[en-GB]ord=omit MnLargeSymbols’000 MnLargeSymbols’005 MnLargeSymbols’102 MnLargeSymbols’107 \DeclareFieldFormat*title#1 \DeclareFieldFormat*journaltitle#1 \DeclareFieldFormat[article]version\IfDecimal#1version #1 \DefineBibliographyStringsenglish withafterword = with an appendix by, \DeclareFieldFormatpages \addbibresourcemain.bib
Modules and generalizations of Joyce vertex algebras
Abstract
Joyce vertex algebras are vertex algebra structures defined on the homology of certain -linear moduli stacks, and are used to express wall-crossing formulae for Joyce’s homological enumerative invariants. This paper studies the generalization of this construction to settings that come from non-linear enumerative problems. In the special case of orthosymplectic enumerative geometry, we obtain twisted modules for Joyce vertex algebras.
We expect that our construction will be useful for formulating wall-crossing formulae for enumerative invariants for non-linear moduli stacks. We include several variants of our construction that apply to different flavours of enumerative invariants, including Joyce’s homological invariants, DT4 invariants, and a version of -theoretic enumerative invariants.
\LibertinusDisplay1 \LibertinusDisplayIntroduction
Para 1.1.1
Joyce vertex algebras, introduced by \textcitejoyce-hall,joyce-homological,joyce-surfaces, are a class of vertex algebras defined on the homology of certain moduli stacks. They are motivated by the study of wall-crossing in enumerative geometry, and it has been mysterious whether they are related to usual sources of vertex algebras, such as conformal field theories.
In some cases, this construction gives familiar vertex algebras, such as the Heisenberg vertex algebra, which arises as the homology of the space , or certain Kac–Moody vertex algebras, which arise as the homology of moduli stacks of representations of Dynkin quivers. See §§ 6.1.3 and 6.3.1 below for these examples, and \textcitelatyntsev-thesis for more details.
Para 1.1.2
Motivated by the problem of generalizing Joyce’s [joyce-homological] enumerative invariants and wall-crossing formulae from the setting of abelian categories to more general algebraic stacks, \textcitebu-self-dual-ii considered, as a first step, the orthosymplectic case, involving moduli stacks of orthogonal or symplectic objects and enumerative invariants counting them.
In this case, \textcitebu-self-dual-ii constructed twisted modules for Joyce vertex algebras, defined on the homology of moduli stacks of orthogonal and symplectic objects. These twisted module structures were used to write down wall-crossing formulae for orthosymplectic enumerative invariants. This structure was later also studied by \textcitedehority-latyntsev.
Para 1.1.3
More recently, the author, Halpern-Leistner, Ibáñez Núñez, and Kinjo [epsilon-i, epsilon-ii, epsilon-iii] developed intrinsic Donaldson–Thomas theory, a new framework for enumerative geometry designed for generalizing results from abelian categories to non-linear moduli stacks. We expect this framework to help formulating a generalization of Joyce’s [joyce-homological] formalism to a larger class of quasi-smooth stacks over , including stacks that do not come from linear categories.
Although such a generalized enumerative theory is still elusive, in this paper, we write down a natural generalization of Joyce’s vertex algebras for general stacks, which we call vertex induction, and which also generalizes the twisted modules in [bu-self-dual-ii].
Para 1.1.4
We expect the vertex induction to be the key operation in wall-crossing formulae for these conjectural invariants, which relate the invariants defined for different stability conditions. Moreover, these wall-crossing formulae should have the same structure as those for motivic Donaldson–Thomas invariants in [epsilon-iii]. These expectations generalize Joyce’s work [joyce-homological] in the linear case.
Wall-crossing formulae are a powerful tool in enumerative geometry, as they provide a strong constraint on the structure of the invariants, and can sometimes be used for direct computations which can otherwise be difficult. See \textcitegross-joyce-tanaka-2022, \textcitebu-2023-curves, and \textcitebojko-lim-moreira-2024 for examples of this approach in the linear case, where Joyce vertex algebras were an essential tool.
Para 1.1.5
We hope that the vertex induction defined in the current work, combined with the formalism of stability and wall-crossing formulae for general stacks in [epsilon-i, epsilon-ii, epsilon-iii], will enable us to formulate a precise conjecture for the existence and behaviour of the generalized invariants in § 1.1.3, and will allow us to compute them in some cases, assuming the conjectural wall-crossing formulae.
Moreover, apart from Joyce’s invariants, several other types of enumerative invariants also exhibit very similar wall-crossing behaviours, including DT4 invariants and certain -theoretic invariants, which we will discuss in §§ 1.1.9 to 1.1.10. Their wall-crossing formulae should be expressed using their corresponding variants of the vertex induction, which we also introduce in this paper.
Para 1.1.6 (Joyce vertex algebras)
Before introducing the general construction of vertex induction, we will first explain its special cases where we obtain Joyce vertex algebras, and modules and twisted modules for them, to better motivate the general construction.
Roughly speaking, a vertex algebra is a vector space , equipped with a unit and multiplication maps denoted by
| (1.1.6.1) |
which we may regard as a family of multiplications depending meromorphically on the formal variables , with possible poles along for . This multiplication should be unital, associative, and commutative in suitable senses.
In Section 2, we recall the construction of Joyce vertex algebras, with a precise statement in Theorem 2.3.3. We start with a moduli stack of objects in a -linear abelian category . Then the graded vector space admits a vertex algebra structure given by
| (1.1.6.2) |
where denotes the virtual dimension of , are homology classes, is the direct sum map, is a translation operator depending on the variables , and is the inverse of the equivariant Euler class of the (virtual) normal bundle of the map . This -fold multiplication will be a special case of vertex induction applied to the moduli stack .
Para 1.1.7 (Modules)
In Section 3, we construct modules and twisted modules for Joyce vertex algebras.
The motivating case is from orthosymplectic enumerative geometry, where we start with a moduli stack of objects in an abelian category , equipped with a contravariant involution , which induces a -action on . The homotopy fixed locus is then the moduli of orthogonal or symplectic objects. See \textcitebu-osp-dt for more details on this set-up.
In this case, the involution induces a twisted involution of the Joyce vertex algebra , meaning that it satisfies
| (1.1.7.1) |
for . The space has the structure of a twisted module for , given by the action
| (1.1.7.2) |
where and are homology classes, is the map sending , and now denotes the normal bundle of .
Here, having a twisted module means that we have
| (1.1.7.3) |
where we allow extra poles at for , which are not present in a usual module, and we require the relation
| (1.1.7.4) |
for and . This forces us to allow the extra poles, as it implies relations such as . Note that our notion of twisted modules is different from the one in \textcite[§5.6]frenkel-ben-zvi-2004.
Para 1.1.8 (Vertex induction)
The vertex induction is the main construction of this paper, and generalizes the constructions of Joyce vertex algebras and modules above. We will introduce it in Section 4, with precise statements in Theorems 4.2.3 and 4.3.7.
The motivating situation is when we are given an algebraic stack over , and we consider its stack of graded points
| (1.1.8.1) |
following \textcitehalpern-leistner-instability. It is equipped with a forgetful map .
For example, if is the moduli stack of objects in an abelian category , then is usually the moduli stack of -graded objects in , and the forgetful map forgets the -grading. As another example, if for a linear algebraic group over , then is a disjoint union of stacks of the form for Levi subgroups . See [halpern-leistner-instability, epsilon-i] for details and more examples.
In this case, the vertex induction is the operation
where is a connected component; denotes pushing forward along the forgetful map ; are formal variables, where depends on ; is a certain translation operator; and is the normal bundle of the map .
For example, if is the moduli stack of objects in an abelian category , then each is isomorphic to an -fold product of components of for some , and the vertex induction gives the -fold multiplication in the Joyce vertex algebra. If, moreover, is equipped with a contravariant involution as in § 1.1.7, then each component is isomorphic to a component of for some , and the vertex induction for gives the multiplication in the twisted module.
Para 1.1.9 (DT4 invariants)
In Section 5.1, we discuss a variant of Joyce vertex algebras which appear in the context of DT4 invariants. These are invariants counting sheaves on Calabi–Yau fourfolds, developed by \textcitecao-leung-dt4, \textciteborisov-joyce-2017, and \textciteoh-thomas-2023-i,oh-thomas-ii. Their existence and wall-crossing formulae were conjectured by \textcite[§4.4]gross-joyce-tanaka-2022, and have not yet been proved at the time of this writing.
