a

r
X
i
v
:
m
a
t
h
/
0
6
1
2
2
5
5
v
5


[
m
a
t
h
.
Q
A
]


4

F
e
b

2
0
0
9
Cardy condition for open-closed eld
algebras
Liang Kong
Abstract
Let V be a vertex operator algebra satisfying certain reductivity and niteness
conditions such that (
V
, the category of V -modules, is a modular tensor category.
We study open-closed eld algebras over V equipped with nondegenerate invariant
bilinear forms for both open and closed sectors. We show that they give algebras
over certain C-extension of the so-called Swiss-cheese partial dioperad, and we can
obtain Ishibashi states easily in such algebras. Cardy condition can be formulated
as an additional condition on such open-closed eld algebras in terms of the action
of the modular transformation S :
1

on the space of intertwining operators

of V . We then derive a graphical representation of S in the modular tensor
category (
V
. This result enables us to give a categorical formulation of the Cardy
condition and the modular invariance condition for 1-point correlation functions
on torus. Then we incorporate these two conditions and the axioms of open-closed
eld algebra over V equipped with nondegenerate invariant bilinear forms into a
tensor-categorical notion called Cardy (
V
[(
V V
-algebra. In the end, we give a
categorical construction of Cardy (
V
[(
V V
-algebra in Cardy case.
Contents
1 Introduction 2
2 Partial dioperads 7
2.1 Partial dioperads . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7
2.2 Conformal full eld algebras . . . . . . . . . . . . . . . . . . . . . . . . . 11
2.3 Open-string vertex operator algebras . . . . . . . . . . . . . . . . . . . . 13
3 Swiss-cheese partial dioperad 15
3.1 2-colored (partial) dioperads . . . . . . . . . . . . . . . . . . . . . . . . . 15
3.2 Swiss-cheese partial dioperads . . . . . . . . . . . . . . . . . . . . . . . . 18
3.3 Open-closed eld algebras over V . . . . . . . . . . . . . . . . . . . . . . 20
3.4 Ishibashi states . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25
4 Cardy condition 28
4.1 The rst version . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 28
4.2 The second version . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 34
1
5 Modular tensor categories 41
5.1 Preliminaries . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 42
5.2 Graphical representation of S :
1

. . . . . . . . . . . . . . . . . . . 46
6 Categorical formulations and constructions 53
6.1 Modular invariant (
V
L
V
R-algebras . . . . . . . . . . . . . . . . . . . . . 53
6.2 Cardy (
V
[(
V V
-algebras . . . . . . . . . . . . . . . . . . . . . . . . . . . 56
6.3 Constructions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 61
A The Proof of Lemma 5.30 64
1 Introduction
This work is a continuation of the works [HKo1]-[HKo3][Ko1][Ko2] and a part of an
open-string extension of a program on the closed conformal eld theory via the theory of
vertex operator algebra. This program was initiated by I. Frenkel and largely developed
by Huang [H1]-[H11]. Zhus work [Z] is also very inuential in this development.
Segal dened the (closed) conformal eld theory [Se1] as a projective monoidal func-
tor from the category of nite ordered sets with morphisms being the conformal equiva-
lent classes of Riemann surfaces with parametrized boundaries to the category of locally
convex complete topological vector spaces. This denition is very dicult to work with
directly. Taking advantage of the theory of vertex operator algebra, Huang suggested
to rst construct all necessary structure on a densed subspace of the relevant complete
topological vector space [H3], then complete it properly later [H5][H6]. This idea guided
us in all our previous works [HKo1]-[HKo3][Ko1][Ko2], in particular in our formulation
of the notion of algebra over the Swiss-cheese partial operad which catches only some
genus-zero information of the whole structure on the densed subspace. This seemingly
temporary structure on the densed subspace does not necessarily follow from that on
the complete topological vector space, thus has its own independent values and is worth-
while to be formulated properly and extended to a theory of all genus. We will call such
a theory on the densed subspace as (closed) partial conformal eld theory.
More precisely, a partial conformal eld theory is a projective monoidal functor
T : 1o (1 between two partial categories 1o, (1 in which the compositions of
morphisms are not always well-dened. 1o is the category of nite ordered sets, and
Mor
RS
(S
1
, S
2
) for any pair of such sets S
1
, S
2
is the set of the conformal equivalent classes
of closed Riemann surfaces with [S
1
[ positively oriented punctures and local coordinates
and [S
2
[ negatively oriented punctures and local coordinates [H3], and the compositions
of morphisms in 1o are given by the sewing operations on oppositely oriented punctures
([H3]). Such sewing operations are only partially dened. (1 is the category of graded
vector spaces with nite dimensional homogeneous spaces and a weak topology induced
from the restricted dual spaces. For any pair of A, B Ob((1), Mor
GV
(A, B) is the
set of continuous linear maps from A to B :=

nG
B
B
(n)
, where the abelian group
G
B
gives the grading on B. For any pair of morphisms A
g
B and B
f
C in (1,
using the projector P
n
: B B
(n)
, we dene f g(u) :=

nG
B
f(P
n
g(u)), u A,
and f g is well-dened only when the sum is absolutely convergent for all u A. We
2
remark that by replacing the surfaces with parametrized boundaries in Segals denition
by the surfaces with oriented punctures and local coordinates in 1o we have enlarged
the morphism sets. Thus a Segals functor may not be extendable to a functor on 1o.
The above denition can be easily extended to include open strings by adding to
the morphism sets of 1o the conformal equivalent classes of Riemann surfaces with
(unparametrized!) boundaries and both oriented interior punctures and oriented punc-
tures on the boundaries. We will call this open-string extended theory as open-closed
partial conformal eld theory. We denote the graded vector spaces associated to interior
punctures (closed strings at innity) and boundary punctures (open strings at innity)
by V
cl
and V
op
respectively.
The sets of all genus-zero closed surfaces (spheres) with arbitrary number of posi-
tively and negatively oriented punctures form a structure of sphere partial dioperad K
(see the denition in Section 1.1). It includes as a substructure the sphere partial operad
[H3], which only includes spheres with a single negatively oriented puncture. Sphere par-
tial dioperad allows all sewing operations as long as the surfaces after sewing are still
genus-zero. Hence to construct genus-zero partial conformal eld theory amounts to
construct projective K-algebras or algebras over certain extension of K. In [HKo2], we
introduced the notion of conformal full eld algebra over V
L
V
R
equipped with a non-
degenerate invariant bilinear form, where V
L
and V
R
are two vertex operator algebras
of central charge c
L
and c
R
respectively and satisfy certain niteness and reductivity
conditions. Theorem 2.7 in [Ko1] can be reformulated as follow: a conformal full eld
algebra V
cl
over V
L
V
R
equipped with a nondegenerate invariant bilinear form canon-
ically gives on V
cl
an algebra over

K
c
L

K
c
R
, which is a partial dioperad extension of
K.
The sets of all genus-zero surfaces with one (unparametrized!) boundary component
(disks) and arbitrary number of oriented boundary punctures form the so-called disk
partial dioperad denoted by D (see the denition in Section 1.1). T restricted on D
induces a structure of algebra over certain extension of D on V
op
. In [HKo1], Huang and
I introduced the notion of open-string vertex operator algebra. We will show in Section
1.3 that an open-string vertex operator algebra of central charge c equipped with a
nondegenerate invariant bilinear form canonically gives an algebra over

D
c
, which is a
partial dioperad extension of D.
The sets of all genus-zero surfaces with only one (unparametrized!) boundary com-
ponent and arbitrary number of oriented interior punctures and boundary punctures
form the so-called Swiss-Cheese partial dioperad S (see the denition in Section 2.2).
A typical elements in S is depicted in Figure 1, where boundary punctures are drawn
as an innitely long strip (or an open string) and interior punctures are drawn as an
innitely long tube (or a closed string). Let V be a vertex operator algebra of central
charge c satisfying the conditions in Theorem 1.1. We will show in Section 2.3 that an
open-closed eld algebra over V , which contains an open-string vertex operator algebra
V
op
and a conformal full eld algebra V
cl
, satisfying a V -invariant boundary condition
[Ko2] and equipped with nondegenerate invariant bilinear forms on both V
op
and V
cl
,
canonically gives an algebra over

S
c
, which is an extension of S.
Note that all surfaces of any genus with arbitrary number of boundary components,
interior punctures and boundary punctures can be obtained by applying sewing oper-
3
Figure 1: a typical element in Swiss-Cheese partial dioperad S
ations to elements in S. Therefore, except some compatibility conditions coming from
dierent decompositions of the same surfaces with higher genus, an algebra over

S
c
if
extendable uniquely determine the entire theory of all genus. In particular, the famous
Ishibashi states [I] can be obtained in an open-closed eld algebra equipped with nonde-
generate invariant bilinear forms. An Ishibashi state is a coherent state in V
cl
such that
(L
L
(n) L
R
(n)) = 0, n Z. In physics, Ishibashi states are obtained by solving
above equation. In Section 2.4, we show how to obtain Ishibashi states constructively
and geometrically from vacuum-like states in V
op
.
Other surfaces which are not included in S only provide additional compatibility con-
ditions. In 2-d topological eld theories, only three additional compatibility conditions
are needed to ensure the consistency of a theory of all genus [La][Mo1][Se2][MSeg][AN][LP].
The rst compatibility condition says that both V
op
and V
cl
are nite dimensional. This
guarantee the convergence of all higher genus correlation functions. The second condi-
tion is the modular invariance condition for 1-point correlation functions on torus. It
is due to two dierent decompositions of the same torus as depicted in Figure 2. This
condition is automatically satised in 2-d topological eld theories and but nontrivial
in conformal eld theories. The third condition is the famous Cardy condition which is
again due to two dierent decompositions of a single surface as shown in Figure 3.
Now we turn to the compatibility conditions in conformal eld theory. In this case,
both V
cl
and V
op
in any nontrivial theory are innite dimensional. We need require the
convergence of all correlation functions of all genus. This is a highly nontrivial condition
and not easy to check for examples. So far the only known convergence results are in
genus-zero [H7] and genus-one theories [Z][DLM][Mi1][Mi2][H9]. We recall a theorem by
Huang.
Theorem 1.1 ([H10][H11]). If (V, Y, 1, ) is a simple vertex operator algebra V satis-
4
Figure 2: modular invariance condition for 1-pt correlation functions on torus
=
Figure 3: Cardy condition
fying the following conditions: 1. V is C
2
-conite; 2. V
n
= 0 for n < 0, V
(0)
= C1, V

is isomorphic to V as V -module; 3. all N-gradable weak V -modules are completely re-
ducible, then the direct sum of all inequivalent irreducible V -modules has a natural struc-
ture of intertwining operator algebra [H4], and the category (
V
of V -modules has a struc-
ture of vertex tensor category [HL1]-[HL4][H2] and modular tensor category [RT][T].
For the intertwining operator algebra given in Theorem 1.1, Huang also proved in
[H7][H9] that the products of intertwining operators and q-traces of them have certain
nice convergence and analytic extension properties. These properties are sucient for
the construction of genus-zero and genus-one correlation functions in closed partial con-
formal eld theory [HKo2][HKo3]. Since the modular tensor category (
V
supports an
action of mapping class groups of all genus, it is reasonable to believe that the above
conditions on V are also sucient to guarantee the convergence of correlation functions
of all genus.
Assumption 1.2. In this work, we x a vertex operator algebra V with central charge c,
which is assumed to satisfy the conditions in Theorem 1.1 without further announcement.
Besides the convergence condition and the axioms of projective K-algebra, Sonoda
[So] argued on a physical level of rigor that it is sucient to check the modular invariance
condition in Figure 2 in order to have a consistent partial conformal eld theory of all
genus. This modular invariance condition was studied in [HKo3]. In the framework of
conformal full eld algebra over V
L
V
R
, it was formulated algebraically as a modular
invariance property of an intertwining operator of V
L
V
R
.
For an open-closed theory, besides the convergence condition, the axioms of algebra
over

S
c
and the modular invariance condition, Lewellen [Le] argued on a physical level of
rigor that the only remaining compatibility condition one needs is the Cardy condition.
Cardy condition in open-closed conformal eld theory is more complicated than that
in topological theory and has never been fully written down by physicists. In Section
3.1, we derive the Cardy condition from the axioms of open-closed partial conformal
eld theory by writing out two sides of the Cardy condition in Figure 3 explicitly in
5
terms of the ingredients of open-closed eld algebra (see Denition 3.4). Using results in
[H9][H10], we can show that the Cardy condition can be reformulated as an invariance
condition of the modular transformation S :
1

on intertwining operators.
There are still more compatibility conditions which were not discussed in [So][Le].
One also need to proved certain algebraic version of uniformization theorems (see [H1]
[H3] for the genus-zero case). Such results for genus large than 0 are still not avail-
able. But it seems that no additional assumption on V is needed. They should follow
automatically from the properties of Virasoro algebra and intertwining operators. This
uniformization problem and convergence problems are not pursued further in this work.
In order to take the advantages of some powerful tools, such as the graphic calculus
in tensor category, in the study of the modular invariance condition and the Cardy con-
dition, we would like to obtain categorical formulations of these conditions. It requires
us to know the action of the modular transformation S in (
V
. Although the action of
SL(2, Z) in a modular tensor category is explicitly known [MSei3][V][Ly][Ki][BK2], its
relation to the modular transformation of the q-trace of the products of intertwining
operators of V is not completely clear. This relation was rst suggested by I. Frenkel
and studied by Moore and Seiberg in [MSei3] but only on a physical level of rigor.
Using Huangs results on modular tensor category [H10][H11], we derive a graphical
representation of S in (
V
. This result enables us to give categorical formulations of the
modular invariance condition and the Cardy condition. We incorporate them with the
categorical formulation of open-closed eld algebra over V equipped with nondegenerate
invariant bilinear forms into a tensor-categorical notion called Cardy (
V
[(
V V
-algebra.
As we discussed in previous paragraphs, it is reasonable to believe that open-closed par-
tial conformal eld theories of all genus satisfying the V -invariant boundary condition
[Ko2] are classied by Cardy (
V
[(
V V
-algebras. However, to construct the high-genus
theories explicitly is still a hard open problem which is not pursued in this work. In the
end, we give an explicit construction of Cardy (
V
[(
V V
-algebra in the so-called Cardy
case in physics literature (see for example [FFRS]).
Note that this work is somewhat complementary to the works of Fjelstad, Fuchs,
Runkel, Schweigert [FS][FRS1]-[FRS4][FjFRS1][FjFRS2]. We will leave a detailed study
of the relationship between these two approaches in [KR].
The layout of this work is as follow: in Section 1, we introduce the notion of sphere
partial dioperad and disk partial dioperad and study algebras over them; in Section 2,
we introduce the notions of Swiss-cheese partial dioperad S and its C-extension

S
c
, and
show that an open-closed conformal eld algebra over V equipped with nondegenerate
invariant bilinear forms canonically gives an algebra over

S
c
, and we also construct
Ishibashi states in such algebras; in Section 3, we give two formulations of the Cardy
condition; in Section 4, we derive a graphic representation of the modular transformation
S; in Section 5, we give the categorical formulations of the nondegenerate invariant
bilinear forms, the modular invariance condition and the Cardy condition. Then we
introduce the notion of Cardy (
V
[(
V V
-algebra and give a construction.
Convention of notations: N, Z, Z
+
, R, R
+
, C denote the set of natural numbers, inte-
gers, positive integers, real numbers, positive real numbers, complex numbers, respec-
tively. Let H = z C[Imz > 0 and H = z C[Imz < 0. Let

R,

C and

H be the
one point compactication of real line, complex plane and up-half plane (including the
6
R-boundary) respectively. Let R
+
and C

be the multiplication groups of positive real

and nonzero complex numbers respectively. The ground eld is always chosen to be C.
Throughout this work, we choose a branch cut for logarithm as follow:
log z = log [z[ + iArg z, 0 Arg z < 2. (1.1)
We dene power functions of two dierent types of complex variables as follow:
z
s
:= e
s log z
, z
s
:= e
s log z
, s R. (1.2)
Acknowledgment The results in Section 2.4 and 3.1 are included in authors thesis.
I want to thank my advisor Yi-Zhi Huang for introducing me to this interesting eld
and for his constant support and many important suggestions for improvement. I thank
C. Schweigert for telling me the meaning of boundary states from a physical point of
view. I also want to thank I. Frenkel, J. Fuchs, A. Kirillov, Jr. and C. Schweigert for
some inspiring conversations on the subject of Section 4.2.
2 Partial dioperads
In Section 1.1, we recall the denition of (partial) dioperad and algebra over it, and
introduce sphere partial dioperad K, disk partial dioperad D and their extensions

K
c
L

K
c
R
,

D
c
as examples. In Section 1.2, we discuss an algebra over

K
c
L

K
c
R
from a
conformal full eld algebra over V
L
V
R
. In Section 1.3, we discuss an algebra over

D
c
from an open-string vertex operator algebra.
2.1 Partial dioperads
Let us rst recall the denition of dioperad given by Gan [G]. Let S
n
be the automor-
phism group of the set 1, . . . , n for n Z
+
. Let m = m
1
+ + m
n
be an ordered
partition and S
n
. The block permutation
(m
1
,...,mn)
S
m
is the permutation which
permutes n intervals of lengths m
1
, . . . , m
n
in the same way as permutes 1, . . . , n. Let

i
S
m
i
, i = 1, . . . , k. We view the element (
1
, . . . ,
k
) S
m
1
S
m
k
naturally as
an element in S
m
by the canonical embedding S
m
1
S
m
k
S
m
. For any S
n
and 1 i n, we dene a map

i : 1, . . . , n 1 1, . . . , n by

i(j) = j if j < i and

i(j) = j + 1 if j i and an element

i() S
n1
by

i()(j) :=

i
1

1
(i)(j).
Denition 2.1. A dioperad consists of a family of sets T(m, n)
m,nN
with an action
of S
m
S
n
on T(m, n) for each pair of m, n Z
+
, a distinguished element I
P
T(1, 1)
and substitution maps
T(m, n) T(k
1
, l
1
) T(k
n
, l
n
)

(i
1
,...,in)
T(mn + k
1
+ k
n
, l
1
+ + l
n
)
(P, P
1
, . . . , P
n
)
(i
1
,...,in)
(P; P
1
, . . . , P
n
) (2.1)
for m, n, l
1
, . . . , l
n
N, k
1
, . . . , k
n
Z
+
and 1 i
j
k
j
, j = 1, . . . , n, satisfying the
following axioms:
7
1. Unit properties: for P T(m, n),
(a) left unit property:
(i)
(I
P
; P) = P for 1 i m,
(b) right unit property:
(1,...,1)
(P; I
P
, . . . , I
P
) = P.
2. Associativity: for P T(m, n), Q
i
T(k
i
, l
i
), i = 1, . . . , n, R
j
T(s
j
, t
j
), j =
1, . . . , l = l
1
+ + l
n
, we have

(q
1
,...,q
l
)
_

(p
1
,...,pn)
(P; Q
1
, . . . , Q
n
); R
1
, . . . , R
l
_
=
(p
1
,...,pn)
(P; P
1
, . . . , P
n
) (2.2)
where P
i
=
(q
l
1
++l
i1
+1
,...,q
l
1
++l
i
)
(Q
i
; R
l
1
++l
i1
+1
, . . . , R
l
1
++l
i
) for i = 1, . . . , n.
3. Permutation property: For P T(m, n), Q
i
P(k
i
, l
i
), i = 1, . . . , n, (, )
S
m
S
n
, (
i
,
i
) S
k
i
S
l
i
, i = 1, . . . , n,

(i
1
,...,in)
((, )(P); Q
1
, . . . , Q
n
)
= ((,
(k
1
1,...,kn1)
),
(l
1
,...,ln)
)
(i
(1)
,...,i
(n)
)
(P; Q
(1)
, . . . , Q
(n)
),

(i
1
,...,in)
(P; (
1
,
1
)(Q
1
), . . . , (
n
,
n
)(Q
n
))
= ((id,

i
1
(
1
), . . . ,

i
n
(
n
)), (
1
, . . . ,
n
))
(
1
1
(i
1
),...,
1
n
(in))
(P; Q
1
, . . . , Q
n
).
We denote such dioperad as (T,
P
, I
P
) or simply T.
Remark 2.2. We dene compositions i j as follow:
Pi j Q :=
(1,...,1,j,1,...1)
(P; I
P
, . . . , I
P
, Q, I
P
, . . . , I
P
).
It is easy to see that
(i
1
,...,in)
can be reobtained from i j . In [G], the denition of
dioperad is given in terms of i j instead of
(i
1
,...,in)
.
Denition 2.3. A partial dioperad has a similar denition as that of dioperad except
the map
(i
1
,...,in)
or i j is only partially dened and the same associativity hold whenever
both sides of (2.2) exist. A (partial) nonassociative dioperad consists of the same data
as those of (partial) dioperad satisfying all the axioms of (partial) dioperad except the
associativity.
The notion of homomorphism and isomorphism of (partial pseudo-) dioperad are
naturally dened.
Remark 2.4. In the case of partial dioperad, the denition using
(i
1
,...,in)
or i j may
have subtle dierences in the domains on which
(i
1
,...,in)
or i j is dened (see appendix
C in [H2] for more details). These dierences have no eect on those algebras over
partial dioperads considered in this work. So we will simply ignore these dierences.
Remark 2.5. Notice that a (partial nonassociative) dioperad T(m, n)
m,nN
naturally
contains a (partial nonassociative) operad T(1, n)
nN
as a substructure.
Denition 2.6. A subset G of T(1, 1) is called a rescaling group of T if the following
conditions are satised:
8
1. For any g, g
1
, . . . , g
n
G, Q P(m, n),
(i)
(g; Q) and
(1,...,1)
(Q; g
1
, . . . , g
n
) are
always well-dened for 1 i m.
2. I
P
G and G together with the identity element I
P
and multiplication map

(1)
: GG G is a group.
Denition 2.7. A G-rescalable partial dioperad is partial dioperad T such that for any
P T(m, n), Q
i
T(k
i
, l
i
), i = 1, . . . , n there exist g
i
G, i = 1, . . . , n such that

(i
1
,...,in)
_
P;
(i
1
)
(g
1
; Q
1
), . . . ,
(in)
(g
n
; Q
n
)
_
is well-dened.
The rst example of partial dioperad in which we are interested in this work comes
from K = K(n

, n
+
)
n

,n
+
N
[Ko2], a natural extension of sphere partial operad K
[H3]. More precisely, K(n

, n
+
) is the set of the conformal equivalent classes of sphere
with n

(n
+
) ordered negatively (positively) oriented punctures and local coordinate
map around each puncture. In particular, K(0, 0) is an one-element set consisting of the
conformal equivalent class of a sphere with no additional structure. We simply denote
this element as

C. We use
Q = ( (z
1
; a
(1)
0
, A
(1)
), . . . , (z
n

; a
(n

)
0
, A
(n

)
) [
(z
1
; a
(1)
0
, A
(1)
) . . . (z
n
+
; a
(n
+
)
0
, A
(n
+
)
) )
K
, (2.3)
where z
i


C, a
(i)
0
C

, A
(i)
C

for i = n

, . . . , 1, 1, . . . , n
+
, to denote a sphere

C with positively (negatively) oriented punctures at z

i


C for i = 1, . . . , n
+
(i =
1, . . . , n

), and with local coordinate map f

i
around each punctures z
i
given by:
f
i
(w) = e
P

j=1
A
(i)
j
x
j+1 d
dx
(a
(i)
0
)
x
d
dx
x

x=wz
i
if z
i
C, (2.4)
= e
P

j=1
A
(i)
j
x
j+1 d
dx
(a
(i)
0
)
x
d
dx
x

x=
1
w
if z
i
= . (2.5)
We introduce a useful notation

Q dened as follow:

