(J. S. Milne) Arithmetic Duality Theorems (B-Ok - Xyz)

Arithmetic Duality Theorems
Second Edition
J.S. Milne
Copyright c 2004, 2006 J.S. Milne.
The electronic version of this work is licensed under a Creative Commons Li-
cense:
http://creativecommons.org/licenses/by-nc-nd/2.5/
Briefly, you are free to copy the electronic version of the work for noncommercial
purposes under certain conditions (see the link for a precise statement).
Single paper copies for noncommercial personal use may be made without ex-
plicit permission from the copyright holder.
All other rights reserved.
First edition published by Academic Press 1986.
A paperback version of this work is available from booksellers worldwide and

from the publisher:
BookSurge, LLC, www.booksurge.com, 1-866-308-6235,
[email protected]
BibTeX information
@book{milne2006,
author={J.S. Milne},
title={Arithmetic Duality Theorems},
year={2006},
publisher={BookSurge, LLC},
edition={Second},
pages={viii+339},
isbn={1-4196-4274-X}
}
QA247 .M554
Contents
Contents iii
I Galois Cohomology 1
0 Preliminaries . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2
1 Duality relative to a class formation . . . . . . . . . . . . . . . . 17
2 Local fields . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 26
3 Abelian varieties over local fields . . . . . . . . . . . . . . . . . . 40
4 Global fields . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 48
5 Global Euler-Poincaré characteristics . . . . . . . . . . . . . . . . 66
6 Abelian varieties over global fields . . . . . . . . . . . . . . . . . 72
7 An application to the conjecture of Birch and Swinnerton-Dyer . . 93
8 Abelian class field theory . . . . . . . . . . . . . . . . . . . . . . 101
9 Other applications . . . . . . . . . . . . . . . . . . . . . . . . . . 116
Appendix A: Class field theory for function fields . . . . . . . . . . . . 126
II Etale Cohomology 139

0 Preliminaries . . . . . . . . . . . . . . . . . . . . . . . . . . . . 139
1 Local results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 148
2 Global results: preliminary calculations . . . . . . . . . . . . . . 163
3 Global results: the main theorem . . . . . . . . . . . . . . . . . . 176
4 Global results: complements . . . . . . . . . . . . . . . . . . . . 188
5 Global results: abelian schemes . . . . . . . . . . . . . . . . . . . 197
6 Global results: singular schemes . . . . . . . . . . . . . . . . . . 205
7 Global results: higher dimensions . . . . . . . . . . . . . . . . . 208
III Flat Cohomology 217

0 Preliminaries . . . . . . . . . . . . . . . . . . . . . . . . . . . . 218
1 Local results: mixed characteristic, finite group schemes . . . . . 232
2 Local results: mixed characteristic, abelian varieties . . . . . . . . 245
iii
iv
3 Global results: number field case . . . . . . . . . . . . . . . . . . 252

4 Local results: mixed characteristic, perfect residue field . . . . . . 257
5 Two exact sequences . . . . . . . . . . . . . . . . . . . . . . . . 266
6 Local fields of characteristic p . . . . . . . . . . . . . . . . . . . 272
7 Local results: equicharacteristic, finite residue field . . . . . . . . 280
8 Global results: curves over finite fields, finite sheaves . . . . . . . 289
9 Global results: curves over finite fields, Néron models . . . . . . . 294
10 Local results: equicharacteristic, perfect residue field . . . . . . . 300
11 Global results: curves over perfect fields . . . . . . . . . . . . . . 304
Appendix A: Embedding finite group schemes . . . . . . . . . . . . . . 307
Appendix B: Extending finite group schemes . . . . . . . . . . . . . . 312
Appendix C: Biextensions and Néron models . . . . . . . . . . . . . . 316
Bibliography 328
Index 337
v
Preface to the first edition.

In the late fifties and early sixties, Tate (and Poitou) found some important du-
ality theorems concerning the Galois cohomology of finite modules and abelian
varieties over local and global fields.
About 1964, Artin and Verdier extended some of the results to étale cohomol-
ogy groups over rings of integers in local and global fields.
Since then many people (Artin, Bester, Bégueri, Mazur, McCallum, the au-
thor, Roberts, Shatz, Vvedens’kii) have generalized these results to flat cohomol-
ogy groups.
Much of the best of this work has not been fully published. My initial purpose
in preparing these notes was simply to write down a complete set of proofs before
they were forgotten, but I have also tried to give an organized account of the
whole subject. Only a few of the theorems in these notes are new, but many
results have been sharpened, and a significant proportion of the proofs have not
been published before.
The first chapter proves the theorems on Galois cohomology announced by
Tate in his talk at the International Congress at Stockholm in 1962, and describes
later work in the same area. The second chapter proves the theorem of Artin and
Verdier on étale cohomology and also various generalizations of it. In the final
chapter improvements using flat cohomology are described.
As far as possible, theorems are proved in the context in which they are stated:
thus theorems on Galois cohomology are proved using only Galois cohomology,
and theorems on étale cohomology are proved using only étale cohomology.
Each chapter begins with a summary of its contents; each section ends with a
list of its sources.
It is a pleasure to thank all those with whom I have discussed these questions
over the years, but especially M. Artin, P. Berthelot, L. Breen, S. Bloch, K. Kato,
S. Lichtenbaum, W. McCallum, B. Mazur, W. Messing, L. Roberts, and J. Tate.
Parts of the author’s research contained in this volume have been supported
by the National Science Foundation.
Finally, I mention that, thanks to the computer, it has been possible to produce
this volume without recourse to typist, copy editor1 , or type-setter.
1 Inevitably, the sentence preceding this in the original contained a solecism

vi
Preface to the second edition.

A perfect new edition would fix all the errors, improve the exposition, update
the text, and, of course, being perfect, it would also exist. Unfortunately, these
conditions are contradictory. For this version, I have translated the original word-
processor file into TEX, fixed all the errors that I am aware of, made a few minor
improvements to the exposition, and added a few footnotes.
Significant changes to the text have been noted in the footnotes. The number-
ing is unchanged from the original (except for II 3.18). All footnotes have been
added for this edition except for those on p 26 and p 284.
There are a few minor changes in notation: canonical isomorphisms are often
denoted ' rather than , and, lacking a Cyrillic font, I use III as a substitute for
the Russian letter shah.
I thank the following for providing corrections and comments on earlier ver-
sions: Ching-Li Chai, Matthias Föhl, Cristian Gonzalez-Aviles, David Harari,
Eugene Kushnirsky, Bill McCallum, Bjorn Poonen, Joël Riou, and others.
Since most of the translation was done by computer, I hope that not many
new misprints have been introduced. Please send further corrections to me at
[email protected].
20.02.2004. First version on web.

07.08.2004. Proofread against original again; fixed many misprints and minor
errors; improved index; improved TEX, including replaced III with the correct
Cyrillic X.
01.07.2006. Minor corrections; reformatted for reprinting.
vii
Notations and Conventions

We list our usual notations and conventions. When they are not used in a particu-
lar section, this is noted at the start of the section.
A global field is a finite extension of Q or is finitely generated and of finite
transcendence degree one over a finite field. A local field is R, C, or a field that
is locally compact relative to a discrete valuation. Thus it is a finite extension
of Qp , Fp ..T //, or R. If v is a prime of a global field, then j jv denotes the
valuation at v normalized in the usual way so that the product formula holds, and
Ov D fa 2 K j jajv 1g. The completions of K and Ov relative to j jv are
denoted by Kv and O bv :
For a field K, K a and K s denote the algebraic and separable algebraic clo-
sures of K, and K ab denotes the maximal abelian extension of K. For a local
field K, K un is the maximal unramified extension of K. We sometimes write GK
for the absolute Galois group Gal.K s =K/ of K and GF =K for Gal.F=K/. By
char.K/ we mean the characteristic exponent of K, that is, char.K/ is p if K has
characteristic p ¤ 0 and is 1 otherwise. For a Hausdorff topological group G,
G ab is the quotient of G by the closure of its commutator subgroup. Thus, G ab is
the maximal abelian Hausdorff quotient group of G, and GK ab
D Gal.K ab =K/.
If M is an abelian group (or, more generally, an object in an abelian category)
and m is an integer, then Mm and M .m/ are the kernel and cokernel S of multi-
plication by m on M . Moreover, M.m/ isTthe m-primary component m Mmn
and Mmdiv is the m-divisible subgroup n Im.mn W M ! M /. The divisible
T
subgroup2 Mdiv of M is m Mm-div . We write Tm M for lim Mmn and M c for the

completion of M with respect to the topology defined by the subgroups of finite
index (sometimes the subgroups are restricted to those of finite index a power
of a fixed integer m, and sometimes to those that are open with respect to some
topology on M ). When M is finite, ŒM denotes its order. A group M is of
cofinite-type if it is torsion and Mm is finite for all integers m.
As befits a work with the title of this one, we shall need to consider a great
many different types of duals. In general, M will denote Homcts .M; Q=Z/, the
group of continuous characters of finite order of M . Thus, if M is discrete torsion
abelian group, then M is its compact Pontryagin dual, and if M is a profinite
abelian group, then M is its discrete torsion Pontryagin dual. If M is a module
over GK for some field K, then M D denotes the dual Hom.M; K s /; when M
is a finite group scheme, M D is the Cartier dual Hom.M; Gm /. The dual (Picard
2 Thisshould be called the subgroup of divisible elements — it contains the largest divisible
subgroup of M but it need not be divisible itself. A similar remark applies to the m-divisible
subgroup.
viii
variety) of an abelian variety is denoted by At . For a vector space M , M _ denotes

the linear dual of M .
All algebraic groups and group schemes will be commutative (unless stated
otherwise). If T is a torus over a field k, then X .T / is the group Homks .Gm ; Tks /
of cocharacters (also called the multiplicative one-parameter subgroups).
There seems to be no general agreement on what signs should be used in
homological algebra. Fortunately, the signs of the maps in these notes will not be
important, but the reader should be aware that when a diagram is said to commute,
it may only commute up to sign. I have generally followed the sign conventions
in Berthelot, Breen, and Messing 1982, Chapter 0.
We sometimes use D to denote a canonical isomorphism,3 and the symbols
df
X D Y and X Ddf Y mean that X is defined to be Y , or that X equals Y by
definition.
In Chapters II and III, we shall need to consider several different topologies
on a scheme X (always assumed to be locally Noetherian or the perfection of a
locally Noetherian scheme). These are denoted as follows:
Xet (small étale site) is the category of schemes étale over X endowed with
the étale topology;
XEt (big étale site) is the category of schemes locally of finite-type over X
endowed with the étale topology;
Xsm (smooth site) is the category of schemes smooth over X endowed with
the smooth topology (covering families are surjective families of smooth maps);
Xqf (small fpqf site) is the category of schemes flat and quasi-finite over X
endowed with the flat topology;
Xfl (big flat site) is the category of schemes locally of finite-type over X
endowed with the flat topology;
Xpf (perfect site) see (III 0).
The category of sheaves of abelian groups on a site X is denoted by S.X /.
3 And sometimes, in this edition, '.

Chapter I
Galois Cohomology
In 1 we prove a very general duality theorem that applies whenever one has a
class formation. The theorem is used in 2 to prove a duality theorem for modules
over the Galois group of a local field. This section also contains an expression for
the Euler-Poincaré characteristic of such a module. In 3, these results are used
to prove Tate’s duality theorem for abelian varieties over a local field.
The next four sections concern global fields. Tate’s duality theorem on mod-
ules over the Galois group of a global field is obtained in 4 by applying the
general result in 1 to the class formation of the global field and combining the
resulting theorem with the local results in 2. Section 5 derives a formula for
the Euler-Poincaré characteristic of such a module. Tate’s duality theorems for
abelian varieties over global fields are proved in 6, and in the following section
it is shown that the validity of the conjecture of Birch and Swinnerton-Dyer for
an abelian variety over a number field depends only on the isogeny class of the
variety.
The final three sections treat rather diverse topics. In 8 a duality theorem
is proved for tori that implies the abelian case of Langlands’s conjectures for a
nonabelian class field theory. The next section briefly describes some of the ap-
plications that have been made of the duality theorems: to the Hasse principle
for finite modules and algebraic groups, to the existence of forms of algebraic
groups, to Tamagawa numbers of algebraic tori over global fields, and to the cen-
tral embedding problem for Galois groups. In the appendix, a class field theory is
developed for Henselian local fields whose residue fields are quasi-finite and for
function fields in one variable over quasi-finite fields.
In this chapter, the reader is assumed to be familiar with basic Galois coho-
mology (the first two chapters of Serre 1964 or the first four chapters of Shatz
1972), class field theory (Serre 1967a and Tate 1967a), and, in a few sections,
1
2 CHAPTER I. GALOIS COHOMOLOGY
abelian varieties (Milne 1986b).

Throughout the chapter, when G is a profinite group, “G-module” will mean
“discrete G-module”, and the cohomology group H r .G; M / will be defined us-
ing continuous cochains. The category of discrete G-modules is denoted by
ModG :
0 Preliminaries
Throughout this section, G will be a profinite group. By a torsion-free G-module,
we mean a G-module that is torsion-free as an abelian group.
Tate (modified) cohomology groups

(Serre 1962, VIII; Weiss (1969).)
When G is finite, there are Tate cohomology groups HTr .G; M /, r 2 Z, M a
G-module, such that
HTr .G; M / D H r .G; M /; r > 0;

P
HT0 .G; M / D M G =NG M; where NG D 2G ;
P P
HT1 .G; M / D Ker.NG /=IG M , where IG D f n j n D 0g ;
HTr .G; M / D Hr1 .G; M /; r < 1:
A short exact sequence of G-modules gives rise to a long exact sequence of Tate
cohomology groups (infinite in both directions).
A complete resolution for G is an exact sequence
d2 d1 d0 d1
L D ! L2 ! L1 ! L0 ! L1 ! L2 !
of finitely generated free ZŒG-modules, together with an element e 2 LG

1 that
r
generates the image of d0 . For any complete resolution of G, HT .G; M / is the
r th cohomology group of the complex HomG .L ; M /. The map d0 factors as

L0 ! Z ! L1
with .x/e D d0 .x/ and .m/ D me. If we let
d2 d1
LC
D ! L2 ! L1 ! L0
d1
L
D L1 ! L2 ! L3 ! ;
0. PRELIMINARIES 3
then
H r .G; M / D H r .HomG .LC

; M //; r 0;
r1
Hr .G; M / D H .HomG .L
; M //; r 0:
By the standard resolution LC C

for G we mean the complex with Lr D ZŒG
r
and the usual boundary map, so that Hom.LC ; M / is the complex of nonhomo-
geneous cochains of M (see Serre 1962, VII 3). By the standard complete reso-
lution for G, we mean the complete resolution obtained by splicing together LC
with its dual (see Weiss 1969, I-4-1).
Except for Tate cohomology groups, we always set H r .G; M / D 0 for r < 0:
For any bilinear G-equivariant pairing of G-modules
M N !P
there is a family of cup-product pairings
.x; y/ 7! x Y yW HTr .G; M / HTs .G; N / ! HTrCs .G; P /
with the following properties:

(0.1.1) dx Y y D d.x Y y/I
(0.1.2) x Y dy D .1/deg.x/ d.x Y y/I
(0.1.3) x Y .y Y z/ D .x Y y/ Y zI
(0.1.4) x Y y D .1/deg.x/ deg.y/ y Y xI
(0.1.5) Res.x Y y/ D Res.x/ Y Res.y/I
(0.1.6) Inf.x Y y/ D Inf.x/ Y Inf.y/I
(d Dboundary map, Res Drestriction map; Inf Dinflation map).
T HEOREM 0.2 (TATE -NAKAYAMA ) Let G be a finite group, C a G -module, and

u an element of H 2 .G; C /. Suppose that for all subgroups H of G
(a) H 1 .H; C / D 0, and
(b) H 2 .H; C / has order equal to that of H and is generated by Res.u/.
Then, for any G -module M such that TorZ1 .M; C / D 0, cup-product with u
defines an isomorphism
x 7! x Y uW HTr .G; M / ! HTrC2 .G; M ˝ C /
for all integers r .
P ROOF. Serre 1962, IX 8. 2

Extensions of G-modules
For G-modules M and N , define ExtrG .M; N / to be the set of homotopy classes
of morphisms M ! N of degree r, where M is any resolution of M by
G-modules and N is any resolution of N by injective G-modules. One sees
readily that different resolutions of M and N give rise to canonically isomorphic
groups ExtrG .M; N /. On taking M to be M itself, we see that ExtrG .M; N / D
H r .HomG .M; N //, and so ExtrG .M; / is the r th right derived functor of
N 7! HomG .M; N /W ModG ! Ab. In particular, ExtrG .Z; N / D H r .G; N /.
There is a canonical product
.f; g/ 7! f gW ExtrG .N; P / ExtsG .M; N / ! ExtrCs

G .M; P /
such that f g is obtained from f W N ! P and gW M ! N by composition

(here N and P are injective resolutions of N and P /. For r D s D 0, the
product can be identified with composition
.f; g/ 7! f ı gW HomG .N; P / HomG .M; N / ! HomG .M; P /:
When we take M D Z, and replace N and P with M and N , the pairing becomes
ExtrG .M; N / H s .G; M / ! H rCs .G; N /:
An r-fold extension of M by N defines in a natural way a class in ExtrG .M; N /

(see Bourbaki Alg. X 7.3 for one correct choice of signs). Two such extensions
define the same class if and only if they are equivalent in the usual sense, and
for r 1, every element of ExtrG .M; N / arises from such an extension (ibid.
X 7.5). Therefore ExtrG .M; N / can be identified with the set of equivalence
classes of r-fold extensions of M by N . With this identification, products are ob-
tained by splicing extensions (ibid. X 7.6). Let f 2 ExtrG .N; P /; then the map
g 7! f gW ExtrG .M; N / ! ExtrCsG .M; P / is the r-fold boundary map defined
by any r -fold extension of N by P representing f:
A spectral sequence for Exts

Let M and N be G-modules, and write Hom.M; N / for the set of homomor-
phisms from M to N as abelian groups. For f 2 Hom.M; N / and 2 G, define
f to be m 7! .f . 1m//. Then Hom.M; N / is a G-module, but it is not in
general a discrete G-module. For a closed normal subgroup H of G, set
[
HomH .M; N / D Hom.M; N /U (union over the open subgroups H U G/
U
D ff 2 Hom.M; N / j f D f for all in some U g:
0. PRELIMINARIES 5
Then HomH .M; N / is a discrete G=H -module, and we define ExtH

r .M; N / to
be the r th right derived functor of the left exact functor
N 7! HomH .M; N /W ModG ! ModG=H :
In the case that H D f1g, we drop it from the notation; in particular,

S
Hom.M; N / D U Hom.M; N /U
with U running over all the open subgroups of G. If M is finitely generated, then
HomH .M; N / D HomH .M; N /, and so
r
ExtH .M; N / D ExtrH .M; N /I
in particular,
Hom.M; N / D Hom.M; N /
(homomorphisms as abelian groups).
T HEOREM 0.3 Let H be a closed normal subgroup of G , and let N and P be

G -modules. Then, for any G=H -module M such that TorZ1 .M; N / D 0, there is
a spectral sequence
ExtrG=H .M; ExtH

s
.N; P // H) ExtrCs
G .M ˝Z N; P /:
This will be shown to be the spectral sequence of a composite of functors, but

first we need some lemmas.
L EMMA 0.4 For any G -modules N and P and G=H -module M , there is a canon-
ical isomorphism
'
HomG=H .M; HomH .N; P // ! HomG .M ˝Z N; P /:
P ROOF. There is a standard isomorphism

'
HomG=H .M; Hom.N; P // ! Hom.M ˝Z N; P /:
Take G-invariants. On the left we get HomG .M; Hom.N; P //, which equals
HomG .M; HomH .N; P //
because M is a G=H -module, and equals HomG .M; HomH .N; P // because M
is a discrete G=H -module. On the right we get HomG .M ˝Z N; P /: 2
L EMMA 0.5 If I is an injective G -module and N is a torsion-free G -module,

then HomH .N; I / is an injective G=H -module.
P ROOF. We have to check that
HomG=H .; HomH .N; I //W ModG=H ! Ab
is an exact functor, but (0.4) expresses it as the composite of the two exact functors
˝Z N and HomG .; I /. 2
L EMMA 0.6 Let N and I be G -modules with I injective, and let M be a G=H -
module. Then there is a canonical isomorphism
'
ExtrG=H .M; HomH .N; I // ! HomG .TorZr .M; N /; I /.
P ROOF. We use a resolution of N
0 ! N1 ! N0 ! N ! 0
by torsion-free G-modules to compute TorZr .M; N /. Thus TorZ1 .M; N / and

TorZ0 .M; N / D M ˝Z N fit into an exact sequence
0 ! TorZ1 .M; N / ! M ˝Z N1 ! M ˝Z N0 ! TorZ0 .M; N / ! 0;
and TorZr .M; N / D 0 for r 2. For each open subgroup U of G containing H ,

there is a short exact sequence
0!
HomG .ZŒG=U ˝Z N; I / !
HomG .ZŒG=U ˝Z N0 ; I / !
HomG .ZŒG=U ˝Z N1 ; I / !
0

HomU .N; I / HomU .N0 ; I / HomU .N1 ; I /
The direct limit of these sequences is an injective resolution
0 ! HomH .N; I / ! HomH .N0 ; I / ! HomH .N1 ; I / ! 0
of HomH .N; I /, which we use to compute ExtrG=H .M; HomH .N; I //. In the
diagram
˛
HomG=H .M; HomH .N0 ; I // ! HomG=H .M; HomH .N1 ; I //
? ?
?' ?'
y y
ˇ
HomG .M ˝Z N0 ; I / ! HomG .M ˝Z N1 ; I /:
0. PRELIMINARIES 7
we have
Ker.˛/ D HomG=H .M; HomH .N; I //;
Coker.˛/ D Ext1G=H .M; HomH .N; I //
Ker.ˇ/ D HomG .TorZ0 .M; N /; I /;
Coker.ˇ/ D HomG .TorZ1 .M; N /; I /:
Thus the required isomorphisms are induced by the vertical maps in the diagram.2
We now prove the theorem. Lemma 0.4 shows that HomG .M ˝Z N; / is the
composite of the functors HomH .N; / and HomG=H .M; /, and Lemma 0.6
shows that the first of these maps injective objects I to objects that are acyclic for
the second functor. Thus the spectral sequence arises in the standard way from a
composite of functors (Hilton and Stammbach 1970)1 .
E XAMPLE 0.7 Let M D N D Z, and replace P with M . The spectral sequence
then becomes the Hochschild-Serre spectral sequence
H r .G=H; H s .H; M // H) H rCs .G; M /:
E XAMPLE 0.8 Let M D Z and H D f1g, and replace N and P with M and N .
The spectral sequence then becomes
H r .G; Ext s .M; N // H) ExtrCs
G .M; N /:
When M is finitely generated, this is simply a long exact sequence

0 ! H 1 .G; Hom.M; N // ! Ext1G .M; N / !
H 0 .G; Ext1 .M; N // ! H 2 .G; Hom.M; N // ! :
In particular, when we also have that N is divisible by all primes occurring as the
order of an element of M , then Ext1 .M; N / D 0, and so
H r .G; Hom.M; N // D ExtrG .M; N /.
E XAMPLE 0.9 In the case that N D Z, the spectral sequence becomes

ExtrG=H .M; H s .H; P // H) ExtrCs
G .M; P /.
The map ExtrG=H .M; P H / ! ExtrG .M; P / is obviously an isomorphism for

r D 0; the spectral sequence shows that it is an isomorphism for r D 1 if
H 1 .H; P / D 0, and that it is an isomorphism for all r if H r .H; P / D 0 for
all r > 0:
1 Better Shatz 1972, p50.
R EMARK 0.10 Assume that M is finitely generated. It follows from the long
exact sequence in (0.8) that ExtrG .M; N / is torsion for r 1. Moreover, if G
and N are written compatibly as G D lim Gi and N D lim Ni (Ni is a Gi -
!
module) and the action of G on M factors through each Gi , then
ExtrG .M; N / D lim ExtrGi .M; N /.

!
R EMARK 0.11 Let H be a closed subgroup of G, and let M be an H -module.
The corresponding induced G-module M is the set of continuous maps aW G !
M such that a.hx/ D h a.x/ all h 2 H , x 2 G. The group G acts on M by the
rule: .ga/.x/ D a.xg/. The functor M 7! M W ModH ! ModG is right adjoint
to the functor ModG ! ModH “regard a G-module as an H -module”; in other
words,
'
HomG .N; M / ! HomH .H; N /; N a G-module, M an H -module.
Both functors are exact, and therefore M 7! M preserves injectives and the
'
isomorphism extends to isomorphisms ExtrG .N; M / ! ExtrH .N; M / all r. In
'
particular, there are canonical isomorphisms H r .G; M / ! H r .H; M / for all
r. (Cf. Serre 1964, I 2.5.)
Augmented cup-products
Certain pairs of pairings give rise to cup-products with a dimension shift.
P ROPOSITION 0.12 Let
0 ! M 0 !M ! M 00 ! 0
0 ! N 0 !N ! N 00 ! 0
be exact sequences of G -modules. Then a pair of pairings
M0 N ! P
M N0 ! P
coinciding on M 0 N 0 defines a canonical family of (augmented cup-product)

pairings
H r .G; M 00 / H s .G; N 00 / ! H rCsC1 .G; /.
P ROOF. See Lang 1966, Chapter V. 2

0. PRELIMINARIES 9
R EMARK 0.13 (a) The augmented cup-products have properties similar to those
listed in (0.1) for the usual cup-product.
(b) Augmented cup-products have a very natural definition in terms of hyper-
cohomology. The tensor product
dM dN
.M 0 ! M 1 / ˝ .N 0 ! N 1 /
of two complexes is defined to be the complex with
d0 d1
M0 ˝ N0 ! M1 ˝ N0 ˚ M0 ˝ N1 ! M1 ˝ N1
with
d 0 .x ˝ y/ D dM .x/ ˝ y C x ˝ dN .y/;
d 1 .x ˝ y C x 0 ˝ y 0 / D x ˝ dN .y/ dM .x 0 / ˝ y 0 :
With the notations in the proposition, let M D .M 0 ! M / and N D .N 0 !
N /. Also write P Œ1 for the complex with P in the degree one and zero
elsewhere. Then the hypercohomology groups Hr .G; M /, Hr .G; N /, and
Hr .G; P Œ1/ equal H r1 .G; M 00 /, H r1 .G; N 00 /, and H r1 .G; P / respec-
tively, and to give a pair of pairings as in the proposition is the same as to give a
map of complexes
M ˝ N ! P Œ1:
Such a pair therefore defines a cup-product pairing
Hr .G; M / Hs .G; N / ! HrCs .G; P Œ1/;
and this is the augmented cup-product.
Compatibility of pairings
We shall need to know how the Ext and cup-product pairings compare.
P ROPOSITION 0.14 (a) Let M N ! P be a pairing of G -modules, and con-
sider the maps M ! Hom.N; P / and
H r .G; M / ! H r .G; Hom.N; P // ! ExtrG .N; P /
induced by the pairing and the spectral sequence in (0.3). Then the diagram
H r .G; M / H s .G; N / ! H rCs .G; P / (cup-product)

# k k
ExtrG .N; P / H s .G; N / ! H rCs .G; P / (Ext pairing)
commutes (up to sign).

(b) Consider a pair of exact sequences
0 ! M 0 ! M ! M 00 ! 0
0 ! N 0 ! N ! N 00 ! 0
and a pair of pairings
M0 N ! P
M N0 ! P
coinciding on M 0 N 0 . These data give rise to canonical maps H r .G; M 00 / !

ExtrC1 00
G .N ; P /, and the diagram
H r .G; M 00 / H s .G; N 00 / ! H rCsC1 .G; P / (augmented cup-product)

# k k
ExtrC1 00 H s .G; N 00 / ! H rCsC1 .G; P /
G .N ; P / (Ext pairing)
commutes (up to sign).
P ROOF. (a) This is standard, at least in the sense that everyone assumes it to be
true. There is a proof in a slightly more general context in Milne 1980, V 1.20,
and Gamst and Hoechsmann 1970, contains a very full discussion of such things.
(See also the discussion of pairings in the derived category in III 0.)
(b) The statement in (a) holds also if M , N , and P are complexes. If we
regard the pair of pairings in (b) as a pairing of complexes M N ! P Œ1
(notations as (0.13b)) and replace M , N , and P in (a) with M , N , and P Œ1,
then the diagram in (a) becomes that in (b). Explicity, the map H r .G; M 00 / !
ExtrC1 00
G .N ; P / is obtained as follows: the pair of pairings defines a map of
complexes M ! Hom.N ; P Œ1/, and hence a map
Hr .G; M / ! Hr .G; Hom.N ; P Œ1/I
but
Hr .G; M / D Hr1 .G; M /;
and there is an edge morphism
Hr .G; Hom.N ; P Œ1/! ExtrG .N ; P Œ1/ D ExtrG .N 00; P / 2

0. PRELIMINARIES 11
Conjugation of cohomology groups

Consider two profinite groups G and G 0 , a G-module M , and a G 0 -module M 0 .
A homomorphism f W G 0 ! G and an additive map hW M ! M 0 are said to be
compatible if h.f .g 0 / m/ D g 0 h.m/ for g 0 2 G 0 and m 2 M . Such a pair
induces homomorphisms .f; h/r W H r .G; M / ! H r .G 0 ; M 0 / for all r:
P ROPOSITION 0.15 Let M be a G -module, and let 2 G . The maps ad. / D
.g 7! g 1 /W G ! G and 1 D .m 7! 1 m/W M ! M are compatible,
and
.ad. /; 1 /r W H r .G; M / ! H r .G; M /
is the identity map for all r .
P ROOF. The first assertion is obvious, and the second needs only to be checked
for r D 0, where it is also obvious (see Serre 1962, VII 5). 2
The proposition is useful in the following situation. Let K be a global field

and v a prime of K. The choice of an embedding K s ! Kvs over K amounts
to choosing an extension w of v to K s , and the embedding identifies GKv with
the decomposition group Dw of w in GK . A second embedding is the composite
of the first with ad. / for some 2 G (because GK acts transitively on the
extensions of v to K s ). Let M be a GK -module. An embedding K s ! Kvs
defines a map H r .GK ; M / ! H r .GKv ; M /, and the proposition shows that the
map is independent of the choice of the embedding.
Extensions of algebraic groups

Let k be a field, and let G D Gal.k s =k/. The category of algebraic group
schemes over k is an abelian category Gpk (recall that all group schemes are as-
sumed to be commutative), and therefore it is possible to define Extrk .A; B/ for
objects A and B of Gpk to be the set of equivalence classes of r-fold extensions
of A by B (see Mitchell 1965, VII). Alternatively, one can chose a projective res-
olution A of A in the pro-category Pro-Gpk , and define Extrk .A; B/ to be the
set of homotopy classes of maps A ! B of degree r (see Oort 1966, I 4, or
Demazure and Gabriel 1970, V 2). For any object A of Gpk , A.k s / is a discrete
G-module, and we often write H r .k; A/ for H r .G; A.k s //.
P ROPOSITION 0.16 Assume that k is perfect.
(a) The functor A 7! A.k s /W Gpk ! ModG is exact.
(b) For all objects A and B in Gpk , there exists a canonical pairing
Extrk .A; B/ H s .k; A/ ! H rCs .k; B/:

P ROOF. (a) This is obvious since k s is algebraically closed.

(b) The functor in (a) sends an r-fold exact sequence in Gpk to an r-fold
exact sequence in ModG , and it therefore defines a canonical map Extrk .A; B/ !
ExtrG .A.k s /; B.k s //. We define the pairing to be that making
ExtrG .A; B/ H s .k; A/ ! H rCs .k; A/

# k k
ExtrG .A.k s /; B.k s // H s .G; A.k s // ! H rCs .G; B.k s //
commute. 2
P ROPOSITION 0.17 Assume that k is perfect, and let A and B be algebraic group
schemes over k . Then there is a spectral sequence
H r .G; Extsks .A; B// H) ExtrCs

k
.A; B/:
P ROOF. See Milne 1970a. 2
C OROLLARY 0.18 If k is perfect and N is a finite group scheme over k of order

prime to char.k/, then Extrk .N; Gm / ' ExtrG .N.k s /; k s / all r:
P ROOF. Clearly Homks .N; Gm / D HomG .N.k s /; k s /, and the table Oort 1966,
p II 14-2, shows that Extsks .N; Gm / D 0 for s > 0. Therefore the propo-
sition implies that Extrk .N; Gm / D H r .G; HomG .N.k s /; k s /, which equals
ExtrG .N.k s /; k s / by (0.8). 2
Topological abelian groups

Let M be an abelian group. In the next proposition we write ^
Q M for the m-
adic completion lim n M=m M of M , and we let Zm D `jm Z` D Z^ and
n
Q
Qm D `jm Q` D Zm ˝Z Q.
P ROPOSITION 0.19 (a) For any abelian group M , M ^ D .M=Mmdiv /^ ; if

M is finite, then M ^ D M.m/, and if M is finitely generated, then M ^ D
M ˝Z Z m .
n
(b) For any abelian group M , lim M .m / D .M ˝Z Q=Z/.m/, which is zero
!
if M is torsion and is isomorphic to .Qm =Zm /r if M is finitely generated of rank
r.
(c) For any abelian group, Tm M D Hom.Qm =Zm ; M / D Tm .Mmdiv /; it is
torsion-free.
0. PRELIMINARIES 13
(d) Write M D Homcts .M; Qm =Zm /; then for any finitely generated abelian
group M , M D .M ^ / and M D M ^ .
(e) Let M be a discrete torsion abelian group and N a totally disconnected
compact abelian group, and let
M N ! Q=Z
be a continuous pairing that identifies each group with the Pontryagin dual of the
other. Then the exact annihilator of Ntors is Mdiv , and so there is a nondegenerate
pairing
M=Mdiv Ntors ! Q=Z:
P ROOF. Easy. 2
Q
Note that the proposition continues to hold if we take m D“ p”, that is,
we take M ^ be the profinite completion of M , Mmdiv to be Mdiv , M.m/ to be
Mtor , and so on.
We shall be concerned with the exactness of completions and duals of exact
sequences. Note that the completion of the exact sequence
0 ! Z ! Q ! Q=Z ! 0
for the profinite topology is
0!b
Z ! 0 ! 0 ! 0;
which is far from being exact. To be able to state a good result, we need the
notion of a strict morphism. Recall (Bourbaki Tpgy, III 2.8) that a continuous
homomorphism f W G ! H of topological groups is said to be a strict mor-
phism if the induced map G= Ker.f / ! f .G/ is an isomorphism of topological
groups. Equivalently, f is strict if the image of every open subset of G is open in
f .G/ for the subspace topology on f .G/. Every continuous homomorphism of
a compact group to a Hausdorff group is strict, and obviously every continuous
homomorphism from a topological group to a discrete group is strict. The Baire
category theorem implies that a continuous homomorphism from a locally com-
pact -compact group onto2 a locally compact group is a strict morphism (Hewitt
and Ross 1963, 5.29; a space is -compact if it is a countable union of compact
subspaces).
^
Recall also that it is possible to define the completion G of a topological
group when the group has a basis of neighbourhoods .Gi / for the identity element
2 The original had “to” for “onto”, but the inclusion of the discrete group Z into Z is continu-
p
ous without being strict.
consisting of normal subgroups; in fact, G D lim i G=Gi . In the next proposition,

we write G for the full Pontryagin dual of a topological group G:

f g
G 0 ! G ! G 00
be an exact sequence of abelian topological groups and strict morphisms.
(a) Assume that the topologies on G 0 , G , and G 00 are defined by neighbourhood
bases consisting of subgroups; then the sequence of completions is also
exact.
(b) Assume that the groups are locally compact and Hausdorff and that the
image of G is closed in G 00 ; then the dual sequence3
G 00 ! G ! G 0
is also exact.
P ROOF. We shall use that a short exact sequence
0!A!B!C !0
of topological groups and continuous homomorphisms remains exact after com-

pletion provided the topology on B is defined by a neighbourhood basis consist-
ing of subgroups and A and C have the induced topologies (Atiyah and MacDon-
ald 1969, 10.3).
By assumption, we have a diagram
'
G= Im.f / ! Im.g/
x ?
? ?
? yb
f g
G0 ! G ! G 00
? x
?a ?
y ?
'
G 0 = Ker.f / ! Im.f /:
When we complete, the map a remains surjective, the middle column remains a
short exact sequence, and b remains injective because in each case a subgroup
has the subspace topology and a quotient group the quotient topology. Since the
isomorphisms obviously remain isomorphisms, (a) is now clear.
3 Here denotes the full Pontryagin dual, which coincides with Hom.; Q=Z/ on abelian
profinite groups.
0. PRELIMINARIES 15
The proof of (b) is similar, except that it makes use of the fact that for any
closed subgroup K of a locally compact abelian group G, the exact sequence
0 ! K ! G ! G=K ! 0
gives rise to an exact dual sequence
0 ! .G=K/ ! G ! K ! 0: 2
Note that in (b) of the theorem, the image of G in G 00 will be closed if it is

the kernel of a homomorphism from G 00 into a Hausdorff group.
The right derived functors of the inverse limit functor

The category of abelian groups satisfies the condition Ab5: the direct limit of an
exact sequence of abelian groups is again exact. Unfortunately, the corresponding
statement for inverse limits is false, although the formation of inverse limits is
always a left exact operation (and the product of a family of exact sequences is
exact).4
4 For an inverse system of abelian groups .An / indexed by N,
un
! An ! An1 ! ;
lim An and lim1 An are the kernel and cokernel respectively of

Q 1u Q
n An ! n An , ..1 u/.ai //n D an unC1 anC1 ;
and limi An D 0 for i > 1. Using the snake lemma, we find that a short exact sequence of abelian

groups
.fn / .gn /
0 ! .An / ! .Bn / ! .Cn / ! 0
gives rise to a six-term exact sequence
0 ! lim An ! lim Bn ! ! lim 1 Cn ! 0.

It is known (and easy to prove) that if an inverse system of abelian groups .An /n2N satisfies the
Mittag-Lœffler condition, then lim1 An D 0, however, the “well-known” generalization of this to

abelian categories satisfying Ab4 (see, for example, Jannsen, Uwe, Continuous étale cohomology.
Math. Ann. 280 (1988), no. 2, 207–245, Lemma 1.15, p. 213) is false: Neeman and Deligne (A
counterexample to a 1961 ”theorem” in homological algebra. With an appendix by P. Deligne.
Invent. Math. 148 (2002), no. 2, 397–420) construct an abelian category A in which small products
and direct sums exist and are exact, i.e., which satisfies Ab4 and Ab4 ; the opposite category has
the same properties, and inside it there is a inverse system .An /n2N with surjective transition maps
(hence .An / satisfies Mittag-Lœffler) such that lim1 An ¤ 0.

P ROPOSITION 0.21 Let A be an abelian category satisfying the condition Ab5

and having enough injectives, and let I be a filtered ordered set. Then for any
object B of A and any direct system .Ai / of objects of A indexed by I , there is a
spectral sequence
lim .r/ ExtsA .Ai ; B/ H) ExtrCs
A .lim Ai ; B/
!
.r/ th
where lim denotes the r right derived functor of lim .

P ROOF. Roos 1961. 2
P ROPOSITION 0.22 Let .Ai / be an inverse system of abelian groups indexed by

N with its natural order.
(a) For r 2, lim .r/ Ai D 0.

(b) If each Ai is finitely generated, then lim .1/ Ai is divisible, and it is un-

countable when nonzero.
(c) If each Ai is finite, then lim .1/ Ai D 0:

P ROOF. (a) See Roos 1961.
(b) See Jensen 1972, 2.5.
(c) See Jensen 1972, 2.3. 2
C OROLLARY 0.23 Let A be an abelian category satisfying Ab5 and having enough
injectives, and let .Ai / be a direct system of objects of A indexed by N. If B is
such that ExtsA .Ai ; B/ is finite for all s and i , then
lim ExtsA .Ai ; B/ D ExtsA .lim Ai ; B/.
!
The kernel-cokernel exact sequence of a pair of maps

The following simple result will find great application in these notes.
P ROPOSITION 0.24 For any pair of maps
f g
A!B!C
of abelian groups, there is an exact sequence
0 ! Ker.f / ! Ker.gıf / ! Ker.g/ ! Coker.f / ! Coker.gıf / ! Coker.g/ ! 0:
P ROOF. An easy exercise. 2
N OTES The subsection “A spectral sequence for Exts” is based on Tate 1966. The rest
of the material is fairly standard.
Since Roos 1961 contains no proofs and some false statements, it would be better to avoid
referring to it. Thus, this subsection should be rewritten. (But see: Roos, Jan-Erik. Derived
functors of inverse limits revisited. J. London Math. Soc. (2) 73 (2006), no. 1, 65–83.)
1. DUALITY RELATIVE TO A CLASS FORMATION 17
1 Duality relative to a class formation

Class formations
Consider a profinite group G, a G-module C , and a family of isomorphisms

invU W H 2 .U; C / ! Q=Z
indexed by the open subgroups U of G. Such a system is said to be a class

formation if
(1.1a) for all open subgroups U G, H 1 .U; C / D 0, and
(1.1b) for all pairs of open subgroups V U G, the diagram
ResV;U
H 2 .U; C / ! H 2 .V; C /
? ?
?inv ?inv
y U y V
n
Q=Z ! Q=Z
commutes with n D .U W V /. The map invU is called the invariant map relative
to U .
When V is a normal subgroup of U of index n, the conditions imply that there
is an exact commutative diagram
ResV;U
0 ! H 2 .U=V; C V / ! H 2 .U; C / ! H 2 .V; C / ! 0
? ? ?
?
yinvU=V
?
yinvU
?
yinvV
1 n
0 ! n Z=Z ! Q=Z ! Q=Z ! 0
in which invU=V is defined to be the restriction of invU . In particular, for a

normal open subgroup U of G of index n, there is an isomorphism

invG=U W H 2 .G=U; C U / ! n1 Z=Z;
and we write uG=U for the element of H 2 .G=U; C U / mapping to 1=n . Thus
uG=U is the unique element of H 2 .G=U; C U / such that invG .Inf.uG=U // D
1=n:
L EMMA 1.2 Let M be a G -module such that TorZ1 .M; C / D 0. Then the map
a 7! a Y uG=U W HTr .G=U; M / ! HTrC2 .G=U; M ˝Z C U /
is an isomorphism for all open normal subgroups U of G and integers r:

P ROOF. Apply (0.2) to G=U , C U , and uG=U : 2
T HEOREM 1.3 Let .G; C / be a class formation; then there is a canonical map
T GW C !
rec G G ab whose image in G ab is dense and whose kernel is the group
U
NG=U C of universal norms.
P ROOF. Take M D Z and r D 2 in the lemma. As HT2 .G=U; Z/ D .G=U /ab

and HT0 .G=U; C U / D C G =NG=U C U , the lemma gives an isomorphism

.G=U /ab ! C G =NG=U C U :
On passing to the projective

T limit over the inverses of these maps, we obtain an
injective map C G = NG=U U ! G abT . The map recG is the composite of this
G G
with the projection of C onto C = NG=U U . It has dense image because,
for all open normal subgroups U of G, its composite with G ab ! .G=U /ab is
surjective. 2
The map recG is called the reciprocity map.

Q UESTION 1.4 Is there a derivation of (1.3), no more difficult than the above
one, that avoids the use of homology groups?
R EMARK 1.5 (a) The following description of recG will be useful. The cup-
product pairing
H 0 .G; C / H 2 .G; Z/ ! H 2 .G; C /
can be identified with a pairing
h ; iW C G Homcts .G; Q=Z/ ! Q=Z
and the reciprocity map is uniquely determined by the equation
hc; i D .recG .c// all c 2 C G , 2 Homcts .G ab ; Q=Z/.
See Serre 1962, XI 3, Pptn 2.

