Additive invariants of orbifolds (Q1637318)

Recall that a \textit{differential graded (dg) category} \(\mathcal A\) is a category enriched over complexes of \(k\)-vector spaces. Also recall that a complex of \(\mathcal O_X\)-modules on a scheme \(X\) is called \textit{perfect} if it is quasi-isomorphic to a bounded complex of locally free \(\mathcal O_X\)-modules. Let \(\mathsf{dgcat}(k)\) denote the category of essentially small \(k\)-linear dg categories. Any dg \(k\)-algebra gives rise to a dg category with a single object. Other examples are given by schemes as the category of perfect complexes \(\mathsf{perf}(X)\) on a suitable algebraic stack \(X\) admits a canonical dg enhancement \(\mathsf{perf}_{\text{dg}}(X).\) The tensor product \(-\otimes_X-\) then makes \(\mathsf{perf}_{\text{dg}}(X)\) to a commutative monoid in the category obtained from \(\mathsf{dgcat}(k)\) by inverting Morita equivalences. An \textit{additive invariant} of dg categories is a functor \(E:\mathsf{dgcat}(k)\rightarrow\mathsf{D}\) with values in an idempotent complete additive category which inverts Morita equivalences and sends semiorthogonal decompositions to direct sums: A Morita equivalence is a dg functor \(F:\mathcal A\rightarrow\mathcal B\) which induces an equivalence on derived categories \(\mathcal D(\mathcal B)\overset{\cong}{\rightarrow}\mathcal D(\mathcal A)\). A semiorthogonal decomposition \(\langle\mathcal T_1,\mathcal T_2\rangle=\mathcal T\) of a triangulated category \(\mathcal T\) consists of full triangulated subcategories \(\mathcal T_1,\mathcal T_2\subset\mathcal T\) such that the inclusions admits left and right adjoints, the category \(\mathcal T\) is generated by the objects in \(\mathcal T_1,\mathcal T_2\), and \(\text{Hom}_{\mathcal T}(\mathcal T_1,\mathcal T_2)=0.\) Examples of additive invariants include several variants of algebraic \(K\)-theory, of cyclic homology, and of topological Hochschild homology. The authors use \(E(X)\) as short for \(E(\text{perf}_{\text{dg}}(X))\) and remark that when \(E\) is lax (symmetric) monoidal, \(E(X)\) is a (commutative) monoid in \(D\). Let \(G\) be a finite group acting on a smooth separated \(k\)-scheme \(X\). Let \([X/G]\) be the associated (global) orbifold, and let \(X/\!\!/G\) be the geometric quotient. The present article is divided into three parts: i) Decomposition of orbifolds: Formulas for \(E([X/G])\) in terms of fixed-point data \(\{E(X^g)\}_{g\in G}\) are presented. In the particular case where \(E\) is algebraic \(K\)-theory or cyclic homology, these formulas reduce to previous results of Vistoli and Baranovsky respectively. ii) Smooth quotients: The article proves that if \(X/\!\!/G\) is smooth, then \(E(X/\!\!/G)\simeq E(X)^G.\) In the particular case where \(E\) is Hochschild homology, this reduces to a previous result of the second author together with \textit{A. Polishchuk} [``Semiorthogonal decompositions of the categories of equivariant coherent sheaves for some reflection groups'', Preprint, \url{arXiv:1503.04160}]. iii) Equivariant Azumaya algebras: The formulas from the above parts are extended to the case when \(X\) is equipped with a \(G\)-equivariant sheaf of Azumaya algebras. Again, in the particular case when \(E\) is Hochschild homology, the formulas reduce to results of \textit{V. Baranovsky} and \textit{T. Petrov} [Adv. Math. 209, No. 2, 547--560 (2007; Zbl 1113.14033)], and \textit{D. Arinkin} et al. [``Formality of derived intersections and the orbifold HKR isomorphism'', Preprint, \url{arXiv:1412.5233}]. The formulas and techniques developed in the article are applied for the proof of Grothendieck's standard conjectures of types \(C^+\) and \(D\), as well as Voevodsky's smash-nilpotence conjecture in the case of low-dimensional orbifolds. The main results are stated by the following: In the part with decomposition of orbifolds, the field \(k\) is of characteristic \(p\geq 0\), \(G\) is a finite group of order \(n\), \(\phi\) is the set of all cyclic subgroups of \(G\), and \(\phi/\sim\) is a fixed set of representatives of the conjugacy classes in \(\phi\). Given \(\sigma\in\phi\), \(N(\sigma)\) denotes its normalizer. It is assumed that \(1/n\in k\), but the \(n^{\text{th}}\) roots