On the topological K-theory of twisted equivariant perfect complexes
Michael K. Brown, Tasos Moulinos

TL;DR
This paper establishes a comparison between topological K-theory of twisted perfect complexes on quotient stacks and twisted equivariant K-theory, proving their equivalence under certain conditions and providing new proofs for existing theorems.
Contribution
It introduces a new comparison map and proves its equivalence in specific cases, extending previous work in twisted equivariant K-theory.
Findings
Comparison map constructed and shown to be an equivalence under certain conditions
New proof of Moulinos's theorem in the non-equivariant case
Generalization of existing constructions in twisted K-theory
Abstract
We construct a comparison map from the topological K-theory of the dg-category of twisted perfect complexes on certain global quotient stacks to twisted equivariant K-theory, generalizing constructions of Halpern-Leistner-Pomerleano and Moulinos. We prove that this map is an equivalence if a version of the projective bundle theorem holds for twisted equivariant K-theory. Along the way, we give a new proof of a theorem of Moulinos that the comparison map is an equivalence in the non-equivariant case.
Peer Reviews
No public reviews on file for this paper yet. If you reviewed it on a platform where reviews are public (OpenReview, ICLR, NeurIPS, ICML), you can paste yours below so the community can read it here.
Videos
No videos yet. Explain this paper in a talk, walkthrough, or lecture? Add one.
On the topological -theory of twisted equivariant perfect complexes
Michael K. Brown
Tasos Moulinos
Abstract
We construct a comparison map from the topological -theory of the dg-category of twisted perfect complexes on certain global quotient stacks to twisted equivariant -theory, generalizing constructions of Halpern-Leistner-Pomerleano and Moulinos ([HLP15], [Mou19]). We prove that this map is an equivalence if a version of the projective bundle theorem holds for twisted equivariant -theory. Along the way, we give a new proof of a theorem of Moulinos that the comparison map is an equivalence in the non-equivariant case.
††MB gratefully acknowledges support from the National Science Foundation (NSF award DMS-1502553).
Contents
-
5.1 Background on invertible algebra bundles and twists of -theory
-
5.2 On the topological -theory of twisted equivariant projective bundles
1 Introduction
Let denote the -category of -linear dg categories, and let denote the -category of spectra. Blanc introduced in [Bla16] a functor
[TABLE]
the topological -theory functor for -linear dg categories, based on a proposal of Toën ([Toë10]). Blanc proves that enjoys the following properties:
- (1)
maps Morita equivalences to weak equivalences,
- (2)
maps short exact sequences of dg categories to fiber sequences, and
- (3)
if is a separated scheme of finite type over , there is a natural equivalence
[TABLE]
in , where is the (ordinary) topological -theory functor for topological spaces, and denotes the complex points of equipped with the analytic topology.
A main source of motivation for the construction of is Katzarkov-Kontsevich-Pantev’s seminal work [KKP08] on noncommutative Hodge theory. The authors predict in [KKP08, Section 2.2.6(b)] that there should be a notion of topological -theory of a -linear dg-category such that, when is smooth and proper, its topological -theory provides a rational lattice inside its periodic cyclic homology, just as in the classical setting of smooth proper complex varieties. This prediction is formulated precisely in Conjecture 1.7 of [Bla16].
Blanc’s comparison theorem (item (3) above) has been extended by Halpern-Leistner-Pomerleano to quotient stacks. Before stating Halpern-Leistner-Pomerleano’s theorem, we recall that, when is a -scheme with action of a complex algebraic group , we say is -quasi-projective if there exists a locally closed immersion , where is a finite-dimensional -representation and is -equivariant.
Theorem 1.1** ([HLP15] Theorem 3.9).**
Let be a complex linear algebraic group, and let be a smooth -quasi-projective scheme over . Choose a decomposition
[TABLE]
where is reductive and is a connected unipotent group. Let be a maximal compact subgroup of , and let denote the -equivariant topological -theory spectrum of . There is a canonical equivalence
[TABLE]
Remark 1.2*.*
It is important to note that denotes the representable -equivariant complex topological -theory spectrum, as defined in [May96, Chapter XIV], as opposed to the -theory with compact support discussed in [Seg68].