(b) The definition of a class formation that we have adopted is slightly stronger
than the usual definition (see Artin and Tate 1961, XIV) in that we require invU
to be an isomorphism rather than an injection inducing isomorphisms

H 2 .U=V; C V / ! .U W V /1 Z=Z
for all open subgroups V U with V normal in U . It is equivalent to the usual

definition plus the condition that the order of G (as a profinite group) is divisible
by all integers n:
E XAMPLE 1.6 (a) Let G be a profinite group isomorphic to b Z (completion of

Z for the topology of subgroups of finite index), and let C D Z with G acting
trivially. Choose a topological generator of G. For each m, G has a unique
open subgroup U of index m, and m generates U . The boundary map in the
cohomology sequence of
0 ! Z ! Q ! Q=Z ! 0
is an isomorphism H 1 .U; Q=Z/ ! H 2 .U; Z/, and we define invU to be the

composite of the inverse of this isomorphism with
f 7!f . m /
H 1 .U; Q=Z/ D Homcts .U; Q=Z/ ! Q=Z:
Note that invU depends on the choice of . Clearly .G; Z/ with these maps is a
class formation. The reciprocity map is injective but not surjective.
(b) Let G be the Galois group Gal.K s =K/ of a nonarchimedean local field K,
and let C D K s . If I D Gal.K s =K un /, then the inflation map H 2 .G=I; K un / !
H 2 .G; K s / is an isomorphism, and we define invG to be the composite of its
inverse with the isomorphisms
ord invG=I
H 2 .G=I; K un / ! H 2 .G=I; Z/ ! Q=Z
where invG=I is the map in defined in (a) (with the choice of the Frobenius auto-
morphism for ). Define invU analogously. Then .G; K s / is a class formation
(see Serre 1967a, 1, or the appendix to this chapter). The reciprocity map is
injective but not surjective.
(c) Let G be the Galois group Gal.K s =K/ of a global field K, and let C D
lim CL where L runs through the finite extensions of K in K s and CL is the idèle
!
class group of L. For each prime v of K, choose an embedding of K s into Kvs
over K. Then there is a unique isomorphism invG W H 2 .G; C / ! Q=Z making
the diagram
invG
H 2 .G; C / ! Q=Z
?
?
y
invv
H 2 .Gv ; Kvs / ! Q=Z
commute for all v (including the real primes) with invv the map defined in (b)
unless v is real, in which case it is the unique injection. Define invU analogously.
Then .G; C / is a class formation (see Tate 1967a, 11). In the number field case,
the reciprocity map is surjective with divisible kernel, and in the function field
case it is injective but not surjective.
(d) Let K be a field complete with respect to a discrete valuation having an

algebraically closed residue field k, and let G D Gal.K s =K/ . For a finite separa-
ble extension L of K, let RL be the ring of integers in L. There is a pro-algebraic
group UL over k such that UL .k/ D RL . Let .U / be the pro-algebraic étale
1 L
fundamental group of UL , and let
1 .U / D lim 1 .UL /; K L K s; ŒLW K < 1:

!
Then 1 .U / is a discrete G-module and .G; 1 .U // is a class formation. In this
case the reciprocity map is an isomorphism. See Serre 1961, 2.5 Pptn 11, 4.1
Thm 1.
(e) Let K be an algebraic function field in one variable over an algebraically
closed field k of characteristic zero. For each finite extension L of K, let CL D
Hom.Pic.XL /; .k//, where XL is the smooth complete algebraic curve over k
with function field L and .k/ is the group of roots on unity in k. Then the duals
of the norm maps Pic.XL0 / ! Pic.XL /, L0 L, make the family .CL / into
a direct system, and we let C be the limit of the system. The pair .G; C / is a
class formation for which the reciprocity map is surjective but not injective. See
Kawada and Tate 1955 and Kawada 1960.
(f) For numerous other examples of class formations, see Kawada 1971.
The main theorem

For each G-module M , the pairings of 0
inv
ExtrG .M; C / H 2r .G; M / ! H 2 .G; C / ! Q=Z
induce maps
˛ r .G; M /W ExtrG .M; C / ! H 2r .G; M /
In particular, for r D 0 and M D Z, we obtain a map
˛ 0 .G; Z/W C G ! H 2 .G; Z/ D Homcts .G; Q=Z/ D G ab :
L EMMA 1.7 In the case that M D Z, the maps ˛ r .G; M / have the following
description:
˛ 0 .G; Z/WC G ! G ab is equal to recG ;

˛ 1 .G; Z/W0 ! 0I
'
˛ 2 .G; Z/WH 2 .G; C / ! Q=Z is equal to invG .
In the case that M D Z=mZ, the maps ˛ r .G; M / have the following description:
the composite of
˛ 0 .G; Z=mZ/W .C G /m ! H 2 .G; Z=mZ/
with H 2 .G; Z=mZ/ .G ab /m is induced by recG ;
˛ 1 .G; Z=mZ/W .C G /.m/ ! .G ab /.m/ is induced by recG ;

1
˛ 2 .G; Z=mZ/W H 2 .G; C /m ! Z=Z is the isomorphism induced by invG :
m
P ROOF. Only the assertion about ˛ 0 .G; Z/ requires proof. As we observed in

(1.5a), recG W H 0 .G; C / ! H 2 .G; Z/ is the map induced by the cup-product
pairing
H 0 .G; C / H 2 .G; C / ! H 2 .G; C / ' Q=Z
and we know that this agrees with the Ext pairing (see 0.14). 2
T HEOREM 1.8 Let .G; C / be a class formation, and let M be a finitely generated
G -module.
(a) The map ˛ r .G; M / is bijective for all r 2, and ˛ 1 .G; M / is bijective for
all torsion-free M . In particular, ExtrG .M; C / D 0 for r 3.
(b) The map ˛ 1 .G; M / is bijective for all M if ˛ 1 .U; Z=mZ/ is bijective for
all open subgroups U of G and all m:
(c) The map ˛ 0 .G; M / is surjective (respectively bijective) for all finite M if
in addition ˛ 0 .U; Z=mZ/ is surjective (respectively bijective) for all U and
m:
The first step in the proof is to show that the domain and target of ˛ r .G; M /
are both zero for large r.
L EMMA 1.9 For r 4, ExtrG .M; C / D 0; when M is torsion-free, Ext3G .M; C /

is also zero.
P ROOF. Every finitely generated G-module M can be resolved
0 ! M1 ! M0 ! M ! 0
by finitely generated torsion-free G-modules Mi . It therefore suffices to prove

that for any torsion-free module M , ExtrG .M; C / D 0 for r 3. Let N D
Hom.M; Z/. Then N ˝Z C ' Hom.M; C / as G-modules, and so (0.8) provides
an isomorphism ExtrG .M; C / ' H r .G; N ˝Z C /. Note that this last group
is equal to lim H r .G=U; N ˝Z C U / where the limit is over the open normal
!
subgroups of G for which N U D N . The theorem of Tate and Nakayama (0.2)
shows that
a 7! a Y uG=U W H r2 .G=U; N / ! H r .G=U; N ˝Z C U /
is an isomorphism for all r 3. The diagram

H r2 .G=U; N / ! H r .G=U; N ˝Z C U /
? ?
? ?
y.U WV /Inf yInf

H r2 .G=V; N / ! H r .G=V; N ˝Z C V /
commutes because Inf.uG=U / D .U W V /uG=V and Inf.aYb/ D Inf.a/YInf.b/.
As H r2 .G=U; N / is torsion for r 2 1, and the order of U is divisible by all
integers n, the limit lim H r2 .G=U; N / (taken relative to the maps .U W V /Inf)
!
is zero for r 2 1, and this shows that H r .G; N ˝Z C / D 0 for r 3. 2
P ROOF ( OF T HEOREM 1.8) Lemma 1.9 shows that the statements of the theo-
rem are true for r 4, and (1.7) shows that they are true for r 2 whenever the
action of G on M is trivial. Moreover, (1.9) shows that Ext3G .Z; C / D 0, and
it follows that Ext3G .Z=mZ; C / D 0 because Ext2G .Z; C/ is divisible. Thus the
theorem is true whenever the action of G on M is trivial. We embed a general M
into an exact sequence
0 ! M ! M ! M1 ! 0
with U an open normal subgroup of G such that M U D M and
M D Hom.ZŒG=U ; M / D ZŒG=U ˝Z M .
As H r .G; M / D H r .U; M / and ExtrG .M ; C / D ExtrU .M; C / (apply (0.3) to

ZŒG=U , M , and C /, there is an exact commutative diagram (1.9.1)
ExtrG .M1 ; C / !
! ExtrU .M; C / ! ExtrG .M; C / ! ExtrC1
G .M1 ; C / !

? ? ? ?
? r ? r ? r ? rC1
y˛ .G;M1 / y˛ .U;M / y˛ .G;M / y˛ .G;M1 /
H 2r .G; M1 / !
! H 2r .U; M / !
H 2r .G; M / !
H 1r .G; M1 / !

The maps ˛ 3 .U; M /, ˛ 4 .G; M1 /, and ˛ 4 .U; M / are all isomorphisms, and so
the five-lemma shows that ˛ 3 .G; M / is surjective. Since this holds for all M ,
˛ 3 .G; M1 / is also surjective, and now the five-lemma shows that ˛ 3 .G; M / is
an isomorphism. The same argument shows that ˛ 2 .G; M / is an isomorphism.

If M is torsion-free, so also are M and M1 , and so the same argument shows
that ˛ 1 .G; M / is an isomorphism when M is torsion-free. The rest of the proof
proceeds similarly. 2
E XAMPLE 1.10 Let .G; Z/ be the class formation defined by a group G b Z

and a generator of G. The reciprocity map is the inclusion n 7! n W Z ! G.
As b Z=Z is uniquely divisible, we see that both ˛ 0 .U; Z=mZ/ and ˛ 1 .U; Z=mZ/
are isomorphisms for all m, and so the theorem implies that ˛ r .G; M / is an iso-
morphism for all finitely generated M , r 1, and ˛ 0 .G; M / is an isomorphism
for all finite M .
In fact, ˛ 0 .G; M / defines an isomorphism HomG .M; Z/^ !H 2 .G; M / for
all finitely generated M . To see this, note that HomG .M; Z/ is finitely generated
and Ext1 .M; Z/ is finite (because H 1 .G; M / is) for all finitely generated M .
Therefore, on tensoring the first four terms of the long exact sequence of Exts
with bZ, we obtain an exact sequence
0 ! HomG .M1 ; Z/^ ! HomU .M; Z/^ ! HomG .M; Z/^ ! Ext1G .M1 ; Z/ ! .
When we replace the top row of (1.9.1) with this sequence, the argument proving
the theorem descends all the way to r D 0.
When M is finite, Extr .M; Z/ D 0 for r ¤ 1 and
Ext1 .M; Z/ D Hom.M; Q=Z/ D M .
Therefore ExtrG .M; Z/ D H r1 .G; M / (by (0.3)), and so we have a non-
degenerate cup-product pairing
H r .G; M / H 1r .G; M / ! H 1 .G; Q=Z/ ' Q=Z:
When M is torsion-free, Extr .M; Z/ D 0 for r ¤ 0 and Hom.M; Z/ is the

linear dual M _ of M . Therefore Extr .M; Z/ D H r .G; M _ /, and so the map
H r .G; M _ / ! H 2r .G; M / defined by cup-product is bijective for r 1,
and induces a bijection H 0 .G; M _ /^ ! H 2 .G; M / in the case r D 0:
E XAMPLE 1.11 Let K be a field for which there exists a class formation .G; C /
with G D Gal.K s =K/, and let T be a torus over K. The character group X .T /
of T is a finitely generated torsion-free G-module with Z-linear dual the cochar-
acter group X .T /, and so the pairing
ExtrG .X .T /; C / H 2r .G; X .T // ! H 2 .G; C / ' Q=Z

defines an isomorphism
ExtrG .X .T /; C / ! H 2r .G; X .T //
for r 1. According to (0.8),
ExtrG .X .T /; C / D H r .G; Hom.X .T /; C //, and
Hom.X .T /; C / D X .T / ˝ C:
Therefore the cup-product pairing
H r .G; X .T / ˝ C / H 2r .G; X .T // ! H 2 .G; C / ' Q=Z
induced by the natural pairing between X .T / and X .T / defines an isomor-
phism
H r .G; X .T / ˝ C / ! H 2r .G; X .T // ; r 1:
R EMARK 1.12 Let .G; C / be a class formation. In Brumer 1966 there is a very
useful criterion for G to have strict cohomological dimension 2. Let V U G
be open subgroups with V normal in U . We get an exact sequence
recV
0 ! Ker.recV / ! C V ! V ab ! Coker.recV / ! 0
of U=V -modules which induces a double connecting homomorphism
d W HTr2 .U=V; Coker.recV // ! HTr .U=V; Ker.recU //.
The theorem states that scdp .G/ D 2 if and only if, for all such pairs V U ,
d induces an isomorphism on the p-primary components for all r. In each of
the examples (1.6a,b,d) and in the function field case of (c), the kernel of recV is
zero and the cokernel is uniquely divisible and hence has trivial cohomology. In
the number field case of (c), the cohomology groups of the kernel are elementary
2-groups, which are zero if and only if the field is totally imaginary (Artin and
Tate 1961, IX 2). Consequently scdp .G/ D 2 in examples (1.6a,b,c,d) except
when p D 2 and K is a number field having a real prime.
On the other hand, let K be a number field and let GS be the Galois group
over K of the maximal extension of K unramified outside a set of primes S. The
statement in Tate 1962, p292 that scdp .GS / D 2 for all primes p that are units at
all v in S (except for p D 2 when K is not totally complex) is still unproven in
general. As was pointed out by A. Brumer, it is equivalent to the nonvanishing of
certain p-adic regulators.5
5 See Corollary 10.3.9,
p538, of Neukirch, Jürgen; Schmidt, Alexander; Wingberg, Kay. Coho-
mology of number fields. Grundlehren der Mathematischen Wissenschaften 323. Springer-Verlag,
Berlin, 2000.
A generalization
We shall need a generalization of Theorem 1.8. For any set P of rational prime
numbers, we define a P -class formation to be a system .G; C; .invU /U / as at the
start of this section except that, instead of requiring the maps invU to be isomor-
phisms, we require them to be injections satisfying the following two conditions:
(a) for all open subgroups V and U of G with V a normal subgroup of U , the
map
invU=V W H 2 .U=V; C V / ! .U W V /1 Z=Z
is an isomorphism, and
(b) for all open subgroups U of G and all primes ` in P , the map on `-primary
components H 2 .U; C /.`/ ! .Q=Z/.`/ induced by invU is an isomor-
phism.
Thus when P contains all prime numbers, a P -class formation is a class forma-
tion in the sense of the first paragraph of this section, and when P is the empty
set, a P -class formation is a class formation in the sense of Artin and Tate 1961.
Note that, in the presence of the other conditions, (b) is equivalent to the order of
G being divisible by `1 for all ` in P . If .G; C / is a class formation and H is
a normal closed subgroup of G, then .G=H; C H / is a P -class formation with P
equal to the set primes ` such that `1 divides .GW H /.
If .G; C / is a P -class formation, then everything said above continues to
hold provided that, at certain points, one restricts attention to the `-primary com-
ponents for ` in P . (Recall (0.10) that ExtrG .M; N / is torsion for r 1:/ In
particular, the following theorem holds.
T HEOREM 1.13 Let .G; C / be a P -class formation, let ` be a prime in P , and
let M be a finitely generated G -module.
(a) The map ˛ r .G; M /.`/W ExtrG .M; C /.`/ ! H 2r .G; M / .`/ is bijective
for all r 2, and ˛ 1 .G; M /.`/ is bijective for all torsion-free M .
(b) The map ˛ 1 .G; M /.`/ is bijective for all M if ˛ 1 .U; Z=`m Z/ is bijective
for all open subgroups U of G and all m:
(c) The map ˛ 0 .G; M / is surjective (respectively bijective) for all finite `-
primary M if in addition ˛ 0 .U; Z=`m Z/ is surjective (respectively bijec-
tive) for all U and m:
p
E XERCISE 1.14 Let K D Q. d / where d is chosen so that the 2-class field
tower of K is infinite. Let K un be the maximal unramified extension of K, and let
H D Gal.K s =K un /. Then .GK =H; C H / is a P -class formation with P D f2g.
Investigate the maps ˛ r .GK =H; M / in this case.
N OTES Theorem 1.8 and its proof are taken from Tate 1966.
2 Local fields
Unless stated otherwise, K will be a nonarchimedean local field, complete with
respect to the discrete valuation ordW K Z, and with finite residue field k.
Let R be the ring of integers in K, and let K un be a largest unramified exten-
sion of K. Write G D Gal.K s =K/ and I D Gal.K s =K un /. As we noted in
(1.6b), .G; K s / has a natural structure of a class formation. The reciprocity map
recG W K ! G ab is known to be injective with dense image. More precisely,
there is an exact commutative diagram
ord
0 ! R ! K ! Z ! 0
? ? ?
? ? ?
y y y
0 ! I ab ! G ab ! b
Z ! 0
in which all the vertical arrows are injective and I ab is the inertia subgroup of
G ab . The norm groups in K are the open subgroups of finite index. See Serre
1962, XIII 4, XIV 6.
In this section N ^ will denote the completion of a group N relative to the
topology defined by the subgroups of N of finite index unless N has a topol-
ogy induced in a natural way from that on K, in which case we allow only sub-
groups of finite index that are open relative to the topology. With this definition,
.R /^ D R , and the reciprocity map defines an isomorphism .K /^ ! GK ab .
s
When M is a discrete G-module, the group HomG .M; K / inherits a topology
from that on K s , and in the next theorem HomG .M; K s /^ denotes its comple-
tion for the topology defined by the open subgroups of finite index6 . As b Z=Z is
uniquely divisible, ˛ 0 .G; Z=mZ/ and ˛ 1 .G; Z=mZ/ are isomorphisms for all m.
Thus most of the following theorem is an immediate consequence of Theorem
1.8.
T HEOREM 2.1 Let M be a finitely generated G -module, and consider
˛ r .G; M /W ExtrG .M; K s / ! H 2r .G; M / :
Then ˛ r .G; M / is an isomorphism for all r 1, and ˛ 0 .G; M / defines an iso-
morphism (of profinite groups)
HomG .M; K s /^ ! H 2 .G; M / .
6 (In original.) If n is prime to the characteristic of K, then K n is an open subgroup of
finite index in K . It follows that every subgroup of K (hence of HomG .M; K s // of finite
index prime to char.K/ is open. In contrast, when the characteristic of K is p ¤ 0, there are

many subgroups Qof finite index in K that are not closed. In fact (see Weil 1967, II 3, Q Pptn
10), 1 C m Zp (product of countably many copies of Zp /, and a proper subgroup of Zp
containing ˚Zp cannot be closed.
2. LOCAL FIELDS 27
The ^ can be omitted if M is finite. The groups ExtrG .M; K s / and H r .G; M /
are finite for all r if M is of finite order prime to char.K/, and the groups
Ext1G .M; K s / and H 1 .G; M / are finite for all finitely generated M whose tor-
sion subgroup is of order prime to char.K/.
P ROOF. We begin with the finiteness statements. For n prime to char.K/, the
cohomology sequence of the Kummer sequence
n
0 ! n .K s / ! K s ! K s ! 0
shows that the cohomology groups are
H r .G; n .K s // D n .K/ K =K n 1
n Z=Z 0
rD 0 1 2 3:
In particular, they are all finite.

Let M be a finite G-module of order prime to char.K/, and choose a finite
Galois extension L of K containing all mth roots of 1 for m dividing the order
of M and such that Gal.K s =L/ acts trivially on M . Then M is isomorphic as
a Gal.K s =L/-module to a direct sum of copies of modules of the form m , and
so the groups H s .Gal.K s =L/; M / are finite for all s, and zero for s 3. The
Hochschild-Serre spectral sequence
H r .Gal.L=K/; H s .Gal.K s =L; M // H) H rCs .G; M /
now shows that the groups H r .G; M / are all finite because the cohomology
groups of a finite group with values in a finite (even finitely generated for r 1)
module are finite. This proves that H r .G; M / is finite for all r and all M of
finite order prime to char.K/, and Theorem 1.8 shows that all the ˛ r .G; M / are
isomorphisms for finite M , and so the groups ExtrG .M; K s / are also finite.
Let M be a finitely generated G-module whose torsion subgroup has order
prime to char.K/. In proving that H 1 .G; M / is finite, we may assume that M is
torsion-free. Let L be a finite Galois extension of K such that Gal.K s =L/ acts
trivially on M . The exact sequence
0 ! H 1 .Gal.L=K/; M / ! H 1 .Gal.K s =K/; M / ! H 1 .Gal.K s =L/; M /
shows that H 1 .G; M / is finite because the last group in the sequence is zero and
the first is finite. Theorem 1.8 implies that ˛ r .G; M / is an isomorphism for r 1
and all finitely generated M , and so Ext1G .M; K s / is also finite.
It remains to prove the assertion about ˛ 0 .G; M /. Note that ˛ 0 .G; Z/ defines
an isomorphism .K /^ ! G ab , and so the statement is true if G acts trivially on
M . Let L be a finite Galois extension of K such Gal.K s =L/ acts trivially on M .

Then HomG .M; K s / D HomG .M; L /, and HomG .M; L / contains an open
compact group HomG .M; Ow /, where O is the ring of integers in L. Using
w
this, it is easy to prove that the maps
0 ! HomG .M1 ; K s / ! HomG .M ; K s / ! HomG .M; K s / !
in the top row of (1.9.1) are strict morphisms. Therefore the sequence remains
exact when we complete the first three terms (see 0.20), and so the same argument
as in (1.8) completes the proof. 2
C OROLLARY 2.2 If M is a countable G -module whose torsion is prime to char.K/,

then
˛ 1 .G; M /W Ext1G .M; K s / ! H 1 .G; M /
is an isomorphism.
P ROOF. Write M as a countable union of finitely generated G-modules Mi and

note that Ext1G .M; K s / D lim Ext1G .Mi ; K s / by (0.23). 2

For any finitely generated G-module M , write M D D Hom.M; K s /. It is
again a discrete G-module, and it acquires a topology from that on K s .
C OROLLARY 2.3 Let M be a finitely generated G -module whose torsion sub-

group has order prime to char.K/. Then cup-product defines isomorphisms
H r .G; M D / ! H 2r .G; M /
for all r 1, and an isomorphism (of compact groups)
H 0 .G; M D /^ ! H 2 .G; M / :
The groups H 1 .G; M / and H 1 .G; M D / are finite.
P ROOF. As K s is divisible by all primes other than char.K/, Extr .M; K s / D

0 for all r > 0, and so ExtrG .M; K s / D H r .G; M D / for all r (see 0.8). 2
C OROLLARY 2.4 Let T be a commutative algebraic group over K whose identity

component T ı is a torus7 . Assume that the order of T =T ı is not divisible by the
characteristic of K , and let X .T / be the group of characters of T . Then cup-
product defines a dualities between:
7 When K has characteristic zero, these are exactly the algebraic groups of multiplicative type.
2. LOCAL FIELDS 29
˘ the compact group H 0 .K; T /^ (completion relative to the topology of open

subgroups of finite index) and the discrete group H 2 .G; X .T //;
˘ the finite groups H 1 .K; T / and H 1 .G; X .T //;
˘ the discrete group H 2 .K; T / and the compact group H 0 .G; X .T //^ (com-
pletion relative to the topology of subgroups of finite index).
In particular, H 2 .K; T / D 0 if and only if X .T /G D 0 (when T is connected,
this last condition is equivalent to T .K/ being compact).
P ROOF. The G-module X .T / is finitely generated without char.K/-torsion,

and X .T /D D T .K s /, and so this follows from the preceding corollary (ex-
cept for the parenthetical statement, which we leave as an exercise — cf. Serre
1964, pII-26). 2
R EMARK 2.5 (a) If the characteristic of K is p ¤ 0 and M has elements of

order p, then Ext1G .M; K s / and H 1 .G; M / are usually infinite. For example
Ext1G .Z=pZ; K s / D K =K p and H 1 .G; Z=pZ/ D K=}K, }.x/ D x p x,
which are both infinite.
(b) If n is prime to the characteristic of K and K contains a primitive nth root
of unity, then Z=nZ n noncanonically and .Z=nZ/D ' n canonically. The
pairing
H 1 .K; Z=nZ/ H 1 .K; n / ! H 2 .K; n / ' Z=nZ
in (2.3) gives rise to a canonical pairing
H 1 .K; n / H 1 .K; n / ! H 2 .G; n ˝ n / ' n :
The group H 1 .K; n / D K =K n , and the pairing can be identified with
.f; g/ 7! .1/v.f /v.g/ f v.g/ =g v.f / W K =K n K =K n ! n
(see Serre 1962, XIV 3).

If K has characteristic p ¤ 0, then the pairing
Ext1G .Z=pZ; K s / H 1 .G; Z=pZ/ ! H 2 .G; K s / ' Q=Z
can be identified with

dg
.f; g/ 7! p 1 Trk=Fp .Res.f //W K =K p K=}K ! Q=Z
g
(see Serre 1962, XIV 5 or III 6 below).

Unramified cohomology
A G-module M is said to be unramified if M I D M . For a finitely gener-
ated G-module, we write M d for the submodule Hom.M; Run / of M D D
Hom.M; K s /. Note that if M is unramified, then H 1 .G=I; M / makes sense
and is a subgroup of H 1 .G; M /. Moreover, when M is finite, H 1 .G=I; M / is
dual to Ext1G=I .M; Z/ (see 1.10).
T HEOREM 2.6 If M is a finitely generated unramified G -module whose torsion

is prime to char.k/, then the groups H 1 .G=I; M / and H 1 .G=I; M d / are the
exact annihilators of each other in the cup-product pairing
H 1 .G; M / H 1 .G; M D / ! H 2 .G; K s / ' Q=Z:
P ROOF. From the spectral sequence (0.3)
ExtrG=I .M; ExtsI .Z; K s // H) ExtrCs s

G .M; K /
and the vanishing of Ext1I .Z; K s / ' H 1 .I; K s /, we find that

Ext1G=I .M; K un / ! Ext1G .M; K s /:
From the split-exact sequence of G-modules
0 ! Run ! K un ! Z ! 0
we obtain an exact sequence
0 ! Ext1G=I .M; Run / ! Ext1G=I .M; K un / ! Ext1G=I .M; Z/ ! 0,
and so the kernel of Ext1G .M; K s / ! Ext1G=I .M; Z/ is Ext1G=I .M; Run /. It
is easy to see from the various definitions (especially the definition of invG in
1.6b) that
˛ 1 .G;M /
Ext1G .M; K s / ! H 1 .G; M /

? ?
? ?
y yInf
˛ 1 .G=I;M /
Ext1G=I .M; Z/ ! H 1 .G=I; M /

commutes. Therefore the kernel of
Ext1G .M; K s / ! H 1 .G=I; M /

2. LOCAL FIELDS 31
is Ext1G=I .M; Run /. Example (0.8) allows us to identify Ext1G .M; K s / with
H 1 .G; M D / and Ext1G=I .M; Run / with H 1 .G=I; M d /, and so the last state-
ment says that the kernel of H 1 .G; M D / ! H 1 .G=I; M / is H 1 .G=I; M d /.
(When M is finite, this result can also be proved by a counting argument; see
Serre 1964, II 5.5.) 2
R EMARK 2.7 A finite G-module M is unramified if and only if it extends to a

finite étale group scheme over Spec.R/. In Chapter III below, we shall see that
flat cohomology allows us to prove a similar result to (2.6) under the much weaker
hypothesis that M extends to a finite flat group scheme over Spec.R/ (see III 1
and III 7).
Euler-Poincaré characteristics
If M is a finite G-module, then the groups H r .G; M / are finite for all r and zero
for r 2. We define
ŒH 0 .G; M /ŒH 2 .G; M /

.G; M / D :
ŒH 1 .G; M /
T HEOREM 2.8 Let M be a finite G -module of order m relatively prime to char.K/.

Then
.G; M / D .RW mR/1:
We first dispose of a simple case.
L EMMA 2.9 If the order of M is prime to char.k/, then .G; M / D 1:
P ROOF. Let p D char.k/. The Sylow p-subgroup Ip of I is normal in I ,

and the quotient I =Ip is isomorphic to b
Z=Zp (see Serre 1962, IV 2, Ex 2). As
H .Ip ; M / D 0 for r > 0, the Hochschild-Serre spectral sequence for I Ip
r
shows that H r .I; M / D H r .I =Ip ; M Ip /, and this is finite for all r and zero
for r > 1 (cf. Serre 1962, XIII 1). The Hochschild-Serre spectral sequence for
G I now shows that H 0 .G; M / D H 0 .G=I; M I /, that H 1 .G; M / fits into
an exact sequence
0 ! H 1 .G=I; M I / ! H 1 .G; M / ! H 0 .G=I; H 1 .I; M // ! 0;
and that H 2 .G; M / D H 1 .G=I; H 1 .I; M //. But G=I ' b Z, and the exact
sequence
1
0 ! H 0 .b
Z; N / ! N ! N ! H 1 .b Z; N / ! 0
(with a generator of b
Z; see Serre 1962, XIII 1) shows that
ŒH 0 .b
Z; N / D ŒH 1 .b
Z; N /
for any finite b

Z-module. Therefore,
ŒH 0 .G=I; M I /ŒH 0 .G=I; H 1 .I; M //

.G; M / D D 1.
ŒH 1 .G=I; M I /ŒH 1 .G=I; H 1 .I; M // 2
Since both sides of equation in (2.8) are additive in M , the lemma allows
us to assume that M is killed by p D char.k/ and that K is of characteristic
zero. We shall prove the theorem for all G-modules M such that M D M GL ,
where L is some fixed finite Galois extension of K contained in K s . Let G D
Gal.L=K/. Our modules can be regarded as Fp ŒG-modules, and we let RFp .G/
, or simply R.G/, be the Grothendieck group of the category of such modules.
Then the left and right hand sides of the equation in (2.8) define homomorphisms
` ; r W R.G/ ! Q>0 . As Q>0 is a torsion-free group, it suffices to show that `
and r agree on a set of generators for RFp .G/ ˝Z Q. The next lemma describes
one such set.
L EMMA 2.10 Let G be a finite group and, for any subgroup H of G , let IndG H
be the homomorphism RFp .H / ˝ Q ! RFp .G/ ˝ Q taking the class of an H -
module to the class of the corresponding induced G -module. Then RFp .G/ ˝ Q
is generated by the images of the IndG
H as H runs over the set of cyclic subgroups
of G of order prime to p .
P ROOF. Write RF .G/ for the Grothendieck group of finitely generated F ŒG-
modules, F any field. Then Serre 1967b, 12.5, Thm 26, shows that, in the case
that F has characteristic zero, RF .G/˝Q is generated by the images of the maps
IndGH with H cyclic. It follows from Serre 1967b, 16.1, Thm 33, that the same
statement is then true for any field F . Finally Serre 1967b, 8.3, Pptn 26, shows
that, in the case that F has characteristic p ¤ 0, the cyclic groups of p-power
order make no contribution. 2
It suffices therefore to prove the theorem for a module M of the form IndGH N.
Let K 0 D LH , let R0 be the ring of integers in K 0 , and let n be the order of N .
Then
.G; M / D .Gal.K s =K 0 /; N /
0
.RW mR/ D .RW nR/ŒK WK D .R0 W nR0 /;
2. LOCAL FIELDS 33
and so it suffices to prove the theorem for N . This means that we can assume that
G is a cyclic group of order prime to p. Therefore H r .G; M / D 0 for r > 0,
and so H r .G; M / D H r .Gal.K s =L/; M /G :
Let 0 be the homomorphism R.G/ ! R.G/ sending a G-module M to
.1/i ŒH i .Gal.K s =L/; M /, where [*] now denotes the class of * in R.G/:
L EMMA 2.11 The following formula holds:
0 .M / D dim.M / ŒKW Qp ŒFp ŒG:
Before proving the lemma, we show that it implies the theorem. Let
W RFp .G/ ! Q>0 be the homomorphism sending the class of a module N to
the order of N G . Then ı 0 D and .ŒFp ŒG/ D p, and so (2.11) shows that
.M / D 0 .M / D p ŒKWQp dim.M / D 1=.RW mR/:
It therefore remains to prove (2.11). On tensoring M with a resolution of

Z=pZ by injective Z=pZŒG-modules, we see that cup-product defines isomor-
phisms of G-modules
H r .Gal.K s =L/; Z=pZ/ ˝ M ! H r .Gal.K s =L/; M /;
and so
0 .M / D 0 .Z=pZ/ ŒM :
Let M0 be the G-module with the same underlying abelian group as M but with
the trivial G-action. The map
˝ m 7! ˝ m
extends to an isomorphism

Fp ŒG ˝ M0 ! Fp ŒG ˝ M;
and so
dim.M / ŒFp ŒG D ŒFp ŒG ŒM :
The two displayed equalities show that the general case of (2.11) is a consequence
of the special case M D Z=pZ.
Note that
H 0 .Gal.K s =L/; Z=pZ/D Z=pZ;

H 1 .Gal.K s =L/; Z=pZ/ D H 1 .Gal.K s =L/; p .K s // D .L =Lp / ;
H 2 .Gal.K s =L/; Z=pZ/ D .p .L// ;
where N denotes Hom.N; Fp / (still regarded as a G-module; as Hom.; Fp /

is exact, it is defined for objects in R.G/). Therefore
0 .Z=pZ/ D ŒZ=pZ ŒL =Lp C Œp .L/:

in R . From the exact sequence
Let U be the group of units RL L
0 ! U=U p ! L =Lp ! Z=pZ ! 0;
we find that
ŒZ=pZ ŒL =Lp D ŒU .p/;
and so
0 .Z=pZ/ D ŒU .p/ C Œp .L/;

D ŒU .p/ C ŒUp :
We need one last lemma.
L EMMA 2.12 Let W and W 0 be finitely generated Zp ŒH -modules for some

finite group H . If W ˝ Qp W 0 ˝ Qp as Qp ŒH -modules, then
ŒW .p/ ŒWp D ŒW 0.p/ ŒWp0
in Fp ŒH :
P ROOF. One reduces the question easily to the case that W W 0 pW , and
for such a module the lemma follow immediately from the exact sequence
0 ! Wp0 ! Wp ! W=W 0 ! W 0.p/ ! W .p/ ! W=W 0 ! 0
given by the snake lemma. 2
The exponential map sends an open subgroup of U onto an open subgroup of

the ring of integers RL of L, and so (2.12) shows that
.p/ .p/
ŒU .p/ ŒUp D ŒRL Œ.RL /p D ŒRL :
The normal basis theorem shows that L Qp ŒGŒKWQp (as G-modules), and so
(2.12) implies that
.p/
ŒRL D ŒK W Qp ŒFp ŒG:
As ŒFp ŒG D ŒFp ŒG, this completes the proof of (2.11).
2. LOCAL FIELDS 35
Archimedean local fields

Corollaries 2.3, 2.4 and Theorem 2.8 all have analogues for R and C.
T HEOREM 2.13 (a) Let G D Gal.C=R/. For any finitely generated G -module
M with dual M D D Hom.M; C /, cup-product defines a nondegenerate pairing
'
HTr .G; M D / HT2r .G; M / ! H 2 .G; C / ! 12 Z=Z
of finite groups for all r:

(b) Let G D Gal.C=R/. For any commutative algebraic group T over
R whose identity component is a torus, cup-product defines dualities between
HTr .G; X .T // and HT2r .G; T .C// for all r .
(c) Let K D R or C, and let G D Gal.C=K/. For any finite G -module M
ŒH 0 .G; M /ŒH 0 .G; M D /

D jmjv .
ŒH 1 .G; M /
P ROOF. (a) Suppose first that M is finite. As G has order 2, the `-primary com-
ponents for ` odd do not contribute to the cohomology groups. We can therefore
assume that M is 2-primary, and furthermore that it is simple. Then M D Z=2Z
with the trivial action of G, and the theorem can be proved in this case by direct
calculation.
When M D Z the result can again be proved by direct calculation, and when
M D ZŒG all groups are zero. Since every torsion-free G-module contains a
submodule of finite index that is a direct sum of copies of Z or ZŒG, this proves
the result for such modules, and the general case follows by combining the two
cases.
(b) Take M D X .T / in (a).
(c) The complex case is obvious because H 0 .G; M / D M and H 0 .G; M D / D
M both have order m, H 1 .G; M / D 0, and jmjv D m2 . In the real case, let
D
generate G, and note that for m 2 M and f 2 M D
..1 /f /.m/ D f .m/= .f . m//

D f .m/ .f . m// (because
D
1 )
D f ..1 C /m/:
Therefore 1 W M D ! M D is adjoint to 1 C W M ! M , and so, in the pairing
M D M ! C ;
.M D /G and NC=R M are exact annihilators. Consequently
ŒH 0 .G; M D /ŒH 0 .G; M /

ŒM D Œ.M D /G ŒNC=R M D ;
ŒHT0 .G; M /
and the periodicity of the cohomology of cyclic groups shows that ŒHT0 .G; M / D
ŒH 1 .G; M /. As ŒM D m D jmjv , this proves the formula. 2
Henselian local fields

Let K be the field of fractions of an excellent Henselian discrete valuation ring
R with finite residue field k. (See Appendix A for definitions.) It is shown in
the Appendix that the pair .GK ; K s / is a class formation, and that the norm
groups are precisely the open subgroups of finite index. The following theorem
generalizes some of the preceding results.
T HEOREM 2.14 Let M be a finitely generated G -module whose torsion sub-

group is prime to char.K/.
(a) The map ˛ r .G; M /W ExtrG .M; K s / ! H 2r .G; M / is an isomor-
phism for all r 1, and ˛ 0 .G; M / defines an isomorphism (of compact groups)
HomG .M; K s /^ ! H 2 .G; M / . The ^ can be omitted if M is finite. The
groups ExtrG .M; K s / and H r .G; M / are finite for all r if M is finite, and the
groups Ext1G .M; K s / and H 1 .G; M / are finite for all finitely generated M:
(b) If K is countable, then for any algebraic group A over K ,
˛ 1 .G; A.K s //W Ext1G .A.K s /; K s / ! H 1 .G; A.K s //
is an isomorphism, except possibly on the p -primary component when char.K/ D

p ¤ 1.
(c) Cup-product defines isomorphisms H r .G; M D / ! H r .G; M / for all
r 1, and an isomorphism H 0 .G; M D /^ ! H 2 .G; M / of compact groups.
The groups H 1 .G; M D / and H 1 .G; M / are both finite.
b be the completion of R. There is a commutative diagram

P ROOF. (a) Let R
0 ! R ! K ! Z ! 0

? ? ?
? ?rec ?
y y y
b ! G ! b
0 ! R Z ! 0:
All the vertical maps are injective, and the two outside vertical maps have cok-
ernels that are uniquely divisible by all primes ` ¤ char.K/. Therefore the
2. LOCAL FIELDS 37
reciprocity map K ! G is injective and has a cokernel that is uniquely divisi-

ble prime to char.K/. The first two assertions now follow easily from (1.8). The
finiteness statements follow from the fact that Gal.K s =K/ D Gal.Kb s =K/
b
s
(b) The group A.K / is countable, and therefore it is a countable union of
finitely generated submodules. The statement can therefore be proved the same
way as (2.2).
(c) The proof is the same as that of (2.3). 2
R EMARK 2.15 (a) Part (a) of the theorem also holds for modules M with p-
torsion, except that it is necessary to complete Ext1G .M; K s /. For example,
when M D Z=pZ, the map ˛ 1 is
K =K p ! Hom.GK ; Z=pZ/:
Because K is excellent, the map K =K p ! K b =K

b p is injective and in-

duces an isomorphism .K =K p /^ ! K b =Kb p . We know Gal.K s =K/ D
Gal.K b and so in this case the assertion follows from the corresponding
b s =K/,
b
statement for K.
(b) As was pointed out to the author by M. Hochster, it is easy to construct
nonexcellent Henselian discrete valuation rings. Let k be a field of characteristic
p, and choose an element u 2 kŒŒt that is transcendental over k.t/. Let R be
the discrete valuation ring k.t; up / \ kŒŒt, and consider the Henselization Rh
of R. Then the elements of Rh are separable over R (Rh is a union of étale
R-subalgebras), and so u … Rh , but u 2 .Rh /^ D kŒŒt.
Complete fields with quasi-finite residue fields

E XERCISE 2.16 Let K be complete with respect to a discrete valuation, but as-
sume that its residue field is quasi-finite rather than finite. (See Appendix A for
definitions.) Investigate to what extent the results of this section continue to hold
for K. References: Serre 1962, XIII, and Appendix A for the basic class field
theory of such fields; Serre 1964, pII-24, pII-29 for statements of what is true;
Vvedens’kii and Krupjak 1976 and Litvak 1980 for a proof of (2.3) for a finite
module in the case the field has characteristic zero.)
d -local fields
A 0-local field is a finite field, and a d -local field for d 1 is a field that is
complete with respect to a discrete valuation and has a .d 1/-local field as
residue field.8 If K is d -local, we shall write Ki , 0 i d , for the i-local
8 These are usually called local fields of dimension d .
field in the inductive definition of K. We write `n for the GK -module f

2
n
K s j
` D 1g, `1 .r/ for lim ˝r , and Z` .r/ for lim ˝r . If M is an `-primary
! `n `n
GK -module, we set M.r/ D M ˝ Z` .r/ and M .r/ D Hom.M; `1 .r//:
T HEOREM 2.17 Let K be a d -local field with d 1, and let ` be a prime ¤
char.K1 /.
(a) There is a canonical trace map

H d C1 .GK ; `1 .d // ! Q` =Z` :
(b) For all GK -modules M of finite order a power of `, the cup-product pairing
H r .GK ; M .d //H d C1r .GK ; M / ! H d C1 .GK ; Q` =Z` .d // ' Q` =Z`
is a nondegenerate pairing of finite groups for all r:
P ROOF. For d D 1, this is a special case of (2.3). For d > 1, it follows by an

easy induction argument from the next lemma. 2
L EMMA 2.18 Let K be any field complete with respect to a discrete valuation,
and let k be the residue field of K . For any finite GK -module of order prime to
char.k/, there is a long exact sequence
! H r .Gk ; M I / ! H r .GK ; M / ! H r1 .Gk ; M.1/I / ! H rC1 .Gk ; M I / !
where I is the inertia group of GK :
P ROOF. Let char.k/ D p, and let Ip be a p-Sylow subgroup

Q of I (so Ip D 1 if
0
p D 1/. Then I Ddf I =Ip is canonically isomorphic to `¤p Z` .1/ (see Serre
1962, IV 2). The same argument that shows that H r .G; M / D M G , MG , 0 for
r D 0, 1, > 2 when G D b Z and M is torsion (Serre 1962, XIII 1), shows in our
case that 8
ˆ
<M ;
I for r D 0
r r 0
H .I; M / D H .I ; M / D M.1/I ; for r D 1 (1)
:̂
0; for r > 1:
The lemma therefore follows immediately from the Hochschild-Serre spectral
sequence for G I: 2
Write Kr R for the r th Quillen K-group of a ring R.

