, ,

1) Let

X = S p e c (A)

, and let

n

an integer with

(n, c h a r (k)) = 1

. If

F

is a finite, locally constant sheaf of

Z / n

-modules for the étale site on

X

, we show

H^{p} (G (A_{s} / A), F (A_{s})) \to \approx H^{p} (X, F)

where

A_{s}

is the separable closure of

A,

the left hand side is the Galois cohomology of

A

with coefficients in the

G (A_{s} / A)

-module

F (A_{s})

and the right hand side is the étale cohomology group of the semi-local scheme

X

with coefficients in

F

2) We extend the Merkuriev-Suslin theorem to a connected, semi-local ring

A

; that is, for

n

relatively prime to the residue characteristics of

A

, the Galois symbol map

K_{2} (A) / n \to \approx H^{2} (A, Z / n (2))

is an isomorphism where, as usual,

Z / n (i) = μ_{n}^{\otimes i}

. Since this implies the cup product map is surjective, we conclude that any Azumaya algebra of order

n

B r (A)

is similar to a tensor product of symbol algebras if

A

contains a primitive

n^{t h}

root of unity.

3) We extend Suslin's computation of the

ℓ

-primary component of the torsion in

K_{2} (k)

k

a field, to

A

Fix notation as follows.

H^{p} (X, F)

(or, if

X = S p e c (A)

H^{p} (A, F)

) denotes the étale cohomology group of

X

with coefficients in the étale sheaf

F

. For connected

A

, Galois cohomology will be indicated by

H^{p} (π_{1} (A, x), M)

where

π_{1} (A, x)

is the algebraic fundamental group of

A

with base point

x : S p e c (k_{s}) \to S p e c (A)

or, equivalently, by

H^{p} (G (A_{s} / A), M)

(

π_{1} (A, x) =

G (A_{s} / A)

, the Galois group of the separable closure of

A

, where the base point is defined by embedding the residue field of

A

into a separably closed field) when

M

is an abelian group equipped with a continuous action.

X^{(d)}

stands for the set of generic points

x

of irreducible components of

X

of codimension

d

. If

M

is a fixed abelian group, we let

M / n

stand for

M / n M

M {ℓ}

stand for the

ℓ

-primary component of

M

, and

D i v (M)

stand for the maximal divisible subgroup of

M

I would like to thank Srinivas, Bruno Kahn, and Wayne Raskind for several very helpful discussions on these and related matters and Chuck Weibel for the last application and unending patience. I also want to express my delight and appreciation to the Tata Institute of Fundamental Research for providing such a fine location and so much stimulation.

1 Local functors

Let

(A, I)

be a pair consisting of a commutative ring with

1

and an ideal

I \subset A .

An étale neighborhood of

(A, I)

is a pair

(A^{'}, I^{'})

and an etale map

u : A \to A^{'}

such that

u (I) A^{'} = I^{'}

and

u

induces an isomorphism

\bar{u} : A / I \to A^{'} / I^{'}

. Geometrically etale neighborhoods of a closed set

W \subset X = S p e c (A)

look like a cartesian diagram where

π

is etale:

\begin{matrix} W^{'} & \subset & X^{'} \end{matrix}

\begin{matrix} ↓ \approx & ↓ π \end{matrix}

\begin{matrix} W & \subset & X \end{matrix}

The set of all étale neighborhoods of

I

A

is a filtered category which we denote

E t (I)

. If we assume that

I \subseteq r a d (A),

then

E t (I)

is used to define the henselization of the pair

(A, I)

[15] by setting

{A_{/ I}}^{h} : = lim_{\to} (A^{'}, I^{'}) \in E t (I) A^{'} .

Then the pair

(A_{/ I}^{h}, I^{h})

is a hensel pair where

I^{h} = I A_{/ I}^{h} .

I = m

is maximal, we get the hensel local ring

A_{m}^{h} .

Let

C

be a category containing

E t (I)

as a full subcategory and

A_{/ I}^{h}

We introduce the definition of a local functor.

Definition 1. A covariant functor

F : C \to A b

is said to be local (for the étale topology) with respect to a closed set

V (I)

S p e c (A)

(1) $I$ is a radical ideal in $A$ , $E t (I)$ is a full subcategory of $C$ , and $A_{/ I}^{h} \in C$

(2) the natural map

\begin{matrix} lim_{\to} (A^{'}, I^{'}) \in E t (I) F (A^{'}) \to F (A / I) \end{matrix}

(1)

is an isomorphism.

We say that

F

is local for the étale topology on

X

if, for all affine open subsets

S p e c (A) \subset X

and all closed sets

V (I) \subset S p e c (A),

F

is local with respect to the closed set

V (I) \subset S p e c (A) .

In many cases of importance the limit condition above can be replaced with a condition involving the henselization of

A

along

I

. Recall that a covariant functor

F : C \to A b

is said to be locally of finite presentation if for any filtered inductive limit

A = {lim}_{\vec{i \in I}} A_{i}

, the natural map

{lim}_{\vec{i \in I}} F (A_{i}) \to F (A)

is an isomorphism.

For such functors we are only interested in a hensel pair condition.

Definition 2. A functor

F : C \to A b

is said to satisfy the hensel pair condition for

I

, resp. epic hensel pair condition for

I

, if

F (A) \to F (A / I)

is an isomorphism, resp. an epimorphism, for any hensel pair

(A, I) .

The functor satisfies the hensel pair condition, resp. epic hensel pair condition if it does so for any hensel pair

(A, I) \in C .

Then if

F

is locally of finite presentation and

F

satisfies the hensel pair condition,

F

is local for the closed set

V (I)

. Thus in applying this definition we will first verify that

F

is locally of finite presentation and then that

F (A^{h}) \to F (A^{h} / I^{h})

is an isomorphism when

(A^{h}, I^{h})

is a hensel pair.

We are primarily interested in three examples of functors satisfying the hensel pair condition.

Example 1.

(1) $H^{i} (-, F)$ where $F$ is a locally constant sheaf of $Z / n$ modules on $S p e c (A)_{e t}$ ([18] or [5] )
(2) $K_{i} (-; Z / n)$ where $n$ is relatively prime to the residue characteristics of $A$ ([4] )and
(3) $K_{2} (-) / n$ where $n$ is relatively prime to the residue characteristics of $A .$

The last example is easily seen to satisfy the hensel pair condition.

Surjectivity of

K_{2} (A) / n \to K_{2} (A / I) / n

for a hensel pair

(A, I)

follows immediately since

K_{2}

is generated by symbols, and injectivity follows by a straightforward calculation done in [11,Appendix] or from Gabber's theorem [4] . Alternatively we could use the universal coefficient theorem, the second example, and note that the third term in the universal coefficient theorem is the

n -

torsion in

K_{1}

which is the

n^{t h}

roots of unity. These examples will be discussed at greater length in the applications.

Our applications are a result of the following theorem.

Theorem 1. Let

A

be a semi-local ring essentially of finite type over a base ring

k

. Suppose

A \sim = B / J

where

B

is a smooth semi-local ring essentially of finite type over

k

. Let

C

be a category of semi-local ring extensions of

B

containing

A

B_{/ J}^{h}

, and the full subcategory

E t (J)

. Suppose

F_{1}, F_{2} : C \to A b

are two covariant functors which are either both local for

V (J)

F_{2}

is local for

V (J)

and

F_{1}

is locally of finite presentation and satisfies the epic hensel pair condition. Let

φ : F_{1} \to F_{2}

be a natural transformation such that

φ (B^{'})

is an isomorphism if

B^{'}

is smooth and essentially of finite type over

k

. Then

φ (A)

is an isomorphism.

