Triangulated categories of mixed motives
aa r X i v : . [ m a t h . AG ] D ec Triangulated categories of mixed motives
Denis-Charles CisinskiFr´ed´eric D´eglise
Universit´e Paul Sabatier, Institut de Math´ematiques de Toulouse, 118 routede Narbonne, 31062 Toulouse Cedex 9, France
E-mail address : [email protected] URL : E.N.S. Lyon - UMPA, 46 all´ee d’Italie, 69364 Lyon Cedex 07, France
E-mail address : [email protected] URL : http://perso.ens-lyon.fr/frederic.deglise/ artially supported by the ANR (grant No. ANR-07-BLAN-042). Abstract.
We construct triangulated categories of mixed motives over a noetherian schemeof finite dimension, extending Voevodsky’s definition of motives over a field. We prove thatmotives with rational coefficients satisfy the formalism of the six operations of Grothendieck.This is achieved by studying descent properties of motives, as well as by comparing differentpresentations of these categories, following insights and constructions of Beilinson, Morel andVoevodsky. Finally, we associate with any mixed Weil cohomology a system of categories ofcoefficients and well behaved realization functors. ontents
Introduction viiA. Historical background viiiA.1. The conjectural theory described by Beilinson viiiA.2. Voevodsky’s motivic complexes viiiA.3. Morel and Voevodsky homotopy theory viiiA.4. Cross functors ixA.5. Grothendieck 6 functors formalism ixB. Voevodsky’s motivic complexes xiC. Beilinson motives xiiiC.1. Definition and fundamental properties xiiiC.2. Constructible Beilinson motives xivC.3. Comparison theorems xivC.4. Realizations xviiD. Detailed organization xviiD.1. Grothendieck six functors formalism (Part 1) xviiiD.2. The constructive part (Part 2) xxiD.3. Motivic complexes (Part 3) xxiD.4. Beilinson motives (Part 4) xxiiiNotations and conventions xxivThanks xxiv
Part 1. Fibred categories and the six functors formalism
11. General definitions and axiomatic 21.1. P -fibred categories 21.1.a. Definitions 21.1.b. Monoidal structures 71.1.c. Geometric sections 91.1.d. Twists 101.2. Morphisms of P -fibred categories 111.2.a. General case 111.2.b. Monoidal case 131.3. Structures on P -fibred categories 141.3.a. Abstract definition 141.3.b. The abelian case 151.3.c. The triangulated case 161.3.d. The model category case 181.4. Premotivic categories 192. Triangulated P -fibred categories in algebraic geometry 212.1. Elementary properties 222.2. Exceptional functors, following Deligne 242.2.a. The support axiom 242.2.b. Exceptional direct image 262.2.c. Further properties 302.3. The localization property 31 iiiv CONTENTS P -fibred model categories 573.1. Extension of P -fibred categories to diagrams 573.1.a. The general case 573.1.b. The model category case 593.2. Hypercovers, descent, and derived global sections 683.3. Descent over schemes 763.3.a. Localization and Nisnevich descent 763.3.b. Proper base change isomorphism and descent by blow-ups 783.3.c. Proper descent with rational coefficients I: Galois excision 803.3.d. Proper descent with rational coefficients II: separation 874. Constructible motives 914.1. Resolution of singularities 914.2. Finiteness theorems 934.3. Continuity 1014.4. Duality 107 Part 2. Construction of fibred categories t -descent model category structure 1195.1.c. Constructible premotivic complexes 1265.2. The A -derived premotivic category 1305.2.a. Localization of triangulated premotivic categories 1305.2.b. The homotopy relation 1345.2.c. Explicit A -resolution 1375.2.d. Constructible A -local premotives 1405.3. The stable A -derived premotivic category 1415.3.a. Modules 1425.3.b. Symmetric sequences 1425.3.c. Symmetric Tate spectra 1445.3.d. Symmetric Tate Ω-spectra 1465.3.e. Constructible premotivic spectra 1506. Localization and the universal derived example 1526.1. Generalized derived premotivic categories 1526.2. The fundamental example 1556.3. Nearly Nisnevich sheaves 1566.3.a. Support property (effective case) 1566.3.b. Support property (stable case) 1586.3.c. Localization for smooth schemes 1597. Basic homotopy commutative algebra 1607.1. Rings 1607.2. Modules 164 Part 3. Motivic complexes and relative cycles
ONTENTS v
8. Relative cycles 1728.1. Definitions 1728.1.a. Category of cycles 1728.1.b. Hilbert cycles 1738.1.c. Specialization 1758.1.d. Pullback 1788.2. Intersection theoretic properties 1828.2.a. Commutativity 1828.2.b. Associativity 1838.2.c. Projection formulas 1848.3. Geometric properties 1858.3.a. Constructibility 1868.3.b. Samuel multiplicities 1889. Finite correspondences 1939.1. Definition and composition 1939.2. Monoidal structure 1979.3. Functoriality 1989.3.a. Base change 1989.3.b. Restriction 1999.3.c. A finiteness property 2009.4. The fibred category of correspondences 20010. Sheaves with transfers 20110.1. Presheaves with transfers 20110.2. Sheaves with transfers 20210.3. Associated sheaf with transfers 20310.4. Examples 20910.5. Comparison results 21010.5.a. Change of coefficients 21110.5.b. Representable qfh-sheaves 21110.5.c. qfh-sheaves and transfers 21111. Motivic complexes 21311.1. Definition and basic properties 21311.1.a. Premotivic categories 21311.1.b. Constructible and geometric motives 21511.1.c. Enlargement, descent and continuity 21611.2. Motivic cohomology 21811.2.a. Definition and functoriality 21811.2.b. Effective motivic cohomology in weight 0 and 1 22011.2.c. The motivic cohomology ring spectrum 22311.3. Orientation and purity 22411.4. The six functors 227
Part 4. Beilinson motives and algebraic K-theory i CONTENTS γ -filtration 24014.2. Definition 24114.3. Motivic proper descent 24514.4. Motivic absolute purity 24615. Constructible Beilinson motives 24715.1. Definition and basic properties 24715.2. Grothendieck 6 functors formalism and duality 24816. Comparison theorems 24916.1. Comparison with Voevodsky motives 24916.2. Comparison with Morel motives 25217. Realizations 25717.1. Tilting 25717.2. Mixed Weil cohomologies 260Bibliography 267Index of terminology 271Index of notations 277Appendix A. Index of properties of P -fibred triangulated categories 279 ntroduction . Historical backgroundA.1. The conjectural theory described by Beilinson. In a landmarking paper, [
Be˘ı87 ],A. Beilinson stated a series of conjectures which offer a complete renewal of the traditional theoryof pure motives invented by A. Grothendieck. Namely, he proposes to extend the notion of puremotives to that of mixed motives with two models in mind: mixed Hodge structures defined byP. Deligne one the one hand, perverse sheaves on the other hand defined in [
BBD82 ]. One ofthe main innovation, considered by Beilinson in analogy with the second model, is to consider atriangulated version of mixed motives in which one could hope to find the more involved theory ofabelian mixed motives through the concept of t-structures. This hoped for theory was conjecturallydescribed by Beilinson in [
Be˘ı87 , 5.10] under the name of motivic complexes .It was modeled (see loc. cit. , paragraph A) on the theory of ´etale l -torsion (resp. l -adic)sheaves and their derived category as introduced fifty years ago by Grothendieck and M. Artin.The major achievement of Grothendieck and his collaborators in [ SGA4 ] was to define a theoryof coefficients systems relative to any scheme with a collection of operations, f ∗ , f ∗ , f ! , f ! , ⊗ , Hom ,satisfying a set of formulas now called the
Grothendieck six functors formalism (see section A.5in this introduction for more details). This formalism, formulated in the language of triangulatedcategories, ultimately encode a very general duality theory. Note however that the complete dualitytheory for l -torsion ´etale sheaves was completed only recently by the work of Gabber [ ILO ].The theory was also conjectured to be deeply linked with Quillen algebraic K-theory (see[
Be˘ı87 , 5.10, § B]). In fact, up to torsion and for a regular scheme S , the ext-groups between twoTate motives over S should coincide with Adams graded parts of Quillen algebraic K-theory. The ideas of Beilinson were very fecund because, not long after the publication of [
Be˘ı87 ], onehad three candidates for a triangulated category of mixed motives, respectively by M. Hanamura,M. Levine, and V. Voevodsky. In this book, we will focus on Voevodsky’s theory.
A.2. Voevodsky’s motivic complexes.
The first attempt of Voevodsky in defining thecategory of motivic complexes, in his 1992 Harvard’s thesis, introduces the fundamental processof A -localization, which amounts to make the affine line contractible in the category of mixedmotives, by analogy with the topological case. It also involves the use of the h -topology whichwas to become fundamental in the area of motives and cohomology. These two ingredients given,Voevodsky defined the triangulated category of (effective) h -motives over any base in [ Voe96 ].However, Voevodsky was aware that his definition will give the correct answer to Beilinson’sconjectural construction only with rational coefficients. In [
VSF00 , chap. 5], he introducesanother definition of motivic complexes over a perfect field with integral coefficients, still usingthe A -localization process but this time introducing the notion of Nisnevich sheaves with transfersand their derived category (see [ MVW06 ] for a detailed exposition). At the time being all theproperties foreseen by Beilinson are established for this integral category over a perfect field, exceptfor the construction of the motivic t -structure. It remains to extend this definition to arbitrarybases and established the Grothendieck six functors formalism.The path in this direction was laid down by Voevodsky in [
Voe10a ] were he uses the theoryof relative cycles invented by Suslin and Voevodsky to extend the definition of transfers. Thisdefinition was also exploited by Ivorra in [
Ivo07 ] to extend the definition of geometric motiviccomplexes of Voevodsky over any base, avoiding the use of sheaves with transfers. Still it entirelyremained to construct Grothendieck six functors formalism for this definition.
A.3. Morel and Voevodsky homotopy theory.
Soon after the introduction of Voevod-sky’s motivic complexes, F. Morel and Voevodsky introduced the more general theory of A -homotopy of schemes ([ MV99 ]) whose design is to extend the framework of algebraic topologyto algebraic geometry and built around the A -localization tool. It is within this theory that wasinvented another important tool in the motivic homotopy theory, the P -stabilization process.From the purely motivic point of view, this amounts to invert the Tate motive Z (1) for the tensorproduct. From the homotopical point of view, this operation is much more involved and reveals See below for the precise statement. This hoped for t -structure is described in [ Voe92 , Hyp. 0.0.21]. . HISTORICAL BACKGROUND ix the theory of spectra, objects which incarnate cohomology theories in algebraic topology. Thesetwo processes, of A -localization and P -stabilization, applied to the category of simplicial Nis-nevich sheaves, led to the stable A -homotopy category of schemes (see [ Jar00 ]) a triangulatedcategory with integral coefficients, defined over any base, which generalizes the category of motiviccomplexes. Over a perfect field, and with rational coefficients, the relation between homotopy and motiveswas clarified in an unpublished paper of Morel ([
Mor06 ]): the rational stable A -homotopycategory contains the stable ( i.e. P -stable) version of the category of motivic complexes as anexplicit direct factor, called the +-part of the stable homotopy category. Then Morel introducesthis +-part as a good candidate for the rational version of the triangulated category of motives([
Mor06 , paragraph at the end of p.2]). We will dubbed the objects of this category the
Morelmotives .On the other hand, with integral coefficients, O. R¨ondigs et P.A. Østvær showed that over afield of characteristic 0, the P -stable category of motivic complexes coincides with the categoryof modules over the ring spectrum which represents motivic cohomology (see [ RØ08 ]). This ringspectrum was introduced by Voevodsky (see [
Voe98 ]) using the theory of relative cycles. It isdefined over any base and one is led to consider the category of modules over this ring spectrumas a possible definition of the integral triangulated category of motives. A.4. Cross functors.
The definitive step towards the six functors formalism in motivichomotopy theory was taken up by Voevodsky in a series of lectures were he laid down the theoryof cross functors . The main theorem of this theory consists in giving a criterion on a systemof triangulated categories indexed by schemes, equipped with a basic functoriality, to be ableto construct exceptional functors ( f ! , f ! ) satisfying the properties required by Grothendieck 6functors formalism. In particular, the system of triangulated categories must satisfy three notableproperties: the A -localization property , the P -stability property and the localization property .Unfortunately, only an introductory part on this theory was released (see [ Del01 ]) in which thebasic setup is established but which does not contain the proof of the main result.The writing of this theory was accomplished by J. Ayoub in his thesis (see [
Ayo07a, Ayo07b ]).Ayoub uses the axioms laid down by Voevodsky: he calls a system of triangulated categories sat-isfying the properties alluded above a homotopy stable functor . However, he goes far beyond theoriginal result of Voevodsky: apart the complete theory of cross functors (concerned with f ! , f ! ),he also studied tensor structures, constructibility properties and their stability under the six op-erations, t-structures and specialization functors such as the vanishing cycle functor. The mainexample of a stable homotopy functor is the stable A -homotopy category. One readily deducesthat the category of Morel motives is also a homotopy stable functor.However, it is by no means obvious that the category of modules over the motivic homotopyring spectrum does meet the requirements of a homotopy stable functor. In fact, it can be seenthat this is equivalent to Conjecture 15 of Voevodsky in [ Voe98 ]. A.5. Grothendieck 6 functors formalism.
A.5.1 . We now give the precise formulation of the
Grothendieck 6 functors formalism . Aspresented here, it is extracted from the properties of the derived category of l -torsion ´etale sheavesobtained in [ SGA4 , tome 3]. A triangulated category T , fibred over the category of schemes, satisfies the Grothendieck 6functors formalism if the following conditions hold: Heuristically, the essential difference between stable A -homotopy and motivic complexes is the presence oftransfers in the later case. See also Theorem 11 in this introduction and its corollary. See also Theorem 8 in this introduction for an extension of their result. In a precise sense, the problem of obtaining the Grothendieck 6 functors formalism with that definitionis almost equivalent to solving affirmatively conjecture 17 of [
Voe02b ]. See below for a partial answer to thatconjecture. It also coincides with formulas gathered by Deligne in an unpublished note which he graciously support uswith. (1) There exists 3 pairs of adjoints functors as follows: f ∗ : T ( X ) ⇄ T ( Y ) : f ∗ , f any morphism, f ! : T ( Y ) ⇄ T ( X ) : f ! , f any separated morphism of finite type,( ⊗ , Hom ) , symetric closed monoidal structure on T ( X ) . (2) There exists a structure of a covariant (resp. contravariant) 2-functors on f f ∗ , f f ! (resp. f f ∗ , f f ! ).(3) There exists a natural transformation α f : f ! → f ∗ which is an isomorphism when f is proper. Moreover, α is a morphism of 2-functors.(4) For any smooth morphism f : X → S in S of relative dimension d , there exists acanonical natural isomorphism of 2-functors p ′ f : f ∗ −→ f ! ( − d )[ − d ]where ?( − d ) denotes the inverse of the Tate twist iterated d -times. Moreover p ′ is anisomorphism of 2-functors.(5) For any cartesian square in S : Y ′ f ′ / / g ′ (cid:15) (cid:15) ∆ X ′ g (cid:15) (cid:15) Y f / / X, such that f is separated of finite type, there exist natural isomorphisms g ∗ f ! ∼ −→ f ′ ! g ′∗ ,g ′∗ f ′ ! ∼ −→ f ! g ∗ . (6) For any separated morphism of finite type f : Y → X , there exist natural isomorphisms Ex ( f ∗ ! , ⊗ ) : ( f ! K ) ⊗ X L ∼ −−→ f ! ( K ⊗ Y f ∗ L ) , Hom X ( f ! ( L ) , K ) ∼ −−→ f ∗ Hom Y ( L, f ! ( K )) ,f ! Hom X ( L, M ) ∼ −−→ Hom Y ( f ∗ ( L ) , f ! ( M )) . (Loc) For any closed immersion i : Z → S with complementary open immersion j , there existsa distinguished triangle of natural transformations as follows: j ! j ! α ′ j −−−→ α i −−−→ i ∗ i ∗ ∂ i −−→ j ! j ! [1]where α ′ ? (resp. α ? ) denotes the counit (resp. unit) of the relevant adjunction.A.5.2 . The next part of Grothendieck 6 functors formalism is concerned with duality. Thiskind of properties appears already in [
Har66 ]. It is considered more axiomatically, in the caseof ´etale sheaves, in [
SGA5 , Exp. I]. In loc. cit. , Grothendieck states the fundamental propertyof absolute purity and indicates its fundamental link with duality. We state these properties asnatural extensions of the properties given in the preceding paragraph; assume T satisfies thesepreceding properties:(7) Absolute purity .– For any closed immersion i : Z → S of regular scheme of (constant)codimension c , there exists a canonical isomorphism: Z ( − c )[ − c ] ∼ −→ i ! ( X )where denotes the unit object for the tensor product. The duality properties are stated in the unpublished notes of Delignes as part of the complete formalism. . VOEVODSKY’S MOTIVIC COMPLEXES xi (8)
Duality .– Let S be regular scheme and K S be any invertible object of T ( S ). For anyseparated finite type morphism f : X → S , put K X = f ! ( K S ). For any object M of T ( X ), put D X ( M ) = Hom ( M, K X ).(a) For any X/S as above, K X is a dualizing object of T ( X ). In other words, thecanonical map: M → D X ( D X ( M ))is an isomorphism.(b) For any X/S as above, and any objects
M, N of T ( X ), we have a canonical iso-morphism D X ( M ⊗ D X ( N )) ≃ Hom X ( M, N ) . (c) For any morphism between separated S -schemes of finite type f : Y → X , we havenatural isomorphisms D Y ( f ∗ ( M )) ≃ f ! ( D X ( M )) f ∗ ( D X ( M )) ≃ D Y ( f ! ( M )) D X ( f ! ( N )) ≃ f ∗ ( D Y ( N )) f ! ( D Y ( N )) ≃ D X ( f ∗ ( N )) . A.5.3 . The last property we want to exhibit as a natural extension of Grothendieck 6 functorsformalism is the compatibility with projective limits of schemes. The basis for the next statementis [
SGA4 , Exp. VI] though it does not appear explicitely. As in the case of the duality property,it should involve some finiteness assumption. Note the formulation belon is valid for an arbitraryfibred triangulated monoidal category T .(9) Continuity .– Let ( S α ) α ∈ A be an essentially affine projective system of schemes. Put S = lim ←− α ∈ A S α .Then the canonical functor2- lim −→ α T ( S α ) → T ( S )is an equivalence of monoidal triangulated categories. B. Voevodsky’s motivic complexes
The primary goal of this treatise is to develop the theory of Voevodsky motives, integrallyover any base scheme , within the framework of sheaves with transfers. Actually, we can defineVoevodsky’s motives with coefficients in an arbitrary ring Λ and prove all the results stated belowin that case but we restrict to integral coefficients for simplicity.After refining and completing Suslin-Voevodsky’s theory of relative cycles, we introduce thecategory S m cor Z ,S of integral finite correspondences over smooth S schemes and the related notionof (Nisnevich) sheaves with transfers over a base scheme S (Def. 10.4.2) as in the usual case ofa perfect base field. Following the idea of stable homotopy, we define the triangulated categoryDM( X ) of stable motivic complexes (see Def. 11.1.1) as the P -stabilization of the A -localizationof the derived category of the (Grothendieck) abelian category of sheaves with transfers over S .One easily gets that the fibred category DM is equipped with the basic functoriality neededby the cross-functor formalism. The main difficulty is the localization property, property (Loc) inParagraph A.5.1. Unfortunately, though all the functors involved in the formulation of (Loc) arewell defined for DM, we can only prove this property when S and Z are smooth over some basescheme (see Prop. 11.4.2). This is not enough to apply Ayoub’s results.However, we are able to construct the 6 operations for DM using the method of Deligne, usedin [ SGA4 , XVII], and partially get the Grothendieck 6 functors formalism: In this introduction, all schemes will be assumed to be noetherian of finite dimension. ii Theorem . The triangulated category DM , fibred over the category ofschemes, satisfies the following part of the properties stated in Paragraph A.5.1: • properties (1), (2), (3), • property (4) when f is an open immersion or f is projective and smooth, • property (5) when g is smooth or f is projective and smooth, • property (6) when f is projective and smooth, • Property (Loc) when S and Z are smooth over some common base scheme. One of the application of this theory is that we get a well defined integral motivic cohomologytheory for any scheme X : H n,m M ( X, Z ) = Hom DM( X ) (cid:0) X , X ( m )[ n ] (cid:1) which enjoy the following properties (see section 11.2): • it admits a ring structure, pullback maps associated with any morphism of schemescompatible with the ring structure, • it admits push-forward maps with respect to projective morphisms between schemessmooth over some common base, or with respect to some finite morphisms (for examplefinite flat; see Paragraph 11.2.4), • it coincides with Voevodsky’s motivic cohomology groups when X is smooth over aperfect field (see Example 11.2.3); in particular one gets the following identification withhigher Chow groups: H n,m M ( X, Z ) = CH m ( X, m − n ) , • it admits Chern class and satisfies the projective bundle formula, • it admits a localization long exact sequence associated with a closed immersion of schemessmooth over some common base.As in the classical case, any smooth S -scheme X admits a motive M S ( X ) over X in DM( S ).Moreover, one defines the Tate motive S (1) as the reduced motive of P S . We defined the categoryof constructible motives DM c ( S ) as the thick triangulated subcategory of DM generated by theobjects of the form M S ( X )( n ) for a smooth S -scheme X and an integer n ∈ Z , where ?( n ) refersto the n -th Tate twist. One gets the following generalization of the classical result obtained byVoevodsky over a perfect field: Theorem . A motive M in DM( S ) is constructible if and only if it iscompact. The category DM c ( S ) is equivalent to the category obtained from the bounded homotopy cate-gory of the additive category S m cor Z ,S in the following way: • take the Verdier quotient modulo the thick triangulated subcategory generated by: – for any Nisnevich distinguished square W k / / g (cid:15) (cid:15) V f (cid:15) (cid:15) U j / / X of smooth S -schemes: [ W ] g ∗ − k ∗ −−−−→ [ U ] ⊕ [ V ] j ∗ + f ∗ −−−−→ [ X ] – for any smooth S -scheme X , p : A X → X the canonical projection: [ A X ] p ∗ −→ [ X ] , • invert the Tate twist, • take the pseudo-abelian enveloppe. The triangulated category DM c ( X ) is stable by the operations f ∗ , f ∗ when f is smoothprojective, and ⊗ but we cannot prove the stability for the other operations of DM and a fortiorido not get the duality properties (7) and (8) of the Grothendieck 6 functors formalism. Recall that M is compact if Hom( M , − ) commutes with arbitrary direct sums. . BEILINSON MOTIVES xiii However, we are able to prove the continuity property (9) for the category DM c :2- lim −→ α DM c ( S α ) ≃ DM c ( S ) , where we only require that the transition morphism of ( X α ) are affine and dominant (see Theorem11.1.24) . Note this result allows us to extend the comparison of motivic cohomology with higherChow groups to arbitrary regular schemes of equal characteristics. C. Beilinson motivesC.1. Definition and fundamental properties.
As anticipated by Morel, the theory ofmixed motives with rational coefficients is much simpler and we succeed in establishing a completeformalism for them. Our initial approach differs slightly from that of Morel. We construct, outof the rational stable homotopy category and the ring spectrum associated with rational QuillenK-theory a Q -linear triangulated category DM B ( X ), which we call the triangulated category ofBeilinson motives (see Def. 14.2.1). Essentially by construction, in the case where X is regular,we have a natural identificationHom DM B ( X ) ( Q X , Q X ( p )[ q ]) ≃ Gr pγ K p − q ( X ) Q , where the right hand side is the graded part of the algebraic K-theory of X with respect to the γ -filtration. These groups were first considered by Beilinson as the rational motivic cohomologygroups. We call them the Beilinson motivic cohomology groups .One of the interest of our definition is that the localization property (Loc) can be easilydeduced from its validity for the stable homotopy category. Therefore, the cross-functor formalismand more generally all the results of Ayoub can be applied to DM B . Using the constructions ofthis book, we obtain a slightly more general and precise formalism. Theorem . All the standard Grothendieck six functorsformalism (see Paragraph A.5.1) is verified by the fibred triangulated category DM B . Concerning duality for Beilinson motives, we first deduce from Quillen’s localization theoremin algebraic K-theory the absolute purity theorem:
Theorem . The absolute purity property (see A.5.2(7)) holds for DM B . As said before, this result is not enough to establish duality for Beilinson motives. We firsthave to use descent theory and resolution of singularities (as first explained by Grothendieck in[
SGA5 , I.3]). Using the existence of trace maps in algebraic K-theory, we prove the followingresult:
Theorem . Consider a finite group G and apullback square of schemes T h / / g (cid:15) (cid:15) Y f (cid:15) (cid:15) Z i / / X in which Y is endowed with an action of G over X . Put U = X − Z and assume the followingthree conditions are satisfied:(a) The morphism f is finite and surjective.(b) The induced morphism f − ( U ) → f − ( U ) /G is flat.(c) The morphism f − ( U ) /G → U is radicial.Put a = f ◦ i . Then, for any object M of DM B ( X ) , we get a canonical distinguished triangle in DM B ( X ) : M −→ i ∗ i ∗ ( M ) ⊕ f ∗ f ∗ ( M ) G −→ a ∗ a ∗ ( M ) G −→ M [1] where ? G means the invariants under the action of G , and the first (resp. second) map of thetriangle is induced by the difference (resp. sum) of the obvious adjunction morphisms. iv In fact, we show that this apparently simple result implies a much stronger descent propery forthe fibred triangulated category DM B , the descent property for the h-topology, thus in particular´etale and even flat descent as well as proper descent. C.2. Constructible Beilinson motives.
The next step towards duality for Beilinson mo-tives is the definition of a suitable finiteness condition. As in the case of Voevodsky motives, wedefine the category of
Beilinson constructible motives , denoted by DM B ,c ( X ), as the thick subcat-egory of DM B ( X ) generated by the motives of the form M X ( Y )( p ) := f ! f ! ( Q X )( p ) for f : Y → X separated smooth of finite type, and p ∈ Z . This category coincides with the full subcategory ofcompact objects in DM B ( X ). The usefulness of this definition comes from the following result, which is the analog of Gab-ber’s finiteness theorem in the l -adic setting. Analogously, its proof relies on absolute purity, (aweak form of) proper descent as well as Gabber’s weak uniformisation theorem. Theorem . The subcategory DM B ,c is stable under the six oper-ations of Grothendieck when restricted to excellent schemes. The final statement concerning Grothendieck 6 functors formalism in the setting of Beilinsonmotives is that, when one restricts to constructible Beilinson motives and separated B -schemes offinite type for an excellent scheme B of dimension less than 2, the complete formalism is available: Theorem . The fibred category DM B ,c over the categoryof schemes described above satisfies the complete Grothendieck 6 functors formalism described insection A.5, in particular the duality property A.5.2(8) and the continuity property
A.5.3(9).
Remark.
Note that the finiteness theorem as well as the duality property are also conse-quences of [
Ayo07a ], respectively Scholie 2.2.34 and Theorem 2.3.73, applied to DM B when onerestricts to quasi-projective schemes over a field or a discrete valuation ring. As ours, the proof ofAyoub uses in an essential way the absolute purity property (Theorem 4 stated above). C.3. Comparison theorems.
In the historical part of this introduction, we saw many ap-proaches for the triangulated category of (rational) motives. We succeed in comparing them allwith our definition of Beilinson motives.Denote by
KGL S the algebraic K-theory spectrum in Morel and Voevodsky’s stable homotopycategory SH( S ). By virtue of a result of Riou, the γ -filtration on K-theory induces a decompositionof KGL S, Q : KGL S, Q ≃ M n ∈ Z H B ,S ( n )[2 n ] . The ring spectrum H B ,S represents Beilinson motivic cohomology. Almost by construction, thecategory DM B ( S ) is the full subcategory of SH Q ( S ) which consists of objects E such that theunit map E → H B ,S ⊗ E is an isomorphism. In fact, our first comparison result relates thetheory of Beilinson motives with the approach of R¨ondigs and Østvær through modules over aring spectrum: Theorem . For any scheme S , there is a canonical equivalence of categories DM B ( S ) ≃ Ho( H B ,S - mod) where the left hand side denotes the homotopy category of modules over the ring spectrum H B ,S . The next comparison involves the the h-topology: this is the Grothendieck topology on thecategory of schemes, generated by ´etale surjective morphisms and proper surjective morphisms.The first published work of Voevodsky on triangulated categories of mixed motives ([
Voe96 ]),introduces the A -homotopy category of the derived category of h-sheaves. We consider a Q -linear and P -stable version of it, which we denote by DM h , Q ( S ). By construction, for any S -scheme of finite type X , there is a h-motive M S ( X ) in DM h , Q ( S ). We define DM h , Q ( S ) as the Note the striking analogy with perfect complexes. i.e. that, locally for the h-topology, any excellent scheme is regular, and any closed immersion betweenexcellent schemes is the embedding of a strict normal crossing divisor into a regular scheme. . BEILINSON MOTIVES xv smallest triangulated full subcategory of DM h , Q ( S ) which is stable by (infinite) direct sums, andwhich contains the objects M S ( X )( p ), for X/S smooth of finite type , and p ∈ Z . Using h-descentin DM B , we get the following comparison result. Theorem . If S is excellent, then we have canonical equivalences of cate-gories DM B ( S ) ≃ DM h , Q ( S ) . In fact, we first prove this result for the variant of DM h , Q ( S ) obtained by replacing everywherethe h-topology by the qfh-topology – for the later, coverings are generated by ´etale covers and finitesurjective morphisms. In particular we get an equivalence of categories: DM h , Q ( S ) ≃ DM qfh , Q ( S ).This result allows us to link Beilinson motives with Voevodsky’s motivic complexes. Let us denoteby DM Q the fibred category of stable motivic complexes alluded to in Paragraph B. Using thepreceding result in the case of the qfh-topology, we prove: Theorem
10 (see Th. 16.1.4) . If S is excellent and geometrically unibranch, then there is acanonical equivalence of categories DM B ( S ) ≃ DM Q ( S ) . In particular, given such a scheme S , we get a description of DM B ,c ( S ) as in Theorem 2 citedabove. Voevodsky’s integral (resp. rational) motivic cohomology is represented in SH( S ) by a ringspectrum H M ,S (resp. H Q M ,S ). The preceding theorem immediately gives an isomorphism of ringspectra: H B ,S ≃ H Q M ,S . As Beilinson motivic cohomology ring spectra over different bases are compatible with pullbacks,we easily deduce the following corollary which solves affirmatively conjecture 17 of [
Voe02b ] insome cases, and up to torsion:
Corollary.
For any morphism f : T → S of excellent geometrically unibranch schemes, thecanonical map f ∗ H Q M ,S → H Q M ,T is an isomorphism of ring spectra. The last comparison statement is concerned with the approach of Morel. According to Morel,the category SH Q ( S ) can be decomposed into two factors, one of them being SH Q ( S ) + , that isthe part of SH Q ( S ) on which the map ǫ : S Q → S Q , induced by the permutation of the factors in G m ∧ G m , acts as −
1. Let S Q + be the unit object of SH Q ( S ) + .Using the presentation of Beilinson motives in terms of H B -modules (Theorem 8 cited above)as well as Morel’s computation of the motivic sphere spectrum in terms of Milnor-Witt K-theory,we obtain another proof of a result of Morel (see [ Mor06 ]):
Theorem
11 (see Th. 16.2.13) . For any scheme S , the canonical map S Q + → H B ,S is anisomorphism. In fact, we even get the following corollary:
Corollary.
For any scheme S , there is a canonical equivalence of categories SH Q ( S ) + ≃ DM B ( S ) . Note in particular that, when S is regular, we get an isomorphism: H p,q M ( S, Z ) ⊗ Q ≃ Gr pγ K p − q ( S ) Q which extends the known isomorphism when S has equal characteristics. It is natural with respect to pullbacksand compatible with products. vi Recall from Morel theory that, when − S , ǫ is equal to − Id on the whole of SH Q ( S ). Thus in that particular case (e.g. S is a scheme overan algebraically closed field), the category of Beilinson motives coincide with the rational stablehomotopy category. In general, we can introduce according to Morel the ´etale variant of SH Q ( S )denoted by D A , ´et ( S, Q ). As locally for the ´etale topology, − B satisfies ´etale descent, we get the following final illuminating comparison statement. Corollary.
For any scheme S , there is a canonical equivalence of categories D A , ´et ( S, Q ) ≃ DM B ( S ) . Let us draw a conclusive picture which summarize most of the comparison results we obtained:
Corollary.
Given any scheme S , the category DM B ( S ) is a full subcategory of the rationalstable homotopy category SH Q ( S ) . Given an rational spectrum E over S , the following conditionsare equivalent:(i) E is a Beilinson motive,(ii) E is an H B ,S -module,(iii) E satisfies ´etale descent,(iii’) ( S excellent) E satisfies qfh -descent,(iii”) ( S excellent) E satisfies h -descent,(iv) ( S excellent geometrically unibranch) E admits transfers,(v) the endomorphism ǫ ∈ End( S Q ) acts by − Id on E i.e. ǫ ⊗ E = − E . Remark. (see Corollary 14.2.16) Points (iv) and (v) are related to the orientation theory forspectra (not only ring spectra). In fact, H B ,S is the universal orientable rational ring spectrumover S .Let Q . Sm S be the Q -linear envelop of the category Sm S . On obtains (see Example 5.3.43 inconjunction with Par. 5.3.35) that the full subcategory of compact objects of SH Q ( S ) is equivalentto the category obtained from the homotopy category K b ( Q . Sm S ) by performing the followingoperations: • take the Verdier quotient modulo the thick triangulated subcategory generated by: – for any Nisnevich distinguished square W k / / g (cid:15) (cid:15) V f (cid:15) (cid:15) U j / / X of smooth S -schemes: Q S ( W ) g ∗ − k ∗ −−−−→ Q S ( U ) ⊕ Q S ( V ) j ∗ + f ∗ −−−−→ Q S ( X ) – for any smooth S -scheme X , p : A X → X the canonical projection: Q S ( A X ) p ∗ −→ Q S ( X ) . • invert the Tate twist, • take the pseudo-abelian enveloppe.Let us denote by D A ,c ( S, Q ) this category. We finally obtain the following concrete descriptionof Beilinson constructible motives: Corollary.
Given any scheme S , the category DM B ,c ( S ) is equivalent to the full subcategoryof D A ,c ( S, Q ) made by the objects E which satisfies one the following equivalent conditions:(i) (Galois descent) given any smooth S -scheme X and any Galois S -cover f : Y → X ofgroup G , the canonical map E ⊗ Q S ( Y ) /G → E ⊗ Q S ( X ) is an isomorphism,(ii) (Orientability) ǫ acts by − Id on E , Recall again the following remarks:(1) When ( −
1) is a sum of square in every residue fields of S , conditions (i), (ii) are true forany rational spectrum E over S . In brief, this is the P -stabilization of the A -localization of the derived category of sheaves of Q -vectorspaces over the lisse-´etale of S . . DETAILED ORGANIZATION xvii (2) When S is excellent and geometrically unibranch, the category DM B ,c ( S ) is equivalentto the category of rational geometric Voevodsky motives (same definition as in Theorem2 but replacing Z by Q ). C.4. Realizations.
The last feature of Beilinson motives is that they are easily realizablein various cohomology theories. To get this fact, we use the setting of modules over a strict ringspectrum. Given such a ring spectrum E in DM B ( S ), one can define, for any S -scheme X , thetriangulated category D( X, E ) = Ho( E X - mod) , where E X = f ∗ E , for f : X → S the structural map.We then have realization functorsDM B ( X ) → D( X, E ) , M
7→ E X ⊗ X M which commute with the six operations of Grothendieck. Using Ayoub’s description of the Bettirealization, we obtain: Theorem . If S = Spec ( k ) with k a subfield of C , and if E Betti represents Betti cohomol-ogy in DM B ( S ) , then, for any k -scheme of finite type, the full subcategory of compact objects of D( X, E Betti ) is canonically equivalent to D bc ( X ( C ) , Q ) . More generally, if S is the spectrum of some field k , given a mixed Weil cohomology E , withcoefficient field (of characteristic zero) K , we get realization functorsDM B ,c ( X ) → D c ( X, E ) , M
7→ E X ⊗ X M (where D c ( X, E ) stands for the category of compact objects of D( X, E )), which commute with thesix operations of Grothendieck (which preserve compact objects on both sides). Moreover, thecategory D c ( S, E ) is then canonically equivalent to the bounded derived category of the abeliancategory of finite dimensional K -vector spaces. As a byproduct, we get the following concretefiniteness result: for any k -scheme of finite type X , and for any objects M and N in D c ( X, E ),the K -vector space Hom D c ( X, E ) ( M, N [ n ]) is finite dimensional, and it is trivial for all but a finitenumber of values of n .If k is of characteristic zero, this abstract construction gives essentially the usual categoriesof coefficients (as seen above in the case of Betti cohomology), and in a sequel of this work, weshall prove that one recovers in this way the derived categories of constructible ℓ -adic sheaves (ofgeometric origin) in any characteristic. But something new happens in positive characteristic: Theorem . Let V be a complete discrete valuation ring of mixed characteristic, with field offunctions K , and residue field k . Then rigid cohomology is a K -linear mixed Weil cohomology, andthus defines a ring spectrum E rig in DM B ( k ) . We obtain a system of closed symmetric monoidaltriangulated categories D rig ( X ) = D c ( X, E rig ) , for any k -scheme of finite type X , such that Hom D rig ( X ) ( X , X ( p )[ q ]) ≃ H q rig ( X )( p ) , as well as realization functors R rig : DM B ,c ( X ) → D rig ( X ) which preserve the six operations of Grothendieck. D. Detailed organization
The book is organized in four parts that we now review in more details. i.e. we say a ring spectrum is strict if it is a commutative monoid in the underlying model category. viii D.1. Grothendieck six functors formalism (Part 1).
The first part is concerned withthe formalism described in section A.5 above. It is the foundational part of this work.We use the language of fibred categories (introduced in [
SGA1 , VI]), complemented by thatof 2-functors (or pseudo-functors), in order to describe the 6 functors formalism. We first describean axioms which allow to derive the core formalism – i.e. the part described in section A.5.1 –from simpler axioms. We do not claim originality in this task: our main contribution is to givea synthesis of the approach of Deligne described in [
SGA4 , XVII] (see also [
Har66 , Appendix])with that of Voevodsky developed by Ayoub in [
Ayo07a ].Recall that a (cleaved) fibred category M over S can be seen as a family of categories M ( S )for every object S of S together with a pullback functor f ∗ : M ( S ) → M ( T ) for any morphism f : T → S of S . Given a suitable class P of morphisms in S , we set up a systematic study ofa particular kind of fibred categories, called P -fibred categories (definition 1.1.10): one where forany f in P , the pullback functor f ∗ admits a left adjoint, generically denoted by f ♯ . The functor f ♯ has to be thought as a variant of the exceptional direct image functor . In section 1, we study basic properties of P -fibred categories which will be the core of the6 functors formalism, such as base change formulas and projection formulas when an additionalmonoidal structure is involved. These formulas are particular case of a compatibility relation be-tween different kind of functors expressed through a canonical comparison morphism. Such kindof comparison morphisms are generically called exchange morphisms . They are very versatile andappears everywhere in the theory (see Paragraphs 1.1.6, 1.1.15, 1.1.24, 1.1.31, 1.1.33, 1.2.5). Infact, they appears fundamentally in Grothendieck 6 functors formalism: in the list of propertiesA.5.1, they are the isomorphisms of (5), (6) and even (4). In the direction of the full Grothendieckfunctoriality, we introduce a core axiomatic for P -fibred categories that we consider as minimal:the categories satisfying this axiomatic are called P -premotivic (section 1.4). P -premotivic cat-egories will form the basic setting in all this work. They will appear in three different flavours,depending on which particular kind of additional structure we consider on categories: abelian,triangulated and model categories.In Section 2, we restrict our attention to the triangulated and geometric case, meaning thatwe consider triangulated P -fibred categories over a suitable category of schemes S . The aimof the section is to develop, and extend, Grothendieck 6 functors formalism in this basic setting.We exhibit many properties of such fibred categories which are indexed in the appendix. Let usconcentrate in this introduction on the two main properties which will corresponds respectivelyto Deligne and Voevodsky’s approach on the 6 functors formalism.The first one, called the support property and abbreviated by (Supp), asserts that the adjointfunctors of the kind f ∗ , for f proper, and j ♯ , for j an open immersion, satisfy a gluing propertythat allows to use the argument of Deligne to construct the exceptional direct image functor f ! . Several properties are derived from (Supp) and the basic axioms of P -fibred categories which leadto a partial version of the 6 functors formalism (see Theorem 2.2.14).The second property, most fundamental in the motivic context, is the localization property abbreviated by (Loc), which is in fact part of the 6 functors formalim (see Paragraph A.5.1). Ithas many interesting consequences and reformulations that are derived in section 2.3.1. Note that(Loc) is also known in the literature as the “gluing formalism”. Some of the properties that weprove in loc.cit. are already classical (see [ BBD82 ]).The most interesting consequence of (Loc) was discovered by Voevodsky: together with theusual A -localization and P -stabilization properties of the motivic context, it implies the completebasic 6 functors formalism as stated in Paragraph A.5.1. This was proved by Ayoub in [ Ayo07a ]. These pullback functors are subject to the usual cocycle condition ; see section 1. This kind of situation frequently happens: analytical case (open immersions), sheaves on the small ´etale site(´etale morphisms), Nisnevich sheaves on the smooth site (smooth morphisms). In the context of torsion ´etale sheaves of [
SGA4 , XVII], property (Supp) is a consequence of the properbase change theorem. . DETAILED ORGANIZATION xix
In section 2.4, we revisit the proof of Ayoub and give some improvement of his theorems (seeTheorem 2.4.50 for the precise statement): • we remove the quasi-projectivity assumption for the existence of f ! , replacing it by theassumption that f is separated of finite type; • we introduce the orientation property which allows to get a simpler more usual form tothe purity isomorphism (the one stated in point (4) of A.5.1); • we give another proof of the main theorem in the oriented case by showing that relativepurity is equivalent to some (strong) duality property in the smooth projective case (seeTheorem 2.4.42); • we directly incorporate the monoidal structure whereas Ayoub gives a separate discussionfor this.Apart from these differences, the material of section 2.4 is very similar to that of [ Ayo07a ].Moreover, in the non oriented case, it should be clear that we rely on the original argument ofAyoub for the proof of Theorem 2.4.42.Concerning terminology, we have called motivic triangulated category (Definition 2.4.45) whatAyoub calls a “monoidal stable homotopy functor”.The remaining of Part 1 is concerned with extensions of Grothendieck 6 functors formalism.In Section 3, we show how to use the setting of P -fibred model categories as a framework toformulate Deligne’s cohomological descent theory.Unless in trivial cases, object of a derived category are not local. To formulate descent theoryin derived categories, the main idea of Deligne was to extend the derived category of a scheme byone relative to a simplicial scheme, usually a hypercover with respect to a Grothendieck topology(see [
SGA4 , Vbis]). The construction consists in first extending the theory of sheaves to the casewhere the base is a simplicial schemes and then consider the associated derived category.We generalize this construction to the case of an arbitrary P -fibred category equipped witha model structure. In fact, we show in Section 3.1 how to extend a P -fibred category over acategory of schemes to the corresponding category of simplicial schemes and even of arbitrarydiagrams of schemes. Most importantly, we show how to extend the fibred model structure tothe case of diagrams of schemes (see Prop. 3.1.11). Concretely, this means that we define aderived functor of the kind L ϕ ∗ (resp. R ϕ ∗ ) for an arbitrary morphism ϕ of diagrams of schemes.Let us underline that these derived functors mingles two different kinds of functoriality: theusual pullback f ∗ (resp. direct image f ∗ ) for a morphism of schemes f together with homotopycolimits (resp. limits) – see the discussion in Paragraph 3.1.12 till Proposition 3.1.16. Withthat extension in hands, we can easily formulate (cohomological) descent theory for arbitraryGrothendieck topologies on the category of schemes for the homotopy category of a P -fibredmodel category: see Definition 3.2.5.The end of Section 3 is devoted to concrete examples of descent in P -fibred model categories,and their relation with properties of the associated homotopy category, assuming it is triangu-lated, as introduced in Section 2. The first and most simple example corresponds to the case of aGrothendieck topology associated with a cd-structure in the sense of Voevodsky (as the Nisnevichand the cdh-topology. See [ Voe10b ] or Paragraph 2.1.10). In that case, descent can be char-acterized as the existence of certain distinguished triangles (Mayer-Vietoris for Zariski topology, The first example of this fact is the circle: any non trivial open subset of S is contractible whereas S itselfis not. Recall that model structures, introduced by Quillen, allow to perform all the usual constructions of derivedcategories by localizing an arbitrary category with respect to a given class of morphisms called weak equivalences.It contains in particular the usual case of complexes of an arbitrary abelian category with quasi-isomorphisms asweak equivalences. The main construction of the theory of Quillen is that of left (resp. right) derived functorswhich can be defined by replacing the usual notion of projective (resp. injective) resolution by that of cofibrant(resp. fibrant) resolution. By restricting the morphisms of diagrams of schemes to a certain class denoted by P cart , we also show howto get a P cart -fibred model category over diagrams of schemes (Rem. 3.1.21) but this is not really needed in thedescent theory. x Brown-Gersten for Nisnevich topology): this is Theorem 3.3.2 which is in fact a reformulation ofthe results of Voevodsky.We then proceed to the most fundamental case of descent in algebraic geometry, that forproper surjective maps which allows in principle the use of resolution of singularities. In fact, themain result of the whole of Section 3 is a characterization of h-descent which allows to reduce it, for P -fibred homotopy triangulated categories which are rational and motivic, to a simple propertyeasily checked in practice : this is Theorem 3.3.37. Along the way, we proved also the followingresults interesting in their own: • several characterization of ´etale descent (Theorems 3.3.23 and 3.3.32); • a characterization of qfh-descent (Theorem 3.3.25) as if it was defined by a cd-structure. In fact, the last point is the heart of the proof of the main Theorem, 3.3.37. Whereas the extensionof fibred homotopy categories to diagrams of schemes is not unprecedented (see [
Ayo07b ]), ourstudy of proper and h-descent seems to be completely new. In our opinion, it is one of the mostimportant technical innovation of this book.In Section 4, we study the extension of Grothendieck 6 functors formalism in rational motiviccategories, mainly duality and continuity. As already mentioned, the general principle is not knewand follows mainly the path laid by Grothendieck in [
SGA5 ].In the case of an abstract motivic triangulated category – which is for the purpose of descenttheory the homotopy category of an underlying fibred model category as seen above – the firsttask is to introduce a correct property of finiteness inherent to any duality theorem. This is donefollowing Voevodsky, as in the work of Ayoub, by introducing the notion of constructiblity inDefinition 4.2.1. The name is inspired by the ´etale case, but the notion of constructibility whichwe consider here is defined a generation property which really corresponds to what Voevodskycalled geometric motives : constructible motives in our sense are generated by twists of motives ofsmooth schemes and are stable by cones, direct factors and finite sums. Let us mention that in goodcases, the property of being constructible coincides with that of being compact in a triangulatedcategory, resounding with the theory of perfect complexes (in the context of l -adic sheaves, thiscorresponds to “constructible of geometric origin”).The main point on constructible motives is the study of their stability under the 6 operationsthat we get from the axioms of a triangulated motivic category. This is done in Section 4.2. Asin the ´etale case, the crucial point is the stability with respect to the operation f ∗ , when f is amorphism of finite type between excellent schemes. In Theorem 4.2.24, we give conditions on amotivic triangulated category so that the stability for f ∗ is guaranteed (then the stability by theother operations follows easily, see 4.2.29). Our proof follows essentially an argument of Gabber.The general principle, going back to [ SGA4 , XIX, 5.1], is to use resolution of singularities toreduce to an absolute purity statement which is among our assumptions. In Section 4.3, we introduce an important property of motivic triangulated categories, called continuity , which allows reasoning that involves projective limits of schemes. In fact, it is shownin Proposition 4.3.4 that this property implies the property (9) of the (extended) Grothendieck6 functors formalism (see Paragraph A.5.3 above). We also give a criterion for continuity (4.3.6)which will be applied later in concrete cases and draw some interesting consequences.Finally, Section 4.4 deals with duality in itself for constructible motives, that is property (8) ofParagraph A.5.2. The main theorem 4.4.21 asserts that, under the same condition than Theorem4.2.24, and if one restricts to schemes that are separated of finite type over an excellent basescheme B of dimension less or equal to 2, then the full duality property holds (see also Corollary4.4.24). The proof follows the line of proof of the analog Th. 2.3.73 of [ Ayo07a ]. In particularthe main point is the fact that constructible motives are generated by some nice motives adaptedto the use of resolution of singularities: see Corollary 4.4.3. The main difference with op. cit. is This is the separation property defined in 2.1.7. Let us mention here it is a consequence of the existence ofwell behaved trace maps (see the proof of Theorem 14.3.3). It is at the origin of the formulation of descent that we gave for DM B in Theorem 5(b) above. Absolute purity will be proved later for Beilinson motives. . DETAILED ORGANIZATION xxi that we use De Jong equivariant resolution of singularities [ dJ97 ], so that our assumptions are alittle bit more general.
D.2. The constructive part (Part 2).
The purpose of this part is to give a method ofconstruction of triangulated categories that satisfies the formalism described in Part 1. We havechosen to mainly use the setting of derived category. Also, we use our notion of P -fibred categories( P -premotivic with a good monoidal structure). Recall this means the pullback functor f ∗ admitsa left adjoint f ♯ when f ∈ P . Essentially, P will be either the class of smooth morphisms offinite type or the class of all morphisms of finite type (eventually separated).In Section 5.1, starting from a P -premotivic abelian category A , we first show how to provethat the derived D( A ) is also a P -premotivic category. This consists in deriving the structuralfunctors of a P -premotivic category, which is done by building a suitable underlying P -fibredmodel category in Proposition 5.1.12. Actually, the proof of the axioms of a model categoryhas already appeared in our previous work [ CD09 ]. Let us mention the flavor of this modelstructure: we can describe explicitely cofibrations as well as fibrations, by the use of an adaptedGrothendieck topology t . This model structure is linked with cohomological t -descent (as shownlater in Proposition 5.2.10). The advantage of our framework is to easily obtain the functorialityof this construction (Paragraph 5.1.23), as well as other homotopical constructions (dg-structure:Rem. 5.1.19, extension to diagrams of schemes: Par. 5.1.20). In paragraph 5.1.c, we also describein suitable cases the constructible objects of the derived category by a presentation similar to thatof Voevodsky’s geometric motives over a perfect field.In Section 5.2 (resp. Section 5.3) we show how to describe the A -localization (resp. P -stabilization) process in P -premotivic derived categories: to any P -premotivic abelian cate-gory A is associated an A -derived category D eff A ( A ) (resp. P -stable and A -derived categoryD A ( A )) in Definition 5.2.16 (resp. 5.3.22). From the model category obtained in Section 5.1, theconstruction uses the classical tools of motivic homotopy theory as introduced by Morel and Vo-evodsky. Again, our framework allows to get the same homotopical constructions as in the simplederived case as well as some nice universal properties. We also get a description of constructibleobjects under suitable assumptions: Section 5.2.d (resp. 5.3.e). These sections are filled withconcrete examples.In Section 6, we focus on the main (in fact universal) example of motivic derived category, the A -derived category of Morel, obtained by the process described above from the abelian premotiviccategory of abelian sheaves over the smooth Nisnevich site. The main point here is that one getsthe localization property for this category by a theorem of Morel and Voevodsky. We give twonew contributions on this topic. First we show in Section 6.1 that the A -derived category canbe embedded in a larger category which naturally contains objects that we can call motives ofsingular schemes. This is useful to state descent properties and will be essential to study h-motives.Second, we show in Section 6.3 how one can use the A -derived category to obtain good propertiesof another premotivic derived category satisfying suitable assumptions. This will be applied tomotivic complexes.In Section 7, we go back to the case of an arbitrary monoidal P -fibred model category M and explain how to use the setting of ring spectra and modules over ring spectra in the premotiviccontext. The main construction associates to a suitable collection of (commutative) ring spectra R in M a P -fibred monoidal category denoted by Ho( R - mod): Proposition 7.2.13. This constructionwill be used several times: • in the study of algebraic K-theory (Section 13): the category of modules over K-theory isthe fundamental technical tool to get motivic proper descent as well as motivic absolutepurity; • in the study of Beilinson motives when we will relate them with modules over motiviccohomology (Theorem 14.2.9); • in the study of realizations associated with a mixed Weil cohomology (Section 17). D.3. Motivic complexes (Part 3).
This part is concerned with the constructions describedabove, in Section B. Our aim is to extend the definition of Voevodsky’s integral motivic complexes xii to any base, then study their functoriality and introduce their non effective, or rather P -stable,counter-part.Our first task, in Section 8, is to revisit Suslin-Voevodsky’s theory of relative cycles exposed in[ SV00b ]. Indeed, they will be at the heart of the general construction. Our presentation is made toprepare the theory of finite correspondences , a particular case of relative cycles. Especially, we wantto give a meaning to the following picture representing the composition of finite correspondences α from X to Y and β from Y to Z : β ⊗ Y α / / (cid:15) (cid:15) β / / (cid:15) (cid:15) Z.α / / (cid:15) (cid:15) YX (see also (9.1.4.1)). More precisely, we want to interpret this as a diagram of cycles. Thus,we are led to consider cycles (with their support), as objects of a category. Concretely, a cycle isconsidered as a multi-pointed scheme, each point being affected with some multiplicity (an integralor rational number).This conceptual shift has the advantage of allowing a treatment of cycles analog to that ofalgebraic varieties, or rather schemes, promoted by Grothendieck via studying morphisms. Thus,we replace the various groups of relative cycles introduced by Suslin and Voevodsky in op. cit. byproperties of morphisms of cycles. Here is a list of the principal ones: • pseudo-dominant (8.1.2), equidimensional (8.1.3 and 8.3.18), • pre-special (8.1.20), • special (8.1.28), • Λ-universal (8.1.48).The most intriguing one, being pre-special , has no counter-part in op. cit.
Its idea comes froma mistake (fortunately insignificant) in the convention of Suslin and Voevodsky. Indeed, Lemma3.2.4 of op. cit. is false whenever the base S is non reduced and irreducible: then any fat point( x , x ) and any flat S -scheme give a counter-example. The explanation is that the operationof specialization along a fat point does not take into account the geometric multiplicities of thebase. On the countrary, when X is flat over an irreducible scheme S , the geometric multiplicityof any irreducible component of X is a multiple of the geometric multiplicity of S . This lead usto the definition of a pre-special morphism of cycles β/α , where a divisibility condition appearsin the multiplicities of β with respect to that of α . The main achievement of Suslin and Voevodsky’s theory is the construction of a pullbackoperation for relative cycles. In our language, it corresponds to a kind of tensor product, moreprecisely a product of cycles relative to a common base cycle (as for example the cycle β ⊗ Y α of the preceding picture). Despite our different presentation, the method to define this operationfollows closely the original idea of Suslin and Voevodsky: use the flatification theorem of Grusonand Raynaud to reduce to the case of flat base change of cycles. Recall that the key point is tofind the correct condition on cycles – or rather morphisms of cycles in our language – so that oneobtains a uniquely defined operation independent of the chosen flatification. This is measured bya specialization procedure (Definition 8.1.25) associated with fat points (Definition 8.1.22) andleads to the central notion of special morphisms of cycles (Definition 8.1.28). An innovation thatwe introduce in the theory is to give, as soon as possible, local definitions at a point in the styleof EGA. This is in particular the case for the property of being special.Once this notion is in place, one defines for a base cycle α , a special α -cycle β and anymorphism φ : α ′ → α the relative product denoted by β ⊗ α α ′ , equivalently the base change of Explicitly, take S = Z = Spec (cid:0) k [ t ] / ( t ) (cid:1) = { η } , R = (cid:0) k [ t ] (cid:1) ( t ) . The left hand side of the equality of 3.2.4 is2 .η while the right hand side is η . To anticipate the remaining of the construction, given a non reduced scheme S , this will allow for theoperation of pullback along the immersion S red → S associated with the reduction of S : it simply corresponds todividing by the geometric multiplicities of S , as the base change to S red does for flat S -schemes. . DETAILED ORGANIZATION xxiii β/α along φ (Definition 8.1.39). This notion is close to the correspondence homomorphisms ofSection 3.2 of op. cit. In particular it usually involves denominators. The last important notion,being Λ -universal , corresponds to cycles β/α with coefficients in a ring Λ ⊂ Q , which keeps theircoefficients in Λ after any base change.One sees that our language is especially convenient when it is time to consider the stabilityof certain properties of morphisms of cycles by composition (Cor. 8.2.6) or base change (Cor.8.1.45). Then the usual statements of intersection theory are proven in Section 8.2, still followingor extending Suslin and Voevodsky: commutativity, associativity, projection formulas. This makesour relative product a good extension of the classical notion of exterior product of cycles (over afield).The focal point of intersection theory is the study of multiplicities. Thus we introduce Suslin-Voevodsky’s multiplicities , as the ones appearing as a corollary of the existence of the relative cycle β ⊗ α α ′ (Definition 8.1.42). A very important result in the theory, already enlightened by Suslinand Voevodsky is the fact these multiplicities can be expressed in terms of Samuel multiplicities . In fact, we even prove a criterion for the property of being special at a point involving Samuel mul-tiplicities at the branches of the point: see Corollary 8.3.25. Roughly speaking, the multiplicitiesarising from Samuel’s definition at each branches of the point must coincide: then this commonvalue is simply the Suslin-Voevodsky’s multiplicity.Finally, still following the treatment of algebraic geometry by Grothendieck, we introducein the theory the study of the constructibility of properties of morphisms of cycles (special andΛ-universal). Explicitly, we prove that given a relative cycle β/α , when α is the cycle associatedwith a scheme S , the locus where β is special (resp. Λ-universal) is an ind-constructible subsetof S (Lemma 8.3.4). This allows to prove the good behavior of these notions with respect toprojective limits of schemes (see in particular 8.3.9). This will be the key point when proving thecontinuity property – (9) of A.5.3 – of the fibred category DM.The rest of Part 3, consists in extending the theory of sheaves with transfers introduced byVoevodsky, originally over a perfect field, to the case of an arbitrary base and apply to it thegeneral procedures studied in Part 2 to get the fibred category DM.In Section 9, we work out the theory of finite correspondences using the formalism of relativecycles. The construction is summarized in Corollary 9.4.1: given a class of morphisms P containedin the class of separated morphisms of finite type and a ring of coefficients Λ, we produce a monoidal P -fibred category, denoted by P cor Λ , whose fiber over a noetherian scheme (eventually singular)is the category of P -schemes with morphisms the finite correspondences.In Section 10, we develop the theory of sheaves with transfers along the very same line as theoriginal treatment of Voevodsky. This time, the outcome can be summarized by Corollaries 10.3.11and 10.3.15: given a class P of morphisms as above and a suitable Grothendieck topology t , weconstruct an abelian premotivic category Sh t ( P , Λ) which is compatible with the topology t (cfPart 2); its fiber over a scheme S is given by t -sheaves of Λ-modules with transfers (in particularpresheaves on P cor Λ ,S ). The section is closed with an important comparison result, essentially dueto Voevodsky, between Nisnevich sheaves with transfers and sheaves for the qfh-topology (withrational coefficients over geometrically unibranch bases): see Theorem 10.5.14.Finally, Section 11 is devoted to gather the work done previously and define the stable derivedcategory of motivic complexes DM Λ , given an arbitrary ring of coefficients Λ. The out-come hasalready been described in Section B above. D.4. Beilinson motives (Part 4).
This part contains the construction of Beilinson motivesas well as the proof of all the properties stated before. It is based on the first and second partsbut independant of the third one – except in the comparison statements of Section 16.1.Section 12 contains a short reminder on the stable homotopy category and the notion oforiented ring spectra. When a correct regularity assumption is added, one reduces to the usual Serre’s Tor-intersection formula:see 8.3.31 and 8.3.32). The most notable topologies t that fit in this result are the Nisnevich and the cdh ones. See Section 10.4. xiv Section 13 is the heart of our construction. It contains a detailed study of the K-theory ringspectrum
KGL and the associated notion of
KGL -modules in the homotopical sense (based onthe formalism introduced in Section 7). Using the works of several authors (most notably: Riou,Nauman, Spitzweck, Østvær), we show how the central results of Quillen on algebraic K-theorygive important properties of
KGL -modules: absolute purity (Th. 13.6.3) and trace maps (Def.13.7.4).In Section 14, we finally introduce the definition of Beilinson motives. Let us describe it indetail now. It is based on the process of Bousfield localization of the stable homotopy categorywith respect to a cohomology. This operation is fundamental in modern algebraic topology. Weapply it in algebraic geometry to the rational stable homotopy category (or, what amount tothe same, to the rational stable A -derived category of Morel, Section 6) and to the rationalK-theory spectrum KGL Q : the Bousfield localization of D A , Λ ( S, Q ) with respect to KGL Q ,S isthe category of Beilinson motives DM B ( S ) over S (Definition 14.2.1). Using the preceding studyof KGL Q together with the decomposition of Riou recalled in the beginning of Section C.3, weget the main properties of the premotivic category DM B : the h-descent theorem (14.3.4) and theabsolute purity theorem (14.4.1).Then the theoretical background laid down in Part 1 is applied to DM B , given in particularthe complete Grothendieck six functors formalism for constructible Beilinson motives (Section 15).Our work closes on the two main subjects described above on Beilinson motives: the comparisonstatements (Section 16) and the study of motivic realizations (Section 17). Notations and conventions
In every section, we will fix a category denoted by S which will contain our geometric objects.Most of the time, S will be a category of schemes which are suitable for our needs; the requiredhypothesis on S are given at the head of each section. In the text, when no precisions are given,any scheme will be assumed to be an object of S .When A is an additive category, we denote by A ♮ the pseudo-abelian enveloppe of A . Wedenote by C( A ) the category of complexes of A . We consider K( A ) (resp. K b ( A )) the categoryof complexes (resp. bounded complexes) of A modulo the chain homotopy equivalences and when A is abelian, we let D( A ) be the derived category of A .If M is a model category, Ho( M ) will denote its homotopy category.We will use the notation α : C ⇄ D : β to mean a pair of functors such that α is left adjoint to β . Similarly, when we speak of an adjointpair of functors ( α, β ), α will always be the left adjoint. We will denote by ad ( α, β ) : 1 → βα (resp. ad ′ ( α, β ) : αβ → α, β ). Considering a natural tranformation η : F → G of functors, we usually denote by the same letter η — when the context is clear — the inducednatural transformation AF B → AGB obtained when considering functors A and B composed onthe left and right with F and G respectively.In section 8, we will assume that equidimensional morphims have constant relative dimension. Thanks
The authors want to thank the following mathematicians for help, motivation, proof-readingduring the elaboration of this text: J. Ayoub, A. Beilinson, P. Deligne, B. Drew, D. H´ebert,J. Hornbostel, A. Huber-Klawitter, B. Kahn, S. Kelly, M. Levine, G. Maltsiniotis, F. Morel,P.A. Østvær, J. Riou, O. R¨ondigs, V. Sala, M. Spitzweck, V. Voevodsky, V. Vologodsky, C. Weibel,J. Wildeshaus. art 1
Fibred categories and the six functorsformalism . General definitions and axiomatic . We assume that S is an arbitrary category.We shall say that a class P of morphisms of S is admissible if it is has the following properties.(Pa) Any isomorphism is in P .(Pb) The class P is stable by composition.(Pc) The class P is stable by pullbacks: for any morphism f : X → Y in P and anymorphism Y ′ → Y , the pullback X ′ = Y ′ × Y X is representable in S , and the projection f ′ : X ′ → Y ′ is in P .The morphisms which are in P will be called the P -morphisms . In what follows, we assume that an admissible class of morphisms P is fixed. P -fibred categories. Definitions.
Let C at be the 2-category of categories.1.1.1 . Let M be a fibred category over S , seen as a 2-functor M : S op → C at ; see [ SGA1 ,Exp. VI]Given a morphism f : T → S in S , we shall denote by f ∗ : M ( S ) → M ( T )the corresponding pullback functor between the corresponding fibers. We shall always assume that(1 S ) ∗ = 1 M ( S ) , and that for any morphisms W g −→ T f −→ S in S , we have structural isomorphisms:(1.1.1.1) g ∗ f ∗ ∼ −→ ( f g ) ∗ which are subject to the usual cocyle condition with respect to composition of morphisms.Given a morphism f : T → S in S , if the corresponding inverse image functor f ∗ has a leftadjoint, we shall denote it by f ♯ : M ( T ) → M ( S ) . For any morphisms W g −→ T f −→ S in S such that f ∗ and g ∗ have a left adjoint, we have anisomorphism obtained by transposition from the isomorphism (1.1.1.1):(1.1.1.2) ( f g ) ♯ ∼ −→ f ♯ g ♯ . Definition . A pre- P -fibred category M over S is a fibred category M over S suchthat, for any morphism p : T → S in P , the pullback functor p ∗ : M ( S ) → M ( T ) has a leftadjoint p ♯ : M ( T ) → M ( S ). Convention . Usually, we will consider that (1.1.1.1) and (1.1.1.2) are identities. Sim-ilarly, we consider that for any object S of S , (1 S ) ∗ = 1 M ( S ) and (1 S ) ♯ = 1 M ( S ) . Example . Let S be an object of S . We let P /S be the full subcategory of the commacategory S /S made of objects over S whose structural morphism is in P . We will usually callthe objects of P /S the P -objects over S .Given a morphism f : T → S in S and a P -morphism π : X → S , we put f ∗ ( π ) = π × S T using the property (Pc) of P (see 1.0). This defines a functor f ∗ : P /S → P /T .Given two P -morphisms f : T → S and π : Y → T , we put f ♯ ( π ) = f ◦ π using the property(Pb) of P . this defines a functor f ♯ : P /T → P /S . According to the property of pullbacks, f ♯ is left adjoint to f ∗ .We thus get a pre- P -fibred category P / ? : S P /S . In practice, S will be an adequate subcategory of the category of noetherian schemes and P will be the classof smooth morphisms (resp. ´etale morphisms, morphisms of finite type, separated or not necessarily separated) in S . We can always strictify globally the fibred category structure so that g ∗ f ∗ = ( fg ) ∗ for any composablemorphisms f and g , and so that (1 S ) ∗ = 1 M ( S ) for any object S of S ; moreover, for a morphism h of S such thata left adjoint of h ∗ exists, and we can choose the left adjoint functor h ♯ which we feel as the most convenient forus, depending on the situation we deal with. For instance, if h = 1 S , we can choose h ♯ to be 1 M ( S ) , and if h = fg ,with f ∗ and g ∗ having left adjoints, we can choose h ♯ to be f ♯ g ♯ (with the unit and counit naturally induced bycomposition). . GENERAL DEFINITIONS AND AXIOMATIC 3 Example . Assume S is the category of noetherian schemes of finite dimension, and P = Sm . For a scheme S of S , let H • ( S ) be the pointed homotopy category of schemes over S defined by Morel and Voevodsky in [ MV99 ]. Then according to op. cit. , H • is a pre- Sm -fibredcategory over S .1.1.6 . Exchange structures I .– Suppose given a weak P -fibred category M .Consider a commutative square of S Y q / / g (cid:15) (cid:15) ∆ X f (cid:15) (cid:15) T p / / S such that p and q are P -morphisms, we get using the identification of convention 1.1.3 a canonicalnatural transformation Ex (∆ ∗ ♯ ) : q ♯ g ∗ ad ( p ♯ ,p ∗ ) −−−−−−→ q ♯ g ∗ p ∗ p ♯ = q ♯ q ∗ f ∗ p ♯ ad ′ ( q ♯ ,q ∗ ) −−−−−−→ f ∗ p ♯ called the exchange transformation between q ♯ and g ∗ . Remark . These exchange transformations satisfy a coherence condition with respect tothe relations ( f g ) ∗ = g ∗ f ∗ and ( f g ) ♯ = f ♯ g ♯ . As an example, consider two commutative squaresin S : Z q ′ / / h (cid:15) (cid:15) Θ Y q / / g (cid:15) (cid:15) ∆ X f (cid:15) (cid:15) W p ′ / / T p / / S and let ∆ ◦ Θ be the commutative square made by the exterior maps – it is usually called the hor-izontal composition of the squares. Then, the following diagram of 2-morphisms is commutative:( qq ′ ) ♯ h ∗ Ex (∆ ◦ Θ) ∗ ♯ / / f ∗ ( pp ′ ) ♯ q ♯ q ′ ♯ h ∗ Ex (Θ ∗ ♯ ) / / q ♯ g ∗ p ′ ♯ Ex (∆ ∗ ♯ ) / / f ∗ p ♯ p ′ ♯ To see this, one divides this diagram as follows: ( qq ′ ) ♯ h ∗ a d ( pp ′ ) / / ( qq ′ ) ♯ h ∗ ( pp ′ ) ♯ ( pp ′ ) ∗ ( qq ′ ) ♯ ( qq ′ ) ∗ f ∗ ( pp ′ ) ∗ a d ′ ( qq ′ ) / / f ∗ ( pp ′ ) ♯ q ♯ q ′ ♯ h ∗ p ′ ∗ p ′ ♯ a d p / / q ♯ q ′ ♯ h ∗ p ′ ∗ p ∗ p ♯ p ′ ♯ q ♯ q ′ ♯ q ′ ∗ q ∗ f ∗ p ♯ p ′ ♯ a d ′ q ′ / / q ♯ q ∗ f ∗ p ♯ p ′ ♯ q ♯ q ′ ♯ q ′ ∗ g ∗ p ′ ♯ a d p / / q ♯ q ′ ♯ q ′ ∗ g ∗ p ∗ p ♯ p ′ ♯ a d ′ q ′ / / q ♯ g ∗ p ∗ p ♯ p ′ ♯ q ♯ q ′ ♯ h ∗ a d p ′ / / q ♯ q ′ ♯ h ∗ p ′ ∗ p ′ ♯ q ♯ q ′ ♯ q ′ ∗ g ∗ p ′ ♯ a d ′ q ′ / / q ♯ g ∗ p ′ ♯ a d p / / q ♯ g ∗ p ∗ p ♯ p ′ ♯ q ♯ q ∗ f ∗ p ♯ p ′ ♯ a d ′ q / / f ∗ p ♯ p ′ ♯ where ad r (resp. ad ′ r ) indicates the morphism obtained from the obvious unit morphism (resp.counit morphism) of the adjunction ( r ♯ , r ∗ ) by eventually adding functors on the left or on theright. The reader can check easily that each cell of the above diagram is commutative, provingour claim.Thus, according to our abuse of notation for natural transformations, Ex behaves as a con-travariant functor with respect to the horizontal composition of squares. The same is true forvertical composition of commutative squares. Remark . In the sequel, we will introduce several exchange transformation betweenvarious functor. We speak of an exchange isomorphism when the transformation is an exchangeisomorphism . When only two kind of functors are involved, say of type a and b, we say that . GENERAL DEFINITIONS AND AXIOMATIC 5 functors of type a and functors of type b commutes when the exchange transformation is anisomorphism.As an example (see also next definition), when the exchange transformation Ex (∆ ∗ ♯ ) is anisomorphism, we simply say that f ∗ and p ♯ commutes – or also that f ∗ commutes with p ♯ .1.1.9 . Under the assumptions of 1.1.6, we will consider the following property:( P -BC) P -base change .– For any cartesian square Y q / / g (cid:15) (cid:15) ∆ X f (cid:15) (cid:15) T p / / S such that p is a P -morphism, the exchange transformation Ex (∆ ∗ ♯ ) : q ♯ g ∗ → f ∗ p ♯ is an isomorphism. Definition . A P -fibred category over S is a pre- P -fibred category M over S whichsatisfies the property of P -base change. Example . Consider the notations of example 1.1.4. Then the transitivity propertyof pullbacks of morphisms in P amounts to say that P / ? satisfies the P -base change property.Thus, P / ? is in fact a P -fibred category, called the canonical P -fibred category . Definition . A P -fibred category M over S is complete if, for any morphism f : T → S , the pullback functor f ∗ : M ( S ) → M ( T ) admits a right adjoint f ∗ : M ( S ) → M ( T ). Remark . In the case where P is the class of isomorphisms a P -fibred category iswhat we usually call a bifibred category over S . Example . The pre- Sm -fibred category H • of example 1.1.5 is a complete Sm -fibredcategory according to [ MV99 ].1.1.15 . Exchange structures II .– Let M be a complete P -fibred category. Consider a com-mutative square Y q / / g (cid:15) (cid:15) ∆ X f (cid:15) (cid:15) T p / / S. We obtain an exchange transformation: Ex (∆ ∗∗ ) : p ∗ f ∗ ad ( g ∗ ,g ∗ ) −−−−−−→ g ∗ g ∗ p ∗ f ∗ = g ∗ q ∗ f ∗ f ∗ ad ′ ( f ∗ ,f ∗ ) −−−−−−→ g ∗ q ∗ . Assume moreover that p and q are P -morphism. Then we can check that Ex (∆ ∗∗ ) is the transposeof the exchange Ex (∆ ∗ ♯ ). Thus, when ∆ is cartesian and p is a P -morphism, Ex (∆ ∗∗ ) is anisomorphism according to ( P -BC).We can also define an exchange transformation: Ex (∆ ♯ ∗ ) : p ♯ g ∗ ad ( f ∗ ,f ∗ ) −−−−−−→ f ∗ f ∗ p ♯ g ∗ Ex (∆ ∗ ♯ ) − −−−−−−−→ f ∗ q ♯ g ∗ g ∗ ad ′ ( g ∗ ,g ∗ ) −−−−−−→ f ∗ q ♯ . Remark . As in remark 1.1.7, we obtain coherence results for these exchange transfor-mations.First with respect to the identifications of the kind f ∗ g ∗ = ( gf ) ∗ , ( f g ) ∗ = f ∗ g ∗ , ( f g ) ♯ = f ♯ g ♯ . In other words, f ∗ commutes with p ♯ . Secondly when several exchange transformations of different kind are involved. As an example,we consider the following commutative diagram in S : Y q , , ❳❳❳❳❳❳❳❳❳❳ Γ ′ Z h (cid:15) (cid:15) q ′ ❢❢❢❢❢❢❢❢❢❢ q ′ ❳❳❳❳ + + ❳❳❳❳ Θ X f (cid:15) (cid:15) ∆ Y q ❢❢❢❢❢ ❢❢❢❢❢ g (cid:15) (cid:15) T p ❲❲❲❲❲ + + ❲❲❲❲❲ Γ Q p ′ ❣❣❣❣ ❣❣❣❣ p ′ + + ❲❲❲❲❲❲❲❲❲❲ ST p ❣❣❣❣❣❣❣❣❣❣ Then the following diagram of natural transformations is commutative: q ♯ g ∗ p ′∗ Ex (∆ ∗ ♯ ) / / Ex (Θ ∗∗ ) (cid:15) (cid:15) f ∗ p ♯ p ′∗ Ex (Γ ♯ ∗ ) ( ( ◗◗◗◗◗◗◗◗ q ♯ q ′∗ h ∗ Ex (Γ ′ ♯ ∗ ) ( ( ◗◗◗◗◗◗◗◗ f ∗ p ∗ p ′ ♯Ex (∆ ∗∗ ) (cid:15) (cid:15) q ∗ q ′ ♯ h ∗ Ex (Θ ∗ ♯ ) / / q ∗ g ∗ p ′ ♯ We left the verification to the reader (it is analogous to that of Remark 1.1.7 except that it involvesalso to the compatibility of the unit and counit of an adjunction).
Definition . Let M be a complete P -fibred category. Consider a commutative squarein S Y q / / g (cid:15) (cid:15) ∆ X f (cid:15) (cid:15) T p / / S. We will say that ∆ is M -transversal if the exchange transformation Ex (∆ ∗∗ ) : p ∗ f ∗ → g ∗ q ∗ of 1.1.15 is an isomorphism.Given an admissible class of morphisms Q in S , we say that M has the transversality (resp. cotransversality ) property with respect to Q -morphisms , if, for any cartesian square ∆ as abovesuch that f is in Q (resp. p is in Q ), ∆ is M -transversal. Remark . Assume S is a sub-category of the category of schemes. When Q is theclass of smooth morphisms (resp. proper morphisms), the cotransversality (resp. transversality)property with respect to Q is usually called the smooth base change property (resp. proper basechange property ). See also Definition 2.2.13.According to Paragraph 1.1.15, we derive the following consequence of our axioms: Proposition . Any complete P -fibred category has the cotransversality property withrespect to P . Let us note for future reference the following corollary:
Corollary . If M is a P -fibred category, then, for any monomorphism j : U → S inP, the functor j ♯ is fully faithful. If moreover M is complete, then the functor j ∗ is fully faithfulas well. Proof.
Because j is a monomorphism, we get a cartesian square in S : U ∆ U j (cid:15) (cid:15) U j / / S. . GENERAL DEFINITIONS AND AXIOMATIC 7 Remark that Ex (∆ ∗ ♯ ) : 1 → j ∗ j ♯ is the unit of the adjunction ( j ♯ , j ∗ ). Thus the P -base changeproperty shows that j ♯ is fully faithful.Assume M is complete. We remark similarly that Ex (∆ ∗∗ ) : j ∗ j ∗ → j ∗ , j ∗ ). Thus, the above proposition shows readily that j ∗ is fully faithful. (cid:3) Monoidal structures.
Let C at ⊗ be the sub-2-category of C at made of symmetric monoidalcategories whose 1-morphisms are (strong) symmetric monoidal functors and 2-morphisms aresymmetric monoidal transformations. Definition . A monoidal pre- P -fibred category over S is a 2-functor M : S → C at ⊗ such that M is a pre- P -fibred category.In other words, M is a pre- P -fibred category such that each of its fibers M ( S ) is endowedwith a structure of a monoidal category, and any pullback morphism f ∗ is monoidal, with theobvious coherent structures. For an object S of S , we will usually denote by ⊗ S (resp. S ) thetensor product (resp. unit) of M ( S ).In particular, we then have the following natural isomorphisms: • for a morphism f : T → S in S , and objects M , N of M ( S ), f ∗ ( M ) ⊗ T f ∗ ( N ) ∼ −→ f ∗ ( M ⊗ S N ); • for a morphism f : T → S in S , f ∗ ( S ) ∼ −→ T . Convention . As in convention 1.1.3, we will generally consider that these structuralisomorphisms are identities.
Example . Consider the notations of example 1.1.4.Using the properties (Pb) and (Pc) of P (see 1.0), for two S -objects X and Y in P /S , thecartesian product X × S Y is an object of P /S . This defines a symmetric monoidal structure on P /S with unit the trivial S -object S . Moreover, the functor f ∗ defined in loc. cit. is monoidal.Thus, the pre- P -fibred category P / ? is in fact monoidal.1.1.24 . Monoidal exchange structures I . Let M be a monoidal pre- P -fibred category M over S . Consider a P -morphism f : T → S , and M (resp. N ) an object of M ( T ) (resp. M ( S )).We get a morphism in M ( S ) Ex ( f ∗ ♯ , ⊗ ) : f ♯ ( M ⊗ T f ∗ ( N )) −→ f ♯ ( M ) ⊗ S N as the composition f ♯ ( M ⊗ T f ∗ ( N )) → f ♯ ( f ∗ f ♯ ( M ) ⊗ T f ∗ ( N )) ≃ f ♯ f ∗ ( f ♯ ( M ) ⊗ S N ) → f ♯ ( M ) ⊗ S N .
This map is natural in M and N . It will be called the exchange transformation between f ♯ and ⊗ T . Remark also that the functor f ♯ , as a left adjoint of a symmetric monoidal functor, is colaxsymmetric monoidal: for any objects M and N of M ( T ), there is a canonical morphism(1.1.24.1) f ♯ ( M ) ⊗ S f ♯ ( N ) → f ♯ ( M ⊗ T N )natural in M and N , as well as a natural map(1.1.24.2) f ♯ ( T ) → S . Remark . As in remark 1.1.7, the preceding exchange transformations satisfy a coher-ence condition for composable morphisms W g −→ T f −→ S . We get in fact a commutative diagram:( f g ) ♯ (cid:0) M ⊗ S ( f g ) ∗ ( N ) (cid:1) Ex (( fg ) ∗ ♯ , ⊗ ) / / (cid:0) ( f g ) ♯ ( M ) (cid:1) ⊗ W Nf ♯ g ♯ (cid:0) M ⊗ S g ∗ f ∗ ( N ) (cid:1) Ex ( g ∗ ♯ , ⊗ ) / / f ♯ (cid:0) g ♯ ( M ) ⊗ T f ∗ ( N ) (cid:1) Ex ( f ∗ ♯ , ⊗ ) / / (cid:0) f ♯ g ♯ ( M ) (cid:1) ⊗ W N As in remark 1.1.16, there is also a coherence relation when different kinds of exchange transfor-mations are involved. Consider a commutative square in S Y q / / g (cid:15) (cid:15) ∆ X f (cid:15) (cid:15) T p / / S such that p and q are P -morphisms and put h = f ◦ q = p ◦ g . Then the following diagram iscommutative: q ♯ g ∗ ( M ⊗ T p ∗ N ) Ex (∆ ∗ ♯ ) / / f ∗ p ♯ ( M ⊗ T p ∗ N ) Ex ( p ∗ ♯ , ⊗ ) / / f ∗ ( p ♯ M ⊗ S N ) q ♯ ( g ∗ M ⊗ Y q ∗ f ∗ N ) Ex ( q ∗ ♯ , ⊗ ) / / ( q ♯ g ∗ M ) ⊗ X f ∗ N Ex (∆ ∗ ♯ ) / / ( f ∗ p ♯ M ) ⊗ X f ∗ N We left the verification to the reader.1.1.26 . Under the assumptions of 1.1.24, we will consider the following property:( P -PF) P -projection formula .– For any P -morphism f : T → S the exchange transformation Ex ( f ♯ , ⊗ T ) : f ♯ ( M ⊗ T f ∗ ( N )) → f ♯ ( M ) ⊗ S N is an isomorphism for all M and N . Definition . A monoidal P -fibred category over S is a monoidal pre- P -fibred cate-gory M : S op → C at ⊗ over S which satisfies the P -projection formula. Example . Consider the canonical monoidal weak P -fibred category P / ? (see example1.1.23). The transitivity property of pullbacks implies readily that P / ? satisfies the property( P -PF). Thus, P / ? is in fact a monoidal P -fibred category called canonical . Definition . A monoidal P -fibred category M over S is complete if it satisfies thefollowing conditions:(1) M is complete as a P -fibred category.(2) For any object S of S , the monoidal category M ( S ) is closed (i.e. has an internal Hom).In this case, we will usually denote by Hom S the internal Hom in M ( S ), so that we havenatural bijections Hom M ( S ) ( A ⊗ S B, C ) ≃ Hom M ( S ) ( A, Hom S ( B, C )) . Example . The P -fibred category H • of example 1.1.14 is in fact a complete monoidal P -fibred category. The tensor product is given by the smash product (see [ MV99 ]).1.1.31 . Monoidal exchange structures II .– Let M be a complete monoidal P -fibred category.Consider a morphism f : T → S in S . Then we obtain an exchange transformation: Ex ( f ∗∗ , ⊗ S ) : ( f ∗ M ) ⊗ S N ad ( f ∗ ,f ∗ ) −−−−−−→ f ∗ f ∗ (cid:0) ( f ∗ M ) ⊗ S N (cid:1) = f ∗ (cid:0) ( f ∗ f ∗ M ) ⊗ T f ∗ N (cid:1) ad ′ ( f ∗ ,f ∗ ) −−−−−−→ f ∗ ( M ⊗ T f ∗ N ) . . GENERAL DEFINITIONS AND AXIOMATIC 9 Remark . As in remark 1.1.25, these exchange transformations are compatible withthe identifications ( f g ) ∗ = f ∗ g ∗ and ( f g ) ∗ = g ∗ f ∗ .Moreover, there is a coherence relation when composing the exchange transformations of the kind Ex ( f ∗∗ , ⊗ ) with exchange transformations of the kind Ex (∆ ∗∗ ) as in loc. cit. Finally, note another kind of coherence relations involving Ex ( f ∗∗ , ⊗ ), Ex (∆ ∗ ♯ ) (resp. Ex ( f ∗ ♯ , ⊗ ))and Ex (∆ ♯ ∗ ).We left the formulation of these coherence relations to the reader, on the model of the precedingones.1.1.33 . Monoidal exchange structures III .– Let M be a complete monoidal P -fibred categoryand f : T → S be a morphism in S .Because f ∗ is monoidal, we get by adjunction a canonical isomorphism Hom S ( M, f ∗ N ) → f ∗ Hom T ( f ∗ M, N ) . Assume that f is a P -morphism. Then from the P -projection formula, we get by adjunction twocanonical isomorphisms: f ∗ Hom S ( M, N ) → Hom T ( f ∗ M, f ∗ N ) , Hom S ( f ♯ M, N ) → f ∗ Hom T ( M, f ∗ N )These isomorphisms are generically called exchange isomorphisms .1.1.c. Geometric sections. . Consider a weak P -fibred category M .Let S be a scheme. For any P -morphism p : X → S , we put M S ( X ) := p ♯ ( X ). Accordingto our conventions, this object is identified with p ♯ p ∗ ( S ). In particular, it defines a covariantfunctor M S : P /S → M ( S ) . Consider a cartesian square in S Y g / / q (cid:15) (cid:15) ∆ X p (cid:15) (cid:15) T f / / S such that p is a P -morphism. With the notations of example 1.1.4, Y = f ∗ ( X ). Then we get anatural exchange transformation Ex ( M T , f ∗ ) : M T ( f ∗ ( X )) = q ♯ ( Y ) = q ♯ g ∗ ( X ) Ex (∆ ∗ ♯ ) −−−−−→ f ∗ p ♯ ( X ) = f ∗ M S ( X ) . In other words, M defines a lax natural transformation P / ? → M .Consider P -morphisms p : X → S , q : Y → S . Let Z = X × S Y be the cartesian productand consider the cartesian square: Z p ′ / / q ′ (cid:15) (cid:15) Θ Y q (cid:15) (cid:15) X p / / S. Using the exchange transformations of the preceding paragraph, we get a canonical morphism Ex ( M S , ⊗ S ) : M S ( X × S Y ) −→ M S ( X ) ⊗ S M S ( Y )as the composition M S ( X × S Y ) = p ♯ q ′ ♯ p ′∗ ( Y ) Ex (Θ ∗ ♯ ) −−−−−→ p ♯ p ∗ q ♯ ( Y ) p ♯ ( X ⊗ X p ∗ q ♯ ( Y )) Ex ( p ♯ , ⊗ X ) −−−−−−−→ p ♯ ( X ) ⊗ S q ♯ ( Y ) = M S ( X ) ⊗ S M S ( Y ) . In other words, the functor M S is symmetric colax monoidal.Remark finally that for any P -morphism p : T → S , and any P -object Y over T , we obtainaccording to convention an identification p ♯ M T ( Y ) = M S ( Y ). Definition . Given a monoidal pre- P -fibred category M over S , the lax naturaltransformation M : P / ? → M constructed above will be called the geometric sections of M .The following lemma is obvious from the definitions above: Lemma . let M be a monoidal P -fibred category. Let M : P / ? → M be the geometricsections of M . Then:(i) For any morphism f : T → S in S , the exchange Ex ( M T , f ∗ ) defined above is anisomorphism.(ii) For any scheme S , the exchange Ex ( M S , ⊗ S ) defined above is an isomorphism.In other words, M is a cartesian functor and M S is a (strong) symmetric monoidal functor. . In the situation of the lemma we thus obtain the following identifications: • f ∗ M S ( X ) ≃ M T ( X × S T ), • p ♯ M T ( Y ) ≃ M S ( Y ), • M S ( X × S Y ) ≃ M S ( X ) ⊗ S M S ( Y ),whenever it makes sense.1.1.d. Twists. . Let M be a pre- P -fibred category of S . Recall that a cartesian section of M ( i.e. acartesian functor A : S → M ) is the data of an object A S of M ( S ) for each object S of S andof isomorphisms f ∗ ( A S ) ∼ −→ A T for each morphism f : T → S , subject to coherence identities; see [ SGA1 , Exp. VI].If M is monoidal, the tensor product of two cartesian sections is defined termwise. Definition . let M be a monoidal pre- P -fibred category. A set of twists τ for M is a set of cartesian sections of M stable by tensor product. For short, we say also that M is τ -twisted .1.1.40 . Let M be a monoidal pre- P -fibred category endowed with a set of twists τ .The tensor product on τ induces a monoid structure that we will denote by + (the unit objectof τ will be written 0).Consider an object i ∈ τ . For any object S of S , we thus obtain an object i S in M ( S )associated with i . Given any object M of M ( S ), we simply put: M { i } = M ⊗ S i S and call this object the twist of M by i . We have, by definition: M { } = M .For any i, j ∈ τ , and any object M of M ( S ), we obtain M { i + j } = ( M { i } ) { j } – using thestructural associativity isomorphism of the monoidal structure. Given a morphism f : T → S , anobject M of M ( S ) and a twist i ∈ τ , we also obtain f ∗ ( M { i } ) = ( f ∗ M ) { i } . If f is a P -morphism,for any object M of M ( T ), the exchange transformation Ex ( f ∗ ♯ , ⊗ T ) of paragraph 1.1.6 inducesa canonical morphism Ex ( f ♯ , { i } ) : f ♯ ( M { i } ) → ( f ♯ M ) { i } . We will say that f ♯ commutes with τ -twists (or simply twists when τ is clear) if for any i ∈ I , thenatural transformation Ex ( f ♯ , { i } ) is an isomorphism. Definition . Let M be a monoidal pre- P -fibred category with a set of twists τ and M : P / ? → M be the geometric sections of M .We say M is τ -generated if for any object S of S , the family of functorsHom M ( S ) ( M S ( X ) { i } , − ) : M ( S ) → S et indexed by a P -object X/S and an element i ∈ τ is conservative.Of course, we do not exclude the case where τ is trivial, but then, we shall simply say that M is geometrically generated . . GENERAL DEFINITIONS AND AXIOMATIC 11 We shall frequently use the following proposition to characterize complete monoidal P -fibredcategories over S : Proposition . Let M : S → C at ⊗ be a -functor such that:(1) For any P -morphism f : T → S , the pullback functor f ∗ : M ( S ) → M ( T ) is monoidaland admits a left adjoint f ♯ in C .(2) For any morphism f : T → S , the pullback functor f ∗ : M ( S ) → M ( T ) admits a rightadjoint f ∗ in C .We consider M as a monoidal weak P -fibred category and denote by M : P / ? → M its associatedgeometric sections. Suppose given a set of twists τ such that M is τ -generated. Then, the followingassertions are equivalent:(3) M satisfies properties ( P -BC) and ( P -PF)( i.e. M is a complete monoidal P -fibred category.)(3’) (a) M is a cartesian functor.(b) For any object S of S , M S is (strongly) monoidal.(c) For any P -morphism f , f ♯ commutes with τ -twists. Proof. ( i ) ⇒ ( ii ): This is obvious (see Lemma 1.1.36).( ii ) ⇒ ( i ): We use the following easy lemma: Lemma . Let C and C be categories, F, G : C → C be two left adjoint functors and η : F → G be a natural transformation. Let G be a class of objects of C which is generating inthe sense that the family of functors Hom C ( X, − ) for X in G is conservative.Then the following conditions are equivalent:(1) η is an isomorphism.(2) For all X in G , η X is an isomorphism. Given this lemma, to prove ( P -BC), we are reduced to check that the exchange transformation Ex (∆ ∗ ♯ ) is an isomorphism when evaluated on an object M T ( U ) { i } for an object U of P /T anda twist i ∈ τ . Then it follows from (ii), 1.1.40 and example 1.1.11. To prove ( P -PF), we proceed in two steps first proving the case M = M T ( U ) { i } and N anyobject of M ( S ) using the same argument as above with the help of 1.1.28. Then, we can provethe general case by another application of the same argument. (cid:3) Suppose given a complete monoidal P -fibered category M with a set of twists τ . Let f : T → S be a morphism of S . Then the exchange transformation 1.1.31 induces for any i ∈ τ anexchange transformation Ex ( f ∗ , { i } ) : ( f ∗ M ) { i } → f ∗ ( M { i } ) . Definition . In the situation above, we say that f ∗ commutes with τ -twists (or simplywith twists when τ is clear) if for any i ∈ τ , the exchange transformation Ex ( f ∗ , { i } ) is anisomorphism.It will happen frequently that twists are ⊗ -invertible. Then f ∗ commutes with twists as itsright adjoint does. P -fibred categories. General case. . Consider two P -fibred categories M and M ′ over S , as well as a cartesian functor ϕ ∗ : M → M ′ between the underlying fibred categories: for any object S of S , we have a functor ϕ ∗ S : M ( S ) → M ′ ( S ) , The cautious reader will use remark 1.1.7 to check that the corresponding map M X ( U × T Y ) { i } → M X ( U × T Y ) { i } is the identity. and for any map f : T → S in S , we have an isomorphism of functors c f M ( S ) ☛☛☛☛ (cid:1) (cid:9) c f ϕ ∗ S / / f ∗ (cid:15) (cid:15) M ′ ( S ) f ∗ (cid:15) (cid:15) M ( T ) ϕ ∗ T / / M ′ ( T ) c f : f ∗ ϕ ∗ S ∼ −→ ϕ ∗ T f ∗ (1.2.1.1)satisfying some cocycle condition with respect to composition in S .For any P -morphism p : T → S , we construct an exchange morphism Ex ( p ♯ , ϕ ∗ ) : p ♯ ϕ ∗ T −→ ϕ ∗ S p ♯ as the composition p ♯ ϕ ∗ T ad ( p ♯ ,p ∗ ) −−−−−−→ p ♯ ϕ ∗ T p ∗ p ♯ c − p −−→ p ♯ p ∗ ϕ ∗ S p ♯ ad ′ ( p ♯ ,p ∗ ) −−−−−−→ ϕ ∗ S p ♯ . Definition . Consider the situation above. We say that the cartesian functor ϕ ∗ : M → M ′ is a morphism of P -fibred categories if, for any P -morphism p , the exchange transformation Ex ( p ♯ , ϕ ∗ ) is an isomorphism. Example . If M is a monoidal P -fibred category, then the geometric sections M : P / ? → M is a morphism of P -fibred categories (1.1.36). Definition . Let M and M ′ be two complete P -fibred categories. A morphism ofcomplete P -fibred categories is a morphism of P -fibred categories ϕ ∗ : M → M ′ such that, for any object S of S , the functor ϕ ∗ S : M ( S ) → M ′ ( S ) has a right adjoint ϕ ∗ ,S : M ′ ( S ) → M ( S ) . When we want to indicate a notation for the right adjoint of a morphism as above, we use thewriting ϕ ∗ : M ⇄ N : ϕ ∗ the left adjoint being in the left hand side.1.2.5 . Exchange structures III . Consider a morphism ϕ ∗ : M → M ′ of complete P -fibredcategories.Then for any morphism f : T → S in S , we define exchange transformations Ex ( ϕ ∗ , f ∗ ) : ϕ ∗ S f ∗ −→ f ∗ ϕ ∗ T , (1.2.5.1) Ex ( f ∗ , ϕ ∗ ) : f ∗ ϕ ∗ ,S −→ ϕ ∗ ,T f ∗ , (1.2.5.2)as the respective compositions ϕ ∗ S f ∗ ad ( f ∗ ,f ∗ ) −−−−−−→ f ∗ f ∗ ϕ ∗ S f ∗ ≃ f ∗ ϕ ∗ T f ∗ f ∗ ad ′ ( f ∗ ,f ∗ ) −−−−−−→ f ∗ ϕ ∗ T ,f ∗ ϕ ∗ ,S ad ( f ∗ ,f ∗ ) −−−−−−→ f ∗ ϕ ∗ ,S f ∗ f ∗ ≃ f ∗ f ∗ ϕ ∗ ,T f ∗ ad ′ ( f ∗ ,f ∗ ) −−−−−−→ ϕ ∗ ,T f ∗ . Remark . We warn the reader that ϕ ∗ : M ′ → M is not a cartesian functor in general,meaning that the exchange transformation Ex ( f ∗ , ϕ ∗ ) is not necessarily an isomorphism, evenwhen f is a P -morphism. . GENERAL DEFINITIONS AND AXIOMATIC 13 Monoidal case.
Definition . Let M and M ′ be monoidal P -fibred categories.A morphisms of monoidal P -fibred categories is a morphism ϕ ∗ : M → M ′ of P -fibredcategories such that for any object S of S , the functor ϕ ∗ S : M ( X ) → N ( S ) has the structure ofa (strong) symmetric monoidal functor, and such that the structural isomorphisms (1.2.1.1) areisomorphisms of symmetric monoidal functors.In the case where M and M ′ are complete monoidal P -fibred categories, we shall say thatsuch a morphism ϕ ∗ is a morphism of complete monoidal P -fibred categories if ϕ ∗ is also amorphism of complete P -fibred categories. Remark . If we denote by M ( − , M ) and M ( − , M ′ ) the geometric sections of M and M ′ respectively, we have a natural identification: ϕ ∗ S ( M S ( X, M )) ≃ M S ( X, M ′ ) . . Monoidal exchange structures IV . Consider a a morphism ϕ ∗ : M → M ′ of completemonoidal P -fibred categories. For objects M (resp. N ) of M ( S ) (resp. M ′ ( S )), we define anexchange transformation Ex ( ϕ ∗ , ⊗ , ϕ ∗ ) : ( ϕ ∗ ,S M ) ⊗ S N → ϕ ∗ ,S ( M ⊗ T ϕ ∗ S N ) , natural in M and N , as the following composite( ϕ ∗ ,S M ) ⊗ S N ad ( ϕ ∗ ,ϕ ∗ ) −−−−−−→ ϕ ∗ ,S ϕ ∗ S (( ϕ ∗ ,S M ) ⊗ S N )= ϕ ∗ ,S (( ϕ ∗ S ϕ ∗ ,S M ) ⊗ T ϕ ∗ S N ) ad ′ ( ϕ ∗ ,ϕ ∗ ) −−−−−−−→ ϕ ∗ ,S ( M ⊗ T ϕ ∗ S N ) . As in remark 1.1.32, we get coherence relations between the various exchange transformationsassociated with a morphism of monoidal P -fibred categories. We left the formulation to thereader.Note also that, because ϕ ∗ is monoidal, we get by adjunction a canonical isomorphism: Hom M ( S ) ( M, ϕ ∗ ,S M ′ ) ∼ −→ ϕ ∗ ,S Hom M ′ ( S ) ( ϕ ∗ S M, M ′ ) . . Consider two monoidal P -fibred categories M , M ′ and a cartesian functor ϕ ∗ : M → M ′ such that, for any scheme S , ϕ ∗ S : M ( S ) → M ′ ( S ) is monoidal.Given a cartesian section K = ( K S ) S ∈ S of M , we obtain for any morphism f : T → S in S a canonical map f ∗ ϕ ∗ S ( K S ) = ϕ ∗ T ( f ∗ ( K S )) → ϕ ∗ T ( K T )which defines a cartesian section of M ′ , which we denote by ϕ ∗ ( K ). Definition . Let ( M , τ ) and ( M ′ , τ ′ ) be twisted monoidal P -fibred categories. Let ϕ ∗ : M → M ′ be a cartesian functor as above (resp. a morphism of monoidal P -fibred categories).We say that ϕ ∗ : ( M , τ ) → ( M ′ , τ ′ ) is compatible with twists if for any i ∈ τ , the cartesiansection ϕ ∗ ( i ) is in τ ′ (up to isomorphism in M ′ ).In particular, ϕ ∗ induces a morphism of monoids τ → τ ′ (if we consider the isomorphism classesof objects). Moreover, for any object K of M ( S ) and any twist i ∈ τ , we get an identification: ϕ ∗ S ( K { i } ) ≃ ( ϕ ∗ S K ) { ϕ ∗ ( i ) } . Moreover, the exchange transformation Ex ( ϕ ∗ , ⊗ ) induces an exchange: Ex ( ϕ ∗ , { i } ) : ϕ ∗ ,S ( K ) { i } → ϕ ∗ ,S (cid:0) K { ϕ ∗ ( i ) } (cid:1) . When this transformation is an isomorphism for any twist i ∈ τ , we say that ϕ ∗ commutes withtwists. Remark . In every examples, the morphism τ → τ ′ will be an explicit injection and wewill cancel it in the notations of twists.Note finally that Lemma 1.1.43 allows to prove, as for Proposition 1.1.42, the following usefullemma: Lemma . Consider two complete monoidal P -fibred categories M , M ′ and denote by M ( − , M ) and M ( − , M ′ ) their respective geometric sections. Let ϕ ∗ : M → M ′ be a cartesianfunctor such that(1) For any scheme S , ϕ ∗ S : M ( S ) → M ′ ( S ) is monoidal.(2) For any scheme S , ϕ ∗ S admits a right adjoint ϕ ∗ ,S .Assume M (resp. M ′ ) is τ -generated (resp. τ ′ -twisted) and ϕ ∗ commutes with twists. Then thefollowing conditions are equivalent:(3) ϕ ∗ is a morphism of complete monoidal P -fibred categories.(3’) For any object X of P /S , the exchange transformation (cf. 1.2.1) ϕ ∗ M S ( X, M ) → M S ( X, M ′ ) is an isomorphism. P -fibred categories. Abstract definition. . We fix a sub-2-category C of C at with the following properties :(1) the 2-functor C at → C at ′ , A A op sends C to C ′ , where C ′ denotes the 2-category whose objects and maps are those of C and whose 2-morphisms are the 2-morphisms of C , put in the reverse direction.(2) C is closed under adjunction: for any functor u : A → B in C , if a functor v : B → A isa right adjoint or a left adjoint to u , then v is in C .(3) the 2-morphisms of C are closed by transposition: if u : A ⇄ B : v and u ′ : A ⇄ B : v ′ are two adjunctions in C (with the left adjoints on the left hand side), a natural trans-formation u → u ′ is in C if and only if the corresponding natural transformation v ′ → v is in C .We can then define and manipulate C -structured P -fibred categories as follows. Definition . A C -structured P -fibred category (resp. C -structured complete P -fibredcategory ) M over S is simply a P -fibred category (resp. a complete P -fibred category) whoseunderlying 2-functor M : S op → C at factors through C .If M and M ′ are C -structured fibred categories over S , a cartesian functor M → M ′ is C -structured if the functors M ( S ) → M ′ ( S ) are in C for any object S of S , and if all the structural2-morphisms (1.2.1.1) are in C as well. Definition . A morphism of C -structured P -fibred categories (resp. C -structuredcomplete P -fibred categories) is a morphism of P -fibred categories (resp. of complete P -fibredcategories) which is C -structured as a cartesian functor.1.3.4 . Consider a 2-category C as in the paragraph 1.3.1. In order to deal with the monoidalcase, we will consider also a sub-2-category C ⊗ of C such that:(1) The objects of C ⊗ are objects of C equipped with a symmetric monoidal structure;(2) the 1-morphisms of C ⊗ are exactly the 1-morphisms of C which are symmetric monoidalas functors;(3) the 2-morphisms of C ⊗ are exactly the 2-morphisms of C which are symmetric monoidalas natural transformations.Note that C ⊗ satisfies condition (1) of 1.3.1, but it does not satisfies conditions (2) and (3) ingeneral. Instead, we get the following properties: See the following sections for examples. . GENERAL DEFINITIONS AND AXIOMATIC 15 (2 ′ ) If u : A → B is a functor in C ⊗ , a right (resp. left) adjoint v is a lax (resp. colax)monoidal functor in C .(3 ′ ) Consider adjunctions u : A ⇄ B : v and u ′ : A ⇄ B : v ′ in C (with the left adjoints on the left hand side). If u → u ′ (resp. v → v ′ ) is a 2-morphism in C ⊗ then v → v ′ (resp. u → u ′ ) is a 2-morphism in C which is a symmetricmonoidal transformation of lax (resp. colax) monoidal functors.We thus adopt the following definition: Definition . A ( C , C ⊗ ) -structured monoidal P -fibred category (resp. a ( C , C ⊗ ) -structuredcomplete monoidal P -fibred category ) is simply a monoidal P -fibred category (resp. a completemonoidal P -fibred category) whose underlying 2-functor M : S op → C at ⊗ factors through C ⊗ .Morphisms of such objects are defined in the same way.Note that, with the hypothesis made on C , all the exchange natural transformations definedin the preceding paragraphs lie in C and satisfy the appropriate coherence property with respectto the monoidal structure.1.3.b. The abelian case. . Let A b be the sub-2-category of C at made of the abelian categories, with the additivefunctors as 1-morphisms, and the natural transformations as 2-morphisms. Obviously, it satisfiesproperties of 1.3.1. When we will apply one of the definitions 1.3.2, 1.3.3 to the case C = A b ,we will use the simple adjective abelian for A b -structured. This allows to speak of morphisms ofabelian P -fibred categories. Let A b ⊗ be the sub-2-category of A b made of the abelian monoidal categories, with 1-morphisms the symmetric monoidal additive functors and 2-morphisms the symmetric monoidalnatural transformations. It satisfies the hypothesis of paragraph 1.3.4. When we will apply def-inition 1.3.5 to the case of ( A b, A b ⊗ ), we will use the simple expression abelian monoidal for( A b, A b ⊗ )-structured monoidal. This allows to speak of morphisms of abelian monoidal P -fibredcategories. Lemma . Consider an abelian P -fibred category A such that for any object S of S , A ( S ) is a Grothendieck abelian category. Then the following conditions are equivalent:(i) A is complete.(ii) For any morphism f : T → S in S , f ∗ commutes with sums.If in addition, A is monoidal, the following conditions are equivalent:(i ′ ) A is monoidal complete.(ii ′ ) (a) For any morphism f : T → S in S , f ∗ is right exact.(b) For any object S of S , the bifunctor ⊗ S is right exact. In view of this lemma, we adopt the following definition:
Definition . A Grothendieck abelian (resp. Grothendieck abelian monoidal) P -fibredcategory A over S is an abelian P -fibred category which is complete (resp. complete monoidal)and such that for any scheme S , A ( S ) is a Grothendieck abelian category. Remark . Let A be a Grothendieck abelian monoidal P -fibred category. Convention-ally, we will denote by M S ( − , A ) its geometric sections. Note that if A is τ -twisted, then anyobject of A is a quotient of a direct sum of objects of shape M S ( X, A ) { i } for a P -object X/S and a twist i ∈ τ .1.3.10 . Consider an abelian category A which admits small sums. Recall the following defi-nition:An object X of T is finitely presented if the functor Hom T ( X, − ) commutes with small filtering For any object a , a ′ in A , F is lax if there exists a structural map F ( a ) ⊗ F ( a ′ ) (1) −−→ F ( a ⊗ a ′ ) satisfyingcoherence relations (see [ Mac98 , XI. 2]). Colax is defined by reversing the arrow (1). colimits. A essentially small G of objects of A is called generating if for any object A of A thereexists an epimorphism of the form: M i ∈ I G i → A where ( G i ) i ∈ I is a family of objects if G . Definition . Let A be an abelian P -fibred category over S .Given a set of twists τ of A , we say A is finitely τ -presented if for any object S of S , forany P -object X/S and any twist i ∈ τ , the object M S ( X ) { i } is finitely presented and the classof such objects form an essentially small generating family of A ( S ). Remark . The triangulated case. . Let T ri be the sub-2-category of C at made of the triangulated categories, with the tri-angulated functors as 1-morphisms, and the triangulated natural transformations as 2-morphisms.Then T ri satisfies the properties of 1.3.1 (property (2) can be found for instance in [ Ayo07a ,Lemma 2.1.23], and we leave property (3) as an exercise for the reader). When we will apply oneof the definitions 1.3.2, 1.3.3 to the case C = T ri , we will use the simple adjective triangulated for T ri -structured. This allows to speak of morphisms of triangulated P -fibred categories .Let T ri ⊗ be the sub-2-category of T ri made of the triangulated monoidal categories, with 1-morphisms the symmetric monoidal triangulated functors and 2-morphisms the symmetric monoidalnatural transformations. It satisfies the hypothesis of paragraph 1.3.4. When we will apply def-inition 1.3.5 to the case of ( T ri, T ri ⊗ ), we will use the expression triangulated monoidal for( T ri, T ri ⊗ )-structured monoidal. This allows to speak of morphisms of triangulated monoidal P -fibred categories . Convention . The set of twists of a triangulated monoidal P -fibred category T willalways be of the form Z × τ where the first factor corresponds to the cartesian sections defined bysuspensions [ n ], n ∈ Z . In the notation, we shall often make the abuse of only indicating τ . Inparticular, the expression T is τ -generated will mean conventionally that T is ( Z × τ )-generatedin the sense of definition 1.1.41.1.3.15 . Consider a triangulated category T which admits small sums. Recall the followingdefinitions:An object X of T is called compact if the functor Hom T ( X, − ) commutes with small sums. Aclass G of objects of T is called generating if the family of functor Hom T ( X [ n ] , − ), X ∈ G , n ∈ Z ,is conservative.The triangulated category T is called compactly generated if there exists a generating set G ofcompact objects of T . This property of being compact has been generalized by A. Neeman tothe property of being α -small for some cardinal α ( cf. [ Nee01 , 4.1.1]) – recall compact= ℵ -small. Then the property of being compactly generated has been generalized by Neeman to theproperty of being well generated ; see [ Kra01 ] for a convenient characterization of well generatedtriangulated categories.
Definition . Let T be a triangulated P -fibred category over S . We say that T is compactly generated (resp. well generated ) if for any object S of S , T ( S ) admits small sumsand is compactly generated (resp. well generated).Given a set of twists τ for T , we say T is compactly τ -generated if it is compactly generatedin the above sense and for any P -object X/S , any twist i ∈ τ , M S ( X ) { i } is compact.1.3.17 . For a triangulated category T which has small sums, given a family G of objectsof T , we denote by hGi the localizing subcategory of T generated by G , i.e. hGi is the smallesttriangulated full subcategory of T which is stable by small sums and which contains all the objectsin G . Recall that, in the case T is well generated (e.g. if T compactly generated), then the family G generates T (in the sense that the family of functors { Hom T ( X, − ) } X ∈G is conservative) if andonly if T = hGi . The following lemma is a consequence of [ Nee01 ]: . GENERAL DEFINITIONS AND AXIOMATIC 17 Lemma . Let T be a triangulated monoidal P -fibred category over S with geometricsections M . Assume T is τ -generated.If T is well generated, then for any object S of S , T ( S ) = h M S ( X ) { i } ; X/S a P -object , i ∈ τ i Moreover, there exists a regular cardinal α such that all the objects of shape M S ( X ) { i } are α -compact. Note finally that the Brown representability theorem of Neeman ( cf. [ Nee01 ]) gives thefollowing lemma (analog of 1.3.7):
Lemma . Consider a well generated triangulated P -fibred category T . Then the fol-lowing conditions are equivalent:(i) T is complete.(ii) For any morphism f : T → S in S , f ∗ commutes with sums.If in addition, T is monoidal, the following conditions are equivalent:(i ′ ) T is monoidal complete.(ii ′ ) (a) For any morphism f : T → S in S , f ∗ is right exact.(b) For any object S of S , the bifunctor ⊗ S is right exact. We finish this section with a proposition which will constitute a useful trick:
Proposition . Consider an adjunction of triangulated categories a : T ⇄ T ′ : b. Assume that T admits a set of compact generators G such that any object in a ( G ) in compact in T ′ . Then b commutes with direct sums. If in addition T ′ is well generated then b admits a rightadjoint. Proof.
The second assertion follows from the first one according to a corollary of the Brownrepresentability theorem of Neeman ( cf. [ Nee01 , 8.4.4]).For the first one, we consider a family ( X i ) i ∈ I of objects of T ′ and prove that the canonicalmorphism ⊕ i ∈ I b ( X i ) → b ( ⊕ i ∈ I X i )is an isomorphism in T . To prove this, it is sufficient to apply the functor Hom T ( G, − ) for anyobject G of G . Then the result is obvious from the assumptions. (cid:3) We shall use often the following standard argument to produce equivalences of triangulatedcategories.
Corollary . Let a : T → T ′ be a triangulated functor between triangulated categories.Assume that the functor a preserves small sums, and that T admits a small set of compactgenerators G , such that a ( G ) form a family of compact objects in T ′ . Then a is fully faithful ifand only if, for any couple of objects G and G ′ in G , the map Hom T ( G, G ′ [ n ]) → Hom T ′ ( a ( G ) , a ( G ′ )[ n ]) is bijective for any integer n . If a is fully faithful, then a is an equivalence of categories if andonly if a ( G ) is a generating family in T ′ . Proof.
Let us prove that this is a sufficient condition. As T is in particular well generated,by the Brown representability theorem, the functor b admits a right adjoint b : T ′ → T . Byvirtue of the preceding proposition, the functor b preserves small sums. Let us prove that a is fullyfaithful. We have to check that, for any object M of T , the map M → b ( a ( M )) is invertible. As a and b are triangulated and preserve small sums, it is sufficient to check this when M runs overa generating family of objects of T (e.g. G ). As G is generating, it is sufficient to prove that themap Hom T ( G, M [ n ]) → Hom T ′ ( a ( G ) , a ( M )[ n ]) = Hom T ′ ( a ( G ) , b ( a ( M ))[ n ]) is bijective for any integer n , which hold then by assumption. The functor a thus identifies T with the localizing subcategory of T ′ generated by a ( G ); if moreover a ( G ) is a generating familyin T ′ , then T ′ = h a ( G ) i , which also proves the last assertion. (cid:3) The model category case. . We shall use Hovey’s book [
Hov99 ] for a general reference to the theory of modelcategories. Note that, following loc. cit. , all the model categories we shall consider will have smalllimits and small colimits.Let M be the sub-2-category of C at made of the model categories, with 1-morphisms the leftQuillen functors and 2-morphisms the natural transformations. When we will apply definition1.3.2 (resp. 1.3.3) to C = M , we will speak of a P -fibred model category for a M -structured P -fibred category M (resp. morphism of P -fibred model categories). Note that according to thedefinition of left Quillen functors, M is then automatically complete.Given a property ( P ) of model categories (like being cofibrantly generated, left and/or rightproper, combinatorial, stable, etc), we will say that a P -fibred model category M over S hasthe property ( P ) if, for any object S of S , the model category M ( S ) has the property ( P ).For the monoidal case, we let M ⊗ be the sub-2-categories of M made of the symmetricmonoidal model categories (see [ Hov99 , Definition 4.2.6]), with 1-morphisms the symmetricmonoidal left Quillen functors and 2-morphisms the symmetric monoidal natural transformations,following the conditions of 1.3.4. When we will apply definition 1.3.5 to the case of ( M , M ⊗ ), wewill speak simply of a monoidal P -fibred model category (resp. morphism of monoidal P -fibredmodel categories ) for a (resp. morphism of) ( M , M ⊗ )-structured monoidal P -fibred category M . Again, M is then monoidal complete. Remark . Let M be a P -fibred model category over S . Then for any P -morphism p : X → Y , the inverse image functor p ∗ : M ( Y ) → M ( X ) has very strong exactness properties:it preserves small limits and colimits (having both a left and a right adjoint), and it preservesweak equivalences, cofibrations, and fibrations. The only non (completely) trivial assertion hereis about the preservation of weak equivalences. For this, one notices first that it preserves trivialcofibrations and trivial fibrations (being both a left Quillen functor and a right Quillen functor). Inparticular, by virtue of Ken Brown Lemma [ Hov99 , Lemma 1.1.12], it preserves weak equivalencesbetween cofibrant (resp. fibrant) objects. Given a weak equivalence u : M → N in M ( Y ), we canfind a commutative square M ′ u ′ / / (cid:15) (cid:15) N ′ (cid:15) (cid:15) M u / / N in which the two vertical maps are trivial fibrations, and where u ′ is a weak equivalence betweencofibrant objects, from which we deduce easily that p ∗ ( u ) is a weak equivalence in M ( X ).1.3.24 . Consider a P -fibred model category M over S . By assumption, we get the followingpairs of adjoint functors:(a) For any morphism f : X → S of S , L f ∗ : Ho( M ( S )) ⇄ Ho( M ( X )) : R f ∗ (b) For any P -morphism p : T → S , the pullback functor L p ♯ : Ho( M ( S )) ⇄ Ho( M ( T )) : L p ∗ = p ∗ = R p ∗ Moreover, the canonical isomorphism of shape ( f g ) ∗ ≃ g ∗ f ∗ induces a canonical isomorphism R ( f g ) ∗ ≃ R g ∗ R f ∗ . In the situation of the P -base change formula 1.1.9, we obtain also that thebase change map L q ♯ L g ∗ → L f ∗ L p ♯ is an isomorphism from the equivalent property of M . Thus, we have defined a complete P -fibredcategory whose fiber over S is Ho( M ( S )). . GENERAL DEFINITIONS AND AXIOMATIC 19 Definition . Given a P -fibred model category M as above, the complete P -fibredcategory defined above will be denoted by Ho( M ) and called the homotopy P -fibred category associated with M .1.3.26 . Assume that M is a monoidal P -fibred model category over S . Then, for any object S of S , Ho( M )( S ) has the structure of a symmetric closed monoidal category; see [ Hov99 ,Theorem 4.3.2]. The (derived) tensor product of Ho( M )( S ) will be denoted by M ⊗ L S N , and the(derived) internal Hom will be written R Hom S ( M, N ), while the unit object will be written S .For any morphism f : T → S in S , the derived functor L f ∗ is symmetric monoidal as followsfrom the equivalent property of its homologue f ∗ .Moreover, for any P -morphism p : T → S and for any object M in Ho( M )( T ) and any object N in Ho( M )( S ), the exchange map of 1.1.24 L p ♯ ( M ⊗ L p ∗ ( N )) → L p ♯ ( M ) ⊗ L N is an isomorphism. Definition . Given a monoidal P -fibred model category M as above, the completemonoidal P -fibred category defined above will be denoted by Ho( M ) and called the homotopymonoidal P -fibred category associated with M . In the present article, we will focus on a particular type of P -fibred category.1.4.1 . Let S be a scheme. Assume S is a full subcategory of the category of S -schemes. Inmost of this work, we will denote by S ft the class of morphisms of finite type in S and by Sm bethe class of smooth morphisms of finite type in S . There is an exception to this rule: throughoutPart 3, S ft will be the class of separated morphisms of finite type in S and Sm will be theclass of separated smooth morphisms of finite type in S . However, the axiomatic which we willpresent in the sequel can be applied identically in each cases so that the reader can freely use therestriction that all morphisms of Sm and S ft are separated.In any case, the classes Sm and S ft are admissible in S in the sense of 1.0 (this is automatic,for instance, if S is stable by pullbacks). Definition . Let P be an admissible class of morphisms in S .A P -premotivic category over S – or simply P -premotivic category when S is clear – isa complete monoidal P -fibred category over S . A morphism of P -premotivic categories is amorphism of complete monoidal P -fibred categories over S .As a particular case, when C is the 2-category T ri of triangulated categories (resp. A b ofabelian categories), a P -premotivic triangulated (resp. abelian) category over S is a ( C , C ⊗ )-structured complete monoidal P -fibred category over S (def. 1.3.5). Morphisms of P -premotivictriangulated (resp. abelian) categories are defined accordingly.We will also say: premotivic for Sm -premotivic and generalized premotivic for S ft -premotivic.The sections of a P -premotivic category will be called premotives . Example . Let S be the category of noetherian schemes of finite dimension.For such a scheme S , recall H • ( S ) is the pointed homotopy category of Morel and Voevodsky; cf. examples 1.1.5, 1.1.14, 1.1.30. Then, according to the fact recalled in these examples the2-functor H • is a geometrically generated premotivic category (recall Definition 1.1.41).For such a scheme S , consider the stable homotopy category SH( S ) of Morel and Voevodsky(see [ Jar00, Ayo07b ]). According to [
Ayo07b ], it defines a triangulated premotivic categorydenoted by SH. Moreover, it is compactly ( Z × Z )-generated in the sense of definition 1.1.41where the first factor refers to the suspension and the second one refers to the Tate twist ( i.e. asa triangulated premotivic category, it is compactly generated by the Tate twists).1.4.4 . Let T be a P -premotivic triangulated category with geometric sections M and τ be aset of twists for T (Definition 1.1.39). Recall from Convention 1.3.14 (resp. and Definition 1.3.16) that T is said to be τ -generated(resp. compactly τ -generated) if for any scheme S , the family of isomorphism of classes of pre-motives of the form M S ( X ) { i } for a P -scheme X over S and a twist i ∈ τ is a set of generators(resp. compact generators) for the triangulated category T ( S ) (in the respective case, we alsoassume T ( S ) admits small sums).Let E be a premotive over S and X be a P -scheme over S . For any ( n, i ) ∈ Z × τ , we definethe cohomology of X in degree n and twist i with coefficients in E as: H n,i T ( X, E ) = Hom T ( S ) (cid:0) M S ( X ) , E { i } ( n ) (cid:1) . The fact T is τ -generated amounts to say that any such premotive E is determined by its coho-mology. Example . The premotivic triangulated category SH of the previous example is com-pactly Z -generated where Z refers to the Tate twist (in other words it is compactly generated byTate twists). Definition . Let M and M ′ be P -premotivic categories.A morphism of P -premotivic categories (or simply a premotivic morphism ) is a morphism ϕ ∗ : M → M ′ of complete monoidal P -fibred categories. We shall also say that ϕ ∗ : M ⇄ M ′ : ϕ ∗ is a premotivic adjunction . When moreover M and M ′ are P -premotivic triangulated (resp.abelian) categories, we will ask ϕ ∗ is a compatible with the triangulated (resp. additive) structure– as in Definition 1.3.3.If we assume that M (resp. M ′ ) is τ -twisted (resp. τ ′ -twisted), we will say as in Definition1.2.11 that ϕ ∗ is compatible with twists if for any i ∈ τ , ϕ ∗ ( i ) belongs up to isomorphism to τ ′ .We say ϕ ∗ is strictly compatible with twists if the induced map ϕ ∗ : τ → τ ′ is bijective.Usually, premotivic categories comes equip with canonical twists (especially the Tate twist)and premotivic morphisms are compatible with twists. Example . With the hypothesis and notations of 1.4.3, we get a premotivic adjunctionΣ ∞ : H • ⇄ SH : Ω ∞ induced by the infinite suspension functor according to [ Jar00 ].1.4.8 . Let T (resp. A ) be a triangulated P -premotivic category with geometric sections M and set of twists τ . For any scheme S , we let T τ,c ( S ) be the smallest triangulated thick subcategory of T ( S ) which contains premotives of shape M S ( S ) { i } (resp. M S ( X, A ) { i } ) for a P -scheme X/S and a twist i ∈ τ . This subcategory is stable by the operations f ∗ , p ♯ and ⊗ . Inparticular, T τ,c defines a not necessarily complete triangulated (resp. abelian) P -fibred categoryover S . We also obtain a morphism of triangulated (resp. abelian) monoidal P -fibred categories,fully faithful as a functor, ι : T τ,c → T Definition . Consider the notations introduced above. We will call T τ,c the τ -constructiblepart of T . For any scheme S , the objects of T τ,c ( S ) will be called τ -constructible .When τ is clear from the context, we will put T c := T τ,c and use the terminology constructible . Remark . The condition of τ -constructibility is a good categorical notion of finitenesswhich extends the notion of geometric motives as introduced by Voevodsky. In the triangulatedmotivic case, it will be studied thoroughly in section 4. Proposition . Let T be a τ -twisted P -premotivic triangulated category. Let S be ascheme such that:(1) The category T ( S ) admits finite sums.(2) For any P -scheme X over S , and any twist i ∈ τ , the premotive M S ( X ) { i } is compact. i.e. stable by direct factors. . TRIANGULATED P -FIBRED CATEGORIES IN ALGEBRAIC GEOMETRY 21 Then, a premotive M over S is τ -constructible if and only if it is compact. Proof. If T is any compactly generated triangulated category, then, for any small family C of compact generators, the thick triangulated category of T generated by C consists exactly ofthe compact objects of T . (cid:3) Thus, when the conditions of this proposition are fulfilled, the category T τ,c ( S ) does notdepend on the particular choice of τ . This will often be the case in practice (see 5.1.33, 5.2.39,5.3.42). Remark . The notion of compact objects in a triangulated category was heavily de-veloped by A. Neeman. Its relation with finiteness conditions is particularly emphasized whenconsidering the derived category of complexes of quasi-coherent sheaves over a quasi-compactseparated scheme: in this triangulated category, being compact is equivalent to being perfect([
Nee96 , Cor. 4.3]).