We define a similar variant of vertex induction on the homology of oriented -shifted symplectic stacks for . For example, -shifted symplectic structures exist on moduli stacks of sheaves on Calabi–Yau fourfolds. The existence of orientations in this case is more subtle, but is recently shown for a class of Calabi–Yau fourfolds by \textcitejoyce-upmeier-bordism.
We expect that a generalized version of DT4 invariants should be defined for a general class of oriented -shifted symplectic stacks, especially in the non-linear cases. We also expect that wall-crossing formulae for these generalized DT4 invariants should be written down using the variant of vertex induction that we define.
Para 1.1.10 (K-theoretic invariants)
In Section 5.2, we consider another variant of Joyce vertex algebras, which are multiplicative vertex algebra structures defined on a version of -homology of moduli stacks, instead of the homology. This was originally due to \textciteliu-k-wall-crossing, who also constructed -theoretic enumerative invariants living in -homology, and proved their wall-crossing formulae using this multiplicative vertex algebra.
We generalize this construction to define a -homology version of vertex induction. We expect that a generalized version of Liu’s -theoretic invariants should be defined for a general class of quasi-smooth stacks, and that they should satisfy wall-crossing formulae expressed using the -theoretic vertex induction.
Para 1.1.11 (Acknowledgements)
The author thanks Dominic Joyce for many helpful comments and suggestions, and Henry Liu for discussions related to the -theory version.
This work was done during the author’s PhD programme supported by the Mathematical Institute, University of Oxford.
Para 1.1.12 (Note)
This work partially supersedes the author’s preprint [bu-self-dual-ii], where the twisted module of Joyce’s vertex algebra was first constructed.
\LibertinusDisplay2 \LibertinusDisplayJoyce vertex algebras
\firamedium2.1 \firamediumVertex algebras
Para 2.1.1
Para 2.1.2 (Formal power series)
Let be a field. For a finite-dimensional -vector space , let be the -algebra of formal power series on with coefficients in , defined as
Elements of are often written as , with a variable . Let be the fraction field of . For a -vector space , define
For finite-dimensional -vector spaces , writing , we have a natural isomorphism , and an injective -module homomorphism
| (2.1.2.1) |
given by power series expansion in the given order. We also denote this map by if is a set of coordinates of . For example, we have
| (2.1.2.2) |
where the notation means that we take , and , are coordinates on , .
This also induces a map for any -vector space .
Para 2.1.3 (Vertex algebras)
We recall the usual definition of a vertex algebra, taken from \textcite[Definition 1.3.1]frenkel-ben-zvi-2004.
A -graded vertex algebra over is a quadruple , consisting of a -graded -vector space , an element of degree , a linear map called the translation operator, and a linear map , satisfying the following axioms:
-
(i)
(Unit) For all , we have , and .
-
(ii)
(Translation) For any , we have
(2.1.3.1) where denotes the commutator, and we have .
-
(iii)
(Locality) For homogeneous elements , the elements
are expansions of the same element in , where denotes the -grading.
Para 2.1.4 (An equivalent definition)
We introduce an alternative definition of a vertex algebra, following ideas from \textcitekim-2011. This definition is perhaps more naturally motivated than the standard one, and will be important for generalizing Joyce vertex algebras to other situations below. See also Theorem 4.1.7 below for a functorial description of this approach.
A -graded vertex algebra over is equivalently the data , consisting of a -graded -vector space and -linear multiplication maps
preserving grading, where , and we invert for . In particular, this includes a map , thought of as the unit. They should satisfy the following properties:
-
(i)
(Unit) For any , we have
(2.1.4.1) More precisely, we have .
-
(ii)
(Commutativity) For any homogeneous elements , and any permutation , we have
(2.1.4.2) where the sign is ‘’ if and only if restricts to an odd permutation on the odd elements.
-
(iii)
(Associativity) For integers and elements , we have
(2.1.4.3) where is the map defined in Eq. 2.1.2.1.
As is common in the literature, we adopt the convenient notation
| (2.1.4.4) |
but the reader should be aware that this is not defined using the individual terms .
Para 2.1.5 (Proof of the equivalence of definitions)
Given a graded vertex algebra as in § 2.1.3, define the maps in § 2.1.4 by
| (2.1.5.1) |
where we take the unique preimage under the embedding
By \textcite[Theorem 3.14]kim-2011, this is well-defined, and satisfies the property § 2.1.4 Item (iii). The properties § 2.1.4 Items (i) to (ii) follow from § 2.1.3 Item (i), Item (iii).
Conversely, given the data as in § 2.1.4, we may define a -graded vertex algebra structure on by setting
for all . Verifying the axioms is straightforward except for Eq. 2.1.3.1, which we prove as follows. Applying the associativity relation Eq. 2.1.4.3 multiple times, we have
as desired. ∎
Para 2.1.6 (Remark)
In fact, in § 2.1.4, it is enough to require the operators for , and only require commutativity and associativity up to terms, since these are enough for converting to the usual definition, and we can then convert it back to obtain all the higher .
Para 2.1.7 (Weak vertex algebras)
We introduce a generalized version of vertex algebras, where for the product , in addition to the usual poles at for , we also allow arbitrary poles along any divisor. As we will see, this type of structure arises naturally from geometry, in the set-up of Section 2.2.
A -graded weak vertex algebra over is the data , consisting of a -graded -vector space and -linear multiplication maps
preserving grading, where , satisfying the properties § 2.1.4 Items (i), (ii), and (iii).
Therefore, such a weak vertex algebra is a vertex algebra if and only if the image of lies in the subspace .
\firamedium2.2 \firamediumModuli spaces
Para 2.2.1
As we mentioned in the introduction, Joyce vertex algebras [joyce-hall, joyce-homological, joyce-surfaces] are vertex algebra structures defined on the homology of certain moduli spaces. Here, we formulate a set of axioms for such moduli spaces.
We use the following terminology and notations:
-
•
A torus is a Lie group isomorphic to for some .
-
•
For a torus , let , be the weight and coweight lattices, and , both isomorphic to .
-
•
is the category whose objects are topological spaces that are homotopy equivalent to CW complexes, and morphisms are homotopy classes of continuous maps.
-
•
is the topological -theory of a space , defined as the abelian group .
Para 2.2.2 (The category )
For convenience of presentation, we introduce a category of spaces with -actions for tori , defined as follows:
-
•
Its objects are triples , where is a torus, is a space, and
is a map, such that it defines a -action on in .
-
•
A morphism consists of a Lie group homomorphism , together with a -equivariant map in .
We often abbreviate the triple as , and call the rank of such an object.
The category admits finite products, giving a symmetric monoidal structure on .
Para 2.2.3 (Weights in K-theory)
Let be an object of . For each weight , let be the subgroup of classes of weight , meaning that , where is the line bundle classified by the map . Let
be the subgroup of classes which admit finite weight decompositions.
Para 2.2.4 (The setting)
We assume given the data , where
-
•
is a space.
-
•
is an action in , giving an object of of rank .
-
•
and are morphisms in , with the diagonal map, defining a commutative monoid structure on in .
-
•
is a class of weight , called the obstruction theory.
In this case, the function is called the virtual dimension of .
For each , let be the -fold product using , and let , where is the -th projection for . Define the (virtual) normal bundle of as the class
| (2.2.4.1) |
We further assume the following condition:
-
•
For any integer , we have
(2.2.4.2) where denotes the part of -weight .
This is a technical condition to ensure that the equivariant Euler class in § 2.3.1 is well-defined, and is satisfied in most of our examples. See Examples 2.2.6 and 2.2.7 below.
Para 2.2.5 (The topological realization of a stack)
To give examples of the data in § 2.2.4, we will use the topological realization functor
| (2.2.5.1) |
as in \textcite[§3.1]blanc-2016, where is the category of affine -schemes, and is the -category of spaces. This is defined as the left Kan extension of the functor sending an affine -scheme to the topological space of its analytification.
For a derived algebraic stack over , we denote by the topological realization of its classical truncation defined as above.
Example 2.2.6
We give a class of examples of the data as in § 2.2.4.
Let be a derived linear moduli stack over in the sense of \textcite[§7.1]epsilon-i and [bu-davison-ibanez-nunez-kinjo-padurariu, §2.4.6], so that it is equipped with a monoid structure and an action
satisfying higher coherence and compatibility conditions. Typical examples include the following:
-
•
The derived moduli stack of representations of a quiver, possibly with potential.
-
•
The derived moduli stack of vector bundles, or coherent sheaves, on a smooth, projective -variety.
Let be the topological realization as in § 2.2.5. The data on are defined by those on . The class is defined as the class of the tangent complex of .
