The six operations in equivariant motivic homotopy theory
TTHE SIX OPERATIONS IN EQUIVARIANT MOTIVIC HOMOTOPY THEORY
MARC HOYOIS
Abstract.
We introduce and study the homotopy theory of motivic spaces and spectra parametrized byquotient stacks [
X/G ], where G is a linearly reductive linear algebraic group. We extend to this equivariantsetting the main foundational results of motivic homotopy theory: the (unstable) purity and gluing theoremsof Morel–Voevodsky and the (stable) ambidexterity theorem of Ayoub. Our proof of the latter is differentthan Ayoub’s and is of interest even when G is trivial. Using these results, we construct a formalism of sixoperations for equivariant motivic spectra, and we deduce that any cohomology theory for G -schemes thatis represented by an absolute motivic spectrum satisfies descent for the cdh topology. Contents
1. Introduction 21.1. Equivariant homotopy theory 21.2. The formalism of six operations 31.3. Summary of the construction 51.4. Some applications 71.5. Acknowledgements 81.6. Notation and terminology 82. Some equivariant geometry 82.1. Invariant subschemes 82.2. Quasi-affine and quasi-projective morphisms 102.3. Linearly reductive groups 122.4. Affine resolutions 122.5. Lifting locally free sheaves 132.6. Lifting smooth quasi-sections 142.7. Linearizations 152.8. Tame group schemes 163. Unstable equivariant motivic homotopy theory 173.1. Preliminaries 173.2. Homotopy invariance 173.3. Nisnevich excision 193.4. Equivariant motivic spaces 203.5. Smooth closed pairs 214. Functoriality 244.1. Easy results 254.2. Exactness properties of pushforwards 274.3. Gluing 305. Ambidexterity for smooth projective morphisms 325.1. Pointed equivariant motivic spaces 325.2. Spheres, Thom spaces, and purity 345.3. The Pontryagin–Thom collapse 366. Stable equivariant motivic homotopy theory 456.1. Equivariant motivic spectra 456.2. The exceptional functors 496.3. Descent properties 53References 54
Date : March 9, 2018. a r X i v : . [ m a t h . AG ] M a r MARC HOYOIS Introduction
The goal of this paper is to develop the formalism of six operations in stable equivariant motivic homotopytheory. An equivariant version of motivic homotopy theory was first considered by Voevodsky in [Del09] andplayed a small part in his proof of the Bloch–Kato conjecture [Voe11], more precisely in the construction ofsymmetric power functors on the A -homotopy category [Voe10, § G -schemes with some specific properties, for G an algebraic group. We will discuss in § Equivariant homotopy theory.
For S a scheme, Voevodsky constructed the ∞ -category SH( S ) ofmotivic spectra over S [Voe98]. It is in many ways an algebraic analog to the ∞ -category SH top ( S ) of sheavesof spectra over a topological space S . In particular, both constructions support a formalism of six operationsas the base S varies.In topology, a basic observation that leads us from parametrized to equivariant homotopy theory is thatthe objects being parametrized (e.g., CW complexes or smooth manifolds) have nontrivial automorphismsand hence can vary in families parametrized by topological stacks rather than just spaces. We can thereforeexpect that the functor S (cid:55)→ SH top ( S ) admits an interesting extension { topological spaces } {∞ -categories } . { topological stacks } SH top Similarly, smooth schemes can vary in families parametrized by algebraic stacks, and hence we might expectan interesting extension of S (cid:55)→ SH( S ) to algebraic stacks: { schemes } {∞ -categories } . { algebraic stacks } SH Of course, it is not hard to construct such extensions, and there are even several sensible possibilities. If atopological stack is presented by a simplicial topological space X • , one can consider the limitSH top ( X • ) = lim n ∈ ∆ SH top ( X n ) . For example, if G is a discrete group and B • G is the usual bar construction on G , then SH top ( B • G ) is the ∞ -category of spectra equipped with a homotopy coherent action of G . It turns out that SH top ( X • ) dependsonly on the topological stack presented by X • . In the same way one can plug into SH( − ) arbitrary simplicialschemes (and indeed arbitrary small diagrams of schemes); this construction was studied in an axiomaticsetting by Ayoub in [Ayo08, § − ) is that, unlike in topology,it depends on presentations of stacks rather than on stacks themselves, because SH( − ) does not satisfy ´etaledescent. Another extension of SH top ( − ) to topological stacks with a more geometric flavor is the so-called HE SIX OPERATIONS IN EQUIVARIANT MOTIVIC HOMOTOPY THEORY 3 “naive” stable equivariant homotopy theory. This coefficient system includes finer cohomological invariantsof topological stacks, like topological K -theory. Its main defect is that it does not satisfy Atiyah duality,which is an important feature of a formalism of six operations. This is rectified by passing to “genuine”stable equivariant homotopy theory. Although one could give a very general definition, this theory onlyworks well for nice enough stacks, such as quotients of topological spaces by actions of compact Lie groups.The principal example is the stack B G = [ ∗ /G ] for G a compact Lie group, in which case SH top ( B G ) is theusual ∞ -category of genuine G -spectra. A partial formalism of six operations in this context, encompassingonly locally constant coefficients, was developed by Hu [Hu03] and May–Sigurdsson [MS06].Our goal in this paper is to extend the assignment S (cid:55)→ SH( S ), together with its formalism of sixoperations, from schemes to a suitable class of algebraic stacks. In other words, our goal is to develop agood theory of “genuine” equivariant motivic spectra over varying base stacks. We will be able to achievethis, essentially, for stacks with affine stabilizers whose derived category is compactly generated by vectorbundles. These are somewhat analogous to quotients of spaces by compact Lie groups, in that they have awell-behaved representation theory compared to more general stacks.At this point we need to come clean about the fact that the word “stack” will not much appear beyond thisintroduction. Indeed, the above-mentioned restriction on stacks implies that we will not lose much generalityby considering only global quotient stacks [ X/G ] for a fixed algebraic group G (e.g., GL n for large n ). It willtherefore be much simpler to work directly with G -schemes rather than the corresponding quotient stacks.For a G -scheme X , we will write SH G ( X ) for SH([ X/G ]). The reader will rightly object that the categoryof G -schemes for fixed G only accounts for morphisms of stacks that are schematic (i.e., representable byschemes). However, this is not a significant drawback because the interesting features of our formalism ofsix operations only exist for schematic morphisms. For example, we do not expect an exceptional adjunction f ! (cid:97) f ! or a left adjoint f (cid:93) to f ∗ unless f is schematic. These restrictions already exist in topology: if G isa nontrivial compact Lie group, pullback along the morphism f : B G → ∗ does not have a left adjoint (inother words, there are no “genuine G -orbits”). Enforcing the existence of such left adjoints naturally leadsto a parametrized version of global homotopy theory in the sense of Schwede [Sch16], which should also havea motivic analog, but we will not discuss it further here.It is worth noting that all the intricacies of equivariant homotopy theory disappear in the ´etale version ofthe theory. Indeed, SH ´et ( − ) satisfies descent for the smooth topology and hence, if X is Artin stack (or evenan Artin ∞ -stack) presented by a simplicial scheme X • with smooth face maps, it is perfectly adequate todefine SH ´et ( X ) as the limit lim n ∈ ∆ SH ´et ( X n ). The result is a theory that is already “genuine” and “global”.This is why, modulo some serious technicalities, it is straightforward to extend the formalism of six operationsin (cid:96) -adic cohomology from schemes to Artin ∞ -stacks (see [LZ14]). Our work is thus motivated by invariantssuch as algebraic K -theory, Chow groups, and algebraic cobordism, which do not satisfy ´etale descent.1.2. The formalism of six operations.
Fix a quasi-compact quasi-separated base scheme B and a flatfinitely presented group scheme G over B . Throughout this introduction, we will assume for simplicity that B has the G -resolution property, i.e., that every finitely generated quasi-coherent G -module over B is thequotient of a locally free one.As explained in § S (cid:55)→ SH G ( S ) = SH([ S/G ]) from G -schemesto ∞ -categories, together with a formalism of six operations. We are only able to develop a good theoryunder the assumption that G is tame (see Definition 2.26). The following are the essential examples of tamegroup schemes: • G is finite locally free of order invertible on B ; • G is of multiplicative type; • G is reductive and B has characteristic 0 (i.e., there exists B → Spec Q ).Moreover, for the same reason as in [Ayo08, § G -schemes that are G -quasi-projective , i.e., that admit a G -equivariant immersion into the projectivization of a G -equivariantvector bundle over B . If G is finite locally free, any quasi-projective G -scheme is G -quasi-projective, and if G is reductive, any normal quasi-projective G -scheme is G -quasi-projective, by Sumihiro’s theorem [Sum75].In the nonequivariant setting, it was shown in [CD12, §
2] how to extend the formalism of six operations fromquasi-projective to more general schemes using Chow’s lemma. Similar tricks will work in our setting, butunfortunately the reach of the equivariant versions of Chow’s lemma is much more limited. For this reason,we will only discuss this generalization when G is discrete (see Remark 6.21). MARC HOYOIS
For G a tame group and S a finitely presented G -quasi-projective scheme (or an arbitrary qcqs G -schemeif G is discrete), we will construct a closed symmetric monoidal stable ∞ -category SH G ( S ). In particular,SH G ( S ) is equipped with a tensor product ⊗ and an internal mapping object Hom. For every G -equivariantmorphism f : T → S , we have a pullback–pushforward adjunction f ∗ : SH G ( S ) (cid:29) SH G ( T ) : f ∗ , where f ∗ is symmetric monoidal. If f is smooth, then f ∗ also admits a left adjoint f (cid:93) . If f is separated andof finite type, we further have the “exceptional” adjunction f ! : SH G ( T ) (cid:29) SH G ( S ) : f ! , where f ! is pushforward with compact support : for any factorization f = p ◦ j where j is an open immersionand p is proper, we have f ! (cid:39) p ∗ ◦ j (cid:93) . Our formalism of six operations is summarized in the following theorem: Theorem 1.1 (Theorem 6.18 and Proposition 6.23) . Let B be a qcqs scheme and G a tame group schemeover B . If G is not finite, we assume that B has the G -resolution property. Then the six operations ( − ) ∗ , ( − ) ∗ , ( − ) ! , ( − ) ! , ⊗ , Hom satisfy the following properties on finitely presented G -quasi-projective B -schemes (or on all qcqs G -schemesif G is discrete), whenever the exceptional functors are defined. (1) (Proper pushforward) If f is a proper G -morphism, there is an equivalence f ! (cid:39) f ∗ . (2) (Smooth pullback) If f is a smooth G -morphism, there is a self-equivalence Tw f and an equivalenceTw f ◦ f ! (cid:39) f ∗ . (3) (Base change) If • •• • gq f p is a cartesian square of G -schemes, there are equivalences f ∗ p ! (cid:39) q ! g ∗ ,f ! p ∗ (cid:39) q ∗ g ! . (4) (Gluing) If i is a closed G -immersion with complementary open G -immersion j , there are cofibersequences j ! j ! → id → i ∗ i ∗ ,i ! i ! → id → j ∗ j ∗ . (5) (Immersive pushforward) If i is a G -immersion, the functors i ∗ and i ! are fully faithful. (6) (Monoidality) If f is any G -morphism, there is an equivalence f ∗ ( − ⊗ − ) (cid:39) f ∗ ( − ) ⊗ f ∗ ( − ) . (7) (Projection formulas) If f is any G -morphism, there are equivalences f ! ( − ⊗ f ∗ ( − )) (cid:39) f ! ( − ) ⊗ − , Hom( f ! ( − ) , − ) (cid:39) f ∗ Hom( − , f ! ( − )) ,f ! Hom( − , − ) (cid:39) Hom( f ∗ ( − ) , f ! ( − )) . (8) (Homotopy invariance) If f is a G -affine bundle, the functors f ∗ and f ! are fully faithful. (9) (Constructible separation) If { f i } is a cover for the G -equivariant constructible topology, the familiesof functors { f ∗ i } and { f ! i } are conservative. HE SIX OPERATIONS IN EQUIVARIANT MOTIVIC HOMOTOPY THEORY 5 If G is finite locally free, standard model-categorical constructions provide a presentation of SH G ( S ) bya symmetric monoidal simplicial model category that is combinatorial and left proper. For more general G , however, we do not know an explicit such presentation, although there exists one by the main result of[NS15]. The language of ∞ -categories is also useful to make sense of the following statement, which is aneasy corollary of Theorem 1.1: Theorem 1.2 (Proposition 6.24) . Under the assumptions of Theorem 1.1, the assignment S (cid:55)→ SH G ( S ) , f (cid:55)→ f ∗ , is a sheaf for the G -equivariant cdh topology. Let us give some brief commentary on the proof of Theorem 1.1. In the nonequivariant case, Ayoub[Ayo08] and Cisinski–D´eglise [CD12] have shown how to derive such a formalism from just a few key results.Similarly, Theorem 1.1 is reduced by abstract nonsense to two nontrivial results: gluing for complementaryopen–closed pairs and ambidexterity for smooth proper morphisms. A third important result is purity forsmooth closed pairs: although purity does not directly enter the proof of Theorem 1.1, it plays a significantrole in our proof of ambidexterity; it also provides an explicit description of the twisting equivalence Tw f ,showing that it depends only on the relative tangent bundle of f . We will discuss these three key results inmore details in § G ( S ).In characteristic zero, the restriction to tame group schemes is not very significant. Indeed, a theoremof Gross [Gro15, Theorem A] states that any qcqs Artin stack with affine stabilizers and the resolutionproperty is the quotient of a quasi-affine scheme by an action of GL n . Thus, if B has characteristic zero, ourformalism of six operations includes all finitely presented Artin stacks over B with affine stabilizers and theresolution property. In arbitrary characteristic, it includes tame Deligne–Mumford stacks that are quotientsof quasi-projective schemes by finite ´etale groups, but also some Artin stacks that are not Deligne–Mumford(e.g., quotients by tori). Unfortunately, we do not know how to set up a satisfying theory for more generalstacks in positive characteristic.The formalism of six operations described above is not the most complete possible as it does not mentiondualizing objects, constructibility, and absolute purity. In ordinary stable motivic homotopy theory, a goodtheory of dualizing and constructible objects requires some form of resolutions of singularities [Ayo08, § loc. cit. can be applied in the present context assuming the existence ofsuitable equivariant resolutions of singularities, which are known to exist over fields of characteristic zero.Absolute purity, on the other hand, seems out of reach since it is not even known for SH( − ).1.3. Summary of the construction.
Let G be a tame group scheme over B and let S be a finitely presented G -quasi-projective B -scheme. As in the nonequivariant case, our construction of SH G ( S ) proceeds in severalsteps. First, we define the unstable equivariant motivic homotopy ∞ -category H G ( S ) as a localization ofthe ∞ -category of presheaves on the category Sm GS of smooth G -quasi-projective S -schemes. An object inH G ( S ) is thus a presheaf F on Sm GS , and it is subject to the following two conditions: • (Homotopy invariance) If Y → X is a G -equivariant torsor under a G -vector bundle, then therestriction map F ( X ) → F ( Y ) is an equivalence. • (Nisnevich excision) F ( ∅ ) (cid:39) ∗ , and if V → X is an ´etale map in Sm GS that is an isomorphism over a G -invariant finitely presented closed subscheme Z ⊂ X , then the following square is cartesian: F ( X ) F ( X (cid:114) Z ) F ( V ) F ( V (cid:114) Z ).In the context of equivariant algebraic K -theory, what we call homotopy invariance is sometimes called strong homotopy invariance to distinguish it from the weaker property of A -homotopy invariance. Thereare several reasons for using the stronger condition: perhaps the most important one is that it plays a crucialrole in our proof of ambidexterity; it also implies that H G ( S ) is generated by the affine schemes in Sm GS , afact that is used in the proofs of all the main results. The Nisnevich excision condition was first consideredby Voevodsky for finite locally free groups in [Del09], where it was shown to be equivalent to a topologicaldescent condition. It was further studied in [Her13, HVØ15, HKØ15].When B is noetherian of finite Krull dimension and G is a finite discrete group, the unstable equivariantmotivic homotopy ∞ -category H G ( B ) is equivalent to the underlying ∞ -category of the model category MARC HOYOIS constructed by Heller, Krishna, and Østvær in [HKØ15]; see Remark 3.14. For more general tame groups G ,however, we do not know if they agree. In any case, our unstable category does not appear to be equivalentto the constructions in [HKO11] and [Her13], where a finer version of the Nisnevich topology is used.The following are our two main results about H G ( S ). They are equivariant generalizations of the purityand gluing theorems of Morel and Voevodsky [MV99, §
3, Theorems 2.23 and 2.21].
Theorem 1.3 (Theorem 3.23) . Let S be a G -scheme and let Z (cid:44) → X be a G -equivariant closed immersionin Sm GS . Deformation to the normal bundle induces a canonical equivalence XX (cid:114) Z (cid:39) N Z XN Z X (cid:114) Z in H G ( S ) , where N Z X is the normal bundle of Z in X with the induced action of G . Theorem 1.4 (Theorem 4.18) . Let i : Z (cid:44) → S be a G -equivariant closed immersion with open complement j : U (cid:44) → S . Then, for every F ∈ H G ( S ) , there is a cocartesian square j (cid:93) j ∗ F FU i ∗ i ∗ F . Theorem 1.3 was also proved in [HKØ15, Theorem 7.6] assuming that S is the spectrum of a perfect field,that G is a finite discrete abelian group acting trivially on S , and that S contains a primitive e th root ofunity, where e is the exponent of G .Our proofs of Theorems 1.3 and 1.4 follow the same ideas as in [MV99, § G -equivariant morphism is not locally the composition of an ´etale map and a vector bundle, even for G finitediscrete. For gluing, an additional complication comes from the fact that, if G is not discrete, the inclusion ofthe subcategory of affine G -schemes in Sm GS is not necessarily cocontinuous for the Nisnevich topology. Weremark that these two theorems already fully exploit the definition of H G ( S ), in the sense that no obviousweakening of the conditions of homotopy invariance and Nisnevich excision would make them work.Let H G • ( S ) denote the undercategory H G ( S ) S/ , i.e., the ∞ -category of pointed presheaves on Sm GS that arehomotopy invariant and Nisnevich excisive. It is a symmetric monoidal ∞ -category under the “smash prod-uct”, which we denote by ⊗ . For every locally free G -module of finite rank E on S , there is a corresponding“representation sphere” S E ∈ H G • ( S ) and a suspension functor Σ E defined byS E = V ( E ) V ( E ) (cid:114) S and Σ E = S E ⊗ ( − ) , where V ( E ) = Spec(Sym( E )). The stable equivariant motivic homotopy ∞ -category SH G ( S ) is then definedby formally adjoining a ⊗ -inverse S − E for each representation sphere S E :SH G ( S ) = H G • ( S )[S − E | E is a locally free G -module of finite rank on S ] . We denote by Σ ∞ : H G • ( S ) → SH G ( S ) the canonical functor. The precise definition of SH G ( S ) is a universalconstruction in the ∞ -category of presentably symmetric monoidal ∞ -categories, and we do not spell it outhere. We only note that the symmetric monoidal ∞ -category of genuine G -spectra can be obtained in thesame way from that of pointed G -spaces. In the equivariant motivic setting, this stabilization procedurerecovers the ones considered in [HKO11] and [Her13] (for finite discrete groups), but it has the advantage thatthe resulting ∞ -category SH G ( S ) comes equipped with a homotopy coherent symmetric monoidal structure.Our final main result is what we call ambidexterity , since it identifies, up to a twist, the left and rightadjoints of the base change functor: Theorem 1.5 (Theorem 6.9) . Let f : X → S be a smooth proper G -morphism. Then there is a canonicalequivalence f ∗ (cid:39) f (cid:93) Σ − Ω f , where f (cid:93) and f ∗ are left and right adjoint to f ∗ : SH G ( S ) → SH G ( X ) . Here, Ω f is the sheaf of differentials of X over S , which is a locally free G -module of finite rank on X .An easy consequence is Atiyah duality , which states that Σ ∞ X + is strongly dual to the Thom spectrum ofthe stable normal bundle of X over S in the symmetric monoidal ∞ -category SH G ( S ). HE SIX OPERATIONS IN EQUIVARIANT MOTIVIC HOMOTOPY THEORY 7
In the nonequivariant case, this ambidexterity theorem was proved by Ayoub [Ayo08, Th´eor`eme 1.7.17].His argument implicitly relies on the fact that any vector bundle is Zariski-locally a sum of line bundles. Thesame argument can be used to prove Theorem 1.5 if X can be embedded into a projective bundle P ( E ) where E is a sum of G -line bundles over S . This suffices to prove Atiyah duality in the stable equivariant motivichomotopy category of a diagonalizable group over an algebraically closed field, but it is far from sufficient ingeneral. Our strategy to prove Theorem 1.5 is to explicitly write down a unit and a counit for an adjunctionbetween f ∗ and f (cid:93) Σ − Ω f , and to verify the triangle identities. This is rather interesting even in the case ofa trivial group, as Ayoub’s proof does not provide a geometric description of this adjunction. The centralconstruction is an algebro-geometric analog of the Pontryagin–Thom collapse map, which is essentially dueto Voevodsky [Voe03, Theorem 2.11], although he only used it to prove a pale motivic-cohomological shadowof Atiyah duality.In fact, we will prove a finer unstable version of Theorem 1.5 (Theorem 5.22), which shows that thepassage from H G • ( − ) to SH G ( − ) is exactly what is needed to enforce ambidexterity for smooth projectivemorphisms. For instance, if f : P nS → S is the projection, we will show that the unit and counit for theadjunction f ∗ (cid:97) f (cid:93) Σ − Ω f , as well as homotopies witnessing the triangle identities, are in the image of Σ ∞ after tensoring with S O at least 2 n + 4 n + 1 times.1.4. Some applications.
Motivic Wirthm¨uller and Adams isomorphisms.
As was observed by Hu [Hu03, § f ∗ (cid:39) f (cid:93) Σ − Ω f . One can also consider these instances in the motivic context. However, they arenot isomorphisms in general, because quotients of tame groups are not always proper.Let G be a tame group scheme over B and let H ⊂ G be a subgroup. Suppose that the quotient B -scheme G/H exists and is smooth and G -quasi-projective over B . Let p : G/H → B be the structural G -morphism,presenting the morphism of stacks B H → B G . The functor p ∗ : SH G ( B ) → SH G ( G/H ) (cid:39) SH H ( B ) isthus the “forgetful” functor from G -spectra to H -spectra, and its left and right adjoints p (cid:93) and p ∗ are theinduction and coinduction functors, respectively. The Wirthm¨uller morphism is then the composition p (cid:93) Σ − Ω G/H (cid:39) p ! → p ∗ ;it is an equivalence if G/H is proper. See [May03, Theorem 1.1] for a statement of the classical Wirthm¨ullerisomorphism in this form.The Adams morphism is more subtle as it involves a nonschematic pushforward (see Remark 6.20). Let H ⊂ G be as above and let N be a smooth normal subgroup of G such that G/N is tame and N ∩ H istrivial. Assume further that G/N H is a
G/N -quasi-projective B -scheme. We then have smooth morphismsof stacks B H p → B G q → B ( G/N ) , with p and qp schematic, and a canonical map ( qp ) ! → q ∗ p ! . Let n be the Lie algebra of N with conjugationaction of G , viewed as a vector bundle over B G . The canonical fiber sequence of cotangent complexes over B H induces an equivalence Σ − Ω p Σ Ω qp (cid:39) Σ − p ∗ ( n ) . The Adams morphism for H -induced G -spectra is thenthe composition ( qp ) (cid:93) (cid:39) ( qp ) ! Σ Ω qp → q ∗ p ! Σ Ω qp (cid:39) q ∗ p (cid:93) Σ − Ω p Σ Ω qp (cid:39) q ∗ Σ − n p (cid:93) ;it is an equivalence if G/H is proper. In classical notation, given an H -spectrum X , the above morphismreads G/N (cid:110) H X → (Σ − n ( G (cid:110) H X )) N . As in the classical case, one can do slightly better and construct anAdams morphism for suitably defined “ N -free G -spectra”, but this requires some variants of our results.1.4.2. An equivariant Lefschetz trace formula.
The formalism of six operations developed here implies anabstract version of the Lefschetz trace formula, as in [Hoy14, Theorem 1.3]. Let S be a G -scheme, let p : X → S be a smooth G -projective morphism, let f : X → X be a G -equivariant endomorphism over S ,and let i : X f (cid:44) → X be the inclusion of the scheme of fixed points of f . Suppose that X f is smooth over S and that the endomorphism id − i ∗ ( df ) of i ∗ (Ω X ) restricts to an automorphism φ of the conormal sheaf N i .Then the trace of Σ ∞ f + in SH G ( S ) is equal to the trace of the automorphism of Σ ∞ X f + induced by φ viathe J -homomorphism.When B is the spectrum of a field and G is finite and discrete, the group of endomorphisms of the G -equivariant motivic sphere spectrum over B was recently computed by Gepner and Heller [Hel16] in terms MARC HOYOIS of a motivic tom Dieck splitting. It would be interesting to express the right-hand side of the trace formulain terms of their computation.1.4.3.
