• 検索結果がありません。

The Chow-Witt ring

N/A
N/A
Protected

Academic year: 2022

シェア "The Chow-Witt ring"

Copied!
38
0
0

読み込み中.... (全文を見る)

全文

(1)

The Chow-Witt ring

Jean Fasel

Received: May 18, 2006 Communicated by Ulf Rehmann

Abstract.

WedenearingstrutureonthetotalChow-Wittgroup

of anyintegralsmoothsheme overa eld ofharateristidierent

from

2

.

2000 Mathematis Subjet Classiation: 14C15, 14C17, 14C99,

14F43

Keywords and Phrases: Chow-Witt groups, Chow groups,

Grothendiek-WittgroupsandWittgroups

Contents

1 Introduction 276

1.1 Conventions . . . 278

2 Preliminaries 278

2.1 Wittgroups . . . 278

2.2 Produts. . . 279

2.3 Supports. . . 281

3 Chow-Witt groups 283

4 The exterior product 290

5 Intersection with a smooth subscheme 296

5.1 TheGysin-Wittmap . . . 296

5.2 Funtoriality . . . 300

6 The ring structure 304

7 Basic properties 307

(2)

1 Introduction

Let

A

beaommutativenoetherianringofKrulldimension

n

and

P

aprojetive

A

-module ofrank

d

. Onean askthefollowingquestion: does

P

admit afree

fator of rank one? Serre proved a long time ago that the answer is always

positivewhen

d > n

. Soinfattherstinterestingaseiswhen

P

isprojetive

ofrankequaltothedimensionof

A

. Supposenowthat

X

isanintegralsmooth

shemeoveraeld

k

ofharateristinot

2

. Todealwiththeabovequestion,

Bargeand Morel introdued theChow-Wittgroups

CH g j (X)

of

X

(alled at

thattimegroupesdeChowdesylesorientés ,see[BM℄)andassoiatedtoeah

vetor bundle

E

of rank

n

an Euler lass

˜ c n (E)

in

CH g n (X )

. It was proved

reently that if

X =

Spe

(A)

we have

˜ c n (P ) = 0

if and only if

P ≃ Q ⊕ A

(see[Mo℄ for

n ≥ 4

, [FS℄ for

n = 3

and [BM℄ or [Fa℄ for the ase

n = 2

). It

is therefore important to provide moretools, suh as a ring struture and a

pull-bakforregular embeddings, to omputetheChow-Witt groupsand the

Eulerlasses.

Todene

CH g p (X )

onsiderthebreprodutoftheomplexinMilnorK-theory

0 // K p M (k(X )) //

M

x 1 ∈X (1)

K p−1 M (k(x 1 )) // . . . //

M

x n ∈X (n)

K p−n M (k(x n )) // 0

andtheGersten-Wittomplexrestritedto thefundamental ideals

0 // I p (k(X )) //

M

x 1 ∈X (1)

I p−1 (O X,x 1 ) // . . . //

M

x n ∈X (n)

I p−n (O X,x n ) // 0

overthequotientomplex

0 // I p /I p+1 (k(X )) // . . . //

M

x n ∈X (n)

I p−n /I p+1−n (O X,x n ) // 0.

Thegroup

CH g p (X)

isdenedasthe

p

-thohomologygroupofthisbreprod-

ut. Roughlyspeaking,anelementof

CH g p (X)

isthelassofasumofvarieties

ofodimension

p

with aquadratiform dened on eah variety. Weoviously

haveamap

CH g p (X) → CH p (X)

.

UsingthefuntorialityofthetwoomplexesweseethattheChow-Wittgroups

satisfy good funtorial properties (see [Fa℄). For example, we have a pull-

bak morphism

f : CH g j (X) → CH g j (Y )

assoiated to eah at morphism

f : Y → X

and a push-forward morphism

g ∗ : CH g j (Y, L) → CH g j+r (X )

assoiatedtoeahpropermorphism

g : Y → X

,where

r =

dim

(X ) −

dim

(Y )

and

L

is a suitableline bundle over

Y

. Using this funtorialbehaviour, it is

possibletoprodueagoodintersetiontheory. Thisiswhatwedointhispaper

(3)

usingthe lassialstrategy(see forexample [Fu℄ or [Ro℄). First wedene an

exteriorprodut

CH g j (X) × CH g i (Y ) → CH g i+j (X × Y )

andthenaGysin-likehomomorphism

i ! : CH g d (X) → CH g d (Y )

assoiatedtoa

losedembedding

i : Y → X

ofsmoothshemes. Theprodutisthendened

astheomposition

CH g j (X) × CH g i (X) // CH g i+j (X × X)

!

// g CH i+j (X)

where

△ : X → X × X

is the diagonal embedding. To dene the exterior

produt, we rst note that Rost already dened an exterior produt on the

homology of the omplex in Milnor K-theory ([Ro℄). Thus it is enough to

deneanexteriorprodutonthehomologyof theGersten-Wittomplexand

showthat bothexteriorprodutsoinideoverthequotientomplex. Weuse

theusualprodutonderivedWittgroups([GN℄)and showthat this produt

passestohomologyusingtheLeibnitz ruleprovedbyBalmer(see[Ba2℄).

The denition of the Gysin-like map is done by following the ideas of Rost

([Ro℄). It usesthedeformationto thenormaloneto modiyanylosed em-

bedding to a nier losed embedding and uses also the long exat sequene

assoiatedtoatriple

(Z, X, U )

where

Z

isalosedsubsetof

X

and

U = X \ Z

.

The produt that we obtain has the meaning of interseting varieties with

quadratiformsdenedonthem. Itisthereforenotasurprisethatthenatural

map

CH g tot (X) → CH tot (X )

turnsoutto bearinghomomorphism. There is howeverasurprise: theprodutthatweobtainisapriorineitherommutative

norantiommutative. This omes from the fat that the produt oftriangu-

latedGrothendiek-Wittgroups

GW i × GW j → GW i+j

doesnotsatisfyany

ommutativityproperty.

