WorksMathematics
An Extremal Vertex Operator Algebra of Central Charge c=64
Abstract
Starting with the standard self-dual -orbifold of the Barnes–Wall lattice VOA of central charge , we construct a simple-current extension of a fixed-point subalgebra of its tensor square. The resulting vertex operator algebra is self-dual of central charge and has no nonzero Virasoro highest-weight vector of weight or . We give the twisted conformal weights, an elementary proof of the group-cohomological detection statement needed to trivialize the twisted-module associator class, and a permutation-cover argument determining the relevant absolute module weights. This identifies the complete weight-two obstruction set in a -dimensional binary quadratic space. An explicit -dimensional totally singular graph avoids that set; its exhaustive verifier is included. The final algebra object and its multiplication are also explicit. To my knowledge, this is the first new central-charge value of an extremal self-dual vertex operator algebra since the examples discussed in my 1995 thesis.
Introduction
Definitions and the main result
All VOAs and their modules are over . By strongly rational we mean simple, rational, -cofinite, self-contragredient, and graded as with . For this grading, rationality together with -cofiniteness is equivalent to regularity [1 Theorem 4.5].
Write for its modular tensor category of ordinary modules [24 Theorem 4.6], for the simple isomorphism classes, and for the tensor product. The notation denotes the contragredient module. We write . A Virasoro highest-weight vector is a nonzero homogeneous vector annihilated by every , .
As in my thesis [21 Definition 2.1.1], a strongly rational VOA is called self-dual (also called holomorphic) if, up to isomorphism, its adjoint module is its only irreducible ordinary module. Equivalently, . This property is stronger than self-contragredience , which is already assumed in the definition of strong rationality.
The notion of extremality is that of [21 Definition 5.2.1]: the graded character agrees with the vacuum Virasoro character through weight , or equivalently there are no nonvacuum Virasoro highest-weight vectors of positive weights at most . The lowest conformal weight of an ordinary module is its lowest -eigenvalue; it does not include the character shift by . A self-dual VOA of central charge 64 is called extremal here if
These conditions exclude Virasoro highest-weight vectors of weights 1 and 2; the first possible positive weight is then .
Let a finite group act faithfully on a strongly rational self-dual VOA , and suppose that is strongly rational. The category of twisted -modules is -graded, with one simple object in each homogeneous component [31 Theorems 4.15 and 4.17]. Its neutral component is , and its simple objects are invertible. Choosing isomorphisms expresses the associativity isomorphisms as a normalized 3-cocycle. Its choice-independent class is denoted by
where the coefficient action is trivial. We call this the twisted-module associator class. The group grading is part of this definition. Passage to the homogeneous components indexed by a subgroup restricts the class; an equivalence of braided centers without this grading would not determine it. Under the positivity hypotheses verified in § 3, the equivalent description using the distinguished function algebra is given in [DNR25, Theorem 6.3 and Remark 7.2] and [9 Theorem 4.10 and Example 4.11(ii)]. For a pointed VOA category, the conformal-weight quadratic form is
When its values are , , we instead use the binary form
Thus means integral conformal weight. This binary form must not be confused with the cyclic type of an involution:
The two normalizations have different purposes and different factors.
Theorem 1.1. Let be the standard self-dual -orbifold, let be the normal elementary abelian subgroup of , and let exchange the two factors of .
Put
The graph subgroup specified in § 6, interpreted in the actual fusion group of by Proposition 5.1, defines a strongly rational self-dual extension with
In particular is extremal.
Extremal VOAs and related literature
In my thesis [21 Section 5.2], I discussed extremal self-dual VOAs of central charges 8, 16, 24, 32, 40; at that time self-duality of the indicated rank-32 and rank-40 orbifolds was still conjectural. I explicitly recorded there that no extremal VOAs of central charge were then known. Subsequent constructions have produced non-extremal self-dual VOAs in higher central charges, including 48 and 72 [16]. To my knowledge, Theorem 1.1 supplies the first new central-charge value for an extremal self-dual VOA since those described in my 1995 thesis. This statement concerns extremality, not the existence of self-dual VOAs at central charge 64 without the extremal condition.
Witten’s proposal [41] prompted extensive subsequent study of the specialization . Three questions must be distinguished: the existence of a modular function with the prescribed initial coefficients, compatibility with further identities involving vertex operators, and existence of a VOA realizing these data. The first two do not by themselves imply the third. Moreover, , so the present central charge lies outside the family originally emphasized by Witten.
Characters and modular reconstruction. Manschot [30] expressed the proposed genus-one functions as modular sums. Manschot and Moore [MM10] developed convergent reconstruction formulae for vector-valued modular forms of nonpositive weight and constraints on their principal parts. Avramis, Kehagias and Mattheopoulou [2] considered , computed candidate extremal characters through , and studied genus-two functions through . Their work thus already includes the candidate character at . Ashrafi [3] studied related modular functions and generalized Hecke operators. The calculations of Maloney and Witten [33] form part of the broader motivation for this programme, but are not VOA existence theorems.
Higher-genus identities, including . Gaiotto and Yin [19] determined candidate genus-two functions at from modular transformation and degeneration conditions. Yin [42] investigated further proposed higher-genus expansions. Gaberdiel, Keller and Volpato [17] systematically studied genus-two vacuum correlation functions and their relation to associativity and genus-one transformation properties. Their construction of a consistent genus-two function for every is not a construction of the underlying VOA. Keller, Mathys and Zadeh [25] obtained additional constraints on multiplicities and structure constants under a positive-definiteness assumption. Their Section 5.6 and Appendix B.2 explicitly treat . Writing for the dimension of the space of Virasoro highest-weight vectors of weight , their formula (5.31) is
The extremal specialization gives , agreeing with the character check in § 8. These calculations concern necessary identities and do not provide an existence proof at .
Obstructions and restrictions on automorphisms. Gaberdiel [11] studied constraints obtained from modular differential equations and the -quotient. The proposed obstruction for , , is not an unconditional nonexistence theorem: Gaiotto [12] gave a counterexample to an implication used in the original argument, and the subsequent analysis of Gaberdiel and Keller [14] retained a holomorphy condition on coefficient functions that was not established in the required generality. Restrictions on possible Monster actions at were considered by Gaiotto [12] and in my work [23]; the orbifold hypotheses attached to these statements must be retained. Lin and Pei [28] obtained divisibility obstructions using the conjectural Segal–Stolz–Teichner correspondence. Their resulting exclusions, including , are consequently not unconditional nonexistence theorems for VOAs under the hypotheses imposed here.
