The theory of C2-cofinite vertex operator algebras
Thomas Creutzig and Terry Gannon
Abstract Despite several efforts, the theory of rational VOAs is still much more complete than that of C2-cofinite ones. We address this by focusing on the analogue of the categorical Verlinde formula and its relation to the modular S-matrix. We explain that the role of the Hopf link is played by the open Hopf link, which defines a representation of the fusion ring on each End(W ). We extract numbers, called logarithmic Hopf link invariants, from those operators. From this a Verlinde formula can be obtained that depends on the endomorphism ring structure. We apply our ideas directly to the triplet algebras Wp. Using the restricted quantum group of sl2, we compute all open Hopf link operators and the logarithmic Hopf link invariants and compare them successfully with the modular S- matrix of Wp. In particular, S-matrix coefficients involving pseudo-trace functions relate to logarithmic Hopf link invariants with nilpotent endomorphism insertion. We show how the S-matrix specifies the fusion ring (not merely the Grothendieck ring) uniquely.
1 The representation theory of nice VOAs
For reasons of simplicity, we will restrict attention in this paper to vertex operator algebras `∞ (VOAs) V of CFT-type — this means V has L0-grading V = n=0 Vn with V0 = C1 and dim Vn < ∞. We also require V, when regarded as a V-module in the usual way, to be simple and isomorphic to its contragredient V∨. By Modgr(V) we mean the category of grading-restricted weak V-modules M, by which we mean L0 decomposes M into a direct sum of finite-dimensional generalized eigenspaces with eigenvalues bounded from below. When L0 is in fact diagonalizable over M, M is called ordinary. This terminology is explained in more detail in Section 2 below. For convenience we will refer to any grading-restricted weak V-module, simply as a V-module.
1.1 The rational modular story Let V be a strongly-rational vertex operator algebra — i.e. regular, in addition to the aforementioned restrictions. The theory of strongly-rational VOAs is well-understood. Such a VOA enjoys three fundamental properties: (Cat) its category of modules Modgr(V) forms a modular tensor category [T, H2]; (Mod) its 1-point toroidal functions are modular [Z]; and (Ver) Verlinde’s formula holds [V, MS, H1].
1 For strongly-rational V, all M ∈ Modgr(V) are ordinary. A strongly-rational V has finitely many simple modules in Modgr(V), up to isomorphism. Fix representatives M0,M1,...,Mn of these classes, with M0 = V. Every (ordinary) V-module will be isomor- k phic to a unique direct sum of these Mj. The fusion coefficients Nij ∈ Z≥0 are defined as the dimension of the space of intertwiners of type Mk . Tensor products of modules Mi Mj can be defined (see e.g. [HL] and references therein), realizing those fusion coefficients:
n ∼ M k Mi ⊗ Mj = Nij Mk . k=0
This defines a ring structure (the tensor ring or fusion ring) on the formal Z-span of the representatives Mj, with unit M0. For each i, define the fusion matrix Ni by (Ni)jk = k Nij . They yield a representation of the tensor ring:
X k Ni Nj = Nij Nk . k
k We also have Nij = dim(HomV (Mi ⊗ Mj,Mk)). Given a V-module M, the 1-point function FM on the torus is
c L0− FM (τ, v) := trM o(v) q 24 , (1) for any τ in the upper half-plane H and v ∈ V, where o(v) ∈ End(M) is the zero-mode. The character of a module is ch[M](τ) = FM (τ, 1). Zhu proved [Z] that for v ∈ V[k] (using the L[0]-grading introduced in [Z]), these 1-point functions form the components of a vector-valued modular form of weight k, holomorphic in H and meromorphic at the cusps, for the modular group SL(2,Z). Hence the characters form a vector-valued modular form of weight 0 for SL(2, Z). (The reason for preferring the more general 1-point functions over the more familiar characters is that the latter are usually not linearly independent.) In particular, the transformation τ 7→ −1/τ is called the modular S-transformation and defines a matrix Sχ via n k X χ FMi (−1/τ, v) = τ Sij FMj (τ, v) . j=0
The eigenvalues of L0 on a simple V-module Mj lie in hj + Z≥0 where hj is called the conformal weight of Mj. [AnM] for instance show that these hj, together with the central charge c, are rational, using the modularity of the characters. k In 1988, Verlinde [V] proposed that these so-called fusion coefficients Nij are related to the matrix Sχ by what is now called Verlinde’s formula:
n χ χ χ −1 X Si`Sj` S N k = `k . (2) ij Sχ `=0 0` This is one of the most exciting outcomes of the mathematics of rational conformal field χ χ −1 theory (CFT). As S is a unitary matrix, (S )`k here can be replaced with the complex χ ∗ conjugate Sk` . The Verlinde formula thus has three aspects:
2 ⊗ (V1) There is a matrix S simultaneously diagonalizing all fusion matrices Ni. Each ⊗ −1 ⊗ diagonal entry ρl(Mi) := (S NiS )ll defines a one-dimensional representation
X k ρl(Mi)ρl(Mj) = Nij ρl(Mk) k
of the tensor ring. All of these ρl are distinct. (V2) The Hopf link invariants
m M ∈ Sij := Mi j C
for any 0 ≤ i, j ≤ n give one-dimensional representations of the fusion ring:
m m n m S Sj` X S i` = N k k` . S m S m ij S m 0` 0` k=0 0`
m m Moreover, each representation ρl appearing in (V1) equals one of these Mi 7→ Si`/S0`. (V3) The deepest fact: these three S-matrices are essentially the same. More precisely, χ χ m ⊗ χ ⊗ m m Sij/S00 = Sij, and S = S works in (V1). For this choice of S , ρl(Mi) = Sil /S0l. Much of (V1)-(V3) is automatic in any modular tensor category. In particular, Theo- rem 4.5.2 in [T] says that in any such category,
m m m ∗ X Si`Sj`Sk` N k = D−2 , ij S m ` 0`
2 P m 2 χ −1 m where D = i S0i . So the remaining content of (2) is that S = D S . Categorically, ∼ the space of 1-point functions FM is identified with the space Hom(1, H) where H = ∨ ∨ ⊕iMi ⊗ Mi (throughout this paper, M denotes the contragredient or dual of M). This object H is naturally a Hopf algebra and a Frobenius algebra. Using this structure, Section 6 of [Ly1] shows End(H) carries a projective action of SL(2, Z); it is the coend of [Ly1], and [Sh] suggests to interpret it (or its dual) as the adjoint algebra of the category. The SL(2, Z)-action on the FM correspond to the subrepresentation on Hom(1, H), which is generated by S-matrix D−1S m and T -matrix coming from the ribbon twist. The other subrepresentations Hom(Mj, H) correspond to the projective SL(2, Z)-representations on the space of 1-point functions with insertions v ∈ Mj, where the vertex operator YM (v, z) M0 implicit in (1) is replaced with an intertwining operator of type ∨ . As in [FS], the Mi Mi character FM of (1) can be interpreted categorically as the partial trace of the ‘adjoint’ action H ⊗ M → M. It turns out to be far easier to construct modular tensor categories, than to construct VOAs or CFTs. This is related to the fact that the subfactor picture is essentially equiva- lent to the categorical one, at least if one assumes unitarity. This can be used to probe just how complete the lists of known (strongly-rational) VOAs and CFTs are. More precisely, given a fusion category (essentially a modular tensor category without the braiding), tak- ing its Drinfeld double or centre construction yields a modular tensor category. Fusion
3 categories can be constructed and classified relatively easily, and their doubles worked out, at least when their fusion rules are relatively simple. Examples of this strategy are provided in e.g. [EG1]. This body of work suggests that the zoo of known modular tensor categories is quite incomplete, and hence that the zoo of known strongly-rational VOAs and CFTs is likewise incomplete.
