Get 20M+ Full-Text Papers For Less Than $1.50/day. Subscribe now for You or Your Team.

Learn More →

Equivariant Hodge theory and noncommutative geometry

Equivariant Hodge theory and noncommutative geometry DANIEL HALPERN-LEISTNER AND DANIEL POMERLEANO Abstract. We develop a version of Hodge theory for a large class of smooth formally proper quotient stacks X/G analogous to Hodge theory for smooth projective schemes. We show that the noncommutative Hodge-de Rham sequence for the category of equivariant coherent sheaves degenerates. This spectral sequence converges to the periodic cyclic homology, which we canonically identify with the topological equivariant K-theory of X with respect to a maximal compact subgroup of G, equipping the latter with a canonical pure Hodge structure. We also establish Hodge-de Rham degeneration for categories of matrix factorizations for a large class of equivariant Landau-Ginzburg models. Contents 1. The noncommutative motive of a quotient stack 7 1.1. Recollections on KN-stratifications 7 1.2. Motivic decompositions via KN stratifications 10 1.3. Λ-modules and the noncommutative Hodge-de Rham sequence 12 1.4. The degeneration property for quotient stacks 14 2. Hodge structures on equivariant K-theory 15 2.1. Equivariant K-theory: Atiyah-Segal versus Blanc 16 2.2. The case of smooth Deligne-Mumford stacks 23 2.3. Equivariant K-theory and periodic cyclic homology 26 2.4. Hodge structure on equivariant K-theory 27 3. Hodge-de Rham degeneration for singularity categories 29 3.1. Preliminaries on categories of singularities on stacks 29 3.2. Motivic decompositions and degeneration for MF 33 3.3. Graded Landau-Ginzburg models 37 4. Computations of Hochschild invariants 41 4.1. Generalities on Hochschild invariants 41 4.2. An HKR theorem and quotients of affine varieties 43 References 45 If X is a smooth projective variety over C, then the cohomology groups H (X;C) can be equipped with a pure Hodge structure of weight n. The theory of Hodge structures then allows one to “lin- earize” many important problems in algebraic geometry. Our goal is to develop such a linearization for the equivariant algebraic geometry of a locally closed algebraic submanifold X ⊂ P which is equivariant with respect to an action of a compact Lie group, M. Note that the complexification G of M, a reductive algebraic group, acts on X as well, and it is natural to ask for a Hodge theory associated intrinsically to the algebraic stack X := X/G. One such linearization follows from the results of [D1], which establishes a canonical mixed Hodge structure on the cohomology of any smooth simplicial scheme and in particular on the equivariant cohomology, H (X), which is the cohomology of the simplicial nerve of the action groupoid of G arXiv:1507.01924v3 [math.AG] 5 Feb 2022 on X. Building on these ideas, one can even associate a motive to the stack X/G as a colimit of motives of schemes as in [MV, Section 4.2]. The present paper lays the groundwork for an alternative approach to equivariant Hodge the- ∗ an ory based on equivariant topological K-theory of the underlying analytic variety K (X ). The equivariant K-theory is a module over the representation ring Rep(G), and the Atiyah-Segal com- pletion theorem canonically identifies its completion at the augmentation ideal of Rep(G) with the Z/2Z-graded equivariant cohomology, even/odd even/odd an ∧ an K (X ) H (X ). M G ∗ an If one completes K (X ) with respect to the evaluation ideal at other g ∈ M, then one recovers the Z/2Z-graded equivariant cohomology of the fixed locus of g with respect to the centralizer of g ∗ an [FHT1, Theorem 3.9]. Thus K (X ) encodes all of these equivariant cohomology groups, as well as the data of how to “spread” them out into a single finitely generated Rep(G)-module. In this ∗ an ∗ sense K (X ) is a much richer invariant than H (X). M G The main challenge is that unlike cohomology, equivariant K-theory is not simply the K-theory of the simplicial scheme arising from the action of G on X, so Deligne’s approach to equivariant Hodge theory does not generalize to K-theory. Instead, our Hodge structures originate in noncommutative algebraic geometry, which views dg-categories as “noncommutative spaces.” We ultimately show that in many cases one can use the dg-enhanced derived category of G-equivariant perfect complexes ∗ an of coherent sheaves on X, Perf(X/G), to construct a pure Hodge structure on K (X ). Noncommutative Hodge-de-Rham degeneration. If A is a dg-category over a field k, the Hochschild chain complex, C (A) plays the role of the Hodge cohomology in noncommutative per algebraic geometry. The periodic cyclic complex C (A), which is a dg-module over k((u)) where u has homological degree −2, behaves like noncommutative de Rham cohomology. There is a per canonical Hodge filtration of the complex C (A) whose associated graded is C (A)⊗k((u)), which • • per leads to a noncommutative Hodge-de Rham spectral sequence converging to H (C (A)) whose first ∗ • per page is H (C (A)⊗k((u))). The Hodge filtration in our theory will be the filtration on H (C (A)) ∗ • ∗ • arising from the degeneration of this spectral sequence. Geometrically, our main tool for establishing degeneration for the category Perf(X/G) will be certain stratifications of X into G-stable locally closed subvarieties. It is a classical observation in geometric invariant theory [H2, N1] that projective varieties with a linearized G-action inherit a canonical stratification. Kirwan used this stratification to deduce many beautiful results concerning the equivariant topology of projective varieties with G-action, including a computation of the Betti K5]. numbers and Hodge numbers for a variety obtained as a GIT quotient [ In this paper, we will consider a certain abstraction of this canonical stratification. These are the (semi-)complete Kirwan-Ness(KN)-stratifications of a G-variety, introduced in [T2] and recalled in Definition 1.1. The chief benefit of this more abstract definition is that it applies in many cases when the ambient variety X is merely quasi-projective. The main classes of examples of G-varieties which admit semi-complete KN-stratifications to keep in mind are: (i) any G-variety X which is projective over an affine G-variety; and (ii) any G-variety X such that X/G admits a good quotient that is projective-over-affine. In the case (i), the KN stratification is complete if and only if dim Γ(X,O ) < ∞. In the case (ii) the stratification is trivial, and it is complete if and only if the good quotient of X/G is projective. Recall that X/G admits a good quotient if there is an algebraic space Y and a G-invariant map π : X → Y such that π : QCoh(X/G) → QCoh(Y ) is exact and (π O ) ≃ O . ∗ ∗ X Y 2 Theorem A (Corollary 1.23). If G is a reductive group and X is a smooth G-quasi-projective vari- ety which admits a complete KN-stratification, then the noncommutative Hodge-de Rham sequence for Perf(X/G) degenerates on the first page. Remark. This builds on [T2], which shows that a version of the Hodge-de Rham spectral sequence ∗ ∗ for H (X) degenerates for such G-schemes and that the (a priori mixed) Hodge structure on H (X) G G is pure in this case. Note that in these examples, the scheme X is not proper, and neither is the quotient stack X/G, so degeneration of the Hodge-de Rham sequence is somewhat unexpected. Likewise from the noncommutative perspective, D. Kaledin’s recent resolution [K1] of a well-known conjecture of M. Kontsevich and Y. Soibelman [KS] shows that the noncommutative Hodge-de Rham sequence degenerates for dg-categories which are smooth and proper. However, the categories Perf(X/G) are typically not smooth even when X is smooth, and they are typically not generated by a single compact generator, so degeneration is again somewhat unexpected. The key observation in establishing the degeneration property for Perf(X/G) is that the forma- tion of the Hochschild complex takes semiorthogonal decompositions of dg-categories to direct sums, and its formation commutes with filtered colimits. Thus if A is a retract of a dg-category which can be built from the derived category of smooth and proper DM stacks via an infinite semiorthogonal decomposition, then the degeneration property holds for A. Example. One simple example is the quotient stack A /G , where G acts with positive weights. m m Then the objects O n{w} ∈ Perf(A /G ), which denote the twist of the structure sheaf by a A m character of G , form an infinite full exceptional collection. Therefore the Hochschild complex of Perf(A /G ) is quasi-isomorphic to a countable direct sum of copies of C (Perf(Spec(k))), and m • the degeneration property follows. We can formulate this most cleanly in terms of G. Tabuada’s universal additive invariant U : dgCat → M [T1,BGT]. Here M is the ∞-category which is the localization of the ∞-category k k of small dg-categories which formally splits all semiorthogonal decompositions into direct sums, and U is the localization map. The following is the main technical result of the paper, and we believe it is of independent interest. Theorem B (See Theorem 1.8). Let X/G be a smooth quotient stack over a field k of characteristic 0 that admits a complete KN stratification. Then there is a smooth projective variety Y such that b ⊕N U (X/G) is a direct summand of U (D (Y )) in M . k k k Connections with (classical) equivariant topology and purity. If G is the complexification of a compact Lie group M as above, we show that one can recover the equivariant topological an + K-theory of the underlying complex analytic space K (X ), as defined in [AS , S4], from the dg-category Perf(X/G). The first ingredient is the recent construction by A. Blanc of a topological K-theory spectrum top K (A) for any dg-category A over C [B3]. Blanc constructs a Chern character natural transforma- top tion ch : K (A) → HP(A), shows that ch⊗C is an equivalence for Perf of a finite type C-scheme, and conjectures this property for any smooth and proper dg-category A. We show that ch⊗C is an isomorphism for all categories of the form Perf(Y), where Y is a smooth DM stack or a smooth quotient stack admitting a semi-complete KN stratification. In fact, we expect that this “lattice conjecture” should hold for a much larger class of dg-categories, such as the categories D (X) for any finite type C-stack and Perf(X/G) for any quotient stack. Following some ideas of Thomason top an T5], we next construct a natural “topologization” map ρ : K (Perf(X/G)) → K (X ) in [ G,X M for any smooth G-quasiprojective scheme X and show: Theorem C (See Theorem 2.10 and Theorem 2.17). For any smooth quasi-projective G-scheme X which admits a semi-complete KN stratification, the topologization map and the Chern character 3 2 provide equivalences G,X ch per ∗ an top oo // K (X )⊗ C π K (Perf(X/G))⊗ C H C (Perf(X/G)) ∗ ∗ • Remark. In fact, Theorem 2.10 shows a bit more. ρ is an equivalence for any smooth G-quasi- G,X projective scheme. For an arbitrary G-quasi-projective scheme X we construct an equivalence c,∨ top b an of spectra ρ : K (D Coh(X/G)) → K (X ), where the latter denotes the M-equivariant G,X an Spanier-Whitehead dual of the spectrum K (X ), sometimes referred to as the equivariant Borel- Moore K-homology of X. ρ is compatible up to homotopy with natural pullback and pushfor- G,X ward maps (to be explained below). This result is of independent interest, and it allows one to “decategorify” theorems regarding equivariant derived categories in a precise way. n an Note that the groups K (X ) are modules over Rep(M), the representation ring of M. We say that a Rep(M)-linear Hodge structure of weight n is a finite Rep(M)-module E along with a finite filtration of the finite Rep(M) -module E ⊗ C inducing a Hodge structure of weight n on the un- per n an derlying abelian group E. Using the previous identification K (X )⊗C ≃ H C (Perf(X/G)), −n • we will show Theorem D (See Theorem 2.20). For any smooth M-quasiprojective scheme admitting a complete n an KN stratification, the noncommutative Hodge-de Rham sequence for K (X ) ⊗ C degenerates n an on the first page, equipping K (X ) with a pure Rep(M)-linear Hodge structure of weight n, functorial in X. There is a canonical isomorphism p n an n−2p der gr K (X ) ≃ H (RΓ(I ,O der)). X I der In this theorem, I denotes the derived inertia stack, sometimes referred to as the “derived loop stack.” As we will see in Lemma 4.2 below, we can express this more concretely as der L G RΓ(I ,O der) ≃ RΓ(G× X × X,O ⊗ O ) , X I Δ −1 where G acts on G × X × X by g · (h, x, y) = (ghg , gx, gy) and the two G-equivariant closed subschemes of G× X × X are defined as Γ = {(g, x, gx)} and Δ = {(g, x, x)} respectively. Example. Along the way, show that the lattice conjecture holds for an arbitrary smooth DM stack, and explicitly compute the Hochschild invariants of Perf(X). For a smooth and proper DM stack, we construct an isomorphism of Hodge structures top 2k−n cl π K (Perf(X))⊗ Q ≃ H (I ;Qhki). Betti X It should be noted that the motivic decompositions of Theorem B play a key role in the proof of Theorem D, but these decompositions do not respect the Rep(M)-linear nature of the Hodge n an structure on K (X ). In Section 4 we spend some time discussing more explicit models for the Hochschild homology and periodic cyclic homology for quotient stacks. For example, we show that when X is smooth and affine, there is an explicit bar-type complex computing the Hochschild homology of Perf(X/G). As an application of Theorem D, we prove an HKR type theorem for the completion of this bar complex at various points of Spec(Rep(G)) when X/G is formally proper. A corollary of this theorem is a description of the completed Hochschild homology modules equipped with the Connes operator in terms of differential forms equipped with the de Rham differential. We will see that these homology level equivalences are induced by suitable chain maps. 4 Extensions to categories of singularities. Another major source of Hodge structures in alge- braic geometry comes from singularity theory. For instance in [S1], Kyoji Saito constructs analogues of Hodge theoretic structures on the universal unfolding of an isolated singularity. More precisely, he describes analogues of the Gauss-Manin connection and period mappings as well as canoni- cal coordinates on the base space of the universal unfolding. Motivated in part by Saito’s work, Katzarkov, Kontsevich, and Pantev [KKP1, KKP2] have proposed a vast generalization of Hodge theory which they call noncommutative (nc) Hodge structures. As the name suggests, they envi- sion that nc Hodge structures should arise naturally from smooth and proper dg-categories(“nc spaces”). Let (X, W ) be a Landau-Ginzburg (LG) model, that is a pair (X, W ) consisting of a smooth quasi- projective variety X and a regular function W : X → A . To any LG model, one may associate the category of matrix factorizations MF(X, W ), which is a 2-periodic (meaning k((β))-linear where β has homological degree −2) dg-category. Applying the theory of nc Hodge structures to these nc spaces is expected to yield a vast generalization of Saito’s theory to pairs (X, W ) with proper critical locus, Crit(W ). For any LG model (X, W ), there is a “dW -twisted” Hodge de-Rham spectral sequence which • • relates the hypercohomologies of the complexes (Ω , dW∧) and (Ω , d+dW∧)(see [KKP1, Section 3.2] for details). Similarly to the classical case, this spectral sequence is known to degenerate when W : X → A is proper by work of Ogus-Vologodsky [OV]. This degeneration result plays a central role in the noncommutative Hodge theory of LG pairs—for example, a version of this result has been used to establish the smoothness of versal deformation spaces of (compactified) LG models (generalizing the universal unfolding space of a singularity)[KKP2]. Efimov and Preygel [P,E] have independently identified the dW -twisted Hodge de-Rham spectral sequence with the k((β))-linear noncommutative Hodge-de Rham spectral sequence for the category MF(X, W )(for closely related results, see also [D2, S3, CT, LP, S5]). It follows that the result of Ogus-Vologodsky can be recast as establishing the degeneration of this spectral sequence of noncommutative origin. We prove the following generalization of this degeneration result, which suggests that in the equivariant context nc Hodge theory should extend to certain dg-categories which are not smooth. We let Crit (W ) denote the critical locus with critical value 0. Theorem E. (Proposition 3.17) If X is a smooth G-quasi-projective scheme which admits a semi- complete KN stratification, and W : X → A is a G-invariant function such that Perf(Crit (W )/G) is a proper dg-category, then the k((β))-linear noncommutative Hodge-de Rham sequence for MF(X/G, W ) degenerates on the first page. Note that by Lemma 1.5, the condition on Crit (W ) in the theorem is equivalent to the induced KN-stratification on Crit (W ) being complete. In the case where X is projective over an affine G-variety, the condition is equivalent to dim Γ(Crit (W ),O ) < ∞, and if X/G admits a Crit (W ) good quotient, the condition is equivalent to the condition that Crit (W )/G admits a projective good quotient. We prove Theorem E by establishing an analog of Theorem B for the k((β))-linear category MF(X/G, W ) in Theorem 3.10. The proof is somewhat more subtle than the case of Perf(X/G), and its formulation is a little more complicated, because at the time of this writing we are not aware of a construction of k((β))-linear additive noncommutative motives. Along the way, we also Noncommutative Hodge structures are also expected to exist in other contexts, notably on the quantum coho- mology of a compact symplectic manifold. 5 establish the degeneration property for MF(X, W ) in the case that X is a smooth quasi-projective DM stack and Crit (W ) is proper (see Section 3). Further questions. The notion of properness in equivariant geometry. Our result on noncommutative Hodge-de Rham degeneration adds to the list of ways in which certain equivariant geometries behave as if they are proper despite not being proper in the sense of algebraic stacks. The intrinsic characterization of which smooth algebraic stacks behave as if they are proper from the perspective of Hodge theory, such as quotient stacks with a complete KN stratification, and which do not, such as BG or BU for a unipotent group U (see Example 1.24), is still somewhat fuzzy. The paper [HLP] studies these properness phenomena systematically by introducing the class of formally proper stacks, with the primary application being the algebraicity of the mapping stack out of a formally proper stack. The examples and counterexamples above for stacks exhibiting noncommutative Hodge-de Rham degeneration are also important examples and counterexamples for stacks which are formally proper in the sense of [HLP]. This raises the natural question. Question 0.1. Do there exist examples of perfect, smooth, and formally proper k stacks X for which the Hodge-de Rham sequence associated to Perf(X) does not degenerate? Hodge structures on equivariant K-theory. We believe that our main theorem for Hodge structures n an on K (X ) raises many questions for further inquiry into the role of Hodge theory in equivariant algebraic geometry. For example, it is plausible that the results above could be extended to con- struct mixed Hodge structures on some version of K-theory for arbitrary finite type stacks. In a different direction, one of the central notions in Hodge theory is that of a variation of Hodge struc- ture. For simplicity, let S be an affine scheme and suppose further that π : X/G → S is a smooth equivariant family over S such that all of the fibers X /G admit complete KN stratifications. Most of the techniques that we have developed work in families, which allows one to establish the exis- per tence of suitable Hodge filtrations on the quasi-coherent sheaf H C (Perf(X/G)). We therefore believe it is quite likely that one can develop a theory of equivariant period maps. Finally, Theorem E suggests that it may be possible to develop a version of noncommutative Hodge theory which applies in the equivariant context. Noncommutative equivariant geometry. Although we make use of noncommutative algebraic geom- etry, all of the differential graded categories in this paper are of commutative origin. It is interesting to try to formulate in noncommutative terms a criterion for the Hodge-de Rham spectral sequence to degenerate. Theorem D suggests the following concrete question: Let A be a proper dg-category which is a module over Perf(BG). Suppose that A⊗ k = Perf(R), where R is a dg- algebra Perf(BG) which is homotopically finitely presented, homologically bounded and such that H (R) is a finitely generated module over HH (R). Question 0.2. Does the Hodge-de Rham spectral sequence always degenerate for such A? Context. Throughout this work, unless explicitly stated otherwise, we work over a fixed subfield ∗ ∗ k ⊂ C. All of our functors are understood to be derived, so we write i for Ri , i for Li , Hom ∗ ∗ for R Hom, etc. We will work with stacks over the ´etale site of k-schemes. By convention, unless otherwise indicated the term quotient stack will denote a quotient of a quasi-projective k-scheme by a linearizable action of an algebraic k-group G, and we denote it X/G. As mentioned above, after the first draft of this paper was circulated, D. Kaledin proved the degeneration conjecture. To the authors’ knowledge, however, the version of the degeneration conjecture for k((β))-linear categories, which is the one which applies to categories of the form MF(X, W ), does not appear in the literature. This is sometimes referred to as a G-quasi-projective scheme. 6 Our stacks will be classical whenever we are studying the derived category of coherent sheaves D Coh(X/G) and its relatives (QC(X),Perf(X), etc.), but when we discuss categories of matrix factorizations MF(X, W ) and its relatives (IndCoh(X), PreMF(X, W ), PreMF (X, W ), etc.), it will be convenient to work with derived stacks. We will work with k-linear dg-categories. For some of the more abstract arguments involving homotopy limits and colimits and symmetric monoidal structures, it will be more convenient to replace them with equivalent stable (i.e. pre-triangulated) dg-categories in the Morita model struc- ture on dg-categories, then to regard them as k-linear stable ∞-categories via the equivalence of [C]. We permit ourselves a bit of fluidity on this point, in that we refer both to the literature on dg-categories and stable ∞-categories as needed for constructions which evidently make sense in either context. Acknowledgements. We would like to thank D. Ben-Zvi, D. Kaledin, F. Morel and A. Preygel for useful conversations. Our thesis advisor C. Teleman explained to us many of the ideas in equivariant K-theory, geometric invariant theory, and noncommutative geometry which this paper builds upon. We are grateful to B. Toen for explaining his computation of the periodic cyclic homology of a Deligne-Mumford stack, and to A. Blanc for useful discussions and for his careful reading of the first draft of this manuscript. Finally, we would like to thank the anonymous referee for several suggestions which improved this article. The first named author was supported by an NSF fellowship, Columbia University, and the Institute for Advanced Study. The second named author was supported by Kavli IPMU, Imperial College, and an EPSRC fellowship. 1. The noncommutative motive of a quotient stack In this section, we show that the noncommutative Hodge-de Rham spectral sequence degenerates for Perf(X) for a large class of smooth quotient stacks subject to a properness condition. Our method for establishing the degeneration property will be to systematically realize the derived category of a smooth quotient stack as being “glued together” from (typically infinitely many) copies of the derived category of smooth Deligne-Mumford stacks. This method will be used several times throughout this paper, so we formulate our main result in a way that can be applied directly in different contexts. We work with the category M of k-linear additive motives in the sense of [T1] (see also [BGT] for a construction using the framework of ∞-categories). This is the ∞-category obtained as the left Bousfield localization of the ∞-category of small k-linear dg-categories localized at the class of morphisms C → A ⊕ B coming from split exact sequences of small dg-categories A → C → B. In other words, objects of M are dg-categories [C], where we have formally adjoined the relation [C] = [A] ⊕ [B] whenever we have a semiorthogonal decomposition C = hA,Bi. We denote the localization functor U : dgCat → M . k k 1.1. Recollections on KN-stratifications. Our primary geometric tool will be a “KN stratifica- tion” of a quotient stack, as defined in [T2, (1.1)] or [HL, Definition 2.2]. This is a decomposition of X as a union of G-equivariant, smooth, locally closed subschemes ss X/G = X /G∪ S /G. (1) For instance, when X is projective-over-affine and G is reductive, a KN-stratification of X/G is induced by a choice of G-linearized ample line bundle L and a Weyl-invariant inner product on the cocharacter lattice of G. Throughout our discussion, we will assume that we have fixed a choice of inner product on the cocharacter lattice of G, and we will refer to the KN-stratification induced by L as the L-stratification. 7 For each i there is a distinguished one parameter subgroup λ of G. If we let L be the centralizer i i of λ , then there is a smooth open subvariety Z ⊂ X which is L -invariant. Then by definition i i i we have n o S := G· x ∈ X| lim λ (t)· x ∈ Z i i i t→0 When the KN stratification arises from GIT, then in fact Z is the semistable locus for the action of L = L /λ (G ) on the closure of Z . i i m i The main object of study in this paper will be quotients stacks admitting a KN stratification of the following form: ss ′ Definition 1.1. A KN-stratification of a quotient stack X/G is semi-complete if X /G and Z /L all admit good quotients which are projective-over-affine. We say that the KN-stratification is complete if all of the qood quotients are projective. ss ′ Remark 1.2. Given a KN-stratification of a G-scheme X, if X /G and Z /L all admit semi- complete (resp. complete) KN stratifications, then the stratification of X can be refined to a semi-complete (resp. complete) KN stratification by replacing each stratum with the preimage of ′ ′ the strata of Z /L under the projection S /G → Z /L and taking the distinguishing one-parameter i i i i i subgroup of each of these new strata to be λ plus a very small rational multiple of the distinguished one-parameter subgroup of the corresponding stratum in Z /L (which can be lifted to L rationally). In a sense the main theorem of GIT is the following: Theorem 1.3. Given a reductive G and any G-ample bundle on a projective-over-affine G-scheme X, the L-stratification is semi-complete. Semi-complete KN stratifications are important because they lead to direct sum decompositions of noncommutative motives. If I is a (possibly infinite) totally ordered set and A is a pretriangulated dg-category, we say that A = hA ; i ∈ Ii forms a semiorthogonal decomposition if objects of the full pre-triangulated dg-subcategories A generate A under cones and shifts, and RHom(A ,A ) = 0 i i j for i > j. In other words, a semiorthogonal decomposition of a pre-triangulated dg-category is by definition a semiorthogonal decomposition of its homotopy category. Lemma 1.4. If X is a smooth G-scheme with a KN stratification, we have an equivalence in M ss U (Perf(X/G)) ≃ U (Perf(X /G))⊕ U (Perf(Z /L )). k k k i i Proof. The main theorem of [HL] provides an infinite semiorthogonal decomposition of Perf(X/G) ss under these hypotheses. One factor of the semiorthogonal decomposition is equivalent to Perf(X /G), and the rest are of the form Perf(Z /L ) , where the subscript denotes the full subcategory of ob- i i w jects whose homology sheaves are concentrated in weight w with respect to λ. The fact that U commutes with filtered colimits implies that the infinite semiorthogonal decomposition maps to an infinite direct sum decomposition of U (Perf(X/G)) ∈ M . On the other hand, the category k k Perf(Z /L ) decomposes as a direct sum of the subcategories Perf(Z /L ) over all w ∈ Z, so i i i i w U (Perf(Z /L ) ) ≃ U (Perf(Z /L )) ∈ M . k i i w k i i k We will also use KN stratifications to compare properness of the dg-category Perf(X/G) to ss properness of the dg-category Perf(X /G) and Perf(Z /L ) for all i. i i Lemma 1.5. Let X be a perfect derived k-stack of finite cohomological dimension, and let Y be cl,red cl,red another perfect derived k stack such that Y ≃ X and O is eventually co-connective. Then the following are equivalent (1) H RΓ(X, F) is finite dimensional for all i and all F ∈ D Coh(X), (2) RΓ(X, F) is finite dimensional for all F ∈ Coh(X), 8 6 (3) Perf(Y) is a proper dg-category. Furthermore, if X is a separated DM stack then this is equivalent to X being proper. Proof. Finite cohomological dimension implies that for any F ∈ D Coh(X) and all i ∈ Z, there is a sufficiently high n such that H RΓ(X, τ F) ≃ H RΓ(X, F), so (2) ⇒ (1). Also, (1) ⇒ (2) i ≤n i because Coh(X) ⊂ D Coh(X) and H RΓ(F) vanishes in all but finitely many degrees. It is clear cl,red cl that (2) can be checked on X because every F ∈ Coh(X) is pushed forward from X , and any cl cl,red F ∈ Coh(X ) has a finite filtration whose associated graded is pushed forward from X . To show that (2) ⇔ (3), it thus suffices to show that (2) is equivalent to Perf(X) being a proper dg-category in the case when X is eventually co-connective. Because X is perfect, for any F ∈ D Coh(X) and any n we can find a perfect complex P such that F is a retract of τ P, ≤n so choosing n large enough shows that H RΓ(X, F) is a retract of H RΓ(X, P), which is finite i i if Perf(X) is a proper dg-category. On the other hand, Perf(X) ⊂ D Coh(X) if X is eventually co-connective, so Hom (E, F) = RΓ(E ⊗ F) is finite dimensional for perfect complexes E and F. For the further claim, it suffices to assume that X is classical. In this case if X is a separated DM stack, one may find a proper surjection from a quasi-projective scheme X → X [O1], and then deduce that X is proper from property (2), and hence X is proper. Lemma 1.6. Let X/G be a quotient stack with a KN stratification. Then Perf(X/G) is a proper ss ′ dg-category if and only if Perf(X /G) and Perf(Z /L ) are proper dg-categories for all i. Proof. It suffices to consider the case of a single closed stratum S ⊂ X with center Z ⊂ S and with open complement U. First assume that Perf(X/G) is a proper dg-category. [H1, Theorem 2.1] a fully faithful embed- ding Perf(U/G) ⊂ D Coh(X/G) (in fact one for each choice of w ∈ Z), and to prove the lemma it will suffice by Lemma 1.5 to show that this embedding preserves RΓ. We will adopt the notation of w − [H1]: this amounts to showing that we can choose a w such that for F ∈ G ⊂ D Coh(X), which − ss ss is identified with D Coh(X ) under restriction, we have RΓ(X, F) ≃ RΓ(X , F). This holds for w = 0 by [H1, Lemma 2.8] Regarding X as a derived stack, we may define the derived fixed locus Z/L, whose underlying classical stack is Z/L. Then [H1, Theorem 2.1] shows that the functor ∗ − ′ − 0 − ˜ ˜ i π : D Coh(Z/L ) ≃ D Coh(Z/L) → D Coh(X/G) ′ ′ is fully faithfull. By Lemma 1.5 the dg-category Perf(Z/L ) is proper, and thus so is Perf(Z/L ). Conversely, assume that Perf(Z/L ) and Perf(U/G) are both proper dg-categories. We will show that Perf(X/G) is proper by invoking Lemma 1.5 and showing that H RΓ(X, F) is finite dimensional for any n and any coherent sheaf F. Again by [H1, Theorem 2.1], we can functorially ′ 0 write F as a finite extension of an object F ∈ G and two objects supported on the unstable − ≥0 − <0 stratum S = S/G, one in D Coh (X) and one in D Coh (X) . In particular as noted above S S ′ ′ we have RΓ(X, F ) ≃ RΓ(U/G, F ), which has finite dimensional homology. ′′ ′′ − Thus it suffices to show that RΓ(X, F ) has finite dimensional homology for any F ∈ D Coh(X) which is set theoretically supported on S. Because X has finite cohomological dimension, we may ′′ b truncate F so that it lies in D Coh(X), and then in can be built out of a sequence of extensions of shifts of objects of the form i E for E ∈ Coh(S/G). Thus it suffices to show that Perf(S/G) is proper. A similar filtration argument using the baric decomposition of [H1, Lemma 2.2] can be used to deduce that Perf(S/G) is proper because Perf(Z/L) is proper. Finally, the projection We will need to consider the derived critical loci below, which is why we have introduced derived stacks here. If X is classical, then there is no need to replace X by an eventually co-connective approximation in (3), but the example Y = X = Spec(k[t]), where t is a variable of homological degree 2, shows that (2) does not imply (3) without the eventually co-connective hypothesis. 9 ′ Z/L → Z/L is a G -gerbe, so the pushforward preserves perfect complexes, and thus Perf(Z/L) is proper if Perf(Z/L ) is proper. Corollary 1.7. Let X/G be a quotient stack with a semi-complete KN stratification. Then Perf(X/G) is a proper dg-category if and only if the stratification is complete. Proof. Combine Lemma 1.6 with Lemma 1.5. 1.2. Motivic decompositions via KN stratifications. We will consider the class of stacks ⊕N which have semi-complete KN stratifications as in Definition 1.1. We use the notation C to denote the direct sum of countably many copies of the dg-category C. Recall also the definition of [K] that a DM stack of finite type with finite inertia over a field of characteristic 0 is quasi-projective if X is a global quotient stack and has a quasi-projective coarse moduli space. We will say that X is furthermore projective-over-affine if its coarse moduli space is projective over an affine variety. Theorem 1.8. Let G be an algebraic group over a field k of characteristic 0. Let X be a smooth G- quasiprojective k-scheme with a semi-complete KN stratification. Then there is a smooth projective- ⊕N over-affine Deligne-Mumford stack Y such that U (Perf(X/G)) is a direct summand of U (Perf(Y)) k k in M . Furthermore if Perf(X/G) is a proper dg-category, then Y can be chosen to be a smooth projective scheme. Note that by Corollary 1.7, Perf(X/G) is proper if and only if the KN stratification is complete. Remark 1.9. The proof is constructive, and actually produces something a bit stronger: if C is the ∞-category of small dg-categories, then Perf(X/G) lies in the smallest subcategory containing Perf(Y) and closed under countable semiorthogonal gluings and passage to semiorthogonal factors. Example 1.10. If X is projective-over-affine with a linearizable G-action, then the condition that Perf(X/G) is a proper dg-category is equivalent to the condition that H RΓ(X,O ) is finite 0 X dimensional, by [HLP, Proposition 4.2.3]. Example 1.11. We can write any algebraic k-group G as a semidirect product G = U ⋊ L, where U is its unipotent radical and L its reductive quotient. Assume that there is a one-parameter subgroup λ : G → L which is central in L and acts with positive weights on Lie(U) in the adjoint representation of G. Then this one-parameter subgroup defines a single KN stratum S = X = {∗}, and Z/L = ∗/L → ∗ is a good quotient. Thus Theorem 1.8 applies to a large class of categories of the form Perf(BG), including when G is a parabolic subgroup of a reductive group. Example 1.12. If G is as in the previous example, and X is a smooth projective-over-affine G- scheme, then one can consider the Bialynicki-Birula stratification of X under the action of λ(G ), λ(G ) L which is a KN stratification. If this is exhaustive, and Γ(X ,O ) is finite dimensional, λ(G ) then the Bialynicki-Birula stratification can be refined to a complete KN stratification of X as in Remark 1.2. Our proof of Theorem 1.8 will proceed by a delicate inductive argument. One of the key tools is the following: Lemma 1.13. Let π : Y → X be a rational morphism of finite-type k-stacks, meaning Rπ O ≃ O . Y X Assume that X is smooth and π preserves D Coh. Then U (Perf(X)) is a summand of U (Perf(Y)) k k in M . Proof. First consider the categories Perf(Y) and Perf(X). The unit of adjunction id → π π is X/G an equivalence in Perf(X), hence π is fully faithful and admits a right adjoint. Hence Perf(X) is a semiorthogonal factor of Perf(Y). The original version of this paper had Y as a projective DM stack, but by subsequent work of Bergh, Lunts, and Schnuerer [BLS] one can further reduce to a smooth projective scheme. 