The condition Eq. 2.2.4.2 usually holds in this case, since if is a moduli stack of objects in a -linear abelian category , then we often have at a point , so only has -weights of the form for , and the weight part agrees with .
Example 2.2.7
Another class of examples of as in § 2.2.4 are given as follows.
Let be a -linear dg-category of finite type in the sense of \textcite[Definition 2.4]toen-vaquie-2007-moduli, and let be the derived moduli stack of right proper objects in , as in [toen-vaquie-2007-moduli, Theorem 3.6]. Examples include the following:
-
•
The derived moduli stack of complexes of representations of a quiver.
-
•
The derived moduli stack of perfect complexes on a smooth, projective -variety.
Let be the topological realization of , and define the data using the scalar product, the direct sum, the zero object, and the tangent complex of .
In this case, the condition Eq. 2.2.4.2 always holds, by the description of the tangent complex of by \textcite[Proposition 3.9]brav-dyckerhoff-2021.
For example, if is the classifying stack of perfect complexes over , as in \textcite[Proposition 3.7]toen-vaquie-2007-moduli, then we have an equivalence , by \textcite[Theorems 4.5 and 4.21]blanc-2016.
\firamedium2.3 \firamediumJoyce vertex algebras
Para 2.3.1 (The equivariant Euler class)
Let , and let be a class such that the weight part is the class of a vector bundle on , possibly of mixed rank. Define
| (2.3.1.1) |
where is a set of coordinates on , and . For each , the part of lying in is a rational function in , with possible poles at for such that is not the class of a vector bundle, although there may not be a uniform bound for the orders of poles as increases.
We have the relation
and in particular, if , then , in which case we also write .
Para 2.3.2 (The translation operator)
Let . Define its translation operator
with and , where we choose an identification , and define by
where is the generator of the -th copy of , dual to the universal first Chern class in .
This construction does not depend on the choice of identification , provided that the coordinates on are transformed accordingly.
Theorem 2.3.3
Let be as in § 2.2.4. Then the assignment
| (2.3.3.1) |
equips the space with the structure of a -graded weak vertex algebra over , where , and is the translation operator of the object .
The product Eq. 2.3.3.1 has poles along for non-zero weights appearing in . In particular, if the class only has -weights and then is a -graded vertex algebra.
-
Proof.
The property Eq. 2.1.4.1 holds since . The relation Eq. 2.1.4.2 holds since the definition Eq. 2.3.3.1 is symmetric (with a sign rule) in the indices .
We now verify Eq. 2.1.4.3. Write and for short. Expanding the left-hand side of Eq. 2.1.4.3, we obtain
where is short for , and is short for . The indices are implicitly summed over the ranges and , respectively. For , we denote by the pullback of from the -th copy of . The second step uses Lemma 2.3.4 below, and the third step uses the fact that has weight zero with respect to the diagonal torus, associated to the variable . This computation agrees with the right-hand side of Eq. 2.1.4.3.
For the final statement on usual vertex algebras, it is enough to show that for each , the class can only have -weights of the form for , where with at the -th position. This is because the assumption implies that for any weight appearing in and any partition , writing and , we have , which implies the desired property. ∎
Lemma 2.3.4
Let , and let , such that its weight part is the class of a vector bundle on . Then for any , we have
| (2.3.4.1) |
where are sets of variables, with , and denotes the expansion map from to .
-
Proof.
We have
where for a choice of identification , and each is the universal first Chern class in the -th copy of , and we expand in positive powers of . In the final line, we pull back along the projection to the second factor, .
Let be the dual of as in § 2.3.2, and equip with a ring structure given by the group structure of in . Then
where denotes the set of variables , and we expand in non-negative powers of . This proves the lemma. ∎
Para 2.3.5 (Morphisms of Joyce vertex algebras)
Finally, we mention a construction of morphisms of Joyce vertex algebras, as in \textcite[§2.5]gross-joyce-tanaka-2022.
Theorem
Let be two spaces as in § 2.2.4, and let be a map in , respecting the monoid structures and -actions. Assume that the class
is the class of a vector bundle on , possibly of mixed rank. Then the map
is a morphism of -graded weak vertex algebras, where denotes the Euler class.
Moreover, if and are maps as above, then .
-
Proof.
Let . Then
where the second last step follows from Lemma 2.3.4 by setting in the lemma, which is possible since is a vector bundle, so does not have negative powers of .
The final statement is elementary. ∎
\LibertinusDisplay3 \LibertinusDisplayModules
We present a construction of modules for Joyce vertex algebras from moduli spaces, which we state in Theorem 3.2.2. We also consider an important variant of this construction in Theorem 3.2.6, arising from orthosymplectic enumerative geometry, where the Joyce vertex algebra admits an involution, and we obtain a twisted module for this vertex algebra.
\firamedium3.1 \firamediumVertex algebra modules
Para 3.1.1 (Modules)
We recall the definition of vertex algebra modules from \textcite[§5.1]frenkel-ben-zvi-2004.
Let be a -graded vertex algebra over , as in § 2.1.3. Then a -module is a -graded -vector space , equipped with an operation , preserving grading with , satisfying the following conditions:
-
(i)
(Unit) For any , we have .
-
(ii)
(Associativity) For any and , the elements
are expansions of the same element in .
Para 3.1.2 (An equivalent definition)
We can also give an alternative definition of a vertex algebra module, in the style of § 2.1.4.
Let be a -graded vertex algebra over , in the sense of § 2.1.4. Then a -module is a -graded -vector space , equipped with operations
preserving grading, where , and we invert for . They should satisfy the following properties:
-
(i)
(Unit) We have .
-
(ii)
(Associativity) For integers and elements , we have
(3.1.2.1) (3.1.2.2) where the maps are defined in Eq. 2.1.2.1.
Again, this definition may look more complicated than § 3.1.1, but it will be more convenient for our purposes, such as for defining weaker notions of modules below.
We adopt the convenient notation
| (3.1.2.3) |
with the same caveats as those below Eq. 2.1.4.4 for vertex algebras.
Para 3.1.3 (Proof of the equivalence of definitions)
We essentially follow the proof of \textcite[Theorem 3.14]kim-2011, adapting it to the case of modules.
Given a -module in the sense of § 3.1.1, define the maps by
| (3.1.3.1) |
where we take the unique preimage under the embedding
| (3.1.3.2) |
To see that such a preimage exists, note that for fixed elements as above, there exists such that in the expression
| (3.1.3.3) |
the order of the operators can be permuted freely without changing the result, up to an appropriate sign. Indeed, this is true when , a proof of which can be found in \textcite[Remark 5.1.4]frenkel-ben-zvi-2004, and the general case follows from swapping two adjacent operators at a time.
This property implies that the expression Eq. 3.1.3.3 must lie in the intersection of the spaces for all , which is . Consequently, the element lies in the image of Eq. 3.1.3.2.
From here, verifying the axioms in § 3.1.2 is straightforward.
Para 3.1.4 (Remark)
Para 3.1.5 (Weak modules)
Similarly to weak vertex algebras introduced in § 2.1.7, we define a weak module over a -graded weak vertex algebra to be a -graded vector space , equipped with multiplication maps
preserving grading, where , satisfying the properties § 3.1.2 Items (i) to (ii).
For example, every weak vertex algebra is a weak module over itself.
Note that if is a vertex algebra, then a weak module is a usual module if and only if the image of lies in the subspace .
Para 3.1.6 (Twisted modules)
We introduce twisted modules for vertex algebras equipped with involutions, which are a special class of weak modules. They will play an important role in the theory of wall-crossing for orthosymplectic enumerative invariants. Note that these are different from the notion of twisted modules in, for example, \textcite[§5.6]frenkel-ben-zvi-2004.
Let be a -graded vertex algebra over . Suppose that is equipped with a twisted involution , where is the vertex algebra with the same underlying graded vector space as , with multiplication
| (3.1.6.1) |
We require that is an isomorphism of -graded vertex algebras, and .
Define a twisted module for as a weak module , such that the image of each lies in the subspace
where we invert for , and such that
| (3.1.6.2) |
for and .
We will construct examples of twisted modules in Section 3.2 below.
Para 3.1.7 (Residues)
Let be a -graded vertex algebra. Recall that the quotient , where is the translation operator in § 2.1.3, admits a -graded Lie algebra structure given by
| (3.1.7.1) |
with sign rules implemented for odd elements in the axioms of a Lie algebra. A usual module for gives rise to a module for , defined by .
