmix Kbos Kbos v + Kbos Kmix χ
−1 †
+ χ † Kfermi − Kmix Kbos Kmix χ
as
−1 −1 †
S (1) = ln Det Kbos − ln Det Kfermi − ln Det 1 − Kfermi Kmix Kbos Kmix . (3.1)
We perform the calculations in the Euclidean formulation (τ = it, S (M) → S (E) = iS (M) )
and rotate the result to Minkowski. Note that indices and the integration symbols are
implicit in the above equations and Eq. (3.2) below. The first two terms of (3.1) are the
ones obtained in the previous paper [16]. We calculate
−1 −1 †
Γmix ≡ Tr ln 1 − Kfermi Kmix Kbos Kmix
∞

1 −1 †
= Tr K K K −1 Kmix (3.2)
bos mix fermi
=1
in this paper.
The calculation is performed using a perturbative expansion. We first separate the
backgrounds for bosonic matrices as

(X0 , Xm , Xa ) = X0 , X
m , ra + X
a , (3.3)
where ra are constants proportional to the identity matrix. (X0 is the Euclidean
continuation of A which is defined by X0 = −iλA.) We divide the action S (quad) into
or fermion background Θ, and the interaction
the ‘free’ part, which does not contain X
vertices. The free part is given by

1 r2 1 a
S (free) = dτ v †
−∂τ + 2
2
v + χ †
−∂τ + γ̃ ra χ , (3.4)
gs s λ λ
where γ̃ a = γ 0 γ a . The propagators are determined from (3.4) as

†
B dk eik(τ −τ )
vI,AA(τ )vJ,B B(τ ) = (gs s )δ δ δ
I J AB A
2π k 2 + r 2

≡ (gs s )δ I J δ AB δ AB /(τ − τ ). (3.5)

†
B
dk eik(τ −τ )
χα,AA(τ )χβ,B (τ ) = (g
s s )δ AB A
δ (/r + ik)αβ
B 2π k 2 + r 2

= (gs s )δ AB δ AB ∂τ + /r αβ /(τ − τ ), (3.6)
where A, B are the U(N) indices, A, B are the U(N4 ) indices and α, β, . . . (= 1, . . . , 8)
are the 6D spinor indices. We have also defined r = r/λ and /r = γ̃ a ra /λ.
Boson–boson vertices are

(int) 1 1
Sbos = dτ 2 vI† VI J (τ )vJ , (3.7)
gs s λ
where

V11 = 2ra X a + Xa2 + X
02 − iλ∂τ X 0 ∂τ + 1 φ1 , φ̄1 + φ2 , φ̄2 ,
0 − 2iλX
2
1

V22 = 2ra Xa + Xa + X0 − iλ∂τ X0 − 2iλX0 ∂τ −
2 2 φ1 , φ̄1 + φ2 , φ̄2 ,
2
V12 = [φ1 , φ2 ], V21 = − φ̄1 , φ̄2 . (3.8)
Fermion–fermion vertices are

(int) 1 1
0 χ.
Sfermi = dτ χ † γ̃ a Xa − iX (3.9)
gs s λ
Boson–fermion mixed vertices are
√
(int) 1 2 †
Smix = dτ χα LI α vI + vI† L†I α χα , (3.10)
gs s λ
where

L1α = γ 0 θ− α , L†1α = − θ−† γ 0 α ,

L2α = − C ∗ θ−† α , L†2α = θ− C α . (3.11)
(int) (int) (int)
The quadratic part of the action is given by S (quad) = S (free) + Sbos + Sfermi + Smix , and
the effective action is derived from the general expression (3.2). Note that Kbos (or Kfermi )
is read from (3.4) and (3.7) (or from (3.4) and (3.9)). Also, Kmix is read from (3.10).
We treat the vertices as an expansion around a reference time τ and obtain the effective
action as a double expansion in the number of vertices and the number of derivatives,
following the procedure described in Ref. [16]. Explicit form of the two-fermion (two L)
terms of the effective action is given as1
∞
∞
n+m+2

1 2
Γθ 2 = (−1)(n+m) m+2n+2
Di ! λ
m,n=0 Di =0 i=2

) (Dn+1 )
(D ) (D ) †(D
× dτ Tr VI1 ,I2 VI2 ,I23 · · · VIn ,Inn+1 LIn+1n+m+1 Xa1 a(D
···X n+m ) (Dm+n+2 )
LI1 ,β
,α m

dk 1 1 Dn+m+1 / r + ik a1
× (i∂k ) D2
· · · (i∂k ) γ̃ (i∂k )Dn+1
2π k 2 + r 2 k2 + r 2 k2 + r 2

/r + ik 1
× · · · γ̃ am (i∂k )Dn+m 2 (i∂k ) Dn+m+2
. (3.12)
k + r 2 k2 + r 2 αβ
1 See Ref. [16] for the terms which do not contain fermions and for more details on the derivation.
For Di = 0, this reduces to
Γθ(d=0)
2
∞

2
= (−1)(n+m) dτ Tr VI1 ,I2 · · · VIn ,In+1 L†In+1 ,α X am LI1 ,β
a1 · · · X
λm+2n+2
m,n=0

dk 1 /r + ik a1 /r + ik r + ik
am / 1
× γ̃ · · · γ̃ .
2π (k 2 + r 2 )n k 2 + r 2 k2 + r 2 k 2 + r 2 αβ k 2 + r 2
(3.13)
In the above equations, we have assumed X 0 = 0. The dependence of the effective action
on X0 can be recovered by replacing ∂τ with the covariant derivative ∂τ + i[X 0 , ], or
can be directly calculated with a slight modification of the above prescription due to the
0 ∂τ vI .
presence of the derivative interaction vI X
3.2. Fermionic terms of the one-loop effective action
Following the method explained above, we calculate the part of the effective action
containing fermionic backgrounds. We describe the θ 2 terms in detail in Section 3.2.1,
and briefly discuss the terms with more θ ’s in Section 3.2.2. The results are presented in
obtaining the Minkowskian effective action from Γ given
Euclidean signature. The rule for
below is to replace dτ with dt, and Dτ with −iD0 .
Interaction starts at 1/r 3 order. Note that only θ− (but not θ+ ) appear non-trivially in
the effective action, for only θ− couple to the quantum fields v or χ as we have seen in
Section 2.
3.2.1. θ 2 terms
N , d) up to N +d < 4,
We summarize the result of θ 2 terms of the effective action Γθ 2 (Xi
where N and d are the numbers of X i ’s and derivatives contained in Γ , respectively. We
present the result by assembling the terms with the same property.
There are two sets of terms which can be regarded as having similar structures as the θ−
part of the classical action:
Γθ 2 (d = 0)A

N4 1
=− dτ Tr θ−† γ̃ a X a , θ− − X a , θ−† γ̃ a θ− (3.14)
4 r3

3N4 rb
+ dτ STr θ−† γ̃ a X a , θ− − X b
a , θ−† γ̃ a θ− ; X (3.15)
4 r 5

3N4 1
+ dτ STr θ−† γ̃ a X a , θ− − X a , θ−† γ̃ a θ− ; (X
b )2 (3.16)
8 r5

15N4 rb rc
− dτ STr θ−† γ̃ a X a , θ− − X b , X
a , θ−† γ̃ a θ− ; X c , (3.17)
8 r 7

N4 λ 1
Γθ 2 (d = 1)A = 3
dτ Tr θ−† Dτ θ− − Dτ θ−† θ− (3.18)
4 r

3N4 λ rb
− b
dτ STr θ−† Dτ θ− − Dτ θ−† θ− ; X (3.19)
4 r5

3N4 λ 1 2
− dτ STr θ−† Dτ θ− − Dτ θ−† θ− ; X b (3.20)
8 r 5

15N4 λ rb rc
+ b , X
dτ STr θ−† Dτ θ− − Dτ θ−† θ− ; X c . (3.21)
8 r7
Here STr(K1 · · · Km ; y1, y2 , . . . , yn ) means that the trace operation is taken after sym-
metrizing all Ki ’s and yj ’s but keeping the location of K1 and the order of Ki ’s. Note
a , X
that each Ki is X m , θ− , or commutators (or covariant derivatives) of them. For exam-
ple,

STr θ−† Dτ θ− ; X c = 1 Tr θ−† Dτ θ− X
b , X b X
c + θ−† X
b X
c Dτ θ− + θ−† Xb Dτ θ− X
c
6

+ θ−† Dτ θ− X c X
b + θ−† X
c X
b Dτ θ− + θ−† Xc Dτ θ− X
b .
The terms in Γθ 2 (d = 0)A and Γθ 2 (d = 1)A are given by the non-Abelian Taylor expansion
of the leading terms (3.14) and (3.18)

1 1 3 1 δab ra rb
→ − ra Xa + −3 + 15 Xa Xb + · · · ,
r3 r3 r5 2 r5 r7
and following the symmetrized trace prescription [1,2], except for the fact that (3.16) and
(3.20) are of the form STr( ∗ ; (X b , X
b )2 ) rather than STr( ∗ ; X b ). Further discussion
on the ordering of matrices will be given in the next section when we consider the
supersymmetry of the effective action.
Terms with multiple number of gamma matrices are given as follows:
Γθ 2 (d = 0)B

3N4 ra
= dτ Tr θ−† γ̃ aa1 a2 X a1 , X
a2 θ− + θ−† γ̃ aa1 a2 θ− Xa1 , X
a2 (3.22)
16 r 5

3N4
+ dτ Tr θ−† γ̃ a1 a2 a3 Xa1 , X
a2 X a3 θ−
16 r 5
+ θ−† γ̃ a1 a2 a3 θ− X a2 X
a1 , X a3 (3.23)

15N4 ra rb
− dτ STr θ−† γ̃ aa1 a2 X a1 , Xa2 θ−
7
16 r
+ θ−† γ̃ aa1 a2 θ− X a1 , X b ,
a2 ; X (3.24)
Γθ 2 (d = 1)B

3N4 λ rb
=− a θ− + θ−† γ̃ ab θ− Dτ X
dτ Tr θ−† γ̃ ab Dτ X a (3.25)
8 r 5

3N4 λ 1
− b Dτ X
dτ Tr θ−† γ̃ ab X a θ− + θ−† γ̃ ab Dτ X
a X
b θ−
16 r 5
b Dτ X
+ θ−† γ̃ ab θ− X a + θ−† γ̃ ab θ− Dτ Xa X
b (3.26)

15N4 λ rb rc
+ dτ STr θ−† γ̃ ac Dτ Xa θ− + θ−† γ̃ ac θ− Dτ X b .
a ; X (3.27)
8 r 7
In addition, there are commutator corrections to Γθ 2 (d = 0, 1)A,B :

Γθ 2 (d = 0)C

3N4 † b
= dτ Tr θ − γ̃ a , X
X a , X
b θ− − θ−† γ̃ b θ− Xa , Xa , X
b (3.28)
16r 5
Γθ 2 (d = 1)C

3N4 λ
= dτ Tr θ−† X a , θ− , Dτ Xa + θ−† , X a , Dτ X a θ− (3.29)
16r 5

N4 λ
+ dτ Tr Dτ θ−† γ̃ ab X b , θ− − θ−† , X
a , X a , X
b γ̃ ab Dτ θ− .
32r 5
(3.30)

Terms containing Xm are collected as:

3N4 ra †
Γθ 2 (d = 0)φ = − dτ Tr Θ− Xm , X
n Γ 0a Γ mn Θ− (3.31)
16 r 5

3N4 †
− dτ Tr Θ− m , X
X n Γ 0a Γ mn Θ− X a (3.32)
16 r 5

15N4 ra rb †
+ dτ STr Θ− m , X
X n Γ 0a Γ mn Θ− ; X b , (3.33)
16 r 7
Γθ 2 (d = 1)φ

N4 λ † mn † mn
= 5
dτ Tr Dτ Θ− Xm , Xn Γ Θ− − Θ− Xm , Xn Γ Dτ Θ− . (3.34)
16 r
Note that we have used the notation of 10D gamma matrices and spinors to write (3.31)–
(3.34). The expressions in 6D notation is given by substituting the particular representation
(A.2) and (A.3) of 10D gamma matrices and the parametrization (A.4) of Majorana–Weyl
spinor Θ setting θ+ = 0. For example, (3.31) is written in 6D notation as

3N4 ra
5
dτ Tr θ−† γ̃ a θ− φ1 , φ̄1 + φ2 , φ̄2 + θ−† φ1 , φ̄1 + φ2 , φ̄2 γ̃ a θ−
16r

− 2θ− B φ1 , φ2 γ̃ a θ− − 2θ−† φ̄1 , φ̄2 γ̃ a B ∗ θ−† .
Terms in Γθ 2 (d = 0)B , Γθ 2 (d = 1)B and Γθ 2 (d = 0)φ have the structure of non-Abelian
Taylor expansion of ra /r 5 :

ra ra δab ra rb
→ + − 5 Xb + · · · .
r5 r5 r5 r7
They obey symmetrized trace prescription except for 1/r 5 -terms (3.23), (3.26) and (3.32),
as in the case of Γθ 2 (d = 0, 1)A .
There also exist terms with two derivatives:

N4 λ2
Γθ 2 (d = 2) = dτ Tr Dτ θ−† γ̃ a Xa , Dτ θ− − 2θ−† γ̃ a Dτ2 X
a , θ− . (3.35)
8r 5
This completes the analysis of Γθ 2 (X N , d) for N + d < 4.

i
To summarize, we obtained
Γθ 2 = Γθ 2 (d = 0, 1)A + Γθ 2 (d = 0, 1)B + Γθ 2 (d = 0, 1)φ

+ Γθ 2 (d = 0, 1)C + Γθ 2 (d = 2) + O Dτd X N θ 2; N + d 4 (3.36)
for two-fermion part of the one-loop effective action.
3.2.2. θ 2n terms
The leading terms of Γθ 2n are given by
n
2 1

Γθ 2n (X) , d = 0 = N4 2
0
dτ Tr L†In α1 LI1 β1 L†I1 α2 LI2 β2 · · · L†In−1 αn L†In βn
λ n

dk /r + ik /r + ik /r + ik
× ··· 2
2π k 2 + r 2 α1 β1 k 2 + r 2 α2 β2 k + r 2 αn βn

N4 λn−1
∝ 3n−1 dτ Tr θ−2n . (3.37)
r
For example, four-fermion terms without Xi insertions are given as

0, d = 0
Γθ 4 (X)

N4 λ 1
=− dτ Tr θ−† θ− θ−† θ− + θ− θ−† θ− θ−† − 2 θ− Bθ− θ−† B ∗ θ−†
16 r 5

5N4 λ ra rb
+ 7
dτ Tr θ−† γ̃a θ− θ−† γ̃b θ− + θ− γ̃a∗ θ−† θ− γ̃b∗ θ−†
16 r

− 2 θ− B γ̃a θ− θ−† γ̃b B ∗ θ−† . (3.38)
This does not vanish unlike the case of Γθ 2 .
4. Supersymmetry of the effective action
In this section, we discuss the supersymmetry of the effective action. The one-loop
effective action which was obtained by integrating out the 0–4 sector fields satisfies the
Ward identity corresponding to the SUSY invariance of the classical action of BD matrix
theory. As we show in Appendix B, at the one-loop level, the Ward identity can be written
in the form
δ (1) S0 + δ (0) S (1) = 0 (4.1)

which states the invariance of the effective action under an effective SUSY transformation.
Let us explain the meaning of (4.1). The tree action S0 is the part of BD action
containing only the 0–0 sector fields which is given in (2.2) and is the same as the BFSS
action. It is invariant under the SUSY transformation which does not involve 0–4 sector
fields δ (0)S0 = 0. The transformation law for δ (0) , which we call the tree-level SUSY
transformation is given by (2.4) with a replacement Xa → X a . Remember that we have
divided the background Xa as Xa = ra + X a in (3.3). In our construction, only X
a is
transformed and ra is kept fixed.2 The discussion in Appendix B shows that δ (1) is given
simply by the one-loop expectation value of the SUSY transformation for the classical
action. Since only δθ− has the part containing the quantum fields (δ θ− defined in (2.6)),
only δ (1) θ− is non-zero and is given by
δ (1) θ− = δ θ−
i † †
= − v1 v1 + v2 v2 η− − 2 v2 v1† B ∗ η† . (4.2)
λ
4.1. Explicit forms of the one-loop corrected SUSY transformation
We give the one-loop corrected SUSY transformation of θ− by explicitly calculating

(4.2). Results at order 1/r 3 and 1/r 4 are given respectively as

1 (1) N4
δ θ− = i 3 φ1 , φ̄1 + φ2 , φ̄2 η − 2 φ̄1 , φ̄2 B ∗ η† , (4.3)
gs ls 1/r 3 4r

1 (1)
δ θ−
gs ls 1/r 4
3N4 ra
a η − 2 Sym φ̄1 , φ̄2 ; X

a B ∗ η†

= −i Sym φ 1 , φ̄ 1 + φ 2 , φ̄ 2 ; X
4 r5
(4.4)
3N4 λ ra † a a∗ †
∗ †
+i θ γ̃ θ− − θ− γ̃ θ− η − 2 θ− B γ̃ θ− B η . a
(4.5)
8 r5 −
Here Sym(K1 · · · Km ; y1 , . . . , yn ) is the symmetrization of all Ki ’s and yj ’s without
changing the order of Ki ’s. Note that there is a relation

STr(K1 K2 · · · Km ; ∗ ) = Tr K1 Sym(K2 · · · Km ; ∗ ) .
(a)
For order 1/r 5 , we divide the result into two parts δ (1)θ− |1/r 5 = δ (1)θ− |1/r 5 +
(b)
δ (1) θ− |1/r 5 :

1 (1) (a)
δ θ−
gs ls 1/r 5
3N4 2
a η − 2 Sym φ̄1 , φ̄2 ; X
2 ∗ †
a B η
= −i 5 Sym φ1 , φ̄1 + φ2 , φ̄2 ; X
8r
(4.6)
15N4 ra rb
a , X

b η
+i Sym φ1 , φ̄1 + φ2 , φ̄2 ; X
8 r7 ∗ †
a , X
− 2 Sym φ̄1 , φ̄2 ; X b B η (4.7)
3N4 λ † b
b γ̃ b∗ θ−† η − 2 θ− B γ̃ b X

b θ− B ∗ η†

+i θ− γ̃ Xb θ− − θ− X (4.8)
8r 5
2 This assignment is possible without losing generality. Note that we have not imposed the tracelessness for
a .
X
15N4 λ ra rb
b η
−i 7
Sym θ−† γ̃ a θ− − θ− γ̃ a∗ θ−† ; X
8 r
− 2 Sym θ− B γ̃ a θ− ; Xb B ∗ η† (4.9)
and

1 (1) (b)
δ θ−
gs ls 1/r 5
3N4
=i 5
[φ1 , φ2 ], φ̄1 , φ̄2 η + φ̄1 , φ̄2 , φ1 , φ̄1 + φ2 , φ̄2 B ∗ η†
16 r
(4.10)
N4 λ2 2
−i D φ1 , φ̄1 + φ2 , φ̄2 η − 2D02 φ̄1 , φ̄2 B ∗ η† (4.11)
16 r 5 0
N4 λ2 †
− 5
θ− D0 θ− − D0 θ− θ−† η − 2(θ− BD0 θ− )B ∗ η† . (4.12)
4r
Note that (4.3), (4.4), (4.6) and (4.7) correspond to the first three terms of non-Abelian
Taylor expansion of 1/r 3 around ra . Also, (4.5), (4.8) and (4.9) correspond to the
expansion of ra /r 5 .
Our corrected SUSY transformation closes to the translation plus a field dependent
gauge transformation [17]. By using the equations of motion for the effective action
S0 + S (1) , we have shown
(0)
δ8 + δ8(1) , δη(0) + δη(1) φ 1/r 4 ∼ DM φ 8̄Γ M η + (EOM for φ) + O (gs ls )2
up to the order 1/r 4 for any field φ in S0 + S (1) .
4.2. On the structure of the supersymmetric action
In this subsection, we try to clarify the structure of the supersymmetric action which
we have obtained, by identifying the blocks which transform within themselves. Firstly,
since ra is kept fixed in the effective SUSY transformation, the invariance of the one-loop
effective action (4.1) holds at each order of 1/r. In addition, the effective action at order
1/r 5 is further divided into two invariant blocks.
We shall examine the θ 0 - and θ 2 -terms at each order of 1/r.3 We include the bosonic
terms obtained in the previous paper [16]. Order 1/r 3 terms of the effective action

λ 1
Γ1/r 3 = 3 dt Tr D0 X a D0 Xa + 1 X b X
a , X b − 1 [φ1 , φ2 ] φ̄1 , φ̄2
a , X
r 4 8λ2 4λ2

1 2 i † 1 † a
+ φ , φ̄ + φ , φ̄ − θ D θ − − θ γ̃ X , θ − (4.13)
2 − 2λ −
1 1 2 2 0 a
16λ2
satisfy (4.1) along with (4.3).
3 Note that Γ which are discussed in this section denote the parts of the Minkowskian effective action S (1) .
Order 1/r 4 terms of the effective action

ra 1
Γ1/r 4 = −3N4 λ 5 dt STr D0 X c D0 Xc + 1 X c , X
d X c , X
d
r 4 8λ2

1 1 2
− 2 [φ1 , φ2 ] φ̄1 , φ̄2 + φ1 , φ̄1 + φ2 , φ̄2 ; Xa
4λ 16λ2
i
+ 8aa1 a2 a3 a4 STr X a2 , X
a3 D0 X a4 ; X a1
16λ
i
+ STr − θ−† D0 θ− − D0 θ−† θ−
4

1 † c
† c

− θ γ̃ Xc , θ− − Xc , θ− γ̃ θ− ; Xa
4λ −

i † ab b γ̃ ab θ−

+ Tr D0 Xb θ− γ̃ θ− + θ−† D0 X
8
1 † abc
b , X

c γ̃ abc θ−

− Xb , Xc θ− γ̃ θ− + θ−† X
16λ
1
− φ1 , φ̄1 + φ2 , φ̄2 θ−† γ̃ a θ− + θ−† φ1 , φ̄1 + φ2 , φ̄2 γ̃ a θ−
16λ

− 2[φ1 , φ2 ]θ− B γ̃ a θ− − 2 φ̄1 , φ̄2 θ−† B ∗ γ̃ a θ−† (4.14)
satisfy (4.1) along with (4.4), (4.5).

By examining δ (0) S (1) , we find that the following terms of the 1/r 5 effective action
transform among themselves (with a certain class of terms in δ (1) S0 ).

15ra rb 3δab 1
(I)
Γ1/r 5 = N4 λ − dt STr D0 X c D0 Xc + 1 X d X
c , X d
c , X
2r 7 2r 5 4 8λ 2

1 1 2
− 2 [φ1 , φ2 ] φ̄1 , φ̄2 + φ ,
1 1φ̄ + φ ,
2 2φ̄ ; X a , X b
4λ 16λ2
i
+ 8aa1 a2 a3 a4 STr X a2 , X
a3 D0 X a4 ; Xa1 , X
b
10λ
i
+ STr − θ−† D0 θ− − D0 θ−† θ−
4

1 † c
c , θ−† γ̃ c θ− ; X a , X
b
− θ− γ̃ Xc , θ− − X
4λ
i
+ STr D0 X c θ−† γ̃ ac θ− + θ−† D0 X b
c γ̃ ac θ− ; X
4
1
− STr X a1 , Xa2 θ−† γ̃ aa1 a2 θ− + θ−† X a2 γ̃ aa1 a2 θ− ; X
a1 , X b
8λ
1
− STr φ1 , φ̄1 + φ2 , φ̄2 θ−† γ̃ a θ− + θ−† φ1 , φ̄1 + φ2 , φ̄2 γ̃ a θ−
8λ

− 2[φ1 , φ2 ]θ− B γ̃ a θ− − 2 φ̄1 , φ̄2 θ−† B ∗ γ̃ a θ−† ; X b . (4.15)
(II)
The other block Γ1/r 5 consists of the rest of the terms of 1/r 5 effective action. They are
written with extra number of commutators compared to (4.15). We do not present all the
terms explicitly, but give an example. The part which contain six X a ’s (and no derivatives
or fermions) takes the form

(II) 3N4 λ 1
a , X

c X
b , X

a , X c
b , X

Γ1/r 5 X a
6 = − 5
dt − 2
Tr X
2r 24λ
1
+ Tr X b , Xa , X
c X a , X b , X
c
12λ 2

1
− Tr a , X
X a , X
c X b , X b , X
c . (4.16)
24λ2
(I) (II)
Criterion for discriminating Γ1/r 5 and Γ1/r 5 is given by the following number associated
with each term:
1
n = d + nθ + nc . (4.17)
2
Here, d is the number of derivatives, nθ is the number of fermions and nc is the
number of commutators in the symmetrized trace. (Symmetrization is applied regarding
the commutator as a single unit.) Under δ (0) S (1) , the number n is preserved (uniformly
increase by 1/2) as we see from the transformation law (2.4), (2.5). Also, terms in δ (1) S0
are classified in the same way. Thus, terms which are connected by SUSY transformation
must have the same n. The first two terms of the RHS of (4.17) is called the ‘order’ [18]
and used to specify the SUSY invariants in the case of Abelian v.e.v. Our discussion here
(I)
is the non-Abelian generalization of that concept. Terms in Γ1/r 3 , Γ1/r4 and Γ1/r 5 have
(II)
n = 2, and terms in Γ1/r 5 have n = 4.
(I)
We note here that the bosonic terms in Γ1/r 3 , Γ1/r4 and Γ1/r 5 are the ones which result
from the Taylor and Van Raamsdonk’s proposal when we apply it to L5-brane background.
We can see that the proposed currents which couple to SUGRA fields have n = 2, from
the explicit forms in Refs. [1,16]. Bosonic terms of the matrix expressions for multipole
moments are obtained from the currents by inserting X i with symmetric ordering, thus also
have n = 2. For a detailed description of the Taylor and Van Raamsdonk’s proposal applied
to the present background, see our previous paper [16].
(I)
We also note that the fermionic terms in Γ1/r 3 , Γ1/r4 and Γ1/r 5 are not of the same
forms as the ones arising from the above proposal, which is based on the analysis of BFSS
model. Whether the two forms of the fermionic terms are physically equivalent or not is
not clear at present.
5. Discussion
We have calculated the one-loop effective action of Berkooz–Douglas matrix theory by

integrating out the 0–4 sector fields. Extending the analysis of Ref. [16], we obtained the
fermionic part of the effective action of multiple D0-branes in the longitudinal 5-brane
background. As we have seen in Section 4, the action is manifestly invariant under an

effective SUSY transformation, which is the consequence of the SUSY invariance of the
classical action of BD matrix theory. The transformation law is given by the expectation
value of the SUSY transformation of classical action evaluated at one-loop order. We
have studied the transformation law and identified the structure which are closed under
the effective SUSY transformation.
Primary motivation of our study is to understand the dynamics and the structure of
symmetries of this matrix model. In addition, we regard our analysis as a first step
towards constructing the supersymmetric action of multiple D-branes in the supergravity
backgrounds. We should note that from the 10D perspective, our model is nothing but the
SYM which takes into account only the lowest modes of open strings. Thus, the effective
action is guaranteed to be valid only when the distance between D0-branes and D4-branes
is short. However, some of the terms in the effective action continues to be valid when the
separation becomes large. It is well known that for the (∂t X i )2 -term in the Abelian case,
the SYM effective action gives exact results, due to the cancellation of the open-string
higher modes [19]. It may be the case that same kind of non-renormalization properties
exist for all the other n = 2 terms which we identified in Section 4.2. Further analysis of
the terms in this class with higher orders in θ or 1/r should be important in this regard.
Of course, to determine which terms of the SYM effective action is valid for large
separation, we must study the cylinder amplitudes between the branes, or prove non-
renormalization theorems in SYM. For the case of multiple branes, both of them are not
well-understood at present. Those are important subjects for future studies.
As mentioned in the introduction, κ-symmetric Born–Infeld action for a single D-brane
can be written in arbitrary background, in principle. We will be able to understand
the structure of the supersymmetric action further by comparing the Abelian part of
our effective action with the gauge-fixed Born–Infeld action of D0-branes in D4-brane
background. Especially, it will be interesting if the difference between the fermionic part
in the effective action of BD matrix theory and the ones present in the proposal based on the
analysis of BFSS model is explained by a difference in the gauge choice for κ-symmetry.
Two-loop study of BD matrix model should give important implications for the
connection between matrix models and gravity. In the first place, we must clarify to what
extent the matrix theory reproduces supergravity result in the case of Abelian matrix
background. Two-loop effective action will also shed light on the coupling of multiple
D-branes to non-weak background fields.
Finally, it is an interesting problem to find classical solutions of the effective action
which preserve SUSY. As argued by Myers [2], background 4-form field strength produced
by D4-branes should give rise to non-Abelian configuration of D0-branes. It will be
possible to discuss the BPS property of those configurations from the effective action which
we have obtained.
Acknowledgements
We would like to thank Y. Kazama, T. Muramatsu, N. Sakai, T. Yoneya for discussions

and comments.
Appendix A. Convention for the representations of spinors and gamma matrices
In this paper, we mainly use the complex 6D Weyl spinors. We summarize here our
conventions for spinors and gamma matrices. The fermions from the 0–0 sector are given
by Majorana–Weyl spinors in 10D. We explain that the fermionic terms in S0 in (2.2) is
given from the familiar 10D notation by choosing an explicit representation.
We consider fermionic matrix field Θ which satisfy the 10D Majorana–Weyl condition
Γ (10)Θ = Θ, Θ † = B (10)Θ, (A.1)
where Γ (10) = Γ 0Γ 1 · · · Γ 9 is the 10D chirality matrix, and B (10) is defined by
B (10) Γ M B (10)−1 = −Γ M∗ (M = 0, . . . , 9).
We make the SO(5, 1) × SO(4) decomposition of the 10D gamma matrices as follows.
Γ µ = γ µ ⊗ γ̂ (µ = 0, 5, . . . , 9),
Γ m
= 1 ⊗ γ̂ m
(m = 1, . . . , 4), (A.2)
where we have defined γ̂ = γ̂ 1 γ̂ 2 γ̂ 3 γ̂ 4 . The SO(5,1) gamma matrices γ µ have 8×8
components and the SO(4) gamma matrices γ̂ m have 4×4 components. We further assume
the explicit representation for the SO(4) gamma matrices

0 σm
γ̂ =
m
, (A.3)
−σ̄ m 0
where σi (i = 1, 2, 3) are Pauli matrices, and σ4 = i1. We also define σ̄i = −σi and
σ̄4 = σ4 .
The chirality in 10D is the product of 6D and 4D chiralities, and in the above
representation of gamma matrices,

12×2 0
Γ(10) = γ̄ ⊗
0 −12×2
where γ̄ = γ 0 γ 5 · · · γ 9 is the 6D chirality matrix. Consequently, 10D Weyl spinor Θ is
decomposed into two pairs of 6D Weyl spinors
 
θ+,1
θ 
Θ =  +,2  .
θ−,1
θ−,2
The matrix B (10) also allows the SO(5,1)×SO(4) decomposition
B (10) = B (6) ⊗ B (4) ,
where B (6) and B (4) satisfy
B (6) γ µ B (6)−1 = −γ µ∗ , B (4) γ̂ m B (4)−1 = −γ̂ m∗ .
We note here the relation for B (6)
B (6)T = −B (6), B (6)∗ = −B (6)−1.
In the text, we omit the superscript (6) on B (6) .

In the representation (A.3), B (4) is given by

8 0 0 1
B =(4)
, where 8 = ,
0 8 −1 0
thus, 10D Majorana condition (A.1) in 10D becomes the SU(2) Majorana condition in 6D.
 † 
θ+,1  (6) 
B θ+,2
 † 
 θ+,2   −B (6)θ+,2 
Θ† =   
 †  =  B (6) θ
.

 θ−,1  −,2
θ† −B θ−,1
(6)
−,2
Since we prefer using unconstrained fermions to perform the loop calculations, we
explicitly eliminate half of the components of Θ (θ+,2 and θ−,2 ) using the SU(2) Majorana
conditions. That is, we take
 
θ+
 −B (6)∗ θ † 
Θ = 
+ .
 (A.4)
θ−
−B (6)∗ θ−†
The fermionic part of the action in (2.2) is obtained from the 10D covariant form

1 i 1
dt Tr
ΘΓ 0
D0 Θ +
ΘΓ i
[Θ, Xi ]
gs s 2 2λ
by taking the above representation of gamma matrices and substituting (A.4).
Appendix B. Derivation of the SUSY invariance of the one-loop effective action
The one-loop effective action of the BD matrix theory is given by integrating out the
0–4 sector fields from the classical action S0 [φ] + S5 [φ, ϕ].

e−(S0 [φ]+S [φ]) = Dϕ e−(S0 [φ]+S5 [φ,ϕ]) .
(1)
(B.1)
Note that we denote the 0–0 sector fields by φ and 0–4 sector fields by ϕ, and that the
integration dτ is implicit throughout this appendix.
As a consequence of the SUSY invariance of the classical action, the following Ward
identity holds:

δS0 [φ] δS5 [φ, ϕ] δS5 [φ, ϕ] −(S0 [φ]+S5 [φ,ϕ])
0 = Dϕ δη φ + δη φ + δη ϕ e , (B.2)
δφ δφ δϕ
where δη φ and δη ϕ are the classical SUSY transformations given in (2.4)–(2.7). We shall
(0)
denote the part of δη φ which contain only the 0–0 sector fields by δη φ, and the part

containing 0–4 sector fields by δη φ.
We first note that the last term of (B.2) vanishes, for it is the infinitesimal form of the
invariance of the integral under the change of variables

−S5 [φ,ϕ] −S5 [φ,ϕ+δϕ]
Dϕ e = D(ϕ + δϕ) e = Dϕ e−S5 [φ,ϕ+δϕ]
where we have taken into account the invariance of the measure.

The first term of (B.2) gives the variation of S0 with respect to the expectation value of
the classical SUSY transformation δφ (times e−(S0 +S ) )
(1)

δS0 −(S0 +S5 ) (0) δS0 [φ] −(S [φ]+S (1) [φ])
Dϕ δη φ e = δη φ + δη(1) φ e 0
δφ δφ
where

Dϕ δη φ e−S5
δη(1) φ ≡ . (B.3)
Dϕ e−S5
Finally, if we evaluate the second term of (B.2) to the one-loop order, it reduces to

δS5 [φ, ϕ] −(S0 [φ]+S5 [φ,ϕ]) δS5 [φ, ϕ] −(S0 [φ]+S5 [φ,ϕ])
Dϕ δη φ e = δη(0) φ Dϕ e .
δφ δφ
(B.4)
For the present model,

δS5 [φ, ϕ] −(S0 [φ]+S5 [φ,ϕ])
Dϕ δη φ e (B.5)
δφ
which would be present in the RHS of (B.4) has at least two loops, for
δS5 [φ, ϕ]
δη φ
δφ
is a four-point vertex as we see from the explicit forms of δ φ (2.6) and S5 (2.3). We
further note that the RHS of (B.4) gives the variation of the one-loop effective action (times
e−(S0 +S ) ) as we see from
(1)
5 −S5
δS (1) Dϕ δS
δφ e
=
δφ Dϕ e−S5
which results from the definition (B.1) of the effective action.
Taking these facts into account, to the one-loop order, (B.2) is rewritten as
δS0 [φ] δS0 [φ] δS (1) [φ]
0 = δη(0) φ + δη(1) φ + δη(0) φ , (B.6)
δφ δφ δφ
which shows that the effective action S0 + S (1) is invariant under effective SUSY
(0) (1)
transformation δη φ + δη φ.
We note that the present discussion is not directly applicable for showing the SUSY of
the effective action at 2-loop order and beyond. At these orders, (B.4) should no longer be
valid, due to the contribution from (B.5).
In addition, we remark here that our simple discussion which involve only the Ward
identity for the SUSY is due to the fact that the gauge fields (which belong to the 0–0
sector) are not integrated. When they are to be integrated, as in the case of the BFSS model,
analysis of the Ward identity for the SUSY plus the one for the gauge symmetry is required
for studying the invariance of the effective action under effective SUSY. It is essentially due
to the fact that the gauge fixing term and the ghost action breaks SUSY. The formalism is
developed in Ref. [20] and explicit form of the effective SUSY transformation laws for the
one-loop effective action of BFSS model (for Abelian v.e.v.) is given in Ref. [21].
References
[1] W. Taylor, M. Van Raamsdonk, Supergravity currents and linearized interactions for matrix theory
configurations with fermionic backgrounds, JHEP 9904 (1999) 013, hep-th/9812239.
[3] M. Cederwall, A. von Gussich, B.E.W. Nilsson, P. Sundell, A. Westerberg, The Dirichlet super-p-branes in
ten-dimensional type IIA and IIB supergravity, Nucl. Phys. B 490 (1997) 179, hep-th/9611159.
[4] M. Aganagic, C. Popescu, J.H. Schwarz, Gauge-invariant and gauge-fixed D-brane actions, Nucl. Phys.
B 495 (1997) 99, hep-th/9612080.
[5] E.A. Bergshoeff, M. de Roo, A. Sevrin, Non-Abelian Born–Infeld and kappa-symmetry, hep-th/0011018.
[6] E.A. Bergshoeff, A. Bilal, M. de Roo, A. Sevrin, Supersymmetric non-Abelian Born–Infeld revisited,
JHEP 0107 (2001) 029, hep-th/0105274.
[7] M. Cederwall, B.E.W. Nilsson, D. Tsimpis, D = 10 super-Yang–Mills at O(α 2 ), JHEP 0107 (2001) 042,
hep-th/0104236.
[8] A. Collinucci, M. de Roo, M.G.C. Eenink, Supersymmetric Yang–Mills theory at order α 3 , JHEP 0206
(2002) 024, hep-th/0205150.
[9] M. de Roo, M.G.C. Eenink, P. Koerber, A. Sevrin, Testing the fermionic terms in the non-Abelian D-brane
effective action through order α 3 , hep-th/0207015.
[10] T. Banks, W. Fischler, S.H. Shenker, L. Susskind, M-theory as a matrix model: a conjecture, Phys. Rev. D 55
(1997) 5112, hep-th/9610043.
[11] D. Kabat, W. Taylor, Linearized supergravity from matrix theory, Phys. Lett. B 426 (1998) 297, hep-
th/9712185.
[12] W. Taylor, M. Van Raamsdonk, Multiple D0-branes in weakly curved backgrounds, Nucl. Phys. B 558
(1999) 63, hep-th/9904095.
[13] W. Taylor, M. Van Raamsdonk, Multiple Dp-branes in weak background fields, Nucl. Phys. B 573 (2000)
703, hep-th/9910052.
[14] I. Klebanov, W. Taylor, M. Van Raamsdonk, Absorption of dilaton partial waves by D3-branes, Nucl. Phys.
B 560 (1999) 207, hep-th/9905174.
[15] M. Berkooz, M.R. Douglas, Five-branes in M(atrix) theory, Phys. Lett. B 395 (1997) 196, hep-th/9610236.
[16] M. Asano, Y. Sekino, Non-Abelian action of D0-branes from matrix theory in the longitudinal 5-brane
background, Nucl. Phys. B 639 (2002) 370, hep-th/0201248.
[17] B. de Wit, D.Z. Freedman, Combined supersymmetric and gauge-invariant field theories, Phys. Rev. D 12
(1975) 2286.
[18] J.A. Harvey, Spin dependence of D0-brane interactions, Nucl. Phys. Proc. Suppl. 68 (1998) 113, hep-
th/9706039.
[19] M.R. Douglas, D. Kabat, P. Pouliot, S.H. Shenker, D-branes and short distances in string theory, Nucl. Phys.
B 485 (1997) 85, hep-th/9608024.
[20] Y. Kazama, T. Muramatsu, On the supersymmetry and gauge structure of matrix theory, Nucl. Phys. B 584
(2000) 171, hep-th/0003161.
[21] Y. Kazama, T. Muramatsu, Fully off-shell effective action and its supersymmetry in matrix theory, Class.
Quantum Grav. 18 (2001) 2277, hep-th/0103116.
D-branes on noncompact Calabi–Yau manifolds:

K-theory and monodromy
Xenia de la Ossa, Bogdan Florea, Harald Skarke
Mathematical Institute, University of Oxford, 24-29 St. Giles’, Oxford OX1 3LB, England, UK
Received 19 June 2002; accepted 23 August 2002
Abstract
We study D-branes on smooth noncompact toric Calabi–Yau manifolds that are resolutions
of Abelian orbifold singularities. Such a space has a distinguished basis {Si } for the compactly
supported K-theory. Using local mirror symmetry we demonstrate that the Si have simple
transformation properties under monodromy; in particular, they are the objects that generate
monodromy around the principal component of the discriminant locus. One of our examples, the
toric resolution of C3 /(Z2 × Z2 ), is a three parameter model for which we are able to give an
explicit solution of the GKZ system.
PACS: 11.25.Mj; 11.25.Sq
1. Introduction
The simplest manifestation of mirror symmetry is an exchange of the Hodge numbers

hii and hi,n−i of a Calabi–Yau n-fold X and its mirror X. Interpreted naively, this would

seem to imply identifications between n-cycles on X and holomorphic cycles on X. This
leads to the following puzzle. Monodromy in the complex structure moduli space of X can
take n-cycles to arbitrary other n-cycles, so this would lead to the counterintuitive picture
of mixing cycles of arbitrary even dimension on X.
Mathematically this puzzle is resolved by Kontsevich’s conjecture [1] that the relevant
objects on X are the elements of the bounded derived category D b of coherent sheaves.
In terms of physics, we now have the following intuitive picture. We should not think
of a cycle as a geometric object per se, but as something that a D-brane can wrap.
E-mail address: florea@maths.ox.ac.uk (B. Florea).
PII: S 0 5 5 0 - 3 2 1 3 ( 0 2 ) 0 0 7 6 2 - 9
X. de la Ossa et al. / Nuclear Physics B 644 (2002) 170–200 171
A D-brane corresponds to a cycle with a vector bundle on it only in a semiclassical limit.

In a more general construction a D-brane can be obtained from higher-dimensional branes
and anti-branes, leading to an interpretation in terms of K-theory [2] that is consistent with
Kontsevich’s approach.
Monodromy in the complexified Kähler moduli space of Calabi–Yau manifolds has
been the object of recent studies both by mathematicians [3–8] and by physicists [9–20].
One particular approach [3,15–17] uses well known results on McKay correspondence
[21–28] to obtain a special basis for the K-theory on X. These authors study noncompact
toric Calabi–Yau manifolds that are resolutions of singularities of the type Cd /Zn (or more
general Calabi–Yau singularities in [19]) with a single exceptional divisor, mainly in order
to describe compact Calabi–Yau manifolds as hypersurfaces in the exceptional divisor.
In this work we study D-branes on noncompact toric Calabi–Yau manifolds in their
own right, with the aim of getting a better understanding of what the fundamental D-brane
degrees of freedom are and how they behave under monodromy. We show how to construct
a distinguished basis for the compactly supported K-theory with a number of remarkable
properties, the most striking being the fact that the elements of this basis seem to generate
the monodromy around the principal component of the discriminant locus in the same way
as the structure sheaf OX does in the compact case. We consider cases with more than
one exceptional divisor, and we test the applicability of the above statements beyond the
realm of McKay correspondence. We do not have general proofs for our statements, but
we demonstrate their validity in various examples with the help of local mirror symmetry.
The outline of this paper is as follows. In Section 2 we present necessary material
on toric varieties, their Mori and Kähler cones, and the secondary fan. In Section 3
we introduce local mirror symmetry, toric moduli spaces and the GKZ system. While
the material in these sections is known, its presentation relying on the holomorphic
quotient approach to toric varieties may be useful; besides, it serves to establish notations
and to introduce some of our examples. Section 4 is the core of this paper. There we
discuss K-theory and known results related to McKay correspondence and proceed to
define the distinguished generators Si of the compactly supported K-theory. We find that
these generators are the ones that are responsible for monodromy around the principal
component of the discriminant locus. In Section 5 we demonstrate that our methods work
in cases that are more complicated than examples of the type Cd /Zn . We consider the case
of C3 /(Z2 × Z2 ) and find that it is possible to solve the corresponding GKZ system, with
results that agree precisely with our assertions.
2. Toric Calabi–Yau manifolds
We start with presenting some general considerations on non-compact toric Calabi–

Yau manifolds and their Kähler and Mori cones that will be useful later. The results
obtained here are standard [29–32], but our derivations from basic facts in toric geometry
are possibly simpler than what can be found in the literature.
The data of a d-dimensional toric variety X can be specified in terms of a fan Σ in a
lattice N isomorphic to Zd . X is smooth whenever each of the d-dimensional cones in Σ
172 X. de la Ossa et al. / Nuclear Physics B 644 (2002) 170–200
is generated over R+ by exactly d lattice vectors that generate N over Z. We will only
consider this case.
Perhaps the simplest way of describing X is as follows: assume that there are k one-
dimensional cones in Σ generated by lattice vectors v1 , . . . , vk . Assign a homogeneous
variable zi to each of the vi and a multiplicative equivalence relation among the zi ,

(z1 , . . . , zk ) ∼ λq1 z1 , . . . , λqk zk (2.1)
with λ ∈ C∗ for any linear relation q1 v1 + · · · + qk vk = 0 among the generators vi . The
qi can be normalized to be integers without common divisor; in the context of a gauged
linear sigma model they are the charges with respect to the U (1) fields. The number of
independent relations of the type (2.1) is k − d.
Define a subset Ck \ FΣ of Ck = {(z1 , . . . , zk )} as the set of all k-tuples of zi with
the following property: if zi vanishes for all i ∈ I ⊂ {1, . . . , k}, then all vi with i ∈ I
belong to the same cone. Then X is (Ck \ FΣ )/(C∗ )k−d , where the division by (C∗ )k−d
is implemented by taking equivalence classes with respect to the multiplicative relations
(2.1).
Every one-dimensional cone generated by vi corresponds in a natural way to the
divisor Di determined by zi = 0. Similarly, an l-dimensional cone spanned by vi1 , . . . , vil
determines the codimension l subspace zi1 = · · · = zil = 0 of X.
a
Monomials of the type z1a1 · · · zk k are sections of line bundles O(a1 D1 + · · · + ak Dk ).
If we denote by M the lattice dual to N and by , the pairing between N and M, it is
v ,m v ,m
easily checked that monomials of the form z1 1 · · · zk k with m ∈ M are meromorphic
functions (i.e., invariant under (2.1)) on X. This implies the linear equivalence relations
v1 , mD1 + · · · + vk , mDk ∼ 0 for any m ∈ M. (2.2)

Conversely, if a divisor of the form a1 D1 + · · · + ak Dk belongs to the trivial class, then
there exists an m ∈ M such that ai = vi , m for all i.
A calculation similar to the way the canonical divisor of Pd is determined shows that
the canonical divisor of X is given by −D1 − · · · − Dk . Thus X is Calabi–Yau if and only
if D1 + · · · + Dk is trivial, i.e., if and only if there exists an m ∈ M such that vi , m = 1
for every i. Therefore, the vi must all lie in the same affine hyperplane. We will make use
of this fact by drawing toric diagrams in dimension d − 1 that display only the endpoints
of the vi .
We will be interested in the Kähler moduli space of X. The dual of the Kähler cone is the
Mori cone spanned by effective curves. Toric curves are determined by (d −1)-dimensional
cones σd−1 in Σ. If a curve is compact, the corresponding cone is the boundary between
(1) (2) (1) (2)
two d-dimensional cones σd , σd . If we denote the integer generators of σd−1 , σd , σd
by {v1 , . . . , vd−1 }, {v1 , . . . , vd−1 , vd }, {v1 , . . . , vd−1 , vd+1 }, respectively (remember that
we are assuming that our cones are simplicial and their generators generate N ), we find
that vd + vd+1 must lie in the intersection of the hyperplane of σd−1 with N and so there
exists a unique linear relation of the form l1 v1 + · · · + ld+1 vd+1 = 0 with ld = ld+1 = 1
and all li integer.
We will now argue that the li are actually the intersection numbers between the curve
C = D1 × · · · × Dd−1 determined by σd−1 and the toric divisors Di . Our general rules
imply that intersection numbers between d different toric divisors are 1 or 0 depending on
whether these divisors form a cone in Σ. This implies C ·Dd = ld = 1, C ·Dd+1 = ld+1 = 1
and C · Di = 0 for i > d + 1. For calculating C · Di with i < d we have to use linear
equivalence relations of the type (2.2). To calculate C · D1 we may choose m to fulfill
v1 , m = 1 and v2 , m = · · · = vd , m = 0. Then

0∼ vi , mDi = v1 , mD1 + vd+1 , mDd+1 + · · ·
= D1 + −l1 v1 − · · · − ld vd , mDd+1 + · · ·
= D1 − l1 Dd+1 + · · · , (2.3)
i.e., D1 ∼ l1 Dd+1 + · · · where ‘· · ·’ stands for Di with i > d + 1 which do not intersect C.
Thus we find that C · D1 = l1 C · Dd+1 = l1 . As our choice of D1 among the Di with i < d
was arbitrary, we have indeed shown that C · Di = li for any i.
A set of generators for the Mori cone is then given by all those curves C (i) whose l (i)
cannot be written as nonnegative linear combinations of the other l (j ) . The matrix L whose
lines are the l (i) of the Mori cone generators has the following remarkable properties:
any Matrix Q consisting of d − k independent (linear combinations of ) lines of L serves
as a ‘charge matrix’ for the relations (2.1). If the Mori cone is simplicial, we just have
L = Q. This will be the case in most of our examples, so we will not distinguish between
L and Q in these
cases. Any column of L is associated with a toric divisor Di . If a linear
combination
j Lij aj of column vectors of Q vanishes, then the corresponding divisor
j aj Dj has vanishing intersection with any effective curve, i.e., it is trivial. Therefore a
diagram displaying the column vectors of L or Q encodes the linear equivalence relations
among the toric divisors Di . We may interpret these vectors as one-dimensional cones of
a fan, the so-called ‘secondary fan’ of X. Note, however, that two distinct but linearly
equivalent toric divisors correspond to the same vector in the secondary fan. As the entries
of L are the intersections between the generators of the Mori cone and the divisors, the
Kähler cone of X is determined by those j aj Dj such that the corresponding linear
combinations of the columns of L only have nonnegative entries.
We should stress that our analysis was in terms of a single fixed triangulation. If we
allow several distinct triangulations, the Mori cone vectors of any of them will lead to
correct charge matrices Q but the Kähler condition will depend on which combinations
of the charge vectors correspond to the Mori cone, i.e., on the choice of triangulation. In
this way several regions of a secondary fan constructed from some charge matrix Q can
correspond to different ‘geometric phases’ in the sense of [33,34].
We will now present some of the examples that we are going to use in this paper.
Example 1. The toric resolution of C2 /Zn .
We have toric divisors D0 , . . . , Dn corresponding to vectors

0 1 n
v0 = , v1 = , . . . , vn = . (2.4)
1 1 1
D0 and Dn are noncompact and correspond to the coordinates of the original C2 on

which Zn acts by (z0 , zn ) → (#z0 , # n−1 zn ) with # = e2πi/n . All other Di are compact
and are nothing but the effective curves. The Mori cone vectors are determined by
vi−1 − 2vi + vi+1 = 0, leading to

 
1 −2 1 0 . . . 0 0
 0 1 −2 1 . . . 0 0 
Q= . (2.5)
...
0 0 0 0 . . . −2 1
Upon dropping the first and the last column, this becomes −MSU(n) , where MSU(n) is
the Cartan matrix of SU(n). Thus the generators of the Kähler cone, corresponding to
linear
combinations of the Di that turn the columns of L into unit vectors, are given by
−1
− n−1j =1 (MSU(n) )ij Dj or, alternatively, by
D0 , D1 + 2D0 , D2 + 2D1 + 3D0 , . . . , Dn−2 + 2Dn−3 + · · · + (n − 1)D0 . (2.6)

Example 2. The toric resolution of Cn /Zn .
The resolution of a singular space of the type Cn /Zn , where Zn acts on the coordinates
of Cn by
(z1 , . . . , zn ) → (#z1 , . . . , #zn ) with # = e2πi/n (2.7)
can be represented torically by vectors v1 , . . . , vn+1 , subject to the single relation v1 + v2 +
· · · + vn = nvn+1 ; the N lattice is just the lattice generated by the vi . The first n vectors
v1 , . . . , vn correspond to the original coordinates zi whereas vn+1 corresponds to the single
exceptional divisor Dn+1 = {zn+1 = 0} isomorphic to Pn−1 . The Mori cone is determined
by the single relation, leading to
Q = (1, 1, . . . , 1, −n). (2.8)
Example 3. The toric resolution of C3 /Z5 .
We first consider a singular space of the type C3 /Z5 , where Z5 acts on the coordinates
of C3 by

(z1 , z2 , z3 ) → #z1 , # 3 z2 , #z3 with # = e2πi/5. (2.9)
As a toric variety C3 /Z5 is determined by three vectors

−1 2 0
v1 = 0 , v2 = 2 , v3 = −1 (2.10)
1 1 1
in a lattice N isomorphic to Z3 , the singularity resulting from the fact that v1 , v2 and v3
generate only a sublattice of N . A complete crepant (i.e., canonical class preserving) toric
resolution X → C3 /Z5 is obtained by adding two further rays

0 1
v4 = 0 , v5 = 1 (2.11)
1 1
and triangulating the resulting diagram (this triangulation is unique in the present case).
The resulting fan, with the redundant third coordinate suppressed, is shown in Fig. 1.
The structure of the resolution is easily read off from this diagram: we have two exceptional
divisors D4 and D5 corresponding to v4 and v5 , respectively. The star fans of v4 and v5 tell
Fig. 1. The resolution of C3 /Z5 .
us that D4 is a P2 and D5 is a Hirzebruch surface F3 . D4 and D5 intersect along a curve

h which is a hyperplane of the P2 and at the same time the negative section of F3 . We
denote by D1 , D2 and D3 the noncompact toric divisors corresponding to the vertices v1 ,
v2 and v3 , respectively (i.e., the zero loci of the coordinates of our original C3 ). Intersection
numbers can be calculated by using the linear equivalences
D1 ∼ D3 ∼ D5 + 2D2 and D1 + D2 + D3 + D4 + D5 ∼ 0 (2.12)

and the fact that three distinct toric divisors have an intersection number of 1 if they belong
to the same cone and 0 otherwise. As D1 is linearly equivalent to D3 we omit expressions
involving D3 in the following. Intersections of divisors are well defined whenever they
involve at least one of D4 and D5 . Triple intersections are given by
D43 = 9, D42 · D5 = −3, D4 · D52 = 1, D53 = 8,
D42 · D1 = −3, D4 · D5 · D1 = 1, D52 · D1 = −2, D52 · D2 = −5,
D4 · D12 = 1, D5 · D12 = 0, D5 · D1 · D2 = 1, D5 · D22 = 3 (2.13)
and the vanishing of D4 · D2 = 0. Intersections of two distinct divisors are determined by
D4 · D5 = D4 · D1 = h, D4 · D2 = 0,
D5 · D1 = f, D5 · D2 = h + 3f, (2.14)
where f is the fibre of the F3 . The self-intersections of D4 and D5 are
D42 = −3h, D52 = −2h − 5f. (2.15)

We also have:
D4 · h = −3, D4 · f = 1, D5 · h = 1, D5 · f = −2,
D1 · h = 1, D1 · f = 0, D2 · h = 0, D2 · f = 1. (2.16)
This implies that (C1 , C2 ) = (h, f ) and (D1 , D2 ) form mutually dual bases of the Mori
cone and the Kähler cone of X. In terms of codimension one (here, two-dimensional)
Fig. 2. The secondary fan of the resolution of C3 /Z5 .
cones σ and the linear relations between the rays in the two cones of maximal dimension
that contain σ , we obtain the following linear relations among the vectors v1 , . . . , v5 of the
fan:
l (1) = (1, 0, 1, −3, 1), l (2) = (0, 1, 0, 1, −2). (2.17)

(j )
As a check on our intersection numbers, we observe that indeed Di · Cj = li .
If we consider the matrix whose lines are the generators (2.17) of the Mori cone and
draw the rays corresponding to the columns of this matrix, we obtain the secondary fan for
X as shown in Fig. 2. The linear relations among the vectors in this fan encode the linear
equivalences (2.12) among the divisors.
3. Local mirror symmetry
In our study of D-brane states we will have to address issues that involve quantum
geometry. A standard tool for this problem is the use of mirror symmetry. In particular,
classical periods in the mirror geometry get mapped to quantum corrected expressions
related to the middle cohomology of the original space. In the noncompact case one has
to use local mirror symmetry. For our applications of this subject we have relied mainly
on [35] and we refer to this paper for further references. The authors of [35] consider
decompactifications of Calabi–Yau hypersurfaces in toric varieties such that the volumes
of certain cycles remain compact. They show that in the decompactification limit these
cycles lead to differential equations that are identical with the GKZ differential systems
of a lower-dimensional geometry. We will assume that this remains true even for cases
where the non-compact Calabi–Yau geometry cannot be identified with a limiting case of
a compact Calabi–Yau hypersurface.
The local mirror of a d-dimensional noncompact Calabi–Yau geometry is determined
by interpreting the diagram of the hyperplane containing the end points of the vi now as
a polytope P in a (d − 1)-dimensional lattice M. A polytope corresponds to a line bundle
to be the origin.
L over a toric variety V by the following construction: fix any point in M
, the lattice
Describe the facets of P by equations Ej (m̃) := ṽj , m̃ + cj = 0, where ṽj ∈ N
dual to M and fix the sign ambiguity about ṽj in such a way that Ej (m̃) is nonnegative for
points m̃ of P . Choose V to be a toric variety whose one-dimensional rays are the ṽj ∈ N
corresponding to a variable xj as in the previous section. To every point m̃ ∈ M assign the
E (m̃)
monomial j xj j . Then L is the bundle whose sections are determined by polynomials
of the type

k E (m̃i )
P (a; x) = ai xj j . (3.1)
i=1 j
The ‘local mirror’ X of X is defined to be the vanishing locus of a section (3.1) of L.

In the present context we can give an alternative description of the Ej : we have
N M ⊕ Z and may choose coordinates such that vi = (m̃i , 1). Then we can write the
affine function Ej (m̃i ) as a linear function of the form vi , ṽj with ṽj ∈ Hom(N, Z) = M
(it is easy to check that the ṽj are the elements of M dual to the (d − 1)-dimensional cones
at the boundary of the support of Σ).
Obviously the complex structure moduli space of X is parametrized by the ai . It is
important to note, however, that different sets of ai need not correspond to different
complex structures. In particular, a scaling xj → λj xj does not amount to a change in
the complex structure but leads to a redefinition of the ai , implying the equivalences
E (m̃ ) E (m̃ ) E (m̃ )

(a1 , a2 , . . . , ak ) ∼ λj j 1 a1 , λj j 2 a2 , . . . , λj j k ak
v1 ,ṽ v2 ,ṽ vk ,ṽ
= λj j a1 , λj j a2 , . . . , λj j ak (3.2)
for any j . Given identifications of this type it is natural to seek a description in terms of
toric geometry. If we interpret the exponents of the λ’s as linear relations among vectors ui
in a toric diagram and notice that the ṽj generate M (at least over the rational numbers),
we find that the ui fulfill
m, v1 u1 + m, v2 u2 + · · · + m, vk uk = 0 (3.3)
for any m ∈ M. These are just the relations among the vectors of the secondary fan which
encodes, as we saw, the linear equivalence relations (2.2) of the divisors Di corresponding
to the vi . There are some subtleties, however: as we saw in the previous section, it is
possible that two distinct (but linearly equivalent) toric divisors lead to the same vector
in the secondary fan. We will show how to interpret this in the context of the examples.
Besides, it is possible that there are identifications in the moduli space that do not come
from rescalings of the type xj → λj xj and hence have a structure different from (3.2).
If this occurs, the toric variety associated with the secondary fan is called the ‘simplified
moduli space’ Msimp . Depending on whether we have extra identifications or not, the toric
variety corresponding to the secondary fan is a compactification of Msmooth (the moduli
space of all smooth local mirror hypersurfaces) or a covering space of a compactification
of Msmooth.
will degenerate over various loci in Msimp where ∂P (a; x)/∂xj = 0 can be solved
X
for all j without violating the conditions on which xi are allowed to vanish simultaneously.
Some of these loci may just be toric divisors, but usually there is also at least one connected
piece given by a polynomial equation in the ai to which we will refer as the primary or
principal component of the discriminant locus.
If we want to relate the mirror geometry to the original one, we have to find a region in
the moduli space where quantum corrections are strongly suppressed. This is the case for
the deep interior of the Kähler cone, the so-called large volume limit, which is dual to the
large complex structure limit. As we saw in Section 2, the Kähler cone can be determined
by writing any divisor as a linear combination of toric divisors and demanding that the
corresponding linear combination of columns of the matrix L contain only nonnegative
entries. If the resulting generators do not belong to the secondary fan, we have to blow up
the moduli space in order to be able to change to the large complex structure variables.
In those cases where the Mori cone is simplicial we can draw the secondary fan by
displaying the columns of L and the generators of the Kähler cone will be nothing but
the unit vectors. If we then write the linear relations among the vectors in the secondary
fan in such a way that we express every vector in terms of the unit vectors and use the
corresponding rules (2.1) to set all variables except the large complex structure variables1
zi to 1, we find that the zi can be expressed as

k
l
(i)
zi = ajj . (3.4)
j =1
Note that we do not include a sign here (compare with, e.g., [36]).
If X is the resolution of an orbifold singularity of the type Cd /Zn there is another
distinguished coordinate patch in the moduli space containing the orbifold locus where all
ai except the ones corresponding to the coordinates of the Cd are set to zero. At this point
the conformal field theory is expected to acquire a quantum symmetry. We find that the
moduli space in this case always has a singularity that looks locally like CdimM /Zn .
The GKZ differential
operators are calculated by using the following recipe: for every
linear relation lj vj = 0, where l corresponds to any curve in the Mori cone (see [35])
we define a differential operator in terms of the ai ,
lj −lj
D= ∂aj − ∂aj . (3.5)
j : lj >0 j : lj <0
µ
Assume that we work in a specific coordinate patch given by some φi = aj ij . In order
to transform (3.5) to a system involving the φi we can rewrite it in terms of operators
Θaj := aj ∂aj , commute all aj to the left using Θaj aj−1 = aj−1 (Θaj − 1) and then express

the Θai as i µij Θφi with Θφi := φi ∂φi .
We stress that the solutions of the GKZ system are not the periods on X but rather the
logarithmic integrals of the periods. While the periods are finite and nonvanishing on the
moduli space wherever X is nondegenerate, the GKZ solutions have extra singularities
at the zero loci of moduli space coordinates coming from the logarithmic integration.
1 We hope that no confusion arises from the fact that we use the same symbol z for the coordinates of X and
i
the large complex structure variables.
The GKZ solutions are multivalued and undergo monodromy transformations around
codimension one loci where they are not holomorphic. We will be interested mainly in
monodromies around the large complex structure divisors zi = 0 and around the principal
component of the discriminant locus. In addition, there is the possibility of a nontrivial
transformation (‘orbifold monodromy’, which, strictly speaking, is not a monodromy) if
the moduli space looks locally like CdimM /Zn .
We will now show how these concepts can be applied to our examples.
Example 1. The mirror geometry of C2 /Zn .
Here V is P1 and the polynomial is given by
a0 x1n + a1 x1n−1 x2 + · · · + an x2n , (3.6)

so the hypersurface X occurs
is just a collection of n points in P1 . A ‘singularity’ of X
whenever two or more of these points coincide. The secondary fan is determined by the
columns of (2.5). For n 3 we have to blow up the moduli space in order to have a
coordinate patch described by the large complex structure variables zi = ai−1 ai+1 /(ai )2
(with 1 i n − 1). The GKZ operators corresponding to the Mori cone generators,
∂a0 ∂a2 − ∂a21 , ∂a1 ∂a3 − ∂a22 , . . . , ∂an−2 ∂an − ∂a2n−1 (3.7)
become
Θa0 Θa2 − z1 (Θa1 − 1)Θa1 , . . . , Θan−2 Θan − zn−1 (Θan−1 − 1)Θan−1 (3.8)
with
Θa0 = Θz1 , Θa1 = −2Θz1 + Θz2 ,

Θai = Θzi−1 − 2Θzi + Θzi+1 for 2 i n − 3,
Θan−1 = Θzn−2 − 2Θzn−1 , Θan = Θzn−1 . (3.9)
We note that the space of solutions of (3.8) is too large unless we introduce further
operators corresponding to linear combinations of the Mori cone generators.
The case of n = 2 allows for an explicit solution [36]: here we have

D = Θz − 2z(2Θz + 1) Θz (3.10)
and D = 0 has a basis of solutions of the form
√
1 1 − 1 − 4z
00 = 1, 01 = ln √ . (3.11)
2πi 1 + 1 − 4z
Special points in the moduli space are the large complex structure limit z = 0, the analog
of the primary component of the discriminant locus at z = 1/4, and the orbifold point
at z = ∞ where we introduce a new coordinate ϕ by zϕ 2 = 1. We find the following
transformation properties upon taking loops around these points:
z=0: 01 → 01 + 1, z = 1/4 : 01 → −01 ,

ϕ→e ϕ: πi
01 → −1 − 01 . (3.12)
Example 2.
of the resolution X of Cn /Zn is just the mirror geometry
The local mirror geometry X
is a
of a compact Calabi–Yau manifold realised as a degree n hypersurface in Pn−1 , i.e., X
degree n hypersurface
a1 x1n + · · · + an xnn + an+1 x1 · · · xn = 0 (3.13)
in Pn−1 /(Zn )n−2 . The GKZ operator ∂a1 · · · ∂an − ∂ann+1 becomes
Θzn − z(−nΘz − n + 1)(−nΘz − n + 2) · · · (−nΘz ) (3.14)
in terms of the large complex structure variable z = a1 · · · an /(an+1 )n .

Example 3.
The mirror geometry of C3 /Z5 , a genus two Riemann surface (see Fig. 3).
Here V is P2 /Z5 with the Z5 acting on the homogeneous coordinates of P2 as
(x1 , x2 , x3 ) → (#x1 , x2 , # −1 x3 ). The polynomial corresponding to Fig. 1 is given by
a1 x15 + a2 x25 + a3 x35 + a4 x12 x2 x32 + a5 x1 x23 x3 , (3.15)
where we have chosen the subscripts of the ai to correspond to those of the vi in Fig. 1.
The local mirror of C3 /Z5 is given by the vanishing locus of (3.15) in P2 /Z5 . The action
of Z5 on P2 has fixed points whenever two of the three xi vanish. The vanishing locus of
(3.15) passes through one of these fixed points if and only if one of a1 , a2 , a3 vanishes.
Thus the generic hypersurface misses the fixed points. A quintic polynomial in P2 defines,
by a standard calculation, a Riemann surface of Euler number χ = −10. As the Z5 acts
without fixed points on this surface, the Euler number is divided by 5, showing that the
local mirror geometry is that of a Riemann surface R with 2 − 2g = χ = −2, i.e., genus
g = 2.
Scalings xi → λi xi imply the equivalences

(a1 , a2 , a3 , a4 , a5 ) ∼ λ51 a1 , λ52 a2 , λ53 a3 , λ21 λ2 λ23 a4 , λ1 λ32 λ3 a5 . (3.16)
Fig. 3. The fan for P2 /Z5 .

If we naively interpret the exponents of the λi as a charge matrix

5 0 0 2 1
0 5 0 1 3 (3.17)
0 0 5 2 1
j j
with entries qi and try to find a fan with rays vi fulfilling i qi vi = 0 for j = 1, 2, 3, we
find that we have to take v1 = v3 , meaning that we should not distinguish between a1 and
a3 . This can be explained by the fact that taking λ3 = λ−11 implies that we can multiply a1
with any nonzero number provided we divide a3 by the same number without affecting the
other ai , i.e., as long as a1 and a3 are nonzero the complex structure of R depends only
on a13 := a1 a3 . This is even true if one of a1 , a3 becomes zero, since an exchange of a1
and a3 can be compensated by exchanging x1 with x3 which does not affect the complex
structure. Thus we can consistently drop the third line and the third column of (3.17) to
obtain a matrix

5 0 2 1
. (3.18)
0 5 1 3
This is just the matrix of linear relations for the secondary fan of Fig. 2. The corresponding
compact toric variety

Mtoric = (a13, a2 , a4 , a5 ) \ (a13 = a4 = 0) ∨ (a2 = a5 = 0) /∼ (3.19)
with ∼ as in (3.16) is closely related to the moduli space Msmooth of smooth hypersurfaces
of the type (3.15): smoothness implies a1 = 0, a2 = 0 and a3 = 0, so

Msmooth ⊂ (a13 , a2 , a4 , a5 ) \ (a13 = 0) ∨ (a2 = 0) /∼ ⊂ Mtoric , (3.20)
i.e., Mtoric is a compactification of Msmooth (other sensible compactifications correspond

to omitting v13 or v2 from Fig. 2). Msmooth is contained in the single coordinate patch of
Mtoric defined by the cone spanned by v4 and v5 . In this patch we can parametrize the
hypersurface as the vanishing locus of
Pψ,φ (x1 , x2 , x3 ) = x15 + x25 + x35 − 5ψx12 x2 x32 − 5φx1 x23 x3 . (3.21)
Having set a1 , a2 and a3 to one has used up most of the freedom coming from (3.16), the
remaining relation being

(ψ, φ) ∼ # 2 ψ, #φ . (3.22)
As we just noticed, R becomes singular along the divisors a13 = 0 and a2 = 0 of
Mtoric . The remaining singularities can be found by looking for values of ψ, φ where
∂Pψ,φ /∂xi = 0 for i = 1, 2, 3 can be solved by some (x1 , x2 , x3 ) = (0, 0, 0). This results
in the equation
16ψ 5 + 40ψ 4 φ 2 + 25ψ 3 φ 4 + 20ψ 2 φ + 45ψφ 3 + 27φ 5 = 1 (3.23)
for the primary component of the discriminant locus.
We note that while Mtoric contains some of the singular loci, it misses others such
as a1 = a4 = 0, a2 = a5 = 0 and any points with three or four of the ai vanishing.
Fig. 4. The moduli space of the resolution of C3 /Z5 .
The divisor a13 = 0 in Mtoric corresponds to two one-dimensional loci a1 = a3 = 0

and a1 a3 = 0, a1 + a3 = 0. Our main concern with the moduli space has to do with
the study of monodromies. Thus we want to know what happens when we move around
singularities at codimension one rather than what happens when we hit them. For example,
the monodromy around a13 = 0 depends only on nonvanishing values of a13 and not on
how we interpret the locus a13 = 0. Therefore Mtoric is sufficient for our purposes.
A schematic representation of Mtoric is given in Fig. 4, with the toric divisors shown
as straight lines and the primary component of the discriminant locus indicated by curved
lines. The locus a13 = 0 is tangent to the discriminant locus at their point of intersection
4a2 a4 = a52 whereas 27a13a5 + a43 = 0 corresponds to a transverse intersection with a2 = 0.
At a4 = a5 = 0 (i.e., φ = ψ = 0) the moduli space has a singularity where the Riemann
surface remains smooth; in addition there are Z2 and Z3 singularities at a13 = a5 = 0 and
a2 = a4 = 0, respectively.
As we saw above, there is a distinguished coordinate patch in Mtoric which contains all
loci where R is smooth. Now we want to study another distinguished set of coordinates
corresponding to the ‘large complex structure limit’. We remember that the Kähler cone
of X (the resolution of C3 /Z5 ) was spanned by D1 ∼ D3 and D2 corresponding to the
vectors v13 and v2 in the secondary fan (Fig. 2), respectively. The large radius limit of X
corresponds to the deep interior of the Kähler cone, so by local mirror symmetry the large
complex structure limit is determined by the v13 − v2 −coordinate patch in Mtoric given by
a5 φ a4 ψ
z1 = a13 = , z2 = a2 =− 2. (3.24)
a43 52 ψ 3 2
a5 5φ
In terms of z1 , z2 the principal component of the discriminant locus is determined by
1 + 27z1 − 8z2 − 225z1z2 + 16z22 + 500z1z2 2 + 3125z12z23 = 0. (3.25)

The GKZ system can be determined and solved with the methods described above. There
are five independent solutions, as expected, which are described in Appendix A.
4. D-branes and tautological bundles
We want to find out about the D-brane vacuum states in type II string theory on X. The
mathematical structure that captures the largest number of properties of brane states is,
at present knowledge, the bounded derived category D b of coherent sheaves on X [37,
38] (but see the remarks in [39,40]). While we will make several remarks concerning
D b , we will work mainly with the somewhat coarser (but easier to handle) concepts of
K-theory. Let K(X) be the Grothendieck group of coherent sheaves on X. We expect
compact brane states on a noncompact space X to correspond to classes of the compactly
supported K-theory group K c (X). Using the duality between K(X) and K c (X) we can
determine a basis for K c (X) by first finding a basis for K(X).
Let us consider the situation where X is a smooth crepant resolution of a singularity
of the type Cd /G, where G is a finite subgroup of SU(d). Since X is smooth, K(X) is
generated by vector bundles (see, e.g., [41]). Moreover, if π : X → Cd /G is a crepant
resolution of an Abelian singularity, K(X) is in fact generated by n line bundles, where n
is the order of G (at least for d 3) [25]. Thus, for finding a basis for the group K c (X)
related to fractional branes it is convenient to first determine a set {Ri } (0 i < n) of line
bundles whose K-theory classes generate K(X). Clearly there is no choice for the Ri that
should be preferred a priori. Rather, there are two distinct constructions, each of which is
related to McKay correspondence:
(1) Mathematicians’ construction [22–25]: there is a vector bundle R (the ‘tautological
vector bundle’) transforming in the regular representation of G whose decomposition into
irreducibles gives the line bundles RiM . In particular, the RiM are generated by their sections
and the action of G on the sections determines a one-to-one correspondence between the
RiM and the characters of the irreducible representations of G. In the case of a resolution
of C2 /G with some finite group G the first Chern classes c1 (RiM ), i 1, form a basis
of H 2 (X, Z) dual to the basis of H2 (X, Z) given by the homology classes of a basis of
effective curves Ci in the resolution. In the case of a singularity of the type C3 /G with G
an Abelian subgroup of SL(3, C) in general there exist several crepant resolutions and not
for every resolution it is possible to define line bundles as above. However, it was shown
in [24] that there exists a distinguished crepant resolution, named G-Hilb, on which it is
still possible to define the tautological line bundles (see also [23,27,28]).2 The advantage
of this approach is that it is rigorously proven for d = 2 and d = 3.
(2) Physicists’ constructions: the authors of [13] suggest to consider, in the style of
[42], the world-volume theory of D0-branes, which is a theory of (n − 1) U (1) gauge
fields and d (n × n) matrices. It is conjectured (and shown in several examples) that the
vacua of such a theory in the different phases corresponding to different choices of Fayet–
Iliopoulos parameters all lead to moduli spaces that are nothing but the geometric phases of
the resolutions X of Cd /G. Now repeat this construction with an extra field transforming in
a specific one-dimensional representation ρi of G. It is conjectured that, independently of
the phase, this should lead to a space that is the total space of a line bundle RiP over X, and
that repeating this for all characters ρi should give a basis {RiP } of K(X). However, this
2 We thank A. Craw for emphasizing the importance of choosing the G-Hilb resolution to us.
construction is extremely tedious to work with. A different method for determining {RiP }
based on the boundary chiral ring associated to a certain two-dimensional gauge theory
has been proposed in [17]. The implications of this approach have been worked out for the
case of a single exceptional divisor that is a weighted projective space W = Pd−1 n1 ,...,nd with
Fermat weights [17] or a Grassmannian3 [19]. In all examples we are aware of, the RiP
have no sections. The advantage of this approach is that it appears to lead to dual classes
SiP whose interpretations in terms of D-branes are very well behaved.
Roughly, the resulting Ri can be summarized in the following way. There is a set of
divisor classes {[Fi ]} containing all Kähler cone generators [Ti ] and the trivial class [0]
such that all Fi are nef, i.e., have nonnegative intersection with any curve in the Mori
cone. If we denote by Ri± the line bundles O(±Fi ), then {RiM } = {Ri+ } and {RiP } = {Ri− }.
In two dimensions the [Fi ] are just the trivial class and the Kähler cone generators. In
higher dimensions we have to add extra divisor classes which are nonnegative integer
linear combinations of the [Ti ]. For the RiM with G-Hilb and d = 3 the authors of [23,
27] have given an explicit construction. In terms of the language used in this paper this can
be summarised in the following way.
Through the sections we can assign a character to any Ti . It is also possible to assign
characters to toric curves. Such a curve C corresponds up to a sign to some m ∈ M leading
to a linear equivalence as in (2.2). By collecting expressions with the same sign this can be
written as D ∼ D where D, D are effective divisors corresponding to the same character.
We then assign this character to C and the corresponding line segment in the diagram,
and find that all the characters obtainable in this way also occur in the list of characters
corresponding to the Ti . Then every interior point I of the toric diagram is of one of the
following types:
(1) There are three pairs of line segments with the same character meeting in I . In this
case we add nothing to the list of [Fi ] (the classes assigned by [23,27] in this case are
already among the Kähler cone generators).
(2) There are two pairs of line segments with characters χm , χn meeting in I (and
possibly an extra line segment). Then add [Tm + Tn ] to the list of [Fi ].
(3) There are three line segments with the same character χm . In this case add [2Tm ] to
the list of [Fi ].
It turns out that this procedure always leads to a one-to-one correspondence between
the Ri+ and the character table of G through the action of G on the sections.
In many cases the [Fi ] are the same in the mathematicians’ and physicists’ construc-
tions, i.e., RiP = (RiM )∗ . However, [17] seems to suggest partial resolutions in the case
with a single interior point where the exceptional divisor is a weighted projective space.
We note that the G-Hilb resolution may be incompatible with such a resolution or any
refinement of it, as the following example shows.
In Fig. 5 we have displayed the G-Hilb resolution of C3 /Z6 constructed according to
the rules of [23,27] and the partial resolution by an exceptional divisor P2(1,2,3) . Clearly the
former cannot be obtained as a refinement of the latter.
In the following we always follow the mathematicians’ approach.
3 P. Mayr informs us that this approach works in more general situations as well.
Fig. 5. G-Hilb and partial resolution of C3 /Z6 .
The next step in our construction of D-brane states is to find a basis for K c (X) that is
dual to the basis of K(X) defined in terms of line bundles Ri . According to [25], there is a
pairing (R, S) between representatives R of K(X) and S of K c (X) that can be evaluated
in terms of Chern characters

(R, S) = ch(R) ∪ chc (S)Td(X), (4.1)
X
c
with ch (S) the localized Chern character4 of the complex S and Td(X) the Todd class
of X. There is also a closely related pairing which will become important when we study
monodromies. It is defined as

R, S = R ∗ , S (4.2)
with R ∗ the line bundle (or, more generally, the complex) dual to R. If we restrict R to
K c (X), these pairings become well defined under the exchange of R and S and we find
that (R, S) is always symmetric whereas R, S is symmetric in even dimensions and skew
in odd dimensions, as a consequence of the fact that Td(X) is even when c1 (X) is trivial.
The generally accepted way of obtaining a basis for K c (X) is to choose classes dual
to those given by the line bundles Ri with respect to ( , ). Following this convention, we
define classes of K c (X) by demanding that their representatives Sj fulfill (Ri , Sj ) = δij .
Thus we obtain Sj+ dual to Ri+ and Sj− dual to Ri− with respect to ( , ) and note that the Sj+
are dual to the Ri− and Sj− are dual to the Ri+ with respect to , .
So far we have not been specific about the representatives Si of the compactly supported
K-theory. In the spirit of [2] we may interpret them as bound states of X-filling branes. In
mathematical terms this amounts to specifying a complex of vector bundles on X that
is exact outside a compact locus Y . It is not hard to check in every example that we
may indeed represent every Si as a formal linear combination of line bundles of the form
OX ( ai Di ) and that the chc (Si ) obtained from the line bundles Ri form a basis for all
Chern characters with support on the compact toric cycles.
Alternatively, one may wish to consider ‘pure’ branes defined in terms of the structure
sheaves of the independent lower-dimensional compact holomorphic cycles. Given the
structure sheaves OCi where the Ci form a basis for all compact holomorphic cycles on
4 Let i : Y >→ X be the embedding of a compact submanifold Y in a noncompact manifold X. Elements of

the compactly supported K-theory can be represented by either coherent sheaves SY on Y or by their finite
resolution by vector bundles on X, that is by complexes S of vector bundles on X which are exact off Y and
whose homology is precisely the push-forward of SY to X [43]. Then, the local Chern character is defined such
that chc (S) = ch(i∗ SY ) [44].
X, applying k push-forwards for every cycle of codimension k leads to sheaves SCi on X.

In order to relate these objects to Chern characters on X we have to use the Grothendieck–
Riemann–Roch theorem,

i∗ ch(SD ) Td(D) = ch(i∗ SD ) Td(X) (4.3)

for embeddings of the type i : D >→ X. Writing chc (S) = n(Ci ) ch( SCi ) allows us to
define the charge vectors n (S). Alternatively we may calculate the charge vectors by first
calculating

Rj , SCi = ch(Rj ) ch
SCi Td(X)
X

= ch(Rj |Ci ) Td(Ci ) = χ(Rj |Ci , Ci ) =: χj i (4.4)
Ci
(C ) (C )
and noticing that (Rj , Sk ) = δj k implies i χj i nk i = δj k , i.e., nk i = (χ −1 )ik . We note
that the compact holomorphic cycles generate the compact homology of X, so the number
of Ci is equal to χ c (X) = 2d−1 i=0 (−1) i H c which is just the number of d-dimensional
i
cones in Σ [29]. √
At the large volume limit the Mukai vector chc (S) Td(X) determines the central
charge5

Z (ti ; S) = − e− ti Ti chc (S) Td(X)
lv
(4.5)
X
of the brane configuration, where the Ti are generators of the Kähler cone. In particular,
we obtain

Z lv t;
Sp = −1 and Z lv t; SCi = ti − 1 (4.6)
for the central charges of D0- and D2-branes wrapping (with trivial bundle) the generators
Ci of the Mori cone dual to the Ti . These are the objects related by local mirror symmetry
to the solutions of the GKZ system at the large complex structure point. More precisely we
expect the exact central charge Z(z; S) to be a linear combination of the GKZ solutions
such that

lim Z(z; S) − Z lv (ti ; S) = 0. (4.7)
z→0
If we demand that Z lv (t; SCi ) measure the complexified Kähler class at the large Kähler
limit we have to make the identification
ln zi
ti − 1 = + O(z). (4.8)
2πi
Note that this is different from the conventions usually adopted in the literature, but we
find that this is precisely the identification that works.
5 This formula occurs implicitly in [45] and explicitly in [13]; see also the remarks in [17].
Linearity implies that the central charge corresponding to any S is given in terms of the
charge vector by

Z(S) = n(Ci ) Z
SCi . (4.9)
Finally, we return to the subject of monodromy. In [1] it was conjectured (and pushed
further in the work of [3,4]) that the monodromies around loci in the moduli space where
the mirror X of a Calabi–Yau threefold X becomes singular induce autoequivalences of
b
D (X), the bounded derived category of coherent sheaves on X. Moreover, in the case of a
Fano surface embedded in a Calabi–Yau threefold, a relationship of these autoequivalences
of D b (X) with mutations of exceptional collections supported on the Fano surface was
pointed out in [4]. For our purposes we will view the various monodromies mainly as
automorphisms of K c (X). However, in some examples we will identify the monodromy
actions on the exceptional collections of coherent sheaves supported on the compact
divisors. As in the case of the local mirror geometry, we will be interested in the following
three types of transformations:
— Monodromy around large Kähler structure divisors in the moduli space;
— Monodromy around the primary component of the discriminant locus;
— ‘Orbifold monodromy’ in the case Cd /G.
Only the monodromy around a divisor zi = 0 in the moduli space where the Kähler
parameter ti (associated with the divisor class [Ti ] in X) becomes infinite allows for
a classical analysis. In this case we just take ti → ti + 1 in (4.5). Because of the
multiplicativity of Chern characters, the fact that the Chern character of a line bundle is
the exponential of its first Chern class and the form of (4.5), this transforms the Sj by
tensoring them with OX (−Ti ). By (4.1), the Ri transform by tensoring with OX (Ti ).
According to the observations in [9,46], ‘orbifold monodromy’ should cyclically
permute the Si if X is a resolution of Cd /Zn .
For the primary component of the discriminant locus we have the following picture:
in the case of a compact Calabi–Yau variety X it is conjectured (see [1,3–5,20]) that a
sheaf F is subjected to a Fourier–Mukai transform whose kernel is the structure sheaf OX ,
implying that the Chern character of F transforms as
ch(F ) −→ ch(F ) − OX , F ch(OX ), (4.10)
where , is the pairing (4.2). In our case of noncompact X this cannot work because it
would violate compact support conditions, but we make the following observation.
In all of our examples we obtain expressions for chc (S0− ) that allow us to choose S0−
in such a way that its restriction S0− |Ci to any compact toric cycle Ci is equal to OCi . For
the case of a resolution π of an orbifold singularity this means that our expressions for
chc (S0− ) are consistent with taking S0− to be the push-forward of the restriction of OX to
π −1 (0).
Wherever we have the possibility of comparison with the mirror geometry, we find that
the monodromy around the primary component of the discriminant locus is given by

ch(F ) −→ ch(F ) − S0− , F chc S0− . (4.11)
More precisely, the following happens: for one parameter models the principal component
is pointlike. If we decompose the GKZ solutions into logarithms and holomorphic pieces
at z = 0, the principal component is at the boundary of the radius of convergence of the

holomorphic pieces. In this case we find that the monodromy is given precisely by (4.11)
provided we choose the simplest anti-clockwise path, ln(−1) = πi and the identification
(4.8). With more than one parameter the discriminant locus consists of several disjoint
pieces in the z coordinate patch (these pieces join in the other coordinate patches), and there
is no unambiguous choice of component or path. We find, however, that at every branch
one of the Si− (possibly transformed by large complex structure monodromy) becomes
massless. This is consistent with the picture that when we take S0− along some nontrivial
paths like the ones corresponding to ‘orbifold monodromy’ we turn it into one of the
other Si− .
If d is even there is a simple consistency check: if we require (4.11) to respect the pairing
, then it is easily checked that this is equivalent to Si− , Si− = 2 (in odd dimensions the
analogous condition Si− , Si− = 0 is fulfilled automatically because of the skew symmetry
of , ). This is indeed true in all of our examples.
Example 1. Resolution of C2 /Zn .
The case of C2 /G with G any discrete subgroup of SU(2) is well understood by
mathematicians in the context of McKay correspondence. If G is Abelian and resolved
by the introduction of a set {Ci } of exceptional curves, and if {[Ci∨ ]} is a basis of divisor
classes dual to {Ci } then the RiM are given by R0M = OX and RiM = OX (Ci∨ ) for i 1. By
(2.6) sections of the Ri+ are given, for example, by 1, z0 , z02 z1 , z03 z12 z2 , . . . , so the action
of Zn on these sections through z0 → #z0 , zn → # n−1 zn indeed reproduces the characters
1, #, # 2 , . . . of Zn .
Using (4.1) and denoting by p the class of a point, we find

n−1

chc S0± = p ∓ Ci , chc Si± = ±Ci for i > 0,
i=1

ch
SCi = p + Ci , ch
Sp = p (4.12)
and therefore
n−1

chc S0− = SCi − (n − 2) ch
ch Sp .
i=1

The restriction of OX to the union of the Ci is the same as SCi except for the n − 2
points of the form Ci · Ci+1 where
SCi has rank two. Upon subtracting the n − 2
sheaves with support on these points we arrive at a class that matches chc (S0− ). It is
easily checked that Si− , Si− = 2 for all i. The large volume central charges are given

by Z lv (t; S0− ) = −1 + i ti and Z lv (t; Si− ) = −ti .
In the case of n = 2 this implies Z(S0− ) = 01 and Z(S1− ) = −1 − 01 and we see that the
principal component and orbifold monodromies found in the mirror geometry are precisely
the ones generated by (4.11) and permutations S0− ↔ S1− , respectively.
Example 2.
For Cn /Zn with Zn : (z1 , . . . , zn ) → (e2πi/n z1 , . . . , e2πi/n zn ) the restrictions of the
Ri to the exceptional divisor D Pn−1 are nothing but O, O(1), . . . , O(n − 1). The
M
independent holomorphic cycles are of the form Pj with 0 j n − 1 and the Ri± restrict
to OPj (±i). This example has been previously considered in [17,47,48]. We include it as
further evidence that the Si have the properties stated above. Defining

χkj := χ OPj (k), P = ch OPj (k) Td Pj
j
Pj
j +1
H
= ekH (4.13)
1 − e−H
Pj
with H the hyperplane divisor, we find that
j +1
H
χkj − χk−1,j = ekH 1 − e−H
1 − e−H
Pj
j j
H H
= e kH
H = e kH
= χk,j −1 . (4.14)
1 − e−H 1 − e−H
Pj Pj−1
k+j
With χk0 = χ0j = 1 this simple recursion is solved by χkj = j and we obtain
j ±i
(Ri± ,
SPj ) = j , implying (R0− ,
SPj ) = 1 for any j , (Ri− ,
SPj ) = 0 for 1 i j and
−
(Ri , SPj ) = (−1) j for i > j . This leads to the following expressions for the S − :
j i−1
S0− =
SPn−1 ,
n−2

n−1 j
(−1)k Sk− = SPn−1 − S j for k 1. (4.15)
k k−1 P
j =k−1
Again the restriction of S0− to any compact toric cycle is the same as the structure sheaf of
that cycle.
Alternatively we may determine the Si− by the ansatz Si− = aik i∗ OPn−1 (k). With

− n−1+k−i
Ri , i∗ OPn−1 (k) = χk−i,n−1 = (4.16)
n−1
we get (a −1 )ki = χk−i,n−1 which leads to

n
i
n
Si− = aik i∗ OPn−1 (k) = (−1)i−k i∗ OPn−1 (k), (4.17)
i −k
k=0 k=0

∗
Si− , Si− = aik ail ch i∗ OPn−1 (k) ch i∗ OPn−1 (l) Td(X)
k,l X

= aik ail 1 − e−nH e−kH elH Td Pn−1
k,l
Pn−1

= aik ail (χl−k,n−1 − χl−k−n,n−1 ). (4.18)
k,l
Using χl−k−n,n−1 = (−1)n−1 χk−l,n−1 and the reciprocity of a and χ we get

− −
Si , Si = aii 1 − (−1)n−1 = 0/2 for n odd/even, (4.19)
as it should be.
For n = 3 we find Z lv (t; S0− ) = −t 2 /2 + 3t/2 − 5/4. The corresponding GKZ system
has been studied at various places in the literature, e.g., in [36] and [9]. In terms of solutions

ln z ln z 2
00 = 1, 01 = + O(z), 02 = + O(z ln z), (4.20)
2πi 2πi
the rule t ∼ 1 + ln z/(2πi) leads to Z(S0− ) = −02 /2 + 01 /2 − 1/4. Comparing with
[9], we find that this is precisely the expression denoted there by td which vanishes at the
discriminant point z = −1/27.
Example 3.
The Kähler cone is generated by [D1 ] and [D2 ] corresponding to the characters # =
e2πi/5 and # 3 , respectively. By applying the rules outlined above, we assign the character #
to each of the three line segments meeting at v4 in Fig. 1 and to the line segment between
v5 and v2 , whereas the remaining two line segments (from v5 to v1 and v3 ) correspond
to # 3 . Thus we get F2 = 2D1 because of the three line segments with equal characters
meeting at v4 and D1 + D2 because of the two pairs of line segments at v5 . Altogether we
get representatives Ri± = O(±Fi ) with
F0 = 0, F1 = D1 , F2 = 2D1 , F3 = D2 , F4 = D1 + D2 , (4.21)
for the bases of K(X), where we have chosen the labels such that sections of Ri+ transform
as # i under (z1 , z2 , z3 ) → (#z1 , # 3 z2 , #z3 ). Using (4.1) we find that the localized Chern
characters of the basis of K c (X) are given by

c ± 3 5 11
ch S0 = D4 + D5 ∓ h + f + p,
2 2 6

3 4
chc S1± = −2D4 − D5 ± 2h + f − p,
2 3
1 1 5 1
chc S2± = D4 ∓ h + p, chc S3± = −D5 ± f − p,
2 2 2 3

c ± 3 1
ch S4 = D5 ∓ f + p, (4.22)
2 3
with p the class of a point.
Let us now consider the branes defined in terms of the structure sheaves Op , Oh , Of ,
OP2 , OF3 of the independent lower-dimensional cycles. Denoting by Sp the result of three
successive inclusion maps acting on Op , etc., we arrive with the help of the Grothendieck–
Riemann–Roch theorem (4.3) at the following result:
3 3 5 4
ch
SD4 = D4 + h + p, ch SD5 = D5 + h + f + p,
2 2 2 3

ch
Sp = p, ch Sh = h + p, ch
Sf = f + p. (4.23)
This allows us to determine the D-brane charges ni = (nD4 , nD5 , np , nh , nf ) with np the
D0-brane charge, nh , nf D2-brane charges and nD4 , nD5 D4-brane charges of the Si− as
n0 = (1, 1, 0, −1, 0), n1 = (−2, −1, 0, 2, 1),

n2 = (1, 0, 0, −1, 0), n3 = (0, −1, 0, 1, 0),
n4 = (0, 1, 1, −1, −1). (4.24)
In particular, this means that S0−= SD4 +
SD5 − Sh . Note how
SD4 + SD5 has rank 1 on
D4 and on D5 except on their intersection h, where it has rank 2 which is compensated by
subtracting
Sh .
At this point we would like to mention that we have performed a similar analysis for
C3 /Zn with arbitrary odd n and an action of the type (z1 , z2 , z3 ) → (#z1 , # n−2 z2 , #z3 ).
In that case the resolution requires a P2 and (n − 3)/2 Hirzebruch surfaces and we again
obtain Ri+ whose sections transform by the characters of Zn and an expression for S0− that
reduces to the structure sheaf on every compact toric cycle.
Returning to C3 /Z5 we now give an alternative description of the compactly supported
K-theory classes in terms of nontrivial sheaves on the exceptional divisors. Again with
the help of the Grothendieck–Riemann–Roch theorem, we find that we may choose
representatives Si in terms of the following combinations of push-forwards of sheaves:

S0− = j∗ OD5 (−h) + g∗ OD4 ,

S1− = −j∗ OD5 (−h − f ) − g∗ V , S2− = g∗ OD4 (−1),
S3− = −j∗ OD5 (−h), S4− = j∗ OD5 (−h − f ), (4.25)
with g : D4 >→ X and j : D5 >→ X inclusion maps. By V we denote a stable bundle on P2
with the Chern character given by
1
ch(V ) = 2 − h − p, (4.26)
2
where p is the class of a point on P2 . Note that {OP2 (−1), V , OP2 } is a foundation of
the helix of exceptional bundles on P2 and that {OF3 (−h − f ), OF3 (−h)} is a regular
exceptional pair on F3 (see [49]).
In terms of the pure brane basis
S the large volume central charge is

Z lv t1 , t2 ;
S = − e−(t1 D1 +t2 D2 ) ch
S Td(X)
X
 − 12 t12 + 32 t1 − 54 
 − 12 (3t22 + 2t1 t2 ) + t1 + 52 t2 − 7 
 6 
= −1 . (4.27)
 
t1 − 1
t2 − 1
We will now discuss monodromy by assuming that the assertions made in this section
are correct. The comparison with the mirror geometry is rather technical and can be found
in Appendix A. We want to find monodromy matrices acting on the charge vectors, n →
n · M, such that n · Z(

S) = n · M · Zmt (
S), where Zmt (
S) = M −1 Z(
S) is the monodromy
transformed version of Z. The monodromy around the orbifold locus cyclically permutes
the charge vectors (4.24). Therefore, we obtain:
 
−2 0 1 1 0
 −2 0 0 1 1 
 
Morb =  0 0 1 0 0. (4.28)
 
−2 1 1 0 0
−1 −2 0 2 1
Also, we can easily compute the large radius limit monodromies, Mt1 and Mt2 . On the
sheaves defined on the exceptional divisors the actions of the monodromies come from
tensoring with the restrictions of OX (D1 ) and OX (D2 ). Therefore, the large radius limit
monodromy Mt1 acts as following: the exceptional collection {OP2 (−n), V, OP2 (−n + 1)}
on P2 is mutated to another exceptional collection, {OP2 (−n + 1), V, OP2 (−n + 2)}, while
on F3 is given by the tensoring with OF3 (f ), therefore taking regular exceptional pairs into
regular exceptional pairs.
The action of the monodromy Mt2 is represented by tensoring any sheaf supported on F3
with OF3 (h + 3f ), hence again transforming regular pairs into regular pairs, while leaving
any sheaf supported on P2 invariant.
Using (4.11) it is possible to compute the action of the monodromy around the principal
component of the discriminant on the generators of K(X): Ri− with i = 1, . . . , 4 are
invariant under this transformation, but R0− "→ OX (−2D1 − D2 ). With the help of (4.4),
we readily obtain the monodromy around the conifold locus:
 
2 1 0 −1 0
 1 2 0 −1 0 
 
Mcon =  0 0 1 0 0  . (4.29)
 
2 2 0 −1 0
1 1 0 −1 1
The conifold monodromy, although preserving exceptional collections, acts in a very
different way on K c (X). For example, we have −[OF3 (−h)] "→ [OP2 ] and [OP2 (−1)] "→
−[OF3 ], that is the D4-branes can ‘jump’ from one exceptional divisor to another.
However, as remarked in [20] , this is not very surprising since autoequivalences of D b (X)
need not preserve the D-branes.
5. Beyond Cd /Zn
Up to now we have only considered cases of the type Cd /Zn with a single triangulation.
We now want to examine the range of validity of our statements regarding the Si− and
monodromy. We first present another example, the resolution of C3 /(Z2 × Z2 ), which
is still an orbifold but has several interesting features: it is not of the simple Zn type, it
allows for more than one triangulation, its resolution involves three new noncompact toric
divisors but no compact toric divisor, and finally it is a three parameter model whose GKZ
system can be solved explicitly. We will be able to show explicitly that the Si− vanish at
Fig. 6. G-Hilb resolution of C3 /Z2 × Z2 .
(branches of) the principal component of the discriminant locus and nowhere else. Aspects
of D-brane states on this model have been studied previously in, e.g., [50,51]. Finally we
examine the possibility of extending our results to cases not of the McKay type. We find
that they still hold in many examples but not in general.
Example 4. A toric resolution of C3 /(Z2 × Z2 ).
A singular space of the type C3 /(Z2 × Z2 ) where every nontrivial element of Z2 × Z2
acts by flipping the sign of two of the three coordinates of C3 can be resolved by
introducing three additional noncompact divisors and three compact curves. There are
several distinct possibilities for choosing the curves.
We use the G-Hilb resolution depicted in Fig. 6. The Mori cone is generated by the
following vectors:
l (1) = (1, 0, 0, 1, −1, −1),

l (2) = (0, 1, 0, −1, 1, −1),
l (3) = (0, 0, 1, −1, −1, 1). (5.1)
The generators of the Kähler cone are the divisors D1 , D2 and D3 corresponding to the
vanishing of the coordinates of C3 . The mirror geometry is determined by
a1 x12 + a2 x22 + a3 x32 + a1 x2 x3 + a2 x1 x3 + a3 x1 x2 = 0. (5.2)

The large complex structure coordinates zi are
a1 a1 a2 a2 a3 a3

z1 = , z2 = , z3 = (5.3)
a2 a3 a1 a3 a1 a2
and the orbifold coordinates are
a 1 a 1
φ1 = √ 1 = √ , φ2 = √ 2 = √ ,
a2 a3 z2 z3 a1 a3 z1 z3
a 1
φ3 = √ 3 = √ . (5.4)
a1 a2 z1 z2
The principal component of the discriminant locus is determined by
4z1 z2 z3 − z1 − z2 − z3 + 1 = 0. (5.5)
The simplest formulation of the GKZ system can be obtained by mixing large complex
structure and orbifold coordinates. We find the operator
D1 = (Θφ2 + Θφ3 )Θφ1 − 2z1 Θφ2 Θφ3 (5.6)

from the first Mori cone vector and the same operator with cyclically permuted indices for
the other two Mori cone vectors. This simply implies
Θφ2 Θφ3 Π = Θφ1 Θφ3 Π = Θφ1 Θφ2 Π = 0 (5.7)

for any solution Π , i.e., there must be a basis of solutions depending only on at most one
of the φi . The sums of two Mori cone vectors lead to operators of the type
D1 = (Θφ1 + Θφ3 )(Θφ1 + Θφ2 ) − 4z1 z2 Θφ1 (Θφ1 − 1), (5.8)
which upon using (5.4) and (5.7) implies

1 − 4φ1−2 Θφ1 + 4φ1−2 Θφ1 Π = 0. (5.9)
The whole GKZ system has three solutions of the type ln((φi + φi2 − 4)/2) and, as
always, a constant solution. Upon returning to large complex structure variables, we obtain
√
1 + 1 − 4z2 z3 1
Π0 = 1, Π1 = ln − (ln z2 + ln z3 ) (5.10)
2 2
and the corresponding index-permuted expressions for Π2 and Π3 .
The divisors Fi determining the line bundles Ri are just D1 , D2 and D3 and we find

chc S0− = p + C1 + C2 + C3 , chc S1− = −C1 ,

chc S2− = −C2 , chc S3− = −C3 , (5.11)
where C1 is the compact curve at the intersection of D2 and D3 , etc. In terms of structure
sheaves we can represent S0− as SC1 + SC2 + SC3 − 2Sp . Noticing that all three curves
intersect in the same point, we find that we can again view S0− as the object whose
restriction to any compact toric cycle is the structure sheaf of that cycle.
The central charges are determined by Z lv (t; S0− ) = −1 + t1 + t2 + t3 and Z lv (t; Si− ) =
−ti for i ∈ {1, 2, 3}, leading to
Π1 + Π2 + Π3
Z S0− = 2 − ,
2πi
−Π1 + Π2 + Π3
Z S1− = −1 + , etc. (5.12)
2πi
At the orbifold point φ1 = φ2 = φ3 = 0 we have the following situation: the moduli
space develops a Z2 × Z2 singularity. Provided we make the right choice of sheets for the
square roots and logarithms, we find Π1 = Π2 = Π3 = 3πi/2 and thus Z(Si− ) = −1/4 for
Fig. 7. Examples not of the McKay type.
Fig. 8. Symmetric triangulation of the half-hexagon.
any i. The ‘orbifold monodromy’ φ2 → −φ2 , φ3 → −φ3 acts as

Π2 ↔ 3πi − Π2 , Π3 ↔ 3πi − Π3 , S0− ↔ S1− , S2− ↔ S3− , (5.13)
and the other elements of the orbifold monodromy act in similar ways.
S0− can become massless only if

1 + 1 − 4z2 z3 1 + 1 − 4z1 z3 1 + 1 − 4z1 z2 = 8z1 z2 z3 . (5.14)
√
We can rewrite this in the form√ 1 − 4z2 z3 (E1 ) = (E2 ) such that E1 and E 2 are
expressions that do not contain 1 − 4z2 z3 . Then a necessary condition for (5.14) to
hold is (1 − 4z2 z3 )(E1 )2 = (E2 )2 and we can proceed to eliminate the other square roots
in the same way. The result is an equation proportional to the square of the expression
determining the principal component of the discriminant locus (5.5). Conversely, if we
solve (5.5), e.g., by setting z1 = (1 − z2 − z3 )/(1 − 4z2 z3 ), plug this into (5.10) and choose
the right sheets, we find that Z(S0− ) indeed vanishes. The same type of analysis works for
the other Si− .
At this point it is natural to ask whether the analog of S0− , i.e., the sheaf that is equal
to the structure sheaf upon restriction to any compact toric cycle but of rank zero away
from these cycles, might lead to monodromies in cases that are not related to McKay
correspondence. It turns out that this is very often the case (at least for sufficiently simple
examples), but not true in general.
Three examples where this works are shown in Fig. 7. The first of these is the resolution
of a conifold singularity and exactly solvable. The other two (anticanonical line bundles
over F0 P1 × P1 and F1 , respectively) are two parameter models that we treated with
analyses similar to the ones used, for example, 3 (C3 /Z5 ).
As a counterexample, consider the Calabi–Yau manifold depicted in Fig. 8, whose GKZ
system is again solvable. Here we find the following: if we choose as the line bundles Ri the
ones determined by the generators of the Kähler cone, we still find that the corresponding
S0− has the same restriction to compact toric cycles as OX . However, only two of the three
generators of K c (X) become massless at the conifold locus (these statements are true for
any triangulation). In particular, for the symmetric triangulation the vanishing locus of the
central charge of S0− does not coincide with the conifold locus.
Acknowledgements
We would like to thank Philip Candelas and Duiliu Diaconescu for very useful
conversations.
Appendix A. Comparison of GKZ solutions and K-theory results for C3 /Z5
The GKZ operators6 corresponding to the Mori cone generators (2.17) are given by
D(1) = ∂a1 ∂a3 ∂a5 − ∂a34 , D(2) = ∂a2 ∂a4 − ∂a25 . (A.1)
This can be turned into a system involving z1 , z2 by standard manipulations described
above. In this way we arrive at the following expressions in terms of Θzi := zi ∂z∂ i :
D1 = Θz21 (Θz1 − 2Θz2 )

− (Θz2 − 3Θz1 + 1)(Θz2 − 3Θz1 + 2)(Θz2 − 3Θz1 + 3)z1 ,
D2 = Θz2 (Θz2 − 3Θz1 ) − (Θz1 − 2Θz2 + 1)(Θz1 − 2Θz2 + 2)z2 . (A.2)
Solutions to this system can be obtained by considering
Π(z1 , z2 ; ρ1 , ρ2 )
∞
n +ρ n +ρ Γ (1 + ρ1 )2 Γ (1 + ρ2 )
:= z1 1 1 z2 2 2
Γ (1 + n1 + ρ1 ) Γ (1 + n2 + ρ2 )
2
n1 ,n2 =0
Γ (1 + ρ1 − 2ρ2 ) Γ (1 + ρ2 − 3ρ1 )
× (A.3)
Γ (1 + n1 − 2n2 + ρ1 − 2ρ2 ) Γ (1 + n2 − 3n1 + ρ2 − 3ρ1 )
(the coefficients of ni , ρi in the Γ -functions are the entries of the Mori cone vectors (2.17))
and its partial derivatives w.r.t. the ρi at ρ1 = ρ2 = 0. We use:

 1, n = 0,
Γ (1 + ρ) $ %−1
= (1 + ρ)(2 + ρ) · · · (n + ρ) , n > 0, (A.4)
Γ (1 + n + ρ) 
ρ(ρ − 1) · · · (ρ + n + 1), n < 0,
6 This GKZ system has also been studied in [52].

&
∂ Γ (1 + ρ) 0, n = 0,
= −Sn /n!, n > 0, (A.5)
∂ρ Γ (1 + n + ρ) ρ=0 (−1)−n−1 (−n − 1)!, n < 0,
& 0, n = 0,
∂2 Γ (1 + ρ)
= finite, n > 0, (A.6)
∂ρ 2 Γ (1 + n + ρ) ρ=0 2(−1)−n S (−n − 1)!, n < 0,
−n−1
where Sn = 1 + 1/2 + · · · + 1/n. This yields the constant solution Π(z1 , z2 ; 0, 0) = 1 and,
with Πi1 ···ik for (∂ k Π/∂ρi1 · · · ∂ρik )|ρ=0 ,
n n n n
Π1 = ln z1 + z1 1 z2 2 An1 n2 − 3 z1 1 z2 2 Bn1 n2 ,
n n n n
Π2 = ln z2 − 2 z1 1 z2 2 An1 n2 + z1 1 z2 2 Bn1 n2 ,

Π11 = (ln z1 ) + 2 ln z1
2
z1n1 z2n2 An1 n2−3 z1n1 z2n2 Bn1 n2
n n n n
−6 z1 1 z2 2 Cn1 n2 + z1 1 z2 2 An1 n2 (−2S2n2 −n1 −1 + 6Sn2 −3n1 − 4Sn1 )
n n
+ z1 1 z2 2 Bn1 n2 (−18S3n1 −n2 −1 + 6Sn1 −2n2 + 12Sn1 ),
n n
n1 n2
Π12 = ln z1 ln z2 + ln z1 −2 z1 z2 An1 n2 + 1 2
z1 z2 Bn1 n2
n n n n
n1 n2
+ ln z2 z1 z2 An1 n2 − 3 z1 z2 Bn1 n2 + 7
1 2
z1 1 z2 2 Cn1 n2
n n
+ z1 1 z2 2 An1 n2 (4S2n2 −n1 −1 − 7Sn2 −3n1 + 4Sn1 − Sn2 )
n n
+ z1 1 z2 2 Bn1 n2 (6S3n1 −n2 − 7Sn1 −2n2 − 2Sn1 + 3Sn2 ),
n n
Π22 = (ln z2 )2 + 2 ln z2 −2 z1n1 z2n2 An1 n2 + z1 1 z2 2 Bn1 n2
n n n n
−4 z1 1 z2 2 Cn1 n2 + z1 1 z2 2 An1 n2 (−8S2n2 −n1 −1 + 4Sn2 −3n1 + 4Sn2 )
n n
+ z1 1 z2 2 Bn1 n2 (−2S3n1 −n2 −1 + 4Sn1 −2n2 − 2Sn2 ),
where
(2n2 − n1 − 1)!
An1 n2 = (−1)2n2 −n1 −1 ,
(n2 − 3n1 )!(n1 !)2 n2 !
(3n1 − n2 − 1)!
Bn1 n2 = (−1)3n1 −n2 −1 ,
(n1 − 2n2 )!(n1 !)2 n2 !
(2n2 − n1 − 1)!(3n1 − n2 − 1)!
Cn1 n2 = (−1)2n2 −n1 −1+3n1 −n2 −1
(n1 !)2 n2 !
and the summations are taken over those values of n1 , n2 where the arguments of all
factorials are nonnegative. Of the three expressions obtained by taking second derivatives
only the first one and the linear combination 3Π22 + 2Π12 of the other two actually solve
the GKZ system (A.2). We note that there is also a linear combination of third derivatives
Fig. 9. The real part of the discriminant locus.
(involving third powers of logarithms) that is annihilated by both operators occurring in

(A.2). The reason is that this system is not yet complete as we have written it: in principle,
we should write down a GKZ operator for every curve in the Mori cone. Taking as an
additional charge vector the sum l (1) + l (2) of our Mori cone generators, we see that the
triple-log solution is excluded.
We now want to study monodromies of the GKZ solutions around the loci where R
degenerates. This is an easy exercise for the divisors z1 = 0, z2 = 0 where the monodromy
is determined by ln zi → ln zi + 2πi. We find that the following set of solutions is well
behaved (i.e., transforms by an SL(5, Z) matrix) under the monodromies zi → e2πi zi :
 
1
 2πi Π1 + const.
1 
 
 
2πi Π2 + const.
1
 . (A.7)
 1
Π − 2(2πi) Π1 + const.
1 
 2(2πi)2 11 
1
2(2πi)2
(3Π22 + 2Π12) − 1
2(2πi) Π2 + const.
In the large complex structure coordinate patch the discriminant locus consists of several
different branches. The slice through real z1 , z2 is shown in Fig. 9. There are two branches
with z1 0. The one with z2 > 0 is tangent to the axis z1 = 0 in z2 = 1/4. Parts of this
branch are at the boundary of the domain of convergence of the Π ’s in such a way that
there is convergence like 1/n2 . Through a numerical analysis we found that at this locus
1 1 1
− (2Π12 + 3Π22 ) + Π2 − = 0, (A.8)
2 2 6
which corresponds to the vanishing of Z(S4− ) if we make the identifications t1 ∼ 1 + ln z1 ,
t2 ∼ ln z2 . This is equivalent to the vanishing of a z2 -monodromy transformed version of
S4− with (4.8). Similarly we find at the other branch with z1 < 0 which intersects z2 = 0 at
z1 = −1/27 that
1 1 1
− Π11 + Π1 − = 0. (A.9)
2 2 4
This corresponds to a z1 -monodromy transformed version of S2− vanishing. The third
branch, with z1 > 0 and z2 < 0 is beyond the region of convergence. We have preliminary
evidence that at this branch a z2 -monodromy transformed version of S0− becomes
massless.
References
[1] M. Kontsevich, Homological algebra of mirror symmetry, alg-geom/9411018.

[2] E. Witten, D-branes and K-theory, JHEP 12 (1998) 019, alg-geom/9411018.
[3] R.P. Horja, Hypergeometric functions and mirror symmetry in toric varieties, math.AG/9912109.
[4] P. Seidel, R.P. Thomas, Braid group actions on derived categories of coherent sheaves, math.AG/0001043.
[5] R. Thomas, Mirror symmetry and actions of braid groups on derived categories, math.AG/0001044.
[6] S. Hosono, Local mirror symmetry and type IIA monodromy of Calabi–Yau manifolds, hep-th/0007071.
[7] B. Szendrői, Diffeomorphisms and families of Fourier–Mukai transforms in mirror symmetry,
math.AG/0103137.
[8] R.P. Horja, Derived category automorphisms from mirror symmetry, math.AG/0103231.
[9] D.-E. Diaconescu, J. Gomis, Fractional branes and boundary states in orbifold theories, JHEP 0010 (2000)
001, hep-th/9906242.
[10] B. Greene, C. Lazaroiu, Collapsing D-branes in Calabi–Yau moduli space: I, hep-th/0001025.
[11] M.R. Douglas, B. Fiol, C. Römelsberger, The spectrum of BPS branes on a noncompact Calabi–Yau, hep-
th/0003263.
[12] B. Fiol, M. Marino, BPS states and algebras from quivers, JHEP 0007 (2000) 031, hep-th/0006189.
[13] D.-E. Diaconescu, M.R. Douglas, D-branes on stringy Calabi–Yau manifolds, hep-th/0006224.
[14] K. Mohri, Y. Onjo, S.-K. Yang, Closed sub-monodromy problems, local mirror symmetry and branes on
orbifolds, hep-th/0009072.
[15] S. Govindarajan, T. Jayaraman, D-branes, exceptional sheaves and quivers on Calabi–Yau manifolds: from
Mukai to McKay, hep-th/0010196.
[16] A. Tomasiello, D-branes on Calabi–Yau manifolds and helices, JHEP 0102 (2001) 008, hep-th/0010217.
[17] P. Mayr, Phases of supersymmetric D-branes on Kähler manifolds and the McKay correspondence,
JHEP 0101 (2001) 018, hep-th/0010223.
[18] B. Andreas, G. Curio, D.H. Ruiperez, S.-T. Yau, Fourier–Mukai transform and mirror symmetry for D-branes
on elliptic Calabi–Yau, math.AG/0012196.
[19] W. Lerche, P. Mayr, J. Walcher, A new kind of McKay correspondence from non-Abelian gauge theories,
hep-th/0103114.
[20] P.S. Aspinwall, Some navigation rules for D-brane monodromy, hep-th/0102198.
[21] J. McKay, Graphs, singularities and finite groups, Proc. Symp. Pure Math. 37 (1980) 183.
[22] G. Gonzales-Sprinberg, J.-L. Verdier, Construction geometrique de la correspondance de McKay, Ann. Sci.
ENS 16 (1983) 409.
[23] M. Reid, McKay correspondence, alg-geom/9702016.
[24] I. Nakamura, Hilbert schemes of Abelian group orbits, preprint.
[25] Y. Ito, H. Nakajima, McKay correspondence and Hilbert schemes in dimension three, math.AG/9803120.
[26] T. Bridgeland, A. King, M. Reid, Mukai implies McKay, math.AG/9908027.
[27] A. Craw, M. Reid, How to calculate A-Hilb C3 , math.AG/9909085.
[28] A. Craw, An explicit construction of the McKay correspondence for A-Hilb C3 , math.AG/0010053.
[29] W. Fulton, Introduction to Toric Varieties, Princeton Univ. Press, Princeton, 1993.
[30] T. Oda, Convex Bodies and Algebraic Geometry, Springer, Berlin, 1988.
[31] D. Cox, The homogeneous coordinate ring of a toric variety, J. Alg. Geom. 4 (1995) 17, alg-geom/9210008.
[32] D. Cox, S. Katz, Mirror Symmetry and Algebraic Geometry, American Mathematical Society, Providence,
1999.
[33] E. Witten, Phases of N = 2 theories in two dimensions, Nucl. Phys. B 403 (1993) 159, hep-th/9301042.
[34] P.S. Aspinwall, B.R. Greene, D.R. Morrison, Multiple mirror manifolds and topology change in string
theory, Phys. Lett. B 303 (1993) 249, hep-th/9301043.
[35] T.-M. Chiang, A. Klemm, S.-T. Yau, E. Zaslow, Local mirror symmetry: calculations and interpretations,
Adv. Theor. Math. Phys. 3 (1999) 495, hep-th/9903053.
[36] P.S. Aspinwall, B.R. Greene, D.R. Morrison, Measuring small distances in N = 2 sigma models, Nucl. Phys.
B 420 (1994) 184, hep-th/9311042.
[37] M.R. Douglas, D-branes, categories and N = 1 supersymmetry, hep-th/0011017.
[38] P.S. Aspinwall, A. Lawrence, Derived categories and zero-brane stability, hep-th/0104147.
[39] C.I. Lazaroiu, Generalized complexes and string field theory, hep-th/0102122.
[40] D.-E. Diaconescu, Enhanced D-brane categories from string field theory, hep-th/0104200.
[41] W. Fulton, Intersection Theory, Springer, 1984.
[42] M.R. Douglas, B.R. Greene, D.R. Morrison, Orbifold resolution by D-branes, Nucl. Phys. B 506 (1997) 84,
hep-th/9704151.
[43] P. Baum, W. Fulton, R. MacPherson, Riemann–Roch and topological K-theory for singular varieties, Acta
Math. 143 (1979) 155.
[44] B. Iversen, Local Chern classes, Ann. Sci. École Norm. Sup. 9 (1976) 155.
[45] D.-E. Diaconescu, C. Römelsberger, D-branes and bundles on elliptic fibrations, Nucl. Phys. B 574 (2000)
245, hep-th/9910172.
[46] P.S. Aspinwall, Resolution of orbifold singularities in string theory, hep-th/9403123.
[47] E. Zaslow, Solitons and helices: the search for a math-physics bridge, Commun. Math. Phys. 175 (1996)
337, hep-th/9408133.
[48] K. Hori, A. Iqbal, C. Vafa, D-branes and mirror symmetry, hep-th/0005247.
[49] A.B. Kvichansky, D.Yu. Nogin, Exceptional collections on ruled surfaces, in: A.N. Rudakov, et al. (Eds.),
Helices and Vector Bundles: Seminaire Rudakov, in: London Math. Soc. Lecture Note Ser., Vol. 148,
Cambridge Univ. Press, Cambridge, 1995, p. 97.
[50] B.R. Greene, D-brane topology changing transitions, Nucl. Phys. B 525 (1998) 284, hep-th/9711124.
[51] P.S. Aspinwall, M.R. Plesser, D-branes, discrete torsion and the McKay correspondence, JHEP 0102 (2001)
009, hep-th/0009042.
[52] S. Mukhopadhyay, K. Ray, Fractional branes on a non-compact orbifold, hep-th/0102146.
Shortcuts for graviton propagation

in a six-dimensional brane world model
Elcio Abdalla, Adenauer Casali, Bertha Cuadros-Melgar
Instituto de Física, Universidade de São Paulo, CP 66.318, CEP 05315-970, São Paulo, Brazil
Received 22 May 2002; received in revised form 29 July 2002; accepted 4 September 2002
Abstract
We consider a six-dimensional brane world model with asymmetric warp factors for time and both
extra spatial coordinates, y and z. We derive the set of differential equations governing the shortest
graviton path and numerically solve it for AdS–Schwarzschild and AdS–Reissner–Nordström bulks.
In both cases we derive a set of conditions for the existence of shortcuts in bulks with shielded
singularities and show some examples of shortcuts obtained under these conditions. Consequences
are discussed.
PACS: 98.80.Hw; 97.60.Lf
Keywords: Extra dimensions; Brane world cosmology; Shortcuts; Graviton propagation
1. Introduction
The ideas of Kaluza and Klein [1], advocating the physical possibility of extra
dimensions in order to achieve the unification of different field theories can be considered
as a landmark in Quantum Field Theory.
Such an importance grew specially half a century after the original works, in the
framework of supergravity and string theories. In the latter, the existence of extra
dimensions is actually enforced by consistency.
Furthermore, new possibilities to realize the extra dimensions permitted to explore new
mechanisms of explaining unified field theories. The possibility of explaining hierarchies in
such a context is specially appealing and has been confirmed in the works of Arkani-Hamed
E-mail address: bertha@fma.if.usp.br (B. Cuadros-Melgar).
PII: S 0 5 5 0 - 3 2 1 3 ( 0 2 ) 0 0 7 9 7 - 6
202 E. Abdalla et al. / Nuclear Physics B 644 (2002) 201–222
and others [2,3]. The hierarchy between the electro-weak (∼ 100 GeV) and the Planck
(1019 GeV) scales has been focused by means of the consideration of extra dimensions at
a submilimeter size, which shows up in a theory of two extra dimensions connecting both
scales. Such an idea replaces the usual one where extra dimensions should only show up at
the Planck scale, and the tower of massive particles thus generated is above that level and
has a wider validity, including cosmology [4].
Such size constraints on the size of the extra dimensions constitute drawbacks in the
formulation of the theory.
More recently, Randall and Sundrum [5,6] proposed a model—or rather a class of
models–where there is a warp factor in the metric, such that even infinitely large extra
dimensions are allowed.
The existence of large extra dimensions with a warp factor naturally raises the question
of whether information can follow a shorter path outside the brane riding on gravitons
[7–10]. We proposed a simple calculation to establish the shortest path followed by a
graviton [11], which propagating in all dimensions in the so-called bulk, could in principle
follow a path which decouples from the brane, that is, from our universe, returning later
to another point, advanced in time with respect to a photon, which by construction must
follow a path in the brane, thus being delayed. In our previous paper we considered
a model constructed in [13] which was basically a generalized Friedmann–Robertson–
Walker universe with cosmological constant, with different scenarios in the brane (where
are living the Standard Model fields) and in the bulk. The result was actually a negative
one, that is, the shortest path followed by the graviton was the same as the one followed by
the photon, namely, inside the brane.
In a model introduced by Csáki et al. [14] the speed of light along flat four-dimensional
sections varies over the extra dimensions due to different warp factors for the space and
time coordinates, a construction similar to the one of Randall and Sundrum. Thus the
authors proposed that gravitational waves might travel faster than photons, which remain
in the brane. The delay between electromagnetic and gravity waves may be experimentally
detected with the gravitational waves detectors under way [10,15].
The models are basically AdS–Schwarzschild or AdS–Reissner–Nordström black holes
in the bulk. Brane models in AdS space with Schwarzschild singularities have been used to
understand the AdS/CFT correspondence and looks like a promising theoretical model [16,
17]. They are based on the Randall–Sundrum scheme [5,6], where a large mass hierarchy
is obtained with uncompactified dimensions from solutions of Einstein equations in higher
dimensions (i.e., in the bulk) with two separated branes. The four-dimensional part of
the metric is multiplied by a “warp” factor which is a rapidly changing function of the
additional dimension.
In this paper we consider a six-dimensional model and look for possible shortcuts
for AdS–Schwarzschild and AdS–Reissner–Nordström bulk configurations. The paper is
organized as follows. In Section 2 we describe a general six-dimensional model, derive
Einstein equations, and find the Israel conditions the metric has to satisfy due to the brane
embedding. At this point we choose a metric describing a six-dimensional black hole and
add a Z2 symmetry. In Section 3 we find the Euler–Lagrange equations which define
the graviton path in this model. Section 4 is devoted to study the numerical solutions
of these equations in the context of AdS–Schwarzschild bulk finding certain analytical
E. Abdalla et al. / Nuclear Physics B 644 (2002) 201–222 203
requirements for the existence of shortcuts. The AdS–Reissner–Nordström bulk is studied

in Section 5, where we perform an analytical discussion to impose a set of conditions
under which shortcuts can coexist with shielded singularities. Finally, consequences are
discussed in Section 6.
2. A six-dimensional model
We consider a six-dimensional model, such as the one constructed by Kanti et al. [18].
We also search for a solution of six-dimensional Einstein equation in AdS space of the
form

ds 2 = −n2 (t, y, z) dt 2 + a 2 (t, y, z) dΣk2 + b2 (t, y, z) dy 2 + c2 (t, y, z) dz2 , (1)
where dΣk2 represents the metric of the three-dimensional spatial sections with k =
−1, 0, 1 corresponding to a hyperbolic, a flat and an elliptic space, respectively.
The total energy–momentum tensor can be decomposed in two parts corresponding to
the bulk and the brane as
M(B) M(b)
T̃NM = T̆N + TN , (2)
where the brane contribution can be written as
M(b) δ(z − z0 )
TN = diag(−ρ, p, p, p, p̂, 0). (3)
bc
In order to have a well-defined geometry, the metric must be continuous across the
brane; however, its derivatives with respect to z can be discontinuous at the position of the
brane, generating a Dirac δ-function in the second derivatives of the metric with respect to
z [12]. These δ function terms must be matched with the components of the brane energy–
momentum tensor (3) in order to satisfy Einstein equations. Thus, we obtain the following
Israel conditions,
2
[∂z a] κ(6)
=− (p − p̂ + ρ),
a 0 b 0 c0 4
2
κ(6)
[∂z b]
2
= − ρ − 3(p − p̂) ,
b 0 c0 4
2
[∂z n] κ(6)
= p̂ + 3(p + ρ) . (4)
b0 c0 n0 4
A metric of the form (1) satisfying six-dimensional Einstein equations is given by
z2
ds 2 = −h(z) dt 2 + dΣk2 + h−1 (z) dz2, (5)
l2
where
dr 2
dΣk2 = + r 2 dΩ(2)
2
+ 1 − kr 2 dy 2, (6)
1 − kr 2
and
z2 M
h(z) = k + − 3 , for AdS–Schwarzschild bulk, (7)
l2 z
z 2 M Q2
h(z) = k + 2 − 3 + 6 , for AdS–Reissner–Nordström bulk, (8)
l z z
with l −2 ∝ −Λ (Λ being the cosmological constant), which describes a black hole in the
bulk, located at z = 0.
Following [14], we find a further solution by means of a Z2 symmetry inverting the
space with respect to the brane position. That is, considering a metric of the form
ds 2 = −A2 (z) dt 2 + B 2 (z) dΣ(4)

2
+ C 2 (z) dz2 (9)
and the brane to be defined at z = z0 , there is a solution given by
A(z), B(z), C(z), for z z0 ,

z2
A z02 /z , B z02 /z , C z02 /z 02 , for z z0 . (10)
z
The Z2 -symmetry corresponds to z → z02 /z.
The static brane still has to obey the Israel conditions (4), which for the metric (5) are
written as
2
[∂z a] κ(6)
= − ρ,
a02 c0 4
2
[∂z n] κ(6)
= (4p + 3ρ), (11)
a0 c0 n0 4
where here
2
[∂z a] = − ,
l
h (z0 )
[∂z n] = − √ . (12)
h(z0 )
3. The shortest cut equation
We consider the metric (1) with k = 0
ds 2 = −n2 (z) dt 2 + a 2 (z)f 2 (r) dr 2 + b2 (z) dy 2 + d 2 (z) dz2 , (13)

where the graviton path is defined equating (13) to zero. Therefore,

r t
n2 (z) − b 2 (z)ẏ 2 − d 2 (z)ż2
f (r ) dr = dt
a(z)
r0 t0
t

≡ L y(t), ẏ(t), z(t), ż(t); t dt (14)
t0
which naturally defines a Lagrangian density. The Euler–Lagrange equations of L define

the graviton path. We first choose to work at a constant y to check on the very possibility
of (13) allowing shortcuts. In this case the resulting equation is simple but far from trivial,

a n d 2 nn a n2
z̈ + −2 + ż + − = 0. (15)
a n d d2 a d2
√ √
For z z0 , a = z/ l, n = h(z), and d = 1/ h(z). For z z0 we have to use the Z2
symmetry showed up in (10).
Notice that this case is equivalent to consider the problem in five dimensions with the
metric shown in [14].
The most general case includes a y dependence on the graviton path; however, as we
will show in what follows, this dependence turns to be superflous and does not affect z-
equation since (15) is independently satisfied. This conclusion is not surprising if we notice
that the metric (13) is y-independent.
The two Euler–Lagrange equations considering a y dependence are then given by

2 a b 2 2
n − d ż ÿ + ż − + 2
2 2
n − d ż − nn + dd ż + d z̈ ẏ
2 2 2
a b

a b
+ b2 − żẏ 3 = 0 (16)
a b
and

a d 2
+
n2 − b2ẏ 2 z̈ + n − b2 ẏ 2 − 2nn + 2bbẏ 2 ż2 + b2 ẏ ÿ ż
a d

a 2 nn − bbẏ 2 2
+ − 2 n − b ẏ +
2 2
n − b2 ẏ 2 = 0. (17)
ad d2
It is clear that the case ẏ = 0 is a solution of this set of equations when at the same time
z obeys (15).
This set of equations can be handled leading to

zżẏ ż2
Fz + h(z) − Fy = 0,
h(z) h(z)

z2 ẏ 2 zżẏ
1− Fz + Fy = 0, (18)
h(z) h(z)
where

2 h (z)
Fy = ÿ + ż − ẏ,
z h(z)

z̈ ż2 3 h (z) h(z) h (z) h(z)

Fz = + 2 1 − z + − . (19)
z z 2 h(z) z 2 z
Since the determinant of the set (18) is non-zero, the solutions of (16) and (17) must satisfy
Fy = 0 and Fz = 0 independently. Furthermore, let us notice that Fy = 0 and Fz = 0 are
the null geodesic equations for y and z respectively obtained from
ẍ α + Γµν ẋ ẋ = λẋ α .
α µ ν
(20)
Thus, a null curve is extreme if and only if it is a null geodesic.
Then, as (19) is the same as (15), our problem is reduced to the previous case with
constant y described by (15).
For k = 0 cases we can also consider (15) as the shortcut equation if we assume
the existence of a y-symmetry in our problem. In this way, our model represents a
generalization of [14].
4. AdS–Schwarzschild bulk
From the Israel conditions (11) together with (12) we have

4 ρ2
κ(6)
h
= , (21)
z02 64
h κ(6)ρ 4 2
=− (4ω + 3), (22)
2z0 64
and we can obtain the black hole mass M as a function of the brane energy density ρ, while
ρ is fixed by a fine-tunning,
κ(6)ρ 4 2
M 2 k
= − (ω + 1) , (23)
z05 2
5 z0 40
4 ρ2
κ(6) 3k 5
=− − , (24)
64 z02 (8ω + 3) (8ω + 3)l 2
where ω = p/ρ.
As we saw in the previous section, the shortcuts in six dimensions are determined
from (15). We should also remember that the brane is static at z = z0 .
If a shortcut exists, there must be a time t = v in the graviton path when ż(v) = 0 and
z̈(v) 0. Thus, (15) evaluated at this point will give

h (zv ) h(zv )
z̈(v) + h(zv ) − = 0. (25)
2 zv
It is obvious that this minimum must be between the brane and the event horizon zh , if
a horizon exists. Otherwise, there is no turning point in the path since the graviton cannot
return after it goes through the event horizon. Hence, h(zv ) > 0. Thus, from (25) we require
h (zv ) h(zv )
F (zv ) = − 0 for zh < zv < z0 . (26)
2 zv
Using (7) this implies
5M k
F (zv ) = 4
− 0. (27)
2 zv zv
This equation has a zero in z = zf = 0 for k = 0
5
zf3 = M.
2k
Thus, for the k = 0 or k = −1 cases there is no positive root. Since the mass, M, is positive,
F (z) > 0 everywhere preventing the coexistence of shortcuts and horizons.
On the other hand, for k = 1 there is one real and positive root, which must satisfy
zf < z0 in order to have shortcuts. This is
5M
− 1 < 0.
2z03
Taking into account (23) and the fact that ε2 must be positive in (24)1
−4(ω + 1)ε2 z02 < 0, (28)
then
ω + 1 > 0. (29)
Now, let us study the conditions under which the event horizon must appear. In general,
the horizons occur at the zeros of h(z), or equivalently at the zeros of
z5 + z3 l 2 − l 2 M.
In the meantime, the non-vanishing zeros of h (z) occur when
2z5 + 3l 2 M = 0.
Since the derivative has no positive zeros with M > 0, there is just one event horizon. Then
as h(z) goes to −∞ at the origin, the conditions
M >0 (30)
and
h(z0 ) > 0 (31)
are necessary and, in fact, enough to have a horizon and assure that the brane lies after it.
1 From now on, we will denote ε 2 = κ 4 ρ 2 /64 in six dimensions.