Theorganizationof this paperis as follows: Insetion 2,wereall someba-

si results on triangular Witt groups. This inludes the onstrution of the

Gersten-Witt omplex, and some results on produts and onsanguinity. In

setion3, weonstrut theChow-Wittgroups, reall someresults and prove

somebasifats. Thedenition oftheexteriorproduttakesplaein setion

4and thedenitionof theGysin-Wittmapin setion5. Inthis part,wealso

prove thefuntorialityof this map. Finallyweput all thepiees togetherin

setion6andprovesomebasiresultsinsetion7.

I would like to thank Paul Balmer, Stefan Gille and Ivo Dell'Ambrogio for

several areful readings of earlier versions of this work. I also would like to

thankthereferee forsomeuseful omments. Thisresearh was supported by

SwissNationalSieneFoundation,grantPP002-112579.

(4)

1.1 Conventions

All shemes are smooth and integral over aeld

k

of harateristi dierent from

2

,orareloalizationsofsuhshemes. Foranytwoshemes

X

and

Y

we

willalwaysdenoteby

X × Y

thebreprodut

X ×

Spe

(k) Y

.

2 Preliminaries 2.1 Witt groups

WereallheresomebasifatsonWitt groupsoftriangulatedategories fol-

lowingtheexpositionof[Ba2℄. Wesupposethatforanytriangulatedategory

C

and anyobjets

A, B

of

C

thegroupHom

(A, B)

isuniquely

2

-divisible. We alsosupposethatalltriangulatedategoriesareessentiallysmall.

Definition

2.1

.

Let

C

beatriangulatedategory. A dualityon

C

is atriple

(D, δ, ̟)

where

δ = ±1

,

D : C → C

is a

δ

-exat ontravariant funtor and

̟ : 1 ≃ D 2

is an isomorphismof funtors satisfying

D(̟ A ) ◦ ̟ DA = id DA

and

T (̟ A ) = ̟ T A

for all

A ∈ C

. A triangulatedategory

C

with aduality

(D, δ, ̟)

iswritten

(C, D, δ, ̟)

.

Example 2.2 . Let

X

bearegularshemeand

P(X )

theategoryofloallyfree

oherent

O X

-modules. Let

D b (P (X))

bethetriangulatedategoryofbounded omplexesof objetsof

P (X )

. Then theusual duality

on

P(X )

dened by

P =

Hom

O X (P, O X )

induesa

1

-exatdualityon

D b (P(X))

. Wealsodenote

thisderiveddualityby

. Moreover,theanonialisomorphism

ev : P → P ∨∨

for any loally free module

P

indues a anonial isomorphism

̟ : 1 → ∨∨

in

D b (P (X))

. More generally, if

L

is anyinvertiblemodule over

X

,then the

dualityHom

O X (

_

, L)

on

P (X )

alsoinduesadualityon

D b (P (X))

.

Definition

2.3

.

Let

(C, D, δ, ̟)

beatriangulatedategorywithduality. For any

i ∈ Z

, dene

(D (i) , δ (i) , ̟ (i) )

by

D (i) = T i ◦ D

,

δ (i) = (−1) i δ

and

̟ (i) = δ i (−1) i(i+1)/2 ̟

. It iseasy tohekthat

(D (i) , δ (i) , ̟ (i) )

isadualityon

C

. It

isalledthe

i th

-shifteddualityof

(D, δ, ̟)

.

Definition

2.4

.

Let

(C, D, δ, ̟)

beatriangulatedategorywithduality,

A ∈ C

and

i ∈ Z

. Amorphism

ϕ : A → D (i) A

is

i

-symmetriifthefollowingdiagram

ommutes:

A ϕ //

̟ A (i)

D (i) A

(D (i) ) 2 (A)

D (i) ϕ // D (i) A.

Theouple

(A, ϕ)

isalledan

i

-symmetripair.

Definition

2.5

.

Wedenote by

Symm i (C)

the monoid ofisometry lasses of

i

-symmetripairs,equippedwiththeorthogonalsum.

(5)

Definition

2.6

.

An

i

-symmetri formis an

i

-symmetripair

(A, ϕ)

where

ϕ

isanisomorphism.

Theorem

2.7

.

Let

(C, D, δ, ̟)

bea triangulated ategory with duality andlet

(A, φ)

be an

i

-symmetripair. Choosean exattriangle ontaining

φ A φ // D (i) A α // C β // T A.

Then there exists an

(i + 1)

-symmetri isomorphism

ψ : C → D (i+1) C

suh

thatthe following diagram ommutes

A φ //

̟ (i)

D (i) A α // C β //

ψ

T A

T ̟ (i)

D (i) (D (i) A)

D (i) φ // D (i) A

δ (i+1) D (i+1) β // D (i+1) C

D (i+1) α // T (D (i) (D (i) A))

where the rows are exat triangles and the seondone is the dual of the rst.

Moreover, the

(i + 1)

-symmetri form

(C, ψ)

is unique up to isometry. It is

denotedby one

(A, φ)

.

Proof. See[Ba1℄,Theorem1.6.

Example 2.8 . Let

A ∈ C

. Forany

i

,themorphism

0 : A → D (i) A

issymmetri

andthenone

(A, 0)

iswelldened.

Corollary

2.9

.

The above onstrution gives awell denedhomomorphism ofmonoids

d i : Symm (i) (C) → Symm (i+1) (C)

suhthat

d i+1 d i = 0

.

Definition

2.10

.

Let

(C, D, δ, ̟)

beatriangulatedategorywithduality. The Wittgroup

W i (C)

isdenedasKer

(d i )/

Im

(d i+1 )

. RemarkthatKer

(d i )

isjust

themonoidofisometry lassesof

i

-symmetriforms.

Definition

2.11

.

Let

(C, D, δ, ̟)

beatriangulatedategorywithduality. The Grothendiek-Wittgroup

GW i (C)

isdenedasthequotientofKer

(d i )

bythe

submonoidgeneratedbytheelementsone

(A, φ) −

one

(A, 0)

where

A ∈ C

and

φ

is

(i − 1)

-symmetri(

0

isalsoseenas an

(i − 1)

-symmetrimorphism).

Example 2.12 . Let

(D b (P (X)), , 1, ̟)

bethetriangulatedategorywithdu- alitydened in Example2.2. Its Witt groupsare theWitt groups

W i (X)

of

thesheme

X

asdened in[Ba1℄.