Superconformal variants. Extremality relative to a superconformal algebra is a different condition from Virasoro extremality for an ordinary VOA. Gaberdiel, Gukov, Keller, Moore and Ooguri [13] studied restrictions imposed by weak Jacobi forms on extremal structures. Benjamin, Dyer, Fitzpatrick and Kachru [4], and Harrison [20], described lattice-orbifold examples at with, respectively, and structures. These are separate extremality problems; their examples do not resolve the ordinary self-dual VOA existence question in higher central charge.
Lattice-orbifold constructions. Gemünden and Keller [15, 16] constructed self-dual lattice orbifolds at with small low-weight subspaces, but these examples are not extremal. More recently, Keller, Ashley Winter Roberts and Jeremy Roberts [KR26] studied orbifolds associated with Barnes–Wall lattices. Their proposed self-dual extension at $c=128 has minimum nonvacuum Virasoro highest weight 4, conditional on triviality of a specified associator class in third group cohomology. This assumption is explicit also in the published version. Even conditional on existence, that example is not extremal, since . None of the conjectural assertions surveyed in this subsection is used as an input to the proof below.
Extremal even unimodular lattices
The analogous lattice existence problem has also seen substantial progress. A positive-definite even unimodular lattice of rank is extremal if it attains the modular-form bound
Relevant developments include Nebe’s rank-72 lattice of minimum 8, announced in 2010 and published in 2012 [34], her fourth extremal rank-48 lattice [35], and Shimada’s further rank-64 constructions [38, 39]. Sakamoto’s recent preprint [36 Theorem 1.1] reports two nonisomorphic extremal rank-88 lattices, obtained using Hermitian tensor products over and computer-assisted minimum calculations. These lattice results do not by themselves produce extremal VOAs: contains an -dimensional Heisenberg subspace in degree 1, so it is not extremal for .
Structure of the proof
The proof has three distinct parts. First we determine the actual orbifold category and its low-weight modules. Second we give an exhaustive finite calculation excluding those modules from an explicit maximal isotropic subgroup. Third the simple-current extension theorem produces the VOA. Character calculations are included only as a check, not as a substitute for these steps. For imported results used in the proof below, theorem numbers refer to the explicitly listed arXiv versions in the bibliography.
The central-charge-32 input and its twisted weights
The Barnes–Wall input and the Albert module
I recorded the expected automorphism group of in my conformal-design paper [22 Example 3.2] as a joint observation with Griess; the identification was explicitly conjectural. Shimakura [37 Theorem 2.30] subsequently proved the group shape ; his introduction also records Griess’s conjecture at the Edinburgh workshop in July 2004. For the detailed structural input used here we refer to Shimakura [37 Proposition 1.8, (2.1), Proposition 2.11, Theorems 2.19 and 2.30]:
The VOA is strongly rational and self-dual: the lattice VOA is strongly rational, and its -twisted minimum is . The three nontrivial -modules have lowest weights , respectively. Thus the cyclic type is , all nontrivial minima are positive, and the integral twisted summand is isotropic. The cyclic orbifold theorem [40 Theorem 5.16] therefore applies. The quotient acts irreducibly on , and
where the indicated nontrivial -characters are distinct and form one orbit. Here and below denotes group shape; no splitting of this extension is assumed.
We use the split Albert module
with invariant cubic
This is the three-matrix model in Cohen–Cooperstein [6]; their minus sign equals a plus sign in characteristic two. The differential
is always regarded as a covector. Pairings are the ordinary coordinate pairing . No identification of with , and no dual-adjoint convention, is needed.
The identification requires more than a group-order comparison. In characteristic two the only nontrivial irreducible -modules of degree at most are the dual modules with highest weights ; see [Lüb01, Theorems 2.3 and 4.4, {xrefext}Appendix A.51]. The same degree bound shows that the irreducible -dimensional -module is absolutely irreducible: a nontrivial scalar extension decomposition would have constituents of smaller nontrivial degree. After possibly composing the chosen quotient identification with the graph automorphism, we may therefore identify with and with .
There are three nonzero orbits, described by
The rank-one stabilizer has shape , and the rank-three stabilizer is [6]. The dual module has the same orbit sizes. Consequently in (4) is precisely the dual rank-one orbit.
A flip calculation fixing the absolute weights
Put . Its irreducible modules form a ten-dimensional plus-type quadratic fusion space . Their lowest weights are
There are 527 nonzero singular labels and 496 nonsingular labels; the singular nonvacuum modules have 16-dimensional weight-one spaces. The latter dimension also follows from transitivity on nonzero singular labels and ; see [Shi14, Proposition 1.8]. That proposition also gives rationality and -cofiniteness of . Its self-contragredience follows from the nondegenerate invariant bilinear form on the lattice VOA: averaging over the involution shows that the restriction to is nondegenerate. Thus is strongly rational, and all its nontrivial irreducible modules have positive lowest weight. There is a full diagonal extension
Let be the literal flip on each summand in (8). It is a genuine VOA automorphism, with no unspecified lift signs. Indeed, after choosing the same intertwining operator in the two identical factors, every projected component of the vertex operator is a scalar multiple of , because the relevant fusion multiplicity is one. Swapping the two tensor positions commutes with this expression. Shimakura’s descriptions (4) and (6) place this flip in .
For , let act on the -summand by . Write
For , write . The trace function and its modular transform are
For the second identity, use [[40 Theorem 3.4]] and the Fourier orthogonality sum of order . Modular covariance of twisted trace functions in the normalization , [[40 Theorem 2.3]], identifies the leading exponent with that of the unique -twisted -module, up to a nonzero scalar. Equating exponents gives
Thus the three possible absolute weights are . This calculation determines lowest weights, but does not determine the dimensions of the corresponding lowest-weight spaces.
Proposition 2.1. For the unique -twisted -module , set , with . Then
For rank-one , .
Proof. The traces of on , for the three types of in (7), are respectively
The vacuum summand contributes zero flip trace in total weight two. Each nonzero singular summand contributes 16 times its sign. For , the singular Fourier sum is ; for , it is 527. The three distinct traces must represent the three nonzero -orbits in .