(6)
Fig. 1. h(z) in six-dimensional AdS–Schwarzschild bulk with the brane located at z = 1/3. Notice that the
singularity is shielded by a horizon.
The condition (31) is automatically satisfied due to Eq. (21).

To fulfill (30) let us substitute (24) into (23) to have

3 z2
ω+ + (ω + 1) 20 < 0.
4 l
If ω + 1 0, this condition is always satisfied, but this configuration does not produce
shortcuts as we would like. However, the condition is also satisfied with ω + 1 > 0 if we
require
3
−1 < ω < − , (32)
4
and
z02 ω + 3/4
<− . (33)
l2 ω+1
If we follow both (32) and (33) together with the fine-tunning for the energy (24), we
will have several shortcuts in AdS–Schwarzschild bulks with shielded singularity. In Figs. 1
and 2 we illustrate an example with ω = −4/5, z0 = 1/3, and l = 1. Notice in Fig. 1 that
the horizon appears before the brane.
Since this case is equivalent to consider the problem in five dimensions with h(z), M
and ρ given in [14], analogous results are obtained. In this case, the fine-tunning in the
energy is given by2

1 1 2
2
ε(5) =− + , (34)
3ω + 1 z02 l 2
2 In this case ε 2 = κ 4 ρ 2 /36.

(5) (5)
Fig. 2. Shortcuts for several initial velocities in six-dimensional AdS–Schwarzschild bulk. Notice that there is a
threshold initial velocity for which the graviton cannot return to the brane and falls into the event horizon.
Fig. 3. h(z) in five-dimensional AdS–Schwarzschild bulk with the brane located at z = 1/2. Notice that the
singularity is shielded by a horizon.
and ω is confined to
2
−1 < ω < − , (35)
3
while the brane position is given by
z02 ω + 2/3
<− . (36)
l2 ω+1
An example is shown in Figs. 3 and 4 for ω = −3/4, z0 = 1/2, and l = 1.
Fig. 4. Shortcuts for several initial velocities in five-dimensional AdS–Schwarzschild bulk. As in the six
dimensional case, there is a threshold initial velocity for which the graviton cannot return to the brane and falls
into the event horizon.
5. AdS–Reissner–Nordström bulk
From the Israel conditions (11) we will have for the black hole mass and charge,
4
κ(6)
M 2k 8
= + + ρ 2 ω,
z05 z02 3l 2 24
4 2
Q2 k 5 8ω + 3 κ(6) ρ
= + + . (37)
z08 z02 3l 2 3 64
At this stage it is convenient to carefully study the possibility of existence of shortcuts
for every value of k.
5.1. k = 0 and k = −1 cases
As it was found in the AdS–Schwarzschild case, (26) determines the existence of

shortcuts. Using (8) we see that (26) has a zero in z = zf = 0 when
5
Mzf3 − 4Q2 − kzf6 = 0. (38)
2
If k = 0, we have a real root in
8Q
zf3 = . (39)
5M
If k = 1, we have two roots in
5 1
zf3 = M ± 25M 2 − 64Q2 . (40)
4 4
Finally, if k = −1, we have

5 1
zf3 = − M ± 25M 2 + 64Q2 . (41)
4 4
Notice that F (z) has at most one real and positive zero if k = 0, −1 and at most two
positive zeros if k = 1.
Analyzing h(z) and its derivative we see that h(z) tends to +∞ both at the singularity
and at infinity, while h (z) tends to −∞ at the singularity and to +∞ at infinity.
The horizons occur at the zeros of h(z), or equivalently, at the zeros of
z8 + l 2 kz6 − l 2 Mz3 + l 2 Q2 = 0. (42)
On the other hand, the non-vanishing zeros of h (z) occur when
2z + 3l Mz − 6l Q = 0.
8 2 3 2 2
(43)
This polynom grows at infinity being negative at the origin. Its derivative has non-vanishing
roots when
16z5 + 9l 2 M = 0. (44)
For M > 0 this equation is never satisfied. Thus, as the derivative of (43) does not vanish
and is positive outside the origin, the polynom (43) grows monotonically and has just
one root. The zeros of this polynom are all non-vanishing zeros of h (z). Therefore, we
conclude that for positive mass there is just one zero for h (z), and hence, at most two
horizons for h(z).
When there is one horizon, h (z) is negative before it and positive after, crossing h(z) at
the very horizon. If there are two horizons, h (z) vanishes at a point between Cauchy and
event horizons, being negative before this point and positive after, while h(z) is positive
at all points except between both horizons. Taking into account both the sign and zeros
of these functions, h (z) crosses h(z) between the Cauchy horizon and the point at which
h (z) vanishes.
Since h (z)/2 has the same sign as h (z) and vanishes at the same point, and in the same
way h(z)/z has the same sign of h(z) and vanishes at the same points, we conclude that,
existing horizons, F (z) necessarily vanishes at some point z = zc such that 0 < zc zh .
However, as we pointed out before, for k = 0 or k = −1 there is only one positive root of
F (z). As F (z) < 0 for z < zc , then F (z) > 0 for z > zc . Thus, because zc zh , F (z) > 0
for z > zh contrary to what was required in (26). This implies that there are no shortcuts
with k = 0 or k = −1 when horizons exist.
In five dimensions the proof is very similar and we arrive to the same conclusion.
5.2. k = 1 case
As we saw in the previous section, F (z) has two real, positive and distinct roots for
k = 1,
5 1
r13 = M − 25M 2 − 64Q2 , (45)
4 4
5 1
r23 = M + 25M 2 − 64Q2 . (46)
4 4
This is the only situation where the shortcuts can coexist with a shielded singularity. In
fact, this situation necessarily requires the second root of F (z) being at some point before
the brane position z0 . This also implies F (z0 ) < 0.
In addition, we must have both Q2 and M positive.
Given the fact that we have horizons, if the brane is not between them or at a horizon
position, then h(z0 ) > 0. Furthermore, in order to guarantee that the brane is located after
the event horizon, we also need h (z0 ) > 0.
From the discussion in the previous section we will have one or two horizons if and
only if h(r1 ) 0.
In summary, shortcuts in bulks with shielded singularities can occur only if k = 1 and
also if the following conditions are supplied:
(1) h(z0 ) > 0 and h (z0 ) > 0 to have both horizons before the brane;
(2) F (z0 ) < 0 and r2 < z0 to have shortcuts with shielded singularity;
(3) Q2 > 0 and M > 0, which assures the positivity of the black hole mass and charge;
(4) h(r1 ) 0 in order to have horizons.
We will analyze each condition and impose certain restrictions on ω, ρ 2 , and z0 .
5.2.1. Existence of both horizons before the brane

These conditions are the simplest to analyze since they restrict ω directly from the Israel
conditions (11) together with (12)
h(z0 )
= ε2 , (47)
z02
h (z0 )
= −(4ω + 3)ε2 . (48)
2z0
The condition (47) is automatically satisfied since ε2 > 0.
From the condition (48)
−(4ω + 3)ε2 > 0, (49)

we have our first restriction
ω < −3/4. (50)
5.2.2. Existence of shortcuts with shielded singularity

From the definition of F (z), (26), and using (47) and (48) we see that
0 > F (z0 ) = −4(ω + 1)ε2 z0 . (51)

Thus we find another condition on ω
ω + 1 > 0. (52)
Besides, from r2 < z0

5M 1
− z03 < − 25M 2 − 64Q2 . (53)
4 4
This equation will be satisfied if3
5M
− z03 < 0, (54)
4
or using (37)
3 3 10 5 10 5 2
z + z0 + z0 ωε < 0, (55)
2 0 3 3
and as ω < −3/4

1 9 z02
2 2
z0 ε > − − 2 . (56)
ω 20 l
5.2.3. Positivity of the black hole mass and charge

Because we require the positivity of the black hole mass, from (37) we have
M z02 3
>0 ⇒ + ωε2 z02 > − , (57)
z03 l2 4
thus,

1 3 z2
z02 ε2 < − − 20 . (58)
ω 4 l
Since 3/4 > 9/20 this condition is certainly compatible with (56).
On the other hand, the positivity of the squared black hole charge requires

Q2 5z02 8
> 0 ⇒ 1 + + ω + 1 z02 ε2 > 0, (59)
z06 3l 2 3
so that

1 5z2
z02 ε2 < −3 − 20 . (60)
8ω + 3 l
In spite of not being trivial, this equation is also compatible with (56). This requires

1 9 z02 1 5z02
− − 2 < −3 − 2 , (61)
ω 20 l 8ω + 3 l
or

1 9 z2
− ω+ − 20 (ω + 1) < 0, (62)
5 4 l
what is always true for −1 < ω < −3/4.
3 We assume that 25M 2 − 64Q2 > 0. We will return to this condition when we discuss the existence of
horizons, where we will impose a stronger restriction, M 2 − 4Q2 > 0.
5.2.4. Existence of horizons

This is the last and the more complicated of our conditions. We must have h(r1 ) 0.
Let x be r13 ,
x 8/3
+ x 2 − Mx + Q2 0. (63)
l2
We do not need to do a complete study of this equation. For our purposes it will be enough
to require
x 2 − Mx + Q2 < 0. (64)
Using (45) this implies
M 2 − 4Q2 > 0. (65)

This condition is necessary but not enough to have horizons. However, this restriction
added to the others developed in this section will be enough to construct shortcuts with
horizons as we will see. Notice that this condition is stronger than that one assumed before,
M > 8/5 Q.
Using (37) in (65),
16 z02
2 2 2
1 − ε2 l 2 + 1 + ωε l > 0. (66)
9 l2
We know from (56) that 1 − ε2 l 2 must be negative. Therefore, we must very carefully
analyze (66). We can interpret (66) as a quadratic equation in the energy

16 z02 32 2 16 2 2 2 4
1+ + ωz − l 2
ε 2
+ l z ω ε > 0, (67)
9 l2 9 0 9 0
what implies

1
z02 ε2 > −32ωz02 + 9l 2 + 3 −64ωz02l 2 + 9l 4 − 64l 2 z02 ω2 ω2 l 2 , (68)
32
or

1
z02 ε2 < −32ωz02 + 9l 2 − 3 −64ωz02l 2 + 9l 4 − 64l 2 z02 ω2 ω2 l 2 . (69)
32
Notice that because −1 < ω < −3/4,
−64ωz02l 2 − 64l 2 z02 ω2 = −64ω(ω + 1)z02 l 2 > 0

and all the previous roots are real and positive.
Summarizing, from the considerations in the previous sections, we must have for ω
−1 < ω < −3/4. (70)

For the energy,

1 9 z2 1 3 z2
− − 20 < z02 ε2 < − − 20 , or (71)
ω 20 l ω 4 l
1 9 z02 1 5z02
− − 2 < z0 ε <
2 2
−3 − 2 , (72)
ω 20 l 8ω + 3 l
depending on which condition is more restrictive.
In addition,

1
z02 ε2 > −32ωz02 + 9l 2 + 3 −64ωz02l 2 + 9l 4 − 64l 2 z02 ω2 ω2 l 2 , (73)
32
or

1
z02 ε2 < −32ωz02 + 9l 2 − 3 −64ωz02l 2 + 9l 4 − 64l 2 z02 ω2 ω2 l 2 . (74)
32
Now we are going to analyze the situations in which all these conditions are compatible.
Let us begin our analysis with Eq. (74). To be compatible with (71) and (72), we just
need

1 9 z02 1
− − 2 < −32ωz02 + 9l 2 − 3 −64ωz02l 2 + 9l 4 − 64l 2 z02 ω2 ω2 l 2 ,
ω 20 l 32
(75)
that is,

9 9 3 z2 z2
− ω− + −64ω 20 + 9 − 64 20 ω2 < 0. (76)
20 32 32 l l
Since 3/4 < |ω| < 1,
9 9
− ω−
20 32
will always be positive and thus, (76) will never be satisfied. Then, we conclude that (74)
is not compatible either with (71) or (72). This implies that z02 ε2 must satisfy (73) together
with (71) or (72).
Let us initially compare (71) with (73). We must have

1 3 z02 1
− − 2 > −32ωz02 + 9l 2 + 3 −64ωz02l 2 + 9l 4 − 64l 2 z02 ω2 ω2 l 2 ,
ω 4 l 32
(77)
that is,

3 9 3 z2 z2
− ω− − −64ω 20 + 9 − 64 20 ω2 > 0. (78)
4 32 32 l l
In this case, since ω is negative and 3/4 <| ω |< 1,
3 9
− ω− >0
4 32
and (78) can be satisfied if

3 9 2 9 z2 z2
− ω− > −64ω 20 + 9 − 64 20 ω2 , (79)
4 32 1024 l l
or
9 z2 9
ω(ω + 1) 20 + (3 + 4ω)ω > 0. (80)
16 l 64
Since ω + 1 > 0 and 3 + 4ω < 0, for positive z0 the inequality will be only fulfilled if

z0 1 3 + 4ω
< − . (81)
l 2 1+ω
So that (73) and (71) can be compatible.
Now we are going to analyze the compatibility between (73) and (72). We must have

1 5z02 z2 9 3
−3 − 2 + 02 − > −64ωz02l 2 + 9l 4 − 64l 2 z02 ω2 (82)
8ω + 3 l ωl 32ω2 32ω2
or using that 8ω + 3 < 0, simplifying, and squaring both sides we can write (82) as
2
2
1 2 z ω2 z2
ω (3 + 4ω)2 + ω2 (ω + 1)2 20 + (3 + 4ω)(ω + 1) 20 > 0. (83)
16 l 2 l
This polynom has just one root for z02 / l 2

z02 1 3 + 4ω
= − .
l2 4 1+ω
Since the coefficient of z04 / l 4 is positive, the inequality is satisfied with the same
condition (81), so we verify that both (72) and (71) are compatible with (73) under the
same restrictions.
Furthermore, let us compare the upper limits of (71) and (72). Suppose

1 3 z2 1 5z2
− − 20 > −3 − 20 , (84)
ω 4 l 8ω + 3 l
what can also be written as
3 3z2
− (4ω + 3) − 20 (ω + 1) > 0. (85)
4 l
Thus, (84) is satisfied if and only if

z02 1 3 + 4ω
<− , (86)
l2 4 1+ω
which is just the same inequality (81), that z0 must satisfy. Hence, between (71) and (72),
it is enough to take into account the latter. Nevertheless, from (83) notice that (72) would
be also compatible with (73) if

z02 1 3 + 4ω
> − ,
l2 4 1+ω
and (84) would be satisfied with a change of sign implying that we should consider (71)
instead of (72); however, as stated before, the compatibility of (71) and (73) requires

z02 1 3 + 4ω
<− ,
l2 4 1+ω
which contradicts our hypothesis. Therefore, the only possible configuration is (86).
At last, we compare the lower limits of (72) and (73). Suppose

1 9 z02 1
− − 2 < −32ωz0
2
+ 9l 2
+ 3 −64ωz0
2 l 2 + 9l 4 − 64l 2 z2 ω2 ,
0
ω 20 l 32ω2l 2
(87)
or

9 9 3 z2 z2
− ω− < −64ω 20 + 9 − 64 20 ω2 .
20 32 32 l l
Squaring and simplifying we obtain

z02 9 4
> − ω + 1 .
l2 20(ω + 1) 5
For −1 < ω < −3/4 this inequality is always satisfied since the right-hand side is negative.
Hence, we conclude that (87) is valid and between the lower limits for the energy in (72)
and (73), we just need to choose the latter.
In short, by purely analytic considerations we conclude that shortcuts in bulks having
no naked singularities and a static brane embedded in can only appear if k = 1 and if the
following conditions are satisfied:
(1) We must choose ω such that −1 < ω < −3/4;

(2) Given ω, the brane must be located at a position such that

z0 1 3 + 4ω
< − , (88)
l 2 1+ω
which is the same condition as AdS–Schwarzschild case (33);
(3) Given (88), the energy ε must satisfy

1 32ωz02 z02 z02 2

− 2 + 9 + 3 −64ω 2 + 9 − 64 2 ω
32ω2 l l l
2
1 5z
< z02 ε2 < −3 − 20 . (89)
8ω + 3 l
In this way, it turns out to be simple to find shortcuts in bulks with shielded singularities.
As an example, let us choose ω = −9/10. From (88) we must have
√
z0 6
< ,
l 2
then we choose l = 1 and z0 = 1.
Fig. 5. h(z) in six-dimensional AdS–Reissner–Nordström bulk with the brane located at z = 1. Notice that the
singularity is shielded by two horizons.
From (89) we have

35 5√ 40
+ 41 < ε2 < ,
24 72 21
√
so we choose ε = 238/125.
In Fig. 5 we plot h(z) with these conditions. Notice that the singularity is protected by
an event horizon and the brane is at z = z0 = 1.
In Fig. 6 we plot the graviton paths obtained from (15) under the previous conditions
for a variety of initial velocities showing that, in fact, shortcuts appear when we choose the
parameters following the complete analysis shown in this section.
The analysis in five dimensions can be performed analogously to six-dimensional one
with
z2 M Q2
h(z) = 1 + − 2 + 4, (90)
l2 z z
and
M 2 3
= + + 3ωε(5)
2
, (91)
z04 z02 l2
Q2 1 2
= + + 3ωε(5)
2
+ ε(5)
2
, (92)
z06 z02 l 2
arriving to the following restrictions:
(1) We must choose ω such that −1 < ω < −2/3;

(2) Given ω, the brane must be located at a position such that

z0 ω + 2/3
< − ; (93)
l 1+ω
Fig. 6. Shortcuts for several initial velocities in six-dimensional AdS–Reissner–Nordström bulk. Notice that there
is threshold initial velocity for which the graviton cannot return to the brane and falls into the event horizon.
2 = κ 4 ρ 2 /36 must satisfy

(3) Given (93), the energy ε(5) (5)
1 z02 z02 1 2z02

2 − 9 ω + 2 1 − 9 ω(ω + 1) < z 2 2
ε < −1 − .
9ω2 l2 l2 0 (5)
3ω + 1 l2
(94)
As an example, we choose ω = −7/8. From (93) we must have
√
z0 15
< ,
l 3
so we choose l = 1 and z0 = 1.
From (94) the energy must fulfill
632 16 √ 24
+ 2
127 < ε(5) < ,
441 441 13
√
then we choose ε(5) = 461/250.
In Fig. 7 we can see h(z) according to the previous conditions. As in the six-dimensional
case, the singularity is protected by an event horizon, and the brane is at z = z0 = 1.
In Fig. 8 we show several graviton paths obtained under the previous conditions for
several initial velocities showing that shortcuts appear when we choose the parameters
following the analysis shown in this section analogously to the six-dimensional case.
Fig. 7. h(z) in five-dimensional AdS–Reissner–Nordström bulk with the brane located at z = 1. We see that the
singularity is protected by two horizons.
Fig. 8. Shortcuts for several initial velocities in five-dimensional AdS–Reissner–Nordström bulk. We see that
there is threshold initial velocity for which the graviton cannot return to the brane and falls into the event horizon.
6. Conclusions
In this paper we have shown that the shortest graviton path is governed by just one
equation involving the “radial” extra coordinate. We have also seen that a symmetry in the
“angular” extra coordinate has permitted us to consider curved spatial sections.
As pointed out in Sections 4 and 5.1 the cases k = 0 and k = −1 display no shortcuts,
leading to the conclusion that k = 1 is a necessary condition for the existence of shortcuts,
what completely agrees with the observation of Chung and Freese [9].
In this way, the AdS–Schwarzschild and AdS–Reissner–Nordström bulks open up the

possibility of having shortcuts provided both the spatial section has positive curvature and
a set of strong restrictions on the brane intrinsic tension must be satisfied. Moreover, its
location in the bulk has to be respected. We should also notice that the energy is already
fine-tunned from the Israel conditions in the case of the AdS–Schwarzschild bulk.
Hence, it is interesting to notice that despite the fact that the charge contributes to have
a negative F (z0 ) and thus facilitates the existence of shortcuts, there are more restrictive
conditions for the energy coming from Q2 > 0 and from the horizons equation (63) which
do not appear in the uncharged case. In this way, the results favor the existence of shortcuts
in bulks with shielded singularities with the same conditions for ω and z0 as the AdS–
Schwarzschild case and also impose what is basically a fine-tunning in the energy. Thus,
both cases seem to be equivalent for the study of shortcuts in static universes with protected
singularities. We should realize that the restrictions to obtain these shortcuts make lose the
main advantage we should have when we study the AdS–Reissner–Nordström case, i.e.,
the absence of fine-tunnings for the intrinsic tension.
In spite of the existence of fine-tunnings, the fact is that shortcuts appear and
consequences are manifold. As mentioned in [7,8], the existence of shortcuts could
partially solve the horizon problem. We should notice that our set of conditions to obtain
shortcuts in AdS–Schwarzschild and AdS–Reissner–Nordström bulks with protected
singularities imposes a restriction on the size of the universe, namely z0 ∼ l, which
corresponds to a primeval universe. So the results shown in the present paper could
contribute to the solution of this important problem.
We can also point out experimental consequences. In particular, gravitational waves
advanced with respect to photons might be found in the proposed gravitational antennas
under way, in case we find a model for the Universe with a physical size.
Acknowledgements
This work has been supported by Fundação de Amparo à Pesquisa do Estado de São
Paulo (FAPESP) and Conselho Nacional de Desenvolvimento Científico e Tecnológico
(CNPq), Brazil.
References
[1] T. Kaluza, Sitzungsber. Preussische Akad. Wiss. K 1 (1921) 966;

O. Klein, Z. Phys. 37 (1926) 895;
O. Klein, Nature 118 (1926) 516.
[2] N. Arkani-Hamed, S. Dimopoulos, G. Dvali, Phys. Lett. B 429 (1998) 263.
[3] I. Antoniadis, N. Arkani-Hamed, S. Dimopoulos, G. Dvali, Phys. Lett. B 436 (1998) 257.
[4] E. Abdalla, L.A. Correa-Borbonet, Phys. Lett. B 489 (2000) 383;
N. Kaloper, A. Linde, Phys. Rev. D 60 (1999) 103509.
[5] L. Randall, R. Sundrum, Phys. Rev. Lett. 83 (1999) 3370.
[7] H. Ishihara, Phys. Rev. Lett. 86 (2001) 381.
[8] R. Caldwell, D. Langlois, Phys. Lett. B 511 (2001) 129, gr-qc/0103070.
[9] D.J. Chung, K. Freese, Phys. Rev. D 62 (2000) 063513.

[10] D.J. Chung, E.W. Kolb, A. Riotto, Phys. Rev. D 65 (2002) 083516, hep-ph/0008126;
D.J. Chung, Talk at Santa Fe 2000 Summer Workshop “Supersymmetry, Branes and Extra Dimensions”.
[11] E. Abdalla, B. Cuadros-Melgar, S. Feng, B. Wang, Phys. Rev. D 65 (2002) 083512, hep-th/0109024;
B. Cuadros-Melgar, Talk at Spanish Relativity Meeting E.R.E., 2001.
[12] P. Binétruy, C. Deffayet, D. Langlois, Nucl. Phys. B 565 (2000) 269, hep-th/9905012.
[13] P. Binétruy, C. Deffayet, U. Ellwanger, D. Langlois, Phys. Lett. B 477 (2000) 285.
[14] C. Csáki, J. Erlich, C. Grojean, Nucl. Phys. B 604 (2001) 312, hep-th/0012143.
[15] M.A. Clayton, J.W. Moffat, Phys. Lett. B 460 (1999) 263;
I.T. Drummond, gr-qc/9908058;
M.A. Clayton, J.W. Moffat, Phys. Lett. B 477 (2000) 269.
[16] B. Wang, E. Abdalla, R.K. Su, Phys. Lett. B 503 (2001) 394;
B. Wang, E. Abdalla, R.K. Su, Mod. Phys. Lett. A 17 (2002) 23.
[17] P. Creminelli, Phys. Lett. B 532 (2002) 284.
[18] P. Kanti, R. Madden, K.A. Olive, Phys. Rev. D 64 (2001) 044021, hep-th/0104177.
Convergence rate and locality of improved overlap

fermions
W. Bietenholz
Institut für Physik, Humboldt Universität zu Berlin, Invalidenstr. 110, D-10115 Berlin, Germany
Received 18 June 2002; accepted 29 August 2002
Abstract
We construct new Ginsparg–Wilson fermions for QCD by inserting an approximately chiral Dirac
operator—which involves ingredients of a perfect action—into the overlap formula. This accelerates
the convergence of the overlap Dirac operator by a factor of 5 compared to the standard construction,
which inserts the Wilson fermion as a point of departure. Taking into account the effort for treating
the improved fermion, we are left with an total computational overhead of about a factor 3. This
remaining factor is likely to be compensated by other virtues; here we show that the level of locality
is clearly improved, so that the exponent of the correlation decay is doubled. We also show that
approximate rotation invariance is drastically improved, but a careful scaling test has to be postponed.
PACS: 11.15.Ha; 11.30.Rd; 12.38.-t
Keywords: Lattice fermions; Chiral symmetry; Locality; Rotation invariance
1. Introduction
Over the recent years, substantial progress has been achieved in the long-standing
problem of constructing a formulation of chiral fermions on the lattice. It turned out that
there is a particularly harmless way to break the full chirality of the lattice Dirac operator,
so that the physical properties related to chirality are still represented correctly. This
breaking term is sufficient to circumvent the Nielsen–Ninomiya No-Go theorem [1], which
refers to full chirality, in the sense that the lattice Dirac operator D anti-commutes with γ5 .
E-mail address: bietenho@physik.hu-berlin.de (W. Bietenholz).
PII: S 0 5 5 0 - 3 2 1 3 ( 0 2 ) 0 0 7 8 9 - 7
224 W. Bietenholz / Nuclear Physics B 644 (2002) 223–247
This harmless breaking is characterized by the Ginsparg–Wilson relation (GWR) [2]
{D, γ5 }x,y = 2(Dγ5 RD)x,y , (1)

where the kernel R has to be local and it must not anti-commute with γ5 . The standard
choice for this term reads
1
st
Rx,y = δx,y , µ > 0. (2)
2µ
The general class of solutions to the GWR with the standard kernel R st is given by the
so-called overlap formula [3]

Dov = Aov + µ, Aov = µA0 A†0 A0 , A0 = D0 − µ. (3)
For D0 one may insert some sensible lattice Dirac operator. It has to be local (in the sense
that the correlations decay exponentially) and free of fermion doubling, but we do not
require any form of chirality for D0 . The resulting operator A0 is then transformed into an
operator Aov , so that µ1 Aov is unitary. This provides the overlap Dirac operator Dov , which
solves the GWR. The choice of the mass parameter µ is constrained; inside its allowed
interval it can be optimized with respect to certain criteria, see below.
From Eq. (3) it is obvious that the spectrum of a GW fermion—with the standard kernel
given in Eq. (2)—is always situated on a circle in the complex plane with center and radius
µ; we denote it as the GW circle. This condition for the spectrum is equivalent to the GWR
with kernel R st .
In almost all the literature, the Wilson operator DW was inserted as D0 into the overlap
formula, without consideration of other options. We denote the resulting particular overlap
fermion as the Neuberger fermion. As any Ginsparg–Wilson fermion (GW fermion), it
has a chiral symmetry which is lattice modified but exact [4]. Since the corresponding
transformation is local—based on the locality of R—the chiral limit is not plagued by
mass renormalization.
The generalization of the overlap fermions to a whole class of solutions of the GWR—
and the motivation for considering alternative operators D0 = DW —were given in Ref. [5].
In fact, there are further issues of importance for a lattice fermion formulation, such as the
quality of the scaling behavior, the level of locality and of approximate rotation invariance.
With these respects, the Neuberger fermion is not quite satisfactory, even though it is local
up to moderate coupling strength [6]. All these properties depend on the choice of D0 ,
hence it is motivated to search for a better starting operator than DW .
Moreover, the practical evaluation of Dov is a formidable numeric task [7]. In particular,
the quenched simulation of the Neuberger fermion is about as expensive as the simulation
of dynamical Wilson fermions [8], i.e., the computational overhead amounts to about two
orders of magnitude. Also with this respect a better choice of D0 can help. Our guide-line
is the following observation: if we insert a GW fermion (with R st ) as D0 , then it is simply
reproduced by the overlap formula (3), Dov = D0 . 1 Therefore, in practice we try to insert
1 Of course, this property holds for any fixed GW kernel R, if one uses the suitably generalized overlap
formula [5].
W. Bietenholz / Nuclear Physics B 644 (2002) 223–247 225
an approximate GW fermion as D0 [5], so that it is altered only a little by the overlap

formula,
Dov ≈ D0 . (4)
This small alteration corresponds to a chiral correction (in the sense of the GWR).2
Thus the first issue is to construct an approximate GW fermion “by hand”. At this
point, we recall that also perfect [2] and classically perfect fermions [10] solve the GWR.
Their construction is very tedious, but in the present context we can use a relatively simple
approximation, namely, a hypercube fermion (HF). The mass renormalization is still quite
strong for simple HFs [11,12], which is a problem in their direct application [13]. However,
here we reach the chiral limit by inserting the HF into the overlap formula.
Of course, the (classically) perfect fermion has further virtues in addition to chirality.
In particular its scaling behavior is (almost) free of lattice artifacts, i.e., dimensionless
ratios of physical observables are (almost) independent of the lattice spacing. Moreover,
the observables are also rotation-invariant for perfect fermions, and rotation symmetry
is approximated very well for the truncated perfect hypercube fermions, as the pion
dispersion shows [11]. Since the overlap formula modifies the HF just a little—see relation
(4)—we expect the good scaling and rotation behavior to persist under the chiral correction.
Then we obtain an improved overlap fermion.
There is yet another virtue to be expected based on relation (4): the hypercube fermion
is short-ranged—its free couplings are restricted to a unit hypercube on the lattice—hence
we also expect a high level of locality for the overlap-HF (given by Dov if we insert
D0 = DHF ). Due to the modest alteration, the long range couplings can be turned on
just slightly, hence their exponential decay will be fast. The Wilson fermion is also short-
ranged, but it changes drastically in the transition to the Neuberger fermion, so we do
not have the above reason to expect a good locality. Indeed, even for the free Neuberger
fermion the decay of couplings (in the separation of fermion and anti-fermion) is a rather
slow exponential [5].
All these expected virtues of the overlap-HF have been tested and impressively
confirmed in the 2-flavor Schwinger model [14]. Now we carry on this program to QCD.
In Section 2, we first describe the construction of a suitable HF in d = 4, and we illustrate
its proximity to a GW fermion by evaluating typical fermionic spectra on small lattices.
In Section 3 we discuss the polynomial evaluation of the inverse square root in Eq. (3).
In Section 4 we investigate the speed of the numeric evaluation of the overlap formula.
We show that for our HF a polynomial of a low order is sufficient to approximate the
sign function or the inverse square root to a high accuracy. For the Neuberger fermion
the same accuracy requires a polynomial of a much higher degree. Section 5 gives results
for the condition numbers and their impact on the convergence rate, now proceeding to
larger lattices. Section 6 presents a comparative study of the level of locality, i.e., of the
exponential decay of the maximal correlation at large distances, and Section 7 compares
the violations of rotation invariance. Section 8 contains a summary and our conclusions.
2 Similarly, one could also insert an approximately chiral D in the 4d space of domain wall fermions [5,9].
0
2. Approximate Ginsparg–Wilson fermions in QCD
For free or perturbatively interacting fermions, perfect actions can be constructed an-
alytically [15]. In the case of non-perturbative interactions, this is possible only numeri-
cally and only in the classical approximation.3 The renormalization group transformation
leading to the (quantum) perfect action would involve effectively a continuum functional
integral. Still its existence is of conceptual interest. It implies, for instance, that also a
topological lattice charge without lattice artifacts exists [17], and that supersymmetry can
in principle be represented continuously on the lattice [18]. Similarly, the perturbatively
perfect action shows that, e.g., the axial anomaly is reproduced correctly [19], but only a
modest improvement persists on the non-perturbative level [13,20]. What is (in principle)
accessible and promising for simulations is an approximation to the classically perfect ac-
tion [21], which works well for fermions in two dimensions [22]. However, the difficult
construction and application in d = 4 is still under investigation [23]. In that case, there is
also a potential for an excellent scaling, but in order to obtain a sufficient chiral quality, it
seems that the group working on it (the Bern Collaboration) also depends on the concept
of Refs. [5,14] (chiral correction).
In the free case, the truncated perfect HF has still an excellent scaling behavior [11,24],
if the renormalization group parameters are optimized for locality. Hence we use the free
HF as the point of departure in our attempt to construct a HF, which is an approximate
GW fermion, and which is also promising with respect to scaling and approximate rotation
invariance. A few elements for its gauging are then added such that the GWR is violated
only modestly at the coupling strength of interest. In this section, we are going to describe
this construction step by step. A synopsis of this construction has been anticipated in Refs.
[25].
The concept formulated in Refs. [5,14] has also been adapted in Ref. [26], where some
progress is reported, although a very simple (“planar”) operator D0 was used, which is
quite far from a GW fermion even in the free case. With this “planar overlap operator” some
results for the finite size scaling of the chiral condensate of Ref. [27] were reproduced,
and the effects of instantons in the chiral symmetry breaking were reconsidered [28].
Other very simple non-standard operators D0 were used in Refs. [29,30]. In contrast, a
very complicated approximate Ginsparg–Wilson fermion was constructed in Ref. [31] by
introducing many parameters and tuning them for a minimal GWR violation at a specific
value of β. This corresponds to the first part of our program (constructing an approximate
GW fermion by hand), but since the overlap formula is not used, chirality is still not exact.
Moreover, there are no ingredients in favor of improving other properties. However, that
work reports some gain if a specific improved gauge action is used. Different improved
gauge actions were applied in Refs. [23,32]. The use of improved gauge actions is also on
our agenda, but in the present work we always use the standard plaquette gauge action.
This allows us to observe unambiguously the progress due to the improved Dirac operator.
3 For recent work on perfect actions for semi-classical effective actions, see Ref. [16].
All the considerations below are based on quenched QCD configurations on periodic
lattices of sizes 44 , 84 (Sections 2, 3,4) and 124 (Sections 5, 6 and 7), and beyond Section 2
we always use β = 6.
2.1. Step 1: Minimal gauging
As we mentioned above, we start from the couplings of the free perfect fermion. It
is optimized for locality, and then truncated to a HF by imposing periodic boundary
conditions. We write the resulting Dirac operator as
D(x − y) = ρµ (x − y)γµ + λ(x − y) (x, y ∈ Z4 , |xν − yν | 1), (5)
and the couplings of the vector term ρµ and the scalar term λ are given in Ref. [11], Table
1. Note that the components of x − y are restricted to −1, 0, 1, and that ρµ is odd in the
µ-direction and even otherwise, while λ is entirely even. In the free case, this HF is an
excellent approximation to a Ginsparg–Wilson fermion [5].
Our first HF for QCD, with a Dirac operator of the form
D(x, y, U ) = ρµ (x, y, U )γµ + λ(x, y, U ) (U : compact gauge field) (6)
is now obtained by “minimal gauging” of the free HF: the sites, which are coupled in the
HF formulation, are connected by shortest lattice path only. The gauging is done by simply
attaching the free coupling to these paths, divided into equal parts where several shortest
paths exist.
This simple prescription already provides an excellent pion dispersion relation [11].
On the other hand, with such a gauging the HF suffers from a strong additive mass
Fig. 1. The fermionic spectrum for the minimally gauged HF and for the Wilson fermion, in a typical configuration
at β = 5 on a 44 lattice.
renormalization [11,12], comparable to the Wilson fermion. We illustrate this at strong

coupling in Fig. 1: it shows the full Dirac operator spectra for the Wilson fermion and the
HF for a typical configuration at β = 5 on a 44 lattice.
In such a rough gauge background, the overlap formula is only applicable if we have an
excellent GW approximation to be inserted as D0 . Fig. 1 shows that the standard Neuberger
fermion, D0 = DW , certainly fails, and the minimally gauged HF cannot handle such a
strong coupling either [33].4
Since we assume the kernel R to be of the form (2), the overlap formula performs a
sort of projection of the eigenvalues onto the GW circle. This projection tends to be close
to radial [14,34]. Hence we need a statistically safe distinction between a left arc and a
right arc of the spectrum of D0 to start with, so that it can be mapped onto a GW circle
in a reliable way. Here we focus on the regime close to the real axis. Eigenvalues have
to cross the center as one changes topological sectors—since we define the topological
charge by the index theorem5—but they have to do so quickly as the deformation proceeds,
so that the distinction is clear for the typical configurations at the coupling strength under
consideration.
However, in the present case the freedom of choosing this center does still not help:
wherever we choose µ, it will frequently happen that too many zeros occur (due to too
many mappings to the left), so that the doubling problem is back, or that too many
Fig. 2. Typical spectra of the minimally gauged HF at β = 5.4 (on the left) and at β = 5.6 (on the right). In the
latter case, the overlap formula seems to be applicable with mass parameter µ ≈ 1.4.
4 Note that in such a situation it could be misguiding to consider only the spectrum of A† A , as it is sometimes
0 0
done in the literature. For the HF in Fig. 1, that spectrum alone would look quite satisfactory, even though the
physically crucial left arc is missing.
5 This is a sensible definition of the topological charge on the lattice for any Ginsparg–Wilson fermion [4].
Up to moderate coupling strength, it is often close to the geometrical charge [35]. In general, it also tends to
agree with the charge identified from cooling the SU(N ) gauge configurations, especially at increasing N [36].
In the case of classically perfect fermions it corresponds to the classically perfect charge [10], and analogously
for (quantum) perfect fermions.
eigenvalues are mapped to the right-hand side, so that mass renormalization reappears.
Also further generalizations of R (beyond the restriction to one site) do not help in this
situation.
A question of interest is at which coupling strength the applicability of the overlap
formula with simple operators like DHF sets in. Typical spectra on a 44 lattice suggest
that for instance at β = 5.4 the coupling is still too strong, but at β 5.6 we are about to
approach to safer grounds, see Fig. 2. The same is true for the Wilson fermion (see first
Ref. in [25], Fig. 8). On larger lattices the minimal β is still likely to rise to about β 5.7.
2.2. Step 2: Critical link amplification
Worried about the strong mass renormalization, we first want to move our HF towards
the chiral limit. We do so by amplifying each link variable by a factor 1/u,
1
Uµ (x) → Uµ (x), u 1. (7)
u
The idea is to compensate the (mean) link suppression due to the gauge field. This can be
viewed as a generalization of the critical hopping parameter used for Wilson fermions, but
it is also related to the spirit of “tadpole improvement” [37].
Fig. 3 illustrates that criticality is reached with u 0.8 at β = 6.6 We see that this
already provides a decent approximation to a GW fermion. This is remarkable, because we
Fig. 3. HF spectra at β = 6 on a 44 lattice at critical link amplification (u = 0.8) for three configurations.
6 The absence of the arc close to 0 is due to the small size of the lattice; it is not due to the choice of the lattice
Dirac operator.
approach the chiral limit in the most economic way, by staying within the framework of
minimal gauging. We did not introduce additional lattice paths yet, hence Step 2 does not
require any computational effort (once the critical value of u is determined).
2.3. Step 3: Fat links
We now want to improve the chiral quality of our HF further by going beyond minimal
gauging. As a non-minimal element we introduce fat links. For a given configurations, we
substitute each link variable according to the following scheme

α
link → (1 − α) · link + staples , α ∈ R, (8)
6
before evaluating the Dirac operator. This is still an economic tool. The substituted link
variable on the right-hand side is not mapped back onto the gauge group, in contrast to the
APE blocking [39], and we do not iterate the substitution (8).
As a first observation, we note that the mass renormalization is enhanced for
increasing α. This may seem counter-intuitive if one imagines that a positive α makes
the configurations “appear smoother”. However, a more precise picture confirms this
observation: the only coupling of range 0 is λ(0, 0, 0, 0) = 1.853 . . . The rest of the scalar
term is negative, and in the free case we have x λ(x) = 0. The gauge field now suppresses
the negative contributions, whereas λ(0, 0, 0, 0) remains unchanged, so that a positive mass
sets in (if we keep u = 1). This is already true for minimal gauging, but if we add fat
links with α > 0, this suppression of the negative part gets even stronger. A fraction of
the negative couplings is now attached to staples instead of single links, and the staple
suppression corresponds to the third power of the mean link suppression.
If we wanted to use fat links to move closer to the chiral limit, we had to take α < 0.
Then the critical value of u rises, and for some strongly negative α (far below −1) it even
arrives at 1, so that criticality could, in principle, be realized solely by means of fat links.
Fig. 4. The effect due to the variation of staple coefficient α in the uncritical HF (on the left). On the right: the
spectrum of the critical HF with and without fat links (everything at L = 4, β = 6).
However, in this case the rest of the spectrum is very far from a GW circle—in particular,
the upper and lower arc are far outside the GW circle—so we do not recommend negative
values of α. Fig. 4 (on the left) shows this effect, and we also see that positive α are
more adequate to make the shape of the spectrum circle-like. So what is really favorable
to improve the proximity to a GW fermion is a positive α along with an adapted (i.e.,
decreased) critical value of u. A good choice is α = 0.3, which requires u = 0.76 at
β = 6. This reduces the radial fluctuation of the eigenvalues—the eigenvalues move closer
together—and the upper and lower arc move closer to the GW circle. This is shown in
Fig. 4 (on the right), which compares the spectra with and without fat links for the same
configuration.
2.4. Step 4: Vector term suppression
The above picture for the mass renormalization ignores the role of the vector term
ρµ (x, y, U ). In fact, some tests confirmed that it has only a very modest influence on the
location of the arc on the left-hand side (and also on the right-hand side). This location is
essentially determined by the scalar term, but the vector term is crucial for the height of
the upper and lower arc, i.e., the term ρµ γµ is responsible for the imaginary part of the
spectrum.
This property will now be used for a further improvement, still without extra
computational effort. We introduce different link amplification factors for the vector term
and the scalar term, since they play a different role. We first keep the critical factor 1/u as
Fig. 5. The HF on a 44 lattice with critical link amplification, fat links and a suppression of the vector term. We
also show the “continuation” around 0 on a 84 lattice with the same parameters, i.e., u = 0.76, α = 0.3, v = 0.92.
an overall link amplification, but then we multiply a link suppression factor v only in the
vector term. So now we multiply the links as follows:
v
Uµ → Uµ in ρµ (x, y, U ),
u
1
Uµ → Uµ in λ(x, y, U ), (9)
u
where u v 1. Still the fat links are useful to suppress the radial fluctuations of the
eigenvalues, so we stay with α = 0.3, u = 0.76, and the suitable vector term suppression
amounts to v = 0.92. Now the upper and lower arc also follow the GW circle, and we
obtain therefore a very satisfactory spectrum, see Fig. 5. It shows again the spectrum on a
44 lattice, but this time we also include part of the spectrum of a typical configuration on
a 84 lattice. From there we show the 100 eigenvalues with the smallest real parts, thus we
also visualize how the spectrum “continues” around zero.
2.5. A comment on the clover term
An obvious candidate for a next step beyond minimal gauging is the clover term. We
performed a sequence of tests with it being added to the HF versions discussed above. We
varied the clover coefficient and also considered both signs, but from spectra on 44 lattices
we did not arrive at a clear further improvement of the GW approximation in this way.
A positive coefficient has generally the effect to pull the spectrum closer to the real axis,
as it was observed before for the Wilson fermion [38], hence the optimal parameter v rises
Fig. 6. The effect of the clover term in the HF spectrum.

a little. A positive clover coefficient does improve the arc around zero—which appears on
the 84 lattice—a little bit, but it distorts the rest. As an example, we show the effect of the
clover coefficient 0.15 in Fig. 6. One could argue that it is precisely the left arc which is
physically crucial. However, we are going to insert our HF into the overlap formula, so we
end up with an exact GW fermion anyhow. In view of the convergence rate in the iterative
evaluation of this formula, the maximal deviation of the HF spectrum from the GW circle
matters, so one should not limit the attention to the left arc only.
Indeed, the systematic study of the transition to the overlap fermion on larger lattices—
which will be presented in Section 5—shows that a clover term with a small coefficient
may help a little to speed up the convergence. So we are going to include it on the 124
lattice (used in Sections 5, 6 and 7), since it is computationally cheap anyhow.
3. Polynomial approximations of the overlap formula
In four dimensions, the evaluation of the overlap formula is a notorious numeric

problem. It can only be done by some approximation of the non-analytic function involved.
In particular, one tries to approximate it by a polynomial in the relevant interval. Here we
apply Chebyshev polynomials for that purpose, as suggested in Ref. [6], and we consider
two possible ways to apply them. For a systematic comparison of various approximation
methods in the case of the Neuberger fermion, see Ref. [40]. We remark that from our
experience, different types of suitable polynomials do not lead to a very different quality
of the approximation (at fixed polynomial degree).
3.1. Approximations of the sign function
We first consider approximations of the sign function, which is done in one way or the
other in most of the literature. One writes the overlap formula as

H
Dov = µ 1 + γ5 √ = µ 1 + γ5 (H ) ,
H 2
H := γ5 (D0 − µ) = H † , (10)
and approximates the sign function

1, x > 0,
(x) = (11)
−1, x < 0,
by a polynomial. As an alternative to the Chebyshev polynomials that we are going to use
here, also an “optimal rational approximation” has been suggested [41] and it was applied,
for instance, in Refs. [7,42].
The eigenvalue distribution of H determines the interval which is relevant for the
approximation. Since this section is only meant to illustrate what the convergence rate
depends on, we consider µ = 1 for simplicity. (Later, when we study the convergence rate
in detail, optimized mass parameters µ will be inserted.)
In Fig. 7(a) we show the eigenvalue histograms for the cases of the Wilson fermion,
HW = γ5 (DW −1), and of our preferred HF on small lattices (α = 0.3, u = 0.76, v = 0.92),
(a) (b)
Fig. 7. (a) The eigenvalue histograms for typical spectra of HHF (grey) and HW (black) at β = 6; (b) the same
after rescaling so that all eigenvalues have absolute values 1.
still at β = 6, on a 44 lattice. We see that the spectrum of HHF is already sharply peaked at
±1, whereas for HW the distribution is very broad.
In order to make the polynomial approximation directly applicable for all eigenvalues,
we first have to rescale the spectra so that they are entirely confined to the interval [−1, 1],
see, for instance, Ref. [43]. The outcome of the minimal rescaling (division by the largest
absolute value of an eigenvalue) is shown in Fig. 7(b). We see that the spectrum of HHF
is not affected too much: the peaks are moved to about ±0.7, but there is still a large
gap around zero. This gap is important, because any polynomial approximation of the sign
function is plagued by its worst errors near the discontinuity at zero (remember for instance
the notorious “Gibbs phenomenon” of the Fourier expansion). On the other hand, for HW
there is (after rescaling) a considerable eigenvalue density around zero. This shows that it
requires much more effort to transform the Wilson fermion into an overlap fermion.
To demonstrate this prediction with an example, we use a linear combination of
Chebyshev polynomials with maximal degree 21 to approximate the sign function in the
overlap formula (10). We consider again a typical configuration at β = 6 on a 44 lattice.
Fig. 8 shows the resulting spectra if we start from the Wilson fermion resp. the HF, and we
see that the latter is clearly superior.
3.2. Approximation of the inverse square root
We now come to a second way to approximate the overlap formula in terms of

polynomials. The idea is not to consider the sign function any more, but to approximate
directly the inverse square root in Eq. (3). This was first applied in Ref. [6]. If we start
from an approximate GW operator D0 , then the square root is close to 1 (or in general
close to µ), and the direct expansion around this constant looks attractive. We have tested
this method in the Schwinger model and we found a very fast convergence if we start
from the 2d HF, but it slows down if we start from the Wilson fermion [14]. To get started
with the polynomial approximation, one rescales the operator A0 so that the spectrum of
A†0 A0 is all contained√in an interval [δ, 1] (0 < δ 1). The virtue of the function to be
expanded, f (x) = 1/ x, is its continuity in this interval. However, if δ becomes very
small we approach a singularity, which is a problem similar to the discontinuity of the sign
Fig. 8. The spectra of approximate overlap fermions on a 44 lattice, where the sign function is approximated by
a polynomial of degree 21. We use a configuration typical at β = 6 and start from D0 = DW (stars) resp. from
D0 = DHF (crosses).
function. Hence, the question here is how small δ is going to be. Actually the question is
the same as in Subsection 3.1 (and also the rescaling is the same), because A†0 A0 = H 2 . So
we can refer again to Fig. 7 where one just has to square the eigenvalues.
Also the comparison of spectra looks very similar to Fig. 8, so we do not repeat it but
turn to a systematic study of the convergence to an overlap fermion.
4. Convergence rate
We now look explicitly at the convergence of the overlap formula approximated by

polynomials.7 As a measure for the deviation from the final overlap fermion, we measure
the maximal deviation of any eigenvalue in the spectrum of µ1 D from the unit circle with
center 1, {z| |z − 1| = 1}. For an exact GW fermion—with respect to the standard kernel
(2)—the spectrum is situated exactly on this circle, as we pointed out in Section 1.
Fig. 9 shows this maximal deviation for a typical configuration on a 44 lattice, as a
function of the polynomial degree. We consider both, the expansion of the sign function as
well as the expansion of the inverse square root. The converge rate is clearly exponential
in the degree of the polynomial. We see further that the HF converges much faster than the
Wilson fermion. For the comparison between the two polynomial expansions one has to
take into account that the square root expansion refers to H 2 , whereas the sign function is
7 The results of this and the following sections have been summarized before in Ref. [44].
Fig. 9. The maximal radial deviation from the normalized Ginsparg–Wilson circle, evaluated from of the full
spectrum on a 44 lattice. We compare the HF at µ = 1 with the Wilson fermion at µ = 1.6, in both cases expanding
the inverse square root as well as the sign function.
expanded in H . Hence the former actually picks up a factor of 2 in the comparison of the
polynomial degree, which makes the sign function look favorable, especially for the HF.
Since the volume considered in Fig. 9 is quite small, we now proceed to a 84 lattice.
Here we cannot evaluate the full spectrum any more, but with the Arnoldi algorithm we
can identify a selected set of eigenvalues. We thus evaluated the 100 eigenvalues with the
least real part (as shown in Fig. 5 before), because they are physically most relevant, and
we measured for this subset again the maximal radial deviation. In Fig. 10 we plot this
maximal deviation, and also the mean deviation, again as a function of the polynomial
degree. For the Wilson fermion we consider two options: we take either the free hopping
parameter and optimize the mass parameter to µ = 1.6, or we take µ = 1, as in the case of
the HF, and insert the critical Wilson fermion. The goal is to make sure that we compare
the HF really to the best application of the Wilson fermion. However, as Fig. 10 shows, the
difference between the different ways to use the Wilson fermion is tiny.8
Due to the arbitrary truncation at just 100 eigenvalues the exponential behavior is not as
clean as in the case of the full spectrum. However, we see that the behavior is very similar,
and the improvement of the HF clearly persists in the same magnitude.
To make these observations more quantitative, we discuss as an example the sign
expansion on the 44 lattice, which has a very precise exponential behavior. For some
configuration typical at β = 6 we measured for the Wilson fermion the maximal deviation
dWmax
(n) = exp(−0.134n), where n is the degree of the Chebyshev polynomial, and for
the mean deviation we found dW mean (n) = 0.13 exp(−0.134n), hence the exponential factor
8 For the Wilson fermion the variation of µ and the variation of the hopping parameter are equivalent, so we
observe here that optimization of the convergence leads very close to criticality.
Fig. 10. The maximal radial deviation and the mean radial deviation from the normalized Ginsparg–Wilson circle,
evaluated from of the 100 energy eigenvalues with the lowest real parts (physical branch) on a lattice volume 84 .
We compare the inverse square root expansion for the HF and for the Wilson fermion. In the latter case we also
compare the case of the free hopping parameter and a suitable mass parameter of µ = 1.6 with the critical Wilson
fermion at µ = 1. They behave very similarly, and the HF converges much faster.
is the same. For the same configuration, the following HF deviations where obtained:
max
dHF (n) = exp(−0.737n), and dHF
mean
(n) = 0.1 exp(−0.737n).
We mention two ways how to arrive at conclusions from these numbers:
• First, we could fix some degree n which we consider affordable in a simulation. The
precision of the GW approximation compares as
max (n)
dW ∼
= e0.6n . (12)
max
dHF (n)
For realistic degrees like n = 20, . . . , 100 this ratio of the accuracies takes a very
considerable magnitude.
• On the other hand, we could fix a certain accuracy d max which we consider necessary
to trust the chiral quality of the approximated GW fermion. Then the polynomial
degrees, which are required to provide this precision, compare as
nW ∼
= 5.5. (13)
nHF
This factor may be regarded as the effective gain of the HF due to the faster
convergence, since the computational effort is essentially proportional to n.
The fluctuation of these ratios over different configurations are modest, even though
d max may vary significantly. For a systematic statistical study we now move on to larger
lattices.
5. Condition numbers
After the explicit convergence study of the last section, we now turn our attention to the
condition numbers of the operators A†0 A0 , which are crucial for the convergence rate. This
allows us to proceed to larger lattices of size 124 , still at β = 6. We first show a history
for the condition numbers for the HF and the Wilson fermion in Fig. 11 (on top). Here we
adapted the parameters so that they are optimal on the larger lattice. For the HF the new
set of optimal parameters reads
u = 0.75, v = 0.99, α = 0.87, c = 0.01, µ = 1.3, (14)

where c is the coefficient for the clover term (without additional multiplication by some
hopping parameter). For the Wilson fermion (with the free hopping parameter), µ = 1.64
turned out to be optimal with respect to the condition number, though the dependence on
µ is weak—as it is also the case for locality, see Section 6. Hence the condition numbers
Fig. 11. On top: a history of the condition numbers of A†0 A0 at β = 6 on a 124 lattice. Below: the corresponding
history of the upper and lower bound of the spectra. It shows that the improvement of the condition number of
the HF is essentially due to the upper bound.
Fig. 12. The precision of the overlap formula approximated by Chebyshev polynomials of the moderate degree
of 60. As a measure for the precision, we show the accuracy of the function fmax (r = 24) (cf. Section 6) for a set
of configurations at L = 12, β = 6. We plot this accuracy against the condition number, in order to illustrate their
monotonous relation, and the progress of the HF over the Wilson fermion.
at µ = 1.4—which is optimal with respect to locality [6]—differ only little from the result
in Fig. 11.
We see that the HF condition number is improved typically by one to two orders of
magnitude. Since it is defined by the ratio of the upper bound divided by the lower bound,
it is instructive to consider these bounds separately. Their histories are shown in Fig. 11
(below). They reveal that the improvement of the condition number is almost entirely an
effect due to the upper bound.
In Fig. 12 we illustrate explicitly how the condition number translates into an
accelerated convergence. As a (somewhat arbitrary) measure for the speed of convergence,
we consider Chebyshev polynomials at some moderate degree, n = 60, and measure the
deviation of the function fmax (r = 24) from the exact result (obtained from huge values of
n). (The quantity fmax (r = 24) represents the maximal correlation over the largest distance
on our lattice; an explicit definition will be given in Section 6.) A polynomial of degree 60
may be affordable in simulations, and we see that it typically approximates fmax (r = 24)
already to a high accuracy for the HF, but not for the Wilson fermion. For the latter we see
that µ = 1.64 is better for the condition number than the locality optimal mass parameter
of µ = 1.4.
However, it is important to note that most practical applications of overlap fermions are
performed such that the lowest few modes are projected out and treated separately. Then
the above polynomial evaluation concerns the rest of the eigenspace. This method helps a
lot, because often very few modes are responsible for a slow convergence. However, their
separate treatment is also tedious, because it requires a very accurate determination of the
corresponding eigenfunctions. Hence the number of modes to be projected out is usually
Fig. 13. The history of the “higher condition numbers” c2 , . . . , c20 (defined in Eq. (15)) for the operator A†0 A0 ,
built from the HF (at µ = 1.3) and from the Wilson fermion (at µ = 1.64).
not more than about 15; beyond that the spectrum becomes quite dense, so projecting out
further eigenvalues raises the remaining lower bound only very little.
To do justice to this situation, we introduce “higher condition numbers” ck , which are
defined by the ratio
largest eigenvalue
ck = . (15)
kth eigenvalue from below
If one projects out 15 eigenvalues, for example, then c16 is relevant for the convergence
of the polynomial evaluation. In Fig. 13 we show the histories for the higher conditions
numbers c2 , . . . , c20 , for the HF and the Wilson fermion. This plot confirms that around
k = 15 the eigenvalue density is large already, and beyond they become even more dense.
Hence it is hardly motivated to increase k much further. We also see that these histories
are much smoother compared to c1 (which was shown in Fig. 11), so here we are able
to take sensible mean values. The results are shown in Fig. 14, and we see that in the
relevant regime the improvement factor for the HF amounts to about 25. The convergence
rate behaves like the square root of the relevant condition number.9 Hence we gain a factor
of ≈ 5. This is amazingly consistent with the explicit result obtained on the small lattices
(but from the full spectrum) in Section 4.
9 Note that the relevant condition number corresponds to 1/δ 2 in the notation of Section 3.
Fig. 14. The expectation values for the higher condition numbers ck (k = 2, . . . , 20) (defined in Eq. (15)) for the
operator A†0 A0 , built from the HF and from the Wilson fermion. In the most popular regime (around k = 15) the
HF gains a factor of ≈ 25 over the Wilson fermion. For the latter we also confirm that µ = 1.64 is a little better
for the condition number than the locality optimal parameter µ = 1.4.
6. Locality
Since there is occasionally some confusion about the term “locality” of a lattice action,
we refer to the definition that the lattice Dirac operator decays at least exponentially in
the distance between ψ̄ and ψ. This is the property which is crucial for providing a safe
continuum limit (since the decay width is fixed in lattice units).
It was conjectured [5]—and later proved [45]—that Ginsparg–Wilson fermions cannot
be “ultralocal”, not even in the free case. This means that their couplings cannot drop to
zero beyond a finite number of lattice spacings. However, locality in the above sense was
shown for the free perfect fermion [15] and for the Neuberger fermion [4] to hold. In Ref.
[5] it was shown that the truncated perfect free HF leads to an overlap-HF, which is more
local than the Neuberger fermion (it has a faster exponential decay).
In the presence of gauge interactions, locality can be proved for smooth configurations,
either by assuming a small upper limit for the deviation of any plaquette variable from 1
[6,46], or by assuming that the eigenvalues of A†0 A0 do not cluster densely in the vicinity of
zero [6]. For realistic configurations, the exponential decay was observed statistically for
the Neuberger fermion in quenched QCD at β = 6 [6]. This property may collapse at much
stronger coupling, but at some point the overlap formula is not applicable any more also
for other reasons, as we discussed in Section 2. The statistical demonstration was done by
showing that the “maximal correlation” between any two lattice sites, separated by a taxi
driver distance r, decays exponentially in r. More precisely, the expectation value of the
Fig. 15. Comparison of the degree of locality for the Neuberger fermion and for the overlap HF. We see that the
latter is clearly more local; the exponent differs by more than a factor of 2.
function

4

fmax (r) := max ψ(y) |xµ − yµ | = r (16)
x, y
µ=1
has to decay exponentially, if a unit source is located at the (arbitrary) site x. We probed 6
sites x for each configuration, then we went through all sites y to determine the maxima.
The 6 options for x were sufficient to stabilize the function fmax (r). (Remember that this
function was mentioned before in Section 5.)
As we explained in the introduction, the property (4) suggests that the higher degree of
locality of the overlap-HF may also persist in the presence of gauge interactions. In fact,
in the Schwinger model a comparison of the function fmax (r) confirmed this conjecture
[14]. Here we extend this comparison to QCD on a periodic 124 lattice, which is the size
that was also used in Ref. [6]. The use of the taxi driver metrics allows us to proceed to
maximal distance 24, and the exponential decay is clearly visible.10 Our result is shown in
Fig. 15. For the Neuberger fermion it agrees well with the result of Ref. [6] (which was also
obtained at β = 6), although we used a different mass parameter. We inserted µ = 1.64
which was optimal for the condition number of the Neuberger fermion (cf. Section 5),
whereas Ref. [6] used µ = 1.4, which is slightly better for the locality—because it was
10 Of course, the decay is also exponential with respect to the Euclidean distance. There the decay is less
smooth, but the asymptotic behavior can be extracted as 0.04 exp(−|x − y|) for the Neuberger fermion, and as
0.7 exp(−2|x − y|) for the overlap HF.
optimized with this respect—but the difference is really small. Also for the overlap-HF we
used the parameter which is optimal for the condition number, in that case µ = 1.3.
We see that the overlap-HF is by far more local. To be explicit, the asymptotic
decay of fmax (r) behaves like 0.017 exp(−0.45r) for the Wilson fermion, and like
0.017 exp(−0.93r) for the HF. This observation suggests that the range of applicability
of the overlap formula extends up to stronger coupling for the overlap-HF.
Interestingly, the very clean exponential decay sets in only in the presence of
interactions. For free fermions, the decay is faster, of course, but the exponential behavior
is not as neat as in Fig. 15.
On the technical side we mention that the convergence of fmax (r) at long distances
requires a very precise evaluation of the inverse square root. This motivated the use of
fmax (r = 24) as a measure for the precision of the approximations to the overlap formula,
see Section 5.
7. Approximate rotation invariance
As a last comparison between the Neuberger fermion and the overlap HF, we consider
the extent of the violations of rotation invariance. To quantify this property, we introduce
the function fmax (|x − y|), which corresponds to definition (16) in Euclidean metrics, as
Fig. 16. Comparison of the violations of the rotation invariance for the Neuberger fermion and for the overlap-HF.
We see that the latter approximates rotation invariance much better. In both cases, the violations of rotation
invariance decay exponentially with the distance, but for the overlap HF the exponent of the decay is almost
doubled.
well as the corresponding function fmin (|x − y|),

4 1/2

fmax |x − y| := max ψ(y) (xµ − yµ ) 2
= |x − y| ,
x,y
µ=1

4 1/2

fmin |x − y| := min ψ(y) (xµ − yµ ) 2
= |x − y| . (17)
x,y
µ=1
We put a unit source on one site x and probe all other sites y for a fixed Euclidean distance.
For each distance |x − y| that occurs, we determine the difference

fmax |x − y| − fmin |x − y|
which represents a measure for the violation of rotation invariance. Fig. 16 shows that
this difference decays exponentially as a function of the Euclidean distance. (Again we
present data from a periodic 124 lattice at β = 6.) For the Neuberger fermion (at µ = 1.64),
the asymptotic decay amounts to 0.065 exp(−0.11 |x − y|). The corresponding asymptotic
decay for the overlap HF (at µ = 1.3) behaves as exp(−2.10 |x − y|).
8. Conclusions
We have constructed a HF for QCD, which approximates the GWR at β = 6. It involves

the couplings of the truncated perfect free fermion, along with four more parameters which
amplify certain links and add fat links as well as a clover term. This HF is designed to be
inserted into the overlap formula, which renders its chirality exact. Since we start off in
the right vicinity, the convergence under polynomial evaluations of the overlap formula is
accelerated compared to the standard Neuberger fermion, which uses the Wilson fermion as
point of departure. We observed consistently over lattice volumes 44 up to 124 an improved
convergence rate the overlap-HF, which allows us to obtain the same chiral accuracy as the
Neuberger fermion if the polynomial degree is reduced by a factor 5. This implies a
correspondingly reduced computational effort.
Our experience with the numeric treatment of the HF is based on a simple implemen-
tation [12], which first computes (hierarchically) a look-up table for all the “hyperlink
variables” across (spatial) diagonals inside the unit hypercube. This increases the number
of SU(3) matrices per site to be stored by a factor of 10 compared to the clover Wilson
fermion. We further estimate that the computational overhead amounts to a factor ≈ 15.
Hence after the gain in the convergence rate we are left with an computational overhead of
a factor of ≈ 3.
However, we should emphasize that the algorithms for the overlap-HF are not optimized
carefully yet, so there may still be space for a further gain due to a better implementation.
Moreover, we expect a number of further virtues of the overlap-HF compared to the
Neuberger fermion, which could well compensate this remaining overhead. We have shown
that locality is clearly improved—the exponent of the correlation decay (in the distance
between ψ̄ and ψ) is doubled—which makes the overlap formula applicable up to stronger
coupling. We have also shown that the overlap-HF approximates the rotation invariance
very well, in contrast to the Neuberger fermion.
However, the really crucial question is if our expectation for an improved scaling can
also be confirmed. If the overlap-HF allows for the use of a somewhat larger physical lattice
spacing, then the remaining overhead could be more than compensated. This hope is based
in particular on the elements of an truncated perfect actions, which are incorporated in our
HF construction. Indeed, a strongly improved scaling was observed for free fermions and
in the 2-flavor Schwinger model [14]. We also run some tests for the meson dispersions in
QCD, but it turned out that the dispersions are quite noisy, in particular for the overlap-HF.
Hence we postpone the delicate question of scaling for further investigation. Interestingly,
the phenomenon of more noise is also known from O(a) improved Wilson fermions with
a clover term. As a general property, GW fermions are also O(a) improved. Hence it is
conceivable that the overlap HF has an even better scaling behavior than the original HF,
since the overlap formula removes the O(a) artifacts.
As long as GW fermions can only be applied in the quenched approximation, the
connection with chiral perturbation theory [47] seems most attractive. However, the
ultimate goal is the use of dynamical Ginsparg–Wilson fermions, which appears as a
tremendous task. Hence it is worthwhile studying the optimal access carefully. We hope
that the current work contributes to this optimization.
Acknowledgements
I am very much indebted to Ivan Hip, who made important contributions to this work. I
also thank him, as well as David Adams, Norbert Eicker, Philippe de Forcrand, Karl Jansen,
Waseem Kamleh, Thomas Lippert, Martin Lüscher, Klaus Schilling, Rainer Sommer and
Urs Wenger for useful discussions. The computations for this work were performed on the
NICSE machines at the Forschungszentrum Jülich.
References
[1] H.B. Nielsen, M. Ninomiya, Phys. Lett. B 105 (1981) 219;

H.B. Nielsen, M. Ninomiya, Nucl. Phys. B 185 (1981) 20.
[2] P.H. Ginsparg, K.G. Wilson, Phys. Rev. D 25 (1982) 2649.
[3] H. Neuberger, Phys. Lett. B 417 (1998) 141;
H. Neuberger, Phys. Lett. B 427 (1998) 353.
[4] M. Lüscher, Phys. Lett. B 428 (1998) 342.
[5] W. Bietenholz, Eur. Phys. J. C 6 (1999) 5837.
[6] P. Hernández, K. Jansen, M. Lüscher, Nucl. Phys. B 552 (1999) 363.
[7] R.G. Edwards, U.M. Heller, R. Narayanan, Phys. Lett. B 540 (1999) 457;
R.G. Edwards, U.M. Heller, R. Narayanan, Phys. Rev. D 59 (1999) 094510, hep-lat/9905028.
[8] P. Hernández, K. Jansen, L. Lellouch, in: A. Frommer, Th. Lippert, B. Medeke, K. Schilling (Eds.),
Numerical Challenges in Lattice Quantum Chromodynamics, Springer, 2000, p. 29, hep-lat/0001008.
[9] Y. Shamir, Phys. Rev. D 62 (2000) 054513.
[10] P. Hasenfratz, V. Laliena, F. Niedermayer, Phys. Lett. B 427 (1998) 125;
P. Hasenfratz, Nucl. Phys. B 525 (1998) 401.
[11] W. Bietenholz, R. Brower, S. Chandrasekharan, U.-J. Wiese, Nucl. Phys. B (Proc. Suppl.) 53 (1997) 921.
[12] W. Bietenholz, N. Eicker, A. Frommer, Th. Lippert, B. Medeke, K. Schilling, G. Weuffen, Comput. Phys.
Commun. 119 (1999) 1.
[13] K. Orginos, et al., Nucl. Phys. B (Proc. Suppl.) 63 (1998) 904.
[14] W. Bietenholz, I. Hip, Nucl. Phys. B 570 (2000) 423;
W. Bietenholz, I. Hip, Nucl. Phys. (Proc. Suppl.) 83–84 (2000) 600.
[15] W. Bietenholz, U.-J. Wiese, Nucl. Phys. B 464 (1996) 319.
[16] S. Kato, N. Nakamura, T. Suzuki, S. Kitahara, Nucl. Phys. B 520 (1998) 323;
S. Fujimoto, S. Kato, T. Suzuki, Phys. Lett. B 476 (2000) 437;
M.N. Chernodub, S. Fujimoto, S. Kato, M. Murata, M.I. Polikarpov, T. Suzuki, Phys. Rev. D 62 (2000)
094506;
M.N. Chernodub, K. Ishiguro, T. Suzuki, hep-lat/0204003.
[17] W. Bietenholz, R. Brower, S. Chandrasekharan, U.-J. Wiese, Phys. Phys. B 407 (1997) 283.
[18] W. Bietenholz, Mod. Phys. Lett. A 14 (1999) 51.
[19] W. Bietenholz, U.-J. Wiese, Phys. Lett. B 378 (1996) 222;
W. Bietenholz, U.-J. Wiese, Nucl. Phys. B (Proc. Suppl.) 47 (1996) 575.
[20] W. Bietenholz, T. Struckmann, Int. J. Mod. Phys. C 10 (1999) 531.
[21] P. Hasenfratz, F. Niedermayer, Nucl. Phys. B 414 (1994) 785;
F. Niedermayer, P. Rüfenacht, U. Wenger, Nucl. Phys. B 597 (2001) 413;
P. Rüfenacht, U. Wenger, Nucl. Phys. B 616 (2001) 163.
[22] C.B. Lang, T.K. Pany, Nucl. Phys. B 513 (1998) 645;
F. Farchioni, V. Laliena, Phys. Rev. D 58 (1998) 054501.
[23] P. Hasenfratz, S. Hauswirth, K. Holland, T. Jörg, F. Niedermayer, U. Wenger, Nucl. Phys. B (Proc. Suppl.) 94
(2001) 627;
P. Hasenfratz, S. Hauswirth, K. Holland, T. Jörg, F. Niedermayer, Nucl. Phys. B (Proc. Suppl.) 106 (2002)
751;
P. Hasenfratz, S. Hauswirth, K. Holland, T. Jörg, F. Niedermayer, Nucl. Phys. B (Proc. Suppl.) 106 (2002)
799, hep-lat/0205010;
S. Hauswirth, Ph.D. thesis, hep-lat/0204015.
[24] W. Bietenholz, U.-J. Wiese, Phys. Lett. B 426 (1998) 114;
W. Bietenholz, Nucl. Phys. A 642 (1998) 275c.
[25] W. Bietenholz, in: X.-Q. Luo, E.B. Gregory (Eds.), Proceedings of the International Workshop on Non-
Perturbative Methods and Lattice QCD, Guangzhou (China), World Scientific, Singapore, 2001, p. 3, hep-
lat/0007017;
W. Bietenholz, N. Eicker, I. Hip, K. Schilling, Nucl. Phys. B (Proc. Suppl.) 94 (2001) 603.
[26] T. DeGrand, Phys. Rev. D 63 (2001) 034503.
[27] P. Hernández, K. Jansen, L. Lellouch, Phys. Lett. B 469 (1999) 198.
[28] T. DeGrand, A. Hasenfratz, Phys. Rev. D 64 (2001) 034512.
[29] A. Boriçi, Phys. Lett. B 453 (1999) 46.
[30] W. Kamleh, D. Adams, D. Leinweber, A. Williams, Phys. Rev. D 66 (2002) 014501.
[31] C. Gattringer, I. Hip, C.B. Lang, Nucl. Phys. B 597 (2001) 451;
For a corresponding work in d = 2, see: C. Gattringer, I. Hip, Phys. Lett. B 480 (2000) 112.
[32] S.J. Dong, F.X. Lee, K.F. Liu, J.B. Zhang, Phys. Rev. Lett. 85 (2000) 5051;
S.J. Dong, F.X. Lee, K.F. Liu, J.B. Zhang, Nucl. Phys. B (Proc. Suppl.) 94 (2001) 752;
F.D.R. Bonnet, P.O. Bowman, D.B. Leinweber, A.G. Williams, J.B. Zhang, Phys. Rev. D 65 (2002) 114503.
[33] W. Bietenholz, in: V. Mitrjushkin, G. Schierholz (Eds.), Lattice Fermions and the Structure of the Vacuum,
Kluwer Academic, 2000, p. 77, hep-lat/0001001.
[34] F. Farchioni, I. Hip, C.B. Lang, Phys. Lett. B 443 (1998) 214.
[35] F. Farchioni, I. Hip, C.B. Lang, M. Wohlgenannt, Nucl. Phys. B 544 (1999) 364.
[36] N. Cundy, M. Teper, U. Wenger, hep-lat/0203030.
[37] G.P. Lepage, P. Mackenzie, Phys. Rev. D 48 (1993) 2250.
[38] C. Gattringer, I. Hip, Nucl. Phys. B 541 (1999) 305.
[39] M. Albanese, et al., Comput. Phys. Commun. 45 (1987) 345.
[40] J. van den Eshof, A. Frommer, Th. Lippert, K. Schilling, H. van der Vorst, hep-lat/0202025;
J. van den Eshof, A. Frommer, Th. Lippert, K. Schilling, H. van der Vorst, Nucl. Phys. B (Proc. Suppl.) 106
(2002) 1070.
[41] H. Neuberger, Phys. Rev. Lett. 81 (1998) 4060.

[42] L. Giusti, C. Hoelbling, C. Rebbi, Phys. Rev. D 64 (2001) 114508.
[43] W.H. Press, S.A. Teukolsky, W.T. Vetterling, B.P. Flannery, Numerical Recipes, 2nd Edition, Cambridge
Univ. Press, Cambridge UK, 1992.
[44] W. Bietenholz, I. Hip, K. Schilling, Nucl. Phys. B (Proc. Suppl.) 106 (2002) 829.
[45] I. Horvath, Phys. Rev. Lett. 81 (1998) 4063;
W. Bietenholz, hep-lat/9901005;
I. Horvath, C. Balwe, R. Mendris, Nucl. Phys. B 599 (2001) 283.
[46] H. Neuberger, Phys. Rev. D 61 (2000) 085015.
[47] S. Chandrasekharan, Phys. Rev. D 60 (1999) 074503;
P.H. Damgaard, M.C. Diamantini, P. Hernández, K. Jansen, Nucl. Phys. B 629 (2002) 445.
Off-shell crosscap state and orientifold planes

with background dilatons ✩
Hiroshi Itoyama a , Shin Nakamura b,∗
a Department of Mathematics and Physics, Graduate School of Science, Osaka City University,
3-3-138 Sugimoto, Sumiyoshi-ku, Osaka 558-8585, Japan
b Theoretical Physics Laboratory, RIKEN (The Institute of Physical and Chemical Research), 2-1 Hirosawa,
Wako, Saitama 351-0198, Japan
Received 31 May 2002; accepted 29 August 2002
Abstract
We show that a non-trivial dilaton condensation alters the dimensions of orientifold planes. An
off-shell crosscap state which naturally interpolates between the usual on-shell crosscap states and
their T-duals plays an important role in the analysis. We present an explicit representation of the off-
shell crosscap state on an RP2 worldsheet in the gauge in which the worldsheet curvature in the bulk
of the fundamental region of the RP2 vanishes. We show that the non-trivial dilaton condensation
reproduces the correct descent relation among orientifold plane tensions.
1. Introduction
Orientifold planes (O-planes) as well as D-branes are important objects to reveal non-
perturbative effects in unoriented string theory. Recent studies on open-string tachyon
condensation have shown that background fields can control the dimensions of D-branes.1
The relationship between the dimensions of the D-branes and the configuration of the
background open-string fields is easily understood from the viewpoint of the worldsheet;
the background fields on the boundaries can alter the boundary conditions on the
✩
This work is supported in part by the Grant-in-Aid for Scientific Research (14540264) from the Ministry of
Education, Science and Culture, Japan.
* Corresponding author.
E-mail addresses: itoyama@sci.osaka-cu.ac.jp (H. Itoyama), nakashin@postman.riken.go.jp
(S. Nakamura).
1 See, for example, Ref. [1] for recent reviews on the related topics.
PII: S 0 5 5 0 - 3 2 1 3 ( 0 2 ) 0 0 7 9 1 - 5
H. Itoyama, S. Nakamura / Nuclear Physics B 644 (2002) 248–262 249
worldsheet. On the other hand, the relationship between the properties of O-planes and
background fields has not been understood well as yet.
In the present work, we investigate the connection between the dimensions of O-pla-
nes and the configuration of the background dilaton field, in unoriented bosonic string
theory. An O-plane is represented as a crosscap on an RP2 worldsheet while a D-brane is
represented as a boundary on a disc. Therefore we consider an RP2 worldsheet in the
presence of the background dilatons. The reason we consider dilatons is based on the
following property. Dilatons couple to the worldsheet curvature and their contribution can
be put on any part of the worldsheet in general. However, we show that the contribution
of the background dilatons localizes on the crosscap if we choose the gauge in which the
worldsheet curvature vanishes in the bulk of the fundamental region of the RP2 . This choice
of gauge is nice; the bulk part of the RP2 becomes free and all the interactions from the
background dilatons appear only on the crosscap. Therefore the effects of the background
dilatons can be translated into the modification of the crosscap conditions.
We introduce a new crosscap state which we refer to as an off-shell crosscap state
in order to analyze the properties of the RP2 worldsheet. Of course, O-planes couple
to only closed strings and we do not have open-string modes on the RP2 worldsheet.
Useful tools to analyze the properties of worldsheets in terms of closed-string modes are
boundary states and crosscap states [2–5]. Usually, boundary states and crosscap states
belong to the closed-string sector which preserves conformal invariance on the worldsheet.
Extensions of these boundary states which do not in general maintain conformal invariance
have been proposed recently [6,7]. (See also [8,9].) We call them off-shell boundary
states in the present article. The off-shell boundary state interpolates between the usual
on-shell boundary states and their T-duals. This state is defined on a disc worldsheet
with quadratic boundary interactions. The dimension of the corresponding D-brane is
controlled by the coupling constants of these interactions. Boundary string field theory
(BSFT) [10,11] states that the coupling constants also parameterize the configuration of
open-string tachyons [12–17]. Calculating the disc partition function by using these states
enables us to obtain the descent relation among the D-brane tensions if we take appropriate
on-shell limits of the partition functions.
An attempt to apply the foregoing idea to crosscap states is given in Ref. [18], in which
the definition of the off-shell crosscap state which interpolates the usual on-shell crosscap
states and their T-duals has been proposed. We present an explicit representation of the off-
shell crosscap state in the present work. We define the off-shell crosscap state on an RP2
worldsheet in the presence of quadratic interactions on the crosscap. The dimension of the
corresponding O-plane is controlled by the coupling constants of these interactions. The
physical meaning of the quadratic interactions on the crosscap is the background dilaton
field of quadratic configuration. The behaviour of the off-shell crosscap state shows that
the background dilatons control the dimension of the O-plane.
The off-shell crosscap state is a useful tool to obtain correlation functions on the RP2
worldsheet in the presence of the quadratic dilatons on the crosscap. The exact partition
function of the RP2 worldsheet which is considered to be proportional to the O-plane
tension can be calculated exactly by using the two-point function. We show that the
condensation of the dilatons of quadratic profile reproduces the correct descent relation
250 H. Itoyama, S. Nakamura / Nuclear Physics B 644 (2002) 248–262
among O-plane tensions by taking appropriate on-shell limits of the partition function.
Note that we have not attempted to find the dynamical origin of the dilaton condensation.
The paper is organized as follows. In Section 2, we construct the RP2 worldsheet on
a complex plane and we show that the contribution of the background dilatons survives
on the crosscap alone if we choose the gauge in which the worldsheet curvature vanishes
in the bulk of the fundamental region of the RP2 . In Section 3 we define the off-shell
crosscap state and we obtain its explicit representation. We find that the behaviour of this
state signifies that a non-trivial dilaton condensation alters the dimensions of O-planes.
We calculate the partition function of the RP2 by using the off-shell crosscap state. In
Section 4, we verify that the non-trivial dilaton condensation reproduces the correct descent
relation of O-plane tensions, by taking appropriate limits of the partition function of the
RP2 worldsheet. In the final section, we summarize the results of this study. We make
comments on some relationship between the RP2 worldsheet and the supersymmetric disc
worldsheet which appears in the analysis of D D̄ systems in Appendix A.
2. RP2 worldsheet with background dilatons
Let us consider an RP2 worldsheet with background dilaton field. We show, in this
section, that the contribution of the background dilatons localizes on the crosscap if we
choose the gauge in which the worldsheet curvature in the bulk of the fundamental region
of the RP2 vanishes.
RP2 is a non-orientable Riemann surface of Euler number one with no hole, no boundary
and one crosscap. We construct the RP2 worldsheet on a complex z-plane by using an
involution where we identify z and − 1z̄ on the complex plane. We choose the fundamental
region Σ to be {z = reiσ | 0 r < 1, 0 σ < 2π} ∪ {z = reiσ | r = 1, 0 σ < π}. The
crosscap C, the non-trivial closed loop of the RP2 worldsheet, is represented as half of unit
circle {z = reiσ | r = 1, 0 σ < π} in this case.
To begin with, we set the metric inside the unit circle (|z| 1) on the complex plane as
1
hzz = hz̄z̄ = 0, hz̄z = hzz̄ = . (2.1)
2
The metric outside the unit circle (|z | 1) is obtained by the involution z = − 1z̄ , z̄ = − 1z
as
∂ z̄ ∂z 1 1
hz z̄ = hz̄z = 2 hz̄z = 4 hz̄z , (2.2)
∂z ∂ z̄ (z z̄ ) r
where r 2 = z z̄ for r 1. Therefore the metric on the entire complex plane can be written
as
1
hzz = hz̄z̄ = 0, hz̄z = hzz̄ = eρ , (2.3)
2
where
ρ(r) = (−4 ln r)θ (r − 1). (2.4)

Next, we rewrite the metric in the polar coordinates (r, σ ) as
gab = ĝab eρ , (2.5)

where ĝrr = 1, ĝσ σ = r2,
ĝσ r = ĝrσ = 0. The worldsheet curvature R is then given by

√ ˆ 1
g R = − ĝ ∇ ρ(r) = −r ∂r + ∂r ρ(r)
2 2
r

= 4 δ (r − 1)r ln r + (2 + ln r)δ(r − 1) . (2.6)
√
Let us calculate the integral Σ() dr dσ g RΦ(r, σ ). We define the integral region Σ()
as {z = reiσ | 0 r 1 + , 0 σ < π} ∪ {z = reiσ | 0 r 1 − , π σ < 2π}, where
is a positive real number. Σ() becomes the fundamental region Σ of the RP2 after we
take the limit → 0. We obtain

√
dr dσ g RΦ(r, σ )
Σ()
π 1+ 2π 1−
√ √
= dσ dr g RΦ(r, σ ) + dσ dr g RΦ(r, σ )
0 0 π 0
π 1+

=4 dσ dr δ (r − 1)r ln r + (2 + ln r)δ(r − 1) Φ(r, σ )
0 0
π
d

=4 dσ − r ln rΦ(r, σ ) r=1 + 2Φ(1, σ )
dr
0
π
=4 dσ Φ(1, σ ), (2.7)
0
which yields
π
1 √ 1
dr dσ g RΦ(r, σ ) = dσ Φ(1, σ ). (2.8)
4π π
Σ 0
Note that (2.8) gives the correct Euler number of the RP2 (which is one) if we set Φ = 1.
Therefore the contribution of the background dilaton concentrates on the crosscap with the
above gauge choice.
3. Off-shell crosscap state
In this section, we define the off-shell crosscap state and obtain its explicit represen-
tation. We apply the off-shell crosscap state to calculate the correlation functions and the
partition function of the RP2 worldsheet with quadratic background dilaton field. We find
that the behaviour of the off-shell crosscap state signifies that a non-trivial dilaton conden-
sation alters the dimensions of O-planes.
3.1. Off-shell crosscap conditions
Let us consider an RP2 worldsheet Σ with the following action:

1 µ¯ 1
I= d z ∂X ∂Xµ +
2
dσ Φ(σ ), (3.1)
2πα π
Σ C
1 µ 2
26
Φ(σ ) = a + uµ X (σ ) , (3.2)
2α
µ=1
where the interaction Φ(σ ) is inserted only on crosscap C. Note that the worldsheet action
is free in the “bulk” region {z = reiσ | 0 r < 1, 0 σ < 2π}. A closed string propagates
freely in the “bulk” toward the crosscap, from the viewpoint of the closed-string channel.
In this sense, Xµ in the “bulk” can be expanded as

iα µ α µ z−n −n
µ µ z̄
X z, z̄ = X0 −
µ
p ln zz̄ + i αn + α̃n . (3.3)
2 2 n n
n=0
The existence of the crosscap, however, makes constraints on the oscillation of the
closed string in the neighborhood of the crosscap (in the region r → 1). For example,
the constraints when we have no interaction on C are given in Ref. [4] as
µ
X z, z̄ − Xµ −1/z̄, −1/z r→1 = 0,

X˙µ z, z̄ + X˙µ −1/z̄, −1/z = 0,
r→1
(3.4)
where

X˙µ z, z̄ ≡ w∂w + w̄ ∂¯w̄ Xµ w, w̄ w=z,w̄=z̄ ,

X˙µ −1/z̄, −1/z ≡ w∂w + w̄ ∂¯w̄ Xµ w, w̄ w=−1/z̄,w̄=−1/z . (3.5)
Note that (3.4) are equivalent to the following constraints:

K0 z, z̄ r→1 = 0,

z∂z + z̄∂¯z̄ K0 z, z̄ = 0,
r→1
(3.6)
where

K0 z, z̄ ≡ Ẋµ z, z̄ + Ẋµ −1/z̄, −1/z . (3.7)
These conditions are rewritten in terms of closed-string modes at r → 1 as
µ
αnµ + (−1)n α̃−n = 0, pµ = 0. (3.8)
The conditions (3.4), (3.6) or (3.8) are referred to as (on-shell) crosscap conditions.
The aim of this subsection is to extend the on-shell crosscap conditions into the case
uµ = 0. We should find, in other words, the constraints on the closed-string modes in the
neighborhood of C in the presence of interaction Φ. We call these constraints off-shell
crosscap conditions. We assert that the off-shell crosscap conditions can be written as

K z, z̄ r→1 = 0,

z∂z + z̄∂¯z̄ K z, z̄ r→1 = 0, (3.9)
where

K z, z̄ ≡ Ẋµ z, z̄ + Ẋµ −1/z̄, −1/z + uµ Xµ z, z̄ + Xµ −1/z̄, −1/z

= w∂w + w̄ ∂¯w̄ Xµ w, w̄ + uµ Xµ w, w̄ w=z,w̄=z̄

+ w∂w + w̄ ∂¯w̄ Xµ w, w̄ + uµ Xµ w, w̄ . (3.10)
w=−1/z̄,w̄=−1/z
The right-hand side of (3.10) indicates the meaning of the off-shell crosscap conditions;
(3.9) are the conditions so that the Xµ in the neighborhood of C, as well as its image by
the involution, connects smoothly with the Xµ on C which obeys

z∂z + z̄∂¯z̄ Xµ z, z̄ + uµ Xµ z, z̄ C = 0, (3.11)
given by varying the worldsheet action (3.1).2

The off-shell crosscap conditions (3.9) are rewritten in terms of closed-string modes as
µ uµ µ µ
− αnµ + (−1)n α̃−n + αn − (−1)n α̃−n = 0,
n
−iα pµ + uµ X0 = 0,
µ
(3.12)
where we do not sum over µ. We can easily check that the conditions (3.12) gives the on-
shell crosscap conditions (3.8) in the limit uµ → 0. We also note that (3.12), in the limit
uµ → ∞, becomes equivalent to the T-dual of (3.8),
µ µ
αnµ − (−1)n α̃−n = 0, X0 = 0, (3.13)
which are rewritten as the T-dual of (3.4):
µ
X z, z̄ + Xµ −1/z̄, −1/z r→1 = 0,

X˙µ z, z̄ − X˙µ −1/z̄, −1/z = 0.
r→1
(3.14)
Note that the conditions (3.9) in the limit uµ → ∞ is (3.14) itself. Therefore the off-
shell crosscap conditions (3.9) or (3.12) naturally interpolate between on-shell crosscap
conditions and their T-duals. The coupling constant uµ , which is a parameter of the
configuration of the background dilaton field, controls the dimension of the corresponding
O-plane.
2 Although RP2 has no boundary, (z∂ + z̄∂¯ )Xµ (z, z̄) which comes from the total derivative survives only
z z̄
on the crosscap due to the involution.
3.2. Off-shell crosscap state and partition function
We define off-shell crosscap state C(u)| using the off-shell crosscap conditions as

µ uµ µ
n µ
C(u) − αn + (−1) α̃−n + n µ
α − (−1) α̃−n = 0,
n n
µ
C(u) −iα pµ + uµ X0 = 0. (3.15)
The explicit form of C(u)| is given as

C(u) = 0|C(u),
∞
1 µ
C(u) ≡ exp − X0 Aµν X0ν exp µ (m) ν
α̃m Cµν αm , (3.16)
2
m=1
where
1
Aµν ≡ A(uµ )δµν = uµ δµν ,
α
(−1)m m − uµ
(m)
Cµν ≡ C (m) (uµ )δµν = − δµν . (3.17)
m m + uµ
We can easily check that this off-shell crosscap state becomes (the T-dual of) the usual
on-shell crosscap state if we take the limit uµ → 0 (uµ → ∞). Therefore the off-shell
crosscap state naturally interpolates between the crosscap state for a higher-dimensional
O-plane and that for a lower-dimensional O-plane.
Next, we show that the off-shell crosscap state is a useful tool to evaluate the quantities
on the RP2 worldsheet. For example, we can calculate the Green’s function and the
partition function on the RP2 worldsheet in the presence of interaction Φ(σ ) on the
crosscap. Let us consider one-dimensional target space and omit the superscript µ of X
and u for simplicity. The Green’s function for this case is given as
C(u)|X(z, z̄)X(w, w̄)|0
G(z, w) =
C(u)|0
α α 2
= − ln|z − w|2 − ln 1 + zw̄
2 2
α ∞
1 k k
+ − α u −zw̄ + −z̄w . (3.18)
u k(k + u)
k=1

In the case z = eiσ and w = eiσ , we obtain
α 2 α 2
G eiσ , eiσ = − ln 1 − ei(σ −σ ) − ln 1 + ei(σ −σ )
2 2
∞
α (−1) k
+ − α u eik(σ −σ ) + e−ik(σ −σ ) . (3.19)
u k(k + u)
k=1
We define composite operator X2 (σ ) as shown in Ref. [11]:

X2 (σ ) ≡ lim X(σ )X(σ + ) − f () ,
→0
α 2
f () = − ln 1 − ei + (const). (3.20)
2
We write the constant in (3.20) as α ln q by using a positive constant q. The value for q
is ambiguous at this stage and depends on the renormalization scheme. We will determine
the value for q later. We can then obtain
∞
α α (−1)k
X2 (σ ) = − ln 22 + − 2α u − α ln q
2 u k(k + u)
k=1

α u
= −α ln(2q) − − 2α Ψ − Ψ (u) , (3.21)
u 2
where
d
du -(u)
Ψ (u) ≡ , (3.22)
-(u)
is a polygamma function for which we have used the relationship
∞

(−1)k 1 u
u = +Ψ − Ψ (u). (3.23)
k(k + u) u 2
k=1
We can next calculate the partition function of the RP2 worldsheet by using the
following relationship:
π
d 1 1
ln Z(u) = −
dσ X2 (σ ) = − X2 (σ )
du 2πα 2α
0

1 α u
= − −α ln(2q) − − 2α Ψ − Ψ (u)
2α u 2

1 d d u d
= ln(2q) + ln u + 2 2 ln - − ln -(u)
2 du du 2 du
√ u√
d 2q u -( u2 )2
= ln + const . (3.24)
du -(u)
We then obtain
√ u√
2q u -( u2 )2
Z(u) = , (3.25)
-(u)
up to the overall normalization factor. In general, the partition function for 26-dimensional
target space with the interaction (3.2) on the crosscap can be written as

26 26 √
u √ u
−a −a −a 2q µ uµ -( 2µ )2
Z(a, u) ≡ e Z(u) ≡ e Z(uµ ) = e , (3.26)
-(uµ )
µ=1 µ=1
up to the overall normalization factor. We note that the partition function (3.26) on the
RP2 has an identical representation with the partition function on the supersymmetric
disc worldsheet which have been considered in the analysis of open-string tachyon
condensation in D D̄ systems [15–17]. We present some comments on the relationship
between the RP2 worldsheet and the supersymmetric disc worldsheet in Appendix A.
4. Derivation of the descent relation among O-plane tensions
In this section, we clarify the physical meaning of the partition function calculated in the
previous section. We show that the condensation of the quadratic dilaton field reproduces
the correct descent relation among the O-plane tensions.
4.1. Sigma model approach and RP2 worldsheet
Let us consider our work from the viewpoint of the sigma model approach. The basic
idea of the sigma model approach is that the spacetime action for string fields is essentially
the renormalized partition function of the worldsheet with corresponding background
string fields. In this sense, the spacetime action S for string field λi may be given as

S(λi ) ∼ [dgab ] dXµ e−Iχ (gab ,X ;λi )
µ
=
χ
Σχ

= Zsphere (λi ) + Zdisc(λi ) + ZRP2 (λi ) + · · · , (4.1)
where Iχ is the action on the worldsheet Σχ of the Euler number χ . Leading term
Zsphere(λi ) is the renormalized partition function on the sphere. This term is of order go−2
where go is the open-string coupling constant. Renormalized partition functions Zdisc(λi )
on the disc and ZRP2 (λi ) on the RP2 are the loop correction terms of order go−1 . In principle,
Zdisc(λi ) is proportional to the tension of the corresponding D-brane and ZRP2 (λi ) is
proportional to the tension of the corresponding O-plane.
It is known, however, that the right-hand side of (4.1) does not give the correct
spacetime action; we need modification of Zsphere(λi ) and Zdisc(λi )3 in order to obtain the
correct off-shell spacetime action. These modifications are closely related to the infinite
Möbius volume of the worldsheets; we need to subtract the divergence from the Möbius
infinity [19–21]. (See also [15,22].)
The situation is different for ZRP2 (λi ) (and for the partition functions of the worldsheets
of χ 0). We should note that the Möbius group of RP2 is SO(3) whose volume is finite,
and we have no Möbius infinity from the RP2 worldsheet. Therefore it is natural to assume
that partition function ZRP2 (λi ) itself is the exact loop correction term from the RP2 graph.
We have calculated the partition function (3.26) on the 2
RP worldsheet in the presence of
quadratic background dilaton field Φ = a + (2α )−1 µ uµ Xµ2 on the crosscap. We have
3 We need not modify Z

disc (λi ) in superstring theory since the volume of the super-Möbius group of the
super-disc is finite [19].
fixed the worldsheet metric so that the “bulk” part of the fundamental region of the RP2
becomes flat and the contribution of the dilatons concentrates on the crosscap. Therefore
we have to calculate the contribution of the ghost field and the anti-ghost field on the RP2
worldsheet in order to obtain the correct overall normalization of (3.26). We write this
overall factor A and then we have the following relationship:
26 √
u √ u 2
−a 2q µ uµ -( 2µ )
ZRP2 (Φ) = Ae ≡ ZRP2 (a, u). (4.2)
-(uµ )
µ=1
The sign of A should be minus. Note that factor e−a is equal to go−1 when the dilaton is
constant.
4.2. Asymptotic behaviour of Z(u)
Let us rewrite Z(u) given in (3.25) as

√ u√
2q u -( u2 )2 4 q u/2 u
Z(u) = =√ F , (4.3)
-(u) u 2 2
where we have defined function F (x) as
4x x-(x)2
F (x) ≡ . (4.4)
2-(2x)
F (x) behaves as follows:

F (x) ∼ 1 + (2 ln 2)x + O x 2 (x → 0), (4.5)
√
F (x) ∼ πx + O x −1/2 x→∞ . (4.6)
Z(u) around u = 0 is then

1 √
Z(u) ∼ 4 √ +O u . (4.7)
u
Thus, Z(u) diverges when u approaches 0. This is an IR divergence which corresponds to
the volume of the spacetime.
On the other hand Z(u) around u = ∞ is
u/2
q π 1
Z(u) ∼ 4 +O √ . (4.8)
2 2 u
We can obtain a finite and non-zero value of Z(u) in the limit u → ∞ if and only if q = 2.
Therefore we assign the value 2 to q. In other words, we have chosen the renormalization
scheme in (3.20) so that we can obtain a finite and non-zero value of Z(u) in the limit
u → ∞. Z(u) is then
√ u√
4 u -( u2 )2 4 u
Z(u) = =√ F , (4.9)
-(u) u 2
up to the overall normalization factor and

26

−a 4 uµ
ZRP2 (a, u) = Ae √ F . (4.10)
uµ 2
µ=1
4.3. Ratio of the O-plane tensions
Let us define quantity Sp as follows:

ZRP2 (a, u) → Sp , (4.11)
where the limit is taken as
u1 , . . . , up+1 → 0, up+2 , . . . , u26 → ∞. (4.12)

We do not touch the parameter a in this subsection. According to the argument in the
previous section, Sp is equal to Vp × Tp where Vp and Tp are the volume and tension of
an Op-plane. Here, the dimension of the O-plane is p + 1 and is defined as the number of
parameters uµ which are taken to zero. We can thus write

S25 = ZRP2 (u25 → 0, ui → 0) = dx 25 V24 T25 ,
S24 = ZRP2 (u25 → ∞, ui → 0) = V24 T24 , (4.13)
where i = 25. Then

S25 Z(u25 → 0) dx 25 T25
= = , (4.14)
S24 Z(u25 → ∞) T24
where Z(u) is given in (4.9). We can rewrite Z(u25 ) as

4 u25
Z(u25 ) = √ F
u25 2

1 u25
= dx 25 e− 2α u25 (x ) √
1 25 2
4F , (4.15)
2α π 2
where x 25 is the zero-mode of X25 . In the second line, we have explicitly rewritten the
√
integral of the zero-mode part of Z(u25 ) [15,23]. (This corresponds to ∼1/ u25 .) We can
therefore obtain

1 u25 4
dx 25 e− 2α u25 (x ) √
1 25 2
lim Z(u25 ) = lim 4F = dx 25 √ · 1.
u25 →0 u25 →0
2α π 2 2α π
(4.16)
We can also obtain

4 u25 4 u25 1
lim Z(u25 ) = lim √ F = lim √ π +O √
u25 →∞ u25 →∞ u25 2 u25 →∞ u25 2 u25

π
=4 . (4.17)
2
Therefore
√
T24 2α π π √
= 4 = π α . (4.18)
T25 4 2
This is precisely the ratio of the tension of an O24-plane and that of an O25-plane. In
general, we can show in a similar manner that
Tp √ q−p
= π α . (4.19)
Tq
5. Conclusion
We have considered the relationship between the configuration of the background dila-
ton field and the dimensions of the O-planes. We showed that the contribution of the dila-
tons on the RP2 worldsheet localizes on the crosscap if we choose the gauge in which the
worldsheet curvature in the “bulk” vanishes. This feature enables us to treat dilatons quite
easily. We have proposed the off-shell crosscap state which naturally interpolates between
the usual crosscap states and their T-duals. The behaviour of the off-shell crosscap state
signifies that the non-trivial dilaton condensation alters the dimensions of O-planes. We
obtained the correlation functions and partition function on the RP2 worldsheet in the pres-
ence of the quadratic dilaton field on the crosscap. We showed that the non-trivial dilaton
condensation reproduces the correct descent relation among O-plane tensions, by taking
the on-shell limits of the partition function of the RP2 . We found the correspondence be-
tween the RP2 worldsheet and the supersymmetric disc which is presented in Appendix A.
We would like to make some comments. We have studied the effects of the non-trivial
dilaton condensation on O-planes. In order to describe the full dynamics of dilatons, we
would need open-closed string field theory.4 Non-perturbative analysis would be necessary
too. However, the present work can be a step to understand the relationship between the
condensation of string fields and the transmutations of O-planes. An extension of the
present work into the supersymmetric case is interesting, and studying the relationship
between our work and the properties of O-planes described by F-theory [25,26] is one
further direction to pursue. Studying the relationship between the configuration of the
dilatons and the O-planes in type I theory [27,28] is also tempting. It has been shown that
non-trivial configurations of dilatons constrain the spacetime positions of D-branes [29].
Effects of the non-trivial dilaton condensation on D-branes are also interesting subjects to
study. Recent studies have shown that the condensation of the closed-string tachyons in
the twisted sector can alter the topology of the orbifolded spacetime [30–33]. We expect
that the present work can also be a step towards understanding the relationship between the
topology of the spacetime and the configuration of the string fields, since the dimensions
of O-planes are closely related to the topology of the orientifolded spacetime.
4 An unoriented open-closed string field theory has been proposed in Ref. [24].
Acknowledgements
We would like to thank H. Kawai, T. Kugo, K. Murakami, T. Nakatsu, A. Tsuchiya

and T. Yokono for fruitful discussions. We are grateful to Santo seminar for providing
an opportunity for collaboration. S.N. wishes to thank M. Laidlaw, M. Rozali, and
G.W. Semenoff for discussions and their hospitality during his stay at University of British
Columbia.
Appendix A. Correspondence between the RP2 worldsheet and the supersymmetric

disc worldsheet
We comment in this appendix on the relationship between the RP2 worldsheet we

have considered and the supersymmetric disc (super-disc) worldsheet which appears in
the analysis of D D̄ systems. Let us consider a unit disc M ({z = reiσ | 0 r 1, 0
σ < 2π}) with the following worldsheet action:

1 2 µ¯ µ¯ µ¯
Isuper-disc = d z ∂X ∂Xµ + ψ ∂ψµ + ψ̃ ∂ ψ̃µ
2
4π α
M

1 µ 2 2
µ ν
ρ 1

+ dσ y X + ψ ∂µ X ψ ∂ν Xρ . (A.1)
4π α µ ∂σ
∂M
This action has been considered in the context of open-string tachyon condensation in D D̄
systems [15–17]. The partition function of the super-disc is obtained in Ref. [15] as

26
1 µ
Z(y) ∝ F y , (A.2)
yµ
µ=1
where function F (x) has been defined in (4.4).
Therefore, the partition function (4.10) of the RP2 worldsheet has an identical form to
the partition function (A.2) of the super-disc. Comparing (4.10) and (A.2), we find the
correspondence
uµ
↔ y µ. (A.3)
2
At the level of integrated operators, we find the correspondence
ORP2 ↔ Odisc , (A.4)
where

2 2
ORP2 ≡ dσ X (σ ), (A.5)
α µ
C
is an operator on the RP2 and

2 1 ν
Odisc (σ ) ≡ dσ Xµ2 (σ ) + ψ µ ∂µ Xρ ψ ∂ν Xρ (σ ) , (A.6)
α ∂σ
∂M
is an operator on the super-disc. We should also note that the descent relation among
D-brane tensions in the D D̄ system can be obtained in the same manner as that we have
shown in Section 4. The correspondence (A.4) can be explained from the calculation of
Xµ2 for the RP2 , in which we have an extra minus sign in the contributions from odd
modes.5 These contributions correspond to those of the fermions on the super-disc, while
the even modes behave like the bosonic part on the super-disc.
References
[1] K. Ohmori, A review on tachyon condensation in open string field theories, hep-th/0102085;
I.Ya. Aref’eva, D.M. Belov, A.A. Giryavets, A.S. Koshelev, P.B. Medvedev, Noncommutative field theories
and (super)string field theories, hep-th/0111208.
[2] P. Horava, Background duality of open-string models, Phys. Lett. B 231 (1989) 251;
P. Horava, Strings on world sheet orbifolds, Nucl. Phys. B 327 (1989) 461.
[3] C.G. Callan, C. Lovelace, C.R. Nappi, S.A. Yost, String loop corrections to beta functions, Nucl. Phys. B 288
(1987) 525.
[4] C.G. Callan, C. Lovelace, C.R. Nappi, S.A. Yost, Adding holes and crosscaps to the superstring, Nucl. Phys.
B 293 (1987) 83.
[5] C.G. Callan, C. Lovelace, C.R. Nappi, S.A. Yost, Loop corrections to superstring equations of motion, Nucl.
Phys. B 308 (1988) 221.
[6] A. Fujii, H. Itoyama, Some computation on g function and disc partition function and boundary string field
[7] T. Lee, Tachyon condensation, boundary state and noncommutative solitons, Phys. Rev. D 64 (2001) 106004,
hep-th/0105115.
[8] E.T. Akhmedov, M. Laidlaw, G.W. Semenoff, On a modification of the boundary state formalism in off-shell
string theory, hep-th/0106033.
[9] M. Laidlaw, G.W. Semenoff, The boundary state formalism and conformal invariance in off-shell string
[10] E. Witten, On background independent open-string field theory, Phys. Rev. D 46 (1992) 5467, hep-
th/9208027.
[11] E. Witten, Some computations in background independent off-shell string theory, Phys. Rev. D 47 (1993)
3405, hep-th/9210065.
[12] J.A. Harvey, D. Kutasov, E. Martinec, On the relevance of tachyons, hep-th/0003101.
[13] A.A. Gerasimov, S.L. Shatashvili, On exact tachyon potential in open string field theory, JHEP 0010 (2000)
034, hep-th/0009103.
[14] D. Kutasov, M. Mariño, G. Moore, Some exact results on tachyon condensation in string field theory,
JHEP 0010 (2000) 045, hep-th/0009148.
[15] D. Kutasov, M. Mariño, G. Moore, Remarks on tachyon condensation in superstring field theory, hep-
th/0010108.
[16] P. Kraus, F. Larsen, Boundary string field theory of the D D̄ system, Phys. Rev. D 63 (2001) 106004, hep-
th/0012198.
[17] T. Takayanagi, S. Terashima, T. Uesugi, Brane–antibrane action from boundary string field theory,
JHEP 0103 (2001) 019, hep-th/0012210.
[18] H. Itoyama, S. Nakamura, Extension of boundary string field theory on disc and RP2 worldsheet geometries,
hep-th/0201035.
[19] O.D. Andreev, A.A. Tseytlin, Partition function representation for the open superstring effective action:
cancellation of Möbius infinities and derivative corrections to Born–Infeld Lagrangian, Nucl. Phys. B 311
(1988) 205.
5 See the third term of the first line on the right-hand side of (3.21).
[20] A.A. Tseytlin, Möbius infinity subtraction and effective action in sigma model approach to closed string
theory, Phys. Lett. B 208 (1988) 221.
[21] A.A. Tseytlin, Conditions of Weyl invariance of two-dimensional sigma model from equations of stationarity
of ‘central charge’ action, Phys. Lett. B 194 (1987) 63.
[22] A.A. Tseytlin, Sigma model approach to string theory effective actions with tachyons, J. Math. Phys. 42
(2001) 2854, hep-th/0011033.
[23] S. Nakamura, Closed-string tachyon condensation and the on-shell effective action of open-string tachyons,
Prog. Theor. Phys. 106 (2001) 989, hep-th/0105054.
[24] T. Kugo, T. Takahashi, Unoriented open-closed string field theory, Prog. Theor. Phys. 99 (1998) 649, hep-
th/9711100.
[25] A. Sen, F-theory and orientifolds, Nucl. Phys. B 475 (1996) 562, hep-th/9605150.
[26] A. Dabholkar, On condensation of closed-string tachyons, hep-th/0109019.
[27] J. Polchinski, E. Witten, Evidence for heterotic—type I string duality, Nucl. Phys. B 460 (1996) 525, hep-
th/9510169.
[28] Y. Arakane, H. Itoyama, H. Kunitomo, A. Tokura, Infinity cancellation, type I compactification and string
S-matrix functional, Nucl. Phys. B 486 (1997) 149, hep-th/9609151.
[29] S. Nakamura, Dirichlet boundary conditions in generalized Liouville theory toward a QCD string, Prog.
Theor. Phys. 104 (2000) 809, hep-th/0004172.
[30] A. Adams, J. Polchinski, E. Silverstein, Don’t panic! Closed string tachyons in ALE spacetimes, JHEP 0110
(2001) 029, hep-th/0108075.
[31] C. Vafa, Mirror symmetry and closed string tachyon condensation, hep-th/0111051.
[32] J.A. Harvey, D. Kutasov, E.J. Martinec, G. Moore, Localized tachyons and RG flows, hep-th/0111154.
[33] A. Dabholkar, C. Vafa, tt ∗ geometry and closed string tachyon potential, JHEP 0202 (2002) 008, hep-
th/0111155.
Higgs-mediated electric dipole moments

in the MSSM: an application to baryogenesis and
Higgs searches
Apostolos Pilaftsis a,b
a Theory Division, CERN, CH-1211 Geneva 23, Switzerland
b Department of Physics and Astronomy, University of Manchester, Manchester M13 9PL, UK
Received 30 July 2002; accepted 10 September 2002
Abstract
We perform a comprehensive study of the dominant two- and higher-loop contributions to the
205 Tl, neutron and muon electric dipole moments induced by Higgs bosons, third-generation quarks
and squarks, charginos and gluinos in the Minimal Supersymmetric Standard Model (MSSM). We
find that strong correlations exist among the contributing CP-violating operators, for large stop,
gluino and chargino phases, and for a wide range of values of tan β and charged Higgs-boson masses,
giving rise to large suppressions of the 205 Tl and neutron electric dipole moments below their present
experimental limits. Based on this observation, we discuss the constraints that the non-observation
of electric dipole moments imposes on the radiatively-generated CP-violating Higgs sector and
on the mechanism of electroweak baryogenesis in the MSSM. We improve previously suggested
benchmark scenarios of maximal CP violation for analyzing direct searches of CP-violating MSSM
Higgs bosons at high-energy colliders, and stress the important complementary rôle that a possible
high-sensitivity measurement of the muon electric dipole moment to the level of 10−24 e cm can
play in such analyses.
PACS: 11.30.Er; 14.80.Er
1. Introduction
The non-observation of electric dipole moments (EDMs) of the thallium atom and
neutron, as well as the absence of large flavour-changing neutral-current (FCNC) decays
E-mail address: pilaftsi@theory.ph.man.ac.uk (A. Pilaftsis).
PII: S 0 5 5 0 - 3 2 1 3 ( 0 2 ) 0 0 8 2 6 - X
264 A. Pilaftsis / Nuclear Physics B 644 (2002) 263–289
put severe constraints on the parameters of a theory. Especially, these constraints become
even more severe for supersymmetric theories, such as the MSSM, in which too large
FCNC and CP-violating effects are generically predicted at the one-loop level, resulting
in gross violations with experimental data. A possible resolution of such FCNC and CP
crises, often considered in the literature [1], makes use of the decoupling properties of the
heavy squarks and sleptons of the first two generations, whose masses should be larger than
∼ 10 TeV. Thus, for sufficiently heavy squarks and sleptons, the one-loop predictions for
FCNC and EDM observables can be suppressed up to levels compatible with experiment.
Also, such a solution poses no serious problem to the gauge hierarchy, as long as the
first two generations of squarks and sleptons are not much heavier than 10 TeV. In this
case, because of their suppressed Yukawa couplings, the radiative effect of the first two
generations of sfermions on the Higgs-boson mass spectrum is still negligible, with respect
to that of TeV scalar top and bottom quarks.1
Recently, it has been shown [2,3] that even third-generation squarks may lead by
themselves to observable effects on the electron and neutron EDMs through Higgs-
boson-mediated two-loop graphs of the Barr–Zee type [4]. This observation offers new
possibilities to probe the CP-violating soft-supersymmetry-breaking parameters related to
the third-generation squarks. Most interestingly, the same CP-violating parameters may
induce radiatively a CP-noninvariant Higgs-sector [5–8], leading to novel signatures at
high-energy colliders [8–11]. It is then obvious that EDM constraints do have important
implications for the phenomenological predictions within the above framework of the
MSSM with explicit CP violation. Moreover, employing upper limits on EDMs, one is,
in principle, able to derive constraints on the phase of the SU(2)L gaugino mass, mW ,
which plays a central rôle in electroweak baryogenesis [12] in the MSSM [13,14].
On the experimental side, the current upper limit on the electron EDM de , as derived
from the absence of a permanent atomic EDM for 205 Tl, has improved by a factor of almost
2 over the last few years [15,16]. Specifically, the reported 2σ upper limit on a thallium
EDM is [16]
|dTl | 1.3 × 10−24 [e cm]. (1.1)
Then, the electron EDM de may be deduced indirectly by means of the effective Lagrangian
1
N ēiγ5 e + CP N

iγ5 N ēe
LEDM = − de ēσµν iγ5 eF µν + CS N
2
+ CT N
σµν iγ5 N ēσ µν e + · · · , (1.2)
where CS , CP , CT and the ellipses denote CP-violating operators of dimension 6 and
higher. With the aid of the effective Lagrangian (1.2), the atomic EDM of 205Tl may be
computed by [17–19]

dTl [e cm] = −585 × de [e cm] + 8.5 × 10−19 [e cm] × CS TeV−2

− 8.0 × 10−22 [e cm] × CT TeV−2 + · · · . (1.3)
1 One should bear in mind that radiative effects on the neutral Higgs-boson masses are proportional to the
fourth power of Yukawa couplings. A simple estimate indicates that the contribution of the second generation of
sfermions is smaller, by a factor of at least 10−7 , than those of the third generation.
A. Pilaftsis / Nuclear Physics B 644 (2002) 263–289 265
In (1.3), the dots denote CP-odd operators of dimension 7 and higher. In our analysis,
we will assume that like CT , the CP-odd operators of dimension 7 and higher give rise
generically to negligible effects on the 205 Tl EDM. Moreover, although the contributions
of the neglected CP-odd operators to other heavy atoms may be comparable to that of de ,
the experimental upper limits are still much weaker than dTl , by at least one order of
magnitude. Consequently, we will only analyze predictions for the thallium EDM dTl
and consider only two operators: the electron EDM de and the CP-odd electron–nucleon
operator CS . From (1.1) and (1.3), it is then not difficult to deduce the following 2σ upper
limits on these two CP-odd operators:

|de | 2.2 × 10−27 [e cm], |CS | 1.5 × 10−6 TeV−2 . (1.4)
In the MSSM under study, the contributions from de and CS to dTl can be of comparable
size and therefore cannot be treated independently. In fact, depending on their relative
sign, one may increase or reduce the EDM bounds on the CP-violating parameters of the
theory. Here, the proposed high-sensitivity measurement of the muon EDM dµ to the level
10−24 e cm [20] may offer new constraints complementary to those obtained by dTl , since
CS and all higher-dimensional CP-odd operators are absent.
Unlike the thallium EDM, the upper limit on the neutron EDM dn is less severe, i.e.,
|dn | 1.2 × 10−25 [e cm], (1.5)

at the 2σ confidence level (CL) [21–23]. Moreover, although promising computations
based on QCD sum rules [24] appeared recently, the theoretical prediction for dn is
rather model-dependent. For example, the predictions between the valence-quark and
quark-parton models may differ, even up to one order of magnitude [25]. Recently, the
experimental upper limit on a permanent EDM of the 199 Hg atom has been improved
by a factor of 4, i.e., |dHg | < 2.33 × 10−28 e cm at the 2σ CL [26]. On the theoretical
side, however, the derivation of bounds [27] from dHg on the chromoelectric dipole
moment (CEDM) operators of u- and d-quarks contains many uncertainties related to
unknown effects of higher-dimensional chiral operators, nucleon-current ambiguities [28],
the neglect of the CP-odd three-gluon operator [29,30], the modelling for extracting the
nuclear Schiff moment [19], the s-quark content in heavy nuclei, etc. Thus, we shall not
implement mercury EDM constraints in our analysis. Instead, we will consider that no
large cancellations [31] below the 10% level occur among the different EDM terms in the
neutron EDM. In a sense, such a procedure takes account of a possible complementarity
relation [27] between the measurements of the neutron and Hg EDMs.
As we have already mentioned above, in the MSSM one-loop EDM effects [25,32–34]
can be greatly suppressed below their experimental limits, if the first two generation of
squarks and sleptons are made heavy enough, typically heavier than 10 TeV [25,33]. Within
such a framework of the MSSM [35], the dominant contributions to EDMs arise from
Higgs-mediated Barr–Zee-type two-loop graphs that involve quarks and squarks of the
third generation, charginos and gluinos.2 In this paper, we improve previous computations
2 Alternatively, one-loop EDM contributions can be suppressed if the CP phases of the trilinear soft-Yukawa
, Bino B
couplings of the first two generations and the CP phases of Wino W and gluino g̃ are all zero, with
of these two-loop contributions to EDMs, by resumming CP-even and CP-odd radiative

effects on the Higgs-boson self-energies and vertices [38,39]. Analogous improvements of
higher-order resummation effects are also considered in the computation of the CP-odd
electron–nucleon operator CS . Within the above resummation approach, we compute the
original Barr–Zee EDM graph induced by t-quarks beyond the two-loop approximation in
the MSSM through one-loop CP-violating threshold corrections to the top-quark Yukawa
coupling. Finally, we compute the Higgs-boson two-loop contribution to EDM induced by
charginos, and discuss the consequences of the derived EDM constraints on electroweak
baryogenesis in the MSSM.
The present paper is organized as follows: in Section 2, we discuss the CP-odd electron–
nucleon operator CS , which gives an enhanced contribution to the thallium EDM dTl in the
large tan β regime [40]. In Section 3, after reviewing the existing dominant Higgs-boson
two-loop contributions to EDMs, we compute the very relevant Barr–Zee contribution
to EDM from t-quarks for the first time in the MSSM. In addition, we critically re-
examine a very recent calculation [41] on Higgs-boson two-loop EDM effects due to
charginos. In Sections 2 and 3, we also improve previous computations of the CP-
odd electron–nucleon operator CS and the electron EDM de , by taking properly into
account higher-order CP-even and CP-odd resummation effects of Higgs-boson self-energy
and vertex graphs. Section 4 is devoted to numerical estimates of EDMs and discusses
the impact of the derived EDM constraints on electroweak baryogenesis and on the
analysis of direct searches for CP-violating Higgs bosons. Our conclusions are drawn in
Section 5.
2. CP-odd electron–nucleon operator CS
Let us first study the contribution of the CP-odd electron–nucleon operator CS [17–19]
to the 205 Tl EDM. At the elementary particle level, CS can be induced by two types of
CP-odd operators in supersymmetric theories: ēiγ5 eq̄q [40] and ēiγ5 eq̃ ∗ q̃, where q and q̃
denote quark and squark fields, respectively. In the MSSM, the above two CP-odd operators
of dimensions 6 and 5 are generated by interactions involving Higgs scalar–pseudoscalar
mixing and CP-violating vertex effects, as those shown in Fig. 1.
However, not all quarks and squarks can give rise to potentially large contributions to
the 205 Tl EDM. Our interest is to consider only enhanced Yukawa and trilinear couplings of
the Higgs bosons to quarks and squarks in the decoupling limit of the first two generation of
squarks. This criterion singles out the CP-odd operators related to top and bottom quarks,
and their supersymmetric partners. In fact, as is shown in Fig. 1, heavy quarks and squarks
do not contribute directly to the CP-odd operator CS , but only through the loop-induced
Bµ and µ being positive according to our CP conventions. In this case, however, if the first two generations of
sfermions are relatively light, e.g., few hundreds of GeV, then additional two-loop EDM graphs [36] exist, such
as those induced by a gluino CEDM, which give non-negligible contributions to the EDMs. Furthermore, there
are two-loop EDM effects induced by a CP-odd γ W + W − operator, which do not decouple in the limit of heavy
squarks and do not depend on Higgs-boson masses [37]. These two-loop EDM contributions are subdominant,
yielding an electron EDM term typically smaller than 10−27 e cm.
Fig. 1. Feynman graphs contributing to a non-vanishing CP-odd electron–nucleon operator CS . At the elementary
particle level, CS is predominantly induced by quantum effects involving (a) t-, b- quarks and (b) t˜-, b̃- squarks.
Blobs and heavy dots denote resummation of self-energy and vertex graphs, respectively.
Higgs-gluon–gluon couplings Hi gg, after they have been integrated out. Thus, the effective
Lagrangian responsible for generating CS is

3
(C ) gw Hi αs a,µν a
LeffS = gHi gg G Gµν + me tan βO3i ēiγ5 e , (2.1)
2MW 8π
i=1
where MW = gw v/2, O is the 3 ×3-mixing matrix that relates the weak to mass eigenstates
of the CP-violating Higgs bosons [6,8], and
2 v2 2 (H ) 2 (H )

gHi gg = g S
+ mq̃2 − mq̃1 ξq + mq̃1 + mq̃2 ζq
2 i 2 i . (2.2)
3 Hi qq 6m2q̃ m2q̃
q=t,b 1 2
S (H ) (H )
In (2.2), the dimensionless coefficients gH i qq
, ξq i , ζq i and the stop and sbottom masses
are given in Appendix A.
The largest contribution to the coupling parameter gHi gg comes from the scalar part of
S
the Hi b̄b coupling, gH i bb
. More explicitly, there are two CP-violating effects that dominate
S 2
gHi bb : (i) the tan β-enhanced threshold effects [40] described by the term

(+hb / hb ) tan2 β
gHi bb ∼ Im
S
O3i , (2.3)
1 + (δhb / hb ) + (+hb / hb ) tan β
and (ii) the scalar–pseudoscalar mixing effects contained in the mixing matrix ele-
ments O1i . The definition of the quantities δhb / hb and +hb / hb may be found in
Appendix A.
At this stage, it is important to observe that if (+hb / hb ) tan β 1, the tan2 β-
S P
dependence of the CP-violating threshold effects on gH i bb
and gH i bb
considerably
S P
modifies. In particular, in the large tan β limit, gHi bb and gHi bb asymptotically approach a
tan β-independent constant, i.e.,

1 + (δhb / hb )
gHS
i bb
→ Im O3i , (2.4)
(+hb / hb )

1 + (δhb / hb )
P
gH i bb
→ Im O1i . (2.5)
(+hb / hb )
Although the above limits may only be attainable in a very large tan β and quasi-
nonperturbative regime of the theory, the onset of a tan β-independent behaviour in gH S
i bb
and gH P
i bb may already start from moderately large values of tan β, i.e., for tan β 30.
Consequently, the limits (2.4) and (2.5) should be regarded as upper bounds on the
S P
CP-violating threshold-enhanced parts of the coupling parameters gH i bb
and gH i bb
. In
our numerical analysis in Section 4, we properly take into account the above-described
S
CP-violating resummation effects on gH i bb
.
The computation of the CP-odd electron–nucleon operator CS can now be performed by
utilizing standard QCD techniques based on the trace anomaly of the energy-momentum
tensor [42]. In the chiral quark mass limit, we then have the simple relation
αs
N| Ga,µν Gaµν |N = −(100 MeV)NN. (2.6)

8π
With the help of (2.6), we can evaluate the effective Hi NN
couplings, and, hence, the
CP-odd operator CS :
me παw gHi gg O3i

3
CS = −(0.1 GeV) tan β 2 2
. (2.7)
MW i=1
MH i
Observe that the operator CS exhibits an enhanced tan3 β dependence [40]; it, therefore,
becomes very significant for intermediate and large values of tan β. Numerical estimates
for this contribution to a thallium EDM will be presented in Section 4.
3. Higgs-boson two-loop contributions to de
We now turn our attention to Higgs-boson two-loop effects [2,3] on the electron EDM
analogous to those first discussed by Barr and Zee [4] in non-supersymmetric theories. As
is shown in Fig. 2, these two-loop EDM effects originate predominantly from graphs that
involve: stop and sbottom squarks (Fig. 2(a,b)) [2], top and bottom quarks (Fig. 2(c)), and
charginos (Fig. 2(d)) [41].
Strictly speaking, the original Barr–Zee graphs induced by top and bottom quarks
in Fig. 2(c) appear beyond the two-loop approximation in the MSSM. However, it is a
formidable task to analytically compute the complete set of the relevant three- and higher-
loop graphs. Therefore, we consider only a subset of higher-loop corrections, in which
the dominant CP-violating terms in the Higgs-boson propagators and the Higgs-quark-
quark vertices are resummed. Such an approach should only be viewed as an effective one,
which is expected to capture the main bulk of the higher-order effects. In the same vein, we
improve previous two-loop EDM calculations related to third-generation squarks [2] and
charginos [41] by resumming dominant CP-violating self-energy terms in the Higgs-boson
propagators.
In the context of the aforementioned resummation approach, the dominant Higgs-boson
two-loop contributions to electron EDM are individually found to be
Fig. 2. Dominant Higgs-boson two-loop contributions to EDM of a light fermion f = e, µ, d in the MSSM
with explicit CP violation (mirror-symmetric graphs are not displayed). Heavy dots indicate resummation of
self-energy and vertex graphs. Two-loop graphs generating a CEDM for a d-quark are also shown.

3 P m2 m2
de 3αem gH i ee q̃1 q̃2
= m e Q2
q qξ (Hi )
F − F
e (a,b) 32π 3 MH 2 MH2 MH2
i=1 i q=t,b i i
m2 m2
q̃1 q̃2
+ ζq(Hi ) F 2
+ F 2
, (3.1)
MH i
M Hi
2
me 2 P
2 3
de 3αem mq
=− 2 2
Qq g g S
Hi ee Hi qq f 2
e 2
8π sin θw MW i=1 q=t,b MH
(c) i

m2q
+ gH
S
gP g
i ee Hi qq 2
, (3.2)
MH i
m + 2
me 1
2 3
de αem χj
=− √ gH
P
ee a H χ −
χ + f 2
e (d) 8 2 π 2 sin2 θw MW i=1 j =1,2 mχj+ i i j j MH i
m2 +
χj
+ gH
S
b − +g
i ee Hi χj χj 2
, (3.3)
MH i
P
where gH i ee
= − tan βO3i , gH
S
i ee
= O1i / cos β, and
1
x(1 − x) x(1 − x)
F (z) = dx ln , (3.4)
z − x(1 − x) z
0
1
z 1 − 2x(1 − x) x(1 − x)
f (z) = dx ln , (3.5)
2 x(1 − x) − z z
0
1
z 1 x(1 − x)
g(z) = dx ln (3.6)
2 x(1 − x) − z z
0
S P
are two-loop functions. The coupling coefficients gH i qq
, gH i qq
, aHi χ − χ + and bHi χ − χ + in
j j j j
(3.1)–(3.3), as well as the squark and chargino masses, are given in Appendix A. Eq. (3.1)
takes on the simpler analytic form presented in [2], if only the CP-odd component a of the
Higgs bosons is considered in an unresummed two-loop calculation of the EDM. In this
(H )
case, the coefficients ζ (Hi ) vanish and ξq i simplifies to
sin 2θq mq Im(µeiδq ) Rq 2m2q Im(µAq )

ξq = Rq = , (3.7)
sin β cos βv 2 sin β cos β v 2 (m2q̃ − m2q̃ )
2 1
where δq = arg(Aq − Rq µ∗ ), with Rt (Rb ) = cot β(tan β).