1.2 The finite logarithmic story An important challenge is to extend the aforementioned results beyond the semi-simplicity of the associated tensor categories. A VOA or CFT is called logarithmic if at least one of its modules is indecomposable but reducible. The name refers to logarithmic singularities appearing in their correlation functions and operator product algebras of intertwining operators. In this paper we are interested in logarithmic VOAs with only finitely many simple modules. Of these, the best studied class is the family of Wp-triplet algebras parameterized by p ∈ Z>1 and this will be our main example too. See e.g. [CR4, FS] for introductions on logarithmic VOAs. By a strongly-finite VOA, we mean a simple C2-cofinite VOA of CFT-type with V isomorphic to V∨. These have finitely many simple modules, though they may have uncountably many indecomposable ones (see Corollary 5.2 below). Any strongly-rational VOA is strongly-finite. Gaberdiel–Goddard have an interesting conjecture that a strongly- finite VOA is strongly-rational iff Zhu’s algebra A0(V) is semi-simple. This subsection reviews what is known about them. See Section 2.2 below for more details. The key hypothesis here is C2-cofiniteness; many of the following aspects will persist even when simplicity or CFT-type is lost. 2 The Wp models are strongly-finite. They have central charge c = 1 − 6(p − 1) /p, and are generated by the conformal vector and 3 other states. The symplectic fermions form a logarithmic vertex operator superalgebra with c = −2d for any d ∈ Z>0 (the number of + pairs of fermions); their even part SFd is a strongly-finite VOA [Ab]. The Wp,p0 -models are C2-cofinite but not simple, so are not strongly-finite. (Cat)0 A tensor product theory for strongly-finite VOAs has been developed by Huang, Lepowsky and Zhang (see e.g. [HLZ] and references therein), see also Miyamoto [Miy2]. The corresponding tensor category Modgr(V) is braided. Rigidity of this category is proven so far only for the Wp models [TW] and symplectic fermions + SFd [DR]. The notion of ordinary V-module was generalized to logarithmic modules (equivalent to what we call grading-restricted weak modules) in [Mil]; see Section 2.2. The space W of intertwiners of type UV , for any V-modules U, V, W , is isomorphic to HomV (U ⊗ V,W ), where the notion of intertwiner must be generalized appropriately [Mil]. In this W nonsemi-simple setting though, we will in general only have the inequality NU,V ≤ dim(HomV (U ⊗ V,W )). [FS] suggest that the generalization of modular tensor category appropriate to strongly-finite VOAs is what we will call a log-modular tensor category and define in Section 1.4, though the notion was first introduced in [KL]. The category of course can no longer be semi-simple, but the subtlety in its definition is the precise form nondegeneracy takes. Miyamoto [Miy3] proposed a weakened (and more complicated) notion of rigidity (‘semi-rigidity’) for nonsemi-simple VOA contexts, but the Wp-models
4 obey the usual rigidity and the nonsemi-simple finite tensor categories [EO] of Etingof– Ostrik have the usual rigidity built in. We expect strongly-finite VOAs to obey rigidity, though of course it may be easier to prove semi-rigidity. Fix representatives M0,...,Mn of isomorphism classes of simple V-modules as before. Let Pi be their projective covers (defined in Section 2.2). Likewise fix representatives of gr isomorphism classes Mλ of indecomposable V-modules. Any V-module M ∈ Mod (V) is isomorphic to a direct sum of finitely many Mλ, in a unique way. The tensor product defines a ring structure Fusfull(V), called the (full) tensor ring, on the formal span over gr Z of the Mλ. This ring is large – Mod (V) can be expected to have continuous families of indecomposables. There are two smaller versions of Fusfull(V). The smallest is the Grothendieck ring FusGr(V), by which is meant the quotient of the tensor ring Fusfull(V) by the formal differences M − M 0 − M 00 for each short exact sequence 0 → M 0 → M → M 00 → 0 in Modgr(V). We will write [M] for the image of a V-module M in the Grothendieck ring, and +, ⊗g for the sum, product in the Grothendieck ring. The Grothendieck ring has gr P basis [M0],..., [Mn] over Z. Explicitly, for any module M ∈ Mod (V), [M] = i mi[Mi] where mi ∈ Z≥0 are the Jordan–H¨older multiplicities of M. Finally, we have the subring simp full Fus (V) of Fus (V) generated by the simple V-modules Mi and their projective covers Pi. In the special case where V is strongly-rational, the Grothendieck ring, full tensor ring and simple-projective subring all coincide. We prefer to avoid calling any of these the fusion ring, as the term ‘fusion ring’ is used in different senses in the literature. The 1-point functions (1) can still be defined, but as we explain in Section 2.2 will P P satisfy FM = i miFMi for any logarithmic module M, where [M] = i mi[Mi]. The
problem is that the C-span of characters ch[Mi](τ) = FMi (τ, 1) is no longer SL(2, Z)- aτ+b invariant, although we still have that each ch[Mi] cτ+d lies in the C[τ]-span of the characters. Many authors interpret this as saying that e.g. Sχ is now τ-dependent. We prefer to say that the ordinary trace functions (1) must be augmented by pseudo-trace functions associated to reducible V-modules. The main result here is due to Miyamoto [Miy1]:
0 (Mod) The C-span of ordinary trace and pseudo-trace functions (see Section 2.2 for more details) is SL(2, Z)-invariant. Corollary 5.10 of [Miy1] uses this to argue that conformal weights and central charges in strongly-finite VOAs must also be rational. The dimension of the resulting SL(2, Z)- representation is dimAm/[Am,Am] − dimAm−1/[Am−1,Am−1], where Ak = Ak(V) is the kth Zhu algebra (a finite-dimensional associative algebra), and m is sufficiently large. It is possible to state explicitly how large m must be, but this isn’t terribly useful as the algebras Ak(V) are very hard to identify in practise. There have been several proposals for a strongly-finite Verlinde formula beyond ra- tionality, mainly in the physics literature, see [R, GabR, FHST]. All these proposals are guided by analogy to the rational setting and they do not connect to the tensor category point of view (recall that equation (2) is a theorem in any modular tensor category, when χ −1 m S there is replaced with D S ). Define the tensor resp. Grothendieck matrices Nλ ν k resp. [N ]i, with entries (Nλ)µν = Nλµ and ([N ]i)jk = [N ]ij defined by the structure
5 constants
M ν X k Mλ ⊗ Mµ = Nλµ ·Mν , [Mi] ⊗g [Mj] = [N ]ij [Mk] . ν j
As before, they define representations of the simple-projective tensor subring respectively Grothendieck ring.
0 ⊗ ⊗g (V1) There are matrices S resp. [S ] which simultaneously put the Nλ resp. [N ]i into block diagonal form. These blocks define indecomposable representations of the tensor resp. Grothendieck rings.
0 (V1) is not deep, and follows because the matrices Nλ resp. [N ]i pairwise commute. In [R], both the tensor matrices Nλ and the Grothendieck matrices [N ]i for the Wp models have been explicitly block-diagonalized: the latter involves only blocks of size 1 and 2 [PRR], while those of the former involve also blocks of size 3 [R]. We also should block- diagonalise the full tensor ring Fusfull(V), in the sense of [CR1],[CR2]. To our knowledge this has never been explored. To our knowledge, no analogue of (V2)0 has been explored. The Hopf link invariants of (V2) are still defined, but vanish on projective modules (at least if the category is rigid). If they did not vanish they would no longer yield tensor or Grothendieck ring representations. The reason for this is that dim End(Mλ) can be > 1. Nothing general about (V3)0 is known or has been explicitly conjectured. However, Lyubashenko [Ly1] explained that the modular group acts also on certain nonsemi-simple braided tensor categories. [FGST1] observed that the SL(2, Z) action on the center of the restricted quantum group Uq(sl2) of sl2 at 2p-th root of unity q, coincides with the one on the space of trace and pseudo-trace functions of the Wp-triplet algebra. The categorical SL(2, Z)-action discussed last subsection extends here as follows. Write C = Modgr(V). According to [Sh], the coend/Hopf algebra/Frobenius algebra/adjoint algebra H is simply UR(V), where U : Z(C) → C is the forgetful functor from the Drinfeld centre of C, and R is its right adjoint. The space of 1-point functions on the torus is HomV (V, H). The ordinary trace functions (1) span a subspace (of dimension equal to the number of simple V-modules).