Definition . Consider a τ -generated premotivic category M .An enlargement of M is the data of a τ ′ -twisted generalized premotivic category M togetherwith a premotivic adjunction ρ ♯ : M −→ M : ρ ∗ (where M is considered as a premotivic category in the obvious way), satisfying the followingproperties:(a) For any scheme S in S , the functor ρ ♯,S : M ( S ) → M ( S ) is fully faithful and its rightadjoint ρ ∗ S : M ( S ) → M ( S ) commutes with sums.(b) ρ ♯ induces an equivalence τ ≃ τ ′ .Again, this notion is defined similarly for a C -structured P -premotivic category.Note that for any smooth S -scheme X , we get in the context of an enlargement as above thefollowing identifications: ρ ♯,S ( M S ( X )) ≃ M S ( X ) ,ρ ∗ S ( M S ( X )) ≃ M S ( X )where M (resp. M ) denote the geometric sections of M (resp. M ).Remember also that for any morphism of schemes f and any smooth morphism p , ρ ♯ commuteswith f ∗ and p ♯ , while ρ ∗ commutes with f ∗ and p ∗ .
2. Triangulated P -fibred categories in algebraic geometry . In this entire section, we fix a base scheme S assumed to be noetherian and a fullsubcategory S of the category of noetherian S -schemes satisfying the following properties:(a) S is closed under finite sums and pullback along morphisms of finite type.(b) For any scheme S in S , any quasi-projective S -scheme belongs to S .In sections 2.2 and 2.4, we will add the following assumption on S :(c) Any separated morphism f : Y → X in S , admits a compactification in S in the senseof [ SGA4 , 3.2.5], i.e. admits a factorization of the form Y j −→ ¯ Y p −→ X where j is an open immersion, p is proper, and ¯ Y belongs to S . Furthermore, if f isquasi-projective, then p can be chosen to be projective.(d) Chow’s lemma holds in S (i.e., for any proper morphism Y → X in S , there exists aprojective birational morphism p : Y → Y in S such that f p is projective as well).A category S satisfying all these properties will be called adequate for future references. For instance, the scheme S can be the spectrum of a prime field or of a Dedekind domain. The category S might be the category of all noetherian S -schemes of finite dimension or simply the category of quasi-projective S -schemes. In all these cases, property (c) is ensured by Nagata’s theorem (see [ Con07 ]) and property (d) byChow’s lemma (see [
EGA2 , 5.6.1]). We also fix an admissible class P of morphisms in S and a complete triangulated P -fibredcategory T . We will add the following assumptions:(d) In section 2.2 and 2.3, P contains the open immersions.(e) In section 2.4, P contains the smooth morphisms of S .In the case T is monoidal, we denote by M : P / ? → T its geometric sections.According to the convention of 1.4.2, we will speak of the premotivic case when P is the classof smooth morphisms of finite type in S and T is a premotivic triangulated category. Definition . We say that T is additive, if for any finite family ( S i ) i ∈ I of schemes in S , the canonical map T a i S i ! → Y i T ( S i )is an equivalence.Recall this property implies in particular that T ( ∅ ) = 0. Lemma . Let S be a scheme, p : A S → S be the canonical projection. The followingconditions are equivalent:(i) The functor p ∗ : T ( S ) → T ( A S ) is fully faithful.(ii) The counit adjunction morphism → p ∗ p ∗ is an isomorphism.In the premotivic case, these conditions are equivalent to the following ones:(iii) The unit adjunction morphism p ♯ p ∗ → is an isomorphism.(iv) The morphism M S ( A S ) p ∗ −→ S induced by p is an isomorphism.(iv’) For any smooth S -scheme X , the morphism M S ( A X ) (1 X × p ) ∗ −−−−−→ M S ( X ) is an isomor-phism. The only thing to recall is that in the premotivic case, p ♯ p ∗ ( M ) = M S ( A S ) ⊗ M and p ∗ p ∗ ( M ) = Hom S ( M S ( A S ) , M ). Definition . The equivalent conditions of the previous lemma will be called the homo-topy property for T , denoted by (Htp).2.1.4 . Recall that a sieve R of a scheme X is a class of morphisms in S /X which is stableby composition on the right by any morphism of schemes (see [ SGA4 , I.4]).Given such a sieve R , we will say that T is R -separated if the class of functors f ∗ for f ∈ R is conservative. Given two sieves R , R ′ of X , the following properties are immediate:(a) If R ⊂ R ′ then T is R -separated implies T is R ′ -separated.(b) If T is R -separated and is R ′ -separated then T is ( R ∪ R ′ )-separated.A family of morphisms ( f i : X i → X ) i ∈ I of schemes defines a sieve R = h f i , i ∈ I i such that f isin R if and only if there exists i ∈ I such that f can be factored through f i . Obviously,(c) T is R -separated if and only if the family of functors ( f ∗ i ) i ∈ I is conservative.Recall that a topology on S is the data for any scheme X of a set of sieves of X satisfying certainstability conditions ( cf. [ SGA4 , II, 1.1]), called t -covering sieves. A pre-topology t on S is thedata for any scheme X of a set of families of morphisms of shape ( f i : X i → X ) i ∈ I satisfyingcertain stability conditions ( cf. [ SGA4 , II, 1.3]), called t -covers. A pre-topology t generated aunique topology t . Definition . Let t be a Grothendieck topology on S . We say that T is t -separated ifthe following property holds: or smooth separated morphisms of finite type when applying this section in Part 3 . TRIANGULATED P -FIBRED CATEGORIES IN ALGEBRAIC GEOMETRY 23 (t-sep) For any t -covering sieve R , T is R -separated in the sense defined above.Obviously, given two topologies t and t ′ on S such that t ′ is finer than t , if T is t -separatedthen it is t ′ -separated.If the topology t on S is generated by a pre-topology t then T is t -separated if and onlyif for any t -covers ( f i ) i ∈ I , the family of functors ( f ∗ i ) i ∈ I is conservative – use [ SGA4 , 1.4] and2.1.4(a)+(c).2.1.6 . Recall that a morphism of schemes f : T → S is radicial if it is injective and for anypoint t of T , the residual extension induced by f at t is radicial ( cf. [ EGA1 , 3.5.4, 3.5.8]) Thefollowing definition is inspired by [
Ayo07a , Def. 2.1.160].
Definition . We say that T is separated (resp. semi-separated ) if T is separated forthe topology generated by surjective families of morphisms of finite type (resp. finite radicialmorphisms) in S . We also denote by (Sep) (resp. (sSep)) this property. Remark . If T is additive, property (Sep) (resp. (sSep)) is equivalent to ask that forany surjective morphism of finite type (resp. finite surjective radicial morphism) f : T → S in S ,the functor f ∗ is conservative. Proposition . Assume T is semi-separated and satisfies the transversality property withrespect to finite surjective radicial morphisms.Then for any finite surjective radicial morphism f : Y → X , the functor f ∗ : T ( X ) → T ( Y ) is an equivalence of categories. Proof.
We first consider the case when f = i is in addition a closed immersion. In this case,we can consider the pullback square below. Y Y i (cid:15) (cid:15) Y i / / Z Using the transversality property with respect to i , we see that the counit i ∗ i ∗ → → i ∗ i ∗ is an isomorphism. As i ∗ isconservative by semi-separability, it is sufficient to check that i ∗ → i ∗ i ∗ i ∗ ( M )is an isomorphism. But this is a section of the map i ∗ i ∗ i ∗ ( M ) → i ∗ ( M ), which is already knownto be an isomorphism.Consider now the general case of a finite radicial extension f . We introduce the pullbacksquare Y × X Y p / / q (cid:15) (cid:15) Y f (cid:15) (cid:15) Y f / / X Consider the diagonal immersion i : Y → Y × X Y . Because Y is noetherian and p is separable, i isfinite ( cf. [ EGA2 , 6.1.5]) thus a closed immersion. As p is a universal homeomorphism, the sameis true for its section i . The preceding case thus implies that i ∗ is an equivalence of categories.Moreover, as pi = qi = 1 Y , we see that p ∗ and q ∗ are both quasi-inverses to i ∗ , which implies thatthey are isomorphic equivalences of categories. More precisely, we get canonical isomorphisms offunctors i ∗ ≃ p ∗ ≃ q ∗ and i ∗ ≃ p ∗ ≃ q ∗ . It is equivalent to ask that f is universally injective. When f is surjective, this is equivalent to ask that f is a universal homeomorphism. We check that the unit map 1 → f ∗ f ∗ is an isomorphism. Indeed, by semi-separability, it issufficient to prove this after applying the functor f ∗ , and we get, using the transversality propertyfor f : f ∗ ≃ i ∗ p ∗ f ∗ ≃ q ∗ p ∗ f ∗ ≃ f ∗ f ∗ f ∗ . We then check that the counit map f ∗ f ∗ → f , we have isomorphisms f ∗ f ∗ ( M ) ≃ q ∗ p ∗ ( M ) ≃ i ∗ i ∗ ( M ) ≃ M. (cid:3) . Recall from [
Voe10b ] that a cd-structure on S is a collection P of commutativesquares of schemes B / / (cid:15) (cid:15) Q Y f (cid:15) (cid:15) A e / / X which is closed under isomorphisms. We will say that a square Q in P is P -distinguished.Voevodsky associates to P a topology t P , the smallest topology such that: • for any P -distinguished square Q as above, the sieve generated by { f : A → X, e : Y → X } is t P -covering on X . • the empty sieve covers the empty scheme. Example . A Nisnevich distinguished square is a square Q as above such that Q iscartesian, f is ´etale, e is an open embedding with reduced complement Z and the induced map f − ( Z ) → Z is an isomorphism. The corresponding cd-structure is called the upper cd-structure (see section 2 of [ Voe10c ]). Because we work with noetherian schemes, the corresponding topologyis the
Nisnevich topology (see proposition 2.16 of loc.cit. ).A proper cdh -distinguished square is a square Q as above such that Q is cartesian, f isproper, e is a closed embedding with open complement U and the induced map f − ( U ) → U isan isomorphism. The corresponding cd-structure is called the lower cd-structure . The topologyassociated with the lower cd-structure is called the proper cdh-topology .The topology generated by the lower and upper cd-structures is by definition (according tothe preceding remark on Nisnevich topology) the cdh-topology .All these three examples are complete cd-structures in the sense of [ Voe10b , 2.3].
Lemma . Let P be a complete cd-structure (see [ Voe10b , def 2.3] ) on S and t P be theassociated topology. The following conditions are equivalent:(i) T is t P -separated.(ii) For any distinguished square Q for P of the above form, the pair of functors ( e ∗ , f ∗ ) isconservative. Proof.
This follows from the definition of a complete cd-structure and 2.1.4(a). (cid:3)
Remark . If we assume that S is stable by arbitrary pullback then any cd-structure P on S such that P -distinguished squares are stable by pullback is complete (see [ Voe10b , 2.4]).
The support axiom. . Consider an open immersion j : U → S . Applying 1.1.15 to the cartesian square U U j (cid:15) (cid:15) U j / / S we get a canonical natural transformation γ j : j ♯ = j ♯ ∗ Ex (∆ ♯ ∗ ) −−−−−→ j ∗ ♯ = j ∗ . . TRIANGULATED P -FIBRED CATEGORIES IN ALGEBRAIC GEOMETRY 25 Recall that the functors j ♯ and j ∗ are fully faithful (see Corollary 1.1.20).Note that according to remark 1.1.7, this natural transformation is compatible with the iden-tifications of the kind ( jk ) ♯ = j ♯ k ♯ and ( jk ) ∗ = j ∗ k ∗ . Lemma . Let S be a scheme, U and V be subschemes such that S = U ⊔ V . We let h : U → S (resp. k : V → S ) be the canonical open immersions.Assume that the functor ( h ∗ , k ∗ ) : T ( S ) → T ( U ) × T ( V ) is conservative and that T ( ∅ ) = 0 .Then the natural transformation γ h (resp. γ k ) is an isomorphism. Moreover, the functor ( h ∗ , k ∗ ) is then an equivalence of categories. Proof. As h ♯ and h ∗ are fully faithful, we have h ∗ h ♯ ≃ h ∗ h ∗ . By P -base change, we alsoget k ∗ h ♯ ≃ k ∗ h ∗ ≃
0. It remains to prove the last assertion. The functor R = ( h ∗ , k ∗ ) has a leftadjoint L defined by L= h ♯ ⊕ k ♯ : L ( M, N ) = h ♯ ( M ) ⊕ k ♯ ( N ) . The natural transformation LR → h ∗ and k ∗ , which gives an isomorphism in T ( U ) and T ( V ) respectively. The natural transformation1 → RL is also an isomorphism because h ♯ and k ♯ are fully faithful. (cid:3) Remark . Assume T is Zariski separated (definition 2.1.5). Then, as a corollary of thislemma, T is additive (definition 2.1.1) if and only if T ( ∅ ) = 0.2.2.4 . Exchange structures V .– Assume T is additive. We consider a commutative square ofschemes V k / / q (cid:15) (cid:15) ∆ T p (cid:15) (cid:15) U j / / S (2.2.4.1)such that j , k are an open immersions and p , q are a proper morphisms.This diagram can be factored into the following commutative diagram: V k % % q ( ( l ◆◆◆ ' ' ◆◆◆ U × S T j ′ / / p ′ (cid:15) (cid:15) Θ T p (cid:15) (cid:15) U j / / S. Then l is an open and closed immersion so that the previous lemma implies the canonical morphism γ l : l ♯ → l ∗ is an isomorphism. As a consequence, we get a natural exchange transformation Ex (∆ ♯ ∗ ) : j ♯ q ∗ = j ♯ p ′∗ l ∗ Ex (Θ ♯ ∗ ) −−−−−→ p ∗ j ′ ♯ l ∗ γ − l −−→ p ∗ j ′ ♯ l ♯ = p ∗ k ♯ using the exchange of 1.1.15. Note that, with the notations introduced in 2.2.1, the followingdiagram is commutative. j ♯ q ∗ Ex (∆ ♯ ∗ ) / / γ j q ∗ (cid:15) (cid:15) p ∗ k ♯p ∗ γ k (cid:15) (cid:15) j ∗ q ∗ ∼ / / ( jq ) ∗ = ( pk ) ∗ p ∗ k ∗∼ o o (2.2.4.2)Indeed one sees first that it is sufficient to treat the case where ∆ is cartesian. Then, as j ♯ is afully faithful left adjoint to j ∗ it is sufficient to check that (2.2.4.2) commutes after having applied j ∗ . Using the cotransversality property with respect to open immersions, one sees then that thisconsists to verify the commutativity of (2.2.4.2) when j is the identity, in which case it is trivial. Definition . Let p : T → S be a proper morphism in S .We say that the triangulated P -fibred category T satisfies the support property with respectto p , denoted by (Supp p ), if it is additive and for any commutative square of shape (2.2.4.1) theexchange transformation Ex (∆ ♯ ∗ ) : j ♯ q ∗ → p ∗ k ♯ defined above is an isomorphism.We say that T satisfies the support property , also denoted by (Supp), if it satisfies (Supp p )for all proper morphism p in S .By definition, it is sufficient to check the last property of property (Supp) in the case where ∆ iscartesian.2.2.b. Exceptional direct image. . We denote by S sep (resp. S open , S prop ) the sub-category of the category S withthe same objects but morphisms are separated morphisms of finite type (resp. open immersions,proper morphisms). We denote by T ∗ : S → T ri ⊗ resp. T ♯ : S open → T ri ⊗ the 2-functor defined respectively by morphisms of type f ∗ and j ♯ ( f any morphism of schemes).The proposition below is essentially based on a result of Deligne [ SGA4 , XVII, 3.3.2]:
Proposition . Assume T is a monoidal P -fibred category and satisfies property (Supp).Then there exists a unique -functor T ! : S sep → T ri ⊗ with the property that T ! | S prop = T ∗ | S prop , T ! | S open = T ♯ and for any commutative square ∆ of shape (2.2.4.1) the composition of the structural isomor-phisms j ♯ q ∗ = j ! q ! ≃ ( jq ) ! = ( pk ) ! ≃ p ! k ! = p ∗ k ♯ is equal to the exchange transformation Ex (∆ ♯ ∗ ) . . Under the assumptions of the proposition, for any separated morphism of finite type f : Y → X , we will denote by f ! : T ( Y ) → T ( X ) the functor T ! ( f ). The functor f ! is called the direct image functor with compact support or the left exceptional functor associated with f . Proof.
We recall the principle of the proof of Deligne. Let f : Y → X be a separatedmorphism of finite type in S .Let C f be the category of compactifications of f in S , i.e. of factorizations of f of the form(2.2.8.1) Y j −→ ¯ Y p −→ X where j is an open immersion, p is proper, and ¯ Y belongs to S . Morphisms of C f are given bycommutative diagrams of the form ¯ Y ′ p ′ + + ❲❲❲❲❲❲ π (cid:15) (cid:15) Y j ′ ❤❤❤❤❤❤ j + + ❲❲❲❲❲❲❲ X. ¯ Y p ❣❣❣❣❣❣❣ (2.2.8.2)in S . To any compactification of f of shape (2.2.8.1), we associate the functor p ∗ j ♯ .To any morphism of compactifications (2.2.8.2), we associate a natural isomorphism p ′∗ j ′ ♯ = p ∗ π ∗ j ′ ♯ Ex (∆ ♯ ∗ ) − −−−−−−−→ p ∗ j ♯ ∗ = p ∗ j ♯ . where ∆ stands for the commutative square made by removing π in the diagram (2.2.8.2), and Ex (∆ ♯ ∗ ) is the corresponding natural transformation (see 2.2.4). The compatibility of Ex (∆ ♯ ∗ )with composition of morphisms of schemes shows that we have defined a functorΓ f : C op f → Hom ( T ( Y ) , T ( X ))which sends all the maps of C f to isomorphisms (by the support property). . TRIANGULATED P -FIBRED CATEGORIES IN ALGEBRAIC GEOMETRY 27 The category C f is non-empty by the assumption 2.0(c) on S , and it is in fact left filtering;see [ SGA4 , XVII, 3.2.6(ii)]. This defines a canonical functor f ! : T ( Y ) → T ( X ), independent ofany choice compactification of f , defined in the category of functors Hom ( T ( Y ) , T ( X )) by theformula f ! = lim −→ C op f Γ f . If f = p is proper, then the compactification Y = −→ Y p −→ X is an initial object of C f , which gives a canonical identification p ! = p ∗ . Similarly, if f = j is anopen immersion, then the compactification Y j −→ X = −→ X is a terminal object of C j , so that we get a canonical identification j ! = j ♯ .This construction is compatible with composition of morphisms. Let g : Z → Y and f : Y → X be two separated morphisms of finite type in S . For any a couple of compactifications Z k −→ ¯ Z q −→ Y and Y j −→ ¯ Y p −→ X of f and g respectively, we can choose a compactification¯ Z h −→ T r −→ Y of jq , and we get a canonical isomorphism f ! g ! ≃ p ∗ j ♯ q ∗ k ♯ ≃ p ∗ r ∗ h ♯ k ♯ ≃ ( pr ) ∗ ( hk ) ♯ ≃ ( f g ) ! . The independence of these isomorphic with respect to the choices of compactification follows from[
SGA4 , XVII, 3.2.6(iii)]. The cocycle conditions (i.e. the associativity) also follows formally from[
SGA4 , XVII, 3.2.6]. The uniqueness statement is obvious. (cid:3) . This construction is functorial in the following sense.Define a 2 -functor with support on T to be a triple ( D , a, b ), where:(i) D : S sep → T ri is a 2-functor (we shall write the structural coherence isomorphisms as c g,f : D ( gf ) ∼ −→ D ( g ) D ( f ) for composable arrows f and g in S sep );(ii) a : T ∗ | S prop → D | S prop and b : T ♯ → D | S open are morphisms of 2-functors which agreeon objects, i.e. such that for any scheme S in S , we have ψ S = a S = b S : T ( S ) → D ( S ) ;(iii) for any commutative square of shape (2.2.4.1) in which j and k are open immersions,while p and q are proper morphisms, the diagram below commutes. ψ S j ♯ q ∗ ψ S Ex (∆ ♯ ∗ ) / / b q ∗ (cid:15) (cid:15) ψ S p ∗ k ♯ak ♯ (cid:15) (cid:15) D ( j ) ψ U q ∗ D ( j ) a (cid:15) (cid:15) D ( p ) ψ T k ♯ D ( p ) b (cid:15) (cid:15) D ( j ) D ( q ) ψ V c − j,q / / D ( jq ) = D ( pk ) ψ V D ( p ) D ( k ) ψ Vc − p,k o o Morphisms of 2-functors with support on T ( D , a, b ) → ( D ′ , a ′ , b ′ )are defined in the obvious way: these are morphisms of 2-functors D → D ′ which preserve all thestructure on the nose.Using the arguments of the proof of 2.2.7, one checks easily that the category of 2-functorswith support has an initial object, which is nothing else but the 2-functor T ! together with theidentities of T ∗ | S prop and of T ♯ respectively. In particular, for any 2-functor D : S sep → T ri , amorphism of 2-functors T ! → D is completely determined by its restrictions to S prop and S open , and by its compatibility with the exchange isomorphisms of type Ex (∆ ♯ ∗ ) in the sense describedin condition (iii) above. Proposition . Assume that T satisfies the support property and consider the notationsof Proposition 2.2.7. For any separated morphism of finite type f in S , there exists a canonicalnatural transformation α f : f ! → f ∗ . The collection of maps α f defines a morphism of -functors α : T ! → T ∗ | S sep , f ( α f : f ! → f ∗ ) whose restrictions to S prop and S open are respectively the identity and the morphism of -functors γ : T ♯ → T ∗ | S open defined in 2.2.1. Proof.
The identities f ∗ = f ∗ for f proper (resp. projective) and the exchange naturaltransformations of type Ex (∆ ♯ ∗ ) turns T ∗ | S sep into a 2-functor with support (resp. restrictedsupport) on T (property (iii) of 2.2.9 is expressed by the commutative square (2.2.4.2)). (cid:3) Proposition . Let T ′ be another triangulated complete P -fibred category over S .Assume that T and T ′ both have the support property, and consider given a triangulated morphismof P -fibred categories ϕ ∗ : T → T ′ (recall definition 1.2.2).Then, there is a canonical family of natural transformations Ex ( ϕ ∗ , f ! ) : ϕ ∗ X f ! → f ! ϕ ∗ Y for each separated morphism of finite type f : Y → X in S , which is functorial with respect tocomposition in S (i.e. defines a morphism of -functors) and such that, the following conditionsare verified:(a) if f is proper, then, under the identification f ! = f ∗ , the map Ex ( ϕ ∗ , f ! ) is the exchangetransformation Ex ( ϕ ∗ , f ∗ ) : ϕ ∗ X f ∗ → f ∗ ϕ ∗ Y defined in 1.2.5;(b) if f is an open immersion, then, under the identification f ! = f ♯ , the map Ex ( ϕ ∗ , f ! ) isthe inverse of the exchange isomorphism Ex ( f ♯ , ϕ ∗ ) : f ♯ ϕ ∗ Y → ϕ ∗ X f ♯ defined in 1.2.1. Proof.
The exchange maps of type Ex ( ϕ ∗ , f ∗ ) define a morphism of 2-functors a : T ∗ | S prop → T ′∗ | S prop = T ′ ! | S prop while the inverse of the exchange isomorphisms of type Ex ( f ♯ , ϕ ∗ ) define a morphism of 2-functors b : T ♯ → T ′ ♯ = T ′ ! | S open , in such a way that the triple ( T ′ ! , a, b ) is a 2-functor with support on T . (cid:3) Corollary . Suppose T satisfies the support property and consider the notations ofproposition 2.2.7.(1) For any cartesian square Y ′ f ′ / / g ′ (cid:15) (cid:15) ∆ X ′ g (cid:15) (cid:15) Y f / / X, such that f is separated of finite type, there exists a canonical natural transformation Ex (∆ ∗ ! ) : g ∗ f ! → f ′ ! g ′∗ compatible with horizontal and vertical compositions of squares, and satisfying the fol-lowing identifications in T ( X ′ ) ( a ) f proper: ( b ) f open immersion: g ∗ f ! Ex (∆ ∗ ! ) / / f ′ ! g ′∗ g ∗ f ! Ex (∆ ∗ ! ) / / f ! g ′∗ g ∗ f ∗ Ex (∆ ∗∗ ) / / f ′∗ g ′∗ , g ∗ f ♯ Ex (∆ ∗ ♯ ) − / / f ′ ♯ g ′∗ . . TRIANGULATED P -FIBRED CATEGORIES IN ALGEBRAIC GEOMETRY 29 Moreover, when g is a P -morphism, Ex (∆ ∗ ! ) is an isomorphism.(2) For any cartesian square ∆ as in (1), assuming f is separated of finite type and g is a P -morphism, there exists a canonical natural transformation Ex (∆ ♯ ! ) : g ♯ f ′ ! → f ! g ′ ♯ compatible with horizontal and vertical compositions of squares, and satisfying the fol-lowing identifications in T ( X ′ ) ( a ) f proper: ( b ) f open immersion: g ♯ f ′ ! Ex (∆ ♯ ! ) / / f ! g ′ ♯ g ♯ f ′ ! Ex (∆ ♯ ! ) / / f ! g ′ ♯ g ♯ f ′ ∗ Ex (∆ ♯ ∗ ) / / f ∗ g ′ ♯ , g ♯ f ′ ♯ f ♯ g ′ ♯ . (3) If furthermore T is monoidal then for any separated morphism of finite type f : Y → X ,there is a natural transformation Ex ( f ∗ ! , ⊗ ) : ( f ! K ) ⊗ L → f ! ( K ⊗ f ∗ L ) which is compatible with respect to composition in S , and such that, in each of thefollowing cases, we have the following identifications: ( a ) f proper: ( b ) f open immersion: ( f ! K ) ⊗ L Ex ( f ∗ ! , ⊗ ) / / f ! ( K ⊗ f ∗ L ) ( f ! K ) ⊗ L Ex ( f ∗ ! , ⊗ ) / / f ! ( K ⊗ f ∗ L )( f ∗ K ) ⊗ L Ex ( f ∗∗ , ⊗ ) / / f ∗ ( K ⊗ f ∗ L ) , ( f ♯ K ) ⊗ L Ex ( f ∗ ♯ , ⊗ ) − / / f ♯ ( K ⊗ f ∗ L ) . As in the previous analogous cases, the natural transformations Ex (∆ ∗ ! ), Ex (∆ ♯, ! ) and Ex ( f ∗ ! , ⊗ )will be called exchange transformations . Proof.
To prove (1), consider a fixed map g : X ′ → X in S . We consider the triangulated P /X -fibred categories T ′ and T ′′ over S /X defined by T ′ ( Y ) = T ( Y ) and T ′′ ( Y ) = T ( Y ′ )for any X -scheme Y (in S ), with g ′ : Y ′ = Y × X X ′ → Y the map obtained from Y → X bypullback along g . The collection of functors g ′ ∗ : T ( Y ) → T ( Y ′ )define an exact morphism of triangulated P /X -fibred categories over S /X (by the P -base changeformula): ϕ ∗ : T ′ → T ′′ . Applying the preceding proposition to the latter gives (1). The fact that we get an isomorphismwhenever g is a P -morphism follows from the P -base change formula and from paragraph 1.1.15.For point (2), we consider the notations above assuming that g is a P -morphism. Thecollection of functors g ′ ♯ : T ( Y ′ ) → T ( Y )associated with an X -scheme Y , g ′ : Y ′ = Y × X X ′ → Y obtained from g as above, define an exactmorphism of triangulated P /X -fibred categories over S /X (applying again the P -base changeformula): ϕ ∗ : T ′′ → T ′ . Applying the preceding proposition to the latter gives (2).The proof of (3) is similar: fix a scheme X in S , as well as an object L in T ( X ). Let T ′ bethe restriction of T to S /X as above. We can consider L as a cartesian section of T ′ , and by the P -projection formula, we then have an exact morphism of triangulated P /X -fibred categoriesover S /X : L ⊗ ( − ) : T ′ → T ′ . Here again, we can apply the preceding proposition and conclude. (cid:3) Further properties.
We will be particularly interested in the following properties of thetriangulated P -fibred category T . Definition . Let f : Y → X be a morphism in S . We introduced the followingproperties for T , assuming in the third case that T is monoidal:(Adj f ) The functor f ∗ admits a right adjoint. Under this assumption, we denote by f ! the rightadjoint of f ∗ .(BC f ) Any cartesian square of S of the form Y ′ f ′ / / g ′ (cid:15) (cid:15) ∆ X ′ g (cid:15) (cid:15) Y f / / X, is T -transversal (Def. 1.1.17) – i.e. the exchange transformation Ex (∆ ∗∗ ) : g ∗ f ∗ → f ′∗ g ′∗ associated with ∆ is an isomorphism.(PF f ) For any object premotive M over Y , and N over X , the exchange transformation (seeparagraph 1.1.31) Ex ( f ∗∗ , ⊗ X ) : ( f ∗ M ) ⊗ X N → f ∗ ( M ⊗ Y f ∗ N )is an isomorphism.We denote by (Adj) (resp. (BC), (PF)) the property (Adj f ) (resp. (BC f ), (PF f )) for any proper morphism f in S and call it the adjoint property (resp. proper base change property , projectionformula ).We can summarize the construction and properties introduced in this section as follows: Theorem . Assume T satisfies the properties (Supp) and (Adj).Then for any separated morphism of finite type f : Y → X in S , there exists an essentiallyunique pair of adjoint functors f ! : T ( Y ) ⇄ T ( X ) : f ! called the exceptional functors , such that:(1) There exists a structure of a covariant (resp. contravariant) -functor on f f ! (resp. f f ! ).(2) There exists a natural transformation α f : f ! → f ∗ compatible with composition in f which is an isomorphism when f is proper.(3) For any open immersion j , j ! = j ♯ and j ! = j ∗ .(4) For any cartesian square Y ′ f ′ / / g ′ (cid:15) (cid:15) ∆ X ′ g (cid:15) (cid:15) Y f / / X, in which f is separated and of finite type, there exists natural transformations Ex (∆ ∗ ! ) : g ∗ f ! → f ′ ! g ′∗ ,Ex (∆ ! ∗ ) : g ′∗ f ′ ! → f ! g ∗ which are isomorphisms in the following three cases: • f is an open immersion. • g is a P -morphism. • T satisfies the proper base change property (BC).Assume that T is in addition monoidal. Then the following property holds: . TRIANGULATED P -FIBRED CATEGORIES IN ALGEBRAIC GEOMETRY 31 (5) For any separated morphism of finite type f : Y → X in S , there exists natural trans-formations Ex ( f ∗ ! , ⊗ ) : ( f ! K ) ⊗ X L −→ f ! ( K ⊗ Y f ∗ L ) , Hom X ( f ! ( L ) , K ) −→ f ∗ Hom Y ( L, f ! ( K )) ,f ! Hom X ( L, M ) −→ Hom Y ( f ∗ ( L ) , f ! ( M )) . which are isomorphisms in the following cases: • f is an open immersion. • T satisfies the projection formula (PF). Indeed the existence of f ! follows from Proposition 2.2.7 while that of f ! follows directly fromassumption (Adj). Assertions (1) and (3) follows from the construction, (2) is Proposition 2.2.10,(4) (resp. (5)) follows from Corollary 2.2.12 and the definition of (BC) (resp. (PF)). Note alsothat the second and third isomorphisms in (5) are obtained by transposition from Ex ( f ! , ⊗ ).2.2.15 . While the properties (BC f ) and (PF f ) are only reasonable in practice for propermorphisms, this is not the case for the property (Adj f ). Recall that an exact functor betweenwell generated triangulated categories admits a right adjoint if and only if it commutes with smallsums: this is an immediate consequence of the Brown representability theorem proved by Neeman( cf. [ Nee01 , 8.4.4]).
Proposition . Assume that T is a compactly τ -generated triangulated premotivic cat-egory over S .Then, for any morphism of schemes f : T → S , the functor f ∗ : T ( T ) → T ( S ) admits a rightadjoint. Proof.
This follows directly from Proposition 1.3.20. (cid:3)
Definition. . Consider a closed immersion i : Z → S in S . Let U = S − Z be the complement opensubscheme of S and j : U → S the canonical immersion. We will use the following consequence ofthe triangulated P -fibred structure on T :(a) The unit 1 → j ∗ j ♯ is an isomorphism.(b) The counit j ∗ j ∗ → i ∗ j ♯ = 0.(d) j ∗ i ∗ = 0.(e) The composite map j ♯ j ∗ ad ′ ( j ♯ ,j ∗ ) −−−−−−→ ad ( i ∗ ,i ∗ ) −−−−−→ i ∗ i ∗ is zero.In fact, the first four relations all follow from the base change property ( P -BC). Relation (e) is aconsequence of (d) once we have noticed that the following square is commutative j ♯ j ∗ / / (cid:15) (cid:15) (cid:15) (cid:15) j ♯ j ∗ i ∗ i ∗ / / i ∗ i ∗ . For the closed immersion i and the triangulated category T , we introduce the property (Loc i )made of the following assumptions:(a) The pair of functors ( j ∗ , i ∗ ) is conservative.(b) The counit i ∗ i ∗ ad ′ ( i ∗ ,i ∗ ) −−−−−−→ Definition . We say that T satisfies the localization property , denoted by (Loc), if:(1) T ( ∅ ) = 0.(2) For any closed immersion i in S , (Loc i ) is satisfied.The main consequence of the localization axiom is that it leads to the situation of the sixgluing functor ( cf. [ BBD82 , prop. 1.4.5]): Proposition . Let i : Z → S be a closed immersion with complementary open immersion j : U → S such that (Loc i ) is satisfied.(1) The functor i ∗ admits a right adjoint i ! .(2) For any K in T ( S ) , there exists a unique map ∂ i,K : i ∗ i ∗ K → j ♯ j ∗ K [1] such that thetriangle j ♯ j ∗ K ad ′ ( j ♯ ,j ∗ ) −−−−−−→ K ad ( i ∗ ,i ∗ ) −−−−−→ i ∗ i ∗ K ∂ i,K −−−→ j ♯ j ∗ K [1] is distinguished. The map ∂ i,K is functorial in K .(3) For any K in T ( S ) , there exists a unique map ∂ ′ i,K : j ∗ j ∗ K → i ∗ i ! K [1] such that thetriangle i ∗ i ! K ad ′ ( i ∗ ,i ! ) −−−−−−→ K ad ( j ∗ ,j ∗ ) −−−−−−→ j ∗ j ∗ K ∂ ′ i,K −−−→ i ∗ i ! K [1] is distinguished. The map ∂ ′ i,K is functorial in K . Under the property (Loc i ), the canonical triangles appearing in (2) and (3) above are calledthe localization triangles associated with i . Proof.
We first consider point (2). For the existence, we consider a distinguished triangle j ♯ j ∗ K ad ′ ( j ♯ ,j ∗ ) −−−−−−→ K π −−→ C +1 −−→ Applying 2.3.1(e), we obtain a factorization K ad ( i ∗ ,i ∗ ) / / π & & ◆◆◆◆◆ i ∗ i ∗ KC w ♠♠♠♠♠ We prove w is an isomorphism. According to the above triangle, j ∗ C = 0. From 2.3.1(d), j ∗ i ∗ i ∗ K = 0 so that j ∗ w is an isomorphism. Applying i ∗ to the above distinguished triangle, weobtain from 2.3.1(c) that i ∗ π is an isomorphism. Thus, applying i ∗ to the above commutativediagram together with (Loc i ) (b), we obtain that i ∗ w is an isomorphism which concludes.Consider a map K u −→ L in T ( S ) and suppose we have chosen maps a and b in the diagram: j ♯ j ∗ K ad ′ ( j ♯ ,j ∗ ) / / u (cid:15) (cid:15) K ad ( i ∗ ,i ∗ ) / / u (cid:15) (cid:15) i ∗ i ∗ K a / / j ♯ j ∗ K [1] u (cid:15) (cid:15) j ♯ j ∗ L ad ′ ( j ♯ ,j ∗ ) / / L ad ( i ∗ ,i ∗ ) / / i ∗ i ∗ L b / / j ♯ j ∗ L [1]such that the horizontal lines are distinguished triangles. We can find a map h : i ∗ i ∗ K → i ∗ i ∗ L completing the previous diagram into a morphism of triangles. Then the map w = h − i ∗ i ∗ ( u )satisfy the relation w ◦ ad ( i ∗ , i ∗ ) = 0. Thus it can be lifted to a map in Hom( j ♯ j ∗ K [1] , i ∗ i ∗ L ). Butthis is zero by adjunction and the relation 2.3.1(d). This proves both the naturality of ∂ i,K andits uniqueness.For point (1) and (3), for any object K of T ( S ), we consider a distinguished triangle D → K ad ( j ∗ ,j ∗ ) −−−−−−→ j ∗ j ∗ K +1 −−→ According to 2.3.1(b), j ∗ D = 0. Thus according to the triangle of point (2) applied to D , we obtain D = i ∗ i ∗ D . Arguing as for point (2), we thus obtain that D is unique and depends functorialy on K so that, if we put i ! K = i ∗ D , point (1) and (3) follows. (cid:3) Remark . Consider the hypothesis and notations of the previous proposition.(1) By transposition from 2.3.1(d), we deduce that i ! j ∗ = 0.(2) Assume that i is a P -morphism. Then the P -base change formula implies that i ∗ j ∗ = 0.Dually, we get that i ! j ♯ = 0. By adjunction, we thus obtain ∂ i,K = 0 and ∂ ′ i,K = 0 forany object K so that both localization triangles are split. In that case, we get that T ( S ) = T ( Z ) × T ( U ). This remark explains why the localization property is too strong for generalized premotivic categories. . TRIANGULATED P -FIBRED CATEGORIES IN ALGEBRAIC GEOMETRY 33 The preceding proposition admits the following reciprocal statement:
Lemma . Consider a closed immersion i : Z → S in S with complementary open im-mersion j : U → S . Then the following properties are equivalent:(i) T satisfies (Loc i ).(ii) (a) The functor i ∗ is conservative.(b) For any object K of T ( S ) , there exists a map i ∗ i ∗ ( K ) → j ♯ j ∗ ( K )[1] which fits intoa distinguished triangle j ♯ j ∗ ( K ) ad ′ ( j ♯ ,j ∗ ) −−−−−−→ K ad ( i ∗ ,i ∗ ) −−−−−→ i ∗ i ∗ ( K ) → j ♯ j ∗ ( K )[1] Proof.
The fact (i) implies (ii) follows from Proposition 2.3.3. Conversely, (ii)(b) impliesthat the pair ( i ∗ , j ∗ ) is conservative and it remains to prove (Loc i ) (b). Let K be an object of T ( S ). Consider the distinguished triangle given by (ii)(b): j ♯ j ∗ ( K ) ad ′ ( j ♯ ,j ∗ ) −−−−−−→ K ad ( i ∗ ,i ∗ ) −−−−−→ i ∗ i ∗ ( K ) → j ♯ j ∗ ( K )[1] . If we apply i ∗ on the left to this triangle, we get using 2.3.1(d) that the morphism i ∗ ( K ) ad ( i ∗ ,i ∗ ) .i ∗ −−−−−−−→ i ∗ i ∗ i ∗ ( K )is an isomorphism. Hence, by the zig-zag equation, the morphism i ∗ i ∗ i ∗ ( K ) i ∗ .ad ′ ( i ∗ ,i ∗ ) −−−−−−−−→ i ∗ ( K )is an isomorphism. Property (ii)(a) thus implies that i ∗ i ∗ ( K ) ≃ K . (cid:3) First consequences of localization.
The following statement is straightforward.
Proposition . Assume T satisfies the localization property and consider a scheme S in S . (1) Let S red be the reduced scheme associated with S . The canonical immersion S red ν −→ S induces an equivalence of categories: ν ∗ : T ( S ) → T ( S red ) . (2) For any any partition(3) partition ( S i ν i −→ S ) i ∈ I of S by locally closed subsets, the family of functors ( ν ∗ i ) i ∈ I isconservative ( S i is considered with its canonical structure of a reduced subscheme of S ). Lemma . If T satisfies the localization property (Loc) then it is additive. Proof.
Note that, by assumption, T ( ∅ ) = 0. Then the assertion follows directly fromLemma 2.2.2. (cid:3) Proposition . If T satisfies the localization property then it satisfies the cdh -separationproperty. Proof.
Consider a cartesian square of schemes B / / (cid:15) (cid:15) Q Y p (cid:15) (cid:15) A e / / X. According to Lemma 2.1.12, we have only to check that the pair of functors ( e ∗ , p ∗ ) is conservativewhen Q is a Nisnevich (or respectively a proper cdh) distinguished square. Let ν : A ′ → X be thecomplementary closed (resp. open) immersion to e , where A ′ has the induced reduced subscheme(resp. induced subscheme) structure. Consider the cartesian square Y p (cid:15) (cid:15) B ′ q (cid:15) (cid:15) o o X A ′ ν o o By assumption on Q , q is an isomorphism. According to (Loc) (ii), ( e ∗ , ν ∗ ) is conservative. Thisconcludes. (cid:3) The following proposition can be found in a slightly less precise and general form in [
Ayo07a ,2.1.162]. Proposition . Assume T satisfies the localization property.Then the following conditions are equivalent:(i) T is separated.(ii) For a morphism f : T → S in S , f ∗ : T ( S ) → T ( T ) is conservative whenever f is:(a) a finite ´etale cover;(b) finite, faithfully flat and radicial. Proof.
Only ( ii ) ⇒ ( i ) requires a proof. Consider a surjective morphism of finite type f : T → S in S . According to [ EGA4 , 17.16.4], there exists a partition ( S i ) i ∈ I of S by (affine)subschemes and a family of maps of the form S ′′ i g i −→ S ′ i h i −→ S i such that g i (resp. h i ) satisfies assumption (a) (resp. (b)) above and such that for any i ∈ I , f × S S ′′ i admits a section. Thus, Proposition 2.3.6 concludes. (cid:3) Localization and exchange properties. . Consider a morphism of complete triangulated P -fibred categories over S : ϕ ∗ : T → T ′ . Recall that for any morphism f : Y → X , there is an exchange transformation (1.2.5.1): Ex ( ϕ ∗ , f ∗ ) : ϕ ∗ X f ∗ −→ f ∗ ϕ ∗ Y . If T and T ′ satisfies the support axiom and f is separated of finite type, we have constructed(Proposition 2.2.11) another exchange transformation: Ex ( ϕ ∗ , f ! ) : ϕ ∗ X f ! −→ f ! ϕ ∗ Y . Proposition . Consider a morphism ϕ ∗ : T → T ′ as above.(1) Let i : Z → X be a closed immersion such that T and T ′ satisfy property (Loc i ).Then the exchange Ex ( ϕ ∗ , i ∗ ) : ϕ ∗ X i ∗ → i ∗ ϕ ∗ Z is an isomorphism.(2) Assume T and T ′ satisfy property (Loc).Then the following conditions are equivalent:(i) For any integer n > and any scheme X in S , the exchange Ex ( ϕ ∗ , p n ∗ ) is anisomorphism where p n : P nX → X is the canonical projection.(ii) For any proper morphism f : Y → X , the exchange Ex ( ϕ ∗ , f ∗ ) is an isomorphism.(3) Assume T and T ′ satisfy properties (Loc) and (Supp).Then conditions (i) and (ii) above are equivalent to the following one:(iii) For any separated morphism f : Y → X of finite type, the exchange Ex ( ϕ ∗ , f ! ) isan isomorphism. Remark . We will simply say that ϕ ∗ commutes with f ! when assertion (iii) is fulfilled.For an important case where this happens, see Proposition 2.4.53. Proof.
Assertion (1) follows easily from the conservativity of ( i ∗ , j ∗ ) where j is the comple-mentary open immersion and the relations of paragraph 2.3.1. Assertion (3) is an easy consequenceof the definition of f ! and the exchange Ex ( ϕ ∗ , f ! ).Concerning assertion (2), we have to prove that (i) implies (ii). We fix a morphism f : Y → X and prove that the exchange Ex ( ϕ ∗ , f ∗ ) : ϕ ∗ Y f ∗ → f ∗ ϕ ∗ X is an isomorphism.We first treat the case where f is projective. According to Proposition 2.3.8, T ′ satisfies theZariski separation property. Using the ( P -BC) property, we see that the problem is local in X so A warning: the proof in loc. cit. seems to require that the schemes are excellent. . TRIANGULATED P -FIBRED CATEGORIES IN ALGEBRAIC GEOMETRY 35 that we can assume X is affine. Then X admits an ample line bundle and there exists an integer n > f can be factored ([ EGA2 , (5.5.4)(ii)]) into a closed immersion i : Y → P nX andthe projection p n : P nX → X . Thus, assertion (1) and assumption (i) allows to conclude.To treat the general case, we argue by noetherian induction on Y , assuming that for anyproper closed subscheme T of Y , the result is known for the restriction of f to T . In fact, the case T = ∅ is obvious because T ( ∅ ) = 0.According to Chow’s lemma [ EGA2 , 5.6.2], there exists a morphism p : Y → Y such that:(a) p and f ◦ p are projective morphisms.(b) There exists a dense open subscheme V of Y over which p is an isomorphism.Let T be the complement of V in Y equipped with its reduced subscheme structure. Let j and i be the respective immersion of T and V in Y . According to point (3) of Proposition 2.3.3, it issufficient to prove that the following natural transformations are isomorphisms: ϕ ∗ Y f ∗ i ∗ → f ∗ ϕ ∗ X i ∗ . (2.3.12.1) ϕ ∗ Y f ∗ j ∗ → f ∗ ϕ ∗ X j ∗ . (2.3.12.2)Concerning the first one, we consider the following commutative diagram: ϕ ∗ Y f ∗ i ∗ Ex ( ϕ ∗ ,f ∗ ) / / f ∗ ϕ ∗ X i ∗ Ex ( ϕ ∗ ,i ∗ ) / / f ∗ i ∗ ϕ ∗ X ϕ ∗ Y ( f i ) ∗ Ex ( ϕ ∗ , ( fi ) ∗ ) / / ( f i ) ∗ ϕ ∗ X . Thus the result follows from assertion (1) and the induction hypothesis.Concerning the natural transformation (2.3.12.2), we consider the pullback square V l / / q (cid:15) (cid:15) Y p (cid:15) (cid:15) V j / / Y. Assumption (b) above says that q is an isomorphism which implies the relation: j ∗ = p ∗ l ∗ q ∗ .In particular, it is sufficient to prove that the natural transformation ϕ ∗ Y f ∗ p ∗ → f ∗ ϕ ∗ X p ∗ is anisomorphism. This follows from the commutativity of the following diagram ϕ ∗ Y f ∗ p ∗ Ex ( ϕ ∗ ,f ∗ ) / / f ∗ ϕ ∗ X p ∗ Ex ( ϕ ∗ ,p ∗ ) / / f ∗ p ∗ ϕ ∗ X ϕ ∗ Y ( f p ) ∗ Ex ( ϕ ∗ , ( fp ) ∗ ) / / ( f p ) ∗ ϕ ∗ X , according to the projective case treated above and assumption (b). The proof is complete. (cid:3) Corollary . In the next statements, we assume T is monoidal when it is needed.(1) Let i : Z → X be a closed immersion such that T satisfies property (Loc i ).Then T satisfies property (Supp i ) (resp. (BC i ), (PF i )).(2) Assume T satisfies the localization property. Then the following properties of T areequivalent:(i) For any integer n > and any scheme X in S , p n : P nX → X being the canonicalprojection, T satisfies (Supp p n ) (resp. (BC p n ), (PF p n )).(ii) T satisfies (Supp) (resp. (BC), (PF)).(3) Assume T is well generated and satisfies the localization property. Then the followingproperties of T are equivalent:(i’) For any integer n > and any scheme X in S , p n : P nX → X being the canonicalprojection, T satisfies (Adj p n ).(ii’) T satisfies (Adj). Proof.
As in the proof of Corollary 2.2.12, each respective case of assertions (1) and (2)follows from the previous proposition applied to a particular type of morphisms ϕ ∗ : T ′ → T ′′ ofcomplete P -fibred triangulated categories over a subcategory S ′ of S . For property (Supp), we proceed as follows. We fix an open immersion j : U → X andlet S ′ = S /X . For any Y /X , we let j Y = Y × X U → Y be the pullback of j . We put T ′ ( Y ) = T ( Y × X U ) and T ′′ ( Y ) = T ( Y ) and let ϕ ∗ Y be the functor: j Y ♯ : T ( Y × X U ) → T ( Y ) . For the property (BC) (resp. (PF)), we refer the reader to the proof of assertion (1) (resp.(2)) in Corollary 2.2.12.Finally we consider assertion (3). It is sufficient to proce that (i’) implies (ii’).According to the Brown representability theorem [
Nee01 , 8.4.4], the property (Adj f ) for a propermorphism f is equivalent to ask that f ∗ preserves small sum.Consider an arbitrary set I . For any scheme S , we put T I ( S ) = T ( S ) I , that is the category offamilies of object of T ( S ) indexed by I . Then T I is obviously a complete triangulated P -fibredcategory over S (limits and colimits are computed termwise). For any scheme S , we consider thefunctor: ϕ ∗ S : T I ( S ) → T ( S ) , ( M i ) i ∈ I X i ∈ I M i . Then ϕ ∗ : T i → T is obviously a morphism of complete P -fibred categories. Thus, givencondition (i’), the preceding proposition applied to ϕ ∗ shows that for any proper morphism f , f ∗ commutes with sums indexed by I . As this is true for any I , we obtain (ii’). (cid:3) Localization and monoidal structure. . Assume T is monoidal and let M denote its geometric sections. Fix a closed immersion i : Z → S in S with complementary open immersion j : U → S . We fix an object M S ( S/S − Z )of T ( S ) and a distinguished triangle(2.3.14.1) M S ( S − Z ) j ∗ −→ S p i −→ M S ( S/S − Z ) d i −→ M S ( S − Z )[1] . Remark that according to 2.3.1(c), the map i ∗ ( p i ) : Z → i ∗ M S ( S/S − Z ) is an isomorphism.Given any object K in T ( S ), we thus obtain an isomorphism i ∗ ( M S ( S/S − Z ) ⊗ S K ) = i ∗ ( M S ( S/S − Z )) ⊗ Z i ∗ ( K ) ( i ∗ p i ) − −−−−−→ Z ⊗ Z i ∗ ( K ) = i ∗ ( K )which is natural in K . It induces by adjunction a map(2.3.14.2) ψ i,K : M S ( S/S − Z ) ⊗ S K → i ∗ i ∗ ( K )which is natural in K .For any P -scheme X/S , we put M S ( X/X − X Z ) = M S ( S/S − Z ) ⊗ S M S ( X ) so that we get from(2.3.14.1) a canonical distinguished triangle: M S ( X − X Z ) j X ∗ −−→ M S ( X ) → M S ( X/X − X Z ) → M S ( X − X Z )[1] . The map (2.3.14.2) for K = M S ( X ) gives a canonical map(2.3.14.3) ψ i,X : M S ( X/X − X Z ) → i ∗ ( M Z ( X Z )) . Proposition . Consider the previous hypothesis and notations. Then the followingconditions are equivalent:(i) T satisfies the property (Loc i ).(ii) (a) The functor i ∗ is conservative.(b) The morphism ψ i,S : M S ( S/S − Z ) → i ∗ ( Z ) is an isomorphism.(c) For any object K of T ( S ) , the exchange transformation Ex ( i ∗∗ , ⊗ ) : ( i ∗ Z ) ⊗ S K → i ∗ i ∗ K is an isomorphism.(iii) (a) The functor i ∗ is conservative.(b) The morphism ψ i,S : M S ( S/S − Z ) → i ∗ ( Z ) is an isomorphism. . TRIANGULATED P -FIBRED CATEGORIES IN ALGEBRAIC GEOMETRY 37 (c) For any objects K and L of T ( S ) , the exchange transformation Ex ( i ∗∗ , ⊗ ) : ( i ∗ K ) ⊗ S L → i ∗ ( K ⊗ Z i ∗ L ) is an isomorphism.Assume in addition that T is well generated and τ -twisted as a triangulated P -fibred category.Then the above conditions are equivalent to the following one:(iv) (a) The functor i ∗ is conservative, commutes with direct sums and with τ -twists.(b) The morphism ψ i,X : M S ( X/X − X Z ) → i ∗ ( M Z ( X Z )) is an isomorphism for any P -scheme X/S . In particular, (Loc i ) implies that for any object K of T ( S ), the localization triangle of 2.3.3 j ♯ j ∗ ( K ) → K → i ∗ i ∗ ( K ) ∂ K −−→ j ♯ j ∗ ( K )[1]is canonically isomorphic (through exchange transformations) to the triangle (2.3.14.1) tensoredwith K . Proof. ( i ) ⇒ ( iii ) : According to (Loc i ) (a), we need only to check that the maps in (iii)(b) and(iii)(c) are isomorphisms after applying i ∗ and j ∗ . This follows easily from (Loc i ) (b).( iii ) ⇒ ( ii ) : Obvious( ii ) ⇒ ( i ) : According to (ii)(b), the distinguished triangle (2.3.14.1) is isomorphic to a triangleof the form j ♯ j ∗ ( S ) ad ′ ( j ♯ ,j ∗ ) −−−−−−→ S ad ( i ∗ ,i ∗ ) −−−−−→ i ∗ i ∗ ( S ) → j ♯ j ∗ ( S ) . According to (ii)(c), this latter triangle tensored with K is isomorphic through exchange transfor-mations to a triangle of the form j ♯ j ∗ ( K ) ad ′ ( j ♯ ,j ∗ ) −−−−−−→ K ad ( i ∗ ,i ∗ ) −−−−−→ i ∗ i ∗ ( K ) → j ♯ j ∗ ( K ) . Thus Lemma 2.3.5 allows to conclude.To end the proof, we remark by using the equations for the adjunction ( i ∗ , i ∗ ) that for anyobject M of T ( S ), the following diagram is commutative: i ∗ i ∗ ( S ) ⊗ K i ∗ ( Z ) ⊗ K Ex ( i ∗∗ , ⊗ ) (cid:15) (cid:15) M S ( S/S − Z ) ⊗ K ψ i ⊗ K ❣❣❣❣❣❣❣❣❣❣❣ ψ i,K + + ❲❲❲❲❲❲❲❲❲❲❲❲❲❲ i ∗ i ∗ ( K ) i ∗ ( Z ⊗ i ∗ i ∗ ( K )) . Note that (i) implies that i ∗ is conservative and commutes with direct sums (see 2.3.3) and (ii)(c)implies it commutes with twists. According to the above diagram, (ii)(b) implies (iv)(b).We prove that reciprocally that (iv) implies (ii). Because (ii)(b) (resp. (ii)(a)) is a particular caseof (iv)(b) (resp. (iv)(a)), we have only to prove (ii)(b). In view of the previous diagram, we arereduced to prove that for any object K of T ( S ), the map ψ i,K is an isomorphism. Consider thefull subcategory U of T ( S ) made of the objects K such that ψ i,K is an isomorphism. Then U istriangulated. Using (iv)(a), U is stable by small sums and τ -twists. By assumption, it containsthe objects of the form M S ( X ) for a P -scheme X/S . Thus, because T is well generated byassumption, Lemma 1.3.18 concludes. (cid:3) Lemma . Consider a closed immersion i : Z → S . We assume the following conditionsare satisfied in addition to that of 2.0: • T is well generated, τ -twisted and satisfies the Zariski separation property. • For any P -scheme X /Z and any point x of X , there exists an open neighbourhood U of x in X and a P -scheme U/S such that U = U × S Z . Then the functor i ∗ is conservative. This property is trivial when P is the class of open immersions or the class of morphisms of finite type in S . It is also true when P is the class of ´etale morphism or P = Sm ( cf. [ EGA4 , 18.1.1]). Proof.
Consider an object K of T ( Z ) such that i ∗ ( K ) = 0. We prove that K = 0.Because T is τ -generated, it is sufficient to prove that for a P -morphism p : X → Z and atwist ( n, m ) ∈ Z × τ , Hom T ( Z ) ( M Z ( X ) { m } [ n ] , K ) = 0 . Because M Z ( X ) = p ♯ ( X ), this equivalent to prove thatHom T ( X ) ( X { m } [ n ] , p ∗ ( K )) = 0 . Using the Zariski separation property on T , this latter assumption is local in X . Thus, accordingto the assumption on the class P , we can assume there exists a P -scheme X/S such that X = X × S Z . Thus M Z ( X ) { m } [ n ] = i ∗ ( M S ( X ) { m } [ n ]) and the initial assumption on K allows toconclude. (cid:3) Note for future applications the following interesting corollaries:
Corollary . Assume T is a premotivic triangulated category which is compactly τ -generated for a group of twists τ ( i.e. any twists in τ admits a tensor inverse) and which satisfiesthe Zariski separation property.Then, for any closed immersion i , the functor i ∗ is conservative, commutes with sums andwith twists. This is a consequence of lemmas 2.3.16 and 2.2.16. In fact, under these conditions, i ∗ commuteswith arbitrary τ -twists because it is true for its (left) adjoint i ∗ . Corollary . Assume T satisfies the assumptions of the preceding corollary. Then thefollowing conditions on a closed immersion i are equivalent:(i) T satisfies the property (Loc i ).(ii) For any scheme S in S and any smooth S -scheme X , the map (2.3.14.3) ψ i,X : M S ( X/X − X Z ) → i ∗ M Z ( X Z ) is an isomorphism. We finish this section with the following useful result:
Proposition . Assume T is τ -twisted and consider a τ ′ -twisted triangulated P -fibredcategory T ′ and a morphism ϕ ∗ : ( T , τ ) ⇄ ( T ′ , τ ′ ) : ϕ ∗ compatible with twists. We assume the following properties:(a) The map τ → τ ′ induced by ϕ ∗ is (essentially) surjective.(b) T ′ is well generated.We consider a closed immersion i : Z → S and further assume the following properties:(c) T satisfies the property (Loc i ).(d) The exchange transformation Ex ( ϕ ∗ , i ∗ ) : ϕ ∗ i ∗ → i ∗ ϕ ∗ is an isomorphism.(e) The functor i ∗ : T ′ ( Z ) → T ′ ( S ) commutes with τ ′ -twists. Then T ′ satisfies the property (Loc i ). Proof.
Note that, under the above assumptions, ϕ ∗ is conservative (in fact, for any P -scheme X/S and any twists i ∈ τ ′ , the premotive M S ( X ) { i } is in the essential image of ϕ ∗ ).Thus, if i ∗ : T ( Z ) → T ( S ) is conservative (resp. commute with sums), then i ∗ : T ′ ( S ) → T ′ ( S )is conservative (resp. commute with sums) using the isomorphism ϕ ∗ i ∗ ≃ i ∗ ϕ ∗ .Let M (resp. M ′ ) be the geometric sections of T (resp. T ′ ). As in 2.3.14, we fix a distinguishedtriangle M S ( S − Z ) j ∗ −→ S p i −→ M S ( S/S − Z ) d i −→ M S ( S − Z )[1] . This will be satisfied if any τ ′ -twists is invertible because the left adjoint of i ∗ commutes with τ ′ -twists. . TRIANGULATED P -FIBRED CATEGORIES IN ALGEBRAIC GEOMETRY 39 and we put M ′ S ( S/S − Z ) = ϕ ∗ M S ( S/S − Z ). According to loc. cit. , we thus get for any P -scheme X/S canonical maps ψ i,X : M S ( X/X − X Z ) → i ∗ M Z ( X Z ) ,ψ ′ i,X : M ′ S ( X/X − X Z ) → i ∗ M ′ Z ( X Z ) . By construction, the following diagram is commutative: ϕ ∗ M S ( X/X − X Z ) ϕ ∗ ψ i,X / / ϕ ∗ i ∗ M Z ( X Z ) Ex ( ϕ ∗ ,i ∗ ) / / i ∗ ϕ ∗ M Z ( X Z ) M ′ S ( X/X − X Z ) ψ ′ i,X / / M ′ Z ( X Z )Thus, Proposition 2.3.15 allows to conclude. (cid:3) Recall we assume P = Sm in this section.2.4.a. The stability property.
The following section is directly inspired by the work of Ayoubin [
Ayo07a , § We claim no originality except for a closer look on the needed axioms.
Definition . A pointed smooth S -scheme will be a couple ( f, s ) of morphisms of S suchthat f : X → S is a smooth separated morphism of finite type and s : S → X is a section of f .We associate with a pointed smooth scheme ( f, s ) the following endofunctor of T ( S ) T h ( f, s ) := f ♯ s ∗ called the associated Thom transformation .If T satisfies (Adj s ) (recall: s ∗ admits a right adjoint denoted by s ! ), we put T h ′ ( f, s ) := s ! f ∗ and call it the associated adjoint Thom transformation . Remark . Note that because f is separated, s is a closed immersion. Example . (1) Let p : E → X be a vector bundle and s be its zero section. Fol-lowing [ Ayo07a ], we put T h ( E ) := T h ( p, s ) and call it simply the Thom transformationassociated with E/X .(2) Consider a pointed smooth S -scheme ( f, s ) such that f is ´etale. Then s is an openand closed immersion. Thus, if T is additive, s ∗ = s ♯ according to Lemma 2.2.2. Inparticular, T h ( f, s ) = Id S . Definition . We will say that T satisfies the stability property , denoted by (Stab), if forany point smooth scheme ( f, s ), the Thom transformation T h ( f, s ) is an equivalence of categories.2.4.5 . Consider a commutative diagram in S of the form S t ●●●●●●●● t ′ (cid:15) (cid:15) Y ′ s ′ / / p ′ (cid:15) (cid:15) ∆ Y g ❋❋❋❋❋❋❋ p (cid:15) (cid:15) S s / / X f / / S (2.4.5.1)such that ∆ is a cartesian square, ( f, s ), ( g, t ) are smooth pointed schemes and g is a smoothseparated morphism of finite type. Then we get a canonical exchange morphism:(2.4.5.2) T h ( g, t ) = f ♯ p ♯ s ′∗ t ′∗ Ex (∆ ♯ ∗ ) −−−−−−−→ f ♯ s ∗ p ′ ♯ t ′∗ = T h ( f, s ) T h ( p ′ , t ′ ) . This is an isomorphism as soon as Ex (∆ ♯ ∗ ) is an isomorphism. The following lemma gives asufficient condition for this to happen. See also [
Del01 , § Lemma . Consider the above notations. If T satisfies (Loc s ) then the natural transfor-mations Ex (∆ ♯ ∗ ) is an isomorphism for any square ∆ as above. This lemma follows easily from the definition of (Loc s ), the relations of paragraph 2.3.1 andthe P -base change formula ( P -BC). It motivates the next definition: Definition . We say that T satisfies the weak localization property (wLoc) if it satisfies(Loc s ) for any closed immersion s which admits a smooth retraction. Proposition . Assume that T satisfies the Nisnevich separation property. Then thefollowing conditions are equivalent:(i) T satisfies (wLoc).(ii) For any scheme S and any closed immersion i : Z → X between smooth S -schemes, T satisfies (Loc i ). Proof.
Of course, (ii) implies (i). We prove the reciprocal statement. The Nisnevich sepa-ration property says that for any Nisnevich cover f : X ′ → X , the functor f ∗ is conservative. Wededuce from that point the properties (Loc i ) (a) and (Loc i ) (b) are local in X with respect tothe Nisnevich topology – for (b), one also uses the smooth projection formula. Thus, we can con-clude as locally for the Nisnevich topology, i admits a smooth retraction (see for example [ D´eg07 ,4.5.11]). (cid:3)
Applying the second point of Example 2.4.3, we easily deduce from that construction thefollowing kind of excision property:
Lemma . Assume that T satisfies (wLoc).Then, given any diagram (2.4.5.1) satisfying the assumption as above and such that p is ´etale,the natural transformation (2.4.5.2) gives an isomorphism: T h ( g, t ) ∼ −−→ T h ( f, s ) . . To any short exact sequence of vector bundles over a scheme S ( σ ) 0 → E ′ ν −→ E π −→ E ′′ → , we can associate a commutative diagram S ! ! ❈❈❈❈❈❈❈ (cid:15) (cid:15) E ′ ν / / (cid:15) (cid:15) ∆ E ❆❆❆❆❆❆ π (cid:15) (cid:15) S / / E ′′ / / S where the non labeled map are either the canonical projections or the zero sections of the rel-evant vector bundles, and ∆ is cartesian. Using the notation of Example 2.4.3, the exchangetransformation (2.4.5.2) associated with this diagram has the following form: T h ( σ ) : T h ( E ) −→ T h ( E ′′ ) ◦ T h ( E ′ ) . Recall from the above that this natural transformation is an isomorphism as soon as T satisfies(wLoc). Proposition . Assume T satisfies (wLoc) and ( Zar -sep). Then the following conditionsare equivalent:(i) The complete triangulated Sm-fibred category T satisfies the stability property.(ii) For any scheme S , the Thom transformation T h ( A S ) is an equivalence of categories. Proof.
We have to prove that (ii) implies (i). Note that according to the above paragraph,we already now that for any scheme S and any integer n ≥ T h ( A nS ) ≃ T h ( A S ) ◦ ,n is anequivalence. . TRIANGULATED P -FIBRED CATEGORIES IN ALGEBRAIC GEOMETRY 41 We consider a smooth pointed scheme ( f : X → S, s ) and we prove that T h ( f, s ) is anequivalence.Recall that (Loc s ) implies (Adj) s (first point of Proposition 2.3.3). In particular, T h ( f, s )admits a right adjoint T h ′ ( f, s ) and we have to prove that the adjunction morphisms are isomor-phisms.Consider an open immersion j : U → S and let ( f , s ) be the restriction of the smooth S -point ( f, s ) over U . Property (Loc s ) implies (BC s ) (Corollary 2.3.13). Thus, using also property( P -BC), we obtain a canonical isomorphism: j ∗ T h ( f, s ) ∼ −−→ T h ( f , s ) j ∗ . Recall also that (Loc s ) implies (Supp s ) (again Corollary 2.3.13). Thus we get a canonical isomor-phism: j ♯ T h ( f , s ) ∼ −−→ T h ( f, s ) j ♯ which gives by adjunction an isomorphism: T h ′ ( f , s ) j ∗ ∼ −−→ j ∗ T h ′ ( f, s ) . Thus, (Zar-sep) shows that the property for T h ( f, s ) to be an equivalence is Zariski local in S .Consider a point a ∈ S , x = s ( a ). As X is smooth over S , there exists an open subscheme U ⊂ X , an integer n ≥ S -morphism π : U → A nS which fits into the followingcartesian square: S / / (cid:15) (cid:15) U π (cid:15) (cid:15) S ν / / A nS where ν is the zero section ( cf. [ EGA4 , 17.12.2]). Note that the scheme S = s − ( U ) is an openneighborhood of a in S . Let us put X = f − ( S ) and U = U ∩ X . Then we get the followingcommutative diagram: X f ' ' ❖❖❖❖❖❖❖❖❖ S s ♦♦♦♦♦♦♦♦♦ ν ' ' ◆◆◆◆◆◆◆◆◆ s ′ / / U ?(cid:31) O O π (cid:15) (cid:15) f ′ / / S A nS ♣♣♣♣♣♣♣♣♣ where π is the restriction of π above S and ν is again the zero section. According to Lemma2.4.9, we get isomorphisms T h ( f , s ) ≃ T h ( f ′ , s ′ ) ≃ T h ( A nS ) . Thus, according to the beginning of the proof, T h ( f , s ) is an equivalence. This concludes because S is an open neighborhood of a in S . (cid:3) Definition . Assume that T is monoidal.(1) For any smooth pointed scheme ( f : X → S, s ), we put M S (cid:0) X/X − s ( S ) (cid:1) := f ♯ s ∗ ( S ).(2) For any vector bundle E/S with projection f and zero section s , we define the Thompremotive associated with E over S as M T h S ( E ) = f ♯ s ∗ ( S ).2.4.13 . We assume T is monoidal and satisfies properties (wLoc) and (Zar-sep).In each case of the previous definition, if we apply f ♯ to the distinguished triangle obtainedfrom point (2) of Proposition 2.3.3 applied to s , we get the following canonical distinguishedtriangles: M S (cid:0) X − s ( S ) (cid:1) → M S ( X ) → M S (cid:0) X/X − s ( S ) (cid:1) +1 −−→ M S ( E × ) → M S ( E ) → M T h S ( E ) +1 −−→ where the first map is induced by the obvious open immersion. Moreover, property (Loc s ) implies (PF s ) (see Corollary 2.3.13). Thus for any premotive K over S , the following composite map is an isomorphism: T h ( f, s ) .K = f ♯ s ∗ ( K ) = f ♯ s ∗ ( S ⊗ S s ∗ f ∗ ( K )) Ex ( s ∗∗ , ⊗ ) − −−−−−−−−→ f ♯ ( s ∗ ( S ) ⊗ X f ∗ ( K )) Ex ( f ∗ ♯ , ⊗ ) −−−−−−→ ( f ♯ s ∗ ( S )) ⊗ S K = M S ( X/X − s ( S )) ⊗ S K (2.4.13.1)Similarly, in the case of a vector bundle E/S , we get a canonical isomorphism: T h ( E ) .K ∼ −−→ M T h S ( E ) ⊗ S K. From these isomorphisms, we deduce easily the following corollary of the previous proposition:
Corollary . Consider the above notations and assumptions. Then the following prop-erties are equivalent:(i) T satisfies the stability property.(ii) For any smooth pointed scheme ( X → S, s ) , the premotive M S ( X/X − s ( S )) is ⊗ -invertible.(iii) For any vector bundle E/S the Thom premotive
M T h S ( E ) is ⊗ -invertible.(iv) For any scheme S , the premotive M T h S ( A S ) is ⊗ -invertible. Remark . Assume that T satisfies the assumptions and the equivalent conditions ofthe previous corollary. Then, under the notations of Paragraph 2.4.10, we associate with the exactsequence ( σ ) a canonical isomorphism(2.4.15.1) T h S ( σ ) : M T h S ( E ) → M T h S ( E ′′ ) ⊗ S M T h S ( E ′ ) . Recall that Deligne introduced in [
Del87 , 4.12] the Picard category K ( S ) of virtual vector bundle over a scheme S .Then, it follows from the above isomorphism and the universal properties of K ( S ) (see [ Del87 ,4.3]) that the functor
M T h S can be extended uniquely to a symmetric monoidal functor: M T h S : K ( S ) → T ( S ) . The reader is refered to [
Ayo07a , th. 1.5.18] for a detailed argument.2.4.16 . Assume T is monoidal. For any scheme S , the canonical projection p : P S → S is a split epimorphism. A splitting is given by the inclusion of the infinite point ν : S → P S .The induced map p ∗ : M S ( P S ) → S is a split epimorphism. Thus it admits a kernel K in thetriangulated category T ( S ). Definition . Under the above assumption and notations, we define the
Tate premotive over S as the object S (1) = K [ −
2] of T ( S ).The monoid generated by the cartesian section ( S ) S defines a canonical N -twist on T calledthe Tate twist . The n -th Tate twist of an object K is denoted by K ( n ).2.4.18 . Consider again the assumption of Paragraph 2.4.13.According to Lemma 2.4.9, we get a canonical isomorphism
M T h S ( A S ) = M S ( A S / A S − { } ) → M S ( P S / P S − { } ) . On the other hand, S (1)[2] is by definition the cokernel of the monomorphism ν ∗ : S → M S ( P S ). Thus we get a canonical morphism:(2.4.18.1) S (1)[2] → M S ( P S / P S − { } ) ∼ −−→ M T h S ( A S ) . From this definition and Corollary 2.4.14 the following result is obvious:
Corollary . Consider the above assumption and notations. Then the following con-ditions are equivalent:(i) T satisfies the homotopy property.(ii) For any scheme S , the arrow (2.4.18.1) is an isomorphism.When these equivalent assertions are satisfied, the following conditions are equivalent: . TRIANGULATED P -FIBRED CATEGORIES IN ALGEBRAIC GEOMETRY 43 (iii) T satisfies the stability property.(iv) For any scheme S , the Tate premotive S (1) is ⊗ -invertible. If S (1) is ⊗ -invertible, we will consider the Tate twist on T as a Z -twist.2.4.b. The purity property. . Let f : X → S be a smooth proper morphism in S . We consider the followingcartesian square: X × S X f ′′ / / f ′ (cid:15) (cid:15) ∆ X f (cid:15) (cid:15) X f / / S (2.4.20.1)where f ′ (resp. f ′′ ) is the projection on the first (resp. second) factor. Let δ : X → X × S X bethe diagonal embedding. Note that ( f ′ , δ ) is a smooth pointed scheme which depends only on f .We put: Σ f := T h ( f ′ , δ ) = f ′ ♯ δ ∗ . We then define a canonical morphism: p f : f ♯ = f ♯ f ′′∗ δ ∗ Ex (∆ ♯ ∗ ) −−−−−→ f ∗ f ′ ♯ δ ∗ = f ∗ ◦ Σ f using the exchange transformation introduced in paragraph 1.1.15. Definition . We say that f is T -pure , or simply pure when T is clear, when thefollowing conditions are satisfied:(1) The natural transformation Σ f is an equivalence.(2) The morphism p f : f ♯ → f ∗ ◦ Σ f is an isomorphism.Then p f is called the purity isomorphism associated with f . We say also that f is universally T -pure if f is pure after any base change along a morphism of S .We introduce the following properties on T : • T satisfies the purity property (Pur) if any proper smooth morphism is pure. • T satisfies the weak purity property (wPur) if for any scheme S and any integer n > p n : P nS → S is pure. Remark . Consider the above notations and assume f is pure.Then f ∗ admits a right adjoint f ! and we deduce by transposition from p f a canonical iso-morphism: p ′ f : f ∗ → Σ − f ◦ f ! . Recall also that, when δ ∗ admits a right adjoint δ ! , Σ f admits as a right adjoint the transformationΩ f := δ ! f ∗ . In particular, Ω f = Σ − f .The following lemma shows the importance of the purity property. Lemma . Assume that T satisfies (wLoc). Let f : Y → X be a proper smooth morphism.If f is universally pure then the following conditions hold:(1) T satisfies (Supp f ) and (BC f ).(2) For any cartesian square Z ˜ f / / h (cid:15) (cid:15) ∆ Y g (cid:15) (cid:15) X f / / S such that g is smooth, the exchange transformation: Ex (∆ ♯ ∗ ) : g ♯ ˜ f ∗ → f ∗ h ♯ is an isomorphism. (3) If moreover T is monoidal then T satisfies (PF f ). Proof.
We first prove condition (2). By assumption, the natural transformation Σ ˜ f is anequivalence. for f and ˜ f : by assumption the natural transformations Σ f = f ′ ♯ δ ∗ and Σ ˜ f = ˜ f ′ ˜ δ ∗ )are equivalences. Thus, it is sufficient to prove that the natural transformation g ♯ ˜ f ∗ Σ ˜ f Ex (∆ ♯ ∗ ) −−−−−→ f ∗ h ♯ Σ ˜ f is an isomorphism.For matter of notations, let us also introduce the following cartesian squares: Z ˜ δ / / h (cid:15) (cid:15) Γ Z × Y Z ˜ f ′ / / k (cid:15) (cid:15) Θ Z h (cid:15) (cid:15) X δ / / X × S X f ′ / / X using the notations of 2.4.20. Thus, by definition: Σ f = f ′ ♯ δ ∗ , Σ ˜ f = ˜ f ′ ˜ δ ∗ . Then we consider thefollowing diagram of exchange transformations: g ♯ ˜ f ♯ p ˜ f / / g ♯ ˜ f ∗ ˜ f ′ ♯ ˜ δ ∗ Ex (∆ ♯ ∗ ) (cid:15) (cid:15) f ♯ h ♯ p f / / f ∗ f ′ ♯ δ ∗ h ♯ f ∗ f ′ ♯ k ♯ ˜ δ ∗ Ex (Γ ♯ ∗ ) o o f ∗ h ♯ ˜ f ′ ♯ ˜ δ ∗ Note that it only involves exchange transformations of type Ex (? ♯ ∗ ): it is commutative by compat-ibility of these exchange transformations with composition. By assumption, the transformations p f and p ˜ f are isomorphisms. Moreover the property (Loc δ ) is satisfied and it implies (Supp δ )according to Corollary 2.3.13. Thus Ex (Γ ♯ ∗ ) is an isomorphism and this concludes the proof of(2). For condition (1), we note that (2) already implies (Supp f ). Thus we have only to prove(BC f ). We consider a square of shape ∆ as in the statement of the lemma without assuming that g is smooth. We have to prove that Ex (∆ ∗∗ ) : g ∗ f ∗ → ˜ f ∗ h ∗ is an isomorphism. We proceed as for condition (2). It is sufficient to prove that Ex (∆ ∗∗ ) is anisomorphism after composition on the right with Σ f . Then we consider the following commutativediagram of exchange transformations: g ∗ f ♯Ex (∆ ∗ ♯ ) (cid:15) (cid:15) p f / / g ∗ f ∗ f ′ ♯ δ ∗ Ex (∆ ∗∗ ) (cid:15) (cid:15) ˜ f ♯ h ∗ p ˜ f / / ˜ f ∗ ˜ f ′ ♯ ˜ δ ∗ h ∗ ˜ f ∗ ˜ f ′ ♯ k ∗ δ ∗ Ex (Γ ∗∗ ) o o ˜ f ∗ h ∗ f ′ ♯ δ ∗ Ex (Θ ∗ ♯ ) o o According to ( P -BC), Ex (∆ ∗ ♯ ) and Ex (Θ ∗ ♯ ) are isomorphisms. By assumption, p f and p ˜ f areisomorphisms. Moreover, property (Loc δ ) is satisfied and this implies Ex (Γ ∗∗ ) is an isomorphismaccording to Corollary 2.3.13. Condition (1) is proved.It remains to prove (3). We consider again the notations of the cartesian diagram (2.4.20.1).For any premotives K over X and L over S , we consider the following commutative diagram of . TRIANGULATED P -FIBRED CATEGORIES IN ALGEBRAIC GEOMETRY 45 exchange transformations (see Remark 1.1.32): f ♯ (cid:0) K ⊗ f ∗ ( L ) (cid:1) Ex ( f ∗ ♯ , ⊗ ) (cid:15) (cid:15) p f / / f ∗ f ′ ♯ δ ∗ (cid:0) K ⊗ δ ∗ f ′∗ f ∗ ( L ) (cid:1) f ∗ f ′ ♯ (cid:0) δ ∗ ( K ) ⊗ f ′∗ f ∗ ( L ) (cid:1) Ex ( f ′∗ ♯ , ⊗ ) (cid:15) (cid:15) Ex ( δ ∗∗ , ⊗ ) O O f ∗ (cid:0) f ′ ♯ δ ∗ ( K ) ⊗ f ∗ ( L ) (cid:1) f ♯ ( K ) ⊗ L p f / / f ∗ f ′ ♯ δ ∗ ( K ) ⊗ L. Ex ( f ∗∗ , ⊗ ) O O By definition, the exchanges Ex ( f ∗ ♯ , ⊗ ) and Ex ( f ′∗ ♯ , ⊗ ) are isomorphisms. By assumption, thearrows labeled p f are isomorphisms. Moreover, the property (Loc δ ) is satisfied: Corollary 2.3.13implies that Ex ( δ ∗∗ , ⊗ ) is an isomorphism. We deduce from this that the arrow Ex ( f ∗∗ , ⊗ ) is anisomorphism. This concludes the proof of (3) as the functor Σ f = f ′ ♯ δ ∗ is an equivalence accordingto the hypothesis on f . (cid:3) . Assume that T satisfies the support property (Supp). Then we can extend Definition2.4.21 to the case of a smooth separated morphism of finite type f : X → S . We still considerthe cartesian square (2.4.20.1) and the diagonal embedding δ : X → X × S X . Again, ( f ′ , δ ) is asmooth pointed scheme so that we can putΣ f := T h ( f ′ , δ ) = f ′ ♯ δ ∗ and we define a canonical morphism:(2.4.24.1) p f : f ♯ = f ♯ f ′′ ! δ ! Ex (∆ ♯ ! ) −−−−−→ f ! f ′ ♯ δ ! = f ! ◦ Σ f . using the exchange transformation of point (2) in Corollary 2.2.12. Definition . Using the notations above, we say that f is T -pure , or simply pure when T is clear, when the following conditions are satisfied:(1) The natural transformation Σ f is an equivalence.(2) The morphism p f : f ♯ → f ! ◦ Σ f is an isomorphism.We can easily deduce from the construction of the exchange transformation Ex (∆ ♯ ! ) that,when T satisfies properties (Stab) and (Pur), any smooth separated morphism of finite type f ispure. The following theorem is a consequence of the formalism developed previously. Theorem . Assume that T satisfies the localization and weak purity properties. Thenthe following conditions hold:(1) T satisfies the stability property.(2) T satisfies the support and base change properties.If moreover T is monoidal, it satisfies the projection formula.(3) Any smooth separated morphism of finite type is pure.(4) For any projective morphism f , the property (Adj f ) holds.If moreover T is well generated, then the adjoint property holds in general. Proof.
We start by proving condition (1). As (Loc) implies (Zar-sep), we can apply Propo-sition 2.4.11 and we have only to prove that for any scheme S , T h ( A S ) is an equivalence. Let s : S → A S be the zero section and j : A S → P S be the canonical open immersion. Put t = j ◦ s .According to Lemma 2.4.9, j induces an isomorphism T h ( A S ) ≃ T h ( p , s ). Consider now the following cartesian squares: S s / / s (cid:15) (cid:15) P S p / / s ′ (cid:15) (cid:15) ∆ S s (cid:15) (cid:15) P S δ / / P S × S P S p ′ / / P S where p ′ (resp. δ ) is the projection on the first factor (resp. diagonal embedding). The property(Loc s ) implies that s ∗ s ∗ = 1 and that the exchange transformation Ex (∆ ♯ ∗ ) is an isomorphismaccording to Corollary 2.3.13. Thus we get an isomorphism of functors: T h ( p , s ) = p ♯ s ∗ = s ∗ s ∗ p ♯ s ∗ Ex (∆ ♯ ∗ ) − −−−−−−−→ s ∗ p ′ ♯ s ′∗ s ∗ = s ∗ p ′ ♯ δ ∗ s ∗ = s ∗ Σ p s ∗ and this proves (1) because p is pure.Condition (2) follows simply from Corollary 2.3.13. In fact, for any scheme S , the weak purityassumption on T implies that p n : P nS → S is universally pure. Thus, Lemma 2.4.23 impliesproperties (Supp p n ) and (BC p n ) so that we can apply Corollary 2.3.13 to get (Supp) and (BC).The same argument applies to the property (PF) in the monoidal case.For condition (3), we consider a smooth separated morphism of finite type g : Y → S and weprove it is pure. According to (1), Σ g is an equivalence. Thus, by definition of p g , it is sufficientto prove that for any cartesian square: Z ˜ f / / h (cid:15) (cid:15) ∆ Y g (cid:15) (cid:15) X f / / S with f separated of finite type, the exchange transformation Ex (∆ ♯ ! ) : g ♯ ˜ f ! → f ! h ♯ is an isomorphism.To do this, we apply Proposition 2.3.11, as in the case of Corollary 2.3.13. We consider theobvious complete Sm -fibred triangulated categories T ′ and T ′′ over S /S which to an S -scheme Y associates: • T ′ ( Y ) = T ( Y × S X ). • T ′′ ( Y ) = T ( Y ).We consider the morphism ϕ ∗ : T ′ → T ′′ such that for any S -scheme Y , ϕ ∗ Y = ( Y × S p ) ♯ . As forany scheme S , p n : P nS → S is universally pure, Lemma 2.4.23 shows that ϕ ∗ satisfies condition(i) of Proposition 2.3.11. According to that Proposition, (i) is equivalent to condition (iii), and(iii) is precisely what we want.It remains only to prove condition (4). According to property (Pur), any smooth propermorphism f satisfies (Adj f ). According to (Loc) and Proposition 2.3.3 any closed immersion i satisfies (Adj i ). It follows easily that any projective morphism f satisfies (Adj f ). When T is wellgenerated, we simply apply point (4) of Corollary 2.3.13. (cid:3) Remark . In particular, in the assumption of the previous theorem, if T satisfiesproperties (Loc), (wPur) and (Adj) , we can apply Theorem 2.2.14 to T so that we get a completeformalism of operations ( f ∗ , f ∗ , f ! , f ! ) satisfying all the desired formulas.Thus the preceding theorem gives another look at the main result of [ Ayo07a , 1.4.2]. In fact,the proof given here is simpler as the assumptions of our theorem are stronger. However, we donot use the homotopy property in our theorem.We end up this section with the theorem of Ayoub [
Ayo07a , 1.4.2], which can be stated in asimpler form according to the preceding theorem: Note that under the assumptions of the previous theorem, we know that for any proper smooth morphism f , f ∗ admits a right adjoint. The same is true for a proper morphism which can be factorized as a closed immersionfollowed by a smooth proper morphism according to (Loc). . TRIANGULATED P -FIBRED CATEGORIES IN ALGEBRAIC GEOMETRY 47 Theorem . Assume T satisfies the localization, homotopy and stability prop-erties.Then T is weakly pure. In fact, this theorem is proved explicitly in op. cit. , Theorem 1.7.9.
Remark . Recall that Ayoub proves more than just this theorem: indeed he constructsthe whole formalism of the 6 functors for quasi-projective morphisms for his monoidal homotopystable functors – see again [
Ayo07a ]. The work we have done here is to isolate the crucialproperties of purity and weak purity. Also, using the construction of Deligne, we have showed howto avoid the assumption of quasi-projectiveness made by Ayoub. Finally, the interest of Theorem2.4.26 is to give a possible approach to the without requiring the homotopyproperty ; this is a question which has been indirectly adressed by many mathematicians (Bloch,Esnault, Barbieri-Viale, ...)2.4.c.
Duality, purity and orientation. . This section is concerned with the relation between purity and duality. We will assumethat T is premotivic.Recall that an object M of a monoidal category M is called strongly dualizable if there existsan object M ′ such that ( M ′ ⊗ − ) is both right and left adjoint to ( M ⊗ − ). Then, M ′ is calledthe strong dual of M .In case M is closed monoidal, we will say that a morphism of the form µ : M ⊗ M ′ → is a perfect pairing if the natural transformation( M ⊗ − ) → Hom ( M ′ , − )obtained from µ by adjunction is an isomorphism. Then M is strongly dualizable with dual M ′ . Proposition . Let f : X → S be a smooth proper morphism.If f is pure then the premotive M S ( X ) is strongly dualizable in T ( S ) with dual: f ∗ ( X ) ≃ f ♯ (cid:0) Ω f ( X ) (cid:1) where Ω f denotes the inverse of Σ f . Proof.
By assumption, Σ f is an automorphism of the category T ( X ). Moreover, the identi-fication (2.4.13.1) can be rewritten as Σ f ( M ) = Σ f ( X ) ⊗ X M for any premotive M over X . Thefact Σ f is an equivalence means that Σ f ( X ) is a ⊗ -invertible object, whose inverse is T := Ω f ( S ).In particular, we get: Ω f ( M ) = T ⊗ M .According to the Sm -projection formula, the functor M S ( X ) ⊗ . is isomorphic to f ♯ f ∗ . Thus,its right adjoint is f ∗ f ∗ . As f is pure by assumption, this last functor is isomorphic to f ♯ Ω f f ∗ .Using the observation at the beginning of the proof and the Sm -projection formula again, weobtain: f ♯ Ω f f ∗ ( N ) = f ♯ ( T ⊗ f ∗ ( N )) = f ♯ ( T ) ⊗ N. Moreover, the right adjoint of f ♯ Ω f f ∗ is f ∗ Σ f f ∗ . Using again the purity isomorphism for f , thislast functor can be identified with f ♯ f ∗ and this concludes. (cid:3) . Assume again that the premotivic triangulated category T satisfies properties (wLoc)and (Nis-sep).Let S be a scheme. A smooth closed S -pair will be pair ( X, Z ) of smooth S -schemes such that Z is closed subscheme of X . We consider the canonical projection p : X → S and the immersion i : Z → X associated with ( X, Z ). Note that according to Proposition 2.4.8, T satisfies property(Loc i ). Then we define the premotive of ( X, Z ) as follows:(2.4.32.1) M S ( X/X − Z ) := p ♯ i ∗ ( Z ) . According to property (Loc i ), we thus get a canonical distinguished triangle:(2.4.32.2) M S ( X − Z ) j ∗ −→ M S ( X ) → M S ( X/X − Z ) +1 −−→ Note that given any smooth morphism p : S → S , we get obviously:(2.4.32.3) p ♯ M S ( X/X − Z ) = M S ( X/X − Z ) . Moreover, given any morphism f : T → S , we get an exchange isomorphism:(2.4.32.4) f ∗ M S ( X/X − Z ) ∼ −→ M T ( X T /X T − Z T ) . A morphism of smooth closed S -pairs ( Y, T ) → ( X, Z ) will be a couple ( f, g ) which fits into acommutative diagram T k / / g (cid:15) (cid:15) ∆ Y f (cid:15) (cid:15) Z i / / X, with i , k the canonical immersions, and such that T = f − ( Z ) as a set. We can associate with( f, g ) a morphism of premotives: M S ( Y /Y − T ) = q ♯ k ∗ g ∗ ( Z ) Ex (∆ ∗∗ ) − −−−−−−−→ q ♯ f ∗ i ∗ ( Z ) Ex ∗ ♯ −−−→ ∗ p ♯ i ∗ ( Z ) = M S ( X/X − Z ) . Indeed, the exchange map Ex (∆ ∗∗ ) is an isomorphism according to (Loc i ) and Corollary 2.3.13.It is easy to check that the triangle (2.4.32.2) is functorial with respect to morphisms of closed S -pairs. Before proving the next theorem, we state the following lemma. Lemma . Consider the assumptions and notations above.Let ( f, g ) : ( Y, T ) → ( X, Z ) be a morphism of smooth closed S -pairs such that f is ´etale and g is an isomorphism. Then the induced map M S ( Y /Y − T ) → M S ( X/X − Z ) is an isomorphism. Proof.
According to the identification 2.4.32.3, it is sufficient to treat the case where X = Z .Let U = X − Z and j : U → X be the obvious immersion. Then ( f, j ) is a Nisnevich cover of X .According to (Nis-sep), it is sufficient to prove that the pullback of M X ( Y /Y − T ) → M X ( X/X − Z )along f and j is an isomorphism. This is obvious using 2.4.32.4. (cid:3) . We consider again the assumption of the paragraph preceding the above lemma.Fix a smooth closed S -pair ( X, Z ). Let B Z X (resp. B Z ( A X ) be the blow-up of X (resp. A X )with center in Z (resp. { } × Z ). We define the deformation space associated with ( X, Z ) as the S -scheme D Z X = B Z ( A X ) − B Z X . Note also D Z Z = A Z is a closed subscheme of D Z X ; thecouple ( D Z X, A Z ) is a smooth closed S -pair.Let N Z X be the normal bundle of Z in X . The scheme D Z X is fibered over A . Moreover, the0-fiber of ( D Z X, A ) is the closed pair ( N Z X, Z ) corresponding to the zero section and the 1-fiberis the closed pair (
X, Z ). In particular, we get the following morphisms of closed pairs:(2.4.34.1) (
X, Z ) d −→ ( D Z X, A Z ) d ←− ( N Z X, Z )We are now ready to state the purity theorem for smooth closed pairs in our abstract formalism.Though our assumptions are more general, this theorem follows exactly from the method of Moreland Voevodsky used to prove this result in the homotopy category H (see [ MV99 , §
3, 2.24]):
Theorem . Consider the above assumptions and notations and suppose that T satisfiesthe homotopy property. Then the morphisms M S ( X/X − Z ) d ∗ −−→ M S ( D Z X/D Z X − A Z ) d ∗ ←−− M S ( N Z X/N × Z X ) =: M T h S ( N Z X ) . are isomorphisms. Proof.
By noetherian induction and the preceding lemma, the statement is local in X forthe Nisnevich topology. Thus, because ( X, Z ) is a smooth closed S -pair, we can assume that thereexists an ´etale map π : X → A n + cS such that π − ( A cS ) = Z – cf. [ EGA4 , 17.12.2]. Consider thepullback square X ′ p / / q (cid:15) (cid:15) X π (cid:15) (cid:15) A n × Z × π | Z / / A n × A cS . . TRIANGULATED P -FIBRED CATEGORIES IN ALGEBRAIC GEOMETRY 49 There is an obvious closed immersion Z → X ′ and its image is contained in q − ( Z ). As q is ´etale, Z is a direct factor of q − ( Z ). Put W = q − ( Z ) − Z and Ω = X ′ − W . Thus Ω is an opensubscheme of X ′ , and the reader can check that p and q induces morphisms of smooth closed S -pairs ( X, Z ) ← (Ω , Z ) → ( A nZ , Z ) . Applying again the preceding lemma, these morphisms induces isomorphisms on the associatedpremotives. Thus we are reduced to the case of the closed S -pair ( A nZ , Z ). A direct computationshows that D Z ( A nZ ) ≃ A × A nZ . Under this isomorphism d (resp. d ) corresponds to the 0-section (resp. 1-section) of A × A nZ corresponding to the first factor. Thus, we conclude usingthe homotopy property. (cid:3) . The interest of the previous theorem is to simplify the purity isomorphism. Let usrestate the assumptions on the triangulated premotivic category T : • T satisfies properties (Nis-sep), (wLoc) and (Htp).Then applying the above theorem, we get for any smooth closed S -pair ( X, Z ) a canonical isomor-phism(2.4.36.1) p X,Z : M S ( X/X − Z ) → M T h S ( N Z X ) Corollary . Consider the assumptions and notations above.(1) For any smooth pointed S -scheme ( f, s ) and any premotive K over S , we get a canonicalisomorphism T h ( f, s ) .K ≃ M S ( X/X − s ( S )) ⊗ S K p X,S −−−→
M T h S ( N s ) ⊗ S K. where the first isomorphism is given by the map (2.4.13.1) and N s is the normal bundleof s .(2) For any smooth separated morphism of finite type f : X → S with tangent bundle T f ,and any premotive K over X , we get a canonical isomorphism: p XX,X : Σ f ( K ) ∼ −−→ M T h X ( T f ) ⊗ X K – here, ( XX, X ) stands for the closed pair corresponding to the diagonal embedding of X/S . In the assumption of point (2), we thus get a canonical map:(2.4.37.1) f ♯ ( K ) p f −→ f ! (Σ f K ) ∼ −−→ f ! (cid:0) M T h X ( T f ) ⊗ X K (cid:1) that we will still denote by p f and call the purity isomorphism associated with f . Definition . Assume the triangulated premotivic category T satisfies (wLoc). Asusual, M (1) denotes the Tate twist of a premotive M .An orientation t of T will be the data for each smooth scheme X and each vector bundle E/X of rank n of an isomorphism t E : M T h X ( E ) → X ( n )[2 n ] , called the Thom isomorphism , satisfying the following coherence properties:(a) Given a scheme X and an isomorphism of vector bundles ϕ : E → F of ranks n over X ,the following diagram is commutative: M T h X ( E ) t E ) ) ❘❘❘❘❘❘❘❘❘ ϕ ∗ / / M T h X ( F ) . t F u u ❦❦❦❦❦❦❦❦❦ X ( n )[2 n ] We define T f as the normal bundle of the diagonal immersion δ : X → X × S X . (b) For any morphism f : Y → X of schemes, and any vector bundle E/X of rank n withpullback F over Y , the following diagram commutes: f ∗ ( M T h X ( E )) ∼ (cid:15) (cid:15) f ∗ t E / / f ∗ ( X ( n )[2 n ]) ∼ (cid:15) (cid:15) M T h Y ( F ) t F / / Y ( n )[2 n ]where the vertical maps are the canonical isomorphisms.(c) For any scheme X and any exact sequence ( σ ) of vector bundles over X → E ′ ν −→ E π −→ E ′′ → , if n (resp. m ) denotes the rank of the vector bundle E ′ (resp. E ′′ ), the following diagramcommutes: M T h X ( E ) t E (cid:15) (cid:15) T h X ( σ ) / / M T h X ( E ′ ) ⊗ M T h X ( E ′′ ) t E ′ ⊗ t E ′′ (cid:15) (cid:15) X ( n + m )[2 n + 2 m ] / / X ( n )[2 n ] ⊗ X ( m )[2 m ]where the map T h X ( σ ) is the isomorphism (2.4.15.1) associated with ( σ ) and the bottomvertical one is the obvious identification.We will also say that T is oriented when the choice of one particular orientation is not essential.Note that the Thom isomorphism can be viewed as a cohomology class in H n,n T ( T h X ( E )) := Hom T ( X ) (cid:0) M T h X ( E ) , S ( n )[2 n ] (cid:1) which in classical homotopy theory is called the Thom class .2.4.39 . Suppose the triangulated premotivic category T satisfies the following properties: • T satisfies properties (Nis-sep), (wLoc), (Htp). • T admits an orientation t .Consider a smooth closed S -pair ( X, Z ) of codimension n . Let p (resp. q ) be the structuralmorphism of X/S (resp.
Z/S ) and i : Z → X the associated immersion. Then we associate with( X, Z ) the following form of the purity isomorphism:(2.4.39.1) p t X,Z : M S ( X/X − Z ) p X,Z −−−→
M T h S ( N Z X ) q ♯ ( t NZX ) −−−−−−→ M S ( Z )( n )[2 n ]where p X,Z is the isomorphism (2.4.36.1). For future reference, note that we deduce from this theso-called Gysin morphism:(2.4.39.2) i ∗ : M S ( X ) π −→ M S ( X/X − Z ) p t X,Z −−−→ M S ( Z )( n )[2 n ]where π is the following map: M S ( X ) = p ♯ ( X ) ad ( i ∗ ,i ∗ ) −−−−−→ p ♯ i ∗ i ∗ ( X ) = M S ( X/X − Z ) . As a particular case, we get using the notation of Corollary 2.4.37, point (2), an isomorphism: p t XX,X : Σ f ( K ) p XX,X −−−−→
M T h X ( T f ) ⊗ K t Tf −−→ K ( d )[2 d ]In particular, when T satisfies property (Supp), the purity comparison map associated with f can be rewritten as:(2.4.39.3) p t f : f ♯ p f −−−→ f ! ◦ Σ f p t XX,X −−−−−→ f ! ( d )[2 d ] Example . Assume as in the above definition that T is premotivic and satisfies prop-erties (wLoc) and (Nis-sep).We suppose the following two additional conditions are fulfilled: . TRIANGULATED P -FIBRED CATEGORIES IN ALGEBRAIC GEOMETRY 51 (a’) There exists a morphism of triangulated premotivic categories: ϕ ∗ : SH ⇆ T : ϕ ∗ where SH is the stable homotopy category of Morel and Voevodsky – see Example 1.4.3.(b’) For any scheme X , let Pic( X ) be the Picard group of X . We assume there exists anapplication c : Pic( X ) → H , T ( X ) := Hom T ( X ) ( M ( X ) , S (1)[2])which is natural with respect to contravariant functoriality – we do not require c is amorphism of abelian groups.Then one can apply the results of [ D´eg08 ] to T ( X ) for any scheme X . All the referenceswhich follows will be within loc. cit. : according to section 2.3.2, the triangulated category T ( X )satisfies the axioms of Paragraph 2.1. Then the existence of the Thom isomorphism followsfrom Proposition 4.3 and, more explicitly, from Paragraph 4.4. Property (a) and (b) of the abovedefinition are easy – explicitly, this is a consequence of 4.10 – and Property (c) follows from Lemma4.30.To sum up, the assumptions (a’) and (b’) guarantees the existence of a canonical orientation of T in the sense of the above definition. Moreover, the purity isomorphism (2.4.39.1) as well as theGysin morphism (2.4.39.2) associated in the preceding paragraph for this particular orientationcoincide with the one defined in [ D´eg08 ] (see in particular the unicity statement of [
D´eg08 , Prop.4.3]).Note moreover that assuming T satisfies all the properties above except (b’), the data of anorientation of T is equivalent to the data of a map c as in (b’). Indeed, if t is an orientation of T , given any line bundle L/X with zero section s , we put c ( L ) = ρ ( t L ) where ρ is the followingcomposite map: H , T ( T h X ( L )) → H , T ( L ) s ∗ −→ H , T ( X )where the first map is induced by the canonical projection M X ( L ) → M T h X ( L ). Then c dependsonly on the isomorphism classes of L/X – property (a) of the above definition – and it is compatiblewith pullbacks – property (c) of the above definition.2.4.41 . We now assume the following conditions on the triangulated premotivic category T : • T satisfies properties (Nis-sep), (wLoc), (Htp) and (Stab). • T admits an orientation t .Let f : X → S be a smooth proper morphism of dimension d . Note we do not need that T satisfies property (Supp) to rewrite the purity comparison map as follows:(2.4.41.1) p t f : f ♯ → f ∗ ( d )[2 d ](see Paragraph 2.4.39).Note also that using the Gysin morphism (2.4.39.2) associated with the diagonal immersion δ : X → X × S X , we get the following morphism:(2.4.41.2) µ t f : M S ( X ) ⊗ M S ( X )( − d )[ − d ] = M S ( X × S X )( − d )[ − d ] δ ∗ −−→ M S ( X ) f ∗ −−−→ S . Theorem . Consider the assumptions and notations above. Then the following condi-tions are equivalent:(i) f is pure: p f is an isomorphism.(i’) The natural transformation p f .f ∗ is an isomorphism.(ii) The premotive M S ( X ) is strongly dualizable and µ t f is a perfect pairing. Note in particular that for any smooth closed S -pair, we obtain a canonical isomorphism in T ( S ) of theform: ϕ ∗ (Σ ∞ X/X − Z ) ≃ M S ( X/X − Z )where one the left hand side X/X − Z stands for the homotopy cofiber of the open immersion ( X − Z ) → X whilethe left hand side is defined by Equality (2.4.32.1). Proof.
In this proof, we put τ ( K ) = K ( d )[2 d ]. As T satisfies property (Stab), f ∗ com-mutes with Tate twist (def. 1.1.44). This means we the following exchange transformation is anisomorphism:(2.4.42.1) Ex τ : τ f ∗ → f ∗ τ. We first prove that (i) is equivalent to (i’). One implication is obvious so that we have onlyto prove that (i’) implies (i). Guided by a method of Ayoub (see [
Ayo07a , 1.7.14, 1.7.15], we willconstruct a right inverse φ and a left inverse φ to the morphism p t f as the following compositemaps: φ : f ∗ τ ad ( f ∗ ,f ∗ ) −−−−−−→ f ∗ f ∗ f ∗ τ Ex − τ −−−→ f ∗ f ∗ τ f ∗ = f ∗ τ f ∗ f ∗ ( p t f .f ∗ f ∗ ) − −−−−−−−−→ f ♯ f ∗ f ∗ ad ′ ( f ∗ ,f ∗ ) −−−−−−→ f ♯ φ : f ∗ τ β f −→ f ∗ τ f ∗ f ♯ ( p t f .f ∗ f ♯ ) − −−−−−−−−→ f ♯ f ∗ f ♯ ad ′ ( f ♯ ,f ∗ ) −−−−−−→ f ♯ . Let us check that p t f ◦ φ = 1. To prove this relation, we prove that the following diagram iscommutative: f ∗ τ ad ( f ∗ ,f ∗ ) / / f ∗ f ∗ f ∗ τ Ex − τ / / f ∗ τ f ∗ f ∗ ( p t f f ∗ f ∗ ) − / / f ♯ f ∗ f ∗ ad ′ ( f ∗ ,f ∗ ) / / (1) f ♯ p t f / / f ∗ τf ∗ τ f ∗ f ∗ ( p t f f ∗ f ∗ ) − / / (2) f ♯ f ∗ f ∗ p t f f ∗ f ∗ / / f ∗ τ f f ∗ ad ′ ( f ∗ ,f ∗ ) / / f ∗ τf ∗ f ∗ f ∗ τ Ex − τ / / (3) f ∗ τ f ∗ f ∗ ad ′ ( f ∗ ,f ∗ ) / / f ∗ τf ∗ τ ad ( f ∗ ,f ∗ ) / / f ∗ f ∗ f ∗ τ ad ′ ( f ∗ ,f ∗ ) / / f ∗ τ. The commutativity of (1) and (2) is obvious and the commutativity of (3) follows from Formula(2.4.42.1) defining Ex τ . Then the result follows from the usual formula between the unit andcounit of an adjunction. The relation φ ◦ p t f = 1 is proved using the same kind of computations.It remains to prove that (i) and (i’) are equivalent to (ii). We already know from Proposition2.4.31 that (i) implies the premotive M S ( X ) is strongly dualizable. Saying that µ t f is a perfectpairing amounts to prove that the natural transformation obtained by adjunction d t f : ( M S ( X ) ⊗ − ) → Hom ( M S ( X ) , − ( d )[2 d ])is an isomorphism. On the other hand, as we have already seen previously, the smooth projectionformula implies an identification of functors: f ♯ f ∗ ≃ ( M S ( X ) ⊗ − ) ,f ∗ f ∗ ≃ Hom ( M S ( X ) , − ) . (2.4.42.2)Thus, to finish the proof, it will be enough to show that the map f ♯ f ∗ p t f f ∗ −−−→ f ∗ τ f ∗ = f ∗ f ∗ τ. is equal to d t f through the identifications (2.4.42.2).Let us consider the following cartesian square X × S X f ′′ / / f ′ (cid:15) (cid:15) ∆ X f (cid:15) (cid:15) X f / / S . TRIANGULATED P -FIBRED CATEGORIES IN ALGEBRAIC GEOMETRY 53 and put g = f ◦ f ′′ . According to the definition of µ t f , and notably Formula (2.4.39.2) for theGysin map δ ∗ , the natural transformation of functors ( µ t f ⊗ − ) can be described as the followingcompositum: f ♯ f ∗ f ♯ f ∗ Ex (∆ ∗ ♯ ) −−−−−→ f ♯ f ′ ♯ f ′′∗ f ∗ = g ♯ g ∗ ad ( δ ∗ ,δ ∗ ) −−−−−−→ g ♯ δ ∗ δ ∗ g ∗ = f ♯ f ′ ♯ δ ∗ f ∗ p t XX,X −−−−→ f ♯ τ f ∗ = f ♯ f ∗ τ ad ′ ( f ♯ ,f ∗ ) −−−−−−→ τ. Note in particular that the base change map Ex (∆ ∗ ♯ ) corresponds to the first identification inFormula (2.4.41.2). Thus we have to prove the preceding composite map is equal to the followingone, obtained by adjunction from p t f : f ♯ f ∗ f ♯ f ∗ = f ♯ f ∗ f ♯ f ′′∗ δ ∗ f ∗ Ex (∆ ♯ ∗ ) −−−−−→ f ♯ f ∗ f ∗ f ′ ♯ δ ∗ f ∗ p t XX,X −−−−→ f ♯ f ∗ f ∗ τ f ∗ = f ♯ f ∗ f ∗ f ∗ τ ad ′ ( f ∗ ,f ∗ ) −−−−−−→ f ♯ f ∗ τ ad ′ ( f ♯ ,f ∗ ) −−−−−−→ τ This amounts to prove, after some easy cancellation, the commutativity of the following diagram: f ∗ f ♯Ex (∆ ∗ ♯ ) (cid:15) (cid:15) f ∗ f ♯ f ′′∗ δ ∗ Ex (∆ ♯ ∗ ) / / f ∗ f ∗ f ′′ ♯ δ ∗ ad ′ ( f ∗ ,f ∗ ) (cid:15) (cid:15) f ′ ♯ f ′′∗ ad ( δ ∗ ,δ ∗ ) / / f ′ ♯ δ ∗ δ ∗ f ′′∗ f ′ ♯ δ ∗ . According to the definition of the exchange transformation Ex (∆ ♯ ∗ ) (cf Paragraph 1.1.14), we candivide this diagram into the following pieces: f ∗ f ♯Ex (∆ ∗ ♯ ) (cid:15) (cid:15) f ∗ f ♯ f ′′∗ δ ∗ ad ( f ∗ ,f ∗ ) / / Ex (∆ ∗ ♯ ) (cid:15) (cid:15) f ∗ f ∗ f ∗ f ♯ f ′′∗ δ ∗ Ex (∆ ∗ ♯ ) / / f ∗ f ∗ f ′ ♯ f ′′∗ f ′′∗ δ ∗ ad ′ ( f ′′∗ ,f ′′∗ ) / / ad ′ ( f ∗ ,f ∗ ) (cid:15) (cid:15) f ∗ f ∗ f ′′ ♯ δ ∗ ad ′ ( f ∗ ,f ∗ ) (cid:15) (cid:15) f ′ ♯ f ′′∗ f ′ ♯ f ′′∗ f ′′∗ δ ∗ ad ( f ∗ ,f ∗ ) ❤❤❤❤❤❤❤❤❤❤❤❤❤❤❤❤❤❤❤❤❤❤❤❤❤❤❤❤❤❤ f ′ ♯ f ′′∗ f ′′∗ δ ∗ ad ′ ( f ′′∗ ,f ′′∗ ) / / f ′ ♯ δ ∗ f ′ ♯ f ′′∗ ad ( δ ∗ ,δ ∗ ) / / ( ∗ ) f ′ ♯ δ ∗ . Every part of this diagram is obviously commutative except for part ( ∗ ). As f ′′ δ = 1, the axiomsof a 2-functors (for f ∗ and f ∗ say) implies that the unit map α : f ′ ♯ f ′′∗ → f ′ ♯ f ′′∗ ( f ′′ δ ) ∗ ( f ′′ δ ) ∗ is the canonical identification that we get using 1 ∗ = 1 and 1 ∗ = 1. We can consider the followingdiagram: f ′ ♯ f ′′∗ α f ′ ♯ f ′′∗ ( f ′′ δ ) ∗ ( f ′′ δ ) ∗ f ′ ♯ f ′′∗ f ′′∗ δ ∗ ad ′ ( f ′′∗ ,f ′′∗ ) (cid:15) (cid:15) f ′ ♯ f ′′∗ ad ( f ′′∗ ,f ′′∗ ) / / f ′ ♯ f ′′∗ f ′′ ♯ f ′′∗ ad ( δ ∗ ,δ ∗ ) / / ad ′ ( f ′′∗ ,f ′′∗ ) (cid:15) (cid:15) f ′ ♯ f ′′∗ ( f ′′ δ ) ∗ ( f ′′ δ ) ∗ ad ′ ( f ′′∗ ,f ′′∗ ) (cid:15) (cid:15) f ′ ♯ f ′′∗ f ′ ♯ f ′′∗ ad ( δ ∗ ,δ ∗ ) / / f ′ ♯ δ ∗ δ ∗ f ′′∗ f ′ ♯ δ ∗ for which each part is obviously commutative. This concludes. (cid:3) As a corollary, together with the results of [
D´eg08 ], we get the following theorem:
Corollary . Let us assume the following conditions on the triangulated premotiviccategory T :(a) T satisfies properties ( Nis -sep), (wLoc), (Htp) and (Stab).(b) T admits an orientation t .(c) There exists a morphism of triangulated premotivic categories: ϕ ∗ : SH ⇆ T : ϕ ∗ . Then any smooth projective morphism is T -pure. In particular, T is weakly pure. Proof.
According to Example 2.4.40, one can apply the results of [
D´eg08 ] to the triangulatedcategory T ( X ). Then it follows from [ D´eg08 , 5.23] that condition (ii) of the above theorem issatisfied. (cid:3)
Remark . This theorem is to be compared with the result of Ayoub recalled in Theorem2.4.28. On the one hand, if T satisfies the localization property, we get another proof of this resultunder the additional assumption that T is oriented. On the other hand, the above theorem doesnot require the assumption that T satisfies (Loc); this is important as we can only prove (wLoc)for the category DM Λ introduced in Definition 11.1.1.2.4.d. Motivic categories.
This section summarizes the main constructions of this part anddraws a conclusive theorem.
Definition . A motivic triangulated category over S is a premotivic triangulated cat-egory over S which satisfies the homotopy, stability, localization and adjoint property. Remark . Without the adjoint property, this definition corresponds to what Ayoubcalled a monoidal stable homotopy -functor (cf [ Ayo07a , def. 2.3.1]). We think our shorterterminology fits well in the spirit of the current theory of mixed motives.