Blanc’s comparison theorem has also been extended by the second author to twisted perfect complexes:
Theorem 1.3** ([Mou19] Theorem 9.6).**
Let be a separated scheme of finite type over , and let be an Azumaya algebra on . Let denote the twist of -theory determined by , and let denote the -twisted -theory spectrum of . There is a canonical equivalence
[TABLE]
A useful consequence of these comparison theorems is that equivariant topological -theory of smooth -quasi-projective schemes over (resp. twisted -theory of separated finite type schemes over ) is invariant under Morita equivalences of dg-categories of equivariant perfect complexes (resp. twisted perfect complexes).
In this article, we study a common generalization of Theorems 1.1 and 1.3. That is, we ask whether the topological -theory of the dg-category of twisted equivariant perfect complexes is equivalent to an associated twisted equivariant topological -theory spectrum. While we do not answer this question in this paper, we construct a comparison map from one to the other, and we reduce the question to a projective bundle-type formula in twisted equivariant -theory.
In more detail: let , , and be as in Theorem 1.1, an Azumaya algebra on the quotient stack , and the twist of the -equivariant topological -theory of determined by . Denote the -twisted -equivariant -theory spectrum of by . For background on twisted equivariant -theory, we refer the reader to [AS04, Section 6] or [FHT11]. Our first goal is to construct a comparison map
[TABLE]
We briefly outline the construction. Let be the Severi-Brauer stack associated to the Azumaya algebra (see Section 2 for details). By Theorem 1.1, we have an equivalence
[TABLE]
In Theorem 3.1, we prove that there is a semi-orthogonal decomposition
[TABLE]
where is the degree of , and the Azumaya algebra is the -fold tensor power of . It follows that
[TABLE]
We also define in Section 5 a canonical map
[TABLE]
When is trivializable, this is the equivalence from the projective bundle formula for (untwisted) equivariant -theory (see Theorem 5.2(1)). We define the map (1.4) by mapping to via (1.6), applying , and then projecting onto via (1.5). Our main result is:
Theorem 1.7**.**
Let be a complex linear algebraic group, a smooth -quasi-projective scheme over , and an Azumaya algebra on . Let be as in the statement of Theorem 1.1, and let be the twist of the -equivariant topological -theory of determined by . There is a natural comparison map
[TABLE]
and it is an equivalence if the map (1.6) is an equivalence.
We give a proof that the map (1.6) is an equivalence in the case where is trivial (Theorem 5.2(3)). This recovers a result of the second author [Mou19, Theorem 1.3]; combining this with Theorem 1.7 gives a new, simpler proof of the comparison theorem for non-equivariant twisted perfect complexes (Theorem 1.3 above); see Remark 5.11 below for details. We also show in Theorem 5.2(2) that the question of whether the map (1.6) is an equivalence may be reduced to the case where .
Remark 1.8*.*
We note that Bergh-Schnürer independently obtained the semi-orthogonal decomposition in Theorem 3.1; this result appeared in the preprint [BS19] a few days after the first version of this article was posted.
Acknowledgements. The authors are grateful to Benjamin Antieau for valuable comments during the preparation of this article. We thank the referee for his or her careful reading and insightful comments, and especially for catching an error in a previous version of this article.
2 Background on Azumaya algebras over algebraic stacks
Let be an algebraic stack over a base scheme . We recall the definition of the lisse-étale site of , denoted ([Ols16, Definition 9.1.6]). The objects of are pairs , where is a scheme and is a smooth morphism. A morphism is given by a morphism of schemes along with a 2-isomorphism . A family of morphisms is a covering if
[TABLE]
is an étale covering. Denote by the lisse-étale topos of . The structure sheaf is given by
[TABLE]
Definition 2.1**.**
An Azumaya algebra of degree over is a locally free -algebra such that is lisse-étale-locally isomorphic to ; that is, for every smooth morphism , where is a scheme, there is an étale covering such that . We shall say is trivializable if there is an isomorphism for some locally free -module . When such an isomorphism has been fixed, we will say is trivial.