10 We will apply Lemma 1.13 in three different situations. Example 1.14. If π : Y → X is a flat morphism of algebraic stacks such that for every k-point p of X the fiber Y satisfies RΓ(Y ,O ) ≃ k, then π is rational. If π is not flat, then the same is true p p Y if we take Y to refer to the derived fiber. Example 1.15. Any representable birational morphism of smooth k-stacks is rational. Indeed we can reduce this to the case for schemes, as birational morphisms are preserved by flat base change and the property of a morphism being rational is fppf-local on the base. Example 1.16. Let G → H → K be an extension of linearly reductive groups, and let K act on a scheme, X. Then the morphism p : X/H → X/K is a G-gerbe – after base change to X this morphism becomes the projection X × BG → X. Thus because G is linearly reductive Rp O ≃ O . X/H X/K Let π : X → X be a projective morphism of smooth projective-over-affine varieties which is equivariant with respect to the action of a reductive group, G. For a G-ample invertible sheaf L on X and a relatively G-ample invertible sheaf M on X , we consider the fractional polarization L = L + ǫM for ǫ ∈ Q. We will need the following: Lemma 1.17. [T2, Lemma 1.2] For any small positive ǫ ∈ Q, the L -stratification of X refines the preimage of the L-stratification of X. Finally, we need another GIT lemma: Lemma 1.18. Let X be a G-quasi-projective scheme which admits a good quotient π : X → Y such ss that Y is projective-over-affine. Then X = X for some linearized projective-over-affine G-scheme X, which can be chosen to be smooth if X is smooth. Proof. The proof of [T2, Lemma 6.1] applies verbatim: one constructs a relative G-compactification for X → Y by choosing a sufficiently large coherent F ⊂ π O so that X embeds in the projectiviza- ∗ X tion of Spec Sym(F). The closure of X is projective over Y , and hence projective-over-affine, and ss it has a linearization for which X = X by the cited argument. Furthermore, one can equivariantly resolve any singularities occuring in X \ X if X is smooth. Proof of Theorem 1.8. Over the course of the proof, we will actually construct a finite set of smooth ⊕N quasi-projective DM stacks Y , . . . ,Y such that U (Perf(X/G)) is a retract of U (Perf(Y )) ⊕ 1 N k k 1 F F ⊕N ···⊕U (Perf(Y )) , and then we may take Y = Y ··· Y at the end. Also, when Perf(X/G) k N 1 N is a proper dg-category, it suffices to show that the Y can be chosen to be smooth proper DM stacks, because [BLS, Theorem 6.6] implies that for any smooth proper DM stack Y, Perf(Y) is geometric, meaning a semiorthogonal factor of Perf of a smooth projective variety. We shall prove the theorem by induction on the rank of G. Note that by Lemma 1.4 and the definition of a semi-complete KN-stratification, it suffices to prove this for quotient stacks which have projective-over-affine good quotients. For our purposes, it will be more convenient to consider smooth G-schemes which are projective-over-affine, and by Lemma 1.18 and Lemma 1.4 it suffices to prove the claim for open unions of KN strata in a quotient stack of this form. We fix a G-ample bundle L on X and consider the L-stratification as in Equation 1. ss Case X = ∅: By Lemma 1.4 we must verify the claims for U (Perf(Z /L )) for all i for which Z ⊂ U. First i i i assume that the inclusion λ(G ) ⊂ L admits a splitting L → G , so that L ≃ G × L where m i i m i m the left factor is λ(G ). Then Z /L ≃ BG × Z /L , so U (Perf(Z /L )) is a direct sum of copies m i i m i k i i )). This is the only point of the proof at which an infinite direct sum enters, and of U (Perf(Z /L k i 11 it is an infinite direct sum of copies of the same category, hence throughout the proof we will only encounter a finite set of distinct DM stacks. If λ(G ) ⊂ L is not split, then we can choose a surjective homomorphism L → L with finite m i i ˜ ˜ kernel, where L ≃ G × L and G ×{1} → L factors through λ(G ). The morphism p : Z /L → m m i m i Z /L is rational, hence Lemma 1.13 reduces the problem to showing the claim for Z /L. By the i i i argument of the previous paragraph it again suffices to prove the claim for Z /L . ¯ i Let Z be the closure of Z , which is a connected component of X and hence smooth and i i ′ ′ projective-over-affine. Then Z is the semistable locus for the action of L on Z , and L has lower i i i i rank than L , so the first claim of the theorem follows from the inductive hypothesis. Note that Lemma 1.6 implies that Perf(Z /L ) is a proper dg-category if Perf(U/G) is, so in this case the inductive hypothesis implies that one can choose the DM stacks Y to be projective. s ss Case X = X 6= ∅: ss The argument in the case where X = ∅ applies here as well, so the inductive hypothesis implies that the conclusion of the theorem holds for Perf(Z /L ) for all Z ⊂ U. By Lemma 1.4 it suffices to i i i ss ss show that the claims hold for X /G. In this case X /G is a smooth separated Deligne-Mumford stack whose coarse moduli space is projective-over-affine. Furthermore if Perf(U/G) is a proper ss ss dg-category then so is Perf(X /G) by Lemma 1.6, and hence X /G is projective by Lemma 1.5. ss ss s ss s ss Case X 6= ∅, X 6= X , and codim(X \ X , X ) ≥ 2: ss As in the previous case, it suffices to show the claims for X /G. Here we use the main result of ′ ′ ss ′ s [K6], which says that there is a birational morphism π : X → X such that (X ) (L ) = (X ) (L ), ǫ ǫ Lemma 1.17 the open subset where L = π L + ǫM for a suitable relatively G-ample M. By ′ −1 ss ′ ss U := π (X (L)) is a union of KN strata, and π : U → X (L) is rational, so by Lemma 1.13 we may reduce the main statement of the theorem to the corresponding claim for U /G, which falls under the previous case. Furthermore, if Perf(U/G) is a proper dg-category the fact that ′ ′ U /G → U/G is proper implies that Perf(U /G) is a proper dg-category, so again we may reduce to the previous case to show that the Y can be chosen to be projective. ss ss s ss Case X 6= ∅ but codim(X \ X , X ) ≤ 1: Let Y be a smooth projective variety with a G-action such that RΓ(Y,O ) ≃ k and for some linearization M we have codim(Y \ Y , Y ) ≥ 2. For instance, Y could be a suitable product of flag varieties, or a large projective space with a suitable linear G action. We linearize the G action on X × Y with L = L + ǫM. By Lemma 1.17, the open subvariety U × Y ⊂ X × Y is a union of KN strata for the L stratification. The projection U ×Y/G → U/G is rational, and so by Lemma 1.13 it suffices to prove the claims for U × Y/G. Note that Perf(U × Y/G) is a proper dg-category if Perf(U/G) is. Finally we have a sequence of inclusions ss s s ss ss X × Y ⊂ (X × Y ) ⊂ (X × Y ) ⊂ X × Y, where the first inlcusion is due to the fact that points in Y have finite stabilizers and thus so do ss s points in X × Y , and the last inclusion follows from Lemma 1.17. This implies that ss s ss ss s ss s codim((X×Y ) \(X×Y ) , (X×Y ) ) ≥ codim(X ×(Y \Y ), X ×Y ) = codim(Y \Y , Y ) ≥ 2, which reduces us to the previous case. 1.3. Λ-modules and the noncommutative Hodge-de Rham sequence. Let us recall the negative cyclic and periodic cyclic homology of a small k-linear dg-category, A. We let C (A) ∈ 12 2 D(Λ) denote the (mixed) Hochschild complex of A, regarded as a dg-module over Λ = k[B]/B where B has homological degree 1 and acts on C (A) by the Connes differential. We have (n) C (A) := C (A)⊗ k[u]/(u ) • • (n) C (A) := lim C ←− per − C (A) := C (A)⊗ k((u)) • • k[[u]] where u is a variable of homological degree −2. The differential on each complex is given by d+uB, where d is the differential on C (A). In fact, these constructions make sense for any Λ-module M. We sometimes denote the negative cyclic construction M and the periodic cyclic construction Tate M . See Lemma 3.2 below. (n) Definition 1.19 ([KS]). The category A is said to have the degeneration property if H (C (A)) ∗ • is a flat k[u]/(u )-module for all n ≥ 1. It is immediate from the definitions that the degeneration property is preserved by filtered colimits of dg-categories. It is also known that the degeneration property holds for categories of the form A = Perf(R), where R is a smooth and proper dg-algebra [K3, K2]. In particular, this holds when A = Perf(X), where X is a smooth and proper Deligne-Mumford stack over k [HR], although a more direct argument in this case follows from Proposition 2.13 below. If A satisfies the degeneration property, then H (C (A)) is a flat k[[u]]-module. (See [KS, Corollary 9.1.3]). The degeneration property owes its name to its relationship with the noncommutative Hodge-de Rham spectral sequence. This is the spectral sequence associated to the filtration of the complex p per p − per F C (A) = u · C (A) ⊂ C (A). • • • per The E page of the spectral sequence is gr C (A) ≃ C (A) ⊗ k((u)). The degeneration property 1 • • implies that this spectral sequence degenerates on the first page, so the associated graded of the per resulting filtration on H (C (A)) is isomorphic to H (C (A))⊗k((u)). Under the assumption that ∗ • ∗ • A is suitably bounded, we can say something more precise: Lemma 1.20. Let A be a dg-category such that H (C (A)) is homologically bounded above and ∗ • which satisfies the degeneration property. Then there exists an (non-canonical) isomorphism H (C (A)) ≃ H (C (A))⊗ k[[u]]. ∗ • Proof. This follows from the remark before Theorem 4.14 of [KKP1]. The hypothesis of Lemma 1.20, that H (C (A)) is homologically bounded above, will apply to ∗ • Perf(X) for all smooth k-stacks X of finite cohomological dimension such that QC(X) is compactly generated. Remark 1.21. The Λ-module C (A) is functorial in A. When A is a symmetric monoidal k-linear ∞-category, exterior tensor product followed by the symmetric monoidal product gives a natural map − − − − C (A)⊗ C (A) → C (A⊗ A) → C (A) • • • • per per and likewise for C (A). On the level of homology, this gives H C (A) the structure of a • ∗ • commutative k((u))-algebra, and for any symmetric monoidal functor between symmetric monoidal per per ∞-categories A → B, the resulting map H C (A) → H C (B) is a map of commutative k((u))- ∗ • ∗ • algebras. 8 p p+1 Note that the filtration is not a filtration of k((u))-modules, as u · F ⊂ F . As explained to us by D. Kaledin, this can be understood by thinking of u as the Tate motive. In other words when k ⊂ C, rather than regarding k[[u]] ∗ ∞ simply as a complex (where u has cohomological degree 2), we regard it as H (P ) with its Hodge structure, which places u in F C[[u]]. 13 1.4. The degeneration property for quotient stacks. In [T2, Theorem 7.3], Teleman estab- lishes the degeneration of a commutative Hodge-de Rham sequence, which converges to the equivari- ant Betti-cohomology H (X), for a smooth quotient stack X/G with a complete KN stratification. The argument in [T2] makes use of the KN stratification and has a similar flavor to the proof of Theorem 1.8. However the proof in the commutative case is substantially simpler. In the noncom- mutative situation, we are not aware of an argument to reduce the proof of degeneration to the case of the quotient of a smooth projective scheme by the action of a reductive group, as was done in [T2]. Using the motivic statement of Theorem 1.8, we can immediately deduce noncommutative HdR degeneration. The main observation is the following Lemma 1.22. The degeneration property is closed under direct summands and arbitrary direct sums in M . Proof. The Hochschild complex C (−) is an additive invariant of dg-categories, hence factors through U uniquely up to contractible choices. The claim follows from the fact that the oper- ation D(Λ) → D(k) taking mapping (M, d, B) 7→ (M ⊗ k[u]/(u ), d + uB) commutes with filtered colimits and in particular infinite direct sums, and the fact that an infinite direct sum of k[u]/(u ) modules is flat if and only if every summand is flat. Corollary 1.23. Let G be a reductive group and let X be a smooth G-quasiprojective scheme which admits a complete KN stratification. Then Perf(X/G) has the degeneration property. Proof. Combine Lemma 1.22 with the conclusion of Theorem 1.8. Example 1.24. As a counterexample, consider Perf(BG ). This category is Morita equivalent to the category Perf(k[ǫ]/(ǫ )) where ǫ has degree −1. By the (graded-commutative) HKR theorem, 2 2 ∗ proposition 5.4.6 of [L1], we have that H C (Perf(k[ǫ]/(ǫ )) = k[ǫ]/(ǫ )⊗ Sym (dǫ), where dǫ has ∗ • degree 0. By theorem 5.4.7 of the same book, the Connes operator goes to the de Rham differential which sends ǫ → dǫ and so the spectral sequence does not degenerate. We also observe, somewhat surprisingly, that the derived category of coherent sheaves on cer- tain singular quotient stacks also has the degeneration property. We will consider the following goeometric set up ss • X/G = X /G ∪ S /G is a complete KN stratification (Definition 1.1) of a smooth quotient stack, • V is a G-equivariant locally free sheaf on X such that V| has λ -weights ≤ 0 for all i, and Z i • σ ∈ Γ(X, V ) is an invariant section. Note that the quantization-commutes-with-reduction theorem [T2] implies that if the λ -weights of G ss G V| are strictly negative, then Γ(X, V ) ≃ Γ(X (L), V ) (this is referred to as adapted in [T2]). Using the methods of [HL] one can show that dim Γ(X, V ) < ∞ even when the λ weight of V| i Z vanishes for some i. Amplification 1.25. In the set up above, if ss (1) σ is regular on X with smooth vanishing locus, and λ=0 (2) for all i the restriction of σ to (V| ) , the summand of V| which is fixed by λ(G ), is Z Z m i i regular with smooth vanishing locus, then there is a smooth and proper quasi-projective DM stack Y such that U (D Coh(X /G)) is a k σ ⊕N retract of U (Perf(Y)) . Proof. We apply the structure theorem for the derived zero locus X in [H1, Theorem 3.2], whose derived category is just the derived category of the sheaf of cdga’s over X/G given by the Koszul 14 algebra A = (Sym(V [1]), dφ = φ(s)). The structure theorem constructs an infinite semiorthogonal decomposition which generalizes the b ss main structure theorem of [HL]. One factor is isomorphic to D Coh(X /G), and the remaining b ′ w ′ factors are isomorphic to D Coh(Z /L ) , where Z denotes the derived zero locus of σ restricted i i λ=0 b ′ to (V| ) , and the superscript w denotes the full subcategory of D Coh(Z /L ) consisting of Z i complexes whose homology is concentrated in weight w. In order to apply this theorem, we must check that after restricting the cotangent complex L X /G to Z /L and looking at the summand with λ-weights < 0, there is no fiber homology in homological degree 1. Because X is a derived zero section, we have h i λ<0 ∨ λ<0 λ<0 ∨ λ<0 ′ ′ ′ L | ≃ (V | ) → (Ω | ) → O ⊗ (g ) . Z X X /G Z Z Z σ i i i i So the weight hypotheses on V| imply that this is a two term complex of locally free sheaves in homological degrees 0 and −1, and hence has no fiber homology in homological degree 1. Lemma 1.4 now applies verbatim Given the structure theorem for D Coh(X /G), the proof of to give a finite direct sum decomposition b b ss b ′ U (D Coh(X /G)) = U (D Coh(X /G))⊕ U (D Coh(Z /L )). k σ k k i σ i Under the hypotheses of the amplification, each factor in this direct sum decomposition is D Coh of a smooth quotient stack satisfying the hypotheses of Theorem 1.8, and the result follows. Remark 1.26. Note that when V is strictly adapted to the KN stratification, then the condition (2) in the previous amplification is vacuous. Corollary 1.27. In the set up of Amplification 1.25, the category D Coh(X /G) has the degener- ation property. We will see another approach to establishing Corollary 1.27 using graded LG-models in Example 3.23 and Proposition 3.26. 2. Hodge structures on equivariant K-theory In this section we consider the action of a reductive group G on a smooth quasi-projective per C-scheme X. Our goal is to identify the periodic cyclic homology C (D Coh(X/G)) with the an complexification of the Atiyah-Segal equivariant topological K-theory K (X ) with respect to a maximal compact subgroup M ⊂ G (see Section 2.1.2). Our final result, Theorem 2.20, will n an allow us to define a pure Hodge structure of weight n on K (X ) in the case where X admits a complete KN stratification. Rather than construct a direct isomorphism, we study an intermediate object, the topological K- top b theory of the dg-category K (D Coh(X/G)), as defined in [B3], which admits natural comparison top isomorphisms with each of these theories. In Blanc’s construction, K (C) is constructed from the geometric realization of the presheaf of spectra on the category, Aff, of affine C-schemes of finite type, K(C) : A 7→ K(A⊗ C). Here K(−) denotes the non-connective algebraic K-theory of a dg-category of [B3, Definition 2.10]. The geometric realization of a presheaf, | • |, is defined to be the left Kan extension of ∞ an the functor A 7→ Σ (SpecA) , regarded as functor with values in spectra, along the Yoneda embedding of the category of finite type C-schemes into presheaves of spectra, Aff → Sp(Aff). The geometric realization functor | • | : Sp(Aff) → Sp admits a right adjoint, which assigns M ∈ Sp 15 ∞ an to the presheaf of spectra H (M) := Hom (Σ (•) , M). The semi-topological K-theory is the B Sp geometric realization st K (C) := |K(C)|, st st regarded as a K (C)-module spectrum. By [B3, Theorem 4.5], we have an isomorphism K (C) ≃ bu, where the latter denotes the connective topological K-theory spectrum. Choosing a generator β ∈ π (bu), one then defines the topological K-theory of a dg-category to be −1 top st −1 K (C) := K (C)[β ] = |K(C)|⊗ bu[β ] bu Remark 2.1. In the definition of the presheaf K(C), A⊗ C denotes the derived tensor product of non-idempotent-complete dg-categories, as in [K4, Section 4.3]. In Section 3 we will also consider a symmetric monoidal structure on small stable idempotent complete dg-categories, which we will denote A⊗B to avoid confusion. In addition, following the convention of [B3, Definition 2.7], throughout this paper we define the algebraic K-theory of a small dg-category C to be the algebraic K-theory of the Waldhausen category Perf(C), so that it is automatically Morita invariant. We warn the reader that as a consequence, if C is a stable dg-category, then K(C) denotes the usual algebraic K-theory of the idempotent completion of C, not the K-theory of C itself. per top We will also use the construction of a Chern character map Ch : K (C) → C (C). First, one per obtains a map of presheaves K(C) → C (C) from the usual Chern character in algebraic K-theory, per per where C (C) denotes the presheaf A 7→ C (A⊗ C). Using a version of the Kunneth formula for • • C per per per periodic cyclic homology, one obtains an equivalence |C (C)| ≃ C (C) ⊗ ± |C (C)|. Then • • • C[u ] per one can construct an isomorphism of presheaves C (C) ≃ H (C[u ]), which leads to a map per |C (C)| → C[u ]. Combining these provides a map st per per per K (C) → C (C)⊗ ± |C (C)| → C (C) C[u ] • • • which give the Chern character after inverting β. The main result we use is [B3, Proposition 4.32], which states that for a finite type C-scheme, X, the Chern character induces an equivalence per top K (Perf(X)) ⊗ C → C (Perf(X)). Furthermore, there is a natural topologization map which top an is an equivalence K (Perf(X)) → K(X ), and under this equivalence Ch can be identified with per an a twisted form of the usual Chern character for X under a canonical isomorphism C (X) → H (X;Q)⊗ C[u ]. More precisely, Blanc’s Chern character provides an equivalence Betti Q top an K (Perf(X))⊗ Q ≃ H (X;Q)⊗ Q(( )) ⊂ H (X ;Q)⊗ C((u)), Betti Betti 2πi which we can alternatively express as an isomorphism top 2p−n an π (K (Perf(X)))⊗ Q ≃ H (X ;Qhpi), where Qhpi ⊂ C denotes the subgroup (2πi) Q. 2.1. Equivariant K-theory: Atiyah-Segal versus Blanc. In this section we consider a reduc- tive group G with maximal compact subgroup M ⊂ G, and a G-quasi-projective scheme X, which need not be smooth. The goal of this section will be to construct a comparison isomorphism be- top b an tween K (D Coh(X/G)) and topological M-equivariant K-homology of X with locally compact supports, whose construction and properties we recall below. per In order to be consistent with the rest of the paper, we use the notation C for the periodic cyclic homology complex of a dg-category, rather than the notation HP used in [B3]. In addition, we use the notation ⊗ rather than top ∧ for the smash product of spectra and module spectra. For example K (C) ⊗ C is the C-module spectrum, which top B3]. we canonically identify with a complex of C-modules, which is denoted K (C) ∧ HC in [ 16 2.1.1. Equivariant Borel-Moore homology theories from invariants of dg-categories. Let us consider an additive invariant of dg-categories E(−). Then we have a Borel-Moore homology theory on the category of G-schemes by defining E (X) := E(D Coh(X/G)) (see Remark 2.4 below). This BM assignment is covariantly functorial with respect to proper G-equivariant maps p : X → X by b ′ b applying E(−) to the pushforward functor p : D Coh(X /G) → D Coh(X/G), and E(X) is contravariantly functorial with respect to flat G-equivariant maps f : X → X by applying E(−) ∗ b b ′ to the pullback functor f : D Coh(X/G) → D Coh(X /G). Note also that if X is a smooth G-space, then we have a canonical equivalence E(Perf(X/G)) → E(D Coh(X/G)), so we can canonically identify E (X) with the “cohomology” theory E(X) := E(Perf(X/G)), which is a BM form of Poincar´e duality. More precisely, let Spaces denote the category of algebraic spaces with a G-action and G- equivariant maps, and let Spaces , where ? = f, p,≃, denote the subcategories consisting of flat maps, proper maps, and isomorphisms respectively. We regard the additive invariant E as giving p p f f op two strict functors into spectra E : Spaces → Sp and E : (Spaces ) → Sp along with BM G BM G an isomorphism of the restriction of these functors to the subcategories Spaces ⊂ Spaces and ≃ ≃ op Spaces ⊂ (Spaces ) . Here we have used the fact that Spaces is a groupoid to canonically G G ≃ ≃ op identify Spaces ≃ (Spaces ) . G G Recall that an additive invariant E(−) of dg-categories is localizing if it takes localization se- quences of dg-categories to exact triangles, and we say that E(−) satisfies equivariant d´evissage ′ b if for any closed immersion i : X ֒→ X of G-spaces the induced map i : E(D Coh(X/G)) → b ′ b ′ E(D Coh (X /G)) is an equivalence, where D Coh (X /G) denotes the derived category of quasi- X X coherent complexes with coherent cohomology sheaves which are set theoretically supported on X. Lemma 2.2. Let E be an additive invariant of dg-categories. Then: ′ ′ ′ (1) The base change formula holds: if p : X → Y is proper, f : Y → Y is flat, and f and p are ∗ ′ ∗ ′ base change of f and p, then we there is an equality of compositions p ◦ f = (f ) ◦ (p ) : ∗ ∗ E (X) → E (Y ) in the homotopy category of spectra Ho(Sp); BM BM (2) If V is a G-equivariant vector bundle on X and π : P(V) → X is the projection, then π : E (X) → E (P(V)) followed by (−) ⊗ O (k) : E (P(V)) → E (P(V)) is a BM BM P(V) BM BM split injection; (3) The previous maps, where k ranges from 0, . . . , n− 1, define a canonical equivalence ⊕n E (P(V)) ≃ E (X) . BM BM Furthermore, if E is localizing and satisfies equivariant d´evissage, then: ′ ′ (4) For any closed immersion of G-spaces i : X ֒→ X with open complement j : U ⊂ X , there is an exact triangle E (X) −→ E (X ) −→ E (U) →; BM BM BM (5) If π : V → X is a G-equivariant torsor for V, then π : E (X) → E (V ) is an BM BM equivalence. Proof. The claim (1) follows immediately by applying E(−) to the usual base change formula. Claims (2) and (3) follow from the fact that E is an additive invariant and there is a canonical semiorthogonal decomposition b ∗ b ∗ b ∗ b D Coh(P(V)/G) = hπ (D Coh(X/G)), π (D Coh(X/G))⊗O (1), . . . , π (D Coh(X/G))⊗O (n−1)i. P(V) P(V) Now let us assume that E satisfies d´evissage. (4) follows immediately from the d´evissage condi- tion and the localization exact triangle b ′ b ′ b ′ E(D Coh (X /G)) → E(D Coh(X /G)) → E(D Coh((X − X)/G)) → . 17 For (5), we note that Thomason’s proof when E(−) is algebraic K-theory in [T4, Theorem 4.1] generalizes to this context as well: the torsor V is classified by an extension 0 → V → V → O → 0 of G-equivariant locally free sheaves on X. V is isomorphic to the complement of the closed ′ ′ embedding P(V) ֒→ P(V ). Using the direct sum decomposition of E(P(V )) and E(P(V)) from (3) one can show that E(i ) : E(P(F)) → E(P(W)) is a split injection, and the pullback map gives an equivalence of the cofiber with E(X). It follows from the localization sequence of (4) that the cofiber of E(i ) can is canonically identified with E(P(V )−P(V)) under the restriction map, so we have that pullback gives an equivalence E(X) ≃ E(V ). Example 2.3. Algebraic K-theory of dg-categories as defined in [B3, Definition 2.10] is a localizing top invariant and satisfies d´evissage, and it follows that K (−) is localizing and satisfies d´evissage as well. Remark 2.4. The theory E (X) gives a Borel-Moore homology theory in the sense of [LM, BM Definition 2.1.2], but whose source is the category of G-spaces and which takes values in Ho(Sp) rather than graded abelian groups. We are forced to use the homotopy category of spectra because we have only formulated the base change formula in this setting. In order to formulate a “strict” Borel-Moore homology theory valued in spectra, one would have to define the source category as an ∞-category of correspondences, as in [GR, Section V]. Remark 2.5. Although we are only interested in quotient stacks here, one can define a Borel-Moore homology theory E (X) := E(D Coh(X)) for arbitrary algebraic stacks, and the statements and BM proofs of the previous lemma extend to this context. 2.1.2. Atiyah-Segal equivariant K-theory. We now turn our attention to topological equivariant K-theory with respect to a maximal compact subgroup M ⊂ G. There is a K-cohomology theory for topological M-spaces constructed in [AS ]. Below we use the more systematic description of K (X) in terms of equivariant stable homotopy theory as the spectrum obtained by taking level- wise M-equivariant mapping spaces from X to the naive M-spectrum underlying the M-spectrum bu . For details on the non-equivariant and equivariant stable homotopy category, we refer the + an reader to [LSM] and [M , Chapters XII, XIV, and XVI]. K (X ) can also be constructed as the K-theory spectrum associated via Quillen’s Q-construction to the exact category of M-equivariant an complex vector bundles on X . We will also consider the Atiyah-Segal equivariant K-homology with locally compact supports c,∨ an AS K (X ). This theory was studied in [T5, Section 5] under the notation G (G, X), and our discussion follows this reference closely. In particular, we refer the reader there for a nice discus- c,∨ sion contextualizing K (−) with respect to several other versions of equivariant K-theory. We c,∨ an have chosen to denote the M-equivariant K-homology with locally compact supports as K (X ) because it is the M-equivariant Spanier-Whitehead dual (see [M , XVI.7]) of the M-spectrum of equivariant K-theory with compact supports constructed in [S4, Page 136], which we denote c an K (X ). In particular, K (X) is strictly covariantly functorial for open immersions [S4, Proposition c,∨ 2.9] and strictly contravariantly functorial for proper maps. Dually, it follows that K (−) can op be regarded as both a strict functor Spaces → Sp and a strict functor (Spaces ) → Sp, G G where Spaces ⊂ Spaces denotes the subcategory of open immersions, and these two functors are isomorphic on the subcategory Spaces . Poincar´e duality in this context gives a canonical c,∨ an an isomorphism K (X ) ≃ K (X ) for complex G-manifolds – a priori this isomorphism depends on a choice of orientation, but complex manifolds are canonically oriented for K-theory (see [S4, Section 3] and [M , XIV]). c,∨ an Lemma 2.2 also hold for K (X ). Namely: Lemma 2.6. The properties of 18 (1) The base change formula holds with respect to a proper map p : X → Y and an open immersion f : U ⊂ Y ; (2) If V is a G-equivariant vector bundle on X of dimension n, then there is a canonical c,∨ c,∨ an an ⊕n equivalence K (P(V) ) ≃ K (X ) . M M ′ ′ (3) For any closed immersion of G-spaces i : X ֒→ X with open complement j : U ⊂ X , there is an exact triangle E (X) −→ E (X ) −→ E (U) →; BM BM BM (4) If π : V → X is a G-equivariant torsor for V, then there is a canonical equivalence E (X) → E (V ). BM BM ′ −1 Proof. The base change formula arises from the observation that if U := p (U) ⊂ X, then there is a commutative diagram of maps of one point compactifications + ′ + // X (U ) , + + // Y U and it is precisely pullback along these maps which is used to define proper pullback and pushfor- ward along an open immersion in K (−). The canonical equivalence (2) is dual to the equivalence c c ⊕n K (P(V)) ≃ K (X) of [S4, Proposition 3.9]. (3) is [S4, Proposition 2.9]. One proves (4) using M M c,∨ c,∨ an the same method as the proof of Lemma 2.2: the canonical functor K (X ) → K (V ) is the M M c,∨ c,∨ th ′ an ⊕n+1 inclusion of the 0 piece of the direct sum decomposition K (P(V )) ≃ K (X ) followed M M by restriction to the open subset V ⊂ P(V ). c,∨ ∗ an Remark 2.7. It is a priori not clear how to define a canonical pullback map f : K (Y ) → c,∨ an K (X ) for a flat map f : X → Y . That is why we have only stated the projection formula for open immersions, and it’s why in (2) and (4) above we can not state that the canonical isomorphisms c,∨ an are given by pullback. Indeed it is not immediate in (4) that the isomorphism K (X ) ≃ c,∨ an K (V ) is independent of the presentation of V as a torsor for V. If f is the restriction of a flat ′ ′ ′ map X → Y along a closed embedding Y ֒→ Y , then one could provisionally define a pullback map using the localization sequence (3), but perhaps the most uniform definition of flat pullback c,∨ for K is via the isomorphism of Theorem 2.10 below. 2.1.3. The comparison map. We now construct a comparison map between Borel-Moore homology c,∨ top b an theories ρ : K (D Coh(X/G)) → K (X ) by first constructing a comparison isomorphism G,X top an between the corresponding cohomology theories K (Perf(X/G)) → K (X ). Observe that we have a natural transformation of presheaves of spectra alg an an K (Perf(X/G× T )) → K (X × T ) which is induced by the functor of exact categories which sends an algebraic G-vector bundle to its underyling complex topological vector bundle equipped with the induced action of M (This functor is symmetric monoidal, and hence induces a map of K-theory spectra [T5, Section 5.4]). By the an an ∞ an an Lemma 2.8 below, the presheaf K (X ×(−) ) is weakly equivalent to Hom (Σ (−) , K (X )), M Sp M where Hom (•,•) denotes the internal function spectrum in the category of spectra. Composing Sp this with the comparison map above gives a map of presheaves alg ∞ an an K (Perf(X/G)⊗ O )) → Hom (Σ T , K (X )), C T Sp M where both sides are regarded simultaneously as presheaves in the G-space X and the affine scheme T . Here we have used the natural Morita equivalence Perf(X/G)⊗ O −→ Perf(X/G×T ). By the C T B3, Definition 3.13] for presheaves on the category adjunction defining the geometric realization [ 19 of affine schemes, this gives a map st an K (Perf(X/G)) → K (X ) of presheaves in the G-space X. This natural transformation of presheaves on Spaces will be an bu-linear by construction. Because K (X ) already satsifies Bott periodicity, this comparison map extends uniquely to the localization, giving our final comparison map perf top st −1 an ρ : K (Perf(X/G)) := K (Perf(X/G))⊗ bu[β ] → K (X ). bu G,X The following technical lemma was used in the construction above: Lemma 2.8. For any topological space Y and topological M-space X, we have a natural weak equivalence in Sp, Hom (Σ Y , K (X)) = K (X × Y ), Sp + M M where on the right Y is regarded as an M-space with trivial M action. Proof. We fix a universe U for forming the equivariant stable homotopy category Sp as in [LSM]. The “change of universe” functor taking an E ∈ Sp to its underlying naive M-spectrum admits a left adjoint, as does the functor from naive M-spectra to spectra which applies the M-fixed point functor level-wise. We will denote the composition of these to functors as (−) : Sp → Sp, and it therefore has a left adjoint, which we denote ι. By definition we have that K (X) := Hom (Σ X , bu ) M Sp + M where bu is the M-spectrum representing equivariant K-theory, Σ is the stabilization functor from pointed M-spaces to M-spectra, and Hom is the internal function spectrum in the symmet- Sp ric monoidal category of M-spectra [LSM, page 72].Thus by the (spectrally enhanced) adjunction and the definition of inner Hom in a symmetric monoidal category we have ∞ ∞ ∞ Hom (Σ Y , K (X)) ≃ Hom (ι(Σ Y )∧ Σ (X ), bu ) Sp + M Sp + + M ∞ ∞ The claim now follows from the natural isomorphism ι(Σ Y ) ≃ Σ (Y ) [LSM, Remark II.3.14(i)], + + where Y is regarded as an M-space with trivial M action, the fact that Σ maps smash products of pointed M-spaces to smash products of M-spectra [LSM, Remark II.3.14(iii)], and the fact that Y ∧ X ≃ (Y × X) for pointed M-spaces. + + + top an Now that we have constructed a comparison map of presheaves K (Perf(X/G)) → K (X ), c,∨ top b an we construct a comparison map for Borel-Moore homology ρ : K (D Coh(X/G)) → K (X ) G,X for a G-quasi-projective scheme X as follows: Choose a G-equivariant closed immersion i : X ֒→ Z into a smooth G-scheme Z with open complement j : U ⊂ X. Then by Lemma 2.2 and Lemma 2.6, and using the canonical equivalence Perf(Z/G) ≃ D Coh(Z/G), there is a dotted arrow completing the map of exact triangles top b top top // // // K (D Coh(X/G)) K (Perf(Z/G)) K (Perf(U/G)) (2) perf perf G,X ρ ρ G,Z G,U i j c,∨ c,∨ c,∨ an an an // // // K (X ) K (Z ) K (U ) M M M c,∨ top b an Lemma 2.9. The map ρ : K (D Coh(X/G)) → K (X ) is canonically defined up to G,X homotopy, independent of the choice of embedding X ֒→ Z, and agrees with the comparison map perf ρ if X is smooth. ρ commutes (up to homotopy) with proper pushforwards, restriction to G,X G,X open G-subschemes, the canonical equivalence between the Borel-Moore homology of X and that of a torsor for a vector bundle over X, and restriction from G-equivariant K-theory to H-equivariant K-theory for a reductive subgroup H ⊂ G such that H is the complexification of H ∩ M. 20 Proof. The fact that the map ρ is canonical up to homotopy follows from the fact that the G,X rest of the diagram is strictly commutative and the formation of homotopy fibers is functorial in the category of spectra. So, the map between homotopy fibers of the two restriction maps j is canonical, and the d´evissage isomorphism with the homotopy fiber of j induced by the Gysin map i c,∨ perf top b an is canonical for both the theories K (D Coh(−/G)) and K ((−) ). The fact that ρ ∼ ρ G,X M G,X when X is smooth follows from considering the identity embedding X ֒→ X. Independence of the choice of embedding is essentially proved in [T5], which is an extension to the equivariant setting of [T3]. For the benefit of the reader, we explain the conceptual core of argument: Define a category Emb of “virtual embeddings” whose objects are G-quasi-projective schemes and whose morphisms X Z consist of a G-equivariant closed subscheme V ֒→ Z along with a G-equivariant map V → X which can be factored as a composition of maps which are torsors for locally free sheaves. Composition is given by pullback of closed subschemes. Then in the proofs of [T5, T3], Thomason shows that given two maps X Z and X Z , there is a linear action of 1 2 n n n n G on A and maps Z A and Z A such that the two compositions X A agree. In 1 2 sm particular, the under-category Emb is filtered, as is the category Emb of virtual embeddings X/ X/ X Z, where Z is smooth. Now let X Z be a virtual embedding into a smooth Z, which corresponds to (V ֒→ Z, π : 1 top 2 an V → X). For ease of notation let E (X) = K (Perf(X/G)) and let E (X) = K (X ) for any smooth G-quasi-projective scheme X. Then we define i i i i E (X Z) := E (X) = fib(E (Z) → E (Z − V )), i = 1, 2. perf Note that by functoriality of the homotopy fiber the comparison maps on cohomology theories ρ G,Z perf and ρ induce a map G,Z−V 1 2 ρ(X Z) : E (X Z) → E (X Z). sm i i We claim that the assignment E (X Z) can be extended to a functor E : Emb → Ho(Sp) X/ sm in which all arrows in Emb map to isomorphisms, and the comparison map ρ(X Z) defines X/ 1 2 a natural transformation of functors E → E . Indeed given a composition X Z → Z , which 1 2 sm we regard as a map Z Z in Emb , consider the following diagram, where the central square 1 2 X/ is cartesian: V . ⑦ ❅ ⑦ ❅ ⑦ ❅ π ⑦ ❅ i ⑦ ❅ ~~⑦ V V 1  p 2  p ❆ ❆ ❆ ❆ ⑦ ⑥ ❆ ❆ ⑦ ⑥ ❆ ❆ ⑦ ⑥ ❆ ⑥ ❆ ❆ ⑥ ❆ ❆ ⑥ ❆ ⑦ ❆ ⑥ ❆ ~~⑦ ~~⑥ /o /o /o /o /o /o /o /o /o // /o /o /o /o /o /o /o /o /o // X Z Z 1 2 π i i i i Then we have canonical isomorphisms E (Z ) −→ E (V ) −→ E (Z ) for i = 1, 2, where the ′ ′ 1 2 2 1 V V 2 2 construction of the latter map uses Poincar´e duality (all of the G-schemes involved are smooth) to convert cohomology to homology with locally compact supports (i.e. Borel-Moore homology), and More precisely, the proof of [T5, Proposition 5.8] shows that for any G-quasi-projective X there is a map X A for some linear represenation of G. Thus it suffices to consider the case of two maps X A , i = 1, 2. Next if V → X is a composition of torsors for locally free sheaves and V ֒→ A , i = 1, 2 are two G-equivariant closed embeddings, n n n 1 2 then the proof of [T3, Lemma 4.2] works equivariantly to constuct an equivariant embeddings A ֒→ A ×A such n n n 1 2 i that the two induced embeddings V ֒→ A × A agree. Thus it suffices to show that for any two maps X A ′ ′ n n ′ n i i corresponding to two fibrations V → X, one can compose with maps A A such that if V ⊂ A corresponds i i n ′ ′ to the compositions X A , then V ≃ V over X. This follows from the proof of [T3, Proposition 4.7], which also 1 2 works equivariantly. 21 i is an isomorphism by d´evissage for the Borel-Moore theory. We leave the somewhat involved i i diagram chase to the reader to show that the isomorphisms E (X Z ) → E (X Z ) commute 1 2 with the comparison maps ρ(X Z ) and ρ(X Z ) up to homotopy. 1 2 Our comparison map ρ arises from choosing an embedding X Z where V = X and com- G,X top b 1 bining ρ(X Z) with the canonical d´evissage isomorphisms K (D Coh(X/G)) ≃ E (X Z) c,∨ an 2 and K (X ) ≃ E (X Z). We observe that using part (5) of Lemma 2.2 and part (4) of top b Lemma 2.6, for any X Z we can compose the canonical equivalence K (D Coh(X/G)) → top b top b 1 K (D Coh(V/G)) followed with the canonical d´evissage equivalence K (D Coh(V/G)) → E (Z), c,∨ and likewise for K (−). Given a map V → Z which is a composition of torsors for vector bun- top b dles, and given a closed G-subscheme X ֒→ Z, the canonical equivalences K (D Coh(Z/G)) → top b top b top b K (D Coh(V/G)) and K (D Coh(X/G)) → K (D Coh(V| /G)) commute with the pushfor- c,∨ ward maps, and likewise for K (−). It follows that for any composition X Z Z , the 1 2 top b 1 1 composition of the canonical maps K (D Coh(X/G)) → E (X Z ) → E (X Z ) is homo- 1 2 c,dual top b 1 topic to the canonical map K (D Coh(X/G)) → E (X Z ), and the same is true for K (−). Combining these facts, we see that the comparison map ρ defined via the canonical equivalences G,X for any smooth virtual embedding X Z ρ(X Z) c,∨ top b 1 2 an K (D Coh(X/G)) ≃ E (X Z) −−−−−→ E (X Z) ≃ K (X ) sm is independed of the choice of X Z, because the category Emb is filtered. X/ c,∨ top b an Now that we have shown that the comparison map ρ : K (D Coh(X/G)) → K (X ) is G,X independent, up to homotopy, of the choice of embedding X Z, it is fairly straightforward to show that ρ commutes with restriction to an open subset and restriction of requivariance from G G,X to a reductive subgroup H ⊂ G which is the complexification of H∩M, because the corresponding perf claims hold for the comparison map ρ for smooth G-schemes. The fact that ρ commutes G,X G,X c,∨ c,∨ top b top b an an with the equivalences K (D Coh(X/G)) ≃ K (D Coh(V/G)) and K (X ) ≃ K (V ) for M M a map V → X which is a torsor for a vector bundle follows from the more general description of ρ above, because any closed embedding into a smooth G-scheme V ֒→ Z defines both a virtual G,X top b 1 embedding X Z and V Z, and the canonical equivalence K (D Coh(X/G)) ≃ E (X Z) top b top b 1 is the composition of the equivalences K (D Coh(X/G)) ≃ K (D Coh(V/G)) ≃ E (V Z) c,∨ and likewise for K (−). Finally, one shows that ρ commutes pushforward along a proper map p : X → Y by choosing G,X a closed immersion Y ֒→ Z into a smooth G-scheme Z and a projective bundle P(V) over Z such that p factors through a closed immersion X ֒→ Y . One shows that ρ commutes with pushforward perf along the projection P(V| ) → Y using d´evissage and the corresponding fact for ρ for smooth G,Z G-schemes. It therefore suffices to show that ρ commutes with closed immersions. For any closed immersion of G-schemes X ֒→ X , we can choose a closed immersion into a smooth G-scheme 1 2 X ֒→ Z. The fact that ρ commutes with pushforward along X ֒→ X follows from the fact 2 1 2 i i i that i is part of a canonical map of exact triangles from E (X ) → E (Z) → E (Z − X ) → to ∗ 1 1 i i i E (X ) → E (Z) → E (Z − X ) for both theories i = 1, 2. 2 2 Note that in Lemma 2.9 we have passed from strict presheaves on the category of algebraic G- spaces with values in spectra, to presheaves on the category of quasi-projective G-schemes with values in the homotopy category Ho(Sp). Presumably neither of these relaxations are necessary, but they simplify our discussion. c,∨ top b an Theorem 2.10. The natural map ρ : K (D Coh(X/G)) −→ K (X ) construced in (2) and G,X Lemma 2.9 is a weak equivalence. 