• 検索結果がありません。

Convergence of Voevodsky’s Slice Tower

N/A
N/A
Protected

Academic year: 2022

シェア "Convergence of Voevodsky’s Slice Tower"

Copied!
36
0
0

読み込み中.... (全文を見る)

全文

(1)

Convergence of Voevodsky’s Slice Tower

Marc Levine1

Received: May 7, 2012 Revised: March 3, 2013 Communicated by Alexander Merkurjev

Abstract. We consider Voevodsky’s slice tower for a finite spectrum E in the motivic stable homotopy category over a perfect fieldk. In casekhas finite cohomological dimension, we show that the slice tower converges, in that the induced filtration on the bi-graded homotopy sheavesΠa,bfnEis finite, exhaustive and separated at each stalk (after inverting the exponential characteristic ofk). This partially verifies a conjecture of Voevodsky.

2010 Mathematics Subject Classification: Primary 14F42; Secondary 55P42

Keywords and Phrases: Morel-Voevodsky stable homotopy category, slice filtration, motivic homotopy theory

Contents

Introduction 908

1. Background and notation 911

2. Voevodsky’s slice tower 914

3. The homotopy coniveau tower 917

4. The simplicial filtration 919

5. The bottom of the filtration 923

6. Finite spectra and cohomologically finite spectra 925

7. The proof of the convergence theorem 930

Appendix A. Norm maps 933

Appendix B. Inverting integers in a triangulated category 936

References 939

1Research supported by the Alexander von Humboldt Foundation

(2)

Introduction

We continue our investigation, begun in [12], of the slice filtration on the bi- graded homotopy sheavesΠ∗,∗(E)for objectsE in the motivic stable homotopy categorySH(k). We refer the reader to §1 for the notation to be used in this introduction.

Letkbe a perfect field, letSH(k)denote Voevodsky’s motivic stable homotopy category of T-spectra over k, SH the classical stable homotopy category of spectra. For a spectrum E∈ SH, the Postnikov tower ofE,

. . .→E(n+1)→E(n)→. . .→E

consists of then−1-connected coversE(n)→E ofE, that is,πmE(n)→πmE is an isomorphism for m≥n andπmE(n)= 0 form < n. Sending E to E(n) defines a functor fromSHto the full subcategoryΣnSHef f ofn−1-connected spectra that is right adjoint to the inclusionΣnSHef f → SH.

Replacing ΣnSHef f with a certain triangulated subcategory ΣnTSHef f(k) of SH(k) that measures a kind of “P1-connectedness” (in a suitable sense, see [30, 31], [23, 24]) or §2 of this paper), Voevodsky has defined a motivic analog of the Postnikov tower; for an objectE ofSH(k)this yields theTate-Postnikov tower(or slice tower)

. . .→fn+1E →fnE →. . .→ E

forE. For integersa, b, we have the stable homotopy sheafΠa,b(E), defined as the Nisnevich sheaf associated to the presheaf

U ∈Sm/k7→[ΣaS1ΣbGmΣTU+,E]SH(k)

and the Tate-Postnikov tower forE gives rise to the filtration FilnTateΠa,b(E) := im(Πa,bfnE →Πa,bE).

Let SHfin(k) ⊂ SH(k) be the thick subcategory of SH(k) generated by the objectsΣnTΣT X+, withX smooth and projective overk,n∈Z. For example, the motivic sphere spectrumSk:= ΣT Speck+is inSHfin(k).

Voevodsky has stated the following conjecture:

Conjecture 1 ([30, conjecture 13]). Let k be a perfect field. Then for E ∈ SHfin(k), the Tate-Postnikov tower of E is convergent in the following sense:

for all a, b, m∈Z, one has

nFTaten Πa,bfmE = 0.

The casesE = ΣqGmSk, a=m= 0 gives some evidence for this conjecture, as we shall now explain.

Forka perfect field, Morel has given a natural isomorphism ofΠ0,−p(Sk)with the Milnor-Witt sheaf KMWp ; this is a certain sheaf on Sm/k with value on each fieldF overk given by the Milnor-Witt groupKpMW(F).2 ForF a field, K0MW(F)is canonically isomorphic to the Grothendieck-Witt group GW(F)

2A presentation of the graded ringKM W(F)may be found in [14, definition 3.1].

(3)

of non-degenerate symmetric bilinear forms over F [14, lemma 3.10]. More generally, Morel has constructed a natural isomorphism3forp, q∈Z

Π0,pqGmSk)∼=KMWq−p.

The isomorphismK0MW(F)∼= GW(F)makesKMW(F)a GW(F)-module; let I(F)⊂GW(F)denote the augmentation ideal. Our main result ofloc. cit. is Theorem 2 ([12, theorem 1]). Let F be a perfect field extension ofk of char- acteristic6= 2. Then

FilnTateΠ0,pqGmSk)(F) =I(F)MKq−pMW(F)⊂Kq−pMW(F) = Π0,pqGmSk)(F) whereM = 0ifn≤porn≤q, andM = min(n−p, n−q)ifn≥pandn≥q.

The following consequence of theorem 2 gives some evidence for Voevodsky’s convergence conjecture:

Proposition3. Letk be a perfect field withchar. k6= 2. For allp, q≥0, and all perfect field extensions F ofk, we have

nFTaten Π0,pΣqGmSk(F) = 0.