We call a morphism of algebraic stacks a Severi-Brauer stack of relative dimension if it is lisse-étale-locally isomorphic to a projectivized vector bundle of relative dimension . We briefly describe a bijection between isomorphism classes of Azumaya algebras of degree and Severi-Brauer stacks of relative dimension . Define to be the group of units in (so ), and set . Conjugation determines an isomorphism
[TABLE]
For any group object in , there are cohomology functors for ; if is an abelian group object, these functors are defined for all . Moreover, the set classifies -torsors on . The isomorphism therefore gives a bijection between isomorphism classes of Azumaya algebras of degree on and the set . Such a torsor determines a Severi-Brauer stack of relative dimension . On the other hand, let be a Severi-Brauer stack of relative dimension . For each smooth morphism , where is a scheme, choose an étale cover such that, for each , the pullback of along is isomorphic to . The line bundles do not necessarily glue to give a line bundle on ; however, applying a construction of Quillen in [Qui73, Section 8.4], one may construct a canonical vector bundle on such that, for each , the pullback of along is isomorphic to . In more detail: one uses that
- •
the bundle on is -equivariant, and
- •
is the bundle associated to a -torsor
to construct the descent data necessary to glue the bundles into a bundle on (Quillen only works with schemes in [Qui73], but the construction adapts to the setting of algebraic stacks). The Azumaya algebra associated to is .
Given an Azumaya algebra on , denote by the dg-category of perfect complexes of left -modules.
3 Semi-orthogonal decompositions for Severi-Brauer stacks
We obtain in this section a semi-orthogonal decomposition of the dg-category of perfect complexes on a Severi-Brauer stack (Theorem 3.1), generalizing a theorem of Bernardera ([Ber09] Theorem 5.1). We emphasize that our proof of Theorem 3.1 is just a matter of concatenating several results of Bergh-Schnürer in [BS20].
Let be an algebraic stack over a scheme , and let be a Severi-Brauer stack of relative dimension , as defined in Section 2. For each smooth morphism , where is a scheme, choose an étale cover such that we have an isomorphism
[TABLE]
of schemes over for each .
As discussed in Section 2, there is a canonical vector bundle on such that the pullback of along is isomorphic to for all , and
[TABLE]
is the Azumaya algebra on corresponding to . For all , the -fold tensor power of is isomorphic to ; note that there is a canonical isomorphism
[TABLE]
In particular, is a trivial Azumaya algebra. Noting that is a right -module, we have an equivalence
[TABLE]
given by
[TABLE]
with inverse given by
[TABLE]
Define dg functors
[TABLE]
Note that each has a right adjoint .
We recall that a dg functor is called quasi-fully faithful if the induced functor on homotopy categories is fully faithful.
Theorem 3.1**.**
The dg functors are quasi-fully faithful, and there is a semi-orthogonal decomposition
[TABLE]
Proof.
To prove that the are fully faithful, we will apply Bergh-Schnürer’s “conservative descent for fully faithfulness” ([BS20, Proposition 4.12]). We recall that a functor between ordinary categories is called conservative if it reflects isomorphisms. Note that a triangulated functor is conservative if and only if it reflects zero objects.
Fix . We have diagrams
[TABLE]
and
[TABLE]
where the products range over each element of each of the étale open covers chosen above. The rightmost horizontal maps are given by pullback, and (resp. ) is the evident analogue of (resp. ). It’s easy to check that the diagrams commute.
The leftmost vertical map in the first diagram is quasi-fully faithful by the projective bundle theorem for schemes, and therefore the middle vertical map in the first diagram is as well. Since the triangulated functor induced by the dg functor
[TABLE]
on the level of homotopy categories reflects 0 objects, it is conservative. It follows from Bergh-Schnürer’s conservative descent for fully faithfulness that is also quasi-fully faithful. The semi-orthogonal decomposition now follows immediately from Bergh-Schnürer’s conservative descent theorem for semi-orthogonal decompositions ([BS20, Theorem 5.16]) and their projective bundle theorem for algebraic stacks ([BS20, Corollary 6.8]).
∎
4 Halpern-Leistner-Pomerleano’s comparison map
In this section, we recall the construction of the equivalence in Halpern-Leistner-Pomerleano’s comparison theorem (Theorem 1.1). Let , , and be as in the setup of Theorem 1.1. Let
[TABLE]
denote the comparison map between the connective -equivariant algebraic -theory of and the (nonconnective) -equivariant topological -theory of ([Tho88, Section 5.4]). We fix some notation: denote by
- •
the category of affine schemes over ,
- •
the suspension spectrum functor,
- •
the operation of adjoining a basepoint to a space, and
- •
the internal mapping object in .