Q = ( ( z
1
; a
(1)
0
, A
(1)
), . . . , ( z
n

; a
(n

)
0
, A
(n

)
)[
( z
1
; a
(1)
0
, A
(1)
) . . . ( z
n
+
; a
(n
+
)
0
, A
(n
+
)
) )
K
, (2.6)
where the overline represents complex conjugations.
We denote the set of all such Q as T
K
(n

, n
+
). Let T
K
:= T
K
(n

, n
+
)
n

,n
+
N
.
There is an action of SL(2, C) on T
K
(n

, n
+
) as Mobius transformations. It is clear that
K(n

, n
+
) = T
K
(n

, n
+
)/SL(2, C). (2.7)
We denote the quotient map T
K
K as
K
. The identity I
K
K(1, 1) is given by
I
K
=
K
( (, 1, 0)[(0, 1, 0) ) (2.8)
where 0 = (0, 0, . . . )

n=1
C. The composition i j is provided by the sewing operation
i j [H3]. In particular, for n
1
, m
2
1, P K(m
1
, n
1
) and Q K(m
2
, n
2
), Pi j Q is
9
the sphere with punctures obtained by sewing the i-th positively oriented puncture of
P with the j-th negatively oriented puncture of Q. The S
n

S
n
+
-action on K(n

, n
+
)
(or T (n

, n
+
)) is the natural one. Moreover, the set
( (, 1, 0)[(0, a, 0) )[a C

(2.9)
together with multiplication 11 is a group which can be canonically identied with
group C

. It is clear that K is a C

-rescalable partial dioperad. We call it sphere

partial dioperad.
The C-extensions of K, such as K
c
and

K
c
L

K
c
R
for c, c
L
, c
R
C, are trivial line
bundles over K with natural C

-rescalable partial dioperad structures (see Section 6.8

in [H2]). Moreover, we denote the canonical section K

K
c
L

K
c
R
as
K
.
The next example of partial dioperad is D = D(n

, n
+
)
n

,n
+
N
, which is an ex-
tension of the partial operad of disk with strips introduced [HKo1]. More precisely,
D(n

, n
+
) is the set of conformal equivalent classes of disks with ordered punctures on
their boundaries and local coordinate map around each puncture. In particular, D(0, 0)
is an one-element set consisting of the conformal equivalent class of a disk with no
additional structure. We simply denote this element as

H. We use
Q = ( (r
n

; b
(n

)
0
, B
(n

)
), . . . , (r
1
; b
(1)
0
, B
(1)
)[
(r
1
; b
(1)
0
, B
(1)
) . . . (r
n
+
; b
(n
+
)
0
, B
(n
+
)
) )
D
, (2.10)
where r
i


R, a
(i)
0
R
+
, B
(i)
R

for i = n

, . . . , 1, 1, . . . , n
+
, to denote a disk

H with positively (negatively) oriented punctures at r

i


R for i = 1, . . . , n
+
(i =
1, . . . , n

), and with local coordinate map g

i
around each punctures r
i
given by:
g
i
(w) = e
P

j=1
B
(i)
j
x
j+1 d
dx
(b
(i)
0
)
x
d
dx
x

x=wr
i
if r
i
R, (2.11)
= e
P

j=1
B
(i)
j
x
j+1 d
dx
(b
(i)
0
)
x
d
dx
x

x=
1
w
if r
i
= . (2.12)
We denote the set of all such Q as T
D
(n

, n
+
). Let T
D
:= T
D
(n

, n
+
)
n

,n
+
N
. The
automorphism group of H, SL(2, R), naturally acts on T
D
(n

, n
+
). It is clear that
D(n

, n
+
) = T
K
(n

, n
+
)/SL(2, R). (2.13)
We denote the quotient map as
D
. The identity I
D
D(1, 1) is given by
I
D
=
D
( (, 1, 0)[(0, 1, 0) ). (2.14)
The composition i j is provided by the sewing operation i j [HKo1]. In particular,
for n
1
, m
2
1, P D(m

, m
+
) and Q D(n

, n
+
), Pi j Q is the disk with strips
obtained by sewing the i-th positively oriented puncture of P with the j-th negatively
oriented puncture of Q. The S
n

S
n
+
-action on D(n

, n
+
) (or T (n

, n
+
)) is the natural
one. The set
( (, 1, 0)[(0, a, 0) )[a R
+
(2.15)
10
together with multiplication 11 is a group which can be canonically identied with
group R
+
. It is clear that D is a R
+
-rescalable partial dioperad. We call it disk partial
dioperad.
D can be naturally embedded to K as a sub-dioperad. The C-extension D
c
of D for
c C is just the restriction of the line bundle K
c
on D. D
c
is also a R
+
-rescalable partial
dioperad and a partial sub-dioperad of K
c
. We denote the canonical section on D D
c
as
D
.
Now we discuss an example of partial nonassociative dioperad which is important
for us. Let U =
nJ
U
(n)
be a graded vector space and J an index set. We denote the
projection U U
(n)
as P
n
. Now we consider a family of spaces of multi-linear maps
E
U
= E
U
(m, n)
m,nN
, where
E
U
(m, n) := Hom
C
(U
m
, U
n
). (2.16)
For f E
U
(m, n), g
j
E
U
(k
j
, l
j
) and u
(j)
p
j
U, 1 p
j
l
j
, j = 1, . . . , n, we say that

(i
1
,...,in)
(f; g
1
, . . . , g
n
)(u
(1)
1
u
(n)
ln
)
:=

s
1
,...,snJ
f
_
P
s
1
g
1
(u
(1)
1
u
(1)
l
1
) P
sn
g
n
(u
(n)
1
u
(n)
ln
)
_
is well-dened if the multiple sum converges absolutely. This give arise to a partially
dened substitution map, for 1 i
j
k
j
, j = 1, . . . , n,

i
1
,...,in
: E
U
(m, n) E
U
(k
1
, l
1
) E
U
(k
n
, l
n
) E
U
(m + k n, l)
where k = k
1
+ + k
n
and l = l
1
+ + l
n
. In general, the compositions of three
substitution maps are not associative. The permutation groups actions on E
U
are the
usual one. Let =
i
1
,...,in
. It is clear that (E
U
, , id
U
) is a partial nonassociative
dioperad. We often denote it simply by E
U
.
Denition 2.8. Let (T,
P
, I
P
) be a partial dioperad. A T-algebra (U, ) consists of a
graded vector space U and a morphism of partial nonassociative dioperad : T E
U
.
When U =
nJ
U
(n)
is a completely reducible module for a group G, J is the set
of equivalent classes of irreducible G-modules and U
(n)
is a direct sum of irreducible
G-modules of equivalent class n J, we denote E
U
by E
G
U
.
Denition 2.9. Let (T,
P
, I
P
) be a G-rescalable partial dioperad. A G-rescalable T-
algebra (U, ) is a T-algebra and the morphism : T E
G
U
is so that [
G
: G End U
coincides with the given G-module structure on U.
2.2 Conformal full eld algebras
Let (V
L
, Y
V
L, 1
L
,
L
) and (V
R
, Y
V
R, 1
R
,
R
) be two vertex operator algebras with central
charge c
L
and c
R
respectively, satisfying the conditions in Theorem 1.1. Let (V
cl
, m
cl
,
cl
)
be a conformal full eld algebra over V
L
V
R
. A bilinear form (, )
cl
on V
cl
is invariant
[Ko1] if, for any u, w
1
, w
2
V
cl
,
(w
2
, Y
f
(u; x, x)w
1
)
cl
= (Y
f
(e
xL
L
(1)
x
2L
L
(0)
e
xL
R
(1)
x
2L
R
(0)
u; e
i
x
1
, e
i
x
1
)w
2
, w
1
)
cl
. (2.17)
11
or equivalently,
(Y
f
(u; e
i
x, e
i
x)w
2
, w
1
)
cl
= (w
2
, Y
f
(e
xL
L
(1)
x
2L
L
(0)
e
xL
R
(1)
x
2L
R
(0)
u; x
1
, x
1
)w
1
)
cl
. (2.18)
We showed in [Ko1] that an invariant bilinear form on V
cl
is automatically symmetric.
Namely, for u
1
, u
2
V
cl
, we have
(u
1
, u
2
)
cl
= (u
2
, u
1
)
cl
. (2.19)
V
cl
has a countable basis. We choose it to be e
i

iN
. Assume that (, )
cl
is also
nondegenerate, we also have the dual basis e
i

iN
. Then we dene a linear map
cl
:
C V
cl
V
cl
as follow:

cl
: 1

iN
e
i
e
i
. (2.20)
The correlation functions maps m
(n)
cl
, n N of V
cl
are canonically determined by Y
and the identity 1
cl
:=
cl
(1
L
1
R
) [Ko2].
For Q T
K
(n

, n
+
) given in (2.3), we dene, for C,

cl
_

K
(
K
(Q))
_
(u
1
u
n
+
) (2.21)
in the following three cases:
1. If z
k
,= for k = n

, . . . , 1, 1, . . . , n
+
, (2.21) is given by

i
1
,...,in

N
_
1
cl
, m
(n

+n
+
)
cl
(e
L
L
+
(A
(1)
)L
R
+
(A
(1)
)
(a
(1)
0
)
L
L
(0)
a
(1)
0
L
R
(0)
e
i
1
,
. . . , e
L
L
+
(A
(n

)
)L
R
+
(A
(n

)
)
(a
(n

)
0
)
L
L
(0)
a
(n

)
0
L
R
(0)
e
in

,
e
L
L
+
(A
(1)
)L
R
+
(A
(1)
)
(a
(1)
0
)
L
L
(0)
a
(1)
0
L
R
(0)
u
1
,
. . . , e
L
L
+
(A
(n
+
)
)L
R
+
(A
(n
+
)
)
(a
(n
+
)
0
)
L
L
(0)
a
(n
+
)
0
L
R
(0)
u
n
+
;
z
1
, z
1
, . . . , z
n

, z
n

, z
1
, z
1
, . . . , z
n
+
, z
n
+
)
_
cl
e
i
1
e
in

(2.22)
where L
L
+
(A) =

j=1
A
j
L
L
+
(j) and L
R
+
(A) =

j=1
A
j
L
R
+
(j) for A = (A
1
, A
2
, . . . )

j=1
C;
2. If k n

, . . . , 1 such that z
k
= (recall (2.5)), (2.21) is given by the
formula obtained from (2.22) by exchanging 1
cl
in (2.22) with
e
L
L
+
(A
(k)
)L
R
+
(A
(k)
)
(a
(k)
0
)
L
L
(0)
a
(k)
0
L
R
(0)
e
i
k
; (2.23)
3. If k 1, . . . , n
+
such that z
k
= (recall (2.5)), (2.21) is given by the formula
obtained from (2.22) by exchanging 1
cl
in (2.22) with
e
L
L
+
(A
(k)
)L
R
+
(A
(k)
)
(a
(k)
0
)
L
L
(0)
a
(k)
0
L
R
(0)
u
k
. (2.24)
12
The following result is proved in [Ko1].
Proposition 2.10. The map
cl
is SL(2, C)-invariant,
Hence
cl
induces a map

K
c
L

K
c
R
E
C

V
cl
, which is still denoted as
cl
. Some
interesting special cases are listed below:

cl
(
K
(

C)) = (1
cl
, 1
cl
)
cl
id
C
,

cl
(
K
(
K
(( (, 1, 0)[ )
K
)) = 1
cl
,

cl
(
K
(
K
(( [(, 1, 0), (0, 1, 0) )
K
)))(u v) = (u, v)
cl
,

cl
(
K
(
K
(( (, 1, 0), (0, 1, 0)[ )
K
))) =
cl
,

cl
(
K
(
K
(( (; 1, 0)[(z; 1, 0), (0; 1, 0) )
K
)))(u v) = Y(u; z, z)v,
where

C is the single element in K(0, 0).
Denition 2.11. A

K
c
L

K
c
R
-algebra (U, ) is called smooth if
1. U =
m,nR
U
(m,n)
is a completely reducible C

-module, where z u = z
m
z
n
u, z
C

, u U
(m,n)
.
2. dimU
(m,n)
< , m, n R and dimU
(m,n)
= 0 for m or n suciently small.
3. is linear on ber and smooth on the base space K.
The Theorem 2.7 in [Ko1] can be restated as the following theorem.
Theorem 2.12. (V
cl
,
cl
) is a smooth

K
c
L

K
c
R
-algebra.
2.3 Open-string vertex operator algebras
Let (V
op
, Y
op
, 1
op
,
op
) be an open-string vertex operator algebra. For r > 0 and v
1
, v
2

V
op
, we dene Y
op
(v
1
, r)v
2
by
Y
op
(v
1
, r)v
2
:= e
rL(1)
Y
op
(v
2
, r)v
1
. (2.25)
Remark 2.13. Taking the analogy between open-string vertex operator algebra and
associative algebra, Y
op
(, r) corresponds to the opposite product [HKo1].
An invariant bilinear form on an open-string vertex operator algebra V
op
is a bilinear
form (, )
op
on V
op
satisfying the following properties:
(v
3
, Y
op
(v
1
, r)v
2
)
op
= (Y
op
(e
rL(1)
r
2L(0)
v
1
, r
1
)v
3
, v
2
)
op
(2.26)
(Y
op
(v
1
, r)v
3
, v
2
)
op
= (v
3
, Y
op
(e
rL(1)
r
2L(0)
v
1
, r
1
)v
2
)
op
(2.27)
for r > 0 and v
1
, v
2
, v
3
V
op
.
Lemma 2.14.
(v
1
, v
2
)
op
= (v
2
, v
1
)
op
(2.28)
13
Proof. The proof is exactly same as that of Proposition 2.3 in [Ko1].
We further assume that (, )
op
is nondegenerate. Let f
i

iR
be a basis of V
op
and
f
i

iR
its dual basis. We dene linear map
op
: C V
cl
V
cl
as follow:

op
: 1

iR
f
i
f
i
. (2.29)
The open-string vertex operator algebra (V
op
, Y
op
, 1
op
,
op
) naturally gives a boundary
eld algebra (V
op
, m
op
, d
op
, D
op
) in which the correlation-function maps m
(n)
op
, n N are
completely determined by Y
op
and 1
op
[Ko2].
For any Q T
D
(n

, n
+
) given in (2.10). Let be a bijective map
n

, . . . , 1, 1, . . . , n
+

1, . . . , n

+ n
+
(2.30)
so that s
1
, . . . , s
n

+n
+
, dened by s
i
:= r

1
(i)
, satisfy s
1
> > s
n

+n
+
0.
Then we dene, for C,

op
(
D
(
D
(Q)))(v
1
v
n
+
) (2.31)
as follow:
1. If r
k
,= , k = n

, . . . , 1, 1, . . . , n
+
, (2.31) is given by

i
1
,...,in

R
_
1
op
, m
(n

+n
+
)
op
(w
1
, . . . , w
n

+n
+
; s
1
, . . . , s
n

+n
+
)
_
op
f
i
1
f
in

(2.32)
where w
(p)
= e
L
+
(B
(p)
)
(b
(p)
0
)
L(0)
f
i
p
and w
(q)
= e
L
+
(B
(q)
)
(b
(q)
0
)
L(0)
v
q
for p =
1, n

and q = 1, . . . , n
+
;
2. If k n

, . . . , 1, 1, . . . , n
+
such that r
k
= , (2.31) is given by the formula
obtained from (2.32) by exchanging the 1
op
with w
(k)
.
Proposition 2.15.
op
is SL(2, R) invariant.
Proof. The proof is same as that of Proposition 2.10.
Hence
op
induces a map

D
c
E
R
+
Vop
, which is still denoted as
op
. Some interesting
special cases are listed explicitly below:

op
(
D
(

H)) = (1
op
, 1
op
)
op
id
C
,

op
(
D
(
D
(( (, 1, 0)[ )))) = 1
op
,

op
(
D
(
D
(( [(, 1, 0), (0, 1, 0) ))))(u v) = (u, v)
op
,

op
(
D
(
D
(( (, 1, 0), (0, 1, 0)[ )))) =
op
,

op
(
D
(
D
(( (; 1, 0)[(r; 1, 0), (0; 1, 0) ))))(u v) = Y
op
(u, r)v.
where

H is the single element in D(0, 0) and r > 0.
Denition 2.16. A

D
c
-algebra (U, ) is called smooth if
14
1. U =
nR
U
(n)
is a completely reducible R
+
-module, where r u = r
n
u, r
R
+
, u U
(m,n)
.
2. dimU
(n)
< , n R and dimU
(n)
= 0 for n suciently small.
3. is linear on ber and smooth on the base space D.
Theorem 2.17. (V
op
,
op
) is a smooth

D
c
-algebra.
Proof. The proof is same as that of Theorem 2.12.
3 Swiss-cheese partial dioperad
In Section 3.1, we introduce the notion of 2-colored partial dioperad and algebra over
it. In Section 3.2, we study a special example of 2-colored partial dioperad called Swiss-
cheese partial dioperad S and its C-extension

S
c
. In Section 3.3, we show that an
open-closed eld algebra over V equipped with nondegenerate invariant bilinear forms
canonically gives an algebra over

S
c
. In Section 3.4, we dene boundary states in such
algebra and show that some of the boundary states are Ishibashi states.
3.1 2-colored (partial) dioperads
Denition 3.1. A right module over a dioperad (Q,
Q
, I
Q
), or a right Q-module, is a
family of sets T(m, n)
m,nN
with an S
m
S
n
-action on each set T(m, n) and substi-
tution maps:
T(m, n) Q(k
1
, l
1
) Q(k
n
, l
n
)

(i
1
,...,in)
T(m + k
1
+ k
n
n, l
1
+ + l
n
)
(P, Q
1
, . . . , Q
n
)
(i
1
,...,in)
(P; Q
1
, . . . , Q
n
) (3.1)
for m, n, l
1
, . . . , l
n
N, k
1
, . . . , k
n
Z
+
and 1 i
j
k
j
, j = 1, . . . , n, satisfying the right
unit property, the associativity and the permutation axioms of dioperad but with the
right action of T on itself in the denition of dioperad replaced by that of Q.
Homomorphism and isomorphism between two right Q-modules can be naturally
dened. The right module over a partial dioperad can also be dened in the usual way.
Denition 3.2. Let Qbe a G-rescalable partial dioperad. A right Q-module is called G-
rescalable if for any P T(m

, m
+
), Q
i
Q(n
(i)

, n
(i)
+
) and 1 j
i
n
(i)

, i = 1, . . . , m
+
,
there exist g
j
G, j = 1, . . . , m
+
such that

(j
1
,...,jm
+
)
_
P;
(j
1
)
(g
1
; Q
1
), . . . ,
(jm
+
)
(g
m
+
; Q
m
+
)
_
(3.2)
is well-dened.
Denition 3.3. A 2-colored dioperad consists of a dioperad (Q,
Q
, I
Q
) and a family of
sets T(n
B

, n
B
+
[n
I

, n
I
+
) equipped with a S
n
B

S
n
B
+
S
n
I

S
n
I
+
-action for n
B

, n
I

N, a
15
distinguished element I
P
T(1, 1[0, 0), and maps
T(m

, m
+
[n

, n
+
) T(k
(1)

, k
(1)
+
[l
(1)

, l
(1)
+
) T(k
(m
+
)

, k
(m
+
)
+
[l
(m
+
)

, l
(m
+
)
+
)

B
(i
1
,...,im
+
)
T(m

m
+
+ k

, k
+
[n

+ l

, n
+
+ l
+
),
T(m

, m
+
[n

, n
+
) Q(p
(1)

, p
(1)
+
) Q(p
(n
+
)

, p
(n)
+
)

I
(j
1
,...,jn
+
)
T(m

, m
+
[n

n
+
+ p

, p
+
) (3.3)
where k

= k
(1)

+ + k
(m
+
)

, l

= l
(1)

+ + l
(m
+
)

and p

= p
(1)

+ + p
(n
+
)

, for
1 i
r
k
(r)

, r = 1, . . . , m
+
, 1 j
s
p
(s)

, s = 1, . . . , n
+
, satisfying the following
axioms:
1. The family of sets T := T(m

, m
+
)
m

,m
+
N
, where
T(m

, m
+
) :=
n

,n
+
N
T(m

, m
+
[n

, n
+
),
equipped with the natural S
m

S
m
+
-action on T(m

, m
+
), together with identity
element I
P
and the family of maps
B
:=
B
(i
1
,...,im
+
)
gives an dioperad.
2. The family of maps
I
:=
I
(j
1
,...,jn
+
)
gives each T(m

, m
+
) a right Q-module
structure for m

, m
+
N.
We denote such 2-colored dioperad as (T[Q, (
B
,
I
)). 2-colored partial dioperad can
be naturally dened. If the associativities of
Q
,
B
,
I
do not hold, then it is called
2-colored (partial) nonassociative dioperad.
Remark 3.4. If we restrict to T(1, m[0, n)
m,nN
and Q(1, n)
nN
, they simply gives
a structure of 2-colored (partial) operad [V][Kont][Ko2].
Now we discuss an important example of 2-colored partial nonassociative dioperad.
Let J
1
, J
2
be two index sets. U
i
=
nJ
i
(U
i
)
(n)
, i = 1, 2 be two graded vector spaces.
Consider two families of sets,
E
U
2
(m, n) = Hom
C
(U
m
2
, U
n
2
),
E
U
1
|U
2
(m

, m
+
[n

, n
+
) = Hom
C
(U
m
+
1
U
n
+
2
, U
m

1
U
n

2
),
for m, n, m

, n

N. We denote both of the projection operators U

1
(U
1
)
(n)
, U
2

(U
2
)
(n)
as P
n
for n J
1
or J
2
. For f E
U
1
|U
2
(k

, k
+
[l

, l
+
), g
i
E
U
1
|U
2
(m
(i)

, m
(i)
+
[n
(i)

, n
(i)
+
), i =
16
1, . . . , k
+
, and h
j
E
U
2
(p
(j)

, p
(j)
+
), j = 1, . . . , l
+
, we say that

B
(i
1
,...,i
k
+
)
(f; g
1
, . . . , g
k
+
)(u
(1)
1
v
(k
+
)
n
(k
+
)
+
v
1
v
l
+
)
:=

s
1
,...,s
k
J
1
f
_
P
s
1
g
1
(u
(1)
1
u
(1)
m
(1)
+
v
(1)
1
v
(1)
n
(1)
+
)
P
s
k
+
g
k
+
(u
(k
+
)
1
u
(k
+
)
m
(k
+
)
+
v
(k
+
)
1
v
(k
+
)
n
(k
+
)
+
) v
1
v
l
+
_

I
(j
1
,...,j
l
+
)
(f; h
1
, . . . , h
l
+
)(u
1
u
k
+
w
(1)
1
w
(l
+
)
p
(l
+
)
+
)
:=

t
1
,...,t
l
J
2
f
_
u
1
u
k
+
P
t
1
h
1
(w
(1)
1
w
(1)
p
(1)
+
)
P
t
l
+
h
l
+
(w
(l
+
)
1
w
(l
+
)
p
(l
+
)
+
)
_
for u
(i)
j
U
1
, v
(i)
j
, w
(i)
j
U
2
, are well-dened if each multiple sum is absolutely conver-
gent. These give arise to partially dened substitution maps:
E
U
1
|U
2
(k

