
From here on, the construction of CFT correlators is the same as in [FRS1], but we go through it in
some detail in Section 5to stress the effect of including parity.
Let Σ be an oriented world sheet, possibly with boundaries and defect lines. Bulk insertions
are labelled by a tuple (
U, ¯
V, δ, ϕ
) and boundary fields by (
W, ν, ψ
), where
U, ¯
V∈ C
(not
b
C
) give
the holomorphic and antiholomorphic representation the bulk field transforms in,
W∈ C
is the
representation for the boundary field,
δ, ν ∈ {±
1
}
describe the parity of the fields, and
ϕ, ψ
take
values in appropriate multiplicity spaces which depend on the theory under consideration.
From Σ one constructs the double
e
Σ
= (Σ
⊔
Σ
rev
)
/∼
, where Σ
rev
is an orientation reversed copy of
Σ and
∼
identifies boundary points of Σ and Σ
rev
. For example, if Σ is a disc, then
e
Σ
is a sphere. A
bulk insertion on Σ splits into two points on
e
Σ
, one on Σ labelled by
U
and one on Σ
rev
labelled by
¯
V
. Boundary insertions result in a single marked point on
e
Σ
labelled by
W
. The first key ingredient
of the TFT construction is:
The correlator for a world sheet Σ of an oriented parity CFT is an element of the space of
conformal blocks on the double e
Σ, i.e. in Bl(Σ) := b
ZC(e
Σ).
Specifying an oriented parity CFT now amounts to, firstly, giving the multiplicity spaces for bulk
fields in terms of
U, ¯
V
and the parity
ϵ
(which may be zero-dimensional), and ditto for boundary
fields. This specifies the field content of the theory. Secondly, one has to give a collection of vectors
Corror
(Σ)
∈Bl
(Σ). The (bi)linear combinations of conformal blocks described by these vectors are
then the correlators of the oriented parity CFT.
The collection
{Corror
(Σ)
}Σ
has to satisfy two consistency conditions: it has to be monodromy free,
which amounts to mapping class group invariance, and it has to be compatible with gluing of world
sheets, see Section 5.4 for details. These consistency conditions can be expressed via the TFT b
ZC.
The second key point of the TFT construction is that consistent collections of correlators for CFTs
of type 2
○can be constructed from a suitable algebraic input:
From a symmetric special Frobenius algebra
B∈b
C
one can construct a consistent collection
of correlators
Corror
B
(Σ)
∈Bl
(Σ). Boundaries of Σ are labelled by
B
-modules and defect
lines by B-B-bimodules.
The correlators
Corror
B
(Σ) are described via the TFT
b
ZC
as follows: From Σ one obtains a bordism
MΣ:∅ −→ e
Σ
which contains a
b
C
-decorated ribbon graph defined in terms of
B
and the field insertions,
boundaries and line defects on Σ. As a 3-manifold,
MΣ
= Σ
×
[
−
1
,
1]
/∼
, where (
z, t
)
∼
(
z, −t
) for
all z∈∂Σ, t∈[−1,1]. For example, if Σ is a disc, then MΣis a 3-ball. One then sets
Corror
B(Σ) := b
ZC(MΣ)∈Bl(Σ) .(1.5)
If in this construction one takes
B∈ C ⊂ b
C
, i.e.
B
is purely parity even, one recovers precisely the
construction in [FRS1] of correlators of oriented CFTs without parity, which is type
1
○
in the above
table.
A more intuitive way to understand this construction is to think of
MΣ
as a fattening of the world
sheet Σ, together with a surface defect inserted in place of the embedded copy of Σ in
MΣ
. The surface
defect determines how the holomorphic and antiholomorphic fields combine to form a consistent
collection of correlators. This interpretation of the construction in [FRS1] was given in [KS], and a
detailed study of surface defects in Reshetikhin-Turaev TFTs can be found in [FSV,CRS2]. We will,
however, not elaborate more on this point of view in the present paper.
5