The map induces a morphism
[TABLE]
of presheaves of spectra on ; in the middle term , the input is considered as a space with trivial -action. The equivalence on the right follows from [HLP15, Lemma 3.10].
Let denote the -category of presheaves of spectra on , and let (resp. ) denote the category of -modules (resp. -modules). Let
[TABLE]
denote the topological realization functor described in [Bla16, Definition 3.13] (Blanc denotes this functor by ). Given a dg-category over , the semi-topological -theory of , denoted , is defined to be . As observed in [Bla16, Definition 3.13], has a right adjoint given by
[TABLE]
The map (4.1) therefore induces a map
[TABLE]
As proven in [Bla16, Section 4], the semi-topological -theory spectrum of any -linear dg-category is a -module. is defined to be . Noting that (4.2) is a morphism of -modules, the adjunction between and given by extension/restriction of scalars yields a map
[TABLE]
This is the map that appears in the Halpern-Leistner-Pomerleano comparison theorem (Theorem 1.1).
5 The comparison map
Let , , , and be as in the setup of Theorem 1.7. Let be the Severi-Brauer stack of relative dimension corresponding to , and let be the vector bundle on introduced in Section 2.
5.1 Background on invertible algebra bundles and twists of -theory
Our reference for this subsection is [Fre12]. We recall that a topological groupoid is a pair of topological spaces that form the objects and morphisms, respectively, in a category in which all morphisms are invertible. Denote by the source and target maps, respectively. For example, the space equipped with its -action determines the global quotient groupoid with and . In this case, and .
There is a notion of a bundle of invertible (i.e. finite dimensional central simple) -algebras over a topological groupoid; we refer the reader to [Fre12, Definition 1.59] for details. We observe that, if is a -equivariant Azumaya algebra over , its analytification is an invertible algebra bundle over the groupoid . Note that, in [Fre12, Definition 1.59], the fibers in an invertible algebra bundle are allowed to be -graded, but, in our setting, all invertible algebra bundles will be trivially graded. Given -equivariant Azumaya algebras and on , a morphism of the associated invertible algebra bundles is an -equivariant --bimodule. Such a bimodule determines an isomorphism if there is an - bimodule such that and .
Twists of topological -theory of a groupoid can be defined in terms of invertible algebra bundles; see [Fre12, Definition 1.78] for the precise definition. In particular, each determines a twist of the -equivariant -theory of (take the local equivalence in [Fre12, Definition 1.78] to be the identity on ). An isomorphism of twists of the topological -theory of arising from invertible algebra bundles on in this way is just an isomorphism of the invertible algebra bundles, in the sense defined above. We note that an isomorphism of twists determines an equivalence of twisted topological -theory spectra.
5.2 On the topological -theory of twisted equivariant projective bundles
Fix . Recall that , and there is a canonical isomorphism
[TABLE]
The --bimodule therefore determines a canonical isomorphism from to the trivial invertible algebra bundle , and hence an isomorphism from to the zero twist. This isomorphism induces an equivalence
[TABLE]
We define
[TABLE]
The map is the analogue of the functor on the level of twisted equivariant -theory; note that is natural with respect to pullback along morphisms of Severi-Brauer stacks. We have a map
[TABLE]
Theorem 5.2**.**
- (1)
When the twist is trivializable, the map (5.1) recovers the equivalence from the projective bundle formula in (untwisted) equivariant topological -theory. 2. (2)
Suppose
[TABLE]
is an equivalence whenever is a closed subgroup of and is an arbitrary -equivariant twist of the -theory of a point; here, is the projective -representation associated to . In this case, the map (5.1) is an equivalence. 3. (3)
The map (5.1) is an equivalence when is trivial.
Remark 5.3*.*
Theorem 5.2(3) was proven by the second author in [Mou19, Theorem 1.3]; it is a consequence of his comparison theorem for the topological -theory of the dg-category of twisted perfect complexes (Theorem 1.3 above). We give here a direct proof of this fact that does not involve topological -theory of dg-categories.