Cdh descent for the homotopy K -theory of tame stacks. The question of cdh descent for the homotopy K -theory of tame Deligne–Mumford stacks was raised in [KØ12, §
1] (it was answered affirmatively forschemes in [Cis13]). Our formalism of six operations yields a positive answer to this question for nice enoughstacks. We only sketch the proof here, the details of which appear in [Hoy17]. Suppose that G is a tamefinite ´etale group scheme over B . For S a quasi-projective G -scheme, let K G ( S ) denote the nonconnective K -theory of perfect complexes on the quotient stack [ S/G ]. Following Weibel [Wei89], define the homotopy K -theory KH G ( S ) by the formula(1.6) KH G ( S ) = colim n ∈ ∆ op K G ( S × A n ) . It was shown in [KØ12, Corollary 5.6] that K G satisfies Nisnevich excision. Using the projective bundleformula from [Tho87a], one can show that the restriction of KH G to Sm GS is represented by an objectKGL S ∈ SH G ( S ). Furthermore, the tameness of G allows us to generalize the results of [MV99, § G -space of KGL S is equivalentto the group completion of a certain monoid structure on the motivic localization of (cid:96) n ≥ colim E Gr n ( E S ),where E ranges over some cofiltered diagram of locally free G -modules on B , independent of S . It followsfrom this explicit description that S (cid:55)→ KGL S is a cartesian section of SH G ( − ). By Corollary 6.25, wetherefore deduce that KH G satisfies cdh descent on quasi-projective G -schemes.1.5. Acknowledgements.
It is my pleasure to thank Joseph Ayoub, David Gepner, Jeremiah Heller, AdeelKhan, Amalendu Krishna, Marc Levine, and Paul Arne Østvær for many helpful and motivating discussionson equivariant motivic homotopy theory and the six operations. I am especially grateful to Elden Elmantofor carefully reading the paper and making numerous comments. This work was partially supported by theNational Science Foundation under grant DMS-1508096.1.6.
Notation and terminology.
As a matter of terminology, we assume that group schemes are flat andfinitely presented, and that locally free sheaves are of finite (but not necessarily constant) rank. A vectorbundle over a scheme X is a module scheme over X that is isomorphic to V ( E ) = Spec(Sym( E )) for somelocally free sheaf of finite rank E on X . Thus, V is a contravariant equivalence between locally free sheavesand vector bundles. We write P ( E ) = Proj(Sym( E )) for the projective bundle of lines in V ( E ). If f : X → S is a morphism of schemes, we denote by Ω f , Ω X/S , or even Ω X the sheaf of relative differentials of f . If i : Z (cid:44) → X is an immersion, we denote by N i its conormal sheaf.Starting with the definition of the unstable equivariant motivic homotopy ∞ -category in §
3, we will usethe language of ∞ -categories [Lur09]. We denote by S the ∞ -category of ∞ -groupoids and by C at ∞ the ∞ -category of (possibly large) ∞ -categories. Categorical terminology must always be understood in the ∞ -categorical context; for example, a presheaf or sheaf is S -valued unless otherwise specified. We denote by P ( C ) the ∞ -category of presheaves on a small ∞ -category C .2. Some equivariant geometry
Throughout this section, B is an arbitrary base scheme and G is a flat finitely presented group schemeover B .2.1. Invariant subschemes.
Let X be a G -scheme and let Y ⊂ X be a subscheme of X . We say that Y is G -invariant if there exists a morphism G × Y → Y making the square G × Y YG × X X a commute, where a : G × X → X is the action. If such a morphism exists, it is clearly unique and is an actionof G on Y .Note that the action a : G × X → X is isomorphic to the projection onto the second factor, and henceit is flat and finitely presented. In particular, it is open. Thus, if U ⊂ X is any open subscheme, then HE SIX OPERATIONS IN EQUIVARIANT MOTIVIC HOMOTOPY THEORY 9 G · U = a ( G × U ) is an open subscheme of X , and it is clearly the smallest G -invariant open subscheme of X containing U . Moreover, if U is quasi-compact, then G × U and hence G · U are also quasi-compact. Lemma 2.1.
Let X be a G -scheme, Z ⊂ X a closed subscheme, and U = X (cid:114) Z its open complement.Then U is G -invariant if and only if Z admits a G -invariant thickening, which may be chosen to be finitelypresented if X and U are qcqs.Proof. Suppose that Z admits a G -invariant thickening Z (cid:48) , so that we have a cartesian square G × Z (cid:48) Z (cid:48) G × X X . a Then a maps the open complement of G × Z (cid:48) to the open complement of Z (cid:48) , i.e., U is G -invariant.For the converse, we consider more generally a groupoid scheme d , d : X ⇒ X where d and d areflat and finitely presented. Given a closed subscheme Z ⊂ X with invariant open complement U , we willshow that there exists an invariant thickening Z (cid:48) of Z (in this context, a subscheme Y ⊂ X is invariant if d − ( Y ) = d − ( Y )). Let Z (cid:48) be the schematic image of d − ( Z ) by d . Note that there are (isomorphic)cartesian squares of the form: P X d − ( Z ) X , fd d P X d − ( Z ) X . fd d Since flat base change preserves schematic images [Stacks, Tag 081I], the first cartesian square tells us thatthe schematic image of f is d − ( Z (cid:48) ). The second cartesian square then tells us that d − ( Z (cid:48) ) = d − ( Z (cid:48) ) asclosed subschemes of X , i.e., that Z (cid:48) is invariant. It remains to prove that Z (cid:48) ∩ U = ∅ . Since U (cid:44) → X is flat, Z (cid:48) ∩ U is the schematic image of d : d − ( Z ) ∩ d − ( U ) → U , which is empty since U is invariant.The last statement is proved using noetherian approximation, see [Gro66, §
8] and [TT90, Appendix C].Assume that X is qcqs. Then there exists a cartesian morphism of groupoid schemes X • → Y • where Y • is of finite type over Z and the maps Y ⇒ Y are flat and finitely presented [Stacks, Tag 04AI]. If U is quasi-compact, then Z admits a finitely presented thickening [TT90, Lemma 2.6.1 (c)], so we can alsoassume that Z is pulled back from a closed subscheme of Y with an invariant open complement. Applyingthe result of the previous paragraph to Y • shows that Z admits a finitely presented invariant thickening. (cid:3) Lemma 2.2.
Let X be a G -scheme and V ⊂ X a G -invariant subscheme. (1) The largest open subset of X in which V is closed is G -invariant. If moreover V is quasi-compact,there exists a quasi-compact G -invariant open subscheme of X in which V is closed. (2) If V (cid:44) → X is quasi-compact, then the schematic closure of V in X is G -invariant. If moreover X isqcqs and V (cid:44) → X is finitely presented, there exists a finitely presented G -invariant closed subschemeof X in which V is open.Proof. (1) Let U ⊂ X be an open subset in which V is closed. It will suffice to show that V is closed in G · U . Since V is G -invariant, a − ( V ) = G × V , where a : G × X → X is the action. Thus, ( G · U ) (cid:114) V = a (( G × U ) (cid:114) a − ( V )) = a ( G × ( U (cid:114) V )), which is open in X since a is open.(2) As in the proof of Lemma 2.1, we consider the more general situation of a groupoid scheme d , d : X ⇒ X , with d and d flat and finitely presented and V ⊂ X an invariant subscheme. The last part of thestatement can then be proved by noetherian approximation. Let ¯ V be the schematic closure of V in X .Consider the cartesian squares d − ( V ) X V X , d d d − ( V ) X V X , d d and recall that flat base change preserves schematic images [Stacks, Tag 081I]. The first cartesian squaretells us that the schematic closure of d − ( V ) in X is d − ( ¯ V ). The second cartesian square then tells us that d − ( ¯ V ) = d − ( ¯ V ), i.e., that ¯ V is invariant. (cid:3) A G -immersion is a G -morphism that is also an immersion. By Lemma 2.2 (1), a G -immersion can befactored canonically as a closed G -immersion followed by an open G -immersion. Lemma 2.3.
Let f : X → S be a G -morphism and Y ⊂ X a subset. If f is smooth (resp. ´etale) at thepoints of Y , then f is smooth (resp. ´etale) on a G -invariant open neighborhood of Y , which may be chosento be quasi-compact if Y is quasi-compact.Proof. Let U be an open subset of X where f is smooth (resp. ´etale). It will suffice to show that f is smooth(resp. ´etale) on G · U . Let x ∈ G · U be any point. Then x = a ( y ) for some y ∈ G × U , where a : G × X → X is the action. Note that the square G × X XG × S S aa id × f f is cartesian. Since id × f is smooth (resp. ´etale) at y and a is flat, it follows from [Gro66, 17.7.1 (ii)] that f is smooth (resp. ´etale) at x . (cid:3) Lemma 2.4.
Let
Z XY st p be a commutative triangle of G -schemes where s and t are closed immersions and p is unramified. Thenthere exists a G -invariant open subscheme U ⊂ X containing Z such that the square Z UZ Y st p is cartesian. If Z is quasi-compact, then U may be chosen to be quasi-compact.Proof. Since p is unramified, the G -immersion (id , s ) : Z (cid:44) → Z × Y X is open. We can then take U to bethe complement of Z × Y X (cid:114) Z in X , which is a G -invariant open subscheme of X by Lemma 2.1. If Z isquasi-compact, we can replace U by G · V where V ⊂ U is a quasi-compact open subset containing Z . (cid:3) Quasi-affine and quasi-projective morphisms.Definition 2.5.
Let f : X → Y be a G -morphism.(1) f is G -quasi-affine (resp. G -affine ) if there exists a locally free G -module E on Y and a quasi-compact G -immersion (resp. a closed G -immersion) X (cid:44) → V ( E ) over Y .(2) f is G -quasi-projective (resp. G -projective ) if there exists a locally free G -module E on Y and aquasi-compact G -immersion (resp. a closed G -immersion) X (cid:44) → P ( E ) over Y . Lemma 2.6.
Let C be any of the four classes of G -morphisms introduced in Definition 2.5. (1) C is closed under base change. (2) If g is separated and g ◦ f belongs to C , so does f .Proof. Easy. (cid:3)
On the other hand, it is not true that the classes of morphisms from Definition 2.5 are stable undercomposition (even if G is trivial!). HE SIX OPERATIONS IN EQUIVARIANT MOTIVIC HOMOTOPY THEORY 11
Definition 2.7 ([Tho87b, Definition 2.1]) . Let X be a G -scheme. We say that X has the G -resolutionproperty if, for every finitely generated quasi-coherent G -module M on X , there exists a locally free G -module of finite rank E and an epimorphism E (cid:16) M . Example 2.8.
Suppose that B is divisorial (e.g., affine, or noetherian, regular, and separated). Then B hasthe G -resolution property in the following cases:(1) G is finite locally free;(2) G is isotrivial;(3) G is reductive with isotrivial radical and coradical (e.g., G is semisimple).This is proved in [Tho87b] under the assumption that B is noetherian. An easy noetherian approximationargument implies the general case. Remark . Any finitely presented group of multiplicative type over B is isotrivial locally in the Nisnevichtopology. Indeed, if B is a henselian local scheme, any ´etale cover of B is refined by a finite ´etale cover, by[Gro67, Th´eor`eme 18.5.11 (c)]. Lemma 2.10.
Let X be a qcqs G -scheme. Every quasi-coherent G -module on X is the colimit of its finitelygenerated quasi-coherent G -submodules.Proof. Let X • be the groupoid scheme defined by the action of G on X . Since X is qcqs and G is flat andfinitely presented, there exists an affine cartesian morphism of groupoid schemes p : X • → Y • , where Y • is offinite type over Z and the maps Y ⇒ Y are flat and finitely presented. Let M be a quasi-coherent G -moduleon X . By [Stacks, Tag 09VH], p ∗ ( M ) has a structure of quasi-coherent module on the groupoid Y • such thatthe counit map p ∗ p ∗ ( M ) → M is a map of G -modules. Since p is affine, this counit map is an epimorphism.Thus, it suffices to show that p ∗ ( M ) is the union of its finitely generated quasi-coherent submodules on thegroupoid Y • . This is true by [Stacks, Tag 07TU]. (cid:3) Lemma 2.11.
Let f : X → Y be a G -quasi-projective G -morphism. If Y is qcqs and has the G -resolutionproperty, then X has the G -resolution property.Proof. One can repeat the proof of [Tho87b, Lemma 2.6], using Lemma 2.10. Details are omitted. (cid:3)
Lemma 2.12.
Let f : X → Y be a quasi-affine (resp. affine) G -morphism of finite type. If Y is qcqs andhas the G -resolution property, then f is G -quasi-affine (resp. G -affine).Proof. We repeat the argument from [Tho87b, Theorem 3.6]. Since f is quasi-affine, f ∗ ( O X ) is a quasi-coherent G -algebra and we have an open G -immersion X (cid:44) → Spec( f ∗ ( O X )) over Y (an isomorphism if f is affine). Let M be a finitely generated quasi-coherent G -submodule of f ∗ ( O X ) that generates f ∗ ( O X ) asan algebra (such an M exists by Lemma 2.10). By the G -resolution property, there exists a locally free G -module E and an epimorphism E (cid:16) M . Then the epimorphism of G -algebras Sym( E ) (cid:16) f ∗ ( O X ) inducesa closed G -immersion Spec( f ∗ ( O X )) (cid:44) → V ( E ) over Y . Thus, f is G -(quasi)-affine. (cid:3) Lemma 2.13.
Let C be any of the four classes of G -morphisms introduced in Definition 2.5 and let f : X → Y and g : Y → Z belong to C . If Z is qcqs and has the G -resolution property, then g ◦ f belongs to C .Proof. The case of G -(quasi)-affine morphisms follows from Lemma 2.12. Let us treat the case of G -(quasi)-projective morphisms. By definition, there exist locally free G -modules F on Y and G on Z , and quasi-compact (closed) G -immersions X (cid:44) → P ( F ) over Y and Y (cid:44) → P ( G ) over Z . Let us abbreviate F ⊗ O Y (1) ⊗ n to F ( n ). By [Gro61, Proposition 4.6.8] (amended by [Gro64, 1.7.15]), there exists n such that the counit g ∗ g ∗ ( F ( n )) → F ( n ) is an epimorphism. Since F ( n ) is finitely generated and g ∗ ( F ( n )) is the union of its finitelygenerated quasi-coherent G -submodules (Lemma 2.10), there exists a finitely generated quasi-coherent G -submodule M of g ∗ ( F ( n )) such that g ∗ ( M ) → F ( n ) is an epimorphism. Then, by the G -resolution property,there exists a locally free G -module E and an epimorphism E (cid:16) M . Thus, we obtain a closed G -immersion P ( F ) (cid:44) → P ( g ∗ ( E )) over Y , whence a quasi-compact (closed) G -immersion X (cid:44) → P ( E ) × P ( G ) over Z . Weconclude using the Segre embedding P ( E ) × P ( G ) (cid:44) → P ( E ⊗ G ), which is a closed G -immersion. (cid:3) Linearly reductive groups. If S is a G -scheme, we denote by QCoh G ( S ) the category of quasi-coherent G -modules on S . Definition 2.14 ([AOV08, Definition 2.2]) . A group scheme G over B is called linearly reductive if the G -fixed-point functor ( − ) G : QCoh G ( B ) → QCoh( B ) is exact. Lemma 2.15 ([AOV08, Proposition 2.4]) . (1) If { f i : U i → B } is an fpqc cover and G U i is linearly reductive for all i , then G is linearly reductive. (2) If G is linearly reductive and f : B (cid:48) → B is any morphism, then G B (cid:48) is linearly reductive.Proof. Assertion (1) is clear since the family of functors { f ∗ i } detects exactness and each f ∗ i commutes with( − ) G . Given (1), it suffices to prove (2) when f is an open immersion and when f is affine. In the latter casethe result is clear since f ∗ is exact and conservative. Suppose that f is an open immersion. Then both f ∗ and f ∗ commute with ( − ) G , f ∗ is exact, and f ∗ is fully faithful. Let ( M α ) be a finite diagram in QCoh G ( B (cid:48) ).Then the computationcolim α M Gα (cid:39) colim α f ∗ f ∗ ( M Gα ) (cid:39) f ∗ colim α f ∗ ( M Gα ) (cid:39) ( f ∗ colim α f ∗ M α ) G (cid:39) (colim α f ∗ f ∗ M α ) G (cid:39) (colim α M α ) G shows that ( − ) G : QCoh G ( B (cid:48) ) → QCoh( B (cid:48) ) is right exact, i.e., that G B (cid:48) is linearly reductive. (cid:3) Example 2.16. (1) A finite locally free group of order invertible on B is linearly reductive [AOV08, Theorem 2.16].(2) A group of multiplicative type is linearly reductive (combine [DG70, Expos´e I, Th´eor`eme 5.3.3] andLemma 2.15 (1)).(3) If B has characteristic zero, any reductive group over B is linearly reductive (for fields, see [Nag61];the general case follows by Lemma 2.15 since any reductive group is ´etale-locally split).Linear reductivity will be used exclusively through the following lemma: Lemma 2.17.
Suppose that B is affine and that G is linearly reductive, and let p : S → B be an affine G -morphism. If M is a locally free G -module of finite rank on S , then M is projective in QCoh G ( S ) .Proof. We need to show that Hom( M , − ) : QCoh G ( S ) → A b is an exact functor. We haveHom( M , − ) (cid:39) Γ B ◦ ( − ) G ◦ p ∗ ◦ H om( M , − ) , where: • H om : QCoh G ( S ) × QCoh G ( S ) → QCoh G ( S ) is the internal hom object in QCoh G ( S ); • p ∗ : QCoh G ( S ) → QCoh G ( B ) is the pushforward along p ; • ( − ) G : QCoh G ( B ) → QCoh( B ) is the G -fixed point functor; • Γ B : QCoh( B ) → A b is the global section functor.Now, H om( M , − ) is exact because M is locally free of finite rank, p ∗ is exact because p is affine, ( − ) G isexact because G is linearly reductive, and Γ B is exact because B is affine. Thus, Hom( M , − ) is exact. (cid:3) Affine resolutions.Definition 2.18.
Let X be a G -scheme. A G -affine bundle is a G -morphism Y → X that is a torsor undera G -vector bundle V over X , such that the action V × X Y → Y is G -equivariant.If Aff r is the group scheme of affine automorphisms of A r and B Aff = (cid:96) r ≥ B Aff r , then G -affine bundlesover X are classified by morphisms of stacks [ X/G ] → B Aff . Using the obvious embeddings Aff r (cid:44) → GL r +1 , we see that the category of G -affine bundles over a G -scheme X is equivalent to the full subcategory of QCoh G ( X ) / O X spanned by the epimorphisms φ : E (cid:16) O X , where E is locally free of finite rank. The G -affine bundle Y corresponding to φ is the X -scheme that classifiessplittings of φ . In other words, Y is the preimage of the 1-section by the epimorphism V ( E ∨ ) (cid:16) A X , orequivalently the complement of P (ker( φ ) ∨ ) in P ( E ∨ ), and it is a torsor under V (ker( φ ) ∨ ). HE SIX OPERATIONS IN EQUIVARIANT MOTIVIC HOMOTOPY THEORY 13
We now recall Jouanolou’s trick, with an equivariant twist. Let S be a G -scheme, let E be a locally free G -module of rank ≥ S , and let p : P ( E ) → S be the associated projective bundle. We then have ashort exact sequence(2.19) 0 → O S → E ⊗ E ∨ → Q → G ( S ), where the first map is the coevaluation. The naturality of the coevaluation implies that thequotient Q is again locally free. We therefore have a G -invariant hyperplane P ( Q ) (cid:44) → P ( E ⊗ E ∨ )whose open complement is an affine G -scheme over S . Let ˜ P ( E ) be the preimage of P ( E ⊗ E ∨ ) (cid:114) P ( Q ) by theSegre embedding σ : P ( E ) × S P ( E ∨ ) (cid:44) → P ( E ⊗ E ∨ ) , and let π : ˜ P ( E ) → P ( E ) be the restriction of the projection onto the first factor. By construction, ˜ P ( E )is affine over S ; it is even G -affine over S if S has the G -resolution property (Lemma 2.12) or if theexact sequence (2.19) splits equivariantly, for example if B and S are affine and G is linearly reductive(Lemma 2.17).We claim that π is G -affine bundle. Consider the canonical epimorphism φ : p ∗ ( E )( − (cid:16) O P ( E ) in QCoh G ( P ( E )), whose kernel is the sheaf of differentials Ω P ( E ) /S . Over a T -point α : E T (cid:16) L of P ( E ), φ isgiven by the composition E T ⊗ L ∨ α ∨ −−→ E T ⊗ E ∨ T → O T , where the second map is the evaluation. The G -affine bundle corresponding to φ is then the complement of P (Ω ∨ P ( E ) /S ) in P ( p ∗ ( E ) ∨ ). Under the isomorphism P ( p ∗ ( E ) ∨ ) (cid:39) P ( E ) × S P ( E ∨ ), it is clear that this complementis exactly ˜ P ( E ). This shows that π is a G -equivariant torsor under the cotangent bundle of P ( E ). Proposition 2.20 (Jouanolou’s trick) . Let S be a qcqs G -scheme with the G -resolution property. Then, forevery G -quasi-projective S -scheme X , there exists a G -affine bundle ˜ X → X where ˜ X is affine over S .Proof. If X = P ( E ), we have just proved this. It is then clear how to obtain such a bundle if X is G -projective.In general, it suffices to construct an affine G -morphism X → P where P is G -projective: we then let ˜ X bethe pullback of ˜ P . Choose a locally free G -module F and a quasi-compact G -immersion X (cid:44) → P ( F ) over S .Let ¯ X be a G -invariant closed subscheme of P ( F ) of which X is an open subscheme (Lemma 2.2 (2)), let Z = Spec( O ¯ X / I ) be a G -invariant closed subscheme of ¯ X with open complement X (Lemma 2.1), and let P be the blowup of ¯ X at Z . Since ¯ X has the G -resolution property, there exists a locally free G -module G and an epimorphism G (cid:16) I , whence a closed G -immersion P (cid:44) → P ( G ) over ¯ X , so that P is G -projective(Lemma 2.13). Finally, the open G -immersion X (cid:44) → P factors through a closed G -immersion X (cid:44) → V ( J ∨ )over P , where J is the ideal of the exceptional divisor in P , and in particular it is affine. (cid:3) Lifting locally free sheaves.
Given a G -scheme X and a G -invariant closed subscheme Z ⊂ X , weshall say that a G -morphism f : X (cid:48) → X is a G -equivariant ´etale neighborhood of Z if it is locally finitelypresented, if the induced map Z × X X (cid:48) → Z is an isomorphism, and if f is ´etale at all points lying over Z .In that case, by Lemma 2.3, there exists a G -invariant open subscheme V ⊂ X (cid:48) such that f | V is ´etale and isan isomorphism over Z . However, it will be important to allow f itself not to be ´etale, since we often need X (cid:48) to be affine but we cannot guarantee that V is affine.The following theorem is an equivariant generalization of a theorem of Arabia [Ara01, Th´eor`eme 1.2.3]. Theorem 2.21.
Suppose that B is affine and that G is linearly reductive. Let s : Z (cid:44) → X be a closed G -immersion between affine G -schemes and let N be a locally free G -module on Z . If X has the G -resolutionproperty, there exists an affine G -equivariant ´etale neighborhood X (cid:48) → X of Z and a locally free G -module M on X (cid:48) lifting N .Proof. Since s ∗ ( E ) is finitely generated and X has the G -resolution property, there exists a locally free G -module E on X and an epimorphism E (cid:16) s ∗ ( N ), whence an epimorphism s ∗ ( E ) (cid:16) N . Since N is projective inQCoh G ( Z ) by Lemma 2.17, there exists a G -equivariant idempotent endomorphism φ of s ∗ ( E ) whose imageis isomorphic to N . Since E is projective in QCoh G ( X ) and the unit map E → s ∗ s ∗ ( E ) is an epimorphism, there exists a G -equivariant endomorphism ψ of E such that s ∗ ( ψ ) = φ . The idea is now to consider theuniversal deformation of ψ , and the locus where it is idempotent will be the desired ´etale neighborhood.Let I ⊂ O X be the ideal of s . By Lemma 2.10, I is the union of its finitely generated quasi-coherent G -submodules. Since φ is idempotent, ψ − ψ has image in IE , hence in I (cid:48) E for some finitely generated I (cid:48) ⊂ I . If Z (cid:48) ⊂ X is the closed subscheme defined by I (cid:48) , it follows that the idempotent φ and hence N lift to Z (cid:48) . Replacing Z by Z (cid:48) , we can therefore assume that I is finitely generated. In that case, by the G -resolution property, there exists a locally free G -module F on X and an epimorphism F (cid:16) I . Let π denotethe composition F (cid:16) I ⊂ O X . By projectivity of E , we can lift ψ − ψ : E → IE to a morphism α : E → F ⊗ E ,so that ψ − ψ = πα . Let p : V → X be the G -vector bundle whose sheaf of sections is H om( E , F ⊗ E ), let β : p ∗ ( E ) → p ∗ ( F ⊗ E ) be the tautological morphism, and let R = α + (2 ψ − id) β + πβ : p ∗ ( E ) → p ∗ ( F ⊗ E ) . This should be understood as follows: locally where F is free, π is a collection of generators π , . . . , π n of I , α is collection of endomorphisms α , . . . , α n of E such that ψ − ψ = (cid:80) i π i α i , and β is the universal familyof n endomorphisms of E . Moreover, πR = ( ψ + πβ ) − ( ψ + πβ ) + [ ψ, πβ ], so the equation πR = 0 expressesthe idempotency of ψ + πβ wherever ψ and πβ commute.Let i : X (cid:48) (cid:44) → V be the locus where R and [ ψ, πβ ] both vanish, so that ψ + πβ is an idempotent endomor-phism of i ∗ p ∗ ( E ). Let M be its image. By construction, M and N are isomorphic over Z × X X (cid:48) . Since φ isidempotent, 2 φ − id is an automorphism of s ∗ ( E ), and since R ≡ α + (2 ψ − id) β modulo I , it follows that p ◦ i : X (cid:48) → X induces an isomorphism Z × X X (cid:48) (cid:39) Z . It remains to check that p ◦ i is ´etale at all pointslying over Z . Since this question is local on X and does not involve G , we can assume that E and F arefree, so that F specifies global generators π , . . . , π n of I . By induction on n , we can further assume that n = 1, i.e., that I is a principal ideal generated by π . Let X (cid:48)(cid:48) ⊂ V be the locus where R vanishes, so that X (cid:48) ⊂ X (cid:48)(cid:48) is the locus where [ ψ, πβ ] vanishes. Note that X (cid:48)(cid:48) → X is also an isomorphism over Z . In theproof of [Ara01, Th´eor`eme 1.2.3], Arabia shows that X (cid:48)(cid:48) → X is ´etale over Z and that [ ψ, πβ ] vanishes inan open neighborhood of Z in X (cid:48)(cid:48) . This implies that X (cid:48) → X is also ´etale over Z , as desired. (cid:3) Lifting smooth quasi-sections.Theorem 2.22.