Proof. In light of theorem 2, this is asserting that theI(F)-adic filtration on Kq−pMW(F)is separated. By [16, théorème 5.3], form≥0,KmMW(F)fits into a cartesian square ofGW(F)-modules

KmMW(F) //

KmM(F)

P f

I(F)m q //I(F)m/I(F)m+1,

whereKmM(F)is the MilnorK-group,qis the quotient map andP f is the map sending a symbol{u1, . . . , um}to the class of the Pfister form<<u1, . . . , um>>

mod I(F)m+1. Form < 0, KmMW(F) is isomorphic to the Witt groupW(F) ofF, that is, the quotient ofGW(F)by the ideal generated by the hyperbolic form x2−y2. Also, the mapGW(F)→W(F)gives an isomorphism ofI(F)r with its image inW(F)for allr≥1. Thus, forn≥1,

I(F)nKmMW(F) =

(I(F)n⊂W(F) form <0 I(F)n+m⊂W(F) form≥0.

The fact that ∩nI(F)n = 0 in W(F) is a theorem of Arason and Pfister [1,

Korollar 1].

Remarks . 1. The proof in [16] that KmMW(F) fits into a cartesian square as above relies on the Milnor conjecture.

2. As pointed out to me by Igor Kriz, Voevodsky’s convergence conjecture in the generality as stated above is false. In fact, takeEto be the Moore spectrum

3This follows from [14, theorem 6.13, theorem 6.40], using the argument of [14, theorem 6.43].

(4)

Sk/ℓfor some primeℓ6= 2. SinceΠa,qSk = 0fora <0, proposition 6.9 below shows thatΠa,qfnSk = 0fora <0, and thus we have the right exact sequence for alln≥0

Π0,0fnSk−−→×ℓ Π0,0fnSk →Π0,0fnE →0.

In particular, we have

FTaten Π0,0E(k) =im(FTaten Π0,0Sk(k)→Π0,0Sk(k)/ℓ) =

=im(I(k)n→GW(k)/ℓ). Take k=R. Then I(R)⊂GW(R)is isomorphic to Zvia the virtual negative index, and I(R)n = (2n−1) ⊂ Z = I(R). Thus Π0,0E = Z/ℓ⊕Z/ℓ and the filtration FTaten Π0,0E is constant, equal to Z/ℓ=I(R)/ℓ, and is therefore not separated.

The convergence property is thus not a “triangulated" one in general, and there- fore seems to be a subtle one. However, if the I-adic filtration on GW(F)is finite for all finitely generatedFoverk(possibly of varying length depending on F), then the augmentation ideal inGW(F)is two-primary torsion. Our compu- tations (at least in characteristic6= 2) show that the filtrationFTate Π0,pΣTG∧qm is in this case at least locally finite, and thus has better triangulated properties.

In particular, forℓ6= 2,

Π0,0(Sk/ℓ) =Z/ℓ, FTaten Π0,0(Sk/ℓ) = 0forn >0.

One can therefore ask if Voevodsky’s convergence conjecture is true if one assumes the finiteness of the I(F)-adic filtration on GW(F) for all finitely generated fields F over k. The main theorem of this paper is a partial answer to the convergence question along these lines.

Theorem 4. Let kbe a perfect field of finite cohomological dimension and let pdenote the exponential characteristic.4 TakeE inSHfin(k)and takex∈X∈ Sm/k withX irreducible. Let d= dimkX. Then for everyr, q, m∈Z, there is an integerN =N(E, r, d, q) such that

(FilnTateΠr,qfmE)x[1/p] = 0

for alln≥N. In particular, ifF is a field extension ofkof finite transcendence dimension doverk, then FilnTateΠr,qfmE(F)[1/p] = 0 for alln≥N.

For a more detailed and perhaps more general statement, we refer the reader to theorem 7.3.

Remarks. 1. The proof of theorem 4 relies on the Bloch-Kato conjecture.

2. As we have seen, Voevodsky’s convergence conjecture is not true for all base fields k. An interesting class of fields strictly larger than the class of fields of finite cohomological dimension is those of finite virtual cohomological dimension (e.g.,R). We suggest the following formulation:

4That is,p= char. kifchar. k >0,p= 1ifchar. k= 0.

(5)

Conjecture5. Letkbe a field of finite virtual 2-cohomological dimension. Then the I(k)-completed slice tower is weakly convergent: afterI(k)-completion, the filtrationFilTateΠr,qfmE is stalkwise separated for eachE inSHfin(k)and each r, q, m.

This modified conjecture is equivalent to Voevodsky’s convergence conjecture in casekhas finite 2-cohomological dimension, as in this caseI(k)is nilpotent.

One could also ask for a weaker version, in which one assumes thatkhas finite p-cohomological dimension for all odd primesp.

3. It would be interesting to be able to say something about the p-torsion in (FilTateΠr,qE)x.

The paper is organized as follows: We set the notation in §1. In §2 we recall some basic facts about the slice tower, the truncation functorsfninSH(k)and SHS1(k), and the associated filtrationFilTateΠa,b. We recall the construction and basic properties of the homotopy coniveau tower, a simplicial model for the slice tower in SHS1(k), in §3. In §4 we use the simplicial nature of the homotopy coniveau tower to analyze the terms in the slice tower. This leads to the main inductive step in our argument (lemma 4.5), and isolates the particular piece that we need to study. This is analyzed further in §5, where we more precisely identify this piece in terms of a KMW -module structure on the bi- graded homotopy sheaves (see theorem 5.3). In §6 we use a decomposition theorem of Morel and results of Cisinski-Déglise to prove some boundedness properties of the homotopy sheaves Πp,qE and their Q-localizations Πp,qEQ

for E in SHfin(k), under the assumption that the base-field k has finite 2- cohomological dimension. In the final section 7, we assemble all the pieces and prove our main result. We conclude with two appendices; the first collects some results on norm maps for finite field extensions that are used throughout the paper and the second assembles some basic facts on the localization of compactly generated triangulated categories with respect to a collection of non-zero integers.