In addition to the dominant Higgs-boson two-loop graphs we have been studying
here, there are also subdominant two-loop EDM diagrams, where the virtual photon is
replaced by a Z boson in Fig. 2. Another class of Higgs-boson two-loop graphs involve
the coupling of the charged Higgs bosons H ± to the photon and the W ∓ bosons [3]. In
this case, for example, the graph in Fig. 2 will proceed via charginos and neutralino in the
fermionic loop. As has been explicitly shown in [3] for most of the cases, this additional
set of graphs give almost one order of magnitude smaller contributions to EDM. Most
importantly, their dependence on the CP-violating parameters of the theory is rather closely
related to the two-loop EDM graphs depicted in Fig. 2. Thus, suppressing the dominant
Higgs-boson two-loop contributions to EDM will automatically lead to a corresponding
suppression of this additional set of two-loop graphs. Therefore, in our analysis we neglect
the aforementioned set of subdominant two-loop graphs.
So far, we have only been studying the electron EDM de . The two-loop prediction for the
muon EDM dµ can easily be obtained from de by considering the obvious mass rescaling
factor mµ /me ≈ 205, i.e.,

dµ ≈ 205de t, b, t˜, b̃, χ ± , (3.8)
where the different two-loop EDM contributions are indicated within the parentheses.
On the other hand, the dominant contributions to neutron EDM dn come from the
CEDM of the d quark and CP-odd three-gluon operator [30], which was first discussed
by Weinberg [29] in non-supersymmetric multi-Higgs doublet models. In the MSSM, the
CP-odd three-gluon operator decouples as 1/m3g̃ and becomes relevant at gluino masses
below the TeV scale. The CEDM of the d quark may be obtained from (3.1) and (3.2), if
one replaces the colour factor 3 by 1/2, and αem Q2q by αs . The computation of the neutron
EDM dn involves a number of hadronic uncertainties, when the EDMs are translated from
the quark to the hadron level [25]. For example, considering the valence-quark model and
renormalization-group running effects from the electroweak scale MZ down to the low-
energy hadronic scale Λh [3], one may be able to establish an approximate relation between
neutron and electron EDMs. Thus, taking the input values for the involved
√ kinematic
parameters: md (Λh ) = 10 MeV, αs (MZ ) = 0.12 and gs (Λh )/(4π) = 1/ 6, we find

dn ≈ −10de (t, t˜) + 1.2de t, t˜, χ ± + dn3G . (3.9)
On the RHS of (3.9), the first and second terms arise from a d-quark CEDM and EDM,
respectively, and dn3G is the contribution to dn due to the CP-odd three-gluon operator. In
obtaining (3.9), we have made two additional approximations as well. First, we neglected
the contribution of the u-quark EDM du to dn , as it is much smaller than the d-quark EDM
dd for the relevant region tan β 3. Second, we ignored the b- and b̃-quantum corrections
to dd and so to dn . Formulae (3.8) and (3.9) will be used to obtain numerical predictions
for the muon and neutron EDMs in the next section.
4. Numerical estimates and discussion
In Sections 2 and 3, we computed the dominant two-loop and the resummed higher-
loop contributions to EDMs that originate from third-generation quarks and squarks, and
charginos. Based on the derived analytic expressions, we can now analyze numerically
the impact of the experimental constraints due to the non-observation of thallium and
neutron EDMs on the CP-violating parameters of the theory, and hence on electroweak
baryogenesis and direct searches for CP-violating Higgs bosons in the MSSM. Moreover,
we will present predictions for the muon EDM dµ and discuss the implications of a possible
high-sensitivity measurement of dµ to the level 10−24 e cm for our analyses.
Based on the observation that CP-violating quantum effects on the neutral Higgs
sector get enhanced when the product Im(µAt )/MSUSY2 is large [5,6], the authors in [8,9]
introduced a benchmark scenario, called CPX, in which the effects of CP violation are
maximized. In CPX, the following values for the µ- and soft-SUSY-breaking parameters
were adopted:
M t = M
Q = M b = MSUSY , µ = 4MSUSY ,
|At | = |Ab | = 2MSUSY , arg(At,b ) = 90◦ ,
|mg̃ | = 1 [TeV], arg(mg̃ ) = 90◦ ,
= mB
mW = 0.3 [TeV]. (4.1)
Without loss of generality, the µ-parameter is chosen to be real. The predictions of CPX
showed [9] that even a light neutral Higgs boson with a mass as low as 60 GeV could
have escaped detection at LEP2.3 A recent experimental analysis of LEP2 data confirms
3 Similar remarks were made earlier in [6], but the LEP2 data were less restrictive then.
this observation [43]. Here, we wish to investigate the compatibility of the CPX scenario
with the experimental limits on EDMs. For this purpose, we allow variations in the gluino
phase, which enters the Higgs sector at two loops, but keep the At phase in (4.1) fixed.
In addition, we will present numerical results for EDMs, where the µ-parameter is varied
from 100 GeV to 4MSUSY . Finally, we leave unspecified the phases of the gaugino mass
parameters mW and mB . As we will see below, the phase of mW is greatly affected by
constraints from the electron EDM.
We start our numerical analysis by presenting predicted values for the 205 Tl EDM
dTl that arise entirely due to the CP-odd electron–nucleon operator CS and are denoted
as dTl (CS ). In Fig. 3, we display numerical estimates for dTl(CS ) as functions of tan β for
four different versions of the CPX scenario with MSUSY = 1 TeV: (a) MH + = 150 GeV,
arg(mg̃ ) = 0 ◦ ; (b) MH + = 300 GeV, arg(mg̃ ) = 0◦ ; (c) MH + = 150 GeV, arg(mg̃ ) = 90 ◦ ;
(d) MH + = 300 GeV, arg(mg̃ ) = 90◦ . The individual b, t˜, t, b̃ contributions to dTl (CS ),
along with their relative signs, are indicated by different types of lines. We observe that
the largest contribution to dTl comes from the b-quarks for large values of tan β, i.e.,
for tan β 15, for which the CP-violating vertex effects become important (see also
the discussion in Section 2). In particular, these CP-violating threshold effects, which
crucially depend on the term Im(+hb / hb ) tan2 β in (2.3), become even more important
for large gluino phases. Thus, the predictions for dTl (CS ) in panels Fig. 3(a) and (b), with
arg(mg̃ ) = 0◦ , are one order of magnitude larger than the ones in Figs. 3(c) and (d), with
arg(mg̃ ) = 90◦ .
For intermediate and smaller values of tan β, i.e., for tan β 15, CP-violating self-
energy effects are significant, especially for relatively light charged Higgs bosons with
masses in the range 150–200 GeV. In fact, these effects have generically opposite sign to
the CP-violating vertex effects, giving rise to natural cancellations among the contributing
EDM terms, and so lead to smaller values of dTl (CS ). Although our numerical results are
in qualitative agreement with those in Ref. [40], we actually find noticeable quantitative
differences, when resummed CP-violating self-energy and vertex effects are considered at
the same time.
Next, we shall investigate numerically higher-order CP-violating vertex and self-energy
effects induced by t- and b-quarks on the electron EDM de . Fig. 4 shows numerical
estimates for those resummed effects on de as functions of tan β, in variants of the CPX
scenario, with (a) MH + = 150 GeV and (b) MH + = 300 GeV. In particular, we considered
three different choices of the gluino phase: arg(mg̃ ) = 90◦ , 0, −90◦ , denoted as t1,2,3 ,
respectively. We find that CP-violating threshold corrections to the Hi tt coupling as small
as 5% are sufficient to lead to observable EDM values for de . In this respect, we see that
the t-quark effects strongly depend on the gluino phase through the combination Im(µmg̃ )
that occurs in Im(+ht / ht ) [cf. (A.4), (A.5)]. Thus, the t-quark contribution to de is positive
(negative) for negative (positive) gluino phases, while it is one order of magnitude smaller
and negative for vanishing gluino phases, i.e., for arg(mg̃ ) = 0◦ . For comparison, we also
included in Fig. 4 the dependence of positive stop/sbottom contributions to de [2] (long-
dash-dotted lines) on tan β. The sum of the t-, b-quark and t˜-, b̃-squark contributions
to de is given by the solid lines 1, 2, 3 for the same values of gluino phases. As before,
we indicate negative contributions to de with a minus sign. From Figs. 4(a) and (b), it
is interesting to notice that if arg(mg̃ ) = 90◦ in CPX, a cancellation between the t-quark
Fig. 3. Numerical estimates of 205 Tl EDM dTl induced by the CP-odd electron–nucleon operator CS as functions
of tan β, in four selected CPX scenarios with MSUSY = 1 TeV. The values of the CPX parameters are given
in (4.1). The individual b, t˜, t, b̃ contributions to dTl (CS ), along with their relative signs, are also displayed.
and t˜-squark EDM contributions occurs for almost the entire range of the perturbatively
allowed tan β values and for all phenomenologically viable charged Higgs-boson masses.
As a consequence of such a cancellation, the electron EDM de is always smaller than the
current 2σ experimental limit on de , i.e., de < 2.2 × 10−27 e cm, even for large tan β values
up to 30. As we will see below, this prediction may considerably change if contributions
from the CP-violating operator CS or chargino two-loop effects are considered.
In order to further gauge the importance of the t-quark two-loop EDM effects, we
present in Fig. 5 numerical values for de versus the µ-parameter for tan β = 20, and
for two charged Higgs-boson masses: (a) MH + = 150 GeV and (b) MH + = 300 GeV.
The soft-SUSY-breaking parameters are chosen as given in (4.1) for MSUSY = 1 TeV,
Fig. 4. Numerical estimates of resummed Higgs-boson two-loop effects on de , induced by t-, b-quarks and
t˜-, b̃-squarks, as functions of tan β, in two variants of the CPX scenario, with (a) MH + = 150 GeV and
(b) MH + = 300 GeV. The long-dash-dotted lines indicate the stop/sbottom contributions to de . The dotted lines
t1,2,3 correspond to top/bottom contributions, for arg(mg̃ ) = 90◦ , 0◦ , −90◦ , respectively. Likewise, the solid
lines 1, 2, 3 give the sum of all the aforementioned contributions to de for the same values of gluino phases.
Contributions to de that are denoted with a minus sign are negative.
except for the µ-parameter, which has been varied from 0.1–4 TeV. For the sake of
comparison, we also included the Higgs-boson two-loop EDM effects induced by t˜- and
b̃-squarks. The meaning of the various types of lines is exactly the same as those in Fig. 4.
Remarkably enough, we find that even µ values as low as 500 GeV may be sufficient
to lead to an electron EDM at the observable level through the original two-loop Barr–
Zee graph in Fig. 2(c). In this context, we also observe that the resummed Higgs-boson
two-loop contributions to de from t-quarks are comparable and even larger than those
coming from t˜-squarks for maximal gluino phases. In fact, if At,b = 0, the t˜-squark and
dominant CP-violating Higgs-mixing effects may be completely switched off, without
Fig. 5. Numerical values of resummed Higgs-boson two-loop effects on de , induced by t-, b-quarks and t˜-,
b̃-squarks, as functions of µ, in two variants of the CPX scenario, with tan β = 20, and (a) MH + = 150 GeV
and (b) MH + = 300 GeV. The meaning of the different line types is identical to that of Fig. 4. For At,b = 0, the
long-dash-dotted line disappears and so the solid lines collapse to the dotted ones.
much affecting the corresponding t-quark two-loop contributions to de . Note that in this
case the t-quark effects on de and the b-quark effects on the CS operator, which both
formally arise at the two-loop level, are proportional to Im(µmg̃ ). Therefore, they turn out
to be strongly correlated and their combined contribution to dTl should carefully be taken
into account (see also discussions of Figs. 7 and 8 below).
As was already pointed out in [2,3], charginos might also contribute to electron EDM
de at the two-loop level. Recently, a computation of those effects appeared in [41].
The authors derived strict constraints on the CP-violating parameters of a scenario in
which electroweak baryogenesis is mediated by CP-violating currents involving chargino
interactions. Here, we re-examine this issue within a scenario that favours the above
mechanism of electroweak baryogenesis and is not in conflict with LEP2 limits on the
Higgs-boson masses and couplings. Specifically, being conservative, we require that these
be MHi 111 GeV, for gH 2
i ZZ
0.3, where gHi ZZ is the Hi ZZ coupling given in units
of the SM hSM ZZ coupling. In addition, we demand that MHi + MHj 170 GeV. On
the other hand, in order for electroweak baryogenesis to proceed via a sufficiently strong
first-order phase transition, the right-handed stop mass parameter Mt must be rather small,
and the µ and the soft gaugino parameter mW must not be too large, typically smaller
than 0.5 TeV [13,14]. Especially, there is a resonant enhancement up to even 10 times the
observed baryon asymmetry, if the condition µ = mW is met [14]. Further requirements
for a scenario leading to successful electroweak baryogenesis are: (i) a moderate trilinear
At -parameter in the range, 0.2 At /M Q 0.65; (ii) a not very large tan β value,
tan β 20; (iii) a soft-SUSY-breaking parameter M Q of a few TeV, for phenomenological
reasons [14]. More explicitly, the following values for the mass parameters are employed:
Q = 3 TeV,
M t = 0,
M b = 3 [TeV],
M
|At | = |Ab | = 1.8 [TeV], arg(At,b ) = 0◦ , tan β 20,
◦
|mg̃ | = 3 TeV, arg(mg̃ ) = 0 ,
◦
µ = |mW
| 0.5 [TeV], ) = 90 .
arg(mW (4.2)
To be able to compare our predictions with those presented in Fig. 2 of Ref. [41],
we choose in Fig. 6(a) the values: µ = mW = 0.2 TeV and MH + = 170 GeV. Since
CP-violating Higgs-mixing effects in the mass spectrum are generically small for the
chosen values of the parameters in (4.2), our mass input MH + = 170 GeV corresponds to
M‘A’ ≈ 150 GeV for the mass of the almost CP-odd Higgs scalar A. Even though on a very
qualitative basis our numerical results on the linear tan β-increase behaviour of de agree
with those reported in [41], the actual functional dependences of the individual ‘h’, ‘H ’,
‘A’ contributions to de on tan β differ significantly. Unlike [41], we find in Fig. 6(a) that
for tan β 5, the tan β-enhanced effect on de originates from the heavier Higgs bosons
‘H ’ and ‘A’, while the EDM contribution due to the lightest Higgs boson ‘h’ is almost
negligible.4 Since the size of de is set by the heavier Higgs-boson masses, i.e., by MH + , and
by µ and mW , our predictions are rather robust under the different choices of the remaining
soft-SUSY-breaking parameters. Moreover, although our numerical values for the total
contribution to de agree very well with [41] for tan β = 2 (de ≈ 0.63 × 10−26 e cm),
they are smaller by ∼20% for tan β = 6, i.e., we find de ≈ 1.62 × 10−26 e cm, which
should be compared with de ≈ 2 × 10−26 e cm. Finally, the electroweak baryogenesis
scenario (4.2) in the low tan β region, tan β 6, which is studied by the authors in [41],
appears to be highly disfavoured by LEP2 data. In this respect, a phenomenologically
viable model, with MH + = 170 GeV, would require larger values of tan β, i.e., tan β 9.
In this case, one has to consider a factor of 10 suppression in the chargino phase, such
that the chargino two-loop EDM effects are reduced to a level close to the experimental
upper limit on de . Consequently, if no cancellations are assumed with possible one-loop
EDM terms, then a model with suppressed chargino phase of ∼ 5◦ and a relatively light
4 The fact that only ‘H ’ and ‘A’ contributions to d exhibit a linearly enhanced dependence on tan β may also
e
be verified independently by a flavour-graph analysis.
Fig. 6. de versus tan β in a scenario favoured by electroweak baryogenesis, with MSSM parameters
Q = M
M D = 3 TeV, M U = 0, At,b = 1.8 TeV, mg̃ = 3 TeV and arg(At,b ) = arg(mg̃ ) = 0◦ . In (a),
MH + = 170 GeV is used, corresponding to M‘A’ ≈ 150 GeV, and mW ◦
= µ = 0.2 TeV and arg(mW ) = 90 .
Also displayed are the individual ‘h’, ‘H ’, ‘A’ contributions to de and the LEP excluded region from direct
Higgs-boson searches. In (b), numerical values are shown for MH + = 150 GeV (solid), 200 GeV (dashed),
300 GeV (dotted), 500 GeV (dash-dotted) and 1 TeV (long-dash-dotted), in a scenario with mW = µ = 0.4 TeV
◦
and arg(mW ) = 90 .
charged Higgs boson, MH + = 150–200 GeV, might still be possible to account for the
observed baryon asymmetry in the Universe, provided the aforementioned resonant factor
10 is used. However, the above situation may be considerably relaxed for larger values of
MH + , since the chargino two-loop EDM effect on de decreases approximately by 1/MH +
as MH + increases. This dependence of de on MH + can explicitly be seen in the lower
panel (b) of Fig. 6, for increasing charged Higgs-boson masses: MH + = 150 GeV (solid),
200 GeV (dashed), 300 GeV (dotted), 500 GeV (dash-dotted) and 1 TeV (long-dash-
◦
dotted), in a scenario with mW = µ = 0.4 TeV and arg(mW ) = 90 .
Fig. 7. EDMs of dTl , dn and dµ as functions of tan β in two versions of the CPX scenario:
(a) arg(mg̃ ) = arg(mW ◦ ◦ ◦
) = 90 , and (b) arg(mg̃ ) = 35 , arg(mW ) = 90 . Also shown are the different contribu-
tions, along with their relative signs, to dTl from top/stop (long-dash-dotted) and chargino (dotted) Higgs-boson
two-loop graphs, as well as from the CP-odd electron–nucleon coupling CS (dashed).
In the following, we will present predictions for more realistic EDM observables, with
relatively reduced hadronic uncertainties, namely the thallium EDM dTl , the neutron EDM
dn , as well as the muon EDM dµ which was suggested to be measured with a high
sensitivity to the level of 10−24 e cm [20]. In Fig. 7, we display numerical values for dTl , dn
and dµ as functions of tan β in two versions of the CPX scenario, with MH + = 150 GeV:
(a) arg(mg̃ ) = arg(mW ◦ ◦ ◦
) = 90 , and (b) arg(mg̃ ) = 35 , arg(mW ) = 90 . Fig. 7 also shows
the different contributions, along with their relative signs, to dTl from top/stop (long-dash-
dotted) and chargino (dotted) Higgs-boson two-loop graphs, as well as from the CP-odd
electron–nucleon operator CS (dashed). Note that the type of lines used to represent the
numerical results of the individual EDM contributions is given in the parentheses. In

panel (a) of Fig. 7, we see that the contribution of CS prevails in dTl , for large values
of tan β, and exceeds the experimental limit for tan β 12. The prediction for dn always
stays below the current experimental limit, and the predicted values for dµ do not reach
the proposed experimental sensitivity for almost all relevant values of tan β. It is amusing
to remark that no EDM constraints can be imposed on the CPX scenario in the range:
4 tan β 12, which is interesting for analyzing Higgs-boson searches at high-energy
colliders. In fact, if the gluino phase is chosen to be arg(mg̃ ) = 35 ◦ (see Fig. 7(b)), the
different EDM terms contributing to dTl approximately cancel and dTl does not exceed
much the experimental limit. Similarly, since the top/stop-CEDM effects are small in this
CPX scenario, the neutron EDM is always smaller than its conservative experimental upper
bound: 1.2 × 10−25 e cm. However, for tan β ≈ 40, the muon EDM can be significant,
and its value dµ ∼ 4.0 × 10−24 e cm lies well within the proposed explorable range.
This example nicely illustrates the important rôle of complementarity of a high-sensitivity
measurement of a muon EDM in constraining the CP-violating parameter space of the
MSSM.
It is also interesting to examine the dependence of the different EDM contributions
shown in Fig. 7 on the µ-parameter, for large values of tan β. In Fig. 8, we display
numerical values of dTl , dn and dµ as functions of µ for two scenarios with tan β = 40,
MSUSY = 1 TeV, mg̃ = 1 TeV, mW ◦
= mB = 0.3 TeV, arg(mg̃ ) = arg(mW ) = 90 , At,b =
◦
2 TeV, arg(At,b ) = 90 : (a) MH + = 150 GeV; (b) MH + = 300 GeV. In analogy with Fig. 7,
the individual contributions to dTl due to top/stop and chargino two-loop graphs and due to
the CP-odd electron–nucleon operator CS are also shown. In Fig. 8(a), we observe that the
different CP-violating EDM operators may cancel in dTl and dn , even for smaller values of
the µ-parameter, i.e., for µ ≈ 700 GeV. In this region of parameter space, the muon EDM
is predicted to be as large as 0.8 × 10−23 e cm, which falls within the reach of the proposed
dµ measurement. In Fig. 8(b), we give numerical estimates of dTl , dn and dµ for a CPX
scenario with a heavier charged Higgs boson, i.e., for MH + = 300 GeV. Again, we find
that the predicted value for dTl can be close to the experimental limit for a wide range of
µ-values, while dµ always stays above the proposed experimental sensitivity.
Let us summarize the focal points of this section. We have explicitly demonstrated that
the non-observation of the thallium EDM can provide strict constraints on the CP-violating
parameters related to third-generation squarks, charginos and gluinos. The constraints
derived from the neutron EDM limit are less restrictive. Nevertheless, our numerical
analysis has also shown that the constraints from the thallium EDM can be significantly
weakened, if the different CP-violating operators de and CS cancel in dTl . For instance,
this could be the case for the benchmark scenario CPX, for low and intermediate values
of tan β. Such cancellations of the CP-violating operators de and CS can occur for a
wide range of parameters and crucially depend on the choice of the phase combinations:
arg(µAt ), arg(µmg̃ ) and arg(µmW ). In particular, we find that a possible high-sensitivity
measurement of dµ to the proposed level of 10−24 e cm can constrain such uncovered
ranges of CP-violating parameters in a rather complementary way. Finally, unless MH +
is of the TeV order, EDM constraints from dTl on scenarios favoured by electroweak
baryogenesis are rather stringent. They generally imply either suppressed chargino phases,
Fig. 8. Numerical values of dTl , dn and dµ as functions of µ for two large-tan β scenarios, with tan β = 40,
MSUSY = 1 TeV, mg̃ = 1 TeV, mW ◦
= mB = 0.3 TeV, arg(mg̃ ) = arg(mW ) = 90 , At,b = 2 TeV,
arg(At,b ) = 90◦ : (a) MH + = 150 GeV; (b) MH + = 300 GeV. In analogy with Fig. 7, the individual contri-
butions to dTl due to top/stop and chargino two loop graphs and due to the CS operator are also shown.
◦
) 10 , or modest cancellations in 1 part to 10 with one-loop EDM terms
i.e., arg(mW
induced by the first two generations of sleptons.
5. Conclusions
To avoid the known CP and FCNC crises in the MSSM, we have considered a
framework, in which the first two generations of squarks and sleptons are heavier than
∼ 10 TeV, while the third generation is light, with masses not larger than the order of
a TeV. Within this framework of the MSSM, we have performed a systematic study of the
dominant two- and higher-loop contributions to the thallium, neutron and muon EDMs,
which are induced by b-, t-quarks, b̃-, t˜-squarks, charginos and gluinos. At present, the
most severe limits are obtained from the non-observation of a thallium EDM dTl , whereas
experimental upper limits on the neutron EDM dn are less stringent and usually constrain
large contributions from a d-quark CEDM and the CP-odd three-gluon operator. Also,
theoretical predictions for dn are plagued by a number of uncertainties while estimating
hadronic matrix elements.
The largest effects on the thallium EDM dTl result from two operators, the CP-odd
electron–nucleon operator CS and the electron EDM de . These two CP-violating op-
erators are formally induced at the two- and higher-loop levels and involve the ex-
change of CP-mixed Higgs bosons. Thus, strong constraints on the radiatively-generated
CP-violating Higgs sector of the MSSM can be derived from dTl , and hence on the analyses
for direct searches of CP-violating Higgs bosons at high-energy colliders, such as LEP2,
Tevatron and LHC [44]. In this context, we have analyzed the compatibility of an earlier
suggested benchmark scenario of maximal CP violation for LEP2 Higgs studies (CPX) [9]
with the thallium and neutron EDMs. We have observed the existence of strong correla-
tions among the different EDM terms, which enable the suppression of dTl and dn even
below the present experimental limits. Specifically, for 4 tan β 12 in the CPX scenario
with MH + = 150 GeV, the stop, gluino, and chargino phases are all allowed to receive their
maximal values, i.e., arg(At ) = arg(mg̃ ) = arg(mW ◦
) = 90 , without being in conflict with
EDM limits (cf. Fig. 7(a)). Most interestingly, for specific choices of the gluino phase, the
allowed range of tan β values compatible with EDM limits can be enlarged dramatically.
For instance, if arg(mg̃ ) = 35◦ in the aforementioned CPX scenario (see also Fig. 7(b)), the
predicted values for 25 tan β 45 do not contradict upper limits on thallium and neutron
EDMs. For the remaining range of tan β values, the obtained prediction does not exceed
the 2σ upper bound on |dTl | by a factor of ∼ 3. Evidently, the degree of cancellations re-
quired between the one- and two-loop EDM terms in the CPX scenario is not excessive,
for certain choices of the gluino phase.
At this point, it is important to stress that a muon EDM dµ measured at the 10−24 e cm
level will help to sensitively probe CP-violating regions of the MSSM parameter space
which cannot be accessed easily by measurements of the thallium and neutron EDMs.
This complementarity property is mainly a consequence of the fact that dµ is free from
interfering CP-odd electron–nucleus interactions thanks to the CS operator, which can
contribute significantly to dTl . Unlike the neutron EDM dn , dµ does not suffer from
hadronic uncertainties. Given the absence of a signal in the measurements of |dTl | and |dn |,
one may now wonder whether a positive signal in dµ would already imply a positive signal
on g − 2 as well. This is not the case within our framework of the MSSM. If the first two
generations of sfermions are above the TeV scale, the biggest contribution to g − 2 comes
again from related two-loop Barr–Zee-type graphs. However, for phenomenologically
viable charged Higgs-boson masses MH + 120 GeV in the MSSM [43], these effects
on g − 2 are negligible [45]. Then, only post-LEP2 high-energy colliders and the proposed
BNL experiment [20] on the muon EDM dµ might be able to sensitively explore the CP-
violating parameter space of the above framework of the MSSM in a rather complementary
manner.
We have also studied the impact of EDM constraints on the mechanism of electroweak
baryogenesis induced by CP-violating chargino currents. For this purpose, we considered
a scenario in (4.2), which favours the above mechanism of electroweak baryogenesis [14].
In such a scenario, the chargino two-loop graphs of Fig. 2(d) represent the dominant
contribution to de and dTl as well. However, as we detailed in Section 4, our theoretical
predictions for de are at variance with those presented in a recent communication [41].
Moreover, we find that LEP2 direct limits on Higgs-boson masses require intermediate
and larger values of tan β, i.e., tan β 6, for a phenomenologically viable scenario of
electroweak baryogenesis. In this tan β regime, experimental upper limits on |dTl| give
rise to strict constraints, especially when no cancellations between the chargino two-loop
and one-loop EDM terms are assumed. In the latter case, the charged Higgs-boson mass
MH + should be relatively large, i.e., MH + 700 GeV for tan β 6 and arg(mW )
◦
90 . Otherwise, for lighter charged Higgs bosons, either the chargino phase should be
suppressed by a factor of at least 10 or cancellations in 1 part to 10 with one-loop EDM
terms need be invoked.
In our computation of the Higgs-boson loop-induced EDMs, we have considered
resummation effects of higher-order CP-conserving and CP-violating terms in Higgs-boson
self-energies and vertices. In particular, the original t-quark two-loop graph suggested by
Barr and Zee [4] occurs beyond the two-loop approximation through threshold effects in
the Hi t¯t coupling and, depending on the choice of the gluino phase, it might even compete
with the t˜-squark two-loop graph [2]. Since our resummation of higher-order terms relied
on an effective Lagrangian approach, one may worry about the relevance of other higher-
order terms present in a complete computation. At this stage, we can only offer estimates
of those possible higher-order electroweak uncertainties in the calculation of EDMs. Thus,
we have checked our results with and without resumming the Higgs-boson self-energies.
In this way, no large modifications are found in our predictions; the variation of our results
is generally less than 10% for MH + 170 GeV, and becomes even smaller, to less than
1% for MH + 200 GeV. This may be attributed to the fact that the dominant contributions
to EDMs come from the heaviest Higgs bosons, on which the relative impact of radiative
effects is less important. On the other hand, CP-violating threshold effects constitute the
main source of theoretical uncertainties in the calculation of the original Barr–Zee graph
of Fig. 2(c), as they are less controllable for low values of tan β.5 In this context, we
remark that even the computation of the CP-odd three-gluon operator is haunted by relevant
higher-order electroweak uncertainties in the MSSM [30]. The Weinberg operator can be
generated in its original fashion [29] at three and higher loops which involve CP-violating
self-energy and vertex subgraphs of Higgs bosons. It then appears necessary to develop
improved techniques that would enable us to provide accurate estimates of (resummed)
higher-order terms in the calculation of EDMs. The present work is a step towards this
goal.
5 A crude estimate suggests that these additional higher-order effects are smaller than 20%.
Acknowledgements
The author thanks Marcela Carena, Carlos Wagner for discussions on issues related to
electroweak baryogenesis, Adam Ritz for comments on the computation of the mercury
EDM, Maxim Pospelov for clarifying remarks pertinent to [40], Darwin Chang and Wai-
Yee Keung for communications with regard to [41], and Athanasios Dedes for critical
remarks.
Appendix A. Effective Higgs-boson couplings
The couplings of the CP-mixed Higgs bosons H1,2,3 to t-, b-quarks, t˜-, b̃-squarks and
charginos χ + play a key rôle in our calculations. In this appendix we present the effective
Lagrangians describing the above interactions, after including dominant one- and two-loop
CP-even/CP-odd quantum effects on the Higgs-boson masses and their respective mixings.
Following the conventions of [8], we first write down the effective Lagrangian of the
Higgs-boson couplings to top and bottom quarks

3
gw mb S gw mt S
LH q̄q = − Hi b̄ gHi bb + igH
P
γ 5 b + t¯ g + ig P
γ 5 t ,
i bb Hi t t Hi t t
2MW 2MW
i=1
(A.1)
with [8]6

1 + (δhb / hb ) O1i
S
gH = Re
i bb 1 + (δhb / hb ) + (+hb / hb ) tan β cos β

(+hb / hb ) O2i
+ Re
1 + (δhb / hb ) + (+hb / hb ) tan β cos β

(+hb / hb )(tan2 β + 1)
+ Im O3i , (A.2)
1 + (δhb / hb ) + (+hb / hb ) tan β

[1 + (δhb / hb )] tan β − (+hb / hb )

P
gH i bb
= − Re O3i
1 + (δhb / hb ) + (+hb / hb ) tan β

(+hb / hb ) tan β O1i
+ Im

(+hb / hb ) O2i
− Im , (A.3)

1 + (δht / ht ) O2i
S
gH = Re
itt
1 + (δht / ht ) + (+ht / ht ) cot β sin β

(+ht / ht ) O1i
+ Re
6 Here, we have also used the fact that b- and t-quark masses are positive, i.e., Im m ∝ Im[h + (δh ) +
b b b
(+hb ) tan β] = 0 and Im mt ∝ Im[ht + (δht ) + (+ht ) cot β] = 0.

(+ht / ht )(cot2 β + 1)
+ Im O3i , (A.4)
1 + (δht / ht ) + (+ht / ht ) cot β

[1 + (δht / ht )] cot β − (+ht / ht )

P
gH itt
= − Re O3i
1 + (δht / ht ) + (+ht / ht ) cot β

(+ht / ht ) cot β O2i
+ Im

(+ht / ht ) O1i
− Im . (A.5)
In (A.2)–(A.5), O is a 3 × 3-dimensional mixing matrix that relates weak to mass
eigenstates of Higgs bosons in the presence of CP violation [6,8], and δht,b / ht,b and
+ht,b / ht,b represent non-logarithmic threshold contributions to bottom and top Yukawa
couplings [46]. As is shown in Fig. 9, the latter quantities are predominantly induced by
gluino and Higgsino loops. In the presence of CP violation, their analytic forms are [8]
δhb 2αs ∗ |ht |2

=− mg̃ Ab I m2b̃ , m2b̃ , |mg̃ |2 − 2
|µ|2 I m2t˜ , m2t˜ , |µ|2 , (A.6)
hb 3π 1 2 16π 1 2
+hb 2αs ∗ ∗ 2 |ht |2 ∗ ∗ 2 2

= mg̃ µ I mb̃ , m2b̃ , |mg̃ |2 + A µ I mt˜ , mt˜ , |µ|2 , (A.7)
hb 3π 1 2 16π 2 t 1 2
+ht 2αs ∗ ∗ 2 2 |hb |2 ∗ ∗ 2

= mg̃ µ I mt˜ , mt˜ , |mg̃ |2 + 2
Ab µ I mb̃ , m2b̃ , |µ|2 , (A.8)
ht 3π 1 2 16π 1 2
δht 2αs ∗ |hb |2 2 2

=− mg̃ At I m2t˜ , m2t˜ , |mg̃ |2 − 2
|µ| I mb̃ , m2b̃ , |µ|2 , (A.9)
ht 3π 1 2 16π 1 2
where αs = gs2 /(4π) is the SU(3)c fine structure constant, and I (a, b, c) is the one-loop
function
ab ln(a/b) + bc ln(b/c) + ac ln(c/a)
I (a, b, c) = . (A.10)
(a − b)(b − c)(a − c)
In addition, the stop and sbottom masses are given by (with q = t, b)
Fig. 9. Effective one-loop =01,2 b̄b and =01,2 t¯t couplings, δhb,t and +hb,t , generated by the exchange of
(a) gluinos g̃ and (b) Higgsinos h̃±
1,2 .
m2q̃1 (q̃2 )
1 2 2 q
= M Q + Mq + 2mq + Tz cos 2βMZ
2 2
2

+ (−) M q2 + cos 2βM 2 Tzq − 2Qq sin2 θw 2 + 4m2q |Aq − Rq µ∗ |2 ,
2 − M
Q Z
(A.11)
where Qt (Qb ) = 2/3(−1/3), Tzt = −Tzb = 1/2, Rt (Rb ) = cot β(tan β), and sin2 θw =
1 − MW 2 /M 2 .
Z
It is important to remark here that only the CP-violating vertex effects on gH S and
i bb
P 2
gHi bb , which are proportional to Im[(+hb / hb ) tan β] in (A.2) and (A.3), are enhanced for
moderately large values of tan β, i.e., 20 tan β 40. However, for very large values of
tan β, i.e., tan β 40, there is a 1/ tan2 β-dependent damping factor due to CP-violating
resummation effects which cancels the tan2 β-enhanced factor mentioned above. As a
consequence, in the large-tan β limit, the coupling factors gH S P
and gH approach a
i bb i bb
tan β-independent constant. A related discussion is also given in Section 2.
Another important ingredient for our computation of two-loop EDMs is the diagonal
effective couplings of the Higgs bosons to scalar top and bottom quarks. Taking the CP-
violating Higgs-mixing effects into account, the effective Lagrangian of interest to us may
be conveniently written in the form

3
vξq(Hi ) q̃1∗ q̃1 − q̃2∗ q̃2 + vζq(Hi ) q̃1∗ q̃1 + q̃2∗ q̃2 ,
diag
LH q̃ ∗ q̃ = Hi (A.12)
i=1 q=t,b
where

(H ) 2m2t O3i O1i ∗ O2i
ξt i = Im(µAt ) 2 − Re(µXt ) + Re(At Xt ) ,
v 2 (m2t˜ − m2t˜ ) sin β sin β sin β
2 1
(A.13)
2
2m O 3i O 2i ∗ O 1i
ξb(Hi ) = b
Im(µA b ) − Re(µX b ) + Re(A b X b ) ,
v 2 (m2 − m2 ) cos2 β cos β cos β
b̃2 b̃1
(A.14)
2
2m O2i 2
2m O1i
, g 2 , , g 2 , (A.15)
(H ) (H )
ζt i = − 2 t + O gw 2
ζb i = − 2 b + O gw 2
v sin β v cos β
with Xq = Aq − Rq µ∗ (q = t, b). Although we assumed m2q̃ > m2q̃ , the effective
1 2
Lagrangian (A.12) exhibits the nice feature that it is fully independent of the hierarchy
of squark masses.
Finally, we present the effective couplings of the CP-mixed Higgs bosons H1,2,3 to
+
charginos χ1,2 [7,47]. These may be conveniently described by the effective Lagrangian
gw
3
LH χ + χ − = − √ Hi χ̄j+ (aHi χ − χ + + bHi χ − χ + iγ5 )χk+ , (A.16)
2 2 i=1 j,k=1,2 j k j k
where
R∗ L R∗ L
aHi χ − χ + = O1i C2j C1k + C2k C1j + O2i C1j
R L∗
C2k + C1k
R L∗
C2j
j k
R∗ L
− iO3i sin β C2j C1k − C2kR L∗
C1j
R∗ L
+ cos β C1j C2k − C1k
R L∗
C2j , (A.17)
R∗ L
bHi χ − χ + = iO1i C2j C1k − C2k C1j + iO2i C1j
R L∗
C2k − C1k
R∗ L R L∗
C2j
j k
R∗ L
+ O3i sin β C2j C1k + C2k C1j + cos β C1j
R L∗
C2k + C1k
R∗ L R L∗
C2j . (A.18)
In the above, C R and C L are 2 × 2 unitary matrices, which diagonalize the chargino mass
matrix:

mW gw φ20∗
MC = , (A.19)
gw φ10 µ
√ √
with φ10 = v1 / 2 and φ20∗ = v2 / 2, through the bi-unitary transformation
C R† MC C L = diag(mχ + , mχ + ). (A.20)
1 2
In (A.20), the chargino mass-eigenvalues are given by

1
mχ + (χ + ) = |m2W | + |µ| + 2MW
2 2
1 2 2
− (+) m2W
+ |µ|2 + 2M 2 2 − 4m µ − M 2 sin 2β 2 ,
W W W (A.21)
while the analytic expressions for the mixing matrices C L,R are quite lengthy in the
presence of CP violation, and will not be presented here; they can be computed using
standard techniques [7].
For completeness, we give the corresponding effective couplings of the would-be
+
Goldstone boson G0 to charginos χ1,2 :
R∗ L R∗ L
aG0 χ − χ + = i cos β C2j C1k − C2k C1j − i sin β C1j
R L∗
C2k − C1k
R L∗
C2j ,
j k
R∗ L R∗ L
bG0 χ − χ + = − cos β C2j C1k + C2k C1j + sin β C1j
R L∗
C2k + C1k
R L∗
C2j . (A.22)
j k
A non-trivial consistency check for the correctness of our analytic results is the vanishing
of the diagonal scalar couplings of the G0 boson to charginos, i.e., aG0 χ − χ + = 0.
j j
References
[1] G.F. Giudice, S. Dimopoulos, Phys. Lett. B 357 (1995) 573;

G. Dvali, A. Pomarol, Phys. Rev. Lett. 77 (1996) 3728;
A.G. Cohen, D.B. Kaplan, A.E. Nelson, Phys. Lett. B 388 (1996) 588;
P. Binétruy, E. Dudas, Phys. Lett. B 389 (1996) 503.
[2] D. Chang, W.-Y. Keung, A. Pilaftsis, Phys. Rev. Lett. 82 (1999) 900;
D. Chang, W.-Y. Keung, A. Pilaftsis, Phys. Rev. Lett. 83 (1999) 3972(E).
[3] A. Pilaftsis, Phys. Lett. B 471 (1999) 174;

D. Chang, W.-F. Chang, W.-Y. Keung, Phys. Lett. B 478 (2000) 239.
[4] S.M. Barr, A. Zee, Phys. Rev. Lett. 65 (1990) 21.
[5] A. Pilaftsis, Phys. Rev. D 58 (1998) 096010;
A. Pilaftsis, Phys. Lett. B 435 (1998) 88.
[6] A. Pilaftsis, C.E.M. Wagner, Nucl. Phys. B 553 (1999) 3;
D.A. Demir, Phys. Rev. D 60 (1999) 055006;
S.Y. Choi, M. Drees, J.S. Lee, Phys. Lett. B 481 (2000) 57;
G.L. Kane, L.-T. Wang, Phys. Lett. B 488 (2000) 383;
S.Y. Choi, K. Hagiwara, J.S. Lee, Phys. Rev. D 64 (2001) 032004;
S.Y. Choi, K. Hagiwara, J.S. Lee, Phys. Lett. B 529 (2002) 212;
S. Heinemeyer, Eur. Phys. J. C 22 (2001) 521;
T. Ibrahim, P. Nath, hep-ph/0204092;
S.W. Ham, S.K. Oh, E.J. Yoo, C.M. Kim, D. Son, hep-ph/0205244.
[7] T. Ibrahim, P. Nath, Phys. Rev. D 63 (2001) 035009.
[8] M. Carena, J. Ellis, A. Pilaftsis, C.E.M. Wagner, Nucl. Phys. B 586 (2000) 92.
[9] M. Carena, J. Ellis, A. Pilaftsis, C.E.M. Wagner, Phys. Lett. B 495 (2000) 155.
[10] M. Carena, J. Ellis, A. Pilaftsis, C.E.M. Wagner, Nucl. Phys. B 625 (2002) 345.
[11] For phenomenological applications of radiative Higgs-sector CP violation in the MSSM, see:
S.Y. Choi, M. Drees, Phys. Rev. Lett. 81 (1998) 5509;
S. Baek, P. Ko, Phys. Rev. Lett. 83 (1999) 488;
W. Bernreuther, A. Brandenburg, M. Flesch, hep-ph/9812387;
K.S. Babu, C. Kolda, J. March-Russell, F. Wilczek, Phys. Rev. D 59 (1999) 016004;
B. Grzadkowski, J.F. Gunion, J. Kalinowski, Phys. Rev. D 60 (1999) 075011;
S.Y. Choi, M. Guchait, H.S. Song, W.Y. Song, Phys. Lett. B 483 (2000) 168;
P. Osland, Acta Phys. Polon. B 30 (1999) 1967;
T. Han, T. Huang, Z.H. Lin, J.X. Wang, X. Zhang, Phys. Rev. D 61 (2000) 015006;
B. Grzadkowski, J. Pliszka, Phys. Rev. D 60 (1999) 115018;
S.Y. Choi, J.S. Lee, Phys. Rev. D 61 (2000) 015003;
C.-S. Huang, L. Wei, Phys. Rev. D 61 (2000) 116002;
K. Freese, P. Gondolo, hep-ph/9908390;
S.Y. Choi, J.S. Lee, Phys. Rev. D 61 (2000) 115002;
A. Dedes, S. Moretti, Phys. Rev. Lett. 84 (2000) 22;
A. Dedes, S. Moretti, Nucl. Phys. B 576 (2000) 29;
M.S. Berger, Phys. Rev. Lett. 87 (2001) 131801;
D.A. Demir, Nucl. Phys. Proc. Suppl. 101 (2001) 431;
S.W. Ham, S.K. Oh, E.J. Yoo, H.K. Lee, J. Phys. G 27 (2001) 1;
A.G. Akeroyd, A. Arhrib, Phys. Rev. D 64 (2001) 095018;
M. Boz, N.K. Pak, Phys. Rev. D 65 (2002) 075014;
A. Arhrib, D.K. Ghosh, O.C.W. Kong, Phys. Lett. B 537 (2002) 217;
E. Christova, S. Fichtinger, S. Kraml, W. Majerotto, Phys. Rev. D 65 (2002) 094002, hep-ph/0205227;
C.-H. Chen, hep-ph/0206143;
M.N. Dubinin, A.V. Semenov, hep-ph/0206205;
A. Bartl, K. Hidaka, T. Kernreiter, W. Porod, hep-ph/0207186.
[12] V.A. Kuzmin, V.A. Rubakov, M.E. Shaposhnikov, Phys. Lett. B 155 (1985) 36.
[13] For recent studies, see:
M. Carena, M. Quirós, C.E.M. Wagner, Nucl. Phys. B 524 (1998) 3;
M. Laine, K. Rummukainen, Phys. Rev. Lett. 80 (1998) 5259;
M. Laine, K. Rummukainen, Nucl. Phys. B 535 (1998) 423;
M. Laine, K. Rummukainen, Nucl. Phys. B 597 (2001) 23;
K. Funakubo, Prog. Theor. Phys. 102 (1999) 389;
J. Grant, M. Hindmarsh, Phys. Rev. D 59 (1999) 116014;
M. Losada, Nucl. Phys. B 537 (1999) 3;
M. Losada, Nucl. Phys. B 569 (2000) 125;
S. Davidson, M. Losada, A. Riotto, Phys. Rev. Lett. 84 (2000) 4284;

A.B. Lahanas, V.C. Spanos, V. Zarikas, Phys. Lett. B 472 (2000) 119;
N. Rius, V. Sanz, Nucl. Phys. B 570 (2000) 155;
J.M. Cline, M. Joyce, K. Kainulainen, JHEP 0007 (2000) 018;
J.M. Cline, M. Joyce, K. Kainulainen, hep-ph/0110031, Erratum;
S.J. Huber, P. John, M.G. Schmidt, Eur. Phys. J. C 20 (2001) 695;
H. Murayama, A. Pierce, hep-ph/0201261.
[14] M. Carena, J.M. Moreno, M. Quirós, M. Seco, C.E.M. Wagner, Nucl. Phys. B 599 (2001) 158, and work in
progress.
[15] P.G. Harris, et al., Phys. Rev. Lett. 82 (1999) 904.
[16] B.C. Regan, E.D. Commins, C.J. Schmidt, D. DeMille, Phys. Rev. Lett. 88 (2002) 071805.
[17] S. Barr, Phys. Rev. Lett. 68 (1992) 1822;
S. Barr, Int. J. Mod. Phys. A 8 (1993) 209.
[18] W. Fischler, S. Paban, S. Thomas, Phys. Lett. B 289 (1992) 373.
[19] I.B. Khriplovich, S.K. Lamoreaux, CP Violation Without Strangeness, Springer, New York, 1997.
[20] Y.K. Semertzidis, et al., Letter of Intent to BNL, Spring of 2000, and Fall of 1997;
Y.K. Semertzidis, et al., Sensitive search for a permanent muon electric dipole moment, hep-ph/0012087;
For recent theoretical studies, see:
K.S. Babu, B. Dutta, R.N. Mohapatra, Phys. Rev. Lett. 85 (2000) 5064;
T. Ibrahim, P. Nath, Phys. Rev. D 64 (2001) 093002;
J.L. Feng, K.T. Matchev, Y. Shadmi, Nucl. Phys. B 613 (2001) 366;
J.R. Ellis, J. Hisano, M. Raidal, Y. Shimizu, Phys. Lett. B 528 (2002) 86.
[21] K.F. Smith, et al., Phys. Lett. B 234 (1990) 191.
[22] P.G. Harris, et al., Phys. Rev. Lett. 82 (1999) 904.
[23] S.K. Lamoreaux, R. Golub, Phys. Rev. D 61 (2000) 051301.
[24] M. Pospelov, A. Ritz, Phys. Rev. Lett. 83 (1999) 2526;
M. Pospelov, A. Ritz, Nucl. Phys. B 573 (2000) 177.
[25] S.A. Abel, S. Khalil, O. Lebedev, Nucl. Phys. B 606 (2001) 151.
[26] M.V. Romalis, W.C. Griffith, E.N. Fortson, Phys. Rev. Lett. 86 (2001) 2505.
[27] T. Falk, K.A. Olive, M. Pospelov, R. Roiban, Nucl. Phys. B 60 (1999) 3.
[28] A. Ritz, private communication.
[29] S. Weinberg, Phys. Rev. Lett. 63 (1989) 2333.
[30] J. Dai, H. Dykstra, R.G. Leigh, S. Paban, D.A. Dicus, Phys. Lett. B 237 (1990) 216;
J. Dai, H. Dykstra, R.G. Leigh, S. Paban, D.A. Dicus, Phys. Lett. B 242 (1990) 547(E).
[31] T. Ibrahim, P. Nath, Phys. Lett. B 418 (1998) 98;
T. Ibrahim, P. Nath, Phys. Rev. D 58 (1998) 019901(E);
M. Brhlik, G.J. Good, G.L. Kane, Phys. Rev. D 59 (1999) 115004;
M. Brhlik, L. Everett, G.L. Kane, J. Lykken, Phys. Rev. Lett. 83 (1999) 2124;
M. Brhlik, L. Everett, G.L. Kane, J. Lykken, Phys. Rev. D 62 (2000) 035005;
S. Abel, S. Khalil, O. Lebedev, Phys. Rev. Lett. 86 (2001) 5850.
[32] J. Ellis, S. Ferrara, D.V. Nanopoulos, Phys. Lett. B 114 (1982) 231;
W. Buchmüller, D. Wyler, Phys. Lett. B 121 (1983) 321;
J. Polchinski, M. Wise, Phys. Lett. B 125 (1983) 393;
F. del Aguila, M. Gavela, J. Grifols, A. Mendez, Phys. Lett. B 126 (1983) 71;
D.V. Nanopoulos, M. Srednicki, Phys. Lett. B 128 (1983) 61;
T. Falk, K.A. Olive, M. Srednicki, Phys. Lett. B 354 (1995) 99;
M. Dugan, B. Grinstein, L. Hall, Nucl. Phys. B 255 (1985) 413;
R. Garisto, J.D. Wells, Phys. Rev. D 55 (1997) 1611.
[33] P. Nath, Phys. Rev. Lett. 66 (1991) 2565;
Y. Kizukuri, N. Oshimo, Phys. Rev. D 46 (1992) 3025.
[34] For recent analyses of one-loop EDM effects, see:

S. Pokorski, J. Rosiek, C.A. Savoy, Nucl. Phys. B 570 (2000) 81;
E. Accomando, R. Arnowitt, B. Dutta, Phys. Rev. D 61 (2000) 115003;
A. Bartl, T. Gajdosik, W. Porod, P. Stockinger, H. Stremnitzer, Phys. Rev. D 60 (1999) 073003.
[35] In this respect, our analysis of EDM constraints on a CP-violating MSSM Higgs sector is complementary to
the one presented by T. Ibrahim, Phys. Rev. D 64 (2001) 035009.
[36] A. Pilaftsis, Phys. Rev. D 62 (2000) 016007.
[37] T.H. West, Phys. Rev. D 50 (1994) 7;
T. Kadoyoshi, N. Oshimo, Phys. Rev. D 55 (1997) 1481.
[38] In our numerical analysis, we employ the Fortran code cph+.f available at http://home.cern.ch/p/
pilaftsi/www/. The code is based on a recent RG-improved diagrammatic calculation of Higgs-boson pole
masses in the MSSM with explicit CP violation [10].
[39] For recent two-loop studies of an effective CP-conserving Higgs potential in the MSSM, see:
M. Carena, H.E. Haber, S. Heinemeyer, W. Hollik, C.E.M. Wagner, G. Weiglein, Nucl. Phys. B 580 (2000)
29;
J.R. Espinosa, R.J. Zhang, JHEP 0003 (2000) 026;
M. Carena, S. Mrenna, C.E.M. Wagner, Phys. Rev. D 62 (2000) 055008;
J.R. Espinosa, I. Navarro, Nucl. Phys. B 615 (2001) 82;
A. Brignole, G. Degrassi, P. Slavich, F. Zwirner, Nucl. Phys. B 631 (2002) 195, hep-ph/0206101;
S.P. Martin, hep-ph/0206136.
[40] O. Lebedev, M. Pospelov, hep-ph/0204359.
[41] D. Chang, W.-F. Chang, W.-Y. Keung, hep-ph/0205084.
[42] M.A. Shifman, A.L. Vainshtein, V.I. Zakharov, Phys. Lett. B 78 (1978) 443;
See also, J. Ellis, M.K. Gaillard, D.V. Nanopoulos, Nucl. Phys. B 106 (1976) 292;
S. Dawson, H.E. Haber, Int. J. Mod. Phys. A 7 (1992) 107.
√ √
[43] OPAL Collaboration, Interpretation of the search for neutral Higgs bosons from s = 91 GeV to s =
209 GeV in a CP-violating MSSM scenario, OPAL Physics Note PN505, available from http://opal.web.
cern.ch/Opal/pubs/physnote/html/pn505.html.
[44] M. Carena, J. Ellis, S. Mrenna, A. Pilaftsis, C.E.M. Wagner, in preparation.
[45] D. Chang, W.-F. Chang, C.-H. Chou, W.-Y. Keung, Phys. Rev. D 63 (2001) 091301;
A. Dedes, H.E. Haber, JHEP 0105 (2001) 006.
[46] E. Ma, Phys. Rev. D 39 (1989) 1922;
R. Hempfling, Phys. Rev. D 49 (1994) 6168;
L.J. Hall, R. Rattazzi, U. Sarid, Phys. Rev. D 50 (1994) 7048;
T. Blazek, S. Raby, S. Pokorski, Phys. Rev. D 52 (1995) 4151;
M. Carena, M. Olechowski, S. Pokorski, C.E.M. Wagner, Nucl. Phys. B 426 (1994) 269;
S. Heinemeyer, W. Hollik, G. Weiglein, Phys. Lett. B 440 (1998) 296;
S. Heinemeyer, W. Hollik, G. Weiglein, Eur. Phys. J. C 9 (1999) 343;
F. Borzumatti, G. Farrar, N. Polonsky, S. Thomas, Nucl. Phys. B 555 (1999) 53;
J.R. Espinosa, R.J. Zhang, Nucl. Phys. B 586 (2000) 3.
[47] S.Y. Choi, M. Drees, J.S. Lee, J. Song, hep-ph/0204200.
Heavy triplet leptons and new gauge boson

Ernest Ma a , D.P. Roy a,b
a Physics Department, University of California, Riverside, CA 92521, USA
b Tata Institute of Fundamental Research, Mumbai (Bombay) 400005, India
Received 5 August 2002; accepted 6 September 2002
Abstract
A heavy triplet of leptons (Σ + , Σ 0 , Σ − )R per family is proposed as the possible anchor of a
small seesaw neutrino mass. A new U (1) gauge symmetry is then also possible, and the associated
gauge boson X may be discovered at or below the TeV scale. We discuss the phenomenology of
this proposal, with and without possible constraints from the NuTeV and atomic parity violation
experiments, which appear to show small discrepancies from the predictions of the standard model.
PACS: 12.60.Cn; 14.60.St; 13.15.+g; 14.60.Hi
1. Introduction
To obtain nonzero neutrino masses so as to explain the observed atmospheric [1] and
solar [2] neutrino oscillations, the minimal standard model of particle interactions is often
extended to include three neutral fermion singlets, usually referred to as right-handed
singlet neutrinos. If they have large Majorana masses, then the famous seesaw mechanism
[3] allows the observed neutrinos to acquire naturally small Majorana masses. On the other
hand, there are other equivalent ways [4,5] to realize this effective dimension-five operator
[6] for neutrino mass. For example, if we replace each neutral fermion singlet by a triplet
[5,7]:

Σ = Σ + , Σ 0 , Σ − ∼ (1, 3, 0) (1)
under SU(3)C × SU(2)L × U (1)Y , the seesaw mechanism works just as well.
If the Majorana mass of Σ is very large, then its effect at low energies is indistin-
guishable from that of the canonical seesaw. On the other hand, if it is at or below the
E-mail address: ma@phyun8.ucr.edu (E. Ma).
PII: S 0 5 5 0 - 3 2 1 3 ( 0 2 ) 0 0 8 1 5 - 5
E. Ma, D.P. Roy / Nuclear Physics B 644 (2002) 290–302 291
TeV energy scale, which is a natural possibility if there exists a second Higgs doublet, as
shown recently [8], then there are interesting new experimental signatures for the origin of
neutrino mass. In Section 2, the phenomenology of this scenario is discussed.
It is well known [9] that in the case of one additional right-handed singlet neutrino per
family of quarks and leptons, it is possible to promote B − L (baryon number – lepton
number) from being a global U (1) symmetry to an U (1) gauge symmetry. Similarly a new
U (1)X gauge symmetry [10] is also possible here. The model is described in Section 3.
Since the X gauge boson may be at or below the TeV scale, it may be responsible
for some of the possible discrepancies observed in recent experiments. In Section 4, we
use it to explain the NuTeV result [11] and explore its phenomenological implications. In
Section 5, we do the same but using atomic parity nonconservation [12] as a constraint. In
Section 6, the Higgs sector is discussed and its difference from other proposals is noted.
We then conclude in Section 7.
2. Heavy triplet leptons
Instead of the usual singlet NR , consider the addition of a fermion triplet

(Σ + , Σ 0 , Σ − )R per family to the particle content of the standard model. The Yukawa
interaction

1
LY = f −φ − ν̄L ΣR+ + √ φ̄ 0 ν̄L + φ − ēL ΣR0 + φ̄ 0 ēL ΣR− + h.c. (2)
2
means that a seesaw neutrino mass matrix is obtained, i.e.,
fv
0 √
MνΣ = f v 2
, (3)
√ MΣ
2
together with e–Σ mixing [7], i.e.,

me f v
MeΣ = . (4)
0 MΣ
For v = φ 0 = 174 GeV as in the standard model, either MΣ has to be very large or
f very small for this to give realistic neutrino masses. However, if a symmetry exists
which replaces Φ in Eq. (2) with a second Higgs doublet η = (η+ , η0 ) having a very small
vacuum expectation value, then MΣ may be at or even below the TeV scale. Such a model
has already been described [8], using NR . It is straightforward to apply it here in the context
of ΣR .
The idea is very simple. Assign lepton number L = −1 to η but L = 0 to Φ and Σ, then
MΣ is an allowed Majorana mass, and Φ is replaced with η in Eq. (2). Let L be broken by
explicit soft terms in the Lagrangian, i.e., µ212 Φ † η + h.c., then for m2η > 0 and large,
0 µ2 v
η = u
− 122 . (5)
mη
For µ212 ∼ 10 GeV2 , v ∼ 102 GeV, mη ∼ 1 TeV, we get u ∼ 1 MeV and mν
f 2 u2 /2MΣ ∼ 1 eV or less if MΣ ∼ 1 TeV. Note that after the explicit breaking of L,

292 E. Ma, D.P. Roy / Nuclear Physics B 644 (2002) 290–302
a residual symmetry is still conserved, i.e., the conventional multiplicative lepton number,
where ν, e, and Σ are odd, but Φ and η are even. In other words, there are no unwanted
L = 1 interactions even though η0 = 0.
Whereas Σ 0 has no coupling to either the photon or the Z boson (as is the case with
NR ), Σ ± interacts with both. Hence our proposal is more easily tested experimentally than
the canonical seesaw. If the mass of Σ ± is below that of η, the former decays only via its
mixing with e± . Thus we expect the decay modes
Σi+ → ej+ Z, ν̄j W + , Σi− → ej− Z, νj W − , (6)

which would map out the Yukawa coupling matrix fij , leading to a good determination of
the neutrino mass matrix itself [8], up to an overall scale. On the other hand, if MΣ > mη ,
then the decays
Σi+ → ej+ η0 , ν̄j η+ (7)
will also determine fij . The subsequent decays of η+ and η0 occur through their small
mixings with φ + and φ 0 , so they are dominated by t b̄ and t t¯ final states and should be
easily identifiable.
Since Σ and η have distinctive signatures once they are produced, their discoveries are
primarily controlled by the size of the signal. We have estimated their pair production cross
sections at the LHC and at the Tevatron via the standard Drell–Yan mechanism. The spin
and color averaged matrix element squares are given by

2
2
1 Q q L q Lη Q q R q Lη
M 2 q q̄→η+ η− = e4 ut − m4η + + + , (8)
3 s s − MZ2 s s − MZ2
where
1
− sin2 θW Iq3 − Qq sin2 θW −Qq sin2 θW
Lη = 2
, Lq = , Rq = . (9)
sin θW cos θW sin θW cos θW sin θW cos θW
The analogous matrix element squares for Σ ± pair production are

1
2 2 Qq Lq RΣ 2
M q q̄→Σ + Σ − = e t − MΣ
2 4
+
3 s s − MZ2

2 2 Qq Rq RΣ 2
+ u − MΣ + , (10)
s s − MZ2
where
1 − sin2 θW
RΣ = . (11)
sin θW cos θW
Fig. 1 shows the LHC and Tevatron production cross sections of the heavy scalar pair η±
and Fig. 2 shows those of the heavy lepton pair Σ ± as functions of their mass. We see
from these figures that the final luminosity of about 300 fb−1 at the LHC will correspond
to a modest discovery limit of both η± and Σ ± up to a mass of about 1 TeV.
Fig. 1. Cross sections for η± pair production at LHC and Tevatron.
Fig. 2. Cross sections for Σ ± pair production at LHC and Tevatron.

3. New gauge boson
Consider SU(3)C × SU(2)L × U (1)Y × U (1)X as a possible extension of the standard

model, under which each family of quarks and leptons transforms as follows:
(u, d)L ∼ (3, 2, 1/6; n1), uR ∼ (3, 1, 2/3; n2), dR ∼ (3, 1, −1/3; n3),
(ν, e)L ∼ (1, 2, −1/2; n4), eR ∼ (1, 1, −1; n5), ΣR ∼ (1, 3, 0; n6). (12)
It has been shown recently [10] that U (1)X is free of all anomalies [13–15] for the
following assignments:
1 1
n2 = (7n1 − 3n4 ), n3 = (n1 + 3n4 ),
4 4
1 1
n5 = (−9n1 + 5n4 ), n6 = (3n1 + n4 ). (13)
4 4
This is a remarkable and highly nontrivial result.
As shown in Ref. [10], there are 6 conditions to be satisfied for the gauging of U (1)X .
Three of them do not involve n6 and have 2 solutions:
(I): n3 = 2n1 − n2 , n4 = −3n1 , n5 = −2n1 − n2 , (14)

1 1 1
(II): n2 = (7n1 − 3n4 ), n3 = (n1 + 3n4 ), n5 = (−9n1 + 5n4 ). (15)
4 4 4
The other 3 involve n6 , and they are given by
1 1
(3n1 + n4 ) = p(p + 1)(2p + 1)n6 , (16)
2 3
6n1 − 3n2 − 3n3 + 2n4 − n5 = (2p + 1)n6 , (17)
6n31 − 3n32 − 3n33 + 2n34 − n35 = (2p + 1)n36 , (18)
where an extra right-handed fermion multiplet transforming as (1, 2p + 1, 0; n6 ) has been
added to each family of quarks and leptons.
To find solutions to the above 3 equations, consider first p = 0, then Eq. (16) forces
one to choose solution (I), and all other equations are satisfied with n1 = n2 = n3 and
n4 = n5 = n6 , i.e., U (1)B−L has been obtained. Consider now p = 0, then if solution (I) is
again chosen, n6 = 0 is required, which leads to U (1)Y , so there is nothing new.
Now consider p = 0 and solution (II). From Eqs. (16), (17), and (18), it is easily shown
that

1/3
4n6 6 3 3
= = = , (19)
3n1 + n4 p(p + 1)(2p + 1) 2p + 1 2p + 1
which clearly gives the unique solution of p = 1, i.e., a triplet. The fact that such a solution
even exists (and for an integer value of p) for the above overconstrained set of conditions
is certainly not a “trivial” or even “expected” result.
The U (1)X charges of the possible Higgs doublets are:
3 1
n1 − n3 = n2 − n1 = n6 − n4 = (n1 − n4 ), n4 − n5 = (9n1 − n4 ), (20)
4 4
which means that two distinct Higgs doublets are sufficient for all possible Dirac fermion
masses in this model. If n4 = −3n1 is chosen, then again U (1)X will be proportional to
U (1)Y . However, for n4 = −3n1 , a new class of models is now possible with U (1)X as a
genuinely new gauge symmetry.
4. Scenario A: neutrino–quark scattering
Consider νq and ν̄q deep inelastic scattering. It has recently been reported [11] by
the NuTeV Collaboration that their measurement of the effective sin2 θW , i.e., 0.2277 ±
0.0013 ± 0.0009, is about 3σ away from the standard-model prediction of 0.2227 ±
0.00037. In this model, the X gauge boson also contributes with

µ 1 − γ5 1 − γ5
JX = n1 ūγ µ u + n1 d̄γ µ d
2 2

1 + γ5 1 + γ5 1 − γ5
+ n2 ūγ µ u + n3 d̄γ µ d + n4 ν̄γ µ ν. (21)
2 2 2
Assuming very small X–Z mixing (| sin θ | 1), the effective neutrino–quark interactions
are then given by
GF q q
Hint = √ ν̄γ µ (1 − γ5 )ν /L q̄γµ (1 − γ5 )q + /R q̄γµ (1 + γ5 )q , (22)
2
where

1 2 2
/L = (1 − ξ )
u
− sin θW + n1 ζ, (23)
2 3

1 1 2
/L = (1 − ξ ) − + sin θW + n1 ζ,
d
(24)
2 3

2 2
/R = (1 − ξ ) − sin θW n2 ζ,
u
(25)
3

1 2
/R = (1 − ξ )
d
sin θW + n3 ζ, (26)
3
with

M 2 gX
ξ = 2n4 sin θ 1 − Z2 , (27)
MX gZ

2
2
MZ2 gX MZ gX
ζ = − sin θ 1 − 2 + 2n4 . (28)
MX gZ MX2 gZ2
The parameter ξ is constrained by data at the Z resonance to be very small. Using the
general analysis of Z–X mixing [16], we find
2s 2 c2 (c2 − s 2 )2 s 2 (−1 − 2s 2 + 4s 4 )
ξ= /1 + /2 + 2 /3
c −s
2 2 2c 2 c c2 − s 2
= 0.624/1 + 0.198/2 − 0.644/3, (29)
where s ≡ sin θW and c ≡ cos θW . Given that |/i | is of order 0.001, ξ is too small to make
much difference in the above [17]. We thus assume ξ = 0 (sin θ = 0) for our subsequent
discussion.
To account for the NuTeV result, i.e.,
eff 2 u 2 d 2
gL = /L + /L = 0.3005 ± 0.0014, (30)
eff 2 u 2 d 2
gR = /R + /R = 0.0310 ± 0.0011, (31)
against the standard-model prediction, i.e.,
eff 2 eff 2
gL SM = 0.3042, gR SM = 0.0301, (32)
consider the following specific model as an illustration:
3 5
n1 = 1, n2 = , n3 = ,
4 4
4 7 13
n4 = , n5 = − , n6 = . (33)
3 12 12
Then
2 2
∆ gLeff = − sin2 θW ζ + 2ζ 2 , (34)
3
eff 2 1 17
∆ gR = − sin2 θW ζ + ζ 2 . (35)
6 8
To fit the experimental values, we need a negative ∆(gLeff )2 . From Eq. (34) we see that it
reaches its maximum value at
1
ζ = sin2 θW , (36)
6
for which
2 1
∆ gLeff = − sin4 θW = −0.0028, (37)
18
2 1
∆ gReff = sin4 θW = +0.0016, (38)
32
in very good agreement with the experimental values of −0.0037 ± 0.0014 and +0.0009 ±
0.0011, respectively.
Using Eqs. (28), (33), and (36), we find that
2
gX sin2 θW gZ2
= . (39)
MX2 16 MZ2
Thus the production of the new gauge boson X may be studied as a function of the single
parameter MX in this scenario. We note first that if the U (1)X assignments of Eq. (33)
apply to electrons as well, then Eq. (39) is in serious conflict with atomic parity violation
and e+ e− cross sections. We must therefore attribute the NuTeV anomaly as being due to
the muon (and perhaps also the tau) sector [17,18] only. In the context of U (1)X , this may
be accomplished as follows. We change the electron’s assignments under U (1)X to zero so
that it does not couple to X at all. To preserve the cancellation of anomalies, we add heavy
fermions at the TeV energy scale, i.e.,
(N, F )L ∼ (1, 2, −1/2; n4), (N, F )R ∼ (1, 2, −1/2; 0), (40)
EL ∼ (1, 1, −1; 0), ER ∼ (1, 1, −1; n5). (41)
These are prevented from coupling to the known leptons by a discrete symmetry to forbid
terms such as EL eR , etc. As a result, the lightest among them is stable, in analogy to the
lightest supersymmetric particle of R-parity conserving supersymmetry.
The spin and color averaged matrix element square for the X boson signal at the
Tevatron and at the LHC is given by
gX4
1
M 2 q q̄→X→f f¯ =
3 (s − MX2 )2 + MX2 6X
2
2 2 2
× nqL u nfL + t 2 n2fR + n2qR u2 n2fR + t 2 n2fL , (42)
where
2
gX
6X = MX 18n21 + 9n22 + 9n23 + 4n24 + 2n25 . (43)
24π
Substituting the required value of gX from Eq. (39) and the X charges (ni ) from Eq. (33)
we see that for MX > 1 TeV, gX > 1 and its width becomes comparable to its mass. Fig. 3
shows the total X boson production cross sections at the LHC and at the Tevatron as
Fig. 3. Cross sections for X production at LHC and Tevatron in Scenario A.

functions of its mass. We see that the predicted signal cross sections are really large if
the X boson is to account for the NuTeV anomaly. It may be noted here that there is a 95%
confidence-level upper limit of

σ (X)B X → e+ e− , µ+ µ− = 40 fb (44)
from the CDF experiment [19] at the Tevatron. The X charges of Eq. (33) correspond to a
branching fraction B(X → µ+ µ− ) = 4–5%. Thus assuming the CDF detection efficiency
to be roughly similar for the e+ e− and µ+ µ− channels, the above constraint would imply
σ (X) < 1000 fb at the Tevatron. On the other hand we see from Fig. 3 that σ (X) > 1000 fb
at the Tevatron right up to MX = 2 TeV (the finite value at the kinematic boundary is due
to the large width). Thus consistency with this limit will require MX > 2 TeV. Fig. 3 also
shows a very large signal cross section at the LHC up to MX = 3 TeV, corresponding to
gX
3. It remains large at larger values of MX as well, the cutoff being provided by the
perturbation theory limit on gX .
5. Scenario B: atomic parity violation
Instead of trying to accommodate the NuTeV discrepancy, we now consider the

possibility of having a small effect in atomic parity violation from U (1)X . Using a recently
reported precise atomic calculation of the 6s–7s parity violating E1 transition in cesium
[20], a slight deviation from the standard-model prediction is obtained, i.e.,
QW = 0.91 ± 0.29 ± 0.36. (45)

In the U (1)X model, with the definition r ≡ n1 /n4 , we have [21]
2 M2
3 gX
QW = −376C1u − 422C1d = Z
(1041r + 23)(1 − 9r). (46)
4 MX2 gZ2
This shows that there is only a narrow range for r which yields a positive value of QW .
For illustration, we choose r = 0, i.e.,
3 3 5 1
n1 = 0, n2 = − , n3 = , n4 = 1, n5 = , n6 = . (47)
4 4 4 4
From Eqs. (45) and (46), we then find
2 M2
gX Z
0.015 < < 0.09. (48)
MX2 gZ2
Since this range of values will lead to large effects in the muon sector if X couples in the
same way, we assume in this scenario that X couples only to electrons and quarks. By
doing so, we also avoid serious constraints from e+ e− → µ+ µ− , τ + τ − measurements at
LEP2 [22,23].
We have estimated the X boson production cross sections at the Tevatron and at the LHC
using Eq. (42) for the lower limit of the X coupling from Eq. (48), which is practically
identical to that of Eq. (39). They are shown in Fig. 4. They are somewhat smaller than
Fig. 4. Cross sections for X production at LHC and Tevatron in Scenario B.
those of Fig. 3 due to the different X charges. On the other hand, we expect a high
B(X → e+ e− )
18% in this case. Thus the CDF limit of Eq. (44) would again imply
MX > 2 TeV. Nonetheless we expect a very large signal cross section at the LHC up to a
mass range of several TeV, till the value of gX is again cut off by the perturbation theory
limit.
Note added: after we have completed our calculations, a new report [24] appeared which
shifted the central value of QW to 0.38, but with the same uncertainties. Since our choice
of the lower limit of Eq. (48) corresponds to it being 0.33, it is also consistent with this
new result.
6. Higgs sector
As shown in Section 3, U (1)X requires two distinct Higgs doublets for fermion masses,
i.e., Φ1 = (φ1+ , φ10 ) with U (1)X charge (9n1 − n4 )/4 which couples to charged leptons, and
Φ2 = (φ2+ , φ20 ) with U (1)X charge (3n1 − 3n4 )/4 which couples to up and down quarks
as well as to Σ. [The leptonic Higgs doublet η of Section 2 may also be introduced so
that only it couples to Σ, while Φ2 couples only to quarks.] To break the U (1)X gauge
symmetry spontaneously, we add a singlet χ with U (1)X charge −2n6 , so that the Yukawa
term χΣΣ would allow Σ to acquire a large Majorana mass at the U (1)X breaking scale.
The Higgs potential of this model is then given by

V = m21 Φ1† Φ1 + m22 Φ2† Φ2 + m23 χ † χ + f χ † Φ1† Φ2 + h.c.
1 2 1 2 1 2
+ λ1 Φ1† Φ1 + λ2 Φ2† Φ2 + λ3 χ † χ + λ4 Φ1† Φ1 Φ2† Φ2
2 2 2
† † † †
+ λ5 Φ1 Φ2 Φ2 Φ1 + λ6 χ χ Φ1 Φ1 + λ7 χ † χ Φ2† Φ2 . (49)
Note that the U (1)X charges allow the trilinear term χ † Φ1† Φ2 , without which V would
have 3 global U (1) symmetries, but only 2 U (1) gauge symmetries, resulting in an
unwanted Goldstone boson. We have not included η because m2η is large and positive as
discussed in Section 2. After the heavy χ [with χ ∼ 1 TeV] has been integrated out,
the reduced two-doublet Higgs potential is of the usual form. The difference from other
proposals is in their Yukawa couplings, i.e., Φ1 couples to charged leptons whereas Φ2
couples to both up and down quarks.
Let us briefly discuss the distinctive phenomenological features of this two-Higgs-
doublet model. While the vacuum expectation value φ20 is required to be ∼ 100 GeV
because of mt , φ10 can be anywhere between ∼ 100 GeV and 1–2 GeV. In terms of
the ratio tan β = φ20 /φ10 , they correspond to the limits tan β
1 and tan β 1. In
the former case, the phenomenological implications are similar to those of the standard
two-Higgs-doublet scenario where Φ2 couples to the up quarks and Φ2 couples to the
charged leptons as well as to the down quarks. In the latter case however, there are
distinctive differences between our proposal and the standard scenario, because Higgs
couplings proportional to mb are now multiplied by cot β instead of tan β. Let us
consider in particular the charged and the pseudoscalar neutral Higgs bosons, H ± and
A0 , which correspond to the linear combination Φ1 sin β − Φ2 cos β. Their distinctive
phenomenological features are summarized below.
(i) The H − t loop contribution to the b → sγ decay amplitude is suppressed. It has the
factor cot2 β instead of tan β cot β = 1 in the standard scenario. This means that the
charged Higgs boson in this model can be relatively light.
(ii) The H − → τ − ν̄ decay dominates over H − → t¯b for any charged Higgs mass.
(iii) The H ± production via t b̄ fusion is no longer the main production mechanism at
hadron colliders. It will instead be pair production via the Drell–Yan mechanism
discussed in Section 2. The plots of Fig. 1 apply equally to H ± pair production in
this case. Thus we expect a visible signal at the LHC up to a Higgs mass of about
1 TeV.
(iv) The A0 → τ + τ − decay dominates over A0 → b b̄ and A0 → t t¯.
(v) The A0 production is again no longer dominated by bb̄ or t t¯ fusion at hadron colliders,
but rather by associated production of A0 H 0 (H ± ) through Z(W ± ) exchange.
There are analogous distinctions for the two physical neutral scalars h0 and H 0 .
However, they depend on the additional mixing angle α which is not necessarily close
to β.
7. Conclusion
In conclusion, we have elaborated on a recently proposed model [10] of neutrino mass,

where a heavy triplet of leptons (Σ + , Σ 0 , Σ − )R per family is added as the anchor of the
seesaw mechanism in place of the canonical singlet NR . Using a second “leptonic” Higgs
doublet η [8], both MΣ and mη can be at the TeV scale and be produced at the LHC.
Experimental determination of their masses and decays to charged leptons will then map
out the neutrino mass matrix.
The existence of the triplet Σ allows a new U (1) gauge symmetry [10] to be defined,
with the associated gauge boson X at or below the TeV scale. The U (1)X charges of
the usual quarks and leptons are fixed up to one free parameter. If gX 2 /M 2 is not too
X
small compared to gZ2 /MZ2 , there will be corrections to the standard model at low energies
coming from this new gauge symmetry. We consider two scenarios. (A) The recent NuTeV
anomaly [11] is explained by U (1)X . (B) The possible slight discrepancy [20] in atomic
parity violation [12] is explained by U (1)X . We find that (A) and (B) are mutually
exclusive, i.e., we can accommodate one but not the other. The resulting constraint from
either (A) or (B) on gX2 /M 2 is such that the production of X has to be very large at hadron
X
colliders. Present limits at the Tevatron imply that MX is larger than about 2 TeV. On the
other hand, it is quite possible that these anomalies have other explanations, in which case
there is no hard constraint on MX or gX . Nevertheless, for gX of order the electroweak
coupling, the production of MX up to a few TeV can be reached at the LHC.
The U (1)X gauge symmetry requires two Higgs doublets of different U (1)X charges,
such that Φ1 couples to charged leptons and Φ2 couples to both up and down quarks.
This differs from the standard scenario and allows in particular the charged Higgs boson
to be light in the case of large tan β, resulting in a number of distinct phenomenological
predictions.
Acknowledgements
This work was supported in part by the US Department of Energy under Grant No. DE-
FG03-94ER40837.
References
[1] Super-Kamiokande Collaboration, S. Fukuda, et al., Phys. Rev. Lett. 85 (2000) 3999, and references therein.
[2] Super-Kamiokande Collaboration, S. Fukuda, et al., Phys. Rev. Lett. 86 (2001) 5656, and references therein;
See also: SNO Collaboration, Q.R. Ahmad, et al., Phys. Rev. Lett. 87 (2001) 071301;
SNO Collaboration, Q.R. Ahmad, et al., Phys. Rev. Lett. 89 (2002) 011301;
SNO Collaboration, Q.R. Ahmad, et al., Phys. Rev. Lett. 89 (2002) 011302.
[3] M. Gell-Mann, P. Ramond, R. Slansky, in: P. van Nieuwenhuizen, D.Z. Freedman (Eds.), Supergravity,
North-Holland, Amsterdam, 1979, p. 315;
T. Yanagida, in: O. Sawada, A. Sugamoto (Eds.), Proceedings of the Workshop on the Unified Theory and
the Baryon Number in the Universe, KEK, Tsukuba, Japan, 1979, p. 95;
R.N. Mohapatra, G. Senjanovic, Phys. Rev. Lett. 44 (1980) 912.
[4] E. Ma, U. Sarkar, Phys. Rev. Lett. 80 (1998) 5716.
[5] E. Ma, Phys. Rev. Lett. 81 (1998) 1171.

[6] S. Weinberg, Phys. Rev. Lett. 43 (1979) 1566.
[7] R. Foot, H. Lew, X.-G. He, G.C. Joshi, Z. Phys. C 44 (1989) 441.
[8] E. Ma, Phys. Rev. Lett. 86 (2001) 2502.
[9] R.E. Marshak, R.N. Mohapatra, Phys. Lett. 91B (1980) 222.
[10] E. Ma, Mod. Phys. Lett. A 17 (2002) 535.
[11] NuTeV Collaboration, G.P. Zeller, et al., Phys. Rev. Lett. 88 (2002) 091802.
[12] S.C. Bennett, C.E. Wieman, Phys. Rev. Lett. 82 (1999) 2484.
[13] S.L. Adler, Phys. Rev. 177 (1969) 2426;
J.S. Bell, R. Jackiw, Nuovo Cimento A 60 (1969) 47;
W.A. Bardeen, Phys. Rev. 184 (1969) 1848.
[14] R. Delbourgo, A. Salam, Phys. Lett. 40B (1972) 381;
T. Eguchi, P.G.O. Freund, Phys. Rev. Lett. 37 (1976) 1251;
L. Alvarez-Gaume, E. Witten, Nucl. Phys. B 234 (1984) 269.
[15] E. Witten, Phys. Lett. B 117 (1982) 324.
[16] G. Altarelli, R. Barbieri, S. Jadach, Nucl. Phys. B 369 (1992) 3;
G. Altarelli, R. Barbieri, S. Jadach, Nucl. Phys. B 376 (1992) 444, Erratum.
[17] S. Davidson, et al., JHEP 0202 (2002) 037.
[18] E. Ma, D.P. Roy, Phys. Rev. D 65 (2002) 075021.
[19] F. Abe, et al., Phys. Rev. Lett. 79 (1997) 2192.
[20] V.A. Dzuba, V.V. Flambaum, J.S.M. Ginges, hep-ph/0204134, and references therein.
[21] See, for example V. Barger, et al., Phys. Rev. D 57 (1998) 3833.
[22] ALEPH Collaboration, R. Barate, et al., Eur. Phys. J. C 12 (2000) 183.
[23] Particle Data Group, Eur. Phys. J. C 15 (2000) 1.
[24] M.Yu. Kuchiev, V.V. Flambaum, hep-ph/0206124.
Massless higher spins and holography

E. Sezgin a , P. Sundell b
a Center for Theoretical Physics, Texas A&M University, College Station, TX 77843, USA
b Department for Theoretical Physics, Uppsala Universitet, Sweden
Received 17 May 2002; accepted 12 August 2002
Abstract
We treat free large N superconformal field theories as holographic duals of higher spin (HS)
gauge theories expanded around AdS spacetime with radius R. The HS gauge theories contain
massless and light massive AdS fields. The HS current correlators are written in a crossing
symmetric form including only exchange of other HS currents. This and other arguments point to
the existence of a consistent truncation to massless HS fields. A survey of massless HS theories with
32 supersymmetries in D = 4, 5, 7 (where the 7D results are new) is given and the corresponding
composite operators are discussed. In the case of AdS4 , the cubic couplings of a minimal bosonic
massless HS gauge theory are described. We examine high energy/small tension limits giving rise
to massless HS fields in the type IIB string on AdS5 × S 5 and M-theory on AdS4/7 × S 7/4 . We
discuss breaking of HS symmetries to the symmetries of ordinary supergravity, and a particularly
natural Higgs mechanism in AdS5 × S 5 and AdS4 × S 7 where the HS symmetry is broken by
finite gYM . In AdS5 × S 5 it is shown that the supermultiplets of the leading Regge trajectory cross
over into the massless HS spectrum. We propose that gYM 2 = 0 corresponds to a critical string
tension of order 1/R 2 and a finite string coupling of order 1/N. In AdS7 × S 4 we give a rotating
membrane solution coupling to the massless HS currents, and describe these as limits of Wilson
surfaces in the AN−1 (2, 0) SCFT, expandable in terms of operators with anomalous dimensions
that are asymptotically small for large spin. The minimal energy configurations have semi-classical
energy E = s for all s and the geometry of infinitely stretched strings with energy and spin density
concentrated at the endpoints.
E-mail addresses: sezgin@physics.tamu.edu (E. Sezgin), p.sundell@teorfys.uu.se (P. Sundell).
PII: S 0 5 5 0 - 3 2 1 3 ( 0 2 ) 0 0 7 3 9 - 3
304 E. Sezgin, P. Sundell / Nuclear Physics B 644 (2002) 303–370
1. Introduction
The strong form of the Maldacena conjecture states that type IIB closed string theory
on AdS5 × S 5 with N units of five-form flux and string coupling gs corresponds to d = 4,
N = 4 SYM theory with SU(N) gauge group and Yang–Mills coupling gYM 2 = g [1–3].
s
This conjecture has been primarily tested for N gYM N 1, where supergravity is a
2
valid approximation [4,5]. It is natural to study the correspondence for N 1 gYM

2 N,
and possibly gYM = 0, where the SYM theory becomes a theory of afree SU(N) valued
2
d = 4, N = 4 vector singletons. At weak ’t Hooft coupling λ = gYM 2 N the natural
gauge invariant operators are composite single-trace operators which can be arranged into
‘trajectories’ according to the value of the twist E − s, where E is the conformal dimension
and s is the spin. The twist is the anomalous contribution to E, which becomes small at
weak ’t Hooft coupling and large N .
A basic observation [6] is the non-intersection principle in a CFT which states that as the
coupling varies there cannot be any mixing between operators that are not mixing already
at the free level. This applies to both the spectrum of composite operators of 4D SYM
in the limit N 1 gYM 2 N and the spectrum of vertex operators of the sigma model
for N gYM N 1. Thus an important test of Maldacena conjecture is to verify that the
2
trajectories of SYM operators with constant twist cross over into the closed string Regge
trajectories.
In this paper we shall show that this is indeed the case for the leading trajectories,
which consist of the states with minimal E for fixed s. In fact, on the SYM side the
leading trajectory, i.e., the operators with minimal twist, consists of bilinear higher spin
(HS) tensors. In the free limit, these have twist 2 and the s 1 sector coincides with the
space of conserved HS currents. General aspects of these currents have been discussed
in [7,8]. The precise spectrum of twist 2 operators and the corresponding HS symmetry
algebra extension of the conformal/AdS group was constructed in [9,10] using group
theoretic methods which shows that the twist 2 operators in fact form an irreducible ‘gauge’
multiplet of the HS algebra.
In [9,10] it was also shown how to describe the HS gauge multiplet on the bulk side
at the level of a linearized AdS field theory containing HS gauge fields as well as other
interesting HS fields generalizing the self-dual two-form of the supergravity multiplet
contained in the spectrum. This immediately raises the following questions; is it possible
to extend this picture to an interacting theory of massless HS fields in AdS5 , and if so, is
this theory the result of a consistent truncation of the full closed string theory in the limit
N 1 gYM 2 N?
There is increasing evidence for the existence of an interacting 5D massless HS gauge

theory [9–13]. This theory has been constructed at the linearized level [9,10], and certain
cubic interactions of the minimal bosonic theory have already been constructed [12]. The
structures involved are natural generalizations of those in D = 4, and we expect that a
similar development will unfold in D = 5.
For those readers not too familiar with massless HS gauge theories, in this paper we
review some of their basic properties in dimensions of interest, namely, D = 4, 5, 7. The
general formulation of interacting massless HS gauge theory has been known in D = 4 for
E. Sezgin, P. Sundell / Nuclear Physics B 644 (2002) 303–370 305
quite some time [14] (see, [15] for a review). In testing the free CFT/HS gauge theory
correspondence ideas, it is important to exhibit the couplings of the HS gauge theory.
The D = 4, N = 8 theory has been examined in great detail in [16,17]. In D = 4 the
basic interactions are contained in a minimal bosonic model which can be embedded as
a consistent truncation into HS gauge theories with N 0. The explicit couplings of the
minimal bosonic model in D = 4 are given in a generally covariant curvature expansion
scheme in [18,19]. Here we shall summarize the results of [19] at the level of cubic
couplings. The analogous bosonic truncation in D = 5 was given in [9] and in D = 7
in [20], though the full interactions still remains to be found. In this paper we also give the
symmetry algebra and massless spectrum of the D = 7, N = 2 HS gauge theory.
The issue of consistent truncation is crucial since the subleading trajectories in the
gauge theory correspond to massive AdS fields which are light, meaning that their AdS
energies are not separated from the massless ones by a mass-gap. Here it is important to
note that regardless of the detailed structure of the bulk interactions, it is still possible
[21] to arrange the effective bulk action into a 1/N 2 expansion such that its extremum
reproduces the 1/N 2 expansion of the correlators of the composite operators of the SU(N)
invariant singleton theory. In fact, this expansion remains highly non-trivial even in the
limit gYM2 = 0 [22,23]. In particular, if one sets to zero all the massive fields on the
boundary, then the extremum of the full effective action should reproduce the correlators
of the bilinear twist 2 operators. The massive fields may still become excited in the bulk,
if massless fields act as sources for massive fields. If this is the case, then the massless HS
gauge theory cannot serve as a good approximation for studying these processes, not even
as an effective theory since it is not possible to eliminate the light massive fields while
preserving locality (the non-localities which one encounters in massless HS theory are not
that bad). Thus, for the massless HS gauge theory to be relevant, it must be possible to
consistently set the massive fields to zero in the full theory, at least in the leading non-
trivial order in the 1/N 2 expansion.
There are several ways to test this consistent truncation. Firstly, it requires consistent
interactions among massless fields, for which there are many indications as already
mentioned. Given the consistent equations of motion or action for the massless fields, one
must then compute the bulk tree amplitudes, which by definition will only contain massless
excitations in the internal lines, and check that they correspond to the correlators of bilinear
composite operators computed in the singleton theory [19,24]. This direct method is
technically rather involved, however, and in this paper we instead provide indirect evidence
for consistent truncation by examining the nature of the correlators between bilinear
operators in singleton theories with large N . We also suggest that the arguments given
in [25–27] for the consistent truncation of type IIB and eleven-dimensional supergravities
on AdS4/7 × S 7/4 to gauged supergravity carry over to the HS context.
In this paper, we also emphasize the fact that the relations between the closed string
parameters gs and α in AdS5 × S 5 and the SYM parameters gYM 2 and N have so far been
tested only in the limit N gs N 1. In this regime the relations can be derived, e.g., by
first identifying the gauge theory parameters with the closed string parameters in flat 10D
spacetime, and then use D3-brane soliton description to interpolate from flat spacetime
down to AdS5 × S 5 . Since only 16 supersymmetries are preserved globally by the D3
brane, there may be string corrections to this computation.
It is important to note that the strong coupling tests of the AdS/CFT duality which
are based on exact calculations on the SYM side (see, for example, [28] and references √
therein) are still limited on the bulk side in that they do not go beyond the leading λ
approximation to the closed string theory in AdS background. As gYM 2 N becomes small
(keeping N large), we do not know the precise relations between closed string parameters
in AdS and the gauge theory parameters. It is clear that the string coupling gs decreases and
the sigma model coupling α /R 2 increases as gYM 2 decreases. We shall speculate that the
bulk parameters approach critical values as gYM 2 = 0 where the bulk theory is described
by closed string theory with coupling 1/N 2 and a singleton worldsheet CFT based on
critical level k affine PSU(2, 2|4) algebra, and that the left- and right-moving singleton
spin fields can be used in the construction of vertex operators describing massless HS fields
in the bulk. The level k is related to the worldsheet sigma model coupling constant, i.e.,
α /R 2 = ls2 /R 2 . The corrected relations between the closed string parameters in AdS5 × S 5
and the gauge theory parameters we propose are given by
gs = f1 (λ)gYM
2
, ls = f2 (λ)R, (1.1)
f1 (λ) ∼ 1, f2 (λ) ∼ λ−1/4 for λ 1,

1
f1 (λ) ∼ , f2 (λ) ∼ 1, for λ 1. (1.2)
λ
Another aspect of massless higher spins and holography which we emphasize in this paper
is a Higgs mechanism by which the HS symmetries are spontaneously broken [21] down
to the symmetries of ordinary supergravity. This phenomenon is best studied in the case of
AdS5 ×S 5 , primarily due to the fact that the free boundary SYM theory can be continuously
deformed by switching on the coupling constant gYM . As a result, the HS currents with
spin s > 2 will no longer be conserved. The resulting anomalies in the conservation
laws for these currents are encoded in operators which can be coupled to Higgs fields
which undergo their landmark shift transformations. Consequently, the Higgs mechanism
mentioned above is expected to take place.
Given that the full interacting HS theory in 5D is still not known, of course we cannot
work out the details of the Higgs mechanism here. However, we do provide kinematic
framework for it which suggests that the Higgsing phenomenon takes place in an infinite
number of massless N = 8, D = 5 multiplets containing HS fields, which together with the
supergravity multiplet make up the spectrum of the massless HS gauge theory. In particular,
we focus on the Higgsing of the Konishi multiplet, which has smax = 4 and is expected to
play an important role in study of the first massive type IIB closed string level, and outline
how the Higgs mechanism can be extended to all the HS multiplets. This phenomenon of
anomalies in the HS current conservation laws in the boundary having a holographic dual
description in the bulk as spontaneous breaking of the corresponding HS gauge symmetries
is similar to a phenomenon of a chiral U (1) anomaly having its gravity dual in a particular
AdS5 supergravity, as has been recently shown in [29].
Switching our discussion to the case of M-theory, we first recall that the Maldacena
conjecture [1,4] states the equivalence between M-theory on AdS4/7 × S 7/4 with N
units of 7/4-form flux on S 7/4 , and superconformal field theories (SCFT) with 16
supercharges describing the low energy dynamics of N parallel coinciding M2/5 branes in
flat eleven-dimensional spacetime. Apparently, these SCFTs are isolated fixed points of the
renormalization groups (RG) that do not admit any marginal deformations, with or without
preservation of supersymmetry. Consequently they do not admit any coupling constants
and Lagrangian descriptions. The main window for viewing these strongly coupled theories
is therefore through the bulk supergravity, which is a valid approximation to M-theory at
fixed energies provided N 1. This corresponds to a subset of the SCFT operators with
fixed conformal dimensions as N 1 [1,4]. Recently other limits of the correspondence
based on considering large internal spin have been proposed [30].
In analogy with type IIB closed string theory on AdS5 × S 5 , it is natural to ask whether
M-theory on AdS4/7 × S 7/4 has an unbroken phase in which M-theory corrections become
relevant at fixed energy and the effective description of the bulk theory becomes a HS
gauge theory with holographic dual given by a free SCFT in d = 3 or d = 6. In other
words, we wish to examine whether it is possible to have a ‘phase diagram’ with two fixed
points, one corresponding to the free singleton SCFT describing the unbroken HS phase
and another one corresponding to the strongly coupled SCFT describing the broken phase.
From the bulk point of view the broken phase is described by membranes interacting in
the flat eleven-dimensional center of AdS4/7 × S 7/4 , while the unbroken phase, which is
specific to AdS, is described by membranes interacting close to the boundary of AdS.
By examining the RG flows on M2/5 branes and D2/4 branes we are led to propose
that the relevant free SCFTs in d = 3, 6 are described by free SU(N) valued OSp(8|4)
singletons and free SU(N) valued d = 6, N = (2, 0) tensor singletons. These theories
have of course figured in the literature before (see, for example, [27]), and have been
used in many circumstances in order to unravel information about the strongly coupled
SCFTs.1 Our point here is that due to the salient features of the large N limit the free
SCFTs make sense on their own as holographic images of the interesting unbroken phases
of M-theory. Technically speaking, large N implies factorization and 1/N expansion of
correlators which can be matched with the expansion of the bulk amplitudes in terms of
the fundamental Planck scale.
As in the case of the type IIB theory, an important issue is whether there is a consistent
truncation down to a massless sector. The ideas for examining this are similar to those
described above for the type IIB theory. The D = 4 case is particularly tractable as in this
case we already know the full form of the interactions among massless HS fields, which
makes it possible to test directly the consistent truncation without first having to construct
the interactions.
An intriguing feature of the proposed unbroken phases of M-theory on AdS4/7 × S 7/4
is that the spectrum is discrete and that there is a finite coupling, 1/N . Thus the unbroken
phases of M-theory appears to be on the same footing as the unbroken phase of the type IIB
theory on AdS5 × S 5 . This suggests that the unbroken phases in AdS4/7 × S 7/4 are theories
of M2-branes with fixed tension.
1 Free singletons, which form N − 1 plets of the Weyl group of SU(N ), appear in various ‘trivial’ IR limits
describing stacks of separated branes sitting at certain orbifold singularities [31]. These free singletons should not
be confused with the SU(N ) valued singletons, though they are curious from the HS perspective and they should
presumably be included into the phase diagram as separate HS phases.
To gather further evidence for this, we examine a family of rotating membrane solutions
in AdS7 × S 4 that are curved space generalizations of those given in flat spacetime in
[32,33] and membrane analogs of the string solutions found recently in [34] (which in
fact describe the leading Regge trajectory states). The minimal energy configurations have
semi-classical energy E = s for all s, and the geometry of infinitely stretched membranes
of zero width, whose energy and spin densities are concentrated in the asymptotic region.
By examining the supersymmetry enhancement in this region we can further show that the
rotating membranes indeed couple to the bilinear HS currents in the SCFT.
There is an important difference between the membrane solitons and the string solitons
given in [34]. The string solitons couple to operators whose anomalous dimensions become
asymptotically small only for large s, (E − s)/s → 0 as sα /R 2 → ∞. The membrane
solitons, on the other hand, couple to anomaly free operators for any value of s. This
is because they arise by taking the limit of zero width which has the dual interpretation
of shrinking a Wilson surface which means that the holographic dual flows to the free
singleton SCFT in d = 6.
We find it rather compelling that relatively simple, free SCFTs contain information
about the unbroken, and perhaps more fundamental, phases of type IIB closed string and
M-theory. Moreover, this means that the results on free SCFT which are scattered over the
literature can now be given a more direct physical interpretation.
In AdS/CFT correspondence, it is important that both the bulk and the boundary
theories admit 1/N expansions which define the physically relevant, i.e., asymptotically
convergent, expansions. In the unbroken HS phase, the bulk side may also admit a strongly
coupled closed string/membrane sigma model description, which we propose has large,
but fixed, coupling given by a critical tension, as mentioned above. In any event, consistent
truncation makes it possible to directly test the AdS/CFT correspondence using only the
action for the massless HS fields which does not require strongly coupled sigma model
computations.
The breaking of the HS symmetries requires the inclusion of Higgs fields whose
interactions require us to go beyond the consistent truncation to massless fields. Whether
this can be done at the level of some effective field theoretical construction in the bulk or
whether it requires extracting information from the strongly coupled sigma model is not
clear at present. Here we can only speculate that the large amount of symmetry present in
the unbroken phase should make the critical string and membrane sigma models amenable
to exact methods.
This paper is organized as follows. In Section 2, the properties of HS gauge theories in
D = 4, 5, 7 are reviewed, including their underlying symmetry algebras and field contents.
The results for the HS superalgebra and spectrum in 7D are new. In Section 3, the
composite singleton operators corresponding to the massless states of HS gauge theories,
their KK towers and Higgs multiplets are discussed. In Section 4, important aspects of
the CFT/HS gauge theory correspondence, and, in particular, the 1/N expansion in the
free CFT on the boundary are described. In Section 5, the 5D HS gauge theory as the
bulk theory arising in the critical limit of type IIB string theory and a Higgs mechanism
breaking the HS gauge symmetries down to those of ordinary supergravity are discussed.
In Section 6, first the CFT 3 /HS gauge theory correspondence for M-theory on AdS4 × S 7
is described. Then, the minimal bosonic truncation of the theory and its cubic interactions
are described. In Section 7, first the CFT 6 /HS gauge theory correspondence for M-theory
on AdS7 × S 4 is discussed. Then our rotating membrane solution in AdS7 × S 4 is given
and its properties and relevance to the 7D HS gauge theory are described. Section 8 is
devoted to a summary and discussion. In Appendix A, we present several tables which
show various sectors of the massless HS gauge theory spectra in D = 5, 7. In Appendix B,
we summarize the UIRs and BPS states of the maximal AdS superalgebras in D = 4, 5, 7.
In Appendices C and D, we collect further group theoretical information that is useful for
Sections 2 and 3.
2. Massless higher spin gauge theories in D = 4, 5, 7
HS gauge theories are generally covariant theories which admit AdS as a vacuum and
have an infinite number of local HS supersymmetries based on HS superalgebras which
are infinite-dimensional extension of the finite-dimensional AdS superalgebras [35] . The
fundamental UIRs of the HS super algebras in D = d + 1 = 4, 5, 7 dimensions are ultra-
short d-dimensional conformal supermultiplets, which we will refer to as singletons.2
Gauging of such a HS superalgebra yields a D-dimensional theory based on a massless
HS supermultiplet given by the symmetric product of two singletons. In this paper we shall
focus our attention on the HS extension of the AdS superalgebras in D = 4, 5, 7 with 32
real supersymmetries because these are the most natural ones to explore from the string/M-
theory point of view. In Section 8, we shall comment on possible extensions to higher D
and higher number of supersymmetries.
The massless HS multiplet is an infinite tower of massless AdS supermultiplets with
supergravity at the lowest level. One key property is the fact that a HS gauge theory in
D > 3 cannot be consistently truncated to an AdS supergravity. Basically, this is due to the
fact that derivatives of lower spin fields serve as sources for HS fields, and it can also be
seen from the structure of the OPE of free field theory stress-energy tensors in d > 2 [7].
However, in D = 4, 5, 7 there exist minimal bosonic truncations which have remarkably
simple physical field content, namely, massless fields of spin s = 0, 2, 4, 6, . . . described
by doubly traceless, symmetric tensors φµ1 ···µs . The embedding of these theories in their
supersymmetric extensions is explained in Tables 1, 2 and 3.
As for the full and covariant (i.e., background independent) interactions among the
massless fields, they are known in the 4D theory [14,18,19]. A condensed account of how
to extract cubic couplings in D = 4 will be given in Section 6.2. The fully interacting
theories in D = 5, 7 have not yet been constructed, though the results obtained so far are
promising [9,10,12,20].
We next list the HS superalgebras in D = 4, 5, 7, their singleton and massless
representations, and how the latter ones are assembled into master 1-form and master
0-form fields. The results in D = 4, 5 were obtained in [10,16]. The minimal bosonic HS
algebra in 7D was obtained in [20]. The results presented here for its supersymmetric
extension are new.
2 In d = 4, 6, these are usually referred to as doubletons, due to the fact that their oscillator construction is
based on two sets of oscillators as opposed to a single set of oscillators used in d = 3.
2.1. The D = 4, N = 8 massless HS gauge theory
The 4D HS algebra hs(8|4) is realized in terms of oscillators obeying the following

algebra [16,17,35]3
yα yβ = yα yβ + iαβ , yα ȳα̇ = yα ȳα̇ , (yα )† = ȳα̇ , (2.1)
i †
θ θ =θ θ +δ ,
i j i j ij
θ = θi, (2.2)
where yα (α = 1, 2) is a Weyl spinor which is a Grassmann even generator of a Heisenberg
algebra, and θ i (i = 1, . . . , 8) is a Grassmann odd generator of an SO(8) Clifford algebra.
The denotes the associative product between oscillators. The products on the right-hand
sides are Weyl ordered, so that, for example, yα yβ = yβ yα and θ i θ j = −θ j θ i . Using
the above contraction rules it is straightforward to compute the -product between two
arbitrary Weyl ordered polynomials of oscillators.
The algebra hs(8|4) consists of arbitrary Grassmann even and antihermitian polynomi-
als P (y, ȳ, θ ) that are sums of monomials of degree 4! + 2 where ! = 0, 1, 2, . . . , which
will be referred to as the level index. The Lie bracket between P , Q ∈ hs(8|4) is given by
[P , Q] . Thus, denoting by P (!) an !th level monomial, the commutation relations have
the schematic form
(! ) (! )
P 1 ,P 2 = P (!) . (2.3)
|!1 −!2 |!!1 +!2
In particular, the zeroth level of hs(8|4) is the maximal finite subalgebra OSp(8|4) whose
generators schematically take the form
Qαi = yα θi , α̇i = ȳα̇ θi ,
Q Uij = θi θj ,
Mα β̇ = yα ȳβ̇ , Mαβ = yα yβ , Mα̇ β̇ = ȳα̇ ȳβ̇ . (2.4)
A generator P (!) in the !th level of hs(8|4) can be expanded as
P (!) (y, ȳ, θ )
1
= ȳ α̇1 · · · ȳ α̇m y β1 · · · y βn θ i1 · · · θ ip Pα̇1 ···α̇m β1 ···βn i1 ···ip . (2.5)
m+n+p
m! n! p!
= 4!+2
The spins of the components are given by s = 12 (m + n). The components with integer
spin are Grassmann even and those with half-integer spin are Grassmann odd. Bosons are
in the 1, 28 and 35± irreps of SO(8) and fermions in the 8 and 56 irreps. The reality
properties follow from P † = −P .
A UIR of OSp(8|4) is denoted by D(E0 , s; a1 , a2 , a3 , a4 ), where the notation is
explained in Appendix B. The fundamental UIR of OSp(8|4), which is also a UIR of
hs(8|4), is the ultra-short singleton [36,37]

D 12 , 0; 0, 0, 0, 1 ⊕ D 1, 12 ; 0, 0, 1, 0 . (2.6)
3 The algebra hs(8|4) is called shsE (8|4) in [9] and shsE (8, 4|0) in [35], where it is also shown to be
isomorphic to ho(8; 8|4).
By taking products of singletons we obtain further unitary representations of hs(8|4).

Two singletons yield OSp(8|4) weight spaces corresponding to massless AdS4 fields with
E0 = s + 1 [38–40].
The massless sector of the hs(8|4) gauge theory is formulated in terms of an hs(8|4)
valued master gauge field Aµ (y, ȳ, θ ) (with expansion given by (2.5)) and a master zero-
form Φ(y, ȳ, θ ) in a quasi-adjoint representation of hs(8|4) with expansion
Φ(y, ȳ, θ )
1
= ȳ α̇1 · · · ȳ α̇m y β1 · · · y βn θ i1 · · · θ ip Φα̇1 ···α̇m β1 ···βn i1 ···ip . (2.7)
−m+n+p
m! n! p!
= 0 mod 4
The reality condition on Φ is discussed in detail in [16,17]. The gauging gives rise to
a set of field equations for physical fields (the action still remains to be found) whose
spectrum is given by the symmetric product of two singletons which is given in Table 1.
The physical spin s 1 fields are the gauge fields in Aµ (y, ȳ, θ ) that correspond to hs(8|4)
generators in (2.5) satisfying |m − n| 1. Those with m = n contain the vierbein and
its HS generalizations, while those with |m − n| = 1 contain the gravitini and their HS
generalizations. The physical fields with s 12 arise in Φ(y, ȳ, θ ) as the components in
(2.7) with m + n 1. The remaining fields in Aµ and Φ are auxiliary and given in terms
of derivatives of the independent fields.
So far we have discussed the free massless HS gauge theory. The general formulation of
interacting massless HS gauge theory has been given in D = 4 [14] (see, [15] for a review),
and examined in detail for N = 8 [16,17]. There exists a minimal bosonic truncation of this
theory whose spectrum consist the physical states with spin s = 0, 2, 4, . . . , each occurring
once. This theory exhibits the basics of any HS gauge theory rather well and it will be
discussed in considerable detail in Section 6.2, which is based on [19].
Table 1
The SO(3, 2) × SO(8) content of the symmetric tensor product of two d = 3, N = 8 singletons. Each entry refers
to the SO(8) content. All SO(8) irreps are irreducible except 70 = 35+ + 35− and all the states have E0 = s + 1
except the scalars in one of the 35-plets at level ! = 0 and one of the scalars at level ! = 1. The representations
have been arranged into a tower of OSp(8|4) supermultiplets labeled by a level index !. The zeroth level is the
D = 4, N = 8 supergravity multiplet with 28 degrees of freedom. The level ! 1 multiplets have 2 × 28 degrees
of freedom. The spin s 1 fields arise in the hs(8|4) valued master gauge field and the spin s 12 arise in
the quasi-adjoint master zero-form. The minimal bosonic truncation of the spectrum is obtained by keeping the
maximum spin fields at each level and the (non-pseudo)scalar at level ! = 1
!\s 0 1 1 3 2 5 3 7 4 9 5 11 6 ···
2 2 2 2 2 2
0 70 56 28 8 1
1 1+1 8 28 56 70 56 28 8 1
2 1 8 28 56 70 56 28 8 1
3 1 8 28 56 70 ...
4 1 ...
..
.
The 5D HS superalgebra hs(2, 2|4) [10] is realized in terms of the following oscillators4

yα ȳβ = yα ȳβ + Cαβ , yα yβ = yα yβ , y † iΓ 0 C α
= ȳα , (2.8)
i †
θ i θ̄j = θ i θ̄j + δji , θi θj = θiθj , θ = θ̄i , (2.9)
where yα (α = 1, . . . , 4) is a Grassmann even Dirac spinor and θ i (i = 1, . . . , 4)

is a Grassmann odd SO(6) SU(4) spinor. The charge conjugation matrix Cαβ is
antisymmetric. The algebra hs(2, 2|4) consists of Grassmann even and antihermitian
polynomials P (y, ȳ, θ, θ̄) that are sums of monomials of degree 4! + 2 (! = 0, 1, 2, . . .)
that are invariant under the U (1)Z generated by
1
Z = (ȳy + θ̄ θ ); (2.10)
2
and traceless in their spinor indices:
P (!) (y, ȳ, θ, θ̄)

1
= ȳ α1 · · · ȳ αm y β1 · · · y βn θ i1 · · · θ ip θ̄j1 · · · θ̄jq Pα1 ···ip j1 ···jq ,
m+n+p+q
m! n! p! q!
= 4!+2
m+p=n+q
(2.11)
where
C α1 β1 Pα1 ···αm β1 ···βn i1 ···ip j1 ···jq = 0, Pi i = 0. (2.12)
The tracelessness of Pi j means the removal of the outer U (1)Y automorphism generator
Y = θ̄ θ. (2.13)
The Lie bracket between P , Q ∈ hs(2, 2|4) is given by [P , Q] /I where I is the ideal
generated by elements of the form
∞

Pn (y, ȳ, θ, θ̄ ) Z ·
· · Z
, (2.14)
n=1 n factors
where Pn are polynomials which are traceless in their spinor indices. The structure of the
Lie bracket is similar to (2.3).
The zeroth level of hs(2, 2|4) is the maximal finite subalgebra
PSU(2, 2|4) = PU(2, 2|4)/U (1)Z , (2.15)
4 The algebra hs(2, 2|4) is called ho (1, 0|8) in [12].

0
where PU(2, 2|4) is the centrally extended superalgebra (with 31 bosonic generators). The
PSU(2, 2|4) generators are realized schematically as
iα = ȳα θ i , 1
Qαi = yα θ̄i , Q Mαβ = ȳα yβ − Cαβ (ȳy),
4
1
U i j = θ̄ i θj − δji (θ̄θ ). (2.16)
4
The Lorentz spin of a generator in (2.11) is given by (jL , jR ) = ( 12 m, 12 n) and the U (1)Y
charge by Y = p − q. The components with integer jL + jR are Grassmann even and those
with half-integer jL + jR are Grassmann odd. Bosons are in the 10 , 150 , 200 , 62 , 102 and
14 irreps of SU(4) × U (1)Y and fermions in the 41 , 43 and 103 irreps. The reality properties
follow from the condition P † = −P which, in particular, implies that the irreps with Y = 0
are real. The generators of the algebra are summarized in Tables 4 and 5 (see Appendix A).
A UIR of SU(2, 2|4) is denoted by D(E0 , jL , jR ; a1 , a2 , a3 )Y where the notation
is explained in Appendix B. The fundamental UIRs of SU(2, 2|4) are the ultra-short
singletons given in Table 6 in Appendix A [41]. Due to the modding out of the ideal I
generated by elements of the form (2.14) the fundamental UIR of hs(2, 2|4) is the singleton
with vanishing Z charge, i.e., the Maxwell supermultiplet [41–44]

D(1, 0, 0; 0, 1, 0)0 ⊕ D 32 , 12 , 0; 1, 0, 0 −1 ⊕ D 32 , 0, 12 ; 0, 0, 1 1
⊕ D(2, 1, 0; 0, 0, 0)−2 ⊕ D(2, 0, 1; 0, 0, 0)2. (2.17)
By taking products of this multiplet we obtain further unitary representations of hs(2, 2|4).
In particular, the product of two singletons yields massless AdS5 fields whose energies,
which are given by E0 = 2 + jL + jR saturate the unitarity bound of a continuous series
(denoted as series A in Appendix B) [41,42].
The massless sector of the hs(2, 2|4) gauge theory is formulated in terms of an hs(2, 2|4)
valued master gauge field Aµ (y, ȳ, θ, θ̄) and a master zero-form Φ(y, ȳ, θ, θ̄ ) in a certain
quasi-adjoint representation of hs(8|4) [9,10], which contains the Weyl tensors and the
extra ‘matter’ fields given in Table 7 in Appendix A. The gauging gives rise to physical
fields whose spectrum is given by the symmetric product of two singletons given in
Table 2. The fields with |Y | 1 and |Y | 2, jL + jR 12 carry SO(4, 1) weights
such that the analysis of the curvature constraints in this sector is analogous to that in
D = 4. In the |Y | 2, jL + jR 1 sector the fields carry SO(4, 1) weights that require
a separate analysis. One finds that [10] the physical fields arise as two-form potentials in
Φ obeying the odd-dimensional self-duality equation B2 = dB2 or higher-spin analogs
of this equation (such equations have been more recently studied in the lightcone gauge
in [45]). The gauge fields in Aµ with |Y | 2 are auxiliary fields which are related to the
independent two-forms in Φ by generalized Hodge dualization rules [10].
So far we have discussed the free massless HS gauge theory. The full interacting theory
based on hs(2, 2|4) has not been constructed yet. However, the kinematics established
in [10] and summarized above, together with the already established principles [14] that
govern the structure of the interacting HS gauge theory in D = 4, suggest that the full
5D interacting theory is perfectly within reach. Indeed certain cubic interactions of the
minimal bosonic HS theory in 5D have already been constructed by Vasiliev [12].
Table 2
The symmetric tensor product of two d = 4, N = 4 SYM singletons arranged into levels ! = 0, 1, 2, . . . of
PSU(2, 2|4) multiplets. The entries denote USp(8) representations: 28 = 27 + 1, 56 = 48 + 8, 70 = 42 + 27 + 1,
which branch under SU(4) × U (1)Y as follows: 1 = 10 , 8 = 41 + 4̄−1 , 27 = 150 + 62 + 6̄−2 , 42 = 200 + 102 +
10−2 +14 + 1̄−4 and 48 = 201 +20−1 +43 + 4̄−3 . Each entry also carries SO(4) SU(2)L ×SU(2)R ⊂ SO(4, 2)
spins (jL , jR ). The total spin s is defined as s = jL + jR and the U (1)Y charge is given by Y = 2(jR − jL ).
The level ! = 0 multiplet is the D = 5, N = 8 supergravity multiplet. The level ! = 1 multiplet is the massless
Konishi multiplet. The level ! 0 multiplets have (4! + 1) × 28 degrees of freedom. The states in the s 12
sector arise as the physical states in the master scalar field Φ, as shown in Table 7. For s 1, the states with
Y = 0, ±1 arise in the sector of the master gauge field Aµ corresponding to the generators of hs(2, 2|4) listed
in Table 4. Those with Y = ±2, ±3, ±4 arise in the master scalar field Φ. With the exception noted in Table 7,
these have dual gauge fields corresponding to the generators of hs(2, 2|4) listed in Table 5. The minimal bosonic
truncation of the spectrum is obtained by keeping the maximum spin fields at each level and the scalar at level
!=1
!\s 0 1 1 3 2 5 3 7 4 9 5 11 6 ···
2 2 2 2 2 2
0 42 48 27 8 1
1 1 8 28 56 70 56 28 8 1
2 1 8 28 56 70 56 28 8 1
3 1 8 28 56 70 ...
4 1 ...
..
.
The linearized gauge theory of the minimal bosonic HS subalgebra, hs(8∗ ) in D = 7

was introduced in [20]. Here we shall construct its supersymmetric extension hs(8∗ |4).
This algebra is realized in terms of the following oscillators:
† 0
yα ȳβ = yα ȳβ + Cαβ , yα yβ = yα yβ , y iΓ C α = ȳα , (2.18)
†
θ i θ̄ j = θ i θ̄ j + Ω ij , θi θj = θiθj , θ i = θ̄ j Ωj i , (2.19)
where yα (α = 1, . . . , 8) is a Grassmann even Dirac spinor and θ i (i = 1, . . . , 4) is a
Grassmann odd Dirac spinor of SO(5) USp(4). The charge conjugation matrix Cαβ is
symmetric and Ω ij is the antisymmetric USp(4) invariant tensor. The algebra hs(8∗ |4)
consists of Grassmann even and antihermitian polynomials P (y, ȳ, θ, θ̄) that are sums
of monomials of degree 4! + 2 (! = 0, 1, 2, . . .) which are invariant under the SU(2)Z
generated by

Z3 = 14 ȳ α yα + θ̄ i θi , Z+ = 14 y α yα + θ i θi , Z− = 14 ȳ α ȳα + θ̄ i θ̄i .
(2.20)
and traceless in their spinor indices. The Lie bracket between P , Q ∈ hs(8∗ |4) is given by
[P , Q] /I where I is the ideal generated by elements of the form
∞

PnI1 ···In y, ȳ, θ, θ̄ ZI1 · · · ZIn , (2.21)
n=1
where PnI1 ···In (y, ȳ, θ, θ̄ ) has an expansion in terms of traceless, Weyl ordered multispinors
and the SU(2)Z indices I1 · · · In are symmetric. The structure of the Lie bracket is again
similar to (2.3). The zeroth level of hs(8∗ |4) is the maximal finite subalgebra OSp(8∗ |4)
realized schematically as
Qαi = yα θ̄i − ȳα θi , Mαβ = ȳ[α yβ] , Uij = θ(i θ̄j ) . (2.22)
An !th level generator P (!) in hs(8∗ |4) can be expanded as
P (!) (y, ȳ, θ, θ̄)

1
= ȳ α1 · · · ȳ αm y β1 · · · y βn θ i1 · · · θ ip θ̄ j1 · · · θ̄ jq Pα1 ···jq , (2.23)
m+n+p+q
m! n! p! q!
= 4!+2
m+p=n+q
where the components are traceless in their Lorentz spinor indices and belong to super-
Young tableaux with two rows of length 2! + 1. A single box in the super-Young tableaux
represents the superoscillator ξ A = (y α , θ i ) or ξ̄ A = (ȳ α , θ̄ i ). An arbitrary Weyl ordered
monomial in these superoscillators corresponds to a super-Young tableaux with two rows.
The restriction m + n = p + q in (2.23) (i.e., equal number of ξ A and ξ̄ A ) follows from the
condition [Z3 , P ] = 0, while the condition [Z± , P ] = 0 rules out super-Young tableaux
with rows of unequal length. The resulting super-Young tableaux of width 2! + 1 splits
into a set of Young tableaux of spinors. Each SO(6, 2) Young tableaux branches into a set
of Young tableaux of SO(6, 1) spinors. The spinorial SO(6, 1) × SO(5) Young tableaux
can be converted into tensorial ones by multiplying with appropriate Dirac matrices of
both groups. The resulting SO(5) irreps are 10 , 50 , 100, 140 , 12 , 52 , 102, 14 in the bosonic
sector and 41 , 161 , 43 in the fermionic sector, where the subscripts denote the U (1)Y charge
defined as
Y = nθ̄ − nθ , (2.24)
with nθ̄ = q and nθ = p, as specified in the expansion (2.23). The SO(6, 1) highest weights
(m1 , m2 , m3 ) are given by
m1 = 2! + 1 − 12 (nθ + nθ̄ ) m2 m3 = 12 |Y |. (2.25)

Note that since P is assumed to be Grassmann even the components in (2.23) with integer
weights are Grassmann even and those with half-integer weights are Grassmann odd. The
reality properties follow from P † = −P . As a result, all SO(6, 1) × SO(5) representations
obey symplectic reality conditions. For example, the supercharge Qαi obey a symplectic
Majorana condition so that it has 32 real components:

αi ≡ (Qβj )† iΓ 0 C Ωj i = Qαi .
Q (2.26)
βα
These results are summarized in Tables 8 and 9 in Appendix A.

