Code Conformal Field Theory and Framed Algebra Arxiv:2104.10094V1

Code conformal ﬁeld theory and framed algebra

Yuto Moriwaki *

Research Institute for Mathematical Sciences, Kyoto University Kyoto, Japan

Abstract. It is known that there are 48 Virasoro algebras acting on the monster conformal field theory. We call conformal field theories with such a property, which are not necessarily chiral, code conformal field theories. In this paper, we introduce a notion of a framed algebra, which is a finite-dimensional non-associative algebra, and showed that the category of framed algebras and the category of code conformal field theories are equivalent. We have also constructed a new family of integrable conformal field theories using this equivalence. These conformal field theories are expected to be useful for the study of moduli spaces of conformal field theories.

Introduction

In physics, the moduli space of two-dimensional conformal field theories has a geo- metric structure by deformations, and plays an important role in string theory. Conformal field theories consisting of only holomorphic fields (chiral conformal field theories) were mathematically formulated by Borcherds and are called vertex algebras or vertex operator algebras (VOAs) [B, FLM]. Since chiral conformal field theories are not deformable, their moduli space is discrete. For example, it has been predicted by physics that the moduli space of modular invariant chiral conformal field theories with central charge (24, 0) arXiv:2104.10094v2 [math.QA] 14 Sep 2021 consists of 71 VOAs [S], which has been studied by mathematicians (see for example [DM, LS3]). One of them is the monster VOA V\ whose automorphism group is the monster group, which is the largest sporadic finite simple group. The VOA V\ contains the mutually 1 ⊕48 commutative 48 Virasoro algebras with central charges 2 , Vir 1 [DMZ]. Dong, Griess 2 and Hohn¨ introduced an algebra called a framed VOA and investigated such an algebra in l general [DGH]. A framed VOA is a VOA with central charge 2 that contains the l-tensor 1 product of Virasoro VOAs with central charge 2 as a sub VOA. From a framed VOA V, l subgroups CV , DV of Z2 called codes are constructed, which play an important role in the study of framed VOAs (see for example [DGH, Mi1, LY]).

*email: moriwaki.yuto (at) gmail.com

1 2

l r In this paper, we call a conformal field theory with central charge ( 2 , 2 ) that contains the l-tensor produce of Virasoro VOA in the holomorphic part and the r-tensor product of Virasoro VOA in the anti-holomorphic part an (l, r)-code conformal field theory. The monster group is an exceptional symmetry, but code conformal field theories are not exceptional, but are considered to be ubiquitous. In fact, of the 71 chiral conformal field theories with central charge (24, 0) that are expected to exist, 56 are known to be (48, 0) code conformal field theories (framed VOA) [La, LS1, LS2]. Figure 1 shows the moduli space of conformal field theories with central charge (1, 1) as expected in physics [Gi, DVV1, DVV2]. Among them, (Ising)2 and Dirac are (2, 2)-code conformal field theories. The line in the figure corresponds to a deformation of conformal field theories. In the case of the central charge (1, 1), by deforming the code conformal field theory (Ising)2 and Dirac, all conformal field theories can be constructed except for the three points in the upper right of the figure. The purpose of this paper is to construct a family of mod- igure ular invariant code conformal field theories, and also to con- F 1. struct new conformal field theories by considering their current-current deformation, which is mathematically constructed in our previous paper [Mo3]. There, we introduced an algebra called a full vertex operator algebra and gave a mathematical formulation of two-dimensional conformal field theories. In this paper, based on the full vertex operator algebra, we introduce the notion of an (l, r)-framed full VOA. An (l, r)-framed full VOA is a mathematical formulation of the (l, r)-code conformal field theory. We also introduced a notion of an (l, r) framed algebra which is a non-associative finite dimensional algebra. The first main result is the following theorem (Theorem 4.3 and Theorem 7.1):

Main Theorem 1. The category of (l, r)-code conformal ﬁeld theories ((l, r)-framed full VOA) and the category of (l, r)-framed algebras are equivalent as categories.

Therefore, the construction and classification of code conformal field theories can be reduced to the construction and classification of framed algebras. r For a positive integer r > 0, a subgroup of Z2 is called a code (a linear code). For g = (g1,..., gr) ∈ Z2, |g|∆ = #{i | gi = 1} ∈ Z≥0 is called the code length. A subgroup r of Z2 is determined only by the rank, but by considering the structure of the code length, r there are various possible codes for the same rank. For any code G ⊂ Z2 satisfying (1,..., 1) ∈ G, we explicitly give a finite dimensional vector space S G and a product

· : S G × S G → S G on it. The second result is the following (Theorem 5.7).

r Main Theorem 2. For any code G ⊂ Z2 satisfying (1,..., 1) ∈ G,S G is a simple (r, r)- framed algebra. 3

Together with the first result, we can construct a code conformal field theory FG with r r central charge ( 2 , 2 ). We also showed that the torus partition function of this conformal field theory is a real analytic modular function on the upper half plane and is modular invariant (Corollary 5.21). The modular invariance is a physically required condition for the theory to be well-defined at higher genus [FMS]. r r Let ha1,..., ani be the subgroup of Z2 generated by a1,..., an ∈ Z2. For example, if 1 r = 1, G = h1i = Z2, then the corresponding conformal field theory is the critical Ising 1 1 model with central charge ( 2 , 2 ). 2 2 For r = 2, the central charge of the corresponding code conformal field theory is ( 2 , 2 ). The G = h(1, 1)i corresponds to Dirac and the G = h(1, 0), (0, 1)i to Ising2 in the figure. Also, if r = 3, G = h(1, 1, 1)i, then it is the level 2 SO(3)-WZW model. In general, if r ≥ 4, then G = h(1, 1,..., 1)i is the level 1 SO(r)-WZW model. Also, by taking more nontrivial codes G, we can construct many new families of conformal field theories. For example, in Section 5.4, we classify the all nontrivial codes r G ⊂ Z2 for r ≤ 6 as in Table 1 and study the dimension of the frame algebra S G and the current of the conformal field theory FG (the Lie algebra of the chiral VOA).

Table 1. all indecomposable code CFTs of rank r ≤ 6

r code G current dim S G name 1 h1i 0 3 critical Ising model √ 2 h11i SO(2) 10 toroidal compactiﬁcation R = 2 3 h111i SO(3) 36 SO(3)-WZW model at level 2 4 h1111i SO(4) 136 SO(4)-WZW model at level 1 ⊥ even h1111i 0 82 G4 5 h11111i SO(5) 528 SO(5)-WZW model at level 1 2;1,1 h11000, 00111, 01100i U(1) 276 G5 6 h11111i SO(6) 2080 SO(6)-WZW model at level 1 2;1,1 h110000, 001111, 101000i SO(3) 1000 G6 2 2;1,2 h110000, 001111, 101100i U(1) 936 G6 4;2 ⊥ h110000, 001100, 000011, 101010i 0 756 (E6 ) ⊥ even h111111i 0 730 G6

An important feature of code conformal ﬁeld theory is that important physical quantities such as correlation functions can be described combinatorially using codes. For example, an example of a four-point correlation function in code conformal ﬁeld theory, excluding the scalar factor, is as follows (Proposition 5.12):

r − 0 1 0 2 0 3 1 2 1 3 2 3 −r 8 r−|α α +α α +α α +α α +α α +α α |l 2 Π0≤i< j≤3 (zi − z j)(¯zi − z¯ j) F(z0, z1, z2, z3)

0 1 2 3 0 2 1 3 0 3 1 2 |α α +α α |l |α α +α α |l |α α +α α |l G01,23(z0, z1, z2, z3) G02,13(z0, z1, z2, z3) G03,12(z0, z1, z2, z3) 4 where F(z0, z1, z2, z3) and Gi j,kl(z1, z2, z3, z4) are deﬁned by

1 1 1 1 2 F(z0, z1, z2, z3) = |(z0 − z1)(z2 − z3)| 2 + |(z0 − z2)(z1 − z3)| 2 + |(z0 − z3)(z1 − z2)| 2 ,

1 1 1 1 2 G01,23(z0, z1, z2, z3) = −|(z0 − z1)(z2 − z3)| 2 + |(z0 − z2)(z1 − z3)| 2 + |(z0 − z3)(z1 − z2)| 2 ,

1 1 1 1 2 G02,13(z0, z1, z2, z3) = |(z0 − z1)(z2 − z3)| 2 − |(z0 − z2)(z1 − z3)| 2 + |(z0 − z3)(z1 − z2)| 2 ,

1 1 1 1 2 G03,12(z0, z1, z2, z3) = |(z0 − z1)(z2 − z3)| 2 + |(z0 − z2)(z1 − z3)| 2 − |(z0 − z3)(z1 − z2)| 2 .

1 Although Gi j,kl(z1, z2, z3, z4) is not single-valued on the configuration space X4(CP ) = 1 4 {(z0, z1, z2, z3) ∈ (CP ) | zi , z j} due to the existence of square roots, their monodromies cancel each other and the correlation function defines a single-valued real analytic func- 1 0 1 2 3 r+r 0 1 2 3 tion on X4(CP ). Also, α , α , α , α are elements in Z2 , and |α α + α α |l etc. are integers determined by the code length. Thus, the code conformal field theories give a family of integrable (various physical quantities can be computed) conformal field theories. It is believed in physics that conformal field theory can be deformed by a field of conformal weight (1, 1) which satisfies a good property called exactly marginal. If a (1, 1) field is the product of a (1, 0) field and a (0, 1) field (called current fields), then the conditions for the field to be exactly marginal is known in physics [CS]. The deformation generated by such a (1, 1) field is called a current-current deformation in physics [FR]. In [Mo3], the current-current deformation of a conformal field theory is mathematically constructed in terms of a full VOA. The conditions for a code conformal field theory FG to admit a current-current deformation can be described combinatorially in terms of codes. ∨ ∨ r To be more precise, the dual code G is defined by G = {α ∈ Z2 | |αG|∆ ⊂ 2Z}. Then, we obtained the following result (Theorem 6.8).

Main Theorem 3. Assume that there are mutually orthogonal length two elements α1, . . . , αN ∨ i in the dual code G , i.e., |αiα j|∆ = 0 for i , j and |α | = 2 for any i = 1,..., N.

Then, the framed full VOA FG admits a current-current deformation parametrized by O(N, N)/O(N) × O(N). Here, O(a, b) is the orthogonal group with the signature (a, b).

∨ For example, if G = h(1, 1)i, then G = h(1, 1)i and thus FG admits a one-dimensional deformation family O(1, 1)/O(1) × O(1) R. In fact, this family of deformations corresponds to the transversal line through the Dirac in the ﬁgure. Combining this result with Main theorem 2, we can construct a continuous family of conformal ﬁeld theories. The four-point correlation function of the deformed CFT

FG can also be calculated explicitly in terms of codes again. An example of the four- point correlation function of the deformed code conformal ﬁeld theory corresponding to σ ∈ O(N, N) is as follows (see Section 6.1 for the precise deﬁnition):

1 pσ−1(si,si),pσ−1(s j,s j) + 1 N− r −r+3N r−2N 4 ( )l 4 8 2 F(z0, z1, z2, z3) Π0≤i< j≤3 (zi − z j)(¯zi − z¯ j) . 5

In the following, we will briefly explain the basic tools for study coded conformal field theories: finite dimensional algebra, the framed algebra, and Main theorem 1. 0.1. What is the framed algebra? The fields of a two-dimensional conformal field theory are operator-valued real analytic functions and have an algebra structure. The aforementioned chiral conformal field theory (vertex operator algebra) is a subalgebra of the conformal field theory consisting of holomorphic fields (see Proposition 2.6). A conformal field theory is a module of the vertex operator algebras of holomorphic part VL and anti-holomorphic part VR. Therefore, the construction of conformal field theories is done by first classifying the modules of the vertex operator algebras and then combining the modules to satisfy the axiom [MS]. The combination of modules satisfying the axiom corresponds to the frobenius algebra object in the representation category of VL ⊗ VR [FRS]. The above considerations in physics were considered mathematically by Huang and Lepowsky. In particular, they showed that the representation category of a good vertex operator algebra has the structure of a braided tensor category [HL1, HL2, HL3, Hu1]. Fur- thermore, in [HK], they constructed a conformal field theory called the diagonal model, which is the diagonal combination of all the irreducible modules of a VOA. On the other hand, there are many important non-diagonal conformal field theories. Our code conformal field theory is one such example. Another example is the case where the vertex operator algebra of the holomorphic and anti-holomorphic parts are the affine Heisenberg vertex operator algebra. Then, the frobenius algebra object in the representation category can be classified by using a certain class of associative algebras that we called AH-pairs [Mo1]. AH pairs are classified using the cohomology of groups, and from an AH-pair we can construct a non-diagonal conformal field theory [Mo2]. These non-diagonal conformal field theories are geometrically meaningful conformal field theories arising from the toroidal compactification of string theory (flat target space) [P]. However, the braided tensor category structure of the representation category of the affine Heisenberg vertex operator algebra is almost trivial, and the resulting theory is also trivial. What we investigate in this paper are conformal field theories with symmetry of the ⊕l ⊕ ⊕r 1 Virasoro algebras Vir 1 Vir 1 . Let L( 2 , 0) be a simple Virasoro vertex operator algebra 2 2 1 of central charge 2 [FZ] and Ll,r(0) be a full vertex operator algebra defined by Ll,r(0) = 1 ⊗l ⊗ 1 ⊗r 1 ⊗l 1 ⊗r L( 2 , 0) L( 2 , 0) , the tensor product of VOA L( 2 , 0) and conjugate VOA L( 2 , 0) (see Section 2). Following the framed vertex operator algebra, we call a full vertex operator algebra which contains Ll,r(0) as a sub full VOA an (l, r)-framed full vertex operator algebra. 1 The module of the Virasoro vertex operator algebra L( 2 , 0) is classified and is known 1 1 1 1 1 to be the direct sum of three irreducible modules: L( 2 , 0), L( 2 , 2 ), andL( 2 , 16 ). It is also known that its representation category has a nontrivial braided tensor category structure l+r [FF, DMZ, W]. Thus Ll,r(0) has 3 irreducible modules and the representation category is nontrivial. 6

Now let us recall the example of conformal field theories constructed from the above affine Heisenberg vertex operator algebra. It is important to note that in the algebra F of the conformal field theory, the lowest weight space ΩF,H of the affine Heisenberg Lie algebra Hˆ has an algebra structure [Mo3]. In particular, if the affine Heisenberg vertex operator algebra is sufficiently large in F, then ΩF,H becomes an AH pair, i.e., an associative algebra [Mo1, Mo3]. For a code conformal field theory F, the lowest weight space

S F of the Virasoro algebras is expected to inherit an algebra structure. The axiomatization of this algebra structure is the framed algebra (see Section 4). The framed algebra is a non-commutative and non-associative finite dimensional alge- 1 bra, reflecting the fact that the representation category of L( 2 , 0) is nontrivial. The Main Theorem 1 states that the functor which takes the lowest weight space of the Virasoro algebras gives the equivalence between the category of framed full VOAs and the category of framed algebras. 0.2. Definition of framed algebra. In this section, we will give a more detailed definition of a framed algebra. For simplicity, nh,h¯ L 1 1 ¯ let l = r = 1. In this case, the code conformal field theory is F = ¯ 1 1 L( , h) ⊗ L( , h) , h,h∈{0, 2 , 16 } 2 2 where nh,h¯ ∈ Z≥0 is the multiplicity. For convenience of explanation, we further assume ¯ 1 ⊗ 1 ¯ ∈ { 1 1 } that nh,h¯ = δh,h¯ and set L1,1(h, h) = L( 2 , h) L( 2 , h) for h IS = 0, 2 , 16 . That is, L ¯ F = h∈IS L1,1(h, h) (diagonal model). A full vertex operator algebra is a vector space F equipped with a linear operator, called a vertex operator X Y(−, z): F → EndF[[z, z¯, |z|R]], a 7→ Y(a, z) = a(r, s)z−r−1z¯−s−1, r,s∈R which satisfies some axiom (Section 2).

From the decomposition of F as an L1,1(0)-module, the vertex operator is also decomposed into the sum of ! (h1, h¯ 1) Y (−, z): L1,1(h2, h¯ 2) → Hom L1,1(h3, h¯ 3), L1,1(h1, h¯ 1) . (h2, h¯ 2), (h3, h¯ 3)

¯ ¯ By [DMZ], Y (h1,h1) (−, z) = Y h1 (−, z) ⊗ Y h1 (−, z¯), that is, the tensor product (h2,h¯2),(h3,h¯3) h2,h3 h¯2,h¯3 of a holomorphic vertex operator and an anti-holomorphic vertex operator (see Section 2.6). From the axioms of the full vertex operator algebra, Y h1 (−, z) is an intertwining h2,h3 operator of modules of a vertex operator algebra introduced in [FHL] (Lemma 2.9). The 1 intertwining operators among the L( 2 , 0)-modules are classiﬁed (see for example [DMZ]). 1 2 For example, there is no Y 1 1 (−, z). The allowed triples are called the fusion rule, which 2 , 16 is a fundamental quantity in the representation theory of a vertex operator algebra. For

λ2, λ3, by setting λ2 ? λ3 to be the set of all possible λ1, the fusion rule forms the product ? : IS × IS → P(IS). The most important condition of the axiom of a full vertex operator algebra is, roughly speaking, the following: 7

∨ FL) For any a1, a2, a3 ∈ F and u ∈ F (the dual vector space of F), The formal u Y a z Y a z a u Y a z Y a z a power series ( ( 1, 1) ( 2, 2) 3) and ( ( 2, 2) ( 1, 1) 3) with formal vari- × ables z1, z¯1, z2, z¯2 converge to a single-valued real analytic function on Y2(C ) in × × {(z1, z2) ∈ Y2(C ) | |z1| > |z2|} and {(z1, z2) ∈ Y2(C ) | |z2| > |z1|}, respectively, × and their analytic continuations are the same function. Here, Y2(C ) = {(z1, z2) ∈

C | z1 , z2, z1 , 0, z2 , 0}. u Y a z Y b z c correlation function The above formal power series ( ( , 1) ( , 2) ) is called a . Corre- lation functions are the most important physical quantity of conformal field theory. Fol- lowing [MS] and [Hu1, Hu2, HK], the code conformal field theory can be constructed in the following two steps: S1) We first compute the correlation functions of all possible combinations of intertwining operators Y h1 (−, z) from the representation theory of the Virasoro ver- h2,h3 tex operator algebra. These functions are called conformal blocks and has mon- × odromies, i.e., is a multivalued holomorphic function on Y2(C ). S2) By combining holomorphic and anti-holomorphic intertwining operators, we second achieve a monodromy invariant correlation function that satisfies [FL]. The first step (S1) was basically done by [BPZ]. In this paper, the calculation will be done in Section 3. It is important to note that even if we determine the input states, h , h , h , h ∈ IS and u ∈ L( 1 , h )∨ and a ∈ L( 1 , h )(i = 1, 2, 3), the conformal block is 0 1 2 3 2 0 i 2 i h0 h not determined. In fact, the conformal block is given by u(Y (a1, z1)Y (a2, z2)a3), h1,h h2,h3 so it depends on the intermediate state h ∈ IS. The set of all possible intermediate states in the four states scattering is given by

.A(h0, h1, h2, h3) = {h0 ∈ h1 ? h, h ∈ h2 ? h3}.

From the Ward identity (Lemma 1.8), we may assume that u, a1, a2, a3 are the lowest weight vector hh0| , |h1i , |h2i , |h3i. In this case, the nontrivial conformal blocks ! ! h h Ch (z , z ) = hh | Y 0 (|h i , z )Y (|h i , z ) |h i h0,h1,h2,h3 1 2 0 1 1 2 2 3 h1, h h2, h3 for h ∈ A(h0, h1, h2, h3) are given in table below (for the precise statement, see Propo- sition 3.8). As you can see from the table, the conformal block has branches along × {z1 = 0} ∪ {z2 = 0} ∪ {z1 = z2}. In particular, it is not single-valued on Y2(C ). Note that in general, u(Y(a1, z1)Y(a2, z2)a3) converges at |z1| > |z2|. So we take a path γ : [0, 1] → × Y2(C ) connecting a point in the region |z1| > |z2| with a point in the region |z2| > |z1|. Then, the analytic continuation of Ch (z , z ) along the path γ is a linear sum of h0,h1,h2,h3 1 2 h0 h,h0 {C (z , z )} 0∈ . Thus, we can deﬁne a connection matrix B (γ) ∈ C h0,h2,h1,h3 2 1 h A(h0,h2,h1,h3) h0,h1,h2,h3 0 for (h, h ) ∈ A(h0, h1, h2, h3) × A(h0, h2, h1, h3) by X 0 0 A Ch (z , z ) = Bh,h (γ)Ch (z , z ) γ h0,h1,h2,h3 1 2 h0,h1,h2,h3 h0,h2,h1,h3 2 1 0 h ∈A(h0,h2,h1,h3) 0 (see Section 3 for more precise deﬁnition). In Section 3.3, we determine Bh,h (γ) ∈ C h0,h1,h2,h3 × for some path γ0 : [0, 1] → Y2(C ) (Proposition 3.13). It is easy to turn the above result 8 Table 2. Conformal blocks

(h , h , h , h ) h Ch (z , z ) 0 1 2 3 h0,h1,h2,h3 1 2 1 1 1 1 1 1 1 1 1 − 1 1 1 { − } 8 2 2 2 2 − 2 2 ( 16 , 16 , 16 , 16 ) 0 2 z1z2(z1 z2) (z1 + z2 ) + (z1 z2 ) 1 1 1 1 1 1 − 1 1 1 { − } 8 2 2 2 − 2 − 2 2 2 2 z1z2(z1 z2) (z1 + z2 ) (z1 z2 ) 1 1 1 1 { − }−1 2 − 2 ( 2 , 2 , 2 , 2 ) 0 z1z2(z1 z2) (z1 z1z2 + z2) 1 1 1 1 1 1 − 8 − z { − } 2 − 2 ( 2 , 2 , 16 , 16 ) 0 z2 z1(z1 z2) (z1 2 ) 1 1 1 1 1 1 1 1 − 8 − { − } 2 − ( 2 , 16 , 2 , 16 ) 16 2 z1 z2(z1 z2) (z1 2z2) 1 1 1 1 1 1 − 1 −1 2 − ( 16 , 2 , 2 , 16 ) 16 2 (z1z2) (z1 z2) (z1 + z2) 1 1 1 1 1 1 − 1 − 1 { } 2 − 8 ( 2 , 16 , 16 , 2 ) 16 2 z1z2 (z1 z2) (z1 + z2) 1 1 1 1 1 1 −1 − 1 { − } 2 − ( 16 , 2 , 16 , 2 ) 16 2 z1 z2(z1 z2) (z1 2z2) 1 1 1 1 1 −1 − z2 { − } 2 − ( 16 , 16 , 2 , 2 ) 0 z2 z1(z1 z2) (z1 2 )

(l,r) l r at l = r = 1 into a general l, r ∈ Z≥0. Set IS = IS × IS . This set parametrizes all 1 ⊗l ⊗ 1 ⊗r i i i ¯ i ¯ i ∈ the irreducible modules of Ll,r(0) = L( 2 , 0) L( 2 , 0) . For λ = (h1,..., hl, h1 ..., hr) (l,r) ¯ 0 1 2 3 0 0 0 ¯ 0 ¯ 0 IS , λ = (h1,..., hl, h1,..., hr) ∈ A(λ , λ , λ , λ ) and λ = (h1,..., hl , h1,..., hr) ∈ A(λ0, λ2, λ1, λ3), a multi-index connection matrix is deﬁned by

0 0 0 h ,h h¯ j,h¯ λ,λ ≡ l i i r j ∈ C Bλ0,λ1,λ2,λ3 Πi=1B 0 1 2 3 Π j=1B¯0 ¯1 ¯2 ¯3 , hi ,hi ,hi ,hi h j ,h j ,h j ,h j

¯ ¯0 ¯ ¯0 h j,h j h j,h j where B¯0 ¯1 ¯2 ¯3 is the complex conjugate of B¯0 ¯1 ¯2 ¯3 ∈ C. It is easy to show that the h j ,h j ,h j ,h j h j ,h j ,h j ,h j multi-index connection matrix gives the connection matrix for the multi-index conformal blocks, which is defined naturally for Ll,r(0)-modules and their intertwining operators. Based on the above preparation, the framed algebra can be defined as follows. Let L (l,r) S = λ∈IS(l,r) S λ be a finite-dimensional IS -graded vector space equipped with a linear map · : S ⊗ S → S and a distinguished non-zero element 1 ∈ S 0 such that: ¯ (l,r) Pl Pr ¯ FA1) For any λ = (h1,..., hl, h1,..., hr) ∈ IS , S λ = 0 unless i=1 hi − j=1 h j ∈ Z. FA2) S 0l+r = C1 and for any a ∈ S , a · 1 = 1 · a = a; 1 2 (l,r) L FA3) For any λ , λ ∈ IS , a ∈ S λ1 and b ∈ S λ2 , a · b ∈ λ∈λ1?λ2 S µ. i (l,r) 0 0 2 1 3 FA4) For any λ ∈ IS , ai ∈ S λi and λ ∈ A(λ , λ , λ , λ ) (i=1,2,3),

X λ,λ0 · · 0 · · a2 λ0 (a1 λ a3) = Bλ0,λ1,λ2,λ3 a1 λ0 (a2 λ a3), λ∈A(λ0,λ1,λ2,λ3)

where ·λ is the composition of · : S ⊗ S → S and the projection S → S λ.

For a framed algebra S , we can naturally deﬁne a vertex operator YS (−, z) on the Ll,r(0)- L ⊗ module FS = λ∈IS(l,r) Ll,r(λ) S λ. Condition (FA1) implies that all the correlation functions of YS (−, z) are single-valued around z2 = 0. And (FA2) means that all the correlation functions of YS (−, z) are single-valued around z1 = z2. The scale transfor- × × mation and rotation C acts on Y2(C ) with (x, y) 7→ (ρx, ρy). Hence, using conformal invariance of the vertex operator, the correlation functions can be viewed as a function on × 1 1 Y2/C = CP \{0, 1, ∞}. Since the fundamental group π1(CP \{0, 1, ∞}) is generated 9 by two elements, (FA1) and (FA2) imply that all the correlation functions is single-valued × real analytic functions on Y2(C ) and satisfy (FL). Conversely, given a framed full vertex operator algebra F, we can show that its lowest weight space S F becomes a framed algebra (Theorem 4.3). The most important observation of this paper is that the multi-index connection matri- λ,λ0 ces Bλ0,λ1,λ2,λ3 have a very simple combinatorial representation as follows (Theorem 3.21):

2 3 1 3 0 1 2 1 2 0 3 2 1 0 3 1 2 2 1 1 2 0 3 B(d +d ,c),(d +d ,c ) − |c c | − |d c (c +c )|+|d c (c +c )|i−|d c |−|d c |+|d d (c +c )| (d0,c0),(d1,c1),(d2,c2),(d3,c3) = ( 1) ( 1)

−πi 1 + i 1 2 3 1 − i 1 2 3 1 2 3 0 exp( |d1d2|)( )|d d d |l ( )|d d d |r (−i)|d d d (c+c )|. 8 2 2

i i 0 l+r i i i The meaning of the symbols is not explained here, but d , c , c, c ∈ Z2 and λ = (d , c ) l+r and | − |l, | − |r, | − | : Z2 → Z are maps deﬁned by the code length (see Section 3.4). By using this combinatorial representation, the structure of a framed algebra can be studied, and in particular Main Theorem 2 is proved. Finally we give the simplest non-trivial example of a framed algebra here. The following combination of conformal blocks deﬁne a single-valued real analytic function on × Y2(C ):

1 1 − 1 1 1 1 2 (z z¯ z z¯ (z − z )(¯z − z¯ )) 8 (z z¯ ) 2 + (z z¯ ) 2 + ((z − z )(¯z − z¯ )) 2 2 1 1 2 2 1 2 1 2 1 1 2 2 1 2 1 2 1 1 (0.1) = C0 (z , z )C0 (z , z ) + C 2 (z , z )C 2 (z , z ). 1 , 1 , 1 , 1 1 2 1 1 1 1 1 2 1 1 1 1 1 2 1 1 1 1 1 2 16 16 16 16 16 , 16 , 16 , 16 16 , 16 , 16 , 16 16 , 16 , 16 , 16

Therefore, it can be a correlation function. In fact, there is a framed algebra and a code conformal field theory that realize this correlation function. Let S Ising = C1 ⊕ Ca ⊕ Cd (1,1) be a three-dimensional vector space with the IS -grading, S 0,0 = C1, S 1 1 = Ca and 2 , 2 S 1 1 = Cd. Define a product on S Ising by a · a = 1, a · d = d · a = 1, and d · d = 1 + a 16 , 16 with unit 1. It is easy to verify that this algebra is a framed algebra. Also, the product d ·d = 1+a corresponds to (0.1). The conformal field theory corresponding to S Ising is the critical Ising model in physics. In fact, (0.1) agrees with the known correlation function of the critical Ising model [BPZ]. As mentioned in the introduction, S Ising is one of the family of framed algebras constructed in this paper with r = 1, G = h1i.

This paper is organized as follows. In Section 1, we collect some fundamental results 1 concerning the representation theory of the Virasoro algebra of central charge 2 []. In Section 2, we recall the definition and some results of a full vertex operator algebra and introduce notions of framed full vertex operator algebra and Virasoro conformal blocks 1 of central charge 2 (see also [BPZ]). The Virasoro conformal blocks will be studied in Section 3. We explicitly calculate all the Virasoro conformal blocks and the connection matrices. In Section 4, a notion of a framed algebra will be introduced and their general structure will be studied. In particular, Main theorem 1 will be shown. In Section 5, we r will construct a (r, r)-framed algebra S G from a code G ⊂ Z2 (Main Theorem 2). Many 10 properties of the associated full framed VOA will be studied there, for example, the modular invariance, correlation functions, the chiral symmetry. Also, we give a classification of codes up to r ≤ 6. In Section 6.1, the current-current deformation of framed full vertex operator algebra will be studied (Main Theorem 3). Acknowledgements I would like to offer my gratitude to my supervisor Professor Masahito Yamazaki for his support, encouragement and valuable discussions. This work was supported by the Research Institute for Mathematical Sciences, an In- ternational Joint Usage/Research Center located in Kyoto University. This work was also partially supported by World Premier International Research Center Initiative (WPI Ini- tiative), MEXT, Japan and the Program for Leading Graduate Schools, MEXT, Japan. The author gratefully acknowledges the Kavli Institute for the Physics and Mathematics of the Universe, the University of Tokyo, where a part of this paper was written and its hospitality during 2020.

Contents 1. Notations and preliminary results 11 1.1. Notations I – formal calculus 11 1.2. Notations II – correlators 13 1.3. Vertex operator algebras, modules and intertwining operators 15 1.4. Virasoro vertex operator algebra 18 1.5. Virasoro conformal block 19 2. Framed full vertex operator algebra 21 2.1. Full vertex operator algebras 21 2.2. Full vertex algebra and vertex algebra 23 2.3. Locality axiom 24 2.4. Deﬁnition of framed full vertex operator algebra 25 2.5. Multi-index Virasoro conformal block 25 2.6. Lowest weight space and conformal block 26 2.7. Induced vertex operator 27 3. Conformal block of Virasoro algebras 28 3.1. Conformal block involving vacuum state 29 3.2. Conformal block – general cases 30 3.3. Connection matrix 34 3.4. Multi-index connections 39 4. Framed algebra 43 4.1. Deﬁnition of framed algebra 43 4.2. Associativity, ideal and bilinear form 46 4.3. Code subalgebra 48

4.4. Classiﬁcation of C[CˆS ]-modules 51 11

4.5. Products among C[CˆS ]-modules 52 4.6. Code vertex operator algebra 53 5. Code conformal ﬁeld theory 55 5.1. Two-cocycle and representations 55 5.2. Construction of product 57 5.3. Modular invariance 66 5.4. Examples and classiﬁcation of codes 69 6. Deformation of code CFTs 72 6.1. Current-current deformation of full vertex algebras 72 6.2. Current-current deformation of code CFTs 75 7. Appendix 78 References 79

1. Notations and preliminary results In this section, we recall some notations from [Mo3] and some basic results of vertex operator algebras.

1.1. Notations I – formal calculus. Throughout this paper, we will use the following notations. We assume that the base ﬁeld is C unless otherwise stated. Let z andz ¯ be independent formal variables. We will use the notation z for the pair (z, z¯) and |z| for zz¯. For a vector space V, we denote by V[[zR, z¯R]] the set of formal sums

X s s¯ as,s¯z z¯ s,s¯∈R

R R such that as,s¯ ∈ V. The space V[[z , z¯ ]] contains various useful subspaces:

R X s V[[z ]] = { asz | as ∈ V} s∈R ± X n V[[z ]] = { anz | an ∈ V} n∈Z R X s s¯ V[[z, z¯, |z| ]] = { as,s¯z z¯ | as,s¯ = 0 unless s − s¯ ∈ Z} s,s¯∈R X n n¯ V[[z, z¯]] = { an,n¯ z z¯ | an,n¯ ∈ V}. n,n¯∈R We also denote by V((z, z¯, |z|R)) the subspace of V[[z, z¯, |z|R]] consisting of the series P s s¯ R s,s¯∈R as,s¯z z¯ ∈ V[[z, z¯, |z| ]] such that: 2 (1) For any H ∈ R,#{(s, s¯) ∈ R | as,s¯ , 0 and s + s¯ ≤ H} is ﬁnite.

(2) There exists N ∈ R such that as,s¯ = 0 unless s ≥ N ands ¯ ≥ N ± P n ± and V((z)) the subspace of V[[z ]] consisting of the series n∈Z anz ∈ V[[z ]] such that:

(1) There exists N ∈ R such that an = 0 unless n ≥ N. 12

The space V((z)) is called the space of formal Laurent series. Thus, V((z, z¯, |z|R)) is a generalization of the Laurent series to two-variables. R Let f (z) ∈ V((z, z¯, |z| )). By the assumption, there exists r0, r1, r2, · · · ∈ R such that

(1) r0 < r1 < r2 < ··· ;

(2) ri → ∞; (3) f (z) could be written as

X∞ X∞ i n m ri an,mz z¯ |z| , i=0 n,m=0

i where an,m ∈ C.

Remark 1.1. As seen above, C((z, z¯, |z|R)) is a Novikov ring with polynomial coeﬃcients.