I am grateful to the referee for a number of comments and suggestions for improving an earlier version of this paper and to S. Kelly for discussions on aspects of localization.

1. Background and notation

Unless we specify otherwise,kwill be a fixed perfect base field, without restric- tion on the characteristic. For details on the following constructions, we refer the reader to [5, 7, 8, 14, 15, 17, 18].

We write[n] for the set {0, . . . , n} with the standard order (including[−1] =

∅) and let ∆ be the category with objects [n], n = 0,1, . . ., and morphisms [n]→[m]the order-preserving maps of sets. Given a categoryC, the category of simplicial objects inCis as usual the category of functors∆op→ C, and the category of cosimplicial objects the functor categoryC.

Spc will denote the category of simplicial sets, Spc the category of pointed simplicial sets,H:=Spc[W E−1]the classical unstable homotopy category and

(6)

H :=Spc[W E−1]the pointed version; here W E is the usual class of weak equivalences, that is, morphismsA→Bthat induce an isomorphism on allπn

for all choice of base-point. Sptis the category of spectra, that is, spectrum objects in Spc with respect to the left suspension functor ΣS1 :=S1∧(−).

WithsW E denoting the class of stable weak equivalences, that is, morphisms f :E→F in Sptthat induce an isomorphism on all stable homotopy groups, SH:=Spt[sW E−1]is the classical stable homotopy category.

For a simplicial object in Spc, resp. Spc, resp. Spt, S : ∆op → Spc,Spc,Spt, we let|S| ∈Spc,Spc,Spt denote respective homotopy col- imithocolimopS.

The motivic versions are as follows: Sm/kis the category of smooth finite type k-schemes. Spc(k)is the category ofSpc-valued presheaves onSm/k,Spc(k) theSpc-valued presheaves, and SptS1(k) theSpt-valued presheaves. These all come with “motivic” model structures as simplicial model categories (see for example [8]); we denote the corresponding homotopy categories by H(k), H(k)and SHS1(k), respectively. SendingX ∈Sm/k to the sheaf of sets on Sm/k represented by X (which we also denote by X) gives an embedding of Sm/k to Spc(k); we have the similarly defined embedding of the category of smooth pointed schemes overk into Spc(k). Sending a (pointed) simplicial set A to the constant presheaf with value A (also denoted by A) defines an embedding ofSpcinSpc(k)and ofSpc inSpc(k).

LetGmbe the pointedk-scheme(A1\0,1). We letT :=A1/(A1\ {0})and let SptT(k)denote the category ofT-spectra, i.e., spectra inSpc(k)with respect to the leftT-suspension functor ΣT :=T∧(−). SptT(k)has a motivic model structure (see [8]) and SH(k) is the homotopy category. We can also form the category of spectra in SptS1(k)with respect to ΣT; with an appropriate model structure the resulting homotopy category is equivalent to SH(k). We will identify these two homotopy categories without further mention.

For each A ∈ Spc(k), the suspension functor ΣA : Spc(k) → Spc(k), ΣA(B) :=B ∧ A, extends to the suspension functorΣA:SptS1(k)→SptS1(k) or ΣA:SptT(k)→SptT(k). ForAcofibrant, this gives the suspension func- tors ΣA : H(k) → H(k), ΣA : SHS1(k) → SHS1(k) and ΣA : SH(k) → SH(k)by applyingΣAto a cofibrant replacement.

BothSHS1(k)andSH(k)are triangulated categories with translation functor ΣS1. OnH(k),SHS1(k)andSH(k), we haveΣT ∼= ΣS1◦ΣGm; the suspension functorsΣT andΣGm onSH(k)are invertible. For A ∈Spc(k), we have an enriched Hom onSptS1(k)andSptT(k)with values in spectra; we denote the enriched Hom functor byHom(A,−). This passes to the homotopy categories H(k),SHS1(k)andSH(k)to give forA ∈ H(k)an enriched HomHom(A,−) with values inSH. ForX∈Sm/k,E∈SptS1(k), Hom(X+, E) =E(X).

(7)

We have the triangle of infinite suspension functorsΣand their right adjoints Ω

H(k) Σ

S1

//

ΣT

%%

JJ JJ JJ JJ J

SHS1(k)

ΣT

SH(k)

H(k) SHS1(k)

S1

oo

SH(k)

T

OO

T

eeJJ

JJ J

JJ JJ

both commutative up to natural isomorphism. These are all left, resp. right derived versions of Quillen adjoint pairs of functors on the underlying model categories.

For X ∈ H(k), we have the bi-graded homotopy sheaf Πa,bX, defined for a, b≥0, as the Nisnevich sheaf associated to the presheaf onSm/k

U 7→HomH(k)aS1ΣbGmU+,X);

note the perhaps non-standard indexing. We have the bi-graded homotopy sheaves Πa,bE for E ∈ SHS1(k), b ≥ 0, a ∈ Z, and Πa,bE for E ∈ SH(k), a, b∈Z, by taking the Nisnevich sheaf associated to

U7→HomSHS1(k)aS1ΣbGmΣS1U+, E)orU 7→HomSH(k)aS1ΣbGmΣTU+,E), as the case may be. We write πn forΠn,0; for E ∈SptS1(k)fibrant, πnE is the Nisnevich sheaf associated to the presheafU 7→πn(E(U)).

SH(k)has the set of compact generators

nS1ΣmTΣS1X+, n, m∈Z, X∈Sm/k}

andSHS1(k)has the set of compact generators

nS1ΣmTΣS1X+, n∈Z, m≥0, X ∈Sm/k}.

ForSH(k), this is [4, theorem 9.2]; the proof of this result goes through without change to yield the statement for SHS1(k). As these triangulated categories are both homotopy categories of stable model categories, both admit arbitrary small coproducts.