A UIR of OSp(8∗ |4) is denoted by D(E0 , J1 , J2 , J3 ; a1 , a2 )Y where the notation is
explained in Appendix B. The fundamental UIRs of OSp(8∗ |4) are the singletons given
in Table 10 in Appendix A [46]. The singleton which is singlet of SU(2)Z also forms an
UIR of hs(8∗ |4). This singleton is the (2, 0) tensor multiplet [44,46–48]

D(2, 0, 0, 0; 0, 1)0 ⊕ D 52 , 1, 0, 0; 1, 0 1 ⊕ D 52 , 0, 0, 1; 1, 0 −1
⊕ D(3, 0, 0, 2; 0, 0)2 ⊕ D(3, 2, 0, 0; 0, 0)−2. (2.27)
By taking products of this singleton one obtains further unitary representations of hs(8∗ |4).
In particular, the square yields massless AdS7 fields with energy E0 = 4 + s where s ≡ J1 .
These energies belong to an isolated series (denoted as series B in Appendix B) [48],
unlike in D = 4, 5 where the massless fields have energies that saturate a continuous series
(the continuous series is saturated by lowest weight spaces arising in the product of three
singletons).
The superalgebra hs(8∗ |4) has a minimal bosonic HS subalgebra hs(8∗ ) whose
representation theory and gauging was described in [20]. We shall assume that the
massless sector of the hs(8∗ |4) gauge theory is formulated in terms of an hs(8∗ |4) valued
master gauge field Aµ (y, ȳ, θ, θ̄) and a master zero-form Φ(y, ȳ, θ, θ̄ ) in a quasi-adjoint
representation5 of hs(8∗ |4) and that the gauging gives rise to physical fields whose
spectrum is given by the symmetric product of two tensor singletons listed in Table 3.
The gauge fields with Y = 0 carry SO(6, 1) weights which are similar to those in the
minimal bosonic theory [20]. The gauge fields with |Y | 1 carry SO(6, 1) weights which
are analogous to those carried by the |Y | 1 fields in the hs(2, 2|4) theory in D = 5.
Thus we expect that the physical fields with |Y | 1, s 1 arise in Aµ . The remaining
physical fields, which have s 12 , or s 1 and |Y | 2, must arise in Φ and be those given
Table 3
The symmetric tensor product of two d = 6, N = (2, 0) tensor singletons arranged into levels ! = 0, 1, 2, . . .
of OSp(8∗ |4) multiplets. The entries denote SO(5) × U (1)Y representations as follows: 14 = 140 , 16 = 161 ,
15 = 100 + 52 , 4 = 41 , 16 = 100 + 52 + 12 , 24 = 161 + 41 + 43 , 36 = 140 + 50 + 10 + 102 + 52 + 14 . The
SO(6) ⊂ SO(6, 2) highest weights (n1 , n2 , n3 ) associated with each entry are given by n1 = s, n2 = 12 |Y | and
n3 = 12 Y . The level ! = 0 multiplet is the D = 7, N = 2 supergravity multiplet. The level ! 0 supermultiplets
contain 13 (! + 1)(2! + 1)(4! + 3) × 28 degrees of freedom. The states with |Y | 1, s 1 are expected to arise in
the sector of the master gauge field Aµ corresponding to the generators given in Table 8. The states with s 12 ,
or |Y | 2 and s 1, which are listed in Table 11, are expected to arise in a quasi-adjoint master zero-form Φ.
With a few low-lying exceptions which are given in Table 11, these are generalized Hodge duals of the |Y | 2
sector of the master gauge field Aµ which corresponds to the hs(8∗ |4) generators listed in Table 9. The minimal
bosonic truncation of the spectrum is obtained by keeping the maximum spin fields at each level and the scalar at
level ! = 1
!\s 0 1 1 3 2 5 3 7 4 9 5 11 6 ···
2 2 2 2 2 2
0 14 16 15 4 1
1 1 4 16 24 36 24 16 4 1
2 1 4 16 24 36 24 16 4 1
3 1 4 16 24 36 ...
4 1 ...
..
.
5 This representation was defined for hs(8∗ ) in [20]. Its generalization to hs(8∗ |4) will not be given here.
by Table 11 in Appendix A. In particular, we expect that the physical fields with |Y | 2,

s 1 arise as three-form potentials in Φ obeying the odd-dimensional self-duality equation
B3 = dB3 and their HS analogs, and that the corresponding gauge fields in Aµ are related
to these three-forms by generalized Hodge dualization rules analogous to those found in
the hs(2, 2|4) theory in D = 5 [10].
So far we have discussed the free massless HS gauge theory. The interacting theory has
not been constructed yet. However, the kinematics of theory established here, together with
what we know in lower dimensions about interacting HS gauge theories should help a great
deal in such a construction. In particular, there are many parallels with the kinematics of
the 5D HS gauge theory which is noteworthy.
3. Composite operators in singleton theories
In this section we describe the singleton theories in d = 3, 4, 6 with 16 supersymmetries

that are of relevance to the HS gauge theories in D = 4, 5, 7 described in the previous
section. We shall also identify the superfield realization of the HS currents in terms of
these singletons, whenever possible.
Explicit expressions for the supersymmetric currents have been constructed so far in
d = 3, and for a minimal bosonic truncation, in arbitrary dimensions. The bosonic currents
are formed out of a set of real scalar singletons and that are primary fields carrying SO(d, 2)
lowest weights (E0 ; m1 , . . . , m[d/2] ) = (d − 2 + s; s, 0, . . . , 0), where s = 0, 2, 4, . . . .
These tensors are conserved currents for s 2. The minimal bosonic HS theories are still
‘maximal’ in the sense that the twist d − 2 currents with even spin are the only composites
which are both conserved and primary. There are conserved currents with E0 − s > d − 2
as well as E0 − s = d − 2 and odd spin, though these can be shown to be descendants of
those with twist d − 2 and even spin.
3.1. The d = 3, N = 8 singleton and its composites
The fundamental UIR of OSp(8|4), which is also a UIR of hs(8|4), is the ultra-short
singleton specified in (2.6). This is just the d = 3, N = 8 scalar multiplet, and its superfield
realization has been known for sometime. In particular, it has arisen in the superembedding
formulation of M2-branes [49]. Following [50], let us work with a realization related to the
one in [49] by triality. The singleton superfield is then carries a spinor representation of
SO(8) and obeys the constraint
Dαi ΦA = (Γi )A Ḃ χα Ḃ , (3.1)

where χα Ḃ is a spinor superfield, i, A, Ḃ = 1, . . . , 8 label the 8v , 8s , 8c representations of
SO(8), respectively, and Γ -matrices are the chirally projected SO(8) Dirac matrices. The
singleton superfield ΦA carries the irrep D(1/2, 0; 0, 0, 0, 1), which belongs to series B
and it is BPS-1/2 multiplet. See Appendix B for notation and further details.
Several composite operators built out of two singletons superfields ΦA and their
derivatives are known [50–54]. Let us identify those which correspond to the spectrum
of massless field in the D = 4 HS gauge theory based on hs(8|4) as shown in Table 1. The
level ! = 0 supercurrent is realized as [50]

1
JAB = ΦA ΦB − δAB Φ 2 , Φ 2 := Φ A ΦA . (3.2)
8
The superfield JAB carries the irrep D(1, 0; 0, 0, 2, 0) which belongs to series B and it
is BPS-1/2 multiplet (see Appendix B). Its lowest component carries the 35s irrep of
SO(8). Together with the scalars in 35c that arise in the θ -expansion, they form the 70-
plet corresponding to the 70 scalar fields of level ! = 0 supergravity multiplet in AdS4
which has 28 degrees of freedom.
At level ! = 1, we have the supercurrent [50]
J = Φ2 (3.3)
which, as a consequence of the basic singleton constraint (3.1), obeys [50]
D ij J − trace = 0, D ij := D αi Dαj . (3.4)
The superfield J carries the irrep D(1, 0; 0, 0, 0, 0), which is current-like semi-short IUR
that saturates the unitarity bound of series A. Its lowest component is a scalar, and another
scalar arises in the θ -expansion. Altogether, 2 × 28 degrees of freedom arise [53] and they
correspond to the massless fields of level ! = 1 shown in Table 1.
Finally, the level ! 2 supercurrents can be realized in terms of the singleton superfield
as follows [50]

2!−2

Jα1 ···α4!−4 = (−1)k 32i∂(α1α2 ··· ∂α2k−1 α2k Φ A ∂α2k+1 α2k+2 ···∂α4!−5 α4!−4 ) ΦA
k=0
+ Γ i Γ j AB ∂(α1 α2 ··· ∂α2k−1 α2k Dαi 2k+1 Φ A

× Dαj 2k+2 ∂α2k+3 α2k+4 ···∂α4!−5 α4!−4) Φ B . (3.5)
These currents obey the constraint [50]
D iα Jαα2 ···α4!−4 = 0, ! 2. (3.6)
The superfield Jαα2 ···α4!−4 carries the irreps D(2! − 1, 2! − 2; 0, 0, 0, 0), which is semi-
short IUR that saturates the unitarity bound of series A. Its lowest component is the current
with spin smin = 2! − 2 and higher components go up to smax = 2! + 2. They correspond
to the level ! 2 massless multiplets listed in Table 1.
As discussed in the introduction, and to be elaborated further in Section 5, if interactions
can be switched on in the 3D CFT such that the HS gauge symmetry breaks down to
OSp(8|4), then the HS currents for level ! 1 will no longer be conserved. Assuming
such breaking, we can characterize the anomalies in conservation law for these currents as
D ij J − trace = gΣ ij − trace, (3.7)
D Jαα2 ···α4!−4 = gΣ
iα i
α2 ···α4!−4 , (3.8)
where g is some coupling constant, and the right-hand sides denote superfields which are
to be determined. These superfields carry the following irreps
Σ ij : D(2, 0; 2, 0, 0, 0), (3.9)

Σ α2 ···α4!−4 : D 2! − 12 , 2! − 52 ; 1, 0, 0, 0 .
i
(3.10)
This means that Σ ij describes a BPS-1/8 multiplet and it satisfies the unitarity condition
of series B. In [50], a BPS short multiplet of this type is built out of four singletons using
harmonic superspace technique. In terms of ordinary superfields we write it as

Σ ij = (Γimnp )AB Γj mnp CD Φ A Φ B Φ C Φ D − trace. (3.11)
This is just the 35-plet contained in the symmetric product (8s × 8s × 8s × 8s )S . Since the
superfield Σ ij represents a BPS-1/8 multiplet, its components go up to smax = 7/2. There-
fore, it is natural to consider this superfield as a candidate for coupling to Higgs superfield
in the bulk which can be eaten by the massless Konishi multiplet to become massive.
Turning to the candidate anomaly superfield Σ i α2 ···α4!−4 given in (3.10), we observe that
it carries a semi-short IUR that saturates the unitarity bound of series A. In general, such
multiplets have been constructed as [50]
S [ai ] = Φ 2 BPS[ai ] , (3.12)
{µ1 ···µs }[ai ] {µ1 ···µs } [ai ]
S =J BPS , (3.13)
where BPS[ai ] is any one of the BPS short multiplets listed in (2.4)–(2.6), and J {µ1 ···µs }
is a spin s current. Assuming that the candidate anomaly superfield Σ i α2 ···α4!−4 belongs
to an irreducible representation of OSp(8|4), since it is an 8-plet of SO(8), it requires the
BPS-1/8 multiplet D(1, 0; 1, 0, 0, 0) and a spin s = 2! − 12 current in (3.13). However, the
BPS-1/8 multiplet cannot be built out of one type of singleton field. Thus, the construction
of Σ i α2 ···α4!−4 , which is important for a Higgs mechanism that can work at all levels ! 1,
remains an open problem.
The BPS multiplets that can be constructed from the product of one type of singletons
are all the BPS-1/2 and BPS-1/4 multiplets listed in (B.4) and (B.5), and all those BPS-
1/8 multiplets listed in (B.6) with integer s [51]. These multiplets, as well as the semi-
short multiplets discussed above which make use of them, are likely to play a significant
role in the description of the full HS gauge theory based on hs(8|4). In particular, the
KK supermultiplets associated with level ! supermultiplets of the massless HS theory are
expected to be BPS-1/2 multiplets. For example, the level ! = 0 multiplet and its KK
towers are realized as [53]
D(k/2, 0; 0, 0, k, 0): Φ(A1 ΦA2 · · · ΦAk ) − traces, k = 2, 3, . . . . (3.14)
Taking k = 2 gives the massless supergravity multiplet and k = 3, 4, . . . give their massive
KK descendants. Similarly, the semi-short multiplets (3.12) and (3.13) with BPS-1/2
composites carrying the irrep D(k/2, 0; 0, 0, k, 0) are candidates for KK descendants of
the level ! > 0 massless multiplets of the HS gauge theory based on hs(8|4).
3.2. The d = 4, N = 4 singleton and its composites
The fundamental UIR of SU(2, 2|4), which is also a UIR of hs(2, 2|4), is the ultra-
short singleton specified in (2.17). This is the d = 4, N = 4 Maxwell multiplet realized in
terms of superfield Wij , where i = 1, . . . , 4 labels the 4-plet of SU(4) and W ij = −Wj i .6
6 This is the unique singleton multiplet of PSU(2, 2|4) and it has vanishing U (1) central charge. The
Z
centrally extended PU(2, 2|4) superalgebra admits an infinite number of singleton multiplets. These have jR = 0
It satisfies the following constraints and reality condition

Dα(i W j )k = 0, αi W j k − trace = 0,
D
† 1
Wij ≡ W ij = ij kl W kl . (3.15)
2
The singleton superfield Wij carries the irrep D(1, 0, 0; 0, 1, 0). It belongs to series C and
it describes a BPS-1/2 multiplet. There are several papers which deal with the construction
of the composite operators built out of the Maxwell (or SYM) singleton. See, for example,
[55–65]. Here we shall follow closely the treatment of [50].
For each state in the spectrum of the HS gauge theory listed in Table 2, one can construct
the corresponding conserved current out of two Maxwell singletons and their derivatives.
To begin with, the level ! = 0 supercurrent is contained in the superfield Jij,kl , which is in
20 of SU(4) and is given by [55,56]
1
Jij,kl = Wij Wkl − ij kl W mn Wmn . (3.16)
12
Defining W a ≡ (Γ a )ij W ij (a = 1, 2, . . . , 6), where Γ a are the chirally projected SO(6)
Dirac matrices, the current superfield (3.16) can equivalently be written as Jab = Wa Wb −
6 δab Wc Wc . Defining Jij =
1 kl klmn J
ij,mn , on the other hand, it obeys the constraint [56]
[k l]m
Dαm Jijkl = χαij
mkl
+ δ[im λkl
αj ] + δ[i λαj ] , (3.17)
where λ and χ are both totally antisymmetric in lower and upper indices and totally
traceless. The superfield Jij,kl carries the irrep D(2, 0, 0; 0, 2, 0). It belongs to series C and
it describes a BPS-1/2 multiplet. Its components can be shown to contain the composite
operators that correspond to the level ! = 0 supergravity multiplet shown in Table 2, and
that the components with spin s 1 are conserved currents.
The level ! = 1 supercurrent is also a special one and is known as the massless Konishi
multiplet. It has the simple form [56]
J = Wij W ij . (3.18)
As a result of the basic singleton constraint (3.15), this current obeys the constraint
D ij J = 0, D ij := D α(i Dαj ) . (3.19)
This multiplet has and they precisely correspond to the level ! = 1 massless states
5 × 28
shown in Table 2. It is characterized by the irrep D(2, 0, 0; 0, 0, 0) carried by its lowest
component. It is a semi-short multiplet which saturates the unitarity bound of series A.
(their complex conjugates have jL = 0), and E0 = jL + 1. Each singleton multiplet forms a massless UIR of
the d = 4, N = 4 Poincaré superalgebra and is characterized by central charge ! = 2|Z| = 0, 1, 2, . . . . Viewed
as massless states of d = 4 Poincaré group, they carry Lorentz spin (i.e., maximum SO(2) helicity) s = jL .
The first three levels of the singleton spectrum shown in Table 6 are special because they are the only singleton
multiplets which contain scalar fields. They are D(1, 0, 0; 0, 1, 0), D(1, 0, 0; 1, 0, 0) and D(1, 0, 0; 0, 0, 0) with Z
charges (0, 1/2, 1) and they can be described by superfields (Wij , W i , W ), respectively. The level ! singletons
D( 2! , 2! − 1, 0; 0, 0, 0) have central charge Z = 2! and can be described by superfield ωα1 ···α!−2 . The constraints
satisfied by all singleton superfields can be found in [50].
In the Poincaré limit, the states are labeled by the little group SO(3) × SU(4). Denoting
the irreps by Rs , where R is denotes an USp(8) irrep (which should be decomposed into
SU(4) irreps) and s is the SO(3) spin, the level ! = 1 massless Konishi multiplet can be
obtained by tensoring the level ! = 0 supergravity multiplet with an SU(4) singlet spin
s = 2 state as follows:
Massless Konishi: (420 + 481/2 + 271 + 83/2 + 12 ) × 12

= 10 + 81/2 + (27 + 1)1 + (48 + 8)3/2 + (42 + 27 + 1)2
+ (48 + 8)5/2 + (27 + 1)3 + 87/2 + 14 . (3.20)
The massless multiplets arising at level ! 2 in the spectrum shown in Table 2 are generic
in their structure. The corresponding conserved currents are contained in a superfield
Jµ1 µ2 ···µ2!−2 , ! 2, (3.21)

which obey the constraints [50]
µ β̇
σ̄ µ1 D iβ Jµ1 µ2 ···µ2!−2 = 0, σ 1 α D i β̇ Jµ1 µ2 ···µ2!−2 = 0. (3.22)
α̇β
The superfield Jµ1 µ2 ···µ2!−2 carries the irrep D(4! − 2, 2! − 2, 2! − 2; 0, 0, 0) and its
components have spins that range from (2! − 2) to (2! + 2). This superfield saturates
the unitarity bound of series A and it describes a semi-short multiplet.
The explicit construction of all the supercurrents in terms of Maxwell singleton is
straightforward but tedious exercise which apparently has not been carried so far. They
are known, however, for the minimal bosonic truncation of the massless HS gauge theory
in D = 5 discussed above. They take the form [8,13]

2!−2
(−1)k
jµ1 ···µ2!−2 = ∂µ · · · ∂µk φ ∗ ∂µk+1 · · · ∂µ2!−2 φ − traces. (3.23)
(k!)2 !((s − k)!)2 1
k=0
So far we have considered free SYM singletons. Switching on the SYM interactions, the
currents listed above for ! 1 will no longer be conserved. The resulting anomalies can
be characterized as follows
√
D ij J = λ Σ ij ,
µ √
σ̄ 1 α̇β D iβ Jµ1 µ2 ···µ!−2 = λ Σµi 2 ···µ2!−2 ,α̇ ,
µ β̇ √
σ 1 α D i β̇ Jµ1 µ2 ···µ2!−2 = λ Σiµ2 ···µ!−2 ,α , (3.24)
where the constant normalization factor is introduced for later convenience (see Section 5).
The superfields on the right-hand side carry the following UIRs of SU(2, 2|4)
Σ ij : D(3, 0, 0; 2, 0, 0), (3.25)

Σµ2 ···µ2!−2 ,α̇ : D 2! − 32 , ! − 32 , ! − 1; 0, 0, 1 , (3.26)

Σµ2 ···µ2!−2 ,α : D 2! − 32 , ! − 1, ! − 32 ; 1, 0, 0 . (3.27)
In the interacting SYM singleton theory the anomaly superfield Σ ij takes the well known
form (see, for example, [61,65]):
4
Σ ij = Tr W k(i W j )! Wk! , (3.28)
N 3/2
where the constant normalization factor is introduced for later convenience (see Section 5).
This superfield belongs to series B and it describes a BPS-1/8 multiplet. Consequently its
components go up to smax = 7/2 and therefore it is a candidate for coupling to Higgs
superfield in the bulk which can be eaten by the massless Konishi multiplet to become
massive. All the components of the massive Konishi multiplet of PSU(2, 2|4) have been
tabulated in [57].
The candidate anomaly superfields Σαi 2 ···α2!−2 ,α , on the other hand carries a semi-short
IUR that satisfy the unitarity bounds of series A or B. In general, such multiplets have been
constructed as [50]
S [ai ] = Φ 2 BPS[ai ] ,
S {µ1 ···µs }[ai ] = J {µ1 ···µs } BPS[ai ] , (3.29)
[ai ]
where BPS is any one of the BPS operators listed in (B.11)–(B.13), and J {µ1 ···µs } is a
spin s current, to be constructed out of the free SYM singleton in our case. For the BPS-
1/2 and BPS-1/4 cases, both of the above operators saturate the series A unitarity bound
(B.8), while in the case of BPS-1/8, they belong to series B. Assuming that the candidate
anomaly superfield Σαi 2 ···α2!−2 ,α carries an irreducible representation, and given that it is in
(100) of SO(4), attempting to construct it as in (3.29) requires the use of BPS-1/8 multiplet
D(3/2, 0, 0; 1, 0, 0), as follows from (B.13). However, these BPS multiples cannot be built
out of SYM singletons alone [50].
The BPS multiplets that can be constructed out of products of SYM singleton alone
are all the BPS-1/2 and BPS-1/4 multiplets listed in (B.11) and (B.12), and all those
BPS-1/8 multiplets listed in (B.13) with integer r [51]. These multiplets, and the semi-
short multiplets discussed above which make use of them, are likely to play a role in
finding the massive states of the full HS gauge theory based on hs(2, 2|4). In particular,
the KK supermultiplets associated with level ! supermultiplets of the massless HS theory
are expected to make use of the BPS-1/2 states. For example, the level ! = 0 multiplet and
its KK towers are realized as [3,58,59]
D(k, 0; 0, k, 0): W(a1 Wa2 · · · Wak ) − traces, k = 2, 3, . . . . (3.30)
Setting k = 2 gives the massless supergravity multiplet and k = 3, 4, . . . their massive KK
descendants. Similarly, the semi-short multiplets (3.12) and (3.13) involving the BPS-1/2
composites carrying the irrep D(k, 0; 0, k, 0) are candidates for KK descendants of the
level ! > 0 massless multiplets of the HS gauge theory based on hs(2, 2|4).
3.3. The d = 6, N = (2, 0) tensor singleton and its composites
The fundamental UIRs of OSp(8∗ |4) are the singletons given in Table 10 in Appendix A
[46,48]. Each row in the table denotes an irreducible singleton multiplet. The superfield
realization of the 6D singletons have been studied by several authors. Here we shall follow
[50,51] where several references to earlier literature can also be found. There exist several
papers on the construction of the composite operators out of the 6D singletons as well; see
[50,51,66,67], for example.
There exist an infinite set of singletons of OSp(8∗ |4). They are shown in Table 10 and
listed in Appendix B. The (2, 0) tensor singleton is the only one which is singlet under an
SU(2)Z defined in Section 2.3. Here we shall focus our attention to the level ! = 0 singleton
described by the superfield W ij which forms the tensor multiplet of d = 6, N = (2, 0)
Poincaré supersymmetry, since all the HS gauge theory states will be formed out of them.
To begin with, we shall take a single copy of the tensor multiplet. Abelian nature of
the singletons is essential for the construction of conserved currents. The superfield Wij
satisfies the following constraints and reality condition [66]
Dα(i W j )k = 0, ij = Ωik Ωj l W kl .

W (3.31)
The singleton superfield Wij carries the irrep D(2; 0, 0, 0; 0, 1) which belongs to series D
and it is BPS-1/2 supermultiplet. For each state in the spectrum of the HS gauge theory
listed in Table 3, one can construct the corresponding conserved current out of two tensor
singletons and their derivatives. To begin with, the level ! = 0 supercurrent is contained in
the superfield Jij,kl , which is in 14-plet of USp(4) and is given by
1
Jij,kl = Wij Wkl − Ωk[i Ωj ]! W mn Wmn . (3.32)
6
Defining W a ≡ (Γ a )ij W ij (a = 1, 2, . . . , 5), where Γ a are the SO(5) Dirac matrices, the
current superfield (3.32) can equivalently be written as Jab = Wa Wb − 16 δab Wc Wc .
The superfield Jij,kl carries the irrep D(4; 0, 0, 0; 0, 2), which belongs to series D and
it describes a BPS-1/2 multiplet. This is the level ! = 0 supergravity multiplet shown in
Table 3.
The level ! = 1 supercurrent is similar to the ones in d = 3, 4 and it takes the form
J = Wij W ij . (3.33)
This current obeys the constraint [50]
j
αβγ δ Dα(i Dβ Dγk) J = 0. (3.34)
The superfield J carries the irrep D(4; 0, 0, 0; 0, 0). It has 14 × 28 components and it can
be obtained group theoretically by tensoring the level ! = 0 supergravity multiplet with the
graviton state which has 14 degrees of freedom. It is a semi-short multiplet which belongs
to series B.
The massless multiplets arising at level ! 2 in the spectrum shown in Table 2 are
generic and the corresponding conserved currents are contained in the superfield
Jα1 ···α2!−2 ,β1 ···β2!−2 , ! 2, (3.35)

where the α and β indices are symmetrized separately. These current superfields obey the
constraint [50]
δγ α1 β1 Dγi Jα1 ···α2!−2 ,β1 ···β2!−2 = 0. (3.36)

The superfield Jα1 ···α2!−2 ,β1 ···β2!−2 carries the irrep D(2! + 2; 0, 2! − 2, 0; 0, 0). It is a semi-
short multiplet which belongs to series B.
An explicit construction of these supercurrents in terms of the (2, 0) tensor singleton
apparently has not been carried out so far. They are known, however, for the minimal
bosonic truncation of the massless HS gauge theory in D = 7 discussed earlier. They take
the form [8,13]

2!−2
(−1)k
jµ1 ···µ2!−2 = ∂µ · · · ∂µk φ ∗ ∂µk+1 · · · ∂µ2!−2 φ
k!(k + 1)!(s − k)!(s − k + 1)! 1
k=0
− traces. (3.37)
So far we have considered free (2, 0) tensor singletons. Interactions for multi-copies of
these singletons are not known and they are expected to be radically different than those
familiar from ordinary field theory. These interactions are also expected to break the HS
gauge symmetries down to those of level ! = 0 supergravity. Let us characterize the break-
down in the conservation laws of the supercurrents of level ! 1 as follows
j
αβγ δ Dα(i Dβ Dγk) J = gΣ δij k , (3.38)
δγ α1 β1
Dγi Jα1 ···α2!−2 ,β1 ···β2!−2 = gΣαδi2 ···α2!−2 ,β1 ···β2!−2 , (3.39)
where g is some coupling constant. Unlike in the cases of d = 3, 4, here we see that the
representation content of the candidate anomaly superfields do not correspond to any BPS
short or semi-short multiplets listed in Appendix B. Of course, here we are assuming that
these anomaly superfields are irreducible. Their computation from first principles may
in principle reveal that they are reducible, and possibly derivatives of some irreducible
superfields. The nature of the anomaly superfields should also reflect the fact that there
are no local non-Abelian interactions for tensor fields that can be described by continuous
deformations of the free theory [90]. This is a qualitative difference between d = 6 and
d = 3, 4, where the free fields admit SYM deformations (after dualization of a scalar in
d = 3).
The semi-short multiplets, as in 3D and 4D cases, have also been constructed in terms
of building blocks discussed above, and they take the form [50]
S [ai ] = Φ 2 BPS[ai ] , (3.40)

{µ1 ···µs }[ai ] {µ1 ···µs } [ai ]
S =J BPS , (3.41)
where BPS[ai ] is any one of the BPS operators listed in (B.20) and (B.21), and J {µ1 ···µs }
is a spin s current, which is to be constructed out of the free (2, 0) tensor singleton in our
case. Both of these saturate the unitary bound of series B.
The BPS multiplets that can be constructed out of products of the tensor singleton alone
are all the BPS-1/2 multiplets listed in (B.20) and all those BPS-1/4 multiplets listed in
(B.21) with integer q [51]. In particular, the level ! = 0 multiplet and its KK towers are
realized as [50]
D(2k, 0, 0, 0; 0, k): W(a1 Wa2 · · · Wak ) − traces, k = 2, 3, . . . . (3.42)

As in the cases of 3D and 4D, here too, setting k = 2 gives the massless supergravity
multiplet and k = 3, 4, . . . give their massive KK descendants. Similarly, the semi-short
multiplets (3.40) and (3.41) with BPS-1/2 composites carrying the irrep D(2k, 0, 0, 0; 0, k)
are candidates for KK descendants of the level ! > 0 massless multiplets of the HS gauge
theory based on hs(8∗ |4).
4. Higher spin gauge theory and holography
We shall first discuss some general features of HS gauge theory/singleton correspon-

dence before we turn to the cases of interest in type IIB string theory and M-theory. In
particular, the properties of the free boundary CFTs which indicate that the massless HS
gauge theories in the bulk provide effective descriptions of the full HS gauge theories trun-
cated to their massless sector will be emphasized.
Consider a CFTd consisting of N supersingletons W i , where i = 1, . . . , N is an
internal index and each W i belongs to some singleton representation of the superconformal
group. Let each singleton belongs to an irreducible representation of some internal
symmetry group G and consider G invariant composite operators O. Our first basic
assumption is that the correlation functions of invariant composite operators factorize as
N → ∞. For example, by using the operator product expansion, a four-point function
O1 O2 O3 O4 ! can be decomposed as
O1 O2 O3 O4 ! = O1 O2 ! O3 O4 ! + O1 O2 O3 O4 !conn , (4.1)
O1 O2 Or ! Or O3 O4 !
O1 O2 O3 O4 !conn = , (4.2)
r
Or Or !
where the disconnected terms are the contributions from the unit operator and the
connected terms are the contributions from the remaining operators. The factorization
means that the connected terms are suppressed by powers of 1/N :
O1 O2 O3 O4 !conn
→ 0 as N → ∞. (4.3)
O1 O2 ! O3 O4 !
In general, there can be several parameters in addition to N in CFTd . Fortunately,
supersymmetry puts considerable amount of constraint on these possibilities. With
application to type IIB string and M-theory in mind, we shall assume that G = SU(N) and
consider SU(N) valued singleton scalar superfields denoted by W I , I = 1, . . . , n. In this
case we have N = N 2 − 1 and the singletons transform in the fundamental representation
of the R-symmetry group SO(n). For the cases of interest, namely in d = 3, 4, 6, we have in
mind the R-symmetry groups SO(8), SO(6) and SO(5), respectively, which correspond to
16 ordinary plus 16 special supersymmetries in the CFTd . The SU(N) valued singletons in
d = 4 are adequate for discussing the tensionless limit of the type IIB theory on AdS5 × S 5 .
The extent to which SU(N) valued singletons in d = 3, 6 may encode the properties of (an
unbroken phase of ) M-theory on AdS4/7 × S 7/4 is discussed in Sections 6 and 7.
The basic composite operators in CFTd are primary bilinear single-trace operators
O(2)r , where the index r labels collectively the set of (SO(d, 2) × R)-representations
involved [7,8]. These operators do not mix with any other operators and provide conserved
HS currents with spin s 1, and certain composite operators of lower spin s < 1. Together
they form an HS multiplet that corresponds in a one-to-one fashion to an HS multiplet of
physical massless bulk fields,7 φ(2)r . In the supersymmetric singleton models of special
interest to type IIB/M-theory the bilinear primaries are discussed in Section 3 and the
corresponding massless spectra are listed in Tables 1, 2 and 3.
The free CFT d also contains composite operators which are pth order monomials in
the basic singleton and its derivatives. Those composites which are not normal ordered
products of other composites as N → ∞ are interpreted as massive single-particle states
in AdS. We shall denote these operators and the corresponding massive bulk fields by
O(p)r and φ(p)r , respectively, where p 3 and r is an additional set of indices labeling the
SO(d, 2) × R weights. The massiveness means that there is no shortening of the associated
SO(d, 2) weight spaces. This implies that the massive operators are not conserved and
hence there are no gauge symmetries associated with the corresponding massive AdS
fields. However, as discussed in the previous section, some of the massive operators belong
to shortened supermultiplets, provided that the superconformal weights saturate certain
unitarity bounds or belong to discrete series. This is the case, for example, for 1/2 BPS
KK modes and the Higgs multiplets listed in the previous section.
For fixed p the space of massive operators O(p)r clearly decomposes into irreducible
HS multiplets, though the representation theory of HS algebras, such as their root structure,
has not yet been developed far enough to characterize the precise ‘lowest’ weights carried
by these multiplets (see [20] for a discussion of this point).
Composite operators which are normal ordered products of other composite operators
as N → ∞ are interpreted as many-particle states. In the case of SU(N) valued singletons,
the single-particle states, O(p)r (p = 2, 3, . . .) are given in the large N limit by single-trace
operators. The n-particle states, which we shall denote by O(p1 ,...,pn )r are given in this limit
by multi-trace operators in the form of normal ordered products of single trace operators
O(pi )ri and their derivatives, pi = 2, 3, . . . , i = 1, . . . , n.
For finite N there is mixing between the single-trace and multi-trace operators [7,23].
This is because n-particle states in the bulk couple to operators that diagonalize the two-
point function:
OR OS ! = ηRS , (4.4)
where R = (p1 , . . . , pn )r and ηRS is an N -independent diagonal matrix. For example,
consider the minimal bosonic truncation based on a single SU(N) valued singleton field W .
The bilinear and tri-linear composites, which have to be single-traces, do not mix. However,
the quartic composites do mix, and they do so as follows. The diagonal scalar states of
energy ∆ = 2d − 4 are given schematically by
2f
O(4) = J(4) + f J(2,2), O(2,2) = J(2,2) − J(4) , (4.5)
1+f2
7 In the minimal bosonic truncation this dictionary has been extended to also include local currents
corresponding to the auxiliary HS gauge fields of the bulk theory [8]. This offers an opportunity to compute
bulk amplitudes in a first order formalism.
where f (N) = a
N + b
N3
, with a and b being some constants, and
1 4
J(4) ∼ 2
:tr W :, J(2,2) ∼ :tr W 2 tr W 2 :, (4.6)
N
are assumed to be normalized such that
J(4) J(4) ! = ∆4 , J(4) J(2,2)! = f (N)∆4 , J(2,2) J(2,2)! = ∆4 , (4.7)

where ∆ = |x|−d+2 is the singleton propagator.
Having introduced the main notation and kinematics, we now continue with the
discussion of the factorization of correlators as N → ∞. From (4.4) it follows that as
N → ∞ a general n-point correlator either vanishes if n is odd or can be written as the
sum of products of n/2 two-point functions. Thus, in the limit N → ∞ the singleton
CFT d describes an ‘antiholographic’ bulk theory of free n-particle states corresponding to
O(p1 ,...,pn )r . For finite N , the 1/N corrections to the singleton CFT give rise to non-trivial
connected parts of the correlation functions which we wish to represent as antiholographic
interactions. To be more precise we wish to examine whether the SU(N) valued singleton
field theory is the holographic dual of an interacting (d + 1)-dimensional theory based on
an effective action, consisting of a bulk term plus a boundary term
Γeff [φ(p)r ] = Γeff,bulk[φ(p)r ] + Γeff,boundary[φ(p)r ], (4.8)

which admit perturbative expansions in powers of 1/N around an AdSd+1 vacuum [21].
The boundary term plays a role in representing certain correlators, such as the extremal
correlators discussed below, that cannot be reproduced from a bulk action. This boundary
term is needed because the variational principle requires Γeff [φ(p)r ] to be stationary when
the fields are varied subject to Dirichlet conditions. The variation of Γeff,boundary[φ(p)r ]
should therefore cancel the total derivatives from the variation of Γeff,bulk[φ(p)r ] that give
rise to boundary terms that involve normal derivatives of the variations. For example,
Γeff,bulk[φ(p)r ] is expected to contain an ordinary R-term for the spin 2 fields and
consequently that Γeff,boundary[φ(p)r ] contains the corresponding Brown–York term.
In the case of 16 supersymmetries (4.8) can be expressed formally as
d 16
eiΓeff [φ(p)r (V )] = ei p,r d x d θ O(p)r V(p)r , (4.9)
where the effective action on the left-hand side is evaluated subject to boundary conditions
dictated by superconformal tensors V(p)r . These superfields are prepotentials for super-
Weyl multiplets containing the boundary conditions on the AdS curvatures.
The correlators of composite operators on the right-hand side of (4.9) are well-behaved
functions of the insertion points as long as they are separated. However, as these points
coincide, the correlators are in general rather badly behaved distributions. Thus a more
careful definition of the generating functional of correlators requires the choice of a
regularization scheme. This leaves room for anomalous effects, even though the singleton
theory is free, which may serve the purpose of selecting critical field content and number
of supersymmetries. In other words, consistency of the right-hand side of (4.9) in the case
of a free SCFT in d dimensions with finite sources for composite operators should be
about as restrictive as consistency of an interacting SCFT in d dimensions. Moreover,
the fact that a successful definition of (4.9) in principle would give rise to a consistent
bulk theory including quantum gravity8 suggests that only the special supersymmetric
singletons corresponding to limits of string/M-theory will be viable in the above sense.
Thus we shall assume that ultimately (4.9) makes sense only for free SCFTs in d 6 with
less than or equal to 16 supersymmetries.9 We address these issues further below when
we discuss the subleading 1/N corrections to the definition of the vacuum used in the
correlator on the right-hand side of (4.9).
The generating functional makes sense only as an asymptotic expansion in 1/N in
which a given order is a formal power series expansion in φ(p)r , which has a finite radius
of convergence by the combinatorial counting rules for double line diagrams of fixed
topology. From the normalization (4.4) and assuming that O! = 0 it follows that as far
as the 1/N counting goes the effective action has the form

1 1 1 1
Γeff [φ] = φ + f3
2
φ + 2 f4
3
φ4 + · · · ,
N N2 N N2

1 1
fn ∼1+O . (4.10)
N2 N2
The singleton field theory determines Γeff [Φ(p)r ] up to non-linear field redefinitions of the
type φ → φ + N1 φ 2 + · · · . After rescaling the fields as
φ = NΦ, (4.11)
we define the classical action as follows

Γeff [Φ] = Γcl [Φ] + O 1/N 2 , (4.12)

N2
Γcl [Φ] = d−1 d d+1 x L Φ, R∂Φ, (R∂)2 Φ, . . . + boundary term, (4.13)
R
where R is the AdS radius. We can now state the properties of the HS gauge theories as
follows. They possess:
(a) a set of one-particle states forming HS multiplets;

(b) a corresponding set of ‘vertex operators’ of a free CFT d ;
(c) a fundamental mass scale, 1/R where R is the AdS radius, and a fundamental
expansion parameter, lPl /R where the Planck length lPl determines the normalization
8 As the basic mechanism behind holography is general covariance, this raises the question whether
holography exhibits any new features as general covariance is extended by HS symmetries. To analyze this,
we presumably need to refine our present, mainly algebraic, understanding of HS symmetries by formulating
these in a more geometric language, perhaps by extending the set of spacetime coordinates as to realize HS gauge
transformations as extended reparametrizations [11].
9 Massless HS fields admit background independent self-interactions in D = 4, and it is most likely that
this is the case for all D (though interactions in D > 7 bring in symplectic spacetime symmetries). However,
the theories of massless HS fields in higher dimensions are presumably not consistent truncations of quantum
consistent theories.
of the effective AdS action to be10
1 N
d−1
= . (4.14)
lPl R d−1
Given these facts we would like to determine the effective action Γeff [φ(p)r ] from a set of
bulk interactions, without any direct reference to the boundary singleton. The basic issue is
whether the interactions can be derived from a string or membrane sigma model, that can
be coupled to the HS background fields. The mass-scale of the HS spectrum is set by the
AdS radius R, which is suggestive of a sigma-model with a fixed critical tension of order
1 in units where the AdS radius R = 1, as we shall discuss further in Sections 5, 6 and 7.
Due to the absence of mass-gap it is not possible to separate the massless fields, φ(2)r ,
from the massive AdS fields, φ(p)r , p > 2, by taking a low energy limit. In a local process in
AdS with energies of the order E ∼ n/R, n 1, the massive modes with E0 < n/R behave
essentially as the KK modes which arise in an AdS compactification of string/M-theory.
Thus the only reasonable possibility in which the massless modes can be separated from the
massive modes in a HS theory is by consistent truncation to the massless sector,11 which
is similar to what happens in the (maximally supersymmetric) sphere compactifications of
type IIB and eleven-dimensional supergravities. There are examples, however, of compact
manifolds, such as T 1,1 , where the higher-dimensional supergravity theory does not admit
a consistent truncation despite the fact that there does exist a lower-dimensional gauged
supergravity.12
Thus we propose that the HS gauge theories in D = 4, 5, 7 with gauge groups hs(8|4),
hs(2, 2|4) and hs(8∗ |4) admit consistent truncation down to the corresponding massless
theories, which we described in Section 2. This consistent truncation can be directly
tested by verifying that the massless bulk theory reproduces exactly the correlators of the
corresponding bilinear operators in the singleton theory. This is a non-trivial test since
nothing is known about higher-dimensional covariant description of the HS theory so far.
Consistent truncation of the full HS gauge theory to its massless sector requires that
there are no terms in the effective bulk action of the form φ(p) φ(2) · · · φ(2) for p 3. Let
us show this in the case of scalar bulk fields. Then the corresponding singleton correlators
are non-zero provided that ∆(p) n∆(2) where n 2 is the number of massless fields.
The case ∆(p) = n∆(2) is called an extremal correlator. The extremality condition implies
p = 2n and in that case it is straightforward to use free field contraction rules to show that
n
2
O(p) (x)O(2)(x1 ) · · · O(2) (xn ) = ∆(x − xi ) , (4.15)
i=1
10 In the case of SU(N ) valued singletons N = N 2 − 1, which means that the Planck constant in the bulk is
given by h̄ = 1/N 2 . The 1/N corrections to the bulk theory are therefore weighted by positive integer powers of
the Planck’s constant.
11 We thank L. Rastelli for helpful discussions on this point.
12 We thank C. Pope for pointing this to us.
where ∆(x) = |x|−d+2 is the singleton propagator. Consider, on the other hand, the bulk
integral
d+1
d z
I= d+1
K∆(p) (z, x)K∆(2) (z, x1 ) · · · K∆(2) (z, xn ), (4.16)
z0
where K∆ (z, xi ) is the standard bulk-to-boundary propagator. K∆ (z, xi ) ∼ z0d−∆ δ d (z − xi )

for small z0 , and as z → xi ,

dz0 −∆+n∆(2)
n
2
I∼ z0 ∆(x − xi ) . (4.17)
z0
i=1
Thus, in the extremal case this integral diverges logarithmically, and the residue of the pole,
treating ∆ as a variable, has the same structure as the extremal correlation function. By
assumption, the antiholographic dual should, however, give rise to finite amplitudes. The
resolution is that a term which diverges logarithmically is scale-invariant, which means
that it can be represented equivalently by a boundary term which is finite. Thus extremal
correlators give rise to couplings that are boundary terms and therefore they do not upset
the consistent truncation.
A similar argument applies to the near-extremal case, when d − 2 < ∆ < n∆(2) .
Here the integral I is finite, but the dependence on the x’s is not of the same form
as the singleton CFT correlator. There are exchange diagrams, though, with the correct
structure of the x-dependence [5]. Thus the near-extremal correlators must be represented
antiholographically in terms of exchange diagrams, and there cannot be any contact term
in the bulk action that can upset consistent truncation we are examining.
The above evidence for consistent truncation is similar to the one given for ordinary type
IIB supergravity AdS × S 5 [25,26] and eleven-dimensional supergravity on AdS4/7 × S 7/4
[27]. The main difference is that whereas the arguments in SUGRA only holds for 1/2
BPS states, the arguments given here for HS theory hold for more general operators since
the holographic dual is by assumption a singleton.
To provide further evidence for consistent truncation, we examine the correlator of four
massless scalar operators Oi = O(2) (xi ), i = 1, . . . , 4. Using free field theory contraction
rules it can be written on manifestly crossing symmetric form as
O1 O2 O3 O4 ! = η12 η34 + η14 η23 + η13 η24 + O1 O2 O3 O4 !conn , (4.18)

O1 O2 O3 O4 !conn = A(s,t
1234
)
+ A(t,u)
1324 + A(u,s)
1243 , (4.19)
where
(x,y)
Aij kl = :Oi Ok ::Oj Ol :!conn (4.20)
(x,y)
and x and y denote in which of the s-, t- and u-channels the quantity Aij kl has
singularities. In the limit x12 , x34 → 0, the correlator can be expanded in the s-channel
by using the OPE
O1 O2 = η12 + C12 (2)r O(2)r (x2 ) + C12 (4)r O(4)r (x2 ) + C12 (2,2)r O(2,2)r (x2 ), (4.21)
where we recall that O(2)r denotes the set of all primary bilinear single-trace operators
labeled by an index r, and O(4)r and O(2,2)r are as given in (4.5). The resulting s-channel
expansion is given by
O1 O2 O3 O4 !s -ch = η12 η34 + C12 (2)r C34,(2)r + O1 O2 O3 O4 !s -ch,finite , (4.22)

where CRST = CRS U ηU T = OR OS OT !, and O1 O2 O3 O4 !s -ch,finite , which is finite in the
s-channel, is given by
(t,u)
O1 O2 O3 O4 !s -ch,finite = :O1 O2 ::O3 O4 :! = η13 η24 + η14 η23 + A1324, (4.23)
(t,u)
A1324 = C12 (4)r C34,(4)r + C12 (2,2)r C34,(2,2)r . (4.24)
The structure of (4.24) appears to be problematic for consistent truncation, and cannot be
ignored in the large N limit as follows from
1 1
C(2)(2)(2)r ∼ , C(2)(2)(4)r ∼ , C(2)(2)(2,2)r ∼ 1. (4.25)
N N
It is possible, however, to write (4.24) in a more tractable form as a manifestly crossing
symmetric sum of terms involving only exchange of bilinear operators. To this end we
first note that the crossing symmetry of the singleton theory implies that the complete
s-channel expansion (4.22) is equal in the sense of analytical continuation to the complete
t- and u-channel expansions in the limits x14 , x23 → 0 or x13 , x24 → 0, respectively. Thus
(4.22) must contain contributions that are singular in the t- and u-channels. From the
form of (4.19) we therefore deduce that the singular part of (4.22) actually must consist
of two separate contributions, one which becomes singular in the t-channel and another
one which becomes singular in the u-channel. We also see that the problematic term in
(4.22) must have singularities in both the t- and u-channels, which by crossing symmetry
should describe massless exchanges. In fact, from (4.21) and free field theory contraction
rules it follows that
A(s,t )
1234 ≡ :O1 O3 ::O2 O4 :!conn = 2 C12
1 (2)r
C34,(2)r = 12 C32 (2)r C14,(2)r , (4.26)
A(t,u)
1324 ≡ :O1 O2 ::O3 O4 :!conn = 1
2 C13
(2)r
C24,(2)r = 1
2 C14
(2)r
C23,(2)r , (4.27)
(u,s)
A1243 ≡ :O1 O4 ::O2 O3 :!conn = 1
2 C12
(2)r
C43,(2)r = 1
2 C13
(2)r
C42,(2)r . (4.28)
To show the second equality in (4.26) we first use (4.21) to expand the single contraction
connecting O1 to O2 in terms of C12 (2)r O(2)r and similarly for 3 and 4. The remaining two
contractions that contribute to the connected part give rise to 12 ηrs , where the factor of 12
arises due to the normal ordering prescription which forbids contractions connecting 1 with
2 and 3 with 4, respectively. The third equality in (4.26) follows by instead using (4.21) to
expand the single contractions connecting 1 to 4 and 3 to 2. The relations (4.27) and (4.28)
are obtained analogously. Eqs. (4.26) and (4.28) imply that the finite contribution (4.24)
can be rewritten in terms of partial wave expansions involving only exchange of bilinear
operators in the crossed channels. We also see that the crossing symmetric form of the
singular contribution C12 (2)r C34(2)r in (4.22) is given by 12 C12 (2)r C34(2)r + (1 ↔ 2). Thus
the complete four-point correlator can be written in a manifestly crossing symmetric form
involving only massless partial waves:

O1 O2 O3 O4 ! = η12 η34 + η23 η14 + η24 η13 + O1 O2 O3 O4 !conn , (4.29)
O1 O2 O3 O4 !conn
1 1 1
= C12 (2)s C34,(2)s + C23 (2)t C14,(2)t + C13 (2)u C42,(2)u. (4.30)
2 2 2
This generalizes so that any correlator of bilinear operators can be written as a manifestly
channel duality invariant sum of conformal blocks involving only exchange of bilinear
operators.
The test of holography requires that the result (4.30) is consistent with that obtained
from the corresponding Witten diagram that uses the classical action of the HS gauge
theory truncated to its massless sector. Now it has been shown in [68] that a Witten
diagram with four external scalars and exchange of an internal scalar φ equals the sum
of the conformal block with exchange of the scalar operator O coupling to φ plus terms
which have the same structure as, but do not exactly agree with, the conformal blocks with
exchange of operators corresponding to the two-particle states formed out of the external
scalar states. In [68] it has also been shown that a Witten contact diagram with four external
scalars has the form of two-particle exchange.
Thus (4.30) has the form required by consistent truncation, provided that the quartic
bulk interactions in the cases of interest lead to cancellation of the parts in bulk four-point
amplitudes that have the structure of conformal blocks with massless two-particle state
exchange. The remaining terms, which come from the Witten diagrams with exchange of
massless bulk fields, can then be written in manifestly s–t–u channel duality invariant form
as conformal blocks with exchange of the corresponding bilinear operators, as in (4.30).
Thus, the bulk side of the story remains to be established. It would be interesting to examine
to what extent the requirement that two-particle partial waves must cancel determines the
structure of higher order interactions in the action for massless fields. We shall return to
this point below in discussing the interaction ambiguity in the massless sector.
Having gathered evidence for the consistent truncation, let us now proceed to explore
some of its consequences. In the above discussion, we have implicitly made the assumption
that the correlators in the free singleton theory are given by ordinary vacuum expectation
values on a conformal plane. Let us assume that this is indeed correct in the large N limit.
The generating functional for correlators of bilinear operators is then given for large N by a
one-loop functional determinant, i.e., the connected n-point correlators are planar diagrams
that scale like N −(n−2) . Thus, the effective bulk action for the massless fields is ‘classical’
and takes the form

N2
Γcl [Φ(2)r ] = d−1 d d+1 x L Φ(2)r , R∂Φ(2)r , (R∂)2 Φ(2)r , . . .
R
+ boundary terms, (4.31)
where R is the AdS radius. The Lagrangian contains higher derivative interactions and the
quadratic part is ghost and tachyon free. It is important to note that the quantity R∇ is not
small in an expansion around AdS.
By construction, both Γcl [Φ(2)r ], and the full classical action Γcl [Φ(p)r ] defined in
(4.12), reproduce the correlators of bilinear operators holographically to the leading orders
in the 1/N expansion, i.e., the extrema of the two actions are equal provided the massive
modes Φ(p)r , p 3 are set to zero at the boundary of AdS. The consistent truncation can
now be phrased as the stronger condition
Γcl [Φ(2)r ] = Γeff [Φ(p)r , Φ(3) = 0, . . .]. (4.32)

This offers the following possibility to test consistent truncation directly. Based on the
results in D = 4, we expect that HS gauge symmetry together with the requirement of
manifest local Lorentz symmetry determines a family of actions
S[φ(2)r ; V] (4.33)
for massless fields where V represents a set of arbitrary parameters. As explained in Sec-
tion 6.2, and in more detail in [19], there exist an interaction ambiguity
in the 4D HS gauge
theory which involves the introduction of an odd function V(x) = ∞ n=1 b 2n+1 x 2n+1 . Al-
ready the simplest choice V(x) = b1 x gives rise to a highly non-trivial model with a struc-
ture of the type indicated in (4.31). The nth order term in V(x) results in higher order
derivative corrections starting at order 2n + 2 in the Lagrangian. Thus, in D = 4 the con-
sistent truncation (4.32) implies a specific choice V(x) = VΓ (x) such that
Γcl [φ(2)r ] = S[φ(2)r ; VΓ ]. (4.34)

Thus, consistent truncation means that there exists a set of parameters VΓ for which the
extremum of S[φ(2)r ; VΓ ] corresponds to the generating functional of correlators of bilin-
ear operators in the singleton theory. A perturbative scheme for obtaining the interactions
in D = 4 to any desired order is given in [18,19], and described in Section 6 for the case
of quadratic terms in the field equations. We are still lacking the description of the full
interactions for massless fields in D > 4, though we expect that the basic building blocks
are of the kind described in [9,10,12,20].
The supersymmetric HS gauge theories in D = 4, 5, 7 can be truncated consistently to
a minimal bosonic HS theory with massless and massive fields. Moreover the massless
minimal bosonic theory is a consistent truncation of the massless supersymmetric HS
theory. Hence, if the truncation of the massive modes is consistent in the supersymmetric
theory then this must also be the case in the minimal bosonic theory. In particular, in D = 4
the interaction ambiguities in the supersymmetric theory and the minimal bosonic theory
are parametrized by the same function V.
Let us now examine more closely the qualitative behavior of the 1/N dependence of
some singleton correlators involving higher than second order traces of singletons. For
example, the connected part of the correlator of four cubic scalar single-trace operators,
O(3) ∼ N13/2 :tr(W 3 ): (these operators do not mix with any other operators) contains both
planar and non-planar double-line graphs which scale like 1/N 2 and 1/N 4 , respectively,
in the large N limit. Another interesting example is the correlator of three O(4) operators,
which contain 1/N and 1/N 3 contributions. In the supersymmetric case, one can arrange
the cyclic orders of R-symmetry indices carried by the singletons to cancel the leading
1/N contribution, and thus the corresponding cubic coupling in the effective action [22].
Hence, the full singleton theory encodes information about a non-trivial 1/N expansion of
the antiholographic dual.
As we have already mentioned, we think of these corrections as being generated by

a quantum theory in the bulk which is generated by a string theory or some other sigma
model which can be coupled to the massless HS fields. From this point of view, it would
be natural to have subleading 1/N corrections also to the interactions in the massless
sector, so that (4.31) would only be valid for large N . We would also expect corrections
to Γeff [Φ(p)r ] which violate the consistent truncation (4.32). These effects do not arise,
however, if we treat the correlators in the singleton theory as ordinary vacuum expectation
values of operators inserted on the conformal plane.
We conclude this section by speculating on possible subleading in 1/N corrections to
the free singleton correlators on the right-hand side of (4.9). To this end, let us assume
that the free singleton theory in question is an actual limit of a CFT describing the low
energy dynamics of open string modes in string theory or ‘open membrane’ modes in
M-theory. For concreteness, let us consider the case of the SU(N) invariant singleton
theory that arises as a limit of the d = 4, N = 4 SYM theory. For finite open string length
the prescription for computing open string theory amplitudes is to attach open string vertex
operators to open string boundaries and sum over all open string fluctuations. This includes
virtual processes including formation of closed string loops. A closed string loop can be
created by inserting a ‘sewing operator’

Rstr = s (0)
Vs (z)V (4.35)
s
on the string worldsheet where the sum runs over a complete set of physical closed string
states. In taking the low energy limit leading to the conformal SYM theory, the physical
effect of the sewing operation is included into the 1/N expansion of the SYM theory with
2 . Thus the limit g 2 → 0 is not smooth in the sense that the closed string sewing
finite gYM YM
operations, which are present for any finite gYM 2 2 = 0, simply because
are absent for gYM
there are no virtual processes in the singleton theory that leads to the addition of internal
‘handles’ in the 1/N expansion. This is reminiscent of the fact that the deformation of the
free singleton theory corresponding to switching on finite gYM2 cannot be described directly
at the level of the composite operators built from the singleton superfield, which contains
the Abelian field strength but not any explicit gauge potential. In fact, this requires that we
introduce gauge couplings by hand, after which gYM 2 can be shifted to any finite value by
marginal deformations.
The above arguments suggest that we modify the definition of the generating functional
in the singleton theory by working with full singleton correlators given schematically by
1
O1 · · · On !full = R k O1 · · · On , (4.36)
k!
k
where the (super)conformally invariant singleton sewing operator R is defined as the sum
over a complete set single trace operators describing a virtual closed string process:
dd x dd y
R= ηrs (x − y)O(p)r (x)O(p)s (y). (4.37)
p
|x − y| 2d
Since each power of R adds an extra power of 1/N 2 , the above definition does not affect
the classical limit though it yields the desired non-trivial subleading 1/N corrections to the
correlators. The insertion of R formally corresponds to taking a trace, which in turn implies
that the correlation function becomes periodic along a cycle on the conformal plane. In
string theory, Rstr acts similarly, and has the geometric effect of adding a handle to the
two-dimensional worldsheet. This suggests that R insertions describe large fluctuations
of the D3 brane worldvolume in the singleton limit. As in the closed string theory, the
consistency of the sewing operation in the free singleton theory may lead to restrictions on
the spacetime superdimension.
In summary, we propose to use HS symmetries in diverse dimensions to determine
actions (or field equations) for massless HS multiplets up to certain well-defined interaction
ambiguities and then to compare the resulting Witten amplitudes with correlators of
bilinear operators in corresponding large N singleton theories. The next step in this
program is to explain the consistent singleton/HS correspondences as limits of string
and M-theories, which, in particular, require the identifications of possible schemes for
breaking HS symmetries.
We emphasize that the tests of CFT/AdS in the HS regime involve a free CFT on the
boundary, unlike the tests in the supergravity regime where the boundary CFT is strongly
coupled. This is possible due to the proposed consistent truncation and the fact that there
still remains the expansion parameter 1/N .
It is not clear exactly how the state of affairs will change once the HS symmetries are
broken. In Section 3 we have identified candidate Higgs multiplets in d = 3, 4. Presumably
this can be done also in d = 6 provided that we develop the proper mathematical language
for describing the interactions on the M5-brane. In general, we expect that the Higgsing
upsets the consistent truncation to the massless sector alone. Moreover, it is not obvious
if there exists a generalized consistent truncation scheme that retains the massless, Higgs
and other relevant massive fields. In any event, it will be interesting to see whether HS
field theoretic methods can be used to describe the Higgsing or one has to resort to some
more basic definition of the bulk interactions, based on some sigma model. We believe it
is too early to make any conclusive remarks on this, though it seems possible to describe
couplings between massless HS fields and Higgs fields, which should form HS multiplets
fitting into master fields of the type discussed in Section 2.
In Sections 5–7 we shall discuss these issues in more detail, and case by case for the
theories described in Section 2.
5. Type IIB on AdS5 × S 5 and 5D higher spin gauge theory
According to the strong version of the Maldacena conjecture [1,4,5] d = 4, N = 4 SYM

theory with SU(N) gauge group, gauge coupling gYM 2 and ’t Hooft coupling λ = Ng 2 is
YM
equivalent to type IIB string theory on AdS5 × S of radius R with string coupling gstr and
5
string length lstr given by

gstr = f1 (λ)gYM
2
, f1 (λ) ∼ 1 for λ 1,
−1/4
lstr = f2 (λ)R, f2 (λ) ∼ λ for λ 1. (5.1)
For large λ, these relations are deduced by interpolating between the AdS5 vacuum × S5
with radius R and dilaton eφ = gs , and the ten-dimensional Minkowski vacuum, using
the classical D3-brane solution with harmonic function H (r) = 1 + 4πNgs ls4 r −4 . The
functions f1,2 (λ) account for possible string corrections to the interpolating region, where
only 16 supersymmetries are preserved. The type IIB string/4D SYM correspondence is
an AdS/CFT correspondence whereby the 4D SYM theory is identified as the holographic
dual of the type IIB closed string theory. The closed string theory is based on a non-linear
sigma-model with coupling constant ls /R. A (dimensionless) closed string amplitude A(str)
has the doubly asymptotic expansion
∞
2g−2
A(str) = gs A(str)
g (ls /R), (5.2)
g=0
(str)
where the amplitude Ag (ls /R), which is obtained from worldsheet perturbation theory
on a Riemann surfaces of fixed genus g, is given by an asymptotic expansion in ls /R. The
5D Planck length is given by
1 N2
3
= . (5.3)
lPl R3
Thus the perturbative string expansion in AdS5 × S 5 makes sense provided that
N 1, gs 1, ls R. (5.4)
The ’t Hooft expansion of the corresponding correlation function A(SYM) in the SYM
theory reads
∞

A(SYM) = N 2−2g A(SYM)
g (λ), (5.5)
g=0
where the amplitude A(SYM) g (λ) is obtained from double-line Feynman graphs with
fixed topology and is given by an analytical expansion in λ. Hence the conjectured
correspondence A(str) = A(SYM) can be examined order-by-order in string loop expansion
and SYM 1/N expansion, leading to a set of strong/weak coupling dualities between
A(str)
g (ls /R) and Ag
(SYM)
(λ).
As discussed earlier, it has been proposed that the HS gauge theory emerges in the
description of the type IIB string theory on AdS5 × S 5 in the limit [9,21–23]
gs → 0, ls → ∞; N 1, R fixed. (5.6)
In this limit the dual free SYM theory is described by an SU(N) valued d = 4, N = 4
SYM singleton. As discussed in the previous section, the bulk physics is conjectured to
be an HS gauge theory in 5D which admits a consistent truncation to an effective action
Γcl [Φ(2)r ] for massless fields. The HS gauge group hs(2, 2|4) and its massless gauge theory
has been described in [10]. We emphasize that there should be direct agreement between
the individual terms in the 1/N expansions of massless gauge theory amplitudes and the
correlators of bilinear currents in the free CFT as described in (4.34) (without having to
first obtain strong coupling results).
There still remains the task of constructing the full interacting HS gauge theory in
5D, though cubic interactions for massless spin s = 2, 4, 6, . . . fields have already been
constructed in [12]. These form a subset of the cubic interactions of the minimal bosonic
truncation Sbos of S[φ(2) ; V] provided that it is consistent to set the scalar field φ in Sbos
equal to zero at the cubic level. This requirement means that Sbos must not have any cubic
interactions that are linear in φ and quadratic in spin s 2 fields. On the other hand, from
the known stress-energy tensor OPEs (see, for example, Eq. (4.58) in [7]), it follows that
the effective action Γeff [φ(2)] should give rise to a non-zero cubic graviton–graviton–scalar
amplitude. Thus the scalar can only be consistently truncated at the cubic level if this
amplitude is represented by a boundary term in Γeff [φ(2)], i.e., if the correlator in question
is extremal or near-extremal. Whether or not this is the case remains to be seen.
We next discuss breaking of the HS symmetry. The level ! = 0 supergravity multiplet
of the massless spectrum of the hs(2, 2|4) theory contains a dilaton, ϕ which is an SU(4)
singlet with energy ∆ = 4 and AdS mass m2 = 0. Since m2 = 0 it is consistent to give ϕ
a VEV in the linearized theory, and we shall assume that this is possible also in the full
HS gauge theory. This corresponds to switching on a finite gYM 2 in the 4D SYM theory. As
i
result the 4D supercovariant derivative Dα becomes also gauge covariant. This does not
upset the stress-energy conservation law (3.17), as it is first order in the superderivative,
while it breaks the Konishi multiplet conservation law (3.19), which is second order in
derivatives. Using the relation

D ij W kl = −2gYM W k(i , W j )l (5.7)
which follows from the superspace formulation of the N = 4 SYM system in 4D, one finds
that the anomalous conservation law for the Konishi current is given by (see, for example,
[61,65]):
4gYM √
D ij J = tr W k(i W j )l Wkl ≡ λ Σ ij . (5.8)
N
The operator Σ ij belongs to the massive Higgs multiplet with smax = 7/2 discussed in
2 the anomalous conservation law (5.8) describes how Σ ij is
Section 3. Thus, for finite gYM
‘eaten’ by the massless Konishi operator J to form a massive operator which belongs to the
long massive Konishi multiplet with smax = 4 containing 216 states. The coupling between
the corresponding bulk fields, which are described on the boundary by prepotentials V and
Vij , and the massless Konishi operator J and its Higgs descendant Σ ij is described by

Sboundary = d 4 x d 16 θ J V + Σ ij Vij . (5.9)
2 , the action S
For finite gYM boundary is invariant under modified gauge transformations
involving a Stückelberg shift transformation of the massive Higgs field,
√
δV = D ij Λij , δVij = − λ Λij . (5.10)
We thus expect that for finite ϕ! = gs the√
effective action Γeff [φ(p)r ] contains kinetic terms
of the schematic form |dφ(2) |2 + |dφ(3) + λ φ(2) |2 , describing a single massive gauge field
with non-critical mass [21]
λ
m2 − m2crit ∼ 2 , (5.11)
R
where (D 2 − m2crit )φ = 0 for an AdS massless field φ.
As discussed in Section 3, the massive spectrum also contains 1/2 BPS massive states
that have the interpretation of KK modes built on the massless HS multiplets. We shall
assume that the Higgs mechanism can be described at the level of KK towers as well, and
that the remaining massive HS multiplets can be organized into massive HS multiplets and
their KK towers. This picture is suggestive of a covariant theory in D = 10 with ‘critical’
length scale l10 and coupling constant g = 1/N which admits AdS5 × S 5 with radius
R = l10 as a vacuum. Since HS interactions in AdS spaces blow up in the flat limit for
finite g, we do not expect the 10D HS theory to admit 10D Minkowski space as a vacuum
for finite g > 0. For g = 0 we get a quadratic Lagrangian, however, which is second order
in derivatives, and as it contains no positive powers of R, it does admit a flat space limit.
Thus, the tensionless limit of the type IIB string theory in 10D flat spacetime is trivial.
Higgsing of the critical theory leads to a non-critical theory with l10 < R which for
l10 R should be identified with type IIB string theory in AdS5 × S 5 with ls ∼ l10 .
For small ls /R the spectrum of string states with AdS energy (measured in units of
1/R) satisfying the condition E R 2 / ls2 , and spin s R 2 / ls2 , can be obtained by
KK reducing the 10D Minkowski space spectrum on S 5 by means of group theoretical
methods (at the classical level these states are described by ‘short’ strings with energy
E = Rl/ ls2 and length l R). In particular, for fixed SO(4) × SO(6) highest weight, the
worldsheet Hamiltonian has a ground state which is the ‘lightest’ state carrying that highest
weight. The lightest states correspond to the leading Regge trajectory in 10D Minkowski
space and form supermultiplets in both 10D Minkowski space and in AdS5 × S 5 with
smax = 2, 4, 6, . . . . In 10D Minkowski space these arise at closed string level ! = 12 smax − 1
(see, for example, [69]), where all multiplets are massive except for level ! = 0 where the
supergravity multiplet resides. For example, the lightest smax = 4 multiplet is the massive
Konishi multiplet which resides at level ! = 1.
As ls /R varies from ls /R 1 to ls /R 1 the different Regge trajectories do not mix
[6] even though the five-form flux and other terms of order 1 in units of R will become
comparable to the mass-term. This follows from the fact that in an exact CFT that admits
a perturbative formulation, such as the worldsheet theory and the boundary SYM theory,
there cannot be mixing between two operators that do not mix in the free theory. Note that
such an admixture would require the introduction of a mass-parameter in the perturbative
formulation, which is not compatible with conformal invariance.
Indeed, there is an exact agreement between the supermultiplet structures of the leading
Regge trajectory for large string tension and the set of massless states of the critical
hs(2, 2|4) theory, such that the level ! multiplet on the leading Regge trajectory flows,
after reversed Higgsing, to the level ! multiplet of the massless spectrum given in Table 2.
We have already argued in Sections 4 and 5 that there should exist a consistent
truncation of the full hs(2, 2|4) theory down to its massless sector. There is no analogous
truncation of the non-critical string theory down to the leading Regge trajectory because the
lightest states of level ! 1 consist of massless states plus Higgs states. The Higgs states
belong to the massive sector of the hs(2, 2|4) theory and therefore break the consistent
truncation.
Since the HS symmetries are broken spontaneously it would be interesting to construct
a HS field theoretic description in AdS of the couplings between the massless fields and the
Higgs fields. Clearly the master field formalism described in Section 2 should be useful in
doing this, though one presumably needs to invoke some additional information, perhaps
from the structure of the factorization of the SYM correlation functions for λ 1. Thus we
should try to find a HS action S(Φ(2)r , Hr ; V, M) for massless fields Φ(2)r and Higgs fields
Hr , where V are the parameters describing the gauge interactions, as will be discussed in
Section 6.2, and M are the parameters describing the coupling of the gauge multiplet
to the massive Higgs fields. We can then study the issue of whether the ‘weak/weak’
version of the AdS/CFT correspondence, which is valid for the massless sector at λ = 0,
can generalized to include the leading Regge trajectory for λ > 0.
The above identification of the leading Regge trajectory states with long strings was also
made recently by the authors of [34], who conjecture that it is possible to follow the bilinear
HS currents with large spin s from weak to strong ’t Hooft coupling, where they correspond
to long strings of length l ∼ R which describe the portion of the leading Regge trajectory
with large spins s R 2 / ls2 1 and AdS energies E = s + R 2 / ls2 log(sls2 /R 2 ). As the
long strings grow infinite in size they become open strings of infinite energy which couple
to bi-local, light-like Wilson lines whose operator product expansion contains the bilinear
HS currents. The long strings have a relatively √ large ratio of energy to spin as compared
to the short strings which have energies E = s R/ ls . Thus, for finite gs and small string
length, a string scattering process at high energies is described by incoming long strings
which fall into AdS spacetime where they fragment into short strings which interact and
then recombine into outgoing long strings. Moreover, the high-energy scattering process
with leading Regge trajectory states corresponds to a CFT correlator with bilinear currents
of large spin and relatively small anomalous contribution, (E − s)/s → 0 as s → ∞. This
suggests that the extreme high energy scattering can be described by the massless HS gauge
theory.
The bilinear currents do not mix with other operators in the free field theory, which
means that they cannot mix at any finite order in perturbation theory either. In the world
sheet sigma-model the counterpart to this statement is that the vertex operators describing
the insertion of long string states at small sigma-model coupling ls /R should ‘flow’ without
mixing to vertex operators at large sigma-model coupling [6]. Moreover, it is expected that
the long strings, which are quantum-mechanically unstable for finite gs , become stable as
! → 0 since ls /R increases (which removes the short string states from the spectrum) and
gs becomes small (which suppress the decay process). This suggests that as λ → 0 there
remains a non-trivial worldsheet sigma-model describing stable long string states which
correspond to the singleton single-trace operators.
From the above discussions we are led to propose that there is a cross-over from large
to small λ in the expressions for the AdS string length and string coupling in terms of the
gauge theory quantities given in (5.1), such that
f1 (λ) ∼ 1/λ, f2 (λ) ∼ 1 + O(λ) for λ 1. (5.12)
Then ls /R ∼ 1 and gs ∼ 1/N as λ → 0. This suggests that the hs(2, 2|4) higher spin
gauge theory is described by a string theory which has a left-moving and right-moving
PSU(2, 2|4) KM algebra with critical level k = kcrit ∼ 1 which admits a singleton
representation and an affine hs(2, 2|4) extension. To be more precise, the critical value for
the level should be such that there exists a maximally reducible Verma module based on the
singleton which contains a maximal number of null-states. In fact, it has been shown [70]
that the affine SO(3, 2) Sp(4) algebra admits singleton-like representations for k = 5/2.
It would be interesting to generalize this result to SO(D − 1, 2) and supersymmetric cases.
For critical level the closed string spectrum then contains physical massless HS states
formed by multiplying a left-moving and a right-moving singleton. The algebra hs(2, 2|4)
can be identified with the following coset

hs(2, 2|4) = Env PSU(2, 2|4) /R, (5.13)
where R is a certain ideal generated by elements in Env(PSU(2, 2|4)) which vanish
identically when the PSU(2, 2|4) generators are realized in terms of a single super-
oscillator as described in Section 2.1. For k = kcrit this construction should lift to the
affine case. The symmetry enhancement from AdS group to HS algebra for critical level,
i.e., critical radius in units of fixed string length, would be similar in spirit to the SU(2)
enhancement occurring at the self-dual radius for string theory on a circle.
The possibility to realize massless higher spins directly in the bulk as products of
left-moving and right-moving singleton representations at critical KM level is rather
appealing. Perhaps the close resemblance between the HS gauge theories in D = 4, 5, 7
is an indication of that singletons play a similar role on critical membranes in D = 4, 7.
6. M-theory on AdS4 × S 7 and 4D higher spin gauge theory
6.1. Holography
Already in [71] it was observed that the OSp(8|4) singleton may play a role in the
description of the supermembrane on AdS4 × S 7 . In [40] the quantization of the d = 3,
N = 8 singleton theory corresponding to a single membrane was shown to yield the infinite
set of massless HS fields contained in the symmetric tensor product of two singleton weight
spaces [38]. Moreover, it was conjectured in [40] that these massless states, as well as the
massive states contained in the higher order tensor products, arise in the supermembrane
theory.13 Subsequently, the group theoretical HS/singleton connection was utilized in [35]
and the fully interacting massless HS field equations in D = 4 were constructed in [14].
In the light of [1], the 4D HS/singleton connection found in [40] was revived as an actual
AdS/CFT correspondence in [16,17].
Importantly, the role of large N discussed in Section 4 was not emphasized in these
early formulations of the correspondence. Thus we need to refine the formulation of the
correspondence by identifying the appropriate dependence on N of the free OSp(8|4)
singleton.
Let us first recall the Maldacena conjecture [1] on the correspondence between
M-theory on AdS4 × S 7 with N units of 7-form flux on S 7 and the low energy dynamics
of N parallel M2-branes in flat eleven-dimensional spacetime, which is described by a
strongly coupled d = 3, N = 8 CFT with SO(8)R symmetry [1,4]. This theory defines a
13 To describe the S 7 compactified M-theory all higher tensor products are needed. The resulting theory lives
on the double cover of AdS4 times S 7 . It is consistent to truncate the theory to only even powers of the singleton.
This corresponds to M-theory on the single cover of AdS4 times S 7 /Z2 RP7 .
non-trivial IR fixed point of d = 3, N = 8 SYM theory with SU(N) gauge group. The
resulting SO(7)R -invariant flow has an antiholographic description as a D2-brane near-
horizon geometry, which is reliable in the UV where the dilaton is small. In the IR the
dilaton blows up and the IIA solution lifts to the SO(8)R invariant AdS4 × S 7 near horizon
region of a stack of N coinciding M2-branes. The resulting antiholographic description of
the strongly coupled SCFT is conjectured to be M-theory on AdS4 × S 7 . For large N the
membrane tension scales like
√
1 N
TM2 = 3 ∼ 3 , (6.1)
lM2 R
where R is the AdS radius, and the 4D Planck length is given by
1 N 3/2
2
= . (6.2)
lPl R2
Hence, for large N ,
R lM2 lPl . (6.3)
For AdS energies E obeying 1 E R/ lM2 the low-energy dynamics of the
antiholographic dual is conjectured [1] to be described by D = 4, N = 8 gauged
supergravity. In particular, it follows from the normalization (6.2) that the strongly coupled
SCFT has ∼ N 3/2 massless degrees of freedom for large N [72,73].
In the UV limit of the D2-brane geometry the dilaton eφ vanishes and the 10D
gravitational curvature diverges (which one might interpret as the appearance of the new
massless HS states that we shall define below). The D2-brane field theory becomes a
SU(N) invariant theory of free 3D super-Maxwell multiplets. Here we note that the
Yang–Mills coupling in the dual SYM theory on the stack of N coinciding D2-branes,
2 = g / l , is held fixed in taking the near-horizon limit. This coupling also coincides
gYM s s
with the ‘local’ Yang–Mills coupling on a stack of probe D2-branes placed at energy
scale u in the near-horizon region, gYM 2 (u) ≡ e φ(u) −g (u)/ l 2 = g 2 , as required for
00 s YM
interpreting the stack of probe branes as describing a Higgs branch of the dual SYM theory.
Thus both the dilaton and running string length vanishes in the UV limit, which is why we
can trust the free SU(N) field theory even though the gravitational curvature diverges.
Dualizing the vector fields and using gYM 2 to rescale the fields, we obtain a free SU(N)
valued OSp(8|4) singleton Φ ∈ 8v described by the SO(8)R invariant Lagrangian14

i

2
d 3 x tr ∂Φ i + fermions . (6.4)
Conversely, assuming that this Lagrangian describes a fixed point on the membrane we can
break SO(8)R → SO(7)R by taking the M-theory to have a finite radius R11 and take Φ 8
to be periodic:
Φ 8 ∼ Φ 8 + g, (6.5)
14 The singleton consists of 8 scalars in 8 and 8 spinors in 8 of SO(8) . By triality one can also obtain a
v s R
singleton multiplet in which the scalars are in 8s and the spinors in 8c .
where the radius g is a constant with dimension 1/2 which we identify as g = R11 /(l11 )3/2
and l11 is the eleven-dimensional Planck length. We recover the free OSp(8|4) invariant
singleton in the decompactification limit R11 → ∞. We may instead use g to dualize Φ 8
and introduce Yang–Mills interactions with gYM = g. The effective coupling is geff 2 =
2
g /u, where u is the 3D energy scale, and as a result the theory now decompactifies in
the IR [1,4,25,31,74]. Thus we have two decompactification limits, the free SU(N) valued
OSp(8|4) singleton field theory which resides in the UV and the strongly coupled SO(8)R
invariant d = 3, N = 8 SCFT in the IR.
Thus it is natural to describe the low energy dynamics of M2-branes in terms of an UV
fixed point of free SU(N) valued OSp(8|4) singletons and an IR fixed point of strongly
coupled OSp(8|4) singletons. We note that the number of massless degrees of freedom
indeed decreases along the RG flow, from N 2 to N 3/2 .
We conjecture that the free singleton theory at the UV fixed point mentioned above
is the holographic dual of the hs(4|8) gauge theory which admits the massless hs(4|8)
gauge theory described in Section 2.2 as a consistent truncation. This theory describes an
unbroken phase of M-theory with N units of M2-brane charge. The strongly coupled fixed
point is the holographic image of a broken phase, which admits an effective supergravity
description at low energies.
There are also IR fixed points containing free OSp(8|4) singletons forming (N − 1)-
dimensional representation of the Weyl group of SU(N) [31]. These are curious points
from the point of view of HS dynamics, and it may be that one should also include them as
non-trivial points in the phase diagram.
As discussed in the previous section, the unbroken phase of the type IIB theory on
AdS5 × S 5 √ arises either as the critical limit λ → 0 at fixed E and s, or as the high energy
limit s λ at fixed λ and N 1. Moreover, as we shall see in the next section,
the unbroken phase of M-theory on AdS7 × S 4 arises at high energies whereby certain
membrane solitons propagate close to the boundary of AdS7 . This suggests that also the
unbroken phase of M-theory on AdS4 × S 7 arises in a high energy limit in which bulk
membranes couple to HS operators in the strongly coupled SCFT 3 with asymptotically
small anomalous dimensions, (E − s)/s → 0, as s → ∞. The four-form flux in the AdS4
directions results ensures the M2-brane equations admit spherical membrane solutions in
AdS4 × S 7 [75–77]. These solutions carry internal SO(8) spin, and are hence closely related
to the matrix-model found in the pp-wave limit [30]. It is natural to expect that these
solutions can be deformed into time-dependent membrane solutions carrying also AdS
spin, in analogy with the string solutions in AdS3 with NS-fluxes [78]. We also expect the
anomalous part of the energy to be minimized and certain fractional supersymmetry to be
restored by taking large AdS radius, i.e., large bulk energies, such that the solution couples
to the conserved HS currents of the hs(8|4) theory. The fact that the holographic dual
resides at a UV fixed point should be encoded into the local geometry of the solution and
to how it minimizes the AdS energy, as in the case of the rotating membrane in AdS7 × S 4 .
It will be interesting to examine the above picture in more detail and in particular to
examine the fluctuation spectrum about this solution, where we expect to find some critical
membrane theory with fixed tension, and perhaps singletons in the worldvolume, giving
rise to the massless HS states.
We expect that the Higgsing of the massless HS fields and the resulting spontaneous
breaking of the hs(4|8) is described by a radially dependent solution to the HS theory which
is the antiholographic dual of the 3D SYM flow obtained by switching on a finite gYM 2 as
discussed above. It will be interesting to see whether HS field theoretic methods are still
relevant for describing this solution, which would then yield ‘weak/weak’ correspondence
between the HS theory coupled to Higgs sector and the SYM theory with expansions in
both 1/N and gYM 2 . It may also be necessary to exhibit in more detail the nature of the
above-mentioned critical membrane.
6.2. Cubic couplings in the 4D higher spin gauge theory
In this section we shall outline the structure of the minimal bosonic HS gauge theory
in D = 4 which is a consistent truncation of the supersymmetric HS theory discussed
in Section 2.1. The spectrum consists of massless fields with spin s = 0, 2, 4, . . . , each
occurring once. The underlying algebra, called hs(4), is an infinite-dimensional extension
of the bosonic AdS4 group. Similar truncation exists also in D = 5, 7 at the spectrum level
but only in D = 4 a full interacting theory is known, both supersymmetric and minimal
bosonic.
The 4D minimal bosonic model is of great interest because it is the simplest interacting
HS gauge theory (with propagating HS degrees of freedom), and yet it exhibits all the
essential principles that underlie such theories. It is a very good starting point for finding
ways to construct the D = 5, 7 HS gauge theories as well. Moreover, it is amenable to
calculations and it is possible to test directly in this model the consistent truncation of the
kind discussed in Section 4 which is required for the holography picture to make sense.
Here, we will not go as far as carrying out these tests [24] but we will nonetheless exhibit
the structure the couplings to give the reader an idea about how they actually look like, as
well as providing enough ingredients to facilitate the required holography computations.
Here we shall focus our attention on the quadratic terms in all the field equations, which,
of course, mean all the cubic couplings at the action level. In an accompanying paper
[19], we shall give a more detailed treatment involving an expansion scheme where the
gravitational gauge fields are treated exactly and the gravitational curvatures and the HS
gauge fields as weak perturbations to all orders. The 4D HS/3d singleton correspondence
in the hs(4) theory at the level of quadratic field equation/cubic action will be provided
elsewhere [24].
The massless field equations (including general interaction ambiguities) have been
given in [14] and studied in more detail in [16–18] and more recently in [19]. These studies
are based on a curvature expansion scheme. The most important step in the expansion
scheme is the linearized analysis which shows that all auxiliary fields are non-propagating.
As a result it is possible to solve iteratively for the auxiliary fields and obtain the physical
field equations to any order. In fact, this scheme yields field equations in terms of only the
physical fields.
The HS spin algebra hs(4) is obtained from hs(4|8) defined in Section 2.1 by setting the
fermionic generators θ i equal to zero. To describe the field equations in 4D spacetime,
which has coordinates x µ , one introduces an auxiliary set of coordinates (zα , z̄α̇ )
which are Grassmann even spinors that are non-commutative in nature, and consider
extensions ϕ(x; z, z̄) of the basic spacetime fields ϕ(x). One then imposes an integrable
curvature constraint in the extended space, whose (x; z, z̄)-components determine the (z, z̄)
dependence of the extended fields ϕ(x; z, z̄) in terms of “initial” conditions φ(x). Setting
z = z̄ = 0 in the remaining x-components of the curvature constraint leads to reduced
curvature constraints in spacetime, which are integrable by construction and one can show
that they contain the physical field equations of the HS gauge theory. Since (z, z̄) are
non-commutative, the reduced constraints contain interactions even though the original
constraint in (x; z, z̄) space has a simple form.
The basic building blocks of the theory are a master 0-form Φ and a master 1-form
Â = dx µ Âµ + dzα Âα + d z̄α̇ Âα̇ , (6.6)

where the hats are used to indicate quantities that depend on (z, z̄). The hatted fields are
given as expansions order by order in z and z̄, with expansion coefficients which are
functions of (x, y, ȳ) with the (y, ȳ) expansions determined by suitable group theoretical
conditions. These conditions are engineered such that at z = z̄ = 0, the pulled-back
components
Aµ = Âµ |Z=0 , Z=0

Φ = Φ| (6.7)
define an hs(4) valued spacetime one-form and a spacetime zero-form in a certain quasi-
adjoint representation of hs(4) [18]:
1 1 α̇1
Aµ (x; y, ȳ) = ȳ · · · ȳ α̇m y α1 · · · y αn Aµα1 ···αn α̇1 ···α̇m (x), (6.8)
2i m!n!
m+n=2 mod 4
1 α̇1
Φ(x; y, ȳ) = ȳ · · · ȳ α̇m y α1 · · · y αn Φα1 ···αn α̇1 ···α̇m (x). (6.9)
m!n!
|m−n|=0 mod 4
The curvature constraints giving rise to the spacetime field equations read
= Â + Â Â = i dzα ∧ dzα Φ
F
i
κ + d z̄α̇ ∧ d z̄α̇ Φ
κ̄, (6.10)
4 4
Φ
D = dΦ + Â Φ− Φ π̄(Â) = 0, (6.11)
where the operators κ, κ̄ are defined as

κ = exp iy α zα , κ̄ = κ † = exp −i ȳ α̇ z̄α̇ , (6.12)
the π -map, and its complex conjugate π̄ , acting on an arbitrary polynomial f (y, ȳ; z, z̄)
are defined as

π f (y, ȳ; z, z̄) = f (−y, ȳ; −z, z̄), π̄ f (y, ȳ; z, z̄) = f (y, −ȳ; z, −z̄), (6.13)
and the -product between two arbitrary polynomials f (y, ȳ, x; z̄) and g(y, ȳ; z, z̄) is
defined as
←− − −
← → →
− ←− ←− − → →
−
∂ ∂ ∂ ∂ ∂ ∂ ∂ ∂
f ∗ g = f exp i + − +i − + g.
∂zα ∂yα ∂zα ∂y α ∂ z̄α̇ ∂ ȳα̇ ∂ z̄α̇ ∂ ȳ α̇
(6.14)
The constraints (6.10) and (6.11) have the gauge symmetry
δ Â = d ˆ + [Â, ˆ ] , = ˆ φ̂ − Φ
δΦ .
ˆ (6.15)
Given the initial conditions (6.7), the components of the constraints (6.10), (6.11) which
in powers of Φ, which
have at least one α or α̇ index can be solved by expanding Â and Φ
contains curvatures and the scalar field, as follows:
∞
∞
∞

=
Φ (n) ,
Φ Âα = Â(n)
α , Âµ = Â(n)
µ , (6.16)
n=1 n=1 n=0
where

Φ, n = 1,
(n)
Φ = (6.17)
Z=0 0, n = 2, 3, . . . ,

Aµ , n = 0,

µ Z=0 = 0,
Â(n)
n = 1, 2, 3, . . . ,
(6.18)

Âα Z=0 = 0, (6.19)
and Φ (n) (n = 2, 3, . . .), Âα (n = 1, 2, 3, . . .) and Âµ (n = 2, 3, . . .) are nth order in Φ.

(n) (n)
Note that Â(n)

µ are linear in Aµ . The condition (6.18) is a physical gauge condition, which
can be imposed by using the gauge symmetry (6.15).
As shown in detail in [19], one first solves iteratively the constraints F µα = F αβ =

Fα̇ β̇ = 0 and Dα Φ = 0 to determine the Φ expansions of Âµ , Âα and Φ, which
schematically take the form
Âµ = Âµ [Aµ , Φ], Âα = Âα [Φ], = Φ[Φ].

Φ (6.20)
µν = 0
Having solved the Z-space part of (6.10) and (6.11), the remaining constraints F

and Dµ Φ = 0 yield spacetime field equations of the form
∞
n
(j ) (n−j )
Fµν = − Â[µ Âν] , (6.21)
Z=0
n=1 j =0
∞
n

) (j )
Dµ Φ = (j ) π Â(n−j
Φ µ
)
− Â(n−j
µ Φ , (6.22)
Z=0
n=2 j =1
where F = dA + A A and DΦ = dΦ + A Φ − Φ π̄(A).

Next, we define the physical scalar φ and expand the master gauge field Aµ (x, y, ȳ) as
Φ|Y =0 = φ. (6.23)
As for the vierbein and HS gauge fields, requiring that they transform homogeneously
under Lorentz transformation, one is led to the following expansion scheme for the master
gauge field

Aµ = eµ + ωµ + Wµ + iωµαβ Âα Âβ − h.c. Z=0 , (6.24)
where the vielbein and the Lorentz connections are defined as

1 α α̇ 1 αβ
eµ = e yα ȳα̇ , ωµ = ωµ yα yβ + h.c. (6.25)
2i µ 4i
and Wµ contains the fields with spin s = 4, 6, 8, . . . and their corresponding auxiliary fields.
We are now ready to state the result for the cubic couplings [19]. They give rise to
quadratic terms in the field equations given by [19]

2 i µ α α̇ ∂ ∂
∇ + 2 φ = ∇ Pµ − σ µ (2)
P (2)
, (6.26)
2 ∂y α ∂ ȳ α̇ µ Y =0

µνρ β̇ µνρ β̇ ∂ ∂ (2)
σ R
α νρ β̇ γ̇
= σ α
J νρ , (6.27)
∂ ȳ β̇ ∂ ȳ γ̇ Y =0
µνρ β̇ (1)
σ F
α1 νρα2 ···αs−1 β̇ γ̇ α̇2 ···α̇s−1

β̇ ∂ ∂ ∂ ∂
= σ µνρ α · · · · · · J (2)
, (6.28)
1 ∂y α2 ∂y αs−1 ∂ ȳ β̇ ȳ α̇s−1 νρ Y =0
where Rµν α̇ β̇ ≡ Fµν α̇ β̇ is the (self-dual part of) the AdS4 valued Riemann curvature, while
the curvature associated with spin s = 4, 6, 8, . . . fields is defined as
(1)
Fνρα
2 ···αs−1 β̇ γ̇ α̇2 ···α̇s−1
= 2∇[ν Wρ]α2 ···αs−1 β̇ γ̇ α̇2 ···α̇s−1 − (s − 2)(σνρ σµ )α2 δ Wµα3 ···αs−1 β̇ γ̇ δ̇ α̇2 ···α̇s−1
− s(σµ σνρ )β̇ γ Wµγ α2 α3 ···αs−1 γ̇ α̇2 ···α̇s−1 . (6.29)
The covariant derivatives in (6.26) and (6.29) are with respect to lo the Lorentz
connection ω. Furthermore, in (6.28) and (6.29), separate symmetrization in the dotted
and undotted indices is understood. Further definitions are
Pµ(2) = Φ π̄ (Wµ ) − Wµ Φ

(2) π̄(eµ ) − eµ Φ
+ Φ π̄ êµ(1) − êµ(1) Φ + Φ (2) , (6.30)
Z=0

(2)
Jµν =− ν (1) + Wµ , êν(1)
êµ(1) êν(1) + eµ , êν(2) + eµ , W Z=0

(1)
+ iRµν αβ Â(1) α Âβ + h.c. Z=0 + Wµ Wν + (µ ↔ ν) (6.31)
and the hatted quantities occurring in the above equations are given by [19]
1
i
α = − zα
Â(1) t dt Φ(−tz, ȳ)κ(tz, y), (6.32)
2
0
1 1
(1)
Â(2)
α = zα t dt Â (1)β
Â(1)
β z→t z,z̄→t z̄ + z̄ β̇
t dt Â(1)
α , Âβ̇ z→t z,z̄→t z̄ , (6.33)
0 0
1
dt ∂Wµ α(1) ∂Wµ
µ(1)
W = −i , Â + Âα̇(1), ,
t ∂y α ∗ ∂ ȳ α̇ ∗ z→t z,z̄→t z̄
0
1

Φ(2) =z α
dt Φ π̄ Â(1)
α − Â(1)
α Φ t →t z,z̄→t z̄
0
1
(1) (1)
+ z̄α̇ dt Φ π Âα̇ − Âα̇ Φ t →t z,z̄→t z̄ ,
0
1
dt (1)
êµ(1) = −ieµα α̇ α ∗ + Âα̇ , yα ∗ z→t z,z̄→t z̄ ,
ȳα̇ , Â(1)
t
0
1
dt (2)
êµ(2) = −ieµα α̇ α ∗ + Âα̇ , yα ∗ z→t z,z̄→t z̄ ,
ȳα̇ , Â(2)
t
0
1 1
dt dt
− eµα α̇
t t
0 0

∂ ∂ (1)
× Â β(1)
− ȳ α̇ , Â (1)
α ∗
+ Â α̇ , y α ∗ z→t z,z̄→t z̄
∂zβ ∂y β

∂ ∂ (1)
+ Âβ̇(1) + ȳα̇ , Â(1)
α ∗
+ Â α̇ , y α ∗ z→t z,z̄→t z̄
∂ z̄β̇ ∂ ȳ β̇

∂ ∂ (1)
+ β
+ β α ∗ + Âα̇ , yα ∗
ȳα̇ , Â(1) Âβ(1)
∂z ∂y z→t z,z̄→t z̄

∂ ∂
+ −
∂ z̄β̇ ∂ ȳ β̇

(1)
× ȳα̇ , Â(1)
α ∗ + Â α̇ , y α ∗
Â β̇(1)
. (6.34)
z→t z,z̄→t z̄ z→t z,z̄→t z̄
In the above formulae, the replacement of (z, z̄) by (tz, t z̄) is to be made inside the integrals
and after performing the -products. Note also the quantity Â(1) α is a basic building block
which occurs in many of the formulae above and that it is first order in Φ.
It is important to note that not all the fields occurring in (6.8) and (6.9) are independent.
An analysis of the constraints (6.10) and (6.11) shows (a) Φα1 ···α2s (s = 2, 4, . . .) are the
Weyl tensors which can be in terms of the curvatures, (b) Φα(m)α̇(n) for m + n > 2 can
αβ
be solved in terms of φ, the Weyl tensors and their derivatives, (c) ωµ is, of course,
the Lorentz spin connection which can be solved in terms of the vierbein eµα α̇ , and (d)
Wµα(m)α̇(n) for |m − n| 2 are auxiliary gauge fields which can be solved in terms of the
physical fields Wα(s−1)α̇(s−1) [19].
The general solution for the auxiliary fields is given by [19]
Wα α̇,β1 ···βm β̇1 ···β̇n

2
= ∇Wβ̇1 β̇2 ,αβ1 ···βm α̇β̇3 ···β̇n
m+1

2n n−1
+ α̇ β̇1 ∇Wβ̇2 γ̇ , αβ1 ···βm γ̇ β̇3 ···β̇n
n+1 m+n+2
n+1
+ ∇W(α γ ,β1 ···βm )γ β̇2 ···β̇n
m+n+2

m
− αβ ∇W γ δβ2 ···βm β̇2 ···β̇n
γδ
(m + 1)(m + 2) 1
+ mαβ1 ξβ̇2 ···βm α̇β̇1 ···β̇n , n > m 0, (6.35)
α1 α̇1 Φα2 ···αm α̇2 ···α̇n ,
Φα1 ···αm α̇1 ···α̇n = −i ∇ (6.36)
where the modified covariant derivatives are defined by

αβ,γ1 ···γm γ̇1 ···γ̇n = 1 σ µν
∇W
(2)
∇µ Wν,γ1 ···γm γ̇1 ···γ̇n − 12 Jµν,γ1 ···γm γ̇1 ···γ̇n , (6.37)
2 αβ

α1 α̇1 Φα2 ···αm α̇2 ···α̇n = ∇α1 α̇1 Φα2 ···αm α̇2 ···α̇n − P (2)
∇ α1 α̇1 ,α2 ···αm α̇2 ···α̇n , (6.38)
and separate total symmetrization of dotted and undotted indices is understood. Since J
and P depend on the auxiliary fields, Eqs. (6.35) and (6.36) must be iterated within the
curvature expansion scheme. This leads to explicit expressions of all auxiliary components
of Wµ and Φ in terms of the remaining physical fields.
Further comments about the above results are in order:
(1) The z-dependence of all the fields involved are exhibited. The above results are
explicit and the remaining task is reduced to performing certain star products and doing
some elementary parameter integrals. These steps, as well as the derivation of the above
results and their generalization to all orders will be provided elsewhere [19].
(2) It is easy to rewrite the field equation (6.27) for the graviton as15

αβ ∂ ∂ (2)
Rµν (ω) − gµν = σµ λ J + (µ ↔ ν) + h.c. , (6.39)
∂y α ∂y β λν Y =0
where Rµν (ω) is the Ricci tensor obtained from the Riemann tensor associated with the
Lorentz connection ωµ . It is important to note that this connection contains torsion as can
be seen from (6.35) and (6.37) which for m = 0, n = 2 give
ωµ ab = ωµ ab (e) + κµ ab , (6.40)
where κµ ab is the con-torsion tensor related to the torsion tensor Tµν a as
κµ ab = Tµ ab − Tµ ba + T ab µ , (6.41)
15 We have set the AdS radius R = 1 but it is straightforward to re-introduce R by dimensional analysis in
which the master 0-form and the master 1-form fields are dimensionless.
where

∂ ∂ (2)
Tµν a = σ a α β̇ J . (6.42)
∂yα ∂ ȳβ̇ µν Y =0
(3) The elimination of the auxiliary fields by means of Eqs. (6.35) and (6.36) gives rise
to higher derivative interactions. In particular, in a given spin sector, the auxiliary fields
are Wµα1 ···αk α̇k+1 ···α̇2s−2 with k = 0, 1, . . . , s/2 and they are related to the physical fields
Wµα(s−1)α̇(s−1) schematically as
Wµ,α(m)α̇(n) ∼ ∂ |m−n|/2 Wµ,α(s−1)α̇(s−1), m + n = 2s − 2. (6.43)

Similarly, the components Φα(m)α̇(n) of the master scalar field are related to the Weyl
tensors which are purely chiral, their derivatives as well as the derivatives of the scalar
as (taking m > n without loss of generality)
Φα(m)α̇(m) ∼ ∂ m φ,
Φα(m)α̇(n) ∼ ∂ (m−n)/2 Φα(m−n) , m − n = 0 mod 4. (6.44)
(4) Whether the master constraints (6.10) and (6.11) are unique is an important question.
In fact, there exist a generalization of (6.10) in which [19] we let

Φκ → V Φ κ , Φ κ̄ → VΦ κ̄ , (6.45)
† ) = (V(X))† . In [19] we
where V(X) is a -function, with its complex conjugate V(X
argue that this function must be of the form
∞

V(X) = b2n+1 X2n+1 , |b1 | = 1. (6.46)
n=0
A similar interaction ambiguity is expected to arise in HS theories yet to be constructed in

5D and 7D as well, and implications of this are discussed in Section 4, in the context of
5D HS gauge theory and holography. In particular, we argued that the freedom in choosing
b2n+1 is important in order to find the precise agreement between the bulk amplitudes and
the boundary correlators required for the massless theory to be a consistent truncation.
7. M-theory on AdS7 × S 4 and 7D higher spin gauge theory
The low-energy dynamics of a stack of N parallel coinciding M5-branes in flat eleven-

dimensional spacetime is described by a strongly coupled d = 6, N = (2, 0) SCFT with
SO(5)R symmetry group [1,4], known as the AN−1 (2, 0) theory [31,79–81]. The theory
is conjectured to have no marginal operators, which means that it describes an isolated
UV fixed point of the renormalization group. It is not known whether the theory has
any relevant operators which preserve the R-symmetry (the supergravity dual description
provides relevant operators which break the R-symmetry). Conversely, starting in the IR
with a number, N say, of free d = 6, N = (2, 0) tensor singletons, it is not known
how to describe non-Abelian interactions among tensor fields; in fact, there are no local
perturbations with this effect [82].16 This is believed to reflect the fact that open membranes
ending on coinciding M5-branes give rise to tensionless closed strings and that the proper
language for formulating the dynamics on the fivebrane is therefore not ordinary field
theory but rather some nonlocal extension of it.
However, if we are willing to give up 6D covariance, then we can use lower-dimensional
RG flows based on ordinary interacting field theories to define the AN−1 (2, 0) theory [1,
4,25,31,74]. In particular, circle reductions of the 6D theory describes RG flows of 4D
and 5D SYM theories with SU(N) gauge group. The SO(4)R invariant RG5 flow has a
type IIA supergravity dual description in terms of the near horizon region of a D4-brane
solution. In the UV limit the dilaton diverges and the solution uplifts to the AdS7 × S 4
near horizon region of the stack of M5-branes. The resulting antiholographic description
of the AN−1 (2, 0) theory is conjectured to be M-theory on AdS7 × S 4 [1]. For large N the
membrane tension scales like
N
TM2 ∼ , (7.1)
R3
where R is the AdS radius, and the 7D Planck length is given by
1 N3
5
= . (7.2)
lPl R5
For large N the Planck length is much smaller than the M2 length scale, lM2 which in
turn is much smaller than the radius. Thus, for energies E obeying 1 E R/ lM2 the
low-energy dynamics of the antiholographic dual is described by D = 7, N = 2 gauged
supergravity. The AN−1 (2, 0) theory has been conjectured to admit an expansion in terms
of integer powers of 1/N which factorize for large N [4].17 From (7.2) it follows that the
AN−1 (2, 0) theory has ∼ N 3 massless degrees of freedom for large N which contain the
N − 1 massless (2, 0) tensor multiplets of the ‘Higgs branch’ of the theory.
In the IR limit of the D4-brane geometry the dilaton eφ vanishes and the gravitational
curvature diverges. As for the D2-brane discussed in Section 6.1, the dual SYM coupling
2 = g l is held fixed in taking the near horizon limit and equals the local Yang–Mills
gYM s s
coupling gYM2 (u) ≡ e φ(u) / −g (u)/ l 2 = g 2 . Hence the local string length diverges in
00 s YM
the IR (unlike the case of the D2-brane where the local string length disappears together
with the dilaton in the UV). Hence, naively the D4-brane field theory becomes a free
SU(N) valued d = 5, N = 2 Maxwell theory with SO(5)R symmetry and finite Yang–Mills
coupling gYM2 . This theory can be made scale invariant by absorbing g 2
YM into the fields,
but this symmetry is superficial since it cannot be lifted to superconformal invariance.
Instead a more natural interpretation is that superconformal invariance is restored by
uplifting to a free SU(N) valued d = 6, N = (2, 0) tensor singleton described by the
16 A single tensor multiplet admits self-interactions, such as, for example, those describing the motion of a
single M5-brane [83–85].
17 From (7.1) it follows that M-theory on AdS × S 4 has an expansion in terms of integer powers of 1/T
7 M2
rather than integer powers of the 7D Plank’s constant. The same remark applies to M-theory on AdS4 × S 7 , which
√
has M2 tension given by (6.1) and has been conjectured to have an expansion in terms of integer powers of 1/ N
[4].
superconformal action18

2
S6 = d 6 x tr dΦ a + |dB|2 + fermions , (7.3)
where Φ a (a = 1, . . . , 5) and Bµν have dimension 2. The superspace formulation of this

theory is described in more detail in Section 3.3. The 6D superconformal invariance can
be spontaneously broken by compactifying (7.3) on a circle of radius R11 (after which R11
can of course be used to rescale the fields):

1
S5+1 = d 5 x tr dφ a + |dA|2 + KK modes and fermions , (7.4)
R11
where all bosonic fields have dimension 1. Taking S5+1 as the generic starting point
for describing the fivebrane dynamics there are thus two ways in which the theory can
decompactify and become superconformally invariant: either by directly taking R11 → ∞
which yields back S6 ; or by throwing away the KK modes and switching on the Yang–
2
Mills interaction with gYM = R11 , which then reaches the AN−1 (2, 0) limit in the UV
limit R11 → ∞. Note that the Yang–Mills deformation does not lead to loss of degrees
of freedom since the KK modes are exchanged with monopoles with mass proportional to
2 = R −1 .
gYM 11
We conclude that it is natural to describe the low energy dynamics of N coinciding
M5-branes in terms of an IR fixed point of free SU(N) valued d = 6, N = (2, 0) tensor
singletons and a UV fixed point given by the AN−1 (2, 0) theory in the unbroken HS phase.
We note that the number of massless degrees of freedom indeed decreases along the RG
flow, from N 3 to N 2 .
We conjecture that the free singleton theory at the IR fixed point mentioned above is the
holographic dual of an hs(8∗ |4) gauge theory which admits a consistent truncation to the
massless hs(8∗ |4) gauge theory in D = 7 described in Section 2.2. This theory describes an
unbroken phase of M-theory with N units of M5-brane charge. The strongly coupled fixed
point is the holographic image of a broken phase which admits an effective supergravity
description at low energies.
As in the case of M-theory on AdS4 × S 7 , there are also curious IR fixed points
consisting of N − 1 free tensor multiplets acted upon by the Weyl group of SU(N) [31]
which should also be included as non-trivial points in the phase diagram of M-theory on
AdS7 × S 4 .
We can motivate further our proposal by constructing and examining the properties of
‘long membrane’ solutions to the M2-brane action [86]

SM2 = N d σ − det γ + N C3 ,
3
(7.5)
where we have set the fermions equal to zero, γαβ = ∂α XM ∂β XN gMN and C3 is the pull-
back of the M-theory three-form potential which has non-zero components only in S 4 . The
18 Tensor self-duality and supersymmetry can be restored at the level of the field equations [85].
worldvolume field equations are

√ 1√
∂α −γ γ αβ ∂β XM gMQ − −γ γ αβ ∂α XM ∂β XN ∂Q gMN
2
+ αβγ ∂α XM ∂β XN ∂β XP HQMNP = 0, (7.6)
where H4 = dC3 . In order to describe the solution, which is similar to the string solution
of [34], we use the global coordinates in AdS7 :

ds 2 = − cosh2 ρ dt 2 + dρ 2 + sinh2 ρ dθ 2 + sin2 θ dφ 2 + sin2 φ dΩ32 . (7.7)
The M2-brane worldvolume coordinates are (τ, σ, ϕ) and our rotating membrane solution
is given by
t = τ, ρ = ρ(σ ), θ = θ (ϕ), φ = ωτ, fixed point in S 3 , (7.8)

where the membrane has the topology of a cylinder −1 < σ < 1, 0 ϕ < 2π , which has
been flattened such that the portion with 0 < ϕ < π is folded on top of the portion with
π < ϕ < 2π . The induced metric becomes
2 2
ds 2 = − cosh2 ρ − ω2 sinh2 ρ sin2 θ dθ 2 + ρ dσ 2 + sinh2 ρ θ dϕ 2 , (7.9)
where ρ ≡ dρ/dσ and θ ≡ dθ/dϕ. It is straightforward to verify the non-trivial
components of the field equations (7.6) which are the t, ρ, θ, φ component. The energy
and spin of the configuration (7.8) are given by
ρ0 θ2
cosh2 ρ sinh ρ
E = 4N dρ dθ , (7.10)
cosh2 ρ − ω2 sinh2 ρ sin2 θ
0 θ1
ρ0 θ2
sinh3 ρ sin2 θ
s = 4Nω dρ dθ , (7.11)
cosh2 ρ − ω2 sinh2 ρ sin2 θ
0 θ1
where
θ1 = θ (0) = θ (2π), θ2 = θ (π ), ρ0 = ρ(±1). (7.12)

The solution which minimizes the energy for a fixed spin and fixed width ! = θ2 − θ1 (we
shall minimize the energy with respect to the width below) is obtained by centering θ (ϕ)
around θ = 0 and maximizing the extension in the ρ-direction by taking
coth ρ0 = ω. (7.13)
If we assume that ! is small then
!N
E= 2
2
2 F1 2, 1; 3/2; 1/ω , (7.14)
ω
2!N
s= 2
2 F1 2, 2; 5/2; 1/ω . (7.15)
3ω3
For ω 1 this describes short membranes with length ρ0 ∼ 1/ω and energy and spins
given by
E 3 = 8!Ns 2 , E, s !N. (7.16)
In flat eleven-dimensional spacetime an analogous relation holds between mass and spin
for all values of the spin (in flat space this relation follows from dimensional analysis).
Thus, in flat spacetime the mass is minimized for given spin by sending ! → 0 and ω → 0
(keeping s fixed). The flat space spectrum therefore contains massless states arbitrary spin,
which can be thought of as infinitely long, thin string-like membranes which are virtually
at rest.
In fact, long ago bosonic open membrane (a disk) rotating simultaneously about two
axis was considered in [32] where the relation a relation like (7.16) was derived. Such
solutions are possible for D 5. Later, this solution was generalized in [33] for the D = 11
supermembrane [86], by gluing two copies of the open membrane of [32] along their edges
to obtain a ‘pancake’ membrane. The zero-point energy of this membrane was studied by
these authors and later in [87]. It was conjectured in [33] that the (semi-classical) energy-
angular momentum relation of the kind (7.16) would be modified by an integral or half
integral number due to the fact that the fermionic coordinates of the supermembrane also
carry intrinsic angular momentum. See [88] for a review of this fascinating subject.
Going back to AdS7 × S 4 , for slow rotation, ω ∼ 1, ω > 1 and finite width !, the solution
(7.8) describes long membranes whose energy and spin now obeys
3π 2/3
E−s = (!N)2/3 s 1/3 , E, s !N. (7.17)
21/3
For ω → 1 the energy and spin diverges and the rotating membrane develops a boundary
given by a folded closed string of length ! which trace out a Wilson surface in the stack of
five-branes. Thus, the long membranes of width ! with finite energy describe operators in
the AN−1 (2, 0) theory which arise in the operator product expansion of the Wilson surface.
The shape of the Wilson surface together with (2.21) suggest that its expansion contains
bilinear higher spin operators which have asymptotically small anomalous dimensions,
(E − s)/s 1 for high spin, s !N 1. In the limit s → ∞ their interactions should be
equivalent to those described by the singletons.
Suppose there is no boundary condition which fixes ! to a finite value. The prescription
is then to vary ! keeping s fixed as to minimize E. The minimal energy configuration for
given spin s is obtained by taking ! → 0, ω → 1 which results in an infinitely long string-
like membrane with energy E = s (the ratio E/s is larger for short wide membranes than
for long thin ones). Note that this geometry is assumed for any value of√s, unlike in the case
of the type IIB closed string which became infinitely long only as s/ λ → ∞. As ! → 0
the dual Wilson surface collapses and the higher derivative corrections to the AN−1 (2, 0)
theory becomes suppressed, resulting in a flow down to the free tensor theory describing
the unbroken phase with hs(8∗ |4) gauge symmetry.
Let us examine the supersymmetry of this solution. The condition for worldvolume
supersymmetry is [75]
1 1 αβγ
Γ = , Γ =√ ∂α XM ∂β XN ∂γ XN , (7.18)
− det γ 3!
and that is the Killing spinor of the AdS7 × S 4 background. An important property of
these Killing spinors is that as we approach the boundary of AdS7 , i.e., as ρ → ∞, they
become an eigenstate of a constant Γ -matrix as follows [75]
Γ = , Γ = Γ012345, (7.19)
where Γa are flat Dirac matrices and a = 0, . . . , 5 are the indices tangent to the boundary
of AdS7 . We have relabeled the coordinates of AdS7 as

t, φ, θ, ψ , θ , φ , ρ → (x0 , x1 , . . . , x5 , ρ), (7.20)
where (ψ , θ , φ ) are the S 3 angles. Now, inserting the solution (7.8) into the definition of
Γ in (7.18) gives
(cΓ0 + (ωs) sin θ Γ1 )Γ62
Γ = , (7.21)
c2 − ω2 s 2 sin2 θ
where c = cosh ρ and s = sinh ρ. Next, we find that [Γ, Γ ] = 0. Therefore the
worldvolume supersymmetries can be written as
= (1 + Γ )(1 + Γ )η, (7.22)

for arbitrary η. We conclude that in the limit ! → 0, ω → 1, ρ → ρ0 → ∞ keeping
s fixed, the solution (7.8) preserves the 8 supersymmetries described by (7.22). The
remaining 8 supersymmetries are broken, which means that the limiting solutions belong to
semi-short multiplets which we identify as the massless HS multiplets arising in the tensor
product of two tensor singletons listed in Table 3. Hence the corresponding anomaly free
operators in the dual SCFT must fall into the same semi-short multiplet. This suggests
that the dual operators are the Konishi-like superfields (3.33) and (3.35) containing the
conserved HS currents described Section 3.
In the above limit the energy and the spin of the membrane accumulate at its ends
which in turn move along light cones at the boundary of AdS7 . The condition (7.22) has a
natural interpretation as the supersymmetry condition for an intersection between the five-
brane and the boundary of an open membrane. This suggests that the relevant part of the
membrane dynamics are the fluctuations in this asymptotic region. It will be interesting
to study more carefully the fluctuation spectrum about these worldvolume singletons, and
in particular to examine whether they exhibit features such as fixed critical tension and
discrete spectrum.
From the fact that the membrane interactions are concentrated at the boundary of the
AdS spacetime we conclude that the hs(8∗ |4) gauge theory is a high energy limit of
M-theory on AdS7 × S 4 .
We remark that the rotating membrane limit is a Lorentzian analog of the pp-wave limit
on AdS7 × S 4 [30] which can be thought of as a collapsed membrane rotating around the
equator of S 4 . A difference that might be important is that the collapsed membrane has
spherical topology while the rotating membrane has cylindrical topology.
One important test of the free CFT 6 /7D HS gauge theory correspondence is the
matching of the holographic Weyl anomaly [73]. It was shown in [89] that the 6d trace-
anomaly of (free) tensor singletons does not match the holographic anomaly computed
in gauged supergravity in D = 7 [73]. To be more precise, the relative strength of the

Euler invariant and the remaining invariant differs in the two cases by a factor of 4/7. It
was argued in [89], however, that the trace anomaly of any CFT 6 picks up contributions
from four-point stress-energy correlators. This makes the 6d trace anomaly sensitive to the
actual interactions of the bulk theory, unlike in, e.g., d = 4, where the trace anomaly can
be computed at weak coupling. Thus the 6d free field trace anomaly should be compared
with the holographic anomaly of the corresponding 7D massless hs(8∗ |4) gauge theory
(which does not admit any consistent truncation to gravity). Note that the corrections from
massless HS exchange in the bulk are higher order in derivatives but of the same order in
1/N . The matching of the overall strength is a consequence of the normalization (4.14),
though the interactions should correct the factor of 4/7 [89].
Finally we comment on the breaking of HS gauge symmetries in AdS7 . The situation
is much less clear here than in D = 4, 5, basically due to the fact that we do not know
how to deform the boundary theory. As was discussed in Section 3.3, we do not have
any candidates for the Higgs fields, which may have to do with the fact that we are
writing local expressions whereas a more drastic, perhaps non-local construction, is what
is actually required to break the HS symmetries in D = 7. Another important difference
between the massless spectra in D = 7 and in D = 4, 5 is the fact that the latter saturate the
unitarity bound for UIRs belonging to certain continuous series of the corresponding AdS
supergroups, while the former belongs to an isolated series (see Section 3). In fact, this can
be used to show that there can be no continuous (marginal) deformations taking the free
SCFT to the strongly coupled fixed point [90]. Another curious fact is that the massless HS
theory described in Section 2.3 does not make use of the ‘massless’ states which saturate
the unitarity bound for UIRs belonging to the continuous series A (see Appendix B). These
operators are described by cubic tensor singletons, and it will be interesting to attempt
to incorporate these into the master field formulation for the 7D HS theory described in
Section 2.3.
8. Summary and discussion
We have proposed that type IIB string theory with N units of D3-brane charge and
M-theory with N units of M2-brane or M5-brane charge have unbroken phases described
by HS gauge theories which admit consistent truncations to massless HS gauge theories
in D = 4, 5, 7 with holographic duals given by SU(N) valued scalar singleton theories in
d = 3, 4, 6 with 16 supersymmetries. The corresponding HS algebras are
hs(8|4) ⊃ OSp(8|4), (8.1)

hs(2, 2|4) ⊃ PSU(2, 2|4), (8.2)
∗ ∗
hs(8 |4) ⊃ OSp(8 |4), (8.3)
which are described in Section 2 together with the corresponding massless HS gauge
theories. These theories also contain massive fields, some of which are Higgs fields that
can be eaten by the massless fields. Both massless and massive fields also have KK towers
which can be used to re-construct the spectrum of the type IIB string and M-theory in
appropriate limits as discussed in more detail in Section 5.
In the case of type IIB string on AdS5 × S 5 , we have conjectured that the hs(2, 2|4)
gauge theory arises in a critical limit of the type IIB theory in which
gstr ∼ 1/N, lstr ∼ R fixed R, N 1, (8.4)
and that this limit corresponds to the free 4D, N = 4, SU(N) SYM in which gYM = 0. This
means that the relations between the closed string parameters in AdS5 × S 5 and the gauge
theory parameters for λ 1, which are read off from the D3-brane solution obtained in the
supergravity approximation, are renormalized, as discussed in Section 5, and summarized
in (1.1) and (1.2).
In the case of M-theory on AdS4/7 × S 7/4 , we have conjectured the holographic
boundary theories to be a SU(N) valued OSp(8|4) singleton field theory which resides
at a UV fixed point in 3d, and a free SU(N) valued (2, 0) tensor singleton field theory
residing at a IR fixed point in 6d.
The spectrum of massless states in all the HS gauge theories discussed here have the
universal property that they all arise in the symmetric product of two singletons. This
motivates a worldsheet sigma model description of these theories based on an affine
extension of AdS superalgebras in D = 4, 5, 7 with critical KM levels leading to left-
moving and right-moving singleton Verma modules with a maximal number of null-states.
In this respect, the existence of a singleton-like representations of affine SO(3, 2) with level
kcrit = 5/2 found in [70] is encouraging.
The idea of obtaining the massless states of a D = 4, N = 8 HS theory starting from the
free OSp(8|4) singleton theory, which in turn was obtained from the eleven-dimensional
supermembrane on AdS4 × S 7 , already appeared long ago [40]. We recall that all the
massless fields in this theory, with the exception of a pseudoscalar, satisfy the energy-spin
relation E0 = s + 1. More recently, long rotating strings that extend to the boundary of
AdS5 and couple to operators which are asymptotically anomaly free, i.e., (E − s)/s → 0
as E, s → ∞, have been studied [34].
Motivated by above the considerations, we have found rotating long membrane
solutions (7.8) to the equations which describe the M2-brane in AdS7 × S 4 background.
These membranes have width ! and the geometry of infinitely stretched strings with energy
and spin density concentrated at the end points. They satisfy the semi-classical energy-spin
relation E = s. A feature not present in the string case is that the energy is minimized for
fixed spin by sending the angular velocity ω → 1 and the width ! → 0 keeping s fixed,
resulting in infinitely long membranes with string-like geometry and semi-classical energy
E = s. In Section 7, we have interpreted these as the lowest weight states of the massless
supermultiplets of the 7D HS gauge theory discussed in Section 2 (see Table 3). Further
aspects of this picture, especially the quantization issue, remain to be studied.
It would also be interesting to study the spherical membrane in AdS4 and examine
whether it admits ‘breathing’ and ‘rotation’ modes similar to those of strings in AdS3 with
NS-fluxes [78].
As there is effectively no separation in AdS energy between the massless HS fields
and the massive HS fields, we have proposed that the massless HS theories (based on HS
extension of the 32 supercharge AdSd+2 superalgebras in d = 3, 4, 6) arise as a result of
consistent truncation of the full HS theories. This proposal can be tested explicitly since
for large N , the singleton theory and the HS gauge theory can be compared order by
order in the 1/N expansion: consistent truncation implies that the massless HS theory
action is the generating functional of correlators of bilinear operators. Indeed, a correlation
function of four bilinear operators in a singleton theory can be written in a manifestly
s–t–u symmetric form in terms of two- and three-point functions involving only bilinear
operators, as discussed in Section 4.
We have also examined mechanisms for spontaneous breaking of HS gauge symmetry
down to the symmetries underlying ordinary supergravity. In D = 4, 5 the ‘order
parameter’ for breaking of HS gauge symmetry is the holographic Yang–Mills coupling.
In 4D this is a marginal deformation which corresponds to a finite dilaton VEV in the
bulk. The broken theory has an AdS vacuum in which the broken gauge fields have non-
critical masses m2 − m2crit ∼ NgYM 2 /R 2 . Using the non-intersection principle we argue
these cross over into the leading Regge trajectory as NgYM 2 . We have also identified the
Higgs multiplets at arbitrary level in the HS spectrum, and the realization of the level-
one Higgs multiplet in terms of composite operators (i.e., anomaly multiplets) in the free
singleton SCFT.
Also in 3d, where the Yang–Mills coupling is a relevant perturbation, we have identified
the Higgs multiplets at arbitrary level in the HS spectrum, and the realization of the level-
one Higgs multiplet in terms of composite operators (i.e., anomaly multiplets) in the free
OSp(8|4) singleton field theory.
In D = 7 we do not know what is the order parameter for breaking HS gauge symmetry,
nor have we identified the Higgs multiplets. This is presumably related to the fact that the
massless gauge fields in D = 7 belong to the discrete B series (see (B.16) in Appendix B).
We believe this issue should have a simple resolution in a framework where the nature of
the mysterious interactions on the five-brane is well understood. We stress that the Higgsing
of the 7D HS gauge theory is dual to weak irrelevant perturbations of the tensor theory in
the IR, which should be describable using a field theoretic, perhaps non-local, construction
in 6d. One may also speculate that the continuous A series (see (B.15)) could play a role
in this, since the corresponding fields can be Higgsed, which signals the existence of the
corresponding anomaly multiplets. This, in turn, would provide valuable data on the details
of the interactions in 6d.
An interesting open problem is to use the HS gauging techniques described in Sections 2
and 6 to construct interactions between massless HS fields and Higgs fields. Clearly, the
issue of consistent truncation becomes moot once we include (massive) Higgs fields. It is
therefore a challenge to examine whether some generalized truncation scheme, perhaps of
the type described in [27], may temper the fluctuations in the massive sector.
In testing various aspects of the AdS/HS gauge theory correspondences discussed in this
paper, it will be very useful to develop a deeper understanding of the geometrical nature
of HS interactions, possibly formulating them in a generalized superembedding approach.
This would provide a universal tool for studying the HS dynamics [91] which would not
only simplify the task of coupling Higgs master fields to HS gauge theories but also yield
a superfield formulation [91] that would simplify the treatment of the bulk interaction and
the computations of the attendant Witten diagrams. On the boundary side, the existing
literature on the OPE computations involving free fields should be extended to cases where
subleading in 1/N contributions will arise [22]. We have described few examples of such
correlators in Section 4.
In this paper we have focused our attention on HS gauge theories in D = 4, 5, 7. No
doubt these results can be extended to AdS6 as well. In D = 3 the HS gauge fields do
not propagate physical degrees of freedom. Nonetheless, physical matter fields of spin
s = 0, 1/2 can be coupled to massless HS gauge theory [92,93]. The advantage here is
that an action principle is known and the mathematics is much simpler than in higher
dimensions. It would be interesting to study this model in the context of massless higher
spins and holography.
At the algebraic level there is in principle no bound on the number of supersymmetries
in HS gauge theories and we expect consistent massless interactions for any N in D 7,
though certain restrictions follow from the requirement of an R symmetry neutral vierbein
[91]. As discussed in Section 4, the restrictions on the spacetime superdimension are
instead expected to be related to the consistency of the full HS quantum theory, including
both massless and massive states, which requires the full generating functional (4.9) of the
free singleton SCFT with finite sources for composite operators. Effectively, the condition
that this quantity exists is expected to be as restrictive in the free singleton SCFT as in the
(strongly) interacting singleton SCFT. This may lead to the restriction that the holographic
dual cannot have more than 16 supersymmetries in d 6. Similar restrictions should
follow from the quantum consistency of the yet to be constructed dual bulk sigma models.
In Section 4, similar effects were argued to arise in the holographic theory due to insertions
of sewing operators in the free singleton field theory required for unitarity.
Another particularly interesting class of singleton CFTs, which we have not considered
here, are the free 4d conformal HS theories constructed in [11]. Here the singleton field
is a master field comprising an infinite set of ordinary singletons which together form an
irreducible representation of a HS extension of the d-dimensional conformal group. In that
case the relevant HS symmetry algebra is an infinite-dimensional extension of Sp(8, R)
which contains the AdS group in 5D.
To conclude, we believe that the remarkable algebraic and geometric structures
underlying HS gauge symmetry are natural extensions of supergravity and will be
important guides towards the true foundations of string and M-theory. In particular, the
simplicity of their holographic duals together with the fact that the bulk physics can still
be phrased in a relatively simple language is both gratifying and compelling. Clearly much
remains to be done in this subject which may be viewed as being still in its infancy.
Acknowledgements
We thank D. Berman, U. Danielsson, S. Frolov, I. Klebanov, F. Kristiansen, J. Minahan,

P. Rajan, D. Roest, K. Skenderis, M. Staudacher, I.Y. Park, A. Tseytlin and K. Zarembo
for stimulating discussions. In particular, we would like to thank L. Rastelli for valuable
discussions on the issue of consistent truncation discussed in Section 4 and J. Fjelstad for
discussions and ideas on critical singleton closed string theory. This research project has
been supported in part by NSF Grant PHY-0070964.
Appendix A. Spectra of massless higher spin gauge theories in D = 5, 7
In this appendix, we tabulate the spectra of singletons and the generators of the super-
HS groups and the field content of the master scalar fields in AdS5 and AdS7 . The case of
Table 4
The hs(2, 2|4) generators with Y = 0, ±1 arranged into levels labeled by ! = 14 (ny + nȳ + nθ + nθ̄ − 2). The
entries are SU(4) × U (1)Y representations as follows: 15 = 150 , 4 = 41 , 1 = 10 , 16 = 150 + 10 , 24 = 201 + 41
and 36 = 200 + 150 + 10 , where the U (1)Y charge is defined by Y = ny − nȳ . The SO(4, 1) content is given by
the highest weights m1 m2 12 |Y | where m1 = 12 (ny + nȳ ). Upon gauging, these generators give rise to spin
s = m1 + 1 gauge fields which can be used to write a canonical set of covariant curvature constraints. As a result
the gauge fields for m2 12 |Y | + 1, s 2 are auxiliary while those for m2 = 12 |Y | contain physical degrees of
freedom
!\s 1 3 2 5 3 7 4 9 5 11 6 ···
2 2 2 2 2
0 15 4 1
1 16 24 36 24 16 4 1
2 1 4 16 24 36 24 16 4 1
3 1 4 16 24 36 ...
4 1 ...
..
.
Table 5
The hs(2, 2|4) generators with Y = ±2, ±3, ±4. The entries are SU(4) × U (1)Y representations as follows:
16 = 102 + 62 , 4 = 43 , 6 = 62 and 1 = 14 . Further notation is defined in Table 4. These generators are associated
with gauge fields dual to generalized antisymmetric tensor fields contained in the scalar master field Φ; see
Table 7 for s 1
!\s 2 5 3 7 4 9 5 11 6 ···
2 2 2 2
1 16 4 6
2 6 4 16 + 1 4 6
3 6 4 16 + 1 ...
..
.
Table 6
The d = 4, N = 4 singletons. The quantity Z is the SU(2, 2|4) central charge carried by the supermultiplet. The
entries in the table denote SU(4) representations. Each entry carries an SO(4) ⊂ SO(6) ⊂ SO(6, 2) representation
(jL , 0), and their complex conjugates (0, jR ). The states for each value of |Z| form a single massless irrep of
d = 4, N = 4 Poincaré superalgebra, and the states carry spin s = jL . For all the states E0 = s + 1, where E0 is
the lowest AdS5 energy. There exists an outer automorphism group U (1)Y of SU(2, 2|4), and the U (1)Y charges
of 6, 4 and 1 are 0, ±1 and ±2, respectively. The Z = 0 multiplet is the d = 4, N = 4 SYM singleton multiplet
which has 8 + 8 degrees of freedom. All the other singleton multiplets have 16 + 16 degrees of freedom. For
superfield realization of all the singletons listed in this table, see Section 3.2
|Z|\s 0 1 1 3 2 5 3 ···
2 2 2
0 6 4 1
1 4 6+1 4 1
2
1 1 4 6 4 1
3 1 4 6 4 1
2
2 1 4 6 4 1
.. ..
. .
Table 7
The physical fields contained in the master scalar field Φ arising in the hs(2, 2|4) gauge theory in D = 5.
The entries are the following SU(4) × U (1)Y representations for s < 1: 42 = 200 + 102 + 10−2 + 14 + 1̄−4 ,
48 = 201 + 20−1 + 43 + 4̄−3 , 8 = 41 + 4̄−1 and 10 ; for s 1: 62 , 43 , 16 = 102 + 62 and 14 . The spin s 1
sector is realized in the field theory in terms of two-form potentials and their higher spin generalizations. These
fields obey self-duality in D = 5 and have dual one-form gauge fields corresponding to the generators given
in Table 5, with the exception of the underlined representations, which have no one-form duals. Here the form
degree refers to the number of curved indices as opposed to the tangential multi-spinor indices arising from the
(y, ȳ)-expansion
!\s 0 1 1 3 2 5 3 7 4 9 5 11 6 ···
2 2 2 2 2 2
0 42 48 6
1 1 8 6 4 16 + 1 4 6
2 6 4 16 + 1 4 6
3 6 4 16 + 1 ...
..
.
Table 8
The hs(8∗ |4) generators with Y = 0, ±1 arranged into levels labeled by ! = 14 (ny + nȳ + nθ + nθ̄ − 2).
The entries are SO(5) × U (1)Y representations as follows: 10 = 100 , 4 = 41 , 1 = 10 , 20 = 161 + 41 and
20 = 140 + 50 + 10 , where the U (1)Y charge is defined by Y = ny − nȳ . The SO(6, 1) content is labeled
by highest weights m1 m2 m3 = 12 |Y |, where m1 = 12 (ny + nȳ ). Upon gauging, these generators give rise
to spin s = m1 + 1 gauge fields which can be used to write a canonical set of covariant curvature constraints.
As a result the gauge fields for m2 12 |Y | + 1, s 2 are auxiliary while those for m2 = 12 |Y | contain physical
degrees of freedom
!\s 1 3 2 5 3 7 4 9 5 11 6 ···
2 2 2 2 2
0 10 4 1
1 10 20 20 20 10 4 1
2 1 4 10 20 20 20 10 4 1
3 1 4 10 20 20 ...
4 1 ...
..
.
Table 9
The hs(8∗ |4) generators with Y = ±2, ±3, ±4. The entries are SO(6) × U (1)Y representations as follows:
15 = 52 + 102 , 4 = 43 , 6 = 52 + 12 and 16 = 102 + 52 + 14 . These generators are associated with gauge fields
dual to generalized antisymmetric three-form tensor fields contained in the scalar master field Φ; see Table 11 for
s 1. Further notation is defined in Table 8
!\s 2 5 3 7 4 9 5 11 6 ···
2 2 2 2
1 15 4 6
2 6 4 16 4 6
3 6 4 16 ...
..
.
Table 10
The d = 6, N = (2, 0) singletons. The quantity Z denotes the SU(2)Z spin defined in Section 2.3. The entries
denote USp(4)Y SO(5) × U (1)Y representations, which are irreducible except 6 = 5 + 1. The U (1)Y charges
of 1, 4, 5 are 0, ±1 and ±2, respectively. The SO(6) highest weights (n1 , n2 , n3 ) associated with each entry are
given by n1 = n = 2 = n3 = s, and the AdS7 energy by E0 = s + 2. The level ! = 0 (Z = 0) multiplet is the
d = 6, N = (2, 0) tensor singleton; see Section 3.3 for superfield realization of all the singletons shown in the
table, and composites formed out of the tensor singleton
|Z|\s 0 1 1 3 2 5 3 ···
2 2 2
0 5 4 1
1 4 6 4 1
2
1 1 4 6 4 1
3 1 4 6 4 1
2
2 1 4 6 4 1
. .
.. ..
Table 11
The physical fields expected to arise in the master scalar field Φ in the hs(8∗ |4) gauge theory in D = 7. The
entries are SO(6) × U (1)Y representations, where 6 = 5 + 1 and 15 = 10 + 5. The spin s 1 sector is expected
to be realized in Φ in terms of three-form potentials and their higher spin generalizations. These fields obey
self-duality in D = 7 and have dual one-form gauge fields corresponding to the generators given in Table 9,
with the exception of the underlined representations, which have no one-form duals. Here the form degree
refers to the number of curved indices as opposed to the tangential multi-spinor indices arising from the (y, ȳ)-
expansion
!\s 0 1 1 3 2 5 3 7 4 9 5 11 6 ···
2 2 2 2 2 2
0 140 161 52
1 10 41 62 43 152 + 14 43 62
2 62 43 152 + 14 43 62
3 62 43 152 + 14 ...
..
.
D = 4 is relatively simpler and has been presented in Section 2.1. The spectrum of physical
states described by the master gauge fields in D = 4, 5, 7 are also given in Section 2.
Appendix B. UIRs of AdS superalgebras in D = 4, 5, 7
In this appendix, we define notation for the irreps of AdS superalgebras in D = 4, 5, 7,

and list these irreps as well as the BPS short supermultiplets. These results are especially
useful for the discussions of Sections 2 and 3.
B.1. The UIRs of OSp(8|4) and BPS multiplets
Recall that a UIR of OSp(8|4) is a multiplet of SO(3, 2) × SO(8) UIRs denoted by
D(E0 , s; a1 , a2 , a3 , a4 ), (B.1)
where E0 is the minimum eigenvalue of the AdS energy generator M05 , s denotes SO(3) ⊂
SO(3, 2) spin and (a1 , a2 , a3 , a4 ) are the Dynkin labels of the SO(8) irrep carried by the
lowest energy state. There exist two series of supermultiplets [48]:
(A) E0 1 + s + a1 + a2 + 12 (a3 + a4 ), (B.2)

(B) E0 = a 1 + a 2 + 1
2 (a3 + a4 ), s = 0. (B.3)
These are the irreps carried by the lowest components of the supermultiplets, and the entire
OSp(8|4) supermultiplets are obtained by acting with supercharges.
The lowest components of the massless supermultiplets shown in Table 1 saturate the
unitarity bound of series A as E0 = s + 1, except in level ! = 0 supergravity multiplet,
in which case D(1, 0; 0, 0, 2, 0) belongs to series B. The discrete series B contains
the BPS multiplets. In particular, the singleton multiplet is characterized by the irrep
D(1/2, 0; 0, 0, 0, 1) carried by its lowest component and it belongs to series B. It can
be described by a suitably constrained superfield. Taking a suitably symmetrized and
constrained product of 2E0 singleton superfields one can construct BPS superfields whose
lowest components carry the following irreps [50]
BPS-1/2: D(p/2, 0; 0, 0, p, 0), (B.4)

BPS-1/4: D (p + 2q)/2, 0; 0, q, p, 0 , (B.5)

BPS-1/8: D (p + 2q + 3r + 4s)/2, 0; r + 2!, q, p, r . (B.6)
All of these belong to series B. In particular, the lowest components of the KK towers of
the level ! = 0 supergravity multiplet carry the irrep D(k/2, 0; 0, 0, k, 0) for k = 3, 4, . . .
[53].
B.2. The UIRs of PSU(2, 2|4) and BPS multiplets
A UIR of SU(2, 2|4) consists of UIRs of SO(4, 2) × SO(6) denoted by
D(E0 , jL , jR ; a1 , a2 , a3 )Y , (B.7)
where E0 is the eigenvalue of the AdS energy generator M06 and (jL , jR ) label the
SO(4) ⊂ SO(4, 2) irrep, (a1 , a2 , a3 ) denote the Dynkin labels of the SO(6) SU(4)
R-symmetry irrep carried by the minimum energy states and Y denotes the outer U (1)Y
automorphism charge, which will often be suppressed when it is vanishing. There exist
three series of supermultiplets [42]:
(A) E0 2 + JL + JR + a1 + a2 + a3 , JL − JR 12 (a3 − a1 ); (B.8)

(B) E0 = 1
2 (a1 + 2a2 + 3a3) 2 + 2JL + 1
2 (3a1 + 2a2 + a3 ), JR = 0 (B.9)
which implies
E0 1 + JL + a1 + a2 + a3 , 1 + JL 12 (a3 − a1 );
(C) E0 = 2a1 + a2 , a3 = a1 , JL = JR = 0. (B.10)
In the case of series B, irreps with (JL ↔ JR , a1 ↔ a3 ) must also be included. The irreps
listed above are carried by the lowest components of the supermultiplets, and the entire
PSU(2, 2|4) supermultiplets are obtained by acting with supercharges.
The lowest components of the massless supermultiplets shown in Table 2 saturate the
unitarity bound of series A as E0 = s + 2, with JL = JR = s/2, except in level ! = 0
supergravity multiplet in which case D(2, 0, 0; 0, 2, 0) belongs to series C. The discrete
series C contains the BPS multiplets. In particular the Maxwell singleton multiplet is
characterized by D(1, 0, 0; 0, 1, 0) carried by its lowest component and it belongs to
series C. It can be described by a suitably constrained superfield. Taking a properly
symmetrized and constrained product of E0 singletons superfields one can construct BPS
superfields whose lowest components carry the following irreps [50]
BPS-1/2: D(p, 0, 0; 0, p, 0), (B.11)
BPS-1/4: D(p + 2q, 0, 0; q, p, q), (B.12)
BPS-1/8: D(p + 2q + 3r, 0, 0; q, p, q + 2r). (B.13)
The BPS-1/2 and BPS-1/4 multiplets belong to series C, and the BPS-1/8 multiplets
belong to series B. The KK towers of the level ! = 0 supergravity are the BPS-1/2
multiplets given by D(k, 0, 0; 0, k, 0) with k = 3, 4, . . . [3,58,59].
There exists an extensive literature on the OPEs of various BPS-1/2 operators. The
UIRs which can appear in these OPEs belong to series A with JL = JR = s/2, and series C
[51].
B.3. The UIRs of OSp(8∗ |4) and BPS multiplets
A UIR of OSp(8∗ |4) consists of UIRs of SO(6, 2) × USp(4) denoted by

D(E0 , J1 , J2 , J3 ; a1 , a2 )Y , (B.14)
where E0 is the eigenvalue of the AdS energy generator M08 , (J1 , J2 , J3 ) denote the
Dynkin labels of the SU(4) SO(6) ⊂ SO(6, 2) irrep, (a1 , a2 ) denote the Dynkin labels
of the USp(4) irrep carried by the minimum energy states and Y denotes the outer U (1)Y
automorphism charge, which will often be suppressed when it is vanishing. There exist
four series of supermultiplets [48]:
(A) E0 6 + 12 (J1 + 2J2 + 3J3 ) + 2(a1 + a2 ), (B.15)
(B) E0 = 4 + 12 (J1 + 2J2 ) + 2(a1 + a2 ), J3 = 0, (B.16)
(C) E =2+ 1
2 J1 + 2(a1 + a2 ), J3 = J2 = 0, (B.17)
(D) E0 = 2(a1 + a2 ), J3 = J2 = J1 = 0. (B.18)
These are the irreps carried by the lowest components of the supermultiplets, and the entire
OSp(8∗ |4) supermultiplets are obtained by acting with supercharges.
The lowest components of the massless supermultiplets shown in Table 3 have E0 =

s + 4 and belong to series B, while the level ! = 0 supergravity multiplet carries the
irrep. D(4, 0, 0, 0; 0, 2) which belongs to series D. The discrete series D contains the BPS
multiplets. The singletons are contained in series C and D. The superfields in terms of
which they are realized, and the UIRs carried by their lowest components are as follows
[46,48,50]:
(D) W ij D(2, 0, 0, 0; 0, 1),

(D) Wi D(2, 0, 0, 0; 1, 0),
(C) W D(2,
! 0, 0, 0; 0, 0),
(C) ωα1 ···α!−2 D 2 + 1, ! − 2, 0, 0; 0, 0 . (B.19)
The index i = 1, . . . , 4, labels the 4-plet of USp(4), the index α = 1, . . . , 4 labels the chiral
spinor of SO(6), W ij = −W j i and symplectic traceless, Ω ij Wij = 0, and ωα1 ···α!−2 is
totally symmetric in its indices (see Table 10 for further details). The superspace constraints
imposed on these superfields can be found in [50]. The superfield Wij represents the
well known (2, 0) tensor singleton and it is singlet under an SU(2)Z group defined in
Section 2.3. The singleton superfields (W i , W, ωα1 ···α!−2 ), on the other hand, carry SU(2)Z
spins (1/2, 1, !/2), respectively. These are the level ! = 1, 2 and ! 3 singletons shown
in Table 10.
Taking suitably symmetrized and constrained products of singleton superfields, one can
construct BPS superfields whose lowest components carry the following UIRs [50]:
BPS-1/2: D(2p, 0, 0, 0; 0, p), (B.20)

BPS-1/4: D(2p + 4q, 0, 0, 0; 2q, p). (B.21)
Both of these belong to series D. The KK towers of the level ! = 0 supergravity are the
BPS-1/2 multiplets given by D(2k, 0, 0, 0; 0, k) with k = 3, 4, . . . [50].
The OPEs of BPS-1/2 operators have been studied [51,90]. The supermultiplets that
can appear in the OPE of two BPS-1/2 operators belong to series A with (J1 , J2 , J3 ) =
(0, s, 0); series B with (J1 , J2 ) = (0, s) and E0 = 4 + s + 2(a1 + a2 ); series C with J1 = 0
and E0 = 2 + 2(a1 + a2 ), and series D [51].
Appendix C. Labeling of USp(8), SU(4), USp(4) and SO(8) irreps
C.1. USp(8)
The highest weight state (HWS) labels (n1 , n2 , n3 , n4 ) of USp(8) satisfy n1 n2

n3 n4 and are related to the Dynkin labels [a1, a2 , a3 , a4 ] as follows:
n1 = a1 + a2 + a3 + a4 , n2 = a2 + a3 + a4 ,
n3 = a3 + a4 , n4 = a4 .
C.2. SU(4) ∼ SO(6)
The HWS labels of SU(4) irreps are (n1 , n2 , n3 ). They satisfy n1 n2 n3 and they
are related to the Dynkin labels [a1 , a2 , a3 ] as follows:
n1 = a1 + a2 + a3 , n2 = a2 + a3 , n3 = a3 .
The SO(6) HW labels by (m1 , m2 , m3 ) obey m1 m2 |m3 | and they are related to the
SO(6) Dynkin labels [b1 , b2 , b3 ] as
m1 = b1 + 12 (b2 + b3 ), m2 = 12 (b2 + b3 ), m3 = 12 (−b2 + b3 ).