22 perf Proof of Theorem 2.10. By the commutativity of the diagram (2) and the fact that ρ ∼ ρ G,X G,X when X is smooth, it suffices to show that ρ is a weak equivalence when X is smooth G-quasi- G,X projective scheme. Let T ⊂ G be a maximal torus which is the complexification of a compact maximal torus T = T ∩ M, and let B ⊂ G be a Borel subgroup containing T . For both G- equivariant cohomology theories, pullback along the map G × X → X is an injection which is canonically split by the pushforward map. The map which forgets from G-equivariance to T - equivariance, then restricts along the T -equivariant map {1}×X → G× X induces an equivalences K (G× X) → K (X) because topologically G× X = M × X and the same maps induces M B T B T c c top top an equivalence K (Perf(G× X/G)) → K (Perf(X/T )) because X/T → G× X/G ≃ X/B is B B perf a composition of torsors for line bundles on X/B. The comparison map ρ commutes with the G,X operations of pullback, proper pushforward, and restriction to subgroups, so ρ is compatible with top top an an the splitting of the inclusions K (Perf(X/G)) ֒→ K (Perf(X/T )) and K (X ) ֒→ K (X ), M T so it suffices to prove the claim when G = T is a torus. ′ ′ We can stratify X/T by smooth T -schemes of the form U× (T/T ), where T ⊂ T is an algebraic subgroup and T acts trivially on U. Using the localization sequences for closed immersions in Lemma 2.2 and Lemma 2.6 and the compatibility of ρ with pushforward along closed immersions G,X and restriction to open subsets, Lemma 2.9, it suffices to prove the claim for schemes of this ′ ′ form. The fact that Perf(U × (T/T )/T ) ≃ Perf(U × BT ) ≃ Perf(U), where χ ranges over ′ top ′ the group of characters of the diagonalizable group T , implies that K (Perf(U × (T/T )/T )) ≃ top ′ K (Perf(U)). There is an analogous decomposition of K (U×T/T ), and ρ ′ respects T T,U×T/T χ c this direct sum decomposition because the summands are the essential image of pullback along the ′ ′ map U × T/T → U followed by tensoring with the various characters of T . We note that when the group is trivial, our comparison map agrees with the one constructed in [B3, Proposition 4.32], therefore it is an equivalence, and the claim follows. Remark 2.11. If G is not necessarily reductive, then one can choose a decomposition G = U ⋊ H, where H is reductive and U is a connected unipotent group. As in the first step in the proof T5, Theorem 5.9], one shows that the map of stacks X/H → X/G can be factored as a of [ top sequence of torsors for vector bundles, so the canonical restriction map K (Perf(X/G)) → top K (Perf(X/H)) is an equivalence by Lemma 2.2. Combining this with the previous theorem shows that for a maximal compact subgroup M ⊂ H ⊂ G, the topologization functor is an equiva- top an lence K (Perf(X/G)) → K (X ) as presheaves of spectra on Sm , and we have a comparison M G top b c∨ an isomorphism ρ : K (D Coh(X/G)) → K (X ). G,X 2.2. The case of smooth Deligne-Mumford stacks. Here we provide an explicit computation of the periodic cyclic homology of Perf(X) for a smooth Deligne-Mumford stack of finite type over C and study its noncommutative Hodge theory when it is proper. The results of this section are likely known to experts. Given a smooth scheme U, we can consider its de Rham complex, 0 → O → Ω → ··· , a −p complex of vector spaces. We can regard this as a Λ-module Ω (U) by defining Ω (U) := Ω and • p letting B act via the de Rham differential. Even though the Λ-module structure is not O -linear, it still defines a sheaf of Λ-modules on the small site X for any smooth DM stack X. We define et the de Rham cohomology of a smooth Deligne-Mumford stack X to be the Λ-module H (X) := RΓ(X , Ω ). dR et • There are several other ways to present H (X). First note that we can equivalently restrict to dR aff the sub-site of ´etale maps U → X for which U is affine, which we denote X , because it has an et equivalent topos of sheaves, i.e. the canonical map is an equivalence aff H (X) −→ RΓ(X , Ω ). dR • et 23 aff We can consider the sheaf of Λ-modules on X given by U 7→ C (O ), the Hochschild com- • U et plex of coordinate algebra. This admits a canonical map to the presheaf of Λ-modules given by U 7→ C (Perf(U)). Likewise, for any smooth affine scheme the map HKR is a map of Λ-modules C (O ) → Ω (U) and compatible with ´etale base change, so they induce maps of presheaves on • U • aff et Lemma 2.12. The canonical maps aff aff aff RΓ(X , Ω ) ← RΓ(X , C (O )) → RΓ(X , C (Perf(−))) • • − • et et et are all equivalences of Λ-modules. Proof. These maps are all equivalences for affine U at the level of underlying complexes. The result follows formally from the fact that a map of Λ-modules is an equivalence if and only if the underlying map of complexes is an equivalence, and the forgetful functor taking a Λ-module to its underlying complex commutes with limits, hence commutes with RΓ. The following is due to Toen, and essentially follows the argument of [T6] in the case of algebraic K-theory. We will need to use both the derived inertia stack I and its underlying classical stack cl I ⊂ I . cl Proposition 2.13 (Toen, unpublished). Let X be a smooth Deligne-Mumford stack, and let I denote its classical inertia stack. There is a natural isomorphism of Λ-modules C (Perf(X)) → cl H (I ). dR The idea of the proof is to show that the formation of both complexes is local in the etale topology over the coarse moduli space of X, so one can reduce to the case of a global quotient. Thus a key observation is that the formation of the derived inertia stack I is ´etale local. Lemma 2.14. Let X → X be a map from a stack to a separated algebraic space, and let U → U be the base change along an etale map U → X. Then I ≃ I × U. U X X Proof. This can be seen for the derived inertia stack from a functor-of-points definition of I . We let U(T ) denote the ∞-groupoid of maps from T to U for a derived affine scheme T . I (T ) = U(T )× U(T ) U(T )×U(T ) ∼ Map(S ,U(T )) 1 1 ∼ Map(S ,X(T ))× 1 Map(S , U(T )) Map(S ,X(T )) So in order to show that I (T ) ≃ I (T ) × U(T ), it will suffice to show that I ≃ I × U U X X(T ) U X X in the derived sense. Consider the following diagram, in which each square is Cartesian and the vertical arrows are closed immersions: ** // // // U U × U Γ X // // U × U U × X X × X Here Γ denotes the graph of the morphism U → X. Then by definition I is the derived self intersection of the closed subspace X → X × X, so in order to prove the claim it will suffice to show that Γ× Γ is isomorphic to I as a derived scheme over U. U×X The map U → Γ is an isomorphism on underlying classical algebraic spaces, and it follows from the fact that U → U × U is an etale closed immersion of closed substacks of U × U that the induced map I → Γ× Γ induces an isomorphism on cotangent complexes as well, hence it is U U×X an isomorphism. 24 cl Proof of Proposition 2.13. The pullback functor along the projection I → X induces a map cl cl C (Perf(X)) → C (Perf(I )). For any ´etale U/I , the pullback functor induces a natural map • • X X cl C (Perf(I )) → C (Perf(U)). Thus we get a map of presheaves of Λ-modules • • cl aff cl C (Perf(X)) → RΓ((I ) , C (Perf(−))) ≃ H (I ). • • dR X et X Note that if p : X → X is the coarse moduli space of X, then the map constructed above is functorial with respect to pullback along maps U → X. We claim that C (Perf(X)), regarded as a presheaf over X, has ´etale descent. Indeed, consider any ´etale map U → X, and let U = X × U. Because the derived category of U is compactly generated [HR], we can identify C (Perf(U)) ≃ RΓ(U, Δ Δ (O )) ≃ RΓ(U, (p ) O ), • ∗ U U ∗ I where Δ : U → U× U is the diagonal, p : U → U is the base change of p, and O is the structure U I sheaf of the derived inertia stack, regarded as a finite algebra over O . In the previous lemma, we saw that the formation of I commutes with ´etale base change, so this combined with the projection formula implies that RΓ(U, p O ) ≃ RΓ(U, p (O )| ), functorially in U. The presheaf ∗ I ∗ I U U X U/X 7→ RΓ(U, p (O )| ) has ´etale descent, so U 7→ C (Perf(U)) does as well. ∗ I U • cl Thus in order to show that C (Perf(X)) → H (I ) is an equivalence, it suffices to verify this • dR after base change to an ´etale cover of X. We can find such a U → X such that U = X × U is a global quotient of a scheme by a finite group action. In that case, the result is shown in [B1, Proposition 4]. Finally after applying the Tate construction, i.e. passing to periodic cyclic homology, we can an compare this to the cohomology of |X |, the geometric realization of the underlying topological stack (in the analytic topology) associated to X [N2], as well as the cohomology of a coarse moduli space X → X. Lemma 2.15. Let X be a Noetherian separated DM stack of finite type over a Noetherian base scheme. Assume that X has finite dimension. Then X has finite ´etale cohomological dimension with Q-linear coefficients, and the functor RΓ(X ,−) commutes with filtered colimits. et Proof. We first claim that the pushforward along the projection to the coarse moduli space p : X → X is exact. Indeed this can be checked ´etale locally on X, and so we may assume that X is a global quotient U/G, where G is a finite group. One can factor p as U/G → X × BG → X – pushforward along the first is exact by [S6, Tag 03QP], and the second is exact because we are using characteristic 0 coefficients. It now suffices to prove the claim when X = X is a Noetherian separated algebraic space of finite type over a Noetherian base scheme. In this case, we can apply the induction principle of [S6, Tag 08GP] and the fact that ´etale cohomology takes elementary excision squares to homotopy cartesian squares to reduce to the case of affine schemes. In this case, the result follows from the fact that derived global sections of characteristic 0 sheaves on a Noetherian scheme in the ´etale topology argees with that in the Nisnevich topology, and the Nisnevich topology has cohomological dimension ≤ d. Finally, the implication that finite cohomological dimension implies commutation with filtered colimits in the unbounded derived category is [CD, Lemma 1.1.7]. Lemma 2.16. There are natural isomorphisms Tate ∗ an ∗ H (X) ≃ C (|X |;Q)⊗ C((u)) ≃ C (X;Q)⊗ C((u)) dR Q Q sing sing 25 Proof. The de Rham isomorphism gives a canonical isomorphism of pre-sheaves of C((u))-modules aff Tate ∗ an on X between U 7→ (Ω (U)) and U 7→ C (U ;C)((u)), so we have a canonical isomorphism et sing ∗ an aff Tate C (|Y |;C)⊗ C((u)) ≃ RΓ(X , Ω (−) ). sing et It therefore suffices to show that the Tate construction commutes with taking derived global sections for the sheaf of Λ-modules Ω . For this we observe that the functor M 7→ M commutes with Tate S homotopy limits, and hence with derived global sections, and M is the filtered colimit of M → 1 1 S S M [2] → M [4] → ··· , so its formation commutes with RΓ by the previous lemma. ∗ an ∗ an Finally, one can check that the pullback map C (Y ;Q) → C (|Y |;Q) is an equivalence sing sing an an locally in the analytic topology on Y . Locally Y is isomorphic to a global quotient of a scheme by a finite group, for which the fact is well-known. 2.3. Equivariant K-theory and periodic cyclic homology. For a dg-category, C, it is natural per top to ask if the Chern character induces an equivalence K (C)⊗C → C (C). This is referred to as the lattice conjecture in [B3], where it is conjectured to hold for all smooth and proper dg-categories. Here we observe some situations in which the lattice conjecture holds, even for categories which are not smooth and proper. Theorem 2.17 (Lattice conjecture for smooth quotient stacks). Let G be an algebraic group acting on a smooth quasi-projective scheme X. If X/G admits a semi-complete KN stratifica- top tion (Definition 1.1), then the Chern character induces an equivalence K (Perf(X/G)) ⊗ C → per C (Perf(X/G)). Lemma 2.18. Let X be a smooth Deligne-Mumford stack, and let i : Z ֒→ X be a smooth closed substack. Then the pushforward functor fits into a fiber sequence per ∗ per per C (Perf(Z)) −→ C (Perf(X)) −→ C (Perf(X− Z)). • • • Proof. This follows from Proposition 2.13, combined with Lemma 2.16 and the usual Gysin se- quence for the regular embedding of inertia stacks I ֒→ I . Z X per top Proof of Theorem 2.17. Because K (−) ⊗ C and C (−) are both additive invariants, proving that the natural transformation top per K (Perf(X/G))⊗ C → C (Perf(X/G)) is an equivalence for smooth projective-over-affine X and reductive G reduces to the case where X/G is Deligne-Mumford by Theorem 1.8. Note that the only point in the proof of Lemma 2.2 which does not immediately apply to an arbi- trary additive invariant is the localization sequence for a closed immersion. Therefore Lemma 2.18 per implies that Lemma 2.2 applies to the presheaf C (Perf(−)), because the only stacks that appear in the proof are DM. We can now imitate the proof of Theorem 2.10: Perf(X/G) is a retract of Perf(X/B), and per top Perf(X/B) → Perf(X/T ) induces an equivalence for both invariants K (•) and C (•), by Lemma 2.2. Thus it suffices to consider smooth DM stacks of the form X/T . Any such stack admits a stratification by smooth stacks of the form U × BΓ for some finite group Γ, and by Lemma 2.18 it suffices to prove the theorem for such stacks. Thus Ch⊗C is an equivalence because it is an equivalence for smooth schemes and Perf(U × BΓ) ≃ Perf(U), the sum ranging over characters of Γ. Corollary 2.19 (Lattice conjecture for smooth DM stacks). Let X be a smooth Deligne-Mumford per top stack. Then the Chern character induces an equivalence K (Perf(X))⊗ C → C (Perf(X)). All of the singular complexes we will encounter have finite dimensional total cohomology, so M((u)) ≃ M⊗ C((u)). 26 Proof. We have established a localization sequence for closed immersions of smooth DM stacks for top K (Perf(−)) in Lemma 2.2. We do not know if the localizing invariant of dg-categories HP(−) satisfies d´evissage in the sense of Lemma 2.2 for arbitrary closed immersions of stacks, but the localization sequence for closed immersions of smooth DM stacks follows from the comparison results Proposition 2.13 and Lemma 2.16 and the corresponding fact for singular cohomology. We therefore have a map of fiber sequences for any closed immersion of smooth DM stacks over C, top b top b top b // // K (D Coh(Z))⊗ C K (D Coh(X))⊗ C K (D Coh(U))⊗ C . Ch Ch Ch b b b // // HP(D Coh(Z)) HP(D Coh(X)) HP(D Coh(U)) From [LMB, Corollaire 6.1.1] every smooth DM stack of finte type admits a stratification by locally closed substacks which are quotients of a smooth affine scheme by a finite group. The corollary follows by applying Theorem 2.17 and the fiber sequence above inductively to this stratification. 2.4. Hodge structure on equivariant K-theory. We can now prove the final result of this paper, the construction of a pure Hodge structure on the equivariant K-theory. What we mean by a pure Hodge structure on a spectrum E in this case is simply a Hodge structure on the homotopy groups of that spectrum π (E), i.e. for each n a weight n Hodge structure on π (E) is a descending ∗ n filtration of π (E)⊗ C such that n+1−p π (E)⊗ C = F π (E)⊗ C⊕ F π (E)⊗ C,∀p n n ∗ Theorem 2.20. Let X be a smooth quasi-projective C-scheme, let M be a compact Lie group whose complexification G acts on X. Then if X/G admits a complete KN stratification, the Chern character isomorphism an per K (X )⊗ C → C (Perf(X/G)) combined with the noncommutative Hodge-de Rham sequence induces a pure Hodge structure of n an weight n on K (X ) with a canonical isomorphism n an n−2p gr (K (X )⊗ C) ≃ H RΓ(I ,O ), X I M X Hodge n an where I denotes the derived inertia stack of X := X/G. The Hodge filtration on K (X ) is compatible with pullback maps, and in particular it is a filtration of Rep(M)-modules. Remark 2.21. As we will see in the proof, this claim also holds for arbitrary smooth and proper DM stacks over C, without requiring that X is a global quotient. Proof. The degeneration property follows from Corollary 1.23, and we have a Chern character isomorphism from Theorem 2.10 combined with Theorem 2.17, so all we have to do is check that per the filtration on C (Perf(X/G)) coming from the HdR spectral sequence combined with the rational structure coming from the Chern character defines a weight n pure Hodge structure on top π (K (Perf(X/G))⊗Q. This claim is closed under arbitrary direct sums and summands in M , −n k so by Theorem 1.8 it suffices to prove this claim for smooth and proper DM stacks which are global quotients of a G-quasi-projective scheme by a reductive group G. cl For a smooth and proper DM stack, Proposition 2.13 gives an isomorphism H (I ) ≃ C (Perf(X)) dR • cl of Λ-modules. Note that I is itself a smooth and proper DM stack, and for any smooth DM stack Tate Y the complex H (Y) is canonically equivalent to the usual de Rham complex of [S2] tensored dR with C((u)), RΓ Y, 0 → O → Ω → ··· ⊗ C((u)). Y C 27 However, the usual Hodge filtration differs slightly from the noncommutative one. We have a canonical isomorphism n Tate l H (H (Y) ) ≃ H (Y;C) dR l≡n mod 2 p • n Tate i gr F H (H (Y) ) ≃ RΓ(Y, Ω [i− 2p])) dR nc Y Because the cyclic complex H (Y) has the degeneration property, we may commute taking coho- dR n p mology H and taking associated graded gr , so we have p • n Tate n+i−2p i gr F H (H (Y) ) ≃ H (Y, Ω ) dR nc Y l n Tate l−p Thus on each direct summand H (Y;C) of H (H (Y) ), the subquotient H (Y, Ω ) shows dR ′ ′ ′′ ′′ up in F if and only if l − p = n + p − 2p for some p ≥ p. In other words the subquotients nc ′ l−n appearing are those for which p ≥ p + . It follows that under the direct sum decomposition above we have M l−n p+ p n Tate 2 l F H (H (Y) ) ≃ F H (Y;C) dR nc classical l≡n mod 2 n Tate l Thus under the isomorphism H (H (Y) ) ≃ H (Y;Q) ⊗ C of Lemma 2.16, the dR l≡n mod 2 l l−n noncommutative Hodge filtration corresponds to the Hodge filtration on H (Y;Q)h i. l≡n mod 2 n cl Tate We claim that this rational structure on H (H (I ) ) agrees with the one induced by the dR top cl Tate equivalence K (Perf(X)) ⊗ C ≃ H (I ) of Theorem 2.17 and Lemma 2.16, so that we have dR an isomorphism of Hodge structures l − n top l cl π (K (Perf(X)))⊗ Q ≃ H (I ;Q)h i. (3) −n l≡n mod 2 th The Hodge structure on the l rational cohomology of the de Rham complex of a smooth DM stack top has weight l (see [S7]), so it would follow that π (K (Perf(X))) ⊗ Q has a Hodge structure of −n weight n. cl cl Tate For the claim about the rational structure of H (I ), note that the isomorphism H (I ) ≃ dR dR X X top K (Perf(X))⊗ C results from applying the derived global sections functor to isomorphic sheaves cl on the ´etale site of I , cl aff top cl aff per RΓ((I ) , K (Perf(−))⊗ C) ≃ RΓ((I ) , C (Perf(−))) et et • X X cl aff ∗ an ≃ RΓ((I ) , C ((−) ;C)((u))). X et sing But according to [B3, Proposition 4.32], the noncommutative Chern character for smooth C- per schemes factors through the twisted Chern character under the natural equivalence C (Perf(X)) ≃ Tate ∗ H (X) ≃ C (X;C((u))). It follows that the isomorphism above is the complexification of a dR sing cl aff map of presheaves of Q-complexes on (I ) et top ∗ an K (Perf(−))⊗ Q → C ((−) ;Q)⊗ Q(( )), sing 2πi n cl Tate which is also a level-wise weak equivalence. Thus the rational structure on H (H (I ) ) agrees dR with that of the Hodge structure of Equation 3. Remark 2.22. For any of the quotient stacks appearing in Amplification 1.25, the theorem above still holds for D Coh(X) with the same proof, with the exception of the explicit computation of per p n b gr H (C (D Coh(X))) when X is not smooth. In particular we have: 28 per an b • A canonical isomorphism K (X )⊗ C → C (D Coh(X/G)) which factors through the M • per top b b complexification of the Chern character K (D Coh(X/G)) → C (D Coh(X/G)); • The degeneration property for D Coh(X/G); and top b • A pure Hodge structure of weight n on π K (D Coh(X)) coming from the degeneration −n of the noncommutative Hodge-de Rham sequence which is Rep(G)-linear. 3. Hodge-de Rham degeneration for singularity categories In this section we extend our methods to establish the degeneration property for certain “dg- categories of singularities” MF(X/G, W ) associated to an equivariant Landau-Ginzburg model, i.e., a smooth G-variety X and a G-invariant function W : X → A . The notation MF is more frequently used for categories of matrix factorizations, which are equivalent to singularity categories for LG models on smooth schemes [O2]. We have chosen to use MF to denote singularity categories for consistency with [P]. The categories MF(X/G, W ) will be Z/2Z-graded, and we will need a suitable ∞-categorical model to work with these. The ∞-category of (essentially) small idempotent complete stable ∞- perf perf categories, Cat , admits a symmetric monoidal structure [BGT, Section 3.1]. For A,B ∈ Cat , ∞ ∞ ˙ b A⊗B is the category of compact objects in Ind(A)⊗ Ind(B), which is idempotent complete. perf For any E -algebra R, Perf(R) is canonically a commutative algebra object in Cat . We let sm ⊗ perf LinCat = (Perf(R) )-Mod(Cat ) R ∞ ⊗ sm denote the ∞-category of Perf(R) -module objects. LinCat is equivalent, via the ind-completion functor, to the ∞-category of (R-Mod) -module objects in the ∞-category of compactly generated presentable stable infinity categories with functors which preserve colimits and compact objects, ⊗ sm as in [L2, Definition 6.2]. Because (R-Mod) is a commutative algebra object, LinCat has a canonical symmetric monoidal structure [L3, Thm. 4.5.2.1]. In addition, if R → R is a map of E - ⊗ ′ ⊗ algebras, the tensor product induces a map of commutative algebra objects Perf(R) → Perf(R ) , and this induces a symmetric monoidal pullback functor [L3, Thm. 4.5.3.1] which we denote ′ sm sm (−)⊗ R : LinCat → LinCat R R ⊗ ′ ⊗ These constructions work just as well with Perf(R) and Perf(R ) replaced by any other essentially small stable idempotent complete symmetric monoidal ∞-categories. sm Any ∞-category C in LinCat is canonically enriched over R-Mod, i.e. RHom(E, F) ∈ R-Mod sm for any E, F ∈ C, via an inner-Hom construction. In fact, by [C], LinCat is equivalent to the ∞-category of categories enriched in R-module spectra. Regarding k((β)), where β is variable of homological degree −2, as an E -algebra via the forgetful functor from dg-algebras to E -algebras, ∞ ∞ sm this identifies LinCat with the ∞-category of categories enriched over dg-k((β))-modules, or k((β)) sm Z/2Z-graded dg-categories. This justifies using LinCat as our model for Z/2Z-graded dg- k((β)) categories. The main results of this section, Theorem 3.10 and Proposition 3.17, will establish a k((β))-linear version of the degeneration property for some categories of singularities on quotient stacks. 3.1. Preliminaries on categories of singularities on stacks. There have been several concrete approaches to developing the general theory of singularity categories [EP,LP,O2]. We will mostly use the perspective of [P], extended more recently in [BRTV], which is more abstract but has the advantage of allowing one to deduce results about singularity categories directly from the analogous statement for derived categories of coherent sheaves (see Lemma 3.9). We will summarize the main definitions and lemmas we will use from [P]. As elsewhere in the paper k denotes a field of characteristic zero. 29 Definition 3.1. A Landau-Ginzburg (LG) model is a pair (X, W ), where X is a smooth finite type k-stack such that the automorphism groups of its geometric points are affine and W is a morphism W : X → A . In particular, X is a QCA stack over k in the sense of [DG]. Our primary examples of interest will be quotient stacks X := X/G over k. By generic smoothness, if W is non-constant on every component of X, then W has only finitely many critical values in A . Throughout this paper we let Crit (W ) denote the component of the vanishing locus of dW ∈ Γ(X, Ω ) which lies set theoretically in X := X× 1 {0}. 0 A We now equip the bounded derived category of coherent sheaves on the zero fiber, D Coh(X ), with a k[[β]]-linear structure, where β is a variable of homological degree -2. This arises from a 1 b homotopical S -action on the category D Coh (X), in the terminology of [P], which concretely refers to a natural action of H (S ; k) ≃ Λ on the Hom-complexes of the category. The formal variable β arises via the same construction which leads to the formal variable u acting on C (A), but we use different variable names to avoid confusion between these two S -actions, especially when we discuss the k((β))-linear negative cyclic homology below. Spec(Λ) admits the structure of a derived group scheme, so the ∞-category D Coh(Λ), as well as its ind-completion IndCoh(Λ), admits a symmetric monoidal structure given by the convolution product “◦”: Given F, G ∈ D Coh(Λ), F◦G := m (F⊠G), where m : Spec(Λ)×Spec(Λ) → Spec(Λ) is the group multiplication. So the underlying complex of F ◦G is F ⊗ G, with Λ-module structure given by letting B act on homogeneous elements by |m| B (m⊗ n) := B (m)⊗ n + (−1) m⊗ B (n). F⊗G F G The following is an enrichment of standard Koszul duality results. Lemma 3.2. The functor D Coh(Λ) → Perf(k[[β]]) V 7→ V := RHom (k, V ) extends to a symmetric monoidal equivalence, leading to a symmetric monoidal equivalence ⊗ ⊗ IndCoh(Λ) = (k[[β]]-Mod) Proof. This is [P, Proposition 3.1.4], see also [BRTV, Remark 2.38, Lemma 2.39]. The proof of Lemma 3.2 relies on an elementary but important observation. Let (V, d) be a complex with a Λ-action. There is a quasi-isomorphism of complexes V = (V [[β]], d + βB). Formal completion does not commute with the formation of tensor products of complexes, but the formation of the complex (V [β], d +βB) does commute with forming tensor products of complexes. So the crux of the proof of Lemma 3.2 is the following: Lemma 3.3. The natural inclusion of complexes (V [β], d + βB) → (V [[β]], d + βB) is a quasi-isomorphism whenever (V, d) is homologically-bounded above. Proof. By definition (V [[β]], d+βB) is the inverse limit of the complexes (V [β]/(β ), d+βB). If V is homologically bounded above, then for any m the canonical map (V [β], d + βB) → (V [β]/(β ), d + βB) induces an isomorphism in homology of degree ≥ m for n ≫ 0, which implies the claim. 30 As described in [P, Construction 3.1.5], [BRTV, Remark 2.38], the stack X admits an action by the derived group scheme Spec(Λ) which defines the upper horizontal arrow in the cartesian square: // X × Spec(Λ) X (4) // X X The action (4) can be described ([P, Remark 3.1.7]) concretely by a cosimplicial, commutative dg-O algebra whose cosimplicial degree n piece is given by A := hO [B , B ,··· , B ], dB = X n X X 1 n i 0, dB = Wi, where each of the variables B have degree one. X i Definition 3.4. We define PreMF(X, W ) := D Coh(X ) with the additional k[[β]]-linear structure b ⊗ b induced by Lemma 3.2 and the D Coh(Λ) -module structure on D Coh(X ) induced by the action (4). It is useful to note that the k[[β]]-linear structure has a concrete dg-model which is described in [P] and which we now recall. Observe that O = A := (O [B ], dB = W ), where B is a variable X X X X X of degree one. Pushforward defines a canonical equivalence: b ⊗ D Coh(X ) = A-Mod(Perf(X) ), where the right hand side denotes the category of coherent A-modules. There are natural adjoint functors: ⊗ b i : A-Mod(Perf(X) ) → D Coh(X) ∗ b ⊗ i : D Coh(X) → A-Mod(Perf(X) ) given by forgetting the A-module structure and tensoring with A, respectively. Now given two dg-A-modules M, N whose underlying complex of O -modules has bounded coherent homology, the Hom-complex Hom (i M, i N) inherits a Λ-module structure given by X ∗ ∗ |φ| B : φ 7→ B ◦ φ− (−1) φ◦ B (5) X X The following proposition shows that this Λ-module structure is enough to recover Homs as A- modules: Lemma 3.5 ([P, Section 3.3]; see also [P, Proposition 3.2.1]). Given objects M, N ∈ A-Mod(Perf(X) ), S 1 we have an equivalence Hom (i M, i N) = Hom (M, N), where the S -action is given by (5). X ∗ ∗ A Proof. For any object M ∈ A-Mod(Perf(X) ), we can construct a complex of B -modules, M[B], where as usual B is a variable of degree 1 with B = 0 and the action of B is given by (B ) +B, X X M where left multiplication by B on M[B] anti-commutes with the action of (B ) . There is an X M adjunction: Hom (i M, i N) = Hom (M[B], N) X ∗ ∗ A ′ ′ |φ| ′ φ(m) 7→ φ(m + Bm ) := φ(m− B · m ) + (−1) B φ(m ) (6) X X ψ(m) := ψ(m + B · 0) ←[ ψ(m + Bm ) |φ| The Λ = k[B]-module structure, which on the left hand side is given by B : φ 7→ B ◦φ−(−1) φ◦B , X X corresponds under this isomorphism to the Λ-module structure given by B : φ(−) 7→ −φ(B · (−)). 12 ∗ More conceptually, we have that M[B] is isomorphic to i i M = A⊗ M. Both are isomorphic to M ⊕M[1] as ∗ O complexes of O -modules, and the isomorphism A ⊗ M → M[B] which intertwines the action of B is (m, m ) 7→ X O X ′ ′ (m + B m , m ). 31 We then extend M[B] to a resolution of M as an A-module by forming the complex // (M[B]⊗ (k((β))/βk[[β]]); d = d − Bβ · (−)) M (7) k M −n B,β 7→0 where β is a variable of homological degree −2, which we refer to as the Koszul-Tate resolution of M. Using Equations (6) and (7), we have that Hom (M, N) can be computed as ∼ ∼ Hom (M[B]⊗ (k((β))/βk[[β]]), N) = Hom (M[B], N)[[β]] = Hom (i M, i N) , ∗ ∗ A k A X where it is evident that the differential on the first term agrees with the differential used to compute the invariants for the S -action defined in (5). The natural k[[β]]-linear structure on the complex Hom (i M, i N) provides an explicit model X ∗ ∗ for the k[[β]]-linear structure from Definition 3.4. Note that in (7), we constructed a canonical quasi-isomorphism of A-modules M = M[B] ⊗ (k((β))/βk[[β]]), where the latter has an explicit action by k[[β]]. Under the resulting quasi-isomorphism Hom (M, N) ≃ Hom (M[B]⊗ (k((β))/βk[[β]]), N), A A k the action of β corresponds to φ(−) 7→ φ(β · (−)). The fact that composition is k[[β]]-linear fol- lows from the elementary calculation that composition Hom (i M, i N) ⊗ Hom (i N, i P) → ∗ ∗ ∗ ∗ X k X Hom (i M, i P) is Λ-linear, and Lemma 3.2. ∗ ∗ With all of this in place, we turn to defining our main objects of interest, the categories of singularities: Definition 3.6. We define the category MF(X, W ) to be MF(X, W ) := PreMF(X, W )⊗ k((β)). k[[β]] This definition is justified by the following lemma: Lemma 3.7 ([P, Proposition 3.4.1]). MF(X, W ) is a dg-enhancement of the idempotent completion of the triangulated category 0 b 0 H (D Coh(X ))/H (Perf(X )). 0 0 Proof. Let M ∈ D Coh(X ). The lemma can be easily reduced to the following assertion: M ∈ Perf(X ) iff β = 0 ⊂ Hom(M, M) for large enough n. To prove this assertion, recall that M is perfect iff it is compact in QC(X ), because X is QCA [DG, Cor. 1.4.3]. 0 0 7) to simply multi- On one hand β ∈ Hom (M, M) corresponds under the quasi-isomorphism ( plying by β . We observe that the kernel of the surjective map n −n−1 0 ker (β : M[B]⊗ (k((β))/βk[[β]]) → M[B]⊗ (k((β))/βk[[β]])) = M[B]· β ⊕···⊕ M[B]· β , is a compact A-module because the associated graded of the β-adic filtration is a direct sum of finitely many copies of the compact A-module M[B] ≃ A⊗ M. Giving a null-homotopy of β , which is equivalent to giving a null homotopy of the composition β ◦ id , is equivalent to giving a factorization of id through the subcomplex ker(β ). In this case M is a homotopy M[B]⊗(k((β))/βk[[β]]) retract of the compact A-module ker(β ) and is thus compact. Conversely, if M[B]⊗(k((β))/βk[[β]]) = ker(β ) is a compact A-module, the identity morphism factors through some ker(β ) for some n. This proves the lemma. There is another point of view on the k[[β]] linear structure which will be useful below. Let (X, W ) be an LG-model. According to [BZNP, Theorem 1.1.3], there is an equivalence of categories b ex D Coh(X ) Fun (Perf(k), Perf(X)) = 1 ⊗ Perf(A ) 32 b ⊗ It is not difficult to check that the D Coh(Λ) -module structure on the left-hand side of this ex equivalence corresponds to the natural Fun (Perf(k), Perf(k))-module structure on the right- 1 ⊗ Perf(A ) hand side. If A is a module category for some symmetric monoidal ∞-category C , and A = hA ; i ∈ Ii is a possibly infinite semiorthogonal decomposition indexed by a totally ordered set I, we say that the semiorthogonal decomposition is C -linear if C⊗A → A factors through A , in i i which case it does so uniquely up to contractible choices. ⊗ ⊗ Lemma 3.8. Let C be a symmetric monoidal stable ∞-category, and let B and A be C -module perf ⊗ ⊗ categories with B compact in C -Mod(Cat ). If A = hA ; i ∈ Ii is a C -linear semiorthogonal ex ex decomposition, then Fun (B,A ) → Fun (B,A) is a fully faithful functor, and identifying the ⊗ i ⊗ C C former with its essential image in the latter, we have a semiorthogonal decomposition ex ex Fun (B,A) = hFun (B,A ); i ∈ Ii ⊗ ⊗ i C C ⊗ ex Proof. The fact that B is compact as a C -module category allows us to commute Fun (B,−) with filtered colimits and therefore reduce to the case of a finite index set I. Then by an inductive argument it suffices to prove the claim in the case where we have a two term semiorthogonal decomposition A = hA ,A i. If we let ι : A ֒→ A denote the inclusion, and we let ι (respectively 0 1 i i ι ) denote the right (respectively left) adjoint whose existence is guaranteed by the semiorthogonal ex ex decomposition. One can check that the composition functor ι ◦(−) : Fun (B,A) → Fun (B,A ) ⊗ ⊗ 1 C C ex ex is a right adjoint to the composition functor ι ◦ (−) : Fun (B,A ) → Fun (B,A), and likewise 1 ⊗ 1 ⊗ C C ι ◦ (−) is a left adjoint to ι ◦ (−). It is also straightforward to check that the canonical maps L R ι ◦ ι ◦ (−) → id and id → ι ◦ ι ◦ (−) are equivalences, and Map(ι ◦ F, ι ◦ G) is contractible for 0 1 1 0 0 1 ex ex any functors F ∈ Fun (B,A ) and G ∈ Fun (B,A ). The claim follows. ⊗ 1 ⊗ 0 C C An immediate corollary of this is the following: 1 ⊗ Lemma 3.9. Let (X, W ) be an LG-model. Suppose that Perf(X) admits a Perf(A ) -linear semiorthog- onal decomposition hA ; i ∈ Ii. Then MF(X, W ) admits a semiorthogonal decomposition by k((β))- linear subcategories D E ex MF(X, W ) = k((β))⊗ Fun (Perf(k),A ) . 1 ⊗ k[[β]] i Perf(A ) Proof. By the previous lemma applied to A = Perf(X), we obtain a semiorthogonal decomposition D E ex PreMF(X, W ) = Fun (Perf(k),A ) . The semiorthogonal decomposition is k[[β]]-linear 1 ⊗ Perf(A ) 1 ⊗ because β acts via endo-functors of Perf(k) as a Perf(A ) -module category. Finally, the localiza- tion functor from k[[β]]-linear categories to k((β))-linear categories commutes with filtered colimits, so one gets the desired semiorthogonal decomposition of MF(X, W ) := k((β))⊗ PreMF(X, W ) k[[β]] by base changing semiorthogonal decompositions. 3.2. Motivic decompositions and degeneration for MF. In this section we prove an ana- sm log of Theorem 1.8 for the k((β))-linear dg-category MF(X/G, W ). For any A ∈ LinCat , let k((β)) sm G(A) ⊂ LinCat denote the smallest ∞-subcategory containing A that is closed under splitting k((β)) sm countable semiorthogonal decompositions in the following sense: for any C ∈ LinCat which has k((β)) a Z-indexed semiorthogonal decomposition C = hC i , C ∈ G(A) if and only if C ∈ G(A) for all i i∈Z i A. Theorem 3.10. Let G be an algebraic group. Let X be a smooth G-quasiprojective k-scheme with a semi-complete KN stratification, and let W : X/G → A be a morphism. Then there is a smooth ′ 1 projective-over-affine Deligne-Mumford stack Y with a map W : Y → A such that ′ sm MF(X/G, W ) ∈ G(MF(Y, W )) ⊂ LinCat . k((β)) 33 ′ Furthermore if Perf(Crit (W )/G) is a proper dg-category, then the pair (Y, W ) can be chosen so ′ 1 that Y is a smooth variety and W : Y → A is projective. Note that by Corollary 1.7, Perf(Crit (W )/G) is a proper dg-category if and only if the induced KN stratification on Crit (W ) is complete. Remark 3.11. There is a slightly cleaner formulation of Theorem 3.10 using k((β))-linear additive noncommutative motives, analogous to Theorem 1.8. [T1] constructs additive noncommutative motives for dg-categories which are linear over a commutative ring, and [BGT] constructs additive noncommutative motives over the sphere spectrum. The methods of [BGT] appear to apply verba- tim to construct the ∞-category of additive noncommutative motives over an arbitrary E -algebra R, such as k((β)), but in the interest of space we have formulated Theorem 3.10 to avoid developing this additional machinery. Before proving the theorem, we note the following analogues of Lemma 1.4 and Lemma 1.13: Lemma 3.12. If X is a smooth G-scheme with a KN stratification, and W : X/G → A is an sm ss LG-model, then MF(X/G, W ) ∈ G MF(X /G, W )⊕ MF(Z /L , W| ) in LinCat . i i Z /L i i i k((β)) Proof. The main semiorthogonal decomposition of [HL] extends to categories of singularities by Lemma 3.9, and hence the argument of Lemma 1.4 applies verbatim to MF(X/G, W ). Lemma 3.13. Let π : Y → X be a rational morphism of finite-type k-stacks, i.e., Rπ O ≃ O , ∗ Y X 1 b and let W : X → A be a morphism. Assume that X is smooth and π preserves D Coh. Then MF(X, W ) is a semiorthogonal factor of MF(Y, W ). ∗ 1 Proof. The functors π and π are are Perf(A )-linear, and it follows from Lemma 3.9 that the semiorthogonal decomposition of Perf(Y) in the proof of Lemma 1.13 induces a semiorthogonal decomposition of MF(Y, W ). Proof of Theorem 3.10. The proof of Theorem 1.8 mostly applies verbatim, with the following sub- stitutions: Lemma 3.12 in place of Lemma 1.4; Lemma 3.13 in place of Lemma 1.13; the properness of the dg-category Perf(Crit (W )/G) in place of the properness of the dg-category Perf(U/G); and the result Proposition 3.14 below in place of [BLS, Theorem 6.6] to reduce from the case of a projective-over-affine DM stack to a quasi-projective scheme. The only parts of the proof of Theorem 1.8 which require modification in the current context are those which have to do with the properness of the dg-category Perf(Crit (W )/G), and this only affects two cases of the inductive proof. We thus re-write these cases, indicating the necessary modifications: ss ss s ss s ss Case X 6= ∅, X 6= X , and codim(X \ X , X ) ≥ 2: ss It suffices by the inductive hypothesis to prove the claim for (X /G, W ). We apply the inductive ss K6]: Let Y ⊂ X be the locus of points whose stabilizer has partial resolution procedure of [ ′ ss maximal dimension. Then Y is a smooth closed subvariety, and the blow up X := Bl (X ) has ′ ss a KN stratification induced by a relatively ample bundle such that (X ) has lower-dimensional stabilizers. Consider a point x ∈ Y , let R ⊂ G be the (reductive) stabilizer subgroup. The G-invariance of W 1 R 1 ss G 1 R implies that (dW ) ∈ (Ω ) ⊂ Ω . Because (X ) ⊂ Y the restriction map (Ω ) → ss ss ss X ,x X ,x X ,x 1 1 R 1 Ω is injective, and hence the pullback map (Ω ) → Ω is injective for any point y in the ss ′ Y,x X ,x X ,y ′ ss ss −1 fiber of x under p : X = Bl X → X . It follows that for any y ∈ p (x), (dW ) = 0 if and only Y x −1 ′ ′ if (d(W| )) = 0, and therefore Crit (W| ) = p (Crit (W )). X y 0 X 0 34 ′ Now by Lemma 3.13 and Lemma 3.12, if Z are the centers of the KN stratification of X , then ss ′ ss MF(X /G, W ) ∈ G MF((X ) /G, W| )⊕ MF(Z /L , W| ) . X i i Z /L i i Note that if Perf(Crit (W )/G) is a proper dg-category, then so is Perf(Crit (W| ′)/G), using 0 0 X criterion (2) of Lemma 1.5 and the fact that Crit (W| ) → Crit (W ) is proper. It follows 0 X 0 from Lemma 1.6 that in this case Perf(Crit (W| ′ ss)/G) and Perf(Crit (W| )/L ) are proper 0 (X ) 0 Z i dg-categories as well. The claim holds for the categories MF(Z /L , W| ) by the inductive hy- i i Z /L i i ′ ss pothesis, so it suffices to show the claim for MF((X ) /G, W| ). Because the dimension of the ′ ss ss stabilizer groups of (X ) are strictly smaller than that of the stabilizer groups in X , we can re- ss ′ ss ss s place X with (X ) and iterate this construction until X = X , which is handled in a previous case. ss ss s ss Case X 6= ∅ but codim(X \ X , X ) ≤ 1: Here the only modification needed to the proof of Theorem 1.8 is the observation that Crit (W| ) = 0 U×Y Crit (W| )× Y , so Perf(Crit (W| )) is a proper dg-category if Perf(Crit (W| )/G) is. 0 U 0 U×Y 0 U In the proof of Theorem 3.10, we used the following proposition to reduce from the case of a DM stack to the case of a variety. The proof is an adaptation of the methods of [BLS] to categories of matrix factorizations. Proposition 3.14. Let W : X → A be an LG-model such that X is a smooth quasi-projective (in the sense of [K]) DM stack and Crit (W ) is proper. Then there is a smooth variety Y and a ′ 1 ′ sm projective morphism W : Y → A such that MF(X, W ) ∈ G(MF(Y, W )) ⊂ LinCat . k((β)) Proof. As observed in [K], X = X/G for some reductive G and G-equivariant locally closed em- N s bedding X ֒→ (P ) for some linearized action of G on a projective space. Considering the graph 1 N 1 s of the map W : X → A gives a G-equivariant locally closed embedding X ֒→ (P × A ) , where G acts trivially on the A factor. Using Kirwan’s resolution algorithm [K6], as in [K], we modify N 1 ss N 1 s (P ×A ) by a series of smooth blow ups away from (P ×A ) such that the resulting semistable locus is a DM stack Y which is smooth and projective over A . Thus taking the closure of X in Y ¯ ¯ and resolving singularities, we can produce a Nagata compactification W : X → A of the original 1 1 morphism W : X → A such that X is projective over A and smooth over the ground field. Because X and X are DM and the critical locus Crit (W ) is proper, it follows that Crit (W ) is a 0 0 ¯ ¯ ¯ union of connected components of Crit (W ). Therefore, the category MF(X, W ) splits as a direct sum of the subcategory consisting of objects supported on Crit (W ) and the subcategory consisting of objects supported on other components of Crit (W) [P, Prop. 4.1.6]. So it suffices to prove the ¯ ¯ claim for MF(X, W ), i.e. we may assume that the potential itself W : X → A is projective rather than assuming that just Crit (W ) is proper. First we reduce to the case where X has generically trivial stabilizer. Because X is a global quotient stack, we may find a vector bundle V over X on which the automorphism groups act faithfully. Then P(V ⊕O ) → X is a rational morphism, so by Lemma 3.13 the claim for P(V ⊕O ) X X implies the claim for X, and the generic stabilizer of the former is trivial. Furthermore, the function W restricted to P(V ⊕ O ) will still be proper. Next we consider Bergh’s destackification, constructed in [B2] X , ⑥ ❅ ⑥ ❅ ⑥ ❅ ~~⑥ X X 35 ′ where X is the coarse moduli space and is smooth, the morphism π is a composition of root stacks along smooth divisors and f is a composition of root stacks along smooth divisors and blow ups along smooth centers. Note that the morphism f is rational, so by Lemma 3.13 it suffices to prove ′ ′ 1 ′ the claim for the composition W : X → X → A . Note that W will still be proper. ′ ′ ′ ′ ′ 1 Finally, because X is the coarse space of X , the function W descends uniquely to W : X → A , ′ ′ and this map is still proper. Because π : X → X is a composition of root stacks along smooth divisors, it suffices to prove the following claim: if Y is a smooth DM stack and X → Y is a root stack along a smooth divisor in Y and W : Y → A is a proper map, then the claim of the proposition for Y implies the claim for X. Let D ֒→ Y be the Cartier divisor used to form the root construction, and consider the diagram // D× Bµ X . D π // D Y Then [IU, Proposition 6.1] shows that we have a semiorthogonal decomposition ∗ r−1 ∗ ∗ Perf(X) = i π (Perf(D))⊗ M , . . . , i π (Perf(D))⊗ M, π (Perf(Y)) , ∗ ∗ D D where M is the universal invertible sheaf coming from the root stack construction. Given a map 1 1 W : Y → A and a module F ∈ Perf(A ), it is evident from the description of this semiorthogonal ∗ ∗ decomposition that each subcategory is preserved by the functor π W (F) ⊗ (−). If follows that 1 ⊗ this is a semiorthogonal decomposition of Perf(A ) -module categories, and thus by Lemma 3.9 there are induced semiorthogonal decompositions ∗ r−1 ∗ ∗ Perf(X ) = i π (Perf(D ))⊗ M , . . . , i π (Perf(D ))⊗ M, π (Perf(Y )) 0 ∗ 0 ∗ 0 0 D D 0 0 D E b ∗ b r−1 ∗ b ∗ b D Coh(X ) = i π (D Coh(D ))⊗ M , . . . , i π (D Coh(D ))⊗ M, π (D Coh(Y )) , 0 ∗ 0 ∗ 0 0 D D 0 0 where the subscript 0 refers to the derived zero fiber of W and its restrictions to D, X, and D×Bµ . Thus we have a semiorthogonal decomposition of k((β))-linear categories MF(X, W ◦ π) ≃ hMF(D, W| ), . . . , MF(D, W| ), MF(Y, W )i D D and the claim of the proposition for X follows from the claim for Y and induction, because D is smooth of one lower dimension. sm 3.2.1. The degeneration property for quotient stacks. For any A ∈ LinCat , we may form k((β)) k((β)) the k((β))-linear Hochschild complex C (A), which is a module over the CDGA Λ((β)) = k((β)) k((β))[B]/(B ). C (A) is computed by a Barr complex k((β)) C (A) := Hom(o , o )⊗ Hom(o , o )⊗ ···⊗ Hom(o , o ), (8) • 1 2 k((β)) 2 3 k((β)) k((β)) n 1 o ,o ,··· ,o 1 2 n where o are objects of A, the differential is given by the usual formula for the Hochschild complex of a k-linear dg-category [M], and B acts by the Connes differential. Using the Λ((β))-module k((β)),(n) k((β)),− k((β)),per structure, we may form the associated complexes C (A), C (A), and C (A) just • • • as in the k-linear case (see Section 1.3). Definition 3.15. A k((β))-linear dg-category A is said to have the k((β))-linear degeneration prop- k((β)),(n) erty if H (C (A)) is a flat k[u]/(u )-module for all n ≥ 1. ∗ • sm Lemma 3.16. Let A ∈ LinCat satisfy the k((β))-linear degeneration property, then any other k((β)) category C ∈ G(A) satisfies the k((β))-linear degeneration property. 