Let be the reverse-orbifold grading involution of
The centralizer order in [[37 Proposition 2.1]] gives
so its -orbit has size 139503 and it has rank one. By [[37 Proposition 2.1 and Lemma 2.3]], its weight-two trace is
The 8432 class therefore has rank one and twisted weight 1. Since , character averaging gives
Hence and . The latter equals
so the classes of weights have ranks two and three. Finally the reverse-orbifold construction after [40 Theorem 5.16] gives
The lattice is rootless, so is precisely the 32-dimensional Heisenberg space. The -twisted lattice Fock space has minimum and half-integer oscillator increments [40 Theorem 7.6]. With the integral-weight eigenspace denoted by , the other eigenspace starts at . Hence , without using the unknown scalar in the modular transformation. Conjugacy gives the same dimension throughout the rank-one orbit.
Group cohomology and the twisted-module associator
Cyclic detection for elementary abelian 2-groups
The following statement is essential. Vanishing of cyclic restrictions does not detect for arbitrary groups.
Lemma 3.1. Let . Restriction to all subgroups of order two gives an injective map
With the latter factors identified with , its image consists of the functions
Proof. The split integral homology Künneth theorem gives
For clarity, the recurrence for the last exponent is obtained from : the contributions are , , , and . They add . The splitting excludes hidden -extensions.
Since is divisible, universal coefficients identifies with . It is therefore elementary abelian of the displayed rank.
For , consider the normalized multiplicative cocycles
Their exponents are trilinear over , so the cocycle equation holds directly. On the cyclic subgroup generated by , they restrict to the standard nontrivial -cocycle to the powers , , , respectively. In particular a normalized coboundary evaluates to on , whereas the nontrivial normalized -cocycle evaluates to .
The squarefree Boolean monomials of degrees one, two, and three are linearly independent as functions on . Thus these cocycle classes are independent; their number equals the computed dimension of , so they form a basis. Both claims follow.
For an involution of a strongly rational self-dual VOA, cyclic orbifold theory gives and assigns the type , as in (2); see [40 Theorem 5.11, definition of type, and Theorem 5.13]. Its identification with can be checked directly in the order-two case. Write that restriction as , where . A simple object of nontrivial -degree in has a projective centralizer scalar satisfying , by the half-braiding hexagon. Its twist square is therefore . The VOA twist square, for a constituent of , is
since constituent weights differ by half-integers and [40 Theorem 5.11]. This proves . A possible inverse-cocycle convention has no effect on a class of order two. The ribbon twist here is , not the character -eigenvalue .
In particular, . Proposition 2.1 and Lemma 3.1 identify its restriction function as the cubic . In squarefree coordinates, if , a representative is
One must use ordered monomials here, not the full six-term polarization of , whose diagonal is zero in characteristic two. The representative need not be -invariant as a cochain; its cohomology class is invariant, by cyclic detection.
The doubled action
Put
Write , . The nonidentity elements are for and for all . Their twisted lowest weights are
Indeed . Also is conjugate to by , and a transposition twist satisfies . More generally the -cycle formula is
with the central charge of one tensor factor [5 proof of Theorem 3.9(3), Theorem 4.4]. The lowest-weight space of each is one-dimensional.
| 0 | 0 | 0 | 0 | 0 |
| rank 1 | 1 | 0 | 2 | 0 |
| rank 2 | 0 | 3 | 0 | |
| rank 3 | 1 | 0 |
Table 1.
Every has weight 2 and type 0.
Proposition 3.2. The full associator class is zero.
Proof. For , its cyclic type is . For , it is . Thus every cyclic restriction vanishes, and Lemma 3.1 applies.
Checking and separately would not by itself exclude mixed cohomology terms. The calculation checks every , and the detection lemma is what excludes those terms. In particular, the calculation retains the contribution of , represented by (12).
Regularity, positivity, and the actual category
The groups used here are finite 2-groups, hence solvable. Carnahan–Miyamoto [8 Theorem 5.24 and Corollary 5.25] gives regularity and self-contragredience of the fixed-point VOAs
The nonnegative grading and the one-dimensional degree-zero space are inherited from the original VOA. By [1 Theorem 4.5], these fixed-point VOAs are strongly rational. Completeness of the irreducible constituents of twisted modules, and the corresponding equivariantization equivalence, follow from [31 Theorems 4.15 and 4.17], without first imposing positivity on every irreducible fixed-point module.
Every -twisted -module for has positive minimum by Proposition 2.1; among the constituents of , only contains the vacuum vector. Hence every nonvacuum -module has positive weight. The same argument using (13) works for . For , untwisted constituents come from , with . Only the symmetric vacuum constituent contains weight zero. Permutation-twisted constituents have weights at least .
Thus the positivity hypotheses in the center-description theorem [DNR25, Theorem 6.2(4) and Remark 7.2] are now satisfied. Together with Proposition 3.2, it gives
Positivity also identifies the categorical dimensions with the positive Frobenius–Perron dimensions, so the equivalence respects the positive spherical structure and the conformal-weight ribbon twist. All simples are invertible, and is a nondegenerate hyperbolic quadratic space. Define the twisting-label homomorphism
Its kernel consists of the irreducible constituents of and is canonically indexed by . For and , their polar pairing satisfies . The same theory gives , even though the class is nonzero.
Finally,
Indeed -invariance leaves just the two conformal vectors in , and -invariance leaves their sum.
Permutation-twisted modules and exact fusion weights
The category alone sees weights modulo , not their absolute minima. The following argument supplies the additional information.
Lemma 4.1 (Lowest-weight summands of a tensor square). Let be strongly rational, and put . Let be the simple -constituent of containing its one-dimensional homogeneous subspace of weight . For every , the fusion product contains a simple summand of lowest weight , arising from .
Proof. The fixed-point theorem [8 Corollary 5.25], together with [1 Theorem 4.5], makes strongly rational. In particular is a rigid semisimple vertex tensor category. All modules used below restrict to finite-length ordinary -modules.
Assign the modules , , to the marked points of the base sphere, with monodromies . The connected covering is ; the modules at the marked points of the covering sphere are , , , . Propagation of the vacuum modules reduces the conformal block to the contragredient pairing between and . Here conformal blocks and propagation are used in the algebraic-analytic definitions of [18 Sections 1.2 and 1.3].
Choose lowest-weight vectors with . The cover block is nonzero on . At the two unramified marked points the changes to lifted coordinates have nonzero derivatives; positive Virasoro modes annihilate lowest-weight vectors, so these changes only multiply the value by nonzero scalars. Propagation of the vacuum vectors also preserves its nonzero value.