Remark . Assume T is a premotivic triangulated category such that:(1) T is well generated.(2) T satisfies the homotopy and stability properties.(3) T satisfies the localization property.Then T is a motivic triangulated category in the above sense. Indeed, property (Adj) is provedunder the above assumptions in point (4) of Theorem 2.4.26. Note also that if T is compactly τ -generated, we simply obtain property (Adj) from Lemma 2.2.16. Example . According to the previous remark, the premotivic category SH of example1.4.3 is a motivic category. In fact, property (1) is proved in [
Ayo07a , 4.5.67], property (2) followsby definition and property (3) is proved in [
Ayo07a , 4.5.44].2.4.49 . In the next theorem, we summarize what is now called the
Grothendieck 6 functorsformalism . In fact, this is a consequence of the axioms in the above definition, as a result of thework done in previous sections. More precisely: • We apply Theorem 2.4.26 using the theorem of Ayoub recalled in 2.4.28, and use thegeneralized theorem of Morel and Voevodsky, Theorem 2.4.35, to get the form (2.4.37.1)of the purity isomorphism. • In the case where T is oriented, we use the form (2.4.41.1) of the purity isomorphism.Recall that, when T satisfies assumption (c) of Corollary 2.4.43, then we have given adifferent proof of the Theorem of Ayoub and the theorem below follows from 2.4.26 and2.4.43. In our examples, (1) will always be satisfied, (2) will be obtained by construction and (3) will be the hardpoint. . TRIANGULATED P -FIBRED CATEGORIES IN ALGEBRAIC GEOMETRY 55 Theorem . Let T be a motivic triangulated category.Then, for any separated morphism of finite type f : Y → X in S , there exists a pair of adjointfunctors, the exceptional functors , f ! : T ( Y ) ⇄ T ( X ) : f ! such that:(1) There exists a structure of a covariant (resp. contravariant) -functor on f f ! (resp. f f ! ).(2) There exists a natural transformation α f : f ! → f ∗ which is an isomorphism when f isproper. Moreover, α is a morphism of -functors.(3) For any smooth separated morphism of finite type f : X → S in S with tangent bundle T f , there are canonical natural isomorphisms p f : f ♯ −→ f ! (cid:0) M T h X ( T f ) ⊗ X . (cid:1) p ′ f : f ∗ −→ M T h X ( − T f ) ⊗ X f ! which are dual to each other – the Thom premotive M T h X ( T f ) is ⊗ -invertible withinverse M T h X ( − T f ) .If T admits an orientation t and f has dimension d then there are canonical naturalisomorphisms p t f : f ♯ −→ f ! ( d )[2 d ] p ′ t f : f ∗ −→ f ! ( − d )[ − d ] which are dual to each other.(4) For any cartesian square: Y ′ f ′ / / g ′ (cid:15) (cid:15) ∆ X ′ g (cid:15) (cid:15) Y f / / X, such that f is separated of finite type, there exist natural isomorphisms g ∗ f ! ∼ −→ f ′ ! g ′∗ ,g ′∗ f ′ ! ∼ −→ f ! g ∗ . (5) For any separated morphism of finite type f : Y → X in S , there exist natural isomor-phisms Ex ( f ∗ ! , ⊗ ) : ( f ! K ) ⊗ X L ∼ −−→ f ! ( K ⊗ Y f ∗ L ) , Hom X ( f ! ( L ) , K ) ∼ −−→ f ∗ Hom Y ( L, f ! ( K )) ,f ! Hom X ( L, M ) ∼ −−→ Hom Y ( f ∗ ( L ) , f ! ( M )) . Remark . It is important to precise that in the case where the morphisms in S areassumed to be quasi-projective, this theorem is proved by Ayoub in [ Ayo07a ] if we except thecase where T is oriented in point (3). With regards to this theorem, our contribution is to extend the result of Ayoub to the nonquasi-projective case and to consider the oriented case – which is crucial in the theory of motives.Recall also we have given another proof of this result in the case where the motivic category T satisfies in addition the assumptions of Corollary 2.4.43 – which will always be the case for thedifferent categories of motives introduced here. This theorem was first announced by Voevodsky but only notes covering the basic setting were to be foundby the time Ayoub wrote the proof. Remark . The purity isomorphism is compatible with composition. Given smoothseparated morphisms of finite type Y g −→ X f −→ S we obtain ( cf. [ EGA4 , 17.2.3]) an exact sequence of vector bundles over Y ( σ ) 0 → g − T f → T fg → T g → . which according to Remark 2.4.15 induces an isomorphism: ǫ σ : M T h Y ( T fg ) MT h Y ( σ ) −−−−−−→ M T h Y ( T g ) ⊗ Y M T h Y ( g − T f ) ∼ −→ g ∗ M T h X ( T f ) ⊗ Y M T h Y ( T g ) . One can check the following diagram is commutative:( f g ) ♯ ( K ) p fg (cid:15) (cid:15) f ♯ g ♯ ( K ) p f ◦ p g (cid:15) (cid:15) f ! (cid:16) M T h X ( T f ) ⊗ X g ! (cid:0) M T h Y ( T g ) ⊗ Y K (cid:1)(cid:17) Ex ( g ∗ ! , ⊗ ) − (cid:15) (cid:15) f ! g ! (cid:0) g ∗ M T h Y ( T f ) ⊗ Y M T h Y ( T g ) ⊗ Y K (cid:1) ǫ − σ (cid:15) (cid:15) ( f g ) ! ( M T h ( T fg ) ⊗ K ) f ! g ! ( M T h ( T fg ) ⊗ K ) . This is not an easy check. In fact, this is one of the key technical point in the proof of the mainTheorem of Ayoub ([
Ayo07a , 1.4.2]). We refer the reader to [
Ayo07a , 1.5] for details.Note also that given the commutativity of the above diagram, if T admits an orientation t ,it readily follows from axiom (c) of Definition 2.4.38 that the following diagram is commutative:( f g ) ♯ ( K ) p t fg (cid:15) (cid:15) f ♯ g ♯ ( K ) p t f ◦ p t g (cid:15) (cid:15) ( f g ) ! ( K )( n + m )[2 n + 2 m ] f ! g ! ( K )( n + m )[2 n + 2 m ]where n (resp. m ) is the relative dimension of f (resp. g ).Morphisms of triangulated motivic categories are compatible with Grothendieck 6 operationsin the following sense: Proposition . Let T and T ′ be motivic triangulated categories and ϕ ∗ : T ⇄ T ′ : ϕ ∗ be an adjunction of premotivic categories.Then ϕ ∗ (resp. ϕ ∗ ) commutes with the operations f ∗ (resp. f ∗ ) for any morphism of schemes f , p ! (resp. p ! ) for any separated morphism p of finite type.Moreover, ϕ ∗ is monoidal and for any premotive M ∈ T ( S ) , N ∈ T ′ ( S ) , the canonical mapHom ( M, ϕ ∗ ( N )) → ϕ ∗ Hom ( ϕ ∗ ( M ) , N ) is an isomorphism. Proof.
The only thing to prove is that ϕ ∗ commutes with p ! as the other statements followseither from the definitions or by adjunction. This follows from Proposition 2.3.11, the purityproperty in T and T ′ (property (3) in the above theorem) and the fact ϕ ∗ commutes with p ♯ when p is smooth by assumption. (cid:3) The main point is to check that the isomorphism of Theorem 2.4.35 is compatible with composition (ofclosed immersions). On that particular point, see [
D´eg08 , Th. 4.32, Cor. 4.33]. . DESCENT IN P -FIBRED MODEL CATEGORIES 57 Remark . With additional assumptions on T and T ′ and over a field, we will see that ϕ ∗ commutes with all of the six operations (see Theorem 4.4.25).
3. Descent in P -fibred model categories . In this section, S is an abstract category and P an admissible class of morphisms in S .In section 3.3 however, we will consider as in 2.0 a noetherian base scheme S and we willassume that S is an adequate category of S -schemes satisfying the following condition on S :(a) Any scheme in S is finite dimensional.Moreover, in sections 3.3.c and 3.3.d, we will even assume:(a ′ ) Any scheme in S is quasi-excellent and finite dimensional.We fix an admissible class P of morphisms in S which contains the class of ´etale morphismsin S and a stable combinatorial P -fibred model category M over S .In section 3.3.d, we will assume furthermore that:(b) The stable model P -fibred category M is Q -linear (see 3.2.14). P -fibred categories to diagrams. The general case. . Assume given a P -fibered category M over S . Then M can be extended to S -diagrams (i.e. functors from a small category to S ) as follows. Let I be a small category, and X a functor from I to S . For an object i of I , we will denote by X i the fiber of X at i (i.e.the evaluation of X at i ), and, for a map u : i → j in I , we will still denote by u : X i → X j themorphism induced by u . We define the category M ( X , I ) as follows.An object of M ( X , I ) is a couple ( M, a ), where M is the data of an object M i in M ( X i ) forany object i of I , and a is the data of a morphism a u : u ∗ ( M j ) → M i for any morphism u : i → j in I , such that, for any object i of I , the map a i is the identity of M i (we will always assumethat 1 ∗ i is the identity functor), and, for any composable morphisms u : i → j and v : j → k in I ,the following diagram commutes. u ∗ v ∗ ( M k ) u ∗ ( a v ) (cid:15) (cid:15) ≃ / / ( vu ) ∗ ( M k ) a vu (cid:15) (cid:15) u ∗ ( M j ) a u / / M i A morphism p : ( M, a ) → ( N, b ) is a collection of morphisms p i : M i → N i in M ( X i ), for each object i in I , such that, for any morphism u : i → j in I , the following diagramcommutes. u ∗ ( M j ) u ∗ ( p j ) / / a u (cid:15) (cid:15) u ∗ ( N j ) b u (cid:15) (cid:15) M i p i / / N i In the case where M is a monoidal P -fibred category, the category M ( X , I ) is naturally endowedwith a symmetric monoidal structure. Given two objects ( M, a ) and (
N, b ) of M ( X , I ), theirtensor product ( M, a ) ⊗ ( N, b ) = ( M ⊗ N, a ⊗ b )is defined as follows. For any object i of I ,( M ⊗ N ) i = M i ⊗ N i , and for any map u : i → j in I , the map ( a ⊗ b ) u is the composition of the isomorphism u ∗ ( M j ⊗ N j ) ≃ u ∗ ( M j ) ⊗ u ∗ ( N j ) with the morphism a u ⊗ b u : u ∗ ( M j ) ⊗ u ∗ ( N j ) → M i ⊗ N i . Note finally that if M is a complete monoidal P -fibred category, then M ( X , I ) admits an internalHom.3.1.2 . Evaluation functors . Assume now that for any S , M ( S ) admits small sums.For each object i of I , we have a functor(3.1.2.1) i ∗ : M ( X , I ) → M ( X i )( M, a ) M i called the evaluation functor associated with i . This functor i ∗ has a left adjoint(3.1.2.2) i ♯ : M ( X i ) → M ( X , I )defined as follows. If M is an object of M ( X i ), then i ♯ ( M ) is the data ( M ′ , a ′ ) such that for anyobject j of I ,(3.1.2.3) ( i ♯ ( M )) j = M ′ j = a u ∈ Hom I ( j,i ) u ∗ ( M ) , and, for any morphism v : k → j in I , the map a ′ v is the canonical map induced by the collectionof maps(3.1.2.4) v ∗ u ∗ ( M ) ≃ ( uv ) ∗ ( M ) → a w ∈ Hom I ( k,i ) w ∗ ( M )for u ∈ Hom I ( j, i ).If we assume that M is a complete P -fibred category and that M ( S ) admits small productsfor any S , then i ∗ has a right adjoint(3.1.2.5) i ∗ : M ( X i ) → M ( X , I )given, for any object M of M ( X i ) by the formula(3.1.2.6) ( i ∗ ( M )) j = Y u ∈ Hom I ( i,j ) u ∗ ( M ) , with transition map given by the dual formula of 3.1.2.4.3.1.3 . Functoriality . Assume that M if a P -fibred category suth that for any object S of S , M ( S ) has small colimits.Remember that, if X and Y are S -diagrams, indexed respectively by small categories I and J , a morphism of S -diagrams ϕ : ( X , I ) → ( Y , J ) is a couple ϕ = ( α, f ), where f : I → J is afunctor, and α : X → f ∗ ( Y ) is a natural transformation (where f ∗ ( Y ) = Y ◦ f ). In particular,for any object i of I , we have a morphism α i : X i → Y f ( i ) in S . This turns S -diagrams into a strict 2-category: the identity of ( X , I ) is the couple(1 X , I ), and, if ϕ = ( α, f ) : ( X , I ) → ( Y , J ) and ψ = ( β, g ) : ( Y , J ) → ( Z , K ) are twocomposable morphisms, the morphism ψ ◦ ϕ : ( X , I ) → ( Z , K ) is the couple ( gf, γ ), where foreach object i of I , the map γ i : X i → Z g ( f ( i )) is the composition X i α i −−−→ Y f ( i ) β f ( i ) −−−→ Z g ( f ( i )) . There is also a notion of natural transformation between morphisms of S -diagrams: if ϕ = ( α, f )and ϕ ′ = ( α ′ , f ′ ) are two morphisms from ( X , I ) to ( Y , J ), a natural transformation t from ϕ to ϕ ′ is a natural transformation t : f → f ′ such that the following diagram of functors commutes. X α | | ①①①①①①①①① α ′ ●●●●●●●●● Y ◦ f t / / Y ◦ f ′ . DESCENT IN P -FIBRED MODEL CATEGORIES 59 This makes the category of S -diagrams a (strict) 2-category.To a morphism of diagrams ϕ = ( α, f ) : ( X , I ) → ( Y , J ), we associate a functor ϕ ∗ : M ( Y , J ) → M ( X , I )as follows. For an object ( M, a ) of M ( Y ), ϕ ∗ ( M, a ) = ( ϕ ∗ ( M ) , ϕ ∗ ( a )) is the object of M ( X )defined by ϕ ∗ ( M ) i = α ∗ i ( M f ( i ) ) for i in I , and by the formula ϕ ∗ ( a ) u = α ∗ i ( a f ( u ) ) : α ∗ i f ( u ) ∗ ( M f ( j ) ) = u ∗ α ∗ j ( M f ( j ) ) → α ∗ i ( M f ( i ) )for u : i → j in I .We will say that a morphism ϕ : ( X , I ) → ( Y , J ) is a P -morphism if, for any object i in I ,the morphism α i : X i → Y f ( i ) is a P -morphism. For such a morphism ϕ , the functor ϕ ∗ has aleft adjoint which we denote by ϕ ♯ : M ( X , I ) → M ( Y , J ) . For instance, given a S -diagram X indexed by a small category I , each object i of I defines a P -morphism of diagrams i : X i → ( X , I ) (where X i is indexed by the terminal category), sothat the corresponding the functor i ♯ corresponds precisely to (3.1.2.2).Assume that M is a complete P -fibred category such that M ( S ) has small limits for anyobject S of S . Then the functor ϕ ∗ has a right adjoint which we denote by ϕ ∗ : M ( X , I ) → M ( Y , J ) . In the case where ϕ is the morphism i : X i → ( X , I ) defined by an object i of I , i ∗ correspondsprecisely to (3.1.2.5). Remark . This construction can be applied in particular to any Grothendieck abelian(monoidal) P -fibred category ( cf. definition 1.3.8). The triangulated case cannot be treated ingeneral without assuming a thorough structure – this is the purpose of the next section.3.1.b. The model category case. . Let M be a P -fibred model category over S ( cf. S -diagram X indexed by a small category I , we will say that a morphism of M ( X , I ) is a termwise weakequivalence (resp. a termwise fibration , resp. a termwise cofibration ) if, for any object i of I , itsimage by the functor i ∗ is a weak equivalence (resp. a fibration, resp. a cofibration) in M ( X i ). Proposition . If M is a cofibrantly generated P -fibred model category over S , then, forany S -diagram X indexed by a small category I , the category M ( X , I ) is a cofibrantly generatedmodel category whose weak equivalences (resp. fibrations) are the termwise weak equivalences (resp.the termwise fibrations). This model category structure on M ( X , I ) will be called the projectivemodel structure. Moreover, any cofibration of M ( X , I ) is a termwise cofibration, and the family of functors i ∗ : Ho( M )( X , I ) → Ho( M )( X i ) , i ∈ Ob( I ) , is conservative.If M is left proper (resp. right proper, resp. combinatorial, resp. stable), then so is theprojective model category structure on M ( X ) . Proof.
Let X δ be the S -diagram indexed by the set of objects of I (seen as a discretecategory), whose fiber at i is X i . Let ϕ : ( X δ , Ob I ) → ( X , I ) be the inclusion (i.e. the mapwhich is the identity on objects and which is the identity on each fiber). As ϕ is clearly a P -morphism, we have an adjunction ϕ ♯ : M ( X δ , Ob I ) ≃ Y i M ( X i ) ⇄ M ( X , I ) : ϕ ∗ . The functor ϕ ♯ can be made explicit: it sends a family of objects ( M i ) i (with M i in M ( X i ))to the sum of the i ♯ ( M i )’s indexed by the set of objects of I . Note also that this proposition istrivially verified whenever X δ = X . Using the explicit formula for i ♯ given in 3.1.2, it is thenstraightforward to check that the adjunction ( ϕ ♯ , ϕ ∗ ) satisfies the assumptions of [ Cra95 , Theorem M ( X , I ). Furthermore, thegenerating cofibrations (resp. trivial cofibrations of M ( X , I )) can be described as follows. Foreach object i of I , let A i (resp. B i ) be a generating set of cofibrations (resp. of trivial cofibrationsin M ( X i ). The class of termwise trivial fibrations (resp. of termwise fibrations) of M ( X , I ) is theclass of maps which have the right lifting property with respect to the set A = ∪ i ∈ I i ♯ ( A i ) (resp.to the set B = ∪ i ∈ I i ♯ ( B i )). Hence, the set A (resp. B ) generates the class of cofibrations (resp.of trivial cofibrations). In particular, as any element of A is a termwise cofibration (which followsimmediately from the explicit formula for i ♯ given in 3.1.2), and as termwise cofibrations are stableby pushouts, transfinite compositions and retracts, any cofibration is a termwise cofibration (bythe small object argument).As any fibration (resp. cofibration) of M ( X , I ) is a termwise fibration (resp. a termwisecofibration), it is clear that, whenever the model categories M ( X i ) are right (resp. left) proper,the model category M ( X , I ) has the same property.The functor ϕ ∗ preserves fibrations and cofibrations, while it also preserves and detects weakequivalences (by definition). This implies that the induced functor ϕ ∗ : Ho( M )( X , I ) → Ho( M )( X δ , Ob I ) ≃ Y i Ho( M )( X i )is conservative (using the facts that the set of maps from a cofibrant object to a fibrant object inthe homotopy category of a model category is the set of homotopy classes of maps, and that amorphism of a model category is a weak equivalence if and only if it induces an isomorphism inthe homotopy category). As ϕ ∗ commutes to limits and colimits, this implies that it commutesto homotopy limits and to homotopy colimits (up to weak equivalences). Using the conservativityproperty, this implies that a commutative square of M ( X , I ) is a homotopy pushout (resp. ahomotopy pullback) if and only if it is so in M ( X δ , Ob I ). Remember that stable model categoriesare characterized as those in which a commutative square is a homotopy pullback square if andonly if it is a homotopy pushout square. As a consequence, if all the model categories M ( X i ) arestable, as M ( X δ , Ob I ) is then obviously stable as well, the model category M ( X , I ) has thesame property.It remains to prove that, if M ( X, I ) is a combinatorial model category for any object X of S , then M ( X , I ) is combinatorial as well. For each object i in I , let G i be a set of accessiblegenerators of M ( X i ). Note that, for any object i of I , the functor i ♯ has a left adjoint i ∗ whichcommutes to colimits (having itself a right adjoint i ∗ ). It is then easy to check that the set ofobjects of shape i ♯ ( M ), for M in G i and i in I , is a small set of accessible generators of M ( X , I ).This implies that M ( X , I ) is accessible and ends the proof. (cid:3) Proposition . Let M be a combinatorial P -fibred model category over S . Then, forany S -diagram X indexed by a small category I , the category M ( X , I ) is a combinatorial modelcategory whose weak equivalences (resp. cofibrations) are the termwise weak equivalences (resp.the termwise cofibrations). This model category structure on M ( X , I ) will be called the injectivemodel structure . Moreover, any fibration of the injective model structure on M ( X , I ) is atermwise fibration.If M is left proper (resp. right proper, resp. stable), then so is the injective model categorystructure on M ( X , I ) . Proof.
See [
Bar10 , Theorem 2.28] for the existence of such a model structure (if, for anyobject X in S , all the cofibrations of M ( X ) are monomorphisms, this can also be done followingmutatis mutandis the proof of [ Ayo07a , Proposition 4.5.9]). Any trivial cofibration of the pro-jective model structure being a termwise trivial cofibration, any fibration of the injective modelstructure is a fibration of the projective model structure, hence a termwise fibration.The assertions about properness follows from their analogs for the projective model structureand from [
Cis06 , Corollary 1.5.21] (or can be proved directly; see [
Bar10 , Proposition 2.31]). Quite unfortunately, this corresponds to the ‘semi-projective’ model structure introduced in [
Ayo07a , Def.4.5.8]. . DESCENT IN P -FIBRED MODEL CATEGORIES 61 Similarly, the assertion on stability follows from their analogs for the projective model structure. (cid:3) . From now on, we assume that a combinatorial P -fibred model category M over S isgiven. Then, for any S -diagram ( X , I ), we have two model category structures on M ( X , I ), andthe identity defines a left Quillen equivalence from the projective model structure to the injectivemodel structure. This fact will be used for the understanding of the functorialities coming frommorphisms of diagrams of S -schemes.3.1.9 . The category of S -diagrams admits small sums. If { ( Y j , I j ) } j ∈ J is a small family of S -diagrams, then their sum is the S -diagram ( X , I ), where I = a j ∈ J I j , and X is the functor from I to S defined by X i = Y j whenever i ∈ I j . Proposition . For any small family of S -diagrams { ( Y j , I j ) } j ∈ J , the canonical functor Ho( M ) (cid:16) a j ∈ J Y j (cid:17) → Y j ∈ J Ho( M )( Y j ) is an equivalence of categories. Proof.
The functor M (cid:16) a j ∈ J Y j (cid:17) → Y j ∈ J M ( Y j )is an equivalence of categories. It thus remains an equivalence after localization. To conclude, itis sufficient to see that the homotopy category of a product of model categories is the product oftheir homotopy categories, which follows rather easily from the explicit description of the homotopycategory of a model category; see e.g. [ Hov99 , Theorem 1.2.10]. (cid:3)
Proposition . Let ϕ = ( α, f ) : ( X , I ) → ( Y , J ) be a morphism of S -diagrams.(i) The adjunction ϕ ∗ : M ( Y , J ) ⇄ M ( X , I ) : ϕ ∗ is a Quillen adjunction with respect tothe injective model structures. In particular, it induces a derived adjunction L ϕ ∗ : Ho( M )( Y , J ) ⇄ Ho( M )( X , I ) : R ϕ ∗ . (ii) If ϕ is a P -morphism, then the adjunction ϕ ♯ : M ( X , I ) ⇄ M ( Y , J ) : ϕ ∗ is a Quillenadjunction with respect to the projective model structures, and the functor ϕ ∗ preservesweak equivalences. In particular, we get a derived adjunction L ϕ ♯ : Ho( M )( X , I ) ⇄ Ho( M )( Y , J ) : L ϕ ∗ = ϕ ∗ = R ϕ ∗ . Proof.
The functor ϕ ∗ obviously preserves termwise cofibrations and termwise trivial cofi-brations (we reduce to the case of a morphism of S using the explicit description of ϕ ∗ givenin 3.1.3), which proves the first assertion. Similarly, the second assertion follows from the factthat, under the assumption that ϕ is a P -morphism, the functor ϕ ∗ preserves termwise weakequivalences (see Remark 1.3.23), as well as termwise fibrations. (cid:3) . The computation of the (derived) functors R ϕ ∗ (and L ϕ ♯ whenever it makes sense)given by Proposition 3.1.11 has to do with homotopy limits (and homotopy colimits). It is easierto first understand this in the non derived version as follows.Consider first the trivial case of a constant S -diagram: let X be an object of S , and I asmall category. Then, seeing X as the constant functor I → S with value X , we have a projectionmap p I : ( X, I ) → X . From the very definition, the category M ( X, I ) is simply the category ofpresheaves on I with values in M ( X ), so that the inverse image functor(3.1.12.1) p ∗ I : M ( X ) → M ( X, I ) = M ( X ) I op is the ‘constant diagram functor’, while its right adjoint(3.1.12.2) lim ←− I op = p I, ∗ : M ( X, I ) → M ( X )is the limit functor, and its left adjoint,(3.1.12.3) lim −→ I op = p I,♯ : M ( X, I ) → M ( X )is the colimit functor.Let S be an object of S . A S -diagram over S is the data of a S -diagram ( X , I ), togetherwith a morphism of S -diagrams p : ( X , I ) → S (i.e. its a S /S -diagram). Such a map p factorsas(3.1.12.4) ( X , I ) π → ( S, I ) p I → S , where π = ( p, I ). Then one checks easily that, for any object M of M ( X , I ), and for any object i of I , one has(3.1.12.5) π ∗ ( M ) i ≃ p i, ∗ ( M i ) , where p i : X i → S is the structural map, from which we deduce the formula(3.1.12.6) p ∗ ( M ) ≃ lim ←− i ∈ I op π ∗ ( M ) i ≃ lim ←− i ∈ I op p i, ∗ ( M i ) , Remark that, if I is a small category with a terminal object ω , then any S -diagram X indexed by I is a S -diagram over X ω , and we deduce from the computations above that, if p : ( X , I ) → X ω denotes the canonical map, then, for any object M of M ( X , I ),(3.1.12.7) p ∗ ( M ) ≃ M ω . Consider now a morphism of S -diagrams ϕ = ( α, f ) : ( X , I ) → ( Y , J ). For each object j ,we can form the following pullback square of categories. I/j u j / / f/j (cid:15) (cid:15) I f (cid:15) (cid:15) J/j v j / / J (3.1.12.8)in which J/j is the category of objects of J over j (which has a terminal object, namely ( j, j ),and v j is the canonical projection; the category I/j is thus the category of pairs ( i, a ), where i isan object of I , and a : f ( i ) → j a morphism in J . From this, we can form the following pullbackof S -diagrams ( X /j, I/j ) µ j / / ϕ/j (cid:15) (cid:15) ( X , I ) ϕ (cid:15) (cid:15) ( Y /j, J/j ) ν j / / ( Y , J )(3.1.12.9)in which X /j = X ◦ u j , Y /j = Y ◦ v j , and the maps µ j and ν j are the one induced by u j and v j respectively. For an object M of M ( X , I ) (resp. an object N of M ( Y , J )), we define M/j (resp.
N/j ) as the object of M ( X /j, I/j ) (resp. of M ( Y /j, J/j )) obtained as M/j = µ ∗ j ( M )(resp. N/j = ν ∗ j ( N )). With these conventions, for any object M of M ( X , I ) and any object j ofthe indexing category J , one gets the formula(3.1.12.10) ϕ ∗ ( M ) j ≃ ( ϕ/j ) ∗ ( M/j ) ( j, j ) ≃ lim ←− ( i,a ) ∈ I/j op α i, ∗ ( M i ) . This implies that the natural map(3.1.12.11) ϕ ∗ ( M ) /j = ν ∗ j ϕ ∗ ( M ) → ( ϕ/j ) ∗ µ ∗ j ( M ) = ( ϕ/j ) ∗ ( M/j ) . DESCENT IN P -FIBRED MODEL CATEGORIES 63 is an isomorphism: to prove this, it is sufficient to obtain an isomorphism from (3.1.12.11) afterevaluating by any object ( j ′ , a : j ′ → j ) of J/j , which follows readily from (3.1.12.10) and fromthe obvious fact that (
I/j ) / ( j ′ , a ) is canonically isomorphic to I/j ′ .In order to deduce from the computations above their derived versions, we need two lemmata. Lemma . Let X be a S -diagram indexed by a small category I , and i an object of I .Then the evaluation functor i ∗ : M ( X , I ) → M ( X i ) is a right Quillen functor with respect to the injective model structure, and it preserves weakequivalences. Proof.
Proving that the functor i ∗ is a right Quillen functor is equivalent to proving that itsleft adjoint (3.1.2.2) is a left Quillen functor with respect to the injective model structure, whichfollows immediately from its computation (3.1.2.3), as, in any model category, cofibrations andtrivial cofibrations are stable by small sums. The last assertion is obvious from the very definitionof the weak equivalences in M ( X , I ). (cid:3) Lemma . For any pullback square of S -diagrams of shape (3.1.12.9) , the functors µ ∗ j : M ( X , I ) → M ( X /j, I/j ) , M M/jν ∗ j : M ( Y , I ) → M ( Y /j, J/j ) , N N/j are right Quillen functors with respect to the injective model structure, and they preserve weakequivalences.
Proof.
It is sufficient to prove this for the functor µ ∗ j (as ν ∗ j is simply the special case where I = J and f is the identity). The fact that µ ∗ j preserves weak equivalences is obvious, so that itremains to prove that it is a right Quillen functor. We thus have to prove that left adjoint of µ ∗ j , µ j,♯ : M ( X /j, I/j ) → M ( X , I ) , is a left Quillen functor. In other words, we have to prove that, for any object i of I , the functor i ∗ µ j,♯ : M ( X , I ) → M ( X )is a left Quillen functor. For any object M of M ( X , I ), we have a natural isomorphism i ∗ µ j,♯ ( M ) ≃ a a ∈ Hom J ( f ( i ) ,j ) ( i, a ) ♯ ( M i ) . But we know that the functors ( i, a ) ♯ are left Quillen functors, so that the stability of cofibrationsand trivial cofibrations by small sums and this description of the functor i ∗ µ j,♯ achieves theproof. (cid:3) Proposition . Let S be an object of S , and p : ( X , I ) → S a S -diagram over S ,and consider the canonical factorization (3.1.12.4) . For any object M of Ho( M )( X , I ) , there arecanonical isomorphisms and Ho( M )( S ) : R π ∗ ( M ) i ≃ R p i, ∗ ( M i ) and R p ∗ ( M ) ≃ R lim ←− i ∈ I op R p i, ∗ ( M i ) . In particular, if furthermore the category I has a terminal object ω , then R p ∗ ( M ) ≃ R p ω, ∗ ( M ω ) . Proof.
This follows immediately from (3.1.12.5) and (3.1.12.6) and from the fact that deriv-ing (right) Quillen functors is compatible with composition. (cid:3)
Proposition . We consider the pullback square of S -diagrams (3.1.12.9) (as well asthe notations thereof ). For any object M of Ho( M )( X , I ) , and any object j of J , we have naturalisomorphisms R ϕ ∗ ( M ) j ≃ R lim ←− ( i,a ) ∈ I/j op R α i, ∗ ( M i ) and R ϕ ∗ ( M ) /j ≃ R ( ϕ/j ) ∗ ( M/j ) in Ho( M )( Y j ) and in Ho( M )( Y /j, J/j ) respectively. Proof.
Using again the fact that deriving right Quillen functors is compatible with compo-sition, by virtue of Lemma 3.1.13 and Lemma 3.1.14, this is a direct translation of (3.1.12.10) and(3.1.12.11). (cid:3)
Proposition . Let u : T → S be a P -morphism of S , and p : ( X , I ) → S a S -diagram over S . Consider the pullback square of S -diagrams ( Y , I ) ϕ / / q (cid:15) (cid:15) ( X , I ) p (cid:15) (cid:15) T u / / S (i.e. Y i = T × S X i for any object i of I ). Then, for any object M of Ho( M )( X , I ) , the canonicalmap L u ∗ R p ∗ ( M ) → R q ∗ L v ∗ ( M ) is an isomorphism in Ho( M )( T ) . Proof.
By Remark 1.3.23, the functor ν ∗ is both a left and a right Quillen functor whichpreserves weak equivalences, so that the functor L ν ∗ = ν ∗ = R ν ∗ preserves homotopy limits.Hence, by Proposition 3.1.15, one reduces to the case where I is the terminal category, i.e. to thetransposition of the isomorphism given by the P -base change formula ( P -BC) for the homotopy P -fibred category Ho( M ) (see 1.1.19). (cid:3) . A morphism of S -diagrams ν = ( α, f ) : ( Y ′ , J ′ ) → ( Y , J ), is cartesian if, for anyarrow i → j in J ′ , the induced commutative square Y ′ i / / α i (cid:15) (cid:15) Y ′ jα j (cid:15) (cid:15) Y f ( i ) / / Y f ( j ) is cartesian.A morphism of S -diagrams ν = ( α, f ) : ( Y ′ , J ′ ) → ( Y , J ) is reduced if J = J ′ and f = 1 J . Proposition . Let ν : ( Y ′ , J ) → ( Y , J ) be a reduced cartesian P -morphism of S -diagrams, and ϕ = ( α, f ) : ( X , I ) → ( Y , J ) a morphism of S -diagrams. Consider the pullbacksquare of S -diagrams ( X ′ , I ) µ / / ψ (cid:15) (cid:15) ( X , I ) ϕ (cid:15) (cid:15) ( Y ′ , J ) ν / / ( Y , J ) (i.e. X ′ i = Y ′ f ( i ) × Y f ( i ) X i for any object i of I ). Then, for any object M of Ho( M )( X , I ) , thecanonical map L ν ∗ R ϕ ∗ ( M ) → R ψ ∗ L µ ∗ ( M ) is an isomorphism in Ho( M )( Y ′ , J ) . Proof.
By virtue of Proposition 3.1.6, it is sufficient to prove that the map j ∗ L ν ∗ R ϕ ∗ ( M ) → j ∗ R ψ ∗ L µ ∗ ( M )is an isomorphism for any object j of J . Let p : ( X /j, I/j ) → Y j and q : ( X ′ /j, J, j ) → Y ′ j bethe canonical maps. As ν is cartesian, we have a pullback square of S -diagrams( X ′ /j, I/j ) µ/j / / q (cid:15) (cid:15) ( X /j, I/j ) p (cid:15) (cid:15) Y ′ j ν j / / Y j . DESCENT IN P -FIBRED MODEL CATEGORIES 65 But ν j being a P -morphism, by virtue of Proposition 3.1.17, we thus have an isomorphism L ν ∗ j R p ∗ ( M/j ) ≃ R q ∗ L ( µ/j ) ∗ ( M/j ) = R q ∗ ( L µ ∗ ( M ) /j ) . Applying Proposition 3.1.16 and the last assertion of Proposition 3.1.15 twice, we also have canon-ical isomorphisms j ∗ R ϕ ∗ ( M ) ≃ R p ∗ ( M/j ) and j ∗ R ψ ∗ L µ ∗ ( M ) ≃ R q ∗ ( L µ ∗ ( M ) /j ) . The obvious identity j ∗ L ν ∗ = L ν ∗ j j ∗ achieves the proof. (cid:3) Corollary . Under the assumptions of Proposition 3.1.19, for any object N of thecategory Ho( M )( Y ′ , j ) , the canonical map L µ ♯ L ψ ∗ ( N ) → L ϕ ∗ L ν ♯ ( N ) is an isomorphism in Ho( M )( X , I ) . Remark . The class of cartesian P -morphisms form an admissible class of morphismsin the category of S -diagrams, which we denote by P cart . Proposition 3.1.11 and the precedingcorollary thus asserts that Ho( M ) is a P cart -fibred category over the category of S -diagrams.3.1.22 . We shall deal sometimes with diagrams of S -diagrams. Let I be a small category,and F a functor from I to the category of S -diagrams. For each object i of I , we have a S -diagram ( F ( i ) , J i ), and, for each map u : i → i ′ , we have a functor f u : J i → J i ′ as well as anatural transformation α u : F ( i ) → F ( i ′ ) ◦ f u , subject to coherence identities. In particular,the correspondance i J i defines a functor from I to the category of small categories. Let I F be the cofibred category over I associated to it; see [ SGA1 , Exp. VI]. Explicitely, I F isdescribed as follows. The objects are the couples ( i, x ), where i is an object of I , and x is anobject of J i . A morphism ( i, x ) → ( i ′ , x ′ ) is a couple ( u, v ), where u : i → i ′ is a morphism of I , and v : f u ( x ) → x ′ is a morphism of J i ′ . The identity of ( i, x ) is the couple (1 i , x ), and,for two morphisms ( u, v ) : ( i, x ) → ( i ′ , x ′ ) and ( u ′ , v ′ ) : ( i ′ , x ′ ) → ( i ′′ , x ′′ ), their composition( u ′′ , v ′′ ) : ( i, x ) → ( i ′′ , x ′′ ) is defined by u ′′ = u ′ ◦ u , while v ′′ is the composition of the map f u ′′ ( x ) = f u ′ ( f u ( x )) f u ′ ( v ) −−−−→ f u ′ ( x ′ ) v ′ −−→ x ′′ . The functor p : I F → I is simply the projection ( i, x ) i . For each object i of I , we get acanonical pullback square of categories J iq (cid:15) (cid:15) ℓ i / / I F p (cid:15) (cid:15) e i / / I (3.1.22.1)in which i is the functor from the terminal category e which corresponds to the object i , and ℓ i isthe functor defined by ℓ i ( x ) = ( i, x ).The functor F defines a S -diagram ( R F , I F ): for an object ( i, x ) of I F , ( R F ) ( i,x ) = F ( i ) x ,and for a morphism ( u, v ) : ( i, x ) → ( i ′ , x ′ ), the map( u, v ) : ( R F ) ( i,x ) = F ( i ) x → ( R F ) ( i ′ ,x ′ ) = F ( i ′ ) x ′ is simply the morphism induced by α u and v . For each object i of I , there is a natural morphismof S -diagrams(3.1.22.2) λ i : ( F ( i ) , J i ) → ( R F , I F ) , given by λ i = (1 F ( i ) , ℓ i ) Proposition . Let X be an object of S , and f : F → X a morphism of functors (where X is considered as the constant functor from I to S -diagrams with value the functor from e to S defined by X ). Then, for each object i of I , we have a canonical pullback square of S -diagrams ( F ( i ) , J i ) λ i / / ϕ i (cid:15) (cid:15) ( R F , I F ) ϕ (cid:15) (cid:15) X i / / ( X, I ) in which ϕ and ϕ i are the obvious morphisms induced by f (where, this time, ( X, I ) is seen as theconstant functor from I to S with value X ).Moreover, for any object M of Ho( M )( R F , I F ) , the natural map i ∗ R ϕ ∗ ( M ) = R ϕ ∗ ( M ) i → R ϕ i, ∗ λ ∗ i ( M ) is an isomorphism. In particular, if we also write by abuse of notation f for the induced map of S -diagrams from ( R F , I F ) to X , we have a natural isomorphism R f ∗ ( M ) ≃ R lim ←− i ∈ I op R ϕ i, ∗ λ ∗ i ( M ) . Proof.
This pullback square is the one induced by (3.1.22.1). We shall prove first that themap i ∗ R ϕ ∗ ( M ) = R ϕ ∗ ( M ) i → R ϕ i, ∗ λ ∗ i ( M )is an isomorphism in the particular case where I has a terminal object ω and i = ω . By virtue ofPropositions 3.1.15 and 3.1.16, we have isomorphisms(3.1.23.1) ω ∗ R ϕ ∗ ( M ) ≃ R lim ←− i ∈ I op R ϕ ∗ ( M ) i ≃ R lim ←− ( i,x ) ∈ I op F R ϕ i,x, ∗ ( M ( i,x ) ) , where ϕ i,x : F ( i ) x → X denotes the map induced by f . We are thus reduced to prove that thecanonical map(3.1.23.2) R lim ←− ( i,x ) ∈ I op F R ϕ i,x, ∗ ( M ( i,x ) ) → R lim ←− x ∈ J op ω R ϕ ω,x, ∗ ( M ( ω,x ) ) ≃ R ϕ ω, ∗ λ ∗ ω ( M )is an isomorphim. As I F is cofibred over I , and as ω is a terminal object of I , the inclusionfunctor ℓ ω : J ω → I F has a left adjoint, whence is coaspherical in any weak basic localizer (i.e.is homotopy cofinal); see [ Mal05 , 1.1.9, 1.1.16 and 1.1.25]. As any model category defines aGrothendieck derivator ([
Cis03 , Thm. 6.11]), it follows from [
Cis03 , Cor. 1.15] that the map(3.1.23.2) is an isomorphism.To prove the general case, we proceed as follows. Let F /i be the functor obtained by com-posing F with the canonical functor v i : I/i → I . Then, keeping track of the conventions adoptedin 3.1.12, we check easily that ( I/i ) F /i = ( I F ) /i and that R ( F /i ) = ( R F ) /i . Moreover, thepullback square (3.1.22.1) is the composition of the following pullback squares of categories. J i a i / / q (cid:15) (cid:15) I F /i u i / / p/i (cid:15) (cid:15) I F p (cid:15) (cid:15) e ( i, i ) / / I/i v i / / I The pullback square of the proposition is thus the composition of the following pullback squares.( F ( i ) , J i ) α i / / ϕ i (cid:15) (cid:15) ( R F /i, I F /i ) µ i / / ϕ/i (cid:15) (cid:15) ( R F , I F ) ϕ (cid:15) (cid:15) X ( i, i ) / / ( X, I/i ) v i / / ( X, I ) . DESCENT IN P -FIBRED MODEL CATEGORIES 67 The natural transformations( i, i ) ∗ R ( ϕ/i ) ∗ → R ϕ i, ∗ α ∗ i and v ∗ i R ϕ ∗ → R ( ϕ/i ) ∗ µ ∗ i are both isomorphisms: the first one comes from the fact that ( i, i ) is a terminal object of I/i ,and the second one from Proposition 3.1.16. We thus get: i ∗ R ϕ ∗ ( M ) ≃ ( i, i ) ∗ v ∗ i R ϕ ∗ ( M ) ≃ ( i, i ) ∗ R ( ϕ/i ) ∗ µ ∗ i ( M ) ≃ R ϕ i, ∗ α ∗ i µ ∗ i ( M ) ≃ R ϕ i, ∗ λ ∗ i ( M ) . The last assertion of the proposition is then a straightforward application of Proposition 3.1.15. (cid:3)
Proposition . If M is a monoidal P -fibred combinatorial model category over S ,then, for any S -diagram X indexed by a small category I , the injective model structure turns M ( X , I ) into a symmetric monoidal model category. In particular, the categories Ho( M )( X , I ) are canonically endowed with a closed symmetric monoidal structure, in such a way that, for anymorphism of S -diagrams ϕ : ( X , I ) → ( Y , J ) , the functor L ϕ ∗ : Ho( M )( Y , J ) → Ho( M )( X , I ) is symmetric monoidal. Proof.
This is obvious from the definition of a symmetric monoidal model category, asthe tensor product of M ( X , I ) is defined termwise, as well as the cofibrations and the trivialcofibrations. (cid:3) Proposition . Assume that M is a monoidal P -fibred combinatorial model categoryover S , and consider a reduced cartesian P -morphism ϕ = ( α, f ) : ( X , I ) → ( Y , I ) . Then, forany object M in Ho( M )( X , I ) and any object N in Ho( M )( Y , I ) , the canonical map L ϕ ♯ ( M ⊗ L ϕ ∗ ( N )) → L ϕ ♯ ( M ) ⊗ L N is an isomorphism. Proof.
Let i be an object of I . It is sufficient to prove that the map i ∗ L ϕ ♯ ( M ⊗ L ϕ ∗ ( N )) → i ∗ L ϕ ♯ ( M ) ⊗ L N is an isomorphism in Ho( M )( X i ). Using Corollary 3.1.20, we see that this map can be identifiedwith the map L ϕ i,♯ ( M i ⊗ L ϕ ∗ i ( N i )) → L ϕ i,♯ ( M i ) ⊗ L N i , which is an isomorphism according to the P -projection formula for the homotopy P -fibred cate-gory Ho( M ). (cid:3) . Let ( X , I ) be a S -diagram. An object M of M ( X , I ) is homotopy cartesian if, forany map u : i → j in I , the structural map u ∗ ( M j ) → M i induces an isomorphism L u ∗ ( M i ) ≃ M j in Ho( M )( X , I ) (i.e. if there exists a weak equivalence M ′ j → M j with M ′ j cofibrant in M ( X j )such that the map u ∗ ( M ′ j ) → M i is a weak equivalence in M ( X i )).We denote by Ho( M )( X , I ) hcart the full subcategory of Ho( M )( X , I ) spanned by homotopycartesian sections. Definition . A cofibrantly generated model category V is tractable if there exist sets I and J of cofibrations between cofibrant objects which generate the class of cofibrations and theclass of trivial cofibrations respectively. Remark . If M is a combinatorial and tractable P -fibred model category over S , thenso are the projective and the injective model structures on M ( X , I ); see [ Bar10 , Thm. 2.28 and2.30]. Proposition . If M is tractable, then the inclusion functor Ho( M )( X , I ) hcart → Ho( M )( X , I ) admits a right adjoint. Proof.
This follows from the fact that the cofibrant homotopy cartesian sections are thecofibrant objects of a right Bousfield localization of the injective model structure on M ( X , I );see [ Bar10 , Theorem 5.25]. (cid:3)
Definition . Let M and M ′ two P -fibred model categories over S . A Quillenmorphism γ from M to M ′ is a morphism of P -fibred categories γ : M → M ′ such that γ ∗ : M ( X ) → M ′ ( X ) is a left Quillen functor for any object X of S . Remark . If γ : M → M ′ is a Quillen morphism between P -fibred combinatorialmodel categories, then, for any S -diagram ( X , I ), we get a Quillen adjunction γ ∗ : M ( X , I ) ⇄ M ′ ( X , I ) : γ ∗ (with the injective model structures as well as with the projective model structures). Proposition . For any Quillen morphism γ : M → M ′ , the derived adjunctions L γ ∗ : Ho( M )( X ) ⇄ Ho( M ′ )( X ) : R γ ∗ define a morphism of P -fibred categories Ho( M ) → Ho( M ′ ) over S . If moreover M and M ′ are combinatorial, then the morphism Ho( M ) → Ho( M ′ ) extends to a morphism of P cart -fibredcategories over the category of S -diagrams. Proof.
This follows immediately from [
Hov99 , Theorem 1.4.3]. (cid:3) . Let S be an essentially small category, and P an admissible class of morphisms in S . We assume that a Grothendieck topology t on S is given. We shall write S ∐ for the fullsubcategory of the category of S -diagrams whose objects are the small families X = { X i } i ∈ I ofobjects of S (seen as functors from a discrete category to S ). The category S ∐ is equivalent tothe full subcategory of the category of presheaves of sets on S spanned by sums of representablepresheaves. In particular, small sums are representable in S ∐ (but note that the functor from S to S ∐ does not preserve sums). Finally, we remark that the topology t extends naturally toa Grothendieck topology on S ∐ such that the topology t on S is the topology induced from theinclusion S ⊂ S ∐ . The covering maps for this topology on S ∐ will be called t -covers (note thatthe inclusion S ⊂ S ∐ is continuous and induces an equivalence between the topos of t -sheaveson S and the topos of t -sheaves on S ∐ ).Let ∆ be the category of non-empty finite ordinals. Remember that a simplicial object of S ∐ is a presheaf on ∆ with values in S ∐ . For a simplicial set K and an object X of S ∐ , we denoteby K × X the simplicial object of S ∐ defined by( K × X ) n = a x ∈ K n X , n ≥ . We write ∆ n for the standard combinatorial simplex of dimension n , and i n : ∂ ∆ n → ∆ n for itsboundary inclusion.A morphism p : X → Y between simplicial objects of S ∐ is a t -hypercover if, locally forthe t -topology, it has the right lifting property with respect to boundary inclusions of standardsimplices, which, in a more precise way, means that, for any integer n ≥
0, any object U of S ∐ ,and any commutative square ∂ ∆ n × U x / / i n × (cid:15) (cid:15) X p (cid:15) (cid:15) ∆ n × U y / / Y , . DESCENT IN P -FIBRED MODEL CATEGORIES 69 there exists a t -covering q : V → U , and a morphism of simplicial objects z : ∆ n × V → X , suchthat the diagram bellow commutes. ∂ ∆ n × V x (1 × q ) / / i n × (cid:15) (cid:15) X p (cid:15) (cid:15) ∆ n × V y (1 × q ) / / z : : ✉✉✉✉✉✉✉✉✉ Y A t -hypercover of an object X of S ∐ is a a t -hypercover p : X → X (where X is considered as aconstant simplicial object). Remark . This definition of t -hypercover is equivalent to the one given in [ SGA4 ,Exp. V, 7.3.1.4].3.2.3 . Let X be a simplicial object of S ∐ . It is in particular a functor from the category∆ op to the category of S -diagrams, so that the constructions and considerations of 3.1.22 applyto X . In particular, there is a S -diagram ˜ X associated to X , namely ˜ X = ( R X , (∆ op ) X ).More explicitely, for each integer n ≥
0, there is a family { X n,x } x ∈ K n of objects of S , such that(3.2.3.1) X n = a x ∈ K n X n,x . In fact, the sets K n form a simplicial set K , and the category (∆ op ) X can be identified over ∆ op tothe category (∆ /K ) op , where ∆ /K is the fibred category over ∆ whose fiber over n is the set K n (seen as a discrete category), i.e. the category of simplices of K . We shall call K the underlyingsimplicial set of X , while the decomposition (3.2.3.1) will be called the local presentation of X .The construction X ˜ X is functorial. If p : X → Y is a morphism of simplicial objects of S ∐ ,we shall denote by ˜ p : ˜ X → ˜ Y the induced morphism of S -diagrams. However, for a morphismof p : X → X , where X is an object of S ∐ , we shall still denote by p : ˜ X → X the correspondingmorphism of S -diagrams.Let M be a P -fibred combinatorial model category over S . Given a simplicial object X of S ∐ , we define the category Ho( M )( X ) by the formula:(3.2.3.2) Ho( M )( X ) = Ho( M )( R X , (∆ op ) X ) . Given an object X of S ∐ and a morphism p : X → X , we have a derived adjunction(3.2.3.3) L p ∗ : Ho( M )( X ) ⇄ Ho( M )( X ) : R p ∗ . Proposition . Consider an object X of S , a simplicial object X of S ∐ , as well as amorphism p : X → X . Denote by K the underlying simplicial set of X , and for each integer n ≥ and each simplex x ∈ K n , write p n,x : X n,x → X for the morphism of S ∐ induced bythe local presentation of X (3.2.3.1) . Then, for any object M of Ho( M )( X ) , there are canonicalisomorphisms R p ∗ R p ∗ ( M ) ≃ R lim ←− n ∈ ∆ R p n, ∗ L p ∗ n ( M ) ≃ R lim ←− n ∈ ∆ (cid:16) Y x ∈ K n R p n,x, ∗ L p ∗ n,x ( M ) (cid:17) . Proof.
The first isomorphism is a direct application of the last assertion of Proposition 3.1.23for F = X , while the second one follows from the first one by Proposition 3.1.10. (cid:3) Definition . Given an object Y of S ∐ , an object M of Ho( M )( Y ) will be said to satisfy t -descent if it has the following property: for any morphism f : X → Y and any t -hypercover p : X → X , the map R f ∗ L f ∗ ( M ) → R f ∗ R p ∗ L p ∗ L f ∗ ( M )is an isomorphism in Ho( M )( Y ).We shall say that M (or by abuse, that Ho( M )) satisfies t -descent if, for any object Y of S ∐ , any object of Ho( M )( Y ) satisfies t -descent. Proposition . If Y = { Y i } i ∈ I is a small family of objects of S (seen as an objectof S ∐ ), then an object M of Ho( M )( Y ) satisfies t -descent if and only if, for any i ∈ I , anymorphism f : X → Y i of S , and any t -hypercover p : X → X , the map R f ∗ L f ∗ ( M i ) → R f ∗ R p ∗ L p ∗ L f ∗ ( M i ) is an isomorphism in Ho( M )( Y i ) . Proof.
This follows from the definition and from Proposition 3.1.10. (cid:3)
Corollary . The P -fibred model category M satisfies t -descent if and only if, for anyobject X of S , and any t -hypercover p : X → X , the functor L p ∗ : Ho( M )( X ) → Ho( M )( X ) is fully faithful. Proposition . If M satisfies t -descent, then, for any t -cover f : Y → X , the functor L f ∗ : Ho( M )( X ) → Ho( M )( Y ) is conservative. Proof.
Let f : Y → X be a t -cover, and u : M → M ′ a morphism of Ho( M )( X ) whoseimage by L f ∗ is an isomorphism. We can consider the ˇCech t -hypercover associated to f , that isthe simplicial object Y over X defined by Y n = Y × X Y × X · · · × X Y | {z } n + 1 times . Let p : Y → X be the canonical map. For each n ≥
0, the map p n : Y n → X factor through f ,from wich we deduce that the functor L p ∗ n : Ho( M )( X ) → Ho( M )( Y n )sends u to an isomorphism. This implies that the functor L p ∗ : Ho( M )( X ) → Ho( M )( Y )sends u to an isomorphism as well. But, as Y is a t -hypercover of X , the functor L p ∗ is fullyfaithful, from which we deduce that u is an isomorphism by the Yoneda Lemma. (cid:3) . Let V be a complete and cocomplete category. For an object X of S , define PSh ( S /X, V )as the category of presheaves on S /X with values in V . Then PSh ( C/ − , V ) is a P -fibred cate-gory (where, by abuse of notations, S denotes also the class of all maps in S ): this is a specialcase of the constructions explained in 3.1.2 applied to V , seen as a fibred category over the ter-minal category. To be more explicit, for each object X of S ∐ , we have a V -enriched Yonedaembedding(3.2.9.1) S ∐ /X × V → PSh ( S /X, V ) , ( U, M } 7→ U ⊗ M , where, if U = { U i } i ∈ I is a small family of objects of S /X , U ⊗ M is the presheaf(3.2.9.2) V a i ∈ I a a ∈ Hom S /S ( V,U i ) M .
For a morphism f : X → Y in S , the functor f ∗ : PSh ( S /Y, V ) → PSh ( S /X, V )is the functor defined by composition with the corresponding functor S /X → S /Y . The functor f ∗ has always a left adjoint f ♯ : PSh ( S /X, V ) → PSh ( S /Y, V ) , which is the unique colimit preserving functor defined by f ♯ ( U ⊗ M ) = U ⊗ M , . DESCENT IN P -FIBRED MODEL CATEGORIES 71 where, on the left hand side U is considered as an object over X , while, on the right hand side, U is considered as an object over Y by composition with f . Similary, if all the pullbacks by f arerepresentable in S (e.g. if f is a P -morphism), the functor f ∗ can be described as the colimitpreserving functor defined by the formula f ∗ ( U ⊗ M ) = ( X × Y U ) ⊗ M . If V is a cofibrantly generated model category, then, for each object X of S , the categoryPSh ( S /X, V ) is naturally endowed with the projective model category structure , i.e. with thecofibrantly generated model category structure whose weak equivalences and fibrations are definedtermwise (this is Proposition 3.1.6 applied to V , seen as a fibred category over the terminalcategory). The cofibrations of the projective model category structure on PSh ( S /X, V ) will becalled the projective cofibrations. If moreover V is combinatorial (resp. left proper, resp. rightproper, resp. stable), so is PSh ( S /X, V ). In particular, if V is a combinatorial model category,then PSh ( S / − , V ) is a P -fibred combinatorial model category over S .According to Definition 3.2.5, it thus makes sense to speak of t -descent in PSh ( S / − , V ).If U = { U i } i ∈ I is a small family of objects of S over X , and if F is a presheaf over S /X , wedefine(3.2.9.3) F ( U ) = Y i ∈ I F ( U i ) . the functor F F ( U ) is a right adjoint to the functor E U ⊗ E .We remark that a termwise fibrant presheaf F on S /X satisfies t -descent if and only if, forany object Y of S ∐ , and any t -hypercover Y → Y over X , the map F ( Y ) → R lim ←− n ∈ ∆ F ( Y n )is an isomorphism in Ho( V ). Proposition . If V is combinatorial and left proper, then the category of presheaves PSh ( S /X, V ) admits a combinatorial model category structure whose cofibrations are the projec-tive cofibrations, and whose fibrant objects are the termwise fibrant objects which satisfy t -descent.This model category structure will be called the t -local model category structure , and the corre-sponding homotopy category will be denoted by Ho t (PSh ( S /X, V )) .Moreover, any termwise weak equivalence is a weak equivalence for the t -local model structure,and the induced functor a ∗ : Ho(PSh ( S /X, V )) → Ho t (PSh ( S /X, V )) admits a fully faithful right adjoint a ∗ : Ho t (PSh ( S /X, V )) → Ho(PSh ( S /X, V )) whose essential image consists precisely of the full subcategory of Ho(PSh ( S /X, V )) spanned bythe presheaves which satisfy t -descent. Proof.
Let H be the class of maps of shape(3.2.10.1) hocolim n ∈ ∆ op Y n ⊗ E → Y ⊗ E , where Y is an object of S ∐ over X , Y → Y is a t -hypercover, and E is a cofibrant replacement ofan object which is either a source or a target of a generating cofibration of V . Define the t -localmodel category structure as the left Bousfield localization of Pr ( S /X, V ) by H ; see [ Bar10 ,Theorem 4.7]. We shall call t -local weak equivalences the weak equivalences of the t -local modelcategory structure. For each object Y over X , the functor Y ⊗ ( − ) is a left Quillen functor from V to Pr ( S /X, V ). We thus get a total left derived functor Y ⊗ L ( − ) : Ho( V ) → Ho t (PSh ( S /X, V )) whose right adjoint is the evaluation at Y . For any object E of V and any t -local fibrant presheaf F on S /X with values in V , we thus have natural bijections(3.2.10.2) Hom( E, F ( Y )) ≃ Hom( Y ⊗ L E, F ) , and, for any simplicial object Y of S /X , identifications(3.2.10.3) Hom( E, R lim ←− n ∈ ∆ F ( Y n )) ≃ Hom( L lim −→ n ∈ ∆ Y n ⊗ L E, F ) , One sees easily that, for any t -hypercover Y → Y and any cofibrant object E of V , the map(3.2.10.4) L lim −→ n ∈ ∆ Y n ⊗ L E → Y ⊗ L E is an isomorphism in the t -local homotopy category Ho t (PSh ( S /X, V )): by the small object ar-gument, the smallest full subcategory of Ho(PSh ( S /X, V )) which is stable by homotopy colimitsand which contains the source and the targets of the generating cofibrations is Ho t (PSh ( S /X, V ))itself, and the class of objects E of V such that the map (3.2.10.4) is an isomorphism in Ho( V )is sable by homotopy colimits. Similarly, we see that, for any object E , the functor ( − ) ⊗ L E preserves sums. As a consequence, we get from (3.2.10.2) and (3.2.10.3) that the fibrant objects ofthe t -local model category structure are precisely the termwise fibrant objects F of the projectivemodel structure which satisfy t -descent. The last part of the proposition follows from the generalyoga of left Bousfield localizations. (cid:3) . Let M be a P -fibred combinatorial model category over S , and S an object of S .Denote by S : S /S → S the canonical forgetful functor. Then there is a canonical morphism of S -diagrams(3.2.11.1) σ : ( S , S /S ) → ( S, S /S )(where ( S, S /S ) stands for the constant diagram with value S ). This defines a functor(3.2.11.2) R σ ∗ : Ho( M )( S , S /S ) → Ho( M )( S, S /S ) = Ho(PSh ( S /S, M ( S ))) . For an object M of Ho( M )( S ), one defines the presheaf of geometric derived global sections of M over S by the formula(3.2.11.3) R Γ geom ( − , M ) = R σ ∗ L σ ∗ ( M ) . This is a presheaf on S /S with values in M ( S ) whose evaluation on a morphism f : X → S is,by virtue of Propositions 3.1.15 and 3.1.16,(3.2.11.4) R Γ geom ( X, M ) ≃ R f ∗ L f ∗ ( M ) . Proposition . For an object M of Ho( M )( S ) , the following conditions are equivalent.(a) The object M satisfies t -descent.(b) The presheaf R Γ geom ( − , M ) satisfies t -descent. Proof.
For any morphism f : X → S and any t -hypercover p : X → X over S , we have, byProposition 3.2.4 and formula (3.2.11.4), an isomorphism R f ∗ R p ∗ L p ∗ L f ∗ ( M ) ≃ R lim ←− n ∈ ∆ R Γ geom ( X n , M ) . From there, we see easily that conditions (a) and (b) are equivalent. (cid:3) . The preceding proposition allows to reduce descent problems in a fibred model categoryto descent problems in a category of presheaves with values in a model category. On can evengo further and reduce the problem to category of presheaves with values in an ‘elementary modelcategory’ as follows.Consider a model category V . Then one can associate to V its corresponding prederivator Ho ( V ), that is the strict 2-functor from the 2-category of small categories to the 2-category ofcategories, defined by(3.2.13.1) Ho ( V )( I ) = Ho( V I op ) = Ho(PSh ( I, V )) . DESCENT IN P -FIBRED MODEL CATEGORIES 73 for any small category I . More explicitly: for any functor u : I → J , one gets a functor u ∗ : Ho ( V )( J ) → Ho ( V )( I )(induced by the composition with u ), and for any morphism of functors I u ' ' v ✤✤ ✤✤ (cid:11) (cid:19) α J , one has a morphism of functors Ho ( V )( I ) Ho ( V )( J ) v ∗ l l u ∗ r r ✤ ✤✤ ✤ K S α ∗ . Moreover, the prederivator Ho ( V ) is then a Grothendieck derivator; see [ Cis03 , Thm. 6.11]. Thismeans in particular that, for any functor between small categories u : I → J , the functor u ∗ hasa left adjoint(3.2.13.2) L u ♯ : Ho ( V )( I ) → Ho ( V )( J )as well as a right adjoint(3.2.13.3) R u ∗ : Ho ( V )( I ) → Ho ( V )( J )(in the case where J = e is the terminal category, then L u ♯ is the homotopy colimit functor, while R u ∗ is the homotopy limit functor).If V and V ′ are two model categories, a morphism of derivators Φ : Ho ( V ) → Ho ( V ′ )is simply a morphism of 2-functors, that is the data of functors Φ I : Ho ( V )( I ) → Ho ( V ′ )( I )together with coherent isomorphisms u ∗ ( Φ J ( F )) ≃ Φ I ( u ∗ ( F ))for any functor u : I → J and any presheaf F on J with values in V (see [ Cis03 , p. 210] for aprecise definition).Such a morphism Φ is said to be continuousmorphism!continuous if, for any functor u : I → J ,and any object F of Ho ( V )( I ), the canonical map(3.2.13.4) Φ J R u ∗ ( F ) → R u ∗ Φ I ( F )is an isomorphism. One can check that a morphism of derivators Φ is continuous if and only ifit commutes with homotopy limits (i.e. if and only if the maps (3.2.13.4) are isomorphisms inthe case where J = e is the terminal category); see [ Cis08 , Prop. 2.6]. For instance, the totalright derived functor of any right Quillen functor defines a continuous morphism of derivators; see[
Cis03 , Prop. 6.12].Dually a morphism Φ of derivators is cocontinuous if, for any functor u : I → J , and anyobject F of Ho ( V )( I ), the canonical map(3.2.13.5) L u ! Φ I ( F ) → Φ J L u ! ( F )is an isomorphism.3.2.14 . We shall say that a stable model category V is Q -linear if all the objects of thetriangulated category Ho( V ) are uniquely divisible. Theorem . Let V be a model category (resp. a stable model category, resp. a Q -linearstable model category), and denote by S the model category of simplicial sets (resp. the stablemodel category of S -spectra, resp. the Q -linear stable model category of complexes of Q -vectorspaces). Denote by the unit object of the closed symmetric monoidal category Ho( S ) .Then, for each object E of Ho( V ) , there exists a unique continuous morphism of derivators R Hom( E, − ) : Ho ( V ) → Ho ( S ) such that, for any object F of Ho( V ) , there is a functorial bijection Hom
Ho( S ) ( , R Hom(
E, F )) ≃ Hom
Ho( V ) ( E, F )) . Proof.
Note that the stable Q -linear case follows from the stable case and from the factthat the derivator of complexes of Q -vector spaces is (equivalent to) the full subderivator of thederivator of S -spectra spanned by uniquely divisible objects.It thus remains to prove the theorem in the case where V be a model category (resp. a stablemodel category) and S is the model category of simplicial sets (resp. the stable model category of S -spectra). The existence of R Hom( E, − ) follows then from [ Cis03 , Prop. 6.13] (resp. [
CT11 ,Lemma A.6]).For the unicity, as we don’t really need it here, we shall only sketch the proof (the case ofsimplicial sets is done in [
Cis03 , Rem. 6.14]). One uses the universal property of the derivator Ho ( S ): by virtue of [ Cis08 , Cor. 3.26] (resp. of [
CT11 , Thm. A.5]), for any model category (resp.stable model category) V ′ there is a canonical equivalence of categories between the categoryof cocontinous morphisms from Ho ( S ) to Ho ( V ′ ) and the homotopy category Ho( V ). As aconsequence, the derivator Ho ( S ) admits a unique closed symmetric monoidal structure, and anyderivator (resp. triangulated derivator) is naturally and uniquely enriched in Ho ( S ); see [ Cis08 ,Thm. 5.22]. More concretely, this universal property gives, for any object E in Ho( V ′ ), a uniquecocontinuous morphism of derivators Ho ( S ) → Ho ( V ′ ) , K K ⊗ E such that ⊗ E = E . For a fixed K in Ho ( S )( I ), this defines a cocontinuous morphism ofderivators Ho ( V ′ ) → Ho ( V ′ I op ) , E K ⊗ E which has a right adjoint Ho ( V ′ I op ) → Ho ( V ′ ) , F F K . Let R Hom( E, − ) : Ho ( V ) → Ho ( S )be a continuous morphism such that, for any object F of V , there is a functorial bijection i F : Hom Ho( S ) ( , R Hom(
E, F )) ≃ Hom
Ho( V ) ( E, F )) . Then, for any object K of Ho ( S )( I ), and any object F of Ho ( V )( I ) a canonical isomorphism R Hom(
E, F K ) ≃ R Hom(
E, F ) K which is completely determined by being the identity for K = (this requires the full univer-sal property of Ho ( S ) given by by [ Cis08 , Thm. 3.24] (resp. by the dual version of [
CT11 ,Thm. A.5])). We thus get from the functorial bijections i F the natural bijections:Hom Ho ( S )( I ) ( K, R Hom(
E, F )) ≃ Hom
Ho( S ) ( , R Hom(
E, F ) K ) ≃ Hom
Ho( S ) ( , R Hom(
E, F K )) ≃ Hom
Ho( V ) ( E, F K ) ≃ Hom Ho ( V )( I ) ( K ⊗ E, F ) . In other words, R Hom( E, − ) has to be a right adjoint to ( − ) ⊗ E . (cid:3) . DESCENT IN P -FIBRED MODEL CATEGORIES 75 Remark . The preceding theorem mostly holds for abstract derivators. The only prob-lem is for the existence of the morphism R Hom( E, − ) (the unicity is always clear). However, thisproblem disapears for derivators which have a Quillen model (as we have seen above), as well asfor triangulated derivators (see [ CT11 , Lemma A.6]). Hence Theorem 3.2.15 holds in fact for anytriangulated Grothendieck derivator.In the case when V is a combinatorial model category (which, in practice, will essentiallyalways be the case), the enrichment over simplicial sets (resp, in the stable case, over spectra) canbe constructed via Quillen functors by Dugger’s presentation theorems [ Dug01 ] (resp. [
Dug06 ]).
Corollary . Let M be a P -fibred combinatorial model category (resp. a stable P -fibred combinatorial model category, resp. a Q -linear stable P -fibred combinatorial model category)over S , and S the model category of simplicial sets (resp. the stable model category of S -spectra,resp. the Q -linear stable model category of complexes of Q -vector spaces).Consider an object S of S , a morphism f : X → S , and a morphism of S -diagrams p :( X , I ) → X over S . Then, for an object M of Ho( M )( S ) , the following conditions are equivalent.(a) The map R f ∗ L f ∗ ( M ) → R f ∗ R p ∗ L p ∗ L f ∗ ( M ) is an isomorphism in Ho( M )( S ) .(b) The map R Γ geom ( X, M ) → R lim ←− i ∈ I op R Γ geom ( X i , M ) is an isomorphism in Ho( M )( S ) .(c) For any object E of Ho( M )( S ) , the map R Hom( E, R Γ geom ( X, M )) → R lim ←− i ∈ I op R Hom( E, R Γ geom ( X i , M )) is an isomorphism in Ho( S ) . Proof.
The equivalence between (a) and (b) follows from Propositions 3.1.15 and 3.1.16,which give the formula R f ∗ R p ∗ L p ∗ L f ∗ ( M ) ≃ R lim ←− i ∈ I op R Γ geom ( X i , M ) . The identification Hom
Ho( S ) ( , R Hom(
E, F )) ≃ Hom
Ho( M )( S ) ( E, F )and the Yoneda Lemma show that a map in Ho( M )( S ) is an isomorphism if and only its imageby R Hom( E, − ) is an isomorphism for any object E of Ho( M )( S ). Moreover, as R Hom( E, − )is continuous, for any small category I and any presheaf F on I with values in M ( S ), there is acanonical isomorphism R Hom( E, R lim ←− i ∈ I op F i )) ≃ R lim ←− i ∈ I op R Hom(
E, F i )) . This proves the equivalence between contitions (b) and (c). (cid:3)
Corollary . Under the assumptions of Corollary 3.2.17, given an object S of S , anobject M of Ho( M )( S ) satisfies t -descent if and only if, for any object E of Ho( M )( S ) the presheafof simplicial sets (resp. of S -spectra, resp. of complexes of Q -vector spaces) R Hom( E, R Γ geom ( − , M )) satisfies t -descent over S /S . Proof.
This follows from the preceding corollary, using the formula given by Proposition3.2.4. (cid:3) Remark . We need the category S to be small in some sense to apply the two precedingcorollaries because we need to make sense of the model projective category structure of Proposition3.2.10. However, we can use these corollaries even if the site S is not small as well: we can eitheruse the theory of universes, or apply these corollaries to all the adequate small subsites of S . Asa consequence, we shall feel free to use Corollaries 3.2.17 and 3.2.18 for non necessarily small sites S , leaving to the reader the task to avoid set-theoretic difficulties according to her/his taste. Definition . For an S -spectrum E and an integer n , we define its n th cohomologygroup H n ( E ) by the formula H n ( E ) = π − n ( E ) , where π i stands for the i th stable homotopy group functor.Let M be a monoidal P -fibred stable combinatorial model category over S . Given an object S of S as well as an object M of Ho( M )( S ), we define the presheaf of absolute derived globalsections of M over S by the formula R Γ( − , M ) = R Hom( S , R Γ geom ( − , M )) . For a map X → S of S , we thus have the absolute cohomology of X with coefficients in M , R Γ( X, M ), as well as the cohomology groups of X with coefficients in M : H n ( X, M ) = H n ( R Γ( X, M )) . We have canonical isomorphisms of abelian groups H n ( X, M ) ≃ Hom
Ho( M )( S ) ( S , R f ∗ L f ∗ ( M )) ≃ Hom
Ho( M )( X ) ( X , L f ∗ ( M )) . Note that, if moreover M is Q -linear, the presheaf R Γ( − , M ) can be considered as a presheafof complexes of Q -vector spaces on S /S . The aim of this section is to give natural sufficient conditionsfor M to satisfy descent with respect to various Grothendieck topologies .3.3.a. Localization and Nisnevich descent. . Recall from example 2.1.11 that a
Nisnevich distinguished square is a pullback squareof schemes V l / / g (cid:15) (cid:15) Y f (cid:15) (cid:15) U j / / X (3.3.1.1)in which f is ´etale, j is an open immersion with reduced complement Z and the induced morphism f − ( Z ) → Z is an isomorphism.For any scheme X in S , we denote by X Nis the small Nisnevich site of X . Theorem . Let V be a (combinatorial) model category and T ascheme in S . For a presheaf F on T Nis with values in V , the following conditions are equivalent.(i) F ( ∅ ) is a terminal object in Ho( V ) , and for any Nisnevich distinguished square (3.3.1.1) in T Nis , the square F ( X ) / / (cid:15) (cid:15) F ( Y ) (cid:15) (cid:15) F ( U ) / / F ( V ) is a homotopy pullback square in V . In fact, using remark 3.2.16, all of this section (results and proofs) holds for an abstract algebraic prederivatorin the sense of Ayoub [
Ayo07a , Def. 2.4.13] without any changes (note that the results of 3.1.b are in fact a proofthat (stable) combinatorial fibred model categories over S give rise to algebraic prederivators). The only interestof considering a fibred model category over S is that it allows to formulate things in a little more naive way. . DESCENT IN P -FIBRED MODEL CATEGORIES 77 (ii) The presheaf F satisfies Nisnevich descent on T Nis . Proof.
By virtue of corollaries 3.2.17 and 3.2.18, it is sufficient to prove this in the casewhere V is the usual model category of simplicial sets, in which case this is precisely Morel andVoevodsky’s theorem; see [ MV99, Voe10b, Voe10c ]. (cid:3) . Consider a Nisnevich distinguished square (3.3.1.1) and put a = jg = f l . According toour general assumption 3.0, the maps a , j and f are P -morphisms. For any object M of M ( X ),we obtain a commutative square in M (which is well defined as an object in the homotopy ofcommutative squares in M ( X )): L a ♯ a ∗ M / / (cid:15) (cid:15) L f ♯ f ∗ ( M ) (cid:15) (cid:15) L j ♯ j ∗ ( M ) / / M. (3.3.3.1)We also obtain another commutative square in M by appyling the functor R Hom X ( − , X ): M / / (cid:15) (cid:15) R f ∗ f ∗ ( M ) (cid:15) (cid:15) R j ∗ j ∗ ( M ) / / R a ∗ a ∗ ( M ) . (3.3.3.2) Proposition . If the category
Ho( M ) has the localization property, then for any Nis-nevich distinguished square (3.3.1.1) and any object M of Ho( M )( X ) , the squares (3.3.3.1) and (3.3.3.2) are homotopy cartesians. Proof.
Let i : Z → X be the complement of the open immersion j ( Z being endowed withthe reduced structure) and p : f − ( Z ) → Z the map induced by f .We have only to prove that one of the squares (3.3.3.1), (3.3.3.2) are cartesian. We choose thesquare (3.3.3.1).Because the pair of functor ( L i ∗ , j ∗ ) is conservative on Ho( M )( X ), we have only to check thatthe pullback of (3.3.3.1) along j ∗ or L i ∗ is homotopy cartesian. But, using the P -base changeproperty, we see that the image of (3.3.3.1) by j ∗ is (canonically isomorphic to) the commutativesquare L g ♯ a ∗ ( M ) (cid:15) (cid:15) L g ♯ a ∗ ( M ) (cid:15) (cid:15) j ∗ ( M ) j ∗ ( M )which is obviously homotopy cartesian.Using again the P -base change property, we obtain that the image of (3.3.3.1) by L i ∗ isisomorphic in Ho( M ) to the square 0 / / p ♯ p ∗ L i ∗ ( M ) (cid:15) (cid:15) / / L i ∗ ( M )which is again obviously homotopy cartesian because p is an isomorphism (note for this last reason, p ♯ = L p ♯ ). (cid:3) Corollary . If Ho( M ) has the localization property then it satisfies Nisnevich descent. Proof.
This corollary thus follows immediately from Corollary 3.2.17, Theorem 3.3.2 andProposition 3.3.4. (cid:3) Remark . Note that using Theorem 3.3.2, if we assume only that Ho( M ) satisfies Nis-nevich descent, then the squares (3.3.3.1) and (3.3.3.2) are homotopy cartesians for any Nisnevichdistinguished square (3.3.1.1).Assume that M is monoidal with geometric sections M . Let S be a base scheme and considera Nisnevich distinguished square (3.3.1.1) of smooth S -schemes. Then the fact that the square(3.3.3.1) is homotopy cartesian implies there exists a canonical distinguished triangle: M S ( V ) g ∗ + l ∗ −−−−→ M S ( U ) ⊕ M S ( Y ) f ∗ + j ∗ −−−−→ M S ( X ) −→ M S ( V )[1]It is called the Mayer-Vietoris triangle associated with the square (3.3.1.1).3.3.b.
Proper base change isomorphism and descent by blow-ups. . Recall from example 2.1.11 that a cdh -distinguished square is a pullback square ofschemes T k / / g (cid:15) (cid:15) Y f (cid:15) (cid:15) Z i / / X (3.3.7.1)in which f is proper surjective, i a closed immersion and the induced map f − ( X − Z ) → X − Z is an isomorphism.Recall from Example 2.1.11 the cdh -topology is the Grothendieck topology on the category ofschemes generated by Nisnevich coverings and by coverings of shape { Z → X, Y → X } for anycdh-distinguished square (3.3.7.1). Theorem . Let V be a (combinatorial) model category. For a presheaf F on S with values in V , the following conditions are equivalent.(i) The presheaf F satisfies cdh -descent on S .(ii) The presheaf F satisfies Nisnevich descent and, for any cdh -distinguished square (3.3.7.1) of S , the square F ( X ) / / (cid:15) (cid:15) F ( Y ) (cid:15) (cid:15) F ( Z ) / / F ( T ) is a homotopy pullback square in V . Proof.
It is sufficient to prove this in the case where V is the usual model category of simpli-cial sets; see corollaries 3.2.17 and 3.2.18. As the distinguished cdh-squares define a bounded reg-ular and reduced cd -structure on S , the equivalence between (i) and (ii) follows from Voevodsky’stheorems on descent with respect to topologies defined by cd -structures [ Voe10b, Voe10c ]. (cid:3) . Consider a cdh-distinguished square (3.3.7.1) and put a = ig = f k . For any object M of M ( X ), we obtain a commutative square in M (which is well defined as an object in thehomotopy of commutative squares in M ( X )): M / / (cid:15) (cid:15) R f ∗ L f ∗ ( M ) (cid:15) (cid:15) R i ∗ L i ∗ ( M ) / / R a ∗ L a ∗ ( M )(3.3.9.1) Proposition . Assume
Ho( M ) satisfies the localization property and the transversalityproperty with respect to proper morphisms. Then the following conditions hold:(i) For any cdh -distinguished square (3.3.7.1) , and any object M of Ho( M )( X ) the commu-tative square (3.3.9.1) is homotopy cartesian.(ii) The P -fibred model category Ho( M ) satisfies cdh -descent. . DESCENT IN P -FIBRED MODEL CATEGORIES 79 Proof.
We first prove (i). Consider a cdh-distinguished square (3.3.7.1) and let j : U → X bethe complement open immersion of i . As the pair of functor ( L i ∗ , j ∗ ) is conservative on Ho( M )( X ),we have only to check that the image of (3.3.9.1) under L i ∗ and j ∗ is homotopy cartesian.Using projective transversality, we see that the image of (3.3.9.1) by the functor L i ∗ is (iso-morphic to) the homotopy pullback square L i ∗ ( M ) / / R g ∗ L g ∗ L i ∗ ( M ) L i ∗ ( M ) / / R g ∗ L g ∗ L i ∗ ( M ) . Let h : f − ( U ) → U be the pullback of f over U . As j is an open immersion, it is byassumption a P -morphism and the P -base change formula implies that the image of (3.3.9.1) by j ∗ is (isomorphic to) the commutative square L j ∗ ( M ) / / (cid:15) (cid:15) R h ∗ L h ∗ L j ∗ ( M ) (cid:15) (cid:15) h is an isomorphism.We then prove (ii). We already know that M satisfies Nisnevich descent (Corollary 3.3.5).Thus, by virtue of the equivalence between conditions (i) and (ii) of Theorem 3.3.8, the computa-tion above, together with corollaries 3.2.17 and 3.2.18 imply that M satisfies cdh-descent. (cid:3) . To any cdh-distinguished square (3.3.7.1), one associates a diagram of schemes Y over X as follows. Let be the category freely generated by the oriented graph a / / (cid:15) (cid:15) bc (3.3.11.1)Then Y is the functor from to S /X defined by the following diagram. T k / / g (cid:15) (cid:15) YZ (3.3.11.2)We then have a canonical map ϕ : Y → X , and the second assertion of Theorem 3.3.10 can bereformulated by saying that the adjunction map M → R ϕ ∗ L ϕ ∗ ( M )is an isomorphism for any object M of Ho( M )( X ): indeed, by virtue of Proposition 3.1.15, R ϕ ∗ L ϕ ∗ ( M ) is the homotopy limit of the diagram R f ∗ L f ∗ ( M ) (cid:15) (cid:15) R i ∗ L i ∗ ( M ) / / R a ∗ L a ∗ ( M )in Ho( M )( X ). In other words, if M has the properties of localization and of projective transver-sality, then the functor L ϕ ∗ : Ho( M )( X ) → Ho( M )( Y , )is fully faithful. Proper descent with rational coefficients I: Galois excision.
From now on, we assumethat any scheme in S is quasi-excellent (in fact, we shall only use the fact that the normalizationof a quasi-excellent schemes gives rise to a finite surjective morphism, so that, in fact, universallyjapanese schemes would be enough). We fix a scheme S in S , and we shall work with S -schemesin S (assuming these form an essentially small category).3.3.12 . The h -topology (resp. the qfh -topology) is the Grothendieck topology on the category ofschemes associated to the pretopology whose coverings are the universal topological epimorphisms(resp. the quasi-finite universal topological epimorphisms). This topology has been introducedand studied by Voevodsky in [
Voe96 ].The h-topology is finer than the cdh-topology and, of course, finer than the qfh-topology. Theqfh-topology is in turn finer than the ´etale topology. An interesting feature of the h-topology (resp.of the qfh-topology) is that any proper (resp. finite) surjective map is an h-cover. In fact, theh-topology (resp. the qfh-topology) can be described as the topology generated by the Nisnevichcoverings and by the proper (resp. finite) surjective maps; see Lemma 3.3.28 (resp. Lemma 3.3.27)below for a precise statement.3.3.13 . Consider a morphism of schemes f : Y → X . Consider the group of automorphisms G = Aut Y ( X ) of the X -scheme Y .Assuming X is connected, we say according to [ SGA1 , exp. V] that f is a Galois cover if itis finite ´etale (thus surjective) and G operates transitively and faithfully on any (or simply one)of the geometric fibers of Y /X . Then G is called the Galois group of Y /X . When X is not connected, we will still say that f is a Galois cover if it is so over any connectedcomponent of X . Then G will be called the Galois group of X . If ( X i ) i ∈ I is the family connectedcomponents of X , then G is the product of the Galois groups G i of f × X X i for each i ∈ I . Thegroup G i is equal to the Galois group of any residual extension over a generic point of X i .The following definition is an extension of the definition 5.5 of [ SV00b ]: Definition . A pseudo-Galois cover is a finite surjective morphism of schemes f : Y → X which can be factored as Y f ′ −→ X ′ p −→ X where f ′ is a Galois cover and p is radicial (such a p is automatically finite and surjective).Note that the group G defined by the Galois cover f ′ is independent of the choice of thefactorization. In fact, if ¯ X denotes the semi-localization of X at its generic points, considering thecartesian squares ¯ Y / / (cid:15) (cid:15) ¯ X ′ / / (cid:15) (cid:15) ¯ X (cid:15) (cid:15) Y f ′ / / X ′ p / / X then G = Aut ¯ X ( ¯ Y ) – for any point y ∈ ¯ Y , x ′ = f ′ ( y ), x = f ( y ), κ x ′ /κ x is the maximal radicialsub-extension of the normal extension κ y /κ x . It will be called the Galois group of Y /X .Remark also that Y is a G -torsor over X locally for the qfh-topology (i.e. it is a Galois objectof group G in the qfh-topos of X ): this comes from the fact that finite radicial epimorphisms areisomorphisms locally for the qfh-topology (any universal homeomorphism has this property by[ Voe96 , prop. 3.2.5]).Let f : Y → X be a finite morphism, and G a finite group acting on Y over X . Note that, as Y is affine on X , the scheme theoretic quotient Y /G exists; see [
SGA1 , Exp. V, Cor. 1.8]. Suchscheme-theoretic quotients are stable by flat pullbacks; see [
SGA1 , Exp. V, Prop. 1.9]. See 4.1.1 below for a reminder on quasi-excellent schemes. The map f induces a one to one correspondence between the generic points of Y and that of X . For anygeneric point y ∈ Y , x = f ( y ), the residual extension κ y /κ x is a Galois extension with Galois group G . See 2.1.6 for a reminder on radicial morphisms. . DESCENT IN P -FIBRED MODEL CATEGORIES 81 Definition . Let G be finite group. A qfh -distinguished square of group G is a pullbacksquare of S -schemes of shape T h / / g (cid:15) (cid:15) Y f (cid:15) (cid:15) Z i / / X (3.3.15.1)in which Y is endowed with an action of G over X , and satisfying the following three conditions.(a) The morphism f is finite and surjective.(b) The induced morphism f − ( X − Z ) → f − ( X − Z ) /G is flat.(c) The morphism f − ( X − Z ) /G → X − Z is radicial.Immediate examples of qfh-distinguished squares of trivial group are the following. The scheme Y might be the normalization of X , and Z is a nowhere dense closed subscheme out of which f isan isomorphism; or Y is dense open subscheme of X which is the disjoint union of its irreduciblecomponents; or Y is a closed subscheme of X inducing an isomorphism Y red ≃ X red .A qfh-distinguished square of group G (3.3.15.1) will be said to be pseudo-Galois if Z isnowhere dense in X and if the map f − ( X − Z ) → X − Z is a pseudo-Galois cover of group G .The main examples of pseudo-Galois qfh-distinguished squares will come from the followingsituation. Proposition . Consider an irreducible normal scheme X , and a finite extension L ofits field of functions k ( X ) . Let K be the inseparable closure of k ( X ) in L , and assume that L/K is a Galois extension of group G . Denote by Y the normalization of X in L . Then the action of G on k ( Y ) = L extends naturally to an action on Y over X . Furthermore, there exists a closedsubscheme Z of X , such that the pullback square T / / (cid:15) (cid:15) Y f (cid:15) (cid:15) Z i / / X is a pseudo-Galois qfh -distinguished square of group G . Proof.
The action of G on L extends naturally to an action on Y over X by functoriality.Furthermore, Y /G is the normalization of X in K , so that Y /G → X is finite radicial and surjective(see [ Voe96 , Lemma 3.1.7] or [
Bou98 , V, §
2, n o
3, lem. 4]). By construction, Y is generically aGalois cover over Y /G , which implies the result (see [
EGA4 , Cor. 18.2.4]). (cid:3) . For a given S -scheme T , we shall denote by L ( T ) the corresponding representableqfh-sheaf of sets (remember that the qfh-topology is not subcanonical, so that L ( T ) has to bedistinguished from T itself). Beware that, in general, there is no reason that, given a finite group G acting on T , the scheme-theoretic quotient L ( T /G ) (whenever defined) and the qfh-sheaf-theoreticquotient L ( T ) /G would coincide. Lemma . Let f : Y → X be a separated morphism, G a finite group acting on Y over X , and Z a closed subscheme of X such that f is finite and surjective over X − Z , and such thatthe quotient map f − ( X − Z ) → f − ( X − Z ) /G is flat, while the map f − ( X − Z ) /G → X − Z is radicial. For g ∈ G , write g : Y → Y for the corresponding automorphism of Y , and define Y g as the image of the diagonal Y → Y × X Y composed with the automorphism Y × X g : Y × X Y → Y × X Y . Then, if T = Z × X Y , we get a qfh -cover of Y × X Y by closed subschemes: Y × X Y = ( T × Z T ) ∪ [ g ∈ G Y g . Proof.
Note that, as f is separated, the diagonal Y → Y × X Y is a closed embedding, sothat the Y g ’s are closed subschemes of Y × X Y . As the map Y × Y/G Y → Y × X Y is a universalhomeomorphism, we may assume that Y /G = X . It is sufficient to prove that, if y and y ′ are two geometric points of Y whose images coincide in X and do not belong to Z , there exists an element g of G such that y ′ = gy (which means that the pair ( y, y ′ ) belongs to Y g ). For this purpose,we may assume, without loss of generality, that Z = ∅ . Then, by assumption, Y is flat over X ,from which we get the identification ( Y × X Y ) /G ≃ Y × X ( Y /G ) ≃ Y (where the action of G on Y × X Y is trivial on the first factor and is induced by the action on Y on the second factor). Thisachieves the proof. (cid:3) Proposition . For any qfh -distinguished square of group G (3.3.15.1) , the commutativesquare L ( T ) /G / / (cid:15) (cid:15) L ( Y ) /G (cid:15) (cid:15) L ( Z ) / / L ( X ) is a pullback and a pushout in the category of qfh -sheaves. Moreover, if X is normal and if Z isnowhere dense in X , then the canonical map L ( Y ) /G → L ( Y /G ) ≃ L ( X ) is an isomorphism of qfh -sheaves (which implies that L ( T ) /G → L ( Z ) is an isomorphism as well). Proof.
Note that this commutative square is a pullback because it was so before taking thequotients by G (as colimits are universal in any topos). As f is a qfh-cover, it is sufficient to provethat L ( T ) × L ( Z ) L ( T ) /G / / (cid:15) (cid:15) L ( Y ) × L ( X ) L ( Y ) /G (cid:15) (cid:15) L ( T ) / / L ( Y )is a pushout square. This latter square fits into the following commutative diagram L ( T ) / / (cid:15) (cid:15) L ( Y ) (cid:15) (cid:15) L ( T ) × L ( Z ) L ( T ) /G / / (cid:15) (cid:15) L ( Y ) × L ( X ) L ( Y ) /G (cid:15) (cid:15) L ( T ) / / L ( Y )in which the two vertical composed maps are identities (the vertical maps of the upper commutativesquare are obtained from the diagonals by taking the quotients under the natural action of G onthe right component). It is thus sufficient to prove that the upper square is a pushout. As thelower square is a pullback, the upper one shares the same property; moreover, all the maps in theupper commutative square are monomorphisms of qfh-sheaves, so that it is sufficient to prove thatthe map ( L ( T ) × L ( Z ) L ( T ) /G ) ∐ L ( Y ) → L ( Y ) × L ( X ) L ( Y ) /G is an epimorphism of qfh-sheaves.According to Lemma 3.3.18, this follows from the commutativity of the diagram L ( T × Z T ) ∐ (cid:16) ` g ∈ G L ( Y g ) (cid:17) / / (cid:15) (cid:15) L ( Y × X Y ) (cid:15) (cid:15) ( L ( T ) × L ( Z ) L ( T ) /G ) ∐ L ( Y ) / / L ( Y ) × L ( X ) L ( Y ) /G in which the vertical maps are obviously epimorphic.Assume now that X is normal and that Z is nowhere dense in X , and let us prove that thecanonical map L ( Y ) /G → L ( X ) is an isomorphism of qfh-sheaves. This is equivalent to provethat, for any qfh-sheaf of sets F , the map f ∗ : F ( X ) → F ( Y ) induces a bijection F ( X ) ≃ F ( Y ) G . . DESCENT IN P -FIBRED MODEL CATEGORIES 83 Let F be a qfh-sheaf. The map f ∗ : F ( X ) → F ( Y ) is injective because f is a qfh-cover, and it isclear that the image of f ∗ lies in F ( Y ) G .Let a be a section of F over Y which is invariant under the action of G . Denote by pr , pr : Y × X Y → Y the two canonical projections. With the notations introduced in Lemma 3.3.18, wehave pr ∗ ( a ) | Y g = a = a.g = pr ∗ ( a ) | Y g for every element g in G . As Z does not contain any generic point of X , the scheme T × Z T doesnot contain any generic point of Y × X Y neither: as any irreducible component of Y dominatesan irreducible component of X , and, as X is normal, the finite map Y → X is universally open; inparticular, the projection pr : Y × X Y → Y is universally open, which implies that any genericpoint of Y × X Y lies over a generic point of Y . By virtue of [ Voe96 , prop. 3.1.4], Lemma 3.3.18thus gives a qfh-cover of Y × X Y by closed subschemes of shape Y × X Y = [ g ∈ G Y g . This implies that pr ∗ ( a ) = pr ∗ ( a ) . The morphism Y → X being a qfh-cover and F a qfh-sheaf, we deduce that the section a lies inthe image of f ∗ . (cid:3) Corollary . For any qfh -distinguished square of group G (3.3.15.1) , we get a bicarte-sian square of qfh -sheaves of abelian groups Z qfh ( T ) G / / (cid:15) (cid:15) Z qfh ( Y ) G (cid:15) (cid:15) Z qfh ( Z ) / / Z qfh ( X ) (where the subscript G stands for the coinvariants under the action of G ). In other words, thereis a canonical short exact sequence of sheaves of abelian groups → Z qfh ( T ) G → Z qfh ( Z ) ⊕ Z qfh ( Y ) G → Z qfh ( X ) → . Proof.
As the abelianization functor preserves colimits and monomorphisms, the precedingproposition implies formally that we have a short exact sequence of shape Z qfh ( T ) G → Z qfh ( Z ) ⊕ Z qfh ( Y ) G → Z qfh ( X ) → , while the left exactness follows from the fact that Z → X being a monomorphism, the mapobtained by pullback, L ( T ) /G → L ( Y ) /G , is a monomorphism as well. (cid:3) . Let V be a Q -linear stable model category (see 3.2.14).Consider a finite group G , and an object E of V , endowed with an action of G . By viewing G as a category with one object we can see E as functor from G to V and take its homotopy limitin Ho( V ), which we denote by E hG (in the literature, E hG is called the object of homotopy fixedpoints under the action of G on E ). One the other hand, the category Ho( V ) is, by assumption,a Q -linear triangulated category with small sums, and, in particular, a Q -linear pseudo-abeliancategory so that we can define E G as the object of Ho( V ) defined by(3.3.21.1) E G = Im p , where p : E → E is the projector defined in Ho( V ) by the formula(3.3.21.2) p ( x ) = 1 G X g ∈ G g.x . The inclusion E G → E induces a canonical isomorphism(3.3.21.3) E G ∼ → E hG in Ho( V ): to see this, by virtue of Theorem 3.2.15, we can assume that V is the model categoryof complexes of Q -vector spaces, in which case it is obvious. Corollary . Let C be a presheaf of complexes of Q -vector spaces on the category of S -schemes. Then, for any qfh -distinguished square of group G (3.3.15.1) , the commutative square R Γ qfh ( X, C qfh ) / / (cid:15) (cid:15) R Γ qfh ( Y, C qfh ) G (cid:15) (cid:15) R Γ qfh ( Z, C qfh ) / / R Γ qfh ( T, C qfh ) G is a homotopy pullback square in the derived category of Q -vector spaces. In particular, we get along exact sequence of shape H n qfh ( X, C qfh ) → H n qfh ( Z, C qfh ) ⊕ H n qfh ( Y, C qfh ) G → H n qfh ( T, C qfh ) G → H n +1qfh ( X, C qfh ) If furthermore X is normal and Z is nowhere dense in X , then the maps H n qfh ( X, C qfh ) → H n qfh ( Y, C qfh ) G and H n qfh ( Z, C qfh ) → H n qfh ( T, C qfh ) G are isomorphisms for any integer n . Proof.
Let C qfh → C ′ be a fibrant resolution in the qfh-local injective model category struc-ture on the category of qfh-sheaves of complexes of Q -vector spaces; see for instance [ Ayo07a ,Cor. 4.4.42]. Then for U = Y, T , we have a natural isomorphism of complexesHom( Q qfh ( U ) G , C ′ ) = C ′ ( U ) G which gives an isomorphism R Hom( Q qfh ( U ) G , C qfh ) ≃ R Γ qfh ( U, C qfh ) G in the derived category of the abelian category of Q -vector spaces. This corollary thus followsformally from Corollary 3.3.20 by evaluating at the derived functor R Hom( − , C qfh ).If furthermore X is normal, then one deduces the isomorphism H n qfh ( X, C qfh ) ≃ H n qfh ( Y, C qfh ) G from the fact that L ( Y ) /G ≃ L ( Y /G ) ≃ X (Proposition 3.3.19), which implies that Z qfh ( Y ) G ≃ Z qfh ( X ). The isomorphism H n qfh ( Z, C qfh ) ≃ H n qfh ( T, C qfh ) G then comes as a byproduct of the longexact sequence above. (cid:3) Theorem . Let X be a scheme, and C be a presheaf of complexes of Q -vector spaceson the small ´etale site of X . Then C satisfies ´etale descent if and only if it has the followingproperties.(a) The complex C satisfies Nisnevich descent.(b) For any ´etale X -scheme U and any Galois cover V → U of group G , the map C ( U ) → C ( V ) G is a quasi-isomorphism. Proof.
These are certainly necessary conditions. To prove that they are sufficient, notethat the Nisnevich cohomological dimension and the rational ´etale cohomological dimension ofa noetherian scheme are bounded by the dimension; see [
MV99 , proposition 1.8, page 98] and[
Voe96 , Lemma 3.4.7]. By virtue of [
SV00a , Theorem 0.3], for τ = Nis , ´et, we have stronglyconvergent spectral sequences E p,q = H pτ ( U, H q ( C ) τ ) ⇒ H p + qτ ( U, C τ ) . Condition (a) gives isomorphisms H p + q ( C ( U )) ≃ H p + q Nis ( U, C
Nis ), so that it is sufficient to provethat, for each of the cohomology presheaves F = H q ( C ), we have H p Nis ( U, F
Nis ) ≃ H p ´et ( U, F ´et ) . As the rational ´etale cohomology of any henselian scheme is trivial in non-zero degrees, it issufficient to prove that, for any local henselian scheme U (obtained as the henselisation of an ´etale . DESCENT IN P -FIBRED MODEL CATEGORIES 85 X -scheme at some point), F Nis ( U ) ≃ F ´et ( U ). Let G be the absolute Galois group of the closedpoint of U . Then we have F Nis ( U ) = F ( U ) and F ´et ( U ) = lim −→ α F ( U α ) G α , where the U α ’s run over all the Galois covers of U corresponding to the finite quotients G → G α .But it follows from (b) that F ( U ) ≃ F ( U α ) G α for any α , so that F Nis ( U ) ≃ F ´et ( U ). (cid:3) Lemma . Any qfh -cover admits a refinement of the form Z → Y → X , where Z → Y is a finite surjective morphism, and Y → X is an ´etale cover. Proof.
This property being clearly local on X with respect to the ´etale topology, we canassume that X is strictly henselian, in which case this follows from [ Voe96 , Lemma 3.4.2]. (cid:3)
Theorem . A presheaf of complexes of Q -vector spaces C on the category of S -schemessatisfies qfh -descent if and only if it has the following two properties:(a) the complex C satisfies Nisnevich descent;(b) for any pseudo-Galois qfh -distinguished square of group G (3.3.15.1) , the commutativesquare C ( X ) / / (cid:15) (cid:15) C ( Y ) G (cid:15) (cid:15) C ( Z ) / / C ( T ) G is a homotopy pullback square in the derived category of Q -vector spaces. Proof.
Any complex of presheaves of Q -vector spaces satisfying qfh-descent satisfies prop-erties (a) and (b): property (a) follows from the fact that the qfh-topology is finer than the ´etaletopology; property (b) is Corollary 3.3.22.Assume now that C satisfies these two properties. Let ϕ : C → C ′ be a morphism of presheavesof complexes of Q -vector spaces which is a quasi-isomorphism locally for the qfh-topology, andsuch that C ′ satisfies qfh-descent (such a morphism exists thanks to the qfh-local model categorystructure on the category of presheaves of complexes of Q -vector spaces; see Proposition 3.2.10).Then the cone of ϕ also satisfies conditions (a) and (b). Hence it is sufficient to prove the theoremin the case where C is acyclic locally for the qfh-topology.Assume from now on that C qfh is an acyclic complex of qfh-sheaves, and denote by H n ( C )the n th cohomology presheaf associated to C . We know that the associated qfh-sheaves vanish,and we want to deduce that H n ( C ) = 0.We shall prove by induction on d that, for any S -scheme X of dimension d and for any integer n , the group H n ( C )( X ) = H n ( C ( X )) vanishes. The case where d < H n ( C ) send finite sums to finite direct sums, so that, in particular, H n ( C )( ∅ ) = 0. Before going further, notice that condition (b) implies H n ( C )( X red ) = H n ( C )( X )for any S -scheme X (consider the case where, in the diagram (3.3.15.1), Z = Y = T = X red ), sothat it is always harmless to replace X by its reduction. Assume now that d ≥
0, and that thevanishing of H n ( C )( X ) is known whenever X is of dimension < d and for any integer n . Underthis inductive assumption, we have the following reduction principle.Consider a pseudo-Galois qfh-distinguished square of group G (3.3.15.1). If Z and T are ofdimension < d , then by condition (b), the map H n ( C )( X ) → H n ( C )( Y ) G is an isomorphism:indeed, we have an exact sequence of shape H n − ( C )( T ) G → H n ( C )( X ) → H n ( C )( Z ) ⊕ H n ( C )( Y ) G → H n ( C )( T ) G , which implies our assertion by induction on d .We shall prove now the vanishing of H n ( C )( T ) for normal S -schemes T of dimension d . Let a be a section of H n ( C ) over such a T . As H n ( C ) qfh ( T ) = 0, there exists a qfh-cover g : Y → T such that g ∗ ( a ) = 0. But, by virtue of Lemma 3.3.24, we can assume g is the composition of afinite surjective morphism f : Y → X and of an ´etale cover e : X → T . We claim that e ∗ ( a ) = 0.To prove it, as, by (a), the presheaf H n ( C ) sends finite sums to finite direct sums, we can assume that X is normal and connected. Refining f further, we can assume that Y is the normalizationof X in a finite extension of k ( X ), and that k ( Y ) is a Galois extension of group G over theinseparable closure of k ( X ) in k ( Y ). By virtue of Proposition 3.3.16, we get by the reductionprinciple the identification H n ( C )( X ) = H n ( C )( Y ) G , whence e ∗ ( a ) = 0. As a consequence, therestriction of the presheaf of complexes C to the category of normal S -schemes of dimension ≤ d isacyclic locally for the ´etale topology (note that this is quite meaningful, as any ´etale scheme overa normal scheme is normal; see [ EGA4 , Prop. 18.10.7]). But C satisfies ´etale descent (by virtueof Theorem 3.3.23 this follows formally from property (a) and from property (b) for Z = ∅ ), sothat H n ( C )( T ) = H n ´et ( T, C ´et ) = 0 for any normal S -scheme T of dimension ≤ d and any integer n . Consider now a reduced S -scheme X of dimension ≤ d . Let p : T → X be the normalization of X . As p is birational (see [ EGA2 , Cor. 6.3.8]) and finite surjective (because X is quasi-excellent),we can apply the reduction principle and see that the pullback map p ∗ : H n ( C )( X ) → H n ( C )( T ) =0 is an isomorphism for any integer n , which achieves the induction and the proof. (cid:3) Lemma . ´Etale coverings are finite ´etale coverings locally for the Nisnevich topology:any ´etale cover admits a refinement of the form Z → Y → X , where Z → Y is a finite ´etale coverand Y → X is a Nisnevich cover. Proof.
This property being local on X for the Nisnevich topology, it is sufficient to provethis in the case where X is local henselian. Then, by virtue of [ EGA4 , Cor. 18.5.12 and Prop.18.5.15], we can even assume that X is the spectrum of field, in which case this is obvious. (cid:3) Lemma . Any qfh -cover admits a refinement of the form Z → Y → X , where Z → Y is a finite surjective morphism, and Y → X is a Nisnevich cover. Proof.
As finite surjective morphisms are stable by pullback and composition, this followsimmediately from lemmata 3.3.24 and 3.3.26. (cid:3)
Lemma . Any h -cover of an integral scheme X admits a refinement of the form U → Z → Y → X , where U → Z is a finite surjective morphism, Z → Y is a Nisnevich cover, Y → X is a propersurjective birational map, and Y is normal. Proof.
By virtue of [
Voe96 , Theorem 3.1.9], any h-cover admits a refinement of shape W → V → X , where W → V is a qfh-cover, and V → X is a proper surjective birational map. By replacing V by its normalization Y , we get a refinement of shape W × V Y → Y → X where W × V Y → Y is a qfh-cover, and Y → X is proper surjective birational map. We concludeby Lemma 3.3.27. (cid:3) Lemma . Let C be a presheaf of complexes of Q -vector spaces on the category of S -schemes satisfying qfh -descent. Then, for any finite surjective morphism f : Y → X with X normal, the map f ∗ : H n ( C )( X ) → H n ( C )( Y ) is a monomorphism. Proof.
It is clearly sufficient to prove this when X is connected. Then, up to refinement, wecan assume that f is a map as in Proposition 3.3.16. In this case, by virtue of Corollary 3.3.22,the Q -vector space H n ( C )( X ) ≃ H n ( C )( Y ) G is a direct factor of H n ( C )( Y ). (cid:3) Theorem . A presheaf of complexes of Q -vector spaces on the category of S -schemessatisfies h -descent if and only if it satisfies qfh -descent and cdh -descent. . DESCENT IN P -FIBRED MODEL CATEGORIES 87 Proof.
This is certainly a necessary condition, as the h-topology is finer than the qfh-topology and the cdh-topology. For the converse, as in the proof of Theorem 3.3.25, it is suf-ficient to prove that any presheaf of complexes of Q -vector spaces C on the category of S -schemessatisfying qfh-descent and cdh-descent, and which is acyclic locally for the h-topology, is acyclic.We shall prove by noetherian induction that, given such a complex C , for any integer n , and any S -scheme X , for any section a of H n ( C ) over X , there exists a cdh-cover X ′ → X on which a vanishes. In other words, we shall get that C is acyclic locally for the cdh-topology, and, as C sat-isfies cdh-descent, this will imply that H n ( C )( X ) = H n cdh ( X, C cdh ) = 0 for any integer n and any S -scheme X . Note that the presheaves H n ( C ) send finite sums to finite direct sums (which follows,for instance, from the fact that C satisfies Nisnevich descent). In particular, H n ( C )( ∅ ) = 0 forany integer n .Let X be an S -scheme, and a ∈ H n ( C )( X ). We have a cdh-cover of X of shape X ′ ∐ X ′′ → X ,where X ′ is the sum of the irreducible components of X red and X ′′ is a nowhere dense closedsubscheme of X , so that we can assume X is integral. Let a be a section of the presheaf H n ( C )over X . As H n ( C ) h = 0, by virtue of Lemma 3.3.28, there exists a proper surjective birationnalmap p : Y → X with Y normal, a Nisnevich cover q : Z → Y , and a surjective finite morphism r : U → Z such that r ∗ ( q ∗ ( p ∗ ( a ))) = 0 in H n ( C )( U ). But then, Z is normal as well (see [ EGA4 ,Prop. 18.10.7]), so that, by Lemma 3.3.29, we have q ∗ ( p ∗ ( a )) = 0 in H n ( C )( Z ). Let T be anowhere dense closed subscheme of X such that p is an isomorphism over X − T . By noetherianinduction, there exists a cdh-cover T ′ → T such that a | T ′ vanishes. Hence the section a vanisheson the cdh-cover T ′ ∐ Z → X . (cid:3) Proper descent with rational coefficients II: separation.
From now on, we assume thatHo( M ) is Q -linear. Proposition . Let f : Y → X be a morphism of schemes in S , and G a finite groupacting on Y over X . Denote by Y the scheme Y considered a functor from G to the category of S -schemes, and denote by ϕ : ( Y , G ) → X the morphism induced by f . Then, for any object M of Ho( M )( X ) , there are canonical isomorphisms ( R f ∗ L f ∗ ( M )) G ≃ ( R f ∗ L f ∗ ( M )) hG ≃ R ϕ ∗ L ϕ ∗ ( M ) . Proof.
The second isomorphism comes from Proposition 3.1.15, and the first, from (3.3.21.3). (cid:3)
Theorem . If Ho( M ) satisfies Nisnevich descent, the following conditions are equiva-lent: (i) Ho( M ) satisfies ´etale descent.(ii) for any finite ´etale cover f : Y → X , the functor L f ∗ : Ho( M )( X ) → Ho( M )( Y ) is conservative;(iii) for any finite Galois cover f : Y → X of group G , and for any object M of Ho( M )( X ) ,the canonical map M → ( R f ∗ L f ∗ ( M )) G is an isomorphism. Proof.
The equivalence between (i) and (iii) follows from Theorem 3.3.23 by corollaries 3.2.17and 3.2.18, and Proposition 3.2.8 shows that (i) implies (ii). It is thus sufficient to prove that (ii)implies (iii). Let f : Y → X be a finite Galois cover of group G . As the functor f ∗ = L f ∗ isconservative by assumption, it is sufficient to check that the map M → ( R f ∗ L f ∗ ( M )) G becomesan isomorphism after applying f ∗ . By virtue of Proposition 3.1.17, this just means that it issufficient to prove (iii) when f has a section, i.e. when Y is isomorphic to the trivial G -torsor over X . In this case, we have the (equivariant) identification L g ∈ G M ≃ R f ∗ L f ∗ ( M ), where G actson the left term by permuting the factors. Hence M ≃ ( R f ∗ L f ∗ ( M )) G . (cid:3) Proposition . Assume that
Ho( M ) has the localization property. The following con-ditions are equivalent:(i) Ho( M ) is separated.(ii) Ho( M ) is semi-separated and satisfies ´etale descent. Proof.
This follows from Proposition 2.3.9 and Theorem 3.3.32. (cid:3)
Corollary . Assume that all the residue fields of S are of characteristic zero, and that M has the property of localization. Then the following conditions are equivalent:(i) Ho( M ) is separated.(ii) Ho( M ) satisfies ´etale descent. Proof.
In this case, a finite surjective morphism f : Y → X is radicial if and only if it inducesan isomorphism after reduction Y red ≃ X red . But it is clear that, by the localization property,such a morphism f induces an equivalence of categories L f ∗ , so that Ho( M ) is automaticallysemi-separated. We conclude by Proposition 3.3.33. (cid:3) Proposition . Assume that
Ho( M ) is separated, satisfies the localization property theproper transversality property. Then, for any pseudo-Galois cover f : Y → X of group G , and forany object M of Ho( M )( X ) , the canonical map M → ( R f ∗ L f ∗ ( M )) G is an isomorphism. Proof.
By Proposition 3.3.33, this is an easy consequence of Proposition 2.1.9 and of condi-tion (iii) of Theorem 3.3.32. (cid:3) . From now on, we assume furthermore that any scheme in S is quasi-excellent. Theorem . Assume that
Ho( M ) satisfies the localization and proper transversality prop-erties. Then the following conditions are equivalent:(i) Ho( M ) is separated;(ii) Ho( M ) satisfies h -descent;(iii) Ho( M ) satisfies qfh -descent;(iv) for any qfh -distinguished square (3.3.15.1) of group G , if we write a = f h = ig : T → X for the composed map, then, for any object M of Ho( M )( X ) , the commutative square M / / (cid:15) (cid:15) ( R f ∗ L f ∗ ( M )) G (cid:15) (cid:15) R i ∗ L i ∗ ( M ) / / ( R a ∗ L a ∗ ( M )) G (3.3.37.1) is homotopy cartesian;(v) the same as condition (iv), but only for pseudo-Galois qfh -distinguished squares. Proof. As M satisfies cdh-descent (Theorem 3.3.10), the equivalence between conditions (ii)and (iii) follows from Theorem 3.3.30 by Corollary 3.2.18. Similarly, Theorem 3.3.25 and corollaries3.3.22, 3.2.17 and 3.2.18 show that conditions (iii), (iv) and (v) are equivalent. As ´etale surjectivemorphisms as well as finite radicial epimorphisms are qfh-coverings, it follows from Proposition3.2.8, Theorem 3.3.32 and Proposition 3.3.33, that condition (iii) implies condition (i). It thusremains to prove that condition (i) implies condition (v). So let us consider a pseudo-Galois qfh-distinguished square (3.3.15.1) of group G , and prove that (3.3.37.1) is homotopy cartesian. Usingproper transversality, we see that the image of (3.3.37.1) by the functor L i ∗ is (isomorphic to) thehomotopy pullback square L i ∗ ( M ) / / ( R g ∗ L g ∗ L i ∗ ( M )) G L i ∗ ( M ) / / ( R g ∗ L g ∗ L i ∗ ( M )) G . . DESCENT IN P -FIBRED MODEL CATEGORIES 89 Write j : U → X for the complement open immersion of i , and b : f − ( U ) → U for the mapinduced by f . As j is ´etale, we see, using Proposition 3.1.17, that the image of (3.3.9.1) by j ∗ = L j ∗ is (isomorphic to) the square j ∗ ( M ) / / (cid:15) (cid:15) ( R b ∗ L b ∗ j ∗ ( M )) G (cid:15) (cid:15) . in which the upper horizontal map is an isomorphism by Proposition 3.3.35. Hence it is a homotopypullback square. Thus, because the pair of functors ( L i ∗ , j ∗ ) is conservative on Ho( M )( X ), thesquare (3.3.37.1) is homotopy cartesian. (cid:3) Corollary . Assume that all the residue fields of S are of characteristic zero, and that Ho( M ) has the localization and proper transversality properties. Then Ho( M ) satisfies h -descentif and only if it satisfies ´etale descent. Proof.
This follows from Corollary 3.3.34 and Theorem 3.3.37. (cid:3)
Corollary . Assume that
Ho( M ) is separated and has the localization and propertransversality properties. Let f : Y → X be a finite surjective morphism, with X normal, and G agroup acting on Y over X , such that the map Y /G → X is generically radicial (i.e. radicial overa dense open subscheme of X ). Consider at last a pullback square of the following shape. Y ′ / / f ′ (cid:15) (cid:15) Y f (cid:15) (cid:15) X ′ / / X Then, for any object M of Ho( M )( X ′ ) , the natural map M → ( R f ′∗ L f ′∗ ( M )) G is an isomorphism. Proof.
For any presheaf C of complexes of Q -vector spaces on S /X , one has an isomorphism R Γ qfh ( X ′ , C qfh ) ≃ R Γ qfh ( Y ′ , C qfh ) G . This follows from the fact that we have an isomorphism of qfh-sheaves of sets L ( Y ) /G ≃ L ( X )(the map Y → Y /G being generically flat, this is Proposition 3.3.19), which implies that the map L ( Y ′ ) /G → L ( X ′ ) is an isomorphism of qfh-sheaves (by the universality of colimits in topoi), andimplies this assertion (as in the proof of 3.3.22).By virtue of Theorem 3.3.37, Ho( M ) satisfies qfh-descent, so that the preceding computationsimply the result by corollaries 3.2.17 and 3.2.18. (cid:3) Corollary . Assume that
Ho( M ) is separated and has the localization and propertransversality properties. Then for any finite surjective morphism f : Y → X with X normal, themorphism M → R f ∗ L f ∗ ( M ) is a monomorphism and admits a functorial splitting in Ho( M )( X ) . Furthermore, this remainstrue after base change by any map X ′ → X . Proof.
It is sufficient to treat the case where X is connected. We may replace Y by anormalization of X in a suitable finite extension of its field of function, and assume that a finitegroup G acts on Y over X , so that the properties described in the preceding corollary are fulfilled(see 3.3.16). (cid:3) Remark . The condition (iv) of Theorem 3.3.37 can be reformulated in a more globalway as follows (this won’t be used in these notes, but this might be useful for the reader who mightwant to formulate all this in terms of (pre-)algebraic derivators [
Ayo07a , Def. 2.4.13]). Given aqfh-distinguished square (3.3.15.1) of group G , we can form a functor F from category I =(3.3.11.1) to the category of diagrams of S -schemes corresponding to the diagram of diagrams of S -schemes ( T , G ) ( h, G ) / / g (cid:15) (cid:15) ( Y , G ) Z in which T and Y correspond to T anf Y respectively, seen as functor from G to S /X . Theconstruction of 3.1.22 gives a diagram of X -schemes ( R F , I F ) which can be described explicitelyas follows. The category I F is the cofibred category over associated to the functor from tothe category of small categories defined by the diagram G G / / (cid:15) (cid:15) Ge in which e stands for the terminal category, and G for the category with one object associated to G . It has thus three objects a, b, c (see (3.3.11.1)), and the morphisms are determined byHom I F ( x, y ) = * if y = c ; ∅ if x = y and x = b, c ;G otherwise.The functor F sends a, b, c to T, Y, Z respectively, and simply encodes the fact that the diagram T h / / g (cid:15) (cid:15) YZ is G -equivariant, the action on Z being trivial. Now, by propositions 3.1.23 and 3.3.31, if ϕ : ( F , I F ) → ( X, ) denotes the canonical map, for any object M of Ho( M )( X ), the ob-ject R ϕ ∗ L ϕ ∗ ( M ) is the functor from = op to M ( X ) corresponding to the diagram below (ofcourse, this is well defined only in the homotopy category of the category of functors from to M ( X )). ( R f ∗ L f ∗ ( M )) G (cid:15) (cid:15) R i ∗ L i ∗ ( M ) / / ( R a ∗ L a ∗ ( M )) G As a consequence, if ψ : ( R F , I F ) → X denotes the structural map, the object R ψ ∗ L ψ ∗ ( M ) issimply the homotopy homotopy limit of the diagram of M ( X ) above, so that condition (iv) ofTheorem 3.3.37 can now be reformulated by saying that the map M → R ψ ∗ L ψ ∗ ( M )is an isomorphism, i.e. that the functor L ψ ∗ : Ho( M )( X ) → Ho( M )( R F , I F )is fully faithful. . CONSTRUCTIBLE MOTIVES 91
4. Constructible motives . Consider as in 2.0 a base scheme S and a sub-category S of the category of S -schemes.In section 4.4, and for the main theorem of section 4.2, we will assume:(a) Any scheme in S is quasi-excellent. Apart in Definition 4.3.2 and the subsequent proposition, where we will consider an abstractsituation, we will be concerned with the study of a fixed premotivic triangulated category T over S (recall Definition 2.4.45) such that:(b) T is motivic (see Definition 2.4.45).(c) T is τ -generated for a set of twists τ (see Paragraph 1.4.4) which is stable under negativeTate twists.(d) T is the homotopy category associated with a stable combinatorial Sm -fibred modelcategory M over S . As usual, the geometric section of T will be denoted by M .Unless explicitly referring to the underlying model category M , we will not indicate in thenotation of the six operations that the functors are derived functors. The aim of this subsection is to gather the results fromthe theory of resolution of singularities that will be used subsequently.4.1.1 . In [
EGA4 , IV, 7.8.2], Grothendieck defined the notion of an excellent ring . Matsumuraintroduced in [
Mat70 ] the weaker notion of a quasi-excellent ring A . Recall A is quasi-excellentif the following conditions hold:(1) A is noetherian.(2) For any prime ideal p , ˆ A p being the completion of A at p , the canonical morphism A → ˆ A p is regular (see 4.1.4 below).(3) For any A -algebra B of finite type, the regular locus of Spec ( B ) is open.Then a ring A is excellent if it is quasi-excellent and universally catenary. Following Gabber, wesay a scheme X is quasi-excellent (resp. excellent ) if it admits an open cover by affine schemeswhose ring is quasi-excellent (resp. excellent). Theorem . Let X be a quasi-excellent scheme, and Z ⊂ X a nowhere dense closed subscheme. Then there exists a finite h -cover { f i : Y i → X } i ∈ I such that for all i in I , f i is a morphism of finite type, the scheme Y i is regular, and f − i ( Z ) iseither empty or the support of a strict normal crossing divisor in Y i . See [
Ill08 ] for a sketch of proof. A complete argument can be found in [
ILO ]. Note that, if weare only interested in schemes of finite type over Spec ( R ), for R either a field, a complete discretevaluation ring, or a Dedekind domain whose field of functions is a global field, this is an immediateconsequence of de Jong’s resolution of singularities by alterations; see [ dJ96 ]. One can also deducethe case of schemes of finite type over an excellent noetherian scheme of dimension lesser or equalto 2 from [ dJ97 ]; see Theorem 4.1.10 and Corollary 4.1.11 below for a precise statement. Remark . This theorem will be used in the proof of Lemma 4.2.14 which is the key pointfor the proof of Theorem 4.2.16.4.1.4 . Recall that a morphism of rings u : A → B is regular if it is flat, and if, for anyprime ideal p in A , with residue field κ ( p ), the κ ( p )-algebra κ ( p ) ⊗ A B is geometrically regular(equivalently, this means that, for any prime ideal q of B , the A -algebra B q is formally smoothfor the q -adic topology). We recall the following great generalization of Neron’s desingularisationtheorem: Theorem . A morphism of noetherian rings u : A → B is regularif and only if B is a filtered colimit of smooth A -algebras of finite type. See Paragraph 4.1.1. The reader can safely restrict his attention to the more classical notion of an excellentscheme ([
EGA4 , IV, 7.8.5]). We use this assumption to use freely the descent theory described in section 3.3. Proof.