36 Proof. The proof of Lemma 1.22 applies verbatim to show that if C = hC i is a Z-indexed semiorthog- onal decomposition of k((β))-linear dg-categories, then C satisfies the degeneration property if and sm only if each of the C do. This implies that the full ∞-subcategory of LinCat consisting of cat- k((β)) egories that satisfy the k((β))-linear degeneration property contains A and is closed under splitting countable semiorthogonal decompositions, so it contains G(A) by definition. Our main degeneration result for categories of matrix factorizations is the following: Proposition 3.17. Let W : X/G → A be an LG-model, where X is a smooth G-quasi-projective scheme which admits a semi-complete KN stratification. If Perf(Crit (W )/G) is a proper dg- category, then the k((β))-linear degeneration property holds for MF(X/G, W ). Proof. By Theorem 3.10 and Lemma 3.16, this reduces to the k((β))-linear degeneration property for MF(Y, W ) where Y is a smooth scheme and W : Y → A is a projective map. This amounts to the degeneration of the W -twisted Hodge-de Rham complex by [E, Theorem 1.3], which is established in [OV, Theorem 4.22]. Example 3.18. The hypotheses of the proposition are satisfied if X is projective-over-affine and Γ(X,O ) is finite dimensional – see Example 1.10. Crit (W ) We also have the following, which was left as an assumption in the original version of this paper: Corollary 3.19. If X is a smooth quasi-projective DM stack and W : X → A is a map such that Crit (W ) is proper, then the k((β))-linear degeneration property holds for MF(X, W ). Proof. Proposition 3.14 reduces this to the case of a projective morphism W : Y → A , with Y smooth, which as noted above follows from [OV, Theorem 4.22]. k((β)) Remark 3.20. We briefly discuss more concrete descriptions of C (MF(X, W )) when X is a smooth separated Deligne-Mumford stack. Namely, for any affine U with an ´etale map U → X, i 1 let Ω (U, W| ) denote the Tate construction on Ω (U)[i] with respect to the S -action given by • U −dW∧, i.e. Ω (U, W| ) is the Λ((β))-module Ω ((β))[i] with differential −β·dW∧. Letting B act • U on Ω (U, W| ) via the de Rham differential as usual, Ω (−, W ) defines a sheaf of Λ((β))-modules • U • aff on the small ´etale site X . We define the global de Rham complex to be the Λ((β))-module et aff Ω (X, W ) := RΓ X , Ω (−, W ) . • • et Assume for simplicity that Crit (W ) = Crit(W ). Then by combining the approach of Proposition 2.13 with [P, Theorem 8.2.6], it is not difficult to show that there is a natural isomorphism of Λ((β))- modules k((β)) cl C (MF(X, W )) ≃ Ω (I , W ), • • k((β)),per cl Tate which induces an equivalence C (MF(X, W )) ≃ Ω (I , W ) . Thus, Corollary 3.19 is equiv- • • alent to the statement that the k((β)) ⊗ Λ-module Ω (X, W ) has the degeneration property. This constitutes a slight generalization of the degeneration results of [OV] to DM stacks. 3.3. Graded Landau-Ginzburg models. We will use these results to establish a large class of examples of k-linear dg-categories for which the usual k-linear degeneration property holds. Definition 3.21. A graded LG-model is a map W : X → A /G , where X is a smooth algebraic k-stack whose automorphism groups at geometric points are affine, and G acts on A with weight one. Note that the data of a graded LG-model is equivalent to specifying an invertible sheaf L on X/G, which is classified by the composition X → A /G → BG , and a section W ∈ Γ(X/G, L). m m ′ ′ ∨ Denote by φ : X → X the G -torsor over X associated to L, i.e., X = Tot (L ) \ 0. To any m X 37 ′ ∗ ′ 1 graded LG-model, we use the term associated LG-model to denote the pair (X , φ W : X → A ). We will see in Proposition 3.24 that, in a precise sense, the graded LG-model is a refinement of its associated LG-model. In the setting of graded LG-models, for F ∈ D Coh(X ), there is a distinguished triangle: −1 ∗ F ⊗ L [1] → i i F → F giving rise to a natural transformation β : −⊗L[−2] → id which in the non-graded case, where L ≃ O , is the natural transformation id[−2] → id induced by the k[[β]]-linear structure of Definition 3.4. To make this natural transformation more explicit, we introduce an analog of the Koszul-Tate resolution (7), KT(M) = M, and construct a very concrete natural transformation β : KT(−)⊗ L[−2] → KT(−). −1 First consider the dg-algebra A := O ⊕ L · B where B is a formal variable of degree 1, X X X and dB = W , i.e., the differential is trivial on O and acts on the second summand by the X X −1 map L → O that is the defining section W . An A-module consists of an O -module M along X X −1 with a map B : L ⊗ M → M[−1] of O modules satisfying the Leibnitz rule dB (s ⊗ m) = X X X −1 Wsm− B (s⊗ dm), where s is a local section of L and m is a homogeneous local section of M. The ∞-category of A-modules is equivalent to the ∞-category D Coh(X ). Given a module M over A, we can form the analogue of M[B] above, which we denote −1 M := M ⊕ (L ⊗ M · B), −1 M −1 where the differential is just d ⊕ L ⊗ d . We have two operators B : L ⊗ M → M[−1], M M −1 −1 defined by the A-module structure of M, and B : L ⊗ M → L ⊗ M, which is just the identity. Using these two operators we can form a canonical action of A on M, where the action of B on M −1 M the first component is just B + B, and the action on the second component is L ⊗ B . X X −r −r+1 −r c c c Let Q(r) : L ⊗ M → L ⊗ M be the map which sends the first component of L ⊗ M to −r+1 the second component of L ⊗ M by the identity. Then we define the Koszul-Tate resolution as   M X −r   KT(M) := L ⊗ M[−2r]; d = d + Q(r) . r≥0 r>0 After pulling back to X , this complex can be identified with M[B] ⊗ (k((β))/βk[[β]]), so it follows that the canonical map KT(M) → M, which annihilates all r > 0 summands and B, is a quasi- isomorphism. We now define the natural map β : KT(M)⊗ L[−2] → KT(M) st to be the map of A-modules which annihilates the r = 0 summand and identifies the (r + 1) th summand of KT(M)⊗ L[−2] with the r summand of KT(M). Definition 3.22. We define the graded singularity category D Sing(X, W ) to be the idempotent completion of the dg-category with the same objects as D Coh(X ) but with morphisms between M, N given by −p Hom b (M, N) := hocolim Hom b (M, N ⊗ L )[2p], D Sing(X,W ) D Coh(X ) −p where the homotopy colimit is formed with respect to the natural maps β : L ⊗ N[2p] → −p−1 L ⊗ N[2p + 2]. Note also that given an LG-model (X, W ), we can forget the data of the trivialization of L to obtain a graded LG-model. This will correspond to forgetting the k((β))-linear structure on the category MF(X,W ). 38 Example 3.23. Let Y be a smooth variety over k, E a vector bundle over Y and let s ∈ Γ(E) be a regular section. We have an action of G on Tot(E ) by scaling in the fibers. The function ∨ 1 s therefore determines a mapping W : X = Tot(E )/G → A /G . The main theorem of [I] s m m b ∨ b −1 gives an equivalence of (Z-graded) dg-categories D Sing(Tot(E )/G , W ) ≃ D Coh(s (0)). The m s construction of this equivalence works equally well when Y has an action of a linear algebraic group G, E is a G-equivariant locally free sheaf, and s is G-invariant (see for instance [H3]). This gives an equivalence of k-linear dg-categories b ∨ b −1 D Sing(Tot(E )/(G× G ), W ) ≃ D Coh(s (0)/G). m s ′ ∨ 1 In this case the associated LG-model is simply the map W : Tot(E )/G → A obtained by forget- ting the G action. For a Z-graded k-linear dg-category C, we may tensor with k((β)), thereby collapsing the grading on Hom(E, F) to a Z/2Z-grading. The following proposition describes the relationship between a graded LG-model and its associated LG-model. 1 ′ ′ 1 Proposition 3.24. Let W : X → A /G be a graded LG-model, and let W : X → A be the associated LG-model. Then we have a canonical equivalence of Z/2Z-graded dg-categories b ′ ′ D Sing(X, W )⊗ k((β)) ≃ MF(X , W ). Lemma 3.25. Let π : X → Y be a smooth affine morphism of QCA stacks. Then objects of the form ∗ b b b π F for some F ∈ D Coh(Y) split generate D Coh(X), i.e. D Coh(X) is the smallest subcategory containing these objects which is closed under shifts, cones, and retracts. Proof. First note that the analogous claim holds for perfect stacks using Perf instead of D Coh and assuming only that π is affine. Indeed, the fact that the pushforward functor π : QC(X) → QC(Y) is conservative implies that objects of the form π F with F ∈ Perf(Y) split generate QC(X). In order to conclude the same for D Coh, we must imitate this argument for the categories IndCoh, i.e., show that π is conservative. The pushforward π again has a left IndCoh,∗ IndCoh,∗ ∗ b adjoint π which preserves D Coh and agrees with the usual pullback functor there. Because IndCoh π satisfies base change with respect to the shriek pullback [G, 5.2.5], and IndCoh satisfies IndCoh,∗ fppf descent with respect to shriek pullback, it suffices to show this when Y = Y is an affine derived scheme and hence X = X is as well. In this case [G, Proposition 4.5.3] shows that the essential image of the functor QC(X)⊗ IndCoh(Y) → IndCoh(X) generates the latter category QC(Y) (in fact this functor is an equivalence). It follows that IndCoh(X) is generated by objects of the IndCoh,∗ b form E ⊗ π (F) for E ∈ Perf(X) and F ∈ D Coh(Y). Furthermore, because X is affine the category Perf(X) is split generated by O , so IndCoh(X) is generated by objects of the form IndCoh,∗ b π (F) with F ∈ D Coh(Y). It follows by adjunction that π is conservative. IndCoh,∗ ′ b Proof of Proposition 3.24. Pullback along φ : X → X defines a functor of dg-categories D Coh(X ) → b ′ D Coh(X ) which intertwines the action of β and β . Making use of the canonical trivialization 0 L| ′ L| ′ ≃ O ′, we get a canonical map X X ∗ −p ∗ ∗ φ : hocolim Hom b (M, N ⊗ L )[2p] → hocolim Hom b ′ (φ (M), φ (N))[2p]. p p D Coh(X ) D Coh(X ) The latter can be identified with ∗ ∗ ∗ ∗ Hom ′ (φ (M), φ (N))⊗ k((β)) Hom ′ (φ (M), φ (N)). PreMF(X ,W ) k[[β]] MF(X ,W ) b b Because D Sing(X, W ) is generated by objects of D Coh(X ), this extends to a dg-functor on ∗ b ′ ′ idempotent completions φ : D Sing(X, W ) → MF(X , W ). From the universal property of the base-change category, φ admits an essentially unique k((β))-linear extension ∗ b ′ φ : D Sing(X, W )⊗ k((β)) → MF(X , W ), k((β)) 39 and we will show that this functor is an isomorphism. ∗ b b Concretely, φ maps an object F ∈ D Coh(X ), regarded as a generator for D Sing(X, W )⊗ k((β)), 0 k k((β)) ∗ b ′ ′ ′ to the object φ (F) ∈ D Coh(X ), regarded as a generator of MF(X , W ). Lemma 3.25 implies ∗ b ′ that objects of the form φ (F) with F ∈ D Coh(X ) generate MF(X , W ), so it suffices to show ∗ b that φ is fully faithful on the full subcategory of D Sing(X, W )⊗ k((β)) spanned by objects of k((β)) D Coh(X ). For M, N ∈ D Coh(X ), we consider q−p Hom b (M, N)⊗ k((β)) := hocolim Hom b (M, N ⊗ L )[2(p− q)][2q]. p−q D Sing(X,W ) D Coh(X ) Commuting colimits and reshuffling indices, this is isomorphic to q −p hocolim Hom (M, L ⊗ N ⊗ L )[2p] = b D Coh(X ) In this presentation, the operator β corresponds to the isomorphism between Hom b (M, L ⊗ D Coh(X ) −p q+1 −(p+1) N ⊗ L ) and Hom (M, L ⊗ N ⊗ L ). The Hom-complex is in turn isomorphic to D Coh(X ) −p hocolim Hom (M, φ (O ′ )⊗ N ⊗ L )[2p], p QC(X ) ∗ X where this last isomorphism uses the identification φ (O ′ ) ≃ L . It also uses the fact that X n∈Z −p M is coherent and φ (O )⊗ N ⊗ L is homologically bounded above, so that we may commute Hom (M,−) with this infinite direct sum. Using the projection formula and the adjunction QC(X ) between φ and φ , we finally obtain a natural equivalence ∗ ∗ ∗ −p = hocolim Hom b ′ (φ (M), φ (N)⊗ φ (L) )[2p] p D Coh(X ) the multiplication by β on Hom b (M, N) ⊗ k((β)) now corresponds to the canonical iso- D Sing(X,W ) morphisms: ∗ ∗ ∗ −p ∗ ∗ ∗ −(p+1) Hom b ′ (φ (M), φ (N)⊗ φ (L) ) → Hom b ′ (φ (M), φ (N)⊗ φ (L) ) D Coh(X ) D Coh(X ) 0 0 ∗ −1 which arise from the canonical trivialization O ′ → φ (L ). This operator is identified with the operator β in Definition 3.4 and therefore we have naturally identified the Hom-complex in b ∗ ∗ D Sing(X, W )⊗ k((β)) with Hom ′ ′ (φ (M), φ (N)) as required. MF(X ,W ) We can now establish our main result on the degeneration property for graded singularity cate- gories: Proposition 3.26. Let W : X/G → A /G be a graded LG-model with associated LG-model ′ ′ 1 ′ ′ W : X /G → A . If X admits a semi-complete KN stratification and Perf(Crit (W )/G) is a proper dg-category, then D Sing(X/G, W ) satisfies the k-linear degeneration property. Lemma 3.27. Let C be a k-linear (i.e., Z-graded) dg-category. Then the degeneration property for C is equivalent to the k((β))-linear degeneration property for C⊗ k((β)). k((β)) Proof. When D = C⊗k((β)), then the Barr complex (8) computing C (D) is quasi-isomorphic to the subcomplex in which all of the objects o lie in the generating set of objects of D of the form E ⊗ k((β)), with E ∈ C. It follows that k((β)) C (D) C (C)⊗ k((β)) • • canonically as dg-Λ((β))-modules. We therefore have that k((β)),(n) (n) C (D) C (C)⊗ k((β)) • • 40 L on the level of chain complexes as well. The result follows since as a complex k((β)) k[2n], n∈Z (n) (n) so the homology H (C (C)⊗ k((β))) = H (C (C))⊗ k((β)) will be flat over k[u]/u if and only ∗ • ∗ • (n) if H (C (C)) is flat over the same ring. ∗ • Proof of Proposition 3.26. By Lemma 3.27 and Proposition 3.24, the degeneration property for b ′ ′ D Sing(X/G, W ) is equivalent to the k((β))-linear degeneration property for MF(X /G, W ), so the result follows from Proposition 3.17. Remark 3.28. Using Example 3.23, Proposition 3.26 implies noncommutative Hodge-de Rham −1 degeneration for the derived category of certain complete intersections s (0)/G, where s is a G- invariant section of a locally free sheaf E on a smooth G-scheme Y . In fact, the hypotheses of ∨ 1 Amplification 1.25 imply that in the associated LG-model W : Tot(E )/G → A , Crit (W ) admits a complete KN-stratification, and thus Perf(Crit (W )/G) is a proper dg-category by Corollary 1.7. Proposition 3.26 therefore provides an alternate proof of Corollary 1.27. 4. Computations of Hochschild invariants 4.1. Generalities on Hochschild invariants. In this section we identify the Hochschild homol- ogy with functions on the derived inertia stack (or loop stack) X× X, which we denote by I , X×X and we give an explicit description when X is a quotient stack. Proposition 4.1. Let X be a smooth algebraic k-stack which is perfect, i.e., X is quasi-compact with affine diagonal, and QC(X) = Ind(Perf(X)), and let Δ : X → X× X be the diagonal. Then we have an identification C (Perf(X)) RΓ(I ,O ). • X I Proof. Morita theory for perfect stacks [BZFN] identifies the identity functor with Δ O in the ∗ X category QCoh(X× X). To compute C , we use the Morita invariant definition of the Hochschild homology of a compactly generated dg-category as the trace of the the identity functor. Thus we must compute the trace tr : QCoh(X× X) → QCoh(Speck). ∗ ∗ On sheaves of the form π (P )⊗ π (P ), with P , P ∈ Perf(X), we have that the trace is given by 1 2 1 2 1 2 ∗ ∗ tr(π (P )⊗ π (P )) := RHom(Hom(P ,O ), P ) 1 2 1 X 2 1 2 ∗ ∗ ∗ = RΓ(X, Δ (π (P )⊗ π (P ))). 1 2 1 2 Since the category QC(X× X) is the colimit completion of sheaves of this form, we have that for an arbitrary object F ∈ QC(X× X), the trace can be computed by F → RΓ(X, Δ (F)). (9) It follows that we have an isomorphism C (Perf(X)) = RΓ(X, Δ Δ O ). • ∗ X Now let X/G be a global quotient stack. We consider the scheme P := G× X × X. Denote by Δ : G× X → P the map (g, x) 7→ (g, x, x), and by Γ : G× X → P the map (g, x) 7→ (g, x, g · x). Both are closed immersions, and we will also use the notation Δ and Γ to denote the corresponding subschemes of P. Lemma 4.2. Let X/G be a smooth quotient stack. Both Γ and Δ are equivariant with respect to −1 the G action on P which sends h· (g, x , x ) → (hgh , hx , hx ). We have 1 2 1 2 L G C (Perf(X/G)) = RΓ(X,O ¯ ⊗ O ) . • Γ 41 Proof. By Proposition 4.1 we must compute the derived global sections of the structure sheaf of the derived inertia stack. First note the alternate presentation for the stack X/G ≃ G × X/G , where the G action in the second presentation is given by −1 (h , h )· (g, x) = (h gh , h x). 1 2 2 1 In this presentation the diagonal X/G → X/G × X/G corresponds to the G -equivariant map G× X → X × X given by (g, x) 7→ (x, gx). Let G act on P by −1 (h , h )· (g, x , x ) = (h gh , h x , h x ). 1 2 1 2 2 1 1 2 2 Then Γ is G -equivariant, and using the presentation above we see that the diagonal factors as the 2 2 2 2 closed immersion Γ : G× X/G → P/G followed by the projection P/G → X × X/G , which is −1 smooth and affine. It follows that the derived inertia stack is the derived intersection of p Γ and −1 2 p Γ in P× 2 P/G . Now P× P ≃ G× G× X × X with G -action given by −1 −1 (h , h )· (g , g , x , x ) = (h g h , h g h , h x , h x ) 1 2 1 2 1 2 2 1 2 2 1 1 2 2 1 1 The projections p , p : P× P → P are given by forgetting g and g respectively. We claim that 1 2 2 1 P× P/G ≃ P/G, where G acts on P as in the statement of the lemma. Indeed we can present P/G as the quotient of G× P by the G -action −1 −1 (h , h )· (g , g , x , x ) = (h g h , h g h , h x , h x ), 1 2 1 2 1 2 2 1 1 2 1 1 1 2 1 1 and we have a G -equivariant isomorphism G× P → P× 2 P given by (g , g , x , x ) 7→ (g , g g , x , g x ) 1 2 1 2 1 1 2 1 1 2 The resulting isomorphism P/G → P× 2P/G is given by the map (g, x , x ) 7→ (1, g, x , x ) which X 1 2 1 2 is equivariant with respect to the diagonal homomorphism G → G . −1 −1 2 2 In order to complete the proof, we must identify the closed substacks p (Γ/G ) and p (Γ/G ) 1 2 −1 in P× 2P/G under the isomorphism with P/G. The first is the closed subscheme p Γ∩({1}×P) = −1 Δ, regarded as a G-equivariant closed subscheme of P, and the second is p (Γ)∩ ({1} × P) = Γ as a G-equivariant closed subscheme of P. The case when X is a vector space, V, and G acts on V via a linear action, is of interest in two-dimensional gauge theory. In this case we make the above derived intersection explicit using a Koszul resolution. Denote by α : G× V → V the action morphism (g, v) 7→ g · v. We choose linear coordinates on V and identify V× V with Spec(k[x , y ]). i i The Koszul complex for the regular sequence K (x − y ) gives a resolution of the diagonal V×V i i on V× V. An important point is that, in this case, this resolution is G-equivariant with respect to the diagonal G-action because the G-action on V is linear. Then K (x − y ) → O G×V×V i i ¯ is a resolution of O over P. ∗ G Corollary 4.3. C (Perf(V/G)) (K (x − α (x ))) • G×V i i Proof. By the above lemma, C (Perf(V/G)) is isomorphic to G ∗ G (K (x − y )⊗ O ) ≃ (Γ K (x − y )) . G×V×V i i Γ G×V×V i i ∗ ∗ ∗ Pulled back to G × V, the function Γ (y ) = α (x ), hence Γ K (x − y ) = K (x − i i G×V×V i i G×V i α (x )). Remark 4.4. We expect that one could describe similar models for the Hochschild homology of general gauged linear sigma models (V/G, W ). The key step needed to do this would be to P, Theorem 4.2.3] to quotient (or more generally QCA) stacks. generalize [ 42 4.2. An HKR theorem and quotients of affine varieties. In [BG], Block and Getzler construct for any compact smooth M-manifold X an explicit model for the M-equivariant cyclic homology of the algebra C (X) using differential forms on X. Our goal is to translate their construction into algebraic geometry and establish their version of the equivariant Hochschild-Kostant-Rosenberg theorem when X = Spec(A) is smooth and affine, G is a reductive group, and X/G is formally proper, which is equivalent to the condition that A is finite dimensional over k. Our proof is an application of Theorem 1.8 and Theorem 2.17. For simplicity, we let k = C throughout this section. To compute the derived intersection appearing in Lemma 4.2, we may use the bar resolution B(A) of A as an A− A bimodule. Namely, ⊗n B (A) := A⊗ A ⊗ A where the differential can be described as the sum b = Σ (−1) ∂ , where i i ′ ′′ a a ⊗··· a ⊗ a , i = 0 1 n 0 0 ′ ′′ ′ ′′ ∂ (a ⊗ a ⊗··· a ⊗ a ) := a ⊗···⊗ a a ⊗···⊗ a , i 6= 0,6= n i 1 n i i+1 0 0 0 0 ′ ′′ a ⊗···⊗ a a , i = n 0 0 Our notation is meant to highlight the fact that the first and last variables in the bar complex play a distinguished role from the other a . We then have that O ⊗ B(A) is a resolution of O which i G we may restrict to Γ. The result is a complex where the n-th graded piece is ⊗n+1 n+1 C (A, G) := O ⊗ A = Γ(G× X ,O n+1). n G G×X For any c ∈ Γ(O ), the differentials ∂ above now take the form n+1 G×X  c(g, x , x , x ,··· , x ) i = 0 0 0 1 n−1 ∂ c(g, x , x ,··· , x ) := c(g, x ,··· , x , x ,··· , x ), i 6= 0,6= n i 0 1 n−1 0 i i n−1 c(g, x ,··· , x , g · x ), i = n 0 n−1 0 We define the Λ-module C (A) := C (A, G) , and note the following corollary of Lemma 4.2. •,G • Corollary 4.5. C (Perf(X/G)) = C (A) • •,G Let C (A) denote the completion of C (A) as a complex of modules over the representation •,G •,G ring Rep(G) ⊗ C = Γ(O ) at the conjugacy class [g]. Let us work for the moment with a fixed normal element g ∈ G. Let Y = Spec(A) denote the fixed point locus of g and B = Γ(O ). The letter Z will designate the centralizer of g and z denotes its Lie algebra, and normality of g ensures that Z is the complexification of Z := Z ∩ M for a maximal compact subgroup M ⊂ G. We have embeddings j : Z → G and k : Y → X. Lemma 4.6. When Spec(A)/G is formally proper, the natural restriction map gives rise to an ∗ ∧ ∧ isomorphism k : C (A) → C (B) . •,G •,Z g g Proof. Note that because Spec(A)/G is formally proper, each C (A) is a coherent Rep(G) ⊗ n,G C-module, so completion commutes with taking homology in this case, and it suffices to prove the result on the level of homology. It is known [FHT2, Proposition 3.10] that the map k : ∗ an ∧ an ∧ K (X ,C) → K (Y ,C) is an isomorphism. The comparison maps M g c g per ∗ an // K (X ,C) H (C (A)) •,G ∗ ∗ k k per ∗ an // K (Y ,C) H (C (B)) Zc •,Z 43 per per ∗ ∧ ∧ are maps of Rep(G)⊗C modules so we conclude that the map k : H (C (A)) → H (C (B)) ∗ ∗ g g •,G •,Z per per is an isomorphism as well. By Theorem 2.20, the vector spaces H (C (A)) and H (C (B)) admit ∗ ∗ •,G •,Z compatible Hodge structures. We observe that the Hodge decompositions per H (C (A)) H (C (A)) ∗ n •,G •,G n≡∗ mod 2 per H (C (B)) H (C (B)) ∗ ∗ •,Z •,Z n≡∗ mod 2 are decompositions of Rep(G) and Rep(Z) modules respectively. This follows because both the Hodge filtration and the conjugate filtration are filtrations of Rep(G) modules as can be seen for per example by examining the explicit model for C (A). The lemma now follows by taking completions •,G of these decompositions. ∧ n Next we construct a model for C (B) based on algebraic differential forms Ω , regarded as a •,Z projective B-module. Recall that the Cartan differential ∗ n ∗ n−1 i : Sym(z )⊗ Ω → Sym(z )⊗ Ω n−1 n ∗ is the unique extension of the contraction map Ω → z ⊗ Ω to a differential satisfying the Y Y ∗ n Liebniz rule. Alternatively, regarding ω ∈ Sym(z )⊗ Ω as a section of a quasi-coherent sheaf over z, we have (iω)(z) = i ω(z). We thus have a chain complex, in fact a CDGA, • ∗ ∗ n Ω [z ] = Sym(z )⊗ Ω [n], i . Y Y ∗ k Note that i is Z-equivariant, and that it descends to the quotient Sym(z) /m . Thus we can define • ∗ ∗ k n Ω [[z ]] := Sym(z )/m ⊗ Ω [n], i Y Y • ∗ Z ∗ k n Z Ω [[z ]] := (Sym(z )/m ⊗ Ω [n]) , i Y k Y Proposition 4.7. The comparison map of Construction 4.8 below is a quasi-isomorphism of Λ- modules ∧ ∧ • ∗ Z HKR : C (B) → lim(Ω [[z ]] ). •,Z g g Y k ←− Hence when Spec(A)/G is formally proper, we have a quasi-isomorphism of Λ-modules ∧ ∗ ∧ • ∗ Z HKR ◦k : C (A) → lim(Ω [[z ]] ). •,G g g Y k ←− Let Z denote the k-th infinitesimal neighborhood of the identity in Z. The exponential map (k) ∗ k provides a compatible system of isomorphisms exp : Spec(Sym(z )/m )) → Z . Note that under k (k) this equivalence, G acts algebraically on Z by scaling, and this action actually extends to (k) 1 ∗ k an action of the monoid A . This is encoded algebraically via a coaction map Sym(z )/m → ∗ k Sym(z )/m ⊗ C[t]. 44 Construction 4.8. For any b ∈ B, the coaction of O on B, the exponential map exp , and the G -action on z define an element ∗ k exp (−t· z)· b ∈ B ⊗ Sym(z )/m ⊗ C[t]. n+1 k We define C (B, Z) := O ⊗ B , i.e. the reduction of C (B, Z) modulo m , and introduce n,k Z • (k) n ∗ the map HKR : C (B, Z) → Ω [[z ]] given by g,k n,k k ′ ′ ψ⊗b ⊗b ⊗··· b 7→ ψ(g·exp (z)) b d(exp (−t z)·b )∧···∧d(exp (−t z)·b )dt dt ··· dt . (10) 1 n k k 1 1 k n n 1 2 n 0 0 ∗ k Here d(−) denotes the Sym(z )/m ⊗ C[t]-linear extension of the exterior derivative ∗ k 1 ∗ k d : B ⊗ Sym(z )/m ⊗ C[t] → Ω ⊗ Sym(z )/m ⊗ C[t]. ∗ k n The integrand is regarded as an element of Sym(z )/m ⊗ Ω ⊗ C[t , . . . , t ], and the integral over 1 n the standard n-simplex Δ is regarded formally as a linear map C[t , . . . , t ] → C. This formula is n 1 n identical to the one used in [BG], so it follows formally from the computations there that HKR g,k is a chain map (See for instance [BG, Theorem 3.2]). This map is Z-equivariant, so it restricts to a chain maps Z • ∗ Z HKR : C (B, Z) → Ω [[z ]] , and g,k •,k Y k ∧ ∧ • ∗ Z HKR := lim HKR : C (B) → lim Ω [[z ]] . g,k •,Z g g Y k ←− ←− k k Proof of Proposition 4.7. By the compatibility of the HKR maps with translation by the central • ∗ element g, it suffices to consider the case g = id. The maps HKR : C (B, Z) → Ω [[z ]] are id,k •,k k ∗ k a compatible family of maps of bounded complexes with coherent homology over B ⊗ Sym(z )/m . HKR is the classical HKR map id,1 ′ ′ b ⊗···⊗ b 7→ b db ··· db , n 1 n 0 0 n! which is an equivalence of Λ-modules. Hence by Nakayama’s lemma each HKR is a quasi- id,k isomorphism, and the same is true after taking Z-invariants. Hence HKR is a quasi-isomorphism. id The final statement of the proposition combines this with the previous lemma. References [AS ] Michael F Atiyah, Graeme B Segal, et al., Equivariant k-theory and completion, J. Differential Geometry 3 (1969), no. 1-18, 9. [B1] Vladimir Baranovsky, Orbifold cohomology as periodic cyclic homology, International Journal of Mathematics 14 (2003), no. 08, 791–812. [B2] Daniel Bergh, Functorial destackification of tame stacks with abelian stabilisers, arXiv preprint arXiv:1409.5713 (2014). [B3] Anthony Blanc, Topological k-theory of complex noncommutative spaces, arXiv preprint arXiv:1211.7360, to appear in Compositio Mathematica (2012). [BG] Jonathan Block and Ezra Getzler, Equivariant cyclic homology and equivariant differential forms, Annales scientifiques de l’ecole normale sup´erieure, 1994, pp. 493–527. [BGT] Andrew J Blumberg, David Gepner, and Gonc¸alo Tabuada, A universal characterization of higher algebraic k-theory, Geometry & Topology 17 (2013), no. 2, 733–838. [BLS] Daniel Bergh, Valery A Lunts, and Olaf M Schnu¨rer, Geometricity for derived categories of algebraic stacks, arXiv preprint arXiv:1601.04465 (2016). [BRTV] Anthony Blanc, Marco Robalo, Bertrand To¨en, and Gabriele Vezzosi, Motivic realizations of singularity categories and vanishing cycles, J. Ec. polytech. Math. 5 (2018), 651–747. MR3877165 [BZFN] David Ben-Zvi, John Francis, and David Nadler, Integral transforms and drinfeld centers in derived algebraic geometry, Journal of the American Mathematical Society 23 (2010), no. 4, 909–966. [BZNP] David Ben-Zvi, David Nadler, and Anatoly Preygel, Integral transforms for coherent sheaves, arXiv preprint arXiv:1312.7164 (2013). 45 [C] Lee Cohn, Differential graded categories are k-linear stable infinity categories, arXiv preprint arXiv:1308.2587 (2013). [CD] Denis-Charles Cisinski and Fr´ed´eric D´eglise, \’etale motives, arXiv preprint arXiv:1305.5361 (2013). [CT] Andrei C˘alda˘raru and Junwu Tu, Curved A algebras and Landau-Ginzburg models, New York J. Math. 19 (2013), 305–342. MR3084707 [D1] Pierre Deligne, Th´eorie de hodge: Iii, Publications Math´ematiques de l’IHES 44 (1974), 5–77. [D2] Tobias Dyckerhoff, Compact generators in categories of matrix factorizations, Duke Math. J. 159 (2011), no. 2, 223–274. MR2824483 [DG] Vladimir Drinfeld and Dennis Gaitsgory, On some finiteness questions for algebraic stacks, Geometric and Functional Analysis 23 (2013), no. 1, 149–294. [E] Alexander I. Efimov, Cyclic homology of categories of matrix factorizations, Int. Math. Res. Not. IMRN 12 (2018), 3834–3869. MR3815168 [EP] Alexander I. Efimov and Leonid Positselski, Coherent analogues of matrix factorizations and relative singu- larity categories, Algebra Number Theory 9 (2015), no. 5, 1159–1292. MR3366002 [FHT1] Daniel S. Freed, Michael J. Hopkins, and Constantin Teleman, Twisted equivariant K-theory with complex coefficients, J. Topol. 1 (2008), no. 1, 16–44. MR2365650 [FHT2] , Twisted equivariant K-theory with complex coefficients, J. Topol. 1 (2008), no. 1, 16–44. MR2365650 (2009c:19011) [G] Dennis Gaitsgory, ind-coherent sheaves, Mosc. Math. J. 13 (2013), no. 3, 399–528, 553. MR3136100 [GR] Dennis Gaitsgory and Nick Rozenblyum, A study in derived algebraic geometry: Volume i: correspondences and duality, American Mathematical Soc., 2017. [H1] D. Halpern-Leistner, Remarks on Theta-stratifications and derived categories, ArXiv e-prints (February 2015), available at 1502.03083. [H2] Wim H. Hesselink, Uniform instability in reductive groups, J. Reine Angew. Math. 303(304) (1978), 74–96. MR514673 [H3] Yuki Hirano, Derived Kno¨rrer periodicity and Orlov’s theorem for gauged Landau-Ginzburg models, Compos. Math. 153 (2017), no. 5, 973–1007. MR3631231 [HL] Daniel Halpern-Leistner, The derived category of a GIT quotient, J. Amer. Math. Soc. 28 (2015), no. 3, 871–912. MR3327537 [HLP] Daniel Halpern-Leistner and Anatoly Preygel, Mapping stacks and categorical notions of properness, arXiv preprint arXiv:1402.3204 (2014). [HR] Jack Hall and David Rydh, Perfect complexes on algebraic stacks, arXiv preprint arXiv:1405.1887 (2014). [I] M. Isik, Equivalence of the derived category of a variety with a singularity category, Int Math Res Notices 12 (2013), 2787–2808. [IU] A. Ishii and K. Ueda, The special McKay correspondence and exceptional collection, ArXiv e-prints (April 2011), available at 1104.2381. [K1] D Kaledin, Spectral sequences for cyclic homology, Algebra, geometry, and physics in the 21st century, 2017, pp. 99–129. [K2] D. Kaledin, Spectral sequences for cyclic homology, Algebra, geometry, and physics in the 21st century, 2017, pp. 99–129. MR3702384 [K3] Dmitry Kaledin, Non-commutative hodge-to-de rham degeneration via the method of deligne-illusie, Pure and Applied Mathematics Quarterly 4 (2008), no. 3, 785–875. [K4] Bernhard Keller, On differential graded categories, International Congress of Mathematicians. Vol. II, 2006, pp. 151–190. MR2275593 [K5] Frances Clare Kirwan, Cohomology of quotients in symplectic and algebraic geometry, Mathematical Notes, vol. 31, Princeton University Press, Princeton, NJ, 1984. MR766741 [K6] , Partial desingularisations of quotients of nonsingular varieties and their Betti numbers, Ann. of Math. (2) 122 (1985), no. 1, 41–85. MR799252 (87a:14010) [KKP1] L. Katzarkov, M. Kontsevich, and T. Pantev, Hodge theoretic aspects of mirror symmetry, Proceedings of Symposia in Pure Mathematics 78 (2008), 87–174. [KKP2] Ludmil Katzarkov, Maxim Kontsevich, and Tony Pantev, Bogomolov-Tian-Todorov theorems for Landau- Ginzburg models, J. Differential Geom. 105 (2017), no. 1, 55–117. MR3592695 [K] Andrew Kresch, On the geometry of deligne-mumford stacks, Algebraic geometry—seattle 2005. part 1, 2009, pp. 259–271. [KS] Maxim Kontsevich and Yan Soibelman, Notes on a-infty algebras, a-infty categories and non-commutative geometry, Homological mirror symmetry, 2009, pp. 1–67. [L1] Jean-Louis Loday, Cyclic homology, Springer, 1992. [L2] Jacob Lurie, Dag-vii: Spectral schemes, Preprint (2011). 46 [L3] , Higher algebra. 2014, Preprint, available at http://www. math. harvard. edu/˜ lurie (2016). [LMB] G´erard Laumon and Laurent Moret-Bailly, Champs alg´ebriques, Vol. 39, Springer Science & Business Media, [LM] M. Levine and F. Morel, Algebraic cobordism, Springer Monographs in Mathematics, Springer, Berlin, 2007. MR2286826 (2008a:14029) [LP] Kevin H. Lin and Daniel Pomerleano, Global matrix factorizations, Math. Res. Lett. 20 (2013), no. 1, 91–106. MR3126725 [LSM] L Gaunce Lewis, Mark Steinberger, and J Peter May, Equivariant stable homotopy theory, Springer, 1986. [M] Barry Mitchell, Rings with several objects, Advances in Math. 8 (1972), 1–161. MR0294454 [M ] J Peter May et al., Equivariant homotopy and cohomology theory, 1996. [MV] Fabien Morel and Vladimir Voevodsky, A 1-homotopy theory of schemes, Publications Math´ematiques de l’IHES 90 (1999), no. 1, 45–143. [N1] Linda Ness, A stratification of the null cone via the moment map, Amer. J. Math. 106 (1984), no. 6, 1281– 1329. With an appendix by David Mumford. MR765581 [N2] Behrang Noohi, Homotopy types of topological stacks, Adv. Math. 230 (2012), no. 4-6, 2014–2047. MR2927363 [O1] Martin C. Olsson, On proper coverings of Artin stacks, Adv. Math. 198 (2005), no. 1, 93–106. MR2183251 (2006h:14003) [O2] Dmitri Orlov, Matrix factorizations for nonaffine LG-models, Math. Ann. 353 (2012), no. 1, 95–108. MR2910782 [OV] Arthur Ogus and Vadim Vologodsky, Nonabelian hodge theory in characteristic p, Publications math´ematiques 106 (2007), no. 1, 1–138. [P] Anatoly Preygel, Thom–sebastiani and duality for matrix factorizations, arXiv preprint: arxiv:1101.5834 (2011). [S1] Kyoji Saito, Period mapping associated to a primitive form, Publ. Res. Inst. Math. Sci. 19 (1983), no. 3, 1231–1264. MR723468 [S2] Matthew Satriano, De Rham theory for tame stacks and schemes with linearly reductive singularities, Ann. Inst. Fourier (Grenoble) 62 (2012), no. 6, 2013–2051 (2013). MR3060750 [S3] Ed Segal, The closed state space of affine Landau-Ginzburg B-models, J. Noncommut. Geom. 7 (2013), no. 3, 857–883. MR3108698 [S4] Graeme Segal, Equivariant k-theory, Publications Math´ematiques de l’IHES 34 (1968), no. 1, 129–151. [S5] D. Shklyarov, Non-commutative Hodge structures: towards matching categorical and geometric examples, Trans. Amer. Math. Soc. 366 (2014), no. 6, 2923–2974. MR3180736 [S6] The Stacks Project Authors, stacks project, 2015. [S7] J. H. M. Steenbrink, Mixed Hodge structure on the vanishing cohomology, Real and complex singularities (Proc. Ninth Nordic Summer School/NAVF Sympos. Math., Oslo, 1976), 1977, pp. 525–563. MR0485870 (58 #5670) [T1] Gonc¸alo Tabuada, Higher K-theory via universal invariants, Duke Math. J. 145 (2008), no. 1, 121–206. MR2451292 (2009j:18014) [T2] Constantin Teleman, The quantization conjecture revisited, Annals of Mathematics 152 (2000), no. 1, 1–43. [T3] Robert W Thomason, Riemann-roch for algebraic versus topological k-theory, Journal of Pure and Applied Algebra 27 (1983), no. 1, 87–109. [T4] , Algebraic k-theory of group scheme actions, Algebraic Topology and Algebraic K-theory (Ann. Math. Stud. 113, pp. 539–563) Princeton, NJ: Princteon U (1987). [T5] , Equivariant algebraic vs. topological k-homology atiyah-segal-style, Duke Mathematical Journal 56 (1988), no. 3, 589–636. [T6] B. Toen, Th´eor`emes de Riemann-Roch pour les champs de Deligne-Mumford, K-Theory 18 (1999), no. 1, 33–76. MR1710187 (2000h:14010) http://www.deepdyve.com/assets/images/DeepDyve-Logo-lg.png Mathematics arXiv (Cornell University)