C OROLLARY 2.19 Let K be a 2-local field, and let m be an integer prime to
char.K1 / and such that K contains the mth roots of 1. Then there is a canonical
injective homomorphism .K2 K/.m/ ! Gal.K ab =K/.m/ with dense image.
2. LOCAL FIELDS 39
P ROOF. On taking M D Z=mZ in the theorem, we obtain an isomorphism
H 2 .G; m ˝ m / ! H 1 .G; Z=mZ/ :
But H 1 .G; Z=mZ/ D Homcts .G; Z=mZ/, and so this gives us an injection
H 2 .G; m ˝m / ! .G ab /.m/ with dense image. Now the theorem of Merkur’ev

and Suslin (1982) provides us with an isomorphism .K2 K/.m/ ! H 2 .G; m ˝
m /. 2
Theorem 2.17 is a satisfactory generalization of Theorem 2.3 in the case that

the characteristic drops from p to zero at the first step. The general case is not yet
fully understood.
Some exercises
E XERCISE 2.20 (a) Let G be a profinite group, and let M be a finitely generated
G-module. Write T D Hom.M; C /, and regard it as an algebraic torus over C.
Let G act on T through its action on M . Show that
Ext0G .M; Z/DX .T /G ;

Ext1G .M; Z/ D 0 .T G /;
ExtrG .M; Z/ D H r1 .G; T / for r 2:
If M is torsion-free, show that ExtrG .M; Z/ D H r .G; X .T //.

(b) Let K be a local field (archimedean or nonarchimedean), and let T be a
torus over K. Let T _ be the torus such that X .T _ / D X .T /. Show that the
finite group H 1 .K; T / is dual to 0 .T _G / and that H 1 .K; T _ / is canonically
isomorphic to the group T .K/ of continuous characters of finite order of T .K/.
(In 8 we shall obtain a similar description of the group of generalized characters
of T .K/.) [Hint: To prove the first part of (a), use the spectral sequence (0.8)
H r .G; Exts .M; C // H) ExtrCs

G .M; C /
and the exponential sequence 0 ! Z ! C ! C ! 0. ]

Reference: Kottwitz 1984.
E XERCISE 2.21 Let K be a 2-local field of characteristic zero such that K1 has
characteristic p ¤ 0. Assume
(a) K has p-cohomological dimension 3 and there is a canonical isomor-
phism
H 3 .G; p n ˝ p n / ! Z=p n Z
(Kato 1979, 5, Thm 1);
(b) if K contains a primitive p th root of 1, then the cup-product pairing is a

nondegenerate pairing of finite groups (ibid. 6).
Prove then that (2.17) holds for K with ` D p:
E XERCISE 2.22 Let K D k..t1 ; :::; td // with k a finite field, and let p D char.k/.
C 1
Define .r/ D Ker.˝K=k;d
r
D0
! ˝K=k
r /, where C is the Cartier operator (see
'
Milne 1976). Show that there is a canonical trace map H 1 .GK ; .d // ! Z=pZ,
and show that the cup-product pairings
H r .GK ; .r// H 1r .GK ; .d r// ! H 1 .GK ; .d // ' Z=pZ
are nondegenerate in the sense that their left and right kernels are zero. Let d D 2,
and assume that there is an exact sequence
p
0 ! K2 K ! K2 K ! .2/ ! 0
with the second map being dlog ^ dlogW K2 K ! .2/ . (In fact such a sequence
exists: the exactness at the first term is due to Suslin 1983; the exactness at the
middle term is a theorem of Bloch (Bloch and Kato 1986); and the exactness at
the last term has been proved by several people.) Deduce that there is a canonical
injective homomorphism .K2 K/.p/ ! .GK ab /.p/ . (These results can be extended
to groups killed by powers of p rather that p itself by using the sheaves n .r/ of
Milne 1986a.)
N OTES The main theorems concerning local fields in the classical sense are due to Tate.
The proofs are those of Tate except for that of (2.8), which is due to Serre (see Serre
1964, II 5). Theorem 2.17 is taken from Deninger and Wingberg 1986.
3 Abelian varieties over local fields

We continue with the notations at the start of the last section. In particular, K
is a local field, complete with respect to a discrete valuation ord, and with finite
residue field k. When G and H are algebraic groups over a field f , we write
ExtrF .G; H / for the group formed in the category GpF (see 0).
Let A be an abelian variety over K, and let At be the dual abelian variety. The
Barsotti-Weil formula (Serre 1959, VII, 3) states that At .K s / D Ext1K s .A; Gm /.
L EMMA 3.1 For any abelian variety A over a perfect field F , there is a canonical
isomorphism
H r .F; At / ! ExtrC1
F .A; Gm /;
all r 0:
3. ABELIAN VARIETIES OVER LOCAL FIELDS 41
P ROOF. The group ExtrF s .A; Gm / is shown to be zero for r 2 in (Oort 1966,
Pptn 12.3), and HomF s .A; Gm / D 0 because all maps from a projective variety
to an affine variety are constant. This together with the Barsotti-Weil formula
shows that the spectral sequence (0.17)
H r .Gal.F s =F /; ExtsF s .A; Gm // H) ExtrCs

F .A; Gm /

degenerates to a family of isomorphisms H r .F; At / ! ExtrC1
F .A; Gm /: 2
In particular Ext1K .A; Gm / D At .K/ when K has characteristic zero, and

the Ext group therefore acquires a topology from that on K. Recall that there is
a canonical pairing (0.16)
ExtrK .A; Gm / H 2r .K; A/ ! H 2 .K; Gm /;

'
and an isomorphism invG W H 2 .K; Gm / ! Q=Z (1.6b). Therefore there is a
canonical map
˛ r .K; A/W ExtrK .A; Gm / ! H 2r .K; A/ .
T HEOREM 3.2 If K has characteristic zero, then ˛ 1 .K; A/ is an isomorphism of

compact groups

Ext1K .A; Gm / ! H 1 .K; A/
and ˛ 2 .K; A/ is an isomorphism of torsion groups of cofinite type

Ext2K .A; Gm / ! A.K/ :
For r ¤ 1; 2, ExtrK .A; Gm / and H 2r .K; A/ are both zero.
We first need a lemma.
L EMMA 3.3 In the situation of the theorem, A.K/ contains an open subgroup of
finite index isomorphic to Rdim.A/ ; therefore A.K/ D A.K/^ (completion for
the profinite topology), and
ŒA.K/.n/ =ŒA.K/n D .R W nR/dim.A/ :
P ROOF. The existence of the subgroup follows from the theory of the logarithm
(see Mattuck 1955 or Tate 1967b, p168–169), and the remaining statements are
obvious. 2
P ROOF ( OF 3.2) From

n
0 ! An ! A ! A ! 0
we get the rows of the following diagram
0 ! ExtrK .A; Gm /.n/ ! ExtrK .An ; Gm / ! ExtrC1

K .A; Gm /n ! 0
? ? ?
? r ? r ? rC1
y˛ .K;A/.n/ y˛ .K;An / y˛ .K;A/n
0 ! .H 2r .K; A/n / ! H 2r .K; An / ! H 1r .K; A/.n/ ! 0

As is explained in (0.18), ExtrK .An ; Gm / ! ExtrG .An .K s /; K s / for all r, and
if we take ˛ r .G; An / to be the map ˛ r .G; An .K s // of 2, then it is clear that the
diagram commutes. As ˛ r .G; An / is an isomorphism of finite groups for all r,
we see that
˛ r .K; A/.n/ W ExtrK .A; Gm /.n/ ! .H 2r .K; A/n /
is an injective map of finite groups for all r, and, in the limit,
lim ˛ r .K; A/.n/ W lim ExtrK .A; Gm /.n/ ! .H 2r .K; A/tors /

is injective. As Ext1K .A; Gm / D At .K/, the lemma shows that
Ext1K .A; Gm / D lim Ext1K .A; Gm /.n/ :

Thus we have shown that ˛ 1 .K; A/ is injective.
We next show that H r .K; A/ D 0 for r 2. For r > 2, this follows from
the fact that G has cohomological dimension 2 (see 2.1). On taking r D 0 in the
above diagram, we get an exact commutative diagram
0 ! 0 ! HomK .An ; Gm / ! Ext1K .A; Gm /

? ? ?
? ? ?
y y y
0 ! .H 2 .K; A/n / ! H 2 .K; An / ! H 1 .K; A/ :
As the right hand vertical arrow is injective, the snake lemma shows that
H 2 .K; A/n D 0, and therefore that H 2 .K; A/ D 0. Because
H r .G; At / ' ExtrC1

K .A; Gm /;
this also shows that ExtrK .A; Gm / D 0 for r ¤ 1; 2:

We now prove that ˛ 1 .K; A/ is an isomorphism. We have already seen that it

is an injective map At .K/ ! H 1 .K; A/ , and it remains to show that the maps
At .K/.n/ ! .H 1 .K; A/n / are surjective for all integers n. As these maps are
injective, this can be accomplished by showing that the groups have the same
order. Let M D An .K s / and M D D Atn .K s /. Then (2.8) shows that
.G; M / D .R W nR/2d D .G; M D /;
where d is the dimension of A, and (3.3) shows that
ŒA.K/.n/ =ŒA.K/n D .R W nR/d D ŒAt .K/.n/ =ŒAt .K/n :
From the cohomology sequence of

n
0 ! M ! A.K s / ! A.K s / ! 0
we find that
ŒA.K/n ŒH 2 .G; M /
.G; M / D
ŒA.K/.n/ ŒH 1 .K; A/n
or
1 1 ŒH 0 .G; M D /
D :
.RW nR/2d .R W nR/d ŒH 1 .G; A/n
As H 0 .G; M D / At .K/n , this can be rewritten as
ŒH 1 .G; A/n D .R W nR/d ŒAt .K/n D ŒAt .K/.n/ ;
which completes the proof that ˛ 1 .K; A/ is an isomorphism.

It remains to show that ˛ 2 .K; A/ is an isomorphism. The diagram at the
start of the proof shows that ˛ 2 .K; A/ is surjective, and we know that it can be
identified with a map H 1 .K; At / ! A.K/ . The above calculation with A and
At interchanged shows that ŒH 1 .K; At /n D ŒA.K/.n/ for all n, which implies
that ˛ 2 .K; A/ is an isomorphism. 2
C OROLLARY 3.4 If K has characteristic zero, then there is a canonical pairing
H r .K; At / H 1r .K; A/ ! Q=Z;

which induces an isomorphism of compact groups At .K/ ! H 1 .K; A/ (case

r D 0) and an isomorphism of discrete groups of cofinite-type H 1 .K; At / !
A.K/ (case r D 1). For r ¤ 0; 1, the groups H r .K; A/ and H r .K; At / are
zero.
P ROOF. Lemma 3.1 allows us to replace ExtrK .A; Gm / in the statement of the
theorem with H r1 .K; At /. 2
R EMARK 3.5 There is an alternative approach to defining the pairings
H r .K; At / H 1r .K; A/ ! H 2 .K; Gm /
of (3.4). For any abelian variety A over K, write Z.A/ forP the group of zero cycles
on AP K s of degree zero (that is, the set of formal sums ni Pi with Pi 2 A.K s /
and ni D 0). There is a surjective map SW Z.A/ A.K s / sending a formal
sum to the corresponding actual sum on A, and we write Y.A/ for its kernel.
There are exact sequences
0 ! Y.At / ! Z.At / !At .K s / ! 0;

0 ! Y.A/ ! Z.A/ !A.K s / ! 0:
Let D be a divisor on At A, and let a and b be elements of Y.At / and Z.A/ such
that the support of D does not meet the support of a b. The projection D.a/ of
D .a b/ onto A is then defined and, because a is in Y.A/, it is principal, say
D.a/ D div.f /. It is now possible to define
df Q ordb .b/
D.a; b/ D f .b/ D b2supp.b/ f .b/ 2 K s :
Now let D be a Poincaré divisor on At A, and let D t be its transpose. A

reciprocity law (Lang 1959, VI 4, Thm 10), shows that the pairings
.a; b/ 7! D.a; b/W Y.At / Z.A/ ! K s

.b; a/ 7! D t .b; a/W Y.A/ Z.At / ! K s
satisfy the equality D.a; b/ D D t .b; a/ if a 2 Y.At / and b 2 Y.A/. They

therefore give rise to augmented cup-product pairings (0.12)
H r .K; At / H 1r .K; A/ ! H 2 .K; Gm /:
It is possible to show that these pairings agree with those in (3.4) (up to sign)
by checking that each is compatible with the pairings
H r .K; Atn / H 2r .K; An / ! H 2 .K; Gm /
defined by the en -pairing Atn An ! Gm . Alternatively, one can show directly

that the maps H r .K; A/ ! ExtrC1 t s s
G .A .K /; K / defined by this pair of pair-
ings (see 0.14b) equal those defined by the Barsotti-Weil formula. (In fact the
best way of handling these pairings is to make use of biextensions and derived
categories, see Chapter III, especially Appendix C.)
R EMARK 3.6 When K has characteristic p ¤ 0, (3.4) can still be proved by

similarly elementary methods provided one omits the p-parts of the groups. More
precisely, write A.K/.non-p/ for lim A.K/.n/ where n runs over all integers not

divisible by p, and let
L
H r .K; A/.non-p/ D `¤p H r .K; A/.`/ for r > 0:
Then the pair of pairings in (3.5) defines augmented cup-products
H r .K; At / H 1r .K; A/ ! Q=Z,
which induce an isomorphism of compact groups

At .K/.non-p/ ! H 1 .K; A/.non-p/ (case r D 0)
and an isomorphism of discrete groups of cofinite type

H 1 .K; At /.non-p/ ! A.K/.non-p/ (case r D 1).
For r ¤ 0; 1, the groups H r .K; At /.non-p/ and H r .K; A/.non-p/ are zero.
Probably this can be proved by the same method as above, but I have not
checked this. Instead I give a direct proof.
There is a commutative diagram
0 ! At .K/.n/ ! H 1 .K; Atn / ! H 1 .K; At /n ! 0
? ? ?
? ? ?
y y y
0 ! H 1 .K; A/n ! H 1 .K; An / ! A.K/.n/ ! 0
for all n prime to p. As the middle vertical arrow is an isomorphism (2.3), we see
on passing to the limit that
H 1 .K; At /.non-p/ ! A.K/.non-p/
is surjective. To show that it is injective, it suffices to show that for any n prime
to p, H 1 .K; At /n ! A.K/.n/ is injective, and this we can do by showing that
the two groups have the same order. There is a subgroup of finite index in A.K/
that is uniquely divisible by all integers prime to p (namely, the kernel of the
specialization map A.R/ ! A0 .k/, where A is the Néron model of A; it follows
from Hensel’s lemma that this is uniquely divisible prime to the characteristic of
k). Consequently ŒA.K/n D ŒA.K/.n/ . Now the same argument as in the proof
of (3.2) shows that the groups in question have the same order. The rest of the
proof is exactly as in (3.2).
In 7 of Chapter III, we shall use flat cohomology to prove that (3.4) is valid
even for the p-components of the groups.
R EMARK 3.7 The duality in (3.4) extends in a rather trivial fashion to archimedean
local fields. Let G D Gal.C=R/ and let A be an abelian variety over R. Then the
pair of pairing in (3.5) defines a pairing of finite groups
HTr .G; At .C// HT1r .G; A.C// ! HT2 .G; C / ' 12 Z=Z
for all integers r. The pairing can be seen to be nondegenerate from the following
diagram
0 ! HTr1 .G; At / ! HTr .G; At2 / ! HTr .G; At / ! 0
? ? ?
? ? ?
y y y
0 ! HTr1 .G; A/ ! HT2r .G; A2 / ! HT1r .G; A/ ! 0:
Part (a) of (2.13) shows that the middle arrow is an isomorphism, and the two
ends of the diagram show respectively that
HTr .G; At / ! HT1r .G; A/
is injective for all r and surjective for all r.

The group A.R/ı is a connected, commutative, compact real Lie group of di-
mension dim.A/, and therefore it is isomorphic to .R=Z/dim.A/ . The norm map
A.C/ ! A.R/ is continuous and A.C/ is compact and connected, and so its
image is a closed connected subgroup of A.R/. Since it contains the subgroup
2A.R/ of A.R/, which has finite index in A.R/, it must also be open, and there-
fore it equals A.R/ı. Consequently HT0 .G; A/ D 0 .A.R//. The exact sequence
0 ! A.R/ı2 ! A.R/2 ! 0 .A/ ! 0
shows that
Œ0 .A/ 2dim.A/ D ŒA.R/2 ,
and so H 1 .R; At / ¤ 0 if and only if ŒA.R/2 > 2dim.A/ . For example, when A
is an elliptic curve, H 1 .R; A/ ¤ 0 if and only if (in the standard form) the graph
of A in R R intersects the x-axis in three points.
When A is an algebraic group over a field k, we now write 0 .A/ for the
set of connected components (for the Zariski topology) of A over k s ; that is,
0 .A/ D Aks =Aıks regarded as a G-module.
P ROPOSITION 3.8 Let A be an abelian variety over K , and let A be its Néron
model over R. Then
H 1 .G=I; A.K un // D H 1 .G=I; 0 .A0 //

where A0 is the closed fibre of A=R. In particular, if A has good reduction, then
H 1 .G=I; A.K un // D 0:
P ROOF. Let Aı be the open subgroup scheme of A whose generic fibre is A

and whose special fibre is the identity component of A0 . Because A is smooth
over R, Hensel’s lemma implies that the reduction map A.Run / ! A0 .k s / is
surjective (see, for example, Milne 1980, I 4.13), and it follows that there is an
exact sequence
0 ! Aı .Run / ! A.Run / ! 0 .A0 / ! 0:
Moreover A.Run / D A.K un / (because A ˝R Run is the Néron model of A ˝K

K un /, and so it remains to show that H r .G=I; Ao .Run // D 0 for r D 1; 2.
An element ˛ of H 1 .G=I; Aı .Run // can be represented by an Aı -torsor P .
As Aı0 is a connected algebraic group over a finite field, Lang’s lemma (Serre
1959, VI.4) shows that the Aı0 -torsor P ˝R k is trivial, and so P .k/ is nonempty.
Hensel’s lemma now implies that P .R/ is nonempty, and so ˛ D 0:
Finally, for each n, H 2 .G=I; Aı .Run =mn // D 0 because G=I has cohomo-
logical dimension 1, and this implies that H 2 .G=I; Aı .Run // D 0 (Serre 1967a,
I 2, Lemma 3). 2
R EMARK 3.9 The perceptive reader will already have observed that the proof of
the proposition becomes much simpler if one assumes that A has good reduction.
R EMARK 3.10 (a) Let R be an excellent Henselian discrete valuation ring with
finite residue field, and let K be the field of fractions of R. For any abelian variety
A over K, let A.K/^ be the completion of A.K/ for the topology defined by K.
Let Kb be the completion of K. Then
(i) the map A.K/^ ! A.K/ b is an isomorphism;
(ii) the map H .K; A/ ! H 1 .K;
1 b A/ is an isomorphism.
Therefore the augmented cup-product pairings
H r .K; At / H 1r .K; A/ ! H 2 .K; Gm /
induce isomorphisms At .K/^ ! H 1 .K; A/ and H 1 .K; At / ! A.K/ :

To prove (i) we have to show that every element of A.K/b can be approxi-
mated arbitrarily closely by an element of A.K/, but Greenberg’s approximation
b can be approximated
theorem (Greenberg 1966) says that every element of A.R/
b b
arbitrarily closely by an element of A.R/, and A.R/ D A.K/:
1 1 b
The injectivity of H .K; A/ ! H .K; A/ also follows from Greenberg’s
theorem, because an element of H 1 .K; A/ is represented by a torsor P over K,
which extends to a flat projective scheme P over R; if P .K/b is nonempty, then

P.R=m / is nonempty for all i, which (by Greenberg’s theorem) implies that
i
P.R/ is nonempty. For the surjectivity, one endows H 1 .K; b A/ with its natural
1
topology, and observes that H .K; A/ is dense in it (because, for any finite Ga-
lois extension L of K, Z 1 .L=K; A/ has a natural structure as an algebraic group
(Milne 1980, p115), and so Greenberg’s theorem can be applied again). Proposi-
b A/ is discrete.
tion 3.8 then shows that the topology on H 1 .K;
E XERCISE 3.11 Investigate to what extent the results of this section continue to
hold when K is replaced by a complete local field with quasi-finite residue field.
N OTES The duality between H 1 .K; At / and H 1 .K; A/ in (3.4) was the first major theo-
rem of the subject (see Tate 1957/58); it was proved before (2.3), and so can be regarded
as the forerunner of the rest of the results in this chapter. The proof of Theorem 3.2 is
modelled on a proof of Tate’s of (3.4) (cf. Milne 1970/72, p276). The description of the
pairing in (3.4) given in (3.5) is that of Tate’s original paper. Proposition 3.8 can be found
in Tate 1962 in the case of good reduction; the stronger form given here is well known.
4 Global fields
Throughout this section, K will be a global field, and S will be a nonempty set
of primes of K, containing the archimedean primes in the case the K is a number
field. If F K, then the set of primes of F lying over primes in S will also be
denoted by S (or, occasionally, by SF /. We write KS for the maximal subfield
of K s that is ramified over K only at primes in S, and GS for Gal.KS =K/. Also
T
RK;S D v…S Ov D fa 2 K j ordv .a/ 0 for all v … Sg
denotes the ring of S-integers in K. For each prime v we choose an embedding

(over K/ of K s into Kvs , and consequently an extension w of v to K s and an
identification of Gv Ddf Gal.Kvs =Kv / with the decomposition group of w in
GK .
Let P denote the set of prime numbers ` such that `1 divides the degree of
KS over K. If K is a function field, then P contains all prime numbers because
KS contains Kk s where k s is the separable closure of the field of constants of
K. If K is a number field, then P contains at least all the primes ` such that
`RK;S D RK;S (that is, such that S contains all primes dividing `/ because for
such primes, KS contains the `m th roots of 1 for all m. (It seems not to be known9
9 Haberland (1978) simply assumes that P is always the set of all prime numbers. Thus, many
of his theorems are false, or at least, not proven. This is unfortunate, since his book has been widely
used as a reference.
4. GLOBAL FIELDS 49
how large P is in the number field case; for example, if K D Q and S D f`; 1g,
is P the set of all prime numbers?10 )
For a finite extension F of K contained in KS , we use the following notations:
JF D the group of idèles of F I
Q
JF ;S D f.aw / 2 JF j aw D 1 for w … Sg ' 0w2S Fw (restricted topologi-
cal product relative to the subgroups O bw );
T
RF ;S D w…S Ow D ring of SF -integers (D integral closure of RK;S in
F );
EF ;S D RF
;S D group of SF -units;
CF ;S D JF ;S =EF ;S D group of SF -idèle classes;
U D f.aw / 2 JF j aw 2 O bw for w … S, aw D 1 otherwiseg '
Q F ;S
b
w…S O w :
Define
JS D lim JF ;S , RS D lim RF ;S , ES D lim EF ;S ,
! ! !
CS D lim CF ;S , US D lim UF ;S ;
! !
where the limit in each case is over all finite extensions F of K contained in KS :
When S contains all primes of K, we usually drop it from the notation. In
this case KS D K s , GS D GK , and P contains all prime numbers. Moreover
JF ;S D JF , RF ;S D F , EF ;S D F , and CF ;S D CF is the idèle class group
of F . Since everything becomes much simpler in this case, the reader is invited
to assume S contains all primes on a first reading.
A duality theorem for the P -class formation .GS ; CS /

Let CS .F / D CF =UF ;S ; we shall show that .GS ; CS / is a P -class formation
Gal.KS =F /
with CS D CS .F /. Note that when S contains all primes, .GS ; CS / is
the class formation .G; C / considered in (1.6c) and CS .F / D CF :
L EMMA 4.1 There is an exact sequence
0 ! CF ;S ! CF =UF ;S ! IdF ;S ! 0;
where IdF ;S is the ideal class group of RF ;S . In particular, if S omits only finitely
'
many primes, then IdF ;S D 1 and CF ;S ! CF =UF ;S :
P ROOF. Note that F \ UF ;S D f1g and JF ;S \ .F UF ;S / D EF ;S (intersec-

tions inside JF /. Therefore UF ;S can be regarded as a subgroup of CF and the
10 As far as I know, this question is still unanswered.

injection JF ;S ,! JF induces an injection CF ;S ,! CF =UF ;S . The cokernel of

this last map is
L
JF =JF ;S UF ;S F ' . v…S Z/=Im.F /;
which can be identified with the ideal class group of RF ;S . If S omits only
finitely many primes, then RK;S is a Dedekind domain with only finitely many
prime ideals, and any such ring is principal. 2
P ROPOSITION 4.2 The pair .GS ; CS / is a P -class formation and
CSGS D CK =UK;S :
P ROOF. As we observed in (1.6c), .G; C / is a class formation, and so .GS ; C HS /,

where HS D Gal.K s =KS /, is a P -class formation (see the discussion preced-
ing 1.13). The next two lemmas show that there is a canonical isomorphism
H r .GS ; C HS / ! H r .GS ; CS / for all r 1, and since the same is true for any
open subgroup of GS , it follows that .GS ; CS / is also a P -class formation. 2
P ROPOSITION 4.3 There is a canonical exact sequence
0 ! US ! C HS ! CS ! 0:
P ROOF. When S is finite, on passing to the direct limit over the isomorphisms

CF ;S ! CF =UF ;S we obtain an isomorphism CS ! C HS =US , which gives the
exact sequence. In the general case, we have to show that lim IdF ;S D 0. Let L
!
be the maximal unramified extension of F (in K s / in which all primes of S split,
and let F 0 be the maximal abelian subextension of L=F . Thus F 0 is the maximal
abelian unramified extension of F in which all primes of S split (that is, such that
all primes in S are mapped to 1 by the reciprocity map). Class field theory (Tate
1967a, 11.3) gives us a commutative diagram
'
IdF ;S ! Gal.L=F /ab Gal.F 0 =F /
? ?
? ?
y yV
'
idF 0 ;S ! Gal.L=F 0 /ab
with V the transfer (that is, Verlagerung) map. The principal ideal theorem (Artin
and Tate 1961, XIII 4) shows that V is zero. Since similar remarks hold for all
finite extensions F of K contained in KS , we see that lim IdF ;S D 0 (direct
!
limit over such F /, and this completes the proof. 2
4. GLOBAL FIELDS 51
L EMMA 4.4 With the above notations, H r .GS ; US / D 0 for r 1. Therefore

the cohomology sequence of the sequence in (4.3) gives isomorphisms CS .K/ !
CSGS and H r .GS ; C HS / ! H r .GS ; CS /, r 1:
P ROOF. By definition,
H r .GS ; US / D lim H r .Gal.F=K/; UF ;S /
!F
Q
D lim H r .Gal.F=K/; w…SF O bw /:
!F
The cohomology of finite groups commutes with products, and so
Q Q Q
H r .Gal.F=K/; bw / D
O H r .Gal.F=K/; b w /;
O
w…SF v…SK wjv
Q Q
D H r .Gal.Fw =Kv /; b
O w /;
v…SK wjv
where in the last product w denotes the chosen prime w lying over v. Now
b
H r .Gal.Fw =Kv /; O w / D 0; r 1;
because v is unramified in F (cf. Serre 1967a, Pptn 1), and this completes the
proof because .C HS /GS D CK (Tate 1967a, 8.1). 2
We write DS .F / and DF for the identity components of CS .F / and CF .

When K is a function field, the idèle groups are totally disconnected, and so their
identity components reduce to the identity element.
L EMMA 4.5 Assume that K is a number field. Then DS .K/ D DK UK;S =UK;S .
It is divisible, and there is an exact sequence
rec
0 ! DS .K/ ! CS .K/ ! GSab ! 0:
P ROOF. When S contains all primes of K, this is a standard part of class field
theory; in fact DK is the group of divisible elements in CK (Artin and Tate 1961,
VII, IX). The identity component of CS .K/ is the closure of the image of the
identity component of CK . As UK;S is compact, CK ! CS .K/ is a proper
map, and so the image of the identity component is already closed. This proves
the first statement, and DS .K/ is divisible because it is a quotient of a divisible
group. The image of UK;S in G ab is the subgroup fixing KS \ K ab , which is also
the kernel of G ab ! GSab , and the existence of the exact sequence follows from
applying the snake lemma to the diagram
UK;S ! Gal.K ab =KS \ K ab / ! 0
? ?
? ?
y y
0 ! DK ! CK ! Gal.K ab =K/ ! 0: 2
T HEOREM 4.6 Let M be a finitely generated GS -module, and let ` 2 P .

(a) The map
˛ r .GS ; M /.`/W ExtrGS .M; CS /.`/ ! H 2r .GS ; M / .`/
is an isomorphism for all r 1.

(b)11 Let K be a number field, and choose a finite totally imaginary Galois
extension L of K contained in KS and such that Gal.KS =L/ fixes M ; if P
contains all prime numbers or if M is a finite module such that ŒM RS D RS ,
then there is an exact sequence
NL=K ˛0
Hom.M; DS .L// ! HomGS .M; CS / ! H 2 .GS ; M / ! 0:
(c) Let K be a function field; for any finitely generated GS -module, there is
an isomorphism
HomGS .M; CS /^ ! H 2 .GS ; M /
where ^ denotes the completion relative to the topology of open subgroups of
finite index.
11 For
(b) to be true, it is necessary to take L sufficiently large. What follows is an email from
Bill McCallum.
Here is the point: suppose you have a totally imaginary field K which does not satisfy Leopoldt’s
conjecture for some prime p, i.e., there is a non-trivial kernel to the map
E 0 ˝ Zp ! completion at p:
Let S be the primes above p and infinity. Let CS .K/ and DS .K/ be as [in the book]. Then
DS .K/ will have an infinitely p-divisible part coming from the primes at infinity, and an extra p-
divisible part coming from the Leopoldt kernel (given an element of the Leopoldt kernel, construct
a sequence of idèles by taking its p n -th roots at primes above p and infinity, 1 everywhere else).
Now, I believe that the kernel of the map
˛ 0 .GS ; Z=pZ/W Hom.Z=pZ; CS / ! H 2 .GS ; Z=pZ/
should be just the part of DS .K/ coming from the infinite primes, no more. I think the statement
would be correct if you replaced “choose a finite totally imaginary Galois extension L of K” with
“choose a sufficiently large finite ...”
Specifically, sufficiently large would be to adjoin enough p-power roots of unity, because the
inverse limit of the Leopoldt kernels under norm is zero as you go up the cyclotomic tower. So the
norms would capture just the infinite part, as required.
I noticed that Tate says “sufficiently large” in his Congress announcement of 1962, although as
you point out in the book he later makes the mistake of saying that GS has strict cohomological
dimension 2, which means that Leopoldt’s conjecture is satisfied and the hedge isn’t necessary!
If I am right, you would have to correct the statement [in the proof of (b)]: “It follows easily
that the sequence is exact whenever GS acts trivially on M and L D K.” I tried to verify this and
ran into the problem that I had to assume H 3 .GS ; Z/ D 0, which is the point of Tate’s mistake.
You would have to replace this sentence with a more detailed argument, and the hypothesis would
be “whenever GS acts trivially on M and L is sufficiently large.”
4. GLOBAL FIELDS 53
P ROOF. Assume first that K is a number field. Lemma 4.5 shows that, for all
` 2 P and all m, ˛ 1 .GS ; Z=`m Z/ is bijective and ˛ 0 .GS ; Z=`m Z/ is surjective.
Thus it follows from (1.13) that part (a) of the theorem is true for number fields
and that ˛ 0 .GS ; M /.`/ is surjective for finite M .
For (b), note first that when M D Z and L D K, the sequence becomes
that in the lemma. It follows easily that the sequence is exact whenever GS acts
trivially on M and L D K. Let M and L be as in (b), and consider the diagram
(1.9.1) in the proof of (1.8):
HomGS .M1 ; CS / ! HomU .M; CS / ! HomGS .M; CS / ! !

? ? ? ?
? ? ? ?
y y y y
H 2 .GS ; M1 / ! H 2 .U; M / ! H 2 .GS ; M / ! ! :
Here U D Gal.KS =L/. All vertical maps in the diagram are surjective, and so
we get an exact sequence of kernels:
Ker.˛ 0 .GS ; M1 // ! Ker.˛ 0 .U; M // ! Ker.˛ 0 .GS ; M // ! 0:
We have already observed that the kernel of ˛ 0 .U; M / is Hom.M; DS .L//, and
therefore the kernel of ˛ 0 .GS ; M / is the image NL=K .Hom.M; DS .L// of this
in HomG .M; CS /:
When K is a function field, recG W CK ! G ab is injective with dense image.
More precisely, there is an exact sequence
0 ! CK ! G ab ! b
Z=Z ! 0
and the first arrow induces a topological isomorphism of fa 2 CK j jaj D 1g onto

the open subgroup Gal.K s =Kk s / of G ab (Artin and Tate 1961, 8.3). From this
we again get an exact sequence
0 ! CSGS ! GSab ! b
Z=Z ! 0
As bZ=Z is uniquely divisible, part (a) of the theorem follows in this case directly
from (1.8). Part (c) can be proved by a similar argument to that which completes
the proof of (2.1). 2
We next reinterpret (4.6) as aQstatement about the cohomology of an algebraic

torus T over K. Let AF ;S D 0w2S Fw be the ring of S-adèles of F , and let
AS D lim AF ;S , where the limit is again over finite extensions of K contained in
!
KS . As for any algebraic group over K, it is possible to define the set T .AF ;S /
of points of T with values in AF ;S , and we let T .AS / D lim T .AF ;S /. If T is
!
split by KS , then T .AS / D X .T / ˝Z JS . This suggests the definition T .RS / D

X .T / ˝Z ES . A cocharacter 2 X .T / defines compatible maps
T .RS / ! ES ; T .AS / ! JS ;
and hence a map T .AS /=T .RS / ! JS =ES D CS . We have therefore a pairing
X .T / T .AS /=T .RS / ! CS ;
which induces cup-product pairings
H r .GS ; X .T // H 2r .GS ; T .AS /=T .RS // ! H 2 .GS ; CS / ! Q=Z.
C OROLLARY 4.7 Let T be a torus over K split by KS , and let ` 2 P . Then the
cup-product pairings defined above induce dualities between:
the compact group H 0 .GS ; X .T //^ (`-adic completion) and the discrete
group
H 2 .GS ; T .AS /=T .RS //.`/I
the finite groups H 1 .GS ; X .T //.`/ and H 1 .GS ; T .AS /=T .RS //.`/;
and, when P contains all the prime numbers, the discrete group H 2 .GS ; X .T //
and the compact group H 0 .GS ; T .AS /=T .RS //^ (completion of the topology of
open subgroups of finite index).
P ROOF. As we saw in (1.11), ExtrGS .X .T /; CS / D H r .GS ; X .T /˝CS /. On
tensoring the exact sequence
0 ! ES ! JS ! CS ! 0
with X .T /, we find that X .T / ˝ CS D T .AS /=T .RS /. Therefore
ExtrGS .X .T /; CS / D H r .GS ; T .AS /=T .RS //;
and part (a) of the theorem gives us isomorphisms
H r .GS ; T .AS /=T .RS //.`/ ! H 2r .GS ; X .T // .`/; r 1:
As H s .GS ; X .T // is obviously finite for s D 1 and is finitely generated for
s D 0, this proves the first two assertions. In the function field case, we also have
an isomorphism
H s .GS ; T .AS /=T .RS //^ ! H 2 .GS ; X .T // :
In the number field case, completing the exact sequence
Hom.X .T /; DS .L// ! H 0 .GS ; T .AS /=T .RS // ! H 2 .GS ; X .T // ! 0
given by (4.6b) yields the required isomorphism because the first group is divisi-
ble. 2
4. GLOBAL FIELDS 55
Statement of the main theorem

The rest of this section is devoted to stating and proving Tate’s theorem (Tate
1962, Thm 3.1), which combines the dualities so far obtained for local and global
fields. From now on, M is a finitely generated GS -module the order of whose
torsion subgroup is a unit in RS .
For v a prime of K, let Gv D Gal.Kvs =Kv /. In the nonarchimedean case, we
write k.v/ for the residue field at v, and gv D Gal.k.v/s =k.v// D Gv =Iv . The
choice of the embedding K s ,! Kvs determines maps Gv ! GK ! GS , and
using these maps we obtain localization maps HSr .GS ; M / ! H r .Gv ; M / for
each GS -module M . We write H r .Kv ; M / D H r .Gv ; M / except in the case
that v is archimedean, in which case we set H r .Kv ; M / D HTr .Gv ; M /. Thus
H 0 .R; M / D M Gal.C=R/ =NC=R M and H 0 .C; M / D 0. When v is archimedean
and M is unramified at v, we write Hun r .K ; M / for the image of H r .g ; M / in
v v
H .Gv ; M /. Thus, Hun .Kv ; M / D H 0 .Kv ; M /, Hun
r 0 1 .K ; M / ' H 1 .g ; M /,
v v
and, unless M has elements of infinite order, Hun2 .K ; M / D 0. A finitely gener-
v
ated GS -module M is unramified for all but finitely many v in S, and we define
PSr .K; M / to be the restricted topological product of the H r .Kv ; M / relative to
the subgroups Hun r .K ; M /. Thus
v
Y
PS0 .K; M / D H 0 .Kv ; M /
v2S
with the product topology (it is compact if M is finite);

Y0
PS1 .K; M / D H 1 .Kv ; M /
v2S
with the restricted product topology (it is always locally compact because each
H 1 .Kv ; M / is finite by (2.1)). If M is finite, then
L
PSr .K; M / D v2S H r .Kv ; M /
(discrete topology) for r ¤ 0; 1.

L EMMA 4.8 For any finitely generated GS -module M , the image of
Y
H r .GS ; M / ! H r .Kv ; M /
v2S
is contained in PSr .K; M /.
P ROOF. Every 2 H r .GS ; M / arises from an element 0 of H r .Gal.L=K/; M /

for some finite Galois extension L of K contained in KS , and for all v that are
unramified in L, the image of in H r .Kv ; M / lies in Hun
r .K ; M /.
v 2
The lemma provides us with maps ˇ r W H r .GS ; M / ! PSr .K; M / for all r.
When necessary, we write ˇSr .K; M / for ˇ r .
L EMMA 4.9 Assume that M is finite. Then the inverse image of every compact
subset of PS1 .K; M / under the map ˇS1 .K; M / is finite (in other words, the map
is proper relative to the discrete topology on H 1 .GS ; M /).
P ROOF. After replacing K with a finite extension contained in KS , we can as-

sume that GS acts trivially on M . Let T be a subset of S omitting only finitely
many elements, and let
Y Y
P .T / D H 1 .Kv ; M / Hun1
.Kv ; M /.
v2S XT v2T
Then P .T / is compact by Tikhonov’s theorem, and every compact neighbour-

hood of 1 in PS1 .K; M / is contained in such a set. It suffices therefore to show
that the inverse image of P .T / is finite. An element of this set is a homomor-
phism f W GS ! M such that KSKer.f / is unramified at all primes v in T . There-
Ker.f /
fore KS is an extension of K of degree dividing the fixed integer ŒM and
unramified outside the finite set S X T . It is a well known consequence of Her-
mite’s theorem (see, for example, Serre 1964, pII-48) that there are only finitely
many such extension fields, and therefore there are only finitely many maps f: 2
Define
XrS .K; M / D Ker.ˇ r W H r .GS ; M / ! PSr .K; M //:
For a finite GS -module M , we write
M D D Hom.M; KS / D Hom.M; ES /:
It is again a finite GS -module, and if the order of M is a unit in RK;S , then

M D D Hom.M; K s / and M DD is canonically isomorphic to M .
The results (2.3), (2.6), and (2.13) combine to show that for all r 2 Z,
PS .K; M / is the algebraic and topological dual of PS2r .K; M D /. Therefore
r
there are continuous maps
r D Sr .K; M D /W PSr .K; M D / ! H 2r .GS ; M /
with r the dual of ˇ 2r :
T HEOREM 4.10 Let M be a finite GS -module whose order is a unit in RK;S .

4. GLOBAL FIELDS 57
(a) The groups X1S .K; M / and X2S .K; M D / are finite and there is a canoni-
cal nondegenerate pairing
X1S .K; M / X2S .K; M D / ! Q=Z:
(b) The map ˇS0 .K; M / is injective and S2 .K; M D / is surjective; for r D
0; 1; 2,
Im.ˇSr .K; M // D Ker.Sr .K; M D //:
(c) For r 3, ˇ r is a bijection
Y
H r .GS ; M / ! H r .Kv ; M /:
v real
Consequently, there is an exact sequence of locally compact groups and continu-

ous homomorphisms
ˇ0 0
0 ! H 0 .GS ; M / ! PS0 .K; M / ! H 2 .GS ; M D /
?
?
y
1 ˇ1
H 1 .GS ; M D / PS1 .K; M / H 1 .GS ; M /
?
?
y
ˇ2 2
H 2 .GS ; M / ! PS2 .K; M / ! H 0 .GS ; M D / ! 0.
The groups in this sequence have the following topological properties:
finite compact compact

compact locally compact discrete
discrete discrete finite
The finiteness of X1S .K; M / is contained in (4.9); that of X2S .K; M D / will
follow from the existence of the nondegenerate pairing in (a). The vertical arrows
in the above diagram will be defined below; alternatively they can be deduced
from the nondegenerate pairings in (a) because the cokernels of 0 and 1 are
X2S .K; M D / and X1S .K; M D / respectively.
E XAMPLE 4.11 (i) For any integer m > 1 and any set of primes S of density
greater than 1=2, X1S .K; Z=mZ/ D 0; consequently, X2S .K; m / D 0 under
the same condition provided m is a unit in RK;S .
(ii) If S omits only finitely many primes of K and m is a unit in RK;S , then
X1S .K; m / D 0 or Z=2Z; consequently, X2S .K; Z=mZ/ D 0 or Z=2Z under
the same conditions.
To see (i), note that
H 1 .GS ; Z=mZ/D Hom.GS ; Z=mZ/

H 1 .Kv ; Z=mZ/ D Hom.Gv ; Z=mZ/:
Therefore an element of X1S .K; Z=mZ/ corresponds to a cyclic extension of K

in which all primes of S split. The Chebotarev density theorem shows that such
an extension must be trivial when S has density greater that 1=2:
To see (ii), note that X1S .K; m / is the kernel of
L
K =K m ! v2S Kv =Kvm ;
that is, it is the set of elements of K that are local mth powers modulo those that
are global mth powers. This set is described in Artin and Tate 1961, X.1, where
the “special case” in which X1S .K; m / ¤ 0 is also determined.
Proof of the main theorem

The proof of the theorem will consist of identifying the exact sequence in the
statement of the theorem with the ExtGS .M D ; /-sequence of
0 ! ES ! JS ! CS ! 0;
except that, in the number field case, HomGS .M D ; JS / and HomGS .M D ; CS /

must be replaced by their quotients by
Q
NL=K Hom.M D ; v arch L w/
and NL=K Hom.M D ; DL;S / for any field L as in (4.6b).

In fact we shall consider more generally a finitely generated GS -module M .
In this case we write M d for the dual of M loosely regarded as a group scheme
over Spec.RK;S /. More precisely, when M is being regarded as a GS -module,
we let M d D Hom.M; ES /. For v … S, M is a gv -module, and we write
M d D Hom.M; O b un
v /; for v 2 S, M is a Gv -module, and we write M
d D
s
Hom.M; Kv /. It will always be clear from the context, which of these three we
mean.
L EMMA 4.12 Let M be a finitely generated GS -module such that the order of
Mtors is a unit in RK;S .
4. GLOBAL FIELDS 59
(a) The group ExtrGS .M; ES / D H r .GS ; M d /, all r 0:

b un
(b) For v … S , H r .gv ; M d / D Extrgv .M; O v /; for r 2, both groups are
0.
P ROOF. (a) As ES is divisible by all integers that are units in RK;S , this is a
special case of (0.8).
(b) As Ob un
v is divisible by all integers dividing the order of Mtors , this is
again a special case of (0.8). The gv -module O b un
v is cohomologically trivial
(Serre 1967a, 1.2), and so an easy generalization to profinite groups of (Serre
1962, IX 6, Thm 11) shows that there exists a short exact sequence of gv -modules
b un
0!O v ! I0 ! I1 ! 0
b un
with I 0 and I 1 injective. It is obvious from this that Extrgv .M; O v / D 0 for
r 2: 2
L EMMA 4.13 In addition to the hypotheses of (4.12), assume either that M is

finite or that S omits only finitely many primes. Then
Q
HomGS .M; JS / D v2S H 0 .Gv ; M d /
(D PS0 .K; M d / if K is a function field) and
ExtrGS .M; JS / D PSr .K; M d /; r 1:
P ROOF. We consider finite subsets T of S satisfying the same hypotheses as

S relative to M , namely, T contains all archimedean primes plus those nonar-
chimedean primes at which M is ramified, and the order of Mtors is a unit in
RK;T . Let
Q Q
JF ;S T D w2T Fw w2S XT O bv.
Then JS D lim J (limit over F and T with F KT and splitting M ),

!F ;T F ;S T
and so (0.10) shows that
ExtrGS .M; JS / D lim ExtrGal.F =K/ .M; JF ;S T /.