Suppose that B is affine and that G is linearly reductive. Let X Z XZ S t ps be a cartesian square of G -schemes where X is affine and s is a closed immersion. Let V ⊂ X Z be a finitelypresented G -invariant closed subscheme of X Z that is smooth (resp. ´etale) over Z . Suppose that p is smoothat each point of V and that X has the G -resolution property. Then there exists an affine G -equivariant ´etaleneighborhood X (cid:48) → X of V and a finitely presented G -invariant closed subscheme ˆ V ⊂ X (cid:48) lifting V suchthat ˆ V → S is smooth (resp. ´etale) at each point of V .Proof. Let i : V (cid:44) → X Z be the inclusion and I ⊂ O X Z its ideal. Since p is smooth at V , the conormalsheaf N i is locally free. Replacing X by an affine G -equivariant ´etale neighborhood of V , we can assumeby Theorem 2.21 that there exists a locally free G -module E on X and an isomorphism i ∗ t ∗ ( E ) (cid:39) N i . Let φ : E (cid:16) t ∗ i ∗ ( N i ) be the adjoint morphism. By Lemma 2.17, E is projective in QCoh G ( X ). Thus, we can findsuccessive lifts in the diagram E O X t ∗ i ∗ ( N i ) t ∗ ( I ) t ∗ ( O X Z ). χφ ψ By Nakayama’s lemma, since I is finitely generated, ψ is surjective in a neighborhood of V . Let J ⊂ O X be the image of χ and let ˆ V be the G -invariant closed subscheme of X defined by J . By construction, theimage of J in t ∗ ( O X Z ) equals t ∗ ( I ) in a neighborhood of V , so that t − ( ˆ V ) = V (cid:113) K . Using the projectivityof O X in QCoh G ( X ), we can lift the morphism (1 ,
0) : O X → O V × O K to an endomorphism of O X , which HE SIX OPERATIONS IN EQUIVARIANT MOTIVIC HOMOTOPY THEORY 15 gives a G -invariant function f on X such that the affine open subscheme X f contains V and is disjoint from K . Replacing X by X f , we can therefore assume that t − ( ˆ V ) = V .It remains to show that ˆ V → S is smooth (resp. ´etale) at each point v ∈ V . Let z ∈ Z be the image of v . Since ˆ V z (cid:39) V z is smooth (resp. ´etale) over z by assumption, it suffices to show that ˆ V → S is flat at v [Gro67, 17.5.1 (b)] (resp. [Gro67, 17.6.1 (c (cid:48) )]). Let c be the codimension of V in X Z at v , or equivalentlythe rank of E at v . Choose a basis f , . . . , f c of the κ ( v )-vector space I ( v ). For each 1 ≤ r ≤ c , choose alocal section ω r of E such that ψ ( ω r )( v ) = f r , and let g r = χ ( ω r ). Since the dimension of J ( v ) is at mostthe rank of E , the epimorphism J ( v ) (cid:16) I ( v ) is an isomorphism. Hence, the functions g r generate J ( v ) as a κ ( v )-vector space. By Nakayama’s lemma, there exists an open neighborhood of v in X where g , . . . , g c generate J as an O X -module. On the other hand, since I is a regular ideal [Gro67, 17.12.1], the images of g , . . . , g c in the local ring O X z ,v form a regular sequence. It follows from [Gro66, 11.3.8 (c)] that ˆ V → S isflat at v . (cid:3) Corollary 2.23.
Suppose that B is affine and that G is linearly reductive. Let S be an affine G -scheme withthe G -resolution property, s : Z (cid:44) → S a closed G -immersion, and X a smooth (resp. ´etale) affine G -schemeover Z . Then there exists a finitely presented affine G -scheme ˆ X over S lifting X such that ˆ X → S is smooth(resp. ´etale) at each point of X .Proof. By Lemma 2.12, there exists a locally free G -module E over S and a closed G -immersion X (cid:44) → V ( E )over S . In particular, we obtain the following diagram: X V ( s ∗ ( E )) V ( E ) Z S . s Now apply Theorem 2.22. (cid:3)
Corollary 2.24.
Suppose that B is affine and that G is linearly reductive. Let Z XS s pq be a commutative triangle of G -schemes where s is a closed G -immersion and q is smooth and separated.Suppose that X × S Z is affine and has the G -resolution property. Then there exists an affine G -equivariant´etale neighborhood X (cid:48) → X of Z such that Z (cid:44) → X (cid:48) admits a G -retraction.Proof. Consider the following diagram:
Z Z × S Z X × S ZZ X . δ π s × id π s The vertical projections are smooth since q is smooth, and δ is a closed immersion since q is separated.Applying Theorem 2.22 to this diagram, we obtain a G -scheme X (cid:48) with the desired properties. (cid:3) Linearizations.Proposition 2.25.
Suppose that B is affine and that G is linearly reductive. Let S be an affine G -schemeand let p : X → S be a finitely presented affine G -morphism with a quasi-regular G -section s : S (cid:44) → X . Thenthere exists a G -morphism h : X → V ( N s ) over S such that: (1) the triangle S X V ( N s ) sz h commutes, where z is the zero section; (2) h is ´etale at each point of s ( S ) .Proof. Let I ⊂ O X be the vanishing ideal of s . Since s is quasi-regular, N s is locally free. By Lemma 2.17, N s is projective in QCoh G ( S ). Thus, the epimorphism p ∗ ( I ) (cid:16) p ∗ ( I / I ) (cid:39) N s admits a G -equivariant section φ : N s (cid:44) → p ∗ ( I ). Let σ : p ∗ ( O X ) → O S be the morphism of G -algebrascorresponding to the section s . Then ker( σ ) = p ∗ ( I ), so that the triangle of G -modules O S p ∗ ( O X ) N sσ φ commutes. Let ψ : Sym( N s ) → p ∗ ( O X ) be the morphism of G -algebras induced by φ , and let h : X → V ( N s )be the corresponding G -morphism over S . It is then clear that the triangle of G -algebras O S p ∗ ( O X )Sym( N s ) σ ψ commutes, which proves (1).Let us check (2). By [Gro67, 17.12.1 (c (cid:48) )], p is smooth at each point of s ( S ). By [Gro67, 17.11.2 (c (cid:48) )], ittherefore suffices to show that the map s ∗ ( dh ) : z ∗ (Ω V ( N s ) /S ) → s ∗ (Ω X/S )is an isomorphism. By definition of h , this map is the composition N s φ → p ∗ ( I ) → p ∗ ( I / I ) (cid:39) s ∗ (Ω X/S ) , and it is an isomorphism by choice of φ . (cid:3) Tame group schemes.Definition 2.26.
A flat finitely presented group scheme G over B is called tame if the following conditionshold: • B admits a Nisnevich covering by schemes having the G -resolution property (Definition 2.7); • G is linearly reductive (Definition 2.14). Example 2.27.
Let G be a group scheme over B . Then G is tame in the following cases (see Example 2.8,Remark 2.9, and Example 2.16):(1) B is arbitrary and G is finite locally free of order invertible on B .(2) B is arbitrary and G is of multiplicative type (e.g., G = G m or G = µ n ).(3) B has characteristic zero and G is reductive (e.g., G = GL n ).Note that our definition of tameness does not imply linearity or even affineness (for example, an ellipticcurve is tame). However, we will soon restrict our attention to G -quasi-projective schemes, and these areonly interesting if G can act nontrivially on vector bundles. HE SIX OPERATIONS IN EQUIVARIANT MOTIVIC HOMOTOPY THEORY 17 Unstable equivariant motivic homotopy theory
As a first step towards the formalism of six operations, we define in this section the unstable equivariantmotivic homotopy ∞ -category H G ( S ) associated with a G -scheme S .3.1. Preliminaries.
For the remainder of this paper, we fix a qcqs base scheme B and a tame group scheme G over B . From now on we will only consider G -schemes S that are finitely presented over B and areNisnevich-locally G -quasi-projective, i.e., for which there exists a Nisnevich cover { U i → B } such that S U i is G -quasi-projective over U i . We denote by Sch GB the category of such G -schemes; this is the category onwhich the six operations will eventually be defined, although we will make one additional minor simplifyingassumption on G at the beginning of § S ∈ Sch GB and T → S is a finitely presented G -quasi-projective G -morphism, then T belongsto Sch GB , by the definition of tameness, Lemma 2.6 (1), and Lemma 2.13. Moreover, any morphism in Sch GB is G -quasi-projective Nisnevich-locally on B , by Lemma 2.6 (2).If S ∈ Sch GB , we denote by Sch GS the slice category over S , by Sm GS ⊂ Sch GS the full subcategory spannedby the smooth S -schemes, and by Et GS ⊂ Sm GS the full subcategory spanned by the ´etale S -schemes. Itis clear that Sch GS admits finite limits and finite sums, and that they are computed in the usual way. Inparticular, Sm GS admits finite products and finite sums. Definition 3.1. A G -scheme S is small if there exists a G -quasi-projective G -morphism S → U where U isan affine scheme with trivial G -action and with the G -resolution property.After having defined the G -equivariant Nisnevich topology, we will see that any S ∈ Sch GB admits aNisnevich covering by small G -schemes (Lemma 3.11). A small G -scheme S has several convenient properties: • S is separated and has the G -resolution property (Lemma 2.11). • Any (quasi)-affine G -scheme of finite type over S is G -(quasi)-affine (Lemma 2.12). • If X → S is G -quasi-projective, then X is small (Lemma 2.13). • There exists a G -affine bundle ˜ S → S where ˜ S is affine (Proposition 2.20). Remark . Suppose that G is a finite discrete group whose order is invertible on B . In that case, manysimplifications are possible throughout the paper. In fact, it is possible to remove all quasi-projectivityassumptions and ultimately obtain the formalism of six operations for arbitrary qcqs B -schemes with G -action, so that we might as well take B = Spec Z [1 / | G | ]. The main simplification comes from the fact thatevery qcqs G -scheme is locally affine in the G -equivariant Nisnevich topology (see Remark 3.10). The readerwho cares for this generality will have no difficulty in adapting the proofs of the main results. For furtherremarks that are relevant to this case, see Remarks 3.10, 3.13, 3.14, 4.15, 6.8, 6.21, and 6.26.3.2. Homotopy invariance.Definition 3.3.
A presheaf F on Sm GS is called homotopy invariant if every G -affine bundle Y → X in Sm GS induces an equivalence F ( X ) (cid:39) F ( Y ). We denote by P htp (Sm GS ) ⊂ P (Sm GS ) the full subcategory spanned bythe homotopy invariant presheaves.Since homotopy invariance is defined by a small set of conditions, the inclusion P htp (Sm GS ) ⊂ P (Sm GS )is an accessible localization. We will denote by L htp the corresponding localization endofunctor of P (Sm GS )whose image is P htp (Sm GS ). We say that a morphism f in P (Sm GS ) is a homotopy equivalence if L htp ( f ) isan equivalence. Note that a colimit of homotopy invariant presheaves is homotopy invariant. In particular, L htp preserves colimits.A morphism f : X → Y in P (Sm GS ) will be called a strict A -homotopy equivalence if there exists amorphism g : Y → X , a sequence of A -homotopies between g ◦ f and id X , and a sequence of A -homotopiesbetween f ◦ g and id Y . For example, any G -affine bundle possessing a G -equivariant section is a strict A -homotopy equivalence. It is clear that any strict A -homotopy equivalence is a homotopy equivalence.Our goal for the remainder of this subsection is to obtain an explicit description of the localization functor L htp . To that end, we first consider a more general situation. Let C be a small ∞ -category and let A be a setof morphisms in C . A presheaf F on C will be called A -invariant if it sends morphisms in A to equivalences.We denote by P A ( C ) ⊂ P ( C ) the full subcategory spanned by the A -invariant presheaves, and by L A thecorresponding localization endofunctor of P ( C ). A morphism f in P ( C ) is called an A -equivalence if L A ( f )is an equivalence. We say that A is stable under pullbacks if, for every morphism X (cid:48) → X in C and every f : Y → X in A ,there exists a cartesian square Y (cid:48) YX (cid:48) X f (cid:48) f in C where f (cid:48) is in A . Note that this condition on A does not imply that the class of A -equivalences is stableunder pullbacks.Let D be a presentable ∞ -category with universal colimits. We recall from [GK15, §
1] that a localizationendofunctor L : D → D is called locally cartesian if it commutes with local base change, i.e., if the canonicalmap L ( A × B X ) → A × B L ( X ) is an equivalence for any span A → B ← X in D with A, B ∈ L ( D ). This isthe natural condition that guarantees that the localization L ( D ) has universal colimits. Proposition 3.4.
Let C be a small ∞ -category and A a set of morphisms in C that is stable under pullbacks.Then: (1) The localization functor L A is given by the formula L A ( F )( X ) (cid:39) colim Y ∈ A op X F ( Y ) , where A X is either the full subcategory of C /X spanned by compositions of A -morphisms or the widesubcategory of the latter whose morphisms are compositions of A -morphisms. (2) L A is a locally cartesian localization functor. (3) L A preserves finite products.Proof. Let i X : A X → C be the forgetful functor. Given F ∈ P ( C ), let ˜ F be the presheaf on C defined by˜ F ( X ) = colim i ∗ X F . To prove (1), we must show that:(i) ˜ F is A -invariant;(ii) the map F → ˜ F is an A -equivalence.If f : X (cid:48) → X is an A -morphism, there is an adjunction f ∗ : A op X (cid:29) A op X (cid:48) : f (cid:93) (with either definition of A X ), and ˜ F ( f ) : ˜ F ( X ) → ˜ F ( X (cid:48) ) is the obvious mapcolim i ∗ X F → colim( i ∗ X F ◦ f (cid:93) ) . It is an equivalence because precomposition with f (cid:93) is left Kan extension along f ∗ [Lur09, Lemma 5.2.6.6].This proves (i).We claim that, for every X ∈ C , the restriction functor i ∗ X : P ( C ) → P ( A X ) preserves A -invariantpresheaves (this is obvious) as well as A -equivalences. Since i ∗ X preserves colimits, it suffices to show that,for every A -morphism Y (cid:48) → Y , i ∗ X ( Y (cid:48) ) → i ∗ X ( Y ) is an A -equivalence in P ( A X ). Since colimits are universalin this ∞ -category, it suffices to show that, for every U ∈ A X and every map U → Y in C , the projec-tion i ∗ X ( Y (cid:48) ) × i ∗ X ( Y ) U → U is an A -equivalence. But this projection can be identified with the morphism Y (cid:48) × Y U → U in A X , which is indeed (like any map in A X ) an A -equivalence.To prove (ii), it therefore suffices to prove that, for every X ∈ C , i ∗ X F → i ∗ X ˜ F is an A -equivalence. Since A X has a final object, P A ( A X ) ⊂ P ( A X ) is the subcategory of constant presheaves, so L A : P ( A X ) → P ( A X )sends i ∗ X F to the constant presheaf with value colim Y ∈ A op X F ( Y ). Since ˜ F is A -invariant and has the samevalue on X , i ∗ X F → i ∗ X ˜ F is an A -equivalence, as was to be shown. This completes the proof of (1), whichimmediately implies (2) by the universality of colimits in S . Since A is stable under pullbacks, the firstversion of the ∞ -category A X has finite products and hence is cosifted, so (3) follows from (1). (cid:3) Corollary 3.5.
The localization functor L htp is locally cartesian and preserves finite products.Proof. Since G -affine bundles are stable under pullbacks, this is a special case of Proposition 3.4. (cid:3) HE SIX OPERATIONS IN EQUIVARIANT MOTIVIC HOMOTOPY THEORY 19
Nisnevich excision.
Let X be a G -scheme. A Nisnevich square over X is a cartesian square(3.6) W VU X i p of G -schemes where i is an open G -immersion, p is ´etale, and p induces an isomorphism V × X Z (cid:39) Z , where Z is the reduced closed complement of U in X . By Lemma 2.1, this implies that there exists a finitelypresented G -invariant closed subscheme Z (cid:48) ⊂ X , complementary to U , such that V × X Z (cid:48) (cid:39) Z (cid:48) .In this subsection, we will denote by C S an arbitrary full subcategory of Sch GS containing ∅ and with theproperty that, if X ∈ C S and Y → X is an ´etale G -morphism, then Y ∈ C S . For example, C S can be any ofthe three categories Sch GS , Sm GS , and Et GS . Definition 3.7.
Let S be a G -scheme. A presheaf F on C S is called Nisnevich excisive if: • F ( ∅ ) is contractible; • for every Nisnevich square Q in C S , F ( Q ) is cartesian.We denote by P Nis ( C S ) ⊂ P ( C S ) the full subcategory of Nisnevich excisive presheaves.Since the property of being Nisnevich excisive is defined by a small set of conditions, P Nis ( C S ) is anaccessible localization of P ( C S ). We denote by L Nis the corresponding localization endofunctor of P ( C S )whose image is P Nis ( C S ). We say that a morphism f in P ( C S ) is a Nisnevich equivalence if L Nis ( f ) is anequivalence. Note that a filtered colimit of Nisnevich excisive presheaves is Nisnevich excisive, so that L Nis preserves filtered colimits.The
Nisnevich topology on C S is the coarsest topology for which: • the empty sieve covers ∅ ; • for every Nisnevich square (3.6) in C S , { U i → X, V p → X } generates a covering sieve.A family of G -morphisms { p i : U i → X } i ∈ I is called a basic Nisnevich cover if I is finite, each p i is ´etale,and there exists a chain of G -invariant finitely presented closed subschemes ∅ = Z ⊂ Z ⊂ · · · ⊂ Z n − ⊂ Z n = X such that, for each 1 ≤ j ≤ n , the morphism (cid:96) i U i → X splits G -equivariantly over Z j (cid:114) Z j − . By the proofof [HKØ15, Proposition 2.15], a basic Nisnevich cover is indeed a cover for the Nisnevich topology, and it isthen clear that basic Nisnevich covers form a basis for the Nisnevich topology on C S .If U = { U i → X } is a family of maps in C S , let ˇ C ( U ) denote the ˇCech nerve of the morphism (cid:97) i U i → X, where the coproduct is taken in P ( C S ). Note that colim ˇ C ( U ) is equivalent to the image of the abovemorphism, which by definition is the sieve generated by U . We say that a presheaf F satisfies U -descent if F ( X ) is the limit of the cosimplicial diagram Map( ˇ C ( U ) , F ), i.e., if F ( X ) (cid:39) Map(
U, F ), where
U (cid:44) → X isthe sieve generated by U . Proposition 3.8.
Let S be a G -scheme and let F be a presheaf on C S . The following are equivalent: (1) F is Nisnevich excisive. (2) F satisfies U -descent for every basic Nisnevich cover U . (3) F is a sheaf for the Nisnevich topology.Proof. The equivalence of (1) and (3) is a special case of [AHW15, Theorem 3.2.5]. The equivalence of (2)and (3) is a special case of [Hoy14, Corollary C.2]. (cid:3)
Corollary 3.9.
The localization functor L Nis is left exact and P Nis ( C S ) is an ∞ -topos.Remark . Let S ∈ Sch GB , and let C denote the enlargement of Sm GS consisting of all finitely presentedsmooth G -schemes over S . Define the Nisnevich topology on C via Nisnevich squares or basic Nisnevichcovers, as above. Then the inclusion i : Sm GS (cid:44) → C is continuous for the Nisnevich topology (in the strongsense that the restriction functor preserves sheaves of ∞ -groupoids), but we do not know if it is cocontinuousin general. If G is finite and discrete, however, we claim that i induces an equivalence between the ∞ -topoi of sheaves. Any X ∈ C admits a G -invariant open cover by G -schemes that map to affine open subschemesof B , which have the G -resolution property. For such a G -scheme, any cover by affine schemes has a ˇCechnerve that belongs to Sch GB , by Lemma 2.12. To prove our claim, it therefore suffices to show that any X ∈ C admits a Nisnevich cover by affine schemes that are smooth over X , and we can clearly assume that X is noetherian. For each x ∈ X , let G x = { g ∈ G | gx = x } and let U x be an affine G x -invariant openneighborhood of x . By [HKØ15, Corollary 2.19], { ( G × U x ) /G x → X } x ∈ X is then a Nisnevich cover of X by affine schemes. Lemma 3.11.
For every S ∈ Sch GB , there exists a basic Nisnevich cover { p i : U i → S } where each U i issmall and each p i is G -quasi-affine.Proof. Since G is tame, there exists a (nonequivariant) basic Nisnevich cover { B i → B } where each B i hasthe G -resolution property. By Lemma 2.11, we may assume that each B i is affine, so that each B i → B isquasi-affine and hence G -quasi-affine. Since S is Nisnevich-locally G -quasi-projective, we may further assumethat S × B B i → B i is G -quasi-projective. Then the projections p i : U i = S × B B i → S have all the desiredproperties. (cid:3) Equivariant motivic spaces.Definition 3.12.
Let S be a G -scheme. A motivic G -space over S is a presheaf on Sm GS that is homotopyinvariant and Nisnevich excisive. We denote by H G ( S ) ⊂ P (Sm GS ) the full subcategory of motivic G -spacesover S .It is clear that the inclusion H G ( S ) ⊂ P (Sm GS ) is an accessible localization. We denote by L mot : P (Sm GS ) → P (Sm GS ) the corresponding localization endofunctor. A morphism f in P (Sm GS ) is called a motivic equivalence if L mot ( f ) is an equivalence. Since the subcategories P Nis (Sm GS ) and P htp (Sm GS ) of P (Sm GS ) are stable underfiltered colimits, we have L mot ( F ) = colim n →∞ ( L htp ◦ L Nis ) n ( F )for every presheaf F . Indeed, it is clear that the right-hand side is a motivic G -space. We will sometimesomit the functor L mot from the notation and tacitly view presheaves on Sm GS as motivic G -spaces, dependingon the context. Remark . Suppose that every X ∈ Sm GS is Nisnevich-locally affine (e.g., G is finite locally free or locallydiagonalizable). Then a Nisnevich excisive presheaf on Sm GS is a motivic G -space if and only if it takesprojections X × A → X to equivalences. Indeed, because G is linearly reductive, every G -affine bundle overa small affine G -scheme has a G -section, by Lemma 2.17, and hence is a strict A -homotopy equivalence. Remark . Combining Remarks 3.10 and 3.13, we deduce that, if B is noetherian of finite Krull dimensionand if G is finite discrete, the ∞ -category H G ( B ) coincides with that defined in [HKØ15]. We do not knowif this is true for more general tame groups G . Proposition 3.15.
The localization functor L mot is locally cartesian and preserves finite products. Inparticular, colimits in H G ( S ) are universal.Proof. This follows from Corollaries 3.5 and 3.9. (cid:3)
Proposition 3.16.
Let S be a G -scheme. (1) The ∞ -category H G ( S ) is generated under sifted colimits by the small affine G -schemes in Sm GS . (2) If a morphism in P (Sm GS ) is an equivalence on small affine G -schemes, it is a motivic equivalence. (3) Every X ∈ Sm GS is compact in H G ( S ) .Proof. (1) It is clear that H G ( S ) is generated under sifted colimits by Sm GS , since Nisnevich excisive presheavestransform finite coproducts in Sm GS (which exist) into finite products. Let X ∈ Sm GS , and let { U i → X } be a G -quasi-projective basic Nisnevich cover of X by small G -schemes (Lemma 3.11). By Lemma 2.13, the fiberproducts U i × X · · · × X U i k are also small and hence are equivalent to small affine schemes by Jouanolou’strick. Hence, the finite sum (cid:96) i ,...,i k U i × X · · · × X U i k is also equivalent to a small affine scheme. Thus, inH G ( S ), X is the colimit of a simplicial diagram whose terms are small affine schemes.(2) If f is an equivalence on small affine G -schemes, then L htp ( f ) is an equivalence on all small G -schemes,by Proposition 3.4 (1), hence L Nis L htp ( f ) is an equivalence, by Lemma 3.11.(3) This follows from the fact that the inclusion H G ( S ) ⊂ P (Sm GS ) preserves filtered colimits. (cid:3) HE SIX OPERATIONS IN EQUIVARIANT MOTIVIC HOMOTOPY THEORY 21
Smooth closed pairs.