See [
Spi99 , theorems 1.1 and 1.2]. (cid:3) . Recall that an alteration is a proper surjective morphism p : X ′ → X which is genericallyfinite, i.e. such that there exists a dense open subscheme U ⊂ X over which p is finite. Definition . Let X be a noetherian scheme endowed with an action of a finitegroup G . A Galois alteration of the couple (
X, G ) is the data of a finite group G ′ , of a surjectivemorphism of groups G ′ → G , of an alteration X ′ → X , and of an action of G ′ on X ′ , such that:(i) the map X ′ → X is G ′ -equivariant;(ii) for any irreducible component T of X , there exists a unique irreducible component T ′ of X ′ over T , and the corresponding finite field extension k ( T ) G ⊂ k ( T ′ ) G ′ is purely inseparable.In practice, we shall keep the morphism of groups G ′ → G implicit, and we shall say that ( X ′ → X, G ′ ) is a Galois alteration of ( X, G ).Given a noetherian scheme X , a Galois alteration of X is a Galois alteration ( X ′ → X, G )of (
X, e ), where e denotes the trivial group. In this case, we shall say that X ′ → X is a Galoisalteration of X of group G . Remark . If p : X ′ → X is a Galois alteration of group G over X , then, if X and X ′ arenormal, irreducible and quasi-excellent, p can be factored as a radicial finite surjective morphism X ′′ → X , followed by a Galois alteration X ′ → X ′′ of group G , such that k ( X ′′ ) = k ( X ′ ) G (justdefine X ′′ as the normalization of X in k ( X ′ ) G ). In other words, up to a radicial finite surjectivemorphism, X is generically the quotient of X ′ under the action of G . Definition . A noetherian scheme S admits canonical dominant resolution of singu-larities up to quotient singularities if, for any Galois alteration S ′ → S of group G , and forany G -equivariant nowhere dense closed subscheme Z ′ ⊂ S ′ , there exists a Galois alteration( p : S ′′ → S ′ , G ′ ) of ( S ′ , G ), such that S ′′ is regular and projective over S , and such that theinverse image of Z ′ in S ′′ is contained in a G ′ -equivariant strict normal crossing divisor (i.e. astrict normal crossing divisor whose irreducible components are stable under the action of G ′ ).A noetherian scheme S admits canonical resolution of singularities up to quotient singularities if any integral closed subscheme of S admits canonical dominant resolution of singularities up toquotient singularities.A noetherian scheme S admits wide resolution of singularities up to quotient singularities if,for any separated S -scheme of finite type X , and any nowhere dense closed subscheme Z ⊂ X ,there exists a projective Galois alteration p : X ′ → X of group G , with X ′ regular, such that,in each connected component of X ′ , Z ′ = p − ( Z ) is either empty, either the support of a strictnormal crossing divisor. Theorem . If an excellent noetherian scheme of finite dimension S admitscanonical resolution of singularities up to quotient singularities, then any separated S -scheme offinite type admits canonical resolution of singularities up to quotient singularities. Proof.
Let X be an integral separated S -scheme of finite type. There exists a finite morphism S ′ → S , with S ′ integral, an integral dominant S ′ -scheme X ′ and a radicial extension X ′ → X over S , such that X ′ has a geometrically irreducible generic fiber over S ′ . It follows then from(the proof of) [ dJ97 , theorem 5.13] that X ′ admits canonical dominant resolution of singularitiesup to quotient singularities, which implies that X has the same property. (cid:3) Corollary . Let S be an excellent noetherian scheme of dimension lesseror equal to . Then any separated scheme of finite type over S admits canonical resolution ofsingularities up to quotient singularities. In particular, S admits wide resolution of singularitiesup to quotient singularities. Proof.
See [ dJ97 , corollary 5.15]. (cid:3) . CONSTRUCTIBLE MOTIVES 93
The aim of this section is to study the notion of τ -constructibilityin the triangulated motivic case and to study its stability properties under Grothendieck six op-erations. Recall the following particular case of Definition 1.4.9: Definition . For a scheme X in S , we denote by T c ( X ) the thick triangulated sub-category of T ( X ) generated by premotives of the form M X ( Y ) { i } for a smooth X -scheme Y anda twist i ∈ τ . We will say that a premotive in T c ( X ) is constructible . Remark . Let us mention that our main examples: • the stable homotopy category SH (cf. Example 1.4.3), • the category of Voevodsky motives DM (cf. Definition 11.1.1), • the category of Beilinson motives DM B (cf. Definition 14.2.1)are all generated by the Tate twists ( i.e. τ = Z ). Recall also Proposition 1.4.11: it applies to allthese examples so that constructible premotives coincides with compact objects. Proposition . If M and N are constructible in T ( X ) , so is M ⊗ X N . Proof.
For a fixed M , the full subcategory of T ( X ) spanned by objects such that M ⊗ X N is constructible is a thick triangulated subcategory of T ( X ). In the case M is of shape M X ( Y ) { n } for Y smooth over X and n ∈ τ , this proves that M ⊗ X N is constructible whenever N is. By thesame argument, using the symmetry of the tensor product, we get to the general case. (cid:3) Similarly, one has the following conservation property.
Proposition . For any morphism f : X → Y of schemes, the functor f ∗ : T ( Y ) → T ( X ) preserves constructible objects. If moreover f is smooth, the functor f ♯ : T ( X ) → T ( Y ) also preserves constructible objects. Corollary . The categories T c ( X ) form a thick triangulated monoidal Sm-fibred sub-category of T . Proposition . Let X a scheme, and X = S i ∈ I U i a cover of X by open subschemes.An object M of T ( X ) is constructible if and only if its restriction to U i is constructible in T ( U i ) for all i ∈ I . Proof.
This is a necessary condition by 4.2.4. For the converse, as X is noetherian, it issufficient to treat the case where I is finite. Proceeding by induction on the cardinal of I it issufficient to treat the case of a cover by two open subschemes X = U ∪ V . For an open immersion j : W → X , write M W = j ♯ j ∗ ( M ). If the restrictions of M to U and V are constructible, then sois its restriction to U ∩ V . According to Proposition 3.3.4, we get a distinguished triangle M U ∩ V → M U ⊕ M V → M → M U ∩ V [1]in which M W is constructible for W = U, V, U ∩ V (using 4.2.4 again). Thus the premotive M isconstructible. (cid:3) Corollary . For any scheme X and any vector bundle E over X , the functors T h ( E ) and T h ( − E ) preserve constructible objects in T ( X ) . Proof.
To prove that T h ( E ) and T h ( − E ) preserves constructible objects, by virtue of thepreceding proposition, we may assume that E is trivial of rank r . It is thus sufficient to prove that M ( r ) is constructible whenever M is so for any integer r . For we may assume that M = X { n } for some n ∈ τ (using 4.2.4), this is true by assumption on τ ; see 4.0(c). (cid:3) Notice however this fact is not true for ´etale motivic complexes. Corollary . Let f : X → Y a morphism of finite type. The property that the functor f ∗ : T ( X ) → T ( Y ) preserves constructible objects is local on Y with respect to the Zariski topology. Proof.
Consider a finite Zariski cover { v i : Y i → Y } i ∈ I , and write f i : X i → Y i for thepullback of f along v i for each i in I . Assume that the functors f i, ∗ preserves constructibleobjects; we shall prove that f ∗ has the same property. Let M be a constructible object in T ( X ).Then for i ∈ I , using the smooth base change isomorphism (for open immersions), we see that therestriction of f ∗ ( M ) to Y i is isomorphic to the image by f i, ∗ of the restriction of M to X i , henceis constructible. The preceding proposition thus implies that f ∗ ( M ) is constructible. (cid:3) Proposition . For any closed immersion i : Z → X , the functor i ∗ : T ( Z ) → T ( X ) preserves constructible objects. Proof.
It is sufficient to prove that for any smooth Z -scheme Y and any twist n ∈ τ , thepremotive i ∗ ( M Z ( Y ) { n } ) is constructible in T ( X ). According to the Mayer-Vietoris triangle (seeRemark 3.3.6), this assertion is local in X . Thus we can assume there exists a smooth X -scheme Y such that Y = Y × X Z (apply [ EGA4 , 18.1.1]). Put U = X − Z and let j : U → X be theobvious open immersion. From the localization property, we get a distinguished triangle M X ( Y × X U ) { n } → M X ( Y ) { n } → i ∗ ( M Z ( Y ) { n } ) → M X ( Y × X U ) { n } [1]and this concludes. (cid:3) Corollary . Let i : Z → X be a closed immersion with open complement j : U → X .an object M of T ( X ) is constructible if and only if j ∗ ( M ) and i ∗ ( M ) are constructible in T ( U ) and T ( Z ) respectively. Proof.
We have a distinguished triangle j ♯ j ∗ ( M ) → M → i ∗ i ∗ ( M ) → j ♯ j ∗ ( M )[1] . Hence this assertion follows from propositions 4.2.4 and 4.2.9. (cid:3)
Proposition . If f : X → Y is proper, then the functor f ∗ : T ( X ) → T ( Y ) preserves constructible objects. Proof.
We shall first consider the case where f is projective. As this property is local on Y (Corollary 4.2.8), we may assume that f factors as a closed immersion i : X → P nY followedby the canonical projection p : P nY → Y . By virtue of Proposition 4.2.9, we can assume that f = p . In this case, the functor p ∗ is isomorphic to p ♯ composed with the quasi-inverse of theThom endofunctor associated to the cotangent bundle of p ; see 2.4.50 (3). Therefore, the functor p ∗ preserves constructible objects by virtue of Proposition 4.2.4 and of Corollary 4.2.7. The casewhere f is proper follows easily from the projective case, using Chow’s lemma and cdh-descent(the homotopy pullback squares (3.3.9.1)), by induction on the dimension of X . (cid:3) Corollary . If f : X → Y is separated of finite type, then the functor f ! : T ( X ) → T ( Y ) preserves constructible objects. Proof.
It is sufficient to treat the case where f is either an open immersion, either a propermorphism, which follows respectively from 4.2.4 and 4.2.11. (cid:3) Proposition . Let X be a scheme. The category of constructible objects in T ( X ) isthe smallest thick triangulated subcategory which contains the objects of shape f ∗ ( X ′ { n } ) , where f : X ′ → X is a (strictly) projective morphism, and n ∈ τ . . CONSTRUCTIBLE MOTIVES 95 Proof.
See [
Ayo07a , lemma 2.2.23]. (cid:3)
The following lemma is the key geometrical point for the finiteness Theorem 4.2.16
Lemma . Let j : U → X be a dense open immersion such that X is quasi-excellent.Then, there exists the following data:(i) a finite h -cover { f i : Y i → X } i ∈ I such that for all i in I , f i is a morphism of finite type,the scheme Y i is regular, and f − i ( U ) is either Y i itself or the complement of a strictnormal crossing divisor in Y i ; we shall write f : Y = a i ∈ I Y i → X for the induced global h -cover;(ii) a commutative diagram X ′′′ g / / q (cid:15) (cid:15) Y f (cid:15) (cid:15) X ′′ u / / X ′ p / / X (4.2.14.1) in which: p is a proper birational morphism, X ′ is normal, u is a Nisnevich cover, and q is a finite surjective morphism.Let T (resp. T ′ ) be a closed subscheme of X (resp. X ′ ) and assume that for any irreduciblecomponent T of T , the following inequality is satisfied: (4.2.14.2) codim X ′ ( T ′ ) ≥ codim X ( T ) , Then, possibly after shrinking X in an open neighbourhood of the generic points of T in X , onecan replace X ′′ by an open cover and X ′′′ by its pullback along this cover, in such a way that wehave in addition the following properties:(iii) p ( T ′ ) ⊂ T and the induced map T ′ → T is finite and pseudo-dominant; (iv) if we write T ′′ = u − ( T ′ ) , the induced map T ′′ → T ′ is an isomorphism. Proof.
The existence of f : Y → X as in (i) follows from Gabber’s weak uniformisationtheorem (see 4.1.2), while the commutative diagram (4.2.14.1) satisfying property (ii) is ensuredby Lemma 3.3.28.Consider moreover closed subschemes T ⊂ X and T ′ ⊂ X ′ satisfying (4.2.14.2).We first show that, by shrinking X in an open neighbourhood of the generic points of T and by replacing the diagram (4.2.14.1) by its pullback over this neighbourhood, we can assumethat condition (iii) is satisfied. Note that shrinking X in this way does not change the condition(4.2.14.2) because codim X ( T ) does not change and codim X ′ ( T ′ ) can only increase. Note first that, by shrinking X , we can assume that any irreducible component T ′ of T ′ dominates an irreducible component T of T . In fact, given an irreducible component T ′ whichdoes not satisfies this condition, p ( T ′ ) is a closed subscheme of X disjoint from the set of genericpoints of T and replacing X by X − f ( T ′ ), we can throw out T ′ .Further, shrinking X again, we can assume that for any pair ( T ′ , T ) as in the precedingparagraph, p ( T ′ ) ⊂ T . In fact, in any case, as p ( T ′ ) is closed we get that T ⊂ p ( T ′ ). Let Z bethe closure of p ( T ′ ) − T in X . Then Z does not contain any generic point of T (because p ( T ′ ) isirreducible), and p ( T ′ ) ∩ ( X − Z ) ⊂ T . Thus it is sufficient to replace X by X − Z to ensure thisassumption. Recall from 8.1.3 that this means that any irreducible component of T ′ dominates an irreducible componentof T . Remember that for any scheme X , codim X ( ∅ ) = + ∞ . Consider again a pair ( T ′ , T ) as in the two preceding paragraphs and the induced commutativesquare: T ′ / / p (cid:15) (cid:15) X ′ p (cid:15) (cid:15) T / / X (4.2.14.3)We show that the map p is generically finite. In fact, this will conclude the first step, becauseif it is true for any irreducible component T ′ of T ′ , we can shrink X again so that the dominantmorphism p : T ′ → T becomes finite.Denote by c ′ (resp. c ) the codimension of T in X (resp. T ′ in X ′ ). Note that (4.2.14.2) givesthe inequality c ′ ≥ c . Let t be the generic point of T , Ω the localization of X at t and considerthe pullback of (4.2.14.3): W ′ / / q (cid:15) (cid:15) Ω ′ q (cid:15) (cid:15) { t } / / Ω . (4.2.14.4)We have to prove that dim( W ′ ) = 0. Consider an irreducible component Ω ′ of Ω ′ containing W ′ .As q is still proper birational, Ω ′ corresponds to a unique irreducible component Ω of Ω such that q induces a proper birational map Ω ′ → Ω . According to [ EGA4 , 5.6.6], we get the inequalitydim(Ω ′ ) ≤ dim(Ω ) . Thus, we obtain the following inequalities:dim( W ′ ) ≤ dim(Ω ′ ) − codim Ω ′ ( W ′ ) ≤ dim(Ω ) − codim Ω ′ ( W ′ ) ≤ dim(Ω) − codim Ω ′ ( W ′ ) . As this is true for any irreducible component Ω ′ of Ω ′ , we finally obtain:dim( W ′ ) ≤ dim(Ω) − codim Ω ′ ( W ′ ) ≤ c − c ′ and this concludes the first step.Keeping T ′ and T as above, as the map from T ′′ to T ′ is a Nisnevich cover, it is a splitepimorphism in a neighbourhood of the generic points of T ′ in X ′ . Hence, as the map X ′ → X is proper and birational, we can find a neighbourhood of the generic points of T in X over whichthe map T ′′ → T ′ admits a section s : T ′ → T ′′ . Let S be a closed subset of X ′′ such that T ′′ = s ( T ′ ) ∐ S (which exists because X ′′ → X ′ is ´etale). The map ( X ′′ − T ′′ ) ∐ ( X ′′ − S ) → X ′ is then a Nisnevich cover. Replacing X ′′ by ( X ′′ − T ′′ ) ∐ ( X ′′ − S ) (and X ′′′ by the pullback of X ′′′ → X ′′ along ( X ′′ − T ′′ ) ∐ ( X ′′ − S ) → X ′ ), we may assume that the induced map T ′′ → T ′ isan isomorphism, without modifying further the data f , p , T and T ′ . This gives property (iv) andends the proof the lemma. (cid:3) . Let T be a full Open -fibred subcategory of T (where Open stands for the class ofopen immersions). We assume that T has the following properties.(a) for any scheme X in S , T ( X ) is a thick subcategory of T ( X ) which contains theobjects of the form X { n } , n ∈ τ ;(b) for any separated morphism of finite type f : X → Y in S , T is stable under f ! ;(c) for any dense open immersion j : U → X , with X regular, which is the complement of astrict normal crossing divisor, j ∗ ( U { n } ) is in T ( U ) for any n ∈ τ .Properties (a) and (b) have the following consequences: any constructible object belongs to T ;given a closed immersion i : Z → X with complement open immersion j : U → X , an object M of T ( X ) belongs to T ( X ) if and only if j ∗ ( M ) and i ∗ ( M ) belongs to T ( U ) and T ( Z ) respectively;for any scheme X in S , the condition that an object of T ( X ) belongs to T ( X ) is local on X forthe Zariski topology. . CONSTRUCTIBLE MOTIVES 97 Theorem . Consider the above hypothesis and assume that T is Q -linear and separated.Let Y be a quasi-excellent scheme and f : X → Y be a morphism of finite type.Then for any constructible object M of T ( X ) , the object f ∗ ( M ) belongs to T ( Y ) . Proof.
It is sufficient to prove that, for any dense open immersion j : U → X , and for any n ∈ τ , the object j ∗ ( U { n } ) is in T . Indeed, assume this is known. We want to prove that f ∗ ( M ) is in T ( Y ) whenever M is constructible. We deduce from property (b) of 4.2.15 and fromProposition 4.2.13 that it is sufficient to consider the case where M = X { n } , with n ∈ τ . Then,as this property is assumed to be known for dense open immersions, by an easy Mayer-Vietorisargument, we see that the condition that f ∗ ( X { n } ) belongs to T is local on X with respect tothe Zariski topology. Therefore, we may assume that f is separated. Consider a compactificationof f , i.e. a commutative diagram Y j / / f (cid:15) (cid:15) ¯ Y ¯ f (cid:127) (cid:127) ⑦⑦⑦⑦⑦⑦⑦⑦ X with j a dense open immersion, and ¯ f proper. By property (b) of 4.2.15, we may assume that f = j is a dense open immersion.Let j : U → X be a dense open immersion. We shall prove by induction on the dimensionof X that , for any n ∈ τ , the object j ∗ ( U { n } ) is in T . The case where X is of dimension ≤ j is then an isomorphism, which implies that j ♯ ≃ j ∗ , and allows toconclude (because T is Open -fibred).Assume that dim
X >
0. Following an argument used by Gabber [
Ill07 ] in the context of ℓ -adic sheaves, we shall prove by induction on c ≥ T ⊂ X ofcodimension > c such that, for any n ∈ τ , the restriction of j ∗ ( U { n } ) to X − T is in T ( X − T ).As X is of finite dimension, this will obviously prove Theorem 4.2.16.The case where c = 0 is clear: we can choose T such that X − T = U . If c >
0, we choose aclosed subscheme T of X , of codimension > c −
1, such that the restriction of j ∗ ( U { n } ) to X − T is in T . It is then sufficient to find a dense open subscheme V of X , which contains all the genericpoints of T , and such that the restriction of j ∗ ( U { n } ) to V is in T : for such a V , we shall obtainthat the restriction of j ∗ ( U { n } ) to V ∪ ( X − T ) is in T , the complement of V ∪ ( X − T ) beingthe support of a closed subscheme of codimension > c in X . In particular, using the smooth basechange isomorphism (for open immersions), we can always replace X by a generic neighbourhoodof T . It is sufficient to prove that, possibly after shrinking X as above, the pullback of j ∗ ( U { n } )along T → X is in T (as we already know that its restriction to X − T is in T ).We may assume that T is purely of codimension c . We may assume that we have data as inpoints (i) and (ii) of Lemma 4.2.14. We let j ′ : U ′ → X ′ denote the pullback of j along p : X ′ → X .Then, we can find, by induction on c , a closed subscheme T ′ in X ′ , of codimension > c −
1, suchthat the restriction of j ′∗ ( U ′ { n } ) to X ′ − T ′ is in T . By shrinking X , we may assume thatconditions (iii) and (iv) of Lemma 4.2.14 are fulfilled as well.For an X -scheme w : W → X and a closed subscheme Z ⊂ W , we shall write ϕ ( W, Z ) = w ∗ i ∗ i ∗ j W, ∗ j ∗ W ( W { n } ) , where i : Z → W denotes the inclusion, and j W : W U → W stands for the pullback of j along w .This construction is functorial with respect to morphisms of pairs of X -schemes: if W ′ → W is amorphism of X -schemes, with Z ′ and Z two closed subschemes of W ′ and W respectively, suchthat Z ′ is sent to Z , then we get a natural map ϕ ( W, Z ) → ϕ ( W ′ , Z ′ ). Remember that we wantto prove that ϕ ( X, T ) is in T . This will be done via the following lemmas (which hold assumingall the conditions stated in Lemma 4.2.14 as well as our inductive assumptions). Lemma . The cone of the map ϕ ( X, T ) → ϕ ( X ′ , T ′ ) is in T . The map ϕ ( X, T ) → ϕ ( X ′ , T ′ ) factors as ϕ ( X, T ) → ϕ ( X ′ , p − ( T )) → ϕ ( X ′ , T ′ ) . By the octahedral axiom, it is sufficient to prove that each of these two maps has a cone in T .We shall prove first that the cone of the map ϕ ( X ′ , p − ( T )) → ϕ ( X ′ , T ′ ) is in T . Given animmersion a : S → X ′ , we shall write M S = a ! a ∗ ( M ) . We then have distinguished triangles M p − ( T ) − T ′ → M p − ( T ) → M T ′ → M p − ( T ) − T ′ [1] . For M = j ′∗ ( U ′ { n } ) (recall j ′ is the pullback of j along p ) the image of this triangle by p ∗ givesa distinguished triangle p ∗ ( M p − ( T ) − T ′ ) → ϕ ( X ′ , p − ( T )) → ϕ ( X ′ , T ′ ) → p ∗ ( M p − ( T ) − T ′ )[1] . As the restriction of M = j ′∗ ( U ′ { n } ) to X ′ − T ′ is in T by assumption on T ′ , the object M p − ( T ) − T ′ is in T as well (by property (b) of 4.2.15 and because T is Open -fibred), from whichwe deduce that p ∗ ( M p − ( T ) − T ′ ) is in T (using the condition (iii) of Lemma 4.2.14 and the property(b) of 4.2.15).Let V be a dense open subscheme of X such that p − ( V ) → V is an isomorphism. Wemay assume that V ⊂ U , and write i : Z → U for the complement closed immersion. Let p U : U ′ = p − ( U ) → U be the pullback of p along j , and let ¯ Z be the reduced closure of Z in X .We thus get the commutative squares of immersions below, Z k / / i (cid:15) (cid:15) ¯ Z l (cid:15) (cid:15) U j / / X and Z ′ k ′ / / i ′ (cid:15) (cid:15) ¯ Z ′ l ′ (cid:15) (cid:15) U ′ j ′ / / X ′ where the square on the right is obtained from the one on the left by pulling back along p : X ′ → X .As p is an isomorphism over V , we get by cdh-descent (Proposition 3.3.10) the homotopy pullbacksquare below. U { n } / / (cid:15) (cid:15) p U, ∗ ( U ′ { n } ) (cid:15) (cid:15) i ∗ i ∗ ( Z { n } ) / / i ∗ i ∗ p U, ∗ ( U ′ { n } )If a : T → X denotes the inclusion, applying the functor a ∗ a ∗ j ∗ to the commutative square above,we see from the proper base change formula and from the identification j ∗ i ∗ ≃ l ∗ k ∗ that we geta commutative square isomorphic to the following one ϕ ( X, T ) / / (cid:15) (cid:15) ϕ ( X ′ , p − ( T )) (cid:15) (cid:15) ϕ ( ¯ Z, ¯ Z ∩ T ) / / ϕ ( ¯ Z ′ , p − ( ¯ Z ∩ T )) , which is thus homotopy cartesian as well. It is sufficient to prove that the two objects ϕ ( ¯ Z, ¯ Z ∩ T )and ϕ ( ¯ Z ′ , p − ( ¯ Z ∩ T )) are in T . It follows from the proper base change formula that the object ϕ ( ¯ Z, ¯ Z ∩ T ) is canonically isomorphic to the restriction to T of l ∗ k ∗ ( Z { n } ). As dim ¯ Z < dim X ,we know that the object k ∗ ( Z { n } ) is in T . By property (b) of 4.2.15, we obtain that ϕ ( ¯ Z, ¯ Z ∩ T )is in T . Similarly, the object ϕ ( ¯ Z ′ , p − ( ¯ Z ∩ T )) is canonically isomorphic to the restriction of p ∗ l ′∗ k ′∗ ( Z ′ { n } ) to T , and, as dim ¯ Z ′ < dim X ′ (because, p being an isomorphism over the denseopen subscheme V of X , ¯ Z ′ does not contain any generic point of X ′ ), k ′∗ ( Z ′ { n } ) is in T . Wededuce again from property (b) of 4.2.15 that ϕ ( ¯ Z ′ , p − ( ¯ Z ∩ T )) is in T as well, which achievesthe proof of the lemma. Lemma . The map ϕ ( X ′ , T ′ ) → ϕ ( X ′′ , T ′′ ) is an isomorphism in T ( X ) . . CONSTRUCTIBLE MOTIVES 99 Condition (iv) of Lemma 4.2.14 can be reformulated by saying that we have the Nisnevich distin-guished square below. X ′′ − T ′′ / / (cid:15) (cid:15) X ′′ v (cid:15) (cid:15) X ′ − T ′ / / X ′ This lemma follows then by Nisnevich excision (Proposition 3.3.4) and smooth base change (for´etale maps).
Lemma . Let T ′′′ be the pullback of T ′′ along the finite surjective morphism X ′′′ → X ′′ .The map ϕ ( X ′′ , T ′′ ) → ϕ ( X ′′′ , T ′′′ ) is a split monomorphism in T ( X ) . We have the following pullback squares T ′′′ t / / r (cid:15) (cid:15) X ′′′ q (cid:15) (cid:15) U ′′′ j ′′′ o o q U (cid:15) (cid:15) T ′′ s / / X ′′ U ′ j ′′ o o in which j ′′ and j ′′′ denote the pullback of j along pu and puq respectively, while s and t are theinclusions. By the proper base change formula applied to the left hand square, we see that themap ϕ ( X ′′ , T ′′ ) → ϕ ( X ′′′ , T ′′′ ) is isomorphic to the image of the map j ′′∗ ( U ′′ { n } ) → q ∗ q ∗ j ′′∗ ( U ′′ { n } ) → q ∗ j ′′′∗ ( U ′′′ { n } ) . by f ∗ s ∗ , where f : T ′′ → T is the map induced by p (note that f is proper as T ′′ ≃ T ′ byassumption). As q ∗ j ′′′∗ ≃ j ′′∗ q U, ∗ , we are thus reduced to prove that the unit map U ′′ { n } → q U, ∗ ( U ′′′ { n } )is a split monomorphism. As X ′′ is normal (because X ′ is so by assumption, while X ′′ → X ′ is´etale), this follows immediately from Corollary 3.3.40.Now, we can finish the proof of Theorem 4.2.16. Consider the Verdier quotient D = T ( X ) / T ( X ) . We want to prove that, under the conditions stated in Lemma 4.2.14, we have ϕ ( X, T ) ≃ D .Let π : T ′′′ → T be the map induced by puq : X ′′′ → X . If a : T ′′′ → Y denotes the map inducedby g : X ′′′ → Y , and j Y : Y U → Y the pullback of j by f , we have the commutative diagrambelow. ϕ ( X, T ) / / ( ( PPPPPPPPPPPP ϕ ( X ′′′ , T ′′′ ) π ∗ a ∗ j Y, ∗ ( Y U { n } ) ♠♠♠♠♠♠♠♠♠♠♠♠♠ By virtue of lemmas 4.2.17, 4.2.19, and 4.2.18, the horizontal map is a split monomorphism in D .It is thus sufficient to prove that this map vanishes in D , for which it will be sufficient to provethat π ∗ a ∗ j Y, ∗ ( Y U { n } ) is in T . The morphism π is finite (by construction, the map T ′′ → T ′ is an isomorphism, while the maps T ′′′ → T ′′ and T ′ → T are finite). Under this condition, T is stable under the operations π ∗ and a ∗ . To finish the proof of the theorem, it remains to checkthat j Y, ∗ ( Y U { n } ) is in T , which follows from property (c) of 4.2.15 (and additivity). (cid:3) Definition . We shall say that T is τ -compatible if it satisfies the following two con-ditions.(a) For any closed immersion i : Z → X between regular schemes in S , the image of X { n } , n ∈ τ , by the exceptional inverse image functor i ! : T ( X ) → T ( Z ) is constructible.(b) For any scheme X , any n ∈ τ , and any constructible object M in T ( X ), the objectHom X ( X { n } , M ) is constructible.As usual, when τ is the monoid generated by the Tate twist, we say compatible with Tate twists . Remark . Condition (b) of the definition above will come essentially for free if theobjects X { n } are ⊗ -invertible with constructible ⊗ -quasi-inverse (which will hold in practice,essentially by definition). Example . In pratice, condition (a) of the definition above will be a consequence of the absolute purity theorem . In particular, the category of Beilinson motives DM B is compatible withTate twist as a corollary of the fact the Tate twist is invertible and Theorem 14.4.1. Lemma . Assume that T is τ -compatible. Let i : Z → X a closed immersion, with X regular, and Z the support of a strict normal crossing divisor. Then i ! ( X { n } ) is constructiblefor any n ∈ τ . As a consequence, if j : U → X denotes the complement open immersion, then j ∗ ( U { n } ) is constructible for any n ∈ τ . Proof.
The first assertion follows easily by induction on the number of irreducible compo-nents of Z , using Proposition 4.2.6. The second assertion follows from the distinguished triangles i ∗ i ! ( M ) → M → j ∗ j ∗ ( M ) → i ∗ i ! ( M )[1]and from Lemma 4.2.9. (cid:3) Theorem . Assume that T is Q -linear, separated, and τ -compatible.Then, for any morphism of finite type f : X → Y such that Y is quasi-excellent, the functor f ∗ : T ( X ) → T ( Y ) preserves constructible objects. Proof.
By virtue of propositions 4.2.4 and 4.2.11 as well as of Lemma 4.2.23, if T is τ -compatible, we can apply Theorem 4.2.16, where T stands for the subcategory of constructibleobjects. (cid:3) Corollary . Under the assumptions of the above theorem, for any quasi-excellentscheme X , and for any couple of constructible objects M and N in T ( X ) , the object Hom X ( M, N ) is constructible. Proof.
It is sufficient to treat the case where M = f ♯ ( Y { n } ), for n ∈ τ and f : Y → X asmooth morphism. But then, we have, by transposition of the Sm -projection formula, a naturalisomorphism: Hom X ( M, N ) ≃ f ∗ Hom ( Y { n } , f ∗ ( N )) . This corollary follows then immediately from Proposition 4.2.4 and from Theorem 4.2.24. (cid:3)
Corollary . Under the assumptions of the above theorem, for any closed immersion i : Z → X such that X is quasi-excellent, the functor i ! : T ( X ) → T ( Z ) preserves constructible objects. Proof.
Let j : U → X be the complement open immersion. For an object M of T ( X ), wehave the following distinguished triangle. i ∗ i ! ( M ) → M → j ∗ j ∗ ( M ) → i ∗ i ! ( M )[1] . By virtue of Proposition 4.2.6 and Theorem 4.2.24, if M is constructible, then j ∗ j ∗ ( M ) have thesame property, which allows to conclude. (cid:3) Lemma . Let f : X → Y be a separated morphism of finite type. The condition that thefunctor f ! preserves constructible objects in T is local over X and over Y for the Zariski topology. Proof. If u : X ′ → X is a Zariski cover, then we have, by definition, u ∗ = u ! , so that, byProposition 4.2.6, the condition that f ! preserves τ -constructibility is equivalent to the condition . CONSTRUCTIBLE MOTIVES 101 that u ∗ f ! ≃ ( f u ) ! preserves τ -constructibility. Let v : Y ′ → Y be a Zariski cover, and considerthe following pullback square. X ′ u / / g (cid:15) (cid:15) X f (cid:15) (cid:15) Y ′ v / / Y We then have a natural isomorphism u ∗ f ! ≃ g ! v ∗ , and, as u is still a Zariski cover, we deduceagain from Proposition 4.2.6 that, if g ! preserves τ -constructibility, so does f ! . (cid:3) Corollary . Under the assuptions of the above theorem, for any separated morphismof finite type f : X → Y , the functor f ! : T ( Y ) → T ( X ) preserves constructible objects. Proof.
By virtue of the preceding lemma, we may assume that f is affine. We can thenfactor f as an immersion i : X → A nY followed by the canonical projection p : A nY → Y . The caseof an immersion is reduced to the case of an open immersion (4.2.4) and to the case of a closedimmersion (4.2.26). Thus we may assume that f = p , in which case p ! ≃ p ∗ ( − )( n )[2 n ] (accordingto point (3) of Theorem 2.4.50), so that we conclude by 4.2.4 and 4.2.9. (cid:3) In conclusion, we have proved the following finiteness theorem:
Theorem . Assume the motivic triangulated category T is Q -linear, separated and τ -compatible. Then constructible objects of T are closed under the six operations of Grothendieck whenrestricted to the subcategory S ′ of S made of quasi-excellent schemes and morphisms of finitetype. In particular, T c is a motivic category over S ′ . . For the next definition, we consider an admissible class P of morphisms in S and anabstract symmetric monoidal P -fibred model category M over S .Let ( S α ) α ∈ A be a projective system of schemes in S , with affine dominant transition maps,and such that S = lim ←− α ∈ A S α is representable in S (we assume that A is a partially ordered set tokeep the notations simple). For each index α , we denote by p α : S → S α the canonical projection.Given an index α ∈ A and an object E α in Ho( M )( S α ), we write E α for the pullback of E α along the map S α → S α , and put E = L p ∗ α ( E α ). Definition . Consider the assumptions above and let τ be a set of twists of Ho( M ).We say that Ho( M ) is τ -continuous if given any projective system of schemes { S α } as above,for any index α , any object E α in Ho( M )( S α ), and any twist n ∈ τ , the canonical maplim −→ α ≥ α Hom
Ho( M )( S α ) ( S α { n } , E α ) → Hom
Ho( M )( S ) ( S { n } , E ) , is bijective.We will simply say continuous instead of τ -continuous in one of the following situations: • τ = ∅ ; • Ho( M ) is a motivic triangulated category and τ = Z is the group generated by the Tatetwist. Example . The main examples of τ -continuous categories will be seen afterwards: • the A -derived category D A , Λ (Example 6.1.13); • the category of motivic complexes DM Λ , and its effective counterpart DM eff Λ (Theorem11.1.24); • the motivic category DM B of Beilinson motives (Proposition 14.3.1). Remember also that T is associated with a combinatorial stable premotivic model category. The interest of this property is to allow a description of constructible objects over S in termsof constructible objects over the S α ’s. Proposition . Assume, under the hypothesis of 4.3.1, that
Ho( M ) is τ -continuous.Consider a scheme S in S , as well as a projective system of schemes { S α } in S with affinetransition maps and such that S = lim ←− α S α .Then, for any index α , and for any objects C α and E α in Ho( M )( S α ) , if C α is con-structible, then the canonical map (4.3.4.1) lim −→ α ≥ α Hom
Ho( M )( S α ) ( C α , E α ) → Hom
Ho( M )( S ) ( C, E ) is bijective. Moreover, the canonical functor (4.3.4.2) 2 - lim −→ α Ho( M ) c ( S α ) → Ho( M ) c ( S ) is an equivalence of monoidal triangulated categories. Proof.
To prove the first assertion, we may assume, without loss of generality, that C α = M S α ( X α ) { n } for some some smooth S α -scheme of finite type X α , and n ∈ τ . Consider anobject E α in Ho( M )( S α ). For α ≥ α , write X α (resp. E α ) for the pullback of X α (resp. of E α ) along the map S α → S α . Similarly, write X (resp. E ) for the pullback of X α (resp. of E α ) along the map S → S α . We shall also write E ′ α (resp. E ′ ) for the pullback of E α (resp. E )along the smooth map X α → S α (resp. X → S ). Then, { X α } is a projective system of schemesin S , with affine transition maps, such that X = lim ←− α X α . Therefore, by continuity, we have thefollowing natural isomorphism, which proves the first assertion.lim −→ α Hom
Ho( M )( S α ) ( M S α ( X α ) { n } , E α ) ≃ lim −→ α Hom
Ho( M )( X α ) ( X α { n } , E ′ α ) ≃ Hom
Ho( M )( X ) ( X { n } , E ′ ) ≃ Hom
Ho( M )( S ) ( M S ( X ) { n } , E )Note that the first assertion implies that the functor (4.3.4.2) is fully faithful. Note that pseudo-abelian triangulated categories are stable by filtered 2-colimits. In particular, the source of thefunctor (4.3.4.2) can be seen as a thick subcategory of Ho( M )( S ). The essential surjectivity of(4.3.4.2) follows from the fact that, for any smooth S -scheme of finite type X , there exists someindex α , and some smooth S α -scheme X α , such that X ≃ S × S α X α ; see [ EGA4 , 8.8.2 and17.7.8]: this implies that the essential image of the fully faithful functor (4.3.4.2) contains all theobjects of shape M S ( X ) { n } for n ∈ τ and X smooth over S , so that it contains Ho( M ) c ( S ), bydefinition. (cid:3) . Before showing how the assumption of continuity can be used in the case of motiviccategories, we state a proposition which later on will allow us to show continuity in concrete cases.We consider again the assumptions and notations of 4.3.1 assuming the transition maps ofthe pro-scheme ( S α ) are P -morphisms. For each index α ∈ A , we choose a small set I α (resp. J α ) of generating cofibrations (resp. of generating trivial cofibration) in Ho( M )( S α ). We alsochoose a small set I (resp. J ) of generating cofibrations (resp. of generating trivial cofibration) inHo( M )( S ).Consider the following assumptions:(a) We have I ⊂ S α ∈ A p ∗ α ( I α ) and J ⊂ S α ∈ A p ∗ α ( J α ).(b) For any index α , if C α and E α are two objects of M ( S α ), with C α either a sourceor a target of a map in I α ∪ J α , the natural maplim −→ α ∈ A Hom M ( S α ) ( C α , E α ) → Hom M ( S ) ( C, E )is bijective. . CONSTRUCTIBLE MOTIVES 103
Proposition . Under the assumptions of 4.3.5, for any index α ∈ A , the pullbackfunctor p ∗ α : M ( S α ) → M ( S ) preserves fibrations and trivial fibrations. Moreover, given anindex α ∈ A , as well as two objects C α and E α in M ( S α ) , if C α belongs to smallest fullsubcategory of Ho( M )( S α ) which is closed under finite homotopy colimits and which contains thesource and targets of I α , then, the canonical map lim −→ α ∈ A Hom
Ho( M )( S α ) ( C α , E α ) → Hom
Ho( M )( S ) ( C, E ) is bijective. Proof.
We shall prove first that, for any index α ∈ A , the pullback functor p ∗ α preservesfibrations and trivial fibrations. By assumption, for any α ≥ α , the pullback functor along the P -morphism S α → S α is both a left Quillen functor and a right Quillen functor. Let E α → F α be a trivial fibration (resp. a fibration) of M ( S α ). Let i : C → D a generating cofibration (resp.a generating trivial cofibration) in M ( S ). By condition (a) of 4.3.5, we may assume that thereexists α ∈ A , a cofibration (resp. a trivial cofibration) i α : C α → D α , such that i = p ∗ α ( i α ).We want to prove that the mapHom( D, E ) → Hom(
C, E ) × Hom(
C,F ) Hom(
D, F )is surjective. But, by condition (b) of 4.3.5, this map is isomorphic to the filtered colimit of thesurjective maps Hom( D α , E α ) → Hom( C α , E α ) × Hom( C α ,F α ) Hom( D α , F α )with α ≥ sup( α , α ), which proves the first assertion.To prove the second assertion, we may assume that C α is cofibrant and that E α if fibrant.The set of maps from a cofibrant object to a fibrant object in the homotopy category of a modelcategory can be described as homotopy classes of maps. Therefore, using the fact that p ∗ α preservescofibrations and fibrations, as well as the trivial ones, we see it is sufficient to prove that the maplim −→ α ∈ A Hom M ( S α ) ( C α , E α ) → Hom M ( S ) ( C, E )is bijective for some nice cofibrant replacement of C α . But the assumptions on C α imply thatit is weakly equivalent to an object C ′ α such that the map ∅ → C ′ α belongs to the smallestclass of maps in M ( S α ), which contains I α , and which is closed under pushouts and (finite)compositions. We may thus assume that C α = C ′ α . In that case, C α is in particular containedin the smallest full subcategory of M ( S α ) which is stable by finite colimits and which containsthe source and targets of I α . As filtered colimits commute with finite limits in the category ofsets, we conclude by using again condition (a) of 4.3.5. (cid:3) We now go back to the situation of a motivic triangulated category T satisfying our generalassumptions 4.0 Lemma . Let a : X → Y be a morphism in S . Assume that X = lim ←− α X α , where { X α } is a projective system of smooth affine Y -schemes. If T is τ -continuous, then, for any objects E and F in T ( Y ) , with E constructible, there is a canonical isomorphism a ∗ Hom Y ( E, F ) ≃ Hom X ( a ∗ ( E ) , a ∗ ( F )) . Proof.
We have a ∗ Hom X ( a ∗ ( E ) , a ∗ ( F )) ≃ Hom Y ( E, a ∗ a ∗ ( F )) , so that the map F → a ∗ a ∗ ( F )) induces a map Hom Y ( E, F ) → a ∗ Hom X ( a ∗ ( E ) , a ∗ ( F )) , hence, by adjunction, a map a ∗ Hom Y ( E, F ) → Hom X ( a ∗ ( E ) , a ∗ ( F )) . We already know that the later is an isomorphism whenever a is smooth. Let us write a α : X α → Y for the structural maps. Let C be a constructible object in T ( X ).By Proposition 4.3.4, we may assume that there exists an index α , and a constructible object C α in T ( X α ), such that, if we write C α for the pullback of C α along the map X α → X α for α ≥ α , we have isomorphisms:Hom( C, a ∗ Hom Y ( E, F )) ≃ lim −→ α Hom( C α , a ∗ α Hom Y ( E, F )) ≃ lim −→ α Hom( C α , Hom X ( a ∗ α ( E ) , a ∗ α ( F ))) ≃ lim −→ α Hom( C α ⊗ X α a ∗ α ( E ) , a ∗ α ( F )) ≃ Hom( C ⊗ X a ∗ ( E ) , a ∗ ( F )) ≃ Hom( C, Hom X ( a ∗ ( E ) , a ∗ ( F ))) . As constructible objects generate T ( X ), this proves the lemma. (cid:3) . Let X be a scheme in S . Assume that, for any point x of X , the correspondingmorphism i x : Spec (cid:0) O hX,x (cid:1) → X is in S (where O hX,x denotes the henselisation of O X,x ). Considerat last a scheme of finite type Y over X , and write a x : Y x = Spec (cid:0) O hX,x (cid:1) × X Y → Y for the morphism obtained by pullback. Finally, for an object E of T ( Y ), let us write E x = a ∗ x ( E ) . Proposition . Under the assumptions of 4.3.8, if moreover T is τ -continuous, then,the family of functors T ( Y ) → T ( Y x ) , E E x , x ∈ X , is conservative.
Proof.
Let E be an object of T ( Y ) such that E x ≃ x of X . For anyconstructible object C of T ( Y ), we have a presheaf of S -spectra on the small Nisnevich site of X : F : U F ( U ) = Hom( M Y ( U × X Y ) , Hom Y ( C, E )) . It is sufficient to prove that F ( X ) is acyclic. As T satisfies Nisnevich descent (3.3.4), it is sufficientto prove that F is acyclic locally for the Nisnevich topology, i.e. that, for any point x of X , thespectrum F (Spec (cid:0) O hX,x (cid:1) ) is acyclic. Writing Spec (cid:0) O hX,x (cid:1) as the projective limit of the Nisnevichneighbourhoods of x in X , we see easily, using Proposition 4.3.4 and Lemma 4.3.7, that, for anyinteger i , π i ( F (Spec (cid:0) O hX,x (cid:1) ) ≃ Hom( C x , E x [ i ]) ≃ (cid:3) Proposition . Let S be a quasi-excellent noetherian and henselian scheme. Write ˆ S for its completion along its closed point, and assume that both S and ˆ S are in S . Consider an S -scheme of finite type X , and write i : ˆ S × S X → X for the induced map. If T is τ -continuous,then the pullback functor i ∗ : T ( X ) → T ( ˆ S × S X ) is conservative. Proof. As S is quasi-excellent, the map ˆ S → S is regular. By Popescu’s theorem, we canthen write ˆ S = lim ←− α S α , where { S α } is a projective system of schemes with affine transition maps,and such that each scheme S α is smooth over S . Moreover, as ˆ S and S have the same residuefield, and as S is henselian, each map S α has a section. Write X α = S α × S X , so that we have X = lim ←− α X α . Consider a constructible object C and an object E in T ( X ). Then, as the maps X α → X have sections, it follows from the first assertion of Proposition 4.3.4 that the mapHom T ( X ) ( C, E ) → Hom T ( ˆ S × S X ) ( i ∗ ( C ) , i ∗ ( E ))is a monomorphism (as a filtered colimit of such things). Hence, if i ∗ ( E ) ≃
0, for any constructibleobject C in T ( X ), we have Hom T ( X ) ( C, E ) ≃
0. Therefore, as τ -constructible objects generate T ( X ), we get E ≃ (cid:3) . CONSTRUCTIBLE MOTIVES 105 Proposition . Let a : X → Y be a regular morphism in S . If T is τ -continuous,then, for any objects E and F in T ( Y ) , with E constructible, there is a canonical isomorphism a ∗ Hom Y ( E, F ) ≃ Hom X ( a ∗ ( E ) , a ∗ ( F )) . Proof.
We want to prove that the canonical map a ∗ Hom Y ( E, F ) → Hom X ( a ∗ ( E ) , a ∗ ( F ))is an isomorphism, while we already know it is so whenever a is smooth. Therefore, to prove thegeneral case, we see that the problem is local on X and on Y with respect to the Zariski topology.In particular, we may assume that both X and Y are affine. By Pospecu’s Theorem 4.1.5, we thushave X = lim ←− α X α , where { X α } is a projective system of smooth affine Y -schemes. We concludeby Lemma 4.3.7. (cid:3) Proposition . Consider the following pullback square in S . X ′ a / / g (cid:15) (cid:15) X f (cid:15) (cid:15) Y ′ b / / Y Assume that f is separated of finite type and that b is regular. Then, if T is τ -continuous, forany object E in T ( Y ) , there is a canonical isomorphism in T ( X ′ ) : a ∗ f ! ( E ) ≃ g ! b ∗ ( E ) . Proof.
We have a canonical map f ! ( E ) → a ∗ g ! b ∗ ( E ) ≃ f ! b ∗ b ∗ ( E ) , which gives, by adjunction, a natural morphism a ∗ f ! ( E ) → g ! b ∗ ( E ) . The latter is invertible whenever b is smooth: this is obvious in the case of an open immersion, sothat, by Zariski descent, it is sufficient to treat the case where b is smooth with trivial cotangentbundle of rank d ; in this case, by relative purity (2.4.50 (3)), this reduces to the canonical isomor-phism a ! f ! ≃ g ! b ! evaluated at E ( − d )[ − d ]. To prove the general case, as the condition is local on X and on Y for the Zariski topology, we may assume that f factors as an immersion X → P nY ,followed by the canonical projection P nY → Y . We deduce from there that it is sufficient to treatthe case where f is either a closed immersion, either a smooth morphism of finite type. The casewhere f (hence also g ) is smooth follows by relative purity (2.4.50): we can then replace f ! and g ! by f ∗ and g ∗ respectively, and the formula follows from the fact that a ∗ f ∗ ≃ g ∗ b ∗ . We maythus assume that f is a closed immersion. As g is a closed immersion as well, the functor g ! isconservative (it is fully faithful). Therefore, it is sufficient to prove that the map b ∗ f ! f ! ( E ) ≃ g ! a ∗ f ! ( E ) → g ! g ! b ∗ ( E )is invertible. Then, using Proposition 4.3.11 (which makes sense because f ! preserves τ -construc-tibility by 4.2.11), and the projection formula, we have b ∗ f ! f ! ( E ) ≃ b ∗ Hom Y ( f ! ( X ) , E ) ≃ Hom Y ′ ( b ∗ f ! ( X ) , b ∗ ( E )) ≃ Hom Y ′ ( g ! ( X ′ ) , b ∗ ( E )) ≃ g ! g ! b ∗ ( E ) , which achieves the proof. (cid:3) Lemma . Let f : X → Y be a morphism in S . Assume that X = lim ←− α X α and Y = lim ←− α Y α , where { X α } and { Y α } are projective systems fo schemes with affine transition maps,while f is induced by a system of morphisms f α : X α → Y α . Let α be some index, C α aconstructible object of T ( Y α ) , and E α an object of T ( X α ) . If T is τ -continuous, then we havea natural isomorphism of abelian groups lim −→ α ≥ α Hom T ( Y α ) ( C α , f α, ∗ ( E α )) ≃ Hom T ( Y ) ( C, f ∗ ( E )) . Proof.
By virtue of Proposition 4.3.4, we have a natural isomorphismlim −→ α ≥ α Hom T ( X α ) ( f ∗ α ( C α ) , E α ) ≃ Hom T ( Y ) ( f ∗ ( C ) , E ) . The expected formula follows by adjunction. (cid:3)
Proposition . Consider the following pullback square in S . X ′ a / / g (cid:15) (cid:15) X f (cid:15) (cid:15) Y ′ b / / Y with b regular. If T is τ -continuous, then, for any object E in T ( X ) , there is a canonicalisomorphism in T ( Y ′ ) : b ∗ f ∗ ( E ) ≃ g ∗ a ∗ ( E ) . Proof.
This proposition is true in the case where b is smooth (by definition of Sm -fibredcategories), from which we deduce, by Zariski separation, that this property is local on Y and on Y ′ for the Zariski topology. In particular, we may assume that both Y and Y ′ are affine. Then, byPopescu’s Theorem 4.1.5, we may assume that Y ′ = lim ←− α Y ′ α , where { Y ′ α } is a projective systemof smooth Y -algebras. Then, using the preceding lemma as well as Proposition 4.3.4, we reduceeasily the proposition to the case where b is smooth. (cid:3) Proposition . Assume that T is τ -continuous, Q -linear and semi-separated, and con-sider a field k , with inseparable closure k ′ , such that both Spec ( k ) and Spec ( k ′ ) are in S . Givena k -scheme X write X ′ = k ′ ⊗ k X , and f : X ′ → X for the canonical projection. Then the functor f ∗ : T ( X ) → T ( X ′ ) is an equivalence of categories. Proof.
It follows immediately from Proposition 4.3.4 and from Proposition 2.1.9 that thefunctor f ∗ : T c ( X ) → T c ( X ′ )is an equivalence of categories. Similarly, for any objects C and E in T ( X ), if C is constructible,the map Hom T ( X ) ( C, E ) → Hom T ( X ) ( f ∗ ( C ) , f ∗ ( E ))is bijective. As constructible objects generate T ( X ), this implies that the functor f ∗ : T ( X ) → T ( X ′ )is fully faithful. As the latter is essentially surjective on a set of generators, this implies that it isan equivalence of categories (see 1.3.21). (cid:3) . CONSTRUCTIBLE MOTIVES 107 The aim of this section is to prove a local duality theorem in Ho( M ) (see4.4.21 and 4.4.24).If we work with rational coefficients, resolution of singularities up to quotient singularities isalmost as good as classical resolution of singularities: we have the following replacement of theblow-up formula. Theorem . Assume that
Ho( M ) is Q -linear and separated. Let X be a scheme in S .Consider a Galois alteration p : X ′ → X of group G , as well as a closed subscheme Z ⊂ X , suchthat U = X − Z is normal, and such that the induced map p U : U ′ = p − ( U ) → U is a finitemorphism. Then the pullback square Z ′ i ′ / / q (cid:15) (cid:15) X ′ p (cid:15) (cid:15) Z i / / X (4.4.1.1) induces an homotopy pullback square M / / (cid:15) (cid:15) ( R p ∗ L p ∗ ( M )) G (cid:15) (cid:15) R i ∗ L i ∗ ( M ) / / ( R i ∗ R q ∗ L q ∗ L i ∗ ( M )) G (4.4.1.2) for any object M of Ho( M )( X ) . Proof.
We already know that, for any object N of Ho( M )( U ), the map N → ( R p U ∗ L p ∗ U ( N )) G is an isomorphism (Corollary 3.3.39). The proof is then similar to the proof of condition (iv) ofTheorem 3.3.37. (cid:3) Remark . Under the assumptions of the preceding theorem, applying the total derivedfunctor R Hom X ( − , E ) to the homotopy pullback square (4.4.1.2) for M = X , we obtain thehomotopy pushout square ( i ! q ! q ! i ! ( E )) G / / (cid:15) (cid:15) ( p ! p ! ( E )) G (cid:15) (cid:15) i ! i ! ( E ) / / E (4.4.2.1)for any object E of Ho( M )( X ) . Corollary . Assume that
Ho( M ) is Q -linear and separated. Let B be a scheme in S , admitting wide resolution of singularities up to quotient singularities. Consider a separated B -scheme of finite type S , endowed with a closed subscheme T ⊂ S . The category of constructibleobjects in Ho( M )( S ) is the smallest thick triangulated subcategory which contains the objects ofshape R f ∗ ( X { n } ) for n ∈ τ , and for f : X → S a projective morphism, with X regular andconnected, such that f − ( T ) red is either empty, either X itself, either the support of a strictnormal crossing divisor. Proof.
Let Ho( M )( S ) ′ be the smallest thick triangulated subcategory of Ho( M )( S ) whichcontains the objects of shape R f ∗ ( X { n } ) for n ∈ τ and f : X → S a projective morphism with X regular and connected, while f − ( T ) red is empty, or X itself, or the support of a strict normalcrossing divisor. We clearly have Ho( M )( S ) ′ ⊂ Ho( M ) c ( S ) (Proposition 4.2.11). To prove thereverse inclusion, by virtue of Proposition 4.2.13, it is sufficient to prove that, for any n ∈ τ , andany projective morphism f : X → S , the object R f ∗ ( X { n } ) belongs to Ho( M )( S ) ′ . We shallproceed by induction on the dimension of X . If X is of dimension ≤
0, we may replace it byits reduction, which is regular. If X is of dimension >
0, by assumption on B , there exists a Galois alteration p : X ′ → X of group G , with X ′ regular and projective over S (and in which T becomes either empty, either X ′ itself, either the support of a strict normal crossing divisor, ineach connected component of X ′ ). Choose a closed subscheme Z ⊂ X , such that U = X − Z isa normal dense open subscheme, and such that the induced map r : U ′ = p − ( U ) → U is a finitemorphism, and consider the pullback square (4.4.1.1). As Z and Z ′ = p − ( Z ) are of dimensionsmaller than the dimension of X , we conclude from the homotopy pullback square obtained byappyling the functor R f ∗ to (4.4.1.2) for M = X { n } , n ∈ τ . (cid:3) Definition . Let S be a scheme in S . An object R of Ho( M )( S ) is τ -dualizing if itsatisfies the following conditions.(i) The object R is constructible.(ii) For any constructible object M of Ho( M )( S ), the natural map M → R Hom S ( R Hom S ( M, R ) , R )is an isomorphism. Remark . If Ho( M ) is τ -compatible, Q -linear and separated, then, in particular, the sixoperations of Grothendieck preserve τ -constructibility in Ho( M ) (4.2.29). Under this assumtion,for any scheme X in S , and any ⊗ -invertible object U in Ho( M )( X ) which is constructible, itsquasi-inverse is constructible: the quasi-inverse of U is simply its dual U ∧ = R Hom ( U, X ), whichis constructible by virtue of 4.2.25. Proposition . Assume that
Ho( M ) is τ -compatible, Q -linear and separated, and con-sider a scheme X in S .(i) Let R be a τ -dualizing object, and U be a constructible ⊗ -invertible object in Ho( M )( X ) .Then U ⊗ L S R is τ -dualizing.(ii) Let R and R ′ be two τ -dualizing objects in Ho( M )( X ) . Then the evaluation map R Hom S ( R, R ′ ) ⊗ L S R → R ′ is an isomorphism. Proof.
This follows immediately from [
Ayo07a , 2.1.139]. (cid:3)
Proposition . Consider an open immersion j : U → X in S . If R is a τ -dualizingobject in Ho( M )( X ) , then j ! ( R ) is τ -dualizing in Ho( M )( U ) . Proof. If M is a constructible object in Ho( M )( U ), then j ! ( M ) is constructible, and themap(4.4.7.1) j ! ( M ) → R Hom X ( R Hom X ( j ! ( M ) , R ) , R )is an isomorphism. Using the isomorphisms of type M ≃ j ∗ j ! ( M ) = j ! j ! ( M ) and j ∗ R Hom X ( A, B ) ≃ R Hom U ( j ∗ ( A ) , j ∗ ( B )) , we see that the image of the map (4.4.7.1) by the functor j ∗ = j ! is isomorphic to the map(4.4.7.2) M → R Hom U ( R Hom U ( M, j ! ( R )) , j ! ( R )) , which proves the proposition. (cid:3) Proposition . Let X be a scheme in S , and R an object in Ho( M )( X ) . Assume thereexists an open cover X = S i ∈ I U i such that the restriction of R on each of the open subschemes U i is τ -dualizing in Ho( M )( U i ) . Then R is τ -dualizing. Proof.
We already know that the property of τ -constructibility is local with respect to theZariski topology (4.2.6). Denote by j i : U i → X the corresponding open immersions, and put R i = j ! i ( R ). Let M be a constructible object in Ho( M )( X ). Then, for all i ∈ I , the image by j ∗ i = j ! i of the map M → R Hom X ( R Hom X ( M, R ) , R ) . CONSTRUCTIBLE MOTIVES 109 is isomorphic to the map j ∗ i ( M ) → R Hom U i ( R Hom U i ( j ∗ i ( M ) , R i ) , R i ) . This proposition thus follows from the property of separation with respect to the Zariski topology. (cid:3)
Corollary . Let f : X → Y be a separated morphism of finite type in S . Given anobject R of Ho( M )( Y ) , the property for f ! ( R ) of being a τ -dualizing object in Ho( M )( X ) is localover X and over Y for the Zariski topology. Proposition . Assume that
Ho( M ) is τ -compatible. Let i : Z → X be a closedimmersion and R be a τ -dualizing object in Ho( M )( X ) . Then i ! ( R ) is τ -dualizing in Ho( M )( Z ) . Proof.
As Ho( M ) is τ -compatible, we already know that i ! ( R ) is constructible. For anyobjects M and R of Ho( M )( Z ) and Ho( M )( X ) respectively, we have the identification: i ! R Hom Z ( M, i ! ( R )) ≃ R Hom X ( i ! ( M ) , R ) . Let j : U → X be the complement immersion. Then we have j ! R Hom X ( i ! ( M ) , R ) ≃ R Hom U ( j ∗ i ! ( M ) , j ! ( R )) ≃ , so that R Hom X ( i ! ( M ) , R ) ≃ i ! L i ∗ R Hom X ( i ! ( M ) , R ) . As i ! is fully faithful, this provides a canonical isomorphism L i ∗ R Hom X ( i ! ( M ) , R ) ≃ i ! R Hom X , ( i ! ( M ) , R ) . Under this identification, we see easily that the map i ! ( M ) → R Hom X ( R Hom X ( i ! ( M ) , R ) , R )is isomorphic to the image by i ! of the map M → R Hom Z ( R Hom Z ( M, i ! ( R )) , i ! ( R )) . As i ! is fully faithful, it is conservative, and this ends the proof. (cid:3) Proposition . Assume that
Ho( M ) is τ -compatible, Q -linear and separated, and con-sider a scheme B in S which admits wide resolution of singularities up to quotient singularities.Consider a separated B -scheme of finite type S , and a constructible object R in Ho( M )( S ) . Thefollowing conditions are equivalent.(i) For any separated morphism of finite type f : X → S , the object f ! ( R ) is τ -dualizing.(ii) For any projective morphism f : X → S , the object f ! ( R ) is τ -dualizing.(iii) For any projective morphism f : X → S , with X regular, the object f ! ( R ) is τ -dualizing.(iv) For any projective morphism f : X → S , with X regular, and for any n ∈ τ , the map (4.4.11.1) X { n } → R Hom X ( R Hom X ( X { n } , f ! ( R )) , f ! ( R )) is an isomorphism in Ho( M )( X ) .If, furthermore, for any regular separated B -scheme of finite type X , and for any n ∈ τ , the object X { n } is ⊗ -invertible, then these conditions are equivalent to the following one.(v) For any projective morphism f : X → S , with X regular, the map (4.4.11.2) X → R Hom X ( f ! ( R ) , f ! ( R )) is an isomorphism in Ho( M )( X ) . Proof.
It is clear that (i) implies (ii), which implies (iii), which implies (iv). Let us checkthat condition (ii) also implies condition (i). Let f : X → S be a morphism of separated B -schemes of finite type, with S regular. We want to prove that f ! ( S ) is τ -dualizing, while wealready know it is true whenever f is projective. In the general case, by virtue of Corollary 4.4.9,we may assume that f is quasi-projective, so that f = pj , where p is projective, and j is an openimmersion. As f ! ≃ j ! p ! , we conclude with Proposition 4.4.7. Under the additional assumption,the equivalence between (iv) and (v) is obvious. It thus remains to prove that (iv) implies (ii). It is in fact sufficient to prove that, under condition (iv), the object R itself is τ -dualizing. To provethat the map(4.4.11.3) M → R Hom X ( R Hom X ( M, R ) , R )is an isomorphism for any constructible object M of Ho( M )( S ), it is sufficient to consider the casewhere M = R f ∗ ( X { n } ) = f ! ( X { n } ), where n ∈ τ and f : X → S is a projective morphism with X regular (Corollary 4.4.3). For any object A of Ho( M )( X ), we have canonical isomorphisms R Hom S ( f ! ( A ) , R ) ≃ R f ∗ R Hom X ( A, f ! ( R ))= f ! R Hom X ( A, f ! ( R )) , from which we get a natural isomorphism: R Hom S ( R Hom S ( f ! ( A ) , R ) , R ) ≃ f ! R Hom X ( R Hom X ( A, f ! ( R )) , f ! ( R )) . Under these identifications, the map (4.4.11.3) for M = f ! ( X { n } ) is the image of the map(4.4.11.1) by the functor f ! . As (4.4.11.1) is invertible by assumption, this proves that R is τ -dualizing. (cid:3) Lemma . Let X be a scheme in S , and R be an object of Ho( M )( X ) . The property for R of being ⊗ -invertible is local over X with respect to the Zariski topology. Proof.
Let R ∧ = R Hom ( R, X ) be the dual of R . The object R is ⊗ -invertible if and onlyif the evaluation map R ∧ ⊗ L X R → X is invertible. Let j : U → X be an open immersion. Then, for any objects M and N in Ho( M )( X ),we have the identification j ∗ R Hom X ( M, N ) ≃ R Hom U ( j ∗ ( M ) , j ∗ ( N )) . In particular, we have j ∗ ( R ∧ ) ≃ j ∗ ( R ) ∧ . As j ∗ is monoidal, the lemma follows from the fact thatHo( M ) has the property of separation with respect to the Zariski topology. (cid:3) Definition . We shall say that Ho( M ) is τ -dualizable if it satisfies the following con-ditions:(i) Ho( M ) is τ -compatible (4.2.20);(ii) for any closed immersion between regular schemes i : Z → S in S , the object i ! ( S ) is ⊗ -invertible (i.e. the functor i ! ( S ) ⊗ L S ( − ) is an equivalence of categories);(ii) for any regular scheme X in S , and for any n ∈ τ , the map X { n } → R Hom X ( R Hom X ( X { n } , X ) , X )is an isomorphism.As in other similar situations, we simply say dualizable with respect to Tate twist when theset of twists τ is generated by the Tate twist. Example . In practice, the property of being dualizable with respect to Tate twist is aconsequence of the absolute purity theorem. Our main example is the motivic category DM B ofBeilinson motives over excellent noetherian schemes, as a consequence of Theorem 14.4.1. Remark . Note that, whenever the objects X { n } are ⊗ -invertible (which will be thecase in practice), conditions (i) and (ii) of the preceding definition are equivalent to the conditionthat i ! ( X ) is constructible and ⊗ -invertible for any closed immersion i between regular separatedschemes in S , while condition (iii) is then automatic. This principle gives easily the property of τ -purity when S is made of schemes of finite type over some field: Proposition . Assume that S consists exactly of schemes of finite type over a field k ,and that one of the following conditions is satisfied:(a) the field k is perfect;(b) Ho( M ) is semi-separated (2.1.7).If the objects { n } are ⊗ -invertible in Ho( M )(Spec ( k )) for all n ∈ τ , then Ho( M ) is τ -dualizable. . CONSTRUCTIBLE MOTIVES 111 Proof.
For any k -scheme of finite type f : X → Spec ( k ), as the functor L f ∗ is symmetricmonoidal, the objects X { n } are ⊗ -invertible in Ho( M )( X ) for all n ∈ τ . Therefore, as stated inremark 4.4.15, we have only to prove that, for any closed immersion i : Z → X between regular k -schemes of finite type, the object i ! ( X ) is ⊗ -invertible and constructible. We may assume that X and Z are smooth (under condition (a), this is clear, and under condition (b), by virtue ofProposition 2.1.9, we may replace k by any of its finite extensions). Using 4.4.12 and 4.2.6, wemay also assume that X is quasi-projective and that Z is purely of codimension c in X , while thenormal bundle of i is trivial. This proposition is then a consequence of relative purity (2.4.50),which gives a canonical isomorphism i ! ( X ) ≃ Z ( − c )[ − c ]. (cid:3) Proposition . Assume that S consists of schemes of finite type over a field k and that Ho( M ) has the following properties:(a) it is τ -dualizable;(b) for any n ∈ τ , { n } is rigid;(c) either k is perfect, either Ho( M ) is continuous.Then, any constructible object of Ho( M )( k ) is rigid. Proof.
By 4.3.15, it is sufficient to treat the case where k is perfect. It is well known thatrigid objects form a thick subcategory of Ho( M ). Thus we conclude easily from Corollary 4.4.3and Proposition 2.4.31. (cid:3) Lemma . Assume that
Ho( M ) is τ -dualizable. Then, for any projective morphism f : X → S between regular schemes in S , the object f ! ( S ) is ⊗ -invertible and constructible. Proof.
As, for any open immersion j : U → X , one has j ∗ = j ! , we deduce easily fromLemma 4.4.12 (resp. Proposition 4.2.6) that the property for f ! ( S ) of being ⊗ -invertible (resp.constructible) is local on S for the Zariski topology. Therefore, we may assume that S is separatedover B and that f factors as a closed immersion i : X → P nS followed by the canonical projection p : P nS → S . Using relative purity for p , we have the following computations: f ! ( S ) ≃ i ! p ! ( S ) ≃ i ! ( P nS ( n )[2 n ]) ≃ i ! ( P nS )( n )[2 n ] . As i is a closed immersion between regular schemes, the object i ! ( P nS ) is ⊗ -invertible and con-structible by assumption on Ho( M ), which implies that f ! ( S ) is ⊗ -invertible and constructibleas well. (cid:3) Definition . Let B a scheme in S . We shall say that local duality holds over B in Ho( M ) if, for any separated morphism of finite type f : X → S , with S regular and of finite typeover B , the object f ! ( S ) is τ -dualizing in Ho( M )( X ). Remark . By definition, if Ho( M ) is τ -compatible, and if local duality holds over B inHo( M ), then the restriction of Ho( M ) to the category of B -schemes of finite type is τ -dualizable.A convenient sufficient condition for local duality to hold in Ho( M ) is the following (in particular,using the result below as well as Proposition 4.4.16, local duality holds almost systematically overfields). Theorem . Assume that
Ho( M ) is τ -dualizable, Q -linear and separated, and considera scheme B in S which admits wide resolution of singularities up to quotient singularities (e.g. B might be any scheme which is separated and of finite type over an excellent noetherian schemeof dimension lesser or equal to in S ; see 4.1.11). Then local duality holds over B in Ho( M ) . Proof.
Let S be a regular separated B -scheme of finite type. Then, for any separated mor-phism of finite type f : X → S , the object f ! ( S ) is τ -dualizing: Lemma 4.4.18 implies immediatelycondition (iv) of Proposition 4.4.11. The general case (without the separation assumption on S )follows easily from Corollary 4.4.8. (cid:3) Proposition . Consider a scheme B in S . Assume that Ho( M ) is τ -dualizable, andthat local duality holds over B in Ho( M ) . Consider a regular B -scheme of finite type S .(i) An object of Ho( M )( S ) is τ -dualizing if and only if it is constructible and ⊗ -invertible. (ii) For any separated morphism of S -schemes of finite type f : X → Y , and for any τ -dualizing object R in Ho( M )( Y ) , the object f ! ( R ) is τ -dualizing in Ho( M )( X ) . Proof.
As the unit of Ho( M )( S ) is τ -dualizing by assumption, Proposition 4.4.6 impliesthat an object of Ho( M )( S ) is τ -dualizing if and only if it is constructible and ⊗ -invertible.Consider a regular B -scheme of finite type S , as well as a separated morphism of S -schemesof finite type f : X → Y , as well as a τ -dualizing object R in Ho( M )( Y ). To prove that f ! ( R ) is τ -dualizing, by virtue of Corollary 4.4.8, we may assume that Y is separated over S . Denote by u and v the structural maps from X and Y to S respectively. As we already know that v ! ( S ) is τ -dualizing, by virtue of Proposition 4.4.6, there exists a constructible and ⊗ -invertible object U in Ho( M )( Y ) such that U ⊗ L Y R ≃ v ! ( S ). As the functor L f ∗ is symmetric monoidal, it preserves ⊗ -invertible objects and their duals, from which we deduce the following isomorphisms: u ! ( S ) ≃ f ! v ! ( S ) ≃ f ! ( U ⊗ L Y R ) ≃ f ! R Hom Y ( U ∧ , R ) ≃ R Hom X ( L f ∗ ( U ∧ ) , f ! ( R )) ≃ R Hom X ( L f ∗ ( U ) ∧ , f ! ( R )) ≃ L f ∗ ( U ) ⊗ L X f ! ( R ) . The object a ! ( S ) being τ -dualizing, while L f ∗ ( U ) is constructible and invertible, we deduce fromProposition 4.4.6 that f ! ( R ) is τ -dualizing as well. (cid:3) . Assume that Ho( M ) is τ -dualizable, Q -linear and separated.Consider a scheme B in S , such that local duality holds over B in Ho( M ) – this is the caseif B admits wide resolution of singularities up to quotient singularities according to the aboveTheorem. Consider a fixed regular B -scheme of finite type S , as well as a constructible and ⊗ -invertible object R in Ho( M )( S ) (in the case S is of pure dimension d , it might be wise to consider R = S ( d )[2 d ], but an arbitrary R as above is eligible by 4.4.22). Then, for any separated S -schemeof finite type f : X → S , we define the local duality functor D X : Ho( M )( X ) op → Ho( M )( X )by the formula D X ( M ) = R Hom X ( M, f ! ( R )) . This functor D X is right adjoint to itself. Corollary . Under the above assumptions, we have the following properties of themotivic triangulated category
Ho( M ) :(a) For any separated S -scheme of finite type X , the functor D X preserves constructibleobjects.(b) For any separated S -scheme of finite type X , the natural map M → D X ( D X ( M )) is an isomorphism for any constructible object M in Ho( M )( X ) .(c) For any separated S -scheme of finite type X , and for any objects M and N in Ho( M )( X ) ,if N is constructible, then we have a canonical isomorphism D X ( M ⊗ L X D X ( N )) ≃ R Hom X ( M, N ) . . CONSTRUCTIBLE MOTIVES 113 (d) For any morphism between separated S -schemes of finite type f : Y → X , we havenatural isomorphisms D Y ( f ∗ ( M )) ≃ f ! ( D X ( M )) f ∗ ( D X ( M )) ≃ D Y ( f ! ( M )) D X ( f ! ( N )) ≃ f ∗ ( D Y ( N )) f ! ( D Y ( N )) ≃ D X ( f ∗ ( N )) for any constructible objects M and N in Ho( M )( X ) and Ho( M )( Y ) respectively. This corollary sums up what must be called the
Grothendieck duality property for the motivictriangulated category Ho( M ) with respect to the set of twists τ . Proof.
Assertions (a) and (b) are only stated for the record ; see 4.2.25. To prove (c), wesee that we have an obvious isomorphism D X ( M ⊗ L X P ) ≃ R Hom X ( M, D X ( P ))for any objects M and P . If N is constructible, we may replace P by D X ( N ) and get the expectedformula using (b). The identification D Y f ∗ ≃ f ! D X is a special case of the formula R Hom Y ( f ∗ ( A ) , f ! ( B )) ≃ f ! R Hom X ( A, B ) . Therefore, we also get: f ∗ D X ≃ D Y f ∗ D X ≃ D Y f ! D X ≃ D Y f ! . The two other formulas of (d) follow by adjunction. (cid:3)
Theorem . Assume that S consists of schemes of finite type over a field k , and considera τ -twisted motivic triangulated category T ′ over S , as well as a premotivic morphism compatiblewith τ -twists ϕ ∗ : T = Ho( M ) → T ′ . We suppose that the following properties hold:(a) T is τ -dualizable and Q -linear and separated;(b) the object { i } is rigid in T ( k ) for any i ∈ τ .Then, the premotivic morphism ϕ ∗ : T c → T ′ c commutes with the six operations. Remark . Remark that, as a corollary, we obtain immediately in the condition of thetheorem that the functor ϕ ∗ commutes with the duality functor on T and T ′ respectively obtainedby applying the above corollary in the case B = Spec ( k ). Proof.
Given a morphism of finite type f : X → Spec ( k ), let us consider the followingproperty.( ∗ ) f For any constructible object M in T ( X ) , the natural exchange map ϕ ∗ f ∗ ( M ) → f ∗ ϕ ∗ ( M ) is invertible. We have put to a lot of assumptions here: in fact, if Ho( M ) is τ -dualizable and if local duality holds over B in Ho( M ), the six Grothendieck operations preserve constructible objects on the restriction of Ho( M ) to B -schemesof finite type; we leave this as a formal exercice for the reader. We will first prove the theorem assuming that property ( ∗ ) f holds for any f .Let u : X → Y be a k -morphism of finite type. We claim that the exchange map ϕ ∗ u ∗ ( M ) → u ∗ ϕ ∗ ( M )is invertible for any τ -constructible object M of T ( X ).It is sufficient to prove that, for any smooth separated k -morphism of finite type g : T → X ,any constructible object M in T ( X ) and any twist i , the natural mapHom T ′ ( X ) ( g ♯ ( T { i } ) , ϕ ∗ u ∗ ( M )) → Hom T ′ ( X ) ( g ♯ ( T { i } ) , u ∗ ( M ))is bijective. Consider the following commutative diagram of morphisms of schemes: V v / / h (cid:15) (cid:15) T g (cid:15) (cid:15) X a ! ! ❈❈❈❈❈❈ u / / Y b } } ⑤⑤⑤⑤⑤⑤ Spec ( k )in which the square is cartesian. Recall that the functor v ∗ preserves constructible objects byvirtue of Theorem 4.2.16. Then we conclude by the computations below:Hom T ′ ( Y ) ( g ♯ ( T { i } ) , ϕ ∗ u ∗ ( M )) = Hom T ′ ( T ) ( Y { i } , g ∗ ϕ ∗ u ∗ ( M ))= Hom T ′ ( T ) ( Y { i } , ϕ ∗ g ∗ u ∗ ( M ))= Hom T ′ ( T ) ( g ∗ b ∗ ( k { i } ) , ϕ ∗ g ∗ u ∗ ( M ))= Hom T ′ ( T ) ( g ∗ b ∗ ( k { i } ) , ϕ ∗ v ∗ h ∗ ( M ))= Hom T ′ ( k ) ( k { i } , ( bg ) ∗ ϕ ∗ v ∗ h ∗ ( M ))= Hom T ′ ( k ) ( k { i } , ϕ ∗ ( bg ) ∗ v ∗ h ∗ ( M )) (by ( ∗ ) bg )= Hom T ′ ( k ) ( k { i } , ( bgv ) ∗ ϕ ∗ h ∗ ( M )) (by ( ∗ ) bgv )= Hom T ′ ( k ) ( k { i } , ( bg ) ∗ g ∗ u ∗ ϕ ∗ ( M ))= Hom T ′ ( Y ) ( g ♯ ( T { i } ) , u ∗ ϕ ∗ ( M ))From there, we see that, for any k -scheme of finite type X and any τ -constructible objects M and N of T ( X ), the natural map ϕ ∗ ( Hom X ( M, N )) → Hom X ( ϕ ∗ ( M ) , ϕ ∗ ( N ))is invertible in T ′ ( X ). For this, we may assume that M = f ♯ ( Y { i } ) for a smooth morphism offinite type f : Y → X and a twist i , in which case we have ϕ ∗ ( Hom X ( M, N )) = ϕ ∗ f ∗ f ∗ ( N ) ≃ f ∗ f ∗ ϕ ∗ ( N ) = Hom X ( ϕ ∗ ( M ) , ϕ ∗ ( N )) . It remains to prove that for any separated k -morphism f : X → Y of finite type and any con-structible object N in T ( X ), the exchange map: ϕ ∗ f ! ( N ) → f ! ϕ ∗ ( N )is an isomorphism. It is sufficient to prove that for any constructible object M in T ( X ), theinduced map: Hom T ′ ( Y ) ( ϕ ∗ ( M ) , ϕ ∗ f ! ( N )) → Hom T ′ ( Y ) ( ϕ ∗ ( M ) , f ! ϕ ∗ ( N ))is bijective, because ϕ ∗ is essentially surjective on the family of generators of T ′ c ( Y ) of the form g ♯ ( W { i } ) for a smooth morphism g and a twist i . This follows formally by applying the functor . CONSTRUCTIBLE MOTIVES 115 Hom T ′ ( Y ) ( Y , − ) to the isomorphisms: f ∗ Hom X ( ϕ ∗ ( M ) , ϕ ∗ f ! ( N )) = ϕ ∗ f ∗ Hom X ( M, f ! ( N ))= ϕ ∗ Hom X ( f ! ( M ) , N )= Hom X ( f ! ϕ ∗ ( M ) , ϕ ∗ ( N ))= f ∗ Hom X ( ϕ ∗ ( M ) , f ! ϕ ∗ ( N ))It remains to prove property ( ∗ ) f for any morphism f of finite type.We claim it is sufficient to prove that, for any k -scheme of finite type X with structuralmorphism f , the following property holds:( ∗∗ ) X For any twist i ∈ τ , the natural exchange map ϕ ∗ f ∗ ( X { i } ) → f ∗ ϕ ∗ ( X { i } ) is invertible. Indeed, by virtue of Theorem 4.2.13, we may assume that M = w ∗ ( W { i } ) for w : W → X aprojective k -morphism, and i ∈ τ . As the exchange map ϕ ∗ w ∗ → w ∗ ϕ ∗ is invertible (Proposition2.4.53), we see that we may assume that M = X { i } for some twist i .Let us prove property ( ∗∗ ) X in the case X is in addition smooth over k . As ϕ ∗ is monoidal,for any rigid object M of T ( k ), we get the identification: ϕ ∗ ( M ∨ ) = ϕ ∗ ( M ) ∨ . On the other hand, according to assumption (b), the object f ♯ ( X ) is rigid in T ( k ) as well as in T ′ ( k ) (because the functor ϕ ∗ is symmetric monoidal and commutes with the operations of theform f ♯ for f smooth). Thus we get: f ∗ ( X { i } ) = Hom k ( f ♯ ( X ) , k { i } ) = f ♯ ( X ) ∨ { i } . Then property ( ∗∗ ) X readily follows.We finally prove property ( ∗∗ ) X for any algebraic k -scheme X . We will proceed by inductionon the dimension of X .In case dim( X ) <
0, the result is obvious. Let us assume dim( X ) ≥
0. According to thelocalization property, we can assume that X is reduced. Let ¯ k be an inseparable closure of k and¯ X = X ⊗ k ¯ k . According to De Jong theorem applied to ¯ X (see Th. 4.1.10 for S = Spec (cid:0) ¯ k (cid:1) ), thereexists a Galois alteration ¯ X ′ → ¯ X of group G such that ¯ X ′ is smooth over ¯ k .We can assume that such a smooth alteration exists over a finite inseparable extension field E/k . Because T (resp. T ′ ) is Q -linear and separated, the base change functor φ ∗ associated withthe finite morphism φ : Spec ( E ) → Spec ( k ) and relative to the premotivic category T (resp. T ′ )is an equivalence of categories (see Proposition 2.1.9). Thus we can replace k by E and assumethat there exists a Galois alteration p : X ′ → X of group G such that X ′ is a smooth k -scheme.Using the localization property, we can assume X is reduced. Then there exists a nowhere denseclosed subscheme ν : Z → X such that U = X − Z is regular (thus normal) and the induced map p | U : p − ( U ) → U is finite. Thus we can apply Theorem 4.4.1 to the cartesian square: Z ′ ν ′ / / q (cid:15) (cid:15) X ′ p (cid:15) (cid:15) Z ν / / X and we get the distinguished triangle in T ( X ) (thus in T ′ ( X ) as well, as the functor ϕ ∗ ismonoidal and commutes with the operations of the form u ∗ for any proper morphism u ) of theform: X { i } → p ∗ ( X ′ { i } ) G ⊕ ν ∗ ( Z { i } ) → ( νq ) ∗ ( Z ′ { i } ) G +1 −−→ for any twist i . If we consider the triangles in T ( k ) and T ′ ( k ) obtained by applying the functor f ∗ ,where f is the structural morphism of X/k , we deduce that property ( ∗∗ ) X follows from properties( ∗∗ ) X ′ , ( ∗∗ ) Z , ( ∗∗ ) Z ′ . Thus we can conclude applying either the case of a smooth k -scheme treatedabove or the induction hypothesis as dim( Z ) = dim( Z ′ ) < dim( X ). (cid:3) art 2 Construction of fibred categories . Fibred derived categories . In this entire section, we fix a full subcategory S of the category of noetherian S -schemessatisfying the following properties:(a) S is closed under finite sums and pullback along morphisms of finite type.(b) For any scheme S in S , any quasi-projective S -scheme belongs to S .We fix an admissible class of morphisms P of S . All our P -premotivic categories ( cf. definition 1.4.2) are defined over S . Moreover, for any abelian P -premotivic category A in thissection, we assume the following:(c) A is a Grothendieck abelian P -premotivic category (see definition 1.3.8 and the recallbelow).(d) A is given with a generating set of twists τ . We sometimes refer to it as the twists of A .(e) We will denote by M S ( X, A ), or simply by M S ( X ), the geometric section over a P -scheme X/S .Without precision, any scheme will be assumed to be an object of S .In section 5.2, except possibly for 5.2.a, we assume further:(f) P contains the class of smooth finite type morphisms.In section 5.3, we assume (f) and instead of (d) above.5.0.27 . We will refer sometimes to the canonical dg-structure of the category of complexesC( A ) over an abelian category A . Recall that to any complexes K and L over A , we associatea complex of abelian groups Hom • A ( K, L ) whose component in degree n ∈ Z is Y p ∈ Z Hom A ( K p , L p + n )and whose differential in degree n ∈ Z is defined by the formula:( f p ) p ∈ Z (cid:0) d L ◦ f p − ( − n .f p +1 ◦ d K ) (cid:1) p ∈ Z . In other words, this is the image of the bicomplex Hom A ( K, L ) by the Tot-product functorwhich we denote by Tot π . Of course, the associated homotopy category is the category K( A )of complexes up to chain homotopy equivalence. Abelian premotives: recall and examples.
Consider an abelian P -premotivic category A . According to the convention of 5.0, for any scheme S , A S is a Grothendieck abelian closedsymmetric monoidal category. Moreover, if τ denotes the twists of A , the essentially small family (cid:0) M S ( X ) { i } (cid:1) X ∈ P /S,i ∈ τ is a family of generators of A S in the sense of [ Gro57 ]. Example . Consider a fixed ring Λ. Let PSh( P /S, Λ) be the category of Λ-presheaves(i.e. presheaves of Λ-modules) on P /S . For any P -scheme X/S , we let Λ S ( X ) be the freeΛ-presheaf on P /S represented by X . Then PSh( P /S, Λ) is a Grothendieck abelian categorygenerated by the essentially small family (cid:0) Λ S ( X ) (cid:1) X ∈ P /S .There is a unique symmetric closed monoidal structure on PSh( P /S, Λ) such thatΛ S ( X ) ⊗ S Λ S ( Y ) = Λ S ( X × S Y ) . Finally the existence of functors f ∗ , f ∗ and, in the case when f is a P -morphism, of f ♯ , followsfrom general sheaf theory ( cf. [ SGA4 ]).Thus, PSh( P , Λ) defines an abelian P -premotivic category.5.1.2 . Consider an abstract abelian P -premotivic category A . To any premotive M of A S ,we can associate a presheaf of abelian groups X Hom A S ( M S ( X ) , M ) . FIBRED DERIVED CATEGORIES 119 which we denote by γ ∗ ( M ).This defines a functor γ ∗ : A S → PSh( P /S, Z ). It admits the following left adjoint: γ ∗ : PSh( P /S, Z ) → A S , F lim −→ X/F M S ( X, A )where the colimit runs over the category of representable presheaves over F .It is now easy to check we have defined a morphism of (complete) abelian P -premotiviccategories:(5.1.2.1) γ ∗ : PSh( P , Z ) ⇄ A : γ ∗ . Moreover PSh( P , Z ) appears as the initial abelian P -premotivic category.Remark that the functor γ ∗ : A S → PSh( P /S, Z ) is conservative if the set of twists τ of A istrivial. Definition . A P -admissible topology t is a Grothendieck pretopology t on the category S , such that any t -covering family consists of P -morphisms.Note that, for any scheme S in S , such a topology t induces a pretopology on P /S (whichwe denote by the same letter). For any morphism (resp. P -morphism) f : T → S , the functor f ∗ (resp. f ♯ ) preserves t -covering families.As P is fixed in all this section, we will simply say admissible for P -admissible. Example . Let t be an admissible topology. We denote by Sh t ( P /S, Λ) the categoryof t -sheaves of Λ-modules on P /S . Given a P -scheme X/S , we let Λ tS ( X ) be the free Λ-linear t -sheaf represented by X . Then, Sh t ( P /S, Λ) is an abelian Grothendieck category with generators(Λ tS ( X )) X ∈ P /S .As in the preceding example, the category Sh t ( P /S, Λ) admits a unique closed symmetricmonoidal structure such that Λ tS ( X ) ⊗ S Λ tS ( Y ) = Λ tS ( X × S Y ). Finally, for any morphism f : T → S of schemes, the existence of functors f ∗ , f ∗ (resp. f ♯ when f is a P -morphism) follows fromthe general theory of sheaves (see again [ SGA4 ]: according to our assumption on t and [ SGA4 ,III, 1.6], the functors f ∗ : P /S → P /T and f ♯ : P /T → P /S (for f in P ) are continuous).Thus, Sh t ( P , Λ) defines an abelian P -premotivic category (with trivial set of twists).The associated t -sheaf functor induces a morphism(5.1.4.1) a ∗ t : PSh( P , Λ) ⇄ Sh t ( P , Λ) : a t, ∗ . Remark . Recall the abelian category Sh t ( P /S, Z ) is a localization of the categoryPSh( S, Z ) in the sense of Gabriel-Zisman. In particular, given an abstract abelian P -premotiviccategory A , the canonical morphism γ ∗ : PSh( P /S, Z ) ⇄ A S : γ ∗ induces a unique morphism Sh t ( P /S, Z ) ⇄ A S if and only if for any presheaf of abelian groups F on P /S such that a t ( F ) = F t = 0, one has γ ∗ ( F ) = 0.We leave to the reader the exercise which consists to formulate the universal property of theabelian P -premotivic category Sh t ( P , Z ). The t -descent model category structure. . Consider an abelian P -premotivic category A with set of twists τ .We let C( A ) be the P -fibered abelian category over S whose fibers over a scheme S is thecategory C( A S ) of (unbounded) complexes in A S . For any scheme S , we let ι S : A S → C( A S )the embedding which sends an object of A S to the corresponding complex concentrated in degreezero.If A is τ -twisted, then the category C( A S ) is obviously ( Z × τ )-twisted. The following lemmais straightforward : We will formulate a derived version in the paragraph on descent properties for derived premotives ( cf. Lemma . With the notations above, there is a unique structure of abelian P -premotiviccategory on C( A ) such that the functor ι : A → C( A ) is a morphism of abelian P -premotiviccategories. . For a scheme S , let ( P /S ) ∐ be the category introduced in 3.2.1. The functor M S ( − )can be extended to ( P /S ) ∐ by associating to a family ( X i ) i ∈ I of P -schemes over S the premotive M i ∈ I M S ( X i ) . If X is a simplicial object of ( P /S ) ∐ , we denote by M S ( X ) the complex associated with thesimplicial object of A S obtained by applying degreewise the above extension of M S ( − ). Definition . Let A be an abelian P -premotivic category and t be an admissible topol-ogy.Let S be a scheme and C be an object of C( A S ) :(1) The complex C is said to be local (with respect to the geometric section) if, for any P -scheme X/S and any pair ( n, i ) ∈ Z × τ , the canonical morphismHom K( A S ) ( M S ( X ) { i } [ n ] , C ) → Hom D( A S ) ( M S ( X ) { i } [ n ] , C )is an isomorphism.(2) The complex C is said to be t -flasque if for any t -hypercover X → X in P /S , for any( n, i ) ∈ Z × τ , the canonical morphismHom K( A S ) ( M S ( X ) { i } [ n ] , C ) → Hom K( A S ) ( M S ( X ) { i } [ n ] , C )is an isomorphism.We say the abelian P -premotivic category A satifies cohomological t -descent if for any t -hypercover X → X of a P -scheme X/S , and for any i ∈ τ , the map M S ( X ) { i } → M S ( X ) { i } is a quasi-isomorphism (or equivalently, if any local complex is t -flasque).We say that A is compatible with t if A satisfies cohomological t -descent, and if, for anyscheme S , any t -flasque complex of A S is local. Example . Consider the notations of 5.1.4.Consider the canonical dg-structure on C(Sh t ( P /S, Λ)) (see 5.1.1). By definition, for anycomplexes D and C of sheaves, we get an equality:Hom K(Sh t ( P /S, Λ)) ( D, C ) = H (Hom • Sh t ( P /S, Λ) ( D, C )) = H (Tot π Hom Sh t ( P /S, Λ) ( D, C )) . In the case where D = Λ tS ( X ) (resp. D = Λ tS ( X )) for a P -scheme X/S (resp. a simplicial P -scheme over S ) we obtain the following identification:Hom K(Sh t ( P /S, Λ)) (Λ tS ( X ) , C ) = H ( C ( X )) . (resp. Hom K(Sh t ( P /S, Λ)) (Λ tS ( X ) , C ) = H (Tot π C ( X )) ) . Thus, we get the following equivalences: C is local ⇔ for any P -scheme X/S , H nt ( X, C ) ≃ H n ( C ( X )) .C is t -flasque ⇔ for any t -hypercover X → X , H n ( C ( X )) ≃ H n (Tot π C ( X )) . According to the computation of cohomology with hypercovers ( cf. [ Bro74 ]), if the complex C is t -flasque, it is local. In other words, we have the expected property that the abelian P -premotivic category Sh t ( P , Λ) is compatible with t .5.1.11 . Consider an abelian P -premotivic category A and an admissible topology t .Fix a base scheme S . A morphism p : C → D of complexes on A S is called a t -fibration if itskernel is a t -flasque complex and if for any P -scheme X/S , any i ∈ τ and any integer n ∈ Z , themap of abelian groups Hom A S ( M S ( X ) { i } , C n ) → Hom A S ( M S ( X ) { i } , D n ) . FIBRED DERIVED CATEGORIES 121 is surjective.For any object A of A S , we let S n A (resp. D n A ) be the complex with only one non trivialterm (resp. two non trivial terms) equal to A in degree n (resp. in degree n and n + 1, with theidentity as only non trivial differential). We define the class of cofibrations as the smallest classof morphisms of C( A S ) which :(1) contains the map S n +1 M S ( X ) { i } → D n M S ( X ) { i } for any P -scheme X/S , any i ∈ τ ,and any integer n ;(2) is stable by pushout, transfinite composition and retract.A complex C is said to be cofibrant if the canonical map 0 → C is a cofibration. For instance, forany P -scheme X/S and any i ∈ τ , the complex M S ( X ) { i } [ n ] is cofibrant.Let G S be the essentially small family made of premotives M S ( X ) { i } for a P -scheme X/S and a twist i ∈ τ , and H S be the family of complexes of the form Cone( M S ( X ) { i } → M S ( X ) { i } )for any t -hypercover X → X and any twist i ∈ τ . By the very definition, as A is compatiblewith t (definition 5.1.9), ( G S , H S ) is a descent structure on A S in the sense of [ CD09 , def. 2.2].Moreover, it is weakly flat in the sense of [
CD09 , par. 3.1]. Thus the following proposition is aparticular case of [
CD09 , theorem 2.5, proposition 3.2, and corollary 5.5] :
Proposition . Let A be an abelian P -premotivic category, which we assume to becompatible with an admissible topology t . Then for any scheme S , the category C( A S ) with thepreceding definition of fibrations and cofibrations, with quasi-isomorphisms as weak equivalencesis a proper symmetric monoidal model category. . We will call this model structure on C( A S ) the t -descent model category structure (over S ). Note that, for any P -scheme X/S and any twist i ∈ τ , the complex M S ( X ) { i } concentratedin degree 0 is cofibrant by definition, as well as any of its suspensions and twists. They form afamily of generators for the triangulated category D( A S ).Observe also that the fibrant objects for the t -descent model category structure are exactly the t -flasque complexes in A S . Moreover, essentially by definition, a complex of A S is local if andonly if it is t -flasque (see [ CD09 , 2.5]).5.1.14 . Consider again the notations and hypothesis of 5.1.11.Consider a morphism of schemes f : T → S . Then the functor f ∗ : C( A S ) → C( A T )sends G S in G T , and H S in H T because the topology t is admissible. This means it satisfiesdescent according to the definition of [ CD09 , 2.4]. Applying theorem 2.14 of op. cit. , the functor f ∗ preserves cofibrations and trivial cofibrations, i.e. the pair of functors ( f ∗ , f ∗ ) is a Quillenadjunction with respect to the t -descent model category structures.Assume that f is a P -morphism. Then, similarly, the functor f ♯ : C( A T ) → C( A S )sends G S (resp. H S ) in G T (resp. H T ) so that it f ♯ also satisfies descent in the sense of op. cit .Therefore, it preserves cofibrations and trivial cofibrations, and the pair of adjoint functors ( f ♯ , f ∗ )is a Quillen adjunction for the t -descent model category structures.In other words, we have obtained the following result. Corollary . Let A be an abelian P -premotivic category compatible with an admis-sible topology t . The P -fibred category C( A ) with the t -descent model category structure definedin 5.1.12 is a symmetric monoidal P -fibred model category. Moreover, it is stable, proper andcombinatorial. . Recall the following consequences of this corollary (see also 1.3.24 for the generaltheory). Consider a morphism f : T → S of schemes. Then the pair of adjoint functors ( f ∗ , f ∗ )admits total left/right derived functors L f ∗ : D( A S ) ⇄ D( A T ) : R f ∗ . More precisely, f ∗ (resp. f ∗ ) preserves t -local (resp. cofibrant) complexes. For any complex K on A S , R f ∗ ( K ) = f ∗ ( K ′ ) (resp. L f ∗ ( K ) = f ∗ ( K ′′ )) where K ′ → K (resp. K → K ′′ ) is a t -local(resp. cofibrant) resolution of K . When f is a P -morphism, the functor f ∗ is even exact and thus preserves quasi-isomorphisms.This implies that L f ∗ = f ∗ . The functor f ♯ admits a total left derived functor L f ♯ : D( A T ) ⇄ D( A S ) : R f ∗ defined by the formula L f ♯ ( K ) = f ♯ ( K ′′ ) for a complex K on A T and a cofibrant resolution K ′′ → K .Note also that the tensor product (resp. internal Hom) of C( A S ) admits a total left derivedfunctor (resp. total right derived functor). For any complexes K and L on A S , this derivedfunctors are defined by the formula: K ⊗ L S L = K ′′ ⊗ S L ′′ R Hom S ( K, L ) =
Hom S ( K ′′ , L ′ )where K → K ′′ and L → L ′′ are cofibrant resolutions and L ′ → L is a t -local resolution.It is now easy to check that these functors define a triangulated P -premotivic category D( A ),which is τ -generated according to 5.1.13. Definition . Let A be an abelian P -premotivic category compatible with an admis-sible topology t .The triangulated P -premotivic category D( A ) defined above is called the derived P -premotiviccategory associated with A . The geometric section of a P -scheme X/S in the category D( A ) is the complex concentratedin degree 0 equal to the object M S ( X ). The triangulated P -fibred category is τ -generated andwell generated in the sense of 1.3.16. Recall this means that D( A S ) is equal to the localizing subcategory generated by the family(5.1.17.1) { M S ( X ) { i } ; X/S P -scheme , i ∈ τ } . Example . Given any admissible topology t , the abelian P -premotivic category Sh t ( P , Λ)introduced in example 5.1.4 is compatible with t ( cf. P -premotiviccategory D(Sh t ( P , Λ)).Remark also that the abelian P -premotivic category PSh( P , Λ) introduced in example 5.1.1 iscompatible with the coarse topology and gives the derived P -premotivic category D(PSh( P , Λ)).
Remark . Recall from 5.0.27 there exists a canonical dg-structure on C( A S ). Then wecan define a derived dg-structure by defining for any complexes K and L of A S , the complex ofmorphisms: R Hom A S ( K, L ) = Hom • A S ( Q ( K ) , R ( L ))where R and Q are respectively some fibrant and cofibrant (functorial) resolutions for the t -descentmodel structure. The homotopy category associated with this new dg-structure on C( A S ) is thederived category D( A S ). Moreover, for any morphism (resp. P -morphism) of schemes f , thepair ( L f ∗ , R f ∗ ) (resp. ( L f ♯ , f ∗ )) is a dg-adjunction. The same is true for the pair of bifunctors( ⊗ L S , R Hom S ).5.1.20 . Consider an abelian P -premotivic category A compatible with a topology t . Accord-ing to section 3.1.b, the 2-functor D( A ) can be extended to the category of S -diagrams: to anydiagram of schemes X : I → S indexed by a small category I , we can associate a symmetricmonoidal closed triangulated category D( A )( X , I ) which coincides with D( A )( X ) when I = e , X = X for a scheme X . Recall also that fibrant/cofibrant resolutions can be made functorially, because our model categories arecofibrantely generated, so that the left or right derived functors are in fact defined at the level of complexes. Indeed remark that D( A ) does not depend on the topology t . i.e. triangulated and stable by sums. . FIBRED DERIVED CATEGORIES 123 Let us be more specific. The fibred category A admits an extension to S -diagrams: a sectionof A over a diagram of schemes X : I → S , indexed by a small category I , is the following data:(1) A family ( A i ) i ∈ I such that A i is an object of A X i .(2) A family ( a u ) u ∈ F l ( I ) such that for any arrow u : i → j in I , a u : u ∗ ( A j ) → A i is amorphism in A X i and this family of morphisms satisfies a cocyle condition (see paragraph3.1.1).Then, D( A )( X , I ) is the derived category of the abelian category A ( X , I ). In particular, objectsof D( A )( X , I ) are complexes of sections of A over ( X , I ) (or, what amount to the same thing,families of complexes ( K i ) i ∈ I with transition maps ( a u ) as above, relative to the fibred categoryC( A )).Recall that a morphism of S -diagrams ϕ : ( X , I ) → ( Y , J ) is given by a functor f : I → J and a natural transformation ϕ : X → Y ◦ f . We say that ϕ is a P -morphism if for any i ∈ I , ϕ i : X i → Y f ( i ) is a P -morphism. For any morphism (resp. P -morphism) ϕ , we have defined in3.1.3 adjunctions of (abelian) categories: ϕ ∗ : A ( Y , J ) ⇄ A ( X , I ) : ϕ ∗ resp. ϕ ♯ : A ( X , I ) ⇄ A ( Y , J ) : ϕ ∗ which extends the adjunctions we had on trivial diagrams.According to Proposition 3.1.11, these respective adjunctions admits left/right derived func-tors as follows: L ϕ ∗ : D( A )( Y , J ) ⇄ D( A )( X , I ) : R ϕ ∗ (5.1.20.1) resp. L ϕ ♯ : D( A )( X , I ) ⇄ D( A )( Y , J ) : L ϕ ∗ = ϕ ∗ (5.1.20.2)Again, these adjunctions coincide on trivial diagrams with the map we already had.Note also that the symmetric closed monoidal structure on C( A ( X , I )) can be derived andinduces a symmetric monoidal structure on D( A )( X , I ) (see Proposition 3.1.24). Recall from 3.2.5 and 3.2.7 that, given a topology t ′ (not necessarily admissible) over S , wesay that D( A ) satisfies t ′ -descent if for any t ′ -hypercover p : X → X (here X is considered as a S -diagram), the functor(5.1.20.3) L p ∗ : D( A )( X ) → D( A )( X )is fully faithful (see Corollary 3.2.7). Proposition . Consider the notations and hypothesis introduced above. Let t ′ be anadmissible topology on S . Then the following conditions are equivalent:(i) D( A ) satisfies t ′ -descent.(ii) A satisfies cohomological t ′ -descent. Proof.
We prove (i) implies (ii). Consider a t ′ -hypercover p : X → X in P /S . This isa P -morphism. Thus, by the fully faithfulness of (5.1.20.3), the counit map L p ♯ p ∗ → X of D( A X ), we thus obtain that M X ( X ) → X is an isomorphism in D( A X ). If π : X → S is the structural P -morphism, by applying the functor L π ♯ to this isomorphism, we obtain that M S ( X ) → M S ( X )is an isomorphism in D( A S ) and this concludes.Reciprocally, to prove (i), we can restrict to t ′ -hypercovers p : X → X which are P -morphisms because t ′ is admissible. Because R p ∗ = p ∗ admits a left adjoint L p ♯ , we have toprove that the counit L p ♯ p ∗ → In fact, D( A ) is then a monoidal P cart -fibred category over the category of S -diagrams (remark 3.1.21). is an isomorphism. This is a natural transformation between triangulated functors which commuteswith small sums. Thus, according to (5.1.17.1), we have only to check this is an isomorphismwhen evaluated at a complex of the form M X ( Y ) { i } for a P -scheme Y /X and a twist i ∈ τ .But the resulting morphism is then M X ( X × X Y ) { i } → M X ( Y ) { i } and we can conclude because X × X Y → Y is a t ′ -hypercover in P /S (again because t ′ is admissible). (cid:3) . . Consider the situation of 5.1.20 Let S be a scheme. An interesting particular caseis given for constant S -diagrams over S ; for a small category I , we let I S be the constant S -diagram I → S , i S, u S . Then the adjunctions (5.1.20.1) for this kind of diagrams definea Grothendieck derivator I D( A )( I S ) . Recall that, if f : I → e is the canonical functor to the terminal category and ϕ = f X : I X → X the corresponding morphism of S -diagrams, for any I -diagram K • = ( K i ) i ∈ I of complexes over A S , we get right derived limits and left derived colimits: R ϕ ∗ ( K • ) = R lim ←− i ∈ I K i . L ϕ ♯ ( K • ) = L lim −→ i ∈ I K i . . The associated derived P -premotivic category is functorial in the following sense.Consider an adjunction ϕ : A ⇄ B : ψ of abelian P -premotivic categories. Let τ (resp. τ ′ ) be the set of twists of A (resp. B ), andrecall that ϕ induces a morphisms of monoid τ → τ ′ still denoted by ϕ . Consider two topologies t and t ′ such that t ′ is finer than t . Suppose A (resp. B ) is compatible with t (resp. t ′ ) and let( G A S , H A S ) (resp. ( G B S , H B S )) be the descent structure on A S (resp. B S ) defined in 5.1.11.For any scheme S , consider the evident extensions ϕ S : C( A S ) ⇄ C( B S ) : ψ S of the above adjoint functors to complexes. Recall that for any P -scheme X/S and any twist i ∈ τ , ϕ S ( M S ( X, A ) { i } ) = M S ( X, B ) { ϕ ( i ) } by definition. Thus, ϕ S sends G A S to G A S . Because t ′ is finer than t , it sends also H A S to H B S . In other words, it satifies descent in the sense of [ CD09 ,par. 2.4] so that the pair ( ϕ S , ψ S ) is a Quillen adjunction with respect to the respective t -descentand t ′ -descent model structure on C( A S ) and C( B S ).Considering the derived functors, it is now easy to check we have obtained a P -premotivicadjunction L ϕ : D( A ) ⇄ D( B ) : R ψ. Example . Let t be an admissible topology. Consider an abelian P -premotivic category A compatible with t . Then the morphism of abelian P -premotivic categories (5.1.2.1) induces amorphism of triangulated P -premotivic categories:(5.1.24.1) L γ ∗ : D(PSh( P , Z )) ⇄ D( A ) : R γ ∗ Similarly, the morphism (5.1.4.1) induces a morphism of triangulated P -premotivic categories(5.1.24.2) a ∗ t : D(PSh( P , Λ)) ⇄ D(Sh t ( P , Λ)) : R a t, ∗ . Note that a ∗ t = L a ∗ t on objects, because the functor a ∗ t is exact. Remark also that this adjunction extends on S -diagrams considering the situation described in 5.1.20: forany diagram X : I → S , we get an adjunction L ϕ X : D( A )( X ) ⇄ D( B )( X ) : R ψ X and this defines a morphism of triangulated monoidal P cart -fibred categories over the S -diagrams ( cf. Proposition3.1.32). . FIBRED DERIVED CATEGORIES 125
Example . Consider an admissible topology t . Let ϕ : Λ → Λ ′ be a morphism of rings.For any scheme S , it induces a pair of adjoint functors:(5.1.25.1) ϕ ∗ : Sh t ( P S , Λ) ⇆ Sh t ( P S , Λ ′ ) : ϕ ∗ such that ϕ ∗ (resp. ϕ ∗ ) is induced by the obvious extension (resp. restriction) of scalars functor.By definition, for any P -scheme X/S , the functor ϕ ∗ sends the representable sheaf of Λ-modulesΛ tS ( X ) to the representable sheaf of Λ ′ -modules Λ ′ tS ( X ). Thus ( ϕ ∗ , ϕ ∗ ) defines an adjunction ofabelian P -premotivic categories. Applying the results of Paragraph 5.1.23, one deduces a P -premotivic adjunction: L ϕ ∗ : D(Sh t ( P , Λ)) ⇆ D(Sh t ( P , Λ ′ )) : R ϕ ∗ . The functor ϕ ∗ is exact so that R ϕ ∗ = ϕ ∗ . Similarly when Λ ′ / Λ is flat, L ϕ ∗ = ϕ ∗ .The following result can be used to check the compatibility to a given admissible topology: Proposition . Let t be an admissible topology. Consider a morphism of abelian P -premotivic categories ϕ : A ⇄ B : ψ such that:(a) For any scheme S , ψ S is exact.(b) The morphism ϕ induces an isomorphism of the underlying set of twists of A and B .According to the last property, we identify the set of twists of A and B to a monoid τ in such away that ϕ acts on τ by the identity.Assume that A is compatible with t . Then the following conditions are equivalent:(i) B is compatible with t .(ii) B satisfies cohomological t -descent, Proof.
The fact ( i ) implies ( ii ) is clear from the definition and we prove the converse usingthe following lemma : Lemma . Consider a morphism of P -premotivic abelian categories ϕ : A ⇄ B : ψ satisfying conditions (a) and (b) of the above proposition and a base scheme S .Given a simplicial P -scheme X over S , a twist i ∈ τ and a complex C over B S , we denoteby ǫ X ,i,C : Hom C( B S ) (cid:0) M S ( X , B ) { i } , C (cid:1) → Hom C( A S ) (cid:0) M S ( X , A ) { i } , ψ S ( C ) (cid:1) the adjunction isomorphism obtained for the adjoint pair ( ϕ S , ψ S ) .Then there exists a unique isomorphism ǫ ′X ,i,C making the following diagram commutative: Hom C( B S ) (cid:0) M S ( X , B ) { i } , C (cid:1) ǫ X ,i,C / / (cid:15) (cid:15) Hom C( A S ) (cid:0) M S ( X , A ) { i } , ψ S ( C ) (cid:1) (cid:15) (cid:15) Hom K( B S ) (cid:0) M S ( X , B ) { i } , C (cid:1) ǫ ′X ,i,C / / Hom K( A S ) (cid:0) M S ( X , A ) { i } , ψ S ( C ) (cid:1) . Assume moreover that B satisfies cohomological t -descent.Then there exists an isomorphism ǫ ′′X ,i,C making the following diagram commutative: Hom K( B S ) (cid:0) M S ( X , B ) { i } , C (cid:1) ǫ ′X ,i,C / / π B X ,i,C (cid:15) (cid:15) Hom K( A S ) (cid:0) M S ( X , A ) { i } , ψ S ( C ) (cid:1) π A X ,i,C (cid:15) (cid:15) Hom D( B S ) (cid:0) M S ( X , B ) { i } , C (cid:1) ǫ ′′X ,i,C / / Hom D( A S ) (cid:0) M S ( X , A ) { i } , ψ S ( C ) (cid:1) , (5.1.27.1) where π A X ,i,C and π B X ,i,C are induced by the obvious localization functors. The existence and unicity of isomorphism ǫ ′X ,i,C follows from the fact that the functors ϕ S and ψ S are additive. Indeed, this implies that the isomorphism ǫ X ,i,C is compatible with chainhomotopies.Consider the injective model structure on C( A S ) and C( B S ) (see for example [ CD09 , 1.2]for the definition). We first treat the case when C is fibrant for this model structure on C( B S ).Because the premotive M S ( X , B ) { i } is cofibrant for the injective model structure, we obtain thatthe canonical map π B X ,i,C is an isomorphism. This implies there exists a unique map ǫ ′′X ,i,C makingdiagram (5.1.27.1) commutative. On the other hand, the isomorphism ǫ ′X ,i,C obtained previouslyis obviously functorial in X . Thus, because B satisfies t -descent, we obtain that ψ S ( C ) is t -flasque. Because A is compatible with t , this implies ψ S ( C ) is t -local, and because M S ( X , B ) { i } is cofibrant for the t -descent model structure on C( A S ), this implies π B X ,i,C is an isomorphism.Thus finally, ǫ ′′X ,i,C is an isomorphism as required.To treat the general case, we consider a fibrant resolution C → D for the injective modelstructure on C( B S ). Because ψ S is exact, it preserves isomorphisms. Using the previous case, Wedefine ǫ ′′X ,i,C by the following commutative diagram:Hom D( B S ) (cid:0) M S ( X , B ) { i } , C (cid:1) ǫ ′′X ,i,C / / ∼ (cid:15) (cid:15) Hom D( A S ) (cid:0) M S ( X , A ) { i } , ψ S ( C ) (cid:1) ∼ (cid:15) (cid:15) Hom D( B S ) (cid:0) M S ( X , B ) { i } , D (cid:1) ǫ ′′X ,i,D / / Hom D( A S ) (cid:0) M S ( X , A ) { i } , ψ S ( D ) (cid:1) . The required property for ǫ ′′X ,i,C then follows easily and the lemma is proved.To finish the proof that (ii) implies (i), we note the lemma immediately implies, under (ii),that the following two conditions are equivalent : • C is t -flasque (resp. local) in C( B S ); • ψ S ( C ) is t -flasque (resp. local) in C( A S ).This concludes. (cid:3) Constructible premotivic complexes.
Definition . Let A be an abelian P -premotivic category compatible with an ad-missible topology t . We will say that t is bounded in A if for any scheme S , there exists anessentially small family N tS of bounded complexes which are direct factors of finite sums of objectsof type M S ( X ) { i } in each degree, such that, for any complex C of A S , the following conditionsare equivalent.(i) C is t -flasque.(ii) For any H in N tS , the abelian group Hom K ( A S ) ( H, C ) vanishes.In this case, we say the family N tS is a bounded generating family for t -hypercoverings in A S . Example . (1) Assume P contains the open immersions so that the Zariski topol-ogy is admissible. Let M V S to be the family of complexes of the formΛ S ( U ∩ V ) l ∗ − k ∗ −−−−→ Λ S ( U ) ⊕ Λ S ( V ) i ∗ + j ∗ −−−−→ Λ S ( X )for any open cover X = U ∪ V , where i , j , k , l denotes the obvious open immersions. It fol-lows then from [ BG73 ] that
M V S is a bounded generating family of Zariski hypercoversin Sh Zar ( P /S, Λ).(2) Assume P contains the ´etale morphisms so that the Nisnevich topology is admissible.We let BG S be the family of complexes of the formΛ S ( W ) g ∗ − l ∗ −−−−→ Λ S ( U ) ⊕ Λ S ( V ) j ∗ + f ∗ −−−−→ Λ S ( X ) . FIBRED DERIVED CATEGORIES 127 for a Nisnevich distinguished square in S ( cf. W l / / g (cid:15) (cid:15) V f (cid:15) (cid:15) U j / / X. Then, by applying 3.3.2, we see that BG S is a bounded generating family for Nisnevichhypercovers in Sh Nis ( P /S, Λ).(3) Assume that P = S ft is the class of morphisms of finite type in S . We let P CDH S be the family of complexes of the formΛ S ( T ) g ∗ − k ∗ −−−−→ Λ S ( Z ) ⊕ Λ S ( Y ) i ∗ + f ∗ −−−−→ Λ S ( X )for a cdh-distinguished square in S ( cf. T k / / g (cid:15) (cid:15) Y f (cid:15) (cid:15) Z i / / X. Then, by virtue of 3.3.8,
CDH S = BG S ∪ P CDH S is a bounded generating family forcdh-hypercovers in Sh cdh (cid:0) S ft /S, Λ (cid:1) .(4) The ´etale topology is not bounded in Sh ´et ( Sm , Λ) for an arbitray ring Λ. However, ifΛ = Q , it is bounded: by virtue of Theorem 3.3.23, a bounded generating family for´etale hypercovers in Sh ´et ( Sm , Q ) S is the union of the class BG S and that of complexesof the form Q S ( Y ) G → Q S ( X ) for any Galois cover Y → X of group G .(5) As in the case of ´etale topology, the qfh-topology is not bounded in general, but it is sowith rational coefficients. Let P QF H S be the family of complexes of the form Q S ( T ) G g ∗ − k ∗ −−−−→ Q S ( Z ) ⊕ Q S ( Y ) G i ∗ + f ∗ −−−−→ Q S ( X )for a qfh-distinguished square of group G in S ( cf. T k / / g (cid:15) (cid:15) Y f (cid:15) (cid:15) Z i / / X. Then, by virtue of Theorem 3.3.25,
QF H S = P QF H S ∪ BG S is a bounded generatingfamily for qfh-hypercovers in Sh qfh (cid:0) S ft /S, Q (cid:1) .(6) Similarly, by Theorem 3.3.30, H S = CDH S ∪ QF H S is a bounded generating family forh-hypercovers in Sh h (cid:0) S ft /S, Q (cid:1) . Proposition . Let A be an abelian P -premotivic category compatible with an admis-sible topology t . We make the following assumptions:(a) t is bounded in A ;(b) for any P -morphism X → S and any n ∈ τ , the functor Hom A S ( M S ( X ) { n } , − ) pre-serves filtered colimits.Then t -local complexes are stable by filtering colimits. Proof.
Let N tS is a bounded generating family for t -hypercovers in A S . Then a complex C of A S is t -flasque if and only if for any H ∈ N tS , the abelian group Hom K ( A S ) ( H, C ) is trivial.Hence it is sufficient to prove that the functor C Hom K ( A S ) ( H, C )preserves filtering colimits of complexes. This will follow from the fact that the functor C Hom C( A S ) ( H, C )preserves filtering colimits. As H a is bounded complex that is degreewise compact, this latterproperty is obvious. (cid:3) . Consider an abelian P -premotivic category A compatible with an admissible topology t , with generating set of twists τ . Assume that t is bounded in A and consider a boundedgenerating family N tS for t -hypercovers in A S .Let M ( P /S, A ) be the full subcategory of A S spanned by direct factors of finite sums ofpremotives of shape M S ( X ) { i } for a P -scheme X/S and a twist i ∈ τ . This category is additiveand we can associate with it its category of complexes up to chain homotopy. We get an obvioustriangulated functor(5.1.31.1) K b (cid:0) M ( P /S, A ) (cid:1) → D( A S ) . Then the previous functor induces a triangulated functor K b (cid:0) M ( P /S, A ) (cid:1) / N tS → D( A S )where the left hand side stands for the Verdier quotient of K b (cid:0) M ( P /S, A ) (cid:1) by the thick subcat-egory generated by N tS .The category K b (cid:0) M ( P /S, A ) (cid:1) / N tS may not be pseudo-abelian while the aim of the previousfunctor is. Thus we can consider its pseudo-abelian envelope and the induced functor(5.1.31.2) (cid:16) K b (cid:0) M ( P /S, A ) (cid:1) / N tS (cid:17) ♮ → D( A S ) . According to Definition 1.4.9, the image of this functor is the subcategory of τ -constructiblepremotives of the triangulated P -premotivic category D( A S ). Then the following proposition isa corollary of [ CD09 , theorem 6.2] :
Proposition . Consider the hypothesis and notations above.If A is finitely τ -presented then D( A ) is compactly τ -generated. Moreover, the functor (5.1.31.2) is fully faithful. Let us denote by D c ( A ) the subcategory of D( A ) made of τ -constructible premotives in thesense of Definition 1.4.9. Taking into account Porposition 1.4.11, the previous proposition admitsthe following corollary: Corollary . Consider the situation of 5.1.31, and assume that A is finitely τ -presented.For any premotive M in D( A S ) , the following conditions are equivalent:(i) M is compact.(ii) M is τ -constructible.Moreover, the functor (5.1.31.2) induces an equivalence of categories: (cid:16) K b (cid:0) M ( P /S, A ) (cid:1) / N tS (cid:17) ♮ → D c ( A S ) . Example . According to example 5.1.29, we get the following examples:(1) Let Λ( Sm /S ) = M ( Sm /S, A ) for A = Sh Nis ( Sm /S, Λ). We obtain a fully faithfulfunctor (cid:16) K b (Λ( Sm /S )) /BG S (cid:17) ♮ → D (cid:0) Sh Nis ( Sm /S, Λ) (cid:1) . which is essentially surjective on compact objects.(2) Let Λ( S ft /S ) = M ( Sm /S, A ) for A = Sh cdh (cid:0) S ft /S, Λ (cid:1) . We obtain a fully faithfulfunctor (cid:16) K b (cid:0) Λ( S ft /S ) (cid:1) /BG S ∪ CDH S (cid:17) ♮ → D (cid:16) Sh cdh ( S ft /S, Λ) (cid:17) . which is essentially surjective on compact objects.(3) Let Q ´et ( Sm /S ) = M ( Sm /S, A ) for A = Sh ´et ( Sm /S, Q ). We obtain a fully faithfulfunctor (cid:16) K b ( Q ´et ( Sm /S )) /BG S (cid:17) ♮ → D (cid:0) Sh ´et ( Sm /S, Q ) (cid:1) . which is essentially surjective on compact objects.5.1.35 . Consider an abelian P -premotivic category A . We introduce the following propertyof A : . FIBRED DERIVED CATEGORIES 129 (C) Consider a projective system ( S α ) α ∈ A of schemes in S with affine transition maps suchthat S = lim ←− α ∈ A S α belongs to S . For any index α ∈ A , any object A α in A S α , andany twist n ∈ τ , the canonical maplim −→ α ∈ A/α Hom A Sα ( S α { n } , A α ) → Hom A S ( S { n } , A )is an isomorphism where A α (resp. A ) is the pullback of A α along the canonical map S α → S α (resp. S → S α ). Proposition . Consider an abelian P -premotivic category A compatible with an ad-missible topology t and satisfying the assumption (C) above.Then the derived premotivic category D( A ) is τ -continuous. Proof.