These are related to the SU(4) HW labels (n1 , n2 , n3 ) and SU(4) Dynkin labels [a1 , a2 , a3 ]
as
m1 = 12 (n1 + n2 − n3 ), m2 = 12 (n1 − n2 + n3 ), m3 = 12 (−n1 + n2 + n3 ),
b1 = a2 , b2 = a1 , b3 = a3 .
C.3. USp(4) ∼ SO(5)
The USp(4) irreps have the HWS labels (n1 , n2 ) which satisfy n1 n2 and they are
related to the Dynkin labels [a1 , a2 ] as
n1 = a1 + a2 , n2 = a2 .
The irreps of SO(5) have HW labels (m1 , m2 ) which satisfy m1 m2 0 and they are
related to the SO(5) Dynkin labels [b1, b2 ] as
m1 = b1 + 12 b2 , m2 = 12 b2 .
These are related to the USp(4) HW labels (n1 , n2 ) and USp(4) Dynkin labels [a1 , a2 ] as
m1 = 12 (n1 + n2 ), m2 = 12 (n1 − n2 ),
b1 = a2 , b2 = a1 .
C.4. SO(8)
The irreps of SO(8) have HWS labels (n1 , n2 , n3 , n4 ) which satisfy n1 n2 0 n3

|n4 | and they are related to the SO(8) Dynkin labels [a1 , a2 , a3 , a4 ] as
n1 = a1 + a2 + 12 (a3 + a4 ), n2 = a2 + 12 (a3 + a4 ),
n3 = 1
2 (a3 + a4 ), n4 = 12 (−a3 + a4 ).
Appendix D. Compact and non-compact bases for SO(d, 2)
We write the SO(d, 2) algebra in canonical form as (A = 0, . . . , d, d + 2):
[MAB , MCD ] = iηBC MAD + 3 more, (D.1)

where η = diag(−, +, . . . , +, −). The compact basis, which is suitable for describing
physical AdS fields, consists of the AdS energy E = −M0,d+2 , the SO(d) generators Mij
(i = 1, . . . , d) and the spin-boosts L±
i = Mi,d+2 ∓iM0i , which shift the AdS energy by ±1.
In compact basis the SO(d, 2) weight spaces D(E0 ; m1 , . . . , m[d/2]) are obtained by acting
with L+ i on lowest weight states, which have minimal energy E = E0 and carry SO(d)
highest weights (m1 , . . . , m[d/2]). Note that the label m1 is the SO(3) ⊂ SO(3, 2) spin in
the case of AdS4 and the sum jL + jR of SU(2)L × SU(2)R SO(4) ⊂ SO(4, 2) spins
in the case of AdS5 . The non-compact basis, which is suitable for describing conformal
fields, consists of the dilatation generator D = Md,d+2 , the SO(d − 1, 1) generators
Mµν (µ = 0, 1, . . . , d − 1), and the d-dimensional momentum Pµ = Mµd + Mµ,d+2 and
generator of special conformal transformations Kµ = Mµd − Mµ,d+2 . The compact basis
(E, Mij , L± i ) and non-compact basis (D, Mµν , Kµ , Pµ ) are related [94] by a similarity
transformation executed by the (non-unitary) operator
S = exp iL+
d, (D.2)
with the following properties
1
SDS −1 = −iE + L− , (D.3)
2 d
i
SM0a S −1 = −iMa,d − L− , SMab S −1 = Mab , (D.4)
2 a
i 1
SK0 S −1 = − L− , SKa S −1 = − L− , (D.5)
2 d 2 a
where we have split the indices as follows
a a

i = 1, 2, . . . , d − 1, d, µ = 0, 1, 2, . . ., d − 1 . (D.6)
Hence (d + 1)-dimensional time-evolution and spatial rotation are equivalent to d-dimen-
sional dilatation and Lorentz rotation. Thus
S −1 D(E0 ; m1, . . . , m[d/2]) = O∆ (0)|0!,

O∆ (0)e−ix
µP µP
where |0! is the vacuum of the CFT d and O∆ (x) = eix µ µ is a conformal
tensor with scaling dimension
∆ = E0
and Lorentz spin given by (m1 , . . . , m[d/2]).
References
[1] J.M. Maldacena, The large N limit of superconformal field theories and supergravity, JHEP 9807 (1998)
013, hep-th/9711200.
[2] S.S. Gubser, I.R. Klebanov, A.M. Polyakov, Gauge theory correlators from noncritical strings, Phys. Lett.
B 428 (1998) 105, hep-th/9802109.
[3] E. Witten, Anti-de Sitter space and holography, Adv. Theor. Math. Phys. 2 (1998) 253, hep-th/9802150.
[4] O. Aharony, S.S. Gubser, J. Maldacena, H. Ooguri, Y. Oz, Large N field theories, string theory and gravity,
Phys. Rep. 323 (2000) 183, hep-th/9905111.
[5] E. D’Hoker, D.Z. Freedman, Supersymmetric gauge theories and the AdS/CFT correspondence, TASI
Lectures, hep-th/0201253.
[6] A.M. Polyakov, Gauge fields and spacetime, hep-th/0110196.
[7] D. Anselmi, Theory of higher spin tensor currents and central charges, Nucl. Phys. B 541 (1999) 323, hep-
th/9808004.
[8] S.E. Konstein, M.A. Vasiliev, V.N. Zaikin, Conformal higher spin currents in any dimension and AdS/CFT
correspondence, JHEP 0012 (2000) 018, hep-th/0010239.
[9] E. Sezgin, P. Sundell, Doubletons and 5D higher spin gauge theory, JHEP 0109 (2001) 036, hep-th/0105001.
[10] E. Sezgin, P. Sundell, Towards massless higher spin extension of D = 5, N = 8 gauged supergravity,
JHEP 0109 (2001) 025, hep-th/0107186.
[11] M.A. Vasiliev, Conformal higher spin symmetries of 4D massless supermultiplets and osp(L, 2M) invariant
equations in generalized (super)space, hep-th/0106149.
[12] M.A. Vasiliev, Cubic interactions of bosonic higher spin gauge fields in AdS5 , Nucl. Phys. B 616 (2001)
106, hep-th/0106200.
[13] A. Mikhailov, Notes on higher spin symmetries, hep-th/0201019.
[14] M.A. Vasiliev, Consistent equations for interacting gauge fields of all spins in 3 + 1 dimensions, Phys. Lett.
B 243 (1990) 378.
[15] M.A. Vasiliev, Higher spin gauge theories: star-product and AdS space, hep-th/9910096.
[16] E. Sezgin, P. Sundell, Higher spin N = 8 supergravity, JHEP 9811 (1998) 016, hep-th/9805125.
[17] E. Sezgin, P. Sundell, Higher spin N = 8 supergravity in AdS4 , hep-th/9903020.
[18] E. Sezgin, P. Sundell, On curvature expansion of higher spin gauge theory, Class. Quantum Grav. 18 (2001)
3241, hep-th/0012168.
[19] E. Sezgin, P. Sundell, Analysis of higher spin field equations in four dimensions, JHEP 0207 (2002) 055,
hep-th/0205132.
[20] E. Sezgin, P. Sundell, 7D higher spin gauge theory: bosonic algebra and linearized constraints, hep-
th/0112100.
[21] E. Witten, Talk given at J.H. Schwarz’ 60th Birthday Conference, Cal Tech, 2–3 November 2001.
[22] P. Haggi-Mani, B. Sundborg, Free large N supersymmetric Yang–Mills theory as a string theory, JHEP 0004
(2000) 031, hep-th/0002189.
[23] B. Sundborg, Stringy gravity, interacting tensionless strings and massless higher spins, Nucl. Phys. Proc.
Suppl. 102 (2001) 113, hep-th/0103247.
[24] U. Danielsson, F. Kristiansson, P. Rajan, E. Segin, P. Sundell, in preparation.
[25] S. Lee, S. Minwalla, M. Rangamani, N. Seiberg, Three-point functions of chiral operators in D = 4, N = 4
SYM at large N , Adv. Theor. Math. Phys. 2 (1998) 697, hep-th/9806074.
[26] E. D’Hoker, D.Z. Freedman, S.D. Mathur, A. Matusis, L. Rastelli, Extremal correlators in the AdS/CFT
correspondence, hep-th/9908160.
[27] E. D’Hoker, B. Pioline, Near-extremal correlators and generalized consistent truncation for AdS4|7 × S 7|4 ,
JHEP 0007 (2000) 021, hep-th/0006103.
[28] G.W. Semenoff, K. Zarembo, Wilson loops in SYM theory: from weak to strong coupling, Nucl. Phys. Proc.
Suppl. 108 (2002) 106, hep-th/0202156.
[29] I.R. Klebanov, P. Ouyang, E. Witten, A gravity dual of the chiral anomaly, hep-th/0202056.
JHEP 0204 (2002) 013, hep-th/0202021.
[31] N. Seiberg, Notes on theories with 16 supercharges, Nucl. Phys. Proc. Suppl. 67 (1998) 158, hep-th/9705117.
[32] K. Kikkawa, M. Yamasaki, Can the membrane be a unification model?, Prog. Theor. Phys. 76 (1986) 1379.
[33] L. Mezincescu, R.I. Nepomechie, P. van Nieuwenhuizen, Do supermembranes contain massless particles?,
Nucl. Phys. B 309 (1988) 317.
[34] S.S. Gubser, I.R. Klebanov, A.M. Polyakov, A semi-classical limit of the gauge/string correspondence, hep-
th/0204051.
[35] S.E. Konstein, M.A. Vasiliev, Extended higher-spin superalgebras and their massless representations, Nucl.
Phys. B 331 (1990) 475.
[36] M. Günaydin, Oscillator like unitary representations of non-compact gauge groups and supergroups and
extended supergravity theories, in: E. Inönü, M. Serdaroglu (Eds.), Lecture Notes in Physics, Vol. 180,
1983.
[37] H. Nicolai, E. Sezgin, Singleton representations of OSp(N, 4), Phys. Lett. B 143 (1984) 389.
[38] M. Flato, C. Fronsdal, One massless particle equals two Dirac singletons, Lett. Math. Phys. 2 (1978) 421.
[39] M. Günaydin, N.P. Warner, Unitary supermultiplets of OSp(8/4, R) and the spectrum of the S 7
compactification of eleven-dimensional supergravity, Nucl. Phys. B 272 (1986) 99.
[40] E. Bergshoeff, A. Salam, E. Sezgin, Y. Tanii, Singletons, higher spin massless states and the supermembrane,
Phys. Lett. B 205 (1988) 237.
[41] M. Günaydin, D. Minic, M. Zagermann, Doubleton conformal theories, CPT and IIB string on AdS5 × S 5 ,
Nucl. Phys. B 534 (1998) 96;
M. Günaydin, D. Minic, M. Zagermann, Nucl. Phys. B 538 (1999) 531, Erratum, hep-th/9806042.
[42] V.K. Dobrev, V.B. Petkova, All positive energy unitary irreducible representations of extended conformal
supersymmetry, Phys. Lett. B 162 (1985) 127.
[43] M. Günaydin, N. Marcus, The spectrum of the S 5 compactification of the chiral N = 2, D = 10 supergravity
and the unitary supermultiplets of U (2, 2/4), Class. Quantum Grav. 2 (1985) L11.
[44] H. Nicolai, E. Sezgin, Y. Tanii, Conformally invariant supersymmetric field theories on S p × S 1 , Nucl. Phys.
B 305 (1988) 483.
[45] R.R. Metsaev, Massless arbitrary spin fields in AdS5 , hep-th/0201226.
[46] M. Günaydin, S. Takemae, Unitary supermultiplets of OSp(8∗ |4) and the AdS7 /CFT 6 duality, Nucl. Phys.
B 578 (2000) 405, hep-th/9910110.
[47] M. Günaydin, P. van Nieuwenhuizen, N.P. Warner, General construction of anti-de Sitter superalgebras and
the spectrum of the S 4 compactification of eleven-dimensional supergravity, Nucl. Phys. B 255 (1985) 63.
[48] S. Minwalla, Restrictions imposed by superconformal invariance on quantum field theories, Adv. Theor.
Math. Phys. 2 (1998) 781, hep-th/9712074.
[49] P.S. Howe, E. Sezgin, Superbranes, Phys. Lett. B 390 (1997) 133, hep-th/9607227.
[50] S. Ferrara, E. Sokatchev, Superconformal interpretation of BPS states in AdS geometries, Int. J. Theor.
Phys. 40 (2001) 935, hep-th/0005151.
[51] S. Ferrara, E. Sokatchev, Universal properties of superconformal OPEs for 1/2 BPS operators in 3 D 6,
hep-th/0110174.
[52] S. Ferrara, E. Sokatchev, Conformal primaries of OSp(8/4, R) and BPS states in AdS4 , JHEP 0005 (2000)
038, hep-th/0003051.
[53] L. Andrianopoli, S. Ferrara, “Non-chiral” primary superfields in the AdSd+1 /CFT d correspondence, Lett.
Math. Phys. 46 (1998) 265, hep-th/9807150.
[54] S. Minwalla, Particles on AdS4/7 and primary operators on M2/5 brane worldvolumes, JHEP 9810 (1998)
002, hep-th/9803053.
[55] E. Bergshoeff, M. de Roo, B. de Wit, Extended conformal supergravity, Nucl. Phys. B 182 (1981) 173.
[56] P. Howe, K.S. Stelle, P.K. Townsend, Supercurrents, Nucl. Phys. B 192 (1981) 332.
[57] L. Andrianopoli, S. Ferrara, On short and long SU(2, 2/4) multiplets in the AdS/CFT correspondence, Lett.
Math. Phys. 48 (1999) 145, hep-th/9812067.
[58] S. Ferrara, C. Fronsdal, A. Zaffaroni, On N = 8 supergravity on AdS5 and N = 4 superconformal Yang–
Mills theory, Nucl. Phys. B 532 (1998) 153, hep-th/9802203.
[59] L. Adrianopoli, S. Ferrara, KK excitations on AdS5 × S 5 as N = 4 “primary” superfields, Phys. Lett. B 430
(1998) 248, hep-th/9803171.
[60] S. Ferrara, M.A. Lledó, A. Zaffaroni, Born–Infeld corrections to D3-brane action in AdS5 × S 5 and
N = 4, d = 4 primary superfields, Phys. Rev. D 58 (1998) 105029, hep-th/9805082.
[61] S. Ferrara, A. Zaffaroni, Bulk gauge fields in AdS supergravity and supersingletons, hep-th/9807090.
[62] S. Ferrara, A. Zaffaroni, Superconformal field theories, multiplet shortening, and the AdS5 /SCFT 4
correspondence, hep-th/9908163.
[63] S. Ferrara, E. Sokatchev, Short representations of SU(2, 2/N ) and harmonic superspace analyticity, Lett.
Math. Phys. 52 (2000) 247, hep-th/9912168.
[64] L. Adrianopoli, S. Ferrara, E. Sokatchev, B. Zupnik, Shortening of primary operators in N -extended SCFT 4
and harmonic analyticity, Adv. Theor. Math. Phys. 3 (1999) 1149, hep-th/9912007.
[65] P.J. Heslop, P.S. Howe, A note on composite operators in N = 4 SYM, Phys. Lett. B 516 (2001) 367, hep-
th/0106238.
[66] P.S. Howe, G. Sierra, P.K. Townsend, Supersymmetry in six dimensions, Nucl. Phys. B 221 (1983) 331.
[67] S. Ferrara, E. Sokatchev, Representations of (1, 0) and (2, 0) superconformal algebras in six dimensions:
massless and short superfields, Lett. Math. Phys. 51 (2000) 55, hep-th/0001178.
[68] H. Liu, Scattering in anti de Sitter space and operator product expansion, Phys. Rev. D 60 (1999) 106005,
hep-th/9811152.
[69] P. Ramond, Group theory for string states, in: M.J. Bowick, F. Gürsey (Eds.), in: High Energy Physics 1985,
Vol. 1, World Scientific, Singapore, 1986, p. 274.
[70] V.K. Dobrev, E. Sezgin, A remarkable representation of the SO(3, 2) Kac–Moody algebra, Int. J. Mod. Phys.
A 6 (1991) 4699.
[71] M.J. Duff, Supermembranes: the first fifteen weeks, Class. Quantum Grav. 5 (1988) 189.
[72] I.R. Klebanov, A.A. Tseytlin, Entropy of near-extremal black p-branes, Nucl. Phys. B 475 (1996) 164,
hep-th/9604089.
[73] M. Henningson, K. Skenderis, The holographic Weyl anomaly, JHEP 9807 (1998) 023, hep-th/9806087.
[74] K. Intriligator, Maximally supersymmetric RG flows and AdS duality, Nucl. Phys. B 580 (2000) 99.
[75] E. Bergshoeff, M.J. Duff, C.N. Pope, E. Sezgin, Supersymmetric supermembrane vacua and singletons,
Phys. Lett. B 199 (1987) 69.
[76] E. Bergshoeff, M.J. Duff, C.N. Pope, E. Sezgin, Compactifications of the eleven-dimensional supermem-
brane, Phys. Lett. B 224 (1989) 71.
[77] M.J. Duff, C.N. Pope, E. Sezgin, A stable supermembrane vacuum with a discrete spectrum, Phys. Lett.
B 225 (1989) 319.
[78] J. Maldacena, H. Ooguri, Strings in AdS3 and the SL(2, R) WZW model. Part 1: The spectrum, J. Math.
Phys. 42 (2001) 2929, hep-th/0001053.
[79] E. Witten, Some comments on string dynamics, in: STRINGS 95: Future Perspectives in String Theory,
hep-th/9507121.
[80] A. Strominger, Open p-branes, Phys. Lett. B 383 (1996) 44, hep-th/9512059.
[81] E. Witten, Five-branes and M-theory on an orbifold, Nucl. Phys. B 463 (1996) 383, hep-th/9512219.
[82] X. Bekaert, M. Henneaux, A. Sevrin, Deformations of chiral two-forms in six dimensions, Phys. Lett. B 468
(1999) 228, hep-th/9909094.
[83] P.S. Howe, E. Sezgin, D = 11, p = 5, Phys. Lett. B 394 (1997) 62, hep-th/9611008.
[84] P.S. Howe, E. Sezgin, P.C. West, Covariant field equations of the M-theory five-brane, Phys. Lett. B 399
(1997) 49, hep-th/9702008.
[85] M. Cederwall, B.E.W. Nilsson, P. Sundell, An action for the super-5-brane in D = 11 supergravity,
JHEP 9804 (1998) 007, hep-th/9712059.
[86] E. Bergshoeff, E. Sezgin, P.K. Townsend, Supermembranes and eleven-dimensional supergravity, Phys. Lett.
B 189 (1987) 75.
[87] S.K. Gandhi, K.S. Stelle, Vanishing of the supermembrane partition function, Class. Quantum Grav. 5 (1988)
L127.
[88] P.K. Townsend, Three lectures on supermembranes, in: M. Green, M. Grisaru, R. Iengo, E. Sezgin,
A. Strominger (Eds.), Superstrings’88, World Scientific, Singapore, 1989.
[89] F. Bastianelli, S. Frolov, A.A. Tseytlin, Conformal anomaly of (2, 0) tensor multiplet in six dimensions and
AdS/CFT correspondence, JHEP 0002 (2000) 013, hep-th/0001041.
[90] G. Arutyunov, E. Sokatchev, Implications of superconformal symmetry for interacting (2, 0) tensor
multiplets, hep-th/0201145.
[91] J. Engquist, P. Sundell, E. Sezgin, N = 1 superspace formulation of D = 4, N = 1, 2, 4, 8 higher spin gauge
theories, to appear.
[92] S. Prokushkin, M. Vasiliev, Higher spin gauge interactions for massive matter fields in 3D AdS spacetime,
Nucl. Phys. B 545 (1999) 385, hep-th/9806236.
[93] S. Prokushkin, A. Segal, M. Vasiliev, Coordinate-free action for AdS3 higher-spin–matter systems, Phys.
Lett. B 478 (2000) 333, hep-th/9912280342.
[94] M. Günaydin, AdS/CFT dualities and the unitary representations of non-compact groups and supergroups:
Wigner versus Dirac, hep-th/0005168.
Isocurvature perturbations in the Ekpyrotic Universe

A. Notari a , A. Riotto b
a Scuola Normale Superiore, Piazza dei Cavalieri 7, Pisa I-56126, Italy
b INFN, sezione di Padova, Via Marzolo 8, Padova I-35131, Italy
Received 6 May 2002; received in revised form 20 August 2002; accepted 23 August 2002
Abstract
The Ekpyrotic scenario assumes that our visible Universe is a boundary brane in a five-dimensional
bulk and that the hot Big Bang occurs when a nearly supersymmetric five-brane travelling along
the fifth dimension collides with our visible brane. We show that the generation of isocurvature
perturbations is a generic prediction of the Ekpyrotic Universe. This is due to the interactions in the
kinetic terms between the brane modulus parametrizing the position of the five-brane in the bulk and
the dilaton and volume moduli. We show how to separate explicitly the adiabatic and isocurvature
modes by performing a rotation in field space. Our results indicate that adiabatic and isocurvature
perturbations might be cross-correlated and that curvature perturbations might be entirely seeded by
isocurvature perturbations.
1. Introduction
The Ekpyrotic cosmology [1,2] has recently received a great deal of attention because
it represents an alternative to inflationary cosmology [3]. It addresses the cosmological
horizon, flatness and monopole problems and generates a spectrum of density perturbations
without invoking any superluminal expansion. Being realized in the context of eleven-
dimensional heterotic M-theory, the Ekpyrotic scenario is also based on strong particle
physics grounds. It assumes that our visible Universe is a boundary brane in the five-
dimensional bulk obtained compactifying six of the eleven dimensions on a Calabi–Yau
manifold. The hot Big Bang occurs when a nearly supersymmetric five-brane travelling
along the fifth dimension collides with our visible brane. The five-brane is attracted towards
the boundary brane where we live by an inter-brane potential which is exponentially
E-mail address: antonio.riotto@pd.infn.it (A. Riotto).
PII: S 0 5 5 0 - 3 2 1 3 ( 0 2 ) 0 0 7 6 5 - 4
372 A. Notari, A. Riotto / Nuclear Physics B 644 (2002) 371–382
suppressed when the two branes are far apart. The dynamics of the system is described
on the four-dimensional point of view in terms of a scalar field ϕ, the brane modulus,
parametrizing the separation between the two branes.
The branes are assumed to start widely separated almost at rest and the four-dimensional
observer experiences a contraction of the cosmological scale factor characterized by a
singularity at the time when the two branes hit each other. Since the Hubble radius contracts
faster than comoving scales, microscopic sub-horizon fluctuations during the contracting
phase may well produce curvature perturbations on cosmological scales today [1,4]. The
spectrum of these adiabatic perturbations was claimed to be scale-independent if the
brane modulus potential scales as e−cϕ/Mp with c 1. This finding has been recently
challenged in a series of papers [5–10]. The difficulty in determining the final spectrum
of perturbations arises from the fact that there is no prescription on how to match the
perturbations generated in a contracting phase across the ‘bounce’ to those in the expanding
hot Big Bang phase when radiation dominates.1 For instance, if the matching is performed
at ρ + δρ = const hypersurfaces, the Ekpyrotic scenario leads to a blue spectrum with
spectral index for scalar perturbations ns = 3. However, if the matching hypersurface is
chosen to be differently the authors of Ref. [12] claim that one may obtain a flat spectrum
with spectral index ns = 1.
In this paper we wish to take a modest step and observe that in the Ekpyrotic scenario
both adiabatic and isocurvature perturbations may be generated during the contraction
phase when the five-brane slowly approaches our visible world. A cross-correlation
between entropy and curvature perturbations may left imprinted. We will also show
that curvature perturbations may be entirely sourced by isocurvature perturbations, thus
providing a way to produce a scale-invariant spectrum of adiabatic perturbations, at least
before the bounce.
The paper is organized as follows. In Section 2 we briefly summarize how the effective
action for the five-brane modulus is derived emphasizing the coupling between the five-
brane modulus and the dilaton and the volume modulus in the kinetic term. This coupling
is responsible for the generation of entropy modes. In Section 3 we describe the generation
of adiabatic and isocurvature perturbations using the powerful technique of rotation in
field space and obtaining the equation for the gravitational potential in terms of properly
defined adiabatic and isocurvature fields. Finally, we end with some concluding remarks in
Section 4.
2. The effective action of the five-brane modulus
Compactification of eleven-dimensional M-theory on a Calabi–Yau threefold X6 times

S 1 /Z2 leads to N = 1 supersymmetry in four space–time dimensions. Five-branes
configurations preserve this supersymmetry if their world-volume W6 is suitably aligned:
it should enclose four-dimensional Minkowski space M4 and a holomorphic two-cycle
1 At present, there is no known mechanism to reverse the bounce. This problem has been circumvented by
adopting the so-called ‘cyclic’ model where the two colliding branes are the boundary branes [11].
A. Notari, A. Riotto / Nuclear Physics B 644 (2002) 371–382 373
C2 in the Calabi–Yau threefold, i.e., W6 = M4 × C2 [13–15]. With this embedding, the

excitations of the five-brane are described by a D = 6 tensor supermultiplet [16,17].
The fields are a chiral antisymmetric tensor Bmn (m, n = 0, 1, . . . , 5), five scalar fields
X(1) , . . . , X(5) specifying the position of the world-volume W6 in the eleven-dimensional
space M11 and their fermionic partners. The antisymmetric tensor has a self-dual field
strength. This symmetry has important implications in four dimensions since the effective
theory will have a massless antisymmetric tensor dual to a pseudoscalar or, in terms of
supermultiplets a chiral multiplet dual to a linear multiplet.
Some of the D = 6 tensor modes are deeply related to the Calabi–Yau geometry. There
are, however, universal modes which can be more easily described, the most obvious
example being the real scalar X(x µ ) (µ = 0, 1, . . . , 3) associated to the position of the
five-brane on the orbifold S 1 /Z2 . Upon reduction to four-dimensions, the bosonic five-
brane excitations which remain are Bµν (x µ ), B45 (x µ ) and X(x µ ) where the self-duality
condition inherited from the D = 6 tensor relates Bµν and B45 . Therefore, each five-brane
generates in M4 two bosonic degrees of freedom. By N = 1 supersymmetry, they belong
to a linear multiplet (or a chiral multiplet by the so-called chiral-linear duality).
To describe the dynamics of the five-brane modulus X parametrizing the position of
the five-brane in S 1 /Z2 and its couplings to four-dimensional supergravity, one has to
couple the linear multiplet containing the field X to four-dimensional supergravity. To do
so, one writes the action for the five-brane coupled to eleven-dimensional supergravity
using the Pasti–Sorokin–Tonin formalism to write covariant Lagrangians for self-dual (or
anti-self-dual) tensors [18]. Upon reduction, one can compute the effective supergravity
couplings of the five-brane modulus X to the N = 1 supergravity and dilaton multiplets,
the modulus of the Calabi–Yau volume and to the E6 × E8 gauge fields and chiral matter
on the boundaries [19–21].
The dilaton and the universal volume modulus are respectively described in four-
dimensions by two vector multiplets V and VT . Bianchi identities in M11 constrain V
to be linear and VT = T + T where T is a chiral multiplet.
After properly rescaling the five-brane modulus X = κ 2 ϕ, where κ = Mp −1 is the
inverse of the four-dimensional Planck scale, the part of the Lagrangian which is of interest
to us is [20,21]

1 3 τ
Lkin = ∂µ χ∂ µ χ + ∂µ β∂ µ β + e(β−χ)/Mp ∂µ ϕ∂ µ ϕ , (2.1)
4 4 2
where τ is a dimensionless constant parametrizing the five-brane tension in units of κ and
χ and β are the dilaton and the volume modulus, respectively. The factor e(β−χ)/Mp in
front of the kinetic term of the field ϕ is originated from the fact that the kinetic terms are
normalized by the world-volume induced metric [20].2
The form (2.1) of the kinetic term of the modulus ϕ parametrizing the position of
the five-brane in S 1 /Z2 makes it explicit that in the Ekpyrotic scenario the moving five-
2 In the chiral version of the theory where the dilaton field is described by the real part Re S of the scalar
component of the chiral multiplet S and the five-brane modulus belongs to a chiral multiplet
S, the kinetic term
S+
τ ( S)2

(2.1) can be understood as coming from the Kähler potential K = − log S + S − 16 − 3 log(T + T).
T +T
brane induces a nontrivial dynamics of the dilaton and volume moduli since solutions
with exactly constant β and χ do not exist. Similarly, the perturbations of the five-brane
modulus, of the dilaton and of the volume modulus might mix as soon as the five-brane
moves. This implies not only that both adiabatic and isocurvature perturbations may be
created, but also that they might be cross-correlated and that isocurvature perturbations
may seed curvature perturbations on super-horizon scales.
3. Cosmological perturbations
In order to simplify the treatment of the cosmological perturbations, from now on we

consider the evolution of two scalar fields, the brane modulus ϕ and the dilaton χ . This
amounts to saying that the volume modulus β is tightly fixed at the minimum of its potential
V (β), that is its mass squared ∂ 2 V /∂β 2 is much larger than the time-dependent term
e(β−χ)/Mp ϕ̇ 2 /Mp 2 . Needless to say, one might equivalently consider the case in which
the dilaton χ is fixed at its minimum and the volume modulus β is free to move. Our
following analysis applies to both cases.
√ −
β/(2Mp
By shifting the dilaton field χ → 2 χ and the brane modulus ϕ → e √τ ) ϕ, we
may write the Lagrangian of the system as
1 1
L = ∂µ χ∂ µ χ + e−αχ ∂µ ϕ∂ µ ϕ − V (ϕ) − V (χ), (3.1)
2 2
√
where α = 2/Mp ,

− p2 ϕ/Mp
V (ϕ) = −V0 e (p 1) (3.2)
is the potential of the brane modulus in the low energy effective action of the Ekpyrotic
scenario [2,4] and V (χ) is the dilaton potential. Since the field ϕ is related to the brane
separation [2], at early times when the five-brane and the boundary brane are separated by
a large distance the scalar field ϕ is very large and positive and the two branes approach
each other very slowly.
Varying the Lagrangian (3.1) we obtain the equations of motion for the homogeneous
fields χ(t) and ϕ(t)
α 2 −αχ
χ̈ + 3H χ̇ + ϕ̇ e + Vχ = 0, (3.3)
2
ϕ̈ + 3H ϕ̇ − α ϕ̇ χ̇ + e Vϕ = 0,
αχ
(3.4)
where Vx denotes the derivative ∂V /∂x and H the Hubble rate.
In order to study the evolution of linear perturbations in the scalar fields coupled to
gravity, we choose the longitudinal gauge, where the line element perturbed to first order
in the scalar metric perturbations is (in the absence of anisotropic stress perturbations)

ds 2 = (1 + 2Φ) dt 2 − a 2 (t) (1 − 2Φ)δij dx i dx j , (3.5)
where Φ is the gravitational potential [22–24].
Scalar field perturbations, with comoving wavenumber k = 2πa/λ for a mode with
physical wavelength λ, then obey the perturbation equations
2
¨ + 3H δϕ
δϕ ˙ + k δϕ + Vϕϕ eαχ δϕ − 4ϕ̇ Φ̇ − 2Vϕ eαχ Φ
a2
˙ + ϕ̇ δχ)
+ α(χ̇ δϕ ˙ − αVϕ eαχ δχ = 0 (3.6)
and
k2
¨ + 3H δχ
δχ ˙ + δχ + Vχχ δχ − 4χ̇ Φ̇ − 2Vχ Φ
a2
α 2 2 −αχ ˙ −αχ = 0.
− ϕ̇ e δχ + α ϕ̇ δϕe (3.7)
2
The perturbed Einstein equations read
4π
Φ̈ + 4H Φ̇ + 2Ḣ + 3H 2 Φ = δp, (3.8)
Mp 2
4π
Φ̇ + H Φ = − δq, (3.9)
Mp 2
k2 4π
3H Φ̇ + 3H 2 Φ + 2
Φ=− δρ, (3.10)
a Mp 2
where the total energy and momentum perturbations are given by [25]

˙ ˙ −αχ α 2 −αχ
δp = δTi = χ̇ δχ + ϕ̇ δϕe
i
− ϕ̇ e δχ
2
2
− Φ χ̇ + ϕ̇ 2 e−αχ − Vϕ δϕ − Vχ δχ, (3.11)

α
δρ = δT00 = χ̇ δχ ˙ + ϕ̇ δϕe
˙ −αχ − ϕ̇ 2 e−αχ δχ
2
2 2 −αχ

− Φ χ̇ + ϕ̇ e + Vϕ δϕ + Vχ δχ, (3.12)

δq = δT0 = − χ̇δχ + ϕ̇δϕe−αχ .
i
(3.13)
We obtain a useful relation if we take the Eq. (3.9) times −3H and then sum it to Eq. (3.10)
[26]
k2 4π
2
Φ =− -m , (3.14)
a Mp 2
where
-m ≡ δρ − 3H δq (3.15)
is used to represent the total matter perturbation [25].
Let us now turn to isocurvature perturbation. When two scalar fields are present
in the dynamics of the system, isocurvature perturbations are generated [25]. In our
case, inserting the expressions for the pressure and the energy density as well as their
perturbations, the entropy perturbation is measured by the quantity

2
S=
3(2V̇ + 3H (χ̇ 2 + ϕ̇ 2 e−αχ ))(χ̇ 2
+ ϕ̇ 2 e−αχ )

× V̇ + 3H χ̇ 2 + ϕ̇ 2 e−αχ (Vχ δχ + Vϕ δϕ)

α 2 −αχ 2
˙ ˙
+ V̇ χ̇ δχ + ϕ̇ δϕe −αχ
− ϕ̇ e 2 −αχ
δχ − Φ χ̇ + ϕ̇ e . (3.16)
2
3.1. Rotation in field space
In order to clarify the role of adiabatic and entropy perturbations, their evolution and
their interconnection, from now on we will follow Ref. [26] and define new adiabatic and
entropy fields by a rotation in field space. The “adiabatic field”, σ , represents the path
length along the classical trajectory, such that
σ̇ = cos θ χ̇ + sin θ ϕ̇e−αχ/2 , (3.17)

where
χ̇ ϕ̇e−αχ/2
cos θ = , sin θ = . (3.18)
χ̇ 2 + ϕ̇ 2 e−αχ χ̇ 2 + ϕ̇ 2 e−αχ
The equation of motion of the field σ is remarkably simple
σ̈ + 3H σ̇ + Vσ = 0, (3.19)
where
Vσ = cos θ Vχ + sin θ Vϕ eαχ/2 . (3.20)

The fluctuation δσ is the component of the two-field perturbation vector along the direction
of the background fields’ evolution [26]. Fluctuations orthogonal to the background
classical trajectory represent nonadiabatic perturbations, and we define the “entropy field”
s such that
δs = − sin θ δχ + cos θ δϕe−αχ/2 . (3.21)

From this definition, it follows that s = const along the classical trajectory, and hence
entropy perturbations are automatically gauge-invariant [27]. Perturbations in δσ , with
δs = 0, describe adiabatic field perturbations, and this is why we refer to σ as the “adiabatic
field”.
The angle θ changes with time along the classical trajectory according to the equation
Vs 1
θ̇ = − + α sin θ σ̇ , (3.22)
σ̇ 2
where
Vs = − sin θ Vχ + cos θ Vϕ eαχ/2 . (3.23)

Armed with these definitions, we are now ready to compute the equation of motion of
the gravitational potential Φ. A straightforward computation leads to

σ̈ σ̈ k2 8π
Φ̈ + H − 2 Φ̇ + 2Ḣ − 2H Φ + 2Φ =− Vs δs. (3.24)
σ̇ σ̇ a Mp 2
This equation makes manifest the neat separation of the roles played by the adiabatic and
isocurvature perturbations. Indeed, on the left-hand side only the adiabatic field σ appears
while on the right-hand side there is a source which is proportional only to the relative
entropy perturbations δs between the brane-modulus and the dilaton field.
Eq. (3.24) shows explicitly how on large scales entropy perturbations can source the
gravitational potential in the Ekpyrotic cosmology. When the five-brane slowly approaches
the boundary brane and the brane modulus decreases in time, perturbations in the brane
modulus, in the dilaton field and in the gravitational potential are generated. Furthermore,
adiabatic and isocurvature perturbations are inevitably correlated. These results are similar
to what found in standard inflation when two or more scalar fields are present [26,28–31].
Using now V̇ = Vσ σ̇ , δV = Vσ δσ + Vs δs and
χ̇ δχ ˙ −αχ − α ϕ̇ 2 e−αχ δχ = σ̇ δσ
˙ + ϕ̇ δϕe ˙ + Vs δs, (3.25)
2
we can write the isocurvature source S in Eq. (3.16) as
˙ + Vs δs) − ΦVσ σ̇ 2
(Vσ + 3H σ̇ )(Vσ δσ + Vs δs) + Vσ (σ̇ δσ
S =2 . (3.26)
3(2Vσ + 3H σ̇ )σ̇ 2
Note this expression is the same as in the standard case α = 0 [26]. We can rewrite the last
expression as
˙ + Vs δs) − ΦVσ σ̇ 2
−σ̈ (Vσ δσ + Vs δs) + Vσ (σ̇ δσ
S=2
3(2Vσ + 3H σ̇ )σ̇ 2
˙ − Φ σ̇ − σ̈ δσ ) + Vs δs(Vσ − σ̈ )
Vσ (σ̇ δσ 2
=2 . (3.27)
3(2Vσ + 3H σ̇ )σ̇ 2
Making use of Eq. (3.14) we can express -m in terms of the new fields δσ and δs [26]
˙ − σ̇ Φ) − σ̈ δσ + 2Vs δs.
-m = σ̇ (δσ (3.28)
Finally, using Eqs. (3.14) and (3.28) we get
Vσ -m − 2Vσ Vs δs + Vs δs(2Vσ + 3H σ̇ )
S=2
3(2Vσ + 3H σ̇ )σ̇ 2
2
Mp 2 Vσ ak 2 Φ H Vs δs
=− + . (3.29)
6π(2Vσ + 3H σ̇ )σ̇ 2 (2Vσ + 3H σ̇ )σ̇
Eq. (3.24) might be rewritten as [25]
ṗ
Ṙ = −3H S. (3.30)
ρ̇
The change in the curvature perturbation on large scales can therefore be directly related
to the nonadiabatic part of the pressure perturbation [25].
Clearly, there can be significant changes in the gravitational potential on large scales
and a large cross-correlation between the adiabatic and the isocurvature modes only if the
entropy perturbation is not suppressed. The next step is therefore to compute the equation
of motion of the entropy field δs. The computation is lengthy, but straightforward. One
finds
2
¨ ˙ k Vs Mp 2 k 2
δs + 3H δs + − θ̇ 2
+ V ss + Γ α δs = − Φ, (3.31)
a2 σ̇ 2 2πa 2
Vss = cos2 θ eαχ Vϕϕ + sin2 θ Vχχ (3.32)

and

α 1 α2
Γα = − cos θ Vσ − 2 sin θ Vs − sin 2θ Vϕ eαχ/2 − σ̇ 2 cos2 θ. (3.33)
2 2 4
Eq. (3.31) is the same found in Ref. [26] apart from the new α-dependent terms present
in Γα . On large scales the inhomogeneous term proportional to the gravitational potential
Φ becomes negligible and the field δs satisfies a homogeneous second-order equation of
motion for the entropy perturbation which is decoupled from the adiabatic modes and
metric perturbations. This amounts to saying that if the initial entropy perturbation is zero
on large scales, it will remain zero at later times as well.
Eqs. (3.24) and (3.31) are the key equations which govern the evolution of the adiabatic
and entropy perturbations in the Ekpyrotic scenario before the collision of the five-brane
and the boundary brane. In principle, they allow us to follow the effect on the adiabatic
curvature perturbation due to the presence of entropy perturbations before the collision
and they provide the “initial” conditions for the matching technique through the bounce in
a four-dimensional approach.
The general super-Hubble solution of Eqs. (3.24) and (3.31) can be written as
Φ = f+ (t) + f− (t) + S(t), (3.34)

δs = g+ (t) + g− (t), (3.35)
where the functions f± and g± are the growing/decaying (or constant) modes of the
homogeneous equations and S(t) is a particular integral of the inhomogenous Eq. (3.31).
From Eq. (3.24) one finds that the amplitude of the particular integral S(t) is correlated
with the amplitude of the entropy perturbation
2π 2
Φ(k)δs ∗ (k ) = 3 CΦδs δ (3)(k − k ) ∝ S(t). (3.36)
k
A nonvanishing cross-correlation between the adiabatic and isocurvature modes are
expected before collision. One can now envisage various possibilities.
If |χ̇/ϕ̇| 1 and V (χ) = 0, the dynamics of the system is close to the single field
version of the Ekpyrotic Universe [1], i.e., the √ scale factor is given by a(t) ∼ (−t)p
while the brane modulus scales like ϕ(t) ∼ 2p √ln(−Mt), where M (V0 /pMp )
2 2
(p 1). It is easy to show that |χ̇/ϕ̇| = |αMp p/2| 1. The adiabatic field σ is
given by exp(−αχ/2)ϕ and the solution to the homogeneous part of the equation for
the gravitational potential (3.24) leads to Φ = A(k) Ha + B(k) at superhorizon scales in
the collapsing phase. The A-growing mode has a scale-independent spectrum, |A|2 k 3 is
k-independent, while the constant mode has a blue spectrum. One then needs to match this
solution to the usual (approximately constant) gravitational potential in the radiation era.
If the matching from the collapsing phase to the radiation era is performed on constant
energy surfaces, the gravitational potential in the radiation era inherits the blue spectrum
from the constant mode in the collapsing phase. However, a nonzero surface tension—
provided by some high-energy theory ingredient—is needed to go through the bounce.
This implies that the transition surface needs not to be a constant energy surface [12].
For instance, imposing the matching on a surface where its shear vanish, one finds that the
gravitational potential in the radiation phase inherits the flat spectrum of A [12]. On the top
of that, a cross-correlation between adiabatic and isocurvature modes is generated before
the collapse with CΦδs ∝ αpMp k 2 = O(p)k 2 , i.e., a blue spectrum.
If |χ̇/ϕ̇| 1 and V (χ) = −M 4 exp(−cχ/Mp ) with M 4 V0 0, one obtains a
sort of modified version of the pre-Big Bang model [32] with a potential for the dilaton
field. The adiabatic field σ is identified
√ with the dilaton field χ and the final spectrum of
curvature perturbations is flat if c = 3 and the matching through the bounce is done onto
a constant energy surface [33].3 Even in this a case cross-correlation between adiabatic and
isocurvature modes is present. Yet, it is suppressed.
A much more interesting situation is realized if the brane-modulus potential is very tiny,
V0 0 and V (χ) = −M 4 exp(−βχ/Mp ), with M some mass scale. Going to conformal
time dτ = dt/a and integrating Eq. (3.4), one finds ϕ = C exp(αχ)/a 2 , where primes
mean derivatives with respect to conformal time τ and C is an integration constant.
The ansatz a(τ ) = (−Mp (1 − q)τ )q/(1−q) and χ(τ ) = A ln(−Mp (1 − q)τ ) satisfy the
equations of motion if αA = 2(3q − 1)/(1 − q), βA/Mp = 2/(1 − q) and αMp /β =
3q − 1. Suppose now that the energy density of the system is dominated by the kinetic term
of the brane-modulus. This will be the case if, for instance, A C/Mp . After introducing
the new variable δS = aδs, Eq. (3.31) reduces to

a α2
δS + k 2 − − ϕ 2 e−αχ δS = 0. (3.37)
a 4
√
Since C 2 2qMp 4 and α = 2/Mp , one finds

2q 2
δS + k 2 − δS = 0. (3.38)
(1 − q)2 τ 2
A nearly invariant spectrum for the entropy perturbations is obtained if 2q 2/(1 − q)2 2,
or q 1/2. This is a desirable output since it means that adiabatic perturbations are entirely
sourced by entropy perturbations inducing a flat spectrum for curvature perturbations with
maximum cross-correlation.
3 If the matching is done using the prescription discussed in Ref. [12], the spectrum is flat if c 1.
4. Concluding remarks
Let us conclude with some comments. The generation of adiabatic plus isocurvature
perturbations and their cross-correlation we have described in this paper occur in the
Ekpyrotic scenario before the moving five-brane collapses onto the boundary brane.
Isocurvature perturbations might not survive after the bounce if during the subsequent
period of reheating both the brane modulus and the dilaton decay into the same species.
In order to have isocurvature perturbations deep in the radiation era after the collapse it
is necessary to have at least one nonzero isocurvature perturbation Sαβ ≡ δα /(1 + wα ) −
δβ /(1 + wβ ) = 0, where δα = δρα /ρα and wα = pα /ρα (the ratio of the pressure to the
energy density) for some components α and β of the system. This may happen if the
fields responsible for the isocurvature perturbations decay into radiation and cold dark
matter at different epochs. On the other hand, we have seen that curvature perturbations
may be entirely seeded by isocurvature perturbations, thus providing a novel mechanism
to produce a scale-independent spectrum of adiabatic perturbations in the Ekpyrotic
Universe.
All previous discussions make it clear that the features of the cosmological perturbations
after the Big Bang as well as the way reheating takes place depend strongly on the details of
the transition from the collapsing to the expanding phase when the five-brane is absorbed
by the boundary brane. In this absorption process the degree of freedom represented by the
brane modulus gets replaced by new degrees of freedom.
In eleven-dimensional M-theory, E8 × E8 gauge fields with strength Fzi zj living on
the boundary of the eleventh dimension and in the six-dimensional Calabi–Yau manifold
(zi with i = 1, 2, 3 are the complex coordinates of such a manifold) satisfy equations of
motion of the type Fzz = Fz̄z̄ = g zz̄ Fzz̄ = 0. Gauge field configurations satisfying these
equations are generically called instantons (for instance, if the Calabi–Yau manifold is
a two-dimensional torus times a four-dimensional variety K3, one gets the traditional
quantum field theory instanton equations F = ±F ). In nonstandard compactifications of
M-theory with a certain number N of five-branes, one schematically obtains the following
constraint

N + F ∧ F = const, (4.1)

that is the number of five-branes plus the “number” of instantons given by F ∧ F is
conserved. From Eq. (4.1) one can infer that in M-theory itis possible to replace one five-
brane with one instanton and vice versa. An instanton with F ∧ F = a can shrink to zero
size, becoming a so-called small instanton with F ∧ F = aδ(zi ) and leave the boundary
brane along the eleventh dimension under the form of a five-brane [34]. A unit of instanton
flux is replaced by a unit of five-brane flux still satisfying the constraint (4.1).
This phenomenon is similar to what happens in Type IIA string theory where D4-
branes may get emitted or absorbed by a set of D8-brane plus O8-orientifold. A crucial
role for describing this phenomenon is played by the strings stretched between the D4-
brane and the (D8 + O8) system. If they are massive, i.e., if their length is nonzero,
it means that the D4-brane is in the bulk away from the (D8 + O8) system. On the
contrary, if the strings are tensionless, their length is zero and the D4-brane touches the
(D8 + O8) system. The transition may be described from the four-dimensional point of
view as a Higgs mechanism. The D4-brane can be described by an N = 2 vector multiplet
{Wα , φ} while the strings stretched between the D4-brane and the (D8 + O8) system are
described in terms of some hypermultiplets Y . They appear as particles on the D4-brane.
Upon compactification to four-dimensions the Lagrangian (in the N = 1 supersymmetry
language) can be written as

+ d 4 θ (φ̄φ + Y
d 2 θ Wα2 + Y φ Y Y ). (4.2)
When
φ = 0, the Y -field is massive. This means that the strings between the D4-brane
and the (D8 + O8) system have a nonvanishing length (or tension): the D4-brane is in the
bulk. At the transition point
Y =
φ = 0, tensionless (massless) strings appear in the
spectrum and the D4-brane is absorbed by the (D8 + O8) system. This transition gives
rise to gauge field configurations whose moduli space contain a number of free parameters
matching the number of the degrees of freedom before the absorption. Some of the Y -fields
are now interpreted as instanton moduli.
Going back to the case of the Ekpyrotic scenario, a crucial role is played by two-
dimensional branes, membranes, stretched between the boundary brane and the slowly
approaching five-brane [35]. These membranes are massive when the five-brane is in the
bulk and the vacuum expectation value of the brane modulus is nonzero. When the five-
brane touches the boundary brane, these membranes have zero length. Therefore, one might
hope to describe the transition from the five-brane to the small instanton in terms of a four-
dimensional Higgs mechanism as done for Type IIA string theory. The problem is that
the massless membranes appear as tensionless strings in the five-brane world-volume even
after compactifying down to four-dimensions. At the transition point the relevant degrees
of freedom of the theory are therefore tensionless strings, the anti-self-dual tensors of the
five-brane, the brane modulus and some instanton gauge field configurations. They are the
fundamental degrees of freedom which allow the description of the system during the last
moments before the collision and the subsequent Big Bang. In terms of these degrees of
freedom the theory does not admit a simple and perturbative four-dimensional description.
Nevertheless, there might be cases in which the system is tractable. One might hope to
do some progress if the transition starts when the five-brane is sufficiently far from the
boundary brane and the membranes—or better the corresponding strings—are sufficiently
massive to admit a description in terms of four-dimensional scalar fields Y . Work along
these lines is in progress [36].
Acknowledgements
We thank N. Bartolo, R. Brandenberger, E. Copeland, F. Finelli, D. Lyth, S. Matarrese

and especially A. Zaffaroni for useful discussions.
References
[1] J. Khoury, B.A. Ovrut, P.J. Steinhardt, N. Turok, Phys. Rev. D 64 (2001) 123522.
[2] J. Khoury, B.A. Ovrut, N. Seiberg, P.J. Steinhardt, N. Turok, hep-th/0108187.

[3] D.H. Lyth, A. Riotto, Phys. Rep. 314 (1999) 1.
[4] J. Khoury, B.A. Ovrut, P.J. Steinhardt, N. Turok, hep-th/0109050.
[5] D. Lyth, Phys. Lett. B 524 (2002) 1.
[6] R. Brandenberger, F. Finelli, JHEP 0111 (2001) 056.
[7] J. Hwang, Phys. Rev. D 65 (2002) 063514.
[8] J. Hwang, H. Noh, hep-th/0203193.
[9] S. Tsujikawa, Phys. Lett. B 526 (2002) 179.
[10] J. Martin, P. Peter, N. Pinto Neto, D.J. Schwarz, hep-th/0112128;
J. Martin, P. Peter, N. Pinto Neto, D.J. Schwarz, hep-th/0204222.
[11] P.J. Steinhardt, N. Turok, Phys. Rev. D 65 (2002) 126003, hep-th/0111098.
[12] R. Durrer, F. Vernizzi, hep-ph/0203275.
[13] K. Becker, M. Becker, A. Strominger, Nucl. Phys. B 456 (1995) 130.
[14] M. Bershadsky, C. Vafa, V. Sadov, Nucl. Phys. B 463 (1996) 420.
[15] E. Witten, Nucl. Phys. B 471 (1996) 135.
[16] G.W. Gibbons, P.K. Townsend, Phys. Rev. Lett. 71 (1993) 3754.
[17] D. Kaplan, J. Michelson, Phys. Rev. D 53 (1996) 3474.
[18] P. Pasti, D. Sorokin, M. Tonin, Phys. Lett. B 398 (1997) 41;
I. Bandos, K. Lechner, A. Nurmagambetov, P. Pasti, D. Sorokin, M. Tonin, Phys. Rev. Lett. (1997) 78.
[19] A. Lukas, B.A. Ovrut, D. Waldram, Phys. Rev. D 59 (1999) 106005.
[20] J.P. Derendinger, R. Sauser, Nucl. Phys. B 598 (2001) 87.
[21] M. Brandle, A. Lukas, Phys. Rev. D 65 (2002) 064024.
[22] J.M. Bardeen, Phys. Rev. D 22 (1980) 1882.
[23] V.F. Mukhanov, H.A. Feldman, R.H. Brandenberger, Phys. Rep. 215 (1992) 203.
[24] H. Kodama, M. Sasaki, Prog. Theor. Phys. Suppl. 78 (1984) 1.
[25] For a review, see A.R. Liddle, D.H. Lyth, Cosmological Inflation and Large-Scale Structure, Cambridge
Univ. Press, Cambridge, 2000.
[26] C. Gordon, D. Wands, B.A. Bassett, R. Maartens, Phys. Rev. D 63 (2001) 023506.
[27] J.M. Stewart, M. Walker, Proc. R. Soc. London A 341 (1974) 49.
[28] N. Bartolo, S. Matarrese, A. Riotto, Phys. Rev. D 64 (2001) 083514.
[29] N. Bartolo, S. Matarrese, A. Riotto, Phys. Rev. D 64 (2001) 123504.
[30] N. Bartolo, S. Matarrese, A. Riotto, hep-ph/0112261.
[31] D. Wands, N. Bartolo, S. Matarrese, A. Riotto, in preparation.
[32] G. Veneziano, Phys. Lett. B 265 (1991) 287;
M. Gasperini, G. Veneziano, Astropart. Phys. 1 (1992) 1;
M. Gasperini, G. Veneziano, Phys. Rev. D 50 (1994) 2519.
[33] F. Finelli, R. Brandenberger, hep-th/0112249.
[34] O.J. Ganor, A. Hanany, Nucl. Phys. B 474 (1996) 122.
[35] A. Strominger, Phys. Lett. B 383 (1996) 44.
[36] A. Riotto, A. Zaffaroni, in preparation.
More about the axial anomaly on the lattice

Hiroshi Igarashi, Kiyoshi Okuyama, Hiroshi Suzuki
Department of Mathematical Sciences, Ibaraki University, Mito 310-8512, Japan
Received 3 July 2002; received in revised form 2 August 2002; accepted 5 September 2002
Abstract
We study the axial anomaly defined on a finite-size lattice by using a Dirac operator which obeys
the Ginsparg–Wilson relation. When the gauge group is U(1), we show that the basic structure of
axial anomaly on the infinite lattice, which can be deduced by a cohomological analysis, persists even
on (sufficiently large) finite-size lattices. For non-Abelian gauge groups, we propose a conjecture on
a possible form of axial anomaly on the infinite lattice, which holds to all orders in perturbation
theory. With this conjecture, we show that a structure of the axial anomaly on finite-size lattices is
again basically identical to that on the infinite lattice. Our analysis with the Ginsparg–Wilson–Dirac
operator indicates that, in appropriate frameworks, the basic structure of axial anomaly is quite robust
and it persists even in a system with finite ultraviolet and infrared cutoffs.
PACS: 11.10.-z; 11.15.Ha; 11.30.Rd
Keywords: Renormalization; Regularization and renormalons; Lattice gauge field theories; Gauge symmetry;
Anomalies in field and string theories
1. Introduction
In Ref. [1], Lüscher pointed out that a cohomological analysis can be used to determine
a basic structure of the axial anomaly in Abelian gauge theories with finite lattice spacings.
This work paved a way to study the axial anomaly in a system with a finite ultraviolet cutoff
and then the technique was applied for various cases [2–6]. The crucial properties which
make this analysis possible are the locality, the gauge invariance and a topological property
of the axial anomaly. The axial anomaly defined by the gauge covariant Dirac operator [7,8]
which satisfies the Ginsparg–Wilson relation [9], especially the overlap-Dirac operator [8],
E-mail addresses: igarashi@serra.sci.ibaraki.ac.jp (H. Igarashi), okuyama@serra.sci.ibaraki.ac.jp

(K. Okuyama), hsuzuki@mx.ipc.ibaraki.ac.jp (H. Suzuki).
PII: S 0 5 5 0 - 3 2 1 3 ( 0 2 ) 0 0 8 1 2 - X
384 H. Igarashi et al. / Nuclear Physics B 644 (2002) 383–394
in fact possesses the required properties [10–14]. A further elaborate analysis with this
recognition finally led to a non-perturbative construction of anomaly-free Abelian chiral
gauge theories on the lattice [15].
The cohomological analysis, however, is limited to the case of a lattice with an infinite
size. A direct cohomological analysis for finite-size lattices is not feasible because:
(i) The analysis is based on the lattice Poincaré lemma [1], which is a lattice analogue of
the Poincaré lemma being valid for Rd . When the topology of the lattice is non-trivial
(as is the case for the periodic lattice), one expects a non-trivial d-cohomology on the
lattice.
(ii) The cohomology relevant to an analysis of axial anomaly is a local cohomology, for
which the concept of the locality is vital. The meaning of the locality, however, is not
clear on a lattice with a finite size because a Dirac operator which obeys the Ginsparg–
Wilson relation has to have exponentially decaying tails [16,17].
In this paper, we study the axial anomaly defined on a finite-size lattice by using the
Ginsparg–Wilson–Dirac operator. This analysis provides an approach to the axial anomaly
in a system with ultraviolet and infrared cutoffs. As already noted, a direct generalization of
the technique of Ref. [1] is not feasible. Instead, we point out that it is possible to determine
the structure of axial anomaly using an argument similar to that of Ref. [15] at least in
Abelian gauge theories. For non-Abelian theories, we propose a conjecture on a possible
form of axial anomaly on the infinite lattice, which is correct within perturbation theory.
Under this conjecture, a similar argument can be applied to non-Abelian cases too. These
results indicate that the structure of axial anomaly is quite robust even with ultraviolet
and infrared cutoffs in appropriate formulations (in the present case, a formulation based
on the Ginsparg–Wilson relation). We consider an even-dimensional lattice Γ whose size
is L, Γ = {x ∈ Zd | 0 xµ < L}, and the gauge field U (x, µ) ∈ G (G is the gauge group)
is assumed to be periodic on Γ , U (x + Lν̂, µ) = U (x, µ).1 The lattice spacing a is set to
be unity, except when the classical continuum limit is considered.
2. Preliminaries
The axial anomaly for the Ginsparg–Wilson–Dirac operator is defined by (see, for
example, Refs. [11,12] for the background)

1
A(x) = tr γd+1 1 − D(x, x) . (2.1)
2
The kernel of the Dirac operator D(x, y) satisfies the Ginsparg–Wilson relation
γd+1 D(x, y) + D(x, y)γd+1 = Dγd+1 D(x, y). (2.2)
1 µ̂ denotes the unit vector in direction µ.

H. Igarashi et al. / Nuclear Physics B 644 (2002) 383–394 385
The salient feature of A(x) is a lattice analogue of the analytic index theorem [10]

A(x) = n+ − n− , (2.3)
x∈Γ
which follows from the algebraic relation (2.2) alone; here n+ (n− ) is the number of zero-
modes of γd+1 D with the positive (negative) chirality. The index theorem (2.3) implies
that the Dirac operator cannot be a smooth function of the gauge field in general, because
the configuration space of lattice gauge field is arcwise connected and, barring a possibility
that n+ − n− is constant for all configurations, the integer n+ − n− jumps at certain points
in the configuration space. A sufficient condition for the smoothness of the overlap-Dirac
operator [8] is the admissibility [13,14]

1 − U (x, µ, ν) < for all x, µ, ν, (2.4)
where U (x, µ, ν) is the plaquette variable and is a constant smaller than (2 −
√
2)/d(d − 1) [14].2 After imposing this admissibility, the space of allowed gauge field
configurations may have non-trivial topology. This condition also guarantees the locality
of the operator [13,14]

D(x, y) C 1 +
x − y
p e−
x−y
/ , (2.5)
where C and p are constants and is a localization range of the Dirac operator. In
addition to the gauge covariance and the locality of the Dirac operator, we assume
that it has the same transformation law as the standard Wilson–Dirac operator under
discrete symmetries of the lattice (rotations, reflections, etc.). In particular, we require the
translational invariance, i.e., D(x, y) is identical to D(x + z, y + z) if the gauge field is
shifted at the same time U (x, µ) → U (x + z, µ).
Suppose that we have constructed a Dirac operator on a lattice with the size L. When
L → ∞, D(x, y) is promoted to a Dirac operator on the infinite lattice D(x, y) →
D ∞ (x, y). This operator also obeys the Ginsparg–Wilson relation
γd+1 D ∞ (x, y) + D ∞ (x, y)γd+1 = D ∞ γd+1 D ∞ (x, y). (2.6)

In what follows, when we compare objects on the finite lattice Γ and on the infinite
lattice, we always take repeated copies of a configuration of the gauge field on Γ as
the gauge-field configuration on the infinite lattice. D(x, x) on Γ and D ∞ (x, x) with the
argument x restricted to Γ may somewhat differ because they have exponentially decaying
tails. However, we assume that this “finite size correction” is exponentially small,

D(x, x) − D ∞ (x, x) κLν e−L/ for x ∈ Γ, (2.7)
where κ and ν are constants. The overlap-Dirac operator [8] possesses all the required
properties we assumed above.3
2 When the mass parameter m in the overlap-Dirac operator is unity, |m| = 1.

3 For the overlap-Dirac operator, whose basic building block is the Wilson–Dirac operator, one can show the

relation D(x, y) = n∈Zd D ∞ (x, y + Ln). We thank Yoshio Kikukawa and Martin Lüscher for clarifying this
point. From this relation and the locality (2.5), one obtains the bound (2.7).
Now, on the infinite lattice, the axial anomaly is given by

∞ 1 ∞
A (x) = tr γd+1 1 − D (x, x) . (2.8)
2
This is a topological field in the sense that

δA∞ (x) = 0, (2.9)
x∈Zd
where δ denotes a local variation of the gauge field. This property can be shown from
the Ginsparg–Wilson relation (2.6) (see Ref. [2], for example). A∞ (x) is thus a local
topological gauge-invariant pseudoscalar field.4 When the gauge group is U(1), we can
then apply the cohomological analysis [1,2] to this quantity. The result is5
A∞ (x) = q(x) + ∂µ∗ kµ∞ (x), (2.11)

where kµ∞ (x) is a local gauge-invariant axial vector current (which is translational-
invariant) and the topological density q(x) is given by
N i d/2
q(x) = µ ν ···µ ν Fµ ν (x)Fµ2 ν2 (x + µ̂1 + ν̂1 ) · · ·
(4π)d/2(d/2)! 1 1 d/2 d/2 1 1
× Fµd/2 νd/2 (x + µ̂1 + ν̂1 + · · · + µ̂d/2−1 + ν̂d/2−1 ), (2.12)
with an integer N . The Abelian field strength is defined by6
1
Fµν (x) = ln U (x, µ, ν), −π < Fµν (x) π. (2.13)
i
Strictly speaking, the cohomological analysis alone admits a more general form of q(x)
than Eq. (2.12); for example, βµν Fµν (x) with anti-symmetric constants βµν is also
possible. However, since A∞ (x) is a pseudoscalar under lattice rotations and reflections,
one infers that it must be proportional to the Levi-Civita symbol. Also the numerical
coefficient in Eq. (2.12) is left undermined in the cohomological analysis. We can, however,
use a matching with the result in the classical continuum limit; the integer N is given by a
sum of chiral charges of massless degrees of freedom [18–24].
Note that Eq. (2.11) is a statement for finite lattice spacings. Eq. (2.11) states that,
even when the lattice spacing is finite, the main part of the axial anomaly is given by the
topological density q(x) which has a quite analogous form to the continuum counterpart.
On the other hand, the total divergence term ∂µ∗ kµ∞ (x) represents “lattice artifacts” in the
4 A field φ(x) is termed local, when its dependence on the gauge field at a point y is exponentially suppressed
as
x − y
→ ∞. For a more precise definition, see Ref. [1].
5 ∂ and ∂ ∗ denote the forward and the backward difference operators, respectively:
µ µ
∂µ f (x) = f (x + µ̂) − f (x), ∗ f (x) = f (x) − f (x − µ̂).

∂µ (2.10)
6 For the cohomological argument to apply, the constant in Eq. (2.4) has to be smaller than 1 and
|Fµν (x)| < π/3 [1].
axial anomaly which depend on the details of the Dirac operator adopted. Our aim in this
paper is to show or argue that the structure represented by Eqs. (2.11) and (2.12) persists
even on finite-size lattices and for general gauge groups G.
3. Abelian gauge theory G = U(1)
For the axial anomaly defined on a finite lattice (2.1), a direct cohomological analysis
is not feasible. Nevertheless, we can show the following
Theorem 3.1. When G = U(1), if the lattice is sufficiently large compared to the
localization range of the Dirac operator, say L/ n,
A(x) = q(x) + ∂µ∗ kµ (x), (3.1)

where kµ (x) is a gauge-invariant periodic current on Γ . The current kµ (x), moreover,
satisfies the bound

kµ (x) − k ∞ (x) κ1 Lν1 e−L/ , (3.2)
µ
with constants κ1 and ν1 .
We emphasize that, for a sufficiently large L, Eq. (3.1) is an exact statement for the axial
anomaly A(x). Eq. (3.2) shows that the current kµ (x) differs from the local current kµ∞ (x)
defined on the infinite lattice only by an exponentially small amount. Hence, when the
lattice size becomes large compared to ρ and thus when the concept of the locality becomes
meaningful, the current kµ (x) can be regarded as a local current. In this way, Eq. (3.1)
shows that the structure of axial anomaly on finite-size lattices is basically identical to that
on the infinite lattice (2.11). The validity of this theorem has been argued intuitively by
Chiu [25].
Proof. The configuration space of the gauge fields allowed by the admissibility (2.4)
consists of many components. Each component is uniquely characterized [15] by the
magnetic flux
1
L−1
mµν = Fµν (x + s µ̂ + t ν̂), (3.3)
2πi
s,t =0
which is an integer. For a configuration with the magnetic flux mµν , from Eq. (2.12), one
has [26]
N (−1)d/2
A∞ (x) = q(x) = µ ν ···µ ν mµ ν mµ ν · · · mµd/2 νd/2
2d/2(d/2)! 1 1 d/2 d/2 1 1 2 2
x∈Γ x∈Γ
= an integer, (3.4)
where the first equality follows from the translational invariance of kµ∞ (x) (namely, kµ∞ (x)
is a periodic current on Γ , when
the gauge field is periodic).7 Combined with the index
theorem (2.3), we see that x∈Γ A(x) − x∈Γ A∞ (x) is an integer. This integer is,
however, bounded by an exponentially small quantity: from the assumed property (2.7),
one infers that

A(x) − A∞ (x) κ2 Lν2 e−L/ . (3.5)

x∈Γ
Therefore, when L is greater than some multiple of , one has

A(x) − A∞ (x) = 0. (3.6)

x∈Γ
For this, we can apply the following
Lemma 3.1. For a periodic field c(x) on Γ satisfying

c(x) = 0, (3.7)
x∈Γ
there exists a periodic current bµ (x) which is given by a sum of c(y), the precise meaning
of which is given in Eq. (3.9) below, such that
∂µ∗ bµ (x) = c(x), |bµ (x)| 2L max |c(x)|. (3.8)

x∈Γ
Applying this lemma to Eq. (3.6), we see that there exists a gauge-invariant periodic
current +kµ (x) such that A(x) − A∞ (x) = ∂µ∗ +kµ (x). This field is exponentially small,
|+kµ (x)| κ1 Lν1 e−L/ , thus kµ (x) = kµ∞ (x) + +kµ (x) which proves the theorem. ✷
The assertions of the Lemma 3.1 immediately follow from the explicit construction
of bµ (x) (though this is not unique):
xµ
1
L−1
L−1
bµ (x) = ··· c(x1, . . . , xµ−1 , yµ , . . . , yd )
Ld−µ
yµ =0 yµ+1 =0 yd =0
xµ + 1
L−1 L−1
− d−µ+1 ··· c(x1, . . . , xµ−1 , yµ , . . . , yd ). (3.9)
L
yµ =0 yd =0
Note that since bµ (x) is given by a sum of the field c(x), bµ (x) is gauge-invariant if so
is c(x).
7 Eq. (2.3) and Theorem 3.1 show that the index is given by the combination (3.4) in terms of the magnetic
flux. For the overlap-Dirac operator, this relation has been verified numerically for d = 2 and d = 4 [27,28] and
proven analytically for d = 2 [28].
4. Non-Abelian cases
For general gauge groups G, a cohomological argument in a non-perturbative level is

not known even on the infinite lattice. Thus we propose a conjecture on a possible form
of A∞ (x):
Conjecture 4.1. For general G,
A∞ (x) = q(x) + ∂µ∗ kµ∞ (x), (4.1)

where kµ∞ (x)is a local gauge-invariant axial vector current (which is translational-
invariant) and the topological density q(x) is given by Lüscher’s topological density [29]
and its higher-dimensional extensions.
The explicit expression of Lüscher’s topological density is known only for d = 2 and for
d = 4. In our context, it is given by N times Eq. (32) of Ref. [29]. We simply assume
that the construction can be pursued for higher-dimensional cases.8 The construction of
Ref. [29] does not provide a pseudoscalar q(x). However, we may always enforce this
pseudoscalar property by taking average over lattice symmetries; we assume that this has
been done and q(x) is a pseudoscalar. The topological density has the classical continuum
limit
1 N i d/2
lim q(x) = µ ν ···µ ν tr Fµ1 ν1 Fµ2 ν2 · · · Fµd/2 νd/2 (x). (4.2)
a→0 a d (4π)d/2(d/2)! 1 1 d/2 d/2
At the moment, we cannot prove the above conjecture in the non-perturbative level.
However, we see that the conjecture holds to all orders in perturbation theory; the
following theorem guarantees that a gauge-invariant topological field is unique (up to a
total divergence) under certain conditions.
Theorem 4.1. Let p(x) be a local gauge-invariant pseudoscalar field (which is translational-
invariant) on the infinite lattice whose dependences on the lattice spacing a arise only
though the gauge field.9 If it is topological

δp(x) = 0, (4.3)
x∈Rd
and the classical continuum limit lima→0 p(x)/a d vanishes, then to all orders in
perturbation theory,
p(x) = ∂µ∗ ,µ (x), (4.4)

where ,µ (x) is a local gauge-invariant axial vector current.
8 For G = (1), the construction of Ref. [29] can be generalized to arbitrary dimensions [26]. The equivalence
U
of Eq. (4.1) with Eq. (2.11) for G = U(1) has been shown [26]. See also Ref. [30].
9 Recall that in the classical continuum limit the gauge potential is introduced as U (x, µ) =

P exp a 01 dt Aµ (x + (1 − t)a µ̂) where a is the lattice spacing.

Proof. Our proof is rather similar to the cohomological argument of Ref. [5]. We expand
p(x) with respect to the bare gauge coupling constant g0 introduced by U (x, µ) =
eg0 Aµ (x) :
∞

p(x) = p(k) (x),
k=1
g0k (k)
p(k) (x) = p (x, y1 , . . . , yk )aµ11···a
···µk Aµ1 (y1 ) · · · Aµk (yk ),
k a1 ak
(4.5)
k! y ,...,y
1 k
where Aµ (x) = Aaµ (x)T a .

First consider p(1) (x). Since p(x) is gauge-invariant, p(1) (x) is invariant under the
linearized gauge transformation
Aµ (x) → Aµ (x) + ∂µ ω(x), (4.6)

and also under the constant gauge transformation

Aµ (x) → Aµ (x) + ω, Aµ (x) . (4.7)
Moreover, since p(1) (x) is a local topological pseudoscalar field and Eq. (4.6) is the gauge
transformation in Abelian theory, one can invoke the cohomological analysis in Abelian
theory. The result is

p(1) (x) = ∂µ∗ ,(1)
µ (x), µ (x) = g0
,(1) ,(1) a a
µ (x, y)ν Aν (y). (4.8)
y
The local axial vector current ,(1)µ (x) is invariant under Eqs. (4.6) and (4.7). A key
(1) (1)
observation is that, from ,µ (x), one can construct a field ,̂µ (x) such that it is invariant
under the original non-Abelian gauge transformation and its lowest-order O(g0 ) term
coincides with ,(1) a
µ (x). This can be accomplished by substituting the gauge potential Aµ (y)
in Eq. (4.8) by the expression [5]
2 a

Âaµ (x, y) = tr T 1 − W (x, y)U (y, µ)W (x, y + µ̂)−1 , (4.9)
g0
where W (x, y) is the ordered product of the link variables from y to x along the shortest
path that goes first in direction 1, then direction 2, and so on. Note that Âaµ (x, y) behaves
gauge covariantly under the original non-Abelian gauge transformation. Thus the resulting
expression,

µ (x) = g0
,̂(1) ,(1) a a
µ (x, y)ν Âν (x, y), (4.10)
y
(1)
is invariant under the non-Abelian gauge transformation due to the invariance of ,µ (x)
under Eq. (4.7). Moreover, since
Âµ (x, y) = Aµ (y) + ∂µy ω(x, y) + O(g0 ), (4.11)

with ω(x, y) the oriented line sum of the gauge potential from y to x, the invariance under
Eq. (4.6) implies that ,̂(1) (1) (1)
µ (x) = ,µ (x) + O(g0 ). Using ,̂µ (x), we may define a local
2 10
gauge-invariant pseudoscalar field

p(x) − ∂µ∗ ,̂(1)
µ (x). (4.12)
which has identical properties with p(x) except that it starts with O(g02 ) term. Thus we
can repeat the above argument from Eq. (4.5) for the field (4.12). This time, however, the
perturbation series analogous to Eq. (4.5) starts from k = 2.
In this way, we repeat the steps from Eq. (4.5) to Eq. (4.12) by eliminating the lowest-
d/2
order term of the topological field until the first order term becomes O(g0 ); here a new
situation arises. The cohomological analysis (with the fact that it is a pseudoscalar) tells
that
ca1 ···ad/2 µ1 ν1 ···µd/2 νd/2 Fµa11 ν1 (x)Fµa22 ν2 (x + µ̂1 + ν̂1 ) · · ·
a ∗ (d/2)
× Fµd/2
d/2
νd/2 (x + µ̂1 + ν̂1 + · · · + µ̂d/2−1 + ν̂d/2−1 ) + ∂µ ,µ , (4.13)
a (x) = ∂ Aa (x) − ∂ Aa (x) denotes the linearized field strength) is a possible form of
(Fµν µ ν ν µ
d/2−1 ∗ (k)
p(x) − k=1 ∂µ ,̂µ (x). However, since the continuum limit of p(x), lima→0 p(x)/a d
vanishes, we infer that the constants ca1 ···ad/2 vanish, ca1 ···ad/2 = 0. Thus we again have
a total divergence. Further, repeating the above procedure, we finally establish p(x) =

∂µ∗ ∞ (k)
k=1 ,̂µ (x). ✷
Going back to Eq. (4.1), we note that both A∞ (x) and q(x) are a local gauge-invariant
topological pseudoscalar field (for the latter, those properties follow from the construction
of q(x) [29]). Moreover, they have the same classical continuum limit (4.2). Thus, applying
Theorem 4.1 to A∞ (x) − q(x), we see that the conjecture holds to all orders in perturbation
theory.
Now, in the proof of Theorem 3.1 in Abelian theory, every stepsare valid even for
non-Abelian theories, except for the crucial relation (3.4), namely, x∈Γ A∞ (x) is an
integer.
With our Conjecture 4.1 for non-Abelian cases, this last condition is also satisfied;
x∈Γ q(x) is Lüscher’s topological charge on a periodic lattice which is an integer. So,
repeating the proof for Theorem 3.1, we have
Corollary of conjecture 4.1. For general G, if the lattice is sufficiently large compared to
the localization range of the Dirac operator, say L/ n,
A(x) = q(x) + ∂µ∗ kµ (x), (4.14)
where kµ (x) is a gauge-invariant periodic current on Γ . The current kµ (x), moreover,
satisfies the bound

kµ (x) − k ∞ (x) κ1 Lν1 e−L/ , (4.15)
µ
(1)
10 The current ,̂ (x) so constructed is not an axial vector under the lattice symmetries. However, we can
µ
always enforce this by taking average over lattice symmetries.
with constants κ1 and ν1 . The topological density q(x) is given by Lüscher’s topological
density [29] and its higher-dimensional extensions.
This corollary states that the basic structure of axial anomaly on finite lattices is identical
that on the infinite lattice. Summing Eq. (4.14) over the lattice Γ , one has an equality
between the index of the Dirac operator (2.3) and the geometrically-defined lattice
topological charge [29]. This equivalence (“lattice index theorem”) has been thought to
be true for long time since the analyses in Refs. [27,31]. Our argument provides a further
support for this equivalence.
5. Conclusion
In this paper, we have studied the axial anomaly defined on a finite-size lattice by using
a Ginsparg–Wilson–Dirac operator. For G = U(1), we show that the basic structure of
axial anomaly on the infinite lattice, which has a quite analogous form to the continuum
counterpart, persists even on a sufficiently large finite-size lattices. For general G, we
conjectured that the axial anomaly on the infinite lattice is basically given by Lüscher’s
topological density; actually this holds to all orders in perturbation theory. With this
conjecture, we showed that this structure again persists even on finite-size lattices. Since
Lüscher’s topological density is a geometrically natural definition of the Chern form in
lattice gauge theory (note that it is proportional to str T a1 · · · T ad/2 ), our analysis indicates
that the basic structure of axial anomaly in continuum theory is quite robust and it persists
even in a system with finite ultraviolet and infrared cutoffs. Of course, we indicated this
persistency only in a framework with the Ginsparg–Wilson relation. To understand precise
conditions on the formulation for this persistency to hold is an interesting open question;
for example, one may enlarge the set of formulations by using the generalized Ginsparg–
Wilson relation [32].
In the gauge-invariant lattice formulation of Abelian chiral gauge theories [15], a
knowledge on the structure of U(1) gauge anomaly on finite-size lattices was of crucial
importance. Recalling this fact, we believe that our analyses will be useful in extending the
construction of Ref. [15] to non-Abelian gauge theories.
Acknowledgements
H.S. would like to thank Takanori Fujiwara, Takahiro Fukui, Yoshio Kikukawa and
Martin Lüscher for valuable discussions. We are grateful to Kazuo Fujikawa for a careful
reading of the manuscript.
References
[1] M. Lüscher, Topology and the axial anomaly in abelian lattice gauge theories, Nucl. Phys. B 538 (1999)
515, hep-lat/9808021.
[2] T. Fujiwara, H. Suzuki, K. Wu, Noncommutative differential calculus and the axial anomaly in abelian lattice
gauge theories, Nucl. Phys. B 569 (2000) 643, hep-lat/9906015;
T. Fujiwara, H. Suzuki, K. Wu, Axial anomaly in lattice abelian gauge theory in arbitrary dimensions, Phys.
Lett. B 463 (1999) 63, hep-lat/9906016.
[3] H. Suzuki, Anomaly cancellation condition in lattice gauge theory, Nucl. Phys. B 585 (2000) 471, hep-
lat/0002009;
H. Igarashi, K. Okuyama, H. Suzuki, Errata and addenda to “Anomaly cancellation condition in lattice gauge
theory”, hep-lat/0012018.
[4] Y. Kikukawa, Y. Nakayama, Gauge anomaly cancellation in SU(2)L × U(1)Y electroweak theory on the
lattice, Nucl. Phys. B 597 (2001) 519, hep-lat/0005015.
[5] M. Lüscher, Lattice regularization of chiral gauge theories to all orders of perturbation theory, J. High
Energy Phys. 06 (2000) 028, hep-lat/0006014.
[6] Y. Kikukawa, Domain wall fermion and chiral gauge theories on the lattice with exact gauge invariance,
Phys. Rev. D 65 (2002) 074504, hep-lat/0105032.
[7] P. Hasenfratz, Prospects for perfect actions, Nucl. Phys. (Proc. Suppl.) 63 (1998) 53, hep-lat/9709110;
P. Hasenfratz, Lattice QCD without tuning, mixing and current renormalization, Nucl. Phys. B 525 (1998)
401, hep-lat/9802007.
[8] H. Neuberger, Exactly massless quarks on the lattice, Phys. Lett. B 417 (1998) 141, hep-lat/9707022;
H. Neuberger, More about exactly massless quarks on the lattice, Phys. Lett. B 427 (1998) 353, hep-
lat/9801031.
[9] P.H. Ginsparg, K.G. Wilson, A remnant of chiral symmetry on the lattice, Phys. Rev. D 25 (1982) 2649.
[10] P. Hasenfratz, V. Laliena, F. Niedermayer, The index theorem in QCD with a finite cut-off, Phys. Lett. B 427
(1998) 125, hep-lat/9801021.
[11] M. Lüscher, Exact chiral symmetry on the lattice and the Ginsparg–Wilson relation, Phys. Lett. B 428 (1998)
342, hep-lat/9802011.
[12] F. Niedermayer, Exact chiral symmetry, topological charge and related topics, Nucl. Phys. (Proc. Suppl.) 73
(1999) 105, hep-lat/9810026.
[13] P. Hernández, K. Jansen, M. Lüscher, Locality properties of Neuberger’s lattice Dirac operator, Nucl. Phys.
B 552 (1999) 363, hep-lat/9808010.
[14] H. Neuberger, Bounds on the Wilson Dirac operator, Phys. Rev. D 61 (2000) 085015, hep-lat/9911004.
[15] M. Lüscher, Abelian chiral gauge theories on the lattice with exact gauge invariance, Nucl. Phys. B 549
(1999) 295, hep-lat/9811032.
[16] I. Horvath, Ginsparg–Wilson relation and ultralocality, Phys. Rev. Lett. 81 (1998) 4063, hep-lat/9808002;
I. Horvath, Ginsparg–Wilson–Lüscher symmetry and ultralocality, Phys. Rev. D 60 (1999) 034510, hep-
lat/9901014.
[17] W. Bietenholz, On the absence of ultralocal Ginsparg–Wilson fermions, hep-lat/9901005.
[18] Y. Kikukawa, A. Yamada, Weak coupling expansion of massless QCD with a Ginsparg–Wilson fermion and
axial U(1) anomaly, Phys. Lett. B 448 (1999) 265, hep-lat/9806013.
[19] K. Fujikawa, A continuum limit of the chiral Jacobian in lattice gauge theory, Nucl. Phys. B 546 (1999) 480,
hep-th/9811235.
[20] D.H. Adams, Axial anomaly and topological charge in lattice gauge theory with overlap-Dirac operator,
Ann. Phys. (N.Y.) 296 (2002) 131, hep-lat/9812003;
D.H. Adams, On the continuum limit of fermionic topological charge in lattice gauge theory, J. Math.
Phys. 42 (2001) 5522, hep-lat/0009026.
[21] H. Suzuki, Simple evaluation of chiral Jacobian with the overlap Dirac operator, Prog. Theor. Phys. 102
(1999) 141, hep-th/9812019.
[22] T.W. Chiu, T.H. Hsieh, Perturbation calculation of the axial anomaly of Ginsparg–Wilson fermion, hep-
lat/9901011.
[23] T. Reisz, H.J. Rothe, The axial anomaly in lattice QED: a universal point of view, Phys. Lett. B 455 (1999)
246, hep-lat/9903003.
[24] M. Frewer, H.J. Rothe, Universality of the axial anomaly in lattice QCD, Phys. Rev. D 63 (2001) 054506,
hep-lat/0004005.
[25] T.W. Chiu, The axial anomaly of Ginsparg–Wilson fermion, Phys. Lett. B 445 (1999) 371, hep-lat/9809013.
[26] T. Fujiwara, H. Suzuki, K. Wu, Topological charge of lattice Abelian gauge theory, Prog. Theor. Phys. 105
(2001) 789, hep-lat/0001029.
[27] R. Narayanan, H. Neuberger, Chiral fermions on the lattice, Phys. Rev. Lett. 71 (1993) 3251, hep-
lat/9308011;
R. Narayanan, H. Neuberger, A construction of lattice chiral gauge theories, Nucl. Phys. B 443 (1995) 305,
hep-th/9411108.
[28] T. Fujiwara, A numerical study of spectral flows of Hermitian Wilson–Dirac operator and the index theorem
in Abelian gauge theories on finite lattices, Prog. Theor. Phys. 107 (2002) 163, hep-lat/0012007;
H. Kurokawa, T. Fujiwara, Spectrum of the hermitian Wilson–Dirac operator for a uniform magnetic field
in two dimensions, hep-lat/0206014.
[29] M. Lüscher, Topology of lattice gauge fields, Commun. Math. Phys. 85 (1982) 39.
[30] A. Phillips, Characteristic numbers of U(1) valued lattice gauge fields, Ann. Phys. (N.Y.) 161 (1985) 399.
[31] R. Narayanan, P. Vranas, A numerical test of the continuum index theorem on the lattice, Nucl. Phys. B 506
(1997) 373, hep-lat/9702005.
[32] K. Fujikawa, Algebraic generalization of the Ginsparg–Wilson relation, Nucl. Phys. B 589 (2000) 487, hep-
lat/0004012;
K. Fujikawa, M. Ishibashi, Chiral anomaly for a new class of lattice Dirac operators, Nucl. Phys. B 587
(2000) 419, hep-lat/0005003;
K. Fujikawa, M. Ishibashi, Locality properties of a new class of lattice Dirac operators, Nucl. Phys. B 605
(2001) 365, hep-lat/0102012;
K. Fujikawa, M. Ishibashi, A perturbative study of a general class of lattice Dirac operators, Phys. Rev. D 65
(2002) 114504, hep-lat/0201016.
Testable (g − 2)µ contribution due to a light

stabilized radion in the Randall–Sundrum model
Prasanta Das, Uma Mahanta
Mehta Research Institute, Chhatnag Road, Jhusi Allahabad 211019, India
Received 4 April 2002; received in revised form 9 July 2002; accepted 13 August 2002
Abstract
In this paper we calculate the (g − 2)µ contribution due to a light stabilized radion using the radion
couplings both to the kinetic energy and the mass of the muon. We find that the radion mediated muon
anomaly (ar ) is a calculable quantity free from powerlike ultraviolet divergences. We have estimated
ar both for Λ mφ mµ and Λ mφ ≈ mµ . Our results show that under the first (second)
condition the radion mediated muon anomaly can be detected with the ultimate future precision for
measuring the muon anomaly provided φ is less than 425 (600) GeV. Whereas with the present
precision ar can be detected provided φ is less than 250 GeV in the first case and 375 GeV in the
second case.
Recently there has been a lot of interest in studying the phenomenology of models of
large [1] and small [2] extra dimensions. Phenomenological data from a variety of sources
have been used to constrain the unknown (free) parameters of models of extra dimensions.
In particular, the precision measurement of muon anomaly by the BNL Collaboration has
been used to constrain the free parameters of extra dimension models [3]. In this paper we
calculate the muon anomaly due to a light stabilized radion [4] in the Randall–Sundrum
model. Using the radion couplings both to the kinetic energy (K.E.) and the mass of the
muon we find that the radion mediated muon anomaly is a calculable quantity free from
powerlike ultraviolet divergences. This is unlike the Kaluza–Klein graviton contribution to
the oblique electroweak parameters S, T and U which is plagued by uncalculable powerlike
divergences [5]. We also find that the radion coupling to the K.E. of the muon leads to a
muon anomaly contribution that could be tested with the present (future) experimental
precision even for mφ mµ provided φ is less than 250 (425) GeV. Our result is
E-mail addresses: pdas@mri.ernet.in (P. Das), mahanta@mri.ernet.in (U. Mahanta).
PII: S 0 5 5 0 - 3 2 1 3 ( 0 2 ) 0 0 7 4 3 - 5
396 P. Das, U. Mahanta / Nuclear Physics B 644 (2002) 395–400
(a) (b)
(c) (d)
Fig. 1. Feynman diagrams that gives rise to the radion contribution to muon anomaly.
significantly greater than previous estimates which have reported very small values of
radion mediated muon anomaly for mφ mµ . The reason the previous estimates had
obtained very small values of ar is that they used only the radion coupling to the mass
of the muon neglecting its coupling to the K.E. of the muon.
The Feynman diagrams that give could give rise to radion mediated muon anomaly are
shown in Fig. 1.
The Feynman rules for evaluating these diagrams can be found in Ref. [6]. The results
of our calculation are briefly presented below.

9e d 4l
−ieΓ1µ = x dx dy /l − p
/ (1 − x) − 23 + xy mµ
2φ2 (2π)4

× /l + p/ (1 − xy) − p
/ (1 − x) + mµ γµ [/l − p / xy + p
/ x + mµ ]
1
× /l − p/ xy − 53 − x mµ
[(l + p) − mµ ][(l + p )2 − m2µ ](l 2 − m2φ )
2 2
9iemµ
=
64π 2φ2

Λ2 3
× x dx dy ln 2 − C1 (x, y)(pµ + pµ ) + C2 (x, y)mµ γµ
A 2
P. Das, U. Mahanta / Nuclear Physics B 644 (2002) 395–400 397

3iem3µ 1
− x dx dy D1 (x, y)(pµ + pµ ) + D2 (x, y)mµ γµ , (1)
32π 2φ2 A 2

3e d 4 l [ 32 (/l + 2/p ) − 4mµ ][/l + p
/ + mµ ]
−ieΓ2µ = − γµ
φ2 (2π)4 [(l + p )2 − m2µ ][l 2 − m2φ ]
1
−3ie 3 2 3 2 Λ2
= dx Λ + B 1 − 2 ln 2
16π 2φ2 2 2 B
0
3 Λ2
+ 1 + x (2 − x)m2µ ln 2 − 1 γµ , (2)
2 B

3e d 4l [/l + p
/ + mµ ][ 32 (/l + 2/
p) − 4mµ ]
−ieΓ3µ = − 2 γ µ
φ (2π) 4 [(l + p) − mµ ][l − m2φ ]
2 2 2
1
−3ie 3 2 3 2 Λ2
= dx Λ + B 1 − 2 ln 2
16π 2φ2 2 2 B
0
3 Λ2
+ 1 + x (2 − x)mµ ln 2 − 1 γµ
2
(3)
2 B
and

3e d 4l 1 3ie Λ2
−ieΓ4µ = γµ = − Λ − mφ ln 2 γµ .
2 2
(4)
φ2 (2π)4 (l 2 − m2φ ) 16φ2 mφ
In the above
A2 = (1 − x + xy)2m2µ + x(1 − y)m2φ , (5)

B 2 = x 2 m2µ + (1 − x)m2φ , (6)
2
C1 (x, y) = 2xy − 2x + , (7)
3
76
C2 (x, y) = 6x 2 y 2 − 12x 2 y + 6x 2 − , (8)

9
2
D1 (x, y) = xy x − 3 + xy 3 + xy − 2x
2 7

+ (1 − x) x 2 y(y − 2) − (1 + x) 53 − x + 43 xy(1 − x), (9)

D2 (x, y) = 32 x 2 − 73 + xy 23 + xy − 2x x 2 y(y − 2) − (1 + x) 53 − x

− (1 − x) x 2 y(y − 2) − (1 + x) 53 − x

− 2xy(1 − x) − xy x 2 − 73 + xy 23 + xy − 2x , (10)
Ci (x, y) and Di (x, y) are therefore dimensionless functions of x and y. Λ is an ultraviolet

momentum cutoff (regulator) beyond which the radion + SM system can no longer be
considered in isolation and new physics (Kaluza–Klein excitations of the graviton) effects
become important. In the Randall–Sundrum model Λ can be identified with the mass of
lightest Kaluza–Klein graviton. According to naive-dimensional analysis estimates [7] the

cut-off Λ should be equal to 4πφ since φ plays here the role of the expansion parameter
similar to fπ in the theory of chiral Lagrangians. Further we have replaced p p ) in the
/ (/
numerator of each vertex function by mµ when it stands to the extreme right (left). Note
that Figs. 2–4 are just proportional to a vector current γµ . Hence they do not contribute
to the muon anomaly. The radion mediated muon anomaly therefore arises only from the
first diagram. After pulling out one external momentum (pµ or pµ ) out of the integral and
Fig. 2. ρ parameter constraints on radion VEV φ and radion mass mφ . The allowed region lies above the curve.
Fig. 3. Plot of muon anomaly ar against the radion VEV in the first case. The horizontal line corresponds to the
ultimate future precision of the experiment.
P. Das, U. Mahanta / Nuclear Physics B 644 (2002) 395–400 399
Fig. 4. Plot of muon anomaly ar against the radion VEV in the second case. The horizontal line corresponds to
the ultimate future precision of the experiment.
applying power counting it can be shown that the anomalous magnetic moment term is at
most log divergent. Using the Gordon identity to replace the convective current (pµ + pµ )
by a vector current (γµ ) and a spin current (iσµν q ν ) we get

9m2µ Λ2 3
ar = x dx dy ln 2 − C1 (x, y)
32π 2 φ2 A 2
2 2
3mµ mµ
− x dx dy 2 D1 (x, y). (11)
16π φ
2 2 A
We shall now estimate the radion mediated muon anomaly under the following two
conditions:
Case I: Λ mφ mµ
In this case we shall consider radion masses in the range 1 GeV to a few hundred GeV.
The reason being such radion masses do not require a large fine tuning of the parameters
in the Golberger and Wise stabilization scheme. Under this condition m2µ /A2 1 and the
second term becomes negligible in comparision to the first term (logarithmic) term. We find
that for φ ≈ 250 GeV the radion mediated muon anomaly turns out to be 1.24 × 10−9 .
On the other hand for φ ≈ 500 GeV the value of ar drops to 3.2 × 10−10 .
Case II: Λ mφ ≈ mµ
Under this condition the first term becomes comparable to the second and both have
to be kept in estimating ar . We find that for φ ≈ 375 GeV the radion mediated muon
anomaly is about 1.1 ×10−9 . On the other hand for φ ≈ 600 GeV the value of ar falls to
4 × 10−10 .
The precision measurement of muon anomaly currently constitutes one of the most
stringent tests for new physics beyond the standard model (SM), particularly in the light
of the recent and future promised results from E821 experiment at BNL. The most recent
experimental value of the muon anomaly is aexp = (11659203 ± 15) × 10−10 [8]. The
µ
SM prediction is aSM = (11659176 ± 6.7) × 10−10 [9]. The SM prediction for the muon
µ
anomaly therefore differs from the experimental value by (27 ± 16.4) × 10−10 , i.e., by 1.6
standard deviation. Since this is less than a 2σ effect we shall not use it to set limits on the
unknown parameters mφ and φ. Rather we shall use the precision of the BNL experiment
as a benchmark for testability of radion mediated muon anomaly. It is hoped that the BNL
Collaboration will be able to reach a precision of 4 × 10−10 . Our results show that in the
first case the radion mediated muon anomaly can be detected with the present (future)
precision for measuring the anomaly provided φ is less than 250 (425) GeV. Whereas in
the second case the detectability of ar with the present (furture) precision requires φ to
be less than 375 (600) GeV. It is worthwhile to compare these bounds with those obtained
from precision measurement of the oblique electroweak parameters. At 2σ difference from
the central value of the T parameter one obtains a lower bound of 400 GeV on φ for
radion masses lying between 1 and 100 GeV [10] The muon anomaly bounds on φ are
therefore comparable with those obtained from the T parameter.
We would like to note that in the first case (mφ mµ ) our result for ar is significantly
greater than those of previously published estimates. Previous estimates of radion mediated
muon anomaly had neglected the radion coupling to the KE of the muon and used only
the radion coupling to the mass of the muon. As a result they got only a subdominant
contribution proportional to m2µ × m2µ /m2φ . It can be shown that the radion coupling to the
muon reduces to the mass of the muon only if both muon lines are on shell. However, in
calculating the loop diagrams shown in Fig. 1 one certainly cannot assume that the muon
lines at each vertex are on shell. The previous estimates of radion mediated muon anomaly
are therefore not trustable.
References
[1] N. Arkani-Hamed, S. Dimopoulos, G. Dvali, Phys. Lett. B 429 (1998) 263;

I. Antoniadis, N. Arkani-Hamed, G. Dvali, Phys. Lett. B 463 (1998) 257.
[3] M.L. Graesser, hep-ph/9902310;
U. Mahanta, S. Rakshit, Phys. Lett. B 480 (2000) 176;
S.C. Park, H.S. Song, Phys. Lett. B 506 (2001) 99;
C.S. Kim, J.D. Kim, J. Song, Phys. Lett. B 511 (2001) 251.
[4] W.D. Goldberger, M.B. Wise, Phys. Rev. Lett. 83 (1999) 4922.
[5] P. Das, S. Raychaudhuri, hep-ph/9908205;
T. Han, D. Marfatia, R. Zhang, Phys. Rev. D 62 (2000) 125018.
[6] P. Das, U. Mahanta, hep-ph/0110309.
[7] H. Georgi, A. Manohar, Nucl. Phys. B 234 (1984) 189;
Z. Chacko, M. Luty, E. Ponton, JHEP 07 (2000) 036.
[8] H.N. Brown, et al., Phys. Rev. Lett. 86 (2001) 2227.
[9] D.H. Hertzog, hep-ex/0202024.
[10] C. Csaki, M. Graesser, G.D. Kribs, Phys. Rev. D 63 (2001) 065002;
P. Das, U. Mahanta, hep-ph/0107162.
Erratum
Erratum to: “A complete calculation of the order αs2

correction to the Drell–Yan K-factor”
[Nucl. Phys. B 359 (1991) 343] ✩
R. Hamberg a , W.L. van Neerven a , T. Matsuura b
a Instituut-Lorentz, University of Leiden, PO Box 9506, 2300 RA Leiden, The Netherlands
b II Institut für Theoretische Physik, Universität Hamburg, D-2000 Hamburg 50, Germany
Received 30 August 2002
There is an error in the integrated cross-sections due to the one-loop corrections to the
reactions q + q̄ → V + g and q(q̄) + g → V + q(q̄). Therefore the coefficient functions
(2),C (2),C (2),C (2),C
∆q q̄ A Eq. (B.9), ∆q q̄ F Eq. (B.10), ∆q q̄ A Eq. (B.19), ∆q q̄ F Eq. (B.20), are changed.
The correct coefficient functions are obtained by the following modifications.
To Eq. (B.9) one has to add
2
αs
CA CF −16x ln x ln(1 − x) + 8x ln2 x − 16x Li2 (1 − x) .
4π
2
αs
CF2 −(48 + 16x) ln x ln(1 − x) − 16 ln x
4π

+ (24 + 8x) ln2 x − (48 + 16x) Li2 (1 − x) .
2
αs
CA Tf 8x ln x ln(1 − x) − 8x ln x − 4x ln2 x + 8x Li2 (1 − x) .
4π
✩
PII of original article: S0550-3213(91)00343-8.
E-mail address: neerven@lorentz.leidenuniv.nl (W.L. van Neerven).
PII: S 0 5 5 0 - 3 2 1 3 ( 0 2 ) 0 0 8 1 4 - 3
404 R. Hamberg et al. / Nuclear Physics B 644 (2002) 403–404
2
αs
CF Tf −(24 − 8x) ln x ln(1 − x) − 24(1 − x) ln(1 − x)
4π
+ (28 − 44x) ln x + (12 − 4x) ln2 x

− (24 − 8x) Li2 (1 − x) + 12(1 − x) .
The new coefficient functions are now in agreement with the results found in [1]. Further
there are some misprints. The function Li3 ( 1+x
1−x ) on line 8 in Eq. (B.19) and on line 4
in Eq. (B.24) has to be read as Li3 ( 1−x
1+x ). Likewise the function Li3 (− 1+x
1−x ) on line 8 in
1−x
Eq. (B.19) and on line 4 in Eq. (B.24) has to be read as Li3 (− 1+x ).
Acknowledgement
W.L. van Neerven would like to thank W.B. Kilgore for discussions. We also would like
to thank V. Ravindran in recalculating the coefficient functions which led to the agreement
with the results in [1].
References
[1] R.V. Harlander, W.B. Kilgore, Phys. Rev. Lett. 88 (2002) 201801.
Erratum
Erratum to: “Harmonic superpotentials and

symmetries in gauge theories with eight
supercharges”
[Nucl. Phys. B 554 (1999) 365] ✩
Boris Zupnik
Bogoliubov Laboratory of Theoretical Physics, Joint Institute for Nuclear Research, Dubna,
Moscow region, 141980, Russia
Received 18 September 2002
In connection with the field-component description of the 5D superconformal non-

Abelian action [1,2], our harmonic-superspace formulation of this model should be
corrected. The correct version of the last part of Section 2 preserves the original numeration
of the corresponding formulas and references:
It is not difficult to construct the non-Abelian analog of the 5D Chern–Simons action
(2.25) starting from the following variational formula:

5
δSCS = k5 d 5 x d 8 θ du Tr δV ++ V −− , D (+2) V −− , (2.26)
which guarantees the gauge invariance taking into account Eqs. (1.7), (1.11), (2.20)
and (2.21)

5
δλ SCS = k5 dζ (−4) du Tr λD (+4) D (+2) V −− , ∇ ++ V −− = 0. (2.27)
Stress that the possible term with [V −− , D (+2) V −− ] in δSCS

5 vanishes as an integral of the
total spinor derivative.

The scale-invariant non-Abelian action SCS 5 has the following form:

k5
5
SCS = d 5 x d 8 θ du Tr V ++ V1−− V ++ , D (+2) V1−− V ++ + · · · , (2.28)
3
✩
PII of original article: S0550-3213(99)00267-9.
E-mail address: zupnik@thsun1.jinr.ru (B. Zupnik).
PII: S 0 5 5 0 - 3 2 1 3 ( 0 2 ) 0 0 8 4 0 - 4
406 B. Zupnik / Nuclear Physics B 644 (2002) 405–406
where higher-order terms are omitted and the linear approximation of the perturbative
solution for V −− is used

V ++ (z, u1 )
V1−− V ++ = du1 . (2.29)
(u+ u+
1)
2
References
[1] T. Kugo, K. Ohashi, Prog. Theor. Phys. 105 (2001) 323, hep-ph/0010288.
[2] E. Bergshoeff, et al., hep-th/0205230.
Nuclear Physics B 644 [FS] (2002) 409–432
Quasi-classical descendants of disordered vertex

models with boundaries
Antonio Di Lorenzo a , Luigi Amico a , Kazuhiro Hikami b
Andreas Osterloh a , Gaetano Giaquinta a
a NEST-INFM & Dipartimento di Metodologie Fisiche e Chimiche (DMFCI), Università di Catania,
viale A. Doria 6, I-95125 Catania, Italy
b Department of Physics, Graduate School of Science, University of Tokyo, Hongo 7-3-1, Bunkyo,
Tokyo 113-0033, Japan
Received 27 June 2002; received in revised form 26 August 2002; accepted 5 September 2002
Abstract
We study descendants of inhomogeneous vertex models with boundary reflections when the spin–
spin scattering is assumed to be quasi-classical. This corresponds to consider certain power expansion
of the boundary-Yang–Baxter equation (or reflection equation). As final product, integrable su(2)-
spin chains interacting with a long range with XXZ anisotropy are obtained. The spin–spin coupling
constants are non-uniform, and a non-uniform tunable external magnetic field is applied; the latter
can be obtained when the boundary conditions are assumed to be quasi-classical as well. The exact
spectrum is achieved by algebraic Bethe ansatz. Having realized the su(2) operators in terms of
fermions, the class of models we found turns out to describe confined fermions with pairing force
interactions. The class of models presented in this paper is a one-parameter extension of certain
Hamiltonians constructed previously. Extensions to su(n)-spin open chains are discussed.
PACS: 02.30.Ik; 75.10.Jm
1. Introduction
Integrable vertex models (VM) in two-dimensional classical statistical mechanics are

the common seed of many relevant exactly-solved quantum models in one dimension
[1,2]. Famous examples are the XXX, XXZ, and XY Z Heisenberg chains that find more
E-mail address: lamico@dmfci.unict.it (L. Amico).
PII: S 0 5 5 0 - 3 2 1 3 ( 0 2 ) 0 0 8 1 1 - 8
410 A. Di Lorenzo et al. / Nuclear Physics B 644 [FS] (2002) 409–432
and more applications in contemporary physics. The key toward this powerful synthesis
is to notice that the “scattering” of the degrees of freedom of both the VM and the spin
chains is described by the same matrix. The Quantum Inverse Scattering Method (QISM)
exploits this fact systematically [3]. The method relies on the observation that transfer
matrices tˆ = Tr(T ) span a one-parameter family of commuting operators if a (scattering)
matrix R exists such that T , R satisfy the celebrated Quantum Yang–Baxter equation. The
equivalence between VM and Heisenberg chains consists in the fact that these models have
the same R-matrix. Due to the property of the scattering R(u, v) = R(u − v), ∀u, v ∈ C
the integrability of VM is preserved if disorder is added at each lattice site such that the
scattering “wave momenta” u, v result to be shifted arbitrarily. In this case, however, it
is difficult to extract a Hamiltonian. A route to simplify the problem is to resort the so-
called “quasi-classical” limit of the QISM. The term “quasi-classical” here indicates that
the scattering between the degrees of freedom of the model is assumed to be quasi-classical.
Quantitatively, this means that a parameter η does exist such that the scattering matrix is
of the form R(u) ∝ 1 ⊗ 1 + ηr(u) in the limit η → 0 (η plays the role of h̄). The quantity
r(u) fulfills the classical Yang–Baxter equation (that is a restatement of the Jacobi identity
for the Poisson brackets of suitable action-angle variables). It is worthwhile to mention,
however, that the systems obtained by this quasi-classical expansion consist of quantum
spins (by no means quasi-classical). The quasi-classical expansion of the transfer matrix
of disordered VM (in the lowest spin representation) is non-trivial and it produces the
Gaudin’s magnet Hamiltonians [4,5] containing a long range spin interaction (in contrast
with the range of the Heisenberg chains which involves nearest neighbour spins).
A richer variety of integrable models by QISM comes from imposing non-trivial
boundary conditions different from the periodic ones. Twisted boundary conditions, for
example, imposed to the six vertex model [6] produce the Gaudin magnet in a non-uniform
local magnetic field, which is very important for physical applications. In fact having
realized the (pseudo)spin algebra in terms of fermions the XXX Gaudin Hamiltonians
in a non-uniform magnetic field are the constants of the motion of the BCS model [7] that
describes pairs of electrons (in time reversed states) interacting with a long range uniform
pairing coupling. The exact solution of the BCS model was found long ago by Richardson
[8] and rediscovered recently. In particular it was used to study small metallic grains
[9,10]; the picture was merged in the scenario of QISM in Ref. [11]. Connections with
WZNW models in field theory have been deeply investigated [12] based on the relation
between solution of KZ equation and Gaudin model found in Refs. [13,14]. The class
of pairing Hamiltonians was generalized by investigating the quasi-classical expansion
of the disordered twisted six vertex model with XXZ R-matrix. In terms of fermions
this class of Hamiltonians represents interacting electron pairs with certain non-uniform
long-range coupling strengths [15–18]. Twisted rings can be cut to open chains and loops
include two reflections at the boundaries. The possibility to include such reflections in
integrable theory was founded and systematically investigated by Sklyanin [19]. The quasi-
classical limit of the disordered six vertex model with boundaries was investigated first
by one of the authors [20,21]. This led to a model where the spin couplings contain
an additional parameter with respect to the original Gaudin magnet, and in a vanishing
external magnetic field (see Eq. (27) and the relative discussion below of it). In the present
work, we proceed along this line. We still consider an inhomogeneous six-vertex model
A. Di Lorenzo et al. / Nuclear Physics B 644 [FS] (2002) 409–432 411
with boundary reflection, following closely Ref. [20]. By properly choosing the reflection
parameters, we introduce an external non-uniform magnetic field of tunable strength in the
Hamiltonian. The trick consists in the assumption that also the boundary conditions have
a quasi-classical expansion (see Eq. (25) and Section 3.2). At best of our knowledge, this
idea is pursued for the first time in the present paper. In the following we summarize the
main results obtained in the bulk of the paper. The class of spin-Sj (j = 1, . . . , N ) models
that we find has Hamiltonian of the form
(z)
H= 2hj Sjz − Jj k
Sjz
Skz + Jj k
Sj+
Sk− + h.c. . (1)
j j,k
j =k
We agree that the Latin indices j, k will run from 1 to N , where N is the number of spins.
The operators S α , α = ±, z are su(2) operators. The couplings are
(z)
Jj k = Ij k cosh(2pzj ) + cosh(2pzk ) − 2 cos (2pt) ,

Jj k = Ij k sinh p(zj − it) sinh p(zk + it) ,
z hj − hk
Ij k = J S ,
cosh(2pzj ) − cosh(2pzk )
z −1
J S = J 1 − J Sz , (2)
where S z is the total z-component of the spin. The quantities hj , zj are two arbitrary
sets of real parameters; t is also a real arbitrary parameter and it directly comes from the
boundary terms (see Eq. (16) with ξ = it); finally, p can be 1, i or can be tending to zero
corresponding to hyperbolic, trigonometric,and rational couplings, respectively.
The eigenstates in the sector with S z = j Sj − M, are

M
|Ψ =
S − (eα )|H , (3)
α=1
where
cosh[p(u + zj + 2it)] − cosh[p(u − zj )]

S − (u) =
Sj− (4)
cosh(2pu) − cosh(2pzj )
j
and

N

z
|H =
S = Sj .
j
j =1
The corresponding eigenvalues are

E= 2hj τi , (5)
j
where

z 1 − xj xk z 1 − xj λα
τj = Sj 1 − J S Sk +J S , (6)
xj − xk xj − λα
k=j α
where we defined
1 + xj
= cosh(2pzj ) − cos(2pt),
1 − xj
1 + λα
= cosh(2peα ) − cos(2pt). (7)
1 − λα
The rapidities λα satisfy, in the sector having total spin S z , the Bethe equations:
1 Sj
−
λα − λβ λα − xj
β=α j

1 1 + J (S z )(1 + S z ) 1 − J (S z )(1 + S z )
+ + = 0. (8)
2J (S z ) 1 + λα 1 − λα
We point out that the dependence on the reflection parameters comes only in the coupling
and eigenvectors (Eqs. (2) and (3)), while the eigenvalues depend on t only implicitely
(through Eq. (7)).
The rational limit of the models is recovered for p → 0.
For t = 0 and pt = π/2, the models reduce to the ones that we presented in Ref. [15]
(see Section 3.3). Thus the class of models we discuss in the present paper is a one-
parameter extension of the former class.
Using the fermionic realizations of su(2) the Hamiltonian (1) can be rephrased to
describe confined fermions interacting with pairing and exchange forces (see Eq. (48)).
The paper is organized as follows. In the next section we summarize the main
ingredients of the inverse scattering of VM with boundaries. In Section 3 we construct the
integrable models we deal with together with their exact solution. In Section 4 we use the
fermionic realization of the su(2) algebra to rewrite the Hamiltonians in a second quantized
form. Section 5 is devoted to final remarks. In Appendix A we review basic properties of
VM. In Appendix B we prove the integrability of a class of models when a more general
(off-diagonal) reflection at the boundary is applied (see Eqs. (B.1)–(B.4)). We also discuss
a generalization to su(n) case in Appendix C.
2. Integrable boundary conditions
In this section, we review how the QISM is applied to VM, in order to obtain a family
of commuting transfer matrices. VM describe a system of interacting classical objects on a
two-dimensional lattice. As described in Appendix A, the partition function of the system
can be written as Z = Tr{tˆ(1) · · · tˆ(K)}, where tˆ(i) are operators in some appropriate
many-body linear space (in the sense that it is the direct product of N elementary linear
spaces). The VM is exactly solvable if [tˆ(i), tˆ(i )] = 0. Usually, it is assumed that the
dependence on the ith row of the lattice comes through a parameter ui , which takes values
on some domain of the complex plane. Then the requirement for exact solvability becomes
[tˆ(u), tˆ(v)] = 0, ∀u, v belonging to the domain.
The QISM provides a way of constructing classes of commuting operators tˆ(u), finding
their eigenvalues and their common eigenstates, and extracting Hamiltonians whose
integrals of motions are tˆ(u). The QISM is a procedure which starts from the R-matrix
and from the Lax operator to yield the transfer matrix tˆ(u). From the transfer matrix, a
class of Hamiltonians can be extracted in various ways, to be depicted below. The QISM
has a built-in Algebraic Bethe Ansatz (ABA) which provides the diagonalization of the
tˆ(u), and hence of the Hamiltonian.
The XXZ R-matrix is
 
a(u, v) 0 0 0
 0 b(u, v) c(u, v) 0 
R(u, v) = 
 0
, (9)
c(u, v) b(u, v) 0 
0 0 0 a(u, v)
where
sinh [p(u − v + η)] sinh [p(−v)]
a(u, v) = , b(u, v) = ,
p p
sinh (pη)
c(u, v) = .
p
It is connected to the Ř-matrix defined for VM by R(u, v) = P12 Ř(1, 2), where
 
1 0 0 0
0 0 1 0
P12 = 
0 1 0 0

0 0 0 1
is the permutation operator, and it is assumed that the dependence of R upon the rows
comes through a parameter assigned to each row.
The corresponding Lax operators are

1 sinh p u + η Sjz Sj−
sinh(pη)
Lj (u) = . (10)
p sinh(pη)Sj+ sinh p u − η
Sjz
Here p is the anisotropy parameter, in the sense that, when p = 0—in which case one can
put either p = 1 or p = i—the QISM yields a Hamiltonian with XXZ-type couplings,
while in the limit p → 0, the hyperbolic/trigonometric functions reduce to rational ones,
and the QISM generates a Hamiltonian having XXX couplings; η, instead, is the so-called
quantum parameter which plays the role of h̄; as we shall later see, it gives the degree of
deformation of the classical algebra su(2) into the quantum algebra suq (2). We remark that
the terminology is somehow misleading: since we associate the algebra su(2) with spins,
realized either by true spins or by pairs of time-reversed electrons, in the limit η → 0 we
obtain genuine quantum Hamiltonians.
The Lax operators act on the auxiliary two-dimensional vector space V, and on the
quantum space Hn . They obey the fundamental Yang–Baxter relation Eq. (A.1), which in
terms of the R-matrix now reads
1 2 2 1
R(u − v)Lj (u − zj )Lj (v − zj ) = Lj (v − zj )Lj (u − zj )R(u − v). (11)
Due to the additive property of the R-matrix R(u, v) = R(u − v), parameters zj taking
into account on-site disorder through the lattice can be introduced.
1 2
As customary Lj (u) = Lj (u) ⊗ 1, and Lj (u) = 1 ⊗ Lj (u); the external product is
meant between two copies of the space V, while the multiplication of the elements of Lj ,
which are operators on Hj , is an internal product. The relation (11) is actually obeyed only
for 1/2 spins, i.e., for dim(Hj ) = 2. The order of the representation (that is the dimension
of Hj ) can be extended to larger values, keeping the dimension of V fixed to 2; however,
one has to renounce to the algebra su(2), and introduce rather the quantum algebra suq (2),
which ensures that the relation (11) is obeyed whatever is the representation of the algebra.
The parameter q is related to the parameters p and η by q = exp(pη). The commutation
rules are
sinh(2pη
Sjz )

Sjz ,
Sj± =
Sj± ,
Sj+ ,
Sj− = . (12)
sinh(pη)
In the quasi-classical limit η → 0, or in the isotropic limit p → 0, suq (2) reduces to su(2).
Next, we consider the monodromy matrix T (u) ≡ L1 (u − z1 ) · · · LN (u − zN ). We have
an internal product over V and an external one over Hj and Hj ; thus T (u) is an operator
over V ⊗ H1 ⊗ · · · ⊗ HN . It has the form

A(u) B(u)
T (u) = ,
C(u) D(u)

with A, B, C, D operators over H = j Hj .
j (u), Lk (v)] = 0 for j = k,
The local relation (11), and the ultra-locality property, [Lab cd
imply that T (u) fulfills the global Yang–Baxter equation

1 2 2 1
R(u − v)T (u)T (v) = T (v)T (u)R(u − v). (13)
In the case of periodic boundary conditions (on the auxiliary matrix space), the quantities
Tr{T (u)} = A(u) + D(u), where the trace is on the auxiliary space V, generate a one
parameter commuting family of operators which underlies an integrable model.
Remarkably, the property of integrability is preserved for a wider class of non-trivial
boundary conditions. Integrable boundary conditions are introduced by the so-called
“boundary K-matrices” satisfying the reflection equations [19,22]
1 2 2 1
R(u − v)K − (u)R(u + v)K − (v) = K − (v)R(u + v)K − (u)R(u − v), (14)
1 2
R(−u + v)K t+ (u)R(−u − v − 2η)K t+ (v)
2 1
= K t+ (v)R(−u − v − 2η)K t+ (u)R(−u + v). (15)
Among the solutions of the reflection equations, we consider the diagonal ones, which
yield Hamiltonians preserving the total spin
S z . They depend upon the free parameters ξ±
K− (u) = K(u, ξ− ),
K+ (u) = K(u + η, ξ+ ),
where

1 sinh p(u + ξ ) 0
K(u, ξ ) = . (16)
p 0 − sinh p(u − ξ )
The family of commuting transfer matrices is

tˆ(u) = Tr K+ (u)U (u) , (17)
where

−1
A(u) B(u)
U (u) = T (u)K− (u)T (−u) = .
C(u) D(u)
The inverse of the monodromy matrix is defined as [3]
T −1 (−u) = σ y T t (−u + η)σ y det−1

q T (−u + η/2),
where σ y is the Pauli matrix in the representation where σ z is diagonal, and detq T (u) =
A(u − η/2)D(u + η/2) − C(u − η/2)B(u + η/2) is the quantum determinant, which is a
su(2)-number.
The eigenvectors of tˆ(u), in the sector with S z = j Sj − M, are given by [19]

M
B(eα )|H ,
α=1
where

|H =
S z = Sj
j
j
is the pseudo-vacuum state having all maximum Sjz eigenvalues. The eigenvalues are
sinh[2p(u + η)]
t (u) = cosh 2p(u + σ ) − cosh(2pδ)
sinh[p(2u + η)]
M
sinh[p(u − eα − η)] sinh[p(u + eα )]
× a(u)d(−u + η)
sinh[p(u + eα + η)] sinh[p(u − eα )]
α=1
sinh(2pu)
− cosh 2p(u − σ + η) − cosh(2pδ)
sinh[p(2u + η)]
M
sinh[p(u − eα + η)] sinh[p(u + eα + 2η)]
× a(−u + η)d(u), (18)
sinh[p(u + eα + η)] sinh[p(u − eα )]
α=1

where we put ξ+ + ξ− = 2σ , ξ+ − ξ− = 2δ and a(u) = N j =1 sinh[p(u − zj + ηSj )]/p,
N
d(u) = j =1 sinh[p(u − zj − ηSj )]/p. The eα satisfy the Bethe equations:
cosh[2p(eα + σ )] − cosh(2pδ)
cosh[2p(eα − σ + η)] − cosh(2pδ)
sinh[p(eα − eβ − η)] sinh[p(eα + eβ )]
×
sinh[p(eα − eβ + η)] sinh[p(eα + eβ + 2η)]
β=α
sinh[p(eα − zj − ηSj )] sinh[p(eα + zj − η(Sj + 1))]
= . (19)
sinh[p(eα − zj + ηSj )] sinh[p(eα + zj + η(Sj − 1))]
j
The final step to obtain integrable models from the procedure above is to observe that
transfer matrices can be used as generating functional of Hamiltonians. A possibility is

H≡ ln tˆ(u)
. (20)
∂u u=uc
In the homogeneous case zj = 0, ∀j , Heisenberg Hamiltonians with nearest-neighbour

interaction are obtained for uc = 0. The presence of disorder makes the application of
Eq. (20) quite difficult. A particular value of uc for which the calculations can be done is
uc → ∞; in this case the interaction in the Hamiltonian is long range [23]. Another way to
face the problem is to resort to the quasi-classical expansion. The trick consists in obtaining
a set of commuting operators as coefficients of the power-η expansion of tˆ(u)—from which
a Hamiltonian turns out can be built as a polynomial. The quasi-classical expansion of the
R-matrix and Lax operators gives
sinh(pu)
Lj (u) = 1 + ηlj(1)(u) + η2 lj(2) (u) + O η3 ,
p

R(u) = 1 ⊗ 1 + ηr(u) + O η2 ,
where 1 is the (2 × 2) identity and r(u) reads

 
cosh(pu) 0 0 0
 0 0 1 0 
r(u) = 

.