Let S be a G -scheme. A smooth closed pair ( X, Z ) over S is a smooth G -scheme X ∈ Sm GS together with a G -invariant closed subscheme Z ⊂ X that is also smooth over S . A morphism ofsmooth closed pairs f : ( Y, W ) → ( X, Z ) is a G -morphism f : Y → X over S such that f − ( Z ) = W . Wesay that f is Nisnevich if it is ´etale and induces an isomorphism W (cid:39) Z . Definition 3.17.
Let S be a G -scheme. A morphism of smooth closed pairs f : ( X (cid:48) , Z (cid:48) ) → ( X, Z ) over S is weakly excisive if the square(3.18) Z (cid:48) X (cid:48) / ( X (cid:48) (cid:114) Z (cid:48) ) Z X/ ( X (cid:114) Z ) f in P (Sm GS ) is motivically cocartesian, i.e., becomes cocartesian in H G ( S ). Remark . Let f : ( X (cid:48) , Z (cid:48) ) → ( X, Z ) be a morphism of smooth closed pairs. If f induces a motivicequivalence Z (cid:48) → Z , then f is weakly excisive if and only if X (cid:48) / ( X (cid:48) (cid:114) Z (cid:48) ) → X/ ( X (cid:114) Z ) is a motivicequivalence. On the other hand, if f is weakly excisive and induces a motivic equivalence X (cid:48) (cid:114) Z (cid:48) → X (cid:114) Z ,then Z/Z (cid:48) → X/X (cid:48) is a motivic equivalence.By definition of the ∞ -category H G ( S ), it is clear that G -affine bundles and Nisnevich morphisms areweakly excisive. The goal of this subsection is to obtain two other families of weakly excisive morphisms: blowups and deformations to the normal bundle . Lemma 3.20.
Let ( X (cid:48)(cid:48) , Z (cid:48)(cid:48) ) g → ( X (cid:48) , Z (cid:48) ) f → ( X, Z ) be morphisms of smooth closed pairs over S . (1) If g is weakly excisive, then f is weakly excisive iff f ◦ g is weakly excisive. (2) If f and f ◦ g are weakly excisive and f induces a motivic equivalence Z (cid:48) → Z , then g is weaklyexcisive. (3) If { U i → X } is a basic Nisnevich cover of X such that f U i × X ···× X U ik is weakly excisive for everynonempty family of indices ( i , . . . , i k ) , then f is weakly excisive.Proof. (1) and (2) are obvious. By Nisnevich descent, the square (3.18) associated with f is a colimit inH G ( S ) of squares associated with f U i × X ···× X U ik , whence (3). (cid:3) Lemma 3.21.
Let ( X (cid:48) , Z (cid:48) , W (cid:48) ) → ( X, Z, W ) be a morphism of smooth closed triples over S . Suppose that ( X (cid:48) , Z (cid:48) ) → ( X, Z ) , ( Z (cid:48) , W (cid:48) ) → ( Z, W ) and ( X (cid:48) (cid:114) W (cid:48) , Z (cid:48) (cid:114) W (cid:48) ) → ( X (cid:114) W, Z (cid:114) W ) are weakly excisive. Then ( X (cid:48) , W (cid:48) ) → ( X, W ) is weakly excisive.Proof. We must show that the boundary of the following diagram is motivically cocartesian: W (cid:48) Z (cid:48) / ( Z (cid:48) (cid:114) W (cid:48) ) X (cid:48) / ( X (cid:48) (cid:114) W (cid:48) ) W Z/ ( Z (cid:114) W ) X/ ( X (cid:114) W ).The first square is motivically cocartesian since ( Z (cid:48) , W (cid:48) ) → ( Z, W ) is weakly excisive. The second square isthe cofiber of the obvious morphism from the square associated with ( X (cid:48) (cid:114) W (cid:48) , Z (cid:48) (cid:114) W (cid:48) ) → ( X (cid:114) W, Z (cid:114) W )to the square associated with ( X (cid:48) , Z (cid:48) ) → ( X, Z ), and hence it is also motivically cocartesian. (cid:3)
Let (
X, Z ) be a smooth closed pair over S . If p : B Z X → X is the blowup of X at Z , then ( B Z ( X ) , p − ( Z ))is a smooth closed pair over S , depending functorially on ( X, Z ). Moreover, p is a morphism of smooth closedpair ( B Z X, p − ( Z )) → ( X, Z ), natural in (
X, Z ). We denote by N Z X the normal bundle of Z in X . Thedeformation space [Ful98, Chapter 5] of ( X, Z ) is a smooth G -scheme D Z X over S × A whose restrictionsto S × S × X and N Z X , respectively. Explicitly, D Z X = B Z × ( X × A ) (cid:114) B Z × ( X × . The closed G -immersion Z × A (cid:44) → X × A lifts uniquely to a closed G -immersion Z × A (cid:44) → D Z X . It isclear that the smooth closed pair ( D Z X, Z × A ) varies functorially with ( X, Z ), and that the inclusions1 (cid:44) → A ← (cid:45) i : ( X, Z ) → ( D Z X, Z × A ) ← ( N Z X, Z ) : i . Lemma 3.22.
Let S be a G -scheme and let P be a class of smooth closed pairs over S . Suppose that thefollowing conditions hold for every smooth closed pair ( X, Z ) : (1) If { U i → X } is a basic Nisnevich cover of X and ( U i × X · · · × X U i k , Z × X U i × X · · · × X U i k ) ∈ P for every nonempty family of indices ( i , . . . , i k ) , then ( X, Z ) ∈ P . (2) If ( Y, W ) → ( X, Z ) is a G -affine bundle and ( Y, W ) ∈ P , then ( X, Z ) ∈ P . (3) If ( Y, W ) → ( X, Z ) is a Nisnevich morphism, then ( X, Z ) ∈ P if and only if ( Y, W ) ∈ P . (4) If E is a locally free G -module on Z , then ( V ( E ) , Z ) ∈ P .Then P contains all smooth closed pairs.Proof. Let (
X, Z ) be an arbitrary smooth closed pair over S . To show that it belongs to P , we can assume,by (1) and Lemma 3.11, that X is small. By Jouanolou’s trick and (2), we can then also assume that X isaffine. By Corollary 2.24, we can find a cartesian square Z X (cid:48)
Z X t p where X (cid:48) is affine, p is ´etale at t ( Z ), and t admits a G -retraction r : X (cid:48) → Z . Since t is a quasi-regular G -immersion, we can apply Proposition 2.25 to r and get a G -morphism h : X (cid:48) → V ( N t ) over Z that is ´etaleat t ( Z ) and such that ht = z , where z : Z (cid:44) → V ( N t ) is the zero section. Let U ⊂ X (cid:48) be the intersection of the´etale locus of p and the ´etale locus of h . By Lemma 2.3, U is a G -invariant open subscheme of X (cid:48) . Shrinking U if necessary, we can assume that it is quasi-compact and that the square Z UZ V ( N t ) tz h is cartesian, by Lemma 2.4. We therefore have two ´etale G -morphisms X ← U → V ( N t )that are isomorphisms over Z . By (4), the pair ( V ( N t ) , Z ) belongs to P . By two applications of (3), wededuce that ( X, Z ) ∈ P . (cid:3) Theorem 3.23.
Let S be a G -scheme. For every smooth closed pair ( X, Z ) over S , the morphisms ( B Z X, p − ( Z )) p −→ ( X, Z ) and ( X, Z ) i −→ ( D Z X, Z × A ) i ←− ( N Z X, Z ) are weakly excisive.Proof. Let P be the class of smooth closed pairs ( X, Z ) over S for which the conclusion of the theoremholds. We will show that P satisfies conditions (1)–(4) of Lemma 3.22. Since blowups commute with flatbase change, it is clear that the functors B , D , and N preserve G -affine bundles and Nisnevich morphisms.If f : ( Y, W ) → ( X, Z ) is such a morphism, it follows from Lemma 3.20 (1,2) that (
X, Z ) ∈ P iff ( Y, W ) ∈ P ,which proves conditions (2) and (3). Condition (1) follows easily from Lemma 3.20 (3).It remains to show that P satisfies condition (4), i.e., that the theorem holds when X = V ( E ) for somelocally free G -module E on Z . In that case, ( B Z X, p − ( Z )) can be identified with the smooth closed pair HE SIX OPERATIONS IN EQUIVARIANT MOTIVIC HOMOTOPY THEORY 23 ( V ( O P ( E ) (1)) , P ( E )). The square (3.18) for p is the outer square in the following diagram: p − ( Z ) B Z X B Z XB Z X (cid:114) p − ( Z ) Z X XX (cid:114) Z . p The first square is motivically cocartesian since the horizontal maps are sections of G -vector bundles andhence are homotopy equivalences. The second square is already cocartesian in P (Sm GS ), since B Z X (cid:114) p − ( Z ) → X (cid:114) Z is a monomorphism. This proves that p is weakly excisive.Similarly, when X = V ( E ), ( D Z X, Z × A ) can be identified with the smooth closed pair ( V ( L ) , V ( L ) Z ),where L is the restriction to X of O P ( E ⊕ O ) (1). On the other hand, ( N Z X, Z ) is canonically isomorphic to(
X, Z ). Under these identifications, i and i are both sections of the G -line bundle ( V ( L ) , V ( L ) Z ) → ( X, Z )and hence are weakly excisive by Lemma 3.20 (2). (cid:3)
By the second part of Theorem 3.23, the morphisms i and i induce an equivalenceΠ = Π X,Z : XX (cid:114) Z (cid:39) N Z XN Z X (cid:114) Z in H G ( S ), natural in the smooth closed pair ( X, Z ). It is called the purity equivalence .Let (
X, Z, W ) be a smooth closed triple over S . Then W × A can be identified with a G -invariant closedsubscheme of D Z X , and we can form the smooth closed pair ( D W × A ( D Z X ) , W × A ) over S × A . Pullingit back along the closed immersions (1 ,
1) 1 × A (1 , A × A A × ,
1) 0 × A (0 , S :(3.24) ( X, W ) ( D Z X, W × A ) ( N Z X, W )( D W X, W × A ) ( D W × A ( D Z X ) , W × A ) ( D W ( N Z X ) , W × A )( N W X, W ) ( N W × A ( D Z X ) , W × A ) ( N W ( N Z X ) , W ).Note that the bottom row of (3.24) is canonically isomorphic to( N W X, W ) i −→ ( D N W Z ( N W X ) , W × A ) i ←− ( N N W Z ( N W X ) , W ) . Corollary 3.25.
For every smooth closed triple ( X, Z, W ) over S , all morphisms in (3.24) are weaklyexcisive.Proof. Each column is weakly excisive by Theorem 3.23. The top row is seen to be weakly excisive byapplying Lemma 3.21 to the morphisms of triples(
X, Z, W ) → ( D Z X, Z × A , W × A ) ← ( N Z X, Z, W ) , using Theorem 3.23 twice. The remaining rows are automatically weakly excisive by Lemma 3.20 (1,2). (cid:3) The top row of (3.24) provides an equivalence XX (cid:114) W (cid:39) N Z XN Z X (cid:114) W in H G ( S ), which we also denote by Π X,Z . It is more precisely the composition XX (cid:114) W = X/ ( X (cid:114) Z )( X (cid:114) W ) / ( X (cid:114) Z ) (cid:39) N Z X/ ( . . . ) N Z (cid:114) W ( X (cid:114) W ) / ( . . . ) = N Z XN Z X (cid:114) W , where the middle equivalence is Π
X,Z in the numerator and Π X (cid:114) W,Z (cid:114) W in the denominator.Let X ∈ Sm GS and let 0 → U → V → W → G -vector bundles on X . The top row of (3.24) for the smooth closed triple( V, U, X ) is a pair of morphisms(3.26) (
V, X ) → ( D U V, X × A ) ← ( U × X W, X )that are weakly excisive by Corollary 3.25. In particular, we obtain an equivalenceΨ = Π
V,U : VV (cid:114) X (cid:39) U × X W ( U × X W ) (cid:114) X in H G ( S ). More generally, if Z ⊂ X is a smooth G -invariant closed subscheme, the morphisms (3.26) arealso weakly excisive after replacing X by Z , so that we have a canonical equivalenceΨ : VV (cid:114) Z (cid:39) U × X W ( U × X W ) (cid:114) Z in H G ( S ), compatible with the previous one. Note that the diagram (3.26) and hence the equivalence Ψ arenatural for universal monomorphisms of short exact sequences of G -vector bundles. Remark . Suppose given a short exact sequence as above and a splitting σ : V (cid:39) U × X W . Then theequivalence Ψ is induced by σ . Indeed, identifying V with U × X W by means of σ , the morphisms (3.26)can be identified with sections of the same G -line bundle over V , so that Ψ σ − is the identity in H G ( S ).With this notation in place, the main point of Corollary 3.25 is that it implies the commutativity of thefollowing square of equivalences in H G ( S ), for every smooth closed triple ( X, Z, W ): XX (cid:114) W N Z XN Z X (cid:114) WN W XN W X (cid:114) X N W N Z XN W N Z X (cid:114) W . Π X,Z Π X,W Π NZX,W Ψ Here, Ψ is induced by the canonical short exact sequence0 → N W Z → N W X → N Z X × Z W → G -vector bundles on W . Remark . More generally, for a smooth closed tuple ( X , . . . , X n ), one can show using an n -fold defor-mation to the normal bundle that all the equivalences X X (cid:114) X n (cid:39) N X n N X n − . . . N X X N X n N X n − . . . N X X (cid:114) X n that one can define using Theorem 3.23 fit in a commutative n -cube. Note that Corollary 3.25 implies thecommutativity of this n -cube in the homotopy 1-category, which is sufficient for many purposes.4. Functoriality
We discuss the functoriality in S of the ∞ -category H G ( S ). HE SIX OPERATIONS IN EQUIVARIANT MOTIVIC HOMOTOPY THEORY 25
Easy results.
It is clear that the empty scheme, Nisnevich squares, and G -affine bundles are preservedby base change. As a result, if f : T → S is a G -morphism, then the functor f ∗ : P (Sm GT ) → P (Sm GS ) , f ∗ ( F )( X ) = F ( X × S T ) , preserves Nisnevich excisive presheaves, homotopy invariant presheaves, and motivic G -spaces. We stilldenote by f ∗ : H G ( T ) → H G ( S ) the restriction of f ∗ . Since it preserves limits, it admits a left adjoint. Wewill overload the notation f ∗ to denote a left adjoint to f ∗ in any context, and when f ∗ itself admits a leftadjoint we will denote it by f (cid:93) . Note that f ∗ : H G ( T ) → H G ( S ) preserves filtered colimits since the inclusionH G ( S ) ⊂ P (Sm GS ) does.If f : T → S is a smooth G -morphism, then the forgetful functor Sm GT → Sm GS preserves the empty scheme,Nisnevich squares, and G -affine bundles. It follows that f ∗ : P (Sm GS ) → P (Sm GT ), which in this case is precom-position with the forgetful functor, preserves Nisnevich excisive presheaves, homotopy invariant presheaves,and motivic G -spaces. In particular, it restricts to a functor H G ( S ) → H G ( T ) that is automatically leftadjoint to f ∗ . We deduce that, for f smooth, f ∗ : H G ( S ) → H G ( T ) has a left adjoint f (cid:93) .In summary, for a G -morphism f : T → S , we have commutative squares P (Sm GT ) P (Sm GS )H G ( T ) H G ( S ), f ∗ f ∗ P (Sm GT ) P (Sm GS )H G ( T ) H G ( S ). f ∗ f ∗ If f is smooth, we moreover have commutative squares P (Sm GS ) P (Sm GT )H G ( S ) H G ( T ), f ∗ f ∗ P (Sm GS ) P (Sm GT )H G ( S ) H G ( T ). f (cid:93) f (cid:93) Proposition 4.1 (Monoidality) . Let f : Y → X be a G -morphism. Then f ∗ : H G ( X ) → H G ( Y ) preservesfinite products.Proof. Since L mot preserves finite products (Proposition 3.15), this follows from the analogous fact at thelevel of presheaves, which is obvious. (cid:3) Proposition 4.2 (Smooth base change) . Let Y (cid:48) YX (cid:48) X gq f p be a cartesian square of G -schemes where p is smooth. Then the exchange transformationsEx ∗ (cid:93) : q (cid:93) g ∗ → f ∗ p (cid:93) : H G ( Y ) → H G ( X (cid:48) ) , Ex ∗∗ : p ∗ f ∗ → g ∗ q ∗ : H G ( X (cid:48) ) → H G ( Y ) are equivalences.Proof. The second exchange transformation is the mate of the first one, so it suffices to show that q (cid:93) g ∗ → f ∗ p (cid:93) is an equivalence. This transformation is L mot of the analogous exchange transformation at the level ofpresheaves, which is clearly an equivalence. (cid:3) Proposition 4.3 (Smooth projection formula) . Let f : Y → X be a smooth G -morphism and let B → A bea morphism in H G ( X ) . For every C ∈ H G ( Y ) /f ∗ A and every D ∈ H G ( X ) /B , the canonical maps f (cid:93) ( f ∗ B × f ∗ A C ) → B × A f (cid:93) C,f ∗ Hom A ( B, D ) → Hom f ∗ A ( f ∗ B, f ∗ D ) are equivalences in H G ( X ) and H G ( Y ) , respectively.Proof. By adjunction, it suffices to show that the first map is an equivalence. Since L mot is locally cartesian(Proposition 3.15), this follows from the projection formula at the level of presheaves. (cid:3) Proposition 4.4 (Homotopy invariance) . Let f : T → S be a G -affine bundle. (1) The functor f ∗ : P (Sm GS ) → P (Sm GT ) detects homotopy and motivic equivalences. (2) The functor f ∗ : H G ( S ) → H G ( T ) is fully faithful.Proof. Since f is smooth, f ∗ has a left adjoint f (cid:93) that preserves homotopy and motivic equivalences. If X ∈ Sm GS , the counit (cid:15) X : f (cid:93) f ∗ X → X is the G -affine bundle X × S T → X . Since f (cid:93) and f ∗ preservecolimits, (cid:15) X is a homotopy equivalence for every X ∈ P (Sm GS ). This easily implies the results. (cid:3) Proposition 4.5 (Nisnevich separation) . Let { f i : U i → S } be a Nisnevich cover of a G -scheme S . (1) The family of functors { f ∗ i : P (Sm GS ) → P (Sm GU i ) } detects Nisnevich and motivic equivalences. (2) The family of functors { f ∗ i : H G ( S ) → H G ( U i ) } is conservative.Proof. Clearly, (1) implies (2). To prove (1), we may assume that each f i is smooth. Let h be a morphismin P (Sm GS ) such that f ∗ i ( h ) is a Nisnevich (resp. motivic) equivalence for all i . Denote by f i ...i n : U i × S · · · × S U i n → S an n -fold fiber product of the given covering. Then f ∗ i ...i n ( h ) is a Nisnevich (resp. motivic) equivalence.Consider the augmented simplicial object C • → id whose n th term C n is the endofunctor F (cid:55)→ (cid:97) i ,...,i n ( f i ...i n ) (cid:93) f ∗ i ...i n F. For every X ∈ Sm GS , colim C • ( X ) → X is a Nisnevich covering sieve. Thus, for every F ∈ P (Sm GS ),colim C • ( F ) → F is a Nisnevich equivalence. Since C • ( h ) is a degreewise Nisnevich (resp. motivic) equiva-lence, it follows by 2-out-of-3 that h is a Nisnevich (resp. motivic) equivalence, as desired. (cid:3) Proposition 4.6 (Affine resolutions) . Let S be a G -scheme. Then there exists a family of smooth G -quasi-affine morphisms { f i : U i → S } , where each U i is small and affine, such that: (1) the family of functors { f ∗ i : P (Sm GS ) → P (Sm GU i ) } detects motivic equivalences; (2) the family of functors { f ∗ i : H G ( S ) → H G ( U i ) } is conservative.Proof. Let { V i → S } be a G -quasi-affine basic Nisnevich cover of S by small G -schemes (Lemma 3.11). ByJouanolou’s trick (Proposition 2.20), there exist G -affine bundles U i → V i where U i is affine. These U i formthe desired family by Propositions 4.4 and 4.5. (cid:3) Proposition 4.7.
Let j : U (cid:44) → X be an open G -immersion. Then the functors j ∗ : H G ( U ) → H G ( X ) ,j (cid:93) : H G ( U ) → H G ( X ) are fully faithful.Proof. It is clear that the unit id → j ∗ j (cid:93) is an equivalence. (cid:3) To make sense of the next proposition, we remark that one can successively construct functors (Sch GB ) op → C at ∞ that send a G -scheme S to Sm GS , P (Sm GS ), P Nis (Sm GS ), and H G ( S ). We refer to [Rob14, § Proposition 4.8 (Nisnevich descent) . (1) The assignment S (cid:55)→ P Nis (Sm GS ) , f (cid:55)→ f ∗ , is a Nisnevich sheaf of ∞ -categories on Sch GB . (2) The assignment S (cid:55)→ H G ( S ) , f (cid:55)→ f ∗ , is a Nisnevich sheaf of ∞ -categories on Sch GB . HE SIX OPERATIONS IN EQUIVARIANT MOTIVIC HOMOTOPY THEORY 27
Proof. (1) It suffices to show that V (cid:55)→ P Nis (Sm GV ) is a Nisnevich sheaf on Et GS for all S ∈ Sch GB . Note thatif V ∈ Et GS , then Sm GV (cid:39) (Sm GS ) /V , since if gf is smooth and g is ´etale then f is smooth. Thus, P Nis (Sm GV ) (cid:39) P Nis (Sm GS ) /V . The fact that V (cid:55)→ P Nis (Sm GS ) /V is a Nisnevich sheaf on Et GS (and even on Sm GS ) follows from Corollary 3.9and general descent theory for ∞ -topoi [Lur09, Theorem 6.1.3.9 (3)].(2) Again, it suffices to show that V (cid:55)→ H G ( V ) is a Nisnevich sheaf on Et GS for all S ∈ Sch GB . Theinclusions H G ( V ) ⊂ P Nis (Sm GV ) are the components of a natural transformation on Et GS , since all morphismsin Et GS are smooth. Given (1) and the fact that limits of fully faithful functors are fully faithful, the sheafcondition is reduced to the following statement: for every basic Nisnevich cover { p i : U i → V } in Et GS andevery F ∈ P Nis (Sm GV ), if p ∗ i ( F ) is homotopy invariant for all i , then F is homotopy invariant. This is clear,since G -affine bundles are stable under base change. (cid:3) Exactness properties of pushforwards.Lemma 4.9.
Let f : T → S be an affine G -morphism where S has the G -resolution property, and let Y → T be a G -affine bundle. Then there exists a G -affine bundle V → S and a G -morphism V T → Y over T .Proof. Let E (cid:16) O T be the epimorphism of locally free G -modules corresponding to Y (see § M be defined by the cartesian square M O S f ∗ ( E ) f ∗ ( O T ) η in QCoh G ( S ). Since f is affine, the horizontal arrows are epimorphisms. Since M is the colimit of its finitelygenerated quasi-coherent G -submodules (Lemma 2.10), there exists a finitely generated quasi-coherent G -submodule N ⊂ M such that N → M → O S is surjective. By the G -resolution property, there exists a locallyfree G -module F and an epimorphism F (cid:16) N . Then F (cid:16) O S defines a G -affine bundle V over S and themorphism f ∗ ( F ) → E over O T defines a G -morphism V T → Y over T , as desired. (cid:3) Lemma 4.10.
Let f : T → S be an affine G -morphism where S has the G -resolution property. Then thefunctor f ∗ : P (Sm GT ) → P (Sm GS ) preserves homotopy equivalences.Proof. Since f ∗ preserves colimits, it suffices to show that it sends G -affine bundles to homotopy equivalences.Let Y → X be a G -affine bundle in Sm GT , and let us prove that f ∗ Y → f ∗ X is a homotopy equivalence. Byuniversality of colimits in P (Sm GS ), it suffices to prove that, for every smooth G -morphism p : U → S andevery U T → X , the projection f ∗ Y × f ∗ X U → U is a homotopy equivalence. Consider the cartesian square U T UT S . gq pf Since p is smooth, we have p ∗ f ∗ (cid:39) g ∗ q ∗ and the projection f ∗ Y × f ∗ X U → U can be identified with p (cid:93) ofthe projection g ∗ Y U × g ∗ X U U → U in P (Sm GU ). Since p (cid:93) preserves homotopy equivalences, we may as wellassume that U = S , so that U T = T . Since f ∗ preserves limits, we then have f ∗ Y × f ∗ X S (cid:39) f ∗ ( Y × X T ) . We are thus reduced to proving the following statement: if Y → T is a G -affine bundle, then f ∗ Y ishomotopically contractible. By Lemma 4.9, there exists a G -affine bundle r : V → S and a G -morphism V T → Y over T . By Proposition 4.4 (1), it remains to prove that r ∗ f ∗ Y is homotopically contractible in P (Sm GV ). Since r is smooth, this presheaf can be identified with the pushforward along V T → V of the G -affine bundle V T × T Y → V T . But by choice of V , this G -affine bundle has a G -section and hence is a strict A -homotopy equivalence, and it is clear that pushforwards preserve strict A -homotopy equivalences. (cid:3) Let C be a small ∞ -category with an initial object ∅ (resp. with finite coproducts). We denote by P ∅ ( C )(resp. P Σ ( C )) the full subcategory of P ( C ) consisting of those presheaves F such that F ( ∅ ) (cid:39) ∗ (resp. thattransform finite coproducts into finite products). Recall that the ∞ -category P Σ ( C ) is freely generated by C under sifted colimits [Lur09, Proposition 5.5.8.15]. We say that an ∞ -category A is weakly contractible if themap A → ∗ is cofinal, or, equivalently, if the ∞ -groupoid completion of A is contractible; colimits indexed bysuch ∞ -categories will be called weakly contractible colimits . Clearly, a presheaf F ∈ P ( C ) belongs to P ∅ ( C )if and only if its ∞ -category of elements is weakly contractible. It follows that P ∅ ( C ) is freely generated by C under weakly contractible colimits. We denote by a ∅ : P ( C ) → P ∅ ( C ) the left adjoint to the inclusion. Lemma 4.11.