We will also consider their combinations, e.g., V((y/x, y¯/x¯, |y/x|R))[x±, x¯±, |x|R], which is spanned by Xk Xl X i n+ri m+ri s s¯ an,m,r,s x x¯ (y/x) (¯y/x¯) i=1 n,m=−l s,s¯∈R

i i for some k, l ∈ Z>0 and ri ∈ R and an,m,s,s¯ ∈ V such that an,m,s,s¯ = 0 unless s − s¯ ∈ Z and i i there exists N such that an,m,s,s¯ = 0 unless s ≥ N ands ¯ ≥ N and {(s, s¯) ∈ R | an,m,s,s¯ , 0 and s + s¯ ≤ H} is ﬁnite for any H ∈ R. d d | |R Let dz and dz¯ be formal diﬀerential operators acting on V[[z, z¯, z ]] by d X X a zsz¯s¯ = sa zs−1z¯s¯ dz s,s¯ s,s¯ s,s¯∈R s,s¯∈R d X X a zsz¯s¯ = sa¯ zsz¯s¯−1. dz¯ s,s¯ s,s¯ s,s¯∈R s,s¯∈R

d | |s | |s −1 d d Since dz z = s z z , the diﬀerential operators dz and dz¯ acts on all the above vector spaces. ∈ | |R d ∈ We note that if f (z) V((z, z¯, z )) satisﬁes dz¯ f (z) = 0, then f (z) V((z)). R For any R ∈ R>0, set AR = {z ∈ C | 0 < |z| < R}, an annulus. Let f (z) ∈ C((z, z¯, |z| )). Then, there exists N ∈ R such that

N X s s¯ (1.1) |z| f (z) = as,s¯z z¯ . s,s¯∈R s,s¯≥0

We say the series f (z) is absolutely convergent around 0 if there exists R ∈ R>0 such P s+s¯ that the sum s,s¯∈R |as,s¯|R is convergent. In this case, f (z) is compactly absolutely- convergent to a continuous function deﬁned on the annulus AR. We note that the deﬁnition of the convergence is independent of the choice of N (for more details see [Mo3, Section 1.2]). 13

1.2. Notations II – correlators. The notion of a conformal singularity is introduced in [Mo3] in order to formulate two-dimensional conformal field theory. In this section, we briefly recall the definition of a conformal singularity and define a space of correlators in two-dimensional conformal field theory. 1 1 Let α1, . . . , αn ∈ CP and f be a C-valued real analytic function on CP \{α1, . . . , αn}. A chart (χ, α) of CP1 at a point α ∈ CP1 is a biholomorphism χ from an open subset U of CP1 to an open subset of C such that α ∈ U and χ(α) = 0. We say that f has a conformal 1 singularity at αi if for any chart (χ, αi) of CP at αi, there exists a formal power series

X s s¯ R as,s¯z z¯ ∈ C((z, z¯, |z| ))(1.2) s,s¯∈R −1 such that it is compactly absolutely-convergent to f ◦ χ (z) on the annulus AR = {z ∈

C | R > |z| > 0} for some R ∈ R>0. It is clear that the above condition is independent of a choice of a chart and the coeﬃcients of the series is uniquely determined by the chart. Let R f have a conformal singularity at αi. Denote by j(χ, f ) ∈ C((z, z¯, |z| )) the formal power −1 series which is compactly absolutely-convergent to f ◦ χ (z), and by F0,1,∞ the space of real analytic functions on CP1 \{0, 1, ∞} with possible conformal singularities at {0, 1, ∞}.

Examples of functions belonging to F0,1,∞ are

r r n n n |z| , |1 − z| , z , (1 − z) , (1 − z¯) ∈ F0,1,∞, where r ∈ R and n ∈ Z. For instance, the expansions of |1 − z|r are ∞ ! ! X r r j(z, |1 − z|r) = znz¯m, n m n,m=0 j(1 − z, |1 − z|r) = |z|r, ∞ ! ! X r r j(z−1, |1 − z|r) = zn−rz¯m−r, n m n,m=0 −1 ± ± ± where z, 1−z, z are charts of 0, 1, ∞, respectively. In fact, F0,1,∞ is a C[z , (1−z) , z¯ , (1− z)±, |1 − z|R]-module.

A non-trivial example of a function in F0,1,∞ is 1 √ √ (1.3) f (z) = |1 − 1 − z|1/2 + |1 + 1 − z|1/2 , Ising 2 which appears in a four point function of the two-dimensional critical Ising model. In this paper, we will show that (1.3) is actually a correlator of a full vertex algebra. The expansion of fIsing(z) around 0 with the chart z is (1.4) 2 + (zz¯)1/2/2 − z/4 − z¯/4 + (zz¯)1/2(z + z¯)/16 + zz¯/32 − 5z2/64 − 5¯z2/64 + ....

1/4 Since fIsing(z) satisﬁes the equations fIsing(z) = fIsing(1 − z) = (zz¯) fIsing(1/z), the expansions around 1 and ∞ are also of the form (1.2). Thus, fIsing(z) ∈ F0,1,∞.

Finally, we remark on the case that f ∈ F0,1,∞ is a holomorphic function. Recall that the ring of regular functions on the aﬃne scheme CP1 \{0, 1, ∞} is C[z±, (1 − z)±]. It is easy to show that a function in C[z±, (1 − z)±] has conformal singularities at {0, 1, ∞}. Thus, 14

± − ± ⊂ ∈ d 1 d − d C[z , (1 z) ] F0,1,∞. Conversely, let f F0,1,∞ satisfy dz¯ f = 2 ( dx i dy ) f = 0. Then, the condition for the conformal singularity implies that {0, 1, ∞} is a pole of f , thus, f is a meromorphic function on CP1. Hence, f ∈ C[z±, (1 − z)±].

1 Proposition 1.2. If f ∈ F0,1,∞ is a holomorphic function on CP \{0, 1, ∞}, then f ∈ C[z±, (1 − z)±].

In the definition of a vertex algebra (chiral conformal field theory), C[z±, (1−z)±] serves as a space of four point correlation functions and plays an important role (see for example the introduction of [Mo3]). Based on this fact, we introduced a space of correlation functions in non-chiral conformal field theory. Set U(x, y) = C((y/x, y¯/x¯, |y/x|R))[x±, x¯±, |x|R] and × 2 Y2(C ) = {(z1, z2) ∈ C | z1 , z2, z1 , 0, z2 , 0}.

× 1 Let η(z1, z2): Y2(C ) → CP \{0, 1, ∞} be the real analytic function deﬁned by η(z1, z2) = z2 × . For f ∈ F0,1,∞, f ◦ η is a real analytic function on Y2(C ). Denote by GCor2 the space z1 × of real analytic functions on Y2(C ) spanned by

α β (1.5) z1 z¯1 f ◦ η(z1, z2), where f ∈ F0,1,∞ and α, β ∈ R satisfy α − β ∈ Z. d d d d It is clear that GCor2 is closed under the product and the derivations , , , . dz1 dz¯1 dz2 dz¯2 d d α β α β d d α β Since (z1 +z2 )z z¯ f ◦η(z1, z2) = αz z¯ f ◦η(z1, z2) and (¯z1 +z¯2 )z z¯ f ◦η(z1, z2) = dz1 dz2 1 1 1 1 dz¯1 dz¯2 1 1 α β βz1 z¯1 f ◦ η(z1, z2), by using a formal calculus, we have:

d d d Lemma 1.3. Let µ(z1, z2) ∈ GCor2 satisfy (z1 + z2 )µ(z1, z2) = αµ(z1, z2) and (¯z1 + dz1 dz2 dz¯1 d z¯2 )µ(z1, z2) = βµ(z1, z2) for some α, β ∈ R. Then, there exists unique f ∈ F0,1,∞ such dz¯2 α β z1 that µ(z1, z2) = z z¯ f ( ). 1 1 z2

α β Let µ(z1, z2) = z1 z¯1 f ◦ η(z1, z2) in (1.5). The expansions of µ(z1, z2) in {(z1, z2) ∈ × × Y2(C ) | |z1| > |z2|} and {(z1, z2) ∈ Y2(C ) | |z2| > |z1|} are respectively given by

α β z1 z¯1 lim j(z, f ) z→z2/z1 α β −1 z1 z¯1 lim j(z , f ), z→z1/z2 which deﬁne maps

||z1|>|z2| : GCor2 → U(z1, z2), µ(z1, z2) 7→ µ(z1, z2)||z1|>|z2| and

||z2|>|z1| : GCor2 → U(z2, z1), µ(z1, z2) 7→ µ(z1, z2)||z2|>|z1|. 15

z2 z2 × Since f ( ) = f ( ), the expansions of µ in {(z1, z2) ∈ Y2(C ) | |z2| > |z1 − z2|} is z1 z2+(z1−z2) given by ! ! α β X α β i j −1 z2 z¯2 (z0/z2) (¯z0/z¯2) lim j(1 − z , f ), i j z→−z0/z2 i, j≥0 where z0 = z1 − z2. We denote it by

||z2|>|z0| : GCor2 → U(z2, z0), µ(z1, z2) 7→ µ(z0 + z2, z2)||z2|>|z0|.

The following lemma connects a full vertex algebra (real analytic) and a vertex algebra (holomorphic):

d d d Lemma 1.4 ([Mo3]). Let µ(z1, z2) ∈ GCor2 satisﬁes µ(z1, z2) = 0, (z1 +z2 )µ(z1, z2) = dz¯1 dz1 dz2 d d αµ(z1, z2) and (¯z1 + z¯2 )µ(z1, z2) = βµ(z1, z2) for some α, β ∈ R. Then, µ(z1, z2) ∈ dz¯1 dz¯2 ± ± ± ± R d ± ± C[z , (z1 −z2) , z , z¯ , |z2| ]. Furthermore, if µ(z1, z2) = 0, then µ(z1, z2) ∈ C[z , z , (z1 − 1 2 2 dz¯2 1 2 ± z2) ].

The space of holomorphic generalized two-point correlation functions is denoted by hol GCor2 , that is, hol ± ± ± GCor2 = C[z1 , z2 , (z1 − z2) ].

1.3. Vertex operator algebras, modules and intertwining operators. In this section, we brieﬂy recall the deﬁnitions of a vertex operator algebra, modules and intertwining L ∨ L ∗ ∗ operators. For a Z-graded vector space V = n∈Z Vn, set V = n∈Z Vn , where Vn is the dual vector space of Vn. L A Z-graded vertex algebra is a Z-graded C-vector space V = n∈Z Vn equipped with a linear map X Y(−, z): V → End(V)[[z±]], a 7→ Y(a, z) = a(n)z−n−1 n∈Z and a non-zero element 1 ∈ V0 satisfying the following conditions: V1) For any a, b ∈ F, Y(a, z)b ∈ V((z));

V2) For any a ∈ V, Y(a, z)1 ∈ V[[z, z¯]] and limz→0 Y(a, z)1 = a(−1)1 = a; V3) Y(1, z) = id ∈ EndV; ∨ hol V4) For any a, b, c ∈ V and u ∈ V , there exists µ(z1, z2) ∈ GCor2 such that

u(Y(a, z1)Y(b, z2)c) = µ(z1, z2)||z1|>|z2|,

u(Y(Y(a, z0)b, z2)c) = µ(z0 + z2, z2)||z2|>|z0|,

u(Y(b, z2)Y(a, z1)c) = µ(z1, z2)||z2|>|z1|;

V5) Vn(r)Vm ⊂ Vn+m−r−1 for any n, m, r ∈ Z. The Fourier modes a(n) is called the n-th product. A vertex operator algebra (of CFT type) is a Z-graded vertex algebra with a distinguished vector ω ∈ V such that 16

(1) There exists a scalar c ∈ C such that m3 − m (1.6) [L(m), L(n)] = (m − n)L(m + n) + δ c 12 m+n,0 holds for any n, m ∈ Z, where L(n) = ω(n + 1); d (2)[ L(−1), Y(a, z)] = dz Y(a, z) for any a ∈ V;

(3) L(0)|Vn = n for any n ∈ Z;

(4) Vn = 0 for any n < 0;

(5) Vn is a ﬁnite dimensional vector space for any n ∈ Z;

(6) V0 is spanned by 1. The scalar c ∈ C in (2.1) is called the central charge of the vertex operator algebra. Let V be a vertex operator algebra. A V-module is deﬁned similarly by a linear map Y(−, z): V → EndM[[z±]] (see for example [LL]). Throughout of this paper, we assume that a V-module M satisﬁes the following conditions: M1) The action of L(0) = ω(1) on M is semisimple with real eigenvalues.

M2) Mr = {m ∈ Mi | L(0)m = rm} is a ﬁnite dimensional vector space for any r ∈ R;

M3) There exists N ∈ R such that Mr = 0 for any r ≤ N. L For non-zero V-module M, by (M3), there exists h ∈ R such that M = r≥h Mr and Mh , 0. Such h ∈ R is called a conformal weight of M. We note that if M is a simple L V-module, then M = Mh+k. k∈Z≥0 Let M be a V-module and set ∨ M ∗ M = Mr , r∈R ∗ a restricted dual space, where Mr is the dual vector space of Mr. Denote the canonical pairing M∨ ⊗ M → C by h−i.A V-module structure on M∨ is deﬁned by (1.7) hY(a, z) f, vi = h f, Y exp(L(1)z)(−z−2)L(0)a, z−1 vi for a ∈ V, f ∈ M∨ and v ∈ M [FHL]. It is called a dual module. If there exists a V- module isomorphism φ : M M∨, then the canonical pairing together with φ deﬁnes a non-degenerate bilinear form on M by

(1.8) hφm1, m2i = (m1, m2)M for m1, m2 ∈ M, which satisﬁes the following condition:

−2 L(0) −1 (Y(a, z)m1, m2) = (m1, Y exp(zL(1))(−z ) a, z m2) for a ∈ V and m1, m2 ∈ M (see [FHL, Remark 5.3.3.]). Now, we recall the deﬁnition of an intertwining operator among modules of a vertex operator algebra from [FHL, HL1]. Let M , M , M be V-modules. An intertwining oper- 1 2 3 ator of type M1 is a linear map M2 M3 R X −r−1 I(−, z): M2 → Hom(M3, M1)[[z ]], v 7→ I(v, z) = v(r)z r∈R such that: 17

(1) For any v2 ∈ M2 and v3 ∈ M3, there exists N ∈ R such that v2(r)v3 = 0 for any r ≥ N; d (2)[ L(−1), I(v, z)] = dz I(v, z) for any v ∈ M2; (3) For any v ∈ M2, a ∈ V and n ∈ Z, ! X n [a(n), I(v, z)] = I(a(i)v, z)zn−i i i≥0 ! X n I(a(n)v, z) = (−1)i a(n − i)I(v, z)zi − (−1)nI(v, z)a(i)zn−i . i i≥0 The space of all intertwining operators of type M1 forms a vector space, which is M2 M3 denoted by I M1 . M2 M3 M1 Let I(−, z) ∈ I and let hi ∈ R be the conformal weight of Mi and assume that M2 M3 L M1 Mi = (Mi)h +k for any i = 1, 2, 3. Since [L(0), I (v, z)] = for any r ∈ R and k∈Z≥0 i M2 M3 v ∈ (M2)r, we have:

Lemma 1.5. For any v2 ∈ M2 and v3 ∈ M3,

h1−h2−h3 I(v2, z)v3 ∈ z (M1)((z)). We quote some lemmas from [FHL]:

Lemma 1.6 ([FHL]). Let M be a V-module.

(1) The vertex operator YM(−, z) which defines the module structure on M is an inter- M twining operator of type V,M ; M (2) A vertex operator Y M,V (−, z) defined by ! M Y (b, z)a = exp(L(−1)z)Y (a, −z)b M, V M M for a ∈ V and b ∈ M, is an intertwining operator of type M,V ; ∨ ∨ V (3) If V V and M M , then a vertex operator Y M,M (−, z) defined by ! ! V M (Y (m , z)m , v) = (m , Y (exp(zL(1))(−z−2)L(0)m , z−1)v) M, M 1 2 V 2 M, V 1 M V for a ∈ V and m1, m2 ∈ M, is an intertwining operator of type M,M , where (−, −)V and (−, −)M are defined by (1.8). M1 Lemma 1.7. [FHL, Lemma 5.2.3] Let Y(−, z) ∈ I . For any r ∈ R and m ∈ Mr with M2 M3 L(1)m = 0, exp(L(1)x)Y(m, y) exp(−L(1)x) = (1 − xy)−2rY(m, y/(1 − xy)).

+ P −n−1 − P n Set T (z) = n≥0 Ln−1z and T (z) = n≥0 L−n−2z . The following lemma immedi- ately follows from the deﬁnition of an intertwining operator:

M Lemma 1.8. Let I(−, z) ∈ I 1 and a ∈ (M2)∆a for ∆a ∈ R. Then, for any n ∈ Z and (M2 M3) m ≥ 0, the following equalities hold: 18

n+1 d n P n+1 n−k (1) [L(n), I(a, z)] = (z dz + (n + 1)z ∆a)I(a, w) + k≥1 k+1 I(L(k)a, z)z ; (2)I (L(−1)a, z) = d I(a, z); dz − − 1 d m − 1 d m + (3)I (L( m 2)a, z) = m! dz T (z) I(a, z) + I(a, z) m! dz T (z) ; + P 1 | − (4) [T (z), I(a, w)] = k≥0 (z−w)k+1 |z|>|w|I(L(k 1)a, w); − P 1 | − (5) [I(a, w), T (z)] = k≥0 (z−w)k+1 |w|>|z|I(L(k 1)a, w). 1.4. Virasoro vertex operator algebra. The Virasoro algebra is the Lie algebra Vir = L n∈Z CL(n) ⊕ Cc with commutation relations m3 − m [L(m), L(n)] = (m − n)L(m + n) + δ c, m+n,0 12 [L(m), c] = 0. Given complex numbers h and c, the Verma module V(c, h) is a free module generated by a vector v = |hi satisfying L(0) · |hi = h |hi , c · |hi = c |hi , and L(n) · |hi = 0 for n > 0. The Verma module V(c, h) admits a unique maximal proper submodule J(c, h) with the irreducible quotient L(c, h) = V(c, h)/J(c, h) and L(c, 0) is a simple vertex operator algebra [FZ]. In the case that the central charge c takes values of the form (p − q)2 c = 1 − 6 , p,q pq where p, q is coprime integers such that p, q ≥ 2, L(cp,q, 0) is a regular vertex operator algebra, i.e., it has only ﬁnitely many isomorphism classes of irreducible modules and every module is completely reducible [W, DLM]. The irreducible modules of L(cp,q, 0) is listed by {L(cp,q, hr,s)}r=1,...,p−1,s=1,...,q−1 [IK], where (qr − ps)2 − (q − p)2 h = . r,s 4pq Note that there are symmetric properties

hr,s = hp−r,q−s = hr+p,s+q. By [BPZ], we have:

Lemma 1.9. For any coprime integers p, q ≥ 2, q L(−1)2 − L(−2) |h i ∈ J p 2,1 cp,q,h2,1 p L(−1)2 − L(−2) |h i ∈ J . q 1,2 cp,q,h1,2 Hereafter, we only consider the case that (p, q) is equal to (3, 4), the Ising model. In 1 this case, the central charge is c3,4 = 2 and the conformal weights of irreducible modules 1 1 { 1 1 } are h1,1 = 0 = h2,3 and h2,1 = 2 = h1,3 and h1,2 = 16 = h2,2. Set IS = 0, 2 , 16 . Let h ∈ IS. By the construction, L(0)-grading of L( 1 , h) start from h, that is, L( 1 , h) = L 2 2 1 h | ∈ 1 k≥0 L( 2 , h)h+k. Let h HomC(L( 2 , h), C) be the unique linear map satisfying (1) hh| |hi = 1; 19

h | 1 ≥ (2) h L( 2 , h)h+k = 0 for any k 1. h | 1 ∨ Then, h is a vector in the dual module L( 2 , h) . 1 1 The q-character of an L( 2 , 0)-module L( 2 , h) is a q-series deﬁned by

X 1 k+h− 1 h− 1 χ (q) = dim L( , h) q 48 ∈ q 48 Z[[q]]. h 2 k+h k≥0 Then, we have (see for example [IK]):

Lemma 1.10.

1 X (24k+1)2 (24k+7)2 χ (q) = q 48 − q 48 0 η(q) k∈Z 1 X (24k+2)2 (24k+10)2 χ 1 (q) = q 48 − q 48 2 η(q) k∈Z 1 X (24k+5)2 (24k+11)2 χ 1 (q) = q 48 − q 48 , 16 η(q) k∈Z 1 where η(q) is the Dedekind eta function. Furthermore, the matrix S such that χi(− τ ) = P j S i jχ j(τ) is  √   1 1 2  1  √  S =  1 1 − 2 2  √ √   2 − 2 0 

1 1 where q = exp(2πiτ) and the matrix is ordered as 0, 2 , 16 . 1.5. Virasoro conformal block. In this section, we recall the result of intertwining oper- 1 1 { 1 1 } ators among L( 2 , 0)-modules. The fusion rule of L( 2 , 0)-modules IS = 0, 2 , 16 is a map ? : IS × IS → P(IS), where P(IS) is a power set of IS and the map ? is described in Table 1.5: Table 3. Fusion rule

1 1 ? 0 2 16 1 1 0 0 2 16 1 1 1 2 2 0 16 1 1 1 { 1 } 16 16 16 0, 2

1 h1 L( 2 ,h1) 1 Set I h h = I 1 1 , the space of intertwining operators between L( 2 , 0)-modules. 2 3 L( 2 ,h2)L( 2 ,h3) Then, the following proposition is obtained in [DMZ]:

Proposition 1.11. For h1, h2, h3 ∈ IS,  !  ∈ h1 1, h1 h2 ? h3, dim I =  h2h3 0, otherwise. h1 Furthermore, if h1 ∈ h2 ? h3 and Y(−, z) ∈ I \{0}, then hh1| Y(|h2i , z) |h3i 0. h2h3 , 20 0h1 h1 For h1, h2, h3 ∈ IS with h1 ∈ h2 ? h3, let I (−, z) ∈ I be an intertwining operator h2h3 h2h3 satisfying hh | I0h1 (|h i , z) |h i = zh1−h2−h3 . 1 h2h3 2 3 Such an intertwining operator uniquely exists by Proposition 1.11. The above normalization, 1 0h1 h1−h2−h3 h1−h2−h3 I (|h2i , z) |h3i = z |h1i + higher terms ∈ z L( , h1)[[z]] h2h3 2 seems natural, however, we will later see that this normalization is not a natural choice. The correct normalization is given by setting  √  −1 0h1 1 1 1  2 I (−, z), (h1, h2, h3) is a permutation of ( , , ), (1.9) Ih1 (−, z) =  h2h3 2 16 16 h2h3  I0h1 (−, z), otherwise.  h2h3

For h0, h1, h2, h3 ∈ IS, set A(h0, h1, h2, h3) = {h ∈ IS | h0 ∈ h1 ? h and h ∈ h2 ? h3}. We note that since the fusion rule is commutative and associative, A(h0, h1, h2, h3) , ∅ if and only if h0 ∈ h1 ? h2 ? h3.

For h0, h1, h2, h3 ∈ IS with h ∈ A(h0, h1, h2, h3), let

h 1 ∨ 1 1 1 C : L( , h0) ⊗ L( , h1) ⊗ L( , h2) ⊗ L( , h3) → C{{x}}{{y}} h0,h1,h2,h3 2 2 2 2 be the linear map deﬁned by

(1.10) Ch (a∗, a , a , a ; x, y) = ha∗, Ih0 (a , x)Ih (a , y)a i, h0,h1,h2,h3 0 1 2 3 0 h1,h 1 h2,h3 2 3 ∗ ∈ 1 ∨ ∈ 1 for a0 L( 2 , h0) and ai L( 2 , hi)(i = 1, 2, 3). In Section 3, we will see that the formal × power series is absolutely convergent to a multivalued holomorphic function on Y2(C ). This function is called a Virasoro conformal block in physics.

Remark 1.12. More precisely, this function is a limit of four variables function,

Ch (z , z , z , z ) = hIh0 (a , z )Ih1 (a , z )Ih (a , z )Ih4 (a , z )1i, h0,h1,h2,h3 1 2 3 4 h1,h1 1 1 h2,h 2 2 h3,h4 3 3 h4,0 4 4 in (z , z , z , z ) 7→ (∞, x, y, 0). Since this function can be recovered from C0h (x, y) 1 2 3 4 h0,h1,h2,h3 (see for example [Mo2]), we only consider Ch (x, y). h0,h1,h2,h3

We note that if A(h0, h1, h2, h3) = ∅ then the conformal block does not make sense.

Thus, throughout of this paper we assume that h0, h1, h2, h3 ∈ IS satisfy A(h0, h1, h2, h3) ,

∅ when we consider the conformal block. The set A(h0, h1, h2, h3) is called intermediate states (under the interaction of four states). If we exchange the order of the interaction, we obtain another intermediate state. For 1 1 1 1 { } 1 1 1 1 { 1 } example, A( 2 , 2 , 16 , 16 ) = 0 and A( 2 , 16 , 2 , 16 ) = 16 . However, the number of intermediate states is independent of the order of the interaction,

#A(h0, h1, h2, h3) = #A(h0, h2, h1, h3). ≥ 1 1 1 1 We remark that #A(h0, h1, h2, h3) 2 if and only if (h0, h1, h2, h3) = ( 16 , 16 , 16 , 16 ). In this case, there are two diﬀerent intermediate states, which makes the CFT non-trivial. 21

2. Framed full vertex operator algebra In this section, we briefly recall the definition and results of a full vertex algebra and a full vertex operator algebra from [Mo3]. In Section 2.4, we introduce a notion of an (l, r)-framed full vertex operator algebra, which is a full vertex operator algebra with the ⊕l ⊕r Virasoro symmetry Vir 1 ⊕Vir 1 . In Section 2.6, we define an algebra structure on a lowest 2 2 weight space of a framed full VOA. The axiom of this algebra will study in Section 4.

2.1. Full vertex operator algebras. In this section, we recall the deﬁnition of a full vertex algebra and a full vertex operator algebra. For an R2-graded vector space F = L ∨ L ∗ ∗ F ¯ F F F F ¯ h,h¯∈R2 h,h, set = h,h¯∈R2 h,h¯ , where h,h¯ is the dual vector space of h,h. A full 2 L vertex algebra is an R -graded C-vector space F = h,h¯∈R2 Fh,h¯ equipped with a linear map X Y(−, z): F → End(F)[[z±, z¯±, |z|R]], a 7→ Y(a, z) = a(r, s)z−r−1z¯−s−1 r,s∈R and an element 1 ∈ F0,0 satisfying the following conditions: FV1) For any a, b ∈ F, Y(a, z)b ∈ F((z, z¯, |z|R)); ¯ FV2) Fh,h¯ = 0 unless h − h ∈ Z;

FV3) For any a ∈ F, Y(a, z)1 ∈ F[[z, z¯]] and limz→0 Y(a, z)1 = a(−1, −1)1 = a; FV4) Y(1, z) = id ∈ EndF; ∨ FV5) For any a, b, c ∈ F and u ∈ F , there exists µ(z1, z2) ∈ GCor2 such that

hu Y a z Y b z ci z z | , ( , 1) ( , 2) = µ( 1, 2) |z1|>|z2|, hu Y Y a z b z ci z z z | , ( ( , 0) , 2) = µ( 0 + 2, 2) |z2|>|z0|, hu Y b z Y a z ci z z | , ( , 2) ( , 1) = µ( 1, 2) |z2|>|z1|; 0 ¯ ¯ 0 FV6) Fh,h¯ (r, s)Fh0,h¯0 ⊂ Fh+h0−r−1,h¯+h¯0−s−1 for any r, s, h, h , h, h ∈ R.

Let (F1, Y1, 11) and (F2, Y2, 12) be full vertex algebras. A full vertex algebra homomorphism from F1 to F2 is a grading preserving linear map f : F1 → F2 such that

(1) f (11) = 12

(2) f (Y1(a, z)b) = Y2( f (a), z) f (b) for any a, b ∈ F1. The notions of a subalgebra and an ideal are deﬁned in the usual way. The diﬀerence between vertex algebras and full vertex algebras is summarized in the following table:

vertex algebra full vertex algebra L L vector space V = n∈Z Vn F = h,h¯∈R Fh,h¯ vertex operator EndV[[z±]] EndF[[z, z¯, |z|R]] singularity C((z)) C((z, z¯, |z|R)) ± ± correlation function C[z , (1 − z) ] F0,1,∞ 22

Let (V, Y, 1) be a Z-graded vertex algebra. Then, by a standard result of the theory of a vertex algebra (see for example [FLM]), u(Y(a1, z1)Y(a2, z2)a3) is an expansion of a hol ± ± −1 ∨ rational polynomial in GCor2 = C[z1 , z2 , (z1 −z2) ] ⊂ GCor2 in |z1| > |z2| for any u ∈ V and a1, a2, a3 ∈ V. Thus, we have:

Proposition 2.1. A Z-graded vertex algebra is a full vertex algebra. ¯ Let (F, Y, 1) be a full vertex algebra. Set F¯ = F and F¯h,h¯ = Fh¯,h for h, h ∈ R. Deﬁne R P −s−1 −r−1 Y¯(−, z): F¯ → End(F¯)[[z, z¯, |z| ]] by Y¯(a, z) = r,s∈R a(r, s)z z¯ . Then, we have: Proposition 2.2. (F¯, Y¯, 1) is a full vertex algebra.

We call it a conjugate full vertex algebra of (F, Y, 1). ¯ 2 Let F be a full vertex algebra. The set {(h, h) ∈ R | Fh,h¯ , 0} is called a spectrum. The spectrum of F is said to be bounded below if there exists N ∈ R such that Fh,h¯ = 0 for any ¯ P h ≤ N or h ≤ N and discrete if for any H ∈ R, h+h¯≤H dim Fh,h¯ is ﬁnite. Then, we have:

Proposition 2.3. Let (F1, Y1, 11) and (F2, Y2, 12) be full vertex algebras. If the spectrum of F1 is discrete and F2 is bounded below, then (F1 ⊗ F2, Y1 ⊗ Y2, 11 ⊗ 12) is a full vertex algebra.

Let F be a full vertex algebra and D and D¯ denote the endomorphism of F deﬁned by Da = a(−2, −1)1 and Da¯ = a(−1, −2) for a ∈ F, i.e., Y(a, z)1 = a + Daz + Da¯ z¯ + ....

P r−s r s Deﬁne Y(a, −z) by Y(a, −z) = r,s∈R(−1) a(r, s)z z¯ , where we used a(r, s) = 0 for r − s < Z, which follows from (FV2) and (FV6). Then, we have: Proposition 2.4. For a ∈ F, the following properties hold: (1)Y (Da, z) = d/dzY(a, z) and Y(Da¯ , z) = d/dzY¯ (a, z); (2)D 1 = D¯ 1 = 0; (3) [D, D¯ ] = 0; (4)Y (a, z)b = exp(zD + z¯D¯ )Y(b, −z)a; (5)Y (Da¯ , z) = [D¯ , Y(a, z)] and Y(Da, z) = [D, Y(a, z)].

We introduced a notion of a full vertex operator algebra similarly to the deﬁnition of a vertex operator algebra. An energy-momentum tensor of a full vertex algebra is a pair of vectors ω ∈ F2,0 andω ¯ ∈ F0,2 such that (1) D¯ ω = 0 and Dω¯ = 0; (2) There exist scalars c, c¯ ∈ C such that m3 − m (2.1) [L(m), L(n)] = (m − n)L(m + n) + δ c 12 m+n,0 m3 − m [L(m), L(n)] = (m − n)L(m + n) + δ c¯ 12 m+n,0 holds for any n, m ∈ Z, where L(n) = ω(n + 1, −1) and L(n) = ω¯ (−1, n + 1); 23

(3) L(−1) = D and L(−1) = D¯ ; | | ¯ ¯ ∈ (4) L(0) Fh,h¯ = h and L(0) Fh,h¯ = h for any h, h R; (5) The spectrum of F is discrete and bounded below;

(6) F0,0 = C1. A full vertex operator algebra is a pair of a full vertex algebra and its energy momentum tensor.

Let (F1, Y1, 11, ω1, ω¯ 1) and (F2, Y2, 12, ω2, ω¯ 2) be full vertex operator algebras. A conformal embedding is a full vertex algebra homomorphism i : F1 → F2 such that:

(1) i(ω1) = ω2 and i(ω ¯ 1) = ω¯ 2; (2) i is injective.

2.2. Full vertex algebra and vertex algebra. In this section, we brieﬂy recall the rela- tion between a vertex algebra and a full vertex algebra. Let F be a full vertex algebra. A vector a ∈ F is said to be a holomorphic vector (resp. an anti-holomorphic vector) if Da¯ = 0 (resp. Da = 0). Let a ∈ ker D¯ . Then, since 0 = Y(Da¯ , z) = d/dzY¯ (a, z), we have P −n−1 a(r, s) = 0 unless s = −1. Hence, Y(a, z) = n∈Z a(n, −1)z . Then, by Lemma 1.4, we have (see [Mo3] for more detail):

Lemma 2.5. Let a, b ∈ F. If Da¯ = 0, then X Y(a, z) = a(n, −1)z−n−1 n∈Z and for any n ∈ Z, ! X n [a(n, −1), Y(b, z)] = Y(a(i, −1)b, z)zn−i, i i≥0 ! ! X n X n Y(a(n, −1)b, z) = (−1)ia(n − i, −1)ziY(b, z) − Y(b, z) (−1)i+na(i, −1)zn−i. i i i≥0 i≥0 By Proposition 2.4, DY¯ (a, z)b = Y(Da¯ , z)b + Y(a, z)Db¯ = 0. Thus, the restriction of Y on ker D¯ deﬁne a linear map Y(−, z) : ker D¯ → End ker D¯ [[z±]]. By the above Lemma and Lemma 1.4, we have:

Proposition 2.6. ker D¯ is a vertex algebra and F is a ker D-module.¯

Lemma 2.7. For a holomorphic vector a ∈ F and an anti-holomorphic vector b ∈ F, [Y(a, z), Y(b, w¯ )] = 0, that is, [a(n, −1), b(−1, m)] = 0 and a(k, −1)b = 0 for any n, m ∈ Z and k ∈ Z≥0.

By Proposition 2.6, ker D and ker D¯ is a subalgebra of F and ker D¯ ⊗ ker D is a full vertex algebra. Deﬁne a linear map t : ker D¯ ⊗ ker D → F by t(a ⊗ b) = a(−1, −1)b for a ∈ ker D¯ and b ∈ ker D. Then, we have:

Proposition 2.8. The map t : ker D¯ ⊗ ker D → F is a full vertex algebra homomorphism. 24

We note that if ker D¯ and ker D are simple vertex algebras, then the map t is injective. R Let Yc(−, z): F → EndF[[z ]] be the vertex operator deﬁned by

X −r−s−2 (2.2) Yc(a, z) = lim Y(a, z) = a(r, s)z z¯→z r,s∈Z for a ∈ F. The identification z = z¯ is well-defined since it gives a well-defined linear map z = z¯ : C((z, z¯, |z|)) → C[[zR]].