!F ;T
Since Exts commute with products in the second place (to see this, compute them
by taking a projective resolution of the term in the first place), on applying (0.11)
we find that

Q Q
ExtrGF=K .M; JF ;S T / D ExtrGF =K .M; Fw / ExtrGF =K .M; O b / :
w v w v Fw
v2T v2S XT
un
bv
As O is cohomologically trivial, (0.9) shows that for v 2 S X T;
ExtrGF b / D Extrg .M; O

.M; O b un
v /;
w =Kv Fw v
and we have already seen that
b un
Extrgv .M; O v D H r .gv ; M d /:
On combining these statements, we find that

r Q r Q r d
ExtGS .M; JS / D lim ExtGF =K .M; Fw / H .gv ; M / :
!F ;T v2T w v
v2S XT
For r 1, (0.9) shows that we can replace the group ExtrGF .M; Fw /
w =Kv
/, which equals H r .G ; M d / by (0.8). Hence
with ExtrGv .M; Kv;s v

Q Q
ExtrGS .M; JS / D lim r d
H .Gv ; M / r d
H .gv ; M / ;
!T v2T v2S XT
Q Q
which equals v2S H 0 .Gv ; M d / when r D 0, and equals 0v2S H 1 .Gv ; M / D
df
PS1 .K; M / when r D 1.

For r 2, H r .gv ; M d / D 0 by (4.12), and so

L
ExtrGS .M; JS / D lim ExtrGF =K .M; Fw / (limit over all F K s ; F K/
!F v2S w v
L
D lim ExtrGF =K .M; Fw / :
v2S !F w v
In the case that S contains almost all primes, lim Fw D Kvs and so
!F
lim ExtrGF =K .M; Fw / D ExtrGv .M; Kvs /.
!F w v
In the case that M is finite, we know that if ` divides the order of M , then S
contains all primes lying over `. Therefore lim H 2 .Gal.Kvs =Fw /; Kvs /.`/ D
!F
lim Br.Fw /.`/ D 0, and the spectral sequence (0.9)
!
ExtrGal.Fw =Kv / .M; H s .Gal.Kvs =Fw /; Kvs // H) ExtrGv .M; Kvs /
shows that again lim ExtrGF =K .M; Fw / D ExtrGv .M; Kvs /. From (0.8) we
!F w v
know that ExtrGv .M; Kvs / D H r .Gv ; M d / (D H r .Kv ; M d /), and so this com-
pletes the proof of the lemma. 2
4. GLOBAL FIELDS 61
R EMARK 4.14 Without the additional hypotheses, (4.13) is false. For example,
let K D Q, S D f1g, and let M D Z. Then GS D f1g, and so ExtrGS .Z; JS / D
0 for r > 0, but
PS2 .K; M d / D H 2 .Gal.C=R/; C / D 12 Z=Z:
Now assume that M is finite. On using (4.12), (4.13), and (4.6) to replace the
terms in the sequence
! ExtrGS .M D ; ES / ! ExtrGS .M D ; JS / ! ExtrGS .M D ; CS / ! ;
Y
0 ! H 0 .GS ; M / ! H 0 .Kv ; M / ! HomGS .M D ; CS /
v2S
! H 1 .GS ; M / ! PS1 .K; M / ! H 1 .GS ; M D /
L
! H 2 .GS ; M / ! PS2 .K; M / ! H 0 .GS ; M D / ! H 3 .GS ; M / ! H 3 .Gv ; M / ! 0
v real
and isomorphisms
' L
H r .GS ; M / ! H r .Kv ; M /; r 4:
v real
This is the required exact sequence except for the first three terms in the number
field case and the surjectivity of PS2 .K; M D / ! H 0 .GS ; M / . But this last map
is dual to H 0 .GS ; M / ! PS0 .K; M /, which is injective. (Note that if M ¤ 0 in
the number field case, then S must contain at least one nonarchimedean prime.)
For the first three terms of the sequence in the number field case, consider the
exact commutative diagram:
Q
Hom.M D ; v arch L v/ !
Hom.M D ; CS .L//
? ?
?N ?N
y L=K y L=K
Q
0 !
H 0 .GS ; M / ! 0
v2S H .Gv ; M / !
HomGS .M D ; CS / ! Ker.ˇ 1 / ! 0
? ?
? ?
y y
PS0 .K; M / !
H 2 .GS ; M /
? ?
? ?
y y
0 0 :
Q
The map Hom.M D ; v arch L v / ! Hom.M ; DS .L// is always an isomor-
D
phism on torsion, and therefore it is an isomorphism in our case. The snake

lemma now gives us an exact sequence
0 ! H 0 .GS ; M / ! PS0 .K; M / ! H 2 .GS ; M D / ! Ker.ˇ 1 / ! ,
which completes the proof of the theorem. (An alternative approach is to note
that the first half of the sequence can be obtained as the algebraic and topological
dual of the second half.)
Consequences
C OROLLARY 4.15 If S is finite and M is a finite GS -module whose order is a
unit in RK;S , then the groups H r .GS ; M / are finite for all r:
P ROOF. In this case the groups PSr .K; M / are finite, and so the finiteness of
H 0 .GS ; M / is obvious and that of H 1 .GS ; M / and H 2 .GS ; M / follows from
the finiteness of X1S .K; M / and X2S .K; M /: 2
C OROLLARY 4.16 Let M be a finite GS -module whose order is a unit in RK;S .

Then, for any finite subset T of S omitting at least one finite prime of S , the map
L 2
H 2 .GS ; M / ! H .Gv ; M /
v2T
is surjective. In particular, in the number field case the map

L 2
H 2 .K; M / ! H .Kv ; M /
v real
is surjective.
P ROOF. Let v0 be a finite prime of K not in T . In order to prove the corollary,

it suffices to show that for any element a D .av / of PS2 .K; M /, it is possible to
modify av0 so as to get an element in the image of ˇ 2 . Theorem 4.10 shows that,
in the duality between PS0 .K; M D / and PS2 .K; M /, the image of ˇ 2 is the or-
thogonal complement of the image of ˇ 0 . Let be the character of PS0 .K; M D /
defined by a, and let 0 be its restriction to H 0 .GS ; M D /. The map
H 0 .GS ; M / ! H 0 .Kv0 ; M /
is injective, and every character of H 0 .GS ; M D / extends to one of H 0 .Kv0 ; M D /.

Choose such an extension of 0 and let av0 0 be the element of H 2 .Kv0 ; M / corre-
sponding to it by duality. When the component av0 of a is replaced by av0 av0 0 ,
then a becomes orthogonal to Im ˇ 0 and is therefore in the image of ˇ 2 : 2
4. GLOBAL FIELDS 63
C OROLLARY 4.17 For any number field K ,
H 0 .GK ; Z/ D Z;
H 2 .GK ; Z/ D Hom.CK =DK ; Q=Z/;
H 2r .GK ; Z/ D .Z=2Z/t for 2r 4, where t is the number of real primes of K;
and
H r .GK ; Z/ D 0 for r odd.
df
P ROOF. The assertions for r 2 are obvious. According to (1.12), G D GK
contains an open subgroup U of index 2 having strict cohomological dimension
2. Therefore H r .G; ZŒG=U / D H r .U; Z/ D 0 for r 3. Let generate G=U .
The exact sequence
1C 1 7!1
0 ! Z ! ZŒG=U ! ZŒG=U ! Z ! 0
gives rise to isomorphisms H r .G; Z/ ! H rC2 .G; Z/ for r 3. For r 4,
H r .G; Z/ D H r1 .G; Q=Z/

D lim H r1 .G; n1 Z=Z/
!
4:10 Q
D lim v real H r1 .Kv ; n1 Z=Z/
Q!
D v real H r .Kv ; Z/:
(We applied (4.10) with S the set of all primes of K.) If r is odd, H r .R; Z/ D 0,
and if r is even, H r .R; Z/ ' Z=2Z, and so this completes the proof. 2
C OROLLARY 4.18 For any prime ` that is a unit in RS ,

L r
H r .GS ; ES /.`/ ! H .Gv ; Kv;s /.`/
v real
is an isomorphism, all r 3. In particular, when r 3, H r .GS ; ES /.`/ D 0 if

` or r is odd.
P ROOF. From the sequence

L
0 ! ES ! KS ! Z!0
v…S
(here S denotes the set of primes of KS lying over a prime of S/, we get an exact
sequence L
H 2 .GS ; ES / ! Br.K/ ! Br.Kv /
v…S
L
(cf. A.7). Therefore, the map H 2 .GS ; ES / ! Br.Kv / is surjective. The
v real
Kummer sequence
`n
0 ! `n ! ES ! ES ! 0
gives us the first row of the next diagram
H 2 .GS ; ES / ! H 3 .GK ; `n / ! H 3 .GK ; ES /`n ! 0

? ? ?
?surj. ? ?
y y y
L L 3 L 3
Br.Kv / ! H .Gv ; `n / ! H .Gv ; Kvs /`n ! 0;
v real v real v real
and the five-lemma proves our assertion for r D 3. One now proceeds by induc-
tion, using the continuation of the diagram. 2
R EMARK 4.19 Let F be a finite extension of K contained in KS , and let HS D

Gal.KS =F /. Then the following diagram is commutative:
H 0 .GS ; M / !
0 ! H 2 .GS ; M D / !
PS0 .K; M / ! H 1 .GS ; M / !

? ? ? ?
? ? ? ?
yRes yRes yCor yRes
0 !
H 0 .HS ; M / ! H 2 .HS ; M D / !
PS0 .F; M / ! H 1 .HS ; M / !

This follows from the commutativity of the following diagrams:
! ExtrGS .M; ES / ! ExtrGS .M; JS / ! ExtrGS .M; CS / !
? ? ?
? ? ?
y y y
! ExtrHS .M; ES / ! ExtrHS .M; JS / ! ExtrHS .M; CS / !
and
ExtrGS .M; CS / ! H 2r .GS ; M D /
? ?
? ?
y yCor
ExtrHS .M; CS / ! H 2r .HS ; M D / :
An explicit description of the pairing between X1 and X2

Finally we shall give an explicit description of the pairing
X1S .K; M / X2S .K; M D / ! Q=Z:

4. GLOBAL FIELDS 65
Represent a 2 X1S .K; M / and a0 2 X2S .K; M D / by cocycles ˛ 2 Z 1 .GS ; M /

and ˛ 0 2 Z 2 .GS ; M D /. Write ˛v and ˛v0 for the restrictions of ˛ and ˛ 0 to
Gv . Then for each v 2 S, we have a 0-cochain ˇv and 1-cochain ˇv0 such that
dˇv D ˛v and dˇv0 D ˛v0 . The cup-product ˛ Y ˛ 0 2 Z 3 .GS ; ES /, and as
H 3 .GS ; ES / has no nonzero elements of order dividing the ŒM , there is a 2-
cochain (for GS / with coefficients in ES such that ˛ Y ˛ 0 D d. Then d.ˇv Y
˛v0 / D dv D d.˛v Y ˇv0 / and d.ˇv Y ˇv0 / D ˛v Y ˇv0 ˇv Y ˛v0 , and so, for
each v, .˛v Y ˇv0 / v and .˛v0 Y ˇv / v are Pcocycles representing the same
class, say cv , in H 2 .Gv ; Kvs /. Set ha; a0 i D invv .cv /. It is easy to see that
this element is independent of the choices made, and one can show that it is equal
to the image of .a; a0 / under the pairing constructed in the proof of the theorem.
Generalization to finitely generated modules

We note that in the course of the proof of (4.10) we have shown the following
result.
T HEOREM 4.20 Assume that S omits only finitely many primes of K , and let M
be a finitely generated12 module over GS such that the order of Mtors is a unit in
RK;S :
(a) The group X2S .K; M d / is finite and is dual to X1S .K; M /.
(b) There is an exact sequence of continuous homomorphisms
Q0
H 1 .GS ; M / H 1 .Kv ; M d / H 1 .GS ; M d /
?
?
y
L
H 2 .GS ; M d / ! H 2 .Gv ; M d / ! H 0 .GS ; M / ! 0;
and for r 3 there are isomorphisms
' Q
H r .GS ; M d / ! v real H
r
.Kv ; M d /:
(c) In the function field case, the sequence in (b) can be extended by
Q0
0 ! H 0 .GS ; M d /^ ! H 0 .Gv ; M d /^ ! H 2 .GS ; M / ! : : :
where ^ denotes completion with respect to the topology of open subgroups

of finite index.
12 Meaning finitely generated as an abelian group.
P ROOF. To obtain (b) and (c), write down the Ext.M; /-sequence of
0 ! ES ! JS ! CS ! 0
and use (4.6), (4.12), and (4.13) to replace various of the terms. Part (a) is a
restatement of the fact that the sequence in (b) is exact at H 2 .GS ; M d /: 2
C OROLLARY 4.21 Let T be a torus over K . If S omits only finitely many

primes, then there are isomorphisms
' M
H r .GS ; T / ! H r .Kv ; T /
v real
for all r 3. In particular H r .GS ; Gm / D 0 for all odd r 3:
P ROOF. Take M D X .T / in the theorem. 2
I do not know to what extent Theorem 4.20 holds with M and M d inter-
changed, but R. Kottwitz has shown that for any torus T over a number field K,
and r D 1, 2, there is a canonical nondegenerate pairing of finite groups
Xr .K; T / X3r .K; X .T // ! Q=Z:
For r ¤ 1; 2, Xr .K; T / and Xr .K; X r

Q .T // rare zero. Here X .K; T / is de-
fined to be the kernel of H .K; T / ! all v H .Kv ; T /. See also (II 4) below.
r
N OTES Theorem 4.10 is due to Tate (see Tate 1962 for an announcement with a brief
indication of proof). Parts of the theorem were found independently by Poitou (1966,
1967). The above proof of (4.10) generalizes that in Tate 1966, which treats only the
case that S contains all primes of K. There is also a proof in Haberland 1978 similarly
generalizing Poitou 1967. Corollaries 4.16 and 4.17 are also due to Tate (cf. Borel and
Harder 1978, 1.6, and Serre 1977, 6.4).
Proofs of parts of the results in this section can also be found in Takahashi 1969,
Uchida 1969, Bashmakov 1972, and Langlands 1983, VII 2.
5 Global Euler-Poincaré characteristics

Let K be a global field, and let S be a finite nonempty set of primes including all
archimedean primes. As in 4, we write KS for the largest subfield of K s that is
T only at primes in S, GS for Gal.KS =K/, and RK;S for the ring
ramified over K
of S-integers v…S Ov . Let M be a finite GS -module whose order is a unit in
RS . We know from (4.15) that the groups H r .GS ; M / are finite for all r, and
we would like to define .GS ; M / to be the alternating product of their orders.
5. GLOBAL EULER-POINCARÉ CHARACTERISTICS 67
However, when K is a real number field, the cohomology groups will in general
be nonzero for an infinite number of values of r (see 4.10c), and so this is not
possible. Instead, we abuse notation, and set
ŒH 0 .GS ; M /ŒH 2 .GS ; M /

.GS ; M / D .
ŒH 1 .GS ; M /
T HEOREM 5.1 With the above definition,13

Y ŒH 0 .Gv ; M /
.GS ; M / D .
jŒM jv
v arch
R EMARK 5.2 (a) In the function field case, the theorem says simply that
.GS ; M / D 1:
In the number field case, (2.13c) shows that
ŒH 0 .Gv ; M / ŒH 1 .Gv ; M /
D ;
jŒM jv ŒH 0 .Gv ; M D /
and (2.13a) shows that ŒH 1 .Gv ; M / D ŒH 1 .Gv ; M D /, which equals

ŒHT0 .Gv ; M D / because the Herbrand quotient of a finite module is 1. There-
fore the formula can also be written as
Y ŒH 0 .Gv ; M D /
T
.GS ; M / D :
ŒH 0 .Gv ; M /
v arch
(b) Because S is finite, all groups in the complex in Theorem 4.10 are finite, and
so the exactness of the complex implies that
Q
.GS ; M / .GS ; M D / D .Kv ; M / (5.2.1)
v2S
13 Since the notation is confusing, I should give an example. Let M D Z=2Z and let R D
S
OK Œ 12 , so that S consists of the infinite primes and those dividing 2. The group H 0 .Gv ; M / D
M Gv D Z=2Z (it is not the Tate cohomology). On the other hand j jv is the normalized valuation
(the one that goes into the product formula). Thus jŒM jv D 2 if v is real, and jŒM jv D 4 if v is
complex. Thus, the formula says that
1
.GS ; M / D
2s
where s is the number of complex primes.
where .Kv ; M / D ŒH 0 .Kv ; M /ŒH 1 .Kv ; M /1 ŒH 2 .Kv ; M / (notations as

in 4). According to (2.8), .Kv ; M / D jŒM jv if v is nonarchimedean, and
obviously
.Kv ; M / D ŒH 0 .Kv ; M / D ŒHT0 .Gv ; M /
if v is archimedean. By assumption jŒM jv D 1 if v … S, and so the product
formula shows that
Y Y ŒH 0 .Kv ; M /
.Kv ; M / D .
jŒM jv
v2S v arch
Now (2.13c) allows us to rewrite this as
Y ŒH 0 .Gv ; M /ŒH 0 .Gv ; M D /

T T
:
ŒH 0 .Gv ; M /ŒH 0 .Gv ; M D /
v arch
Therefore (5.2.1) is also implied by (5.1), and conversely, in the case that M
M D , (5.2.1) implies the theorem.
(c) The theorem can sometimes be useful in computing the order of H 1 .GS ; M /.
It says that
Y
ŒH 1 .GS ; M / D ŒH 0 .GS ; M / ŒH 2 .GS ; M / jŒM jv =ŒH 0 .Gv ; M /;
v arch
and we know by (4.10) that H 2 .GS ; M / fits into an exact sequence

L
0 ! X2S .K; M / ! H 2 .GS ; M / ! H 2 .Kv ; M / ! H 0 .GS ; M D / ! 0:
v2S
By duality, ŒX2S .K; M / D ŒX1S .K; M D / and ŒH 2 .Kv ; M / D ŒH 0 .Kv ; M D /,

and so the theorem is equivalent to the statement
ŒH 1 .GS ; M / D
ŒH 0 .GS ; M / Y 0 Y jŒM jv
ŒX1S .K; M D / 0 D
ŒH .Kv ; M D / 0
:
ŒH .GS ; M / ŒH .Gv ; M /
v2S v arch
The method of the proof of Theorem 5.1 is similar to that of (2.8). Let '.M /
be the quotient of .GS ; M / by the right hand side of the equation. We have
to show that '.M / D 1. The argument in (5.2b) shows that (4.10) implies that
'.M /'.M D / D 1, and so in order to prove the theorem for a module M , it
suffices to show that '.M / D '.M D /
L EMMA 5.3 The map ' from the category of finite GS -modules to Q>0 is addi-
tive.
P ROOF. Let
0 ! M 0 ! M ! M 00 ! 0
be a short exact sequence, and consider the truncated cohomology sequence
0 ! H 0 .GS ; M 0 / ! ! H 4 .GS ; M 00 / ! H 5 .GS ; M 0 /0 ! 0;
where H 5 .GS ; M 0 /0 is the kernel of the map H 5 .GS ; M 0 / ! H 5 .GS ; M /. Ac-

cording to (4.10), for r 3, we can replace H r .GS ; / with
L
PSr .K; / D H r .Gv ; /:
v archimedean
Now ŒPS3 .K; M / D ŒPS4 .K; M / because the Herbrand quotient of a finite mod-
ule is 1, and so the sequence leads to the equality
.M 0 / .M 00 / D .M / ŒPS5 .K; M 0 /0 ;
where PS5 .K; M 0 /0 denotes the kernel of the map PS5 .K; M 0 / ! PS5 .K; M /. Be-
cause of the periodicity of the cohomology of a finite cyclic group, ŒPS5 .K; M 0 /0 D
ŒC , where
L 1 L 1
C D Ker. H .Gv ; M 0 / ! H .Gv ; M //:
v real v real
From the exact sequence

L 0 L 0 L 0
0! H .Gv ; M 0 / ! H .Gv ; M / ! H .Gv ; M 00 / ! C ! 0
v arch v arch v arch
we see that
Y ŒH 0 .Gv ; M 0 / ŒH 0 .Gv ; M 00 /
ŒC D :
ŒH 0 .Gv ; M /
v arch
As ŒM 0 ŒM 00 D ŒM , it is now clear that '.M 0 /'.M 00 / D '.M /: 2
The lemma shows that it suffices to prove the theorem for a module M killed
by some prime p, and the assumptions on M require that p be a unit in RS .
Choose a finite Galois extension L of K, L KS , that splits M and contains
a primitive p th root of 1 (primitive 4th root in the case that p D 2/. Let G be
Gal.L=K/. We need only consider modules M split by L. Note that ' defines
a homomorphism from the Grothendieck group RFp .G/ to Q>0 . An argument
as in the proof of Theorem 2.8 (using 2.10) allows us to replace K by a larger

field, and consequently assume that G is a cyclic group of order prime to p. Note
that L is totally imaginary, and so H r .Gal.KS =L/; M / D 0 for r 3 (by 4.10).
It follows that there is a well-defined homomorphism 0 W RFp .G/ ! RFp .G/
sending the class ŒM of M in RFp .G/ to
ŒH 0 .Gal.KS =L/; M / ŒH 1 .Gal.KS =L/; M / C ŒH 2 .Gal.KS =L/; M /:
As Hom.; Fp / is exact, it also defines a functor

W RFp .G/ ! RFp .G/:
L EMMA 5.4 For a finite Fp ŒG-module M , there are the following formulas:
(a) 0 .M D / D ŒM 0 .p /:
(b) ŒM ŒFp ŒG D dimFp .M / ŒFp ŒG:
P ROOF. (a) On tensoring a resolution of p by Hom.M; Fp /, we see that the

cup-product pairing arising from
.
; f / 7! .x 7!
f .x/ /W p Hom.M; Fp / ! M D
defines isomorphisms
H r .Gal.KS =L/; p / ˝ Hom.M; Fp / ! H r .Gal.KS =L/; M D /
for all r (recall that Gal.KS =L/ acts trivially on M and p /. This gives the
formula.
(b) Let M0 denote M regarded as a G-module with the trivial action. As we
observed in 2, ˝ m 7! ˝ m extends to an isomorphism Fp ŒG ˝ M0 !
Fp ŒG ˝ M , and this gives (b). 2
On applying both parts of the lemma, we see that
0 .M D / ŒFp ŒG D ŒM ŒFp ŒG 0 .p / D dim.M / ŒFp ŒG 0 .p /:
Similarly
0 .M / ŒFp ŒG D dim.M / ŒFp ŒG 0 .p /:
Let be the homomorphism RFp .G/ ! Q>0 sending the class of a module N to
the order of N G . Then ı 0 D , and so on applying to the above equalities,
we find that .M / D .M D /.
Let v be a real prime of K. If Lw ¤ Kv , then p must be odd, and so
ŒM v D ŒMGv , which equals Œ.M D /Gv . This shows that the factors of '.M /
G
and '.M D / corresponding to v are equal. It is now clear that '.M / D '.M D /,
and we have already noted that this is implies that '.M / D 1:
R EMARK 5.5 In the function field case there is a completely different approach
to the theorem. Let K D Kk s (composite inside KS ), and let H D Gal.KS =K/.
Let g.K/ be the genus of K, and let s be the number of primes of K lying over
primes in S. Then H is an extension of a group H 0 having 2g.K/ C s generators
and a single well-known relation (the tame fundamental group of the curve over
k s obtained by omitting the points of S/ by a pro-p group, p D char.K/. Using
this, or a little étale cohomology, it is possible to show that H r .H; M / is finite
for all finite H -modules M of order prime to p (cf. Milne 1980, V 2). Also,
it follows from (4.10) that H r .H; M / D 0 for r > 2. The Hochschild-Serre
spectral sequence for H G gives short exact sequences
0 ! H r1 .H; M /g ! H r .GS ; M / ! H r .H; M /g ! 0
in which g D G=H D Gal.k s =k/ D h i and the two end groups are defined by
the exactness of
1
0 ! N g ! N ! N ! Ng ! 0:
It follows from the first set of exact sequences that
ŒH 0 .H; M /g ŒH 1 .H; M /g ŒH 2 .H; M /g
.GS ; M / D
ŒH 0 .H; M /g ŒH 1 .H; M /g ŒH 2 .H; M /g
and from the second that this product is equal to 1.
An extension to infinite S
As we observed above, in the case that S is finite, all groups in the complex
in (4.10) are finite, and therefore the alternating product of their orders is one.
Oesterlé (1982/83) shows that, when S is infinite, it is possible to define natural
Haar measures on the groups in the complex, and prove that (in an appropriate
sense) the alternating product of the measures is again one. For example, the mea-
sure to take on PS1 .K; M / is the Haar measure for which the compact subgroup
Q 1 Iv
v H .gv ; M / (product over all nonarchimedean v/ has measure 1 (note that
H .gv ; M v / D Hun
1 I 1 .K ; M / if M is unramified at v/. The main result of
v
Oesterlé 1982/83 can be stated as follows.
T HEOREM 5.6 Let K be a global field, let S be a (possibly infinite) set of primes
of K , and let M be a finite GS -module. Assume that S contains all archimedean
primes and all primes for which ŒM is not a unit. Relative to the Haar measure
on PS1 .K; M / defined above, a fundamental domain for PS1 .K; M / modulo the
action of the discrete subgroup H 1 .GS ; M /=X1S .K; M / has finite measure
ŒX1S .K; M /ŒH 0 .GS ; M D / Y
ŒH 0 .Gv ; M /:
ŒX1S .K; M D /ŒH 0 .GS ; M / v archimedean
P ROOF. Suppose first that S is finite. Then the groups are all finite, and the
measure of the fundamental domain in question is
Q 1
ŒH .Kv ; M / ŒX1S .K; M /
Q 1 :
ŒH .gv ; M I / ŒH 1 .GS ; M /
From (5.2c) we know that this is equal to
ŒX1S .K; M /ŒH 0 .GS ; M D / Y ŒH 1 .Kv ; M / Y ŒH 0 .Gv ; M /

:
ŒX1S .K; M D /ŒH 0 .GS ; M / v2S ŒH 1 .gv ; M I / ŒH 0 .Kv ; M D / jŒM jv
v arch
As
ŒH 1 .gv ; M I / D ŒH 0 .gv ; M I / D ŒH 0 .Gv ; M /
for v nonarchimedean (we set it to zero for v archimedean) and ŒH 0 .Kv ; M D / D
ŒH 2 .Kv ; M /, we see that the middle term is
Y Y
.Kv ; M /1 ŒHT0 .Gv ; M /:
v2S v archimedean
In (5.2b) we showed that

Y Y
.Kv ; M /1 D jŒM jv =ŒHT0 .Gv ; M /:
v2S v arch
This verifies the theorem in this case. For an infinite set S, one chooses a suitably
large finite subset S 0 of S and shows that the theorem for S is equivalent to the
theorem for S 0 (see Oesterlé 1982/83, 7). 2
N OTES Theorem 5.1 is due to Tate (see Tate 1965/66, 2.2, for the statement together
with hints for a proof). Detailed proofs are given in Kazarnovskii 1972 and Haberland
1978, 3. The above proof differs from previous proofs in that it avoids any calculation
of the cohomology of n .
In his original approach to Theorem 4.10, Tate proved it first in the case that S is
finite by making use of a counting argument involving (presumably) the formula (5.2.1)
for .GS ; M /.GS ; M D / in order to show that X1S .K; M / and X2S .K; M D / have the
same order. He deduced it for an infinite S by passing to the limit. (See Tate 1962, p192.)
Theorem 5.6 is taken from Oesterlé 1982/83.
6 Abelian varieties over global fields

Throughout this section K will be a global field, and A will be an abelian variety
over K. The letter S will always denote a nonempty set of primes of K containing
6. ABELIAN VARIETIES OVER GLOBAL FIELDS 73
all archimedean primes and all primes at which A has bad reduction. We continue
to write KS for the largest subfield of K s containing K thatTis ramified only at
primes in S, GS for Gal.KS =K/, and RK;S for the subring v…S Ov of K. The
letter m is reserved for an integer that is a unit in RK;S ; thus jmjv D 1 for all
v … S. For example, it is always permitted to take S to be the set of all primes of
K, and in that case m can be any integer prime to char.K/. As usual, we fix an
embedding of K s into Kvs for each prime v of K:
For an abelian group M , M ^ denotes the m-adic completion lim M=mn M .
n
If X is an algebraic group over K, then we often write H r .GS ; X / for the group
H r .GS ; X.KS // (equal to H r .K; X / Ddf H r .GK ; X.K s // in the case that
S contains all primes of K/. When X is an algebraic group over Kv , we set
H r .Kv ; X / D H r .Gv ; X.Kvs // except when v is archimedean, in which case we
set it equal to HTr .Gv ; X.Kvs //. By H r .; X.m// we mean lim H r .; Xmn /
!n
and by H r .; Tm X / we mean lim H r .; Xmn /:
n
The weak Mordell-Weil theorem

The Mordell-Weil theorem says that A.K/ is finitely generated. The first step in
its proof is the weak Mordell-Weil theorem: for some integer n > 1, A.K/=nA.K/
is finite. We prove a stronger result in (6.2) below.
L EMMA 6.1 Let A and B be abelian varieties over K having good reduction
outside S , and let f W A ! B be an isogeny whose degree is a unit in RK;S .
Write Af for Ker.f /. Then all points in Af .K s / have their coordinates in KS ,
and there is an exact sequence
f
0 ! Af .KS / ! A.KS / ! B.KS / ! 0:
In particular, there is an exact sequence
m
0 ! Am .KS / ! A.KS / ! A.KS / ! 0:
P ROOF. Let P 2 B.K/; its inverse image f 1 .P / in A is a finite subscheme of
A. We shall show that this finite subscheme splits over KS , which implies that P
lies in the image of A.KS / ! B.KS /. When P is taken to be zero, f 1 .P / is
Af , and so this shows that Af is split over KS , i.e., that Af .KS / D Af .K s /.
By assumption, A and B extend to abelian schemes A and B over Spec.RK;S /.
The map f extends to a finite flat map f W A ! B which, because its degree is
prime to the residue characteristics of RK;S , is also étale. Our point P extends to
a section P of B over SpecRS , and f 1 .P/ is a finite étale subscheme of A over
Spec.RS /. Any such scheme splits over RS , which implies that f 1 .P / splits,
and proves the lemma. (For more details on such things, see Milne 1986b, 20.)2
The lemma yields exact sequences
f
! H r .GS ; Af / ! H r .GS ; A/ ! H r .GS ; B/ !
m
! H r .GS ; Am / ! H r .GS ; A/ ! H r .GS ; A/ !
P ROPOSITION 6.2 (Weak Mordell-Weil theorem) For any integer n prime to the
characteristic of K , A.K/=nA.K/ is a finite group.
P ROOF. Given n, we can choose a finite set S of primes of K satisfying the

conditions in the first paragraph and such that n is a unit in RK;S . Then (6.1)
provides us with an exact sequence
n
0 ! An .KS / ! A.KS / ! A.KS / ! 0:
The cohomology sequence of this gives an injection A.KS /.n/ ,! H 1 .GS ; An /,

and we have seen in (4.15) that this last group is finite. 2
To deduce the full Mordell-Weil theorem from (6.2), one uses heights (see
Lang 1983, V):
The Selmer and Tate-Shafarevich groups

The Tate-Shafarevich group A classifies the forms of A for which the Hasse prin-
ciple fails. The Selmer group gives a computable upper bound for the rank of
A.K/. The difference between the upper bound and the actual rank is measured
by the Tate-Shafarevich group.
L EMMA 6.3 Let a be an element of H 1 .K; A/. Then for all but finitely many
primes v of K , the image of a in H 1 .Kv ; A/ is zero.
P ROOF. As H 1 .K; A/ is torsion, na D 0 for some n, and as H 1 .K; An / !

H 1 .K; A/n is surjective, there is a b 2 H 1 .K; An / mapping to a. For almost all
v, An .K s / is an unramified Gv -module and b maps into Hun 1 .K ; A / (see 4.8).
v n
Therefore a maps into H 1 .gv ; A.Kvun // for almost all v, but (3.8) shows that this
last group is zero unless v is one of the finitely many primes at which A has bad
reduction. 2
The Tate-Shafarevich group XS .K; A/ is defined to be the kernel of

L
H 1 .GS ; A/ ! v2S H
1
.Kv ; A/:
The Selmer groups SS .K; A/m and SS .K; A; m/ are defined by the exact se-
quences
L
0 ! SS .K; A/m ! H 1 .GS ; Am / ! v2S H 1 .Kv ; A/
L
0 ! SS .K; A; m/ ! H 1 .GS ; A.m// ! v2S H 1 .Kv ; A/:
The second sequence can be obtained by replacing m with mn in the first sequence
and passing to the direct limit. Therefore
SS .K; A; m/ D lim SS .K; A/mn :

!n
When S contains all primes of K, we drop it from the notation. Thus,
Y
X.K; A/ D Ker.H 1 .K; A/ ! H 1 .Kv ; A//
all v
Y
1
S.K; A; m/ D Ker.H .K; A.m// ! H 1 .Kv ; A//:
all v
P ROPOSITION 6.4 There is an exact sequence
0 ! A.K/.m/ ! SS .K; A/m ! XS .K; A/m ! 0:
P ROOF. Apply the snake lemma to the diagram

0 0
? ?
? ?
y y
SS .K; A/m ! XS .K; A/m
? ?
? ?
y y
0 ! A.K/.m/ ! H 1 .GS ; Am / ! H 1 .GS ; A/m ! 0
? ? ?
? ? ?
y y y
L D L
0 ! v H .Kv ; A/ !
1
v H .Kv ; A/ ! 0
1
P ROPOSITION 6.5 There are exact sequences

L
0 ! H 1 .GS ; A.m// ! H 1 .K; A.m// ! v…S H
1
.Kv ; A/
1 1 L 1
0 ! H .GS ; A/.m/ ! H .K; A/.m/ ! v…S H .Kv ; A/
P ROOF. For v … S, there is a commutative diagram

H 1 .GS ; A.m// ! H 1 .gv ; A.Kvun //
? ?
? ?
y y
H 1 .K; A.m// ! H 1 .Gv ; A.Kvs //:
According to (3.8), H 1 .gv ; A.Kvun // D 0, and so the diagram shows that the
image of H 1 .GS ; A.m// in H 1 .K; A.m// is contained in the kernel of
L
H 1 .K; A.m// ! v…S H 1 .Kv ; A/:
Conversely, let a lie in this kernel. We may assume that a is the image of an
element b of H 1 .K; Am / (after possibly replacing m by a power). To prove that
the first sequence is exact, it suffices to show that b (hence a) is split by a finite
extension of K unramified outside S. After replacing K by such an extension,
we can assume that Am .K/ D Am .K s / (because of 6.1). Then b corresponds
to a homomorphism f W Gal.K s =K/ ! Am .K/, and it remains to show that the
subfield Kf of K s fixed by the kernel of f is unramified outside S. This can be
checked locally. If v … S, then, by assumption, the image of b in H 1 .Kv ; Am /
maps to zero in H 1 .Kv ; A/. It therefore arises from an element cv of A.Kv /.
The closure of Kf in Kvs is Kv .m1 cv /, which is unramified by (6.1).
The exactness of the second exact sequence can be derived from the first. In
the diagram
lim A.K/.m/ lim A.K/.m/
! ? ! ?
? ?
y y
L
0 ! H 1 .GS ; A.m// ! H 1 .K; A.m// ! v…S H 1 .Kv ; A/
? ? ?
? ? ?
y y y
L
0 ! H 1 .GS ; A/.m/ ! H 1 .K; A/.m/ ! v…S H 1 .Kv ; A/
? ?
? ?
y y
0 0
the exactness of the bottom row follows from the exactness of the rest of the
diagram (use the snake lemma, for example). 2
C OROLLARY 6.6 For all S and m (as in the first paragraph),

XS .K; A/.m/ D X.K; A/.m/
SS .K; A; m/ D S.K; A; m/:
P ROOF. The kernel-cokernel sequence (see 0.24) of the pair of maps
ˇ L pr L
H 1 .K; A/.m/ ! all v H
1
.Kv ; A/.m/ ! v…S H
1
.Kv ; A/.m/
is
L
0 ! X.K; A/.m/ ! H 1 .GS ; A/.m/ ! v2S H
1
.Kv ; A/.m/ ! ,
because (6.5) allows us to replace Ker.pr ı ˇ/ with H 1 .GS ; A/.m/. This se-
quence identifies X.K; A/.m/ with XS .K; A/.m/. The second equality is proved
by replacing H 1 .K; A/.m/ in the proof with H 1 .K; A.m//: 2
R EMARK 6.7 Recall (4.15) that H 1 .GS ; Am .KS // is finite when S is finite.
Therefore its subgroup SS .K; A/m is finite when S is finite, and (6.4) then shows
that XS .K; A/m is finite. It follows now from (6.6) that X.K; A/m is finite,
and (6.4) in turn shows that S.K; A/m is finite. Consequently, S.K; A/.m/ and
X.K; A/.m/ are extensions of finite groups by divisible groups isomorphic to
direct sums of copies of Q` =Z` , ` dividing m. It is widely conjectured that
X.K; A/ is in fact finite.
Definition of the pairings

The main results in this section will concern the continuous homomorphisms
Q
ˇ 0 WA.K/^ ! 0
v2S H.Kv ; A/^ (compact groups)
L
ˇ r WH r .GS ; A/.m/ ! v2S H r .Kv ; A/.m/, r ¤ 0; (discrete groups).
Write XrS .K; A; m/ D Ker.ˇ r /. Thus X1S .K; A; m/ D XS .K; A/.m/, which
we have shown to be independent of S. We also write
XrS .K; A.m// D lim XrS .K; Amn /

!n Y
D Ker.H r .GS ; A.m// ! H r .Kv ; A.m///
v2S
XrS .K; Tm A/ D lim n XrS .K; Amn /
Y
D lim n Ker.H r .GS ; Amn / ! H r .Kv ; Amn //:

v2S
L EMMA 6.8 For any r 2, there is a canonical isomorphism
XrS .K; A; m/ ! XrS .K; A.m//:

P ROOF. For each r 2, there is an exact commutative diagram:

0 !
H r1 .GS ; A/ ˝ Qm =Zm ! H r .GS ; A.m// ! H r .GS ; A/.m/ ! 0
? ? ?
? r1 ? r ? r
yˇ .A/˝1 yˇ .A.m// yˇ .A/.m/
L r1 L r L r
0 !
H .Kv ; A/ ˝ Qm =Zm !
H .Kv ; A.m// ! H .Kv ; A/.m/ ! 0:
v2S v2S v2S
As r 1 1, the groups H r1 .GS ; A/ and H r1 .Kv ; A/ are both torsion, and so
their tensor products with Qm =Zm are both zero. The diagram therefore becomes

H r .GS ; A.m// ! H r .GS ; A/.m/
? ?
? r ? r
y ˇ .A.m// yˇ .A/.m/
L r L r
H .Kv ; A.m// ! H .Kv ; A/.m/;
v2S v2S
from which the result is obvious. 2
P ROPOSITION 6.9 For r D 0, 1, 2, there are canonical pairings
h ; iW XrS .K; A; m/ X2r t

S .K; A ; m/ ! Q=Z:
P ROOF. There is a unique pairing making the diagram
X0S .K; A; m/ X2S .K; At ; m/ ! Q=Z

# " k
X1S .K; Tm A/ X2S .K; At .m// ! Q=Z
commute. Here the bottom pairing is induced by the em -pairing and the pairings
in 4, the first vertical arrow is induced by the map
H r .GS ; A/ ! lim H rC1 .GS ; Am /

, and the second vertical map is the isomorphism in (6.8). This defines the pairing
in the case r D 0, and the case r D 2 can be treated similarly.
The definition of the pairing in the case r D 1 is more difficult. We will in
fact define a pairing
h ; iW XS .K; A/m XS .K; At /m ! Q=Z:
Since the Tate-Shafarevich groups are independent of S, we take S to be the set

of all primes of K. If
is a global cohomology class, cocycle, or cochain, we
write
v for the corresponding local object.
Let a 2 X.K; A/m and a0 2 X.K; At /m . Choose elements b and b 0 of

H 1 .GK ; Am / and H 1 .GK ; Atm / mapping to a and a0 respectively. For each v, a
maps to zero in H 1 .Kv ; A/, and so it is obvious from the diagram
A.Kv / ! H 1 .Kv ; Am / ! H 1 .Kv ; A/
x
?
?
A.Kv / ! H 1 .Kv ; Am2 /
that we can lift bv to an element bv;1 2 H 1 .Gv ; Am2 / that is in the image of
A.Kv /:
Suppose first that a is divisible by m in H 1 .GK ; A/, say a D ma1 , and
choose an element b1 2 H 1 .GK ; Am2 / mapping to a1 . Then bv;1 b1;v maps to
zero under H 1 .Kv ; Am2 / ! H 1 .Kv ; Am /, and so it is the image of an element
cv in H 1 .Kv ; Am /. We define
P
ha; a0 i D invv .cv Y bv0 / 2 Q=Z
where the cup-product is induced by the em -pairing Am Atm ! Gm , and invv
'
is the canonical map H 2 .Kv ; Gm / ! Q=Z.
In the general case, let ˇ be a cocycle representing b, and lift it to a cochain
ˇ1 2 C 1 .GK ; Am2 /. Choose a cocycle ˇv;1 2 Z 1 .Gv ; Am2 / representing bv;1 ,
and a cocycle ˇ 0 2 Z 1 .GK ; Atm / representing b 0 . The coboundary dˇ1 of ˇ1
takes values in Am , and dˇ1 Y ˇ 0 represents an element of H 3 .GK ; K s /. But
this last group is zero (by 4.18 or 4.21), and so dˇ1 Yˇ 0 D d for some 2-cochain
. Now14 .ˇv;1 ˇ1;v / Y ˇv0 v is a 2-cocycle, and we can define
P
ha; a0 i D invv ..ˇ1;v ˇ1;v / Y ˇv0 "v / 2 Q=Z:
It is not too difficult to check that the pairing is independent of the choices made.2
R EMARK 6.10 (a) If B is a second abelian variety over K having good reduction
outside S and f W A ! B is an isogeny, then
hf .a/; bi D ha; f t .b/i; a 2 XrS .K; A; m/; b 2 X2r t
S .K; B ; m/:
This follows from the fact that the local pairings are functorial.
(b) Let D be a divisor on A rational over K, and let 'D W A ! At be the
corresponding homomorphism sending a 2 A.K s / to the class of Da D, where
Da is the translate D C a of D. Then hc; 'D .c/i D 0 for all c 2 X1S .K; A; m/.
See Tate 1962, Thm 3.3. This can be proved by identifying the pairing defined
in (6.9) with that defined in (6.11) below, which we check has this property. See
also (II 5).
14 Poonen suggests ˇv;1 and ˇ1;v should be interchanged.
R EMARK 6.11 There is a more geometric description of a pairing on the Tate-

Shafarevich groups, which in the case of elliptic curves reduces to the original
definition of Cassels 1962, 3. An element a of X.K; A/ can be represented
by a locally trivial principal homogeneous space X over K. Let K s .X / be the
function field of X ˝K K s . Then the exact sequence
0 ! K s ! K s .X / ! Q ! 0
leads to a commutative diagram
Br.K/ ! H 2 .GK ; K s .X / / ! H 2 .GK ; Q/ ! 0

? ? ?
? ? ?
y y y
L L 2 L
0 ! Br.Kv / ! H .Gv ; Kvs .X / / ! H 2 .Gv ; Q/:
all v all v all v
The zero at top right comes from the fact that H 3 .GK ; K s / D 0 (see 4.21). The
zero at lower left is a consequence of the local triviality of X . Indeed, consider
an arbitrary smooth variety Y over a field k. The map Br.Y / ! Br.k.Y // is
injective (Milne 1980, II 2.6). The structure map Y ! Spec.k/ induces a map
Br.k/ ! Br.Y /, and any element of Y.k/ defines a section to this map, which is
then injective. In our situation, we have a diagram
0 ! Br.Kv / ! Br.XKv / ! Br.XKvs /

? ?
?inj ?inj
y y
Br.Kv / ! Br.Kv .X // ! Br.Kvs .X //
from which the claimed injectivity is obvious.