Let i : Z (cid:44) → S be a closed G -immersion between G -schemes that are affine over B . Then i ∗ : P ∅ (Sm GZ ) → P ∅ (Sm GS ) preserves Nisnevich equivalences.Proof. We start with a preliminary observation. Let C be an ∞ -category with an initial object. If K is anysimplicial set and p : K → C is a diagram, then a colimit of p is the same thing as a colimit of an extendeddiagram K (cid:47) → C that takes the initial vertex of K (cid:47) to an initial object of C , and the simplicial set K (cid:47) is ofcourse weakly contractible.By Nisnevich separation and smooth base change, we may assume that B is affine and has the G -resolutionproperty. Let C ⊂ P (Sm GZ ) ∆ be the full subcategory consisting of: • equivalences; • the map 0 → ∅ , where 0 is the empty presheaf and ∅ is the empty scheme; • the map K Q → X for every Nisnevich square Q over X in Sm GZ , where K Q denotes the pushout of Q in P (Sm GZ ).Note that C contains the initial object of P (Sm GZ ) ∆ , since it is an equivalence. By the preceding observation,the class of Nisnevich equivalences in P ∅ (Sm GZ ) is the closure of a ∅ ( C ) under 2-out-of-3, pushouts, and weaklycontractible colimits. Since i ∗ : P ∅ (Sm GZ ) → P ∅ (Sm GS ) preserves weakly contractible colimits, it suffices toshow that i ∗ a ∅ sends elements of C to Nisnevich equivalences. This is obvious for equivalences and for0 → ∅ , since a ∅ (0) = ∅ . By universality of colimits in P ∅ (Sm GS ), it remains to show that, for every smooth G -morphism p : U → S and every map U Z → X , the projection i ∗ K Q × i ∗ X U → U is a Nisnevich equivalence. Consider the cartesian square U Z UZ S . i (cid:48) p (cid:48) i p Since p is smooth, we have i (cid:48)∗ p (cid:48)∗ (cid:39) p ∗ i ∗ and the above projection can be identified with p (cid:93) of the projection i (cid:48)∗ ( K p (cid:48)∗ Q ) × i (cid:48)∗ p (cid:48)∗ X U → U. Since p (cid:93) preserves Nisnevich equivalences, we may as well assume that U = S , so that U Z = Z . It is thenclear that i ∗ K Q × i ∗ X S (cid:39) i ∗ K Q × X Z . Thus, we are reduced to proving the following statement:( ∗ ) For every Nisnevich square Q over Z , i ∗ K Q is Nisnevich contractible.Let Q be the square U × Z V VU Z .Let us first prove ( ∗ ) with “Nisnevich” replaced by “Zariski”, so that V → Z is an open G -immersion. Then S (cid:114) ( Z (cid:114) U ) and S (cid:114) ( Z (cid:114) V ) form a Zariski cover of S . By the Zariski version of Proposition 4.5, it sufficesto prove that the restrictions of i ∗ K Q to these two open subsets are Zariski contractible. By smooth basechange, those restrictions are pushforwards of the restrictions of K Q to U and V , which are clearly final HE SIX OPERATIONS IN EQUIVARIANT MOTIVIC HOMOTOPY THEORY 29 objects in P (Sm GU ) and P (Sm GV ), respectively. Since i ∗ preserves final objects, this shows that i ∗ K Q is Zariskicontractible. Altogether, we have proved that i ∗ : P ∅ (Sm GZ ) → P ∅ (Sm GS ) preserves Zariski equivalences.We now prove ( ∗ ) in general. By Nisnevich separation and smooth base change, we may assume that V is G -quasi-projective, since this is true Nisnevich-locally on B . Let W (cid:44) → Z be a G -invariant closedcomplement of U in Z such that V × Z W (cid:39) W . Since B is affine and has the G -resolution property andsince V is G -quasi-projective, we can use Jouanolou’s trick (Proposition 2.20) to obtain a G -affine bundle˜ V → V where ˜ V is affine. Let ˜ W = ˜ V × V W . We now consider the following diagram, the dashed parts ofwhich will be described below: W ˆ W ˜ W ˜ V ˆ VW Z S . q p First of all, since ˜ V → Z is smooth, Corollary 2.23 implies that there exists an affine G -scheme ˆ V over S fitting into a cartesian square as above and such that p is smooth at ˜ V . Since B and W are affine and G is linearly reductive, the G -affine bundle ˜ W → W admits a G -section (Lemma 2.17), which is a closed G -immersion W (cid:44) → ˜ W . We now apply Theorem 2.22: replacing if necessary ˆ V , ˜ V , and ˜ W by affine G -equivariant ´etale neighborhoods of W , we obtain an affine G -scheme ˆ W such that all squares in the abovediagram are cartesian, and such that q : ˆ W → S is ´etale at W . By Lemma 2.3, there exists a quasi-compact G -invariant open neighborhood ˆ W ◦ of W on which q is ´etale. The open G -immersion j : S (cid:114) W (cid:44) → S andthe ´etale G -morphism ˆ W ◦ → S then form a Nisnevich cover of S . By Proposition 4.5, it will thereforesuffice to prove that j ∗ i ∗ K Q and p ∗ i ∗ K Q are Nisnevich contractible. By smooth base change, we have j ∗ i ∗ K Q (cid:39) i (cid:48)∗ K Q U and p ∗ i ∗ K Q (cid:39) i (cid:48)(cid:48)∗ K Q ˜ V , where i (cid:48) : U (cid:44) → S (cid:114) W and i (cid:48)(cid:48) : ˜ V (cid:44) → ˆ V . Now, K Q U is clearly afinal object in P (Sm GU ). It remains to prove that i (cid:48)(cid:48)∗ K Q ˜ V ∈ P (Sm G ˆ V ) is Nisnevich contractible. The Nisnevichsquare Q ˜ V is isomorphic to the square ˜ V U (cid:113) ˜ Y ˜ V (cid:113) ˜ Y ˜ V U ˜ V ,where Y is the complement of the diagonal in V × Z V and ˜ Y = Y × V ˜ V . It is then clear that K Q ˜ V → ˜ V isa Zariski equivalence. As we have already proved that i (cid:48)(cid:48)∗ preserves Zariski equivalences, we are done. (cid:3) Theorem 4.12.
Let i : Z (cid:44) → S be a closed G -immersion. Then i ∗ : P ∅ (Sm GZ ) → P ∅ (Sm GS ) preserves motivicequivalences.Proof. By Proposition 4.6 and smooth base change, we may assume that B and S are affine and that S hasthe G -resolution property. The observation at the beginning of the proof of Lemma 4.11 shows that theclass of motivic equivalences in P ∅ (Sm GZ ) is the closure under 2-out-of-3, pushouts, and weakly contractiblecolimits of the union of the classes of homotopy equivalences and of Nisnevich equivalences. The result nowfollows from Lemmas 4.10 and 4.11. (cid:3) Corollary 4.13.
Let i : Z (cid:44) → S be a closed G -immersion. Then i ∗ : H G ( Z ) → H G ( S ) preserves weaklycontractible colimits.Remark . Unless S (cid:114) Z is empty, i ∗ : H G ( Z ) → H G ( S ) does not preserve the initial object. Thisinconvenience disappears when we pass to the ∞ -category of pointed motivic G -spaces. We will see someremarkable consequences of this fact in § Remark . Suppose that G is finite locally free. Then, if f : Y → X is a finite G -morphism, the functor f ∗ : P Σ (Sm GY ) → P Σ (Sm GX ) preserves Nisnevich equivalences. Modulo a noetherian approximation argument, this can be proved exactly as in [MV99, §
3, Proposition 1.27], using the fact that, when B is noetherian offinite Krull dimension, the family of functors P (Sm GS ) → S , F (cid:55)→ F ( X × X/G ( X/G ) hx ) ( X ∈ Sm GS and x ∈ X/G ) , which is well-defined by Lemma 2.12, detects Nisnevich equivalences [Del09, Proposition 13]. It follows that f ∗ : H G ( Y ) → H G ( X ) preserves sifted colimits. We do not know if these facts hold for more general G .4.3. Gluing.
Let
Z (cid:44) → S be a closed G -immersion with open complement U ⊂ S . If X ∈ Sm GS and t : Z (cid:44) → X Z is a G -equivariant section of the projection X Z → Z , we define a presheaf of sets Φ S ( X, t ) onSm GS by: Φ S ( X, t )( Y ) = (cid:40) Hom S ( Y, X ) × Hom Z ( Y Z ,X Z ) ∗ if Y Z (cid:54) = ∅ , ∗ if Y Z = ∅ ,where the map ∗ → Hom Z ( Y Z , X Z ) hits the composition Y Z → Z t (cid:44) → X Z . More succinctly,Φ S ( X, t ) = ( X (cid:97) X U U ) × i ∗ X Z S. Note that Φ S ( X, t ) is functorial in the pair (
X, t ). Moreover, if f : T → S is a G -morphism, there is a naturalmap f ∗ Φ S ( X, t ) → Φ T ( X T , t T ) , which is an isomorphism when f is smooth. Lemma 4.16.
Let p : X (cid:48) → X be a G -morphism in Sm GS and let t : Z (cid:44) → X Z and t (cid:48) : Z (cid:44) → X (cid:48) Z be closed G -sections such that pt (cid:48) = t . Suppose that p is ´etale at each point of t (cid:48) ( Z ) . Then Φ S ( p ) : Φ S ( X (cid:48) , t (cid:48) ) → Φ S ( X, t ) is a Nisnevich equivalence.Proof. By Lemmas 2.3 and 2.4 and 2-out-of-3, we can assume that p is ´etale and that the square Z X (cid:48) Z Z X Zt (cid:48) t p is cartesian. We must show that L Nis Φ S ( p ) is an equivalence. Since it is a map between 0-truncated objectsin an ∞ -topos, it suffices to show that it is 1-connective, i.e., that both it and its diagonal are effectiveepimorphisms.Let f : Y → X be an element of Φ S ( X, t )( Y ) and let Y (cid:48) = Y × X X (cid:48) . The pair of G -morphisms π : Y (cid:48) → Y and Y × X ( X (cid:114) t ( Z )) (cid:44) → Y form a Nisnevich cover of Y , and we claim that f lifts to Φ S ( X (cid:48) , t (cid:48) ) on that cover.The G -morphism π : Y (cid:48) → X (cid:48) defines an element of Φ S ( X (cid:48) , t (cid:48) )( Y (cid:48) ) lifting f π . By definition of Φ S ( X, t ),the open G -immersion Y U (cid:44) → Y × X ( X (cid:114) t ( Z )) is an isomorphism. Thus, f lifts trivially on Y × X ( X (cid:114) t ( Z )).This shows that L Nis Φ S ( p ) is an effective epimorphism.Let f, g : Y → X (cid:48) be elements in Φ S ( X (cid:48) , t (cid:48) )( Y ) that become equal in Φ S ( X, t )( Y ). Then f and g inducetwo G -sections of the ´etale G -morphism X (cid:48) × X Y → Y . Their equalizer V is a quasi-compact G -invariantopen subscheme of Y . By definition of Φ S ( X (cid:48) , t (cid:48) ), V contains Y Z . Thus, V and Y U form a Zariski cover of Y on which f and g agree. This shows that the diagonal of L Nis Φ S ( p ) is an effective epimorphism. (cid:3) Lemma 4.17.
Let E be a locally free G -module on S and let t : S → V ( E ) be the zero section. Then Φ S ( V ( E ) , t Z ) → S is a homotopy equivalence.Proof. The map A × Φ S ( V ( E ) , t Z ) → Φ S ( V ( E ) , t Z ) , ( a, f ) (cid:55)→ af, is an A -homotopy between the identity and the compositionΦ S ( V ( E ) , t Z ) → S t → Φ S ( V ( E ) , t Z ) . (cid:3) HE SIX OPERATIONS IN EQUIVARIANT MOTIVIC HOMOTOPY THEORY 31
Theorem 4.18 (Gluing) . Let i : Z (cid:44) → S be a closed G -immersion with open complement j : U (cid:44) → S . Then,for every F ∈ H G ( S ) , the square j (cid:93) j ∗ F FU i ∗ i ∗ F (cid:15) ! ! η is cocartesian, where ! denotes a unique map.Proof. By Nisnevich separation and smooth base change, we can assume that S is separated. All functorsof F in this square preserve weakly contractible colimits: this is obvious for the top left, top right, andbottom left corner, and it follows from Corollary 4.13 for the bottom right corner. By Proposition 3.16 (1)and our assumption that S is separated, it suffices to prove the theorem when F is the motivic localizationof a scheme X ∈ Sm GS that is affine over S . By Theorem 4.12, i ∗ i ∗ F is then the motivic localization of i ∗ X Z ∈ P (Sm GS ). It therefore suffices to show that the canonical map X (cid:97) X U U → i ∗ X Z in P (Sm GS ) is a motivic equivalence. By universality of colimits, it suffices to show that for every Y ∈ Sm GS and every map t : Y Z → X Z in Sm GZ , the projection( X (cid:97) X U U ) × i ∗ X Z Y → Y is a motivic equivalence. Using smooth base change and the smooth projection formula (as in the proofs ofLemmas 4.10 and 4.11), we may assume without loss of generality that Y = S , so that Y Z = Z . We are thusreduced to proving the following statement:( ∗ ) For every X ∈ Sm GS that is affine over S and every G -section t : Z (cid:44) → X Z , Φ S ( X, t ) is motivicallycontractible.By Proposition 4.6, it suffices to prove ( ∗ ) when S is small and affine. Moreover, by Lemma 4.16, we arefree to replace X by any affine G -equivariant ´etale neighborhood of t ( Z ). By Theorem 2.22, we can thusfind a commutative diagram of G -schemes Z VX Z XZ S , t i p where V (cid:44) → X is a closed G -immersion, p : V → S is ´etale at Z , and all squares are cartesian. By Lemma 2.3,there exists a quasi-compact G -invariant open neighborhood V ◦ of Z on which p is ´etale. Denote by v : V ◦ (cid:44) → V the inclusion. The open G -immersion j : U (cid:44) → S and the ´etale G -morphism pv : V ◦ → S then form aNisnevich cover of S . By Nisnevich separation, it will suffice to show that j ∗ Φ S ( X, t ) and v ∗ p ∗ Φ S ( X, t ) aremotivically contractible. Since Z U = ∅ , j ∗ Φ S ( X, t ) (cid:39) Φ U ( X U , t U ) is a final object in P (Sm GU ). Since v and pv are smooth, we have v ∗ p ∗ Φ S ( X, t ) (cid:39) v ∗ Φ V ( X V , t V ), so it suffices to show that Φ V ( X V , t V ) is motivicallycontractible. But by construction of V , there exists a G -section V (cid:44) → X V extending t V : Z → X V . Thus, itremains to prove ( ∗ ) when S is small and affine and X → S has a G -section s extending t . By Proposition 2.25,there exists a G -morphism h : X → V ( N s ) that is ´etale at s ( S ) and such that hs = z , where z is the zerosection. By Lemma 4.16, h induces a motivic equivalence Φ S ( X, t ) → Φ S ( V ( N s ) , z Z ). Finally, Φ S ( V ( N s ) , z Z )is motivically contractible by Lemma 4.17. (cid:3) Corollary 4.19.
Let i : Z (cid:44) → S be a closed G -immersion. Then the functor i ∗ : H G ( Z ) → H G ( S ) is fullyfaithful. Proof.
Let j : U (cid:44) → S be the complementary open G -immersion. It is clear that i ∗ j (cid:93) : H G ( U ) → H G ( Z ) isthe constant functor with value the initial object. Its right adjoint j ∗ i ∗ is thus the constant functor withvalue the terminal object. Let F ∈ H G ( S ). Applying Theorem 4.18 to i ∗ F , we obtain a cocartesian square U i ∗ FU i ∗ i ∗ i ∗ F , η in H G ( S ), showing that ηi ∗ is an equivalence. It follows from a triangle identity that i ∗ (cid:15) is an equivalence.It remains to show that i ∗ is conservative. By Proposition 4.6 and smooth base change, we can assume that S is small and affine. Let f be a morphism in H G ( Z ) such that i ∗ ( f ) is an equivalence, and let us showthat f is an equivalence. By Proposition 3.16 (1), it suffices to show that f is an equivalence on every affinescheme X ∈ Sm GZ . This follows at once from Corollary 2.23 and Lemma 2.3. (cid:3) Corollary 4.20 (Closed base change) . Let
W YZ X kg i f be a cartesian square of G -schemes where i is a closed G -immersion. Then the exchange transformationEx ∗∗ : f ∗ i ∗ → k ∗ g ∗ : H G ( Z ) → H G ( Y ) is an equivalence.Proof. By Corollary 4.19, it suffices to show that Ex ∗∗ i ∗ is an equivalence. This follows easily from Theo-rem 4.18 and smooth base change. (cid:3) Ambidexterity for smooth projective morphisms
Pointed equivariant motivic spaces.
Let S be a G -scheme. A pointed motivic G -space over S is amotivic G -space X over S equipped with a global section S → X . We denote by H G • ( S ) the ∞ -category ofpointed motivic G -spaces, i.e., the undercategory H G ( S ) S/ . We denote by ( − ) + : H G ( S ) → H G • ( S ) the leftadjoint to the forgetful functor.By [Lur16, Proposition 4.8.2.11], the cartesian symmetric monoidal structure on H G ( S ) extends uniquelyto a symmetric monoidal structure on H G • ( S ) that is compatible with colimits. Its tensor product will bedenoted by ⊗ , and its unit by S .If f : T → S is a G -morphism, both f ∗ : H G ( T ) → H G ( S ) and its left adjoint f ∗ preserve the final object,and hence they lift to an adjunction f ∗ : H G • ( S ) (cid:29) H G • ( T ) : f ∗ . If moreover f is smooth, then f ∗ preserves limits and we therefore have an adjunction f (cid:93) : H G • ( T ) (cid:29) H G • ( S ) : f ∗ . The left adjoint f (cid:93) is characterized by f (cid:93) ( X + ) (cid:39) ( f (cid:93) X ) + . The following pointed version of the smoothprojection formula follows immediately from Proposition 4.3: for B ∈ H G • ( S ) and C ∈ H G • ( T ), the canonicalmap f (cid:93) ( f ∗ B ⊗ C ) → B ⊗ f (cid:93) C is an equivalence. All the other results of § Proposition 5.1.
Let i : Z (cid:44) → S be a closed G -immersion. Then the functor i ∗ : H G • ( Z ) → H G • ( S ) preservescolimits.Proof. One deduces immediately from Corollary 4.13 that i ∗ preserves weakly contractible colimits. It alsotrivially preserves the initial object, and hence it preserves all colimits. (cid:3) HE SIX OPERATIONS IN EQUIVARIANT MOTIVIC HOMOTOPY THEORY 33
It follows from Proposition 5.1 that, for i : Z (cid:44) → S a closed G -immersion, i ∗ has a right adjoint i ! : H G • ( S ) → H G • ( Z ) . Proposition 5.2 (Pointed gluing) . Let i : Z (cid:44) → S be a closed G -immersion with open complement j : U (cid:44) → S .For every X ∈ H G • ( S ) , j (cid:93) j ∗ X → X → i ∗ i ∗ X is a cofiber sequence, and i ∗ i ! X → X → j ∗ j ∗ X is a fiber sequence.Proof. The second statement follows from the first one by adjunction. Denote by u : H G • ( S ) → H G ( S ) theforgetful functor. We consider the following diagram in H G ( S ): j (cid:93) j ∗ u ( X ) j (cid:93) j ∗ u ( X ) (cid:97) U S u ( X ) U S i ∗ i ∗ u ( X ). j The outside square is cocartesian by Theorem 4.18, and the first square is formally cocartesian. Thus, thesecond square is also cocartesian. But the second square is the image by u of the given sequence. It remainsto observe that u reflects weakly contractible colimits. (cid:3) Corollary 5.3 (Smooth–closed base change) . Let
W TZ S tq s p be a cartesian square of G -schemes, where p is smooth and s is a closed immersion. Then the exchangetransformations Ex (cid:93) ∗ : p (cid:93) t ∗ → s ∗ q (cid:93) : H G • ( W ) → H G • ( S ) , Ex ∗ ! : q ∗ s ! → t ! p ∗ : H G • ( S ) → H G • ( W ) are equivalences.Proof. The second transformation is the mate of the first one, so it suffices to show that the first transfor-mation is an equivalence. Since t ∗ is fully faithful, it suffices to show that Ex (cid:93) ∗ t ∗ is an equivalence. Thisfollows easily from Proposition 5.2 and smooth base change. (cid:3) Corollary 5.4 (Closed projection formula) . Let i : Z (cid:44) → S be a closed G -immersion and let A ∈ H G • ( S ) .For every B ∈ H G • ( Z ) and every C ∈ H G • ( S ) , the canonical maps A ⊗ i ∗ B → i ∗ ( i ∗ A ⊗ B ) , Hom( i ∗ A, i ! C ) → i ! Hom(
A, C ) are equivalences.Proof. The first equivalence follows at once from Proposition 5.2 and the smooth projection formula. Thesecond equivalence follows by adjunction. (cid:3)
Spheres, Thom spaces, and purity.
Let S be a G -scheme and let M be a locally free G -module on S . Let p : V ( M ) → S be the associated vector bundle with zero section s . We denote byΣ M : H G • ( S ) (cid:29) H G • ( S ) : Ω M the adjunction p (cid:93) s ∗ (cid:97) s ! p ∗ . We call Σ M X the M -suspension of X and Ω M X the M -loop space of X . Thepointed G -space Σ M S ∈ H G • ( S ) is called the M -sphere and is also denoted by S M .Let f : T → S be a G -morphism and let M be a locally free G -module on S . By smooth and closed basechange, we obtain equivalences f ∗ Σ M (cid:39) Σ f ∗ ( M ) f ∗ and Ω M f ∗ (cid:39) f ∗ Ω f ∗ ( M ) . If moreover f is smooth (resp. a closed immersion), we also have an equivalenceΣ M f (cid:93) (cid:39) f (cid:93) Σ f ∗ ( M ) (resp. f ! Ω M (cid:39) Ω f ∗ ( M ) f ! )by Corollary 5.3.By the smooth and closed projection formulas, we have canonical equivalencesΣ M (cid:39) S M ⊗ ( − ) and Ω M (cid:39) Hom(S M , − ) . By Proposition 5.2, we have S M (cid:39) V ( M ) V ( M ) (cid:114) S , i.e., there is a cofiber sequence ( V ( M ) (cid:114) S ) + → V ( M ) + → S M in H G • ( S ). It follows from this description that the assignments M (cid:55)→ Σ M and M (cid:55)→ Ω M are functors on thecategory of locally free G -modules and epimorphisms (this functoriality can also be described more directlyusing exchange transformations).Let 0 → N → M → P → G -modules on a G -scheme S . In § M (cid:39) S N ⊗ S P . We will also denote by Ψ the induced equivalence of functors Σ M (cid:39) Σ N Σ P . Remark . Given a short exact sequence as above, we can form the diagram S V ( P ) V ( M ) S V ( N ) S , t s paq u b r where p, q, r are the structure maps of the associated vector bundles and s, t, u are their respective zerosections. By Corollary 5.3, the transformation Ex (cid:93) ∗ : b (cid:93) a ∗ → u ∗ q (cid:93) is an equivalence. Using Remark 3.27 andthe fact that any short exact sequence of locally free G -modules splits when pulled back to an appropriate G -affine bundle, it is easy to show that the equivalence Ψ agrees with the transformationΣ M = p (cid:93) s ∗ (cid:39) r (cid:93) b (cid:93) a ∗ t ∗ Ex (cid:93) ∗ −−−→ r (cid:93) u ∗ q (cid:93) t ∗ = Σ N Σ P . We will not need this alternative description.If M is a locally free G -module on X ∈ Sm GS , we letTh X ( M ) = p (cid:93) S M ∈ H G • ( S ) , where p : X → S is the structure map. The pointed G -space Th X ( M ) is called the Thom space of M . HE SIX OPERATIONS IN EQUIVARIANT MOTIVIC HOMOTOPY THEORY 35
We will now recast the purity equivalence Π of § Z XS sq p be a commutative triangle of G -schemes where p and q are smooth and s is a closed immersion. To such asmooth closed pair ( X, Z ) over S we can associate the functor p (cid:93) s ∗ , and to a morphism of smooth closedpairs f : ( X (cid:48) , Z (cid:48) ) → ( X, Z ) we can associate the natural transformation ψ ( f ) : p (cid:48) (cid:93) s (cid:48)∗ f ∗ → p (cid:93) s ∗ (with obvious notation), which is adjoint to the composition s (cid:48)∗ f ∗ Ex ∗∗ ←−− f ∗ s ∗ η −→ f ∗ p ∗ p (cid:93) s ∗ (cid:39) p (cid:48)∗ p (cid:93) s ∗ . By Proposition 5.2, p (cid:93) s ∗ Z is the pointed space X/ ( X (cid:114) Z ), and ψ ( f ) Z is simply the map X (cid:48) / ( X (cid:48) (cid:114) Z (cid:48) ) → X/ ( X (cid:114) Z ) induced by f .Consider the deformation to the normal bundle: Z XS s p i Z × A D Z XS × A s ˆ p i Z V ( N s ) S , s p and denote by r : S × A → S and r : Z × A → Z the projections. We then obtain a zig-zag of naturaltransformations p (cid:93) s ∗ i ∗ ψ ( i ) −−−→ r (cid:93) ˆ p (cid:93) ˆ s ∗ ψ ( i ) ←−−− q (cid:93) Σ N s i ∗ . Precomposing with r ∗ , we obtain the zig-zag p (cid:93) s ∗ → r (cid:93) ˆ p (cid:93) ˆ s ∗ r ∗ ← q (cid:93) Σ N s . Proposition 5.7 (Purity) . With the above notation, the four natural transformations p (cid:93) s ∗ → r (cid:93) ˆ p (cid:93) ˆ s ∗ r ∗ ← q (cid:93) Σ N s ,s ! p ∗ ← r ∗ ˆ s ! ˆ p ∗ r ∗ → Ω N s q ∗ are equivalences.Proof. The second zig-zag is the mate of the first one, so it suffices to show that the first zig-zag is anequivalence. Since s ∗ is fully faithful and all functors involved preserve colimits, it suffices to show that it isan equivalence on s ∗ ( Y + ) for Y ∈ Sm GX . But in that case, by Proposition 5.2, it can be identified with thezig-zag X × X Y ( X (cid:114) Z ) × X Y → D Z X × X Y ( D Z X (cid:114) ( Z × A )) × X Y ← N Z X × X Y ( N Z X (cid:114) Z ) × X Y , which is an equivalence in H G • ( S ) by Theorem 3.23. (cid:3) With every triangle (5.6) are therefore associated canonical equivalencesΠ s : p (cid:93) s ∗ (cid:39) q (cid:93) Σ N s and Π s : s ! p ∗ (cid:39) Ω N s q ∗ , which will be called the purity equivalences , generalizing their namesake from § § The Pontryagin–Thom collapse.