, k
+
[l

, l
+
) E
U
1
|U
2
(m
(1)

, m
(1)
+
[n
(1)

, n
(1)
+
) E
U
1
|U
2
(m
(k
+
)

, m
(k
+
)
+
[n
(k
+
)

, n
(k
+
)
+
)

B
(i
1
,...,i
k
+
)
E
U
1
|U
2
(k

k
+
+ m

, m
+
[l

+ n

, n
+
).
E
U
1
|U
2
(k

, k
+
[l

, l
+
) E
U
2
(p
(1)

, p
(1)
+
) E
U
2
(p
(l
+
)

, p
(l

)
+
)

I
(j
1
++j
l
+
)
E
U
1
|U
2
(k

, k
+
[l

l
+
+ p
(1)

+ + p
(l
+
)

, p
(1)
+
+ + p
(l
+
)
+
).
where m

= m
(1)

+ + m
(k
+
)

and n

= n
(1)

+ + n
(k
+
)

. In general,
B
(i
1
,...,i
k
+
)
and

I
(j
1
++j
l
+
)
do not satisfy the associativity. Let
E
U
1
|U
2
= E
U
1
|U
2
(m

, m
+
[n

, n
+
)
m

,n

N
,
E
U
2
= E
U
2
(n)
nN
,
B
:=
B
(i
1
,...,in)
and
I
:=
I
(j
1
,...,jn)
. It is obvious that
(E
U
1
|U
2
[E
U
2
, (
B
,
I
)) is a 2-colored partial nonassociative operad.
Let U
1
be a completely reducible G
1
-modules and U
2
a completely reducible G
2
-
modules. Namely, U
1
=
n
1
J
1
(U
1
)
(n
1
)
, U
2
=
n
2
J
2
(U
2
)
(n
2
)
where J
i
is the set of
equivalent classes of irreducible G
i
-modules and (U
i
)
(n
i
)
is a direct sum of irreducible
G
i
-modules of equivalent class n
i
for i = 1, 2. In this case, we denote E
U
1
|U
2
by E
G
1
|G
2
U
1
|U
2
.
Denition 3.5. A homomorphism between two 2-colored (partial) dioperads
(T
i
[Q
i
, (
B
i
,
I
i
)), i = 1, 2
consists of two (partial) dioperad homomorphisms:

P
1
|Q
1
: T
1
T
2
, and
Q
1
: Q
1
Q
2
such that
P
1
|Q
1
: T
1
T
2
, where T
2
has a right Q
1
-module structure induced by
dioperad homomorphism
Q
1
, is also a right Q
1
-module homomorphism.
17
Denition 3.6. An algebra over a 2-colored partial dioperad (T[Q, (
B
,
I
)), or a T[Q-
algebra consists of two graded vector spaces U
1
, U
2
and a 2-colored partial dioperad
homomorphism (
P|Q
,
Q
) : (T[Q, (
B
,
I
)) (E
U
1
|U
2
[E
U
2
, (
B
,
I
)). We denote this
algebra as (U
1
[U
2
,
P|Q
,
Q
).
Denition 3.7. If a 2-colored partial dioperad (T[Q, ) is so that T is a G
1
-rescalable
partial operad and a G
2
-rescalable right Q-module, then it is called G
1
[G
2
-rescalable.
Denition 3.8. A G
1
[G
2
-rescalable T[Q-algebra (U
1
[U
2
,
P|Q
,
Q
) is a T[Q-algebra so
that
P|Q
: T E
G
1
|G
2
U
1
|U
2
and
Q
: Q E
G
2
U
2
are dioperad homomorphisms such that

P|Q
: G
1
End U
1
coincides with the G
1
-module structure on U
1
and
Q
: G
2

End U
2
coincides with the G
2
-module structure on U
2
.
3.2 Swiss-cheese partial dioperads
A disk with strips and tubes of type (m

, m
+
; n

, n
+
) [Ko2] is a disk consisting of m
+
(m

) ordered positively (negatively) oriented punctures on the boundary of the disk,

and n
+
(n

) ordered positively (negatively) oriented punctures in the interior of the disk,

and local coordinate map around each puncture. Two disks are conformal equivalent
if there exists a biholomorphic map between them preserving order, orientation and
local coordinates. We denote the moduli space of disks with strips and tubes of type
(m

, m
+
; n

, n
+
) as S(m

, m
+
[n

, n
+
).
We use the following notation
_
( (r
m

, b
m

0
, B
(m

)
), . . . , (r
1
, b
1
0
, B
(1)
) [
(r
1
, b
1
0
, B
(1)
), . . . , (r
m
+
, b
m
+
0
, B
(m
+
)
) )
_
_
( (z
n

, a
n

0
, A
(n

)
), . . . , (z
1
, a
1
0
, A
(1)
) [
(z
1
, a
1
0
, A
(1)
), . . . , (z
n
+
, a
n
+
0
, A
(n
+
)
) )

S
(3.4)
where r
i


R, b
(i)
0
R

, B
(i)
k
R and z
j
H, a
(j)
0
C

, A
(j)
l
C for all i =
m

, . . . , 1, 1, . . . , m
+
, j = n

, . . . , 1, 1, . . . , n
+
and k, l Z
+
, to represent a disk
with strips at r
i
with local coordinate map f
i
and tubes at z
j
with local coordinate map
g
j
given as follow:
f
i
(w) = e
P
k
B
(i)
k
x
k+1 d
dx
(b
(i)
0
)
x
d
dx
x[
x=wr
i
if r
i
R , (3.5)
= e
P
k
B
(i)
k
x
k+1 d
dx
(b
(i)
0
)
x
d
dx
x[
x=
1
w
if r
i
= , (3.6)
g
j
(w) = e
P
k
A
(j)
k
x
k+1 d
dx
(a
(j)
0
)
x
d
dx
x[
x=wz
j
. (3.7)
We denote the set of all such disks given in (3.4) as T
S
(m

, m
+
[n

, n
+
). The auto-
morphisms of the upper half plane, which is SL(2, R), change the disk (3.4) to a dierent
but conformal equivalent disk. It is clear that we have
S(m

, m
+
[n

, n
+
) = T
S
(m

, m
+
[n

, n
+
)/SL(2, R). (3.8)
Let S = S(m

, m
+
[n

, n
+
)
m

,m
+
,n

,n
+
N
. The permutation groups
(S
m

S
m
+
) (S
n

S
n
+
)
18
acts naturally on S(m

, m
+
[n

, n
+
). There are so-called boundary sewing operations
[HKo1][Ko2] on S, denoted as i
B
j
, which sews the i-th positively oriented boundary
puncture of the rst disk with the j-th negatively oriented boundary puncture of the
second disk. Boundary sewing operations naturally induce the following maps:
S(m

, m
+
[n

, n
+
) S(k
(1)

, k
(1)
+
[l
(1)

, l
(1)
+
) S(k
(m
+
)

, k
(m
+
)
+
[l
(m
+
)

, l
(m
+
)
+
)

B
(i
1
,...,im
+
)
S(m

m
+
+ k

, k
+
[n

+ l

, n
+
+ l
+
),
where k

= k
(1)

+ + k
(m
+
)

, l

= l
(1)

+ + l
(m
+
)

for 1 i
r
k
(r)

, r = 1, . . . , m
+
. It
is easy to see that boundary sewing operations or
B
(i
1
,...,im
+
)
, together with permutation
group actions on the order of boundary punctures, provide S with a structure of partial
dioperad.
There are also so-called interior sewing operations [HKo1][Ko2] on S, denoted as
i
I
j
, which sew the i-th positively oriented interior puncture of a disk with the j-th
negatively oriented puncture of a sphere. The interior sewing operations dene a right
action of K on S:
S(m

, m
+
[n

, n
+
) K(p
(1)

, p
(1)
+
) K(p
(n
+
)

, p
(n
+
)
+
)

I
(j
1
,...,jn
+
)
S(m

, m
+
[n

n
+
+ p

, p
+
) (3.9)
where p

= p
(1)

+ +p
(n
+
)

, for 1 j
s
p
(s)

, s = 1, . . . , n
+
. Such action gives S a right
K-module structure.
Let
B
=
B
(i
1
,...,in)
and
I
=
I
(i
1
,...,in)
. The following proposition is clear.
Proposition 3.9. (S[K, (
B
,
I
)) is a R
+
[C

-rescalable 2-colored partial dioperad.

We call (S[K, (
B
,
I
)) Swiss-cheese partial dioperad. When it is restricted on S =
S(1, m[0, n)
m,nN
, it is nothing but the so-called Swiss-cheese partial operad [HKo1][Ko2].
In [HKo1][Ko2], we show that the Swiss-cheese partial operad S can be naturally
embedded into the sphere partial operad K via the so-called doubling map, denoted as
: S K. Such doubling map obviously can be extended to a doubling map S K,
still denoted as . In particular, the general element (3.4) maps under to
( (z
1
, a
(1)
0
, A
(1)
), . . . , (z
n

, a
(n

)
0
, A
(n

)
),
( z
1
, a
(1)
0
, A
(1)
), . . . ( z
n

, a
(n

)
0
, A
(n

)
),
(r
1
, b
(1)
0
, B
(1)
), . . . , (r
m

, b
(m

)
0
, B
(m

)
) [
(z
1
, a
(1)
0
, A
(1)
), . . . , (z
n
+
, a
(n
+
)
0
, A
(n
+
)
),
( z
1
, a
(1)
0
, A
(1)
), . . . , ( z
n
+
, a
(n
+
)
0
, A
(n
+
)
),
(r
1
, b
(1)
0
, B
(1)
), . . . , (r
m
+
, b
(m
+
)
0
, B
(m
+
)
) )
K
. (3.10)
The following proposition is clear.
19
Proposition 3.10. Let P
i
S(m
(i)

, m
(i)
+
[n
(i)

, n
(i)
+
), i = 1, 2 and Q K(m

, m
+
). If
P
1
i
B
j
P
2
and P
1
k
I
l
Q exists for 1 i m
(1)
+
, 1 j m
(2)

and 1 k n
(1)
+
, 1 l
m

, then we have
(P
1
i
B
j P
2
) = (P
1
)
2n
(1)
+
+i

(2n
(2)

+j)
(P
2
),
(P
1
k
I
l
Q) = ((P
1
) kl Q)
n
(1)
+
+m
+
1+k
l

Q. (3.11)
By above Proposition, we can identify S as its image under in K with boundary
sewing operations replaced by ordinary sewing operations in K and interior sewing
operations replaced by double-sewing operations in K as given in (3.11).
The C-extension

S
c
(m

, m
+
[n

, n
+
) of S(m

, m
+
[n

, n
+
) is dened to be the pull-
back bundle of

K
c
(2n

+m

, 2n
+
+m
+
). We denote the canonical section on

S
c
, which
is induced from that on

K
c
, as
S
. The boundary (interior) sewing operations can be
naturally extended to

S
c
. We denote them as
B
(
I
) and corresponding substitution
maps as
B
(
I
). There is a natural right action of

K
c

K
c
on

S
c
dened by
I
.
The following proposition is also clear.
Proposition 3.11. (

S
c
[

K
c

K
c
, (
B
,
I
)) is a R
+
[C

-rescalable 2-colored partial diop-

erad.
We will call the structure (

S
c
[

K
c

K
c
, (
B
,
I
)) as Swiss-cheese partial dioperad with
central charge c.
Denition 3.12. An algebra over

S
c
viewed as a dioperad, (U, ), is called smooth if
1. U =
nR
U
(n)
is a completely reducible R
+
-module, where r u = r
n
u, r
R
+
, u U
(m,n)
.
2. dimU
(n)
< , n R and dimU
(n)
= 0 for n suciently small.
3. is linear on ber and smooth on the base space S.
Denition 3.13. An

S
c
[

K
c

K
c
-algebra (U
1
[U
2
,

S
c
|

K
c

K
c
,

K
c

K
c
) is smooth if both
(U
1
,

S
c
|

K
c

K
c
) and (U
2
,

K
c

K
c
) as algebras over dioperads are smooth.
3.3 Open-closed eld algebras over V
Let (V
op
, Y
op
,
op
) be an open-string vertex operator algebra over V and (V
cl
, Y,
cl
) a
conformal full eld algebra over V V . Let
( (V
cl
, Y,
cl
), (V
op
, Y
op
,
op
), Y
clop
) (3.12)
be an open-closed eld algebra over V [Ko2]. We denote the formal vertex operators
associated with Y
op
and Y as Y
f
op
and Y
f
respectively. Let
L
=
cl
(1),
R
=
cl
(1)
20
and
op
=
op
(). We have
Y
f
(
L
; x, x) = Y
f
(
L
, x) =

nZ
L(n) 1x
n2
,
Y
f
(
R
; x, x) = Y
f
(
R
, x) =

nZ
1 L(n) x
n2
,
Y
f
op
(
op
, x) =

nZ
L(n)x
n2
. (3.13)
We also set L
L
(n) = L(n) 1 and L
R
(n) = 1 L(n) for n Z.
For u V , we showed in [Ko2] that Y
clop
(
cl
(u 1); z, z) and Y
clop
(
cl
(1 u); z, z)
are holomorphic and antiholomorphic respectively. So we also denote them simply by
Y
clop
(
cl
(u 1), z) and Y
clop
(
cl
(1 u), z) respectively.
By the V -invariant boundary condition [Ko2], we have
Y
clop
(
L
, r) = Y
op
(
op
, r) = Y
clop
(
R
, r). (3.14)
We assume that both V
op
and V
cl
are equipped with nondegenerate invariant bilinear
forms (, )
op
and (, )
cl
respectively.
Lemma 3.14. For any u V
cl
and v
1
, v
2
V
op
and z H, we have
(v
2
, Y
clop
(u; z, z)v
1
)
op
= (Y
clop
(e
zL(1)
z
2L(0)
e
zL(1)
z
2L(0)
u; z
1
, z
1
)v
2
, v
1
)
op
(3.15)
Proof. Using (2.26), for xed z H, we have,
(v
2
, Y
clop
(u; z, z)v
1
)
op
= (v
2
, Y
clop
(u; z, z)Y
op
(1, r)v
1
)
op
= (v
2
, Y
op
(Y
clop
(u; z r, z r)1, r)v
1
)
op
= (e
r
1
L(1)
Y
op
(v
2
, r
1
)e
rL(1)
r
2L(0)
Y
clop
(u; z r, z r)1, v
1
)
op
(3.16)
for [z[ > r > [zr[ > 0. Notice that e
r
1
L(1)
Aut(V
op
), r C. By taking v
2
= 1
op
,
it is easy to see that
e
rL(1)
r
2L(0)
Y
clop
(u; z r, z r)1 (3.17)
is a well-dened element in V
op
for [z[ > r > [z r[ > 0. Because of the chirality
splitting property of Y
clop
(see (1.72)(1.73) in [Ko2]), it is easy to show that (3.17)
equals to
Y
clop
(e
zL(1)
z
2L(0)
e
zL(1)
z
2L(0)
u, r
1
z
1
, r
1
z
1
)1 (3.18)
for r > [z r[ > 0. By the commutativity I of analytic open-closed eld algebra proved
in [Ko2], we know that for xed z H,
e
r
1
L(1)
Y
op
(v
2
, r
1
)Y
clop
(e
zL(1)
z
2L(0)
e
zL(1)
z
2L(0)
u, r
1
z
1
, r
1
z
1
)1
21
and
e
r
1
L(1)
Y
clop
(e
zL(1)
z
2L(0)
e
zL(1)
z
2L(0)
u, r
1
z
1
, r
1
z
1
)Y
op
(v
2
, r
1
)1
= e
r
1
L(1)
Y
clop
(e
zL(1)
z
2L(0)
e
zL(1)
z
2L(0)
u, r
1
z
1
, r
1
z
1
)e
r
1
L(1)
v
2
(3.19)
converge in dierent domains for r, but are analytic continuation of each other along a
path in r R
+
. Moreover, using L(1) property of intertwining operator and chirality
splitting property of Y
clop
again, the right hand side of (3.19) equals to
Y
clop
(e
zL(1)
z
2L(0)
e
zL(1)
z
2L(0)
u, z
1
, z
1
)v
2
for [r
1
z
1
[ > [r
1
[. Therefore, the both sides of (3.15) as constant functions of r are
analytic continuation of each other. Hence (3.15) must hold identically for all z H.
For u
1
, . . . , u
l
V
cl
, v
1
, . . . , v
n
V
cl
, r
1
, . . . , r
n
R, r
1
> > r
n
and z
1
, . . . , z
l
H,
we dene
m
(l;n)
clop
(u
1
, . . . , u
l
; v
1
, . . . , v
n
; z
1
, z
1
, . . . , z
l
, z
l
; r
1
, . . . , r
n
)
:= e
rnL(1)
m
(l;n)
clop
(u
1
, . . . , u
l
; v
1
, . . . , v
n
; z
1
r
n
, z
1
r
n
, . . . , z
l
r
n
, z
l
r
n
;
r
1
r
n
, . . . , r
n1
r
n
, 0). (3.20)
We simply extend the denition of m
(l;n)
clop
to a domain where some of r
i
can be negative.
Note that such denition is compatible with L(1)-properties of m
clop
.
Lemma 3.15. For u
1
, . . . , u
l
V
cl
, v, v
1
, . . . , v
n
V
cl
, r
1
, . . . , r
n
R, r
1
> > r
n
= 0
and z
1
, . . . , z
l
H,
(v, m
(l;n)
clop
(u
1
, . . . , u
l
; v
1
, . . . , v
n
; z
1
, z
1
, . . . , z
l
, z
l
; r
1
, . . . , r
n
))
op
= (v
n
, m
(l;n)
clop
(F
1
u
1
, . . . , F
l
u
l
; v, G
1
v
1
, . . . , G
n1
v
n1
;
z
1
1
, z
1
1
, . . . , z
1
l
, z
1
l
; 0, r
1
1
, . . . , r
1
n1
))
op
(3.21)
where
F
i
= e
z
i
L(1)
z
2L(0)
i
e
z
i
L(1)
z
2L(0)
i
, i = 1, . . . , l,
G
j
= e
r
j
L(1)
r
2L(0)
j
, j = 1, . . . , n 1. (3.22)
Proof. By Lemma 3.14, (3.21) is clearly true for l = 0, 1; n = 0, 1. By (2.26) and
(2.27), (3.21) is true for l = 0, n = 2. We then prove the Lemma by induction. Assume
that (3.21) is true for l = k 0, n = m 2 or l = k 1, n = m 1.
Let l = k and n = m + 1. It is harmless to assume that 0 < r
n1
, [z
i
[ < r
n2
, i =
22
l
1
+ 1, . . . , l for some l
1
l. Using the induction hypothesis, we obtain
(v, m
(l;n)
clop
(u
1
, . . . , u
l
; v
1
, . . . , v
n
; z
1
, z
1
, . . . , z
l
, z
l
; r
1
, . . . , r
n
))
op
= (v, m
(l
1
;n1)
clop
(u
1
, . . . , u
l
1
; v
1
, . . . , v
n2
, m
(ll
1
;2)
clop
(u
l
1
+1
, . . . , u
l
; v
n1
, v
n
;
z
l
1
+1
, z
l
1
+1
, . . . , z
l
, z
l
; r
n1
, 0); z
1
, z
1
, . . . , z
l
1
, z
l
1
; r
1
, . . . , r
n2
, 0))
op
= (m
(l
1
;n1)
clop
(F
1
u
1
, . . . , F
l
1
u
l
1
; v, G
1
v
1
, . . . , G
n1
v
n1
;
z
1
1
, z
1
1
, . . . , z
1
l
1
, z
1
l
1
; 0, r
1
1
, . . . , r
1
n1
),
m
(ll
1
;2)
clop
(u
l
1
+1
, . . . , u
l
; v
n1
, v
n
; z
l
1
+1
, z
l
1
+1
, . . . , z
l
, z
l
; r
n1
, 0))
op
= (m
(ll
1
;2)
clop
(F
l
1
+1
u
l
1
+1
, . . . , F
l
u
l
; m
(l
1
;n1)
clop
(F
1
u
1
, . . . , F
l
1
u
l
1
;
v, G
1
v
1
, . . . , G
n1
v
n1
; z
1
1
, z
1
1
, . . . , z
1
l
1
, z
1
l
1
; 0, r
1
1
, . . . , r
1
n2
),
G
n1
v
n1
; z
1
l
1
+1
, z
1
l
1
+1
, . . . , z
1
l
, z
1
l
; 0, r
1
n1
), v
n
)
op
=

sR
(e
r
1
n1
L(1)
m
(ll
1
;2)
clop
(F
l
1
+1
u
l
1
+1
, . . . , F
l
u
l
; P
s
e
r
1
n2
L(1)
m
(ll
1
;n1)
clop
(F
1
u
1
,
. . . , F
l
1
u
l
1
; v, G
1
v
1
, . . . , G
n1
v
n1
; z
1
1
+ r
1
n2
, z
1
1
+ r
1
n2
,
. . . , z
1
l
1
+ r
1
n2
, z
1
l
1
+ r
1
n2
; r
1
n2
, r
1
1
+ r
1
n2
, . . . , r
1
n3
+ r
1
n2
, 0),
G
n1
v
n1
; z
1
l
1
+1
+ r
1
n1
, z
1
l
1
+1
+ r
1
n1
,
. . . , z
1
l
+ r
1
n1
, z
1
l
+ r
1
n1
; r
1
n1
, 0), v
n
)
op
(3.23)
Note that the position of P
s
and e
r
1
n2
L(1)
can not be exchanged in general. Because
if we exchange their position, the sum may not converge and then the associativity law
does not hold. We want to use analytic continuation to move it to a domain such that
we can freely apply associativity law. By our assumption on V , both sides of (3.23) are
restrictions of analytic function of z
l
1
+1
,
l
1
+1
, . . . , z
l
,
l
, r
n1
on
l
1
+1
= z
l
1
+1
, . . . ,
l
= z
l
.
Let z
l
1
+1
, . . . , z
l
, r
n1
satisfy the following conditions:
[ z
1
p
+ r
1
n2
[, r
1
n1
r
1
n2
> [ z
1
i
+ r
1
n2
[, r
1
n2
, r
1
j
+ r
1
n2
, (3.24)
for all i = 1, . . . , l, j = 1, . . . , n 3 and p = l
1
+ 1, . . . , l. Note that such condition
dene a nonempty open subset on H
l
R
+
. Choose a path
1
in the complement of
the diagonal in H
l
from initial point (z
l
1
+1
, . . . , z
l
) to ( z
l
1
+1
, . . . , z
l
) and a path
2
in R
+
from r
n1
to r
n1
. We also denote the complex conjugate of path
1
as
1
, which is a
path in H
l
. Combine
1
,
1
with
2
, we obtain a path from (z
l
1
+1
, z
l
1
+1
, . . . , z
l
, z
l
, r
n1
)
to ( z
l
1
+1
,

z
l
1
+1
, . . . , z
l
,

z
l
, r
n1
). Analytically continuating the right hand side of (3.23)
along the path , we obtain, by the properties of m
(l;n)
clop
[Ko2],
(e
r
1
n1
L(1)
m
(ll
1
;2)
clop
(F
l
1
+1
u
l
1
+1
, . . . , F
l
u
l
; P
s
m
(ll
1
;n1)
clop
(F
1
u
1
, . . . , F
l
1
u
l
1
;
v, G
1
v
1
, . . . , G
n1
v
n1
; z
1
1
+ r
1
n2
, z
1
1
+ r
1
n2
,
. . . , z
1
l
1
+ r
1
n2
, z
1
l
1
+ r
1
n2
; r
1
n2
, r
1
1
+ r
1
n2
, . . . , r
1
n3
+ r
1
n2
, 0),