Proof. If $A$ is essentially of finite type over $k$ , then $A$ has a presentation as $B / J$ where $B$ is essentially of finite type and smooth over $k$ . Thus $φ (B)$ is an isomorphism as is $φ (B^{'})$ for any $B^{'} \in E t (J)$ and so, by ( 1 ), $φ (A)$ must be an isomorphism in the first case. In the second case, $F_{1} (B_{/ J}^{h}) \to F_{1} (A)$ must be an isomorphism since $F_{2}$ is local for $V (J)$ and $φ (B_{/ J}^{h})$ is an isomorphism. □

Corollary 1. In the above situation

F_{1}

is local for

V (J)

if it is locally of finite presentation and satisfies the epic hensel pair condition.

2 Applications

We have three applications of this perspective. They all rely on first establishing an isomorphism for semi-local rings smooth over a base scheme and then using the examples above to extend the result to arbitrary semi-local schemes over the base.

2.1 Galois cohomology = Étale cohomology

As a first application we consider the relationship between étale and Galois cohomology of semi-local rings. While it is a generally accepted fact for smooth local rings over a field, the details do not, as far as I know, appear in the literature although there is an argument, due to Bruno Kahn, when the generalized Kato conjecture holds [11] . We begin by reviewing and reinterpreting Galois cohomology and then proving the isomorphism in this case.

Let

S

be an arbitrary connected scheme. Define a site

S_{r e t}

by considering the category of schemes

T \to S

which are finite and étale (= revêtement étale) over

S

. A covering morphism is a surjection of schemes over

S

T_{2} \to T_{1}

, which will necessarily be finite and étale.

This is the same as considering the class

E = (r e t)

of all finite, étale morphisms [12,ChapterII,Section1] .

S_{r e t}

is then the small site

(E / S)_{E}

where

E

consists of ”revêtements étales” as above. This site is discussed in [12,Chapter1] where it is called

F E t

and used to describe the fundamental group of

S

. Cohomology in

S_{r e t}

can be identified with Galois cohomology according to the following recipe.

Suppose

F

is a sheaf on

S_{r e t}

. Then

F

takes finite, disjoint unions of schemes to direct products. Thus the discussion in [12,ChapterIII,Example2.6] applies to the covering

T \to S

S_{r e t}

where

T

is Galois over

S

with group

G

. Hence

\begin{matrix} {\overset{ˇ}{H}}^{p} (T / S, F) \sim = H^{p} (G, F (T)) \end{matrix}

(2)

where

F (T)

is a left

G

-module via the action of

G

T

Let

ι : S (S_{r e t}) \to P (S_{r e t})

be the forgetful functor that regards a sheaf as a contravariant functor defined on

S_{r e t}

. When we pass to the limit over all coverings

T / S

S_{r e t}

, the left hand side of ( 2 ) becomes

{\overset{ˇ}{H}}^{p} (S_{r e t}, ι F)

. If we fix a geometric point

x : S p e c (k_{s}) \to S

where

k_{s}

is a separably closed field, then the theory of the fundamental group expresses any finite, étale covering

T \to S

as a quotient of a Galois covering. Consequently the right hand side of ( 2 ) is

lim_{\to} H^{p} (G_{i}, F (T_{i}))

where

G_{i}

ranges through the finite quotients of

π_{1} (S, x)

and

T_{i} \to S

is the corresponding étale covering with group

G_{i}

. Let

F (S_{s})

denote the abelian group

lim_{\to} F (T_{i})

. If

F

is a locally constant sheaf on

S

, then the subgroup of

π_{1} (S, x)

defining the covering

T \to S

such that

F |_{T}

is constant is of finite index in

π_{1} (S, x)

and acts trivially on

F (S_{s})

. Hence

F (S_{s})

has a continuous

π_{1} (S, x)

-action. In general this need not be the case, but we have enough to conclude that for

F

a locally constant sheaf on

S_{r e t}

\begin{matrix} H^{p} (S_{r e t}, F) \sim = H^{p} (π_{1} (S, x), F (S_{s})) \end{matrix}

(3)

(Note that a finite, locally constant sheaf for the étale topology is the same as a finite, locally constant sheaf on

S_{r e t}

.) In view of the isomorphism ( 2 ) on

S_{r e t}

, this is essentially the statement that sheaf cohomology coincides with Čech cohomology. Corollary 2.5 of Chapter III [12] gives sufficient conditions for this; namely, for every surjection

F \to F^{''}

of sheaves, the map

lim_{\to} \prod F (U_{i_{0} . . . i_{p}}) \to lim_{\to} \prod F^{''} (U_{i_{0} . . . i_{p}})

is surjective where the limit is over all coverings

{\prod U_{i} \to S}

S

and

U_{i_{0} . . . i_{p}} = U_{i_{0}} \times_{S} \dots \times_{S} U_{i_{p}} .

But if

T \to S

is a Galois covering of

S

with group

G

, then

F (T_{i_{0} . . . i_{p}}) = F (\prod_{G^{\times p}} T) = \prod_{G^{\times p}} F (T)

. Thus taking limits of

F (T_{i_{0} . . . i_{p}})

over all coverings means taking limits over all coverings of products of copies of

F (T)

and so surjectivity is immediate.

This discussion reduces our first application to the following theorem about the change of sites morphism

τ : S_{e t} \to S_{r e t}

Theorem 2. Let

A

be a connected semi-local ring containing a field

k

. If

F

is a finite, locally constant sheaf of

Z / n

-modules for the étale site on

A

where

(n, c h a r (k)) = 1

, then

\begin{matrix} H^{p} (A_{r e t}, τ_{*} F) \to H^{p} (A_{e t}, F) \end{matrix}

(4)

is an isomorphism for all p.

Proof. We wish to apply our extension theorem. We begin by treating the case of a smooth, semi-local ring

R

essentially of finite type over

k

, an algebraically closed field first. We will show that

R^{q} τ_{*} F = 0

for

q > 0

and any finite, locally constant sheaf

F

Z / n

-modules. Suppose

x \in H^{q} (R, F)

with

q > 0 .

We may assume

F = C_{X}

is a constant sheaf of

Z / n

-modules with value

C

since

F

becomes constant after a finite, etale extension of

R

. By assumption there is a smooth, connected variety

X

over

k

, closed points

t_{i} \in X, 1 \leq i \leq m

, such that

R \sim = O_{X, t_{1} . . . t_{m}}

and an element

x^{'} \in H^{q} (X, C_{X})

such that

x^{'} |_{S p e c (O_{X, t_{1} . . . t_{m}})} = x

We construct a finite, étale covering

ρ : X^{'} \to X

such that

ρ^{*} (x^{'}) = 0

using induction on

dim (X)

and the existence of Artin neighborhoods.

dim (X) = 0

, the assertion is obvious. If

dim (X) = d

, the existence of an Artin neighborhood relative to

k

means there is a diagram ([12,p.117] or [16] )

\begin{matrix} X & \to j & \bar{X} & \leftarrow i & Y \end{matrix}

\begin{matrix} f ↘ & ↓ \bar{f} & ↙ g \end{matrix}

\begin{matrix} X_{0} \end{matrix}

in which

(1) $j$ is an open immersion, dense in each fibre and $Y = \bar{X} - X;$
(2) $\bar{f}$ is smooth and projective with geometrically irreducible fibres of dimension one;
(3) $g$ is finite and étale and each fibre of $g$ is non-empty.