0 1 0 0
0 0 0 cosh(pu)
Unfortunately, if one wants to extend the results to spins higher than 1/2, one has to
increase the dimension of the auxiliary space accordingly Ref. [24]. There is, though,
the remarkable exception of isotropic models, i.e., the ones obtained in the p → 0 limit,
for which the quantum algebra reduces to su(2). The XXX model with higher spin was
introduced in Refs. [24,25]. A central point of the approach is that, the quasi-classical limit
η → 0 reduces the quantum algebra to ordinary su(2), whatever is the dimension of the
quantum space Hj . This implies that the operators thus found are realized through spin
operators
Sjz ,
Sj± , and that they commute with each other for arbitrary spins.
3. Quasi-classical expansion of vertex models with boundary reflections
In this section we investigate systematically the power η-expansion of the transfer

matrix of disordered vertex models with boundary. As final product of this procedure we
obtain a class of integrable models describing interacting quantum spins with non-uniform
long range interaction.
The quasi-classical expansion of the transfer matrix reads

tˆ(u) = τ̂ (0) + ητ̂ (1)(u) + η2 τ̂ (2)(u) + O η3 . (21)
The property [tˆ(u), tˆ(v)] = 0 ensures the existence of hierarchy of integrable models in the
quasi-classical expansion. We have indeed
∞

tˆ(u), tˆ(v) = ηl Cl (u, v) = 0,
l=0
which implies Cl (u, v) = 0. We give the first five terms

C0 (u, v) = τ̂ (0) (u), τ̂ (0) (v) ,

C1 (u, v) = τ̂ (0) (u), τ̂ (1) (v) + τ̂ (1)(u), τ̂ (0) (v) ,

C2 (u, v) = τ̂ (0) (u), τ̂ (2) (v) + τ̂ (2)(u), τ̂ (0) (v) + τ̂ (1) (u), τ̂ (1)(v) ,

C3 (u, v) = τ̂ (0) (u), τ̂ (3) (v) + τ̂ (3)(u), τ̂ (0) (v) + τ̂ (1) (u), τ̂ (2)(v)

+ τ̂ (2) (u), τ̂ (1) (v) ,

C4 (u, v) = τ̂ (0) (u), τ̂ (4) (v) + τ̂ (4)(u), τ̂ (0) (v) + τ̂ (1) (u), τ̂ (3)(v)

+ τ̂ (3) (u), τ̂ (1) (v) + τ̂ (2) (u), τ̂ (2)(v) . (22)
From the expansion above, one finds that the first non-trivial term τ̂ (n) (u) (i.e., which is
not just a C-number or an invariant of the algebra) gives rise to a family of commuting
operators. For example, if τ̂ (0) = C-number (as usually is the case) the first class of
integrable models is generated by [τ̂ (1)(u), τ̂ (1) (v)] = 0. In the presence of boundary
conditions corresponding to generic choice of ξ± it turns out that τ̂ (1)(u) is non-trivial.
This is not what we wish, since τ̂ (1) (u) only contains non-interacting spins.1 In the next
section we will see how the parameters ξ± can be suitably chosen to yield “trivial” τ̂ (1)(u),
such that the first non-trivial term in the quasi-classical expansion of tˆ will be τ̂ (2)(u),
which yields a spin–spin interaction.
The η expansion of the Lax operators is
sinh(pu) (1) (2)
Lj (u) 1 + ηlj (u) + η2 lj (u) , (23)
p
where
p + −
lj(1) (u) = p coth(pu)
Sjz σ z + Sj σ +
Sj− σ + ,
sinh (pu)
1 In general, the nth order terms will contain up to n-body terms.

p2 z 2
lj(2) (u) =
S 1. (24)
2 j
Thus, the monodromy matrix is:
p − z z
T (u) P (u) 1 + η + − − +
− cosh puj Sj σ + Sj σ + Sj σ
j
sinh (puj )
− z z + − − + − + + −
cosh (pu− j ) cosh (puk )Sj Sk 1 + Sj Sk σ σ + Sj Sk σ σ
+ p2 η2
j <k
sinh (pu− −
j ) sinh (puk )

1 z 2
+ Sj 1 + irrelevant terms ,
2
j

+
detq T (−u + η/2) P (−u) 1 − pη
2
coth puj ,
j
1 z z
T −1
(−u) P −1
(−u) 1 + pη cosh pu+ + − − +
j Sj σ + Sj σ + Sj σ
j
sinh (pu+
j )
+ z z + − − + − + + −
cosh (pu+
j ) cosh (puk )Sj Sk 1 + Sj Sk σ σ + Sj Sk σ σ
+ p2 η2
j <k
sinh (pu+ +
j ) sinh (puk )

1 z 2
+ Sj 1 + irrelevant terms ,
2
j

where we defined u±
j ≡ u ± zj , P (u) =
−
j sinh(puj )/p. The irrelevant terms are either
C-numbers or off-diagonal matrices, contributing to the trace with terms order η3 . The
expansion of K± reads
K± (u, ξ± )
(0)
1 sinh p u + ξ± 0
(0)
p 0 − sinh p u − ξ±
(1) (0)
δ±,+ + ξ± cosh p u + ξ± 0
+η (1) (0)
,
0 − δ±,+ − ξ± cosh p u − ξ±
where δ±,+ is the usual Kronecker δ, and we took into account the expansion
(0) (1)
ξ± ξ± + ηξ± . (25)
(0) (0) (0) (0) (1) (1)
We define for convenience 2ξ = ξ+ − ξ− , 2Σ = ξ+ + ξ− , and 2ς = ξ+ + ξ− . Then
the terms in the expansion of the transfer matrix given in Eq. (21) read
1
τ̂ (0) (u) = 2
P (u)P −1 (−u) cosh (2pu) cosh (2pΣ) − cosh (2pσ ) ,
p
1
τ̂ (1) (u) = C-numbers + P (u)P −1 (−u) sinh (2pu) sinh(2pΣ)
p
− z
× coth puj + coth pu+
j Sj ,
j
τ̂ (2)
(u) = C-numbers

+ P (u)P −1 (−u) 2ς sinh (2pu) cosh(2pΣ)
+ z
× coth pu−j + coth puj Sj

j

+ cosh (2pu) sinh (2pΣ) − sinh (2pξ )
+ z
× coth pu−j + coth puj Sj

j
1 + z
+ sinh (2pu) sinh(2pΣ) coth pu+
j coth puk Sj
2
jk
1
+ cosh (2pu) cosh (2pΣ) − cosh (2pξ )
2
+ + z z
× coth pu− coth pu−
j + coth puj k + coth puk Sj Sk
jk

1 1 1
+ +
2 sinh(pu− −
j ) sinh(puk ) sinh(pu+ +
j ) sinh(puk )

+ −
× Sj Sj−
Sk + Sk+
1 Sk− +
Sj+ Sj−
Sk+
− cosh (2pu) cosh (2pξ ) − cosh (2pΣ)
2 sinh (pu− +
j ) sinh (puk )
jk

1 Sk− −
Sj+ Sj−
Sk+
+ sinh (2pu) sinh (2pξ ) . (26)
2 sinh (pu−
jk
+
j ) sinh (puk )
As can be seen by Eqs. (22), the operators τ̂ (2)(u) commute with each other if
(1)
τ̂ (u), τ̂ (3)(v) + τ̂ (3) (u), τ̂ (1)(v) = 0.
A sufficient condition for this relation to be fulfilled is that τ̂ (1) (u) is just a C-number. This
requires that Σ = 0.
3.1. Classical boundary
At first, we assume classical boundary. With this term we mean that the boundary
parameters ξ± are assumed to be independent of η, i.e., ς = 0. This case was analyzed
by one of the authors in Ref. [20]. The commuting family of operators τ̂ (2)(u) given above
reduces to

Sz
τ̂0(2) (u) = 2P (u)P −1 (−u) sinh2 (2pu) Sz
cosh (2pu) − cosh (2pzj ) j
j
1
+
(cosh(2pu) − cosh(2pzj ))(cosh(2pu) − cosh(2pzk ))
j,k
j =k

1 z z
× cosh(2pzj ) + cosh(2pzk ) − 2 cosh(2pξ ) Sj
Sk
2
1 + −
+ cosh p(zj + zk ) − cosh(2pξ ) cosh p(zj − zk ) Sk +
Sj Sj−
Sk+
2

1 + −
− sinh(2pξ ) sinh p(zj − zk ) Sj
Sk − Sj−
Sk+ ,
2
where we dropped the C-numbers and the Casimir coming from the term j = k in the
sums. A finite subset of u-independent operators in involution can be obtained taking the
limits u → zj of τ̂ (2) (u), and dividing by the factor

sinh(2pzj )P −1 (−zj ) sinh p(zj − zk ) /p.
k=j
They are
1
τ̂0j =
Skz
Sjz +
k k
k=j
z z
Sk
+ −
+ cosh p(zj + zk ) − cosh(2pξ ) cosh p(zj − zk ) Sj Sj−
Sk + Sk+
+ − !
− sinh(2pξ ) sinh p(zj − zk ) Sj
Sk − Sj−
Sk+ . (27)
The τ̂j form a complete set, in the sense that any τ̂ (2)(u) can be built from them according
to the formula
1
τ̂0 (u) = 2P (u)P −1 (−u) sinh2 (2pu)
(2)
τ̂0j .
j

z z S z
≡ Sjz
We notice the term k Sk Sj in the operators τ̂j . It describes a self-interaction of
the spins with the magnetic field generated by the spins themselves. In the next section we
shall see how to add an external magnetic field.
The eigenstates of operators (27) are given by

M
|Ψ =
S − (eα )|H , (28)
α=1
where
cosh[p(u + zj + 2ξ )] − cosh[p(u − zj )]

S − (u) = Sj− ∝
B(u)
.
cosh(2pu) − cosh(2pzj ) dη η=0
j
The rapidities eα fulfill the first order term in the expansion of Eqs. (19) around η = 0
1 Sj
−
cosh(2peα ) − cosh(2peβ ) cosh(2peα ) − cosh(2pzj )
β=α j
1/2
+ = 0. (29)
cosh(2peα ) − cosh(2pξ )
Putting cosh(2peα ) = exp(2Eα ) + cosh(2pξ ) and cosh(2pzj ) = exp(2wj ) + cosh(2pξ ),
the equations above reduce to the modified Gaudin’s equations presented in Ref. [15]:

coth(Eα − Eβ ) − Sj coth(Eα − wj ) + S z = 0. (30)
β=α j
The eigenvalues are

τ0j = Sj S z + Sk coth (wj − wk ) − coth (wj − Eα ) . (31)
k=j α
3.2. Non-classical boundary
In this section we show how to include a scalable term proportional to Sjz in the
operators τ̂j . Such a term is crucial for physical applications since, as we shall see, it allows
to introduce a non-uniform magnetic field in the Hamiltonian. Furthermore, when the spins
are realized by pairs of time-reversed electrons Sj+ = cj ↑ cj ↓ a
Sjz = − 12 (n̂j ↑ + n̂j ↓ − 1),
non-uniform magnetic field corresponds to a kinetic energy term (see Section 4). In order
(1) (1)
to reach our goal, we have exploited the fact that ξ± can depend on η, i.e., ξ+ + ξ− is not
necessarily zero. We refer to this kind of boundary conditions as a non-classical boundary.
Thus, we put ς = 0. We obtain

2ς + Sz
τ̂ (2) (u) = 2P (u)P −1 (−u) sinh2 (2pu) Sz
cosh (2pu) − cosh (2pzj ) j
j
1
+
(cosh(2pu) − cosh(2pzj ))(cosh(2pu) − cosh(2pzk ))
j,k
j =k

1 z z
Sk
2
1 + −
Sk + Sk+
2
1 + −
− sinh(2pξ ) sinh p(zj − zk ) Sk −
Sj Sj−
Sk+ . (32)
2
The integrals of motion are obtained again taking the limits u → zj , dividing now also by
2ς +
S z:
. z z 1
τ̂j = τj
S = Sj − J
S
k
k=j
z z
Sk
+ −
Sk + Sk+
+ − !
Sj Sj−
Sk+ , (33)
where we put J ( S z ) = J /(1 − J

S z ), with J = −1/(2ς ). The operators τ̂ (2)(u) can be
built from the τ̂j according to
1
τ̂ (2) (u) = 2 2ς +
S z P (u)P −1 (−u) sinh2 (2pu) τ̂j .
j
For real J , the τ̂j are Hermitian if zj are real and ξ is pure imaginary, or vice versa. Their
eigenstates are still given by Eq. (28)

M
|Ψ =
S − (eα )|H , (34)
α=1
where
cosh[p(u + zj + 2ξ )] − cosh[p(u − zj )]
−
S (u) = −
Sj ∝ B(u)
.
cosh(2pu) − cosh(2pzj ) dη η=0
j
The difference with the previous subsection, is that the first order term in the expansion of
Eqs. (19) around η = 0 contains an additional term
1 Sj
−
cosh(2peα ) − cosh(2peβ ) cosh(2peα ) − cosh(2pzj )
β=α j
2 (1 + 1/J )
1
+ = 0. (35)
cosh(2peα ) − cosh(2pξ )
Putting cosh(2peα ) = exp(2Eα ) + cosh(2pξ ) and cosh(2pzj ) = exp(2wj ) + cosh(2pξ ),
the equations above reduce as well to the modified Gaudin’s equations presented in Ref.
[15]:
1
coth(Eα − Eβ ) − Sj coth(Eα − wj ) + = 0. (36)
J (S z )
β=α j
The eigenvalues are

z
τj = Sj 1 − J S Sk coth (wj − wk ) − coth (wj − Eα ) . (37)
k=j α
There are parameterizations yielding rational Bethe equations. Among these, we consider
1 + xj
= cosh(2pzj ) − cosh(2pξ ),
1 − xj
1 + λα
= cosh(2peα ) − cosh(2pξ ). (38)
1 − λα
Then the eigenvalues are

z 1 − xj xk z 1 − xj λα
τj = Sj 1 − J S Sk +J S . (39)
xj − xk xj − λα
k=j α
The Bethe equations (35) become

1 Sj
−
λ − λβ λα − xj
β=α α j

1 1 + J (S z )(1 + S z ) 1 − J (S z )(1 + S z )
+ + = 0. (40)
2J (S z ) 1 + λα 1 − λα
In this form, they admit a two-dimensional electrostatic interpretation [26].
3.3. On the equivalence with Gaudin’s model in external magnetic field
We show that, at ξ = 0 and pξ = iπ/2, these integrals of motion are equivalent to the
modified Gaudin’s Hamiltonians introduced in Ref. [15]. The integrals of motion reduce to

z cosh(2pzj ) + cosh(2pzk ) ∓ 2 z z
τ̂j =
Sjz − J S S
S
cosh(2pzj ) − cosh(2pzk ) j k
k=j

cosh[p(zj + zk )] ∓ cosh[p(zj − zk )] + −
+ Sj−
Sk +
Sj Sk+ , (41)
where the upper sign refers to ξ = 0 and the lower one to pξ = iπ/2. We make the change
of variable sinh (pzj ) = exp wj , if ξ = 0, cosh (pzj ) = exp wj , if pξ = iπ/2, obtaining

z 1 + − − +
z
τ̂j = Sj − J S coth(wj − wk )
Sjz
Skz +
S S + Sj Sk .
2 sinh(wj − wk ) j k
k=j
Thus, apart a sector-dependent rescaling of the coupling J → J (

S z ) = J (1 − J
S z )−1 ,
the operators given above are equivalent to modified Gaudin’s Hamiltonians.
3.4. Construction of the Hamiltonian
The simplest Hamiltonian that it is possible to build up is a first degree polynomial in

τ̂j with arbitrary real parameters hj
z hj − hk
H= 2hj τ̂j = 2hj
Sjz − J
S
j j j,k
j =k
z z
× cosh(2pzj ) + cosh(2pzk ) − 2 cos (2pt) Sj
Sk
+ −
+ cosh p(zj + zk ) − cos(2pt) cosh p(zj − zk ) Sj Sj−
Sk + Sk+
+ − !
− i sin(2pt) sinh p(zj − zk ) Sk −
Sj Sj−
Sk+ , (42)
where we put ξ = it, with real t.
4. Second quantized Hamiltonians
In this section we employ the fermionic realization of su(2) to write the integrable
models we found Eq. (42) in second quantization. The two orthogonal Dj - and Dl -
dimensional realizations are
Dj
+ †
+ =
K cj,δj ↓ cj δj ↑ , − = K
K ,
j j j
δj =1
Dj

z = 1
K (1 − n̂j δj ↑ − n̂j δj ↓ ), (43)
j
2
δj =1
and
Dl Dl
† 1
Sl+
= †
clρ clρl ↓ , Sl−
= Sl+ ,

Slz = (n̂lρl ↑ − n̂lρl ↓ ), (44)
l↑ 2
ρl =1 ρl =1
where operators c, and n ≡

c† , c† c
are fermionic operators. We arbitrarily grouped
the levels in the subsets j = 1, . . . , ΩK , each containing Dj levels, and in the subsets
l = 1, . . . , ΩS , each containing Dl levels. The maximum values of the z components of
the spin are Kj = Dj /2 and Sl = Dl /2, respectively. Thus, a level a will be characterized
alternatively by the pairs (j (a), δj (a)(a)) or (l(a), ρl(a)(a)). We write a Hamiltonian of the
form
H = H K + H S + E0 , (45)
where E0 is a constant and

ΩK
ΩK
HK = − +
2ηj τj K j2 ,
gjj K (46)
j =1 j =1

ΩS
ΩS
HS = 2ζl τl
S + Jllxx
S l2 , (47)
l=1 l=1
O
where operators τ (O), = K,
S are defined in Eq. (33). Due to the orthogonality of the
realizations (43), (44) we observe that [HK , HS ] = 0. Furthermore, HK and HS are block-
diagonal, and their common eigenstates are the direct product of the eigenstates of HK
and of HS , each restricted to the subspace corresponding to one of its blocks [16]. The
integrability together with the exact solution of the Hamiltonian (45) follows from the
integrability of each HK , HS proved in Section 3.2 and from Eqs. (34), (36), and (37).
Finally, the second quantized form of the Hamiltonian (45) reads
† †
H= εaσ n̂aσ + Uab (n̂a↑ + n̂a↓ )(n̂b↑ + n̂b↓ ) + gab ca↑ ca↓ cb↓ cb↑
aσ ab
xx † †

+ Jab
z
(n̂a↑ − n̂a↓ )(n̂b↑ − n̂b↓ ) + Jab ca↑ cb↓ cb↑ ca↓ , (48)
where Ω number the levels, , Ω and ca,σ ≡ cj (a),δj (a)(a),σ ; the constant in
a, b = 1, . . .
Eq. (45) turns out to be E0 = j Dj εj + j k Dj Dk Uj k .
The kinetic energy term reads
1 1

εaσ n̂aσ = (εa↑ + εa↓ )(n̂a↑ + n̂a↓ ) + (εa↑ − εa↓ )(n̂a↑ − n̂a↓ ) .
aσ a
2 2
We choose a partition—in equivalence classes—of the single particle levels in such a
way that all levels having the same value of εa ≡ 12 (εa↑ + εa↓ ) belong to the same class
(hence we write εj instead of εa , where j individuates the class2 ), and, analogously, a
second partition is defined in such a way that all the levels having the same value of
ζa ≡ 12 (εa↑ − εa↓ ) belong to the same class (hence we write the common value as ζl ).3
The couplings between levels a and b depend only on the equivalence classes of the two
levels. For j ≡ j (a) = k ≡ k(b) and l ≡ l(a) = m ≡ m(b), they are
cosh[p(zj + zk )] − cosh[p(zj − zk − 2itK )]
gab = gj k = 2JK K z (ηj − ηk ) ,
cosh(2pzj ) + cosh(2pzk ) − 2 cos (2ptK )
4Uab = 4Uj k = JK K z (ηj − ηk ) ,
cosh[p (yl + ym )] − cosh[p (yl − ym + 2itS )]
Jabxx
= Jlmxx
= −2JS S z (ζl − ζm ) ,
cosh(2p yl ) − cosh(2p ym )
z z cosh(2p yl ) + cosh(2p ym ) − 2 cos (2p tS )
Jab = Jlm = −JS S z (ζl − ζm ) . (49)
cosh(2p yl ) − cosh(2p ym )
For j = k, we have the relation gjj = 4Ujj , and gjj can be chosen arbitrarily.4
Analogously, for l = m, we have Jllz = Jllxx .
2 η = ε + 2 D U + g /2 have to be determined consistently; they must satisfy a system of linear

j j k k jk jj
equations, as discussed in [27].
3 The partitioning of the levels that we chose above guarantee that the interaction does not vanish between
levels having the same value of εa or ζa .
4 In particular, they can be chosen in such a way that η = ε [28].
j j
5. Conclusions
In this paper we have studied integrable disordered vertex models in presence of

boundary reflections. The quasi-classical expansion of the models has been thoroughly
investigated. This expansion produces a hierarchy of models which import the integrability
of the original vertex models. The extraction of the energy from the generating functionals
(which is infeasible, in general) is very simplified by the quasi-classical limit and the
Hamiltonian is given as polynomial of the integrals of motion. The class of models
we obtain describes interacting spins with non-uniform couplings, and in a non-uniform
external magnetic field. In this sense the present models generalize those ones found in
Ref. [20]. On the other hand, these Hamiltonians constitutes also a one-parameter (t in
the text) extension of the class of models found in Ref. [15] that are recovered when
the boundary terms give rise of twisted periodic boundary conditions. As a result, the
integrability of these latter models has a firm ground within the Sklyanin procedure [19].
The presence of the external magnetic field is an effect of the boundary terms which are
assumed, in turn, quasi-classical. We also obtained the exact solution of the class of models
presented here through algebraic Bethe ansatz. The Bethe equations can be recast in a form
which allows the electrostatic analogy as was done in Ref. [26].
An important point is that the models apply to any spin Sj (not only to spin 1/2). The
reason is that, in the present case, the integrals of motion can contain only spin S operators
since the quantum algebra suq (2) reduces to su(2) in the quasi-classical limit (whatever
the dimension of the representation is).
By realizing the spin operators in terms of fermions, the class of models we found
describes confined fermions in degenerate levels with pairing force interaction.
Appendix A. Inhomogeneous vertex models
VM are models of 2D classical statistical mechanics. They consist in a (K × N) array

of vertices (see Fig. 1), where the nearest neighbours are linked by horizontal and vertical
legs. The legs can be of several species, each identified by a number hi,j = 1, . . . , Hi ,
for the horizontal legs of the ith row, and vi,j = 1, . . . , Vj for the vertical ones of the
j th column (the number of species can depend on the row or column; in this case, we
have an inhomogeneous model). Here we are using the convention that the pair (i, j )
individuates the vertical (horizontal) leg above (left of) the vertex (i, j ). A statistical weight
w(legsi,j ; i, j ) = exp [−β ε(legsi,j ; i, j )] is assigned to each vertex, depending on the
legs configurations (legsi,j = {hi,j , hi+1,j , vi,j , vi,j +1 }) around it. If the weight depends
explicitly on the positionof thevertex, we have a disordered model. The goal is to find the
partition function Z = {legs} i,j w(legsi,j ; i, j ).
To each vertex, one can associate a matrix, whose elements are the weights correspond-
ing to the possible configurations of legs, in the following way: fix the horizontal legs
around site (i, j ) to their minimal values, say hi,j = hi+1,j = 1; then vary the values of the
vertical legs, associating the upper one to a row, and the lower one to a column; a (Vj × Vj )
matrix is thus obtained, which we indicate by Lj1,1 (i), whose entries are the weights corre-
sponding to the possible legs configurations with horizontal legs fixed to 1; then repeat the
(a) (b)
Fig. 1. (a) A vertex configuration. (b) Numbering of the lattice.
procedure changing the values of the horizontal legs, associating the left one to a row, and
the right one to a column; a block matrix Lj (i) is finally obtained, which is conventionally
hi,j ,h
called the Lax operator. It is a (Hi × Hi ) matrix whose entries Lj i+1,j
(i) are in turn
(Vj × Vj ) matrices, i.e., operators over the linear space Hj . The partition function of the
(1 × N) lattice with periodic boundary conditions in vertical and horizontal direction is,
in terms of the Lax operators, Z1 = TrV TrH {L1 (1) · · · LN (1)} ≡ TrV tˆ(1), where by TrH
we mean the trace over the horizontal space, and by TrV the trace over the vertical ones;
we introduced the transfer matrix tˆ(i) ≡ TrH {L1 (i) · · · L N (i)}, where the hat is meant to
remind that the transfer matrix is an operator over H = j Hj , the direct product of the
linear spaces associated to the vertical legs. For a (K × N) lattice, the partition function is
tˆ(i),tˆ(i )] = 0, ∀i, i , it is possible to simultaneously diagonal-
Z = TrV {tˆ(1) · · · tˆ(K)}. If [
ize the tˆ(i), obtaining Z = r K ˆ
i=1 tr (i), where tr (i) is the rth eigenvalue of t (i). In this
case, the VM is exactly solvable.
From the tˆ(i), it is commonly possible to extract many-body Hamiltonians of interest.5
Thus, a given exactly solvable vertex model corresponds uniquely to a family of
commuting many-body operators.
It turns out that the transfer matrices commute with each other, and thus the
corresponding vertex model is exactly solvable, if Hi = Hi = H , ∀i, i , and a family of
(H 2 × H 2 ) matrices, the Ř-matrices, exists, such that the Lax operators obey the relation
Ř(i, i )Lj (i) ⊗ Lj (i ) = Lj (i ) ⊗ Lj (i)Ř(i, i ). (A.1)
Given a Ř-matrix, this is a very strict requirement, which in general implies that many legs
configurations are not allowed, i.e., their weight is zero, while the allowed ones are related
to each other by some parametrization.
A relevant case is when the dimensions of vertical and horizontal space are equal, and
they do not depend on the row or column: Hi = Vj ≡ 2S + 1. Then, the Ř-matrices are
but the Lax operators where the matrix elements have been written down explicitly in
their matrix representation. It turns out that the entries of the Lax operators are matrices
5 In general, such Hamiltonians are not by any means related to the Hamiltonian of the VM.
belonging to the (2S + 1)-dimensional realization of su(2), i.e., spins over the Hilbert6
space Hj . There is the drawback that the Ř-matrix is difficult to determine and to handle,
since its dimension increases very fast with S [29,30]. A technique to build larger Ř-
matrices using the (4 × 4) Ř-matrices (the simplest ones) as building blocks was devised
by Kulish, Reshetikhin, and Sklyanin [31]. In the present paper, by means of the quasi-
classical expansion, we will build up operators for spins higher than 1/2 still making use
of (4 × 4) Ř-matrices.
Appendix B. General K+ matrix
In this appendix we construct the Hamiltonian when the general solution of the
reflection equation (15) is considered [33,34]. In this case we have

1 sinh p(u + η + ξ+ ) κ+ sinh 2p(u + η)
K+ (u) = . (B.1)
p κ+ sinh 2p(u + η) − sinh p(u + η − ξ+ )
Since we want τ̂ (1) to be a C-number, we must impose κ+ iηc. Thus, the final effect of
the general reflection results in an additional term in the second order of the transfer matrix
2ic
τ̂ (2) (u) → τ̂ (2) (u) + P (u)P −1 (−u) sinh2 (2pu)
p

sinh[p(zj − ξ )]
×
S+
cosh(2pu) − cosh(2pzj ) j
j

sinh[p(zj + ξ )]
−
Sj− . (B.2)
j
The Hamiltonian is again built according to

H= 2hj τ̂j , (B.3)
j
where the integrals of motion are

z sinh[p(zj − ξ )] + sinh[p(zj + ξ )] −
τ̂j = 1 − J Sz Sj − icJ
Sj −
Sj
p p
1
−J
k
k=j
z z
Sk
+ −
Sk + Sk+
+ − !
Sj Sj−
Sk+ . (B.4)
6 It is a finite-dimensional vector space. We denote it as Hilbert space in foresight of its interpretation as a

quantum space.
We point out that the Hamiltonian is Hermitian for real c and ξ = it with real t. The
diagonalization of this class of Hamiltonians might be achieved by functional Bethe ansatz
[6]. Nevertheless, it seems worth to study the models (B.3), (B.4) since their potential
application to condensed matter (see also Eqs. (43), (44)).
Appendix C. Generalization to su(n)
In this appendix we briefly discuss on a generalization of the Gaudin model to the

su(n) case. We can define the Gaudin model for other Lie algebras (see, e.g., Ref. [32] for
recent works). Following the method depicted in Section 3 we obtain a modified Gaudin
Hamiltonian to include a scalable term proportional to the Cartan generators of su(n) (as
far as we know, previously obtained models do not contain this term).
The trigonometric R-matrix for su(n) chains is given by
1 1
n n

R(λ) = sinh p(λ + η) E aa ⊗ E aa + sinh(pλ)E aa ⊗ E bb
p p
a=1 a=b
1 −pλ sgn(a−b)
n
+ e sinh(pη)E ab ⊗ E ba , (C.1)
p
a=b
where E ab denotes n × n matrix with unity at (a, b) element. They satisfy

ab cd bc ad
,E
E = δ E − δ da Ecb .
The corresponding diagonal solution of the reflection equation (14) is [33]
1 pλ 1 −pλ
F− n

K− (λ) = e sinh p(ξ− − λ) E aa + e sinh p(ξ− + λ) E aa ,
p p
a=1 a=F− +1
1 −pλ−2pηa
F+

K+ (λ) = e sinh p(ξ+ + λ) E aa
p
a=1
1
n

+ epλ+pη(n−2a) sinh p(ξ+ − ηn − λ) E aa , (C.2)
p
a=F+ +1
where F± is arbitrary, 1 F± n. Hereafter we set F+ = F− = F for simplicity. With these

K-matrices, the Hamiltonian of the su(n) homogenous spin chain with nearest neighbour
interaction with open boundary was computed in Refs. [35,36] by using the formula (20).
The Hamiltonian with the long range interaction is constructed following the procedure
presented in Section 3.4, where the constants of motion are calculated by the formula (21).
They read

n
τ̂j = 2 jaa
aE
a=1
F
(1) (1)
+ ξ− coth p(ξ − zj ) + ξ+ coth p(ξ + zj )
a=1

sinh(2pzj )
+ (n − F) aa
E
sinh(p(ξ − zj )) sinh(p(ξ + zj )) j

n
! aa
+ ξ−(1) coth p(ξ + zj ) + ξ+(1) coth p(ξ − zj ) E
j
a=F+1

n
+ coth p(zj − zk ) + coth p(zj + zk ) aa E
E aa
j k
k=j a

n
ep(zj −zk ) sgn(a−b)
+ ba E
E ab
sinh(p(zj − zk )) j k
a=b

F
n
sinh(p(ξ + zj )) ep(zk −zj )
+ E ab
ba E
sinh(p(ξ − zj )) sinh(p(zj + zk )) j k
a=1 b=F+1

n
F
sinh(p(ξ − zj )) ep(zj −zk )
+ ba E
E ab
sinh(p(ξ + zj )) sinh(p(zj + zk )) j k
a=F+1 b=1

F
ep(zj +zk ) sgn(b−a) ba ab
n
ep(zj +zk ) sgn(b−a) ba ab
+ E E + E E ,
sinh(p(zj + zk )) j k sinh(p(zj + zk )) j k
a,b=1 a,b=F+1
a=b a=b
where Eab are site-k su(n) operators.

k
The eigenvalue of τ̂j is given as the quasi-classical limit of the su(n)-transfer matrix
(see Eq. (6) of [35]) τ̂ (u) and with u → zj :
(1) (1)
τj = ξ+ coth p(zj + ξ ) − coth(pξ ) − ξ− coth p(zj + ξ ) + coth(pξ )
sinh(2pzj ) e−2zj
+ (F − n) +1−n
sinh[p(zj − ξ )] sinh[p(zj + ξ )] sinh(2zj )

N

+ coth p(zj + zk ) + coth p(zj − zk )
k=j

M1
(1) (1)
+ coth p zj + ek + coth p zj − ek . (C.3)
k
The Bethe ansatz equations can be obtained in the same limit of a result in Ref. [35], and
we have (for a = 1, 2, . . . , n − 1)

Ma
(a) (a) (a) (a)
2 coth p ej + ek + coth p ej − ek
k=j

+ δa,F n + ξ−(1) − ξ+(1) coth p(z − ξ ) + coth p(z + ξ )

Ma+1

= coth p ej(a) + ek(a+1) + coth p ej(a) − ek(a+1)
k

Ma−1

+ coth p ej(a) + ek(a−1) + coth p ej(a) − ek(a−1) . (C.4)
k
Here we assume ej(0) = zj and M0 = N , Mn = 0.

The fermionic models can be obtained by using the fermionic realization
jab = c† cj,b − 1 δab ,

E (C.5)
j,a
n
with a constraint
†
cj,a cj,a = 1.
a
References
[1] R.J. Baxter, Exactly Solved Models in Statistical Mechanics, Academic Press, 1982.
[2] M. Gaudin, La fonction d’onde de Bethe, Masson, 1983.
[3] V.E. Korepin, N.M. Bogoliubov, A.G. Izergin, Quantum Inverse Scattering Method and Correlation
Functions, Cambridge Univ. Press, 1993.
[4] M. Gaudin, J. Physique 37 (1976) 1087.
[5] K. Hikami, P.P. Kulish, M. Wadati, J. Phys. Soc. Jpn. 61 (1992) 3071.
[6] E.K. Sklyanin, J. Sov. Math. 47 (1989) 2473.
[7] M.C. Cambiaggio, A.M.F. Rivas, M. Saraceno, Nucl. Phys. A 624 (1997) 157.
[8] R.W. Richardson, Phys. Lett. 3 (1963) 277.
[9] J. von Delft, D.C. Ralph, Phys. Rep. 345 (2001) 61.
[10] A. Mastellone, G. Falci, R. Fazio, Phys. Rev. Lett. 80 (1998) 4542.
[11] L. Amico, G. Falci, R. Fazio, J. Phys. A 34 (2001) 6425–6434.
[12] G. Sierra, Nucl. Phys. B 572 (2000) 517–534.
[13] H.M. Babujian, J. Phys. A 26 (1993) 6981.
[14] N. Reshetikhin, A. Varchenko, in: Geometry Topology, and Physics, 1995, p. 293.
[15] L. Amico, A. Di Lorenzo, A. Osterloh, Phys. Rev. Lett. 86 (2001) 5759.
[16] L. Amico, A. Di Lorenzo, A. Osterloh, Nucl. Phys. B 614 (2001) 449.
[17] J. Dukelsky, C. Esebbag, P. Schuck, Phys. Rev. Lett. 87 (2001) 66403.
[18] J. von Delft, R. Poghossian, Algebraic Bethe ansatz for a discrete-state BCS pairing model, cond-
mat/0106405.
[19] E.K. Sklyanin, J. Phys. A 21 (1988) 2375.
[20] K. Hikami, J. Phys. A 28 (1995) 4997.
[21] K. Hikami, J. Phys. A 28 (1995) 4053.
[22] I. Cherednik, Theor. Math. Phys. 61 (1984) 35.
[23] H.J. de Vega, Nucl. Phys. B 240 (1984) 495.
[24] H.M. Babujian, Phys. Lett. A 90 (1982) 479.
[25] L.A. Takhtajan, Phys. Lett. A 87 (1982) 479.
[26] L. Amico, A. Di Lorenzo, A. Mastellone, A. Osterloh, R. Raimondi, Ann. Phys. 299 (2002) 228.
[27] A. Di Lorenzo, A new class of exactly solvable models, PhD thesis, Università di Catania, Italy, 2001.
[28] R.W. Richardson, private communication.
[29] V.I. Fateev, A.B. Zamolodchikov, Sov. J. Nucl. Phys. 32 (1980) 298.
[30] K. Sogo, Y. Akutsu, T. Abe, Prog. Theor. Phys. 70 (1983) 730.
[31] P.P. Kulish, N. Reshetikhin, E.K. Sklyanin, Lett. Math. Phys. 5 (1981) 393.
[32] P.P. Kulish, N. Manojlovic, Lett. Math. Phys. 55 (2001) 77.
[33] H.J. de Vega, A. Gonzalez-Ruiz, J. Phys. A 26 (1993) L519.
[34] S. Ghoshal, A. Zamolodchikov, Int. J. Mod. Phys. A 9 (1994) 3841.
[35] H.J. de Vega, A. González-Ruiz, Mod. Phys. Lett. A 9 (1994) 2207.
[36] A. Doikou, R.I. Nepomechie, Nucl. Phys. B 530 (1998) 641.
Critical exponent ω at O(1/N) in O(N) × O(m)

spin models
J.A. Gracey
Theoretical Physics Division, Department of Mathematical Sciences, University of Liverpool, PO Box 147,
Liverpool, L69 3BX, United Kingdom
Received 8 July 2002; received in revised form 3 September 2002; accepted 6 September 2002
Abstract
We compute the O(1/N) correction to the stability critical exponent, ω, in the Landau–Ginzburg–
Wilson model with O(N) × O(m) symmetry at the stable chiral fixed point and the stable direction
at the unstable antichiral fixed point. Several constraints on the O(1/N) coefficients of the four loop
perturbative β-functions are computed.
1. Introduction
The renormalization group analysis of spin models has proven to be an important

tool in understanding and predicting critical properties of phase transitions in a variety
of phenomena. For a recent review see, for example, [1]. For instance, the Heisenberg
model based on an O(3) nonlinear sigma model or O(3) φ 4 theory has been widely
used to understand ferromagnetic phase transitions in nature. Indeed the perturbative field
theory techniques in this instance have been developed to five loops in MS in standard
d-dimensional regularization in φ 4 theory, [2], and equally impressively to six loops in
the three-dimensional fixed dimension renormalization, [3–5]. Given such success these
methods have been applied to similar models of other critical phenomena. Over a period
of years various calculations have been carried out in the Landau–Ginzburg–Wilson model
which essentially is an extension of φ 4 theory where the O(N) symmetry is replaced by
an O(N) × O(2) symmetry, [6–11]. The critical structure in this generalized model is
richer than the usual φ 4 theory in that theoretically there are several fixed points over and
E-mail address: jag@amtp.liv.ac.uk (J.A. Gracey).
PII: S 0 5 5 0 - 3 2 1 3 ( 0 2 ) 0 0 8 1 8 - 0
434 J.A. Gracey / Nuclear Physics B 644 [FS] (2002) 433–450
above the Heisenberg one, depending on the value of N , which are either stable or unstable,
[10,11]. However, the properties of the phase transitions in this Landau–Ginzburg–Wilson
class of models is controversial, [9,11–21]. First, the two-dimensional nonlinear sigma
model with the same symmetry is believed to reside in the same universality class, [10,
11]. Therefore, it ought to be possible to use either model to compute useful information
on the critical properties of the physically interesting stable chiral fixed point. However,
it has been pointed out in [12] that the results from a (d = 2 + ε)-dimensional study do
not match those deduced from the higher-dimensional theory. Second, in field theoretical
calculations such as perturbation theory the phase transitions are regarded as second order
whilst numerical or Monte Carlo simulations would appear to indicate transitions are first
order in nature, [9,13–18]. Moreover, the particular behaviour depends on the value of N
though the precise range where this occurs is still undetermined. Further, one recent work
has suggested an interesting point of view for the origin of the disagreements for N = 2
and 3 in both two and three dimensions. In [22] it is argued that it is due to the fact that
one flows to the stable chiral fixed point along a spiral-like trajectory in contrast to the
usual descent. To endeavour to clarify the issue the perturbative analysis of the Landau–
Ginzburg–Wilson critical behaviour has recently been extended to a higher loop order in
[11]. Previous one and two loop computations were carried out in [6–8]. The new three
loop MS calculations of the renormalization group functions such as the β-functions of
both coupling constants and anomalous dimensions, [11], have provided more accurate
information on the fixed point locations and the range of parameters for which they exist
and are stable or not. Indeed in this respect the models with the more general symmetry
group of O(N) × O(m) were studied with m only set to m = 2 at the end, [11]. Such three
loop calculations represent the current perturbative status of the model.
However, it is in principle possible to extend this three loop MS dimensionally
regularized calculation to the next order, though it will involve a huge number of Feynman
diagrams. In previous work in the simpler O(N) models the perturbative computations
were complemented with large N calculations of the same renormalization group functions
to several orders in powers of 1/N . For instance, various critical exponents are available
at both O(1/N 2 ) and O(1/N 3 ), [23–26], as functions of d, with 2 < d < 4, which
correspond through the critical renormalization group equation with the renormalization
group functions. More correctly these critical exponents were computed in d-dimensions
at the nontrivial Wilson–Fisher fixed point of the d-dimensional β-function which
corresponds to the Heisenberg fixed point in three dimensions in O(N) φ 4 theory or the
O(N) nonlinear sigma model which does lie in the same universality class. The coefficients
of the powers of in the -expansion of such exponents in d = 4 − 2 dimensions are in
exact agreement with the perturbative coefficients in the renormalization group function
to the perturbative order they are known, at a particular order in 1/N . More significantly
the large N critical exponents contain new higher order information in the uncomputed
coefficients which would therefore assist future perturbative calculations. Indeed in the
Landau–Ginzburg–Wilson context various critical exponents have already been computed
in the model with O(N) × O(m) symmetry at the two nontrivial fixed points which exist
in addition to the Heisenberg fixed point, [6,11]. These are known as the chiral stable, (CS),
and antichiral unstable, (AU), fixed points. Moreover, the results for the critical exponents η
and ν at O(1/N 2 ) at both CS and AU are in agreement with the new perturbative results of
J.A. Gracey / Nuclear Physics B 644 [FS] (2002) 433–450 435
[11]. However, to extract the new information encoded in the exponents at four and higher
loops in these d-dimensional functions in relation to the four-dimensional theory one
requires knowledge of the location of the fixed points to the same loop and large N order.
From the critical renormalization group equation such information is encoded in the critical
exponent ω which relates to the critical β-function slope of the model in the universality
class which is renormalizable in four dimensions. In the Landau–Ginzburg–Wilson model
this has not yet been computed at O(1/N) at either CS or AU. Therefore, it is the purpose
of this article to rectify this gap and thereby unlock the door to higher order information on
the structure of the perturbative renormalization group equations such as the β-function.
In O(N) φ 4 theory this problem has already been resolved at O(1/N 2 ), [26], where the
O(1/N) value for ω is relatively trivial to establish, [27], with the elegant machinery
of the large N critical point method of [23,24]. However, for the CS and AU Landau–
Ginzburg–Wilson fixed points the leading order, O(1/N), analysis is much more involved
since one is studying a model with two independent coupling constants. Therefore, whilst
our calculation also opens the road to an O(1/N 2 ) computation, it represents a nontrivial
example of how one treats the large N formalism for ω exponents explicitly in a quantum
field theory with more than one coupling constant which deserves detailed treatment.
The paper is organized as follows. In Section 2 we recall the background details of
the model we are interested in and derive explicit expressions for the location of the
various fixed points from the explicit three loop perturbative results at O(1/N) as well
as the perturbative values of the eigenexponents of the stability matrix at criticality. These
are related to the exponents ω which we are interested in. Section 3 is devoted to the
development of the large N formalism for computing these various ω and the explicit
d-dimensional expressions are given at O(1/N). Various concluding remarks are given in
Section 4.
2. Background
The Lagrangian of the massless Landau–Ginzburg–Wilson model involves a scalar field

with two quartic self interaction terms with an O(N) × O(m) symmetry and is given by
1 2 u ai ai 2 v ai bi 2
L= ∂µ φ ai + φ φ + φ φ − φ ai φ ai φ bj φ bj , (2.1)
2 4! 4!
where 1 i N , 1 a m, and u and v are the bare coupling constants. As in [11]
we rewrite (2.1) in order to perform the large N expansion. This involves introducing two
auxiliary scalar fields one of which, T ab , is a symmetric traceless tensor under O(m). Thus
(2.1) is equivalent to
1 1 1 3σ 2 3
L = (∂µ φ ai )2 + σ φ ai φ ai + T ab φ ai φ bi − − T ab T ab , (2.2)
2 2 2 2w 2v
where w = u + (m − 1)v/m in our notation. If one uses the equations of motion for σ
and T ab then (2.1) is recovered. The coupling constants are defined in the kinetic terms
to ensure that the vertices are in the right form for applying the uniqueness technique
to compute the large N Feynman diagrams [23,28]. One can understand the fixed point
structure of (2.1) by considering the β-functions for each coupling which have been
computed to several orders, [8,11]. At three loops these are
βu (u, v)

1 (mn + 8) 2 1 v
= (d − 4)u + u − (m − 1)(n − 1)v u −
2 6 3 2

1 11 13 5
− (3mn + 14)u3 + (m − 1)(n − 1) u2 − uv + v 2 v
6 9 12 18

u4
+ 33m2 n2 + 922mn + 2960 + ζ(3)(480mn + 2112)
432

− 4 79mn + 1318 + 768ζ(3) u3

− 555mn − 460(m + n) + 6836 + 4032ζ(3) u2 v

+ 2 213mn − 358(m + n) + 1933 + 960ζ(3) uv 2

(m − 1)(n − 1)v
− 121mn − 309(m + n) + 817 + 216ζ(3) v 3
864

+ n3 [a1 − a2 − a3 − a4 − a5 − a6 ]u5 + a2 u4 v + a3 u3 v 2

+ a4 u2 v 3 + a5 uv 4 + a6 v 5

6 1
+O u ; 2 (2.3)
N
and
βv (u, v)
1 1 1
= (d − 4)v + 2uv + (m + n − 8)v 2 − (5mn + 82)u2 v
2 6 18
1 2 1
+ 5mn − 11(m + n) + 53 uv − 13mn − 35(m + n) + 99 v 3
9 36

+ 52m2 n2 − 57mn(m + n) − 2206mn

− 111 m2 + n2 + 4291(m + n) − 8084
v 4
− ζ (3) 1416mn − 3216(m + n) + 7392
864

− 39m2 n2 − 35mn(m + n) − 1302mn − 36 m2 + n2 + 2401(m + n)
v 3 u
− 5725 − ζ (3) 768mn − 1824(m + n) + 4896
216

+ 78m2 n2 − 35mn(m + n) − 2114mn + 3182(m + n) − 12520
u2 v 2
− ζ (3) 1152mn − 2304(m + n) + 10368
432

u3 v
− 13m n − 368mn − 3284 − ζ(3)(192mn + 2688)
2 2
216

+ N (b2 − b3 − b4 − b5 − b6 )u v + b3 u v + b4 u v + b5 uv 4 + b6 v 5
3 4 3 2 2 3

1
+ O u6 ; 2 , (2.4)
N
where we have rescaled the coupling constants by a numerical factor to ensure the
expressions are in the correct format for comparing with the Heisenberg large N value
for ω and the ones we compute here. Also we have included the d-dependent terms as we
are interested in the fixed point structure in d-dimensions. To assist with determining new
information on the four loop structure of both β-functions at O(1/N) we have introduced
parameters, {ai } and {bi }, for the coefficients of the possible terms.
By examining the solutions to βu = 0 and βv = 0, [8,11], several fixed points emerge.
First, there are the two obvious ones of the Gaussian fixed point, uc = 0, vc = 0, and the
Heisenberg fixed point, uc = 0, vc = 0. For the latter point setting v = 0 and m = 1 in
(2.3) one recovers the usual O(N) symmetric φ 4 theory whose β-function is known at five
loops in MS in four dimensions, [2]. Indeed in our choice of parametrization of βv (u, v)
at four loops we used this fact to restrict the function to be proportional to v. These two
fixed points clearly lie on the axes of the (u, v) coupling plane. However, for a range of
values of N and m there are two other fixed points which both have uc = 0 and vc = 0.
One is known as CS and the other AU. The ultraviolet renormalization group flow of the
four fixed points in the (u, v) plane is shown graphically in Fig. 1. The range of values for
N and m for which such a renormalization group flow is present has been detailed in [8,
11], for example. However, for the purposes of the large N calculation we will require the
values of uc and vc to several order in and powers of 1/N where we take the convention
Fig. 1. Renormalization group flow on the (u, v) plane illustrating the Gaussian (G), Heisenberg (H), stable chiral
(CS) and unstable antichiral (AU) fixed points.
d = 4 − 2 . In [8,11] the explicit functions of N and m were presented at two loops. The
full expressions at three loops can also be derived but are large, [11], and the full form is
not necessary for our purposes. Indeed if we write
∞
ur1 ur2 r 1
uc = + 2 +O ,
N N N3
r=1
∞
vr1 vr2 r 1
vc = + 2 + (2.5)
N N N3
r=1
for the 1/N expansion of the critical couplings in powers of the four fixed points are
determined to O( 4 ) and O(1/N 2 ) as follows. For the Gaussian fixed point uri = vri
= 0. At the Heisenberg fixed point u11 = 6/m, ur1 = 0 for r = 1, u12 = − 48/m2 ,
u22 = 108/m2, u32 = − 99/m2 , u42 = − 7776(a1 − a2 − a3 − a4 − a5 − a6 )/m5 and
vri = 0. At the stable fixed point, CS,
u11 = 6, u21 = 0, u31 = 0, u41 = 0, u12 = −6m − 42,

1
u22 = 18m + 90, u32 = − (39m + 159), u42 = −7776a1,
2
v11 = 6, v21 = 0, v31 = 0,
v41 = 0v12 = −6m − 24, v22 = 18m + 54,
1
v32 = − (39m + 99), v42 = −7776b2. (2.6)
2
Finally, at the unstable fixed point, AU,
6(m − 1)
u11 = , u21 = 0, u31 = 0,
m
108 96
u12 = −6m − 6 + − 2,
m m
282 252
u22 = 18m + 12 − + 2,
m m
39m 513 171
u32 = − − 66 + − 2,
2 2m m
7776
u42 = 5 (m − 1) a1 + (m − 1)4 a2 + (2m − 1)(m − 1)3 a3
5
m

+ 3m2 − 3m + 1 (m − 1)2 a4 + (2m2 − 2m + 1)(2m − 1)(m − 1)a5

+ 5m4 − 10m3 + 10m2 − 5m + 1 a6 − 2(m − 1)5 b2 − 2(m − 1)4 b3

− 2(2m − 1)(m − 1)3 b4 − 2 3m2 − 3m + 1 (m − 1)2 b5

− 2 2m2 − 2m + 1 (2m − 1)(m − 1)b6 ,
72
v11 = 6, v21 = 0, v31 = 0, v12 = −6m − 24 + ,
m
204 39m 99 243

v22 = 18m + 54 − , v32 = − − + ,
m 2 2 m
7776
v42 = − (m − 1)4 b2 + (m − 1)3 b3 + (2m − 1)(m − 1)2 b4
m4

+ 3m2 − 3m + 1 (m − 1)b5 + 2m2 − 2m + 1 (2m − 1)b6 . (2.7)
These agree to two loops with the expressions given in [8,11]. As we will be computing
the critical exponents ω which relate to the critical slope of the β-functions in large N we
can use these values to determine the O(1/N) form of the critical exponents. As we are
working with a two coupling model the stability exponents of each fixed point are related
to the eigenvalues, λI , of the matrix of derivatives, Ω(u, v), evaluated at the appropriate
fixed point where
∂βu (u,v) ∂βu (u,v)
Ω(u, v) = ∂βv∂u (u,v)
∂v
∂βv (u,v) . (2.8)
∂u ∂v
For the Gaussian case this is trivial and will not concern us here. For the Heisenberg fixed
point Ω(u, v) becomes triangular because βv (u, v) has no terms involving only u at any
order which implies

∂βv (u, v)
Heis
= 0. (2.9)
∂u
uc ,vc
Thus, the critical point eigenvalues in the Heisenberg case are1

1
λHeis
+ = − 18 2 − 33 3
mN

3888 1
− 3 [a1 − a2 − a3 − a4 − a5 − a6 ] 4 + O 5 O ,
m N2

1 4 1
λHeis
− = − + 12 − 10 2
− 13 3
+ O + . (2.10)
mN N2
One of these corresponds to the stable direction in the renormalization group flow, as
indicated in Fig. 1, whilst the other corresponds to the unstable direction which will
not concern us here. It is for the former for which the corrections at O(1/N 2 ) have
been computed in d-dimensions, [26], and we will record the O(1/N) value later. For
the remaining two fixed points the critical matrix Ω(uc , vc ) is not triangular. However,
computing the eigenvalues at criticality of (2.8) we find

1 3(m2 + 4m + 7) 2 (13m4 + 72m3 + 202m2 + 216m + 25) 3
λCS
+ = − −
N (m + 1) 2(m + 1)3
1 It is worth stressing that our convention, d = 4 − 2 , and the form we take for the β-functions, (2.3) and
(2.4), implies that for a stable direction the eigenexponent is (4 − d)/2 + O(1/N ). This differs from the standard
result for the stability eigenexponent of (4 − d) + O(1/N ) by a factor of 2 which ought to be taken into account
when comparing with other calculations. (See, for example, [2].)

− 4 1944(m + 1)4 a1 + 972(m − 1)(m + 1)4 b2

4 5
− 2m + m − 13 (m + 2)(m − 1)
2
+O
(m + 1)5

1
+O ,
N2

1 (13m2 + 26m + 25) 2
λCS
− = − 6(m + 1) −
N (m + 1)
(3m4 + 12m3 + 82m2 + 116m − 5) 3
+
2(m + 1)3

+ 4 324(2m − 1)(m − 1)(m + 1)4 a1
+ 324(m + 1)5 a2 + 648(m + 1)5 a3
+ 972(m + 1)5 a4 + 1296(m + 1)5 a5 + 1620(m + 1)5 a6
− 648m(m − 2)(m + 1)4 b2 − 324(m + 1)5 b3
− 648(m + 1)5 b4 − 972(m + 1)5 b5 − 1296(m + 1)5 b6

4 5
− 2m2 + m − 13 (m + 2)(m − 1) + O
(m + 1)5

1
+O ,
N2

1 2 1
λAU
+ = − 3m + 9m − 34 2 − 13m2 + 33m − 162 3
mN 2

− 3888 (m − 1)4 b2 + (m − 1)3 b3 + (2m − 1)(m − 1)2 b4

+ 3m2 − 3m + 1 (m − 1)b5
4
+ (2m − 1) 2m2 − 2m + 1 b6 3
m

5 1
+O +O ,
N2

(m − 1)(m + 2) 3 3 4 1
λAU
− = − + 6 − 13 2
+ + O + O , (2.11)
mN 2 N2
where the sign of the O( ) terms relates to the stability property of the fixed point
when viewed from the ultraviolet renormalization group flow of Fig. 1. For the exponents
corresponding to the stable direction we have included the O( 4 ) terms which depend on
the unknown parameters of the O(1/N) four loop β-functions of (2.3) and (2.4) and which
will be constrained by our O(1/N) critical exponents.
3. Large N formalism
To compute the same critical eigenexponents in the large N formalism in d-dimensions

one follows the programme of [23,24] where the appropriate Schwinger–Dyson equations
are analyzed at the respective d-dimensional fixed points of the theory. At this point the
propagators scale asymptotically to power law behaviour. In other words, in coordinate
space, we have
B
φ ai,bj (x) = δ ab δ ij φ(x), σ (x) ∼ , σTab,cd (x) = Xab,cd σT (x), (3.1)
(x 2 )β
where
A C
φ(x) ∼ , σT (x) ∼ (3.2)
(x 2 )α (x 2 )γ
and

1 ac bd 2
Xab,cd = δ δ + δ ad δ bc − δ ab δ cd . (3.3)
2 m
This group theory factor satisfies the projector property
Xab,pq Xpq,cd = Xab,cd (3.4)

which implies that in labeling the internal indices on a diagram one only needs to put
indices on the φ-field lines. The quantities A, B and C are the x-independent amplitudes
of the fields and the exponents, α, β and γ in our notation, are related to the wave function
anomalous dimension η by
1
α = µ − 1 + η, β = 2 − η − χ, γ = 2 − η − χT (3.5)
2
where χ and χT are the respective anomalous dimensions of the vertices involving σ and
T ab and d = 2µ. In [11] η was computed at O(1/N 2 ) at both fixed points CS and AU.
For the stable chiral one both the σ and T ab fields propagate and couple. However, at the
Heisenberg fixed point only the σ field propagates since clearly the T ab field is absent in
the usual formulation of φ 4 theory. Interestingly at the unstable antichiral fixed point the
opposite situation emerges in that only the T ab field is present and the σ field is omitted
from the calculation of the large N exponents, [11]. Since the amplitudes appear in the
calculations in the combinations z = A2 B and y = A2 C throughout and as they will be
required for computing our ω exponents we have determined their values at O(1/N). For
reference, at CS they are
2Γ (2µ − 2)
z1 = − , y1 = mz1 (3.6)
mΓ (2 − µ)Γ (µ − 2)
and at AU we have
2Γ (2µ − 2)
z1 = 0, y1 = − . (3.7)
Γ (2 − µ)Γ (µ − 2)
Moreover, for completeness the exponent η at O(1/N) is given by
∞
ηi
η= , (3.8)
Ni
i=1
where
2(m + 1)Γ (2µ − 2)
η1CS = − ,
Γ (µ + 1)Γ (µ − 1)Γ (µ − 2)Γ (2 − µ)
2(m − 1)(m + 2)Γ (2µ − 2)
η1AU = − . (3.9)
mΓ (µ + 1)Γ (µ − 1)Γ (µ − 2)Γ (2 − µ)
Any subsequent exponent at either fixed point will be expressed in terms of their respective
value for η1 .
Since the exponents ωI relate to corrections to scaling then to compute them one
considers corrections to the asymptotic scaling forms (3.2), [23,24,26]. In coordinate space
we take
A ω B ω
φ(x) ∼ 2 α 1 + A x 2 , σ (x) ∼ 2 β 1 + B x 2 ,
(x ) (x )
C
σT (x) ∼ 2 γ 1 + C x 2
ω
(3.10)
(x )
where A , B and C are the x-independent correction to scaling amplitudes whose
values are not important here. In addition to (3.10) one requires the scaling form of the
inverse propagators which are determined by inverting the Fourier transform of (3.10) in
momentum space. Thus
p(α) ω
φ −1 (x) ∼ 1 − q(α, ω) x 2 A ,
(x 2 )2µ−α A
p(β) ω
σ −1 (x) ∼ 1 − q(β, ω) x 2 B ,
(x 2 )2µ−β B
p(γ ) ω
σT−1 (x) ∼ 1 − q(γ , ω) x 2 C , (3.11)
(x 2 )2µ−γ C
where the functions p(x) and q(x, y) are defined by
a(x − µ) a(x − y)a(x + y − µ)
p(x) = , q(x, y) = (3.12)
a(x) a(x)a(x − µ)
with a(x) = Γ (µ − x)/Γ (x). Further, in our notation each exponent ωI in the stable
direction will have the 1/N expansion
∞
ωi
ω = (µ − 2) + (3.13)
Ni
i=1
Fig. 2. Schwinger–Dyson equation for φ field at O(1/N ).
Fig. 3. Basic topologies for the corrections to the σ and T ab Schwinger–Dyson equations to determine ω at
O(1/N ).
and are therefore related to the eigenexponents of Ω(uc , vc ) by ωI = − λI respectively in

our conventions. Moreover, as noted before our choice of ω differs from the usual definition
by a factor of (−1/2).
To solve for ω1 one substitutes the asymptotic scaling values into the lines of the
Schwinger–Dyson equations of each of the fields. These are illustrated in Figs. 2 and 3.
In the latter we have only indicated the possible topologies that arise due to the nature of
the expansion in powers of 1/N and each wiggly line represents all possible combinations
of σ and T ab fields. If we ignore for the moment the two and three loop corrections in
Fig. 3 we find the equations are represented by
ω ω
0 = p(α) 1 − q(α, ω) x 2 A + z 1 + (A + B ) x 2
(m − 1)(m + 2)y ω
+ 1 + (A + C ) x 2 ,
2m
ω 1 ω
0 = p(β) 1 − q(β, ω) x 2 B + Nmz 1 + 2 x 2 A ,
2
2 ω 1 ω
0 = p(γ ) 1 − q(γ , ω) x C + Ny 1 + 2 x 2 A (3.14)
2
at the CS fixed point which we consider first for illustration. These equations decouple
on dimensional grounds into a set which determine η at CS and a set involving ω. The
former lead to the values for η1 , z1 and y1 quoted above. For the latter set of equations for
consistency the determinant of the 3 × 3 matrix defined with respect to the basis vector
{A , B , C } has to vanish which naively leads to the equation
(m − 1)(m + 2)
0= yq(β, ω)
m
(m − 1)(m + 2)
− q(γ , ω) 2z − 1 + q(α, ω) q(β, ω) z + y . (3.15)
2m
This is similar to the consistency equation which determines the exponent denoted by λ
in [23,24] where the correction to scaling exponent has canonical dimension λ0 = µ − 1.
However, for this value q(α, λ0 ) with α = (µ − 1) or 2 canonically is always an O(1)
quantity. By contrast, for the four-dimensional model q(α, ω0 ) with ω0 = (µ − 2) and α =
(µ − 1) or 2 it has values which are O(N) and O(1/N) respectively and therefore affects
the consistency equation. For instance, whilst the product q(α, ω)q(β, ω)q(γ , ω) in (3.15)
will be O(1/N), this equation, (3.15), would give an incorrect value for the exponents
since the contributions in (3.15) from σ −1 and σT−1 are of the same order in 1/N as those
from the two and three loop correction graphs of Fig. 3 which we had naively omitted.
Thus to determine ω1 correctly one must include the contributions from these diagrams to
the B and C parts of the Schwinger–Dyson representation in the asymptotic approach to
the fixed points. Hence, the σ and T ab equations in (3.14) are modified to
ω 1 ω
0 = p(β) 1 − q(β, ω) x 2 B + Nmz 1 + 2 x 2 A
2
1 1
+ Nmz Π1 B + (m + 2)(m − 1)NyzΠ1 C
2
2 4

1 ω
+ N 2 m2 z3 Π2 B + (m + 2)(m − 1)N 2 y 2 zΠ2 C x 2 ,
2
2 ω 1 ω
0 = p(γ ) 1 − q(γ , ω) x C + Ny 1 + 2 x 2 A
2
1 (m − 2) 2
+ NyzΠ1 B + Ny Π1 C + N 2 y 2 zΠ2 B
2 4m

(m − 2)(m + 4) 2 3 ω
+ N 2 y 2 zΠ2 C + N y Π2 C x 2 . (3.16)
4m
In these equations to ensure the correct contribution to the ω Schwinger–Dyson equation
after decoupling one correction term, (x 2 )ω , is on one internal σ or T ab line. There are
corrections to the φ lines but these only contribute to ω2 after examining where they appear
in the consistency determinant. This feature of having to include higher order graphs to
obtain the correct ω1 is not peculiar to the CS fixed point as it already occurs at the
Heisenberg point, [26]. With these additional diagrams the correct consistency equation
is given by setting the determinant of the matrix
 1 + q(α, ω)z + (m−1)(m+2) y z (m−1)(m+2)
y 
2m 2m
 2 q(β, ω) + zΠ1 + 2mN z2 Π2 (m−1)(m+2)
y(Π1 + 2NyΠ2 ) 

2m 
 2 z(Π1 + 2NyΠ2 ) q(γ , ω) + (m−2)(m+4) Ny 2 Π2 
2m

+ (m−2)
2m yΠ1 + 2NyzΠ2
(3.17)
to zero.
Therefore, to solve for the respective ω’s from (3.17) the values of these additional
diagrams must be computed. As the graphs are similar to those used in the O(N) φ 4
calculation of ω if the group theory of each graph is suppressed we merely quote the
d-dimensional values for the respective two and three loop graphs of Fig. 3. They are,
ignoring symmetry factors,
2
Π1 = ν(2, µ − 1, µ − 1) ,
2
Π2 = ν(2, µ − 1, µ − 1) ν(1, 2, 2µ − 3)ν(4 − µ, µ − 1, 2µ − 3), (3.18)
where ν(x, y, z) = a(x)a(y)a(z). These were calculated using the method of uniqueness
of [28] which was extended from three dimensions to d-dimensions in [23,24]. Further,
since the O(1/N) values of q(β, ω) and q(γ , ω) now need to be included it transpires that
the O(1/N) expression for the vertex anomalous dimensions, χ and χT , are required. This
is due to

a(4 − µ)Γ (µ) [ω1 − η1 − χ1 ] 1
q(β, ω) = +O . (3.19)
a(2)a(2 − µ) N N2
We have computed each vertex anomalous dimension at each fixed point and find
µ(4µ − 5)η1CS
χ1CS = − ,
(µ − 2)
µ[(2µ − 3)m + (4µ − 5)]η1CS
,1 = −
χTCS ,
(µ − 2)(m + 1)
µ(m − 2)[(m + 4)(2µ − 3) + 1]η1AU
,1 = −
χTAU . (3.20)
(m − 1)(m + 2)(µ − 2)
The expression for χ1CS is formally the same as that for the O(1/N) vertex dimension at
the Heisenberg fixed point. For each exponent we have computed the value by applying
the technique of [29] of large N critical point renormalization of 3-point functions to each
vertex at the appropriate fixed point. For χ we have checked that the value agrees with that
given by the scaling law which emerges from the theory which is in the same universality
class as (2.2). This is believed to be the O(N) × O(m) two-dimensional nonlinear sigma
model where instead of the term quadratic in σ of (2.2) one has a term linear in σ where its
coupling constant is related to the critical exponent ν. However, in such a model it is clear
from group theory that there can be no linear term in T ab and therefore both expressions
for χT cannot be derived from a scaling law but only direct calculation.
With these values we can now determine the solution for the consistency equation for
each fixed point. For CS since the matrix is 3 × 3 two values for ω emerge. These are
(2µ − 1)η1CS
CS
ω± = (µ − 2) +
2(m + 1)(µ − 2)N

× m(µ − 1)(µ − 4) + 2µ2 − 7µ + 4
1/2
± µ (m2 − 1)(µ − 1)2 + 2(m − 1)(2µ − 3)(µ − 1) + (5µ − 8)2

1
+O , (3.21)
N2
where the ± subscript refers to the sign in front of the discriminant. Both are perfectly
acceptable solutions since one corresponds to one stable direction at CS and the other to
the eigenexponent from the second direction.
For AU the derivation of the consistency equation follows the same pattern as that for
CS in that two and three loop graphs have also to be included due to the large N counting.
The consistency equation itself can be deduced from (3.17) by deleting the second row and
column from the matrix and setting z = 0 in the remaining entries. Since only the T ab field
is present only one value for ω emerges which is

µ(m − 2)[2µ − 5 − 2(m + 4)(2µ − 3)] η1AU
ω+AU
= (µ − 2) − 2µ2 − 3µ − 1 +
(m − 1)(m + 2) N

1
+O (3.22)
N2
which corresponds to the eigenexponent of the stable direction at AU. For completeness
we note that at the Heisenberg fixed point for the stable direction we have, [26,27],

4(2µ − 1)2 Γ (2µ − 2) 1
ω+ = (µ − 2) −
Heis
+O (3.23)
Γ (2 − µ)Γ (µ − 1)Γ (µ − 2)Γ (µ + 1)mN N2
which is deduced by deleting the third row and column of (3.17) and setting y = 0.
In order to check the correctness of each result one can set d = 4 − 2 , or µ = 2 − ,
and expand each exponent to O( 3 ) and compare with the values of explicit perturbation
theory, (2.11). We find total agreement for all three exponents and therefore regard (3.21)
and (3.22) as correct. Indeed equipped with (3.21) and (3.22) we can now reverse the check
argument and use the information contained in the d-dimensional exponents to determine
constraints on the structure of the four loop MS corrections to (2.3) and (2.4). First, we
record the expansion of each of the exponents in the stable directions which are consistent
with (2.11). From (3.21), (3.22) and (3.23) we have

3(m2 + 4m + 7) 2 (13m4 + 72m3 + 202m2 + 216m + 25) 3
ω+,1 =
CS
−
(m + 1) 2(m + 1)3

(3m6 + 10m5 + 21m4 − 36m3 − 387m2 − 326m + 395) 4 5
+ + O ,
4(m + 1)5

(13m2 + 26m + 25) 2 (3m4 + 12m3 + 82m2 + 116m − 5) 3
CS
ω−,1 = 6(m + 1) − +
(m + 1) 2(m + 1)3
[48(m + 1)6 ζ (3) + 3m6 + 18m5 + 29m4 + 76m3 + 397m2 + 322m − 397] 4
+
4(m + 1)5

+ O 5 ,

1 1 1
ω+,1 =
AU
(3m2 + 9m − 34) 2 − (13m2 + 33m − 162) 3 + (3m2 − 5m − 54) 4
m 2 4

5
+O ,

1 5 4 5
ω+,1 =
Heis
18 − 33 − + O .
2 3
(3.24)
m 2
Next we compare the O( 4 ) coefficients with those of the explicit perturbative expansion
which have been parametrized by {ai } and {bi }. From the Heisenberg exponent we have
5m2
a1 − a2 − a3 − a4 − a5 − a6 = (3.25)
7776
which agrees with the explicit four loop φ 4 computation when one sets m = 1 and
determines the O(1/N) coefficient of the u5 term of βu (u, v). From the CS fixed point
we have
(3m + 7)(m − 3)
2a1 + (m − 1)b2 = − (3.26)
15552
and

(m + 1) a2 + 2a3 + 3a4 + 4a5 + 5a6 − b3 − 2b4 − 3b5 − 4b6
+ (2m − 1)(m − 1)a1 − 2m(m − 2)b2
ζ (3) (3m2 + 6m + 19)
= + . (3.27)
108(m + 1)3 5184(m + 1)5
The final constraint arises from the stable direction at the AU fixed point which gives
(m − 1)4 b2 + (m − 1)3 b3 + (2m − 1)(m − 1)2 b4 + (3m2 − 3m + 1)(m − 1)b5
+ (2m − 1)(2m2 − 2m + 1)b6
[3m2 − 5m − 54]
=− . (3.28)
15552
One can now deduce the values for the exponents in the four stable directions in three
dimensions. These are, with our conventions,

1 4(m + 4) 1
ω+CS
=− + + O ,
2 3π 2 N N2

1 16(m + 1) 1
ω−CS
=− + 2
+O ,
2 3π N N2

1 4[m2 + 4m − 8] 1
ω+AU
=− + + O ,
2 3π 2 mN N2

1 32 1
ω+Heis
=− + + O . (3.29)
2 3π 2 mN N2
For CS the corrections both have the same sign and interestingly neither involves a square
root which appears in the d-dimensional expression. For the specific case of m = 2 we
have from (3.29)

CS
1 8 1
ω+ m=2 = − + 2 + O ,
2 π N N2

CS
1 16 1
ω− m=2
=− + 2 +O ,
2 π N N2

AU
1 8 1
ω+ m=2
=− + +O ,
2 3π 2 N N2

Heis
1 16 1
ω+ m=2
= − + + O . (3.30)
2 3π 2 N N2
With these values we can comment on the possible breakdown of stability in this model. In
our computations so far have relied on the fact that the stability picture for the fixed points
represented in Fig. 1 is valid for the full range of N in the large N expansion. However, it
has been suggested that for certain values of N this scenario may be different and that
the underlying assumption of the existence of a second order phase transition, around
which the large N critical point formalism is built, could break down. Indeed there is
some controversy in this model in three dimensions about the existence of CS and whether
the phase transition is first or second order for relatively low values of N . Moreover, the
precise value of N where the order changes has not been determined consistently from
different methods. (A recent review is given in [17].) For instance, a value has been found
for this critical value of N by using (d = 4 − 2 )-dimensional perturbation theory and
it was extended to three dimensions using standard resummation techniques, [30], giving
Nc = 3.39. By contrast, Monte Carlo methods and another -expansion extraction have
suggested Nc < 2, [6–8]. With the large N corrections (3.30) we can naively examine the
range of values of N for which either CS stability exponent changes sign. For ω− CS
|m=2
this will occur when Nc = 3.24 whilst ω+ |m=2 changes sign when Nc = 1.62. The
CS
former would suggest a critical N in a similar range to that of [30]. However, these
remarks ought to be qualified with various observations. First, we have only computed
the O(1/N) correction to stability where N is assumed to be large. Therefore, one has to
ask whether the approximation will still be valid for such a low value of N . Second, the
nature of the large N expansion is a reordering of perturbation theory such that a certain
class of diagrams are summed first. Therefore, if one could compute to all orders one would
reproduce ordinary perturbation theory and so obtaining a value for Nc in three dimensions
which is not inconsistent with the resummed value determined from several loop orders in
ordinary perturbation theory would seem only to reinforce that particular value. In other
words if nonperturbative effects become significant at CS for low values of N to affect the
precise location of Nc these will have been omitted in perturbation theory. Third, for our
value of Nc we have naively assumed the large N series is convergent and therefore that
our simple assumption that when the O(1/N) correction exceeds 1/2 the character of the
fixed point changes is valid. Only a higher order calculation would resolve this.
4. Discussion
We have computed the correction to scaling exponents ω in all the stable directions
of the Landau–Ginzburg–Wilson model with O(N) × O(m) symmetry at O(1/N) in
d-dimensions. This allows one to extract information in all the available large N exponents
of [11] at O(1/N) in relation to the four-dimensional theory in the same way that the two-
dimensional critical slope information contained in the exponent ν does for the underlying
two-dimensional theory. It is also worth stressing again, [11], that from the point of view
of the large N formalism a consistent picture emerges in terms of the active fields of the
theory formulated in terms of the auxiliary fields σ and T ab of (2.2). At the Heisenberg and
AU fixed points only σ and T ab , respectively, propagate which corresponds in the large
N formalism developed here to one stable direction and hence only one eigenexponent
emerged. However, at the only fully stable fixed point both fields are relevant leading
to two independent stability exponents. This is a natural way to picture this particular
model which we assume persists to higher orders in large N . Moreover, our consistency
with three loop MS perturbation theory provides an important internal cross check on the
values of quantities, such as the vertex anomalous dimensions, which we had to compute en
route to our expressions for ωi . Further, the information contained within the expressions
(3.21), (3.22) and (3.23) will provide important checks on any future explicit four loop MS
perturbative calculations which would improve the accuracy of the numerical estimates
deduced from (2.3), (2.4) and other renormalization group functions. Such four loop
calculations are certainly viable since five loop results are available in MS in ordinary
O(N) φ 4 theory. For example, the integration routines for the four loop Feynman diagrams
have already been constructed. However, one can also attack this problem from the large
N point of view. For instance, we have demonstrated the elegance of the formalism to
produce the critical eigenexponents at O(1/N). However, this machinery has already been
extended in [26] to compute ω2 in the Heisenberg model. Therefore, we would expect
there to be no serious obstacles to extending the present computation to find ω2CS and ω2AU .
For example, the values of the underlying three, four, five and six loop Feynman diagrams
which are analogous to the corrections of Fig. 3 for the O(1/N 2 ) calculation have been
determined. We hope to return to this in a future article.
References
[1] A. Pelissetto, E. Vicari, cond-mat/0012164 .

[2] H. Kleinert, J. Neu, V. Schulte-Frohlinde, K.G. Chetyrkin, S.A. Larin, Phys. Lett. B 256 (1991) 81;
H. Kleinert, J. Neu, V. Schulte-Frohlinde, K.G. Chetyrkin, S.A. Larin, Phys. Lett. B 319 (1993) 545.
[3] G.A. Baker Jr., B.G. Nickel, M.S. Green, D.I. Meiron, Phys. Rev. Lett. 36 (1977) 1351.
[4] G.A. Baker Jr., B.G. Nickel, D.I. Meiron, Phys. Rev. B 17 (1978) 1365.
[5] S.A. Antonenko, A.I. Sokolov, Phys. Rev. E 51 (1995) 1894.
[6] H. Kawamura, Phys. Rev. B 38 (1988) 4916.
[7] H. Kawamura, Phys. Rev. B 42 (1990) 2610.
[8] H. Kawamura, J. Phys. Soc. Jpn. 59 (1990) 2305.
[10] H. Kawamura, J. Phys. C 10 (1998) 4707.
[11] A. Pelissetto, P. Rossi, E. Vicari, Nucl. Phys. B 607 (2001) 605.
[12] P. Azaria, B. Delamotte, Th. Jolicœur, Phys. Rev. Lett. 64 (1990) 3175.
[13] T. Bhattacharya, A. Billoire, R. Lacaze, Th. Jolicœur, J. Phys. 4 (1994) 181;
M.L. Plumer, A. Mailhot, Phys. Rev. B 50 (1994) 16113;
A. Mailhot, M.L. Plumer, A. Caillé, Phys. Rev. B 50 (1994) 6854;
D. Loison, H.T. Diep, Phys. Rev. B 50 (1994) 16453;
E.H. Boubcheur, D. Loison, H.T. Diep, Phys. Rev. B 54 (1994) 4165.
[17] H. Kawamura, Can. J. Phys. 79 (2001) 1447.
[18] D. Loison, A.I. Sokolov, B. Delamotte, S.A. Antonenko, K.D. Schotte, H.T. Diep, JETP Lett. 72 (2000) 337.
[19] A. Pelissetto, P. Rossi, E. Vicari, Phys. Rev. B 63 (2001) 140414.
[20] M. Tissier, B. Delamotte, D. Mouhanna, Phys. Rev. Lett. 84 (2000) 5208.
[21] M. Itakura, cond-mat/0110306.
[22] P. Calabrese, P. Parruccini, A.I. Sokolov, cond-mat/0205046.
[23] A.N. Vasil’ev, Yu.M. Pis’mak, J.R. Honkonen, Theor. Math. Phys. 46 (1981) 157.
[26] D.J. Broadhurst, J.A. Gracey, D. Kreimer, Z. Phys. C 75 (1997) 559.

[27] A.N. Vasil’ev, A.S. Stepanenko, Theor. Math. Phys. 95 (1992) 471.
[28] M. d’Eramo, L. Peliti, G. Parisi, Lett. Nuovo Cimento 2 (1971) 878.
[29] A.N. Vasil’ev, M.Yu. Nalimov, Theor. Math. Phys. 56 (1982) 643.
[30] S.A. Antonenko, A.I. Sokolov, K.B. Varnashev, Phys. Lett. A 208 (1995) 3785.
Fermion Schwinger’s function for the

SU(2)-Thirring model
Benjamin Doyon a,∗ , Sergei Lukyanov a,b
a Department of Physics and Astronomy, Rutgers University, Piscataway, NJ 08855-0849, USA
b L.D. Landau Institute for Theoretical Physics, Kosygina 2, Moscow, Russia
Received 16 May 2002; received in revised form 29 August 2002; accepted 4 September 2002
Abstract
We study the Euclidean two-point function of Fermi fields in the SU(2)-Thirring model on the
whole distance (energy) scale. We perform perturbative and renormalization group analyses to obtain
the short-distance asymptotics, and numerically evaluate the long-distance behavior by using the
form factor expansion. Our results illustrate the use of bosonization and conformal perturbation
theory in the renormalization group analysis of a fermionic theory, and numerically confirm the
validity of the form factor expansion in the case of the SU(2)-Thirring model.
PACS: 11.10.Kk
Keywords: Integrable quantum field theory; Correlation function; Conformal perturbation theory;
Renormalization group; SU(2)-Thirring model
1. Introduction
Correlation functions in 2D integrable models have attracted much attention from

experts in Quantum Field Theory (QFT). The possibility of their exploration on the whole
length (energy) scale is of great importance. It gives a rare opportunity to probe non-
perturbatively general principles of QFT. There is also a pragmatic reason for this interest.
The past two decades have witnessed experimental work to identify and study quasi
one-dimensional systems (for a review, see [1,2]). There were collective efforts of many
physicists to apply integrable QFT to describe such physical systems [3]. For this purpose,
E-mail address: doyon@physics.rutgers.edu (B. Doyon).
PII: S 0 5 5 0 - 3 2 1 3 ( 0 2 ) 0 0 7 9 5 - 2
452 B. Doyon, S. Lukyanov / Nuclear Physics B 644 [FS] (2002) 451–475
a non-perturbative treatment of the correlation functions in integrable models seems to be

valuable. It is worth noting that in recent years, angle resolved photoemission spectroscopy
has matured into a powerful experimental method for probing the electronic Green’s
functions in quasi one-dimensional systems [4]. Hence, two-point fermion correlators in
integrable theories deserve special consideration.
In this paper we are studying Schwinger’s function (Green’s function in the Euclidean
region)

Ψσ (x)Ψ̄σ (0)
in the SU(2)-Thirring model, which is described by the Euclidean action1

πg
A= d x 2
Ψ̄σ γ ∂µ Ψσ +
µ
(Ψ̄ γµ τΨ ) .
2
(1.1)
8
σ =↑,↓
Here Ψσ is a doublet of Dirac Fermi fields, and the Pauli matrices τ = (τ 1 , τ 2 , τ 3 )

act on the “colour” indices σ = ↑, ↓. The QFT (1.1) possesses a variety of interesting
properties [5]. For instance, it is an asymptotically free theory (for g > 0) with unbroken
chiral symmetry, and its mass scale M appears through dimensional transmutation. The
model belongs to the very special class of field theories which admit an infinite number of
conservation laws preventing particle production in scattering processes. The Hamiltonian
of (1.1) was diagonalized by the Bethe ansatz techniques in [6,7].
Also, it is a popular model for the interacting one-dimensional electron gas; as is
known [8–12], (1.1) describes the scaling limit of the half-filled Hubbard chain,
+∞
†
HHub = − cj,σ cj +1,σ + cj†+1,σ cj,σ
j =−∞ σ =↑,↓

† 1 † 1
+ U cj,↑ cj,↑ − cj,↓ cj,↓ − ,
2 2
†
where {cj,σ , cj ,σ } = δσ σ δjj . More precisely, if one sends the coupling constant U →
+0, the correlation length
π 2π
Rc = √ e U
2 U
diverges and the correlation functions in the Hubbard chain at large lattice separations
assume certain scaling forms. In particular, if |j − j | 1, the equal-time fermion
correlator can be written as
† sin( π2 (j − j )
cj ,σ cj,σ → δσ σ
F |j − j |/Rc . (1.2)
π(j − j )
The scaling function F here is directly related to the field-theoretic correlation function:
δσ σ γµ x µ
Ψσ (x)Ψ̄σ (0) = F M|x| . (1.3)
2π |x| 2
1 The definition of the coupling constant g is not conventional, but convenient for our purposes.
B. Doyon, S. Lukyanov / Nuclear Physics B 644 [FS] (2002) 451–475 453
Our analysis of the fermion correlator is based, on the one hand, on recently
proposed expressions for the form factors of soliton-creating operators (or topologically
charged fields) in the sine-Gordon model [13],2 and on the other hand, on a conformal
perturbative analysis of two-point correlation functions involving such fields. The form
factor expressions can be used to obtain the long-distance behavior of these two-point
functions, whereas conformal perturbation theory gives their short-distance expansion
([16]). The interest in some of these topological fields stems from their rôle in fermionic
theories. For instance, it is well-known that the sine-Gordon model is equivalent to the
massive Thirring model [17]. The components of the Thirring fermion field are then
associated with soliton-creating operators of topological charge ±1 and Lorenz spin
± 12 , and correlators of these operators in the sine-Gordon model are related to fermion
correlators in the massive Thirring model [18]. More interestingly, the sine-Gordon
theory is closely related to a model which is an integrable deformation of (1.1) [5].
This “deformed” (or anisotropic) SU(2)-Thirring model exhibits the so-called spin-charge
separation, which is translated by its representation in terms of two bosonic theories, one
for the charge part, one for the spin part. The spin part of the fermion field corresponds
to soliton-creating operators of topological charge ±1 and Lorenz spin ± 14 in the sine-
Gordon model, and its charge part is related to similar operators in a free massless bosonic
theory.
Although form factor expansions and conformal perturbation theory are very effective
tools for the study of, respectively, the long-distance and the short-distance asymptotics of
Schwinger’s functions [16,19–21], one usually gets into trouble when trying to compare
both predictions in a region where they are expected to be accurate enough. Indeed,
in general, one has the freedom of choosing the overall multiplicative normalization in
the expansion arising from conformal perturbation theory as well as in the form factor
expansion, and there is no systematic way of relating both normalizations. For the case of
the soliton-creating operators, the constant relating both normalizations was conjectured
in [13]. It allows one to make unambiguous numerical predictions on the correlation
functions of soliton-creating fields on the whole distance scale using the combined
conformal perturbation theory and form factor data. We performed this calculation for
the case of the SU(2)-Thirring fermion.
The paper is organized as follows. In Section 2, we recall some standard results
concerning the anisotropic SU(2)-Thirring model and its relation to the sine-Gordon
theory. In Section 3, the short-distance behavior of correlators of the soliton-creating
operators is examined by means of conformal perturbation theory. Here we also perform a
Renormalization Group (RG) resummation of the perturbative expansion in the vicinity
of the Kosterlitz-Thouless point which corresponds to the SU(2) limit of the fermion
theory. In Section 4, the perturbative calculation is adapted to the momentum space fermion
Schwinger’s function; we give the two-point function in the SU(2)-Thirring model to third
order in the running coupling. This particular result was recently obtained by standard
perturbation theory in the modified Minimal Subtraction (MS) scheme [22] (calculations
2 Without taking normalization into consideration, some of such form factors were considered previously in
Refs. [14,15]
in [22] concern, in fact, fermion correlators in a general non-abelian Thirring model). We

then compare the result of Ref. [22] with ours and explicitly relate our RG scheme to
the MS scheme. In Section 5, the long-distance behavior of the fermion correlator in the
anisotropic SU(2)-Thirring model is analyzed by means of the form factors given in [13].
In Section 6, we examine properties of the fermion spectral density in the SU(2)-Thirring
model. The outcome of our calculations is discussed in Section 7, where we numerically
compare the short-distance behavior of the scaling function F (1.2), (1.3) (from the RG
analysis) with its long-distance behavior (from its form factor expansion).
Finally, we note that this work has an essential overlap with Ref. [23], where (1.1) was
considered as a model of one-dimensional Mott insulators.
2. Bosonization of the anisotropic SU(2)-Thirring model
The SU(2)-invariant Thirring model admits an integrable generalization such that the
underlying SU(2) symmetry is explicitly broken down to U (1) ⊗ Z2 :

πg 3 3 πg⊥ 1 1
AATM = d x 2
Ψ̄σ γµ ∂ Ψσ +
µ
J J + Jµ Jµ + Jµ Jµ ,
2 2
(2.1)
8 µ µ 8
σ =↑,↓
where
JµA = Ψ̄ γµ τ A Ψ (2.2)
are vector currents (and, as before, τ A are Pauli matrices). The model (2.1) is renormal-
izable, and its coupling constants g , g⊥ should be understood as “running” ones. In par-
ticular, in the RG-invariant domain g |g⊥ |, all RG trajectories originate from the line
g⊥ = 0 of UV stable fixed points, and (2.1) indeed defines a quantum field theory.3 Hence,
in this domain (which is the only one that we discuss here), each RG trajectory is uniquely
characterized by the limiting value
1
ρ= lim g () (2.3)
2 →0
of the running coupling g () at extremely short distances ( stands for the length scale),
i.e., the theory (2.1) depends only on the dimensionless parameter ρ, besides the mass
scale M appearing through dimensional transmutation.
As is well-known (see, e.g., [3,5]), the model (2.1) can be bosonized in terms of the
sine-Gordon field ϕ(x),

1
AsG = d x 2
(∂ν ϕ) − 2µ cos(βϕ) ,
2
(2.4)
16π
with the coupling constant β in (2.4) related to ρ (2.3) by
1
β2 = , (2.5)
1+ρ
3 The Hamiltonians corresponding to opposite choices of the sign of g are unitary equivalent, so the sign of
⊥
this coupling does not affect the physical observables.
and a free massless boson. Then the mass scale M is identified with the mass of the sine-
Gordon solitons, which is related to the parameter µ by [24]
√
1
Γ ( 1+ρ ) 1 2ρ
π Γ ( 12 + 2ρ ) 1+ρ
µ= ρ M 1
. (2.6)
πΓ ( 1+ρ ) 2Γ ( 2ρ )
The precise operator relations between (2.1) and (2.3) can be found in [13]. In particular,
for the two-point fermion correlator, the bosonization implies that
δσ σ γµ x µ (1)
Ψσ (x)Ψ̄σ (0) = F (r), (2.7)
2π |x|3/2 1/4
(n)
where we use the notation Fω (n = 1, ω = 1/4) for the real function which depends
only on the distance r = |x| (and implicitly on the mass scale M and the parameter ρ),
and which, in essence, coincides with the Euclidean correlator of non-local topologically
charged fields in the model (2.4):

n −n z̄ ωn (n)
O−ωβ (x)Oωβ (0) = eiπ Fω (r), (2.8)
z
where z = x1 + ix2 , z̄ = x1 − ix2 . Again we refer the reader to the paper [13] for the precise
definition of the field Oan (a = ωβ). Here we note that it carries an integer topological
charge n, a scale dimension
2ω2 n2
d= + (1 + ρ), (2.9)
1+ρ 8
and a Lorentz spin ωn.
3. Short-distance expansion
3.1. Conformal perturbation theory
We now turn to the analysis of the short-distance behavior of the correlator (2.8). In
general, one can examine this behavior via the operator product expansion, for instance:
Fω(n) (r) = CI (r) + Ccos(βϕ) (r)cos(βϕ) + · · · . (3.1)
The structure functions (CI (r), Ccos(βϕ) (r), etc.) admit power series expansions in µ2 ,
which can be obtained by using the standard rules of conformal perturbation theory [16,25],
whereas the vacuum expectation values of the associated operators are in general non-
analytical at µ = 0. In the perturbative treatment, we regard the sine-Gordon model (2.4)
as a Gaussian conformal field theory

1
AGauss = d 2 x (∂ν ϕ)2 (3.2)
16π
perturbed by the relevant operator cos(βϕ). Notice that in the limit µ → 0, the non-local
topologically charged fields Oan can be expressed in terms of the right and left moving parts
of a free massless field ϕ = ϕR (z) − ϕL (z̄) governed by the action (3.2):

n n
Oa |µ→0 → Oa = exp i a −
n n
ϕR (z) − i a + ϕL (z̄) . (3.3)
4β 4β
Conformal perturbation theory gives the structure function CI (3.1) in the form

z̄ ωn −n (x)O
n (0) exp 2µ
eiπ CI (r) = O ωβ −ωβ d 2
y cos(βϕ) , (3.4)
z Gauss
where . . .Gauss is the expectation value in the Gaussian theory AGauss and the exponential
is understood as a perturbative series in µ. In the perturbative series, the integrals will
have power law IR divergences which should be thrown away [16]. Such a regularization
prescription is indicated by the prime near the integral symbol. In the absence of
logarithmic divergences, throwing away the divergences is equivalent to treating the
integrals as analytical continuations in the field dimensions [16]. Considering only the
(n)
part of Fω perturbative in µ, it is a simple matter to obtain

Fω(n) (r) = r −2d 1 + Jn 2ωβ 2 , −2β 2 µ2 r 4−4β + O r 8−8β , r 2 ,
2 2
(3.5)
where d is given by (2.9) and

n n n n
Jn (a, c) = d 2 x d 2 y x a+ 2 x̄ a− 2 (1 − x)−a− 2 (1 − x̄)−a+ 2
n n n n
× y −a− 2 ȳ −a+ 2 (1 − y)a+ 2 (1 − ȳ)a− 2 |x − y|2c . (3.6)
Two comments are in order here. First, the next omitted term in the short-distance
expansion (3.5) comes from either the next term in the perturbative series for CI
2
(O(r 8−8β )) or from the leading contribution of cos(βϕ) (O(r 2 )) in (3.1). Therefore, the
µ2 term written in (3.5) is a leading correction to the scale invariant part of the correlation
function for 12 < β 2 < 1 only. Second, in writing (3.5) we specify the overall multiplicative
normalization of the non-local topologically charged field Oωβ n by the condition
Fω(n) (r) → r −2d as r → 0. (3.7)

The integral (3.5) can be calculated using, for instance, techniques illustrated in [26].
The result can be expressed in terms of two generalized hypergeometric functions at unity:
A(q, c) = 3 F2 (−c, −c − 1, 1 − q; −c − q, 2; 1),

B(q, c) = 3 F2 (q, q + 1, c + 2; c + q + 2, c + q + 3; 1). (3.8)
With q = a + n/2 and q̄ = a − n/2, we found:
Jn (a, c) = J (1) + J (2) + J (3) + J (4), (3.9)

where
J (1) = q q̄Γ (1 − q)Γ (1 − q̄)Γ (1 + c + q)Γ (1 + c + q̄)Γ 2 (−1 − c)

× cos π(q − q̄) − cos(πc) cos π(q + q̄ + c) A(q, c)A(q̄, c),
π 2 qΓ (1 + c)Γ (1 + q̄)Γ (1 + c + q)Γ (−1 − c − q̄)

J (2) = A(q, c)B(q̄, c),
Γ (q)Γ (−q̄)Γ (3 + c + q̄)
π 2 q̄Γ (1 + c)Γ (1 + q)Γ (1 + c + q̄)Γ (−1 − c − q)
J (3) = B(q, c)A(q̄, c),
Γ (q̄)Γ (−q)Γ (3 + c + q)
π 2 Γ 2 (1 + q)Γ 2 (2 + c)Γ (−1 − c − q̄)Γ (−2 − c − q̄)
J (4) = − B(q, c)B(q̄, c).
Γ 2 (−q̄)Γ 2 (−c)Γ (2 + c + q)Γ (3 + c + q)
Notice that for n = 0, the integral (3.6) was calculated previously in the work [27] (see also
Ref. [28]).
3.2. Renormalization group resummation
Here we discuss the short-distance expansion of the correlator (2.8) for β 2 sufficiently
close to unity. For this purpose, it is convenient to use the notation
. = 1 − β 2 1. (3.10)
Our previous short-distance analysis suggests the following expansion for the structure
function CI :
∞

2. 2k
CI (r) = r 2d
1+ ck µr , (3.11)
k=1
where the coefficients ck are given by certain 4k-fold Coulomb-type integrals. Evidently,
this expansion cannot be directly applied in the limit . → 0, where the perturbation
cos(βϕ) of the Gaussian action (3.2) becomes marginal. However, being expressed as
a function of the scaling distance Mr, the structure function CI (r) should admit the
following form:
CI (r) = Zn,ω C(ren)

I (Mr), (3.12)
where the r-independent renormalization constant Zn,ω absorbs all divergences at . = 0

(ren)
and renders the renormalized structure function CI finite in this limit. The divergences
(ren)
of the renormalization constant Zn,ω should be directly related to the singularities of CI
at Mr = 0; they point out that the power law asymptotic behavior (3.5) is modified by
logarithmic corrections at . = 0.
In order to explore the short-distance behavior for . 1, it is convenient to return
to the fermion description. Being essentially the corresponding structure function in the
renormalizable QFT (2.1), CI (r) obeys the Callan–Symanzik equation. Therefore, it can
be written in the form:
r
−2d dr
CI (r) = r exp −2 (Γg − d) . (3.13)
r
0
Here the function Γg is supposed to have a regular power series expansion in terms of the
running coupling constants g,⊥ = g,⊥ (r):
∞

Γg = γlk gl g⊥
2k
, (3.14)
l,k=0
where γlk are constant coefficients. Notice that only even powers of the coupling g⊥
appear in this expansion (see footnote 3). In writing (3.13), we use the normalization
condition (3.7), and take into account that the UV limiting value of Γg coincides with
the scale dimension (2.9),
lim Γg = d. (3.15)
r→0
We have also assumed that there is no resonance mixing of the operator Oωβ n with other
fields, so it is renormalized as a singlet. One can easily check that this is indeed the case
for the operators with |ω| < 12 + |n|4 .
Condition (3.15) already encloses an important restriction on the series (3.14). Indeed,
using Eqs. (2.3) and (2.9) along with the condition that the line of UV stable fixed points
corresponds to g⊥ = 0, one obtains
6
Γg = Γ (0) (g ) + Γ (1)(g )g⊥
2
+ Γ (2) (g )g⊥
4
+ O g⊥ , (3.16)
where

2ω2 n2 g
Γ (0) (g ) = g + 1+ .
1+ 2
8 2
The values of the other coefficients γl,k1 appearing in (3.14) essentially depend on
the choice of a renormalization scheme, i.e., on the precise specification of the running
coupling constants. The latter obey the RG equations
g⊥2
dg dg⊥ g g⊥
r = , r = . (3.17)
dr f (g , g⊥ ) dr f⊥ (g , g⊥ )
Perturbatively, f (g , g⊥ ) and f⊥ (g , g⊥ ) admit loop expansions as power series in g and
g⊥ . In this work, we will use the scheme introduced by Al.B. Zamolodchikov [24,29]. He
showed that under a suitable diffeomorphism in g and g⊥ , the functions f and f⊥ can be
chosen to be equal to each other, and furthermore, to be equal to
g
f = f⊥ = 1 + . (3.18)
2
With this choice for the β-function, the RG equations (3.17) can be integrated. To do this,
we note that this system of differential equations has a first integral, the numerical value of
which is determined through the condition (2.3),
g2 − g⊥
2
= (2ρ)2 . (3.19)
Using (3.19), (3.10) and (2.5), Eq. (3.17) are solved as
√
1+q 4 q
g = 2ρ , g⊥ = ρ , (3.20)
1−q 1−q
where

−2.
1−q
q = (rΛ)4. . (3.21)
ρ
The normalization scale Λ is another integration constant of the system (3.17). It is of the
order of the physical mass scale and supposed to have a regular loop expansion,

Λ = M exp τ0 + τ1 ρ + τ2 ρ 2 + · · · . (3.22)
It should be noted that the even coefficients τ0 , τ2 , . . . are essentially ambiguous and can
be chosen at will. A variation of these coefficients corresponds to a smooth redefinition of
the coupling constants which does not affect the β-function. By contrast, the odd constants
τ2k+1 are unambiguous and precisely specified once the form of the RG equations is fixed.
It is possible to show [24,29] that the odd constants vanish in the Zamolodchikov’s scheme:
τ2k+1 = 0 (k = 0, 1 . . .).
Once the coefficients τ2k in (3.22) are chosen, the running coupling constants are
completely specified, and all coefficients in the power series expansion (3.14) are
determined unambiguously. They can be explicitly calculated by comparing the conformal
perturbative result (3.5) with the form (3.13). From (3.13),
1
Γg = − r∂r log(CI )
2
and, as it follows from the general conformal perturbative expansion (3.11) and the
definition (3.21) of q, the function Γg can be expanded in powers of q. Explicitly, using
the conformal perturbative result (3.5),

√ 4.
ρ
Γg = d − 2. µ2 Jn 2ω(1 − .), 2. − 2 q + O q 2 . (3.23)
Λ
Moreover, the coefficients in this expansion are power series in ρ. For example, using
Eqs. (2.6) and (3.22), it is easy to show that

√ 2.

πµ ρ 1 2
= exp −2τ̄0 ρ + 2τ̄0 − ρ
. Λ 2

2 1 3
− 2τ2 + 2τ̄0 − ζ(3) − ρ + O ρ4 . (3.24)
3 2
Here and after, we set for convenience

π γE +τ̄0
eτ0 = e , (3.25)
8
where γE = 0.5772 . . . is the Euler constant. The integral Jn (2ω(1 − .), 2. − 2) appearing
in (3.23) can also be expanded in powers of ρ, using . = ρ/(1 + ρ). In Appendix A,
we quote the first few terms in the expansion of Jn (a, c) (3.6) around c = −2, which are
obtained through the use of (3.9). From this expansion, it is easy to obtain the expansion
of Jn (2ω(1 − .), 2. − 2) in powers ρ. Then, one can compare the conformal perturbative
expansion of Γg in q and ρ (3.23) with the corresponding expansion (3.14) coming from
the RG analysis (where of course one should expand g and g⊥ 2 in q and ρ from (3.20)).
This determines the coefficients γl,1 for l = 0, 1, 2. If we want an expression valid to order
g 4 , we need one more coefficient: γ0,2 . In principle, it can be obtained from the expansion
in ρ of the coefficients c2 in the series (3.11). In Section 5, we describe a way to find γ0,2
without the cumbersome calculation beyond the lowest order in conformal perturbation
theory.
In order to simplify the form of the structure function (3.13), it is convenient, instead of
using the coefficients γl,k , to parametrize the first few terms of the power series expansions
Γ (1,2)(g ) (3.16) as:
2

1 n u1 3u2 2 3
Γ (g ) = −
(1)
g − + v1 g + v2 − g + O g ,
1 + 2 32 2 2
v2
Γ (2) (g ) = − + O(g ). (3.26)
2
The explicit values of the coefficients u1 , u2 , v1 and v2 in (3.26) are given in Appendix B.
Let us substitute (3.16) and (3.26) into Eq. (3.13). The RG flow Eq. (3.17) allow one to
evaluate the integral and to write the structure function in the form (3.12) with
2 ω2 −n2 (1−ρ 2 )/16
= (Mr)−4ω
(ren) 2 −n2 (1+ρ 2 )/4
CI g⊥

× e−u1 g −u2 g 1 + g⊥
3
2
(v1 + v2 g ) + O g 4 , (3.27)
and
√ n2 /2−2d 2ρu +(2ρ)3 u +···
Zn,ω = M 2d 2ρ+1 ρeτ0 ρ+τ2 ρ +···
3
e 1 2 . (3.28)
Notice that the transformation
Zn,ω → ew0 +w1 (2ρ)

2 +w (2ρ)4 +···
2 Zn,ω , (3.29)
where the series contains only even powers of ρ with arbitrary coefficients wk ,
accompanied by the transformation
2 2 −w g 2 −g 2 2 +··· (ren)
C(ren)
I → e−w0 −w1 g −g⊥ 2 ⊥ CI
does not affect the structure function CI (3.12) due to relation (3.19).
Our prime interest in this work is the correlation function (2.7). For n = 1 and ω = 1/4,
the relations obtained above lead to the following perturbative expansion for the two-point
fermion correlator in the anisotropic SU(2)-Thirring model:

ZΨ δσ σ γµ x µ 2 ρ 2 3 τ̄0 3
Ψσ (x)Ψ̄σ (0) = g⊥
16 exp − g − g
2π ρ2 16 32
|x|2+ 4

3 1 2 3 1 1
× exp τ̄0 − g⊥ − τ̄02 − τ̄0 − 2
g g⊥ + O g4 ,
16 4 16 6 16
(3.30)
where

− ρ 3
ρ2
− 8(1+ρ) π 4(1+ρ) 3ρ γE 3 4
ZΨ = (4ρ) M exp − ρ +O ρ .
2 8 4
In Eq. (3.30), we use the notation τ̄0 defined by (3.25).
We now set ρ = 0 and g = g⊥ = g in (3.30) to obtain the perturbative expansion of the
scaling function F (1.3) for the SU(2)-Thirring model,

3 3 1 2 3 1 4
F (pert)
= exp − g + τ̄0 − g − τ̄ −
2
g +O g .
3
(3.31)
16 16 4 16 0 16
Here the running coupling constant g solves the equation

−1 1 π γE +τ̄0
−g + ln(g) = ln e Mr , (3.32)
2 2
which is the limit ρ = 0 of Eqs. (3.20) and (3.21).
Let us stress here that, if the perturbation series could be summed, then the function F
should not depend on the auxiliary parameter τ̄0 :
∂F
= 0.
∂ τ̄0
This is, however, not true if we truncate the series (3.31) at some order N (for instance, if
one leaves only the terms explicitly written in (3.31)). In this case,
∂ (pert)
FN = O g N+1 ,
∂ τ̄0
(pert)
where the truncated series is denoted by FN . In fitting numerical data with (3.31), we
may treat τ̄0 as an optimization parameter, allowing us to minimize or at least develop a
feeling for the effects of the remainder of the series. Similar ideas have been discussed for
QCD in Ref. [30].
It may be worth mentioning that Eq. (3.27), along with explicit values of the coefficients
quoted in Appendix B, allows one to immediately determine the short-distance expansion
of some other conventional correlators in the (anisotropic) SU(2)-Thirring model. For
example, since the sine-Gordon field ϕ (2.4) itself can be defined by the relation

∂
ϕ = −i Oan ,
∂a n=0
a=0
and the spin current Jµ3 (2.2) is bosonized as

β
Jµ3 = ∂µ ϕ, (3.33)
2π
we can use (3.27) to obtain the short-distance expansion of the current-current correlator.
For the SU(2)-Thirring model (1.1) one has,

A δ AB 2xµ xν
Jµ (x)JνB (0) = 2 2 δµν − I1 + δ µν 2 ,
I (3.34)
π |x| |x|2
where

g 1 g2 g3
I1 = 1 − + τ̄0 + − τ̄0 (τ̄0 + 1)
2 4 2 2

3
τ̄0 τ̄0 13 7
+ + τ̄02 + − − ζ(3) g 4 + O g 5 ,
2 4 128 16
2

g 7 2 τ̄0 13 7
I2 = 1 − 2τ̄0 g + τ̄0 (3τ̄0 + 1)g − 4τ̄0 + τ̄0 + −
2 3
− ζ(3) g 3
4 2 2 16 2

+ O g4 ,
and the running coupling constant g is the same as in (3.32).
4. Perturbative expansion of the momentum-space correlation function
4.1. Large-momentum asymptotics
Perturbative calculations of fermion Green’s functions in renormalizable 2D models

with four-fermion interaction are widely covered in the literature (see [22,31] and
references therein). The results in this domain are usually expressed in momentum space.
Hence it seems appropriate at this point to adapt the calculation of the previous section to
the momentum-space fermion correlator, giving a large-momentum expansion.
The RG analysis performed in the previous section can be applied in essentially the
same way to the Fourier transform of the fermion correlator (2.7):

γ µ pµ 2
d 2 x e−ipx Ψσ (x)Ψ̄σ (0) = −iδσ σ F p . (4.1)
p2
Here and after we use the notation p2 = pµ pµ . From the result of conformal perturbation
theory, (3.5), one immediately obtains the large momentum expansion of this Fourier
transform:

= Q(dΨ ) p2 dΨ − 2 1 + Q(dΨ − 2.) J1 β 2 /2, −2β 2 µ2 p2 −2.
1
F
Q(dΨ )

2 −4. −2
+O p ,p , (4.2)
where
Γ ( 32 − a)
Q(a) = 21−2a ,
Γ ( 12 + a)
and
1 ρ2
dΨ = +
2 4(1 + ρ)
is the scale dimension of the fermion field. The factor Q(dΨ − 2.)/Q(dΨ ) is essentially
the only source of differences between the RG treatments in coordinate space and in
momentum space. The RG analysis in momentum space goes as in the previous section.
The perturbative part in µ of F obeys the Callan–Symanzik equation, so it can be written
as
∞
2 dΨ − 1 ds

F (pert)
= Q(dΨ ) p 2 exp − Γg − dΨ , (4.3)
s
p2
where the function Γg admits a power series expansion in terms of the momentum-space
running coupling constants g,⊥ = g,⊥ (p2 ) depending on the Lorentz invariant p2 :
∞

Γg = γ̃l,k gl g⊥
2k
. (4.4)
l,k=0
Notice that, with some abuse of notations, we use here the same symbols g,⊥ for
the momentum-space running couplings as we used for the coordinate-space running
couplings. In order to fix the coefficients in (4.4), we have to choose a renormalization
scheme. Substituting r by 1/ p2 in (3.21) defines Zamolodchikov’s scheme in momentum
space. It is a simple matter to repeat the steps of the previous section in order to determine
the first few coefficients γ̃l,1 in (4.4). Just compare the logarithmic derivatives of the
expressions (4.2) and (4.3); the only difference is that the factor Q(dΨ − 2.)/Q(dΨ )
in (4.2) will have to be expanded in ρ, giving non-trivial contributions. As for the
coefficients γ̃l,2 , one would in principle need the next order in conformal perturbation
theory. However, again as in the previous section, it is possible to determine γ̃0,2 without
this calculation, as described in the next section. From these coefficients, and from the form
of the RG flow equation, one can evaluate the integral in (4.3) and obtain the asymptotic
behavior of the two-point function in the Euclidean region at p2 → +∞. We quote here
the result in the case of the SU(2)-Thirring model,

3 3 1 2 3 1 4
F(pert)
= exp − g + τ̃0 − g − τ̃ −
2
g +O g .
3
(4.5)
16 16 4 16 0 16
Here
√
1
−g −1 + ln(g) = ln 2π Meτ̃0 / p2 , (4.6)
2
and τ̃0 is an arbitrary parameter which can be chosen at will. Notice the strong similarity
between (4.5) and (3.31).
We also quote here the corresponding function Γg (4.4) in the case g = g⊥ :

1 3 3 3 3
Γg = + g 2 − τ̃0 g 3 + 3τ̃02 + τ̃0 − g4 + O g5 . (4.7)
2 32 16 32 16
4.2. Comparison with the four-loop conventional perturbation calculations
In [22], the anomalous dimension for the fermion field in the MS scheme was found
to fourth order for a general non-Abelian Thirring model (see also [31] and references
therein for a discussion of various aspects of dimensional regularization in the non-

Abelian Thirring model and for results to lower order). In contrast, we have calculated, in
coordinate space, the two-point functions of more general operators, including the fermion
fields, in the particular case of the SU(2)-Thirring model (and an anisotropic deformation
of it), and we have sketched the equivalent calculation in momentum space for the fermion
fields. We would now like to compare Eq. (4.5) with the SU(2) case of the Ali–Gracey
result [22]. In order to perform the comparison, we need to find the relation between our
running coupling constant g and theirs, which will be denoted gAG = −λ,4 and then find
the relation between our function Γg (4.3) and their anomalous dimension, which we will
denote γλ .
The coupling λ corresponds to the MS scheme; the associated β-function was found
in [31] to fourth order:
dλ λ2 λ3 83 4
2p2 2
= β λ = − + 2
− 3
λ + O λ5 . (4.8)
dp π 2π 128π
By comparison, in the scheme that we use, the β-function (3.17), (3.18) is
dg g2 g3 g4
2p2 = βg = − = −g 2 + − + O g5 . (4.9)
dp 2 1 + g/2 2 4
The difference in the factor multiplying the square of the coupling in these two expressions
results only from a different normalization of the coupling in the action (see Eq. (1.1)). The
relation between the couplings g and λ that corresponds to these different β-functions is

λ τ 51
= g − τg 2 + τ 2 + + g3 + O g4 . (4.10)
π 2 128
Here τ is some numerical factor which cannot be determined by comparing the β-
functions: its variation modifies the choice of the normalization scale and does not affect
the β-functions. The normalization scale for the MS scheme is defined by imposing the
following condition on the subleading asymptotics of the solution of the RG flow Eq. (4.8):

2 2
λ 1 1 ln ln( p2 /ΛMS ) ln ln( p /ΛMS )
= + +O . (4.11)
π ln( p2 /ΛMS ) 2 ln2 ( p2 /ΛMS ) ln3 ( p2 /ΛMS )

(This implies that the term O(1/ ln2 ( p2 /ΛMS )) does not appear in the expansion of λ.)
From (4.6), (4.10) and (4.11), we find that
√
ΛMS = 2π Meτ̃0 −τ . (4.12)
In [22], the perturbative part of the function F (4.1) was calculated up to the overall
multiplicative normalization to third order in λ. The result can be written in the following
form

2
p
(pert) 1 1 ds
F ∝ exp − γλ ,
hλ 2 s
4 Notice that in [22], the coupling constant g

AG is assumed to be negative, so λ > 0, which agrees with the
sign of our coupling constant g.
where the function hλ and the anomalous dimension γλ were given in [22] to fourth order
in λ for the Thirring model with a general non-Abelian symmetry. In the particular case of
the SU(2)-symmetry, they specialize to
15 2 11 3 3(80ζ(3) − 511) 4
hλ = 1 + 2
λ − 3
λ + 4
λ + O λ5 , (4.13)
128π 512π 32768π
and
3 2 15 3 3
γλ = − 2
λ + 3
λ + 4
λ4 + O λ5 . (4.14)
16π 64π 1024π
Comparing (4.3) in the case ρ = 0 with the above expressions, one has the following
relation:
1 γ λ βλ d
Γg = − − log(hλ ). (4.15)
2 2 2 dλ
Using Eqs. (4.10)–(4.14), one can check that our result (4.7) agrees with (4.15), provided
that
τ = τ̃0 . (4.16)
Notice that the relation between the normalization scale ΛMS and M,
√
ΛMS = 2π M, (4.17)
which is a consequence of (4.12) and (4.16), was previously found in Ref. [12].
5. Long-distance behavior
Here we concentrate on the long-distance behavior of Schwinger’s function (2.8) for

n = 1 and ω = 1/4. Let us recall that for 1/2 < β 2 1, there are only solitons and
antisolitons in the spectrum of the sine-Gordon model. We will denote them by A− and
A+ , respectively. The conservation of the topological charge,
∞
β
dx ∂x ϕ,
2π
−∞
+1
implies that the non-vanishing form factors of the operator Oβ/4 are of the form

+1
vac|Oβ/4 (0)A− (θ1 ) · · · A− (θN+1 )A+ (θ1 ) · · · A+ (θN ) , (5.1)
where θi and θj denote rapidities of solitons and antisolitons respectively. Up to an overall
normalization, all these form factors can be written down in closed form, as certain N -fold
integrals [14,32,33]. The spectral decomposition for the correlation function (2.7) then
gives
+∞
dθ −Mr cosh(θ) 2
+1
(0)A− (θ )
(1)
F1/4 (r) = e vac | Oβ/4
2π
−∞
∞
1 dθ1d θ2 dθ3 −Mr 3 cosh(θk )
+ e k=1
3! (2π)3
−∞

× vac|O+1 (0)Aσ (θ1 )Aσ (θ2 )Aσ (θ3 ) 2 + · · · , (5.2)
β/4 1 2 3
σ1 +σ2 +σ3 =−1
where the dots stand for the five-particle and higher contributions, which are of the order
of e−5Mr . The long-distance asymptotic behavior of the correlation function is dominated
by the contribution of the one-particle states,

+1
(0)A− (θ ) = Z1 (β/4) e 4 (θ+ 2 ) ,
1 iπ
vac|Oβ/4
and has an especially simple form,
−Mr
e
(1)
F1/4 (r) = Z1 (β/4) √ + O e−3Mr . (5.3)
2πMr
Here we use the notation Zn (a)(a = ωβ) from work [13] for the field-strength renormal-
ization which controls the long-distance asymptotics of the correlation function (2.8). Let
us stress here that the overall multiplicative normalization of the field Oβ/4
1 was already
fixed by the condition (3.7), hence, the constant Z1 (β/4) is totally unambiguous. In [13],
the following explicit formula for Zn (ωβ) was proposed:
Zn (ωβ)

n
n2 √π MΓ ( 3 + 1 ) 2d
C2 2 C2 − 4 2 2ρ
=
2C12 16ρ 2Γ (1 + 2ρ )
1
∞
dt cosh(4ωt)e−(1+ρ)nt − 1 n −2t
× exp + − 2de .
t 2 sinh(t) sinh((1 + ρ)t) cosh(tρ) 2 sinh(t)
0
(5.4)
In this formula, d is the scale dimension (2.9) and the constants C1 , C2 read
∞
2− 12 e 4 Γ ( 14 )
5 1
dt sinh2 ( 2 )e−t
tρ
C1 = √ exp ,
π A3G t 2 cosh2 (tρ) sinh(t)
0
∞
dt sinh2 ( 2 )e−t
tρ
Γ 4 ( 14 )
C2 = exp −2 ,
4π 3 t cosh(tρ) sinh(t)
0
where AG = 1.282427 . . . is the Glaisher constant.
We do not write down explicitly the general formula for the three-particle contribution
in (5.3) because it is a rather mechanical substitution of relations presented in [13]. (For
β 2 = 1 the corresponding formulas can be found in Appendix C.) Here we make the
following observation concerning the β 2 → 1 limit. The examination of (5.2) based on

explicit formulas for the form factors shows that the function [Z1 (β/4)]−1 F1/4
(1)
admits
2
an asymptotic power series expansion in terms of the variable ρ . In other words, all
(1)
divergences at ρ 2 → 0 of F1/4 , considered as a function of the variables ρ 2 and Mr,
are absorbed by the normalization constant Z1 (β/4). Using Eq. (5.4), one can check that
the constant Z1 (β/4) admits exactly the same type of singular behavior at ρ 2 = 0 as the
constant Zn,ω (3.28) for n = 1, ω = 1/4, and also that
Z1 (β/4) 1√
= 2− 3 π e− 4 A3G exp w1 ρ 2 + O ρ 3 .
1
(5.5)
Z1,1/4
The explicit form of the coefficient w1 is not essential here. What is important is that
the linear term in ρ does not appear in the expansion (5.5). This observation can be
immediately generalized and checked for any n and ω. Furthermore, we expect that

∞
Zn (ωβ)
log = wk ρ 2k + O ρ ∞ , (5.6)
Zn,ω
k=0
where Zn (ωβ) is the normalization constant (5.4). In other words, by means of the
transformation (3.29) with properly chosen coefficients wk , the constant Zn,ω in (3.12)
can be set to be equal (in a sense of formal power series) to Zn (ωβ). At the moment, we
do not have a rigorous proof of (5.6). But it leads to some interesting prediction to be
checked. As was already mentioned, the calculations performed in the leading order in
conformal perturbation theory determine only the combination v2 − 3u2 /2, but do not fix
the individual values of the coefficients u2 and v2 in the series (3.27). Accepting (5.6),
one can immediately find the values of the coefficients u2 (see Appendix B). In the case
n = 1, ω = 1/4, it allows one to extend the perturbative expansion (3.31), as well as the
equivalent expansion (4.5), to order g 3 . As was discussed in Section 4, Eq. (4.5) is in a
complete agreement with the result of four-loop perturbative calculations from [22]. This
in fact shows that the ρ 3 -term really is absent in the series (5.5).
6. Spectral density
The spectral density is an important quantity related to the two-point function and its
analytical structure in momentum space. It is often what is measured in actual condensed
matter experiments [4,23], and it allows one to completely reconstruct the two-point
function. In this section, we discuss the properties of the spectral density in the SU(2)-
Thirring model.
The spectral decomposition of the fermion Green’s function yields the following form
(4.1):
for the function F
+∞ (s)
AF
p2 = 1 −
F ds 2 . (6.1)
p +s
M2
The notation AF for the spectral density reminds us that F

, considered as a function of
one complex variable p , has a branch cut in the Minkowski region p2 < 0 starting at
2
p2 = −M 2 , and that the spectral density can be recovered from the discontinuity along
this cut:
(s) = 1 iπ −iπ
AF F e s −F e s . (6.2)
2πi
The easiest way to obtain the large s asymptotics of the spectral density is to use
the expansion (4.5) along with knowledge of the analytical properties of the coupling
constant g (4.6) as a function of the complex variable p2 . Notice that g can be expressed
in terms of the principal branch of the product log (or Lambert) function, which gives the
solution for W in W eW = z (see, e.g., [34]):

2 −2τ̃0
−1 p e
g = 2W . (6.3)
πM 2
The principal branch of the W -function analytically maps the complex z-plane minus the
branch cut z ∈ ] − ∞, −e−1] to the part of the complex W -plane enclosing the real axis and
delimited by the curve !e W = − "m W cot("m W ) for −π < "m W < π . The analyticity
implies that the power series
∞

1 d n
iφz W (z)
n! dz z=s
n=0
converges for real positive s > e−1 and |φ| π and coincides with W (eiφ s). Similar
considerations are, of course, valid for the coupling constant g (6.3). In particular, for
sufficiently large s,
∞

1 d n
g e±iπ s = ±iπp2 2 g p2 .
n! dp p 2 =s
n=0
This then gives us, with (4.5) and the RG flow equation (4.9), the asymptotic expansion of
the spectral density for large s. It can be written in the following form:
2
(pert)
(s) = − g 1 − g − π − 1 g 2 + O g 3 ∂ F
2
AF . (6.4)
2 2 4 ∂g p2 =s
Here the function F(pert) is given by (4.5) and g is defined by Eq. (4.6).
Now let us consider the threshold behavior of the spectral density. According to the
analyses of the previous section, the long-distance asymptotic behavior of the scaling
function F (1.3) is described by the expansion

F = F (1) + F (3) + O e−5Mr , (6.5)
where
F (1) = Ce−Mr ,
with the constant

C = 2− 6 e− 4 A3G = 0.921862 . . . .
5 1
The function F (3) in (6.5) gives the three-particle contribution to the correlation function.
Using the definitions (1.3), (4.1) and the above relation, one can obtain:

2 1

F p =C 1− + ···. (6.6)
1 + p2 /M 2
Here the dots stand for contributions of the massive multiparticle intermediate states. The
last relation implies that the spectral density (6.2) can be written as
C Θ(s − M 2 )
(s) =
AF (3) (s),
+ Θ s − 9M 2 AF (6.7)
π s/M 2 − 1
where

1 for s 0,
Θ(s) =
0 for s < 0
and AF (3) is some function which contributes to the spectral density only above the
threshold s = 9M 2 .
7. Numerics
In Table 1 we present results of numerical evaluation of the function F (1.3) as a

function of the scaling distance Mr (r = |x|). To estimate the short-distance behavior,
we use the perturbative expansion (3.31). As was already mentioned, the parameter τ̄0
allows one to have control over the accuracy of the truncated series, so we calculate (3.31)
for two different values of τ̄0 : −0.25 and +0.25. To determine the long-distance behavior
of the function F , we use the formula (6.5), where the three-particle contribution F (3)
was obtained by means of Eq. (5.2) along with formulas for the three-particle form factors
quoted in [13] (see Appendix C). It is interesting to see that the sum of the one- and three-
particle contributions to F is very near to unity at r = 0 (to within 1%), which indicates that
this three-particle computation of the correlation function is in fact accurate to about 1%
for all distance scales (more accurate, of course, for larger r). Also, note that the crossover
between the long- and short-distance asymptotics appears to be at the scaling distances
Mr ∼ 0.001–0.01, where both asymptotics coincide to within about 0.1%.
8. Conclusion
In this paper, we studied the fermion Schwinger’s function in the SU(2)-Thirring

model and in an anisotropic integrable deformation of it. We used and compared two
well-established methods: the form factor expansion, and conformal perturbation theory
combined with a RG analysis, leading respectively to the long-distance and the short-
distance behavior of the correlation function. A proper comparison of these opposite limit
Table 1
The scaling function F (1.2), (1.3). The first columns give the results of the long-distance expansion which
includes contributions of the one-, three- and one + three-particle states. The data in the last two columns
correspond to the perturbative expansion (3.31) for the two different values of the auxiliary parameter τ̄0
Mr F (1) F (3) F (1) + F (3) F (pert) (τ̄0 = −0.25) F (pert) (τ̄0 = 0.25)
0 0.921862 0.068 0.990 1.00000 1.00000
0.00001 0.921853 0.0553 0.9771 0.980129 0.980130
0.00005 0.921816 0.0522 0.9740 0.976311 0.976314
0.0001 0.921770 0.0504 0.9722 0.974192 0.974196
0.0002 0.921678 0.0483 0.9700 0.971674 0.971678
0.001 0.920941 0.0415 0.9624 0.963508 0.963520
0.002 0.920020 0.0375 0.9575 0.958435 0.958454
0.01 0.912689 0.0252 0.9379 0.939386 0.939460
0.025 0.899101 0.0168 0.9159 0.919294 0.919494
0.05 0.876902 0.0106 0.8875 0.894050 0.894547
0.075 0.855251 0.00738 0.86263 0.871796 0.872717
0.1 0.834135 0.00541 0.83955 0.850520 0.852013
0.15 0.793454 0.00317 0.79662 0.808380 0.811548
0.2 0.754757 0.00200 0.75676 0.765139 0.770842
0.25 0.717947 0.00131 0.71926 0.719980 0.729252
0.3 0.682932 0.000889 0.683822 0.672640 0.686654
0.35 0.649625 0.000617 0.650243 0.623153 0.643171
0.4 0.617942 0.000436 0.618379 0.571774 0.599063
0.45 0.587805 0.000313 0.588118 0.518942 0.554677
0.5 0.559137 0.000227 0.559365 0.465257 0.510405
behaviors needs an adjustment of the normalization used in one method with respect
to that used in the other method. The necessary formula for such an adjustment was
proposed in [13]: there the exact form factors, with appropriate normalization constant,
were conjectured assuming a “conformal” normalization of the fields. This allowed us to
numerically compare both methods and to observe an agreement to within 1% in the region
0 < Mr < 0.05. Moreover, using these exact form factors we conjectured an infinite set of
relations between expansion coefficients of the fermion anomalous dimension arising in
the RG treatment of the anisotropic model. This was done essentially by identifying the
singular part in ρ 2 (see Eqs. (2.3) and (2.5)) of the normalization constant conjectured
in [13] with the singular part of the normalisation constant obtained in the RG treatment
(see (5.6)). Using one of these relations along with a first non-trivial order calculation in
conformal perturbation theory, we obtained the desired fermion Schwinger’s function to
third order in the coupling of the SU(2) model, and observed agreement with what was
obtained in [22] by standard perturbation theory. These results, numerical and analytical,
suggest the validity of the conjectured exact form factors of [13] in the case of the SU(2)-
Thirring model. It might be interesting to apply similar methods to Thirring-like models
with other symmetries.
Acknowledgements
S.L. is grateful to F. Essler, A. Tsvelik for their helpful collaboration during the
early stages of this work. We have also extremely benefited from discussions with A.
Zamolodchikov, Al. Zamolodchikov.
This research is supported in part by DOE grant #DE-FG02-96ER10919. B.D. is
supported in part by a NSERC Postgraduate Scholarship.
Appendix A
In this appendix, we give the first few terms in the expansion of Jn (a, c) (3.6) around
c = −2. The coefficients in this expansion involve standard functions of a, which could
2ω −2
then easily be used to obtain an expansion of Jn ( 1+ρ , 1+ρ ) in powers of ρ, as is needed
in (3.23). To simplify the result, we will use the parameter
b = c + 2.
We find the following expansions in b of the functions A(q, b − 2), B(q, b − 2) (3.8)
involved in (3.9):

2
Γ (2 − b − q)Γ (b) 2 π ψ(1 − q) + γE 3
A(q, b − 2) = 1+b + +O b ,
Γ (2 − b)Γ (1 + b − q) 6 q

Γ (q + b)Γ (b) b b2
B(q, b − 2) = 1+ + 2 q 2 ψ (q) − 1
Γ (q)Γ (2b) 2q 2q

b3
+ qψ (q) + 2ψ (q) + O b4 .
4q
Hence,
π 2 4a 2 − n2 + 2nb
Jn (a, b − 2) =
2 b2 (1 − b)2

Gn (a) aGn (a) + Gn (a) 10
× exp −Gn (a)b + −2 + ζ(3) b 3
12 n2 − 4a 2 3

4
+O b ,
where
Gn (a) = ψ(a + n/2) + ψ(−a + n/2) + 2γE ,
and Gn (a) = d
da Gn (a), Gn (a) = d2
G (a).
da 2 n
Appendix B
In this appendix, we write down explicit expressions for the coefficients u1 , u2 , v1 and
v2 taking part in the expansion (3.16), (3.26) of the function Γg .
On the one hand, from the assumption (5.6), the coefficients of odd powers of ρ in the
exponential factor of Zn,ω (3.28) are completely fixed by the conjectured constant Zn (ωβ)
(5.4). This fixes u1 , u2 uniquely, giving

n2 3 n(n − 2)
u1 = ω 2 − Tn (2ω) − + ,
16 2 16

1 n2 n2 1 ω(4ω2 − 1)
u2 = ω2 − ω2 + − Tn (2ω) + Tn (2ω)
12 16 16 2 12

1 2 1 n(n + 4) 11ω2 1 τ2 n2
+ ω − Tn (2ω) − − + τ̄0 + ω −
2
,
4 12 768 48 24 2 16
(B.1)
where
Tn (a) = ψ(a + n/2) + ψ(−a + n/2) + 2γE + 2τ̄0 ,

d d2
Tn (a) = Tn (a), Tn (a) = 2 Tn (a).
da da
On the other hand, the expansion in powers of ρ of (3.23) uniquely determines the
coefficients u1 , v1 and v2 − 32 u2 in the first equation of (3.26). The coefficient u1 thus
obtained is in agreement with (B.1), verifying the assumption (5.6) to first order. The
coefficients v1 and (using the expressions u1,2 from (B.1)) v2 are

ω 2 n2 1 2 n2
v1 = ω − T (2ω) + ω − T 2 (2ω)
2 16 n 4 16 n

3 2 n(5n − 4) 7 2 n(17n − 20) u1
− ω − Tn (2ω) + ω − + ,
4 48 8 112 2

2
2
n 1 ω
v2 = ω2 − − Tn (2ω)
16 24 4

n2 Tn (2ω) 1 2 n(n − 4) 1
− ω2 − + ω + − ωTn (2ω)
16 2 4 16 2

1 n2 1 2 n(n − 4)
− ω −
2
T (2ω) −
3
ω + Tn2 (2ω)
12 16 n 8 16

1 2 n(n − 2) 1
− ω − − Tn (2ω)
8 8 2

1 2 n2 n(n − 8) u1 v1 3u2 τ̄0
− ω − 2τ2 − 14ζ(3) − 3 − + + + − .
8 16 256 8 2 2 8
Appendix C
In this appendix, we give the formula for the three-particle contribution F (3) (6.5) to the
fermion two-point function in the SU(2)-Thirring model that we used for our numerical
calculations. We first specialize the expression written in [13] to the case of three-particle
form factors of the field Oβ/4
1 for β 2 = 1:

vac|O1/4
1
(0)A− (θ1 ) · · · A+ (θk ) · · · A− (θ3 ) in
9
AG2 Γ 3 ( 14 ) iπ 1
3
θm
=− 15 3 9
G(θm − θj )
e 8 M4 e 4
2 e π
4 8
m=1 4
m<j

dγ − γ
k 3
× e 2 W (θp − γ ) W (γ − θp )
2π
C+ p=1 p=k+1

dγ − γ
k−1 3
+ e 2 W (θp − γ ) W (γ − θp ) . (C.1)
2π
C− p=1 p=k
Here the functions G and W are

∞
2− 12 e 4 Γ ( 14 )
5 1
dt sinh2 t (1 − iθ/π)e−t
G(θ ) = i √ sinh(θ/2) exp (C.2)
π A3G t sinh(2t) cosh(t)
0
and
Γ ( 34 − iθ
2π )Γ (− 4
1
+ iθ
2π )
W (θ ) = 2 . (C.3)
Γ 2 ( 14 )
The contour C+ starts from −∞ on the real axis of the complex γ -plane, goes above the
poles located at γ = θp + iπ/2, p = 1, . . . , k, and below those located at γ = θp − iπ/2,
p = k + 1, . . . , 3, always staying in the strip −π/2 − 0 < "m γ < π/2 + 0, and finally
extends to +∞ on the real axis. Similarly, the contour C− goes above the poles located at
γ = θp + iπ/2, p = 1, . . . , k − 1, and below those at γ = θp − iπ/2, p = k, . . . , 3. Notice
that the integrals in (C.1) can be expressed in terms of the generalized hypergeometric
function 3 F2 at unity.
Using the expressions (C.1) and performing one of the rapidity integrals in (5.2), one
can obtain the following form for the function F (3) in (6.5):
∞ √
2e− 4 A9G
3
e−Mr 3+2 cosh x+2 cosh y+2 cosh(x−y)
F (3)
= dx dy
3πΓ 6 ( 14 )
1
(3 + 2 cosh x + 2 cosh y + 2 cosh(x − y)) 4
−∞
2
× 2|R1 (x, y)|2 + |R2 (x, y)|2 G(x)G(y)G(x − y)

−x 1
x+y e + e−y + 1 4
×e 2 .
ex + ey + 1
The functions R1 and R2 here are
y
R2 (x, y) = e− 2 + 4 R1 (−x, y − x) − e− 2 − 4 R1∗ (−y, x − y)
x iπ iπ
and

cosh x2 cosh y2 1 1 ix 1 iy ix iy
R1 (x, y) = − U − ,− − ,− − ;− ,−
2 sinh x sinh y 2 2 2π 2 2π 2π 2π
cosh y−x
2 cosh 2
x
+ e− 2
x
2 sinh(y − x) sinh x

1 1 i(y − x) 1 ix i(y − x) ix
× , − , + ;1− ,2 +
2 2 2π 2 2π 2π 2π
y cosh x−y
2 cosh 2
y
+ e− 2
2 sinh(x − y) sinh y

1 1 i(x − y) 1 iy i(x − y) iy
×U , − , + ;1− ,2 + ,
2 2 2π 2 2π 2π 2π
where U (a, b, c; d, e) is related to the generalized hypergeometric function 3 F2 by
Γ (a)Γ (b)Γ (c)
U (a, b, c; d, e) = 3 F2 (a, b, c; d, e; 1).
Γ (d)Γ (e)
References
[1] G. Gruner, Density Waves in Solids, Addison–Wesley, 1994.

[2] C. Bourbonnais, P. Jerome, The normal phase of quasi-one-dimensional organic superconductors, in:
P. Bernier, S. Lefrant, E. Bidan (Eds.), Advances in Synthetic Metals, Twenty Years of Progress in Science
and Technology, Elsevier, New York, 1999.
[3] A.O. Gogolin, A.A. Nersesyan, A.M. Tsvelik, Bosonization and Strongly Correlated Systems, Cambridge
Univ. Press, 1999.
[4] Y. Hwu, et al., Photoemission near the Fermi energy in one dimension, Phys. Rev. B 46 (1992) 13624–13626;
F. Zwick, et al., Absence of quasiparticles in the photoemission spectra of quasi-one-dimensional Berchgaard
salts, Phys. Rev. Lett. 79 (1997) 3982–3985.
[5] J. Zinn-Justin, Quantum Field Theory and Critical Phenomena, third Edition, Clarendon, Oxford, 1996.
[6] N. Andrei, J.H. Lowenstein, Diagonalization of the chiral-invariant Gross–Neveu hamiltonian, Phys. Rev.
Lett. 43 (1979) 1698–1701.
[7] A.A. Belavin, Exact solution of the two-dimensional model with asymptotic freedom, Phys. Lett. B 87
(1979) 117–121.
[8] V.M. Filev, Spectrum of two-dimensional relativistic model, Teoret. Mat. Fiz. 33 (1977) 119–124.
[9] V.J. Emery, A. Luther, I. Peschel, Solution of the one-dimensional electron gas on a lattice, Phys. Rev. B 13
(1976) 1272–1276.
[10] V.E. Korepin, F.H.L. Essler, SU(2) ⊗ SU(2) invariant scattering matrix of the Hubbard model, Nucl. Phys.
B 426 (1994) 505–533.
[11] E. Melzer, On the scaling limit of the 1d Hubbard model at half filling, Nucl. Phys. B 443 (1995) 553–564.
[12] F. Woynarovich, P. Forgacs, Scaling limit of the one-dimensional attractive Hubbard model: the half-filled
band case, Nucl. Phys. B 498 (1997) 65–603.
[13] S. Lukyanov, A. Zamolodchikov, Form factors of soliton-creating operators in the sine-Gordon model, Nucl.
Phys. B 607 (2001) 437–455.
[14] H. Babujian, A. Fring, M. Karowski, A. Zapletal, Exact form-factors in integrable quantum field theories,
the sine-Gordon model, Nucl. Phys. B 538 (1999) 535–586.
[15] G. Delfino, Off critical correlations in the Ashkin–Teller model, Phys. Lett. B 450 (1999) 196–201.
[16] Al.B. Zamolodchikov, Two-point correlation function in scaling Lee–Yang model, Nucl. Phys. B 348 (1991)
619–641.
[17] S. Coleman, The quantum sine-Gordon equation as the massive Thirring model, Phys. Rev. D 11 (1975)
2088–2097.
[18] S. Mandelstam, Soliton operators for the quantized sine-Gordon equation, Phys. Rev. D 11 (1975) 3026–
3030.
[19] B. Berg, M. Karowski, P. Weisz, Construction of Green’s functions from an exact S-matrix, Phys. Rev. D 19
(1979) 2477–2479.
[20] J. Balog, M. Niedermaier, Off-shell dynamics of the O(3) non-linear sigma model beyond Monte Carlo and
perturbation theory, Nucl. Phys. B 500 (1997) 421–461.
[21] C. Acerbi, G. Mussardo, A. Valleriani, Form-factors and correlation functions of the stress-energy tensor in
(1) (1) (2)
massive deformation of the minimal models E(N) × E(N) /E(N) , Int. J. Mod. Phys. A 11 (1996) 5327–5364.
[22] D.B. Ali, J.A. Gracey, Four loop wave function renormalization in the non-Abelian Thirring model, Nucl.
Phys. B 605 (2001) 337–364.
[23] F.H.L. Essler, A.M. Tsevelik, Weakly coupled one-dimensional Mott insulator, cond-mat/0108382.
[24] Al.B. Zamolodchikov, Mass scale in the sine-Gordon model and its reduction, Int. J. Mod. Phys. A 10 (1995)
1125–1150.
[25] R. Guida, N. Magnoli, All order IR finite expansion for short distance behavior of massless theories
perturbed by a relevant operator, Nucl. Phys. B 471 (1996) 361–388.
[26] S.D. Mahur, Quantum Kac–Moody symmetry in integrable field theories, Nucl. Phys. B 369 (1992) 433–
460.
[27] V. Dotsenko, M. Picco, P. Pujoi, Renormalization group calculation of correlation functions for the 2D
random bound Ising and Potts models, Nucl. Phys. B 455 (1995) 701–723.
[28] R. Guida, N. Magnoli, Tricritical Ising model near criticality, Int. J. Mod. Phys. A 13 (1998) 1145–1158.
[29] Al.B. Zamolodchikov, unpublished.
[30] P.M. Stevenson, Optimized perturbation theory, Phys. Rev. D 23 (1981) 2916–2943.
[31] J.F. Bennett, J.A. Gracey, Three-loop renormalization of the SU(Nc ) non-Abelian Thirring model, Nucl.
Phys. B 563 (1999) 390–436.
[32] F.A. Smirnov, Form-factors in Completely Integrable Models of Quantum Field Theory, World Scientific,
Singapore, 1992.
[33] S. Lukyanov, Free field representation for massive integrable models, Commun. Math. Phys. 167 (1) (1995)
183–226;
S. Lukyanov, Form-factors of exponential fields in the sine-Gordon model, Mod. Phys. Lett. A 12 (1997)
2543–2550.
[34] R.M. Corless, G.H. Gonnet, D.E.G. Hare, D.J. Jeffrey, D.E. Knuth, On the Lambert W function, Adv.
Comput. Math. 5 (4) (1996) 329–359;
D.J. Jeffrey, D.E.G. Hare, R.M. Corless, Unwinding the branches of the Lambert W function, Math. Sci. 21
(1996) 1–7.
Asymptotic Bethe ansatz for the Sutherland–Römer

model with open boundary conditions
Jun-Peng Cao a,∗ , Rui-Hong Yue b , Yu-Peng Wang a
a Institute of Physics & Center for Condensed Matter Physics, Chinese Academy of Sciences,
PO Box 603, Beijing 100080, China
b Institute of Modern Physics, Northwest University, PO Box 105, Xi’an 710069, China
Received 23 May 2002; received in revised form 5 August 2002; accepted 12 September 2002
Abstract
By constructing the reflection spin-Dunkl operators, the integrable Sutherland–Römer model
(SRM) with open boundary condition is established, which describes a one-dimensional, two-
component, quantum many-particle system in which like particles interact with a pair potential
g(g + 1)/ sinh2 (r), while unlike particles interact with a pair potential −g(g + 1)/ cosh2 (r). By
solving the Schrödinger equation and using the properties of the hypergeometric functions and
gamma functions, the two-particle scattering matrix and the reflection matrix are obtained in the
framework of the asymptotic Bethe ansatz method. The Bethe ansatz equations of the system are
obtained. The Hamiltonians of SRM with some other open boundary conditions are expressed
explicitly. Our method can be generalized, as a example, to the boundary Calogero–Sutherland model
which is also constructed by the reflection spin-Dunkl operators.
PACS: 75.10.-w; 05.50.Q; 0530
Keywords: Sutherland–Römer model; Reflection spin-Dunkl operator; Asymptotic Bethe ansatz; Reflection
equation
1. Introduction
It is believed that strong correlation effect plays an important role in high Tc

superconductors, heavy fermion systems and low-dimensional systems. In metals the
E-mail address: caojp@phy.nwu.edu.cn (J.-P. Cao).
PII: S 0 5 5 0 - 3 2 1 3 ( 0 2 ) 0 0 8 2 9 - 5
J.-P. Cao et al. / Nuclear Physics B 644 [FS] (2002) 476–494 477
screening of the Coulomb potential is expected to lead to short-range interactions.