For F a finitely generated field extension of k, we may view SpecF as the generic point of some X ∈Sm/k (since k is perfect). Thus, for a Nisnevich sheaf S on Sm/k, we may define S(F) as the stalk of S at SpecF ∈ X.

For an arbitrary field extensionF ofk (not necessarily finitely generated over k), we define S(F)as the colimit overS(Fα), as Fα runs over subfields of F containingkand finitely generated overk. For a finitely generated fieldF over k, we consider objects such as SpecF, or AnF as pro-objects inSpc(k)by the usual system of finite-type models; the same holds for related objects such as SpecF+ inH(k)orΣS1SpecF+ inSHS1(k), etc. We extend this to arbitrary field extensions of k by taking the system of finitely generated subfields. We will usually not explicitly insert the “pro-” in the text, but all such objects, as well as morphisms and isomorphisms between them, should be so understood.

(8)

2. Voevodsky’s slice tower

We begin by recalling definition and basic properties of the Tate-Postnikov tower inSHS1(k)and inSH(k). We then define the main object of our study:

the filtration on the bi-graded homotopy sheaves of a T-spectrum or an S1- spectrum induced by the respective Tate-Postnikov towers.

For n≥0, we let ΣnTSHS1(k)be the localizing subcategory of SHS1(k)gen- erated by the (compact) objectsΣmTΣS1X+, withX ∈Sm/kand m≥n. We note that Σ0TSHS1(k) =SHS1(k). The inclusion functor in : ΣnTSHS1(k) → SHS1(k) admits, by results of Neeman [21, theorem 4.1], a right adjointrn; define the functor fn :SHS1(k)→ SHS1(k)byfn :=in◦rn. The co-unit for the adjunction gives us the natural morphism

ρn:fnE→E

forE∈ SHS1(k); similarly, the inclusionΣmTSHS1(k)⊂ΣnTSHS1(k)forn < m gives the natural transformation fmE → fnE, forming the Tate-Postnikov tower

. . .→fn+1E→fnE →. . .→f0E=E;

we define fn := id forn <0. We complete fn+1E → fnE to a distinguished triangle

fn+1E→fnE→snE→fn+1E[1];

this distinguished triangle actually characterizes snE up to unique isomor- phism, hence this defines a distinguished triangle that is functorial inE. The objectsnEis thenth slice ofE.

There is an analogous construction inSH(k): Forn∈Z, let ΣnTSHef f(k)⊂ SH(k)

be the localizing category generated by the T-suspension spectra ΣmTΣT X+, for X ∈ Sm/k and m ≥n; write SHef f(k)for Σ0TSHef f(k). As above, the inclusion in : ΣnTSHef f(k)→ SH(k) admits a right adjoint rn, giving us the truncation functorfn,n∈Z, and the Tate-Postnikov tower

. . .→fn+1E →fnE →. . .→ E.

We define the layer snE by a distinguished triangle as above. For integers N ≥n, we let ρn,N :fN →fn andρn :fn →id denote the canonical natural transformations. We mention the following elementary but useful result.

Lemma2.1. For integersN, n, the diagram of natural endomorphisms ofSH(k) fn◦fN

fnN)

ρn(fN)

//fN

ρN

fn ρn //id

(9)

commutes. Moreover, for N ≥n, the map ρn(fN) is a natural isomorphism, and for N ≤ n, the map fnN) is a natural isomorphism. The same holds with SHS1(k)replacing SH(k).

Proof. The first assertion is just the naturality ofρn with respect to the mor- phismρN :fN →id.

Suppose N ≥ n. Then ΣNTSHef f(k) ⊂ ΣnTSHef f(k) and thus for all E ∈ SH(k), id : fNE → fNE satisfies the universal property of ρn(fNE) : fn(fNE) → fNE, namely, fNE is in ΣnTSHef f(k) and id : fNE → fNE is universal for maps T → fNE with T ∈ ΣnTSHef f(k). Thus, ρn(fNE) is an isomorphism.

If N ≤ n, then for E ∈ SH(k), fn(fNE) is in ΣnTSHef f(k) and ρn(fNE) : fn(fNE)→fNE is universal for maps T →fNE withT ∈ΣnTSHef f(k). Since ΣnTSHef f(k) ⊂ ΣNTSHef f(k), the universal property of ρN(E) : fNE → E shows that ρN(E)◦ρn(fNE) : fn(fNE) → E is universal for maps T → E with T ∈ΣnTSHef f(k), and thusfnN(E))is an isomorphism. The proof for

SHS1(k)is the same.

Lemma 2.2. Forn∈Z, there is a natural isomorphism (2.1) fnT E ∼= ΩT fnE.

Proof. First suppose thatn≥0. It follows from [10, theorem 7.4.1] thatΩT fnE is in ΣnTSHS1(k)and thus we need only show that Ωρn : ΩT fnE → ΩTE satisfies the universal property of fnT E → ΩT E. ΣnTSHS1(k)is generated as a localizing subcategory of SHS1(k) by objectsΣnTG, G∈ SHS1(k), so it suffices to check for objects of this form. We have

HomSHS1(k)nTG,ΩT fnE)∼= HomSH(k)TΣnTG, fnE)

∼= HomSH(k)nTΣTG, fnE) ρn // HomSH(k)nTΣT G,E)

∼= HomSH(k)TΣnTG,E)∼= HomSHS1(k)nTG,ΩTE).

It is easy to check that this sequence of isomorphisms is induced by(ΩT ρn). Now suppose that n <0. Then fnT E ∼=f0TE ∼= ΩT f0E, so it suffices to show that the map f0E → fnE induces an isomorphism ΩT f0E → ΩT fnE.