1.3 Our observations We propose a general answer to (V2)0, using what we call logarithmic Hopf link invariants. Very much like ordinary Hopf link invariants in a (semi-simple) modular tensor category immediately imply the Verlinde formula [T], the logarithmic Hopf link invariants also define a logarithmic Verlinde formula for structure constants of the tensor or Grothendieck ring provided that a certain S-matrix is invertible. Before detailing this idea below, let us provide some historical background. One of us (TC) has several collaborations with Antun Milas, David Ridout and Simon Wood on logarithmic VOAs with infinitely many simple objects. These VOAs have the advantage over the C2-cofinite ones that the generic indecomposable module is simple and there is a natural Verlinde formula, which we have verified to a certain extent in the cases of the affine vertex algebra of sl2 at admissibible but non-integer level [CR1, CR2], the affine vertex super algebra of gl(1|1) [CR3] and its extensions [AC] as well as the
6 singlet algebra [CM1, RW1]. But there has not yet been any connection made to the categorical picture. Since the non-generic singlet algebra module characters are partial theta functions, they have no nice modular S-transformation. This forced us in [CM1] to regularize characters and then the modular transformation had a correction term. In the rational setting (at least when there is a unique Mi with lowest conformal weight), the Verlinde formula implies that the map
ch[M](τ) C → C,M 7→ lim , τ→0 ch[M0](τ) where M0 is the vacuum module, is a representation of the fusion ring. In the setting studied in [CM1], this asymptotic dimension depended on the regime of the regularization parameter. If the correction term dominated, then the asymptotic dimensions seemed to correspond to the semi-simplification of the fusion ring. This picture was further elaborated in [CMW], where the new insight is that in these non-finite logarithmic theories the asymptotic dimensions of characters conjecturally define representations of the tensor ring and they contain detailed information on that ring and its quotient structure. Higher rank analogues are currently being studied [CM2] and reviews on our ideas are given in [CR4, RW2, C]. A natural question is thus: Can we give a categorical interpretation for these representations of the fusion ring? Our answer, in the context of strongly-finite VOAs, is discussed next. Let C be a finite rigid and braided but nonsemi-simple tensor category (see Section 2.1 for the definitions). For each object W , define the map
ΦW : C → End(W ),U 7→ ΦU,W , where ΦU,W is the open Hopf link operator
ΦU,W = U (3)
W
As we prove in Theorem 2.2 below, naturality of the braiding implies that ΦU,W defines a representation of the tensor ring in the finite-dimensional algebra End(W ): ΦU⊗V,W = full ΦU,W ◦ ΦV,W . We would expect that the representation U 7→ ΦU,W of Fus (V) is indecomposable iff W is. One might also expect that Φ?,W1 and Φ?,W2 define equivalent full Fus (V)-representations iff Wi are equivalent V-modules; we however find that if W1 and W2 are projective modules of the same extension block then also Φ?,W1 and Φ?,W2 define equivalent Fusfull(V)-representations. We show in Section 4 that all the blocks appearing in Rasmussen’s block-diagonalization of the Grothendieck matrices for Wp, come from these ΦU,W . We have not completed the analysis for the 3 × 3 blocks appearing in his decomposition of the tensor ring for Wp. We expect though that the analogous statement holds for the tensor and Grothendieck matrices for any strongly-finite VOA, as well as the indecomposable subrepresentations of Fusfull(V). Together, this would provide the full analogue of (V2) in the strongly-finite context.
7 The big question of course is to relate these open Hopf link operators with the modu- larity of pseudo-trace functions. It is tempting to guess that for projective V-modules M, there is a natural isomorphism between End(M) and the space of pseudo-traces on M. Now suppose the category C contains a subcategory P that is a tensor ideal closed under retract. Further, we want this subcategory to be semi-simple. Incidentally, the following discussion can easily be generalized to tensor categories with infinitely many simple objects as well as to more than just one such ideal. The key requirement is that we also want P to be generated by an ambidextrous element so that there exists a modified trace t on P. This idea of modified trace has been developed by Geer, Patureau-Mirand and collaborators [GKP1, CGP, GKP2, GPT, GPV]. If W is not simple, then its endomorphism ring is not one-dimensional (but since C is (1) (n) finite, it will be finite-dimensional). Fix a basis {xw , . . . , xw } of the endomorphism ring of W . The map ΦU,W can then be expanded as
n X (i) (i) ΦU,W = ΦU,W xw , i=1 and these coefficients satisfy
n X ijk (i) (j) X X (k) aW ΦU,W ΦV,W = NUV ΦX,W i,j=1 X∈C for every k = 1, . . . , n. Here the structure constants of the endomorphism ring of W and the tensor ring of C are defined via
n (i) (j) X ijk (k) M X xw ◦ xw = aW xw ,U ⊗ V = NUV X. k=1 X∈C
How can we relate all this to modular properties of a C2-cofinite VOA? Let’s take the modified trace of W assuming that W is in P. Then we define the logarithmic Hopf link invariant to be n n (j) xw (j) X (i) (i) (j) X (i) (i) (j) SU,W := tW ΦU,W xw = ΦU,W tw x ◦ x = ΦU,W tw Φ1,W x ◦ x . i=1 i=1 Here 1 is the identity in C. In other words, the logarithmic Hopf link invariants relate to our expansion coefficients. A really nice situation for the VOA case would be if the pseudo-trace c L0− (j) trW q 24 x transforms under τ 7→ −1/τ into a linear combination of true characters and the expansion coefficients (the S-matrix coefficients) have a very direct relation (equality up to overall normalization) to the logarithmic Hopf link invariants. In this article, we will see that this really nice situation works for Wp. Since the triplet algebra Wp is the only fairly well-understood strongly-finite VOA that is not strongly-rational, this is the natural example to test our ideas and conjectures. gr Although Mod (Wp) is braided and rigid, the braiding is not explicitly known so that we have to take an indirect route, through the related quantum group Uq(sl2).
8 In a nutshell, we have the following results. In section three, we compute the open Hopf link invariants ΦV,W of Uq(sl2) in two ways: first following the strategy of [CGP], which uses as input the known tensor products; and secondly using the ideas on logarithmic knot invariants of Murakami and Nagatomo [Mur, MN]. The result is given in Theorem 3.8. The logarithmic Hopf link invariants give us a Verlinde formula for certain structure constants of the tensor ring. This is done in subsection 3.5. Although this doesn’t directly yield all tensor coefficients, Theorem 4.3 shows that the logarithmic Hopf link invariants together with little additional information completely determine the tensor ring. In Section 4, we compare to the modular S-matrix of the Wp triplet algebra and find perfect agreement with the logarithmic Hopf link invariants, see Theorem 4.4. In particu- lar, logarithmic Hopf link invariants with nilpotent endomorphism insertion correspond to S-matrix coefficients involving pseudo-trace functions, while those with identity insertion correspond to the weight zero part of the characters. We also compare there the Jordan blocks of [R] with the open Hopf link operators, for Wp. We thus see that the logarithmic Hopf link invariants correspond to the S-matrix of an SL(2, Z)-action, analogous to the Hopf link invariants in (semi-simple) modular tensor categories. It would be interesting future work to connect this to Lyubashenko [Ly1] as well as giving an independent proof of the SL(2, Z)-action using logarithmic Hopf link invariants if possible. A more important observation is the nice matching of S-matrix co- efficients, especially that S-matrix coefficients involving the pseudo-trace functions relate to logarithmic Hopf link invariants with nilpotent endomorphism insertion. We believe that this is no coincidence and our hope is that one can prove a theorem as above for C2-cofinite VOAs whose tensor category is rigid and which have an ideal possessing a modified trace. In Section 5 we give several general results...
1.4 Questions and optimistic guesses
Let V be a strongly-finite VOA, i.e. a simple C2-cofinite VOA of CFT-type with V isomorphic to V∨. Based on the preceding remarks, one could hope that the following are true, at least in broadstrokes, though surely some details will have to be modified. It is perhaps premature to make precise conjectures, but we propose the following as a way to help guide future work.
Question. We need more examples of C2-cofinite VOAs! Most fundamental is the question as to what replaces the role of modular tensor category for strongly-finite VOAs.
Definition 1.1. A log-modular tensor category C is a finite tensor category which is ribbon, whose double is isomorphic to the Deligne product C ⊗ Copp.
These terms are defined in Section 2.1. Modular tensor categories C are precisely the ribbon fusion categories whose double is isomorphic to C⊗Copp, and a finite tensor category is the natural nonsemi-simple generalization of a fusion category, so this is the natural definition. A log-modular tensor category is meant to be the category of V-modules for some strongly-finite VOA, so the definition is conjectural in that sense and should be tweaked if necessary as we learn more.