(Although the statement of the existence of Artin neighborhoods given in [12] refers only to a local ring, the proof as given in [16,ExposeXI,Section3] clearly extends to the semi-local case. In fact, the proof of Proposition 3.3, the key statement, is footnoted to that effect. Since base extension of a local ring from an arbitrary field

k

to its separable closure

k_{s}

may produce a semi-local ring we need this more general result.) Now

(Y, \bar{X})

is a smooth

X_{0} -

pair. The Gysin sequence [12,ChapterVI,Corollary5.3andRemark5.4(a)] becomes

{\bar{f}}_{*} C_{\bar{X}} \sim = f_{*} C_{X},

0 \to R^{1} {\bar{f}}_{*} C_{\bar{X}} \to R^{1} f_{*} C_{X} \to g_{*} (C_{Y} \otimes T_{Y / \bar{X}}) \to R^{2} {\bar{f}}_{*} C_{\bar{X}} \to R^{2} f_{*} C_{X} \to 0,

and

R^{p} {\bar{f}}_{*} C_{\bar{X}} \sim = R^{p} f_{*} C_{X}

for

p > 2

. The base change maps for the proper morphisms

\bar{f}

and

g

are isomorphisms. Consequently the base change map for

f

is also an isomorphism. Since the fibres of

f

are non-complete curves, we find

R^{p} f_{*} C = 0

for

p > 1

. Moreover

\bar{f}

is smooth and so

R^{1} {\bar{f}}_{*} C_{\bar{X}}

is finite and locally constant on

X_{0}

as is

g_{*} (C_{Y} \otimes T_{Y / X})

since

g

is an etale covering space and

T_{Y / X}

is locally constant on

Y

. Therefore we conclude that

R^{p} f_{*} C_{X}

is a finite, locally constant sheaf of

Z / n -

modules for

p = 0

1

This information shows that the Leray spectral sequence for

f

degenerates into the long exact sequence

\begin{matrix} \begin{matrix} . . . \to H^{q} (X_{0}, f_{*} C_{X}) \to H^{q} (X, C_{X}) \to & H^{q - 1} (X_{0}, R^{1} f_{*} C_{X}) & \to H^{q + 1} (X_{0}, f_{*} C_{X}) \to . . . \end{matrix} \end{matrix}

(5)

and we may apply the induction hypothesis. Then for any

y \in H^{q - 1} (X_{0}, R^{1} f_{*} C_{X}),

there is, after shrinking

X_{0}

if necessary, a finite, étale covering space

ρ^{'} : X_{0}^{'} \to X_{0}

such that

ρ^{' *} (y) = 0

. Now by using the pullbacks of appropriate coverings of

X_{0}

X

and a diagram chase, it is a straightforward matter to produce a finite, étale covering

ρ : X^{'} \to X

such that

ρ^{*} (x^{'}) = 0

This argument is, of course, the essence of Remark 3.16 [12,ChapterIII] and shows that

R^{q} τ_{*} F = 0

for

q > 0

. Hence the Leray spectral sequence for the change of site morphism

τ : X_{e t} \to X_{r e t}

collapses, and we conclude

H^{p} (R_{r e t}, τ_{*} F) \sim = H^{p} (R_{\overset{´}{e} t}, F)

Now suppose that

R

is a semi-local ring which is smooth and essentially of finite type over

k

, a not necessarily algebraically closed field. Then for any scheme

X

there is a Hochschild-Serre spectral sequence [12,ChapterIII,Remark2.21(b)]

\begin{matrix} H^{p} (G (k_{s} / k), H^{q} (X_{s}, F)) \Rightarrow H^{n} (X, F) \end{matrix}

(6)

where

X_{s} = X \times_{k} S p e c (k_{s})

. Similarly the theory of algebraic fundamental groups constructs a profinite group extension

1 \to G ((R \otimes_{k} k_{s})_{s} / R \otimes_{k} k_{s}) \to G (R_{s} / R) \to G (k_{s} / k) \to 1

and so there is a corresponding Hochschild-Serre spectral sequence

\begin{matrix} \begin{matrix} H^{p} (G (k_{s} / k), H^{q} (G ((R \otimes_{k} k_{s})_{s} / R \otimes_{k} k_{s}), F ((R \otimes_{k} k_{s})_{s}))) & \Rightarrow & H^{n} (G (R_{s} / R), F (R_{s})) \end{matrix} \end{matrix}

(7)

The change of site morphism defines a homomorphism

Ψ

from ( 6 ) to ( 7 ).

Since

F

is a sheaf of

Z / n

-modules where

n

is relatively prime to char

(k)

, we conclude, as usual, that

H^{p} (X \times_{k} S p e c (k_{s}), F) \sim = H^{p} (X \times_{k} S p e c (\bar{k}), F)

where

\bar{k}

is the algebraic closure of

k

. Consequently, by the first case,

Ψ

is an isomorphism on the

E_{2}^{p, q}

terms. This shows that ( 4 ) is an isomorphism when

A

is essentially smooth over a not necessarily algebraically closed field

k .

Some preparation is required in order to apply Theorem 1 to an arbitrary semi-local ring

A

whether we work with the small or large étale site. Our hypothesis on

F

shows that

F

is representable in either the small or large site by a semi-local scheme

F

which is finite and étale over

A .

Now

A

is a filtered limit of connected semi-local rings

A_{i}

essentially of finite type over

k .

Consequently there is a finite, etale covering space

F_{i_{0}}

over

A_{i_{0}}

for a sufficiently large

A_{i_{0}}

such that

F_{i_{0}} \times_{A_{i_{0}}} A ≊ F

. We use

F_{i_{0}}

to define a locally constant sheaf

F_{i_{0}}

over

A_{i_{0}}

whose restriction to

A

F .

Next find a presentation of

A_{i_{0}}

by a smooth connected semi-local ring

B_{i_{0}}

essentially of finite type over

k

so that

A_{i_{0}} \sim = B_{i_{0}} / I

. Since any étale covering space of

A_{i_{0}}

may be lifted to an étale covering space of

B_{i_{0}}

(by the theorem of the primitive element all one has to do is lift a separable polynomial), we may assume there is a finite, locally constant sheaf

G

S p e c (B_{i_{0}})

whose restriction to

S p e c (A)

F

Let

C

be the category of connected semi-local rings over

B

F_{1} = H^{p} (-_{r e t}, τ_{*} G)

, and

F_{2} = H^{p} (-, G)

where now we consider

G

on the large site over

S p e c (B)

. Note that

G

is represented by a finite étale covering space

B^{'}

B

B_{e t}

and so

G

is also represented by

B^{'}

on the big étale site over

B

. Since

B^{'}

also defines a covering in

B_{r e t}

the same observation applies to the sheaf

τ_{*} G

in the big site

B_{r e t}

. After showing that the hypotheses of our extension theorem are met we can conclude that ( 4 ) is an isomorphism for any

B -

algebra

A

with the property that

G |_{A} = F

and

τ_{*} G |_{A} = τ_{*} F

Both

F_{1}

and

F_{2}

are locally of finite presentation. For

F_{2}

this is [12,ChapterIII,Lemma1.16] suitably interpreted. For

F_{1}

we need a different argument. Suppose

R = lim_{\to} R_{i}

with

R_{i}

a connected semi-local ring over

B

where

i \in I

, a filtered set. Any étale covering space of

lim_{\to} R_{i}

comes from an étale covering space defined over one of the

R_{i}

and homomorphisms between any two étale covering spaces of

lim_{\to} R_{i}

descend to a homomorphism between covering spaces over one of the

R_{i}

. Thus we have an isomorphism

{lim}_{R_{i}} {lim}_{T_{i} / R_{i}} {\overset{ˇ}{H}}^{p} (T_{i} / R_{i}, G) \to {lim}_{T / R} {\overset{ˇ}{H}}^{p} (T / R, G)

since the Čech cohomology groups depend only on the Galois group of the covering by ( 2 ) and

G

is locally constant. Thus, using the Galois cohomology interpretation of

F_{1}

, we also have

H^{p} (R_{r e t}, τ_{*} G) \sim = lim_{\to} H^{p} (R_{i r e t}, τ_{*} G)

and so

F_{1}

is locally of finite presentation.

Now we turn to the hensel pair condition. Suppose

(R, J)

is a hensel couple and

G

is a locally constant sheaf of

Z / n -

modules on

S p e c (R)

Then

G

is represented by a finite, étale group scheme over

R

, and, by Example 1 , we have an isomorphism for a hensel couple

(R, J)

and such a

G

H^{p} (R_{\overset{´}{e} t}, G) \sim = H^{p} (R / J_{\overset{´}{e} t}, G |_{R / J}) .