We use Proposition 4.3.6 applied to the t -descent model structure on C( A T ) for T = S or T = S α . (see Paragraph 5.1.13). Recall from Paragraph 5.1.11 that this model structureis associated with a descent structure. Thus according to [ CD09 , 2.3], there exist an explicitgenerating set I (resp. J ) for cofibrations (resp. trivial cofibrations). Moreover, the source ortarget of any map in I ∪ J is a complex C satisfying the following assumption:(rep) for any integer i ∈ Z , C i is a sum of premotives of the form M T ( X ) { n } where X/T is a P -scheme and n ∈ τ .Thus, to check the assumption of 4.3.6 for C( A ), we fix a projective system ( S α ) α ∈ A satisfyingthe assumptions of property (C) above; we have to prove that for any index α ∈ A and anycomplexes C α and E α such that C α satisfies (rep), the natural map:lim −→ α ∈ A/α Hom C( A Sα ) ( C α , E α ) → Hom C( A S ) ( C, E )is bijective.Given the definition of morphisms in a category of complexes, it is sufficient to check thiswhen the Hom groups are computed as morphisms of Z -graded objects. Thus it is sufficient totreat the case where C α and E α are concentrated in degree 0. Thus, as C α satisfies property(rep), we are exactly reduced to assumption (C) on A . (cid:3) Example . (1) Assume P is contained in the class of morphisms of finite type.Then the abelian P -premotivic category PSh( P , Λ) of example 5.1.1 satisfies as-sumption (C). Indeed, property (C) when A is a representable presheaf follows from theassumption on P : P -schemes over some base S always are of finite presentation over S – S is noetherian according to our general assumption 5.0. Then the case of a gen-eral presheaf A follows because A is an inductive limit of representable presheaf and theglobal sections functor commutes with inductive limit of presheaves.(2) Let S ft be the class of morphisms of finite type and let t be one of the followingtopologies: Nis , ´et , cdh , qfh , h.Then the generalized abelian premotivic category Sh t (cid:0) S ft , Λ (cid:1) of example 5.1.4 satisfiesassumption (C).Indeed, according to the preceding example, we have only to prove that for anymorphism f : X → S , the functor f ∗ : PSh( S ftS , Λ) → PSh( S ftT , Λ)preserves the property of being a t -sheaf.If f is a morphism of finite type, the functor f ∗ admits as a left adjoint the functor f ♯ , which preserves t -covers. Thus the assertion is clear in that case.In the general case, we use the fact that X/S is a projective limit of a projectivesystem ( X α ) α ∈ A where X α is an S -scheme affine and of finite type over S . To check thatfor a t -sheaf F over S , the presheaf f ∗ ( F ) is a t -sheaf, we fix a t -cover ( W i ) i ∈ I of X in S ftX . As X is noetherian, we can assume I is finite. Moreover, there exists an index α ∈ A such that for the t -cover ( W i ) i ∈ I can be lifted to X α . Then, using property (C) of PSh( S ft , Λ) applied to F and ( X α ), we reduce to check that f ∗ α ( F ) is a t -sheaf for α ≥ α . This follows from the first case treated.(3) Let Sm be the class of smooth morphisms and t be one of the topologies: Nis , ´et.As we will see in Example 6.1.1, there exists a canonical enlargement of abelian premotiviccategories (see (6.1.1.1)): ρ ♯ : Sh t ( Sm , Λ) ⇄ Sh t (cid:0) S ft , Λ (cid:1) : ρ ∗ . As the functor ρ ♯ is fully faithful and commutes with f ∗ for any morphism of schemes f , we deduce from the preceding point that the abelian premotivic category Sh t ( Sm , Λ)satisfies the above condition (C).As an application of the previous proposition, we thus obtain that the derived pre-motivic category D(Sh t ( Sm , Λ)) is τ -continuous. -derived premotivic category. Localization of triangulated premotivic categories. . Let A be an abelian P -premotivic category compatible with an admissible topology t and D( A ) be the associated derived P -premotivic category.Suppose given an essentially small family of morphisms W in C( A ) which is stable by theoperations f ∗ , f ♯ (in other words, W is a sub- P -fibred category of C( A )). Remark that thelocalizing subcategory T of D( A ) generated by the cones of arrows in W is again stable by theseoperations. Moreover, as for any P -morphism f : X → S we have f ♯ f ∗ = M S ( X ) ⊗ S ( − ), thecategory T is stable by tensor product with a geometric section.We will say that a complex K over A S is W -local if for any object T of T and any integer n ∈ Z , Hom D( A S ) ( T, K [ n ]) = 0. A morphism of complexes p : C → D over A S is a W -equivalence if for any W -local complex K over A S , the induced mapHom D( A S ) ( D, K ) → Hom D( A S ) ( C, K )is bijective.A morphism of complexes over A S is called a W -fibration if it is a t -fibration with a W -localkernel. A complex over A S will be called W -fibrant if it is t -local and W -local.As consequence of [ CD09 , 4.3, 4.11 and 5.6], we obtain :
Proposition . Let A be an abelian P -premotivic category compatible with an admissibletopology t and W be an essentially small family of morphisms in C( A ) stable by f ∗ and f ♯ .Then the category C( A S ) is a proper closed symmetric monoidal category with the W -fibrationsas fibrations, the cofibrations as defined in 5.1.11, and the W -equivalences as weak equivalences. The homotopy category associated with this model category will be denoted by D( A S )[ W − S ].It can be described as the Verdier quotient D( A S ) / T S .In fact, the W -local model category on C( A S ) is nothing else than the left Bousfield localizationof the t -local model category structure. As a consequence, we obtain an adjunction of triangulatedcategories:(5.2.2.1) π S : D( A S ) ⇄ D( A S )[ W − S ] : O S such that O S is fully faithful with essential image the W -local complexes. In fact, the modelstructure gives a functorial W -fibrant resolution 1 → R W R W : C( A S ) → C( A S ) , which induces O S .Note that the triangulated category D( A S )[ W − S ] is generated by the complexes concentrated indegree 0 of the form M S ( X ) { i } – or, equivalently, the W -local complexes R W ( M S ( X ) { i } ) – for a P -scheme X and a twist i ∈ τ . Remark . Another very useful property is that W -equivalences are stable by filteringcolimits; see [ CD09 , prop. 3.8]. . FIBRED DERIVED CATEGORIES 131 . Recall from 5.1.14 that for any morphism (resp. P -morphism) f : T → S , the functor f ∗ (resp. f ♯ ) satisfies descent; as it also preserves W , it follows from [ CD09 , 4.9] that theadjunction f ∗ : C( A S ) → C( A T ) : f ∗ (resp. f ♯ : C( A S ) → C( A T ) : f ∗ )is a Quillen adjunction with respect to the W -local model structures. This gives the followingcorollary. Corollary . The P -fibred category C( A ) with the W -local model structure on its fibersdefined above is a monoidal P -fibred model category, which is moreover stable, proper and combi-natorial. We will denote by D( A )[ W − ] the triangulated P -premotivic category whose fiber over ascheme S is the homotopy category of the W S -local model category C( A S ). The adjunction(5.2.2.1) readily defines an adjuntion of triangulated P -premotivic categories(5.2.5.1) π : D( A ) ⇄ D( A )[ W − ] : O . The P -fibred categories D( A ) and D( A )[ W − ] are both τ -generated (and this adjunction iscompatible with τ -twists in a strong sense). Remark . For any scheme S , the category D( A S )[ W − S ] is well generated and has acanonical dg-structure (see also 5.1.19).5.2.7 . With the notations above, let us put T = D( A )[ W − ] to clarify the following notations.As in 5.1.20, the fibred category T has a canonical extension to S -diagrams X : I → S .If we define W X as the class of morphisms ( f i ) i ∈ I in C( A ( X , I )) such that for any object i , f i is a W -equivalence, then T ( X ) is the triangulated category D( A ( X , I ))[ W − X ].Again, this triangulated category is symmetric monoidal closed and for any morphism (resp. P -morphism) ϕ : ( X , I ) → ( Y , J ), we get (derived) adjunctions as in 5.1.20: L ϕ ∗ : T ( Y , J ) ⇄ T ( X , I ) : R ϕ ∗ (5.2.7.1) (resp. L ϕ ♯ : T ( X , I ) ⇄ T ( Y , J ) : L ϕ ∗ = ϕ ∗ )(5.2.7.2)In fact, T is then a complete monoidal P cart -fibred category over the category of diagrams ofschemes and the adjunction (5.2.5.1) extends to an adjunction of complete monoidal P cart -fibredcategories. Example . Suppose we are under the hypothesis of example 5.1.24.2.Let W t,S denote the family of maps which are of the form Λ S ( X ) → Λ S ( X ) for a t -hypercover X → X in P /S . Then W t is obviously stable by f ∗ and f ♯ .Recall now that a complex of t -sheaves on P /S is local if and only if its t -hypercohomologyand its hypercohomology computed in the coarse topology agree ( cf. a ∗ t : D(PSh( P , Λ)) ⇄ D(Sh t ( P , Λ)) : R a t, ∗ . induces an equivalence of triangulated P -premotivic categoriesD(PSh( P , Λ))[ W − t ] ⇄ D(Sh t ( P , Λ)) . Recall R a t, ∗ is fully faithful and identifies D(Sh t ( S, Λ)) with the full subcategory of D(PSh( S, Λ))made by t -local complexes.5.2.9 . A triangulated P -premotivic category ( T , M ) such that there exists:(1) an abelian P -premotivic category A compatible with an admissible topology t on Sm .(2) an essentially small family W of morphisms in C( A ) stable by f ∗ and f ♯ (3) an adjunction of triangulated P -premotivic categories D( A )[ W − ] ≃ T will be called for short a derived P -premotivic category . According to convention 5.0(d) and fromthe above construction, T is τ -generated for some set of twists τ . Let us denote simply by M S ( X ) the geometric sections of T . In this case, using the morphisms(5.1.24.1) and (5.2.5.1), we get a canonical morphism of triangulated P -premotivic categories:(5.2.9.1) ϕ ∗ : D(PSh( P , Z )) ⇄ T : ϕ ∗ . By definition, for any premotive M , any scheme X and any integer n ∈ Z , we get a canonicalidentification:(5.2.9.2) Hom T ( S ) ( M S ( X ) , M [ n ]) = H n Γ( X, ϕ ∗ ( M )) . Given any simplicial scheme X , we put M S ( X ) = ϕ ∗ (cid:0) Z S ( X ) (cid:1) , so that we also obtain:(5.2.9.3) Hom T ( S ) ( M S ( X ) , M [ n ]) = H n (cid:0) Tot π Γ( X , R γ ∗ ( M )) (cid:1) . Proposition . Consider the above notations and t an admissible topology. The follow-ing conditions are equivalent.(i) For any t -hypercover X → X in P /S , the induced map M S ( X ) → M S ( X ) is an isomor-phism in T ( S ) .(i ′ ) For any t -hypercover p : X → X in P /S , the induced functor L p ∗ : T ( X ) → T ( X ) isfully faithful.(i ′′ ) T satisfies t -descent.(ii) There exists an essentially unique map ϕ ∗ t : D(Sh t ( P /S, Z )) → T ( S ) making the fol-lowing diagram essentially commutative: D(PSh( P /S, Z )) ϕ ∗ / / a t (cid:15) (cid:15) T ( S )D(Sh t ( P /S, Z )) ϕ ∗ t ❦❦❦❦❦❦❦❦❦❦❦ (ii ′ ) For any complex C ∈ C(PSh( P /S, Z )) such that a t ( C ) = 0 , ϕ ∗ ( C ) = 0 .(ii ′′ ) For any map f : C → D in C(PSh( P /S, Z )) such that a t ( f ) is an isomorphism, ϕ ∗ ( f ) is an isomorphism.(iii) There exists an essentially unique map ϕ t ∗ : T ( S ) → D(Sh t ( P /S, Z )) making the fol-lowing diagram essentially commutative: D(PSh( P /S, Z )) T ( S ) ϕ ∗ o o ϕ t ∗ u u ❦❦❦❦❦❦❦❦❦❦❦ D(Sh t ( P /S, Z )) R O t O O (iii ′ ) For any premotive M in T ( S ) , the complex ϕ ∗ ( M ) is local.(iii ′′ ) For any premotive M in T ( S ) , any P -scheme X/S and any integer n ∈ Z , Hom T ( S ) ( M S ( X ) , M [ n ]) = H nt ( X, ϕ ∗ ( M )) . When these conditions are fulfilled for any scheme S , the functors appearing in (ii) and (iii) inducea morphism of triangulated P -premotivic categories: ϕ ∗ t : D(Sh t ( P , Z )) ⇄ T : ϕ t ∗ . Proof.
The equivalence between conditions ( i ), ( i ′ ) and ( i ′′ ) is clear (we proceed as in theproof of 5.1.21). The equivalences ( ii ) ⇔ ( ii ′ ) ⇔ ( ii ′′ ) and ( iii ) ⇔ ( iii ′ ) follows from example 5.2.8and the definition of a localization. The equivalence ( i ) ⇔ ( ii ′′ ) follows again from loc. cit. Theequivalences ( i ) ⇔ ( iii ′ ) ⇔ ( iii ′′ ) follows finally from (5.2.9.2), (5.2.9.3), and the characterisationof a local complex of sheaves ( cf. (cid:3) We will formulate in some remarks below universal properties of some derived P -premotivic categories.When doing so, we will restrict to morphisms of derived P -premotivic categories which can be written as L ϕ : D( A )[ W − ] → D( A )[ W − ]for a morphism ϕ : A → A of abelian P -premotivic categories compatible with suitable topologies. More naturaluniversal properties could be obtained if one considers the framework of dg-categories or triangulated derivator. . FIBRED DERIVED CATEGORIES 133 Remark . The preceding proposition express the fact that the category D(Sh t ( P , Z ))is the universal derived P -premotivic category satisfying t -descent.5.2.12 . We end this section by making explicit two particular cases of the descent propertyfor derived P -premotivic categories.Consider a derived P -premotivic category T with geometric sections M . Considering anydiagram X : I → P /S of P -schemes over S , with projection p : X → S , we can associate apremotive in T : M S ( X ) = L p ♯ ( S ) = L lim −→ i ∈ I M S ( X i ) . In particular, when I is the category • → • , we associate to every S -morphism f : Y → X of P -schemes over S a canonical bivariant premotive M S ( X f −→ Y ) . When f is an immersion, we will also write M S ( Y /X ) for this premotive. Note that in any case,there is a canonical distinguished triangle in T ( S ): M S ( X ) f ∗ −→ M S ( Y ) π f −−→ M S ( X f −→ Y ) ∂ f −→ M S ( X )[1] . This triangle is functorial in the arrow f – with respect to commutative squares.Given a commutative square of P -schemes over SB e ′ / / g (cid:15) (cid:15) Y f (cid:15) (cid:15) A e / / X (5.2.12.1)we will say that the image square in T ( S ) M S ( B ) e ′∗ / / g ∗ (cid:15) (cid:15) M S ( Y ) f ∗ (cid:15) (cid:15) M S ( A ) e ∗ / / M S ( X )is homotopy cartesian if the premotive associated with diagram 5.2.12.1 is zero. Proposition . Consider a derived P -premotivic category T . We assume that P contains the ´etale morphisms (resp. P = S ft ). Then, with the above definitions, the followingconditions are equivalent:(i) T satisfies Nisnevich (resp. proper cdh ) descent.(ii) For any scheme S and any Nisnevich (resp. proper cdh ) distinguished square Q of S -schemes, the square M S ( Q ) is homotopy cartesian in T ( S ) .(iii) For any Nisnevich (resp. proper cdh ) distinguished square of shape (5.2.12.1) , the canon-ical map M S ( Y /B ) ( f/g ) ∗ −−−−→ M S ( X/A ) is an isomorphism.Moreover, under these conditions, to any Nisnevich (resp. proper cdh ) distinguished square Q ofshape (5.2.12.1) , we associate a map ∂ Q : M S ( X ) π e −→ M S ( X/A ) ( f/g ) − ∗ −−−−−→ M S ( Y /B ) ∂ e ′ −−→ M S ( Y )[1] which defines a distinguished triangle in T ( S ) : M S ( B ) (cid:18) e ′∗ − g ∗ (cid:19) −−−−→ M Z ( Y ) ⊕ M S ( A ) ( f ∗ ,e ∗ ) −−−−→ M S ( X ) ∂ Q −−→ M S ( Y )[1] . In fact, if T = D( A )[ W − ] for an abelian P -premotivic category A , then we can define M S ( X → Y ) asthe cone of the morphism of complexes (concentrated in degree 0) M S ( X ) f ∗ −−→ M S ( Y ). If T = D( A )[ W − ], this amount to say that the diagram obtained of complexes by applying the functor M S ( − ) is homotopy cartesian in the W -local model category C( A ). Proof.
The equivalence of (i) and (ii) follows from the theorem of Morel-Voevodsky 3.3.2(resp. the theorem of Voevodsky 3.3.8). To prove the equivalence of (ii) and (iii), we assume T = D( A )[ W − ]. Then, the homotopy colimit of a square of shape 5.2.12.1 is given by thecomplex Cone (cid:0) Cone( M S ( B ) → M S ( Y )) → Cone( M S ( A ) → M S ( X )) (cid:1) . This readily proves the needed equivalence, together with the remaining assertion. (cid:3)
Remark . In the first of the respective cases of the proposition, condition (ii) is whatwe usually called the
Brown-Gersten property (BG) for T , whereas condition (iii) can be calledthe excision property . In the second respective case, condition (ii) will be called the proper cdhproperty for the generalized premotivic category T . We say also that T satisfies the (cdh) propertyif it satisfies condition (ii) with respect to any cdh distinguished square Q .5.2.b. The homotopy relation. . Let A be an abelian P -premotivic category compatible with an admissible topology t . We consider W A to be the family of morphisms M S ( A X ) { i } → M S ( X ) { i } for a P -scheme X/S and an twist i in τ . The family W A is obviously stable by f ∗ and f ♯ . Definition . Let A be an abelian P -premotivic category compatible with an admis-sible topology t . With the notation above, we define D eff A ( A ) = D( A )[ W − A ] and refer to it as the(effective) P -premotivic A -derived category with coefficients in A .By definition, the category D eff A ( A ) satisfies the homotopy property (Htp) (see 2.1.3). Ac-cording to the general facts about localization of derived premotivic categories, the triangulatedpremotivic category D eff A ( A ) is τ -generated. Example . We can divide our examples into two types:1) Assume P = Sm :Consider the admissible topology t = Nis. Following F. Morel, we define the (effective) A -derived category over S to be D eff A (Sh Nis ( Sm /S, Λ)). Indeed we get a triangulated premotiviccategory (see also the construction of [
Ayo07b ]):(5.2.17.1) D eff A , Λ := D eff A (Sh Nis ( Sm , Λ)) . We shall also write its fibres(5.2.17.2) D eff A ( S, Λ) := D eff A , Λ ( S ) = D eff A (Sh Nis ( Sm /S, Λ))for a scheme S . For Λ = Z , we shall often write simply(5.2.17.3) D eff A := D eff A (Sh Nis ( Sm , Z )) . Another interesting case is when t = ´et; we get a triangulated premotivic category of effective´etale premotives : D eff A (Sh ´et ( Sm , Λ)) . In each of these cases, we denote by Λ tS ( X ) the premotive associated with a smooth S -scheme X .2) Assume P = S ft :Consider the admissible topology t = h (resp. t = qfh). In [ Voe96 ], Voevodsky has introducedthe category of h-motives (resp. qfh-motives). In our formalism, one defines the category of effective h -motives (resp. effective h -motives ) over S with coefficients in Λ as: DM eff h ( S, Λ) = D eff A (cid:0) Sh h (cid:0) S ft /S, Λ (cid:1)(cid:1) , resp. DM eff qfh ( S, Λ) = D eff A (cid:0) Sh qfh (cid:0) S ft /S, Λ (cid:1)(cid:1) . In other words, this is the A -derived category of h-sheaves (resp. qfh-sheaves) of Λ-modules.Moreover, these categories for various schemes S are the fibers of a generalized premotivic . FIBRED DERIVED CATEGORIES 135 triangulated category. What we have added to the construction of Voevodsky is the functors ofthe generalized premotivic structure.We will denote simply by Λ tS ( X ) the corresponding premotive associated with X in DM eff t ( S, Λ).Another interesting case is obtained when t = cdh. We get an A -derived generalized pre-motivic category D eff A (cid:0) Sh cdh (cid:0) S ft , Λ (cid:1)(cid:1) whose premotives are simply denoted by Λ cdh S ( X ) for anyfinite type S -scheme X .5.2.18 . Let C be a complex with coefficients in A S . According to the general case, we say that C is A -local if for any P -scheme X/S and any ( i, n ) ∈ τ × Z , the map induced by the canonicalprojection Hom D( A S ) ( M S ( X ) { i } [ n ] , C ) → Hom D( A S ) ( M S ( A X ) { i } [ n ] , C )is an isomorphism. The adjunction (5.2.2.1) defines a morphism of triangulated P -premotiviccategories D( A ) ⇄ D eff A ( A )such that for any scheme S , D eff A ( A S ) is identified with the full subcategory of D( A S ) made of A -local complexes.Fibrant objects for the model category structure on C( A S ) appearing in Proposition 5.2.2relatively to W A , simply called A -fibrant objects, are the t -flasque and A -local complexes.We say a morphism f : C → D of complexes of A S is an A -equivalence if it becomes anisomorphism in D eff A ( A S ). Considering moreover two morphisms f, g : C → D of complexes of A S , we say they are A -homotopic if there exists a morphism of complexes H : M S ( A S ) ⊗ S C → D such that H ◦ ( s ⊗ C ) = f and H ◦ ( s ⊗ C ) = g , where s and s are respectively inducedby the zero and the unit section of A S /S . When f and g are A -homotopic, they are equal asmorphisms of D eff A ( A S ). We say the morphism p : C → D is a strong A -equivalence if thereexists a morphism q : D → C such that the morphisms p ◦ q and q ◦ p are A -homotopic to theidentity. A complex C is A -contractible if the map C → A -equivalence.As an example, for any integer n ∈ N , and any P -scheme X/S , the map p ∗ : M S ( A nX ) → M S ( X )induced by the canonical projection is a strong A -equivalence with inverse the zero section s , ∗ : M S ( X ) → M S ( A nX ).5.2.19 . The category D eff A ( A ) is functorial in A .Let ϕ : A ⇄ B : ψ be an adjunction of abelian P -premotivic categories. Consider twotopologies t and t ’ such that t ′ is finer than t . Suppose A (resp. B ) is compatible with t (resp. t ′ ). For any scheme S , consider the evident extensions ϕ S : C( A S ) ⇄ C( B S ) : ψ S of the aboveadjoint functors to complexes. We easily check that the functor ψ S preserves A -local complexes.Thus, applying 5.1.23, the pair ( ϕ S , ψ S ) is a Quillen adjunction for the respective A -localizedmodel structure on C( A S ) and C( B S ); see [ CD09 , 3.11]. Considering the derived functors, it isnow easy to check we have obtained an adjunction L ϕ : D eff A ( A ) ⇄ D eff A ( B ) : R ψ of triangulated P -premotivic categories. Example . Consider the notations of 5.2.17. In the case where P = Sm , we get fromthe adjunction of (5.1.24.2) the following adjunction of triangulated premotivic categories a ∗ ´et : D eff A , Λ ⇄ D eff A (Sh ´et ( Sm , Λ)) : R a ´et , ∗ . Example . Let T be a derived P -premotivic category as in 5.2.9. If T satisfies theproperty (Htp), then the canonical morphism (5.2.9.1) induces a morphismD eff A (PSh( P , Z )) ⇄ T . If moreover T satisfies t -descent for an admissible topology t , we further obtain as in 5.2.10 amorphism D eff A (Sh t ( P , Z )) ⇄ T . Particularly interesting cases are given by D eff A (resp. D eff A (cid:0) Sh cdh (cid:0) S ft , Z (cid:1)(cid:1) ) which is the univer-sal derived premotivic category (resp. generalized premotivic category), i.e. initial premotiviccategory satisfying Nisnevich descent (resp. cdh descent) and the homotopy property.5.2.22 . As in Example 5.1.25, let t be an admissible topology and ϕ : Λ → Λ ′ be an extensionof rings. Then, from the P -premotivic adjunction (5.1.25.1) and according to Paragraph 5.2.19,we get an adjunction of triangulated P -premotivic categories: L ϕ ∗ : D eff A (cid:0) Sh t ( P , Λ) (cid:1) ⇆ D eff A (cid:0) Sh t ( P , Λ ′ ) (cid:1) : R ϕ ∗ . Consider also complexes C and D of t -sheaves of Λ-modules over P S . Then there exists a canonicalmorphism of Λ ′ -modules:(5.2.22.1) Hom D eff A (Sh t ( P S , Λ)) (cid:0)
C, D (cid:1) ⊗ Λ Λ ′ −→ Hom D eff A (Sh t ( P S , Λ ′ )) (cid:0) L ϕ ∗ ( C ) , L ϕ ∗ ( D ) (cid:1) There are two notable cases where this map is an isomorphism:
Proposition . Consider the above assumptions. Then the map (5.2.22.1) is an iso-morphism in the two following cases:(1) If Λ ′ is a free Λ -module and C is compact;(2) If Λ ′ is a free Λ -module of finite rank. Proof.
Note that in any case, the functor ϕ ∗ admits a right adjoint ϕ ! . We can assume that Λ ′ = I. Λ for a set I . In this case, we get for any sheaf F of Λ-modules: ϕ ∗ ϕ ∗ ( F ) = F ⊗ Λ Λ ′ = I.F.
Moreover, for any P -scheme X/S , we get: ϕ ∗ (Λ ′ tS ( X )) = Λ ′ tS ( X ) = I. Λ tS ( X ) . In particular, the functor ϕ ∗ : C(Sh t ( P S , Λ ′ )) → C(Sh t ( P S , Λ)) satisfies descent in the senseof [
CD09 , 2.4] and preserves the family W A . Thus it is a left Quillen functor with respect tothe A -local model structures. In particular, because it is also a right Quillen functor, we get: R ϕ ∗ = ϕ ∗ = L ϕ ∗ . In particular, we get in D eff A (Sh t ( P S , Λ)): R ϕ ∗ L ϕ ∗ ( D ) = L ϕ ∗ L ϕ ∗ ( D ) = L ( ϕ ∗ ϕ ∗ )( D ) = I.D.
Thus the Proposition follows as the functor Hom( C, − ) commutes with direct sums if C is compactand with finite direct sums in any case. (cid:3) We remark the following useful property.
Proposition . Consider a morphism ϕ ∗ : A ⇄ B : ϕ ∗ of abelian P -premotivic categories such that A (resp. B ) is compatible with an admissible topology t (resp. t ′ ). Assume t ′ is finer than t .Let S be a base scheme. Assume that ϕ ∗ : A S → B S commutes with colimits . Then ϕ ∗ : C( A S ) → C( B S ) respects A -equivalences. In other words, the right derived functor R ϕ ∗ : D eff A ( B S ) → D eff A ( A S ) satisfies the relation R ϕ ∗ = ϕ ∗ . It is defined by the formula: ϕ ! ( F ) = Hom Λ (Λ ′ , F )equipped with its canonical structure of sheaf of Λ ′ -modules. This amounts to ask that ϕ ∗ is exact and commutes with direct sums. . FIBRED DERIVED CATEGORIES 137 Proof.
In this proof, we write ϕ ∗ for ϕ ∗ ,S . We first prove that ϕ ∗ preserves strong A -equivalences (see 5.2.18).Consider two maps u, v : K → L in C( B S ). To give an A -homotopy H : M S ( A S , B ) ⊗ S K → L between u and v is equivalent by adjunction to give a map H ′ : K → Hom B S ( M S ( A S , B ) , L )which fits into the following commutative diagram: K H ′ (cid:15) (cid:15) u v v ♥♥♥♥♥♥♥♥♥♥♥♥♥♥♥♥♥♥♥ v ( ( PPPPPPPPPPPPPPPPPPP L Hom B S ( M S ( A S , B ) , L ) s ∗ o o s ∗ / / L where s and s are the respective zero and unit section of A S /S .Because M S ( A S , B ) = ϕ ∗ S ( M S ( A S , A )), we get a canonical isomorphism (see paragraph1.2.9) ϕ ∗ ( Hom B S ( M S ( A S , B ) , L )) ≃ Hom B S ( M S ( A S , A ) , ϕ ∗ ( L )) . Thus, applying ϕ ∗ to the previous commutative diagram and using this identification, we obtainthat ϕ ∗ ( u ) is A -homotopic to ϕ ∗ ( v ).As a consequence, for any P -scheme X over S , and any B -twist i , the map ϕ ∗ ( M S ( A X , B ) { i } ) → ϕ ∗ ( M S ( X, B ) { i } )induced by the canonical projection is a strong A -equivalence, thus an A -equivalence.The functor ϕ ∗ : B S → A S commutes with colimits. Thus it admits a right adjoint that wewill denote by ϕ ! . Consider the injective model structure on C( A S ) and C( B S ) (see [ CD09 , 2.1]).Because ϕ ∗ is exact, it is a left Quillen functor for these model structures. Thus, the right derivedfunctor R ϕ ! is well defined. From the result we just get, we see that R ϕ ! preserves A -localobjects, and this readily implies L ϕ ∗ = ϕ ∗ preserves A -equivalences. (cid:3) . To relate the category D eff A ( S ) with the homotopy category of schemes of Morel andVoevodsky [ MV99 ], we have to consider the category of simplicial Nisnevich sheaves of setsdenoted by ∆ op Sh( Sm /S ). Considering the free abelian sheaf functor, we obtain an adjunctionof categories ∆ op Sh( Sm /S ) ⇄ C(Sh( Sm /S, Z )) . If we consider Blander’s projective A -model structure [ Bla03 ] on the category ∆ op Sh( Sm /S ),we can easily see that this is a Quillen pair, so that we obtain a P -premotivic adjunction of simple P -premotivic categories N : H ⇄ D eff A : K. Note that the functor N sends cofiber sequences in H ( S ) to distinguished triangles in D eff A ( S ).5.2.c. Explicit A -resolution. . Consider an abelian P -premotivic category A compatible with an admissible topology t .Consider the canonically split exact sequence0 → S s −→ M S ( A S ) → U → s : S → M S ( A S ) is induced by the zero section of A . The section correspondingto 1 in A defines another map s : S → M S ( A S )which does not factor through s , so that we get canonically a non trivial map u : S → U . Thisdefines for any complex C of A S a map, called the evaluation at Hom ( U, C ) = S ⊗ S Hom ( U, C ) u ⊗ −−−→ U ⊗ S Hom ( U, C ) ev −→ C. We define the complex R (1) A ( C ) to be R (1) A ( C ) = Cone (cid:0) Hom ( U, C ) → C (cid:1) . We have by construction a map r C : C → R (1) A ( C ) . This defines a morphism of functors from the identity functor to R (1) A . For an integer n ≥
1, wedefine by induction a complex R ( n +1) A ( C ) = R (1) A ( R ( n ) A ( C )) , and a map r R ( n ) A ( C ) : R ( n ) A ( C ) → R ( n +1) A . We finaly define a complex R A ( C ) by the formula R A ( C ) = lim −→ n R ( n ) A ( C ) . We have a functorial map C → R A ( C ) . Lemma . With the above hypothesis and notations, the map C → R A ( C ) is an A -equivalence. Proof.
For any closed symmetric monoidal category C and any objects A , B , C and I in C ,we have Hom( I ⊗ Hom ( B, C ) , Hom ( A, C )) = Hom(
Hom ( B, C ) , Hom ( I, Hom ( A, C )))= Hom(
Hom ( B, C ) , Hom ( I ⊗ A, C )) . Hence any map I ⊗ A → B induces a map I ⊗ Hom ( B, C ) → Hom ( A, C ) for any object C . Ifwe apply this to C = C( A S ) and I = M S ( A ), we see immediately that the functor Hom ( − , C )preserves strong A -homotopy equivalences. In particular, for any complex C , the map C → Hom ( M S ( A X ) , C ) is a strong A -homotopy equivalence. This implies that Hom ( U, C ) → A -equivalence, so that the map r C is an A -equivalence as well. As A -equivalences are stableby filtering colimits, this implies our result. (cid:3) Proposition . Consider the above notations and hypothesis, and assume that t isbounded in A .For any t -flasque complex C of A S , the complex R A ( C ) is t -flasque and A -local. Moreover,the morphism C → R A ( C ) is an A -equivalence. If furthermore C is t -flasque, so is R A ( C ) . Proof.
The last assertion is a particular case of Lemma 5.2.27. The functor R (1) A preserves t -flasque complexes. By virtue of 5.1.30, the functor R A has the same gentle property. It thusremains to prove that the functor R A sends t -flasque complexes on A -local ones. We shall usethat the derived category D( A S ) is compactly generated; see 5.1.30.Let C be a t -flasque complex of A S . To prove R A ( C ) is A -local, we are reduced to provethat the map R A ( C ) → Hom ( M S ( A X ) , R A ( C ))is a quasi-isomorphism, or, equivalently, that the complex Hom ( U, R A ( C )) is acyclic. As U is adirect factor of M S ( A X , A ), for any P -scheme X over S and any i in I , the object Z S ( X ; A ) { i }⊗ S U is compact. This implies that the canonical maplim −→ n Hom ( U, R ( n ) A ( C )) → Hom ( U, R A ( C ))is an isomorphism of complexes. As filtering colimits preserve quasi-isomorphisms, the complex Hom ( U, R A ( C )) (resp. R A ( C )) can be considered as the homotopy colimit of the complexes Hom ( U, R ( n ) A ( C )) (resp. R ( n ) A ( C )). In particular, for any compact object K of D( A S ), the canonicalmorphisms lim −→ n Hom( K, Hom ( U, R ( n ) A ( C ))) → Hom( K, Hom ( U, R A ( C )))lim −→ n Hom(
K, R ( n ) A ( C )) → Hom(
K, R A ( C )) . FIBRED DERIVED CATEGORIES 139 are bijective.By construction, we have distinguished triangles Hom ( U, R ( n ) A ( C )) → R ( n ) A ( C ) → R ( n +1) A ( C ) → Hom ( U, R ( n ) A ( C ))[1] . This implies that the evaluation at 1 morphism ev : Hom ( U, R A ( C )) → R A ( C )induces the zero mapHom D( A S ) ( K, Hom ( U, R A ( C ))) → Hom D( A S ) ( K, R A ( C ))for any compact object K of D( A S ). Hence the induced map a = Hom ( U, ev ) : Hom ( U, Hom ( U, R A ( C ))) → Hom ( U, R A ( C ))has the same property: for any compact object K , the mapHom D( A S ) ( K, Hom ( U, Hom ( U, R A ( C )))) → Hom D( A S ) ( K, Hom ( U, R A ( C )))is zero.The multiplication map A × A → A induces a map µ : U ⊗ S U → U such that the composition of µ ∗ : Hom ( U, R A ( C )) → Hom ( U ⊗ S U, R A ( C )) = Hom ( U, Hom ( U, R A ( C )))with a is the identity of Hom ( U, R A ( C )). As D( A S ) is compactly generated, this implies that Hom ( U, R A ( C )) = 0 in the derived category D( A S ). (cid:3) Remark . Consider a t -flasque resolution functor ( i.e. a fibrant resolution for the t -localmodel structure) R t : C( A S ) → C( A S ), 1 → R t . As a corollary of the proposition, the compositefunctor R A ◦ R t is a resolution functor by t -local and A -local complexes. Example . Consider an admissible topology t and the P -premotivic A -derived cate-gory D = D eff A (Sh t ( P , Λ)). Suppose that t is bounded for abelian t -sheaves (for example, this isthe case for the Zariski and the Nisnevich topologies, see 5.1.29).Let C be a complex of abelian t -sheaves on P /S . If C is A -local, thenHom D ( S ) (Λ tS ( X ) , C ) = H nt ( X ; C )(this is true without any condition on t ).Consider a t -local resolution C t of C in C (cid:0) Sh t ( P /S, Λ) (cid:1) . Then we get the following formula:Hom D ( S ) (cid:0) Λ tS ( X ) , C [ n ] (cid:1) = H n (cid:0) Γ (cid:0) X, R A ( C t ) (cid:1)(cid:1) . Corollary . Consider a morphism of abelian P -premotivic categories ϕ : A ⇄ B : ψ Suppose there are admissible topologies t and t ′ , with t ′ finer than t , such that the followingconditions are verified.(i) A is compatible with t and B is compatible with t ′ .(ii) B and D( B ) are compactly τ -generated.(iii) For any scheme S , the functor ψ S : B S → A S preserves filtering colimits.Then, ψ S : C( B S ) → C( A S ) preserves A -equivalences between t ′ -flasque objects. If moreover ψ S is exact, the functor ψ S preserves A -equivalences. Proof.
We already know that ψ S is a right Quillen functor, so that it preserves local objectsand A -fibrant objects. This implies also that ψ S preserves A -equivalences between A -fibrantobjects (this is Ken Brown’s lemma [ Hov99 , 1.1.12]). Let D be a t ′ -flasque complex of B S . Then ψ S ( D ) is a t -flasque complex of A S . It follows from Proposition 5.2.28 that R A ( D ) is A -localand that D → R A ( D ) is an A -equivalence. Lemma 5.2.27 implies the map ψ S ( D ) → R A ( ψ S ( D )) = ψ S ( R A ( D )) is a an A -equivalence. This implies the first assertion.The last assertion is a direct consequence of the first one. (cid:3) . Consider the usual cosimplicial scheme ∆ • defined by∆ n = Spec ( Z [ t , . . . , t n ] / ( t + · · · + t n − ≃ A n (see [ MV99 ]). For any scheme S , we get a cosimplicial object of A S , namely M S (∆ • S ). Given anycomplex C of A S , we define its associated Suslin singular complex as(5.2.32.1) C ∗ ( C ) = Tot ⊕ Hom ( M S (∆ • S ) , C ) , where Hom ( M S (∆ • S ) , C ) is considered as a bicomplex by the Dold-Kan correspondence. Thecanonical map M S (∆ • S ) → S induces a map C → C ∗ ( C ) . Lemma . For any complex C of A S , the map C ∗ ( C ) → Hom ( M S ( A S ) , C ∗ ( C )) = C ∗ ( Hom ( M S ( A S ) , C )) is a chain homotopy equivalence. Proof.
The composite morphism( s p × Id ) ∗ : M S ( A × ∆ • S ) → M S ( A × ∆ • S ) , where s is the map induced by the zero section, and p is the map induced by the obvious projectionof A on its base, is chain homotopic to the identity. Indeed, the homotopy relation is given bythe formula s n = n X i =0 ( − i . (1 ⊗ S ψ i ) : M S ( A × ∆ n +1 S ) → M S ( A × ∆ nS )where 1 is the identity of M S ( A S ), and ψ i is induced by the map ∆ n +1 S → A × ∆ nS which sendsthe j -th vertex v j,n +1 to either 0 × v j,n , if j ≤ i , or to 1 × v j − ,n otherwise. This implies thelemma. (cid:3) Lemma . For any t -flasque complex C of A S , we have a canonical isomorphism C ∗ ( C ) ≃ L lim −→ n R Hom ( M S (∆ nS ) , C ) in D( A S ) . This is a variation on the Dold-Kan correspondence. As a direct consequence, we get:
Lemma . For any complex C of A S , the map C → C ∗ ( C ) is an A -equivalence. Proposition . If t is bounded in A , then, for any t -flasque complex C of A S , C ∗ ( C ) is A -local. Proof.
Using the first premotivic adjunction of example 5.2.21 and the fact that D( A ) iscompactly generated (5.1.30), we can reduce the proposition to the case where A S is the categoryof presheaves of abelian groups over P /S , in which case this is well known. (cid:3) Constructible A -local premotives. . Consider an abelian P -premotivic category A compatible with an admissible topology t . Assume that t is bounded in A (see Definition 5.1.28) and consider a bounded generating family N tS for t -hypercovers in A S .Let T A S be the family of complexes of C( A S ) of shape M S ( A X ) { i } → M S ( X ) { i } for a P -scheme X over S and a twist i ∈ I . Then the functor (5.1.31.1) obviously induces thefollowing functor(5.2.37.1) (cid:16) K b (cid:0) M ( P /S, A ) (cid:1) / N tS ∪ T A S (cid:17) ♮ → D eff A ( A S ) , . FIBRED DERIVED CATEGORIES 141 where the category on the left is the pseudo-abelian category associated to the Verdier quotientof K b (cid:0) M ( P /S, A ) (cid:1) by the thick subcategory generated by N tS ∪ T A S . Applying Thomason’slocalization theorem [ Nee01 ], we get from Proposition 5.1.32 the following result:
Proposition . Consider the previous hypothesis and notations and assume that A isfinitely τ -presented.Then D eff A ( A ) is compactly τ -generated. Moreover, the functor (5.2.37.1) is fully faithful. Let us denote by D eff A ,c ( A ) the subcategory of D eff A ( A ) made of τ -constructible premotives inthe sense of Definition 1.4.9. Taking into account Proposition 1.4.11, we deduce from the aboveproposition the following corollary: Corollary . Under the assumptions of 5.2.38, for any premotive M in D eff A ( A S ) , thefollowing conditions are equivalent:(i) M is compact;(ii) M is τ -constructible.Moreover, the functor (5.2.37.1) induces an equivalence of categories: (cid:16) K b (cid:0) M ( P /S, A ) (cid:1) / N tS ∪ T A S (cid:17) ♮ → D eff A ,c ( A S ) . Example . With the notations of 5.1.34, we get the following equivalences of categories: (cid:16) K b (Λ( Sm /S )) / ( BG S ∪ T A S ) (cid:17) ♮ → D eff A ,c ( S, Λ) . (cid:16) K b (cid:0) Λ( S ft /S ) (cid:1) /CDH S ∪ T A S (cid:17) ♮ → D eff A ,c (cid:16) Sh cdh ( S ft /S, Λ) (cid:17) . This statement is the analog of the embedding theorem [
VSF00 , chap. 5, 3.2.6].
Proposition . Assume P = S ft is the class of finite type (resp. separated and offinite type) morphisms.Let A be an abelian generalized premotivic category compatible with an admissible topology t and satisfying the property (C) of Paragraph 5.1.35.Then the triangulated generalized premotivic category D eff A ( A ) is τ -continuous. Proof.
The proof relies on the following lemma:
Lemma . Under the assumptions of the preceding proposition, for any morphism ofschemes f : T → S , the functor L f ∗ : D( A S ) → D( A T ) preserves A -local complexes. When f is a morphism of finite type (resp. separated of finite type), the functor L f ∗ admits L f ♯ as a left adjoint and the lemma is clear. In the general case, one can write f as a projectivelimit of a projective system of morphisms of scheme ( f α : T α → S ) α ∈ A such that f α is affine offinite type. Recall from Proposition 5.1.36, D( A ) is τ -continuous. Thus, to check that for an A -local complexe C in D( A S ), the complex L f ∗ ( C ) is A -local, we thus are reduced to provethat L f ∗ α ( C ) is A -local which follows from the first treated case. The lemma is proven.Given the full embedding D eff A ( A ) → D( A ) whose image is made of A -local complexes, theproposition now directly follows from the previous lemma and the fact D( A ) is τ -continuous. (cid:3) Example . Taking into account the second point of Example 5.1.37, the previous propo-sition can be applied to the category Sh t (cid:0) S ft , Z (cid:1) where t = Nis , ´et , cdh , qfh , h. Remark . The previous proposition will be extended to the (non generalized) premotiviccase in Corollary 6.1.12. -derived premotivic category. Modules.
Let A be an abelian P -premotivic category with generating set of twists τ .A cartesian commutative monoid R of A is a cartesian section of the fibred category A over S such that for any scheme S , R S has a commutative monoid structure in A S and for any morphismof schemes f : T → S , the structural transition maps φ f : f ∗ ( R S ) → R T are isomorphisms ofmonoids.Let us fix a cartesian commutative monoid R of A .Consider a base scheme S . We denote by R S - mod the category of modules in the monoidalcategory A S over the monoid R S . For any P -scheme X/S and any twist i ∈ τ , we put R S ( X ) { i } = R S ⊗ S M S ( X ) { i } endowed with its canonical R S -module structure. The category R S - mod is a Grothendieck abeliancategory such that the forgetful functor U S : R S - mod → A S is exact and conservative. A familyof generators for R S - mod is given by the modules R S ( X ) { i } for a P -scheme X/S and a twist i ∈ τ . As A S is commutative, R S - mod has a unique symmetric monoidal structure such that thefree R S -module functor is symmetric monoidal. We denote by ⊗ R this tensor product. Note that R S ( X ) ⊗ R R S ( Y ) = R S ( X × S Y ). Finally the categories of modules R S - mod form a symmetricmonoidal P -fibred category, such that the following proposition holds (see 7.2.10). Proposition . Let A be a τ -generated abelian P -premotivic category and R be a carte-sian commutative monoid of A .Then the category R - mod equipped with the structures introduced above is a τ -generated abelian P -premotivic category.Moreover, we have an adjunction of abelian P -premotivic categories: (5.3.1.1) R ⊗ ( − ) : A ⇄ R - mod : U .
Remark . With the hypothesis of the preceding proposition, for any morphism of schemes f : T → S , the exchange transformation f ∗ U S → U T f ∗ is an isomorphism by construction of R - mod (7.2.10). Proposition . Let A be a τ -generated abelian P -premotivic category compatible with anadmissible topology t . Consider a cartesian commutative monoid R of A such that for any scheme S , tensoring quasi-isomorphisms between cofibrant complexes by R S gives quasi-isomorphisms (e.g. R S might be cofibrant (as a complex concentrated in degree zero), or flat). Then the abelian P -premotivic category R - mod is compatible with t . Proof.
In view of Proposition 5.1.26, we have only to show that R - mod satisfies coho-mological t -descent. Consider a t -hypercover p : X → X in P /S . We prove that the map p ∗ : R S ( X ) → R S ( X ) is a quasi-isomorphism in C( R S - mod). The functor U S is conservative, and U S ( p ∗ ) is equal to the map: R S ⊗ S M S ( X ) → R S ⊗ S M S ( X ) . But this is a quasi-isomorphism in C( A S ) by assumption on R S . (cid:3) Remark . According to Lemma 5.1.27, for any simplicial P -scheme X over S , any twist i ∈ τ and any R S -module C , we get canonical isomorphisms:Hom K( R S - mod) (cid:0) R S ( X ) { i } , C (cid:1) ≃ Hom K( A S ) ( M S ( X ) { i } , C )(5.3.4.1) Hom D( R S - mod) ( R S ( X ) { i } , C ) ≃ Hom D( A S ) ( M S ( X ) { i } , C ) . (5.3.4.2)5.3.b. Symmetric sequences.
Let A be an abelian category.Let G be a group. An action of G on an object A ∈ A S is a morphism of groups G → Aut A ( A ) , g γ Ag . We say that A is a G -object of A . A G -equivariant morphism A f −→ B of G -objects of A is a morphism f in A such that γ Bg ◦ f = f ◦ γ Ag .If E is any object of A , we put G × E = L g ∈ G E considered as a G -object via the permutationisomorphisms of the summands.If H is a subgroup of G , and E is an H -object, G × E has two actions of H : the first one,say γ , is obtained via the inclusion H ⊂ G , and the second one denoted by γ ′ , is obtained using . FIBRED DERIVED CATEGORIES 143 the structural action of H on E . We define G × H E as the coequalizer of the family of morphisms( γ σ − γ ′ σ ) σ ∈ H , and consider it equipped with its induced action of G . Definition . Let A be an abelian category.A symmetric sequence of A is a sequence ( A n ) n ∈ N such that for each n ∈ N , A n is a S n -object of A . A morphism of symmetric sequences of A is a collection of S n -equivariant morphism( f n : A n → B n ) n ∈ N .We let A S be the category of symmetric sequences of A .It is straightforward to check A S is abelian. For any integer n ∈ N , we define the n -thevaluation functor as follows: ev n : A S → A , A ∗ A n . Any object A of A can be considered as the trivial symmetric sequence ( A, , . . . ). The functor i : A ( A, , . . . ) is obviously left adjoint to ev and we obtain an adjunction(5.3.5.1) i : A ⇄ A S : ev . Remark i is also right adjoint to ev . Thus i preserves every limits and colimits.For any integer n ∈ N and any symmetric sequence A ∗ of A , we put( A ∗ {− n } ) m = (cid:26) S m × S m − n A m − n if m ≥ n A S , and we have A ∗ {− n }{− m } = A ∗ {− n − m } (through acanonical isomorphism). Remark finally that for any integer n ∈ N , the functor i n : A → A S , A ( i ( A )) {− n } is left adjoint to ev n . Remark . Let S be the category of finite sets with bijective maps as morphisms. Thenthe category of symmetric sequences is canonically equivalent to the category of functors S → A .This presentation is useful to define a tensor product on A S . Definition . Let A be a symmetric closed monoidal abelian category.Given two functors A ∗ , B ∗ : S → A , we put: E ⊗ S F : S A N L N = P ⊔ Q E ( P ) ⊗ F ( Q ) . If A is the unit object of the monoidal category A , the category A S is then a symmetricclosed monoidal category with unit object i ( A ).5.3.8 . Let A be an object of A . Then the n -th tensor power A ⊗ n of A is endowed with acanonical action of the group S n through the structural permutation isomorphism of the symmetricstructure on A . Thus the sequence Sym( A ) = ( A ⊗ n ) n ∈ N is a symmetric sequence.Moreover, the isomorphism A ⊗ n ⊗ A ⊗ m → A ⊗ n + m is S n × S m -equivariant. Thus it inducesa morphism µ : Sym( A ) ⊗ S Sym( A ) → Sym( A ) of symmetric sequences. We also consider theobvious morphism η : i ( A ) = i ( A ⊗ ) → Sym( A ). One can check easily that Sym( A ) equippedwith the multiplication µ and the unit η is a commutative monoid in the monoidal category A S . Definition . Let A be an abelian symmetric monoidal category. The commutativemonoid Sym( A ) of A S defined above will be called the symmetric monoid generated by A . Remark . One can describe Sym( A ) by a universal property: given a commutativemonoid R in A S , to give a morphism of commutative monoids Sym( A ) → R is equivalent to givea morphism A → R in A . . Consider an abelian P -premotivic category A .Consider a base scheme S . According to the previous paragraph, the category A S S is anabelian category, endowed with a symmetric tensor product ⊗ S S . For any P -scheme X/S and anyinteger n ∈ N , using (5.3.5.2), we put M S ( X, A S ) {− n } = i ( M S ( X, A )) {− n } . It is immediate that the class of symmetric sequences of the form M S ( X, A S ) {− n } for a smooth S -scheme X and an integer n ≥ A S S which istherefore a Grothendieck abelian category. It is clear that for any P -scheme X and Y over S , M S ( X, A S ) {− n } ⊗ S S M S ( Y, A S ) {− n } = M S ( X × S Y, A S ) {− n } . Given a morphism (resp. P -morphism) of schemes f : T → S and a symmetric sequence A ∗ of A S , we put f ∗ S ( A ∗ ) = ( f ∗ A n ) n ∈ N (resp. f S ♯ ( A ∗ ) = ( f ♯ A n ) n ∈ N ). This defines a functor f ∗ S : A S S → A S T (resp. f S ♯ : A S T → A S S ) which is obviously right exact. Thus the functor f ∗ S admits a right adjoint which we denote by f S ∗ . When f is in P , we check easily the functor f S ♯ is left adjoint to f ∗ S .From criterion 1.1.42 and Lemma 1.2.13, we check easily the following proposition: Proposition . Consider the previous hypothesis and notations.The association S A S S together with the structures introduced above defines an N × τ -generated abelian P -premotivic category.Moreover, the different adjunctions of the form (5.3.5.1) over each fibers over a scheme S define an adjunction of P -premotivic categories: (5.3.12.1) i : A ⇄ A S : ev Indeed, i is trivially compatible with twists. Proposition . Let A be an abelian P -premotivic category, and t be an admissibletopology. If A is compatible with t then A S is compatible with t . Proof.
This is based on the following lemma (see [
CD09 , 7.5, 7.6]):
Lemma . For any complex C of A S , any complex E of A S S and any integer n ≥ ,there are canonical isomorphisms: Hom K( A S S ) ( i ( C ) {− n } , E ) ≃ Hom K( A S ) ( C, E n )(5.3.14.1) Hom D( A S S ) ( i ( C ) {− n } , E ) ≃ Hom D( A S ) ( C, E n )(5.3.14.2)If A is compatible with t , this implies that E is local (resp. t -flasque) if and only if for any n ≥ E n is local (resp. t -flasque). This concludes. (cid:3) Symmetric Tate spectra. . Consider an abelian P -premotivic category A .For any scheme S , the unit point of G m,S defines a split monomorphism of A -premotives S → M S ( G m,S ). We denote by S { } the cokernel of this monomorphism and call it the suspended Tate S -premotive with coefficients in A . The collection of these objects for any scheme S is a cartesian section of A denoted by { } . For any integer n ≥
0, we denote by { n } its n -the tensor power.With the notations of 5.3.9, we define the symmetric Tate spectrum over S as the symmetricsequence S {∗} = Sym ( S { } ) in A S S . The corresponding collection defines a cartesian commu-tative monoid of the fibred category A S , called the absolute Tate spectrum . Definition . Consider an abelian P -premotivic category A .We denote by Sp( A ) the abelian P -premotivic category of modules over {∗} in the category A S . The objects of Sp( A ) are called the abelian (symmetric) Tate spectra. As we will almost never consider non symmetric spectra, we will cancel the word ”symmetric” in ourterminology. . FIBRED DERIVED CATEGORIES 145
The category Sp( A ) is ( N × τ )-generated. Composing the adjunctions (5.3.1.1) and (5.3.12.1),we get an adjuntion(5.3.16.1) Σ ∞ : A ⇄ Sp( A ) : Ω ∞ of abelian P -premotivic categories.Let us explicit the definition. An abelian Tate spectrum ( E, σ ) is the data of :(1) for any n ∈ N , an object E n of A S endowed with an action of S n (2) for any n ∈ N , a morphism σ n : E n { } → E n +1 in A S such that the composite map E m { n } σ m { n − } −−−−−−→ E m +1 { n − } → ... σ m + n − −−−−−→ E m + n is S n × S m -equivariant with respect to the canonical action of S n on S { n } and the structuralaction of S m on E m . By definition, ev ( E ) = E . Recall that ev is exact.Given an object A of A S , the abelian Tate spectrum Σ ∞ A is defined such that (Σ ∞ A ) n = A { n } with the action of S n given by its action on S { n } by permutations of the factors.Be careful we consider the category Sp( A S ) as N -twisted by negative twists. For any abelianTate spectrum E ∗ , ( E ∗ {− n } ) m = S n × S m − n E m − n for n ≥ m .5.3.17 . Consider a morphism ϕ : A → B of abelian P -premotivic categories. Then as ϕ ( A { } ) = B { } , ϕ can be extended to abelianTate spectra in such a way that the following diagram commutes: A ϕ / / Σ ∞ A (cid:15) (cid:15) B Σ ∞ B (cid:15) (cid:15) Sp( A ) Sp( ϕ ) / / Sp( B ) . (Of course the obvious diagram for the corresponding right adjoints also commutes.) Definition . For any scheme S , a complex of abelian Tate spectra over S will be calledsimply a Tate spectrum over S .A Tate spectrum E is a bigraded object. In the notation E mn , the index m corresponds to the(cochain) complex structure and the index n to the symmetric sequence structure.From propositions 5.3.3 and 5.3.13, we get the following: Proposition . Let A be an abelian P -premotivic category compatible with an admis-sible topology t . Then Sp( A ) is compatible with t . Note also that remark 5.3.4 and Lemma 5.3.14 implies that for any simplicial P -scheme X over S , any integer n ∈ N , and any Tate spectrum E , we have canonical isomorphisms:Hom K(Sp( A S )) (Σ ∞ M S ( X , A ) {− n } , E ) ≃ Hom K( A S ) (Σ ∞ M S ( X , A ) , E n )(5.3.19.1) Hom D(Sp( A S )) (Σ ∞ M S ( X , A ) {− n } , E ) ≃ Hom D( A S ) (Σ ∞ M S ( X , A ) , E n )(5.3.19.2)According to the proposition, the category C(Sp( A S )) of Tate spectra over S has a t -descent modelstructure. The previous isomorphisms allow to describe this structure as follows:(1) For any simplicial P -scheme X over S , and any integer n ≥
0, the Tate spectrumΣ ∞ M S ( X , A ) {− n } is cofibrant.(2) A Tate spectrum E over S is fibrant if and only if for any integer n ≥
0, the complex E n over A S is local ( i.e. t -flasque).(3) Let f : E → F be a morphism of Tate spectra over S . Then f is a fibration (resp.quasi-isomorphism) if and only if for any integer n ≥
0, the morphism f n : E n → F n ofcomplexes over A S is a fibration (resp. quasi-isomorphism).Note that properties (2) and (3) follows from (5.3.4.1) and (5.3.14.1). . We can also introduce the A -localization of this model structure. The correspondinghomotopy category is the A -derived P -premotivic category D eff A (Sp( A )) introduced in 5.2.16.The isomorphism (5.3.19.2) gives the following assertion: From the above, a Tate spectrum E is A -local if and only if for any integer n ≥ E n is A -local.(1) A Tate spectrum E over S is A -local if and only if for any integer n ≥
0, the complex E n over A S is A -local.(2) Let f : E → F be a morphism of Tate spectra over S . Then f is a A -local fibration (resp.weak A -equivalence) if and only if for any integer n ≥
0, the morphism f n : E n → F n of complexes over A S is a A -local fibration (resp. weak A -equivalence).As a consequence, the isomorphism (5.3.19.2) induces an isomorphismHom D eff A (Sp( A S )) (Σ ∞ M S ( X , A ) {− n } , E ) ≃ Hom D eff A ( A S ) (Σ ∞ M S ( X , A ) , E n ) . (5.3.20.1)Similarly, the adjunction (5.3.16.1) induces an adjunction of triangulated P -premotivic categories(5.3.20.2) L Σ ∞ : D eff A ( A ) ⇄ D eff A (Sp( A )) : R Ω ∞ . Symmetric Tate Ω -spectra. . The final step is to localize further the category D eff A (Sp( A )). The aim is to relatethe positive twists on D eff A ( A ) obtained by tensoring with S { } and the negative twists onD eff A (Sp( A )) induced by the consideration of symmetric sequences.Let X be a P -scheme over S . From the definition of Σ ∞ , there is a canonical morphism ofabelian Tate spectra: (cid:2) Σ ∞ (cid:0) S { } (cid:1)(cid:3) {− } → Σ ∞ S . Tensoring this map by Σ ∞ M S ( X, A ) {− n } for any P -scheme X over S and any integer n ∈ N ,we obtain a family of morphisms of Tate spectra concentrated in cohomological degree 0: (cid:2) Σ ∞ (cid:0) M S ( X, A ) { } (cid:1)(cid:3) {− n − } → Σ ∞ M S ( X, A ) {− n } . We denote by W Ω this family and put W Ω , A = W Ω ∪ W A . Obviously, W Ω , A is stable by theoperations f ∗ and f ♯ . Definition . Let A be an abelian P -premotivic category compatible with an admissi-ble topology t . With the notations introduced above, we define the stable A -derived P -premotiviccategory with coefficients in A as the derived P -premotivic categoryD A ( A ) := D(Sp( A ))[ W − , A ]defined in Corollary 5.2.5.5.3.23 . According to this definition, we get the following identification:D A ( A ) = D eff A (Sp( A ))[ W − ] . Using the left Bousfield localization of the A -local model structure on C(Sp( A )), we thus obtaina canonical adjunction of triangulated P -fibred premotivic categoriesD eff A (Sp( A )) ⇄ D eff A (Sp( A ))[ W − ]which allows to describe D A ( A S ) as the full subcategory of D eff A (Sp( A S )) made of Tate spectrawhich are W Ω -local in D eff A (Sp( A S )). Recall a Tate spectrum E is a sequence of complexes ( E n ) n ∈ N over A S together with suspension maps in C( A S ) σ n : S { } ⊗ E n → E n +1 . From this, we deduce a canonical morphism S { } ⊗ L E n → E n +1 in D eff A ( A ) whose adjointmorphism we denote by(5.3.23.1) u n : E n → R Hom D eff A ( A S ) ( S { } , E n +1 )According to (5.3.20.1), the condition that E is W Ω -local in D eff A (Sp( A )) is equivalent to ask thatfor any integer n ≥
0, the map (5.3.23.1) is an isomorphism in D eff A (Sp( A )). . FIBRED DERIVED CATEGORIES 147 Considering the adjunction (5.3.20.2), we obtain finally an adjunction of triangulated P -fibredcategories:(5.3.23.2) Σ ∞ : D eff A ( A ) ⇄ D eff A (Sp( A )) ⇄ D A ( A ) : Ω ∞ . Note that tautologically, the Tate spectrum Σ ∞ ( S { } ) has a tensor inverse given by the spectrum(Σ ∞ S ) {− } in D A ( A S ). Thus, we have obtained from the abelian premotivic category A atriangulated premotivic category D A ( A S ) which satisfies the properties: • the homotopy property (Htp); • the stability property (Stab); • the t -descent property.As we will see in the followings, the construction satisfies a universality property that the readercan already guess. Definition . Consider the assumptions of definition 5.3.22.For any scheme S , we say that a Tate spectrum E over S is a Tate Ω -spectrum if the followingconditions are fulfilled:(a) For any integer n ≥ E n is t -flasque and A -local.(b) For any integer n ≥
0, the adjoint of the structural suspension map E n → Hom C( A S ) ( S { } , E n +1 )is a quasi-isomorphism.In particular, a Tate Ω-spectrum is W Ω -local in D eff A (Sp( A S )). In fact, it is also W Ω , A -localin the category D(Sp( A S )) so that the category D A ( A ) is also equivalent to the full subcategoryof D(Sp( A S )) spanned by Tate Ω-spectra.Fibrant objects of the W Ω , A -local model category on C(Sp( A )) obtained in definition 5.3.22are exactly the Tate Ω-spectra. Proposition . Consider the above notations. Let S be a base scheme.(1) If the endofunctor D eff A ( A S ) → D eff A ( A S ) , C R Hom D eff A ( A S ) ( S { } , C ) is conservative, then the functor Ω ∞ S is conservative.(2) If the Tate twist E E (1) is fully faithful in D eff A ( A S ) , then Σ ∞ S is fully faithful.(3) If the Tate twist E E (1) induces an auto-equivalence of D eff A ( A S ) , then (Σ ∞ S , Ω ∞ S ) areadjoint equivalences of categories. Remark . Similar statements can be obtained for the derived categories rather thanthe A -derived categories. We left their formulation to the reader. Proof.
Consider point (1). We have to prove that for any W Ω -local Tate spectrum E inD eff A (Sp( A S )), if R Ω ∞ ( E ) = 0, then E = 0. But R Ω ∞ ( E ) = Ω ∞ ( E ) = E (see 5.3.20). Becausefor any integer n ≥
0, the map (5.3.23.1) is an A -equivalence, we deduce that for any integer n ∈ Z , the complex E n is (weakly) A -acyclic. According to (5.3.20.1), this implies E = 0 –because D A ( A S ) is N -generated.Consider point (2). We want to prove that for any complex C over A S , the counit map C → R Ω ∞ L Σ ∞ ( C ) is an isomorphism. It is enough to treat the case where C is cofibrant.Considering the left adjoint L Σ ∞ of (5.3.20.2), we first prove that L Σ ∞ ( C ) is W Ω -local.Because C is cofibrant, this Tate spectrum is equal in degree n to the complex C { n } (with itsnatural action of S n ). Moreover, the suspension map is given by the isomorphism (in the monoidalcategory C( A S )) σ n : S { } ⊗ S C { n } → C { n + 1 } . In particular, the corresponding map in D eff A ( A S ) σ ′ n : S { } ⊗ LS C { n } → C { n + 1 } . is canonically isomorphic to S { } ⊗ LS C { n } ⊗ −−→ S { } ⊗ LS C { n } . Thus, because the Tate twist is fully faithful in D eff A ( A S ), the adjoint map to σ ′ n is an A -equivalence. In other words, L Σ ∞ ( C ) is W Ω -local. But then, as C is cofibrant, C = Ω ∞ Σ ∞ ( C ) = R Ω ∞ L Σ ∞ ( C ), and this concludes.Point (3) is then a consequence of (1) and (2). (cid:3) Remark . (1) The construction of the triangulated category D A ( A ) can also beobtained using the more general construction of [ CD09 , §
7] – see also [
Hov01 , 7.11] and[
Ayo07b , chap. 4] for even more general accounts. Here, we exploit the simplificationarising from the fact that we invert a complex concentrated in degree 0: this allowed usto describe D A ( A ) simply as a Verdier quotient of the derived category of an abeliancategory. However, we can also consider the category of symmetric spectra in C( A S )with respect to one of the complexes S (1)[2] or S (1) and this leads to the equivalentcategories; see [ Hov01 , 8.3].(2) Point (3) of Proposition 5.3.25 is a particular case of [
Hov01 , 8.1].5.3.28 . Consider a morphism of abelian P -premotivic categories ϕ : A ⇄ B : ψ such that A (resp. B ) is compatible with a system of topology t (resp. t ′ ). Suppose t ′ is finerthan t . According to 5.3.17, we obtain an adjunction of abelian P -premotivic categories ϕ : C(Sp( A )) ⇄ C(Sp( B )) : ψ. The pair ( ϕ S , ψ S ) is a Quillen adjunction for the stable model structures (apply again [ CD09 ,prop. 3.11]). Thus we obtain a morphism of triangulated P -premotivic categories: L ϕ : D A ( A ) ⇄ D A ( B ) : R ψ. Remark . Under the light of Proposition 5.3.25, the category D A ( A ) might be consid-ered as the universal derived P -premotivic category T with a morphism D( A ) → T , and suchthat T satisfies the homotopy and the stability property. This can be made precise in the settingof algebraic derivators or of dg-categories (or any other kind of stable ∞ -categories). Proposition . Let t and t ′ be two admissible topologies, with t ′ finer than t . Then D A (Sh t ′ ( P , Λ)) is canonically equivalent to the the full subcategory of D A (Sh t ( P , Λ)) spannedby the objects which satisfy t ′ -descent. Proof.
It is sufficient to prove this proposition in the case where t is the coarse topology.We deduce from [ Ayo07b , 4.4.42] that, for any scheme S in S , we haveD A (Sh t ′ ( P /S, Λ)) = D (PSh( P /S, Λ)) [ W − ] , with W = W t ′ ∪ W A ∪ W Ω , where W t ′ is the set of maps of shapeΣ ∞ M S ( X ) { n } [ i ] → Σ ∞ M S ( X ) { n } [ i ] , for any t ′ -hypercover X → X and any integers n ≤ i . The assertion is then a particularcase of the description of the homotopy category of a left Bousfield localization. (cid:3) Example . We have the stable versions of the P -premotivic categories introduced inexample 5.2.17:1) Consider the admissible topology t = Nis. Following F. Morel, we define the stable A -derivedpremotivic category as (see also the construction of [ Ayo07b ]):D A , Λ := D A (Sh Nis ( Sm , Λ)) and D A , Λ := D A (cid:0) Sh Nis (cid:0) S ft , Λ (cid:1)(cid:1) , as well as the generalized stable A -derived premotivic category (5.3.31.1) D A , Λ := D A (cid:0) Sh Nis (cid:0) S ft , Λ (cid:1)(cid:1) . We will see in Example 6.1.10 that the generalized version contains the usual one as a full subcategory. . FIBRED DERIVED CATEGORIES 149
Given a scheme S , we shall also write:(5.3.31.2) D A ( S, Λ) := D A , Λ ( S ) and D A ( S, Λ) := D A , Λ ( S ) . In the case when t = ´et, we get the triangulated premotivic categories of ´etale premotives :D A (Sh ´et ( Sm , Λ)) and D A (cid:0) Sh ´et (cid:0) S ft , Λ (cid:1)(cid:1) . In each of these cases, we denote by Σ ∞ Λ tS ( X ) the premotive associated with a smooth S -scheme X . From the adjunction (5.1.24.2), we get an adjunction of triangulated premotivic categories: a ´et : D A , Λ ⇄ D A (Sh ´et ( Sm , Λ)) : R O ´et .
2) Assume P = S ft :Consider the S ft -admissible topology t = h (resp. t = qfh). In [ Voe96 ], Voevodsky hasintroduced the category of effective h-motives (resp. qfh-motives). According to the theory pre-sented above, one can extends this definition to the stable setting: one defines the category ofstable h -motives (resp. qfh -motives ) over S with coefficients in Λ as:DM h ( S, Λ) := D A (cid:0) Sh h (cid:0) S ft /S, Λ (cid:1)(cid:1) . resp. DM qfh ( S, Λ) := D A (cid:0) Sh qfh (cid:0) S ft /S, Λ (cid:1)(cid:1) . In other words, this is the stable A -derived category of h-sheaves (resp. qfh-sheaves) of Λ-modules. Moreover, we get the generalized triangulated premotivic category of h -motives (resp. qfh -motives) with coefficients in Λ over S :DM h , Λ := D A (cid:0) Sh h (cid:0) S ft , Λ (cid:1)(cid:1) . resp. DM qfh , Λ := D A (cid:0) Sh qfh (cid:0) S ft , Λ (cid:1)(cid:1) . For an S -scheme of finite type X , we will denote by Σ ∞ Λ h S ( X ) (resp Σ ∞ Λ qfh S ( X )) the correspondingpremotive associated with X in DM t ( S, Λ). Note that the h -sheafification functor induces apremotivic adjunction (see Paragraph 5.3.28):(5.3.31.3) DM qfh , Λ ⇄ DM h , Λ . These generalized premotivic categories are too big to be reasonable (in particular for the local-ization property – see Remark 2.3.4). Therefore, we introduce the triangulated category DM t ( S, Λ)as the localizing subcategory of DM t ( S, Λ) generated by objects of shape Σ ∞ Λ tS ( X )( p )[ q ] for anysmooth S -scheme of finite type X and any integers p and q . The fibred category DM h , Λ (resp.DM qfh , Λ ) defined above is premotivic. We call it the premotivic category h -motives (resp. qfh -motives) . The family of inclusions a premotivic morphism(5.3.31.4) DM t ( S, Λ) → DM t ( S, Λ)indexed by a scheme S defines a premotivic morphism (the existence of right adjoints is ensuredby the Brown representability theorem). Remark . When Λ = Q , we will show that the categories DM h , Q and DM qfh , Q areequivalent and satisfies the axioms of a motivic category. In fact, they are equivalent to thecategory of Beilinson motives. See Theorem 16.1.2 for all these results. Proposition . Consider the notations of the second point in the above example. Thenthe premotivic category DM t, Λ satisfies t -descent. Proof.
This is true for DM t, Λ by construction, which implies formally the assertion forDM t, Λ . (cid:3) Remark . According to Proposition 5.2.10 and Remark 5.3.29, for any admissible topol-ogy t , D A (Sh t ( P , Z )) is the universal derived P -premotivic category satisfying t -descent as wellas the homotopy and stability properties. A crucial example for us: the stable A -derived premotivic category D A is the universalderived premotivic category satisfying the properties of homotopy, of stability and of Nisnevichdescent.5.3.35 . We assume P = Sm .Let Sh • ( Sm ) be the category of pointed Nisnevich sheaves of sets. Consider the pointed versionof the adjunction of P -premotivic categories N : ∆ op Sh • ( Sm ) ⇄ C(Sh
Nis ( Sm , Z )) : K constructed in 5.2.25.If we consider on the left hand side the A -model category defined by Blander [ Bla03 ],( N S , K S ) is a Quillen adjunction for any scheme S .We consider ( G m ,
1) as a constant pointed simplicial sheaf. The construction of symmetric G m -spectra respectively to the model category ∆ op Sh • ( Sm ) can now be carried out following[ Jar00 ] or [
Ayo07b ] and yields a symmetric monoidal model category whose homotopy categoryis the stable homotopy category of Morel and Voevodsky SH( S ).Using the functoriality statements [ Hov01 , th. 8.3 and 8.4], we finally obtain a P -premotivicadjunction(5.3.35.1) N : SH ⇄ D A : K. The functor K is the analog of the Eilenberg-Mac Lane functor in algebraic topology; infact, this adjunction is actually induced by the Eilenberg-MacLane functor (see [ Ayo07b , chap.4]). In particular, as the rational model category of topological (symmetric) S -spectra is Quillenequivalent to the model category of complexes of Q -vector spaces, we have a natural equivalenceof premotivic categories(5.3.35.2) SH Q ⇄ D A , Q , (where SH Q ( S ) denotes the Verdier quotient of SH( S ) by the localizing subcategory generated bycompact torsion objects).5.3.36 . We can extend the considerations of Example 5.1.25 and Paragraph 5.2.22 on changingcoefficients in categories of sheaves.Let t be an admissible topology and ϕ : Λ → Λ ′ be an extension of rings. Using the P -premotivic adjunction (5.1.25.1) and according to Paragraph 5.3.28, we get an adjunction oftriangulated P -premotivic categories: L ϕ ∗ : D A (cid:0) Sh t ( P , Λ) (cid:1) ⇄ D A (cid:0) Sh t ( P , Λ ′ ) (cid:1) : R ϕ ∗ . Givew two Tate spectra C and D of t -sheaves of Λ-modules over P S , we get a canonical morphismof Λ ′ -modules:(5.3.36.1) Hom D A (Sh t ( P S , Λ)) (cid:0)
C, D (cid:1) ⊗ Λ Λ ′ −→ Hom D A (Sh t ( P S , Λ ′ )) (cid:0) L ϕ ∗ ( C ) , L ϕ ∗ ( D ) (cid:1) Then the stable version of Proposition 5.2.23 holds (the proof is the same):
Proposition . Consider the above assumptions. Then the map (5.3.36.1) is an iso-morphism in the two following cases:(1) If Λ ′ is a free Λ -module and C is compact;(2) If Λ ′ is a free Λ -module of finite rank. Constructible premotivic spectra.
Lemma . Let A be an abelian P -premotivic category compatible with a topology t andsuch that the category A -derived category D eff A ( A ) satisfies Nisnevich descent.Then, for any scheme S , the non trival cyclic permutation (123) of order acts as the identityon the premotive S { } ⊗ in D eff A ( A S ) . Proof.
Using example 5.2.21, it is sufficient to prove this in D A , Λ ( S ), which is well known;see for example [ Ayo07b , 4.5.65]. (cid:3) . FIBRED DERIVED CATEGORIES 151
Proposition . Consider the hypothesis of the previous lemma and assume that thetriangulated premotivic category D eff A ( A ) is compactly τ -generated.Then, for any scheme S , any couple of integers ( i, a ) , any compact object C of D eff A ( A S ) andany Tate spectrum E in A S , we have a canonical isomorphism Hom D A ( A S ) ( L Σ ∞ ( C ) { a } , E [ i ]) ≃ lim −→ r>> Hom D eff A ( A S ) ( C { a + r } , E r [ i ]) . Proof.
Given the previous lemma, this is a direct consequence of [
Ayo07b , theorems 4.3.61and 4.3.79]. (cid:3)
Corollary . Under the assumptions of the preceding proposition, the triangulated cat-egory D A ( A S ) is compactly ( Z × τ ) -generated where the factor Z corresponds to the Tate twist.More precisely, if D eff A ,c ( A S ) denotes the category of compact objects in D eff A ( A S ) , then thecategory of compact objects in D A ( A S ) is canonically equivalent to the pseudo-abelian completionof the category obtained as the -colimit of the following diagram: D eff A ,c ( A S ) ⊗ S { } −−−−−→ D eff A ,c ( A S ) −→ · · · −→ D eff A ,c ( A S ) ⊗ S { } −−−−−→ D eff A ,c ( A S ) −→ · · · . Let A be an abelian P -premotivic category compatible with an admissible topology t . Assume that: • The topology t is bounded in A (Definition 5.1.28). • The abelian P -premotivic category A is finitely τ -presented.We will denote by N tS a bounded generating family for t -hypercovers in A S .Recall from Proposition 5.2.38 that the category of compact objects of the triangulated cate-gory D eff A ( A S ) is canonically equivalent to the triangulated monoidal category: (cid:16) K b (cid:0) Z S ( Sm /S ; A ) (cid:1) / ( N tS ∪ T A S ) (cid:17) ♮ Let us denote by D A ,gm ( A S ) the category obtained from the monoidal category on the left handside of the above functor by formally inverting the Tate twist Z A S (1). Because D A ( A ) satisfiesthe stability property by construction, we readily obtains a canonical monoidal functor(5.3.41.1) D A ,gm ( A S ) → D A ( A S ) . Then applying Proposition 5.2.38, the above corollary and Proposition 1.4.11, we deduce:
Corollary . Consider the above hypothesis and notations.Then the triangulated premotivic category D A ( A ) is compactly ( Z × τ ) -generated. For anypremotive M in D A ( A S ) the following conditions are equivalent:(i) M is compact;(ii) M is ( Z × τ ) -constructible.Moreover, the functor (5.3.41.1) is fully faithful and has for essential image the compact ( i.e. τ -constructible) objects of D A ( A S ) . Example . From the considerations of Example 5.2.40, we obtain that for any scheme S , the compact objects of the category D A ( S, Λ) (resp. D A (cid:16) Sh cdh ( S ft /S, Λ) (cid:17) ) is obtainedfrom the monoidal triangulated categoryK b (Λ( Sm /S )) (resp. K b (cid:0) Λ( S ft /S ) (cid:1) )by the following steps: • one mods out by the triangulated subcategories T A S and BG S (resp. CDH S ) correspond-ing to the A -homotopy property and the Brown-Gersten triangles (resp. cdh-triangles), • one takes the pseudo-abelian envelope, • one formally inverts the Tate twist. Proposition . Assume P = S ft is the class of finite type (resp. separated and offinite type) morphisms.Let A be an abelian generalized premotivic category compatible with an admissible topology t such that: • A satisfies property (C) of Paragraph 5.1.35. • The A -derived category D eff A ( A ) is compactly τ -generated and satisfies Nisnevich de-scent.Then the stable A -derived premotivic category D A ( A ) is ( Z × τ ) -continuous. Proof.
This is an immediate corollary of Proposition 5.2.41 combined with Proposition5.3.39. (cid:3)
Example . According to the previous proposition and the second point of Example5.1.37, the generalized triangulated premotivic category D A , Λ is continuous. We also refer thereader to Corollary 6.1.12 for an extension of this result to the non generalized case.
6. Localization and the universal derived example . In this section, S is an adequate category of S -schemes as in 2.0. In sections 6.2 and6.3, we assume in addition that the schemes in S are finite dimensional.We will apply the definitions of the preceding section to the admissible class made of morphismsof finite type (resp. smooth morphisms of finite type) in S , denoted by S ft (resp. Sm ).Recall the general convention of section 1.4: • premotivic means Sm -premotivic. • generalized premotivic means S ft -premotivic. Example . Let t be a S ft -admissible topology. For a scheme S , we denote by Sh t (cid:0) S ft /S, Λ (cid:1) the category of sheaves of abelian groups on S ft /S for the topology t S . For an S -scheme of finitetype X , we let Λ tS ( X ) be the free t-sheaf of Λ-modules represented by X . Recall Sh t (cid:0) S ft , Λ (cid:1) isa generalized abelian premotivic category (see 5.1.4).Let ρ : Sm /S → S ft /S be the obvious inclusion functor and let us denote by t S the initialtopology on Sm /S such that ρ is continuous. Then it induces ( cf. [ SGA4 , IV, 4.10]) a sequenceof adjoint functors Sh t ( Sm /S, Λ) ρ ♯ - - ρ ∗ Sh t (cid:0) S ft /S, Λ (cid:1) ρ ∗ o o and we checked easily that this induces an enlargement of abelian premotivic categories:(6.1.1.1) ρ ♯ : Sh t ( Sm , Λ) ⇄ Sh t (cid:0) S ft , Λ (cid:1) : ρ ∗ . Remark . Note that for any scheme S , the abelian category Sh t ( Sm /S, Λ) can be de-scribed as the Gabriel quotient of the abelian category Sh t (cid:0) S ft /S, Λ (cid:1) with respect to the sheaves F over S ft /S such that ρ ∗ ( F ) = 0.An example of such a sheaf in the case where t = Nis and dim( S ) > S ( Z ) on S ft /S represented by a nowhere dense closed subscheme Z of S is zero when restrictedto Sm /S .6.1.3 . Consider an abelian premotivic category A compatible with an admissible topology t on Sm and a generalized abelian premotivic category A compatible with an admissible topology t ′ on S . We denote by M (resp. M ) the geometric sections of A (resp. A ). We assume that t ′ restricted to Sm is finer that t , and consider an adjunction of abelian premotivic categories: ρ ♯ : A ⇄ A : ρ ∗ . . LOCALIZATION AND THE UNIVERSAL DERIVED EXAMPLE 153 Let S be a scheme in S . The functors ρ ♯ and ρ ∗ induce a derived adjunction (see 5.2.19): L ρ ♯ : D eff A ( A S ) ⇄ D eff A ( A S ) : R ρ ∗ (where A is considered as an Sm -fibred category). Proposition . Consider the previous hypothesis, and fix a scheme S . Assume further-more that we have the following properties.(i) The functor ρ ♯ : A S → A S is fully faithful.(ii) The functor ρ ∗ : A S → A S commutes with small colimits.Then, the following conditions hold :(a) The induced functor ρ ∗ : C( A S ) → C( A S ) preserves A -equivalences.(b) The A -derived functor L ρ ♯ : D eff A ( A S ) → D eff A ( A S ) is fully faithful. Proof.
Point (a) follows from Proposition 5.2.24. To prove (b), we have to prove that theunit map M → ρ ∗ L ρ ♯ ( M )is an isomorphism for any object M of D eff A ( A S ). For this purpose, we may assume that M iscofibrant, so that we have M ≃ ρ ∗ ρ ♯ ( M ) ≃ ρ ∗ L ρ ♯ ( M )(where the first isomorphism holds already in C( A S )). (cid:3) Corollary . Consider the hypothesis of the previous proposition. Then the family ofadjunctions L ρ ♯ : D eff A ( A S ) → D eff A ( A S ) : R ρ ∗ indexed by a scheme S induces an enlargement oftriangulated premotivic categories L ρ ♯ : D eff A ( A ) ⇆ D eff A ( A ) : R ρ ∗ . Example . Considering the situation of 6.1.1, we will be particularly interested in thecase of the Nisnevich topology. We denote by D eff A , Λ the generalized A -derived premotivic cat-egory associated with Sh (cid:0) S ft , Λ (cid:1) (see also Example 5.3.31). The preceding corollary gives acanonical enlargement:(6.1.6.1) D eff A , Λ ⇄ D eff A , Λ . Consider again the hypothesis of 6.1.3. We denote simply by M (resp. M ) the geometricsections of the premotivic triangulated category D A ( A ) (resp. D A ( A )).Recall from 5.3.15 that we have defined S { } (resp. S { } ) as the cokernel of the canonicalmap S → M S ( G m,S ) (resp. S → M S ( G m,S )). Thus, it is obvious that we get a canonicalidentification ρ ♯ ( S { } ) = S { } . Therefore, the enlargement ρ ♯ can be extended canonically toan enlargement ρ ♯ : Sp( A ) ⇄ Sp( A ) : ρ ∗ of abelian premotivic categories in such a way that for any scheme S , the following diagramcommutes: A S ρ ♯ / / Σ ∞ A (cid:15) (cid:15) A S Σ ∞ A (cid:15) (cid:15) Sp( A S ) ρ ♯ / / Sp( A S ) . According to Proposition 5.3.13, Sp( A ) (resp. Sp( A )) is compatible with t (resp. t ′ ), and weobtain an adjoint pair of functors (5.3.28): L ρ ♯ : D A ( A S ) ⇄ D A ( A S ) : R ρ ∗ . From the preceding commutative square, we get the identification:(6.1.7.1) L ρ ♯ ◦ Σ ∞ A = Σ ∞ A ◦ L ρ ♯ As in the non effective case, we get the following result:
Proposition . Keep the assumptions of Proposition 6.1.4, and suppose furthermore thatboth D eff A ( A ) and D eff A ( A ) are compactly τ -generated. Then the derived functor L ρ ♯ : D A ( A S ) → D A ( A S ) is fully faithful. Proof.
We have to prove that for any Tate spectrum E of D A ( A S ), the adjunction morphism E → L ρ ∗ R ρ ♯ ( E )is an isomorphism. According to Proposition 1.3.20, the functor L ρ ∗ admits a right adjoint. Thus,applying Lemma 1.1.43, it is sufficient to consider the case where E = M S ( X ) { i } [ n ] for a smooth S -scheme X , and a couple ( n, i ) ∈ Z × τ .Moreover, it is sufficient to prove that for another smooth S -scheme Y and an integer j ∈ Z ,the induced morphismHom(Σ ∞ M S ( Y ) { j } , Σ ∞ M S ( X ) { i } [ n ]) → Hom(Σ ∞ M S ( Y ) { j } , Σ ∞ M S ( X ) { i } [ n ])is an isomorphism. Using the identification (6.1.7.1), propositions 5.3.39 and 6.1.4 allows toconclude. (cid:3) Corollary . If the assumptions of Proposition 6.1.8 hold for any scheme S in S , thenwe obtain an enlargement of triangulated premotivic categories L ρ ♯ : D A ( A ) ⇄ D A ( A ) : R ρ ∗ . Example . Considering again the situation of 6.1.1, in the case of the Nisnevich topol-ogy. We denote by D A , Λ the generalized stable A -derived premotivic category associated withSh (cid:0) S ft , Λ (cid:1) . The preceding corollary gives a canonical enlargement:(6.1.10.1) L ρ ♯ : D A , Λ ⇄ D A , Λ : R ρ ∗ which is compatible with the enlargement (6.1.6.1) in the sense that the following diagram isessentially commutative: D eff A , Λ / / Σ ∞ (cid:15) (cid:15) D eff A , ΛΣ ∞ (cid:15) (cid:15) D A , Λ / / D A , Λ Corollary . Consider a Grothendieck topology t on our category of schemes S . Let S be a scheme in S , and M an object of D A , Λ ( S ) . Then M satisfies t -descent in D A , Λ ( S ) ifand only if L ρ ♯ ( M ) satisfies t -descent in D A , Λ ( S ) . Proof.
Let f : X → S be a diagram of S -schemes of finite type. Define H q ( X , M ( p )) = Hom D A , Λ ( S ) (Λ X , L f ∗ ( M )( p )[ q ]) H q ( X , M ( p )) = Hom D A , Λ ( S ) (Λ X , L f ∗ L ρ ♯ ( M )( p )[ q ])for any integers p and q . The full faithfulness of L ρ ♯ ensures that the comparison map H q ( X , M ( p )) → H q ( X , M ( p ))is always bijective. This proposition follows then from the fact that M (resp. L ρ ♯ ( M )) satisfies t -descent if and only if, for any integers p and q , for any S -scheme of finite type X , and any t -hypercover X → X , the induced map H q ( X, M ( p )) → H q ( X , M ( p )) (resp. H q ( X, M ( p )) → H q ( X , M ( p )) )is bijective. (cid:3) We end-up this section with another interesting application of the preceding results.
Corollary . Consider the hypothesis and assumptions of Proposition 6.1.4. We sup-pose furthermore that the generalized abelian premotivic category A satisfies condition (C) ofParagraph 5.1.35. . LOCALIZATION AND THE UNIVERSAL DERIVED EXAMPLE 155 (1) Then the triangulated premotivic category D eff A ( A ) is τ -continuous.(2) Assume furhtermore that D eff A ( A ) and D eff A ( A ) are compactly τ -generated. Then thetriangulated premotivic category D A ( A ) is τ -continuous. Proof.
According to Proposition 5.2.41, the category D eff A ( A ) is τ -continuous. Accordingto Corollary 6.1.5, the functor L ρ ♯ : D eff A ( A ) → D eff A ( A ) : R ρ ∗ is fully faithful and commuteswith L f ∗ . Thus Point (1) follows.In the assumption of Point (2), we deduce from Proposition 5.3.44 that D A ( A ) is ( Z × τ )-continuous. Thus it is sufficient to apply Corollary 6.1.9 as in the effective case to get the assertionof Point (2). (cid:3) Example . According to the second point of Example 5.1.37, we can apply this corollaryto the enlargement Sh
Nis ( Sm , Λ) → Sh Nis (cid:0) S ft , Λ (cid:1) . Thus, we deduce that the triangulated premotivic categories D eff A , Λ and D A , Λ both are continuous. Recall from [
Ayo07b ] the following theorem of Moreland Voevodsky:
Theorem . The triangulated premotivic categories D eff A , Λ and D A , Λ satisfy the localiza-tion property. Corollary . (1) The premotivic category D A , Λ is a motivic category.(2) It is compactly generated by the Tate twist.(3) Suppose that T is a derived premotivic category (see 5.2.9) which is a motivic category.Then there exists a canonical morphism of derived premotivic categories: D A , Z → T . Proof.
The first assertion follows from the previous theorem and Remark 2.4.47. The secondone follows from Corollary 5.3.42. The last one follows from Proposition 3.3.5 and Example5.3.34. (cid:3)
Remark . Thus, Theorem 2.4.50 can be applied to D A , Λ . In particular, for any sepa-rated morphism of finite type f : T → S , there exists a pair of adjoint functors f ! : D A , Λ ( T ) ⇄ D A , Λ ( S ) : f ! as in the theorem loc. cit. so that we have removed the quasi-projective assumption in [ Ayo07a ].6.2.4 . Because the cdh topology is finer than the Nisnevich topology, we get an adjunction ofgeneralized premotivic categories: a ∗ cdh : D A , Λ ⇄ D A (cid:0) Sh cdh (cid:0) S ft , Λ (cid:1)(cid:1) : R a cdh , ∗ . Corollary . For any scheme S , the composite functor D A ( S, Λ) → D A ( S, Λ) a cdh −−−→ D A (cid:0) Sh cdh (cid:0) S ft /S, Λ (cid:1)(cid:1) is fully faithful.Moreover, it induces an enlargement of premotivic categories: (6.2.5.1) D A , Λ ⇄ D A (cid:0) Sh cdh (cid:0) S ft , Λ (cid:1)(cid:1) Remark . This corollary is a generalisation in our derived setting of the main theoremof [
Voe10c ]. Note that if dim( S ) >
0, there is no hope that the above composite functor isessentially surjective because as soon as Z is a nowhere dense closed subscheme of S , the premotive M cdh S ( Z, Λ) does not belong to its image ( cf. remark 6.1.2). Proof.
According to Corollary 6.2.2 and Proposition 3.3.10, any Tate spectrum E of D A ( S, Λ)satisfies cdh-descent in the derived premotivic category D A , Λ , and this implies the first assertionby 5.3.30 and 6.1.11. The second one then follows from the fact the forgetful functorD A (cid:0) Sh cdh (cid:0) S ft /S, Λ (cid:1)(cid:1) → D A ( S, Λ) . commutes with direct sums (its left adjoint preserves compact objects). (cid:3) . In all this section, we fix an abelian premotivic category A and we consider the canonicalpremotivic adjunction (5.1.2.1) associated with A .We assume A satisfies the following properties.(i) A is compatible with Nisnevich topology, so that we have from (5.1.2.1) a premotivicadjunction:(6.3.1.1) γ ∗ : Sh Nis ( Sm , Z ) ⇄ A : γ ∗ . (ii) A is finitely presented ( i.e. the functors Hom A S ( M S ( X ) , − ) preserve filtered colimitsand form a conservative family, Def. 1.3.11).(iii) For any scheme S , and for any open immersion U → X of smooth S -schemes, the map M S ( U ) → M S ( X ) is a monomorphism.(iv) For any scheme S , the functor γ ∗ : A S → Sh Nis ( Sm /S, Z ) is exact.Note that the functor γ ∗ : A S → Sh Nis ( Sm /S, Z ) is exact and conservative. As it also preservesfiltered colimits, this functor preserves in fact small colimits.Observe also that, according to these assumption, the abelian premotivic category of Tatespectra Sp( A ) is compatible with Nisnevich topology, N -generated. Moreover, we get a canonicalpremotivic adjunction(6.3.1.2) γ ∗ : Sp(Sh Nis ( Sm , Z )) ⇄ Sp( A ) : γ ∗ such that γ ∗ is conservative and preserves small colimits.In the following, we show how one can deduce properties of the premotivic triangulated cate-gories D eff A ( A ) and D A ( A ) from the good properties of D eff A , Z and D A , Z .6.3.a. Support property (effective case).
Proposition . For any scheme S , the functor γ ∗ : C( A S ) → C(Sh
Nis ( Sm /S, Z )) pre-serves and detects A -equivalences. Proof.
It follows immediately from Corollary 5.2.31 that γ ∗ preserves A -equivalences. Thefact it detects them can be rephrased by saying that the induced functor γ ∗ : D eff A ( A S ) → D eff A , Z ( S )is conservative. This is obviously true once we noticed that its left adjoint is essentially surjectiveon generators. (cid:3) Corollary . The right derived functor R γ ∗ = γ ∗ : D eff A ( A S ) → D eff A , Z ( S ) is conservative. Proposition . Let f : S ′ → S be a finite morphism of schemes. Then the inducedfunctor f ∗ : C( A S ′ ) → C( A S ) preserves colimits and A -equivalences. . LOCALIZATION AND THE UNIVERSAL DERIVED EXAMPLE 157 Proof.
We first prove f ∗ preserves colimits. We know the functors γ ∗ preserve colimits andare conservative. As we have the identification γ ∗ f ∗ = f ∗ γ ∗ , it is sufficient to prove the propertyfor A = Sh Nis ( Sm , Z ). Let X be a smooth S -scheme. It is sufficient to prove that, for any point x of X , if X hx denotes the henselianization of X at x , the functorSh Nis ( Sm /S ′ , Z ) → A b , F f ∗ ( F )( X hx ) = F ( S ′ × S X hx )commutes to colimits. Moreover the scheme S ′ × S X hx is finite over X hx , so that we have S ′ × S X hx = ∐ i Y i , where the Y i ’s are a finite family of henselian local schemes over S ′ × S X hx . Hence we haveto check that the functor F L i F ( Y i ) preserves colimits. As colimits commute to sums, it isthus sufficient to prove that the functors F F ( Y i ) commute to colimits. This follows from thefact that the local henselian schemes Y i are points of the topos of sheaves over the small Nisnevichsite of X .We are left to prove that the functor f ∗ : C( A S ′ ) → C( A S ) respects A -equivalences. For this,we shall study the behaviour of f ∗ with respect to the A -resolution functor constructed in 5.2.26.Note that f ∗ commutes to limits because it has a left adjoint. In particular, we know that f ∗ isexact. Moreover, one checks easily that f ∗ R ( n ) A = f ∗ R ( n ) A . As f ∗ commutes to colimits, this givesthe formula f ∗ R A = R A f ∗ . Let C be a complex of Nisnevich sheaves of abelian groups on Sm /S ′ .Choose a quasi-isomorphism C → C ′ with C ′ a Nis-flasque complex. Applying Proposition 5.2.28,we know that R A ( C ′ ) is A -fibrant and that we get a canonical A -equivalence f ∗ ( C ) → f ∗ ( C ′ ) → f ∗ ( R A ( C ′ )) = R A ( f ∗ ( C ′ )) . Hence we are reduced to prove that f ∗ preserves A -equivalences between A -fibrant objects.But such A -equivalences are quasi-isomorphisms, so that we can conclude using the exactness of f ∗ . (cid:3) Proposition . For any open immersion of schemes j : U → S , the exchange transfor-mation j ♯ γ ∗ → γ ∗ j ♯ is an isomorphism of functors. Proof.
Let X be a scheme, and F a Nisnevich sheaf of abelian groups on Sm /X . Definethe category C F as follows. The objects are the couples ( Y, s ), where Y is a smooth schemeover X , and s is a section of F over Y . The arrows ( Y, s ) → ( Y ′ , s ′ ) are the morphisms f ∈ Hom Sh Nis ( Sm /X, Z ) ( Z X ( Y ) , Z X ( Y ′ )) such that f ∗ ( s ′ ) = s . We have a canonical functor ϕ F : C F → Sh Nis ( Sm /X, Z )defined by ϕ F ( Y, s ) = Z X ( Y ), and one checks easily that the canonical maplim −→ C F ϕ F = lim −→ ( Y,s ) ∈ C F Z X ( Y ) → F is an isomorphism in Sh Nis ( Sm /X, Z ) (this is essentially a reformulation of the Yoneda lemma).Consider now an object F in the category A U . We get two categories C γ ∗ ( F ) and C γ ∗ ( j ♯ ( F )) .There is a functor i : C γ ∗ ( F ) → C γ ∗ ( j ♯ ( F )) which is defined by the formula i ( Y, s ) = (
Y, j ♯ ( s )). To explain our notations, let us say that we see s as a morphism from M S ( U, A ) to F , so that j ♯ ( s ) is a morphism from M S ( Y, A ) = j ♯ M S ( U, A )to j ♯ ( F ). This functor i has right adjoint i ′ : C γ ∗ ( j ♯ ( F )) → C γ ∗ ( F ) defined by i ′ ( Y, s ) = ( Y U , s U ), where Y U = Y × S U , and s U is the section of γ ∗ ( F ) over Y U that corresponds to the section j ∗ ( s ) of j ∗ j ♯ γ ∗ ( F ) over Y U under the canonical isomorphism γ ∗ ( F ) ≃ j ∗ j ♯ γ ∗ ( F ) (here, we use strongly the fact the functor j ♯ is fully faithful). The existenceof a right adjoint implies i is cofinal. This latter property is sufficient for the canonical morphismlim −→ C γ ∗ ( F ) ϕ γ ∗ ( j ♯ ( F )) ◦ i → lim −→ C γ ∗ ( j♯ ( F )) ϕ γ ∗ ( j ♯ ( F )) = γ ∗ ( j ♯ ( F )) to be an isomorphism. But the functor ϕ γ ∗ ( j ♯ ( F )) ◦ i is exactly the composition of the functor j ♯ with ϕ γ ∗ ( F ) . As the functor j ♯ commutes to colimits, we havelim −→ C γ ∗ ( F ) ϕ γ ∗ ( j ♯ ( F )) ◦ i = lim −→ C γ ∗ ( F ) j ♯ ϕ γ ∗ ( F ) ≃ j ♯ lim −→ C γ ∗ ( F ) ϕ γ ∗ ( F ) ≃ j ♯ ( γ ∗ ( F )) . Hence we obtain a canonical isomorphism j ♯ ( γ ∗ ( F )) ≃ γ ∗ ( j ♯ ( F )). It is easily seen that thecorresponding map γ ∗ ( F ) → j ∗ ( γ ∗ ( j ♯ ( F ))) = γ ∗ ( j ∗ j ♯ ( F )) is the image by γ ∗ of the unit map F → j ∗ j ♯ ( F ). This shows the isomorphism we have constructed is the exchange morphism. (cid:3) Corollary . For any open immersion of schemes j : U → S , the functor j ♯ : A U → A S is exact. Moreover, the induced functor j ♯ : C( A U ) → C( A S ) preserves A -equivalences. Proof.
Using the fact γ ∗ is exact and conservative, and propositions 6.3.2 and 6.3.5, it issufficient to prove this corollary when A = Sh Nis ( Sm , Z ). It is straightforward to prove exactnessusing Nisnevich points. The fact j ♯ preserves A -equivalences follows from the exactness propertyand from the obvious fact it preserves strong A -equivalences. (cid:3) Corollary . Let j : U → S be an open immersion of schemes. For any object M of D eff A ( A U ) the exchange morphism (6.3.7.1) L j ♯ ( R γ ∗ ( M )) → R γ ∗ ( L j ♯ ( M )) is an isomorphism in D eff A ( S, Z ) . Support property (stable case). . Recall from 5.3.17 that the premotivic adjunction ( γ ∗ , γ ∗ ) induces a canonical adjunctionof abelian premotivic categories that we denote by:˜ γ ∗ : Sp(Sh Nis ( Sm , Z )) ⇄ Sp( A S ) : ˜ γ ∗ Proposition . For any scheme S , the functor induced functor ˜ γ ∗ : C (cid:0) Sp( A S ) (cid:1) ⇄ C (cid:0) Sp(Sh
Nis ( Sm /S, Z )) (cid:1) preserves and detects stable A -equivalences. Proof.
Using the equivalence between symmetric Tate spectra and non symmetric Tatespectra, we are reduced to prove this for complexes of non symmetric Tate spectra. Consider anon symmetric Tate spectrum ( E n ) n ∈ N with suspension maps σ n : E n { } → E n +1 . The nonsymmetric Tate spectrum ˜ γ ∗ ( E ) is equal to γ ∗ ( E n ) in degree n ∈ Z , and the suspension map isgiven by the composite: S { } ⊗ S γ ∗ ( E n ) → γ ∗ ( γ ∗ ( S { } ) ⊗ S E n ) = γ ∗ ( E n { } ) γ ∗ ( σ n ) −−−−→ E n +1 . Thus, propositions 6.3.2 and 5.3.40 allows to conclude. (cid:3)
Corollary . The right derived functor R γ ∗ = γ ∗ : D A ( A S ) → D A , Z ( S ) is conservative. Proposition . Let j : U → X be an open immersion of schemes. For any object M of D A ( A U ) , the exchange morphism L j ♯ ( R γ ∗ ( M )) → R γ ∗ ( L j ♯ ( M )) is an isomorphism in D A , Z ( X ) . Proof.
From Corollary 6.3.6 and the P -base change formula for the open immersion j , onededuces easily that j ♯ preserves stable A -equivalences of (non symmetric) Tate spectra. Moreover,Proposition 6.3.5 shows that j ♯ γ ∗ = γ ∗ j ♯ at the level of Tate spectra. This concludes. (cid:3) . LOCALIZATION AND THE UNIVERSAL DERIVED EXAMPLE 159 Corollary . The triangulated premotivic category D A ( A ) satisfies the support prop-erty. Proof.
According to corollary 6.3.10, the functor R γ ∗ is conservative. Thus, by virtue ofthe preceding proposition, to prove the support property in the case of D A ( A ) it is sufficient toprove it in the case where A = Sh Nis ( Sm , Z ). This follows from theorems 6.2.1 and 2.4.50. (cid:3) Localization for smooth schemes.
Lemma . Let i : Z → S be a closed immersion which admits a smooth retraction p : S → Z . Then the exchange transformation L γ ∗ R i ∗ → R i ∗ L γ ∗ is an isomorphism in D eff A ( A S ) (resp. D A ( A S ) ). Proof.
We first remark that for any object C of C( A Z ) (resp. C(Sp( A Z ))) the canonicalsequence j ♯ ( pj ) ∗ ( C ) → p ∗ ( C ) → i ∗ ( C )is a cofiber sequence in D eff A ( A S ) (resp. D A ( A ) S )). Indeed, we can check this after applying theexact conservative functor γ ∗ . The sequence we obtain is canonically isomorphic through exchangetransformations to j ♯ j ∗ p ∗ ( γ ∗ C ) → p ∗ ( γ ∗ C ) → i ∗ i ∗ p ∗ ( γ ∗ C )using Corollary 6.3.7, the commutation of γ ∗ with j ∗ , p ∗ and i ∗ (recall it is the right adjoint ofa premotivic adjunction) and the relation pi = 1. But this last sequence is a cofiber sequence inD eff A , Z ( S ) (resp. D A , Z ( S )) because it satisfies the localization property (see 6.2.1).Using exchange transformations, we obtain a morphism of distinguished triangles in DM eff Z ( S ) γ ∗ j ♯ j ∗ p ∗ ( C ) / / γ ∗ p ∗ ( C ) / / γ ∗ i ∗ ( C ) / / Ex ( γ ∗ ,i ∗ ) (cid:15) (cid:15) γ ∗ j ♯ j ∗ p ∗ ( C )[1] j ♯ j ∗ p ∗ ( γ ∗ C ) / / p ∗ ( γ ∗ C ) / / i ∗ ( γ ∗ C ) / / j ♯ j ∗ p ∗ ( γ ∗ C )[1]The first two vertical arrows are isomorphisms as γ ∗ is the left adjoint of a premotivic adjunction;thus the morphism Ex ( γ ∗ , i ∗ ) is also an isomorphism. (cid:3) Proposition . Let i : Z → S be a closed immersion. If i admits a smooth retraction,then D eff A ( A ) satisfies (Loc i ). Proof.
This follows from Proposition 2.3.19 and the preceding lemma. (cid:3)
Corollary . Let S be a scheme. Then the premotivic category D eff A ( A ) (resp. D A ( A ) )satisfies localization with respect to any closed immersion between smooth S -schemes. Proof.
Let i : Z → X be closed immersion between smooth S -schemes. We want to provethat D eff A ( A ) (resp. D A ( A )) satisfies localization with respect to i . According to 2.3.18, it issufficient to prove that for any smooth S -scheme S , the canonical map M S ( X/X − X Z ) → i ∗ M Z ( X Z )is an isomorphism where we use the notation of loc. cit. and M ( ., A ) denotes the geometricsections of D eff A ( A ) (resp. D A ( A )). But the premotivic triangulated category category D A ( A )(resp. D eff A ( A )) satisfies the Nisnevich separation property and the Sm -base change property.Thus, we can argue locally in S for the Nisnevich topology. Thus, the statement is reduced to thepreceding proposition as i admits locally for the Nisnevich topology a smooth retraction (see forexample [ D´eg07 , 4.5.11]). (cid:3)
7. Basic homotopy commutative algebra7.1. Rings.
Definition . A symmetric monoidal model category V satisfies the monoid axiom if, forany trivial cofibration A → B and any object X , the smallest class of maps of V which containsthe map X ⊗ A → X ⊗ B and is stable by pushouts and transfinite compositions is contained inthe class of weak equivalences.7.1.2 . Let V be a symmetric monoidal category. We denote by Mon ( V ) the category ofmonoids in V . If V has small colimits, the forgetful functor U : Mon ( V ) → V has a left adjoint F : V → Mon ( V ) . Theorem . Let V a symmetric monoidal combinatorial model category which satisfies themonoid axiom. The category of monoids Mon ( V ) is endowed with the structure of a combinatorialmodel category whose weak equivalences (resp. fibrations) are the morphisms of commutativemonoids which are weak equivalences (resp. fibrations) in V . In particular, the forgetful functor U : Mon ( V ) → V is a right Quillen functor. Moreover, if the unit object of V is cofibrant, thenany cofibrant object of Mon ( V ) is cofibrant as an object of V . Proof.
This is very a particular case of the third assertion of [
SS00 , Theorem 4.1] (the factthat
Mon ( V ) is combinatorial whenever V is so comes for instance from [ Bek00 , Proposition2.3]). (cid:3)
Definition . A symmetric monoidal model category V is strongly Q -linear if the un-derlying category of V is additive and Q -linear (i.e. all the objects of V are uniquely divisible). Remark . If V is a strongly Q -linear stable model category, then it is Q -linear in thesense of 3.2.14. Lemma . Let V be a strongly Q -linear model category, G a finite group, and u : E → F an equivariant morphism of representations of G in V . Then, if u is a cofibration in V , so is theinduced map E G → F G (where the subscript G denotes the coinvariants under the action of thegroup G ). Proof.
The map E G → F G is easily seen to be a direct factor (retract) of the cofibration E → F . (cid:3) . If V is a symmetric monoidal category, we denote by Comm ( V ) the category of com-mutative monoids in V . If V has small colimits, the forgetful functor U : Comm ( V ) → V has a left adjoint F : V → Comm ( V ) . Theorem . Let V a symmetric monoidal combinatorial model category. Assume that V is left proper and tractable, satisfies the monoid axiom, and is strongly Q -linear. Then thecategory of commutative monoids Comm ( V ) is endowed with the structure of a combinatorialmodel category whose weak equivalences (resp. fibrations) are the morphisms of commutativemonoids which are weak equivalences (resp. fibrations) in V . In particular, the forgetful functor U : Comm ( V ) → V is a right Quillen functor.If moreover the unit object of V is cofibrant, then any cofibrant object of Comm ( V ) is cofibrantas an object of V . Proof.
The preceding lemma implies immediately that V is freely powered in the sense of[ Lur12 , Definition 4.3.17], so that the existence of this model category structure follows from ageneral result of Lurie [
Lur12 , Proposition 4.3.21]. The second assertion is then true by definition. . BASIC HOMOTOPY COMMUTATIVE ALGEBRA 161
The last assertion is proved by a careful analysis of pushouts by free maps in
Comm ( V ) as follows.For two cofibrations u : A → B and v : C → D in V , write u ∧ v for the map u ∧ v : A ⊗ D ∐ A ⊗ C B ⊗ C → B ⊗ D (which is a cofibration by definition of monoidal model categories). By iterating this construction,we get, for a cofibration u : A → B in V , a cofibration ∧ n ( u ) = u ∧ · · · ∧ u | {z } n times : (cid:3) n ( u ) → B ⊗ n . Note that the symmetric group S n acts naturally on B ⊗ n and (cid:3) n ( u ). We define Sym n ( B ) = ( B ⊗ n ) S n and Sym n ( B, A ) = (cid:3) n ( u ) S n . By virtue of Lemma 7.1.6, we get a cofibration of V : σ n ( u ) : Sym n ( B, A ) → Sym n ( B ) . Consider now the free map F ( u ) : F ( A ) → F ( B ) can be filtered by F ( A )-modules as follows.Define D = F ( A ). As A = Sym ( B, A ), we have a natural morphism F ( A ) ⊗ Sym ( B, A ) → F ( A ).The objects D n are then defined by induction with the pushouts below. F ( A ) ⊗ Sym n ( B, A ) F ( A ) ⊗ σn ( u ) / / (cid:15) (cid:15) F ( A ) ⊗ Sym n ( B ) (cid:15) (cid:15) D n − / / D n We get natural maps D n → F ( B ) which induce an isomorphismlim −→ n ≥ D n ≃ F ( B )in such a way that the morphism F ( u ) correspond to the canonical map F ( A ) = D → lim −→ n ≥ D n . Hence, if F ( A ) is cofibrant, all the maps D n − → D n are cofibrations, so that the map F ( A ) → F ( B ) is a cofibration in V . In the particular case where A is the initial object of V , we see thatfor any cofibrant object B of V , the free commutative monoid F ( B ) is cofibrant as an object of V (because the initial object of Comm ( V ) is the unit object of V ). This also implies that, if u isa cofibration between cofibrant objects, the map F ( u ) is a cofibration in V .This description of F ( u ) also allows to compute the pushouts of F ( u ) in Comm ( V ) in V asfollows. Consider a pushout F ( A ) F ( u ) / / (cid:15) (cid:15) F ( B ) (cid:15) (cid:15) R v / / S in Comm ( V ). For n ≥
0, define R n by the pushouts of V : F ( A ) / / (cid:15) (cid:15) D n (cid:15) (cid:15) R / / R n We then have an isomorphism lim −→ n ≥ R n ≃ S .
In particular, if u is a cofibration between cofibrant objects, the morphism of commutative monoids v : R → S is then a cofibration in V . As the forgetful functor U preserves filtered colimits, concludeeasily from there (with the small object argument [ Hov99 , Theorem 2.1.14]) that any cofibration of Comm ( V ) is a cofibration of V . Using again that the unit object of V is cofibrant in V (i.e. thatthe initial object of Comm ( V ) is cofibrant in V ) this proves the last assertion of the theorem. (cid:3) Corollary . Let V a symmetric monoidal combinatorial model category. Assume that V is left proper and tractable, satisfies the monoid axiom, and is strongly Q -linear. Consider asmall set H of maps of V , and denote by L H V the left Bousfield localization of V by H ; see [ Bar10 , Theorem 4.7] . Define the class of H -equivalences in Ho( V ) to be the class of maps whichbecome invertible in Ho( L H V ) . If H -equivalences are stable by (derived) tensor product in Ho( V ) ,then L H V is a symmetric monoidal combinatorial model category (which is again left proper andtractable, satisfies the monoid axiom, and is strongly Q -linear).In particular, under these assumtions, there exists a morphism of commutative monoids → R in V which is a weak equivalence of L H V , with R a cofibrant and fibrant object of L H V . Proof.
The first assertion is a triviality. The last assertion follows immediately: the map → R is simply obtained as a fibrant replacement of in the model category Comm ( L H V )obtained from Theorem 7.1.8 applied to L H V . (cid:3) . Consider now a category S , as well as a closed symmetric monoidal bifibred category M over S . We shall also assume that the fibers of M admit limits and colimits.Then the categories Mon ( M ( X )) (resp. Comm ( M ( X ))) define a bifibred category over S as follows. Given a morphism f : X → Y , the functor f ∗ : M ( Y ) → M ( X )is symmetric monoidal, so that it preserves monoids (resp. commutative monoids) as well asmorphisms between them. It thus induces a functor(7.1.10.1) f ∗ : Mon ( M ( Y )) → Mon ( M ( X ))(resp. f ∗ : Comm ( M ( Y )) → Comm ( M ( X )) ) . As f ∗ : M ( Y ) → M ( X ) is symmetric monoidal, its right adjoint f ∗ is lax monoidal: there is anatural morphism(7.1.10.2) Y → f ∗ ( X ) = f ∗ f ∗ ( Y ) , and, for any objects A and B of M ( X ), there is a natural morphism(7.1.10.3) f ∗ ( A ) ⊗ Y f ∗ ( B ) → f ∗ ( A ⊗ X B )which corresponds by adjunction to the map f ∗ ( f ∗ ( A ) ⊗ Y f ∗ ( B )) ≃ f ∗ f ∗ ( A ) ⊗ f ∗ f ∗ ( B ) → A ⊗ B .
Hence the functor f ∗ preserves also monoids (resp. commutative monoids) as well as morphismsbetween them, so that we get a functor(7.1.10.4) f ∗ : Mon ( M ( X )) → Mon ( M ( Y ))(resp. f ∗ : Comm ( M ( X )) → Comm ( M ( Y )) ) . By construction, the functor f ∗ of (7.1.10.1) is a left adjoint ot the functor f ∗ of (7.1.10.4). Theseconstructions extend to morphisms of S -diagrams in a similar way. Proposition . Let M be a symmetric monoidal combinatorial fibred model categoryover S . Assume that, for any object X of S , the model category M ( X ) satisfies the monoidaxiom (resp. is left proper and tractable, satisfies the monoid axiom, and is strongly Q -linear).(a) For any object X of S , the category Mon ( M )( X ) (resp. Comm ( M )( X ) ) of monoids(resp. of commutative monoids) in M ( X ) is a combinatorial model category structurewhose weak equivalences (resp. fibrations) are the morphisms of commutative monoidswhich are weak equivalences (resp. fibrations) in M ( X ) . This turns Mon ( M ) (resp.Comm ( M ) ) into a combinatorial fibred model category over S . . BASIC HOMOTOPY COMMUTATIVE ALGEBRA 163 (b) For any morphism of S -diagrams ϕ : ( X , I ) → ( Y, J ) , the adjunction ϕ ∗ : Mon ( M )( Y , J ) ⇄ Mon ( M )( X , I ) : ϕ ∗ ( resp. ϕ ∗ : Comm ( M )( Y , J ) ⇄ Comm ( M )( X , I ) : ϕ ∗ ) is a Quillen adjunction (where the categories of monoids Mon ( M )( X , I ) (resp. ofcommutative monoids Comm ( M )( X , I ) ) are endowed with the injective model categorystructure obtained from Proposition 3.1.7 applied to Mon ( M ) (resp. to Comm ( M ) ).(d) If moreover, for any object X of S , the unit X is cofibrant in M ( X ) , then, for mor-phism of S -diagrams ϕ : ( X , I ) → ( Y, J ) , the square Ho(
Mon ( M ))( Y , J ) L ϕ ∗ / / U (cid:15) (cid:15) Ho(
Mon ( M ))( X , I ) U (cid:15) (cid:15) Ho( M )( Y , J ) L ϕ ∗ / / Ho( M )( X , I )(7.1.11.1) is essentially commutative. Similarly, in the respective case, the square Ho(
Comm ( M ))( Y , J ) L ϕ ∗ / / U (cid:15) (cid:15) Ho(
Comm ( M ))( X , I ) U (cid:15) (cid:15) Ho( M )( Y , J ) L ϕ ∗ / / Ho( M )( X , I )(7.1.11.2) is essentially commutative. Proof.