Gui’s cover theorem [18 Theorem 2.25] identifies the base and cover blocks as the same underlying linear functional, not merely as abstract vector spaces. Therefore the base block is nonzero on , where spans the lowest-weight space of .
By [18 Theorem 4.10], this block determines a twisted intertwining operator. Restricting the vertex operators to gives an ordinary -intertwining operator by [18 Remark 4.6]: the relevant -eigenspaces are finite-dimensional and the spectrum is bounded below. The conformal vector is unchanged. Restrict the first and third entries to . The displayed lowest-weight coefficient remains nonzero. Decomposing among simple -constituents gives one with a nonzero three-point block with the other two modules . Its lowest weight is exactly : this weight occurs in , whereas the entire ambient module has no smaller weight.
Contragredient compatibility [18 Theorem 2.23] makes self-contragredient. Its nondegenerate graded pairing pairs its unique lowest-weight line with itself; hence . Rigidity of now implies . The dual has the same minimum. Moreover , and flip identifies the latter with as -modules. Thus also arises from the specified middle module.
Return to . For , let be the unique simple -module containing the one-dimensional subspace of . Then
The automorphism normalizes , fixes pointwise, and sends to .
Proposition 4.2. For all ,
Proof. Write as an -module. Its summands are simple currents indexed by , and
is a simple-current extension of of dimension . For a direct dimension justification, acts on , its fixed algebra is , and acts on the -summand by . The tensor subcategory generated by these summands is [DNR25, Theorem 6.1]. Consequently each summand in (17) is invertible in and has Frobenius–Perron dimension one. Moreover , as an algebra object in this subcategory, corresponds to [DNR25, Theorem 6.3]. In particular the multiplication of any two character summands is nonzero.
The restriction contains the module in Lemma 4.1: the permutation-twisted module has underlying vector space , and its subspace has the permutation-twisted -module structure . Since the fix , every contains .
Let be the -graded fusion category of untwisted and -twisted -modules, and put . The covering calculation gives
Indeed, the covering sphere has the labels , , , , where the duals arise from the target . Removing the two vacuum labels leaves the contragredient pairing. The dimension formula is [18 Corollary 2.28]. To interpret it as the displayed categorical Hom-space, use [18 Remark 4.6 and the discussion preceding Proposition 4.7] and the tensor-product universal property in [31 discussion following Remark 4.16]. The required -modules are finite-length ordinary modules, hence finitely generated, and is -cofinite. Consequently
The stabilizer of is , whose irreducible projective representations are one-dimensional because . Thus the equivariant simple object corresponding to forgets to . The forgetful tensor functor preserves Frobenius–Perron dimension, so .
For the extension , the identification of local algebra modules with VOA modules is braided monoidal by [7 Theorem 3.65]. Both VOAs are strongly rational, and is a finite object of , as required there. The algebra has a nondegenerate invariant Frobenius pairing and trivial twist. The dimension formula [KO01, Theorem 1.18], together with the positive categorical dimensions already verified above, gives
Positivity of Frobenius–Perron dimensions forces
We next justify locality of free induction before using it. Every simple-current summand of acts nontrivially on the simple underlying object . Otherwise the module axiom, the nonzero product in , and the unit action give a contradiction. Thus . Since , the balancing identity shows that the double braiding of and is the identity. Hence centralizes the algebra , as do and all its simple summands. By [7 Proposition 2.65 and Theorem 2.67], free induction is local and braided monoidal on this centralizer. The preceding braided equivalence identifies its local tensor product with VOA fusion. Frobenius reciprocity [KO01, Theorem 1.6], (18), and dimensions therefore give
The monoidality is the usual free-module identity
for objects centralizing the commutative algebra .
Fix , and choose an -constituent of containing a lowest-weight vector. Then , and also occurs in , since . Lemma 4.1 gives a constituent of , also arising in , with . The unit summand of gives an injective -map ; thus some simple -summand of contains a vector of weight .
We claim that . Using , put . Then
Thus occurs in both and . Twisted modules labeled by different -conjugacy classes have disjoint simple -constituents [DNR25, Theorem 5.1(1)]. Since is central in , its conjugacy class is a singleton, so . The displayed induced product decomposition consequently makes one of .
Every module with twisting label occurs in , whose minimum is . Hence . The normalizer translations act transitively on the pairs and preserve conformal weight, proving the equality for every pair. For , it is , since has exponent two.
Coherent coordinates and all weight-two obstructions
Define
Its polar form is denoted by . For , put
Euler’s identity for a cubic in characteristic two gives . Cubic expansion then gives
On the actual fusion group, Propositions 2.1 and 4.2 give
Proposition 5.1. There is a unique linear isometry sending to . For the character
its corresponding simple module has image .
Proof. The vectors span . Indeed, a linear functional vanishing on all of them would give an identically zero Boolean polynomial
In the coordinates (5), is a sum of 45 distinct squarefree cubic monomials. Each of its 27 derivatives has five quadratic monomials, and these 135 monomials are mutually distinct. The derivative polynomials are therefore independent. Constant, linear, quadratic, and cubic squarefree monomials are independent as functions on , so all coefficients vanish.
Because the form on is nondegenerate, the Gram matrix of the has rank 56. The equal Gram matrix of the actual forces these to span the 56-dimensional space . A relation in either family pairs to zero with the full other family, so nondegeneracy gives the same relation there. Thus the assignment extends to an isometry of polar forms. Both quadratic forms vanish on the spanning family; their difference is linear and hence zero. This proves existence and uniqueness.
The module indexed by (5.3) pairs with as . The vector has exactly these pairings with every . Nondegeneracy proves the final assertion.
Lowest-weight spaces for rank-one diagonal elements
For rank-one , put
Since , cubic expansion yields
Lemma 5.2. For every nonzero rank-one , the linear map has rank 10, and is constant on its fibers. Consequently (22) gives exactly distinct vectors as varies.
Proof. The -action preserves and is transitive on rank-one vectors. It suffices to take in (5). Direct differentiation gives
This consists of five disjoint hyperbolic pairs; its remaining seventeen coordinates form a radical on which it is identically zero. Its polar map is , proving the claims for this . Norm invariance gives
which transports the assertions to every rank-one vector.
By Proposition 4.2, all 1024 corresponding distinct -modules have weight two. The full -twisted module has a lowest-weight space of dimension
Each of these distinct modules contributes at least one line to this space. Equality proves that they exhaust it and that each has a one-dimensional lowest-weight space.