9 This is not obvious; for example, Miyamoto [Miy3] proposes rigidity should be replaced for general C2-cofinite VOAs with a notion he calls ‘semi-rigidity’. In any case, there should be several superficially different but equivalent definitions; what we want is to make connection with the deep work of Lyubashenko [Ly1, Ly2].
Conjecture 1.1. The category Modgr(V) of V-modules is a log-modular category.
Some authors have conjectured that any modular tensor category is realized by a strongly-rational VOA. Analogously, one should ask if any log-modular tensor category is realized by a strongly-finite VOA (we don’t want to elevate this to the status of a conjecture yet). In Section 2.1 we define the ‘semi-simplification’ of a category, by quotienting by the negligible morphisms and restricting to the full subcategory generated by the simple modules.
Conjecture 1.2. The semi-simplification of category Modgr(V) is a modular tensor cat- egory, whose Grothendieck ring is the quotient of the Grothendieck ring of Modgr(V) by the image of the V-modules which are both simple and projective.
There are several weaker statements, which we prove in Section 5 assuming e.g. rigidity. We collect some of these in the next two conjectures.
Conjecture 1.3. If M is simple then End(M) = CId. If M is indecomposable then all elements of rad(End(M)) are nilpotent and End(M) = CId + rad(End(M)). The projective V-modules form an ideal in Modgr(V), closed under tensor products and taking contragredient.
Conjecture 1.4. Suppose V is strongly-finite but not strongly-rational. Then: (a) V is not projective as a V-module. (b) V has infinitely many families of indecomposable V-modules, each parametrized by complex numbers. ∼ ∼ i (c) for any grading-restricted weak V-module Y , Pi ⊗Y = Y ⊗Pi = ⊕j,kNk,j∨ hY : MjiPk; (d) the categorical dimension of any projective V-module P is 0.
Most exciting is the question as to what replaces Verlinde’s formula. We propose:
m Conjecture 1.5. (a) S block-diagonalizes all tensor resp. Grothendieck matrices NMj , NPj m −1 m (j) (j) resp. [N ]j: S NM S = ⊕jBM where for each j, M 7→ BM is an indecomposable representation of the tensor ring (similarly for the Grothendieck matrices) (...this seems wrong...). (j) (b) Each indecomposable subrepresentation M 7→ BM of the Grothendieck ring repre- sentation is equivalent to a submodule of the open Hopf link invariant M 7→ ΦM,Pj for some projective cover Pj, each occurring with multiplicity exactly 1. Each indecomposable subrepresentation of the regular representation of the full tensor ring Fusfull(V) occurs in Fusfull(V) with multiplicity 1. (c) There is an isomorphism between the space of pseudo-traces and ambidextrous... such that...
10 We do not know which indecomposable subrepresentations should appear in the regular representation of the full tensor ring Fusfull(V) or even the subring Fussimp(V) generated by the simple V-modules Mj and their projective covers Pj — even for V = Wp this question is mysterious. It is a theorem for Wp that open Hopf link operators do not suffice for Fusfull(V) or Fussimp(V). This interesting question should be explored. Of course, continua of indecomposables will be needed in general. We expect that the space spanned by the characters of V possesses more structure than it is usually credited. The character of a simple module Recall the slash operator
aτ + b a b F | γ(τ, v) = (cτ + d)−kF , v , γ = . k cτ + d c d
Conjecture 1.6. (a) The C-span ChV of the 1-point functions FM (τ, v), as M runs over (0) (1) (N) all simple V-modules, has a grading ChV = ChV ⊕ ChV ⊕ · · · ⊕ ChV , where for any (k) v ∈ V with L[0]v = nv, the space of all F (?, v) for F ∈ ChV is an SL(2, Z)-module for the slash operator |k+n, and this SL(2,Z)-module structure is independent of v. (b) If M is the projective cover of a simple module, then its 1-point function FM lies in (0) (0) ChV . If M is simple with nilpotency m, then its 1-point function FM lies in ChV ⊕ (m) · · · ⊕ ChV . Pm (k) (c) k=0 dim ChV will equal the number of simple V-modules with nilpotency degree ≤ m. (doesn’t seem write)
SF2 already should give a counterexample to this conjecture... We should work out SF2 explicitly. The known space of pseudo-trace is 11-dim, while the modular closure of characters is 6-dimensional. It’ll be good to see the additional pseudo-traces, and see why they’re there. Question. Relation of Runkel’s approach to Miyamoto’s. What is dimension of space of 1-point functions for Wp and SFd? After all, the dimension of Miyamoto’s space can be bigger than expected, as we explain for symplectic fermions. Could Runkel give a submodule? Miyamoto’s pseudo- traces recover the formal space of 1-point functions on torus, whose definition is copied from Zhu, but maybe in the semi-simple case Zhu’s definition should be chopped down? One of the things to gain from the categorical picture is that it is at present much easier to construct examples of modular tensor categories than it is to construct rational VOAs. We expect the analogue to hold for strongly-finite VOAs:
Conjecture 1.7. (a) Any finite tensor category (in the sense of Etingof–Ostrik [EO]) can be realized by endomorphisms on some algebra, as sketched in Section 5.5 below. (b) The centre (or quantum double) construction applied to a finite tensor category yields a log-modular tensor category (as in Conjecture 1.1).
It is a theorem that any unitary fusion category can be realized by ∗-endomorphisms on some ∗-algebra; the generalization to nonunitary fusion categories is the theme of [EG2]. It is a theorem that the centre construction applied to any fusion category is a modular tensor category. Conjecture 1.6 is the obvious generalization to the finite but nonsemi-simple setting. It is known (Thm. 3.34 of [EO]) that the double of a finite tensor
11 category is a finite tensor category. Of course most log-modular tensor categories won’t be centres of finite tensor categories, just as most modular tensor categories are not centres or doubles of fusion categories. Just as we expect that any modular tensor category is the category Modord(V) of ordinary V-modules for some strongly-rational VOA V, we would expect that any log-modular tensor category (at least once that term is defined correctly!) is the category Modgr(V) for some strongly-rational VOA V.
Acknowledgements: We would like to thank J¨urgenFuchs for many explanations of Lyubashenko’s work on modularity within nonsemi-simple tensor categories. TC also thanks Shashank Kanade for many discussions on related topics. TG thanks the Physics Department at Karlstad University for a very pleasant and stimulating work environment. Our research is supported in part by NSERC.
2 Background material 2.1 Braided tensor categories We use the books [T], [EGNO] as references on tensor categories. Some features for nonsemi-simple ones are included in [CGP, GKP1]. Let C be our tensor category. We assume it to be strict so that we don’t have to worry about associativity isomorphisms. The category is called rigid if for each object M in the category, there is a dual M ∗ ∗ ∗ and morphisms bV ∈ Hom(1,V ⊗ V ) (the co-evaluation) and dV ∈ Hom(V ⊗ V, 1) (the evaluation) such that
(IdV ⊗ dV ) ◦ (bV ⊗ IdV ) = IdV , (dV ⊗ IdV ∗ ) ◦ (IdV ∗ ⊗ bV ) = IdV ∗
Rigidity is required for the categorical trace and dimension, as we’ll see shortly. In the vertex operator algebra language, a morphism is an intertwiner and an object is a module. Given any modules V,W in the category, the braiding cV,W is an intertwiner in Hom(V ⊗ W, W ⊗ V ) satisfying
cU,V ⊗W = (IdV ⊗ cU,W ) ◦ (cU,V ⊗ IdW ) , cU⊗V,W = (cU,W ⊗ IdV ) ◦ (IdU ⊗ cV,W ) (4) and also for any intertwiners f ∈ Hom(V,V 0), g ∈ Hom(W, W 0),
(g ⊗ f) ◦ cV,W = cV 0,W 0 ◦ (f ⊗ g) . (5)
This property is called naturality of braiding. It implies cV,1 = c1,V = 1 as well as the Yang-Baxter equation
(IdW ⊗ cU,V ) ◦ (cU,W ⊗ IdV ) ◦ (IdU ⊗ cV,W ) = (cV,W ⊗ IdU ) ◦ (IdV ⊗ cU,W ) ◦ (cU,V ⊗ IdW ) .