2.2 Products

Given apairing

⊗ : C × D → M

of triangulatedategorieswith duality and assumingthat this pairingsatises somenie onditions,theauthors of [GN℄

deneapairingofWittgroups. Webrieyreallsomedenitions(see1.2and

1.11in [GN℄):

(6)

Definition

2.13

.

Let

C, D

and

M

be triangulated ategories. A produt between

C

and

D

withodomain

M

isaovariantbi-funtor

⊗ : C × D → M

exatin both variables and satisfying the following ondition: the funtorial

isomorphisms

r A,B : A ⊗ T B ≃ T(A ⊗ B)

and

l A,B : T A ⊗ B ≃ T (A ⊗B )

make

thediagram

T A ⊗ T B l A,T B //

r T A,B

T (A ⊗ T B)

T(r A,B )

T (T A ⊗ B)

T(l A,B ) // T 2 (A ⊗ B)

skew-ommutative.

Definition

2.14

.

Let

C, D

and

M

betriangulated ategories with dualities.

Wherethere is nopossibleonfusion, we dropthe subsriptsfor

D, δ

and

̟

.

A dualizing pairing between

C

and

D

with odomain

M

is a produt

with

isomorphisms

η A,B : DA ⊗ DB ≃ D(A ⊗ B)

naturalin

A

and

B

whihmakethefollowingdiagramsommute

1.

A ⊗ B ̟ A ⊗̟ B //

̟ A⊗B

D 2 A ⊗ D 2 B

η DA,DB

D 2 (A ⊗ B)

D(η A,B ) // D(DA ⊗ DB)

2.

T(DT A ⊗ DB)

δ C δ M T(η T A,B )

DA ⊗ DB

l DT A,DB

oo

η A,B

r DA,DT B

// T(DA ⊗ DT B)

δ L δ M T (η A,T B )

T D(T A ⊗ B) D(A ⊗ B)

T D(l A,B )

oo T D(r A,B ) // T D(A ⊗ T B).

Theorem

2.15

.

Let

C, D

and

M

be triangulated ategories with duality. Let

⊗ : C × D → M

be a dualizing pairing between

C

and

D

with odomain

M

.

Then

indues forall

i, j ∈ Z

apairing

⋆ : W i (C) × W j (D) → W i+j (M).

(7)

Proof. See[GN℄,Theorem2.9.

Example2.16 . Let

(D b (P(X )), , 1, ̟)

bethetriangulatedategorywithdual- itydenedinExample2.2. Theusualtensorprodutinduesadualizingpair-

ingoftriangulatedategoriesandthenaprodut

W i (X )×W j (X ) → W i+j (X )

.

Suppose that

L

and

N

are invertible modules over

X

. Then Hom

O X (

_

, L)

,

Hom

O X (

_

, N)

andHom

O X (

_

, L ⊗ N )

givedualities

,

and

on

D b (P (X ))

.

Thetensorprodutgivesadualizingpairing

⊗ : (D b (P (X)), , 1, ̟) × (D b (P (X)), , 1, ̟) → (D b (P (X)), , 1, ̟).

2.3 Supports

Webriey reall thenotion of triangulated ategorywith supports following

[Ba2℄.

Definition

2.17

.

Let

X

beatopologialspae. A triangulatedategoryde- ned over

X

is apair

(C,

Supp

)

where

C

is atriangulatedategoryand Supp assigns to eah objet

A ∈ C

a losed subset Supp

(A)

of

X

suh that the

followingrulesaresatised:

(S1) Supp

(A) = ∅ ⇐⇒ A ≃ 0

.

(S2) Supp

(A ⊕ B ) =

Supp

(A) ∪

Supp

(B)

.

(S3) Supp

(A) =

Supp

(T A)

.

(S4) For everydistinguishedtriangle

A // B // C // T A

wehaveSupp

(C) ⊂

Supp

(A) ∪

Supp

(B)

.

When

I

isasaturatedtriangulatedsubategoryof

C

and

S

isthemultipliative systemofmorphismswhose oneis in

I

,then wean onstrutasupport on

theategory

S −1 C := C/I

. Thisisdonein[Ba3℄when

C

hasatensorprodut.

Howeverwewillonlyneedsomebasifats,soweprovethefollowinglemma:

Lemma

2.18

.

let

C

be a triangulated ategory dened over a topologial spae

X

. Let

I

beasaturatedtriangulatedsubategory of

C

andletSupp

(I) =

∪ A∈I

Supp

(A)

. Supposethat Supp

(A) ⊂

Supp

(I)

implies

A ∈ I

. Let

S

bethe

multipliative systemin

C

ofmorphisms

f

suhthat one

(f ) ∈ I

andlet

I // C // C/I

bethe exat sequene of triangulatedategories obtainedby inverting

S

. Then

C/I

isatriangulatedategorydenedover

X = X \

Supp

(I )

(withtheindued

topology).

(8)

Proof. WedeneSupp

S (A) :=

Supp

(A) ∩ X

foranyobjet

A ∈ C/I

andshow

thatSupp

S

satisesthepropertiesofDenition 2.17. Itiseasytoseethatthe

rules(S1), (S2)and(S3) aresatised. Weonlyhavetoprove(S4).

Firstobservethat if

s : A → B

isamorphismin

S

and

A s // B // C // T A

isan exattriangle in

C

ontaining

s

, then Supp

S (A) =

Supp

S (B)

(use(S4)

fortheategory

C

). ThisshowsthatSupp

S (A) =

Supp

S (A )

if

A ≃ A

in

C/I

.

Bydenitionofthetriangulationof

C/I

,anyexattriangle

A α // B // C // T A

in

C/I

is isomorphito theloalization ofanexattrianglein

C

. This shows

thatSupp

S (C) ⊂

Supp

S (A) ∪

Supp

S (B)

.

Example 2.19 . Let

D b (P (X))

be theusual triangulated ategory. Dene the support of an objet

P ∈ D b (P(X ))

as the union of the support of all the

ohomologygroupsof

P

,i.e

Supp

(P) = [

i

Supp

(H i (P )).