Thus

F_{2}

is a local functor for the etale topology satisfying the hensel pair condition. As for

F_{1}

, there is a one-to-one correspondence between étale covering spaces of

R / J

and étale covering spaces of

R

, and so we have an isomorphism

H^{p} (R_{r e t}, τ_{*} G) \sim = H^{p} (R / J_{r e t}, τ_{*} (G |_{R / J})) .

Hence

F_{1}

is a local functor for the etale topology and also satisfies the hensel pair condition. We can now apply Theorem 1 to conclude the proof. □

Corollary 2. Let

A

be a semi-local ring containing a field

k

Suppose

F

is a finite, locally constant sheaf of

Z / n

-modules where

(n, c h a r (k)) = 1

. Then for any

x \in H^{p} (A, F)

p > 0

, there is a Galois extension

A^{'} / A

such that

x |_{A^{'}} = 0 \in H^{p} (A^{'}, F)

2.2 Merkurjev-Suslin theorem

Our second application is an extension of the Mercuriev-Suslin theorem to an arbitrary semi-local ring.

For a semi-local ring

A

and any

n

which is relatively prime to the residue characteristics of

A

, Kummer theory provides a natural isomorphism

K_{1} (A) / n \to H^{1} (A, Z / n (1)) .

The Galois symbol map [20] is obtained by extending this map multiplicatively to

K_{i} (A) / n \to H^{i} (A, Z / n (i)) .

Theorem 3. Let

A

be a semi-local ring. Then if

n

is relatively prime to the residue characteristics of

A

, the Galois symbol map

\begin{matrix} K_{2} (A) / n \to H^{2} (A, Z / n (2)) \end{matrix}

(8)

is an isomorphism.

Corollary 3. Let

A

be a semi-local ring, and suppose

n

is an integer which is relatively prime to the residue characteristics of

A

. Then the cup product map

H^{1} (A, Z / n (1)) \times H^{1} (A, Z / n (1)) \to H^{2} (A, Z / n (2))

is onto. In particular, if

A

contains a primitive

n^{t h}

root of unity, the

n

-torsion in the Brauer group of

A

is generated by symbol algebras

(a, b)_{n}

Corollary 4. Let

X

be a scheme, and suppose

n

is an integer which is relatively prime to the residue characteristics of

X

The symbol map defines an isomorphism of Zariski sheaves

\underset{̲}{K_{2} / n} \to \underset{̲}{H^{2} (Z / n (2)}) .

The proof requires some preliminary material on K-theory with coefficients, Chern classes, and a discussion of results of Gillet which will be used to prove the theorem for semi-local rings smooth and essentially of finite type over a semi-local Dedekind ring.

Fix a ring

A

, a prime

ℓ

that is a unit in

A

, and an integer

n

which is at least

2

ℓ = 2

(See [21,Proposition2.4] for a detailed discussion of the case

ℓ = 2

.) Algebraic K-theory with coefficients,

K_{r} (A; Z / ℓ^{n})

, is a bifunctor in

A

and the cyclic group

Z / ℓ^{n}

, and, for a pair of integers

1 \leq i, 0 \leq j \leq 2 i

, there is a theory of Chern classes given by natural transformations

c_{i, j}^{ℓ^{n}} : K_{2 i - j} (A; Z / ℓ^{n}) \to H^{j} (A, Z / ℓ^{n} (i))

satisfying certain properties. Among the properties these objects satisfy are:

(1) (universal coefficient theorem) There is a natural short exact sequence

\begin{matrix} 0 \to K_{r} (A) / ℓ^{n} K_{r} (A) \to K_{r} (A; Z / ℓ^{n}) \to_{ℓ^{n}} K_{r - 1} (A) \to 0 \end{matrix}

(9)

(2) (functoriality) For any

n > m

, the diagram

\begin{matrix} K_{2 i - j} (A; Z / ℓ^{m}) & \to & K_{2 i - j} (A; Z / ℓ^{n}) \end{matrix}

\begin{matrix} ↓ c_{i j}^{ℓ^{m}} & ↓ c_{i j}^{ℓ^{n}} \end{matrix}

\begin{matrix} H^{j} (A, Z / ℓ^{m} (i)) & \to & H^{j} (A, Z / ℓ^{n} (i)) \end{matrix}

defined from the inclusion

Z / ℓ^{m} ↪ Z / ℓ^{n}

commutes.

(3) (naturality) For any valuation ring

O_{v}

and prime

ℓ

distinct from the residue characteristic of

O_{v}

and any

n

, we have a commutative diagram

\begin{matrix} K_{2 i - j} (F; Z / ℓ^{m}) & \to \bar{\partial} & K_{2 i - j - 1} (κ (v); Z / ℓ^{n}) \end{matrix}

\begin{matrix} ↓ c_{i j}^{ℓ^{n}} & ↓ (1 - i) c_{i - 1 j - 1}^{ℓ^{n}} \end{matrix}

\begin{matrix} H^{j} (F, Z / ℓ^{n} (i)) & \to \partial & H^{j - 1} (κ (v), Z / ℓ^{n} (i - 1)) \end{matrix}

where

κ (v)

is the residue field of

O_{v}

and

F

is its field of fractions. In particular the Chern classes fit together to define a map from the Quillen-Gersten complex for

K_{*} (-; Z / ℓ^{n})

to the Bloch-Ogus complex for

H^{*} (-, Z / ℓ^{n} (\cdot))

. (Quillen's argument in [14] applies unchanged to

K_{i} (-; Z / m)

where

m = ℓ^{n}

Details of the above properties can be found in [17] .

Now the other tool we need is a reformulation and extension to a semi-local ring of a result of Gillet-Levine[7] and Gillet[6] .

Theorem 4 (Gillet-Levine, Gillet). Let

B

be a connected, semi-local ring with quotient field

K

which is smooth and essentially of finite type over a semi-local Dedekind ring

D

. Then, for any integer

n

relatively prime to the residue characteristics of

B

and any

q \geq 0,

we have a Gersten-Quillen resolution of

K_{q} (B; Z / n) :

0 \to K_{q} (B; Z / n) \to K_{q} (K; Z / n) \to \prod_{x \in B^{(1)}} K_{q - 1} (κ (x); Z / n) \to \dots

Proof. The argument of Gillet and Levine and Gillet immediately extends to semi-local rings as is clear from going through their arguments. □

We also need a weak version of this result for etale cohomology. This has been demonstrated by Gillet but remains unpublished[8] . A proof for

n

an odd integer relatively prime to the residue characteristics of

B

is given in an appendix at the end of this paper.

Lemma 1. Let

B

be a connected, semi-local ring which is smooth and essentially of finite type over a semi-local Dedekind ring,

D

. Let

K

be the quotient field of

B .

Then, for any integer

n

which is relatively prime to the residue characteristic of

D,

H^{2} (B, Z / n (2)) \to H^{2} (K, Z / n (2))

is a monomorphism.

We will need the universal coefficient theorem to produce a Gersten-Quillen sequence for

K_{2} (-) / n

using work of Gillet and Gillet-Levine and then the compatibility of the Chern class map with the Quillen-Gersten complex and the Bloch-Ogus complex will do the rest in the smooth case.

Proof. We deal with a semi-local algebra

A

over a mixed characteristic Dedekind ring

D

first. We may assume

A

is connected. Both

K_{2} (-) / n

and

H^{2} (-, Z / n (2))

are locally of finite presentation over

S p e c (D),

and so we may assume

A

is essentially of finite type over

D .

Now realize

A ≊ B / J

where

B

is a smooth, semi-local ring which is essentially of finite type over

O_{v} .

Let

C

be the category of connected semi-local rings essentially of finite presentation over

D .

We wish to apply Theorem 1 to this situation with

k = D,

F_{1} = K_{2} (-) / n,

F_{2} = H^{2} (-, Z / n (2)),

and

φ : F_{1} \to F_{2}

being the Galois symbol map.