If the category C is in addition additive (i.e. it has direct sums), then (U ⊕ V ) ⊗ W resp. W ⊗ (U ⊕ V ) are isomorphic with U ⊗ W ⊕ V ⊗ W resp. W ⊗ U ⊕ W ⊗ V , and using these isomorphisms the naturality (5) of the braiding implies that we can make the identifications cU⊕V,W = cU,W ⊕ cV,W , cW,U⊕V = cW,U ⊕ cW,V . (6)
12 In a (semi-simple) modular tensor category, the braiding is needed for the topological S-matrix, as we’ll see shortly. Given any module V in our category, the twist θV ∈ Hom(V,V ) satisfies, for any module V and intertwiner f ∈ Hom(V,V ),
θV ⊗W = cW,V ◦ cV,W ◦ (θV ⊗ θW ) , θV ◦ f = f ◦ θV
This implies θ1 = 1. The twist directly gives us the (diagonal) topological T -matrix in a modular tensor category: for any simple object V , θV is a number and TV,V = θV . Define morphisms 0 ∗ bV := (IdV ∗ ⊗ θV ) ◦ cV,V ∗ ◦ bV ∈ Hom (1,V ⊗ V ) and 0 ∗ dV := dV ◦ cV,V ∗ ◦ (θV ⊗ IdV ∗ ) ∈ Hom (V ⊗ V , 1) . Given any intertwiner f ∈ Hom(V,V ), the (categorical) trace is defined by
0 tr(f) = dV ◦ (f ⊗ IdV ∗ ) ◦ bV ∈ Hom(1, 1),
We can identify Hom(1, 1) with C. The trace satisfies tr(f ◦ g) = tr(g ◦ f) and tr(f ⊗ g) = tr(f)tr(g). We define the (categorical) dimension to be dim(V ) = tr(IdV ). It will be positive if the category is unitary. Computations are greatly simplified using a graphical calculus, some basic notation is
0 0 V W. bV bV dV dV cV,W
Details on the graphical calculus are found in the textbooks [T, K, BK]. By a finite tensor category [EO] we mean an abelian rigid tensor category over C, where morphism spaces are finite-dimensional, there are only finitely many simple objects, each of which has a projective cover, every object has finite length in the sense of Jordan– H¨older,and the endomorphism algebra of the tensor unit is C. By a fusion category (see e.g. [EGNO]) we mean a finite tensor category which is in addition semi-simple. By a ribbon category [T] we mean a strict braided tensor category equipped with left duality ∨ X and a twist ΘX ∈ End(X). m In a braided fusion category, the categorical S-matrix is defined by SU,V = tr(cU,V ◦ cV,U ) ∈ C, for any simple modules U, V . Its graphical representation is the Hopf link
V W.
By a modular tensor category [T] we mean a fusion category which is ribbon, with invert- ible S m. The categorical S- and T -matrices define a projective representation of SL(2, Z). It m m m satisfies SU,V = SV,U , S1,V = dim(V ), as well as Verlinde’s formula
X W m −1 m m NU,V SW,X = (dim X) SU,X SV,X W
13 Ideally (e.g. when the VOA is strongly-rational), the categorical S- and T -matrices will agree up to scalar factors with the modular S- and T -matrices defined through the VOA characters. Return for now to a finite tensor category C. Assume it is spherical. By a negligible morphism f ∈ HomC(U, V ) we mean one for which the categorical trace TrC(gf) = 0 for neg all g ∈ HomC(V,U). The negligible morphisms form a subspace IC (U, V ) of HomC(U, V ), closed under taking duals, arbitrary compositions, as well as arbitrary tensor products. Consider the category C whose objects are the same as those of C, but whose Hom- neg spaces are HomC(U, V )/IC (U, V ). This category was originally defined in [BW]; see also p.236 of [EGNO]. If U, V are indecomposable in C and f ∈ HomC(U, V ) is not an isomorphism, then f is negligible. Then C is a spherical tensor category, whose simple objects are precisely the indecomposable objects of C with nonzero categorical dimension (the indecomposables with dimension 0 are in C isomorphic to the 0-object). Moreover, two simple objects in C are isomorphic iff they are isomorphic indecomposables in C. So C will generally have infinitely many inequivalent simples, but it is semi-simple in the sense that every object in C is a direct sum of simples. By the semi-simplification Css of C, we mean the full subcategory of C generated by the simples of C. Then Css will also be a semi-simple spherical tensor category. If in C the tensor product of simples is always a direct sum of simples and projectives (as it is for Wp and the symplectic fermions), then Css will be a fusion category, whose Grothendieck ring is the quotient of that of C by the modules in C which are both simple and projective. If C is ribbon, so is both C and Css.
Theorem 2.1. Note though that the semi-simplification of a nonsemi-simple finite tensor category C can never be unitary, as some quantum-dimensions must be negative in order that the projectives of C have 0 quantum-dimension.
2.2 Categories and 1-point functions of strongly-finite VOAs By a weak module M of a VOA V we mean the usual definition of a V-module, except that L need not act semi-simply. Rather, M = ` M where M is a (possibly infinite- 0 r∈C r r k dimensional) generalized eigenspace of L0 with eigenvalue r — i.e. (L0 − rId) (Mr) = 0 for some k = k(r). We call M an ordinary V-module if L0 acts semi-simply. We say a weak module M is grading-restricted if all spaces Mr are finite-dimensional, and also gr there is some l = lM ∈ R such that Mr = 0 whenever Re(r) < l. Write Mod (V) for the category of all grading-restricted weak V-modules. The morphisms are linear maps f : M → N such that f(YM (u, z)) = YN (u, z)f for all u ∈ V, i.e. f(unw) = unf(w) for all u ∈ V, w ∈ M, and n ∈ Z. For example, L0 is always in the centre of End(W). A regular VOA is one in which any weak V-module is a direct sum of simple ordinary V-modules. Given any algebra A (e.g. a VOA), an A-module P is called projective if for every A- modules M,N and surjective A-morphism f : M → N and any A-morphism h : P → N, there is an A-morphism h0 : P → M such that h = f ◦ h0. An A-morphism h : P → M is called a projective cover of an A-module M if P is a projective A-module and any A-morphism g : N → P for which h ◦ g is surjective, is itself surjective.