Proposition 5.3. The nonvacuum simple -modules of integral lowest weight at most two all have weight two. Their images under form the disjoint union
Their respective sizes are
and their total size is 277347807.
Proof. The ordinary module has . Its -eigenspaces and (4) give (24), with the vacuum constituent omitted. The preceding dimension argument gives exactly (25) for rank-one diagonal twisting labels. The other diagonal minima are 3 and . Every has a one-dimensional weight-two subspace, giving (26). The ordinary module has no weight-one vectors, and every nontrivially twisted module has minimum at least two. Completeness of their irreducible -constituents excludes additional cases.
An explicit maximal isotropic graph
For and an alternating linear map , define
The first two coordinates show that this is 28-dimensional. Moreover
Thus . Its intersection with is zero, where the coordinates of Proposition 5.1 are understood.
Use row-major coordinates in , then , then . The entry of these matrices has index , , and , respectively. Encode a vector as . Covectors use the dual basis and the same encoding. All the following numbers are decimal:
The 27 columns , in order, are
Proposition 6.1 (Exact finite certificate). The map in (29) is alternating, and the graph (27) is disjoint from the three obstruction sets in Proposition 5.3.
Proof by exhaustive binary arithmetic. Symmetry and zero diagonal of are checked entrywise. The set has already been excluded.
An intersection with is equivalent to
Indeed contraction with and Euler’s identity then force , the remaining required coordinate equality. Exhaustion of all values of gives no solution of (30).
For the diagonal obstruction, enumerate every nonzero with . There are 139503. Solve
If no solution exists, that gives no intersection. If is a particular solution, Lemma 5.2 shows that the entire solution fiber has the same value . The solution fiber is disjoint from the graph precisely when .
Exactly 139500 systems are inconsistent. The other three are:
| 37467505 | 847 | 0 | 1 |
| 102838768 | 16442 | 1 | 0 |
| 122627027 | 787 | 0 | 1 |
Table 6.
Thus every solution fiber is disjoint from the graph. The complete program in Appendix § 10 implements these tests using only (5), its coordinate gradient, and Gaussian elimination over .
The verifier additionally confirms rank 10 for every rank-one Hessian and checks on all seventeen kernel basis vectors, for a total of 2371551 such checks. Since the restriction of a quadratic form to its polar radical is linear, these basis tests independently establish constancy on every solution fiber. The optimized and undefined-behavior-sanitized builds both passed on 19 November 2026 with GCC 14.2.0. The source in Appendix § 10 has SHA-256
The finite check is exhaustive, not a random search or a sample of rank-one representatives.
Twenty-eight actual simple-current generators
Let denote the irreducible constituent of indexed by the character in (21), and let be the row-major coordinate basis of . Since the cubic is squarefree, and . The following 28 simple currents generate :
Their images are and , respectively. They are independent, pairwise orthogonal, and singular.
The algebra object and the self-dual extension
Two cohomology classes occur here. The ordinary class vanishes by Proposition 3.2. Separately, has an abelian intertwining algebra structure [40 Theorem 4.1]. Its associativity and commutativity data form an abelian 3-cocycle , whose diagonal values, in the convention of that theorem, are
Here takes only the values , so . Hence these diagonal values determine the same pre-metric group as the pointed braided category. This abelian class is nontrivial on the full fusion group , although . Its restriction to an isotropic subgroup is trivial in abelian third cohomology, so both coherences can be corrected at once. The trivialization and the resulting extension are the assertions of [40 Theorems 4.1 and 4.2].
An explicit algebra object is obtained as follows. Write
In the standard hyperbolic pointed model, choose trivial associator and braiding
Let be the strictly upper-triangular part of , so . On the object
define multiplication on the homogeneous summands, relative to the standard fusion identifications, by
Bilinearity makes a normalized 2-cocycle:
Furthermore
which is braided commutativity. The unit is the summand. This finite twisted group-algebra object is separable over (the usual separability idempotent is divided by ). It is connected, has trivial twist, and
Thus it is Lagrangian. The quadratic isometry of Proposition 5.1 lifts to a braided equivalence of pointed categories, by their classification by pre-metric groups [9 Section 5.3]. Thus this algebra transports to . Alternatively the simple-current extension theorem applies directly, without making such a skeletal choice.
Proof of Theorem 1.1. The fixed-point VOA is strongly rational and every nonvacuum irreducible -module has positive lowest weight, as checked before (14). The group is isotropic and self-perpendicular. The extension theorem [40 Theorem 4.2] therefore gives a strongly rational self-dual VOA
extending with the same conformal vector and central charge 64. Every included simple module has integral conformal weight. Propositions 5.3 and 6.1 exclude all nonvacuum modules of weight at most two. Equation (15) consequently gives and , as asserted. □
The graded character
The candidate extremal character at and its first highest-weight multiplicities already occur in [2, 25]. The calculation here is a check on the constructed extension, not an independent existence argument. In this section only, denotes the Fourier variable. Let
The modular-invariance theorem for rational -cofinite VOAs [40 Theorems 2.1 and 2.2] implies
Here the -scalar is 1: it is by the contragredience relation, and evaluation at , where the graded character is strictly positive, fixes the sign. The transformation law of therefore shows that is a modular form of weight 32 for , with no pole at the cusp. The space of such forms has basis
Equivalently, set , so . The three coefficients are uniquely determined by , , and . This gives
The branch of is fixed by .
The dimension of the weight-three Virasoro highest-weight space is 278511. Indeed, since and ,
For a vector in this kernel, vanishes because , and gives . All higher positive modes vanish by the grading. The kernel is therefore exactly the highest-weight space, of dimension .
For an exact formal-series check, put
then . Also define the auxiliary formal series
Exact expansion gives nonnegative integral coefficients of both and through degree 500. This finite calculation is not used in the existence proof. Apart from the weight-three calculation above, no interpretation of the coefficients of as highest-weight multiplicities is needed or asserted here.
References
Standalone exhaustive verifier
The following source is self-contained and uses no nonstandard libraries. Save it as bw64_matrix_certificate.cpp, then run
c++ -O3 -std=c++17 bw64_matrix_certificate.cpp -o c64_check
./c64_check
An additional diagnostic build is
c++ -O1 -g -std=c++17 -fsanitize=undefined \
-fno-sanitize-recover=undefined \
bw64_matrix_certificate.cpp -o c64_check_ubsan
./c64_check_ubsan
The coefficient arrays, bit convention, and norm are exactly those in § 6. The polynomial formula checks on supports of size at most three (respectively two) are exhaustive coefficient checks for squarefree cubics (respectively quadratics). The subsequent open-cell loop ranges over all vectors.