But forF ∈ SHS1(k),ΣT F is inSHef f(k)and HomSHS1(k)(F,ΩT f0E)∼= HomSH(k)T F, f0E)

ρn,0

// HomSH(k)TF, fnE)∼= HomSHS1(k)(F,ΩT fnE).

ForE∈ SHS1(k), we have (by [10, theorem 7.4.2]) the canonical isomorphism (2.2) ΩrGmfnE∼=fn−rrGmE

(10)

for r≥0. As ΩGm :SH(k)→ SH(k)is an auto-equivalence, and restricts to an equivalence

Gm : ΣnTSHef f(k)→Σn−1T SHef f(k), the analogous identity in SH(k)holds as well, for allr∈Z.

Definition 2.3. For a ∈ Z, b ≥ 0, E ∈ SHS1(k), define the filtration FTate Πa,bE ofΠa,bE by

FTaten Πa,bE:=im(Πa,bfnE→Πa,bE); n∈Z.

Similarly, forE ∈ SH(k),a, b, n∈Z, define

FTaten Πa,bE :=im(Πa,bfnE →Πa,bE).

The main object of this paper is to understandFTaten Πa,bEfor suitableE. For later use, we note the following:

Lemma 2.4. 1. For E ∈ SHS1(k), n, p, a, b ∈ Z with p, b, b−p ≥ 0, the adjunction isomorphism Πa,bE∼= Πa,b−ppGmE induces an isomorphism

FTaten Πa,bE∼=FTaten−pΠa,b−ppGmE.

Similarly, for E ∈ SH(k), n, p, a, b∈Z, the adjunction isomorphism Πa,bE ∼= Πa,b−ppGmE induces an isomorphism

FTaten Πa,bE ∼=FTaten−pΠa,b−ppGmE.

2. ForE ∈ SH(k),a, b, n∈Z, with b≥0, we have a canonical isomorphism ϕE,a,b,n: Πa,bfnE →Πa,bfnTE,

inducing an isomorphism FTaten Πa,bE ∼=FTaten Πa,bTE.

Proof. (1) By (2.2), adjunction induces isomorphisms FTaten Πa,bE:=im(Πa,bfnE→Πa,bE)

∼=im(Πa,b−ppGmfnE→Πa,b−ppGmE)

=im(Πa,b−pfn−ppGmE→Πa,b−ppGmE) =FTaten−pΠa,b−ppGmE.

The proof forE ∈ SH(k)is the same.

For (2), the isomorphismϕE,a,b,narises from (2.1) and the adjunction isomor- phism

HomSHS1(k)aS1ΣbGmΣS1U+,fnT E)

∼= HomSHS1(k)aS1ΣbGmΣS1U+,ΩT fnE)

∼= HomSH(k)aS1ΣbGmΣTU+, fnE).

(11)

3. The homotopy coniveau tower

Our computations rely heavily on our model for the Tate-Postnikov tower in SHS1(k), which we briefly recall (for details, we refer the reader to [10]).

We start with the cosimplicial schemen7→∆n, with∆nthealgebraicn-simplex Speck[t0, . . . , tn]/P

iti−1. The cosimplicial structure is given by sending a mapg: [n]→[m]to the map ∆(g) : ∆n →∆mdetermined by

∆(g)(ti) = (P

j,g(j)=itj ifg−1(i)6=∅

0 else.

Afaceof∆mis a closed subschemeF defined by equationsti1=. . .=tir = 0;

we let∂∆n⊂∆n be the closed subscheme defined byQn

i=0ti = 0, i.e.,∂∆n is the union of all the proper faces.

TakeX ∈Sm/k. We letSX(q)(m)denote the set of closed subsetsW ⊂X×∆m such that

codimX×FW ∩X×F ≥q

for all facesF ⊂∆m (including F = ∆m). We makeSX(q)(m)into a partially ordered set via inclusions of closed subsets. SendingmtoSX(q)(m)andg: [n]→ [m] to∆(g)−1:SX(q)(m)→ SX(q)(n)gives us the simplicial posetSX(q).

Now takeE∈SptS1(k). ForX∈Sm/kand closed subsetW ⊂X, we have the spectrum with supportsEW(X)defined as the homotopy fiber of the restriction map E(X)→ E(X \W). This construction is functorial in the pair (X, W), where we define a mapf : (Y, T)→(X, W)as a morphismf:Y →XinSm/k withf−1(W)⊂T. We usually denote the map induced byf : (Y, T)→(X, W) byf:EW(X)→ET(Y), but forf = idX : (X, T)→(X, W),i:W →T the resulting inclusion, we writei:EW(X)→ET(X)foridX.

Define

E(q)(X, m) := hocolim

W∈SX(q)(m)

EW(X×∆m).

The fact that m 7→ SX(q)(m) is a simplicial poset, and (Y, T) 7→ ET(Y) is a functor from the category of pairs to spectra shows that m 7→ E(q)(X, m) defines a simplicial spectrum. We define the spectrumE(q)(X)by

E(q)(X) :=|m7→E(q)(X, m)|:= hocolim

op E(q)(X,−).

Forq≥q, the inclusionsSX(q)(m)⊂ S(q

)

X (m)induce a map of simplicial posets SX(q)⊂ S(q

)

X and thus a morphism of spectra iq,q:E(q)(X)→E(q)(X). Since E(0)(X,0) =E(X), we have the canonical map

ǫX:E(X)→E(0)(X),

(12)

which is a weak equivalence if E is homotopy invariant. Together, this forms theaugmented homotopy coniveau tower

E(∗)(X) :=. . .→E(q+1)(X)−→iq E(q)(X)−−−→iq−1

. . . E(1)(X)−→i0 E(0)(X)←−−ǫX E(X) with iq := iq,q+1. For homotopy invariant E, this gives us the homotopy coniveau tower inSH

E(∗)(X) :=. . .→E(q+1)(X)−→iq E(q)(X)

iq1

−−−→. . . E(1)(X)−→i0 E(0)(X)∼=E(X).