However, low carrier density and their reduced mobility do not provide an effective
mechanism for screening and long-range interaction potentials should be considered. Thus
we need to consider the quantum many-particle effect in those cases. Exactly solvable
models are of great help in the understanding of quantum many-particle physics. The
Calogero–Sutherland model (CSM) [1,2] and its generalization [3–5], which describes
a system of N particles on a circle interacting pairwise through long range potentials,
is generating a lot of attention in this connection. In particular, CSM provides a fully
solvable model in which the ideas of fractional statistics can be tested [6]. It is also shown
that the CSM in one dimension and the edge states of the fractional quantum Hall effect
(FQHE) have numerous common features [6]. For instance, in both cases the ground state
is a Jastrow–Slater wave function and the excited states are constructed by multiplying
polynomials to the ground-state wave function. The spectrum of the CSM Hamiltonian
can be interpreted as the energy of a collection of free quasi-particles obeying a generalized
exclusion principle, which is to be consistent with the anyon statistics [7–9]. Besides these,
the spin-chain relatives of these model, i.e., the so-called “Haldane–Shastry” model (HSM)
[10,11] were discovered, and it was subsequently found how to put internal degrees of
freedom into the CSM [12]. The exact solutions of these models have recently become a
very active and exciting topic of research.
The CSM and its generalization have been extensively studied with various methods,
in particular, with the periodic boundary conditions via the asymptotic Bethe ansatz
method (ABAM) [13]. If a model is integrable, it actually suffices to know the asymptotic
behaviour of the incoming and outgoing wave function at long distances. That is to say,
the scattering matrix can be calculated without the explicit knowledge of the many-
particle wave function. The application of the ABAM requires an independent proof of
integrability. An elegant and compact method to construct integrable models is via Dunkl
operators [3,14].
Sutherland–Römer model (SRM), as a generalization of the CSM, describes a one-
dimensional (1D), two-component, integrable quantum many-particle system in which like
particles interact with potential g(g + 1)/ sinh2 (r), while unlike particles interact with
potential −g(g + 1)/ cosh2 (r). The Hamiltonian of the system is defined by [15]
N
∂2
N g(g + 1)(1 + σ z σ z )
j k
N g(g + 1)(1 − σ z σ z )
j k
H =− + − , (1)
j =1
∂xj j =k sinh (xj − xk )
2 2
j =k
cosh (xj − xk )
2
where xj and σjz are the coordinate and z component Pauli matrix of the particle j ,
respectively, N the total number of particles, g the coupling constant. The two kinds of
quasi-particles are distinguished by a quantum number σjz = ±1, which may be thought of
as either spin or charge. When like quasi-particles are near, the repulsive potential increases
as 1/r 2 , while for large separations, both potentials decay exponentially. This system
was first introduced by Calogero [16], who showed it to be integrable. Sutherland [17]
soon afterward showed that the system could be exactly solved, and gave the solution for
a single-component system. Sutherland and Römer discussed the exact solution (in the
thermodynamic limit) for the system with two-component. All the above works only deal
with the systems with periodic boundary conditions.
478 J.-P. Cao et al. / Nuclear Physics B 644 [FS] (2002) 476–494
Impurities and boundary effects play a relevant role in 1D system. The strong coupling
fixed point of impurity models frequently renormalizes to the equivalent open boundary
impurity problem [18]. This is also expected to be the case of models with long range
potentials and could be of relevance for the edge states of the FQHE. Open boundaries
have been studied in the context of BCN type CSM [19–21] and open Haldane–Shastry
spin chain [22].
In this paper, we construct the Hamiltonian of the SRM with open boundary conditions
(see Eq. (2)) by the reflection spin-Dunkle operators and prove its integrability. By solving
the Schrödinger equation and using the ABAM, we obtain the scattering matrix of the
two particles and the reflecting matrix. Then, with the help of quantum inverse scattering
method, we obtain the eigenvalues of the transfer matrix and the Hamiltonian of the system.
This paper is organized as follows. In Section 2 are some descriptions of the reflection
spin-Dunkl operators and the system consider in this paper is constructed. In Section 3,
the two-particle scattering matrix and reflecting matrix are calculated. The Hamiltonian is
diagonalized in Section 4. Section 5 includes a brief summary and some discussions.
2. The reflection spin-Dunkl operators and the system
The SRM with open boundary conditions is defined by the Hamiltonian

N
∂2 N g(g + 1)(1 + σ z σ z )
j l
N g(g + 1)(1 − σ z σ z )
j l
H =− + −
j =1
∂x 2
j j =l
2 sinh2
(x j − x l ) j =l
2 cosh2
(x j − xl )

N g(g + 1)(1 + σ z σ z )
j l

N g(g + 1)(1 − σ z σ z )
j l
+ −
j =l
2 sinh2 (xj + xl ) j =l
2 cosh2 (xj + xl )

N ρ(ρ + 1)(1 + σ z )
j

N ρ(ρ + 1)(1 − σ z )
j
+ − , (2)
j =1
sinh2 (xj ) j =1
cosh2 (xj )
where xj and σjz are the coordinate and the z component Pauli matrix of the particle
j , respectively, N the total number of particles, g and ρ the coupling constants. The
system (2) describes the SRM with boundary fields. By rescaling xj → πxj /(2L), it is
seen that the last two terms describe two boundary fields at 0 and L, respectively. The
terms of sinh(xj + xl ) and cosh(xj + xl ) represent a typical feature of the open boundary
system, which describe the interaction between the j th electron and the mirror image of the
lth electron or vice versa. The inclusion of the image terms is just equivalent to removing
the infinite wall at the boundary.
First, we show the construction and the integrability of the system (2). For a one-
dimensional system of N particles, we define the reflection spin-Dunkl operators

N

Dj = pj + j l + iuj Mj ,
vj l Mj l + v̄j l M (3)
l=1,l=j
where pj (j = 1, 2, . . . , N) are the momenta of the 1D quantum mechanical particles. The
vj l = v(xj − xl , σjz , σlz ), v̄j l = v(xj + xl , σjz , σlz ) and uj = u(xj , σjz ) are yet undetermined
functions. Mj l are the permutation operators,
Mj l = Mlj = Mj†l , Mj2l = 1,

Mj l Aj = Al Mj l , Mj l Al = Aj Mj l , Mj l Ak = Ak Mj l , for k = j, l,
where Aj is any operator (including Mj l themselves) carrying one or more particle indices
including the coordinates and spins. Mj are the reflection operators,
Mj xj = −xj Mj , Mj pj = −pj Mj , Mj σjz = σjz Mj , [Mi , Mj ] = 0.

j l is the combination of the Mj l and Mj ,
M
j l = Mj Ml Mj l ,
M M j l ,
j l uj = u−l M (4)
where u−l = u(−xl , σlz ).
Notice that our definition of Mj l is slightly different from those in the earlier papers,
where Mj l is only the coordinate permutation operator and finally can be mapped onto
the permutation operator Pj l of the internal degrees of freedoms by quantum mechanic
symmetry Mj l Pj l = ±1. While in our definition, Mj l is the total permutation operator.
It not only exchanges the coordinates but also the spins of two corresponding particles
and simply takes the value 1 acting on a boson wave function and −1 on a fermion wave
function.
The commutator of the reflection spin-Dunkl operators is

N
[Di , Dj ] = Uj l − Wj lk , (5)
k=1,k=j,l
where

Uj l = uj vj l + uj vl,−j − ul v̄j,−l − u−l v̄j,l Mj Mj l

− ul vj l + ul v−j,l − uj v̄−j,l − u−j v̄j,l Ml Mj l ,
Wj lk = (vj k vlj − vlj vlk − vlk vj k )Mlk Mj l − (vlk vj l − vj l vj k − vj k vlk )Mj l Mlk

+ vj k v̄lj − v̄lj v−lk − v̄lk v̄j,−k Mj k M lk

+ vj k v̄lk − v̄lk vj,−l + v̄j l v̄−j,k M lk Mj k

+ v̄j k vl,−j − v̄lj v̄−l,k − vlk v̄j k M j k Mlk

+ v̄j l v−j,k + v̄j k v̄l,−k − vlk v̄j l Mlk M j k

+ v̄j k v̄l,−j − vlj v̄lk − v̄lk vj,−k Mj l M j k

+ vj l v̄j k + v̄j k vl,−k − v̄lk v̄j,−l M j k Mj l .
If the reflection spin-Dunkl operators commute with each other, [Dj , Dl ] = 0, we can
define the commutative
family of the conversed quantities by the reflection spin-Dunkl
operators as In = j (Dj )n . If one of the In is chosen as the Hamiltonian, then the model
is integrable. For SRM, we seek for the solutions of the equation, Uj l = Wj lk = 0. After
tedious calculation, we obtain four different solutions. One of them is following
ig
vj l = coth(xj − xl ) − sgn(j − l) 1 + σjz σlz
2
ig
+ tanh(xj − xl ) − sgn(j − l) 1 − σjz σlz , (6)
2
ig ig
v̄j l = coth(xj + xl ) + 1 1 + σjz σlz + tanh(xj + xl ) + 1 1 − σjz σlz , (7)
2 2

uj = ρ coth(xj ) + 1 1 + σjz + tanh(xj ) + 1 1 − σjz , (8)
where g, ρ are coupling constants and the sign function, sgn(j − l) is defined as

1, for j > l,
sgn(j − l) =
−1, for j < l.
It is necessary to point out that permutation operators Mj l and Mj commute with the sign
function. The other three solutions can be found in Appendix A.
We choose I2 as the Hamiltonian

N
N

H= pj2 + j l
vj2l + v̄j2l + vj l Mj l + v̄j l M
j =1 j =l

N
1 N
1
N
+ uj Mj + u2j + Uj l − Wj lk , (9)
2 3
j =1 j =l j =l=k=j
where uj = ∂j uj , vj l = ∂j vj l and v̄j l = ∂j v̄j l . The Hamiltonian (9) can also be explicitly
written as
N
∂2 N g(g − M )(1 + σ z σ z )
jl j l
N g(g − M )(1 − σ z σ z )
jl j l
H =− + −
j =1
∂x 2
j j =l
2 sinh2
(x j − x l ) j =l
2 cosh 2
(x j − x l )

N g(g − M )(1 + σ z σ z )
jl j l

N g(g − M )(1 − σ z σ z )
jl j l
+ −
j =l
2 cosh2 (xj + xl )

N ρ(ρ − M )(1 + σ z )
j j

N ρ(ρ − M )(1 − σ z )
j j
+ − . (10)
j =1
sinh2 (xj ) j =1
cosh2 (xj )
The permutation operators Mj l acting on the bosonic or fermionic subspace of the Hilbert
space will simply become ±1. In addition [Mj , H ] = 0, which means the Hamiltonian
is invariant under reflections. Hence, Mj can be substituted by its eigenvalues ±1 or by
σjz [22]. Here Mj = ±1 corresponds to a scalar impurity potential, while Mj = σjz yields
a scalar potential and a boundary magnetic field. In the case of Mj l = −1 and Mj = −1 of
system (10), we arrive at the Hamiltonian (2).
3. The scattering and reflecting matrix
The asymptotic Bethe ansatz method (ABAM) is a powerful tool to derive the energy for
model with nonlocal interaction. Since the model is integrable, the ABAM requires only
the asymptotic behavior of the wave function at long distances, i.e., the two-body phase-
shift can be obtained without the full knowledge of the many-particles wave function.
We summarize the results in the center of mass frame and Mj l = −1 and Mj = −1
blow. For like particles, the potential is g(g + 1)/ sinh2 (x). Splitting off the center of
mass (R = (x1 + x2 )/2), the Schrödinger equation is straightforwardly reduced to a single
differential equation for the relative coordinate (x = x1 − x2 )

d 2Ψ 1 g(g + 1)
+ E − Ψ = 0, (11)
dx 2 2 sinh2 x
√
where E = 2k 2 with k = (k1 − k2 )/2. If we put ξ = coth x, + = −E = ik and
Ψ = (ξ 2 − 1)+/2w( 12 (1 + ξ )), the Schrödinger equation (11) can be changed into the
hypergeometric equation

u(1 − u)w (u) + γ − (α + β + 1)u w (u) − αβw(u) = 0, (12)
where we have used the notations α = ik − g, β = ik + g + 1, γ = 1 + ik and u = 12 (1 + ξ ).
The solution of Eq. (12) can be expressed in terms of hypergeometric functions as

ik/2 1
Ψ (ξ ) = C1 ξ 2 − 1 F α, β, γ , (1 + ξ )
2
(1−γ )
ik/2 1 + ξ
+ C2 ξ 2 − 1
2

1
× F β − γ + 1, α − γ + 1, 2 − γ , (1 + ξ ) . (13)
2
As x → 0 the wave function has two terms, one diverging as x −g and another is
proportional to x 1+g . The coefficient of the x −g -term has to vanish for the energy
expectation value to be properly defined. This determines the relation between the
constants C1 and C2 ,
C1 1(1 − ik)1(1 + g + ik)
= −(−1)ik . (14)
C2 1(1 + ik)1(1 + g − ik)
Asymptotically as x → −∞, the wave function becomes
Ψ (x → −∞) = C1 2ik eikx + C2 (−2)ik e−ikx . (15)

Therefore the scattering amplitude between like particles is
1(1 − ik)1(1 + g + ik)
S(k) = − . (16)
1(1 + ik)1(1 + g − ik)
For bosons (fermions), the wave function must be (anti)symmetric, so the scattering
amplitude for transmission will be ±S(k). In follows, we will drop factor of −1 in the
scattering amplitudes and this can be taken by the choice of either the statistics of the
particles or the number of particles are even or odd.
For unlike particles, with the potential −g(g + 1)/ cosh2 (x), the Schrödinger equation
is

d 2Ψ 1 g(g + 1)
+ E− Ψ = 0. (17)
dx 2 2 cosh2 x
√
If we put ξ = tanh x, + = −E = ik and Ψ = (1 − ξ 2 )+/2w( 12 (1 − ξ )), Eq. (17) can be
changed into the hypergeometric equation

u(1 − u)w (u) + γ − (α + β + 1)u w (u) − αβw(u) = 0, (18)
where α = ik − g, β = ik + g + 1, γ = 1 + ik and u = 12 (1 − ξ ). The solution of Eq. (18),

at the point u = 12 (1 − ξ ) = 0, can be expressed in terms of hypergeometric functions

ik/2 1
Ψ (ξ ) = C1 1 − ξ 2 F α, β, γ , (1 + ξ )
2
(1−γ )
ik/2 1 − ξ
+ C2 1 − ξ 2
2

1
× F β − γ + 1, α − γ + 1, 2 − γ , (1 − ξ ) . (19)
2
We assume the wave function satisfies the connection condition Ψ (ξ ) = P12 Ψ (−ξ ),
where P12 = 1 for the triplet state and P12 = −1 for the singlet state. P12 is the permutation
operator. We find that if x → ∞, then ξ → 1, 1 − ξ = 2e−2x , while if x → −∞, then
ξ → −1, 1 +ξ = 2e2x . Using the properties of the hyperbolic functions and the 1 functions
1(γ )1(γ − α − β)
F (α, β, γ , z) = F (α, β, α + β + 1 − γ , 1 − z)
1(γ − α)1(γ − β)
1(γ )1(α + β − γ )
+ (1 − z)γ −α−β
1(α)1(β)
× F (γ − α, γ − β, γ + 1 − α − β, 1 − z), (20)
F (α, β, γ , 0) = 1, (21)
π
1(z)1(1 − z) = , (22)
sin πz
πz
1(1 + z)1(1 − z) = , (23)
sin πz
we obtain the asymptotic behavior of the wave function (x → ∞),
C1 2ik e−ikx + C2 (2)ik eikx

1(1 + ik)1(−ik) 1(1 + ik)1(ik)
= P12 C1 e−ikx + P12 C1 eikx
1(1 + g)1(−g) 1(ik − g)1(ik + g + 1)
1(1 − ik)1(−ik) 1(1 − ik)1(ik)
+ P12 C2 e−ikx + P12 C2 eikx .
1(1 + g − ik)1(−g − ik) 1(−g)1(1 + g)
Then, the scattering amplitude between unlike particles is

1(1 + ik)1(1 + g − ik) sin πg sin πik
S(k) = − − P12 .
1(1 − ik)1(1 + g + ik) sin π(ik + g) sin π(ik + g)
The two-particles scattering matrix of the system can be written as
1(1 + i(kj − kl )/2)1(1 + g − i(kj − kl )/2)
Sj,l (kj − kl ) =
1(1 − i(kj − kl )/2)1(1 + g + i(kj − kl )/2)
 
1 0 0 0
0 sin π(i(kj −kl )/2)
 sin π(1+g+i(kj −kl )/2)
sin π(1+g)
sin π(1+g+i(kj −kl )/2) 0
× sin π(i(kj −kl )/2) . (24)
 0 sin π(1+g+i(k
sin π(1+g)
0
j −kl )/2) sin π(1+g+i(kj −kl )/2)
0 0 0 1
This scattering matrix has the symmetry properties
S12 (k) = P12 S12 (k)P12 = S12 (k)t1 t2 ,

where t denotes transpose. Moreover, it satisfies the unitarity relation
S12 (k)S12 (−k) = I,

where I is 2 × 2 unit matrix, and the crossing relation
S12 (k) = −σ y S12 (−k + ig)t2 σ y ,

where σ y is the y component Pauli matrix. One can easily check that this S matrix satisfies
the Yang–Baxter equation [23,24]
S12 (k1 − k2 )S13 (k1 − k3 )S23 (k2 − k3 ) = S23 (k2 − k3 )S13 (k1 − k3 )S12 (k1 − k2 ), (25)
where S12 (k), S13 (k) and S23 (k) act in C n ⊗ C n ⊗ C n with S12 (k) = S(k) ⊗ 1, S23 (k) =
1 ⊗ S(k), etc.
Now, we consider the reflection with the boundary. For the spin-up particles (σ = 1),
the potential is ρ(ρ + 1)/ sinh2 (x). Using similar method, the wave function is given
asymptotically as
Ψ (x → ∞) → Aeikx + Be−ikx . (26)

The reflection amplitude is
A 1(1 + ik)1(1 + ρ − ik)
= . (27)
B 1(1 − ik)1(1 + ρ + ik)
For the spin-down particles (σ = −1), the potential is −ρ(ρ + 1)/ cosh2 (x). We also
assume the wave function satisfies Ψ (x) = Mj Ψ (−x), where Mj is boundary permutation
operators. The reflection amplitude is

Ā 1(1 + ik)1(1 + ρ − ik) sin πρ sin π(ik)
=− − Mj . (28)

B 1(1 − ik)1(1 + ρ + ik) sin π(ik − ρ) sin π(ik − ρ)
Therefore, the reflecting matrix of the system can be written as

1(1 + ikj )1(1 + ρ − ikj ) 1 0
R(kj ) = sin π(ikj )+sin π(1+ρ) . (29)
1(1 − ikj )1(1 + ρ + ikj ) 0 sin π(1+ρ+ikj )
We find that this reflecting matrix R(k) satisfies the reflection equation [25,26]
S12 (k1 − k2 )R1 (k1 )S21 (k1 + k2 )R2 (k2 ) = R2 (k2 )S12 (k1 + k2 )R1 (k1 )S21 (k1 − k2 ),
(30)
where R1 (k) = R(k) ⊗ I , R2 (k) = I ⊗ R(k) and I is the 2 × 2 unit matrix.
4. Exact diagonalization of the system
In the system (2), we assume that the reflecting matrix of the left boundary is the unit
matrix, which means that the particles reflected by the left boundary only obtain a phase
shift π . The periodic motion of particle j consists of its scattering with each of the particles
to the right, reflecting off the right boundary, then scattering with all other particles while it
moves to the left, it is reflected at the left boundary with a phase shift π , and scattering until
it reaches its original position. The transfer matrix then consists of a product of 2(N − 1)
particle–particle scattering matrices and the reflecting matrix of the right boundary. If the
momentum of the particle j (j = 1, 2, . . . , N) is kj and the initial wave function is ξ0 , we
have
Sj,1 (kj − k1 ) · · · Sj,j −1 (kj − kj −1 )Sj,j +1 (kj − kj +1 ) · · · Sj,N (kj − kN )Rj,0 (kj )
× Sj,N (kj + kN ) · · · Sj,j +1 (kj + kj +1 )Sj,j −1 (kj + kj −1 ) · · · Sj,1 (kj + k1 )ξ0
= e2ikj L ξ0 . (31)
The above N eigenvalue equations are simultaneously
solved by Bethe ansatz, diagonaliz-
ing the Hamiltonian with eigenvalue E = N k
j =1 j
2.
We put
− − − − + + + +
X(kj ) = Sj,1 · · · Sj,j −1 Sj,j +1 · · · Sj,N Rj,0 Sj,N · · · Sj,j +1 Sj,j −1 · · · Sj,1 , (32)
±
where Sj,l = S(kj ± kl ). The monodromy matrix of the system is
− − − − −
T (k) = Sτ,j (k)Sτ,1 (k) · · · Sτ,j −1 (k)Sτ,j +1 (k) · · · Sτ,N (k)Rτ,0 (k)
+ + + + +
× Sτ,N (k) · · · Sτ,j +1 (k)Sτ,j −1 (k) · · · Sτ,1 (k)Sτ,j (k), (33)
±
where Sτ,j (k) = S(k ± kj ). The monodromy matrix satisfies the reflection equation
S1,2 (k1 − k2 )T1 (k1 )S1,2 (k1 + k2 )T2 (k2 ) = T2 (k2 )S1,2 (k1 + k2 )T1 (k1 )S1,2 (k1 − k2 ),
(34)
where the T1 (k) and T2 (k) act on the τ1 and τ2 auxiliary space. T1 (k) = T (k) ⊗ I ,
T2 (k) = I ⊗ T (k) and I is the 2 × 2 unit matrix.
−
We find Sτ,j (kj ) = Pτj and
+
trτ T (kj ) = X(kj ) trτ Pτj Sτj (kj )

1(1 + ikj )1(1 + g − ikj ) sin π(1 + g)
= 1+ X(kj ). (35)
1(1 − ikj )1(1 + g + ikj ) sin π(1 + g + ikj )
Therefore, the eigenstate of the X(kj ) can be obtained by constructing the eigenstate of
the monodromy matrix T (k).
Define
U (k) = Sτ,j (k − kj )Sτ,1 (k − k1 ) · · · Sτ,j −1 (k − kj −1 )
× Sτ,j +1 (k − kj +1 ) · · · Sτ,N (k − kN )

A(k) B(k)
= , (36)
C(k) D(k)
which satisfies the Yang–Baxter relation
S12 (k1 − k2 )U1 (k1 )U2 (k2 ) = U2 (k2 )U1 (k1 )S12 (k1 − k2 ). (37)
Then the monodromy matrix can also be written as

−1 α(k) β(k)
T (k) = U (k)Rτ,0 (k)U (−k) = , (38)
γ (k) δ(k)
where
−1 −1
U −1 (−k) = Sτ,N (−k − kN ) · · · Sτ,j +1 (−k − kj +1 )
−1 −1 −1
× Sτ,j −1 (−k − kj −1 ) · · · Sτ,1 (−k − k1 )Sτ,j (−k − kj )
N
1(1 + i(k + kj )/2)1(1 + g − i(k + kj )/2)
=
1(1 − i(k + kj )/2)1(1 + g + i(k + kj )/2)
j =1
sin π((ik + ikj )/2)
× σ y U t −k + i(1 + g) σ y .
sin π((ik + ikj )/2 + 1 + g)
In the derivation, we have used the property of scattering matrix,
−1
Sτ,l (−k − kl ) = Sτ,l (k + kl ). (39)
The transfer matrix is the trace of the monodromy matrix
t (k) = tr T (k) = α(k) + δ(k)
1(1 + ik)1(1 + g − ik) sin π(1 + g + ik) + sin π(1 + g)
= X(k). (40)
1(1 − ik)1(1 + g + ik) sin π(1 + g + ik)
The eigenvalue of X(k) can be obtained from that of the transfer matrix because
1(1 − ik)1(1 + g + ik)
X(k) =
1(1 + ik)1(1 + g − ik)
sin π(1 + g + ik)
× α(k) + δ(k) . (41)
sin π(1 + g + ik) + sin π(1 + g)
The vacuum state of the system is
|Ω = |↑1 ⊗ · · · ⊗ |↑N . (42)

The U (k) and its inverse matrix acting on the vacuum give

N
1(1 + i(k − kj )/2)1(1 + g − i(k − kj )/2)
U (k)11 |Ω = |Ω,
1(1 − i(k − kj )/2)1(1 + g + i(k − kj )/2)
j =1

N
1(1 + i(k − kj )/2)1(1 + g − i(k − kj )/2)
U (k)22 |Ω =
1(1 − i(k − kj )/2)1(1 + g + i(k − kj )/2)
j =1
sin πi((k − kj )/2)
× |Ω,
sin π((ik − ikj )/2 + 1 + g)
U (k)12 |Ω = 0,

N
1(1 + i(k + kj )/2)1(1 + g − i(k + kj )/2)
U −1 (−k)11 |Ω = |Ω,
1(1 − i(k + kj )/2)1(1 + g + i(k + kj )/2)
j =1

N
1(1 + i(k + kj )/2)1(1 + g − i(k + kj )/2)
−1
U (−k)22 |Ω =
1(1 − i(k + kj )/2)1(1 + g + i(k + kj )/2)
j =1
sin πi((k + kj )/2)
× |Ω,
sin π((ik + ikj )/2 + 1 + g)
U −1 (−k)12 |Ω = 0, (43)
where U (k)nm means the mth row and nth column element of the U (k) matrix. In order
to obtain the eigenvalues of the monodromy acting this vacuum state, we must change the
position of C(k1 ) and B(k2 ). From the Yang–Baxter equation (37), we find the following
commutation relation
sin π(1 + g)
C(k1 )B(k2 ) = D(k2 )A(k1) − D(k1 )A(k2 ) + B(k1 )C(k2 ). (44)
sin π(i(k1 − k2 )/2)
From Eqs. (43) and (44), we obtain the eigenvalues of the monodromy matrix acting on
the vacuum state,
N
1(1 + i(k − kj )/2)1(1 + g − i(k − kj )/2)1(1 + i(k + kj )/2)
α(k)|Ω =
1(1 − i(k − kj )/2)1(1 + g + i(k − kj )/2)1(1 − i(k + kj )/2)
j =1
1(1 + g − i(k + kj )/2) 1(1 + ik)1(1 + ρ − ik)
× |Ω,
1(1 + g + i(k + kj )/2) 1(1 − ik)1(1 + ρ + ik)

N
1(1 + i(k − kj )/2)1(1 + g − i(k − kj )/2)1(1 + i(k + kj )/2)
δ(k)|Ω =
1(1 − i(k − kj )/2)1(1 + g + i(k − kj )/2)1(1 − i(k + kj )/2)
j =1
1(1 + g − i(k + kj )/2) 1(1 + ik)1(1 + ρ − ik)
×
1(1 + g + i(k + kj )/2) 1(1 − ik)1(1 + ρ + ik)

N
sin π((ik + ikj )/2) sin π((ik − ikj )/2)
×
sin π((ik + ikj )/2 + 1 + g) sin π((ik − ikj )/2 + 1 + g)
j =1

sin π(ik) + sin π(1 + ρ) sin π(1 + g)
× −
sin π(ik + 1 + ρ) sin π(ik + 1 + g)

sin π(1 + g)
+ |Ω,
sin π(ik + 1 + g)
γ (k)|Ω = 0,
β(k)|Ω = 0. (45)
The eigenstate of the system is constructed by acting the β on the vacuum state (see
Eq. (54)). In order to get the eigenvalues of the transfer matrix, we must exchange the
position of α(k), δ(k) with β(k). From the reflection equation (34), we obtain the following
commutative relations between the elements of the monodromy matrix, which play an
important role in the algebraic Bethe ansatz method
β(k1 )β(k2 ) = β(k2 )β(k1 ), (46)

sin π((ik1 + ik2 )/2) sin π((ik1 − ik2 )/2 − 1 − g)
α(k1 )β(k2 ) = β(k2 )α(k1 )
sin π((ik1 − ik2 )/2) sin π((ik1 + ik2 )/2 + 1 + g)
sin π((ik1 + ik2 )/2) sin π(1 + g)
+ β(k1 )α(k2 )
sin π(1 + g)
− β(k1)δ(k2 ), (47)
sin π((ik1 + ik2 )/2 + 1 + g)
δ(k1 )β(k2 )
sin π((ik1 − ik2 )/2 + 1 + g) sin π((ik1 + ik2 )/2 + 2 + 2g)
= β(k2 )δ(k1 )
sin π(1 + g) sin π((ik1 + ik2 )/2 + 2 + 2g)
− β(k1 )δ(k2 )
sin π(1 + g) sin π(2 + 2g)
− β(k2 )α(k1 )
sin π(2 + 2g) sin π((ik1 − ik2 )/2)
+ β(k1 )α(k2 ). (48)
sin π(1 + g) sin π((ik1 + ik2 )/2 + 1 + g)
In the right-hand side of Eq. (48), behind α(k), we find not only term δ(k) but also α(k).
This will cause trouble in the proceeding of the algebraic Bethe ansatz method, especially
in the case where the thermodynamic limit of this system is taken. In order to solve this
problem, we should reformulate α(k) and δ(k) as α(k) and δ̄(k) so that when commute
δ̄(k) with β(k), only the term δ̄(k) exists behind β(k). Therefore, we define
δ̄(k) = sin π(ik + 1 + g)δ(k) − sin π(1 + g)α(k), (49)

then, Eqs. (47) and (48) can be simplified as

sin π((ik1 + ik2 )/2) sin π((ik1 − ik2 )/2 − 1 − g)
α(k1 )β(k2 ) = β(k2 )α(k1 )
sin π(1 + g)
− β(k1 )δ̄(k2 )
sin π(ik2 + 1 + g) sin π((ik1 + ik2)/2 + 1 + g)
sin π(1 + g) sin π(ik2 )
+ β(k1 )α(k2 ), (50)
sin π((ik1 − ik2 )/2) sin π(ik2 + 1 + g)
δ̄(k1 )β(k2 )
sin π((ik1 − ik2 )/2 + 1 + g) sin π((ik1 + ik2 )/2 + 2 + 2g)
= β(k2 )δ̄(k1 )
sin π((ik1 − ik2 )/2) sin π((ik1 + ik2)/2 + 1 + g)
sin π(1 + g) sin π(ik2 + 2 + 2g)
− β(k1 )δ̄(k2 )
sin π((ik1 − ik2 )/2) sin π(ik2 + 1 + g)
sin π(1 + g) sin π(ik1 + 2 + 2g) sin π(ik2 )
+ β(k1 )α(k2 ). (51)
sin π(ik2 + 1 + g) sin π((ik1 + ik2)/2 + 1 + g)
With the help of Eq. (49), the eigenvalues of the monodromy matrix, Eq. (45) will take
the following form,
N
1(1 + i(k − kj )/2)1(1 + g − i(k − kj )/2)1(1 + i(k + kj )/2)
α(k)|Ω =
1(1 − i(k − kj )/2)1(1 + g + i(k − kj )/2)1(1 − i(k + kj )/2)
j =1
1(1 + g − i(k + kj )/2) 1(1 + ik))1(1 + ρ − ik)
× |Ω,
1(1 + g + i(k + kj )/2) 1(1 − ik)1(1 + ρ + ik)
δ̄(k)|Ω
N
1(1 + i(k − kj )/2)1(1 + g − i(k − kj )/2)1(1 + i(k + kj )/2)
=
1(1 − i(k − kj )/2)1(1 + g + i(k − kj )/2)1(1 − i(k + kj )/2)
j =1
1(1 + g − i(k + kj )/2) 1(1 + ik))1(1 + ρ − ik)
×
1(1 + g + i(k + kj )/2) 1(1 − ik)1(1 + ρ + ik)

N
×
j =1
[sin π(ik) + sin π(1 + ρ)] sin π(ik + 1 + g)
× − sin π(1 + g) |Ω,
sin π(ik + 1 + ρ)
δ̄(kj )|Ω = 0. (52)
The transfer matrix is still a linear combination of α(k) and δ̄(k) under the condition (49),
sin π(ik + 1 + g) + sin π(1 + g) 1
t (k) = α(k) + δ̄(k). (53)
sin π(ik + 1 + g) sin π(ik + 1 + g)
We assume the eigenstate of the system is
|Φ = β(λ1 )β(λ2 ) · · · β(λM )|Ω. (54)
Transfer matrix acting on the eigenstate, we should exchange the position of α(k), δ̄(k) and
β(λ1 )β(λ2 ) · · · β(λM ). Repeatedly using the commutation relations (46), (50) and (51), we
have
t (k)|Φ
sin π(ik + 1 + g) + sin π(1 + g)
=
sin π(ik + 1 + g)

M
sin π((ik + iλm )/2) sin π((ik − iλm )/2 − 1 − g)
× ΦM α(k)|Ω
sin π((ik − iλm )/2) sin π((ik + iλm )/2 + 1 + g)
m=1

M
sin π(1 + g) j
− Φ δ̄(λm )|Ω
sin π(iλm + 1 + g) sin π((ik + iλm )/2 + 1 + g) M
m=1

M
sin π((iλm − iλl )/2 + 1 + g) sin π((iλm + iλl )/2 + 2 + 2g)
×
sin π((iλm − iλl )/2) sin π((iλm + iλl )/2 + 1 + g)
l=m

M
sin π(1 + g) sin π(iλm ) j
+ ΦM α(λm )|Ω
sin π((ik − iλm )/2) sin π(iλm + 1 + g)
m=1

M
sin π((iλm + iλl )/2) sin π((iλm − iλl )/2 − 1 − g)
×
l=m
1
+
sin π(ik + 1 + g)

M
sin π((ik − iλm )/2 + 1 + g) sin π((ik + iλm )/2 + 2 + 2g)
× ΦM δ̄(k)|Ω
m=1

M
sin π(1 + g) sin π(ik + 2 + 2g) j
− Φ δ̄(λm )|Ω
sin π((ik − iλm )/2) sin π(iλm + 1 + g) M
m=1

M
×
l=m

M
sin π(1 + g) sin π(ik + 2 + 2g) sin π(iλm ) j
+ Φ α(λm )|Ω
sin π(iλm + 1 + g) sin π((ik + iλm )/2 + 1 + g) M
m=1

M
× , (55)
l=m
where
ΦM = β(λ1 )β(λ2 ) · · · β(λM ),

j
ΦM = β(λ1 ) · · · β(λj −1 )β(k)β(λj +1 ) · · · β(λM ).
Therefore, the eigenvalue of the transfer matrix t (k) is

N
1(1 + i(k − kj )/2)1(1 + g − i(k − kj )/2)1(1 + i(k + kj )/2)
Λt (k) =
1(1 − i(k − kj )/2)1(1 + g + i(k − kj )/2)1(1 − i(k + kj )/2)
j =1
1(1 + g − i(k + kj )/2) 1(1 + ik)1(1 + ρ − ik)
×
1(1 + g + i(k + kj )/2) 1(1 − ik)1(1 + ρ + ik)

sin π(1 + g + ik) + sin π(1 + g)
×
sin π(1 + g + ik)
M
×
m=1

sin πik + sin π(1 + ρ) sin π(1 + g)
+ −

N
×
j =1

M
× ,
m=1
(56)
if the following Bethe ansatz equations are satisfied
cos π((iλm )/2) cos π(iλm /2 + 1 + ρ)
cos π(iλm /2 + 1 + g) cos π(iλm /2 + 1 + g − 1 − ρ)

M
×
l=m

N
sin π((iλm + ikj )/2) sin π((iλm − ikj )/2)
= ,
sin π((iλm + ikj )/2 + 1 + g) sin π((iλm − ikj )/2 + 1 + g)
j =1
m = 1, 2, . . . , M. (57)
With the help of Eq. (41), we obtain the eigenvalue of the X(k) matrix,

N
1(1 + i(k − kj )/2)1(1 + g − i(k − kj )/2)1(1 + i(k + kj )/2)
ΛX(k) =
1(1 − i(k − kj )/2)1(1 + g + i(k − kj )/2)1(1 − i(k + kj )/2)
j =1
1(1 + g − i(k + kj )/2)
×
1(1 + g + i(k + kj )/2)
1(1 + ik)1(1 + ρ − ik) 1(1 − ik)1(1 + g + ik)
×
1(1 − ik)1(1 + ρ + ik) 1(1 + ik)1(1 + g − ik)
M
×
m=1
sin π(1 + g + ikj )

+
sin π(1 + g + ikj ) + sin π(1 + g)

sin πik + sin π(1 + ρ) sin π(1 + g)
× −

N
×
j =1

M
× .
m=1
(58)
Let k = kj in Eq. (58) and notice that the eigenvalue of X(kj ) is e2ikj L , we obtain
another set of Bethe ansatz equations,
N
1(1 + i(kj − kl )/2)1(1 + g − i(kj − kl )/2)1(1 + i(kj + kl )/2)
e2ikj L =
1(1 − i(kj − kl )/2)1(1 + g + i(kj − kl )/2)1(1 − i(kj + kl )/2)
l=j
1(1 + g − i(kj + kl )/2) 1(1 + ikj )1(1 + ρ − ikj )
×
1(1 + g + i(kj + kl )/2) 1(1 − ikj )1(1 + ρ + ikj )

M
sin π((ikj + iλm )/2) sin π((ikj − iλm )/2 − 1 − g)
× ,
sin π((ikj − iλm )/2) sin π((ikj + iλm )/2 + 1 + g)
m=1
j = 1, 2, . . . , N. (59)
Let iλm → iλm − (1 + g), the Bethe ansatz equations (57) and (59) will take the form of
cos π((iλm − 1 − g)/2) cos π((iλm − 1 − g)/2 + 1 + ρ)
cos π((iλm + 1 + g)/2) cos π((iλm + 1 + g)/2 − 1 − ρ)

M
sin π((iλm + iλl )/2 − 1 − g) sin π((iλm − iλl )/2 − 1 − g)
×
sin π((iλm + iλl )/2 + 1 + g) sin π((iλm − iλl )/2 + 1 + g)
l=m

N
sin π (iλm + iλl − 1 − g) sin π (iλm − iλl − 1 − g)
= 2 2
,
sin π2 (iλm + iλl + 1 + g) sin π2 (iλm − iλl + 1 + g)
l=1
m = 1, 2, . . . , M, (60)

N
1(1 + i(kj − kl )/2)1(1 + g − i(kj − kl )/2)1(1 + i(kj + kl )/2)
e2ikj L =
1(1 − i(kj − kl )/2)1(1 + g + i(kj − kl )/2)1(1 − i(kj + kl )/2)
l=j
1(1 + g − i(kj + kl )/2) 1(1 + ikj )1(1 + ρ − ikj )
×
1(1 + g + i(kj + kl )/2) 1(1 − ikj )1(1 + ρ + ikj )

M
sin π2 (ikj + iλm − 1 − g) sin π2 (ikj − iλm − 1 − g)
× ,
sin π2 (ikj + iλm + 1 + g) sin π2 (ikj − iλm + 1 + g)
m=1
j = 1, 2, . . . , N. (61)
5. Discussion
In conclusion, by constructing the reflection spin-Dunkl operators, we present the one-

dimensional Sutherland–Römer model with open boundary conditions. After proving the
integrability of the system, we obtain the two-particle scattering matrix and the reflection
matrix by using the asymptotic Bethe ansatz method. Then, we obtain the Bethe ansatz
equations and the eigenvalues of the transfer matrix and the Hamiltonian of the system
with the help of quantum inverse scattering method. Using the results presented in this
paper, one may also calculate the boundary free energy, thermodynamic limit and finite
size corrections by analyzing the Bethe ansatz equations and the eigenvalues of the transfer
matrix. Other physical phenomena are also worth of studying such as surface critical
exponents and scaling, the central charges in conformal field theory, etc.
It is a very effective method to construct integrable model via reflection spin-Dunkl
operators. As an example, we propose the following reflection spin-Dunkl operators

N

Dj = pj + ig/2 coth(xj − xl ) − sgn(j − l) σjz + σlz Mj l
l=1,l=j
+ coth(xj + xl ) + 1 σjz + σlz M j l

+ i (β − δ)(coth xj + 1)σjz + 2δ coth(2xj ) + 1 σjz Mj . (62)
One can check that
[Di , Dj ] = 0. (63)
The conserved
quantities can be constructed by the reflection spin-Dunkl operators as
In = j (Dj )n and they commute among themselves. So the system is integrable. The

Hamiltonian is defined as H = j (Dj )2 , which can also be explicitly written as

N
∂2
N g 2 (1 + σ z σ z ) − g(σ z + σ z )M
j l j l jl
H =− +
∂x 2
j =1 j =l
2 sinh2 (xj − xl )
j l
N g 2 (1 + σ z σ z ) − g(σ z + σ z )M
N β(β − σ z M )
j l j l j j
+ +
j =l
2 sinh2 (xj + xl ) j =1
sinh2 (xj )

N δ(δ − σ z M )
j j
− . (64)
j =1
cosh2 (xj )
It is the Hamiltonian of the hyperbolic two-component CSM with open boundary

condition. If σjz = ±1 and σlz = ∓1, the interacting potential is zero, which means that the
spin-up quasi-particles and the spin-down quasi-particles do not interact with each other.
Therefore, the system shares two similar fractional statistics. The Bethe ansatz equations
and the energy spectrum of the system can be calculated straightforwardly by the method
suggested in this paper.
Appendix A. SRM with general open boundary conditions
The SRM with general open boundary conditions can be constructed by the following
reflection spin-Dunkl operators
ig
N

Dj = pj + coth(xj − xl ) − sgn(j − l) 1 + σjz σlz
2
l=1,l=j
+ tanh(xj − xl ) − sgn(j − l) 1 − σjz σlz Mj l
ig
N

+ coth(xj + xl ) + 1 1 + σjz σlz
2
l=1,l=j
+ tanh(xj + xl ) + 1 1 − σjz σlz M j l + iuj Mj , (A.1)
where uj is a undetermined function. The integrability of the system requires that Dj
commutate with each other. Thus, we should seek for the solution of [Di , Dj ] = 0. After
tedious calculation, we found uj may have the following forms besides Eq. (8),

(1) uj = ρ (tanh xj + 1) + (coth xj + 1) , (A.2)
(2) uj = β(tanh xj − coth xj )σj , (A.3)

(3) uj = ρ (tanh xj + 1) + (coth xj + 1) + β(tanh xj − coth xj )σj , (A.4)
where ρ and β are coupling constants. The corresponding Hamiltonians are
H = H0 + Hb , b = 1, 2, 3, (A.5)
where
N
∂2 N g(g − M )(1 + σ z σ z )
jl j l
N g(g − M )(1 − σ z σ z )
jl j l
H0 = − + −
j =1
∂xj2
j =l
2 sinh (xj − xl )
2
j =l
2 cosh (xj − xl )
2

N g(g − M )(1 + σ z σ z )
jl j l

N g(g − M )(1 − σ z σ z )
jl j l
+ − , (A.6)
j =l
2 cosh2 (xj + xl )

N
ρ(ρ − Mj )
N
ρ(ρ − Mj )
H1 = − , (A.7)
j =1
sinh2 (xj ) j =1
cosh2 (xj )

N
β(β − σj Mj )
N
β(β − σj Mj )
H2 = − , (A.8)
j =1
sinh2 (xj ) j =1
cosh2 (xj )

N
ρ(ρ − Mj )
N
ρ(ρ − Mj )
N
β(β − σj Mj )
H3 = − +
j =1
sinh2 (xj ) j =1
cosh2 (xj ) j =1
sinh2 (xj )

N
β(β − σj Mj )
− . (A.9)
j =1
cosh2 (xj )
All these are interesting integrable systems and can be studied straightforwardly by the
method suggested in this paper.
References
[1] F. Calogero, J. Math. Phys. 10 (1971) 2191;

F. Calogero, J. Math. Phys. 10 (1971) 2197.
[2] B. Sutherland, J. Math. Phys. 12 (1971) 246;
B. Sutherland, J. Math. Phys. 12 (1971) 251;
B. Sutherland, Phys. Rev. A 4 (1971) 2019;
B. Sutherland, Phys. Rev. A 5 (1971) 1372.
[3] A.P. Polychronakos, Phys. Rev. Lett. 69 (1992) 703.
[4] K. Hikami, Phys. Rev. B 45 (1992) 7525;
K. Hikami, Phys. Rev. B 46 (1992) 1005.
[5] Z.N.C. Ha, F.D.M. Haldane, Phys. Rev. B 46 (1992) 9359.
[6] F.D.M. Haldane, Bull. Amer. Phys. Soc. 37 (1992) 377;
P.J. Forrester, B. Jancovici, J. Phys. 45 (1984) L583;
A.P. Polychronakos, Nucl. Phys. B 324 (1989) 597;
F.D.M. Haldane, Phys. Rev. Lett. 67 (1991) 937.
[7] Z.N.C. Ha, Nucl. Phys. B 435 (1995) 604.
[8] F. Lesage, Nucl. Phys. B 435 (1995) 585.
[9] P.J. Forrester, Nucl. Phys. B 388 (1992) 671.
[10] F.D.M. Haldane, Phys. Rev. Lett. 60 (1988) 635.
[11] B.S. Shastry, Phys. Rev. Lett. 60 (1988) 639.
[12] K. Hikami, M. Wadati, J. Phys. Soc. Jpn. 62 (1993) 469.
[13] F.D.M. Haldane, Phys. Rev. Lett. 66 (1991) 1529;
B. Sutherland, Phys. Rev. Lett. 75 (1995) 1248;
F.H.L. Essler, Phys. Rev. B 51 (1995) 1375.
[14] F. Dunkl, Trans. Amer. Math. Soc. 311 (1989) 167.
[15] B. Sutherland, R.A. Römer, Phys. Rev. Lett. 71 (1993) 2789.
[16] F. Calogero, O. Ragnisco, C. Marchioro, Lett. Nuovo Cimento 13 (1975) 383.
[17] B. Sutherland, Rocky Mountain J. Math. 8 (1978) 413.
[18] A. Furusaki, N. Nagaosa, Phys. Rev. Lett. 72 (1994) 892;
A.W.W. Ludwig, I. Affleck, Nucl. Phys. B 428 (1994) 545.
[19] M.A. Olshanetsky, A.M. Perelomov, Phys. Rep. 94 (1983) 313;
T. Yamamoto, J. Phys. Soc. Jpn. 63 (1994) 1223;
T. Yamamoto, N. Kawakami, S.K. Yang, J. Phys. A 29 (1996) 317;
H. Frahm, S.I. Matveenko, J. Phys. B 5 (1998) 671;
A.P. Polychronakos, Nucl. Phys. B 543 (1999) 485.
[20] K. Hikami, J. Phys. Soc. Jpn. 65 (1996) 394.
[21] T. Yamamoto, Phys. Lett. A 208 (1995) 293.
[22] D. Bernard, V. Pasquier, D. Serban, Europhys. Lett. 30 (1995) 301;
B.D. Simons, B.L. Altshuler, Phys. Rev. B 46 (1994) 639.
[23] P.P. Kulish, E.K. Sklyanin, in: Lecture Notes in Physics, Vol. 151, Springer, 1982, p. 61.
[24] V.E. Korepin, N.M. Bogoliubov, A.G. Izergin, Quantum Inverse Scattering Method and Correlation
[25] I.V. Cherednik, Theor. Math. Phys. 17 (1983) 77;
I.V. Cherednik, Theor. Math. Phys. 61 (1984) 911.
Euler–Poincaré characteristic and phase transition in

the Potts model on Z2
Philippe Blanchard a , Santo Fortunato b , Daniel Gandolfo a,c,d
a Fakultät für Physik, Theoretische Physik and BiBos, Universität Bielefeld, Universitätsstrasse, 25,
D-33615 Bielefeld, Germany
b Fakultät für Physik, Theoretische Physik, Universität Bielefeld, Universitätsstrasse, 25,
D-33615 Bielefeld, Germany
c PhyMat, Département de Mathématiques, Université de Toulon et du Var, BP 132,
F-83957 La Garde Cedex, France
d CPT, CNRS, Luminy case 907, F-13288 Marseille Cedex 9, France
Received 7 May 2002; received in revised form 11 July 2002; accepted 7 August 2002
Abstract
Recent results concerning the topological properties of random geometrical sets have been
successfully applied to the study of the morphology of clusters in percolation theory. This approach
provides an alternative way of inspecting the critical behaviour of random systems in statistical
mechanics.
For the 2d, q-states Potts model on the square lattice with q 6, intensive and accurate numerics
indicates that the average of the Euler characteristic (taken with respect to the Fortuin–Kasteleyn
random cluster measure) changes sign at the critical threshold of the magnetization transition.
Keywords: Cluster morphology; Euler–Poincaré characteristic; Phase transition
1. Introduction
Recently, new insights in the study of the critical properties of clusters in percolation
theory have emerged based on ideas coming from mathematical morphology [1] and
integral geometry [2–4]. These mathematical theories provide a set of geometrical
and topological measures allowing to quantify the morphological properties of random
E-mail addresses: blanchard@physik.uni-bielefeld.de (P. Blanchard), fortunat@physik.uni-bielefeld.de

(S. Fortunato), gandolfo@cpt.univ-mrs.fr (D. Gandolfo).
PII: S 0 5 5 0 - 3 2 1 3 ( 0 2 ) 0 0 6 8 1 - 8
496 P. Blanchard et al. / Nuclear Physics B 644 [FS] (2002) 495–508
systems. In particular these tools have been applied to the study of random cluster
configurations in percolation theory and statistical physics [5–9].
One of these measures is the Euler–Poincaré characteristic χ which is a well-known
descriptor of the topological features of geometric patterns. It belongs to the finite set
of Minkowski functionals whose origin lies in the mathematical study of convex bodies
and integral geometry (see [2–4]). These measures, as we shall explain below, share
the following remarkable property: any homogeneous, additive, isometry-invariant and
conditionally continuous functional on a compact subset of the Euclidean space Rd can
be expressed as a linear combination of the Minkowski functionals. This is the well-known
Hadwiger’s theorem [2] of integral geometry which has a wide scope of applications in
mathematical physics due its rather general settings.
The use of these measures in image analysis [10], problems of shape recognition [1],
determination of the large scale structures of the universe [11], modelling of porous media
[12], microemulsions [13] and fractal analysis [14] has been a topic of growing interest
recently.
We also recall that for the problem of bond percolation on regular lattices, Sykes and
Essam [15] were able to show, using standard planar duality arguments, that for the case of
self-dual matching lattices (e.g., Z2 ), the mean value of the Euler–Poincaré characteristic
changes sign at the critical point (this even led them to announce a proof for the value of
the critical probability of bond percolation on Z2 ), see also [16].
More recently, Wagner [8] was able to compute the Euler–Poincaré characteristic on
the set of all plane regular mosaics (the 11 Archimedean lattices) as a function of the
site occupancy probability p ∈ [0, 1]] and showed that a close connection exists between
the threshold for site percolation on these lattices and the point where the Euler–Poincaré
characteristic (expressed as a function of p) changes sign.
The aim of this article is to further investigate the role played by this morphological
indicator in statistical physics and to present new results concerning its behaviour in
the case of the 2-dimensional Potts model. Namely we present here clear evidence,
based on Monte Carlo simulations, that for the 2d Potts model, the Euler–Poincaré
characteristic changes sign at the critical point of the thermal transition. Namely we
find that for q = 1, . . . , 4 it changes sign continuously at the transition point while,
for q = 5, 6 it has a first order transition at the critical point. As far as we know,
this is the first example of a discontinuous behaviour of this parameter in a physical
model.
The paper is organized as follows. In Section 2 we introduce basis facts concerning
Minkowski functionals and the necessary definitions for our model, the numerical results
are presented in Section 3 followed by some comments and discussion in Section 4.
2. The model
We first briefly summarize the basic facts from integral geometry and give the
definition of the Minkowski functionals including the Euler–Poincaré characteristic,
see [2–4] for more complete expositions. We will then show how to compute the
Euler–Poincaré characteristic in the case of a random configuration of sites and bonds
P. Blanchard et al. / Nuclear Physics B 644 [FS] (2002) 495–508 497
produced by a Fortuin–Kasteleyn (FK) transformation of the partition function of the Potts

model [17].
2.1. Minkowski functionals
The Euler–Poincaré characteristic χ is an additive functional on subsets of Rd such that

for A, B ⊂ Rd (see example in Fig. 1)
χ(A ∪ B) = χ(A) + χ(B) − χ(A ∩ B) (2.1)

with

1, if A
= ∅, A convex,
χ(A) =
0, A = ∅.
The set of Minkowski functionals is then defined by (see, e.g., [5,10])

Wα (A) = χ(A ∪ Eα ) dµ(Ea ), α = 0, . . . , d − 1, (2.2)
π d/2
Wd (A) = ωd χ(A), ωd = (2.3)
Γ (1 + d/2)
where Eα is an α-dimensional plane in Rd , dµ(Ea ) its density normalized such that, for
the d-dimensional ball Bd (r) with radius r, Wα (Bd (r)) = ωd r d−α and ωd is the volume
of the unit ball in Rd . Obviously, additivity of the Minkowski functionals is inherited from
(2.1), furthermore they are usually conveniently normalized through
ωd−α
Mα (A) = Wα (A).
ωd ωα
The computation of these normalized functionals in dimensions 1, 2, 3 in terms of the
usual geometric measures (length, area, volume, . . . ) is given in Table 1. An important
result of integral geometry is Hadwiger’s completeness theorem which asserts that,
under not too restrictive and furthermore physically reasonable assumptions, namely:
additivity, motion invariance (under translations and rotations) and conditional continuity
(which states that any convex body can be smoothly approximated by convex polyhedra),
Fig. 1. Examples of calculation of the Euler–Poincaré characteristic in dimension d = 2 for some combinations
of convex subsets of R2 .
Table 1
Values of the normalized Minkowski functionals for d-dimensional subsets of Rd , d = 1, 2, 3 in terms of
geometric measures, L: length, S: area, V : volume, C: circumference, H : integral mean curvature, χd : Euler–
Poincaré characteristic
d M0 M1 M2 M3
1 L χ1 /2 ··· ···
2 S C χ2 /π ···
3 V S H /2π 2 3χ3 /4π
any functional M decomposes as a linear combination of the finite set of Minkowski

functionals

d
M(A) = cν Mν (A),
ν=0
where the cν are real coefficients.

This theorem has very important practical consequences, we refer to [2,3,10]. We shall
only mention here the principal kinematic formula (see [3,9]) which has been used to
estimate the critical thresholds in continuum percolation theory [18,19]. It reads
α
α
Mα (A ∪ gB) dg = Mα−β (B)Mβ (A), α = 0, . . . , d,
β
G β=0
where integration is over the group of motions G (i.e., rotations and translations, g =
(r, Θ)). This formula is very useful to calculate mean values of Minkowski functionals
for random distributions of objects. For instance, it can be applied to the computation of
the excluded volume of convex bodies (leading, in case of spherical objects, to Steiner’s
formula) which has important applications in continuum percolation theory (see [18]).
2.2. Euler–Poincaré characteristic of 2d cell complexes
It is one of the topics of algebraic topology to show that the above general definitions of
Minkowski functionals (including Euler–Poincaré characteristic) extend to cell complexes
to which random bond configurations on regular lattices belong.
The square lattice Z2 can be viewed as a cell-complex L = {L0 , L1 , L2 }, where L0 = Z2
is the set of sites, L1 is the set of bonds and L2 is the set of plaquettes (see [20–23]).
To a set of bonds B ⊂ L1 we associate the subcomplex Λ(B) ⊂ L defined as the
maximal closed subcomplex containing B. A subcomplex K0 of a complex K is said to be
closed if every element of K which precedes (i.e., is less than) some element of K0 is itself
an element of K0 . K0 is maximal in the sense that if K0 is a subcomplex of K and K0 is
strictly contained in K0 , then K0 is not closed.
This subcomplex can be written
Λ(B) = Λ0 (B) ∪ Λ1 (B) ∪ Λ2 (B),

where Λ1 (B) = B, Λ0 (B) = {x ∈ L0 | x ∩ B

= ∅}, i.e., it is the set of sites which are
endpoints of bonds of B, and

Λ2 (B) = p = [x, y, z, t] ∈ L2 : xy, yz, zt, tx ∈ B .
In other words Λ2 (B) is the set of plaquettes all of whose bonds belong to B.
We use the notation N 0 (B) ≡ |Λ0 (B)| to denote the number of sites of Λ0 (B), N 1 (B) ≡
|Λ1 (B)| to denote the number of bonds of Λ1 (B) and N 2 (B) ≡ |Λ2 (B)| the number of
plaquettes of Λ2 (B).
The Euler–Poincaré characteristic of Λ(B) is defined by [21]
χ(B) = N 0 (B) − N 1 (B) + N 2 (B). (2.4)
It satisfies the Euler–Poincaré formula [21]
χ(B) = π 0 (B) − π 1 (B) + π 2 (B), (2.5)
where π 0 (B) and π 1 (B) are respectively the number of connected components and the
number of independent cycles of Λ(B). Here π 2 (B) = 0 because Λ(B) is a 2d closed cell-
complex. π 2 (B) characterizes the number of holes that are not homotopy equivalent to
a point (see [21]), a situation that can arise only from dimension d = 3.
It turns out that this formalism of cell complexes, which has been used at length in order
to prove deep results in statistical mechanics [9,22,24–27], is all what we need to compute
the Euler–Poincaré characteristic of FK random bond configurations for the Potts model.
We produce below the results we get in case of the 2d Potts model but the same
formalism extends to higher dimensions.
2.3. Potts model
The partition function for the q-states Potts model on a lattice Λ ⊂ Zd at inverse
temperature β reads

Potts
Zβ,q (Λ) = exp β δ(σi , σj ) , (2.6)
σ i,j ⊂Λ
where the first sum runs over all configurations σ ⊂ {1, . . . , q}|Λ| , the second one is over
each nearest neighbour pair of Potts spins on Λ and δ is the Kronecker symbol. We
remind that, whenever q is large enough, in any dimension d 2, this model exhibits a
unique (inverse) temperature βc where the mean energy is discontinuous (see [24–26]). In
dimension d = 2 for q 5 this is an exact result [28] and it is expected to be true, in d = 3,
for q 3 [29].
After performing the (FK) transformation [17], the partition function (2.6) leads to the
following random cluster representation

N 1 (X) N (X)
FK
Zβ,q (Λ) = eβ − 1 q Λ . (2.7)
X
Here the summation is over all graphs X which can be drawn inside the domain Λ and
NΛ (X) is the number of connected components of X (including isolated sites). As before,
Fig. 2. A bond configuration X ⊂ BΛ with 25 sites, 19 bonds and 2 plaquettes.
we shall call N 1 (X) the number of bonds of the configuration X, N 0 (X) the number of
sites which are endpoints of a bond in X and N 2 (X) the number of plaquettes of the
configuration X, i.e., the set of cells in Λ having 4 occupied bonds on its boundary (see
Fig. 2). Formula (2.4) leads to
χ(X) = N 0 (X) − N 1 (X) + N 2 (X). (2.8)
this expression will allow us to compute the mean value of the Euler–Poincaré character-
istic with respect to the FK measure.
3. Numerical results
We have performed Monte Carlo simulations of the 2-dimensional q-state Potts model
on the square lattice for q ranging from 2 to 6. We have always simulated the models near
the critical temperature, whose value can be exactly determined through the well-known
√
formula [29]: βc (q) = J /kTc (q) = log(1 + q). In order to extract a value of the Euler
characteristic χ as close as possible to the value at the infinite volume limit, we have
taken rather large lattices: for the three models with a continuous transition (q = 2, 3, 4)
we arrived at lattice sizes up to 20002. The algorithm we used is the Wolff cluster update
[30]. The identification of FK cluster configurations has been performed via the Hoshen–
Kopelman algorithm [31]; we always considered free boundary conditions for the cluster
labeling. The calculation of the Euler characteristic is done during the cluster labeling
procedure and takes basically zero CPU time: the number of active bonds is stored during
the main Hoshen–Kopelman sweep of the lattice, the number of sites joined by bonds
is trivially obtained right after the clusters are labeled (it is just the total size of the
clusters containing at least two sites). The determination of the number of plaquettes is
more involved, but it requires only an additional sweep over the elementary squares of the
lattice. This procedure is relatively fast since the analysis of a plaquette stops as soon as a
“broken” bond is found.
The Wolff algorithm is the less efficient the bigger the number q of states. Particularly
dramatic is what happens when one passes from the 2-state (Ising) to the 3-state model:
in the former case, on the 20002 lattice it is enough to perform few updates (5–10) to get
uncorrelated configurations for the cluster variables, in the latter one needs about 1000
updates! Because of that, the simulations for q = 3, 4 on the 20002 lattice were very
slow, and the relative data could not reach a high statistics. This is also the reason why
we used the Hoshen–Kopelman algorithm instead of extracting the information from the
single cluster of the update: since for q > 2 basically all the CPU time is taken by the
update phase of the program, the analysis of all clusters of a given configuration allows us
to improve considerably the statistics of the percolation data without time losses.
If χ changes sign at the threshold, on a finite lattice the values measured at each iteration
would be distributed around zero, provided the lattice is large enough. Therefore one would
see both positive and negative values. For this reason, it is helpful to look at the distribution
of χ . The values of the Euler characteristic we shall refer to are meant per lattice site; this
has the advantage that we can see if and how the data concentrate around some value by
increasing the lattice size. In the following we present separately the results for q = 2, 3, 4
and q = 5, 6.
3.1. Results for the models with a continuous phase transition
The first case we consider here is the Ising model. Fig. 3 shows the χ distribution
for three different lattice sizes: 5002 , 10002 and 20002 . In each case we have taken
100 000 iterations, measuring the variables of interest every 5 updates. The peak of the
distribution shifts towards χ = 0 the larger the lattice. The average values of χ are:
χ(5002) = 0.00101(2), χ(10002) = 0.00053(2), χ(20002) = 0.00024(1). We notice that
the averages are quite small and decrease sensibly if we go to larger sizes, reducing
themselves to about the half when we pass from a lattice to the next one. This approximate
linear scaling of χ with the lattice side suggests that the Euler characteristic at the infinite
volume limit indeed vanishes.
Fig. 3. Distribution of χ for the FK cluster configurations of the 2d Ising model at the critical point.
Fig. 4. Distribution of χ for the FK configurations of the 2d, 3-state Potts model at the critical point.
Fig. 5. Distribution of χ for the FK cluster configurations of the 2d, 4-state Potts model at the critical point.
Let us now examine the case q = 3. In Fig. 4 we again plot the χ distribution for
the same three lattice sizes we have considered for the Ising model. Since we collected
a different number of measurements for the different lattices, for a real comparison of the
distributions we needed to renormalize the total number of measurements on each lattice
to the same value: we decided to renormalize all data sets to the number of measurements
on the 5002 lattice (10 000). The distributions are broader than the Ising ones but they
appear almost exactly centered at χ = 0. The average values are in fact much smaller than
before: χ(5002) = 0.00024(4), χ(10002) = 0.00012(3), χ(20002) is zero within errors.
We then deduce that also for q = 3, χ = 0 at the critical point. To complete our analysis
we studied the case q = 4. In Fig. 5 we present a comparison of the χ distributions for two
lattice sizes, 8002 and 20002 . There is a clear shift of the center of the distribution towards
zero when one goes from the smaller to the larger lattice. The average values of the Euler
characteristic in the two cases are χ(8002 ) = −0.00145(8) and χ(20002) = −0.0006(2).
Also here there is no apparent convergence to some value, even if the lattices are rather
large: |χ| reduces itself to less than its half by changing the lattice size. From all this
we also deduce that the Euler characteristic of the FK clusters of the 2-dimensional 4-state
Potts model vanishes at criticality.
We also found that the variance 1χ of the χ distribution for the Ising model scales
as L−0.95 ≈L−1 , where L is the lattice side; for the 3-state Potts model it seems that
1χ scales as L−0.78 . We notice that, in the Ising case, if the correct scaling behaviour
of 1χ goes indeed as the inverse lattice side, the fluctuations of the Euler characteristic
would behave like the energy ones. But the fluctuations of the energy are proportional
to the specific heat of the system, and the L−1 behaviour is a consequence of the fact
that the specific heat of the 2D Ising model diverges logarithmically at criticality (α = 0).
That could mean that, in the Ising case, the fluctuation of the Euler characteristic diverges
with temperature like the specific heat. This impression is confirmed by the results on the
3-state Potts model: since in that case α/ν = 2/5, the scaling behaviour of the energy
fluctuations with L goes like L−4/5 = L−0.8 , in excellent agreement with our numerical
estimate.
3.2. Results for the models with a first order phase transition
For q > 4 the 2d q-state Potts model undergoes a first order phase transition, i.e., the
thermal variables vary discontinuously at the critical threshold. The magnetization, for
instance, makes a jump, varying from zero to a non-zero value. Because of that, we expect
that the cluster configurations change abruptly at the critical point, and that the cluster
variables exhibit as well discontinuities. In particular, the Euler characteristic may jump
from a value to another.
We analyze here the 5- and 6-state Potts models. In both cases we have performed
simulations on three lattices: 1002 , 2002 and 3002 . In Figs. 6 and 7 we compare
the distribution histograms on the 3002 lattice of the magnetization M and the Euler
characteristic χ at three different temperatures near Tc . We define the magnetization by
taking the excess of sites in the majority spin state (per lattice site) with respect to the
value 1/q in the paramagnetic phase, when all spin states are equally distributed. Therefore
we always measure M > 0 and that removes the degeneracy of the magnetization states
due to the Z(q) symmetry of the Hamiltonian. In this way, if one finds a double peak
structure in some temperature range, one can suspect that the transition is discontinuous.
Looking at the magnetization histograms of the figures one clearly sees the spontaneous
symmetry breaking by reducing the temperature. The double peak structure of M suggests
that the transition is first order, as it is known. The corresponding histograms of the
Euler characteristic show a perfectly analogous pattern. To check whether the transition
is indeed first-order, we determined the temperature βH at which the two peaks of the
Euler characteristic distribution are equally high. For 5-state Potts we found βH (1002) =
1.17343, βH (2002) = 1.17405 and βH (3002) = 1.17422; for 6-state Potts βH (1002) =
1.23763, βH (2002) = 1.23804 and βH (3002) = 1.23812. Successively we analyzed the
scaling of the hump between the two peaks with the linear dimension L of the lattice. The
height of the hump χm decreases with L according to the law log(χm ) ∝ −L2 , which is
Fig. 6. Distribution histograms of the magnetization M and the Euler characteristic χ for the 2-dimensional
5-state Potts model at three different temperatures. The lattice size is 3002 . The behaviour of χ is driven by M.
the typical behaviour at a first order phase change. As the result is valid for the 5-state and
the 6-state Potts model, it is likely to be valid also for q > 6, when the discontinuity of
M at the threshold is sharper. Looking at both figures we remark that the centers of the
peaks of χ look approximately symmetric with respect to zero. If this symmetry exists,
it would be an interesting feature, and at the moment we have no arguments to justify it.
In order to determine with some accuracy the values of χ in the two coexisting phases
we would need to increase considerably the size of the lattice, but the required computer
time would increase dramatically for the reasons we explained at the beginning of this
section.
Fig. 7. Distribution histograms of the magnetization M and the Euler characteristic χ for the 2-dimensional
6-state Potts model at three different temperatures. The lattice size is 3002 . The behaviour of χ is driven by M.
4. Conclusions
This work clearly indicates that the Euler–Poincaré characteristic χ is indeed an

important indicator of the phase transition for the 2d Potts model, to the extent of the
cases (q = 2, . . . , 6) studied here. For this model, it reveals that the topology of cluster
configurations has a deep meaning concerning criticality.
This result is not trivial. Considering the 2d Ising model, the Euler–Poincaré character-
istic computed from the physical clusters of +/− spins changes sign near Tc but definitely
not at the critical point. For the 11 plane mosaics, Wagner has found [8] that χ changes sign
close to (but not right at) the critical percolation point. For several models with continuous
spins, a similar behaviour as the one described in this paper holds. We know already that
the modification of the clusters topology induced by the FK representation has non-trivial
consequences on the critical behaviour of spin systems [32–34].
The fact that χ changes sign at Tc for the 2d Ising model in FK representation can be
understood heuristically in the following way. From Tc up to T = ∞ the system is in its
disordered phase and the only excitations one can get in the FK-bond representation are
made of isolated bonds (the probability to see any plaquette vanishes exponentially). Ap-
plying the Euler–Poincaré formula (2.5), one sees that χ behaves like π 0 (X) times a term
of the order of the volume of the system. However, from Tc down to T = 0, the system
is in its ordered phase and the corresponding FK-configuration is (with high probability)
made of O(1) connected bond components. Missing bonds constitute the excitations and
their number scales with the volume of the system so, using again (2.5), one gets that χ
behaves like −π 1 (X) times a term of the order of the volume of the system. This explains
in the case of the Ising model the change of sign of χ at Tc . For Potts spins, when the
transition is first order (q 3), a similar argument holds in FK representation at the critical
point.
The striking phenomenon is the vanishing of the Euler–Poincaré characteristic at Tc
when the transition is second order.
Of primary interest is of course to understand how to relate the Euler–Poincaré
characteristic to the order parameter of the phase transition. This is probably not a simple
task and has not been done so far, even for models when an exact formulae can be derived
for the Euler–Poincaré characteristic (see [5]). Scaling properties and critical exponent of
this quantity are also subjects of great interest.
Another important question concerns the critical behaviour in gauge models. For
example a similar study could provide some insights concerning the deconfining transition
in SU(N ) gauge theory. Indeed, some works [35] tend to indicate that this transition could
be probed by percolation of some physical clusters related to color fields in lattice QCD.
These models have been thoroughly investigated in the past and the tools coming from
algebraic topology have been of primary importance to uncover profound duality results
concerning their phase structure [23,36].
Other spin systems have to be investigated in order to see whether this property of the
Euler–Poincaré characteristic is shared by models with continuous symmetries such as the
(X–Y )-model or the Widom–Rowlinson model. How does χ behave in spin glasses for
example?
Acknowledgements
We are indebted to H. Wagner who initiated our interest for the topics developed in this
paper and to J. Ruiz for fruitful discussions.
Financial support from the BiBoS Research Center (University of Bielefeld), TMR
network ERBFMRX-CT-970122 and the DFG under grant FOR 339/1-2 are gratefully
acknowledged.
References
[1] J. Serra, Mathematical Morphology, Academic Press, 1982.

[2] H. Hadwiger, Vorlesungen über Inhalt, Oberfläche und Isoperimetrie, Springer, 1957.
[3] L.A. Santalò, Integral Geometry and Geometric Probability, Addison–Wesley, 1976.
[4] R. Schneider, Convex Bodies: The Brunn–Minkowski Theory, Cambridge Univ. Press, Cambridge, 1993.
[5] K.R. Mecke, H. Wagner, Euler characteristic and related measures for random geometric sets, J. Stat.
Phys. 64 (1991) 843.
[6] B.L. Okun, Euler characteristic in percolation theory, J. Stat. Phys. 59 (1990) 523.
[7] J.P. Jernot, P. Jouannot, in: Mathematical Morphology and Applications to Image Processing, ISMM’94, 35,
1994.
[8] H. Wagner, Euler characteristic for Archimedean lattices, 2000, unpublished.
[9] K.R. Mecke, Applications of Minkowski Functionals in Statistical Physics, in: K.R. Mecke, D. Stoyan
(Eds.), Statistical Physics and Spatial Statistics. The Art of Analyzing and Modeling Spatial Structures and
Pattern Formation, in: Lecture Notes in Physics, Vol. 554, Springer, 2000.
[10] S.B. Gray, IEEE Trans. Comput. 20 (1971) 551.
[11] K.R. Mecke, Th. Buchert, H. Wagner, Astron. Astrophys. 288 (1994) 697.
[12] C.H. Arns, M.A. Knackstedt, W.V. Pinczewski, K. Mecke, Euler–Poincaré characteristics of classes of
disordered media, Phys. Rev. E 63 (2001) 31112.
[13] T. Hofsässm, H. Kleinert, J. Chem. Phys. 86 (1987) 3565.
[14] K.R. Mecke, Complete family of fractal dimensions based on integral geometry, submitted to Phys. A.
[15] M.F. Sykes, J.W. Essam, Exact critical percolation probabilities for site and bond problems in two
dimensions, J. Math. Phys. 5 (1964) 1117–1121.
[16] G. Grimmett, Percolation, Springer, 1999.
[17] C.M. Fortuin, P.W. Kasteleyn, On the random cluster model I. Introduction and relation to other models,
Physica 57 (1972) 536.
[18] I. Balberg, Universal percolation thresholds limits in the continuum, Phys. Rev. B 31 (1985) 4053.
[19] G.E. Pike, C.H. Seager, Percolation and conductivity: a computer study, Phys. Rev. B 10 (1974) 1421.
[20] P.S. Alexandroff, Combinatorial Topology, Graylock Press, Rochester, 1956.
[21] M.K. Agoston, Algebraic Topology, Pure and Applied Mathematics, Dekker, New York, 1976.
[22] C.-E. Pfister, Y. Velenik, Random cluster representation of the Ashkin–Teller model, J. Stat. Phys. 88 (1997)
1295.
[23] K. Drühl, H. Wagner, Algebraic formulation of duality transformations for Abelian lattice models, Ann.
Phys. 141 (1982) 225.
[24] R. Kotecky, S. Shlosman, First order phase transition in large entropy lattice models, Commun. Math.
Phys. 83 (1982) 493.
[25] L. Laanait, A. Messager, J. Ruiz, Phase coexistence and surface tension in the Potts model, Commun. Math.
Phys. 105 (1986) 527.
[26] R. Kotecky, L. Laanait, A. Messager, J. Ruiz, The q-states Potts model in the standard Piogov–Sinaï theory,
surface tension and Wilson loops, J. Stat. Phys. 58 (1990) 199.
[27] L. Laanait, S. Miracle-Solé, A. Messager, J. Ruiz, S. Shlosman, Interfaces in the Potts model I. Pirogov–
Sinaï theory of the Fortuin–Kasteleyn transformation, Commun. Math. Phys. 140 (1991) 81.
[29] F.Y. Wu, The Potts model, Rev. Mod. Phys. 54 (1982) 235.
[30] U. Wolff, Collective Monte Carlo updating for spin systems, Phys. Rev. Lett. 62 (1989) 361.
[31] J. Hoshen, R. Kopelman, Percolation and cluster distribution, Phys. Rev. B 14 (1976) 3438.
[32] A. Coniglio, C.R. Nappi, F. Peruggi, L. Russo, Percolation points and critical points in the Ising model, J.
Phys. A 10 (1977) 205.
[33] A. Coniglio, W. Klein, Clusters and Ising critical droplets: a renormalization group approach, J. Phys. A 13
(1980) 2775.
[34] A. Coniglio, H.E. Stanley, D. Stauffer, Fluctuations in the number of percolation clusters, J. Phys. A 12
(1979).
[35] S. Fortunato, H. Satz, Polyakov loop percolation and deconfinement in SU(2) gauge theory, Phys. Lett.
B 475 (2000) 311;
S. Fortunato, F. Karsch, P. Petreczky, H. Satz, Effective Z(2) spin models of deconfinement and percolation
in SU(2) gauge theory, Phys. Lett. B 502 (2000) 321.
[36] F.J. Wegner, Duality in general Ising models and phase transitions without local order parameter, J. Math.
Phys. 12 (1971) 2259.
Spectrum of boundary states in N = 1 SUSY

sine-Gordon theory
Z. Bajnok, L. Palla, G. Takács
Institute for Theoretical Physics, Eötvös University, H-1117 Budapest, Pázmány Péter sétány 1/A, Hungary
Received 24 July 2002; received in revised form 29 August 2002; accepted 10 September 2002
Abstract
We consider N = 1 supersymmetric sine-Gordon theory (SSG) with supersymmetric integrable
boundary conditions (boundary SSG = BSSG). We find two possible ways to close the boundary
bootstrap for this model, corresponding to two different choices for the boundary supercharge. We
argue that these two bootstrap solutions should correspond to the two integrable Lagrangian boundary
theories considered recently by Nepomechie.
PACS: 11.55.Ds; 11.30.Pb; 11.10.Kk
Keywords: Supersymmetry; Sine-Gordon model; Integrable quantum field theory; Boundary scattering;
Bootstrap
1. Introduction
In this paper we consider the N = 1 supersymmetric sine-Gordon theory with

supersymmetric integrable boundary conditions (BSSG). Our aim is to find a closure of
the boundary bootstrap for this model.
The N = 1 supersymmetric sine-Gordon model is the natural supersymmetric extension
of the ordinary sine-Gordon model. It is an integrable field theory with infinitely many
conserved charges [1]. The S matrix of the theory was obtained in [2], while the integrable
and supersymmetric boundary conditions were considered in [3], but it took a while until
the most general integrable supersymmetric boundary interaction was found [4].
E-mail addresses: bajnok@afavant.elte.hu (Z. Bajnok), palla@ludens.elte.hu (L. Palla),

takacs@ludens.elte.hu (G. Takács).
PII: S 0 5 5 0 - 3 2 1 3 ( 0 2 ) 0 0 8 2 1 - 0
510 Z. Bajnok et al. / Nuclear Physics B 644 [FS] (2002) 509–532
As a result of integrability, the boundary scattering factorizes, and the general solution
of the boundary Yang–Baxter equation was found in [5], but the constraint of supersym-
metry was not imposed. Nepomechie was the first to consider supersymmetric boundary
scattering [4], building on previous results obtained in the case of supersymmetric sinh-
Gordon theory [6]. However, no one has exposed the full structure of the solitonic re-
flection amplitude, although the results obtained in the case of the tricritical Ising model
[7] are closely related to this problem. Besides that, the closure of the bootstrap and the
spectrum of boundary states have not been even touched before. Therefore our aim is to
clear up the issue of supersymmetric boundary scattering in the BSSG model and to find
the complete spectrum of boundary states and their associated reflection factors. The main
idea—motivated by the successful description of the bulk scattering—is to look for the
reflection amplitudes in a form where there is no mixing between the supersymmetric and
other internal quantum numbers. This means an ansatz for the reflection amplitudes as a
product of two terms one of which is the ordinary (bosonic) sine-Gordon reflection ampli-
tude, while the other describes the scattering of the SUSY degrees of freedom.
Within the bootstrap procedure, we consider first the solitonic reflection amplitudes
on the ground state and on the first two excited boundaries. The SUSY factors in these
solutions have no poles in the physical strip, thus the masses of boundary states emerging
are the same as in the bosonic theory, however, SUSY introduces a nontrivial degeneracy.
The spectrum of general higher excited boundaries is easily extracted from these results.
We determine also the various reflection amplitudes on these excited boundaries.
There are two ways to close the bootstrap, starting from two different ground state
reflection amplitudes, corresponding to two possible choices of the boundary supercharge.
Both solutions lead to the same spectrum of boundary states, but the reflection amplitudes
are different. In one case the reflections conserve fermionic parity, while in the other they
do not.
The layout of the paper is as follows. Section 2 recalls briefly some important facts
about supersymmetric sine-Gordon theory. In Section 3 the reader is reminded of the
supersymmetric and integrable boundary interactions that can be added to the theory,
and we also discuss of the boundary supercharge and derive a formula relating it to the
boundary Hamiltonian. Section 4 gives a quick review (containing only the most necessary
facts) of the spectrum and reflection factors of the ordinary (nonsupersymmetric) sine-
Gordon model with integrable boundary conditions. In Section 5 we present the main
results of the paper, which is a conjecture for the spectrum and the full set of reflection
factors of the BSSG model. We consider the breather reflection amplitudes in Section 6
and then give our conclusions in Section 7.
2. Bulk SSG theory
The bulk supersymmetric sine-Gordon theory is defined by the classical action

ASSG = dt dx LSSG (x, t),
1 β m2
γ µ ∂µ Ψ + mΨ
LSSG = ∂µ Φ∂ µ Φ + i Ψ Ψ cos Φ + cos βΦ, (1)
2 2 β2
Z. Bajnok et al. / Nuclear Physics B 644 [FS] (2002) 509–532 511
where Φ is a real scalar, Ψ is a Majorana fermion field, m is a mass parameter and β is

the coupling constant. The theory is invariant under an N = 1 supersymmetry algebra and
has infinitely many commuting local conserved charges [1]. These charges survive at the
quantum level and render the theory integrable, which makes it possible to describe the
exact spectrum and the S matrix.
2.1. Spectrum and bulk scattering amplitudes
The spectrum consists of the soliton/antisoliton multiplet, realizing supersymmetry in a

nonlocal way, and breathers that are bound states of a soliton with an antisoliton.
The building blocks of supersymmetric factorized scattering theory were first described
in [10], using an ansatz in which the full scattering amplitude is a direct product of a part
carrying the SUSY structures and a part describing all the rest of the dynamics. The full
SSG S matrix was constructed in [2].

The supersymmetric solitons are described by RSOS kinks Kab (θ ) of mass M and
rapidity θ , where a, b take the values 0, 2 and 1 with |a − b| = 1/2, and describe
1
the supersymmetric structure, while = ± corresponds to topological charge ±1

(soliton/antisoliton). Multi-particle asymptotic states are built as follows

K 1 (θ1 )K 2 (θ2 ) · · · K N
a0 a1 a1 a2 aN−1 aN (θN ) , (2)
where θ1 > θ2 > · · · > θN for an in state and θ1 < θ2 < · · · < θN for an out state. The
two-particle scattering process
1 2
Kab (θ1 ) + Kbc (θ2 ) → Kad2 (θ2 ) + Kdc1 (θ1 )
has an amplitude of the form
a d

SSUSY θ1 − θ2 × SSG (θ1 − θ2 )11 22 , (3)
b c
i.e., the tensor structure of the scattering amplitude factorizes into a part describing the
SUSY structure (which we call the SUSY factor) and another part corresponding to the
topological charge (the bosonic factor).
The bosonic factor coincides with the usual sine-Gordon S matrix1
SSG (u)++ −−
++ = SSG (u)−−
∞

Γ 2(l − 1)λ − λuπ Γ 2lλ + 1 − π
λu
=− (u → −u) ,
l=1 π Γ (2l − 1)λ + 1 − π
Γ (2l − 1)λ − λu λu
sin(λu) 2π 1
SSG (u)+− −+
+− = SSG (u)−+ = SSG (u)++
++ , λ= − ,
sin(λ(π − u)) β 2 2
sin(λπ)
SSG (u)−+ +−
+− = SSG (u)−+ = SSG (u)++
++ , u = −iθ, (4)
sin(λ(π − u))
1 Note that the relation between the parameter λ and the coupling β is different from the sine-Gordon case.
while the SUSY factor is identical to the S matrix of the tricritical Ising model perturbed
by the primary field of dimension 35 [8]:

0 12 1 12 θ π
SSUSY 1 θ = SSUSY 1 θ =2 (iπ−θ)/2πi
cos − K(θ ),
2 0 2 1 4i 4
1 1
0 1 θ
SSUSY 2 θ = SSUSY 2 θ =2 θ/2πi
cos K(θ ),
1 1 4i
0 2 1 2

0 12 1 12
SSUSY 1 θ = S θ = 2(iπ−θ)/2πi cos θ + π K(θ ),
1 0
SUSY 1
2 2
4i 4
1 1
1 0 θ π
SSUSY 2 1 θ = SSUSY 2 1 θ = 2θ/2πi cos − K(θ ),
0 2 1 2 4i 2
∞
1 Γ (k − 1/2 + θ/2πi)Γ (k − θ/2πi)
K(θ ) = √ .
π Γ (k + 1/2 − θ/2πi)Γ (k + θ/2πi)
k=1
As the SUSY factor has no poles in the physical strip, the solitonic amplitudes (3) have
poles at exactly the same locations as the sine-Gordon soliton S matrix. These correspond
to bound states (breathers) Bn of mass
πn
mn = 2M sin , n = 1, . . . , [λ].
2λ
The S matrix of the breathers was first found in [9]. The breathers form a particle multiplet
composed of a boson and a fermion, on which supersymmetry is represented in a standard
way [10,11].
For the ordinary sine-Gordon theory, the correspondence between the Lagrangian theory
and the bootstrap S matrix (4) is very well established. There is much less evidence for the
correctness of the S matrix (3) as the scattering amplitude of SUSY sine-Gordon theory.
Besides the original construction [2] (based on arguments related to N = 1 supersymmetric
minimal models), another indication is that at a particular value of the coupling β where it
is expected to have a restriction to the SUSY version of Lee–Yang theory (superconformal
minimal model SM(2/8) perturbed by the relevant superconformal primary field Φ(1, 3),
which is equivalent to Virasoro minimal model M(3/8) perturbed by the primary field
Φ(1, 5)), the first breather supermultiplet has the same scattering amplitude as predicted
(2)
from RSOS restriction of imaginary coupled a2 Toda theory in [12] (see also [13]).
2.2. Bulk SUSY charges
The bulk theory has two supersymmetry charges of opposite chirality Q and Q, which
together form a Majorana spinor. They act on one-particle states |Ai (θ ) in the following
way [8,10,11]:
√
QAi (θ ) = mi eθ/2 QAi (θ ) , Ai (θ ) = √mi e−θ/2 Q
Q Ai (θ ) ,
are matrices satisfying
where mi are the particle masses and Q, Q
Q2 = 1, 2 = 1.
Q
In the one-particle basis {K0 1 , K1 1 , K 1 0 , K 1 1 } (we omit the upper index , as the
2 2 2 2
SUSY action does not depend on the topological charge), the supersymmetry algebra is
represented by the matrices
   
0 i 0 0 0 i 0 0
 −i 0 0 0   
Q=  =  −i 0 0 0  .
Q
 0 0 1 0 ,  0 0 −1 0 
0 0 0 −1 0 0 0 1
The SUSY algebra of the sine-Gordon theory has a central charge and the matrix
 
1 0 0 0
1  0 
Z = {Q, Q} = 0 1 0 
2  0 0 −1 0 
0 0 0 −1
describes the SUSY central charge in the above basis. This is not to be confused with
the topological charge T of the sine-Gordon solitons, which is represented by the upper
indices . Z can take the values 0 or ±1, and it distinguishes between solitons/antisolitons
mediating from odd to even and from even to odd vacua of the bosonic potential [14] (this
means that using the terminology of [15] the theory is 2-folded).
The above representation of SUSY describes BPS saturated objects. There was a
controversy in the literature whether the solitons in SSG are BPS saturated, since N = 1
SUSY does not protect their mass from acquiring radiative corrections. However, it was
shown in [16] that they remain BPS saturated at one-loop and probably to all orders, due
to anomalous quantum corrections to the classical formula for the central charge Z.
The fermionic parity operator Γ = (−1)F is given by the matrix
 
0 1 0 0
1 0 0 0
Γ = 0 0 0 1.

0 0 1 0
Using the definition of Γ it is possible to specify a basis of pure bosonic and pure
fermionic states for any given (fixed) number of particles. However, the composition
(coproduct) rules of the kink states as given in (2) are not free and therefore in the boson-
fermion internal space supersymmetry acts nonlocally. The action of supersymmetry on
multi-particle states involves braiding factors depending on Γ that are defined by the
coproduct ∆:
∆(Q) = Q ⊗ I + Γ ⊗ Q,
=Q
∆(Q) ⊗ I + Γ ⊗ Q,

∆(Γ ) = Γ ⊗ Γ.
The action on breather states can be derived using the bootstrap, but can also be obtained
from the representation theory of the SUSY algebra. It turns out that the central charge Z
(as well as the topological charge T ) vanishes identically for the breathers. For further
details we refer to [10,11].
3. Boundary SSG theory
3.1. Integrable and supersymmetric boundary interactions
Boundary SSG (BSSG) theory can be described by an action of the form

∞ 0 ∞
ABSSG = dt dx LSSG (x, t) + LB (t) dt,
−∞ −∞ −∞
where LB (t) is a term local at x = 0. We consider boundary interactions which are both
supersymmetric and integrable. The case when LB (t) depends only on the value of Φ and
Ψ at x = 0 was considered in [3], where it was found that supersymmetry and integrability
restricts the boundary interaction to the form

4m β
LB = ± 2 cos Φ ± ψ̄ψ (5)
β 2 x=0
which gives four discrete choices. Here we write the Majorana spinor in component form

ψ̄
Ψ= .
ψ
However, this is not the whole story. The Majorana fermion in the ultraviolet limit is
described by the c = 12 Ising conformal field theory. It was noted in [17] that in order
for the Majorana fermion to describe correctly the boundary states of the conformal field
theory, one has to include a fermionic boundary degree of freedom a(t) that is related to
the boundary value of the Ising spin. Using this fact, a two-parameter set of integrable
supersymmetric boundary conditions was derived in [4]. The boundary interaction term is
of the form

L±
B = ±ψ̄ψ + ia∂t a − 2f± (Φ) a(ψ ∓ ψ̄) + B (Φ) x=0 . (6)
The functions f and B are fixed by the requirement of boundary integrability and
supersymmetry (cf. [4]):
m β
B(Φ) = α cos (Φ − Φ0 ),
β2 2
√
Cm 1
f+ (Φ) = sin (βΦ − D) , f− (Φ) = if+ (Φ)|α→−α ,
2 4

βΦ0 D α sin βΦ 0
C = α 2 − 8α cos + 16, tan = 2
.
2 2 α cos βΦ0 − 4 2
We denote the theories obtained by adding L± ±
B to the bulk action as BSSG . The most
important property of this action is that it depends on two continuously varying boundary
parameters α and Φ0 , exactly as in the case of the nonsupersymmetric boundary sine-
Gordon theory [18]. This is important for consistency with the bootstrap since the reflection
factors we find depend on two parameters as well. The boundary Lagrangian (5) can be
obtained as a special case of (6) when the parameters are tuned so that f = 0, and therefore
the boundary fermion a decouples.
3.2. The boundary SUSY charge and the Hamiltonian
can be written as integrals of local (fermionic)

The bulk SUSY charges Q and Q
densities q and q̄:
∞ ∞
Q= q(x, t) dx, =
Q q̄(x, t) dx.
−∞ −∞
They have the anticommutation relation

= 2MZ
{Q, Q}
and satisfy the following relation (among others)
QQ + Q Q = 2H, (7)

where H = h(x, t) dx is the Hamiltonian, Z is the SUSY central charge and M is the
soliton mass. In a boundary theory with supersymmetric integrable boundary condition,
the conserved supercharge can be written as follows:
0

± =
Q q(x, t) ± q̄(x, t) dx + QB (x = 0, t),
−∞
where QB is a boundary contribution, localized at x = 0. There are two possible choices

(±) corresponding to the two possible conformal fermionic boundary conditions in the
ultraviolet limit. As shown in [4], these correspond to the two choices of sign in (6) and
therefore to BSSG± .
Similarly, the Hamiltonian takes the form
0
=
H h(x, t) dx + HB (x = 0, t),
−∞
where HB is the boundary interaction. Let us for the moment neglect the contribution from
the central charge Z (this is possible, e.g., in a sector containing only breathers). Then,
using Eq. (7) it is easy to see that
0
2±
Q =2 h(x, t) dx + 2HB (x = 0, t),
−∞
± is conserved in time, and

where HB is again some term localized at x = 0. However, Q
therefore
0
d
2 h(x, t) dx + 2HB (x = 0, t) = 0.
dt
−∞
In the supersymmetric boundary sine-Gordon theory, this property uniquely determines

HB as the boundary interaction HB and therefore
2± = 2H
Q .
Including Z, it is natural to expect that this relation extends as follows:
2± = 2(H
Q ± M Z),
(8)
is an appropriate extension of Z to the boundary situation. We shall see that the
where Z
two bootstrap solutions we propose correctly reproduce this formula.
4. Boundary sine-Gordon model
4.1. Ground state reflection factors
The most general reflection factor—modulo CDD-type factors—of the soliton/antisoli-

ton multiplet |s, s̄ on the ground state boundary, denoted by | , satisfying the boundary
versions of the Yang–Baxter, unitarity and crossing equations was found by Ghoshal and
Zamolodchikov [18]:
+
P (η, ϑ, u) Q(η, ϑ, u)
RSG (η, ϑ, u) =
Q(η, ϑ, u) P − (η, ϑ, u)
+
P0 (η, ϑ, u) Q0 (u) σ (η, u) σ (iϑ, u)
= − R0 (u) ,
Q0 (u) P0 (η, ϑ, u) cos(η) cosh(ϑ)
P0± (η, ϑ, u) = cos(λu) cos(η) cosh(ϑ) ∓ sin(λu) sin(η) sinh(ϑ),
Q0 (u) = − sin(λu) cos(λu), (9)
where η and ϑ are the two real parameters characterizing the solution,
∞

Γ 4lλ − 2λuπ Γ 4λ(l − 1) + 1 − π
2λu
R0 (u) = (u → −u)
l=1
Γ (4l − 3)λ − 2λu
π Γ (4l − 1)λ + 1 − 2λu
π
is the boundary condition independent part and
σ (x, u)
cos x
=
cos(x + λu)
1 x
∞ λu
Γ 12 + πx + (2l − 1)λ − λu
π Γ 2 − π + (2l − 1)λ − π
× 1 x (u → −u)
l=1
Γ 12 − πx + (2l − 2)λ − λu
π Γ 2 + π + 2lλ − π
λu
describes the boundary condition dependence. The reflection factors of the breathers can
be obtained by the bulk bootstrap procedure [19].
4.2. The general spectrum and the associated reflection factors
The spectrum of boundary excited states was determined in [20,21]. It can be

parametrized by a sequence of integers |n1 , n2 , . . . , nk , whenever the π2 νn1 >
wn2 > · · · 0 condition holds, where
η π(2n + 1) η π(2k − 1)
νn = − and wk = π − − .
λ 2λ λ 2λ
The mass of such a state is

m|n1 ,n2 ,...,nk = M cos(νni ) + M cos(wni ). (10)
i odd i even
The reflection factors depend on which sectors we are considering. In the even sector, i.e.,
when k is even, we have

Q|n1 ,n2 ,...,nk (η, ϑ, u) = Q(η, ϑ, u) ani (η, u) ani (η̄, u),
i odd i even
and

P|n±1 ,n2 ,...,nk (η, ϑ, u) = P ± (η, ϑ, u) ani (η, u) ani (η̄, u),
i odd i even
for the solitonic processes, where
n
η
an (η, u) = 2 −l ; η̄ = π(λ + 1) − η
π
l=1
and
y+1 y−1 u
sin + xπ
{y} = y+1
2λ 2λ
y−1 , (x) = 2 2
xπ .
2λ − 1 2λ +1 sin u2 − 2
In the odd sector, i.e., when k is odd, the same formulae apply if in the ground state
reflection factors the η ↔ η̄ and s ↔ s̄ changes are made. The breather sector can be
obtained again by bulk fusion.
5. Supersymmetric boundary sine-Gordon model
5.1. Ground state reflection factors
5.1.1. The general solution for the reflection factor

Following the bulk case, we suppose that the reflection matrix factorizes as
RSUSY (θ ) × RSG (θ ).
In this special form the constraints as unitarity, boundary Yang–Baxter equation and
crossing-unitarity relation [18] can be satisfied separately for the two factors. Since the
sine-Gordon part already fulfills these requirements, we concentrate on the supersymmetric

part.
From the RSOS nature of the bulk S-matrix (3) it is clear that the boundary must also
have RSOS labels and the adjacency conditions between the nearest kink and the boundary
must also hold. Thus the following reflections are possible:

Kba (θ )|Ba = b
Rac (θ )Kbc (−θ )|Bc
c
or in detail
K0 1 (θ )|B 1 = R 01 1 (θ )K0 1 (−θ )|B 1 ; K1 1 (θ )|B 1 = R 11 1 (θ )K1 1 (−θ )|B 1

2 2 2 2 2 2 2 2 2 2 2 2
and
1 1
K 1 a (θ )|Ba = Raa
2
(θ )K 1 a (−θ )|Ba + Rab
2
(θ )K 1 b (−θ )|Bb ,
2 2 2
b = a, a, b = 0, 1 (11)
In the second process the label of the boundary state has changed, which shows that |B0
and |B1 form a doublet. All of the constraints mentioned above factorize in the sense
that they give independent equations for the reflections on the boundary |B1/2 and on the
doublet |B0,1 . Since the ground state boundary is expected to be nondegenerate we first
concentrate on reflection factors off the singlet boundary |B1/2 . The most general solution
of the boundary Yang–Baxter equation is of the form [5]

R 01 1 (θ ) = 1 + A sinh(θ/2) M(θ ); R 11 1 (θ ) = 1 − A sinh(θ/2) M(θ )
2 2 2 2
while unitarity and crossing symmetry give the following restrictions

M(θ )M(−θ ) 1 − A2 sinh2 (θ/2) = 1,

iπ θ iπ iπ
M − θ = cosh − K(2θ )2 (iπ+2θ)/2πi
M +θ .
2 2 4 2
We suppose that the boundary states |Ba , a = 0, 1 can be obtained by boundary bootstrap
from the ground state |B1/2 [7,22]. Therefore we do not consider the Yang–Baxter
1/2
equation and the other constraints for the amplitudes Rab (θ ) as these will be guaranteed
to be fulfilled by the bootstrap. We shall see later that in general the boundary states |Ba ,
a = 0, 1 come in multiple copies, each of which forms a doublet of states with the same
energy.
5.1.2. Action of the boundary supersymmetry charges Q±

We need to construct the action of the boundary supercharge on the asymptotic states.
We expect that the action is given by
± = Q ± Q
Q + QB ,
where Q, Q act on the particles as in the bulk theory (Section 2.2).2 The reason for this
is that they are given by integrals of local (fermionic) densities, and asymptotic particles
are localized far away from the wall, so the action of these charges is not affected by the
presence of the boundary. QB is the action of the boundary contribution, which we take to
be
QB = γ Γ, (12)
where γ is some unknown parameter (related to the energy of the boundary ground
state—see later). The reason for this choice is that we expect the boundary supercharge to
commute with the bulk S-matrix, which is symmetric under the action of Q, Q and Γ by
construction [10,11]. (12) is also supported by classical considerations in [4], showing also
that the classical version of γ is a function of the parameters in the boundary Lagrangian
(6).
Next we need to give the action of Q, Q and Γ on the boundary ground state |B1/2 .
Following [7], we choose

QB 1 = 0, B 1 = 0,
Q Γ B 1 = B 1 . (13)
2 2 2 2
The first two relations express that the boundary ground state is supersymmetric, while
the last one shows that it is an eigenvector of Γ . We expect that because the ground
state is nondegenerate. The choice of the eigenvalue (±1) is not important, as it could
be compensated by a redefinition of γ . It is a consequence of (8) and (13) that the ground
± on
state energy is γ 2 /2, which will be shown later to be consistent with the action of Q
the excited boundary states.
5.1.3. Supersymmetric reflection amplitudes

Now we would like to impose the supersymmetry constraints on the ground state
reflection amplitudes.
The two choices Q ± will give different solutions [7,22]. If the boundary supercharge
+ commutes with the reflections (theory BSSG+ ) then we obtain
Q
∞

Γ k − 2πiθ
Γ k − 2πiθ
−θ/πi
R 1 1 (θ ) = R 1 1 (θ ) = 2
0 1
{θ ↔ −θ }
2 2 2 2
k=1
Γ k − 14 − 2πi
θ
Γ k + 14 − 2πi
θ
= 2−θ/πi P (θ ). (14)
− that commutes with the reflections (BSSG ) then the result is
If, however, it is Q −

ξ θ
R 1 1 (θ ) = cos + i sinh
0
K(θ − iξ )K(iπ − θ − iξ )2−θ/πi P (θ ),
2 2 2 2

ξ θ
R 11 1 (θ ) = cos − i sinh K(θ − iξ )K(iπ − θ − iξ )2−θ/πi P (θ ), (15)
2 2 2 2
differs by a sign from that used in [7], while agrees with the convention in [11]. This
2 Note that our charge Q
is important when comparing our results with those in [7].
where ξ is related to γ as
√ ξ
γ = −2 M cos . (16)
2
Note that symmetry of the reflection under Γ requires
R 01 1 (θ ) = R 11 1 (θ ),
2 2 2 2
thus in the first case (BSSG+ ) the reflections also commute with the operator Γ , while in
the other case (BSSG− ) they do not. We remark that there are no poles in the physical strip
in any of the reflection factors above.
In the case BSSG+ , the supersymmetry constraints do not determine the value of γ
in contrast to the results of [7,22]. The reason is that it is the supersymmetry of the
reflections on |Ba , a = 0, 1 which connects γ with a parameter in the reflection matrix
itself. However, we construct these reflections by the bootstrap, which determines them
completely, and γ is left as a free parameter. As it was argued above and as will also be
seen later γ is connected to the vacuum energy, so it is not a new parameter of the theory
(in principle it is expressible in terms of the Lagrangian parameters). The only independent
parameters introduced by the boundary are η and ϑ which are present in the bosonic sine-
Gordon reflection factors RSG .
5.2. The general spectrum and the associated reflection factors
5.2.1. The Γ symmetric case (BSSG+ )

We start with the analysis of the ground state reflection factors
R a1 1 (θ ) × RSG (θ ),
2 2
where the SUSY component has the form (14). Since the only poles of these reflection
factors are due to the sine-Gordon part their explanation has to be similar to that in the
bosonic theory. However, we have to supplement the formulae for the bosonic theory
with RSOS indices in a consistent way. The sine-Gordon reflection factor has boundary
independent poles at i nπ2λ for n = 1, 2, . . . , which can be described by Fig. 1. This
is identical to the nonsupersymmetric diagram except that it is decorated with RSOS
indices, which are displayed inside circles. Clearly the dashed line denotes the full breather
supermultiplet (now consisting of a boson and a fermion).
The boundary dependent poles of RSG are located at
η (2n + 1)π
−iθ = νn = − , n = 0, 1, . . . .
λ 2λ
At the position of these poles we associate boundary bound states to the reflection
amplitudes R a1 1 , a = 0, 1
2 2

1 1 1
|a, 1/2|n = |1/2
Ka 1 (iνn ) , where ≡ |B 1 , (17)
g|a,1/2|n 2 2 2 2
Fig. 1. Soliton bulk pole.
where the g-factor is the SUSY part of the boundary coupling, coming from the SUSY
component of the reflection factor (for definitions of boundary couplings, see [18]). The
two states (a = 0, 1) for a given n form a doublet which realizes the structure (11), that
is the K 1 a kinks can scatter on it. The action of the boundary supercharge on these states
2
can be calculated using the coproduct rules in [10,11], taking into account the action of the
charges on the boundary ground state:

√ νn

Q+ |0, 1/2|n = r γ − 2i M cos |1, 1/2|n,
2

√ νn

Q+ |1, 1/2|n = r −1
γ + 2i M cos |0, 1/2|n,
2
|1/2
g|1,1/2|n
r= |1/2
. (18)
g|0,1/2|n
The boundary supercharge satisfies
2
2+ |a, 1/2|n = 2 γ + M cos νn + M |a, 1/2|n
Q
2
which is exactly the relation Q 2+ = 2(H + M Z),
since the central charge of this state is

Z = 1 (we take the ground state |1/2 to have Z = 0, while the bulk soliton Ka1/2 has
Z = 1), the ground state has energy γ /2 by virtue of the relations (12), (13) and M cos νn
2
is the energy that the excited state has relative to the ground state (10).
The SUSY reflection factors of K 1 a off |a, 1/2|n can be computed from the bootstrap
2
principle (Figs. 2 and 3):

1 1 1 1 x 1
b

2 2
ga Rab (θ ) = gb
2 2 2
S (θ − iνn )S (θ + iνn )R 1 1 (θ ) ,
x
x=0,1
a 12 x 12 2 2
1
|1/2
where ga2 ≡ g|a,1/2|n . (19)
Fig. 2. Bootstrap I. Fig. 3. Bootstrap II.
The result turns out to be
1
1 gb2 νn θ
Rab (θ ) = P (θ )K(θ + iνn )K(θ − iνn ) 1 δab cos
2
+ δa,1−b sin . (20)
2 2i
ga2
Note the appearance of the g factors in the result. They are the SUSY parts of the boundary
couplings and come in two types: one corresponds to the absorption of the particle while
creating a higher excited boundary state, the other describing the emission of the particle
and transition to some lower excited boundary state. The ones above are of the absorption
type. The residues of the full reflection factor are described by the product of an emission
and an absorption type full boundary coupling (for a definition of boundary couplings and
their relation to the residue of the reflection factor see [18]). Due to the tensor product
structure the full boundary coupling is given by the bosonic part multiplied with the SUSY
g-factor, as in the case of bulk scattering [11]. The product of the appropriate emission
and absorption SUSY g-factors is constrained to coincide with the value of the SUSY part
of the reflection factor at the position of the pole in the bosonic factor. It can be seen in
general that this does not give enough constraints to determine their value unambiguously
due to the degeneracy introduced by the RSOS indices a, b, and as no physical quantity
should explicitly depend on their value (see the example of the relation Q2+ = 2(H + M Z)

discussed above) we do not present any solution for them.
Being constructed by the bootstrap, the reflection factors (20) necessarily satisfy the
constraints of boundary factorization and crossing-unitarity; in addition, they commute
with Q+ which is guaranteed by the fact that the action of the boundary supercharge is also
derived from the bootstrap as in (18). The full reflection factor on the |a, 1/2|n excited
boundary can be obtained by multiplying this result with the appropriate excited bosonic
reflection factor:
1 1
±
2
Rab (θ ) × Q|n (η, ϑ, θ ) 2
or Rab (θ ) × P|n (η, ϑ, θ ). (21)
Clearly (20) has neither pole nor zero in the physical strip. So the poles of the reflection
factors on the first excited wall (21) are exactly the same as in the nonsupersymmetric
theory: that is they are at iνk or at iwm .
The decoration of the nonsupersymmetric diagrams shows, that Fig. 5 explains the ν
type of poles, while Fig. 4 explains the w type, but only for wm > νn . For wm < νn we
have a boundary bound state which we denote by

1
, a, 1 m, n = 1
K 1 a (iwm )|a, 1/2|n,
2 2 |a,1/2|n 2
g1 1
| 2 ,a, 2 |m,n
so this is also a doublet, but now it is the Ka 1 type kinks that are able to reflect on it. It
2
can be checked easily that the relation Q 2+ = 2(H + M Z)
holds for these states as well,
consistently with the previous interpretation of γ (for these states Z = 0).
At this point the question emerges whether the two states a = 0, 1 forming the doublet
| 12 , a, 12 |m, n are physically different or there is a possibility for some identification
so that a single state can explain the pole in the reflection matrix (21). This can be
decided by examining whether one can describe the residue of the reflection factor with a
1/2 1/2 1/2 1/2
single intermediate state, which implies a relation between the R00 , R11 , R10 and R01
components of the reflection factor at the pole. This relation is violated (for generic values
of the parameters) and so one must really introduce the two states above.
Following the same analysis we performed in [21], but now using a decorated version
of the Coleman–Thun diagrams it can be seen that the poles in the reflection matrix on the
above boundary excited state, which cannot be explained by Coleman–Thun diagrams are
located at iνk . Since the poles appear in association with both reflection factors R b1 1 (θ ),
22
the corresponding boundary states, which are denoted by

1
b, , a, 1 k, m, n
2 2
have a fourfold degeneracy.
Fig. 4. w type poles. Fig. 5. ν type poles.

Fig. 6.
It is clear that the general boundary bound state has the structure

ak . . . 1 , a1 , 1 nk . . . , m1 , n1 or 1 , ak . . . 1 , a1 , 1 mk , nk . . . , m1 , n1 . (22)
2 2 2 2 2
From this we see that in the supersymmetric case the boundary excited states have a
nontrivial degeneracy in contrast to the bosonic theory. The degeneracy is labeled by RSOS
sequences starting from 1/2. In both states in (22) the labels ai can freely take the values
0 and 1, and, as a result, the degeneracy of the states is 2k . The associated reflection
factors can be computed from successive application of the bootstrap procedure, which
is illustrated on Fig. 6
The result depends on the Z charge of the scattering particles. In the Z
= 0 case the
result can be written in the following form
1 |bk ... 12 ,b1 , 12 |nk ...,m1 ,n1 1

k−1
aa
R 2 (θ ) = Ra21 b1 (θ ) fbiibi+1
i+1
(wmi , νni+1 , θ ), (23)
|ak ... 12 ,a1 , 12 |nk ...,m1 ,n1
i=1
where fba11ba22 (wm1 , νn2 , θ ) is the contribution of the dotted square summing over x1 = 0, 1
that is
fba11ba22 (wm1 , νn2 , θ )
x1 1
x1 1
= S 1 2
(θ − iwm1 )S 1
(θ + iwm1 )
2
x1 =0,1 2 a1 2 b1
1 1
x1 b2
×S 2
1
(θ − iνn2 )S 2
1
(θ + iνn2 ).
a2 2 x 1 2
Collecting the common factors we have
fba11ba22 (wm1 , νn2 , θ ) = K(θ − iwm1 )K(θ + iwm1 )K(θ − iνn2 )K(θ + iνn2 )
|a1 , 12 |n1 | 12 a1 , 12 |m1 ,n1
g g
| 12 a1 , 12 |m1 ,n1 |a2 , 12 a1 , 12 |n2 ,m1 ,n1 a1 a2
× hb1 b2 (wm1 , νn2 , θ ),
{a ↔ b}
where
hab11 ab22 (wm1 , νn2 , θ )

θ − iwm1 π θ + iwm1 π
=2 cos + (1 − 2δx,a1 ) cos + (1 − 2δx,b1 )
4i 4 4i 4
x=0,1

θ − iνn2 π θ + iνn2 π
× cos − (δx,a2 ) cos − (1 − δx,b2 ) .
4i 2 4i 2
= 1 case, which is indicated with dotted lines on the diagram, the result contains
In the Z
an extra factor
| 21 ,bk ... 12 ,b1 , 12 |mk ...,m1 ,n1 x 1 |bk ... 12 ,b1 , 12 |nk ...,m1 ,n1
R xk (θ ) = hakk ,bk R 2 (θ ) , (24)
| 21 ,ak ... 12 ,a1 , 12 |mk ...,m1 ,n1 |ak ... 12 ,a1 , 12 |nk ...,m1 ,n1
where
|ak 12 ...
g
θ | 12 ak−1 ... θ π
hxakk ,bk = 2− iπ − sin − (δxk ,ak + δxk ,bk )
|bk 1 ... 2i 2
g1 2
| 2 bk−1 ...

wmk π
+ cos + (δxk ,ak − δxk ,bk ) .
2 2
5.2.2. The Γ nonsymmetric (BSSG− ) case

The discussion of the Γ non symmetric (BSSG− ) solution runs entirely parallel to the
previous Γ symmetric case, the only difference being that in the input of the bootstrap
procedure, i.e., in the ground state reflection amplitude
R a1 1 (θ ) × RSG (θ )
2 2
the supersymmetry factors, R a1 1 (θ ), a = 0, 1, are taken now from (15). These factors
2 2
depend explicitly on γ , and this dependence pertains in the (SUSY) reflection amplitudes
on exited boundaries. Nevertheless, since none of these amplitudes has a pole in the
physical strip, following the steps of the previous considerations leads to the same
conclusion regarding the indexing and degeneracies of the boundary states. Therefore we
concentrate here mainly on the differences between the two solutions.
At the position of the νn poles in the ground state reflection amplitude we again associate
boundary bound states |a, 1/2|n to R a1 1 a = 0, 1 as in (17) (though of course the present
2 2
values of boundary couplings may differ from the previous ones). The action of the present
boundary supercharge, Q − , on these states is

√ νn

Q− |0, 1/2|n = r γ + 2 M sin |1, 1/2|n,
2

√
− |1, 1/2|n = r −1 γ − 2 M sin νn |0, 1/2|n.
Q
2
The action of Q 2− on these states is compatible with the relation Q 2− = 2(H − M Z),

2
provided we keep the interpretation of γ /2 as the ground state energy.
Using the SUSY factors Eq. (15) in the bootstrap equation (19) for the reflections of the
K 1 a kinks on these boundary states gives
2

1
1 gb2 ξ νn θ θ
Rab (θ ) = Z− (θ ) 1 δab cos cos
2
+ (−1) i sinh cosh
a
2 2 2 2
ga2

θ ξ νn
− iδa,1−b sinh cos + (−1)b sin ,
2 2 2
where ξ is expressed in terms of γ in Eq. (16), and
Z− (θ ) = P (θ )K(θ + iξ )K(θ − iξ )K(θ + iνn )K(θ − iνn ).