Theorem 2.2. Let V be a strongly-finite VOA. Then:
gr (i) Mod (V) is a C-linear abelian braided tensor category;
14 (ii) Modgr(V) has finitely many simple modules (up to isomorphism), all of which are ordinary modules; (iii) all Hom-spaces in Modgr(V) are finite-dimensional; (iv) any grading-restricted weak V-module has finite Jordan–H¨olderlength (hence is Ar- tinian and Noetherian);
(v) any simple V-module Mi has a projective cover Pi. (vi) the contragredient M 7→ M ∨ defines a contravariant endofunctor from Modgr(V). Thus Modgr(V) is a finite tensor category in the sense of [EO], apart from possibly rigidity. Theorem 2.2 is proved in [H3] (though parts were known earlier). In particular, part (i) is Thm.4.11 there (see also [HLZ] and references therein). Part (ii) is part of Thm.3.24 and Thm. 3.12. Part (iii) is Thm. 3.23. Part (iv) is Corollary 3.16. Part (v) is Thm. 3.23 of [H3] and Proposition 10 of [Miy3]. Part (vi) is explicitly given in Proposition 10.4 of [TW]. Using Theorem 2.1 and Proposition 6-3 of [NT], we can identify the VOA category Modgr(V) with the category of finite-dimensional modules of an explicit finite-dimensional algebra:
Theorem 2.3. Let V be strongly-finite and write P = P0 ⊕ · · · ⊕ Pn for the direct sum of projective covers of simple V-modules. Then AV := EndV (P ) is a finite-dimensional gr fin associative algebra, and Mod (V) is equivalent as an abelian category to Mod (AV ), where V-module M corresponds to the right AV -module HomV (P,M). Equivalence as abelian categories means we ignore tensor products and duals, but the equivalence preserves simples, projectives, indecomposables, composition series, etc. We explore consequences of this in Section 5.1 below. The module P = P0 ⊕ · · · ⊕ Pn is called a progenerator. If we have in addition rigidity, then Modgr(V) is in fact a finite tensor category, and we can say more. In particular, Proposition 2.7 of [EO] says any finite tensor category can be regarded as the tensor category of modules of a weak quasi-Hopf algebra. The simplest example of Theorem 2.3 is a strongly-rational V, whose corresponding algebra AV will be a direct sum of C’s, one for each simple V-module. For a more interesting example, take V to be the triplet algebra Wp. Then from the computations 2 p−1 in Section 6 of [NT], we obtain that AWp is the direct sum C ⊕ A8 of algebras, where ± ± ± ± ∓ ± ∓ A8 is the 8-dimensional algebra with basis {e , τ1 , τ2 , τ1 τ2 = τ2 τ1 } and products ± ± ± ± ∓ ± ± ± ± ∓ s t ± ∓ ± ∓ e τi = τi = τi e , e e = e , e e = 0, and all τi τj = 0 except for τ1 τ2 = τ2 τ1 . ± The two copies of C correspond to the simple projective Wp-modules Xp . Each copy of + − A8 has two simple modules, corresponding to Xi ,Xp−i for some 1 ≤ i < p. Let V be strongly-finite. Write Ek(τ) for the Eisenstein series of SL(2, Z), normalized to have constant term −Bk/k!. Write m = C[E4(τ),E6(τ)] for the space of (holomorphic) modular forms of SL(2, Z). Recall the subspace Uq(V) of V ⊗C m defined e.g. in [Z]. Definition 2.4. By a 1-point function on the torus for V is meant a function
F : V ⊗C m × H → C satisfying:
(i) for u(τ) ∈ V ⊗C m, F (u(τ), τ) is holomorphic for τ ∈ H; P P (ii) F ( i ui ⊗ fi(τ), τ) = i fi(τ)F (ui, τ) for all fi(τ) ∈ m and ui ∈ V;
15 (iii) F (u, τ) = 0 for u ∈ Oq(V); (iv) for all u ∈ V,
∞ 1 d X F (L u, τ) = F (u, τ) + E (τ) F (L u, τ) . [−2] 2πi dτ 2k [2k−2] k=1
Let C(V) denote the space of all 1-point functions. It is easy to show (see Theorem 5.2 in [Miy1]) that C(V) carries an action of SL(2, Z) through M¨obiustransformations on τ ∈ H as usual. The main result (Theorem 5.5) of [Miy1] is that C(V) is finite-dimensional, and spanned by the pseudo-trace functions SM,φ(u, τ), where M is a ‘generalized Verma module interlocked with a symmetric linear functional φ’ of the mth Zhu algebra Am(V) for m sufficiently large. These functions SM,φ(u, τ) are quite difficult to compute in practise, but [AN] revised [Miy1], and this version is both simpler and fits much more nicely with our approach. We describe their work shortly. The background underlying both [Miy1, AN] is the following. Choose any indecom- posable V-module M, and let Mn (n = 0, 1,...) denote the generalized L0-eigenspace of eigenvalue h + n (h is the conformal weight of M). On each subspace Mn, L0 − c/24 acts k as (n + h − c/24)Id + Lnil where Lnil = 0 for some k independent of n. We obtain for the usual character
∞ ∞ k−1 X X X (2πiτ)j F (v, τ) = Tr| (o(v)qL0−c/24) = qn+h−c/24 Tr| (o(v) Lj ) . (7) M Mn j! Mn nil n=0 n=0 j=0
Now, the conformal vector ω of V lies in the centre of the nth Zhu algebra An(V) for any n n, and on ⊕k=0Mn we have o(u ∗n v) = o(u) ◦ o(v) (Theorem 3.2 of [DLM1]) where ∗n is the product in An(V), so on Mn we have o(v) ◦ L0 = o(v ∗n ω) = o(ω ∗n v) = L0 ◦ o(v). Therefore on Mn the nilpotent operator Lnil also commutes with all operators o(v). This j k k k means for any j > 0 and any v ∈ V,(o(v) Lnil) = o(v) Lnil = 0. But the trace of a nilpotent operator is always nilpotent, so (7) collapses to
∞ X n+h−c/24 FM (v, τ) = Tr|Mn (o(v))q . n=0
Thus the ordinary trace can never see the nilpotent part of L0, and FM (v, τ) = FN (v, τ) whenever V-modules M,N have the same composition factors. Nor can we obtain any more 1-point functions if we insert an endomorphism f of V-module M: by definition, f will commute with all zero-modes and hence with L0, so f ◦ o(v) ◦ Lnil is still nilpotent. What we need is a way to generalize the trace of Mn-endomorphisms so that the terms j o(v)Lnil with j > 0 can contribute. There is a classical situation where this happens. Let A be a finite-dimensional asso- ciative algebra over C. A linear functional φ : A → C is called symmetric if φ(ab) = φ(ba) for all a, b ∈ A. Let SLF (A) denote the space of all symmetric linear functionals on A — it can be naturally identified with the dual space (A/[A, A])∗. Now suppose W is a finitely- generated projective A-module. Then an A-coordinate system of W consists of finitely P many ui ∈ W and the same number of fi ∈ HomA(W, A) such that w = i fi(w)ui
16 for any w ∈ W . Given an A-coordinate system, we can associate any endomorphism α ∈ EndA(W ) with a matrix [α] whose ijth entry is [α]ij = fi(α(uj)) ∈ A. Fix any symmetric linear functional φ ∈ SLF (A) and A-coordinate system {ui, fi}, and define φ the pseudo-trace TrW : EndA(W ) → C by
φ X TrW (α) = φ(Tr([α])) = φ (fi(α(ui))) . i
The pseudo-trace φW is independent of the choice of A-coordinate system, and lies in φ φ SLF (EndA(W )). It satisfies TrW (α ◦ β) = TrV (β ◦ α) for any α ∈ HomA(V,W ), β ∈ HomA(W, V ). To apply this generalized notion of trace to our VOA setting, we need to find a finite- dimensional associative algebra A and V-modules M for which the generalized eigenspaces Mn are projective A-modules. In [Miy1], A is the nth Zhu algebras An(V) for n suffi- ciently large, and M are certain projective V-modules. [AN] makes a different choice. In particular, let M be a V-module for a strongly-finite VOA V. It is elementary that M and in fact all of its generalized eigenspaces Mn are modules over the (finite-dimensional asso- ciative) algebra EndV (M). Suppose M is projective over some subalgebra B of EndV (M). Then each subspace Mn will also be projective over B. For any φ ∈ SLF (B), define the pseudo-trace function
∞ k−1 X X (2πiτ)j F B,φ(v, τ) = qr+n−c/24 Trφ (o(v) Lj ) M j! Mn nil n=0 j=0 is a 1-point function, where k is as above. The point is Theorem 4.3.4 of [AN], which says B,φ that these FM (v, τ) are 1-point functions. Quantum field theory teaches that correlation functions should arise from field inser- tions. These generalized traces can be interpreted presumably as inserting a defect, but the defect would have to break the conformal symmetry. The explanation for why it is so difficult to obtain these 1-point functions here is that the factorization of the full CFT into chiral halves here is much more subtle than in RCFT — the true QFT is the full CFT. The price we pay for the bulk space not being factorizable in the RCFT sense, is the need to introduce these generalized traces (or something equivalent).
2.3 Open Hopf links and modified traces
Let C be any rigid braided tensor category. Denote associativity isomorphisms by AX,Y,Z : X ⊗ (Y ⊗ Z) → (X ⊗ Y ) ⊗ Z. Let f be an endomorphism of V ⊗ W in C. Then the left partial trace is
−1 −1 ptrL (f) := (dV ⊗IdW )◦AV ∗,V,W ◦(IdV ∗ ⊗f)◦(IdV ∗ ⊗(ψV ⊗IdW ))◦AV ∗,V ∗∗,W ◦(bV ∗ ⊗IdW ) which is an element of End(W ). What replaces here the Hopf link invariant of modular tensor categories, are the open and logarithmic Hopf link invariants. Definition 2.5. Let V,W be in C, then the open Hopf link operator is
ΦV,W : = ptrL (cV,W ◦ cW,V ) ∈ End(W ) .
17 The name general Hopf link is used in [CGP]. Pictorially the open Hopf link operator is given above in (3). The following theorem is a direct consequence of the graphical calculus and is well-known.
Theorem 2.6. If C is strict, then for any W ∈ C, the map
Φ ? ,W : Obj(C) → End (W ) ,V 7→ ΦV,W is a representation of the tensor ring.