Proof.
Suppose is trivializable. We have , and so is isomorphic to the bundle of endomorphisms of . Just as above, the --bimodule determines an equivalence
[TABLE]
Similarly, the - -bimodule induces an equivalence
[TABLE]
Now we observe that the diagram
[TABLE]
commutes on the level of homotopy groups. The commutativity of the square on the left is clear. As for the triangle on the right, since, in the setup of [Fre12], compositions of morphisms of twists are given by tensor products of bimodules, it’s easy to check that the map is induced by the - -bimodule given by the line bundle . (1) now follows from [FHT11, Proposition 3.4 (ii)].
We now prove (2). Since our space is a smooth manifold, is compact, and acts smoothly on , admits an open cover by -invariant open subsets such that each is -equivariantly homotopy equivalent to an -space of the form for some closed subgroup of ([Bre72, Corollary VI.2.4]; see Section IV.1 for the definition of a “locally smooth” action). Since is second countable, we can assume this cover is countable; write it as . There is an equivalence for each and (here, and throughout the proof, we abuse notation slightly: the superscripts “” really indicate pullbacks of ). Covering with the -invariant open sets , we obtain a commutative square
[TABLE]
where is some projective -representation. By assumption, the bottom horizontal map is an equivalence; it follows that the top horizontal map is an equivalence as well.
For , let
[TABLE]
and
[TABLE]
We observe that each is a projective bundle over . We have maps
[TABLE]
for each . Since the top horizontal maps in the squares (5.4) are equivalences, the Mayer-Vietoris theorem for twisted -theory ([FHT11, Section 3]) implies that the maps (5.5) are all equivalences. Let
[TABLE]
and
[TABLE]
denote the “infinite mapping cylinders”, following the terminology of [FHT11]. The proof of [FHT11, Proposition A.19] implies that, taking the homotopy limit of (5.5) over , we get the map
[TABLE]
we conclude that this map is also an equivalence.
Let
[TABLE]
denote the canonical maps. As in the proof of [FHT11, Proposition A.19], choose a partition of unity subordinate to the open cover of , and use it to construct a section of . Pulling back along , we get an induced partition of unity subordinate to the open cover of and therefore an induced section of such that the diagram
[TABLE]
commutes. The commutativity of diagram (5.7) implies that the map (5.1) is a section of the equivalence (5.6), and so (5.1) is also an equivalence. This proves (2).
Finally, (3) is immediate from (1) and (2), since any twist of the non-equivariant -theory of a point is trivializable.
∎
Question 5.8**.**
Is the map (5.1) an equivalence in general?
Example 5.9**.**
To answer Question 5.8, it suffices, by Theorem 5.2(2), to consider the case where . In this case, is simply a projective -representation. Such an object corresponds to a central extension of the form
[TABLE]
This central extension canonically determines a complex -representation , and we have
[TABLE]
where denotes the unit sphere in . Let denote the unit ball in . By the Thom isomorphism and the long exact sequence of the pair , we have an exact sequence
[TABLE]
where denotes the representation ring of , and .
On the other hand, by [FHT11, Example 1.10], we have
[TABLE]
where denotes the ring of -twisted representations of . So, to prove the projective bundle formula for twisted equivariant -theory in the case of a point, one must show
- (a)
The map induced by (5.1) is an isomorphism, and
- (b)
.
We have been unable to prove either of these two statements.
We remark that the answer to Question 5.8 is “yes” in the case where and is a torus. To see this, say . We have ; thus, every twist of the -equivariant topological -theory of a point is trivializable, and so the statement follows from Theorem 5.2(1).
5.3 Proof of Theorem 1.7
As discussed in the introduction, we define our comparison map
[TABLE]
by mapping to via (5.1), applying the inverse of the Halpern-Leistner-Pomerleano equivalence discussed in Section 4, and then projecting onto the summand of via the equivalence in Theorem 3.1.