Now, suppose that is equipped with an involution as in § 3.1.6, and let be a twisted module for . Then the Lie algebra admits an induced involution , where is the Lie algebra with the opposite Lie bracket . Moreover, the assignment
| (3.1.7.2) |
for and establishes as a twisted module for , in that we have
| (3.1.7.3) | ||||
| (3.1.7.4) |
for homogeneous elements and , where denotes the degree. The identity Eq. 3.1.7.4 can be seen by applying the residue theorem to , giving
| (3.1.7.5) |
where each of the four terms corresponds to a term in Eq. 3.1.7.4, and we use the fact that .
Note that this structure of a twisted module for a Lie algebra also appears in motivic Donaldson–Thomas theory for orthosymplectic structure groups, described in \textcite[§5.2.2]bu-osp-dt. Also, as noted there, when the coefficient ring contains , a twisted module for is equivalent to a usual module for the subalgebra of consisting of elements with , with the action .
\firamedium3.2 \firamediumModules for Joyce vertex algebras
Para 3.2.1 (The setting)
We assume given the data as in § 2.2.4, We also assume given the extra data , where
-
•
is a space, regarded as an object of of rank .
-
•
is a map, exhibiting as an -module in , or equivalently, in .
-
•
is a class, called the obstruction theory.
For each integer , let
| (3.2.1.1) |
denote the -fold module multiplication map. Define the normal bundle of by
| (3.2.1.2) |
where , where , are the projections to and . We further assume the following condition:
-
•
For any integer , we have
(3.2.1.3) where denotes the part of -weight .
Theorem 3.2.2
Let be as in § 3.2.1. Then the assignment
| (3.2.2.1) |
defines a weak module structure on for the weak vertex algebra in Theorem 2.3.3.
The product Eq. 3.2.2.1 has poles along for non-zero weights appearing in . In particular, if the class only has -weights with , so that is a vertex algebra, and if the class only has -weights with , then is a module for .
-
Proof.
The proof is very similar to that of Theorem 2.3.3, and an analogous computation verifies the axioms for a weak module. See also Theorem 4.2.3 below for a more general proof.
For the final statement on usual modules, let , and consider the class . It is enough to show that it only has -weights of the form for , where we set , and for , we set with at the -th position. But the assumption implies that for any with , writing and , we have , which implies the desired property. ∎
Para 3.2.3 (An involution-twisted version)
We now discuss a version that often arises from moduli spaces involving the orthogonal and symplectic groups, which will be important in studying enumerative invariants for such moduli spaces.
Let be as in § 3.2.1. We further assume given the following data:
-
•
An involution , such that in .
It should satisfy the following conditions:
-
•
preserves the monoid structure .
-
•
reverses the -action , in that it is equivariant with respect to the map .
-
•
preserves the module action , in that .
Again, these are conditions in , and do not require higher coherence.
Example 3.2.4
Our main source of examples of the situation of § 3.2.3 is orthosymplectic enumerative geometry, as in \textcitebu-osp-dt.
Let be a self-dual derived linear moduli stack over , that is a derived linear moduli stack as in Example 2.2.6, equipped with an involution , satisfying conditions similar to those in § 3.2.3, as in [bu-osp-dt, §2.2.5]. Typical examples include the following:
-
•
The derived moduli stack of representations of a self-dual quiver , possibly with potential, where being self-dual means that we are given a contravariant involution of , which induces an involution of the stack. See Section 6.3 below for details, and see also [bu-osp-dt, §6.1].
-
•
The derived moduli stack of objects in a quasi-abelian subcategory , where is a smooth projective -variety, such that is stable under the dual functor , where we choose a line bundle and an integer . See also [bu-osp-dt, §6.2] for this set-up.
In this case, the derived fixed locus of the involution is the moduli stack of self-dual objects, whose points correspond to pairs with and , such that . Such objects can often be interpreted as orthogonal or symplectic objects in a linear category, where the difference between being orthogonal and symplectic is encoded in the higher coherence data of the involution on as a -action. Again, see [bu-osp-dt] for details.
We have an action given by , where is equipped with the obvious self-dual structure.
We now set and to be the topological realizations, and with the involution on and the -action on given by the involution on and its action on . Let , be the classes of the tangent complexes , , where we note that is given by the -invariant part of the pullback of to .
Then, assuming that satisfies the conditions in § 2.2.4, all the extra conditions in §§ 3.2.1 and 3.2.3 will be automatically satisfied, where the weight condition Eq. 3.2.1.3 follows from the similar condition Eq. 2.2.4.2 for and the above description of .
Example 3.2.5
We also consider a derived version of Example 3.2.4, similar to Example 2.2.7.
Namely, let be a -linear dg-category as in Example 2.2.7, and suppose we are given a contravariant involution , inducing a -action on the derived moduli stack, . Examples of include the following:
-
•
The derived category of representations of a self-dual quiver, with the involution induced by the self-dual structure of the quiver.
-
•
The category of perfect complexes on a smooth projective -variety , with the involution as in Example 3.2.4.
Again, the derived fixed locus is the moduli stack of self-dual objects, and we have an action given by .
Theorem 3.2.6
Let be as in § 3.2.3. Assume the class only has -weights with , so that is a vertex algebra by Theorem 2.3.3, and the class only has -weights with .
Then the product Eq. 3.2.2.1 establishes as a twisted module for the vertex algebra .
-
Proof.
It follows from Theorem 3.2.2 that is a weak module for , and the relations Eqs. 3.1.6.1 to 3.1.6.2 follow from the constructions. It remains to verify the restriction on poles, which is equivalent to that the class only has -weights of the form for , with notations as in the proof of Theorem 3.2.2. But the assumption implies that for any with , we have , again with notations as in the proof of Theorem 3.2.2, and an elementary argument shows that this implies the desired property on -weights. ∎
Para 3.2.7 (Morphisms of modules)
Finally, we state a result analogous to § 2.3.5, which gives a construction of morphisms between modules for Joyce vertex algebras.
Theorem
Let and be as in § 3.2.1, be a map as in § 2.3.5, and a map compatible with the actions of and , such that the class
is the class of a vector bundle on , possibly of mixed rank. Then the map
is compatible with the weak vertex algebra module structures, in the sense that
for any and . In particular, if and , then is a homomorphism of weak modules.
The proof is very similar to that of § 2.3.5, and we omit it here. See Theorem 4.2.3 below for a proof of a more general result.
\LibertinusDisplay4 \LibertinusDisplayVertex induction
This section presents the main construction of this paper, the vertex induction, which unifies and simultaneously generalizes Theorems 2.3.3, 2.3.5, 3.2.2, 3.2.6, and 3.2.7.
We start by introducing a paracategory of vertex spaces, in which vertex algebras are algebra objects, which we prove in Theorem 4.1.7. This construction is somewhat similar to that in \textciteborcherds-1998, although not precisely the same. This will then provide a functorial approach to reformulating the constructions in Sections 2 to 3, and we will use this to describe the vertex induction in Theorem 4.2.3. We specialize to the case of algebraic stacks in Theorem 4.3.7, which is our motivating case.
\firamedium4.1 \firamediumVertex algebras as algebra objects
Para 4.1.1 (Paracategories)
A paracategory is, roughly speaking, a ‘category with partially defined composition’. The following definition is adapted from \textcite[§2]hermida-mateus-2003-paracategories. Compare the notion of a relaxed multilinear category in \textcite[§4]borcherds-1998.
A paracategory is a directed graph , together with a partially defined function for all , meaning a function from a subset of the -fold fibre product set to , satisfying the following axioms:
-
(i)
The map is fully defined. That is, identity morphisms always exist.
-
(ii)
The map is fully defined, and is the identity map.
-
(iii)
For , we have
(4.1.1.1) in the sense that whenever one side of the equality is defined, so is the other, and both sides are equal.
Any (small) category can be seen as a paracategory in the obvious way.
Note that the axioms imply that composition with an isomorphism is always defined, since for example, if is an isomorphism, and is such that , then , showing that is defined.
A functor of paracategories is a map of directed graphs , such that for any , if is defined, then is defined and is equal to . Note that such functors can always be composed.
A natural isomorphism of functors consists of an isomorphism for any , such that the naturality squares commute. Natural isomorphisms can always be composed.
Para 4.1.2 (Monoidal paracategories)
A (symmetric) monoidal structure on a paracategory consists of an object and a functor , equipped with natural isomorphisms satisfying the usual axioms.
One can also define (symmetric) monoidal functors between (symmetric) monoidal paracategories in the usual way.
For a (coloured) operad and a symmetric monoidal paracategory , an -algebra in is a symmetric monoidal functor , where is the underlying symmetric monoidal category of , as in \textcite[Remark 2.2.4.6]lurie-ha.