The exact sequence
0 ! Q ! Div0 .X ˝ K s / ! Pic0 .X ˝ K s / ! 0
yields a cohomology sequence
H 1 .GK ; Div0 .X ˝ K s // ! H 1 .GK ; Pic0 .X ˝ K s // ! H 2 .GK ; Q/ ! :

A trivialization A˝K s ! X ˝K s determines an isomorphism Pic0 .X ˝K s / !
Pic0 .A ˝ K s /. Because the trivialization is uniquely determined up to translation
by an element of A.K s / and translations by elements in A.K s / act trivially on
Pic0 .A ˝ K s / (Milne 1986b, 9.2), the isomorphism is independent of the choice
of the trivialization. A similar argument shows that it is a GK -isomorphism.
Therefore the sequence gives a map H 1 .GK ; At / ! H 2 .GK ; Q/. Let a0 2

X.K; At /, and let b 0 be its image in H 2 .GK ; Q/. Then b 0 lifts to an element
of H 1 .GK ; K s .X / /, and the image of this in
P˚H .Kv ; Kv .X / // lifts to an
1 s
element .cv / 2 ˚ Br.Kv /. Define ha; a i D0 invv .cv / 2 Q=Z. Note that the
cokernel of Br.K/ ! ˚ Br.Kv / is Q=Z, and so ha; a0 i can also be described
as the image of b 0 under the map defined be the snake lemma. As the principal
homogeneous space X is uniquely determined up to isomorphism by a, this shows
that ha; a0 i is well-defined.
It is easy to prove that if X.K; A/ is mapped to X.K; At / by means of a
polarization defined by a K-rational divisor15 , then the pairing on X.K; A/ is
alternating. Let P 2 X.K s /; then P D P C ˛. / where .˛. // is a cocycle
representing a. The map 'D sends Q 2 A.K s / to the class of DQ D in
Pic0 .A/, and so a0 is represented by the cocycle .˛ 0 . // 2 Z 1 .GK ; At /, where
˛ 0 . / is represented by the divisor E D D˛./ D. Now use the trivialization

Q 7! P CQW A˝K s ! X ˝K s to identify Pic0 .A/ with Pic0 .X /. Then one sees
immediately that .˛0 /, regarded as a crossed homomorphism into Pic0 .XK s /,
lifts to a crossed homomorphism into Div0 .XK s /. Therefore the image of a0 in
H 2 .GK ; Q/ is zero, and so ha; a0 i D 0:
We leave it to the reader to check that this pairing agrees with that defined in
(6.9).
R EMARK 6.12 When A is the Jacobian of a curve X over K, there is yet another
description of a pairing on the Tate-Shafarevich groups. Write S for the canonical
map Div0 .X ˝ K s / ! A.K s /:
Let a 2 X.K; A/ be represented by ˛ 2 Z 1 .GK ; A.K s //, and let ˛v D dˇv
with ˇv 2 Z 0 .Gv ; A.Kvs //. Write
˛ D S.a/; a 2 C 1 .GK ; Div 0 .XK s //

ˇv D S.bv /; bv 2 C 0 .Gv ; Div 0 .XKvs //:
Then av D d bv C .fv / in C 1 .Gv ; Div 0 .XKvs //, where fv 2 C 1 .Gv ; Kvs .X / /.

Moreover d a D .f /, f 2 Z 2 .GK ; K s .X / /. Let a0 be a second element of
X.K; A/ and define a0 , b0v , fv0 , and f 0 as for a. Define
D f 0 Y a f Y a0 2 C 3 .GK ; K s /:
15 This condition was omitted in the original.

Then d D 0, and so, as H 3 .GK ; K s / D 0, D d " for some " 2 C 2 .GK ; K s /.

Set16
v D fv0 Y av b0v Y Resv .f / "v 2 C 2 .GK ; Kvs /
where Y denotes the cup-product pairing induced by .h; c/ 7! h.c/. Then v is a
2-cocycle representing a cv 2 Br.Kv /, and we let
P
ha; a0 i D invv .cv / 2 Q=Z:
One shows without serious difficulty that the choices in the construction can be
made so that ha; a0 i is defined and that it is independent of the choices.17
The main theorem

We shall need to consider the duals of the maps ˇ r . Recall (3.4, 3.6, and 3.7) that
H r .Kv ; A/ is dual to H 1r .Kv ; At /, except possibly for the p-components in
characteristic p. Therefore there exist maps,
L
1 W v2S H 1 .Kv ; A/.m/ ! At .K/ .m/ (discrete groups)
Q
0 W v2S H 0 .Kv ; A/^ ! H 1 .GS ; At /.m/ (compact groups)
such that r .A/ D ˇ 1r .At / .
T HEOREM 6.13 (a) The left and right kernels of the canonical pairing
X1 .K; A/.m/ X1 .K; At /.m/ ! Q=Z
are the divisible subgroups of X1 .K; A/.m// and X1 .K; At /.m/.
(b) The following statements are equivalent:
(i) X1 .K; A/.m/ is finite;
(ii) Im.ˇ 0 / D Ker. 0 / and the pairing between X0S .K; A; m/ and
X2S .K; At ; m/ is nondegenerate.
(c) The map ˇ 2 is surjective with kernel the divisible subgroup of
H 2 .GS ; A/.m/, and for r > 2, ˇ r is an isomorphism
L
H r .GS ; A/.m/ ! H r .Kv ; A/.m/:
v real
16 In the original, the " was omitted (cf. the proof of 6.9). As Bjorn Poonen pointed out to me,
without it, v need not be a cocycle. For more on the pairing, see:
Gonzalez-Aviles, Cristian D. Brauer groups and Tate-Shafarevich groups. J. Math. Sci. Univ.
Tokyo 10 (2003), no. 2, 391–419.
Poonen, Bjorn; Stoll, Michael. The Cassels-Tate pairing on polarized abelian varieties. Ann. of
Math. (2) 150 (1999), no. 3, 1109–1149.
The first reference includes a proof that it coincides with the pairing in (6.9).
17 The original claimed without proof that the pairing is always alternating, but this is not true
— see the paper of Poonen and Stoll mentioned in an earlier footnote.

R EMARK 6.14 (a) Much of the above theorem is summarized by the following
statement: if X.K; A/.m/ and X.K; At /.m/ are finite, then there is an exact
sequence with continuous maps
L 2
0 ! H .Kv ; At /.m/ ! H 2 .GS ; At /.m/ ! H 0 .GS ; A/^
vreal
?
? 0
yˇ
0 Q
H 1 .GS ; A/.m/ H 1 .GS ; At /.m/ H 0 .Kv ; A/^
v2S
?
?
ˇ 1y
L 1
H 1 .Kv ; A/.m/ ! H 0 .GS ; At /^ ! H 2 .GS ; A/.m/
v2S
?
? 2
yˇ
L 2
0 H .Kv ; A/.m/
v real
The unnamed arrows exist because of the nondegeneracy of the pairings defined
in (6.9).
(b) We shall see in (6.23) and (6.24) below that if S contains almost all primes
of K, then ˇ 0 and ˇ 2 are both injective. In this case, the above sequence can be
shortened to a four-term sequence:
M
0 ! X.K; A/.m/ ! H 1 .GS ; A/.m/ ! H 1 .Kv ; A/.m/ ! H 0 .GS ; At /^ ! 0:
v2S
In particular, when S contains all primes of K and the Tate-Shafarevich groups

are finite, then the dual of the exact sequence
M
0 ! X.K; A/ ! H 1 .K; A/ ! H 1 .Kv ; A/ ! B ! 0:
v2S
is an exact sequence
Y
0 X.K; At / H 1 .K; A/ H 0 .Kv ; At / At .K/^ 0
all v
except possibly for the p-components in characteristic p ¤ 0. Here B is defined

to be the cokernel of the preceding map. In the second sequence, H 0 .Kv ; At / D
At .Kv / unless v is archimedean, in which case it equals the quotient of A.Kv /
t ^
Q A0 .K/ ist the profinite completion
by its identity component (see 3.7). The term
t
of A .K/, which is equal to its closure in H .Kv ; A / (see 6.23b).
(c) If X.K; A/ is finite, then so also is X.K; At /. To see this note that there
is an integer m and maps f W A ! At and gW At ! A such that fg D m D gf .
Therefore there are maps
X.f /W X.K; A/ ! X.K; At /

X.g/W X.K; At / ! X.K; A/
whose composites are both multiplication by m. It follows that the kernel of

X.g/ is contained in X.K; At /m . When m is prime to the characteristic, we
observed in (6.7) that X.K; At /m is finite, and an elementary proof of the same
statement for m a power of char.K/ can be found in Milne 1970b (see also Chap-
ter III). Hence the kernel of X.g/ is finite, and this shows that X.At / is finite.
We begin the proof of (6.13) with part (c). As we saw in the proof of (6.8),
when r 2, there is a commutative diagram

H r .GS ; A.m// ! H r .GS ; A/.m/
? ?
? r ? r
yˇ .A.m// yˇ .A/.m/
L r L
H .Kv ; A.m// ! H r .Kv ; A/.m/;
v2S v2S
As H r .Kv ; A/ is zero when r 2 and v is nonarchimedean (see 3.2), the sum at

lower right needs to be taken only over the real primes. When r > 2, ˇ r .A.m//
is an isomorphism (see 4.10c), and so ˇ r .A/ is an isomorphism. When r D 2,
(4.10) shows that the cokernel of ˇ 2 .A.m// is

lim Atmn .K/ D lim Atmn .K/ D Tm At .K/ ;
!
which is zero because At .K/ is finitely generated (by the Mordell-Weil theorem).
Consider the diagram
m
H 2 .GS ; Am / !
H 2 .GS ; A/.m/
! H 2 .GS ; A/.m/ !
H 3 .GS ; Am /
? ? ? ?
? 2 ? 2 ? 2 ? 3
yˇ .Am / yˇ .A/ yˇ .A/ yˇ .Am /
L 2 L 2 L 2 L 3
H .Kv ; Am / !
H .Kv ; A/.m/ !
H .Kv ; A/.m/ !
H .Kv ; Am /:
v real v real v real v real
The first vertical arrow is surjective by (4.16). We have just shown that ˇ 2 .A/ is
surjective, and we know that ˇ 3 .Am / is an isomorphism by (4.10c). Therefore
we have a surjective map of complexes, and so the sequence of kernels is exact,
from which it follows that Ker.ˇ 2 .A// is divisible by m. On repeating this argu-
ment with m replaced by mn we find that Ker.ˇ 2 .A// is divisible by all powers
of m. Since it obviously contains H 2 .GS ; A/.m/div , this shows that it equals
H 2 .GS ; A/.m/div . This completes the proof of part (c).
We next prove part (b). Let v 2 S, and consider the diagram
0 ! A.K/.m/ ! H 1 .GS ; Am / ! H 1 .GS ; A/m ! 0

? ? ?
? ? ?
y y y
0 ! H 0 .Kv ; A/.m/ ! H 1 .Kv ; Am / ! H 1 .Kv ; A/m ! 0:
On replacing m with mn and passing to the inverse limit, and then replacing
the bottom row by the restricted product over all v in S, we obtain an exact
commutative diagram
0 !
A.K/^ !
H 1 .GS ; Tm A/ !
Tm H 1 .GS ; A/ !
0
? ? ?
? 0 ? ? 1
yˇ y yˇ
Q Q0 Q0
0 !
H 0 .Kv ; A/^ !
v2S H 1 .Kv ; Tm A/ !
v2S Tm H 1 .Kv ; A/ !
0:
v2S
The snake lemma now gives an exact sequence
0 ! X0S .K; A/ ! X1S .K; Tm A/ !Tm X.K; A/ !

Q 0
H .Kv ; A/^
0
! .H 1 .GS ; At .m// :
Im.ˇ /
Here we have used (4.10) to identify the cokernel of the middle vertical map with
a subgroup of
lim H 1 .GS ; Atmn / D .lim H 1 .GS ; Atmn // D H 1 .GS ; At .m// :

!
Consider the maps
Q 0 0
H 0 .Kv ; A/^ ! .H 1 .GS ; At /.m/ / ! H 1 .GS ; At .m//
v2S
the second of which is the dual of H 1 .GS ; At .m// H 1 .GS ; At /.m/ and is
therefore injective; consequently, Ker. 0 ı 0 / D Ker. 0 /. The composite 0 ı 0
is the composite of the quotient map
Q 0
Q 0 ^ H .Kv ; A/^
v2S H .K v ; A/ !
Im.ˇ 0 /
with . Since Im.ˇ 0 / goes to zero under 0 ı 0 , we see that it must also be
mapped to zero by 0 , that is, 0 ı ˇ 0 D 0 (without any assumptions). We also
see that Ker. 0 / D Im.ˇ 0 / if and only if is injective, which is equivalent to
X1S .K; Tm A/ ! Tm X.K; A/ being surjective.
Consider on the other hand the first part
0 ! X0S .K; A/ ! X1S .K; Tm A/ ! Tm X.K; A/
of the above exact sequence and the isomorphism

X2S .K; A/.m/ X2S .K; A.m//
in (6.8). Clearly the duality between X1S .K; Tm A/ and X2S .K; A.m// arising
from (4.10) induces a duality between X0 .K; A/ and X2 .K; A/.m/ if and only
if the map X1S .K; Tm A/ ! Tm X.K; A/ is zero.
On combining the conclusions of the last two paragraphs, we find that the
following two statements are equivalent:
(*) X0S .K; A/ and X2S .K; At / are dual and Im.ˇ 0 / D Ker. 0 /
(**) X1S .K; Tm A/ ! Tm X.K; A/ is both surjective and zero.
Clearly (**) is equivalent to Tm X.K; A/ being zero, but Tm X.K; A/ D 0 if and
only if the m-divisible subgroup of X.K; A/.m/ is zero, in which case the group
is finite. This proves the equivalence of statements (i) and (ii) in (b).
In preparing for the proof of (a), we shall need a series of lemmas. Since
the statement of (a) does not involve S, we can choose it to be any set we wish
provided it satisfies the conditions in the first paragraph of this section. We always
take it to be finite.
L
L EMMA 6.15 Let a 2 v2S H 1 .Kv ; Am /, and consider the pairing
P Q L
h ; iv W H 1 .Kv ; Am / H 1 .Kv ; Atm / ! Q=Z; hav ; av0 iv D invv .av Yav0 /:
L
Then ha; a0 i D 0 for all a0 in the image of SS .K; At /m ! v2S H 1 .Kv ; Atm /
Qand0 only if a can
if Q be written a D a1 C a2 with a1 and Q a2 in the images of
H .Kv ; A/ ! H 1 .Kv ; Am / and H 1 .GS ; Am / ! H 1 .Kv ; Am / respec-
tively.
P ROOF. The dual of the diagram

L
H 1 .Kv ; hRAtm /
RRR 1
RRRˇ
RRR
RRR
L 1
t
H .Kv ; A / o H 1 .GS ; Atm / o SS .K; At /m o 0
is
Q
H 1 .K
O v; A
R m/
RRR 1
RRR
RRR
RRR
Q 0 )
H .Kv ; A/ / H 1 .GS ; Atm / / .S .K; At /m / / 0:
S
Q 1
Let a 2 H .Kv ; Am /. If a maps to zero in .SS .K; At /m / , then 1 .a/
is the image of an element b in H 0 .Kv ; A/. Let a1 denote the image of b
in H 1 .Kv ; Am /; then a a1 is in the kernel of 1 . But according to (4.10),
the kernel of 1 is the image of H 1 .GS ; Am /, and so a a1 D a2 for some
a2 2 H 1 .GS ; Am /: 2
L EMMA 6.16 Let X0 .A/ be the subgroup of X.K; A/ of elements that become
divisible by m in H 1 .GS ; A/. Then there is an exact sequence
0 ! X0 .K; A/ ! X.K; A/ ! X2S .K; Am /:
P ROOF. Consider
m
X.K; A/ ! X.K; A/ ! X2S .K; Am /
? ? ?
? ? ?
y y y
m
H 1 .GS ; A/ ! H 1 .GS ; A/ ! H 2 .GS ; Am /
? ? ?
? ? ?
y y y
L 1 m L 1 L 1
H .Kv ; A/ ! H .Kv ; A/ ! H .Kv ; Am /.
v2S v2S v2S
An element a in X.K; A/ maps to zero in X2S .K; A/ if and only if it maps to

zero in H 2 .GS ; Am /, and this occurs if and only if its image in H 1 .GS ; A/ is
divisible by m: 2
L EMMA 6.17 Let a 2 X0 .K; A/. Then a 2 mX.K; A/ if and only if ha; a0 i D
0 for all a0 2 X.K; At /m :
P ROOF. If a D ma0 with a0 2 X.K; A/, then
ha; a0 i D hma0 ; a0 i D ha0 ; ma0 i D 0
for all a0 2 X.K; At /m . Conversely, assume that a satisfies the second condition,
and let a1 2 H 1 .GS ; A/ be such that ma1 D a; we have to show that a1 can be
modified to lie in X.K; A/. Choose a finite set S satisfying the conditions at the
start of this section and containing all v for which a1;v ¤ 0. If a1 is replaced
by its sum with an element of H r .GS ; A/, then it is still zero outside S (see the
proof of 6.5). Define b1 , bv;1 , and cv as in (6.9); thus b1 2 H 1 .GS ; Am2 / and
maps to a1 , bv;1 2 H 1 .Kv ; Am2 / and maps to bv , and cv 2 H 1 .Kv ; Am / and
maps to bv;1 b1;v . We shall show that there is an element b0 2 H 1 .GS ; Am /
such that
b0;v cv b1;v mod A.Kv /.m/
for all v. This will complete the proof, because then a1 C a0 , with a0 the image
of b0 in H 1 .GS ; A/, lies in X.K; A/ and is such that m.a1 aP 0 / D ma1 D a:
According to (6.14), an element b0 will exist if and only if v hcv ; bv0 i D 0
for all b 0 in SSP
.K; A/m . But, by definition of the pairing on the Tate-Shafarevich
groups (6.9), v hcv ; bv0 i D ha; a0 i where a0 is the image of b 0 in X.K; At /m ,
and our assumption on a is that this last term is zero. 2
We now complete the proof of part (a) of the theorem. Note that because the
groups are torsion, the pairing must kill the divisible subgroups. Consider the
diagram
0 !
X0 .K; A/=mX.K; A/ !
X.K; A/=mX.K; A/ ! X2S .K; Am /
? ? ?
? ? ?
y y y

.X.K; At /m = Im X1 .K; Atm // !
0 ! X.K; At /m X1S .K; Atm / .
!
The top row comes from (6.16) and the bottom row is the dual of an obvious
sequence
X1S .K; Atm / ! X.K; At /m ! Coker ! 0:
The first vertical map is the injection given by Lemma 6.17, and the third ver-
tical arrow is the isomorphism of (4.10). A diagram chase now shows that the
middle vertical arrow is also injective. On passing to the limit over powers of
m, we obtain an injection X.K; A/^ ! X.K; At /.m/ . But X.K; A/^ D
X.K; A/=X.K; A/mdiv , and so the left kernel in the pairing
X.K; A/.m/ X.K; At /.m/ ! Q=Z
is X.K; A/mdiv . Therefore
ŒX.K; A/=X.K; A/mdiv ŒX.K; At /=X.K; At /mdiv :
Since this holds for all A, we also have
ŒX.K; At /=X.K; At /mdiv ŒX.K; At t /=X.K; At t /mdiv
D ŒX.K; A/=X.K; A/mdiv :
It follows that all these orders are equal, and therefore that the right kernel is
X.K; At /mdiv .
R EMARK 6.18 If A has dimension one and m is prime, then X2S .K; Am / D 0
(see 9.6), and so X0 .K; A/m D X.K; A/m (see 6.17). Therefore in this case it
is significantly easier both to define the pairing on the Tate-Shafarevich groups
and to prove its nondegeneracy.
Complements in the case that S contains almost all primes

We shall now show how a theorem of Serre (1964/71) can be used to improve
some of these results when S omits only finitely many primes.
P ROPOSITION 6.19 Let m be an integer prime to char.K/, and let G be the im-
age of Gal.K s =K/ in Tm A. Then the group H 1 .G; Tm A/ is finite.
P ROOF. When m is prime, this is proved in Serre 1964/71, II 2, and the result for
a composite m follows immediately. 2
We give a second proof of (6.19) based on a theorem of Bogomolov and

a lemma of Sah. Note that (6.1) shows that the action of GK on Tm A fac-
tors through GS . Also that, because Tm A is a Zm -module, Zm is a subring of
End.Tm A/ and Z m is a subgroup of Aut.Tm A/:
L EMMA 6.20 For any prime ` ¤ char.K/, the image of GS in Aut.T` A/ con-
tains an open subgroup of Z
`
:
P ROOF. Theorem 3 of Bogomolov 1981 shows that (at least when K is a number
field), for any prime `, the Lie algebra of the image of GK in Aut.T` A/ contains
the scalars. This implies that the image of GK is open in Z`
. 2
L EMMA 6.21 Let G be a profinite group and M a G -module. For any element
of the centre of G , H r .G; M / is annihilated by x 7! x x:
P ROOF. We first allow to be any element of G, not necessarily a central ele-

ment. The maps
g 7! g 1 W G ! G; m 7! 1 mW M ! M;
are compatible, and so define automorphisms ˛ r W H r .G; M / ! H r .G; M /. Ac-

cording to (0.15), ˛ r is the identity map. In the case that central, one sees
by looking on cochains that ˛ r is the map induced by the G-homomorphism
1 W M ! M . Consequently, acts as the identity map on H r .G; M /, as
claimed by the lemma. 2
We now (re-)prove 6.19. It follows from (6.20) that there is an integer i and
an element 2 G such that x D .ì 1/x, all x 2 T` A. Now (6.21) shows
that ì H 1 .G; T` A/ D 0. Corollary 4.15 implies that H 1 .G; A`n / is finite for all
n, and so the inverse limit of the exact sequences
ì
H 1 .G; A`n / ! H 1 .G; A`nCi / ! H 1 .G; Aì /
0
H 1 .G; T` A/ ! H 1 .G; T` A/ ! H 1 .G; Aì /;
and so H 1 .G; T` A/ is a subgroup of the finite group H 1 .G; Aì /:
P ROPOSITION 6.22 If S omits only finitely many primes of K , then the map
Q
H 1 .GS ; Tm A/ ! v2S H 1 .Kv ; Tm A/
is injective.
P ROOF. Write X1S .K; Tm A/ for the kernel of the map in the statement of the
proposition. Then there is an exact commutative diagram
0 ! H 1 .G; Tm A/ ! H 1 .GS ; Tm A/ ! H 1 .G 0 ; Tm A/

? ? ?
? ? ?
y y y
L 1 L 1 L 1 0
0 ! H .G v ; Tm A/ ! H .Gv ; Tm A/ ! H .Gv ; Tm A/
v2S v2S v2S
in which G and G v are the images of G and Gv in Aut.Tm A/ and G 0 and Gv0
are the kernels of G G and Gv G v . The two right hand groups consist of
continuous homomorphisms, and so the Chebotarev density theorem shows that
the right hand vertical map is injective. It follows that the subgroup X1S .K; Tm A/
of H 1 .GS ; Tm A/ is contained in H 1 .G; Tm A/, and is therefore torsion. Since
Tm H 1 .GS ; A/ is torsion free, the sequence
0 ! A.K/^ ! H 1 .GS ; Tm A/ ! Tm H 1 .GS ; A/
now shows that any element c of X1S .K; Tm A/ is in A.K/^ . But for any nonar-
chimedean prime v, the map A.K/^ ! A.Kv /^ is injective on torsion points,
and so c D 0: 2
C OROLLARY 6.23 Assume S omits only finitely many primes.

(a) There is an injection

Q
Tm X1 .K; A/ ! v2S H
0
.Kv ; A/^ =A.K/^ :
(b) There is a sequence of injective maps

Q
A.K/^ ! lim SS .K; A/m ! v2S H
0
.Kv ; A/^ .

In particular, X0S .K; A/ D 0. The kernel of A.K/ ! A.K/^ is the
subgroup of elements with finite order prime to m.
P ROOF. Consider the diagram

a
0 !
A.K/^ !
H 1 .GS ; Tm A/
! Tm H 1 .GS ; A/ !
0
? ? ?
? ?c ?
y y yb
Q ^
Q d Q
0 !
v2S H
0 .K
v ; A/ !
v2S H
1 .K
v ; Tm A/
! v2S Tm H
1 .K
v ; A/ !
0:
The vertical arrow marked c is injective, and that marked b has kernel Tm X.K; A/.
Therefore part (a) follows from the snake lemma. The first map in part (b) is the
inclusion Ker.a/ ,! Ker.b ı a/. The second is the injection Ker.d ı c/ ,!
Ker.d /. 2
C OROLLARY 6.24 Let S be as in the proposition. The map

L 2
H 2 .GS ; A/ ! H .Kv ; A/
v real
is an isomorphism; in particular, X2S .K; A/ D 0:
P ROOF. We have a commutative diagram
H 1 .GS ; A/ ˝ Qm =Zm !
0 ! H 2 .GS ; A.m// !
H 2 .GS ; A/.m/ !
0
? ? ?
? ? ?
y y y
˚H 1 .Kv ; A/ ˝ Qm =Zm !
0 ! ˚H 2 .Kv ; A.m// !
˚H 2 .Kv ; A/.m/ !
0:
Because H 1 .Kv ; A/ is torsion, its tensor product with Qm =Zm is zero. Therefore
a nonzero element of X2S .K; A/.m/ would give rise to a nonzero element of
X2S .K; A.m//, but this group is dual to X1S .K; Tm At /, which the proposition
shows to be zero. Therefore the map is injective, and it was shown to be surjective
in (6.13c). 2
R EMARK 6.25 Note that (6.23b) solves the congruence subgroup problem for
subgroups of A.K/ of index prime to the characteristic of K: any such subgroup
contains a subgroup defined by congruence conditions. (In fact, that was Serre’s
purpose in proving (6.19).)
On combining the above results with Theorem 6.13, we obtain the following
theorem.
T HEOREM 6.26 Assume that S omits only finitely many primes of K:

(a) The left and right kernels of the canonical pairing
X.K; A/.m/ X.K; At /.m/ ! Q=Z
are the divisible subgroups of X.K; A/.m/ and X.K; At /.m/:

(b) The Tate-Shafarevich group X.K; A/.m/ is finite if and only if Im.ˇ 0 / D
Ker. 0 /, in which case there is an exact sequence
M
0 ! X.K; A/.m/ ! H 1 .GS ; A/.m/ ! H 1 .Kv ; A/.m/ ! At .K/^ ! 0:
v2S
(c) The groups XrS .K; A; m/ are zero for r ¤ 1, and for r 2, ˇ r is an
isomorphism
Š M
H r .GS ; A/ ! H r .Kv ; A/.m/:
v real
R EMARK 6.27 The Tate-Shafarevich group is not known18 to be finite for a sin-
gle abelian variety over a number field. However, there are numerous exam-
ples where it has been shown that some component X.K; A/.m/ is finite. The
first examples of abelian varieties over global fields known to have finite Tate-
Shafarevich groups are to be found in Milne 1967 and Milne 1968. There it
is shown that, for constant abelian varieties over a function field K, the Tate-
Shafarevich group is finite19 and has the order predicted by the conjecture of
Birch and Swinnerton-Dyer (see the next section for a statement of the conjec-
ture; an abelian variety over a function field K is constant if it is obtained by base
change from an abelian variety over the field of constants of K). See also Milne
18 Only a few months after the book was sent to the publisher, Rubin proved that the Tate-
Shafarevich groups of some elliptic curves over Q with complex multiplication are finite, and not
long after that Kolyvagin proved similar results for some modular elliptic curves.
19 Of course, this implies the finiteness of the Tate-Shafarevich group of an abelian variety that
becomes constant over a finite extension of the ground field.

7. CONJECTURE OF BIRCH AND SWINNERTON-DYER 93
1975, where (among other things) it is shown that the same conjecture is true for
the elliptic curve
Y 2 D X.X 1/.X T /
over k.T /, k finite.
N OTES Theorem 6.13 was proved by Cassels in the case of elliptic curves (Cassels 1962,
1964) and by Tate in the general case (announcement Tate 1962). So far as I know, no
complete proof of it has been published before.20 The survey article Bashmakov 1972
contains proofs of parts of it, and Wake 1986 shows how to deduce (6.22), (6.23), and
(6.24) from (6.19); both works have been helpful in the writing of this section in the
absence of Tate’s original proofs. 21 22
7 An application to the conjecture of Birch and

Swinnerton-Dyer
The results of the preceding two sections will be applied to show that the con-
jecture of Birch and Swinnerton-Dyer, as generalized to abelian varieties by Tate,
is compatible with isogenies (except possibly for isogenies whose degree is di-
visible by the characteristic of K). We begin by reviewing the statement of the
conjecture in Tate 1965/66, 1. Throughout, A and B will be abelian varieties of
dimension d over a global field K, and G D Gal.K s =K/.
L-series.
Let v be a nonarchimedean prime of K, and let k.v/ be the corresponding residue
field. If A has good reduction at v, then it gives rise to an abelian variety A.v/
over k.v/. The characteristic polynomial of the Frobenius endomorphism of A.v/
is a polynomial Pv .TQ / of degree 2d with Qcoefficients in Z such that, when we
factor it as Pv .T / D i .1 ai T /, then i .1 ai / is the number of points on
m
A.v/ with coordinates in the finite field of degree m over k.v/ (see, for example,
Milne 1986b, 19). It can be described also in terms of V` A Ddf Q` ˝ T` A.
Let Dv Iv be the decomposition and inertia groups at v, and let F rv be the
Frobenius element of Dv =Iv . Then (6.1) shows that Iv acts trivially on T` A, and
20 In fact, before the publication of the original version of this book, no proof of Theorem 6.13
was available.
21 David Harari and Tamás Szamuely have shown that the global duality theorems for
tori and abelian varieties can be combined to give a duality theorem for one-motives
(arXiv:math.NT/0304480, April 30,2003).
22 This section should be rewritten in terms of generalized Selmer groups.
it is known (ibid.) that
Pv .A; T / D det.1 .F rv /T jV` A/; ` ¤ char.k/:
For any finite set S of primes of K including the archimedean primes and
those where A has bad reduction, we define the L-series LS .s; A/ by the formula
Q
LS .s; A/ D v…S Pv .A; N v s /1
where N v D Œk.v/. Because the inverse roots ai of Pv .T / have absolute value

q 1=2 , the product is dominated by
K .s 1=2/2d , and it therefore converges for
<.s/ > 3=2. It is widely conjectured that LS .s; A/ can be analytically continued
to a meromorphic function on the whole complex plane. This is known in the
function field case, but in the number field case it has been verified only for mod-
ular elliptic curves,23 abelian varieties with potential complex multiplication, and
some other abelian varieties.
Let ! be a nonzero global differential d -form on A. As .A; ˝A d / has di-
mension 1, ! is uniquely determined up to multiplication by an element of K .

For each nonarchimedean prime v of K, let v be the Haar measure on Kv for
which Ov has measure 1, and for each archimedean prime, take v to be the usual
Lebesgue measure on Kv . With these choices, we have .cU / D jcjv v .U / for
any c 2 K and compact U Kv . Just as a differential on a manifold and a
measure on R define a measure on the manifold, so do ! and v define a measure
on A.Kv /, and we set
Z
v .A; !/ D j!jv dv
A.Kv /
Q
R 1961). Let be the measure v on the adèle ring AK of K, and set
(see Weil
jj D AK =K . For any finite set S of primes of K including all archimedean
primes and those nonarchimedean primes for which A has bad reduction or such
that ! does not reduce to a nonzero differential d -form on A.v/, we define
jjd
LS .s; A/ D LS .s; A/ Q .
v2S v .A; !/
This function is independent of the choice of !; if ! 0 D c! is a second differ-

ential d -form on A having good reduction outside S, then c must be a unit at all
primes outside S, and so the product formula shows that
Q 0 Q
v2S v .A; ! / D v2S v .A; !/:
23 Hence all elliptic curves defined over Q!

The function LS .s; A/ depends on the choice of S, but its asymptotic behaviour
as s approaches 1 does not, because if v is a prime at which A and ! have good
reduction at v, then it is known that v .A; !/ D ŒA.k.v//=.N v/d (ibid., 2.2.5),
and it is easy to see that this equals Pv .A; N v 1 /.
Heights
The logarithmic height of a point x D .x0 W ::: W xm / in Pm .K/ is defined by

Q
h.x/ D log max fjxi jv g :
all v 0im
The product formula shows that this is independent of the representation of x.

Let D be a very ample divisor on A. After replacing D with D C .1/ D, we
may assume that D is linearly equivalent to .1/ D. Let f W A ! Pn be the
embedding defined by D, and for a 2 A.K/, let 'D .a/ be the point in At .K/
represented by the divisor .D Ca/ D. Then there is a unique bi-additive pairing
h ; iW At .K/ A.K/ ! R
such that h'D .a/; ai C 2h.f .a// is bounded on A.K/. The discriminant of the
pairing is known to be nonzero. The pairing is functorial in the sense that if
f W A ! B is an isogeny, then the diagram
At .K/ A.K/ ! R
"f t #f k
B t .K/ B.K/ ! R
commutes. (See Lang 1983, Chapter V:)
Statement.
In order to state the conjecture of Birch and Swinnerton-Dyer we need to assume
that the following two conjectures hold for A:
(a) the function LS .s; A/ has an analytic continuation to a neighbourhood of
1;
(b) the Tate-Shafarevich group X.K; A/ of A is finite.
The conjecture then asserts:
LS .s; A/ ŒX.K; A/ j dethai0 ; aj ij

lim D P P (B-S/D)
s!1 .s 1/r At .K/W Zai0 .A.K/W Zai /
where r is the common rank of A.K/ and At .K/, and .ai0 /1ir and .ai /1ir
are families of elements of At .K/ and A.K/ that are linearly independent over
Z.
L EMMA 7.1 Let A and B be isogenous abelian varieties over a global field K ,
and let S be a finite set of primes including all archimedean primes and all primes
at which A or B has bad reduction.
(a) The functions LS .s; A/ and LS .s; B/ are equal. In particular, if one func-
tion can be continued to a neighbourhood of s D 1, then so also can the
other.
(b) Assume that the isogeny has degree prime to the char.K/. If one of X.A/
of X.B/ is finite, then so also is the other.

P ROOF. (a) An isogeny A ! B defines an isomorphism V` A ! V` B, and so
the polynomials Pv .T / are the same for A and for B:
(b) Let f W A ! B be the isogeny, and let Af be the kernel of f . Enlarge S
so that deg.f / is a unit in RK;S . Then (6.1) gives us an exact sequence
f
! H 1 .GS ; Af / ! H 1 .GS ; A/ ! H 1 .GS ; B/ ! .
According to (4.15), H 1 .GS ; Af / is finite, and so the kernels of f W H 1 .GS ; A/ !

H 1 .GS ; B/ and a fortiori X.f /W X.K; A/ ! X.K; B/ are finite. Therefore
if X.K; B/ is finite, so also is X.K; A/, and the reverse implication follows
by the same argument from the fact there exists an isogeny gW B ! A such that
g ı f D deg.f /: 2
Before stating the main theorem of this section, it is convenient to make an-
other definition. If f W X ! Y is a homomorphism of abelian groups with finite
kernel and cokernel, we define
ŒKer.f /
z.f / D :
ŒCoker.f /
L EMMA 7.2 (a) If X and Y are finite, then z.f / D ŒX =ŒY :
f g
(b) Consider maps of abelian groups X ! Y ! Z ; if any two of z.f /, z.g/,
and z.g ı f / are defined, then so also is the third, and z.g ı f / D z.g/z.f /:
(c) If X D .0 ! X 0 ! ! X n ! 0/ is a complex of finite groups, then
Q r .1/r Q r
ŒX D ŒH r .X /.1/ :
(d) If f W X ! Y is a map of exact sequences of finite length, and z.f r /

r
is defined for all r , then z.f r /.1/ D 1:
P ROOF. Part (b) is obvious from the kernel-cokernel sequence of the two maps.
Part (d) is obvious from the snake lemma when X and Y are short exact se-
quences, and the general case follows. The remaining statements are even eas-
ier. 2
T HEOREM 7.3 Assume that the abelian varieties A and B are isogenous by an
isogeny of degree prime to the char.K/. If the conjecture of Birch and Swinnerton-
Dyer is true for one of A or B , then it is true for both.
P ROOF. We assume that the conjecture is true for B and prove that it is then true
for A. Let f W A ! B be an isogeny of degree prime to the characteristic of K, and
let f t W B t ! At be the dual isogeny. Choose an element !B 2 .B; ˝B=K
d /, and

let !A be its inverse image f !B on A. Fix a finite set S of primes of K including
all archimedean primes, all primes at which A or B has bad reduction, all primes
whose residue characteristic divides the degree of f , and all primes at which
!B or !A does not reduce to a nonzero global differential form. Finally choose
linearly independent families of elements .ai /1ir of A.K/ and .bi0 /1ir of
B t .K/, where r is the common rank of the groups of K-rational points on the
four abelian varieties, and let bi D f .ai / and ai0 D f t .bi0 /. Then .ai0 /1ir
and .bi /1ir are linearly independent families of elements of At .K/ and B.K/.
The proof will proceed by comparing the corresponding terms in the conjectured
formulas for A and for B:
The functoriality of the height pairings shows that
hf t .bj0 /; ai i D hbj0 ; f .ai /i;
and this can be rewritten as
haj0 ; ai i D hbj0 ; bi i:
Therefore,
dethaj0 ; ai i D dethbj0 ; bi i:
The diagram
P P
0 ! Zai ! A.K/ ! A.K/= Zai ! 0
? ? ?
? ? ?
y yf .K/ y
P
0 ! Zbi ! B.K/ ! B.K/=˙ Zbi ! 0
and its analogue for f t , we see that
P t P
.A.K/ W Zai / B .K/ W Zbi0
z.f .K// D P ; z.f .K// D t
t
P :
.B.K/ W Zbi / A .K/ W Zai0
We have seen in (7.1) and (6.14c) that the finiteness of X.B/ implies that of
X.A/; X.At /, and X B t , and so (6.13a) shows that the two pairings in the
following diagram are nondegenerate,
X.A/ X.At / ! Q=Z

#f "f t k
X.B/ X.B t / ! Q=Z
Therefore, ŒCoker X.f / D ŒKer X.f t /, and so we have equalities
ŒX.A/ ŒKer X.f /

D z.X .f // D :
ŒX.B/ ŒKer X.f t /
Finally, consider the map f .Kv /W A.Kv / ! B.Kv /. By definition !A D

f !B , and so v .U; !A / D v .f U; !B / for any subset U of A.Kv / that is
mapped injectively into B.Kv /. Therefore,
v .A.Kv /; !A / D ŒKer f .Kv / v .f .A.Kv //; !B /:
Since
v .f .A.Kv //; !B / D ŒCoker.f .Kv //1 v .B.Kv //; !B /;
we see that
z.f .Kv // D v .A; !A /=v .B; !B /;
and so
Q
L .s; A/ v .B; !B / Q

D Q D z.f .Kv //1 :
L .s; B/ v .A; !A / v2S
On combining all the boxed formulas, we find that to prove the theorem it suffices
to show that
Q ŒKer X.f t / z.f .K//
z.f .Kv // D (7.3.1)
v2S ŒKer X.f / z.f t .K//
Consider the commutative diagram

L 0
0 ! H 0 .GS ; M / ! H .Kv ; M /
v2S
?
?
y
H 2 .GS ; M D /
?
?
y
0 ! Coker.f .K// ! H 1 .GS ; M / ! H 1 .GS ; A/f ! 0
? ? ?
? 0 ?' ? 00
y' y y'
L L 1 L 1
0 ! Coker.f .Kv // ! H .Kv ; M / ! H .Kv ; A/f ! 0
v2S v2S v2S
? ? ?
? 0 ? ? 00
y y y

0 ! H 1 .GS ; B t /f t ! H 1 .GS ; M D / ! .Coker f t .K/ ! 0
in which the rows are extracted from the cohomology sequences of

f
0 ! M ! A.KS / ! B.KS / ! 0
f
0 ! M ! A.Kvs / ! B.Kvs / ! 0
ft
0 ! M D ! B t .KS / ! At .KS / ! 0
respectively, and middle column is part of the exact sequence in Theorem 4.10.
The rows are exact. The duality between B.Kv / and H 1 .Kv ; B t / induces a du-
ality between B.Kv /=fA.Kv / and H 1 .Kv ; B t /f t , and the map 0 is the dual of
the composite
L L
H 1 .GS ; B t /f t ! v2S H
1
.Kv ; B t /f t ! v2S .B.Kv /=fA.Kv // :
The map 00 is the dual of the composite

L L
At .K/=f t B t .K/ ! v2S A
t
.Kv /=f t B t .Kv / ! v2S .H
1
.Kv ; A/f / .
The two outside columns need not be exact, but it is clear from the diagram that
they are complexes.
The serpent lemma and a small diagram chase give us an exact sequence
0 ! Ker.' 0 / ! Ker.'/ ! Ker.' 00 / ! Ker. 0

/= Im.' 0 / ! 0:
As Ker.' 00 / D Ker.X.f //, we obtain the formula
ŒKer ' 0 ŒKer X.f /

D 1:
ŒKer ' ŒKer 0 = Im ' 0
From the first column, we get (using (7.2c) and that Coker. 0/ D .Ker X.f t // /
ŒCoker .f .K// ŒKer ' 0

Q ŒH 1 .GS ; B t /f t D 0 =Im' 0
ŒKer X.f t /:
v2S ŒCoker .f .K v // ŒKer
From the third row, we get
ŒH 1 .GS ; B t /f t
1D 1 D
ŒCoker f t .K/ :
ŒH .GS ; M /
From the middle column we get (using that H 0 .GS ; M / D Ker f .K/, . . . )
ŒKer f .K/ Y ŒH 0 .Kv ; M / ŒH 2 .GS ; M D /

1D Q :
ŒKer f .Kv /
v arch
ŒHT0 .Kv ; M / ŒKer '
Finally, we have the obvious equality
ŒKer f t .K/ D ŒH 0 .GS ; M D /:
On multiplying these five equalities together, we find that

Y ŒKer X.f t / z.f .K// Y ŒH 0 .Kv ; M /
D
z.f .Kv // D .GS ; M / :
ŒKer X.f / z.f t .K// ŒHT0 .Kv ; M /
v2S v arch
Theorem 5.1 (in the form (5.2a)) shows that the product of the last two terms on
the right of the equation is 1, and so this completes the proof of the theorem. 2
R EMARK 7.4 Since in the number field case the conjecture of Birch and Swinnerton-
Dyer is not known24 for a single abelian variety, it is worth pointing out that the
above arguments apply to the m-primary components of the groups involved: if
X.K; A/.m/ is finite and has the order predicted by the conjecture, then the same
is true of any abelian variety isogenous to A.
R EMARK 7.5 We mention two results of a similar (but simpler) nature to (7.3).
Let A be an abelian variety over a finite separable extension F of the global
field K. Then A gives rise to an abelian variety A over K by restriction of
24 As (foot)noted earlier, this is no longer the case.
8. ABELIAN CLASS FIELD THEORY 101
scalars. The conjecture of Birch and Swinnerton-Dyer holds for A over F if and
only if it holds for A over K (see Milne 1972, Thm 1).
Let A be an abelian variety over a number field K, and assume that it acquires
complex multiplication over F , and that F is the smallest extension of K for
which this is true. Under certain hypotheses on A, it is known that the conjecture
of Birch and Swinnerton-Dyer holds for A over K if and only if it holds for AF
over F (ibid. Corollary to Thm 3).
N OTES For elliptic curves, Theorem 7.3 was proved by Cassels (1965). The general case
was proved by Tate (announcement Tate 1965/66, Theorem 2.1). The above proof was
explained to me by Tate in 1967.
8 Abelian class field theory, in the sense of Langlands

Abelian class field theory for a global field K defines a reciprocity map
recK W CK ! Gal.K s =K/ab that classifies the finite abelian extensions of K.
Dually, one can regard it as associating a character ı recK of CK with each
(abelian) character of Gal.K s =K/ of finite order; the correspondence is such
that the L-series of and ı recK are equal. It is this second interpretation
that generalizes to the nonabelian situation. For any reductive group G over a
local or global field K, Langlands has conjectured that it is possible to associate
an automorphic representation of G with each “admissible” homomorphism of
the Weil group WK of K (Weil-Deligne group in the case of a local field) into
a certain complex group L G; the L-series of the automorphic representation is
to equal that of the Weil-group representation. In the case that G D Gm , the
correspondence is simply that noted above. For a general reductive group, the
conjecture is difficult even to state since it requires a knowledge of representation
theory over adèle groups (see Borel 1979). For a torus however the statement of
the conjecture is simple, and we shall prove it in this case. First we prove a dual-
ity theorem (8.6), and then we explain the relation of the theorem to Langlands’s
conjectural class field theory. In contrast to the rest of these notes, in this section
we shall consider cohomology groups H r .G; M / in which G is not a profinite
group. The symbol H r .G; M / will denote the group constructed without regard
for topologies, and Hcts r .G; M / will denote the group defined using continuous
cochains. As usual, when G is finite, HTr .G; M /, r 2 Z, denotes the Tate group.
For a topological group M ,
M D Homcts .M; Q=Z/ D group of characters of M of finite order;
M u D Homcts .M; R=Z/ D group of characters of M (the Pontryagin dual
of M /I
M 0 D Homcts .M; C=Z/ D Homcts .M; C / D group of generalized charac-

ters of M I
M D Hom.M; C / D group of generalized (not necessarily continuous)
characters of M . When M is discrete, M 0 D M .
As usual, when K is a global field, we write CK for the idèle class group of
K. In order to be able to give uniform statements, we sometimes write CK for
K when K is a local field.
Weil groups
First we need to define the Weil group of a local or global field K. This is a
triple .WK ; '; .rF // comprising a topological group WK , a continuous homo-
morphism 'W WK ! Gal.K s =K/ with dense image, and a family of isomor-

phisms rF W CF ! WFab , one for each finite extension F K s of K, where
WF D ' 1 .GF /. (Here, as always, WKab is the quotient of WK by the closure
WKc of its commutator subgroup.)
For any finite extension F of K, define WF =K D WK =WFc ; then, if F is
Galois over K, there is an exact sequence
0 ! CF ! WF =K ! GF =K ! 0
whose class in H 2 .GF =K ; CF / is the canonical class (that is, the element denoted
by uGF=K in the second paragraph of 1). The topology on WF =K is such that
CF receives its usual topology and is an open subgroup of WF =K . The full Weil
group WK is equal to the inverse limit lim WF =K (as a topological group).