Thenit is easyto see that

(D b (P(X )),

Supp

)

isatriangulated ategorywith support. Denoteby

D b (P(X )) (k)

thefullsubategoryof

D b (P (X ))

ofobjets

whosesupportisofodimension

≥ k

. Then

D b (P(X )) (k)

isasaturatedtrian-

gulatedategoryandthefollowingsequene

D b (P (X)) (k) → D b (P (X)) → D b (P(X ))/D b (P (X)) (k)

satisestheonditionsofLemma 2.18. So

D b (P (X))/D b (P (X )) (k)

is atrian-

gulatedategoryover

X = {x ∈ X |

odim

(x) ≤ k − 1}

.

Thefollowingdenitions arealsoduetoBalmer(see[Ba2℄):

Definition

2.20

.

Let

(C,

Supp

)

beatriangulatedategoryover

X

andassume

that

C

hasastrutureoftriangulatedategorywithduality

(C, D, δ, ̟)

. Then

wesaythat

C

is atriangulatedategorywithdualitydenedover

X

if

(S5) Supp

(A) =

Supp

(DA)

foreveryobjet

A

.

Definition

2.21

.

Let

(C,

Supp

C )

,

(D,

Supp

D )

and

(M,

Supp

M )

be triangu-

latedategoriesdenedover

X

. Supposethat

⊗ : C × D → M

isapairingoftriangulatedategories. Thepairing

isdened over

X

if

(S6) Supp

M (A ⊗ B) =

Supp

C (A) ∩

Supp

D (B).

(9)

Example 2.22 . Thetriangulatedategory

D b (P (X))

withthesupportdened

inExample2.19andthepairingofExample2.16satisfytheondition(S5)and

(S6).

Definition

2.23

.

Thedegeneray lousofasymmetripair

(A, α)

isdened

tobethesupportoftheoneof

α

:

DegLo

(α) =

Supp

(

one

(α)).

Definition

2.24

.

Let

(C,

Supp

)

beatriangulatedategorywithdualitydened over

X

. The onsanguinityof two symmetripairs

α

and

β

is dened to be

thefollowingsubsetof

X

:

Cons

(α, β) = (

Supp

(α) ∩

DegLo

(β)) ∪ (

DegLo

(α) ∩

Supp

(β)).

Wearenowready tostatetheLeibnitzformula:

Theorem

2.25 (Leibnitz formula)

.

Assume that we have a dualizing pairing

⊗ : C × D → F

of triangulatedategories with dualities over

X

. Let

α

and

β

betwo symmetri pairs suhthat DegLo

(α) ∩

DegLo

(β) = ∅

. Then we have

anisometry

δ F · d(α ⋆ β) = δ C · d(α) ⋆ β + δ D · α ⋆ d(β)

where

δ C , δ D , δ F

arethe signsinvolvedinthe dualitiesof

C, D

and

F

.

Proof. See[Ba2℄,Theorem5.2.

3 Chow-Witt groups

Let

(D b (P (X)), , 1, ̟)

be the triangulated ategory with the usual duality ofExample 2.2and onsider itsfull subategory

D b (P (X )) (i)

ofobjetswith

supports of odimension

≥ i

(here we use the support dened in Example

2.19).Thenthedualityon

D b (P (X))

induesdualitieson

D b (P (X)) (i)

forany

i

([Ba1℄). Itisalsolearthat

D b (P(X )) (i+1) ⊂ D b (P (X)) (i)

forany

i

.

Definition

3.1

.

For all

i ∈ N

, denote by

D b i (X )

the triangulated ategory

D b (P (X)) (i) /D b (P (X)) (i+1)

.

Suppose that

(A, α)

is an

i

-symmetri form in

D b i (X )

. Then there exists an

i

-symmetripair

(B, β)

suhthattheloalizationof

(B, β)

is

(A, α)

(byloal-

izationwemeanthemap

Symm i (D b (P(X)) (i) ) → Symm i (D b i (X ))

induedby

thefuntor

D b (P (X)) (i) → D b i (X )

). Applying2.7,wegetan

(i + 1)

-symmetri

form

(C, ψ)

. Byonstrution,

C ∈ D b (P (X)) (i+1)

. Loalizingthisformweget

aform

(C, ψ)

in

W i+1 (D b i+1 (X))

. Atrstsight,thisonstrutiondependson

somehoiesbutin fatthisisnotthease(see[Ba1℄,Corollary4.16). Hene

wegetawelldened homomorphism

d i : W i (D b i (X )) → W i+1 (D b i+1 (X)).

(10)

Theorem

3.2

.

Let

X

be a regular sheme of dimension

n

. Then we have a

omplex

0 // W 0 (D 0 b (X)) d

0

// W 1 (D b 1 (X )) d

1

// . . . d n // W n (D b n (X )) // 0.

Proof. See[BW ℄,Theorem3.1 andParagraph8.

Let

A

bearegularloal ring. Wedenoteby

W f l (A)

theWittgroupof nite

lengthmodulesover

A

(see[QSS℄ formoreinformationsaboutWitt groupsof nitelengthmodules). Thefollowingpropositionholds:

Proposition

3.3

.

Wehaveisomorphisms

W i (D b i (X )) ≃ M

x∈X (i)

W f l (O X,x ).

Proof. See[BW ℄,Theorem6.1 andTheorem6.2.

Remark 3.4 . Sineweusetheisomorphismoftheaboveproposition,webriey

reallhowtoobtainasymmetriomplexfromanitelengthmodule. Formore

details,see[BW℄ or [Fa℄, Chapter3. Chooseapoint

x ∈ X (i)

, anitelength

O X,x

-module

M

and a symmetri isomorphism

φ : M →

Ext

i O X,x (M, O X,x )

.

Let

P •

be aresolutionof

M

byloally freeoherent

O X,x

-modules. Then

P •

anbehosenoftheform

0 // P i // . . . // P 0 // M // 0.

Dualizingthisomplexandshifting

i

timesgivesthefollowingdiagram

0 // P i //

. . . // P 0 //

M //

φ

0

0 // P 0 // . . . // P i //

Ext

i O X,x (M, O X,x ) // 0.