For example, an associative algebra in can be described as an object , with -ary multiplication maps for all , satisfying the usual axioms. A commutative algebra is an associative algebra such that the maps are equivariant under permutations.
Para 4.1.3 (Vertex spaces)
We now define a symmetric monoidal paracategory of vertex spaces, where vertex algebras are algebra objects.
Define a vertex space over a field as a triple , consisting of the following:
-
•
A finite-dimensional -vector space , whose dimension is called the rank of the vertex space.
-
•
A -graded -vector space , thought of as a space of fields on .
-
•
A -linear map , called the translation operator, where is a set of coordinates for , and , thought of as translating fields along . It should satisfy
(4.1.3.1) (4.1.3.2) In other words, is a representation of the formal group of the additive group of .
We often write for the vertex space, and write and for and .
Para 4.1.4 (Maps of vertex spaces)
A map of vertex spaces consists of a -linear map , and a -linear map
where , such that
| (4.1.4.1) | ||||
| (4.1.4.2) |
for and . Here, can be thought of as translating fields along the vector , then pulling them back to along .
The identity map is given by and .
Note that the relation Eq. 4.1.4.1 might tempt one think that is determined by via . However, this is not the case, as is often singular at , in which case is undefined.
Para 4.1.5 (Composition)
Maps of vertex spaces cannot always be composed. Given a diagram
| (4.1.5.1) |
of maps of vertex spaces, we say that is the composition of and , written , if , and we have
| (4.1.5.2) |
for and . Note that is unique if it exists, and always exists if or is the identity. Composition is associative and unital whenever all involved compositions are defined.
Moreover, this generalizes to define -fold compositions of maps of vertex spaces for any . This defines the structure of a paracategory.
We denote by the paracategory of vertex spaces over .
Para 4.1.6 (The monoidal structure)
Given vertex spaces and , we define a vertex space structure on by
| (4.1.6.1) | ||||
| (4.1.6.2) |
The tensor product defines a symmetric monoidal structure on , and has a unit given by the vertex space , with and . We include a sign rule when identifying with on the odd graded pieces.
Theorem 4.1.7
-
Proof.
These follow from the definitions. ∎
\firamedium4.2 \firamediumVertex induction
Para 4.2.1
We now introduce the vertex induction, which is a simultaneous generalization of the constructions of Joyce vertex algebras and modules in Sections 2 to 3.
As mentioned in the introduction, we expect this construction to be useful for enumerative geometry, especially for formulating wall-crossing formulae for enumerative invariants of non-linear moduli stacks.
Para 4.2.2 (The paracategory )
To help with stating our main construction in Theorem 4.2.3, we define an auxiliary paracategory as follows. Recall the category from § 2.2.2.
-
•
An object of is a quadruple , where , and is a class of weight .
-
•
A morphism in is a morphism in , such that the normal bundle
satisfies the following conditions:
-
(a)
. Equivalently, .
-
(b)
The part of weight is the class of a vector bundle on .
-
(a)
Composition in is not fully defined, since the condition Item (a) is not preserved by composition in general.
Theorem 4.2.3
-
Proof.
The proof is very similar to that of Theorem 2.3.3, and the key is in proving that preserves composition. Let be composable morphisms in . Then for any , we have
where the third line uses Lemma 2.3.4, and the last step uses the fact that has weight with respect to , and the fact that . ∎
This generalizes the main constructions in Sections 2 to 3, in the sense that the associativity and commutativity axioms of vertex algebras and modules are now encoded as the functoriality of the functor , in view of Theorem 4.1.7. See also § 4.3.8 below for more motivations of this construction from enumerative geometry.
\firamedium4.3 \firamediumVertex induction for stacks
Para 4.3.1
We now discuss the vertex induction applied to algebraic stacks over , which is the case that we expect to be useful for enumerative geometry.
Throughout, let be a derived algebraic stack over , locally finitely presented, such that its classical truncation is a classical algebraic -stack over , quasi-separated, with affine stabilizers and separated inertia.
Para 4.3.2
We use the formalism of the component lattice of an algebraic stack, following \textciteepsilon-i. The component lattice is a combinatorial structure attached to any algebraic stack, and is key to generalizing existing results in enumerative geometry for linear moduli stacks to non-linear moduli stacks. It provides the indexing set for decomposition-type theorems, or for terms in wall-crossing formulae, etc. See [epsilon-ii, epsilon-iii, bu-davison-ibanez-nunez-kinjo-padurariu] for these applications.
Para 4.3.3 (The component lattice)
For a finite rank free -module , let be the torus with coweight lattice , where is the group algebra of . Define the stack of -graded points of as the derived mapping stack
It is also a derived algebraic stack satisfying the conditions in § 4.3.1, and this construction is contravariant in . We also write .
The component lattice of is the functor
where is the category of finite rank free -modules, and denotes the set of connected components. This does not depend on the derived structure, in that we have .
A presheaf on is also called a formal lattice, so that is a formal lattice. Its underlying set is the set .
For example, if is a quotient stack, where is a linear algebraic group over and is a -representation, then the component lattice is the quotient formal lattice , where is the coweight lattice of the maximal torus , and is the Weyl group. In general, is usually glued from copies of for various , along their automorphisms and maps between them. See [epsilon-i] for details and more examples.
Para 4.3.4 (Special faces)
As in [epsilon-i], define a face of as a morphism of formal lattices for some , regarded as a formal lattice via the Yoneda embedding. Such faces naturally form a category . For a face , write
for the connected component corresponding to . For a morphism of faces , there is an induced morphism , giving a functor
| (4.3.4.1) |
where is the -category of derived stacks over .
A special face of is a face such that one cannot enlarge without changing . More precisely, is special if for any morphism of faces such that is an isomorphism, it admits a retraction , so that .
For example, if as in § 4.3.3, then the special faces of are precisely the intersections of hyperplanes in dual to weights of and roots of in . For such a face , the stack is the quotient of the fixed locus of the corresponding subtorus by the Levi subgroup .
Let be the full subcategory of special faces. There is a special closure functor
which is left adjoint to the inclusion. The functor Eq. 4.3.4.1 factors through this functor, and in particular, we have for any , so is a canonical replacement of without changing the stack .
By the finiteness theorem [epsilon-i, Theorem 6.2.3], if is quasi-compact and satisfies a very mild condition called quasi-compact graded points, then has only finitely many special faces.
Para 4.3.5 (Tangent weights)
For any face , write , where is the tangent complex of , and is the forgetful morphism. The canonical -action on induces a -grading of , and we have
| (4.3.5.1) |
where denotes the degree part with respect to the -grading.
Define the set of tangent weights of on by
| (4.3.5.2) |
We say that has finite tangent weights if this set is finite for all faces , or equivalently, for all special faces .
Para 4.3.6 (A functor to )
Theorem 4.3.7
-
Proof.
This follows from Theorem 4.2.3 and § 4.3.6. ∎
Para 4.3.8
We expect that the maps of vertex spaces associated to in Theorem 4.3.7, or the vertex induction maps, together with their residues along their poles in the variables, will be useful in studying the structure of enumerative invariants. In particular, we expect that the wall-crossing formulae for a conjectural version of Joyce’s homological invariants, discussed in § 1.1.3, should be expressed using these maps, and should have the same form as those for intrinsic Donaldson–Thomas invariants to appear in [epsilon-iii].
\LibertinusDisplay5 \LibertinusDisplayVariants
\firamedium5.1 \firamediumA real version
Para 5.1.1
We introduce a generalization of the vertex induction, where we allow the obstruction theory to be an oriented real -theory class instead of a complex -theory class. This will recover the constructions in Sections 2 to 4 if the oriented real -theory class comes from a complex -theory class.
We expect that this construction will be important in studying DT invariants for general -shifted symplectic stacks, especially in the non-linear cases, where the stack is not the moduli stack of objects in a linear category. The linear case is discussed in \textcite[§4.4]gross-joyce-tanaka-2022.
Para 5.1.2 (Real -theory)
For a space , define the real topological -theory of as the commutative ring
| (5.1.2.1) |
There is a forgetful map given by complexification.
Given a class , an orientation of is a trivialization of the -bundle on classified by the composition , with the second map induced by the projection on each component. Such an orientation exists if and only if the class is zero, in which case the orientations form an -torsor.
Para 5.1.3 (The square root Euler class)
For a space and an -bundle on , recall the square root Euler class
which is the Euler class of the underlying oriented real vector bundle, and is only non-zero if is even. It satisfies , where is the top Chern class of the associated complex vector bundle.
Para 5.1.4 (The equivariant square root Euler class)
Let , with connected. Let be a class such that the weight part is the class of an -bundle on . Suppose that is equipped with an orientation, so in particular, is also an -bundle.