Assertion (a) is an immediate consequence of Theorem 7.1.3 (resp. of Theorem7.1.8), and assertion (b) is a particular case of Proposition 3.1.11 (beware that the injective modelcategory structure on
Comm ( M )( X , I ) does not necessarily coincide with the model categorystructure given by Theorem 7.1.3 (resp. of Theorem 7.1.8) applied to the injective model structureon M ( X , I )). For assertion (d), we see by the second assertion of Proposition 3.1.6 that it issufficient to prove it when ϕ : X → Y is simply a morphism of S . In this case, by construction ofthe total left derived functor of a left Quillen functor, this follows from the fact that ϕ ∗ commuteswith the forgetful functor and from the fact that, by virtue of the last assertion of Theorem7.1.3 (resp. of Theorem 7.1.8), the forgetful functor U preserves weak equivalences and cofibrantobjects. (cid:3) Remark . The main application of the preceding corollary will come from assertion(d): it says that, given a monoid (resp. a commutative monoid) R in M ( Y ) and a morphism f : X → Y , the image of R by the functor L f ∗ : Ho( M )( Y ) → Ho( M )( X )is canonically endowed with a structure of monoid (resp. of commutative monoid) in the strongestsense possible. Under the assumptions of assertion (c) of Proposition 7.1.11, we shall often makethe abuse of saying that L f ∗ ( R ) is a monoid (resp. a commutative monoid) in M ( X ) withoutrefereeing explicitely to the model category structure on Mon ( M )( X ) (resp. on Comm ( M )( X )).Similarly, for any monoid (resp. commutative monoid) R in M ( X ), R f ∗ ( R ) will be canonicallyendowed with a structure of a monoid (resp. a commutative monoid) in M ( Y ). In particular, forany monoid (resp. commutative monoid) R in M ( Y ), the adjunction map R → R f ∗ L f ∗ ( R )is a morphism of monoids (i.e. is a map in the homotopy category Ho( Mon ( M ))( X ) (resp.Ho( Comm ( M ))( X ))), and, for any monoid (resp. commutative monoid) R in M ( X ), the adjunc-tion map L f ∗ R f ∗ ( R ) → R is a morphism of monoids (i.e. is a map in the homotopy category Ho( Mon ( M ))( Y ) (resp.Ho( Comm ( M ))( Y ))). Remark . In order to get a good homotopy theory of commutative monoids wihoutthe strongly Q -linear assumption, we should replace commutative monoids by E ∞ -algebras (i.e.objects endowed with a structure of commutative monoid up to a bunch of coherent homotopies).More generally, we should prove the analog of Theorem 7.1.3 and of Theorem 7.1.8 by replacing Mon ( V ) by the category of algebras of some ‘well behaved’ operad, and then get as a consequencethe analog of Proposition 7.1.11. All this is a consequence of the general constructions and resultsof [ Spi01, BM03, BM09 ].However, in the case we are interested in the homotopy theory of commutative monoids insome category of spectra V , it seems that some version of Shipley’s positive stable model structure ( cf. [ Shi04 , Proposition 3.1]) would provide a good model category for commutative monoids,which, by Lurie’s strictification theorem [
Lur12 , Theorem 4.4.4.7], would be equivalent to thehomotopy theory of E ∞ -algebras in V . This kind of technics is available in the context of stablehomotopy theory of schemes, which provides a good setting to speak of motivic commutative ringspectra; see [ Hor10 ]. Therefore, Theorem 7.1.8 and Proposition 7.1.11 are in fact true in SH forgenuine commutative monoids without any Q -linearity assumption. . Given a monoid R in a symmetric monoidal category V , we shall write R - mod( V ) forthe category of (left) R -modules. The forgetful functor U : R - mod( V ) → V is a left adjoint to the free R -module functor R ⊗ ( − ) : V → R - mod( V ) . If V has enough small colimits, and if R is a commutative monoid, the category R - mod( V ) isendowed with a unique symmetric monoidal structure such that the functor R ⊗ ( − ) is naturallysymmetric monoidal. We shall denote by ⊗ R the tensor product of R - mod( V ). Theorem . Let V be a combinatorial symmetric model category which satisfies themonoid axiom.(i) For any monoid R in V , the category of right (resp. left) R -modules is a combinatorialmodel category with weak equivalences (resp. fibrations) the morphisms of R -moduleswhich are weak equivalences (resp. fibrations) in V .(ii) For any commutative monoid R in V , the model category of R -modules given by (i) is acombinatorial symmetric monoidal model category which satisfies the monoid axiom. Proof.
Assertions (i) and (ii) are particular cases of the first two assertions of [
SS00 , The-orem 4.1]. (cid:3)
Definition . A symmetric monoidal model category V is perfect if it has the followingproperties.(a) V is combinatorial and tractable (3.1.27);(b) V satisfies the monoid axiom;(c) For any weak equivalence of monoids R → S , the functor M S ⊗ R M is a left Quillenequivalence from the category of left R -modules to the category of left S -modules.(d) weak equivalences are stable by small sums in V . Remark . If V is a perfect symmetric monoidal model category, then, for any com-mutative monoid R , the symmetric monoidal model category of R -modules in V given by The-orem 7.2.2 (ii) is also perfect: condition (c) is quite obvious, and condition (d) comes from thefact that the forgetful functor U : R - mod → V commutes with small sums, while it preserves anddetects weak equivalences. Note that condition (d) implies that the functor U : Ho( R - mod) → Ho( V ) preserves small sums. Remark . If V is a stable symmetric monoidal model category which satisfies the monoidaxiom, then for any monoid R of V , the model category of (left) R -modules given by Theorem . BASIC HOMOTOPY COMMUTATIVE ALGEBRA 165 R - mod) is given by the derived tensor productby the R -bimodule R [1], which is clearly invertible with inverse R [ − A -derived premotivic categories (cf Def. 5.3.22): Proposition . Let t be an admissible topology. Then, for any scheme S in S , thesymmetric monoidal model structure on C(Sp(Sh t ( P /S, Z ))) underlying the triangulated category D A (Sh t ( P /S, Z )) is perfect. Proof.
The generating family of Sh t ( P /S, Z ) is flat in the sense of [ CD09 , 3.1], so that, byvirtue of [
CD09 , prop. 7.22 and cor. 7.24], the assumptions of Proposition 7.2.9 are fulfilled. (cid:3)
Proposition . Let V be a stable perfect symmetric monoidal model category. Assumefurthermore that Ho( V ) admits a small family G of compact generators (as a triangulated category).For any monoid R in V , the triangulated category Ho( R - mod( V )) admits the set { R ⊗ L E | E ∈ G} as a family of compact generators. Proof.
We have a derived adjunction R ⊗ L ( − ) : Ho( V ) ⇄ Ho( R - mod( V )) : U .
As the functor U preserves small sums the functor R ⊗ L ( − ) preserves compact objects. But U isalso conservative, so that { R ⊗ L E | E ∈ G} is a family of compact generators of Ho( R - mod( V )). (cid:3) Remark . If V is a combinatorial symmetric model category which satisfies the monoidaxiom, then there are two ways to derive the tensor product. The first one consists to derive theleft Quillen bifunctor ( − ) ⊗ ( − ), which gives the usual derived tensor product( − ) ⊗ L ( − ) : Ho( V ) × Ho( V ) → Ho( V ) . Remember that, by construction, A ⊗ L B = A ′ ⊗ B ′ , where A ′ and B ′ are cofibrant replacementsof A and B respectively. On the other hand, the monoid axiom gives that, for any object A of V ,the functor A ⊗ ( − ) preserves weak equivalences between cofibrant objects, which implies that ithas also a total left derived functor A ⊗ L ( − ) : Ho( V ) → Ho( V ) . Despite the fact we have adopted very similar (not to say identical) notations for these two derivedfunctor, there is no reason they would coincide in general: by construction, the second one is definedby A ⊗ L B = A ⊗ B ′ , where B ′ is some cofibrant replacement of B . However, they coincide quiteoften in practice (e.g. for simplicial sets, for the good reason that all of them are cofibrant, or forsymmetric S -spectra, or for complexes of quasi-coherent O X -modules over a quasi-compact andquasi-separated scheme X ). Proposition . Let V be a stable combinatorial symmetric monoidal model categorywhich satisfies the monoid axiom. Assume furthermore that, for any cofibrant object A of V , thefunctor A ⊗ ( − ) preserve weak equivalences (in other words, that the two ways to derive the tensorproduct explained in Remark 7.2.8 coincide), and that weak equivalences are stable by small sumsin V . Then the symmetric monoidal model category V is perfect. Proof.
We just have to check condition (c) of Definition 7.2.3. Consider a weak equivalenceof monoids R → S . We then get a derived adjunction S ⊗ L R ( − ) : Ho( R - mod( V )) ⇄ Ho( S - mod( V )) : U , where S ⊗ L R ( − ) is the left derived functor of the functor M S ⊗ R M . We have to prove that,for any left R -module M , the map M → S ⊗ L R M is an isomorphism in Ho( V ). As this is a morphism of triangulated functors which commuteswith sums, and as Ho( R - mod( V )) is well generated in the sense of Neeman [ Nee01 ] (as thelocalization of a stable combinatorial model category), it is sufficient to check this when M runs over a small family of generators of Ho( R - mod( V )). Let us chose is a small family of generators G of Ho( V ). As the forgetful functor from Ho( R - mod( V )) to Ho( V ) is conservative, we see that { R ⊗ L E | E ∈ G} is a small generating family of Ho( R - mod( V )). We are thus reduced to provethat the map R ⊗ L E → S ⊗ L R ( R ⊗ L E ) ≃ S ⊗ L E is an isomorphism for any object E in G . For this, we can assume that E is cofibrant, and thisfollows then from the fact that the functor ( − ) ⊗ E preserves weak equivalences by assumption. (cid:3) . Let S be a category endowed with an admissible class of morphisms P , and M asymmetric monoidal P -fibred category. Consider a monoid R in the symmetric monoidal category M (1 S , S ) (i.e. a section of the fibred category Mon ( M ) over S ). In other words, R consists ofthe data of a monoid R X for each object X of S , and of a morphism of monoids a f : f ∗ ( R Y ) → R X for each map f : X → Y in S , subject to coherence relations; see 3.1.2.For an object X of S , we shall write R - mod( X ) for the category of (left) R X -modules in M ( X ), i.e. R - mod( X ) = R X - mod( M ( X )) . This defines a fibred category R - mod over S as follows.For a morphism f : X → Y , the inverse image functor(7.2.10.1) f ∗ : R - mod( Y ) → R - mod( X )is defined by(7.2.10.2) M R X ⊗ f ∗ ( R Y ) f ∗ ( M )(where, on the right hand side, f ∗ stands for the inverse image functor in M ). The functor(7.2.10.1) has a right adjoint(7.2.10.3) f ∗ : R - mod( X ) → R - mod( Y )which is simply the functor induced by f ∗ : M ( X ) → M ( Y ) (as the latter sends R X -modules to f ∗ ( R X )-modules, which are themselves R Y -modules via the map a f ).If the map f is a P -morphism, then, for any R X -module M , the object f ♯ ( M ) has a naturalstructure of R Y -module: using the map a f , M has a natural structure of f ∗ ( R Y )-module f ∗ ( R Y ) ⊗ X M → M , and applying f ♯ , we get by the P -projection formula (1.1.26) a morphism R Y ⊗ f ♯ ( M ) ≃ f ♯ ( f ∗ ( R Y ) ⊗ M ) → f ♯ ( M )which defines a natural R Y -module structure on f ♯ ( M ). For a P -morphism f : X → Y , we definea functor(7.2.10.4) f ♯ : R - mod( X ) → R - mod( Y )as the functor induced by f ♯ : M ( X ) → M ( Y ). Note that the functor (7.2.10.4) is a left adjointto the functor (7.2.10.1) whenever the map a f : f ∗ ( R Y ) → R X is an isomorphism in M ( X ).We shall say that R is a cartesian monoid in M over S if R is a monoid of M (1 C , C ) suchthat all the structural maps f ∗ ( R Y ) → R X are isomorphisms (i.e. if R is a cartesian section ofthe fibred category Mon ( M ) over S )If R is a cartesian monoid in M over S , then R - mod is a P -fibred category over S : to seethis, it remains to prove that, for any pullback square of S X ′ g / / f ′ (cid:15) (cid:15) X f (cid:15) (cid:15) Y ′ h / / Y in which f is a P -morphism, and for any R X -module M , the base change map f ′ ♯ g ∗ ( M ) → h ∗ f ♯ ( M ) . BASIC HOMOTOPY COMMUTATIVE ALGEBRA 167 is an isomorphism, which follows immediately from the analogous formula for M .Similarly, we see that whenever R is a commutative monoid of M (1 C , C ) (i.e. R X is acommutative monoid in M ( X ) for all X in S ), then R - mod is a symmetric monoidal P -fibredcategory. Proposition . Let M be a combinatorial symmetric monoidal P -fibred model categoryover S which satisfies the monoid axiom, and R a monoid in M (1 S , S ) (resp. a cartesian monoidin M over S ). Then 7.2.2 (i) applied termwise turns R - mod into a combinatorial fibred modelcategory (resp. a combinatorial P -fibred model category).If moreover R is commutative, then R - mod is a combinatorial symmetric monoidal fibredmodel category (resp. a combinatorial symmetric monoidal P -fibred model category). Proof.
Choose, for each object X of S , two small sets of maps I X and J X which generatethe class of cofibrations and the class of trivial cofibrations in M ( X ) respectively. Then R X ⊗ X I X and R X ⊗ X J X generate the class of cofibrations and the class of trivial cofibrations in R - mod( X )respectively. For a map f : X → Y in S , we see from formula (7.2.10.2) that the functor (7.2.10.1)sends these generating cofibrations and trivial cofibrations to cofibrations and trivial cofibrationsrespectively, from which we deduce that the functor (7.2.10.1) is a left Quillen functor. In therespective case, if f is a P -morphism, then we deduce similarly from the projection formula(1.1.26) in M that the functor (7.2.10.4) sends generating cofibrations and trivial cofibrationsto cofibrations and trivial cofibrations respectively. The last assertion follows easily by applying7.2.2 (ii) termwise. (cid:3) Definition . Let M be a symmetric monoidal P -fibred model category over S . A homotopy cartesian monoid R in M will be a homotopy cartesian section of Mon ( M ). Proposition . Let M be a perfect symmetric monoidal P -fibred model category over S , and consider a homotopy cartesian monoid R in M over S .Then Ho( R - mod) is a P -fibred category over S , and R ⊗ L ( − ) : Ho( M ) → Ho( R - mod) is a morphism of P -fibred categories. In the case where R is commutative, Ho( R - mod) is even asymmetric monoidal P -fibred category.Moreover, for any weak equivalence between homotopy cartesian monoids R → S over S , theQuillen morphism S ⊗ R ( − ) : R - mod → S - mod induces an equivalence of P -fibred categories over S S ⊗ L R ( − ) : Ho( R - mod) → Ho( S - mod) . Proof.
It is sufficient to prove these assertions by restricting everything over S /S , where S runs over all the objects of S . In particular, we may (and shall) assume that S has a terminalobject S . As M is perfect, it follows from condition (c) of Definition 7.2.3 that we can replace R by any of its cofibrant resolution. In particular, we may assume that R S is a cofibrant objectof Mon ( M )( S ). We can thus define a termwise cofibrant cartesian monoid R ′ as the family ofmonoids f ∗ ( R S ), where f : X → S runs over all the objects of S ≃ S /S . There is a canonicalmorphism of homotopy cartesian monoids R ′ → R which is a termwise weak equivalence. We thusget, by condition (c) of Definition 7.2.3, an equivalence of fibred categories R ⊗ L R ′ ( − ) : Ho( R ′ - mod) → Ho( R - mod) . We can thus replace R by R ′ , which just means that we can assume that R is cartesian andtermwise cofibrant. The first assertion follows then easily from Proposition 7.2.11. In the casewhere R is commutative, we prove that Ho( R - mod) is a P -fibred symmetric monoidal categoryas follows. Let f : X → Y a morphism of S . We would like to prove that, for any object M inHo( R - mod)( X ) and any object N in Ho( R - mod)( Y ), the canonical map(7.2.13.1) L f ♯ ( M ⊗ L R f ∗ ( N )) → L f ♯ ( M ) ⊗ L R N is an isomorphism. By adjunction, this is equivalent to prove that, for any objects N and E inHo( R - mod)( Y ), the map(7.2.13.2) f ∗ R Hom R ( N, E ) → R Hom R ( f ∗ ( N ) , f ∗ ( E ))is an isomorphism in Ho( R - mod)( X ) (where R Hom R stands for the internal Hom of Ho( R - mod)).But the forgetful functors U : Ho( R - mod)( X ) → Ho( M )( X )are conservative, commute with f ∗ for any P -morphism f , and commute with internal Hom: byadjunction, this follows immediately from the fact that the functors R ⊗ L ( − ) : Ho( M )( X ) → Ho( R - mod)( X ) ≃ Ho( R ′ - mod)( X )are symmetric monoidal and define a morphism of P -fibred categories (and thus, in particular,commute with f ♯ for any P -morphism f ). Hence, to prove that (7.2.13.2) is an isomorphism, itis sufficient to prove that its analog in Ho( M ) is so, which follows immediately from the fact thatthe analog of (7.2.13.1) is an isomorphism in Ho( M ) by assumption.For the last assertion, we are also reduced to the case where R and S are cartesian and termwisecofibrant, in which case this follows easily again from condition (c) of Definition 7.2.3. (cid:3) Proposition . Let M be a combinatorial symmetric monoidal model category over S which satisfies the monoid axiom. Then, for any cartesian monoid R in M over S we have aQuillen morphism R ⊗ ( − ) : M → R - mod . If, for any object X of S , the unit object X is cofibrant in M ( X ) and the monoid R X is cofibrantin Mon ( M )( X ) , then the forgetful functors also define a Quillen morphism U : R - mod → M . Proof.
The first assertion is obvious. For the second one, note that, for any object X of S ,the monoid R X is also cofibrant as an object of M ( X ); see Theorem 7.1.3. This implies that theforgetful functor U : R X - mod → M ( X )is a left Quillen functor: by the small object argument and by definition of the model categorystructure of Theorem 7.2.2 (i), this follows from the trivial fact that the endofunctor R X ⊗ ( − ) : M ( X ) → M ( X )is a left Quillen functor itself whenever R X is cofibrant in M ( X ). (cid:3) Remark . The results of the preceding proposition (as well as their proofs) are alsotrue in terms of P cart -fibred categories (3.1.21) over the category of S /S -diagrams for any object S of S (whence over all S -diagrams whenever S has a terminal object).7.2.16 . Consider now a noetherian scheme S of finite dimension. We choose a full subcategoryof the category of separated noetherian S -schemes of finite dimension which is stable by finitelimits, contains separated S -schemes of finite type, and such that, for any ´etale S -morphism Y → X , if X is in S /S , so is Y . We denote by S /S this chosen category of S -schemes.We also fix an admissible class P of morphisms of S /S which contains the class of ´etalemorphisms. Definition . A property P of Ho( M ), for M a stable combinatorial P -fibred modelcategory over S /S , is homotopy linear if the following implications are true.(a) If γ : M → M ′ is a Quillen equivalence (i.e. a Quillen morphism which is termwise aQuillen equivalence) between stable combinatorial P -fibred model category over S /S ,then M has property P is and only if M ′ has property P .(b) If M is a stable combinatorial symmetric monoidal P -model category which satisfies themonoid axiom, and such that the unit X of M ( X ) is cofibrant, then, for any cartesianand termwise cofibrant monoid R in M over S /S , R - mod has property P . . BASIC HOMOTOPY COMMUTATIVE ALGEBRA 169 Proposition . The following properties are homotopy linear: A -homotopy invariance, P -stability, the localization property, the property of proper transversality, separability, semi-separability, t -descent (for a given Grothendieck topology t on S /S ). Proof.
Property (a) of the definition above is obvious. Property (b) comes from the factthat the forgetful functors U : Ho( R - mod) → Ho( M )are conservative and commute with all the operations: L f ∗ and R f ∗ for any morphism f , as wellas L f ♯ for any P -morphism (by Proposition 7.2.14). Hence any property formulated in terms ofequations involving only these operations is homotopy linear. (cid:3) art 3 Motivic complexes and relative cycles n this entire part, we adopt the special convention that smooth means smooth separated offinite type. This concerns also the framework of premotivic categories: we assume the admissibleclass Sm is made of smooth separated morphisms of finite type.This assumption is required by the use of the theory of finite correspondences (see moreprecisely Example 9.1.4).
8. Relative cycles . In this entire section, S is the category of noetherian schemes; any scheme is assumedto be noetherian. We fix a subring Λ ⊂ Q which will be the ring of coefficients of the algebraiccycles considered in the following section. Category of cycles. . Let X be a scheme. As usual, an element of the underlying set of X will be called a point and a morphism Spec ( k ) → X where k is a field will be called a geometric point . We oftenidentify a point x ∈ X with the corresponding geometric point Spec ( κ x ) → X . However, theexplicit expression ”the point Spec ( k ) → X ” always refers to a geometric point.As our schemes are assumed to be noetherian, any immersion f : X → Y is quasi-compact. Thus,according to [ EGA1 , 9.5.10], the schematic closure ¯ X of X in Y exists which gives a uniquefactorization of f X j −→ ¯ X i −→ Y such that i is a closed immersion and j is an open immersion with dense image . Note thatwhen Y is reduced, ¯ X coincide with the topological closure of X in Y with its induced reducedsubscheme structure. In this case, we simply call ¯ Y the closure of Y in X . Definition . A Λ -cycle is a couple (
X, α ) such that X is a scheme and α is a Λ-linearcombination of points of X . A generic point of ( X, α ) is a point which appears in the Λ-linearcombination α with a non zero coefficient. The support Supp( α ) of α is the closure of the genericpoints of α .A morphism of Λ-cycles ( Y, β ) → ( X, α ) is a morphism of scheme f : Y → X such that f (Supp( β )) ⊂ Supp( α ). We say this morphism is pseudo-dominant if for any generic point y of( Y, β ), f ( y ) is a generic point of ( X, α ).When considering such a pair (
X, α ), we will denote it simply by α and refer to X as the domain of α . We also use the notation α ⊂ X to mean the domain of the cycle α is the scheme X . The category of Λ-cycle is functorial in Λ with respect to morphisms of integral rings. In whatfollows, cycles are assumed to have coefficients in Λ unless explicitly stated.8.1.3 . Given a property ( P ) of morphisms of schemes, we will say that a morphism f : β → α of cycles satisfies property ( P ) if the induced morphism f | Supp( α )Supp( β ) satisfies property ( P ). Definition . Let X be a scheme. We denote by X (0) the set of generic points of X . Wedefine as usual the cycle associated with X as the cycle with domain X : h X i = X x ∈ X (0) lg( O X,x ) .x. The integer lg( O X,x ), length of an artinian local ring, is called the geometric multiplicity of x in X . When no confusion is possible, we usually omit the delimiters in the notation h X i . As anexample, we say that α is a cycle over X to mean the existence of a structural morphism of cycles α → h X i . Recall the scheme ¯ X is characterized by the property of being the smallest sub-scheme of Y with the existenceof such a factorization. . RELATIVE CYCLES 173 . When Z is a closed subscheme of a scheme X , we denote by h Z i X the cycle h Z i considered as a cycle with domain X .Consider a cycle α with domain X . Let ( Z i ) i ∈ I be the family of the reduced closure of genericpoints of α . Then we can write α uniquely as α = P i ∈ I n i . h Z i i X . We call this writing the standardform of α for short. Definition . Let α = P i ∈ I n i .x i be a cycle with domain X and f : X → Y be anymorphism.For any i ∈ I , put y i = f ( x i ). Then f induces an extension field κ ( x i ) /κ ( y i ) between theresidue fields. We let d i be the degree of this extension field in case it is finite and 0 otherwise.We define the pushforward of α by f as the cycle with domain Yf ∗ ( α ) = X i ∈ I n i d i .f ( x i ) . Thus, when f is an immersion, f ∗ ( α ) is the same cycle as α but seen as a cycle with domain X . Remark also that we obtain the following equality(8.1.6.1) f ∗ (cid:0) h X i (cid:1) = (cid:10) ¯ X (cid:11) Y where ¯ X is the schematic closure of X in Y (indeed X is a dense open subscheme in ¯ X ). When f is clear, we sometimes abusively put: h X i Y := f ∗ ( h X i ).By transitivity of degrees, we obviously have f ∗ g ∗ = ( f g ) ∗ for a composable pair of morphisms( f, g ). Definition . Let α = P i ∈ I n i .x i be a cycle over a scheme S with domain f : X → S and U ⊂ S be an open subscheme. Let I ′ = { i ∈ I | f ( x i ) ∈ U } . We define the restriction of α over U as the cycle α | U = P i ∈ I ′ n i .x i with domain X × S U considered as a cycle over U .If α = P i ∈ I n i . h Z i i X , then obviously α | U = P i ∈ I n i . h Z i × S U i X U . We state the followingobvious lemma for convenience : Lemma . Let S be a scheme, U ⊂ S an open subscheme and X be an S -scheme. Let j : X U → X be the obvious open immersion.(i) For any cycle ( X U , α ′ ) , (cid:0) j ∗ ( α ′ ) (cid:1) | U = α ′ .(ii) Assume ¯ U = S . For any cycle ( X, α ) pseudo-dominant over S , j ∗ ( α | U ) = α . Hilbert cycles. . Recall that a finite dimensional scheme X is equidimensional – we will say absolutelyequidimensional – if its irreducible components have all the same dimension.We will say that a flat morphism f : X → S is equidimensional if it is of finite type and forany connected component X ′ of X , there exists an integer e ∈ N such that for any generic point η in X ′ , the fiber f − [ f ( η )] is absolutely equidimensional of dimension e . Definition . Let S be a scheme.Let α be a cycle over S with domain X . We say that α is a Hilbert cycle over S if there existsa finite family ( Z i ) i ∈ I of closed subschemes of X which are flat equidimensional over S and a finitefamily ( n i ) i ∈ I ∈ Λ I such that α = X i ∈ I n i . h Z i i X . Example . Any cycle over a field k is a Hilbert cycle over Spec ( k ). Let S be thespectrum of a discrete valuation ring. A cycle α = P i ∈ I n i .x i over S is a Hilbert cycle if and onlyif each point x i lies over the generic points of S . Indeed, an integral S -scheme is flat if and onlyif it is dominant.The following lemma follows almost directly from a result of [ SV00b ]: Lemma . Let f : S ′ → S be a morphism of schemes and X be an S -scheme of finitetype. Put X ′ = X × S S ′ .Let ( Z i ) i ∈ I be a finite family of closed subschemes of X such that each Z i is flat equidimen-sional over S . We assume the following relation: (8.1.12.1) X i ∈ I n i . h Z i i X = 0 Then we the following equality holds: X i ∈ I n i . h Z i × S S ′ i X ′ = 0 . Proof.
When we assume that for any index i ∈ I , Z i /S is equidimensional of dimension e ,this lemma is exactly [ SV00b , Prop. 3.2.2]. We show how to reduce to that case.Up to adding more members to the family ( Z i ), we can always assume that Z i is connected.Then, because Z i /S is equidimensional by assumption, there exists an integer e i such that for anypoint x ∈ Z (0) i , the fiber f − [ f ( x )] is absolutely equidimensional of dimension e i . In particularthe transcendence degree d x of the residual extension κ x /κ f ( x ) satisfies the relation: d x = e i .For any integer e ∈ N , we define the following subset of I : I e = { i ∈ I | ∀ x ∈ Z (0) i , d x = e } . Thus ( I e ) e ∈ N is a partition of I .One can rewrite the assumption (8.1.12.1) as follows: for any point x ∈ X , X i ∈ I | x ∈ Z (0) i n i . lg( O Z i ,x ) = 0 . In particular, given any integer e ∈ N , we deduce that the family ( Z i ) i ∈ I e still satisfies the relation(8.1.12.1). As any member of this family is equidimensional of dimension e , we can apply [ SV00b ,Prop. 3.2.2] to ( Z i ) i ∈ I e . This concludes. (cid:3) . Consider a Hilbert S -cycle α ⊂ X and a morphism of schemes f : S ′ → S . Put X ′ = X × S S ′ . We choose a finite family ( Z i ) i ∈ I of flat equidimensional S -schemes and a finitefamily ( n i ) i ∈ I ∈ Λ I such that α = P i ∈ I n i . h Z i i X . The previous lemma says exactly that the cycle X i ∈ I n i . h Z i × S S ′ i X ′ depends only on α and not on the chosen families. Definition . Adopting the preceding notations and hypothesis, we define the pullbackcycle of α along the morphism f : S ′ → S as the cycle with domain X ′ α ⊗ ♭S S ′ = X i ∈ I n i . h Z i × S S ′ i X ′ . In this setting the following lemma is obvious :
Lemma . Let α be a Hilbert cycle over S , and S ′′ → S ′ → S be morphisms of schemes.Then ( α ⊗ ♭S S ′ ) ⊗ ♭S ′ S ′′ = α ⊗ ♭S S ′′ . We will use another important computation from [
SV00b ] (it is a particular case of loc. cit. ,3.6.1).
Proposition . Let R be a discrete valuation ring with residue field k .Let α ⊂ X be a Hilbert cycle over Spec ( R ) and f : X → Y a morphism over Spec ( R ) . We denoteby f ′ : X ′ → Y ′ the pullback of f over Spec ( k ) .Suppose that the support of α is proper with respect to f .Then f ∗ ( α ) is a Hilbert cycle over R and the following equality of cycles holds in X ′ : f ′∗ ( α ⊗ ♭S k ) = f ∗ ( α ) ⊗ ♭S k. . RELATIVE CYCLES 175 Definition . Let p : ˜ S → S be a birational morphism. Let C be the minimal closedsubset of S such that p induces an isomorphism ( ˜ S − ˜ S × S C ) → ( S − C ).Consider α = P i ∈ I n i . h Z i i X a cycle over S written in standard form.We define the strict transform ˜ Z i of the closed subscheme Z i in X along p as the schematicclosure of ( Z i − Z i × S C ) × S ˜ S in X × S ˜ S . We define the strict transform of α along p as the cycleover ˜ S ˜ α = X i ∈ I n i . h ˜ Z i i X × S ˜ S . As in [
SV00b ], we remark that a corollary of the platification theorem of Gruson-Raynaud isthe following :
Lemma . Let S be a reduced scheme and α be a pseudo-dominant cycle over S .Then there exists a dominant blow-up p : ˜ S → S such that the strict transform ˜ α of α along p is a Hilbert cycle over ˜ S . We conclude this part by recalling an elementary lemma about cycles and Galois descentwhich will be used extensively in the next sections :
Lemma . Let
L/K be an extension of fields and X be a K -scheme. We put X L = X × K Spec ( L ) and consider the faithfully flat morphism f : X L → X .Denote by Cycl( X ) (resp. Cycl( X L ) ) the cycles with domain X (resp. X L ).(1) The morphism f ∗ : Cycl( X ) → Cycl( X L ) , β β ⊗ ♭K L is a monomorphism.(2) Suppose L/K is finite. For any K -cycle β ∈ Cycl( X ) , f ∗ ( β ⊗ ♭K L ) = [ L : K ] .β .(3) Suppose L/K is finite normal with Galois group G .The cycles in the image of f ∗ are invariant under the action of G . For any cycle β ∈ Cycl( X L ) G , there exists a unique cycle β K ∈ Cycl( X ) such that β K ⊗ ♭K L = [ L : K ] i .β where [ L : K ] i is the inseparable degree of L/K . Specialization.
The aim of this section is to give conditions on cycles so that one candefine a relative tensor product on them.
Definition . Consider two cycles α = P i ∈ I n i .s i and β = P j ∈ J m i .x j . Let S be thesupport of α .A morphism β f −→ α of cycles is said to be pre-special if it is of finite type and for any j ∈ J ,there exists i ∈ I such that f ( x j ) = s i and n i | m j in Λ. We define the reduction of β/α as thecycle over S β = X j ∈ J,f ( x j )= s i m j n i .x j . Example . Let S be a scheme and α a Hilbert S -cycle. Then the canonical morphismof cycles α → h S i is pre-special. If S is the spectrum of a discrete valuation ring, an S -cycle α ispre-special if and only if it is a Hilbert S -cycle. Definition . Let α be a cycle.A point (resp. trait ) of α will be a morphism Spec ( k ) x −→ α (resp. Spec ( R ) τ −→ α ) such that k is a field (resp. R is a discrete valuation ring). We simply say that x (resp. τ ) is dominant if theimage of the generic point in the domain of α is a generic point of α .Let x : Spec ( k ) → α be a point. An extension of x will be a point y on α of the form Spec ( k ) → Spec ( k ) x −→ α .A fat point of α will be morphismsSpec ( k ) s −→ Spec ( R ) τ −→ α such that τ is a dominant trait and the image of s is the closed point of Spec ( R ).Given a point x : Spec ( k ) → α , a fat point over x is a factorization of x through a dominant traitas above.In the situation of the last definition, we denote simply by ( R, k ) a fat point over x , withoutindicating in the notation the morphisms s and τ . Remark . With our choice of terminology, a point of α is in general an extension of aspecialization of a generic point of α . As a further example, a dominant point of α is an extensionof a generic point of α . Lemma . For any cycle α and any non dominant point x : Spec ( k ) → α , there existsan extension y : Spec ( k ) → α of x and a fat point ( R, k ) over y . Proof.
Replacing α by its support S , we can assume α = h S i . Let s be the image of x in S , κ its residue field. We can assume S is reduced, irreducible by taking one irreducible componentcontaining s , and local with closed point s . Let S = Spec ( A ), K = Frac( A ). According to[ EGA2 , 7.1.7], there exists a discrete valuation ring R such that A ⊂ R ⊂ K , and R/A is anextension of local rings. Then any composite extension k/κ of k and the residue field of R over κ gives the desired fat point ( R, k ). (cid:3) Definition . Let β → α be a pre-special morphism of cycles. Consider S the support of α and X the domain of β . Let β = P j ∈ J m j . h Z j i X be the reduction of β/α written in standardform.(1) Let Spec ( K ) → α be a dominant point. We define the following cycle over Spec ( K )with domain X K = X × S Spec ( K ) : β K = X j ∈ J m j . h Z j × S Spec ( K ) i X K . (2) Let Spec ( R ) τ −→ S be a dominant trait, K be the fraction field of R and j : X K → X R bethe canonical open immersion. We define the following cycle over R with domain X R : β R = j ∗ ( β K ) . According to example 8.1.11, β R is a Hilbert cycle over R .(3) Let x : Spec ( k ) → α be a point on α and ( R, k ) be a fat point over x .We define the specialization of β along the fat point ( R, k ) as the cycle β R,k := β R ⊗ ♭R k using the above notation and definition 8.1.14. It is a cycle over Spec ( k ) with domain X k = X × S Spec ( k ). Remark . Let β ⊂ X be an S -cycle, x : Spec ( K ) → S be a dominant point and U bean open neighborhood of x in S .Then if β is pre-special over S , β | U is pre-special over U and β K = ( β | U ) K .If τ : Spec ( R ) → S (resp. ( R, k )) is a trait (resp. fat point) with generic point x , we also get β R = ( β | U ) R (resp. β R,k = ( β | U ) R,k ).8.1.27 . Let S be a reduced scheme, and β = P i ∈ I n i .x i be an S -cycle with domain X . Forany index i ∈ I , let κ i be the residue field of x i .Consider a dominant point x : Spec ( K ) → S . Let η be its image in S and F be the residuefield of η . We put I ′ = { i ∈ I | f ( x i ) = η } where f : X → S is the structural morphism. Withthese notations, we get β K = X i ∈ I ′ n i . h Spec ( κ i ⊗ F K ) i X K , and for a dominant trait Spec ( R ) → S with generic point x ,(8.1.27.1) β R = X i ∈ I ′ n i . h Spec ( κ i ⊗ F K ) i X R , . RELATIVE CYCLES 177 where Spec ( κ i ⊗ F K ) is seen as a subscheme of X K (resp. X R ).Consider a fat point ( R, k ) with generic point x and write β = P i ∈ I n i . h Z i i X in standardform ( i.e. Z i is the closure of { x i } in X ). Then according to (8.1.6.1), we obtain β R,k = X i ∈ I ′ n i . (cid:10) Z i,K × R Spec ( k ) (cid:11) X k where Z i,K = Z i × S Spec ( K ) is considered as a subscheme of X K and the schematic closure istaken in X R .Considering the description of the schematic closure for the generic fiber of an R -scheme ( cf. [ EGA4 , 2.8.5]), we obtain the following way to compute β R,k . By definition, R is an F -algebra.For i ∈ I ′ , let A i be the image of the canonical morphism κ i ⊗ F R → κ i ⊗ F K. It is an R -algebra without R -torsion. Moreover, the factorizationSpec ( κ i ⊗ F K ) → Spec ( A i ) → Spec ( κ i ⊗ F R )defines Spec ( A i ) as the schematic closure of the left hand side in the right hand side ( cf. [ EGA4 ,2.8.5]). In particular, we get an immersion Spec ( A i ⊗ R k ) → X k and the nice formula : β R,k = X i ∈ I ′ n i . h Spec ( A i ⊗ R k ) i X k . Definition . Consider a morphism of cycles f : β → α and a point x : Spec ( k ) → α .We say that f is special at x if it is pre-special and for any extension y : Spec ( k ) → α of x , forany fat points ( R, k ) and ( R ′ , k ) over y , the equality β R,k = β R ′ ,k holds in X k . Equivalently, wesay that β/α is special at x .We say that f is special (or that β is special over α ) if it is special at every point of α . Remark . (1) Trivially, f is special at every dominant point of α .(2) Given an extension y of x , it is equivalent for f to be special at x or at y (use Lemma8.1.19(1)). Thus, in the case where α = h S i , we can restrict our attention to the points s ∈ S .(3) According to 8.1.26, the property that β/S is special at s ∈ S depends only on an openneighbourhood U of s in S . More precisely, the following conditions are equivalent :(i) β is special at s over S .(ii) β | U is special at s over U . Example . Let S be a scheme and β be a Hilbert cycle over S . We have already seenthat β → h S i is pre-special. The next lemma shows this morphism is in fact special. Lemma . Let S be a scheme and β be a Hilbert cycle over S . Consider a point x :Spec ( k ) → S and a fat point ( R, k ) over x .Then β R,k = β ⊗ ♭S k . Proof.
According to the preceding definition and Lemma 8.1.15 it is sufficient to prove β R = β ⊗ ♭S R . As the two sides of this equation are unchanged when replacing β by the reduction β of β/S , we can assume that S is reduced. By additivity, we are reduced to the case where β = h X i is the fundamental cycle associated with a flat S -scheme X . According to 8.1.6.1, β R = (cid:10) X K (cid:11) X R . Applying now [ EGA4 , 2.8.5], X K is the unique closed subscheme Z of X R suchthat Z is flat over Spec ( R ) and Z × R Spec ( K ) = X K . Thus, as X R is flat over Spec ( R ), we get X K = X R and this concludes. (cid:3) Lemma . Let p : ˜ S → S be a birational morphism and consider a commutative diagram ˜ S p (cid:15) (cid:15) Spec ( k ) / / Spec ( R ) ✐✐✐✐ * * ❱❱❱❱ S This shows that our definition coincide with the one given in [
SV00b ] (p. 23, paragraph preceding 3.1.3)in the case where α = h S i , S reduced. such that ( R, k ) is a fat point of ˜ S and S .Consider a pre-special cycle β over S and ˜ β its strict transform along p . Then, ˜ β is pre-specialand ˜ β R,k = β R,k . Proof.
Using 8.1.26, we reduce to the case where p is an isomorphism which is trivial. (cid:3) Lemma . Let S be a reduced scheme, x : Spec ( k ) → S be a point and α be a pre-specialcycle over S . Let p : ˜ S → S be a dominant blow-up such that the strict transform ˜ α of α along p is a Hilbert cycle over ˜ S . Then the following conditions are equivalent :(i) α is special at x .(ii) for every points x , x : Spec ( k ) → ˜ S such that p ◦ x = p ◦ x and p ◦ x is an extensionof x , ˜ α ⊗ ♭ ˜ S x = ˜ α ⊗ ♭ ˜ S x . Proof.
The case where x is a dominant point follows from the definitions and the fact p isan isomorphism at the generic point. We thus assume x is non dominant.( i ) ⇒ ( ii ) : Applying Lemma 8.1.24 to x i , i = 1 ,
2, we can find an extension x ′ i : Spec ( k i ) → ˜ S of x i and a fat point ( R i , k i ) over x ′ i . Taking a composite extension L of k and k over k , we canfurther assume L = k = k and p ◦ x ′ = p ◦ x ′ . Then for i = 1 ,
2, we get (cid:0) ˜ α ⊗ ♭ ˜ S x i (cid:1) ⊗ ♭k L . . α ⊗ ♭ ˜ S x ′ i . . α R i ,L . . α R i ,L , and this concludes according to 8.1.19(1).( ii ) ⇒ ( i ) : Consider an extension y : Spec ( k ) → α over x and two fat point ( R , k ), ( R , k ) over y . Fix i ∈ { , } . As p is proper birational, the trait Spec ( R i ) on S can be extended (uniquely)to ˜ S . Let x i : Spec ( k ) → Spec ( R i ) → ˜ S be the induced point. Then the following computationallows to conclude : α R i ,k . . α R i ,k . . α ⊗ ♭ x i (cid:3) Pullback. . In this part, we construct a pullback which extends the pullback defined by Suslinet Voevodsky in [
SV00b , 3.3.1] to the case of morphism of cycles. Consider the situation of adiagram of cycles β f (cid:15) (cid:15) X (cid:15) (cid:15) ⊂ α ′ / / α S ′ / / S where the diagram on the right is the domain of the one on the left. Let n be exponentialcharacteristic of Supp( α ′ ).The pullback of β , considered as an α -cycle, over α ′ will be a Λ[1 /n ]-cycle denoted by β ⊗ α α ′ .It will fits into the following commutative diagram of cycles β ⊗ α α ′ / / (cid:15) (cid:15) β (cid:15) (cid:15) X × S S ′ / / (cid:15) (cid:15) X (cid:15) (cid:15) ⊂ α ′ / / α S ′ / / S where the right commutative square is again the support of the left one.It will be defined under an assumption on β/α and is therefore non symmetric . This assump-tion will imply that β/α is pre-special, and the first property of β ⊗ α α ′ is that it is pre-specialover α ′ .We define this product in three steps in which the following properties will be a guideline :(P1) Let S be the support of α and β be the reduction of β/α as an S -cycle. Consider thecanonical factorization α ′ → S → α .Then, β ⊗ α α ′ = β ⊗ S α ′ . See further 8.2.3 for this question. All these properties except (P3) will be particular cases of the associativity of the pullback. . RELATIVE CYCLES 179 (P2) Consider a commutative diagramSpec ( E ) / / Spec ( R ′ ) / / (cid:15) (cid:15) ( ∗ ) Spec ( R ) (cid:15) (cid:15) α ′ / / α such that ( R, E ) (resp. ( R ′ , E )) is a fat point on α (resp. α ′ ).Then, ( β ⊗ α α ′ ) R ′ ,E = β R,E .Assume α ′ → α = h S ′ → S i .(P3) If β is a Hilbert cycle over S , β ⊗ S S ′ = β ⊗ ♭S S ′ .(P4) Consider a factorization S ′ → U j −→ S such that j is an open immersion. Then β ⊗ S S ′ = β | U ⊗ U S ′ .(P5) Consider a factorization S ′ → ˜ S p −→ S such that p is a birational morphism. Then β ⊗ S S ′ = ˜ β ⊗ ˜ S S ′ . Lemma . Consider the hypothesis of 8.1.34 in the case where α ′ = Spec ( k ) is a point x of α .We suppose that f is special at x .Then the pre-special Λ[1 /n ] -cycle β ⊗ α k exists and is uniquely determined by property (P2)above. We also put β k := β ⊗ α k .The properties (P1) to (P5) are fulfilled and in addition :(P6) For any extension fields L/k , β L = β k ⊗ ♭k L . Proof.
According to Lemma 8.1.24 there always exists a fat point (
R, E ) over an extensionof x . Thus the unicity statement follows from 8.1.19(1).For the existence, we first consider the case where α = h S i is a reduced scheme. ApplyingLemma 8.1.18, there exists a blow-up p : ˜ S → S such that the strict transform ˜ β of β along p is aHilbert cycle over ˜ S .As p is surjective, the fiber ˜ S k is a non empty algebraic k -scheme. Thus, it admits a closedpoint given by a finite extension k ′ of k . Let k ′ /k be a normal closure of k ′ /k and G be its Galoisgroup. As β/S is special at x by hypothesis, Lemma 8.1.33 implies that ˜ β ⊗ ♭ ˜ S k ′ is G -invariant.Thus, applying Lemma 8.1.19, there exists a unique cycle β k ⊂ X k with coefficients in Λ[1 /n ] suchthat β k ⊗ ♭k k ′ = ˜ β ⊗ ♭ ˜ S k ′ .We prove (P2). Given a diagram ( ∗ ) with α ′ = Spec ( k ), we first remark that ( β k ) R ′ ,E = β k ⊗ ♭k E . As p is proper birational, the dominant trait Spec ( R ) → S lifts to a dominant traitSpec ( R ) → ˜ S . Let E ′ /k be a composite extension of k ′ /k and E/k . With these notations, we getthe following computation : β R,E ⊗ ♭E E ′ . . β R,E ⊗ ♭E E ′ . . β ⊗ ♭ ˜ S E ′ . .
15( ˜ β ⊗ ♭ ˜ S k ′ ) ⊗ ♭E E ′ β k ⊗ ♭k E ′ , so that we can conclude by applying 8.1.19(1).In the general case, we consider he support S of α abd β /S the reduction of β/α . Accordingto (P1), we are led to put β k := ( β ) k with the help of the preceding case. Considering thedefinition of specialization along fat points, we easily check this cycle satisfies property (P2).Finally, property (P6) (resp. (P3), (P5)) follows from the unicity statement applying lemmas8.1.24, 8.1.19(1) (resp. and moreover Lemma 8.1.31, 8.1.32). (cid:3) Remark . In the case where x is a dominant point, the cycle β k defined in the previousproposition agrees with the one defined in 8.1.25(1). Lemma . Consider the hypothesis of 8.1.34 in the case where α ′ = Spec ( O ) is a traitof α . Let K be the fraction field of O and x the corresponding point on α .We suppose that f is special at x .Then the pre-special Λ[1 /n ] -cycle β ⊗ α O exists and is uniquely defined by the property ( β ⊗ α O ) ⊗ ♭O K = β K with the notations of the preceding lemma. We also put β O := β ⊗ α O . The properties (P1) to (P5) are fulfilled and in addition :(P6’) For any extension O ′ /O of discrete valuation rings, β O ′ = β O ⊗ ♭O O ′ . Proof.
Remark that, with the notation of definition 8.1.7, β O ⊗ ♭O K = β O | Spec( K ) . For thefirst statement, we simply apply Lemma 8.1.8 and put β O = j ∗ ( β K ) where j : X K → X O is thecanonical open immersion.Then properties (P1), (P3), (P4), (P5) and (P6’) of the case considered in this lemma followseasily from the uniqueness statement and the corresponding properties in the preceding lemma(applying again 8.1.8).It remains to prove (P2). According to (P1), we reduce to the case α = h S i for a reducedscheme S . We choose a birational morphism p : ˜ S → S such that the proper transform ˜ β is aHilbert ˜ S -cycles. Consider a diagram of the form ( ∗ ) in this case. According to property (P3), wecan assume R ′ = O .Remark the trait Spec ( R ) → S admits an extension Spec ( R ) → ˜ S as p is proper. The point x admits an extension K ′ /K which lifts to a point x ′ : Spec ( K ′ ) → ˜ S – again ˜ S K is a non emptyalgebraic scheme. The discrete valuation corresponding to O ⊂ K extends to a discrete valuationon K ′ as K ′ /K is finite. Let O ′ ⊂ K ′ be the corresponding valuation ring. The corresponding traitSpec ( O ′ ) → S thus admits a lifting to ˜ S corresponding to the point x ′ as p is proper. Consideringa composite extension E ′ /K of K ′ /K and E/K , we have obtained a commutative diagramSpec ( E ′ ) / / Spec ( O ′ ) / / Spec ( R ) (cid:15) (cid:15) Spec ( O ′ ) / / ˜ S which lifts our original diagram ( ∗ ). Let x (resp. x ) be the point Spec ( E ) ′ → ˜ S correspondingto the the composite through the upper way (resp. lower way) in the preceding diagram.Then, β R,E ⊗ ♭E E ′ = ˜ β x . Moreover, we get( β ⊗ S O ) O,E ⊗ ♭E E ′ . . β ⊗ S O ) ⊗ ♭O E ′ ( P P ′ ) ( ˜ β ⊗ ˜ S O ′ ) ⊗ ♭O ′ E ′ ( P ˜ β x . By hypothesis, β/α is special at Spec ( K ′ ) → S . Thus Lemma 8.1.33 concludes. (cid:3) Theorem . Consider the hypothesis of 8.1.34.Assume f is special at the generic points of α ′ .Then the pre-special Λ[1 /n ] -cycle β ⊗ α α ′ exists and is uniquely determined by property (P2).It satisfies all the properties (P1) to (P5). Proof.
According to Lemma 8.1.24, for any point s of S ′ with residue field κ , there existsan extension E/κ and a fat point (
R, E ) (resp. ( R ′ , E )) of α (resp. α ′ ) over Spec ( E ) → α (resp.Spec ( E ) → α ′ ). The uniqueness statement follows by applying Lemma 8.1.19(1).For the existence, we write α ′ = P i ∈ I n i . h Z i i S ′ in standard form.For any i ∈ I , let K i be the function field of Z i and consider the canonical morphismSpec ( K i ) → α . Let β K i ⊂ X K i be the Λ[1 /n ]-cycle defined in lemma 8.1.35. Let j i : X K i → X ′ be the canonical immersion and put :(8.1.38.1) β ⊗ α α ′ = X i ∈ I n i .j i ∗ ( β K i ) . Then properties (P1), (P3), (P4) and (P5) are direct consequences of this definition and ofthe corresponding properties of Lemma 8.1.35.We check property (P2). Given a diagram of the form ( ∗ ), there exists a unique i ∈ I such thatSpec ( R ′ ) dominates Z i . Thus we get for this choice of i ∈ I that ( β ⊗ α α ′ ) R ′ ,E = (cid:0) j i ∗ ( β K i ) (cid:1) R ′ ,E .Let K ′ be the fraction field of R ′ and consider the open immersion j ′ : X K ′ → X R ′ . The followingcomputation then concludes : (cid:0) j i ∗ ( β K i ) (cid:1) R ′ ,E j ′∗ (cid:0) j i ∗ ( β K i ) K ′ ) ⊗ ♭R ′ E . . j ′∗ ( β K ′ ) ⊗ ♭R ′ E . . β R ′ ⊗ ♭R ′ E . . ( P β R,E . (cid:3) . RELATIVE CYCLES 181 Definition . In the situation of the previous theorem, we call the Λ[1 /n ]-cycle β ⊗ α α ′ the pullback of β/α by α ′ .8.1.40 . By construction, the cycle β ⊗ α α ′ is bilinear with respect to addition of cycles in thefollowing sense:(P7) Consider the hypothesis of 8.1.34. Let α ′ , α ′ be cycles with domain S ′ such that α = α ′ + α ′ . If β/α is special at the generic points of α and α , then the following cyclesare equal in X × S S ′ : β ⊗ α ( α ′ + α ′ ) = β ⊗ α α ′ + β ⊗ α α ′ . (P7’) Consider the hypothesis of 8.1.34. Let β , β be cycles with domain X such that β = β + β . If β and β are special over α at the generic points of α ′ , then β/α is specialat the generic points of α ′ and the following cycles are equal in X × S S ′ :( β + β ) ⊗ α α ′ = β ⊗ α α ′ + β ⊗ α α ′ . In the theorem above, we can assume that X (resp. S , S ′ ) is the support of β (resp. α , α ′ ).Thus the support of β ⊗ α α ′ is included in X × S S ′ . More precisely: Lemma . Consider the hypothesis of 8.1.34 and assume that X (resp. S , S ′ ) is thesupport of β (resp. α , α ′ ). Then, if β/α is special at the generic points of α ′ , we obtain:(i) Let ( X × S S ′ ) (0) be the generic points of X × S S ′ . Then, we can write β ⊗ α α ′ = X x ∈ ( X × S S ′ ) (0) m x .x (ii) For any generic point x of X × S S ′ , if m x = 0 , the image of x in S ′ is a generic point s ′ and the multiplicity of s ′ in α ′ divides m x in Λ[1 /n ] . Proof.
Point (ii) is just a traduction that β ⊗ α α ′ is pre-special over α ′ . For point (i), wereduce easily to the case where α is the scheme S and S is reduced. We can also assume that α ′ is the spectrum of a field k . It is sufficient to check point (i) after an extension of k . Thus wecan apply Lemma 8.1.18 to reduce to that case where β is a Hilbert cycle over S . This case isobvious. (cid:3) Definition . In the situation of the previous lemma, we put m SV ( x ; β ⊗ α α ′ ) := m x ∈ Λ[1 /n ]and we call them the Suslin-Voevodsky multiplicities (in the operation of pullback). Remark . Consider the notations of the previous lemma:(1) Assume that α is the spectrum of a field k . Then the product β ⊗ k α ′ is always definedand agrees with the classical exterior product (according to (P3)).(2) According to the previous lemma, the irreducible components of X × S S ′ which doesnot dominate an irreducible component of S ′ have multiplicity 0: they correspond to the”non proper components” with respect to the operation β ⊗ α α ′ .(3) Assume α ′ → α = h S ′ p −→ S i , β = P i ∈ I n i .x i . Let y be a generic point of X × S S ′ lyingover a generic point s ′ of S ′ . Let S ′ be the irreductible component of S ′ correspondingto s ′ . Consider any irreductible component S of S which contains p ( s ′ ) and let β = P i n i .x i where the sums runs over the indexes i such that x i lies over S . Then, accordingto (8.1.38.1), m SV ( y ; β ⊗ S h S ′ i ) = m SV ( y ; β ⊗ S h S ′ i ) . This is a key property of the Suslin-Voevodsky multiplicities which explains why we haveto consider the property that β/α is special at s ′ (see 8.3.25 for a refined statement). Lemma . Consider a morphism of cycles α ′ → α and a pre-special morphism f : β → α which is special at the generic points of α . Consider a commutative square Spec ( k ′ ) x ′ / / (cid:15) (cid:15) α ′ (cid:15) (cid:15) Spec ( k ) x / / α such that k and k ′ are fields. Then the following conditions are equivalent :(i) f is special at x .(ii) β ⊗ α α ′ → α ′ is special at x ′ . Proof.
This follows easily from Lemma 8.1.24 and property (P2). (cid:3)
Corollary . Let f : β → α be a special morphism.Then for any morphism α ′ → α , β ⊗ α α ′ → α ′ is special. Definition . Let f : β → α be a morphism of cycles and x : Spec ( k ) → α be a point.We say that f is Λ-universal at x if it is special at x and the cycle β ⊗ α k has coefficients in Λ.In the situation of this definition, let s be the image of x in the support of α , and κ s be itsresidue field. Then according to (P6), β k = β κ s ⊗ ♭κ s k . Thus f is Λ-universal at x if and only if itis Λ-universal at s . Furthermore, the following lemma follows easily : Lemma . Let f : β → α be a morphism of cycles. The following conditions are equiva-lent : (i) For any point s ∈ α , f is Λ -universal at s .(ii) For any point x : Spec ( k ) → α , f is Λ -universal at x .(iii) For any morphism of cycles α ′ → α , β ⊗ α α ′ has coefficients in Λ . Definition . We say that a morphism of cycles f is Λ-universal if it satisfies theequivalent properties of the preceding lemma.Of course, Λ-universal morphisms are stable by base change. These definitions will be appliedsimilarly to morphisms of schemes by considering the associated morphism of cycles. Example . According to property (P3) of the pullback, a flat equidimensional morphismof schemes is Λ-universal.
Commutativity.
Lemma . Consider morphisms of cycles with support in the left diagram β (cid:15) (cid:15) X f (cid:15) (cid:15) ⊂ γ / / α T g / / S such that β/α is pre-special and γ/α is pseudo-dominant.Assume α = X i ∈ I n i .s i , β = X j ∈ J m j .x j , γ = X l ∈ H p l .t l and denote by κ s i (resp. κ x j , κ t l ) the residue field of s i (resp. x j , t l ) in S (resp. X , T ). Consid-ering ( i, j, l ) ∈ I × J × H such that f ( x j ) = g ( t l ) = s i , we denote by ν j,l : Spec (cid:0) κ x j ⊗ κ si κ t l (cid:1) → X × S T the canonical immersion.Then the following assertions hold :(i) β is special at the generic points of γ .(ii) The cycle β ⊗ α γ has coefficients in Λ . . RELATIVE CYCLES 183 (iii) The following equality of cycles holds β ⊗ α γ = X i,j,l m j n i p l .ν j,l ∗ (cid:0) h Spec (cid:0) κ y j ⊗ κ xi κ z l (cid:1) i (cid:1) where the sum runs over ( i, j, l ) ∈ I × J × H such that f ( x j ) = g ( t j ) = s i . Proof.
Assertion (i) is in fact the first point of 8.1.29. Assertion (ii) follows from assertion(iii), which is a consequence of the defining formula (8.1.38.1) and remark 8.1.36. (cid:3)
Corollary . Let g : T → S be a flat morphism and β = P j ∈ J m j . h Z j i X be a pre-special S -cycle written in standard form.Then β/S is pre-special at the generic points of T and β ⊗ S h T i = X j ∈ J m j . h Z j × S T i . The pullback β ⊗ α γ , at it is defined only when β/α is special, is in general non symmetric in β and γ . However the previous lemma implies it is symmetric whenever it makes sense : Corollary . Consider pre-special morphisms of cycles β → α and γ → α .Then β (resp. γ ) is special at the generic points of γ (resp. β ) and the following equalityholds: β ⊗ α γ = γ ⊗ α β . Associativity.
Proposition . Consider morphism of cycles β f −→ α , α ′′ → α ′ → α such that f is specialat the generic points of α ′ and of α ′′ . Let n be the exponential characteristic of α ′′ .Then the following assertions hold:(i) The relative cycle ( β ⊗ α α ′ ) /α ′ is special at the generic points of α ′′ .(ii) The cycle ( β ⊗ α α ′ ) ⊗ α ′ α ′′ has coefficients in Λ[1 /n ] .(iii) ( β ⊗ α α ′ ) ⊗ α ′ α ′′ = β ⊗ α α ′′ . Proof.
Assertion (i) is a corollary of Lemma 8.1.44. Assertion (ii) is in fact a corollary ofassertion (iii), which in turn follows easily from the uniqueness statement in theorem 8.1.38. (cid:3)
Lemma . Let γ g −→ β f −→ α be two pre-special morphisms of cycles with domains Y → X → S . Consider a fat point ( R, k ) over α such that γ/β is special at the generic points of β R,k .Then γ/α is pre-special and the following equality of cycles holds in Y k : γ R,k = γ ⊗ β ( β R,k ) . Proof.
The first statement is obvious.We first prove: γ R = γ ⊗ β β R .Remark that β R → β is pseudo-dominant. Thus γ/β is special at the generic points of β R and theright hand side of the preceding equality is well defined. Moreover, according to Lemma 8.2.1, wecan restrict to the case where α = s , β = x and γ = y , with multiplicity 1. Let κ s , κ x , κ y be thecorresponding respective residue fields, and K be the fraction field of R .Then, according to (8.1.27.1), γ R = h κ y ⊗ κ s K i Y R and β R = h κ x ⊗ κ s K i X R . But Lemma 8.2.1implies that γ ⊗ β β R = h κ y ⊗ κ x ( κ x ⊗ κ s K ) i X R . Thus the associativity of the tensor product offields allows to conclude.From this equality and Proposition 8.2.4, we deduce that: γ R ⊗ β R β R,k = ( γ ⊗ β β R ) ⊗ β R β R,k = γ ⊗ β β R,k . Thus, the equality we have to prove can be written γ R ⊗ ♭R k = γ R ⊗ β R ( β R ⊗ ♭R k ) and we arereduced to the case α = Spec ( R ).In this case, we can assume β = h X i with X integral. Let us consider a blow-up ˜ X p −→ X suchthat the proper transform ˜ γ of γ along p is a Hilbert cycle over ˜ X (8.1.18). We easily get (from(P3) and 8.1.15) that ˜ γ k = ˜ γ ⊗ ˜ X h ˜ X k i . Let Y (resp. ˜ Y ) be the support of γ (resp. ˜ γ ), q : ˜ Y → Y the canonical projection. We considerthe cartesian square obtained by pullback along Spec ( k ) → Spec ( R ):˜ Y k q k / / (cid:15) (cid:15) Y k (cid:15) (cid:15) ˜ X k p k / / X k . As X k ⊂ X (resp. Y k ⊂ Y ) is purely of codimension 1, the proper morphism p k (resp. q k ) is stillbirational. As a consequence, q k ∗ (˜ γ ) = γ . Let y be a point in ˜ Y (0) k ≃ Y (0) k which lies above a point x in ˜ X (0) k ≃ X (0) k Then, according to (P5) and using the notations of 8.1.42, we get m SV ( y ; ˜ γ ⊗ ˜ X h ˜ X k i ) = m SV ( y ; γ ⊗ X h X k i ) . This readily implies q k ∗ (˜ γ ⊗ ˜ X h ˜ X k i ) = γ ⊗ X h X k i and allows us to conclude. (cid:3) As a corollary of this lemma using the uniqueness statement in Theorem 8.1.38, we obtained :
Corollary . Let γ g −→ β f −→ α be pre-special morphisms of cycles.Let x : Spec ( k ) → α be a point. If β/α is special (resp. Λ -universal) at x and γ/β is special(resp. Λ -universal) at the generic points of β k , then γ/α is special at x .Let α ′ → α be any morphism of cycles with domain S ′ → S and n be the exponential charac-teristic of α ′ . Then, whenever it is well defined, the following equality of Λ[1 /n ] -cycles holds: γ ⊗ β ( β ⊗ α α ′ ) = γ ⊗ α α ′ . A consequence of the transitivity formulas is the associativity of the pullback :
Corollary . Suppose given the following morphisms of cycles α (cid:29) (cid:29) ❁❁❁❁ β f (cid:1) (cid:1) ✂✂✂✂ (cid:30) (cid:30) ❂❂❂❂ γ g (cid:0) (cid:0) ✂✂✂✂ δ σ such that f and g are pre-specials.Then, whenever it is well defined, the following equality of cycles hold: γ ⊗ σ ( β ⊗ δ α ) = ( γ ⊗ σ β ) ⊗ δ α Proof.
Indeed, by the transitivity formulas 8.2.4 and 8.2.6, both members of the equationare equal to ( γ ⊗ σ β ) ⊗ β ( β ⊗ δ α ). (cid:3) Projection formulas.
Proposition . Consider morphisms of cycles with support in the left diagram β (cid:15) (cid:15) X (cid:15) (cid:15) ⊂ α ′ / / α S ′ q / / S such that β/α is special at the generic points of α ′ .Consider a factorization S ′ g −→ T → S .Then β/α is special at the generic points of g ∗ ( α ) and the following equality of cycles holds in X × S T : β ⊗ α g ∗ ( α ′ ) = (1 X × S g ) ∗ ( β ⊗ α α ′ ) . Proof.
The first assuption is obvious. By linearity, we can assume S ′ is integral and α ′ isthe generic point s of S ′ with multiplicity 1. Let L (resp. E ) be the residue field of s (resp. g ( s )).Consider the pullback square X L g / / j (cid:15) (cid:15) X Ei (cid:15) (cid:15) X × S S ′ g X / / X × S T where i and j are the natural immersions. . RELATIVE CYCLES 185 Let d be the degree of L/E if it is finite and 0 otherwise. We are reduced to prove the equality g X ∗ ( j ∗ ( β L )) = d.i ∗ ( β E ). Using the functoriality of pushforward and property (P6), it is sufficientto prove the equality g ∗ ( β E ⊗ ♭E L ) = d.β E . If d = 0, the morphism g induces an infinite extensionof fields on any point of X L which concludes. If L/E is finite, g is finite flat and β E ⊗ ♭E L is theusual pullback by g . Then the needed equality follows easily (see [ Ful98 , 1.7.4]). (cid:3)
Lemma . Let β → α be a pre-special morphism of cycles with domain X p −→ S . Let ( R, k ) a fat point over α and X f −→ Y → S be a factorization of p . Let f k be the pullback of f over Spec ( k ) .Suppose that the support of β is proper with respect to f . Then f ∗ ( β ) is pre-special over α and the equality of cycles (cid:0) f ∗ ( β ) (cid:1) R,k = f k ∗ ( β R,k ) holds in Y k . Proof.
As usual, considering the support S of α , we reduce to the case where α = h S i . Let K be the fraction field of R . As Spec ( K ) maps to a generic point of S , we can assume S isintegral. Let F be its function field. We can assume by linearity that β is a point x in X withmultiplicity 1.Let L (resp. E ) be the residue field of x (resp. y = f ( x )). Let d be the degree of L/E if it isfinite and 0 otherwise. Consider the following pullback squareSpec ( L ⊗ F K ) j / / f (cid:15) (cid:15) X × S Spec ( R ) = X Rf R (cid:15) (cid:15) Spec ( E ⊗ F K ) i / / Y × S Spec ( R ) = Y R . According to the formula (8.1.27.1), we obtain: f R ∗ ( β R ) = f R ∗ j ∗ ( h L ⊗ F K i ) = i ∗ f ∗ ( h L ⊗ F K i )= i ∗ f ∗ ( f ∗ ( h E ⊗ F K i ) = i ∗ ( d. h E ⊗ F K i ) = h f ∗ ( β ) i R . We are finally reduced to the case S = Spec ( R ) and β is a Hilbert cycle over Spec ( R ). Notethat f ∗ ( β ) is still a Hilbert cycle over Spec ( R ). As β R,k = β ⊗ ♭R k , the result follows now fromProposition 8.1.16. (cid:3) Corollary . Consider morphisms of cycles with support in the left diagram β (cid:15) (cid:15) X p (cid:15) (cid:15) ⊂ α ′ / / α S ′ / / S such that β/α is special at the generic points of α ′ (resp. Λ -universal).Consider a factorization X f −→ Y → S of p .Suppose that the support of β is proper with respect to f . Then f ∗ ( β ) /α is special at thegeneric points of α ′ (resp. Λ -universal) and the following equality of cycles holds in X × S S ′ : ( f × S S ′ ) ∗ ( β ⊗ α α ′ ) = (cid:0) f ∗ ( β ) (cid:1) ⊗ α α ′ . . We introduce a notation which will come often in the next section. Let S be a schemeand α = P i ∈ I n i . h Z i i X an S -cycle written in standard form.Let s be a point of S and Spec ( k ) ¯ s −→ S be a geometric point of S with k separably closed.Let S ′ be one of the following local schemes: the localization of S at s , the Hensel localization of S at s , the strict localization of S at ¯ s .We then define the cycle with coefficients in Λ and domain X × S S ′ as: α | S ′ = X i ∈ I n i h Z i × S S ′ i X × S S ′ . Remark . The canonical morphism S ′ → S is flat. In particular, α/S is special at thegeneric points of S ′ and we easily get: α | S ′ = α ⊗ S S ′ . Constructibility.
Definition . Let S be a scheme and s ∈ S a point. We say that a pre-special S -cycle α is emphtrivial at s if it is special at s and α ⊗ S s = 0.Naturally, we say that α is trivial if it is zero. Thus α is trivial if and only if it is trivial atthe generic points of S .Recall from [ EGA4 , 1.9.6] that an ind-constructible subset of a noetherian scheme X is aunion of locally closed subset of X . Lemma . Let S be a noetherian scheme, and α/S be a pre-special cycle. Then the set T = (cid:8) s ∈ S | α/S is special (resp. trivial, Λ -universal) at s (cid:9) is ind-constructible in S . Proof.
Let s be a point of T , and Z be its closure in S with its reduced subscheme structure.Put α Z = α ⊗ S Z , defined because α is special at the generic point of Z . Given any point t of Z ,we know that α/S is special at t if and only if α Z /Z is special at t ( cf. U s of Z such that α Z | U Z is a Hilbert cycle over U Z . Thus, α/S is specialat each point of U s and U s ⊂ T . This concludes and the same argument proves the respectivestatements. (cid:3) . Let I be a left filtering category and ( S i ) i ∈ I be a projective system of noetherianschemes with affine transition morphisms. We let S be the projective limit of ( S i ) and we assumethe followings:(1) S is noetherian.(2) There exists an index i ∈ I such that the canonical projection S p i −→ S i is dominant.In this case, there exists an index j/i such that for any k/j , the map p k induces an isomorphism S (0) → S (0) k on the generic points ( cf. [ EGA4 , 8.4.1]). Thus, replacing I by I/j , we can assumethat this property is satisfied for all index i ∈ I . As a consequence, the following properties areconsequences of the previous ones:(3) For any i ∈ I , p i : S → S i is pseudo-dominant and p i induces an isomorphism S (0) → S (0) i .(4) For any arrow j → i of I , p ji : S j → S i is pseudo-dominant and p ji induces an isomor-phism S (0) j → S (0) i . Proposition . Consider the notations and hypothesis above. Assume we are given aprojective system of cycles ( α i ) i ∈ I such that α i is a pre-special cycle over S i and for any j → i , α j = α i ⊗ S i S j . Put α = α i ⊗ S i S for an index i ∈ I . The following conditions are equivalent:(i) α/S is special (resp. Λ -universal).(ii) There exists i ∈ I such that α i /S i is special (resp. Λ -universal).(iii) There exists i ∈ I such that for all j/i , α j /S j is special (resp. Λ -universal).Let s be point of S and s i its image in S i . Then the following conditions are equivalent:(i) α/S is special (resp. Λ -universal) at s .(ii) There exists i ∈ I such that α i /S i is special (resp. Λ -universal) at s i .(iii) There exists i ∈ I such that for all j/i , α j /S j is special (resp. Λ -universal) at s j . Proof.
Let P be one of the respective properties: “special”, “trivial”, “Λ-universal”. Usingthe fact that being P at s is an ind-constructible property (from Lemma 8.3.4), it is sufficient toapply [ EGA4 , th. 8.3.2] to the following family of sets: F i = { s i ∈ S i | α i satisfies P at s i } , F = { s ∈ S | α satisfies P at s } . The pullback is well defined because of point (3) and (4) of the hypothesis above. . RELATIVE CYCLES 187
To get the two sets of equivalent conditions of the statement from op. cit. we have to prove thefollowing relations: (1) : ∀ ( j → i ) ∈ Fl( I ) , p − ji ( F i ) ⊂ F j , (2) : F = ∪ i ∈ I p − i ( F i ) . We consider the case where P is the property “special”. For relation (1), we apply 8.1.44 whichimplies the stronger relation p − ji ( F i ) = F j . For relation (2), another application of 8.1.44 gives infact the stronger relation F = p − i ( F i ) for any i ∈ I .Consider a point s j ∈ S and put s i = p ji ( s j ). Assume α i is special at s i . Then, applying 8.2.4and (P3), we get:(8.3.6.1) α j ⊗ S j s j = ( α i ⊗ S i s i ) ⊗ ♭κ ( s i ) κ ( s j ) . Similarly, given s ∈ S j , s i = p i ( s ), and assuming α i is special at s i , we get:(8.3.6.2) α ⊗ S s = ( α i ⊗ S i s i ) ⊗ ♭κ ( s i ) κ ( s ) . We consider now the case where P is the property “trivial”. Then relation (1) follows from(8.3.6.1). Relation (2) follows from (8.3.6.1) and 8.1.19(1).We finally consider the case P is the property “Λ-universal”. Relation (1) in this case is againa consequence of (8.3.6.1). According to (8.3.6.2), we get the inclusion ∪ i ∈ I f − i ( F i ) ⊂ F . Wehave to prove the reciprocal inclusion.Consider a point s ∈ S with residue field k such that α/S is Λ-universal at s . For any i ∈ I , weput s i = p i ( s ) and denote by k i its residue field. It is sufficient to find an index i ∈ I such that α i ⊗ S i s i has coefficients in Λ. Thus we are reduced to the following lemma: Lemma . Let ( k i ) i ∈ I op be an ind-field and put: k = lim −→ i ∈ I op k i .Consider a family ( β i ) i ∈ I such that β i is a k i -cycle of finite type with coefficients in Q and forany j/i , β j = β i ⊗ ♭k i k j . We put β = β i ⊗ ♭k i k .If for an index i ∈ I , β i ⊗ ♭k i k has coefficients in Λ , then there exists j/i such that β j hascoefficients in Λ . We can assume that for any j/i , β j has positive coefficients. Let X j (resp. X ) be the supportof β j (resp. β ). We obtain a pro-scheme ( X j ) j/i such that X = lim ←− i ∈ I X i . The transition mapsof ( X j ) j/i are dominant. Thus, by enlarging i , we can assume that for any j/i , the induced map π ( X i ) → π ( X j ) is a bijection. Thus we can consider each element of π ( X ) separately andassume that all the X i are integrals: for any j/i , β j = n j . h X j i for a positive element n j ∈ Q .Arguing generically, we can further assume X j = Spec ( L j ) for a field extension of finite type L j of k j . By assumption now, for any j/i , L i ⊗ k i k j is an Artinian ring whose reduction is the field L j . Moreover, n j = n i . lg( L i ⊗ k i k j ) and we know that n := n i . lg( L i ⊗ k i k ) belongs to Λ.Let p be a prime not invertible in Λ such that v p ( n i ) < v p denotes the p -adic valuationon Q . It is sufficient to find an index j/i such that v p ( n j ) ≥
0. Let L = ( L i ⊗ k i k ) red . Remarkthat L = lim −→ i ∈ I op L i . It is a field extension of finite type of k . Consider elements a , ..., a n algebraically independant over k such that L is a finite extension of k ( a , ..., a n ). By enlarging i ,we can assume that a , ..., a n belongs to L i . Thus L i is a finite extension of k i ( a , ..., a n ): replacing k i by k i ( a , ..., a n ), we can assume that L i /k i is finite.Let L ′ be the subextension of L over k generated by the p -th roots of elements of k . As L/k isfinite, L ′ /k is finite, generated by elements b , ..., b r ∈ L . consider an index j/i such that b , ..., b r belongs to L j . It follows that v p (lg( L i ⊗ k i k j )) = v p (lg( L i ⊗ k i k )). Thus v p ( n j ) = v p ( n ) ≥ (cid:3) Corollary . Let S be a scheme and α be a pre-special S -cycle.Let ¯ s be a geometric point of S , with image s in S , and S ′ be the strict localization of S at ¯ s .Then the following conditions are equivalent:(i) α/S is special at s .(i’) α/S is special at ¯ s . (ii) (cid:0) α | S ′ (cid:1) /S ′ is special at ¯ s (notation of 8.3.1).(iii) There exists an ´etale neighbourhood V of ¯ s in S such that ( α ⊗ S V ) /V is special at ¯ s . Proof.
The equivalence of (i) and (i’) follows trivially from definition ( cf. α | S ′ = α ⊗ S S ′ . Thus ( i ′ ) ⇒ ( ii ) is easy (see 8.1.44). Moreover, ( ii ) ⇒ ( iii ) isa consequence of the previous proposition applied to the pro-scheme of ´etale neighbourhood of ¯ s .Finally, ( iii ) ⇒ ( i ) follows from Lemma 8.1.44. (cid:3) Proposition . Consider the notations and hypothesis of 8.3.5. Assume that S and S i are reduced for any i ∈ I .Suppose given a projective system ( X i ) i ∈ I of S i -schemes of finite type such that for any j/i , X j = X i × S i S j . We let X be the projective limit of ( X i ) .Then for any pre-special (resp. special, Λ -universal) S -cycle α ⊂ X , there exists i ∈ I and apre-special (resp. special, Λ -universal) S i -cycle α i ⊂ X i such that α = α i ⊗ S i S . Proof.
Using Proposition 8.3.6, we are reduced to consider the first of the respective casesof the proposition. Write α = P r ∈ Θ n r . h Z r i X in standard form.Consider r ∈ Θ. As X is noetherian, there exists an index i ∈ I and a closed subscheme Z r,i ⊂ X i such that Z r = Z r,i × S i S . Moreover, replacing Z r,i by the reduced closure of the image of thecanonical map Z r ( ∗ ) −−→ Z r,i , we can assume that the map ( ∗ ) is dominant. For any j ∈ I/i , weput Z r,j = Z r,i × S i S j . The limit of the pro-scheme ( Z r,j ) j ∈ I/i op is the integral scheme Z r . Thus,applying [ EGA4 , 8.2.2], we see that by enlarging i , we can assume that for any j ∈ I/i , Z r,j isirreducible (but not necessarily reduced).We repeat this construction for every r ∈ Θ, enlarging i at each step. Fix now an element j ∈ I/i .The scheme Z r,j may not be reduced. However, its reduction Z ′ r,j is an integral scheme such that Z ′ r,j × S j S = Z r . We put α j = X r ∈ Θ n r h Z ′ r,j i X j . Let z r,j be the generic point of Z ′ r,j , and s r,j be its image in S j . It is a generic point and correspondsuniquely to a generic point s r of S according to the point (3) of the hypothesis 8.3.5. Thus α j /S j is pre-special. Moreover, we get from the above that κ ( z r,j ) ⊗ κ ( s r,j ) κ ( s r ) = κ ( z r ) where z r is thegeneric point of Z r . Thus the relation α j ⊗ S j S = α follows from lemma 8.2.1. (cid:3) Samuel multiplicities. . We give some recall on Samuel multiplicities, following as a general reference [
Bou93 ,VIII. § A be a noetherian local ring with maximal ideal m . Let M = 0 be a A -module of finite typeand q ⊂ m an ideal of A such that M/ q M has finite length. Let d be the dimension of the supportof M . Recall from loc. cit. that Samuel multiplicity of M at q is defined as the integer: e A q ( M ) := lim n →∞ (cid:18) d ! n d lg A ( M/ q n M ) (cid:19) In the case M = A , we simply put e q ( A ) := e A q ( A ) and e ( A ) := e A m ( A ).We will use the following properties of these multiplicities that we recall for the convenienceof the reader; let A be a local noetherian ring with maximal ideal m :Let Φ be the generic points p of Spec ( A ) such that dim( A/ p A ) = dim A . Then according toproposition 3 of loc. cit. :( S e q ( A ) = X p ∈ Φ lg( A p ) .e q ( A/ p ) . Let B be a local flat A -algebra such that B/ m B has finite length over B . Then according toproposition 4 of loc. cit. :( S e m B ( B ) e ( A ) = lg B ( B/ m B ) . This pullback is defined in any case because of point (3) of the hypothesis above. . RELATIVE CYCLES 189
Let B be a local flat A -algebra such that m B is the maximal ideal of B . Let q ⊂ A be an idealsuch that A/ q A has finite length. Then according to the corollary of proposition 4 in loc. cit. :( S e q B ( B ) = e q ( A ) . Assume A is integral with fraction field K . Let B be a finite local A -algebra such that B ⊃ A .Let k B /k A be the extension of the residue fields of B/A . Then, according to proposition 5 andpoint b) of the corollary of proposition 4 in loc. cit. ,( S e m B ( B ) e ( A ) = dim K ( B ⊗ A K )[ k B : k A ] . Definition . (i) Let S = Spec ( A ) be a local scheme, s = m the closed point of S .Let Z be an S -scheme of finite type with special fiber Z s . For any generic point z of Z s ,denoting by B the local ring of Z at z , we define the Samuel multiplicity of Z at z over S as therational integer: m S ( z, Z/S ) = e m B ( B ) e ( A ) . In the case where Z is integral, we define the Samuel specialization of the S -cycle h Z i at s asthe cycle with rational coefficients and domain Z s : h Z i ⊗ S S s = X z ∈ Z (0) s m S ( z, Z/S ) .z. Consider an S -cycle of finite type α = P i ∈ I n i . h Z i i X written in standard form. We define the Samuel specialization of the S -cycle α at s as the cycle with domain X s : α ⊗ S S s = X i ∈ I n i . h Z i i ⊗ S S s. (ii) Let S be a scheme. For any point s of S , we let S ( s ) be the localized scheme of S at s .Let f : Z → S be an S -scheme of finite type, and z a point of Z which is generic in its fiber.Put s = f ( z ). We define the Samuel multiplicity of
Z/S at z as the integer m S ( z, Z/S ) := m S ( z, Z × S S ( s ) /S ( s ) ) . Consider an S -cycle of finite type α with domain X and a point s of S . We define the Samuelspecialization of the S -cycle α at s as the cycle with rational coefficients: α ⊗ S S s = (cid:0) α | S ( s ) (cid:1) ⊗ S S ( s ) s. Lemma . Let S be a scheme, and p : Z ′ → Z an S -morphism which is a birationaluniversal homeomorphism. Then for any point s ∈ S , h Z ′ i ⊗ S S s = h Z i ⊗ S S s in ( Z ′ s ) red = ( Z s ) red . Proof.
By hypothesis, p induces an isomorphism Z ′ (0) ≃ Z (0) between the generic points.Given any irreducible component T ′ of Z ′ corresponding to the irreducible component T of Z , weget by hypothesis: T ′ red ≃ T red (as schemes), lg (cid:0) O Z ′ ,T ′ (cid:1) = lg (cid:0) O Z,T (cid:1) . Thus, we easily concludes from the definition. (cid:3) . Let Z f −→ S be a morphism of finite type and a z a point of Z , s = f ( z ). Assume z isa generic point of Z s . We introduce the following condition: D ( z, Z/S ) : (cid:26) For any irreducible component T of Z ( z ) , T s = ∅ or dim( T ) = dim( Z ( z ) ) . Remark . This condition is in particular satisfied if Z ( z ) is absolutely equidimensional(and a fortiori if Z is absolutely equidimensional).An immediate translation of ( S
1) gives: Lemma . Let S be a local scheme with closed point s and Z be an S -scheme of finitetype such that Z s is irreducible with generic point z .If the condition D ( z, Z/S ) is satisfied, then h Z i ⊗ S S s = m S ( z, Z/S ) .z . We get directly from ( S
2) the following lemma:
Lemma . Let S be a scheme, s be a point of S , and α = P i ∈ I n i . h Z i i X be an S -cyclein standard form such that Z i is a flat S -scheme of finite type.Then α is a Hilbert S -cycle and α ⊗ S S s = α ⊗ ♭S s . With the notations of 8.3.1, we get from ( S Lemma . Let S be a scheme, s a point of S with residue field k and α an S -cycle offinite type.(i) Let S ′ be the Hensel localization of S at s . Then, α ⊗ S S s = (cid:0) α | S ′ (cid:1) ⊗ S S ′ s .(ii) Let ¯ k a separable closure corresponding and ¯ s the corresponding geometric point of S . Let S (¯ s ) be the strict localization of S at ¯ s . Then, (cid:0) α ⊗ S S s (cid:1) ⊗ ♭k ¯ k = (cid:0) α | S (¯ s ) (cid:1) ⊗ S S (¯ s ) ¯ s. Let us recall from [
EGA4 , 13.3.2] the following definition:
Definition . Let f : X → S be a morphism of finite type between noetherian schemes,and x a point of X .We say f is equidimensional at x if there exists an open neighbourhood U of x in X and aquasi-finite pseudo-dominant S -morphism U → A dS for d ∈ N . The integer d is independant ofthe choice of U : it is called the relative dimension of f at x .We say f is equidimensional if it is equidimensional at every point of X . Remark . A quasi-finite morphism is equidimensional if and only if it is pseudo-dominant. According to [
EGA4 , 12.1.1.5], this definition agrees with the convention stated inparagraph 8.1.9 in the case of flat morphisms.Note that a direct translation of ( S
4) gives:
Lemma . Let S = Spec ( A ) be an integral local scheme with closed point s and fractionfield K . Let Z be a finite equidimensional S -scheme and z a generic point of Z s . Let B be thelocal ring of Z at z .Then, m S ( z, Z/S ) = dim K ( B ⊗ A K )[ κ ( x ) : κ ( s )] . . Recall that a scheme S is said to be unibranch ( resp. geometrically unibranch) at apoint s ∈ S if the henselisation (resp. strict henselisation) of the local ring O S,s is irreducible (see[
EGA4 , 6.15.1, 18.8.16]). The scheme S is said to be unibranch (resp. geometrically unibranch )if it is so at any point s ∈ S .The following result is the key point of this subsection. Proposition . Consider a cartesian square Z ′ g ′ / / f ′ (cid:15) (cid:15) Z f (cid:15) (cid:15) S ′ g / / S and a point s ′ of S ′ , s = g ( s ′ ) . Let k (resp. k ′ ) be the residue field of s (resp. s ′ ). We assumethe following conditions:(1) S (resp. S ′ ) is geometrically unibranch at s (resp. s ′ ).(2) f and f ′ are equidimensional of dimension n .(3) For any generic point z of Z s (resp. z ′ of Z s ′ ) the condition D ( z, Z/S ) (resp. D ( z ′ , Z ′ /S ′ ) )is satisfied. . RELATIVE CYCLES 191 Then, the following equality holds in Z s ′ : h Z ′ i ⊗ S S ′ s ′ = ( h Z i ⊗ S S s ) ⊗ ♭k k ′ . Proof.
According to Lemma 8.3.15, we have to prove the equality:(8.3.22.1) X z ′ ∈ Z (0) s ′ m S ( z ′ , Z ′ /S ′ ) .z ′ = X z ∈ Z (0) s m S ( z, Z/S ) . h Spec ( κ ( z ) ⊗ k k ′ ) i Z s ′ . As f is equidimensional of dimension n , we can assume according to 8.3.18 that there existsa quasi-finite pseudo-dominant S -morphism p : Z → A nS . For any generic point z of Z s , t = p ( z )is the generic point of A ns . Thus applying ( S m S ( z, Z/S ) = m S ( z, Z/ A nS ) . Consider the S ′ morphism p ′ : Z ′ → A nZ ′ obtained by base change. It is quasi-finite. As Z ′ /S ′ is equidimensional of dimension n , p ′ must be pseudo-dominant. For any generic point z ′ of Z s ′ , t ′ = p ′ ( z ′ ) is the generic point of A ns ′ and as in the preceding paragraph, we get m S ( z ′ , Z ′ /S ′ ) = m S ( z ′ , Z ′ / A nS ′ ) . Moreover, the residue field κ t of t (resp. κ t ′ of t ′ ) is k ( t , ..., t n ) (resp. k ′ ( t , ..., t n )) andthis implies Spec ( κ ( z ) ⊗ κ t κ t ′ ) is homeomorphic to Spec ( κ ( z ) ⊗ k k ′ ) and has the same geometricmultiplicities. Putting this and the two preceding relations in (8.3.22.1), we get reduced to thecase n = 0 – indeed, according to [ EGA4 , 14.4.1.1], A nS (resp. A nS ′ ) is geometrically unibranchat t (resp. t ′ ).Assume now n = 0, so that f and f ′ are quasi-finite pseudo-dominant.Let ¯ k be a separable closure of k and ¯ k ′ a separable closure of a composite of ¯ k and k ′ . Itis sufficient to prove relation (8.3.22.1) after extension to ¯ k ′ (Lemma 8.1.19). Thus according to8.3.17 and hypothesis (3), we can assume S and S ′ are integral strictly local schemes.For any z ∈ Z (0) s , the extension κ ( z ) /k is totally inseparable. Moreover, z corresponds to aunique point z ′ ∈ Z (0) s ′ and we have to prove for any z ∈ Z (0) s : m S ( z ′ , Z ′ /S ′ ) = m S ( z, Z/S ) . lg( κ ( z ) ⊗ k k ′ ) . Let S = Spec ( A ), K = Frac( A ) and B = O Z,z (resp. S ′ = Spec ( A ′ ), K ′ = Frac( A ′ ) and B ′ = O Z ′ ,z ′ ). As B is quasi-finite dominant over A and A is henselian, B/A is necessarily finitedominant. The same is true for B ′ /A ′ and ( S
4) gives the formulas: m S ( z, Z/S ) = dim K ( B ⊗ A K )[ κ ( z ) : k ] , m S ( z ′ , Z ′ /S ′ ) = dim K ′ ( B ′ ⊗ A ′ K ′ )[ κ ( z ′ ) : k ′ ] . As B ′ ⊗ A ′ K ′ = ( B ⊗ A K ) ⊗ K K ′ , the numerator of these two rationals are the same. To conclude,we are reduced to the easy relation[ κ ( z ′ ) : k ′ ] . lg( κ ( z ) ⊗ k k ′ ) = [ κ ( z ) : k ] . (cid:3) Definition . Let S be a scheme and α = P i ∈ I n i . h Z i i X be an S -cycle in standardform.We say α/S is pseudo-equidimensional over s if it is pre-special and for any i ∈ I , the structualmap Z i → S is equidimensional at the generic points of the fiber Z i,s . Proposition . Let S be a strictly local integral scheme with closed point s and residuefield k and α be an S -cycle pseudo-equidimensional at s .Then for any extension Spec ( k ′ ) s ′ −→ S of s and any fat point ( R, k ′ ) of S over s ′ , the followingrelation holds: α R,k ′ = (cid:0) α ⊗ S S s (cid:1) ⊗ ♭k k ′ . Proof.
We put S ′ = Spec ( R ) and denote by s ′ its closed point. Reductions .– By additivity, we reduce to the case α = h Z i , Z is integral and the structuralmorphism f : Z → S is equidimensional at the generic points of Z s . Any generic points of S ′ s ′ dominantes a generic point of Z s so that we can argue locally at each generic point x of Z s . Thuswe can assume Z s is irreducible with generic point x . Moreover, as Z is equidimensional at x , wecan assume according to 8.3.18 there exists a quasi-finite pseudo-dominant S -morphism(8.3.24.1) Z p −→ A nS . Note that S is geometrically unibranch at s . Thus, applying [ EGA4 , 14.4.1] (”crit`ere deChevalley”), f is universally open at x . As S ′ is a trait whose close point goes to s in S , it followsfrom [ EGA4 , 14.3.7] that the base change f ′ : Z ′ → S ′ of f along S ′ /S is pseudo-dominant.Let T be an irreducible component of Z ′ , with special fiber T s ′ and generic fiber T K ′ over S ′ .Then T → S ′ is a dominant morphism of finite type. Thus, according to [ EGA4 , 14.3.10], either T s ′ = ∅ or dim( T s ′ ) = dim( T K ′ ). Moreover, the dimension of T η is equal to the transcendantaldegree of the function field of T over K ′ , which is equal to the transcendental degree of Z over K . This is n according to (8.3.24.1). Thus, in any case, T is equidimensional of dimension n over S ′ and this implies Z ′ is equidimensional of dimension n over S ′ . Moreover, either T s ′ = ∅ ordim( T ) = n + 1 = dim( Z ′ ). Note this implies that for any generic point z ′ of Z s ′ , the condition D ( z ′ , Z ′ /S ′ ) is satisfied. Middle step .– We prove: α R,k = h Z ′ i ⊗ S S ′ s ′ .According to Lemma 8.3.16, α R,k = h Z ′ K i ⊗ ♭R k ′ = h Z ′ K i ⊗ S S ′ s ′ . But the canonical map Z ′ K → Z ′ is a birational universal homeomorphism so that we concludethis step by Lemma 8.3.12. Final step .– We have only to point out that the conditions of Proposition 8.3.22 are fulfilled forthe obvious square; this is precisely what we need. (cid:3)
Corollary . Let S be a reduced scheme, s a point of S and α an S -cycle which ispseudo-equidimensional over s .Let ¯ s be a geometric point of S with image s in S and S ′ be the strict localization of S at ¯ s .We let S ′ = ∪ i ∈ I S ′ i be the irreducible components of S ′ and α i be the cycle made by the part ofthe cycle α ⊗ ♭S S ′ whose points dominate S ′ i .Then the following conditions are equivalent:(i) α/S is special at s .(ii) the cycle α λ ⊗ S S ′ i ¯ s does not depend on i ∈ I .Moreover, when these conditions are fulfilled, α ⊗ S ¯ s = α λ ⊗ S S ′ i ¯ s . Proof.
According to Corollary 8.3.8, we reduce to the case S = S ′ . Then this follows directlyfrom the preceding proposition. (cid:3) Corollary . Let S be a reduced scheme, geometrically unibranch at a point s ∈ S , and α an S -cycle. The following conditions are equivalent:(i) α/S is pseudo-equidimensional over s .(ii) α/S is special at s .Under these conditions, α ⊗ S s = α ⊗ S S s . Remark . In particular, over a reduced geometrically unibranch scheme S , every cyclewhose support is equidimensional over S is special. Corollary . Let S be a reduced scheme and s ∈ S a point such that S is geometricallyunibranch at s and e ( O S,s ) = 1 . Then for any S -cycle α , the following conditions are equivalent:(i) α/S is pseudo-equidimensional over s .(ii) α/S is Λ -universal at s . . FINITE CORRESPONDENCES 193 Remark . In particular, over a regular scheme S , every cycle whose support is equidi-mensional over S is Λ-universal. Remark also the following theorem: Theorem . Let S be an excellent scheme, s ∈ S a point. The following conditions areequivalent:(i) S is regular at s .(ii) S is geometrically unibranch at s and e ( O S,s ) = 1 .(iii) S is unibranch at s and e ( O S,s ) = 1 .Bibliographical references for the proof . We can assume S is the spectrum of an excellent local ring A with closed point s . The implication ( i ) ⇒ ( ii ) follows from the fact that a normal local ring isgeometrically unibranch (at its closed point) and from [ Bou93 , AC.VIII. §
7, prop. 2]. ( ii ) ⇒ ( iii )is trivial. Concerning the implication ( iii ) ⇒ ( i ), let ˆ A be the completion of the local ring A . Weknow from [ Bou93 , AC.VIII.108, ex. 24] that when e ( A ) = 1 and ˆ A is integral, A is regular. Note e ( A ) = 1 implies A is reduced. To conclude, we refer to [ EGA4 , 7.8.3, (vii)] which establishedthat if A is local excellent reduced, ˆ A is integral if and only if A is unibranch.Finally, we get the following theorem already proved by Suslin and Voevodsky ([ SV00b ,3.5.9]):
Theorem . Let S be a scheme and s a point with residue field κ s such that the localring A of S at s is regular. Then for any equidimensional S -scheme Z and any generic point z of Z s , m SV ( z, h Z i ⊗ S s ) = X i ( − i lg A Tor Ai ( O Z,z , κ s ) . Proof.
We reduce to the case S = Spec ( A ). Then Z is absolutely equidimensional and wecan apply Lemma 8.3.15 together with Corollary 8.3.26 to get that m SV ( z, h Z i⊗ S s ) = m S ( z, Z/S ).Then the result follows from a theorem of Serre [ Ser75 , IV.12, th. 1]. (cid:3)
Remark . Let S be a regular scheme, X a smooth S -scheme and α ⊂ X an S -cyclewhose support is equidimensional over S . Let s be a point of S and i : X s → X the closedimmersion of the fiber of X at s . Then the cycle i ∗ ( α ) of [ Ser75 , V-28, par. 7] is well defined andwe get: α ⊗ S s = i ∗ ( α ) .
9. Finite correspondences . In this section, S is the category of all noetherian schemes. We fix an admissible class P of morphisms in S and assume in addition that P is contained in the class of separatedmorphisms of finite type.Consider two S -schemes X and Y . To clarify certain formulas, we will denote X × S Y simplyby XY and let p XXY : XY → X be the canonical projection morphism.We fix a ring of coefficients Λ ⊂ Q . . Let S be a base scheme. For any P -scheme X/S , we let c ( X/S,
Λ) be the Λ-modulemade of the finite and Λ-universal S -cycles with domain X . Consider a morphism f : Y → X of P -schemes over S . Then the pushforward of cycles induces a well defined morphism: f ∗ : c ( Y /S, Λ) → c ( X/S, Λ) . Indeed, consider a cycle α ∈ c ( Y /S ). Let us denote by Z its support in Y and by f ( Z ) ⊂ X image of the latter by f . We consider these subsets as reduced subschemes. Note that f ( Z ) isseparated and of finite type over S because X/S is noetherian, separated, and of finite type, byassumption 9.0. Because
Z/S is proper, [
EGA2 , 5.4.3(ii)] shows that f ( Z ) is indeed proper over S . Thus, the cycle f ∗ ( α ) is Λ-universal according to Corollary 8.2.10. Finally, Z/S is finite, wededuce that f ( Z ) is quasi-finite, thus finite, over S . This implies the result. With the notations of [
SV00b ], c ( X/S, Z ) = c equi ( X/S,
0) when S is reduced. Definition . Let X and Y be two P -schemes over S .A finite S -correspondence from X to Y with coefficients in Λ is an element of c S ( X, Y ) Λ := c ( X × S Y /X ) . We denote such a correspondence by the symbol X • α −→ Y .In the case Λ = Z , we simply put c S ( X, Y ) := c S ( X, Y ) Z . Through the rest of thissection, unless explicitely stated, any cycle and any finite S -correspondence are assumed to havecoefficients in Λ. Remark . (1) According to properties (P7) and (P7’) ( cf. c S ( X, Y ) Λ commutes with finite sums in X and Y .(2) Consider α ∈ c S ( X, Y ) Λ . Let Z be the support of α . Then, Z is finite pseudo-dominantover X (by definition 8.1.20). This means that Z is finite equidimensional over X .When X is regular (resp. X is reduced geometrically unibranch and char( X ) ⊂ Λ × ),a cycle α ⊂ X × S Y written in standard form: α = X i n i h Z i i X × S Y defines a finite S -correspondence from X to Y if and only if for any index i ∈ I , thescheme Z i is finite equidimensional over X ( i.e. finite and dominant over an irreduciblecomponent of X ) – cf. c S ( X, Y ) Λ is the free Λ-module generated by theclosed integral subschemes Z of X × S Y which are finite equidimensional over X .(3) Recall that in general, there is only an inclusion c S ( X, Y ) ⊗ Z Λ ⊂ c S ( X, Y ) Λ . This inclusion is an equality if S is regular ( cf. S ) ⊂ Λ × . Given more generally inclusions of rings Λ ⊂ Λ ′ ⊂ Q , we get an inclusion of groups(9.1.3.1) c S ( X, Y ) Λ ⊗ Λ Λ ′ ⊂ c S ( X, Y ) Λ ′ which, for the same reasons, is an equality when S is regular or char( S ) ⊂ Λ × . Example . (1) Let f : X → Y be a morphism in P /S .Because X/S is separated (assumption 9.0), the graph Γ f of f is a closed subschemeof X × S Y . The canonical projection Γ f → X is an isomorphism. Thus h Γ f i XY is aHilbert cycle over X . In particular, it is Λ-universal and also finite over X , thus it definesa finite S -correspondence from X to Y .(2) Let f : Y → X be a finite S -morphism which is Λ-universal (as a morphism of theassociated cycles). Then the graph Γ f of f is closed in X × S Y and the projectionΓ f → X is isomorphic to f . Thus the cycle h Γ f i XY is a finite Λ-universal cycle over X which therefore define a finite S -correspondence t f : X •−→ Y called the transpose ofthe finite Λ-universal morphism f .Suppose we are given finite S -correspondences X • α −→ Y • β −→ Z . Consider the following diagramof cycles : β ⊗ Y α / / (cid:15) (cid:15) β / / (cid:15) (cid:15) Z.α / / (cid:15) (cid:15) YX (9.1.4.1)The pullback cycle is well defined and has coefficients in Λ as β is Λ-universal over Y . Moreover,according to the definition of pullback ( cf. β ⊗ Y α is a finite Λ-universalcycle over X with domain XY Z . Note finally that according to 9.1.1, the pushforward of thislatter cycle by p XZXY Z is an element of c S ( X, Z ) Λ . Indeed Suslin-Voeodsky’s multiplicities of a cycle over a scheme X can only have denominators whose primefactors divide the residue characteristics of X according to 8.1.38. . FINITE CORRESPONDENCES 195 Definition . Using the preceding notations, we define the composition product of β and α as the finite S -correspondence β ◦ α = p XZXY Z ∗ ( β ⊗ Y α ) : X •−→ Z. Remark . In the case where S is regular and X , Y , Z are smooth over S , the compositionproduct defined above agree with the one defined in [ D´eg07 , 4.1.16] in terms of the Tor-formulaof Serre. In fact, this is a direct consequence of 8.3.31 after reduction to the case where α and β are represented by closed integral subschemes (see also point (2) of remark 9.1.3).We sum up the main properties of the composition for finite correspondences in the followingproposition : Proposition . Let X , Y , Z be P -schemes over S .(1) For any finite S -correspondences X • α −→ Y • β −→ Z • γ −→ T , we have ( γ ◦ β ) ◦ α = γ ◦ ( β ◦ α ) .(2) For any X • α −→ Y g −→ Z , h Γ g i Y Z ◦ α = (1 X × S g ) ∗ ( α ) .(3) For any X f −→ Y • β −→ Z , β ◦ h Γ f i XY = β ⊗ Y h X i .Moreover, if f is flat, β ◦ h Γ f i XY = ( f × S Z ) ∗ ( β ) considering the flat pullback of cyclesin the classical sense.(4) For any X f ←− Y • β −→ Z such that f is finite Λ -universal, β ◦ t f = ( f × S Z ) ∗ ( β ) .(5) For any X • α −→ Y g ←− Z such that g is finite Λ -universal, t g ◦ α = h Z i ⊗ Y α .If we suppose that g is finite flat, then t g ◦ α = (1 X × S g ) ∗ ( α ) . Proof. (1) Using respectively the projection formulas 8.2.10 and 8.2.8, we obtain( γ ◦ β ) ◦ α = p XTXY ZT ∗ (cid:0) ( γ ⊗ Z β ) ⊗ Y α (cid:1) γ ◦ ( β ◦ α ) = p XTXY ZT ∗ (cid:0) γ ⊗ Z ( β ⊗ Y α ) (cid:1) . Thus this formula is a direct consequence of the associativity 8.2.7.(2) Let ǫ : Γ g → Y and p XZX Γ g : X Γ g → XZ be the canonical projections. As ǫ is anisomorphism, we have tautologically h Y i = ǫ ∗ ( h Γ g i ). We conclude by the following computation :(1 X × S g ) ∗ ( α ) = (1 X × S g ) ∗ ( h Y i ⊗ Y α ) = (1 X × S g ) ∗ ( ǫ ∗ h Γ g i ⊗ Y α ) ( ∗ ) = (1 X × S g ) ∗ (1 X × S ǫ ) ∗ ( h Γ g i ⊗ Y α ) = p XZX Γ g ∗ ( h Γ g i ⊗ Y α ) ( ∗ ) = p XZXY Z ∗ ( h Γ g i Y Z ⊗ Y α )The equalities labeled ( ∗ ) follow from the projection formula of 8.2.10.(3) The first assertion follows from projection formula of 8.2.8 and the fact that Γ f is isomor-phic to X : β ◦ h Γ f i XY = p XZXY Z ∗ ( β ⊗ Y h Γ f i XY ) = β ⊗ Y p XXY ∗ ( h Γ f i XY ) = β ⊗ Y h X i The second assertion follows from Corollary 8.2.2.(4) and (5): The proof of these assertions is strictly similar to that of (2) and (3) instead thatwe use the projection formula of 8.2.8 (and do not need the commutativity 8.2.3). (cid:3)
As a corollary, we obtain that the composition of S -morphisms coincide with the compositionof the associated graph considered as finite S -correspondences. For any S -morphism f : X → Y ,we will still denote by f : X •−→ Y the finite S -correspondence equal to h Γ f i XY . Note moreoverthat for any P -scheme X/S , the identity morphism of X is the neutral element for the compositionof finite S -correspondences. Definition . We let P cor Λ ,S be the category of P -schemes over S with morphisms thefinite S -correspondences and the composition product of definition 9.1.5.An object of P cor Λ ,S will be denoted by [ X ]. The category P cor Λ ,S is additive, and the direct sumis given by the disjoint union of P -schemes over S . We have a canonical faithful functor(9.1.8.1) γ : P /S → P cor Λ ,S which is the identity on objects and the graph on morphisms. We call it the graph functor .9.1.9 . Given extension of rings Λ ⊂ Λ ′ ⊂ Q , we get according to Remark 9.1.3(3) and thedefinition of composition of finite correspondences a functor of Λ ′ -linear categories:(9.1.9.1) P cor Λ ,S ⊗ Λ Λ ′ → P cor Λ ′ ,S which is the identity on objects and the inclusions of the form (9.1.3.1) on morphisms. Proposition . Consider the above notations. If S is regular or char( S ) ⊂ Λ × then thefunctor (9.1.9.1) is an equality of categories. Indeed according to point (3) of Proposition 9.1.3, the inclusions of groups of correspondencesused to define the above functors are all equalities in each respective cases.9.1.11 . Given two S -morphisms f : Y → X and g : X ′ → X such that g is finite Λ-universal,we get from the previous proposition the equality of cycles in Y X ′ : t g ◦ f = h X ′ i ⊗ X h Y i Y X where Y is seen as a closed subscheme of Y X through the graph of f .In particular, when either f or g is flat, we get (use property (P3) of 8.1.34 or Corollary 8.2.2): t g ◦ f = h X ′ × X Y i Y X ′ . To state the next formulas (the generalized degree formulas), we introduce the following notion:
Definition . Let f : X ′ → X be a finite equidimensional morphism.For any generic point x of X , we define the degree of f at x as the integer:deg x ( f ) = X x ′ /x [ κ x ′ : κ x ]where the sum runs over the generic points of X ′ lying above x . Proposition . Let X be a connected S -scheme and f : X ′ → X be a finite S -morphism.If f is special then there exists an integer d ∈ N ∗ such that for any generic point x of X , deg x ( f ) = d .Moreover, f ◦ t f = d. X . We simply call d the degree of the finite special morphism f . Proof.
Let ∆ ′ be the diagonal of X ′ /S . For any generic point x of X , we let ∆ x be thediagonal of the corresponding irreducible component of X , seen as a closed subscheme of X .According to Proposition 9.1.7, and the definition of pushforwards, we get α := f ◦ t f = ( f × S f ) ∗ ( h ∆ ′ i X ′ X ′ ) = X x ∈ X (0) deg x ( f ) . h ∆ x i XX . Considering generic points x , y of X , we prove deg x ( f ) = deg y ( f ). By induction, we can reduceto the case where x and y have a common specialisation s in X because X is connected andnoetherian. Then, as α/X is special, we get by definition of the pullback (see more precisely8.1.43) α ⊗ S s = deg x ( f ) .s = deg y ( f ) .s as required. The remaining assertion then follows. (cid:3) . FINITE CORRESPONDENCES 197 Proposition . Let f : X ′ → X be an S -morphism which is finite, radicial and Λ -universal.Assume X is connected, and let d be the degree of f .Then t f ◦ f = d. X ′ . In particular, if d is invertible in Λ , f is an isomorphism in P cor Λ ,S . Proof.
According to 9.1.11, t f ◦ f = h X ′ i ⊗ X h X ′ i as cycles in X ′ X ′ . Let x be the genericpoint of X and k be its residue field. Let { x ′ i , i ∈ I } be the set of generic points of X , and for any i ∈ I , k ′ i be the residue field of x ′ i . According to 8.2.1, we thus obtain: t f ◦ f = X ( i,j ) ∈ I h Spec (cid:0) k ′ i ⊗ k k ′ j (cid:1) i X ′ X ′ . The result now follows by the definition of the degree and the fact that for any i ∈ I , k ′ i /k isradicial. (cid:3) Fix a base scheme S . Let X , X ′ , Y , Y ’ be P -schemes over S .Consider finite S -correspondences α : X •−→ Y and α ′ : X ′ •−→ Y ′ . Then αX ′ := α ⊗ X h XX ′ i and α ′ X := α ′ ⊗ X ′ h XX ′ i are both finite Λ-universal cycles over XX ′ . Using stability bycomposition of finite Λ-universal morphisms ( cf. Corollary 8.2.6), the cycle ( αX ′ ) ⊗ XX ′ ( α ′ X ) isfinite Λ-universal over XX ′ . Definition . Using the above notation, we define the tensor product of α and α ′ over S as the finite S -correspondence α ⊗ trS α ′ = ( αX ′ ) ⊗ XX ′ ( α ′ X ) : XX ′ •−→ Y Y ′ . Let us first remark that this tensor product is commutative (use commutativity of the pullback8.2.3) and associative (use associativity of the pullback 8.2.7). Moreover, it is compatible withcomposition :
Lemma . Suppose given finite S -correspondences : α : X → Y, β : Y → Z, α ′ : X ′ → Y ′ , β ′ : Y ′ → Z ′ . Then ( β ◦ α ) ⊗ trS ( β ′ ◦ α ′ ) = ( β ⊗ trS β ′ ) ◦ ( α ⊗ trS α ′ ) . Proof.
We put αX ′ = α ⊗ X h XX ′ i , α ′ X = α ′ ⊗ X h XX ′ i and βY ′ = β ⊗ Y h Y Y ′ i , β ′ Y = β ′ ⊗ Y h Y Y ′ i . We can compute the right hand side of the above equation as follows : p XX ′ ZZ ′ XX ′ Y Y ′ ZZ ′ ∗ (cid:16) ( βY ′ ⊗ Y Y ′ β ′ Y ) ⊗ Y Y ′ ( αX ′ ⊗ XX ′ α ′ X ) (cid:17) (1) = p XX ′ ZZ ′ XX ′ Y Y ′ ZZ ′ ∗ (cid:16) ( βY ′ ⊗ Y Y ′ β ′ Y ) ⊗ Y Y ′ ( α ′ X ⊗ XX ′ αX ′ ) (cid:17) (2) = p XX ′ ZZ ′ XX ′ Y Y ′ ZZ ′ ∗ (cid:16) βY ′ ⊗ Y Y ′ (( β ′ Y ⊗ Y Y ′ α ′ X ) ⊗ XX ′ αX ′ ) (cid:17) (3) = p XX ′ ZZ ′ XX ′ Y Y ′ ZZ ′ ∗ (cid:16) ( βY ′ ⊗ Y Y ′ αX ′ ) ⊗ XX ′ ( β ′ Y ⊗ Y Y ′ α ′ X )) (cid:17) . Equality (1) follows from commutativity 8.2.3, equality (2) from associativity 8.2.7 and equality(3) by both commutativity and associativity.For the left hand side, we note that using the projection formula 8.2.10, the left hand side isequal to p XX ′ ZZ ′ XX ′ Y Y ′ ZZ ′ ∗ (cid:16)(cid:0) ( β ⊗ Y α ) ⊗ X h XX ′ i (cid:1) ⊗ XX ′ (cid:0) ( β ′ ⊗ Y ′ α ′ ) ⊗ X ′ h XX ′ i (cid:1)(cid:17) . We are left to remark that( β ⊗ Y α ) ⊗ X h XX ′ i = (cid:0) ( βY ′ ) ⊗ Y Y ′ α (cid:1) ⊗ X h XX ′ i = βY ′ ⊗ Y Y ′ αX ′ , using transitivity 8.2.4 and associativity 8.2.7. We thus conclude by symmetry of the other partin the left hand side. (cid:3) Definition . We define a symmetric monoidal structure on the category P cor Λ ,S by putting[ X ] ⊗ trS [ Y ] = [ X × S Y ] on objects and using the tensor product of the previous definition formorphisms.9.2.4 . Note that the functor γ : P /S → P cor Λ ,S is monoidal for the cartesian structure on thesource category. Indeed, this is a consequence of property (P3) of the relative product (see 8.1.34)and the remark that for any morphisms f : X → Y and f ′ : X ′ → Y ′ , (Γ f × S X ′ ) × XX ′ (Γ ′ f × S X ) =Γ f × S f ′ . Fix a morphism of schemes f : T → S . For any P -scheme X/S , we put X T = X × S T . For a pair of P -schemes over S (resp. T -schemes) ( X, Y ), we put XY = X × S Y (resp. XY T = X × T Y ).9.3.a. Base change.
Consider a finite S -correspondence α : X •−→ Y . The cycle α ⊗ X h X T i defines a finite T -correspondence from X T to Y T denoted by α T . Lemma . Consider finite S -correspondences X • α −→ Y • β −→ Y .Then ( β ◦ α ) T = β T ◦ α T . Proof.
This follows easily using the projection formula 8.2.10, the associativity formula 8.2.7and the transitivity formula 8.2.4 : p XZXY Z ∗ ( β ⊗ Y α ) ⊗ X h X T i = p XZ T XY Z T ∗ (cid:0) ( β ⊗ Y α ) ⊗ X h X T i (cid:1) = p XZ T XY Z T ∗ (cid:0) β ⊗ Y ( α ⊗ X h X T i ) (cid:1) = p XZ T XY Z T ∗ (cid:0) ( β ⊗ Y h Y T i ) ⊗ Y T ( α ⊗ X h X T i ) (cid:1) . (cid:3) Definition . Let f : T → S be a morphism of schemes. Using the preceding lemma, wedefine the base change functor f ∗ : P cor Λ ,S → P cor Λ ,T [ X/S ] [ X T /T ] c S ( X, Y ) Λ ∋ α α T . We sum up the basic properties of the base change for correspondences in the following lemma.
Lemma . Take the notation and hypothesis of the previous definition.(1) The functor f ∗ is symmetric monoidal.(2) Let f ∗ : P /S → P /T be the classical base change functor on P -schemes over S . Thenthe following diagram is commutative: P /S γ S / / f ∗ (cid:15) (cid:15) P cor Λ ,Sf ∗ (cid:15) (cid:15) P /T γ T / / P cor Λ ,T . (3) Let σ : T ′ → T be a morphism of schemes. Through the canonical isomorphisms ( X T ) T ′ ≃ X T ′ , equality ( f ◦ σ ) ∗ = σ ∗ ◦ f ∗ holds. Proof. (1) This point follows easily using the associativity formula 8.2.7 and the transitivityformulas 8.2.4, 8.2.6.(2) This point follows from the fact that for any S -morphism f : X → Y , there is a canonicalisomorphism Γ f T → Γ f × S T .(3) This point is a direct application of the transitivity 8.2.4. (cid:3) Lemma . Let f : T → S be a universal homeomorphism.Then f ∗ : P cor Λ ,S → P cor Λ ,T is fully faithful. Proof.
Let X and Y be P -schemes over S . Then X T → X is a universal homeomorphism.Any generic point x of X corresponds uniquely to a generic point of X T . Let m x (resp. m ′ x ) be the . FINITE CORRESPONDENCES 199 geometric multiplicity of x in X (resp. X T ). Consider a finite S -correspondence α = P i ∈ I n i .z i .For each i ∈ I , let x i be the generic point of X dominated by z i . Then we get by definition: f ∗ ( α ) = X i ∈ I m ′ x i n i m x i .z i and the lemma is clear. (cid:3) Restriction.
Consider a P -morphism p : T → S . For any pair of T -schemes ( X, Y ), wedenote by δ XY : X × T Y → X × S Y the canonical closed immersion deduced by base change fromthe diagonal immersion of T /S .Consider a finite T -correspondence α : X •−→ Y . The cycle δ XY ∗ ( α ) is the cycle α consideredas a cycle in X × S Y . It defines a finite S -correspondence from X to Y . Lemma . Let X , Y and Z be T -schemes. The following relations are true :(1) For any T -morphism f : X → Y , δ XY ∗ (cid:0) h Γ f i XY T (cid:1) = h Γ f i XY .(2) For all α ∈ c T ( X, Y ) Λ and β ∈ c T ( Y, Z ) Λ , δ XZ ∗ ( β ◦ α ) = ( δ Y Z ∗ ( β )) ◦ ( δ XY ∗ ( α )) . Proof.
The first assertion is obvious.The second assertion is a consequence of the projection formulas 8.2.8 and 8.2.10, and thefunctoriality of pushforwards :( δ Y Z ∗ ( β )) ◦ ( δ XY ∗ ( α )) = p XZXY Z ∗ (cid:0) δ Y Z ∗ ( β ) ⊗ Y δ XY ∗ ( α ) (cid:1) = p XZXY Z ∗ δ XY Z ∗ ( β ⊗ Y α ) = δ XZ ∗ p XZ T XY Z T ∗ ( β ⊗ Y α ) . (cid:3) Definition . Let p : T → S be a P -morphism.Using the preceding lemma, we define a functor p ♯ : P cor Λ ,T → P cor Λ ,S [ X → T ] [ X → T p −→ S ] c T ( X, Y ) Λ ∋ α δ XY ∗ ( α ) . This functor enjoys the following properties:
Lemma . Let p : T → S be a P -morphism.(1) The functor p ♯ is left adjoint to the functor p ∗ .(2) For any composable P -morphisms Z q −→ T p −→ S , ( pq ) ♯ = p ♯ q ♯ .(3) Let p ♯ : P /T → P /S be the functor induced by composition with p . Then the followingdiagram is commutative: P /T γ T / / p ♯ (cid:15) (cid:15) P cor Λ ,Tp ♯ (cid:15) (cid:15) P /S γ S / / P cor Λ ,S . Proof.
For point (1), we have to construct for schemes
X/T and
Y /S a natural isomorphism c S ( p ♯ X, Y ) Λ ≃ c T ( X, p ∗ Y ) Λ . It is induced by the canonical isomorphism of schemes ( p ♯ X ) × S Y ≃ X × T ( p ∗ Y ).Point (2) follows from the associativity of the pushforward functor on cycles. Note also that thisidentification is compatible with the transposition of the identification of 9.3.3(3) according to theadjunction property just obtained.Point (3) is a reformulation of 9.3.5(2). (cid:3) A finiteness property. . We assume here that P is the class of all separated morphisms of finite type in S .Let I be a left filtering category and ( X i ) i ∈ I be a projective system of separated S -schemes offinite type with affine dominant transition morphisms. We let X be the projective limit of ( X i ) i and assume that X is Noetherian over S . Proposition . Let Y be a P -scheme of finite type over S . Then the canonical morphism ϕ : lim −→ i ∈ I op c S ( X i , Y ) Λ → c ( X × S Y / X , Λ) . is an isomorphism. Proof.
Note that according to [
SGA4 , IV, 8.3.8(i)], we can assume the conditions (2) of 8.3.5is verified for ( X i ) i ∈ I . Thus conditions (1) to (4) of loc. cit. are verified. Then the surjectivity of ϕ follows from 8.3.9 and the injectivity from 8.3.6. (cid:3) We can summarize the preceding con-structions:
Proposition . The -functor P cor Λ : S P cor Λ ,S equipped with the pullback defined in 9.3.2 and with the tensor product of 9.2.3 is a monoidal P -fibred category such that the functor γ : P → P cor Λ (see (9.1.8.1) ) is a morphism of monoidal P -fibred category. Proof.
According to Lemma 9.3.7, for any P morphisms p , p ∗ admits a left adjoint p ♯ .We have checked that γ is symmetric monoidal and commutes with f ∗ and p ♯ (see respectively9.2.4, 9.3.3 and 9.3.7). But γ is essentially surjective. Thus, to prove the properties ( P -BC) and( P -PF) for the fibred category P cor Λ , we are reduced to the case of P which is easy (see example1.1.28). This concludes. (cid:3) Remark . Consider the definition above.(1) The category P cor Λ is Λ-linear. For any scheme S , P cor Λ ,S is additive. For any finite familyof schemes ( S i ) i ∈ I which admits a sum S in S , the canonical map P cor Λ ,S → M i ∈ I P cor Λ ,S i is an isomorphism.(2) The functor γ : P → P cor Λ is nothing else than the canonical geometric sections of P cor Λ (see definition 1.1.35).We will apply these definitions in the particular cases P = Sm (resp. P = S ft ) the classof smooth separated (resp. separated) morphisms of finite type. Note that we get a commutativesquare Sm γ / / (cid:15) (cid:15) S m cor Λ (cid:15) (cid:15) S ft γ / / S ft,cor Λ where the vertical maps are the obvious embeddings of monoidal Sm -fibred categories.9.4.3 . Consider extensions of rings Λ ⊂ Λ ′ ⊂ Q . The functors (9.1.9.1) for various schemes S in S are compatible with the operations of a P -fibred category because it is just concernedwith adding denominators in the coefficients of the finite correspondences considered. Thus theyinduce a morphism of monoidal P -fibred categories over S :(9.4.3.1) P cor Λ ⊗ Λ Λ ′ → P cor Λ ′ .