G
n1
v
n1
; z
1
l
1
+1
+ r
1
n1
,

z
1
l
1
+1
+ r
1
n1
,
. . . , z
1
l
+ r
1
n1
,

z
1
l
+ r
1
n1
; r
1
n1
r
1
n2
, 0), v
n
)
op
(3.25)
23
where

G
n1
= e
r
n1
L(1)
r
2L(0)
n1
. Using the associativity of open-closed eld algebra and
L(1)-properties of m
clop
, we see that (3.25) further equals to
(m
(l;n+1)
clop
(F
1
u
1
, . . . , F
l
u
l
; v, G
1
v
1
, . . . ,

G
n1
v
n1
; z
1
1
, z
1
1
, . . . ,
z
1
l
1
, z
1
l
1
, z
1
l
1
+1
,

z
1
l
1
+1
, . . . , z
1
l
, z
1
l
; 0, r
1
1
, . . . , r
1
n1
), v
n
)
op
.
(3.26)
By analytically continuating (3.26) along the path , which is reversed, we obtain
the right hand side of (3.21). Hence (3.23) and the right hand side of (3.21) are analytic
continuation of each other along path () () which is a constant path. Hence (3.21)
holds for l = k, n = m+ 1.
Now let l = k +1, n = m. The proof is similar to the case l = k, n = m+1. We only
point out the dierence. Using the smoothness of m
clop
, it is enough to prove the case
when [z
i
[ , = [z
j
[ for i, j = 1, . . . , l and i ,= j. Without losing generality, we assume that
[z
1
[ > > [z
l
[ > 0. Let n
1
n be the smallest so that 0 < r
j
< [z
l
[ for j n
1
. Then
we have
(v, m
(l;n)
clop
(u
1
, . . . , u
l
; v
1
, . . . , v
n
; z
1
, z
1
, . . . , z
l
, z
l
; r
1
, . . . , r
n
))
op
= (v, m
(l1;n
1
)
clop
(u
1
, . . . , u
l1
; v
1
, . . . , v
n
1
1
, m
(1;nn
1
+1)
clop
(u
l
; v
n
1
, . . . , v
n
;
z
l
, z
l
; r
n
1
, . . . , r
n
); z
1
, z
1
, . . . , z
l1
, z
l1
; r
1
, . . . , r
n
1
1
))
op
(3.27)
We can then apply (3.21) as in (3.23) for the case l k, n m, which is true by our
induction hypothesis. The rest of proof is entirely same as that of the case l = k, n =
m+ 1.
We dene a map, for z, C and z ,= ,
clop
(z, ) : V
cl
V
op
as

clop
(z, )(u) = Y
clop
(u; z, )1
op
.
We denote its adjoint as

clop
(z, ). Namely,

clop
(z, ) : V
op
V
cl
is given by
(

clop
(z, )(w), u)
cl
= (w,
clop
(z, )(u))
op
(3.28)
for any u V
cl
and w V
op
.
Let Q be an element in T
S
(n

, n
+
[m

, m
+
) of form (3.4). Let be the map (2.30) so
that s
1
, . . . , s
n

+n
+
, dened as s
i
:= r

1
(i)
, satisfy s
1
> > s
n

+n
+
0. Then
we dene

clop
(
S
(Q))(u
1
u
m
+
v
1
v
n
+
) (3.29)
as follow:
1. If s
1
,= , (3.29) is given by

i
1
,...,im

;j
1
,...,jn

_
1
op
, m
(m

+m
+
;n

+n
+
)
clop
(u
1
, . . . , u
m

, u
1
, . . . , u
m
+
;
w
1
, . . . , w
n

+n
+
; z
1
, z
1
, . . . , z
m

, z
m

, z
1
, z
1
, . . . , z
m
+
, z
m
+
;
s
1
, . . . , s
n

+n
+
)
_
op
e
i
1
e
in

f
j
1
f
jm

. (3.30)
24
where
u
p
= e
L
L
+
(A
(p)
)L
R
+
(A
(p)
)
(a
(p)
0
)
L
L
(0)
a
(p)
0
L
R
(0)
e
i
p
,
u
q
= e
L
L
+
(A
(q)
)L
R
+
(A
(q)
)
(a
(q)
0
)
L
L
(0)
a
(q)
0
L
R
(0)
u
q
,
w
(k)
= e
L
+
(B
(k)
)
(b
(k)
0
)
L(0)
f
j
k
,
w
(l)
= e
L
+
(B
(l)
)
(b
(l)
0
)
L(0)
v
l
, (3.31)
for p = 1, . . . , n

, q = 1, . . . , n
+
, k = 1, . . . , m

and l = 1, . . . , m
+
.
2. When r
k
= for some k = m

, . . . , 1, 1, . . . , m
+
. (3.29) is given by the
formula obtain from (3.30) by exchanging 1
op
with w
(k)
.
Lemma 3.16.
clop
is SL(2, R)-invariant.
Proof. The SL(2, R) is generated by the following three transformations 1. w
aw, a R
+
; 2. w w b, b R; 3. w
1
w
. That
clop
is invariant under the
rst two transformations simply follows from the L(0)- and L(1)-properties of m
clop
.
That
clop
is invariant under the third transformation is proved in Lemma 3.15.
Hence
clop
induces a map

S
c
E
R
+
|C

Vop|V
cl
, which is still denoted as
clop
. We list a
few interesting cases:

clop
_

S
__
( (, 1, 0)[ )
_
_
( [(z, 1, 0))

S
__
=
clop
(z, z),

clop
_

S
__
( [(, 1, 0))
_
_
((z, 1, 0)[ )

S
__
=

(z, z) (3.32)
and for b R
+
, B R

, a C

, A C

and v V
op
, we have

clop
_

S
__
( [(, b, B))
_
_
((z, a, A)[ )

S
__
= a
L
L
(0)
a
L
R
(0)
e

j=1
(1)
j
[A
j
L
L
(j)+A
j
L
R
(j)]

(z, z)(e
L
+
(B)
b
L(0)
v). (3.33)
Theorem 3.17. (V
op
[V
cl
,
clop
,
cl
) is an R
+
[C

-rescalable smooth

S
c
[

K
c
K
c
-algebra.
Proof. The smoothness is automatic. We showed in [Ko2] that (V
op
[V
cl
,
clop
,
cl
)
is an R
+
[C

-rescalable smooth

S
c
[

K
c
K
c
-algebra. The rest of the proof is similar to
that of Theorem 2.12 in [Ko1]. We omit the detail here.
3.4 Ishibashi states
As we mentioned in the introduction, an open-closed eld algebra over V equipped with
nondegenerate invariant bilinear forms for both open theory and closed theory contains
all the data needed to grow to an open-closed partial eld eld theory of all genus.
Without adding more compatibility conditions, itself is already an interesting object to
study. We show in this subsection that the famous Ishibashi states [I] can be studied
in the framework of such algebra. Throughout this subsection, we x an open-closed
algebra over V given in (3.12) and equipped with nondegenerate invariant bilinear forms
(, )
cl
and (, )
op
.
For u V
op
and z
0
H, we dene the boundary state B
z
0
(u) V
cl
associated with
u and z
0
by
B
z
0
(u) = e
L(1)
( z
0
z
0
)
L(0)
e
L(1)
z
0
z
0
L(0)

clop
(z
0
, z
0
)(u). (3.34)
25
Proposition 3.18. If u V
op
is a vacuum-like vector [LL], i.e. L(1)u = 0, then, for
z
0
H, B
z
0
(u) is an Ishibashi state, i.e.
(L
L
(n) L
R
(n))B
z
0
(u) = 0, n Z. (3.35)
Proof. For v V
cl
, the following two functions of z
(u, Y
clop
(
L
, z + z
0
)Y(v; z
0
, z
0
)1
op
)
op
,
(u, Y
clop
(
R
; z + z
0
)Y(v; z
0
, z
0
)1
op
)
op
(3.36)
can be extended to a holomorphic function and an antiholomorphic function in z[z +
z
0
H, z ,= 0 respectively by our assumption on V . We denote the extended functions
by g
1
(
L
, z) and g
2
(
R
, z) respectively.
The following two limits
lim
z+z
0
r
_
1
z
z
0
z
0
_
4
g
1
(
L
, z)
lim
z+z
0
r
_
1
z
z
0
z
0
_
4
g
2
(
R
, z) (3.37)
exist for all r R. Using (3.15), it is easy to see that above two limits also exist
for r =

R if and only if u is vacuum-like. Hence, by V -invariant (or conformal
invariant) boundary condition, we have
lim
z+z
0
r
_
1
z
z
0
z
0
_
4
g
1
(
L
, z) = lim
z+z
0
r
_
1
z
z
0
z
0
_
4
g
2
(
R
, z) (3.38)
for all r

R when L(1)u = 0.
On the other hand, for [z + z
0
[ > [z
0
[, we have
(u, Y
clop
(
L
, z + z
0
)Y
clop
(v; z
0
, z
0
)1
op
)
op
= (u, Y
clop
(Y(
L
, z)v; z
0
, z
0
)
op
1
op
)
op
= (u,
clop
(z
0
, z
0
)(Y(
L
, z)v))
op
= (

clop
(z
0
, z
0
)(u), Y(
L
, z)v)
cl
= (B
z
0
(u), e
L(1)
( z
0
z
0
)
L(0)
e
L(1)
z
0
z
0
L(0)
Y(
L
, z)v)
cl
. (3.39)
Note that one should check the convergence property of each step in (3.39). In particular,
in the last step, the convergence and equality follow from the convergence of early steps
and the fact that ( z
0
z
0
)
L(0)
e
L(1)
z
0
z
0
L(0)
e
L(1)
Aut V
cl
. For 0 < [z[ <
[Imz
0
[, it is easy to show that
e
L(1)
( z
0
z
0
)
L(0)
e
L(1)
z
0
z
0
L(0)
Y(
L
, z)v
= Y(e
(1
z
z
0
z
0
)L
L
(1)
_
1
z
z
0
z
0
_
2L
L
(0)

( z
0
z
0
)
L
L
(0)

L
,
1
z
0
z
0
z
1
z
z
0
z
0
)v
1
=
_
1
z
z
0
z
0
_
4
( z
0
z
0
)
2
Y(
L
,
1
z
0
z
0
z
1
z
z
0
z
0
)v
1
(3.40)
26
where v
1
= e
L(1)
( z
0
z
0
)
L(0)
e
L(1)
z
0
z
0
L(0)
v. Hence, for all 0 < [z[ < [Imz
0
[ and
[z + z
0
[ > [z
0
[, we obtain
(u, Y
clop
(
L
, z + z
0
)Y
clop
(v; z
0
, z
0
)1
op
)
op
= (B
z
0
(u),
_
1
z
z
0
z
0
_
4
( z
0
z
0
)
2
Y(
L
, f(z))v
1
)
cl
(3.41)
where f is the composition of the following maps:
w w + z
0

(w + z
0
) z
0
(w + z
0
) z
0
=
1
z
0
z
0
w
1
w
z
0
z
0
which maps the domain H z
0
to the unit disk. Since g
1
(
L
, z) is analytic and free of
singularities for z + z
0
Hz
0
, the right hand side of (3.41) can also be extended to
an analytic function in z H z
0
0. If we view f(z) as a new variable , then the
right hand side of (3.41) can be extended to an analytic function on [1 > [[ > 0,
which has a Laurent series expansion. By the uniqueness of Laurent expansion, the
right hand side of (3.41) gives exactly such Laurent expansion and thus is absolutely
convergent in [1 > [[ > 0. Moreover, lim
z+z
0
r
g
1
(
L
, z) exists for all r

R. By the
properties of Laurent series, the right hand side of (3.41) must converge absolutely for
all f(z) [[[ = 1 to the function given by lim
z+z
0
r
g
1
(
L
, z), r

R.
Similarly, for all 0 < [ z[ < [Imz
0
[ and [z + z
0
[ > [z
0
[, we have
(u, Y
clop
(
R
, z + z
0
)Y
clop
(v; z
0
, z
0
)1
op
)
op
= (B
z
0
(u), e
L(1)
( z
0
z
0
)
L(0)
e
L(1)
z
0
z
0
L(0)
Y(
R
, z)v)
cl
= (B
z
0
(u), Y(e
(1
z
z
0
z
0
)L
R
(1)
_
1
z
z
0
z
0
_
2L
R
(0)
(z
0
z
0
)
L
R
(0)

R
, g( z))v
1
)
cl
=
_
1
z
z
0
z
0
_
4
(z
0
z
0
)
2
(B
z
0
(u), Y(
R
, g( z))v
1
)
cl
(3.42)
where g is the composition of the following maps:
w w + z
0

(w + z
0
) z
0
(w + z
0
) z
0

(w + z
0
) z
0
(w + z
0
) z
0
=
1
z
0
z
0
w
1
w
z
0
z
0
which maps the domain H z
0
to the unit disk. Moreover, the right hand side of
(3.42), as a Laurent series of g( z), is absolutely convergent for all g( z) [[[ = 1 to
lim
z+z
0
r
g
2
(
R
, z), r

R.
Also notice that
g(r z
0
) =
1
f(r z
0
)
e
i
[0 < 2 (3.43)
for all r

R. Using (3.38) and by replacing z in (3.41) by r z
0
and z in (3.42) by
r z
0
, we obtain the following identity:
(B
z
0
(u), Y(
L
, e
i
)v
1
)
cl
= (B
z
0
(u), Y(
R
, e
i
)v
1
)
cl
e
4i
(3.44)
27
where e
i
= f(r z
0
), for all 0 < 2. Notice that the existence of both sides of
(3.44) follows directly from (3.38), which further follows from the condition of u being
vacuum-like. Then we obtain

nZ
(B
z
0
(u), L
L
(n)v
1
)
cl
e
i(n2)
=

nZ
(B
z
0
(u), L
R
(n)v
1
)
cl
e
i(n2)
for all 0 < 2. Notice that v
1
can be arbitrary. Therefore, we must have (3.35)
when L(1)u = 0.
In physics, boundary states are usually obtained by solving the equation (3.35). The
solutions of such equation was rst obtained by Ishibashi [I]. They are called Ishibashi
states. The denition of boundary states we gives in (3.34) is more general. Boundary
conditions are also called D-branes in string theory. If u is not a vacuum-like vector,
the boundary state (3.34) associated with u is also very interesting in physics (see for
example [FFFS1]). Such boundary states are associated to the geometry on D-branes.
In Section 5.2, we will give a more natural denition of D-brane (Denition 6.16).
4 Cardy condition
In this section, we derive the Cardy condition from the axioms of open-closed partial
conformal eld theory by writing out the algebraic realizations of the both sides of Figure
3 explicitly. Then we reformulate the Cardy condition in the framework of intertwining
operator algebra. Throughout this section, we x an open-closed eld algebra over V
given in (3.12) equipped with nondegenerate invariant bilinear forms (, )
op
and (, )
cl
.
4.1 The rst version
In the Swiss-cheese dioperad, we exclude an interior sewing operation between two
disks with strips and tubes and a self-sewing operation between two oppositely oriented
boundary punctures on a single disk. The surface obtained after these two types of
sewing operations can be the same cylinder or annulus. The axioms of open-closed
partial conformal eld theory require that the algebraic realization of these two sewing
operations must coincide. This gives a nontrivial condition called Cardy condition (recall
Figure 3).
Although the Cardy condition only involves genus-zero surfaces, its algebraic real-
ization is genus-one in nature. This fact is manifest if we consider the doubling map .
A double of a cylinder is actually a torus. Hence the Cardy condition is a condition on
the equivalence of two algebraic realizations of two dierent decompositions of a torus.
This is nothing but a condition associated to modularity.
That an annulus can be obtained by two dierent sewing operations is also shown in
Figure 4. In particular, the surface (A) in Figure 4 shows how an annulus is obtained
by sewing two oppositely oriented boundary punctures on the same disk with strips and
tubes in S(1, 3[0, 0), and surface (C) in Figure 4, viewed as a propagator of close string,
can be obtained by sewing an element in S(0, 1[1, 0) with an element in S(0, 1[0, 1) along
the interior punctures. We only show in Figure 4 a simple case in which there are only
two boundary punctures and no interior puncture. In general, the number of boundary
28
1
g
(C)
s2
p
s1
p

q
s
q
2 s
q
1
(B)

2
s
s
1
f
0 1
0
1 1
0 1/2
(A)
Figure 4: Cardy condition: two dierent sewings give same annulus.
punctures and interior punctures can be arbitrary. However, all general cases can be
reduced to this simple case by applying associativities. Notice that the two boundary
punctures in this simple case can not be reduced further by the associativities. We only
focus on this case in this work.
The conformal map f between the surface (A) and (B) and g between the surface
(C) and (B) in Figure 4 are given by
f(w) =
1
2i
log w,
g(w) =

2i
log w. (4.1)
It is also useful to know their inverses f
1
(w) = e
2iw
, g
1
(w) = e
2i(
1

)w
. For any
z C, we set q
z
:= e
2iz
and p
z
:= e
2i(
1

)z
. The radius of the outer circle of the surface
(C) in Figure 4 is [p
s
1
[ = 1 and that of inner circle is
[p
s
2
[ = e
i(
1

)
= q
1/2

. (4.2)
As we have argued and will show more explicitly later, Cardy condition deeply
related to modularity. In [H9], Huang introduced the so-called geometrically modied
29
intertwining operators, which is very convenient for the study of modularity. He was
motivated by the fact that it is much easier to study modularity in the global coordinates.
Namely, one should choose the local coordinates at s
1
, s
2
in surface (B) in Figure 4 as
simple as possible. More precisely, we choose the local coordinates at s
1
, s
2
as
w e
i
2
(w s
1
)
w e

i
2
(w s
2
) (4.3)
respectively. Correspondingly, the local coordinates at punctures q
s
1
, q
s
2
are
f
qs
1
(w) = e
i
2
1
2i
log
_
1 +
1
q
s
1
x
_

x=wqs
1
f
qs
2
(w) = e

i
2
1
2i
log
_
1 +
1
q
s
2
x
_

x=wqs
2
(4.4)
respectively.
Notice that both local coordinates f
qs
1
(w) and f
qs
2
(w) are real analytic. Hence
B
(i)
j
R, b
(i)
0
R
+
, i = 1, 2 such that
f
qs
1
(w) = e
P

j=1
B
(1)
j
x
j+1 d
dx
(b
(1)
0
)
x
d
dx
x

x=wqs
1
,
f
qs
2
(w) = e
P

j=1
B
(2)
j
x
j+1 d
dx
(b
(2)
0
)
x
d
dx
x

x=wqs
2
. (4.5)
Then the algebraic realization of the surface (A) gives a map V
op
V
op
C. We assume
1 > [q
s
1
[ > [q
s
2
[ > [q

[ > 0. By the axiom of open-closed conformal eld theory, this

map must be given by (recall (2.25))
v
1
v
2
Tr
Vop
_
Y
op
(T
qs
1
v
1
, q
s
1
)Y
op
(T
qs
2
v
2
, q
s
2
)q
L(0)

_
, (4.6)
where
T
qs
1
= e

j=1
B
(1)
j
L(j)
(b
(1)
0
)
L(0)
,
T
qs
2
= e

j=1
B
(2)
j
L(j)
(b
(2)
0
)
L(0)
. (4.7)
We need rewrite T
qs
1
and T
qs
2
. Let A
j
, j Z
+
, be the complex numbers dened by
log(1 +y) = e
P

j=1
A
j
y
j+1 d
dy
y.
It is clear that A
j
R. Hence we also obtain:
1
2i
log
_
1 +
1
z
x
_
= z
x
d
dx
e
P
jZ
+
A
j
x
j+1 d
dx
(2i)
x
d
dx
x. (4.8)
Lemma 4.1.
T
qs
1
= (q
s
1
)
L(0)
e

j=1
A
j
L(j)
(2i)
L(0)
e

i
2
L(0)
,
T
qs
2
= (q
s
2
)
L(0)
e

j=1
A
j
L(j)
(2i)
L(0)
e
2iL(0)
e
i
2
L(0)
. (4.9)
30
Proof. By results in [BHL], if e
P

j=1
C
j
L(j)
c
L(0)
0
= e
P

j=1
D
j
L(j)
d
L(0)
0
for any C
j
, D
j

C, c
0
, d
0
C, we must have C
j
= D
j
and c
0
= d
0
. Therefore, by moving the factor
(q
s
1
)
L(0)
to the right side of e

j=1
A
j
L(j)
in (4.9) and similarly moving the factor (q
s
2
)
L(0)
in (4.9), we see that it is enough to show
(b
(1)
0
)
L(0)
= (q
s
1
)
L(0)
(2i)
L(0)
e

i
2
L(0)
,
(b
(1)
0
)
L(0)
= (q
s
2
)
L(0)
(2i)
L(0)
e
2iL(0)
e
i
2
L(0)
. (4.10)
Using our conventions (1.1)(1.2), it is easy to check that above identities hold.
Let W be a V -module. Huang introduced the following operator in [H9].
|(x) := x
L(0)
e

j=1
A
j
L(j)
(2i)
L(0)
(EndW)x.
Thus (4.6) can be rewritten as follow:
v
1
v
2
Tr
Vop
_
Y
op
(|(q
s
1
)v

1
, q
s
1
)Y
op
(|(q
s
2
)e
2iL(0)
v

2
, q
s
2
)q
L(0)

_
(4.11)
where v

1
= e

i
2
L(0)
v
1
and v

2
= e
i
2
L(0)
v
2
.
Now we consider the algebraic realization of the surface (C) in Figure 4 obtained
from an interior sewing operation between an element in S(0, 1[1, 0) and an element in
S(0, 1[0, 1).
Lemma 4.2. r (0, 1) R
+
, the surface C in Figure 4 is conformally equivalent to
a surface Q
1
1
I
1
Q
2
, where
1. Q
1
S(0, 1[0, 1) with punctures at z
1
H,

R and local coordinates:
f
z
1
: w
_
p
s
1
r
_
w z
1
w z
1
, (4.12)
f
(1)

: w e
i
2
_

2i
_
log
w z
1
w z
1
; (4.13)
2. Q
2
S(0, 1[1, 0) with punctures at z
2
H,

R with local coordinates
f
z
2
: w
_
r
p
s
2
_
w z
2
w z
2
, (4.14)
f
(2)

: w e

i
2
_

2i
_
log
w z
2
w z
2
. (4.15)
Proof. Let us rst dene two disks D
1
and D
2
. D
1
is the unit disk, i.e. D
1
:= z
C[[z[ 1, which has punctures at p
s
1
, 0 and local coordinates:
g
ps
1
: w e
i
2
_

2i
log w s
1
_
, (4.16)
g
0
: w r
1
w. (4.17)
31
D
2
is the disk z

C[[z[ [p
s
2
[ which has punctures at p
s
2
, and local coordinates:
g
ps
2
: w e

i
2
_

2i
log w s
2
_
, (4.18)
g

: w
r
w
. (4.19)
It is not hard to see that the surface C in Figure 1 can be obtained by sewing the
puncture 0 D
1
with the puncture D
2
according to the usual denition of interior
sewing operation. Then it is enough to show that D
1
and D
2
are conformally equivalent
to P and Q respectively. We dene two maps h
1
: Q
1
D
1
and h
2
: Q
2
D
2
as follow:
h
1
: w p
s
1
w z
1
w z
1
,
h
2
: w p
s
2
w z
2
w z
2
.
It is clear that h
1
and h
2
are both biholomorphic. We can check directly that h
1
and h
2
map punctures to punctures and preserve local coordinates as well.
Using (3.33), we obtain the the algebraic realization the annulus C in Figure 1 as
follow:
v
1
v
2
((T
L
1
T
R
1
)

clop
(z
1
, z
1
)(T
2
v

1
), (T
L
3
T
R
3
)

clop
(z
2
, z
2
)(T
4
v

2
))
cl
(4.20)
where T
L,R
1
, T
2
, T
L,R
3
, T
4
are conformal transformations determined by local coordinates
f
z
1
, f
(2)