Let VF be the image of the map t in Proposition 2.8, VF = t(ker D¯ ⊗ ker D) ⊂ F. For any a ∈ ker D¯ and b ∈ ker D, by Lemma 2.7 and Lemma 2.5,

X −n−m−2 Yc(a(−1, −1)b, z) = a(n, −1)b(−1, m)z = Yc(a, z)Yc(b, z). n,m∈Z ± In particular, Yc(a(−1, −1)b, z) ∈ EndF[[z ]]. If F is a full VOA, then by Lemma 2.5

(VF, Yc) is a vertex operator algebra and F is a VF-module. Furthermore, by Lemma 2.5 and the deﬁnition of an intertwining operator, we have:

Lemma 2.9. Let F be a full vertex operator algebra. For the vertex operator Yc(−, z) deﬁned by (2.2),VF is a vertex operator algebra, F is a VF-module and Yc(−, z) is an F intertwining operator of type FF . Remark 2.10. This lemma is also found in [HK].

2.3. Locality axiom. In the definition of a full vertex algebra, the terms hu Y a z Y b z ci z z | , ( , 1) ( , 2) = µ( 1, 2) |z1|>|z2|, hu Y b z Y a z ci z z | (2.3) , ( , 2) ( , 1) = µ( 1, 2) |z2|>|z1| are symmetric while hu Y Y a z b z ci z z z | , ( ( , 0) , 2) = µ( 0 + 2, 2) |z2|>|z0| is not. So it is convenient to give a definition of a full vertex algebra only by using (2.3). Such a result is obtained in [Mo3] similarly to the case of vertex algebras. We briefly recall this result in this section. 2 L A full prevertex algebra is an R -graded C-vector space F = h,h¯∈R2 Fh,h¯ equipped with a linear map X Y(−, z): F → EndF[[z±, z¯±, |z|R]], a 7→ Y(a, z) = a(r, s)z−r−1z¯−s−1 r,s∈R and a non-zero element 1 ∈ F0,0 such that PV1) For any a, b ∈ F, Y(a, z)b ∈ F((z, z¯, |z|R)); ¯ PV2) Fh,h¯ = 0 unless h − h ∈ Z;

PV3) For any a ∈ F, Y(a, z)1 ∈ F[[z, z¯]] and limz→0 Y(a, z)1 = a(−1, −1)1 = a; PV4) Y(1, z) = id ∈ EndF; ¯ 0 ¯ 0 PV5) Fh,h¯ (r, s)Fh0,h¯0 ⊂ Fh+h0−r−1,h¯+h¯0−s−1 for any r, s, h, h, h , h ∈ R. A full prevertex algebra (F, Y, 1) is said to be translation covariant if there exist linear maps D, D¯ ∈ End F such that 25

T1) D1 = D¯ 1 = 0; ∈ d ¯ d T2) For any a F,[D, Y(a, z)] = dz Y(a, z) and [D, Y(a, z)] = dz¯ Y(a, z); Then, we have (see [Mo3, Proposition 3.8]):

Proposition 2.11. Assume that a translation covariant full prevertex algebra (F, Y, 1, D, D¯ ) satisﬁes the following conditions: L1) The spectrum of F is bounded below; ∨ L2) For any a1, a2, a3 ∈ F and u ∈ F , there exists µ(z1, z2) ∈ GCor2 such that

u Y a z Y a z a z z | ( ( 2, 2) ( 1, 1) 3) = µ( 1, 2) |z1|>|z2| u Y a z Y a z a z z | ( ( 1, 1) ( 2, 2) 3) = µ( 1, 2) |z2|>|z1|

Then, (F, Y, 1) is a full vertex algebra.

2.4. Deﬁnition of framed full vertex operator algebra. Let l, r ∈ Z≥0. By Lemma 2.3, r 1 l ⊗ 1 the tensor product of the Virasoro vertex operator algebras L( 2 , 0) L( 2 , 0) is a full r 1 vertex operator algebra, where L( 2 , 0) is the conjugate full vertex algebra in Proposition 2.2. We denote it by Ll,r(0). An (l, r)-framed full vertex operator algebra is a full vertex operator algebra F together with a conformal embedding i : Ll,r(0) → F.

Let S l, S r be the symmetric group of degree l, r. Then, by the permutation, S l × S r is naturally a subgroup of the full vertex operator algebra automorphism group Aut Ll,r(0).

A morphism from an (l, r)-framed full VOA i1 : Ll,r(0) → F1 to an (l, r)-framed full VOA i2 : Ll,r(0) → F2 is a pair of a full VOA homomorphism φ : F1 → F2 and (g, g¯) ∈

S l × S r such that i2 ◦ (g, g¯) = φ ◦ i1 : Ll,r(0) → F2. We denote the category of framed full vertex operator algebras by Framed full VOAs. By deﬁnition, a framed full vertex ⊕l ⊕r operator algebra is a module of copies of Virasoro algebras Vir 1 ⊕ Vir 1 . We denote the 2 2 ⊕l action of i-th component of the left Virasoro algebras Vir 1 by {Li(n)}n∈Z and the action 2 ⊕r of j-th component of the right Virasoro algebras Vir 1 by {L j(n)}n∈Z for l ≥ i ≥ 1 and 2 r ≥ j ≥ 1.

2.5. Multi-index Virasoro conformal block. In this section, we prepare some notations (l,r) l r (l,r) for Ll,r(0)-modules. Set IS = IS ×IS . For λ = (h1, h2,..., hl, h¯ 1,..., h¯ r) ∈ IS , deﬁne 1 ⊗ · · · ⊗ 1 ⊗ 1 ¯ ⊗ · · · ⊗ 1 ¯ an Lr,s(0)-module by Ll,r(λ) = L( 2 , h1) L( 2 , hl) L( 2 , h1) L( 2 , hr) and set |λi = |h1i ⊗ · · · |hli ⊗ |h¯ 1i ⊗ · ⊗ |h¯ ri ∈ Ll,r(λ), which is a lowest weight vector. By the similar argument in Lemma 2.9, it is easy to show that this gives a bijection between irreducible (l,r) Ll,r(0)-modules and IS . By extending the fusion rule of IS, deﬁne a fusion rule ? : IS(l,r) × IS(l,r) → P(IS(l,r)) by

1 2 1 2 1 2 ¯ 1 ¯ 2 ¯ 1 ¯ 2 λ ? λ = (h1 ? h1) × · · · × (hl ? hl ) × (h1 ? h1) × · · · × (hr ? hr )

i i i ¯ i ¯ i (l,r) for λ = (h1,..., hl, h1 ..., hr) ∈ IS and i = 1, 2. 26 1 2 ∈ (l,r) ∈ 1 2 λ − 1 → 2 R R For λ , λ IS and λ λ ?λ , let Iλ1,λ2 ( , z): Ll,r(λ ) Hom Ll,r(λ ), Ll,r(λ) [[z , z¯ ]] be a linear map deﬁned by the tensor product of intertwining operators: Ol Or λ − hi − ⊗ ¯hi − Iλ1,λ2 ( , z) = I 1 2 ( , z) I 1 2 ( , z¯). hi ,hi hi ,hi i=1 j=1 (l,r) Pl Pr ¯ Let l, r, s : IS → Q be maps deﬁned by l(λ) = i=1 hi and r(λ) = j=1 h j for λ = (l,r) (h1, h2,..., hl, h¯ 1,..., h¯ r) ∈ IS and s = l − r.

Remark 2.12. The map l (resp. r) measures the conformal weights of holomorphic part (resp. anti-holomorphic part) and s corresponds to the spin of the state, which must be an integer for any bosonic state.

By Lemma 1.5, we have:

1 2 (l,r) 1 2 1 2 Lemma 2.13. Let λ , λ ∈ IS , λ ∈ λ ? λ and a ∈ Ll,r(λ ), b ∈ Ll,r(λ ). Then, λ ∈ l(λ)−l(λ1)−l(λ2) r(λ)−r(λ1)−r(λ2) Iλ1,λ2 (a, z)b z z¯ Ll,r(λ)((z, z¯)). 1 2 ∈ Z λ ∈ | |R In particular, if s(λ), s(λ ), s(λ ) , then Iλ1,λ2 (a, z)b Ll,r(λ)((z, z¯, z )). For λ0, λ1, λ2, λ3 ∈ IS(l,r), let A(λ0, λ1, λ2, λ3) be the set of all λ ∈ IS(l,r) such that λ ∈ λ2 ? λ3 and λ0 ∈ λ1 ? λ. We note that since the fusion rule is commutative and associative again, A(λ0, λ1, λ2, λ3) , ∅ if and only if λ0 ∈ λ1 ? λ2 ? λ3. For λ0, λ1, λ2, λ3 ∈ IS(l,r) with λ ∈ A(λ0, λ1, λ2, λ3), let λ 0 ∨ ⊗ 1 ⊗ 2 ⊗ 3 → {{ }}{{ }} Cλ0,λ1,λ2,λ3 : Ll,r(λ ) Ll,r(λ ) Ll,r(λ ) Ll,r(λ ) C x y be the linear map deﬁned by λ ∗ h ∗ λ0 h i (2.4) Cλ0,λ1,λ2,λ3 (a0, a1, a2, a3; x, y) = a0, Iλ1,λ(a1, x)Iλ2,λ3 (a2, y)a3 , ∗ 0 ∨ i for a0 ∈ Ll,r(λ ) and ai ∈ Ll,r(λ )(i = 1, 2, 3). In this paper we call this function a multi- index Virasoro conformal block. It is clear from the deﬁnition that a multi-index Virasoro conformal block is decomposed into the product of Virasoro conformal blocks. Properties of multi-index Virasor conformal blocks will be studied in Section 3. 2.6. Lowest weight space and conformal block. Let F be a framed full vertex oper- 1 l+r ator algebra. Then, F is a module of the vertex operator algebra L( 2 , 0) , where the − 1 l+r module structure is given by Yc( , z) (see Lemma 2.9). Since any module of L( 2 , 0) is completely reducible [W, DLM], F is decomposed into a direct sum of irreducible

Ll,r(0)-modules, M F = Ll,r(λ) ⊗ (S F)λ, λ∈IS(l,r) (l,r) where (S F)λ, for λ = (h1, h2,..., hl, h¯ 1,..., h¯ r) ∈ IS , consists of all vectors v ∈ F such that: for any l ≥ i ≥ 1, r ≥ j ≥ 1

LW1) Li(0)v = hiv, L j(0)v = h¯ jv;

LW2) Li(n)v = L j(n)v = 0 for any n ≥ 1. 27

By the deﬁnition of a full VOA, (S F)λ is a ﬁnite dimensional vector space. Set S F = L ⊂ ⊕l ⊕ ⊕r λ∈IS(l,r) (S F)λ F, which is the lowest weight space of Vir 1 Vir 1 . 2 2 − → R 1 l+r By Lemma 2.9 again, the vertex operator Yc( , z): F EndF[[z ]] is an L( 2 , 0) - F module intertwining operator of type FF . We note that ! 1 ! F M L (λ ) ⊗ S 1 I I l,r λ (2.5) = 2 3 FF Ll,r(λ ) ⊗ S λ2 , Ll,r(λ ) ⊗ S λ3 λ1,λ2,λ3∈IS(l,r) 1 ! M Ll,r(λ ) S 2 , S 3 , S 1 ⊗ I = HomC( λ HomC( λ λ )) 2 3 Ll,r(λ ), Ll,r(λ ) λ1,λ2,λ3∈IS(l,r) l 1 ! r ¯ 1 ! M O hi O h j = HomC(S λ2 ⊗ S λ3 , S λ1 ) ⊗ I ⊗ I , h2, h3 h¯ 2, h¯ 3 λ1,λ2,λ3∈IS(l,r) i=1 i i j=1 j j i i i ¯ i ¯ i where λ = (h1,..., hl, h1,..., hr) for i = 1, 2, 3 and the last equality follows from [DMZ]. By (2.5), the vertex operator Y(−, z) of the framed full vertex operator algebra F is decomposed into the tensor product of holomorphic and anti-holomorphic Virasoro in- h1 h¯1 tertwining operators Ih ,h (−, z), I¯ ¯ (−, z¯). Furthermore, it is uniquely determined by the L 2 3 h2,h3 2 ⊗ 3 1 linear map λ1,λ2,λ3∈IS(l,r) Hom(S λ S λ , S λ ), which gives an algebra structure on S F. We denote this product S F × S F → S F by ·. (l,r) For λ ∈ IS , let πλ : S F → S F be the projection of the graded vector space S F onto

(S F)λ and ·λ : S ⊗ S → S λ given by the composition of the product · : S ⊗ S → S and the

projection πλ. Then, the product · is a unique linear map S F ⊗ S F → S F such that for any 1 2 (l,r) λ , λ ∈ IS and a1 ∈ (S F)λ1 , a2 ∈ (S F)λ2 , X M λ | 1i | 2i ⊗ · ∈ ⊗ Y(a1, z)a2 = Iλ1,λ2 ( λ , z) λ (a1 λ a2) F = Ll,r(λ) (S F)λ. λ∈λ1?λ2 λ∈IS(l,r) The following lemma easily follows from (2.5) and the deﬁnition of a full vertex operator algebra:

Lemma 2.14. For any framed full VOA F, S F satisﬁes the following conditions: (l,r) FA1) For any λ ∈ IS , (S F)λ = 0 unless s(λ) ∈ Z.

FA2)( S F)0 = C1 and for any a ∈ S , a · 1 = 1 · a = a; 1 2 (l,r) L FA3) For any λ , λ ∈ IS , a1 ∈ S λ1 and a2 ∈ S λ2 , a1 · a2 ∈ λ∈λ1?λ2 S λ.

As seen above, we can obtain an algebra structure on S F from the vertex operator Y(−, z). In the next section, we consider this construction in the reverse direction.

2.7. Induced vertex operator. Let S be a ﬁnite-dimensional IS(l,r)-graded vector space, M S = S λ λ∈IS(l,r)

with a linear map · : S ⊗ S → S and a distinguished non-zero element 1 ∈ S 0 satisﬁes (l,r) the conditions (FA1),(FA2) and (FA3) in Lemma 2.14. For λ ∈ IS , let πλ : S → S

be the projection of the graded vector space S onto S λ and denote the composition of the

product and the projection by ·λ : S ⊗ S → S λ. 28 L ⊗ → Set FS = λ∈IS(l,r) Lr,s(λ) S λ, an Ll,r(0)-module. Let i : S FS be the linear map (l,r) deﬁned by i(a) = |λi ⊗ a ∈ Lr,s(λ) ⊗ S λ for any h ∈ IS and a ∈ S λ.

Deﬁne Y(−, z, z¯): FS → EndFS by X ⊗ ⊗ λ ⊗ · Y(u1 a1, z, z¯)u2 a2 = Iλ1λ2 (u1, z)u2 (a1 λ a2). λ∈λ1?λ2 1 2 (l,r) 1 2 1 2 for λ , λ ∈ IS , u1 ∈ Ll,r(λ ), u2 ∈ Ll,r(λ ) and a1 ∈ Ll,r(λ ) and a2 ∈ Ll,r(λ ).

Let D, D¯ ∈ End FS deﬁned by Xl Xl D = Li(−1), D¯ = L j(−1), i=1 j=1 ⊕l ⊕r where Li(−1), L j(−1) are given by the action of Vir ⊕ Vir . Then, we have:

Proposition 2.15. (FS , Y(−, z), 1, D, D¯ ) is a translation covariant full prevertex algebra and its spectrum is bounded below. Proof. (PV1), (PV5), (T1) and (T2) follow from the deﬁnition of the intertwining operators. (PV2) follows from (FA1). (PV3) and (PV4) follows from (FA2) and Lemma 1.6. The spectrum is clearly bounded below. Remark 2.16. In general, there is a linear isomorphism between the following spaces: (1) The space of intertwining operators I FS ; FS FS L (2) The space of linear maps · ∈ HomC(S ⊗ S, S ) such that a · b ∈ λ∈λ1?λ2 S λ for 1 2 (l,r) any λ , λ ∈ IS , a ∈ S λ1 and b ∈ S λ2 .

One of the main result of this paper is to obtain the precise condition when FS becomes a framed full vertex operator algebra. By Proposition 2.11, FS is a full vertex algebra if 0 1 2 3 (l,r) and only if FS satisﬁes the locality axiom (L2). Let λ , λ , λ , λ ∈ IS and ui ⊗ ai ∈ i ∗ ∗ 0 ∨ ∗ Ll,r(λ ) ⊗ S λi and u0 ⊗ a0 ∈ Ll,r(λ ) ⊗ (S λ0 ) . Then, ∗ ∗ hu0 ⊗ a0, Y(u1 ⊗ a1, z1, z¯1)Y(u2 ⊗ a2, z2, z¯2)u3 ⊗ a3i X 0 hu∗ Iλ u z Iλ u z u iha∗ a · a · a i = 0, λ1λ( 1, 1) λ2λ3 ( 2, 2) 3 0, 1 λ0 ( 2 λ 3) λ∈λ2?λ3 X λ h ∨ · · i (2.6) = Cλ0,λ1,λ2,λ3 (u0, u1, u2, u3; z1, z2) a0 , a1 λ0 (a2 λ a3) . λ∈λ2?λ3

Thus, to clarify when FS is a full vertex algebra, it is important to study multi-index Virasoro conformal blocks. The next section is devoted to studying them.

3. Conformal block of Virasoro algebras In this section, we consider the Virasoro conformal blocks. We ﬁrst study some special- ization of single Virasoro conformal blocks. For (h0, h1, h2, h3) ∈ IS and h ∈ A(h0, h1, h2, h3), set Ch (x, y) = Ch (hh | , |h i , |h i , |h i ; x, y) h0,h1,h2,h3 h0,h1,h2,h3 0 1 2 3 = hh | Ih0 (|h i , x)Ih (|h i , y) |h i . 0 h1,h 1 h2,h3 2 3 29

We will later see that all informations of the Virasoro conformal blocks are obtained from these specialized ones. It is useful to introduce non-normalized conformal blocks,

C0h (x, y) = hh | I0h0 (|h i , x)I0h (|h i , y) |h i . h0,h1,h2,h3 0 h1,h 1 h2,h3 2 3 In section 3.1, we study the specialized Virasoro conformal blocks in the case that one of hi is equal to 0. General cases are studied in Section 3.2. Analytic continuations of the Virasoro conformal blocks are studied in Section 3.3 and their multi-index generalizations are given in Section 3.4, which is the most important part of this section.

3.1. Conformal block involving vacuum state. In this section, we study C0h (x, y) h0,h1,h2,h3 in the case that one of hi is equal to 0. In this case, the conformal blocks are very easy to calculate. We note that results obtained in this section is always true for any vertex operator algebra.

First, we assume that h1 = 0. Then, A(h0, 0, h2, h3) is not empty if and only if h0 ∈ h2 ? h3. Note that in this case A(h0, 0, h2, h3) = {h0}. For h ∈ h ?h , by Lemma 1.6 and the deﬁnition of the intertwining operator I0h0 (−, z), 0 2 3 h2,h3 C0h0 (x, y) = hh | I0h0 (|0i , x)I0h0 (|h i , y) |h i h0,0,h2,h3 0 0,h0 h2,h3 2 3 = hh | Y(1, x)I0h0 (|h i , y) |h i 0 h2,h3 2 3 = hh | id I0h0 (|h i , y) |h i 0 h0 h2,h3 2 3 = yh0−h2−h3 .

Similarly, in the case of h2 = 0, for any h0 ∈ h1 ? h3,

C0h3 (x, y) = xh0−h1−h3 . h0,h1,0,h3

Second, we assume that h3 = 0. Then, A(h0, h1, h2, 0) is not empty if and only if h0 ∈ h1 ? h2, and in this case A(h0, h1, h2, 0) = {h2}.

By Lemma 1.6 and hh0| (L(−1)−) = 0,

C0h2 (x, y) = hh | I0h0 (|h i , x)I0h2 (|h i , y) |0i h0,h1,h2,0 0 h1,h2 1 h2,0 2 = hh | I0h0 (|h i , x) exp(L(−1)y)Y(|0i , −y) |h i 0 h1,h2 1 2 = hh | I0h0 (|h i , x) exp(L(−1)y) |h i 0 h1,h2 1 2 = hh | I0h0 (|h i , x − y) |h i 0 h1,h2 1 2 ∞ ! X h0 − h1 − h2 = xh0−h1−h2−kyk = (x − y)h0−h1−h2 | . k |x|>|y| k=0

Finally, we assume that h0 = 0. Then, A(0, h1, h2, h3) is not empty if and only if h1 ∈ h2 ? h3, and in this case A(0, h1, h2, h3) = {h1}. By Lemma 1.6, it is easy to show the following lemma:

∈ ∈ 1 h | 0 | i i h | −1 i · −2h Lemma 3.1. For any h IS and v L( 2 , h), 0 Ih,h( h , z)v = h exp(L(1)z )v z . 30

h2−h1−h3 h1−h2−h3 h3−h1−h2 = x y (x − y) ||x|>|y| Thus, we have:

Proposition 3.2. Let h0, h1, h2, h3 ∈ IS with h ∈ A(h0, h1, h2, h3). In the case that one of hi is equal to 0, the conformal blocks are given in the following table:

Table 4. Conformal blocks with vacuum sector

C0h (x, y) h0,h1,h2,h3 h2−h1−h3 h1−h2−h3 h3−h1−h2 h0 = 0 x y (x − y) h0−h2−h3 h1 = 0 y h0−h1−h3 h2 = 0 x h0−h1−h2 h3 = 0 (x − y)

3.2. Conformal block – general cases. In this section, we study conformal blocks in general cases, that is, none of {hi}i=0,1,2,3 is not 0. We explain the method to calculate the conformal blocks introduced in [BPZ]. Since h hh | Ih0 (|h i , x)Ih (|h i , y) |h i 0 0 h1,h 1 h2,h3 2 3 = hh | L(0)Ih0 (|h i , x)Ih (|h i , y) |h i 0 h1,h 1 h2,h3 2 3 = hh | [L(0), Ih0 (|h i , x)]Ih (|h i , y) |h i + hh | Ih0 (|h i , x)[L(0), Ih (|h i , y)] |h i 0 h1,h 1 h2,h3 2 3 0 h1,h 1 h2,h3 2 3 + hh | Ih0 (|h i , x)Ih (|h i , y)L(0) |h i 0 h1,h 1 h2,h3 2 3 = (x∂ + y∂ + h + h + h )Ch0,h (x, y), x y 1 2 3 (h1,h2,h3) we have (x∂ + y∂ + h + h + h − h )Ch (x, y) = 0. x y 1 2 3 0 h0,h1,h2,h3 Thus, by Lemma 1.3, y x−h0+h1+h2+h3Ch (x, y) ∈ C[[( )R]]. h0,h1,h2,h3 x Thus, the evaluation of x = 1 is well-deﬁned. Set

0h 0h 0h0 0h (3.1) C h ,h ,h ,h (z) = lim C h ,h ,h ,h (x, z) = hh0| I (|h1i , 1)I h ,h (|h2i , z) |h3i , 0 1 2 3 x→1 0 1 2 3 h1,h 2 3 which is a formal power series in zh−h2−h3 C((z)). Then, we have: 31

Lemma 3.3. For any h0, h1, h2, h3 ∈ IS with h ∈ A(h0, h1, h2, h3), the following conditions hold: y (1)C 0h (x, y) = xh0−h1−h2−h3C0h ( ); h0,h1,h2,h3 h0,h1,h2,h3 x (2)C 0h (z) = 1 · zh−h2−h3 + O(zh−h2−h3+1) ∈ zh−h2−h3 C[[z]]. h0,h1,h2,h3 By the above lemma, in order to determine Ch (x, y) it suffices to calculate C0h (z), h0,h1,h2,h3 h0,h1,h2,h3 which is also called a conformal block. It is convenient to set 3 1 h = a − , a 4 2 1 1 for a ∈ R. Then, h 4 = and h 3 = . The following lemma is shown by [BPZ]: 3 2 4 16 ∈ ∈ { 4 3 } ∈ Lemma 3.4. For h0, h1, h2 IS and a 3 , 4 and h A(h0, h1, h2, ha), the conformal block C0h (z) satisfies the following differential equation: h0,h1,h2,ha

2 1 2 0h (1 − z)z∂ + ((2h˜ − 2 + a)z + a)∂z + ((h˜(a + h˜ − 1) − ah )z − ah ) C (z), z (1 − z)z 1 2 h0,h1,h2,ha where h˜ = h0 − h1 − h2 − ha.

Proof. By Lemma 1.8 and Lemma 1.9,

0 = hh | I0h0 (|h i , x)I0h (|h i , y)L(−1)2 − aL(−2) |h i 0 h1,h 1 h2,ha 2 a

2 −1 −1 h1 h2 0h = (∂x + ∂y) + a(x ∂x + y ∂y − − )C (x, y). x2 y2 h0,h1,h2,ha By Lemma 3.3,

2 −1 −1 h1 h2 h˜ 0h z 0 = (∂y + ∂z) + a(y ∂y + z ∂z − − ) y C ( ). y2 z2 h0,h1,h2,h3 y Thus, the assertion holds.

∈ 3 − 1 ∈ { 4 3 } Let h0, h1, h2, h3 IS with h3 = ha = 4 a 2 for a 3 , 4 and set 1 D = (1 − z)z∂2 + ((2h˜ − 2 + a)z + a)∂ + ((h˜(a + h˜ − 1) − ah )z2 − ah ), h0,h1,h2,h3 z z (1 − z)z 1 2 the differential operator in Lemma 3.4. The explicit descriptions of the differential equations are given in Table 3.2: Those differential equations have possible singular points at {0, 1, ∞}, but all singular points are regular. Thus, their formal solutions C0h (z) are absolutely convergent in h0,h1,h2,h3 {z ∈ C | |z| < 1} and have analytic continuation to multivalued holomorphic functions on CP1 \{0, 1, ∞}. Since zr is not single-valued around z = 0, we need to consider a branch cut.

Set R− = {r ∈ R | r ≤ 0} and deﬁne the holomorphic function Log : C \ R− → C by Log(z) = log|z| + Arg(z) with Arg(z) ∈ (−π, π), called the principle value. The following lemma is clear: 32 Table 5. Diﬀerential equations

(h0, h1, h2, h3) Dh0,h1,h2,h3 1 1 1 1 − 2 − − − ( 16 , 16 , 16 , 16 ) z(1 z)∂z + (3/4 3/2z)∂z 3/(64z(1 z)) 1 1 1 1 − 2 − − − ( 2 , 2 , 2 , 2 ) z(1 z)∂z + (4/3 8/3z)∂z 2/(3z(1 z)) 1 1 1 1 − 2 − − 2 − ( 2 , 2 , 16 , 16 ) z(1 z)∂z + (3/4 3/2z)∂z (3 + 21z )/(64z(1 z)) 1 1 1 1 − 2 − − − ( 2 , 16 , 2 , 16 ) z(1 z)∂z + (3/4 3/2x)∂z 3/(8z(1 z)) 1 1 1 1 − 2 − − − 2 − ( 16 , 2 , 2 , 16 ) z(1 z)∂z + (3/4 13/4z)∂z (3 7z )/(8z(1 z)) 1 1 1 1 − 2 − − 2 − ( 2 , 16 , 16 , 2 ) z(1 z)∂z + (4/3 11/12z)∂z (16 + 21z )/(192z(1 z)) 1 1 1 1 − 2 − − − ( 16 , 2 , 16 , 2 ) z(1 z)∂z + (4/3 8/3z)∂z 1/(12z(1 z)) 1 1 1 1 − 2 − − − 2 − ( 16 , 16 , 2 , 2 ) z(1 z)∂z + (4/3 8/3z)∂z (8 7z )/(12z(1 z))

Lemma 3.5. Let µ, ν ∈ C \ R− such that µν < R−. Then,  2πi Arg(α) + Arg(β) > π  Log(µν) = Log(µ) + Log(ν) +  − 0 π < Arg(α) + Arg(β) < π  −2πi Arg(α) + Arg(β) < −π.

r r Then, for µ ∈ C \ R− and r ∈ R, µ is deﬁned by µ = exp(rLog(µ)). By Lemma 3.4, the formal power series Ch (z) is absolutely convergent to a single-valued holomorphic h0,h1,h2,h3 function in this branch V+ = C \ R− ∩ {|z| < 1}. Let denote this holomorphic function by Ch (z)| Since D is a second order diﬀerential equation with regular h0,h1,h2,h3 V+ h0,h1,h2,h3 singularities, by the standard method, we can determine the conformal blocks explicitly.

Proposition 3.6. The conformal blocks are given in Table 6:

Table 6. Conformal blocks

0h (h0, h1, h2, h3) h C (z)|V h0q,h1,h2,h3 + q 1 1 1 1 1 − 1 − 1 √ √ ( , , , ) 0 z 8 (1 − z) 8 1 − z + 1 + z 16 16 16 16 2 q q 1 − 1 − 1 √ √ 8 − 8 − − 2 z (1 z) 1 + z 1 z 1 1 1 1 −1 − −1 − 2 ( 2 , 2 , 2 , 2 ) 0 z (1 z) (1 z + z ) 1 1 1 1 − 1 − 1 z 8 − 2 − ( 2 , 2 , 16 , 16 ) 0 z (1 z) (1 2 ) 1 1 1 1 1 − 1 − 1 2 − 2 − ( 2 , 16 , 2 , 16 ) 16 z (1 z) (1 2z) 1 1 1 1 1 − 1 −1 2 − ( 16 , 2 , 2 , 16 ) 16 z (1 z) (1 + z) 1 1 1 1 1 − 1 − 1 2 − 8 ( 2 , 16 , 16 , 2 ) 16 z (1 z) (1 + z) 1 1 1 1 1 − 1 − 1 2 − 2 − ( 16 , 2 , 16 , 2 ) 16 z (1 z) (1 2z) 1 1 1 1 −1 − 1 z − 2 − ( 16 , 16 , 2 , 2 ) 0 z (1 z) (1 2 )

Here, we need to fix the branches of functions in the table. Since the ranges {1 − µ | µ ∈ q √ r √ V } and {1 ± µ | µ ∈ V } are contained in C \ R−, (1 − z) (resp. 1 ± z) on V is + + √ + − 1 ± defined by exp(rLog(1 z)) (resp. exp( 2 Log(1 z))), which fix the branches. 33

Remark 3.7. It is noteworthy that the differential equation is independent of the choice of an intermediate state h ∈ A(h0, h1, h2, h3). By Lemma 3.3, for h ∈ A(h0, h1, h2, h3), the as- ymptotic behavior of Ch (z) is zh−h2−h3 . Thus, different choices of h ∈ A( 1 , 1 , 1 , 1 ) = h0,h1,h2,h3 16 16 16 16 1 {0, } correspond to linearly independent solution of D 1 1 1 1 . 2 16 , 16 , 16 , 16 Looking at Table 6, we find that the conformal block has some symmetry:

(h0, h1, h2, h3) ↔ (h0, h2, h1, h3) z ↔ z−1. In order to see this symmetry precisely, we have to consider the conformal block of two variables C0h (y, z) and the true normalization of intertwining operators I(−, z) h0,h1,h2,h3 (1.9). Since Ch (x, y) is a scalar multiple of C0h (x, y), the convergence of the h0,h1,h2,h3 h0,h1,h2,h3 formal power series Ch (x, y) follows from the convergence of C0h (z). h0,h1,h2,h3 h0,h1,h2,h3 To be more precise, we ﬁx the branch of the Ch (x, y). h0,h1,h2,h3 The series Ch (x, y) is absolutely convergent to a single-valued holomorphic func- h0,h1,h2,h3 tion in

(3.2) {(x, y) ∈ (C \ R−) × (C \ R−) | |x| > |y|}, which follows from the convergence of the series C0h (z) and Lemma 3.3. However, h0,h1,h2,h3 −1 r −1 −1 in (3.2), (x y) = exp(rLog(x y)) is ill-deﬁned if x y ∈ R− and is not always equal to x−ryr = exp(−rLog(x)) exp(rLog(y)) (see Lemma 3.5). Thus, it is convenient to consider smaller region, like {(x, y) ∈ C2 | Arg(x), Arg(y) are small}, where Log is additive.

Thus, set (x0, y0) = (4 + i, 2). For > 0, set

n 2 o U+() = (x, y) ∈ C | |x − x0| < , |y − y0| < , which is a neighborhood of (x0, y0). If is suﬃciently small, then U+() is contained in × Y2(C ) and {y/x ∈ C | (x, y) ∈ U+} is contained in V+ since y/x ∼ y0/x0 ∈ V+. Take such

> 0 and set U+ = U+(). The precise values of x0, y0, do not matter. In this region U , Ch (x, y) deﬁne a single-valued holomorphic function. Let + h0,h1,h2,h3 denote this holomorphic function by Ch (x, y)| . h0,h1,h2,h3 U+ Then, by Proposition 3.6 and Lemma 3.3, we have:

Proposition 3.8. The conformal blocks Ch (x, y) are given in Table 7: h0,h1,h2,h3

Here, the branches of functions in the table is defined as follows: Let (x, y) ∈ U+. Since 1 1 1 r 1 1 2 Re(x − y) > 0 (resp. Re(x 2 − y 2 ) > 0), (x − y) (resp. x 2 − y 2 ) is well-defined for the principle value. Thus, all the functions in the table are specified. We note that in U+ Log is additive, e.g., (x−1y)r = x−ryr by Lemma 3.5. Hence, there is no ambiguity in the expressions. Now, we can see the following symmetry from the table 7:

(h0, h1, h2, h3) ↔ (h0, h2, h1, h3) (x, y) ↔ (y, x). 34 Table 7. Conformal blocks

(h , h , h , h ) h Ch (x, y)| 0 1 2 3 h0,h1,h2,h3 U+ 1 1 1 1 1 − 1 1 1 1 1 1 1 { − } 8 2 2 2 2 − 2 2 ( 16 , 16 , 16 , 16 ) 0 2 xy(x y) (x + y ) + (x y ) 1 1 − 1 1 1 1 1 1 1 { − } 8 2 2 2 − 2 − 2 2 2 2 xy(x y) (x + y ) (x y ) 1 1 1 1 { − }−1 2 − 2 ( 2 , 2 , 2 , 2 ) 0 xy(x y) (x xy + y ) 1 1 1 1 − 1 − 1 y 8 { − } 2 − ( 2 , 2 , 16 , 16 ) 0 y x(x y) (x 2 ) 1 1 1 1 1 1 − 1 − 1 8 { − } 2 − ( 2 , 16 , 2 , 16 ) 16 2 x y(x y) (x 2y) 1 1 1 1 1 1 − 1 −1 2 − ( 16 , 2 , 2 , 16 ) 16 2 (xy) (x y) (x + y) 1 1 1 1 1 1 − 1 − 1 { } 2 − 8 ( 2 , 16 , 16 , 2 ) 16 2 xy (x y) (x + y) 1 1 1 1 1 1 −1 − 1 { − } 2 − ( 16 , 2 , 16 , 2 ) 16 2 x y(x y) (x 2y) 1 1 1 1 −1 − 1 y { − } 2 − ( 16 , 16 , 2 , 2 ) 0 y x(x y) (x 2 )

The precise statement of this symmetry is given in the next section.