Using

φ

we get asymmetri morphism

ϕ : P • → (P • )

. Thus we haveon-

strutedan

i

-symmetripairintheategory

D b (P (O X,x ))

fromthepair

(M, φ)

.

Sine

D i b (X) ≃ a

x∈X (i)

D b (P (O X,x ))

([BW℄,Proposition7.1),weanseethepair

(P • , ϕ)

asasymmetripairin

D i b (X )

.

Definition

3.5

.

Theomplex

0 // W f l (k(X)) //

M

x 1 ∈X (1)

W f l (O X,x 1 ) // . . . //

M

x n ∈X (n)

W f l (O X,x n ) // 0

isalledtheGersten-Wittomplexof

X

. Wedenoteitby

C(X, W )

.

(11)

This omplex is obtained by using the usual duality

on the triangulated

ategory

D b (P (X))

(Example 2.2). For any invertible module

L

over

X

, we

haveadualityderivedfrom thefuntor

♯ =

Hom

O X,x (

_

, L)

andwean apply

thesameonstrutionto getaGersten-Wittomplex.

Definition

3.6

.

Let

X

bearegularsheme and

L

aninvertible

O X

-module.

Wedenote by

C(X, W, L)

theGersten-Wittomplexobtainedfrom the dual- ity

.

Theorem

3.7

.

Let

A

bearegularloal

k

-algebraand

X =

Spe

(A)

. Then for

any

i > 0

wehave

H i (C(X, W )) = 0

.

Proof. See[BGPW℄,Theorem6.1.

Let

A

be a regularloal ringof dimension

n

. Denote by

F

the residueeld

of

A

. Then any hoie of a generator

ξ ∈

Ext

n A (F, A)

givesan isomorphism

α ξ : W (F) → W f l (A)

. Reallthat

I(F)

isthefundamental idealof

W (F )

. If

n ≤ 0

,put

I n (F ) = W (F)

.

Definition

3.8

.

Forany

n ∈ Z

let

I f l n (A)

betheimageof

I n (F )

by

α ξ

.

Remark 3.9 . It iseasilyseenthat

I f l n (A)

doesnotdependonthehoieofthe

generator

ξ ∈

Ext

n A (F, A)

.

Proposition

3.10

.

The dierential

d

of the Gersten-Witt omplex satises

d(I f l m (O X,x )) ⊂ I f l m−1 (O X,y )

forany

m ∈ Z

,

x ∈ X (i)

and

y ∈ X (i−1)

.

Proof. See[Gi3℄, Theorem6.4or[Fa℄, Theorem9.2.4.

Definition

3.11

.

Let

L

be an invertible

O X

-module. We denote by

C(X, I d , L)

theomplex

0 // I f l d (k(X)) //

M

x 1 ∈X (1)

I f l d−1 (O X,x 1 ) // . . . //

M

x n ∈X (n)

I f l d−n (O X,x n ) // 0.

Remark 3.12 . Inpartiular,wehave

C(X, I 0 , L) = C(X, W, L)

.

Theorem

3.13

.

Let

A

bean essentiallysmoothloal

k

-algebra. Then for any

i > 0

we have

H i (C(X, I d )) = 0

.

Proof. See[Gi3℄, Corollary7.7.

Ofourse,thereis aninlusion

C(X, I d+1 , L) → C(X, I d , L)

andthereforewe

getaquotientomplex.

Definition

3.14

.

Denoteby

C(X, I d )

theomplex

C(X, I d , L)/C(X, I d+1 , L)

.

Remark 3.15 . Foranyinvertiblemodule

L

theomplexes

C(X, I d )/C(X, I d+1 )

and

C(X, I d , L)/C(X, I d+1 , L)

areanonially isomorphi(see[Fa℄, Corollary

E.1.3),sowean dropthe

L

in

C(X, I d )

.

(12)

Remark 3.16 . Theomplex

C(X, I d )

isoftheform

0 // I f l d (k(X ))/I f l d+1 (k(X )) //

M

x 1 ∈X (1)

I f l d−1 (O X,x 1 )/I f l d (O X,x 1 ) // . . . .

Remark 3.17 . As a onsequene of Theorem 3.13, we immediately see that

H i (C(X, I d )) = 0

for

i > 0

if

X =

Spe

(A)

where

A

is anessentiallysmooth loal

k

-algebra.

Let

F

beaeld and denoteby

K i M (F )

the

i

-th MilnorK-theorygroupof

F

.

If

i < 0

itisonvenienttoput

K i M (F ) = 0

.

Definition

3.18

.

Let

X

beasheme. Thenforany

d

wehaveaomplex

0 // K d M (k(X)) //

M

x 1 ∈X (1)

K d−1 M (k(x 1 )) // . . . //

M

x n ∈X (n)

K d−n M (k(x n )) // 0.

Wedenoteitby

C(X, K d M )

.

Proof. See[Ka℄,Proposition1or [Ro℄, Paragraph3.

Wealsohavetheexatnessofthisomplexwhen

X

isthespetrumofasmooth

loal

k

-algebra:

Theorem

3.19

.

Let

A

beasmoothloal

k

-algebra. Thenfor all

i > 0

wehave

H i (C(X, K d M )) = 0

.

Proof. See[Ro℄,Theorem6.1.

If

F

isaeld,reallthatwehaveahomomorphismduetoMilnor

s : K j M (F ) → I j (F )/I j+1 (F )

given by

s({a 1 , . . . , a j }) =< 1, −a 1 > ⊗ . . . ⊗ < 1, −a j >

. The following is

true:

Lemma

3.20

.

The homomorphisms

s

indueamorphismof omplexes

s : C(X, K d M ) → C(X, I d ).

Proof. See[Fa℄,Proposition10.2.5.

Definition

3.21

.

Let

C(X, G d , L)

be the bre produt of

C(X, K d M )

and

C(X, I d , L)

over

C(X, I d )

:

C(X, G d , L) //

C(X, I d , L)

π

C(X, K d M ) s // C(X, I d ).

(13)

Definition

3.22

.

Let

X

be a smooth sheme and

L

an invertible

O X

-

module. The

j

-th Chow-Witt group

CH g j (X, L)

of

X

twisted by

L

is the

group

H j (C(X, G j , L))

.