Letting Sm//k denote the subcategory of Sm/k with the same objects and with morphisms the smooth morphisms, it is not hard to see that sending X to E(∗)(X)defines a functor fromSm//kopto augmented towers of spectra.

On the other hand, forE∈SptS1(k), we have the (augmented) Tate-Postnikov tower

fE:=. . .→fq+1E→fqE→. . .→f0E∼=E

in SHS1(k), which we may evaluate atX ∈Sm/k, giving the towerfE(X) in SH, augmented over E(X).

CallE ∈SptS1(k)quasi-fibrantif, for E→Ef ib a fibrant replacement in the motivic model structure, the map E(X) → Ef ib(X) is a stable weak equiv- alence in Spt for all X ∈ Sm/k. As a general rule, we will represent an E∈ SHS1(k)by a fibrant object inSptS1(k), also denotedE, without making explicit mention of this choice.

As a direct consequence of [10, theorem 7.1.1] we have

Theorem 3.1. Let E be a quasi-fibrant object in SptS1(k), and take X ∈ Sm/k. Then there is an isomorphism of augmented towers inSH

(fE)(X)∼=E(∗)(X)

over the identity on E(X), which is natural with respect to smooth morphisms in Sm/k.

In particular, we may use the modelE(q)(X)to understand(fqE)(X).

Remark 3.2. ForX, Y ∈ Sm/k with given k-points x∈ X(k), y ∈Y(k), we have a natural isomorphism inSHS1(k)

ΣS1(X∧Y)⊕ΣS1(X∨Y)∼= ΣS1(X×Y),

using the additivity of the categorySHS1(k). Thus,ΣS1(X∧Y)is a canonically defined summand ofΣS1(X×Y). In particular forE a quasi-fibrant object of SptS1(k), we have a natural isomorphism in SH

Hom(X∧Y, E)∼= hofib (E(X×Y)→hofib(E(X)⊕E(Y)→E(k))) where the maps are induced by the evident restriction maps. In particular, we may defineE(X∧Y)via the above isomorphism, and our comparison results

(13)

for Tate-Postnikov tower and homotopy coniveau tower extend to values at smash products of smooth pointed schemes overk.

4. The simplicial filtration

In this section, we study the filtration onπrfnE(X)induced by the simplicial structure of the modelE(n)(X).

Lemma 4.1. Let S be a smooth k-scheme, W ⊂S×A1 a closed subset such that p: W →S is finite. Let E ∈ SptS1(k) be quasi-fibrant. Then the map induced by the inclusion i:W →p−1(p(W))induces the zero map

i(EW(S×A1))→π(Ep1(p(W))(S×A1)).

Proof. We steal a proof of Morel’s: LetZ=p(W), and letj0:S×A1→S×P1 be the standard open neighborhood ofS×0 inS×P1. SinceW is finite over S,W is closed inS×P1, so we have the following commutative diagram (4.1) πr(EW(S×P1)) ¯i //

j0

πr(EZ×P1(S×P1))

j0

πr(EW(S×A1)) i

//πr(EZ×A1(S×A1)),

where¯i :W →Z×P1 is the inclusion. Let i :S →S×P1 be the infinity section. SinceW∩S× ∞=∅, the composition

πr(EW(S×P1))−→¯i πr(EZ×P1(S×P1)) i

−−→ πr(EZ×∞(S× ∞)) is the zero map. Letting j :S×A1 →S×P1 be the standard open neigh- borhood ofS× ∞inS×P1, the restriction map

ir(EZ×A1(S×A1))→πr(EZ×∞(S× ∞)) is an isomorphism, hence

j ◦¯ir(EW(S×P1))→πr(EZ×A1(S×A1))

is the zero map. Write˜jfor the inclusions ofS×P1\ {0,∞}intoj0(S×A1) and˜j0 for the inclusions ofS×P1\ {0,∞}intoj(S×A1). Combining (4.1) with the commutativity of the diagram

πr(EW(S×P1)) j

¯i

//

j0

πr(EZ×A1(S×A1))

˜j0

πr(EW(S×A1))

˜j◦i

//πr(EZ×P1\{0,∞}(S×P1\ {0,∞}))

(14)

we see that˜j ◦i= 0. From the long exact localization sequence . . .→πr(EZ×0(S×A1))−−→i0 πr(EZ×A1(S×A1))

˜j

−−→πr(EZ×A1\{0}(S×A1\ {0}))→. . . we see that

ir(EW(S×A1)))⊂i0∗r(EZ×0(S×A1)))⊂πr(EZ×A1(S×A1)).

But i0∗ : πr(EZ×0(S ×A1)) → πr(EZ×A1(S×A1)) is the zero map, since the map i1 : πr(EZ×A1(S×A1))→πr(EZ×1(S×1)) is an isomorphism and

i1◦i0∗= 0.

Lemma 4.2. Suppose F is infinite. Take W ∈ SF(n)(p) and suppose codimp

F(W) > n. Then the canonical map EW(∆pF) → E(n)(SpecF, p) in- duces the zero map on π.

Proof. We identify∆pwithApvia the barycentric coordinatest1, . . . , tp. Sup- pose W has dimension d < p−n. Then d ≤ p−1 and, as F is infinite, a general linear projection L :Ap → Ap−1 restricts toW to a finite morphism W →Ap−1. In addition,W:=L−1(L(W))is inSF(n)(p)forLsuitably general.