, f
(1)
z
2
, f
(2)

, and (T
L
i
T
R
i
)

is the adjoint of T
L
i
T
R
i
with respect to the bilinear
form (, )
cl
for i = 1, 3, and v

1
= e

i
2
L(0)
v
1
and v

2
= e
i
2
L(0)
v
2
.
Lemma 4.3. Recall the convention (1.2), we have
T
L
1
= (z
1
z
1
)
L(0)
e
L(1)
_
p
s
1
r
_
L(0)
, T
R
1
= (z
1
z
1
)
L(0)
e
L(1)
_
p
s
1
r
_L(0)
, (4.21)
T
2
= e
z
1
L(1)
(z
1
z
1
)
L(0)
|(1)
_

_
L(0)
, (4.22)
T
L
3
= ( z
2
z
2
)
L(0)
e
L(1)
_
p
s
2
r
_
L(0)
, T
R
3
= ( z
2
z
2
)
L(0)
e
L(1)
_
p
s
2
r
_L(0)
, (4.23)
T
4
= e
z
2
L(1)
( z
2
z
2
)
L(0)
|(1)
_

_
L(0)
. (4.24)
Proof. From (4.12) and (4.14), we obtain
f
z
1
: w (z
1
z
1
)
x
d
dx
e
x
2 d
dx
_
p
s
1
r
_
x
d
dx
x

x=wz
1
f
z
2
: w ( z
2
z
2
)
x
d
dx
e
x
2 d
dx
_
p
s
2
r
_
x
d
dx
x

x=wz
2
(4.25)
Then (4.21) and (4.23) is obvious. Notice that the expression (4.21) is independent of
our choice of branch cut as long as we keep the convention (1.2).
32
From (4.13) and (4.15), we obtain
f
(1)

(w) = e
z
1
x
2 d
dx
(z
1
z
1
)
x
d
dx
e
P

j=1
A
j
x
j+1 d
dx
(2i)
x
d
dx
_

_
x
d
dx
e
i
2
x
d
dx
x

x=
1
w
(4.26)
f
(2)

(w) = e
z
2
x
2 d
dx
( z
2
z
2
)
x
d
dx
e
P

j=1
A
j
x
j+1 d
dx
(2i)
x
d
dx
_

_
x
d
dx
e

i
2
x
d
dx
x

x=
1
w
(4.27)
Recall that f
(1)

, f
(2)

are both real analytic. Similar to the proof of Lemma 4.1, to show
(4.22) and (4.24) is enough to show that
(b
(1)

)
L(0)
= (z
1
z
1
)
L(0)
(2i)
L(0)
_

_
L(0)
e

i
2
L(0)
(4.28)
(b
(2)

)
L(0)
= ( z
2
z
2
)
L(0)
(2i)
L(0)
_

_
L(0)
e
i
2
L(0)
(4.29)
for some b
(1)

, b
(2)

R
+
. Using our convention (1.1) and (1.2), it is a direct check that
(4.28) and (4.29) holds.
Combining (4.11), (4.20) and additional natural factors q

c
24

, q

c
24

(see [Z][H9]),
which is due to determinant line bundle on torus [Se1][Kr], we obtain the following
formulation of the Cardy condition:
Denition 4.4. The open-closed eld algebra over V given in (3.12) and equipped with
nondegenerate bilinear forms (, )
op
and (, )
cl
is said to satisfy Cardy condition if the
left hand sides of the following formula, z
1
, z
2
H, v
1
, v
2
V
op
,
Tr
Vop
_
Y
op
(|(q
s
1
)v
1
, q
s
1
)Y
op
(|(q
s
2
)e
2iL(0)
v
2
, q
s
2
)q
L(0)c/24

_
=
_
(T
L
1
T
R
1
)

clop
(z
1
, z
1
)(T
2
v
1
), q
c/24

(T
L
3
T
R
3
)

clop
(z
2
, z
2
)(T
4
v
2
)
_
cl
(4.30)
converge absolutely when 1 > [q
s
1
[ > [q
s
2
[ > [q

[ > 0, and the right hand side of (4.30)

converge absolutely for all s
1
, s
2
H satisfying Re s
1
= 0, Re s
2
=
1
2
. Moreover, the
equality (4.30) holds when 1 > [q
s
1
[ > [q
s
2
[ > [q

[ > 0.
Remark 4.5. The dependence of z
1
, z
2
, r of the right hand side of (4.30) is supercial
as required by the independence of z
1
, z
2
, r of the left hand side of (4.30). We will see it
more explicitly later.
Using the denition of boundary states (3.34), (4.30) can also be written as follow:
Tr
Vop
_
Y
op
(|(q
s
1
)v
1
, q
s
1
)Y
op
(|(q
s
2
)e
2iL(0)
v
2
, q
s
2
)q
L(0)c/24

_
=
_
B
z
1
(T
2
v
1
),
_
p
s
2
p
s
1
_
L(0)

_
p
s
2
p
s
1
_
L(0)
q
c/24

B
z
2
(T
4
v
2
)
_
cl
. (4.31)
33
4.2 The second version
In this subsection, we rewrite the Cardy condition (4.30) in the framework of intertwining
operator algebra.
Since V satises the conditions in Theorem 1.1, it has only nite number of in-
equivalent irreducible modules. Let 1 be the set of equivalence classes of irreducible
V -modules. We denote the equivalence class of the adjoint module V as e, i.e. e 1.
Let W
a
be a chosen representative of a 1.
For any V -module (W, Y
W
), we denote the graded dual space of W as W

, i.e. W

nC
(W
(n)
)

. There is a contragredient module structure on W

[FHL] given by a vertex

operator Y

W
, which is dened as follow
Y

W
(u, x)w

, w) := w

, Y
W
(e
xL(1)
x
2L(0)
u, x
1
)w) (4.32)
for u V, w W, w

. (W

, Y

W
) (or simply W) is the only module structure on W

we use in this work. So we can set Y

W
:= Y

W
. We denote the equivalent class of W

a
as a

. It is harmless to set W

a
= W
a
. Moreover, W

is canonically identied with W.

Hence a

= a for a 1.
By assumption on V , V

= V , i.e. e

= e. From [FHL], there is a nondegenerate

invariant bilinear form (, ) on V such that (1, 1) = 1. This bilinear form species a
unique isomorphism from V to V

. In the rest of this work, we identify V

with V using
this isomorphism without mentioning it explicitly.
For any triple V -modules W
1
, W
2
, W
3
, we have isomorphisms

r
: 1
W
3
W
1
W
2
1
W
3
W
2
W
1
, r Z
given as follow:

r
()(w
2
, z)w
1
= e
zL(1)
(w
1
, e
(2r+1)i
z)w
2
, (4.33)
for 1
W
3
W
1
W
2
and w
i
W
i
, i = 1, 2. The following identity

r

r1
=
r1

r
= id (4.34)
is proved in [HL3].
For 1
W
3
W
1
W
2
and r Z, a so-called r-contragredient operator A
r
() was introduced
in [HL3]. Here, we use two slightly dierent operators

A
r
() and

A
r
() introduced in
[Ko1] and dened as follow:

A
r
()(w
1
, e
(2r+1)i
x)w

3
, w
2
) = w

3
, (e
xL(1)
x
2L(0)
w
1
, x
1
)w
2
),

A
r
()(w
1
, x)w

3
, w
2
) = w

3
, (e
xL(1)
x
2L(0)
w
1
, e
(2r+1)i
x
1
)w
2
), (4.35)
for 1
W
3
W
1
W
2
and w
1
W
1
, w
2
W
2
, w

3
W

3
. In particular, when W
1
= V and
W
2
= W
3
= W, we have

A
r
(Y
W
) =

A
r
(Y
W
) = Y

W
= Y
W
, r Z. If 1
W
3
W
1
W
2
, then

A
r
(),

A
r
() 1
W

2
W
1
W

3
for r Z and

A
r


A
r
() =

A
r


A
r
() = . (4.36)
34
Let 1
a
3
a
1
a
2
. We dene
123
:=
r


A
r
. It is easy to see that
w

a
3
, (w
a
1
, x)w
a
2
)
= e
x
1
L(1)

123
()(w

a
3
, x
1
)e
xL(1)
x
2L(0)
w
a
1
, w
a
2
) (4.37)
for w
a
1
W
a
1
, w
a
2
W
a
2
, w

a
3
W

a
3
. It is also clear that
123
is independent of r Z.
It is proved in [Ko1] that
3
123
= id
V
a
3
a
1
a
2
. We also denote
1
123
as
132
. Clearly, we have

132
=

A
r

r1
and

132
()(w
1
, x)w

3
, w
2
)
= w

3
, e
x
1
L(1)
(w
2
, x
1
)e
xL(1)
x
2L(0)
w
1
) (4.38)
for w
a
1
W
a
1
, w
a
2
W
a
2
, w

a
3
W

a
3
.
For any V -module W, we dene a V -module map
W
: W W by

W
: w e
2iL(0)
w. (4.39)
For W
a
, we have
Wa
= e
2iha
id
Wa
where h
a
C is the lowest conformal weight of W
a
.
We denote the graded dual space of V
cl
and V
op
by V

cl
and V

op
respectively. Let

cl
: V
cl
V

cl
and
op
: V
op
V

op
be the isomorphisms induced from (, )
cl
and (, )
op
respectively. Namely, we have
(u
1
, u
2
)
cl
=
cl
(u
1
), u
2
)
(v
1
, v
2
)
op
=
op
(v
1
), v
2
) (4.40)
for u
1
, u
2
V
cl
and v
1
, v
2
V
op
.
V
cl
as a conformal full eld algebra over V V can be expanded as follow:
V
cl
=
N
cl
i=1
W
r
L
(i)
W
r
R
(i)
(4.41)
where r
L
, r
R
: 1, . . . , N
cl
1. For a 1, we choose a basis e
a;

N
of W
a
and a
dual basis e

a;

N
of W

a
. Then
e
r
L
(i),
e
r
R
(i),

i=1,...,N
cl
,,N
(4.42)
is a basis of V
cl
and

1
cl
(e

r
L
(i);
e

r
R
(i),
)
i=1,...,N
cl
,,N
(4.43)
is its dual basis with respect to the nondegenerate bilinear form (, )
cl
.
Let T : (
V V
(
V
be the tensor bifunctor. We showed in [Ko2] that there is a
morphism
clop
: T(V
cl
) V
op
in (
V
(see (3.81),(3.82) in [Ko2] for denition). We
dene a morphism

clop
: T(V

cl
) V
op
as a composition of maps as follow:

clop
: T(V

cl
)
T(
1
cl
)
T(V
cl
)

clop
V
op
. (4.44)
By the universal property of tensor product ([HL1]-[HL4]), 1
W
3
W
1
W
2
and Hom
V
(W
1

W
2
, W
3
) for any three V -modules W
1
, W
2
, W
3
are canonically isomorphic. Given a mor-
phism m Hom
V
(W
1
W
2
, W
3
), we denote the corresponding intertwining operator
as
m
. Conversely, given an intertwining operator , we denote its corresponding mor-
phism as m
Y
. Therefore, we have two intertwining operators

clop
and

clop
corre-
sponding to morphisms
clop
and

clop
respectively.
35
Lemma 4.6. For z H, we have

clop
(z, z)(e
r
L
(i),
e
r
R
(i),
) = e
zL(1)

clop
(e
r
L
(i),
, z z)e
r
R
(i),
. (4.45)
Proof. It is proved in [Ko2] that
m
Y
clop
= m
Y
f
op
(
clop
id
Vop
). (4.46)
Using (4.46), when z H, H and [[ > [z [ > 0, we have

clop
(z, )(e
r
L
(i),
e
r
R
(i),
) = Y
clop
(e
r
L
(i),
e
r
R
(i),
; z, )1
op
= Y
f
op
(

clop
(e
r
L
(i),
, z )e
r
R
(i),
, )1
= e
L(1)

clop
(e
r
L
(i),
, z )e
r
R
(i),
. (4.47)
By the convergence property of the iterate of two intertwining operators, the right hand
side of (4.47) is a power series of absolutely convergent for [[ > [z [ > 0 . By
the property of power series, the right hand side of (4.47) must converge absolutely for
all z H, H. Because analytic extension in a simply connected domain is unique,
we obtain that the equality (4.47) holds for all z H, H. When = z, we obtain
(4.45).
Now we consider the both sides of Cardy condition (4.30) for an open-closed eld
algebra over V . On the left hand side of (4.30), we have q
s
2
< 0. Using (2.25) and
(4.33), we obtain, v
3
V
op
,
Y
op
(|(q
s
2
)v
2
, q
s
2
)v
3
= e
|qs
2
|L(1)
Y
op
(v
3
, [q
s
2
[) |(q
s
2
)v
2
=
1
(Y
f
op
)(|(q
s
2
)v
2
, e
i
[q
s
2
[)v
3
. (4.48)
Hence we can rewrite the left hand side of (4.30) as follow:
Tr
Vop
_
Y
f
op
(|(q
s
1
)v
1
, q
s
1
)
1
(Y
f
op
)(|(q
s
2
)e
2iL(0)
v
2
, e
i
[q
s
2
[)q
L(0)c/24

_
(4.49)
for q
s
1
> [q
s
2
[ > [q

[ > 0.
We have the following result for the right hand side of (4.30).
Proposition 4.7. For s
1
, s
2
H, Re s
1
= 0, Re s
2
= 0,
((T
L
1
T
R
1
)

clop
(z
1
, z
1
)(T
2
v
1
), q

c
24

(T
L
3
T
R
3
)

clop
(z
2
, z
2
)(T
4
v
2
))
cl
=
N
cl

i=1
Tr
W
r
R
(i)
e
2iL(0)

1
_
|(q

s
1
)
_

_
L(0)
v
1
, q

s
1
_

2
_
|(q

s
2
)
_

_
L(0)
v
2
, q

s
2
_
q
L(0)
c
24

(4.50)
where
1
and
2
are intertwining operators of types
_
W
r
R
(i)
VopW

r
L
(i)
_
and
_
W

r
L
(i)
VopW
r
R
(i)
_
respectively
and are given by

1
=
123
(

clop
) (
op
id
W

r
L
(i)
), (4.51)

2
=
0
(
132
(

clop
)) (
op
id
W
r
R
(i)
). (4.52)
36
Proof. Let z
3
:= z
1
z
1
and z
4
:= z
2
z
2
. By (3.28) and (4.45), the left hand side
of (4.50) equals to
((T
L
1
T
R
1
)

clop
(z
1
, z
1
)(T
2
v
1
), (T
L
3
T
R
3
)

clop
(z
2
, z
2
)(T
4
v
2
))
cl
=
N
cl

i=1

,
_
(T
L
3
T
R
3
)

clop
(z
2
, z
2
)(T
4
v
2
), e
r
L
(i),
e
r
R
(i),
_
cl
_

1
cl
(e

r
L
(i),
e

r
R
(i),
), (T
L
1
T
R
1
)

clop
(z
1
, z
1
)(T
2
v
1
)
_
cl
=
N
cl

i=1

,
_
T
4
v
2
,
clop
(z
2
, z
2
)(T
L
3
e
r
L
(i),
T
R
3
e
r
R
(i),
)
_
op
_

clop
(z
1
, z
1
)(
1
cl
(T
L
1
e

r
L
(i),
T
R
1
e

r
R
(i),
)), T
2
v
1
_
op
=
N
cl

i=1

,
_
T
4
v
2
, e
z
2
L(1)

clop
(T
L
3
e
r
L
(i),
, z
2
z
2
)T
R
3
e
r
R
(i),
_
op
_
e
z
1
L(1)

clop
(T
L
1
e

r
L
(i),
, z
1
z
1
)T
R
1
e

r
R
(i),
, T
2
v
1
_
op
=
N
cl

i=1

,

op
(T
4
v
2
), e
z
2
L(1)

clop
(T
L
3
e
r
L
(i),
, z
2
z
2
)T
R
3
e
r
R
(i),
)
e
z
1
L(1)

clop
(T
L
1
e

r
L
(i),
, z
1
z
1
)T
R
1
e

r
R
(i),
,
op
(T
2
v
1
) )
=
N
cl

i=1

,
e
z
1
4
L(1)

123
(

clop
)(
op
(e
z
2
L(1)
T
4
v
2
), z
1
4
)
z
2L(0)
4
e
z
1
4
L(1)
T
L
3
e
r
L
(i),
, T
R
3
e
r
R
(i),
)
T
R
1
e

r
R
(i),
, e
z
1
3
L(1)

123
(

clop
)(
op
(e
z
1
L(1)
T
2
v
1
), z
1
3
)
z
2L(0)
3
e
z
1
3
L(1)
T
L
1
e

r
L
(i),
)
=
N
cl

i=1

,

123
(

clop
)(
op
(e
z
2
L(1)
T
4
v
2
), z
1
4
)z
2L(0)
4
e
z
1
4
L(1)
T
L
3
e
r
L
(i),
,
e
z
1
4
L(1)
T
R
3
e
r
R
(i),
) e

r
R
(i),
, (T
R
1
)

e
z
1
3
L(1)

123
(

clop
)(
op
(e
z
1
L(1)
T
2
v
1
), z
1
3
)z
2L(0)
3
e
z
1
3
L(1)
T
L
1
e

r
L
(i),
)
=
N
cl

i=1

,
e
r
L
(i),
, (T
L
3
)

e
z
1
4
L(1)
z
2L(0)
4

A
0

123
(

clop
)(e
z
1
4
L(1)
z
2L(0)
4

op
(e
z
2
L(1)
T
4
v
2
), e
i
z
4
)e
z
1
4
L(1)
T
R
3
e
r
R
(i),
) e

r
R
(i),
, (T
R
1
)

e
z
1
3
L(1)

123
(

clop
)(
op
(e
z
1
L(1)
T
2
v
1
), z
1
3
)z
2L(0)
3
e
z
1
3
L(1)
T
L
1
e

r
L
(i),
) (4.53)
We dene two intertwining operators as follow:

(0)
1
=
123
(

clop
) (
op
id
W

r
L
(i)
),

(0)
2
=

A
0
(
123
(

clop
)) (
op
id
W

r
R
(i)
). (4.54)
37
Using (4.21),(4.22),(4.23) and (4.24), we further obtain that the left hand side of (4.53)
equals to
N
cl

i=1

,
e
r
L
(i),
,
_
p
s
2
r
_
L(0)
( z
2
z
2
)
L(0)
z
2L(0)
4

(0)
2
(z
2L(0)
4
( z
2
z
2
)
L(0)
|(1)
_

_
L(0)
v
2
, e
i
z
4
)
z
2
z
2
L(0)
_
p
s
2
r
_L(0)
e
r
R
(i),
)e

r
R
(i),
,
_
p
s
1
r
_L(0)
(z
1
z
1
)
L(0)

(0)
2
(z
L(0)
3
|(1)
_

_
L(0)
v
1
, z
1
3
)z
L(0)
3
_
p
s
1
r
_
L(0)
e

r
L
(i),
)
=
N
cl

i=1
Tr
W

r
L
(i)

(0)
2
(|(p
s
2
)(1/)
L(0)
v
2
, E
1
)E
2

(0)
1
(|(p
s
1
)(1/)
L(0)
v
1
, p
s
1
) (4.55)
where E
1
= p
s
2
( z
2
z
2
)z
2
4
e
i
z
4
and
E
2
= p
L(0)
s
2
( z
2
z
2
)
L(0)
z
2L(0)
4
( z
2
z
2
)
L(0)
p
s
2
L(0)
p
s
1
L(0)
z
1
z
1
L(0)
z
L(0)
3
p
L(0)
s
1
.
For E
1
, since z
1
4
is obtained by operations on intertwining operator where z
4
is treated
formally, z
1
4
really means [z
4
[
1
e
i
i
2
. Therefore, we have
E
1
= p
s
2
e
3i
2
e
i
e
i
e
i
2
= e
2i
p
s
2
. (4.56)
For E
2
, keep in mind (1.1) and (1.2), we have
E
2
=

p
s
2
p
s
1

2L(0)
( z
2
z
2
)
L(0)
(z
2
z
2
)
2L(0)
( z
2
z
2
)
L(0)
(z
1
z
1
)
L(0)
(z
1
z
1
)
L(0)
= q
L(0)

e
3i
2
L(0)
e
2
i
2
L(0)
e

3i
2
L(0)
e

i
2
L(0)
e

i
2
L(0)
= q
L(0)

e
2iL(0)
. (4.57)
Therefore, we obtain that the left hand side of (4.53) further equals to
N
cl

i=1
Tr
W

r
L
(i)

2
0
(
(0)
2
)
_
|(p
s
2
)
_

_
L(0)
v
2
, p
s
2
_
q
L(0)

e
2iL(0)

(0)
1
_
|(p
s
1
)
_

_
L(0)
v
1
, p
s
1
_
, (4.58)
where we have used the fact that (, e
2i
x) =
2
0
()(, x) for any intertwining operator
. By using the property of trace, it is easy to see that (4.58) multiplying q

c
24

is nothing
but the right hand side of (4.50).
38
Remark 4.8. It is easy to check that the absolute convergence of the left hand side of
(4.53) by our assumption easily implies the absolute convergence of each step in (4.53).
Notice that the absolute convergence of the right side of (4.50) is automatic because V
is assumed to satisfy the conditions in Theorem 1.1. Hence, by tracing back the steps
in above proof, we see that the absolute convergence of the left hand side of (4.50) is
also automatic.
Now we recall some results in [H9][H10]. We follow the notations in [HKo3]. We
denote the unique analytic extension of
Tr
Wa
1

a
1
;(1)
aa
1
;i
(|(e
2iz
)w
a
, e
2iz
)q
L(0)
c
24

in the universal covering space of 1 > [q

[ > 0 as
E(Tr
Wa
1

a
1
;(1)
aa
1
;i
(|(e
2iz
)w
a
, e
2iz
)q
L(0)
c
24

).
By [Mi2][H9], above formula is independent of z. Consider the map: for w
a
W
a
,