Equivariant Hodge theory and noncommutative geometry

Mathematics , Volume 2022 (1507) – Jul 7, 2015

Loading next page...
 
/lp/arxiv-cornell-university/equivariant-hodge-theory-and-noncommutative-geometry-f48pFZd1jL

References (84)

eISSN
ARCH-3343
DOI
10.2140/gt.2020.24.2361
Publisher site
See Article on Publisher Site

Abstract

DANIEL HALPERN-LEISTNER AND DANIEL POMERLEANO Abstract. We develop a version of Hodge theory for a large class of smooth formally proper quotient stacks X/G analogous to Hodge theory for smooth projective schemes. We show that the noncommutative Hodge-de Rham sequence for the category of equivariant coherent sheaves degenerates. This spectral sequence converges to the periodic cyclic homology, which we canonically identify with the topological equivariant K-theory of X with respect to a maximal compact subgroup of G, equipping the latter with a canonical pure Hodge structure. We also establish Hodge-de Rham degeneration for categories of matrix factorizations for a large class of equivariant Landau-Ginzburg models. Contents 1. The noncommutative motive of a quotient stack 7 1.1. Recollections on KN-stratifications 7 1.2. Motivic decompositions via KN stratifications 10 1.3. Λ-modules and the noncommutative Hodge-de Rham sequence 12 1.4. The degeneration property for quotient stacks 14 2. Hodge structures on equivariant K-theory 15 2.1. Equivariant K-theory: Atiyah-Segal versus Blanc 16 2.2. The case of smooth Deligne-Mumford stacks 23 2.3. Equivariant K-theory and periodic cyclic homology 26 2.4. Hodge structure on equivariant K-theory 27 3. Hodge-de Rham degeneration for singularity categories 29 3.1. Preliminaries on categories of singularities on stacks 29 3.2. Motivic decompositions and degeneration for MF 33 3.3. Graded Landau-Ginzburg models 37 4. Computations of Hochschild invariants 41 4.1. Generalities on Hochschild invariants 41 4.2. An HKR theorem and quotients of affine varieties 43 References 45 If X is a smooth projective variety over C, then the cohomology groups H (X;C) can be equipped with a pure Hodge structure of weight n. The theory of Hodge structures then allows one to “lin- earize” many important problems in algebraic geometry. Our goal is to develop such a linearization for the equivariant algebraic geometry of a locally closed algebraic submanifold X ⊂ P which is equivariant with respect to an action of a compact Lie group, M. Note that the complexification G of M, a reductive algebraic group, acts on X as well, and it is natural to ask for a Hodge theory associated intrinsically to the algebraic stack X := X/G. One such linearization follows from the results of [D1], which establishes a canonical mixed Hodge structure on the cohomology of any smooth simplicial scheme and in particular on the equivariant cohomology, H (X), which is the cohomology of the simplicial nerve of the action groupoid of G arXiv:1507.01924v3 [math.AG] 5 Feb 2022 on X. Building on these ideas, one can even associate a motive to the stack X/G as a colimit of motives of schemes as in [MV, Section 4.2]. The present paper lays the groundwork for an alternative approach to equivariant Hodge the- ∗ an ory based on equivariant topological K-theory of the underlying analytic variety K (X ). The equivariant K-theory is a module over the representation ring Rep(G), and the Atiyah-Segal com- pletion theorem canonically identifies its completion at the augmentation ideal of Rep(G) with the Z/2Z-graded equivariant cohomology, even/odd even/odd an ∧ an K (X ) H (X ). M G ∗ an If one completes K (X ) with respect to the evaluation ideal at other g ∈ M, then one recovers the Z/2Z-graded equivariant cohomology of the fixed locus of g with respect to the centralizer of g ∗ an [FHT1, Theorem 3.9]. Thus K (X ) encodes all of these equivariant cohomology groups, as well as the data of how to “spread” them out into a single finitely generated Rep(G)-module. In this ∗ an ∗ sense K (X ) is a much richer invariant than H (X). M G The main challenge is that unlike cohomology, equivariant K-theory is not simply the K-theory of the simplicial scheme arising from the action of G on X, so Deligne’s approach to equivariant Hodge theory does not generalize to K-theory. Instead, our Hodge structures originate in noncommutative algebraic geometry, which views dg-categories as “noncommutative spaces.” We ultimately show that in many cases one can use the dg-enhanced derived category of G-equivariant perfect complexes ∗ an of coherent sheaves on X, Perf(X/G), to construct a pure Hodge structure on K (X ). Noncommutative Hodge-de-Rham degeneration. If A is a dg-category over a field k, the Hochschild chain complex, C (A) plays the role of the Hodge cohomology in noncommutative per algebraic geometry. The periodic cyclic complex C (A), which is a dg-module over k((u)) where u has homological degree −2, behaves like noncommutative de Rham cohomology. There is a per canonical Hodge filtration of the complex C (A) whose associated graded is C (A)⊗k((u)), which • • per leads to a noncommutative Hodge-de Rham spectral sequence converging to H (C (A)) whose first ∗ • per page is H (C (A)⊗k((u))). The Hodge filtration in our theory will be the filtration on H (C (A)) ∗ • ∗ • arising from the degeneration of this spectral sequence. Geometrically, our main tool for establishing degeneration for the category Perf(X/G) will be certain stratifications of X into G-stable locally closed subvarieties. It is a classical observation in geometric invariant theory [H2, N1] that projective varieties with a linearized G-action inherit a canonical stratification. Kirwan used this stratification to deduce many beautiful results concerning the equivariant topology of projective varieties with G-action, including a computation of the Betti K5]. numbers and Hodge numbers for a variety obtained as a GIT quotient [ In this paper, we will consider a certain abstraction of this canonical stratification. These are the (semi-)complete Kirwan-Ness(KN)-stratifications of a G-variety, introduced in [T2] and recalled in Definition 1.1. The chief benefit of this more abstract definition is that it applies in many cases when the ambient variety X is merely quasi-projective. The main classes of examples of G-varieties which admit semi-complete KN-stratifications to keep in mind are: (i) any G-variety X which is projective over an affine G-variety; and (ii) any G-variety X such that X/G admits a good quotient that is projective-over-affine. In the case (i), the KN stratification is complete if and only if dim Γ(X,O ) < ∞. In the case (ii) the stratification is trivial, and it is complete if and only if the good quotient of X/G is projective. Recall that X/G admits a good quotient if there is an algebraic space Y and a G-invariant map π : X → Y such that π : QCoh(X/G) → QCoh(Y ) is exact and (π O ) ≃ O . ∗ ∗ X Y 2 Theorem A (Corollary 1.23). If G is a reductive group and X is a smooth G-quasi-projective vari- ety which admits a complete KN-stratification, then the noncommutative Hodge-de Rham sequence for Perf(X/G) degenerates on the first page. Remark. This builds on [T2], which shows that a version of the Hodge-de Rham spectral sequence ∗ ∗ for H (X) degenerates for such G-schemes and that the (a priori mixed) Hodge structure on H (X) G G is pure in this case. Note that in these examples, the scheme X is not proper, and neither is the quotient stack X/G, so degeneration of the Hodge-de Rham sequence is somewhat unexpected. Likewise from the noncommutative perspective, D. Kaledin’s recent resolution [K1] of a well-known conjecture of M. Kontsevich and Y. Soibelman [KS] shows that the noncommutative Hodge-de Rham sequence degenerates for dg-categories which are smooth and proper. However, the categories Perf(X/G) are typically not smooth even when X is smooth, and they are typically not generated by a single compact generator, so degeneration is again somewhat unexpected. The key observation in establishing the degeneration property for Perf(X/G) is that the forma- tion of the Hochschild complex takes semiorthogonal decompositions of dg-categories to direct sums, and its formation commutes with filtered colimits. Thus if A is a retract of a dg-category which can be built from the derived category of smooth and proper DM stacks via an infinite semiorthogonal decomposition, then the degeneration property holds for A. Example. One simple example is the quotient stack A /G , where G acts with positive weights. m m Then the objects O n{w} ∈ Perf(A /G ), which denote the twist of the structure sheaf by a A m character of G , form an infinite full exceptional collection. Therefore the Hochschild complex of Perf(A /G ) is quasi-isomorphic to a countable direct sum of copies of C (Perf(Spec(k))), and m • the degeneration property follows. We can formulate this most cleanly in terms of G. Tabuada’s universal additive invariant U : dgCat → M [T1,BGT]. Here M is the ∞-category which is the localization of the ∞-category k k of small dg-categories which formally splits all semiorthogonal decompositions into direct sums, and U is the localization map. The following is the main technical result of the paper, and we believe it is of independent interest. Theorem B (See Theorem 1.8). Let X/G be a smooth quotient stack over a field k of characteristic 0 that admits a complete KN stratification. Then there is a smooth projective variety Y such that b ⊕N U (X/G) is a direct summand of U (D (Y )) in M . k k k Connections with (classical) equivariant topology and purity. If G is the complexification of a compact Lie group M as above, we show that one can recover the equivariant topological an + K-theory of the underlying complex analytic space K (X ), as defined in [AS , S4], from the dg-category Perf(X/G). The first ingredient is the recent construction by A. Blanc of a topological K-theory spectrum top K (A) for any dg-category A over C [B3]. Blanc constructs a Chern character natural transforma- top tion ch : K (A) → HP(A), shows that ch⊗C is an equivalence for Perf of a finite type C-scheme, and conjectures this property for any smooth and proper dg-category A. We show that ch⊗C is an isomorphism for all categories of the form Perf(Y), where Y is a smooth DM stack or a smooth quotient stack admitting a semi-complete KN stratification. In fact, we expect that this “lattice conjecture” should hold for a much larger class of dg-categories, such as the categories D (X) for any finite type C-stack and Perf(X/G) for any quotient stack. Following some ideas of Thomason top an T5], we next construct a natural “topologization” map ρ : K (Perf(X/G)) → K (X ) in [ G,X M for any smooth G-quasiprojective scheme X and show: Theorem C (See Theorem 2.10 and Theorem 2.17). For any smooth quasi-projective G-scheme X which admits a semi-complete KN stratification, the topologization map and the Chern character 3 2 provide equivalences G,X ch per ∗ an top oo // K (X )⊗ C π K (Perf(X/G))⊗ C H C (Perf(X/G)) ∗ ∗ • Remark. In fact, Theorem 2.10 shows a bit more. ρ is an equivalence for any smooth G-quasi- G,X projective scheme. For an arbitrary G-quasi-projective scheme X we construct an equivalence c,∨ top b an of spectra ρ : K (D Coh(X/G)) → K (X ), where the latter denotes the M-equivariant G,X an Spanier-Whitehead dual of the spectrum K (X ), sometimes referred to as the equivariant Borel- Moore K-homology of X. ρ is compatible up to homotopy with natural pullback and pushfor- G,X ward maps (to be explained below). This result is of independent interest, and it allows one to “decategorify” theorems regarding equivariant derived categories in a precise way. n an Note that the groups K (X ) are modules over Rep(M), the representation ring of M. We say that a Rep(M)-linear Hodge structure of weight n is a finite Rep(M)-module E along with a finite filtration of the finite Rep(M) -module E ⊗ C inducing a Hodge structure of weight n on the un- per n an derlying abelian group E. Using the previous identification K (X )⊗C ≃ H C (Perf(X/G)), −n • we will show Theorem D (See Theorem 2.20). For any smooth M-quasiprojective scheme admitting a complete n an KN stratification, the noncommutative Hodge-de Rham sequence for K (X ) ⊗ C degenerates n an on the first page, equipping K (X ) with a pure Rep(M)-linear Hodge structure of weight n, functorial in X. There is a canonical isomorphism p n an n−2p der gr K (X ) ≃ H (RΓ(I ,O der)). X I der In this theorem, I denotes the derived inertia stack, sometimes referred to as the “derived loop stack.” As we will see in Lemma 4.2 below, we can express this more concretely as der L G RΓ(I ,O der) ≃ RΓ(G× X × X,O ⊗ O ) , X I Δ −1 where G acts on G × X × X by g · (h, x, y) = (ghg , gx, gy) and the two G-equivariant closed subschemes of G× X × X are defined as Γ = {(g, x, gx)} and Δ = {(g, x, x)} respectively. Example. Along the way, show that the lattice conjecture holds for an arbitrary smooth DM stack, and explicitly compute the Hochschild invariants of Perf(X). For a smooth and proper DM stack, we construct an isomorphism of Hodge structures top 2k−n cl π K (Perf(X))⊗ Q ≃ H (I ;Qhki). Betti X It should be noted that the motivic decompositions of Theorem B play a key role in the proof of Theorem D, but these decompositions do not respect the Rep(M)-linear nature of the Hodge n an structure on K (X ). In Section 4 we spend some time discussing more explicit models for the Hochschild homology and periodic cyclic homology for quotient stacks. For example, we show that when X is smooth and affine, there is an explicit bar-type complex computing the Hochschild homology of Perf(X/G). As an application of Theorem D, we prove an HKR type theorem for the completion of this bar complex at various points of Spec(Rep(G)) when X/G is formally proper. A corollary of this theorem is a description of the completed Hochschild homology modules equipped with the Connes operator in terms of differential forms equipped with the de Rham differential. We will see that these homology level equivalences are induced by suitable chain maps. 4 Extensions to categories of singularities. Another major source of Hodge structures in alge- braic geometry comes from singularity theory. For instance in [S1], Kyoji Saito constructs analogues of Hodge theoretic structures on the universal unfolding of an isolated singularity. More precisely, he describes analogues of the Gauss-Manin connection and period mappings as well as canoni- cal coordinates on the base space of the universal unfolding. Motivated in part by Saito’s work, Katzarkov, Kontsevich, and Pantev [KKP1, KKP2] have proposed a vast generalization of Hodge theory which they call noncommutative (nc) Hodge structures. As the name suggests, they envi- sion that nc Hodge structures should arise naturally from smooth and proper dg-categories(“nc spaces”). Let (X, W ) be a Landau-Ginzburg (LG) model, that is a pair (X, W ) consisting of a smooth quasi- projective variety X and a regular function W : X → A . To any LG model, one may associate the category of matrix factorizations MF(X, W ), which is a 2-periodic (meaning k((β))-linear where β has homological degree −2) dg-category. Applying the theory of nc Hodge structures to these nc spaces is expected to yield a vast generalization of Saito’s theory to pairs (X, W ) with proper critical locus, Crit(W ). For any LG model (X, W ), there is a “dW -twisted” Hodge de-Rham spectral sequence which • • relates the hypercohomologies of the complexes (Ω , dW∧) and (Ω , d+dW∧)(see [KKP1, Section 3.2] for details). Similarly to the classical case, this spectral sequence is known to degenerate when W : X → A is proper by work of Ogus-Vologodsky [OV]. This degeneration result plays a central role in the noncommutative Hodge theory of LG pairs—for example, a version of this result has been used to establish the smoothness of versal deformation spaces of (compactified) LG models (generalizing the universal unfolding space of a singularity)[KKP2]. Efimov and Preygel [P,E] have independently identified the dW -twisted Hodge de-Rham spectral sequence with the k((β))-linear noncommutative Hodge-de Rham spectral sequence for the category MF(X, W )(for closely related results, see also [D2, S3, CT, LP, S5]). It follows that the result of Ogus-Vologodsky can be recast as establishing the degeneration of this spectral sequence of noncommutative origin. We prove the following generalization of this degeneration result, which suggests that in the equivariant context nc Hodge theory should extend to certain dg-categories which are not smooth. We let Crit (W ) denote the critical locus with critical value 0. Theorem E. (Proposition 3.17) If X is a smooth G-quasi-projective scheme which admits a semi- complete KN stratification, and W : X → A is a G-invariant function such that Perf(Crit (W )/G) is a proper dg-category, then the k((β))-linear noncommutative Hodge-de Rham sequence for MF(X/G, W ) degenerates on the first page. Note that by Lemma 1.5, the condition on Crit (W ) in the theorem is equivalent to the induced KN-stratification on Crit (W ) being complete. In the case where X is projective over an affine G-variety, the condition is equivalent to dim Γ(Crit (W ),O ) < ∞, and if X/G admits a Crit (W ) good quotient, the condition is equivalent to the condition that Crit (W )/G admits a projective good quotient. We prove Theorem E by establishing an analog of Theorem B for the k((β))-linear category MF(X/G, W ) in Theorem 3.10. The proof is somewhat more subtle than the case of Perf(X/G), and its formulation is a little more complicated, because at the time of this writing we are not aware of a construction of k((β))-linear additive noncommutative motives. Along the way, we also Noncommutative Hodge structures are also expected to exist in other contexts, notably on the quantum coho- mology of a compact symplectic manifold. 5 establish the degeneration property for MF(X, W ) in the case that X is a smooth quasi-projective DM stack and Crit (W ) is proper (see Section 3). Further questions. The notion of properness in equivariant geometry. Our result on noncommutative Hodge-de Rham degeneration adds to the list of ways in which certain equivariant geometries behave as if they are proper despite not being proper in the sense of algebraic stacks. The intrinsic characterization of which smooth algebraic stacks behave as if they are proper from the perspective of Hodge theory, such as quotient stacks with a complete KN stratification, and which do not, such as BG or BU for a unipotent group U (see Example 1.24), is still somewhat fuzzy. The paper [HLP] studies these properness phenomena systematically by introducing the class of formally proper stacks, with the primary application being the algebraicity of the mapping stack out of a formally proper stack. The examples and counterexamples above for stacks exhibiting noncommutative Hodge-de Rham degeneration are also important examples and counterexamples for stacks which are formally proper in the sense of [HLP]. This raises the natural question. Question 0.1. Do there exist examples of perfect, smooth, and formally proper k stacks X for which the Hodge-de Rham sequence associated to Perf(X) does not degenerate? Hodge structures on equivariant K-theory. We believe that our main theorem for Hodge structures n an on K (X ) raises many questions for further inquiry into the role of Hodge theory in equivariant algebraic geometry. For example, it is plausible that the results above could be extended to con- struct mixed Hodge structures on some version of K-theory for arbitrary finite type stacks. In a different direction, one of the central notions in Hodge theory is that of a variation of Hodge struc- ture. For simplicity, let S be an affine scheme and suppose further that π : X/G → S is a smooth equivariant family over S such that all of the fibers X /G admit complete KN stratifications. Most of the techniques that we have developed work in families, which allows one to establish the exis- per tence of suitable Hodge filtrations on the quasi-coherent sheaf H C (Perf(X/G)). We therefore believe it is quite likely that one can develop a theory of equivariant period maps. Finally, Theorem E suggests that it may be possible to develop a version of noncommutative Hodge theory which applies in the equivariant context. Noncommutative equivariant geometry. Although we make use of noncommutative algebraic geom- etry, all of the differential graded categories in this paper are of commutative origin. It is interesting to try to formulate in noncommutative terms a criterion for the Hodge-de Rham spectral sequence to degenerate. Theorem D suggests the following concrete question: Let A be a proper dg-category which is a module over Perf(BG). Suppose that A⊗ k = Perf(R), where R is a dg- algebra Perf(BG) which is homotopically finitely presented, homologically bounded and such that H (R) is a finitely generated module over HH (R). Question 0.2. Does the Hodge-de Rham spectral sequence always degenerate for such A? Context. Throughout this work, unless explicitly stated otherwise, we work over a fixed subfield ∗ ∗ k ⊂ C. All of our functors are understood to be derived, so we write i for Ri , i for Li , Hom ∗ ∗ for R Hom, etc. We will work with stacks over the ´etale site of k-schemes. By convention, unless otherwise indicated the term quotient stack will denote a quotient of a quasi-projective k-scheme by a linearizable action of an algebraic k-group G, and we denote it X/G. As mentioned above, after the first draft of this paper was circulated, D. Kaledin proved the degeneration conjecture. To the authors’ knowledge, however, the version of the degeneration conjecture for k((β))-linear categories, which is the one which applies to categories of the form MF(X, W ), does not appear in the literature. This is sometimes referred to as a G-quasi-projective scheme. 6 Our stacks will be classical whenever we are studying the derived category of coherent sheaves D Coh(X/G) and its relatives (QC(X),Perf(X), etc.), but when we discuss categories of matrix factorizations MF(X, W ) and its relatives (IndCoh(X), PreMF(X, W ), PreMF (X, W ), etc.), it will be convenient to work with derived stacks. We will work with k-linear dg-categories. For some of the more abstract arguments involving homotopy limits and colimits and symmetric monoidal structures, it will be more convenient to replace them with equivalent stable (i.e. pre-triangulated) dg-categories in the Morita model struc- ture on dg-categories, then to regard them as k-linear stable ∞-categories via the equivalence of [C]. We permit ourselves a bit of fluidity on this point, in that we refer both to the literature on dg-categories and stable ∞-categories as needed for constructions which evidently make sense in either context. Acknowledgements. We would like to thank D. Ben-Zvi, D. Kaledin, F. Morel and A. Preygel for useful conversations. Our thesis advisor C. Teleman explained to us many of the ideas in equivariant K-theory, geometric invariant theory, and noncommutative geometry which this paper builds upon. We are grateful to B. Toen for explaining his computation of the periodic cyclic homology of a Deligne-Mumford stack, and to A. Blanc for useful discussions and for his careful reading of the first draft of this manuscript. Finally, we would like to thank the anonymous referee for several suggestions which improved this article. The first named author was supported by an NSF fellowship, Columbia University, and the Institute for Advanced Study. The second named author was supported by Kavli IPMU, Imperial College, and an EPSRC fellowship. 1. The noncommutative motive of a quotient stack In this section, we show that the noncommutative Hodge-de Rham spectral sequence degenerates for Perf(X) for a large class of smooth quotient stacks subject to a properness condition. Our method for establishing the degeneration property will be to systematically realize the derived category of a smooth quotient stack as being “glued together” from (typically infinitely many) copies of the derived category of smooth Deligne-Mumford stacks. This method will be used several times throughout this paper, so we formulate our main result in a way that can be applied directly in different contexts. We work with the category M of k-linear additive motives in the sense of [T1] (see also [BGT] for a construction using the framework of ∞-categories). This is the ∞-category obtained as the left Bousfield localization of the ∞-category of small k-linear dg-categories localized at the class of morphisms C → A ⊕ B coming from split exact sequences of small dg-categories A → C → B. In other words, objects of M are dg-categories [C], where we have formally adjoined the relation [C] = [A] ⊕ [B] whenever we have a semiorthogonal decomposition C = hA,Bi. We denote the localization functor U : dgCat → M . k k 1.1. Recollections on KN-stratifications. Our primary geometric tool will be a “KN stratifica- tion” of a quotient stack, as defined in [T2, (1.1)] or [HL, Definition 2.2]. This is a decomposition of X as a union of G-equivariant, smooth, locally closed subschemes ss X/G = X /G∪ S /G. (1) For instance, when X is projective-over-affine and G is reductive, a KN-stratification of X/G is induced by a choice of G-linearized ample line bundle L and a Weyl-invariant inner product on the cocharacter lattice of G. Throughout our discussion, we will assume that we have fixed a choice of inner product on the cocharacter lattice of G, and we will refer to the KN-stratification induced by L as the L-stratification. 7 For each i there is a distinguished one parameter subgroup λ of G. If we let L be the centralizer i i of λ , then there is a smooth open subvariety Z ⊂ X which is L -invariant. Then by definition i i i we have n o S := G· x ∈ X| lim λ (t)· x ∈ Z i i i t→0 When the KN stratification arises from GIT, then in fact Z is the semistable locus for the action of L = L /λ (G ) on the closure of Z . i i m i The main object of study in this paper will be quotients stacks admitting a KN stratification of the following form: ss ′ Definition 1.1. A KN-stratification of a quotient stack X/G is semi-complete if X /G and Z /L all admit good quotients which are projective-over-affine. We say that the KN-stratification is complete if all of the qood quotients are projective. ss ′ Remark 1.2. Given a KN-stratification of a G-scheme X, if X /G and Z /L all admit semi- complete (resp. complete) KN stratifications, then the stratification of X can be refined to a semi-complete (resp. complete) KN stratification by replacing each stratum with the preimage of ′ ′ the strata of Z /L under the projection S /G → Z /L and taking the distinguishing one-parameter i i i i i subgroup of each of these new strata to be λ plus a very small rational multiple of the distinguished one-parameter subgroup of the corresponding stratum in Z /L (which can be lifted to L rationally). In a sense the main theorem of GIT is the following: Theorem 1.3. Given a reductive G and any G-ample bundle on a projective-over-affine G-scheme X, the L-stratification is semi-complete. Semi-complete KN stratifications are important because they lead to direct sum decompositions of noncommutative motives. If I is a (possibly infinite) totally ordered set and A is a pretriangulated dg-category, we say that A = hA ; i ∈ Ii forms a semiorthogonal decomposition if objects of the full pre-triangulated dg-subcategories A generate A under cones and shifts, and RHom(A ,A ) = 0 i i j for i > j. In other words, a semiorthogonal decomposition of a pre-triangulated dg-category is by definition a semiorthogonal decomposition of its homotopy category. Lemma 1.4. If X is a smooth G-scheme with a KN stratification, we have an equivalence in M ss U (Perf(X/G)) ≃ U (Perf(X /G))⊕ U (Perf(Z /L )). k k k i i Proof. The main theorem of [HL] provides an infinite semiorthogonal decomposition of Perf(X/G) ss under these hypotheses. One factor of the semiorthogonal decomposition is equivalent to Perf(X /G), and the rest are of the form Perf(Z /L ) , where the subscript denotes the full subcategory of ob- i i w jects whose homology sheaves are concentrated in weight w with respect to λ. The fact that U commutes with filtered colimits implies that the infinite semiorthogonal decomposition maps to an infinite direct sum decomposition of U (Perf(X/G)) ∈ M . On the other hand, the category k k Perf(Z /L ) decomposes as a direct sum of the subcategories Perf(Z /L ) over all w ∈ Z, so i i i i w U (Perf(Z /L ) ) ≃ U (Perf(Z /L )) ∈ M . k i i w k i i k We will also use KN stratifications to compare properness of the dg-category Perf(X/G) to ss properness of the dg-category Perf(X /G) and Perf(Z /L ) for all i. i i Lemma 1.5. Let X be a perfect derived k-stack of finite cohomological dimension, and let Y be cl,red cl,red another perfect derived k stack such that Y ≃ X and O is eventually co-connective. Then the following are equivalent (1) H RΓ(X, F) is finite dimensional for all i and all F ∈ D Coh(X), (2) RΓ(X, F) is finite dimensional for all F ∈ Coh(X), 8 6 (3) Perf(Y) is a proper dg-category. Furthermore, if X is a separated DM stack then this is equivalent to X being proper. Proof. Finite cohomological dimension implies that for any F ∈ D Coh(X) and all i ∈ Z, there is a sufficiently high n such that H RΓ(X, τ F) ≃ H RΓ(X, F), so (2) ⇒ (1). Also, (1) ⇒ (2) i ≤n i because Coh(X) ⊂ D Coh(X) and H RΓ(F) vanishes in all but finitely many degrees. It is clear cl,red cl that (2) can be checked on X because every F ∈ Coh(X) is pushed forward from X , and any cl cl,red F ∈ Coh(X ) has a finite filtration whose associated graded is pushed forward from X . To show that (2) ⇔ (3), it thus suffices to show that (2) is equivalent to Perf(X) being a proper dg-category in the case when X is eventually co-connective. Because X is perfect, for any F ∈ D Coh(X) and any n we can find a perfect complex P such that F is a retract of τ P, ≤n so choosing n large enough shows that H RΓ(X, F) is a retract of H RΓ(X, P), which is finite i i if Perf(X) is a proper dg-category. On the other hand, Perf(X) ⊂ D Coh(X) if X is eventually co-connective, so Hom (E, F) = RΓ(E ⊗ F) is finite dimensional for perfect complexes E and F. For the further claim, it suffices to assume that X is classical. In this case if X is a separated DM stack, one may find a proper surjection from a quasi-projective scheme X → X [O1], and then deduce that X is proper from property (2), and hence X is proper. Lemma 1.6. Let X/G be a quotient stack with a KN stratification. Then Perf(X/G) is a proper ss ′ dg-category if and only if Perf(X /G) and Perf(Z /L ) are proper dg-categories for all i. Proof. It suffices to consider the case of a single closed stratum S ⊂ X with center Z ⊂ S and with open complement U. First assume that Perf(X/G) is a proper dg-category. [H1, Theorem 2.1] a fully faithful embed- ding Perf(U/G) ⊂ D Coh(X/G) (in fact one for each choice of w ∈ Z), and to prove the lemma it will suffice by Lemma 1.5 to show that this embedding preserves RΓ. We will adopt the notation of w − [H1]: this amounts to showing that we can choose a w such that for F ∈ G ⊂ D Coh(X), which − ss ss is identified with D Coh(X ) under restriction, we have RΓ(X, F) ≃ RΓ(X , F). This holds for w = 0 by [H1, Lemma 2.8] Regarding X as a derived stack, we may define the derived fixed locus Z/L, whose underlying classical stack is Z/L. Then [H1, Theorem 2.1] shows that the functor ∗ − ′ − 0 − ˜ ˜ i π : D Coh(Z/L ) ≃ D Coh(Z/L) → D Coh(X/G) ′ ′ is fully faithfull. By Lemma 1.5 the dg-category Perf(Z/L ) is proper, and thus so is Perf(Z/L ). Conversely, assume that Perf(Z/L ) and Perf(U/G) are both proper dg-categories. We will show that Perf(X/G) is proper by invoking Lemma 1.5 and showing that H RΓ(X, F) is finite dimensional for any n and any coherent sheaf F. Again by [H1, Theorem 2.1], we can functorially ′ 0 write F as a finite extension of an object F ∈ G and two objects supported on the unstable − ≥0 − <0 stratum S = S/G, one in D Coh (X) and one in D Coh (X) . In particular as noted above S S ′ ′ we have RΓ(X, F ) ≃ RΓ(U/G, F ), which has finite dimensional homology. ′′ ′′ − Thus it suffices to show that RΓ(X, F ) has finite dimensional homology for any F ∈ D Coh(X) which is set theoretically supported on S. Because X has finite cohomological dimension, we may ′′ b truncate F so that it lies in D Coh(X), and then in can be built out of a sequence of extensions of shifts of objects of the form i E for E ∈ Coh(S/G). Thus it suffices to show that Perf(S/G) is proper. A similar filtration argument using the baric decomposition of [H1, Lemma 2.2] can be used to deduce that Perf(S/G) is proper because Perf(Z/L) is proper. Finally, the projection We will need to consider the derived critical loci below, which is why we have introduced derived stacks here. If X is classical, then there is no need to replace X by an eventually co-connective approximation in (3), but the example Y = X = Spec(k[t]), where t is a variable of homological degree 2, shows that (2) does not imply (3) without the eventually co-connective hypothesis. 9 ′ Z/L → Z/L is a G -gerbe, so the pushforward preserves perfect complexes, and thus Perf(Z/L) is proper if Perf(Z/L ) is proper. Corollary 1.7. Let X/G be a quotient stack with a semi-complete KN stratification. Then Perf(X/G) is a proper dg-category if and only if the stratification is complete. Proof. Combine Lemma 1.6 with Lemma 1.5. 1.2. Motivic decompositions via KN stratifications. We will consider the class of stacks ⊕N which have semi-complete KN stratifications as in Definition 1.1. We use the notation C to denote the direct sum of countably many copies of the dg-category C. Recall also the definition of [K] that a DM stack of finite type with finite inertia over a field of characteristic 0 is quasi-projective if X is a global quotient stack and has a quasi-projective coarse moduli space. We will say that X is furthermore projective-over-affine if its coarse moduli space is projective over an affine variety. Theorem 1.8. Let G be an algebraic group over a field k of characteristic 0. Let X be a smooth G- quasiprojective k-scheme with a semi-complete KN stratification. Then there is a smooth projective- ⊕N over-affine Deligne-Mumford stack Y such that U (Perf(X/G)) is a direct summand of U (Perf(Y)) k k in M . Furthermore if Perf(X/G) is a proper dg-category, then Y can be chosen to be a smooth projective scheme. Note that by Corollary 1.7, Perf(X/G) is proper if and only if the KN stratification is complete. Remark 1.9. The proof is constructive, and actually produces something a bit stronger: if C is the ∞-category of small dg-categories, then Perf(X/G) lies in the smallest subcategory containing Perf(Y) and closed under countable semiorthogonal gluings and passage to semiorthogonal factors. Example 1.10. If X is projective-over-affine with a linearizable G-action, then the condition that Perf(X/G) is a proper dg-category is equivalent to the condition that H RΓ(X,O ) is finite 0 X dimensional, by [HLP, Proposition 4.2.3]. Example 1.11. We can write any algebraic k-group G as a semidirect product G = U ⋊ L, where U is its unipotent radical and L its reductive quotient. Assume that there is a one-parameter subgroup λ : G → L which is central in L and acts with positive weights on Lie(U) in the adjoint representation of G. Then this one-parameter subgroup defines a single KN stratum S = X = {∗}, and Z/L = ∗/L → ∗ is a good quotient. Thus Theorem 1.8 applies to a large class of categories of the form Perf(BG), including when G is a parabolic subgroup of a reductive group. Example 1.12. If G is as in the previous example, and X is a smooth projective-over-affine G- scheme, then one can consider the Bialynicki-Birula stratification of X under the action of λ(G ), λ(G ) L which is a KN stratification. If this is exhaustive, and Γ(X ,O ) is finite dimensional, λ(G ) then the Bialynicki-Birula stratification can be refined to a complete KN stratification of X as in Remark 1.2. Our proof of Theorem 1.8 will proceed by a delicate inductive argument. One of the key tools is the following: Lemma 1.13. Let π : Y → X be a rational morphism of finite-type k-stacks, meaning Rπ O ≃ O . Y X Assume that X is smooth and π preserves D Coh. Then U (Perf(X)) is a summand of U (Perf(Y)) k k in M . Proof. First consider the categories Perf(Y) and Perf(X). The unit of adjunction id → π π is X/G an equivalence in Perf(X), hence π is fully faithful and admits a right adjoint. Hence Perf(X) is a semiorthogonal factor of Perf(Y). The original version of this paper had Y as a projective DM stack, but by subsequent work of Bergh, Lunts, and Schnuerer [BLS] one can further reduce to a smooth projective scheme. 