Remark 3.9. If we consider four point correlation functions Ch (z , z , z , z ) = hI0 (a , z )Ih0 (a , z )Ih (a , z )Ih3 (a , z )1i, h0,h1,h2,h3 0 1 2 3 h0,h0 0 0 h1,h 1 1 h2,h3 2 2 h3,0 3 3 then the conformal block admits S 4-symmetry or more precisely the symmetry of the pure braided group P4. We note that this four point correlation function can be recovered from the most degenerate one Ch (z) (see [Mo2]). h0,h1,h2,h3 3.3. Connection matrix. We continue to use the notations and the definitions in the × × previous section. Let σ : Y2(C ) → Y2(C ) be an involution defined by σ(x, y) = (y, x). × Let γ : [0, 1] → Y2(C ) be a continuous map such that γ(0) = (x0, y0) and γ(1) = (y0, x0). For any h , h , h , h ∈ IS and h ∈ A(h , h , h , h ), the holomorphic function Ch (x, y)| 0 1 2 3 0 1 2 3 h0,h1,h2,h3 U+ × has an analytic continuation to the multivalued holomorphic function on Y2(C ). Let h h Aγ C (x, y) denote the function obtained by the analytic continuation of C (x, y)|U h0,h1,h2,h3 h0,h1,h2,h3 + along the path γ. The function A Ch (x, y) is defined on σ(U ), a neighborhood γ h0,h1,h2,h3 + × h × × of (y0, x0) ∈ Y2(C ). The pull back of Aγ C (x, y) by σ : Y2(C ) → Y2(C ) defines h0,h1,h2,h3 a holomorphic function on U . We denote it by σ∗A Ch (x, y) . + γ h0,h1,h2,h3 n 0 o ∗ h h | Then, σ Aγ Ch ,h ,h ,h (x, y) is a linear sum of functions Ch ,h ,h ,h (x, y) U+ 0 , 0 1 2 3 0 2 1 3 h ∈A(h0,h2,h1,h3) which can be checked case by case by Table 4 and Table 7. h,h0 More precisely, there exists a unique matrix {B (γ)} 0∈ such that h0,h1,h2,h3 h A(h0,h2,h1,h3) X 0 0 σ∗A Ch (x, y) = Bh,h (γ)Ch (x, y)| . γ h0,h1,h2,h3 h0,h1,h2,h3 h0,h2,h1,h3 U+ 0 h ∈A(h0,h2,h1,h3) h,h0 It is noteworthy that the matrix {B (γ)} 0∈ depends only on the homotopy h0,h1,h2,h3 h A(h0,h2,h1,h3) class of the path γ. × Let γ0 : [0, 1] → Y2(C ) be the path given by x + y x − y x + y x − y γ (t) = (x , y ) = 0 0 + exp(πit) 0 0 , 0 0 − exp(πit) 0 0 . 0 t t 2 2 2 2

Then, γ0(0) = (x0, y0) and γ0(1) = (y0, x0). 35

0 Hereafter, we calculate the connection matrices {Bh,h (γ )} for the path γ . Set h0,h1,h2,h3 0 0 0 0 Bh,h = Bh,h (γ ). h0,h1,h2,h3 h0,h1,h2,h3 0

Since xt −yt = exp(πit)(x0−y0), xt −yt : [0, 1] → C crosses the negative real line R− from the above once, and x , y : [0, 1] → C do not cross R . By using those facts, we can calcu- t t − p q − r p+r q P∞ r − k k| late the connection matrix of a function x y (x y) P(x, y) = P(x, y)x y k=0 k ( 1) (y/x) U+ , where p, q, r ∈ R and P(x, y) ∈ C[x, y], a polynomial. The branch of this function is cho- sen in the same way as in the previous section. Then, we have:

Lemma 3.10. For any p, q, r ∈ R and P(x, y) ∈ C[x, y],

∗ p q r p q r σ Aγ0 (x y (x − y) P(x, y)) = exp(πir)y x (x − y) P(y, x).

1 1 1 1 All connection matrices except for the case of (h0, h1, h2, h3) = ( 16 , 16 , 16 , 16 ) can be calculated by the above lemma. We note that except for this case (h0, h1, h2, h3) = 1 1 1 1 ( 16 , 16 , 16 , 16 ), there is only one possible intermediate state A(h0, h1, h2, h3), and the con- 0 nection matrices Bh,h are not matrices but scalars. Thus we sometimes omit the index h0,h1,h2,h3 0 of Bh,h as B in these cases. h0,h1,h2,h3 h0,h1,h2,h3 By Table 4, we have:

Lemma 3.11. For any h0, h1, h2, h3 ∈ IS and h ∈ h2 ? h3, Bh1,h2 = exp(πi(h − h − h )), 0,h1,h2,h3 3 1 2 Bh0,h3 = 1, h0,0,h2,h3 Bh3,h0 = 1, h0,h1,0,h3 Bh2,h1 = exp(πi(h − h − h )). h0,h1,h2,0 0 1 2 Similarly, by Table 7, we have:

Lemma 3.12. For any h0, h1, h2, h3 ∈ IS and h ∈ A(h0, h1, h2, h3) such that none of

{hi}i=0,1,2,3 is 0, the connection matrices are given in Table 8:

Table 8. Connection matrices

0 (h , h , h , h ) Bh,h 0 1 2 3 h0,h1,h2,h3  1+i 1−i  1 1 1 1 − 1  2 2  ( 16 , 16 , 16 , 16 ) exp( 8 πi)  1−i 1+i  2 2 1 1 1 1 − ( 2 , 2 , 2 , 2 ) 1 1 1 1 1 ( 2 , 2 , 16 , 16 ) i 1 1 1 1 ( 2 , 16 , 2 , 16 ) i 1 1 1 1 − ( 16 , 2 , 2 , 16 ) 1 1 1 1 1 − 1 ( 2 , 16 , 16 , 2 ) exp( 8 πi) 1 1 1 1 ( 16 , 2 , 16 , 2 ) i 1 1 1 1 ( 16 , 16 , 2 , 2 ) i 36

1 1 1 1 Proof. By Lemma 3.10, it suﬃces to consider the case of (h0, h1, h2, h3) = ( 16 , 16 , 16 , 16 ). 1 1 1 1 Along the path γ0, x 2 , y 2 and x 2 + y 2 remain in the region {z ∈ C | Rez > 0}. Thus, the 1 1 1 1 1 1 1 1 1 −1 branches of x 2 , y 2 and (x 2 + y 2 ) 2 do not change. Since x 2 − y 2 = (x 2 + y 2 ) (x − y) cross

R− once,

1 1 1 1 1 1 1 1 1 1 1 1 ∗ 2 2 2 2 2 2 2 2 2 2 2 2 σ Aγ0 (x + y ) + (x − y ) = (x + y ) + i(x − y ) and

1 1 1 1 1 1 1 1 1 1 1 1 ∗ 2 2 2 2 2 2 2 2 2 2 2 2 σ Aγ0 (x + y ) − (x − y ) = (x + y ) − i(x − y ) , which implies the assertion.

Combining Lemma 3.11 and Lemma 3.12, we obtain:

Proposition 3.13. For any ∗,

B∗,0,∗,∗ = B∗,∗,0,∗ = 1

B 1 1 = −1 ∗, 2 , 2 ,∗  i a or a0 1  ( = 2 ) Ba, 1 , 1 ,a0 = Ba, 1 , 1 ,a0 =  2 16 16 2 −i otherwise,  0 1 0 1 (a, a , , a = a )  16  0 1 0 0 1 i (a, a , , a , a ) B(b,b ) = exp(− πi) ×  16 a, 1 , 1 ,a0  16 16 8  1+i (a = a0 = 1 , b = b0)  2 16  1−i 0 1 0  2 (a = a = 16 , b , b ).

We end this section by generalizing the above results to the whole conformal blocks.

Recall that for (h0, h1, h2, h3) ∈ IS and h ∈ A(h0, h1, h2, h3), the conformal block is a linear map

h 1 ∨ 1 1 1 C : L( , h0) ⊗ L( , h1) ⊗ L( , h2) ⊗ L( , h3) → C((x))((y)), h0,h1,h2,h3 2 2 2 2 deﬁned by

Ch (a∗, a , a , a ; x, y) = ha∗, Ih0 (a , x)Ih (a , y)a i, h0,h1,h2,h3 0 1 2 3 0 h1,h 1 a2,h3 2 3

∗ ∈ 1 ∨ ∈ 1 for a0 L( 2 , h0) and ai L( 2 , hi)(i = 1, 2, 3). As we will see in the following lemma, the whole conformal blocks Ch (a∗, a , a , a ; x, y) can be recovered from the spe- h0,h1,h2,h3 0 1 2 3 cialized conformal blocks Ch (x, y) [BPZ]: h0,h1,h2,h3

Lemma 3.14. For any ∆ ∈ R and a∗ ∈ L( 1 , h )∨ and a ∈ L( 1 , h ) (i = 0, 1, 2, 3), the i 0 2 0 ∆0 i 2 i ∆i following equalities hold: 37

(1) For any n ∈ Z,

Ch (L(−n)a∗, a , a , a ; x, y) h0,h1,h2,h3 0 1 2 3 ! n+1 d n+1 d n n h ∗ = x + y + (n + 1)x ∆a + (n + 1)y ∆a C (a , a1, a2, a3; x, y) dx dy 1 2 h0,h1,h2,h3 0 ∞ ! X n + 1 h ∗ n−k h ∗ n−k + C (a , L(k)a1, a2, a3; x, y)x + C (a , a1, L(k)a2, a3; x, y)y k + 1 h0,h1,h2,h3 0 h0,h1,h2,h3 0 k=1 +Ch (a∗, a , a , L(n)a ; x, y); h0,h1,h2,h3 0 1 2 3 (2) For any n ∈ Z,

Ch (a∗, a , a , L(−n)a ; x, y) h0,h1,h2,h3 0 1 2 3 ! −n+1 d −n+1 d −n −n h ∗ = − x + y + (−n + 1)x ∆a + (−n + 1)y ∆a C (a , a1, a2, a3; x, y) dx dy 1 2 h0,h1,h2,h3 0 ∞ ! X −n + 1 h ∗ −n−k h ∗ −n−k − C (a , L(k)a1, a2, a3; x, y)x + C (a , a1, L(k)a2, a3; x, y)y k + 1 h0,h1,h2,h3 0 h0,h1,h2,h3 0 k=1 +Ch (L(n)a∗, a , a , a ; x, y). h0,h1,h2,h3 0 1 2 3 (3)

h ∗ d h ∗ C (a , L(−1)a1, a2, a3; x, y) = C (a , a1, a2, a3; x, y) h0,h1,h2,h3 0 dx h0,h1,h2,h3 0 h ∗ d h ∗ C (a , a1, L(−1)a2, a3; x, y) = C (a , a1, a2, a3; x, y). h0,h1,h2,h3 0 dy h0,h1,h2,h3 0 (4) For m ≥ 0,

Ch (a∗, L(−m − 2)a , a , a ; x, y) h0,h1,h2,h3 0 1 2 3 ! X k k−m h ∗ = x C (L(2 + k)a , a1, a2, a3; x, y) m h0,h1,h2,h3 0 k≥0 ! −k − 1 1 h ∗ + ||x|>|y|C (a , a1, L(k − 1)a2, a3; x, y) m (x − y)−k−m−1 h0,h1,h2,h3 0 ! −k − 1 h ∗ −1−k−m + C (a , a1, a2, L(k − 1)a3; x, y)x . m h0,h1,h2,h3 0 (5) For m ≥ 0,

Ch (a∗, a , L(−m − 2)a , a ; x, y) h0,h1,h2,h3 0 1 2 3 ! X k k−m h ∗ = y C (L(2 + k)a , a1, a2, a3; x, y) m h0,h1,h2,h3 0 k≥0 ! k+m+1 −k − 1 1 h ∗ + (−1) ||x|>|y|C (a , L(k − 1)a1, a2, a3; x, y) m (x − y)−k−m−1 h0,h1,h2,h3 0 ! −k − 1 h ∗ −1−k−m + C (a , a1, a2, L(k − 1)a3; x, y)y . m h0,h1,h2,h3 0 38

Proof. By Lemma 1.8,

Ch (a∗, L(−m − 2)a , a , a ; x, y) h0,h1,h2,h3 0 1 2 3 m ! ! m ! ! ∗ 1 d − h0 h ∗ h0 1 d + h = a T (x) I (a1, x)I (a2, y)a3 + a I (a1, x) T (x) I (a2, y)a3 0 m! dx h1,h a2,h3 0 h1,h m! dx h2,h3 m ! ! ∗ 1 d − h0 h = a T (x) I (a1, x)I (a2, y)a3 0 m! dx h1,h h2,h3     X − − ! ∗  h0  k 1 1 h   + a I (a1, x)  ||x|>|y|I (L(k − 1)a2, y) a3 0  h1,h  m (x − y)−k−m−1 h2,h3   k≥0 m ! ! ∗ h0 h 1 d + + a I (a1, x)I (a2, y) T (x) a3 0 h1,h a2,h3 m! dx ! X k k−m h ∗ = x C (L(2 + k)a , a1, a2, a3; x, y) m h0,h1,h2,h3 0 k≥0 ! −k − 1 1 h ∗ + ||x|>|y|C (a , a1, L(k − 1)a2, a3; x, y) m (x − y)−k−m−1 h0,h1,h2,h3 0 ! −k − 1 h ∗ −1−k−m + C (a , a1, a2, L(k − 1)a3; x, y)x . m h0,h1,h2,h3 0

The other cases are obtained in the same way.

The following proposition says that for any a the connection matrix of Ch (a∗, a , a , a ; x, y) i h0,h1,h2,h3 0 1 2 3 is the same as Ch (x, y). h0,h1,h2,h3

∗ ∈ 1 ∨ ∈ 1 Proposition 3.15. For any a0 L( 2 , h0) and ai L( 2 , hi) (i = 0, 1, 2, 3), the formal power series Ch (a∗, a , a , a ; x, y) is absolutely convergent in U and has the ana- h0,h1,h2,h3 0 1 2 3 + × lytic continuation to a multivalued holmorphic function on Y2(C ). Furthermore,

X 0 0 σ∗A Ch (a∗, a , a , a ; x, y) = Bh,h Ch (a∗, a , a , a ; x, y)| . γ0 h0,h1,h2,h3 0 1 2 3 h0,h1,h2,h3 h0,h2,h1,h3 0 2 1 3 U+ 0 h ∈A(h0,h2,h1,h3)

Proof. We may assume that a∗ ∈ L( 1 , h )∨ and a ∈ L( 1 , h ) (i = 0, 1, 2, 3) for some 0 2 0 h0+k0 i 2 i hi+ki ki ∈ Z≥0. We prove the proposition by the induction on K = k0 + k1 + k2 + k3. For K = 0, 1 | i since L( 2 , h)h is spanned by h , the equation follows from the deﬁnition of the connection 0 matrix. For K > 0, then one of ai can be written as ai = L(−n)a i with n > 0. 0 We only consider the case of a1 = L(−m − 2)a 1 with m ≥ 0. All the other cases can be shown in the same way. We note that for any (single-valued) polynomial p(x, y) ∈ C[x±, y±, (x − y)±] and multivalued function f (x, y),

∗ ∗ ∗ ∗ σ Aγ0 (p(x, y) f (x, y)) = σ Aγ0 (p(x, y)) σ Aγ0 ( f (x, y)) = p(y, x)σ Aγ0 ( f (x, y)) . 39

Thus, by Lemma 3.14, σ∗A Ch (a∗, L(−m − 2)a0 , a , a ; x, y) γ0 h0,h1,h2,h3 0 1 2 3 ! X k ∗ k−m h ∗ 0 = σ Aγ x C (L(2 + k)a , a 1, a2, a3; x, y) m 0 h0,h1,h2,h3 0 k≥0 ! ! −k − 1 ∗ 1 h ∗ 0 + σ Aγ ||x|>|y|C (a , a , L(k − 1)a2, a3; x, y) m 0 (x − y)−k−m−1 h0,h1,h2,h3 0 1 ! −k − 1 ∗ h ∗ 0 −1−k−m + σ Aγ C (a , a , a2, L(k − 1)a3; x, y) x m 0 h0,h1,h2,h3 0 1 ! X k k−m ∗ h ∗ 0 = y σ Aγ C (L(2 + k)a , a 1, a2, a3; x, y) m 0 h0,h1,h2,h3 0 k≥0 ! −k − 1 k+m+1 1 ∗ h ∗ 0 + (−1) ||x|>|y|σ Aγ C (a , a , L(k − 1)a2, a3; x, y) m (x − y)k+m+1 0 h0,h1,h2,h3 0 1 ! −k − 1 −1−k−m ∗ h ∗ 0 + y σ Aγ C (a , a , a2, L(k − 1)a3; x, y) . m 0 h0,h1,h2,h3 0 1 By the induction hypothesis and Lemma 3.14 (3), ! X 0 X k 0 RHS = Bh,h yk−mCh (L(2 + k)a∗, a , a0 , a ; x, y) h0,h1,h2,h3 h0,h2,h1,h3 0 2 1 3 0 m h ∈A(h0,h2,h1,h3) k≥0 ! −k − 1 k+m+1 1 h0 ∗ 0 + (−1) ||x|>|y|C (a , L(k − 1)a2, a , a3; x, y) m (x − y)k+m+1 h0,h1,h2,h3 0 1 ! ! −k − 1 −1−k−m h0 ∗ 0 + y C (a , a2, a , L(k − 1)a3; x, y) , m h0,h2,h1,h3 0 1 which is equal to

X 0 0 Bh,h Ch (a∗, a , L(−m − 2)a0 , a ; x, y) h0,h1,h2,h3 h0,h2,h1,h3 0 2 1 3 0 h ∈A(h0,h2,h1,h3) by Lemma 3.14 (5). Thus, the assertion holds.

i i i ¯ i ¯ i (l,r) 3.4. Multi-index connections. Let l, r ∈ Z>. For λ = (h1,..., hl, h1 ..., hr) ∈ IS , λ = ¯ 0 1 2 3 0 0 0 ¯ 0 ¯ 0 0 2 1 3 (h1,..., hl, h1,..., hr) ∈ A(λ , λ , λ , λ ) and λ = (h1,..., hl , h1,..., hr) ∈ A(λ , λ , λ , λ ), a multi-index connection matrix is deﬁned by

0 0 0 h ,h h¯ j,h¯ λ,λ ≡ l i i r j ∈ C Bλ0,λ1,λ2,λ3 Πi=1B 0 1 2 3 Π j=1B¯0 ¯1 ¯2 ¯3 , hi ,hi ,hi ,hi h j ,h j ,h j ,h j

i (l,r) 0 1 2 3 ∗ Proposition 3.16. Let λ ∈ IS (i = 0, 1, 2, 3) and λ ∈ A(λ , λ , λ , λ ). For any u0 ∈ 1 ∨ ∈ 1 λ ∗ L( 2 , h0) and ui L( 2 , hi) (i = 1, 2, 3), the formal power series Cλ0,λ1,λ2,λ3 (u0, u1, u2, u3; x, y) 40 is absolutely convergent in U+ and has the analytic continuation to a multivalued holmor- × phic function on Y2(C ). Furthermore, X 0 0 ∗ λ ∗ λ,λ λ ∗ | σ Aγ0 Cλ0,λ1,λ2,λ3 (u0, u1, u2, u3; x, y) = Bλ0,λ1,λ2,λ3Cλ0,λ1,λ2,λ3 (u0, u1, u2, u3; x, y) U+ . λ0∈A(λ0,λ2,λ1,λ3) The purpose of this section is to give a convenient combinatorial description of these (l,r) l+r l+r connection matrices. We will ﬁrst identify IS as a subset of Z2 × Z2 . 2 We start from the case of (l, r) = (1, 0). Deﬁne a map wt : Z2 → Q by 1 1 wt(d, c) ≡ d + c 16 2 2 2 for (d, c) ∈ Z2. Then, we can identify IS as the subset {(d, c) ∈ Z2 | dc = 0}, i.e., 1 1 0 ↔ (0, 0), ↔ (0, 1), ↔ (1, 0). 2 16 Then, the fusion rule can be written as (d, c) ? (d0, c0) = (d + d0, (1 + d)c0 + (1 + d0)c + dd0{0, 1})

0 0 for (d, c), (d , c ) ∈ IS, where we used the ring structure on Z2. (l,r) l+r l+r We generalize the above identification to IS and Z2 ×Z2 . For c = (c1,..., cl, c¯1,..., c¯r) 0 0 0 0 0 l+r 0 0 0 0 0 and c = (c1,..., cl , c¯1,..., c¯r) ∈ Z2 , we define cc = (c1c1,..., clcl , c¯1c¯1,..., c¯rc¯r) ∈ l+r l+r l+r Z2 , that is, the usual ring structure on Z2 . We also define maps |−|l, |−|r, |−| : Z2 → Z Pl Pr by |c|l = i=1 ci, |c|r = i=1 c¯i and |c| = |c|l − |c|r.

1 2 l+r 1 2 1 2 1 2 |c1+c2| Remark 3.17. For c , c ∈ Z2 , |c + c | = |c | + |c | − 2|c c |. In particular, (−1) = (−1)|c1|+|c2|. We will frequently use this formula.

l+r ¯ ¯ An element of Z2 is called a codeword. For a codeword d = (d1,..., dl, d1,..., dr), we deﬁne d l+r Z2 ≡ {c ∈ Z2 | dc = c}, l+r which is the subset of Z2 consisting of all codewords supported by d. The all-one vector l+r l+r l+r l+r d is a codeword 1 = (1, 1,..., 1) ∈ Z2 and set d⊥ = 1 + d for d ∈ Z2 . Then, Z2 is the kernel of the left multiplication by d⊥. (l,r) l+r l+r l+r l+r We identify IS as a subset of Z2 × Z2 consisting of elements (d, c) ∈ Z2 × Z2 satisfying dc = (0,..., 0). For µ = (d, c) ∈ IS(l,r), d is called d-part of µ and c is called c-part of µ. l+r l+r (l,r) An explicit bijection is given by {(d, c) ∈ Z2 × Z2 | dc = (0,..., 0)} → IS ,

(d1,..., dl, d¯1,..., d¯r, c1,..., cl, c¯1,..., c¯r) 7→ (wt(d1, c1),..., wt(dl, cl), wt(d¯1, c¯1) ..., wt(d¯r, c¯r)).

(l,r) ` d⊥ Then, IS = ∈ l+r {d} × Z and we have: d Z2 2 Lemma 3.18. The fusion product on IS(l,r) can be written as

1 1 2 2 1 2 1 2 1 2 (d , c ) ? (d , c ) = {(d + d , (d + d )⊥(c + c ) + γ)} ∈ d1d2 . γ Z2

1 2 1 2 1 1 2 2 (l,r) 1 d⊥ 2 d⊥ 1 2 (d +d )⊥ for (d , c ), (d , c ) ∈ IS . In particular, {d } × Z2 ? {d } × Z2 ⊂ {d +d }×Z2 . 41

l+r By the above lemma, the fusion rule can be described by using the product on Z2 . l+r The intermediate states can also be described by the product on Z2 :

Lemma 3.19. Let (d0, c0), (d1, c1), (d2, c2), (d3, c3) ∈ IS(l,r).

l+r (1) For α ∈ Z2 , (0, α) is a simple current, i.e., #(0, α) ? (d, c) = 1 for any (d, c) ∈ IS(l,r); (2) (0, 0) ∈ (d1, c1) ? (d2, c2) if and only if d1 = d2 and c1 = c2; 2 3 1 0 0 1 1 2 2 3 3 1 d d d⊥ 2 (3)A (d , c ), (d , c ), (d , c ), (d , c ) , ∅ if and only if there exits γ ∈ Z2 ,γ ∈ 3 1 2 1 2 3 d d d⊥ 3 d d d⊥ Z2 and γ ∈ Z2 such that

d0 = d1 + d2 + d3 0 1 2 3 1 2 3 1 2 3 c = (d + d + d )⊥(c + c + c ) + γ + γ + γ ;

2 3 1 3 1 2 1 2 3 1 d d d⊥ 2 d d d⊥ 3 d d d⊥ (4) For any γ ∈ Z2 ,γ ∈ Z2 and γ ∈ Z2 ,

1 2 3 1 2 3 1 2 3 1 2 3 1 1 2 2 3 3 A (d + d + d , (d + d + d )⊥(c + c + c ) + γ + γ + γ ), (d , c ), (d , c ), (d , c ) 2 3 2 3 2 3 1 s = {(d + d , (d + d )⊥(c + c ) + γ + γ )} s∈ d1d2d3 γ Z2 and

1 2 3 1 2 3 1 2 3 1 2 3 2 2 1 1 3 3 A (d + d + d , (d + d + d )⊥(c + c + c ) + γ + γ + γ ), (d , c ), (d , c ), (d , c ) 1 3 1 3 1 3 2 u = {(d + d , (d + d )⊥(c + c ) + γ + γ ))} u∈ d1d2d3 . γ Z2 In particular, #A (d0, c0), (d1, c1), (d2, c2), (d3, c3) ≤ 1 if d1d2d3 = 0.

Proof. (1), (2) clearly follows from Lemma 3.18. By Lemma 3.18, (3.3) (d1, c1) ? (d2, c2) ? (d3, c3)

1 1 2 3 2 3 2 3 d2d3 = (d , c ) ? (d + d , (d + d )⊥(c + c ) + Z2 ) 1 2 3 1 2 3 1 2 3 2 3 1 d2d3 d1(d2+d3) = (d + d + d , (d + d + d )⊥(c + (d + d )⊥(c + c )) + d⊥Z2 + Z2 ).

1 2 3 2 3 1 3 1 2 1 2 3 1 d2d3 d (d +d ) d d d⊥ d d d⊥ d d d⊥ 2 2 We note that d⊥Z2 + Z2 = Z2 + Z2 + Z2 . Since d c = 0, we have:

3 1 2 1 2 3 2 3 2 1 2 3 3 2 2 1 3 2 2 d d d⊥ (d + d + d )⊥(d + d )c = (d + d + d )⊥d d⊥c = d d d⊥c ∈ Z2 .

1 2 3 2 3 2 3 1 2 3 2 3 Hence, (3) follows from (d + d + d )⊥(d + d )⊥(c + c ) = (d + d + d )⊥(c + c ) + 1 2 3 2 3 2 3 (d + d + d )⊥(d + d )(c + c ). (4) easily follows from (3.3).

Remark 3.20. More than two intermediate states exist if and only if there is a compo- 1 1 1 1 1 nent which is equal to ( 16 , 16 , 16 , 16 ). Since the d-part corresponds to 16 , all the possible d1d2d3 0 1 2 3 intermediate states are parameterized by Z2 . We note that since d = d + d + d , d0d1d2d3 (d1+d2+d3)d1d2d3 d1d2d3 i i i i l+r Z2 = Z2 = Z2 by d d = d for d ∈ Z2 . Thus, the intermediate states are symmetric for d0, d1, d2, d3. 42

If d1d2d3 = 0, then there is only one intermediate state and the connection matrix is a scalar. So we set

2 3 2 3 2 3 1 3 1 3 1 3 (d +d ,(d +d )⊥(c +c )),(d +d ,(d +d )⊥(c +c )) B 0 0 1 1 2 2 3 3 B (d ,c ),(d ,c ),(d ,c ),(d ,c ) = (d0,c0),(d1,c1),(d2,c2),(d3,c3) for short. Now, we can state the main result on this section, which follows from Proposition 3.13.

2 3 1 3 1 2 1 1 2 2 3 3 (l,r) 1 d d d⊥ 2 d d d⊥ Theorem 3.21. Let (d , c ), (d , c ), (d , c ) ∈ IS and γ ∈ Z2 ,γ ∈ Z2 and 1 2 3 3 d d d⊥ 0 1 2 3 0 1 2 3 1 2 3 1 2 3 γ ∈ Z2 . Set d = d +d +d and c = (d +d +d )⊥(c +c +c )+γ +γ +γ . Then, 0 l+r 2 3 0 0 1 1 2 2 3 3 1 3 0 for any c, c ∈ Z2 with (d +d , c) ∈ A((d , c ), (d , c ), (d , c ), (d , c )) and (d +d , c ) ∈ A((d0, c0), (d2, c2), (d1, c1), (d3, c3)),

2 3 1 3 0 1 2 1 2 0 3 2 1 0 3 1 2 2 1 1 2 0 3 B(d +d ,c),(d +d ,c ) − |c c | − |d c (c +c )|+|d c (c +c )|i−|d c |−|d c |+|d d (c +c )| (d0,c0),(d1,c1),(d2,c2),(d3,c3) = ( 1) ( 1)

−πi 1 + i 1 2 3 1 − i 1 2 3 1 2 3 0 exp( |d1d2|)( )|d d d |l ( )|d d d |r (−i)|d d d (c+c )|. 8 2 2 Furthermore, if d1 = (0,..., 0), then

|c1c2| |c1d2| |d2c1(c0+c3)| B(d0,c0),(0,c1),(d2,c2),(d3,c3) = (−1) (−i) (−1) .

We end this section by showing that the multi-index conformal blocks are linearly independent. We will use the following lemma:

Lemma 3.22. Let f1(y), f2(y), g1(y), g2(y) ∈ C((y, y¯)) such that

(1)f 1(y) and f2(y) are linearly independent over C;

(2)f 1(y)g1(y) + f2(y)g2(y) = 0;

(3) (∂y f1(y))g1(y) + (∂y f2(y))g2(y) = 0 and (∂y¯ f1(y))g1(y) + (∂y¯ f2(y))g2(y) = 0.

Then, g1(y) = g2(y) = 0.

−1 −1 Proof. We assume that g1(y) , 0. By (2) and (3), ∂y g1(y)g2(y) = ∂y¯ g1(y)g2(y) = 0, which contradicts (1).

0 1 2 3 (l,r) λ ∈ Z ∈ { − − − − } 0 1 2 3 Proposition 3.23. For any l, r ≥0 and λ , λ , λ , λ IS , Cλ0,λ1,λ2,λ3 ( , , , ; x, y) λ∈A(λ ,λ ,λ ,λ ) are linearly independent as vectors in 0 ∨ 1 2 3 R R R R HomC Ll,r(λ ) ⊗ Ll,r(λ ) ⊗ Ll,r(λ ) ⊗ Ll,r(λ ), C[[x , x¯ , y , y¯ ]] . over C.

λ { − − − − } 0 1 2 3 Proof. It suﬃces to show that Cλ0,λ1,λ2,λ3 ( , , , ; 1, y) λ∈A(λ ,λ ,λ ,λ ) are linearly inde- 0 ∨ 1 2 3 1 1 pendent as vectors in HomC Ll,r(λ ) ⊗ Ll,r(λ ) ⊗ Ll,r(λ ) ⊗ Ll,r(λ ), C((y 16 , y¯ 16 )) , where 1 1 1 1 C((y 16 , y¯ 16 )) is the ﬁeld of formal Laurent series with the formal valuables y 16 andy ¯ 16 . i 1 l,r 1 1 1 ∈ (l,r) We may assume that all λ are equal to 16 = ( 16 , 16 ,..., 16 ) IS . In this case, 1 l,r 1 l,r 1 l,r 1 l,r { 1 }l+r A( 16 , 16 , 16 , 16 ) = 0, 2 . We prove the assertion by the induction of l + r. If l + r = 1, then the assertion clearly follows. Let l + r > 1 and assume that there exists

{kλ ∈ C} 1 1 such that λ∈{0, 2 }×···×{0, 2 } X λ − − − − kλC 1 l,r 1 l,r 1 l,r 1 l,r ( , , , ; 1, y) = 0. 16 , 16 , 16 , 16 1 l+r λ∈{0, 2 } 43

Then,     0  X λ0  0 = C (−, −, −, −; 1, y)  k 0C (−, −, −, −; 1, y) 1 , 1 , 1 , 1  0,λ 1 l−1,r 1 l−1,r 1 l−1,r 1 l−1,r  16 16 16 16  16 , 16 , 16 , 16   0 1 l+r−1  λ ∈{0, 2 } (3.4)   1  X 0  2  λ  + C 1 1 1 1 (−, −, −, −; 1, y)  k 1 0C l−1,r l−1,r l−1,r l−1,r (−, −, −, −; 1, y) . , , ,  2 ,λ 1 1 1 1  16 16 16 16  16 , 16 , 16 , 16   0 1 l+r−1  λ ∈{0, 2 } | 1 i − | 1 i ∈ 1 1 By Lemma 3.22, substituting suitable vectors 16 , L( 1) 16 L( 2 , 16 ) into the entries P λ0 0 of (3.4), we can show that both 0 1 l+r−1 k0,λ C l−1,r l−1,r l−1,r l−1,r (−, −, −, −; 1, y) and λ ∈{0, 2 } 1 1 1 1 16 , 16 , 16 , 16 P λ0 0 1 l+r−1 k 1 0C l−1,r l−1,r l−1,r l−1,r (−, −, −, −; 1, y) are equal to 0. Thus, by the induc- λ ∈{0, 2 } 2 ,λ 1 1 1 1 16 , 16 , 16 , 16 tion assumption, all kλ are equal to 0. Hence, the assertion holds.

4. Framed algebra Motivated by the study in last section, we introduce a notion of a famed algebra in this section. In Section 4.1, we define a notion of an (l, r)-framed algebra and show that for any framed algebra we can construct a framed full VOA and vice versa. In Section 4.2- 4.5, we develop a general theory of a framed algebra. In Section 4.6, we study VOAs constructed from trivial framed algebras. L ∈ 4.1. Definition of framed algebra. Let l, r Z≥0 and S = λ∈IS(l,r) S λ be a finite- dimensional IS(l,r)-graded vector space equipped with a linear map · : S ⊗ S → S and a distinguished non-zero element 1 ∈ S 0 such that: (l,r) FA1) For any λ ∈ IS , S λ = 0 unless s(λ) ∈ Z.