1
(
a
1
;(1)
aa
1
;i
) : w
a
E(Tr
Wa
1

a
1
;(1)
aa
1
;i
(|(e
2iz
)w
a
, e
2iz
)q
L(0)
c
24

). (4.59)
We denote the right hand side of (4.59) as
1
(
a
1
;(1)
aa
1
;i
)(w
a
; z, ). Notice that we choose
to add z in the notation even though it is independent of z. We dene an action of
SL(2, Z) on the map (4.59) as follow:
__
a b
c d
_
(
1
(
a
1
;(1)
aa
1
;i
))
_
(w
a
; z, ) (4.60)
= E
_
Tr
Wa
1

a
1
;(1)
aa
1
;i
_
|(e
2iz

)
_
1
c + d
_
L(0)
w
a
, e
2iz

_
q
L(0)
c
24

_
where

=
a+b
c+d
and z

=
z
c+d
, for
_
a b
c d
_
SL(2, Z) and w
a
W
a
. The following
Theorem is proved in [Mi2][H7].
Theorem 4.9. There exists a unique A
ij
a
2
a
3
C for a
2
, a
3
1 such that, for w
a
W
a
,
E
_
Tr
Wa
1

a
1
;(1)
aa
1
;i
_
|(e
2iz

)
_
1
c + d
_
L(0)
w
a
, e
2iz

_
q
L(0)
c
24

_
=

a
3
I
A
ij
a
1
a
2
E(Tr
Wa
1

a
2
;(2)
aa
2
;j
(|(e
2iz
)w
a
, e
2iz
)q
L(0)
c
24

) (4.61)
where

=
a+b
c+d
and z

=
z
c+d
.
In particular, the action of S =
_
0 1
1 0
_
on (4.59) induces, for each a 1, an
automorphism on
a
1
I
1
a
1
aa
1
, denoted as S(a). Namely, we have
S(
1
(
a
1
;(1)
aa
1
;i
)) =
1
(S(a)(
a
1
;(1)
aa
1
;i
)) (4.62)
39
Combining all such S(a), we obtain an automorphism on
a,a
1
I
1
a
1
aa
1
. We still denote it
as S, i.e. S =
aI
S(a). Then S can be further extended to a map on
a,a
3
I
1
a
3
aa
3
1
a
a
1
a
2
given as follow:
S(
a
3
;(1)
aa
3
;i

a;(2)
a
1
a
2
;j
) := S(
a
3
;(1)
aa
3
;i
)
a;(2)
a
1
a
2
;j
. (4.63)
There is a fusing isomorphism map [H4]:
T :
aI
1
a
4
a
1
a
1
a
a
2
a
3

=

bI
1
a
4
ba
3
1
b
a
1
a
2
(4.64)
for a
1
, a
2
, a
3
, a
4
1. Using the isomorphism T, we obtain a natural action of S on

b,a
3
1
a
3
a
1
b
1
b
a
2
a
3
.
It is shown in [H7] that the following 2-points genus-one correlation function, for
a
1
, a
2
, a
3
, a
4
1, i = 1, . . . , N
a
1
aa
1
, j = 1, . . . , N
a
a
2
a
3
and w
a
k
W
a
k
, k = 2, 3,
Tr
Wa
1

a
1
;(1)
aa
1
;i
(|(q
z
2
)
a;(2)
a
2
a
3
;j
(w
a
2
, z
1
z
2
)w
a
3
, q
z
2
)q
L(0)
c
24

(4.65)
is absolutely convergent when 1 > [e
2iz
2
[ > [q

[ > 0 and 1 > [e

2i(z
1
z
2
)
[ > 0 and single-
valued in the chosen branch. It can be extended uniquely to a single-valued analytic
function on the universal covering space of
M
2
1
= (z
1
, z
2
, ) C
3
[z
1
,= z
2
+ p + q, p, q Z, H.
This universal covering space is denoted by

M
2
1
. We denote this single-valued analytic
function on

M
2
1
as
E(Tr
Wa
1

a
1
;(1)
aa
1
;i
(|(q
z
2
)
a;(2)
a
2
a
3
;j
(w
a
2
, z
1
z
2
)w
a
3
, q
z
2
)q
L(0)
c
24

).
We denote the space spanned by such functions on

M
2
1
by G
1;2
.
For
a
1
;(1)
aa
1
;i
1
a
1
aa
1
and
a;(2)
a
2
a
3
;j
1
a
a
2
a
3
, we now dene the following linear map:

2
(
a
1
;(1)
aa
1
;i

a;(2)
a
2
a
3
;j
) :
b
2
,b
3
I
W
b
2
W
b
3
G
1;2
as follow: the map restricted on W
b
2
W
b
3
is dene by 0 for b
2
,= a
2
or b
3
,= a
3
, and by
E(Tr
Wa
1

a
1
;(1)
aa
1
;i
(|(q
z
2
)
a;(2)
a
2
a
3
;j
(w
a
2
, z
1
z
2
)w
a
3
, q
z
2
)q
L(0)
c
24

), (4.66)
for all w
a
k
W
a
k
, k = 2, 3. The following identity was proved in [H9].
_

2
(
a
1
;(1)
aa
1
;i

a;(2)
a
2
a
3
;j
)
_
_

_
L(0)
w
a
2

_

_
L(0)
w
a
3
__
_

z
1
,
1

z
2
;
1

_
=
_

2
(S(
a
1
;(1)
aa
1
;i

a;(2)
a
2
a
3
;j
))(w
a
2
w
a
3
)
_
(z
1
, z
2
, ). (4.67)
One can also produce 2-point genus-one correlation functions from a product of two
intertwining operators. It is proved in [H7] that w
a
k
W
a
k
, k = 1, 2,
Tr
Wa
4

a
4
;(1)
a
1
a
3
;i
(|(q
z
1
)w
a
1
, q
z
1
)
a
3
;(2)
a
2
a
4
;j
(|(q
z
2
)w
a
2
, q
z
2
)q
L(0)
c
24

, (4.68)
40
is absolutely convergent when 1 > [q
z
1
[ > [q
z
2
[ > [q

[ > 0. (4.68) has a unique extension

to the universal covering space

M
2
1
, denoted as
E(Tr
Wa
4

a
4
;(1)
a
1
a
3
;i
(|(q
z
1
)w
a
1
, q
z
1
)
a
3
;(2)
a
2
a
4
;j
(|(q
z
2
)w
a
2
, q
z
2
)q
L(0)
c
24

). (4.69)
Such functions on

M
2
1
also span G
1;2
. We dene a map

2
(
a
4
;(1)
a
1
a
3
;i

a
3
;(2)
a
2
a
4
;j
) :
b
1
,b
2
I
W
b
1
W
b
2
G
1;2
(4.70)
as follow: the map restrict on W
b
1
W
b
2
is dened by 0 for b
1
,= a
1
, b
2
,= a
2
, and by
(4.69) for w
a
1
W
a
1
, w
a
2
W
a
2
.
It was proved by Huang in [H7] that the fusing isomorphism (4.64) gives the following
associativity:
E(Tr
Wa
4

a
4
;(1)
a
1
a
3
;i
(|(q
z
1
)w
a
1
, q
z
1
)
a
3
;(2)
a
2
a
4
;j
(|(q
z
2
)w
a
2
, q
z
2
)q
L(0)
c
24

)
=

a
5
I

k,l
F(
a
4
;(1)
a
1
a
3
;i

a
3
;(2)
a
2
a
4
;j
,
a
4
;(3)
a
5
a
4
;k

a
5
;(4)
a
1
a
2
;l
)
E(Tr
Wa
4

a
4
;(3)
a
5
a
4
;k
(|(q
z
2
)
a
5
;(4)
a
1
a
2
;l
(w
a
1
, z
1
z
2
)w
a
2
, q
z
2
)q
L(0)
c
24

), (4.71)
where F(
a
4
;(1)
a
1
a
3
;i

a
3
;(2)
a
2
a
4
;j
,
a
4
;(3)
a
5
a
4
;k

a
5
;(4)
a
1
a
2
;l
) is the matrix representation of T in the basis

a
4
;(1)
a
1
a
3
;i

a
3
;(2)
a
2
a
4
;j

i,j
,
a
4
;(3)
a
5
a
4
;k

a
5
;(4)
a
1
a
2
;l

k,l
. Therefore, w
a
k
W
a
k
, k = 1, 2, we obtain
_

2
(
a
3
;(1)
a
1
a;i

a;(2)
a
2
a
3
;j
)
_
_

_
L(0)
w
a
1

_

_
L(0)
w
a
2
__
_

z
1
,
1

z
2
;
1

_
=
_

2
(S(
a
3
;(1)
a
1
a;i

a;(2)
a
2
a
3
;j
))(w
a
1
w
a
2
)
_
(z
1
, z
2
, ). (4.72)
Combining (4.49), (4.50), (4.51), (4.52) and (4.72) and Remark 4.8, we obtain a
simpler version of Cardy condition.
Theorem 4.10. The Cardy condition can be rewritten as follow:
_

W
r
R
(i)

123
(

clop
) (
op
id
W

r
L
(i)
)
_

0
(
132
(

clop
)) (
op
id
W
r
R
(i)
)
_
= S
1
_
Y
f
op

_

1
(Y
f
op
) (
Vop
id
Vop
)
__
. (4.73)
5 Modular tensor categories
This section is independent of the rest of this work. The tensor product theory of
modules over a vertex operator algebra is developed by Huang and Lepowsky [HL1]-
[HL4][H2]. In particular, the notion of vertex tensor category is introduced in [HL1].
Huang later proved in [H11] that (
V
is a modular tensor category for V satisfying
conditions in Theorem 1.1. In Section 4.1, we review some basic ingredients of modular
tensor category (
V
. In Section 4.2, we show how to nd in (
V
a graphical representation
of the modular transformation S :
1

discussed in Section 3.2.

41
5.1 Preliminaries
We recall some ingredients of vertex tensor category (
V
and those of modular tensor
category structure on (
V
constructed in [H11].
There is an associativity isomorphism /
/ : W
1
(W
2
W
3
) (W
1
W
2
) W
3
for each triple of V -modules W
1
, W
2
, W
3
. The relation between the fusing isomorphism
T (recall (4.64)) and the associativity isomorphism / in (
V
is described by the following
commutative diagram:

a
5
I
1
a
4
a
1
a
5
1
a
5
a
2
a
3

=
//
F

Hom
V
(W
a
1
(W
a
2
W
a
3
), W
a
4
)
(A
1
)

a
6
I
1
a
4
a
6
a
3
1
a
6
a
1
a
2

=
//
Hom
V
((W
a
1
W
a
2
) W
a
3
, W
a
4
)
(5.1)
where the two horizontal maps are canonical isomorphisms induced from the universal
property of .
We recall the braiding structure on (
V
. For each pair of V -modules W
1
, W
2
, there is
also a natural isomorphism, for z > 0, 1
P(z)
+
: W
1

P(z)
W
2
W
2

P(z)
W
1
, dened by
1
P(z)
+
(w
1

P(z)
w
2
) = e
L(1)
T

+
(w
2

P(z)
w
1
), (5.2)
where
+
is a path from z to z inside the lower half plane as shown in the following
graph
z 0 z
. (5.3)
The inverse of 1
P(z)
+
is denoted by 1
P(z)

, which is characterized by
1
P(z)

(w
2

P(z)
w
1
) = e
L(1)
T

(w
1

P(z)
w
2
), (5.4)
where

is a path in the upper half plane as shown in the following graph

z 0 z . (5.5)
We denote 1
P(1)

simply as 1

. The natural isomorphisms 1

give (
V
two dierent
braiding structures. We choose 1
+
as the default braiding structure on (
V
. Sometimes,
we will denote it by ((
V
, 1
+
) to emphasis our choice of braiding isomorphisms.
42
Notice that our choice of 1

follows that in [Ko2], which is dierent from that in

[H8][H11][Ko1]. For each V -module W, (4.39) denes an automorphism
W
: W W
called a twist. A V -module W is said to have a trivial twist if
W
= id
W
. The twist
and braiding 1
+
satisfy the following three balancing axioms

W
1
W
2
= 1
+
1
+
(
W
1

W
2
) (5.6)

V
= id
V
(5.7)

W
= (
W
)

. (5.8)
for any pair of V -modules W
1
, W
2
.
Let
a
ea
be a basis of 1
a
ea
for all a 1 such that it coincides with the vertex
operator Y
Wa
, which denes the V -module structure on W
a
, i.e.
a
ea
= Y
Wa
. We choose
a basis
a
ae
of 1
a
ae
as follow:

a
ae
=
1
(
a
ea
). (5.9)
We also choose a basis
e
aa
of 1
e
aa
as

e
aa
=
e

aa
=

A
0
(
a
ae
) =
132
(
a
ea
). (5.10)
Notice that these choices are made for all a 1. In particular, we have

e
=
1
(
a

ea
),
e
a

a
=
e

a
=

A
0
(
a

e
).
The following relation was proved in [Ko1].

e
a

a
= e
2iha

0
(
e
aa
) = e
2iha

1
(
e
aa
). (5.11)
For any V -modules W
1
, W
2
, W
3
and 1
W
3
W
1
W
2
, we denote by m
Y
the morphism in
Hom
V
(W
1
W
2
, W
3
) associated to under the identication of two spaces induced by
the universal property of .
Now we recall the construction of duality maps [H11]. We will follow the convention
in [Ko1]. Since (
V
is semisimple, we only need to discuss irreducible modules. For a 1,
the right duality maps e
a
: W

a
W
a
V and i
a
: V W
a
W

a
for a 1 are given by
e
a
= m
Y
e
a

a
, m
Y
e
aa

i
a
= dima id
V
where dima ,= 0 for a 1 (proved by Huang in [H10]). The left duality maps e

a
:
W
a
W

a
V and i

a
: V W

a
W
a
are given by
e

a
= m
Y
e
aa

, m
Y
e
a

a
i

a
= dima id
V
.
In a ribbon category, there is a powerful tool called graphic calculus. One can express
various morphisms in terms of graphs. In particular, the right duality maps i
a
and e
a
are denoted by the following graphs:
i
a
=
a a

,
e
a
=
a a

,
43
the left duality maps are denoted by
i

a
=
a

a
,
e

a
=
a

a ,
and the twist and its inverse, for any object W, are denoted by

W
=
W
W ,

1
W
=
W
W .
The identity (5.11) proved in [Ko2] is nothing but the following identity:
a

a
=
a a

=
a a

. (5.12)
This formula (5.12) is implicitly used in many graphic calculations in this work.
A basis
a
3
;(1)
a
1
a
2
;i

N
a
3
a
1
a
2
i=1
of 1
a
3
a
1
a
2
for a
1
, a
2
, a
3
1 induces a basis e
a
3
a
1
a
2
;i
of Hom(W
a
1

W
a
2
, W
a
3
). One can also denote e
a
3
a
1
a
2
;i
as the following graph:
a
1
a
2
a
3
i
.
(5.13)
Note that we will always use a to represent W
a
and a

to represent W

a
in graph for
simplicity. By the universal property of
P(z)
, the map
0
: 1
a
3
a
1
a
2
1
a
3
a
2
a
1
induces a
linear map
0
: Hom
V
(W
a
1
W
a
2
, W
a
3
) Hom
V
(W
a
2
W
a
1
, W
a
3
) given as follow:

0
:
a
1
a
2
a
3
i

a
2
a
1
a
3
i
. (5.14)
Let us choose a basis f
a
1
a
2
a
3
;j

N
a
3
a
1
a
2
j=1
of Hom
V
(W
a
3
, W
a
1
W
a
2
), denoted by
f
a
1
a
2
a
3
;j
=
a
1
a
2
a
3
j
, (5.15)
such that
a
1
a
2
a
3
a
3
i
j
=
ij
a
3
. (5.16)
44
The following identity is proved in [Ko1]
a
1
a
2
=

a
3
I
N
a
3
a
1
a
2

i=1
a
1
a
2
a
1
a
2
a
3
i
i
. (5.17)
We prove a similar identity below.
Lemma 5.1.

a
4
I

l
dima
4
dimb
a

3
a

3
b
b
a
4
k
k
=
a

3
b (5.18)
Proof. Using the rst balancing axiom (5.6), we have
a
3
a
4
b
l
= a
3
a
4
b
l

1
=
a
3
a
4
b
l
= a
3
a
4
b
l
.
(5.19)
Then the Lemma follows from the following relations:
dima
4
dimb
a

3
b
a
4
a
4
k
l
=
1
dimb
a

3
a

4
b
k
l
a
4
(5.20)
=
1
dimb
a

3
a
4
b
l
k
a
4
=
kl
a
4
. (5.21)
Similar to
0
,

A
0
,
123
and
132
can also be described graphically as proved in [Ko1].
We recall these results below.
Proposition 5.2.

A
0
:
a
1
a
2
a
3
i

a
1
a

3
a

2
i
. (5.22)
45

123
:
a
1
a
2
a
3
i

a

3
a
1
a

2
i
, (5.23)

132
:
a
1
a
2
a
3
i

a

1
a
2
a

3
i
. (5.24)
5.2 Graphical representation of S :
1

In [HKo3], we dened an action of , on

2
(
a
1
;(1)
aa
1
;p

a;(2)
a
2
a
3
;q
). More precisely,
(
2
(
a
1
;(1)
aa
1
;p

a;(2)
a
2
a
3
;q
)) :
a
2
,a
3
I
W
a
2
W
a
3
G
1;2
(5.25)
(
2
(
a
1
;(1)
aa
1
;p

a;(2)
a
2
a
3
;q
)) :
a
2
,a
3
I
W
a
2
W
a
3
G
1;2
(5.26)
are dened by
((
2
(
a
1
;(1)
aa
1
;p

a;(2)
a
2
a
3
;q
)))(w
2
w
3
)
= (
2
(
a
1
;(1)
aa
1
;p

a;(2)
a
2
a

2
;q
))(w
2
w
3
; z
1
, z
2
1; )
(((
a
1
;(1)
aa
1
;p

a;(2)
a
2
a
3
;q
)))(w
2
w
3
)
= (
2
(
a
1
;(1)
aa
1
;p

a;(2)
a
2
a
3
;q
))(w
2
w
3
; z
1
, z
2
+ ; ) (5.27)
if w
2
w
3
W
a
2
W
a
3
, and by 0 if otherwise.
We also showed in [HKo3] that induces an automorphism on
aI
1
a
1
aa
1
1
a
a
2
a
3
given as follow:

a
1
;(1)
aa
1
;i

a;(2)
a
2
a
3
;j

b,cA

k,l,p,q
e
2iha
3
F
1
(
a
1
;(1)
aa
1
;i

a;(2)
a
2
a
3
;j
;
a
1
;(3)
a
2
b;k

b;(4)
a
3
a
1
;l
)
F(
a
1
;(3)
a
2
b;k

2
1
(
b;(4)
a
3
a
1
;l
);
a
1
;(5)
ca
1
;p

c;(6)
a
2
a
3
;q
)

a
1
;(5)
ca
1
;p

c;(6)
a
2
a
3
;q
. (5.28)
We still denoted this automorphism and its natural extension on
a,a
1
I
1
a
1
aa
1
1
a
a
2
a
3
by
. The following Lemma follows immediately from (5.28).
Lemma 5.3. can also be expressed graphically as follow:
:
a
2
a
3
a
1
a
1
i
j
a

a
2
a
3
a
1
a
1
i
j
. (5.29)
For , we prefer to use maps

A
0
and

A
0
dened in (4.35) instead of the map A
0
used
in [HKo3]. We obtain the following Lemma, which is proved in appendix.
46
Lemma 5.4. also induces an automorphism on
a,a
1
I
1
a
1
aa
1
1
a
a
2
a
3
given by

a
1
;(1)
aa
1
;i

a;(2)
a
2
a
3
;j

bI

k,l

cI

p,q
F
1
(
a
1
;(1)
aa
1
;i

a;(2)
a
2
a
3
;j
;
a
1
;(3)
a
2
b;k

b;(4)
a
3
a
1
;l
)
F(

A
0
(
a
1
;(3)
a
2
b;k
)

A
0
(
b;(4)
a
3
a
1
;l
),
b

;(5)
cb

;p

c;(6)
a
2
a
3
;q
)

A
0
(
b

;(5)
cb

;p
)
2
0
(
c;(6)
a
2
a
3
;q
). (5.30)
We still denote this automorphism by .
Lemma 5.5. can be expressed graphically as follow:
:
a
2
a
3
a
1
a
a
1
i
j

bI

l
a
2
a
3
b
b
i
j
l
l
a
a
1
a
3
a
1
. (5.31)
Proof. Using (5.30), we can see that is the composition of following maps

a,a
1
I
1
a
1
aa
1
1
a
a
2
a
3
F
1
//

b,a
1
I
1
a
1
a
2
b
1
b
a
3
a
1

A
0

A
0
//

b,a
1
I
1
b

a
2
a

1
1
a

1
a
3
b

b,cI
1
b
cb
1
c
a
2
a
3

b,cI
1
b

cb
1
c
a
2
a
3
.

A
0

2
0
oo
(5.32)
By the commutative diagram (5.1), (5.32) can be rewritten graphically as follow:
:
a
2
a
3
a
1
a
a
1
i
j
=

bI

k
a
2
a
3
a
3
i
j
a
a
1
a
1
a
1
k
k
b
(5.33)

bI

k
a
2
a
3
a

1
b

k
k
b

i
j
a
=

bI

k
a
2
a
3
a
1
b

k
b

i
j
k
a

bI

k
a
2
a
3
b
b
i
j
k
k
a
a
1
a
3
a
1
.
47
We have introduced S, , all as isomorphisms on
a,a
1
I
1
a
1
aa
1
1
a
a
2
a
3
. They satisfy
the following well-known equation [MSei1][MSei3][H10][HKo3]:
S = S. (5.34)
We proved in [HKo3] that S is determined by the identity (5.34) up to a constant S
e
e
.
We will solve the equation (5.34) for S graphically below.
Proposition 5.6.
S(a) :
a a
1
a
1
i

a
2
I
S
e
e
dima
2
a
i
a
1
a
2
(5.35)
Proof. Since we know that the equation (5.34) determine S up to an overall constant
S
e
e
. Hence we only need to check that (5.35) gives a solution to (5.34).
Combining (5.31) with (5.35), we obtain that
S :
a
2
a
3
a
1
a
1
i
j
a

b,a
4
I

k
S
e
e
dima
4
a
2
a
3
b
b
a
3
a
4
k
i
j
k
. (5.36)
The diagram in the right hand side of (5.36) can be deformed as follow:
a
2
a
3
b
b
a
3
a
4
k
i
j
k
=
a
2
a
3
b
b
a
3
a
4
k
i
j
k
(5.37)
By (5.18), we have

a
4
I

k
S
e
e
dima
4
a
2
a
3
b
b
a
3
a
4
k
i
j
k
= S
e
e
dimb
a
2
a
3
b
i
j
(5.38)
48
On the other hand, combining (5.29) with (5.35), we obtain
S :
a
2
a
3
a
1
a
1
i
j
a

bI
S
e
e
dimb
a
2
a
3
b
i
j
a
a
1
. (5.39)
Notice the diagram on the right hand side of (5.39) is exactly a deformation of the
diagram on the right hand side of (5.38). Hence we obtain that the map dened as
(5.35) give a solution to the equation S = S.
To determine S completely, we need determine S
e
e
. This can be done by using other
identities satised by S.
Proposition 5.7.
S
2
(a)(
a
1
;(1)
aa
1
;i
) =

A
0
(
a
1
;(1)
aa
1
;i
) (5.40)
Proof. By the denition of S-cation on
a
1
;(1)
aa
1
;i
1
a
1
aa
1
, we have

1
(S
2
(a)(
a
1
;(1)
aa
1
;i
))(w
a
; z, ) =
1
(
a
1
;(1)
aa
1
;i
)
_

L(0)
_
1

_
L(0)
w
a
; z,
_
(5.41)
for w
a
W
a
. Keep in mind of our convention on branch cut for logarithm. We have

L(0)
_
1

_
L(0)
w
a
= e
iL(0)
w
a
.
Hence we obtain

1
(S
2
(a)(
a
1
;(1)
aa
1
;i
))(w
a
; z, ) =
1
(
a
1
;(1)
aa
1
;i
)(e
iL(0)
w
a
; z, ) (5.42)
By (A.47), we also have

1
(
a
1
;(1)
aa
1
;i
)(e
iL(0)
w
a
; z, )
= E
_
Tr
Wa
1

a
1
;(1)
aa
1
;i
(|(e
2iz
)e
iL(0)
w
a
, e
2iz
)q
L(0)
c
24

_
= E
_
Tr
(Wa
1
)