Let S be a G -scheme and let f : X → S be a smooth G -projectivemorphism. Fix a locally free G -module E on S , everywhere of rank ≥
1, and a closed G -immersion i : X (cid:44) → P ( E ) over S . The goal of this subsection is to define the Pontryagin–Thom collapse map η = η ( X,i ) : S M → Th X ( N )in H G • ( S ). Here, M = M ( X,i ) and N = N ( X,i ) are locally free G -modules on S and X , related by an explicitequivalence α = α ( X,i ) : S Ω f ⊗ S N (cid:39) S f ∗ ( M ) in H G • ( X ). When G is trivial and S is the spectrum of a field, η ( X,i ) coincides with the map defined byVoevodsky in [Voe03, Theorem 2.11], up to some O nS -suspension.We warn the reader that the construction of η is rather involved, so we start with an informal discussion.If X is instead a smooth compact manifold over S = ∗ , the classical Pontryagin–Thom collapse map isdefined as follows. We first choose an embedding of X into a ball B inside some large Euclidean space E .The collapse map η : E/ ( E (cid:114) B ) → E/ ( E (cid:114) X ) is then a map from a sphere S E to the Thom space of thenormal bundle of X in E (by the tubular neighborhood theorem). To replicate this construction for X asmooth G -projective S -scheme, one would need an embedding of X into a projective bundle that misses ahyperplane, which is of course impossible (unless X is finite). The key idea is that we can nevertheless findsuch an embedding “up to homotopy”: there exists a zig-zag X p ←− ˜ X s (cid:44) −→ ˜ P ( F ) q −→ P ( F )in Sm GS , where F is a locally free G -module on S , s is a closed G -immersion, and p and q are compositionsof G -affine bundles, such that the following two properties hold. The first is that the image of ˜ X in P ( F )misses a G -invariant hyperplane P ( G ); this allows us to define a collapse map(5.8) S G (cid:39) P ( F ) P ( G ) → Th ˜ X ( N s )in H G • ( S ) (by the purity theorem). The second is that there exists a locally free G -module N on X such thatS p ∗ ( N ) (cid:39) S N s , at least after tensoring with a sphere defined over S . This gives us a “stable” equivalence(5.9) Th ˜ X ( N s ) (cid:39) Th X ( N ) . The Pontryagin–Thom collapse map η is then the composition of (5.8) and (5.9). Some further care mustbe taken in the actual construction of η below, and although it does fit this basic sketch, this will not bemade explicit.We first give the construction of η and α in the special case where i : X (cid:44) → P ( E ) is the identity. Recallfrom § → O S → E ⊗ E ∨ → Q → , where the first map is the coevaluation and Q is locally free. We let P = P ( E ⊗ E ∨ ) , H = P ( Q ) , V = P (cid:114) H. The map v : V → S is the G -affine bundle that universally splits the short exact sequence (5.10). ByProposition 4.4, the functor v ∗ : H G • ( S ) → H G • ( V )is fully faithful. Thus, implicitly replacing S by V , we may assume given a splitting E ⊗ E ∨ (cid:39) O S ⊕ Q of (5.10), so that V is the G -vector bundle V ( Q ) and P is its projective completion. In particular, we havethe usual zig-zag S Q = VV (cid:114) → PP (cid:114) ← PH , where the first map is a Zariski equivalence and the second is a homotopy equivalence (
H (cid:44) → P (cid:114) P ( Q )). HE SIX OPERATIONS IN EQUIVARIANT MOTIVIC HOMOTOPY THEORY 37
Let X ∨ = P ( E ∨ ) and let σ : X × X ∨ (cid:44) → P be the Segre embedding (here and in what follows, productsare taken over S ). Consider the cartesian squares˜ X X × X ∨ KV P H . jσ σ σ We denote by π : ˜ X → X and ρ : ˜ X → X ∨ the canonical projections; recall from § π is a G -affinebundle. We also consider the cartesian squares˜ K ˜ X × X ∨ ˜ XK X × X ∨ X , π π π and we let s : ˜ X (cid:44) → ˜ X × X ∨ , x (cid:55)→ ( x, ρx ) , be the graph of ρ . We then have the following zig-zag X ˜ X ˜ X × X ∨ X × X ∨ P , π s π σj where the vertical maps are G -affine bundles and the horizontal maps are closed G -immersions. Note that s ( ˜ X ) ∩ ˜ K = ∅ since j ( ˜ X ) ∩ K = ∅ . We therefore obtain the following collapse map in H G • ( S ):(5.11) S Q (cid:39) PH → PH ∪ ( P (cid:114) σ ( X × X ∨ )) Π (cid:39) Th X × X ∨ ( N σ )Th K ( N σ ) π ← Th ˜ X × X ∨ ( π ∗ ( N σ ))Th ˜ K ( π ∗ ( N σ )) → V ( π ∗ ( N σ )) V ( π ∗ ( N σ )) (cid:114) s ( ˜ X ) Π (cid:39) Th ˜ X ( N s ⊕ j ∗ ( N σ )) . Consider the canonical short exact sequence0 → Ω X → f ∗ ( E )( − → O X → G ( X ). Tensoring it with its dual, we obtain the following diagram of short exact sequences:(5.12) 0 0 00 Ω X f ∗ ( E )( − O X
00 Ω X ⊗ f ∗ ( E ∨ )(1) f ∗ ( E ⊗ E ∨ ) f ∗ ( E ∨ )(1) 00 Ω X ⊗ Ω ∨ X f ∗ ( E )( − ⊗ Ω ∨ X Ω ∨ X We define R = f ∗ ( E )( − ⊗ Ω ∨ X . The short exact sequences (5.12) define an equivalence(5.13) Σ O Σ Ω X Σ R (cid:39) Σ f ∗ ( E ⊗ E ∨ ) of endofunctors of H G • ( X ). Remark . In the sequel, we will never use the definitions of R and of the equivalence (5.13). Thus,any choice of a locally free G -module R on X together with such an equivalence can be used to define aPontryagin–Thom collapse map. Different choices will be shown to yield “stably” equivalent maps.The Segre embedding σ induces the short exact sequence(5.15) 0 → N σ → σ ∗ (Ω P ) → Ω X × X ∨ → . With the obvious isomorphismsΩ V (cid:39) v ∗ ( Q ) , Ω ˜ X (cid:39) π ∗ (Ω X ) ⊕ ρ ∗ (Ω X ∨ ) , N s (cid:39) ρ ∗ (Ω X ∨ ) , the short exact sequence j ∗ (5.15) becomes0 → j ∗ ( N σ ) → ˜ f ∗ ( Q ) → π ∗ (Ω X ) ⊕ N s → , where ˜ f = f π : ˜ X → S is the structure map. This induces an equivalence(5.16) Σ π ∗ (Ω X ) Σ j ∗ ( N σ ) ⊕ N s (cid:39) Σ ˜ f ∗ ( Q ) of endofunctors of H G • ( ˜ X ). Combining (5.13) and (5.16), we obtain an equivalence(5.17) Σ O Σ π ∗ (Ω X ) Σ π ∗ ( R ) (cid:39) Σ O Σ π ∗ (Ω X ) Σ j ∗ ( N σ ) ⊕ N s . Combining (5.17) and (5.13), we further obtain an equivalence(5.18) Σ ˜ f ∗ ( E ⊗ E ∨ ) Σ π ∗ ( R ) (cid:39) Σ ˜ f ∗ ( E ⊗ E ∨ ) Σ j ∗ ( N σ ) ⊕ N s . We let M = ( E ⊗ E ∨ ) ⊕ Q , N = f ∗ ( E ⊗ E ∨ ) ⊕ R . From (5.13) we obtain the equivalence α = α ( P ( E ) , id) as follows:(5.19) Σ O Σ Ω X Σ R (cid:39) Σ O Σ f ∗ ( Q ) = ⇒ Σ Ω X Σ R Σ O (cid:124) (cid:123)(cid:122) (cid:125) Σ f ∗ ( E ⊗ E ∨ ) Σ Ω X Σ R (cid:39) Σ Ω X Σ R Σ O (cid:124) (cid:123)(cid:122) (cid:125) Σ f ∗ ( E ⊗ E ∨ ) Σ f ∗ ( Q ) = ⇒ Σ Ω X Σ N α (cid:39) Σ f ∗ ( M ) . Finally, the Pontryagin–Thom collapse map η = η ( P ( E ) , id) is defined as the compositionS M (cid:39) Σ E ⊗ E ∨ S Q (5.11) −−−−→ Σ E ⊗ E ∨ Th ˜ X ( N s ⊕ j ∗ ( N σ )) (5.18) (cid:39) Th ˜ X ( π ∗ ( N )) (cid:39) Th X ( N ) . This concludes the construction of α ( X,i ) and η ( X,i ) when i is the identity. In general, let M ( X,i ) be M ( P ( E ) , id) , and let N ( X,i ) be the conormal sheaf of the immersion X (cid:44) → V ( N ( P ( E ) , id) ), which is canonicallyidentified with N i ⊕ i ∗ ( N ( P ( E ) , id) ). Then the equivalence α ( X,i ) is given byΣ Ω X Σ N ( X,i ) = Σ Ω X Σ N i Σ i ∗ ( N ( P ( E ) , id) ) Ψ (cid:39) Σ i ∗ (Ω P ( E ) ) Σ i ∗ ( N ( P ( E ) , id) ) α ( P ( E ) , id) (cid:39) Σ f ∗ ( M ) , and the Pontryagin–Thom collapse map η ( X,i ) is the compositionS M ( X,i ) η ( P ( E ) , id) −−−−−→ Th P ( E ) ( N ( P ( E ) , id) ) → V ( N ( P ( E ) , id) ) V ( N ( P ( E ) , id) ) (cid:114) i ( X ) Π (cid:39) Th X ( N ( X,i ) ) . Now that η = η ( X,i ) has been defined, we can upgrade it to a natural transformation η : Σ M → f (cid:93) Σ N f ∗ using the projection formulasΣ M (cid:39) S M ⊗ ( − ) and f (cid:93) Σ N f ∗ (cid:39) Th X ( N ) ⊗ ( − ) . We denote by (cid:15) : f ∗ f (cid:93) → Σ Ω f HE SIX OPERATIONS IN EQUIVARIANT MOTIVIC HOMOTOPY THEORY 39 the composition(5.20) f ∗ f (cid:93) (cid:39) π (cid:93) π ∗ → π (cid:93) δ ∗ δ ∗ π ∗ (cid:39) π (cid:93) δ ∗ Π (cid:39) Σ N δ ν (cid:39) Σ Ω f , where π , : X × X ⇒ X are the projections, δ : X (cid:44) → X × X is the diagonal, and ν : N δ → Ω f is theisomorphism sending the class of x ⊗ − ⊗ x to dx . Note that, unlike η , (cid:15) does not depend on a choice ofembedding i : X (cid:44) → P ( E ). For Y ∈ Sm GX , the component of (cid:15) at Y + is given more explicitly by the map( Y × X ) + → Y × X ( Y × X ) (cid:114) Y Π (cid:39) Σ N δ Y + ν (cid:39) Σ Ω f Y + collapsing the complement of the graph of Y → X , where Y × X belongs to Sm GX via the second projection. Lemma 5.21.
Let S be a G -scheme and E a locally free G -module on S . Then the transposition on S E ⊗ S E in H G • ( S ) is homotopic to (cid:104)− (cid:105) ⊗ id , where (cid:104)− (cid:105) : S E → S E is induced by the linear automorphism E → E , x (cid:55)→ − x .Proof. The matrices (cid:18) (cid:19) and (cid:18) − (cid:19) are related by elementary transformations and hence are A -homotopic in SL ( Z ). We conclude using theaction of SL ( Z ) on V ( E ⊕ E ) / ( V ( E ⊕ E ) (cid:114) (cid:3) The following theorem is the last nontrivial result in our approach to the formalism of six operations:
Theorem 5.22 (Unstable ambidexterity) . Let E be a locally free G -module of rank ≥ on S and let X P ( E ) S if be a commuting triangle of G -schemes where i is a closed immersion and f is smooth. Then the compositions f ∗ Σ M η −→ f ∗ f (cid:93) Σ N f ∗ (cid:15) −→ Σ Ω f Σ N f ∗ α (cid:39) Σ f ∗ ( M ) f ∗ (cid:39) f ∗ Σ M , (5.23) Σ M f (cid:93) η −→ f (cid:93) Σ N f ∗ f (cid:93) (cid:15) −→ f (cid:93) Σ N Σ Ω f α (cid:39) f (cid:93) Σ f ∗ ( M ) (cid:39) Σ M f (cid:93) (5.24) are the identity.Proof. We keep using the notation introduced above in the construction of η . In addition, if e : Z (cid:44) → Y is aclosed G -immersion, we will abbreviate Y (cid:114) e ( Z ) to e c or Z c . Let ι be the transformation (5.23) evaluatedon S . It is not difficult to check that (5.23) can be identified with ι ⊗ f ∗ ( − ) and (5.24) with f (cid:93) ( ι ⊗ ( − )).Thus, it will suffice to show that ι is the identity in H G • ( X ). Let us first reduce to the case where i : X (cid:44) → P ( E ) is the identity. In fact, we claim that ι X = i ∗ ( ι P ( E ) ) inH G • ( X ). Recall that N ( X,i ) = N i ⊕ i ∗ ( N ( P ( E ) , id) ). We consider the following diagram in H G • ( X ):Th X ( f ∗ ( M )) V ( N P ( E ) ) × X ( P ( E ) × X ) c V ( N P ( E ) ) × X ( X × X ) c V ( N X ) × X ( X × X ) c V ( N P ( E ) ) × X ∆ cX V ( N X ) × X ∆ cX Th X ( N ( i × id) δ X ⊕ i ∗ ( N P ( E ) )) Th X ( N δ X ⊕ N X )Th X ( i ∗ ( N δ P ( E ) ) ⊕ i ∗ ( N P ( E ) )) Th X (Ω X ⊕ N X )Th X ( i ∗ (Ω P ( E ) ) ⊕ i ∗ ( N P ( E ) )) Th X ( f ∗ ( M )). f ∗ ( η P ( E ) ) Π i Π i Π ( i × id) δ Π δ Ψ (cid:39) νν α X i ∗ ( α P ( E ) )Ψ The lower composition is i ∗ ( ι P ( E ) ) and the upper composition is ι X . The middle rectangle commutes byCorollary 3.25 applied to the smooth closed triple ( V ( N P ( E ) ) × X, X × X, ∆ X ). The commutativity of thetrapezoid follows from the commutative diagram of canonical short exact sequences δ ∗ X ( N i × id ) N ( i × id) δ X N δ X i ∗ ( N δ P ( E ) ) N i i ∗ (Ω P ( E ) ) Ω X . (cid:39) (cid:39) ν ν Finally, the commutativity of the lower triangle is the definition of α ( X,i ) in terms of α ( P ( E ) , id) .From now on, we therefore assume that X = P ( E ). As in the definition of η , we implicitly pull backeverything along the G -affine bundle v : V → S , so as to have a canonical isomorphism E ⊗ E ∨ (cid:39) O S ⊕ Q .Let γ : ˜ X (cid:44) → P × ˜ X be the graph of the G -immersion σj : ˜ X (cid:44) → P . From the short exact sequence0 → N γ → σ ∗ (Ω V ) ⊕ Ω ˜ X dσ +id −−−−→ Ω ˜ X → , we obtain an isomorphism µ : ˜ f ∗ ( Q ) = σ ∗ (Ω V ) (cid:39) N γ sending a section x of σ ∗ (Ω V ) to ( x, − ( dσ )( x )). We claim that the following rectangle commutes in H G • ( X ),where the first row is ι :(5.25) S f ∗ ( M ) f ∗ (Th X ( N )) Σ Ω f S N S f ∗ ( M ) Σ f ∗ ( M ) ˜ X + Σ f ∗ ( E ⊗ E ∨ ) P × ˜ XH × ˜ X Σ f ∗ ( E ⊗ E ∨ ) P × ˜ Xγ c Σ f ∗ ( M ) ˜ X + . f ∗ ( η ) (cid:15) α (cid:39) µ Π π (cid:39) π (cid:39) Assuming this for the moment, let us conclude the proof of the theorem by showing that the bottom rowof (5.25) is the identity. Let ζ : ˜ X (cid:44) → V × ˜ X ⊂ P × ˜ X be the zero section. Consider the map φ : A × V × ˜ X → V × ˜ X, ( t, v, x ) (cid:55)→ ( v − tσ ( x ) , x ) . HE SIX OPERATIONS IN EQUIVARIANT MOTIVIC HOMOTOPY THEORY 41
This is an A -family of linear automorphisms of the G -vector bundle V × ˜ X over ˜ X . Note that φ is theidentity and φ γ = ζ . Its projective completion is an A -familyˆ φ : A × P × ˜ X → P × ˜ X. We now consider the following diagram:Σ f ∗ ( Q ) ˜ X + P × ˜ XH × ˜ X P × ˜ Xγ c Σ f ∗ ( Q ) ˜ X + Σ f ∗ ( Q ) ˜ X + P × ˜ XH × ˜ X P × ˜ Xζ c Σ f ∗ ( Q ) ˜ X + , (cid:39) Π (cid:39) Π φ ˆ φ ˆ φ where for the bottom purity equivalence we use the obvious isomorphism N ζ (cid:39) ˜ f ∗ ( Q ). The commutativityof the first two squares is clear. The last square commutes because the isomorphism N γ (cid:39) N ζ restrictionof γ ∗ ( dφ ) is exactly the composite N γ (cid:39) ˜ f ∗ ( Q ) (cid:39) N ζ of the given isomorphisms. The left vertical arrowis A -homotopic to the identity via φ , and it is clear that the lower row is the identity, by definition of theequivalence S Q (cid:39) P/H .It remains to prove the commutativity of the rectangle (5.25). The proof is mostly straightforward butthere are a few subtle points. We first note that there is a commutative squareS f ∗ ( M ) S f ∗ ( M ) Σ f ∗ ( M ) ˜ X + Σ f ∗ ( M ) ˜ X + ιπ ππ (cid:93) π ∗ ( ι ) in H G • ( X ), and that π (cid:93) π ∗ ( ι ) is the composition of the following four maps:(1) the first part of π (cid:93) ˜ f ∗ ( η ): π (cid:93) ˜ f ∗ of the ( E ⊗ E ∨ )-suspension of (5.11);(2) the second part of π (cid:93) ˜ f ∗ ( η ): π (cid:93) ˜ f ∗ ˜ f (cid:93) of (5.18) ˜ X , followed by the projection π × id : Th ˜ X × ˜ X ( π ∗ π ∗ ( N )) → Th X × ˜ X ( π ∗ ( N ));(3) π (cid:93) π ∗ ( (cid:15) ), which collapses the complement of ˜ δ : ˜ X (cid:44) → X × ˜ X :Th X × ˜ X ( π ∗ ( N )) → V ( π ∗ ( N ))˜ δ c Π (cid:39) Th ˜ X ( N ˜ δ ⊕ π ∗ ( N )) ν (cid:39) Σ Ω f ⊕ N ˜ X + ;(4) finally, the equivalence α : π (cid:93) π ∗ (5.19) X : Σ Ω f ⊕ N ˜ X + (cid:39) Σ f ∗ ( M ) ˜ X + . We must prove that π (cid:93) π ∗ ( ι ) coincides with the lower row of (5.25). Let us first simplify steps (2)–(4); this iswhere the precise definitions of (5.18) and (5.19) come into play. Let ¯ δ : ˜ X × X ˜ X (cid:44) → ˜ X × ˜ X be the obviousclosed immersion, which is the pullback of ˜ δ along π × id. We contemplate the following diagram in H G • ( X ), where Σ (cid:48) stands for Σ f ∗ ( E ⊗ E ∨ ) :Σ (cid:48) Th ˜ X × ˜ X ( π ∗ ( N s ⊕ j ∗ ( N σ ))) Σ (cid:48) Th ˜ X × ˜ X ( π ∗ π ∗ ( R )) Σ (cid:48) Th X × ˜ X ( π ∗ ( R ))Σ (cid:48) V ( π ∗ ( N s ⊕ j ∗ ( N σ )))¯ δ c Σ (cid:48) V ( π ∗ π ∗ ( R ))¯ δ c Σ (cid:48) V ( π ∗ ( R ))˜ δ c Σ (cid:48) Th ˜ X × X ˜ X ( N ¯ δ ⊕ π ∗ ( N s ⊕ j ∗ ( N σ ))) Σ (cid:48) Th ˜ X × X ˜ X ( N ¯ δ ⊕ π ∗ π ∗ ( R )) Σ (cid:48) Th ˜ X ( π ∗ ( N δ ) ⊕ π ∗ ( R ))Σ (cid:48) Th ˜ X ( π ∗ (Ω X ) ⊕ N s ⊕ j ∗ ( N σ )) Σ (cid:48) Th ˜ X ( π ∗ (Ω X ) ⊕ π ∗ ( R ))Σ (cid:48) Th ˜ X ( π ∗ (Ω X ) ⊕ N s ⊕ j ∗ ( N σ )) Σ (cid:48) Th ˜ X ( ˜ f ∗ ( Q )). (5.18)(5.18)(5.18) π × id π × idΠ ¯ δ Π ¯ δ Π ˜ δ νπ (5.18) π π (5.16) (5.19) ν The composition of the top row and the right column is the composition of steps (2)–(4). Note that theparallel equivalences π and π are retractions of the same map and hence are homotopic in H G • ( X ). Thecommutativity of each square is clear except the last one. Unfolding the definitions of the three equivalencesinvolved, we see that its commutativity is equivalent to that of the following rectangle, where τ exchangesthe two occurrences of Σ π ∗ (Ω) :Σ π ∗ (Ω) Σ π ∗ ( R ) Σ O Σ π ∗ (Ω) Σ N s ⊕ j ∗ ( N σ ) Σ π ∗ (Ω) Σ π ∗ ( R ) Σ O Σ ˜ f ∗ ( Q ) Σ π ∗ (Ω) Σ π ∗ ( R ) Σ O Σ π ∗ (Ω) Σ π ∗ ( R ) Σ π ∗ (Ω) Σ π ∗ ( R ) Σ O Σ π ∗ (Ω) Σ N s ⊕ j ∗ ( N σ ) Σ π ∗ (Ω) Σ π ∗ ( R ) Σ O Σ ˜ f ∗ ( Q ) Σ π ∗ (Ω) Σ π ∗ ( R ) Σ O Σ π ∗ (Ω) Σ π ∗ ( R ) . τ τ (5.16) (5.13)(5.16) (5.13) This follows immediately from Lemma 5.21.Thus, the composition of (2)–(4) coincides with the f ∗ ( E ⊗ E ∨ )-suspension of the following composition:(5.26) Th ˜ X × ˜ X ( π ∗ ( N s ⊕ j ∗ ( N σ ))) → V ( π ∗ ( N s ⊕ j ∗ ( N σ )))¯ δ c Π (cid:39) Th ˜ X × X ˜ X ( N ¯ δ ⊕ π ∗ ( N s ⊕ j ∗ ( N σ ))) π → Th ˜ X ( π ∗ ( N δ ) ⊕ N s ⊕ j ∗ ( N σ )) (5.16) ν (cid:39) Σ f ∗ ( Q ) ˜ X + . On the other hand, (1) is the f ∗ ( E ⊗ E ∨ )-suspension of the following composition:(5.27) Σ f ∗ ( Q ) ˜ X + (cid:39) P × ˜ XH × ˜ X → P × ˜ X ( H ∪ σ c ) × ˜ X Π (cid:39) Th X × X ∨ × ˜ X ( π ∗ ( N σ ))Th K × ˜ X ( π ∗ ( N σ )) π ← Th ˜ X × X ∨ × ˜ X ( π ∗ π ∗ ( N σ ))Th ˜ K × ˜ X ( π ∗ π ∗ ( N σ )) → V ( π ∗ π ∗ ( N σ )) × ˜ Xs c × ˜ X Π (cid:39) Th ˜ X × ˜ X ( π ∗ ( N s ⊕ j ∗ ( N σ ))) . We must therefore show that the f ∗ ( E ⊗ E ∨ )-suspension of (5.26) ◦ (5.27) coincides with the bottom rowof (5.25). Consider the closed immersion β = ( j, id) : ˜ X (cid:44) → X × X ∨ × ˜ X, x (cid:55)→ ( πx, ρx, x ) , and the two closed immersions β : ˜ X × X ˜ X (cid:44) → ˜ X × X ∨ × ˜ X, ( x, y ) (cid:55)→ ( x, ρx, y ) ,β : ˜ X × X ˜ X (cid:44) → ˜ X × X ∨ × ˜ X, ( x, y ) (cid:55)→ ( x, ρy, y ) . HE SIX OPERATIONS IN EQUIVARIANT MOTIVIC HOMOTOPY THEORY 43