1 // Standalone exact verifier for the c=64 finite certificate.
2 // Build: c++ -O3 -std=c++17 bw64_matrix_certificate.cpp -o bw64_matrix_certificate
3 // Only N(X,Y,Z)=det(X)+det(Y)+det(Z)+tr(XYZ) over F_2 is used.
4 // Bits 0..8, 9..17, 18..26 are the row-major entries of X,Y,Z.
5 // No octonions, change of coordinates, or dual-adjoint identities are used.
6 #include <algorithm>
7 #include <array>
8 #include <chrono>
9 #include <cstdint>
10 #include <cstdlib>
11 #include <iostream>
12 #include <vector>
13
14 using U = std::uint32_t;
15 constexpr int n = 27;
16 constexpr U limit = U(1) << n;
17 constexpr U ell = 1701044;
18 constexpr std::array<U,n> M = {
19 8745196,110867808,95730417,99134097,22725868,41716247,128519191,
20 37605405,89711618,58241068,12720466,113755244,38528647,78467855,
21 11123089,36920254,91945155,99458066,25831589,111637498,6390710,
22 131850988,3350328,53374031,44499804,27826914,2828622
23 };
24 std::array<std::uint16_t,512> cof;
25 std::array<unsigned char,512> det;
26 std::uint16_t tp[512][512]; // transposed ordinary matrix product
27 U mt[3][512];
28 std::vector<std::array<int,3>> terms;
29
30 void require(bool ok, const char* message) {
31 if (!ok) { std::cerr << "FAIL: " << message << '\n'; std::exit(1); }
32 }
33 int bit(U a, int i) { return (a >> i) & 1; }
34 int parity(U a) { return __builtin_parity(a); }
35 U linear(const std::array<U,n>& cols, U a) {
36 U b=0;
37 while (a) { int i=__builtin_ctz(a); b^=cols[i]; a&=a-1; }
38 return b;
39 }
40 U gradient(U a) {
41 U x=a&511, y=(a>>9)&511, z=a>>18;
42 return U(cof[x]^tp[y][z]) | (U(cof[y]^tp[z][x])<<9)
43 | (U(cof[z]^tp[x][y])<<18);
44 }
45 int norm(U a) {
46 U x=a&511, y=(a>>9)&511, z=a>>18;
47 return det[x]^det[y]^det[z]^parity(tp[x][y]&z);
48 }
49 U gradient_by_terms(U a) {
50 U b=0;
51 for (auto t:terms) for (int j=0;j<3;++j)
52 if (bit(a,t[(j+1)%3]) && bit(a,t[(j+2)%3])) b^=U(1)<<t[j];
53 return b;
54 }
55 int norm_by_terms(U a) {
56 int b=0;
57 for (auto t:terms) b^=bit(a,t[0])&bit(a,t[1])&bit(a,t[2]);
58 return b;
59 }
60 void initialize() {
61 require(ell<limit,"ell outside 27 bits");
62 for (int i=0;i<n;++i) {
63 require(M[i]<limit && !bit(M[i],i),"M not alternating");
64 for (int j=0;j<n;++j) require(bit(M[i],j)==bit(M[j],i),"M not symmetric");
65 }
66 for (int base:{0,9,18}) {
67 std::array<int,3> p={0,1,2};
68 do { terms.push_back({base+p[0],base+3+p[1],base+6+p[2]}); }
69 while (std::next_permutation(p.begin(),p.end()));
70 }
71 for (int i=0;i<3;++i) for (int j=0;j<3;++j) for (int k=0;k<3;++k)
72 terms.push_back({3*i+j,9+3*j+k,18+3*k+i});
73 require(terms.size()==45,"wrong cubic monomial count");
74 for (U x=0;x<512;++x) {
75 std::array<int,3> p={0,1,2};
76 do { det[x]^=bit(x,p[0])&bit(x,3+p[1])&bit(x,6+p[2]); }
77 while (std::next_permutation(p.begin(),p.end()));
78 for (int i=0;i<3;++i) for (int j=0;j<3;++j) {
79 int r[2],c[2],nr=0,nc=0;
80 for (int k=0;k<3;++k) { if(k!=i)r[nr++]=k; if(k!=j)c[nc++]=k; }
81 int v=(bit(x,3*r[0]+c[0])&bit(x,3*r[1]+c[1]))
82 ^(bit(x,3*r[0]+c[1])&bit(x,3*r[1]+c[0]));
83 cof[x]|=U(v)<<(3*i+j);
84 }
85 for (U y=0;y<512;++y) for (int i=0;i<3;++i) for (int j=0;j<3;++j) {
86 int v=0;
87 for (int k=0;k<3;++k) v^=bit(x,3*i+k)&bit(y,3*k+j);
88 tp[x][y]|=U(v)<<(3*j+i);
89 }
90 for (int b=0;b<3;++b) mt[b][x]=linear(M,x<<(9*b));
91 }
92 // The two norm formulas are squarefree cubics, and the two gradient
93 // formulas are squarefree quadratics. Values on supports of size <=d
94 // determine every squarefree Boolean polynomial of degree <=d exactly.