A
0
(
a
1
;(1)
aa
1
;i
)(e
iL(0)
|(e
2iz
)w
a
, e
i
e
2iz
)q
L(0)
c
24

_
= E
_
Tr
(Wa
1
)

A
0
(
a
1
;(1)
aa
1
;i
)(|(e
2iz
)w
a
, e
2iz
)q
L(0)
c
24

_
=
1
(

A
0
(
a
1
;(1)
aa
1
;i
))(w
a
; z, ) (5.43)
Therefore, combining (5.42) and (5.43), we obtain (5.40).
The following lemma is proved in [BK2].
Lemma 5.8. Let D
2
=

aI
dim
2
a. Then D ,= 0 and we have

aI
dima
2
dima
D
2
a
1
a

1
a
a

2
a
2
=
a
1
a
2
a
1
a

1
. (5.44)
49
Proposition 5.9.
(S
e
e
)
2
=
1
D
2
. (5.45)
Proof. By (5.35), we have
S
2
(a) :
a a
1
a
1
i

a
2
,a
3
I
(S
e
e
)
2
dima
2
dima
3
a
i
a
1
a
2
a
3
a
3
. (5.46)
Apply (5.12) to the graph in the right hand side of (5.46), we obtain

a
2
I
dima
2
dima
3
D
2
a
i
a
1
a
2
a
3
a
3
=

a
2
I
dima
2
dima
3
D
2
a
a
1
a
2
a
3
a
3
i
. (5.47)
By (5.44), the right hands side of (5.47) equals to

a
1
a
3
a
a

1
a

1
i
=
a
1
a
3
a a

1
a

1
i
. (5.48)
By (5.40) and (5.22), we obtain (5.45).
So far, we have determined S
e
e
up to a sign. Now we consider the relation between
S and another generator of modular group T : + 1. We dene a T-action on

1
(
a
1
aa
1
;i
) as follow:
T(
1
(
a
1
aa
1
;i
))(w
a
; z, ) :=
1
(
a
1
aa
1
;i
)(w
a
; z, + 1). (5.49)
It is clear that this action induces an action of T on 1
a
1
aa
1
for all a, a
1
1, given by
T[
V
a
1
aa
1
= e
2i(ha
c
24
)
(5.50)
where h
a
is the lowest conformal weight of W
a
.
Lemma 5.10. S and T satisfy the following relation:
(T
1
S)
3
= S
2
= T
1
S
2
T. (5.51)
50
Proof. Let w
a
W
a
. We have
(T
1
S)
3
((
1
(
a
1
;(1)
aa
1
;i
)))(w
a
; z, )
=
1
(
a
1
;(1)
aa
1
;i
)
_
_

L(0)
_
1
1
1
1
_
L(0) _
1
1
_
L(0)
w
a
, z,
_
_
(5.52)
Keeping in mind our choice of branch cut. Then it is easy to show that

L(0)
_
1
1
1
1
_
L(0) _
1
1
_
L(0)
w
a
= e
iL(0)
w
a
.
By (5.42), we obtain the rst equality of (5.51). The proof of the second equality (5.51)
is similar.
Proposition 5.11. Let p

aI
e
2iha
dim
2
a. Then we have
S
e
e
=
1
p

e
2ic/8
=
1
p
+
e
2ic/8
. (5.53)
Proof. In the proof of the Theorem 3.1.16 in [BK2], Bakalov and Kirillov proved an
identity, which, in our own notation, can be written as follow:
1
(S
e
e
)
2
D
2
e
2i
c
24
ST
1
S =
1
S
e
e
p
+
D
2
e
2i
2c
24
TST.
By (5.51) and the fact p

p
+
= D
2
which is proved in [BK2], we simply obtain that
S
e
e
=
1
p

e
2ic/8
.
Using (5.45) and p
+
p

= D
2
, we also obtain the second equality.
We thus dene
D := p

e
2ic/8
= p
+
e
2ic/8
. (5.54)
Notice that this notation is compatible with the denition of D
2
. Then the action of
modular transformation S(a) on
a
1
I
Hom
V
(W
a
W
a
1
, W
a
1
) can be expressed graphi-
cally as follow:
S(a) :
a a
1
a
1
i

a
2
I
dima
2
D
a
i
a
1
a
2
. (5.55)
Proposition 5.12.
S
1
(a) :
a a
1
a
1
i

a
2
I
dima
2
D
a
i
a
1
a
2
. (5.56)
51
Proof. Composing the map (5.55) with (5.56), we obtain a map given as follow:
a a
1
a
1
i

a
2
,a
3
I
dima
2
dima
3
D
2
a
i
a
1
a
2
a
3
. (5.57)
Apply (5.44) to the graph in (5.57), it is easy to see that above map is the identity map.
Remark 5.13. Bakalov and Kirillov obtained the same formula (5.55) in [BK2] by
directly working with modular tensor category and solving equations obtained in the so-
called Lego-Teichm uller game [BK1]. In our approach, we see the direct link between the
modular transformations of q-traces of the product (or iterate) of intertwining operators
and their graphic representations in a modular tensor category.
Proposition 5.14.
(S(a))

:
a a
1
a
1
j

a
2
I
dima
2
D
a
j
a
1
a
2
, (5.58)
(S
1
(a))

:
a a
1
a
1
j

a
2
I
dima
2
D
a
j
a
1
a
2
. (5.59)
Proof. We only prove (5.58). The proof of (5.59) is analogous to that of (5.58).
It is enough to show that the pairing between the image of (5.35) and that of (5.58)
still gives
ij
. This can be proved as follow:

a
2
dima
2
D
2
a
a
1
a
1
a
2
i
j
=

a
2
dima
1
dima
2
dima
1
D
2
a
a
1
a
1
a
2
i
j
(5.60)
By (5.44), the right hand side of (5.60) equals to
1
dima
1
a a
1
a
2
i
j
=
ij
.
52
6 Categorical formulations and constructions
In this section, we give a categorical formulation of modular invariant conformal full
eld algebra over V
L
V
L
, open-string vertex operator algebra over V equipped with
nondegenerate invariant bilinear forms and Cardy condition. Then we introduce a notion
called Cardy (
V
[(
V V
-algebra. In the end, we give a categorical construction of such
algebra in the Cardy case [FFFS2].
6.1 Modular invariant (
V
L
V
R-algebras
We rst recall the notion of coalgebra and Frobenius algebra ([FS]) in a tensor category.
Denition 6.1. A coalgebra A in a tensor category ( is an object with a coproduct
Mor(A, AA) and a counit Mor(A, 1
C
) such that
(id
A
) = (id
A
) , ( id
A
) = id
A
= (id
A
) , (6.1)
which can also be expressed in term of the following graphic equations:
= = =
Denition 6.2. Frobenius algebra in ( is an object that is both an algebra and a
coalgebra and for which the product and coproduct are related by
(id
A
m) (id
A
) = m = (mid
A
) (id
A
), (6.2)
or as the following graphic equations,
= =
. (6.3)
A Frobenius algebra is called symmetric if the following condition is satised.
=
. (6.4)
Let V
L
and V
R
be vertex operator algebras satisfying the conditions in Theorem 1.1.
Then the vertex operator algebra V
L
V
R
also satises the conditions in Theorem 1.1
[HKo1]. Thus (
V
L
V
R also has a structure of modular tensor category. In particular,
53
we choose the braiding structure on (
V
L
V
R to be 1
+
which is dened in [Ko2]. The
twist
+
for each V
L
V
R
-module is dened by

+
= e
2iL
L
(0)
e
2iL
R
(0)
. (6.5)
Duality maps are naturally induced from those of (
V
L and (
V
R.
The following theorem is proved in [Ko1].
Theorem 6.3. The category of conformal full eld algebras over V
L
V
R
equipped with
nondegenerate invariant bilinear forms is isomorphic to the category of commutative
Frobenius algebra in (
V
L
V
R with a trivial twist.
Remark 6.4. In a ribbon category, it was proved in [FFRS] that a commutative Frobe-
nius algebra with a trivial twist is equivalent to a commutative symmetric Frobenius
algebra.
Let 1
L
and 1
R
denote the set of equivalent class of irreducible V
L
-modules and V
R
-
modules respectively. We use a and a
i
for i N to denote elements in 1
L
and we use e
to denote the equivalent class of V
L
. We use a and a
i
for i N to denote elements in
1
R
, and e to denote the equivalent class of V
R
. For each a 1
L
( a 1
R
), we choose a
representative W
a
(W
a
). We denote the vector space of intertwining operators of type
_
Wa
3
Wa
1
Wa
2
_
and
_
W a
3
W a
1
W a
2
_
as 1
a
3
a
1
a
2
and

1
a
3
a
1
a
2
respectively, the fusion rule as N
a
3
a
1
a
2
and N
a
3
a
1
a
2
respectively.
A conformal full eld algebra over V
L
V
R
, denoted as A
cl
, is a direct sum of
irreducible modules of V
L
V
R
, i.e.
A
cl
=
N
=1
W
r()
W
r()
, (6.6)
where r : 1, . . . , N 1
L
and r : 1, . . . , N 1
R
, for some N Z
+
. Let e
c
ab;i

and e
c
a

b;j
be basis for 1
c
ab
and

1
c
a

b
, and f
ab
c;i
and

f
a

b
c;j
be the dual basis respectively.
Then the vertex operator Y can also be expanded as follow:
Y =

,,

i,j
d

(f
r()r()
r();i
,

f
r() r()
r();j
) e
r();i
r()r()
e
r();j
r() r()
, (6.7)
where d

denes a bilinear map

(1
r()
r()r()
)

1
r()
r() r()
)

C
for all , , = 1, . . . , N for some N N.
Since the trace function pick out = terms, we dene Y
diag

by
Y
diag

:=

i,j
d

(f
r()r()
r();i
,

f
r() r()
r();j
) e
r();i
r()r()
e
r();j
r() r()
. (6.8)
Let Y
diag
:=

Y
diag

. Of course, it is obvious to see that such dened Y

diag
is indepen-
dent of the choice of basis. We denote the representation of the modular transformation
S :
1

on
bI
L1
b
ab
and
bI
R1

b
a

b
, by S
L
(a) and S
R
( a) respectively.
54
In [HKo3], we dened the notion of modular invariant conformal full eld algebra
over V
L
V
R
(see [HKo3] for the precise denition). It basically means that n-point
genus-one correlation functions built out of q- q-traces are invariant under the action of
modular group SL(2, Z) for all n N. Moreover, we proved the following results in
[HKo3].
Proposition 6.5. A
cl
, a conformal full eld algebra over V
L
V
R
, is modular invariant
if it satises c
L
c
R
= 0 mod 24 and
S
L
(r()) (S
R
( r()))
1
Y
diag

= Y
diag

(6.9)
for all = 1, . . . , N.
We denote the morphism in
Hom
C
V
L
V
R
_
(W
r()
W
r()
) (W
r()
W
r()
), W
r()
W
r()
_
which corresponds to Y
diag

by m
Y
diag

. Then (6.9) is equivalent to the following categor-

ical condition:
S
L
(r()) (S
R
( r()))
1
m
Y
diag

= m
Y
diag

(6.10)
for all = 1, . . . , N.
Now we choose a basis b
A
cl
a a;i
of Hom
C
V
L
V
R
(W
a
W
a
, A
cl
) and a basis b
a a;i
A
cl
of
Hom
C
V
L
V
R
(A
cl
, W
a
W
a
) as follow:
b
A
cl
ab;i
:=
a a
A
cl
i
,
b
ab;i
A
cl
:=
A
cl
a a
i
, (6.11)
satisfying the following conditions: for all a 1
L
, a 1
R
,
=
ij
a a
a a
A
cl
j
i
and

a, a,i
A
cl
A
cl
a a
i
i
=
A
cl
A
cl
. (6.12)
Then the condition (6.10) can be further expressed graphically as follow:
dima a
D
L
D
R
A
cl
A
cl
A
cl
a a
a a
=

r(k)=a a
A
cl
a a
a a
k
k
A
cl
A
cl
(6.13)
for all a 1
L
, a 1
R
, where D
L
and D
R
are the D dened by (5.54) in (
V
L and (
V
R
respectively.
55
Denition 6.6. Let V
L
and V
R
be so that c
L
c
R
= 0 mod 24. A modular invariant
(
V
L
V
R-algebra is an associative algebra (A,
A
,
A
) satisfying the condition (6.13).
Some properties of modular invariant algebra follow immediately from above deni-
tion, such as the following famous condition [MSei2]:

aI
L
, aI
R
(S
L
(e))
ab
N
b

b
(S
R
( e))
1

b a
= N
a a
(6.14)
where N
a a
is the multiplicity of W
a
W
a
in A
cl
for a 1
L
, a 1
R
. We leave a more
systematic study of modular invariant (
V
L
V
R-algebras to elsewhere.
Theorem 6.7. Let V
L
and V
R
be so that c
L
c
R
= 0 mod 24. The following two
notions are equivalent:
1. Modular invariant conformal full eld algebra over V
L
V
R
equipped with a non-
degenerate invariant bilinear form.
2. Modular invariant commutative Frobenius algebra with a trivial twist.
Proof. The Theorem follows from Theorem 6.3 and the equivalence between (6.9)
and (6.13) immediately.
Remark 6.8. In the case V
L
= C

= V
R
, a modular invariant commutative Frobenius
algebra with a trivial twist in (
V
L
V
R is simply a commutative Frobenius algebra over C,
which is equivalent to a 2-dimensional topological eld theory (see for example [BK2]).
In this case, the modular invariance condition holds automatically.
6.2 Cardy (
V
[(
V V
-algebras
For an open-string vertex operator algebra V
op
over V equipped with a nondegenerate
invariant bilinear form (, )
op
, there is an isomorphism
op
: V
op
V
op
induce from
(, )
op
(recall (4.40)).
In this case, V
op
is a V -module and Y
f
op
is an intertwining operator. By comparing
(2.26) with (4.37), and (2.27) with (4.38), we see that the conditions (2.26) and (2.27)
can be rewritten as
Y
f
op
=
1
op

123
(Y
f
op
) (
op
id
Vop
) (6.15)
=
1
op

132
(Y
f
op
) (id
op

op
). (6.16)
Remark 6.9. The representation theory of open-string vertex operator algebra can be
developed. In that context,
123
(Y
f
op
) gives V

op
a right V
op
-module structure and the
equation (6.15) is equivalent to the statement that
op
is an isomorphism between two
right V
op
-modules. Similarly,
132
(Y
f
op
) gives V

op
a left V
op
-module structure and the
equation (6.16) is equivalent to the statement that
op
is an isomorphism between two
left V
op
-modules. But we do not need it in this work.
Theorem 6.10. The category of open-string vertex operator algebras over V equipped
with a nondegenerate invariant bilinear form is isomorphic to the category of symmetric
Frobenius algebras in (
V
.
56
Proof. We have already shown in [HKo1] that an open-string vertex operator algebra
over V is equivalent to an associative algebra in (
V
.
Let V
op
be an open-string vertex operator algebra over V . Giving a nondegener-
ate invariant bilinear form (recall (2.26) and (2.27)) on V
op
is equivalent to give an
isomorphism
op
: V
op
V

op
satisfying the conditions (6.15) and (6.16). If we dene
V

op
V
op
:=
V

op
V
op
V

op
, (6.17)
V
op
V

op
:=
V

op
V
op
V

op
, (6.18)
then (6.15) and (6.16) can be rewritten as
V
op
V
op
V

op

op
=
V
op
V
op
V

op

op
. (6.19)
V
op
V
op
V

op

op
=
V
op
V
op
V

op

op
, (6.20)
Using the map
op
and its inverse, we can obtain a natural coalgebra structure on F
dened as follow:

Vop
=
V
op
V
op
V
op

op

1
op

1
op

Vop
=

op
V
op
. (6.21)
Similar to the proof of [Ko1, Thm. 4.15], (6.19) and (6.20) imply that such dened
Vop
and
Vop
give V
op
a Frobenius algebra structure. Moreover, we also showed in [Ko1] that
(6.19) implies the equality between
op
and the left hand side of (6.4). Similarly, using
(6.20), we can show that the right hand side of (6.4) also equals to
op
. Thus V
op
has a
structure of symmetric Frobenius algebra.
We thus obtain a functor from the rst category to the second category.
Conversely, given a symmetric Frobenius algebra in (
V
. In [HKo1], we showed that
it gives an open-string vertex operator algebra over V . It is shown in [FRS2] that either
side of (6.4) is an isomorphism. Take
op
to be either side of (6.4). Then (6.19) and
(6.20) follow automatically from the denition (6.17) and (6.18). They are nothing but
the invariance properties (recall (6.15)(6.16)) of the bilinear form associated with
op
.
Thus we obtain a functor from the second category to the rst category.
57
It is routine to check that these two functors are inverse to each other.
Now we consider an open-closed eld algebra over V given in (3.12) equipped with
nondegenerate invariant bilinear forms (, )
op
and (, )
cl
. We assume that V
cl
and V
op
have the following decompositions:
V
cl
=
N
cl
i=1
W
r
L
(i)
W
r
R
(i)
, V
op
=
Nop
i=1
W
r(i)
where r
L
, r
R
: 1, . . . , N
cl
1 and r : 1, . . . , N
op
1. We denote the embedding
b
Vop
(i)
: W
r(i)
V
op
, the projection b
(i)
Vop
: V
op
W
r(i)
and b
V
cl
(i)
: W
r
L
(i)
W
r
R
(i)
T(V
cl
)
by the following graphs:
b
Vop
(i)
=
r(i)
V
op
i
,
b
(i)
Vop
=
V
op
r(i)
i
,
b
V
cl
(i)
=
T(V
cl
)
i i
r
L
(i) r
R
(i) . (6.22)
We denote the map
clop
: T(V
cl
) V
op
[Ko2] by the following graph

clop
:=
T(V
cl
)
V
op
. (6.23)
Now we can express the Cardy condition (4.73) in graphs. The left hand side of
(4.73) can be expressed by:

N
cl
i=1
V
op

op

1
cl
i i
r
R
(i)
(r
L
(i))

V
op

op
i i
r

R
(i)
(r
L
(i))

. (6.24)
By the universal property of tensor product, for a
i
1, i = 1, . . . , 6, we have a
canonical isomorphism:

aI
1
a
4
a
1
a
1
a
a
2
a
3

=
Hom
V
(W
a
1
(W
a
2
W
a
3
), W
a
4
)

1

2
m
Y
1
(id
Wa
1
m
Y
2
). (6.25)
Under this canonical isomorphism, the Cardy condition (4.73) can be viewed as a con-
dition on two morphisms in Hom
V
(V
op
(V
op
W
r
R
(i)
), W
r
R
(i)
). In particular, the left
hand side of (4.73) viewed as a morphism in Hom
V
(V
op
(V
op
W
r
R
(i)
), W
r
R
(i)
) can be
58
expressed as follow:

N
cl
i=1
V
op

op

op
V
op
r
R
(i)
i i
T(
1
cl
)
i i
r
R
(i)
. (6.26)
We dene a morphism

clop
: V
op
T(V
cl
) by

clop
=
V
op
T(V
cl
)
:=

op
T(
1
cl
)
. (6.27)
Then using the morphism

clop
, we can rewrite the graph in (6.26) as follow:

op
V
op
V
op
i i
i i
r
R
(i)
r
R
(i)
=
V
op
V
op
r
R
(i)

op
r
R
(i)
i i
i i
. (6.28)
Using (5.55), (6.26) and (6.28), we obtain a graphic version of the Cardy condition
(4.73) as follow:

r
R
(i)=a
V
op
V
op

op
i i
i i
r
R
(i)
r
R
(i)
=
dima
D
V
op
V
op
V
op
V
op
V
op
a
a (6.29)
Using the Frobenius properties of V
op
, one can show that (6.29) is equivalent to the
following condition:

r
R
(i)=a

V
op
V
op
i i
i i
a
a
=
dima
D
V
op
V
op
a
a . (6.30)
59
The asymmetry between chiral and antichiral parts in (6.29)(6.30) is supercial. Using
(5.44), one can show that (6.30) is further equivalent to the following condition:
dima
D
V
op
V
op
a
a
=

r(k)=a
V
op
V
op
a
k
a
k
. (6.31)
We recall a denition in [Ko2].
Denition 6.11. An open-closed (
V
[(
V V
-algebra, denoted as (A
op
[A
cl
,
clop
), consists
of a commutative symmetric associative algebra A
cl
in (
V V
, an associative algebra A
op
in (
V
and an associative algebra morphism
clop
: T(V
cl
) V
op
, satisfying the following
condition:
V
op
V
op
T(V
cl
)
=
T(V
cl
) V
op
V
op
. (6.32)
The following Theorem is proved in [Ko2].
Theorem 6.12. The category of open-closed eld algebras over V is isomorphic to the
category of open-closed (
V
[(
V V
-algebras.
Denition 6.13. A Cardy (
V
[(
V V
-algebra is an open-closed (
V
[(
V V
-algebra (A
op
[A
cl
,
clop
)
such that A
cl
is modular invariant commutative symmetric Frobenius algebra in (
V V
and A
op
a symmetric Frobenius algebra in (
V
and the Cardy condition (6.30) or (6.31)
hold.
Remark 6.14. Notice that, in the case V = C, Cardy (
V
[(
V V
-algebra (using (6.31))
exactly coincides with the usual algebraic formulation of 2-dimensional open-closed topo-
logical eld theory [La][Mo1][Mo2][Se2][MSeg][AN][LP]. As we discussed in the introduc-
tion, we believe that open-closed partial conformal eld theories of all genus satisfying
the V -invariant boundary condition [Ko2] are classied by Cardy (
V
[(
V V
-algebras.
The following result is clear.
Theorem 6.15. The category of open-closed eld algebras over V equipped with non-
degenerate invariant bilinear forms and satisfying the modular invariance condition and
the Cardy condition is isomorphic to the category of Cardy (
V
[(
V V
-algebras.
Denition 6.16. A V -invariant D-brane associated to a closed algebra A
cl
in (
V V
is
a pair (A
op
,
clop
) such that the triple (A
op
[A
cl
,
clop
) gives a Cardy (
V
[(
V V
-algebra.
D-branes usually form a category as we will see in an example in the next subsection.
60
6.3 Constructions
In this section, we give a categorical construction of Cardy (
V
[(
V V
-algebra. This
construction is called Cardy case in physics literature [FFFS2].
Let us rst recall the diagonal construction of the close algebra V
cl
[FFRS][HKo2][Ko1]
[HKo3]. We will follow the categorical construction given in [Ko1].
Let V
cl
is be the object in (
V V
given as follow
V
cl
=
aI
W
a
W

a
. (6.33)
The decomposition of V
cl
as a direct sum gives a natural embedding V V V
cl
. We
denote this embedding as
cl
. We dene a morphism
cl
Hom
V V
(V
cl
V
cl
, V
cl
) by

cl
=

a
1
,a
2
,a
3
A
N
a
3
a
1
a
2

i,j=1
f
a
1
a
2
a
3
;i
, f
a

1
a

2
a

3
;j
) e
a
3
a
1
a
2
;i
e
a

3
a

1
a

2
;j
, (6.34)
where e
a
3
a
1
a
2
;i
and f
a
1
a
2
;j
a
3
are basis vectors given in (5.13) and (5.15) and , ) is a bilinear
pairing given by
1
dima
3
a
1
i
a
3
a
2
a