Letting i:W →W be the inclusion, it suffices to show that the map iEW(∆pF)→πEW(∆pF)

is the zero map. Via an affine linear change of coordinates on ∆p, we may identify∆pwithAp−1×A1andL:Ap→Ap−1with the projectionAp−1×A1

Ap−1. The result thus follows from lemma 4.1.

Let(∆pF, ∂∆p)(n)be the set of codimensionnpointswof∆pF such that{w}is in SF(n)(p).

Lemma 4.3. Let F be an infinite field. Then the restriction maps EW(∆pF)→ ⊕w∈(∆p

F,∂∆p)(n)∩WEw(SpecOp

F,w) for W ∈ SF(n)(p)defines an injection

πr(E(n)(F, p))→ ⊕w∈(∆p

F,∂∆p)(n)πrEw(SpecOpF,w) for each r∈Z.

Proof. Take W ∈ SF(n)(p). Since ∆pF is affine, we can find a W ∈ SF(n)(p)of pure codimensionnwithW ⊃W: just take a sufficiently general collection of nfunctionsf1, . . . , fn vanishing onW and letW be the common zero locus of thefi. Thus the set of pure codimensionnsubsetsWof∆pFwithW∈ SF(n)(p) is cofinal inSF(n)(p).

(15)

LetW ∈ SF(n)(p)have pure codimensionnon∆pF and letW0⊂W be any closed subset. Then W0 is also in SF(n)(p) and we have the long exact localization sequence

. . .→πrEW0(∆pF)−−−→iW0 πrEW(∆pF)→πrEW\W0(∆pF \W0)→. . . Let SF(n)(p)0 ⊂ SF(n)(p)be the set of all W0 ∈ SF(n)(p) with codimpFW0 > n.

Let

E(n)(F, p)0= hocolim

W0∈S(n)F (p)0

EW0(∆pF).

Passing to the limit over the above localization sequences gives us the long exact sequence

. . .→πrE(n)(F, p)0 i0∗

−−→πrE(n)(F, p)

→ ⊕w∈(∆p

F,∂∆p)(n)πrEw((SpecOp

F,w))→. . . By lemma 4.2, the mapi0∗ is the zero map, which proves the lemma.

Let S : ∆op →Spt be a simplicial spectrum, |S| = hocolimopS ∈ Spt the associated spectrum, giving us the spectral sequence

Ep,q1qS(p) =⇒πp+q|S|.

This spectral sequence induces an increasing filtration Filsimp πr|S| on πr|S|.

We have the q-truncated simplicial spectrum S≤q and Filsimpq πr|S| is just the image of πr|S≤q| in πr|S|. In particular Filsimp−1 πr|S| = 0 and

q=0Filsimpq πr|S| =πrS, so the spectral sequence is weakly convergent, and is strongly convergent if for instance there is an integer q0 such that S(p)is q0-connected for allp.

The isomorphism of theorem 3.1 thus gives us the weakly convergent spectral sequence

(4.2) Ep,q1 (X, E, n) =πqE(n)(X, p) =⇒πp+qfnE(X)

which is strongly convergent if πqE(n)(X, p) = 0for q≤q0, independent ofp.

This defines the increasing filtration Filsimp (E)πrfnE(X) of πrfnE(X)with associated graded grsimpp (E)πrfnE(X) =Ep,r−p .

Lemma 4.4. Suppose thatk is infinite and that Πa,∗E(K) = 0 for a < 0 and all fields K overk. LetF ⊃k be a field extension ofk. Then

(1) Ep,r−p1 (F, E, n) = 0 for p > r + n and Filsimpr+n(E)πrfnE(F) = πrfnE(F).

(2) Ep,q1 (F, E, n) is isomorphic to a subgroup of

w∈(∆p

F,∂∆p)(n)Πq+n,nE(F(w)).

(3) The spectral sequence (4.2)is strongly convergent.

Proof. Since the spectral sequence is weakly convergent, to prove (3) it suffices to show that Ep,q1 (F, E, n) = 0 forq < −nand for (1) it suffices to show that

(16)

Ep,r−p1 = 0forp > r+n. These both follows from (2) as our hypothesis implies that Πa,∗E(F(w)) = 0fora <0,w∈∆pF.

For (2), lemma 4.3 gives us an inclusion Ep,q1qE(n)(F, p)⊂ ⊕w∈(∆p

F,∂∆p)(n)πqEw(SpecOp

F,w).

Take w ∈ (∆pF, ∂∆p)(n). By the Morel-Voevodsky purity isomorphism [18, loc.cit.], we haveEw(SpecOp

F,w)∼=Hom(ΣnS1ΣnGmw+, E), hence πqEw(SpecOp

F,w)∼= Πq+n,nE(F(w)),

which proves (2).

For a field extensionKofk, we write tr. dimkKfor the transcendence dimension ofK overk.

Lemma 4.5. Let E be in SHS1(k)and suppose Πa,∗E(K) = 0 for a <0 and all fields K over k. Let pbe the exponential characteristic of k. Let r be an integer. Suppose we have functions

(d, q)7→Nj(d, q;E)≥0; d, q≥0, j= 0, . . . , r−1,

such that, for each field extension K of k with tr. dimkK ≤ d, each j = 0, . . . , r−1, and all integersq, M ≥0,m≥Nj(d, q;E), we have

(4.3) FTatem Πj,qfME(K)[1/p] = 0.

Let F be a field extension of k with tr. dimkF ≤ d and fix an integern ≥0.

Forr >0, letN = maxr−1j=0Nj(r−j+d, n;E); for r≤0, setN = 0. Then for all integersm≥N,n≥0, , we have

πr(fn(fmE))(F))[1/p] = Filsimpn (fmE)πr(fn(fmE))(F)[1/p].