95 unsigned nc=0,gc=0;
96 auto check=[&](U a,int weight) {
97 require(norm(a)==norm_by_terms(a),"cubic formula mismatch"); ++nc;
98 if(weight<=2) { require(gradient(a)==gradient_by_terms(a),"gradient mismatch"); ++gc; }
99 };
100 check(0,0);
101 for(int i=0;i<n;++i) {
102 check(U(1)<<i,1);
103 for(int j=i+1;j<n;++j) {
104 check((U(1)<<i)|(U(1)<<j),2);
105 for(int k=j+1;k<n;++k) check((U(1)<<i)|(U(1)<<j)|(U(1)<<k),3);
106 }
107 }
108 require(nc==3304 && gc==379,"incomplete polynomial checks");
109 std::cout<<"Exact formula checks: 3304 cubic, 379 gradient; M alternating.\n";
110 }
111
112 int main() {
113 const auto start=std::chrono::steady_clock::now();
114 initialize();
115 std::vector<U> rankone;
116 std::uint64_t tested=0,bigbad=0;
117 for(U z=0;z<512;++z) for(U y=0;y<512;++y) {
118 U yz=(y<<9)|(z<<18), myz=mt[1][y]^mt[2][z];
119 for(U x=0;x<512;++x) {
120 U a=x|yz, ga=gradient(a); ++tested;
121 bigbad+=(ga==ell^mt[0][x]^myz);
122 if(a && !ga) rankone.push_back(a);
123 }
``
}
require(tested==limit,"incomplete big-cell enumeration");
require(bigbad==0,"big-cell collision");
require(rankone.size()==139503,"wrong rank-one count");
std::cout<<"Big cell: "<<tested<<" vectors, "<<bigbad<<" collisions; rank one: "
<<rankone.size()<<’.’<<std::endl;
std::uint64_t kernelchecks=0,solvable=0,outside=0,boundarybad=0;
for(U u:rankone) {
require(!norm(u),"rank-one norm nonzero");
std::array<U,n> h{},piv{},pre{};
std::vector<U> kernel;
int rank=0;
for(int i=0;i<n;++i) h[i]=gradient(u^(U(1)<<i)); // grad(u)=grad(e_i)=0
for(int i=0;i<n;++i) {
require(!bit(h[i],i),"Hessian diagonal nonzero");
for(int j=0;j<n;++j) require(bit(h[i],j)==bit(h[j],i),"Hessian not symmetric");
U v=h[i],a=U(1)<<i;
while(v) {
int k=31-__builtin_clz(v);
if(piv[k]) { v^=piv[k]; a^=pre[k]; }
else { piv[k]=v; pre[k]=a; ++rank; break; }
}
if(!v) {
require((a>>i)==1,"kernel generators not triangular independent");
kernel.push_back(a);
}
}
require(rank==10 && kernel.size()==17,"rank-one Hessian rank not 10");
// beta_u(a)=<grad(a),u> is quadratic with polar form H_u.
// For k in ker(H_u), beta_u(a+k)=beta_u(a)+beta_u(k).
// Its restriction to ker(H_u) is linear. Checking beta on this
// complete kernel basis therefore proves constancy on EVERY fiber.
for(U k:kernel) {
require(k && !linear(h,k),"invalid kernel vector");
require(!parity(gradient(k)&u),"beta not constant on a Hessian fiber");
++kernelchecks;
}
U rhs=linear(M,u),v=rhs,a=0;
while(v) {
int k=31-__builtin_clz(v);
if(!piv[k]) break;
v^=piv[k]; a^=pre[k];
}
if(v) { ++outside; continue; }
++solvable;
require(linear(h,a)==rhs,"incorrect particular solution");
boundarybad+=(parity(gradient(a)&u)==parity(ell&u));
std::cout<<"Solvable boundary u="<<u<<", witness a="<<a
<<", beta="<<parity(gradient(a)&u)<<", ell(u)="<<parity(ell&u)<<’.’<<’\n’;
}
require(kernelchecks==std::uint64_t(139503)*17,"incomplete kernel checks");
require(solvable+outside==139503,"incomplete boundary enumeration");
require(boundarybad==0,"boundary collision");
std::cout<<"Boundary: "<<outside<<" outside Hessian image, "<<solvable
<<" solvable with opposite beta, "<<boundarybad<<" collisions.\n"
<<"Every rank-one Hessian has rank 10; "<<kernelchecks
<<" kernel-basis beta checks passed.\n";
// K={(alpha,a;ell(a),alpha*ell+M*a)} has dimension 28 by its first
// coordinates and Q|K=0 because M is alternating. It contains no nonzero
// vector in the label-map kernel. Equality ell(a)=ell+M*a automatically
// implies ell(a)=N(a), by cubic Euler and a.M.a=0. For rank-one u,
// the entire boundary is (0,b;beta_u(a),H_u*a), exactly the test above.
std::cout<<"PASS: the 28-dimensional totally singular graph avoids all three bad sets.\n"
<<"Elapsed seconds: "<<std::chrono::duration<double>(std::chrono::steady_clock::now()-
start).count()
<<std::endl;
}References
- [1]Toshiyuki Abe, Geoffrey Buhl, and Chongying Dong. Rationality, regularity, and C₂-cofiniteness. Transactions of the American Mathematical Society, 356(8):3391–3402, 2004. Theorem numbers follow arXiv:math/0204021v1; doi:10.1090/S0002-9947-03-03413-5.
- [2]Spyros D. Avramis, Alex Kehagias, and Constantina Mattheopoulou. Three-dimensional AdS gravity and extremal CFTs at c = 8m. Journal of High Energy Physics, 11:022, 2007. arXiv:0708.3386.DOI
- [3]Maryam Ashrafi. Three dimensional pure gravity and generalized Hecke operators. Journal of High Energy Physics, 09:083, 2020. arXiv:2004.04424.DOI
- [4]Nathan Benjamin, Ethan Dyer, A. Liam Fitzpatrick, and Shamit Kachru. An extremal N = 2 superconformal field theory. Journal of Physics A: Mathematical and Theoretical, 48(49):495401, 2015. arXiv:1507.00004.DOI
- [5]Katrina Barron, Chongying Dong, and Geoffrey Mason. Twisted sectors for tensor product vertex operator algebras associated to permutation groups. Communications in Mathematical Physics, 227:349–384, 2002. Theorem numbers follow arXiv:math/9803118v2; doi:10.1007/s002200200633.DOI
- [6]Arjeh M. Cohen and Bruce N. Cooperstein. The 2-spaces of the standard E₆(q)-module. Geometriae Dedicata, 25:467–480, 1988. Author-institute PDF.