0. SHEAVES WITH TRANSFERS 201
According to Proposition 9.1.10, we get the following result:
Proposition . Consider the above notations. Then the above morphism of monoidal P -fibred categories is an equality whenever it is restricted to one of the following subcategories of S : • The category of regular schemes. • The category of noetherian finite dimensional schemes S such that char( S ) ⊂ Λ × . Remark . The restriction of the category P cor Z to the category of regular schemes wasalready defined in [ D´eg07 ]. Indeed, one can check using the comparison of Suslin-Voevodsky’smultiplicites with Serre’s intersection multiplicities (using Tor-formulas ; cf. τ ∗ , τ ♯ , and ⊗ tr defined here coincide with that of [ D´eg07 ]. This remark extends 9.1.6.
10. Sheaves with transfers . The category S is the category of noetherian schemes of finite dimension. We fix anadmissible class P of morphisms in S satisfying the following assumptions:(a) Any morphism in P is separated of finite type.(b) Any ´etale separated morphism of finite type is in P .We fix a topology t on S which is P -admissible and such that:(c) For any scheme S , there is a class of covers E of the form ( p : W → S ) with p a P -morphism such that t is the topology generated by E and the covers of the form( U → U ⊔ V, V → U ⊔ V ) for any schemes U and V in S .We fix a ring of coefficients Λ. Whenever we speak of Λ-cycles (or the premotivic category P cor Λ ,etc...), we mean cycles with coefficients in the localization of Z with respect to invertible primesin Λ.Note that in sections 10.4 and 10.5, we will apply the conventions of section 1.4 by replacingthe class of smooth morphisms of finite type (resp. morphisms of finite type) there by the class ofsmooth separated morphisms of finite type (resp. separated morphisms of finite type). We consider the additive category P cor Λ ,S of definition9.1.8 and the graph functor γ : P /S → P cor Λ ,S of (9.1.8.1). Definition . A presheaf with transfers F over S is an additive presheaf of Λ-modulesover P cor Λ ,S . We denote by PSh (cid:0) P cor Λ ,S (cid:1) the corresponding category.If X is a P -scheme over S , we denote by Λ trS ( X ) the presheaf with transfers represented by X . We denote by ˆ γ ∗ the functor(10.1.1.1) PSh (cid:0) P cor Λ ,S (cid:1) → PSh( P /S, Λ) , F F ◦ γ. Note that PSh (cid:0) P cor Λ ,S (cid:1) is obviously a Grothendieck abelian category generated by the objectsΛ trS ( X ) for a P -scheme X/S . Moreover, the following proposition is straightforward:
Proposition . There is an essentially unique Grothendieck abelian P -premotivic cat-egory PSh ( P cor Λ ) which is geometrically generated (cf. 1.1.41), whose fiber over a scheme S is PSh (cid:0) P cor Λ ,S (cid:1) and such that the functor Λ trS induces a morphism of additive monoidal P -fibredcategories. (10.1.2.1) P cor Λ → PSh ( P cor Λ ) . Moreover, the functor (10.1.1.1) induces a morphism of abelian P -premotivic categories ˆ γ ∗ : PSh( P , Λ) ⇄ PSh ( P cor Λ ) : ˆ γ ∗ . Proof.
To help the reader, we recall the following consequence of Yoneda’s lemma: Lemma . Let F : ( P cor Λ ,S ) op → Λ - mod be a presheaf with transfers. Let I be the categoryof representables preshaves with transfers over F . Then the canonical map lim −→ Λ trS ( X ) → F Λ trS ( X ) → F is an isomorphism. The limit is taken in PSh (cid:0) P cor Λ ,S (cid:1) and runs over I . This lemma allows us to define the structural left adjoint of PSh ( P cor Λ ) (recall f ∗ , p ♯ for p a P -morphism and the tensor product) because they are indeed determined by (10.1.2.1). Theexistence of the structural right adjoints is formal.The same lemma allows to get the adjunction (ˆ γ ∗ , ˆ γ ∗ ). (cid:3) Remark . Note that for any presheaf with transfers F over S , and any morphism f : T → S (resp. P -morphism p : S → S ′ ), we get as usual f ∗ F = F ◦ f ∗ (resp. p ∗ F = F ◦ p ♯ )where the functor f ∗ (resp. p ♯ ) on the right hand side is taken with respect to the P -fibredcategory P cor Λ .Let us state the following lemma for future use. Lemma . Let Let ( S α ) α ∈ A be a projective system of schemes in S , with dominant affinetransition maps, and such that S = lim ←− α ∈ A S α is representable in S .Consider an index α ∈ A and a presheaf with transfers F over S α . For any index α/α ,we denote by F α (resp. F ) the pullback of F α over S α (resp. S ) in the sense of the premotivicstructure on PSh ( P cor Λ ) .Then the canonical map: lim −→ α ∈ A/α F α ( S α ) −→ F ( S ) is an isomorphism. Proof.
The presheaf F α can be written as an inductive limit of representable sheaves ofthe form Λ trS α ( X α ) of a P -scheme X α /S α . As the global section functor on presheaves withtransfers commute with inductive limit, we are reduced to the case where F = Λ trS α ( X α ). In thiscase, the lemma follows directly from Proposition 9.3.9. (cid:3) Definition . A t -sheaf with transfers over S is a presheaf with transfers F such thatthe functor F ◦ γ S is a t -sheaf. We denote by Sh t (cid:0) P cor Λ ,S (cid:1) the full subcategory of PSh( P cor Λ ,S , Λ) ofsheaves with transfers.According to this definition, we get a canonical faithful functor γ ∗ : Sh t (cid:0) P cor Λ ,S (cid:1) → Sh t ( P /S, Λ) , F F ◦ γ. Example . A particularly important case for us is the case when t = Nis is the Nisnevichtopology. According to the original definition of Voevodsky, a Nisnevich sheaf with transfers willbe called simply a sheaf with transfers . Remark . Later on, in the case P = S ft , we will use the notation Λ tr S ( X ) to denotethe presheaf on the big site S ft,cor Λ ,S represented by a separated S -scheme of finite type. Proposition . Let X be an P -scheme over S .(1) The presheaf Λ trS ( X ) is an ´etale sheaf with transfers.(2) If char( X ) ⊂ Λ × , Λ trS ( X ) is a qfh -sheaf with transfers. Proof.
For point (1), we follow the proof of [
D´eg07 , 4.2.4]: the computation of the pullbackby an ´etale map is given in our context by point (3) of Proposition 9.1.7. Moreover, the propertyfor a cycle α/Y to be Λ-universal is ´etale-local on Y according to 8.3.8. For point (2), we refer to[ SV00b , 4.2.7]. (cid:3)
0. SHEAVES WITH TRANSFERS 203
We can actually describe explicitely representable presheaves with transfers in the followingcase:
Proposition . Let S be a scheme and X be a finite ´etale S -scheme. Then for any P -scheme Y over S , Γ( Y, Λ trS ( X )) = π ( Y × S X ) . Λ . This readily follows from the following lemma:
Lemma . Let f : X → S be an ´etale separated morphism of finite type. Let π finite ( X/S ) be the set of connected components V of X such that f ( V ) is equal to a connected component of S ( i.e. f is finite over V ).Then c ( X/S,
Λ) = π finite ( X/S ) . Λ . Proof.
We can assume that S is reduced and connected.We first treat the case where X = S . Consider a finite Λ-universal S -cycle α with domain S .Write α = P i ∈ I n i . h Z i i S in standard form. By definition, Z i dominates an irreducible componentof S thus Z i is equal to that irreducible component.Consider S an irreducible component of S and an index i ∈ I such that S ∩ Z i is not empty.Consider a point s ∈ S ∩ Z i . We have obviously α s = n i . h Spec ( κ ( s )) i 6 = 0. Thus there existsa component of α which dominates S i.e. ∃ j ∈ I such that Z j = S . Moreover, computing α s using alternatively Z i and Z j gives n i = n j .As S is noetherian, we see inductively { Z i | i ∈ I } is the set of irreducible components of S and forany i, j ∈ I , n i = n j . Thus c ( S/S,
Λ) = Z .Consider now the case of an ´etale S -scheme X . By additivity of c , we can assume that X isconnected. Consider the following canonical map: c ( X/S, Λ) → c ( X × S X/X, Λ) , α α ⊗ ♭S X. Note that considering the projection p : X × S X → X , by definition, α ⊗ ♭S X = p ∗ ( α ).Consider the diagonal δ : X → X × S X of X/S . Because
X/S is ´etale and separated, δ is a directfactor of X × S X and we can write X × S X = X ⊔ U . Because c is additive, c ( X × S X/X,
Λ) = c ( X/X, Λ) ⊕ c ( U/X, Λ) . Moreover, the projection on the first factor is induced by the map δ ∗ on cycles. Because δ ∗ p ∗ = 1,we deduce that a cycke in c ( X/S,
Λ) corresponds uniquely to a cycle in c ( X/X,
Λ). According tothe preceding case, this latter group is the free group generated by the cycle h X i . This latter cycleis Λ-universal over S , because X/S is flat. Thus, if
X/S is finite, it is an element of c ( X/S,
Λ) sothat c ( X/S,
Λ) = Λ. Otherwise, not any of the Λ-linear combination of h X i belongs to c ( X/S,
Λ)so that c ( X/S,
Λ) = 0. (cid:3) . Recall from 3.2.1 that we denote by ( P /S ) ∐ the category of I -diagrams of objects in P /S indexed by a discrete category I . Given any simplicial object X of ( P /S ) ∐ , we will considerthe complex Λ trS ( X ) of PSh (cid:0) P cor Λ ,S (cid:1) applying the definition of 5.1.8 to the Grothendieck P -fibredcategory PSh ( P ).Consider a t -cover p : W → X in P /X . We denote by W nX the n -fold product of W over X (in the category P /X ). We denote by ˇ S ( W/X ) the ˇCech simplicial object of P cor Λ ,S suchthat ˇ S n ( W/X ) = W n +1 X . The canonical morphism ˇ S ( W/X ) → X is a t -hypercover according todefinition 3.2.1. We will call these particular type of t -hypercovers the ˇCech t -hypercovers of X . Definition . We will say that the admissible topology t on P is compatible withtransfers (resp. weakly compatible with transfers ) if for any scheme S and any t -hypercover (resp.any ˇCech t -hypercover) X → X in the site P /S , the canonical morphism of complexes(10.3.2.1) Λ trS ( X ) → Λ trS ( X )induces a quasi-isomorphism of the associated t -sheaves on P /S . Obviously, if t is compatible with transfers then it is weakly compatible with transfers.Recall from 10.2.4 that, in the cases t = Nis , ´et, (10.3.2.1) is actually a morphism of complexesof t -sheaves with transfers. The following proposition is a generalisation of [ Voe96 , 3.1.3] but itsproof is in fact the same.
Proposition . The Nisnevich (resp ´etale) topology t on P is weakly compatible withtransfers. Proof.
We consider a t -cover p : W → X , the associated ˇCech hypercover X = ˇ S ( W/X ) of X and we prove that the map (10.3.2.1) is a quasi-isomorphism of t -sheaves. Recall that a point of P /S for the topology t is given by an essentially affine pro-object ( V i ) i ∈ I of P /S . Moreover, itsprojective limit V in the category of schemes is in particular a local henselian noetherian scheme.It will be sufficient to check that the fiber of (10.3.2.1) at the point ( V i ) i ∈ I is a quasi-isomorphism.Thus, according to Proposition 9.3.9, we can assume that S = V is a local henselian scheme andwe are to reduce to prove that the complex of abelian groups . . . → c ( W × X W/S, Λ) → c ( W/S, Λ) p ∗ −→ c ( X/S, Λ) → C .Recall that the abelian group c ( X/S ) is covariantly functorial in X with respect to separatedmorphisms of finite type f : X ′ → X ( cf. paragraph 9.1.1). Moreover, if f is an immersion, f ∗ isobviously injective.Let F be the set of closed subschemes Z of X such that Z/S is finite. Given a closedsubscheme Z in F , we let C Z be the complex of abelian groups(10.3.3.1) . . . → c ( W Z × Z W Z /S, Λ) → c ( W Z /S, Λ) p Z ∗ −−→ c ( Z/S, Λ) → p Z is the t -cover obtained by pullback along Z → X . From what we have just recalled, wecan identify C Z with a subcomplex of C . The set F can be ordered by inclusion, and C is theunion of its subcomplexes C Z . If F is empty, then C = 0 and the proposition is clear. Otherwise, F is filtered and we can write: C = lim −→ Z ∈F C Z . Thus, it will be sufficient to prove that C Z is acyclic for any Z ∈ F . Because S is henselian and Z is finite over S , Z is indeed a finite sum of local henselian schemes. This implies that the t -cover p Z , which is in particular ´etale surjective, admits a splitting s : Z → W Z . Then the complex(10.3.3.1) is contractible with contracting homotopy defined by the family( s × Z W nZ ) ∗ : c ( W nZ /S, Λ) → c ( W n +1 Z /S, Λ) . (cid:3) . Considering an additive abelian presheaf G on P /S , the natural transformation X Hom
PSh( P /S ) (ˆ γ ∗ Λ trS ( X ) , G )defines a presheaf with transfers over S . We will denote by G τ its restriction to the site P /S .Note that this definition can be applied in the case where G is a t -sheaf on P /S , because underthe assumption 10.0 on t , it is in particular an additive presheaf. Definition . We will say that t is mildly compatible with transfers if for any scheme S and any t -sheaf F on P /S , F τ is a t -sheaf on P /S .If t is weakly compatible with transfers then is it mildly compatible with transfers. Remark . Assume t is mildly compatible with transfers. Then for any scheme S , any t -cover p : W → X in P /S induces a morphism p ∗ : Λ trS ( W ) → Λ trS ( X ) Actually, this defines a right adjoint to the functor ˆ γ ∗ .
0. SHEAVES WITH TRANSFERS 205 which is an epimorphism of the associated t -sheaves on P /S . This means that for any correspon-dence α ∈ c S ( Y, X ), there exists a t -cover q : W ′ → Y and a correspondence α ′ ∈ c S ( W ′ , W )making the following diagram commutative: W ′ ˆ α / / • / / q (cid:15) (cid:15) W p (cid:15) (cid:15) Y α / / • / / X (10.3.6.1) Lemma . Assume t is mildly compatible with transfers.Let S be a scheme and P tr be a presheaf with transfers over S . We put P = P tr ◦ γ as apresheaf on P /S . We denote by F the t -sheaf associated with P and by η : P → F the canonicalnatural transformation.Then there exists a unique pair ( F tr , η tr ) such that:(1) F tr is a sheaf with transfers over S such that F tr ◦ γ = F .(2) η tr : P tr → F tr is a natural transformation of presheaves with transfers such that theinduced transformation P = ( P tr ◦ γ ) → ( F tr ◦ γ ) = F coincides with η . Proof.
As a preliminary observation, we note that given a presheaf G on P /S , the data ofa presheaf with transfers G tr such that G tr ◦ γ = G is equivalent to the data for any P -schemes X and Y of a bilinear product(10.3.7.1) G ( X ) ⊗ Z c S ( Y, X ) → G ( Y ) , ρ ⊗ α
7→ h ρ, α i such that:(a) For any morphism f : Y ′ → Y in P /S , f ∗ h ρ, α i = h ρ, α ◦ f i .(b) For any morphism f : X → X ′ in P /S , if ρ = f ∗ ( ρ ′ ), h ρ, α i = h ρ ′ , f ◦ α i .(c) When X = Y , for any ρ ∈ F ( X ), h ρ, X i = ρ .(d) For any finite S -correspondence β ∈ c S ( Z, Y ), hh ρ, α i , β i = h ρ, α ◦ β i .On the other hand, the data of products of the form (10.3.7.1) for any P -schemes X and Y over S which satisfy the conditions (a) and (b) above is equivalent to the data of a natural transformation φ : G → G τ by putting h ρ, α i φ = [ φ X ( ρ )] Y .α .Applying this to the presheaf with transfers P tr , we obtain a canonical natural transformation ψ : P → P τ . By assumption on t , F τ is a t -sheaf. Thus there existe a unique natural transformation ψ suchthat the following diagram commutes: P ψ / / η (cid:15) (cid:15) P τη τ (cid:15) (cid:15) F φ / / F τ Thus we get products of the form 10.3.7.1 associated with φ which satisfies (a) and (b). Thecommutativity of the above diagram asserts they are compatible with the ones correspondingto P tr and the unicity of the natural transformation φ implies the uniqueness statement of thelemma.To finish the proof of the existence, we must show (c) and (d) for the product h ., i φ . Considera couple ( ρ, α ) ∈ F ( X ) × c S ( Y, X ). Because F is the t -sheaf associated with P , there exists a t -cover p : W → X and a section ˆ ρ ∈ P ( W ) such that p ∗ ( ρ ) = η W (ˆ ρ ). Moreover, according toremark 10.3.6, we get a t -cover q : W ′ → Y and a correspondence ˆ α ∈ c S ( W ′ , W ) making thediagram (10.3.6.1) commutative. Then we get using (a) and (b): q ∗ h ρ, α i φ = h ρ, α ◦ q i φ = h ρ, p ◦ ˆ α i φ = h p ∗ ρ, ˆ α i φ = h η W (ˆ ρ ) , ˆ α i φ = h ˆ ρ, ˆ α i ψ . Because q ∗ : F ( X ) → F ( W ) is injective, we deduce easily from this principle the properties (c)and (d) and this concludes. (cid:3) . Let us consider the canonical adjunction a ∗ t : PSh( P /S, Λ) ⇄ Sh t ( P /S, Λ) : O t where O t is the canonical forgetful functor.We also denote by O trt : Sh t (cid:0) P cor Λ ,S (cid:1) → PSh (cid:0) P cor Λ ,S (cid:1) the obvious forgetful functor. Trivially,the following relation holds:(10.3.8.1) ˆ γ ∗ a t, ∗ = a t, ∗ γ ∗ . Proposition . Using the notations above, the following condition on the admissibletopology t are equivalent:(i) t is mildly compatible with transfers.(ii) For any scheme S , the functor O trt admits a left adjoint a ∗ t : PSh (cid:0) P cor Λ ,S (cid:1) → Sh t (cid:0) P cor Λ ,S (cid:1) which is exact and such that the exchange transformation (10.3.9.1) a ∗ t ˆ γ ∗ → γ ∗ a ∗ t induced by the identification (10.3.8.1) is an isomorphism.Under these conditions, the following properties hold for any scheme S :(iii) The category Sh t (cid:0) P cor Λ ,S (cid:1) is a Grothendieck abelian category.(iv) The functor γ ∗ commutes with every limits and colimits. Proof.
The fact (i) implies (ii) follows from the preceding lemma as we can put a trt ( F ) = F tr with the notation of the lemma. The fact this defines a functor, as well as the properties statedin (ii), follows from the uniqueness statement of loc. cit. Let us assume (ii). Then (iii) follows formally because from (ii), from the existence, adjunctionproperty and exactness of a ∗ t , because PSh (cid:0) P cor Λ ,S (cid:1) is a Grothendieck abelian category. Moreover,we deduce from the isomorphism (10.3.9.1) that γ ∗ is exact: indeed, a ∗ t and ˆ γ ∗ are exact. As γ ∗ commutes with arbitrary direct sums, we get (iv).From this point, we deduce the existence of a right adjoint γ ! to the functor γ ∗ . Using againthe isomorphism (10.3.9.1), we obtain for any t -sheaves F on P /S and any P -scheme X/S acanonical isomorphism F τ ( X ) = γ ! F ( X ). This proves (i). (cid:3) . Under the assumption of the previous proposition, given any P -scheme X/S , we willput Λ trS ( X ) t = a ∗ t Λ trS ( X ). The above proposition shows that the family Λ trS ( X ) t for P -schemes X/S is a generating family in Sh t (cid:0) P cor Λ ,S (cid:1) . Moreover, we get easily the following corollary of thepreceding proposition and Proposition 10.1.2: Corollary . Assume that t is mildly compatible with transfers.Then there exists an essentially unique Grothendieck abelian P -premotivic category Sh t ( P cor Λ ) which is geometrically generated, whose fiber over a scheme S is Sh t (cid:0) P cor Λ ,S (cid:1) and such that the t -sheafification functor induces an adjunction of abelian P -premotivic categories: a ∗ t : PSh ( P cor Λ ) ⇄ Sh t ( P cor Λ ) : O trt . Moreover, the functor γ ∗ induces an adjunction of abelian P -premotivic categories: (10.3.11.1) γ ∗ : Sh t ( P , Λ) ⇄ Sh t ( P cor Λ ) : γ ∗ . Remark . Notice moreover that γ ∗ a ∗ t = a ∗ t ˆ γ ∗ . Proof.
In fact, using the exactness of a ∗ t , given any sheaf F with transfers F over S , we geta canonical isomorphism F = lim −→ Λ trS ( X ) t → F Λ trS ( X ) t where the limit is taken in Sh t (cid:0) P cor Λ ,S (cid:1) and runs over the representable t -sheaves with transfersover F . As in the proof of 10.1.2, this allows to define uniquely the structural left adjoints of
0. SHEAVES WITH TRANSFERS 207 Sh t ( P cor Λ ). The existence (and uniqueness) of the structural right adjoints then follows formally.The same remark allows to construct the functor γ ∗ . (cid:3) Remark . Let us explicit the meaning of the preceding Corollary for a topology t whichis compatible with transfers. Given a complex C with coefficients in the category Sh t (cid:0) P cor Λ ,S (cid:1) , thefollowing conditions are equivalent:(i) C is local (Definition 5.1.9),(i’) γ ∗ ( C ) is local,(i”) given any P -scheme X/S and any integer n ∈ Z , the canonical map H n ( C ( X )) → H nt ( X, γ ∗ ( C ))is an isomorphism,(ii) C is t -flasque (Definition 5.1.9),(ii’) γ ∗ ( C ) is t -flasque,(ii”) given any t -hypercover p : X → X in P /S and any integer n ∈ Z , the canonical map p ∗ : H n ( C ( X )) → H n ( C ( X ))is an isomorphism.More precisely, the equivalence of (i) and (ii) is the preceding corollary, while the equivalence of(i) and (i’) (resp. (ii) and (ii’)) follows from the existence of the adjunction (10.3.11.1) and thefact γ ∗ is exact. The equivalence between (i’) and (i”) (resp. (ii’) and (ii”)) is a simple translationof Definition 5.1.9.10.3.14 . Recall from Definition 5.1.9 we say that the abelian P -premotivic category Sh t ( P cor Λ )satisfies cohomological t -descent if for any scheme S , and any t -hypercover X → X in P /S , theinduced morphism of complexes in Sh t (cid:0) P cor Λ ,S (cid:1) Λ trS ( X ) t → Λ trS ( X ) t is a quasi-isomorphism. The preceding corollary thus gives the following one: Corollary . Assume t is mildly compatible with transfers.Then the following conditions are equivalent:(i) The topology t is compatible with transfers.(ii) The abelian P -premotivic category Sh t ( P cor Λ ) satisfies cohomological t -descent.(iii) The abelian P -premotivic category Sh t ( P cor Λ ) is compatible with t (see 5.1.9). Proof.
The equivalence of (i) and (ii) follows easily from the isomorphism (10.3.9.1). Theequivalence of (ii) and (iii) is Proposition 5.1.26 applied to the adjunction (10.3.11.1), in view of10.3.9(iv). (cid:3) . Recall from Paragraph 2.1.10 that a cd-structure P on S is the data of a family ofcommutative squares, called P -distinguished, of the form B k / / g (cid:15) (cid:15) Q Y f (cid:15) (cid:15) A i / / X (10.3.16.1)which is stable by isomorphisms. Further, we will consider the following assumptions on P :(a) P is complete, regular and bounded in the sense of [ Voe10c ].(b) Any P -distinguished square as above is made of P -morphisms and k is an immersion.(c) Any square as above which is cartesian and such that X = A ⊔ Y , i and f being theobvious immersions, is P -distinguished.Then the topology t P associated with P (see 2.1.10) is P -admissible and satisfy assumption10.0(c). Moreover, according to [ Voe10c , 2.9], we obtain the following properties: (d) A presheaf F on P /S is a t P -sheaf if and only if F ( ∅ ) = 0 and for any P -distinguishedsquare (10.3.16.1) in P /S , the sequence0 → F ( X ) f ∗ + e ∗ −−−−→ F ( Y ) ⊕ F ( A ) k ∗ − g ∗ −−−−→ F ( B )is exact.(e) For any P -distinguished square (10.3.16.1) the sequence of representable pre-sheaves on P /S → Λ S ( B ) k ∗ − g ∗ −−−−→ Λ S ( Y ) ⊕ Λ S ( A ) f ∗ + e ∗ −−−−→ Λ S ( X ) → t P -sheaves. Proposition . Consider a cd-structure P satisfying properties (a) and (b) above andassume that t = t P is the topology associated with P . Then the following conditions are equivalent:(i) The topology t is compatible with transfers.(ii) The topology t is mildly compatible with transfers.(iii) For any scheme S and any P -distinguished square (10.3.16.1) in P /S , the short sequenceof presheaves with transfers over S → Λ trS ( B ) k ∗ − g ∗ −−−−→ Λ trS ( Y ) ⊕ Λ trS ( A ) f ∗ + e ∗ −−−−→ Λ trS ( X ) → becomes exact on the associated t -sheaves on P /S . Proof.
The implication (i) ⇒ (ii) is obvious.The implication (ii) ⇒ (iii) follows from point (e) above and the following facts: γ ∗ is rightexact (Corollary 10.3.11), γ ∗ a t = a trt ˆ γ ∗ (remark 10.3.12), k ∗ : Λ trS ( B ) → Λ trS ( Y ) is a monomor-phism of presheaves with transfers (use 9.1.7(2) and the fact k is an immersion from assumption(b)).Assume (iii). Then we obtain (ii) as a direct consequence of the point (d) above. Thus,to prove (i), we have only to prove that the abelian P -premotivic category Sh t ( P cor Λ ) satisfiescohomological t -descent according to 10.3.15.Let S be a scheme. Recall that the category D(Sh t ( P /S, Λ) has a canonical DG-structure (seefor example 5.0.27). The cohomological t -descent for Sh t (cid:0) P cor Λ ,S (cid:1) can be reformulated by sayingthat for any complex K of t -sheaves on P /S , and any t -hypercover X → X , the canonical mapof D(Λ- mod) R Hom • D(Sh t ( P /S, Λ)) ( γ ∗ Λ trS ( X ) t , K ) → R Hom • D(Sh t ( P /S, Λ)) ( γ ∗ Λ trS ( X ) t , K )is an isomorphism. Recall also there is the injective model structure on C(Sh t ( P /S, Λ)) for whichevery object is cofibrant and with quasi-isomorphisms as weak equivalences (see [
CD09 , 2.1] formore details). Replacing K by a fibrant resolution for the injective model structure, we get forany simplicial objects X of P /S ∐ that: R Hom • D(Sh t ( P /S, Λ)) ( γ ∗ Λ trS ( X ) t , K ) = Hom • D(Sh t ( P /S, Λ)) ( γ ∗ Λ trS ( X ) t , K ) . Thus it is sufficient to prove that the presheaf E : P /S op → C(Λ- mod) , X Hom • D(Sh t ( P /S, Λ)) ( γ ∗ Λ trS ( X ) t , K )satisfies t -descent in the sense of 3.2.5.We derive from (iii) that E sends a P -distinguished square to a homotopy cartesian square inD(Λ- mod). Thus the assertion follows from the arguments on t -descent from [ Voe10b, Voe10c ]. (cid:3) Remark . It follows from Remark 10.3.13 that under the equivalent conditions (i),(ii), (iii) of the above corollary, the admissible topology t = t P is bounded in Sh t ( P cor Λ ) in thesense of Definition 5.1.28. Over a scheme S , a bounded generating family is given by the followingcomplexes of Sh t (cid:0) P cor Λ ,S (cid:1) : . . . → → Λ trS ( B ) k ∗ − g ∗ −−−−→ Λ trS ( Y ) ⊕ Λ trS ( A ) f ∗ + e ∗ −−−−→ Λ trS ( X ) → → . . . induced by a P -distignuished square of the form (10.3.16.1) – see also Example 5.1.29.
0. SHEAVES WITH TRANSFERS 209
We end-up this section with a compatibility of certain sheaves with transfers with projectivelimits of schemes. This will be the key point to establish continuity for motivic complexes.
Proposition . Let t be one of the topologies Nis , ´et , cdh .Let ( S α ) α ∈ A be a projective system of schemes in S , with dominant affine transition maps,and such that S = lim ←− α ∈ A S α is representable in S .Consider an index α ∈ A and a t -sheaf with transfers F over S ft,cor Λ ,S . For any index α/α ,we denote by F α (resp. F ) the pullback of F α over S α (resp. S ) in the sense of the premotivicstructure on Sh t ( P cor Λ ) (obtained in Corollary 10.3.11).Then the canonical map: lim −→ α ∈ A/α F α ( S α ) −→ F ( S ) is an isomorphism. Proof.
We consider the forgetful functor: O trt : Sh t (cid:16) S ft,cor Λ (cid:17) → PSh (cid:16) S ft,cor Λ (cid:17) . It is fullyfaithful and it commutes with the global section functor. We want to prove the proposition byusing Lemma 10.1.5. Thus it is sufficient to prove that, for any morphism f : X → S in S ,the functor O trt commutes with f ∗ . In other words, the pullback functor ˆ f ∗ for presheaves withtransfers on S ft,cor Λ preserves t -sheaves with transfers: for any t -sheaf with transfers F over S ,ˆ f ∗ ( F ) is a t -sheaf with transfers.Let us first treat the case where f is separated of finite type. Then ˆ f ∗ admits a left adjointˆ f ♯ which preserves t -covers. Thus the property is clear.In the general case, we write f as a projective limit of morphisms of schemes ( f α : X α → S ) α ∈ A such that the transition morphisms of the projective scheme ( X α ) α ∈ A are affine and dominantand each f α is separated of finite type. To check that ˆ f ∗ ( F ) is a t -sheaf, we consider a t -cover p : W → X of an S -scheme separated of finite type. Because of our choice of topology t , thereexists an index α /α such that p : W → X can be lifted as a t -cover p : W α → X α over S α .Using Lemma 10.1.5 again, we now are reduced to prove that for any α/α , ˆ f ∗ α ( F ) satisfies the t -sheaf property with respect to the pullback of p over S α /S α . This follows from the first casetreated. (cid:3) Remark . The previous proposition generalizes [
D´eg07 , Prop. 2.19]. . Assume that t is the Nisnevich topology. According to Lemma 10.3.3 and Proposition10.3.17, t is then compatible with transfers. With the notation of Corollary 10.3.11, we get thefollowing definition: Definition . We denote bySh tr ( − , Λ) , Sh tr ( − , Λ)the respective abelian premotivic and generalized abelian premotivic categories defined in Corol-lary 10.3.11 in the respective cases P = S m , P = S ft .From now on, the objects of Sh tr ( S, Λ) (resp. Sh tr ( S, Λ)) are called sheaves with transfers over S (resp. generalized sheaves with transfers over S ).Let X be a separated S -scheme of finite type. We let Λ tr S ( X ) be the generalized sheaf withtransfers represented by X ( cf. X is S -smooth, we denote by Λ trS ( X ) its restriction to S m cor Λ ,S – i.e. the sheaf with transfers over S represented by X .An important property of sheaves with transfers is that the abelian premotivic categorySh tr ( − , Λ) (resp. Sh tr ( − , Λ)) is compatible with the Nisnevich topology on S m (resp. S ft )according to Proposition 10.3.17. Note moreover that it is compactly geometrically generated. Write the O S -algebra f ∗ ( O X ) as the filtered union of its finite type sub- O S -algebras, ordered by inclusion. . We also obtained an adjunction (resp. generalized adjunction) of premotivic abeliancategories γ ∗ : Sh( S m, Λ) ⇄ Sh tr ( − , Λ) : γ ∗ γ ∗ : Sh (cid:0) S ft , Λ (cid:1) ⇄ Sh tr ( − , Λ) : γ ∗ . Note that in each case γ ∗ is conservative and exact according to 10.3.9(iv). Remark . An important application of the existence of the pair of adjoint functors( γ ∗ , γ ∗ ) is the following computation: given any complex K of sheaves with transfers over S andany smooth S -scheme X ,Hom D(Sh tr ( S, Λ)) (Λ trS ( X ) , K [ n ]) = Hom D(Sh tr ( S, Λ)) ( L γ ∗ Λ S ( X ) , K [ n ])= Hom D(Sh( S m, Λ)) (Λ S ( X ) , γ ∗ ( K )[ n ]) = H n Nis ( X, γ ∗ ( K )) . This is a generalization of [
VSF00 , chap. 5, 3.1.9] to unbounded complexes and arbitrary bases.10.4.5 . Let S be a scheme. Consider the inclusion functor ϕ : S m cor Λ ,S → S ft,cor Λ ,S . It inducesa functor ϕ ∗ : Sh tr ( S, Λ) → Sh tr ( S, Λ) , F F ◦ ϕ which is obviously exact and commute with arbitrary direct sums. In particular, it commutes witharbitrary colimits. Lemma . With the notations above, the functor ϕ ∗ admits a left adjoint ϕ ! such thatfor any smooth S -scheme X , ϕ ! (Λ trS ( X )) = Λ tr S ( X ) . The functor ϕ ! is fully faithful. In other words, we have defined an enlargement of premotivic abelian categories ( cf. definition1.4.13)(10.4.6.1) ϕ ! : Sh tr ( − , Λ) → Sh tr ( − , Λ) : ϕ ∗ . Proof.
Let F be a sheaf with transfers. Let { X/F } be the category of representable sheafΛ trS ( X ) over F for a smooth S -scheme X . We put ϕ ! ( F ) = lim −→ { X/F } Λ tr S ( X ) . The adjunction property of ϕ ! is immediate from the Yoneda lemma. We prove that for any sheafwith transfers F , the unit adjunction morphism F → ϕ ∗ ϕ ! ( F ) is an isomorphism. As alreadyremarked, ϕ ∗ commutes with colimits so that we are restricted to the case where F = Λ trS ( X )which follows by definition. (cid:3) . Assume now that t = cdh is the cdh-topology, and P = S ft is the class of separatedmorphisms of finite type. Recall the topology t is associated with the lower cd-structure – seeExample 2.1.11. Then the assumptions of Proposition 10.3.17 with respect to the lower cd-structure are fulfilled according to [ SV00b , 4.3.3] combined with [
SV00b , 4.2.9]. Thus we getthe following result:
Proposition . The admissible topology cdh on S ft is compatible with transfers. As a corollary, we get a generalized premotivic abelian category whose fiber over a scheme S is the category Sh tr cdh ( S, Λ) of cdh-sheaves with transfers on S ft . It is compatible with thecdh-topology. Moreover, the restriction of a cdh to Sh tr ( S, Λ) induces a morphism of generalizedpremotivic categories; we get the following commutative diagram of such morphisms:Sh( − , Λ) γ ∗ (cid:15) (cid:15) a ∗ cdh / / Sh cdh ( − , Λ) γ ∗ cdh (cid:15) (cid:15) Sh tr ( − , Λ) a ∗ cdh / / Sh tr cdh ( − , Λ)
0. SHEAVES WITH TRANSFERS 211
Change of coefficients. . Assume the topology t is mildly compatible with transfers and consider a localizationΛ ′ of Λ.Then the morphism (9.4.3.1) of P -premotivic categories extends to an adjunction of abelian P -premotivic categories:(10.5.1.1) Sh t ( P cor Λ ) ⊗ Λ Λ ′ ⇆ Sh t ( P cor Λ ′ )Proposition 9.4.4 immediately yields the following result: Proposition . Consider the above notations. Then the above P -premotivic adjunctionis an equality whenever it is restricted to one of the following subcategories of S : • The category of regular schemes. • The category of noetherian finite dimensional schemes S such that char( S ) ⊂ Λ × . Remark . Remark 9.4.5 can be extended to sheaves with transfers: for any regularscheme S , the category Sh tr ( S, Z ) = Sh Nis (cid:0) S m cor Z ,S (cid:1) defined here coincides with that defined in[ D´eg07 ], as well as its operations of a P -premotivic category when restricted to regular schemes.10.5.b. Representable qfh -sheaves. . Let us denote by Sh qfh ( S, Λ) the category of qfh-sheaves of Λ-modules over S ft /S .Remark that for an S -scheme X , the Λ-presheaf represented by X is not a sheaf for the qfh-topology. We denote the associated sheaf by Λ qfh S ( X ). We let a qfh be the associated qfh-sheaffunctor. Recall that for any S -scheme X , the graph functor (10.4.3) induces a morphism of sheavesΛ S ( X ) γ X/S −−−→ Λ tr S ( X ) . We recall the following theorem of Suslin and Voevodsky (see [
SV00b , 4.2.7+4.2.12]):
Theorem . Let S be a scheme such that char( S ) ⊂ Λ × . Then, for any S -scheme X ,the application of a qfh to the map γ X/S gives an isomorphism in Sh qfh ( S, Λ) : Λ qfh S ( X ) γ qfh X/S −−−→ Λ tr S ( X ) . . Assume char( S ) ⊂ Λ × . Using the previous theorem, we associate to any qfh-sheaf F ∈ Sh qfh ( S, Λ) a presheaf with transfers ρ ( F ) : X Hom Sh qfh ( S, Λ) (Λ tr S ( X ) , F ) . We obviously get γ ∗ ρ ( F ) = F as a presheaf over S ft /S so that ρ ( F ) is a sheaf with transfers andwe have defined a functor ρ : Sh qfh ( S, Λ) → Sh tr ( S, Λ) . For any S -scheme X , ρ (Λ qfh S ( X )) = Λ tr S ( X ) according to the previous proposition. Corollary . Assume char( S ) ⊂ Λ × . Let f : X ′ → X be a morphism of S -schemes.If f is a universal homeomorphism, then the map f ∗ : Λ tr S ( X ′ ) → Λ tr S ( X ) is an isomorphism in Sh tr ( S, Λ) . Proof.
Indeed, according to [
Voe96 , 3.2.5], Λ qfh S ( X ′ ) → Λ qfh S ( X ) is an isomorphism inSh qfh ( S, Λ) and we conclude by applying the functor ρ . (cid:3) -sheaves and transfers. Proposition . Assume char( S ) ⊂ Λ × . Any qfh -sheaf of Λ -modules over the categoryof S -schemes of finite type is naturally endowed with a unique structure of a sheaf with transfers,and any morphism of such qfh -sheaves is a morphism of sheaves with transfers.In particular, the qfh -sheafification functor defines a left exact functor left adjoint to theforgetful functor ρ : Sh qfh ( S, Λ) → Sh tr ( S, Λ) introduced in 10.5.6. Proof.
It follows from Theorem 10.5.5 that the category of Λ-linear finite correspondences iscanonically equivalent to the full subcategory of the category of qfh-sheaves of Λ-modules spannedby the objects of shape Λ qfh S ( X ) for X separated of finite type over S . The first assertion is thusan immediate consequence of Theorem 10.5.5 and of the (additive) Yoneda lemma. The fact thatthe qfh-sheafification functor defines a left adjoint to the restriction functor ρ is then obvious,while its left exactness is a consequence of the facts that it is left exact (at the level of sheaveswithout transfers) and that forgetting transfers defines a conservative and exact functor from thecategory of Nisnevich sheaves with transfers to the category of Nisnevich sheaves. (cid:3) Recall the following theorem:
Theorem . Assume Λ is a Q -algebra. Let F be an ´etale Λ -sheaf on S ft /S . Then forany S -scheme X , and any integer i , the canonical map H i Nis ( X, F ) → H i ´et ( X, F ) is an isomorphism. Proof.
Using the compatibility of ´etale cohomology with projective limits of schemes, weare reduced to prove that H i ´et ( X, F ) = 0 whenever X is henselian local and i >
0. Let k be theresidue field of X , G its absolute Galois group and F the restriction of F to Spec ( k ). Then F isa G -module and according to [ SGA4 , 8.6], H i ´et ( X, F ) = H i ( G, F ). As G is profinite, this groupmust be torsion so that it vanishes by assumption. (cid:3) Remark . The preceding theorem also follows formally from Theorem 3.3.23.
Proposition . Assume Λ is a Q -algebra. Let S be an excellent scheme and F be a qfh -sheaf of Λ -modules on S ft /S . Then for any geometrically unibranch S -scheme X of finitetype, and any integer i , the canonical map H i Nis ( X, F ) → H i qfh ( X, F ) is an isomorphism. Proof.
According to 10.5.9, H i Nis ( X, F ) = H i ´et ( X, F ). Let p : X ′ → X be the normalizationof X . As X is an excellent geometrically unibranch scheme, p is a finite universal homeomorphism.It follows from [ SGA4 , VII, 1.1] that H i ´et ( X, F ) = H i ´et ( X ′ , F ) and from [ Voe96 , 3.2.5] that H i qfh ( X, F ) = H i qfh ( X ′ , F ). Thus we can assume that X is normal, and the result is now exactly[ Voe96 , 3.4.1]. (cid:3)
Corollary . Assume Λ is a Q -algebra. Let S be an excellent scheme.(1) Let X be a geometrically unibranch S -scheme of finite type. For any point x of X ,the local henselian scheme X hx is a point for the category of sheaves Sh qfh ( S, Λ) ( i.e. evaluating at X hx defines an exact functor).(2) The family of points X hx of the previous type is a conservative family for Sh qfh ( S, Λ) . Proof.
The first point follows from the previous proposition. For any excellent scheme X ,the normalization morphism X ′ → X is a qfh-cover. Thus the category Sh qfh ( S, Λ) is equivalentto the category of qfh-sheaves on the site made of geometrically unibranch S -schemes of finitetype. This implies the second assertion. (cid:3) . Given any scheme S , we introduce the following composite functor using the notationsof 10.5.6 and 10.4.5: ψ ∗ : Sh qfh ( S, Λ) ρ −→ Sh tr ( S, Λ) ϕ ∗ −−→ Sh tr ( S, Λ) . Theorem . Assume Λ is a Q -algebra and let S be a geometrically unibranch excellentscheme. Considering the above notation, the following conditions are true :(i) For any S -scheme X of finite type, ψ ∗ (cid:0) Λ qfh S ( X ) (cid:1) = Λ trS ( X ) .(ii) The functor ψ ∗ admits a left adjoint ψ ! .(iii) For any smooth S -scheme X , ψ ! (cid:0) Λ trS ( X ) (cid:1) = Λ qfh S ( X ) .(iv) The functor ψ ∗ is exact and preserves colimits.
1. MOTIVIC COMPLEXES 213 (v) The functor ψ ! is fully faithful. According to property (iii), the functor ψ ! commutes with pullbacks. In other words, we havedefined an enlargement of abelian premotivic categories ( cf. definition 1.4.13) over the categoryof (noetherian) geometrically unibranch schemes:(10.5.14.1) ψ ! : Sh tr ( − , Λ) ⇄ Sh qfh ( − , Λ) : ψ ∗ Proof.
Point (i) follows from the fact Λ trS ( X ) = Λ qfh S ( X ). Recall the enlargement of (10.4.6.1): ϕ ! : Sh tr ( − , Λ) → Sh tr ( − , Λ) : ϕ ∗ . We define the functor ψ ! as the composite :Sh tr ( S, Λ) ϕ ! −→ Sh tr ( S, Λ) γ ∗ −→ Sh( S, Λ) a qfh −−→ Sh qfh ( S, Λ) . According to the properties of the functors in this composite, ψ ! is exact and preserves colimits.Moreover, for any smooth S -scheme X , as Λ trS ( X ) is a qfh-sheaf over S ft /S according to 10.2.4, ψ ! (Λ trS ( X )) = Λ qfh S ( X ) which proves (iii). Property (ii) follows from (iii) and the fact ψ ! commuteswith colimits, while the sheaves Λ trS ( X ) for X/S smooth generate Sh tr ( S, Λ).For any smooth S -scheme X , Γ( X ; ψ ∗ ( F )) = F ( X ). Thus the exactness of ψ ∗ follows fromcorollary 10.5.12. As ψ ∗ obviously preserves direct sums, we get (iv).To check that for any sheaf with transfers F the unit map F → ψ ∗ ψ ! ( F ) is an isomorphism,we thus are reduced to the case where F = Λ trS ( X ) for a smooth S -scheme X which follows from(i) and (iii). (cid:3)
11. Motivic complexes . In this section, S is the category of noetherian finite dimensional schemes. It is adequatein the sense of 2.0. Given a scheme S , we denote by S m S the category smooth separated S -schemesof finite type. It is admissible in the sense of 1.0.We fix a ring of coefficients Λ. Premotivic categories.
According to Proposition 10.3.17 and Corollary 10.3.15, theabelian premotivic category Sh tr ( − , Λ) constructed in 10.4.2 is compatible with Nisnevich topol-ogy. Thus we can apply to it the general definitions of section 5. This gives the following definition:
Definition . We define the (Λ-linear) category of motivic complexes (resp. stablemotivic complexes or simply motives ) following definition 5.3.22 (resp. definition 5.2.16) asDM eff Λ = D eff A (cid:0) Sh tr ( − , Λ) (cid:1) resp. DM Λ = D A (cid:0) Sh tr ( − , Λ) (cid:1) . Given a scheme S , we will put: DM eff ( S, Λ) = DM eff Λ ( S ), DM( S, Λ) = DM Λ ( S ).11.1.2 . Let us unfold the preceding definition. Given a scheme S in S , the triangulatedcategory DM eff ( S, Λ) is equal to the A -localization of the derived category D(Sh tr ( S, Λ)) of thecategory of sheaves with transfers over S .Given a smooth scheme S -scheme X of finite type, we have denoted by Λ trS ( X ) the sheaf withtransfers represented by X over S . We will see this sheaf as an object of DM eff ( S, Λ), as a complexconcentrated in degree 0, and call it the effective motivic complex associated with
X/S .Recall the following operations as part of the premotivic structure: • Given any morphism f : T → S in S , there exists an adjunction of the form: L f ∗ : DM eff ( S, Λ) ⇆ DM eff ( T, Λ) : R f ∗ . • Given a separated smooth morphism of finite type f : T → S in S , there exists anadjunction of the form: L f ♯ : DM eff ( S, Λ) ⇆ DM eff ( T, Λ) : f ∗ . • Given any noetherian finite dimensional scheme S , the category DM eff ( S, Λ) is symmetricclosed monoidal.These operations are subject to the properties of a premotivic category: functoriality, smoothbase change formula, smooth projection formula – see section 1 for more details. By construction,the triangulated premotivic category DM eff Λ satisfies the homotopy property and the Nisnevichdescent properties.By construction (cf. (5.3.23.2)), we get an adjunction of triangulated premotivic categories(11.1.2.1) Σ ∞ : DM eff Λ ⇄ DM Λ : Ω ∞ . Considering the
Tate motivic complex (11.1.2.2) Λ trS (1) := Λ trS ( P S / { } ) , the object Σ ∞ (Λ trS (1)) is ⊗ -invertible in DM( S, Λ) and this property characterizes uniquely thehomotopy category DM( S, Λ) – see Remark 5.3.29. Given a smooth separated S -scheme X offinite type, we put: M S ( X ) := Σ ∞ Λ trS ( X )and simply call it the motive associated with X/S . Usually we denote by S the unit of themonoidal category DM( S, Λ).By construction, the premotivic category DM Λ satisfies the homotopy, stability and Nisnevichdescent properties (see Paragraph 5.3.23). Example . • Let k be a perfect field. Then DM eff ( k, Z ) contains as a fullsubcategory the category DM eff − ( k ) defined by Voevodsky (cf [ VSF00 , Chap. 5]). Thisis the content of the proof of [
VSF00 , Chap. 5, Prop. 3.2.3]. Indeed, recall fromParagraph 5.2.18 that DM( k, Z ) is equivalent to the full subcategory of D(Sh tr ( k, Z ))made by the complexes which are A -local. Over a perfect field, Theorem 3.1.12 of[ VSF00 , Chap. 5] implies that a complex of sheaves with transfers is A -local if andonly if its homotopy sheaves are A -invariant. • Let S be a regular scheme. The triangulated categories DM eff ( S, Z ) and DM( S, Z )introduced here coincide with that constructed in [ CD09 ]. The same is true concerningthe operations of premotivic triangulated categories (see Remark 10.5.3).11.1.4 . Let Λ ′ be a localization of Λ. The premotivic adjunction(11.1.4.1) Sh tr ( − , Λ) ⊗ Λ Λ ′ ⇆ Sh tr ( − , Λ ′ )obtained as a particular case of (10.5.1.1) gives the following adjunctions of triangulated premotiviccategories: DM Λ ⊗ Λ Λ ′ ⇆ DM Λ ′ , DM eff Λ ⊗ Λ Λ ′ ⇆ DM eff Λ ′ . (11.1.4.2)Proposition 10.5.2 gives the following result: Proposition . The above premotivic adjunctions are equalities whenever it is restrictedto one of the following subcategories of S : • The category of regular schemes. • The category of noetherian finite dimensional schemes S such that char( S ) ⊂ Λ × . In other words, when S is a scheme of one of the categories listed above, the triangulatedmonoidal category DM( S, Λ ′ ) (resp. DM eff ( S, Λ ′ )) is the naive localization of the categoryDM( S, Λ) (resp. DM eff ( S, Λ)) with respect to integers invertible in Λ ′ .
1. MOTIVIC COMPLEXES 215
Constructible and geometric motives. . The premotivic triangulated category DM eff Λ is geometrically generated: given anyscheme S , the essentially small set G effS of motivic complexes of the form Λ trS ( X ) for a smooth sep-arated S -scheme X of finite type form a set of generators in the triangulated category DM eff ( S, Λ).Similarly, the premotivic triangulated category DM Λ is Z -generated where Z is the set oftwists corresponding to the Tate twist: given any scheme S , the essentially small set G S of motivesof the form M S ( X )( n ) for a smooth separated S -scheme X of finite type and an integer n ∈ Z form a set of generators in the triangulated category DM( S, Λ).Following the general conventions about premotivic triangulated category (Definition 1.4.9),we define the notion of constructibility for motives as follows:
Definition . Given any scheme S , we define the category of constructible motives (resp. constructible motivic complexes ) over S as the thick triangulated subcategory of DM( S, Λ) (resp.DM eff ( S, Λ)) generated by G S (resp. G effS ). We denote it by DM c ( S, Λ) (resp. DM eff c ( S, Λ)).
Remark . Recall that DM c, Λ (resp. DM eff c, Λ ) is Sm -fibred monoidal subcategory of DM Λ (resp. DM eff Λ ) over S . In other words, constructible motives (resp. motivic complexes) are stableby the operations f ∗ , p ♯ for p smooth and tensor product. This is obvious from definitions.11.1.9 . Let S be a scheme. Consider the triangulated subcategory V S of K b ( S m cor Λ ,S ) generatedby complexes of one the following forms :(1) for any distinguished square W k / / g (cid:15) (cid:15) V f (cid:15) (cid:15) U j / / X of smooth S -schemes,[ W ] g ∗ − k ∗ −−−−→ [ U ] ⊕ [ V ] j ∗ + f ∗ −−−−→ [ X ](2) for any smooth S -scheme X , p : A X → X the canonical projection.[ A X ] p ∗ −→ [ X ] . Definition . We define the category DM eff gm ( S, Λ) of geometric effective motives over S as the pseudo-abelian envelope of the triangulated categoryK b ( S m cor Λ ,S ) / V S . We define the category DM gm, Λ ( S ) of geometric motives over S as the triangulated categoryobtained from DM eff gm ( S, Λ) by formally inverting the Tate complex[ P S ] → [ S ] . Remark . The categories of geometric motives (resp. effective geometric motives) overan arbitrary base, as defined here, already appears in the work of Ivorra [
Ivo07 , sec. 1.3].11.1.12 . According to this definition, we can construct for any scheme S a commutativediagram of functors: DM eff gm ( S, Λ) / / (cid:15) (cid:15) DM eff ( S, Λ) Σ ∞ (cid:15) (cid:15) DM gm ( S, Λ) / / DM( S, Λ)(11.1.12.1)where the right vertical map is the left adjoint of (11.1.2.1).Recall from Remark 10.3.18 that the Nisnevich topology is bounded in Sh tr ( − , Λ). Thus, as acorollary of Proposition 5.2.38, Corollary 5.2.39 and Corollary 5.3.42 we get the following result:
Theorem . The horizontal functors of the square (11.1.12.1) are fully faithful andtheir essential images consist of constructible objects in the sense of Definition 11.1.7.Given any motive (resp. motivic complex) M over S , the following conditions are equivalent:(i) M is geometric ( i.e. in the image of the horizontal map of diagram (11.1.12.1) ), (ii) M is constructible,(iii) M is compact.The triangulated category DM( S, Λ) (resp. DM eff ( S, Λ) ) is compactly generated. More pre-cisely, the objects of the set of generators G S (resp. G effS ) defined in Paragraph 11.1.6 are compact. Remark . If S = Spec ( k ) is the spectrum of a perfect field, then the categoriesDM gm ( S, Λ) and DM eff gm ( S, Λ) coincide with the categories introduced by Voevodsky in [
VSF00 ,chap. 5, Sec. 2.1]. The above theorem is a generalization of [
VSF00 , chap. 5, Th. 3.2.6] to anarbitrary base (and the non effective case).11.1.c.
Enlargement, descent and continuity. . We can apply the definitions of section 5 to the generalized abelian premotivic cate-gory Sh tr ( − , Λ) constructed in 10.4.2
Definition . We define the (Λ-linear) category of generalized motivic complexes (resp. generalized motives ) following definition 5.3.22 (resp. definition 5.2.16) asDM eff Λ = D eff A (cid:0) Sh tr ( − , Λ) (cid:1) resp. DM Λ = D A (cid:0) Sh tr ( − , Λ) (cid:1) . . The advantage of this definition is that any separated S -scheme X of finite typedefines a generalized motivic complex, given by the sheaf with transfers Λ tr S ( X ) seen as a complexconcentrated in degree 0 (see Definition 10.4.2).The category DM eff Λ , as a generalized premotivic category, admits the following operations: • Given any morphism f : T → S in S , there exists an adjunction of the form: L f ∗ : DM eff ( S, Λ) ⇆ DM eff ( T, Λ) : R f ∗ . • Given a separated morphism f : T → S of finite type in S (non necessarily smooth),there exists an adjunction of the form: L f ♯ : DM eff ( S, Λ) ⇆ DM eff ( T, Λ) : f ∗ . • Given any noetherian finite dimensional scheme S , the category DM eff ( S, Λ) is symmetricclosed monoidal.These operations satisfies the properties of a genaralized premotivic category for which we referthe reader to section 1.4.As in the non generalized case, we get from the general construction (see (5.3.23.2)) an ad-junction of triangulated generalized premotivic categories(11.1.17.1) Σ ∞ : DM eff Λ ⇄ DM Λ : Ω ∞ . To any separated S -scheme X of finite type, we associate a generalized motive as: M S ( X ) := Σ ∞ Λ tr S ( X ) . By construction, the generalized premotivic category DM eff Λ (resp. DM Λ ) satisfies the homo-topy property, Nisnevich descent property (resp. and stability property).11.1.18 . According to Remark 10.3.18, the Nisnevich topology is bounded in Sh tr ( − , Λ).Thus, as a corollary of Proposition 5.2.38 (resp. Corollary 5.2.39), we obtain in particular thatDM eff ( S, Λ) (resp. DM( S, Λ)) is compactly generated, with the essentially small family of objectsΛ tr S ( X ) (resp. M S ( X )( n )) for a separated S -scheme of finite type X (resp. and an integer n ∈ Z )as compact generators.Recall that for any scheme S , the obvious restriction functor ϕ ∗ : Sh tr ( S, Λ) → Sh tr ( S, Λ)admits a left adjoint ϕ ! which is fully faithful (Lemma 10.4.6). Moreover, the adjoint pair ( ϕ ! , ϕ ∗ )satisfies the assumption of Proposition 6.1.4 so that applying Corollary 6.1.9 gives the followingproposition:
1. MOTIVIC COMPLEXES 217
Proposition . Given any scheme S , the adjoint pair ( ϕ ! , ϕ ∗ ) can be derived andinduces the following pair of adjoint functors ϕ ! : DM( S, Λ) ⇄ DM( S, Λ) : ϕ ∗ , resp. ϕ ! : DM eff ( S, Λ) ⇄ DM eff ( S, Λ) : ϕ ∗ , (11.1.19.1) such that ϕ ! is fully faithful.More generally, the family of these adjunctions for a noetherian finite dimensional scheme S defines an enlargement of premotivic categories (Definition 1.4.13). The abuse of notations is justified because of the following essentially commutative diagramof functors: DM eff Λ Σ ∞ / / ϕ ! (cid:15) (cid:15) DM Λ ϕ ! (cid:15) (cid:15) DM eff Λ Σ ∞ / / DM Λ (11.1.19.2)Recall that, given a smooth separated S -scheme X , we have the relation: ϕ ! ( M S ( X )) = M S ( X ) . Remark . Beware that the functor ϕ ∗ is far from being conservative. The followingexample was suggested by V.Vologodsky: let Z be a nowhere dense closed subscheme of S . Then ϕ ∗ ( M S ( Z )) = 0. In fact, one can see that DM( S, Λ) is a localization of the category DM( S, Λ)with respect to the objects M such that ϕ ∗ ( M ) = 0.11.1.21 . With rational coefficients, the preceding proposition can be refined. Recall that theqfh-sheafification functor (10.5.8) induces by 5.3.28 a premotivic adjunction α ∗ : DM Q ⇄ DM qfh , Q : α ∗ . Theorem . If S is a geometrically unibranch excellent noetherian scheme of finitedimension then the following composite functor α ∗ ϕ ! : DM( S, Q ) → DM qfh , Q ( S ) is fully faithful. Proof.
Note that DM eff ( S, Q ) and D eff A (Sh qfh ( S, Q )) are compactly generated; see example5.1.29 and Proposition 5.2.38. Hence this corollary follows from Theorem 10.5.14 and Proposition6.1.8. (cid:3) Remark . Recall this theorem can be rephrased by saying that motives over S satisfiesqfh-descent – see Remark 5.2.11 and more generally Section 3. In the next section, we will giveapplications of this fact to motivic cohomology. Theorem . The following assertions hold:(1) The triangulated premotivic categories DM eff Λ and DM Λ are continuous when we restrictourself to pro-schemes whose transition maps are affine and dominant.(2) The generalized triangulated premotivic categories DM eff Λ and DM Λ are continuous withthe same restriction on pro-schemes. Proof.
Note that Proposition 10.3.19 shows precisely that the generalized premotivic abeliancategory Sh tr ( − , Λ) satisfies Property (C) of Paragraph 5.1.35. Therefore the assertion (2) followsfrom Propositions 5.2.41 and 5.3.44. Moreover, the assertion (1) follows from Corollary 6.1.12 given the enlargement obtained inProposition 11.1.19. (cid:3) These propositions are also true with the restriction on pro-schemes considered in the statement of theTheorem. Example . From the previous theorem and Proposition 4.3.4, we obtain in particularthat for any pro-scheme ( S α ) α ∈ A with affine and dominant transition map such that S = lim ←− α ∈ A S α is noetherian finite dimensional, there exists canonical equivalences of categories:2- lim −→ α (cid:0) DM eff gm, Λ ( S α ) (cid:1) → DM eff gm, Λ ( S ) ,
2- lim −→ α (cid:0) DM gm, Λ ( S α ) (cid:1) → DM gm, Λ ( S ) . This result generalizes [
Ivo07 , 4.16].
Definition and functoriality.
Definition . Let S be a scheme and ( n, m ) ∈ Z a couple of integer. We define the motivic cohomology of S in degree n and twist m with coefficients in Λ as the Λ-module H n,m M ( S, Λ) = Hom
DM( S, Λ) (cid:0) S , S ( m )[ n ] (cid:1) . Assuming m ≥
0, we define the effective motivic cohomology of S in degree n and twist m withcoefficients in Λ as the Λ-module H n,m M ,eff ( S, Λ) = Hom DM eff ( S, Λ) (cid:0) Λ trS , Λ trS ( m )[ n ] (cid:1) . Motivic cohomology (resp. effective motivic cohomology) is contravariant with respect tomorphisms of schemes and the monoidal structure on DM Λ (resp. DM eff Λ ) defines a ring structurecompatible with pullbacks: given two cohomology classes: α : S → S ( m )[ n ] , α ′ : S → S ( m ′ )[ n ′ ] , one simply put: α.α ′ = α ⊗ S α ′ . The link between motivic cohomology and effective motivic cohomology is provided by Proposi-tion 5.3.39. Given any scheme S and any couple of integers ( n, m ) ∈ Z , one has a canonicalidentification: H n,m M ( S, Λ) = lim −→ r>> Hom DM eff ( S, Λ) (cid:0) Λ trS ( r ) , Λ trS ( m + r )[ n ] (cid:1) . . Let Λ ′ be a localization of Λ. Then using the left adjoint of the premotivic adjunction(11.1.4.2), we get a canonical morphism H n,m M ( S, Λ) ⊗ Λ Λ ′ → H n,m M ( S, Λ ′ ) . It is obviously compatible with pullbacks and the product structure. According to Proposition11.1.5, this map is an isomorphism (even an identity) when one the following conditions arefulfilled: • S is a regular schemes. • char( S ) ⊂ Λ × . Example . Let k be a perfect field. Given any smooth separated k -scheme S of finitetype, with structural morphism f , and any pair of integers ( n, m ) ∈ Z , motivic cohomology asdefined in the previous definition coincide with motivic cohomology as defined by Voevodsky in[ VSF00 , chap. 5] according to the following computation and Remark 11.1.14: H n,m M ( X, Z ) = Hom DM( X, Z ) ( X , X ( m )[ n ]) = Hom DM( X, Z ) ( X , f ∗ ( k )( m )[ n ])= Hom DM( k, Z ) ( L f ♯ ( X ) , k ( m )[ n ]) = Hom DM( k, Z ) ( M k ( X ) , k ( m )[ n ])= Hom DM gm ( k, Z ) ( M k ( X ) , k ( m )[ n ]) . In particular, it coincides with higher Chow groups (cf [
Voe02a ]) according to the followingformula: H n,m M ( X, Z ) = CH m ( X, m − n ) .
1. MOTIVIC COMPLEXES 219
Recall in particular the following computations: H n,m M ( X, Z ) = Z π ( X ) if n = m = 0 , G m ( X ) if n = m = 1 ,CH m ( X ) if n = 2 m, m < , n > min( m + dim( X ) , m )where CH m ( X ) is the usual Chow group of m -codimensional cycles in X .Note we will extend the identification of motivic cohomology as defined in the previous defi-nition with the general version defined by Voevodsky – [ Voe98 ] – in section 11.2.c.11.2.4 . Consider a separated morphism p : X → S of finite type. Recall from the S ft -fibredstructure of DM Λ that M S ( X ) = L p ♯ p ∗ ( S ). Using the adjunction property of the pair ( L p ♯ , p ∗ ),we easily get: H n,m M ( X, Λ) = Hom
DM( X, Λ) (cid:0) X , X ( m )[ n ] (cid:1) = Hom DM( X, Λ) (cid:0) X , X ( m )[ n ] (cid:1) = Hom DM( S, Λ) (cid:0) M S ( X ) , S ( m )[ n ] (cid:1) . (11.2.4.1)In particular, given any finite S -correspondence α : X •−→ Y between separated S -schemes offinite type, we obtain a pullback α ∗ : H n,m M ( Y, Λ) → H n,m M ( X, Λ)which is, among other properties, functorial with respect to composition of finite S -correspondencesand extends the natural contravariant functoriality of motivic cohomology.In particular, given any finite Λ-universal morphism f : Y → X , we obtain a pushout f ∗ : H n,m M ( Y, Λ) → H n,m M ( X, Λ)by considering the transpose of the graph of f . Proposition . Let f : Y → X be a finite Λ -universal morphism of schemes. Assume X is connected and let d > be the degree of f (cf. 9.1.12). Then for any element x ∈ H n,m M ( X, Λ) , f ∗ f ∗ ( x ) = d.x . This is a simple application of Proposition 9.1.13. We left to the reader the exercice to stateprojection and base change formulas for this pushout.
Example . Let f : Y → X be a finite morphism. Recall that f is Λ-universal in thefollowing particular cases: • f is flat (see Example 8.1.49); • X is regular and f sends the generic points of Y to generic points of X (see Corollary8.3.28).In particular, motivic cohomology is covariant with respect to this kind of finite morphisms.Another important application of the generalized motives is obtained using the Corollary10.5.7: Proposition . Let f : X ′ → X be a separated universal homeomorphism of finite type.Assume that char( X ) ⊂ Λ × . Then the pullback functor H n,m M ( X, Λ) → H n,m M ( X ′ , Λ) is an isomorphism. Remark . The preceding considerations hold similarly for the effective motivic coho-mology.
Example . In characteristic 0, motivic cohomology (effective and non effective) is in-variant under seminormalization ([
Swa80 ]).When restricted to excellent geometrically unibranch scheme X , motivic cohomology (effectiveand non effective) is invariant under normalization. Indeed, the normalization X ′ → X of such ascheme is a universal homeomorphism ([ EGA4 , IV , 23.2.2]) of finite type. Effective motivic cohomology in weight and . . Let S be a scheme and X a smooth S -scheme. For any subscheme Y of X , we denoteby Λ trS ( X/Y ) the cokernel of the canonical morphism of sheaf with transfers Λ trS ( Y ) → Λ trS ( X ). Asthis morphism is a monomorphism, we obtain a canonical distinguished triangle in DM eff ( S, Λ)Λ trS ( Y ) → Λ trS ( X ) → Λ trS ( X/Y ) → Λ trS ( X )[1] . Using this notation and according to Definition 2.4.17, the Tate motivic complex is defined as:Λ trS (1) = Λ trS ( P S / {∞} )[ − trS (1) = Λ trS ( P S / A S )[ −
2] = Λ trS ( A S / G m )[ − Proposition . Suppose S is a normal scheme.Then the sheaf on Sm S represented by G m admits a canonical structure of a sheaf with trans-fers and there is a canonical isomorphism in DM eff ( S, Λ) : G m ⊗ Z Λ ≃ −−→ Λ trS (1)[1] . Proof.
Let U be an open subscheme of A S and X be a smooth S -scheme. Note that X isnormal according to [ EGA4 , 18.10.7]. Consider a cycle α = X i n i . h Z i i of X × S U with n i ∈ Λ and Z i irreducible finite and dominant over an irreducible component of X .Then Z i is a divisor in X × S U and according to [ EGA4 , 21.14.3], it is flat over X . In other words, α is a Hilbert bert cycle which implies it is Λ-universal (Example 8.1.49). As a consequence, weobtain the equality H i Γ( X ; C ∗ Λ trS ( U )) = H sing − i ( X × S U/X ) ⊗ Z Λwhere the functor C ∗ is the associated Suslin singular complex (see (5.2.32.1)) and the right handside is the Suslin homology of X × S U/X ( cf. [ SV00b ]).Suppose in addition that X and U are affine and let Z = P S − U . According to a theorem ofSuslin and Voevodsky ( cf. [ SV00b , th. 3.1]), H sing − i ( X × S U/X ) = (cid:26)
Pic( X × S P S , X × S Z ) if i = 00 otherwise;the group on the left hand side is the relative Picard group . In particular, the complex C ∗ Λ trS ( U ),seen as a complex of presheaves with transfers, is concentrated in cohomological degree 0 and its0-th cohomology is the presheaf X Pic( X × S P S , X × S Z ) ⊗ Z Λ.Consider the following exact sequence of presheaves with transfers:0 → Λ trS ( G m ) → Λ trS ( A S ) → ˜Λ trS ( A S / G m ) → . Applying the functor C ∗ to it, relatively to the category of complexes of presheaves with transfers,we obtain a distinguished triangle in D(PSh tr ( S, Λ)): C ∗ Λ trS ( G m ) → C ∗ Λ trS ( A S ) → C ∗ ˜Λ trS ( A S / G m ) +1 −−→ C ∗ Λ trS ( G m ) . Taking the associated long exact sequence of cohomology presheaves, we obtain that the complexof presheaves with transfers C ∗ ˜Λ trS ( A S / G m ) is concentrated in cohomological degree 0 and − → ˆ H − [ C ∗ ˜Λ trS ( A S / G m )] → ˆ H [ C ∗ Λ trS ( G m )] → ˆ H [ C ∗ Λ trS ( A S )] → ˆ H [ C ∗ ˜Λ trS ( A S / G m )] → . By definition of the relative Picard group, given any smooth (affine) scheme X , we get an exactsequence of abelian groups:(11.2.11.1) 0 → G m ( X ) → Pic( X × S P S , X ⊔ X ∞ ) → Pic( X × S P S , X ) → .
1. MOTIVIC COMPLEXES 221
Thus we deduce that: ˆ H [ C ∗ ˜Λ trS ( A S / G m )] = 0 , ˆ H − [ C ∗ ˜Λ trS ( A S / G m )] = G m ⊗ Z Λ . This gives in particular a canonical isomorphism: C ∗ ˜Λ trS ( A S / G m )[ − ≃ G m ⊗ Z Λin D(PSh tr ( S, Λ)). Taking its image in DM eff ( S, Λ) we obtain a canonical isomorphism which canbe written as: C ∗ Λ trS ( A S / G m )[ − ≃ G m ⊗ Z Λ . Thus we can conclude because, according to Lemma 5.2.35, the canonical mapΛ trS ( A S / G m ) → C ∗ Λ trS ( A S / G m )is an isomorphism in DM eff ( S, Λ). (cid:3)
Remark . In the course of the proof, a canonical structure of a sheaf with transfersover S on G m has naturally appeared – described by the exact sequence (11.2.11.1). This structureis classical (see [ MVW06 , Ex. 2.4]). One can describe it as follows.Let X and Y be smooth S -schemes. Assume X is connected (thus irreducible as it is normal).Let Z be a closed integral subscheme Z of X × S Y which is finite surjective over X . Then Z/X corresponds to an extension of function fields
L/K . The norm map of
L/K induces amorphism of abelian groups: N Z/X : G m ( Z ) → G m ( X ). Then we associate with Z , seen as afinite correspondence from X to Y , the following morphism: G m ( Y ) p ∗ −→ G m ( Z ) N Z/X −−−−→ G m ( X )where p : Y → Z is the natural projection.The following proposition is well known to the expert. We include a proof for completeness. Proposition . For any regular scheme X and any interger i ≥ , H i Nis ( X, G m ) = O X ( X ) × if i = 0 , Pic( X ) if i = 1 , otherwisewhere Pic( X ) is the Picard group of X . Proof.
Let Y be any ´etale scheme over X . We let C ( V ) be the abelian group made bythe invertible rational functions on V and C ( V ) be the group of 1-codimensional cycles in V .Classically, one associates with any rational function f on V its Weil divisor div( f ) ∈ C ( V ).Recall, when V is integral with function field K , f ∈ K , one puts:div V ( f ) = X x ∈ V (1) v x ( f ) .x ;the sum runs over the points of codimension 1 in V and v x ( f ) is the valuation of f correspondingto the valuation ring O X,x .According to this definition, we get a complex:0 → G m ( V ) → C ( V ) div V −−−→ C ( V ) . This sequence is functorial with respect to pullback of ´etale X -schemes. Thus we have defined amorphism of presheaves on X ´et : π : G m → C ∗ . Given any Nisnevich distinguished square Q (Example 2.1.11), one can check easily that the imageof Q by C (resp. C ) is cocartesian. As a consequence C ∗ is a complex of Nisnevich sheaveswhich satisfies the Brown-Gersten property – i.e. it is Nisnevich flasque in the sense of Definition5.1.9 according to Proposition 5.2.13 applied to the derived category of Nisnevich sheaves over X . On the other hand, π is a quasi-isomorphism of Nisnevich sheaves over S : indeed it is wellknown that for any regular local ring A , the sequence0 → A × → Frac( A ) × div A −−−→ Z ( A ) → A is a unique factorization domain – the classicalAuslander-Buchsbaum theorem, (e.g. [ Mat70 , 20.3]).In particular, we get H i ( X, G m ) = H i ( C ∗ ( X )) and this concludes. (cid:3) The following theorem is a generalization of a well-known computation of Voevodsky forsmooth schemes over a perfect field. The second case is a corollary of the two preceding proposi-tions.
Theorem . Let S be a scheme and n ∈ Z an integer. The following computation holds:(1) H n, M ,eff ( S, Λ) = Hom DM eff ( S ) (Λ trS , Λ trS [ n ]) = (cid:26) Λ π ( S ) if n = 00 otherwise;(2) if S is regular, H n, M ,eff ( S, Λ) = Hom DM eff ( S ) (Λ trS , Λ trS (1)[ n ]) = O S ( S ) × ⊗ Z Λ if n = 1Pic( S ) ⊗ Z Λ if n = 20 otherwise Proof.
For the first case, according to Proposition 10.2.5, the sheaf Λ trS is Nisnevich localand A -local as a complex of sheaves. It is obviously acyclic for the Nisnevich topology. Thus, weconclude using again 10.2.5 in the case n = 0.Consider the second case. According to Proposition 11.2.13, the sheaf G m on Sm S is A -local.Thus according to Proposition 11.2.11 G m ⊗ Λ[ −
1] is an A -resolution of Λ trS (1). In particular,Hom DM eff ( S ) (Λ trS , Λ trS (1)[ n ]) = Hom D(Sh tr ( S, Λ)) (Λ trS , G m ⊗ Λ[ n − H n − ( S, G m ) ⊗ Λwhere the second identification follows from Remark 10.4.4. The conclusion follows from anotherapplication of Proposition 11.2.13. (cid:3)
The following corollary is a (very) weak cancellation result in DM eff ( S ) : Corollary . Let S be a regular scheme. Then R Hom (Λ trS (1) , Λ trS (1)) = Λ trS . Moreover, if m = 0 or m = 1 , for any integer n > m , R Hom (Λ trS ( n ) , Λ trS ( m )) = 0 . Proof.
We consider the first assertion. Any smooth S -scheme is regular. Hence it is sufficientto prove that for any connected regular scheme S , for any integer n ∈ Z ,Hom DM eff ( S ) (Λ trS (1) , Λ trS (1)[ n ]) = (cid:26) Λ if n = 00 otherwise.Using the exact triangle(11.2.15.1) Λ trS ( G m ) → Λ trS ( A ) → Λ trS (1)[2] +1 −−→ and the second case of the previous theorem, we obtain the following long exact sequence · · · → Hom(Λ trS ( A ) , Λ trS (1)[ n ]) → Hom(Λ trS ( G m ) , Λ trS (1)[ n ]) → Hom(Λ trS (1) , Λ trS (1)[ n − → Hom(Λ trS ( A ) , Λ trS (1)[ n + 1]) → · · · Then we conclude using the previous theorem and the factPic( A × S ) = Pic( G m × S )whenever S is regular.
1. MOTIVIC COMPLEXES 223
For the last assertion, we are reduced to prove that if S is a regular scheme, for any integers n > i , Hom DM eff ( S ) (Λ trS ( n ) , Λ trS [ i ]) = 0 . This is obviously implied by the case n = 1.Consider the distinguished triangle (11.2.15.1) again. Then the long exact sequence attachedto the cohomological functor Hom DM eff ( S, Λ) ( − , Λ trS ) and applied to this triangle together with thefirst case of the previous theorem allows us to conclude. (cid:3) The motivic cohomology ring spectrum. . According to definition 10.4.2 and paragraph 10.4.3, we have an adjunction of abelianpremotivic categories γ ∗ : Sh( − , Λ) ⇄ Sh tr ( − , Λ) : γ ∗ such that γ ∗ is conservative and exact. According to Paragraph 5.3.28, it induces an adjunctionof triangulated premotivic categories(11.2.16.1) L γ ∗ : D A , Λ ⇄ DM Λ : R γ ∗ . Composing with the premotivic adjunction between the stable homotopy category and the A -derived homtopoy category (5.3.35.1), we finally get a canonical premotivic adjunction:(11.2.16.2) ϕ ∗ : SH ⇄ DM Λ : ϕ ∗ . Recall that, because ϕ ∗ is monoidal, ϕ ∗ is weakly monoidal. In particular, for any scheme S , onegets canonical morphisms S → ϕ ∗ ( S ) , ϕ ∗ ( S ) ∧ ϕ ∗ ( S ) → ϕ ∗ ( S )which gives a structure of a commutative monoid to the spectrum ϕ ∗ ( S ) i.e. a ring spectrum. Definition . Given any scheme S , one defines the motivic cohomology ring spectrum over S with coefficients in Λ as the commutative ring spectrum:H Λ M ,S := ϕ ∗ ( S ) . The properties of the functor ϕ ∗ immediately implies that the ring spectrum H Λ M ,S representsmotivic cohomology. One easily checks now that this ring specutrm coincides with the originalone of Voevodsky (see [ Voe98 , sec. 6.1]) – in the case Λ = Z . Therefore, our definition of motiviccohomology (with Z -coefficients) agrees with that given by Voevodsky in loc. cit. . Consider a localization Λ ′ of Λ. Then one gets an essentially commutative diagramof premotivic adjunctions: D A ( S, Λ) ⊗ Λ Λ ′ t t ✐✐✐✐✐✐✐✐ DM( S, Λ) ⊗ Λ Λ ′ o o SH( S ) D A ( S, Λ ′ ) j j ❯❯❯❯❯❯❯❯ (1) O O DM( S, Λ ′ ) (2) O O o o where the map (1) is the canonical equivalence and the map (2) is the left adjoint of (11.1.4.2) (inthe stable case). Note that (2) is weakly monoidal. Thus this essentially commutative diagramdefines a canonical morphism of ring spectra:(11.2.18.1) H Λ M ,S ⊗ Λ Λ ′ → H Λ ′ M ,S . As a corollary of Proposition 11.1.5, we get the following result:
Proposition . If S ir regular or char( S ) ⊂ Λ × then the map (11.2.18.1) is an iso-morphism. Remark . We do not know what is the nature of the map (11.2.18.1) in the case of ageneral scheme S . . Let f : T → S be a morphism of schemes. Recall from the structure of the premotivicadjunction ( ϕ ∗ , ϕ ∗ ) defined above that we get an exchange morphism: f ∗ ϕ ∗ → ϕ ∗ f ∗ Applying this natural transformation to the unit object S of DM( S, Λ), one gets a canonicalmorphism of ring spectra: τ f : f ∗ (H Λ M ,S ) → H Λ M ,T . Remark that this shows the collection (H Λ M ,S ) is a section of the fibred category SH. Recall alsothe following conjecture of Voevodsky ([ Voe02b , conj. 17]):
Conjecture.
For any morphism f as above, the map τ f is an isomorphism. At least, Voevodsky formulated this conjecture in the case where Λ = Z . As we have warnthe reader in remark 11.2.20, the case Λ = Z does not necessarily implies the case of an arbitraryring Λ ⊂ Q . Remark . We will solve affirmatively a particular case of this conjecture in 16.1.7. . For any scheme S , we let P ∞ S be the ind-scheme S → P S → · · · → P nS → P n +1 S → made of the obvious closed immersions. This ind-scheme has a comultiplication given by the Segreembeddings P ∞ S × S P ∞ S → P ∞ S Define Λ trS ( P ∞ ) = lim −→ Λ trS ( P n ). Applying Theorem 11.2.14 in the case S = Spec ( Z ), weobtain a canonical isomorphism:Hom DM eff (Spec( Z ) , Λ) (Λ tr ( P ∞ ) , Λ tr (1)[2]) = Pic( P ∞ ) ⊗ Z Λ , whose aim is a free Λ-algebra of power series in one variable. Considering the canonical dualinvertible sheaf on P ∞ , we obtain a canonical formal generator of this Λ-algebra and thus amorphism DM eff (Spec ( Z ) , Λ): c : Λ tr ( P ∞ ) → Λ tr (1)[2] . For any scheme S , considering the canonical projection f : S → Spec ( Z ), we obtain by pullbackalong f a morphism of DM eff ( S, Λ) c ,S : Λ trS ( P ∞ S ) → Λ trS (1)[2] . Consider G m as a sheaf of groups over Sm S . Following [ MV99 , part 4], we introduce its clas-sifying space B G m as a simplicial sheaf over Sm S . From proposition 1.16 of loc. cit. , weget Hom H s • ( S ) ( S + , B G m ) = Pic( S ) . Moreover, in H • ( S ), we obtain a canonical isomorphism B G m = P ∞ S of pointed simplicial sheaf ( cf. loc. cit. , prop. 3.7). Thus finally, we obtain acanonical map of pointed setsPic( S ) = Hom H s • ( S ) ( S + , B G m ) → Hom H • ( S ) ( S + , P ∞ ) → Hom DM eff ( S, Λ) (Λ trS , Λ trS ( P ∞ / ∗ )) → Hom DM eff ( S, Λ) (Λ trS , Λ trS ( P ∞ )) . Definition . Consider the above notations. We define the first motivic Chern class asthe following composite morphism c : Pic( S ) −→ Hom DM eff ( S, Λ) (Λ trS , Λ trS ( P ∞ S )) ( c ,S ) ∗ −−−−→ Hom DM eff ( S, Λ) (Λ trS , Λ trS (1)[2]) −→ Hom
DM( S, Λ) ( S , S (1)[2]) = H , M ( S, Λ)The first motivic Chern class is evidently compatible with pullback.
1. MOTIVIC COMPLEXES 225
Remark . Beware that the mapPic( S ) → Hom DM eff ( S, Λ) (Λ trS , Λ trS ( P ∞ S ))defined above is not necessarily a morphism of abelian groups. However, the composite:Pic( S ) −→ Hom DM eff ( S, Λ) (Λ trS , Λ trS ( P ∞ S )) ( c ,S ) ∗ −−−−→ Hom DM eff ( S, Λ) (Λ trS , Λ trS (1)[2])is the isomorphism of Theorem 11.2.14 when S is normal. In particular, it is a morphism of abeliangroups in this case. We will give an argument below for the general case.11.3.4 . The triangulated category DM( S, Λ) thus satisfies all the axioms of [
D´eg08 , 2.1] (seealso 2.3.1 of loc. cit. in the regular case). In particular, we derive from the main results of loc.cit. the following facts:(1) Let p : P → S be a projective bundle of rank n . Let c : S → S (1)[2] be the first Chernclass of the canonical line bundle on P . Then the map M S ( P ) P i p ⊗ c i −−−−−→ n M i =0 S ( i )[2 i ]is an isomorphism. This gives the projective bundle theorem in motivic cohomology forany base scheme.One deduces using the method of Grothendieck that motivic cohomology possessesChern classes of vector bundles which satisfies all the usual properties (see remark belowfor additivity).(2) Let i : Z → X be a closed immersion between smooth separated S -schemes of finite type.Assume i has pure codimension c and let j be the complementary open immersion. Thenthere is a canonical purity isomorphism : p X,Z : M S ( X/X − Z ) → M S ( Z )( c )[2 c ] . This defines in particular the
Gysin triangle M S ( X − Z ) j ∗ −→ M S ( X ) i ∗ −→ M S ( Z )( c )[2 c ] ∂ X,Z −−−→ M S ( X − Z )[1] . (3) Let f : Y → X be a projective morphism between smooth separated S -schemes offinite type. Assume f has pure relative dimension d . Then there is an associated Gysinmorphism f ∗ : M S ( X ) → M S ( Y )( d )[2 d ]functorial in f . We refer the reader to loc. cit for various formulas involving the Gysinmorphism (projection formula, excess intersection,...)Note in particular that we deduce from that Gysin morphism the following map inmotivic cohomology: f ∗ : H n,i M ( Y, Λ) → H n +2 d,i + d M ( X, Λ) . (4) For any smooth projective S -scheme X , the premotive M S ( X ) admits a strong dual . If X has pure relative dimension d over S , the strong dual of M S ( X ) is M S ( X )( − d )[ − d ]. Remark . According to loc. cit. , there exists for any scheme S a formal group law F S ( x, y ) with coefficients in the graded ring H ∗ , ∗M ( S, Λ). If one consider the Segre embeddingΣ : P ∞ S → P ∞ S × S P ∞ S one has: F S ( x, y ) = σ ∗ (1) through the isomorphism: H ∗ , ∗M ( P ∞ S × S P ∞ S , Λ) ≃ H ∗ , ∗M ( S, Λ)[[ x, y ]]which results from the projective bundle formula in motivic cohomology.According to Remark 11.3.3, whenever S is normal, one gets F S ( x, y ) = x + y . In particular, F Spec( Z ) ( x, y ) = x + y . On the other hand, according to the above definition of F S ( x, y ), F S ( x, y )is compatible with pullback. Thus one deduces that F S ( x, y ) = x + y for any scheme S . . According to the properties that we have previously proved, motivic cohomology, andin particular the bigraded part H n,n M ( X, Z ), possesses many of the desired property of a generalizedChow theory for regular schemes (see [ SGA6 , XIV, § K ( X ) ⊗ Z Q ch −→ H ∗ , ∗M ( X, Z ) ⊗ Q ≃ H ∗ , ∗M ( X, Q )where the final isomorphism follows from Paragraph 11.2.2. In particular we will prove in the nextsection (Corollary 16.1.7) that, when X is regular, this map is an isomorphism as expected. Remark . Among the good properties of motivic cohomology is the fact it is defined,with its ring structure and natural functoriality, other arbitrary schemes. On the other hand,even when X is regular, one cannot describe at the moment the cohomology group H n,n M ( X, Z )in terms of classes of n -codimensional cycles in X modulo an appropriate equivalence relation.Let us however mention the two following interesting facts:(1) Let X be a scheme of finite type over Spec ( Z ) and X p be its fiber over a primer p . Thenone has a pullback map: H n,n M ( X, Z ) → H n,n M ( X p , Z ) , σ σ p . When X is an arithmetic scheme (regular and flat over Z ) with good reduction at p ,the target is the Chow group of n -codimensional cycles (see Example 11.2.3). Then σ p should be thought as the specialization of its generic fiber (which lies in H n,n M ( X Q , Z ) = CH n ( X Q ) according to the Example 11.2.3). This construction should coincide withother specialization maps in the arithmetic case (see for example [ Ful98 , § X be a smooth S -scheme. Then any n -codimensional closed subscheme Z of X whichis smooth over S defines using the Gysin morphism an element[ Z ] = i ∗ (1) ∈ H n,n M ( X, Z )which should be called the fundamental class of X . One can extract from [ D´eg08 ]some of the expected properties of these fundamental classes (relation to Chern classes,pullback properties such as compatibility with base change).In particular, any S -point of X defines an element of H d,d M ( X, Z ) where d is thedimension of X (assumed of pure dimension). In particular, the group H d,d M ( X, Z ) isclose to a group of cycles in X of relative dimension 0 over S .11.3.8 . We end up this series of remarks on motivic cohomology with the following constructionthat the reader might enjoy.Let S be any scheme and P S be the category of smooth projective S -schemes. Given anyscheme X and Y in P S , one can use the group H d,d M ( X × S Y, Λ)where d is the relative dimension of Y as a group of correspondences using the properties obtainedso far from motivic cohomology. In particular, one can mimic the construction of the category ofChow motives over a field k using the category P S and these correspondences. One obtains anadditive monoidal category Chow ′ ( S, Λ) of strong Chow motives .According to the duality property of motives (Paragraph 11.3.4, point 4) one also obtains acanonical isomorphism Hom
DM( S, Λ) ( M S ( X ) , M S ( Y )) = H d,d M ( X × S Y, Λ) . Thus one deduces a canonical full embedding of monoidal categories:Chow ′ ( S, Λ) → DM gm ( S, Λ)which extends the well known case when S is a perfect field. Remark . Beware that, with rational coefficients, a sharper notion of Chow motives– in more precise terms, these are motives of weight zero – have been introduced recently (see[
H´eb11 ], [
Bon10 ]).
1. MOTIVIC COMPLEXES 227 . Recall that according to Definition 10.4.2 and Paragraph 10.4.3, we have an adjunctionof abelian premotivic categories γ ∗ : Sh( − , Λ) ⇄ Sh tr ( − , Λ) : γ ∗ such that γ ∗ is exact and conservative. Moreover, for any scheme S , any smooth S -schemes X , Y and any open immersion j : U → X , the canonical map: j ∗ : c S ( Y, U ) → c S ( Y, X )is obviously a monomorphism. Thus the abelian premotivic category Sh tr ( − , Λ) satisfies theassumptions (i)-(iv) of Paragraph 6.3.1. In particular, we deduce from Corollaries 6.3.12 and6.3.15 the following theorem:
Proposition . The premotivic triangulated category DM Λ satisfies the support property.Moreover, for any scheme S and any closed immersion i : Z → X between smooth S -schemes, DM Λ satisfies the localization property with respect to i , (Loc i ). An important corollary of this proposition is that given any separated morphism f : Y → X of finite type, one can construct an adjunction of triangulated categories: f ! : DM( Y, Λ) ⇆ DM( X, Λ) : f ! such that f ! = f ∗ when f is proper (see Section 2.2). We will elaborate on this fact at the end ofthis section.11.4.3 . Note that in particular, the premotivic category DM Λ satisfies the weak localizationproperty (wLoc). According to the premotivic adjunction (11.2.16.2) and the existence of thefirst Chern class in motivic cohomology (Definition 11.3.2), one can apply Example 2.4.40 tothe premotivic triangulated category DM Λ (which satisfies the Nisnevich separation propertyby construction). This implies in particular that DM Λ is oriented as a premotivic triangulatedcategory (Definition 2.4.38).In particular, one can apply Corollary 2.4.43 to DM Λ and get the following result: Proposition . Any smooth projective morphism f is DM Λ -pure: the canonical puritymap (2.4.39.3) f ♯ → f ! ( d )[2 d ] , is an isomorphism where d is the relative dimension of f . In particular, DM Λ is weakly pure. The only property of the premotivic triangulated categoryDM Λ that we cannot prove is the localization property for general closed immersions. However,the properties we have seen so far allows to construct the 6 operations and establish some of itsproperties that are already of interest. Let us summarize this formalism, from Theorem 2.2.14together with Lemma 2.4.23: Theorem . For any separated morphism of finite type f : Y → X , there exists anessentially unique pair of adjoint functors f ! : DM( Y, Λ) ⇄ DM( X, Λ) : f ! such that:(1) There exists a structure of a covariant (resp. contravariant) -functor on f f ! (resp. f f ! ).(2) There exists a natural transformation α f : f ! → f ∗ which is an isomorphism when f isproper. Moreover, α is a morphism of -functors.(3) For any smooth projective morphism f : X → S of relative dimension d , there arecanonical natural isomorphisms p t f : f ♯ −→ f ! ( d )[2 d ] p ′ t f : f ∗ −→ f ! ( − d )[ − d ] which are dual to each other. (4) For any cartesian square: Y ′ f ′ / / g ′ (cid:15) (cid:15) ∆ X ′ g (cid:15) (cid:15) Y f / / X, such that f is separated of finite type, there exist natural transformations g ∗ f ! ∼ −→ f ′ ! g ′∗ ,g ′∗ f ′ ! ∼ −→ f ! g ∗ , which are isomorphisms in the following cases: • g is smooth; • f is projective and smooth.(5) For any smooth projective morphism f : Y → X , there exist natural isomorphisms Ex ( f ∗ ! , ⊗ ) : ( f ! K ) ⊗ X L ∼ −−→ f ! ( K ⊗ Y f ∗ L ) , Hom X ( f ! ( L ) , K ) ∼ −−→ f ∗ Hom Y ( L, f ! ( K )) ,f ! Hom X ( L, M ) ∼ −−→ Hom Y ( f ∗ ( L ) , f ! ( M )) . Remark . As an example of application, let us recall the construction of the generaltrace map (from [
SGA4 ]) in the case of a smooth projective morphism f : Y → X of relativedimension d . It is the following composite map: f ∗ f ∗ α − f −−→ f ! f ∗ p ′ t f −−→ f ! f ! ( d )[2 d ] ad ′ ( f ! ,f ! ) −−−−−−→ d )[2 d ] . This allows one to recover the Gysin map associated with f , already constructed in Paragraph11.3.4, as well as the duality property for the motive M X ( Y ). art 4 Beilinson motives and algebraic K-theory . In all this part, S is assumed to be the category of noetherian schemes of finite dimen-sion.
12. Stable homotopy theory of schemes12.1. Ring spectra.
Consider a base scheme S .Recall that a ring spectrum E over S is a monoid object in the monoidal category SH( S ).We say that E is commutative if it is commutative as a monoid in the symmetric monoidalcategory SH( S ). In what follows, we will assume that all our ring spectra are commutativewithout mentioning it.The premotivic category is Z -graded where the first index refers to the simplicial sphere andthe second one to the Tate twist. According to our general convention, a cohomology theoryrepresentable in SH is Z -graded accordingly: given such a ring spectrum E , for any smooth S -scheme X , and any integer ( i, n ) ∈ Z , we get a bigraded ring: E n,i ( X ) = Hom SH( S ) (cid:0) Σ ∞ X + , E ( i )[ n ] (cid:1) . When X is a pointed smooth S -scheme, it defines a pointed sheaf of sets still denoted by X andwe denote by ˜ E n,i ( X ) for the corresponding cohomology ring.The coefficient ring associated with E is the cohomology of the base E ∗∗ := E ∗∗ ( S ). The ring E ∗∗ ( X ) (resp. ˜ E ∗∗ ( X )) is in fact an E ∗∗ -algebra.12.1.1 . We say E is a strict ring spectrum if there exists a monoid in the category of symmetricTate spectra E ′ and an isomorphism of ring spectra E ≃ E ′ in SH( S ). In this case, a module M over the monoid E in the monoidal category SH( S ) will be said to be strict if there exists an E ′ -module M ′ in the category of symmetric Tate spectra, as well as an isomorphism of E -modules M ≃ M ′ in SH( S ). . Consider the infinite tower P S → P S → · · · → P nS → · · · of schemes pointed by the infinity. We denote by P ∞ S the limit of this tower as a pointed Nisnevichsheaf of sets and by ι : P S → P ∞ S the induced inclusion. Recall the following definition, classicallytranslated from topology: Definition . Let E be a ring spectrum over S . An orientation of E is a cohomologyclass c in ˜ E , ( P ∞ S ) such that ι ∗ ( c ) is sent to the unit of the coefficient ring of E by the canonicalisomorphism ˜ E , ( P S ) = E , .We then say that ( E, c ) is an oriented ring spectrum . We shall say also that E is orientable ifthere exists an orientation c .According to [ MV99 , 1.16 and 3.7], we get a canonical map for any smooth S -scheme X Pic( X ) = H ( X, G m ) → Hom H • ( S ) ( X + , P ∞ ) → Hom
SH( S ) (Σ ∞ X + , Σ ∞ P ∞ )(the first map is an isomorphism whenever S is regular (or even geometrically unibranch)). Giventhis map, an orientation c of a ring spectrum E defines a map of sets c ,X : Pic( X ) → E , ( X )which is natural in X (and from its construction in [ MV99 ], one can check that c = c , P ∞ S ( O (1))).Usually, we put c = c ,X . Example . (1) The original example of an oriented ring spectrum is the algebraiccobordism spectrum MGL introduced by Voevodsky ( cf. [ Voe98 ]).(2) According to Definition 11.3.2, the motivic cohomology ring spectrum H Λ M ,S defined in11.2.17 is an oriented ring spectrum.
2. STABLE HOMOTOPY THEORY OF SCHEMES 231 (3) Consider a triangulated premotivic category T which satisfies the weak localizationproperty (wLoc) and such that there exists an adjunction of triangulated premotiviccategories: ϕ ∗ : SH ⇆ T : ϕ ∗ . Recall that ϕ ∗ is symmetric monoidal. Thus, its right adjoint is weakly symmetricmonoidal and for any the spectrumH T ,S := ϕ ∗ ( S )admits a (commutative) ring structure.Then T is oriented in the sense of Definition 2.4.38 if and only if the ring spectrumH T ,S is oriented in the sense of Definition 12.2.2 – see Example 2.4.40. Moreover, anorientation of T is equivalent to the data of orientations H T ,S for any scheme S whichare stable by pullbacks (on cohomology). Remark . When E is a strict ring spectrum, the category E - mod satisfies the axiomsof [ D´eg08 , 2.1] (see example 2.12 of loc.cit. ).Recall the following result, which first appeared in [
Vez01 ]: Proposition . Let ( E, c ) be an oriented ring spectrum.Then: E ∗∗ ( P ∞ S ) = E ∗∗ [[ c ]] E ∗∗ ( P ∞ S × P ∞ S ) = E ∗∗ [[ x, y ]] where x (resp. y ) is the pullback of c along the first (resp. second) projection. Remark . When E is a strict ring spectrum, this is [ D´eg08 , 3.2] according to remark12.2.4. The proof follows an argument of Morel ([
D´eg08 , lemma 3.3]) and the arguments of op.cit. , p. 634, can be easily used to obtain the proposition arguing directly for the cohomologyfunctor X E ∗ , ∗ ( X ).12.2.7 . Recall that the Segre embeddings P nS × P mS → P n + m + nmS define a map σ : P ∞ S × P ∞ S → P ∞ S . It gives the structure of an H -group to P ∞ S in the homotopy category H ( S ). Consider thehypothesis of the previous proposition. Then the pullack along σ in E -cohomology induces a map E ∗∗ [[ c ]] σ ∗ −→ E ∗∗ [[ x, y ]]and following Quillen, we check that the formal power series σ ∗ ( c ) defines a formal group law overthe ring E ∗∗ . Definition . Let (
E, c ) be an oriented ring spectrum and consider the previous nota-tion.The formal group law F E ( x, y ) := σ ∗ ( c ) will be called the formal group law associated to( E, c ).Recall the formal group law has the form: F E ( x, y ) = x + y + X i + j> a ij .x i y j with a ij = a ji in E − i − j, − i − j .The coefficients a ij describe the failure of additivity of the first Chern class c . Indeed, assumingthe previous definition, we get the following result: Proposition . Let X be a smooth S -scheme.(1) For any line bundle L/X , the class c ( L ) is nilpotent in E ∗∗ ( X ) . (2) Suppose X admits an ample line bundle. For any line bundles L, L ′ over X , c ( L ⊗ L ) = F E ( c ( L ) , c ( L )) ∈ E , ( X ) . We refer to [
D´eg08 , 3.8] in the case where E is strict; the proof is the same in the generalcase.Recall the following theorem of Vezzosi ( cf. [ Vez01 , 4.3]):
Theorem . Let ( E, c ) be an oriented spectra over S , with formal group law F E . Then there exists a bijection between the following sets:(i) Orientation classes c ′ of E .(ii) Morphisms of ring spectra MGL → E in SH( S ) .(iii) Couples ( F, ϕ ) where F is a formal group law over E ∗∗ and ϕ is a power series over E ∗∗ which defines an isomorphism of formal group law: ϕ is invertible as a power series and F E ( ϕ ( x ) , ϕ ( y )) = F ( x, y ) . In what follows, we shall use frequently the equivalence of pre-motivic categories (see 5.3.35) SH Q ⇄ D A , Q , and shall identify freely any rational spectrum over a scheme S with an object of D A ( S, Q ).
13. Algebraic K-theory13.1. The K-theory spectrum.
We consider the spectrum
KGL S which represents homo-topy invariant K-theory in SH( S ) according to Voevodsky (see [ Cis13 ], [
Voe98 , 6.2], [
Rio10 ,5.2] and [
PPR09 ]). It is characterized by the following properties:(K1) For any morphism of schemes f : T → S , there is an isomorphism f ∗ KGL S ≃ KGL T inSH( T ).(K2) For any regular scheme S and any integer n , there exists an isomorphismHom SH( S ) ( S [ n ] , KGL S ) → K n ( S )(where the right hand side is Quillen algebraic K-theory) such that, for any morphism f : T → S of regular schemes, the following diagram is commutative:Hom ( S [ n ] , KGL S ) / / (cid:15) (cid:15) Hom ( f ∗ S [ n ] , f ∗ KGL S ) Hom ( T [ n ] , KGL T ) (cid:15) (cid:15) K n ( S ) f ∗ / / K n ( T )(where the lower horizontal map is the pullback in Quillen algebraic K-theory along themorphism f and the upper horizontal map is obtained by using the functor f ∗ : SH( S ) → SH( T ) and the identification (K1)).(K3) For any scheme S , there exists a unique structure of a commutative monoid on KGL S which is compatible with base change – using the identification (K1) – and induces thecanonical ring structure on K ( S ).Thus, according to (K1) and (K3), the collection of the ring spectrum KGL S for any scheme S form an absolute ring spectrum. As usual, when no confusion can arise, we will not indicate thebase in the notation KGL .Note that (K1) implies formally that the isomorphism of (K2) can be extended for any smooth S -scheme X (with S regular), giving a natural isomorphism:Hom SH( S ) (Σ ∞ X + [ n ] , KGL ) → K n ( X ) .
3. ALGEBRAIC K-THEORY 233 . Recall from the construction the following property of the spectrum
KGL :(K4) the spectrum
KGL is a P -periodic spectrum in the sense that there exists a canonicalisomorphism KGL ∼ −→ R Hom (cid:0) Σ ∞ P S , KGL (cid:1) = KGL ( − − . As usual, P S is pointed by the infinite point.This isomorphism, classically called the Bott isomorphism, is characterized uniquely by the factthat its adjoint isomorphism (obtained by tensoring with S (1)[2]) is equal to the composite(13.2.1.1) γ u : KGL (1)[2] ⊗ u −−−→ KGL ∧ KGL µ −→ KGL . where u : Σ ∞ P → KGL corresponds to the class [ O (1)] − K ( P ) through the isomorphism(K2) and µ is the structural map of monoid from (K3).Using the isomorphism of (K4), the property (K1) can be extended as follows: for any smooth S -scheme X and any integers ( i, n ) ∈ Z , there is a canonical isomorphism:(13.2.1.2) KGL n,i ( X ) ∼ −−→ K i − n ( X ) . Remark . The element u is invertible in the ring KGL ∗ , ∗ ( S ). Its inverse is the Bottelement β ∈ KGL , ( S ). If we chose as an orientation of the ring spectrum KGL ( cf. β. ([ O (1)] − ∈ KGL , ( P ∞ ) , the corresponding formal group law is the multiplicative formal group law: F ( x, y ) = x + y + β − .xy. Theorem . The spectrum KGL can be represented canon-ically by a cartesian monoid KGL ′ , as well as by a homotopy cartesian commutative monoid KGL β in the fibred model category of symmetric P -spectra, in such a way that there exists a morphismof monoids KGL ′ → KGL β which is a termwise stable A -equivalence. Proof.
For any noetherian scheme of finite dimension S , one has a strict commutative ringspectrum KGL βS which is canonically isomorphic to KGL S in SH( S ) as ring spectra; see [ RSØ10 ].One can check that the objects
KGL βS do form a commutative monoid over the diagram of allnoetherian schemes of finite dimension (i.e. a commutative monoid in the category of sections of thefibred category of P -spectra over the category of noetherian schemes of finite dimension), eitherby hand, by following the explicit construction of loc. cit. , either by modifying its constructionvery slightly as follows: one can perform mutatis mutandis the construction of loc. cit. in the P -stabilization of the A -localization of the model category of Nisnevich simplicial sheaves over(any essentially small adequate subcategory of) the category of all noetherian schemes of finitedimension, and get an object KGL β , whose restriction to each of the categories Sm /S is theobject KGL βS . From this point of view, we clearly have canonical maps f ∗ ( KGL βS ) → KGL βT forany morphism of schemes f : T → S . The object KGL β is homotopy cartesian, as the composedmap L f ∗ ( KGL S ) ≃ L f ∗ ( KGL βS ) → f ∗ ( KGL βS ) → KGL βT ≃ KGL T is an isomorphism in SH( T ). Consider now a cofibrant resolution KGL ′ Spec( Z ) → KGL β Spec( Z ) in the model category of monoids of the category of symmetric P -spectra over Spec ( Z ); seeTheorem 7.1.3. Then, we define, for each noetherian scheme of finite dimension S , the P -spectrum KGL ′ S as the pullback of KGL ′ Spec( Z ) along the map f : S → Spec ( Z ). As the functor f ∗ is a left Quillen functor, the object KGL ′ S is cofibrant (both as a monoid and as a P -spectrum),so that we get, by construction, a termwise cofibrant cartesian strict P -ring spectrum KGL ′ , aswell as a morphism KGL ′ → KGL β which is a termwise stable A -equivalence. (cid:3) . For each noetherian scheme of finite dimension S , one can consider the model categoriesof modules over KGL ′ S and KGL βS respectively; see 7.2.2. The change of scalars functor along thestable A -equivalence KGL ′ S → KGL βS defines a left Quillen equivalence, whence an equivalenceof homotopy categories: Ho( KGL ′ S - mod) ≃ Ho(
KGL βS - mod) . We put Ho(
KGL - mod)( S ) = Ho( KGL βS - mod) , and call this category the homotopy category of KGL-modules over S . By definition, for anysmooth S -scheme X , we have a canonical isomorphismHom SH( S ) (Σ ∞ ( X + ) , KGL [ n ]) ≃ Hom
KGL ( KGL S ( X ) , KGL [ n ])(where KGL S ( X ) = KGL S ∧ L S Σ ∞ ( X + ), while Hom KGL stands for Hom
Ho(
KGL - mod)( S ) ).According to (K1) and (K3), for any regular scheme X , we thus get a canonical isomorphism:(13.3.2.1) ǫ S : Hom KGL ( KGL S [ n ] , KGL S ) ∼ −−→ K n ( S ) . Using Bott periodicity (K4), and the compatibility with base change, this isomorphism can beextended for any smooth S -scheme X and any pair ( n, m ) ∈ Z :(13.3.2.2) ǫ X/S : Hom
KGL ( KGL S ( X ) , KGL S ( m )[ n ]) ∼ −−→ K m − n ( X ) . Corollary . The categories
Ho(
KGL- mod)( S ) form a motivic category, and the func-tors SH( S ) → Ho(
KGL- mod)( S ) , M KGL S ∧ L S M define a morphism of motivic categories SH → Ho(
KGL- mod) over the category of noetherian schemes of finite dimension.
Proof.
This follows from the preceding theorem and from 7.2.13 and 7.2.18. (cid:3) . Consider a closed immersion i : Z → S with complementary open immersion j : U → S . Assume S is regular.We use the definition of [ Gil81 , 2.13] for the K-theory of S with support in Z denoted by K Z ∗ ( S ).In other words, we define K Z ( S ) as the homotopy fiber of the restriction map R Γ( S, KGL S ) = K ( S ) → K ( U ) = R Γ( U, KGL U ) , and put: K Zn ( S ) = π n ( K Z ( S )).Applying the derived global section functor R Γ( S, − ) to the homotopy fiber sequence(13.4.1.1) i ! i ! KGL S → KGL S → j ∗ j ∗ KGL S , we get a homotopy fiber sequence(13.4.1.2) R Γ( S, i ! i ! KGL S ) → R Γ( S, KGL S ) → R Γ( U, KGL S )from which we deduce an isomorphism in the stable homotopy category of S -spectra:(13.4.1.3) R Γ( Z, i ! KGL S ) = R Γ( S, i ! i ! KGL S ) ≃ K Z ( S ) . We thus get the following property:(K6) There is a canonical isomorphismHom
SH( S ) (cid:0) S [ n ] , i ! i ! KGL S (cid:1) → K Zn ( S )which satisfies the following compatiblities:
3. ALGEBRAIC K-THEORY 235 (K6a) the following diagram is commutative:Hom( [ n + 1] , j ∗ j ∗ KGL S ) / / (cid:15) (cid:15) Hom (cid:0) [ n ] , i ! i ! KGL S (cid:1) / / (cid:15) (cid:15) Hom( [ n ] , KGL S ) (cid:15) (cid:15) K n +1 ( U ) / / K Zn ( S ) / / K n ( S )where the upper horizontal arrows are induced by the localization sequence (13.4.1.1),and the lower one is the canonical sequence of K-theory with support. The extremeleft and right vertical maps are the isomorphisms of (K2);(K6b) for any morphism f : Y → S of regular schemes, k : T → Y the pullback of i along f , the following diagram is commutative:Hom (cid:0) [ n ] , i ! i ! KGL S (cid:1) / / (cid:15) (cid:15) Hom (cid:0) f ∗ [ n ] , f ∗ i ! i ! KGL S (cid:1) / / Hom (cid:0) [ n ] , k ! k ! KGL Y (cid:1) (cid:15) (cid:15) K Zn ( S ) f ∗ / / K Tn ( Y )where the lower horizontal map is given by the functoriality of relative K-theory (in-duced by the funtoriality of K-theory) and the left one is obtained using the functor f ∗ of SH, the canonical exchange morphism f ∗ i ! i ! → k ! k ! f ∗ and the identification(K1).This property can be extended to the motivic category Ho( KGL - mod) and we get a canonicalisomorphism(13.4.1.4) ǫ i : Hom KGL ( KGL S [ n ] , i ! i ! KGL S ) ∼ −−→ K Zn ( S )satisfying the analog of (K6a) and (K6b). . Consider a cartesian square of regular schemes Z ′ k / / g (cid:15) (cid:15) S ′ f (cid:15) (cid:15) Z i / / S with i a closed immersion. We will say that this square is Tor-independant if Z and S ′ areTor-independant over S in the sense of [ SGA6 , III, 1.5]: for any i >
0, Tor Si ( O Z , O S ′ ) = 0. In this case, when we assume in addition that all the schemes in the previous square areregular and that i is a closed immersion we get from [ TT90 , 3.18] the formula f ∗ i ∗ = k ∗ g ∗ : K ∗ ( Z ) → K ∗ ( S ′ )in Quillen K-theory. An important point for us is that there exists a canonical homotopy betweenthese morphisms at the level of the Waldhausen spectra. According to the localization theoremof Quillen [
Qui73 , 3.1], we get:
Theorem . For any closed immersion i : Z → S between regular schemes,there exists a canonical isomorphism q i : K Zn ( S ) → K n ( Z ) . Moreover, this isomorphism is functorial with respect to the Tor-independant squares as above,with i a closed immersion and all the schemes regular. For example, when i is a regular closed immersion of codimension 1, this happens if and only if the abovesquare is transversal. When all the schemes in the square admit ample line bundles, we can refer to [
Qui73 , 2.11]. In the proof of Quillen, one can also trace back a canonical homotopy with the restriction mentioned in thepreceding footnote. Remark . In the condition of this theorem, the following diagram is commutative byconstruction: K Zn ( S ) + + ❲❲❲❲❲ q i (cid:15) (cid:15) K n ( S ) K n ( Z ) i ∗ ❣❣❣❣❣❣ where the non labelled map is the canonical one. Definition . Let i : Z → S be a closed immersion between regular schemes.We define the fundamental class associated with i as the morphism of KGL -modules: η i : i ∗ KGL Z → KGL S defined by the image of the unit element 1 through the following morphism: K ( Z ) q − i −−→ K Z ( S ) ǫ − i −−→ Hom
KGL ( KGL S , i ! i ! KGL S ) = Hom KGL ( i ∗ KGL Z , KGL S ) . We also denote by η ′ i : KGL Z → i ! KGL S the morphism obtained by adjunction. Remark . The fundamental class has the following functoriality properties.(1) By definition, and applying remark 13.5.3, the composite map
KGL S → i ∗ i ∗ ( KGL S ) = i ∗ KGL
Z η i −→ KGL S corresponds via the isomorphism ǫ S to i ∗ (1) ∈ K ( S ). According to [ SGA6 , Exp. VII,2.7], this class is equal to λ − ( N i ) where N i is the conormal sheaf of the regular immersion i .(2) In the situation of a Tor-independant square as in 13.5.1, remark that f ∗ η i = η k throughthe canonical exchange isomorphism f ∗ i ∗ = k ∗ g ∗ — apply the functoriality of ǫ i from(K6b) and the one of q i .(3) Using the identification i ! i ∗ = 1, we get η ′ i = i ! η i . Consider a cartesian square as in 13.5.1and assume f is smooth. Then the square is Tor-independant and we get g ∗ η ′ i = η ′ k usingthe exchange isomorphism g ∗ i ! = k ! f ∗ . Proposition . For any closed immersion i : Z → S between regular schemes, thefollowing diagram is commutative: Hom
KGL ( KGL Z [ n ] , KGL Z ) η ′ i / / ǫ Z (cid:15) (cid:15) ( ∗ ) Hom
KGL ( KGL Z [ n ] , i ! KGL S ) ǫ i (cid:15) (cid:15) K n ( Z ) q − i / / K Zn ( S ) Proof.
In this proof, we denote by [ − , − ] the bifunctor Hom KGL ( − , − ). Step
1: We assume that i : Z → S admits a retraction p : S → Z .Consider a KGL -linear map α : KGL Z [ n ] → KGL Z . Then, η ′ i ( α ) corresponds by adjunction to thecomposition i ∗ KGL Z [ n ] i ∗ ( α ) −−−→ i ∗ KGL
Z η i −→ KGL S . Applying the projection formula for the motivic category Ho(
KGL - mod), we get: i ∗ ( α ) = i ∗ (1 ⊗ i ∗ p ∗ ( α )) = i ∗ (1) ⊗ p ∗ ( α ) . Here 1 stands for the identity morphism of the
KGL -module
KGL Z . This shows that η ′ i ( α )corresponds by adjunction to the composite map: η i ⊗ p ∗ ( α ) : i ∗ KGL Z [ n ] = i ∗ KGL Z [ n ] ⊗ KGL S → KGL S ⊗ KGL S = KGL S (the tensor product is the KGL -linear one). By assumption, i ∗ : K ∗ ( Z ) → K ∗ ( S ) admits aretraction which implies the canonical map O i : K Z ∗ ( S ) → K ∗ ( S ) admits a retraction ( cf. remark13.5.3). To check that the diagram ( ∗ ) is commutative, we can thus compose with O i .
3. ALGEBRAIC K-THEORY 237
Recall the first point of remark 13.5.5: applying property (K6a) and the fact the isomorphism ǫ S : [ KGL S [ n ] , KGL S ] → K n ( S ) is compatible with the algebra structures, we are finally reducedto prove that i ∗ ( α ) = i ∗ (1) .p ∗ ( α ) ∈ K n ( S ) . This follows from the projection formula in K-theory (see [
Qui73 , 2.10] and [
TT90 , 3.17]).
Step
2: We shall reduce the general case to Step 1. We consider the following deformation to thenormal cone diagram: let D be the blow-up of A S in the closed subscheme { } × Z , P be theprojective completion of the normal bundle of Z in S and s be the canonical section of P/Z ; weget the following diagram of regular schemes: Z s / / i (cid:15) (cid:15) A Z (cid:15) (cid:15) Z s o o s (cid:15) (cid:15) S / / D P o o (13.6.1.1)where s (resp. s ) is the zero (resp. unit) section of A Z over Z . These squares are cartesianand Tor-independant in the sense of 13.5.1. The maps s and s induce isomorphisms in K-theorybecause Z is regular. Thus, the second point of remark 13.5.5 allows to reduce to the case of theimmersion s which was done in Step 1. (cid:3) . Consider a cartesian square T k / / g (cid:15) (cid:15) X f (cid:15) (cid:15) Z i / / S such that S and Z are regular, i is a closed immersion and f is smooth. In this case, the followingdiagram is commutativeHom KGL ( KGL Z ( T )[ n ] , KGL Z ) η ′ i / / Hom
KGL ( KGL Z ( T )[ n ] , i ! KGL S )Hom KGL ( KGL T [ n ] , KGL T ) η ′ k / / Hom
KGL ( KGL T [ n ] , k ! KGL X )using the adjunction ( g ♯ , g ∗ ), the exchange isomorphism g ∗ i ! ≃ k ! f ∗ (which uses relative purity forsmooth morphisms) and the third point of remark 13.5.5. In particular, the preceding propositionhas the following consequences: Theorem . For any closed immersion i : Z → S between regularschemes, the map η ′ i : KGL Z → i ! KGL S is an isomorphism in the category Ho(
KGL- mod)( Z ) (or in SH( Z ) ). Corollary . Given a cartesian square as above, for any pair ( n, m ) ∈ Z , the followingdiagram is commutative: Hom(
KGL S ( X ) , i ∗ KGL Z ( m )[ n ]) η i / / Hom(
KGL S ( X ) , KGL S ( m )[ n ]) ǫ X/S ∼ (cid:15) (cid:15) Hom(
KGL Z ( T ) , KGL Z ( m )[ n ]) ǫ T/Z ∼ (cid:15) (cid:15) K m − n ( T ) k ∗ / / K m − n ( X ) where the vertical maps are the isomorphisms (13.3.2.2) . . Let S be a regular scheme. Let Y be a smooth S -scheme. The canonical mapPic( Y ) → K ( Y ) ∼ −→ Hom
KGL ( KGL S ( Y ) , KGL S ) β ∗ −→ Hom
KGL ( KGL S ( Y ) , KGL S (1)[2])defines Chern classes in the category Ho( KGL - mod)( S ); they corresponds to the orientation de-fined in remark 13.2.2.Let p : P → S be a projective bundle of rank n . Let v = [ O (1)] − K ( P ). It corresponds toa map v : KGL S ( P ) → KGL S . According to [ D´eg08 , 3.2] and our choice of Chern classes, thefollowing map
KGL S ( P ) P i β i . v i ⊠ p ∗ −−−−−−−−→ M ≤ i ≤ n KGL S ( i )[2 i ]is an isomorphism. As β is invertible, it follows that the map(13.7.1.1) ϕ P/S : KGL S ( P ) P i v i ⊠ p ∗ −−−−−−→ M ≤ i ≤ n KGL S is an isomorphism as well. Using this formula, the map Hom( ϕ P/S , KGL S ) is equal to the isomor-phism of Quillen’s projective bundle theorem in K-theory ( cf. [ Qui73 , 4.3]): f P/S : n M i =0 K ∗ ( S ) → K ∗ ( P ) , ( S , ..., S n ) X i p ∗ ( S i ) .v i . Let p ∗ : K ∗ ( P ) → K ∗ ( S ) be the pushout by the projective morphism p . According to theprojection formula, it is K ∗ ( S )-linear. In particular, it is determined by the n + 1-uple ( a , ..., a n )where a i = p ∗ ( v i ) ∈ K ( S ) through the isomorphism f P/S . Let a i : KGL S → KGL S be the mapcorresponding to a i . Definition . Consider the previous notations. We define the trace map associated withthe projection p : P → S as the morphism of KGL -modulesTr
KGL p : p ∗ ( KGL P ) = R Hom ( KGL S ( P ) , KGL S ) ( ϕ ∗ P/S ) − −−−−−−→ n M i =0 KGL S ( a ,..., a n ) −−−−−−→ KGL S . From this definition, it follows that Tr p represents the pushout by p in K-theory:Hom KGL ( KGL S [ n ] , p ∗ KGL P ) Tr KGL p ∗ / / Hom
KGL ( KGL S [ n ] , KGL S ) ǫ S (cid:15) (cid:15) Hom
KGL ( KGL P [ n ] , KGL P ) ǫ P (cid:15) (cid:15) K n ( P ) p ∗ / / K n ( S )Consider moreover a cartesian square: Q q / / g (cid:15) (cid:15) P p (cid:15) (cid:15) Y f / / S such that f is smooth. From the projective base change theorem, we get f ∗ p ∗ p ∗ = q ∗ q ∗ g ∗ . Usingthis identification, we easily obtain that f ∗ Tr KGL p = Tr KGL q . Thus, we conclude that the mapHom KGL ( KGL S ( Y )[ n ] , p ∗ KGL P ) Tr KGL p −−−−→ Hom
KGL ( KGL S ( Y )[ n ] , KGL S )represents the usual pushout map q ∗ : K n ( Q ) → K n ( Y )through the canonical isomorphisms (13.3.2.2).