Remark 3.23 . Denoteby

GW j (D b j (X ), L)

the

j

-thGrothendiek-Wittgroupof theategory

D j b (X )

withthedualityderivedfromHom

O X (

_

, L)

(seeDenition

2.11). Itisnothardto seethat

C(X, G j , L)

isisomorphito

GW j (D b j (X), L)

andthereforetheomplex

C(X, G j , L)

is

. . . // C(X, G j , L) j−1 // GW j (D b j (X ), L) d j // W j+1 (D b j+1 (X ), L) // . . .

Hene

CH g j (X, L)

is a quotient of Ker

(d j )

and a subquotient of

GW j (D b j (X ), L)

.

Wealsohavetheexatness oftheomplex

C(X, G d , L)

in theloal ase:

Theorem

3.24

.

Let

A

be a smooth loal

k

-algebra and

X =

Spe

(A)

. Then

H i (C(X, G j )) = 0

for all

j

andall

i > 0

.

Proof. As

C(X, G j )

is the bre produt of the omplexes

C(X, K j M )

and

C(X, I j )

over

C(X, I j )

,wehaveanexatsequeneofomplexes

0 // C(X, G j ) // C(X, I j ) ⊕ C(X, K j M ) // C(X, I j ) // 0

induingalongexatsequenein ohomology. It followsthen from Theorem

3.13andTheorem3.19that

H i (C(X, G j )) = 0

if

i > 1

. For

i = 1

,wehavean

exatsequene

H 0 (C(X, I j )) ⊕ H 0 (C(X, K j M )) // H 0 (C(X, I j )) // H 1 (C(X, G j )) // 0.

Theexatsequeneofomplexes

0 // C(X, I j+1 ) // C(X, I j ) // C(X, I j ) // 0

showsthat

H 0 (C(X, I j ))

mapsonto

H 0 (C(X, I j ))

.

Definition

3.25

.

Let

X

beasmoothshemeand

L

aninvertible

O X

-module.

Wedenethesheaf

G L j

on

X

by

G L j (U) = H 0 (C(U, G j , L))

.

Wehave:

Theorem

3.26

.

Let

X

be a smooth sheme of dimension

n

. Then for any

i

wehave

H Zar i (X, G j L ) ≃ H i (C(X, G j , L)).

(14)

Proof. Denesheaves

C l

by

C l (U ) = C(U, G j , L) l

forany

l ≥ 0

. Itislearthat

the

C l

areasquesheaves. Wehaveaomplexofsheavesover

X 0 // G L j // C 0 // C 1 // . . . // C n // 0.

Theorem3.24showsthatthisomplexisaasqueresolutionof

G L j

. Thusthe

theoremisproved.

Suppose that

f : X → Y

isaatmorphism. Sine itpreservesodimensions, itinduesamorphismofomplexes

f : C(Y, G j , L) → C(X, G j , f L)

forany

j ∈ N

andanylinebundle

L

over

Y

([Fa℄,Corollary10.4.2). Henewe

have:

Theorem

3.27

.

Let

f : X → Y

bea at morphism and

L

a linebundleover

Y

. Then, forany

i, j

wehave homomorphisms

f : H i (C(Y, G j , L)) → H i (C(X, G j , f L)).

Inpartiular, if

E

isavetor bundle over

Y

and

π : E → Y

isthe projetion,

wehaveisomorphisms

π : H i (C(Y, G j , L)) → H i (C(E, G j , π L)).

Proof. We have amorphism of omplexes

f : C(Y, G j , L) → C(X, G j , f L)

whih givestheinduedhomomorphismsinohomology. For theproof ofho-

motopyinvariane,seeCorollary11.3.2in[Fa℄.

Proposition

3.28

.

Let

f : X → Y

and

g : Y → Z

be at morphisms. Then

(gf ) = f g

.

Proof. See[Fa℄,Proposition3.4.9.

Suppose that

f : X → Y

is a nite morphism with dim

(Y ) −

dim

(X) = r

.

Consider the morphism of loally ringed spaes

f : (X, O X ) → (Y, f ∗ O X )

induedby

f

. If

X

issmooth,then

L = f

Ext

r O Y (f ∗ O X , O Y )

is aninvertible

moduleover

Y

([Gi2 ℄,Corollary6.6)andwegetamorphismofomplexes(of

degreer)

f ∗ : C(X, G j−r , L ⊗ f N ) → C(Y, G j , N)

foranyinvertiblemodule

N

over

Y

([Fa℄,Corollary5.3.7).

(15)

Proposition

3.29

.

Let

f : X → Y

be a nite morphism between smooth

shemes. Let dim

(Y ) −

dim

(X) = r

and

N

be an invertible module over

Y

.

Then themorphism of omplexes

f ∗

indues ahomomorphism

f ∗ : H i−r (C(X, G j−r , L ⊗ f N )) → H i (C(Y, G j , N)).

Inpartiular,wehave([Fa℄,Remark9.3.5):

Proposition

3.30

.

Let

f : X → Y

be a losed immersion of odimensio n

r

betweensmooth shemes. Then

f

induesan isomorph ism

f ∗ : H i−r (C(X, G j−r , L ⊗ f N )) → H X i (C(Y, G j , N ))

forany

i, j

andany invertiblemodule

N

over

Y

.

Importantremark 3.31 . If

f : X → Y

isalosedimmersion,then

f ∗

willalways

bethemapwithsupport:

f ∗ : H i−r (C(X, G j−r , L ⊗ f N )) → H X i (C(Y, G j , N ))

Thetransferfornitemorphismsisfuntorial([Fa℄,proposition5.3.8):

Proposition

3.32

.

Let

f : X → Y

and

g : Y → Z

be nitemorphisms. Then

g ∗ f ∗ = (gf ) ∗

.

Remark 3.33 . Let

X

beasmoothshemeand

D

beasmootheetiveCartier

divisoron

X

. Let

i : D → X

be theinlusion and

L(D)

be the line bundle

over

X

assoiatedto

D

. Thenthere isaanonialsetion

s ∈ L(D)

(see [Fu℄,

AppendixB.4.5)andanexatsequene

0 // O X s

// L(D) // i ∗ O D // 0.