Proof: Linearity under addition is trivial. The ring homomorphism property is then the following identity
U U ΦU⊗V,W = U ⊗ V = V = = ΦU,W ◦ ΦV,W . V
W W W
Here, the third equality holds because of naturality of braiding, and the second one is (4) together with IdU⊗W = IdU ⊗ IdW .
In order to prove this statement for not necessarily strict categories one has to translate the argument to some commutative diagrams.
Theorem 2.7. For any W ∈ C, with not necessarily strict C the map
Φ ? ,W : Obj(C) → End (W ) ,V 7→ ΦV,W is a representation of the tensor ring.
Proof: We first introduce a few short-hand notations. Let P and Q be any two parenthe- sized products of the same objects A, B, C, . . . in the same order. Construct isomorphisms from P to Q just using associativity isomorphisms (and their inverses), then any two such isomorphisms coincide by the associativity axiom of monoidal categories. We denote this isomorphism by α. Further identity morphisms will be surpressed for readability. All squares of the following diagram commute due to naturality of braiding, while the triangle commutes because of last paragraphs discussion.
−1 −1 ∗ ψU ◦ bU ◦ rV ∗ (V ∗ ⊗ V ) ⊗ W / ((V ∗ ⊗ (U ∗ ⊗ U)) ⊗ V ) ⊗ W α α α −1 −1 ψ ◦ b ∗ ◦ r U U V ∗ α + V ∗ ⊗ (V ⊗ W ) / (V ∗ ⊗ (U ∗ ⊗ U)) ⊗ (V ⊗ W ) / (V ∗ ⊗ U ∗) ⊗ (U ⊗ (V ⊗ W ))
cV,W cV,W cV,W −1 −1 ψ ◦ b ∗ ◦ r U U V ∗ α V ∗ ⊗ (W ⊗ V ) / (V ∗ ⊗ (U ∗ ⊗ U)) ⊗ (W ⊗ V ) / (V ∗ ⊗ U ∗) ⊗ (U ⊗ (W ⊗ V ))
18 The next diagram is commutative for the exact same reasons as the previous one.
rV ∗ ◦ dU (V ∗ ⊗ V ) ⊗ W o ((V ∗ ⊗ (U ∗ ⊗ U)) ⊗ V ) ⊗ W O O k α α α
rV ∗ ◦ dU α V ∗ ⊗ (V ⊗ W ) o (V ∗ ⊗ (U ∗ ⊗ U)) ⊗ (V ⊗ W ) o (V ∗ ⊗ U ∗) ⊗ (U ⊗ (V ⊗ W )) O O O cW,V cW,V cW,V
rV ∗ ◦ dU α V ∗ ⊗ (W ⊗ V ) o (V ∗ ⊗ (U ∗ ⊗ U)) ⊗ (W ⊗ V ) o (V ∗ ⊗ U ∗) ⊗ (U ⊗ (W ⊗ V ))
Using that b(U⊗V ) is the map
−1 bV rV bU α 1 / (V ⊗ V ∗) / (V ⊗ 1) ⊗ V ∗ / (V ⊗ (U ⊗ U ∗)) ⊗ V ∗ / (V ⊗ U) ⊗ (U ∗ ⊗ V ∗) and dU⊗V is the map
∗ α dU rV ∗ dV (U ⊗ V ) ⊗ (U ⊗ V ) / (V ∗ ⊗ (U ∗ ⊗ U)) ⊗ V / (V ∗ ⊗ 1) ⊗ V / V ∗ ⊗ V / 1 together with naturality of the pivotal structure the open Hopf link operator can be written as −1 −1 ΦU⊗V,W = `W ◦ dV ◦ A ◦ B ◦ C ◦ ψV ◦ bV ∗ ◦ `W , where A = rV ∗ ◦ dU ◦ α ◦ cW,V is given by the upper right path of the second diagram, −1 −1 −1 B = AA,C,B ◦ cA,C ◦ cC,A ◦ AA,C,B and C = cV,W ◦ α ◦ ψU ◦ bU ∗ ◦ rV ∗ is given by the upper right path of the first diagram. Commutativity of these two diagrams implies that they coincide with the two lower left paths, i.e.
−1 −1 A = α ◦ cW,V ◦ rV ∗ ◦ dU ◦ α, C = α ◦ ψU ◦ bU ∗ ◦ rV ∗ ◦ cV,W ◦ α. Inserting this in the open Hopf link operator, we get
−1 −1 ΦU⊗V,W = `W ◦ dV ◦ α ◦ cW,V ◦ D ◦ cV,W ◦ α ◦ ψV ◦ bV ∗ ◦ `W ,
19 where D is given by the left path of the following third diagram.
V ∗ ⊗ (W ⊗ V ) −1 −1 rV ∗ `W
t α * (V ∗ ⊗ 1) ⊗ (W ⊗ V ) / V ∗ ⊗ ((1 ⊗ W ) ⊗ V )
−1 −1 ψU ◦ bU∗ ψU ◦ bU∗ α (V ∗ ⊗ (U ∗ ⊗ U)) ⊗ (W ⊗ V ) / V ∗ ⊗ (((U ∗ ⊗ U) ⊗ W ) ⊗ V )
α α α (V ∗ ⊗ U ∗) ⊗ ((U ⊗ W ) ⊗ V ) / V ∗ ⊗ ((U ∗ ⊗ (U ⊗ W )) ⊗ V )
cU,W ◦ cW,U cU,W ◦ cW,U α (V ∗ ⊗ U ∗) ⊗ ((U ⊗ W ) ⊗ V ) / V ∗ ⊗ ((U ∗ ⊗ (U ⊗ W )) ⊗ V )
α α α (V ∗ ⊗ (U ∗ ⊗ U)) ⊗ (W ⊗ V ) / V ∗ ⊗ (((U ∗ ⊗ U) ⊗ W ) ⊗ V )
rV ∗ ◦ dU `W ◦ dU * t V ∗ ⊗ (W ⊗ V )
The triangles commute due to the triangle axiom together with naturality of left and right multiplication `W , rV ∗ , all the squares commute because of naturality of associativity and braidings. The right-hand side is the open Hopf link operator ΦU,W , hence
−1 −1 ΦU⊗V,W = `W ◦ dV ◦ α ◦ cW,V ◦ ΦU,W ◦ cV,W ◦ α ◦ ψV ◦ bV ∗ ◦ `W −1 −1 = `W ◦ dV ◦ α ◦ cW,V ◦ cV,W ◦ α ◦ ψV ◦ bV ∗ ◦ `W ◦ ΦU,W = ΦV,W ◦ ΦU,W , by naturality of braiding, associativity and left-multiplication.
To help in identifying the open Hopf link invariants, we have the following simple but useful fact, which follows immediately from the naturality (5) of the braiding isomor- phisms: Lemma 2.8. Hom intertwines the open Hopf link operators: for any U, V, W ∈ C and any f ∈ Hom(V,W ), ΦU,W ◦ f = f ◦ ΦU,V ∈ Hom(V,W ) .
For example we know that ΦU,W , being an endomorphism of W , will restrict to an endomorphism of the socle of W , and will project to an endomorphism of the top of W (recall the socle of a module is the largest semi-simple submodule, while the top is the quotient by the radical); the lemma allows us to extract both of these. In particular, when End(W ) is one-dimensional, ΦU,W equals either 0 or ΦU,M for any simple submodule M. Moreover, if our category C is in addition additive (i.e. has direct sums), then using the ∼ ∼ isomorphisms of (W1 ⊕W2)⊗V = W1 ⊗V ⊕W2 ⊗V and V ⊗(W1 ⊕W2) = V ⊗W1 ⊕V ⊗W2 to write (6), we obtain
ΦV,W1⊕W2 = ΦU,W1 ⊕ ΦV,W2 . (8)
20 Implicit in (8) is the natural embedding of End(W1) ⊕ End(W2) into End(W1 ⊕ W2) (the latter is often larger). (8) means without loss of generality we can restrict to indecompos- able W . Let now P be a full subcategory of C that is a right ideal, meaning that for V in P and W in C, both 1. V ⊗ W is in P, and
2. if α : W → V and β : V → W such that β ◦ α = IdW then W in P. Definition 3.1 of [CGP] (see also [GKP1]) is Definition 2.9. A modified trace on P is a family of linear functions
{ tV : End(V ) → C | V ∈ P } satisfying 1. For U in P and W in C and any f in End (U ⊗ W )
tU⊗W (f) = tV (ptrL (f)) .