First observe that

K_{2} (-) / n

satisfies the epic hensel pair condition since it is generated by symbols and

H^{2} (-, Z / n (2))

satisfies the hensel pair condition by Gabber or Strano's result ([5] or [18] ). It remains to show that

φ (B)

is an isomorphism if

B

is smooth and essentially of finite type over

D .

The exactness of

\begin{matrix} K_{2} (B; Z / n) ↪ K_{2} (K; Z / n) \to \prod_{x \in (S p e c B)^{(1)}} K_{1} (κ (x); Z / n) \end{matrix}

(10)

where

K

is the quotient field of

B

was shown by Gillet [6] using work of Gillet-Levine [7] (see Theorem 2.6). We see that

K_{1} (κ (x); Z / n) ≊ K_{1} (κ (x)) / n,

K_{2} (B; Z / n)

is an extension of

μ_{n} (B) (=_{n} K_{1} (B))

K_{2} (B) / n,

and similarly for

K_{2} (K; Z / n) .

We can now reinterpret ( 10 ) as the exact sequence of the first line below

\begin{matrix} K_{2} (B) / n ↪ & K_{2} (K) / n \to & \prod_{x \in (S p e c B)^{(1)}} K_{1} (κ (x)) / n \end{matrix}

\begin{matrix} ↓ & ↓ ≊ & ↓ ≊ \end{matrix}

\begin{matrix} H^{2} (B, Z / n (2)) ↪ & H^{2} (K, Z / n (2)) \to & \prod_{x \in (S p e c B)^{(1)}} H^{1} (κ (x), Z / n (1)) \end{matrix}

. Here the bottom row is not necessarily exact but is a complex, and the first map is a monomorphism by Lemma 1 , the first vertical isomorphism is the Merkuriev-Suslin theorem for fields, and the second vertical isomorphism is the observation that both groups are isomorphic to

\prod_{x \in (S p e c B)^{(1)}} κ (x)^{*} / κ (x)^{* n} .

The diagram commutes by the naturality condition above. Since the bottom row is a complex we conclude that the first vertical map is an isomorphism as desired. Theorem 1.1 now finishes this case.

The case of a semi-local ring containing a field is similar but simpler since we can use Grayson's version of the Gersten-Quillen sequence [9] and the Bloch-Ogus sequence [3] in place of the argument involving algebraic K-theory with coefficients and Lemma 1 . □

Suppose the Bloch-Kato conjecture holds for fields; that is, the Galois symbol map

\begin{matrix} K_{i}^{M} (K) / n \to H^{i} (K, Z / n (i)) \end{matrix}

(11)

is an isomorphism where

K_{i}^{M} (K)

is the

i^{t h}

Milnor K-group of the field

K

and

n

is relatively prime to

c h a r (K)

. Then the same argument applies since the Gersten-Quillen and Bloch-Ogus sequences hold for

i

as well as 2, at least up to

(i - 1)!

torsion and Gillet's unpublished result.

Thus the Bloch-Kato conjecture for fields in degree

i

and

i - 1

would show that ( 11 ) is an isomorphism for smooth local rings essentially of finite type over

k

. The rest of the argument is then identical, and so we would conclude that

K_{i}^{M} (A) / n

\to H^{i} (A, Z / n (i))

is an isomorphism if

A

is a semi-local ring containing

k

and

(n, (i - 1)! c h a r (k)) = 1

2.3 Torsion in K $_{2}$

The final application was suggested by C. Weibel. Recall that Suslin in [19] used Chern classes to construct an isomorphism

K_{2} (F) {ℓ} \to \approx H^{1} (F, Q_{ℓ} / Z_{ℓ} (2)) / D i v (H^{1} (F, Q_{ℓ} / Z_{ℓ} (2)))

where

F

is either a field of positive characteristic and

ℓ

is a prime distinct from

c h a r (F),

F

is a field of finite type over

Q

. Note that Suslin is using continuous Galois cohomology since he uses the identification

H^{1} (F, Q_{ℓ} / Z_{ℓ} (2)) / D i v (H^{1} (F, Q_{ℓ} / Z_{ℓ} (2))) = H_{c o n t}^{2} (F, Z_{ℓ} (2)) {ℓ}

where continuous cochains must be used to correctly identify

H_{c o n t}^{2} (F, Z_{ℓ} (2)) {ℓ}

H^{1} (F, Q_{ℓ} / Z_{ℓ}) (2))

modulo its maximal divisible subgroup.

Theorem 5. Let

A

be a semi-local ring containing a field

k

and suppose that

A

is either essentially of finite type over

Q

c h a r (k) > 0

. Let

ℓ

be a prime distinct from

c h a r (k)

. Then the Chern class

c_{2, 1}

induces an isomorphism

K_{2} (A) {ℓ} \to \approx H^{1} (A, Q_{ℓ} / Z_{ℓ} (2)) / D i v (H^{1} (A, Q_{ℓ} / Z_{ℓ} (2))) \to \approx H_{c o n t}^{2} (A, Z_{ℓ} (2)) {ℓ}

where

H_{c o n t}^{2} (A, Z_{ℓ}) (2))

is continuous étale cohomology [10] .

Corollary 5. Suppose

A

is a unibranch, e.g. normal, local ring containing a field

k

with

c h a r (k) = p > 0,

and let

ℓ

be as above. Then

K_{2} (A) {ℓ} \to \approx H^{1} (A, Q_{ℓ} / Z_{ℓ} (2))

The conclusion of Theorem 5 is independent of the characteristic of

k

, but the argument required in the two cases is different. We begin by recalling the structure of Suslin's argument in [19] and the role of continuous cohomology. He first observes (in the proof of Proposition 3.8

)

that

c_{2, 1}

appears in a commutative, exact diagram

\begin{matrix} \begin{matrix} K_{3} (F) \to & K_{3} (F; Z / ℓ^{n}) & \to c_{2, 1} & H^{1} (F, Z / ℓ^{n} (2)) \\ ↓ & ↓ \\ H_{c o n t}^{1} (F, Z_{ℓ} (2)) \to & H_{c o n t}^{1} (F, Q_{ℓ} (2)) & \to & H_{c o n t}^{1} (F, Q_{ℓ} / Z_{ℓ} (2)) \to & H_{c o n t}^{2} (F, Z_{ℓ} (2)) {ℓ} \to 0 \end{matrix} \end{matrix}

(12)

where the right hand vertical arrow comes from the coefficient sequence defined by multiplication by

ℓ^{n}

Q_{ℓ} / Z_{ℓ} (2)

. Continuous cohomology is used here to make the bottom sequence exact (the first arrow is defined on

ℓ

-adic sheaves by sending

x \mapsto ℓ^{- n} x

) and to identify the image of

H_{c o n t}^{1} (F, Q_{ℓ} (2))

with the maximal divisible subgroup of

H_{c o n t}^{1} (F, Q_{ℓ} / Z_{ℓ} (2))

and the image of

H_{c o n t}^{1} (F, Q_{ℓ} / Z_{ℓ} (2))

with the

ℓ

-primary torsion in

H_{c o n t}^{2} (F, Z_{ℓ} (2))

. Now by exactness and the universal coefficient theorem

c_{2, 1}

factors through

K_{2} (F) {ℓ}

\begin{matrix} {\bar{c}}_{2, 1} : & K_{2} (F) {ℓ} \to H_{c o n t}^{1} (F, Q_{ℓ} / Z_{ℓ} (2)) / I m a g e (H_{c o n t}^{1} (F, Q_{ℓ} (2))) \end{matrix}

\begin{matrix} \sim = H_{c o n t}^{2} (F, Z_{ℓ} (2)) {ℓ} . \end{matrix}

Suslin has already shown that

H_{c o n t}^{1} (F, Z_{ℓ} (2))

is finite when

c h a r (F) > 0

[19,Corollary2.8] , and so

H_{c o n t}^{1} (F, Q_{ℓ} (2)) = 0

and

H_{c o n t}^{1} (F, Q_{ℓ} / Z_{ℓ} (2)) \sim = H_{c o n t}^{2} (F, Z_{ℓ} (2)) {ℓ}

in this case. He then shows that

{\bar{c}}_{2, 1} : K_{2} (F) {ℓ} \to H_{c o n t}^{1} (F, Q_{ℓ} / Z_{ℓ} (2))

is an isomorphism [19,Theorem3.9] by an induction argument. In characteristic zero