FA2) S 0 = C1 and for any a ∈ S , a · 1 = 1 · a = a; 1 2 (l,r) L FA3) For any λ , λ ∈ IS , a1 ∈ S λ1 and a2 ∈ S λ2 , a1 · a2 ∈ λ∈λ1?λ2 S λ. (l,r) Recall that for λ ∈ IS πλ : S → S is the projection of the graded vector space S onto S λ and the composition of the product · and the projection πλ is denoted by ·λ : S ⊗ S → S λ. L ⊗ By Proposition 2.15, FS = λ∈IS(l,r) Lr,s(λ) S λ is a full prevertex algebra. Proposition 4.1. The following conditions are equivalent:

(1) The full prevertex algebra FS is a full vertex algebra; i (l,r) 0 0 2 1 3 (2) For any λ ∈ IS , ai ∈ S λi (i=1,2,3) and λ ∈ A(λ , λ , λ , λ ), X λ,λ0 · 0 · 0 · 0 · a2 λ (a1 λ a3) = Bλ0,λ1,λ2,λ3 a1 λ (a2 λ a3). λ∈A(λ0,λ1,λ2,λ3)

0 1 2 3 (l,r) i ∗ ∗ 0 ∗ Proof. Let λ , λ , λ , λ ∈ IS and ui ⊗ ai ∈ Ll,r(λ ) ⊗ S λi and u0 ⊗ a0 ∈ Ll,r(λ ) ⊗ (S λ0 ) . First, we assume that FS is a full vertex algebra. Then, by (FV5), there exists µ(z1, z2) ∈

GCor2 such that hu∗ ⊗ a∗ Y u ⊗ a z Y u ⊗ a z u ⊗ a i z z | 0 0, ( 1 1, 1) ( 2 2, 2) 3 3 = µ( 1, 2) |z1|>|z2| hu∗ ⊗ a∗ Y u ⊗ a z Y u ⊗ a z u ⊗ a i z z | 0 0, ( 2 2, 2) ( 1 1, 1) 3 3 = µ( 1, 2) |z2|>|z1|. 44

Let us consider the path γ0 in Section 3.3. By (2.6) and Proposition 3.16, ∗A hu∗ ⊗ a∗ Y u ⊗ a z Y u ⊗ a z u ⊗ a i σ γ0 0 0, ( 1 1, 1) ( 2 2, 2) 3 3    X  = σ∗A  Cλ (u , u , u , u ; z , z )ha∨, a · (a · a )i . γ0  λ0,λ1,λ2,λ3 0 1 2 3 1 2 0 1 λ0 2 λ 3  λ∈A(λ0,λ1,λ2,λ3) (♣) X h ∨ · · i ∗ λ = a0 , a1 λ0 (a2 λ a3) σ Aγ0 Cλ0,λ1,λ2,λ3 (u0, u1, u2, u3; z1, z2) λ∈A(λ0,λ1,λ2,λ3)   X  X 0 0  = ha∨, a · (a · a )i  Bλ,λ Cλ (u , u , u , u ; z , z ) 0 1 λ0 2 λ 3  λ0,λ1,λ2,λ3 λ0,λ2,λ1,λ3 0 2 1 3 1 2  λ∈A(λ0,λ1,λ2,λ3) λ0∈A(λ0,λ2,λ1,λ3)   X 0  X 0  = Cλ (u , u , u , u ; z , z )  Bλ,λ ha∨, a · (a · a )i . λ0,λ2,λ1,λ3 0 2 1 3 1 2  λ0,λ1,λ2,λ3 0 1 λ0 2 λ 3  λ0∈A(λ0,λ2,λ1,λ3) λ∈A(λ0,λ1,λ2,λ3) hu∗ ⊗ a∗ Y u ⊗ a z Y u ⊗ a z u ⊗ a i Since 0 0, ( 1 1, 1) ( 2 2, 2) 3 3 is absolutely convergent to the single- valued real analytic function µ(z1, z2), we have ∗A hu∗ ⊗ a∗ Y u ⊗ a z Y u ⊗ a z u ⊗ a i σ γ0 0 0, ( 1 1, 1) ( 2 2, 2) 3 3

= µ(z2, z1)|U+ hu∗ ⊗ a∗ Y u ⊗ a z Y u ⊗ a z u ⊗ a i| = 0 0, ( 2 2, 1) ( 1 1, 2) 3 3 U+ . Combining this with (♣), we have:

X λ0 ∨ h · · 0 i Cλ0,λ2,λ1,λ3 (u0, u2, u1, u3; z1, z2) a0 , a2 λ0 (a1 λ a3) λ0∈A(λ0,λ2,λ1,λ3)   X 0  X 0  = Cλ (u , u , u , u ; z , z )  Bλ,λ ha∨, a · (a · a )i . λ0,λ2,λ1,λ3 0 2 1 3 1 2  λ0,λ1,λ2,λ3 0 1 λ0 2 λ 3  λ0∈A(λ0,λ2,λ1,λ3) λ∈A(λ0,λ1,λ2,λ3) By Proposition 3.23, we must have

∨ X λ,λ0 ∨ h · · 0 i h · · i a0 , a2 λ0 (a1 λ a3) = Bλ0,λ1,λ2,λ3 a0 , a1 λ0 (a2 λ a3) λ∈A(λ0,λ1,λ2,λ3) for any λ0 ∈ A(λ0, λ2, λ1, λ3), that is, (2) holds. F z z hu∗ ⊗ a∗ Y u ⊗ a z Y u ⊗ Conversely, assume that (2) holds. Set ( 1, 2) = 0 0, ( 1 1, 1) ( 2 a z u ⊗ a i F z z 2, 2) 3 3 . By Proposition 3.16, the formal power series ( 1, 2) is absolutely con- × vergent to a (possibly) multi-valued real-analytic function on Y2(C ). We will ﬁrst show that this function is in fact single-valued. For the sake of simplicity, we assume that i 0 ui = |λ i and u0 = hλ |, that is, the lowest weight vectors. By the similar argument in R R Lemma 3.3, there exists a formal power series G(z) ∈ C[[z , z¯ ]] such that F(z1, z2) = l(λ0)−l(λ1)−l(λ2)−l(λ3) r(λ0)−r(λ1)−r(λ2)−r(λ3) i i i z1 z¯1 G(z2/z1). By (FA1), s(λ ) = l(λ ) − r(λ ) ∈ Z for all i = l(λ0)−l(λ1)−l(λ2)−l(λ3) r(λ0)−r(λ1)−r(λ2)−r(λ3) r(λ0)−r(λ1)−r(λ2)−r(λ3) s(λ0)−s(λ1)−s(λ2)−s(λ3) 0, 1, 2, 3. Thus, z1 z¯1 = (z1z¯1) z1 × is a single-valued real analytic function on Y2(C ). 45

× 1 1 Since the image of the map Y2(C ) → CP , (z1, z2) 7→ z2/z1 is CP \{0, 1, ∞}, it suﬃces to show that the formal power series G(z) is single-valued on CP1 \{0, 1, ∞}. We 1 note that the fundamental group π1(CP \{0, 1, ∞}) is generated by two cycles. One is a cycle around 0 and the other is a cycle around ∞. Since by Lemma 3.3 X h ∨ · · i λ G(z) = a0 , a1 λ0 (a2 λ a3) Cλ0,λ1,λ2,λ3 (z)(4.1) λ∈A(λ0,λ1,λ2,λ3) X l(λ)−l(λ2)−l(λ3) r(λ)−r(λ2)−r(λ3) ∨ = z z¯ ha0 , a1 ·λ0 (a2 ·λ a3)i + O(z, z¯) , λ∈A(λ0,λ1,λ2,λ3) the monodromy of G(z) around z = 0 is trivial by (FA1). In particular, G(z) ∈ C((z, z¯, |z|R)).

Next we consider the monodromy of G(z) around z = ∞. More precisely, let p∞ be a small cycle around ∞ and we will consider the analytic continuation of G(z) along the −1 path γ0 ◦ p∞ ◦ γ0. By (♣) and the assumption (2), (4.2) ∗A hu∗ ⊗ a∗ Y u ⊗ a z Y u ⊗ a z u ⊗ a i σ γ0 0 0, ( 1 1, 1) ( 2 2, 2) 3 3   X 0  X 0  = Cλ (u , u , u , u ; z , z )  Bλ,λ ha∨, a · (a · a )i λ0,λ2,λ1,λ3 0 2 1 3 1 2  λ0,λ1,λ2,λ3 0 1 λ0 2 λ 3  λ0∈A(λ0,λ2,λ1,λ3) λ∈A(λ0,λ1,λ2,λ3) X λ0 ∨ h · · 0 i = Cλ0,λ2,λ1,λ3 (u0, u2, u1, u3; z1, z2) a0 , a2 λ0 (a1 λ a3) λ0∈A(λ0,λ2,λ1,λ3) hu∗ ⊗ a∗ Y u ⊗ a z Y u ⊗ a z u ⊗ a i = 0 0, ( 2 2, 1) ( 1 1, 2) 3 3 .

−1 Thus, similarly to (4.1), by (FA1) again the monodromy of G(z) along the path γ0 ◦p∞◦γ0 is trivial. Denote the analytic continuation of G(z) by g(z) which is a single-valued real analytic function on CP1 \{0, 1, ∞}. Since multi-index Virasoro conformal blocks have expansions around 0, 1, ∞, g(z) has conformal singularities at 0, 1, ∞. Thus, g(z) ∈ F0,1,∞, which implies that there exists µ(z1, z2) ∈ GCor2 such that

hu∗ ⊗ a∗ Y u ⊗ a z Y u ⊗ a z u ⊗ a i z z | 0 0, ( 1 1, 1) ( 2 2, 2) 3 3 = µ( 1, 2) |z1|>|z2|. hu∗ ⊗ a∗ Y u ⊗ a z Y u ⊗ a z u ⊗ a i Furthermore, by (4.2), 0 0, ( 2 2, 2) ( 1 1, 1) 3 3 is also the expansion of the same function µ(z1, z2), that is,

hu∗ ⊗ a∗ Y u ⊗ a z Y u ⊗ a z u ⊗ a i z z | 0 0, ( 2 2, 2) ( 1 1, 1) 3 3 = µ( 1, 2) |z2|>|z1|

Thus, by Proposition 2.11 and Proposition 2.15, (1) holds.

Now, we introduce a notion of an (l, r)-framed algebra. An (l, r)-framed algebra (or (l,r) L framed algebra for short) is a ﬁnite-dimensional IS -graded vector space S = λ∈IS(l,r) S λ equipped with a linear map · : S ⊗ S → S and a distinguished non-zero element 1 ∈ S 0 which satisfy (FA1), (FA2) and (FA3) together with the following condition: 46

i (l,r) 0 0 1 2 3 FA4) For any λ ∈ IS , ai ∈ S λi and λ ∈ A(λ , λ , λ , λ ) (i=1,2,3),

X λ,λ0 · · 0 · · a2 λ0 (a1 λ a3) = Bλ0,λ1,λ2,λ3 a1 λ0 (a2 λ a3). λ∈A(λ0,λ1,λ2,λ3)

Remark 4.2. The framed algebra is nothing but an explicit deﬁnition of a commutative ⊗l ⊗r algebra object in the braided tensor category (RepVir 1 ) ⊗ (RepVir 1 ) . 2 2

Now, by Proposition 2.15, Proposition 4.1 and Lemma 2.14, we have:

Theorem 4.3. If (S, ·, 1) is a framed algebra, then (FS , Y, 1) in Proposition 2.15 is a framed full vertex operator algebra. Conversely, if (F, Y, 1) is a framed full vertex operator algebra, then (S F, ·, 1) in Lemma 2.14 is a framed algebra.

The purpose of this paper is to systematically construct examples of framed algebras. We note that we can deﬁne a category of framed algebra and prove that the category of framed algebras and the category of framed full vertex operator algebras are equivalent. But, we do not need this result for the main purpose of this paper. The interested reader may consult Appendix of this paper.

4.2. Associativity, ideal and bilinear form. The product of a framed algebra is commutative up to the connection matrix. In this section, we show that a framed algebra is “associative” and there always exists an invariant bilinear form. The following lemma says that a framed algebra is associative:

i (l,r) Lemma 4.4. Let S be a framed algebra. For λ ∈ IS and ai ∈ S λi , the following conditions hold:

s(µ)+s(λ1)+s(λ2) 0 1 2 (1)a 1 ·λ0 a2 = (−1) a2 ·λ0 a1 for λ ∈ λ ? λ . (2) For µ0 ∈ A(λ0, λ1, λ2, λ3) and λ0 ∈ IS(l,r),

s(µ0)+s(µ)+s(λ2)+s(λ0) X µ,µ0 · 0 · 0 − · · a1 λ (a2 µ a3) = ( 1) Bλ0,λ3,λ1,λ2 (a1 µ a2) λ0 a3. µ∈A(λ0,λ3,λ1,λ2)

Proof. Applying a3 = 1 to the deﬁnition (4) of a framed algebra and by Lemma 3.11, we have

a2 ·λ0 a1 = a2 ·λ0 (a1 ·λ1 1)

X (µ,λ1) ∨ B < a , a · 0 a · > = λ0,λ1,λ2,0 0 1 λ 2 µ 1 µ∈A(λ0,λ1,λ2,0) (λ2,λ1) B a · 0 a = λ0,λ1,λ2,0 1 λ 2 0 1 2 0 1 2 = exp(πi(l(λ ) − l(λ ) − l(λ ))) exp(−πi(r(λ ) − r(λ ) − r(λ )))a1 ·λ0 a2 s(λ0)+s(λ1)+s(λ2) = (−1) a1 ·λ0 a2, 47 where we used s(λi) ∈ Z. Thus,

a1 ·λ0 (a2 ·µ0 a3) s(µ0)+s(λ2)+s(λ3) = (−1) a1 ·λ0 (a3 ·µ0 a2) X 0 − s(µ0)+s(λ2)+s(λ3) µ,µ · · = ( 1) Bλ0,λ3,λ1,λ2 a3 λ0 (a1 µ a2) µ∈A(λ0,λ3,λ1,λ2) X 0 − s(µ0)+s(λ2)+s(λ3) µ,µ − s(λ0)+s(λ3)+s(µ) · · = ( 1) Bλ0,λ3,λ1,λ2 ( 1) (a1 µ a2) λ0 a3 µ∈A(λ0,λ3,λ1,λ2) X 0 − s(µ0)+s(µ)+s(λ2)+s(λ0) µ,µ · · = ( 1) Bλ0,λ3,λ1,λ2 (a1 µ a2) λ0 a3. µ∈A(λ0,λ3,λ1,λ2) An ideal of a framed algebra S is a IS(l,r)-graded subspace M ⊂ S such that a · m ∈ M for any a ∈ S and m ∈ M. By Lemma 4.4, any ideal of S is automatically a two-sided ideal, that is, m · a ∈ M for any a ∈ S and m ∈ M. A framed algebra is said to be simple if it does not contain a proper ideal. Then, we have:

Proposition 4.5. Let S be a framed algebra. Then, S is simple if and only if (FS , Y, 1) is a simple full vertex algebra.

Proof. Let S be a framed algebra. Suppose that I is an ideal of FS . Since I is decom- (l,r) L posed into simple Ll,r(0)-modules, there exists IS -graded subspace M = (l,r) Mλ ⊂ L L λ∈IS ⊗ λ∈IS(l,r) S λ such that I = λ∈IS(l,r) Ll,r(λ) Mλ. By deﬁnition of FS , M is an ideal of the framed algebra S . The reverse direction can be obtained similarly. We will use the following lemmas later: By the fusion rule, 0 ∈ λ1 ? λ2 if and only if λ1 = λ2. Hence, we have:

i (l,r) 1 2 Lemma 4.6. Let λ ∈ IS and ai ∈ S λi for i = 1, 2. If a1 ·0 a2 , 0, then λ = λ .

i (l,r) Lemma 4.7. Let S be a simple framed algebra and ai ∈ S λi for λ ∈ IS (i = 1, 2). If a1 · a2 = 0, then a1 = 0 or a2 = 0.

Proof. We may assume that a2 , 0. Then, by Lemma 4.4, S a2 ≡ {b · a2}b∈S is an ideal 0 of S . Since S is simple and a2 ∈ S a2, S a2 = S . Thus, there exists a ∈ S λ2 such that 0 0 2s(λ2) 0 a ·0 a2 = 1 by Lemma 4.6 and (FA2). Then, by Lemma 4.4, a2 ·0 a = (−1) a ·0 a2 = 1 0 P 0 and a1 = a1 · (a2 ·0 a ) = µ Coefficient × (a1 ·µ a2) · a = 0. Hence, a1 = 0. Let S be a framed algebra. Define the bilinear form (−, −): S ⊗ S → C by (a, b)1 = s(λ) (−1) a ·0 b for any a ∈ S λ and b ∈ S λ0 , where we use (FA2). Proposition 4.8. The bilinear form (−, −): S ⊗ S → C satisfies the following conditions: i (l,r) for any ai ∈ S λi and λ ∈ IS (i = 1, 2, 3), 1 2 (1)S λ1 and S λ2 are orthogonal to each other for λ , λ ; (2) The bilinear form is symmetric; 48

s(λ1) (3) The bilinear form is invariant, that is, (a1 · a2, a3) = (−1) (a2, a1 · a3). (4) The bilinear form is non-degenerate if and only if S is simple.

s(λ1) Proof. (1) clearly follows from Lemma 4.6. By Lemma 4.4, (a1, a2)1 = (−1) a1 ·0 a2 = s(λ2) (−1) a2 ·0 a1 = (a2, a1)1. Since

(a1 · a2, a3)1 = (a1 ·λ3 a2, a3)1 = (a3, a1 ·λ3 a2)1 s(λ1)+s(λ2)+s(λ3) = (−1) (a3, a2 ·λ3 a1)1 s(λ1)+s(λ2) = (−1) a3 ·0 (a2 ·λ3 a1) s(λ1)+s(λ2) = (−1) B0,λ2,λ3,λ1 a2 ·0 (a3 ·λ2 a1) s(λ3)+s(λ2) = (−1) (a2, a3 ·λ2 a1) s(λ1) = (−1) (a2, a1 · a3),

(3) holds. Let R(−,−) be the radical of the bilinear form, i.e.,

∗ R(−,−) = ker(S → S , a 7→ (a, −)).

s(λ) Let m ∈ R(−,−) and a ∈ S λ. Since (am, −) = (−1) (m, a−) = 0, we have am ∈ R(−,−). By (l,r) (1), R(−,−) is an IS -graded subspace of S . Hence, R(−,−) is an ideal of S and 1 < R(−,−).

Thus, R(−,−) = 0 if S is simple. Conversely, we assume that R(−,−) = 0 and M ⊂ S is an (l,r) ideal with 1 < M. Since M is an IS -graded subspace, M ∩ S 0 = 0. Thus, for any a ∈ S and m ∈ M, a ·0 m = 0. Hence, M ⊂ R(−,−) = 0, which implies that S is simple. Hereafter, we assume that a framed algebra is simple.

4.3. Code subalgebra. Let S be a simple (l, r)-framed algebra. In this section, we in- l+r troduce some subgroups of Z2 , which plays an important role in the study of framed algebras. Such subgroups are called codes and were originally introduce in the study of framed vertex operator algebras [DGH]. Recall that the set IS(l,r) has the following decomposition:

(l,r) a d IS = {d} × Z2. ∈ r+s d Z2 Accordingly, the framed algebra S also has the following decomposition: M S = AS (d) ∈ l+r d Z2 M AS (d) = S d,c. ∈ l+r c Z2 Set

l+r CS = {α ∈ Z2 | S 0,α , 0}, l+r DS = {d ∈ Z2 | AS (d) , 0}, l+r 2 IS = {(d, c) ∈ (Z2 ) | S d,c , 0}. 49

Then, by Lemma 3.18, M 0 0 0 0 0 (4.3) S d,c · S d ,c ⊂ S d+d ,(d+d )⊥(c+c )+γ. ∈ dd0 γ Z2 0 0 0 In particular, AS (d) · AS (d ) ⊂ AS (d + d ) for any d, d ∈ DS . Thus, by Lemma 4.7, DS l+r is a subgroup of Z2 and S is a DS -graded algebra. Similarly, since S 0,α · S 0,β ⊂ S 0,α+β l+r for any α, β ∈ CS , CS is a subgroup of Z2 . Denote S 0,α simply by S α for α ∈ CS . Then, L 0 AS (0) = S α. For v ∈ S d,c and v ∈ S d0,c0 , by (4.3), α∈CS 0 X 0 0 0 0 v · v = v ·(d+d ,(d+d )⊥(c+c )+γ) v . ∈ dd0 γ Z2 0 0 0 0 0 0 0 We sometimes denote v ·(d+d ,(d+d )⊥(c+c )+γ) v by v ·(d+d )⊥(c+c )+γ v for short.

Remark 4.9. Let α ∈ CS , (d, c) ∈ IS and a ∈ S α, v ∈ S d,c. It is noteworthy that a · v consists of only one component, that is, X a · v = a ·d⊥α+c+γ v = a ·d⊥c+α v. ∈ 0 γ Z2 l+r Proposition 4.10. The subsets CS , DS ⊂ Z2 satisfy the following conditions: l+r (1)C S and DS are subgroups of Z2 ; (2) |α| ∈ 2Z and |d| ∈ 8Z for any α ∈ CS and d ∈ DS ;

(3) |αd| ∈ 2Z for any α ∈ CS and d ∈ DS . ∈ 1 | | 1 | | ∈ Proof. (1) was already shown. If (d, c) IS , then by (FA1) s(d, c) = 16 d + 2 c Z. Thus, (2) follows. If (0, α), (d, c) ∈ IS , then by Lemma 3.18 and Lemma 4.7, (d, d⊥α + c) ∈ IS . 1 | | 1 | | ∈ ∈ | | − | | ∈ Thus, 16 d + 2 d⊥α + c Z. Combining this with s(d, c) Z, we have d⊥α + c c 2Z. l+r Since |d⊥α + c| − |c| = |d⊥α| + 2|d⊥cα|, we have |d⊥α| ∈ 2Z. Since |dα| = |1 α| − |d⊥α| =

|α| − |d⊥λ| and |α| ∈ 2Z, |dα| ∈ 2Z.

⊥ Thus, CS and DS are linear codes and DS is a subcode of the dual code CS = {β ∈ l+r Z2 | (β, CS ) ⊂ 2Z}. The purpose of this section is to prove that AS (0) is isomorphic to a twisted group (l,r) algebra C[CˆS ] and AS (d) is an irreducible IS -graded C[CˆS ]-module. The following lemma is very important to study a framed algebra:

Lemma 4.11. Let α1, α2 ∈ CS and (d, c) ∈ IS and a1 ∈ S α1 , a2 ∈ S α2 and v ∈ S d,c. Then, |α1α2| a2 · (a1 · v) = (−1) a1 · (a2 · v) and a1 · (a2 · v) = (a1 · a2) · v. In particular, AS (0) is an associative algebra and S is an AS (0)-module.

Proof. By (FA4), Remark 4.9 and Theorem 3.21, a2·(a1·v) = a2·d,d⊥(α1+α2)+c(a1·d,d⊥α1+cv) = |α1α2| (−1) a1 · (a2 · v) holds. Applying Lemma 4.4, we have (4.4)

s(d,d⊥(α1+α2)+c)−s(d,d⊥α2+c)+s(α2)+s(α1+α2) a1 · (a2 · v) = (−1) B(d,d⊥(α1+α2)+c),(d,c),α1,α2 (a1 · a2) · v 1 |d (α +α )+c|−|d α +c|+|α |+|α +α | 1 2 ⊥ 1 2 ⊥ 2 2 1 2 |α1c| |dα1| |dα1(α2+d⊥(α1+α2)+c)| = (−1) (−1) (−1) 2 (−1) (a1 · a2) · v. 50 1 | |−| | 1 2 d⊥(α1+α2)+c d⊥α2+c |d α | |d α (d α +c)| By using Remark 3.17, we have (−1) = (−1) 2 ⊥ 1 (−1) ⊥ 1 ⊥ 2 and 1 |α +α |+|α | 1 2 1 2 2 |α | |α α | |dα (α +d⊥(α +α )+c)| (−1) = (−1) 2 1 (−1) 1 2 . Since dd⊥ = 0 and dc = 0, we have (−1) 1 2 1 2 = 1 |dα1α2| (|α1|+|d⊥α1|+|dα1|) |α1α2|+|d⊥α1(d⊥α2+c)|+|α1c|+|dα1α2| (−1) . Hence, the RHS of (4.4) is equal to (−1) 2 (−1) (a1· 1 (|α1|+|d⊥α1|+|dα1|) |α1| a2) · v. Since |d⊥α1| + |dα1| = |α1|, by Proposition 4.10 (−1) 2 = (−1) = 1. |d⊥α (d⊥α +c)| |α c|+|d⊥α α | |α α |+|d⊥α (d⊥α +c)|+|α c|+|dα α | Since d⊥c = c,(−1) 1 2 = (−1) 1 1 2 . Thus, (−1) 1 2 1 2 1 1 2 = (−1)2|α1α2|+2|α1c| = 1. Hence, the assertion holds.

Lemma 4.12. For any α ∈ CS , dim S α = 1.

Proof. Let 0 , a ∈ S α for α ∈ CS . By Lemma 4.7, 0 , a · a ∈ S 2α = S 0. By (FA2), we 0 0 0 0 may assume that a·a = 1. Then, for any a ∈ S α, by Lemma 4.11, a = a ·1 = a ·(a·a) = |α| 0 (−1) a · (a · a) ∈ Ca. Thus, dim S λ = 1. L Thus, AS (0) = S α is isomorphic to the group algebra C[CS ] as a vector space. α∈CS Let ε : CS → Z2 be a bilinear map satisfying |α| (1) ε(α, α) = (−1) 2 ; (2) ε(α, β)ε(β, α) = (−1)|αβ|. 2 × By [FLM], such a bilinear map exists and gives an element in Z (CS , C ), the group × cohomology with the coeﬃcient in C (see for example [Mo1]). Let C[CˆS ] be the vector ˆ space with the basis {eα}α∈CS and deﬁne a product on C[CS ] by eαeβ = ε(α, β)eα+β. Since ε(−, −) is a two-cocycle, the product is associative. Since C× is an injective object in the category of abelian groups, similarly to the proof of Proposition in [Mo1], we have:

2 × |αβ| Proposition 4.13. If a two-cocycle µ(−, −) ∈ Z (CS , C ) satisﬁes µ(α, β)µ(β, α) = (−1) |α| × and µ(α, α) = (−1) 2 for any α, β ∈ CS . Then, there exists a map f : CS → C such −1 −1 that µ(α, β) f (α + β) f (α) f (β) = ε(α, β) for any α, β ∈ CS , that is, µ(−, −) = ε(−, −) ∈ 2 × H (CS , C ).

Then, by Lemma 4.7, Lemma 4.11 and the above proposition, we have:

Proposition 4.14. There exists a C-algebra isomorphism between AS (0) and C[CˆS ] which preserves the IS(l,r)-grading.

|α| −1 s(α) Remark 4.15. For α ∈ CS , since eα · eα = (α, α)e0 = (−1) e0, we have eα = (−1) eα. Thus, by Proposition 4.8, the invariant bilinear form on a framed algebra S is C[CˆS ]- invariant, i.e., (eα · v1, eα · v2) = (v1, v2) for any α ∈ CS and v1, v2 ∈ S.

Hereafter, we identify AS (0) as C[CˆS ] and take a basis eα ∈ S α for α ∈ CS such that eα · eβ = ε(α, β)eα+β. By Lemma 4.11, AS (d) is an AS (0)-module.

(l,r) Proposition 4.16. Let d ∈ DS . Suppose that N ⊂ AS (d) is an IS -graded subspace such that a · n ∈ N for any a ∈ AS (0) and n ∈ N. Then, N = 0 or N = AS (d), that is, AS (d) is (l,r) an irreducible IS -graded AS (0)-module. 51

Proof. Let N ⊂ AS (d) be a non-zero sub-AS -module. Then, by Lemma 4.4, SN ≡ {a · n}a∈S,n∈N is an ideal of the framed algebra S . Since S is simple, SN = S . Thus, AS (d) =

AS (d) ∩ SN = {a · n}a∈AS (0),n∈N = N.

We constructed even codes (DS , CS ) from the framed algebra S , which we call the l+r structure codes of the framed algebra. Conversely, let C ⊂ Z2 be an even code, that is, l+r C is a subgroup of Z2 and |α| ∈ 2Z for any α ∈ C. Then, we have: Proposition 4.17. The twisted group algebra C[Cˆ] is a simple framed algebra whose codes are (0, C).

Proof. Since C is even, (FA1) holds. (FA2) and (FA3) follow from the deﬁnition of C[Cˆ]

|α1α2| with 1 = e0. Let α1, α2, α3 ∈ C. Then, by deﬁnition eα1 ·(eα2 ·eα3 ) = (−1) eα2 ·(eα1 ·eα3 ), which implies (FA4). The simplicity is obvious. The above proposition gives trivial examples of framed algebras. General framed algebras can be obtained by module extensions of AS (0) = C[Cˆ]. We end this section by introducing a three-dimensional framed algebra which is not a twisted group algebra and one of the simplest non-trivial framed algebra. (1,1) Let S Ising = C1 ⊕ Ca ⊕ Cd be a three-dimensional vector space with the IS -grading,

S 0 = C1, S 1 1 = Ca and S 1 1 = Cd. We deﬁne a product on S Ising by a · a = 1, 2 , 2 16 , 16 a · d = d · a = 1, and d · d = 1 + a with unit 1.

It is not hard to show that S Ising is a simple framed algebra. This fact will be shown in a general setting in Section 5. So we only check (FA4) for (a1, a2, a3) = (d, d, d) and 0 1 1 0 1 1 λ = ( 2 , 2 ) and λ = ( 16 , 16 ) here. Since

d · 1 1 (d · 1 1 d) = d · a = d 16 , 16 2 , 2 and X 1 1 λ,( 2 , 2 ) B 1 1 1 1 1 1 1 1 d · 1 , 1 (d ·λ d) ( 16 , 16 ),( 16 , 16 ),( 16 , 16 ),( 16 , 16 ) 16 16 1 1 1 1 1 1 1 1 λ∈A(( 16 , 16 ),( 16 , 16 ),( 16 , 16 ),( 16 , 16 )) X 1 1 = d · 1 1 (d ·λ d) = (d · 1 + d · a) = d, 16 , 16 1 1 2 2 λ=(0,0),( 16 , 16 ) (FA5) holds. ⊂ 1+1 ⊂ 1+1 The codes of S Ising are CS Ising =< (1, 1) > Z2 and DS Ising =< (1, 1) > Z2 . The corresponding conformal ﬁeld theory is the critical 2d Ising model.

Note that it is exceptional that S Ising becomes an associative algebra.

4.4. Classiﬁcation of C[CˆS ]-modules. By Proposition 4.16, AS (d) is an irreducible module of AS (0). In this section, we further study the structure of AS (d). The results of this section is essentially obtained in the chiral setting [Mi2] (see also [LY, Section 4.2]). We rewrite them in terms of a framed algebra here. L We ﬁrst recall that AS (d) = S d,c. Let (d, c) ∈ IS and consider the stabilizer (d,c)∈IS d subgroup CS = {α ∈ CS | eα · S d,c ⊂ S d,c}. Then, by the fusion rule and Lemma 4.7, 52

d d it is clear that CS = {α ∈ CS | d⊥α = 0} or CS consists of vectors whose support is in d ˆd L d. Since C is a subgroup of CS , C[C ] = d eα is a subalgebra of C[CˆS ], where S S α∈CS d the two-cocycle of CS is given by the restriction of ε(−, −). The following lemma can be proved similarly to the proof of Proposition 4.16:

d Lemma 4.18. Let (d, c) ∈ IS . If N is a subspace S d,c such that eα · N ⊂ N for any α ∈ CS , ˆd ˆd then N = 0 or N = C[CS ], that is, S d,c is a simple C[CS ]-module.

By the universality of the tensor product, we have a C[CˆS ]-module homomorphism,

φd,c : C[CˆS ] ⊗ ˆd S d,c → AS (d). The map φd,c is surjective since AS (d) is simple. We C[CS ] claim that φd,c is isomorphism. By the surjectivity, it suﬃces to show that dim C[CˆS ]⊗ ˆd C[CS ] ˆ d S d,c = dim AS (d). By the construction, dim C[CS ] ⊗ ˆd S d,c = [CS : CS ] dim S d,c, C[CS ] d where [CS : CS ] is the index of groups. Since for any α ∈ CS , eα· : S d,c → S d,c+d⊥α is isomorphism, dim S d,c+d⊥α is equal to dim S d,c. Thus, we have:

Lemma 4.19. The map φd,c : C[CˆS ] ⊗ ˆd S d,c → AS (d) is a C[CˆS ]-module isomorphism C[CS ] and preserves the IS(l,r)-grading.

ˆd l+r Finally, we study the structure of a C[CS ]-module S d,c. A subgroup H ⊂ Z2 is said d to be isotropic if |αβ| ∈ 2Z for any α, β ∈ H. Let Hd ⊂ CS be a maximal isotropic |αβ| subspace. Since (−1) = 1 for any α, β ∈ Hd, eα and eβ commute with each other in L C[CˆS ]. Thus, Ceα ⊂ C[CˆS ] is isomorphic to the (untwisted) group algebra C[Hd]. α∈Hd We ﬁx an isomorphism C[Hd] → C[CˆS ] and identify it as a subalgebra of C[CˆS ]. Since

Hd is a finite group, S d,c is decomposed into the sum of one-dimensional modules of × Hd. Let χ : Hd → C be a character and let e(d, c, χ) ∈ S d,c satisfy e(d, c, χ) , 0 and eα · e(d, c, χ) = χ(α)e(d, c, χ) for any α ∈ Hd, an eigenvector. Then, by the universality ˆd ˆd of the tensor product again, ψ : C[CS ] ⊗C[Hd] Cχ → S d,c is a (surjective) C[CS ]-module ˆd homomorphism. By the representation theory of extraspecial 2-groups, C[CS ] ⊗C[Hd] Cχ ˆd is an irreducible representation of C[CS ] (see for example [FLM]). Thus, the map is an ˆd isomorphism of C[CS ]-modules. d |α−| We observe that for α ∈ CS , eα·e(d, c, χ) is an eigenvector with the eigenvalue (−1) χ : × |α−| × |αβ| Hd → C , where (−1) : Hd → C is the character defined by β 7→ (−1) . ˆd We summarize the above argument as follows: Let Irr C[CS ] be the set of irreducible ˆd ∨ × ˆd C[CS ]-modules and Hd : Hom(Hd, C ) the set of characters. For M ∈ Irr C[CS ], there ∨ L exists χ ∈ H such that M = d M(−1)|α−|χ. Thus, the correspondence η : M 7→ d α∈CS /Hd ∨ d χM ∈ Hd /CS defines a map. Hence, we have:

ˆd ∨ d Proposition 4.20. The map η : Irr C[CS ] → Hd /CS gives a bijection.