1
j
a

2
a

3
. (6.35)
Notice that V

cl
has the same decomposition as V
cl
in (6.33). They are isomorphic
as V V -modules. There is, however, no canonical isomorphism. Now we choose a
particular isomorphism
cl
: V
cl
V

cl
given by

cl
=
aI
D
dima
e
2iha
id
WaW

a
. (6.36)
The isomorphism
cl
induces a nondegenerate invariant bilinear form on V
cl
viewed as
V V -module.
The following Theorem is a categorical version of Theorem 5.1 in [HKo3]. We give
a categorical proof here.
Theorem 6.17. (V
cl
,
cl
,
cl
) together with isomorphism
cl
gives a modular invariant
commutative Frobenius algebra in (
V V
with a trivial twist.
Proof. It was proved in [Ko1] that (V
cl
,
cl
,
cl
) together with isomorphism
cl
gives
a commutative Frobenius algebra with a trivial twist. It remains to show the modular
invariance.
First, the bilinear pairing , ) given in (6.35) can be naturally extended to a bilinear
form, still denoted as , ), on
a,a
1
I
(W
a
1
, W
a
W
a
1
) as follow:
f
aa
1
a
1
;i
, f
bb
1
b
1
;j
) :=
ab

a
1
b

1
f
aa
1
a
1
;i
, f
a

1
a

1
;j
). (6.37)
61
Then it is easy to see that to prove the modular invariance of V
cl
is equivalent to
prove that the bilinear form , ) on
a,a
1
I
(W
a
1
, W
a
W
a
1
) dened above is invariant
under the action of (S
1
)

. Clearly, when b ,= a

, (S
1
(a))

f
aa
1
a
1
;i
, (S(b))

f
bb
1
b
1
;j
) = 0.
When b = a

, we have (using (5.58), (5.59) and (5.44))

(S
1
(a))

f
aa
1
a
1
;i
, (S(a

))

f
a

b
1
b
1
;j
) =

a
3
I
dima
3
D
2
a
a
1
b
1
i
j
a
3
=

a
3
I
dima
3
D
2
a
a
1
b
1
i
j
a
3
=

a
3
I
dima
3
D
2
a
a
1
b
1
i j
a
3
=

a
3
I
dima
3
D
2
a
a
1
b
1
b

1
i
j
a
3
=

a
1
b

1
dima
1
a
a
1
i
j
a

1
=

a
1
b

1
dima
1
a a
1
i j
a
1
.
Now we dene V
op
. Let X be a V -module. Let e
X
: X

X V and i
X
: V
X X

be the duality maps dened in [Ko1]. V

op
:= X X

has a natural structure of

symmetric Frobenius algebra [FS] with
op
:= i
X
,
op
:= id
X
e
X
id
X
,
op
:= e
X
and

op
:= id
X
i
X
id
X
.
62
Now we dene a map
clop
: T(V
cl
) V
op
by
X X

a a

:=
a a

X X

. (6.38)
Lemma 6.18.
X X

a a

=
dima
D
a a

X X

. (6.39)
Proof. By (6.27) and (6.38), we have
X X

a a

=
dima
D
X X

a a

=
dima
D
X X

a a

(6.40)
It is easy to see that the last gure in (6.40) can be deformed to that on the right hand
side of (6.39).
Theorem 6.19. (V
cl
, V
op
,
clop
) is a Cardy (
V
[(
V V
-algebra.
Proof. Recall that T(V
cl
) together with multiplication morphism
T(V
cl
)
= T(
cl
)
2
and morphism
T(V
cl
)
= T(
cl
)
0
is an associative algebra. We rst prove that
clop
is an algebra morphism. It is clear that
clop

T(V
cl
)
=
op
. It remains to show the
following identity

clop

T(V
cl
)
=
op
(
clop

clop
). (6.41)
By the denition of
2
, that of
cl
and (6.38), we obtain

clop

T(V
cl
)
=

a,b,cI

i,j
1
dimc
c c

a b
i
j
a a

b b

i j
c c

X
(6.42)
It is easy to see that the right hand side of (6.42) equals to

a,b,cI

i,j

c
1
I
a a

b b

i j
c
1
c i
a b j
X X

(6.43)
63
Using (5.17) to sum up the indices c
1
and i, we obtain that (6.43) further equals to

a,b,cI

j
a a

b b

j
c
j
X X

(6.44)
Using (5.17) again, we obtain

a,bI
a a

b b

X X

a,bI
a a

b b

X X

, (6.45)
the right hand side of which is nothing but
op
(
clop

clop
).
The commutativity (6.32) follows from the following identity:
X X

a a

=
X X

a a

. (6.46)
In summary, we have proved that the triple (V
cl
, V
op
,
clop
) is an open-closed (
V
[(
V V
-
algebra. It remains to show that Cardy condition (6.30) holds. We use (6.38), (6.39)
and the denition of
op
and
op
to express both sides of (6.30) graphically. Then it is
easy to see that both sides are the deformations of each other.
Theorem 6.19 reects the general fact that consistent open theories (or D-branes)
for a given closed theory are not unique. Instead they form a category. There are many
good questions one can ask about Cardy (
V
[(
V V
-algebra, for example its relation to the
works of Fuchs, Runkel, Schweigert and Fjelstad [FS][FRS1]-[FRS4][FjFRS1][FjFRS2].
We leave such topics to [KR] and future publications.
A The Proof of Lemma 5.30
Lemma A.1. For w
1
W
a
1
, w
a
W
a
2
and w

a
3
(W
a
3
)

and
a
3
a
1
a
2
1
a
3
a
1
a
2
, we have
w

a
3
, (|(x)w
a
1
, x)w
a
2
)
=

A
r
()(|(e
(2r+1)i
x
1
)e
(2r+1)iL(0)
w
a
1
, e
(2r+1)i
x
1
)w

a
3
, w
a
2
). (A.47)
64
Proof. Using the denition of

A
r
, we see that the left hand side of (A.47) equals to

A
r
()(e
xL(1)
x
2L(0)
|(x)w
a
1
, e
(2r+1)i
x
1
)w

a
3
, w
a
2
). (A.48)
In [H8], the following formula
e
xL(1)
x
2L(0)
e
(2r+1)iL(0)
|(x)e
(2r+1)iL(0)
= |(x
1
) (A.49)
is proved. Applying (A.49) to (A.48), we obtain (A.47) immediately.
Now we are ready to give a proof of Lemma 5.4.
Proof. We have, for w
a
2
W
a
2
, w
a
3
W
a
3
,
((
2
(
a
1
;(1)
aa
1
;i

a;(2)
a
2
a
3
;j
))(w
a
2
w
a
3
))(z
1
, z
2
+ , )
= E
_
Tr
Wa
1

a
1
;(1)
aa
1
;i
(|(e
2i(z
2
+)
)

a;(2)
a
2
a
3
;j
(w
a
2
, z
1
(z
2
+ ))w
a
3
, e
2i(z
2
+)
)q
L(0)
c
24

_
=

bI

k,l
F
1
(
a
1
;(1)
aa
1
;i

a;(2)
a
2
a
3
;j
;
a
1
;(3)
a
2
b;k

b;(4)
a
3
a
1
;l
)
E
_
Tr
Wa
1

a
1
;(3)
a
2
b;k
(|(e
2iz
1
)w
a
2
, e
2iz
1
)

b;(4)
a
3
a
1
;l
(|(e
2i(z
2
+)
)w
a
3
, e
2i(z
2
+)
)q
L(0)
c
24

_
. (A.50)
Using the L(0)-conjugation formula, we can move q

from the right side of

b;(4)
a
3
a
1
;l
to the
left side of
b;(4)
a
3
a
1
;l
. Then using the following property of trace:
Tr
Wa
1
(AB) = Tr
W
b
(BA), (A.51)
for all A : W
b
W
a
1
, B : W
a
1
W
b
whenever the multiple sums in either side of (A.51)
converge absolutely, we obtain that the left hand side of (A.50) equals to

bI

k,l
F
1
(
a
1
;(1)
aa
1
;i

a;(2)
a
2
a
3
;j
;
a
1
;(3)
a
2
b;k

b;(4)
a
3
a
1
;l
)
E
_
Tr
W
b

b;(4)
a
3
a
1
;l
(|(e
2iz
2
)w
a
3
, e
2iz
2
)
a
1
;(3)
a
2
b;k
(|(e
2iz
1
)w
a
2
, e
2iz
1
)q
L(0)
c
24

_
.
(A.52)
Now apply (A.47) to (A.52). We then obtain that (A.52) equals to

bI

k,l
F
1
(
a
1
;(1)
aa
1
;i

a;(2)
a
2
a
3
;j
;
a
1
;(3)
a
2
b;k

b;(4)
a
3
a
1
;l
)
E
_
Tr
(W
b
)

A
r
(
a
1
;(3)
a
2
b;k
)(|(e
(2r+1)i
e
2iz
1
)e
(2r+1)iL(0)
w
a
2
, e
(2r+1)i
e
2iz
1
)

A
r
(
b;(4)
a
3
a
1
;l
)(|(e
(2r+1)i
e
2iz
2
)e
(2r+1)iL(0)
w
a
3
, e
(2r+1)i
e
2iz
2
)q
L(0)
c
24

_
.
65
Now apply the associativity again and be careful about the branch cut as in [H8], then
the left hand side of (A.50) further equals to

bI

k,l

cI

p,q
F
1
(
a
1
;(1)
aa
1
;i

a;(2)
a
2
a
3
;j
;
a
1
;(3)
a
2
b;k

b;(4)
a
3
a
1
;l
)
F(

A
r
(
a
1
;(3)
a
2
b;k
)

A
r
(
b;(4)
a
3
a
1
;l
),
b

;(5)
cb

;p

c;(6)
a
2
a
3
;q
)
E
_
Tr
(W
b
)
q
L(0)
c
24

;(5)
cb

;p
(|(e
(2r+1)i
e
2iz
2
)

c;(6)
a
2
a
3
;q
(e
(2r+1)iL(0)
w
a
2
, e
i
(z
1
z
2
))e
(2r+1)iL(0)
w
a
3
, e
(2r+1)i
e
2iz
2
)
_
.
=

bI

k,l

cI

p,q
F
1
(
a
1
;(1)
aa
1
;i

a;(2)
a
2
a
3
;j
;
a
1
;(3)
a
2
b;k

b;(4)
a
3
a
1
;l
)
F(

A
r
(
a
1
;(3)
a
2
b;k
)

A
r
(
b;(4)
a
3
a
1
;l
),
b

;(5)
cb

;p

c;(6)
a
2
a
3
;q
)
E
_
Tr
W
b
q
L(0)
c
24

A
r
(
b

;(5)
cb

;p
)(e
(2r+1)iL(0)

c;(6)
a
2
a
3
;q
(e
(2r+1)iL(0)
w
a
2
, e
i
(z
1
z
2
))e
(2r+1)iL(0)
w
a
3
, e
2iz
2
)
_
(A.53)
Choosing r = 0 and using (, e
2i
x) =
2
0
()(, x), we obtain
((
2
(
a
1
;(1)
aa
1
;i

a;(2)
a
2
a
3
;j
))(w
a
2
w
a
3
))(z
1
, z
2
+ , )
=

bI

k,l

cI

p,q
F
1
(
a
1
;(1)
aa
1
;i

a;(2)
a
2
a
3
;j
;
a
1
;(3)
a
2
b;k

b;(4)
a
3
a
1
;l
)
F(

A
0
(
a
1
;(3)
a
2
b;k
)

A
0
(
b;(4)
a
3
a
1
;l
),
b

;(5)
cb

;p

c;(6)
a
2
a
3
;q
)
E
_
Tr
W
b

A
0
(
b

;(5)
cb

;p
)(
2
0
(
c;(6)
a
2
a
3
;q
)(w
a
2
, z
1
z
2
)w
a
3
, e
2iz
2
)q
L(0)
c
24

_
. (A.54)
By the linear independency proved in [H10] of the last factor in each term of above sum,
it is clear that induces a map given by (5.30).
References
[AN] A. Alexeevski and S.M. Natanzon, Noncommutative two-dimensional topological
eld theories and Hurwitz numbers for real algebraic curves. To appear in Sel. Math.
(2006). math.GT/0202164.
[BHL] K. Barron, Y.-Z. Huang and J. Lepowsky, Factorization of formal exponential and
uniformization, J. Alg. 228 (2000), 551579.
[BK1] B. Bakalov and A. Kirillov Jr. , On the Lego-Teichm uller game, Transform. Groups
5 (2000) 207.
[BK2] B. Bakalov and A. Kirillov Jr., Lectures on Tensor Categories and Modular Functors,
University Lecture Series, Vol. 21, Amer. Math. Soc., Providence, 2001.
66
[C1] J. L. Cardy, Conformal invariance and surface critical behavior, Nucl. Phys. B240
(1984), 514-532.
[C2] J. L. Cardy, Eect of boundary conditions on the operator content of two-
dimensional conformally invariant theories, Nucl. Phys. B275 (1986), 200218.
[C3] J. L. Cardy, Operator content of two-dimensional conformal invariant theories, Nucl.
Phys. B270 (1986) 186-204.
[C4] J. L. Cardy, Boundary conditions, fusion rules and the Verlinde formula, Nucl. Phys.
B324 (1989), 581596.
[DLM] C.-Y. Dong, H.-S. Li and G. Mason, Modular-invariance of trace functions in orbifold
theory and generalized Moonshine, Comm. Math. Phys. 214 (2000), 156.
[FFFS1] G. Felder, J. Fr ohlich, J. Fuchs and C. Schweigert, The geometry of WZW branes,
J. Geom. Phys. 34 (2000) 162190.
[FFFS2] G. Felder, J. Fr ohlich, J. Fuchs and C. Schweigert, Correlation functions and bound-
ary conditions in rational conformal eld theory and three-dimensional topology,
Compositio Math. 131 (2002), 189237.
[FFRS] J. Fr ohlich, J. Fuchs, I. Runkel, C. Schweigert, Correspondences of ribbon categories,
Adv. Math. 199 (2006), no.1 192329.
[FjFRS1] J. Fjelstad, J. Fuchs, I. Runkel and C. Schweigert, TFT construction of RCFT
correlators V: Proof of modular invariance and factorisation, Theo. and Appl. of
Cat. 16 (2006), 342-433 [hep-th/0503194].
[FjFRS2] J. Fjelstad, J. Fuchs, I. Runkel and C. Schweigert, Uniqueness of open/closed ratio-
nal CFT with given algebra of open states, preprint, arXiv:hep-th/0612306.
[FHL] I. B. Frenkel, Y.-Z. Huang and J. Lepowsky, On axiomatic approaches to vertex
operator algebras and modules, preprint, 1989; Memoirs Amer. Math. Soc. 104,
1993.
[FRS1] J. Fuchs, I. Runkel and C. Schweigert, Conformal correlation functions, Frobenius
algebras and triangulations, Nucl. Phys. B624 (2002), 452468.
[FRS2] J. Fuchs, I. Runkel and C. Schweigert, TFT construction of RCFT correlators I:
Partition functions, Nucl. Phys. B678 (2004) 353497.
[FRS3] J. Fuchs, I. Runkel and C. Schweigert, TFT construction of RCFT correlators III:
Simple currents, Nucl. Phys. B694 (2004) 277
[FRS4] J. Fuchs, I. Runkel and C. Schweigert, TFT construction of RCFT correlators IV:
Structure constants and correlation functions, preprint hep-th/0412290, to appear
in Nucl. Phys. B.
[FS] J. Fuchs and C. Schweigert, Category theory for conformal boundary conditions,
Fields Institute Commun. 39 (2003) 2570.
[G] W.L. Gan, Koszul duality for dioperads, Math. Res. Lett. 10 (2003), no.1, 109124.
67
[H1] Y.-Z. Huang, Geometric interpretation of vertex operator algebras, Proc. Natl. Acad.
Sci. USA 88 (1991), 99649968.
[H2] Y.-Z. Huang, A theory of tensor products for module categories for a vertex operator
algebra, IV, J. Pure Appl. Alg. 100 (1995), 173-216.
[H3] Y.-Z. Huang, Two-dimensional conformal geometry and vertex operator algebras,
Progress in Mathematics, Vol. 148, Birkhauser, Boston, 1997.
[H4] Y.-Z. Huang, Generalized rationality and a Jacobi identity for intertwining oper-
ator algebras, Selecta Math. (N. S.), 6 (2000), 225267.
[H5] Y.-Z. Huang, A functional-analytic theory of vertex (operator) algebras. I. Comm.
Math. Phys. 204 (1999), no. 1, 6184.
[H6] A functional-analytic theory of vertex (operator) algebras. II. Comm. Math. Phys.
242 (2003), no. 3, 425444.
[H7] Y.-Z. Huang, Dierential equations and intertwining operators, Comm. Contemp.
Math. 7 (2005) 375400.
[H8] Y.-Z. Huang, Riemann surfaces with boundaries and the theory of vertex operator
algebras, Fields Institute Commun. 39 (2003) 109.
[H9] Y.-Z. Huang, Dierential equations, duality and modular invariance, Comm. Con-
temp. Math. 7 (2005), 649706.
[H10] Y.-Z. Huang, Vertex operator algebras and the Verlinde conjecture, Comm. Con-
temp. Math. 10 (2008) 103-154.
[H11] Y.-Z. Huang, Rigidity and modularity of vertex tensor categories, preprint,
math.QA/0502533, to appear.
[HKo1] Y.-Z. Huang and L. Kong, Open-string vertex algebra, category and operad,
Comm. Math. Phys., 250 (2004), 433-471.
[HKo2] Y.-Z. Huang and L. Kong, Full eld algebra, Commun. Math. Phys. 272 (2007)
345396 [math.QA/0511328].
[HKo3] Y.-Z. Huang and L. Kong, Modular invariance for conformal full eld algebras,
math.QA/0609570.
[HL1] Y.-Z. Huang and J. Lepowsky, Tensor products of modules for a vertex operator
algebra and vertex tensor categories, in: Lie Theory and Geometry, in honor of
Bertram Kostant, ed. R. Brylinski, J.-L. Brylinski, V. Guillemin, V. Kac, Birkhauser,
Boston, 1994, 349383.
[HL2] Y.-Z. Huang and J. Lepowsky, A theory of tensor products for module categories for
a vertex operator algebra, I, Selecta Math. (N.S.) 1 (1995), 699-756.
[HL3] Y.-Z. Huang and J. Lepowsky, A theory of tensor products for module categories for
a vertex operator algebra, II, Selecta Math. (N.S.) 1 (1995), 757786.
[HL4] Y.-Z. Huang and J. Lepowsky, A theory of tensor products for module categories for
a vertex operator algebra, III, J. Pure Appl. Alg. 100 (1995), 141-171.
68
[I] N. Ishibashi, The boundary and crosscap states in conformal eld theories, Mod.
Phys. Lett. A4 (1989).
[Ko1] L. Kong, Full eld algebras, operads and tensor categories, Adv. Math. 213 (2007)
271-340 [math.QA/0603065]
[Ko2] L. Kong, Open-closed eld algebras, Commun. Math. Phys. 280, 207-261.
[Kont] M. Kontsevich, Operads and motives in deformation quantization, Letters in Math-
ematical Physics 48: 35-72, (1999).
[Ki] A. Kirillov Jr. On an inner product in modular tensor categories, J. Amer. Math.
Soc 9 (1996), no 4, 11351169.
[Kr] I. Kriz, On spin and modularity of conformal eld theory, Ann. Sci.

Ecole Norm.
Sup. (4) 36 (2003), no. 1, 57112.
[KR] L. Kong and I. Runkel, Cardy algebras and sewing constrains, I (arXiv:0807.3356),
II, in preparation.
[La] C. I. Lazaroiu, On the structure of open-closed topological eld theory in two di-
mensions, Nucl. Phys. B603 (2001), 497530.
[Le] D.C. Lewellen, Sewing constraints for conformal eld theories on surfaces with
boundaries, Nucl. Phys., B372 (1992) 654.
[Ly] V. Lyubashenko, Modular transformations for Tensor categories, J. Pure Appl. Alg.
98 (1995) no. 3, 297-327.
[LL] J. Lepowsky, H.-S. Li, Introduction to vertex operator algebras and their represen-
tations. Progress in Mathematics, 227 Birkhauser Boston, Inc. Boston, MA, 2004.
[LP] A. Lauda and H. Pfeier, Open-closed strings: Two-dimensional extended TQFTs
and Frobenius algebras, Topology Appl. 155 (2008) 623666.
[Mi1] M. Miyamoto, Modular invariance of vertex operator algebras satisfying C
2
-
coniteness. Duke Math. J. 122 (2004), no. 1, 51-91.
[Mi2] M. Miyamoto, Intertwining operators and modular invariance, preprint, arXiv:
math.QA/0010180.
[Mo1] G. Moore, Some comments on branes, G-ux, and K-theory, Internat. J. Modern
Phys. A16 (2001), 936944.
[Mo2] G. Moore, D-branes, RR-Fields and K-heory, I, II, III, VI, Lecture notes for the ITP
miniprogram: The duality workshop: a Math/Physics collaboration, June, 2001;
http://online.itp.ucsb.edu/online/mp01/moore1
[MSei1] G. Moore and N. Seiberg, Classical and quantum conformal eld theory, Comm.
Math. Phys. 123 (1989), 177254.
[MSei2] G. Moore and N. Seiberg, Naturality in conformal eld theory, Nucl. Phys. B313
(1989) 1640.
69
[MSei3] G. Moore and N. Seiberg, Lecture on RCFT, Physics Geometry and Topology, Edited
by H.C. Lee, Plenum Press, New York, 1990.
[MSeg] G. Moore and G. Segal, D-branes and K-theory in 2D topological eld theory,
hep-th/0609042.
[RT] N. Reshetikhin and V.G. Turaev, Invariants of 3-manifolds via link polynomials of
quantum groups, Invent. Math. 103, No. 3 (1991) 547-597.
[Se1] G. Segal, The denition of conformal eld theory, preprint, 1988; also in: Topology,
geometry and quantum eld theory, ed. U. Tillmann, London Math. Soc. Lect. Note
Ser., Vol. 308. Cambridge University Press, Cambridge, 2004, 421577.
[Se2] G. Segal, Topological structures in string theory, R. Soc. Lond. Philos. Trans. A359
(2001), 13891398.
[So] H. Sonoda, Sewing conformal eld theories, I, II. Nucl. Phys. B311 (1988) 401416,
417432.
[T] Turaev, Quantum invariant of knots and 3-manifolds, de Gruyter Studies in Math-
ematics, vol. 18, Walter de Gruyter, Berlin, 1994.
[V] A. A. Voronov, The Swiss-cheese operad, in: Homotopy invariant algebraic struc-
tures, in honor of J. Michael Boardman, Proc. of the AMS Special Session on Ho-
motopy Theory, Baltimore, 1998, ed. J.-P. Meyer, J. Morava and W. S. Wilson,
Contemporary Math., Vol. 239, Amer. Math. Soc., Providence, RI, 1999, 365373.
[Z] Y.-C. Zhu, Modular invariance of vertex operator algebras, J. Amer. math. Soc. 9
(1996), 237-302.
Max-Planck-Institute for Mathematics in the Sciences, Inselstrasse 22, D-04103,
Leipzig, Germany
and
Institut Des Hautes

Etudes Scientifiques, Le Bois-Marie, 35, Route De Chartres,
F-91440 Bures-sur-Yvette, France
Max-Planck-Institute for Mathematics, Vivatsgasse 7, D-23111 Bonn, Germany
E-mail address: kong@mpim-bonn.mpg.de
70