Note that β and β do not define the same closed subscheme of ˜ X × X ∨ × ˜ X . Instead, there is a commutativesquare ˜ X × X ∨ × ˜ X ˜ X × X ∨ × ˜ X ˜ X × X ˜ X ˜ X × X ˜ X , ττβ β where τ exchanges the two copies of ˜ X , inducing an isomorphism τ ∗ ( N β ) (cid:39) N β . In particular, we havecanonical isomorphisms π ∗ ( N β ) (cid:39) N β and π ∗ ( N β ) (cid:39) N β . This is used to define the arrows labeled π and τ in the following diagram: P × ˜ X ( H ∪ σ c ) × ˜ X P × ˜ Xγ c Th ˜ X ( N γ )Th X × X ∨ × ˜ X ( π ∗ ( N σ ))Th K × X ( π ∗ ( N σ )) V ( π ∗ ( N σ )) β c Th ˜ X ( N β ⊕ j ∗ ( N σ ))Th ˜ X × X ∨ × ˜ X ( π ∗ π ∗ ( N σ ))Th ˜ K × X ( π ∗ π ∗ ( N σ )) V ( π ∗ π ∗ ( N σ )) β c Th ˜ X × X ˜ X ( N β ⊕ π ∗ j ∗ ( N σ )) V ( π ∗ π ∗ ( N σ )) s c × ˜ X V ( π ∗ π ∗ ( N σ )) β c Th ˜ X × X ˜ X ( N β ⊕ π ∗ j ∗ ( N σ ))Th ˜ X × ˜ X ( π ∗ ( N s ⊕ j ∗ ( N σ ))) V ( π ∗ ( N s ⊕ j ∗ ( N σ )))¯ δ c Th ˜ X × X ˜ X ( N ¯ δ ⊕ π ∗ ( N s ⊕ j ∗ ( N σ )))Th ˜ X ( π ∗ ( N δ ) ⊕ N s ⊕ j ∗ ( N σ ))Th ˜ X ( ˜ f ∗ ( Q )). Π σ π Π σ π Ψ π ( ∗ ) τ Π s × id Π s × id ΨΠ γ Π β Π β Π β Π ¯ δ π (5.16) ν All the unlabeled arrows are quotient maps, and all the labeled ones are equivalences. The commutativityof the two top squares and of the two bottom squares follows from Corollary 3.25. The commutativity ofthe other two small squares is obvious. To complete the proof, we will show that:(a) the composition of the right column is induced by the isomorphism µ : N γ (cid:39) ˜ f ∗ ( Q );(b) despite appearances, the rectangle ( ∗ ) commutes.Let us prove (a). The first equivalence labeled Ψ in the above diagram is induced by the short exactsequence of conormal sheaves(5.28) 0 → j ∗ ( N σ ) → N γ → N β → γ = ( σ × id) ◦ β . The second one is similarly induced by the short exact sequenceassociated with the triangle β = ( s × id) ◦ ¯ δ ; it is the pullback by π of the short exact sequence(5.29) 0 → N s → N β → π ∗ ( N δ ) → associated with the triangle β = (id × s ) ◦ ˜ δ . Thus, we have a commuting squareTh ˜ X × X ˜ X ( N β ⊕ π ∗ j ∗ ( N σ )) Th ˜ X × X ˜ X ( N ¯ δ ⊕ π ∗ ( N s ⊕ j ∗ ( N σ )))Th ˜ X ( N β ⊕ j ∗ ( N σ )) Th ˜ X ( π ∗ ( N δ ) ⊕ N s ⊕ j ∗ ( N σ )), ΨΨ π π allowing us to commute Ψ and π . The resulting composition π τ π − is clearly the identity. To prove (a),it remains to show that the compositionΣ N γ (5.28) (cid:39) Σ N β ⊕ j ∗ ( N σ ) (5.29) (cid:39) Σ π ∗ ( N δ ) ⊕ N s ⊕ j ∗ ( N σ ) ν (cid:39) Σ π ∗ (Ω X ) ⊕ N s ⊕ j ∗ ( N σ ) (5.16) (cid:39) Σ ˜ f ∗ ( Q ) is induced by µ . Recall that (5.16) is induced by the short exact sequence0 → j ∗ ( N σ ) → σ ∗ (Ω V ) dσ −→ Ω ˜ X → ˜ X (cid:39) π ∗ (Ω X ) ⊕ ρ ∗ (Ω X ∨ ) and ρ ∗ (Ω X ∨ ) (cid:39) N s . Note that N β is the conormalsheaf of the diagonal of ˜ X ; let ν (cid:48) : N β → Ω ˜ X be the isomorphism sending the class of x ⊗ − ⊗ x to dx . The claim now follows from the followingisomorphisms of short exact sequences: j ∗ ( N σ ) N γ N β j ∗ ( N σ ) σ ∗ (Ω V ) Ω ˜ X , µ ν (cid:48) N s N β π ∗ ( N δ ) ρ ∗ (Ω X ∨ ) Ω ˜ X π ∗ (Ω X ). (cid:39) ν (cid:48) ν To prove (b), note that both β and β factor through the closed G -immersion λ : ˜ X × X ˜ X × X ∨ (cid:44) → ˜ X × X ∨ × ˜ X. Let us write β , = λ ◦ ζ , , where ζ , : ˜ X × X ˜ X (cid:44) → ˜ X × X ˜ X × X ∨ , and let κ : ˜ X × X ˜ X × X ∨ → X × X ∨ be the projection. We can then break up the rectangle ( ∗ ) as follows: V ( π ∗ π ∗ ( N σ )) β c V ( π ∗ π ∗ ( N σ )) λ c V ( π ∗ π ∗ ( N σ )) β c V ( N λ ⊕ κ ∗ ( N σ )) ζ c Th ˜ X × X ˜ X × X ∨ ( N λ ⊕ κ ∗ ( N σ )) V ( N λ ⊕ κ ∗ ( N σ )) ζ c Th ˜ X × X ˜ X ( N ζ ⊕ ζ ∗ ( N λ ) ⊕ π ∗ j ∗ ( N σ )) Th ˜ X × X ˜ X ( N ζ ⊕ ζ ∗ ( N λ ) ⊕ π ∗ j ∗ ( N σ ))Th ˜ X × X ˜ X ( N β ⊕ π ∗ j ∗ ( N σ )) Th ˜ X × X ˜ X ( N β ⊕ π ∗ j ∗ ( N σ )). ττ Π λ Π λ Π λ Π ζ Π ζ Ψ Ψ
Only the commutativity of the middle rectangle is not clear. However, it is clear that the middle rectanglecommutes if we add in the automorphism τ of Th ˜ X × X ˜ X × X ∨ ( N λ ⊕ κ ∗ ( N σ )) that permutes the two factorsof ˜ X (defined on the Thom space since λτ = τ λ and τ κ = κ ). The claim follows from the observation thatthis automorphism is homotopic to the identity, since the diagonal ˜ X (cid:44) → ˜ X × X ˜ X is a section of a G -vectorbundle. (cid:3) HE SIX OPERATIONS IN EQUIVARIANT MOTIVIC HOMOTOPY THEORY 45 Stable equivariant motivic homotopy theory
In this section, we construct the stable equivariant motivic ∞ -category SH G ( S ) for a G -scheme S . Recallthat we have fixed a qcqs base scheme B and a tame group scheme G over B . Throughout this section, wemoreover assume that one of the following conditions holds: • G is finite locally free; or • B has the G -resolution property.This restriction is not essential, but the definition of SH G ( S ) in general is more complicated; it is howeverdetermined by the requirement that SH G ( − ) be a Nisnevich sheaf.Let P r L denote the ∞ -category whose objects are presentable ∞ -categories and whose morphisms arecolimit-preserving functors. Recall that P r L admits limits and colimits, and that the former are computed in C at ∞ [Lur09, § P r L , ⊗ [Lur16, Proposition 4.8.1.14], andwe call a commutative algebra in P r L , ⊗ a presentably symmetric monoidal ∞ -category. For C ∈ CAlg( P r L , ⊗ ),a C -module will always mean a C -module in the symmetric monoidal ∞ -category P r L , ⊗ ; we denote by Mod C the ∞ -category of C -modules.6.1. Equivariant motivic spectra.
To transform the “suspended adjunction” of Theorem 5.22 into agenuine adjunction, we need to make the spheres S M invertible for the tensor product of pointed motivic G -spaces. To that end, we use the formalism developed by Robalo in [Rob15, § ∞ -category C and a set of objects X in C , there exists a functor C → C [ X − ]with the following universal property in the ∞ -category CAlg( P r L , ⊗ ): any functor f : C → D such that f ( x )is invertible for all x ∈ X factors uniquely through C [ X − ]. In particular, the ∞ -category of C [ X − ]-modules(in P r L , ⊗ ) is the full subcategory of C -modules on which all the objects of X act by equivalences. In loc. cit. ,this construction is considered only when X has a single element, but it is clear that the filtered colimit C [ X − ] = colim F ⊂ XF finite C [( (cid:79) F ) − ]satisfies the required universal property.If M is a C -module and x ∈ C , let Stab x ( M ) denote the colimit of the sequence M −⊗ x −−−→ M −⊗ x −−−→ M −⊗ x −−−→ · · · in Mod C . Note that the underlying ∞ -category of Stab x ( M ) is the limit of the tower · · · Hom( x, − ) −−−−−−→ M Hom( x, − ) −−−−−−→ M Hom( x, − ) −−−−−−→ M . More generally, we define the C -module Stab X ( M ) as follows. Let L denote the 1-skeleton of the nerve ofthe poset N , and let L ( X ) be the simplicial set of almost zero maps X → L . Note that L ( X ) is a filteredsimplicial set and is the union of the simplicial subsets L ( F ) for finite subsets F ⊂ X . Using the symmetricmonoidal structure on C , we can construct a diagram L ( X ) → Mod C sending each vertex to M and eachedge in the x -direction to the functor ( − ) ⊗ x . ∗ We let Stab X ( M ) be the colimit of this diagram. Cofinalityconsiderations show that Stab X ( M ) = colim F ⊂ XF finite Stab (cid:78) F ( M ) . Informally speaking, an object in Stab X ( M ) is an object of M equipped with compatible w -deloopings for w any finite tensor product of elements of X .Recall from [Rob15, Remark 2.20] that an object x ∈ C is n -symmetric if the cyclic permutation of x ⊗ n is homotopic to the identity. If x ∈ C is n -symmetric for some n ≥ M is a C -module, then x acts onStab x ( M ) by an equivalence. If this holds for every x ∈ X , there results a canonical map of C [ X − ]-modules M ⊗ C C [ X − ] → Stab X ( M ) , ∗ More precisely, this amounts to defining a functor P fin ( X ) → Mod C , where P fin ( X ) is the poset of finite subsets of X .Considering the universal case, we have to define a functor P fin ( X ) → B ( (cid:96) n ≥ X n / Σ n ), where the target is a (2 , P fin ( X ) to the simplicial bar construction on themonoidal groupoid (cid:96) n ≥ X n / Σ n , sending a k -simplex Y ⊂ · · · ⊂ Y k to ( Y (cid:114) Y , . . . , Y k (cid:114) Y k − ). which is an equivalence [Rob15, Corollary 2.22]. In particular, we obtain the following explicit descriptionof C [ X − ] as a C -module: C [ X − ] (cid:39) Stab X ( C ) . If C → D is a morphism of presentably symmetric monoidal ∞ -categories, there is an induced base changefunctor ( − ) ⊗ C D : Mod C → Mod D between the ∞ -categories of modules. Below we will need to know that an adjoint pair f : M (cid:29) N : g inMod C (i.e., both f and g are morphisms of C -modules in P r L , ⊗ ) gives rise to an adjoint pair M ⊗ C D (cid:29) N ⊗ C D in Mod D . This follows from the fact that the above base change functor is in fact an ( ∞ , D = C [ X − ] and that each x ∈ X is n -symmetric for some n ≥
2, so that M ⊗ C D (cid:39) Stab X ( M ) as C -modules. The claim is then thatthe functors Stab X ( f ) and Stab X ( g ) form an adjoint pair. If we write Stab X ( M ) and Stab X ( N ) as cofilteredlimits of ∞ -categories indexed by L ( X ) op , the right adjoint to Stab X ( f ) is the functor induced in the limitby the right adjoint to f , and similarly for Stab X ( g ). The claim thus follows from the fact that limits of ∞ -categories preserve adjunctions, since they preserve unit transformations [Lur09, Definition 5.2.2.7]. Inparticular, if f or g is fully faithful, so is Stab X ( f ) or Stab X ( g ).We are now ready to define the stable equivariant motivic homotopy ∞ -category. If S is a G -scheme, wedenote by Sph S the collection of all spheres S E in H G • ( S ), where E is a locally free G -module of finite rankon S . Definition 6.1.
Let S be a G -scheme with structure map p : S → B . The symmetric monoidal ∞ -categoryof motivic G -spectra over S is defined bySH G ( S ) = H G • ( S )[ p ∗ (Sph B ) − ] . We denote by Σ ∞ : H G • ( S ) (cid:29) SH G ( S ) : Ω ∞ the canonical adjunction, where Σ ∞ is symmetric monoidal.We will see in Corollary 6.7 below that SH G ( S ) = H G • ( S )[Sph − S ], which is the intended definition, butthe above definition makes it easier to extend the functorial properties of H G • ( − ) to SH G ( − ). If E is a locallyfree G -module on S , we will denote by S − E the ⊗ -inverse of S E = Σ ∞ S E in SH G ( S ). Note that SH G ( S ) isstable since S O (cid:39) Σ( A (cid:114) , G -morphism f : T → S , the universal property of SH G ( S ) yields an adjunction f ∗ : SH G ( S ) (cid:29) SH G ( T ) : f ∗ where the left adjoint is symmetric monoidal, such that f ∗ Σ ∞ = Σ ∞ f ∗ and Ω ∞ f ∗ = f ∗ Ω ∞ . Lemma 6.2.
For every G -morphism f : T → S , the functor H G • ( T ) ⊗ H G • ( S ) SH G ( S ) → SH G ( T ) induced by f ∗ : SH G ( S ) → SH G ( T ) is an equivalence of symmetric monoidal ∞ -categories.Proof. Compare universal properties. (cid:3)
Lemma 6.3.
Let S be a G -scheme. Every sphere S E ∈ H G • ( S ) is -symmetric.Proof. This follows at once from Lemma 5.21. (cid:3)
Proposition 6.4.
Let S be a G -scheme with structure map p : S → B . (1) There is a canonical equivalence of H G • ( S ) -modules SH G ( S ) (cid:39) Stab p ∗ (Sph B ) H G • ( S ) . (2) The ∞ -category SH G ( S ) is generated under sifted colimits by E − ⊗ Σ ∞ X + , where E ∈ p ∗ (Sph B ) and X ∈ Sm GS is small and affine. (3) For every X ∈ Sm GS , Σ ∞ X + is compact in SH G ( S ) . (4) For every G -morphism f : T → S , the functor f ∗ : SH G ( T ) → SH G ( S ) preserves colimits. HE SIX OPERATIONS IN EQUIVARIANT MOTIVIC HOMOTOPY THEORY 47
Proof. (1) This follows from Lemma 6.3 and the above discussion.(2) By (1) and [Lur09, Lemma 6.3.3.6], every E ∈ SH G ( S ) can be written as a filtered colimit of objectsof the form E − ⊗ Σ ∞ X with X ∈ H G • ( S ). Since the adjunction H G ( S ) (cid:29) H G • ( S ) is monadic, every such X is a simplicial colimit of objects in the image of the left adjoint, and we conclude with Proposition 3.16 (1).(3) By Proposition 3.16 (3), the objects X + for X ∈ Sm GS are compact generators of H G • ( S ). In particular,every Thom space Th X ( E ) is compact in H G • ( S ), being a pushout of compact objects, and it follows thatΩ E = Hom(S E , − ) preserves filtered colimits. By (1), this implies that Ω ∞ preserves filtered colimits andhence that Σ ∞ preserves compact objects.(4) This follows from (2) and (3). (cid:3) It follows from Lemma 6.2 that, if f : X → S is a smooth G -morphism, the H G • ( S )-module adjunction f (cid:93) : H G • ( X ) (cid:29) H G • ( S ) : f ∗ induces by base change along H G • ( S ) → SH G ( S ) an SH G ( S )-module adjunction f (cid:93) : SH G ( X ) (cid:29) SH G ( S ) : f ∗ . Similarly, if i : Z (cid:44) → S is a closed G -immersion, the H G • ( S )-module adjunction i ∗ : H G • ( S ) (cid:29) H G • ( Z ) : i ∗ induces by base change along H G • ( S ) → SH G ( S ) an SH G ( S )-module adjunction i ∗ : SH G ( S ) (cid:29) SH G ( Z ) : i ∗ . In particular, i ∗ preserves colimits and we also have an adjunction i ∗ : SH G ( Z ) (cid:29) SH G ( S ) : i ! . In summary, the functors f (cid:93) and i ∗ , for f smooth and i a closed immersion, extend to SH G ( − ) togetherwith the smooth and closed projection formulas. Moreover, the following properties are immediately deducedfrom their unstable (pointed) analogs using the functoriality of base change along Σ ∞ : smooth base change,closed base change, smooth–closed base change, and the full faithfulness of f ∗ , j ∗ , j (cid:93) , and i ∗ for f a G -affinebundle, j an open G -immersion, and i a closed G -immersion. The gluing and purity theorems follow easilyfrom their unstable versions and Proposition 6.4 (2), and the Nisnevich separation property from the factthat there is, for every S , a conservative family of functors SH G ( S ) → H G • ( S ) that commute with smoothbase change. Thus, all the functoriality discussed so far for H G • ( − ) extends to SH G ( − ). What is perhapsless obvious is that arbitrary spheres are invertible in SH G ( S ): Proposition 6.5.
Let p : X → S be a smooth G -morphism with a closed G -section s . Then the adjunction p (cid:93) s ∗ : SH G ( S ) (cid:29) SH G ( S ) : s ! p ∗ is an equivalence of ∞ -categories.Proof. By purity and the projection formulas, the left adjoint can be identified with S E ⊗ ( − ) where E is theconormal sheaf of s . Thus, the proposition is equivalent to the statement that S E is invertible in SH G ( S ). ByProposition 4.6 and the fact that smooth base change is closed symmetric monoidal, we may assume that S is small and affine. Let r : S → B be the structure map. We claim that there exists a locally free G -module F on B and an epimorphism r ∗ ( F ) (cid:16) E . Assuming this claim, we have S r ∗ ( F ) (cid:39) S E ⊗ S G in H G • ( S ), where G is the kernel of r ∗ ( F ) (cid:16) E , and since S r ∗ ( F ) is invertible in SH G ( S ), so is S E .It therefore suffices to establish the claim. If G is finite, we can find an epimorphism r ∗ ( O nG ) (cid:16) E , where O G is the regular representation of G over B . If G is not finite, then by assumption B has the G -resolutionproperty. As r : S → B is quasi-affine, r ∗ r ∗ ( E ) → E is an epimorphism. By Lemma 2.10, r ∗ ( E ) is the unionof its finitely generated quasi-coherent G -submodules. Hence, there exists a finitely generated quasi-coherent G -module M on B and an epimorphism r ∗ ( M ) (cid:16) E . By the G -resolution property, there exists a locallyfree G -module F on B and an epimorphism F (cid:16) M . We therefore obtain an epimorphism r ∗ ( F ) (cid:16) E , asdesired. (cid:3) Remark . The use of purity in the proof of Proposition 6.5 is not essential: after reducing to the casewhere S is small and affine and p is G -quasi-projective, one can obtain an equivalence X/ ( X (cid:114) S ) (cid:39) S E directly from the results of § Corollary 6.7.
Let p : S → B be a G -morphism. Then the canonical functor SH G ( S ) = H G • ( S )[ p ∗ (Sph B ) − ] → H G • ( S )[Sph − S ] is an equivalence of symmetric monoidal ∞ -categories.Remark . If G is finite locally free, we also haveSH G ( S ) (cid:39) H G • ( S )[(S p ∗ ( O G ) ) − ] , where O G is the regular representation of G . Indeed, the proof of Proposition 6.5 shows that any sphere S E becomes invertible in the right-hand side.Let f : X → S be a smooth separated G -morphism. As in § (cid:15) : f ∗ f (cid:93) Σ − Ω f → id : SH G ( X ) → SH G ( X )by the composition (5.20). By adjunction, we obtain a natural transformation f (cid:93) Σ − Ω f → f ∗ . Theorem 6.9 (Ambidexterity) . Let f : X → S be a smooth proper G -morphism. Then the transformation f (cid:93) Σ − Ω f → f ∗ : SH G ( X ) → SH G ( S ) is an equivalence.Proof. Any proper morphism in Sch GB is G -projective Nisnevich-locally on B . By Nisnevich separation andsmooth base change, we can therefore assume that f is smooth and G -projective. In that case, Theorem 5.22shows that the transformation (cid:15) : f ∗ f (cid:93) Σ − Ω f → id is the counit of an adjunction f ∗ (cid:97) f (cid:93) Σ − Ω f (between thehomotopy 1-categories), whence the result. (cid:3) Corollary 6.10 (Proper base change) . Let Y (cid:48) YX (cid:48) X gq f p be a cartesian square of G -schemes where p is proper. Then the exchange transformationEx ∗∗ : f ∗ p ∗ → q ∗ g ∗ : SH G ( Y ) → SH G ( X (cid:48) ) is an equivalence.Proof. Any proper morphism in Sch GB is G -projective Nisnevich-locally on B . By Nisnevich separation andsmooth base change, we can assume that p is a closed G -immersion or that p is smooth and G -projective. Inthe former case, the result holds by closed base change. In the latter case, we note that the square f ∗ p (cid:93) Σ − Ω p f ∗ p ∗ q (cid:93) Σ − Ω q g ∗ q ∗ g ∗ Ex ∗ (cid:93) Ex ∗∗ is commutative for formal reasons (see the proof of [CD12, Lemma 2.4.23 (1)]). Hence, the desired resultfollows from Theorem 6.9 and smooth base change. (cid:3) Corollary 6.11 (Proper projection formula) . Let p : Y → X be a proper G -morphism. For every A ∈ SH G • ( X ) and B ∈ SH G • ( Y ) , the canonical map A ⊗ p ∗ B → p ∗ ( p ∗ A ⊗ B ) is an equivalence.Proof. This follows from the smooth and closed projection formulas as in the proof of Corollary 6.10. (cid:3)
HE SIX OPERATIONS IN EQUIVARIANT MOTIVIC HOMOTOPY THEORY 49
Corollary 6.12 (Smooth–proper base change) . Let Y (cid:48) YX (cid:48) X gq f p be a cartesian square of G -schemes where p is proper and f is smooth. Then the exchange transformationEx (cid:93) ∗ : f (cid:93) q ∗ → p ∗ g (cid:93) : SH G ( Y (cid:48) ) → SH G ( X ) is an equivalence.Proof. This follows from smooth–closed base change as in the proof of Corollary 6.10. (cid:3)
Corollary 6.13 (Atiyah duality) . Let f : X → S be a smooth proper G -morphism. Then Σ ∞ X + is stronglydual to f (cid:93) S − Ω f in SH G ( S ) .Proof. Recall that A ∈ SH G ( S ) is strongly dualizable if and only if, for every B ∈ SH G ( S ), the canon-ical map Hom( A, S ) ⊗ B → Hom(
A, B ) is an equivalence. By the smooth projection formula, we haveHom( f (cid:93) X , B ) (cid:39) f ∗ f ∗ B , and the canonical map f ∗ X ⊗ B → f ∗ f ∗ B is an equivalence by the proper projec-tion formula. Thus, f (cid:93) X is strongly dualizable, with dual f ∗ X . By Theorem 6.9, f ∗ X (cid:39) f (cid:93) S − Ω f . (cid:3) The exceptional functors.