4.5. Products among C[CˆS ]-modules. In this section, we derive an important formula 0 0 for the product · : AS (d) × AS (d ) → AS (d + d ). 53

1 1 2 2 Lemma 4.21. Let a ∈ S α and v1 ∈ S d1,c1 and v2 ∈ S d2,c2 for α ∈ CS and (d , c ), (d , c ) ∈ d1d2 IS . Then, for γ ∈ Z2 , 1 |d1α| |αc1|+|d1d2α|+|γα|+|d1αc2| 1 2 1 2 2 1 2 1 2 a · (v1 ·(d +d )⊥(c +c )+γ v2) = (−1) (−1) v1 ·(d +d )⊥(c +c +α)+γ (a · v2). 1 1 2 |d⊥α|+|d αc | a · (v1 · v2) = (−1) (a · v1) · v2.

d1d2 Proof. By Theorem 3.21, for γ ∈ Z2

1 2 1 2 1 2 1 2 1 2 1 2 a · (v1 ·(d +d )⊥(c +c )+γ v2) = a ·(d +d )⊥(α+c +c )+γ (v1 ·(d +d )⊥(c +c )+γ v2)

1 1 1 1 2 1 2 1 2 |αc | |d α| |d α(c +γ+(d +d )⊥(c +c +α))| 2 1 2 1 2 = (−1) (−1) (−1) v1 ·(d +d )⊥(c +c +α)+γ (a · v2) |αc1| 1 |d1α| |d1α(c2+γ+d2α)| 2 1 2 1 2 = (−1) (−1) (−1) v1 ·(d +d )⊥(c +c +α)+γ (a · v2) 1 |d1α| |αc1|+|d1d2α|+|γα|+|d1αc2| 2 1 2 1 2 = (−1) (−1) v1 ·(d +d )⊥(c +c +α)+γ (a · v2).

d1d2 By Lemma 4.4, for γ ∈ Z2 ,

1 2 1 2 1 2 1 2 a ·(d +d )⊥(c +c +α)+γ (v1 ·(d +d )⊥(c +c )+γ v2) 1 2 1 2 1 2 1 2 1 2 1 2 1 1 1 1 1 = (−1)s(d +d ,(d +d )⊥(c +c +α)+γ)−s(d +d ,(d +d )⊥(c +c )+γ)+s(d ,d⊥α+c )−s(d ,c )

1 2 1 2 1 2 2 2 1 1 1 2 1 2 B(d +d ,(d +d )⊥(c +c +α)+γ),(d ,c ),α,(d ,c )(a · v1) ·(d +d )⊥(c +c +α)+γ v2 1 |(d1+d2) (c1+c2+α)+γ|−|(d1+d2) (c1+c2)+γ|+|d1 α+c1|−|c1| = (−1) 2 ⊥ ⊥ ⊥

|c2α| 1 |d2α| |d2α(c1+d1α+γ)| 2 1 2 1 2 (−1) (−1) (−1) (a · v1) ·(d +d )⊥(c +c +α)+γ v2 1 2 1 1 2 1 2 1 1 2 2 1 1 |d d α|+|d α|+|α((d +d )⊥(c +c )+γ)|+|d αc | |c α|+|d α(c +d α+γ)| ⊥ ⊥ ⊥ 1 2 1 2 = (−1) (−1) (a · v1) ·(d +d )⊥(c +c +α)+γ v2 |d1 α|+|d1αc2| ⊥ 1 2 1 2 = (−1) (a · v1) ·(d +d )⊥(c +c +α)+γ v2. By the above lemma, we have:

1 2 1 2 Corollary 4.22. Let vd1 ∈ S (d1,0) and vd2 ∈ S (d2,0) for d , d ∈ DS . Then, for any α , α ∈

CS ,

X 1 1 2 1 2 1 2 2 2 1 1 2 |d⊥α|+|d d⊥α α |+|d d α |+|γα |+ |d α | (eα1 · vd1 ) · (eα2 · vd2 ) = (−1) 2 eα1 · (eα2 · (vd1 ·(d1+d2,γ) vd2 )). ∈ d1d2 γ Z2

1 2 1 2 Hence, if we know the product vd1 ·vd2 ∈ AS (d +d ), then the product on AS (d )⊗AS (d ) are uniquely determined by the AS (0)-module structure.

4.6. Code vertex operator algebra. In this section, we consider the case of r = 0, a l chiral conformal ﬁeld theory. If α, β ∈ Z2 satisfy |α|, |β| ∈ 2Z, then |α+β| = |α|+|β|−2|αβ| ∈ l l 2Z. Thus, all even codewords in Z2 forms a (maximal even) subgroup of Z2. We denote it even by C and the corresponding full vertex algebra constructed in Theorem 4.3 by V even . l Cl Since V even consists of only holomorphic ﬁelds, by Proposition 2.6, V even is a vertex Cl Cl operator algebra. Recall that the degree one space of a vertex operator algebra inherits a even even Lie algebra structure. Set (C )[2] = {α ∈ C | |α| = 2}. Then, {e } ∈ even forms a l l α α (Cl )2 54

even l basis of (V even ) . Hence, dim(V even ) = #(C )[2] = . By the construction of V even , Cl 1 Cl 1 l 2 Cl we have:

Lemma 4.23. The 0 and 1-th products on (V even ) are given by: Cl 1   ∈ even ε(α, β)eα+β if α + β (Cl )2, eα(0)eβ =  0 otherwise and   (−1)1 if α = β, eα(1)eβ =  0 otherwise

even for α, β ∈ (Cl )2.

By the above lemma, it is not hard to show that (V even ) is isomorphic to the orthogonal Cl 1 Lie algebra so(l) as Lie algebras. We will examine the structure of vertex algebras V even Cl in more detail. even In the case of l = 1, since C = 0, V even is nothing but the Virasoro vertex operator 1 C1 1 algebra L( , 0). In the case of l = 2, it is well-known that V even is isomorphic to the lattice 2 C2 vertex operator algebra V2Z associated with the rank one lattice Zα with (α, α) = 4 (see [DGH]). even In general for k ≥ 1, since C2k contains mutually orthogonal k vectors (1100 ··· ), (0011 ··· ), (00 ··· 11), V even is an extension of the lattice vertex operator algebra V k and C2k (2Z) thus a lattice vertex operator algebra. Set D1 = 2Z and for k ≥ 2 let Dk be the lattice generated by (±1, ±1, 0, 0, ··· ), (0, ±1, ±1, 0, ··· ), (0, 0, ±1, ±1, ··· ),..., (0, 0, 0,..., ±1, ±1) in k R . If k ≥ 4, then D is the root lattice of type D . It is easy to show that V even is k k C2k isomorphic to the lattice vertex operator algebra VDk for any k ≥ 1.

Let us consider the odd cases starting from l = 3. Then, {e110, e011, e101} forms the basis of (V even ) and {2ie , −e − ie , e − ie } forms a sl -triple {h, e, f } for some C3 1 110 011 101 011 101 2 two-cocycle. By Lemma 4.23, h(1)h = 41. Thus, V even contains a subalgebra isomorphic C3 to VA1,2 , the simple aﬃne vertex operator algebra of level 2 associated with sl2. Since 1 ⊗3 even L( , 0) ⊂ V and (110), (011), (101) generate C , we have V even V . 2 A1,2 3 C3 A1,2 Similarly, for general odd integers l = 2k + 1 with k ≥ 2, V even is isomorphic to V , C2k+1 Bk,1 the simple aﬃne vertex operator algebra of level 1 type Bk. Hence, we have:

Proposition 4.24. The vertex operator algebras {V even } are isomorphic to: Cl l=1,2,... 1 1 (1)V even L( , 0), the simple Virasoro VOA of central charge ; C1 2 2 (2)V even V , the lattice VOA for any k ≥ 1; C2k Dk (3)V even V , the simple aﬃne VOA of type A at level 2; C3 A1,2 1 (4)V even V , the simple aﬃne VOA of type B at level 1 for any k ≥ 2. C2k+1 Bk,1 k 55

5. Code conformal field theory In this section, we construct modular invariant conformal ﬁeld theories from codes. Hereafter, we assume that l = r. In this case, there is a group homomorphism

r r+r ∆ : Z2 → Z2 , g 7→ (g, g).

r r r+r Let G ⊂ Z2 be a subgroup satisfying (1 ) ∈ G and DG = ∆G ⊂ Z2 the image of G under ∆ and set ⊥ r+r CG = (∆G) = {α ∈ Z2 | |αd| ∈ 2Z for any d ∈ DG}.

The purpose of this section is to construct an (r, r)-framed algebra with the codes (DG, CG). 0 r For any γ, γ ∈ ∆Z2, since

0 0 0 (5.1) |γγ | = |γγ |l − |γγ |r = 0, the image of ∆ is isotropic. This property is very useful in the construction. 2 In Section 5.1, we explicitly construct a two-cocycle ε(−, −) ∈ H (CG, Z2) and representations AG(d) of C[CˆG] for each d ∈ DG. In Section 5.2, we construct the product on L S G = AG(d) and prove that S G is a simple (r, r)-framed algebra. In Section 5.3, d∈DG the character of the framed full vertex operator algebra FG associated with S G is shown to satisfy the modular invariance. In Section 5.4, the codes G are classiﬁed up to r ≤ 6.

2 5.1. Two-cocycle and representations. We first construct a two-cocycle ε(−, −) ∈ H (CG, Z2). even r+r r+r ⊥ r+r Set Cr,r = {α ∈ Z2 | |α| ∈ 2Z} = (1 ) , which is the maximal even code in Z2 . Since r r+r r ⊥ even (1 ) ∈ G, 1 = ∆(1 ) ∈ DG and CG = DG ⊂ Cr,r . We will construct a two-cocycle on even Cr,r and define a two-cocycle on CG by restriction. r Let ei ∈ Z2 be the vector whose i-th component is 1 and all the other components are 0 r+r and sete ˜i = (ei, 0) ∈ Z2 . Then, it is clear that {∆e1,..., ∆er, e˜1 −e˜2, e˜2 −e˜3,..., e˜r−1 −e˜r} is even a basis of Cr,r . We consider it as an ordered basis (v1, v2,..., v2r−1) = (∆e1,..., ∆er, e˜1 − even even e˜2, e˜2 − e˜3,..., e˜r−1 − e˜r). Let ε(−, −): Cr,r × Cr,r → Z2 be the bilinear map defined by

 1 |v | (−1) 2 i i = j  ε(v , v ) =  i j 1 i < j  (−1)|viv j| i > j.

(α,β) even Then, ε(−, −) satisﬁes ε(α, β)ε(β, α) = (−1) for any α, β ∈ Cr,r . The following lemma is very important:

0 r 0 0 Lemma 5.1. For any g, g ∈ Z2, |∆g∆g | = 0 and ε(∆g, ∆g ) = 1

r Proof. The first equation follows from (5.1). Since (∆e1,..., ∆er) forms a basis of ∆Z2 ⊂ even Cr,r , the second equation follows from the definition of the bilinear form ε(−, −). L Define the twisted group algebra C[CˆG] = Ceα by the restriction of the two- α∈CG cocycle ε(−, −), i.e., eα · eβ = ε(α, β)eα+β for α, β ∈ CG. 56

r r+r d r+r Let d ∈ ∆Z2. Recall that we set d⊥ = 1 + d and Z2 = {α ∈ Z2 | d⊥α = 0}, the set of codewords whose support is in d. Then, set

d d CG = Z2 ∩ CG d d r ∆ = Z2 ∩ ∆Z2.

In Section 4.4 we show that a framed algebra with the code (CG, DG) is a direct sum of ˆd ˆd irreducible C[CG]-modules and each irreducible C[CG]-module can be constructed from d a one-dimensional representation of a maximal isotropic subspace in CG (see Proposition 4.20). By the virtue of Lemma 5.1, we have:

d d Lemma 5.2. For any d ∈ DG, ∆ is a maximal isotropic subspace of CG.

Proof. Recall that by (5.1), |∆1r+r ∆1r+r | ∈ 2Z, i.e., ∆1r+r is isotropic. Thus, ∆d ⊂ ∆1r+r is 1r+r d d isotropic. By DG ⊂ ∆ , |∆ DG| = 0, which implies ∆ ⊂ CG. d d d It suﬃces to show that ∆ is maximal. Let α ∈ CG satisfy |∆ α| = 0. We assume that r α < ∆Z2. Then, the left and right i-th component of α is diﬀerent for some i. Then, d r d d |α∆ei| = ±1 and ∆ei ∈ ∆ , a contradiction. Thus, α ∈ ∆Z2 ∩ CG. Hence, α ∈ ∆ , which implies that ∆d is maximal.

0 1r+r 0 d For any γ, γ ∈ ∆ , by Lemma 5.1, eγeγ0 = ε(γ, γ )eγ+γ0 = eγ+γ0 . Thus, C[∆ ], the d 1r+r group algebra of ∆ , is canonically a subalgebra of C[CˆG] for any d ∈ ∆ . Let d ∈ DG d and Ctd be the trivial representation of ∆ . Set

AG(d) = C[CˆG] ⊗C[∆d] Ctd, the induced module and M S G = AG(d).

d∈DG

(r,r) We denote the action of C[CˆG] on S G by · and deﬁne an IS -grading on S G by eα · td ∈

(S G)d,d⊥α for any α ∈ CG.

We end this section by determining the dimension of AG(d). Set

r+0 r+r r Z2 = {(α, 0) ∈ Z2 | α ∈ Z2} left r+0 CG = Z2 ∩ CG.

r+r r+r left 1 1 d d⊥ Then, we have CG = CG ⊕ ∆ and ∆ = ∆ ⊕ ∆ for any d ∈ DG. Hence, {eα · left d⊥ d⊥ r−|d|l td} left d⊥ is a basis of AG(d). Thus, dimC AG(d) = #(C ⊕ ∆ ). Since #∆ = 2 α∈CG ⊕∆ G r left #Z2 r−dim G 2r−dim G and #CG = #G = 2 and #CG = 2 , we have 1 2r−dim G−|d|l (5.2) dimC AG(d) = 2 = #CG. 2|d|l 57 L 5.2. Construction of product. In this section, we will deﬁne a product on S G = AG(d). d∈DG For each d ∈ ∆1r+r , set X δd = eα ∈ C[CˆG]. α∈∆d Since by Lemma 5.1 the two-cocycle ε(−, −) is trivial on ∆1r+r , we have:

1r+r d Lemma 5.3. Let d ∈ ∆ and γ ∈ ∆ . Then, δd · eγ = eγ · δd = δd ∈ C[CˆG].

1 2 1 2 For d , d ∈ DG, deﬁne a map md1,d2 : C[CˆG] × C[CˆG] → AG(d + d ) by

1 2 1 2 1 2 2 1 1 2 |d d⊥α α |+|d d α |+ |d α | md1,d2 (eα1 , eα2 ) = (−1) 2 eα1 · δd1d2 · eα2 · td1+d2

1 2 for α , α ∈ CG.

1 d1 2 d2 Lemma 5.4. For any γ ∈ ∆ , γ ∈ ∆ and α, β ∈ CG,

md1,d2 (eαeγ1 , eβeγ2 ) = md1,d2 (eα, eβ).

2 1 2 1 2 1 1 2 1 Proof. Set ζ1 = d⊥γ , ζ12 = d γ and ζ2 = d⊥γ , ζ21 = d γ . Then, γ = ζ1 + ζ12, 2 d1d2 d1+d2 γ = ζ2 + ζ21. Since ζ12, ζ21 ∈ ∆ and ζ1, ζ2 ∈ ∆ , by Lemma 5.3 and deﬁnition of 1 2 1 2 1 2 1 2 1 2 1 2 1 2 AG(d + d ), we have eζ21 · δd d = δd d = δd d · eζ12 and eζ1 · td +d = td +d = eζ2 · td +d . By

1 2 Lemma 5.1, eγ = eζ1 · eζ12 and eγ = eζ2 · eζ21 , we have

1 2 1 2 1 2 md ,d (eαeγ , eβeγ ) = md ,d (eαeζ1 eζ12 , eβeζ2 eζ21 )

1 2 1 2 1 2 2 1 1 2 |d d⊥(α+γ )(β+γ )|+|d d (β+γ )|+ |d (β+γ )| = (−1) 2 (eα · eγ1 ) · δd1d2 · (eβ · eγ2 ) · td1+d2

1 2 1 2 1 1 |d d (α+ζ1)β|+|d d (β+ζ21)|+ |d (β+ζ21)| ⊥ 2 1 2 1 2 = (−1) eα · eζ1 · δd d · eβ · eζ21 · td +d . |βζ | |ζ β| 21 1 2 1 1 2 By deﬁnition of ε(−, −), we have eβ ·eζ21 = (−1) eζ21 ·eβ and eζ1 ·δd d ·eβ = (−1) δd d · eβ · eζ1 . Thus, 1 2 1 2 1 1 |d d⊥(α+ζ1)β|+|d d (β+ζ21)|+ |d (β+ζ21)| |βζ21|+|βζ1| RHS = (−1) 2 (−1) eα · δd1d2 · eβ · td1+d2

1 2 1 2 1 1 |d d⊥αβ|+|ζ1β|+|d d β|+|ζ21|+ (|d β|+|ζ21|−2|βζ21|) |βζ21|+|βζ1| = (−1) 2 (−1) eα · δd1d2 · eβ · td1+d2

1 2 1 2 1 1 |d d⊥αβ|+|d d β|+ |d β| = (−1) 2 eα · δd1d2 · eβ · td1+d2

= md1,d2 (eα, eβ),

1r+r where we used |ζ| = |ζ|l − |ζ|r = 0 for any ζ ∈ ∆ .

1 2 By the above lemma, md1,d2 gives the well-deﬁned map md1,d2 : AG(d ) × AG(d ) → 1 2 1 2 AG(d + d ). Combining md1,d2 for all d , d ∈ DG, we have a product · : S G ⊗ S G → S G.

We note that the action of C[CˆG] on AG(d) and m0,d : C[CˆG] ⊗ AG(d) → AG(d) coincide with each other. In fact, for any α, β ∈ CG and d ∈ DG,

m0,d(eα, eβ · td) = eα · δ0 · eβ · td

= eα · (eβ · td). Thus, we can use the same notation ·.

We will show that S G is an (r, r)-framed algebra. Recall that S G is spanned by {eα · td ∈ } ∈ ∈ 1 | | 1 | | 1 | | (S G)d,d⊥α α∈CG,d∈DG . For any d DG and α CG, s(d, d⊥α) = 16 d + 2 d⊥α = 2 d⊥α . 58

r+r ∈ r+r ∈ ⊥ 1 | | ∈ Since 1 DG, d⊥ = d + 1 DG. Thus, by CG = DG, 2 d⊥α Z, which implies (FA1).

For any d ∈ DG and α ∈ CG,

(eα · td) · e0 = md,0(eα · td, e0)

= eα · δ0 · e0 · td

= eα · td, which implies (FA2).

1 2 1 2 Lemma 5.5. For d , d ∈ DG and α , α ∈ CG, M 1 1 2 2 1 2 1 2 (eα · td ) · (eα · td ) ∈ (S G)(d +d ,(d1+d2)⊥(α +α )+γ). γ∈∆d1d2 P Proof. By deﬁnition, eα1 · δd1d2 · eα2 · td1+d2 = γ∈∆d1d2 eα1 · eγ · eα2 · td1+d2 . Since eα1 · eγ · 1 2 2 1 2 1 2 1 2 1 2 eα · td +d ∈ (S G)(d +d ),(d +d )⊥(α +α +γ). The assertion follows from (d + d )⊥γ = γ. 1r+r Hence, it suﬃces to show that (FA4). Since |γ|l = |γ|r for any γ ∈ ∆ , we have:

1 2 3 1 2 3 1 2 3 1 2 3 ∈ −πi | 1 2| 1+i |d d d |l 1−i |d d d |r 1 |d d d |l Lemma 5.6. For d , d , d DG, exp( 8 d d ) = 1 and ( 2 ) ( 2 ) = ( 2 ) . By the above lemma and Theorem 3.21, the connection matrix of an (r, r)-framed algebra with the code (∆G, (∆G)⊥) is simpler than general cases. 1 2 3 1 2 3 Let α , α , α ∈ CG, d , d , d ∈ DG. We will consider the products (eα1 · td1 ) · (eα2 · td2 ) · (eα3 · td3 ) and (eα2 · td2 ) · (eα1 · td1 ) · (eα3 · td3 ) and show (FA4). By Lemma 5.5, (eα1 · td1 ) · (eα2 · td2 ) · (eα3 · td3 ) is in M M 1 2 3 1 2 3 1 2 3 0 (5.3) (S G)(d +d +d ,(d +d +d )⊥(α +α +α +γ+γ ). γ0∈∆d1(d2+d3) γ∈∆d2d3 1 2 3 2 3 1 2 3 1 2 3 1 2 3 2 3 1 3 1 2 Since (d + d + d )⊥(d d ) = d⊥d d and (d + d + d )⊥d (d + d ) = d⊥d d + d⊥d d , (5.3) can be written as: M M M 1 2 3 1 2 3 1 2 3 1 2 3 (5.4) (S G)(d +d +d ,(d +d +d )⊥(α +α +α )+γ +γ +γ ). 1 2 3 2 3 1 3 1 2 γ1∈∆d⊥d d ) γ2∈∆d⊥d d ) γ3∈∆d⊥d d )

1 2 3 2 1 3 3 1 2 Let γ1 ∈ ∆d⊥d d , γ2 ∈ ∆d⊥d d , γ3 ∈ ∆d⊥d d . By (5.4), it suﬃces to consider the products 1 1 1 2 3 1 2 3 1 2 3 1 2 3 2 2 3 3 (eα · td ) ·d +d +d ,(d +d +d )⊥(α +α +α )+γ +γ +γ ) (eα · td ) · (eα · td ) and 2 2 1 2 3 1 2 3 1 2 3 1 2 3 1 1 3 3 (eα · td ) ·d +d +d ,(d +d +d )⊥(α +α +α )+γ +γ +γ ) (eα · td ) · (eα · td ) . By (5.3), each (γ, γ0) ∈ ∆d2d3 × ∆d1(d2+d3) which can contribute to 1 1 1 2 3 1 2 3 1 2 3 1 2 3 2 2 3 3 (eα · td ) ·d +d +d ,(d +d +d )⊥(α +α +α )+γ +γ +γ ) (eα · td ) · (eα · td ) must satisfy

1 1 γ = d⊥γ 2 2 0 γ = d⊥γ 3 3 0 γ = d⊥γ . 59

Since γ0 ∈ ∆d1(d2+d3), γ0 is uniquely determined by this equation, more speciﬁcally, γ0 = γ2 + γ3. However, there is an ambiguity in the choice of γ. In fact, γ must be of the form 1 s 1 s {γ + γ }γs∈∆d1d2d3 and is not uniquely determined since d⊥γ = 0. s d1d2d3 1 1 1 2 3 1 2 3 1 2 3 1 2 3 2 Let γ ∈ ∆ and we will calculate (eα · td ) ·d +d +d ,(d +d +d )⊥(α +α +α )+γ +γ +γ ) (eα · 2 2 3 2 3 2 3 s 1 3 3 td ) ·d +d ,(d +d )⊥(α +α )+γ +γ (eα · td ) . By the deﬁnition,

2 2 2 3 2 3 s 1 3 3 (eα · td ) ·(d +d ,(d +d )⊥(α2+α3)+γ +γ ) (eα · td ) 2 3 3 2 3 2 3 1 2 3 |d d α |+|d d⊥α α |+ |d α | = (−1) 2 eα2 · eγs+γ1 · eα3 · td2+d3 .

Thus, 1 1 1 2 3 1 2 3 1 2 3 2 2 2 3 s 1 3 3 (eα · td ) ·(d +d +d )⊥(α +α +α )+γ +γ +γ (eα · td ) ·(d +d )⊥(α2+α3)+γ +γ (eα · td ) |d2d3α3|+|d2d3 α2α3|+ 1 |d2α3| |d1(d2+d3)(α2+α3)|+|d1(d2+d3) α1(α2+α3+γs)|+ 1 |d1(α2+α3+γs)| = (−1) ⊥ 2 (−1) ⊥ 2

eα1 · eγ2+γ3 · eα2 · eγ1+γs · eα3 · td1+d2+d3 |d2d3α3|+|d2d3 α2α3|+ 1 |d2α3| |d1(d2+d3)(α2+α3)|+|d1(d2+d3) α1(α2+α3)|+ 1 |d1(α2+α3)| |γs(α1+α2+α3)| = (−1) ⊥ 2 (−1) ⊥ 2 (−1)

|α2(γ2+γ3)|+|γsα3| (−1) eα1 · eα2 · eγ1+γ2+γ3 · eα3 · td1+d2+d3 |d2d3α3|+|d2d3 α2α3|+ 1 |d2α3| |d1(d2+d3)(α2+α3)|+|d1(d2+d3) α1(α2+α3)|+ 1 |d1(α2+α3)| = (−1) ⊥ 2 (−1) ⊥ 2

|α2(γ2+γ3)|+|γs(α1+α2)| (−1) eα1 · eα2 · eγ1+γ2+γ3 · eα3 · td1+d2+d3 and similarly for γu ∈ ∆d1d2d3 2 2 1 2 3 1 2 3 1 2 3 1 1 1 3 1 3 u 2 3 3 (eα · td ) ·(d +d +d )⊥(α +α +α )+γ +γ +γ (eα · td ) ·(d +d ,(d +d )⊥(α1+α3)+γ +γ ) (eα · td ) |d1d3α3|+|d1d3 α1α3|+ 1 |d1α3| |d2(d1+d3)(α1+α3)|+|d2(d1+d3) α2(α1+α3)|+ 1 |d2(α1+α3)| = (−1) ⊥ 2 (−1) ⊥ 2

|α1(γ1+γ3)|+|γu(α1+α2)| |α1α2| (−1) (−1) eα1 · eα2 · eγ1+γ2+γ3 · eα3 · td1+d2+d3 .

1 2 3 We remark that ( 1 )|d d d |l = 1 . Thus, by Lemma 5.6 and Theorem 3.21, 2 #∆d1d2d3

2 3 2 3 2 3 1 s 1 3 1 3 1 3 2 s (d +d ,(d +d )⊥(α +α )+γ +γ ),(d +d ,(d +d )⊥(α +α )+γ +γ ) (5.5) B 1 2 3 1 2 3 1 2 3 1 2 3 1 1 1 2 2 2 3 3 3 (d +d +d ,(d +d +d )⊥(α +α +α )+γ +γ +γ ,(d ,d⊥α ),(d ,d⊥α ),(d ,d⊥α ))

1 1 2 1 2 2 1 1 1 2 2 1 2 3 1 2 3 = (−1)|d⊥d⊥α α |(−i)|d d⊥α |+|d d⊥α |i|d d d⊥(α +α )+γ | #∆d1d2d3 1 2 3 1 2 s u 1 2 2 2 1 1 3 3 1 2 3 1 2 3 1 2 3 (−i)|d d d (α +α )+(γ +γ )|(−1)|(d d⊥α +d d⊥α )(d⊥α +(d +d +d )⊥(α +α +α )+(γ +γ +γ ))|.

Since

2 1 1 1 2 2 1 2 3 1 2 3 1 2 3 1 2 s u (−i)|d d⊥α |+|d d⊥α |i|d d d⊥(α +α )+γ |(−i)|d d d (α +α )+(γ +γ )|

1 2 1 2 1 1 2 1 1 2 2 1 2 1 2 2 1 2 3 1 2 3 1 2 1 2 3 1 2 1 2 s u = (−i)|d d α |+|d α |−2|d d α |+|d d α |+|d α |−2|d d α |i|d d d⊥(α +α )|−2|γ (α +α )|(−i)|d d d (α +α )|−2|(α +α )(γ +γ )|

|d1d2(α1+α2)|+|(α1+α2)(γs+γu)|+|γ3(α1+α2)|+ 1 (|d1α2|+|d2α1|) |d1d2α1|+|d1d2α2| |d1d2d3 (α1+α2)| |d1d2d3(α1+α2)| = (−1) 2 (−i) i ⊥ (−i) 60 and

1 2 1 1 2 2 1 2 3 1 2 1 2 3 1 2 (−i)|d d α |+|d d α |i|d d d⊥(α +α )|(−i)|d d d (α +α )|

1 2 1 1 2 2 1 2 1 2 1 2 3 1 2 1 2 3 1 2 1 2 3 1 2 = (−i)|d d α |+|d d α |i|d d (α +α )|+|d d d (α +α )|−2|d d d (α +α )|(−i)|d d d (α +α )|

1 2 3 1 2 1 2 1 1 2 2 1 2 1 1 2 2 1 2 1 2 = (−1)|d d d (α +α )|(−i)|d d α |+|d d α |i|d d α |+|d d α )|−2|d d α α |

1 2 3 1 2 1 2 1 2 = (−1)|d d d (α +α )|+|d d α α |, we have

2 3 2 3 2 3 1 s 1 3 1 3 1 3 2 s (d +d ,(d +d )⊥(α +α )+γ +γ ),(d +d ,(d +d )⊥(α +α )+γ +γ ) B 1 2 3 1 2 3 1 2 3 1 2 3 1 1 1 2 2 2 3 3 3 (d +d +d ,(d +d +d )⊥(α +α +α )+γ +γ +γ ,(d ,d⊥α ),(d ,d⊥α ),(d ,d⊥α ))

1 1 2 1 2 1 2 2 2 1 1 3 3 1 2 3 1 2 3 1 2 3 = (−1)|d⊥d⊥α α |(−1)|(d d⊥α +d d⊥α )(d⊥α +(d +d +d )⊥(α +α +α )+(γ +γ +γ ))| #∆d1d2d3 |γ3(α1+α2)|+|(γs+γu)(α1+α2)|+|d1d2d3(α1+α2)|+|d1d2(α1+α2+α1α2)|+ 1 (|d1α2|+|d2α1|) (−1) 2 . Since

1 2 2 2 1 1 3 3 1 2 3 1 2 3 1 2 3 1 2 3 1 2 (−1)|(d d⊥α +d d⊥α )(d⊥α +(d +d +d )⊥(α +α +α )+(γ +γ +γ ))|(−1)|d d d (α +α )|

1 1 2 2 1 2 3 1 2 1 2 3 2 3 1 2 3 1 3 1 2 3 2 1 2 3 1 2 3 1 1 2 3 = (−1)|α γ |+|α γ |(−1)|d d d (α +α )|+|d d⊥d⊥α α |+|d⊥d d⊥α α |+|d d⊥d α (α +α +α )|+|d⊥d d α (α +α +α )|

1 1 2 2 1 2 3 1 2 1 2 3 2 3 1 2 3 1 3 1 2 3 2 1 2 3 2 1 3 1 2 3 1 1 2 3 1 2 3 = (−1)|α γ |+|α γ |(−1)|d d d (α +α )|+|d d⊥d⊥α α |+|d⊥d d⊥α α |+|d d⊥d α |+|d d⊥d α (α +α )|+|d⊥d d α |+|d⊥d d α (α +α )|

1 1 2 2 1 3 2 2 3 1 1 2 3 2 3 1 2 3 1 3 1 2 3 2 1 3 1 2 3 1 2 3 = (−1)|α γ |+|α γ |+|d d α |+|d d α |(−1)|d d⊥d⊥α α |+|d⊥d d⊥α α |+|d d⊥d α (α +α )|+|d⊥d d α (α +α )|

1 1 2 2 1 3 2 2 3 1 1 2 2 3 1 2 1 3 1 2 3 1 2 = (−1)|α γ |+|α γ |+|d d α |+|d d α |(−1)|d d⊥α α |+|d⊥d α α |+|(d +d )d α α |, ﬁnally we have

1 1 1 2 2 1 2 2 3 2 1 1 3 1 2 3 1 2 1 3 2 2 3 1 = (−1)|α γ |+|α γ |+|d d⊥α α |+|d d⊥α α |+|(d +d )d α α |+|d d α |+|d d α | #∆d1d2d3 |d1 d2 α1α2| |γ3(α1+α2)|+|(γs+γu)(α1+α2)|+|d1d2(α1+α2+α1α2)|+ 1 (|d1α2|+|d2α1|) (−1) ⊥ ⊥ (−1) 2 Set

2 3 3 2 3 2 3 1 2 3 1 2 3 2 3 1 2 3 1 2 3 1 1 2 3 s |d d α |+|d d⊥α α |+ |d α | |d (d +d )(α +α )|+|d (d +d )⊥α (α +α )|+ |d (α +α )| X12(γ ) = (−1) 2 (−1) 2

2 2 3 s 1 2 × (−1)|α (γ +γ )|+|γ (α +α )|

1 3 3 1 3 1 3 1 1 3 2 1 3 1 3 2 1 3 2 1 3 1 2 1 3 u |d d α |+|d d⊥α α |+ |d α | |d (d +d )(α +α )|+|d (d +d )⊥α (α +α )|+ |d (α +α )| X21(γ ) = (−1) 2 (−1) 2

1 1 3 u 1 2 1 2 × (−1)|α (γ +γ )|+|γ (α +α )|+|α α |

1 1 2 2 1 2 2 3 2 1 1 3 1 2 3 1 2 1 3 2 2 3 1 B(γs, γu) = (−1)|α γ |+|α γ |+|d d⊥α α |+|d d⊥α α |+|(d +d )d α α |+|d d α |+|d d α |

|d1 d2 α1α2| |γ3(α1+α2)|+|(γs+γu)(α1+α2)|+|d1d2(α1+α2+α1α2)|+ 1 (|d1α2|+|d2α1|) (−1) ⊥ ⊥ (−1) 2 . Then, (FA4) is equivalent to prove:

u 1 X s u s X21(γ ) = B(γ , γ )X12(γ )(5.6) #∆d1d2d3 γs∈∆d1d2d3 for any γu ∈ ∆d1d2d3 . 61

u s u s Hence, in order to prove (FA4), it suﬃces to show that 1 = X21(γ )X12(γ )B(γ , γ ). In u s u s X21(γ )X12(γ )B(γ , γ )

|d2d3α3|+|d2d3 α2α3|+ 1 |d2α3| |d1(d2+d3)(α2+α3)|+|d1(d2+d3) α1(α2+α3)|+ 1 |d1(α2+α3)| = (−1) ⊥ 2 (−1) ⊥ 2

2 2 3 s 1 2 (−1)|α (γ +γ )|(−1)|γ (α +α )|

|d1d3α3|+|d1d3 α1α3|+ 1 |d1α3| |d2(d1+d3)(α1+α3)|+|d2(d1+d3) α2(α1+α3)|+ 1 |d2(α1+α3)| (−1) ⊥ 2 (−1) ⊥ 2

1 1 3 u 1 2 1 2 (−1)|α (γ +γ )|(−1)|γ (α +α )|(−1)|α α |

1 1 2 2 1 2 2 3 2 1 1 3 1 2 3 1 2 1 3 2 2 3 1 (−1)|α γ |+|α γ |+|d d⊥α α |+|d d⊥α α |+|(d +d )d α α |+|d d α |+|d d α |

|d1 d2 α1α2| |γ3(α1+α2)|+|(γs+γu)(α1+α2)|+|d1d2(α1+α2+α1α2)|+ 1 (|d1α2|+|d2α1|) (−1) ⊥ ⊥ (−1) 2 , all the terms which involve γ1, γ2, γ3, γs, γu cancel each other. Thus,

|d2d3α3|+|d2d3 α2α3|+ 1 |d2α3| |d1(d2+d3)(α2+α3)|+|d1(d2+d3) α1(α2+α3)|+ 1 |d1(α2+α3)| RHS =(−1) ⊥ 2 (−1) ⊥ 2

|d1d3α3|+|d1d3 α1α3|+ 1 |d1α3| |d2(d1+d3)(α1+α3)|+|d2(d1+d3) α2(α1+α3)|+ 1 |d2(α1+α3)| |α1α2| (−1) ⊥ 2 (−1) ⊥ 2 (−1)

1 2 2 3 2 1 1 3 1 2 3 1 2 1 3 2 2 3 1 (−1)|d d⊥α α |+|d d⊥α α |+|(d +d )d α α |+|d d α |+|d d α |

|d1 d2 α1α2| |d1d2(α1+α2+α1α2)|+ 1 (|d1α2|+|d2α1|) (−1) ⊥ ⊥ (−1) 2 . and similarly all the terms which involve α3 cancel,

|d1(d2+d3)α2|+|d1(d2+d3) α1α2|+ 1 |d1α2| |d2(d1+d3)α1|+|d2(d1+d3) α2α1|+ 1 |d2α1| |α1α2| RHS =(−1) ⊥ 2 (−1) ⊥ 2 (−1)

|(d1+d2)d3α1α2|+|d1d3α2|+|d2d3α1| |d1 d2 α1α2| |d1d2(α1+α2+α1α2)|+ 1 (|d1α2|+|d2α1|) (−1) (−1) ⊥ ⊥ (−1) 2 and thus all terms cancel. Hence, we have:

r r Theorem 5.7. For any code G ⊂ Z2 with (1 ) ∈ G,S G is a simple framed algebra with the codes (∆G, (∆G)⊥).