E XAMPLE 8.1 (a) Let K be a nonarchimedean local field. The Weil group WK is
the dense subgroup of GK consisting of elements that act as an integral multiple of
the Frobenius automorphism on the residue field. It therefore contains the inertia
subgroup IK of GK , and the quotient WK =IK is Z. The topology on WK is that
for which IK receives the profinite topology and is an open subgroup of WK . The
map ' is the inclusion map, and rF is the unique isomorphism F ! WFab such
that rF followed by ' is the reciprocity map.
(b) Let K be an archimedean local field. If K D C, then WK is C , ' is
the trivial map C ! Gal.C=C/, and rK is the identity map. If K is real, then
WK D K s t jK s (disjoint union) with the rules j 2 D 1 and jzj 1 D z
(complex conjugate). The map ' sends K s to 1 and j to the nontrivial element
of GK . The map rK s is the identity map, and rK is characterized by
rK .1/ D jWKc
1
rK .x/ D x 2 WKc for x 2 K, x > 0:
(c) Let K be a function field in one variable over a finite field. The Weil
group WK is the dense subgroup of Gal.K s =K/ of elements that act as an in-
tegral multiple of the Frobenius automorphism on the algebraic closure of the
field of constants. It therefore contains the geometric Galois group GKks D
Gal.K s =Kk s / GK , and the quotient of W by GKks is Z. The topology on
WK is that for which GKks receives the profinite topology and is an open sub-
group of WK . The map ' is the inclusion map, and rF is the unique isomorphism
CF ! WFab such that rF followed by ' is the reciprocity map.
(d) Let K be an algebraic number field. Only in this case, which of course
is the most important, is there no explicit description of the Weil group. It is
constructed as the inverse limit of the extensions corresponding to the canonical
classes uGF=K (see Artin and Tate 1961, XV, where the Weil group is constructed
for any class formation, or Tate 1979).
Let K be a global field. For each prime v of K, it is possible to construct a

commutative diagram
'v
WKv ! GKv
? ?
? ?
y y
'
WK ! GK ;
(see Tate 1979, 1.6.1). We shall assume in the following that one such diagram
has been selected for each v.
Some cohomology
We regard the cohomology and homology groups as being constructed using
the standard complexes. For example, H r .G; M / D H r .C .G; M // where
C .G; M / consists of maps Œg1 ; : : : ; gr 7! ˛.g1 ; : : : ; gr /W G r ! M . When
G is finite, the groups HT1 .G; M / and HT0 .G; M / are determined by the exact
sequence
NG
0 ! HT1 .G; M / ! MG ! M G ! HT0 .G; M / ! 0:
L EMMA 8.2 Let G be a finite group, and let Q be an abelian group regarded as
a G -module with trivial action. If Q is divisible, then for all G -modules M , the
cup-product pairing
HTr1 .G; Hom.M; Q// HTr .G; M / ! HT1 .G; Q/ Q
induces an isomorphism
HTr1 .G; Hom.M; Q// ! Hom.HTr .G; M /; Q/
for all r .
P ROOF. This is proved in Cartan and Eilenberg 1956, XII 6.4. 2
Let .G; C / be a class formation, and let G be the quotient of G by an open

normal subgroup H . The pairing
.f; c/ 7! f .c ˝ /W .C H ˝ M / C H ! M
and the canonical class u 2 H 2 .G; C H / define maps
a 7! a Y uW HTr .G; .C H ˝ M / / ! HTrC2 .G; M /:
L EMMA 8.3 For all finitely generated torsion-free G -modules M and all r , the
map
Y uW HTr .G; .C H ˝ M / / ! HTrC2 .G; M /
is an isomorphism.
P ROOF. The diagram
HTr .G; .C H ˝ M / / HTr1 .G; C H ˝ M // ! HT1 .G; C / C

#Yu "uY k
HTrC2 .G; M / HTr3 .G; M / ! HT1 .G; C / C
commutes because of the associativity of cup products:
.a Y u/ Y b D a Y .u Y b/; a 2 HTr .G; .C H ˝ M / /; b 2 HTr3 .G; M /:
The two pairings are nondegenerate by (8.2), and the second vertical map is an
isomorphism by virtue of the Tate-Nakayama theorem (0.2). It follows that the
first vertical map is an isomorphism. 2
Note that
.C H ˝ M / D Hom.C H ˝ M; C /
df
D Hom.C H ; Hom.M; C //
D Hom.C H ; M /:
Therefore, the isomorphism in the above lemma can also be written

Y uW HTr .G; Hom.C H ; M // ! HTrC2 .G; M /:
Let
0 ! CH ! W ! G ! 1
be an exact sequence of groups corresponding to the canonical class u in
H 2 .G; C H /. For any W -module M , the Hochschild-Serre spectral sequence
gives an exact sequence
Inf Res
0 ! H 1 .G; M C / ! H 1 .W ; M / ! H 1 .C H ; M /G ! H 2 .G; M C /:
The map (the transgression) has the following explicit description: let a 2
H 1 .C H ; M /G , and choose a 1-cocycle ˛ representing it; extend ˛ to a 1-cochain
ˇ on W ; then dˇ is a 2-cocycle on G, and the class it represents is .a/.
L EMMA 8.4 If C H acts trivially on M , then the transgression
W H 0 .G; Hom.C H ; M // ! H 2 .G; M /
is the negative of the map Y u induced by the pairing
Hom.C H ; M / C H ! M:
S H
P ROOF. Write W D C wg (disjoint union of right cosets), and let
wg wg 0 D .g; g 0 /wgg 0 . Then ..g; g 0 // is a 2-cocycle representing u. Let
˛ 2 HomG .C H ; M /, and define ˇ by ˇ.cwg / D ˛.c/, c 2 CH . Then
dˇ.g; g 0 / D dˇ.wg ; wg 0 /
df
D gˇ.wg 0 / ˇ.wg wg 0 / C ˇ.wg /

D 0 ˛..g; g 0 // C 0
D ˛..g; g 0 //;
which equals .˛ Y /.g; g 0 /. Therefore .˛/ D ˛ Y u: 2
The duality theorem

Let K be a local or global field (we could in fact work abstractly with any class
formation), and let F be a finite Galois extension of K. Let M be a finitely
generated torsion-free GF =K -module. Then
M 0 D Homcts .M; C / D Hom.M; C / D M

df
are again GF =K -modules. We shall use the notation M 0 when we wish to em-
phasize that M 0 has a topology. We frequently regard these groups as WF =K -
modules.
F
Write WF =K D wg CF (disjoint union of left cosets). For any homomor-
phism ˛W CF ! M , the map Cor.˛/W WF =K ! M such that
P
.Cor.˛//.w/ D wg ˛.wg1 wwg 0 /; wwg 0 wg mod CF 0
g2G
is a cocycle, and so we have a map
CorW H 1 .CF ; M / ! H 1 .WF =K ; M /;
called the corestriction map. It is independent of the choice of coset representa-

tives (see Serre 1962, VII 7, or Weiss 1969, p81). It is clearly continuous, and so
maps continuous homomorphisms to continuous cocycles.
L EMMA 8.5 The corestriction map CorW H 1 .CF ; M / ! H 1 .WF =K ; M / fac-
tors through H 1 .CF ; M /G , G D GF =K :
P ROOF. Let ˛ 2 Hom.CF ; M / and h 2 G. Then .h˛/.w/ D wh ˛.wh1 wwh /

(this is the definition), and so
P
Cor.h˛/.w/ D g wg wh ˛.wh1 wg1 wwg 0 wh /
where g 0 is such that wwg 0 wg mod CF . The family .wg wh /g2G is also a set
of coset representatives for CF in WF =K , and w.wg 0 wh / .wg wh / mod CF .
Therefore the class of Cor.h˛/ is the same as that of Cor.˛/, and so Cor..h
1/˛/ D 0 in H 1 .WF =K ; M /: 2
T HEOREM 8.6 For any finitely generated torsion-free GF =K -module M , the core-
striction map defines an isomorphism

Homcts .CF ; M 0 /GF=K ! Hcts
1
.WF =K ; M 0 /:
P ROOF. Throughout the proof, we write G for GF =K . We shall first prove that
the corestriction map defines an isomorphism
Hom.CF ; M /G ! H 1 .WF =K ; M /
and then show (in Lemma 8.9) that it makes continuous homomorphisms corre-
spond to continuous cocycles.
Consider the diagram (8.6.1)
NG
HT1 .G; Hom.CF ; M // !
0! Hom.CF ; M /G ! Hom.CF ; M /G !
HT0 .G; Hom.CF ; M //
? ? ? ?
? ? ? ?
y yCor yid y
0!
H 1 .G; M / !
H 1 .WF =K ; M / !
H 1 .CF ; M /G !
H 2 .G; M /:
The top row is the sequence defining the Tate cohomology groups of Hom.CF ; M /.
The bottom row can be deduced from the Hochschild-Serre spectral sequence or
else can be constructed in an elementary fashion. The two isomorphisms are those
in Lemma 8.3. The third square (anti-) commutes because of (8.4). We shall prove
in the next two lemmas that the first two squares in the diagram commute. The
five-lemma will then show that CorW Hom.CF ; M /G ! H 1 .WF =K ; M / is an
isomorphism. Finally Lemma 8.9 will complete the proof. 2
L EMMA 8.7 The first square in (8.6.1) commutes.
P ROOF. We first show that Cor maps an element of HT1 .G; Hom.CF ; M //
into the subgroup H 1 .G; M / of H 1 .WF =K ; M /. Let ˛ be a homomorphism
CF ! M , and let c 2 CF and w 2 W . Then
X
.Cor.˛//.cw/ D wg ˛.wg1 cwg wg1 wwg 0 /
g
X
D .wg ˛/.c/ C .Cor ˛/.w/
g
D .N˛/.c/ C .Cor ˛/.w/:
Therefore, if N˛ D 0 (that is, ˛ 2 HT1 .G; Hom.CF ; M ///, then (Cor ˛/.w/
depends only on the class of w in G, and so Cor.˛/ arises by inflation from an
element of H 1 .G; M /.
It remains to show that the restriction of Cor to HT1 .G; Hom.CF ; M // is
Y u. Note that
X
.Cor.˛//.h/ D g.˛.wg1 wh wh1 g //
g
X
D .g˛/.wh wh1 g wg1 /
g
X
D .g˛/.u.h; h1 g//:
g
To obtain the middle equality, we have used that
wg1 wh wh1 g D c H) wh wh1 g D wg c D .gc/wg H) wh wh1 g wg1 D gc
and that g.˛.c// D .g˛/.gc/:

It is difficult to give explicit descriptions of cup-products when both negative
and positive indices are involved. We shall use the exact sequence
0 ! IG ! ZŒG ! Z ! 0
to shift the problem. This sequence remains exact when tensored with M and
Hom.CF ; M /, and the boundary maps in the resulting cohomology sequences
give the horizontal maps in the following diagram:
d 1
HT1 .G; Hom.CF ; M // ! HT0 .G; Hom.CF ; M / ˝ IG /
? ?
?Yu ?
y yYu˝1
d1
H 1 .G; M / ! H 2 .G; M ˝ IG /:
Both boundary maps are isomorphisms, and Y u is the unique map making
the diagram commute. If we can show that the diagram still commutes when this
map is replaced with Cor, we will have proved the lemma. This we do by an ugly
cocycle calculation.
Note first that d 1 and d 1 have the following descriptions:
d 1 .˛/ D N.˛ ˝ 1/
D N.˛ ˝ 1/ .N˛/ ˝ 1
X
D g˛ ˝ .g 1/; ˛ 2 Hom.CF ; M /; N˛
g
D0
d 1 .ˇ/.g1 ; g2 / D g1 ˇ.g2 / ˝ .g1 1/; ˇ 2 Z 1 .G; M /; g1 ; g2 2 G:

If ˛ 2 Hom.CF ; M / has N˛ D 0, then
X
.d 1 ı Cor ˛/.g1 ; g2 / D g1 .g˛/.u.g2 ; g21 g// ˝ .g1 1/
g2G
and X
.d 1 ˛ Y .u ˝ 1//.g1 ; g2 / D .g˛/.u.g1 ; g2 // ˝ .g 1/:
g
P
An element of M ˝ IG can be written uniquely in the form g mg ˝ .g 1/.
P
Therefore a general element of C 1 .G; M ˝ IG / is of the form Fg ˝ .g 1/
with Fg a map G ! M , and a coboundary in B 2 .G; M ˝ IG / can be written
X X
d. Fg ˝ .g 1//.g1 ; g2 / D .g1 Fg 1 g .g2 /
1
g g
X
Fg .g1 g2 / C Fg .g1 // ˝ .g 1/ g1 :Fg .g2 / ˝ .g1 1/:
g
In obtaining the second expression, we have used that
P P
g1 . Fg .g2 / ˝ .g 1// D g1 Fg .g2 / ˝ .g1 g g1 /
P P
D g1 Fg .g2 / ˝ .g1 g 1/ g1 Fg .g2 / ˝ .g1 1/
P P
D g1 Fg 1 g .g2 / ˝ .g 1/ g1 Fg .g2 / ˝ .g1 1/.
1
Put
Fg .g2 / D .g˛/.u.g2 ; g21 g/I
then
P
.d Fg ˝ .g 1/ .d 1 ˛/ Y .u ˝ 1/ C .d 1 ı Cor.˛//.g1 ; g2 / D
P
.g˛/.g1 u.g2 ; g21 g11 g/ u.g1 g2 ; g21 g11 g/1 u.g1 ; g11 g/ u.g1 ; g2 /1 / ˝ .g 1/:
When we put h D g21 g11 g, this becomes

X
.g˛/.g1 u.g2 ; h/ u.g1 g2 ; h/1 u.g1 ; g2 h/ u.g1 ; g2 /1 / ˝ .g 1/;
and each term in the sum is zero because u is a 2-cocycle. Therefore

P
d .Fg ˝ .g 1// D .d 1 ˛/ Y .u ˝ 1/ .d 1 ı Cor.˛//;
which completes the proof of the lemma. 2
L EMMA 8.8 The composite

Cor Res
H 1 .CF ; M / ! H 1 .WF =K ; M / ! H 1 .CF ; M /
is equal to the norm NG . Hence the second square in (8.6.1) commutes.
P ROOF. For ˛ 2 Z 1 .CF ; M / and w 2 WF =K ,

X
Cor.˛/.w/ D wg ˛.wg1 wwg 0 /:
g
P
When w 2 CF , this becomes Cor.˛/.w/ D g2G g˛.g 1 wg/ D .NG ˛/.w/.2
L EMMA 8.9 Let ˛ 2 Hom.CF ; M 0 /; then ˛ 2 Homcts .CF ; M 0 / if and only if

Cor.˛/ 2 Zcts
1
.WF =K ; M 0 /:
P ROOF. Clearly ˛ 2 Z 1 .WF =K ; M 0 / is continuous if and only if its restriction

to CF is continuous. Therefore (8.8) shows that it suffices to prove that a homo-
morphism f W CF ! M 0 is continuous if and only if NG f is continuous. Since
NG is continuous, there is a commutative diagram
N
Homcts .CF ; M 0 / ! Homcts .CF ; M 0 /G !
HT0 .G; Homcts .CF ; M 0 // !
0
? ? ?
? ? ?
y y y
N
Hom.CF ; M 0 / ! Hom.CF ; M 0 /G HT0 .G; Hom.CF ; M 0 // !
! 0;
from which it follows that it suffices to show that
HT0 .G; Homcts .CF 0 ; M 0 // ! HT0 .G; Hom.CF ; M 0 //
is injective. In fact, following Labesse 1984, we shall prove much more. 2
L EMMA 8.10 For all r , the map
HTr .G; Homcts .CF ; M 0 // ! HTr .G; Hom.CF ; M 0 //
is an isomorphism.
P ROOF. We consider the cases separately.

(a) K LOCAL ARCHIMEDEAN . The only nontrivial case has K D R and
F D C. Here CK D C , and we shall use the exponential sequence
0 ! Z ! C ! C ! 0.
From it we get exact sequences
0 ! Hom.C ; M 0 / ! Hom.C; M 0 / ! Hom.Z; M 0 / ! 0
(because M 0 is divisible) and
0 ! Homcts .C ; M 0 / ! Homcts .C; M 0 / ! Homcts .Z; M 0 / ! 0
(because M 0 is a connected commutative Lie group). The groups Hom.C; M 0 /

and Homcts .C; M 0 / are uniquely divisible, and so are cohomologically trivial.
Therefore, we can replace CF in the statement of the lemma with Z, but then it
becomes obvious because Z is discrete.
(b) K LOCAL NONARCHIMEDEAN . Here CF D F . From Serre 1967a, 1.4,
we know that F contains a cohomologically trivial open subgroup V ; moreover
V contains a fundamental system .Vn / of neighbourhoods of zero with each Vn an
open subgroup, such that V =Vn is cohomologically trivial. (For example, when
F is unramified over K, it is possible to take V D OK :) Now, because M 0 is
divisible, Serre 1962, IX 6, Thm 9, shows that Hom.V; M 0 / and Hom.V =Vn ; M 0 /
are also cohomologically trivial. As Homcts .V; M / D lim Hom.V =Vn ; M 0 /, we
!
see that it also is cohomologically trivial. A similar argument to the above, using
the sequence
0 ! V ! F ! F =V ! 0
shows that it suffices to prove the lemma with CF replaced with F =V , but this
group is discrete.
(c) K GLOBAL . Here CF is the idèle class group. Define V CK to be
Q
Vv where Vv D O bv for v a nonarchimedean prime that is unramified in F and
Vv is a subgroup as considered in (b) for the remaining nonarchimedean primes.
This group has similar properties to the group V in (b). It therefore suffices to
prove the lemma with CF replaced with CF =V . In the function field case this is
discrete, and in the number field case it is an extension of a finite group by R
(with trivial action). In the first case the lemma is obvious, and in the second the
exponential again shows that R is the quotient of a uniquely divisible group by
a discrete group. 2
This completes the proof of the theorem.

C OROLLARY 8.11 Let K be a global or local field, and let M be a finitely gen-
erated torsion-free GK -module. There is a canonical isomorphism
'
..C ˝ M /GK /0 ! Hcts
1
.WK ; M 0 /
S
where C D CF :
P ROOF. Let F be a finite Galois extension of K splitting M . Any continuous
crossed homomorphism f W WK ! M 0 restricts to a continuous homomorphism
on WF . Because M is commutative and Hausdorff, f must be trivial on WFc
and so factors through WK =WFc Ddf WF =K . Consequently, the inflation map
1
Hcts .WF =K ; M 0 / ! Hcts
1
.WK ; M 0 / is bijective.
Next note that Hom.CF ; M 0 / D .CF ˝ M /0 . I claim that the canonical map
.CF ˝ M /0G ! ..CF ˝ M /G /0 is an isomorphism. Note that this is obviously
so when 0 is replaced with , because .CF ˝ M /G is the maximal subgroup of
CF ˝M on which G acts trivially, and so ..CF ˝M /G / is the maximal quotient
group on which G acts trivially, that is, it is ..CF ˝ M / /G . The diagram
0 ! H 1 .G; .CF ˝ M /0 / ! .CF ˝ M /0G ! ..CF ˝ M /0 /G
? ? ?
? ? ?injective
y y y

0 ! H 1 .G; .CF ˝ M / / ! .CF ˝ M /G ! ..CF ˝ M / /G
shows that the middle vertical arrow is injective. Now the diagram
.CF ˝ M /0G ! ..CF ˝ M /G /0

? ?
?injective ?
y y

.CF ˝ M /G ! ..CF ˝ M /G /
shows that .CF ˝ M /0G ! ..CF ˝ M /G /0 is injective, which proves the claim
since the map is obviously surjective. To complete the proof of the corollary, note
that
Hom.CF ; M 0 /GF=K D .CF ˝M /0GF=K D ..CF ˝M /GF=K /0 D ..C ˝M /GK /0 ;
and so the corollary simply restates the theorem. 2
R EMARK 8.12 (a) After making the obvious changes, the above arguments show
that there is a canonical isomorphism
..C ˝ M /GK /u ! Hcts

1
.WK ; M u /:
1 .W ; M 0 / becomes
(b) Replace M in (8.11) with its linear dual. Then Hcts K
1
Hcts .WK ; M ˝ C=Z/ and ..C ˝ M / K / becomes HomGK .M; C /0 . On the
G 0
2
other hand, (4.10) gives us an isomorphism Hcts .GK ; M / ! HomGK .M; C / ,
and
2 1 1
Hcts .GK ; M / ' Hcts .GK ; M ˝ Q=Z/ ' Hcts .WK ; M ˝ Q=Z/:
These results and their relations can be summarized as follows: for any finitely
generated torsion-free G-module M , there is a commutative diagram
1 .W ; M ˝ .Q=Z// / H 1 .W ; M ˝ .R=Z// / H 1 .W ; M ˝ .C=Z//

Hcts K cts K cts K

HomWK .M; C / / HomW .M; C / u / HomW .M; C /0 ;
K K
in which the horizontal maps are defined by the inclusions
Q=Z ,! R=Z ,! C=Z.

Application to tori
Let T be a torus over a field K. The dual torus T _ to T is the torus such that
X .T _ / is the linear dual X .T / of X .T /. When K is a global field, we
say that an element of Hcts 1
.WK ; M 0 / is locally trivial if it restricts to zero in
1
Hcts .WKv ; M 0 / for all primes v:
T HEOREM 8.13 Let K be a local or global field, and let T be a torus over K .
1 .W ; T _ .C// is canonically isomorphic to the group
(a) When K is local, Hcts K
of continuous generalized characters of T .K/:
(b) When K is global, there is a canonical homomorphism from
1 .W ; T _ .C// onto the group of continuous generalized characters of
Hcts K
T .AK /=T .K/. The kernel is finite and consists of the locally trivial classes.
P ROOF. Take M D X .T / in the statement of (8.11). Then Hom.M; R / D
T _ .R/ for any ring R containing a splitting field for T . In particular, M 0 D
T _ .C/, and so Hcts
1
.WK ; M 0 / D Hcts
1
.WK ; T _ .C//:
When K is local, ..X .T / ˝ CF /G /0 D .T .F /G /0 D T .K/0 , which proves
(a). In the global case, on tensoring the exact sequence
0 ! F ! JF ! CF ! 0
with X .T /, we obtain an exact sequence
0 ! T .F / ! T .AF / ! X .T / ˝ CF ! 0;
and hence an exact sequence
0 ! T .K/ ! T .AK / ! .X .T / ˝ CF /G ! H 1 .G; T .F //:
The last group in this sequence is finite, and so we have a surjection with finite
kernel ..X .T / ˝ CF /G /0 .T .AK /=T .K//0 . This, composed with the iso-
morphism

1
Hcts .WF =K ; T _ .C// ! ..X .T / ˝ CF /G /0
of the theorem, gives the map.
There is a commutative diagram:
1
Hcts .WK ; T _ .C// ! .T .AK /=T .K//0
? ?
? ?
y y
Q 1 Q
Hcts .WKv ; T _ .C// ! T .Kv /0 :
We have just seen that the lower horizontal map is an isomorphism, and the sec-
ond vertical map is injective because it is the dual of a surjective map. Therefore,
the kernels of the two remaining maps are equal, as claimed by the theorem. 2
Re-interpretation as class field theory

Let T be a torus over K, let M D X .T /, and let T _ be the torus such that
X .T _ / D M . Let GK act on T _ .C/ D Hom.M; C / through its action on
M , and define L T to be the semi-direct product T _ .C/ Ì GK . It is complex Lie
group with identity component L T ı D T _ .C/. A continuous homomorphism
'W WK ! L T is said to be admissible if it is compatible with the projections
onto GK . Two such homomorphisms ' and ' 0 are said to be equivalent if there
exists a t 2 L T ı such that ' 0 .w/ D t'.w/t 1 for all w. Write ˚K .T / for the
set of equivalence classes of admissible homomorphisms, and define ˘K .T / to
be T .K/0 when K is local and .T .K /=T .K//0 when K is global.
T HEOREM 8.14 There is a canonical map ˚K .T / ! ˘K .T /; when K is local,

the map is an isomorphism, and when K is global, it is surjective with finite
kernel.
P ROOF. Any continuous homomorphism 'W WK ! L T can be written ' D

f with f and maps from WK into L T ı and GK respectively. One checks
immediately that ' is an admissible homomorphism if and only if f is a 1-cocycle
and is the map WK ! GK given as part of the structure of WK . Moreover,
every 1-cocycle arises in this way, and two '’s are equivalent if and only if the
corresponding 1-cocycles are cohomologous. Thus the theorem follows immedi-
ately from (8.13). 2
L-series
Let K be a nonarchimedean local field. For any representation of WK on a
finite-dimensional complex vector space V , the L-series
L.s; / D .det.1 .F r/N./s jV I /1
where F r is an element of WK mapping to 1 under the canonical map WK ! Z,

is a local uniformizing parameter, and I is the inertia group. For a global field
K and representation of L T , the Artin-Hecke L-series L.s; / is defined to be
the product of the local L-series at the nonarchimedean primes. (It is possible also
to define factors corresponding to the archimedean factors, but we shall ignore
them.) For S a finite set of primes, we let LS .s; / be the product of the local
factors over all primes not in S:
Assume now that T splits over an unramified Galois extension F of K. On
tensoring
0 ! O
F !F !Z!0
with X .T /, we obtain an exact sequence
0 ! T .OF / ! T .F / ! X .T / ! 0
with T .OF / a maximal compact subgroup of T .F /. The usual argument (Serre

1967, 1.2) shows that H 1 .GF =K ; T .OF // D 0, and so there is an exact sequence
0 ! T .OK / ! T .K/ ! X .T /GF=K ! 0:
Let be a generalized character of T .K/, and assume that it is trivial on T .OK /

(we then say that is unramified). Such a gives rise to a generalized character
of X .T /GF=K , which we can extend to a generalized character Q of X .T /.
Because
Hom.X .T /; C / D X .T / ˝ C D T _ .C/ D L T 0 ;
we can view Q as an element of this last group. Let r be a representation of L T

(as a pro-algebraic group) on a finite-dimensional complex vector space V . We
define the L-series
Q /N.!/s jV /
L.s; ; r/ D det.1 r.;
where is an element of Gal.K s =K/ restricting to the Frobenius automorphism

on F .
Now let K be a global field, and let be a generalized character of
T .AK /=T .K/. By restriction, we get generalized characters v of Kv for each
v. Let F be a finite Galois extension of K splitting T , and choose a finite set of
primes S of K including all archimedean primes, all primes that ramify in F , and
all primes v for which v is ramified. Define the automorphic L-series
Y
LS .s; ; r/ D L.s; v ; rv /
v…S
where rv is the restriction of r to the local L-group.
T HEOREM 8.15 (a) Let K be a local field, and let T be a torus over K splitting
over an unramified extension of K . For all ' 2 ˚.T / and all representations r of
LT ,
L.x; r ı '/ D L.s; ; r/

where is the character of T .K/ corresponding to ' in (8.14).
(b) Let K be a global field, and let T be a torus over K . Let ' 2 ˚.T /,
and let be the corresponding element of ˘.T /. Choose a set S of primes of K
containing all archimedean primes, all primes that ramify in a splitting field for
T , and all primes v such that v is ramified. Then, for all representations r of
LT ,
LS .s; r ı '/ D LS .s; ; r/:
P ROOF. Only (a) has to be proved, and we leave this as an exercise to the reader.2
The general conjecture
Let K be a global field, and let G be a reductive group over K. Then G is

determined by certain linear data (a root datum), and the group L G 0 is defined
by the dual data. The full L-group L G is defined to be a semi-direct product
L G 0 Ì G . The set ˚.G/ of equivalence classes of admissible homomorphisms
K
WK ! L G is defined analogously to the case of a torus, but the analogue of a
generalized character of T .AK /=T .K/ is more difficult to define. Since G.AK /
is neither commutative nor compact, its interesting representations are infinite
dimensional. The correct notion is that of an irreducible automorphic represen-
tation of G. Langlands conjectures25 that it is possible to associate with each
' 2 ˚.G/ a (nonempty) set of irreducible automorphic representations of G. If
is associated with ', then the L-series of ' and are related as in (8.15b): let
r be a complex representation of L G; corresponding to almost all primes v of K,
it is possible to define a local L-series for and r; for each of these primes v, the
local L-series for and r is equal to the corresponding factor of the Artin-Hecke
L-series of r'. See Borel 1979.
N OTES The results in this section were proved in Langlands 1968 and again in Labesse
1984. While the above proof of (8.6) borrows from the proofs in both papers, it is some-
what simpler than each. For applications of the theorems, see Kottwitz 1984, Labesse
1984, and Shelstad 1986.
9 Other applications
We explain a few of the other applications that have been made of the duality
theorems in 2 and 4.
25 For G D GL , Langlands’s conjecture has been proved for function fields (Drinfeld, Laf-
n
forgue, et al.) and for local fields (Harris, Taylor, Henniart, et al.).
9. OTHER APPLICATIONS 117
The Hasse principle for finite modules

Let K be a global field, and let M be a finite module over GK . We say that the
Hasse principle holds for M if the map
Q
ˇ r .K; M /W H 1 .K; M / ! all v H 1 .Kv ; M /
is injective.
E XAMPLE 9.1 (a) Let F=K be a finite Galois extension of degree n such that the
greatest common divisor r of localp p ŒFw W Kv is strictly less than n. (For
degrees
example, let K D Q and F D Q. 13; 17/; then n D 4 and the local degrees
are all 1 or 2.) Consider the exact sequence

0 ! M ! .Z=nZ/ŒG ! Z=nZ ! 0
P P
in which G D Gal.F=K/ and is the augmentation map n 7! n . From
its cohomology sequence, we obtain an isomorphism Z=nZ ! H 1 .G; M /. Let
c generate H 1 .G; M /; then rc is a nonzero element of H 1 .K; M / mapping to
zero in all the local cohomology groups. Therefore X1 .K; M / ¤ 0, and the
Hasse principle does not hold for M:
(b) From the duality theorem (4.10), we see that, for M as in (a), X2 .K; M / ¤
0. For a more explicit example (based on the failure of the original form of the
Grunwald theorem) see Serre 1964, III 4.7.
In view of these examples, the theorem below is of some interest. For a mod-
ule M , we write K.M / for the subfield of K s fixed by Ker.GK ! Aut.M //.
Thus K.M / is the smallest splitting field of M . A finite group G is said to be
`-solvable if it has a composition series whose factors of order divisible by ` are
cyclic.
T HEOREM 9.2 Let M be a finite simple GK -module such that `M D 0 for some
prime `, and assume that Gal.K.M /=K/ is an `-solvable group.
(a) If S is a set of primes of K with Dirichlet density one, then the mapping
Y
ˇS1 .K; M /W H 1 .K; M / ! H 1 .Kv ; M /
v2S
is injective.
(b) If ` ¤ char.K/, then the mapping
Y
ˇ 2 .K; M /W H 2 .K; M / ! H 2 .Kv ; M /
all v
is injective.
Note that ˇSr is not quite the same as the map in 4. However the next lemma
shows that Ker ˇS1 .K; M / D X1S .K; M /. For any profinite group G and G-
module M , define H1 .K; M / to be the kernel of
Y
H 1 .G; M / ! H 1 .Z; M /;
Z
where the product is over all closed cyclic subgroups Z of G. When G D GK , we

also write H1 .K; M / for H1 .G; M /. The next result explains the significance
of this notion for the theorem. As always, for each prime v of K, we choose an
extension w of v to K s :
L EMMA 9.3 Let M be a finite GK -module, and let F K s be a finite Galois
extension of K containing K.M /. Let S be a set of primes of K with Dirichlet
density one, and let
Y
ˇS1 .F=K; M /W H 1 .GF =K ; M / ! H 1 .GFw =Kv ; M /
v2S
be the map induced by the restriction maps. Then there is a commutative diagram
'
Ker.ˇS1 .F=K; M //
! Ker.ˇS1 .K; M //
\ \
'
H1 .Gal.F=K/; M /
! H1 .GK ; M /
The inclusions become equalities when all the decomposition groups Gal.Fw =Kv /
are cyclic.
P ROOF. There is an exact commutative diagram

Inf
0
! H 1 .Gal.F=K/; M / !
H 1 .K; M /
! H 1 .F; M /
? ? ?
? 1 ? 1 ? 1
yˇS .F =K;M / yˇS .K;M / yˇS .F ;M /
Q 1 Inf Q 1 Q 1
0
! H .Gal.Fw =Kv /; M / ! H .Kv ; M / ! H .Fw ; M /:
v2S v2S v2S
The Chebotarev density theorem shows that ˇS1 .F; M / is injective. The inflation
map therefore defines an isomorphism of the kernels of the first two vertical maps,
which gives us the isomorphism on the top row. The isomorphism on the bottom
row can be proved by a similar argument. The Chebotarev density theorem shows
that all cyclic subgroups of Gal.F=K/ are of the form Gal.Fw =Kv / for some
primes wjv with v 2 S, and so clearly H1 .Gal.F=K/; M / Ker.ˇ 1 .F=K; M //.
The reverse inclusion holds if all the decomposition groups are cyclic. 2
We say that the Hasse principle holds for a finite group G (and the prime `/
if H1 .G; M / D 0 for all finite simple G-modules M (with `M D 0/. Note that
the Hasse principle obviously holds for G if all of its Sylow subgroups are cyclic.
L EMMA 9.4 Let

1 ! G 0 ! G ! G 00 ! 1
be an exact sequence of finite groups. If the Hasse principle holds for G 0 and G 00
relative to the prime `, then it holds for G and `; conversely, if the Hasse principal
holds for G and `, then it holds for G 00 and `:
P ROOF. Let M be a simple G-module such that `M D 0. As G 0 is normal in G,

0 0 0
M G is stable under G, and so either M G D 0 or M G D M . In the first case,
there is a commutative diagram
Res
H 1 .G; M / ! H 1 .G 0 ; M /
? ?
? ?
y y
Q 1 Res Q 1
H .Z; M / ! H .Z \ G 0 ; M /
0
in which the upper restriction map has kernel H 1 .G 00 ; M G / D 0. When regarded
as a G 0 -module, M is semisimple because, for any nonzero simple G 0 -module N
of M , M is a sum of simple modules gN , g 2 G. Therefore, if the Hasse
principle holds for G 0 and `, then the right hand vertical arrow is an injection.
Consequently, the first vertical arrow is also an injection, and this shows that the
Hasse principle holds for G and `.
0
In the case that M G D M , we consider the diagram
0 ! H 1 .G 00 ; M / !
H 1 .G; M / ! H 1 .G 0 ; M /
? ? ?
? ? ?
y y y
Q 1 Q Q
0 ! H .ZG 0 =G 0 ; M / ! H 1 .Z; M / ! H 1 .Z \ G 0 ; M /:
The right hand vertical arrow is an injection because G 0 acts trivially on M

and the groups Z \ G 0 generate G 0 . The left hand vertical arrow has kernel
H1 .G 00 ; M / because the groups ZG 0 =G 0 run through all cyclic subgroups of
G 00 , and so we see that if the Hasse principle holds for G 00 and ` then it holds also
for G and `.
We use the same diagram to prove the converse part of the lemma. A sim-
0
ple G 00 -module can be regarded as a simple G-module such that M G D M .
Therefore, the diagram shows that H1 .G 00 ; M / D 0 if H1 .G; M / D 0. 2
P ROPOSITION 9.5 (a) The Hasse principle holds for a finite group (and the prime
`/ when it holds for all the composition factors of the group (and `).
(b) If G is `-solvable, then the Hasse principle holds for G and `:
(c) A solvable group satisfies the Hasse principle.
P ROOF. Part (a) follows by induction from the lemma. Part (c) follows from
(a) and the obvious fact that the Hasse principle holds for a cyclic group. Part
(b) follows from (a) and (c) and the additional fact that the higher cohomology
groups of a module killed by ` relative to a group of order prime to ` are all zero.2
We now prove Theorem 9.2. Lemma 9.3 shows that
Ker ˇS1 .K; M / D Ker ˇS1 .K.M /=K; M / H1 .GK.M /=K ; M /;
and (9.5b) shows that this last group is zero, which proves part (a) of the theorem.
From (4.10) we know that Ker.ˇ 2 .K; M // is dual to Ker.ˇ 1 .K; M D //. Clearly
M D is simple if M is, and the extension K.M D / is `-solvable if K.M / is be-
cause it is contained in K.M /.`/. Therefore part (b) of the theorem follows
from part (a).
C OROLLARY 9.6 If M Z=`Z Z=`Z (as an abelian group) for some prime `
not equal to the characteristic of K , then X2 .K; M / D 0:
P ROOF. If M is simple (or semisimple) as a GK -module, this follows directly

from the theorem. The remaining case can be proved directly. 2
N OTES The groups Hr .G; M / were introduced by Tate (see Serre 1964/71). Theorem
9.2 and its proof are taken from Jannsen 1982. For an elementary proof of (9.6), see
Cassels 1962, 5.
The Hasse principle for algebraic groups

In this subsection, G will be a connected (not necessarily commutative) linear
algebraic group over a number field K. We say that G satisfies the Hasse principle
Y
H 1 .K; G/ ! H 1 .Kv ; G/
all v
if is injective. It is known (Kneser 1966, 1969; Harder 1965/66) that if G is semi-

simple and simply-connected without26 factors of type E8 , then H 1 .Kv ; G/ D 0
Q
for all nonarchimedean v, and H 1 .K; G/ ! v real H 1 .Kv ; G/:
26 In 1989, Chernousov removed this condition.
T HEOREM 9.7 Let G be a simply connected semisimple group, and let 'W G !
G 0 be a separable isogeny. Let M be the kernel of '.K s /W G.K s / ! G 0 .K s /,
and assume that X2 .K; M / D 0. If the Hasse principle holds for G , then it also
holds for G 0 :
We first prove a lemma.
L EMMA 9.8 Let M be a finite module GK whose order is not divisible by char.K/,
and assume S omits only finitely manyQ primes of K .
(a) The cokernel of H 1 .K; M / ! v…S H 1 .Kv ; M / is canonically isomor-
phic to the dual of .Ker ˇS1 .K; M D //=.Ker ˇ 1 .K; M D //.
(b) If each v … S has a cyclic decomposition group in K.M /, then
M
H 1 .GK ; M / ! H 1 .Kv ; M /
v…S
is surjective. In particular
M
H 1 .K; M / ! H 1 .Kv ; M /
v real
is surjective.
P ROOF. (a) From (4.10) we know there is an exact sequence
ˇ1 Y 1
H 1 .K; M D / ! PS1 .K; M D / H 1 .Kv ; M D / ! H 1 .K; M / :
v…S
Therefore the kernel-cokernel sequence (0.24) of the pair of maps

Y
H 1 .K; M D / ! PS1 .K; M D / H 1 .Kv ; M D / ! PS1 .K; M D /
v…S
Y
0! Ker ˇ 1 .K; M D / ! Ker ˇS1 .K; M D / ! H 1 .Kv ; M D / ! H 1 .K; M / :
v…S
Q / D 1Ker. /, but this

The exactness at the third term says that Ker.ˇS1 /= Ker.ˇ 1
last group is the dual of the cokernel of H .K; M / ! v…S H .Kv ; M /:

1
(b) Let F be a finite Galois extension of K containing K.M D /. Accord-

ing to the Chebotarev density theorem, for each prime v … S having a cyclic
decomposition group in Gal.F=K/, there is a prime v 0 2 S having the same

decomposition group. Therefore if an element c of H 1 .GF =K ; M D / maps to
zero in H 1 .GFw =Kv ; M D / for all v in S, then it maps to zero for all v. Hence
Ker ˇS1 .F=K; M / D Ker ˇ 1 .F=K; M /, and Lemma 9.3 shows that this implies
that Ker ˇS1 .K; M / D Ker ˇ 1 .K; M /. Now (a) implies (b). 2
P ROOF ( OF T HEOREM 9.7) Consider the diagram of pointed sets:
H 1 .K; M / ! H 1 .K; G/ ! H 1 .K; G 0 / ! H 2 .K; M / !