We now show how the results of the previous sections give rise to theformalism of six operations for SH G ( − ). It is not difficult to define a functor(6.14) SH G : (Sch GB ) op → CAlg( P r L , ⊗ )that sends a G -scheme S to SH G ( S ) and a G -morphism f : T → S to f ∗ : SH G ( S ) → SH G ( T ); we refer to[Rob14, § f ! at the level of ∞ -categories, we will use the technologydeveloped in [GR16, Chapter V.1]. A G -morphism f : Y → X in Sch GB is called compactifiable if there existsa proper G -scheme P in Sch GB such that f is the composition of a G -immersion Y (cid:44) → P × B X and theprojection P × B X → X . It is easy to show that the composition of two compactifiable G -morphisms iscompactifiable, so that compactifiable G -morphisms define a wide subcategory (Sch GB ) comp of Sch GB . Notealso that if g ◦ f is compactifiable, then f is compactifiable; in particular, any G -morphism between G -quasi-projective B -schemes is compactifiable. By Lemma 2.2 (2), any compactifiable G -morphism f canbe written as p ◦ j where j is an open G -immersion and p is a proper compactifiable G -morphism; sucha factorization is called a compactification of f . It is then clear that the category (Sch GB ) comp , equippedwith its wide subcategories of open immersions and proper morphisms, satisfies the assumptions of [AGV73,Expos´e XVII, 3.2.4]. By [AGV73, Expos´e XVII, Proposition 3.2.6 (ii)], the category of compactifications ofany compactifiable G -morphism is cofiltered, and in particular weakly contractible.Given S ∈ Sch GB , let Corr(Sch GS ) propall , comp denote the 2-category whose 1-morphisms are spans X ← Y → Z in Sch GS with Y → Z compactifiable and whose 2-morphisms are proper G -morphisms between spans [GR16,Chapter V.1, § GS gives a functor(6.15) (Sch GS ) op → Mod SH G ( S ) . We claim that this functor extends uniquely † to an ( ∞ , GS ) prop , , comp → Mod SH G ( S ) satisfying the following condition:( ∗ ) Let U (cid:44) → X be an open G -immersion in Sch GS . Then the canonical 2-isomorphism between id U andthe composition of the two spans U = U (cid:44) → X and X ← (cid:45) U = U becomes the unit of an adjunctionin Mod SH G ( S ) . † Uniqueness means that the ∞ -groupoid of such extensions is contractible. Here, “2-op” means that we reverse the direction of the 2-morphisms. We first note that (6.15) sends smooth(resp. proper) G -morphisms to left (resp. right) adjointable morphisms in Mod SH G ( S ) , by the smooth pro-jection formula (resp. by the proper projection formula and Proposition 6.4 (4)). Smooth base change foropen immersions allows us to apply [GR16, Chapter V.1, Theorem 3.2.2]: the functor (6.15) admits a uniqueextension to the 2-category Corr(Sch GS ) openall , open , whose 1-morphisms are spans X ← Y → Z with Y → Z an open G -immersion. Its restriction to the (2 , GS ) all , open satisfies condition ( ∗ ), and by[GR16, Chapter V.1, Theorem 4.1.3] it is the unique extension of (6.15) with this property. We now apply[GR16, Chapter V.1, Theorem 5.2.4] with proper compactifiable morphisms as admissible morphisms andopen immersions as co-admissible morphisms: this is justified by the weak contractibility of the categoriesof compactifications, proper base change (Corollary 6.10), and smooth–proper base change for open immer-sions (Corollary 6.12). As a result, there is a unique further extension of (6.15) from Corr(Sch GS ) all , open toCorr(Sch GS ) prop , , comp , as claimed.Let us unpack some of the data encoded by (6.16). Given a compactifiable G -morphism f : Y → X inSch GB , we denote by f ! : SH G ( Y ) → SH G ( X )the image of the span Y id ← Y f → X by (6.16), with S = B . Being a morphism in P r L , f ! admits a rightadjoint f ! . The functors f ! and f ! are called the exceptional functors . If f is proper, there is an adjunctionof spans ( X f ← Y id → Y ) (cid:97) ( Y id ← Y f → X )in Corr(Sch GB ) prop , , comp , so that f ! (cid:39) f ∗ . On the other hand, if f is an open immersion, condition ( ∗ ) impliesthat f ! (cid:39) f (cid:93) . Thus, in general, we have an equivalence f ! (cid:39) p ∗ j (cid:93) for any factorization f = pj with j an open G -immersion and p a proper compactifiable G -morphism.A compactifiable G -morphism f : Y → X may also be viewed as a morphism in Sch GX . By uniqueness ofthe extensions (6.16), there is a commutative squareCorr(Sch GX ) all , comp Mod SH G ( X ) Corr(Sch GB ) all , comp Mod SH G ( B ) ,which shows that f ! can be promoted to an SH G ( X )-module functor. In particular, there is a canonicalequivalence f ! ( − ⊗ f ∗ ( − )) (cid:39) f ! ( − ) ⊗ − . By construction, this equivalence is the smooth projection formula if f is an open immersion and the properprojection formula if f is proper.Given a cartesian square of G -schemes(6.17) Y (cid:48) YX (cid:48) X gq f p with p compactifiable, the functor (6.16) gives a canonical equivalence Ex ∗ ! : f ∗ p ! (cid:39) q ! g ∗ : SH G ( Y ) → SH G ( X (cid:48) ) . By construction, Ex ∗ ! can be identified with the exchange equivalence Ex ∗ (cid:93) if p is an open immersion andwith the exchange equivalence Ex ∗∗ if p is proper. By adjunction, there is also a natural equivalence Ex ! ∗ : p ! f ∗ (cid:39) g ∗ q ! : SH G ( X (cid:48) ) → SH G ( Y ) . Given the cartesian square (6.17) with p compactifiable, we can define a natural transformation Ex ! ∗ : p ! g ∗ → f ∗ q ! : SH G ( Y (cid:48) ) → SH G ( X ) HE SIX OPERATIONS IN EQUIVARIANT MOTIVIC HOMOTOPY THEORY 51 by the composition p ! g ∗ η → f ∗ f ∗ p ! g ∗ Ex ∗ ! (cid:39) f ∗ q ! g ∗ g ∗ (cid:15) → f ∗ q ! . It is an equivalence if f is proper: this is obvious if p is proper, and it follows from Corollary 6.12 if p is anopen immersion. Similarly, we can define a natural transformation Ex ∗ ! : g ∗ p ! → q ! f ∗ : SH G ( X ) → SH G ( Y (cid:48) )by the composition g ∗ p ! η → g ∗ p ! f ∗ f ∗ Ex ! ∗ (cid:39) g ∗ g ∗ q ! f ∗ (cid:15) → q ! f ∗ . It is an equivalence if f is smooth: this is obvious if p is an open immersion, and it follows from Corollary 6.12if p is proper.Finally, given f : X → S compactifiable, we will define an endofunctor Tw f : SH G ( X ) → SH G ( X ) andnatural transformations f ! → f ∗ , Tw f ◦ f ! → f ∗ . Consider the cartesian square of G -schemes X × S X XX S , π π f f and let δ : X → X × S X be the diagonal. Since compactifiable morphisms are separated, δ is proper andhence δ ∗ = δ ! . Then the transformation f ! → f ∗ is the composition f ! (cid:39) f ! π ∗ δ ∗ Ex ! ∗ −−−→ f ∗ π δ ∗ (cid:39) f ∗ . It is an equivalence if f is proper, since Ex ! ∗ is. We set Tw f = δ ! π ∗ . The natural transformation Tw f ◦ f ! → f ∗ is then the composition δ ! π ∗ f ! Ex ∗ ! −−−→ δ ! π !1 f ∗ (cid:39) f ∗ . It is an equivalence if f is smooth, since Ex ∗ ! is. In that case, Tw f itself is an equivalence of ∞ -categories,by Proposition 6.5, and in fact it is canonically equivalent to Σ − Ω f , by purity (Proposition 5.7).The following theorem summarizes the properties of the six operations established so far: Theorem 6.18.
The six operations ( − ) ∗ , ( − ) ∗ , ( − ) ! , ( − ) ! , ⊗ , Hom satisfy the following properties, whenever the exceptional functors are defined. (1) (Proper pushforward) If f is a proper G -morphism, there is a canonical equivalence f ! (cid:39) f ∗ . (2) (Smooth pullback) If f is a smooth G -morphism, there is a canonical equivalence Σ − Ω f ◦ f ! (cid:39) f ∗ . (3) (Base change) If • •• • gq f p is a cartesian square of G -schemes, there are canonical equivalences f ∗ p ! (cid:39) q ! g ∗ ,f ! p ∗ (cid:39) q ∗ g ! . (4) (Gluing) If i is a closed G -immersion with complementary open G -immersion j , there are cofibersequences j ! j ! → id → i ∗ i ∗ ,i ! i ! → id → j ∗ j ∗ . (5) (Immersive pushforward) If i is a G -immersion, the functors i ∗ and i ! are fully faithful. (6) (Monoidality) If f is any G -morphism, there is a canonical equivalence f ∗ ( − ⊗ − ) (cid:39) f ∗ ( − ) ⊗ f ∗ ( − ) . (7) (Projection formulas) If f is any G -morphism, there are canonical equivalences f ! ( − ⊗ f ∗ ( − )) (cid:39) f ! ( − ) ⊗ − , Hom( f ! ( − ) , − ) (cid:39) f ∗ Hom( − , f ! ( − )) ,f ! Hom( − , − ) (cid:39) Hom( f ∗ ( − ) , f ! ( − )) . (8) (Homotopy invariance) If f is a G -affine bundle, the functors f ∗ and f ! are fully faithful. Corollary 6.19.
Let f : Y → X be a compactifiable G -morphism. Then the functor f ! : SH G ( X ) → SH G ( Y ) preserves colimits.Proof. By Nisnevich separation and Theorem 6.18 (2), the question is Nisnevich-local on B . Hence, we canassume that f is G -quasi-projective. By Lemma 2.2 (1), we can further assume that f is smooth or a closed G -immersion. The result then follows from Theorem 6.18 (2) and (4), respectively. (cid:3) Remark . As we explained in § { schemes } op → { symmetric monoidal ∞ -categories } to algebraic stacks. It is easy to see from thedefinitions (and Corollary 6.7) that the symmetric monoidal ∞ -categories H G ( S ), H G • ( S ), and SH G ( S ) areindeed intrinsic invariants of the quotient stack [ S/G ]. Moreover, it is clear that the adjunction f ∗ (cid:97) f ∗ , with f ∗ symmetric monoidal, exists for an arbitrary morphism of stacks f . For example, if f : B G → B is theunique map, f ∗ : SH( B G ) → SH( B ) is the motivic analog of the “genuine G -fixed points” functor in stableequivariant homotopy theory. From this perspective, the base change property of Theorem 6.18 admits thefollowing generalization (with the same proof). Given any cartesian square of stacks Y (cid:48) YX (cid:48) X gq f p for which SH( − ), p ! , and q ! are defined, there is a canonical equivalence f ∗ p ! (cid:39) q ! g ∗ . Remark . Suppose that the tame group G is finite and discrete. In that case, SH G ( S ) can be defined forarbitrary qcqs G -schemes S , using for Sm GS the category of all finitely presented smooth G -schemes over S (see Remark 3.10) and inverting the regular representation sphere (see Remark 6.8). The proper base changetheorem and related results then hold for arbitrary proper G -morphisms. Following the proof of [Hoy14,Proposition C.13], we only need a suitable equivariant version of Chow’s lemma [AGV73, XII, § G -morphismof finite type f : X → S and a projective morphism π : X (cid:48) → X such that f ◦ π is quasi-projective and π − ( U ) (cid:39) U for some nonempty open subset U ⊂ X . Replacing U by a maximal nonempty intersection ofits G -translates, we can assume that V = (cid:96) g ∈ G/H gU is an open subscheme of X , where H is the stabilizerof U . Replacing X (cid:48) by the fiber product of the | H | maps h ◦ π : X (cid:48) → X , h ∈ H , we can assume that π is H -equivariant. ‡ Let Z ⊂ X be an H -invariant closed complement to V (cid:114) U . Then one can replace X (cid:48) bythe induced G -scheme ( G × X (cid:48) Z ) /H and U by V .Similarly, the functor f ! can be defined for any G -morphism f : X → S that is separated and of finite type,since Nagata’s compactification theorem immediately generalizes to the G -equivariant setting. Indeed, givena nonequivariant compactification f = p ◦ j where j : X (cid:44) → ¯ X , one obtains an equivariant one by replacing ‡ In terms of stacks, this fiber product is the Weil restriction of X (cid:48) along X → [ X/H ]. HE SIX OPERATIONS IN EQUIVARIANT MOTIVIC HOMOTOPY THEORY 53 ¯ X by the schematic closure of X in the fiber product of the | G | maps g ◦ p : ¯ X → S , g ∈ G (the schematicclosure will be G -invariant by Lemma 2.2 (2)).In summary, if G is a finite discrete group of order d , Theorem 6.18 applies to all qcqs G -schemes overSpec Z [1 /d ], the exceptional adjunction f ! (cid:97) f ! being defined for f separated and of finite type.6.3. Descent properties.
We investigate the descent properties of the functor S (cid:55)→ SH G ( S ). We begin byintroducing the equivariant analogs of the cdh topology and the constructible topology.Let X be a G -scheme. An abstract blowup square over X is a cartesian square(6.22) W YZ X i p of G -schemes where i is a closed G -immersion, p is proper, and p induces an isomorphism Y × X ( X (cid:114) Z ) (cid:39) X (cid:114) Z . The cdh topology on Sch GB is the coarsest topology finer than the Nisnevich topology and such that,for every abstract blowup square (6.22), { i, p } generates a covering sieve of X . The same argument as in theproof of Proposition 3.8 shows that a presheaf F on Sch GB is a sheaf for the cdh topology if and only if F ( ∅ )is contractible and F sends Nisnevich squares and abstract blowup squares to cartesian squares.The constructible topology on Sch GB is the coarsest topology such that: • the empty sieve covers the empty scheme; • if Z (cid:44) → X is a closed G -immersion with open complement U (cid:44) → X , { U (cid:44) → X, Z (cid:44) → X } generates acovering sieve.Note that the constructible topology is finer than the cdh topology. Proposition 6.23.
Let { f i : U i → S } be a constructible cover of a G -scheme S . Then the families offunctors { f ∗ i : SH G ( S ) → SH G ( U i ) } , { f ! i : SH G ( S ) → SH G ( U i ) } are conservative (assuming each f i compactifiable in the latter case).Proof. This is an immediate consequence of gluing. (cid:3)
Proposition 6.24.
The functors (Sch GB ) op → C at ∞ , S (cid:55)→ SH G ( S ) , f (cid:55)→ f ∗ , (Sch GB ) opcomp → C at ∞ , S (cid:55)→ SH G ( S ) , f (cid:55)→ f ! , are sheaves for the cdh topology.Proof. We give the proof for the first functor. We must show that:(a) SH G ( ∅ ) (cid:39) ∗ ;(b) SH G ( − ) takes Nisnevich squares to cartesian squares;(c) SH G ( − ) takes abstract blowup squares to cartesian squares.Assertion (a) is obvious. We will prove (c) and omit the proof of (b) which is entirely similar. § Let Q be theabstract blowup square W YZ X . kq i p By [Lur11, Lemma 5.17], SH G ( Q ) is cartesian if and only if:(d) the pair ( i ∗ , p ∗ ) is conservative; § Assertion (b) also follows from Proposition 4.8 (2), but this alternative argument is not available for the second functor. (e) given E Z ∈ SH G ( Z ), E Y ∈ SH G ( Y ), E W ∈ SH G ( W ), and q ∗ E Z (cid:39) E W (cid:39) k ∗ E Y , if E X = i ∗ E Z × ( pk ) ∗ E W p ∗ E Y , then the maps i ∗ E X → E Z and p ∗ E X → E Y induced by the canonical projections are equivalences.Assertion (d) follows from Proposition 6.23, since { i, p } is a constructible cover of X . Let us prove (e).Proper base change and the full faithfulness of i ∗ immediately imply that i ∗ E X → E Z is an equivalence. Toshow that p ∗ E X → E Y is an equivalence, it suffices to show that k ∗ p ∗ E X → k ∗ E Y and j ∗ p ∗ E X → j ∗ E Y areequivalences, where j is the open G -immersion complementary to k . The former is q ∗ of the map just shownto be an equivalence, and the latter is an equivalence by smooth base change. (cid:3) Corollary 6.25.
Let S be a G -scheme and let E ∈ SH G ( S ) . Then the functor (Sch GS ) op → SH G ( S ) , ( f : X → S ) (cid:55)→ f ∗ f ∗ E, is a sheaf for the cdh topology, and the functor (Sch GB ) comp /S → SH G ( S ) , ( f : X → S ) (cid:55)→ f ! f ! E, is a cosheaf for the cdh topology. In particular, taking mapping spaces or mapping spectra, we deduce that any space-valued or spectrum-valued invariant of G -schemes represented by a cartesian section of SH G ( − ) satisfies cdh descent. Remark . If G is finite and discrete, the results of this subsection hold if we replace Sch GB by the categoryof all qcqs G -schemes over B and (Sch GB ) comp by the wide subcategory of separated G -morphisms of finitetype (see Remark 6.21). References [AGV73] M. Artin, A. Grothendieck, and J.-L. Verdier,
Th´eorie des Topos et Cohomologie ´Etale des Sch´emas, Tome 3 , LectureNotes in Mathematics, vol. 305, Springer, 1973[AHW15] A. Asok, M. Hoyois, and M. Wendt.,
Affine representability results in A -homotopy theory I: vector bundles , 2015, arXiv:1506.07093v2 [math.AG] [AOV08] D. Abramovich, M. Olsson, and A. Vistoli, Tame stacks in positive characteristic , Ann. I. Fourier (2008), no. 4,pp. 1057–1091[Ara01] A. Arabia, Rel`evements des alg`ebres lisses et de leurs morphismes , Comment. Math. Helv. (2001), pp. 607–639[Ayo08] J. Ayoub, Les six op´erations de Grothendieck et le formalisme des cycles ´evanescents dans le monde motivique, I ,Ast´erisque (2008)[CD12] D.-C. Cisinski and F. D´eglise,
Triangulated categories of mixed motives , 2012, arXiv:0912.2110v3 [math.AG] [CJ14] G. Carlsson and R. Joshua,
Equivariant Motivic Homotopy Theory , 2014, arXiv:1404.1597v1 [math.AT] [Cis13] D.-C. Cisinski,
Descente par ´eclatements en K -th´eorie invariante par homotopie , Ann. Math. (2013), no. 2,pp. 425–448[DG70] M. Demazure and A. Grothendieck, Propri´et´es g´en´erales des sch´emas en groupes , Lecture Notes in Mathematics, vol.151, Springer, 1970[Del09] P. Deligne,
Voevodsky’s Lectures on Motivic Cohomology 2000/2001 , Algebraic Topology, Abel Symp., vol. 4, 2009[Ful98] W. Fulton,
Intersection Theory , Springer, 1998[GK15] D. Gepner and J. Kock,
Univalence in locally cartesian closed ∞ -categories , 2015, arXiv:1208.1749v4 [math.CT] [GR16] D. Gaitsgory and N. Rozenblyum, A study in derived algebraic geometry , book project, 2016, [Gro61] A. Grothendieck, ´El´ements de G´eom´etrie Alg´ebrique: II. ´Etude globale ´el´ementaire de quelques classes de mor-phismes , Publ. Math. I.H.´E.S. (1961)[Gro64] , ´El´ements de G´eom´etrie Alg´ebrique: IV. ´Etude locale des sch´emas et des morphismes de sch´emas, Premi`erepartie , Publ. Math. I.H.´E.S. (1964)[Gro66] , ´El´ements de G´eom´etrie Alg´ebrique: IV. ´Etude locale des sch´emas et des morphismes de sch´emas, Troisi`emepartie , Publ. Math. I.H.´E.S. (1966)[Gro67] , ´El´ements de G´eom´etrie Alg´ebrique: IV. ´Etude locale des sch´emas et des morphismes de sch´emas, Quatri`emepartie , Publ. Math. I.H.´E.S. (1967)[Gro15] P. Gross, Tensor generators on schemes and stacks , 2015, arXiv:1306.5418v2 [math.AG] [HKO11] P. Hu, I. Kriz, and K. Ormsby,
The homotopy limit problem for Hermitian K-theory, equivariant motivic homotopytheory and motivic Real cobordism , Adv. Math. (2011), no. 1, pp. 434–480[HKØ15] J. Heller, A. Krishna, and P. A. Østvær,
Motivic homotopy theory of group scheme actions , J. Topol. (2015), no. 4,pp. 1202–1236 HE SIX OPERATIONS IN EQUIVARIANT MOTIVIC HOMOTOPY THEORY 55 [HVØ15] J. Heller, M. Voineagu, and P. A. Østvær,
Equivariant cycles and cancellation for motivic cohomology , Doc. Math. (2015), pp. 269–332[Hel16] J. Heller, Endomorphisms of the equivariant motivic sphere , Oberwolfach Rep. (2016), no. 1[Her12] P. Herrmann, Stable Equivariant Motivic Homotopy Theory and Motivic Borel Cohomology , Ph.D. Thesis, 2012[Her13] ,
Equivariant Motivic Homotopy Theory , 2013, arXiv:1312.0241v1 [math.AT] [Hoy14] M. Hoyois,
A quadratic refinement of the Grothendieck–Lefschetz–Verdier formula , Alg. Geom. Top. (2014), no. 6,pp. 3603–3658[Hoy17] , Cdh descent in equivariant homotopy K -theory , 2017, arXiv:1604.06410 [math.KT] [Hu03] P. Hu, Duality for Smooth Families in Equivariant Stable Homotopy Theory , SMF, 2003[KØ12] A. Krishna and P. A. Østvær,
Nisnevich descent for K -theory of Deligne–Mumford stacks , J. K-theory (2012),no. 2, pp. 291–331[LZ14] Y. Liu and W. Zheng, Enhanced six operations and base change theorem for Artin stacks , 2014, arXiv:1211.5948v2[math.AG] [Lur09] J. Lurie,
Higher Topos Theory , Annals of Mathematical Studies, vol. 170, Princeton University Press, 2009[Lur11] ,
Derived Algebraic Geometry VII: Spectral Schemes , 2011, [Lur16] ,
Higher Algebra , 2016, [MS06] J. P. May and J. Sigurdsson,
Parametrized Homotopy Theory , Mathematical Surveys and Monographs, vol. 132,AMS, 2006[MV99] F. Morel and V. Voevodsky, A -homotopy theory of schemes , Publ. Math. I.H.´E.S. (1999), pp. 45–143[May03] J. P. May, The Wirthm¨uller isomorphism revisited , Theory Appl. Categ. (2003), no. 5, pp. 132–142[NS15] T. Nikolaus and S. Sagave, Presentably symmetric monoidal ∞ -categories are represented by symmetric monoidalmodel categories , 2015, arXiv:1506.01475v2 [math.AT] [Nag61] M. Nagata, Complete reducibility of rational representations of a matric group , J. Math. Kyoto Univ. (1961),pp. 87–99[Rob14] M. Robalo, Th´eorie Homotopique Motivique des Espaces Noncommutatifs , Ph.D. Thesis, 2014, http://webusers.imj-prg.fr/~marco.robalo/these.pdf [Rob15] , K -theory and the bridge from motives to noncommutative motives , Adv. Math. (2015), pp. 399–550[Sch16] S. Schwede, Global homotopy theory , 2016, [Stacks] The Stacks Project Authors,
Stacks Project , 2016, http://stacks.math.columbia.edu [Sum75] H. Sumihiro,
Equivariant completion II , J. Math. Kyoto Univ. (1975), no. 3, pp. 573–605[TT90] R. W. Thomason and T. Trobaugh, Higher algebraic K-theory of schemes and of derived categories , The GrothendieckFestschrift III, Progress in Mathematics, vol. 88, Birkh¨auser, 1990, pp. 247–435[Tho87a] R. W. Thomason,
Algebraic K-theory of group scheme actions , Algebraic Topology and Algebraic K-theory (W. Brow-der, ed.), Annals of Mathematical Studies, vol. 113, Princeton University Press, 1987[Tho87b] ,
Equivariant Resolution, Linearization, and Hilbert’s Fourteenth Problem over Arbitrary Base Schemes , Adv.Math. (1987), pp. 16–34[Voe98] V. Voevodsky, A -Homotopy Theory , Doc. Math., Extra Volume: Proceedings of the International Congress ofMathematicians, Volume I (1998), pp. 579–604[Voe03] , Motivic cohomology with Z / -coefficients , Publ. Math. I.H.´E.S. (2003), no. 1, pp. 59–104[Voe10] , Motivic Eilenberg–MacLane spaces , Publ. Math. I.H.´E.S. (2010), pp. 1–99[Voe11] ,
On motivic cohomology with Z /l -coefficients , Ann. Math. (2011), no. 1, pp. 401–438[Wei89] C. A. Weibel, Homotopy algebraic K -theory , Algebraic K-Theory and Number Theory, Contemp. Math., vol. 83,AMS, 1989, pp. 461–488 Department of Mathematics, Massachusetts Institute of Technology, Cambridge, MA, USA
E-mail address : [email protected] URL ::