ApplyingHom

O X (

_

, L(D))

andshifting,weobtainthefollowingdiagram

0 // O X s

//

L(D) //

i ∗ O D //

0

0 //

Hom

O X (L(D), L(D)) s //

Hom

O X (O X , L(D)) //

Ext

1 O X (i ∗ O D , L(D)) // 0

whih showsthat Ext

1

O X (i ∗ O D , O X ) ⊗ L(D) ≃ i ∗ O D

. Proposition3.30shows thatwethenhaveanisomorphism

i ∗ : H i−1 (C(D, G j−1 , i L(D))) → H D i (C(X, G j )).

Lemma

3.34

.

Let

g : X → Y

be a at morphism and

f : Z → Y

a nite

morphism. Consider the following breprodut

(16)

V f

//

g

X

g

Z f // Y.

Then

(f ) ∗ (g ) = g f ∗

.

Proof. See[Fa℄,Corollary12.2.8.

Remark 3.35 . Ofourse,in theabovebreprodutwesupposethat

V

isalso

smoothandintegral. Suhastrongassumptionisnotneessaryingeneral,but

thisaseissuientforourpurposes.

Remark 3.36 . Itispossibletodeneamap

f ∗

whenthemorphism

f

isproper

(see[Fa℄)butwedon'tusethisfathere.

4 The exterior product

Let

X

and

Y

betwoshemes. Thebreprodut

X × Y

omesequippedwith

twoprojetions

p 1 : X × Y → X

and

p 2 : X × Y → Y

.

Lemma

4.1

.

Forany

i, j ∈ N

, the pairing

⊠ : D i b (X) × D b j (Y ) → D i+j b (X × Y )

given by

P ⊠ Q = p 1 P ⊗ p 2 Q

isa dualizing pairing of triangulated ategories withduality.

Proof. Straightveriation.

Corollary

4.2

.

Forany

i, j ∈ N

, the pairing

⊠ : D i b (X) × D b j (Y ) → D i+j b (X × Y )

indues apairing

⋆ : W i (D b i (X )) × W j (D j b (Y )) → W i+j (D b i+j (X × Y )).

Proof. ClearbyTheorem2.15.

Corollary

4.3

.

Let

ψ ∈ W j (D b j (Y ))

. Then wehave ahomomorphism

µ ψ : W i (D b i (X )) → W i+j (D b i+j (X × Y ))

given by

µ ψ (ϕ) = ϕ ⋆ ψ

.

Reallthat we have isomorphisms

W i (D i b (X )) ≃ M

x∈X (i)

W f l (O X,x )

(Proposi-

tion3.3).

(17)

Definition M

4.4

.

For any

s ∈ Z

, denote by

I s (D i b (X))

the preimage of

x∈X (i)

I f l s (O X,x )

undertheaboveisomorphism.

Proposition

4.5

.

Forany

m, p ∈ N

the produt

⋆ : W i (D i b (X )) × W j (D j b (Y )) → W i+j (D b i+j (X × Y ))

indues aprodut

⋆ : I m (D i b (X)) × I n (D j b (Y )) → I m+n (D b i+j (X × Y )).

Proof. Let

x ∈ X (i)

and

y ∈ Y (j)

. Itislearthattheprodutanbeomputed

loally (use [GN℄, Theorem 3.2). So wean suppose that

X =

Spe

(A)

and

Y =

Spe

(B)

where

A

and

B

areloalin

x

and

y

respetively. Reallthat we havethefollowingdiagram

X × Y p 2 //

p 1

Y

X //

Spe

(k).

Let

P

bean

A

-projetiveresolutionof

k(x)

and

Q

bea

B

-projetiveresolution

of

k(y)

. Considerasymmetriform

ρ : k(x) →

Ext

i A (k(x), A)

andasymmet-

riform

µ : k(y) →

Ext

j B (k(y), B)

. Then

p 1 (ρ)

is a symmetriisomorphism supportedbytheomplex

P ⊗ k B

and

p 2 (µ)

isasymmetriisomorphismsup- portedbytheomplex

A ⊗ k Q

. Theomplex

(P ⊗ k B) ⊗ A⊗ k B (A ⊗ k Q)

(whih

isisomorphito

P ⊗ k Q

)hasitshomologyonentratedin degree

0

,and this

homologyisisomorphito

k(x) ⊗ k k(y)

. Let

u

beapointofSpe

(k(x) ⊗ k k(y))

.

Thentherestritionof

p 1 ρ⊗p 2 µ

to

u

isanitelengthmodule

M

whosesupport

ison

u

withasymmetriform

M →

Ext

i+j (A⊗B)

u (M, (A ⊗ B) u ).

Taking its lass in the Witt group, we obtain a

k(u)

-vetor spae

V

with a

symmetri form

ψ : V →

Ext

i+j (A⊗B) u (V, (A ⊗ B) u )

. Now hoose a unit

a ∈ k(x) ×

. Consider the image

a u

of

a

under the homomorphism

k(x) → k(u)

.

Thelassof

p 1 (aρ) ⊗ p 2 (µ)

isthesymmetriform

a u ψ : V →

Ext

i+j (A⊗B)

u (V, (A ⊗ B) u ).

Asthesamepropertyholdsforanyunit

b ∈ k(y) ×

,weonludethat

p 1 (< 1, −a 1 > ⊗ . . . ⊗ < 1, −a n > ρ) ⊗ p 2 (< 1, −b 1 > ⊗ . . . ⊗ < 1, −b m > µ)

isequalto

< 1, −(a 1 ) u > ⊗ . . . ⊗ < 1, −(b m ) u > ψ

.

(18)

Reallthatforanysheme

X

wehaveaGersten-Wittomplex(Denition3.5)

C(X, W ) : . . . // W r (D b r (X )) d

r

X // W r+1 (D r+1 b (X)) // . . .

andaomplex

C(X, I d )

:

. . . //

M

x r ∈X (r)

I f l d−r (O X,x r ) //

M

x r+1 ∈X (r+1)

I f l d−r−1 (O X,x r+1 ) // . . . .