Proof. It suﬃces to show that S G is simple, which follows from the fact that AG(d) is an 0 0 irreducible C[CˆG]-module and the product AG(d) ⊗ AG(d ) → AG(d + d ) is non-zero.

By Theorem 4.3, FG = FS G is a full vertex operator algebra. Thus, we have:

r r Corollary 5.8. For any code G ⊂ Z2 with (1 ) ∈ G,FG is a simple framed full vertex operator algebra.

We can calculate all the four point correlation functions of FG explicitly by using the r code G. For this purpose, it is convenient to introduce the length function on G ⊂ Z2. r Pr r Deﬁne a map | − |∆ : Z2 → Z≥0 by |g|∆ = i=1 gi for g = (g1, g2,..., gr) ∈ Z2. For the sake of simplicity, we only compute some special correlators. General correlators can be obtained similarly. Recall that by the discussion in Section 4.4 (S G)(d1+d2+d3,0) d1+ˆd2+d3 is an irreducible representation of C[CG ] and decomposed into the direct sum of distinct one dimensional representations of C[∆d1+ˆd2+d3 ], i.e., M (5.7) (S G)(d1+d2+d3,0) = ((S G)(d1+d2+d3,0))χ. χ∈(∆d1+d2+d3 )∨ 62

d1+d2+d3 Furthermore by construction td1+d2+d3 spans the trivial representation of C[∆ ], ∨ 1 2 3 C 1 2 3 ∈ C (S G)(d +d +d ,0))trivial = td +d +d . Let td1+d2+d3 HomC(S G, ) be the linear map deﬁned by the composition of projections M S G = (S G)d,c → (S G)(d1+d2+d3,0) (d,c) and M (S G)(d1+d2+d3,0) = ((S G)(d1+d2+d3,0))χ → (S G)(d1+d2+d3,0))trivial C, χ∈(∆d1+d2+d3 )∨

∨ 1 2 3 C 1 2 3 C 1 2 3 where (S G)(d +d +d ,0))trivial = td +d +d is normalized by td1+d2+d3 (td +d +d ) = 1.

Remark 5.9. We recall that by Proposition 4.8, S G has a natural non-degenerate sym- 0 d1+d2+d3 ∨ metric bilinear form (−, −): S G × S G → C. By the invariance, for χ, χ ∈ (∆ ) , 0 ∨ 1 2 3 1 2 3 0 − ((S G)(d +d +d ,0))χ and ((S G)(d +d +d ,0))χ are orthogonal if χ , χ . Thus, td1+d2+d3 ( ) = ∗ (td1+d2+d3 , −) ∈ S G.

Then, we have:

1 2 3 0 1 2 3 Proposition 5.10. For d , d , d ∈ DG, set d = d + d + d .

∗ ht Y t 1 z Y t 2 z t 3 i d0 , ( d , 1) ( d , 2), d 1 1 2 3 1 1 3 1 2 3 1 1 2 − |d d d |l − |d d |l − |d d |l − |d d |l = 2 2 (z1z¯1) 8 (z2z¯2) 8 ((z1 − z2)(¯z1 − z¯2)) 8

1 1 2 3 √ √ p 2 |d d d |l z1z¯1 + z2z¯2 + (z1 − z2)(¯z1 − z¯2) .

Proof. By (2.6),

∗ ht Y t 1 z Y t 2 z t 3 i d1+d2+d3 , ( d , 1) ( d , 2), d X (d2+d3,γ) ∗ C z , z ht , t 1 · t 2 · 2 3 t 3 i = (d1+d2+d3,0),(d1,0),(d2,0),(d3,0)( 1 2) d1+d2+d3 d d (d +d ,γ) d γ∈∆d2d3 X (d2+d3,γ) ∗ C z , z ht , t 1 · e · t 2 3 i = (d1+d2+d3,0),(d1,0),(d2,0),(d3,0)( 1 2) d1+d2+d3 d ( γ d +d ) γ∈∆d2d3 X (d2+d3,γ) ∗ |d1(d2+d3)γ|+ 1 |d1γ| C z , z ht , − 2 δ 1 2 3 · e · t 1 2 3 i = (d1+d2+d3,0),(d1,0),(d2,0),(d3,0)( 1 2) d1+d2+d3 ( 1) d (d +d ) γ d +d +d γ∈∆d2d3 X 2 3 C(d +d ,γ) z , z . = (d1+d2+d3,0),(d1,0),(d2,0),(d3,0)( 1 2) γ∈∆d1d2d3 By Table 4,

C 1 1 (x, y) = C 1 1 (x, y) = C 1 1 (x, y) = 1 16 , 16 ,0,0 16 ,0, 16 ,0 16 ,0,0, 16 − 1 C 1 1 (x, y) = (x − y) 8 0, 16 , 16 ,0 − 1 − 1 C 1 1 (x, y) = x 8 , C 1 1 (x, y) = y 8 . 0, 16 ,0, 16 0,0, 16 , 16 63

Thus, we have

∗ ht Y t 1 z Y t 2 z t 3 i d1+d2+d3 , ( d , 1) ( d , 2), d 1 2 3 1 1 3 1 2 3 1 1 2 −|d d d |l − |d d |l − |d d |l − |d d |l = 2 (z1z¯1) 8 (z2z¯2) 8 ((z1 − z2)(¯z1 − z¯2)) 8

1 2 3 q √ √ √ √ q √ √ √ √ !|d d d |l z1 + z2)( z¯1 + z¯2) + z1 − z2)( z¯1 − z¯2) , where we used, by Table 6,

1 1 C0 (z , z )C0 (z , z ) + C 2 (z , z )C 2 (z , z ) 1 , 1 , 1 , 1 1 2 1 1 1 1 1 2 1 1 1 1 1 2 1 1 1 1 1 2 16 16 16 16 16 , 16 , 16 , 16 16 , 16 , 16 , 16 16 , 16 , 16 , 16 q q ! 1 − 1 √ √ √ √ √ √ √ √ = (z z¯ z z¯ (z − z )(¯z − z¯ )) 8 z + z )( z¯ + z¯ ) + z − z )( z¯ − z¯ ) . 2 1 1 2 2 1 2 1 2 1 2 1 2 1 2 1 2 and q √ √ √ √ q √ √ √ √ !2 z1 + z2)( z¯1 + z¯2) + z1 − z2)( z¯1 − z¯2) √ √ p = 2( z1z¯1 + z2z¯2 + (z1 − z2)(¯z1 − z¯2))

The correlator computed the above proposition is a limit of the four point correlation function (see Remark 3.9) As expected, the four point correlation function is a more L symmetric form. Let h−i : FG → C be the projection FG = h,h¯∈R2 (FG)h,h¯ → (FG)0,0 = C1 C, where C1 C is normalized by h1i = 1. The linear map h−i is called the vacuum expectation value and the formal power series of the form,

hY a z Y a z Y a z Y a z 1i ( 0, 0) ( 1, 1) ( 2, 2) ( 3, 3) , is called a four point correlation function. It was proved in [Mo2] that under certain condition all four point correlation functions are convergent to a real analytic function on

1 1 4 X4(CP ) = {(z1, z2, z3, z4) ∈ (CP ) | zi , z j} and can be calculated from the specialize one like

∗ ht Y t 1 z Y t 2 z t 3 i g0 , ( g , 1) ( g , 2), g . By using these results, we have:

0 1 2 3 Corollary 5.11. Let d , d , d , d ∈ DG. Then, hY t 0 z Y t 1 z Y t 2 z Y t 3 z 1i ( d , 0) ( d , 1) ( d , 2) ( d , 3) 1 i j 0 1 2 3 − |d d |l 0 1 2 3 −|d d d d |l 8 |d d d d |l = 2 δd0+d1+d2+d3,0Π0≤i< j≤3 (zi − z j)(¯zi − z¯ j) F(z0, z1, z2, z3) where we set

1 1 1 1 2 F(z0, z1, z2, z3) = |(z0 − z1)(z2 − z3)| 2 + |(z0 − z2)(z1 − z3)| 2 + |(z0 − z3)(z1 − z2)| 2 . 64

For example, let r = 1 and G = h1i. Then, the corresponding (1, 1)-framed algebra is the three dimensional framed algebra S Ising = C1 ⊕ Ca ⊕ Cd with the product d · d = 1 + a and a·d = d ·a = d and the grading (S Ising)0,0 = C1, (S Ising) 1 1 = Ca and (S Ising) 1 1 = Cd. 2 , 2 16 , 16 4 By the above result, the four point correlation function for (d, d, d, d) ∈ S Ising is

1 1 1 − 8 1 1 1 2 Π (z − z )(¯z − z¯ ) |(z − z )(z − z )| 2 + |(z − z )(z − z )| 2 + |(z − z )(z − z )| 2 , 2 0≤i< j≤3 i j i j 0 1 2 3 0 2 1 3 0 3 1 2 which is exactly equal to the correlator of the critical Ising model in the physics literature (see for example [BPZ]). Another example of a four point correlation function is given in the following proposition:

0 1 2 3 left Proposition 5.12. Let α , α , α , α ∈ CG . Then,

hY e 0 · t r+r z Y e 1 · t r+r z Y e 2 · t r+r z Y e 3 · t r+r z 1i ( α 1 , 0) ( α 1 , 1) ( α 1 , 2) ( α 1 , 3) − r −r |α1α3| 8 =δα0+α1+α2+α3,02 (−1) (1, eα0 · eα1 · eα2 · eα3 )Π0≤i< j≤3 (zi − z j)(¯zi − z¯ j)

0 1 0 2 0 3 1 2 1 3 2 3 r−|α α +α α +α α +α α +α α +α α |l × F(z0, z1, z2, z3)

0 1 2 3 0 2 1 3 0 3 1 2 |α α +α α |l |α α +α α |l |α α +α α |l G01,23(z0, z1, z2, z3) G02,13(z0, z1, z2, z3) G03,12(z0, z1, z2, z3) where Gi j,kl(z1, z2, z3, z4) are deﬁned by

1 1 1 1 2 G01,23(z0, z1, z2, z3) = −|(z0 − z1)(z2 − z3)| 2 + |(z0 − z2)(z1 − z3)| 2 + |(z0 − z3)(z1 − z2)| 2 ,

1 1 1 1 2 G02,13(z0, z1, z2, z3) = |(z0 − z1)(z2 − z3)| 2 − |(z0 − z2)(z1 − z3)| 2 + |(z0 − z3)(z1 − z2)| 2 ,

1 1 1 1 2 G03,12(z0, z1, z2, z3) = |(z0 − z1)(z2 − z3)| 2 + |(z0 − z2)(z1 − z3)| 2 − |(z0 − z3)(z1 − z2)| 2 .

0 1 2 3 Proof. e ·t r+r Y e ·t r+r z Y e ·t r+r z e ·t r+r We will ﬁrst calculate ( α 1 , ( α 1 , 1) ( α 1 , 2) α 1 ). By Proposition 4.8 and Remark 4.15, e 0 · t r+r Y e 1 · t r+r z Y e 2 · t r+r z e 3 · t r+r ( α 1 , ( α 1 , 1) ( α 1 , 2) α 1 ) X (0r+r,α2+α3+γ) 1 3 2 |α | = C(1r+r,0),(1r+r,0),(1r+r,0),(1r+r,0)(z1, z2)(−1) eα0 · t1r+r , (eα1 · t1r+r ) · (eα2 · eγ · eα3 ) γ∈∆1r+r X (0r+r,α2+α3+γ) 1 3 2 |α | = C(1r+r,0),(1r+r,0),(1r+r,0),(1r+r,0)(z1, z2)(−1) (eα1 · t1r+r ) · (eα0 · t1r+r ), eα2 · eγ · eα3 γ∈∆1r+r X (0r+r,α2+α3+γ) 1 (|α3|+|α0|) 2 0 = C(1r+r,0),(1r+r,0),(1r+r,0),(1r+r,0)(z1, z2)(−1) eα1 · eγ · eα0 , eα2 · eγ · eα3 γ,γ0∈∆1r+r X (0r+r,α2+α3+γ) 1 (|α3|+|α1|) 2 0 = C(1r+r,0),(1r+r,0),(1r+r,0),(1r+r,0)(z1, z2)(−1) 1, eα0 · eγ · eα1 · eα2 · eγ · eα3 . γ,γ0∈∆1r+r

The left-hand-side is zero unless γ + γ0 = α0 + α1 + α2 + α3. Since γ + γ0 ∈ ∆1r+r , 0 1 2 3 left left 1r+r 0 0 1 2 3 α +α +α +α ∈ CG and CG ∩∆ = 0, we may assume that γ = γ and α = α +α +α . 65

Then,

1 3 1 X 1 2 (0r+r,α2+α3+γ) 2 (|α |+|α |) |γ(α +α )| = (1, eα0 · eα1 · eα2 · eα3 ) (−1) (−1) C(1r+r,0),(1r+r,0),(1r+r,0),(1r+r,0)(z1, z2). γ∈∆1r+r By Table 6,

1 1 C0 (z , z )C0 (z , z ) − C 2 (z , z )C 2 (z , z ) 1 , 1 , 1 , 1 1 2 1 1 1 1 1 2 1 1 1 1 1 2 1 1 1 1 1 2 16 16 16 16 16 , 16 , 16 , 16 16 , 16 , 16 , 16 16 , 16 , 16 , 16 q q ! 1 − 1 √ √ √ √ √ √ √ √ = (z z¯ z z¯ (z − z )(¯z − z¯ )) 8 z + z )( z¯ − z¯ ) + z − z )( z¯ + z¯ ) 2 1 1 2 2 1 2 1 2 1 2 1 2 1 2 1 2 (5.8)

1 1 C 2 (z , z )C0 (z , z ) + C0 (z , z )C 2 (z , z ) 1 1 1 1 1 2 1 1 1 1 1 2 1 , 1 , 1 , 1 1 2 1 1 1 1 1 2 16 , 16 , 16 , 16 16 , 16 , 16 , 16 16 16 16 16 16 , 16 , 16 , 16 q q ! 1 − 1 √ √ √ √ √ √ √ √ = √ (z1z¯1z2z¯2(z1 − z2)(¯z1 − z¯2)) 8 z1 + z2)( z¯1 + z¯2) − z1 − z2)( z¯1 − z¯2) , 2

1 1 C 2 (z , z )C0 (z , z ) − C0 (z , z )C 2 (z , z ) 1 1 1 1 1 2 1 1 1 1 1 2 1 , 1 , 1 , 1 1 2 1 1 1 1 1 2 16 , 16 , 16 , 16 16 , 16 , 16 , 16 16 16 16 16 16 , 16 , 16 , 16 q q ! 1 − 1 √ √ √ √ √ √ √ √ = √ (z1z¯1z2z¯2(z1 − z2)(¯z1 − z¯2)) 8 z1 + z2)( z¯1 − z¯2) − z1 − z2)( z¯1 + z¯2) , 2 We would like to rewrite (5.8) in more symmetric way. An important observation is that q √ √ √ √ q √ √ √ √ !2 z1 + z2)( z¯1 − z¯2) + z1 − z2)( z¯1 + z¯2) √ √ p = z1z¯1 − z2z¯2 + (z1 − z2)(¯z1 − z¯2) and similarly q √ √ √ √ q √ √ √ √ !2 z1 + z2)( z¯1 + z¯2) − z1 − z2)( z¯1 − z¯2) √ √ p = z1z¯1 + z2z¯2 − (z1 − z2)(¯z1 − z¯2) q √ √ √ √ q √ √ √ √ !2 z1 + z2)( z¯1 − z¯2) − z1 − z2)( z¯1 + z¯2) √ √ p = z1z¯1 − z2z¯2 − (z1 − z2)(¯z1 − z¯2).

Hence, by the results in [Mo2], the assertion holds.

Remark 5.13. As it is mentioned in Remark 3.9, the pure braid group P4 acts on the functions Gi j,kl(z1, z2, z3, z4). The action of the standard generators σ12, σ23, σ34 ∈ P4 can be written as:

σ12 :(G12,34, G13,24, G14,23) 7→ (G12,34, −G14,23, G13,24)

σ23 :(G12,34, G13,24, G14,23) 7→ (G13,24, −G12,34, G14,23)

σ34 :(G12,34, G13,24, G14,23) 7→ (G12,34, G14,23, −G13,24),

2 which generates S 4 Z2 o S 3. 66

The following proposition follows from the construction:

r1 r2 ri Proposition 5.14. Let r1, r2 ∈ Z> and G1 ⊂ Z2 and G2 ⊂ Z2 be codes with (1 ) ∈ Gi for i = 1, 2. Then, FG1⊥G2 is isomorphic to FG1 ⊗ FG2 as a full vertex operator algebra, where r1+r2 G1 ⊥ G2 is a subcode of Z2 .

Let us consider the chiral part (holomorphic part) of FG (see Proposition 2.6). Since the 1 16 -part is diagonal, it does not contribute to the holomorphic part. Thus, the holomorphic left ∩ part of FG is equal to the holomorphic part of FC[CˆG] and is determined by CG = CG r (Z2 × 0), which is exactly equal to ⊥ r G = {α ∈ Z2 | |αG|∆ ⊂ 2Z}, r the dual code of G ⊂ Z2 with respect to | − |∆. Set ⊥ ⊥ G [2] = {α ∈ G | |α|∆ = 2}. ⊥ Then, it is clear that dim(FG)1,0 = #G [2]. Hence, we have:

Proposition 5.15. The holomorphic part of FG is isomorphic to FC[Gˆ⊥]. Furthermore, ⊥ dim(FG)1,0 = dim(FG)0,1 = #G [2]. Since (1r) ∈ G, G⊥ is an even code. Thus, G⊥ is a subcode of the maximal even code even r C ⊂ Z . The vertex operator algebra structure of F evenˆ is determined in Proposi- r,0 2 C[Cr,0 ] tion 4.24. Thus, we can determine the vertex operator algebra structure of FC[Gˆ⊥] as a subalgebra of F evenˆ . C[Cr,0 ] r r 5.3. Modular invariance. Let G ⊂ Z2 be a code satisfying (1 ) ∈ G. The (q, q¯)-character of FG is deﬁned by

r r L(0)− 48 L(0)− 48 R R ChqFG = trq q¯ |FG ∈ Z[[q , q¯ ]] X h− r h¯− r = dim(FG)h,h¯ q 48 q¯ 48 . h,h¯∈R We will soon see that the (q, q¯)-character is a real analytic function on the upper half-plane H with the identiﬁcation q = exp(2πiτ) andq ¯ = exp(−2πiτ¯) for τ ∈ H. In this section, we prove the modular invariance of the (q, q¯)-character of a full vertex operator algebra

FG. (r,r) ⊗r Let CIS be the C-vector space with the basis θ0, θ 1 , θ 1 and set CIS = (CIS) ⊗ 2 16 ⊗r L (CIS) = (r,r) Cθµ. µ∈IS L A character of a framed algebra S = µ∈IS(r,r) S µ is deﬁned by X (r,r) ZS = dim S µ θµ ∈ CIS . µ∈IS(r,r) Let C[Xr,r] be the space of real analytic functions on H spanned by the product of { a1 b1 a2 b2 a3 b3 } (r,r) → theta functions χ0 (τ)χ0 (τ)χ 1 (τ)χ 1 (τ)χ 1 (τ)χ 1 (τ) a1,a2,a3,b1,b2,b3∈Z≥0 and F : C[IS ] 2 2 16 16 r,r r ∈ C[X ] the linear map deﬁned by F(θµ) = Πi=1χµi (τ)χµi (τ) for µ = (µ1, . . . , µr, µ¯ 1,..., µ¯ r) IS(r,r). By Lemma 1.10, we have: 67

Lemma 5.16. The (q, q¯)–character of the full vertex operator algebra FG is equal to

F(ZS G ).

Recall that AG(d) is the induced module C[CˆG] ⊗C[∆d] Ctd for any d ∈ DG. Hence, we have:

P P −|d|l Lemma 5.17. For the framed algebra S G,ZS G = d∈DG α∈CG 2 θ(d,d⊥α). In particular, d d −|d| d dim(S G)d,d⊥α = #[CG : ∆ ] = 2 #CG for any d ∈ DG and α ∈ CG.

Set ZG = ZS G . By the assumption (FA1), the invariance of T-transformation, i.e.,

F(ZD)(τ + 1) = F(ZD)(τ), is clear. Motivated by Lemma 1.10, we deﬁne a linear map S : CIS → CIS by 1 1 1 S (θ0) = θ0 + θ 1 + √ θ 1 2 2 2 2 16 1 1 1 S (θ 1 ) = θ0 + θ 1 − √ θ 1 2 2 2 2 2 16 1 1 S (θ 1 ) = √ θ0 − √ θ 1 . 16 2 2 2 ⊗r+r (r,r) (r,r) −1 We also denote the map S : IS → IS by S . Since F(ZG)( τ ) = F(S (ZG))(τ), in order to prove the modular invariance of the character ChqFG, it suﬃces to show that

S (ZG) = ZG.

Remark 5.18. We remark that F is not injective, since, for example, F(θ 1 ) = χ0χ 1 = ( 2 ,0) 2 F(θ 1 ). Thus, the condition S (ZD) = ZD is stronger than the modular invariance of the (0, 2 ) character. In fact, this stronger condition is needed in order to deﬁne the conformal ﬁeld theory FG on genus one surfaces.

r+r We will prepare some elementary lemma from linear algebra. For d ∈ Z2 and a d subgroup H ⊂ Z2, set

⊥d d H = {α ∈ Z2 | |αH| ⊂ 2Z}.

⊥ ⊥ ⊥ d d ⊥ It is clear that H ⊂ d(H ) = {dα}α∈H . For α ∈ H and h ∈ H, since H ⊂ Z2, |αh| = |α(dh)| = |(dα)h|. Thus, dα ∈ H⊥d . Then, we have:

r+r r+r ⊥ d ⊥d Lemma 5.19. For d ∈ Z2 and a subgroup A ⊂ Z2 , d(A ) = (A ) .

⊥ d d d Proof. Let α ∈ A . Since A = A ∩ Z2 ⊂ A, for any a ∈ A , |αa| ∈ 2Z. Since |αa| = |α(da)| = |(dα)a|, dα ∈ (Ad)⊥d , which implies that dA⊥ ⊂ (Ad)⊥d .

d ⊥d ⊥ d ⊥d We will show that (A ) ⊂ A . Let b ∈ (A ) . Let fb : A → Z2 be a Z2-linear map |bα| d d ⊥d defined by fb(α) = (−1) for α ∈ A. Then, fb(A ) = 1 by definition of (A ) . If fb ≡ 1, ⊥ d⊥ then b ∈ A and there is nothing to prove. We will claim that there exits γ ∈ Z2 such that b + γ ∈ A⊥. If such γ exists, then d(b + γ) = db = b, which implies that (Ad)⊥d ⊂ A⊥. d Note that A is the kernel of the left multiplication d⊥∗ : A → d⊥A, α → d⊥α. d d Thus, A/A d⊥A and fb defines a Z2-linear map f˜b : d⊥A A/A → Z2. Since 68

d⊥ d⊥ |αβ| d⊥ | − | : Z2 × Z2 → Z2, (α, β) 7→ (−1) is a perfect pairing, by d⊥A ⊂ Z2 , there exists d⊥ ˜ |γα| |α(b+γ)| γ ∈ Z2 such that fb(α) = (−1) for any α ∈ d⊥A. For any α ∈ A,(−1) = |αb| |αγ| |α(d⊥γ)| |(d⊥α)γ| (−1) (−1) = fb(α)(−1) = f˜b(d⊥α)(−1) = 1. Then, we have:

r r Proposition 5.20. For any code G ⊂ Z2 with (1 ) ∈ G,S (ZG) = ZG.

(r,r) (r,r) Proof. For µ ∈ IS , define a linear map πµ : CIS → C by taking the coefficient of θµ. 0 0 r+r 0 0 By Lemma 5.17, it suffices to show that for d , c ∈ Z2 with d c = 0,

 0 0  −|d |l d 0 ∈ 0 0 ∈ 2 #CG if d DG and c = d⊥α for some α CG, π(d0,c0)S (ZG) =  0 otherwise. 0 By the deﬁnition of S , πd0,c0 (S (θd,c)) = 0 unless d d = (0,..., 0), i.e., the support of d and d0 are disjoint. 0 0 d⊥ Since {d ∈ DG | dd = (0,..., 0)} is equal to DG , by the deﬁnition of the map S , 0 0 √ 0 0 0 0 |d c|+|dc | −|d |l−|d |r−|d|l−|d|r −|(d+d )⊥|l−|(d+d )⊥|r π(d0,c0)S (θd,c) = (−1) 2 2

√ 0 0 0 0 = 2|d |l+|d |r 4−r(−1)|d c|+|dc |2|d|l . Thus, by Lemma 5.17,

X X −|d| 0 0 l 0 0 π(d ,c )S (ZG) = 2 π(d ,c )(S (θd,d⊥α)) 0 d⊥ α∈CG d∈DG √ 0 0 |d |l+|d |r 2 X X 0 0 − |dc |+|d⊥d α| = r ( 1) 4 0 d⊥ α∈CG d∈DG √ 0 0 |d |l+|d |r 2 X X 0 0 − |dc |+|d α| = r ( 1) , 4 0 d⊥ α∈CG d∈DG where in the last equality we used dd0 = 0. Since   0 ⊥ X 0 #CG if d ∈ C (−1)|d α| =  G  α∈CG 0 otherwise,

0 d⊥ and the similar result for DG , we have

 0 d0 d0  |d |l−2r ⊥ 0 ∈ 0 ∈ ⊥ ⊥ 2 #CG#DG if d DG and c (DG ) π(d0,c0)S (ZG) =  0 otherwise.

0 0 0 d d ⊥ 0 d ⊥ 0 0 0 0 ⊥ ⊥ 0 ⊥ d⊥ ⊥ d⊥ Since d c = 0, c ∈ (DG ) is equivalent to c ∈ (DG ) . By Lemma 5.19, (DG ) = 0 0 d d ⊥ 0 0 ⊥ 0 ⊥ 0 ⊥ d⊥ d⊥(DG) = d⊥CG, as desired. Furthermore, by Lemma 5.19, dim DG = 2|d⊥|l−dim(DG ) = 0 0 0 0 0 0 d −|d |l d 2|d⊥|l −dim d⊥CG = 2r−2|d |l −dim CG +dim CG . Hence, π(d0,c0)S (ZG) is equal to 2 #CG 0 0 0 if d ∈ DG and c ∈ d⊥CG and zero otherwise. Hence, we have: 69

Corollary 5.21. The character of the full vertex operator algebra FG satisﬁes the modular invariance.

r We end this section by determining the dimension of S G. For a code G ⊂ Z2, a code Pr k enumerate is a polynomial PG(t) = k=0 Nk(G)t ∈ Z[t] where the coeﬃcients are deﬁned by Nk(G) = #{g ∈ G | |g|∆ = k}. For d ∈ D , by Lemma 5.17, dim A (d) = #CG . G G 2|d|l Thus, we have:

2r−dim G 1 Proposition 5.22. The dimension of the framed algebra S G is equal to 2 PG( 2 ). 5.4. Examples and classification of codes. In this section, we consider examples of the code conformal field theories constructed in Theorem 5.7. By Proposition 5.14, it r suffices to study the case that the code G ⊂ Z2 is indecomposable. We will also classify indecomposable codes up to r ≤ 6. r Let hα1, . . . , αki be the code generated by α1, . . . , αk ∈ Z2. We introduce the following min r r even codes: Let Gr be the subcode of Z2 generated by (1 ) = (1, 1,..., 1) and Gr is the r even subcode of Z2 consisting of all even codewords if r is even. Note that Gr is generated by r − 1 codewords, (1, 1, 0,..., 0), (0, 1, 1, 0,..., 0),..., (0,..., 0, 1, 1) and contains 1r+r since r is even. min min min min ⊥ min even G D D min C D C C We first study r . Set r = Gr and r = ( r ) . Then, r = r,r , the r+r min r r maximal even code in Z2 . We note that Gr is the minimal code which contains 1 ∈ Z2. min min Thus, Gr maximizes the dual code Cr . The code enumerate and the dimension are in this case

r P min t t Gr ( ) = 1 + r−1 r S min dim Gr = 2 (2 + 1)

r+r even S min A min ⊕ A min A min C C ˆ and Gr = Gr (0) Gr (1 ), where Gr (0) = [ r ] is a twisted group algebra r+r r−1 A min and Gr (1 ) is a 2 -dimensional representation. By proposition 5.15, the chiral part of the theory is isomorphic to F evenˆ and the dimension of the Lie algebra (F min ) is C[Cr ] Gr 1,0 even r #Cr = 2 , which is equal to the dimension of SO(r). In fact, by Proposition 4.24, they F min are isomorphic. Moreover, we can identify the full vertex algebra Gr as the algebra of SO(r) WZW model at level one for r ≥ 4. We remark that if r = 3, then F min is SU(2) G3 WZW model at level two.

Important informations of a code conformal ﬁeld theory FG are the dimension of the framed algebra S G and the structure of the Lie algebra (FG)1,0 (FG)0,1, which we call currents. In this section, we classify codes G and determine dim S G and (FG)1,0.

even even For example, for r ≥ 2Z≥1, the code enumerate of Gr and the dimension of S Gr are (1 + t)r + (1 − t)r P even (t) = Gr 2 r even dim S Gr = 3 + 1.

even ⊥ r In this case, there are no currents if r ≥ 4 since (Gr ) is h1 i. 70

Now, we will classify codes. The following lemma is clear:

Lemma 5.23. If a code G contains a codeword α ∈ G with |α|∆ = 1, then G is decompos- 0 r−1 0 able, in fact, G = hαi ⊥ G ⊂ Z2 ⊥ Z2 , where G = {β ∈ G | |αβ|∆ = 0}. If the dimension or the codimension of G is low, then the classiﬁcation is easy. In fact, we have:

r r Proposition 5.24. Let G ⊂ Z2 be a code with 1 ∈ G. Then, the following properties are hold: (1) If dim G = 1, then G = h1ri; r (2) If dim G = r, then G = Z2; (3) If dim G = 2, then G is decomposable; r (4) If dim G = r − 1 and G is indecomposable, then r is even and G = Geven. Proof. (1), (2) and (3) are clear. Assume that dim G = r − 1. Then, dim G⊥ = 1. Let r ⊥ k r−k α ∈ Z2 generate G and set |α|∆ = k. We may assume that α = (1 0 ). If k < r, then (0r−111) ∈ (G⊥)⊥ = G. Thus, by Lemma 5.23, if G is indecomposable, k = r and thus G = h1ri⊥. Furthermore, since 1r ∈ G, r must be even. All codes with r ≤ 3 are listed in Table 5.4: Table 9. all code CFTs of low rank

r code G current dim S G CFT 1 h1i 0 3 Ising 2 h10, 01i 0 9 Ising⊗2 h11i U(1) 10 Dirac 3 h100, 010, 001i 0 27 Ising⊗3 h100, 011i U(1) 30 Ising ⊗ Dirac h111i SU(2) 36 SU(2) WZW-model at level 2

Now, we consider an indecomposable code G with dim G = 3 for any r ∈ Z>0. Let

α1 ∈ G be the non-zero shortest codeword and set a = |α1|∆, that is, a = minα∈G,α,0 |α|∆. a r−a r Without loss of generality, we may assume that α1 = (1 0 ) ∈ Z2, the ﬁrst a components a r−a a r−a r are 1. Then, G is generated by α1 = (1 0 ) and α2 = (0 1 ) and some element α3 ∈ Z2. b a−b c r−a−c We may assume that α3 = (1 0 1 0 ) for some a ≥ b ≥ 0 and r − a ≥ c ≥ 0. By a ≥ r−a ≥ adding α1 and α2, we may further assume that 2 b and 2 c. If b = 0, then the code a c r−a−c r−a a r−a G is decomposed into h(1 )i ⊥ h(1 0 ), (1 )i ⊂ Z2Z2 . Thus, b > 0 and similarly c > 0. Hence, by Lemma 5.23, we can assume that a, b, c satisfy r a r − a (5.9) ≥ b + c ≥ a ≥ 2, ≥ b ≥ 1, ≥ c ≥ 1. 2 2 2 r Then, such integers a, b, c uniquely determine the code G. We denote it by Ga;b,c. r Proposition 5.25. If G is a indecomposable code with dim G = 3, then G = Ga;b,c for some a, b, c ∈ Z≥0 satisfying (5.9). 71

In the case of r = 5, all possible a, b, c ∈ Z≥0 satisfy (5.9) are only (a, b, c) = (2, 1, 1). 5 5 5 Thus, by Lemma 5.23, all possible indecomposable codes G ⊂ Z2 with (1 ) ∈ G are G2;1,1 and h11111i.