We define an equivariant square root Euler class , which is a real analogue of the class in § 2.3.1.
Consider the hyperplane arrangement on given by hyperplanes dual to non-zero -weights appearing in . For an open chamber of this hyperplane arrangement, define
| (5.1.4.1) |
where is equipped with the orientation induced by writing , where the last term has the orientation given by , that is, factoring as , with the first map given by , and the second map given by .
One can verify that does not depend on the choice of , while the orientation of depends on . The class is only non-zero if has even rank, and we have the relations
| (5.1.4.2) | ||||
| (5.1.4.3) |
where denotes with the opposite orientation.
When is not connected, we define on each connected component of as above.
We have the relation
| (5.1.4.4) |
where is equipped with the induced orientation, which does not depend on the order of and if they have even rank. In particular, if , then , in which case we also write .
Para 5.1.5 (Orientation of the normal bundle)
For a map of spaces and classes and , equipped with orientations, the normal bundle is equipped with the induced orientation, defined by the property that the sum preserves orientation.
This depends on the order of the sum, in that if we used instead, the orientation would be changed by a factor of . The orientation is reversed whenever the orientation of either or is reversed.
Para 5.1.6 (The paracategory )
Recall the category from § 4.2.2. We define a similar paracategory , using real -theory, as follows:
-
•
An object is a quintuple , where , and is a class of weight , and is an orientation of .
-
•
A morphism is a morphism in , such that the normal bundle satisfies , and the part of weight is the class of a vector bundle on .
There is a forgetful functor , although this will not be used below.
Example 5.1.7 (Shifted symplectic stacks)
Let be an -shifted symplectic stack over , with . An important special case is when , such as when is a moduli stack of coherent sheaves on a Calabi–Yau fourfold.
In this case, the symplectic structure gives rise to an orthogonal structure on the complex , and hence a class , where is the topological realization. An orientation of , as defined in § 5.1.2, is equivalent to an isomorphism that squares to the isomorphism given by the shifted symplectic structure, which agrees with \textcite[Definition 2.12]borisov-joyce-2017.
Theorem 5.1.8
There is a functor of paracategories
where is equipped with the translation operator defined in § 2.3.2, and is equipped with the induced orientation as in § 5.1.5.
In particular, we have the following:
-
(i)
Let be as in § 2.2.4, but with , replaced by , instead, and assume that is equipped with an orientation. Then there is a -graded weak vertex algebra structure on , defined by
(5.1.8.1) where are homogeneous elements, each supported on a single connected component of , and denotes the rank of on the support of .
-
(ii)
Let be as in § 3.2.1, but with , replaced by , instead, and assume that , are equipped with orientations. Then there is a weak module structure on for the weak vertex algebra in Item (i), defined by
(5.1.8.2) where and are homogeneous elements supported on single components, is as in Item (i), and is the rank of on the support of .
-
(iii)
Let be as in Theorem 4.3.7, equipped with an -shifted symplectic structure for , and an orientation. Then there is a functor
(5.1.8.3)
-
Proof.
For the main statement, the proof of Theorem 4.2.3 can be adapted to this situation without much change, and we only need to check the real version of Lemma 2.3.4, that is, for and , with an orientation, such that is the class of a vector bundle, we have
(5.1.8.4) for any . But it is straightforward to adapt the proof of Lemma 2.3.4 to this situation.
The statement Item (i) follows from the main statement, where we precompose the vertex induction map with the map given by
with notations as in the statement of Item (i), with the sign inserted so that it is equivariant under (signed) permutations of the factors. The sign is needed because in , compared to homology, the parity of elements is reversed on components with odd virtual dimension.
The statement Item (ii) holds analogously.
For Item (iii), as in § 4.3.6, there is a functor
sending a face to the object , where we choose an arbitrary orientation of for each , which exists by Eq. 4.3.5.1, as the class of is orientable and in . ∎
Para 5.1.9
We can also apply Theorem 5.1.8 Items (i) to (ii) to shifted symplectic stacks as follows.
Suppose that is an oriented -shifted symplectic stack over , with , and that it is also equipped with the structure of a derived linear moduli stack, as in Example 2.2.6. Suppose that the weight condition Eq. 2.2.4.2 is satisfied, which is usually the case, as discussed in Example 2.2.6. Then Eq. 5.1.8.1 defines a vertex algebra structure on . If, moreover, is equipped with a self-dual structure as in Example 3.2.4, such that its -action preserves the shifted symplectic structure, then the fixed locus is also -shifted symplectic, and if it is also orientable, then Eq. 5.1.8.2 defines a twisted module structure on .
For example, when is an open substack of the moduli stack of perfect complexes on a Calabi–Yau fourfold , an orientation is given by \textcite[Theorem 13.7]joyce-upmeier-bordism, under a topological assumption on . In this case, self-dual structures may be obtained as in Example 3.2.4, but we do not know if orientations exist on .
Alternatively, we could also consider derived versions as in Examples 2.2.7 and 3.2.5. Namely, if and are as in Example 2.2.7, and if is equipped with a -Calabi–Yau structure, with , then has an -shifted symplectic structure by \textcite[Theorem 5.6]brav-dyckerhoff-2021. If it is orientable, then Eq. 5.1.8.1 defines a vertex algebra structure on , and if is self-dual as in Example 3.2.5, where the involution is compatible with the Calabi–Yau structure, then is also -shifted symplectic, and if it is also orientable, then Eq. 5.1.8.2 defines a twisted module structure on .
\firamedium5.2 \firamediumA K-theory version
Para 5.2.1
We introduce a -theory version of vertex induction, generalizing Liu’s [liu-k-wall-crossing] construction of a -theory version of Joyce vertex algebras. We expect this to be useful for generalizing the -theoretic invariants of \textciteliu-k-wall-crossing to more general quasi-smooth stacks.
An important difference between the -theory version and the homology version is that the former is multiplicative in its nature, rather than additive. This is reflected in the fact that we obtain multiplicative vertex algebras and vertex spaces, rather than the usual versions.
Para 5.2.2 (Multiplicative vertex algebras)
Define a multiplicative vertex algebra over to be the data , consisting of a -vector space and -linear multiplication maps
where we write , satisfying the following properties:
-
(i)
(Unit) For any , we have
(5.2.2.1) -
(ii)
(Commutativity) For any homogeneous elements , and any permutation , we have
(5.2.2.2) -
(iii)
(Associativity) For integers and elements , we have
(5.2.2.3)
Similarly, we define weak multiplicative vertex algebras as above, but with the codomain of enlarged to .
As usual, we abbreviate the product as .
Para 5.2.3 (Multiplicative modules)
We have a similar notion of a module over a multiplicative vertex algebra , defined as a -vector space equipped with -linear multiplication maps
where we write , with the following properties:
-
(i)
(Unit) We have .
-
(ii)
(Associativity) For integers and elements , we have
(5.2.3.1) (5.2.3.2)
Define a weak module over a weak multiplicative vertex algebra in the same way as above, but with the codomain of enlarged to .
Similarly to § 3.1.6, we also have a notion of twisted modules. Namely, if a multiplicative vertex algebra is equipped with a twisted involution, meaning an involution such that
| (5.2.3.3) |
for all , then we define a twisted module for as a weak module such that the image of lies in the subspace , and such that for all and . Here, inverting means we allow poles also at for .
As usual, we abbreviate as .
Para 5.2.4 (A version of -homology)
We now define the underlying vector space where our multiplicative vertex algebras and vertex spaces will live.
Let . Define
| (5.2.4.1) |
where is the ideal consisting of elements with for all .
In other words, is the continuous dual of with respect to the topology on given by the neighbourhood basis of consisting of .
This is a modification of the construction of \textcite[§2.2]liu-k-wall-crossing, and the purpose of imposing the vanishing condition is to ensure that the equivariant exterior power in § 5.2.7 is well-defined.
We have a cap product
| (5.2.4.2) |
defined by for all and .
For example, as in \textcite[Proposition 2.3.3]liu-k-wall-crossing, we have
| (5.2.4.3) |
where acts on itself by translation, and we identify , where is the universal line bundle.
Para 5.2.5 (The multiplicative translation operator)
Let . Let be a set of coordinates on , and write .
Following \textcite[§3.2.3]liu-k-wall-crossing, define the multiplicative translation operator
where is a class of weight , and we write . This is well-defined, since as in \textcite[Lemma 3.2.3]liu-k-wall-crossing, we have the alternative description
| (5.2.5.1) |
where , and is the element in Eq. 5.2.4.3.