of unity are not necessarily in \(k\). Let \(X\) be a smooth, not necessarily quasi-projective, separated \(k\)-scheme equipped with a \(G\)-action. Let \(R(G)\) be the representation ring of \(G\), then for a \(G\)-representation \(V\), \([V]\mapsto V\otimes_k-\) defines an action of \(R(G)\) on \(E([X/G])\) for every additive invariant \(E\). The first main result is stated verbatim: Theorem 1.1. For every additive invariant \(E:\mathsf{dgcat}(k)\rightarrow D,\) with values in a \(\mathbb Z[1/n]\)-linear category, we have an isomorphism \[ E([X/G])\simeq\underset{\sigma\in\phi/\sim}{\bigoplus}\tilde E([X^\sigma/\sigma])^{N(\sigma)} \] induced by pullback with respect to the morphisms \([X^\sigma/\sigma]\rightarrow[X/G].\) Moreover, if \(E\) is lax monoidal, then this is an isomorphism of monoids. The authors remark that the isomorphism in the theorem can be rewritten as \(E([X/G])\simeq\underset{\sigma\in\phi/\sim}{\bigoplus}\tilde E(X^\sigma\times\text{Spec}(k[\sigma]))^{N(\sigma)},\) which gives the intuitive meaning that every additive invariant of orbifolds can be computed using solely ordinary schemes. However, when \(E\) is lax monoidal, this isomorphism is no longer an isomorphism of monoids. For a commutative ring \(R\) and an \(R\)-linear idempotent complete additive category \(D\), the canonical action of the category of finitely generated projective \(R\)-modules on \(D\) is denoted \(-\otimes_R-.\) The interpretation of this action gives several corollaries and examples proving the advantage of \(G\)-invariants over \(G\)-coinvariants: It preserves monoid structures. It is remarked that the arguments of Vistoli cannot be used in this situation as they depend on the dévissage property of \(G\)-theory which does not hold for many additive invariants, in particular Hochschild homology. Also, the relation to the McKay correspondence is remarked. In the part considering Smooth quotients, let \(\pi:X\rightarrow X/\!\!/G\) be the quotient morphism. The second main result states that if \(E:\mathsf{dgcat}(k)\rightarrow D\) is an additive invariant with values in a \(\mathbb Z[1/n]\)-linear category, if \(X/\!\!/G\) is \(k\)-smooth, then the induced morphism \(\pi^\ast:E(X/\!\!/G)\rightarrow E(X)^G\) is invertible. In the final case of equivariant Azumaya algebras, let \(\mathcal F\) be a flat quasicoherent sheaf of algebras over \([X/G]\) and let \(\mathsf{perf}_{\text{dg}}([X/G];\mathcal F)\) be the canonical enhancement of the category of \(G\)-equivariant perfect \(\mathcal F\)-modules \(\mathsf{perf}([X/G];\mathcal F).\) Let \(E([X/G];\mathcal F)\) be short for \(E(\mathsf{perf}_{\text{dg}}([X/G];\mathcal F))\) when an additive invariant \(E\) is applied. When a cyclic sungroup \(\sigma\in\phi\) is given, \(\mathcal F_\sigma\) denotes the pullback of \(\mathcal F\) along the morphism \([X^\sigma/\sigma]\rightarrow[X/G].\) Then the third main result is stated as follows (verbatim): Theorem 1.25. For every additive invariant \(E:\mathsf{dgcat}(k)\rightarrow D\), with values in a \(\mathbb Z[1/n]\)-linear category, we have an isomorphism \[ E([X/G];\mathcal F)\simeq\underset{\sigma\in\phi/\sim}{\bigoplus}\tilde E([X^\sigma/\sigma];\mathcal F_\sigma)^{N(\sigma)} \] induced by pullback with respect to the morphisms \([X^\sigma/\sigma]\rightarrow[X/G].\) As a corollary to this result, it is proved that when \(\mathcal F\) is a sheaf of Azumaya algebras over \([X/G]\), the structural morphism \(Y_\sigma\rightarrow X^\sigma\) can be explicitly described as a Galois cover. Also, if \(k\) contains the \(n^{\text{th}}\) roots of unity and \(D\) is \(\mathbb Z[1/nr]\)-linear, then the isomorphism in the theorem reduces to \[ E([X/G];\mathcal F)\simeq\underset{\sigma\in\phi/\sim}{\bigoplus}(e_\sigma E(Y_\sigma))^{N(\sigma)} \] and even equivariantly if in addition \(D\) is \(l\)-linear for a field \(l\) which contains both the \(n^{\text{th}}\) roots of unity and \(1/nr\in l\). The article is rather voluminous, covering a lot of abstract theory. It is rather explicit, though it depends on some knowledge of invariant theory, including orbifolds. It contains a lot of new results which are true generalizations and unify known results about known additive invariants.