? ? ? ?
? ?injective ? ?injective
y y y y
Q 1 Q 1 Q 1 Q
H .Kv ; M / !
H .Kv ; G/ ! H .Kv ; G 0 / !
H 2 .Kv ; M / !:
all v all v all v all v
(See Serre 1961, VII, Annexe.) If c 2 H 1 .K; G 0 / maps to zero in H 1 .Kv ; G 0 /

for all v, then it lifts to an element b 2 H 1 .K; G/. As we observed above,
H 1 .Kv ; G/ D 0 for all nonarchimedean v. For each archimedean prime v, the
image bv of b in H 1 .Kv ; G/ lifts to an element av of H 1 .Kv ; M /. According to
(9.8), there is an element a 2 H 1 .K; M / mapping to av for all archimedean v.
Now b a0 , where a0 is the image of a in H 1 .K; G/, maps to c in H 1 .K; G 0 /
and to 0 in H 1 .Kv ; G/ for all v. The last condition shows
Q that b a0 (hence c/ is
zero. This shows that the kernel of H 1 .K; G 0 / ! H 1 .Kv ; G/ is zero, and a
standard twisting argument (cf. Kneser 1969, I 1.4) now allows one to show that
the map is injective. 2
C OROLLARY 9.9 Let G be a semisimple algebraic group over K without factors

of type E8 . Then the Hasse principle holds for G under each of the following
hypotheses:
(a) G has trivial centre;
(b) G is almost absolutely simple;
(c) G is split by a finite Galois extension F of K such that all Sylow subgroups
of Gal.F=K/ are cyclic;
(d) G is an inner form of a group satisfying (a), (b), or (c).
P ROOF. A group with trivial centre is a product of groups of the form RF =K G

with G an absolutely simple group over F , and so (a) follows from (b). An
absolutely almost simple group is an inner form of a quasi-split almost simple
group, and such a group is split by a extension whose Galois group is a subgroup
of the group of automorphisms of its Dynkin diagram. But this automorphism
group is either trivial or is Z=2Z or S3 . Therefore (b) follows from (c) and (d).
Q s/ !
Let G be split by an extension F as in (c), and let M be the kernel of G.K
G.K s / where GQ is the universal covering group of G. Then M is a sum of

Gal.Fs =F / modules of the form m , and so Gal.Fs =F / acts trivially on M D .
Therefore
X1 .K; M D / D Ker ˇ 1 .F=K; M D / H1 .Gal.F=K/; M D / D 0;
and so X2 .K; M / D 0. Finally (d) is obvious from the fact that the Gal.K s =K/-
module M is unchanged when G is replaced by an inner form. 2
N OTES Theorem 9.7 is proved in Harder 1967/68, Theorem 4.3.2, and in Kneser 1969,
pp77-78. Part (a) of Corollary 9.9 is proved in Langlands 1983, VII 6.27 All of the results
in this subsection are contained in Sansuc 1981.
Forms of an algebraic group

The next result shows that (under certain conditions) a family of local forms of
an algebraic group arises from a global form.
T HEOREM 9.10 Let K be an algebraic number field, S a finite set of primes of

K , and G an absolutely almost simple algebraic group over K that is either simply
connected or has trivial centre. Then the canonical map
Q
H 1 .K; Aut.G// ! v2S H
1
.Kv ; Aut.G//
is surjective.
P ROOF ( SKETCH ) Let GQ be the universal covering group of G, and let M D

Q s / ! G.K s //. Consider the diagram
Ker.G.K
! Q
H 1 .K; G/ ! H 1 .K; G/ ! H 2 .K; M / !
? ? ?
? ? ?
y y y
L 1 L 1 L 2
! H .Kv ; G/Q ! H .Kv ; G/ ! H .Kv ; M / !
v2S v2S v2S
Corollary 4.16 shows that the final vertical map is surjective. We have already
noted that the first vertical map is surjective when G has no factors of type
E8 , but in fact this condition is unnecessary. Next one shows that the map
H 1 .K; G/ ! H L
2 .K; M / is surjective, and a diagram chase then shows that
H 1 .K; G/ ! 1
v2S H .Kv ; G/ is surjective. One shows that it suffices to
27 See also Satz 4.3.2 of: Harder, Günter, Bericht über neuere Resultate der Galoiskohomologie
halbeinfacher Gruppen. Jber. Deutsch. Math.-Verein. 70 1967/1968 Heft 4, Abt. 1, 182–216.

prove the theorem for a split G, in which case Aut.G/ is the semi-direct product
G Ì Aut.D/ of G with the automorphism group of the Dynkin diagram of G.
The proof of the theorem then is completed by showing that H 1 .K; Aut.D// !
L
H 1 .Kv ; Aut.D// is surjective. 2
For the details, see Borel and Harder 1978, where the theorem is used to prove
the existence of discrete cocompact subgroups in the groups of rational points of
reductive groups over nonarchimedean local fields of characteristic zero.28
The Tamagawa numbers of tori

We refer the reader to Weil 1961 for the definition of the Tamagawa number .G/
of a linear algebraic group G over a global field.
T HEOREM 9.11 For any torus T over a global field K
ŒH 1 .K; T _ /
.G/ D
ŒX1 .K; T /
Q
where X1 .K; T / is the kernel of H 1 .K; T / ! all v H 1 .Kv ; T / and T _ is the
dual torus defined by the relation X .T / D X .T _ /:
P ROOF. Let
ŒX1 .K; T /
'.T / D .T / :
ŒH 1 .K; T _ /
The proof has three main steps:
(i) ' is an additive function on the category of tori over KI
(ii) '.Gm / D 1;
(iii) for any finite separable extension F of K, '.ResF =K T / D '.T /:
Once these fact have been established the proof is completed as follows. The
functor T 7! X .T / defines an equivalence between the category of tori over
28 In
the same paper (Theorem 1.7), Borel and Harder prove that, for a connected semisimple
group G over a number field K, the canonical map
M
H 1 .K; G/ ! H 1 .Kv ; G/
v2S
is surjective for any finite set S of primes. However, as Prasad and Rapinchuk have noted, their
proof only uses that S omits at least one nonarchimedean prime. Thus, they in fact prove that
M
H 1 .K; G/ ! H 1 .Kv ; G/
v¤v0
is surjective for any nonarchimedean v0 .

K and the category RepZ .GK / of continuous representations of GK on free Z-

modules of finite rank, and so we can regard ' as being defined on the latter
category. Then (i) says that ' induces a homomorphism K0 .RepZ .GK // !
Q>0 . A theorem (Swan 1960) shows that ŒX ŒX 0 is a torsion element of
K0 .RepZ .GK // if X ˝ Q D X 0 ˝ Q, and, as Q>0 is torsion-free, ' is zero on
torsion elements of K0 .RepZ .GK //. Therefore ' takes equal values on isogenous
tori. Artin’s theorem on characters (Serre 1967b, 9.2) implies that for any torus
T , there exists and
Q integer m and finite separable Q extensions Fi and Ej of K
such that T m ResFi =K Gm is isogenous to ResEj =K Gm . Now (ii) and
(iii) show that '.T /m D 1, and therefore '.T / D 1.
Statements (ii) and (iii) are easily proved (they follow almost directly from
the definitions), and so the main point of the proof is (i). This is proved by an
argument, not dissimilar to that used to prove Theorem 7.3, involving the duality
theorems. See Ono 1961, 1963, and also Oesterlé 1984, which corrects errors in
Ono’s treatment of the function field case. 2
The central embedding problem

Let S be a finite set of primes of a global field K, and let GS be the Galois group
over K of the maximal extension of K unramified outside S. Let
1!M !E!G!1
be an extension of finite groups with M in the centre of E, and let 'W GS G

be a surjective homomorphism. The embedding problem for E and ' is the prob-
lem of finding a surjective homomorphism ˚ W GS E lifting '. Concretely
this means the following: the homomorphism ' realizes G as the Galois group
of an extension F of K that is unramified outside S, and the embedding problem
asks for a field F 0 that is Galois over K with Galois group E, is also unramified
outside S, contains F , and is such that the map E G induced by the inclusion
of F into F 0 is that in the sequence. For each v in S, let Gv be the image of
Gal.Kvs =K/ in G, and let Ev be the inverse image of Gv in E. Then the (lo-
cal) embedding problem asks for a homomorphism Gal.Kvs =Kv / Ev lifting
Gal.Kvs =Kv / Gv :
Let be the class of the extension in H 2 .G; M /. If the embedding prob-
lem has a solution, then clearly is sent to zero by the map H 2 .G; M / !
H 2 .GS ; M / defined by '. The converse is also true if ¤ 0 and M is a simple
G-module. Thus Theorem 9.2 has the following consequence.
1!M !E!G!1
be a nonsplit central extension of finite groups, and let 'W GK G be a sur-

jective homomorphism. If M is a simple G -module and G is solvable, then the
embedding problem for E and ' has a solution if and only if the corresponding
local problem has a solution for all v .
P ROOF. The necessity of the condition is obvious. For the sufficiency, note that
when the local problem has a solution, the image of in H 2 .Kv ; M / is zero for
all v. According to (9.2), this implies that is zero. 2
Unfortunately, the proposition does not lead to a proof of Shafarevich’s the-

orem (Shafarevich 1954): for any number field K and finite solvable group G,
there exists an extension F of K with Galois group G.
For other applications of the duality theorems to the embedding problem, see
for example Haberland 1978, Neumann 1977, and Klingen 1983.
Abelian varieties defined over their fields of moduli

Let A be a polarized abelian variety defined over Q. The obstruction to A having
a model over its field of moduli is a class in H 2 .G; Aut.A//. In the case that
Aut.A/ is abelian, the duality theorems can sometimes be helpful in studying this
element.
Abelian varieties and Zp -extensions

The duality theorems (and their generalizations to flat cohomology) have been
used in the study of the behaviour of the points on an abelian variety as one
progresses up a Zp -tower of number fields. See for example Manin 1971, Mazur
1972, Harris 197929 , and Rubin 1985.
Appendix A: Class field theory for function fields

Most of the accounts of class field theory either omit the case of a function field
or make it appear harder than the number field case. In fact it is easier, at least for
those knowing a little algebraic geometry. In this appendix we derive the main
results of class field theory (except for the existence theorem) for a function field
over a finite field. As a preliminary, we derive class field theory for a Henselian
local field with quasi-finite residue field. We also investigate to what extent the
global results hold for a function field over a quasi-finite field.
29 See also: Harris, Michael, Correction to “p-adic representations arising from descent on
abelian varieties” [Compositio Math. 39 (1979), 177–245]. Compositio Math. 121 (2000), no. 1,
105–108.
APPENDIX A: CLASS FIELD THEORY FOR FUNCTION FIELDS 127
Local class field theory

A field k is quasi-finite if it is perfect and if the Galois group G.k s =k/ is iso-
morphic to the profinite completion b Z of Z. The main examples of quasi-finite
fields are the finite fields and the power series fields k0 ..t// with k0 an alge-
braically closed field of characteristic zero, but there are others. For example, any
algebraic extension k 0 of a quasi-finite field k whose degree Œk 0 W k is divisible
by only a finite power of each prime number is quasifinite. Also, given an alge-
braically closed field K, one can always find a quasi-finite field k having K as its
algebraic closure (Serre 1962, XIII 2, Ex 3).
Whenever a quasi-finite field k is given, we shall always assume that there is
also given as part of its structure a generator k of Gal.k s =k/, or equivalently,
a fixed isomorphism 'k W .Gk / ! Q=Z where .Gk / is the character group
Homcts .Gk ; Q=Z/ of Gk . The relation between and 'k is that 'k ./ D . /
for all 2 .Gk /. A finite extension ` of a quasi-finite field k is again quasi-
Œ`Wk
finite with generator ` D k . When k is finite, we always take k to be the
Frobenius automorphism a 7! aq , q D Œk:
Note that the Brauer group of a quasi-finite field is zero, because Gk has
cohomological dimension one and k s is divisible. Let R be a discrete valuation
ring with residue field k D R=m. Write f for the reduction of an element of R
or RŒX modulo m. We say that R is Henselian if it satisfies the conclusion of
Hensel’s lemma: whenever f is a monic polynomial with coefficients in R such
that f factors as f D g0 h0 with g0 and h0 monic and relatively prime, then f
itself factors as f D gh with g and h monic and such that g D g0 and h D h0 .
Hensel’s lemma says that complete discrete valuation rings are Henselian, but not
all Henselian rings are complete. For example, let v be a prime in a global field
K, and let Ov be the ring of elements of K that are integral at v. Choose an
extension w of v to K s , let K dec be the decomposition field of w in K s , and let
Ovh be the ring elements of K dec that are integral with respect to w. Alternatively,
choose an embedding of K s into Kvs , and let Ovh D K s \ O b v . Then Ovh is a
Henselian local ring, called the Henselization of Ov . See, for example, Milne
1980, I 4.
Now let R be a Henselian discrete valuation ring with quasi-finite residue
field k, and write K for its field of fractions. Many results usually stated only
for complete discrete valuation rings hold in fact for Henselian discrete valua-
tion rings (often the proof uses only that the ring satisfies Hensel’s lemma). For
example, the valuation v on K has a unique extension to a valuation (which we
shall also write v/ on K s . As usual we write K un for the maximal unramified
subextension of K s over K, and Run for the integral closure of R in K un .
P ROPOSITION A.1 There is a canonical isomorphism invK W Br.K/ ! Q=Z:
We first show that Br.K un / D 0. Let D be a skew field of degree n2 over

K un . Because Run is also a Henselian discrete valuation ring, the valuation on
K un has a unique extension to each commutative subfield of D, and therefore it
has a unique extension to D. The usual argument in the commutative case shows
that, for this extension, n2 D ef . Let ˛ in D have value 1=e; then K un Œ˛, being
a commutative subfield of D, has degree at most n, and so e n. On the other
hand f D 1 because the residue field of R is algebraically closed, and it follows
that n D 1:
The exact sequence
0 ! H 2 .Gal.K un =K/; K un / ! Br.K/ ! Br.K un /
now shows that Br.K/ D H 2 .Gal.K un =K/; K un /.

ord
L EMMA A.2 The map H 2 .Gal.K un =K/; K un / ! H 2 .Gal.K un =K/; Z/ is an
isomorphism.
P ROOF. As
ord
0 ! Run ! K un ! Z ! 0
is split as a sequence of Gal.K un =K/-modules, the map in question is surjective.
Let c lie in its kernel, and let be a cocycle representing c. Associated with c
there is a central simple algebra B over K (Herstein 1968, 4.4), and if is chosen
to take values in R , then the same construction that gives B gives an Azumaya
algebra B0 over R that is an order in B. The reduction B0 ˝R k of B0 is a
central simple algebra over k, and therefore is isomorphic to a matrix algebra. An
elementary argument (Milne 1980, IV 1.6) shows then that B0 is also isomorphic
to a matrix algebra, and this implies that c D 0: 2
We define invK to be the unique map making

invK
H 2 .Gal.K un =K/; Z/ D Br.K/ ! Q=Z
x x
?
'?d
?
'?'k
H 1 .Gal.K un =K/; Q=Z/ .Gk /

commute. It is an isomorphism. If F is a finite separable extension of K, then the
integral closure RF of R in F is again a Henselian discrete valuation ring with
quasi-finite residue field, and one checks easily from the definitions that
invF .Res.a// D ŒF W K invK .a/; a 2 Br.K/:

Therefore .GK ; K s / is a class formation in the sense of 1.

We identify the cup-product pairing
H 0 .GK ; K s / H 2 .GK ; Z/ ! H 2 .GK ; K s /
with a pairing
h ; iW K .GK / ! Br.K/:
T HEOREM A.3 (Local reciprocity law). There is a continuous homomorphism
.; K/W K ! Gal.K ab =K/
such that
(a) for each finite abelian extension F K s of K , .; K/ induces an isomor-
phism
.; F=K/W K =NF =K F ! Gal.F=K/I
(b) for any 2 .G/ and a 2 K , .a; K/ D invK ha; i:
P ROOF. As is explained in 1, this theorem is a formal consequence of the fact

that .GK ; K s / is a class formation. 2
C OROLLARY A.4 Let F1 and F2 be extensions of K such that F1 \ F2 D K ,

and let F D F1 F2 . If all three fields are finite abelian extensions of K , then
NF D NF1 \ NF2 and .NF1 /.NF2 / D K :
P ROOF. According the theorem, a 2 NF if and only if .a; K/ acts trivially

on F ; this is equivalent to .a; K/ acting trivially on F1 and F2 , or to it lying in
NF1 \ NF2 . The second equality can be proved similarly. 2
R EMARK A.5 When F=K is unramified, lifts to a unique (Frobenius) element

Q in Gal.F=K/, and .; K/W K ! Gal.F=K/ sends an element a of K to
Q ord.a/ . In particular, .a; F=K/ D 1 if a 2 R . When F=K is ramified, the
description of .a; F=K/ is much more difficult (see Serre 1967a, 3.4).
We say that a subgroup N of K is a norm group if there exists a finite

abelian extension F of K such that N D NF =K F . (The name is justified by the
following result: if F=K is any finite separable extension of K, then NF =K F D
NL=K L where L is the maximal abelian subextension of F ; see Serre 1962, XI
4.)
R EMARK A.6 The reciprocity map defines an isomorphism lim K =N ! G ab

(inverse limit over the norm groups in K /, and so to understand G ab fully it is
necessary to determine the norm groups. This is what the existence theorem does.
C ASE 1: K is complete and k is finite. This is the classical case. Here the
norm groups of K are precisely the open subgroups of finite index. Every sub-
group of finite index prime to char.K/ is open. The image of the reciprocity map
is the subgroup of G ab of elements that act as an integral power of the Frobenius
automorphism on k s . The reciprocity map is injective, and it defines an isomor-
phism of the topological group R onto the inertia subgroup of G ab . See Serre
1962, XIV 6.
C ASE 2: K is Henselian with finite residue field. We assume that R is excel-
lent. This is equivalent to the completion K b of K being separable over K. From
the description of it given above, it is clear that the Henselization of the local ring
at a prime in a global field is excellent. Under this assumption:
b is of the form F
(i) every finite separable extension of K b for a finite separable
b
extension F of K with ŒF W K D ŒF W KI b
b and hence K
(ii) K is algebraically closed in K, b is linearly disjoint from Ka
over K:
To prove (i), write Fb D KŒ˛,
b and let F D KŒˇ with ˇ a root of a polynomial
in KŒX that is close to the minimal polynomial of ˛ over K (cf. Lang 1970, II
2). To prove (ii), note that if K is not algebraically closed in K, b then there is an
b
element ˛ of K that is integral over R but which does not lie in R. Let f .X /
be the minimal polynomial of ˛ over K. As ˛ is integral over R, b it lies in R,
b
b
and so f has a root in R. An approximation theorem (Greenberg 1966) now says
that f has a root in R, but f was chosen to be irreducible over K. Thus K is
algebraically closed in K, b and combined with the separability of K b over K, this
b a
implies that K is linearly disjoint from K (Lang 1958, III 1, Thm 2).
On combining these two assertions, we find that F 7! F b defines a degree-
preserving bijection from the set of finite separable extensions F of K to the set
of similar extensions of K. b Moreover, NF D N F b \ K for each F because
NF is dense in N F b and Greenberg’s theorem implies that NF is open in K .
It follows that the norm groups of K are again precisely the open subgroups of
finite index.
C ASE 3: K is complete with quasi-finite residue field. In this case every
subgroup of K of finite index prime to char.k/ is a norm group, but not every
open subgroup of index a power of the characteristic of k is. In Whaples 1952–
54, various characterizations of the norm groups are given using the pro-algebraic
structure on K :
C ASE 4: K is Henselian with quasi-finite residue field. We leave this case to
the reader to investigate.
Global class field theory

Let X be a complete smooth curve over a quasi-finite field .k; /, and let K D
k.X /. The set of closed points of X will be denoted by30 X 0 (thus X 0 omits
only the generic point of X ). To each point v of X 0 , there corresponds a val-
uation (also written v/ of K, and we write Kv for the completion of K with
respect to v and Rv for the ring of integers in Kv . The residue field k.v/ is a
quasi-finite field of degree deg.v/ over k with deg.v/ as the chosen generator of
Gal.k.v/s =k.v//. Write av for the image of P an element a of Br.K/ in Br.Kv /,
and define invK W Br.K/ ! Q=Z to be a 7! v invv .av / where invv is invKv .
Let X Ddf X ˝k k s be X regarded as curve over k s , and let K D k s .X/ be its
function field. We write JacX be the Jacobian variety of X:
T HEOREM A.7 There is an exact sequence

M invK
0 ! H 1 .Gk ; JacX .k s // ! Br.K/ ! Br.Kv / ! Q=Z ! 0:
v2X 0
P ROOF. We use the exact sequence of Gk -modules

0 ! k s ! K ! Div.X / ! Pic.X/ ! 0
L
where Div.X/ D v2X 0 Z is the group of (Weil) divisors on X . From the
cohomology sequence of its truncation

0 ! k s ! K ! Q ! 0
'
we obtain an isomorphism H 2 .Gk ; K / ! H 2 .Gk ; Q/. Note that

H 2 .Gk ; K / D Ker.Br.K/ ! Br.K//;
and recall that Tsen’s theorem (Shatz 1972, Theorem 24) states that Br.K/ D 0,

and so H 2 .Gk ; Q/ D H 2 .Gk ; K / D Br.K/:
The cohomology sequence of the remaining segment
0 ! Q ! Div.X/ ! Pic.X/ ! 0
30 Ithas been suggested to me that, since many authors write Xi (resp. X i ) for the set of
(schematic) points of dimension (resp. codimension) i, X0 would be a better notation for the set of
closed points.
of the sequence is
H 1 .Gk ; Div.X // ! H 1 .Gk ; Pic.X// ! H 2 .Gk ; Q/

! H 2 .Gk ; Div.X // ! H 2 .Gk ; Pic.X// ! 0:
L L
But H r .Gk ; Div.X // D v2X 0 H .Gk ; Dv /, where Dv Ddf
r
w7!v Z is the
Gk -module induced by the trivial Gk.v/ -module Z, and so
L
H r .Gk ; Div.X // D v2X 0 H r .Gal.k s =k.v//; Z/:
L
In particular, H 1 .Gk ; Div.X // D 0 and H 2 .Gk ; Div.X // D v2X 0 .Gk.v/ /:
Almost by definition of JacX , there is an exact sequence
0 ! JacX .k s / ! Pic.X / ! Z ! 0:
As JacX .k s / is divisible (Milne 1986b, 8.2) and k has cohomological dimension

one, H 2 .Gk ; JacX .k s // D 0, and so H 2 .Gk ; Pic.X// D H 2 .Gk ; Z/ D .Gk /.
These results allow us to identify the next sequence with the required one:
H 1 .Gk ; Pic.X// !
! H 2 .Gk ; Q/ !
H 2 .Gk ; Div.X // !
H 2 .Gk ; Pic.X // !
0

P
L
0 !
H 1 .Gk ; JacX / !
Br.K/ !
.Gk.v/ / ! .Gk / !
0:
L
The map ˙ can be identified with invK W v Br.Kv / ! Q=Z: 2
Q
We define the group of idèles JK of K to be the subgroup of v2X 0 Kv
comprising those elements a D .av / such that av 2 Rv for all but finitely many
v. The quotient of JK by K (embedded diagonally) is the idèle class group of
CK of K. We set J D lim JF and C D lim CF (limit over all finite extension F
! !
of K, F K s ).
C OROLLARY A.8 If H 1 .Gk0 ; JacX .k s // D 0 for all finite extensions k 0 of k ,

then it is possible to define on .GK ; C / a natural structure of a class formation.
P ROOF. An argument similar to that in (4.13) shows that

L
H r .GK ; J / D v2X 0 H r .Gv ; Kvs /; r 1;
where for each v in X 0 a choice w of an extension of v to K s has been made in

order to identify K with a subfield of Kvs and Gv Ddf GKv with a decomposition
group in G. Consider the diagram
L invK
0 !
Br.K/ !
v2X 0 Br.Kv / ! Q=Z !
0

L
H 1 .GK ; C / !
0 ! H 2 .GK ; K s / !
v2X 0 H 2 .Gv ; Kvs / !
H 2 .GK ; C / !
0
whose top row is the sequence in (A.7) and whose bottom row is the cohomology
sequence of
0 ! K s ! J ! C ! 0:
The zero at lower left comes from Hilbert’s theorem 90, and the zero at lower right
comes from the fact that GK has cohomological dimension cd.k/ C 1 D 2.
This diagram shows that H 1 .GK ; C / D 0 and that there is a unique isomorphism

invK W H 2 .GK ; C / ! Q=Z making
invK
H 2 .GK ; C / ! Q=Z
x
?
?
invv
H 2 .Gv ; Kvs / ! Q=Z
commute for all v. The same assertions are true for any finite separable extension
F of K, and it obvious that the maps invF satisfy the conditions (1.1). 2
C OROLLARY A.9 If k is algebraic over a finite field, then .GK ; C / is a class

formation.
P ROOF. Lang’s lemma shows that H 1 .Gk ; A/ D 0 for any connected algebraic
group A if k is finite. If k is algebraic over a finite field, then any element of
H 1 .Gk ; A/ is represented by a principal homogeneous space, which is defined
over a finite field and is consequently trivial by what we have just observed. 2
In (A.14) below, we shall see examples of fields K=k for which the conditions
of (A.8) fail. We now investigate how much of class field theory continues to hold
in such cases.
Fix an extension of each v to K s , and hence embeddings iv W Gv ! GK .
Define
.; K/W JK ! Gal.K ab =K/
by Q
.a; K/ D iv .av ; Kv /; a D .av /:
v2X 0
For any finite abelian extension F of K, this induces a mapping
.; F=K/W JK =NF =K JF ! Gal.F=K/

Q
such that .a; F=K/ D iv .av ; Fw =Kv / where Fw denotes the completion of F
at the chosen prime lying over v. It follows from (A.5) above, and the fact that
only finitely many primes of K ramify in F , that this last product is finite (and
that the previous product converges).
L EMMA A.10 For all a in K , .a; K/ D 0:
P ROOF. Consider the diagram:
/
L invK
/ Q=Z
JK .G/ Br.Kv /
O
H 0 .GO K ; J / H 2 .GK ; Z/ / H 2 .G ; J /
OK
H 0 .GK ; K s / H 2 .GK ; Z/ / H 2 .GK ; K s /
The two lower P pairings are defined by cup-product, and the top pairing sends
.a; / to invK . hav ; jGv i/ D ..a; K//, a D .av / (here h ; i is as in A.3).
It is obvious from the various definitions that the maps are compatible with the
pairings. If a 2 K , then the diagram shows that .a; K/ lies in the image of
Br.K/ in Q=Z, but Br.K/ is the kernel of invK , and so .a; K/ D 0 for all .
This implies that .a; K/ D 0: 2
The lemma shows that there exist maps
.; K/WCK ! Gal.K ab =K/

.; F=K/WCK =NCF ! Gal.F=K/; F=K finite abelian:
We shall say that the reciprocity law holds for K=k if, for all finite abelian exten-
sions F=K, this last map is an isomorphism.
Unfortunately the reciprocity law does not always hold because there can
exist abelian extensions F=K in which all primes of K split, that is, such that
Fw D Kv for all primes v. This suggests the following definition: let F be a
finite abelian extension of K, and let K 0 be the maximal subfield of F containing
K and such that all primes of K split in K 0 ; the reduced Galois group G F =K of
F over K is the subgroup Gal.F=K 0 / of Gal.F=K/:
P ROPOSITION A.11 For any finite abelian extension F of K , the map .; F=K/
'
induces an isomorphism CK =NCF ! G F =K :
P ROOF. For each prime v, the image of Gal.Fw =Kv / is contained in G F =K , and
so the image of J in Gal.F=K/ is also contained in G F =K :
It clearly suffices to prove the surjectivity of .; F=K/ in the case that F=K 0
is cyclic of prime order. Then there exists a prime v such that Fw ¤ Kv , and
Kv ! Gal.Fw =Kv / G F =K is surjective by local class field theory.
To prove the injectivity, we count. If F1 and F2 are finite abelian extensions
of F such that F1 \ F2 D K and F1 F2 D F , then it follows from (A.4) that
NCF1 \ NCF2 D NCF and .NCF1 /.NCF2 / D NCF . As G F1 =K \ G F2 =K D 1
and G F1 =K G F2 =K D G F =K , it suffices to prove that CK =NCF and G F =K have
the same order for F=K cyclic of prime power order.
Let F=K be cyclic of prime power, and consider the diagram
L invK
Br.F=K/ ! Br.Fw =Kv / ! Q=Z
v2X 0
x x
? ?
? ?
K =NF ! JK =NJF ! CK =NCF ! 0:
The top row is part of the sequence in (A.7), and the bottom row is part of the
Tate cohomology sequence of
0 ! F ! JF ! CF ! 0:
The first two vertical arrows are the isomorphisms given by the periodicity of the
cohomology of cyclic groups. The order of the image of invK is the maximum of
the orders of the Br.Fw =Kv /, and the order of Br.Fw =Kv / is ŒFw W Kv . Thus
the order of the image equals the order of G.L=K/. From the diagram, we see
that it is also the order of CK =NCF : 2
For any curve Y over a quasi-finite

L field, we define the Brauer group Br.Y /
of Y to be the kernel of Br.F / ! v2Y 0 Br.Fv /, where F is the function field
of Y . This definition will be justified in A.15 below. Note that Theorem A.7
shows that Br.Y / D H 1 .Gk ; JacY .k s //:
P ROPOSITION A.12 The following statements are equivalent:
(a) the reciprocity law holds for K=kI
(b) for all finite cyclic extensions F=K , the sequence
L
Br.F=K/ ! v2X 0 Br.Fw =Kv / ! ŒF W K1 Z=Z ! 0
is exact;
(c) for all finite cyclic extensions F=K , H 1 .Gal.F=K/; Br.Y // D 0, where
Br.Y / is the Brauer group of the projective smooth curve with function
field F:
P ROOF. It follows from (A.11) that the reciprocity law holds for K=k if and
only if G F =K D GF =K for all finite abelian extensions F=K, and it suffices to
check this for cyclic extensions. But, as we saw in the above proof, for such
an extension the order of G.F=K/ is the order of the cokernel of Br.F=K/ !
L
Br.Fw =Kv /. The equivalence of (a) and (b) is now clear.
Consider the exact commutative diagram
L
Br.X / ! Br.K/ !
Br.Kw / !
Q=Z ! 0
? ? ? ?
? ?˛ ? ?n
y y y y
L
0 !
.Br.F /= Br.Y //Gal.F =K/ ! . Br.Fw //Gal.Fw =Kv / !
Q=Z
where n D ŒF W K. From the Hochschild-Serre spectral sequences, we get exact

sequences
0 ! Br.F=K/ ! Br.K/ ! Br.F /Gal.F =K/ ! H 3 .Gal.F=K/; F /

0 ! Br.Fw =Kv / ! Br.Kv / ! Br.Fw /Gal.Fw =Kv / ! H 3 .Gal.Fw =Kv /; Fw /;
and from the periodicity of the cohomology of cyclic groups, we see that
H 3 .Gal.F=K/; F / D H 1 .Gal.F=K/; F / D 0;
H 3 .Gal.Fw =Kv /; Fv / D H 1 .Gal.Fw =Kv /; Fw / D 0:
Thus the preceding diagram gives an exact sequence of kernels and cokernels,
L
Br.F=K/ ! Br.Fw =Kv / ! n1 Z=Z ! Coker.˛/ ! 0:
But, as Br.K/ ! Br.F /Gal.F =K/ is surjective,

Coker.˛/ D Coker Br.F /Gal.F =K/ ! .Br.F /= Br.Y //Gal.F =K/ ;
which equals H 1 .Gal.F=K/; Br.Y // because H 1 .Gal.F=K/; Br.F // D 0 (look

at the Hochschild-Serre spectral sequence). Thus (b) is equivalent to (c). 2
L We say that the Hasse principle holds for K=k if the map K =NF !
Kv =NFw is injective for all finite cyclic field extensions F of K:
P ROPOSITION A.13 The following are equivalent:

(a) the Hasse principle holds for KI

(b) H 1 .Gk ; JacX .k s // D 0I
(c) Br.X / D 0:
In particular, the Hasse principle holds for K=k if k is algebraic over a finite field.
P ROOF. As K =NF Br.F=K/ for F=K finite and cyclic, Lwe see that that
the Hasse principle holds for K=k if and only if Br.F=K/ ! S w =Kv / is
Br.F
injective for all F=K finite and cyclic. As Br.K/ D 0, Br.K/ D Br.F=K/
where the union runs over all finite cyclic
L extensions. Thus the Hasse principle
holds for K if and only if Br.K/ ! Br.Kv / is injective, but the kernel of this
map is Br.X / D H 1 .Gk ; JacX .k s //: 2
R EMARK A.14 Let k0 be an algebraically closed field of characteristic zero, and

let k be the quasi-finite field k0 ..t//. In this case there exist elliptic curves E over
k with H 1 .k; E/ ¤ 0, and there exist function fields K over k with finite exten-
sions F linearly disjoint from k s such that every prime of K splits completely in
F (see Rim and Whaples 1966). In lectures in 1966, Rim asked (rather pessimisti-
cally) whether the following conditions on a quasi-finite field k are equivalent:
(a) k is algebraic over a finite field;
(b) H 1 .Gk ; A/ D 0 for all connected commutative algebraic group varieties
over k;
(c) the reciprocity law holds for all K=kI
(d) the Hasse principle holds for all K=k:
We have seen that (a) H) (b) H) (c),(d), but (b) does not imply (a). In fact
Jardan has shown (1972, 1974) that if k is finitely generated over Q, then for
almost all 2 Gal.k s =k/, the fixed field k. / of is quasi-finite and has the
property that every absolutely irreducible variety over it has a rational point; thus
(b) holds for k. /:
R EMARK A.15 We use étale cohomology to show that the group Ker.Br.K/ !
˚ Br.Kv // is indeed the Brauer group of X . Let W X ! Spec.k/ be the struc-
ture morphism, and consider the exact sequence of sheaves
0 ! Gm ! g Gm ! DivX ! 0
on Xet (see Milne 1980, II 3.9; here g is the inclusion of the generic point into X
and DivX is the sheaf of Weil divisors). On applying the right derived functors of
, we get a long exact sequence of sheaves on Spec.k/et which we can regard
as Gk -modules. The sequence is

0 ! k s ! K ! Div.X/ ! Pic.X/ ! 0;
which is exactly the sequence considered in the proof of Theorem A.7. Here it
tells us that Gm D k s , R1 Gm D Pic.X /, and Rr Gm D 0 for r 2.
Therefore the Leray spectral sequence for reduces to a long exact sequence
! H r .Gk ; k s / ! H r .Xet ; Gm / ! H r1 .Gk ; Pic.X // ! :
From this we can read off that H 2 .Xet ; Gm / D H 1 .Gk ; Pic.X //, which proves
what we want because H 2 .Xet ; Gm / is equal to the Brauer group of X:
E XERCISE A.16 Investigate to what extent the results in the second section re-
main true when the fields Kv are replaced by the Henselizations of K at its primes.
N OTES Class field theory for complete fields with quasi-local residue fields was first
developed in Whaples 1952/54 (see also Serre 1962). The same theory for function fields
over quasi-finite fields was investigated in Rim and Whaples 1966.
Chapter II
Etale Cohomology
In 1 we prove a duality theorem for Z-constructible sheaves on the spectrum

of a Henselian discrete valuation ring with finite residue field. The result is ob-
tained by combining the duality theorems for modules over the Galois groups of
the finite residue field and the field of fractions. After making some preliminary
calculations in 2, we prove in 3 a generalization of the duality theorem of Artin
and Verdier to Z-constructible sheaves on the spectrum of the ring of integers in a
number field or on curves over finite fields. In the following section, the theorems
are extended to certain nonconstructible sheaves and to tori; also the relation be-
tween the duality theorems in this and the preceding chapter is examined. Section
5 treats duality theorems for abelian schemes, 6 considers singular schemes, and
in 7 the duality theorems are extended to schemes of dimension greater than one.
In this chapter, the reader is assumed to be familiar with the more elementary
parts of étale cohomology, for example, with Chapters II and III of Milne 1980.
All schemes are endowed with the étale topology.
0 Preliminaries
We begin by reviewing parts of Milne 1980. Recall that S.Xet / denotes the cate-
gory of sheaves of abelian groups on Xet (small étale site).
Cohomology with support on a closed subscheme

(Milne 1980, p73-78, p91-95)
Consider a diagram
i j
Z ,! X -U
139
140 CHAPTER II. ETALE COHOMOLOGY
in which i and j are closed and open immersions respectively, and X is the
disjoint union of i.Z/ and j.U /. There are the following functors between the
categories of sheaves:
i jŠ
i j
S.Zet / ! S.Xet / ! S.Uet /:
iŠ j
Each functor is left adjoint to the one listed below it; for example, HomZ .i F; F 0 / '
HomX .F; i F 0 /. The functors i , i , jŠ , and j are exact, and i Š and j are left
exact. The functors i , i Š , j , and j map injective sheaves to injective sheaves.
For any sheaf F on X , there is a canonical exact sequence
0 ! jŠ j F ! F ! i i F ! 0: (1)
P ROPOSITION 0.1 (a) For any sheaves F on U and F 0 on X , the restriction

map
ExtrX .jŠ F; F 0 / ! ExtrU .F; j F 0 /
is an isomorphism for r 0; in particular,
ExtrX .jŠ Z; F 0 / ' H r .U; F 0 jU /; r 0:
(b) For any sheaves F on X and F 0 on U , there is a spectral sequence
ExtrX .F; Rs j F 0 / H) ExtrCs 0

U .F jU; F /:
(c) For any sheaves F on X and F 0 on Z , there is a canonical isomorphism

'
ExtrX .F; i F 0 / ! ExtrZ .i F; F 0 /; r 0:
(d) For any sheaf F on X , there is a canonical isomorphism
HZr .X; F / ' ExtrX .i Z; F /; r 0I
consequently, for any sheaf F on Z ,
HZr .X; i F / ' H r .Z; F /; r 0:
(e) For any sheaves F on Z and F 0 on X , there is a spectral sequence
ExtrZ .F; Rs i Š F 0 / H) ExtrCs 0

X .i F; F /:
0. PRELIMINARIES 141
(f) For any sheaf F on X , there is a long exact sequence
! HZr .X; F / ! H r .X; F / ! H r .U; F / ! :
P ROOF. (a) If F 0 ! I is an injective resolution of F 0 , then j F 0 ! j I is

an injective resolution of j F 0 , because j is exact and preserves injectives. On
passing to the cohomology groups in
HomX .jŠ F; I / ' HomU .F; j I /;
we obtain isomorphisms ExtrX .jŠ F; F 0 / ' ExtrU .F; j F 0 /:

(b) As j is left exact and preserves injectives, and HomX .F; / ı j '
HomU .jŠ F; /, this is the spectral sequence of a composite of functors (Milne
1980, Appendix B, Theorem 1).
(c) The proof is the same as that of (a): as i is exact and preserves injec-
tives, on forming the derived functors of HomX .F; i .// ' HomZ .i F; /,
we obtain canonical isomorphisms ExtrX .F; i .// ' ExtrZ .i F; /.
(d) From the exact sequence (see (1))
0 ! HomX .i Z; F / ! HomX .Z; F / ! HomX .jŠ Z; F /
we see that
df
HomX .i Z; F / ' Ker. .X; F / ! .U; F // D Z .X; F /:
Hence HomX .i Z; / ' Z .X; /, and on passing to the derived functors we
obtain the required canonical isomorphism. The second statement can be obtained
by combining the first with (c):
(e) As i Š is left exact and preserves injectives, and HomZ .F; / ı i Š '
HomU .i F; /, this is the spectral sequence of a composite of functors.
(f) Form the ExtX .; F /-sequence of
0 ! jŠ Z ! Z ! i Z ! 0
(see (1)) and apply (a) and (d). 2
The exact sequence in (0.1f) is referred to as the cohomology sequence of the

pair X U:
Extensions of sheaves
We generalize Theorem 0.3 of Chapter I to the étale topology.
If Y is a Galois covering of a scheme X with Galois group G, then for
any G-module M , there is a unique locally constant sheaf FM on X such that
.Y; FM / D M (as a G-module) (Milne 1980, III 1.2). In the next theorem we
use the same letter for M and FM .
For any étale map f W U ! X , let ZU D fŠ Z. Then every sheaf on Xet is
a quotient of a direct sum of sheaves of the form ZU (cf. the second proof of
III 1.1, Milne 1980), and such a sheaf is flat. Therefore, flat resolutions exist in
S.Xet /. The condition T orrZ .M; N / D 0 for r > 0 in the next theorem means
that, for any flat resolution F ! N of N , M ˝Z F ! M ˝Z N is a resolution
of M ˝Z N .
T HEOREM 0.2 Let Y be a finite Galois covering of X with Galois group G ,

and let N and P be sheaves on Xet . Then, for any G -module M such that
T orrZ.M; N / D 0 for r > 0, there is a spectral sequence
ExtrG .M; ExtsY .N; P // H) ExtrCs

X .M ˝Z N; P /:
In particular, there is a spectral sequence
H r .G; ExtsY .N; P // H) ExtrCs

X .N; P /:
The second spectral sequence is obtained from the first by taking M D Z.