Para 5.2.6 (Multiplicative vertex spaces)
Similarly to § 4.1.3, we also define multiplicative vertex spaces, which we use to describe multiplicative versions of the constructions in Section 4.
A multiplicative vertex space over a field is a triple , consisting of a finite-dimensional -vector space , a -vector space , and a -linear map , called the translation operator, where we set for , for a set of coordinates for , satisfying
| (5.2.6.1) | ||||
| (5.2.6.2) |
In other words, is a representation of the formal group .
Para 5.2.7 (The equivariant exterior power)
Let , and be a class such that is the class of a vector bundle.
Heuristically, the equivariant exterior power is defined by
| (5.2.7.1) |
where is a set of coordinates on , and we write , etc. However, the expression Eq. 5.2.7.1 does not make sense yet, since the sum can be infinite when is not the class of a vector bundle,
To fix this, we expand the sum in the variables as
| (5.2.7.2) |
where is a vector bundle by assumption, , denotes the completion of with respect to (see § 5.2.4), and
| (5.2.7.3) |
which satisfies . For example, if is a sum of line bundles, then .
We take Eq. 5.2.7.2 as the definition of from now on. We have the relation
| (5.2.7.4) |
whenever the right-hand side is defined.
Theorem 5.2.8
There is a functor of paracategories
where is equipped with the translation operator defined in § 5.2.5.
In particular, we have the following:
-
(i)
Let be as in § 2.2.4. Then the assignment
(5.2.8.1) defines a weak multiplicative vertex algebra structure on , with poles along the locus for non-zero weights appearing in . In particular, if the class only has -weights and then is a multiplicative vertex algebra.
-
(ii)
Let be as in § 3.2.1. Then the assignment
(5.2.8.2) defines a weak module structure on for the weak multiplicative vertex algebra , with poles along the locus for non-zero weights appearing in .
-
(iii)
Let be as in Theorem 4.3.7. Then there is a functor
(5.2.8.3)
-
Proof.
For the main statement, we need to verify that the construction preserves composition. Let be composable morphisms in . Then for any , we have
where the third line uses Lemma 5.2.9 below, and the last step uses the fact that has weight with respect to .
It is straightforward to check that the other statements follow from the main statement. ∎
Lemma 5.2.9
Let , and let such that is the class of a vector bundle. Then for any , we have
| (5.2.9.1) |
This is a -theory version of Lemma 2.3.4. See also \textcite[Lemma 3.2.8]liu-k-wall-crossing.
-
Proof.
As is multiplicative in , it is enough to prove this when is of pure weight . If , then both sides are equal to . If , then it is enough to show that
(5.2.9.2) for any , where is the operator that acts as on the weight component. Both sides of Eq. 5.2.9.2 are expansions of , which is a holomorphic function in the region , and we expand it at , where it has a pole along of order depending on . The two expansions are thus equal. ∎
Para 5.2.10 (An algebraic version)
The construction in Theorem 5.2.8 also has a version for algebraic -theory, which is closer to the original construction of \textciteliu-k-wall-crossing.
Namely, for a derived algebraic stack over as in § 4.3.1, equipped with an action , define the subgroup
of the Grothendieck group of perfect complexes on to be the preimage of under the topological realization map . Then, define as in § 5.2.4, with Chern characters valued in .
In particular, if admits a linear structure as in Example 2.2.6 or Example 2.2.7, then has the structure of a multiplicative vertex algebra, which is a version of Liu’s original construction. If, moreover, admits a self-dual structure as in Example 3.2.4 or Example 3.2.5, then has a twisted involution, and is a twisted module for this involution.
\LibertinusDisplay6 \LibertinusDisplayExamples
\firamedium6.1 \firamediumClassifying stacks
Para 6.1.1
In this section, we discuss the following examples of our construction:
-
(i)
A vertex algebra structure on the homology of , the classifying stack of perfect complexes over . This was due to \textcitejoyce-hall and discussed in \textcite[§2.6.29]latyntsev-thesis, and is related to the Heisenberg vertex algebra.
-
(ii)
Twisted module structures on the homology of and .
-
(iii)
Vertex induction for the homology of for reductive groups over .
Para 6.1.2 (For reductive groups)
We begin with the example Item (iii) above.
Let be a linearly reductive algebraic group over , with maximal torus . Let be the neutral component of the centre of . Consider the object , where acts on by translation. We identify
where is the -th standard generator upon an identification , and is the Weyl group. Dually, we have
where denotes taking coinvariants, and regarding , as (co)homology classes of , acts as via the cap product.
We set , with the adjoint action of on its Lie algebra .
In fact, we are now in the situation of Example 5.1.7, where admits a -shifted symplectic structure given by choosing a Weyl-invariant inner product on . This symplectic structure is orientable if and only if acts trivially on via the adjoint action of , which is always true if has an odd number of connected components.
Assuming such an orientation exists, for a Levi subgroup , we have
| (6.1.2.1) | ||||
| (6.1.2.2) |
where is the set of roots of such that for a cocharacter with , and the ‘’ sign depends on the choice of orientations on and and the choice of . For example, reversing will change the sign by .
We can then write down the vertex induction map
| (6.1.2.3) |
defined in Theorem 5.1.8 explicitly using Eq. 6.1.2.1, where is a set of coordinates on , the coweight lattice of the neutral component of the centre of , and the map has no poles in this case. Namely, it is given by
| (6.1.2.4) |
where is a polynomial in , and is a linear combination of the induced by the inclusion , and the ‘’ sign depends on the choice of orientations. Similarly, this can be done for the version in Theorem 4.3.7 using Eq. 6.1.2.2.
Para 6.1.3 (Perfect complexes)
Let be the classifying stack of perfect complexes over , and consider the space
as a special case of Example 2.2.7. Set , where is the universal class. The class agrees with the class of the tangent complex , where is the universal perfect complex on . We thus have , where denotes the component of rank .
The class lifts to a class using the -shifted symplectic structure on given by \textcite[§2.3]pantev-toen-vaquie-vezzosi-2013, as discussed in § 5.1.9. It admits an orientation given by a choice of identification on the component of rank .
We discuss two versions of Joyce vertex algebras in this case: The usual version from Section 2.3, and the real version from Section 5.1. We have
| (6.1.3.1) |
where are variables dual to the universal Chern characters, in that using the cap product, the -th Chern character acts as . We have
| (6.1.3.2) |
where is the pullback of the universal class of the -th factor, so that using the universal relation
| (6.1.3.3) |
where is the -th Chern class, and using a suitable choice of orientation, we have
| (6.1.3.4) | ||||
| (6.1.3.5) |
where run over the range , and corresponds to a connected component of , and is the -th Chern character pulled back along the -th projection .
From these, one can write down explicit formulae for the vertex algebra structures given by Theorems 2.3.3 and 5.1.8. For example, the latter version is defined on the space
whose vertex algebra structure is given by
| (6.1.3.6) |
where each is homogeneous and supported on a single component , and regarded as a polynomial in the variables , and
As in \textcite[§§2.6.25–29]latyntsev-thesis, this is a lattice vertex algebra on the lattice , and the subalgebra at is isomorphic to the Heisenberg vertex algebra.
Para 6.1.4 (Orthosymplectic complexes)
The space can be seen as a homotopy fixed locus of a -action on or given by complex conjugation. See \textcite[Corollary 7.6]dugger-2005 for the type B and D cases, and taking the -fold loop space gives the case of type C by Bott periodicity.
Define by
where and are the universal classes, and we use the operation .
The class is orientable in types B and C, which can be checked via the class, where we have for any , and . However, the same reasoning shows that it is not orientable in type D. This is also similar to the orientability criterion that we obtained in § 6.1.2.
Let act on via the map given by , with equipped with the obvious orthosymplectic structure. More precisely, we define it as the homotopy -fixed loci of the map , where acts on by complex conjugation followed by swapping the first and third factors.
We have
| (6.1.4.1) |
where again, is a formal symbol, and or is dual to the -th Chern character, in a sense similar to that of § 6.1.3.
In types B and C, we have
| (6.1.4.2) |
and the twisted module structure given by Theorem 5.1.8 can be explicitly written as
| (6.1.4.3) |
where and are homogeneous and supported on components and , and ‘’ means ‘’ in type B and ‘’ in type C, and ‘’ vice versa, and
and throughout. The change of variables at the end of Eq. 6.1.4.3 is the effect of pushing forward along .
There is also the usual version given by Theorem 3.2.6, which is defined for all the types B, C, and D, with a similar expression to Eq. 6.1.4.3.