10 We will apply Lemma 1.13 in three different situations. Example 1.14. If π : Y → X is a flat morphism of algebraic stacks such that for every k-point p of X the fiber Y satisfies RΓ(Y ,O ) ≃ k, then π is rational. If π is not flat, then the same is true p p Y if we take Y to refer to the derived fiber. Example 1.15. Any representable birational morphism of smooth k-stacks is rational. Indeed we can reduce this to the case for schemes, as birational morphisms are preserved by flat base change and the property of a morphism being rational is fppf-local on the base. Example 1.16. Let G → H → K be an extension of linearly reductive groups, and let K act on a scheme, X. Then the morphism p : X/H → X/K is a G-gerbe – after base change to X this morphism becomes the projection X × BG → X. Thus because G is linearly reductive Rp O ≃ O . X/H X/K Let π : X → X be a projective morphism of smooth projective-over-affine varieties which is equivariant with respect to the action of a reductive group, G. For a G-ample invertible sheaf L on X and a relatively G-ample invertible sheaf M on X , we consider the fractional polarization L = L + ǫM for ǫ ∈ Q. We will need the following: Lemma 1.17. [T2, Lemma 1.2] For any small positive ǫ ∈ Q, the L -stratification of X refines the preimage of the L-stratification of X. Finally, we need another GIT lemma: Lemma 1.18. Let X be a G-quasi-projective scheme which admits a good quotient π : X → Y such ss that Y is projective-over-affine. Then X = X for some linearized projective-over-affine G-scheme X, which can be chosen to be smooth if X is smooth. Proof. The proof of [T2, Lemma 6.1] applies verbatim: one constructs a relative G-compactification for X → Y by choosing a sufficiently large coherent F ⊂ π O so that X embeds in the projectiviza- ∗ X tion of Spec Sym(F). The closure of X is projective over Y , and hence projective-over-affine, and ss it has a linearization for which X = X by the cited argument. Furthermore, one can equivariantly resolve any singularities occuring in X \ X if X is smooth. Proof of Theorem 1.8. Over the course of the proof, we will actually construct a finite set of smooth ⊕N quasi-projective DM stacks Y , . . . ,Y such that U (Perf(X/G)) is a retract of U (Perf(Y )) ⊕ 1 N k k 1 F F ⊕N ···⊕U (Perf(Y )) , and then we may take Y = Y ··· Y at the end. Also, when Perf(X/G) k N 1 N is a proper dg-category, it suffices to show that the Y can be chosen to be smooth proper DM stacks, because [BLS, Theorem 6.6] implies that for any smooth proper DM stack Y, Perf(Y) is geometric, meaning a semiorthogonal factor of Perf of a smooth projective variety. We shall prove the theorem by induction on the rank of G. Note that by Lemma 1.4 and the definition of a semi-complete KN-stratification, it suffices to prove this for quotient stacks which have projective-over-affine good quotients. For our purposes, it will be more convenient to consider smooth G-schemes which are projective-over-affine, and by Lemma 1.18 and Lemma 1.4 it suffices to prove the claim for open unions of KN strata in a quotient stack of this form. We fix a G-ample bundle L on X and consider the L-stratification as in Equation 1. ss Case X = ∅: By Lemma 1.4 we must verify the claims for U (Perf(Z /L )) for all i for which Z ⊂ U. First i i i assume that the inclusion λ(G ) ⊂ L admits a splitting L → G , so that L ≃ G × L where m i i m i m the left factor is λ(G ). Then Z /L ≃ BG × Z /L , so U (Perf(Z /L )) is a direct sum of copies m i i m i k i i )). This is the only point of the proof at which an infinite direct sum enters, and of U (Perf(Z /L k i 11 it is an infinite direct sum of copies of the same category, hence throughout the proof we will only encounter a finite set of distinct DM stacks. If λ(G ) ⊂ L is not split, then we can choose a surjective homomorphism L → L with finite m i i ˜ ˜ kernel, where L ≃ G × L and G ×{1} → L factors through λ(G ). The morphism p : Z /L → m m i m i Z /L is rational, hence Lemma 1.13 reduces the problem to showing the claim for Z /L. By the i i i argument of the previous paragraph it again suffices to prove the claim for Z /L . ¯ i Let Z be the closure of Z , which is a connected component of X and hence smooth and i i ′ ′ projective-over-affine. Then Z is the semistable locus for the action of L on Z , and L has lower i i i i rank than L , so the first claim of the theorem follows from the inductive hypothesis. Note that Lemma 1.6 implies that Perf(Z /L ) is a proper dg-category if Perf(U/G) is, so in this case the inductive hypothesis implies that one can choose the DM stacks Y to be projective. s ss Case X = X 6= ∅: ss The argument in the case where X = ∅ applies here as well, so the inductive hypothesis implies that the conclusion of the theorem holds for Perf(Z /L ) for all Z ⊂ U. By Lemma 1.4 it suffices to i i i ss ss show that the claims hold for X /G. In this case X /G is a smooth separated Deligne-Mumford stack whose coarse moduli space is projective-over-affine. Furthermore if Perf(U/G) is a proper ss ss dg-category then so is Perf(X /G) by Lemma 1.6, and hence X /G is projective by Lemma 1.5. ss ss s ss s ss Case X 6= ∅, X 6= X , and codim(X \ X , X ) ≥ 2: ss As in the previous case, it suffices to show the claims for X /G. Here we use the main result of ′ ′ ss ′ s [K6], which says that there is a birational morphism π : X → X such that (X ) (L ) = (X ) (L ), ǫ ǫ Lemma 1.17 the open subset where L = π L + ǫM for a suitable relatively G-ample M. By ′ −1 ss ′ ss U := π (X (L)) is a union of KN strata, and π : U → X (L) is rational, so by Lemma 1.13 we may reduce the main statement of the theorem to the corresponding claim for U /G, which falls under the previous case. Furthermore, if Perf(U/G) is a proper dg-category the fact that ′ ′ U /G → U/G is proper implies that Perf(U /G) is a proper dg-category, so again we may reduce to the previous case to show that the Y can be chosen to be projective. ss ss s ss Case X 6= ∅ but codim(X \ X , X ) ≤ 1: Let Y be a smooth projective variety with a G-action such that RΓ(Y,O ) ≃ k and for some linearization M we have codim(Y \ Y , Y ) ≥ 2. For instance, Y could be a suitable product of flag varieties, or a large projective space with a suitable linear G action. We linearize the G action on X × Y with L = L + ǫM. By Lemma 1.17, the open subvariety U × Y ⊂ X × Y is a union of KN strata for the L stratification. The projection U ×Y/G → U/G is rational, and so by Lemma 1.13 it suffices to prove the claims for U × Y/G. Note that Perf(U × Y/G) is a proper dg-category if Perf(U/G) is. Finally we have a sequence of inclusions ss s s ss ss X × Y ⊂ (X × Y ) ⊂ (X × Y ) ⊂ X × Y, where the first inlcusion is due to the fact that points in Y have finite stabilizers and thus so do ss s points in X × Y , and the last inclusion follows from Lemma 1.17. This implies that ss s ss ss s ss s codim((X×Y ) \(X×Y ) , (X×Y ) ) ≥ codim(X ×(Y \Y ), X ×Y ) = codim(Y \Y , Y ) ≥ 2, which reduces us to the previous case. 1.3. Λ-modules and the noncommutative Hodge-de Rham sequence. Let us recall the negative cyclic and periodic cyclic homology of a small k-linear dg-category, A. We let C (A) ∈ 12 2 D(Λ) denote the (mixed) Hochschild complex of A, regarded as a dg-module over Λ = k[B]/B where B has homological degree 1 and acts on C (A) by the Connes differential. We have (n) C (A) := C (A)⊗ k[u]/(u ) • • (n) C (A) := lim C ←− per − C (A) := C (A)⊗ k((u)) • • k[[u]] where u is a variable of homological degree −2. The differential on each complex is given by d+uB, where d is the differential on C (A). In fact, these constructions make sense for any Λ-module M. We sometimes denote the negative cyclic construction M and the periodic cyclic construction Tate M . See Lemma 3.2 below. (n) Definition 1.19 ([KS]). The category A is said to have the degeneration property if H (C (A)) ∗ • is a flat k[u]/(u )-module for all n ≥ 1. It is immediate from the definitions that the degeneration property is preserved by filtered colimits of dg-categories. It is also known that the degeneration property holds for categories of the form A = Perf(R), where R is a smooth and proper dg-algebra [K3, K2]. In particular, this holds when A = Perf(X), where X is a smooth and proper Deligne-Mumford stack over k [HR], although a more direct argument in this case follows from Proposition 2.13 below. If A satisfies the degeneration property, then H (C (A)) is a flat k[[u]]-module. (See [KS, Corollary 9.1.3]). The degeneration property owes its name to its relationship with the noncommutative Hodge-de Rham spectral sequence. This is the spectral sequence associated to the filtration of the complex p per p − per F C (A) = u · C (A) ⊂ C (A). • • • per The E page of the spectral sequence is gr C (A) ≃ C (A) ⊗ k((u)). The degeneration property 1 • • implies that this spectral sequence degenerates on the first page, so the associated graded of the per resulting filtration on H (C (A)) is isomorphic to H (C (A))⊗k((u)). Under the assumption that ∗ • ∗ • A is suitably bounded, we can say something more precise: Lemma 1.20. Let A be a dg-category such that H (C (A)) is homologically bounded above and ∗ • which satisfies the degeneration property. Then there exists an (non-canonical) isomorphism H (C (A)) ≃ H (C (A))⊗ k[[u]]. ∗ • Proof. This follows from the remark before Theorem 4.14 of [KKP1]. The hypothesis of Lemma 1.20, that H (C (A)) is homologically bounded above, will apply to ∗ • Perf(X) for all smooth k-stacks X of finite cohomological dimension such that QC(X) is compactly generated. Remark 1.21. The Λ-module C (A) is functorial in A. When A is a symmetric monoidal k-linear ∞-category, exterior tensor product followed by the symmetric monoidal product gives a natural map − − − − C (A)⊗ C (A) → C (A⊗ A) → C (A) • • • • per per and likewise for C (A). On the level of homology, this gives H C (A) the structure of a • ∗ • commutative k((u))-algebra, and for any symmetric monoidal functor between symmetric monoidal per per ∞-categories A → B, the resulting map H C (A) → H C (B) is a map of commutative k((u))- ∗ • ∗ • algebras. 8 p p+1 Note that the filtration is not a filtration of k((u))-modules, as u · F ⊂ F . As explained to us by D. Kaledin, this can be understood by thinking of u as the Tate motive. In other words when k ⊂ C, rather than regarding k[[u]] ∗ ∞ simply as a complex (where u has cohomological degree 2), we regard it as H (P ) with its Hodge structure, which places u in F C[[u]]. 13 1.4. The degeneration property for quotient stacks. In [T2, Theorem 7.3], Teleman estab- lishes the degeneration of a commutative Hodge-de Rham sequence, which converges to the equivari- ant Betti-cohomology H (X), for a smooth quotient stack X/G with a complete KN stratification. The argument in [T2] makes use of the KN stratification and has a similar flavor to the proof of Theorem 1.8. However the proof in the commutative case is substantially simpler. In the noncom- mutative situation, we are not aware of an argument to reduce the proof of degeneration to the case of the quotient of a smooth projective scheme by the action of a reductive group, as was done in [T2]. Using the motivic statement of Theorem 1.8, we can immediately deduce noncommutative HdR degeneration. The main observation is the following Lemma 1.22. The degeneration property is closed under direct summands and arbitrary direct sums in M . Proof. The Hochschild complex C (−) is an additive invariant of dg-categories, hence factors through U uniquely up to contractible choices. The claim follows from the fact that the oper- ation D(Λ) → D(k) taking mapping (M, d, B) 7→ (M ⊗ k[u]/(u ), d + uB) commutes with filtered colimits and in particular infinite direct sums, and the fact that an infinite direct sum of k[u]/(u ) modules is flat if and only if every summand is flat. Corollary 1.23. Let G be a reductive group and let X be a smooth G-quasiprojective scheme which admits a complete KN stratification. Then Perf(X/G) has the degeneration property. Proof. Combine Lemma 1.22 with the conclusion of Theorem 1.8. Example 1.24. As a counterexample, consider Perf(BG ). This category is Morita equivalent to the category Perf(k[ǫ]/(ǫ )) where ǫ has degree −1. By the (graded-commutative) HKR theorem, 2 2 ∗ proposition 5.4.6 of [L1], we have that H C (Perf(k[ǫ]/(ǫ )) = k[ǫ]/(ǫ )⊗ Sym (dǫ), where dǫ has ∗ • degree 0. By theorem 5.4.7 of the same book, the Connes operator goes to the de Rham differential which sends ǫ → dǫ and so the spectral sequence does not degenerate. We also observe, somewhat surprisingly, that the derived category of coherent sheaves on cer- tain singular quotient stacks also has the degeneration property. We will consider the following goeometric set up ss • X/G = X /G ∪ S /G is a complete KN stratification (Definition 1.1) of a smooth quotient stack, • V is a G-equivariant locally free sheaf on X such that V| has λ -weights ≤ 0 for all i, and Z i • σ ∈ Γ(X, V ) is an invariant section. Note that the quantization-commutes-with-reduction theorem [T2] implies that if the λ -weights of G ss G V| are strictly negative, then Γ(X, V ) ≃ Γ(X (L), V ) (this is referred to as adapted in [T2]). Using the methods of [HL] one can show that dim Γ(X, V ) < ∞ even when the λ weight of V| i Z vanishes for some i. Amplification 1.25. In the set up above, if ss (1) σ is regular on X with smooth vanishing locus, and λ=0 (2) for all i the restriction of σ to (V| ) , the summand of V| which is fixed by λ(G ), is Z Z m i i regular with smooth vanishing locus, then there is a smooth and proper quasi-projective DM stack Y such that U (D Coh(X /G)) is a k σ ⊕N retract of U (Perf(Y)) . Proof. We apply the structure theorem for the derived zero locus X in [H1, Theorem 3.2], whose derived category is just the derived category of the sheaf of cdga’s over X/G given by the Koszul 14 algebra A = (Sym(V [1]), dφ = φ(s)). The structure theorem constructs an infinite semiorthogonal decomposition which generalizes the b ss main structure theorem of [HL]. One factor is isomorphic to D Coh(X /G), and the remaining b ′ w ′ factors are isomorphic to D Coh(Z /L ) , where Z denotes the derived zero locus of σ restricted i i λ=0 b ′ to (V| ) , and the superscript w denotes the full subcategory of D Coh(Z /L ) consisting of Z i complexes whose homology is concentrated in weight w. In order to apply this theorem, we must check that after restricting the cotangent complex L X /G to Z /L and looking at the summand with λ-weights < 0, there is no fiber homology in homological degree 1. Because X is a derived zero section, we have h i λ<0 ∨ λ<0 λ<0 ∨ λ<0 ′ ′ ′ L | ≃ (V | ) → (Ω | ) → O ⊗ (g ) . Z X X /G Z Z Z σ i i i i So the weight hypotheses on V| imply that this is a two term complex of locally free sheaves in homological degrees 0 and −1, and hence has no fiber homology in homological degree 1. Lemma 1.4 now applies verbatim Given the structure theorem for D Coh(X /G), the proof of to give a finite direct sum decomposition b b ss b ′ U (D Coh(X /G)) = U (D Coh(X /G))⊕ U (D Coh(Z /L )). k σ k k i σ i Under the hypotheses of the amplification, each factor in this direct sum decomposition is D Coh of a smooth quotient stack satisfying the hypotheses of Theorem 1.8, and the result follows. Remark 1.26. Note that when V is strictly adapted to the KN stratification, then the condition (2) in the previous amplification is vacuous. Corollary 1.27. In the set up of Amplification 1.25, the category D Coh(X /G) has the degener- ation property. We will see another approach to establishing Corollary 1.27 using graded LG-models in Example 3.23 and Proposition 3.26. 2. Hodge structures on equivariant K-theory In this section we consider the action of a reductive group G on a smooth quasi-projective per C-scheme X. Our goal is to identify the periodic cyclic homology C (D Coh(X/G)) with the an complexification of the Atiyah-Segal equivariant topological K-theory K (X ) with respect to a maximal compact subgroup M ⊂ G (see Section 2.1.2). Our final result, Theorem 2.20, will n an allow us to define a pure Hodge structure of weight n on K (X ) in the case where X admits a complete KN stratification. Rather than construct a direct isomorphism, we study an intermediate object, the topological K- top b theory of the dg-category K (D Coh(X/G)), as defined in [B3], which admits natural comparison top isomorphisms with each of these theories. In Blanc’s construction, K (C) is constructed from the geometric realization of the presheaf of spectra on the category, Aff, of affine C-schemes of finite type, K(C) : A 7→ K(A⊗ C). Here K(−) denotes the non-connective algebraic K-theory of a dg-category of [B3, Definition 2.10]. The geometric realization of a presheaf, | • |, is defined to be the left Kan extension of ∞ an the functor A 7→ Σ (SpecA) , regarded as functor with values in spectra, along the Yoneda embedding of the category of finite type C-schemes into presheaves of spectra, Aff → Sp(Aff). The geometric realization functor | • | : Sp(Aff) → Sp admits a right adjoint, which assigns M ∈ Sp 15 ∞ an to the presheaf of spectra H (M) := Hom (Σ (•) , M). The semi-topological K-theory is the B Sp geometric realization st K (C) := |K(C)|, st st regarded as a K (C)-module spectrum. By [B3, Theorem 4.5], we have an isomorphism K (C) ≃ bu, where the latter denotes the connective topological K-theory spectrum. Choosing a generator β ∈ π (bu), one then defines the topological K-theory of a dg-category to be −1 top st −1 K (C) := K (C)[β ] = |K(C)|⊗ bu[β ] bu Remark 2.1. In the definition of the presheaf K(C), A⊗ C denotes the derived tensor product of non-idempotent-complete dg-categories, as in [K4, Section 4.3]. In Section 3 we will also consider a symmetric monoidal structure on small stable idempotent complete dg-categories, which we will denote A⊗B to avoid confusion. In addition, following the convention of [B3, Definition 2.7], throughout this paper we define the algebraic K-theory of a small dg-category C to be the algebraic K-theory of the Waldhausen category Perf(C), so that it is automatically Morita invariant. We warn the reader that as a consequence, if C is a stable dg-category, then K(C) denotes the usual algebraic K-theory of the idempotent completion of C, not the K-theory of C itself. per top We will also use the construction of a Chern character map Ch : K (C) → C (C). First, one per obtains a map of presheaves K(C) → C (C) from the usual Chern character in algebraic K-theory, per per where C (C) denotes the presheaf A 7→ C (A⊗ C). Using a version of the Kunneth formula for • • C per per per periodic cyclic homology, one obtains an equivalence |C (C)| ≃ C (C) ⊗ ± |C (C)|. Then • • • C[u ] per one can construct an isomorphism of presheaves C (C) ≃ H (C[u ]), which leads to a map per |C (C)| → C[u ]. Combining these provides a map st per per per K (C) → C (C)⊗ ± |C (C)| → C (C) C[u ] • • • which give the Chern character after inverting β. The main result we use is [B3, Proposition 4.32], which states that for a finite type C-scheme, X, the Chern character induces an equivalence per top K (Perf(X)) ⊗ C → C (Perf(X)). Furthermore, there is a natural topologization map which top an is an equivalence K (Perf(X)) → K(X ), and under this equivalence Ch can be identified with per an a twisted form of the usual Chern character for X under a canonical isomorphism C (X) → H (X;Q)⊗ C[u ]. More precisely, Blanc’s Chern character provides an equivalence Betti Q top an K (Perf(X))⊗ Q ≃ H (X;Q)⊗ Q(( )) ⊂ H (X ;Q)⊗ C((u)), Betti Betti 2πi which we can alternatively express as an isomorphism top 2p−n an π (K (Perf(X)))⊗ Q ≃ H (X ;Qhpi), where Qhpi ⊂ C denotes the subgroup (2πi) Q. 2.1. Equivariant K-theory: Atiyah-Segal versus Blanc. In this section we consider a reduc- tive group G with maximal compact subgroup M ⊂ G, and a G-quasi-projective scheme X, which need not be smooth. The goal of this section will be to construct a comparison isomorphism be- top b an tween K (D Coh(X/G)) and topological M-equivariant K-homology of X with locally compact supports, whose construction and properties we recall below. per In order to be consistent with the rest of the paper, we use the notation C for the periodic cyclic homology complex of a dg-category, rather than the notation HP used in [B3]. In addition, we use the notation ⊗ rather than top ∧ for the smash product of spectra and module spectra. For example K (C) ⊗ C is the C-module spectrum, which top B3]. we canonically identify with a complex of C-modules, which is denoted K (C) ∧ HC in [ 16 2.1.1. Equivariant Borel-Moore homology theories from invariants of dg-categories. Let us consider an additive invariant of dg-categories E(−). Then we have a Borel-Moore homology theory on the category of G-schemes by defining E (X) := E(D Coh(X/G)) (see Remark 2.4 below). This BM assignment is covariantly functorial with respect to proper G-equivariant maps p : X → X by b ′ b applying E(−) to the pushforward functor p : D Coh(X /G) → D Coh(X/G), and E(X) is contravariantly functorial with respect to flat G-equivariant maps f : X → X by applying E(−) ∗ b b ′ to the pullback functor f : D Coh(X/G) → D Coh(X /G). Note also that if X is a smooth G-space, then we have a canonical equivalence E(Perf(X/G)) → E(D Coh(X/G)), so we can canonically identify E (X) with the “cohomology” theory E(X) := E(Perf(X/G)), which is a BM form of Poincar´e duality. More precisely, let Spaces denote the category of algebraic spaces with a G-action and G- equivariant maps, and let Spaces , where ? = f, p,≃, denote the subcategories consisting of flat maps, proper maps, and isomorphisms respectively. We regard the additive invariant E as giving p p f f op two strict functors into spectra E : Spaces → Sp and E : (Spaces ) → Sp along with BM G BM G an isomorphism of the restriction of these functors to the subcategories Spaces ⊂ Spaces and ≃ ≃ op Spaces ⊂ (Spaces ) . Here we have used the fact that Spaces is a groupoid to canonically G G ≃ ≃ op identify Spaces ≃ (Spaces ) . G G Recall that an additive invariant E(−) of dg-categories is localizing if it takes localization se- quences of dg-categories to exact triangles, and we say that E(−) satisfies equivariant d´evissage ′ b if for any closed immersion i : X ֒→ X of G-spaces the induced map i : E(D Coh(X/G)) → b ′ b ′ E(D Coh (X /G)) is an equivalence, where D Coh (X /G) denotes the derived category of quasi- X X coherent complexes with coherent cohomology sheaves which are set theoretically supported on X. Lemma 2.2. Let E be an additive invariant of dg-categories. Then: ′ ′ ′ (1) The base change formula holds: if p : X → Y is proper, f : Y → Y is flat, and f and p are ∗ ′ ∗ ′ base change of f and p, then we there is an equality of compositions p ◦ f = (f ) ◦ (p ) : ∗ ∗ E (X) → E (Y ) in the homotopy category of spectra Ho(Sp); BM BM (2) If V is a G-equivariant vector bundle on X and π : P(V) → X is the projection, then π : E (X) → E (P(V)) followed by (−) ⊗ O (k) : E (P(V)) → E (P(V)) is a BM BM P(V) BM BM split injection; (3) The previous maps, where k ranges from 0, . . . , n− 1, define a canonical equivalence ⊕n E (P(V)) ≃ E (X) . BM BM Furthermore, if E is localizing and satisfies equivariant d´evissage, then: ′ ′ (4) For any closed immersion of G-spaces i : X ֒→ X with open complement j : U ⊂ X , there is an exact triangle E (X) −→ E (X ) −→ E (U) →; BM BM BM (5) If π : V → X is a G-equivariant torsor for V, then π : E (X) → E (V ) is an BM BM equivalence. Proof. The claim (1) follows immediately by applying E(−) to the usual base change formula. Claims (2) and (3) follow from the fact that E is an additive invariant and there is a canonical semiorthogonal decomposition b ∗ b ∗ b ∗ b D Coh(P(V)/G) = hπ (D Coh(X/G)), π (D Coh(X/G))⊗O (1), . . . , π (D Coh(X/G))⊗O (n−1)i. P(V) P(V) Now let us assume that E satisfies d´evissage. (4) follows immediately from the d´evissage condi- tion and the localization exact triangle b ′ b ′ b ′ E(D Coh (X /G)) → E(D Coh(X /G)) → E(D Coh((X − X)/G)) → . 17 For (5), we note that Thomason’s proof when E(−) is algebraic K-theory in [T4, Theorem 4.1] generalizes to this context as well: the torsor V is classified by an extension 0 → V → V → O → 0 of G-equivariant locally free sheaves on X. V is isomorphic to the complement of the closed ′ ′ embedding P(V) ֒→ P(V ). Using the direct sum decomposition of E(P(V )) and E(P(V)) from (3) one can show that E(i ) : E(P(F)) → E(P(W)) is a split injection, and the pullback map gives an equivalence of the cofiber with E(X). It follows from the localization sequence of (4) that the cofiber of E(i ) can is canonically identified with E(P(V )−P(V)) under the restriction map, so we have that pullback gives an equivalence E(X) ≃ E(V ). Example 2.3. Algebraic K-theory of dg-categories as defined in [B3, Definition 2.10] is a localizing top invariant and satisfies d´evissage, and it follows that K (−) is localizing and satisfies d´evissage as well. Remark 2.4. The theory E (X) gives a Borel-Moore homology theory in the sense of [LM, BM Definition 2.1.2], but whose source is the category of G-spaces and which takes values in Ho(Sp) rather than graded abelian groups. We are forced to use the homotopy category of spectra because we have only formulated the base change formula in this setting. In order to formulate a “strict” Borel-Moore homology theory valued in spectra, one would have to define the source category as an ∞-category of correspondences, as in [GR, Section V]. Remark 2.5. Although we are only interested in quotient stacks here, one can define a Borel-Moore homology theory E (X) := E(D Coh(X)) for arbitrary algebraic stacks, and the statements and BM proofs of the previous lemma extend to this context. 2.1.2. Atiyah-Segal equivariant K-theory. We now turn our attention to topological equivariant K-theory with respect to a maximal compact subgroup M ⊂ G. There is a K-cohomology theory for topological M-spaces constructed in [AS ]. Below we use the more systematic description of K (X) in terms of equivariant stable homotopy theory as the spectrum obtained by taking level- wise M-equivariant mapping spaces from X to the naive M-spectrum underlying the M-spectrum bu . For details on the non-equivariant and equivariant stable homotopy category, we refer the + an reader to [LSM] and [M , Chapters XII, XIV, and XVI]. K (X ) can also be constructed as the K-theory spectrum associated via Quillen’s Q-construction to the exact category of M-equivariant an complex vector bundles on X . We will also consider the Atiyah-Segal equivariant K-homology with locally compact supports c,∨ an AS K (X ). This theory was studied in [T5, Section 5] under the notation G (G, X), and our discussion follows this reference closely. In particular, we refer the reader there for a nice discus- c,∨ sion contextualizing K (−) with respect to several other versions of equivariant K-theory. We c,∨ an have chosen to denote the M-equivariant K-homology with locally compact supports as K (X ) because it is the M-equivariant Spanier-Whitehead dual (see [M , XVI.7]) of the M-spectrum of equivariant K-theory with compact supports constructed in [S4, Page 136], which we denote c an K (X ). In particular, K (X) is strictly covariantly functorial for open immersions [S4, Proposition c,∨ 2.9] and strictly contravariantly functorial for proper maps. Dually, it follows that K (−) can op be regarded as both a strict functor Spaces → Sp and a strict functor (Spaces ) → Sp, G G where Spaces ⊂ Spaces denotes the subcategory of open immersions, and these two functors are isomorphic on the subcategory Spaces . Poincar´e duality in this context gives a canonical c,∨ an an isomorphism K (X ) ≃ K (X ) for complex G-manifolds – a priori this isomorphism depends on a choice of orientation, but complex manifolds are canonically oriented for K-theory (see [S4, Section 3] and [M , XIV]). c,∨ an Lemma 2.2 also hold for K (X ). Namely: Lemma 2.6. The properties of 18 (1) The base change formula holds with respect to a proper map p : X → Y and an open immersion f : U ⊂ Y ; (2) If V is a G-equivariant vector bundle on X of dimension n, then there is a canonical c,∨ c,∨ an an ⊕n equivalence K (P(V) ) ≃ K (X ) . M M ′ ′ (3) For any closed immersion of G-spaces i : X ֒→ X with open complement j : U ⊂ X , there is an exact triangle E (X) −→ E (X ) −→ E (U) →; BM BM BM (4) If π : V → X is a G-equivariant torsor for V, then there is a canonical equivalence E (X) → E (V ). BM BM ′ −1 Proof. The base change formula arises from the observation that if U := p (U) ⊂ X, then there is a commutative diagram of maps of one point compactifications + ′ + // X (U ) , + + // Y U and it is precisely pullback along these maps which is used to define proper pullback and pushfor- ward along an open immersion in K (−). The canonical equivalence (2) is dual to the equivalence c c ⊕n K (P(V)) ≃ K (X) of [S4, Proposition 3.9]. (3) is [S4, Proposition 2.9]. One proves (4) using M M c,∨ c,∨ an the same method as the proof of Lemma 2.2: the canonical functor K (X ) → K (V ) is the M M c,∨ c,∨ th ′ an ⊕n+1 inclusion of the 0 piece of the direct sum decomposition K (P(V )) ≃ K (X ) followed M M by restriction to the open subset V ⊂ P(V ). c,∨ ∗ an Remark 2.7. It is a priori not clear how to define a canonical pullback map f : K (Y ) → c,∨ an K (X ) for a flat map f : X → Y . That is why we have only stated the projection formula for open immersions, and it’s why in (2) and (4) above we can not state that the canonical isomorphisms c,∨ an are given by pullback. Indeed it is not immediate in (4) that the isomorphism K (X ) ≃ c,∨ an K (V ) is independent of the presentation of V as a torsor for V. If f is the restriction of a flat ′ ′ ′ map X → Y along a closed embedding Y ֒→ Y , then one could provisionally define a pullback map using the localization sequence (3), but perhaps the most uniform definition of flat pullback c,∨ for K is via the isomorphism of Theorem 2.10 below. 2.1.3. The comparison map. We now construct a comparison map between Borel-Moore homology c,∨ top b an theories ρ : K (D Coh(X/G)) → K (X ) by first constructing a comparison isomorphism G,X top an between the corresponding cohomology theories K (Perf(X/G)) → K (X ). Observe that we have a natural transformation of presheaves of spectra alg an an K (Perf(X/G× T )) → K (X × T ) which is induced by the functor of exact categories which sends an algebraic G-vector bundle to its underyling complex topological vector bundle equipped with the induced action of M (This functor is symmetric monoidal, and hence induces a map of K-theory spectra [T5, Section 5.4]). By the an an ∞ an an Lemma 2.8 below, the presheaf K (X ×(−) ) is weakly equivalent to Hom (Σ (−) , K (X )), M Sp M where Hom (•,•) denotes the internal function spectrum in the category of spectra. Composing Sp this with the comparison map above gives a map of presheaves alg ∞ an an K (Perf(X/G)⊗ O )) → Hom (Σ T , K (X )), C T Sp M where both sides are regarded simultaneously as presheaves in the G-space X and the affine scheme T . Here we have used the natural Morita equivalence Perf(X/G)⊗ O −→ Perf(X/G×T ). By the C T B3, Definition 3.13] for presheaves on the category adjunction defining the geometric realization [ 19 of affine schemes, this gives a map st an K (Perf(X/G)) → K (X ) of presheaves in the G-space X. This natural transformation of presheaves on Spaces will be an bu-linear by construction. Because K (X ) already satsifies Bott periodicity, this comparison map extends uniquely to the localization, giving our final comparison map perf top st −1 an ρ : K (Perf(X/G)) := K (Perf(X/G))⊗ bu[β ] → K (X ). bu G,X The following technical lemma was used in the construction above: Lemma 2.8. For any topological space Y and topological M-space X, we have a natural weak equivalence in Sp, Hom (Σ Y , K (X)) = K (X × Y ), Sp + M M where on the right Y is regarded as an M-space with trivial M action. Proof. We fix a universe U for forming the equivariant stable homotopy category Sp as in [LSM]. The “change of universe” functor taking an E ∈ Sp to its underlying naive M-spectrum admits a left adjoint, as does the functor from naive M-spectra to spectra which applies the M-fixed point functor level-wise. We will denote the composition of these to functors as (−) : Sp → Sp, and it therefore has a left adjoint, which we denote ι. By definition we have that K (X) := Hom (Σ X , bu ) M Sp + M where bu is the M-spectrum representing equivariant K-theory, Σ is the stabilization functor from pointed M-spaces to M-spectra, and Hom is the internal function spectrum in the symmet- Sp ric monoidal category of M-spectra [LSM, page 72].Thus by the (spectrally enhanced) adjunction and the definition of inner Hom in a symmetric monoidal category we have ∞ ∞ ∞ Hom (Σ Y , K (X)) ≃ Hom (ι(Σ Y )∧ Σ (X ), bu ) Sp + M Sp + + M ∞ ∞ The claim now follows from the natural isomorphism ι(Σ Y ) ≃ Σ (Y ) [LSM, Remark II.3.14(i)], + + where Y is regarded as an M-space with trivial M action, the fact that Σ maps smash products of pointed M-spaces to smash products of M-spectra [LSM, Remark II.3.14(iii)], and the fact that Y ∧ X ≃ (Y × X) for pointed M-spaces. + + + top an Now that we have constructed a comparison map of presheaves K (Perf(X/G)) → K (X ), c,∨ top b an we construct a comparison map for Borel-Moore homology ρ : K (D Coh(X/G)) → K (X ) G,X for a G-quasi-projective scheme X as follows: Choose a G-equivariant closed immersion i : X ֒→ Z into a smooth G-scheme Z with open complement j : U ⊂ X. Then by Lemma 2.2 and Lemma 2.6, and using the canonical equivalence Perf(Z/G) ≃ D Coh(Z/G), there is a dotted arrow completing the map of exact triangles top b top top // // // K (D Coh(X/G)) K (Perf(Z/G)) K (Perf(U/G)) (2) perf perf G,X ρ ρ G,Z G,U i j c,∨ c,∨ c,∨ an an an // // // K (X ) K (Z ) K (U ) M M M c,∨ top b an Lemma 2.9. The map ρ : K (D Coh(X/G)) → K (X ) is canonically defined up to G,X homotopy, independent of the choice of embedding X ֒→ Z, and agrees with the comparison map perf ρ if X is smooth. ρ commutes (up to homotopy) with proper pushforwards, restriction to G,X G,X open G-subschemes, the canonical equivalence between the Borel-Moore homology of X and that of a torsor for a vector bundle over X, and restriction from G-equivariant K-theory to H-equivariant K-theory for a reductive subgroup H ⊂ G such that H is the complexification of H ∩ M. 20 Proof. The fact that the map ρ is canonical up to homotopy follows from the fact that the G,X rest of the diagram is strictly commutative and the formation of homotopy fibers is functorial in the category of spectra. So, the map between homotopy fibers of the two restriction maps j is canonical, and the d´evissage isomorphism with the homotopy fiber of j induced by the Gysin map i c,∨ perf top b an is canonical for both the theories K (D Coh(−/G)) and K ((−) ). The fact that ρ ∼ ρ G,X M G,X when X is smooth follows from considering the identity embedding X ֒→ X. Independence of the choice of embedding is essentially proved in [T5], which is an extension to the equivariant setting of [T3]. For the benefit of the reader, we explain the conceptual core of argument: Define a category Emb of “virtual embeddings” whose objects are G-quasi-projective schemes and whose morphisms X Z consist of a G-equivariant closed subscheme V ֒→ Z along with a G-equivariant map V → X which can be factored as a composition of maps which are torsors for locally free sheaves. Composition is given by pullback of closed subschemes. Then in the proofs of [T5, T3], Thomason shows that given two maps X Z and X Z , there is a linear action of 1 2 n n n n G on A and maps Z A and Z A such that the two compositions X A agree. In 1 2 sm particular, the under-category Emb is filtered, as is the category Emb of virtual embeddings X/ X/ X Z, where Z is smooth. Now let X Z be a virtual embedding into a smooth Z, which corresponds to (V ֒→ Z, π : 1 top 2 an V → X). For ease of notation let E (X) = K (Perf(X/G)) and let E (X) = K (X ) for any smooth G-quasi-projective scheme X. Then we define i i i i E (X Z) := E (X) = fib(E (Z) → E (Z − V )), i = 1, 2. perf Note that by functoriality of the homotopy fiber the comparison maps on cohomology theories ρ G,Z perf and ρ induce a map G,Z−V 1 2 ρ(X Z) : E (X Z) → E (X Z). sm i i We claim that the assignment E (X Z) can be extended to a functor E : Emb → Ho(Sp) X/ sm in which all arrows in Emb map to isomorphisms, and the comparison map ρ(X Z) defines X/ 1 2 a natural transformation of functors E → E . Indeed given a composition X Z → Z , which 1 2 sm we regard as a map Z Z in Emb , consider the following diagram, where the central square 1 2 X/ is cartesian: V . ⑦ ❅ ⑦ ❅ ⑦ ❅ π ⑦ ❅ i ⑦ ❅ ~~⑦ V V 1  p 2  p ❆ ❆ ❆ ❆ ⑦ ⑥ ❆ ❆ ⑦ ⑥ ❆ ❆ ⑦ ⑥ ❆ ⑥ ❆ ❆ ⑥ ❆ ❆ ⑥ ❆ ⑦ ❆ ⑥ ❆ ~~⑦ ~~⑥ /o /o /o /o /o /o /o /o /o // /o /o /o /o /o /o /o /o /o // X Z Z 1 2 π i i i i Then we have canonical isomorphisms E (Z ) −→ E (V ) −→ E (Z ) for i = 1, 2, where the ′ ′ 1 2 2 1 V V 2 2 construction of the latter map uses Poincar´e duality (all of the G-schemes involved are smooth) to convert cohomology to homology with locally compact supports (i.e. Borel-Moore homology), and More precisely, the proof of [T5, Proposition 5.8] shows that for any G-quasi-projective X there is a map X A for some linear represenation of G. Thus it suffices to consider the case of two maps X A , i = 1, 2. Next if V → X is a composition of torsors for locally free sheaves and V ֒→ A , i = 1, 2 are two G-equivariant closed embeddings, n n n 1 2 then the proof of [T3, Lemma 4.2] works equivariantly to constuct an equivariant embeddings A ֒→ A ×A such n n n 1 2 i that the two induced embeddings V ֒→ A × A agree. Thus it suffices to show that for any two maps X A ′ ′ n n ′ n i i corresponding to two fibrations V → X, one can compose with maps A A such that if V ⊂ A corresponds i i n ′ ′ to the compositions X A , then V ≃ V over X. This follows from the proof of [T3, Proposition 4.7], which also 1 2 works equivariantly. 21 i is an isomorphism by d´evissage for the Borel-Moore theory. We leave the somewhat involved i i diagram chase to the reader to show that the isomorphisms E (X Z ) → E (X Z ) commute 1 2 with the comparison maps ρ(X Z ) and ρ(X Z ) up to homotopy. 1 2 Our comparison map ρ arises from choosing an embedding X Z where V = X and com- G,X top b 1 bining ρ(X Z) with the canonical d´evissage isomorphisms K (D Coh(X/G)) ≃ E (X Z) c,∨ an 2 and K (X ) ≃ E (X Z). We observe that using part (5) of Lemma 2.2 and part (4) of top b Lemma 2.6, for any X Z we can compose the canonical equivalence K (D Coh(X/G)) → top b top b 1 K (D Coh(V/G)) followed with the canonical d´evissage equivalence K (D Coh(V/G)) → E (Z), c,∨ and likewise for K (−). Given a map V → Z which is a composition of torsors for vector bun- top b dles, and given a closed G-subscheme X ֒→ Z, the canonical equivalences K (D Coh(Z/G)) → top b top b top b K (D Coh(V/G)) and K (D Coh(X/G)) → K (D Coh(V| /G)) commute with the pushfor- c,∨ ward maps, and likewise for K (−). It follows that for any composition X Z Z , the 1 2 top b 1 1 composition of the canonical maps K (D Coh(X/G)) → E (X Z ) → E (X Z ) is homo- 1 2 c,dual top b 1 topic to the canonical map K (D Coh(X/G)) → E (X Z ), and the same is true for K (−). Combining these facts, we see that the comparison map ρ defined via the canonical equivalences G,X for any smooth virtual embedding X Z ρ(X Z) c,∨ top b 1 2 an K (D Coh(X/G)) ≃ E (X Z) −−−−−→ E (X Z) ≃ K (X ) sm is independed of the choice of X Z, because the category Emb is filtered. X/ c,∨ top b an Now that we have shown that the comparison map ρ : K (D Coh(X/G)) → K (X ) is G,X independent, up to homotopy, of the choice of embedding X Z, it is fairly straightforward to show that ρ commutes with restriction to an open subset and restriction of requivariance from G G,X to a reductive subgroup H ⊂ G which is the complexification of H∩M, because the corresponding perf claims hold for the comparison map ρ for smooth G-schemes. The fact that ρ commutes G,X G,X c,∨ c,∨ top b top b an an with the equivalences K (D Coh(X/G)) ≃ K (D Coh(V/G)) and K (X ) ≃ K (V ) for M M a map V → X which is a torsor for a vector bundle follows from the more general description of ρ above, because any closed embedding into a smooth G-scheme V ֒→ Z defines both a virtual G,X top b 1 embedding X Z and V Z, and the canonical equivalence K (D Coh(X/G)) ≃ E (X Z) top b top b 1 is the composition of the equivalences K (D Coh(X/G)) ≃ K (D Coh(V/G)) ≃ E (V Z) c,∨ and likewise for K (−). Finally, one shows that ρ commutes pushforward along a proper map p : X → Y by choosing G,X a closed immersion Y ֒→ Z into a smooth G-scheme Z and a projective bundle P(V) over Z such that p factors through a closed immersion X ֒→ Y . One shows that ρ commutes with pushforward perf along the projection P(V| ) → Y using d´evissage and the corresponding fact for ρ for smooth G,Z G-schemes. It therefore suffices to show that ρ commutes with closed immersions. For any closed immersion of G-schemes X ֒→ X , we can choose a closed immersion into a smooth G-scheme 1 2 X ֒→ Z. The fact that ρ commutes with pushforward along X ֒→ X follows from the fact 2 1 2 i i i that i is part of a canonical map of exact triangles from E (X ) → E (Z) → E (Z − X ) → to ∗ 1 1 i i i E (X ) → E (Z) → E (Z − X ) for both theories i = 1, 2. 2 2 Note that in Lemma 2.9 we have passed from strict presheaves on the category of algebraic G- spaces with values in spectra, to presheaves on the category of quasi-projective G-schemes with values in the homotopy category Ho(Sp). Presumably neither of these relaxations are necessary, but they simplify our discussion. c,∨ top b an Theorem 2.10. The natural map ρ : K (D Coh(X/G)) −→ K (X ) construced in (2) and G,X Lemma 2.9 is a weak equivalence. 