To prove Theorem 1.7, we will need the following
Lemma 5.10**.**
Assume that the answer to Question 5.8 is “yes”. Denote by the composition
[TABLE]
where the second map is projection onto the component, so that is a canonical splitting of . In this case, in the setting of Theorem 1.7, Halpern-Leistner-Pomerleano’s comparison map respects the decompositions of and arising from Theorems 3.1 and the map (5.1), respectively. More precisely, if and , the map is trivial, i.e. it induces the zero map in the stable homotopy category.
Proof.
Recall the comparison map from Section 4. Our first step is to show
[TABLE]
is trivial. The map factors as
[TABLE]
where (resp. ) denotes the connective algebraic -theory of the exact category of -equivariant vector bundles on (resp. -equivariant complex vector bundles on ). Note that, since is -quasi-projective, has the resolution property; this is why the map is an equivalence.
We have a commutative diagram
[TABLE]
The middle two vertical maps are given by the same formula as , and we abuse notation by referring to them with the same symbol. The horizontal maps are all the canonical ones. Observing that the composition
[TABLE]
is trivial, we conclude that is trivial.
We now show is trivial. Since is trivial, the composition
[TABLE]
is a trivial map of presheaves of spectra on (i.e. the map is trivial pointwise), where the middle map is as constructed in Section 4. It follows that the induced map
[TABLE]
is also trivial (here, denotes the semi-topological -theory functor for dg categories, whose definition is recalled in Section 4). Finally, we conclude that the induced map
[TABLE]
is trivial. ∎
Proof of Theorem 1.7.
Our hypothesis is that the answer to Question 5.8 is “yes”. Define maps as in the statement of Lemma 5.10. We have maps
[TABLE]
where is Halpern-Leistner-Pomerleano’s comparison map. By Theorems 3.1 and our assumption that the map (5.1) is an equivalence, the composition
[TABLE]
is an equivalence, i.e. the induced map
[TABLE]
of -graded abelian groups is an isomorphism. By Lemma 5.10, the off-diagonal entries of this matrix are 0, and so the diagonal entries are all isomorphisms; that is, each comparison map is an equivalence. Finally, observe that our equivalence is the inverse of . ∎
Remark 5.11*.*
In the construction of the comparison map, the smoothness assumption is only necessary so that we can (1) use Halpern-Leistner-Pomerleano’s comparison map, and (2) choose an equivariantly contractible open cover of . In particular, the smoothness assumption is not necessary when is trivial, as we can just use Blanc’s comparison map in this case, and, of course, is locally contractible as well. Similarly, “quasi-projective” can be replaced with “separated of finite type” when is trivial. With this in mind, notice that Theorem 5.2(3) and Theorem 1.7 imply that our comparison map is an equivalence in the non-equivariant case. This gives a new, simpler proof of the second author’s comparison theorem for the topological -theory of the dg-category of (non-equivariant) twisted perfect complexes (Theorem 1.3).
The reference list from the paper itself. Each links out to its DOI / PubMed record.
- 1[AS 04] Michael Atiyah and Graeme Segal, Twisted K-theory , Ukr. Mat. Vis. (2004), no. 1, 287–330.
- 2[Ber 09] Marcello Bernardara, A semiorthogonal decomposition for Brauer-Severi schemes , Mathematische Nachrichten 282 (2009), no. 10, 1406–1413.
- 3[Bla 16] Anthony Blanc, Topological K-theory of complex noncommutative spaces , Compositio Mathematica 152 (2016), no. 3, 489–555.
- 4[Bre 72] Glen E. Bredon, Introduction to compact transformation groups , Pure and Applied Mathematics, Vol. 46, Academic Press, New York-London, 1972.
- 5[BS 19] Daniel Bergh and Olaf M. Schnürer, Decompositions of derived categories of gerbes and of families of Brauer-Severi varieties , ar Xiv preprint ar Xiv:1901.08945 (2019).
- 6[BS 20] , Conservative descent for semi-orthogonal decompositions , Advances in Mathematics 360 (2020), 106882.
- 7[FHT 11] Daniel S. Freed, Michael J. Hopkins, and Constantin Teleman, Loop groups and twisted K 𝐾 K -theory I , J. Topol. 4 (2011), no. 4, 737–798.
- 8[Fre 12] Daniel Freed, Lectures on twisted K-theory and orientifolds , Erwin Schrödinger Institute program K-theory and Quantum Fields (2012).