Para 6.1.5 (Remark)
In § 6.1.4, we can also take to be the topological realization of classifying stacks of orthosymplectic perfect complexes. More precisely, define -actions on or by taking the dual complex. The data of a -action includes a choice of identification , which can be taken to be , which gives the distinction between types C and D. Define classifying stacks
of orthogonal (resp. symplectic, orthogonal) perfect complexes of odd (resp. even, even) rank, as derived fixed loci of these -actions. We have an equivalence given by , with homotopy inverse .
\firamedium6.2 \firamediumPrincipal bundles
Para 6.2.1
We describe the vertex induction for moduli stacks of principal -bundles or perfect complexes on a variety. We consider the following versions:
-
(i)
An algebraic version, involving the moduli stack of principal -bundles on a -variety.
-
(ii)
A topological version, using the topological mapping space from a -variety to .
-
(iii)
A version for moduli stacks of perfect complexes on a -variety, where we obtain the Joyce vertex algebra.
-
(iv)
A version for moduli stacks of orthosymplectic perfect complexes on a -variety, where we obtain twisted modules for the Joyce vertex algebra.
Para 6.2.2 (The algebraic version)
Let be a connected, smooth, projective variety over , and let be a linearly reductive algebraic group over , with Lie algebra . Let
| (6.2.2.1) |
be the derived mapping stack, or the moduli stack of -bundles on . Its tangent complex satisfies that at a -point corresponding to a -bundle , we have , where is the adjoint vector bundle of , with fibres isomorphic to .
By Theorem 4.3.7, there is a functor
In particular, for each Levi subgroup , meaning the centralizer of a cocharacter, there is a vertex induction map
| (6.2.2.2) |
where is a set of coordinates on , and we invert for non-zero elements that are images of roots of . The vertex induction map respects composition, in that for Levi subgroups , the induction for is the composition of the other two inductions in .
If, moreover, is Calabi–Yau of dimension , then admits a -shifted symplectic structure, and we are in the situation of § 5.1.9. For a Levi subgroup , if we are given orientations of and , then there is a vertex induction map
| (6.2.2.3) |
given by Theorem 5.1.8, with same notations as in Eq. 6.2.2.2.
Para 6.2.3 (The topological version)
Now let be a compact complex manifold, and let be a linearly reductive algebraic group over . Consider the topological mapping space
| (6.2.3.1) |
which is the space of topological -bundles on . Define the class
| (6.2.3.2) |
where is the class of the adjoint representation, is the evaluation map, is the projection, and is the Gysin map, defined in \textcite[§IV.5.27]karoubi-1978-k-theory for manifolds, and defined here by taking the mapping space from to the Gysin map .
Then for each Levi subgroup , there is a vertex induction map
| (6.2.3.3) |
similar to Eq. 6.2.2.2, given by Theorem 4.2.3. It respects composition in the same sense as in Eq. 6.2.2.2.
There is also a real version, similar to Eq. 6.2.2.3. Let be a compact spin manifold of dimension for some , such as a Calabi–Yau -fold, and let be as above. Then using the class , and the Gysin map for such spin manifolds from \textcite[§IV.5.27]karoubi-1978-k-theory, we obtain a class , where . If this class is orientable, then we obtain a real version of the vertex induction map
| (6.2.3.4) |
similar to Eq. 6.2.2.3.
Para 6.2.4 (Perfect complexes)
Let be a smooth proper -scheme, and let
be the derived moduli stack of perfect complexes on , as in \textcite[Definition 3.28]toen-vaquie-2007-moduli, which admits a perfect tangent complex . Then
-
•
Theorem 2.3.3 defines a vertex algebra structure on , originally due to \textcitejoyce-hall; see also \textcite[Theorem 4.4]gross-homology.
-
•
Theorem 5.2.8 defines a multiplicative vertex algebra structure on , essentially due to \textciteliu-k-wall-crossing.
-
•
If is Calabi–Yau of dimension , then is -shifted symplectic, and if we are given an orientation of , then Theorem 5.1.8 defines a vertex algebra structure on , also originally due to \textcitejoyce-hall.
Alternatively, we can consider a topological version of the above. Let now be a compact complex manifold, and let
be the topological mapping space. Set , where is classified by the evaluation map to , and is the projection. Again,
-
•
Theorem 2.3.3 defines a vertex algebra structure on .
-
•
Theorem 5.2.8 defines a multiplicative vertex algebra structure on .
-
•
If is Calabi–Yau of dimension , then lifts to a class in , and given an orientation, Theorem 5.1.8 defines a vertex algebra structure on .
By \textcite[Lemma 4.10 and Proposition 4.14]gross-homology, if is of class D in the sense of [gross-homology, Definition 4.8], which includes all curves and algebraic surfaces, then there is an equivalence identifying and , so the algebraic and topological versions agree.
Para 6.2.5 (Orthosymplectic complexes)
Let be one of the three stacks in § 6.1.5.
Let be a smooth proper -scheme, and let be as in § 6.2.4. Let
be the derived mapping stack, using one of the three options, which admits a perfect tangent complex . Let act on by setting , similarly to § 6.1.4. Then
-
•
Theorem 3.2.6 defines a twisted module structure on for the vertex algebra .
-
•
Theorem 5.2.8 defines a twisted module structure on for the multiplicative vertex algebra .
-
•
If is Calabi–Yau of dimension , then is -shifted symplectic, and if we are given orientations of and , then Theorem 5.1.8 defines a twisted module structure on for the vertex algebra .
Similarly, there are topological versions defined using topological mapping spaces to
and we avoid repeating the details here.
\firamedium6.3 \firamediumQuivers
Para 6.3.1 (Quivers)
We very briefly describe the Joyce vertex algebra for quivers and its variants, which already exist in the literature.
Let be a quiver, where , are finite sets of vertices and edges, and are the source and target maps.
Consider the moduli stack of representations of ,
| (6.3.1.1) |
where and . It is a smooth algebraic stack over , and has a perfect tangent complex . Its topological realization is given by .
We have a vertex algebra structure on given by Theorem 2.3.3, due to \textcitejoyce-hall, and a multiplicative vertex algebra structure on given by Theorem 5.2.8, due to \textciteliu-k-wall-crossing.
There is also a derived version, where we consider the derived moduli stack of complexes of representations of , as in \textcite[Definition 3.33]toen-vaquie-2007-moduli or \textcite[§4.3.2]latyntsev-thesis. Its topological realization is .
Again, we have Joyce’s vertex algebra structure on given by Theorem 2.3.3, and Liu’s multiplicative vertex algebra structure on given by Theorem 5.2.8.
By \textcite[§4.3.7 and Theorem 4.4.8]latyntsev-thesis, the vertex algebra is a lattice vertex algebra, and when is a Dynkin quiver of type A/D/E, it gives the simple quotient of the Kac–Moody vertex algebra of the same type at level .
Para 6.3.2 (Self-dual quivers)
We now discuss orthosymplectic twisted modules for Joyce vertex algebras for quivers described above.
Let be a quiver as above. A self-dual structure on is the following data:
-
(i)
A contravariant involution
where is the opposite quiver of , such that .
-
(ii)
Choices of signs
such that for all , and for all .
For details, see \textcite[§6.1]bu-osp-dt, \textciteyoung-2015-self-dual,young-2016-hall-module,young-2020-quiver, and \textcitederksen-weyman-2002.
As in [bu-osp-dt, §6.1.3], there is a moduli stack of self-dual representations of ,
| (6.3.2.1) |
where is the subset of dimension vectors such that for all , and is even if and . The vector space and the group are given by
| (6.3.2.2) | ||||
| (6.3.2.3) |
where is the set of vertices with , and the sets of vertices with and . Similarly, is the set of edges with , and the sets of edges with and . There is a natural action , given by , where is the dual representation of ; see [bu-osp-dt, §6.1.2].
We have a twisted module for the Joyce vertex algebra , given by Theorem 3.2.2, and a twisted module for the multiplicative vertex algebra , given by Theorem 5.2.8.
There is also a derived version, where we consider the derived stack , the fixed locus of the -action on given by the self-dual structure of . Its topological realization is given by
| (6.3.2.4) |
where and as in § 6.1.5.
We have a twisted module for the Joyce vertex algebra , given by Theorem 3.2.2, and a twisted module for the multiplicative vertex algebra , given by Theorem 5.2.8.
As discussed in § 6.1.5, we expect to have and , but in any case, the above construction still works if we replace by the product Eq. 6.3.2.4 with , replaced by and .
Chenjing Bu [email protected]
Mathematical Institute, University of Oxford, Oxford OX2 6GG, United Kingdom.