3. ALGEBRAIC K-THEORY 239 . Consider a projective morphism f : T → S between regular schemes and choose afactorization T i −→ P p −→ S where i is a closed immersion and p is the projection of a projective bundle. Let us define amorphism Tr KGL ( p,i ) : f ∗ KGL T = p ∗ i ∗ KGL
T p ∗ η i −−−→ p ∗ KGL P Tr KGL p −−−−→ KGL S . According to 13.6.4 and the previous paragraph, for any cartesian square Y g / / b (cid:15) (cid:15) X a (cid:15) (cid:15) T f / / S such that a is smooth, the following diagram is commutative.Hom KGL ( KGL S ( X ) , f ∗ KGL T ( m )[ n ]) Tr KGL ( p,i ) ∗ / / Hom
KGL ( KGL S ( X ) , KGL S ( m )[ n ]) ǫ X/S ≃ (cid:15) (cid:15) Hom
KGL ( KGL T ( Y ) , KGL Z ( m )[ n ]) ǫ Y/T ≃ (cid:15) (cid:15) K m − n ( Y ) g ∗ / / K m − n ( X )(13.7.3.1) Definition . Considering the above notations, we define the trace map associated to f as the morphism Tr KGL f = Tr KGL ( p,i ) : f ∗ f ∗ KGL S → KGL S . Remark . By definition, the trace map Tr
KGL f is a morphism of KGL -modules. As aconsequence, the map obtained by adjuntion η ′ f : KGL T ≃ f ∗ KGL S → f ! KGL S is also a morphism of KGL -module. This implies that the morphism η ′ f (and thus also Tr KGL f ) iscompletely determined by the element η ′ f ∈ Hom
KGL ( KGL T , f ! KGL S ) ≃ Hom
SH( T ) ( T , f ! KGL S ) . Moreover, as p is smooth, there is a canonical isomorphism p ! KGL S ≃ KGL P (by relative purityfor p and by periodicity; see [ Rio10 , lemma 6.1.3.3]). From there, we deduce from Theorem 13.6.3that we have a canonical isomorphism f ! KGL S ≃ i ! KGL P ≃ KGL T . This implies that we have an isomorphism:Hom
SH( T ) ( T , f ! KGL S ) ≃ K ( T ) . Hence the map η ′ f is completely determined by a class in K ( T ). The problem of the functoriality oftrace maps in the motivic category Ho( KGL - mod) is thus a matter of functoriality of this element η ′ f in K , which can be translated faithfully to the problem of the functoriality of pushforwardsfor K .However, the only property of trace maps we shall use here is the following. Proposition . Let f : T → S be a finite flat morphism of regular schemes such thatthe O S -module f ∗ O T is (globally) free of rank d . Then the following composite mapKGL S → f ∗ f ∗ KGL S Tr KGL f −−−−→ KGL S is equal to d. KGL S in Ho(
KGL- mod)( S ) (whence in SH( S ) ). Proof.
Let ϕ be the composite map of Ho( KGL - mod)( S ) KGL S → f ∗ f ∗ KGL S Tr f −−→ KGL S . As ϕ is KGL S -linear by construction, it corresponds to an element ϕ ∈ Hom
KGL ( KGL S , KGL S ) ≃ Hom
SH( S ) ( S , KGL S ) ≃ K ( S ) . According to the commutative diagram (13.7.3.1), if we apply the functor Hom
SH( S ) ( S , − ) to ϕ ,we obtain through the evident canonical isomorphisms the composition of the usual pullback andpushforward by f in K-theory: K ( S ) f ∗ −→ K ( T ) f ∗ −→ K ( S ) . With these notations, the element of K ( S ) corresponding to ϕ is the pushforward of 1 T = f ∗ (1 S ) by f , while the element corresponding to the identity of KGL S is of course 1 S . Under ourassumptions on f , it is obvious that we have the identity f ∗ (1 T ) = d. S ∈ K ( S ). This meansthat ϕ is d times the identity of KGL S . (cid:3)
14. Beilinson motives14.1. The γ -filtration. . We denote by
KGL Q the Q -localization of the absolute ring spectrum KGL , consideredas a cartesian section of D A , Q . From [ Rio10 , 5.3.10], this spectrum has the following property:(K5) For any scheme S , there exists a canonical decomposition, called the Adams decomposi-tion KGL Q ,S ≃ M i ∈ Z KGL ( i ) S compatible with base change and such that for any regular scheme S , the isomorphismof (K2) induces an isomorphism:Hom D A ( S, Q ) (cid:16) Q S ( X )[ n ] , KGL ( i ) S (cid:17) ≃ K ( i ) n ( S ) := Gr iγ K n ( S ) Q where the right hand side is the i -th graded piece of the γ -filtration on K-theory groups.We will denote by π i : KGL Q ,S → KGL ( i ) S , resp. ι i : KGL ( i ) S → KGL Q ,S the projection (resp. inclusion) defined by the decomposition (K3) and we put p i = ι i π i for thecorresponding projector on KGL Q ,S . Definition . We define the
Beilinson motivic cohomology spectrum as the ra-tional Tate spectrum H B ,S = KGL (0) S . Remark . Note that, by definition, for any morphism of schemes f : T → S , we have f ∗ H B ,S ≃ H B ,T . Lemma . The isomorphism γ u of (13.2.1.1) is homogenous of degree +1 with respectto the graduation (K5). In other words, for any integer i ∈ Z , the following composite map is anisomorphism KGL ( i ) (1)[2] ι i −→ KGL Q (1)[2] γ u −→ KGL Q π i −→ KGL ( i +1) . For any integer i ∈ Z , we thus get a canonical isomorphism(14.1.4.1) H B ( i )[2 i ] ∼ −→ KGL ( i ) .
4. BEILINSON MOTIVES 241
Proof.
It is sufficient to check that, for j = i + 1, ( p j ◦ γ u ◦ p i = 0 ,p j ◦ γ − u ◦ p i = 0in Hom D A ( S, Q ) ( KGL Q , KGL Q ). But according to [ Rio10 , 5.3.1 and 5.3.6], we have only tocheck these equalities for the induced endomorphism of K (seen as a presheaf on the category ofsmooth schemes over Spec ( Z )). This follows then from the compatibility of the projective bundleisomorphism with the γ -filtration; see [ SGA6 , Exp. VI, 5.6]. (cid:3) . Recall from [
NSØ09 ] that
KGL Q is canonically isomorphic (with respect to the ori-entation 13.2.2) to the universal oriented rational ring spectrum with multiplicative formal grouplaw introduced in [ NSØ09 ]. The isomorphism of the preceding corollary shows in particular that H B is obtained from KGL Q by killing the elements β n for n = 0. In particular, this shows that H B is canonically isomorphic to the spectrum denoted by L Q in [ NSØ09 ], which corresponds tothe universal additive formal group law over Q . This implies that H B has a natural structure ofa (commutative) ring spectrum. Proposition . The multiplication map µ : H B ⊗ H B → H B is an isomorphism. This trivially implies that the following map is an isomorphism:(14.1.6.1) 1 ⊗ η : H B → H B ⊗ H B . Proof.
It is enough to treat the case S = Spec ( Z ). We will proove that the projector ψ : H B ⊗ H B µ −→ H B ⊗ η −−→ H B ⊗ H B is an isomorphism (in which case it is in fact the identity). We do that for the isomorphic ringspectrum L Q .Let H top Q be the topological spectrum representing rational singular cohomology. In theterminology of [ NSØ09 ], L Q is a Tate spectrum representing the Landweber exact cohomologywhich corresponds to the Adams graded M U ∗ -algebra Q obtained by killing every generators ofthe Lazard ring M U ∗ . The corresponding topological spectrum is of course H top Q .According to [ NSØ09 , 9.2], the spectrum E = L Q ⊗ L Q is a Landweber exact spectrum cor-responding to the M U ∗ -algebra Q ⊗ MU ∗ Q = Q . In particular, the corresponding topologicalspectrum is simply H top Q . Thus, according to [ NSØ09 , 9.7], applied with F = E = L Q ⊗ L Q ,we get an isomorphism of Q -vector spacesEnd( L Q ⊗ L Q ) = Hom Q ( Q , E ∗∗ ) = Q . Thus ψ = λ.Id for λ ∈ Q . But λ = 0 is excluded because ψ is a projector on a non trivial factor,so that we can conclude. (cid:3) Definition . Let S be any scheme.We say that an object E of D A ( S, Q ) is H B -acyclic if H B ⊗ E = 0 in D A ( S, Q ). A morphismof D A ( S, Q ) is an H B -equivalence if its cone is H B -acyclic (or, equivalently, if its tensor productwith H B is an isomorphism).An object M of D A ( S, Q ) is H B -local if, for any H B -acyclic object E , the group Hom( E, M )vanishes.We denote by DM B ( S ) the Verdier quotient of D A ( S, Q ) by the localizing subcategory madeof H B -acyclic objects (i.e. the localization of D A ( S, Q ) by the class of H B -equivalences).The objects of DM B ( S ) are called the Beilinson motives . Proposition . An object E of D A ( S, Q ) is H B -acyclic if and only if we have KGL Q ⊗ E = 0 . Proof.
This follows immediately from property (K5) (see 14.1.1) and Lemma 14.1.4. (cid:3)
Proposition . The localization functor D A ( S, Q ) → DM B ( S ) admits a fully faithfulright adjoint whose essential image in D A ( S, Q ) is the full subcategory spanned by H B -local ob-jects. More precisely, there is a left Bousfield localization of the stable model category of symmetricTate spectra Sp( S, Q ) by a small set of maps whose homotopy category is precisely DM B ( S ) . Proof.
For each smooth S -scheme X and any integers n, i ∈ Z , we have a functor with valuesin the category of Q -vector spaces F X,n,i = Hom D A ( S, Q ) (Σ ∞ Q S ( X ) , H B ⊗ ( − )( i )[ n ]) : Sp( S, Q ) → Q - modwhich preserves filtered colimits. We define the class of H B -weak equivalences as the class of mapsof Sp( S, Q ) whose image by F X,n,i is an isomorphism for all X and n , i . By virtue of [ Bek00 ,Prop. 1.15 and 1.18], we can apply Smith’s theorem [
Bek00 , Theorem 1.7] (with the class ofcofibrations of Sp( S, Q )), which implies the proposition. (cid:3) Remark . We shall often make the abuse of considering DM B ( S ) as a full subcategoryin D A , Q ( S ), with an implicit reference to the preceding proposition.Note that H B -acyclic objects are stable by the operations f ∗ , f ♯ and ⊗ , so that applyingCorollary 5.2.5, we obtain a premotivic category DM B together with a premotivic adjunction:(14.2.4.1) β ∗ : D A , Q ⇄ DM B : β ∗ . Proposition . The spectrum H B ,S is H B -local and the unit map η H B : → H B ,S is an H B -equivalence in D A ( S, Q ) . Proof.
The unit map η : S → H B ,S is an H B -equivalence by 14.1.6.Consider a rational spectrum E over S such that E ⊗ H B = 0 and a map f : E → H B . Itfollows trivially from the commutative diagram E f / / ⊗ η (cid:15) (cid:15) H B ,S ⊗ η (cid:15) (cid:15) PPPPPPPPPPP PPPPPPPPPPP E ⊗ H B ,S f ⊗ / / H B ,S ⊗ H B ,S µ / / H B ,S that f = 0, which shows that H B ,S is H B -local. (cid:3) Corollary . The family of ring spectra H B ,S comes from a cofibrant cartesian com-mutative monoid (7.2.10) of the symmetric monoidal fibred model category of Tate spectra overthe category of schemes. Proof.
By virtue of Proposition 14.2.5 and of Corollary 7.1.9, there exists a cofibrant commu-tative monoid in the model category of symmetric Tate spectra over Spec ( Z ) which is canonicallyisomorphic to H B , Z in D A (Spec ( Z ) , Q ) (as commutative ring spectrum). For a morphism ofschemes f : S → Spec ( Z ), we can then define H B ,S as the pullback of H B , Z (at the level of themodel categories); using Proposition 7.1.11, we see that this defines a cofibrant cartesian commu-tative monoid on the fibred category of spectra which is isomorphic to H B ,S as commutative ringspectra in D A ( S, Q ). (cid:3) . From now on, we shall assume that H B is given by a cofibrant cartesian commutativemonoid of the symmetric monoidal fibred model category of Tate spectra over the category ofschemes. By virtue of propositions 7.2.11 and 7.2.18), we get the motivic category Ho( H B - mod)of H B -modules, together with a commutative diagram of morphisms of premotivic categoriesD A , Q β ' ' ❖❖❖❖❖❖ H B ⊗ ( − ) / / Ho( H B - mod)DM B ϕ ❦❦❦❦❦❦❦❦ (any H B -acyclic object becomes null in the homotopy category of H B -modules by definition, sothat H B ⊗ ( − ) factors uniquely through DM B by the universal property of localization).
4. BEILINSON MOTIVES 243
Proposition . The forgetful functor U : Ho( H B - mod)( S ) → D A ( S, Q ) is fully faithful. Proof.
We have to prove that, for any H B ,S -module M , the map H B ,S ⊗ M → M is an isomorphism in D A , Q ( S ). As this is a natural transformation between exact functors whichcommute with small sums, and as D A , Q is a compactly generated triangulated category, it issufficient to check this for M = H B ,S ⊗ E , with E a (compact) object of D A , Q ( S ) (see 7.2.7). Inthis case, this follows immediately from the isomorphism (14.1.6.1). (cid:3) Theorem . The functor DM B ( S ) → Ho( H B ,S - mod) is an equivalence of triangulatedmonoidal categories. Proof.
This follows formally from the preceding proposition by definition of DM B (see forinstance [ GZ67 , Chap. I, Prop. 1.3]). (cid:3)
Remark . The preceding theorem shows that the premotivic category Ho( H B - mod)as well as the morphism D A , Q → Ho( H B - mod) are completely independent of the choice of thestrictification of the (commutative) monoid structure on H B given by Corollary 14.2.6. Corollary . The premotivic category DM B ≃ Ho( H B - mod) is a Q -linear motiviccategory. Proof.
It follows from Proposition 7.2.18 and Theorem 14.2.9 that DM B satisfies the homo-topy, stability and localization properties (because this is true for D A , Q by 6.2.2). It is also wellgenerated because it is a localization of D A , Q . Thus we can apply Remark 2.4.47 to conclude. (cid:3) Remark . One can also prove that DM B is motivic much more directly: this followsfrom the fact that D A , Q is motivic and that the six Grothendieck operations preserve H B -acyclicobjects, so that all the properties of D A , Q induce their analogs on DM B by the 2-universalproperty of localization (we leave this as an easy exercise for the reader). Definition . For a scheme X , we define its Beilinson motivic cohomology by theformula: H q B ( X, Q ( p )) = Hom DM B ( X ) ( X , X ( p )[ q ]) . In fact, according to the preceding corollary, the cohomology theory defined above is repre-sented by the ring spectrum H B . In particular, we can now justify the terminology of Beilinsonmotives: Corollary . For any regular scheme X , we have a canonical isomorphism H q B ( X, Q ( p )) ≃ Gr pγ K p − q ( X ) Q . . Recall from Paragraph 14.1.5 that H B ,S is canonically oriented for any scheme S .Moreover, these orientations are compatible with pullbacks with respect to S . This means inparticular that the motivic triangulated category DM B is oriented (see Example 12.2.3).In particular, the fibred category DM B satisfies the usual Grothendieck 6 functors formalism.We refer the reader to Theorem 2.4.50 for the precise statement.It was remarked in Paragraph 14.1.5 that H B ,S is the universal oriented ring spectrum withadditive formal group law over S . This property can be expressed by the following nice descriptionof Beilinson motives: Corollary . Let E be a rational spectrum over S . The following conditions areequivalent:(i) E is a Beilinson motive (i.e. is in the essential image of the right adjoint of the local-ization functor D A , Q → DM B );(ii) E is H B -local;(iii) the map η ⊗ E : E → H B ⊗ E is an isomorphism; (iv) E is an H B -module in D A , Q ;(v) E is admits a strict H B -module structure.If, in addition, E is a commutative ring spectrum, these conditions are equivalent to the followingones:(Ri) E is orientable;(Rii) E is an H B -algebra;(Riii) E admits a unique structure of H B -algebra;And, if E is a strict commutative ring spectrum, these conditions are equivalent to the followingconditions:(Riv) there exists a morphism of commutative monoids H B → E in the stable model categoryof Tate spectra;(Rv) there exists a unique morphism H B → E in the homotopy category of commutativemonoids of the category of Tate spectra. Proof.
The equivalence between statements (i)–(v) follows immediately from 14.2.9. If E isa ring spectrum, the equivalence with (Ri), (Rii) and R(iii) is a consequence of 12.2.10 and of thefact that MGL Q is H B -local; see [ NSØ09 , Cor. 10.6]. It remains to prove the equivalence with(Riv) and (Rv). Then, E is H B -local if and only if the map E → H B ⊗ E is an isomorphism. Butthis map can be seen as a morphism of strict commutative ring spectra (using the model structureof 7.1.8 applied to the model category of Tate spectra) whose target is clearly an H B -algebra, sothat (Riv) is equivalent to (ii). It remains to check that there is at most one strict H B -algebrastructure on E (up to homotopy), which follows from the fact that H B is the initial object inthe homotopy category of commutative monoids of the model category given by Theorem 7.1.8applied to the model structure of Proposition 14.2.3. (cid:3) Corollary . One has the following properties.(1) The ring structure on the spectrum H B is given by the following structural maps (withthe notations of 14.1.1). H B ⊗ H B ι ⊗ ι −−−→ KGL Q ⊗ KGL Q µ KGL −−−→
KGL Q π −→ H B , Q η KGL −−−→
KGL Q π −→ H B . (2) The map ı : H B → KGL Q is compatible with the monoid structures.(3) Let H B [ t, t − ] = L i ∈ Z H B ( i )[2 i ] be the free H B -algebra generated by one invertible gener-ator t of bidegree (2 , . Then the section u : Q (1)[2] → KGL Q induces an isomorphismof H B -algebras: γ ′ u : H B [ t, t − ] → KGL Q . Proof.
Property (1) follows from properties (2) and (3). Property (2) is a trivial consequenceof the previous corollary. Using the isomorphisms (14.1.4.1) of Lemma 14.1.4, we get a canonicalisomorphism H B ,S [ t, t − ] ∼ −→ M i ∈ Z KGL ( i ) . Through this isomorphism, the map γ ′ u corresponds to the Adams decomposition (i.e. to theisomorphism (K5) of 14.1.1) from which we deduce property (3). (cid:3) Remark . One deduces easily, from the preceding proposition and from 14.1.6, anotherproof of the fact that
KGL Q is a strict commutative ring spectrum.The isomorphism (3) is in fact compatible with the gradings of each term: the factor H B .t i issent to the factor KGL ( i ) . Recall also the parameter t corresponds to the unit β − in KGL ∗ , ∗ . Corollary . The Adams decomposition is compatible with the monoid structure onKGL Q : for any integer i, j, l such that l = i + j , the following map is zero.KGL ( i ) ⊗ KGL ( j ) ι i ⊗ ι j −−−−→ KGL Q ⊗ KGL Q µ −−→ KGL Q π l −−→ KGL ( l )4. BEILINSON MOTIVES 245 . Let R be a Q -algebra with structural morphism ϕ . Recall from Paragraph 5.3.36that we get an adjunction of premotivic triangulated categories: ϕ ∗ : D A , Q → D A ,R : ϕ ∗ . Moreover, for any object M and N of D A , Q ( S ), the canonical map(14.2.20.1) Hom( M, N ) ⊗ Q R → Hom( ϕ ∗ ( M ) , ϕ ∗ ( N )) . is an isomorphism provided M is compact or R is a finite Q -vector space.In particular, the ring spectrum KGL R := ϕ ∗ ( KGL Q ) represents Quillen algebraic K-theorywith coefficients in R over regular schemes. We can repeat Definition 14.2.1 with R -coefficientsand this gives the category DM B ( S, R ) of Beilinson motives with R -coefficients together with anadjuntion: ϕ ∗ : DM B → DM B ( − , R ) : ϕ ∗ . Moreover, using the canonical map (14.2.20.1) and the fact it is an isomorphism when M isa constructible Beilinson motives, we immediately extend all the properties proved so far from Q -coefficients to R -coefficients. Recall from Definition 4.3.2 we have defined the notion ofcontinuity for a triangulated premotivic category which is the homotopy category of a premotivicmodel category, such as the triangulated motivic category DM B – in this case, the notion ofcontinuity is relative to the Tate twist. Proposition . The motivic triangulated category DM B is continuous. Proof.
We consider the adjunction (14.2.4.1). According to Theorem 14.2.9, the functor β ∗ commutes with pullbacks by arbitrary morphisms. Thus the continuity property for DM B followsfrom the continuity property for D A , Q which was established in Example 6.1.13. (cid:3) We will give the main applications of continuity in the section on constructible Beilinsonmotives. Recall from 4.3.9 the following corollary of the continuity property of the motivic categoryDM B : Corollary . Let X be a scheme, and consider an X -scheme Y of finite type. Givena point x ∈ X , we denote by X hx the spectrum of the local henselian ring of X at the point x . Let a x : Y × X X hx → Y be the canonical map. Then the family of functors DM B ( Y ) → DM B ( Y × X X hx ) , E a ∗ x ( E ) is conservative. As the reader might expect, this proposition is very useful to reduce global properties ofthe motivic category DM B to local properties. This is in particular illustrated by the followingproposition. Theorem . The motivic category DM B is separated (on the category of noetherianschemes of finite dimension). Proof.
According to Proposition 2.3.9, it is sufficient to check that, for any finite surjectivemorphism f : T → S , the pullback functor f ∗ : DM B ( S ) → DM B ( T )is conservative.We argue by induction on the dimension of S .Let us first treat the case where dim( S ) = 0. Using the localization property, we can assumethat S and T are reduced ( cf. S is a disjoint sum of spectra of fields. In particular, f is not only finite surjective but also flat. Moreover, it is also globally free. It will be sufficientto prove that, for any Beilinson motive E over S , the adjunction map E → f ∗ f ∗ ( E ) is a monomorphism in DM B . Using the projection formula in DM B applied to the finite morphism f (point (5) of Theorem 2.4.50), this latter map is isomorphic to (cid:0) H B → f ∗ f ∗ ( H B ) (cid:1) ⊗ E . We are finally reduced to prove that the map H B ,S → f ∗ f ∗ H B ,S is a monomorphism in DM B (any monomorphism of a triangulated category splits). As H B ,S is a direct factor of KGL Q ,S , itis sufficient to find a retraction of the adjunction map KGL Q ,S → f ∗ f ∗ KGL Q ,S , and this follows from Proposition 13.7.6.Let us finally solve the induction process. Applying the preceding proposition, we can assumethat S is local henselian. Let s be the closed point of S and U be the open complement. Let f s (resp. f U ) be the pullback of f above s (resp. U ). Using the localization property of DM B andthe base change isomorphisms (point (4) of Theorem 2.4.50), it is sufficient to treat the case ofthe finite morphisms f U and f s . The case of f U follows by the induction hypothesis while the caseof f s follows from the case treated previously. This ends up the induction process. (cid:3) According respectively to Proposition 3.3.33 and Theorem 3.3.37, we deduce from the preced-ing proposition the following result:
Theorem . (1) The motivic category DM B satisfies ´etale descent.(2) The motivic category DM B satisfies h -descent when restricted to quasi-excellent schemes. Recall this means that for any ´etale hypercover (resp. h-hypercover of a quasi-excellentscheme) p : X → X and for any Beilinson motive E over X , the map p ∗ : R Γ( X, E ) → R Γ( X , E ) = R lim ←− n R Γ( X n , E )is an isomorphism in the derived category of the category of Q -vector spaces (see Corollary 3.2.17taking into account Definition 3.2.20). Theorem . Let i : Z → S be a closed immersion between regularschemes. Assume i has pure codimension n .Then, considering the notations of 14.1.1, definition 13.5.4, and the identification (14.1.4.1) ,the composed map H B ,Z ι −→ KGL Q ,Z η ′ i −→ i ! KGL Q ,S π n −−→ i ! H B ,S ( n )[2 n ] is an isomorphism. This isomorphism, of equivalently the map obtained by adjunction: i ∗ ( H B ,Z ) → H B ,S ( n )[2 n ]is called the fundamental class associated with i . In fact, this is a canonical class in the Beilinsonmotivic cohomology of X with support in Z of bidegree (2 n, n ). Remark . It follows from Remark 13.5.5 that the fundamental class in Beilinson motiviccohomology is compatible with pullback with respect to Tor-independant square.
Proof.
We have only to check that the above composition induces an isomorphism afterapplying the functor Hom( Q S ( X ) , − ( a )[ b ]) for a smooth S -scheme X and a couple of integers( a, b ) ∈ Z . Using Remark 13.5.5(3), this composition is compatible with smooth base change andwe can assume X = S . Let us consider the projector p a : K Zr ( S ) Q = K r ( S/S − Z ) Q → K r ( S/S − Z ) Q induced by π a ◦ ι a : KGL Q → KGL Q , and denote by K ( a ) r ( S/S − Z ) (with r = 2 a − b ) its image. Byvirtue of Propostion 13.6.1, we only have to check that the following composite is an isomorphism: ρ i : K ( a ) r ( Z ) ι a −→ K r ( Z ) Q q − i −−−→ K r ( S/S − Z ) Q π a −→ K ( a + n ) r ( S/S − Z ) .
5. CONSTRUCTIBLE BEILINSON MOTIVES 247
From 13.5.2, the morphism ρ i is functorial with respect to Tor-independant cartesian squares ofregular schemes ( cf. K ( a ) r ( Z ) / / ρ i (cid:15) (cid:15) K ( a ) r ( A Z ) (cid:15) (cid:15) K ( a ) r ( Z ) ρ s (cid:15) (cid:15) o o K ( a + n ) r ( S/S − Z ) / / K ( a + n ) r ( D/D − A Z ) K ( a + n ) r ( P/P − Z ) o o in which any of the horizontal maps is an isomorphism (as a direct factor of an isomorphism).Thus, we are reduced to the case of the closed immersion s : Z → P , canonical section of theprojectivisation of a vector bundle E (where E is the normal bundle of the closed immersion i ).Moreover, as the assertion is local on Z , we may assume E is a trivial vector bundle.Let p : P → Z be the canonical projection, j : P − Z → P the obvious open immersion.Consider the element v = [ O (1)] − ∈ K ( P ). Because v has virtual rank 0, it belongs to K (1)0 ( P ).Recall that, according to the projective bundle formula, the horizontal lines in the followingcommutative diagram are split short exact sequences:0 / / K r ( P/P − Z ) Q ν / / (cid:15) (cid:15) K r ( P ) Q j ∗ / / (cid:15) (cid:15) K r ( P − Z ) Q (cid:15) (cid:15) / / / / K ( a + n ) r ( P/P − Z ) ν ′ / / K ( a + n ) r ( P ) / / K ( a + n ) r ( P − Z ) / / . By assumption on E , v n lies in the kernel of j ∗ and the diagram allows to identify the gradedpiece K ( a + n ) r ( P/P − Z ) with the submodule of K ( a + n ) r ( P ) of the form K ( a ) r ( Z ) .v n .On the other hand, j ∗ s ∗ = 0: there exists a unique element ǫ ∈ K ( Z ) such that s ∗ (1) = p ∗ ( ǫ ) .v n in K ( P ). From the relation p ∗ s ∗ (1) = 1, we obtain that ǫ is a unit in K ( Z ), withinverse the element p ∗ ( v n ). By virtue of [ SGA6 , Exp. VI, Cor. 5.8], p ∗ ( v n ) belongs to the 0-th γ -graded part of K ( P ) Q so that the same holds for its inverse ǫ . In the end, for any element z ∈ K r ( Z ), we get the following expression: s ∗ ( z ) = s ∗ (1 .s ∗ p ∗ ( z )) = s ∗ (1) .p ∗ ( z ) = p ∗ ( ǫ.z ) .v n . Thus, the commutative diagram K ( a ) r ( Z ) / / K r ( Z ) Q q − s / / s ∗ ( ( ◗◗◗◗◗◗◗◗◗◗◗ K r ( P/P − Z ) Q ν (cid:15) (cid:15) / / K ( a + n ) r ( P/P − Z ) ν ′ (cid:15) (cid:15) K r ( P ) Q / / K ( n ) r ( P )implies that the isomorphism q − s preserves the γ -filtrations (up to a shift by n ). Hence it inducesan isomorphism on the graded pieces by functoriality. (cid:3)
15. Constructible Beilinson motives15.1. Definition and basic properties.
In this section, we apply the general results ofSection 4 to the triangulated motivic category DM B . Let us first recall the definition of con-structibility (Def. 4.2.1) which corresponds to the Tate twist. Definition . Given any scheme S , we define the category DM B ,c ( S ) of constructibleBeilinson motives over S as the thick triangulated subcategory of DM B ( S ) generated by themotives of the form M S ( X )( i ) for a smooth S -scheme X and an integer i ∈ Z . Remark . Constructible Beilinson motives plays towards Beilinson motives the samerole than complexes of ´etale sheaves with bounded cohomology and constructible cohomologysheaves plays agains complexes of ´etale sheaves (in the case of torsion coefficients primer to theresidue characteristics. This fact will be even more striking after Theorems 15.2.1 and 15.2.4. . Recall from Corollary 6.2.2 that D A , Q is compactly generated by the Tate twist.According to Theorem 14.2.9, the same is true for the motivic category DM B . Thus Proposition1.4.11 gives the following criterion of constructibility for Beilinson motives: Proposition . Given any base scheme S , a Beilinson motive M over S is constructibleif and only if it is compact. Remark . In the sequel, we will give several concrete descriptions of the category ofconstructible Beilinson motives (see Corollaries 16.1.6 and 16.2.16).Recall from Proposition 14.3.1 that DM B is continuous (with respect to the Tate twist).Proposition 4.3.4 thus implies the following properties of constructibles Beilinson motives: Proposition . Let ( S α ) α ∈ A be a pro-object of noetherian finite dimensional schemeswith affine transition maps and such that the scheme S = lim ←− α ∈ A S α is noetherian of finite di-mension.Then the canonical functor: (15.1.6.1) 2 - lim −→ α DM B ,c ( S α ) → DM B ,c ( S ) is an equivalence of monoidal triangulated categories. Example . Under the assumptions of the abobe proposition, for any couple of integers( p, q ), the canonical map lim −→ α H q B ( S α , Q ( p )) → H q B ( S, Q ( p ))is an isomorphism. The motivic triangulated cate-gory DM B is separated (14.3.3) and weakly pure (see Definition 4.2.20 ; this follows directly fromTheorem 14.4.1). Thus the abstract Theorem 4.2.29 gives the finiteness theorem, which we statehere in an explicit way to help the reader: Theorem . The triangulated subcategory DM B ,c of DM B is stable by the following op-erations:(1) f ∗ for any morphism of schemes f .(2) f ∗ for any morphism f : Y → X of finite type such that X is quasi-excellent (resp. anyproper morphism f ).(3) f ! for any separated morphism of finite type f .(4) f ! for any separated morphism of finite type f .(5) ⊗ X for any scheme X .(6) Hom X for any quasi-excellent scheme X . To be more precise, point (1) and (5) are obvious, the non respe condition of point (2) is thehardest fact and follows from Theorem 4.2.24, point (3) as well as the respe condition of point (2)is Corollary 4.2.12, point (4) is Corollary 4.2.28 and point (6) is Corollary 4.2.25.15.2.2 . Let B be an excellent scheme such that dim( B ) ≤
2. Recall that B satisfies wide reso-lution of singularities up to quotient singularities (see Def. 4.1.9 and the result of De Jong recalledin 4.1.11). Thus according to Corollary 4.4.3, we get the following description of constructibleBeilinson motives: Proposition . Let S be a separated B -scheme of finite type, and T ⊂ S a closedsubscheme. Then the triangulated category DM B ,c ( S ) is the smallest triangulated category of DM B ( S ) which contained motives of the form f ∗ ( X )( n ) where n is an integer and f : X → S is a projective morphism such that X is regular connectedand f − ( T ) red is either empty, either X of the support of a strict normal crossing divisor. This result is to be compared with [
Qui73 , Sec. 7, 2.2] – it concerns homotopy invariant K-theory ratherthan K-theory.
6. COMPARISON THEOREMS 249
The main motivation to introduce the notion of constructibility is Grothendieck duality. Weobtain this duality from the theoretical result on motivic triangulated categories, more preciselyCorollary 4.4.24:
Theorem . Let B be an excellent scheme such that dim( B ) ≤ and S be a regularseparated B -scheme of finite type.Then for any separated morphism f : X → S of finite type, the premotive f ! ( S ) is a dual-izing object of DM B ,c ( X ) . In fact, if we put D X ( M ) := Hom X ( M, f ! ( S )) for any constructibleBeilinson motives M , the following properties hold:(a) For any separated S -scheme of finite type X , the functor D X preserves constructibleobjects.(b) For any separated S -scheme of finite type X , the natural map M → D X ( D X ( M )) is an isomorphism for any constructible Beilinson motive M .(c) For any separated S -scheme of finite type X , and for any Beilinson motive M and N over X , if N is constructible then we have a canonical isomorphism D X ( M ⊗ X D X ( N )) ≃ Hom X ( M, N ) . (d) For any morphism between separated S -schemes of finite type f : Y → X , we havenatural isomorphisms D Y ( f ∗ ( M )) ≃ f ! ( D X ( M )) f ∗ ( D X ( M )) ≃ D Y ( f ! ( M )) D X ( f ! ( N )) ≃ f ∗ ( D Y ( N )) f ! ( D Y ( N )) ≃ D X ( f ∗ ( N )) where M (resp. N ) is a constructible Beilinson motive over X (resp. Y ). . Let R be a Q -algebra. We define the premotivic triangulated category of constructible Beilinson motives with coefficientsin R as the category of constructible objects of the category DM B ( − , R ) defined in Paragraph14.2.20.According to loc. cit. , for any constructible Beilinson motives with coefficients in Q , we getan isomorphism:Hom DM B ,c ( S ) ( M, N ) ⊗ Q R −→ Hom DM B ,c ( S,R ) (cid:0) L ϕ ∗ ( M ) , L ϕ ∗ ( N ) (cid:1) . It is straightforward to see that this isomorphism allows to extend all the results proved so far forBeilinson motives with coefficient in Q to the case of R -coefficients.
16. Comparison theorems16.1. Comparison with Voevodsky motives. . We consider the premotivic adjunction of 11.4.1(16.1.1.1) γ ∗ : D A , Q ⇄ DM Q : γ ∗ . For a scheme S , γ ∗ ( S ) is a (strict) commutative ring spectrum, and, for any object M of DM Q ( S ), γ ∗ ( M ) is naturally endowed with a structure of γ ∗ ( S )-module. On the other hand, as we havethe projective bundle formula in DM Q ( S ) (11.3.4), γ ∗ ( S ) is orientable (12.2.10), which impliesthat, for any object M of DM Q ( S ), γ ∗ ( M ) is an H B ,S -module, whence is H B -local (14.2.16). Asconsequence, we get a canonical factorization of (16.1.1.1):(16.1.1.2) D A , Q β ∗ −→ DM B ϕ ∗ −−→ DM Q . The examples we have in mind are: R = E is a number field, R = C , R = Q l , ¯ Q l for a prime l . Consider the commutative diagram of premotivic categoriesD A , Q γ ∗ / / ρ ♯ (cid:15) (cid:15) DM Q ψ ♯ (cid:15) (cid:15) D A , Q γ ∗ / / DM Q (16.1.1.3)in which the two vertical maps are the canonical enlargements, and, in particular, are fully faithful(see 6.1.8).Let t denotes either the qfh-topology or the h-topology. We also have the following commu-tative triangle D A , Q γ ∗ / / a ∗ DM Q α ∗ / / DM t, Q (16.1.1.4)in which both a ∗ and α ∗ are induced by the t -sheafification functor; see 5.3.31 and 11.1.21. Weobtain from (16.1.1.2), (16.1.1.3), and (16.1.1.4) the commutative diagram of premotivic categoriesbelow, in which χ ♯ = ϕ ∗ α ∗ ψ ♯ . D A , Q β ∗ / / ρ ♯ (cid:15) (cid:15) DM B χ ♯ (cid:15) (cid:15) D A , Q a ∗ / / DM t, Q (16.1.1.5)From now on, we shall fix an excellent noetherian scheme of finite dimension S . Theorem . We have canonical equivalences of categories DM B ( S ) ≃ DM qfh , Q ( S ) ≃ DM h , Q ( S ) (recall that, for t = qfh , h , DM t, Q ( S ) stands for the localizing subcategory of DM t, Q ( S ) , spannedby the objects of shape Σ ∞ Q S ( X )( n ) , where X runs over the family of smooth S -schemes, and n ≤ is an integer; see 5.3.31). Proof.
Let t denote the qfh-topology or the h-topology. We shall prove that the functor χ ♯ : DM B ( S ) → DM t, Q ( S )is fully faithful, and that its essential image is precisely DM t, Q . The functor β ∗ : DM B → D A , Q ( S )is fully faithful, so that its composition with its left adjoint β ∗ is canonically isomorphic to theidentity. In particular, we get isomorphisms of functors: χ ♯ ≃ χ ♯ β ∗ β ∗ ≃ a ∗ ρ ♯ β ∗ . The right adjoint of a ∗ is fully faithful, and its essential image consists of the objects of D A , Q ( S )which satisfy t -descent (5.3.30). On the other hand, the functor ρ ♯ is fully faithful, and an object ofD A , Q ( S ) satisfies t -descent if and only if its image by ρ ♯ satisfies t -descent (6.1.11). By virtue ofTheorem 14.3.4, this implies immediately that χ ♯ is fully faithful. Let DM t, Q ( S ) be the localizingsubcategory of DM t, Q ( S ) spanned by the objects of shape Σ ∞ Q ( X )( n ), where X runs over thefamily of smooth S -schemes, and n ≤ t, Q ( S ) is compactlygenerated (see 5.1.29, 5.2.38 and 5.3.40), and that χ ♯ is a fully faithful exact functor which preservessmall sums as well as compact objects from DM B ( S ) to DM t, Q ( S ). As, by construction, thereexists a generating family of compact objects of DM t, Q ( S ) in the essential image of χ ♯ , this impliesthat χ ♯ induces an equivalence of triangulated categories DM B ( S ) ≃ DM t, Q ( S ) (see 1.3.21). (cid:3) Let us underline the following result which completes Corollary 14.2.16:
Theorem . Let E be an object of D A ( S, Q ) . The following conditions are equivalent:
6. COMPARISON THEOREMS 251 (i) E is a Beilinson motive;(ii) E satisfies h -descent;(iii) E satisfies qfh -descent; Proof.
We already know that condition (i) implies condition (ii) (second point of Theorem14.3.4), and condition (ii) implies obviously condition (iii). It is thus sufficient to prove thatcondition (iii) implies condition (i). If E satisfies qfh-descent, then ρ ♯ ( E ) satisfies qfh-descent inDM( S, Q ) as well. The commutativity of (16.1.1.4) implies then that ρ ♯ ( E ) belongs to the essentialimage of γ ∗ (the right adjoint of γ ∗ ). As ρ ♯ is fully faithful, the commutativity of (16.1.1.3) thusimplies that E itself belongs to the essential image of γ ∗ (the right adjoint to γ ∗ ). In particular, E is then a module over the ring spectrum γ ∗ ( S ), which is itself an H B -algebra. We conclude byCorollary 14.2.16. (cid:3) Theorem . If S is geometrically unibranch, then the comparison functor ϕ ∗ : DM B ( S ) → DM Q ( S ) is an equivalence of triangulated monoidal categories. Proof. If S is geometrically unibranch, then we know that the composed functorDM Q ( S ) ψ ♯ −→ DM Q ( S ) α ∗ −−→ DM qfh , Q ( S )is fully faithful (11.1.22). The commutative diagramDM B ( S ) ϕ ∗ / / χ ♯ DM Q ( S ) α ∗ ψ ♯ / / DM qfh , Q ( S )and Theorem 16.1.2 imply that ϕ ∗ is fully faithful. As ϕ ∗ is exact, preserves small sums as wellas compact objects, and as DM Q ( S ) has a generating family of compact objects in the essentialimage of ϕ ∗ , the functor ϕ ∗ has to be an equivalence of categories (1.3.21). (cid:3) Remark . Some version of the preceding theorem (the one obtained by replacing DM B by Ho( H B - mod)) was already known in the case where S is the spectrum of a perfect field; see[ RØ08 , theorem 68]. The proof used de Jong’s resolution of singularities by alterations andPoincar´e duality in a crucial way. The proof of the preceding theorem we gave here relies onproper descent but does not use any kind of resolution of singularities.The preceding theorem allows to give the following description of constructible Beilinsonmotives over geometrically unibranch schemes:
Corollary . For any geometrically unibranch scheme S , the functor ϕ ∗ induces anequivalence of triangulated monoidal categories: DM B ,c ( S ) ∼ −−→ DM gm ( S, Q ) where the right hand side is the Q -linear version of the category of geometric (Voevodsky) motives(Definition 11.1.10). Note that we also applied Proposition 11.1.5 to get this corollary.We finally point out the following important fact about Voevodsky’s motivic cohomologyspectrum H M ,S = γ ∗ ( S ) with rational coefficients: Corollary . (1) For any geometrically unibranch excellent scheme S , the canon-ical map H B ,S → H Q M ,S is an isomorphism of ring spectra.(2) For any morphism f : T → S of excellent geometrically unibranch schemes, the canonicalmap f ∗ H Q M ,S → H Q M ,T is an isomorphism of ring spectra. The second part is the last conjecture of Voevodsky’s paper [
Voe02b ] with rational coefficients(and geometrically unibranch schemes) – see also Paragraph 11.2.21.
Proof.
The first part is a trivial consequence of the previous theorem, and the second followsfrom the first, as the Beilinson motivic cohomology spectrum is stable by pullbacks. (cid:3) . Let S be a scheme. The permutation isomorphism(16.2.1.1) τ : Q (1)[1] ⊗ LQ Q (1)[1] → Q (1)[1] ⊗ LQ Q (1)[1]satisfies the equation τ = 1 in D A ( S, Q ). Hence it defines an element ǫ in End D A ( S, Q ) ( Q ) whichalso satisfies the relation ǫ = 1. We define two projectors(16.2.1.2) e + = ǫ −
12 and e − = ǫ + 12 . As the triangulated category D A ( S, Q ) is pseudo abelian, we can define two objects by theformulæ:(16.2.1.3) Q + = Im e + and Q − = Im e − . Then for an object M of D A ( S, Q ), we set(16.2.1.4) M + = Q + ⊗ LQ M and M − = Q − ⊗ LQ M .
It is obvious that for any objects M and N of D A ( S, Q ), one has(16.2.1.5) Hom D A ( S, Q ) ( M i , N j ) = 0 for i, j ∈ { + , −} with i = j .Denote by D A ( S, Q ) + (resp. D A ( S, Q ) − ) the full subcategory of D A ( S, Q ) made of objectswhich are isomorphic to some M + (resp. some M − ) for an object M in D A ( S, Q ). Then (16.2.1.5)implies that the direct sum functor ( M + , M − ) M + ⊕ M − induces an equivalence of triangulatedcategories(16.2.1.6) (D A ( S, Q ) + ) × (D A ( S, Q ) − ) ≃ D A ( S, Q ) . We shall call D A ( S, Q ) + the category of Morel motives over S . The aim of this section is tocompare this category with DM B ( S ) (see Theorem 16.2.13). This will consists essentially ofproving that Q + is nothing else than Beilinson’s motivic spectrum H B (which was announcedby Morel in [ Mor06 ]). The main ingredients of the proof are the description of DM B ( S ) as fullsubcategory of D A ( S, Q ), the homotopy t -structure on D A ( S, Q ), and Morel’s computation ofthe endomorphism ring of the motivic sphere spectrum in terms of Milnor-Witt K-theory [ Mor03,Mor04a, Mor04b, Mor12 ].16.2.2 . For a little while, we shall assume that S is the spectrum of a field k .Recall that the algebraic Hopf fibration is the map A − { } → P , ( x, y ) [ x, y ] . This defines, by desuspension, a morphism η : Q (1)[1] → Q in D A ( S, Q ); see [ Mor03 , 6.2] (recall that we identify D A ( S, Q ) with SH Q ( S ) and that, underthis identification, Q (1)[1] corresponds to Σ ∞ ( G m )). Lemma . We have η = ǫη in Hom D A ( S, Q ) ( Q (1)[1] , Q ) . Proof.
See [
Mor03 , 6.2.3]. (cid:3) . Recall the homotopy t -structure on D A ( S, Q ); see [ Mor03 , 5.2]. To remain close tothe conventions of loc. cit. , we shall adopt homological notations, so that, for any object M ofD A ( S, Q ), we have the following truncation triangle τ > M → M → τ ≤ M → τ > M [1] .
6. COMPARISON THEOREMS 253
We whall write H for the zeroth homology functor in the sense of this t -structure. This t -structure can be described in terms of generators, as in [ Ayo07a , definition 2.2.41]: the categoryD A ( S, Q ) ≥ is the smallest full subcategory of D A ( S, Q ) which contains the objects of shape Q S ( X )( m )[ m ] for X smooth over S , m ∈ Z , and which satisfies the following stability conditions:(a) D A ( S, Q ) ≥ is stable under suspension; i.e. for any object M in D A ( S, Q ) ≥ , M [1] isin D A ( S, Q ) ≥ ;(b) D A ( S, Q ) ≥ is closed under extensions: for any distinguished triangle M ′ → M → M ′′ → M ′ [1] , if M ′ and M ′′ are in D A ( S, Q ) ≥ , so is M ;(c) D A ( S, Q ) ≥ is closed under small sums.With this description, it is easy to see that D A ( S, Q ) ≥ is also closed under tensor product (be-cause the class of generators has this property). The category D A ( S, Q ) ≤ is the full subcategoryof D A ( S, Q ) which consists of objects M such thatHom D A ( S, Q ) ( Q S ( X )( m )[ m + n ] , M ) ≃ X/S smooth, m ∈ Z , and n >
0; see [
Ayo07a , 2.1.72].Note that the heart of the homotopy t -structure is symmetric monoidal, with tensor product ⊗ h defined by the formula: F ⊗ h G = H ( F ⊗ L S G )(the unit object is H ( Q )).We shall still write η : H ( Q (1)[1]) → H ( Q ) for the map induced by the algebraic Hopffibration. Proposition . Tensoring by Q ( n )[ n ] defines a t -exact endofunctor of D A ( S, Q ) forany integer n . Proof.
As tensoring by Q ( n )[ n ] is an equivalence of categories, it is sufficient to prove thisfor n ≥
0. This is then a particular case of [
Ayo07a , 2.2.51]. (cid:3)
Proposition . For any smooth S -scheme X of dimension d , and for any object M of D A ( S, Q ) , the map Hom( Q S ( X ) , M ) → Hom( Q S ( X ) , M ≤ n ) is an isomorphism for n > d . Proof.
Using [
Mor03 , lemma 5.2.5], it is sufficient to prove the analog for the homotopy t -structure on D eff A , Q ( S ), which follows from [ Mor05 , lemma 3.3.3]. (cid:3)
Proposition . The homotopy t -structure is non degenerated. Even better, for anyobject M of D A ( S, Q ) , we have canonical isomorphisms L lim −→ n τ >n M ≃ M and R lim ←− n τ >n M ≃ , as well as isomorphisms L lim −→ n τ ≤ n M ≃ and M ≃ R lim ←− n τ ≤ n M .
Proof.
The first assertion is a direct consequence of propositions 16.2.5 and 16.2.6 (becausethe objects of shape Q S ( X )( m )[ i ], for X/S smooth, and m, i ∈ Z , form a generating family).As the objects Q S ( X )( m )[ m + n ] are compact in D A ( S, Q ), the category D A ( S, Q ) ≤ is closedunder small sums. As D A ( S, Q ) ≥ is also closed under small sums, we deduce easily that thetruncation functors τ > and τ ≤ preserve small sums, which implies that the homology functor H has the same property. Moreover, if C → · · · → C n → C n +1 → · · · is a sequence of maps in D A ( S, Q ), then C = L lim −→ n C n fits in a distinguished triangle of shape M n C n − s → M n C n → C → M n C n [1] , where s is the map induced by the maps C n → C n +1 . This implies that, for any integer i , we havelim −→ n H i ( C n ) ≃ H i ( C )(where the colimit is taken in the heart of the homotopy t -structure). As the homotopy t -structureis non degenerated, this proves the two formulas L lim −→ n τ >n M ≃ M and L lim −→ n τ ≤ n M ≃ . Let X be a smooth S -scheme of finite type, and p , q be some integer. To prove that the mapHom( Q S ( X )( m )[ i ] , M ) → Hom( Q S ( X )( m )[ i ] , R lim ←− n τ ≤ n M )is bijective, we may assume that m = 0 (replacing M by M ( − m )[ − m ] and i by i − m , and usingProposition 16.2.5). Consider the Milnor short exact sequence below, with A = Q S ( X )[ i ]:0 → lim ←− n Hom( A [1] , τ ≤ n M ) → Hom( A, R lim ←− n τ ≤ n M ) → lim ←− n Hom(
A, τ ≤ n M ) → . Using Proposition 16.2.6, as lim ←− of a constant functor vanishes, we get that the mapHom( A, M ) → Hom( A, R lim ←− n τ ≤ n M )is an isomorphism. This gives the isomorphism M ≃ R lim ←− n τ ≤ n M .
Using the previous isomorphism, and by contemplating the homotopy limit of the homotopy cofibersequences τ >n M → M → τ ≤ n M , we deduce the isomorphism R lim ←− n τ >n M ≃ (cid:3) Lemma . We have H B ∈ D A ( S, Q ) ≥ , so that we have a canonical map H B → H ( H B ) in D A ( S, Q ) . In particular, for any object M in the heart of the homotopy t -structure, if M isendowed with an action of the monoid H ( H B ) , then M has a natural structure of H B -module in D A ( S, Q ) . Proof. As H B is isomorphic to the motivic cohomology spectrum in the sense of Voevod-sky (16.1.7), the first assertion is the first assertion of [ Mor03 , theorem 5.3.2]. Therefore, thetruncation triangle for H B gives a triangle τ > H B → H B → H ( H B ) → τ > H B [1] , which gives the second assertion. For the third assertion, consider an object M in the heart ofthe homotopy t -structure, endowed with an action of H ( H B ). Note that D A ( S, Q ) ≥ is closedunder tensor product, so that H B ⊗ L S M is in D A ( S, Q ) ≥ . Hence we have natural maps H B ⊗ L S M → H ( H B ⊗ L S M ) → H ( H ( H B ) ⊗ L S M ) = H ( H B ) ⊗ h M .
Then the structural map H ( H B ) ⊗ h M → M defines a map H B ⊗ L S M → M which gives theexpected action (observe that, as we already know that H B -modules do form a thick subcategoryof D A ( S, Q ) (14.2.8), we don’t even need to check all the axioms of an internal module: it issufficient to check that the unit Q → H B induces a section M → H B ⊗ L S M of the map constructedabove). (cid:3)
6. COMPARISON THEOREMS 255
Lemma . We have the following exact sequence in the heart of the homotopy t -structure. H ( Q (1)[1]) η −→ H ( Q ) → H ( H B ) → Proof.
Using the equivalence of categories from the heart of the homotopy t -structure tothe category of homotopy modules in the sense of [ Mor03 , definition 5.2.4], by virtue of Corol-lary 16.1.7 and [
Mor03 , theorem 5.3.2], we know that H ( H B ) corresponds to the homotopymodule K M ∗ ⊗ Q associated with Milnor K-theory, while H ( Q ) corresponds to the homotopymodule K MW ∗ ⊗ Q associated with Milnor-Witt K-theory (which follows easily from [ Mor12 ,theorems 2.11, 6.13 and 6.40]). Considering K M ∗ and K MW ∗ as unramified sheaves in the sense ofMorel [ Mor12 ], this lemma is then a reformulation of the isomorphism K MW ∗ ( F ) /η ≃ K M ∗ ( F )for any field F ; see [ Mor12 , remark 2.2]. (cid:3)
Proposition . We have H B + ≃ H B , and the induced map Q + → H B gives a canonicalisomorphism H ( Q + ) ≃ H ( H B ) . Proof.
The map ǫ (1)[1] : Q (1)[1] → Q (1)[1] can be described geometrically as the morphismassociated with the pointed morphism ı : G m → G m , t t − (see the second assertion of [ Mor03 , lemma 6.1.1]). In the decomposition K ( G m ) ≃ k [ t, t − ] × ≃ k × ⊕ Z , the map ı induces multiplication by − Z . Using the periodicity isomorphism KGL (1)[2] ≃ KGL , we get the identifications: K ( G m ) ⊃ Hom
SH( k ) (Σ ∞ ( G m )[1] , KGL ) ≃ Hom
KGL ( KGL , KGL ) ≃ K ( k ) ≃ Z . Therefore, ǫ acts as the multiplication by − KGL Q , whence on H B as well. Thismeans precisely that H B + ≃ H B . By Lemma 16.2.3, the class 2 η vanishes in Q + , so that, appylingthe ( t -exact) functor M M + to the exact sequence of Lemma 16.2.9, we get an isomorphism H ( Q + ) ≃ H ( H B + ) ≃ H ( H B ). (cid:3) Corollary . For any object M in the heart of the homotopy t -structure, M + is aBeilinson motive. Proof.
The object M is an H ( Q )-module, so that M + is an H ( Q + )-module. By virtueof Proposition 16.2.10, M + is then a module over H ( H B ), so that, by Lemma 16.2.8, M + isnaturally endowed with an action of H B . (cid:3) Remark . Until now, we did not really use the fact we are in a Q -linear context(replacing H B by Voevodsky’s motivic spectrum, we just needed 2 to be invertible in the precedingcorollary). However, the following result really uses Q -linearity (because, in the proof, we seeDM B ( S ) as a full subcategory of D A ( S, Q ); see Proposition 14.2.3). Theorem . For any noetherian scheme of finite dimension S , the map Q + → H B isan isomorphism in D A ( S, Q ) . As a consequence, we have a canonical equivalence of triangulatedmonoidal categories D A ( S, Q ) + ≃ DM B ( S ) . This theorem has already been proved by Morel when S is the spectrum of a perfect field –where the left hand side is the rational category of Voevodsky motives. Morel announced thatthe category D A ( S, Q ) + should be the category of rational motives and this theorem confirm hisinsight. Proof.
Observe that, if ever Q + ≃ H B , we have D A ( S, Q ) + ≃ DM B ( S ): this follows fromthe fact that an object M of D A ( S, Q ) belongs to D A ( S, Q ) + (resp. to DM B ( S )) if and only ifthere exists an isomorphism M ≃ M + (resp. M ≃ H B ⊗ L S M ; see 14.2.16). It is thus sufficient toprove the first assertion.As both Q + and H B are stable by pullback, it is sufficient to treat the case where S = Spec ( Z ).Using Corollary 14.3.2, we may replace S by any of its henselisations, so that, by the localizationproperty, it is sufficient to treat the case where S is the spectrum of a (perfect) field k .We shall prove directly that, for any object M of D A ( S, Q ), M + is an H B -module (or,equivalently, is H B -local). Note that DM B ( S ) is closed under homotopy limits and homotopycolimits in D A ( S, Q ): indeed the inclusion functor DM B → D A , Q has a left adjoint whichpreserves a family of compact generators, whence it also has a left adjoint (1.3.20). By virtueof Proposition 16.2.7, we may thus assume that M is bounded with respect to the homotopy t -structure. As DM B ( S ) is certainly closed under extensions in D A ( S, Q ), we may even assumethat M belongs to the heart the homotopy t -structure. We conclude with Corollary 16.2.11. (cid:3) Corollary . For any noetherian scheme of finite dimension S , if − is a sum ofsquares in all the residue fields of S (e.g. if S is a scheme over a finite field), then Q − ≃ in D A ( S, Q ) , and we have a canonical equivalence of triangulated monoidal categories D A ( S, Q ) ≃ DM B ( S ) . Proof.
It is sufficient to prove that, under this assumption, Q − ≃
0. As in the precedingproof, we may replace S by any of its henselisations (4.3.9), so that, by the localization property(and by induction on the dimension), it is sufficient to treat the case where S is the spectrum ofa field k . We have to check that, if − k , then we have ǫ = −
1. Using[
Mor03 , remark 6.3.5 and lemma 6.3.7], we see that, if k is of characteristic 2, we always have ǫ = −
1, while, if the characteristic of k is distinct from 2, we have a morphism of rings GW ( k ) → Hom D A , Q (Spec( k )) ( Q , Q ) , where GW ( k ) denotes the Grothendieck-Witt ring over k . This morphism sends the class of thequadratic form − X to − ǫ and this proves the result. (For a more precise version of this, withintegral coefficients, see [ Mor12 , proposition 2.13].) (cid:3) . Recall from Example 5.3.43 that we can describe the category D A ,c ( S, Q ) of compactobjects of D A ( S, Q ) as the triangulated monoidal category obtained from (cid:16) K b ( Q ( Sm /S )) / ( BG S ∪ T A S ) (cid:17) ♮ by formally inverting the Tate twist. The operation ǫ still acts on this category and the decom-position in + and − part of a motive respects constructibility as this is a decomposition by directfactors. The preceding theorem gives the following description of constructible Beilinson motives: Corollary . For any noetherian scheme of finite dimension S , there is a canonicalequivalence of triangulated monoidal categories DM B ,c ( S ) ≃ D A ,c ( S, Q ) + When − is a sum of square in all the residue fields of S , this equivalence can be written: DM B ,c ( S ) ≃ D A ,c ( S, Q ) . . Consider the Q -linear ´etale motivic category D A , ´et ( − , Q ), defined byD A , ´et ( S, Q ) = D A (Sh ´et ( Sm /S, Q ))(see 5.3.31). The ´etale sheafification functor induces a morphism of motivic categories(16.2.17.1) D A ( S, Q ) → D A , ´et ( S, Q ) . We shall prove the following result, as an application of Theorem 16.2.13. i.e. the Grothendieck group of quadratic forms
7. REALIZATIONS 257
Theorem . For any noetherian scheme of finite dimension S , there is a canonicalequivalence of categories DM B ( S ) ≃ D A , ´et ( S, Q ) . As for Theorem 16.2.13, the idea of this result is from F. Morel who already proved it at leastin the case of a base field.In order prove the above Theorem, we shall study the behaviour of the decomposition (16.2.1.3)in D A , ´et ( S, Q ): Lemma . We have Q − ≃ in D A , ´et ( S, Q ) . Proof.
Proceeding as in the proof of Theorem 16.2.13, we may assume that S is the spectrumof a perfect field k . By ´etale descent, we see that we may replace k by any of its finite extension.In particular, we may assume that − k . But then, by virtue of Corollary16.2.14, Q − ≃ A ( S, Q ), so that, by functoriality, Q − ≃ A , ´et ( S, Q ). (cid:3) Proof of Theorem 16.2.18.
Note that the functor (16.2.17.1) has a fully faithful rightadjoint, whose essential image consists of objects of D A ( S, Q ) which satisfy ´etale descent. As anyBeilinson motive satisfies ´etale descent (first point of 14.3.4), DM B ( S ) can be seen naturally as afull subcategory of D A , ´et ( S, Q ). On the other hand, by virtue of the preceding lemma, any objectof D A ( S, Q ) which satisfies ´etale descent belongs to D A ( S, Q ) + . Hence, by Theorem 16.2.13,any object of D A ( S, Q ) which satisfies ´etale descent is a Beilinson motive. This achieves theproof. (cid:3) Remark . If S is excellent, and if all the residue fields of S are of characteristic zero,one can prove Theorem 16.2.18 independently of Morel’s theorem: this follows then directly froma descent argument, namely from Corollary 3.3.38 and from Theorem 16.1.3. Corollary . For any regular noetherian scheme of finite dimension S , we have canon-ical isomorphisms Hom D A , ´et ( S, Q ) ( Q S , Q S ( p )[ q ]) ≃ Gr pγ K p − q ( S ) Q . Proof.
This follows immediately from Theorem 16.2.18, by definition of DM B (14.2.14). (cid:3) Corollary . For any geometrically unibranch excellent noetherian scheme of finitedimension S , there is a canonical equivalence of symmetric monoidal triangulated categories D A , ´et ( S, Q ) ≃ DM( S, Q ) . Proof.
This follows from theorems 16.1.4 and 16.2.18. (cid:3)
Remark . The preceding corollary extends immediately to the case of coefficients ina Q -algebra R (cf. Example 5.3.36 for the left hand side and Paragraph 14.2.20 for the right handside). Corollary . Let S be an excellent noetherian scheme of finite dimension. An objectof D A ( S, Q ) satisfies h -descent if and only if it satisfies ´etale descent. Proof.
This follows from theorems 16.1.3 and 16.2.18. (cid:3)
17. Realizations17.1. Tilting. . Let M be a stable perfect symmetric monoidal Sm -fibred combinatorial model categoryover an adequate category of S -schemes S , such that Ho( M ) is motivic, with generating set oftwists τ .Consider a homotopy cartesian commutative monoid E in M . Then E - mod is an Sm -fibredmodel category, such that Ho( E - mod) is motivic, and we have a morphism of motivic categories(see 7.2.13 and 7.2.18) Ho( M ) → Ho( E - mod) , M
7→ E ⊗ L M . In practice, all the realization functors are obtained in this way (at least over fields), which can beformulated as follows (for simplicity, we shall work here in a Q -linear context, but, if we are readyto consider higher categorical constructions, there is no reason to make such an assumption).17.1.2 . Consider a quasi-excellent noetherian scheme S of finite dimension, as well as two stablesymmetric monoidal Sm -fibred combinatorial model categories M and M ′ over the category of S -schemes of finite type such that Ho( M ) and Ho( M ′ ) are motivic (as triangulated premotiviccategories). We also assume that both Ho( M ) and Ho( M ′ ) are Q -linear and separated, and aregenerated by their Tate twists (1.1.41).Consider a Quillen adjunction(17.1.2.1) ϕ ∗ : M ⇄ M ′ : ϕ ∗ , inducing a morphism of Sm -fibred categories(17.1.2.2) L ϕ ∗ : Ho( M ) → Ho( M ′ ) . Note that the functor L ϕ ∗ preserves constructible objects, whence defines a morphism of motiviccategories(17.1.2.3) L ϕ ∗ : Ho( M ) c → Ho( M ′ ) c . Proposition . Under the assumptions of 17.1.2, if, for any regular S -scheme of finitetype X , and for any integers p and q , the map Hom
Ho( M )( X ) ( X , X ( p )[ q ]) → Hom
Ho( M ′ )( X ) ( X , X ( p )[ q ]) is bijective, then the morphism (17.1.2.3) is an equivalence of motivic categories. Moreover, if both Ho( M ) and Ho( M ′ ) are compactly generated by their Tate twists, then the morphism (17.1.2.2) is an equivalence of motivic categories. Proof.
Note first that, for any S -scheme of finite type X , and for any integers p and q , themap Hom Ho( M )( X ) ( X , X ( p )[ q ]) → Hom
Ho( M ′ )( X ) ( X , X ( p )[ q ])is bijective: by h-descent (3.3.37) and by virtue of Gabber’s weak uniformization Theorem 4.1.2,it is sufficient to treat the case where X is regular, which is done by assumption. Let S be a S -scheme of finite type. To prove that the functor L ϕ ∗ : Ho( M ) c ( S ) → Ho c ( M ′ )( S )is an equivalence, by virtue of Theorem 4.2.16, it is sufficient to prove that, for any projectivemorphisms f : X → S and g : Y → S , and for any integers p and q , the mapHom Ho( M )( X ) ( R f ∗ ( X ) , R g ∗ ( Y )( p )[ q ]) → Hom
Ho( M ′ )( X ) ( R f ∗ ( X ) , R g ∗ ( Y )( p )[ q ])is an isomorphism. Consider the pullback square X × S Y pr / / pr (cid:15) (cid:15) Y g (cid:15) (cid:15) X f / / S From Proposition 2.4.53, the functor ϕ ∗ commutes with f ! when f is a separated morphism of finitetype. One then easily concludes using this fact and the isomorphisms (obtained by adjunctionand proper base change)Hom( R f ∗ ( X ) , R g ∗ ( Y )( p )[ q ]) ≃ Hom( L g ∗ R f ∗ ( X ) , X ( p )[ q ]) ≃ Hom( R pr , ∗ L pr ∗ ( X ) , X ( p )[ q ]) ≃ Hom( R pr , ∗ ( X × S Y ) , X ( p )[ q ]) , that (17.1.2.3) is an equivalence of motivic categories. If both Ho( M ) and Ho( M ′ ) are compactlygenerated by their Tate twists, then the sum preserving exact functor L ϕ ∗ : Ho( M )( S ) → Ho( M ′ )( S )
7. REALIZATIONS 259 is an equivalence at the level of compact objects, whence it is an equivalence of categories (1.3.21). (cid:3) . Under the assumptions of 17.1.2, assume that M and M ′ are strongly Q -linear (7.1.4),left proper, tractable, satisfy the monoid axiom, and have cofibrant unit objects. Let E ′ be a fibrantresolution of in M ′ (Spec ( k )). By virtue of Theorem 7.1.8, we may assume that E ′ is a fibrantand cofibrant commutative monoid in M ′ . Then R ϕ ∗ ( ) = ϕ ∗ ( E ′ ) is a commutative monoid in M . Let E be a cofibrant resolution of ϕ ∗ ( E ′ ) in M (Spec ( k )). Using Theorem 7.1.8, we mayassume that E is a fibrant and cofibrant commutative monoid, and that the map E → R ϕ ∗ ( E ′ )is a morphism of commutative monoids (and a weak equivalence by construction). We can see E and E ′ as cartesian commutative monoids in M and M ′ respectively (by considering their pullbacksalong morphisms of finite type f : X → Spec ( k )). We obtain the essentially commutative diagramof left Quillen functors below (in which the lower horizontal map is the functor induced by ϕ ∗ andby the change of scalars functor along the map ϕ ∗ ( E ) → E ′ ): M / / (cid:15) (cid:15) M ′ (cid:15) (cid:15) E - mod / / E ′ - mod(17.1.4.1)where E - mod and E ′ - mod are respectively the model premotivic categories of E -modules and E ′ -modules (see Proposition 7.2.11).Note furthermore that the right hand vertical left Quillen functor is a Quillen equivalenceby construction (identifying M ′ ( X ) with X -modules, and using the fact that the morphism ofmonoids X → E ′ X is a weak equivalence in M ′ ( X )). Theorem . Consider the assumptions of 17.1.4, with S = Spec ( k ) the spectrum of afield k . We suppose furthermore that one of the following conditions is verified.(i) The field k is perfect.(ii) The motivic categories Ho( M ) and Ho( M ′ ) are continuous and semi-separated.Then the morphism Ho( E - mod) c → Ho( E ′ - mod) c ≃ Ho( M ′ ) c is an equivalence of motivic categories. Under these identifications, the morphism (17.1.2.3) cor-responds to the change of scalar functor Ho( M ) c → Ho( E - mod) c , M
7→ E ⊗ L M .
If moreover both
Ho( M ) and Ho( M ′ ) are compactly generated by their Tate twists, then theseidentifications extend to non-constructible objects, so that, in particular, the morphism (17.1.2.2) corresponds to the change of scalar functor Ho( M ) → Ho( M ′ ) ≃ Ho( E - mod) , M
7→ E ⊗ L M .
Remark . This theorem can be thought as (a part of) a tilting theory for motivic(homotopy) categories.
Proof.
For any regular k -scheme of finite type X , and for any integers p and q , the mapHom Ho( M )( X ) ( X , E X ( p )[ q ]) → Hom
Ho( M ′ )( X ) ( X , E ′ X ( p )[ q ])is bijective: this is easy to check whenever X is smooth over k , which proves the assertion undercondition (i), while, under condition (ii), we see immediately from Proposition 4.3.15 that we mayassume condition (i). The first assertion is then a special case of the first assertion of Proposition17.1.3. Similarly, by Proposition 7.2.7, the second assertion follows from the second assertion ofProposition 17.1.3. (cid:3) Example . Let M be the stable Sm -fibred model category of Tate spectra, so thatHo( M ) = D A , Q , and write M B for the left Bousfield localization of M by the class of H B -equivalences (see 14.2.3), so that Ho( M B ) = DM B .Let k be a field of characteristic zero, endowed with an embedding σ : k → C . Given acomplex analytic manifold X , let M an ( X ) be the category of complexes of sheaves of Q -vectorspaces on the smooth analytic site of X (i.e. on the category of smooth analytic X -manifolds,endowed with the Grothendieck topology corresponding to open coverings), endowed with its localmodel structure (see [ Ayo07b , 4.4.16] and [
Ayo10 ]). We shall write M effBetti ( X ) for the stableleft Bousfield localization of M an ( X ) by the maps of shape Q ( U × D ) → Q ( U ) for any analyticsmooth X ( C )-manifold U (where D = { z ∈ C , | z | < } denotes the open disc). We define at last M Betti ( X ) as the stable model category of analytic G m -spectra over X ( C ) (see [ Ayo10 , section1]). Given a k -scheme of finite type X , we shall write(17.1.7.1) D Betti ( X ) := Ho( M Betti ( X ))(where the topological space X ( C ) is endowed with its canonical analytic structure). Accordingto [ Ayo10 , 1.8 and 1.10], there exists a canonical equivalences of categories(17.1.7.2) D
Betti ( X ) ≃ Ho( M effBetti ( X )) ≃ D( X ( C ) , Q ) , where D( X ( C ) , Q ) stands for the (unbounded) derived category of the abelian category of sheavesof Q -vector spaces on the small site of X ( C ). By virtue of [ Ayo10 , section 2], there existsa symmetric monoidal left Quillen morphism of monoidal Sm -fibred model categories over thecategory of k -schemes of finite type(17.1.7.3) An ∗ : M → M Betti , which induces a morphism of motivic categories over the category of k -schemes of finite type. Hence R An ∗ ( ) is a ring spectrum in D A , Q (Spec ( k )) which represents Betti cohomology of smooth k -schemes. As D Betti satisfies ´etale descent, it follows from Corollary 3.3.38 that it satisfies h-descent,from which, by virtue of Theorem 16.1.3, the morphism (17.1.7.3) defines a left Quillen functor(17.1.7.4) An ∗ : M B → M Betti , hence gives rise to a morphism of motivic categories(17.1.7.5) DM B → D Betti , the Betti realization functor of Beilinson motives.Appyling Theorem 17.1.5 to (17.1.7.4), we obtain a commutative ring spectrum E Betti = R An ∗ ( ) which represents Betti cohomology of smooth k -schemes, such that the restriction of thefunctor (17.1.7.5) to constructible objects corresponds to the change of scalars functors(17.1.7.6) DM B ,c ( X ) → Ho( E Betti - mod) c ( X ) ≃ D bc ( X ( C ) , Q ) , M
7→ E
Betti ⊗ L M .
In other words, once Betti cohomology of smooth k -schemes is known, one can reconstruct canon-ically the bounded derived categories of constructible sheaves on X ( C ) for any k -scheme of finitetype X , from the theory of mixed motives. We expect all the realization functors to be of thisshape (which should follow from (some variant of) Theorem 17.1.5): the (absolute) cohomologyof smooth k -schemes with constant coefficients determines the derived categories of constructiblesheaves over any k -scheme of finite type, whatever this means. For instance, the whole theory ofvariations of mixed Hodge structures should be obtained from Deligne cohomology, seen as a ringspectrum in DM B ( k ) (or, more precisely, in M B ( k )). Let S be an excellent (regular) noetherian scheme offinite dimension, and K a field of characteristic zero, called the field of coefficients .17.2.1 . Let E be a Nisnevich sheaf of commutative differential graded K -algebras (i.e. is acommutative monoid in the category of sheaves of complexes of K -vector spaces). We shall write H n ( X, E ) = Hom D eff A , Q ( X ) ( Q X , E [ n ])
7. REALIZATIONS 261 for any smooth S -scheme of finite type X , and for any integer n (note that, if E satisfies Nisnevichdescent and is A -homotopy invariant, which we can always assume, using 7.1.8, then H n ( X, E ) = H n ( E ( X ))).We introduce the following axioms :W1 Dimension .— H i ( S, E ) ≃ ( K if i = 0,0 otherwise.W2 Stability .— dim K H i ( G m , E ) = ( i = 0 or i = 1,0 otherwise.W3 K¨unneth formula .— For any smooth S -schemes X and Y , the exterior cup productinduces an isomorphism M p + q = n H p ( X, E ) ⊗ K H q ( Y, E ) ∼ → H n ( X × S Y, E ) . W3 ′ Weak K¨unneth formula .— For any smooth S -scheme X , the exterior cup product inducesan isomorphism M p + q = n H p ( X, E ) ⊗ K H q ( G m , E ) ∼ → H n ( X × S G m , E ) . . Under assumptions W1 and W2, we will call any non zero element c ∈ H ( G m , E ) a stability class . Note that such a class corresponds to a non trivial map c : Q S (1) → E in D eff A , Q ( S ) (using the decomposition Q ( G m ) = Q ⊕ Q (1)[1]). In particular, possibly after re-placing E by a fibrant resolution (so that E is homotopy invariant and satisfies Nisnevich descent),such a stability class can be lifted to an actual map of complexes of presheaves. Such a lift willbe called a stability structure on E . Definition . A sheaf of commutative differential graded K -algebras E as above is a mixed Weil cohomology (resp. a stable cohomology ) if it satisfies the properties W1, W2 and W3(resp. W1, W2 and W3 ′ ) stated above. Proposition . Let E be a stable cohomology. There exists a (commutative) ring spec-trum E in DM B ( S ) with the following properties.(i) For any smooth S -scheme X , and any integer i , there is a canonical isomorphism of K -vector spaces H i ( X, E ) ≃ Hom DM B ( S ) ( M S ( X ) , E [ i ]) . (ii) Any choice of a stability structure on E defines a map Q (1) → E in DM B ( S ) , whichinduces an E -linear isomorphism E (1) ≃ E . Proof.
By [
CD12 , Proposition 2.1.6], there exists a commutative ring spectrum E in D A , Q ( S )such that, for any smooth S -scheme X , and any integer i , there is a canonical isomorphism of K -vector spaces H i ( X, E ) ≃ Hom D A , Q ( S ) ( M S ( X ) , E [ i ]) , and such that any choice of a stability structure on E defines an isomorphism E (1) ≃ E . By virtueof [ CD12 , corollary 2.2.8] and of Theorem 12.2.10, this ring spectrum E is oriented, so that, byCorollary 14.2.16, E is an H B -module, i.e. belongs to DM B ( S ). (cid:3) . Given a stable cohomology E and its associated ring spectrum E , we can see E as acartesian commutative monoid: we define, for an S -scheme X , with structural map f : X → S : E X = L f ∗ ( E ) , and put(17.2.5.1) D( X, E ) := Ho( E - mod)( X ) = Ho( E X - mod) . We thus have realization functors(17.2.5.2) DM B ( X ) → D( X, E ) , M
7→ E X ⊗ L X M which commute with the six operations of Grothendieck. Furthermore, D( − , E ) is a motiviccategory which is Q -linear (in fact K -linear), separated, and continuous. Observe furthermorethat, if S is the spectrum of a field , then D( − , E ) is also pure (4.4.16), so that the six Grothendieckoperations preserves constructible objects in D( − , E ) (4.2.29).For an S -scheme X , define H q ( X, E ( p )) = Hom DM B ( X ) ( Q X , E ( p )[ q ]) ≃ Hom D( X, E ) ( E X , E X ( p )[ q ])(this notation is compatible with 17.2.1 by virtue of Proposition 17.2.4). Corollary . Any stable cohomology (in particular, any mixed Weil cohomology) ex-tends naturally to S -schemes of finite type, and this extension satisfies cohomological h -descent(in particular, ´etale descent as well as proper descent). Proof.
This follows immediately from the construction above and from Theorem 14.3.4. (cid:3) . We denote by D ∨ ( S, E ) the localizing subcategory of D( S, E ) generated by its rigidobjects (i.e. by the objects which have strong duals). For instance, for any smooth and proper S -scheme X , E ( X ) = E ⊗ L S M S ( X ) belongs to D ∨ ( S, E ); see 2.4.31.If we denote by D( K ) the (unbounded) derived category of the abelian category of K -vectorspaces, we get the following interpretation of the K¨unneth formula. Theorem . If E is a mixed Weil cohomology, then the functor R Hom E ( E , − ) : D ∨ ( S, E ) → D( K ) is an equivalence of symmetric monoidal triangulated categories. Proof.
This is [
CD12 , theorem 2.6.2]. (cid:3)
Theorem . If S is the spectrum of a field, then D ∨ ( S, E ) = D( S, E ) . Proof.
This follows then from Corollary 4.4.17. (cid:3)
Remark . It is not reasonnable to expect the analog of Theorem 17.2.9 to hold when-ever S is of dimension >
0; see (the proof of) [
CD12 , corollary 3.2.7]. Heuristically, for higherdimensional schemes X , the rigid objects of D( X, E ) are extensions of some kind of locally constantsheaves (in the ℓ -adic setting, these correspond to Q ℓ -faisceaux lisses ). Corollary . If E is a mixed Weil cohomology, and if S is the spectrum of a field,then the functor R Hom E ( E , − ) : D( S, E ) → D( K ) is an equivalence of symmetric monoidal triangulated categories. Remark . This result can be thought as a tilting theory for the spectra associatedwith stable cohomologies.17.2.13 . Assume that E is a mixed Weil cohomology, and that S is the spectrum of a field k .For each k -scheme of finite type X , denote by D c ( X, E ) the category of constructible objects ofD( X, E ): by definition, this is the thick triangulated subcategory of D( X, E ) generated by objectsof shape E ( Y ) = E ⊗ L X M X ( Y ) for Y smooth over X (we can drop Tate twists because of 17.2.4 (ii)).The category D c ( X, E ) also coincides with the category of compact objects in D( X, E ); see 1.4.11.Write D b ( K ) for the bounded derived category of the abelian category of finite dimensional K -vector spaces. Note that D b ( K ) is canonically equivalent to the homotopy category of perfectcomplexes of K -modules, i.e. to the category of compact objects of D( K ). Corollary . Under the assumptions of 17.2.13, we have a canonical equivalence ofsymmetric monoidal triangulated categories D c (Spec ( k ) , E ) ≃ D b ( K ) .
7. REALIZATIONS 263
Proof.
This follows from 17.2.11 and from the fact that equivalences of categories preservecompact objects. (cid:3)
Corollary . Under the assumptions of 17.2.13, if E ′ is another K -linear stable coho-mology with associated ring spectrum E ′ , any morphism of presheaves of commutative differential K -algebras E → E ′ inducing an isomorphism H ( G m , E ) ≃ H ( G m , E ′ ) gives a canonical iso-morphism E ≃ E ′ in the homotopy category of commutative ring spectra. In particular, we getcanonical equivalences of categories D( X, E ) ≃ D( X, E ′ ) for any k -scheme of finite type X (and these are compatible with the six operations of Grothendieck,as well as with the realization functors). Proof.
This follows from Theorem 17.2.9 and from [
CD12 , theorem 2.6.5]. (cid:3)
Remark . The preceding result can be stated in the following way: if E and E ′ aretwo (strict) commutative ring spectra associated to K -linear mixed Weil cohomologies definedon smooth k -schemes, then any morphism E → E ′ in the homotopy category of (commutative)monoids in the model category of K -linear Tate spectra is invertible. Theorem . Under the assumptions of 17.2.13, the six operations of Grothendieckpreserve constructibility in D( − , E ) . Proof.
Observe that D( − , E ) is Q -linear and separated (because DM B is so, see 7.2.18), aswell as pure (by Proposition 4.4.16). We conclude with 4.2.29. (cid:3) . As a consequence, we have, for any k -scheme of finite type X , a realization functorDM B ,c ( X ) → D c ( X, E )and we deduce from Theorem 4.4.25 that it preserves all of Grothendieck six operations. For X = Spec ( k ), by virtue of Corollary 17.2.14, this corresponds to a symmetric monoidal exactrealization functor R : DM B ,c (Spec ( k )) → D b ( K ) . This leads to a finiteness result:
Corollary . Under the assumptions of 17.2.13, for any k -scheme of finite type X ,and for any objects M and N in D c ( X, E ) , Hom E ( M, N [ n ]) is a finite dimensional K -vector space,and it is trivial for all but a finite number of values of n . Proof.
Let f : X → Spec ( k ) be the structural map. By virtue of 17.2.17, as M and N areconstructible, the object R f ∗ R Hom X ( M, N ) is constructible as well, i.e. is a compact object ofD(Spec ( k ) , E ). But R Hom E ( M, N ) is nothing else than the image of R f ∗ R Hom X ( M, N ) by theequivalence of categories given by Corollary 17.2.11. Hence R Hom E ( M, N ) is a compact object ofD( K ), which means that it belongs to D b ( K ). (cid:3) . For a K -vector space V and an integer n , define V ( n ) = ( V ⊗ K Hom K ( H ( G m , E ) ⊗ n , K ) if n > V ⊗ K H ( G m , E ) ⊗ ( − n ) if n ≤ K ( −
1) = H ( G m , E ) ≃ H ( P k , E ) defines a natural isomorphism V ( n ) ≃ V for any integer n . We have canonical isomorphisms H q ( X, E ( p )) ≃ H q ( X, E )( p )(using the fact that the equivalence of Corollary 17.2.14 is monoidal). The realization functors(17.2.5.2) induce in particular cycle class maps cl X : H q B ( X, Q ( p )) → H q ( X, E )( p )(and similarly for cohomology with compact support, for homology, and for Borel-Moore homol-ogy). Example . Let k be a field of characteristic zero. We then have a mixed Weil coho-mology E dR defined by the algebraic de Rham complex E dR ( X ) = Ω ∗ A/k for any smooth affine k -scheme of finite type X = Spec ( A ) (algebraic de Rham cohomology ofsmooth k -schemes of finite type is obtained by Zariski descent); see [ CD12 , 3.1.5]. We obtainfrom 17.2.4 a commutative ring spectrum E dR , and, for a k -scheme of finite type X , we defineD dR ( X ) = D c ( X, E dR ) . We thus get a motivic category D dR , and we have a natural definition of algebraic de Rhamcohomology of k -schemes of finite type, given by H n dR ( X ) = Hom D dR ( X ) ( E dR ,X , E dR ,X [ n ]) . This definition coincides with the usual one: this is true by definition for smooth k -schemes of finitetype, while the general case follows from h-descent (17.2.6) and from de Jong’s Theorem 4.1.11(or resolution of singularities `a la Hironaka). We have, by construction, a de Rham realizationfunctor R dR : DM B ,c ( X ) → D dR ( X )which preserves the six operations of Grothendieck (Theorem 4.4.25). In particular, we have cycleclass maps H q B ( X, Q ( p )) → H q dR ( X )( p ) . Note that, for any field extension k ′ /k , we have natural isomorphisms H n dR ( X ) ⊗ k k ′ ≃ H n dR ( X × Spec( k ) Spec ( k ′ )) . Example . Let k be a field of characteristic zero, which is algebraically closed andcomplete with respect to some valuation (archimedian or not). We can then define a stablecohomology E dR , an as analytic de Rham cohomology of X an , for any smooth k -scheme of finitetype X ; see [ CD12 , 3.1.7]. As above, we get a ring spectrum E dR , an , and for any k -scheme offinite type, a category of coefficientsD dR , an ( X ) = D c ( X, E dR , an ) , which allows to define the analytic de Rham cohomology of any k -scheme of finite type X by H n dR , an ( X ) = Hom D dR , an ( X ) ( E dR , an ,X , E dR , an ,X [ n ]) . We also have a realization functor R dR , an : DM B ,c ( X ) → D dR , an ( X )which preserves the six operations of Grothendieck.We then have a morphism of stable cohomologies E dR → E dR , an which happens to be a quasi-isomorphism locally for the Nisnevich topology (this is Grothendieck’stheorem in the case where K is archimedian, and Kiehl’s theorem in the case where K is non-archimedian; anyway, one obtains this directly from Corollary 17.2.15). This induces a canonicalisomorphism E dR ≃ E dR , an in the homotopy category of commutative ring spectra. In particular, E dR , an is a mixed Weilcohomology, and, for any k -scheme of finite type, we have natural equivalences of categoriesD dR ( X ) → D dR , an ( X ) , M
7→ E dR , an ⊗ L E dR M which commute with the six operations of Grothendieck and are compatible with the realizationfunctors.Note that, in the case k = C , E dR , an coincides with Betti cohomology (after tensorization by C ), so that we have canonical fully faithful functorsD Betti ,c ( X ) ⊗ Q C → D dR , an ( X )
7. REALIZATIONS 265 which are compatible with the realization functors. More precisely, we have equivalencesD bc ( X ( C ) , C ) ≃ Ho( E Betti ⊗ Q C - mod) c ( X ) ≃ D dR , an ( X ) . In particular, by the Riemann-Hilbert correspondence, D dR , an ( X ) is equivalent to the boundedderived category of analytic regular holonomic D -modules on X which are constructible of geo-metric origin. (A purely algebraic proof of this equivalence would furnish a new proof of theRiemann-Hilbert correspondence, using Corollary 17.2.15.) Example . Let V be a complete discrete valuation ring of mixed characteristic withperfect residue field k and field of functions K . The Monsky-Washnitzer complex defines a stablecohomology E MW over smooth V -schemes of finite type, defined by E MW ( X ) = Ω ∗ A † /V ⊗ V K for any affine smooth V -scheme X = Spec ( A ) (the case of a smooth V -scheme of finite type isobtained by Zariski descent); see [ CD12 , 3.2.3]. Let E MW be the corresponding ring spectrum inDM B (Spec ( V )), and write j : Spec ( K ) → Spec ( V ) and i : Spec ( k ) → Spec ( V ) for the canonicalimmersions. As we obviously have j ∗ E MW = 0 (the Monsky-Washnitzer cohomology of a smooth V -scheme with empty special fiber vanishes), we have a canonical isomorphism E MW ≃ R i ∗ L i ∗ E MW . We define the rigid cohomology spectrum E rig in DM B (Spec ( k )) by the formula E rig = L i ∗ E MW . This is a ring spectrum associated to a K -linear mixed Weil cohomology: cohomology with co-efficients in E rig coincides with rigid cohomology in the sense of Berthelot, and the K¨unnethformula for rigid cohomology holds for smooth and projective k -schemes (as rigid cohomology co-incides then with cristalline cohomology), from which we deduce the K¨unneth formula for smooth k -schemes of finite type; see [ CD12 , 3.2.10]. As before, we defineD rig ( X ) = D c ( X, E rig )for any k -scheme of finite type X , and put H n rig ( X ) = Hom D rig ( X ) ( E rig ,X , E rig ,X [ n ]) . Here again, we have, by construction, rigid realization functors R rig : DM B ,c ( X ) → D rig ( X )which preserve the six operations of Grothendieck (Theorem 4.4.25), as well as (higher) cycle classmaps H q B ( X, Q ( p )) → H q rig ( X )( p ) . ibliography [Ayo07a] J. Ayoub, Les six op´erations de Grothendieck et le formalisme des cycles ´evanescents dans le mondemotivique (I) , Ast´erisque, vol. 314, Soc. Math. France, 2007.[Ayo07b] ,
Les six op´erations de Grothendieck et le formalisme des cycles ´evanescents dans le mondemotivique (II) , Ast´erisque, vol. 315, Soc. Math. France, 2007.[Ayo10] J. Ayoub,
Note sur les op´erations de Grothendieck et la r´ealisation de Betti , J. Inst. Math. Jussieu (2010), no. 2, 225–263.[Bar10] C. Barwick, On left and right model categories and left and right Bousfield localizations , Homology,Homotopy Appl. (2010), no. 2, 245–320.[BBD82] A. Be˘ılinson, J. Bernstein, and P. Deligne, Faisceaux pervers , Ast´erisque (1982), 5–171.[Be˘ı87] A. Be˘ılinson,
Height pairing between algebraic cycles , Current trends in arithmetical algebraic geometry(Arcata, Calif., 1985), Contemp. Math., vol. 67, Amer. Math. Soc., 1987, pp. 1–24.[Bek00] T. Beke,
Sheafifiable homotopy model categories , Math. Proc. Camb. Phil. Soc. (2000), 447–475.[BG73] K. S. Brown and S. M. Gersten,
Algebraic K -theory as generalized sheaf cohomology , Higher K-theoriesI (Proc. Conf., Battelle Memorial Inst., Seattle, Wash., 1972), Lecture Notes in Mathematics, vol. 341,Springer-Verlag, 1973, pp. 266–292.[Bla03] B. Blander, Local projective model structures on simplicial presheaves , K -Theory (2003), no. 3,283–301.[BM03] C. Berger and I. Moerdijk, Axiomatic homotopy theory for operads , Comment. Math. Helv. (2003),805–831.[BM09] , On the derived category of an algebra over an operad , Georgian Math. J. (2009), no. 1,13–28.[Bon10] M.V. Bondarko, Weights for relative motives; relation with mixed complexes of sheaves ,arXiv:1007.4543v3, 2010.[Bou93] N. Bourbaki, ´El´ements de math´ematique. Alg`ebre commutative. Chapitres 8 et 9 , Masson, 193, Reprintof the 1983 original.[Bou98] ,
Commutative algebra. Chapters 1–7 , Elements of Mathematics (Berlin), Springer-Verlag, Berlin,1998, Translated from the French, Reprint of the 1989 English translation.[Bro74] K.S. Brown,
Abstract homotopy theory and generalized sheaf cohomology , Trans. Amer. Math. Soc. (1974), 419–458.[CD09] D.-C. Cisinski and F. D´eglise,
Local and stable homological algebra in Grothendieck abelian categories ,Homology, Homotopy and Applications (2009), no. 1, 219–260.[CD12] D.-C. Cisinski and F. D´eglise, Mixed Weil cohomologies , Adv. Math. (2012), no. 1, 55–130.[Cis03] D.-C. Cisinski,
Images directes cohomologiques dans les cat´egories de mod`eles , Annales Math´ematiquesBlaise Pascal (2003), 195–244.[Cis06] , Les pr´efaisceaux comme mod`eles des types d’homotopie , Ast´erisque, vol. 308, Soc. Math. France,2006.[Cis08] ,
Propri´et´es universelles et extensions de Kan d´eriv´ees , Theory and Applications of Categories (2008), no. 17, 605–649.[Cis13] D.-C. Cisinski, Descente par ´eclatements en K-th´eorie invariante par homotopie , Ann. of Math. (2013).[Con07] B. Conrad,
Deligne’s notes on Nagata compactifications , J. Ramanujan Math. Soc. (2007), no. 3,205–257.[Cra95] S. E. Crans, Quillen closed model structures for sheaves , J. Pure Appl. Algebra (1995), 35–57.[CT11] D.-C. Cisinski and G. Tabuada,
Non connective K -theory via universal invariants , Compos. Math. (2011), no. 4, 1281–1320.[D´eg07] F. D´eglise, Finite correspondences and transfers over a regular base , Algebraic cycles and motives. Vol.1, London Math. Soc. Lecture Note Ser., vol. 343, Cambridge Univ. Press, Cambridge, 2007, pp. 138–205.[D´eg08] ,
Around the Gysin triangle II , Doc. Math. (2008), 613–675.[Del87] P. Deligne, Le d´eterminant de la cohomologie , Current trends in arithmetical algebraic geometry (Arcata,Calif., 1985), Contemp. Math., vol. 67, Amer. Math. Soc., 1987, pp. 93–177.[Del01] P. Deligne,
Voevodsky lectures on cross functors , ,Fall 2001.[dJ96] A. J. de Jong, Smoothness, semi-stability and alterations , Publ. Math. IHES (1996), 51–93. [dJ97] , Families of curves and alterations , Annales de l’institut Fourier (1997), no. 2, 599–621.[Dug01] D. Dugger, Combinatorial model categories have presentations , Adv. Math (2001), no. 1, 177–201.[Dug06] ,
Spectral enrichment of model categories , Homology, Homotopy, and Applications (2006), no. 1,1–30.[EGA1] A. Grothendieck and J. Dieudonn´e, ´El´ements de g´eom´etrie alg´ebrique. I. Le langage des sch´emas , Publ.Math. IHES (1960).[EGA2] , ´El´ements de g´eom´etrie alg´ebrique. II. ´Etude globale ´el´ementaire de quelques classes de mor-phismes , Publ. Math. IHES (1961).[EGA4] , ´El´ements de g´eom´etrie alg´ebrique. IV. ´Etude locale des sch´emas et des morphismes de sch´emasIV , Publ. Math. IHES
20, 24, 28, 32 (1964-1967).[Ful98] W. Fulton,
Intersection theory , second ed., Springer, 1998.[Gil81] H. Gillet,
Riemann-Roch theorems for higher algebraic K -theory , Adv. in Math. (1981), no. 3, 203–289.[Gro57] A. Grothendieck, Sur quelques points d’alg`ebre homologique , Tˆohoku Math. J. (1957), no. 2, 119–221.[GZ67] P. Gabriel and M. Zisman, Calculus of fractions and homotopy theory , Ergebnisse der Mathematik,vol. 35, Springer-Verlag, 1967.[Har66] R. Hartshorne,
Residues and duality , Lecture notes of a seminar on the work of A. Grothendieck, givenat Harvard 1963/64. With an appendix by P. Deligne. Lecture Notes in Mathematics, No. 20, Springer-Verlag, Berlin, 1966. MR 0222093 (36
Structure de poids `a la Bondarko sur les motifs de Beilinson , Compos. Math. (2011),no. 5, 1447–1462.[Hor10] J. Hornbostel,
Preorientations of the derived motivic multiplicative group , arXiv:1005.4546, 2010.[Hov99] M. Hovey,
Model categories , Math. surveys and monographs, vol. 63, Amer. Math. Soc., 1999.[Hov01] ,
Spectra and symmetric spectra in general model categories , J. Pure Appl. Algebra (2001),no. 1, 63–127.[Ill07] L. Illusie,
On Gabbers uniformization theorems : outline and applications to ´etale co-homology , notes from the Conference on Ramification and vanishing cycles, available at , 2007.[Ill08] ,
On gabber’s refined uniformization , notes from Talks at the Univ. Tokyo, available at , 2008.[ILO] L. Illusie, Y. Lazslo, and F. Orgogozo,
Travaux de Gabber sur l’uniformisation locale et la cohomologie´etale des sch´emas quasi-excellents , S´eminaire `a l’´Ecole Polytechnique 2006–2008; arXiv:1207.3648.[Ivo07] F. Ivorra,
R´ealisation l -adique des motifs triangul´es g´eom´etriques. I , Doc. Math. (2007), 607–671.[Jar00] J. F. Jardine, Motivic symmetric spectra , Doc. Math. (2000), 445–553.[Kra01] H. Krause, On Neeman’s well generated triangulated categories , Doc. Math. (2001), 121–126.[Lip78] J. Lipman, Desingularization of two dimensional schemes , Annals of Math. (1978), 151–207.[Lur12] J. Lurie,
Higher algebra , preprint, 2012.[Mac98] S. MacLane,
Categories for the working mathematician , second ed., Graduate Texts in Mathematics,vol. 5, Springer-Verlag, New York, 1998.[Mal05] G. Maltsiniotis,
La th´eorie de l’homotopie de Grothendieck , Ast´erisque, vol. 301, Soc. Math. France,2005.[Mat70] H. Matsumura,
Commutative algebra , W. A. Benjamin, Inc., New York, 1970.[Mor03] F. Morel,
An introduction to A -homotopy theory , Contemporary Developments in Algebraic K-theory,ICTP Lecture notes, vol. 15, 2003, pp. 357–441.[Mor04a] , On the motivic π of the sphere spectrum , Axiomatic, enriched and motivic homotopy theory,NATO Sci. Ser. II Math. Phys. Chem., vol. 131, Kluwer Acad. Publ., Dordrecht, 2004, pp. 219–260.[Mor04b] , Sur les puissances de l’id´eal fondamental de l’anneau de Witt , Comment. Math. Helv. (2004),no. 4, 689–703.[Mor05] , The stable A -connectivity theorems , K-theory (2005), 1–68.[Mor06] , Rational stable splitting of Grassmanians and the rational motivic sphere spectrum , statementof results, 2006.[Mor12] , A -algebraic topology over a field , Lecture Notes in Mathematics, vol. 2052, Springer, 2012.[MV99] F. Morel and V. Voevodsky, A -homotopy theory of schemes , Publ. Math. IHES (1999), 45–143.[MVW06] C. Mazza, V. Voevodsky, and C. Weibel, Lecture notes on motivic cohomology , Clay MathematicsMonographs, vol. 2, American Mathematical Society, Providence, RI, 2006. MR 2242284 (2007e:14035)[Nee96] A. Neeman,
The Grothendieck duality theorem via Bousfield’s techniques and Brown representability , J.Amer. Math. Soc. (1996), no. 1, 205–236.[Nee01] , Triangulated categories , Annals of Mathematics Studies, vol. 148, Princeton University Press,Princeton, NJ, 2001.[NSØ09] N. Nauman, M. Spitzweck, and P.A. Østvær,
Motivic Landweber exactness , Doc. Math. (2009),551–593.[PPR09] I. Panin, K. Pimenov, and O. R¨ondigs, On Voevodsky’s algebraic K-theory spectrum , Algebraic Topology,Abel. Symp., vol. 4, Springer, 2009, pp. 279–330.
IBLIOGRAPHY 269 [Qui73] D. Quillen,
Higher algebraic K-theory , Higher K-theories I (Proc. Conf., Battelle Memorial Inst., Seattle,Wash., 1972), Lecture Notes in Mathematics, vol. 341, Springer-Verlag, 1973, pp. 85–147.[Rio10] J. Riou,
Algebraic K-theory, A -homotopy and Riemann-Roch theorems , J. Topol. (2010), no. 2,229–264.[RØ08] O. R¨ondigs and P. A. Østvær, Modules over motivic cohomology , Adv. Math. (2008), no. 2, 689–727.[RSØ10] O. R¨ondigs, M. Spitzweck, and P.A. Østvær,
Motivic strict ring models for K-theory , Proc. Amer. Math.Soc. (2010), no. 10, 3509–3520.[Ser75] J.-P. Serre,
Alg`ebre locale , multiplicit´es , Lecture Notes in Mathematics, vol. 11, Springer-Verlag, 1975,third edition.[SGA1] ,
Revˆetements ´etales et groupe fondamental , Documents Math´ematiques, vol. 3, Soc. Math.France, 2003, S´eminaire de G´eom´etrie Alg´ebrique du Bois–Marie 1960–61 (SGA 1). ´Edition recompos´eeet annot´ee du LNM 224, Springer, 1971.[SGA4] M. Artin, A. Grothendieck, and J.-L. Verdier,
Th´eorie des topos et cohomologie ´etale des sch´emas ,Lecture Notes in Mathematics, vol. 269, 270, 305, Springer-Verlag, 1972–1973, S´eminaire de G´eom´etrieAlg´ebrique du Bois–Marie 1963–64 (SGA 4).[SGA5] ,
Cohomologie l-adique et Fonctions L , Lecture Notes in Mathematics, vol. 589, Springer-Verlag,1977, S´eminaire de G´eom´etrie Alg´ebrique du Bois–Marie 1965–66 (SGA 5).[SGA6] P. Berthelot, A. Grothendieck, and L. Illusie,
Th´eorie des intersections et th´eor`eme de Riemann-Roch ,Lecture Notes in Mathematics, vol. 225, Springer-Verlag, 1971, S´eminaire de G´eom´etrie Alg´ebrique duBois–Marie 1966–67 (SGA 6).[Shi04] B. Shipley,
A convenient model category for commutative ring spectra , Contemp. Math. (2004),473–483.[Spi99] M. Spivakovsky,
A new proof of D. Popescu’s theorem on smoothing of ring homomorphisms , Journalof the Amer. Math. Soc. (1999), no. 2, 381–444.[Spi01] M. Spitzweck, Operads, Algebras and Modules in Monoidal Model Categories and Motives , PhD the-sis, Mathematisch Naturwissenschatlichen Fakult¨at, Rheinischen Friedrich Wilhelms Universit¨at, Bonn,2001.[SS00] S. Schwede and B. Shipley,
Algebras and modules in monoidal model categories , Proc. London Math.Soc. (2000), 491–511.[SV00a] A. Suslin and V. Voevodsky, Bloch-Kato conjecture and motivic cohomology with finite coefficients ,NATO Sciences Series, Series C: Mathematical and Physical Sciences, vol. 548, pp. 117–189, Kluwer,2000.[SV00b] ,
Relative cycles and Chow sheaves , Annals of Mathematics Studies, vol. 143, ch. 2, pp. 10–86,Princeton University Press, 2000.[Swa80] R. G. Swan,
On seminormality , J. Algebra (1980), no. 1, 210–229.[TT90] R. W. Thomason and Thomas Trobaugh, Higher algebraic K -theory of schemes and of derived categories ,The Grothendieck Festschrift, Vol. III, Progr. Math., vol. 88, Birkh¨auser Boston, Boston, MA, 1990,pp. 247–435.[Vez01] G. Vezzosi, Brown-Peterson spectra in stable A -homotopy theory , Rend. Sem. Mat. Univ. Padova (2001), 47–64.[Voe92] V. Voevodsky, A letter to beilinson , K-theory preprint archive 0033, 1992.[Voe96] V. Voevodsky,
Homology of schemes , Selecta Math. (N.S.) (1996), no. 1, 111–153.[Voe98] , A -homotopy theory , Proceedings of the International Congress of Mathematicians, Vol. I(Berlin, 1998), 1998, pp. 579–604 (electronic).[Voe02a] , Motivic cohomology groups are isomorphic to higher chow groups in any characteristic , Int.Math. Res. Not. (2002), 351–355.[Voe02b] , Open problems in the motivic stable homotopy theory. I , Motives, polylogarithms and Hodgetheory, Part I (Irvine, CA, 1998), Int. Press Lect. Ser., vol. 3, Int. Press, Somerville, MA, 2002, pp. 3–34.[Voe10a] ,
Cancellation theorem , Doc.Math. Extra volume: Andrei A. Suslin sixtieth birthday (2010),671–685.[Voe10b] ,
Homotopy theory of simplicial sheaves in completely decomposable topologies , J. Pure Appl.Algebra (2010), no. 8, 1384–1398.[Voe10c] ,
Unstable motivic homotopy categories in Nisnevich and cdh-topologies , J. Pure Appl. Algebra (2010), no. 8, 1399–1406.[VSF00] V. Voevodsky, A. Suslin, and E. M. Friedlander,
Cycles, transfers and motivic homology theories , Annalsof Mathematics Studies, vol. 143, Princeton Univ. Press, 2000. ndex of terminology acyclic, H B -acyclic, 241adequate, category of schemes, 21adjunctionof P -fibred categories, see also morphism ofof premotivic categories, see also morphism ofQuillen adjunction, 61admissible topology, see also topologyadmissible, class of morphisms, 2algebra E ∞ -algebra, 164 H B -algebra, 244alteration, 92Galois alteration, 92, 107Auslander-Buchsbaum theorem, 222base change P -base change, 5proper base change, 6smooth base change, 6bifibred category, 5Bott isomorphism, 233bounded (topology), 126bounded generating family, 126Brown representability theorem, 17, 31, 36bundlenormal, 48, 49tangent, 49, 55virtual vector bundle, 42cartesian morphism, see also morphismcd-structure, 24, 208lower, 24upper, 24Chow’s lemma, 21, 35classChern, 224, 238fundamental, 236, 246coefficients, for Beilinson motives, 245, 249cofibration, 121termwise, 59coherence, 3, 5, 8, 13cohomologyalgebraic De Rham, 264analytic De Rham, 264Beilinson motivic, 243Betti, 260Chow group, see also groupeffective motivic, 218higher Chow group, see also groupK-theory, see also K-theoryLandweber exact, 241mixed Weil, 261 Monsky-Washnitzer, 265motivic, 218representable, 230rigid, 265stable, 261commute, see also functorcompact, 16, 128, 141, 216compatible with (a topology) t , 120, 125compatible with transfers, see also topologycompatible with twists, 13, 20complexalgebraic De Rham, 264Monsky-Washnitzer, 265conservative, 33, 36, 104constructibility, see also constructibleconstructible, see also τ -constructible( Z × τ )-constructible, 151 τ -constructible, 20, 93–113, 128, 141Beilinson motive, 247–249motive, 215, 216motivic complex, 215 A -contractible, 135cotransversality property, 6cover, 68Galois cover, 80h-cover, 86, 91pseudo-Galois cover, 80qfh-cover, 85, 86cycleΛ-cycle, 172, see also cycleΛ-universal (morphism of), 182associated, 172Hilbert, 173pre-special (morphism of), 175pseudo-equidimensional, 191pullback, 178associativity, 183commutativity, 182of Hilbert cycles, 174projection formulas, 184pushforward, 173restriction, 173Samuel specialization, 189special (morphism of), 177specialization, 176standard form, 173trivial, 186decomposition, Adams, 240, 244deformation space, 48derivator, Grothendieck, 66, 73, 75, 124derived derived P -premotivic category, 132descentcdh-descent, 78, 86cohomological h-descent, 262cohomological t -descent, 120, 123´etale, 84, 86, 89, 246, 257Galois, 175h-descent, 86, 88, 89, 246, 257Nisnevich, 77qfh-descent, 85, 86, 88, 217 t -descent, 69, 123, 132, 148dg-structure, 120, 131diagram S -diagram, 57, 122, 131direct image with compact support, see also functor, left exceptionaldivisorWeil, 221domain (of a Λ-cycle), 172dual, strong, 47, 225dualitylocal duality, 111duality, Grothendieck, 113, 249dualizable, strongly, 47, 51dualizing τ -dualizing, 108embedding, Segre, 231enlargement, of premotivic categories, see also premotivicequidimensionalabsolutely, 173flat morphism, 173equivalence A -equivalence, 135 H B -equivalence, 241of motivic categories, 258of triangulated monoidal categories, 243, 251strong A -equivalence, 135termwise weak equivalence, 59weak equivalence of commutative monoids, 160weak equivalence of modules, 164weak equivalence of monoids, 160 W -equivalence, 130equivalence, of categories, 23exceptional functor, see also functorexchangeisomorphism, 4, 9, 34, 157–159morphism, see also exchange transformationtransformation, 3, 5, 7, 8, 12, 29fibrant A -fibrant, 135fibration t -fibration, 120algebraic Hopf, 252of commutative monoids, 160of modules, 164of monoids, 160termwise, 59 W -fibration, 130fibredfibred category, 2monoidal pre- P -fibred category, 7monoidal P -fibred category, 8 model (category), 18of finite correspondences, 200 P -fibred category, 5 τ -generated, see also generatedabelian, 15abelian monoidal, 15canonical, 5canonical monoidal, 8complete, 5complete monoidal, 8finitely τ -presented , see also finitely presentedgeometrically generated, see also generatedGrothendieck abelian, 15Grothendieck abelian monoidal, 15homotopy, 19homotopy monoidal, 19model, 18monoidal P -fibred model category, 121triangulated, 16triangulated monoidal, 16pre- P -fibred category, 2filtration, γ -filtration, 240finite correspondence, 194composition, 195finite S -correspondence, see also finitecorrespondencegraph functor, see also functortensor porduct, see also tensor producttranspose, see also morphismfinitely presentedfinitely τ -presented, 16, 128, 141, 151object of a category, 15finiteness theorem, 101, 248flasque, t -flasque complex, 120formalism, Grothendieck 6 functors, 54, 243functorcommutes, 5evaluation, 58, 63, 143exceptional, 30, 55graph, 196infinite suspension, 20left exceptional, 26Quillen, 63 t -exact endofunctor, 253Galois group, see also groupgenerated τ -generated, 10, 20compactly ( Z × τ )-generated, 151compactly τ -generated, 20triangulated P -fibred, 16, 31, 128, 141compactly generated, 16, 248, 258triangulated P -fibred, 16geometrically generated, 10well generated, 16triangulated P -fibred, 16global section, see also sectiongroupChow, 219Galois group, 80 H -group, 231higher Chow, 218Picard, 221relative Picard, 220 NDEX OF TERMINOLOGY 273 henselisation, 104homeomorphism, universal, 23, 189, 219homotopic, A -homotopic, 135homotopycolimit, 61limit, 61object of homotopy fixed points, 83homotopy cartesian, 78, 84, 85, 88, 107, 133object over a diagram, 67square, 77homotopy category, 3, 19, 48, 137homotopy linear, 168homtopy pullback, see also homotopy cartesianhypercover, 68, 132ˇCech t -hypercovers, 203ind-constructible, 186infinite suspension, see also functorK-theoryhomotopy invariant, 232Milnor, 255Milnor-Witt, 255Quillen, 232with support, 234law, formal group, 231linear Q -linear (stable model category), 73strongly Q -linear, 160local, 120 W -local, 130 A -local, 135, 138 H B -local, 241localizationtriangle, seetriangle, 32map, trace, 228, 238model structure t -descent, 121injective (diagrams), 60positive stable model structure, 164projective (diagrams), 59 W -local, 130module H B -module, 242, 244strict H B -module, 244modules KGL -modules, 234over a homotopy cartesian commutative monoid,167, 259over a monoid, 164monoid, 160, 167cartesian, 166, 168cartesian commutative monoid, 142commutative monoid, 160homotopy cartesian, 167monoid axiom, 160, 164, 165, 168monoidal stable homotopy 2-functor, 54morphism T -pure, see also morphism, purecartesian —- of S -diagrams, 64cocontinuous, 73degree, 196faithfully flat, 34 finite Λ-universal, 219Gysin, 225of P -fibred categories, 12of P -fibred model categories, 18of P -premotivic categories, 19, 20of Λ-cycles, 172of abelian P -fibred categories, 15of abelian P -premotivic categories, 19of abelian monoidal P -fibred categories, 15of complete P -fibred categories, 12of derivators, 73of monoidal P -fibred model category, 18of S -diagrams, 58of triangulated P -fibred categories, 16of triangulated P -premotivic categories, 19of triangulated monoidal P -fibred categories, 16of triangulated premotivic categories, 56pseudo-dominant, 172pure, 45, 47, 51pure (proper case), 43Quillen —- of P -fibred model categories, 68radicial, 23, 34separated, 26, 172transpose, 194universally T -pure (proper case), 43motive, 213Beilinson, 241, 255Chow (strong), 226constructible, see also constructibleeffective h-motives, 134effective qfh-motives, 134generalized, 216geometric, 20, 215, 251geometric effective, 215h-motive, 149qfh-motive, 149Morel, 252–257motivic complex, 213constructible, see also constructiblegeneralized, 216stable, 213, see also motivemultiplicitygeometric, 172Samuel (of a cycle), 189Samuel (of a module), 188Suslin-Voevodsky, 181nilpotent, 231Nisnevichdistinguished square, see also distinguishedtopology, see also topologyorientable, 244orientation, 51of a ring spectrum, 230of a triangulated premotivic category, 49, 51perfect, 164, 167perfect pairing, 47Picard category, 42point, 172fat point (of a cycle), 175generic (of a cycle), 172geometric, 172of a cycle, 175
74 INDEX OF TERMINOLOGY pointed, smooth S -scheme, 39prederivator, 72premotive, 19Tate premotive, 42premotiviccase, 22category, 19category of h-motives, 149category of qfh-motives, 149enlargement of —- category, 21, 154, 213, 217generalized —- category, 19morphism, see also morphism of premotiviccategories P -premotivic A -derived category, 134abelian category, 19category, 19derived category, 122stable A -derived category, 146triangulated category, 19stable A -derived premotivic category, 148presentationlocal presentation of a simplicial object, 69presented, see also finitely presentedpresheafΛ-presheaf, 118with transfers, 201projection formula P -projection formula, 8projective system, of schemes, xi, 101, 129, 186, 202,209pseudo-Galois, see also cover or distinguishedpullbackof fundamental class, 246purityabsolute, 237, 246isomorphism (relative), 43, 49, 50, 225quasi-excellent, 91quotientGabriel, 152radicial, see also morphismrealization functor(associated with a stable cohomology), 262Betti, 260de Rham, 264of construcible motives, 263rigid, 265resolution of singularities, 91canonical —- up to quotient singularities, 92canonical dominant —- up to quotientsingularities, 92wide —- up to quotient singularities, 92, 248Riemann-Hilbert, 265ringGrothendieck-Witt, 256schematic closure, 172schemeexcellent, 91, 212, 217geometrically unibranch, 190, 194, 212, 217, 251quasi-excellent, 91, 97, 246regular, 193–195, 214, 222strictly local, 191 unibranch, 190sectionabsolute derived global section, 76cartesian, 142geometric, 10, 12, 36geometric derived global section, 72sequencesymmetric sequence, 143sheaf´etale sheaf with transfers, 202generalized sheaf with transfers, 209h-sheaf, 134, 149qfh-sheaf, 134, 149, 211sheaf with transfers, 202, 209 t -sheaf of Λ-modules, 119 t -sheaf with transfers, 202sieve, 22, 24singularSuslin singular complexe, 140specialization, 226spectra, see also spectrumspectrumabelian Tate spectrum, 144absolute Tate spectrum, 144algebraic cobordism, 230Beilinson motivic cohomology spectrum, 240motivic cohomology ring spectrum, 223, 230rational, 232ring —-, 230ring —- (associated with a stable cohomology),261strict ring —-, 230Tate spectrum, 145Tate Ω-spectrum, 147universal oriented ring —- with additive formalgroup law, 243spheresimplicial, 230square P -distinguished, 24cdh-distinguished, 78, 127Nisnevich distinguished, 24, 76, 127proper cdh-distinguished, 24pseudo-Galois qfh-distinguished, 81qfh-distinguished, 81, 88, 127Tor-independant, 235, 246stable homotopy category of schemes, 19strict transform, 175, 178strictification theorem, 164strongly dualizable, see also dualizableTatemotivic complex, 214twist, see also twist, 230tensor productof finite correspondences, 197Thomadjoint transformation, 39class, 50isomorphism, 49premotive, 41transformation, 39tilting, 259, 262topology NDEX OF TERMINOLOGY 275 admissible, 119cdh-topology, 24compatible with transfers, 203, 208h-topology, 80, 250mildly compatible with transfers, 204, 206, 208Nisnevich, 24 P -admissible, 119proper cdh, 24qfh-topology, 80, 250weakly compatible with transfers, 203tractable, 67, 160traitof a cycle, 175transfer, see also presheaf or sheaftransversal M -transversal square, 6transversality property, 6triangleGysin, 225localization triangle, 32Mayer-Vietoris triangle, 78 t -structureheart, 254non degenerated, 253 t -structure, homotopy, 252twist, 10, 19commutes with τ -twists ( or twists), 10, 11, 13of a triangulated monoidal P -fibred category, 16Tate, 19, 20, 42, 147 τ -twisted, 10underlying simplicial setof a simplicial object, 69universal, 133weak equivalence, see also equivalence ndex of notations α ⊗ ♭S S ′ , 177 α ⊗ S S s , 192 α ⊗ trS α ′ , 200˜ α , 178 A S , 144 β ◦ α , 198 β R,k , 179 β ⊗ α α ′ , 181 h Z i X , 175 c S ( X, Y ) Λ , 197 C ∗ , 141 c ( X/S,
Λ), 197D A ( A ), 147D A ,gm ( A S ), 152D A , Λ , 149D A , Λ , 150D A ( S, Λ) + , 256D Betti ( X ), 264deg x ( f ), 199D eff A ( A ), 135D eff A , Λ , 135DM B ,c ( S ), 251DM B ( S ), 245DM eff gm ( S, Λ), 219DM gm ( S, Λ), 219DM h , Λ , 150DM Λ , 217DM Λ , 219DM eff Λ , 217DM eff Λ , 219DM qfh , Λ , 150 DM eff h , Λ , 135 DM eff qfh , Λ , 135 •−→ , 197D( X, E ), 266 e A q ( M ), 191 H B , 246 H q B ( X, Q ( p )), 247 H n,m M ,eff ( S, Λ), 221 H n,m M ( S, Λ), 221Hom • ( − , − ), 118 H • ( S ), 3 KGL β , 237 KGL ′ , 237 KGL Q , 244 KGL S , 236 KGL ( i ) S , 244Λ tS ( X ), 119Λ trS ( X ), 205Λ tr S ( X ), 206 M an ( X ), 264 M effBetti ( X ), 264 M Betti ( X ), 264 MGL , 234 m SV ( x ; β ⊗ α α ′ ), 184 M S ( X ), 217 M S ( X ), 220 P cor Λ ,S , 199 P cart , 65PSh( P /S, Λ), 118 P cor Λ ,S , 205 R A , 139SH( S ), 19Sh t ( P /S, Λ), 119Sh t (cid:16) P cor Λ ,S (cid:17) , 206Sh tr ( − , Λ), 213Sh tr ( − , Λ), 213Sym( A ), 144 ⊗ S , 144Tot π , 118 t f , 198Tr KGL p , 242Λ tS ( X ), 135 PPENDIX A
Index of properties of P -fibred triangulated categories Name Symbol Def. related Remarkresultadditive 2.1.1adjoint property (Adj) 2.2.13 2.2.14adjoint property for f (Adj f ) 2.2.13 f morphism of schemescotransversality property 1.1.17 defined for any P -fibred categoryhomotopy property (Htp) 2.1.3localization property (Loc) 2.3.2 2.4.266.3.15localization property for i (Loc i ) § i closed immersionmotivic 2.4.45 2.4.50 for premotivic triangulated categories,14.2.11 means: (Htp), (Stab), (Loc), (Adj)oriented 2.4.38 2.4.43 for triangulated premotivic categoriessatisfying (wLoc)projection formula (PF) 2.2.13projection formula for f (PF f ) 2.2.13 2.4.26 f morphism of schemesproper base change property (BC) 2.2.13 2.4.26proper base change property for f (BC f ) 2.2.13 f morphism of schemespurity property (Pur) 2.4.21 2.4.26separated (Sep) 2.1.7 4.2.244.4.2114.3.3semi-separated (sSep) 2.1.7 3.3.33stability property (Stab) 2.4.4support property (Supp) 2.2.5 2.2.122.2.1411.4.2 τ -compatible 4.2.20 4.2.29 τ set of twists τ -continuous 4.3.2 6.1.13 for homotopy P -fibred categories,11.1.24 τ set of twists14.3.1 τ -dualizable 4.4.13 4.4.21 τ set of twists t -descent property 3.2.5 for homotopy P -fibred categories, t topologytransversality property 1.1.17 for any P -fibred category t -separated (t-sep) 2.1.5 t topologyweak localization property (wLoc) 2.4.7 11.4.2weak purity property (wPur) 2.4.21 2.4.262.4.43topologyweak localization property (wLoc) 2.4.7 11.4.2weak purity property (wPur) 2.4.21 2.4.262.4.43