2. For U, V in P and morphisms f : V → U and g : U → V
tV (g ◦ f) = tU (f ◦ g) .
For example, the categorical trace on the full category C is a (trivial) example of a modified trace. A more interesting example of a right ideal is spanned by the projective modules in C. By Theorem 2.5.1 of [GKP1], a way to get modified traces is through the existence of ambidextrous objects in C. We pursue this strategy in Section 3 below. Definition 2.10. Let V in C, W in P and x in End (W ), then the logarithmic Hopf link invariant is m ;P,x SV,W := tW (ΦV,W ◦ x) = tW (x ◦ ΦV,W ) . The second equality in the definition comes from property 2 of modified trace, and tells us we only need to apply an endomorphism to one side within the trace. When P = C, we will drop it from the superscript. Likewise we will drop the endomorphism x if x is the identity. The logarithmic Hopf link invariants extract some numbers from the operators ΦU,V , but they lose information too. The endomorphisms x can be used to extract more. We discuss partial traces and the more general symmetric linear functionals φ, in Section 5.1. Lemma 2.11. Let V and W in P then
m ;P m ;P SV,W = SW,V . Proof: This follows from combining the properties of the modified trace, namely
tW (ΦV,W ) = tW (ptrL (cV,W ⊗ cW,V )) = tW ⊗V (cV,W ⊗ cW,V )
= tV ⊗W (cW,V ⊗ cV,W ) = tV (ptrL (cW,V ⊗ cV,W )) = tV (ΦW,V ) .
21 2.4 The triplet algebra Wp
Our main reference on the triplet algebra is [TW]. The triplet models Wp (p ≥ 2 an integer) are a family of VOAs with central charge cp = 1 − 6(p − 1)/p. They are loga- rithmic and strongly-finite [AdM]. As mentioned in Section 2.2, this implies they have finitely many simple (ordinary) modules, whose characters (when augmented by pseudo- traces) form a vector-valued modular function, and which yields a braided tensor category gr Mod (Wp). The tensor product and rigidity for Wp were determined in [TW] (the tensor product had been previously conjectured in [FHST],[GabR]). + − The simple modules are Xs = Λ(s) and Xs = Π(s) for 1 ≤ s ≤ p. The tensor + + − unit (vacuum) is X1 . Their projective covers are Ps = R0(s) and Ps = R1(s), where ± ± Pp = Xp and
+ + + − − − 0 → Ys → Ps → Xs → 0, 0 → Ys → Pp−s → Xp−s → 0
± for 1 ≤ s ≤ p − 1, where Ys denotes the reducible but indecomposable modules + + − − − + 0 → Xs → Ys → 2·Xp−s → 0 , 0 → Xp−s → Ys → 2·Xs → 0 .
± ± We prefer the [TW] notation Xs and Ps . These 4p − 2 irreducible and/or projective ± ± Wp-modules Xs ,Ps are the most important ones. They are closed under tensor product simp and form the subring Fus (Wp). gr [NT] proved the equivalence as abelian categories of Mod (Wp) with the category fin Mod (Uq(sl2)) of finite-dimensional modules of the restricted quantum group Uq(sl2) πi/p fin at q = e , and [KS] determined all indecomposables in Mod (Uq(sl2)) (although the key lemma, describing pairs of matrices up to simultaneous conjugation, is really due to Frobenius (1890)). Hence:
Theorem 2.12. [TW] The complete list of indecomposable grading-restricted weak Wp- modules, up to equivalence, is: ± (i) for each 1 ≤ j ≤ p and each sign, the simple modules Xj and their projective covers ± ± ± Pj (Pp = Xp ); ± ± ± 1 (ii) for each 1 ≤ j < p, each sign, and each d ≥ 1, Gj,d, Hj,d, and Ij,d(λ), where λ ∈ CP . ± ± [KS] also give their socle series. Note that Ys = Hs,1. As with any strongly-finite VOA, any other Wp-module will be the direct sum of (finitely many) indecomposable modules. By Proposition 6-6 of [NT], all indecomposable Wp-modules are ordinary (i.e. L0 is ± diagonalizable) except for Pj for 1 ≤ j < p, whose L0 has 2 × 2 and 1 × 1 Jordan blocks. ± ± All indecomposables except those Pj have radical=socle; Pj for 1 ≤ j < p has radical ∓ ± ± ± ± ∼ ∓ ± ± Gj,1. Pj also contains a copy of each Ij,1(λ), and Pj /Ij,1(λ) = Ip−j,1(−λ). The Xs ,Ps are self-dual (Proposition 39 of [TW]). By exactness of duality (e.g. Proposition 4.2.9 of ∨ ∨ ∼ ∓ ± ∨ ∼ ∓ ± ∨ ∼ ∓ [EGNO]) we obtain (Gj,d) = Hp−j,d,(Hj,d) = Gp−j,d, and (Ij,d(λ)) = Ip−j,d(−λ). This fin recovers the duals in Mod (Uq(sl2)) of [KS], where rigidity followed from the elementary and independent reason that Uq(sl2) is a finite-dimensional Hopf algebra. For now, turn to the full subcategory relevant to our discussion, whose objects are ± ± finite direct sums of Xs ,Ps for 1 ≤ s ≤ p. We return to the other indecomposables at the end of this subsection.
22 The fusions are completely determined from the following:
0 0 0 0 X1 ⊗ Xs = Xs ,X1 ⊗ Ps = Ps , + X2 s = 1 + + + + X2 ⊗ Xs = Xs−1 ⊕ Xs+1 2 ≤ s < p , + Pp−1 s = p + − P2 ⊕ 2 · Xp s = 1 + + + + X2 ⊗ Ps = Ps−1 ⊕ Ps+1 2 ≤ s < p − 1 , + + Pp−2 ⊕ 2 · Xp s = p − 1 together with associativity and commutativity. For example one can show
0 + + − + Ps ⊗ Pt = 2·Xs ⊗ Pt ⊕ 2·Xp−s ⊗ Pt .
+ + + + + − In W2, X2 ⊗ Ps must be replaced with X2 ⊗ P1 = 2·X2 ⊕ 2·X2 . ± ± As always, the Z-span of the projective modules {Ps }1≤s≤p ∪ {Xp } form a fusion ± ± ideal. The quotient of the tensor subring Z-span{Xs ,Pt } by that ideal is easily seen to be two copies of the sl(2)d fusion ring at level p − 2, more precisely it is isomorphic − to the fusion ring of the rational VOA L1(sl2) ⊗ Lp−2(sl2), where X1 corresponds to + the integrable highest weight module L(Λ0 + Λ1) of sl(2)d 1 and Xs corresponds likewise to the module L((p − 1 − s)Λ0 + (s − 1)Λ1) of sl(2)d p−2 for 1 ≤ s < p. The proof is elementary, obtained by comparing the tensors by generators. In fact, more importantly, gr ss this also persists categorically: the semi-simplification (Mod (Wp)) is a modular tensor category, namely the twist of that of sl(2)d p−2 ⊕ sl(2)d 1 by a simple-current of order 2 as discussed in Section 3.2. ± The characters of Xs are 1 s 1 s ch[X+](τ) = θ (τ) + 2θ0 (τ) , ch[X−](τ) = θ (τ) − 2θ0 (τ) s η(τ) p p−s,p p−s,p s η(τ) p s,p s,p where X 2πiτj2/2 2πizj θs,p(τ, z) = e e √ j∈ √s + 2p 2p Z √ √s ∗ is the theta series associated to the coset 2p + L ∈ L /L of the even lattice L = 2pZ and 0 1 ∂ θs,p(τ) = θs,p(τ, 0) , θs,p(τ) = √ θs,p(τ, z) . 2πi 2p∂z z=0 + − The characters of Ps and Pp−s are both
+ − + − ch[Ps ](τ) = ch[Pp−s](τ) = 2ch[Xs ](τ) + 2ch[Xp−s](τ).
Theta functions of an even lattice are vectors of the dual of the Weil representation of its discriminant, hence