H_{c o n t}^{1} (F, Q_{ℓ} (2))

is only zero if

F_{0}

, the algebraic closure of

Q

F

, has only real embeddings in

C

, but he can reduce the theorem to

F_{0}

[19,Proposition3.3] and here the desired result was proven earlier by Tate. Thus extending this result to a semi-local ring containing a field which is essentially of finite type over

Q

in characteristic zero requires first defining the factorization of

c_{2, 1}

through

{\bar{c}}_{2, 1}

(which will require continuous étale cohomology [10] ) and then showing that

{\bar{c}}_{2, 1}

is an isomorphism.

c_{2, 1}

is defined for arbitrary semi-local rings

A

(at least if

n > 2

when

ℓ = 2

) and Suslin's factorization argument applies which allows us to define

\begin{matrix} {\bar{c}}_{2, 1} (A) : K_{2} (A) {ℓ} \to H_{c o n t}^{1} (A, Q_{ℓ} / Z_{ℓ} (2)) / D i v^{1} (A) \end{matrix}

(13)

where, for simplicity,

D i v^{1} (A)

denotes the maximal

ℓ -

divisible subgroup of

H_{c o n t}^{1} (A, Q_{ℓ} / Z_{ℓ} (2))

. The exact sequence needed for this factorization requires using continuous étale cohomology which Jannsen [10] has developed.

It agrees with continuous group cochain cohomology when

A

is a field and has all the needed properties to construct the exact sequence in ( 12 ) and so to show that

{\bar{c}}_{2, 1}

is properly defined. Moreover, for a coefficient system of sheaves

F_{n}

, when

H^{i - 1} (A, F_{n})

satisfies the Mittag-Leffler condition,

H_{c o n t}^{i} (A, {lim}_{\leftarrow} F_{n})

is the usual

ℓ

-adic cohomology. Thus

H_{c o n t}^{1} (A, Z_{ℓ} (2)) = H^{1} (A, Z_{ℓ} (2))

and

H_{c o n t}^{1} (A, Q_{ℓ} / Z_{ℓ} (2)) = H^{1} (A, Q_{ℓ} / Z_{ℓ} (2))

In addition, [10,Theorem5.14] is the identification

H_{c o n t}^{2} (A, Z_{ℓ} (2)) {ℓ} \sim = I m (H^{1} (A, Q_{ℓ} / Z_{ℓ} (2))) .

The proof of Theorem 3, while different in

c h a r (A) > 0

and

c h a r (A) = 0

, starts from the comparison of the Gersten-Quillen exact sequence with a modification of the Bloch-Ogus sequence which is still exact.

Assume

A

is a semi-local ring which is smooth and essentially of finite type over a field

k

(

k = Q

in characteristic

0

). Consider, as in the proof earlier,

\begin{matrix} \begin{matrix} 0 \to K_{2} (A) {ℓ} \to & K_{2} (F) {ℓ} \to & \prod_{x \in S p e c (A)^{1}} K_{1} (κ (x)) {ℓ} \\ ↓ {\bar{c}}_{2, 1} (A) & ↓ {\bar{c}}_{2, 1} (F) & ↓ = \\ 0 \to H^{1} (A, (2)) / D i v^{1} (A) \to & H^{1} (F, (2)) / D i v^{1} (F) \to & \prod_{x \in S p e c (A)^{1}} H^{0} (κ (x), (1)) \end{matrix} \end{matrix}

(14)

where

H^{1} (A, (2))

denotes

H^{1} (A, Q_{ℓ} / Z_{ℓ} (2))

, etc. for typographical reasons. If

c h a r (A)

is positive, then

H^{1} (A, Z_{ℓ} (2))

is finite and so

H^{1} (A, Q_{ℓ} (2))

which maps onto the maximal divisible subgroup of

H^{1} (A, Q_{ℓ} / Z_{ℓ} (2))

vanishes. Thus the bottom sequence in ( 14 ) is just the Bloch-Ogus sequence and so

{\bar{c}}_{2, 1} (A)

is an isomorphism. If

c h a r (A) = 0

and

A

is essentially of finite type over

Q

, then

H^{0} (κ (x), Q_{ℓ} / Z_{ℓ} (1))

has no divisible part for any

x \in S p e c (A)^{(1)}

. Thus

D i v^{1} (A) = D i v^{1} (F)

, and so the bottom sequence is an exact quotient of the Bloch-Ogus sequence.

Thus in either case

{\bar{c}}_{2, 1} (A)

is an isomorphism.

Now suppose

A

is an arbitrary semi-local ring of positive characteristic.

Since both

K_{2} (A)

and

H^{1} (A, Q_{ℓ} / Z_{ℓ} (2))

commute with direct limits of rings we may assume that

A

is a semi-local ring essentially of finite type over a field. We can then use Theorem 1 since

H^{1} (A, Q_{ℓ} / Z_{ℓ} (2))

is local for the étale topology with respect to closed sets by 1) in Example ( 1 ) and

K_{2} (A) {ℓ}

is locally of finite presentation and satisfies the epic hensel pair condition. This latter condition follows since

K_{3} (A; Z / ℓ^{n})

is local for closed sets by 2) in Example ( 1 ) and so

_{ℓ^{n}} K_{2} (A),

being a quotient, satisfies the epic hensel pair condition. Note that the Corollary to Theorem 1 shows that

K_{2} (A) {ℓ}

is then local for the étale topology with respect to closed sets.

The characteristic

0

case is essentially the same but is complicated by the fact that

H^{1} (A, Q_{ℓ} / Z_{ℓ} (2)) / D i v^{1} (A)

does not behave nicely with respect to limits of rings. However Jannsen has shown that

H^{1} (A, Q_{ℓ} / Z_{ℓ} (2)) / D i v^{1} (A) ≊ H_{c o n t}^{2} (A, Z_{ℓ} (2)) {ℓ} .

In addition he shows that the sequence,

0 \to {lim}_{\leftarrow}^{1} H^{2} (B, Z / ℓ^{n} (2)) \to H_{c o n t}^{2} (B, Z_{ℓ} (2)) \to {lim}_{\leftarrow} H^{2} (B, Z / ℓ^{n} (2)) \to 0,

is exact where

{lim}_{\leftarrow}^{1}

refers to the first left derived functor of the system

{H^{2} (B, Z / ℓ^{n} (2))} .

Consequently

H_{c o n t}^{2} (B^{h}, Z_{ℓ} (2)) ≊ H_{c o n t}^{2} (A, Z_{ℓ} (2))

B^{h}

is the henselization with respect to

I

of a presentation

A ≊ B / I

as a quotient of a semi-local ring

B

smooth over

k .

This and the corresponding isomorphism

H^{1} (B^{h}, Q_{ℓ} / Z_{ℓ} (2)) ≊ H^{1} (A, Q_{ℓ} / Z_{ℓ} (2))

then allow us to conclude that

H_{c o n t}^{2} (B^{h}, Z_{ℓ} (2)) {ℓ} ≊ H_{c o n t}^{2} (A, Z_{ℓ} (2)) {ℓ} .

Now we may apply Theorem to show

{\bar{c}}_{2, 1} : K_{2} (A) {ℓ} \to H_{c o n t}^{2} (A, Z_{ℓ} (2)) {ℓ}

is an isomorphism.

Corollary 4 is obtained by noting that Suslin [19,Corollary2.8] also proved that

H^{1} (F, Z_{ℓ} (2))

is finite if

F

is a field of positive characteristic.