In the case of r = 6, all possible a, b, c ∈ Z≥0 satisfy (5.9) are (a, b, c) = (2, 1, 2), (2, 1, 1). 6 6 Thus, G = G2;1,2 = h(110000), (001111), (101100)i or G = G2;1,1 = h(110000), (001111), (101000)i. Finally, we consider a indecomposable code G with dim G = r − 2. In this case, dim G⊥ = 2. It is clear that G⊥ is an indecomposable code. We note that the condition (1r) ∈ G is equivalent to the condition that G⊥ is even. Similarly to the above, any two dimensional code is generated by (1a0r−a), (1b0a−b1c0r−a−c). If r − a − c > 0, then (0r−111) ∈ G, which contradicts to the condition that G is indecomposable. Set

r a r−a b a−b r−a r Ea;b = h(1 0 ), (1 0 1 )i ⊂ Z2

⊥ r r for r > a ≥ b ≥ 0. Then, G is isomorphic to Ea;b for some a, b ∈ Z. We note that Ea;b consists of {(0r), (1a0r−a), (1b0a−b1r−a), (0b1a−b1r−a)}. Since G⊥ is even, a, b + r − a, a − b + r − a ∈ 2Z. Since G⊥ is indecomposable, r > a > b > 0. We may assume that (1a0r−a) has a ≤ the minimal length and 2 b. Thus, a, b, c satisﬁes a (5.10) a, b − r ∈ 2Z, r > a > b > 0, ≥ b, b + r ≥ 2a, r − b ≥ a. 2 Hence, we have:

r ⊥ Proposition 5.26. If G is a indecomposable code with dim G = r − 2, then G = (Ea;b) for some a, b ∈ Z≥0 satisfying (5.10).

In the case of r = 6, all possible a, b, c ∈ Z≥0 satisfy (5.10) are (a, b) = (4, 2). Then, the 6 ⊥ weight enumerator of (E4;2) is

2 3 4 6 P 6 ⊥ (t) = 1 + 3t + 8t + 3t + t . (E4,2) By an easy computation, we have:

Lemma 5.27. For r ≥ 5 and k ≥ 1,

2 k+1 r−k−1 r−2 r P r (t) = 1 + t + 2t + 2t + t + t . G2;1,k We summarize the property of these codes in Table 10:

Table 10. Some series of code CFTs

code enumerator dim S G current min r r−1 r Gr 1 + t 2 (2 + 1) SO(r) even (1+t)r+(1−t)r r ≥ Gr 2 3 + 1 0 (r 4) 2;1,1 2 r−2 r r−3 r−2 Gr 1 + 3t + 3t + t 2 (7 · 2 + 13) SO(r − 3) (r ≥ 4) 2;1,2 2 3 r−3 r−2 r r−3 r−3 Gr 1 + t + 2t + 2t + t + t 2 (12 · 2 + 21) U(1) × SO(r − 4) (r ≥ 4)

All indecomposable codes with r ≤ 6 are listed in Table 11 72 Table 11. all indecomposable code CFTs of rank 4,5,6

r code current dim S G min 4 h1111i SO(4) 136 G4 ⊥ even h1111i 0 82 G4 min 5 h11111i SO(5) 528 G5 2;1,1 h11000, 00111, 01100i U(1) 276 G5 min 6 h11111i SO(6) 2080 G6 2;1,1 h110000, 001111, 101000i SO(3) 1000 G6 2 2;1,2 h110000, 001111, 101100i U(1) 936 G6 4;2 ⊥ h110000, 001100, 000011, 101010i 0 756 (E6 ) ⊥ even h111111i 0 730 G6

Remark 5.28. From the table, it seems that for given r ∈ Z>0 the smallest (resp. the r r second smallest) dimension of S G is 3 (resp. 3 + 1 if r is even) which is given by the CFT ⊗r − r r 1 r Ising (resp. FGeven ). Furthermore, it seems that the largest dimension of S G is 2 (2 +1) which is given by the SO(r)-WZW model.

6. Deformation of code CFTs In our previous paper [Mo3], we construct an exactly marginal deformation of a conformal ﬁeld theory in terms of a full vertex algebra. The deformation is called a current- current deformation in physics. In this section, we study the current-current deformation of code CFTs and calculate deformed four point correlation functions. In Section 6.1, we recall the current-current deformation of a full vertex algebra in a general setting and in Section 6.2 apply it to code CFTs.

6.1. Current-current deformation of full vertex algebras. In this section, we recall the deﬁnition of the current-current deformation of a full vertex algebra from [Mo3].

Let Hl and Hr be real ﬁnite dimensional vector subspaces equipped with non-degenerate symmetric bilinear forms (−, −)l : Hl ×Hl → R and (−, −)l : Hr ×Hr → R. Let MHl (0) and

MHr (0) be aﬃne Heisenberg vertex algebras associated with (Hl, (−, −)l) and (Hr, (−, −)r).

Set H = Hl ⊕ Hr and let p, p : H → H be projections on Hl and Hr and

MH,p = MHl (0) ⊗ MHr (0), the tensor product of the vertex algebra MHl (0) and the conjugate vertex algebra MHr (0) (see Proposition 2.2 and Proposition 2.3).

In this section, we consider a class of a full vertex algebra which is an MH,p-module (like an algebra over a ring). More precisely, let F be a full vertex algebra and we assume that MH,p is a subalgebra of F, MH,p ⊂ F. Then, since Hl = (MH,p)1,0 and Hr = (MH,p)0,1,

F ⊂ F1,0 and Hr ⊂ F0,1. 0 We note that the subspaces Hl and Hr satisfy the following conditions: For any hl, hl ∈ 0 Hl and hr, hr ∈ Hr, 73

H1) Hl ⊂ F1,0 and Hr ⊂ F0,1;

H2) DH¯ l = 0 and DHr = 0; 0 0 0 0 H3) hl(1, −1)hl = (hl, hl )l1, hr(−1, 1)hr = (hr, hr)r1; 0 0 H4) hl(n, −1)hl = 0, hr(−1, n)hr = 0 for any n = 0 or n ∈ Z≥2. P −n−1 Since hl ∈ Hl is a holomorphic vector, by Lemma 2.5, Y(hl, z) = n∈Z hl(n, −1)z . For h ∈ H, set

h[n]0 = (ph)(n, −1) + (ph ¯ )(−1, n). By Lemma 2.5 and Lemma 2.7, for any h, h0 ∈ H,

0 0 0 [h[n]0, h [m]0] = (ph, ph )l + (ph ¯ , ph¯ )r δn+m,0.

Thus, {h[n]0}h∈H,n∈Z deﬁnes an action of the aﬃne Heisenberg Lie algebra Hl ⊕ˆ Hr on F. For α ∈ H, set

α F = {v ∈ F | h[0]0v = ((ph, pα)l + (ph ¯ , p¯α)r) v for any h ∈ H}.

A full H-vertex operator algebra, denoted by (F, H, p, ω, ω¯ ), is a full vertex operator

algebra (F, ω, ω¯ ) with a subalgebra MH,p such that

FHO1) h[0]0 are semisimple on F with real eigenvalues for any h ∈ H; α FHO2) For any α ∈ H, there exists N ∈ R such that Ft,t¯ ∩ F = 0 for t ≤ N or t¯ ≤ N;

FHO3) L(n)Hl = 0 and L(n)Hr = 0 for any n ≥ 1.

Remark 6.1. For good full vertex algebras, the split Cartan subalgebras of the Lie alge-

bras F1,0,F0,1 deﬁne a full H-vertex algebra structure on F.

Let (F, Y, H, p, ω, ω¯ ) be a full H-vertex algebra. By (FHO1) and (FHO2) and the representation theory of the aﬃne Heisenberg Lie algebra ([FLM, Theorem 1.7.3]), F is generated by lowest weight vectors as a module of the aﬃne Heisenberg Lie algebra. α α More precisely, let ΩF,H be the set of all vectors v ∈ F such that

(1) h[n]0v = 0 for any h ∈ H and n ≥ 1 and set M α ΩF,H = ΩF,H. α∈H L α Then, F is isomorphic to MH,p ⊗ ΩF,H = α∈H MH,p ⊗ ΩF,H as an MH,p-module. ph1[0]0 ph¯ 2[0]0 R R For h1, h2 ∈ H, deﬁne z z¯ ∈ EndF[z , z¯ ] by

zph1[0]0 z¯ph¯ 2[0]0 v = z(ph1,pα)l z¯(ph ¯ 2,p¯α)r v

for α ∈ H and v ∈ Fα.

Deﬁne a bilinear form (−, −)lat on H by

(α, β)lat = (pα, pβ)l − (p ¯α, p¯β)r

for α, β ∈ H and let O(Hl ⊕ −Hr) be the orthogonal group on the space (H, (−, −)lat). 74

Theorem 6.2. [Mo3, Section 5] Let (F, Y, H, p, ω, ω¯ ) be a full H-vertex algebra. There exists a family of vertex operators on F parametrized by σ ∈ O(Hl ⊕ −Hr)

R Yσ(−, z): F → EndF[[z, z¯, |z| ]] such that for any σ ∈ O(Hl ⊕ −Hr),

(1) (F, Yσ, 1, ω, ω¯ ) is a full vertex operator algebra; (2) For any h ∈ H,

−1 −1 −1 −1 Yσ(h, z) = Y(h, z) + (pσ h − ph)[0]0z + (p ¯σ h − ph¯ )[0]0z¯ ;

α (3) For any α ∈ H and v ∈ ΩF,H, X zn z¯n Y (v, z) = exp (pσ−1α − pα)[−n] + (p ¯σ−1α − p¯α)[−n] σ 0 n 0 n n≥1 −n −n X z z¯ −1 −1 Y(v, z) exp (pσ−1α − pα)[n] + (p ¯σ−1α − p¯α)[n] z(pσ α−pα)[0]0 z¯(p ¯σ α−p¯α)[0]0 ; 0 −n 0 −n n≥1

α Furthermore, for any t, t¯ ∈ R, the (L(0), L(0))-weight of v ∈ ΩF,H ∩ Ft,t¯ with respect to the full vertex operator algebra structure (F, Yσ, 1, ω, ω¯ ) is ! (pσ−1α, pσ−1α) − (pα, pα) (p ¯σ−1α, p¯σ−1α) − (p ¯α, p¯α) t + l l , t¯+ r r 2 2 and for any h−i : F → C such that hh(−n)−i for n ≥ 0, the four point correlation function αi for αi ∈ H and vi ∈ ΩF,H satisﬁes hY v z Y v z Y v z Y v z 1i σ( 1, 1) σ( 2, 2) σ( 3, 3) σ( 4, 4) −1 −1 (pσ αi,pσ α j)l−(pαi,pα j)l z − z z − z hY v z Y v z Y v z Y v z 1i = Π1≤i< j≤4 ( i j)(¯i ¯ j) ( 1, 1) ( 2, 2) ( 3, 3) ( 4, 4) . The family of full vertex operator algebras constructed in the above theorem is called the current-current deformation of a full H-vertex algebra in [Mo3]. We gives some remarks. First, if σ = id ∈ O(Hl ⊕ −Hr), then σα − α = 0 and Yσ(−, z) = Y(−, z).

Furthermore, if σ ∈ O(Hl)×O(Hr) ⊂ O(Hl ⊕−Hr), then σ commutes with the projection p −1 −1 and (pσ α, pσ α)l = (pα, pα)l, so nothing essentially changes. In fact, the isomorphic classes of the current-current deformation are parametrized by the double coset

(6.1) DF,H\O(Hl ⊕ −Hr)/O(Hl) × O(Hr), where DF,H is a subgroup of O(Hl ⊕ −Hr) introduced in [Mo3] and is called the duality group (for more detail see [Mo3, Theorem 5.5]).

It is noteworthy that since (zi − z j)(¯zi − z¯ j) ∈ R for any point of the conﬁguration space 1 4 X4 = {(z1, z2, z3, z4) ∈ (CP ) | zi , z j},

−1 −1 (pσ αi,pσ α j)l−(pαi,pα j)l (zi − z j)(¯zi − z¯ j) −1 −1 = exp (pσ αi, pσ α j)l − (pαi, pα j)l log (zi − z j)(¯zi − z¯ j) is a single-valued function on X4. 75

Remark 6.3. The lowest weight space ΩF,H of a full H-algebra i : MH,p ,→ F is similar to the lowest weight space S F for a framed full vertex operator algebra i : Ll,r(0) ,→ F.

As with the framed algebra structure on S F, ΩF,H inherits an algebra structure which is called a generalized full vertex algebra. This algebra structure plays a key role in the construction of the current-current deformation.

r 6.2. Current-current deformation of code CFTs. Let G ⊂ Z2 be a code such that r ⊥ ⊥ ⊥ 1 ∈ G and G the dual code. Set G [2] = {α ∈ G | |α|∆ = 2}. By Proposition 5.15, ⊥ G [2] determine the Lie algebra (FG)1,0 (FG)0,1. 1 N ⊥ i j Let α , . . . , α ∈ G [2] be the maximal mutually orthogonal codewords, i.e., |α α |∆ = i i r+r i i r+r 0 for any i , j and set αl = (α , 0) ∈ Z2 and αr = (0, α ) ∈ Z2 for i = 1, 2,..., N. Then, i i { ∈ ˆ } αl and αr are in CG. Let eα C[CG] α∈CG be the basis of the twisted group algebra given explicitly in Section 5. i ⊥ Then, by Proposition 4.24, each α ∈ G [2] deﬁnes subalgebras of FG which are isomorphic to the lattice vertex operator algebra V2Z and its conjugate V2Z. Furthermore,

(V2Z)1,0 is spanned by e i and (V2Z)0,1 is spanned by e i . αl αr |αi| l √ 2 ˆ i We note that by deﬁnition eαi · eαi = (−1) = −1 in C[CG]. Set h = −1eαi and √ l l l l i i − i { } hr = 1eαr for i = 1,..., N. Then, hl i=1,...,N satisfy i j (1) hl(n, −1)hl = 0 for any n ≥ 2 or n = 0; i j (2) hl(1, −1)hl = δi, j i N N and the similar result holds for {hr}i=1,...,N. Let Hl (resp. Hr ) be the subspace of FG i i N N spanned by {hl}i=1,2,...,N (resp. {hr}i=1,2,...,N) and deﬁne a bilinear form on Hl (resp. Hr ) i j i j N N N N by (hl, hl )l = δi, j (resp. (hr, hr)r = δi, j). Set H = Hl ⊕ Hr and let p ∈ EndH be the N projection onto Hl . Then, we have a injective homomorphism MHN ,p ,→ FG. We note that this embedding factors through

(6.2) MHN ,p ,→ V(2Z)N ⊗ V(2Z)N ,→ FG

We will show that MHN ,p ,→ FG is a full H-vertex operator algebra. Since (FG)t,t¯ = 0 if t < 0 or t¯ < 0, (FHO2) is clear. From an easy computation, (FHO3) follows. (FHO1) follows from (6.2) and the representation theory of a lattice vertex operator algebra (see for example [LL]). Hence, we have:

1 N ⊥ Proposition 6.4. If α , . . . , α ∈ G [2] are mutually orthogonal codewords, then MHN ,p ,→

FG is a full H-vertex operator algebra and the code CFT FG admits the current-current deformation parametrized by O(N, N)/O(N) × O(N).

In order to calculate the deformed four point correlation function, we need to decom- pose FG as an V(2Z)N ⊗ V(2Z)N -module. For the sake of simplicity, we ﬁrst consider the case 1 r 1 1 0 of N = 1 and set α = α = (1100 ... 0) ∈ Z2 and hl = hl , hr = hr . For , = ±, set

,0 0 S G = {hl · v = v, hr · v = v}. 76

Since hl · hl = 1 and hr · hr = 1 in C[CˆG] and S G is a C[CˆG]-module by Lemma 4.11,

M ,0 (6.3) S G = S G . ,0=± Recall that M FG = Lr,r(d, c) ⊗ (S G)d,c. (d,c) ¯ ¯ r+r r+r Let d = (d1, d2,..., dr, d1,..., dr) ∈ Z2 and c = (c1, c2,..., cr, c¯1,..., c¯r) ∈ Z2 satisfy ˆ ¯ ¯ r+r−4 (S G)d,c , 0. Set d = (d3, d4,..., dr, d3,..., dr) ∈ Z2 andc ˆ = (c3, c4,..., cr, c¯3,..., c¯r) ∈ r+r−4 ¯ ¯ Z2 . Then, by Proposition 4.10, |dαl|, |dαr| ∈ 2Z. In fact, by construction, (d1, d2, d1, d2) = (0, 0, 0, 0) or (1, 1, 1, 1).

Assume that (d1, d2, d¯1, d¯2) = (1, 1, 1, 1). Then, by the fusion rule, (S G)d,c is stable ,0 ,0 under the ·-multiplication by hl and hr. Set (S G)d,c = S G ∩ (S G)d,c. Then, (S G)d,c = L ,0 ,0 ,0=±(S G)d,c . It is easy to show that by [DGH, Lemma 3.1] (S G)d,c is a direct sum of the copies of V 1 ⊗ V 0 1 . In fact, 2Z+ 2 2Z+ 2 ,0 ,0 Lr,r(d, c) ⊗ (S G) V 1 ⊗ V 0 1 ⊗ Lr−2,r−2(dˆ, cˆ) ⊗ (S G) d,c 2Z+ 2 2Z+ 2 d,c as a V2Z ⊗ V2Z ⊗ Lr−2,r−2(0)-module.

Lemma 6.5. Assume that (d1, d2, d¯1, d¯2) = (1, 1, 1, 1). Then, t˜d = (1 + hl) · td satisﬁes

(1)h l[n]0t˜d = 0 = hr[n]0t˜d for any n ≥ 1;

(2)h l[0]0t˜d = t˜d = hr[0]0t˜d. 1 (h +h ) − 1 (h +h ) In particular, t˜ = (1 + h ) · t ∈ Ω 2 l r . Furthermore, (1 − h ) · t ∈ Ω 2 l r . d l d FG,Rhl⊕Rhr l d FG,Rhl⊕Rhr

Proof. (1) follows from the representation theory of the lattice vertex algebra V2Z. Since hl · hl = 1, we have hl · (1 + hl) · td = (hl + 1) · td. Similarly, we have hr · (1 + hl) · d td = (hr + hr · hl) · td = (hr − 1αl, αreαl+αr )td. Since αl + αr ∈ ∆ , by the deﬁnition of AG(d), we have hr · (1 + hl) · td = (hr − αl, αr)td. By deﬁnition of the two cocycle

(−, −), (αl, αr) = (αl, αl + ∆(1100 ... )) = (αl, αl)(αl, ∆(1100 ... )) = −1. Hence, hr · td = hr · eαl+αr · td = hl · td. Thus, hr · (1 + hl) · td = (1 + hl) · td. The case of (1 − hl) · td can be shown similarly.

Remark 6.6. Readers may wonder that the eigenvalues of hl(0, −1) and hr(−1, 0) on 1 0 1 V 1 ⊗ V 0 1 are 2Z + and 2Z + , however the eigenvalues of the ·-multiplication 2Z+ 2 2Z+ 2 2 2 of hl and hr on S G are ±. This diﬀerence happens because of the normalization (1.9) for 1 16 the intertwining operator I 1 1 (−, z). 2 , 16

4 We brieﬂy discuss in the case of (d1, d2, d¯1, d¯2) = (0, 0, 0, 0). There are 2 possibilities 4 for (c1, c2, c¯1, c¯2) ∈ Z2 and for the sake of simplicity we only consider the holomorphic part (c1, c2).

Assume that (c1, c2) is equal to (0, 1) or (1, 0). Then, (S G)02r,(01∗∗) ⊕ (S G)02r,(10∗∗) is stable +,∗ −,∗ under the ·-multiplication of hl and the S part (resp. S -part) corresponds to V 1 G G 2Z+ 2 (resp. V 1 ). 2Z− 2 77

Finally, assume that (c1, c2) is equal to (0, 0) or (1, 1). In this case, hl does not act as the zero-mode. More speciﬁcally, let v ∈ (S G)02r,(00∗∗). Then, hl(−1, −1)v is again in

(S G)02r,(11∗∗) and thus hl · v = hl(−1, −1)v. Similarly, for w ∈ (S G)02r,(11∗∗), hl(1, −1)w is in

(S G)02r,(00∗∗) and hl · w = hl(1, −1)w. By using Lemma 6.5, we can construct lowest weight vector for the action of the Heisenberg Lie algebra HˆN for any N > 0. Let α1, . . . , αN ∈ G⊥[2] be mutually or- N N i thogonal codewords. For s = (s1,..., sN) ∈ {±} , set h(s) = Πi=1(1 + sihl). Then, by 1 PN i i 2 i=1 si(hl+hr) r+r ∈ Lemma 6.5, h(s)t1 ΩF,HN .

Lemma 6.7. For s0, s1, s2, s3 ∈ {±}N,

0 1 2 3 hY h s · t r+r z Y h s · t r+r z Y h s · t r+r z Y h s · t r+r z 1i ( ( ) 1 , 0) ( ( ) 1 , 1) ( ( ) 1 , 2) ( ( ) 1 , 3) 1 (pi,p j)+ 1 N− r −r+3N r−2N 4 4 8 = 2 δs0+s1+s2+s3,0F(z0, z1, z2, z3) Π0≤i< j≤3 (zi − z j)(¯zi − z¯ j) .

Proof. We ﬁrst consider the case of s0 = s1 = s2 = s3 = +N. By dividing N into 8 regions, by Proposition 5.12, we have

0 1 2 3 Y h s · t r+r z Y h s · t r+r z Y h s · t r+r z Y h s · t r+r z 1i ( ( ) 1 , 0) ( ( ) 1 , 1) ( ( ) 1 , 2) ( ( ) 1 , 3) − r X −r 8 r N! = 2 Π0≤i< j≤3 (zi − z j)(¯zi − z¯ j) F(z0, z1, z2, z3) k1!k2! ··· k8! k1,k2,...,k8 G (z , z , z , z )!2k1+2k4 G (z , z , z , z )!2k2+2k5 G (z , z , z , z )!2k3+2k6 × − 01,23 0 1 2 3 − 02,13 0 1 2 3 − 03,12 0 1 2 3 F(z0, z1, z2, z3) F(z0, z1, z2, z3) F(z0, z1, z2, z3)  2 2 2 N − r  G + G + G  = 2−rΠ (z − z )(¯z − z¯ ) 8 F(z , z , z , z )r 2 − 2 01,23 02,13 03,12  0≤i< j≤3 i j i j 0 1 2 3  F2 

2 2 2 2 Since G01,23 + G02,13 + G03,12 = F , the four point correlation function is equal to zero. We next consider the case of s0 = s1 = s2 = +N and s3 = −N. Then, the four point correlation function is

− r X N! −r 8 r k3+k4+k5+k7 2 Π0≤i< j≤3 (zi − z j)(¯zi − z¯ j) F(z0, z1, z2, z3) (−1) k1!k2! ··· k8! k1,k2,...,k8 G (z , z , z , z )!2k1+2k4 G (z , z , z , z )!2k2+2k5 G (z , z , z , z )!2k3+2k6 × − 01,23 0 1 2 3 − 02,13 0 1 2 3 − 03,12 0 1 2 3 F(z0, z1, z2, z3) F(z0, z1, z2, z3) F(z0, z1, z2, z3) = 0. 78

Finally, we consider the case of s0 = s1 = +N and s2 = s3 = −N. Then, the four point correlation function is − r X −r 8 r N! 2 Π0≤i< j≤3 (zi − z j)(¯zi − z¯ j) F(z0, z1, z2, z3) k1!k2! ··· k8! k1,k2,...,k8 G (z , z , z , z )!2k1+2k4 G (z , z , z , z )!2k2+2k5 G (z , z , z , z )!2k3+2k6 × − 01,23 0 1 2 3 + 02,13 0 1 2 3 + 03,12 0 1 2 3 F(z0, z1, z2, z3) F(z0, z1, z2, z3) F(z0, z1, z2, z3)  2 2 2 N − r  G − G − G  = 2−rΠ (z − z )(¯z − z¯ ) 8 F(z , z , z , z )r 2 − 2 01,23 02,13 03,12  0≤i< j≤3 i j i j 0 1 2 3  F2  − r −r+3N 8 r−2N 1 N = 2 Π0≤i< j≤3 (zi − z j)(¯zi − z¯ j) F(z0, z1, z2, z3) ((z0 − z1)(z2 − z3)(¯z0 − z¯1)(¯z2 − z¯3)) 2 It is easy to derive the general formula from the above computations. By Lemma 6.5, Lemma 6.7 and Theorem 6.2, we have:

Theorem 6.8. Assume that there exists mutually orthogonal codewords α1, . . . , αN ∈ ⊥ G [2]. Then, the code CFT FG admits a current-current deformation parametrized by O(N, N)/O(N) × O(N). Furthermore, for σ ∈ O(N, N) and s0, s1, s2, s3 ∈ {±}N, the deformed four point function satisﬁes

0 1 1 3 hY h s · t r+r z Y h s · t r+r z Y h s · t r+r z Y h s · t r+r z 1i σ( ( ) 1 , 0) σ( ( ) 1 , 1) ( ( ) 1 , 2) σ( ( ) 1 , 3) 1 pσ−1(si,si),pσ−1(s j,s j) + 1 N− r −r+3N r−2N 4 ( )l 4 8 = 2 δp0+p1+p2+p3,0F(z0, z1, z2, z3) Π0≤i< j≤3 (zi − z j)(¯zi − z¯ j) .

7. Appendix

Let l, r ∈ Z≥0. In Appendix, we will deﬁne the category of (l, r)-framed algebras and prove that it is equivalent to the category of (l, r)-framed full vertex operator algebras.

The symmetric group S l (reps. S r) naturally acts on the left component (resp. the right component) of IS(l,r) = ISl × ISr. Similarly to the case of framed full vertex operator algebras, a morphism from a (l, r)-framed algebra S to T is a pair of a permutation (g, g¯) ∈

S l × S r and a linear map f : S → T such that: (l,r) (1) For any µ ∈ IS , f (S µ) = T(g,g¯)µ. (2) f (1) = 1; (3) f (a · b) = f (a) · f (b). We denote the category of framed algebras by Framed algebras. For any framed full vertex operator algebra F, by Lemma 2.15 and Proposition 4.1 S F is a framed algebra. It is easy to show that this correspondence

S : Framed full VOAs → Framed algebras, F 7→ S F gives a functor. Conversely, for any framed algebra S , by Theorem 5.7, FS is a framed full vertex operator algebra. Let ( f, (g, g¯)) : S → T be a morphism between framed algebras

S and T. Then, we have a unique linear map f˜ : FS → FT such that: (1) the restriction of f˜ on S coincides with f ; 79

(2) f˜ is a Virl × Virr-module homomorphism. By using Lemma 3.14, we can show that f˜ is a full vertex algebra homomorphism. Hence, the correspondence

F : Framed algebras → Framed full VOAs, S 7→ FS also gives a functor. Then, we have:

Theorem 7.1. The functor F : Framed algebras → Framed full VOAs gives an equivalence of categories. The inverse functor is given by S : Framed full VOAs → Framed algebras.

References

[B] R.E. Borcherds, Vertex algebras, Kac-Moody algebras, and the Monster, Proc. Nat. Acad. Sci. U.S.A., 83, 1986, (10), 3068–3071. [BPZ] A. A. Belavin, A. M. Polyakov, A. B. Zamolodchikov, Infinite conformal symmetry in two- dimensional quantum field theory, Nuclear Phys. B, 241, 1984, (2), 333–380. [CS] S. Chaudhuri and J. A. Schwartz, A criterion for integrably marginal operators, Phys. Lett. B, 219, 1989, (2)-(3), 291–296. [DGH] C. Dong, R.L. Griess and G. Hohn,¨ Framed vertex operator algebras, codes and the Moonshine module, Comm. Math. Phys., 193, 1998, (2), 407–448. [DLM] C. Dong, H. Li and G. Mason, Regularity of rational vertex operator algebras, Adv. Math., 132, 1997, (1), 148–166. [DM] C. Dong and G. Mason, Holomorphic vertex operator algebras of small central charge, Pacific J. Math., 213, 2004, (2), 253–266. [DMZ] C. Dong, G. Mason and Y. Zhu, Discrete series of the Virasoro algebra and the moonshine module, Algebraic groups and their generalizations: quantum and infinite-dimensional methods (University Park, PA, 1991), Proc. Sympos. Pure Math., 56, 295–316. [DVV1] R. Dijkgraaf, E. Verlinde, and H. Verlinde, On moduli spaces of conformal field theories with c ≥ 1, Perspectives in string theory (Copenhagen, 1987), 1988, 117–137. [DVV2] R. Dijkgraaf, E. Verlinde, and H. Verlinde, C = 1 conformal field theories on Riemann surfaces, Comm. Math. Phys., 115, 1988, (4), 649–690. [FLM] I. Frenkel, J. Lepowsky, and A. Meurman, Vertex operator algebras and the Monster, Pure and Applied Mathematics, 134, Academic Press, Inc., Boston, MA, 1988. [FR]S.F orste¨ and D. Roggenkamp, Current-current deformations of conformal field theories, and WZW models, J. High Energy Phys., 2003, 5. [FHL] I. Frenkel, Y. Huang and J. Lepowsky, On axiomatic approaches to vertex operator algebras and modules, Mem. Amer. Math. Soc., 104, 1993, (494). [FMS] P. Di Francesco, P. Mathieu and D. Sen´ echal,´ Conformal field theory, Graduate Texts in Contempo- rary Physics, Springer-Verlag, New York, 1997. [FRS] J. Fuchs, I. Runkel and C. Schweigert, Conformal correlation functions, Frobenius algebras and triangulations, Nucl. Phys. 624 2002, 452–468. [FZ] I. Frenkel and Y. Zhu, Vertex operator algebras associated to representations of affine and Virasoro algebras, Duke Math. J., 66, 1992, (1), 123–168. [FF] B.L. Fe˘ıgin and D.B. Fuchs, Cohomology of some nilpotent subalgebras of the Virasoro and Kac- Moody Lie algebras, J. Geom. Phys., 5, 1988, (2), 209–235. [Gi] P. Ginsparg, Curiosities at c = 1, Nuclear Phys. B, 295, 1988, (2), FS21, 153–170. 80

[Hu1] 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. [Hu2] Y.-Z. Huang, Virasoro vertex operator algebras, (nonmeromorphic) operator product expansion and the tensor product theory, J. Alg., 182, (1996), 201–234. [HK] Y.-Z. Huang, L. Kong, Full field algebras, Comm. Math. Phys., 272, 2007, (2), 345–396. [HL1] Y.-Z. Huang and J. Lepowsky, A theory of tensor products for module categories for a vertex operator algebra, I, Selecta Mathematica (New Series), 1, 1995, 699–756. [HL2] Y.-Z. Huang and J. Lepowsky, A theory of tensor products for module categories for a vertex operator algebra, II, Selecta Mathematica (New Series), 1, 1995, 757–786. [HL3] 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. [IK] K. Iohara and Y. Koga, Representation theory of the Virasoro algebra, Springer Monographs in Mathematics, 2011. [La] C.H. Lam, On the constructions of holomorphic vertex operator algebras of central charge 24, Comm. Math. Phys., 305, 2011, (1), 153–198. [LL] J. Lepowsky and H. Li, Introduction to vertex operator algebras and their representations, Progress in Mathematics, 227, Birkhauser¨ Boston, Inc., Boston, MA, 2004. [LS1] C. Lam and H. Shimakura, Quadratic spaces and holomorphic framed vertex operator algebras of central charge 24, Proc. Lond. Math. Soc. (3), 104, 2012, (3), 540–576. [LS2] C. Lam and H. Shimakura, Classification of holomorphic framed vertex operator algebras of central charge 24, Amer. J. Math., 137, 2015, (1), 111–137. [LS3] C. Lam and H. Shimakura, 71 holomorphic vertex operator algebras of central charge 24, Bull. Inst. Math. Acad. Sin. (N.S.), 14, 2019, (1), 87–118. [LY] C. Lam and H. Yamauchi, On the structure of framed vertex operator algebras and their pointwise frame stabilizers, Comm. Math. Phys., 277, 2008, (1), 237–285. [Mo1] Y. Moriwaki, Genus of vertex algebras and mass formula, Math. Z, 2021, doi:10.1007/s00209-021- 02702-0, arXiv:2004.01441 [q-alg]. [Mo2] Y. Moriwaki, Full vertex algebra and bootstrap – consistency of four point functions in 2d CFT, arXiv:2006.15859 [q-alg]. [Mo3] Y. Moriwaki, Two-dimensional conformal field theory, current-current deformation and mass formula, arXiv:2007.07327 [q-alg]. [Mi1] M. Miyamoto, A new construction of the Moonshine vertex operator algebra over the real number field, Ann. of Math. (2), 159, 2004, (2), 535–596. [Mi2] M. Miyamoto, Representation theory of code vertex operator algebra, J. Algebra, 201, 1998, (1), 115–150. [MS] G. Moore and N. Seiberg, Classical and quantum conformal field theory, Comm. Math. Phys. 123, 1989, 177–254. [P] J. Polchinski, String theory. Vol. I, Cambridge Monographs on Mathematical Physics, Cambridge University Press, Cambridge, 1998. [S] A. N. Schellekens, Meromorphic c = 24 conformal field theories, Comm. Math. Phys., 153, 1993, (1), 159–185. [W] W. Wang, Rationality of Virasoro vertex operator algebras, Internat. Math. Res. Notices, 1993, (7), 197–211.