22 perf Proof of Theorem 2.10. By the commutativity of the diagram (2) and the fact that ρ ∼ ρ G,X G,X when X is smooth, it suffices to show that ρ is a weak equivalence when X is smooth G-quasi- G,X projective scheme. Let T ⊂ G be a maximal torus which is the complexification of a compact maximal torus T = T ∩ M, and let B ⊂ G be a Borel subgroup containing T . For both G- equivariant cohomology theories, pullback along the map G × X → X is an injection which is canonically split by the pushforward map. The map which forgets from G-equivariance to T - equivariance, then restricts along the T -equivariant map {1}×X → G× X induces an equivalences K (G× X) → K (X) because topologically G× X = M × X and the same maps induces M B T B T c c top top an equivalence K (Perf(G× X/G)) → K (Perf(X/T )) because X/T → G× X/G ≃ X/B is B B perf a composition of torsors for line bundles on X/B. The comparison map ρ commutes with the G,X operations of pullback, proper pushforward, and restriction to subgroups, so ρ is compatible with top top an an the splitting of the inclusions K (Perf(X/G)) ֒→ K (Perf(X/T )) and K (X ) ֒→ K (X ), M T so it suffices to prove the claim when G = T is a torus. ′ ′ We can stratify X/T by smooth T -schemes of the form U× (T/T ), where T ⊂ T is an algebraic subgroup and T acts trivially on U. Using the localization sequences for closed immersions in Lemma 2.2 and Lemma 2.6 and the compatibility of ρ with pushforward along closed immersions G,X and restriction to open subsets, Lemma 2.9, it suffices to prove the claim for schemes of this ′ ′ form. The fact that Perf(U × (T/T )/T ) ≃ Perf(U × BT ) ≃ Perf(U), where χ ranges over ′ top ′ the group of characters of the diagonalizable group T , implies that K (Perf(U × (T/T )/T )) ≃ top ′ K (Perf(U)). There is an analogous decomposition of K (U×T/T ), and ρ ′ respects T T,U×T/T χ c this direct sum decomposition because the summands are the essential image of pullback along the ′ ′ map U × T/T → U followed by tensoring with the various characters of T . We note that when the group is trivial, our comparison map agrees with the one constructed in [B3, Proposition 4.32], therefore it is an equivalence, and the claim follows. Remark 2.11. If G is not necessarily reductive, then one can choose a decomposition G = U ⋊ H, where H is reductive and U is a connected unipotent group. As in the first step in the proof T5, Theorem 5.9], one shows that the map of stacks X/H → X/G can be factored as a of [ top sequence of torsors for vector bundles, so the canonical restriction map K (Perf(X/G)) → top K (Perf(X/H)) is an equivalence by Lemma 2.2. Combining this with the previous theorem shows that for a maximal compact subgroup M ⊂ H ⊂ G, the topologization functor is an equiva- top an lence K (Perf(X/G)) → K (X ) as presheaves of spectra on Sm , and we have a comparison M G top b c∨ an isomorphism ρ : K (D Coh(X/G)) → K (X ). G,X 2.2. The case of smooth Deligne-Mumford stacks. Here we provide an explicit computation of the periodic cyclic homology of Perf(X) for a smooth Deligne-Mumford stack of finite type over C and study its noncommutative Hodge theory when it is proper. The results of this section are likely known to experts. Given a smooth scheme U, we can consider its de Rham complex, 0 → O → Ω → ··· , a −p complex of vector spaces. We can regard this as a Λ-module Ω (U) by defining Ω (U) := Ω and • p letting B act via the de Rham differential. Even though the Λ-module structure is not O -linear, it still defines a sheaf of Λ-modules on the small site X for any smooth DM stack X. We define et the de Rham cohomology of a smooth Deligne-Mumford stack X to be the Λ-module H (X) := RΓ(X , Ω ). dR et • There are several other ways to present H (X). First note that we can equivalently restrict to dR aff the sub-site of ´etale maps U → X for which U is affine, which we denote X , because it has an et equivalent topos of sheaves, i.e. the canonical map is an equivalence aff H (X) −→ RΓ(X , Ω ). dR • et 23 aff We can consider the sheaf of Λ-modules on X given by U 7→ C (O ), the Hochschild com- • U et plex of coordinate algebra. This admits a canonical map to the presheaf of Λ-modules given by U 7→ C (Perf(U)). Likewise, for any smooth affine scheme the map HKR is a map of Λ-modules C (O ) → Ω (U) and compatible with ´etale base change, so they induce maps of presheaves on • U • aff et Lemma 2.12. The canonical maps aff aff aff RΓ(X , Ω ) ← RΓ(X , C (O )) → RΓ(X , C (Perf(−))) • • − • et et et are all equivalences of Λ-modules. Proof. These maps are all equivalences for affine U at the level of underlying complexes. The result follows formally from the fact that a map of Λ-modules is an equivalence if and only if the underlying map of complexes is an equivalence, and the forgetful functor taking a Λ-module to its underlying complex commutes with limits, hence commutes with RΓ. The following is due to Toen, and essentially follows the argument of [T6] in the case of algebraic K-theory. We will need to use both the derived inertia stack I and its underlying classical stack cl I ⊂ I . cl Proposition 2.13 (Toen, unpublished). Let X be a smooth Deligne-Mumford stack, and let I denote its classical inertia stack. There is a natural isomorphism of Λ-modules C (Perf(X)) → cl H (I ). dR The idea of the proof is to show that the formation of both complexes is local in the etale topology over the coarse moduli space of X, so one can reduce to the case of a global quotient. Thus a key observation is that the formation of the derived inertia stack I is ´etale local. Lemma 2.14. Let X → X be a map from a stack to a separated algebraic space, and let U → U be the base change along an etale map U → X. Then I ≃ I × U. U X X Proof. This can be seen for the derived inertia stack from a functor-of-points definition of I . We let U(T ) denote the ∞-groupoid of maps from T to U for a derived affine scheme T . I (T ) = U(T )× U(T ) U(T )×U(T ) ∼ Map(S ,U(T )) 1 1 ∼ Map(S ,X(T ))× 1 Map(S , U(T )) Map(S ,X(T )) So in order to show that I (T ) ≃ I (T ) × U(T ), it will suffice to show that I ≃ I × U U X X(T ) U X X in the derived sense. Consider the following diagram, in which each square is Cartesian and the vertical arrows are closed immersions: ** // // // U U × U Γ X // // U × U U × X X × X Here Γ denotes the graph of the morphism U → X. Then by definition I is the derived self intersection of the closed subspace X → X × X, so in order to prove the claim it will suffice to show that Γ× Γ is isomorphic to I as a derived scheme over U. U×X The map U → Γ is an isomorphism on underlying classical algebraic spaces, and it follows from the fact that U → U × U is an etale closed immersion of closed substacks of U × U that the induced map I → Γ× Γ induces an isomorphism on cotangent complexes as well, hence it is U U×X an isomorphism. 24 cl Proof of Proposition 2.13. The pullback functor along the projection I → X induces a map cl cl C (Perf(X)) → C (Perf(I )). For any ´etale U/I , the pullback functor induces a natural map • • X X cl C (Perf(I )) → C (Perf(U)). Thus we get a map of presheaves of Λ-modules • • cl aff cl C (Perf(X)) → RΓ((I ) , C (Perf(−))) ≃ H (I ). • • dR X et X Note that if p : X → X is the coarse moduli space of X, then the map constructed above is functorial with respect to pullback along maps U → X. We claim that C (Perf(X)), regarded as a presheaf over X, has ´etale descent. Indeed, consider any ´etale map U → X, and let U = X × U. Because the derived category of U is compactly generated [HR], we can identify C (Perf(U)) ≃ RΓ(U, Δ Δ (O )) ≃ RΓ(U, (p ) O ), • ∗ U U ∗ I where Δ : U → U× U is the diagonal, p : U → U is the base change of p, and O is the structure U I sheaf of the derived inertia stack, regarded as a finite algebra over O . In the previous lemma, we saw that the formation of I commutes with ´etale base change, so this combined with the projection formula implies that RΓ(U, p O ) ≃ RΓ(U, p (O )| ), functorially in U. The presheaf ∗ I ∗ I U U X U/X 7→ RΓ(U, p (O )| ) has ´etale descent, so U 7→ C (Perf(U)) does as well. ∗ I U • cl Thus in order to show that C (Perf(X)) → H (I ) is an equivalence, it suffices to verify this • dR after base change to an ´etale cover of X. We can find such a U → X such that U = X × U is a global quotient of a scheme by a finite group action. In that case, the result is shown in [B1, Proposition 4]. Finally after applying the Tate construction, i.e. passing to periodic cyclic homology, we can an compare this to the cohomology of |X |, the geometric realization of the underlying topological stack (in the analytic topology) associated to X [N2], as well as the cohomology of a coarse moduli space X → X. Lemma 2.15. Let X be a Noetherian separated DM stack of finite type over a Noetherian base scheme. Assume that X has finite dimension. Then X has finite ´etale cohomological dimension with Q-linear coefficients, and the functor RΓ(X ,−) commutes with filtered colimits. et Proof. We first claim that the pushforward along the projection to the coarse moduli space p : X → X is exact. Indeed this can be checked ´etale locally on X, and so we may assume that X is a global quotient U/G, where G is a finite group. One can factor p as U/G → X × BG → X – pushforward along the first is exact by [S6, Tag 03QP], and the second is exact because we are using characteristic 0 coefficients. It now suffices to prove the claim when X = X is a Noetherian separated algebraic space of finite type over a Noetherian base scheme. In this case, we can apply the induction principle of [S6, Tag 08GP] and the fact that ´etale cohomology takes elementary excision squares to homotopy cartesian squares to reduce to the case of affine schemes. In this case, the result follows from the fact that derived global sections of characteristic 0 sheaves on a Noetherian scheme in the ´etale topology argees with that in the Nisnevich topology, and the Nisnevich topology has cohomological dimension ≤ d. Finally, the implication that finite cohomological dimension implies commutation with filtered colimits in the unbounded derived category is [CD, Lemma 1.1.7]. Lemma 2.16. There are natural isomorphisms Tate ∗ an ∗ H (X) ≃ C (|X |;Q)⊗ C((u)) ≃ C (X;Q)⊗ C((u)) dR Q Q sing sing 25 Proof. The de Rham isomorphism gives a canonical isomorphism of pre-sheaves of C((u))-modules aff Tate ∗ an on X between U 7→ (Ω (U)) and U 7→ C (U ;C)((u)), so we have a canonical isomorphism et sing ∗ an aff Tate C (|Y |;C)⊗ C((u)) ≃ RΓ(X , Ω (−) ). sing et It therefore suffices to show that the Tate construction commutes with taking derived global sections for the sheaf of Λ-modules Ω . For this we observe that the functor M 7→ M commutes with Tate S homotopy limits, and hence with derived global sections, and M is the filtered colimit of M → 1 1 S S M [2] → M [4] → ··· , so its formation commutes with RΓ by the previous lemma. ∗ an ∗ an Finally, one can check that the pullback map C (Y ;Q) → C (|Y |;Q) is an equivalence sing sing an an locally in the analytic topology on Y . Locally Y is isomorphic to a global quotient of a scheme by a finite group, for which the fact is well-known. 2.3. Equivariant K-theory and periodic cyclic homology. For a dg-category, C, it is natural per top to ask if the Chern character induces an equivalence K (C)⊗C → C (C). This is referred to as the lattice conjecture in [B3], where it is conjectured to hold for all smooth and proper dg-categories. Here we observe some situations in which the lattice conjecture holds, even for categories which are not smooth and proper. Theorem 2.17 (Lattice conjecture for smooth quotient stacks). Let G be an algebraic group acting on a smooth quasi-projective scheme X. If X/G admits a semi-complete KN stratifica- top tion (Definition 1.1), then the Chern character induces an equivalence K (Perf(X/G)) ⊗ C → per C (Perf(X/G)). Lemma 2.18. Let X be a smooth Deligne-Mumford stack, and let i : Z ֒→ X be a smooth closed substack. Then the pushforward functor fits into a fiber sequence per ∗ per per C (Perf(Z)) −→ C (Perf(X)) −→ C (Perf(X− Z)). • • • Proof. This follows from Proposition 2.13, combined with Lemma 2.16 and the usual Gysin se- quence for the regular embedding of inertia stacks I ֒→ I . Z X per top Proof of Theorem 2.17. Because K (−) ⊗ C and C (−) are both additive invariants, proving that the natural transformation top per K (Perf(X/G))⊗ C → C (Perf(X/G)) is an equivalence for smooth projective-over-affine X and reductive G reduces to the case where X/G is Deligne-Mumford by Theorem 1.8. Note that the only point in the proof of Lemma 2.2 which does not immediately apply to an arbi- trary additive invariant is the localization sequence for a closed immersion. Therefore Lemma 2.18 per implies that Lemma 2.2 applies to the presheaf C (Perf(−)), because the only stacks that appear in the proof are DM. We can now imitate the proof of Theorem 2.10: Perf(X/G) is a retract of Perf(X/B), and per top Perf(X/B) → Perf(X/T ) induces an equivalence for both invariants K (•) and C (•), by Lemma 2.2. Thus it suffices to consider smooth DM stacks of the form X/T . Any such stack admits a stratification by smooth stacks of the form U × BΓ for some finite group Γ, and by Lemma 2.18 it suffices to prove the theorem for such stacks. Thus Ch⊗C is an equivalence because it is an equivalence for smooth schemes and Perf(U × BΓ) ≃ Perf(U), the sum ranging over characters of Γ. Corollary 2.19 (Lattice conjecture for smooth DM stacks). Let X be a smooth Deligne-Mumford per top stack. Then the Chern character induces an equivalence K (Perf(X))⊗ C → C (Perf(X)). All of the singular complexes we will encounter have finite dimensional total cohomology, so M((u)) ≃ M⊗ C((u)). 26 Proof. We have established a localization sequence for closed immersions of smooth DM stacks for top K (Perf(−)) in Lemma 2.2. We do not know if the localizing invariant of dg-categories HP(−) satisfies d´evissage in the sense of Lemma 2.2 for arbitrary closed immersions of stacks, but the localization sequence for closed immersions of smooth DM stacks follows from the comparison results Proposition 2.13 and Lemma 2.16 and the corresponding fact for singular cohomology. We therefore have a map of fiber sequences for any closed immersion of smooth DM stacks over C, top b top b top b // // K (D Coh(Z))⊗ C K (D Coh(X))⊗ C K (D Coh(U))⊗ C . Ch Ch Ch b b b // // HP(D Coh(Z)) HP(D Coh(X)) HP(D Coh(U)) From [LMB, Corollaire 6.1.1] every smooth DM stack of finte type admits a stratification by locally closed substacks which are quotients of a smooth affine scheme by a finite group. The corollary follows by applying Theorem 2.17 and the fiber sequence above inductively to this stratification. 2.4. Hodge structure on equivariant K-theory. We can now prove the final result of this paper, the construction of a pure Hodge structure on the equivariant K-theory. What we mean by a pure Hodge structure on a spectrum E in this case is simply a Hodge structure on the homotopy groups of that spectrum π (E), i.e. for each n a weight n Hodge structure on π (E) is a descending ∗ n filtration of π (E)⊗ C such that n+1−p π (E)⊗ C = F π (E)⊗ C⊕ F π (E)⊗ C,∀p n n ∗ Theorem 2.20. Let X be a smooth quasi-projective C-scheme, let M be a compact Lie group whose complexification G acts on X. Then if X/G admits a complete KN stratification, the Chern character isomorphism an per K (X )⊗ C → C (Perf(X/G)) combined with the noncommutative Hodge-de Rham sequence induces a pure Hodge structure of n an weight n on K (X ) with a canonical isomorphism n an n−2p gr (K (X )⊗ C) ≃ H RΓ(I ,O ), X I M X Hodge n an where I denotes the derived inertia stack of X := X/G. The Hodge filtration on K (X ) is compatible with pullback maps, and in particular it is a filtration of Rep(M)-modules. Remark 2.21. As we will see in the proof, this claim also holds for arbitrary smooth and proper DM stacks over C, without requiring that X is a global quotient. Proof. The degeneration property follows from Corollary 1.23, and we have a Chern character isomorphism from Theorem 2.10 combined with Theorem 2.17, so all we have to do is check that per the filtration on C (Perf(X/G)) coming from the HdR spectral sequence combined with the rational structure coming from the Chern character defines a weight n pure Hodge structure on top π (K (Perf(X/G))⊗Q. This claim is closed under arbitrary direct sums and summands in M , −n k so by Theorem 1.8 it suffices to prove this claim for smooth and proper DM stacks which are global quotients of a G-quasi-projective scheme by a reductive group G. cl For a smooth and proper DM stack, Proposition 2.13 gives an isomorphism H (I ) ≃ C (Perf(X)) dR • cl of Λ-modules. Note that I is itself a smooth and proper DM stack, and for any smooth DM stack Tate Y the complex H (Y) is canonically equivalent to the usual de Rham complex of [S2] tensored dR with C((u)), RΓ Y, 0 → O → Ω → ··· ⊗ C((u)). Y C 27 However, the usual Hodge filtration differs slightly from the noncommutative one. We have a canonical isomorphism n Tate l H (H (Y) ) ≃ H (Y;C) dR l≡n mod 2 p • n Tate i gr F H (H (Y) ) ≃ RΓ(Y, Ω [i− 2p])) dR nc Y Because the cyclic complex H (Y) has the degeneration property, we may commute taking coho- dR n p mology H and taking associated graded gr , so we have p • n Tate n+i−2p i gr F H (H (Y) ) ≃ H (Y, Ω ) dR nc Y l n Tate l−p Thus on each direct summand H (Y;C) of H (H (Y) ), the subquotient H (Y, Ω ) shows dR ′ ′ ′′ ′′ up in F if and only if l − p = n + p − 2p for some p ≥ p. In other words the subquotients nc ′ l−n appearing are those for which p ≥ p + . It follows that under the direct sum decomposition above we have M l−n p+ p n Tate 2 l F H (H (Y) ) ≃ F H (Y;C) dR nc classical l≡n mod 2 n Tate l Thus under the isomorphism H (H (Y) ) ≃ H (Y;Q) ⊗ C of Lemma 2.16, the dR l≡n mod 2 l l−n noncommutative Hodge filtration corresponds to the Hodge filtration on H (Y;Q)h i. l≡n mod 2 n cl Tate We claim that this rational structure on H (H (I ) ) agrees with the one induced by the dR top cl Tate equivalence K (Perf(X)) ⊗ C ≃ H (I ) of Theorem 2.17 and Lemma 2.16, so that we have dR an isomorphism of Hodge structures l − n top l cl π (K (Perf(X)))⊗ Q ≃ H (I ;Q)h i. (3) −n l≡n mod 2 th The Hodge structure on the l rational cohomology of the de Rham complex of a smooth DM stack top has weight l (see [S7]), so it would follow that π (K (Perf(X))) ⊗ Q has a Hodge structure of −n weight n. cl cl Tate For the claim about the rational structure of H (I ), note that the isomorphism H (I ) ≃ dR dR X X top K (Perf(X))⊗ C results from applying the derived global sections functor to isomorphic sheaves cl on the ´etale site of I , cl aff top cl aff per RΓ((I ) , K (Perf(−))⊗ C) ≃ RΓ((I ) , C (Perf(−))) et et • X X cl aff ∗ an ≃ RΓ((I ) , C ((−) ;C)((u))). X et sing But according to [B3, Proposition 4.32], the noncommutative Chern character for smooth C- per schemes factors through the twisted Chern character under the natural equivalence C (Perf(X)) ≃ Tate ∗ H (X) ≃ C (X;C((u))). It follows that the isomorphism above is the complexification of a dR sing cl aff map of presheaves of Q-complexes on (I ) et top ∗ an K (Perf(−))⊗ Q → C ((−) ;Q)⊗ Q(( )), sing 2πi n cl Tate which is also a level-wise weak equivalence. Thus the rational structure on H (H (I ) ) agrees dR with that of the Hodge structure of Equation 3. Remark 2.22. For any of the quotient stacks appearing in Amplification 1.25, the theorem above still holds for D Coh(X) with the same proof, with the exception of the explicit computation of per p n b gr H (C (D Coh(X))) when X is not smooth. In particular we have: 28 per an b • A canonical isomorphism K (X )⊗ C → C (D Coh(X/G)) which factors through the M • per top b b complexification of the Chern character K (D Coh(X/G)) → C (D Coh(X/G)); • The degeneration property for D Coh(X/G); and top b • A pure Hodge structure of weight n on π K (D Coh(X)) coming from the degeneration −n of the noncommutative Hodge-de Rham sequence which is Rep(G)-linear. 3. Hodge-de Rham degeneration for singularity categories In this section we extend our methods to establish the degeneration property for certain “dg- categories of singularities” MF(X/G, W ) associated to an equivariant Landau-Ginzburg model, i.e., a smooth G-variety X and a G-invariant function W : X → A . The notation MF is more frequently used for categories of matrix factorizations, which are equivalent to singularity categories for LG models on smooth schemes [O2]. We have chosen to use MF to denote singularity categories for consistency with [P]. The categories MF(X/G, W ) will be Z/2Z-graded, and we will need a suitable ∞-categorical model to work with these. The ∞-category of (essentially) small idempotent complete stable ∞- perf perf categories, Cat , admits a symmetric monoidal structure [BGT, Section 3.1]. For A,B ∈ Cat , ∞ ∞ ˙ b A⊗B is the category of compact objects in Ind(A)⊗ Ind(B), which is idempotent complete. perf For any E -algebra R, Perf(R) is canonically a commutative algebra object in Cat . We let sm ⊗ perf LinCat = (Perf(R) )-Mod(Cat ) R ∞ ⊗ sm denote the ∞-category of Perf(R) -module objects. LinCat is equivalent, via the ind-completion functor, to the ∞-category of (R-Mod) -module objects in the ∞-category of compactly generated presentable stable infinity categories with functors which preserve colimits and compact objects, ⊗ sm as in [L2, Definition 6.2]. Because (R-Mod) is a commutative algebra object, LinCat has a canonical symmetric monoidal structure [L3, Thm. 4.5.2.1]. In addition, if R → R is a map of E - ⊗ ′ ⊗ algebras, the tensor product induces a map of commutative algebra objects Perf(R) → Perf(R ) , and this induces a symmetric monoidal pullback functor [L3, Thm. 4.5.3.1] which we denote ′ sm sm (−)⊗ R : LinCat → LinCat R R ⊗ ′ ⊗ These constructions work just as well with Perf(R) and Perf(R ) replaced by any other essentially small stable idempotent complete symmetric monoidal ∞-categories. sm Any ∞-category C in LinCat is canonically enriched over R-Mod, i.e. RHom(E, F) ∈ R-Mod sm for any E, F ∈ C, via an inner-Hom construction. In fact, by [C], LinCat is equivalent to the ∞-category of categories enriched in R-module spectra. Regarding k((β)), where β is variable of homological degree −2, as an E -algebra via the forgetful functor from dg-algebras to E -algebras, ∞ ∞ sm this identifies LinCat with the ∞-category of categories enriched over dg-k((β))-modules, or k((β)) sm Z/2Z-graded dg-categories. This justifies using LinCat as our model for Z/2Z-graded dg- k((β)) categories. The main results of this section, Theorem 3.10 and Proposition 3.17, will establish a k((β))-linear version of the degeneration property for some categories of singularities on quotient stacks. 3.1. Preliminaries on categories of singularities on stacks. There have been several concrete approaches to developing the general theory of singularity categories [EP,LP,O2]. We will mostly use the perspective of [P], extended more recently in [BRTV], which is more abstract but has the advantage of allowing one to deduce results about singularity categories directly from the analogous statement for derived categories of coherent sheaves (see Lemma 3.9). We will summarize the main definitions and lemmas we will use from [P]. As elsewhere in the paper k denotes a field of characteristic zero. 29 Definition 3.1. A Landau-Ginzburg (LG) model is a pair (X, W ), where X is a smooth finite type k-stack such that the automorphism groups of its geometric points are affine and W is a morphism W : X → A . In particular, X is a QCA stack over k in the sense of [DG]. Our primary examples of interest will be quotient stacks X := X/G over k. By generic smoothness, if W is non-constant on every component of X, then W has only finitely many critical values in A . Throughout this paper we let Crit (W ) denote the component of the vanishing locus of dW ∈ Γ(X, Ω ) which lies set theoretically in X := X× 1 {0}. 0 A We now equip the bounded derived category of coherent sheaves on the zero fiber, D Coh(X ), with a k[[β]]-linear structure, where β is a variable of homological degree -2. This arises from a 1 b homotopical S -action on the category D Coh (X), in the terminology of [P], which concretely refers to a natural action of H (S ; k) ≃ Λ on the Hom-complexes of the category. The formal variable β arises via the same construction which leads to the formal variable u acting on C (A), but we use different variable names to avoid confusion between these two S -actions, especially when we discuss the k((β))-linear negative cyclic homology below. Spec(Λ) admits the structure of a derived group scheme, so the ∞-category D Coh(Λ), as well as its ind-completion IndCoh(Λ), admits a symmetric monoidal structure given by the convolution product “◦”: Given F, G ∈ D Coh(Λ), F◦G := m (F⊠G), where m : Spec(Λ)×Spec(Λ) → Spec(Λ) is the group multiplication. So the underlying complex of F ◦G is F ⊗ G, with Λ-module structure given by letting B act on homogeneous elements by |m| B (m⊗ n) := B (m)⊗ n + (−1) m⊗ B (n). F⊗G F G The following is an enrichment of standard Koszul duality results. Lemma 3.2. The functor D Coh(Λ) → Perf(k[[β]]) V 7→ V := RHom (k, V ) extends to a symmetric monoidal equivalence, leading to a symmetric monoidal equivalence ⊗ ⊗ IndCoh(Λ) = (k[[β]]-Mod) Proof. This is [P, Proposition 3.1.4], see also [BRTV, Remark 2.38, Lemma 2.39]. The proof of Lemma 3.2 relies on an elementary but important observation. Let (V, d) be a complex with a Λ-action. There is a quasi-isomorphism of complexes V = (V [[β]], d + βB). Formal completion does not commute with the formation of tensor products of complexes, but the formation of the complex (V [β], d +βB) does commute with forming tensor products of complexes. So the crux of the proof of Lemma 3.2 is the following: Lemma 3.3. The natural inclusion of complexes (V [β], d + βB) → (V [[β]], d + βB) is a quasi-isomorphism whenever (V, d) is homologically-bounded above. Proof. By definition (V [[β]], d+βB) is the inverse limit of the complexes (V [β]/(β ), d+βB). If V is homologically bounded above, then for any m the canonical map (V [β], d + βB) → (V [β]/(β ), d + βB) induces an isomorphism in homology of degree ≥ m for n ≫ 0, which implies the claim. 30 As described in [P, Construction 3.1.5], [BRTV, Remark 2.38], the stack X admits an action by the derived group scheme Spec(Λ) which defines the upper horizontal arrow in the cartesian square: // X × Spec(Λ) X (4) // X X The action (4) can be described ([P, Remark 3.1.7]) concretely by a cosimplicial, commutative dg-O algebra whose cosimplicial degree n piece is given by A := hO [B , B ,··· , B ], dB = X n X X 1 n i 0, dB = Wi, where each of the variables B have degree one. X i Definition 3.4. We define PreMF(X, W ) := D Coh(X ) with the additional k[[β]]-linear structure b ⊗ b induced by Lemma 3.2 and the D Coh(Λ) -module structure on D Coh(X ) induced by the action (4). It is useful to note that the k[[β]]-linear structure has a concrete dg-model which is described in [P] and which we now recall. Observe that O = A := (O [B ], dB = W ), where B is a variable X X X X X of degree one. Pushforward defines a canonical equivalence: b ⊗ D Coh(X ) = A-Mod(Perf(X) ), where the right hand side denotes the category of coherent A-modules. There are natural adjoint functors: ⊗ b i : A-Mod(Perf(X) ) → D Coh(X) ∗ b ⊗ i : D Coh(X) → A-Mod(Perf(X) ) given by forgetting the A-module structure and tensoring with A, respectively. Now given two dg-A-modules M, N whose underlying complex of O -modules has bounded coherent homology, the Hom-complex Hom (i M, i N) inherits a Λ-module structure given by X ∗ ∗ |φ| B : φ 7→ B ◦ φ− (−1) φ◦ B (5) X X The following proposition shows that this Λ-module structure is enough to recover Homs as A- modules: Lemma 3.5 ([P, Section 3.3]; see also [P, Proposition 3.2.1]). Given objects M, N ∈ A-Mod(Perf(X) ), S 1 we have an equivalence Hom (i M, i N) = Hom (M, N), where the S -action is given by (5). X ∗ ∗ A Proof. For any object M ∈ A-Mod(Perf(X) ), we can construct a complex of B -modules, M[B], where as usual B is a variable of degree 1 with B = 0 and the action of B is given by (B ) +B, X X M where left multiplication by B on M[B] anti-commutes with the action of (B ) . There is an X M adjunction: Hom (i M, i N) = Hom (M[B], N) X ∗ ∗ A ′ ′ |φ| ′ φ(m) 7→ φ(m + Bm ) := φ(m− B · m ) + (−1) B φ(m ) (6) X X ψ(m) := ψ(m + B · 0) ←[ ψ(m + Bm ) |φ| The Λ = k[B]-module structure, which on the left hand side is given by B : φ 7→ B ◦φ−(−1) φ◦B , X X corresponds under this isomorphism to the Λ-module structure given by B : φ(−) 7→ −φ(B · (−)). 12 ∗ More conceptually, we have that M[B] is isomorphic to i i M = A⊗ M. Both are isomorphic to M ⊕M[1] as ∗ O complexes of O -modules, and the isomorphism A ⊗ M → M[B] which intertwines the action of B is (m, m ) 7→ X O X ′ ′ (m + B m , m ). 31 We then extend M[B] to a resolution of M as an A-module by forming the complex // (M[B]⊗ (k((β))/βk[[β]]); d = d − Bβ · (−)) M (7) k M −n B,β 7→0 where β is a variable of homological degree −2, which we refer to as the Koszul-Tate resolution of M. Using Equations (6) and (7), we have that Hom (M, N) can be computed as ∼ ∼ Hom (M[B]⊗ (k((β))/βk[[β]]), N) = Hom (M[B], N)[[β]] = Hom (i M, i N) , ∗ ∗ A k A X where it is evident that the differential on the first term agrees with the differential used to compute the invariants for the S -action defined in (5). The natural k[[β]]-linear structure on the complex Hom (i M, i N) provides an explicit model X ∗ ∗ for the k[[β]]-linear structure from Definition 3.4. Note that in (7), we constructed a canonical quasi-isomorphism of A-modules M = M[B] ⊗ (k((β))/βk[[β]]), where the latter has an explicit action by k[[β]]. Under the resulting quasi-isomorphism Hom (M, N) ≃ Hom (M[B]⊗ (k((β))/βk[[β]]), N), A A k the action of β corresponds to φ(−) 7→ φ(β · (−)). The fact that composition is k[[β]]-linear fol- lows from the elementary calculation that composition Hom (i M, i N) ⊗ Hom (i N, i P) → ∗ ∗ ∗ ∗ X k X Hom (i M, i P) is Λ-linear, and Lemma 3.2. ∗ ∗ With all of this in place, we turn to defining our main objects of interest, the categories of singularities: Definition 3.6. We define the category MF(X, W ) to be MF(X, W ) := PreMF(X, W )⊗ k((β)). k[[β]] This definition is justified by the following lemma: Lemma 3.7 ([P, Proposition 3.4.1]). MF(X, W ) is a dg-enhancement of the idempotent completion of the triangulated category 0 b 0 H (D Coh(X ))/H (Perf(X )). 0 0 Proof. Let M ∈ D Coh(X ). The lemma can be easily reduced to the following assertion: M ∈ Perf(X ) iff β = 0 ⊂ Hom(M, M) for large enough n. To prove this assertion, recall that M is perfect iff it is compact in QC(X ), because X is QCA [DG, Cor. 1.4.3]. 0 0 7) to simply multi- On one hand β ∈ Hom (M, M) corresponds under the quasi-isomorphism ( plying by β . We observe that the kernel of the surjective map n −n−1 0 ker (β : M[B]⊗ (k((β))/βk[[β]]) → M[B]⊗ (k((β))/βk[[β]])) = M[B]· β ⊕···⊕ M[B]· β , is a compact A-module because the associated graded of the β-adic filtration is a direct sum of finitely many copies of the compact A-module M[B] ≃ A⊗ M. Giving a null-homotopy of β , which is equivalent to giving a null homotopy of the composition β ◦ id , is equivalent to giving a factorization of id through the subcomplex ker(β ). In this case M is a homotopy M[B]⊗(k((β))/βk[[β]]) retract of the compact A-module ker(β ) and is thus compact. Conversely, if M[B]⊗(k((β))/βk[[β]]) = ker(β ) is a compact A-module, the identity morphism factors through some ker(β ) for some n. This proves the lemma. There is another point of view on the k[[β]] linear structure which will be useful below. Let (X, W ) be an LG-model. According to [BZNP, Theorem 1.1.3], there is an equivalence of categories b ex D Coh(X ) Fun (Perf(k), Perf(X)) = 1 ⊗ Perf(A ) 32 b ⊗ It is not difficult to check that the D Coh(Λ) -module structure on the left-hand side of this ex equivalence corresponds to the natural Fun (Perf(k), Perf(k))-module structure on the right- 1 ⊗ Perf(A ) hand side. If A is a module category for some symmetric monoidal ∞-category C , and A = hA ; i ∈ Ii is a possibly infinite semiorthogonal decomposition indexed by a totally ordered set I, we say that the semiorthogonal decomposition is C -linear if C⊗A → A factors through A , in i i which case it does so uniquely up to contractible choices. ⊗ ⊗ Lemma 3.8. Let C be a symmetric monoidal stable ∞-category, and let B and A be C -module perf ⊗ ⊗ categories with B compact in C -Mod(Cat ). If A = hA ; i ∈ Ii is a C -linear semiorthogonal ex ex decomposition, then Fun (B,A ) → Fun (B,A) is a fully faithful functor, and identifying the ⊗ i ⊗ C C former with its essential image in the latter, we have a semiorthogonal decomposition ex ex Fun (B,A) = hFun (B,A ); i ∈ Ii ⊗ ⊗ i C C ⊗ ex Proof. The fact that B is compact as a C -module category allows us to commute Fun (B,−) with filtered colimits and therefore reduce to the case of a finite index set I. Then by an inductive argument it suffices to prove the claim in the case where we have a two term semiorthogonal decomposition A = hA ,A i. If we let ι : A ֒→ A denote the inclusion, and we let ι (respectively 0 1 i i ι ) denote the right (respectively left) adjoint whose existence is guaranteed by the semiorthogonal ex ex decomposition. One can check that the composition functor ι ◦(−) : Fun (B,A) → Fun (B,A ) ⊗ ⊗ 1 C C ex ex is a right adjoint to the composition functor ι ◦ (−) : Fun (B,A ) → Fun (B,A), and likewise 1 ⊗ 1 ⊗ C C ι ◦ (−) is a left adjoint to ι ◦ (−). It is also straightforward to check that the canonical maps L R ι ◦ ι ◦ (−) → id and id → ι ◦ ι ◦ (−) are equivalences, and Map(ι ◦ F, ι ◦ G) is contractible for 0 1 1 0 0 1 ex ex any functors F ∈ Fun (B,A ) and G ∈ Fun (B,A ). The claim follows. ⊗ 1 ⊗ 0 C C An immediate corollary of this is the following: 1 ⊗ Lemma 3.9. Let (X, W ) be an LG-model. Suppose that Perf(X) admits a Perf(A ) -linear semiorthog- onal decomposition hA ; i ∈ Ii. Then MF(X, W ) admits a semiorthogonal decomposition by k((β))- linear subcategories D E ex MF(X, W ) = k((β))⊗ Fun (Perf(k),A ) . 1 ⊗ k[[β]] i Perf(A ) Proof. By the previous lemma applied to A = Perf(X), we obtain a semiorthogonal decomposition D E ex PreMF(X, W ) = Fun (Perf(k),A ) . The semiorthogonal decomposition is k[[β]]-linear 1 ⊗ Perf(A ) 1 ⊗ because β acts via endo-functors of Perf(k) as a Perf(A ) -module category. Finally, the localiza- tion functor from k[[β]]-linear categories to k((β))-linear categories commutes with filtered colimits, so one gets the desired semiorthogonal decomposition of MF(X, W ) := k((β))⊗ PreMF(X, W ) k[[β]] by base changing semiorthogonal decompositions. 3.2. Motivic decompositions and degeneration for MF. In this section we prove an ana- sm log of Theorem 1.8 for the k((β))-linear dg-category MF(X/G, W ). For any A ∈ LinCat , let k((β)) sm G(A) ⊂ LinCat denote the smallest ∞-subcategory containing A that is closed under splitting k((β)) sm countable semiorthogonal decompositions in the following sense: for any C ∈ LinCat which has k((β)) a Z-indexed semiorthogonal decomposition C = hC i , C ∈ G(A) if and only if C ∈ G(A) for all i i∈Z i A. Theorem 3.10. Let G be an algebraic group. Let X be a smooth G-quasiprojective k-scheme with a semi-complete KN stratification, and let W : X/G → A be a morphism. Then there is a smooth ′ 1 projective-over-affine Deligne-Mumford stack Y with a map W : Y → A such that ′ sm MF(X/G, W ) ∈ G(MF(Y, W )) ⊂ LinCat . k((β)) 33 ′ Furthermore if Perf(Crit (W )/G) is a proper dg-category, then the pair (Y, W ) can be chosen so ′ 1 that Y is a smooth variety and W : Y → A is projective. Note that by Corollary 1.7, Perf(Crit (W )/G) is a proper dg-category if and only if the induced KN stratification on Crit (W ) is complete. Remark 3.11. There is a slightly cleaner formulation of Theorem 3.10 using k((β))-linear additive noncommutative motives, analogous to Theorem 1.8. [T1] constructs additive noncommutative motives for dg-categories which are linear over a commutative ring, and [BGT] constructs additive noncommutative motives over the sphere spectrum. The methods of [BGT] appear to apply verba- tim to construct the ∞-category of additive noncommutative motives over an arbitrary E -algebra R, such as k((β)), but in the interest of space we have formulated Theorem 3.10 to avoid developing this additional machinery. Before proving the theorem, we note the following analogues of Lemma 1.4 and Lemma 1.13: Lemma 3.12. If X is a smooth G-scheme with a KN stratification, and W : X/G → A is an sm ss LG-model, then MF(X/G, W ) ∈ G MF(X /G, W )⊕ MF(Z /L , W| ) in LinCat . i i Z /L i i i k((β)) Proof. The main semiorthogonal decomposition of [HL] extends to categories of singularities by Lemma 3.9, and hence the argument of Lemma 1.4 applies verbatim to MF(X/G, W ). Lemma 3.13. Let π : Y → X be a rational morphism of finite-type k-stacks, i.e., Rπ O ≃ O , ∗ Y X 1 b and let W : X → A be a morphism. Assume that X is smooth and π preserves D Coh. Then MF(X, W ) is a semiorthogonal factor of MF(Y, W ). ∗ 1 Proof. The functors π and π are are Perf(A )-linear, and it follows from Lemma 3.9 that the semiorthogonal decomposition of Perf(Y) in the proof of Lemma 1.13 induces a semiorthogonal decomposition of MF(Y, W ). Proof of Theorem 3.10. The proof of Theorem 1.8 mostly applies verbatim, with the following sub- stitutions: Lemma 3.12 in place of Lemma 1.4; Lemma 3.13 in place of Lemma 1.13; the properness of the dg-category Perf(Crit (W )/G) in place of the properness of the dg-category Perf(U/G); and the result Proposition 3.14 below in place of [BLS, Theorem 6.6] to reduce from the case of a projective-over-affine DM stack to a quasi-projective scheme. The only parts of the proof of Theorem 1.8 which require modification in the current context are those which have to do with the properness of the dg-category Perf(Crit (W )/G), and this only affects two cases of the inductive proof. We thus re-write these cases, indicating the necessary modifications: ss ss s ss s ss Case X 6= ∅, X 6= X , and codim(X \ X , X ) ≥ 2: ss It suffices by the inductive hypothesis to prove the claim for (X /G, W ). We apply the inductive ss K6]: Let Y ⊂ X be the locus of points whose stabilizer has partial resolution procedure of [ ′ ss maximal dimension. Then Y is a smooth closed subvariety, and the blow up X := Bl (X ) has ′ ss a KN stratification induced by a relatively ample bundle such that (X ) has lower-dimensional stabilizers. Consider a point x ∈ Y , let R ⊂ G be the (reductive) stabilizer subgroup. The G-invariance of W 1 R 1 ss G 1 R implies that (dW ) ∈ (Ω ) ⊂ Ω . Because (X ) ⊂ Y the restriction map (Ω ) → ss ss ss X ,x X ,x X ,x 1 1 R 1 Ω is injective, and hence the pullback map (Ω ) → Ω is injective for any point y in the ss ′ Y,x X ,x X ,y ′ ss ss −1 fiber of x under p : X = Bl X → X . It follows that for any y ∈ p (x), (dW ) = 0 if and only Y x −1 ′ ′ if (d(W| )) = 0, and therefore Crit (W| ) = p (Crit (W )). X y 0 X 0 34 ′ Now by Lemma 3.13 and Lemma 3.12, if Z are the centers of the KN stratification of X , then ss ′ ss MF(X /G, W ) ∈ G MF((X ) /G, W| )⊕ MF(Z /L , W| ) . X i i Z /L i i Note that if Perf(Crit (W )/G) is a proper dg-category, then so is Perf(Crit (W| ′)/G), using 0 0 X criterion (2) of Lemma 1.5 and the fact that Crit (W| ) → Crit (W ) is proper. It follows 0 X 0 from Lemma 1.6 that in this case Perf(Crit (W| ′ ss)/G) and Perf(Crit (W| )/L ) are proper 0 (X ) 0 Z i dg-categories as well. The claim holds for the categories MF(Z /L , W| ) by the inductive hy- i i Z /L i i ′ ss pothesis, so it suffices to show the claim for MF((X ) /G, W| ). Because the dimension of the ′ ss ss stabilizer groups of (X ) are strictly smaller than that of the stabilizer groups in X , we can re- ss ′ ss ss s place X with (X ) and iterate this construction until X = X , which is handled in a previous case. ss ss s ss Case X 6= ∅ but codim(X \ X , X ) ≤ 1: Here the only modification needed to the proof of Theorem 1.8 is the observation that Crit (W| ) = 0 U×Y Crit (W| )× Y , so Perf(Crit (W| )) is a proper dg-category if Perf(Crit (W| )/G) is. 0 U 0 U×Y 0 U In the proof of Theorem 3.10, we used the following proposition to reduce from the case of a DM stack to the case of a variety. The proof is an adaptation of the methods of [BLS] to categories of matrix factorizations. Proposition 3.14. Let W : X → A be an LG-model such that X is a smooth quasi-projective (in the sense of [K]) DM stack and Crit (W ) is proper. Then there is a smooth variety Y and a ′ 1 ′ sm projective morphism W : Y → A such that MF(X, W ) ∈ G(MF(Y, W )) ⊂ LinCat . k((β)) Proof. As observed in [K], X = X/G for some reductive G and G-equivariant locally closed em- N s bedding X ֒→ (P ) for some linearized action of G on a projective space. Considering the graph 1 N 1 s of the map W : X → A gives a G-equivariant locally closed embedding X ֒→ (P × A ) , where G acts trivially on the A factor. Using Kirwan’s resolution algorithm [K6], as in [K], we modify N 1 ss N 1 s (P ×A ) by a series of smooth blow ups away from (P ×A ) such that the resulting semistable locus is a DM stack Y which is smooth and projective over A . Thus taking the closure of X in Y ¯ ¯ and resolving singularities, we can produce a Nagata compactification W : X → A of the original 1 1 morphism W : X → A such that X is projective over A and smooth over the ground field. Because X and X are DM and the critical locus Crit (W ) is proper, it follows that Crit (W ) is a 0 0 ¯ ¯ ¯ union of connected components of Crit (W ). Therefore, the category MF(X, W ) splits as a direct sum of the subcategory consisting of objects supported on Crit (W ) and the subcategory consisting of objects supported on other components of Crit (W) [P, Prop. 4.1.6]. So it suffices to prove the ¯ ¯ claim for MF(X, W ), i.e. we may assume that the potential itself W : X → A is projective rather than assuming that just Crit (W ) is proper. First we reduce to the case where X has generically trivial stabilizer. Because X is a global quotient stack, we may find a vector bundle V over X on which the automorphism groups act faithfully. Then P(V ⊕O ) → X is a rational morphism, so by Lemma 3.13 the claim for P(V ⊕O ) X X implies the claim for X, and the generic stabilizer of the former is trivial. Furthermore, the function W restricted to P(V ⊕ O ) will still be proper. Next we consider Bergh’s destackification, constructed in [B2] X , ⑥ ❅ ⑥ ❅ ⑥ ❅ ~~⑥ X X 35 ′ where X is the coarse moduli space and is smooth, the morphism π is a composition of root stacks along smooth divisors and f is a composition of root stacks along smooth divisors and blow ups along smooth centers. Note that the morphism f is rational, so by Lemma 3.13 it suffices to prove ′ ′ 1 ′ the claim for the composition W : X → X → A . Note that W will still be proper. ′ ′ ′ ′ ′ 1 Finally, because X is the coarse space of X , the function W descends uniquely to W : X → A , ′ ′ and this map is still proper. Because π : X → X is a composition of root stacks along smooth divisors, it suffices to prove the following claim: if Y is a smooth DM stack and X → Y is a root stack along a smooth divisor in Y and W : Y → A is a proper map, then the claim of the proposition for Y implies the claim for X. Let D ֒→ Y be the Cartier divisor used to form the root construction, and consider the diagram // D× Bµ X . D π // D Y Then [IU, Proposition 6.1] shows that we have a semiorthogonal decomposition ∗ r−1 ∗ ∗ Perf(X) = i π (Perf(D))⊗ M , . . . , i π (Perf(D))⊗ M, π (Perf(Y)) , ∗ ∗ D D where M is the universal invertible sheaf coming from the root stack construction. Given a map 1 1 W : Y → A and a module F ∈ Perf(A ), it is evident from the description of this semiorthogonal ∗ ∗ decomposition that each subcategory is preserved by the functor π W (F) ⊗ (−). If follows that 1 ⊗ this is a semiorthogonal decomposition of Perf(A ) -module categories, and thus by Lemma 3.9 there are induced semiorthogonal decompositions ∗ r−1 ∗ ∗ Perf(X ) = i π (Perf(D ))⊗ M , . . . , i π (Perf(D ))⊗ M, π (Perf(Y )) 0 ∗ 0 ∗ 0 0 D D 0 0 D E b ∗ b r−1 ∗ b ∗ b D Coh(X ) = i π (D Coh(D ))⊗ M , . . . , i π (D Coh(D ))⊗ M, π (D Coh(Y )) , 0 ∗ 0 ∗ 0 0 D D 0 0 where the subscript 0 refers to the derived zero fiber of W and its restrictions to D, X, and D×Bµ . Thus we have a semiorthogonal decomposition of k((β))-linear categories MF(X, W ◦ π) ≃ hMF(D, W| ), . . . , MF(D, W| ), MF(Y, W )i D D and the claim of the proposition for X follows from the claim for Y and induction, because D is smooth of one lower dimension. sm 3.2.1. The degeneration property for quotient stacks. For any A ∈ LinCat , we may form k((β)) k((β)) the k((β))-linear Hochschild complex C (A), which is a module over the CDGA Λ((β)) = k((β)) k((β))[B]/(B ). C (A) is computed by a Barr complex k((β)) C (A) := Hom(o , o )⊗ Hom(o , o )⊗ ···⊗ Hom(o , o ), (8) • 1 2 k((β)) 2 3 k((β)) k((β)) n 1 o ,o ,··· ,o 1 2 n where o are objects of A, the differential is given by the usual formula for the Hochschild complex of a k-linear dg-category [M], and B acts by the Connes differential. Using the Λ((β))-module k((β)),(n) k((β)),− k((β)),per structure, we may form the associated complexes C (A), C (A), and C (A) just • • • as in the k-linear case (see Section 1.3). Definition 3.15. A k((β))-linear dg-category A is said to have the k((β))-linear degeneration prop- k((β)),(n) erty if H (C (A)) is a flat k[u]/(u )-module for all n ≥ 1. ∗ • sm Lemma 3.16. Let A ∈ LinCat satisfy the k((β))-linear degeneration property, then any other k((β)) category C ∈ G(A) satisfies the k((β))-linear degeneration property. 