But if

A

is a unibranch local ring, e.g. a normal local ring, then

H^{1} (A, Z_{ℓ} (2)) ↪ H^{1} (F, Z_{ℓ} (2))

is also finite. Consequently

H^{1} (A, Q_{ℓ} (2)) = 0

, and so we have an isomorphism

H^{1} (A, Q_{ℓ} / Z_{ℓ} (2)) \sim = H^{2} (A, Z_{ℓ} (2)) {ℓ}

In this appendix

ζ_{n}

will always stand for a primitive

n^{t h}

root of unity. We wish to prove the following theorem. (Note that this is part of [11,Theorem1] and is proved without assuming the generalized Kato conjecture.)

Theorem 6. Let

A

be a connected, regular semi-local ring with quotient field

K,

and

n

an odd positive integer relatively prime to the residue characteristics of

A .

Then

H^{2} (A, Z / n (2)) \to j H^{2} (K, Z / n (2))

is injective.

We begin with a series of lemmas to calculate the cohomology of the extension obtained by adjoining an

n^{t h}

root of

1

. Fix the following situation in order to describe

Z / n (k)

and the action of the cyclotomic character on this module. Fix an integer

n

such that

n = d m

and

m = d e

for some positive integers

d, e,

and let

A

be a connected ring with

ζ_{m} \in A

and

d

a unit in

A .

Assume

A [ζ_{n}]

is connected. Then

X^{d} - ζ_{m} =^{d} \prod_{i = 1} (X - ζ_{n} {(ζ_{m}^{e})}^{i})

A [ζ_{n}],

and

A [ζ_{n}]

is a Galois extension of

A

with cyclic Galois group

G_{d} : = Z / d = 〈 σ 〉 \subset {(Z / n)}^{\times} .

Construct a dictionary for the action of

G_{d}

Z / n (k)

σ

sends

ζ_{n} \mapsto ζ_{n} (ζ_{m}^{e}) .

Since

n | m^{2},

we have

\begin{matrix} G_{d} on Z / n (k) & Z / d on Z / n \end{matrix}

\begin{matrix} (ζ_{n}, ζ_{m} = ζ_{n}^{d}, ζ_{d} = ζ_{m}^{e}) \in Z / n (1)^{\times 3} & (1, d, m = d e) \in Z / n^{\times 3} \end{matrix}

\begin{matrix} σ (ζ_{n}) = ζ_{n} (ζ_{m}^{e}) = ζ_{n}^{1 + m} \in Z / n (1) & σ (1) = 1 + m \in Z / n \end{matrix}

\begin{matrix} σ^{k} (ζ_{n}) = {(ζ_{n}^{1 + m})}^{k} = ζ_{n}^{1 + k m} \in Z / n (k) & σ^{k} (1) = {(1 + m)}^{k} = 1 + k m \in Z / n \end{matrix}

\begin{matrix}  \end{matrix}

This dictionary makes the computation of the cohomology of

G_{d}

Z / n (k)

straightforward.

Lemma 2.

H^{0} (G_{d}, Z / n (k)) = Z / (m (d, k)),

and

H^{2 r} (G_{d}, Z / n (k)) = {\begin{matrix} Z / (d, k) & if & d is odd or k is even or 4 | m \\ Z / 2 (d, k) & if & d is even and k is odd and 4 ∤ m \end{matrix}

and

H^{2 r + 1} (G_{d}, Z / n (k)) = {\begin{matrix} Z / (d, k) & if & d is odd or k is even or 4 | m \\ Z / 2 (d, k) & if & d is even and k is odd and 4 ∤ m \end{matrix}

Proof. The standard resolution used to calculate the cohomology of a cyclic group

G_{d}

Z / n (k)

is used to carry out this calculation. Our dictionary allows us to calculate

N_{k} = \sum_{i = 0}^{d - 1} σ^{i k} : Z / n \to Z / n

and

T_{k} = 1 - σ^{k} : Z / n \to Z / n

\begin{matrix} T_{k} (j) & = & j - j (1 + k m) = - k m j \end{matrix}

\begin{matrix} N_{k} (j) & = & j (\sum_{i = 0}^{d - 1} (1 + i k m)) \end{matrix}

\begin{matrix} = & j (d + k m \frac{d (d - 1)}{2}) \end{matrix}

\begin{matrix} = & {\begin{matrix} j d & if & d is odd or k is even \\ j (d - m \frac{d}{2}) & if & d is even and k is odd \end{matrix} \end{matrix}

In the case when

d

is even and

k

is odd, we note that

(m, 1 - \frac{m}{2}) = {\begin{matrix} 2 & if 4 ∤ m \\ 1 & otherwise \end{matrix} .

Then the sequence that calculates the cohomology groups is

\dots Z / n \to N_{k} Z / n \to T_{k} Z / n \to N_{k} Z / n \to T_{k} Z / n \dots,

and we find

\begin{matrix} Ker T_{k} & = & \frac{d}{(d, k)} Z / n \sim = Z / (m (d, k)) \end{matrix}

\begin{matrix} Im T_{k} & = & k m Z / n = (d, k) \cdot m Z / n \end{matrix}

\begin{matrix} Ker N_{k} & = & {\begin{matrix} m Z / n & if & d is odd or k is even or 4 | m \\ \frac{m}{2} Z / n & if & d is even, k is odd, and 4 ∤ m \end{matrix} \end{matrix}

\begin{matrix} Im N_{k} & = & {\begin{matrix} d Z / n & if & d is odd or k is even or 4 | m \\ 2 d Z / n & if & d is even, k is odd, and 4 ∤ m \end{matrix} \end{matrix}

The rest is straightforward. □

Corollary 6. Let

A

be a regular, connected semi-local domain, and let

n

be an odd positive integer relatively prime to the residue characteristics of

A

. Suppose that

m

is the largest divisor of

n

such that

ζ_{m} \in A

. Let

A_{n} = A [ζ_{n}],

and let

G =

Gal

(A_{n} / A) .

Then

H^{p} (G, Z / n (2)) = {\begin{matrix} Z / m (2) & if & p = 0 \\ 0 & if & p > 0 \end{matrix}

Proof. If $m = n,$ we are done. Suppose that $n = ℓ n^{'}$ where $ℓ$ is the smallest prime divisor of $\frac{n}{m} .$ Then there is a cyclic subgroup $H ⊳ G$ such that $A_{n^{'}} : = A^{H} = A [ζ_{n^{'}}],$ and $A_{n^{'}}$ is a Galois extension of $A$ with abelian Galois group $G / H .$ If $ζ_{ℓ} \in A_{n^{'}},$ then $A_{n}$ and $A_{n^{'}}$ satisfy the hypothesis of the lemma.
Hence $H^{p} (H, Z / n (2)) = {\begin{matrix} Z / n^{'} (2) & if & p = 0 \\ 0 & if & p > 0 \end{matrix} .$ If $ζ_{ℓ} \neq A_{n^{'},}$ then $[A_{n} : A_{n^{'}}] = ℓ - 1$ is relatively prime to $n^{'} .$ Hence $H^{p} (H, Z / n (2) = {\begin{matrix} Z / n^{'} (2) & if & p = 0 \\ 0 & if & p > 0 \end{matrix} .$ Now by using the Hochschild-Serre spectral sequence, $E_{2}^{p, q} = H^{p} (G / H, H^{q} (H, Z / n (2))) \Rightarrow H^{p + q} (G, Z / n (2)),$ and the above calculations, we conclude that $H^{p} (G / H, H^{0} (H, Z / n (2))) \to \approx H^{p} (G, Z / n (2)) .$ Since $H^{0} (H, Z / n (2)) = Z / n^{'} (2),$ we may apply an induction hypothesis and thus prove the corollary. □

Lemma 3. Let

A

be a regular, connected, semi-local domain with quotient field

K,

n

a positive integer relatively prime to the residue characteristics of

A

. Let

B

be a Galois extension of

A

with quotient field

L

, and let

G = G a l (B / A) .

Suppose

ζ_{n} \in B .

Then

H^{1} (G, H^{1} (B, Z / n (2))) \to H^{1} (G, H^{1} (L, Z / n (2)))

is injective.