Theabovepropositiongives:

Corollary

4.6

.

Theprodut

⋆ : C(X, W ) × C(Y, W ) → C(X × Y, W )

indues for any

r, s ∈ N

aprodut

⋆ : C(X, I r ) × C(Y, I s ) → C(X × Y, I r+s ).

Now we investigate the relations between

and the dierentialsof the om- plexes.

Proposition

4.7

.

Let

ψ ∈ W j (D j b (Y ))

be suh that

d j Y (ψ) = 0

. Then the

following diagramommutes

W i (D b i (X)) d

i

X //

(−1) j µ ψ

W i+1 (D i+1 b (X ))

µ ψ

W i+j (D i+j b (X × Y ))

d i+j X×Y

// W i+j+1 (D b i+j+1 (X × Y )).

Proof. Let

ϕ ∈ W i (D b i (X ))

. Let

X (≥i+1)

bethesetofpointsof

X

ofodimen-

sion

≥ i+1

,

Y (≥j+1)

thepointsof

Y

ofodimension

≥ j +1

and

(X ×Y ) (≥i+j+1)

theset of points of

X × Y

of odimension

≥ i + j + 1

. ByLemma 2.18, the

triangulatedategories

D i b (X), D b j (Y )

and

D b i+j (X × Y )

are dened overthe

topologialspaes

X \ X (≥i+1)

,

Y \ Y (≥j+1)

and

(X ×Y ) \(X ×Y ) (≥i+j+1)

. Let

α ∈ Symm i (D b (P(X )) (i) )

and

β ∈ Symm j (D b (P (Y )) (j) )

besymmetripairs

representing

ϕ

and

ψ

. By denition, DegLo

(α)

is of odimension

≥ i + 1

,

DegLo

(β)

is ofodimension

≥ j + 1

and

is neutral. Itis easilyseenthat

Supp

(dp 1 α)∩

Supp

(dp 2 β) = ∅

inthetopologialspae

(X ×Y )\(X ×Y ) (≥i+j+1)

.

Theorem2.25impliesthat

(−1) i+j d(p 1 α ⋆ p 2 β ) = (−1) i dp 1 α ⋆ p 2 β + (−1) j p 1 α ⋆ dp 2 β.

(19)

UsingTheorem2.15,weseethat wehavein

W i+j (D b i+j (X × Y ))

theequality

(−1) j d i+j X×Y (p 1 ϕ ⋆ p 2 ψ) = p 1 d i X (ϕ) ⋆ p 2 ψ.

Thefollowingorollaryisobvious.

Corollary

4.8

.

Let

ψ ∈ I m (D j b (Y ))

be suh that

d Y j (ψ) = 0

. Then the

following diagramommutes

I p (D b i (X )) d

i

X //

(−1) j µ ψ

I p−1 (D i+1 b (X ))

µ ψ

I p+m (D b i+j (X × Y ))

d i+j X×Y

// I p+m−1 (D b i+j+1 (X × Y )).

Wenow haveto dealwith theomplex in MilnorK-theory. Let

C(X, K r M )

,

C(Y, K s M )

and

C(X ×Y, K r+s M )

betheomplexesinMilnorK-theoryassoiated

to

X, Y

and

X × Y

. In[Ro℄,Rost denesaprodut

⊙ : C(X, K r M ) i × C(Y, K s M ) j → C(X × Y, K r+s M ) i+j

asfollows: Let

u ∈ (X ×Y ) (i+j)

,

x ∈ X (i)

,

y ∈ Y (j)

besuhthat

x

and

y

arethe

projetionsof

u

. Let

ρ = {a 1 , . . . , a r−i } ∈ K r−i M (k(x))

and

µ = {b 1 , . . . , b s−j } ∈ K s−j M (k(y))

. Then

(ρ ⊙ µ) u = l((k(x) ⊗ k k(y)) u ){(a 1 ) u , . . . , (a r−i ) u , (b 1 ) u , . . . , (b s−j ) u }

wherethe

(a l ) u

and

(b t ) u

aretheimagesof the

a l

and

b t

under theinlusions

k(x) → k(u)

and

k(y) → k(u)

, and

l((k(x) ⊗ k k(y)) u )

is the length of the

module

k(x) ⊗ k k(y)

loalizedin

u

.

Lemma

4.9

.

Forany

ρ ∈ C(X, K r M ) i

and

µ ∈ C(Y, K s M ) j

wehave

d(ρ ⊙ µ) = d(ρ) ⊙ µ + (−1) j ρ ⊙ d(µ).

Proof. See[Ro℄,Paragraph14.4.

Corollary

4.10

.

Let

µ ∈ C(Y, K s M ) j

besuhthat

dµ = 0

. Thenthefollowing

diagramommutes:

C(X, K r M ) i d

i

X //

⊙µ

C(X, K r M ) i+1

⊙µ

C(X × Y, K r+s M ) i+j

d i+j X×Y

// C(X × Y, K M r+s ) i+j+1 .

参照

関連したドキュメント

For the above case, we show that “every uncountable system of linear homogeneous equations over Z , each of its countable subsystems having a non-trivial solution in Z , has

(In the sequel we shall restrict attention to homology groups arising from normalising partial isometries, this being appropriate for algebras with a regular maximal

Witt Nyström gives a similar looking integral formula that relates the Monge-Ampère energy of Hermitian metrics on a line bundle L over a compact complex manifold, to an integral

Applications of msets in Logic Programming languages is found to over- come “computational inefficiency” inherent in otherwise situation, especially in solving a sweep of

In order to be able to apply the Cartan–K¨ ahler theorem to prove existence of solutions in the real-analytic category, one needs a stronger result than Proposition 2.3; one needs

Our method of proof can also be used to recover the rational homotopy of L K(2) S 0 as well as the chromatic splitting conjecture at primes p &gt; 3 [16]; we only need to use the

Classical definitions of locally complete intersection (l.c.i.) homomor- phisms of commutative rings are limited to maps that are essentially of finite type, or flat.. The

In this paper we focus on the relation existing between a (singular) projective hypersurface and the 0-th local cohomology of its jacobian ring.. Most of the results we will present