- [7]Thomas Creutzig, Shashank Kanade, and Robert McRae. Tensor categories for vertex operator superalgebra extensions. Memoirs of the American Mathematical Society, 295(1472), 2024. Theorem numbers follow arXiv:1705.05017v2; doi:10.1090/memo/1472.DOI
- [8]Scott Carnahan and Masahiko Miyamoto. Regularity of fixed-point vertex operator subalgebras. arXiv:1603.05645v4, 2018. Theorem and corollary numbers follow this version.arxiv.org/abs/1603.05645
- [9]Alexei Davydov, Michael Müger, Dmitri Nikshych, and Victor Ostrik. The Witt group of non-degenerate braided fusion categories. arXiv:1009.2117v2, 2011. Theorem and example numbers follow this version.arxiv.org/abs/1009.2117
- [11]Matthias R. Gaberdiel. Constraints on extremal self-dual CFTs. Journal of High Energy Physics, 11:087, 2007. arXiv:0707.4073.DOI
- [12]Davide Gaiotto. Monster symmetry and extremal CFTs. arXiv:0801.0988, 2008.arxiv.org/abs/0801.0988
- [13]Matthias R. Gaberdiel, Sergei Gukov, Christoph A. Keller, Gregory W. Moore, and Hirosi Ooguri. Extremal N = (2, 2) 2D conformal field theories and constraints of modularity. Communications in Number Theory and Physics, 2(4):743–801, 2008. arXiv:0805.4216.arxiv.org/abs/0805.4216
- [14]Matthias R. Gaberdiel and Christoph A. Keller. Modular differential equations and null vectors. Journal of High Energy Physics, 09:079, 2008. arXiv:0804.0489v3.DOI
- [15]Thomas Gemünden and Christoph A. Keller. Orbifolds of lattice vertex operator algebras at d = 48 and d = 72. Journal of Algebra, 523:93–118, 2019. arXiv:1802.10581.arxiv.org/abs/1802.10581
- [16]Thomas Gemünden and Christoph A. Keller. Non-abelian orbifolds of lattice vertex operator algebras. Journal of Algebra, 585:656–696, 2021. doi:10.1016/j.jalgebra.2021.06.023.DOI
- [17]Matthias R. Gaberdiel, Christoph A. Keller, and Roberto Volpato. Genus two partition functions of chiral conformal field theories. Communications in Number Theory and Physics, 4(2):295–363, 2010. arXiv:1002.3371v2.arxiv.org/abs/1002.3371
- [18]Bin Gui. Genus-zero permutation-twisted conformal blocks for tensor product vertex operator algebras: the tensor-factorizable case. Communications in Contemporary Mathematics, 28(2):2550034, 2026. Theorem numbers follow arXiv:2111.04662v4 (16 March 2025); doi:10.1142/S0219199725500348.DOI
- [19]Davide Gaiotto and Xi Yin. Genus two partition functions of extremal conformal field theories. Journal of High Energy Physics, 08:029, 2007. arXiv:0707.3437.DOI
- [20]Sarah M. Harrison. Extremal chiral N = 4 SCFT with c = 24. Journal of High Energy Physics, 11:006, 2016. arXiv:1602.06930.DOI
- [21]Gerald Höhn. Selbstduale Vertexoperatorsuperalgebren und das Babymonster. PhD thesis, Rheinische Friedrich-Wilhelms-Universität Bonn, 1995. Bonner Mathematische Schriften 286 (1996), 1–85; arXiv:0706.0236.
- [22]Gerald Höhn. Conformal designs based on vertex operator algebras. Advances in Mathematics, 217(5):2301–2335, 2008. Preprint first posted 23 January 2007; arXiv:math/0701626; doi:10.1016/j.aim.2007.11.003.DOI
- [23]Gerald Höhn. Self-dual vertex operator superalgebras of large minimal weight. arXiv:0801.1822, 2008.
- [24]Yi-Zhi Huang. Rigidity and modularity of vertex tensor categories. arXiv:math/0502533v2, 2007. Theorem numbers follow this version.arxiv.org/abs/math/0502533
- [25]Christoph A. Keller, Grégoire Mathys, and Ida G. Zadeh. Bootstrapping chiral CFTs at genus two. Advances in Theoretical and Mathematical Physics, 22(6):1447–1487, 2018. arXiv:1705.05862.arxiv.org/abs/1705.05862
- [28]Ying-Hsuan Lin and Du Pei. Holomorphic CFTs and topological modular forms. Communications in Mathematical Physics, 401(1):325–332, 2023. arXiv:2112.10724.DOI
- [30]Jan Manschot. AdS₃ partition functions reconstructed. Journal of High Energy Physics, 10:103, 2007. arXiv:0707.1159.
- [31]Robert McRae. Twisted modules and G-equivariantization in logarithmic conformal field theory. Communications in Mathematical Physics, 383(3):1939–2019, 2021. Theorem numbers follow arXiv:1910.13226v2; doi:10.1007/s00220-020-03882-2.DOI
- [33]Alexander Maloney and Edward Witten. Quantum gravity partition functions in three dimensions. Journal of High Energy Physics, 02:029, 2010. arXiv:0712.0155.DOI
- [34]Gabriele Nebe. An even unimodular 72-dimensional lattice of minimum 8. Journal für die reine und angewandte Mathematik, 673:237–247, 2012. Preprint first posted August 2010; arXiv:1008.2862.arxiv.org/abs/1008.2862
- [35]Gabriele Nebe. A fourth extremal even unimodular lattice of dimension 48. Discrete Mathematics, 331:133–136, 2014. arXiv:1312.3780; doi:10.1016/j.disc.2014.05.011.DOI
- [36]Ryotaro Sakamoto. Two extremal even unimodular lattices of rank 88. arXiv:2609.15633v2, 2026. Version 2, 27 September 2026.arxiv.org/abs/2609.15633
- [37]Hiroki Shimakura. The automorphism group of the Z₂-orbifold of the Barnes–Wall lattice vertex operator algebra of central charge 32. Mathematical Proceedings of the Cambridge Philosophical Society, 156(2):343–361, 2014. Theorem numbers follow arXiv:1310.6832; doi:10.1017/S0305004113000704.DOI
- [38]Ichiro Shimada. An even extremal lattice of rank 64. Journal of Number Theory, 185:1–15, 2018. arXiv:1710.02629; doi:10.1016/j.jnt.2017.10.028.DOI
- [39][Shi22] Ichiro Shimada. A note on Quebbemann’s extremal lattices of rank 64. Journal de théorie des nombres de Bordeaux, 34(3):813–826, 2022. arXiv:2106.04828; doi:10.5802/jtnb.1229.DOI
- [40][vEMS20] Jethro van Ekeren, Sven Möller, and Nils R. Scheithauer. Construction and classification of holomorphic vertex operator algebras. Journal für die reine und angewandte Mathematik, 759:61–99, 2020. Theorem numbers follow arXiv:1507.08142v3; doi:10.1515/crelle-2017-0046.
- [41][Wit07] Edward Witten. Three-dimensional gravity revisited. arXiv:0706.3359, 2007.arxiv.org/abs/0706.3359
- [42][Yin08] Xi Yin. Partition functions of three-dimensional pure gravity. Communications in Number Theory and Physics, 2(2):285–324, 2008. arXiv:0710.2129.arxiv.org/abs/0710.2129