36 Proof. The proof of Lemma 1.22 applies verbatim to show that if C = hC i is a Z-indexed semiorthog- onal decomposition of k((β))-linear dg-categories, then C satisfies the degeneration property if and sm only if each of the C do. This implies that the full ∞-subcategory of LinCat consisting of cat- k((β)) egories that satisfy the k((β))-linear degeneration property contains A and is closed under splitting countable semiorthogonal decompositions, so it contains G(A) by definition. Our main degeneration result for categories of matrix factorizations is the following: Proposition 3.17. Let W : X/G → A be an LG-model, where X is a smooth G-quasi-projective scheme which admits a semi-complete KN stratification. If Perf(Crit (W )/G) is a proper dg- category, then the k((β))-linear degeneration property holds for MF(X/G, W ). Proof. By Theorem 3.10 and Lemma 3.16, this reduces to the k((β))-linear degeneration property for MF(Y, W ) where Y is a smooth scheme and W : Y → A is a projective map. This amounts to the degeneration of the W -twisted Hodge-de Rham complex by [E, Theorem 1.3], which is established in [OV, Theorem 4.22]. Example 3.18. The hypotheses of the proposition are satisfied if X is projective-over-affine and Γ(X,O ) is finite dimensional – see Example 1.10. Crit (W ) We also have the following, which was left as an assumption in the original version of this paper: Corollary 3.19. If X is a smooth quasi-projective DM stack and W : X → A is a map such that Crit (W ) is proper, then the k((β))-linear degeneration property holds for MF(X, W ). Proof. Proposition 3.14 reduces this to the case of a projective morphism W : Y → A , with Y smooth, which as noted above follows from [OV, Theorem 4.22]. k((β)) Remark 3.20. We briefly discuss more concrete descriptions of C (MF(X, W )) when X is a smooth separated Deligne-Mumford stack. Namely, for any affine U with an ´etale map U → X, i 1 let Ω (U, W| ) denote the Tate construction on Ω (U)[i] with respect to the S -action given by • U −dW∧, i.e. Ω (U, W| ) is the Λ((β))-module Ω ((β))[i] with differential −β·dW∧. Letting B act • U on Ω (U, W| ) via the de Rham differential as usual, Ω (−, W ) defines a sheaf of Λ((β))-modules • U • aff on the small ´etale site X . We define the global de Rham complex to be the Λ((β))-module et aff Ω (X, W ) := RΓ X , Ω (−, W ) . • • et Assume for simplicity that Crit (W ) = Crit(W ). Then by combining the approach of Proposition 2.13 with [P, Theorem 8.2.6], it is not difficult to show that there is a natural isomorphism of Λ((β))- modules k((β)) cl C (MF(X, W )) ≃ Ω (I , W ), • • k((β)),per cl Tate which induces an equivalence C (MF(X, W )) ≃ Ω (I , W ) . Thus, Corollary 3.19 is equiv- • • alent to the statement that the k((β)) ⊗ Λ-module Ω (X, W ) has the degeneration property. This constitutes a slight generalization of the degeneration results of [OV] to DM stacks. 3.3. Graded Landau-Ginzburg models. We will use these results to establish a large class of examples of k-linear dg-categories for which the usual k-linear degeneration property holds. Definition 3.21. A graded LG-model is a map W : X → A /G , where X is a smooth algebraic k-stack whose automorphism groups at geometric points are affine, and G acts on A with weight one. Note that the data of a graded LG-model is equivalent to specifying an invertible sheaf L on X/G, which is classified by the composition X → A /G → BG , and a section W ∈ Γ(X/G, L). m m ′ ′ ∨ Denote by φ : X → X the G -torsor over X associated to L, i.e., X = Tot (L ) \ 0. To any m X 37 ′ ∗ ′ 1 graded LG-model, we use the term associated LG-model to denote the pair (X , φ W : X → A ). We will see in Proposition 3.24 that, in a precise sense, the graded LG-model is a refinement of its associated LG-model. In the setting of graded LG-models, for F ∈ D Coh(X ), there is a distinguished triangle: −1 ∗ F ⊗ L [1] → i i F → F giving rise to a natural transformation β : −⊗L[−2] → id which in the non-graded case, where L ≃ O , is the natural transformation id[−2] → id induced by the k[[β]]-linear structure of Definition 3.4. To make this natural transformation more explicit, we introduce an analog of the Koszul-Tate resolution (7), KT(M) = M, and construct a very concrete natural transformation β : KT(−)⊗ L[−2] → KT(−). −1 First consider the dg-algebra A := O ⊕ L · B where B is a formal variable of degree 1, X X X and dB = W , i.e., the differential is trivial on O and acts on the second summand by the X X −1 map L → O that is the defining section W . An A-module consists of an O -module M along X X −1 with a map B : L ⊗ M → M[−1] of O modules satisfying the Leibnitz rule dB (s ⊗ m) = X X X −1 Wsm− B (s⊗ dm), where s is a local section of L and m is a homogeneous local section of M. The ∞-category of A-modules is equivalent to the ∞-category D Coh(X ). Given a module M over A, we can form the analogue of M[B] above, which we denote −1 M := M ⊕ (L ⊗ M · B), −1 M −1 where the differential is just d ⊕ L ⊗ d . We have two operators B : L ⊗ M → M[−1], M M −1 −1 defined by the A-module structure of M, and B : L ⊗ M → L ⊗ M, which is just the identity. Using these two operators we can form a canonical action of A on M, where the action of B on M −1 M the first component is just B + B, and the action on the second component is L ⊗ B . X X −r −r+1 −r c c c Let Q(r) : L ⊗ M → L ⊗ M be the map which sends the first component of L ⊗ M to −r+1 the second component of L ⊗ M by the identity. Then we define the Koszul-Tate resolution as   M X −r   KT(M) := L ⊗ M[−2r]; d = d + Q(r) . r≥0 r>0 After pulling back to X , this complex can be identified with M[B] ⊗ (k((β))/βk[[β]]), so it follows that the canonical map KT(M) → M, which annihilates all r > 0 summands and B, is a quasi- isomorphism. We now define the natural map β : KT(M)⊗ L[−2] → KT(M) st to be the map of A-modules which annihilates the r = 0 summand and identifies the (r + 1) th summand of KT(M)⊗ L[−2] with the r summand of KT(M). Definition 3.22. We define the graded singularity category D Sing(X, W ) to be the idempotent completion of the dg-category with the same objects as D Coh(X ) but with morphisms between M, N given by −p Hom b (M, N) := hocolim Hom b (M, N ⊗ L )[2p], D Sing(X,W ) D Coh(X ) −p where the homotopy colimit is formed with respect to the natural maps β : L ⊗ N[2p] → −p−1 L ⊗ N[2p + 2]. Note also that given an LG-model (X, W ), we can forget the data of the trivialization of L to obtain a graded LG-model. This will correspond to forgetting the k((β))-linear structure on the category MF(X,W ). 38 Example 3.23. Let Y be a smooth variety over k, E a vector bundle over Y and let s ∈ Γ(E) be a regular section. We have an action of G on Tot(E ) by scaling in the fibers. The function ∨ 1 s therefore determines a mapping W : X = Tot(E )/G → A /G . The main theorem of [I] s m m b ∨ b −1 gives an equivalence of (Z-graded) dg-categories D Sing(Tot(E )/G , W ) ≃ D Coh(s (0)). The m s construction of this equivalence works equally well when Y has an action of a linear algebraic group G, E is a G-equivariant locally free sheaf, and s is G-invariant (see for instance [H3]). This gives an equivalence of k-linear dg-categories b ∨ b −1 D Sing(Tot(E )/(G× G ), W ) ≃ D Coh(s (0)/G). m s ′ ∨ 1 In this case the associated LG-model is simply the map W : Tot(E )/G → A obtained by forget- ting the G action. For a Z-graded k-linear dg-category C, we may tensor with k((β)), thereby collapsing the grading on Hom(E, F) to a Z/2Z-grading. The following proposition describes the relationship between a graded LG-model and its associated LG-model. 1 ′ ′ 1 Proposition 3.24. Let W : X → A /G be a graded LG-model, and let W : X → A be the associated LG-model. Then we have a canonical equivalence of Z/2Z-graded dg-categories b ′ ′ D Sing(X, W )⊗ k((β)) ≃ MF(X , W ). Lemma 3.25. Let π : X → Y be a smooth affine morphism of QCA stacks. Then objects of the form ∗ b b b π F for some F ∈ D Coh(Y) split generate D Coh(X), i.e. D Coh(X) is the smallest subcategory containing these objects which is closed under shifts, cones, and retracts. Proof. First note that the analogous claim holds for perfect stacks using Perf instead of D Coh and assuming only that π is affine. Indeed, the fact that the pushforward functor π : QC(X) → QC(Y) is conservative implies that objects of the form π F with F ∈ Perf(Y) split generate QC(X). In order to conclude the same for D Coh, we must imitate this argument for the categories IndCoh, i.e., show that π is conservative. The pushforward π again has a left IndCoh,∗ IndCoh,∗ ∗ b adjoint π which preserves D Coh and agrees with the usual pullback functor there. Because IndCoh π satisfies base change with respect to the shriek pullback [G, 5.2.5], and IndCoh satisfies IndCoh,∗ fppf descent with respect to shriek pullback, it suffices to show this when Y = Y is an affine derived scheme and hence X = X is as well. In this case [G, Proposition 4.5.3] shows that the essential image of the functor QC(X)⊗ IndCoh(Y) → IndCoh(X) generates the latter category QC(Y) (in fact this functor is an equivalence). It follows that IndCoh(X) is generated by objects of the IndCoh,∗ b form E ⊗ π (F) for E ∈ Perf(X) and F ∈ D Coh(Y). Furthermore, because X is affine the category Perf(X) is split generated by O , so IndCoh(X) is generated by objects of the form IndCoh,∗ b π (F) with F ∈ D Coh(Y). It follows by adjunction that π is conservative. IndCoh,∗ ′ b Proof of Proposition 3.24. Pullback along φ : X → X defines a functor of dg-categories D Coh(X ) → b ′ D Coh(X ) which intertwines the action of β and β . Making use of the canonical trivialization 0 L| ′ L| ′ ≃ O ′, we get a canonical map X X ∗ −p ∗ ∗ φ : hocolim Hom b (M, N ⊗ L )[2p] → hocolim Hom b ′ (φ (M), φ (N))[2p]. p p D Coh(X ) D Coh(X ) The latter can be identified with ∗ ∗ ∗ ∗ Hom ′ (φ (M), φ (N))⊗ k((β)) Hom ′ (φ (M), φ (N)). PreMF(X ,W ) k[[β]] MF(X ,W ) b b Because D Sing(X, W ) is generated by objects of D Coh(X ), this extends to a dg-functor on ∗ b ′ ′ idempotent completions φ : D Sing(X, W ) → MF(X , W ). From the universal property of the base-change category, φ admits an essentially unique k((β))-linear extension ∗ b ′ φ : D Sing(X, W )⊗ k((β)) → MF(X , W ), k((β)) 39 and we will show that this functor is an isomorphism. ∗ b b Concretely, φ maps an object F ∈ D Coh(X ), regarded as a generator for D Sing(X, W )⊗ k((β)), 0 k k((β)) ∗ b ′ ′ ′ to the object φ (F) ∈ D Coh(X ), regarded as a generator of MF(X , W ). Lemma 3.25 implies ∗ b ′ that objects of the form φ (F) with F ∈ D Coh(X ) generate MF(X , W ), so it suffices to show ∗ b that φ is fully faithful on the full subcategory of D Sing(X, W )⊗ k((β)) spanned by objects of k((β)) D Coh(X ). For M, N ∈ D Coh(X ), we consider q−p Hom b (M, N)⊗ k((β)) := hocolim Hom b (M, N ⊗ L )[2(p− q)][2q]. p−q D Sing(X,W ) D Coh(X ) Commuting colimits and reshuffling indices, this is isomorphic to q −p hocolim Hom (M, L ⊗ N ⊗ L )[2p] = b D Coh(X ) In this presentation, the operator β corresponds to the isomorphism between Hom b (M, L ⊗ D Coh(X ) −p q+1 −(p+1) N ⊗ L ) and Hom (M, L ⊗ N ⊗ L ). The Hom-complex is in turn isomorphic to D Coh(X ) −p hocolim Hom (M, φ (O ′ )⊗ N ⊗ L )[2p], p QC(X ) ∗ X where this last isomorphism uses the identification φ (O ′ ) ≃ L . It also uses the fact that X n∈Z −p M is coherent and φ (O )⊗ N ⊗ L is homologically bounded above, so that we may commute Hom (M,−) with this infinite direct sum. Using the projection formula and the adjunction QC(X ) between φ and φ , we finally obtain a natural equivalence ∗ ∗ ∗ −p = hocolim Hom b ′ (φ (M), φ (N)⊗ φ (L) )[2p] p D Coh(X ) the multiplication by β on Hom b (M, N) ⊗ k((β)) now corresponds to the canonical iso- D Sing(X,W ) morphisms: ∗ ∗ ∗ −p ∗ ∗ ∗ −(p+1) Hom b ′ (φ (M), φ (N)⊗ φ (L) ) → Hom b ′ (φ (M), φ (N)⊗ φ (L) ) D Coh(X ) D Coh(X ) 0 0 ∗ −1 which arise from the canonical trivialization O ′ → φ (L ). This operator is identified with the operator β in Definition 3.4 and therefore we have naturally identified the Hom-complex in b ∗ ∗ D Sing(X, W )⊗ k((β)) with Hom ′ ′ (φ (M), φ (N)) as required. MF(X ,W ) We can now establish our main result on the degeneration property for graded singularity cate- gories: Proposition 3.26. Let W : X/G → A /G be a graded LG-model with associated LG-model ′ ′ 1 ′ ′ W : X /G → A . If X admits a semi-complete KN stratification and Perf(Crit (W )/G) is a proper dg-category, then D Sing(X/G, W ) satisfies the k-linear degeneration property. Lemma 3.27. Let C be a k-linear (i.e., Z-graded) dg-category. Then the degeneration property for C is equivalent to the k((β))-linear degeneration property for C⊗ k((β)). k((β)) Proof. When D = C⊗k((β)), then the Barr complex (8) computing C (D) is quasi-isomorphic to the subcomplex in which all of the objects o lie in the generating set of objects of D of the form E ⊗ k((β)), with E ∈ C. It follows that k((β)) C (D) C (C)⊗ k((β)) • • canonically as dg-Λ((β))-modules. We therefore have that k((β)),(n) (n) C (D) C (C)⊗ k((β)) • • 40 L on the level of chain complexes as well. The result follows since as a complex k((β)) k[2n], n∈Z (n) (n) so the homology H (C (C)⊗ k((β))) = H (C (C))⊗ k((β)) will be flat over k[u]/u if and only ∗ • ∗ • (n) if H (C (C)) is flat over the same ring. ∗ • Proof of Proposition 3.26. By Lemma 3.27 and Proposition 3.24, the degeneration property for b ′ ′ D Sing(X/G, W ) is equivalent to the k((β))-linear degeneration property for MF(X /G, W ), so the result follows from Proposition 3.17. Remark 3.28. Using Example 3.23, Proposition 3.26 implies noncommutative Hodge-de Rham −1 degeneration for the derived category of certain complete intersections s (0)/G, where s is a G- invariant section of a locally free sheaf E on a smooth G-scheme Y . In fact, the hypotheses of ∨ 1 Amplification 1.25 imply that in the associated LG-model W : Tot(E )/G → A , Crit (W ) admits a complete KN-stratification, and thus Perf(Crit (W )/G) is a proper dg-category by Corollary 1.7. Proposition 3.26 therefore provides an alternate proof of Corollary 1.27. 4. Computations of Hochschild invariants 4.1. Generalities on Hochschild invariants. In this section we identify the Hochschild homol- ogy with functions on the derived inertia stack (or loop stack) X× X, which we denote by I , X×X and we give an explicit description when X is a quotient stack. Proposition 4.1. Let X be a smooth algebraic k-stack which is perfect, i.e., X is quasi-compact with affine diagonal, and QC(X) = Ind(Perf(X)), and let Δ : X → X× X be the diagonal. Then we have an identification C (Perf(X)) RΓ(I ,O ). • X I Proof. Morita theory for perfect stacks [BZFN] identifies the identity functor with Δ O in the ∗ X category QCoh(X× X). To compute C , we use the Morita invariant definition of the Hochschild homology of a compactly generated dg-category as the trace of the the identity functor. Thus we must compute the trace tr : QCoh(X× X) → QCoh(Speck). ∗ ∗ On sheaves of the form π (P )⊗ π (P ), with P , P ∈ Perf(X), we have that the trace is given by 1 2 1 2 1 2 ∗ ∗ tr(π (P )⊗ π (P )) := RHom(Hom(P ,O ), P ) 1 2 1 X 2 1 2 ∗ ∗ ∗ = RΓ(X, Δ (π (P )⊗ π (P ))). 1 2 1 2 Since the category QC(X× X) is the colimit completion of sheaves of this form, we have that for an arbitrary object F ∈ QC(X× X), the trace can be computed by F → RΓ(X, Δ (F)). (9) It follows that we have an isomorphism C (Perf(X)) = RΓ(X, Δ Δ O ). • ∗ X Now let X/G be a global quotient stack. We consider the scheme P := G× X × X. Denote by Δ : G× X → P the map (g, x) 7→ (g, x, x), and by Γ : G× X → P the map (g, x) 7→ (g, x, g · x). Both are closed immersions, and we will also use the notation Δ and Γ to denote the corresponding subschemes of P. Lemma 4.2. Let X/G be a smooth quotient stack. Both Γ and Δ are equivariant with respect to −1 the G action on P which sends h· (g, x , x ) → (hgh , hx , hx ). We have 1 2 1 2 L G C (Perf(X/G)) = RΓ(X,O ¯ ⊗ O ) . • Γ 41 Proof. By Proposition 4.1 we must compute the derived global sections of the structure sheaf of the derived inertia stack. First note the alternate presentation for the stack X/G ≃ G × X/G , where the G action in the second presentation is given by −1 (h , h )· (g, x) = (h gh , h x). 1 2 2 1 In this presentation the diagonal X/G → X/G × X/G corresponds to the G -equivariant map G× X → X × X given by (g, x) 7→ (x, gx). Let G act on P by −1 (h , h )· (g, x , x ) = (h gh , h x , h x ). 1 2 1 2 2 1 1 2 2 Then Γ is G -equivariant, and using the presentation above we see that the diagonal factors as the 2 2 2 2 closed immersion Γ : G× X/G → P/G followed by the projection P/G → X × X/G , which is −1 smooth and affine. It follows that the derived inertia stack is the derived intersection of p Γ and −1 2 p Γ in P× 2 P/G . Now P× P ≃ G× G× X × X with G -action given by −1 −1 (h , h )· (g , g , x , x ) = (h g h , h g h , h x , h x ) 1 2 1 2 1 2 2 1 2 2 1 1 2 2 1 1 The projections p , p : P× P → P are given by forgetting g and g respectively. We claim that 1 2 2 1 P× P/G ≃ P/G, where G acts on P as in the statement of the lemma. Indeed we can present P/G as the quotient of G× P by the G -action −1 −1 (h , h )· (g , g , x , x ) = (h g h , h g h , h x , h x ), 1 2 1 2 1 2 2 1 1 2 1 1 1 2 1 1 and we have a G -equivariant isomorphism G× P → P× 2 P given by (g , g , x , x ) 7→ (g , g g , x , g x ) 1 2 1 2 1 1 2 1 1 2 The resulting isomorphism P/G → P× 2P/G is given by the map (g, x , x ) 7→ (1, g, x , x ) which X 1 2 1 2 is equivariant with respect to the diagonal homomorphism G → G . −1 −1 2 2 In order to complete the proof, we must identify the closed substacks p (Γ/G ) and p (Γ/G ) 1 2 −1 in P× 2P/G under the isomorphism with P/G. The first is the closed subscheme p Γ∩({1}×P) = −1 Δ, regarded as a G-equivariant closed subscheme of P, and the second is p (Γ)∩ ({1} × P) = Γ as a G-equivariant closed subscheme of P. The case when X is a vector space, V, and G acts on V via a linear action, is of interest in two-dimensional gauge theory. In this case we make the above derived intersection explicit using a Koszul resolution. Denote by α : G× V → V the action morphism (g, v) 7→ g · v. We choose linear coordinates on V and identify V× V with Spec(k[x , y ]). i i The Koszul complex for the regular sequence K (x − y ) gives a resolution of the diagonal V×V i i on V× V. An important point is that, in this case, this resolution is G-equivariant with respect to the diagonal G-action because the G-action on V is linear. Then K (x − y ) → O G×V×V i i ¯ is a resolution of O over P. ∗ G Corollary 4.3. C (Perf(V/G)) (K (x − α (x ))) • G×V i i Proof. By the above lemma, C (Perf(V/G)) is isomorphic to G ∗ G (K (x − y )⊗ O ) ≃ (Γ K (x − y )) . G×V×V i i Γ G×V×V i i ∗ ∗ ∗ Pulled back to G × V, the function Γ (y ) = α (x ), hence Γ K (x − y ) = K (x − i i G×V×V i i G×V i α (x )). Remark 4.4. We expect that one could describe similar models for the Hochschild homology of general gauged linear sigma models (V/G, W ). The key step needed to do this would be to P, Theorem 4.2.3] to quotient (or more generally QCA) stacks. generalize [ 42 4.2. An HKR theorem and quotients of affine varieties. In [BG], Block and Getzler construct for any compact smooth M-manifold X an explicit model for the M-equivariant cyclic homology of the algebra C (X) using differential forms on X. Our goal is to translate their construction into algebraic geometry and establish their version of the equivariant Hochschild-Kostant-Rosenberg theorem when X = Spec(A) is smooth and affine, G is a reductive group, and X/G is formally proper, which is equivalent to the condition that A is finite dimensional over k. Our proof is an application of Theorem 1.8 and Theorem 2.17. For simplicity, we let k = C throughout this section. To compute the derived intersection appearing in Lemma 4.2, we may use the bar resolution B(A) of A as an A− A bimodule. Namely, ⊗n B (A) := A⊗ A ⊗ A where the differential can be described as the sum b = Σ (−1) ∂ , where i i ′ ′′ a a ⊗··· a ⊗ a , i = 0 1 n 0 0 ′ ′′ ′ ′′ ∂ (a ⊗ a ⊗··· a ⊗ a ) := a ⊗···⊗ a a ⊗···⊗ a , i 6= 0,6= n i 1 n i i+1 0 0 0 0 ′ ′′ a ⊗···⊗ a a , i = n 0 0 Our notation is meant to highlight the fact that the first and last variables in the bar complex play a distinguished role from the other a . We then have that O ⊗ B(A) is a resolution of O which i G we may restrict to Γ. The result is a complex where the n-th graded piece is ⊗n+1 n+1 C (A, G) := O ⊗ A = Γ(G× X ,O n+1). n G G×X For any c ∈ Γ(O ), the differentials ∂ above now take the form n+1 G×X  c(g, x , x , x ,··· , x ) i = 0 0 0 1 n−1 ∂ c(g, x , x ,··· , x ) := c(g, x ,··· , x , x ,··· , x ), i 6= 0,6= n i 0 1 n−1 0 i i n−1 c(g, x ,··· , x , g · x ), i = n 0 n−1 0 We define the Λ-module C (A) := C (A, G) , and note the following corollary of Lemma 4.2. •,G • Corollary 4.5. C (Perf(X/G)) = C (A) • •,G Let C (A) denote the completion of C (A) as a complex of modules over the representation •,G •,G ring Rep(G) ⊗ C = Γ(O ) at the conjugacy class [g]. Let us work for the moment with a fixed normal element g ∈ G. Let Y = Spec(A) denote the fixed point locus of g and B = Γ(O ). The letter Z will designate the centralizer of g and z denotes its Lie algebra, and normality of g ensures that Z is the complexification of Z := Z ∩ M for a maximal compact subgroup M ⊂ G. We have embeddings j : Z → G and k : Y → X. Lemma 4.6. When Spec(A)/G is formally proper, the natural restriction map gives rise to an ∗ ∧ ∧ isomorphism k : C (A) → C (B) . •,G •,Z g g Proof. Note that because Spec(A)/G is formally proper, each C (A) is a coherent Rep(G) ⊗ n,G C-module, so completion commutes with taking homology in this case, and it suffices to prove the result on the level of homology. It is known [FHT2, Proposition 3.10] that the map k : ∗ an ∧ an ∧ K (X ,C) → K (Y ,C) is an isomorphism. The comparison maps M g c g per ∗ an // K (X ,C) H (C (A)) •,G ∗ ∗ k k per ∗ an // K (Y ,C) H (C (B)) Zc •,Z 43 per per ∗ ∧ ∧ are maps of Rep(G)⊗C modules so we conclude that the map k : H (C (A)) → H (C (B)) ∗ ∗ g g •,G •,Z per per is an isomorphism as well. By Theorem 2.20, the vector spaces H (C (A)) and H (C (B)) admit ∗ ∗ •,G •,Z compatible Hodge structures. We observe that the Hodge decompositions per H (C (A)) H (C (A)) ∗ n •,G •,G n≡∗ mod 2 per H (C (B)) H (C (B)) ∗ ∗ •,Z •,Z n≡∗ mod 2 are decompositions of Rep(G) and Rep(Z) modules respectively. This follows because both the Hodge filtration and the conjugate filtration are filtrations of Rep(G) modules as can be seen for per example by examining the explicit model for C (A). The lemma now follows by taking completions •,G of these decompositions. ∧ n Next we construct a model for C (B) based on algebraic differential forms Ω , regarded as a •,Z projective B-module. Recall that the Cartan differential ∗ n ∗ n−1 i : Sym(z )⊗ Ω → Sym(z )⊗ Ω n−1 n ∗ is the unique extension of the contraction map Ω → z ⊗ Ω to a differential satisfying the Y Y ∗ n Liebniz rule. Alternatively, regarding ω ∈ Sym(z )⊗ Ω as a section of a quasi-coherent sheaf over z, we have (iω)(z) = i ω(z). We thus have a chain complex, in fact a CDGA, • ∗ ∗ n Ω [z ] = Sym(z )⊗ Ω [n], i . Y Y ∗ k Note that i is Z-equivariant, and that it descends to the quotient Sym(z) /m . Thus we can define • ∗ ∗ k n Ω [[z ]] := Sym(z )/m ⊗ Ω [n], i Y Y • ∗ Z ∗ k n Z Ω [[z ]] := (Sym(z )/m ⊗ Ω [n]) , i Y k Y Proposition 4.7. The comparison map of Construction 4.8 below is a quasi-isomorphism of Λ- modules ∧ ∧ • ∗ Z HKR : C (B) → lim(Ω [[z ]] ). •,Z g g Y k ←− Hence when Spec(A)/G is formally proper, we have a quasi-isomorphism of Λ-modules ∧ ∗ ∧ • ∗ Z HKR ◦k : C (A) → lim(Ω [[z ]] ). •,G g g Y k ←− Let Z denote the k-th infinitesimal neighborhood of the identity in Z. The exponential map (k) ∗ k provides a compatible system of isomorphisms exp : Spec(Sym(z )/m )) → Z . Note that under k (k) this equivalence, G acts algebraically on Z by scaling, and this action actually extends to (k) 1 ∗ k an action of the monoid A . This is encoded algebraically via a coaction map Sym(z )/m → ∗ k Sym(z )/m ⊗ C[t]. 44 Construction 4.8. For any b ∈ B, the coaction of O on B, the exponential map exp , and the G -action on z define an element ∗ k exp (−t· z)· b ∈ B ⊗ Sym(z )/m ⊗ C[t]. n+1 k We define C (B, Z) := O ⊗ B , i.e. the reduction of C (B, Z) modulo m , and introduce n,k Z • (k) n ∗ the map HKR : C (B, Z) → Ω [[z ]] given by g,k n,k k ′ ′ ψ⊗b ⊗b ⊗··· b 7→ ψ(g·exp (z)) b d(exp (−t z)·b )∧···∧d(exp (−t z)·b )dt dt ··· dt . (10) 1 n k k 1 1 k n n 1 2 n 0 0 ∗ k Here d(−) denotes the Sym(z )/m ⊗ C[t]-linear extension of the exterior derivative ∗ k 1 ∗ k d : B ⊗ Sym(z )/m ⊗ C[t] → Ω ⊗ Sym(z )/m ⊗ C[t]. ∗ k n The integrand is regarded as an element of Sym(z )/m ⊗ Ω ⊗ C[t , . . . , t ], and the integral over 1 n the standard n-simplex Δ is regarded formally as a linear map C[t , . . . , t ] → C. This formula is n 1 n identical to the one used in [BG], so it follows formally from the computations there that HKR g,k is a chain map (See for instance [BG, Theorem 3.2]). This map is Z-equivariant, so it restricts to a chain maps Z • ∗ Z HKR : C (B, Z) → Ω [[z ]] , and g,k •,k Y k ∧ ∧ • ∗ Z HKR := lim HKR : C (B) → lim Ω [[z ]] . g,k •,Z g g Y k ←− ←− k k Proof of Proposition 4.7. By the compatibility of the HKR maps with translation by the central • ∗ element g, it suffices to consider the case g = id. The maps HKR : C (B, Z) → Ω [[z ]] are id,k •,k k ∗ k a compatible family of maps of bounded complexes with coherent homology over B ⊗ Sym(z )/m . HKR is the classical HKR map id,1 ′ ′ b ⊗···⊗ b 7→ b db ··· db , n 1 n 0 0 n! which is an equivalence of Λ-modules. Hence by Nakayama’s lemma each HKR is a quasi- id,k isomorphism, and the same is true after taking Z-invariants. Hence HKR is a quasi-isomorphism. id The final statement of the proposition combines this with the previous lemma. References [AS ] Michael F Atiyah, Graeme B Segal, et al., Equivariant k-theory and completion, J. Differential Geometry 3 (1969), no. 1-18, 9. [B1] Vladimir Baranovsky, Orbifold cohomology as periodic cyclic homology, International Journal of Mathematics 14 (2003), no. 08, 791–812. [B2] Daniel Bergh, Functorial destackification of tame stacks with abelian stabilisers, arXiv preprint arXiv:1409.5713 (2014). [B3] Anthony Blanc, Topological k-theory of complex noncommutative spaces, arXiv preprint arXiv:1211.7360, to appear in Compositio Mathematica (2012). [BG] Jonathan Block and Ezra Getzler, Equivariant cyclic homology and equivariant differential forms, Annales scientifiques de l’ecole normale sup´erieure, 1994, pp. 493–527. [BGT] Andrew J Blumberg, David Gepner, and Gonc¸alo Tabuada, A universal characterization of higher algebraic k-theory, Geometry & Topology 17 (2013), no. 2, 733–838. [BLS] Daniel Bergh, Valery A Lunts, and Olaf M Schnu¨rer, Geometricity for derived categories of algebraic stacks, arXiv preprint arXiv:1601.04465 (2016). [BRTV] Anthony Blanc, Marco Robalo, Bertrand To¨en, and Gabriele Vezzosi, Motivic realizations of singularity categories and vanishing cycles, J. Ec. polytech. Math. 5 (2018), 651–747. MR3877165 [BZFN] David Ben-Zvi, John Francis, and David Nadler, Integral transforms and drinfeld centers in derived algebraic geometry, Journal of the American Mathematical Society 23 (2010), no. 4, 909–966. [BZNP] David Ben-Zvi, David Nadler, and Anatoly Preygel, Integral transforms for coherent sheaves, arXiv preprint arXiv:1312.7164 (2013). 45 [C] Lee Cohn, Differential graded categories are k-linear stable infinity categories, arXiv preprint arXiv:1308.2587 (2013). [CD] Denis-Charles Cisinski and Fr´ed´eric D´eglise, \’etale motives, arXiv preprint arXiv:1305.5361 (2013). [CT] Andrei C˘alda˘raru and Junwu Tu, Curved A algebras and Landau-Ginzburg models, New York J. Math. 19 (2013), 305–342. MR3084707 [D1] Pierre Deligne, Th´eorie de hodge: Iii, Publications Math´ematiques de l’IHES 44 (1974), 5–77. [D2] Tobias Dyckerhoff, Compact generators in categories of matrix factorizations, Duke Math. J. 159 (2011), no. 2, 223–274. MR2824483 [DG] Vladimir Drinfeld and Dennis Gaitsgory, On some finiteness questions for algebraic stacks, Geometric and Functional Analysis 23 (2013), no. 1, 149–294. [E] Alexander I. Efimov, Cyclic homology of categories of matrix factorizations, Int. Math. Res. Not. IMRN 12 (2018), 3834–3869. MR3815168 [EP] Alexander I. Efimov and Leonid Positselski, Coherent analogues of matrix factorizations and relative singu- larity categories, Algebra Number Theory 9 (2015), no. 5, 1159–1292. MR3366002 [FHT1] Daniel S. Freed, Michael J. Hopkins, and Constantin Teleman, Twisted equivariant K-theory with complex coefficients, J. Topol. 1 (2008), no. 1, 16–44. MR2365650 [FHT2] , Twisted equivariant K-theory with complex coefficients, J. Topol. 1 (2008), no. 1, 16–44. MR2365650 (2009c:19011) [G] Dennis Gaitsgory, ind-coherent sheaves, Mosc. Math. J. 13 (2013), no. 3, 399–528, 553. MR3136100 [GR] Dennis Gaitsgory and Nick Rozenblyum, A study in derived algebraic geometry: Volume i: correspondences and duality, American Mathematical Soc., 2017. [H1] D. Halpern-Leistner, Remarks on Theta-stratifications and derived categories, ArXiv e-prints (February 2015), available at 1502.03083. [H2] Wim H. Hesselink, Uniform instability in reductive groups, J. Reine Angew. Math. 303(304) (1978), 74–96. MR514673 [H3] Yuki Hirano, Derived Kno¨rrer periodicity and Orlov’s theorem for gauged Landau-Ginzburg models, Compos. Math. 153 (2017), no. 5, 973–1007. MR3631231 [HL] Daniel Halpern-Leistner, The derived category of a GIT quotient, J. Amer. Math. Soc. 28 (2015), no. 3, 871–912. MR3327537 [HLP] Daniel Halpern-Leistner and Anatoly Preygel, Mapping stacks and categorical notions of properness, arXiv preprint arXiv:1402.3204 (2014). [HR] Jack Hall and David Rydh, Perfect complexes on algebraic stacks, arXiv preprint arXiv:1405.1887 (2014). [I] M. Isik, Equivalence of the derived category of a variety with a singularity category, Int Math Res Notices 12 (2013), 2787–2808. [IU] A. Ishii and K. Ueda, The special McKay correspondence and exceptional collection, ArXiv e-prints (April 2011), available at 1104.2381. [K1] D Kaledin, Spectral sequences for cyclic homology, Algebra, geometry, and physics in the 21st century, 2017, pp. 99–129. [K2] D. Kaledin, Spectral sequences for cyclic homology, Algebra, geometry, and physics in the 21st century, 2017, pp. 99–129. MR3702384 [K3] Dmitry Kaledin, Non-commutative hodge-to-de rham degeneration via the method of deligne-illusie, Pure and Applied Mathematics Quarterly 4 (2008), no. 3, 785–875. [K4] Bernhard Keller, On differential graded categories, International Congress of Mathematicians. Vol. II, 2006, pp. 151–190. MR2275593 [K5] Frances Clare Kirwan, Cohomology of quotients in symplectic and algebraic geometry, Mathematical Notes, vol. 31, Princeton University Press, Princeton, NJ, 1984. MR766741 [K6] , Partial desingularisations of quotients of nonsingular varieties and their Betti numbers, Ann. of Math. (2) 122 (1985), no. 1, 41–85. MR799252 (87a:14010) [KKP1] L. Katzarkov, M. Kontsevich, and T. Pantev, Hodge theoretic aspects of mirror symmetry, Proceedings of Symposia in Pure Mathematics 78 (2008), 87–174. [KKP2] Ludmil Katzarkov, Maxim Kontsevich, and Tony Pantev, Bogomolov-Tian-Todorov theorems for Landau- Ginzburg models, J. Differential Geom. 105 (2017), no. 1, 55–117. MR3592695 [K] Andrew Kresch, On the geometry of deligne-mumford stacks, Algebraic geometry—seattle 2005. part 1, 2009, pp. 259–271. [KS] Maxim Kontsevich and Yan Soibelman, Notes on a-infty algebras, a-infty categories and non-commutative geometry, Homological mirror symmetry, 2009, pp. 1–67. [L1] Jean-Louis Loday, Cyclic homology, Springer, 1992. [L2] Jacob Lurie, Dag-vii: Spectral schemes, Preprint (2011). 46 [L3] , Higher algebra. 2014, Preprint, available at http://www. math. harvard. edu/˜ lurie (2016). [LMB] G´erard Laumon and Laurent Moret-Bailly, Champs alg´ebriques, Vol. 39, Springer Science & Business Media, [LM] M. Levine and F. Morel, Algebraic cobordism, Springer Monographs in Mathematics, Springer, Berlin, 2007. MR2286826 (2008a:14029) [LP] Kevin H. Lin and Daniel Pomerleano, Global matrix factorizations, Math. Res. Lett. 20 (2013), no. 1, 91–106. MR3126725 [LSM] L Gaunce Lewis, Mark Steinberger, and J Peter May, Equivariant stable homotopy theory, Springer, 1986. [M] Barry Mitchell, Rings with several objects, Advances in Math. 8 (1972), 1–161. MR0294454 [M ] J Peter May et al., Equivariant homotopy and cohomology theory, 1996. [MV] Fabien Morel and Vladimir Voevodsky, A 1-homotopy theory of schemes, Publications Math´ematiques de l’IHES 90 (1999), no. 1, 45–143. [N1] Linda Ness, A stratification of the null cone via the moment map, Amer. J. Math. 106 (1984), no. 6, 1281– 1329. With an appendix by David Mumford. MR765581 [N2] Behrang Noohi, Homotopy types of topological stacks, Adv. Math. 230 (2012), no. 4-6, 2014–2047. MR2927363 [O1] Martin C. Olsson, On proper coverings of Artin stacks, Adv. Math. 198 (2005), no. 1, 93–106. MR2183251 (2006h:14003) [O2] Dmitri Orlov, Matrix factorizations for nonaffine LG-models, Math. Ann. 353 (2012), no. 1, 95–108. MR2910782 [OV] Arthur Ogus and Vadim Vologodsky, Nonabelian hodge theory in characteristic p, Publications math´ematiques 106 (2007), no. 1, 1–138. [P] Anatoly Preygel, Thom–sebastiani and duality for matrix factorizations, arXiv preprint: arxiv:1101.5834 (2011). [S1] Kyoji Saito, Period mapping associated to a primitive form, Publ. Res. Inst. Math. Sci. 19 (1983), no. 3, 1231–1264. MR723468 [S2] Matthew Satriano, De Rham theory for tame stacks and schemes with linearly reductive singularities, Ann. Inst. Fourier (Grenoble) 62 (2012), no. 6, 2013–2051 (2013). MR3060750 [S3] Ed Segal, The closed state space of affine Landau-Ginzburg B-models, J. Noncommut. Geom. 7 (2013), no. 3, 857–883. MR3108698 [S4] Graeme Segal, Equivariant k-theory, Publications Math´ematiques de l’IHES 34 (1968), no. 1, 129–151. [S5] D. Shklyarov, Non-commutative Hodge structures: towards matching categorical and geometric examples, Trans. Amer. Math. Soc. 366 (2014), no. 6, 2923–2974. MR3180736 [S6] The Stacks Project Authors, stacks project, 2015. [S7] J. H. M. Steenbrink, Mixed Hodge structure on the vanishing cohomology, Real and complex singularities (Proc. Ninth Nordic Summer School/NAVF Sympos. Math., Oslo, 1976), 1977, pp. 525–563. MR0485870 (58 #5670) [T1] Gonc¸alo Tabuada, Higher K-theory via universal invariants, Duke Math. J. 145 (2008), no. 1, 121–206. MR2451292 (2009j:18014) [T2] Constantin Teleman, The quantization conjecture revisited, Annals of Mathematics 152 (2000), no. 1, 1–43. [T3] Robert W Thomason, Riemann-roch for algebraic versus topological k-theory, Journal of Pure and Applied Algebra 27 (1983), no. 1, 87–109. [T4] , Algebraic k-theory of group scheme actions, Algebraic Topology and Algebraic K-theory (Ann. Math. Stud. 113, pp. 539–563) Princeton, NJ: Princteon U (1987). [T5] , Equivariant algebraic vs. topological k-homology atiyah-segal-style, Duke Mathematical Journal 56 (1988), no. 3, 589–636. [T6] B. Toen, Th´eor`emes de Riemann-Roch pour les champs de Deligne-Mumford, K-Theory 18 (1999), no. 1, 33–76. MR1710187 (2000h:14010)

Journal

MathematicsarXiv (Cornell University)

Published: Jul 7, 2015

There are no references for this article.