Proof. We delete the “[1/p]" from the notation in the proof, using the conven- tion that we have inverted the exponential characteristicpthroughout.

Ifkis a finite field, fix a primeℓand letkbe the union of allℓ-power extensions of k. If we know the result for k, then using proposition A.1(2) for each k ⊂ k ⊂ k with k finite over k proves the result for k, after inverting ℓ.

Doing the same for someℓ6=ℓ, we reduce to the case of an infinite fieldk.

By [12, proposition 3.2], the hypothesis Πa,∗E(K) = 0 for a < 0 and all K implies Πa,∗fmE(K) = 0 for a < 0, all K, and all m ≥ 0. In particular, πr(fn(fmE))(F) = 0 for r < 0 and all n, m ≥ 0, so for r < 0, the lemma is trivially true. We therefore assume r ≥ 0. In addition, it follows from lemma 4.4(1) that the spectral sequence (4.2), E∗,∗ (F, fmE, n), is strongly convergent for all m, n≥ 0 and Ep,r−p1 (F, fmE, n) = 0 for p > r+n. Since Filsimp (fmE)πr(fn(fmE))(F)is by definition the filtration onπr(fn(fmE))(F) induced by this spectral sequence, we need only show that

Ep,r−p1 (F, fmE, n) = 0forn < p≤r+nandm≥N.

In particular, the result is proved forr= 0; we now assumer >0.

(17)

Letpbe an integer,n < p≤r+n. By lemma 4.4(2), Ep,r−p1 (F, fmE, n)⊂ ⊕w∈(∆p

F,∂∆p)(n)Πr−p+n,nfmE(F(w)).

Forw∈(∆pF, ∂∆p)(n),whas codimensionnon∆pF, hence tr. dimFF(w) =p−n and thus tr. dimkF(w)≤p−n+d. We havem≥Nr−p+n(p−n+d, n;E)since 0≤r−p+n < r, and so our hypothesis (4.3) implies

FTatem Πr−p+n,nfmE(F(w)) = 0.

But FTatem Πa,bfmE(F(w)) = im(Πa,bfmfmE(F(w)) −−−−−−−→ρm(fmE) Πa,bfmE(F(w))) by definition, and ρm(fmE)is an isomorphism (lemma 2.1), hence

FTatem Πa,bfmE(F(w)) = Πa,bfmE(F(w)).

Thus Πr−p+n,nfmE(F(w)) = 0 and hence Ep,r−p1 = 0 forn < p ≤r+n, as

desired.

5. The bottom of the filtration

In this section, k will be a fixed perfect base field. We study the subgroup Filsimpn (E)πrfnE(F)isolated in lemma 4.5.

Lemma5.1. LetEbe inSHS1(k). ThenFilsimpn−1(E)πrfnE(F) = 0for all fields F overk.

Proof. For anyX ∈Sm/k,Filsimpq πrE(n)(X)is by definition the image of πr|E(n)(X,− ≤q)| →πrE(n)(X),

where E(n)(X,− ≤ q) is the q-truncation (or q-skeleton) of E(n)(X,−). For X = SpecF, we clearly haveSF(n)(p) =∅forp < n, as∆pFhas no closed subsets of codimension > p. Thus |E(n)(X,− ≤q)| is the 0-spectrum for q < nand

hence Filsimpn−1 πrE(n)(F) = 0.

To study the first non-zero layerFilsimpn (E)πrfnE(F)in Filsimp (E)πrfnE(F), we apply the results of [12]. For this, we recall some of these results and constructions.

We letVn= (∆1F\∂∆1)n. The function−t1/t0on∆1gives an open immersion ρn:Vn→AnF, identifying Vn with(A1F\ {0,1})n.

Suppose that E is an n-fold T-loop spectrum, that is, there is an object ω−nT E ∈ Spt(k) and an isomorphism E ∼= ΩnTωT−nE in SHS1(k). Given an n-fold deloopingωT−nE ofE, we have explained in [12, §5] how to construct a

“transfer map”

TrF(w)/FE(w)→πE(F), for each closed pointw∈AnF, separable overF.

If nowE = ΩT E for someT-spectrum E ∈ SH(k), then the bi-graded homo- topy sheavesΠ∗,∗Eadmit a canonical right action by the bi-graded homotopy sheaves of the sphere spectrum Sk ∈ SH(k):

Πa,bE⊗Πp,q(Sk)→Πa+p,b+qE

参照

関連したドキュメント

In the case, say, of showing that a genus 2 algebraically slice knot is not concordant to a knot of genus 1, we have to prove that it is not concordant to any knot in an innite

Kraaikamp [7] (see also [9]), was introduced to improve some dio- phantine approximation properties of the regular one-dimensional contin- ued fraction algorithm in the following

We first prove that a certain full subcategory of the category of finite flat coverings of the spectrum of the ring of integers of a local field equipped with coherent

The coefficients for the recursion for A(2m, 2m) given in Theorem 7.9 are as follows.. A proof of the finite filling conjecture. Differ- ential Geom. Every nontrivial knot in S 3

Thus as a corollary, we get that if D is a finite dimensional division algebra over an algebraic number field K and G = SL 1,D , then the normal subgroup structure of G(K) is given

Theorem 5 (strongly visible ⇒ multiplicity-free). The slice plays a crucial role when we formulate a multiplicity-free theorem in the vector bundle case, as we have seen in Theorem

In Subsection 2.9, we proved that there is an adjunction S R : sFib 0 → rCat 0 between the category of stable meet semilattice fibrations and the category of restriction

On the other hand, conjecture C for a smooth projective variety over a finite field allows to compute the Kato homology of X s in (1-3), at least in the case of semi- stable