Although this reflection amplitude has a slightly more complicated form than the one in
Eq. (20), it also solves the boundary Yang–Baxter equation and the other constraints by
construction. The difference between the two comes from the fact that (20) commutes with
Q − .
+ , while the present reflection factor is invariant under Q
Finally we point out that the expressions (23), (24) for the kink reflection amplitudes on
the general higher excited boundary states remain valid in the case BSSG− as well, if the
ground state reflection factor (14) is replaced by (15).
6. Breather reflection factors
The breathers have vertex type scattering matrices in contrast to the RSOS type ones
of the kinks. These scattering matrices enter into the equations determining the reflection
factors of the breathers, nevertheless there is no need for their explicit form as the breather
reflection factors on the various boundaries can be obtained from that of the soliton
kinks by using the (bulk) fusion and the bootstrap [19]; the procedure is summarized
schematically on Fig. 7.
Fig. 7. The bootstrap procedure for the breather reflection factors on the boundary ground state | 12 .
If two bulk kinks form a bound state at a rapidity difference iρ (0 < ρ < π ) the bound
state is identified with a supermultiplet (φ, ψ) of mass 2M cos(ρ/2). (In case of the kth
breather ρ = ρk = π − kπ λ .) The fusing coefficients of these processes are defined via [11]:

Kab (θ + iρ/2)Kbc (θ − iρ/2) = f φ abc φ(θ ) + f ψ abc ψ(θ )
with the nonvanishing coefficients being:
f φ 0 1 0 = f φ 1 1 1 = 2(π−2ρ)/4π f φ 1 0 1 = 2(π−2ρ)/4π f φ 1 1 1
2
! 2
2 2

2 2
ρ−π
= K(iρ)2(π−ρ)/2π cos
4
and
f ψ 1 1 0 = −f ψ 0 1 1 = 2(π−2ρ)/4π if ψ 1 0 1 = −2(π−2ρ)/4π if ψ 1 1 1
2
! 2

2 2

2 2
ρ+π
= K(iρ)2(π−ρ)/2π cos .
4
To describe the ground state reflection amplitudes of the bosonic (φ) and fermionic (ψ)
components we represent them as

1 1
φ(θ ) = φ
K 1 0 (θ + iρ/2)K0 1 (θ − iρ/2)
2 2f 1 0 1 2 2
2 2

1
+ K 1 1 (θ + iρ/2)K1 1 (θ − iρ/2) ,
2 2 2

1 1
ψ(θ ) = K 1 0 (θ + iρ/2)K0 1 (θ − iρ/2)
2 2f ψ 1 0 1 2 2
2 2

1
− K 1 1 (θ + iρ/2)K1 1 (θ − iρ/2) .
2 2 2
These expressions show that they also provide an ordinary doublet representation of the
boundary supercharge Q ± and that the fermionic parity Γ act on them in the standard
way. The actual reflection factors are obtained from the bootstrap equation on Fig. 7, where
the dashed lines represent either φ or ψ. The bosonic and fermionic reflection factors are
qualitatively different in the Γ symmetric and Γ nonsymmetric cases, since the bootstrap
equations contain both the R 01 1 and the R 11 1 ground state kink reflection amplitudes, and
2 2 2 2
these are significantly different in the two cases. Writing the breather reflection factors on
the ground state boundary as

φ 1 A+ B φ 1

(θ ) = (−θ )
ψ 2
B A − ψ 2
in the Γ symmetric (BSSG+ ) case one obtains

= 0, θ π θ π
B=B A+ = Z(θ ) cos − , A− = Z(θ ) cos + ,
2i 4 2i 4
√
Z(θ ) = P (θ + iρ/2)P (θ − iρ/2) 2 K(2θ )2−θ/ iπ ,
while in the Γ nonsymmetric (BSSG− ) case we get

= −Z(θ
) √ γ "
B = −B cos(ρ/2) sinh(θ ),
2 M
2
) cosh θ
A± = Z(θ
γ
− sin2
ρ
+ sinh2
θ
2 4M 4 2
2
θ γ ρ θ
∓ i sinh + sin2 + sinh2 ,
2 4M 4 2
with
) = K(2θ )2−θ/ iπ F (θ − iρ/2)F (θ + iρ/2),
Z(θ
F (θ ) = P (θ )K(θ + iξ )K(θ − iξ ).
The two cases are indeed qualitatively different: in the BSSG+ solution the conservation
of fermionic parity forbids the φ → ψ reflection on the ground state boundary, while in
the BSSG− solution this reflection is possible. The form of the BSSG+ solution is the
same as the one obtained in [24] by imposing fermion number conservation. The structural
form of the reflection factors in the case BSSG− are identical to the ones obtained in [23]
from the eight vertex free fermion model with boundary. In [4] it was proposed that γ ,
which appears explicitly in the reflection matrix, could be fixed in terms of η and ϑ by
the boundary bootstrap. Here we see that this is not the case, as the bootstrap gives no
constraint for γ , due to degeneracies appearing in the boundary excited states. We recall
however that γ is not a free parameter, as it is determined by the ground state boundary
energy (or equivalently, by the boundary supercharge), thus it can be expressed in terms of
the Lagrangian parameters in principle.
A more explicit description of the reflection factor of the first breather on the ground
state boundary in the BSSG− case was given in boundary sinh-Gordon model studied in
[6], but the precise connection between the parameters used in that paper and the present
one is yet to be determined.
In the bosonic theory the ground state reflection amplitude of the kth breather has poles
at

η π π k−1
−iθ = − + (k − 2l − 1) , l = 0, . . . , .
λ 2 2λ 2
In the supersymmetric theory, these poles signal the presence of the excited boundary states

1
, a, 1 l, k − l
2 2
as intermediate states in the breather reflection process. Since there are two intermediate
states (a = 0, 1), the determinant of the 2 × 2 reflection matrix should not vanish at the
position of these poles (as the residue of the reflection matrix at the pole is proportional to
the projector on the subspace of on-shell intermediate states). It is straightforward to verify
that this is indeed the case for both the BSSG+ and the BSSG− solutions.
Using the bootstrap procedure it is also possible to obtain the breather reflection factors
on excited boundaries. If the RSOS sequence characterizing the boundary state ends with
a label a (a = 0 or a = 1), then, e.g., the bosonic breather can be represented by

φ(θ )a, 1 , . . .n1 , . . .
2

1 1
= φ Ka 1 (θ + iρ/2)K 1 a (θ − iρ/2)a, , . . .n1 , . . .
f a1a 2 2 2
2
in the bootstrap procedure. To emphasize that even in the BSSG+ case there are φ → ψ
type reflections on excited boundaries we give here the reflection matrix of breathers
on the |a1/2|n states. In the basis of |φ(θ )0, 1/2|n, |φ(θ )1, 1/2|n, |ψ(θ )0, 1/2|n
|ψ(θ )1, 1/2|n it can be written as
 
C+ 0 0 −D/r
 C+ rD 0 
# ) 0
Z(θ , (25)
 0 D/r C− 0 
−rD 0 0 C−
$
ρ ν
where D = √1 cos 2 cos 2n sin θi and
2

νn θ π 1 ρ θ θ π
C± = cos2 cos ∓ + cos − cos cos ± .
2 2i 4 2 2 i 2i 4
It is easy to show that in spite of the nontrivial boson fermion reflection the operator Γ
commutes with this reflection matrix.
A nontrivial check on the consistency of the bootstrap solution can be obtained by
considering the pole structure of the full reflection amplitude containing the SUSY factor
(25). The bosonic reflection factor of Bk on the bosonic boundary excited state |n has a
pole at −iθ = π2 − λη + 2λπ
(k + 2n + 1) [21]. In the supersymmetric case, this means that
a boundary excited state of the form

1
a, n + k (26)
2
enters as an on-shell intermediate state in the scattering of Bk on |a, 1/2|n. However, due
to the doublet (boson/fermion) structure of the breather naively one would expect 4 states
to explain the residue of the 4 × 4 reflection factor. In the conjectured spectrum, on the
other hand, the only possible process goes via (26) and it allows for only two intermediate
states (a = 0, 1). Therefore one expects that the determinant of the matrix (25) should have
a double zero there. It can be verified by direct calculation that this double zero is indeed
there without imposing any restriction on the parameters.
In the reflection of the kth breather on the nth excited boundary, there is another family
of poles at −iθ = ηλ − 2λ π
(k − 2l + 1), l = 0, . . . , n − 1, [21] that in the supersymmetric
case should correspond to intermediate states of the form

1
b, , a, 1 l, k − l, n .
2 2
At these poles, the number of intermediate states is 4 (a, b = 0, 1) and so we expect that
the determinant of the SUSY factor does not vanish, which indeed turns out to be the case.
7. Conclusions
To start we summarize the results of this paper. We considered the boundary scattering
amplitudes in boundary supersymmetric sine-Gordon theory (BSSG). Imposing the con-
straint of supersymmetry on solutions of the boundary Yang–Baxter equation, we found
two consistent sets of amplitudes that describe the reflection of solitons off the boundary in
its ground state. Then we considered the two bootstrap systems built from these fundamen-
tal amplitudes and conjectured the closure of this bootstrap, i.e., the set of boundary states
and the reflection factors on them. We also derived a relation between the boundary super-
charge and the Hamiltonian and checked that this relation holds for the bootstrap solutions.
Although the reflection amplitudes are different, the spectrum of states is the same in the
two bootstrap solutions. This common spectrum is characterized partly by a sequence of
integers, just like in the case of the ordinary sine-Gordon model [21], but also by an RSOS
sequence of length k + 1 (if the length of the integer sequence is k) starting from 1/2. The
energy of the state depends only on the integer labels, the different RSOS sequences cor-
respond to degenerate states. It is interesting to note that the nonsupersymmetric boundary
spectrum allows for a tensor product type supersymmetrization, and no further constraints
are obtained in accord with the bulk case [11].
In the case of the BSSG+ theory, the reflection amplitudes depend on two parameters η
and ϑ, that are inherited from the bosonic reflection factors and were originally introduced
in [18]. In the bosonic case it is known how these parameters are related to the parameters
of the boundary Lagrangian in the perturbed CFT formalism [25]. Besides that, the SUSY
algebra introduces a further parameter γ , which is related to the energy of the boundary
ground state and so must be a function of the parameters of the BSSG Lagrangian. In the
BSSG− theory the difference is that γ appears also in the expression for the reflection fac-
tors themselves. In the bosonic case the expression for the boundary energy in terms of La-
grangian parameters is also known [25]. The existence of two different families of solutions
and the number of parameters are in accordance with the expectations that they describe
the scattering in the Lagrangian theories corresponding to the boundary interaction (6) [4].
It is a very interesting and important issue to connect the bootstrap parameters η, θ
and the vacuum energy parameter γ to the parameters of the Lagrangian description for
the supersymmetric case as well. In the case of the nonsupersymmetric boundary sine-
Gordon theory this was achieved by considering it as a combined bulk and boundary
perturbation of a c = 1 free massless boson with Neumann boundary condition. However,
even the interpretation of the bulk SSG theory as a perturbed CFT is nontrivial, and we are
investigating this problem. We are also working on getting more evidence to link the bulk
S matrix and the reflection factors to the Lagrangian theory. Work is in progress in these
directions and we hope to report on the results in the very near future.
Acknowledgements
The authors would like to thank G.M.T. Watts for very useful discussions and
R.I. Nepomechie for comments on the manuscript. G.T. thanks the Hungarian Ministry
of Education for a Magyary Postdoctoral Fellowship, while B.Z. acknowledges partial
support from a Bolyai János Research Fellowship. This research was supported in part by
the Hungarian Ministry of Education under FKFP 0043/2001 and the Hungarian National
Science Fund (OTKA) grants T037674/02 and T34299/01.
References
[1] S. Ferrara, L. Girardello, S. Sciuto, An infinite set of conservation laws of the supersymmetric sine-Gordon
theory, Phys. Lett. B 76 (1978) 303.
[2] C. Ahn, Complete S-matrices of supersymmetric sine-Gordon theory and perturbed superconformal minimal
model, Nucl. Phys. B 354 (1991) 57–84.
[3] T. Inami, S. Odake, Y.-Z. Zhang, Supersymmetric extension of the sine-Gordon theory with integrable
boundary interactions, Phys. Lett. B 359 (1995) 118–124, hep-th/9506157.
[4] R.I. Nepomechie, The boundary supersymmetric sine-Gordon model revisited, Phys. Lett. B 509 (2001)
183–188, hep-th/0103029.
[5] C. Ahn, W.M. Koo, Exact boundary S matrices of the supersymmetric sine-Gordon theory on a half line,
J. Phys. A 29 (1996) 5845–5854, hep-th/9509056.
[6] C. Ahn, R.I. Nepomechie, Exact solution of the supersymmetric sinh-Gordon model with boundary, Nucl.
Phys. B 586 (2000) 611–640, hep-th/0005170.
[7] R.I. Nepomechie, Supersymmetry in the boundary tricritical Ising field theory, preprint UMTG-234 hep-
th/0203123.
[8] A.B. Zamolodchikov, Fractional-spin integrals of motion in perturbed conformal field theory, in: H. Guo,
Z. Qiu, H. Tye (Eds.), Fields, Strings and Quantum Gravity, Gordon and Breach, 1989.
[9] R. Shankar, E. Witten, The S matrix of the supersymmetric nonlinear sigma model, Phys. Rev. D 17 (1978)
2134.
[10] K. Schoutens, Supersymmetry and factorizing scattering, Nucl. Phys. B 344 (1990) 665–695.
[11] T.J. Hollowood, E. Mavrikis, The N = 1 supersymmetric bootstrap and Lie algebras, Nucl. Phys. B 484
(1997) 631–652, hep-th/9606116.
[12] G. Takács, A new RSOS restriction of the Zhiber–Mikhailov–Shabat model and Φ(1, 5) perturbations of
nonunitary minimal models, Nucl. Phys. B 489 (1997) 532–556, hep-th/9604098.
[13] C. Ahn, R.I. Nepomechie, The scaling supersymmetric Yang–Lee model with boundary, Nucl. Phys. B 594
(2001) 660–684, hep-th/0009250.
[14] E. Witten, D.I. Olive, Supersymmetry algebras that include topological charges, Phys. Lett. B 78 (1978) 97.
[15] Z. Bajnok, L. Palla, G. Takács, F. Wágner, The k-folded sine-Gordon model in finite volume, Nucl. Phys.
B 587 (2000) 585–618, hep-th/0004181.
[16] N. Graham, R.L. Jaffe, Energy, central charge, and the BPS bound for (1 + 1)-dimensional supersymmetric
solitons, Nucl. Phys. B 544 (1999) 432–447, hep-th/9808140;
M.A. Shifman, A.I. Vainshtein, M.B. Voloshin, Anomaly and quantum corrections to solitons in two-
dimensional theories with minimal supersymmetry, Phys. Rev. D 59 (1999) 045016, hep-th/9810068;
A. Litvintsev, P. van Nieuwenhuizen, Once more on the BPS bound for the SUSY kink, preprint YITP-00-18
hep-th/0010051;
A. Rebhan, P. van Nieuwenhuizen, R. Wimmer, The anomaly in the central charge of the supersymmetric
kink from dimensional regularization and reduction, preprint TUW-02-16, YITP-02-35, hep-th/0207051.
[17] R. Chatterjee, A.B. Zamolodchikov, Local magnetization in critical Ising model with boundary magnetic
field, preprint RU-93-54, hep-th/9311165.
[18] S. Ghoshal, A.B. Zamolodchikov, Boundary S matrix and boundary state in two-dimensional integrable
quantum field theory, Int. J. Mod. Phys. A 9 (1994) 3841–3886;
S. Ghoshal, A.B. Zamolodchikov, Boundary S matrix and boundary state in two-dimensional integrable
quantum field theory, Int. J. Mod. Phys. A 9 (1994) 4353, hep-th/9306002, Erratum.
[19] S. Ghoshal, Bound state boundary S matrix of the sine-Gordon model, Int. J. Mod. Phys. A 9 (1994) 4801,
hep-th/9310188;
A. Fring, R. Köberle, Factorized scattering in the presence of reflecting boundaries, Nucl. Phys. B 421
(1994) 159, hep-th/9304141.
[20] P. Mattsson, P. Dorey, Boundary spectrum in the sine-Gordon model with Dirichlet boundary conditions, J.
Phys. A 33 (2000) 9065–9094, hep-th/0008071;
P. Mattsson, Integrable Quantum Field Theories in the Bulk and with a Boundary, PhD thesis, hep-
th/0111261.
[21] Z. Bajnok, L. Palla, G. Takács, G.Z. Tóth, The spectrum of boundary states in sine-Gordon model with
integrable boundary conditions, Nucl. Phys. B 622 (2002) 548–564, hep-th/0106070.
[22] L. Chim, Boundary S matrix for the tricritical Ising model, Int. J. Mod. Phys. A 11 (1996) 4491–4512,
hep-th/9510008.
[23] C. Ahn, W.M. Koo, Supersymmetric sine-Gordon model and the eight-vertex free fermion model with
boundary, Nucl. Phys. B 482 (1996) 675, hep-th/9606003.
[24] M. Moriconi, K. Schoutens, Reflection matrices for integrable N = 1 supersymmetric theories, Nucl. Phys.
B 487 (1997) 756–778, hep-th/9605219.
[25] Al.B. Zamolodchikov, unpublished;
Z. Bajnok, L. Palla, G. Takács, Finite size effects in boundary sine-Gordon theory, Nucl. Phys. B 622 (2002)
565–592, hep-th/0108157.
Numerical study for the c-dependence of fractal

dimension in two-dimensional quantum gravity
Noboru Kawamoto, Kenji Yotsuji 1
Department of Physics, Faculty of Science, Hokkaido University, Sapporo, 060-0810, Japan
Received 17 July 2002; accepted 10 September 2002
Abstract
We numerically investigate the fractal structure of two-dimensional quantum gravity coupled to
matter central charge c for −2 c 1. We reformulate Q-state Potts model into the model which
can be identified as a weighted percolation cluster model and can make continuous change of Q,
which relates c, on the dynamically triangulated lattice. The c-dependence of the critical coupling
is measured from the percolation probability and susceptibility. The c-dependence of the string
susceptibility of the quantum surface is evaluated and has very good agreement with the theoretical
predictions. The c-dependence of the fractal dimension based on the finite size scaling hypothesis is
measured and has excellent agreement with one of the theoretical predictions previously proposed
except for the region near c ≈ 1.
1. Introduction
It could probably be one of the most annoying questions: “what is the observable
of quantum gravity to check if the quantum theory of gravity is well defined?” In two-
dimensional quantum gravity the answer to this question is ready: “the fractal dimension
of the quantum surface is the well-defined observable which can be analytically and
numerically calculated.”
The importance of the fractal nature of two-dimensional quantum gravity was first
recognized by KPZ [1] where the critical exponent is recognized to represent the fractal
structure of the two-dimensional quantum surface. It has, however, been recognized later
E-mail addresses: kawamoto@particle.sci.hokudai.ac.jp (N. Kawamoto), yotsuji@masant.tokai.jaeri.go.jp

(K. Yotsuji).
1 Present address: Visible Information Center, Inc. 440, Muramatsu, Tokai-mura, Ibaraki, Japan.
PII: S 0 5 5 0 - 3 2 1 3 ( 0 2 ) 0 0 8 2 2 - 2
534 N. Kawamoto, K. Yotsuji / Nuclear Physics B 644 [FS] (2002) 533–567
that more direct observable to detect the quantum fractal nature of space time would be the
Housdorff dimension, or fractal dimension [2,3].
The serious numerical study to measure the fractal nature of two-dimensional quantum
surface was initiated by Agistein and Migdal [4]. They have proceeded the direct
measurements of the fractal dimension of two-dimensional quantum surface by proposing
a recursive sampling algorithm for c = 0 model but could not observe the fractal nature of
the quantum gravity in two dimension. The size of the triangulation was not large enough
to observe the fractal nature of the quantum surface of c = 0 model. The first numerical
confirmation of the fractal structure of two-dimensional quantum gravity was carried out
by Kawamoto, Kazakov, Saeki and Watabiki by using the recursive sampling algorithm for
c = −2 model [5]. In these numerical analyses the large size lattice triangulation (up to
5 million triangles) was necessary to confirm the fractal nature by the direct measurement
of the fractal dimension. This numerical confirmation of the two-dimensional quantum
space–time triggered wide varieties of numerical and analytic investigations of the fractal
nature of two-dimensional quantum gravity.
There are three analytic derivations of the c-dependence of Housdorff dimension or
fractal dimension of two-dimensional quantum gravity coupled to matter central charge c,
by Distler, Hlousek and Kawai [2], Kawai and Ninomiya [3] and later by Watabiki,
Kawamoto and Saeki [6]. It was, however, pointed out that the measured fractal dimension
of c = −2 model is very close to the third formulae given by Watabiki et al. [6]. In the
meantime Kawai, Kawamoto, Mogami and Watabiki tried to understand the fractal nature
of quantum gravity from the Matrix model point of view and succeeded to derive the
transfer matrix of the quantum surface of two-dimensional pure gravity (c = 0) [7]. The
formulation made it possible to derive the fractal dimension of pure gravity to be exactly
dF = 4 which is consistent with the value of the first and third formulae. This analytic
investigation of the random surface triggered further analytic and numerical investigations
of two-dimensional gravity [8–11].
Baby universe idea was proposed and proved to be useful to calculate string
susceptibility numerically [12,13]. Then finite size scaling hypothesis was proposed later
by being inspired by the analytic derivation of the two-point function of pure gravity [14].
The finite size scaling hypothesis made it possible to measure the fractal dimension very
accurately. Using this formulation we made systematic and the most reliable numerical
measurement of the fractal dimension for c = −2 model [15,16]. Then the very accurately
measured fractal dimension of c = −2 model perfectly agreed with the theoretical value of
the third formula, dF = 3.56 ± 0.04(numerical) 3.561 . . .(theoretical).
So far the numerical investigations of the fractal dimensions were carried out mainly
for c = −2 and c = 0 model and for several unitary series of conformal field theory [17]
and several values in the region c > 1 [18] and for c = −5 [19].
Here in this article we investigate the systematic investigation on the c-dependence
of the fractal dimension of two-dimensional quantum surface. For c = −2, 0 analytic
formulae by the help of matrix model was available while for the other continuous value
of matter central charge it was not obvious how to formulate the models to be convenient
for the numerical study of the fractal dimension. It has, however, been known that the
continuos central charge dependence can be accommodated by Q-state Potts model on the
flat lattice. Here in this paper we reformulate the Q-state Potts model into the model which
N. Kawamoto, K. Yotsuji / Nuclear Physics B 644 [FS] (2002) 533–567 535
is a generalization of percolation cluster model, weighted percolation cluster model [20,

21]. We formulate the model on the dynamically triangulated lattice and extend to take non-
integer value Q which is a function of c. By this model we can investigate c-dependence
of the fractal nature of the two-dimensional quantum gravity coupled to matter central
charge c. There have already been several calculations of critical indices of Ising model,
three-state Potts models, and large Q values of Q-state Potts model coupled to quantum
gravity [22,23].
This paper is organized as follows: in Section 2 we summarize the analytic derivations
of Hausdorff dimension and fractal dimension. In Section 3 we explain the equivalence
of Potts model and a weighted percolation cluster model which we call generalized Potts
model. We then provide the definition of the generalized Potts model on the dynamically
triangulated lattice. In Section 4 we explain the details of the Metropolis argolism of the
generalized Potts model. In Section 5 numerical results of the c-dependence of critical
coupling constant, string susceptibility, and fractal dimension are shown. Conclusions and
discussions are given in the last section.
2. Analytic derivation of fractal dimension and previous numerical results
The fractal nature of the two-dimensional quantum gravity was first recognized by KPZ
[1]. It was, however, not clear what kind of fractal it meant in the beginning. Serious
analytic study on the fractal dimension of quantum gravity has been given by Distler,
Hlousek and Kawai [2], and Kawai and Ninomiya [3] and later by Watabiki, Kawamoto
and Saeki [6]. Here we summarize the analytic derivation of the fractal dimension.
In the derivation of the fractal dimension of two-dimensional quantum gravity coupled
with matter central charge c, we use the formulation of Liouville theory in particular the
formulation of conformal gauge given by DDK [24]. We first summarize the main results
of the formulation.
Formally the continuum partition function for matter coupled to two-dimensional
gravity is given by

√
Z(A) = Dg δ d 2 x g − A ZM [g], (2.1)
where ZM [g] is a matter part of the partition function with gravitational background and
A is the total area.
Regularized counterpart of the above partition function by dynamical triangulation is

Zreg (A) = ZM [G] δNa 2 ,A ∼ ZM [G0 ], (2.2)
G
where N is the number of equilateral triangles and a 2 is the area of the triangle. G denotes
a triangulation and G0 is the typical triangulation which we select from the huge set of
triangulations. The last approximate equality in Eq. (2.2) is valid up to the normalization
factor and if the selection of the typical surface is carried out by a correct procedure. Since
the path integration of the metric is carried out after the selection of the typical surface, G0
carries the information of the quantum fluctuation of spacetime effectively.
Following David, Distler and Kawai (DDK) [24], we obtain the gauge fixed version of
the two-dimensional gravity coupled to matter central charge c with a conformal gauge
gµν = ĝµν eφ :

c − 25
Z(A) = Dĝ φ ∆FP [ĝ]ZM [ĝ]δ 2
d x ĝ e α1 φ
− A exp SL [φ, ĝ] , (2.3)
48π
where ∆FP is the Fadeev–Popov contribution. SL [φ, ĝ] is the Liouville term given by
1
SL [φ, ĝ] = d x ĝ
2 µν
ĝ ∂µ φ∂ν φ + φ R , (2.4)
2
where we set the renormalized cosmological constant equal to zero for simplicity. α1
appeared in Eq. (2.3) is obtained from the following general formula:
√
25 − c − (25 − c)(25 − c − 24n)
αn = . (2.5)
12
DDK have shown that the primary conformal field of weight n − 1 can be made Weyl
invariant at the quantum level with a quantum correction:

d 2 x ĝ eαn φ Φn , (2.6)
where Φn transforms as Φn |ĝeσ = e(n−1)σ Φn |ĝ . The term eα1 φ in the delta function of Eq.
(2.3) is needed to keep the world sheet volume to be A at the quantum level.
Let us define an expectation value of an observable O(g) by

−1
O(g) A = Z(A) Dĝ φ ∆FP [ĝ]ZM [ĝ]δ 2
d x ĝ e α1 φ
−A

c − 25
× O(ĝ, φ) exp SL [φ, ĝ] . (2.7)
48π
2.1. Hausdorff dimension and fractal dimension
Here we define two types of critical exponents which specialize the fractal nature of the
two-dimensional random surface.
Let us first define an intrinsic area A(r) of the random surface as a function of the mean
square average size of the world sheet r 2 as viewed in the embedding space. We define
Hausdorff dimension dH as

dH
A(r) = r2 . (2.8)
It should be noted that this definition of Hausdorff dimension refers to the embedded space.
Let us next define N(r) as the number of lattice points or number of triangles inside r
steps from a marked site. A step on the original lattice is one link step of a triangle while
a step on the dual lattice is one dual link (edge) step on the dual lattice. We define fractal
dimension dF as
N(r) = r dF . (2.9)
This definition of the fractal dimension characterizes the connectivity of the random
surface, how the random surface is composed by the connection of triangles, and thus
could be called connectivity dimension.
Distler, Hlousek and Kawai [2] evaluated the mean square size of the quantum surface
embedded in a D-dimensional space by calculating the two-point Green’s function of
vertex operator

1 2 ∂
x A = 2 2 ln d 2 x1 ĝ(x1 ) d 2 x2 ĝ(x2 ) eik(X(x1)−X(x2 ))
D ∂k A k=0

1 2
= 2 d x1 ĝ(x1 ) d x2 ĝ(x2 ) X(x1 ) − X(x2 )
2 2
A Z(A) A
|γs |
A
= (A → ∞), (2.10)
A0
where γs is the string susceptibility given by
√
D − 1 − (25 − D)(1 − D)
γs = . (2.11)
12
We can then obtain the first formula of Housdorff dimension as a function of D.
(1) 2
dH = , (2.12)
|γs |
where D could later be identified as matter central charge c in two dimension.
Let us next consider a derivation of fractal dimension using fermion as a test particle in
the gravitational background following by Kawai and Ninomiya [3]. The Lagrangian for
matter fermion coupled to gravity can be given by
1 √ √
L=− g R + Λ g + eψ̄iDψ / − meψ̄ψ, (2.13)
16π
where Λ and e are, respectively, cosmological constant and the determinant of the vielbein
in D dimensions. In D = 2 + . dimensions gravitational quantum corrections can be
evaluated by the .-expansion formulation [25]. Then the fermion mass term is expected
to acquire anomalous dimension via wave function renormalization of the matter fermion.
Here we try to identify the anomalous dimension of the fermion mass term by the use
of DDK formulation for Liouville theory. Under the scaling of the cosmological constant
Λ → β −D Λ, (2.14)
the change in the Lagrangian can be absorbed by the field redefinition
gµν → β 2 gµν . (2.15)
Since the vielbein transforms like square root of metric, the fermion kinetic term transforms
/ → β D−1 eψ̄iDψ.
as eψ̄iDψ / Then the scaling parameter can be absorbed by the field
redefinition ψ → β (1−D)/2ψ. The fermion mass term changes as meψ̄ψ → βmeψ̄ψ, in
particular ψ̄ψ → β −1 ψ̄ψ in D = 2.
In two dimensions the fermion mass term m d 2 x eψ̄ψ is expected to have the form
√
of Eq. (2.6) after the introduction of the gravitational quantum correction. Since g and
e acquire the same scale change, ψ̄ψ can be identified as primary conformal field of Φ1/2
because of the same scale change: Φ1/2 |ĝβ 2 = β −1 Φ1/2 |ĝ and ψ̄ψ → β −1 ψ̄ψ. Then the
Weyl invariant fermion mass term with the quantum correction is given by

m d 2 x ê eα1/2 φ ψ̄ψ. (2.16)
We now consider the following quantum average of the fermion mass term:

m d 2 x ê eα1/2 φ ψ̄ψ , (2.17)
A
where the quantum average · · ·A is defined in Eq. (2.7). Under the constant shift of the
conformal field φ → φ − 2 ln β/α1 , the quantum average should be unchanged and yet the
following relation holds:

m d 2 x êeα1/2 φ ψ̄ψ = mβ −2α1/2 /α1 d 2 x ê eα1/2 φ ψ̄ψ , (2.18)
A β2A
where the following change of delta function is taken into account:

δ d 2 x ĝ eα1 φ − A → β 2 δ d 2 x ĝ eα1 φ − β 2 A .
This relation suggests that the theory with two different sets of parameters (A, m) and
(β 2 A, mβ −2(α1/2 )/α1 ) are equivalent. The two-dimensional volume measured by the length
(2)
scale of the fermion field leads to another definition of fractal dimension dF [3],
d (2)
A ∼ L2 = L2α1/2 /α1 F , (2.19)
where
√ √
(2) α1 25 − c + 13 − c
dF (c) = =2× √ √ , (2.20)
α1/2 25 − c + 1 − c
with αn given by the formula (2.5).
2.2. Fractal dimension from diffusion equation of random walk
Here we provide yet another derivation of the fractal dimension by using the solution of
diffusion equation on the random surface.
We define the Laplacian on the dynamically triangulated lattice. We first define
adjacency matrix Kij on the dynamically triangulated lattice. For a chosen typical surface
G0 we number the sites of the triangulated lattice. Then the (i, j ) component of the
adjacency matrix Kij is defined as: Kij = 1 if ith site and j th site are connected by a link,
Kij = 0 if they are not connected by a link. It is interesting to note that (n, n0 ) component
of (K T )i,j counts the number of possible random walks reaching from a marking site n0
to a site n after T steps. The Laplacian defined on the dynamically triangulated lattice is
given by
1
∆L = 1 − S, Sij = Kij , (2.21)
qj
where qj is called coordination number and denotes a number of links connected to the
site j . Sij is thus a probability of one step random walk from the site j to the neighboring
site i. The diffusion equation on a triangulated surface G0 with N triangles is given by
∂T ΨN(G0 ) (T ; n, n0 ) = ∆L (G0 )ΨN(G0 ) (T ; n, n0 ), (2.22)

(G )
where ∂T is a difference operator in T and ΨN 0 (T ; n, n0 ) is a wave function of the
diffusion equation and denotes the probability of finding the random walker at the site n
after T steps from the starting site n0 . A solution of the diffusion equation can be obtained
(G )
as ΨN 0 (T ; n, n0 ) = eT ∆L (G0 ) (δn,n0 ), where (δn,n0 ) is N -component vector with unit n0
entry.
We now consider the continuum limit of this diffusion equation. First of all we recover
the lattice constant a. In taking continuum limit, the total physical area A = ai2 Ni is fixed
and i → ∞; ai → 0, Ni → ∞ is taken, where Ni is the number of triangles and ai2 is the
area of a triangle. In each step of the limiting process we select a typical surface Gi for
the given number of triangles Ni , on which the lattice Laplacian ∆L (Gi ) of Eq. (2.21) is
defined. Now the lattice version of the diffusion equation (2.22) can be rewritten by
1 (Gi ) (Gi ) 1
Ψ A T + a 2
i ; x, x 0 − Ψ A (T ; x, x 0 ) = 2 ∆L (Gi )ΨA(Gi ) (T ; x, x0),
ai2 ai
(2.23)
where the location of the site x is measured with respect to the lattice constant ai . Thus
we identify the dimension of T as that of area: dim[T ] = dim[A]. In the continuum limit
the solution of the diffusion equation (2.23) is expected to approach the continuum wave
(G ) (G )
function: ΨA i (T ; x, x0) → ΨA ∞ (T ; x, x0). Numerically we approximate the limiting
surface as the typical surface (G0 ) of the maximum size triangulation: G∞ G0 . As
we have already noted in Eq. (2.2), the metric integration is effectively carried out for
Eq. (2.23) since we have chosen a typical surface. This means that the quantum effect
is included for the wave function of Eq. (2.23). On the other hand the solution of the
continuum counterpart of the diffusion equation: ∂τ Ψ (τ ; x, x0) = ∆(g)Ψ (τ ; x, x0) is still
background metric dependent in general. Furthermore the dimensions of T and τ may not
necessarily be equal.
Let us now define the comeback probability of random walk on the triangulated lattice
and relate it with the continuum expression of Liouville theory as follows:
(G )
G(T ) ≡ ΨA 0 (T ; x = x0 )

√ √
d 2 x g Ψ (τ ; x = x0 ) d 2x g
A A

1 √
= d 2 x g eτ ∆ Ψ (0; x = x0 )
A A
1 1
∼ ∼ , (2.24)
A T
where · · ·A is the quantum average defined in Eq. (2.7) and the last similarity relations
are dimensional relations. We should remind of the fact that the metric integration is
effectively carried out since we have chosen the typical surface G0 for the wave function of
the comeback probability. The initial wave function can be formally written as Ψ (0; x =
√
x0 ) = limx→x0 δ. (x − x0 )1/ g, where the regularized delta function is needed such as:
δ. (x − x0 ) = (1/π) × ./((x − x0 )2 + . 2 ).
We next consider how to accommodate the Weyl invariance into the diffusion equation
of random walk at the quantum level by using the formulation of Liouville theory. Let us
consider the following quantity by Liouville theory:

√
d 2x g Ψ (τ ; x = x0 )
A

√ √
= d 2 x g Ψ (0; x = x0 ) +τ d x g ∆Ψ (0; x = x0 )
2
+ ···, (2.25)
A A
where the solution of the diffusion equation is expanded by τ .

Taking a conformal gauge gµν (x) = ĝµν eφ(x) and introducing DDK arguments, we can
rewrite the first and second terms of the Eq. (2.25) as
1
√
d x g Ψ (0; x = x0 )
2
= d x ĝ δ. (x − x0 ) 2
= 1,
A ĝ x=x0 A

2 √
−−→
α−1 φ 1
d x g ∆Ψ (0; x = x0 ) = 2
d x ĝ e ∆x δ. (x − x0 ) ,
A ĝ x=x0 A
(2.26)
where the term eα−1 φ is introduced to keep the Weyl invariance of the second line of
Eq. (2.26) in accordance with the arguments of Eq. (2.6). Here it should be noted that
∆ Ψ (0; x = x0 ) in the second term of Eq. (2.26) can be identified as a primary conformal
field of Φ−1 .
Similar to the treatment for the Weyl invariance of the fermion mass term of Eq. (2.18),
the quantum average of the comeback probability should be unchanged under the constant
shift of the conformal field φ → φ − 2 ln β/α1 . In particular the second term of Eq. (2.25)
should be unchanged and yet the following parameter change is generated:

α−1 φ
τ 2
d x ĝ e ∆Ψ (0; x = x0 )
A

= τβ −2α−1 /α1 2
d x ĝ e α−1 φ (0; x = x0 )
∆Ψ . (2.27)
β2A
This relation suggests that the theory with two different sets of parameters (τ, A) and
(β −2α−1 /α1 τ, β 2 A) are equivalent. Then we obtain the following dimensional relation:
dim τ = dim A−α−1 /α1 . (2.28)

We now point out that the expectation value of the mean squared geodesic distance is
evaluated by the standard continuum treatment

√ 2
d 2 x g r(x, x0) Ψ (τ ; x, x0)

2 √
2 1 1
= d x g r(x, x0 ) ˆ
δ(x − x0 ) + τ ∆x δ(x − x0 ) + · · ·
ĝ ĝ
2
= −4τ + O τ , (2.29)
which is related to the quantum version of the mean squared geodesic distance in the small
τ region
2 2
r ≡ r(x, x0 ) ΨA(G0 ) (T ; x, x0)
x

√ √ 2 √
2
d x g d x0 g r(x, x0 ) Ψ (τ ; x, x0)
2 2
d x g
A A
−α−1 /α1 −α−1 /α1
∼τ ∼A ∼T . (2.30)
The last similarity relations are dimensional relations. Here we give the third definition of
fractal dimension

d (3)
A= r2 F , (2.31)
where
√ √
(3) α1 25 − c + 49 − c
dF (c) = −2 =2× √ √ , (2.32)
α−1 25 − c + 1 − c
with αn given by Eq. (2.5).
2.3. Previous numerical results
Serious numerical investigations of the fractal nature of two-dimensional quantum

gravity (c = 0 model) was initiated by Agistein and Migdal for c = 0 model [4]. They
proposed recursive sampling algorithm which partially use an analytic formula for tree
diagrams and rainbow diagrams to generate planar Feynman diagrams corresponding to
different types of two-dimensional sphere. They could not, however, observe the fractal
structure numerically for this model. The fractal nature of two-dimensional quantum
gravity was numerically first confirmed by Kawamoto, Kazakov, Saeki and Watabiki
by using the recursive sampling algorithm for c = −2 model [5]. It was recognized in
this numerical analyses that large number of triangles is necessary to confirm the fractal
nature by the parameterization of the fractal dimension given by (2.9). In fact they needed
5 million triangles to confirm the fractal nature. It was then recognized that the number of
triangles were not large enough to measure the fractal dimension of c = 0 model.
The numerical fractal dimension of c = −2 model was given in this numerical analyses
as dF 3.55. After deriving the third formula of the fractal dimension dF(3)(c) of
Eq. (2.32), we have recognized that the numerical value of the fractal dimension is close to
√
the theoretical value of the third formula dF(3)(−2) = (3 + 17)/2 = 3.561 . . ..
Except for the analytic derivation of the fractal dimension, one obtains the following
analytic relations of comeback probability in (2.24) and mean squared geodesic distance
in (2.30) from the diffusion equation:
(1) G(T )T ∼ const.,

(3)
(2) r 2 = T 2/dF (−2) T −0.56 ,
which can be numerically checked. These relations were numerically confirmed for c = −2
model in [6].
An alternative analytic investigation of the fractal structure of the pure gravity (c = 0)
were carried out by deriving transfer matrix of random surface by Kawai, Kawamoto,
Mogami and Watabiki. They found out the beautiful scaling function ρ(L; D) which
counts the number of boundaries whose boundary lengths lie between L and L + dL
located at geodesic distance D measured from a marked point. It is evaluated by taking the
continuum limit from the transfer matrix and disk amplitude of dynamical triangulation.
The functional form of ρ(L; D) for c = 0 model is given by

3 1 14
ρ(L; D)D 2 = √ x −5/2 + x −3/2 + x 1/2 e−x , (2.33)
7 π 2 3
where x = L/D 2 is a scaling parameter. This quantity ρ(L; D)D 2 for c = 0 model was
measured numerically and had excellent agreement with the theoretical scaling function
(2.33) [11]. One of the important result of this analysis is that the fractal dimension of the
c = 0 model turns out to be dF = 4 which is consistent with the first and third formulae.
Based on these theoretical and numerical evidences the third formula of the fractal
dimension would be the correct formula for the c-dependence of the fractal dimension.
In order to clear up the situation we started serious systematic and very accurate numerical
study of c = −2 model by using finite size scaling hypothesis [15,16]. It was concluded that
the measured fractal dimension from this analysis is dF = 3.56 ± 0.04 which is perfectly
consistent with the theoretical value of the third formula.
3. Generalized Potts model ≡ weighted percolation cluster model
Potts model [20] was defined as a generalization of the Ising model [26]. Fortuin and
Kasteleyn [21] showed that the Q-state Potts model is equivalent to a weighted percolation
cluster model which we explain in this section. Their construction allows the Potts model
to be generalized to non-integral values of Q. We may call this model as generalized Potts
model or equivalently weighted percolation cluster model. These models were originally
formulated on the square lattice while we extend to formulate the model on a dynamically
triangulated lattice.
We define a planar ϕ 3 -graph G dual to a triangulated lattice. Let a graph GN have
N vertices which are dual to triangles in the triangulated lattice. With each vertex i, we
associate a spin that can take Q different values σi = 1, 2, . . . , Q. Two adjacent spins σi
and σj interact with interaction energy −J δ(σi , σj ), where

1, σi = σj ,
δ(σi , σj ) = (3.1)
0, σi = σj .
Thus the Hamiltonian is

H = −K δ(σi , σj ), (3.2)
i,j
where K = J /kB T can be reinterpreted as a coupling constant and the summation runs
over all the pairs of nearest-neighbor vertices i, j . Then the partition function of this
model is given by

ZN (K; GN ) = exp K δ(σi , σj ) , (3.3)
{σ } on GN i,j
where the σ -summation runs over all possible values {σi = 1, 2, . . . , Q} for the spin
variables σ1 , σ2 , . . . , σN on GN . Thus there are QN terms in the summation.
It has been shown [21,27] that the partition function (3.3) can be expressed as a
dichromatic polynomial [28]. In order to see this, let us expand the partition function as
a product of terms associated with nearest-neighbor vertices. This can be worked out by
using the relation [δ(σi , σj )]m = δ(σi , σj ) (m = 1, 2, . . .) as follows

ZN (K; GN ) = 1 + vδ(σi , σj ) , (3.4)
{σ } i,j
where we set v = exp(K) − 1.

Let E be the number of the pairs of the nearest-neighbor vertices which we simply call
edges on the graph GN . It is equal to the number of original links in the triangulated lattice,
i.e., 3N/2. Then the summand in Eq. (3.4) is a product of E factors. Each factor is the sum
of two terms (1 and v δ(σi , σj )), so the product can be expanded as the sum of 2E terms.
Each of these 2E terms can be associated with a bond-cluster-graph (from now on call this
a cluster configuration) on GN . Note that the term is a product of E factors, one for each
edge. The factor for edge i, j is either 1 or vδ(σi , σj ): if it is the former, leave the edge
empty, if the latter, place a bond on the edge with weight v δ(σi , σj ). Do this for all edges
i, j . We then have a one-to-one correspondence between cluster configurations on GN
and terms in the expansion of the product in Eq. (3.4).
Consider a typical cluster configuration,
containing C connected components (1
C N ), namely, clusters and b = C i=1 bi bonds (0 b 3N/2). See Fig. 1. ith cluster
includes bi bands and an isolated vertex may be regarded as a cluster with bi = 0. Then
the corresponding term in the expansion contains a factor v b , and the effect of the delta
functions is that all vertices within a component must have the same spin σ . Summing
over independent spins, it follows that terms gives a contribution QC v b to the partition
function ZN . Summing over all such terms, i.e., over all cluster configurations, we have
Q

ZN (K; GN ) = QC v b . (3.5)
{cluster}
on GN
Fig. 1. A fragment of a cluster configuration on a planar ϕ 3 -graph. An isolated vertex is regarded as a cluster.
The summation is over all cluster configurations that can be drawn on GN . The expression
(3.5) is a dichromatic polynomial. Note that Q in Eq. (3.5) need not be an integer. We can
allow it to be any positive real number, and this can be a useful generalization of Q-state
Potts model to non-integer real number Q. We may call this model as generalized Q-state
Potts model or weighted percolation cluster model.
Let us consider Eq. (3.5) for a few particular values of Q. Firstly, we consider a
Q=1
model ZN (K) formulated on a given triangulated lattice. For the Q → 1 limit, if we
set v = p/(1 − p) the partition function is
E
Q=1 1
ZN (K) = vb = pb (1 − p)E−b . (3.6)
1−p
{cluster} {cluster}
Q=1
Therefore, ZN (K) becomes a sum over all possible bond percolation configurations
with the correct weight where p is the probability of a bond being present [21]. This result
holdsin any dimension and any lattice on which one defines the Potts models. Since the
sum {cluster} in Eq. (3.6) is the total probability and thus equal to 1, this model coupled
to two-dimensional quantum gravity corresponds to the pure gravity model (c = 0).
Next, let us examine for the Q → 0 limit. At the critical point of the Q-state Potts
model on the two-dimensional square lattice, it is known that v ∼ Q1/2 [20]. In general,
on a graph GN we can assume v ∼ Qα in the Q → 0 limit (0 < α < 1). Then the partition
Q→0
function ZN (K) becomes
Q→0

ZN (K) ∼ QαN Qαl+(1−α)C , (3.7)
{cluster}
where l is the number of independent loops in a cluster configuration. We have used the
Euler relation (See Fig. 1):
b = N + l − C. (3.8)
For 0 < α < 1 the leading terms in Eq. (3.7) in the Q → 0 limit can be obtained by
taking C = 1 (one cluster) and l = 0 (no loops). These dominant configurations are just the
spanning trees of the graph GN [21,29]. Then each spanning tree configuration contributes
Q→0
to ZN (K) with an equal weight. Therefore, this model coupled to two-dimensional
quantum gravity is equivalent to the c = −2 scalar fermion model coupled to quantum
gravity [5].
Temperley and Lieb [30] used the result of Fortuin and Kasteleyn to prove the
equivalence of the Potts model to the six-vertex or square-ice model, with staggered
polarizations. Baxter, Kelland and Wu (BKW) [27] have later found a very elegant
derivation of the result of Temperley and Lieb. They use a construction known as the BKW
construction which makes many exact results obvious, including the critical temperature,
self-duality and energy at criticality of the Potts model.
Thus the Q-state Potts models are analytically solved and the relations with the
conformal field theories are well known [31,32]. Q is related to central charge c in the
following particular form:

π 6
Q = 4 cos 2
, c=1− . (3.9)
m+1 m(m + 1)
The minimal unitary conformal field theories with central charge c between 0 and 1
correspond to integer m; m = 2, 3, 4, . . . . The generalization of Q to any positive real
number corresponds to a continuous change of the central charge c, a generalization from
minimal to non-minimal series of conformal field theories.
Within the framework of dynamical triangulations the generalized Potts model, or
equivalently the weighted percolation cluster models, coupled to two-dimensional quantum
gravity is described by the following partition function:
Q
1
ZN (K) = QC v b , (3.10)
SGN
GN ∈{ϕ 3 (TN )} {cluster}
where {ϕ 3 (TN )} denotes the set of ϕ 3 (TN )-graphs dual to triangulations TN of fixed
topology (which we always assume to be sphere) and SGN is a symmetry factor.
4. The Metropolis algorithm of the generalized Potts model
We intend to evaluate numerically the fractal dimensions of two-dimensional quantum

gravity coupled to matter central charge c by the generalized Potts model, equivalently
the weighted percolation cluster model formulated in the previous section (3.10). In the
process of Metropolis updating we need double step updating: firstly we update a cluster
configuration on a given graph GN , secondly the graph GN itself should be updated.
Here we first formulate the Metropolis algorithm of cluster configuration on a given

graph GN . Eq. (3.5) leads us to choose the probability distribution of the generalized Potts
model for a cluster configuration {C}k by

(k) (k)
QC v b
P {C}k = Q
, (4.1)
ZN (K; GN )
where C (k) and b(k) are the number of cluster and the total number of edges b(k) =
C (k) (k)
i=1 bi for the given cluster configuration {C}k , respectively.
We need to define a transition function t[{C}k , {C}l ] for a transition {C}k → {C}l ,
which satisfies ergodicity and the following detailed balance condition:

P {C}k t {C}k , {C}l = P {C}l t {C}l , {C}k . (4.2)
Here we choose to use Glauber function [33] as the transition function
δWlk
t {C}k , {C}l = , (4.3)
1 + δWlk
where
t[{C}k , {C}l ]
= δWlk = QδClk v δblk (4.4)
t[{C}l , {C}k ]
with δClk = C (l) − C (k) and δblk = b (l) − b(k) .
For a given cluster configuration {C}k , our updating proceeds as follows:
(1) We randomly pick up an edge on the graph GN and change the edge by the following
procedure and then the cluster configuration changes into {C}l .
(2) We have to find out a change of the probability distribution δWlk = QδClk v δblk when
we change the edge, where δblk is the change in the total number of bonds. If the edge
originally has a bond, remove the bond thus δblk = −1. If the edge originally does not
have a bond, add a bond to the edge thus δblk = +1. δClk is the change in the number
of clusters depending on the corresponding change of the edge.
(3) Next, we generate a pseudorandom number r uniformly distributed from 0 to 1 and
change the edge following to the procedure (2) if and only if
δWlk
r . (4.5)
1 + δWlk
This procedure ensures the transition {C}k → {C}l with the correct probability. We
use the Glauber function for the transition function because of a faster convergence to
the equilibrium distribution in our model.
(4) Return to (1) unless the system is sufficiently equilibrated.
By this updating, ergodicity and detailed balance condition can be ensured.

Let us point out that it is non-trivial to evaluate δClk in the step (2). For a given cluster
configuration {C}k , we pick up an edge on a graph GN . Suppose the edge does not have a
bond we add a bond on the edge with the probability of the step (3) and thus δblk = +1 if
Fig. 2. Cluster configurations: (a), (b), (c), (d). Doded lines are edges without bond while solid lines are edges
with bond. A and B are vertices in (a) and (b) while C and D are vertices in (c) and (d). Li denotes ith surrounding
loop. The arrows are pointers and compose the segments of the surrounding loops.
the bond is added. For example we pick up the edge A–B which has vertices A and B in
Fig. 2(a) or the edge C–D which has vertices C and D in Fig. 2(c), where those edges A–B
and C–D does not have a bond. After the bond A–B and the bond C–D are added, (a) and
(c) of Fig. 2 turn into (b) and (d) of Fig. 2, respectively. In order to find δClk we need to
know if the both vertices of the edge belong to the same cluster or not before the bond is
added. For example the vertices A and B in Fig. 2(a) belong to the different clusters while
the vertices C and D in Fig. 2(c) belong to the same cluster. It is not time consuming to
classify this difference numerically since we just need to know the data set of the collection
of numbered vertices belonging to the same cluster. If both vertices originally belong to
the different clusters then δClk = −1, while δClk = 0 if they originally belong to the same
cluster. It is thus numerically not difficult to identify δClk in the case of δblk = +1.
Let us suppose that we pick up an edge which already has a bond. Then we remove
the bond with the probability of the step (3) and thus δblk = −1 if it is removed. We may
pick up cluster configurations (b) and (d) of Fig. 2 as particular examples which already
has a bond at the edge A–B and C–D, respectively, and turn into (a) and (c) of Fig. 2,
respectively, after the bond A–B and C–D are removed. In order to find δClk we need to
know the information if the both vertices of the edge belong to the same cluster or not
after the bond is removed. The vertices A and B belong to the different cluster in (a) and
thus δClk = +1, while the vertices C and D belong to the same cluster in (c) and thus
δClk = 0. The crucial difference from the case δblk = +1 to the case δblk = −1 is that the
collection of the data set to classify the different cluster is not enough to differentiate if
two vertices are in the same cluster or not after the bond is removed. The straightforward
way to determine the connectedness of two vertices is to start from the first vertex and
enumerate all vertices connected to it until either the second vertex is reached, or an entire
connected region will be enumerated. This method is adequate if the clusters are small
enough (tens of vertices), but if large clusters are involved (in the vicinity of the critical
point) the CPU time requirements can grow unreasonable. Since the connectedness, a non-
local property, must be determined with every iteration to know δClk , it is important to find
faster algorithm to evaluate δClk . In order to quickly determine the connectedness of large
clusters, we have implemented an algorithm [34] using an auxiliary data structure based
on the BKW construction [27].
Let us reconsider a ϕ 3 -graph GN , with N vertices, together with its dual triangular lattice
TN . If a bond is present on GN , then its dual bond on TN is absent, and vice versa. The
boundaries between clusters on GN and their dual clusters on TN will form a collection of
closed loops. Now we call this closed loops the surrounding loops. We then have the Euler
relation,
b + 2C = N + L, (4.6)
which relates the number of clusters and bonds to the number of surrounding loops L.
By saving information of the surrounding loops as a data set, we transform the problem
of determining connectedness of two vertices to the problem of determining whether two
surrounding loop segments are part of the same surrounding loop or not. The surrounding
loops are represented as a chain of pointers in the computer. A pointer is a memory location
containing the address of the next pointer in the chain. There are three surrounding pointers
represented by arrows for each vertex, as is shown in Fig. 3. We have shown surrounding
loops composed of pointers for the panels (a), (b), (c), and (d) of Fig. 2. For example
in Fig. 2(b) pointer P1 points to pointer P2 , P2 points to P3 , etc. In this manner the
loops are represented by chains of pointers. Because of the (differential) Euler relation,
δblk + 2δClk = δLlk , we can determine δClk if we can determine δLlk , the change in the
number of surrounding loops.
Now, let us consider cases of removing a bond in the process of a Metropolis updating
in Fig. 2. The change; Fig. 2(b) → Fig. 2(a), illustrates the case in which removing the
bond A–B will divide the surrounding loop L1 into two surrounding loops L1 and L2
in Fig. 2(a), while the change; Fig. 2(d) → Fig. 2(c), illustrates the case in which two
surrounding loops L4 and L6 in Fig. 2(d) will be joined into the loop L4 in Fig. 2(c) if
the bond C–D is removed. In either case, two cuts must be made in chains of pointers
and the four ends rejoined if the bond is removed. In the case of A–B bond in Fig. 2(b),
the P1 → P2 and the P4 → P5 connections must be replaced by P1 → P5 and P4 → P2
Fig. 3. Three pointers surrounding each vertex.
connections, respectively, in Fig. 2(a). Similarly in the case of the bond C–D in Fig. 2(d),
the P6 → P7 and the P8 → P9 connections must be replaced by P6 → P9 and P8 → P7
connections, respectively, in Fig. 2(c). By collecting the information of chains we can
immediately conclude that δClk = +1 (for δblk = −1) since P1 and P4 are in the same
chain for Fig. 2(b) while P6 and P8 are in the different chains for Fig. 2(d) and thus
δClk = 0 (for δblk = −1). It is important to note that the information of the connectedness
of the cluster configuration of (a) and (c) can be obtained by the data set of the chains of
pointers of (b) and (d), respectively. It is numerically much easier to find if two pointers
are in the same loop chain or not.
In summary, in order to find δWlk for the Metropolis algorithm of the generalized Potts
model in Monte Carlo simulation, we first pick up an edge and find out δblk depending
on whether the edge has a bond or not. We first cut and rejoin our chains of pointers
depending on addition or removal of the bond on the edge. Using the (differential) Euler
relation, δblk + 2δClk = δLlk , we evaluate δClk by judging whether two pointers attached
to the different vertices of the given edge belong to the same chain or not.
When we apply Monte Carlo simulations to quantum gravity coupled to the generalized
Potts model or equivalently the weighted percolation clusters, we have to update the
cluster configuration for a given triangulation and at the same time we have to update the
triangulation for a given cluster configuration. The updating of cluster configurations on a
given triangulation can be carried out just as described in the above. In order to update
triangulations corresponding to the given ϕ 3 -graphs GN , we use the standard flip-flop
algorithm. We first choose two neighboring triangles randomly and flip the common link to
generate a new triangulation. This flip move changes the triangulation locally as in Fig. 4.
This move is enough to make the process ergodic for the chosen class of triangulations;
fixed number of triangles and fixed topology [35].
In our simulations we avoid to generate configurations corresponding to tadpole and
self-energy graphs. In other words we consider the class of triangulations satisfying the
following conditions:
(1) no link has coinciding end;

(2) two sites may be connected by no more than one link (i.e., parallel links are forbidden).
Fig. 4. The flip move. This move is ergodic in the class of triangulations with a fixed number of triangles and
fixed topology.
Fig. 5. Bond assignment after a flip move in a fragment of a cluster configuration. The bond is assigned to the
new edge with 50 % probability.
In the updating of triangulations, if a triangulation in a forbidden class is generated we

ignore the attempt to make the flip and instead choose a new link randomly and attempt a
new flip.
It is important to recognize that the flip procedure to generate random surface changes
the cluster configuration. By the flip procedure the chosen edge which is the dual of
the common link of the chosen neighboring triangles changes into a new edge which
may or may not have a bond. In the present algorithm we generate a bond on the new
edge with 50% probability. For example a cluster configuration in Fig. 5 changes into
two possible cluster configurations with an equal probability by the flip procedure. In
this way a change in connectivity of two triangles concerned with updating entails a
change in the assignments of neighbors and therefore a change in the weight function
(k) (k) Q
P [{C}k ] = QC v b /ZN (K; GN ) and then the transition function, δWlk /(1 + δWlk ), can
be estimated according to the Metropolis scheme described in the above.
5. Numerical simulations and results
The Monte Carlo simulations of generalized Potts models on dynamically triangulated

surfaces are performed using the algorithm mentioned in the previous section. In our
simulations we only generate planar ϕ 3 -graphs of spherical topology with no tadpole or
self-energy configurations. And we perform simulations on graphs with the fixed number
N vertices.
We first define one sweep of our Monte Carlo simulation. In the process of updating for
a given cluster configuration with a fixed triangulation one cluster-sweep means checking
of the Metropolis procedure on each edge of the graph in a given order. On the other hand,
in the process of updating dynamical triangulation, one triangulation-sweep means that
we randomly pick up edges one after another with roughly 3N/2 (total number of edges)
times and try to flip them and at the same time we check to eliminate every triangulations
containing tadpole or self-energy graph.
After generating an initial cluster configuration, we thermalize our system by perform-
ing 2000 triangulation-sweeps and 1000 cluster-sweeps, respectively. For the observables
to be measured, this thermalization time is enough to thermalize our system according to
the measurement of the autocorrelation time at criticality. To obtain a typical cluster sam-
ple configuration on a typical random surface background we perform 100 triangulation-
sweeps and 20 cluster-sweeps. Then we measure several observables for the given con-
figuration. We iterate these operations enough times until independent samples of a given
number, which depends on the lattice size and on the observable to be measured, are ob-
tained.
5.1. The critical coupling constants
It is a well accepted observation that lattice models with a second order phase transition
lead to corresponding continuum theories at the second order phase transition point.
In particular minimal conformal lattice models lead to the corresponding continuum
conformal field theory models at the critical point in two dimensions. It is well known that
the Q-state Potts models on regular two-dimensional lattice correspond to the field theories
of minimal unitary conformal series at the corresponding critical coupling constant Kc . The
correspondence is given in Eq. (3.9). It is analytically known that the Q-state Potts models
make continuous (second order or even higher order) phase transition at the critical point
for 0 Q 4, while they have first order phase transition point for Q > 4.
It is analytically not known if the nature of the phase transition may be changed when
gravity coupled to the Potts models. For several examples of minimal unitary models, it
is known that the order of phase transition can be raised to higher order when gravity
is coupled [36,37]. In the case of Q = 10 and 200, numerical results suggests a strong
evidence in favor of continuous transitions [23]. In our simulations we assume that the
Q state Potts models coupled to gravity presented by Eq. (3.10) have continuous phase
transitions for 0 Q 4 even at the non-unitary value of c.
In order to locate the critical coupling by using finite-size scaling method, we investigate
the percolation probability P (K) and the cluster size distribution ns (K) of percolation
theory [38]. Let us briefly summarize the physical meaning of the percolation probability
P (K) and the cluster size distribution ns (K) by studying the simplest Potts model of Q = 1
Q=1
case. The partition function of Q = 1 Potts model ZN (K) is given by Eq. (3.6). By the
last expression of Eq. (3.6) we can recognize that p with v = eK − 1 = p/(1 − p) can be
identified as the probability of a bond being present at an edge [21]. When K → 0, v → 0
and thus p → 0, i.e., the probability of a bond being present at an edge is getting small,
then the probability of finding a maximum cluster (the bonds of the maximum cluster
reaches from one end to the other of the lattice extension) is expected to be zero. Let
us define a quantity PN (K) = {the maximum cluster size}/N , where the cluster size is
defined as a number of bond of the cluster. PN (K) is a probability of a vertex being on
the maximum cluster, where N is the total number of vertices. P (K) = limN→∞ PN (K)
is called percolation probability. This quantity is the order parameter of the percolation
transition and is expected to show the following critical singularity in the infinite lattice for
|K − Kc | → 0

P (K) ∼ (K − Kc ) , K Kc ,
βp
(5.1)
0, K Kc ,
where βp is the critical exponent associated to the percolation probability.
In a finite lattice with the size N , the finite-size percolation probability PN (K) cannot
vanish at any K > 0, and its behavior depends on the lattice size. In Fig. 6 we show the
size dependence of PN (K) for Q = 2.5 and 0.6 as examples. As we can see, PN (K)
with finite size dependence does not have sharp rise in contrast with Eq. (5.1) but has
milder rise with respect to K. In these simulations we have performed with the lattice sizes
N = 100, 200, 400, 800 and 1600 for Q = 0.2, 0.4, 0.6, 0.8, 1.0, 1.5, 2.0, 2.5, 3.0, 3.5 and
4.0, respectively. For each lattice size the number of independent samples is 5000. The
range of the coupling constant K in which we measure the observables depends on the
models, and they are shown in Table 1. So far as the behavior of PN (K) is concerned, the
order of phase transition is consistent with second order.
For the behavior of PN (K) near K → Kc we suppose the following scaling behavior
based on the finite-size scaling hypothesis [39]

PN (K) = L−βp /ν FP (K − Kc ) L1/ν , (5.2)
Fig. 6. Size-dependence of PN (K) for Q = 2.5 and 0.6 as examples. The sizes of the systems are N = 100 (the
highest curve), 200, 400, 800 and 1600 (the lowest curve).
Table 1
The ranges and intervals of the coupling constant K in which we
measure the finite-size percolation probability PN (K) for various Q
Q Range of K Interval of K
0.2 0.25 ∼ 0.82 0.03 × 20 (points)
0.4 0.50 ∼ 1.07 0.03 × 20 (points)
0.6 0.70 ∼ 1.27 0.03 × 20 (points)
0.8 0.80 ∼ 1.37 0.03 × 20 (points)
1.0 0.83 ∼ 1.40 0.03 × 20 (points)
1.5 1.05 ∼ 1.62 0.03 × 20 (points)
2.0 1.23 ∼ 1.65 0.03 × 15 (points)
2.5 1.30 ∼ 1.87 0.03 × 20 (points)
3.0 1.42 ∼ 1.84 0.03 × 15 (points)
3.5 1.45 ∼ 2.02 0.03 × 20 (points)
4.0 1.55 ∼ 1.97 0.03 × 15 (points)
Fig. 7. The functions ΦN,N (K), defined by Eq. (5.3) for all pairs of Ni ’s where N = 100, 200, . . . , 1600. The
curves are Q = 2.5 and 0.6 as examples.
where (5.1) is assumed for the infinite system of N → ∞. L is the linear extension of
the system and ν is the critical exponent associated to the correlation length ξ(K) ∼
1/|K − Kc |ν . Since the total volume is proportional to N , we should make an identification
N = LdF with dF as fractal dimension. We may view ξ(K) as an average geodesic
distance. In order to extract Kc using Eq. (5.2), we define the following function ΦN,N (K)
[40]
ln[PN (K)/PN (K)]
ΦN,N (K) = , (5.3)
ln[N/N ]
for a pair of size (N, N ). The functions ΦN,N (K) and ΦN ,N (K) for two different pairs
of sizes (N, N ) and (N , N ) should thus intersect at Kc , and the intersection point should
yield −βp /νdF if corrections to finite-size scaling can be neglected. In Fig. 7 we plot the
functions ΦN,N (K) for all pairs of given sizes in the cases of Q = 2.5 and 0.6. As we
can see from Fig. 6, PN (K) grows with the increase of Q, while the intersection point is
relatively clear for larger Q. It is getting more difficult to find the intersection point for
smaller Q.
Fig. 8. Size-dependence of χN (K) for Q = 2.5 and 0.6 as examples. The sizes of system are N = 100 (the lowest
peak), 200, 400, 800, 1600 and 3200 (the highest peak).
Let us next define a quantity ns (K) = ps (K)/s, where ps (K) is a probability of a vertex
being on a cluster of size s. ns (K) can be recognized as a cluster size distribution. This
can be understood as follows: suppose we have ms clusters with seize s, we can obtain a
relation ps (K) = ms s/N and thus ns (K) = ps (K)/s = ms /N which is proportional to the
cluster number and thus can be understood as a cluster size distribution. Using the cluster
size distribution we define the so-called percolation susceptibility χ(K)

χ(K) = s 2 ns (K), (5.4)
s
where the prime-sum means that the maximum cluster is omitted from the summation.
This quantity is the average number of bonds of a (finite) cluster and is expected to show
the following critical singularity in the infinite lattice for |K − Kc | → 0

(K − Kc )−γp , K Kc ,
χ(K) ∼ (5.5)
(Kc − K)−γp , K Kc ,
where γp and γp are the critical exponents associated to the percolation susceptibility.
Since these quantities P (K) and χ(K) play a major role in usual percolation theory, we
use them as crucial quantities of determining the critical point of generalized Potts models.
In a finite lattice with the size N , the finite-size percolation susceptibility χN (K)
cannot diverge at the critical point Kc but reaches a maximum of finite height only. The
magnitude of this maximum depends on the size of the lattice. In Fig. 8 we show the size
dependence of χN (K) for Q = 2.5 and 0.6 as examples. In these simulations we take lattice
sizes N = 100, 200, 400, 800, 1600 and 3200 for various values of Q’s. The range of the
coupling constant K and the number of independent samples in which we measure the
observables depend on the models and the lattice sizes, and they are shown in Tables 2–4.
c (N) instead of the true critical
In a finite lattice there is so called pseudo-critical point K
point Kc , where χN (K) reaches the maximum. The finite size scaling hypothesis for
the correlation length [39] means that at the pseudo-critical point the correlation length
coincides with the linear extension of the system, i.e.,

ξ Kc (N) ∼ L ∼ N 1/dF , (5.6)
Table 2
The lattice sizes, ranges of the coupling constant K and the number of samples in which we measure the finite-size
percolation susceptibility χN (K) for Q = 0.2, 0.4, 0.6 and 0.8
Q # triangles Range of K Interval of K # samples
0.2 100 0.20 ∼ 0.58 0.02 × 20 (points) 20000
200 0.30 ∼ 0.58 0.02 × 15 (points) 20000
400 0.36 ∼ 0.64 0.02 × 15 (points) 20000
800 0.40 ∼ 0.67 0.03 × 10 (points) 20000
1600 0.48 ∼ 0.66 0.02 × 10 (points) 20000
3200 0.47 ∼ 0.74 0.03 × 10 (points) 5000
0.4 100 0.42 ∼ 0.80 0.02 × 20 (points) 20000
200 0.53 ∼ 0.81 0.02 × 15 (points) 20000
400 0.61 ∼ 0.89 0.02 × 15 (points) 20000
800 0.66 ∼ 0.93 0.03 × 10 (points) 20000
1600 0.75 ∼ 0.93 0.02 × 10 (points) 20000
3200 0.74 ∼ 1.01 0.03 × 10 (points) 5000
0.6 100 0.57 ∼ 1.01 0.02 × 23 (points) 20000
200 0.71 ∼ 1.05 0.02 × 18 (points) 20000
400 0.77 ∼ 1.09 0.02 × 17 (points) 20000
800 0.83 ∼ 1.10 0.03 × 10 (points) 20000
1600 0.86 ∼ 1.16 0.03 × 11 (points) 20000
3200 0.86 ∼ 1.19 0.03 × 12 (points) 10000
0.8 100 0.70 ∼ 1.14 0.02 × 23 (points) 20000
200 0.83 ∼ 1.11 0.02 × 15 (points) 20000
400 0.91 ∼ 1.19 0.02 × 15 (points) 20000
800 0.95 ∼ 1.22 0.03 × 10 (points) 20000
1600 0.97 ∼ 1.27 0.03 × 11 (points) 20000
3200 1.02 ∼ 1.29 0.03 × 10 (points) 10000
which, using ξ(K) ∼ |K − Kc |−ν , yields for N → ∞

Kc (N) − Kc ∼ N −1/νdF . (5.7)
In order to extract Kc we make three-parameter fit to Eq. (5.7). In Fig. 9 we show the best
c (N) for each lattice size have been estimated
linear fits for Q = 2.5 and 0.6. The error of K
using a polynomial fit to K c (N). Extrapolating to N = ∞ yields the true critical coupling
Kc .
For large Q the critical couplings Kc determined by using two methods as above are
in agreement within the error. The discrepancy in the values is presumably due to the
fact that larger lattices are needed before the methods converge completely. In Fig. 10 we
have plotted the best values of Kc versus values of Q’s. In this figure we have drawn the
fitting function to data points with a polynomial of fifth-order in Q1/2 . We also show the
theoretical critical couplings for the Q-state Potts models on the honeycomb [41,42] and
square [20] lattice (i.e., no longer coupled to gravity). As these two curves show a similar
trend, this figure suggests the existence of exact solutions for the Q-state Potts models
coupled to gravity.
In the following subsections we use these values of the critical value Kc obtained from
the best-fit curve of Fig. 10.
Table 3
The same of Table 2 for Q = 1.0, 1.5, 2.0 and 2.5
1.0 100 0.82 ∼ 1.20 0.02 × 20 (points) 20000
200 0.94 ∼ 1.22 0.02 × 15 (points) 20000
400 1.00 ∼ 1.32 0.02 × 17 (points) 20000
800 1.05 ∼ 1.32 0.03 × 10 (points) 20000
1600 1.13 ∼ 1.31 0.02 × 10 (points) 20000
3200 1.14 ∼ 1.34 0.02 × 11 (points) 10000
1.5 100 1.01 ∼ 1.39 0.02 × 20 (points) 20000
200 1.13 ∼ 1.41 0.02 × 15 (points) 20000
400 1.19 ∼ 1.47 0.02 × 15 (points) 20000
800 1.28 ∼ 1.46 0.02 × 10 (points) 20000
1600 1.31 ∼ 1.49 0.02 × 10 (points) 20000
3200 1.33 ∼ 1.51 0.02 × 10 (points) 10000
2.0 100 1.10 ∼ 1.60 0.02 × 26 (points) 20000
200 1.20 ∼ 1.60 0.02 × 21 (points) 20000
400 1.28 ∼ 1.64 0.02 × 19 (points) 20000
800 1.37 ∼ 1.61 0.02 × 13 (points) 20000
1600 1.41 ∼ 1.63 0.02 × 12 (points) 20000
3200 1.45 ∼ 1.63 0.02 × 10 (points) 10000
2.5 100 1.26 ∼ 1.64 0.02 × 20 (points) 20000
200 1.37 ∼ 1.65 0.02 × 15 (points) 20000
400 1.42 ∼ 1.70 0.02 × 15 (points) 20000
800 1.50 ∼ 1.68 0.02 × 10 (points) 20000
1600 1.53 ∼ 1.71 0.02 × 10 (points) 20000
3200 1.54 ∼ 1.72 0.02 × 10 (points) 10000
Table 4
The same of Table 2 for Q = 3.0, 3.5 and 4.0
3.0 100 1.35 ∼ 1.73 0.02 × 20 (points) 20000
200 1.46 ∼ 1.74 0.02 × 15 (points) 20000
400 1.50 ∼ 1.78 0.02 × 15 (points) 20000
800 1.59 ∼ 1.77 0.02 × 10 (points) 20000
1600 1.60 ∼ 1.80 0.02 × 11 (points) 20000
3200 1.62 ∼ 1.80 0.02 × 10 (points) 10000
3.5 100 1.40 ∼ 1.78 0.02 × 20 (points) 20000
200 1.53 ∼ 1.81 0.02 × 15 (points) 20000
400 1.58 ∼ 1.86 0.02 × 15 (points) 20000
800 1.66 ∼ 1.84 0.02 × 10 (points) 20000
1600 1.68 ∼ 1.86 0.02 × 10 (points) 20000
3200 1.69 ∼ 1.87 0.02 × 10 (points) 10000
4.0 100 1.42 ∼ 1.90 0.02 × 25 (points) 20000
200 1.56 ∼ 1.90 0.02 × 18 (points) 20000
400 1.65 ∼ 1.93 0.02 × 15 (points) 20000
800 1.73 ∼ 1.91 0.02 × 10 (points) 20000
1600 1.74 ∼ 1.92 0.02 × 10 (points) 20000
3200 1.75 ∼ 1.93 0.02 × 10 (points) 10000
Fig. 9. The best linear fit to Eq. (5.7) for Q = 2.5 and 0.6, where δ = 1/νdF .
Fig. 10. The best values of Kc versus Q. The fat solid line is the fitting curve to the data points with polynomials
of fifth-order in Q1/2 (gravity). We also show the theoretical critical coupling for the Q-state Potts models on the
honeycomb and square lattice (flat).
5.2. The string susceptibility
The string susceptibility exponent γs is one of the simplest quantities which characterize
the fractal structure of quantum gravity. This quantity is introduced as the exponent of
the subleading correction to the canonical partition function for random surfaces of fixed
volume A
Z(A) ∼ eΛc A Aγs −3 , (5.8)
for A → ∞, where Λc denotes the critical cosmological constant. As we show in Eq. (2.2),
the total area A is proportional to the number of triangles N in the regularized counterpart
of the Eq. (5.8). In the case of pure gravity, Z(A = N) (a 2 = 1) is equal to the number of
inequivalent triangulations with volume N .
Physically the string susceptibility can be identified as an order parameter for the
branching probability of random surface. This could be understood by the following
relation:
A
1
dB BZ(B)(A − B)Z(A − B) ∼ Aγs , (5.9)
Z(A)
0
where BZ(B) can be identified as a possible number of triangulation with a marked point
on a triangulated surface of area B. Thus the left-hand side of Eq. (5.9) measures the
average branching rate of the total surface branching into two parts [43]. For γs > 0 the
surface has tendency to branch more while for γs < 0 the surface tends to be smooth.
Using the analytical approach in a continuum framework, c-dependence of γs was first
derived by KPZ [1] and later rederived by DDK [24] by using conformal gauge formulation
of Liouville theory,
√
c − 1 − (25 − c)(1 − c)
γs (c) = , (5.10)
12
for two-dimensional quantum gravity coupled to the matter central charge c with spherical
topology.
In this subsection we investigate the string susceptibility exponent γs by measuring the
distributions of so-called baby universes [12,13,18]. It has already been pointed out that
the numerical values of string susceptibility are in perfect agreement with the theoretical
results (5.10) of Q-state Potts models for integral values of Q’s. We then expect that the
agreement will be perfect even for the non-integral values of Q’s. We intend to use the
numerical investigations of γs (c) as the cross check of the critical values of Kc calculated
in the previous subsection.
The branching probability of the surface with total area A into B and A − B is given by
the integrand of Eq. (5.9). The lattice counterpart of this branching probability is given by
3(B + 1)Z(B + 1)(N − B + 1)Z(N − B + 1)
bN (B) ∼
Z(N)
γs −2
B
∼N B 1− (1 B < N), (5.11)
N
where two baby universes are divided by single triangle in the current formulation of
triangulations. We may call the smaller part of the minimum neck as a baby universe
(minbu). In numerical simulations it is easy to find the shortest loops in a given
triangulation and then to enumerate the area of the corresponding minbu’s.
The simulations were done with lattice sizes N = 1000 and 2000 for various values
of Q. For each lattice size the number of independent samples is 100K. The values of Q’s
and the critical couplings Kc (Q) are shown in Table 5.
In order to extract γs we have fitted the distributions expressed in the form

B a2
ln bN (B) = a1 + (γs − 2) ln B 1 − + , (5.12)
N B
where a1 and a2 are fitting parameters and the last term is a finite size correction term
for small B [13]. We have introduced a lower cut-off Bc , because Eq. (5.11) is only
Table 5
The critical coupling constants Kc for various values of Q
Q Kc
0.00 0.000
0.05 0.428
0.20 0.764
0.50 1.053
0.80 1.212
1.00 1.288
1.50 1.434
2.00 1.547
2.50 1.647
3.00 1.732
3.50 1.800
4.00 1.848
Fig. 11. The measured string susceptibility γs versus central charge c and the theoretical curve. No logarithmic
corrections are introduced in the fits.
asymptotically correct, deviations can be expected for small B. Moreover we have cut
large B part, because the distributions of minbu’s are not universal for large B.
The values γs (Bc ) extracted from Eq. (5.12) approach exponentially to a limiting value
for large Bc . Thus in order to extract the limiting value γs we fit the values γs (Bc ) in such
a form
γs (Bc ) = γs − a1 e−a2 Bc . (5.13)
In Fig. 11 we plotted the limiting value γs obtained by the above method versus central
charge c together with the theoretical curve (5.10).
The reason why the results for c ≈ 1 disagree with the theoretical curve is possibly due
to the fact that logarithmic corrections are not yet introduced. It is well known [12,18] that
logarithmic corrections play an important role in the vicinity of c ≈ 1. Thus we assume
that the partition function has the following asymptotic behavior for large N
Z(N) ∼ eλc N N γs −3 (ln N)α , (5.14)
Fig. 12. The measured string susceptibility γs versus central charge c and the theoretical curve. Logarithmic
corrections are introduced in the fits for Q = 2.5, 3.0, 3.5 and 4.0.
where α is an additional parameter. The measured distributions bN (B) can now be fitted
to the following parameterization:

B a2
ln[bN (B)] = a1 + (γs − 2) ln B 1 − + α ln ln B ln(N − B) + , (5.15)
N B
for Q = 2.5, 3.0, 3.5 and 4.0. Then γs with the logarithmic corrections versus central
charge c are plotted in Fig. 12.
The string susceptibility γs with the logarithmic corrections for the various values of
Q’s are in very good agreement with the theoretical curve. We can then conclude that the
values of critical coupling Kc (Q) evaluated numerically in the previous subsection are
correct within errors.
5.3. The fractal dimension
The most straightforward definition of the fractal dimension is given by
N(r) = r dF , (5.16)
where we count the number of triangles N(r) inside r steps from a marked triangle. In
fact the fractal structure of the two-dimensional quantum gravity was first confirmed in
this way by Kawamoto, Kazakov, Saeki and Watabiki for c = −2 scalar fermion model
[5]. In these analyses they needed 5 million triangles to obtain the reliable value of the
fractal dimension. It was later recognized that finite size scaling hypothesis is very useful
to evaluate the fractal dimension numerically and thus relatively small number of triangles
is enough to obtain very accurate fractal dimension for c = −2 model [15,16]. Here we use
finite size scaling hypothesis to obtain the c-dependence of the fractal dimension.
It has already been recognized numerically in [5] that the total perimeter length sN (r)
measured at geodesic distance r from a marked triangle grows
sN (r) = r dF −1 . (5.17)
This fact triggered the investigation of analytic derivation of transfer matrix for two-
dimensional random surface of quantum gravity for c = 0 model [7]. It has then been
recognized that two point function of random surface can be related to the measurement of
sN (r) [9,14,44].
In the case of pure gravity (c = 0) “two-point function” with fixed volume A is defined
by

1 2 √ 1 √
SA (R) = D[g] δ d x g−A d 2x g
Z(A) A

√
× d 2 x g δ Dg (x, x ) − R , (5.18)
where Dg (x, x ) denotes the geodesic distance between the points labeled by x and x
measured with respect to g. Then SA (R) dR is the average area of a spherical shell of
thickness dR and radius R from a marked point on the manifold. We recognize that the
lattice triangulation version of SA (R) corresponds to sN (r). According to the numerical
result, we expect to have a relation
SA (R) ∼ R dF −1 for R ∼ 0. (5.19)
In case of pure gravity (c = 0) SA (R) was calculated analytically

SA (R) = R 3 f R/A1/4 , (5.20)
where f (x) ∼ e−x

4/3
for large x [9,14]. This analytic result is consistent with the
calculation of the fractal dimension dF = 4 for c = 0 model derived by transfer matrix
formalism [7]. This result, however, strongly suggests the following scaling hypothesis for
general matter central charge c coupled two-dimensional quantum gravity

SA (R) = A1−1/dF F R/A1/dF , (5.21)
where
F (x) ∼ x dF −1 , x 1. (5.22)
This beautiful scaling behavior was observed with a high accuracy in the very systematic
numerical analyses of the fractal structure of two-dimensional quantum gravity coupled to
c = −2 matter [15,16]. Here we assume that the scaling hypothesis works in general for
the general matter central charge −2 c 1.
In our numerical simulations we define a geodesic distance as a “triangle distance”
instead of link distance on a triangulation for saving the CPU time. The triangle distance
denotes the shortest path along neighboring triangles between the two separate triangles.
Therefore the triangle distance is equal to the “edge distance” in the dual ϕ 3 -graph. Then
the discretized analogue of SA (R), sN (r) with A = Na 2 (a 2 = 1), consists of all triangles
with triangle distance r measured from a marked triangle. Corresponding to the above
continuum description we expect the following behavior for sN (r):

r
sN (r) ∼ N 1−1/dF F (x), x= , (5.23)
N 1/dF
and we expect F (x) to behave as x dF −1 for small x.
In order to measure correlation function sN (r) the simulations were carried out
with lattice sizes ranging; N = 100, 200, 400, 800, 1600, 3200 and 6400 for Q =
0.0, 0.05, 0.2, 0.5 and 0.8. For Q = 1.0, 1.5, 2.0, 2.5, 3.0, 3.5 and 4.0 we have performed
the simulations with lattice sizes ranging; N = 100 ∼ 6400 and 12800. For each lattice size
ranging; N = 100 ∼ 1600 the number of independent samples are 100K, and we choose
20 initial random vertices (rv) on each configuration. Then the number of independent
samples is 50K × 40(rv) for N = 3200, 20K × 100(rv) for N = 6400 and 10K × 200(rv)
for N = 12800, respectively. The values of Q’s and its critical couplings Kc (Q) are shown
in Table 5.
In order to extract the fractal dimensions dF from the scaling relation (5.23), It is
crucial to introduce the so-called shift parameter a [14–16] which accommodate the
finite size effects. This parameter a is considered as the leading order correction in the
scaling variable x = r/N 1/dF . This is reasonable from the point of view that the shortest
distances may include lattice artifacts, and thus we cannot expect an exact agreement
with the continuum formulae. On the other hand, in order to incorporate the higher order
corrections in the scaling variable x we need to introduce a second shift parameter b [14]:
N 1/dF → N 1/dF + b. In this way we are led to a “phenomenological” scaling variable x
r +a
x= . (5.24)
N 1/dF + b
In the present simulations with small lattice sizes (at most N ∼ 103 ) we need to introduce
the parameter b, however, it is numerically very hard to determine the optimal values of
three parameters a, b and dF in the case of the gravity coupled to matter. Although such
optimal values of parameters are obtained, they keep the dependence on lattice size.
We have extracted the fractal dimensions dF by the following method. First we use the
decay of the peak of sN (r)/N with N (this is similar to the use of finite-size scaling in
the case of the percolation susceptibility). From Eq. (5.23) we expect the following scaling
behavior:
sN (r)/N ∼ N −1/dF . (5.25)

We have divided N ’s into three successive divisions such as N ’s ((100, 200, 400),
(200, 400, 800), etc.), and we have performed to fit sN (r)/N = a1 N −1/dF for each
successive divisions by two parameters. In this way the dF dependence on N becomes
clear. Then we further assume the following scaling behavior:

1 1
−α
−
dF (N) dF (∞) ∼ N , (5.26)
where we can extrapolate 1/dF (∞) by a linear fit to Eq. (5.26). This is shown in Fig. 13 for
Q = 2.5 and 0.5 as examples, where the estimation of error is based on a non-linear fits to
sN (r)/N . The values of dF ’s obtained in this way for the various values of Q’s are plotted
Fig. 13. The best linear fit to Eq. (5.26) for Q = 2.5 and 0.5 as examples, by the decay of the peak of sN (r)/N
with N .
Fig. 14. The best linear fit to Eq. (5.27) for Q = 2.5 and 0.5 as examples, by the decay of the inverse average
radius 1/rN with N .
Fig. 15. The measured fractal dimension dF by the decay of peak versus central charge c and the three theoretical
curves given by Eqs. (2.12), (2.20) and (2.32).
in Fig. 15. Three theoretical curves given by the formulae; (2.12), (2.20) and (2.32), are
shown to be compared. It is clear that the formula (2.32) is closer to the numerical values
of the fractal dimension.
Fig. 16. The measured fractal dimension dF by the average radius versus central charge c and the three theoretical
curves given by Eqs. (2.12), (2.20) and (2.32).
Secondly, in order to make use of the whole information of sN (r) we have used the
average radius rN of universes with volume N
∞
1
rN ≡ rsN (r) ∼ N 1/dF . (5.27)
N
r=0
We can then expect 1/rN ∼ N −1/dF , where dF can be determined in the same way as
above. In Fig. 14 we show the corresponding linear fits to Eq. (5.27). The values of dF ’s
obtained in this way for the various values of Q’s are plotted in Fig. 16.
These two independent results are consistent with each other and support the theoretical
prediction dF(3) (c).
6. Conclusion and discussions
In this article we have shown the results of numerical investigations of the fractal
dimension of two-dimensional quantum gravity coupled to the matter central charge c for
−2 c 1. The c-dependence of the matter central charge is introduced by reformulating
Q-state Potts model into the model which can be identified as a generalization of
percolation cluster model, weighted percolation cluster model, on the random lattice.
In this formulation Q can be generalized into non-integer value and thus continuous
c-dependence is realized and then we have called this model simply as the generalized
Potts model. Since the model has a percolation cluster feature, we have formulated a new
Metropolis algorithm to generate clusters on dynamically triangulated surface.
The c-dependence of the critical coupling Kc is not known theoretically. We have
evaluated the c-dependence of Kc by measuring percolation probability and percolation
susceptibility with the help of scaling hypothesis. The c-dependence of the critical coupling
has similar behaviour as those of flat lattice. It is then very natural to expect that there
is a theoretical solution of theoretical coupling Kc in two-dimensional quantum gravity

coupled to matter central charge c. The order of the phase transition is assumed to be
second or even higher order at the critical point and we have not observed any evidence
against this assumption. The string susceptibility is measured by the baby universe
technique and has excellent agreement with the theoretical curve. It is recognized that
the next leading logarithmic correction is important to improve the agreement with the
theoretical prediction in particular for the region near c ≈ 1. Several parameters were
needed to incorporate the finite size effects. The string susceptibility measurement is
carried out as a cross check of the critical values of Kc and the result was perfectly
consistent with the independent measurement of the critical exponent in the above.
The c-dependence of the fractal dimension is measured based on the two-point function
of quantum surface with the finite size scaling hypothesis. We needed two extra parameters
to accommodate the finite size corrections to the scaling parameter. Measurements are
carried out by two methods; the decay behaviour of the peak of two-point function and the
average radius of universes. The results agree well each other. The c-dependence of the
fractal dimension is excellent agreement with the following theoretical prediction except
for the region c ≈ 1:
√ √
(3) 25 − c + 49 − c
dF (c) = 2 × √ √ . (6.1)
25 − c + 1 − c
We consider that the deviation of the agreement of the numerical values of the fractal
dimensions near c ≈ 1 from the theoretical values is possibly due to the fact that the size
of the lattice is not large enough to observe the possible discrete jump from the fractal
dimension c < 1 of Eq. (6.1) to the value of branched polymer phase dF = 2, γs = 1/2
[45–47].
It is interesting to measure the change of fractal dimension very accurately in this
delicate region near c ≈ 1. The theoretical curve of the c-dependence of the fractal
dimension has infinite slope at c = 1 and then turns into imaginary value. We conjecture
that the fractal phase of two-dimensional quantum gravity c < 1 turns into branched
polymer phase in 1 < c. We expect that there is a discrete jump of the fractal dimension
at c = 1. In order to measure this discrete jump numerically we may need huge number of
triangles in the simulation.
It is interesting to compare with the measurement of the string susceptibility in
three-dimensional simplicial gravity. In three dimensions tetrahedron is the fundamental
simplex which corresponds to the triangle of two-dimensional quantum gravity. In three-
dimensional dynamical triangulation the gravitational constant can be a free parameter and
plays a role of central charge c of two-dimensional quantum gravity. It was measured that
the string susceptibility changes from negative region to the positive region where branched
polymer phase is expected [48]. Here again the phase change from the fractal phase to the
branched polymer phase is expected. For realistic higher dimensional simplicial quantum
gravity it would be important to understand the phase change such as the fractal-branched
polymer phase change.
Acknowledgements
We would like to thank I. Kostov and Y. Watabiki for useful discussions at the very
early stage of this work. This work is supported in part by Japanese Ministry of Education,
Science, Sports and Culture under the grant number 13640250.
References
[1] V. Knizhnik, A.M. Polyakov, A. Zamolodchikov, Mod. Phys. Lett. A 3 (1988) 819.
[2] J. Distler, Z. Hlousek, H. Kawai, Mod. Phys. Lett. A 5 (1990) 1093.
[3] H. Kawai, M. Ninomiya, Nucl. Phys. B 336 (1990) 115.
[4] M.E. Agishtein, A.A. Migdal, Int. J. Mod. Phys. C 1 (1990) 165;
M.E. Agishtein, A.A. Migdal, Nucl. Phys. B 350 (1991) 690.
[5] N. Kawamoto, V.A. Kazakov, Y. Saeki, Y. Watabiki, Phys. Rev. Lett. 68 (1992) 2113.
[6] N. Kawamoto, Y. Saeki, Y. Watabiki, in preparation;
Y. Watabiki, Prog. Theor. Phys. Suppl. 114 (1993) 1;
N. Kawamoto, Quantum gravity, in: K. Kikkawa, M. Ninomiya (Eds.), Proceedings of Nishinomiya–Yukawa
Workshop, Nishinomiya, World Scientific, 1992, p. 112;
N. Kawamoto, Current topics in theoretical physics, in: Y.M. Cho (Ed.), First Asia-Pacific Winter School
for Theoretical Physics, World Scientific, 1993.
[7] H. Kawai, N. Kawamoto, T. Mogami, Y. Watabiki, Phys. Lett. B 306 (1993) 19.
[8] Y. Watabiki, Nucl. Phys. B 441 (1995) 119;
Y. Watabiki, Phys. Lett. B 346 (1995) 46.
[9] J. Ambjørn, Y. Watabiki, Nucl. Phys. B 445 (1995) 129.
[10] J. Ambjørn, C.F. Kristjansen, Y. Watabiki, Nucl. Phys. B 504 (1997) 555.
[11] S. Oda, N. Tsuda, T. Yukawa, Prog. Theor. Phys. 99 (1998) 875.
[12] S. Jain, S.D. Mathur, Phys. Lett. B 286 (1992) 239.
[13] J. Ambjørn, S. Jain, G. Thorleifsson, Phys. Lett. B 307 (1993) 34.
[14] J. Ambjørn, J. Jurkiewicz, Y. Watabiki, Nucl. Phys. B 454 (1995) 313.
[15] J. Ambjørn, K.N. Anagnostopoulos, T. Ichihara, L. Jensen, N. Kawamoto, Y. Watabiki, K. Yotsuji, Phys.
Lett. B 397 (1997) 177.
[16] J. Ambjørn, K.N. Anagnostopoulos, T. Ichihara, L. Jensen, N. Kawamoto, Y. Watabiki, K. Yotsuji, Nucl.
Phys. B 511 (1998) 673.
[17] J. Ambjørn, K.N. Anagnostopoulos, Nicl. Phys. B 497 (1997) 445.
[18] J. Ambjørn, G. Thorleifsson, Phys. Lett. B 323 (1994) 7.
[19] K.N. Anagnostopoulos, P. Bialas, G. Thorleifsson, J. Stat. Phys. 94 (1999) 321.
[20] R.B. Potts, Proc. Cambridge Philos. Soc. 48 (1952) 106.
[21] C.M. Fortuin, P.W. Kasteleyn, J. Phys. Soc. Jpn. Suppl. 26 (1969) 11;
C.M. Fortuin, P.W. Kasteleyn, Physica 57 (1972) 536.
[22] J. Jurkiewicz, A. Krzywicki, B. Petersson, B. Soderberg, Phys. Lett. B 213 (1988) 511;
C.F. Baillie, D.A. Johnston, Phys. Lett. B 286 (1992) 44;
S. Catterall, J. Kogut, R. Renken, Phys. Lett. B 292 (1992) 277;
J. Ambjørn, B. Durhuus, T. Jonsson, G. Thorleifsson, Nucl. Phys. B 398 (1993) 568;
J. Ambjørn, G. Thorleifsson, M. Wexler, Nucl. Phys. B 439 (1995) 187.
[23] J. Ambjørn, G. Thorleifsson, M. Wexler, Nucl. Phys. B 439 (1995) 187.
[24] F. David, Mod. Phys. Lett. A 3 (1988) 651;
J. Distler, H. Kawai, Nucl. Phys. B 321 (1989) 509.
[25] S. Weinberg, in: S.W. Hawking, W. Israel (Eds.), General Relativity, an Einstein Centenary Survey,
Cambridge Univ. Press, 1979, p. 790;
R. Gastmans, R. Kallosh, C. Truffin, Nucl. Phys. B 133 (1978) 417;
S.M. Christensen, M.J. Duff, Phys. Lett. B 79 (1978) 213.
[26] E. Ising, Phys. 31 (1925) 253.

[27] R.J. Baxter, S.B. Kelland, F.Y. Wu, J. Phys. A 9 (1976) 397.
[28] W.T. Tutte, J. Combin. Theory 2 (1967) 301.
[29] F.Y. Wu, J. Phys. A 10 (1977) L113.
[30] J.N.V. Temperley, E.H. Lieb, Proc. R. Soc. London Ser. A 322 (1971) 251.
[31] Vl.S. Dotsenko, Nucl. Phys. B 235 (1984) 54.
[32] Vl.S. Dotsenko, V.A. Fateev, Nucl. Phys. B 240 (1984) 312.
[33] R.J. Glauber, J. Math. Phys. 4 (1963) 294.
[34] M. Sweeny, Phys. Rev. B 27 (1983) 4445.
[35] V.A. Kazakov, I.K. Kostov, A.A. Migdal, Phys. Lett. B 157 (1985) 295;
D.V. Boulatov, V.A. Kazakov, I.K. Kostov, A.A. Migdal, Nucl. Phys. B 275 (1986) 641.
[36] V.A. Kazakov, Phys. Lett. A 119 (1986) 140.
[37] D.V. Boulatov, V.A. Kazakov, Phys. Lett. B 186 (1987) 379.
[38] D. Stauffer, Introduction to Percolation Theory, Taylor & Francis, 1985.
[39] M.E. Fisher, M.N. Barber, Phys. Rev. Lett. 28 (1972) 1516.
[40] M.N. Barber, W. Selke, J. Phys. A 15 (1982) L617.
[41] T. Kihara, Y. Midzuno, T. Shizume, J. Phys. Soc. Jpn. 9 (1954) 681.
[42] D. Kim, R. Joseph, J. Phys. C 7 (1974) L167.
[43] H. Kawai, Nucl. Phys. B (Proc. Suppl.) 26 (1992) 93.
[44] J. Ambjørn, B. Durhuus, T. Jonsson, Quantum Geometry, Cambridge Univ. Press, 1997.
[45] J. Ambjørn, B. Durhuus, T. Jonsson, Phys. Lett. B 244 (1990) 403.
[46] B. Durhuus, Nucl. Phys. B 426 (1994) 203.
[47] J. Jurkiewicz, A. Krzywicki, Phys. Lett. B 392 (1997) 291.
[48] J. Ambjørn, D.V. Boulatov, N. Kawamoto, Y. Watabiki, Phys. Lett. B 480 (2000) 319.
A(1)
n−1 reflection K-matrices
A. Lima-Santos
Universidade Federal de São Carlos, Departamento de Física, Caixa Postal 676,
CEP 13569-905 São Carlos, Brazil
Received 17 July 2002; received in revised form 6 September 2002; accepted 6 September 2002
Abstract
We investigate the possible regular solutions of the boundary Yang–Baxter equation for the vertex
(1)
models associated with the An−1 affine Lie algebra. We have classified them in two classes of
solutions. The first class consists of n(n − 1)/2 K-matrix solutions with three free parameters. The
second class are solutions that depend on the parity of n. For n odd there exist n reflection K-matrices
with 2 + [n/2] free parameters. It turns out that for n even there exist n/2 K-matrices with 2 + n/2
free parameters and n/2 K-matrices with 1 + n/2 free parameters.
PACS: 75.10.Jm; 05.90.+m
Keywords: Reflection equation; K-matrix
1. Introduction
The search for integrable models through the Yang–Baxter equation [1–3]
R12 (u − v)R13 (u)R23 (v) = R23 (v)R13 (u)R12 (u − v) (1.1)

has been performed by the quantum group approach in [4]. Thus the problem is reduced
to a linear one. Indeed, R matrices corresponding to vector representations of all non-
exceptional affine Lie algebras were determined in this way in [5].
A similar approach is clearly desirable for finding solutions K(u) of the boundary
Yang–Baxter equation [6,7]
R12 (u − v)K1 (u)R21 (u + v)K2 (v) = K2 (v)R12 (u + v)K1 (u)R21 (u − v). (1.2)
E-mail address: dals@df.ufscar.br (A. Lima-Santos).
PII: S 0 5 5 0 - 3 2 1 3 ( 0 2 ) 0 0 8 1 9 - 2
A. Lima-Santos / Nuclear Physics B 644 [FS] (2002) 568–584 569
With this goal in mind, the study of boundary quantum groups was initiated in
[8]. However, as observed by Nepomechie [9], an independent systematic method of
constructing the boundary quantum group generators is not yet available. In contrast to
the bulk case [5], one cannot exploit boundary affine Toda field theory, since appropriated
classical integrable boundary conditions are not yet known [10].
We are also sharing the hope that by studying the known examples of boundary quantum
group generators, it may become possible to uncover their basic algebraic structure, and
to find generalizations to all affine Lie algebras. Independent of the lack of an algebraic
solution from the quantum group approach, there has been an increasing amount of
effort towards the understanding of two-dimensional integrable theories with reflecting
boundaries via solutions of the reflection equation (1.2). In field theory, attention is focused
on the boundary S-matrix. In statistical mechanics, the emphasis has been on deriving
solutions of (1.2) and the calculation of various surface critical phenomena, both at and
away from criticality [11]. In condensed matter physics the actual target is the impurity
problem.
The classification of all possible solutions of the reflection equation (1.2) by direct
computation has been seen as a very difficult problem. However, recently we have proposed
(2)
a method which allows the classification of the Dn+1 reflection K-matrices [12] as well
as the K-matrices of the 19-vertex models [13]. In spite of these papers we decided to
continue in this line in order to include the A(1)
n−1 reflection K-matrices which will reveal
us its algebraic structure.
(1)
We have organized this paper as follows. In Section 2 we choose the An−1 reflection
equations and in Section 3 their solutions are derived and classified in two types. The
last section is reserved for the conclusion. The first models have its K-matrices written
explicitly in Appendices A–D.
2. The A(1)
n−1 reflection equations
The R-matrix for the vertex models associated with the A(1)n−1 (n 2 ) affine Lie algebra
was originally found in the articles [14,15] and as presented in [5] it has the form

R(u) = a1 (u) Eii ⊗ Eii + a2 (u) Eii ⊗ Ejj
i=j

+ a3 (u) Eij ⊗ Ej i + a4 (u) Eij ⊗ Ej i , (2.1)
i<j i>j
where Eij denotes the elementary n by n matrices ((Eij )ab = δia δib ) and the Boltzmann
weights with functional dependence on the spectral parameter u are given by

a1 (u) = eu − q 2 , a2 (u) = q eu − 1 ,

a3 (u) = − q 2 − 1 , a4 (u) = −eu q 2 − 1 . (2.2)
Here q denotes an arbitrary parameter.
570 A. Lima-Santos / Nuclear Physics B 644 [FS] (2002) 568–584
For n > 2, the R-matrix (2.1) does not enjoy P and T symmetry but just PT invariance
P12 R12 (u)P12 ≡ R21 (u) = R12 (u)t1 t2 (2.3)

and unitarity
R12 (u)R21 (−u) = ζ (u) = a1 (u)a1 (−u). (2.4)

It is not crossing invariant either but it obeys the weaker property [16]
−1 t2 −1 ζ (u + ρ)
R12 (u)t2 = M2 R12 (u + 2ρ)M2−1 , (2.5)
ζ (u + 2ρ)
where M is a symmetry of the R-matrix
[R(u), M ⊗ M] = 0, Mij = δij q n+1−2i , ρ = n ln q. (2.6)

The matrix K− (u) satisfies the left boundary Yang–Baxter equation, also known as the
reflection equation,
1 2 2 1
R12 (u − v) K − (u)R21 (u + v) K − (v) =K − (v)R12 (u + v) K − (u)R21 (u − v),
(2.7)
which governs the integrability at boundary for a given bulk theory. A similar equation
(1)
should also hold for the matrix K+ (u) at the opposite boundary. However, for the An−1
models, one can see from [17] that the corresponding quantity
K+ (u) = K− (−u − ρ)t M, (2.8)

satisfies the right boundary Yang–Baxter equation. Here t = t1 t2 and ti stands for
transposition taken in the ith space.
Therefore, we can start for searching the matrices K− (u). In this paper only regular
(1)
solutions will be considered, although there is much interest for non-regular An−1 solutions
[18,19].
Regular solutions mean that the matrix K− (u) has the form

n
K− (u) = kij (u) Eij (2.9)
i,j =1
and satisfies the condition
kij (0) = δij , i, j = 1, 2, . . . , n. (2.10)

Substituting (2.1) and (2.9) into (2.7), we will get n4 functional equations for the kij
matrix elements, many of which are dependent. In order to solve them, we shall proceed
in the following way. First we consider the (i, j ) component of the matrix Eq. (2.7). By
differentiating it with respect to v and taking v = 0, we get algebraic equations involving
the single variable u and n2 parameters

dkij (v)
βij = , i, j = 1, 2, . . . , n. (2.11)
dv v=0
Second, these algebraic equations are denoted by E[i, j ] = 0 and collected into blocks
B[i, j ] , i = 1, . . . , n2 − i and j = i, i + 1, . . . , n2 − i, defined by


 E[i, j ] = 0,

 E[j, i] = 0,
B[i, j ] = (2.12)

 E n2 + 1 − i, n2 + 1 − j = 0,


E n2 + 1 − j, n2 + 1 − i = 0.
For a given block B[i, j ], the equation E[n2 + 1 − i, n2 + 1 − j ] = 0 can be obtained from
the equation E[i, j ] = 0 by interchanging
kij ↔ kn+1−i n+1−j , βij ↔ βn+1−i n+1−j
, a3 (u) ↔ a4 (u) (2.13)
and the equation E[j, i] = 0 is obtained from the equation E[i, j ] = 0 by the interchanging
kij ↔ kj i , βij ↔ βj i . (2.14)
In this way, we can control all equations and a particular solution is simultaneously
connected with at least four equations.
3. The A(1)
n−1 K-matrix solutions
Analyzing the A(1) n−1 reflection equations one can see that they possess a very special
structure. Several equations exist involving only the elements out of the diagonal, kij
(i = j ), these are the simplest equations and we will solve them first.
By direct inspection one can see that the diagonal blocks B[i, i] are uniquely solved by
the relations
βij kj i (u) = βj i kij (u), ∀ i = j. (3.1)
It means that we only need to find the n(n − 1)/2 elements kij (i < j ). Now we choose
a particular kij (i < j ) to be different from zero, with βij = 0, and try to express all
remaining elements in terms of this particular element. We have verified that this is possible
provided that

 a4 (u) βpq kij (u), if p > i and q > j,
a3 (u) βij
kpq (u) = β (p = q). (3.2)
 pq kij (u), if p > i and q < j
βij
Combining (3.1) with (3.2) we will obtain a very strong entail for the elements out of the
diagonal

kpj (u) = 0, for p = i,
kij (u) = 0 ⇒ (3.3)
kiq (u) = 0, for q = j.
It means that for a given kij , the only elements different from zero in the ith-row and in
the j th-column of K− (u) are kii , kij , kjj and kj i .
Analyzing more carefully these equations with the conditions (3.1) and (3.3), we have
found from the n(n − 1)/2 matrix elements kij (i < j ) that there are two possibilities to
choose a particular kij = 0:
• Only one non-diagonal element and its symmetric are allowed to be different from
zero. Thus we have n(n − 1)/2 reflection K-matrices with n + 2 non-zero elements.
Here we will denote by KIij (i < j ), the K-matrix for which the non-diagonal element
kij is the one chosen to be the non-zero matrix element. These matrices will be named
type-I solutions.
• For each kij = 0, additional non-diagonal elements and its asymmetric are allowed to
be different from zero provided they satisfy the equations
kij (u)kj i (u) = krs (u)ksr (u), with i + j = r + s mod n. (3.4)

It means that we will get n reflection K-matrices with the number of non-zero
elements depending on the parity of n. Next, we choose the n possible particular
elements as being k1j , j = 1, 2, . . . , n and k2n . We will also denote the corresponding
K-matrices by KII 1j , j = 1, 2, . . . , n and K2n , respectively. These matrices are named
II
type-II solutions.
For example, the A(1)

2 model has only type-I solutions. The K-matrices are
   
k11 k12 0 k11 0 k13
KI12 =  k21 k22 0 , KI13 =  0 k22 0 ,
0 0 k33 k31 0 k33
 
k11 0 0
K23 =
I  0 k22 k23  . (3.5)
0 k32 k33
(1)
One can expect that these are the three possibilities to write the same solution for the A2
model.
For the A(1)
3 model we have six type-I solutions {K12 , K13 , K14 , K23 , K24 , K34 } all with
I I I I I I
six non-zero elements. In this model we also have two type-II solutions {KII 12 , K14 }:
II
   
k11 k12 0 0 k11 0 0 k14
 0   0 
KII  k21 k22 0   0 k22 k23 ,
12 =  14 = 
and KII
0 0 k33 k34  0 k32 k33 0 
0 0 k43 k44 k41 0 0 k44
k12 k21 = k34 k43 , k14k41 = k23k32 . (3.6)

For n 5, in addition to the n(n − 1)/2 type-I solutions with n + 2 non-zero matrix
elements, we have also n type-II solutions with the following property: if n is odd these
K-matrices have 2n − 1 non-zero elements, but if n is even, half of these K-matrices have
2n non-zero elements and the remaining ones are matrices with 2n − 2 non-zero elements.
Although we already know as counting the K-matrices for the A(1) n−1 models we
still have to identify among them which are similar. Indeed we can see a Zn similarity
transformation which maps their matrix elements positions:
K (α) = hα K (0) hn−α , α = 0, 1, 2, . . . , n − 1, (3.7)
where hα are the Zn matrices
(hα )ij = δi,i+α mod n. (3.8)
In order to do this we can choose K (0) as KII

12 and the similarity transformations (3.7)
give us the K (α) matrices whose matrix elements are in the same positions of the matrix
elements of the KII1j and K2n matrices. However, due to the fact that the relations (3.2)
II
involve the ratio aa43 (u)

(u) = e , as well as the additional constraints (3.4), we could not find a
u
similarity transformation among these K’s matrices, even after a gauge transformation.
Even for the type-I solutions the similarity account is not simple due to the presence
of three types of scalar functions and the constraint equations for the parameters βij .
Nevertheless, as we have found a way to write all solutions, we can leave the similarity
account to the reader.
Having identified these possibilities we may proceed in order to find the n diagonal
elements kii (u) in terms of the non-diagonal elements kij (u) for each Kij matrix.
These procedure is now standard [13]. For instance, if we are looking for KII 12 , the non-
diagonal elements kij , (i + j = 3 mod n) in terms of k12 are given by
β


ij
for i + j = 3,
 β12 k12 (u),
kij (u) = βij u
for i + j = 3 mod n, (3.9)

 β12 e k12 (u),

0, otherwise,
for i, j = 1, 2, . . . , n (i = j ).
Substituting (3.9) into the reflection equations we can now easily find the kii elements
up to an arbitrary function, here identified as k12 (u). Moreover, their consistency relations
will yield us some constraints equations for the parameters βij .
After we have found all diagonal elements in terms of kij (u), we can, without loss of
generality, choose the arbitrary functions as
1
kij (u) = βij e2u − 1 , i < j. (3.10)
2
This choice allows us to work out the solutions in terms of the functions fii (u) and hij (u)
defined by
1
fii (u) = βii eu − 1 + 1 and hij (u) = βij e2u − 1 , (3.11)
2
for i, j = 1, 2, . . . , n.
Now, we will simply present the general solutions and write them explicitly for the first
values of n in Appendices A–D. Let us start considering the type-I solutions.
3.1. The type-I K-matrices
Here we have n(n − 1)/2 reflection K-matrices with n + 2 non-zero elements. For
1 < i < j n we get (n − 2)(n − 1)/2 solutions
KIij = fii (u)Eii + e2u fii (−u)Ejj + hij (u)Eij + hj i (u)Ej i

i−1 j −1

n
(i)
+ Zi (u) Ell + Yi+1 (u) Ell + e2u Zi (u) Ell , (3.12)
l=1 l=i+1 l=j +1
(i)
where Zi (u) and Yi+1 (u) are scalar functions defined by
1
Zi (u) = fii (−u) + (βii + β11 )e−u e2u − 1 (3.13)
2
and
1
Yl(i) (u) = fii (u) + (βll − βii ) e2u − 1 . (3.14)
2
For i = 1 and 1 < j n we get the n − 1 remaining solutions
KI1j = f11 (u)E11 + e2u f11 (−u)Ejj + h1j (u)E1j + hj 1 (u)Ej 1
j −1

n
+ Y2(1) (u) Ell + Xj +1 (u) Ell , (3.15)
l=2 l=j +1
where a new scalar function appears,

1
Xj +1 (u) = e2u f11 (−u) + (βj +1 j +1 + β11 − 2)eu e2u − 1 . (3.16)
2
The number of free parameters is fixed by the constraint equations which depend on the
presence of these scalar functions: when Yl(i) (u) is present in Kij we have constraint
equations of the type
βij βj i = (βll + βii − 2)(βll − βii ), (3.17)
but, when Zi (u) is present the corresponding constraints are of the type
βij βj i = (β11 + βii )(β11 − βii ). (3.18)
The presence of at least one Xj +1 (u) yields a third type of constraints,
βij βj i = (βj +1j +1 + β11 − 2)(βj +1j +1 − β11 − 2). (3.19)
From (3.12) and (3.15) we can see that in each KIij
we have at most two scalar functions.
It means that all these Kij matrices are 3-parameter solutions of the reflection equation.
I
Finally, we observe that the solution with i = 1 and j = n, i.e.,

KI1n = f11 (u)E11 + e2u f11 (−u)Enn + h1n (u)E1n
j −1

(1)
+ hn1 (u)En1 + Y2 (u) Ell (3.20)
l=2
has the constraint
β1n βn1 = (β22 + β11 − 2)(β22 − β11 ) (3.21)

and, it is the solution derived by Abad and Rios [20].
3.2. The type-II K-matrices
Due to the property (3.4) we have found three type-II general solutions for each A(1)
n−1
model:

Type-IIa = KII 12p , Type-IIb = KII 12p+1 , Type-IIc = KII2n ,

n
p = 1, 2, . . . , (3.22)
2
where [ n2 ] being the integer part of n2 .
For n-odd, the type-IIa solution is
[ n2 ]+p

p
n
KII
12p = f11 (u) Ejj + e f11 (−u)
2u
Ejj + e2u f11 (u) Ejj
j =1 j =p+1 j =[ n2 ]+p+2
+ X[ n2 ]+p+1 (u)E[ n2 ]+p+1 [ n2 ]+p+1

+ + u
e hij (u)Eij , (3.23)
i+j =1+2p i+j =1+2p mod n
i=j i=j
with constraint equations

βrs βsr = β[ n2 ]+p+1[ n2 ]+p+1 + β11 − 2 β[ n2 ]+p+1 [ n2 ]+p+1 − β11 − 2 ,
r + s = 1 + 2p mod n. (3.24)
For the type-IIb solutions we have obtained the following matrices
[ n2 ]+p+1

p
n
KII
12p+1 = f11 (u) Ejj + e f11 (−u)
2u
j =1 j =p+2 j =[ n2 ]+p+2

+ Xp+1 (u)Ep+1 p+1 + + e u
hij (u)Eij ,
i=j i=j
(3.25)
together with their constraint equations
βrs βsr = (βp+1 p+1 + β11 − 2)(βp+1 p+1 − β11 ),

r + s = 2 + 2p mod n. (3.26)
Finally, the type-IIc solution is the matrix KII

2n
[ n2 ]+1

n
KII
2n = Z2 (u)E11 + e f22 (−u)
2u
j =2 j =[ n2 ]+2

+ hij (u)Eij , (3.27)
i+j =2 mod n
i=j
for which the constraint equations are
βrs βsr = (β11 + β22 )(β11 − β22 ),

r + s = 2 mod n. (3.28)
The function Z2 (u) is given by (3.13) and the functions Xj (u) by (3.16), while the
functions f11 (u), f22 (u) and hij (u) are given by (3.11). Therefore we have n reflection
K-matrices for the A(1)
n−1 models (n odd). They are (2 + [ 2 ])-free parameter solutions with
n
2n − 1 non-zero matrix elements.

When n is even we have a similar identification but substantial differences exist.
In the n-even case the type-IIa solutions are the matrices
2 +p
n

p
n
KII
12p = f11 (u) Ejj + e f11 (−u)
2u
j =1 j =p+1 j = n2 +p+1

+ + e u
hij (u)Eij , (3.29)
i+j =2 i+j =1+2p mod n
i=j i=j
with constraint equations
β12p β2p1 = βrs βsr , r + s = 1 + 2p mod n. (3.30)

For the type-IIb solutions we have
n

p 2 +p

(1)
KII
12p+1 = f11 (u) Ejj + Yp+1 (u)Ep+1 p+1 + e f11 (−u)
2u
Ejj
j =1 j =p+2

n
+ X n2 +p+1 (u)E n2 +p+1 n
2 +p+1
+ e2u f11 (u) Ejj
j = n2 +p+2

+ + e u
hij (u)Eij , (3.31)
i=j i=j
with the following constraint equations

βrs βsr = (βp+1 p+1 + β11 − 2)(βp+1 p+1 − β11 )

= β n2 +p+1 n2 +p+1 + β11 − 2 β n2 +p+1 n2 +p+1 − β11 − 2 ,
r + s = 2 + 2p mod n. (3.32)
Again, the type-IIc solution is the matrix KII
2n
n

2
(2)
KII
2n = Z2 (u)E11 + f22 (u) Ejj + Y n +1 (u)E n2 +1 n
2 +1
2
j =2

n
+ e2u f22 (−u) Ejj + hij (u)Eij , (3.33)
j = n2 +2 i+j =2 mod n
i=j
with the constraint equations

βrs βsr = (β11 − β22 )(β11 − β22 )

= β n +1 n +1 + β22 − 2 β n +1 n +1 − β22 ,
2 2 2 2
r + s = 2 mod n, (3.34)
(2)
where the scalar functions Z2 (u) and Yp+1 (u)
are given by (3.13) and (3.14), respectively.
Here we observe that for n even, the type-IIa is a (2 + n2 )-free parameter solution with
2n non-zero matrix elements, while the type-IIb and the type-IIc are (1 + n2 )-free parameter
solutions with 2(n − 1) non-zero matrix elements.
4. Conclusion
The absence of an algebraic method such as the quantum group approaches leads us to
believe that a direct computation from their reflection equations should be a starting point
to obtain its classification.
After a systematic study of the functional equations we find that there are two types of
solutions for the A(1)
n−1 models. We call of type-I the K-matrices with three free parameters
and n + 2 non-zero matrix elements. These solutions were denoted by KIij to emphasize
the non-zero element out of the diagonal and its symmetric, which results in n(n − 1)/2
reflection K-matrices.
The type-II solutions are more interesting because their have many free parameters. The
(1)
An−1 models for n odd, in addition to the type-I solutions, have n type-II solutions with
2n − 1 non-zero matrix elements and (2 + [ n2 ]) free parameters. It turns out that for n even
we also have n type-II solutions but half of them are K-matrices with 2n non-zero matrix
elements and (2 + n2 ) free parameters, while the remaining ones have 2(n − 1) non-zero
matrix elements with (1 + n2 ) free parameters.
The corresponding K+ (u) are obtained from the isomorphism (2.8). Out of this
classification we have the trivial solution (K− = 1, K+ = M) for these models. Thus we
ended our discussion on the reflection matrices for the vertex models associated with the
A(1)
n−1 affine Lie algebra.
To complete the classification for all non-exceptional Lie algebras we still have to
(1) (1) (1) (2) (2)
consider the vertex models associated with the Bn , Cn , Dn , A2n and A2n−1 Lie
algebras.
Acknowledgement
This work was supported in part by Fundação de Amparo à Pesquisa do Estado de São
Paulo-FAPESP-Brasil and by Conselho Nacional de Desenvolvimento-CNPq-Brasil.
Appendix A. The A(1)

1 reflection K-matrices
(1)
This is a very special case among the An−1 models. We note that there is only one
general K-matrix with 4 non-zero matrix elements [21,22]. From the type-IIa solutions
(3.29) or from the type-I solutions (3.15) it is the K12 matrix

f11 (u) h12 (u)
KI12 = .
h21 (u) e2u f11 (−u)
Although there is no constraint equation in this case, the regular condition (2.10) has fixed
in three the number of free parameters, in agreement with all type-I reflection K-matrices.
Appendix B. The A(1)

2 reflection K-matrices
This is also a special case because it has only the type-I solutions KI12 , KI13 and KI23 .
From (3.15) we have
KI12 = f11 (u)E11 + e2u f11 (−u)E22 + h12 (u)E12 + h21 (u)E21 + X3 (u)E33
 
f11 (u) h12 (u) 0
=  h21 (u) e2u f11 (−u) 0 , (B.1)
0 0 X3 (u)
with the four parameters β11 , β12 , β21 and β33 satisfied the constraint equation
β12 β21 = (β33 − β11 − 2)(β33 + β11 − 2). (B.2)

Two diagonal solutions are derived from (B.1) due to this constraint equation
lim X3 (u) = e2u f11 (−u)

β33 →−β11 +2

⇒ D1 = diag f11 (u), e2u f11 (−u), e2u f11 (−u) (B.3)
and
lim X3 (u) = e2u f11 (u)

β33 →β11 +2

⇒ D2 = diag f11 (u), e2u f11 (−u), e2u f11 (u) . (B.4)
The matrix KI13 is also given by (3.15)
(1)
KI13 = f11 (u)E11 + e2u f11 (−u)E33 + h13 (u)E13 + h31 (u)E31 + Y2 (u)E22
 f (u) 0 h13 (u) 
11
= 0 (1)
Y2 (u) 0 , (B.5)
h31 (u) 0 e2u f11 (−u)
but now the constraint equation is
β13 β31 = (β22 + β11 − 2)(β22 − β11 ), (B.6)

and the corresponding diagonal reductions are
(1)
lim Y2 (u) = e2u f11 (−u),
β22 →−β11 +2

D3 = diag f11 (u), e2u f11 (−u), e2u f11 (−u) (B.7)
and
(1)
lim Y2 (u) = f11 (u),
β22 →β11

D4 = diag f11 (u), f11 (u), e2u f11 (−u) . (B.8)
For the solution named KI23 we recall Eq. (3.12) with i = 2 and j = 3
KI23 = f22 (u)E22 + e2u fii (−u)E33 + h23 (u)E23 + h32 (u)E32 + Z2 (u)E11
 
Z2 (u) 0 0
= 0 f22 (u) h23 (u)  , (B.9)
0 2u
h32 (u) e f22 (−u)
with the constraint
β23 β32 = (β11 + β22 )(β11 − β22 ) (B.10)

we get more two diagonal solutions
lim Z2 (u) = f22 (−u),

β22 →−β11

D5 = diag f22 (−u), f22 (u), e2u f22 (−u) (B.11)
and
lim Z2 (u) = f22 (u),

β22 →β11

D6 = diag f22 (u), f22 (u), e2u f22 (−u) . (B.12)
Due to the constraint equations these reflection K-matrices have only three free
parameters and the corresponding diagonal solutions have only one free parameter.
Here we observe that only four of these diagonal solutions are independents because
D6 = D4 and D3 = D1 . Here we also note that the solutions D1 and D4 are the diagonal
solutions derived by the first time in [21] and KI13 is the non-diagonal solution derived in
[20].
In a certain sense these solution are particular because they do not reveal us all properties
shared by the regular A(1)n−1 reflection K-matrices for n odd. Before we consider the next
odd case, let us consider the case n = 4.
(1)
Appendix C. The A3 reflection K-matrices
In this case the structure of the general solution begins to appear but it is still particular
because half of the type-II solutions are type-I solutions.
The K1j matrices for the type-I solutions are given by (3.15). For KI12 we get
 
f11 (u) h12 (u) 0 0
 h21 (u) e2u f11 (−u) 0 0 
KI12 =  0
 (C.1)
0 X3 (u) 0 
0 0 0 X3 (u)
with the constraint
β12 β21 = (β33 + β11 − 2)(β33 − β11 − 2). (C.2)

For KI13 we have
 f (u) 0 h13 (u) 0 
11
 0
(1)
Y2 (u) 0 0 
KI13 = 
 h (u)

 (C.3)
31 0 e2u f11 (−u) 0
0 0 0 X4 (u)
with the constraint
β13 β31 = (β44 + β11 − 2)(β44 − β11 − 2) = (β22 + β11 − 2)(β22 − β11 ). (C.4)
The KI14 matrix is
 
f11 (u) 0 0 h14 (u)
 0 Y2(1) (u) 
 0 0 
KI14 =   (C.5)
 0 0
(1)
Y2 (u) 0 
h41 (u) 0 0 e2u f11 (−u)
with
β14 β41 = (β22 + β11 − 2)(β22 − β11 ). (C.6)

The remaining type-I K-matrices are given by (3.12). For KI23 we get
 
Z2 (u) 0 0 0
 0 f22 (u) h23 (u) 0 
KI23 = 
 0 2u

 (C.7)
h32 (u) e f22 (−u) 0
0 0 0 e2u Z2 (u)
with constraint
β23 β32 = (β11 + β22 )(β11 − β22 ). (C.8)

For KI24 we have
 
Z2 (u) 0 0 0
 0 f22 (u) h24 (u) 
 0 
KI24 =  (2)  (C.9)
 0 0 Y3 (u) 0 
0 h42 (u) 0 e2u f22 (−u)
with the constraint
β24 β42 = (β11 + β22 )(β11 − β22 ) = (β33 + β22 − 2)(β33 − β22 ) (C.10)
and finally for KI34
 
Z3 (u) 0 0 0
 0 Z3 (u) 0 0 
KI34 =  0
 (C.11)
0 f33 (u) h34 (u) 
0 0 h34 (u) e2u f33 (−u)
with
β34 β43 = (β11 + β33 )(β11 − β33 ). (C.12)

For the type-IIa solutions we get from (3.29) more two K-matrices
 
f11 (u) h12 (u) 0 0
 h21 (u) e2u f11 (−u) 0 0 
KII =   (C.13)
12  0 0 e f11 (−u) e h34 (u) 
2u u
0 0 eu h43 (u) e2u f11 (u)

with the eight non-zero elements satisfying a constraint equation
β12 β21 = β34 β43 . (C.14)

The another K-matrix is given by
 
f11 (u) 0 0 h14 (u)
 0 f (u) h (u) 0 
KII  11 23 
14 =  2u  (C.15)
0 h32 (u) e f11 (−u) 0
h41 (u) 0 0 2u
e f11 (−u)
with a constraint equation

β14 β41 = β23 β32 . (C.16)
Note that both type-II solutions (C.13) and (C.15) have four free parameters.
Next, we can solve these constraint equations to derive eighteen diagonal solutions.
Using the following reductions for the scalar functions Xj +1 (u), Yl(i) (u) and Zi (u)
lim Xj +1 (u) = e2u f11 (−u),
βj+1 j+1 →−β11 +2
lim Xj +1 (u) = e2u f11 (u),

βj+1 j+1 →β11 +2
lim Yl(i) (u) = e2u fii (−u), lim Yl(i) (u) = fii (u),
βll →−βii +2 βll →βii
lim Zi (u) = fii (−u), lim Zi (u) = fii (u). (C.17)
β11 →−βii β11 →βii
we can see that only half of these diagonal solutions are independents:

D1 = diag f (u), e2u f (−u), e2u f (−u), e2u f (−u) ,

D2 = diag f (u), e2u f (−u), e2u f (u), e2u f (u) ,

D3 = diag f (u), f (u), e2u f (−u), e2u f (−u) ,

D4 = diag f (u), e2u f (−u), e2u f (−u), e2u f (u) ,

D5 = diag f (u), f (u), e2u f (−u), e2u f (u) ,

D6 = diag f (u), f (u), f (u), e2u f (−u) ,

D7 = diag f (−u), f (u), e2u f (−u), e2u f (−u) ,

D8 = diag f (−u), f (u), f (u), e2u f (−u) ,

D9 = diag f (−u), f (−u), f (u), e2u f (−u) , (C.18)
where we have used a compact notation for the functions fii (u)

fii (u) ≡ f (u) = β eu − 1 + 1 (C.19)
where β is the free parameter.
Appendix D. The A(1)

4 type-II reflection K-matrices
Here we will only write explicitly the five type-II solutions and their constraint equations
(1)
for the A4 model. They have nine non-zero matrix elements and four free parameters:
 f (u) h (u) 0 0 0 
11 12
 h21 (u) e2u f11 (−u) 0 0 0 
 
K12 = 
II
 0 0 e 2u f (−u)
11 0 e u h (u)  ,
35 
 0 0 0 X4 (u) 0 
0 0 u
e h53 (u) 0 2u
e f11 (u)
β12 β21 = β35 β53 = (β44 + β11 − 2)(β44 − β11 − 2), (D.1)
 f (u) 0 0 h14 (u) 0 

11
 0 f11 (u) h23 (u) 0 0 
 
KII  0 0 
14 =  h32 (u) e2u f11 (−u) 0 ,
 h (u) 0 0 2u
e f11 (−u) 0 
41
0 0 0 0 X5 (u)
β14 β41 = β23 β32 = (β55 + β11 − 2)(β55 − β11 − 2), (D.2)
 f (u) 0 h13 (u) 0 0 
11
 0 X2 (u) 0 0 0 
 
KII  ,
13 =  h31 (u) 0 2u
e f11 (−u) 0 0 
 0 0 0 e2u f11 (−u) e h45 (u) 
u
0 0 0 eu h54 (u) e2u f11 (u)

β13 β31 = β45 β54 = (β22 + β11 − 2)(β22 − β11 ), (D.3)
 f (u) 0 0 0 h15 (u) 
11
 0 f11 (u) 0 h24 (u) 0 
 
KII  0 ,
15 =  0 X3 (u) 0 0 
 0 h42 (u) 0 e2u f11 (−u) 0 
h51 (u) 0 0 0 e2u f11 (−u)
β15 β51 = β24 β42 = (β33 + β11 − 2)(β33 − β11 ), (D.4)
 Z (u) 0 0 0 0 
2
 0 f11 (u) 0 0 h25 (u) 
 
KII  ,
25 =  0 0 f11 (u) h34 (u) 0 
 0 0 2u
h43 (u) e f11 (−u) 0 
0 h52 (u) 0 0 2u
e f11 (−u)
β25 β52 = β34 β43 = (β11 + β22 )(β11 − β22 ). (D.5)
The corresponding diagonal solutions are also one-parameter solutions.
References
[2] V.E. Korepin, A.G. Izergin, N.M. Bogoliubov, Quantum Inverse Scattering Method and Correlation
[3] E. Abdalla, M.C.B. Abdalla, K. Rothe, Nonperturbative Methods in Two-Dimensional Quantum Field
Theory, 2 edn., World Scientific, Singapore, 2001.
[4] P.P. Kulish, N.Yu. Reshetikhin, J. Sov. Math. 23 (1983) 2435.
[5] M. Jimbo, Commun. Math. Phys. 102 (1986) 537.
[6] I.V. Cherednik, Theor. Math. Phys. 61 (1984) 977.
[8] L. Mezincescu, R.I. Nepomechie, Int. J. Mod. Phys. A 13 (1998) 2747.
[9] R.I. Nepomechie, Boundary quantum group generators of type A, hep-th/0204181.
[10] P. Bowcock, E. Corrigan, P.E. Dorey, R.H. Rietdijk, Nucl. Phys. B 445 (1995) 469.
[11] M.T. Batchelor, V. Fridkin, A. Kuniba, Y.K. Zhou, Phys. Lett. B 376 (1996) 266.
[12] A. Lima-Santos, Nucl. Phys. B 612 (2001) 446.
[13] A. Lima-Santos, Nucl. Phys. B 558 (1999) 637.
[14] I.V. Cherednik, Theor. Math. Phys. 43 (1980) 356.
[15] O. Babelon, H.J. de Vega, C.M. Viallet, Nucl. Phys. B 180 (1981) 542.
[16] N.Yu. Reshetikhin, M. Semenov-Tian-Shansky, Lett. Math. Phys. 19 (1990) 133.
[17] L. Mezincescu, R.I. Nepomechie, Int. J. Mod. Phys. A 7 (1992) 5657.
(1)
[18] G.M. Gandenberger, New non-diagonal solutions to the an boundary Yang–Baxter equation, hep-
th/9911178.
[19] G.W. Delius, N.J. Mackay, Quantum group symmetry in sine-Gordon and affine Toda field theories on the
half-line, hep-th/0112023.
[20] J. Abad, M. Rios, Phys. Lett. B 352 (1995) 92.
[21] H.J. de Vega, A. González-Ruiz, J. Phys. A 26 (1993) 519.
[22] S. Ghoshal, A.B. Zamolodchikov, Int. J. Mod. Phys. A 9 (2001) 3841.
CUMULATIVE AUTHOR INDEX B641–B644
Abdalla, E. B644 (2002) 201 De Forcrand, P. B642 (2002) 290

Agashe, K. B643 (2002) 172 Degrassi, G. B643 (2002) 79
Akhmedov, E.Kh. B643 (2002) 339 de la Ossa, X. B644 (2002) 170
Amico, L. B644 (2002) 409 Delgado, A. B643 (2002) 172
Arias, O. B643 (2002) 187 Dermíšek, R. B641 (2002) 327
Arutyunov, G. B643 (2002) 49 Di Bari, P. B643 (2002) 367
Asano, M. B644 (2002) 151 Diehl, M. B643 (2002) 431
Di Francesco, P. B641 (2002) 519
Bagnoud, M. B641 (2002) 61 Dijkgraaf, R. B644 (2002) 3
Bajnok, Z. B644 (2002) 509 Dijkgraaf, R. B644 (2002) 21
Bastianelli, F. B642 (2002) 372 Di Lorenzo, A. B644 (2002) 409
Beneke, M. B643 (2002) 431 Dittmaier, S. B642 (2002) 307
Doyon, B. B644 (2002) 451
Berglund, P. B641 (2002) 351
Bietenholz, W. B644 (2002) 223 Ellis, J. B643 (2002) 229
Bilal, A. B641 (2002) 61
Blanchard, P. B644 (2002) 495 Faraggi, A.E. B641 (2002) 93
Blumenhagen, R. B641 (2002) 235 Faraggi, A.E. B641 (2002) 111
Bouttier, J. B641 (2002) 519 Feldmann, Th. B643 (2002) 431
Brandhuber, A. B641 (2002) 351 Fjelstad, J. B641 (2002) 376
Brignole, A. B643 (2002) 79 Florea, B. B644 (2002) 170
Brodsky, S.J. B642 (2002) 344 Foerster, A. B642 (2002) 501
Buchmüller, W. B643 (2002) 367 Forte, S. B643 (2002) 477
Fortunato, S. B644 (2002) 495
Cao, J.-P. B644 (2002) 476 Fradkin, E. B642 (2002) 483
Cardenas, R. B643 (2002) 187
Gaillard, M.K. B643 (2002) 201
Carlevaro, L. B641 (2002) 61
Gandolfo, D. B644 (2002) 495
Casali, A. B644 (2002) 201 Garavuso, R. B641 (2002) 111
Chapovsky, A.P. B643 (2002) 431 Garland, L.W. B642 (2002) 227
Corley, S. B641 (2002) 131 Gehrmann, T. B642 (2002) 227
Cremades, D. B643 (2002) 93 Ghilencea, D.M. B641 (2002) 35
Cristofano, G. B641 (2002) 547 Giaquinta, G. B644 (2002) 409
Cuadros-Melgar, B. B644 (2002) 201 Giedt, J. B643 (2002) 201
Curio, G. B643 (2002) 131 Glover, E.W.N. B642 (2002) 227
Cvetič, M. B642 (2002) 139 Gluza, J. B642 (2002) 157
Cvetič, M. B644 (2002) 65 González, J. B642 (2002) 407
Czakon, M. B642 (2002) 157 Gracey, J.A. B644 (2002) 433
Groot Nibbelink, S. B641 (2002) 35
Dabholkar, A. B641 (2002) 223 Guan, X.-W. B642 (2002) 501
Das, P. B644 (2002) 395 Guitter, E. B641 (2002) 519
0550-3213/2002 Published by Elsevier Science B.V.

PII: S 0 5 5 0 - 3 2 1 3 ( 0 2 ) 0 0 8 9 0 - 8
586 Nuclear Physics B 644 (2002) 585–587
Hamberg, R. B644 (2002) 403 Lunardini, C. B643 (2002) 339

Hasenfratz, P. B643 (2002) 280 Lunin, O. B642 (2002) 91
Hatsuda, M. B644 (2002) 40 Lüst, D. B641 (2002) 235
Hauswirth, S. B643 (2002) 280
Hejczyk, J. B642 (2002) 157 Ma, E. B644 (2002) 290
Henkel, M. B641 (2002) 405 Magnea, L. B643 (2002) 477
Hikami, K. B644 (2002) 409 Mahanta, U. B644 (2002) 395
Hoffmann, L. B641 (2002) 188 Maiella, G. B641 (2002) 547
Holland, K. B643 (2002) 280 Maillet, J.M. B641 (2002) 487
Hwang, D.S. B642 (2002) 344 Maillet, J.M. B642 (2002) 433
Hwang, S. B641 (2002) 376 Maltoni, M. B643 (2002) 321
Månsson, T. B641 (2002) 376
Ibáñez, L.E. B643 (2002) 93 Marchesano, F. B643 (2002) 93
Igarashi, H. B644 (2002) 383 Marotta, V. B641 (2002) 547
Isidro, J.M. B641 (2002) 111 Mathur, S.D. B642 (2002) 91
Itoyama, H. B644 (2002) 248 Matsuura, T. B644 (2002) 403
Mesref, L. B641 (2002) 188
Jahn, O. B642 (2002) 357 Meziane, A. B641 (2002) 188
Jansen, K. B643 (2002) 517 Mueller, A.H. B643 (2002) 501
Janssen, B. B643 (2002) 399
Jegerlehner, F. B641 (2002) 285 Nair, V.P. B641 (2002) 533
Jejjala, V. B642 (2002) 483 Nakamura, S. B644 (2002) 248
Jörg, T. B643 (2002) 280 Nakatsu, T. B642 (2002) 13
Nandi, S. B641 (2002) 327
Kalmykov, M.Yu. B641 (2002) 285 Niccoli, G. B641 (2002) 547
Kamimura, K. B644 (2002) 40 Niedermayer, F. B643 (2002) 280
Karabali, D. B641 (2002) 533 Notari, A. B644 (2002) 371
Kawamoto, N. B644 (2002) 533
Ohlsson, T. B643 (2002) 247
Khveshchenko, D.V. B642 (2002) 515
Okuda, T. B641 (2002) 393
Kiem, Y. B641 (2002) 256
Okuyama, K. B644 (2002) 383
Kiem, Y. B642 (2002) 389
Ooguri, H. B641 (2002) 3
Kim, N. B643 (2002) 31
Osterloh, A. B644 (2002) 409
Kim, S.-S. B641 (2002) 256
Ott, T. B641 (2002) 235
Kim, Y. B642 (2002) 389
Kitanine, N. B641 (2002) 487 Palla, L. B644 (2002) 509
Kitanine, N. B642 (2002) 433 Palumbo, F. B643 (2002) 391
Kitazawa, Y. B642 (2002) 210 Park, J. B642 (2002) 389
Kogut, J.B. B642 (2002) 181 Parvizi, S. B641 (2002) 223
Körs, B. B641 (2002) 235 Pawlowski, J.M. B642 (2002) 357
Koukoutsakis, A. B642 (2002) 227 Penati, S. B643 (2002) 49
Krause, A. B643 (2002) 131 Petkou, A.C. B643 (2002) 49
Kristjansen, C. B643 (2002) 3 Philipsen, O. B642 (2002) 290
Piccione, A. B643 (2002) 477
Langacker, P. B642 (2002) 139 Pilaftsis, A. B644 (2002) 263
Latorre, J.J. B643 (2002) 477 Plefka, J. B643 (2002) 3
Lee, S. B642 (2002) 389 Plefka, J. B643 (2002) 31
Leigh, R.G. B642 (2002) 483 Plümacher, M. B643 (2002) 367
Levman, G. B642 (2002) 3 Ponsot, B. B642 (2002) 114
Lima-Santos, A. B644 (2002) 568 Pope, C.N. B644 (2002) 65
Links, J. B642 (2002) 501
Lozano, Y. B643 (2002) 399 Quiros, I. B643 (2002) 187
Lü, H. B642 (2002) 173
Lü, H. B644 (2002) 65 Raby, S. B641 (2002) 327
Lukyanov, S. B644 (2002) 451 Raidal, M. B643 (2002) 229
Nuclear Physics B 644 (2002) 585–587 587
Ramgoolam, S. B641 (2002) 131 Takács, G. B642 (2002) 456

Remiddi, E. B642 (2002) 227 Takács, G. B644 (2002) 509
Rey, S.-J. B641 (2002) 256 Terras, V. B641 (2002) 487
Riotto, A. B644 (2002) 371 Terras, V. B642 (2002) 433
Roth, M. B642 (2002) 307 Tórtola, M.A. B643 (2002) 321
Roy, D.P. B644 (2002) 290 Toublan, D. B642 (2002) 181
Rühl, W. B641 (2002) 188
Vafa, C. B641 (2002) 3
Sakaguchi, M. B644 (2002) 40
Vafa, C. B644 (2002) 3
Santambrogio, A. B643 (2002) 49
Sato, H.-T. B641 (2002) 256 Vafa, C. B644 (2002) 21
Schmidt, I. B642 (2002) 344 Valle, J.W.F. B643 (2002) 321
Schwetz, T. B643 (2002) 321 van Neerven, W.L. B644 (2002) 403
Seidl, G. B643 (2002) 247 Veretin, O. B641 (2002) 285
Sekino, Y. B644 (2002) 151 Vives, O. B641 (2002) 93
Semenoff, G.W. B643 (2002) 3
Sezgin, E. B644 (2002) 303 Wang, Y.-P. B644 (2002) 476
Shiu, G. B642 (2002) 139 Watts, G.M.T. B642 (2002) 456
Sinclair, D.K. B642 (2002) 181 Wei, Z.-T. B642 (2002) 263
Skarke, H. B644 (2002) 170 Wosiek, J. B644 (2002) 85
Slavich, P. B643 (2002) 79
Slavnov, N.A. B641 (2002) 487
Yang, M.-Z. B642 (2002) 263
Slavnov, N.A. B642 (2002) 433
Yoshida, K. B644 (2002) 113
Smirnov, A.Yu. B643 (2002) 339
Yoshida, K. B644 (2002) 128
Sokatchev, E. B643 (2002) 49
Yotsuji, K. B644 (2002) 533
Sommer, R. B643 (2002) 517
Yue, R.-H. B644 (2002) 476
Staudacher, M. B643 (2002) 3
Sugiyama, K. B644 (2002) 113
Sugiyama, K. B644 (2002) 128 Zarembo, K. B643 (2002) 157
Sundell, P. B644 (2002) 303 Zhou, H.-Q. B642 (2002) 501
Sundrum, R. B643 (2002) 172 Zirotti, A. B642 (2002) 372
Suyama, T. B641 (2002) 341 Zupnik, B. B644 (2002) 405
Suzuki, H. B644 (2002) 383 Zwirner, F. B643 (2002) 79

Nucl - Phys.B v.644

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Nucl - Phys.B v.644

Uploaded by

Copyright:

Available Formats

Nuclear Physics B 644 (2002) 3–20

Matrix models, topological strings, and

Received 15 August 2002; accepted 23 August 2002

E-mail address: rhd@science.uva.nl (R. Dijkgraaf).

0550-3213/02/$ – see front matter  2002 Published by Elsevier Science B.V.

2. Large-N topological string conjectures

In this section we formulate a general class of large-N conjectures involving B-model

2.1. Geometry of the generalized conifold transition

φ0 = φ0 , z φ1 = zφ1 + W (φ0 ), z = 1/z. (2.2)

To relate it to the geometry given by (2.1) we make the identifications

uv − 4ω2 + 4ωW (x) = 0. (2.4)

uv + y 2 + W 2 (x) + f (x) = 0, (2.5)

y 2 + W (x)2 + f (x) = 0. (2.7)

2.2. Lifting to topological string dualities

with Y a 3-chain connecting P1 and C.

2.3. Operator formalism

2.4. Precise conjecture

FW,f (gs ) = FW,Ni (gs )

1 We thank E. Witten for a useful discussion on this point.

3. Matrix models and special geometry

3.1. Matrix integrals in the planar limit

We consider the one-matrix integral over N × N Hermitean matrices Φ

In this way we end up with a collection of N variables λ1 , . . . , λN in a potential W (λ).

the equation of motion satisfied by the eigenvalues is

3.2. The spectral curve

y 2 − W (x)2 + f (x) = 0. (3.7)

3.3. Filling fractions and periods

Here the Ai are the cuts in the complex x-plane.

and the deformation as

then by plugging definition (3.4) in the matrix integral we obtain

3.4. Higher genus

3.5. Domain walls and eigenvalue tunneling

4. Double scaling limits and other further generalizations

for some coefficients cg , in perfect accord with the expected answer.

4.1. Final remarks

[34] R. Dijkgraaf, C. Vafa, in preparation.

On geometry and matrix models

Received 15 August 2002; accepted 23 August 2002

In [1] we have shown how the effective superpotential in N = 1 supersymmetric

E-mail address: rhd@science.uva.nl (R. Dijkgraaf).

0550-3213/02/$ – see front matter  2002 Published by Elsevier Science B.V.

y 2 − W (x)2 + µf (x) = 0, (1.3)

(This equation can be considered as analogous to the Bohr–Sommerfeld quantization

2. Unitary matrix models and the Seiberg–Witten solution

We have reviewed how the effective superpotential in N = 1 supersymmetric gauge

2.1. Geometrical engineering of N = 2 theories

2.2. Unitary matrix models

y 2 − W (x)2 + 4µf (x) = 0. (2.4)

2.3. The Gross–Witten model

y 2 − , 2 sin2 x + 4µ,(cos x + u) = 0. (2.7)

where A is a one-cycle homologous to A1 − A2 . This equation describes the exact relation

2.4. Double scaling limit

2.5. Generalizations for general groups

engineering in [17]. This will give rise to a curve of the form

2.6. Connections with A-model topological strings

It is natural to ask if there is a connection with A-model topological strings, and in

3. Quiver matrix models

3.1. N = 1 quiver gauge theories and topological strings

For such a gauge theory we can write a general tree-level superpotential

where Φi (with i = 1, . . . , r) is an Ni × Ni Hermitian matrix and for every linked indices

3.2. Saddle points and dual CY geometry

then the associated condition reads

The the associated rational planar curves

(i) giving the correct three-form, d B[2] = H[3] for dH[3] = 0,

[2] = i +I −I Γ+ (τ1 )I J Π L