Proof. Since $A$ is a UFD, we have an exact sequence $0 \to A^{*} \to K^{*} \to \prod o r d_{x} \prod_{x \in A^{(1)}} i_{x *} Z \to 0$ where $A^{(1)} = {x \in S p e c (A) / h t (x) = 1}$ and $o r d_{x} : K^{*} \to i_{x *} Z$ calculates the order of $f \in K^{*}$ at the discrete valuation associated to $x,$ and $i_{x} : S p e c (k (x)) \to S p e c (A)$ is the inclusion of the residue field of $A_{x} .$ If we map this sequence to itself by multiplication by $n$ , the cokernels form the short exact sequence $0 \to A^{*} / A^{* n} \to K^{*} / K^{* n} \to \prod o r d_{x} \prod_{x \in A^{(1)}} i_{x *} Z / n \to 0 .$ There is a sequence of $G$ -modules obtained by twisting the sequence for $B$ with the cyclotomic character $0 \to B^{*} / B^{* n} (1) \to L^{*} / L^{* n} (1) \to \prod_{y \in B^{(1)}} i_{y *} Z / n (1) \to 0,$ relating $B /^{*} B^{* n} (1) = H^{1} (B, Z / n (2))$ to $L^{*} / L^{* n} (1) = H^{1} (L, Z / n (2))$ since $ζ_{n} \in L^{*} .$ Thus we need only show that $H^{0} (G, L^{*} / L^{n *} (1)) \to H^{0} (G, \prod_{y \in B^{(1)}} i_{y *} Z / n (1))$ is onto to complete the proof of the lemma. But $H^{0} (G, \prod_{y \in B^{(1)}} i_{y *} Z / n (1)) ≊ \prod_{x \in A^{(1)}} H^{0} (D_{y | x}, i_{y *} Z / n (1)) ≊ \prod_{x \in A^{(1)}} i_{x *} Z / m_{x} (1)$ where $D_{y | x}$ is the decomposition group of a choice of $y \in B^{(1)}$ lying over $x \in A^{(1)}$ and $Z / m_{x} (1)$ is the group of roots of unity of order $n$ in $k (y)$ fixed by $D_{y | x}$ (expressed in the above notation).
Any element $α \in \prod_{x \in A^{(1)}} i_{x *} Z / m_{x} (1)$ can be written as $α = \sum_{x \in A^{(1)}} δ_{x} \cdot ζ_{n}^{r_{x}}$ where the sum is finite and $δ_{x} = 0$ for $x^{'} \in A^{(1)}$ if $x^{'} \neq x$ and $δ_{x} = 1 \in i_{x *} Z / n$ at $x \in A^{(1)} .$ Then there is $f_{x} \in K^{*} / K^{* n}$ with $o r d_{x} (f_{x}) = δ_{x}$ and $o r d_{x^{'}} (f_{x}) = 0$ if $x^{'} \neq x .$ Hence $\sum_{x \in A^{(1)}} (\sum_{σ \in G / D_{y | x}} f_{x} σ (ζ_{n}^{r_{x}}))$ has image $α .$ □

Remark 1. The proof amounts to first reducing to the case of a dvr

A

by using the UFD sequence and then solving the problem at the completion where

D_{y | x}

becomes the Galois group and finally spreading that solution around using the transitivity of the action of

G

{y \in B^{(1)} / y lies over a fixed x \in A^{(1)}} .

Proof. If $ζ_{n} \in A,$ the theorem reduces to the well known result $B r (A) \subset B r (K)$ since $H^{2} (A, Z / n (k)) ≊ H^{2} (A, Z / n (1)) ≊_{n} B r (A)$ and similarly for $K .$ We reduce the theorem to this case by analyzing the cohomology of a cyclic Galois covering $A [ζ_{n}] / A$ with Galois group $G .$ There is a spectral sequence for this covering $E_{2}^{p, q} = H^{p} (G, H^{q} (A [ζ_{n}], Z / n (2))) \Rightarrow H^{n} (A, Z / n (2)) .$ Suppose $x \in K e r [H^{2} (A, Z / n (2)) \to H^{2} (K, Z / n (2))] .$ Then $x = 0$ in $H^{2} {(A [ζ_{n}], Z / n (2))}^{G}$ by the above remarks. But Lemma 3 shows that $x = 0$ also in $E_{\infty}^{1, 1} \subseteq H^{1} (G, H^{1} (A [ζ_{n}], Z / n (2))) .$ Finally Corollary 6 shows that $x \in E_{\infty}^{2, 0} \subseteq H^{2} (G, Z / m (2))$ where $ζ_{m} \in A .$ But $H^{2} (G, Z / m (2))$ is the same for the Galois covering $A [ζ_{n}] / A$ and $K [ζ_{n}] / K .$ Hence $x = 0$ and the theorem follows. □

References

S. Bloch and A. Ogus “Gersten's conjecture and the homology of schemes”, Ann Sci Ecole Norm Sup., 4th sér. 7(1974). 181-202.
J-L Colliot-Thélène and M. Ojanguren “Espaces Principaux homogènes localement triviaux”, Publications Mathematiques de l'IHES, 75(1992), 97-122.
J.-L. Colliot-Thélène, R.T. Hoobler, and B. Kahn, “The Bloch—Ogus—Gabber Theorem”, Algebraic K-Theory, Fields Inst. Commun., Vol. 16, Amer. Math. Soc., Providence, RI, 1997, 31–94.
O. Gabber, “K-theory of henselian local rings and henselian pairs”, a letter from Ofer Gabber to M. Karoubi, 83, AMS Contemp Math, 1989.
O. Gabber, “Affine analog of the proper base change theorem”, Israel J. of Math., 87(1994),325-335.
H. Gillet, “Gersten's conjecture for the K-theory with torsion coefficients of a discrete valuation ring”, J Algebra, 103(1986), 377-380.
H. Gillet,.and M. Levine, “The relative form of Gersten's conjecture over a discrete valuation ring:The smooth case”, J Pure Appl Algebra, 46(1987), 59-71.
H. Gillet, personal communication, July 1995.
D. Grayson, “Universal exactness in algebraic K-theory”, J Pure Appl Algebra, 36(1985), 139-141.
U. Jannsen, “Continuous étale cohomology”, Math Ann, 280(1988), 207-245.
B. Kahn, “Deux théorèmes de comparision en cohomologie étale; applications”, Duke Math J, 69(1993), 137-165.
J. Milne, Étale Cohomology, Princeton University Press, 1980, Princeton, NJ.
C. Pedrini, and C. Weibel, “Invariants of real curves”, Rend Sem Mat Univers Politecn Torino, 49(1991), 139-173.
D. Quillen, “Higher Algebraic K-theory I”, Algebraic K-theory I: Higher K-theories, Lecture Notes in Math, 341, Springer-Verlag, Berlin, 1973, 85-147.
Michel Raynaud, Anneaux Locaux Hensélien, Lecture Notes in Math, 169, Springer-Verlag, Berlin, 1970.
SGA4:Theorie des Topos et Cohomologie Étale des Schemas, Lecture Notes in Math, 305, Springer-Verlag, Berlin, 1973.
C. Soulé, “K-théorie des anneaux d'entiers de corps de nombres et cohomologie étale”, Inventiones math, 55(1979), 251-295.
R. Strano,“On the étale cohomology of hensel rings”, Comm in Alg, 12, 2195-2211.
A. Suslin, “Torsion in $K_{2}$ of fields”, K-theory, 1(1987), 5-29.
J. Tate, “Relations between $K_{2}$ and Galois cohomology”, Inv Math 36(1976), 257-274.
C. Weibel, “Étale Chern classes at the prime 2”, Proceedings of the Lake Louise Conference on Algebraic K-Theory, NATO ASI Series C, 407, Kluwer Academic Publishers, Dordrecht 1993, 249-286.

City College of New York, New York, New York 10031

Raymond T. Hoobler