After a few preliminaries, the first will be shown to be the spectral sequence of a
composite of functors.
L EMMA 0.3 For any sheaves N and P on X and G -module M , there is a canon-
ical isomorphism
HomG .M; HomY .N; P // ' HomX .M ˝Z N; P /:
P ROOF. Almost by definition of tensor products, there is a canonical isomor-

phism
HomY .M; HomY .N; P // ' HomY .M ˝Z N; P /:
Because M becomes the constant sheaf on Y ,
HomY .M; HomY .N; P // ' Hom.M; HomY .N; P //
(homomorphisms of abelian groups). On taking G-invariants, we get the required

isomorphism. 2
L EMMA 0.4 If I is an injective sheaf on X and F is a flat sheaf on X , then

HomY .F; I / is an injective G -module.
P ROOF. We have to check that the functor HomG .; HomY .F; I //W S.Xet / !
Ab is exact, but Lemma 0.3 expresses it as the composite of two exact functors
˝Z F and HomX .; I /: 2
L EMMA 0.5 Let N and I be sheaves on X with I injective, and let M be a G -

module. If T orrZ .M; N / D 0 for r > 0, then ExtrG .M; HomY .N; I // D 0 for
r > 0:
P ROOF. By assumption, a flat resolution F ! N of N gives a resolution

M ˝Z F ! M ˝Z N of M ˝Z N , and it follows from the injectivity of
I that HomX .M ˝Z N; I / ! HomX .M ˝Z F ; I / is then a resolution of
HomX .M ˝ N; I /. In particular, on taking M D ZŒG, we get a resolution
HomX .ZŒG ˝Z N; I / ! HomX .ZŒG ˝Z F ; I / (2)
of HomX .ZŒG ˝Z N; I /. But
HomX .ZŒG ˝Z F; I / ' HomG .ZŒG; HomY .F; I // ' HomY .F; I /
for any sheaf F , and so (2) can be regarded as a resolution
HomY .N; I / ! HomY .F ; I /
of the G-module HomY .N; I /. In fact (0.4), this is an injective resolution of

HomY .N; I /, which we use to compute ExtrG .M; HomY .N; I //. From (0.3) we
know that
HomG .M; HomY .F ; I // ' HomX .M ˝Z F ; I /;
and we have already noted that this last complex is exact except at the first step.
Consequently ExtrG .M; HomY .N; I // D 0 for r > 0: 2
We now prove Theorem 0.2. Lemma 0.3 shows that
HomX .M ˝Z N; /
is the composite of the functors HomY .N; / and HomG .M; /, and Lemma 0.5
shows that the first of these sends injective objects I to objects that are acyclic for
the second functor. We therefore obtain the spectral sequence as that associated
with a composite of functors (Milne 1980, Appendix B, Theorem 1).
C OROLLARY 0.6 Let M and N be sheaves on X , and let Y be a finite Galois

covering of X . Then
ExtrY .M; N / ' ExtrX . M; N /:
P ROOF. On applying Theorem 0.2 with M D ZŒG, we find that

HomG .ZŒG; ExtrY .M; N // ' ExtrX .ZŒG ˝Z M; N /;
but the first group is ExtrY .M; N /, and, in the second, ZŒG ˝ M D M: 2
Pairings
For any sheaves M , N , and P on X , there are canonical pairings
ExtrX .N; P / ExtsX .M; N / ! ExtrCs
X .M; P /;
which can be defined in the same way as the pairings in (I 0). Also, if X is
quasi-projective over an affine scheme (as all our schemes will be), then we can
identify the cohomology groups with the Čech groups (Milne 1980, III 2.17) and
use the standard formulas (ibid. V 1.19) to define cup-product pairings
H r .X; M / H s .X; N / ! H rCs .X; M ˝ N /:
Recall also (ibid. III 1.22) that there is a spectral sequence
H r .X; ExtX
s
.M; N // H) ExtrCs
X .M; N /
whose edge morphisms are maps H r .X; HomX .M; N // ! ExtrX .M; N /. As
we noted in the proof of Lemma 0.3, a pairing M N ! P corresponds to a
map M ! HomX .N; P /:
P ROPOSITION 0.7 Let M N ! P be a pairing of sheaves on X , and consider
the composed map
H r .X; M / ! H r .X; HomX .N; P // ! ExtrX .M; N /.
Then the diagram
H r .X; M / H s .X; N / ! H rCs .X; P / cup-product pairing

# k k
ExtrX .N; P / H s .X; N / ! H rCs .X; P / Ext pairing
commutes.
P ROOF. See Milne 1980, V 1.20. 2

The Čech complex

Let F be a sheaf on Xet . For any U étale over X , let C .V =U; F / be the Cech
complex corresponding to a covering .V ! U /, and define C .F /.U / to be
lim C .V =U; F / (direct limit over the étale coverings .V ! U //. Then C .F /
!
is a complex of presheaves on X , and we let C .U; F / D .U; C .F // be the
complex of its sections over U . In the next proposition, we write Hr .F / for the
presheaf U 7! H r .U; F /:
P ROPOSITION 0.8 Assume that X is quasi-projective over an affine scheme.
(a) For any sheaf F on X , H r .C .F // ' Hr .F / and H r .C .X; F // '
H r .X; F /:
(b) For any morphism f W Y ! X , there is a canonical map f C .F / !
C .f F /, which is a quasi-isomorphism if f is étale.
(c) For any pair of sheaves F and F 0 , there is a canonical pairing C .F /
C .F 0 / ! C .F ˝ F 0 / inducing the cup-product on cohomology.
(d) Let X be the spectrum of a field K , and let F be the sheaf on X corre-
sponding to the GK -module M . Then C .X; F / is the standard resolution
of M (defined using inhomogeneous cochains).
P ROOF. (a) By definition, H r .C .V =U; F // D HL r .V =U; F /, and therefore
H r .C .F /.U // D lim H r .C .V =U; F // D lim HL r .V =U; F / D HL r .U; F /:
! !
Under our assumptions, the Čech groups agree with derived-functor groups, and
so this says that
H r . .U; C .F /// ' H r .U; F / D .U; Hr .F //;

df
which proves both the equalities.

(b) For any étale map V ! X , there is a canonical map
.V; F / ! .V.Y / ; f .F //:
In particular, when U is étale over X and .V ! U / is an étale covering of U ,
r ; f .F // (here V r denotes
then there is a canonical map .V r ; F / ! .V.Y /
V U V U :::/, all r. On passing to the limit over V , we obtain a map
.U; C r .F // ! .U.Y / ; C r .f F //
for all r, and these maps give a map of complexes C .F / ! f C .f F /. By
adjointness, we get a map f C .F / ! C .f F /. The last part of the statement
is obvious because, when f is étale,
H r .f C .F // ' Hr .F /jU ' Hr .F jU / ' H r .C .f F //:
(c) For each U ! X , the standard formulas define a pairing of complexes
.U; C .F // .U; C .F 0 // ! .U; C .F ˝ F 0 //;
and these pairings are compatible with the restriction maps.

(d) If U is a finite Galois covering of X with Galois group G, then it is
shown in Milne 1980, III 2.6, that C .U=X; F / is the standard complex for the
G-module F .U /. The result follows by passing to the limit. 2
Constructible sheaves
Let X be a scheme of Krull dimension one. A sheaf F on such a scheme is
constructible if there is a dense open subset U of X such that
(a) for some finite étale covering U 0 ! U , the restriction of F to U 0 is the
constant sheaf defined by a finite group;
(b) for all x … U , the stalk Fx of F is finite.
It is said to be Z-constructible if its restriction to some such U 0 is the constant
sheaf defined by a finitely generated group and the stalks Fx are finitely gen-
erated. Note that a constructible sheaf is Z-constructible and a Z-constructible
sheaf is constructible if and only if it is torsion.
The constructible sheaves form an abelian subcategory of S.Xet / and if
0 ! F 0 ! F ! F 00 ! 0
is exact, then F is constructible if and only if F 0 and F “ are constructible. For

a morphism locally of finite type, carries constructible sheaves to con-
structible sheaves, and when is finite, has the same property. Similar state-
ments hold for Z-constructible sheaves.
P ROPOSITION 0.9 If X is quasi-compact, then every sheaf on X is a filtered di-

rect limit of Z-constructible sheaves. Moreover, every torsion sheaf is a filtered
direct limit of constructible sheaves.
P ROOF. Let F be a sheaf on X , and consider all pairs .gW U ! X; s/ with g

étale, U affine, and s a section of F over U . For each such pair, we have a
map Z ! F jU sending 1 to s. This induces a map g Z ! F and g Z is
Z-constructible (because g is finite over a dense open subset of X ). Therefore
the image of g Z in F is Z-constructible, and it is clear that the union of all
subsheaves of this form is F . When F is torsion, then each of the subsheaves,
being torsion and Z-constructible, is constructible. 2
Mapping cones
For a complex A , A Œ1 denotes the complex with .A Œ1/r D ArC1 and the
differential d r D dArC1 . Let uW A ! B be a map of complexes. The mapping
cone C .u/ corresponding to u is the complex A Œ1 ˚ B with the differential
d r D dArC1 C urC1 C dBr . Thus C r .u/ D ArC1 ˚ B r , and the differential
is .a; b/ 7! .da; ua C db/. There is an obvious injection iW B ,! C .u/
and an obvious projection pW C .u/Œ1 A , and the distinguished triangle
corresponding to u is
p u i
C .u/Œ1 ! A ! B ! C .u/:
By definition, every distinguished triangle is isomorphic to one of this form.

A short exact sequence
u v
0 ! A ! B ! C ! 0
gives rise to a distinguished triangle

w u v
C Œ1 ! A ! B ! C
in which w is defined as follows: let qW C .u/ ! C be v on B and zero

on A Œ1; then q is a quasi-isomorphism, and so we can define w to be .p/ ı
q 1 Œ1. A distinguished triangle of complexes of sheaves on Xet
p u i
C Œ1 ! A ! B ! C ;
gives rise to a long exact sequence of hypercohomology groups
! Hr .X; A / ! Hr .X; B / ! Hr .X; C / ! HrC1 .X; A / !
P ROPOSITION 0.10 (a) A morphism of exact sequences of complexes
0 ! A ! B ! C ! 0

? ? ?
?a ? ?c
y yb y
0 ! D ! E ! F ! 0
defines a distinguished triangle
C .c/Œ1 ! C .a/ ! C .b/ ! C .c/:

(b) Assume that the rows of the diagram
C Œ1 ! A ! B ! C

? ?
? ?
y y
F Œ1 ! D ! E ! F
are distinguished triangles and that the diagram commutes; then the diagram can
be completed to a morphism of distinguished triangles.
(c) For any maps
u v
A ! B ! C
of complexes, there is a distinguished triangle
C .v/Œ1 ! C .u/ ! C .v ı u/ ! C .v/:
P ROOF. The statements are all easy to verify. (Note that (b) and (c) are special
cases of the axioms (TR2) and (TR3) (Hartshorne 1966, I 1) for a triangulated
category; also that the distinguished triangle in (c) is the analogue for complexes
of the kernel-cokernel sequence (I 0.24) of a pair of maps.) 2
1 Local results
Except when stated otherwise, X will be the spectrum of an excellent Henselian
discrete valuation ring R with field of fractions K and residue field k. For ex-
ample, R could be a complete discrete valuation ring or the Henselization of the
local ring at a prime in a global field. We shall use the following notations:
Ks
j I D Gal.K s =K un /
ks Run K un G D Gal.K s =K/
j j j g D Gal.k s =k/ D G=I
k R K
i j
Spec k D x ! X u D Spec K
Preliminary calculations
We compute the cohomology groups of Z and Gm :
1. LOCAL RESULTS 149
P ROPOSITION 1.1 (a) Let F be a sheaf on u; then H r .X; jŠ F / D 0 for all r 0,

and consequently there is a canonical isomorphism
'
H r .u; F / ! HxrC1 .X; jŠ F /; all r 0:
(b) For any sheaf F on X , the map H r .X; F / ! H r .x; i F / is an isomor-

phism all r 0:
P ROOF. (a) The cohomology sequence of the pair X u (see 0.1f)
! Hxr .X; jŠ F / ! H r .X; jŠ F / ! H r .u; jŠ F ju/ !
shows that the first part of the statement implies the second.
Let M be the stalk Fu of F at u regarded as a G-module. The functor F 7!
i j F can be identified with
M 7! M I W ModG ! Modg :
The equality Homg .N; M I / D HomG .N; M / for N a g-module shows that
i j has an exact left adjoint, namely, “regard the g-module as a G-module”,
and so i j preserves injectives. Consider the exact sequence (1)
0 ! jŠ F ! j F ! i i j F ! 0:
If F is injective, this is an injective resolution of jŠ F because j and i preserve

'
injectives. As H 0 .X; j F / ! H 0 .X; i i j F / is the isomorphism M G !
.M I /g , the cohomology sequence of the sequence shows that H 0 .X; jŠ F / D 0
for all F and that H r .X; F / D 0 for all r 0 if F is injective. In particular, we
see that if F is injective, then jŠ F is acyclic for .X; /:
Let F ! I be an injective resolution of F . Then jŠ F ! jŠ I is an acyclic
resolution of jŠ F , and so H r .X; jŠ F / ' H r . .X; jŠ I //. But H r . .X; jŠ I // '
.Rr f /.F / where f is the functor F 7! .X; jŠ F / D 0, and so H r .X; jŠ F / D 0
for all r:
(b) The cohomology sequence of
0 ! jŠ j F ! F ! i i F ! 0
yields the required isomorphisms. 2
C OROLLARY 1.2 For all r , H r .X; Z/ D H r .g; Z/; in particular, when k is

finite,
H r .X; Z/ D Z 0 Q=Z 0
r D 0 1 2 3:
P ROOF. This follows immediately from part (b) of the proposition. 2
L EMMA 1.3 If k is algebraically closed, then H r .K; Gm / D 0 for all r 1:
P ROOF. The assumption that R is excellent entails that K b is separable1 over K,

b Therefore K
and we have seen in (I A.6) that K is algebraically closed in K. b is
a regular extension of K, and so we can apply Shatz 1972, Theorem 27, p116,
to obtain that K is a C1 field. It follows that K has cohomological dimension at
most 1, and so H r .K; Gm / D 0 for r 2 (Serre 1964, II.3). 2
L EMMA 1.4 If k is perfect, then Rr j Gm D 0 for all r > 0; therefore
ExtrX .F; j Gm / ' Extru .F ju; Gm /
for all sheaves F on X and all r:
P ROOF. The stalks of Rr j Gm are (see Milne 1980, III 1.15)
.Rr j Gm /x ' H r .Kun ; Gm / D 0; r > 0 (by 1.3)

r r
.R j Gm /x ' H .Ks ; Gm / D 0; r > 0:
This proves the first assertion, and the second follows from (0.1b). 2
P ROPOSITION 1.5 Assume that k is finite.

(a) For all r > 0, H r .X; Gm / D 0:
(b) We have
Hxr .X; Gm / D 0 Z 0 Q=Z 0
r D 0 1 2 3 > 3:
P ROOF. (a) As Rr j Gm D 0 for r 0, H r .X; j Gm / ' H r .K; Gm / all r.

Moreover, H r .K; Gm / D 0 for r ¤ 0; 2 and H 2 .K; Gm / ' H 2 .k; Z/ ' Q=Z
(cf. Lemma 1.3 or I A.1). On the other hand, H r .X; i Z/ ' H r .x; Z/ for all r
(0.3c), and so the exact sequence
ord
0 ! Gm ! j Gm ! i Z ! 0
1A field K is separable over a subfield k if either the characteristic is 0 or the characteristic is
1
p ¤ 0 and K is linearly disjoint from k p over k (Jacobson 1964, p166). It is a regular extension
of k if it is separable over k and k is algebraically closed in K (Zariski and Samuel 1960, p229).
gives rise to an exact sequence

ord
0 ! H 0 .X; Gm / ! K ! Z ! H 1 .X; Gm / ! 0 ! 0 !
'
H 2 .X; Gm / ! H 2 .K; Gm / ! H 2 .k; Z/ ! 0 ! H 3 .X; Gm / ! 0 ! ;
which yields the result.

(b) Consider the cohomology sequence of the pair X u (see 0.1f),
Hx0 .X; Gm / !
0 ! H 0 .X; Gm / !
H 0 .K; Gm / !
Hx1 .X; Gm / !
H 1 .X; Gm / !

ord
R ! K 0:
From the part we have displayed, it is clear that Hx0 .X; Gm / D 0 and Hx1 .X; Gm / '
Z. The remainder of the sequence gives isomorphisms H r .K; Gm / ' HxrC1 .X; Gm /
for r 1, from which the values of Hxr .X; Gm /, r 2, can be read off. 2
C OROLLARY 1.6 Assume that k is finite. If n is prime to char.k/, then Hxr .X; n / '
Z=nZ for r D 2; 3, and Hxr .X; n / D 0 otherwise.
P ROOF. As n is prime to char.k/, the sequence

n
0 ! n ! G m ! G m ! 0
is exact, and the result follows immediately from (1.5). 2
R EMARK 1.7 (a) Part (a) of (1.5) can also be obtained as a consequence of
the following more general result (Milne 1980, III 3.11(a)): if G is a smooth
commutative group scheme over the spectrum of a Henselian local ring, then
H r .X; G/ ' H r .x; G0 / for r 1, where x is the closed point of X , and G0 is
the closed fibre of G=X .
Alternatively, (1.1b) shows that H r .X; Gm / ' H r .x; i Gm /. Obviously

i Gm corresponds to the g-module Run , and it is not difficult to show that
H r .g; Run / D 0 for r > 0 (cf. I A.2).
(b) There is an alternative way of computing the groups Hxr .X; Gm /. When-
ever k is perfect, (1.4) shows that H r .X; j Gm / ! H r .u; Gm / is an isomor-
phism for all r, and it follows immediately that Hxr .X; j Gm / D 0 for all r.
Therefore the exact sequence
ord
0 ! Gm ! j Gm ! i Z ! 0

leads to isomorphisms Hxr .X; i Z/ ! HxrC1 .X; Gm /, and we have seen in (0.1d)
that Hxr .X; i Z/ ' H r .x; Z/. Therefore Hxr .X; Gm / ' H r1 .x; Z/ for all r:
This argument works whenever k is perfect. For example, when k is alge-
braically closed, it shows that Hxr .X; Gm / D 0, Z, 0, ... for r D 0, 1, 2, ... from
which it follows that, if n is prime to char.k/, then Hxr .X; n / D 0, 0, Z=nZ, 0,
... for r D 0, 1, 2, ... . These values of H r .X; n / are those predicted by the
purity conjecture (Artin, Grothendieck, and Verdier 1972/73, XIX).
Since Rr i Š Gm is the sheaf on z associated with the presheaf z 0 7! Hzr0 .X 0 ; Gm /
(here X 0 ! X is the étale covering of X corresponding to z 0 ! z/, we see that
if k is any perfect field, then R1 i Š Gm ' Z and Rr i Š Gm D 0 for r ¤ 1, and
consequently that if .n;char.k// D 1, then R2 i Š n ' Z=nZ and Rr i Š n D 0 for
r ¤ 2. For this last statement concerning n it is not even necessary to assume
that k is perfect.
The duality theorem

We now assume that the residue field k is finite. There are two natural candi-
dates for a trace map Hx3 .X; Gm / ! Q=Z. The first is that in (1.5), namely,
'
the composite of the inverse of H 2 .u; Gm / ! Hx3 .X; Gm / with H 2 .u; Gm / D
invK
H 2 .GK ; Ks / ! Q=Z. The second is that in (1.7b), namely, the compos-
'
ite of the inverse of H 2 .x; Z/ D Hx2 .X; Z/ ! Hx3 .X; Gm / with H 2 .x; Z/ D
invk
H 2 .Gk ; Z/ ! Q=Z. From the definition of invK (see I 1.6) it is clear that the
two methods lead to the same map.
Recall (0.1d) that for any sheaf F on X , Hxr .X; F / ' ExtrX .i Z; F /, and so
there is a canonical pairing
ExtrX .F; Gm / Hx3r .X; F / ! Hx3 .X; Gm /:
On combining the pairing with the trace map, we obtain a map
˛ r .X; F /W ExtrX .F; Gm / ! Hx3r .X; F / :
Before we can state the theorem, we need to endow HomX .F; Gm / with a
topology. We shall see below that the restriction map
HomX .F; Gm / ! Homu .F ju; Gm / ' HomG .Fu ; K s /
is injective. The last group inherits a topology from that on Ks , and we give
HomX .F; Gm / the subspace topology. When F is Z-constructible, all subgroups
of HomX .F; Gm / of finite index prime to char.K/ are open.
T HEOREM 1.8 (a) Let F be a Z-constructible sheaf on X , without p -torsion

if K has characteristic p ¤ 0.
i) The map ˛ 0 .X; F / defines an isomorphism
HomX .F; Gm /^ ! H 3 .X; F /
(completion for the topology of open subgroups of finite index).
ii) The group Ext1X .F; Gm / is finitely generated, and ˛ 1 .X; F / defines
an isomorphism
Ext1X .F; Gm /^ ! H 3 .X; F /
(completion for the topology of subgroups of finite index).
iii) For r 2, the groups ExtrX .F; Gm / are torsion of cofinite-type, and
˛ r .X; F / is an isomorphism.
(b) Let F be a constructible sheaf on X , and assume that K is complete or that
pF D F for p D charK . Then the pairing
ExtrX .F; Gm / Hx3r .X; F / ! Hx3 .X; Gm / ' Q=Z:
is nondegenerate; if pF D F , then all the groups are finite.
We first consider a sheaf of the form i F , F a Z-constructible sheaf on x.

Recall (0.1d), that for such a sheaf Hxr .X; i F / ' H r .x; F /, all r:
L EMMA 1.9 For any sheaf F on x , there is a canonical isomorphism
'
Extr1 r
x .F; Z/ ! ExtX .i F; Gm /, all r 1.
P ROOF. From the exact sequence

0 ! Gm ! j Gm ! i Z ! 0
! ExtrX .i F; Gm / ! ExtrX .i F; j Gm / ! ExtrX .i F; i Z/ ! :
But ExtrX .i F; j Gm / ' Extru .i F ju; Gm / in view of (0.1b) and (1.4), and the
second group is zero because i F ju D 0. Also ExtrX .i F; i Z/ ' Extrx .F; Z/
by (0.1c), and so the sequence gives the required isomorphisms. 2
It is obvious from the various definitions that the diagram
ExtrX .i F; Gm / Hx3r .X; i F / ! Hx3 .X; Gm / ' Q=Z

"' k "'
Extr1
x .F; Z/ H 3r .x; F / ! H 2 .x; Z/ ' Q=Z
commutes. The lower pairing can be identified with the pairing in (I 1.10). We
deduce: Ext1 .i F; Gm / is finitely generated and ˛ 1 .X; i F / defines an isomor-
phism Ext1 .i F; Gm /^ ! Hx2 .X; i F / (completion for the profinite topol-
ogy); ˛ 2 .X; i F / is an isomorphism of finite groups; ˛ 3 .X; i F / is an isomor-
phism of torsion groups of cofinite-type; for all other values of r, the groups are
zero. When F is constructible, it corresponds to a finite g-module, and so all the
groups are finite (and discrete). This completes the proof of the theorem for a
sheaf of the form i F:
We next consider a sheaf of the form jŠ F , with F a Z-constructible sheaf on
u without p-torsion. Consider the diagram
ExtrX .jŠ F; Gm / Hx3r .X; jŠ F / ! Hx3 .X; Gm / ' Q=Z

#' "' "'
Extru .F; Gm / H 2r .u; F / ! H 2 .u; Gm / ' Q=Z
in which the first isomorphism is restriction from X to u (see 0.1a), and the two
remaining isomorphisms are boundary maps in the cohomology sequence of the
pair X u (see (1.1) and (1.5)). It is again clear from the various definitions that
the diagram commutes. The lower pairing can be identified with that in (I 2.1).
We deduce:
˘ HomX .jŠ F; Gm / is finitely generated and ˛ 0 .X; jŠ F / defines an isomor-
phism
HomX .jŠ F; Gm /^ ! Hx3 .X; jŠ F /
(completion for the topology of open subgroups of finite index);
˘ ˛ 1 .X; jŠ F / is an isomorphism of finite groups;
˘ ˛ 2 .X; i F / is an isomorphism of torsion groups of cofinite type;
˘ for all other values of r, the groups are zero.
When F is constructible, it corresponds to a finite G-module, and all the groups
are finite (and discrete).
This completes the proof of the theorem when F i i F or F jŠ j F
and F is without p-torsion. For a general Z-constructible sheaf F without p-
torsion, we use the exact sequence
0 ! jŠ .F jU / ! F ! i i F ! 0;
and apply the five-lemma to the diagram

ExtrX .i i F; Gm / !
! ExtrX .F; Gm / !
ExtrX .jŠ .F jU /; Gm / !

? ? ? ? ?
? ? ? ? ?
y y y y y
Hx3r .X; i i F / !
! Hx3r .X; F / !
Hx3r .X; jŠ .F jU // !

This leads immediately to a proof of (a) of the theorem for r 2. For r < 2, one
only has to replace the first four terms in the top row of the diagram with their
completions:
Hom.F; Gm /^ !
0 ! Hom.jŠ .F jU /; Gm /^ !
Ext1X .i i F; Gm /^ !

? ? ?
? ? ?
y y y
Hx3 .X; F /
0 ! Hx3 .X; jŠ .F jU //
! Hx2 .X; i i F /
! !

Note that the top row is exact by virtue of (I 0.20a).
The remaining case, where K is complete and F is constructible with p tor-
sion, can be treated similarly.
C OROLLARY 1.10 Let p D char k:
(a) Let F be a locally constant constructible sheaf on X such that pF D F ,
and let F D D Hom.F; Gm /. Then there is a canonical nondegenerate
pairing of finite groups
H r .X; F D / Hx3r .X; F / ! Hx3 .X; Gm / ' Q=Z:
(b) Let F be a constructible sheaf on u such that pF D F , and let F D D

Homu .F; Gm /. The pairing F D F ! Gm extends to a pairing j F D
j F ! Gm , and the resulting pairing
H r .X; j F D / Hx3r .X; j F / ! Hx3 .X; Gm / ' Q=Z
is nondegenerate.
P ROOF. (a) We shall use the spectral sequence (Milne 1980, III.1.22)
H r .X; ExtX
s
.F; Gm // H) ExtrCs
X .F; Gm /
to show that the term ExtrX .F; Gm / in the theorem can be replaced with H r .X; F D /.
According to Milne 1980, III 1.31, the stalk of ExtXs .F; G / at x is Exts .F ; R /
m x un
(Ext as abelian groups). This group is zero for s > 0 because Run is divis-
ible by all primes dividing the order of Fx . The stalk of ExtX s .F; G / at u
m
s
is Ext .Fx ; Ks /, which is zero for s > 0 by the same argument. Therefore
ExtX s
.F; Gm / is zero for s > 0, and the spectral sequence collapses to give iso-
morphisms H r .X; F D / ' ExtrX .F; Gm /:
'
(b) On applying j to the isomorphism F D ! Homu .F; Gm /, we obtain an
'
isomorphism j F D ! j Homu .F; Gm /. But
j Homu .F; Gm / ' j Hom.j j F; Gm / ' HomX .j F; j Gm /

(see Milne 1980, II 3.21), and from the Ext sequence of
0 ! Gm ! j Gm ! i Z ! 0
and the vanishing of HomX .j F; i Z/ ' Homx .i j F; Z/ we find that

HomX .j F; j Gm / ' HomX .j F; Gm /. Thus j F D ' HomX .j F; Gm /
and the existence of the pairing j F D j F ! Gm is obvious.
Next we shall show that ExtX r .j F; G / D 0 for r > 0. Let M D F .
m u
If M D M , then j F is locally constant, and we showed in the proof of part
I
(a) of the corollary that the higher Ext’s vanish for such sheaves. If M I D
0, then j F D jŠ F , and ExtX r .j F; G / ' j Ext r .F; G /. The stalk of
Š m u m
j Extu .F; Gm / at x is H .K un ; M D /. Because M has order prime to p,
r r
H r .I; M D / ' H r .I =Ip ; M D /, where Ip is the p-Sylow subgroup of I . But

H r .I =Ip ; M D / is zero for r > 1, and H 1 .I =Ip ; M D / is dual to H 0 .I =Ip ; M /,
which equals M I (cf. the proof of I 2.18). By assumption, this is zero. Because
I is normal in GK , every GK -module has a composition series whose quotients
Q are such that either QI D Q or QI D 0. Our arguments therefore show that
ExtXr .j F; G / D 0 for r > 0:
m
The spectral sequence
H r .X; ExtX
s
.j F; Gm // H) ExtrCs
X .j F; Gm /
'
therefore reduces to a family of isomorphisms H r .X; j F D / ! ExtrX .j F; Gm /,
and the corollary follows from the theorem. 2
R EMARK 1.11 (a) Part (a) of the theorem is true without the condition that F has
no p torsion, p D char K, provided one endows Ext1X .F; Gm / with a topology
deduced from that on K and defines Ext1X .F; Gm /^ to be the completion with
respect to the topology of open subgroups of finite index. Note that ˛ 1 .X; i Z/
is the natural inclusion Z ,! b Z, and that ˛ 1 .X; jŠ Z=pZ/ for p D char K is
an isomorphism of infinite compact groups K =K p ! .K=}K/ when K is
complete. The first example shows that it is necessary to complete Ext1X .F; Gm /
in order to obtain an isomorphism, and the second shows that it is necessary to
endow Ext1X .F; Gm / with a topology coming from K because not all subgroups
of finite index in K =K p are open.
(b) By using derived categories, it is possible to restate (1.8) in the form of
(1.10) for any constructible sheaf F such that pF D F . Simply set F D D
RHom.F; Gm / (an object in the derived category of the category of constructible
sheaves on X /, and note that Hr .X; F D / D ExtrX .F; Gm /. (The point of the
proof of (1.10) is to show that H r .RHom.F; Gm // D 0 for r > 0 when F is
locally constant.)
Singular schemes
We now let X D Spec R with R the Henselization of an excellent integral local
ring of dimension 1 with finite residue field k. Then R is again excellent, but it
need not be reduced. Let u D fu1 ; :::; um g be the set of points of X of dimension
0. Then OX;ui is a field Ki , and the normalization RQ of R is a product of excellent
Henselian discrete valuation rings Ri such that Ri has field of fractions Ki (see
Raynaud 1970, IX). We have a diagram
Qi Q
fx1 ; : : : ; xm g / Q o j u
X

j

x
i /X
with XQ D Spec RQ and x and xi the closed points of X and Spec Ri respectively.
For h in the total ring of fractions of R, define
P
ord.h/ D Œk.xi /W k.x/ ordi .h/
where ordi is the valuation on Ki . One can define a similar map for any U étale
over X , and so obtain a homomorphism ordW j Gm ! i Z. Define G be the
complex of sheaves j Gm ! i Z on X:
L EMMA 1.12 (a) For all r > 0, Rr j Gm D 0; therefore
ExtrX .F; j Gm / ' Extru .F ju; Gm /
for all sheaves F on X and all r .

(b) For all r , there is a canonical isomorphism H r1 .x; Z/ ! Hxr .X; G/:
P ROOF. (a) The map is finite, and therefore is exact. As j D jQ , this
shows that Rr j Gm ' Rr jQ Gm , which is zero by (1.4).
(b) From (a) and (0.1) we see that
Hxr .X; j Gm / ' ExtrX .i Z; j Gm / ' Extru .i Zju; Gm / D 0;
all r. The exact sequence
! Hxr .X; G/ ! Hxr .X; j Gm / ! Hxr .X; i Z/ !
now leads immediately to the isomorphism. 2
'
We define the trace map Hx3 .X; Gm / ! Q=Z to be the composite of the
' '
inverse of H 2 .x; Z/ ! Hx3 .X; Gm / and invk W H 2 .g; Z/ ! Q=Z:
T HEOREM 1.13 For any constructible sheaf F on X ,

ExtrX .F; Gm / Hx3r .X; F / ! Hx3 .X; Gm / ' Q=Z;
is a nondegenerate pairing of finite groups.
As in the case that X is regular, it suffices to prove this for sheaves of the
form i F and jŠ F:
L EMMA 1.14 For all r , ExtrX .i F; j Gm / D 0; therefore the boundary maps
Extr1 r
x .F; Z/ ! ExtX .i F; Gm / are isomorphisms.
P ROOF. The proof is the same as that of (1.9). 2
The theorem for i F now follows from the diagram:
ExtrX .i F; G/ Hx3r .X; i F / ! Hx3 .X; G/ ' Q=Z

"' "' "'
Extr1
x .F; Z/ H 3r .x; F / ! H 2 .x; Z/ ' Q=Z:
We next consider a sheaf of the form jŠ F:
L EMMA 1.15 For all r , H r .X; jŠ F / D 0; therefore the maps
H r1 .u; F / ! Hxr .X; jŠ F /
are isomorphisms.
0 ! jQŠ F ! jQ F

0 ! jŠ F ! j F ! i i j F ! 0:
Because . jQŠ F /x D 0, the map of jQŠ F into i i j F is zero, and therefore
the image of jQŠ F is contained in jŠ F . The resulting map jQŠ F ! jŠ F in-
duces isomorphisms on the stalks and therefore is itself an isomorphism. The first
assertion now follows from (1.1), and the second is an immediate consequence of
the first. 2
The theorem for jŠ F now follows from the diagram:
ExtX .jŠ F; G/ Hx3r .X; jŠ F / ! Hx3 .X; G/ ' Q=Z

P
#' "' " "
L L L
i Extrui .F; Gm / i H 2r .ui ; Fi / ! i H 2 .ui ; Gm / ' .Q=Z/m :
Higher dimensional schemes

We obtain a partial generalization of (1.8) to d -local fields. Recall from (I 2) that
a 0-local field is a finite field, and that a d -local field is a field that is complete
with respect to a discrete valuation and has a .d 1/-local field as residue field. If
p is either 1 or a prime and M is an abelian group or sheaf, we write M.nonp/
for lim Mm where the limit is over all integers prime to p. We also write 1 .r/
!
for the sheaf lim ˝r on Xet (limit over all integers m/:
! m
T HEOREM 1.16 Let K be a d -local field with d 2, and let p D char.K1 /
where K1 is the 1-local field in the inductive definition of K . Let X be Spec R
with R the discrete valuation ring in K , and let x and u be the closed and open
points Spec k and Spec K of X .
(a) There is a canonical isomorphism
'
Hxd C2 .X; 1 .d //.nonp/ ! Q=Z.nonp/:
(b) For any constructible sheaf F on X such that pF D F;
ExtrX .F; 1 .d // Hxd C2r .X; F / ! Hxd C2 .X; 1 .d // ! Q=Z
is a nondegenerate pairing of finite groups, all r:
P ROOF. (a) Let i and j be the inclusions of x and u respectively into X . As
we observed in (1.7), Rr i Š m ' Z=mZ for r D 2 and is zero otherwise.
On tensoring both sides with ˝dm
1 , we find that R r i Š ˝d ' ˝d 1 for
m m
r D 2 and is zero otherwise. Next, on passing to the direct limit, we find that
Rr i Š 1 .d /.nonp/ ' 1 .d 1/.nonp/ for r D 2 and is zero otherwise.
Now the spectral sequence
H r .x; Rs i Š 1 .d // H) HxrCs .X; 1 .d //
shows that Hxd C2 .X; 1 .d //.nonp/ ' H d .x; 1 .d 1//.nonp/, which
equals .Q=Z/.non-p/ by (I 2.17).
We give a second derivation of this trace map. Note that (1.1) implies that
'
H d C1 .u; 1 .d // ! Hxd C2 .X; jŠ 1 .d //. Moreover Hxd C2 .X; jŠ 1 .d // !
Hxd C2 .X; 1 .d // is an isomorphism because k has cohomological dimension d
(this is implied by (I 2.17) applied to k/. Therefore we have a trace map
Hxd C2 .X; 1 .d //.non-p/ H d C1 .u; 1 .d //.non-p/ .Q=Z/.non-p/:
The inductive approach we adopted to define the trace map in (I 2.17) shows that
the two definitions give the same trace map are equal.
(b) As in the previous cases, it suffices to prove this for sheaves of the form
i F and jŠ F: 2
L EMMA 1.17 For all r , there are canonical isomorphisms
Extr2 r
x .F; 1 .d 1// ! ExtX .i F; 1 .d //:
P ROOF. Because Rr i Š 1 .d / D 0 for r ¤ 2 and R2 i Š 1 .d / ' 1 .d 1/,

the spectral sequence,
Extrx .F; Rs i Š 1 .d // H) ExtrCs

X .i F; 1 .d //;
collapses to give the required isomorphisms. 2
The theorem for i F now follows from the diagram,
ExtrX .i F; 1 .d // Hxd C2r .X; i F / ! Hxd C2 .X; 1 .d // ! Q=Z

" k "
Extr2 d C2r .x; F / ! H d .x; .d 1// ! Q=Z
x .F; 1 .d 1// H 1
and (I 2.17) applied to k:

Let F be a sheaf on u. The theorem for jŠ F follows from the diagram,
ExtrX .jŠ F; 1 .d // Hxd C2r .X; jŠ F / ! Hxd C2 .X; 1 .d // ' Q=Z

# " "
Extru .F; 1 .d // H d C1r .u; F / ! H d C1 .u; 1 .d // ' Q=Z
and (I.2.17) applied to K. This completes the proof of Theorem 1.16.

For any ring A we write Kr A for the r th Quillen K-group of A, and for
any scheme X , we write Kr for the sheaf on Xet associated with the presheaf
U 7! Kr .U; OU /.
T HEOREM 1.18 Let K and p be as in (1.16).

(a) There is a canonical isomorphism
'
Hxd C2 .X; K2d 1 /.nonp/ ! .Q=Z/.nonp/:
(b) For any constructible sheaf F on X such that pF D F ,
ExtrX .F; K2d 1 / Hxd C2r .X; F / ! Hxd C2 .X; K2d 1 / ! Q=Z

The main part of the proof is contained in the next lemma. Recall (Browder
1977) that for any ring A, there are K-groups with coefficients Kr .A; Z=mZ/
fitting into exact sequences
0 ! Kr .A/.m/ ! Kr .A; Z=mZ/ ! Kr1 .A/m ! 0:
Also that for any ring A and integer m that is invertible in A, there is a canon-
ical map m ! K2 .A; Z=mZ/. Using the product structure on the groups
Kr .A; Z=mZ/, we obtain a canonical map m .r/ ! K2r .A; Z=mZ/ ! K2r1 .A/:
L EMMA 1.19 Let X be any scheme. If m is invertible on X , then there is an

exact sequence
m
0 ! m .d / ! K2d 1 ! K2d 1 ! 0
of sheaves on Xet .
P ROOF. This is fairly direct consequence of the following two theorems.

(a) Let k be an algebraically closed field; then K2r k is uniquely divisible for
all r, and K2r1 k is divisible with torsion subgroup equal to 1 .r/ (Suslin
1983b, 1984).
(b) If R is a Henselian local ring with residue field k and m is invertible in
'
R, then Kr .R; Z=mZ/ ! Kr .k; Z=mZ/ for all r (Gabber 1983; see also
Suslin 1984).
For any field extension L=k of degree p n , there are maps
f W Kr .k/ ! Kr .L/, f W Kr .L/ ! Kr .k/
such that f ıf D p n D f ıf . Therefore Kr .k/.nonp/ ! Kr .L/.nonp/

is an isomorphism. This remark, together with (a) and a direct limit argument, im-
plies that for a separably closed field k with char.k/ D p, .K2r k/.nonp/ is
uniquely divisible for all r and .K2r1 k/.nonp/ is divisible with torsion sub-
group equal to 1 .r/.nonp/. In terms of K-groups with coefficients, this says
that K2r .k; Z=mZ/ ' K2r1 .k/m ' m .r/ and K2r1 .k; Z=mZ/ D 0 for all
m prime to p:
Let R be a strictly Henselian local ring. From the diagram
0 ! K2r .R/.m/ ! K2r .R; Z=mZ/ ! K2r1 .R/m ! 0
? ? ?
? ?' ?
y y y

0 ! K2r .k/.m/ ! K2r .k; Z=mZ/ ! K2r1 .k/m ! 0
'
we see that K2r1 .R/m ! K2r1 .k/m , and therefore that the map m .r/ !
K2r1 .R/m is an isomorphism. As K2r1 .R; Z=mZ/ ' K2r1 .k; Z=mZ/ D 0,
we know that K2r1 .R/.m/ D 0, and therefore that the sequence
m
0 ! m .r/ ! K2r1 .R/ ! K2r1 .R/ ! 0:
is exact. This implies the lemma because the exactness of a sequence of sheaves
can be checked on the stalks. 2
L EMMA 1.20 Let X be the spectrum of a Henselian discrete valuation ring with
closed point x ; then for any sheaf F on X , Hxr .X; F / is torsion for r 2:
P ROOF. Let u be the open point of X , and consider the exact sequence
! H 1 .u; F / ! Hx2 .X; F / ! H 2 .X; F / ! :
'
From (1.1) we know that H r .X; F / ! H r .x; i F / (excellence is not used in
the proof of (1.1)), and H r .x; i F / and H r .u; F ju/ are both torsion for r > 0
because they are Galois cohomology groups. The lemma follows. 2
We now complete the proof of Theorem 1.18. The exact sequence in the
lemma leads to an exact sequence (ignoring p-torsion)
0 ! Hxd C1 .X; K2d 1 /˝Q=Z ! Hxd C2 .X; 1 .d // ! Hxd C2 .X; K2d 1 / ! 0;
which shows that
Hxd C2 .X; 1 /.nonp/ ! Hxd C2 .X; K2d 1 /.nonp/
is an isomorphism because the first term in the sequence is zero. Similarly,
'
ExtrX .F; 1 / ! ExtrX .F; K2d 1 /
for all r, and so (1.16) implies (1.18).
R EMARK 1.21 The corollary is a satisfactory generalization of (1.8) in the case
that K1 has characteristic zero. The general case, where the characteristic jumps
from p to zero at some later stage, is not yet understood. For a discussion of what
the best result should be, see 7 below.
N OTES Part (b) of Theorem 1.8 is usually referred to as the local form of the duality
theorem of Artin and Verdier, although Artin and Verdier 1964 only discusses global
results. In Deninger 1986c it is shown that the result extends to singular schemes when
Gm is replaced by G. The extension to higher dimensional schemes in Theorems 1.16
and 1.18 is taken from Deninger and Wingberg 1986. The key lemma 1.19 has probably
been proved by several people.
2. GLOBAL RESULTS: PRELIMINARY CALCULATIONS 163
2 Global results: preliminary calculations

Throughout this section, K will be a global field. When K is a number field,
X denotes the spectrum of the ring of integers in K, and when K is a function
field, k denotes the field of constants of K and X denotes the unique connected
smooth complete curve over k having K as its function field. The inclusion of
the generic point into X is denoted by gW Spec K ! X , and we sometimes write
for Spec K. For any open subset U of X , U 0 is the set of closed points of
U , often regarded as the set of primes of K corresponding to points of U . The
residue field at a nonarchimedean prime v is denoted by k.v/. The field Kv is
the completion of K at v if v is archimedean, and it is the field of fractions of
the Henselization Ohv of Ov otherwise. For a sheaf F on X or an open subset
of X , we sometimes write Fv for the sheaf on Spec Kv obtained by pulling back
relative to the obvious map
fv W Spec Kv ! Spec K ! X:
Note that when v is nonarchimedean, there is a commutative diagram

Spec Kv ! Spec K
? ?
? ?
y y
Spec Ovh ! X:
For an archimedean prime v of K and a sheaf F on Spec.Kv /, we set H r .Kv ; F / '
HTr .Gv ; M / (notation as in I 0) where M is the Gv -module corresponding to F
and Gv D Gal.Kv;s =Kv /. Therefore H r .K; F / is zero for all r 2 Z when v is
complex, and H r .Kv ; F / is isomorphic to HT0 .Gv ; M / or H 1 .Gv ; M / accord-
ing as r is even or odd when v is real. We let gv D Gal.k.v/s =k.v//:
The cohomology of Gm
P ROPOSITION 2.1 Let U be an open subset of X , and let S denote the set of all
primes of K (including the archimedean primes) not corresponding to a point of
U . Then
H 0 .U; Gm / D .U; O U / ;
H 1 .U; Gm / D Pic.U /;
there is an exact sequence
M P
2 invv
0 ! H .U; Gm / ! Br.Kv / ! Q=Z ! H 3 .U; Gm / ! 0:
v2S
and M
H r .U; Gm / D H r .Kv ; Gm /; r 4:
v real
P ROOF. Let gW ,! U be the inclusion of the generic point of U . There is an

exact sequence (Milne 1980, II 3.9)
0 ! Gm ! g Gm; ! DivU ! 0 (3)

L
with DivU D v2U 0 iv Z the sheaf of Weil divisors on U . The same argument
as in (1.4) shows that Rs g Gm D 0 for s > 0, and so the Leray spectral sequence
for g degenerates to a family of isomorphisms
'
H r .U; g Gm; / ! H r .K; Gm /; r 0:
Clearly,
L L
H r .U; DivU / D v2U 0 H
r
.v; i Z/ D v2U 0 H
r
.k.v/; Z/;
and we know from (I A.2) that
H r .k.v/; Z/ ' Z 0 Br.Kv / 0

r D 0 1 2 3:
The cohomology sequence of (3) therefore gives exact sequences

L
0 ! H 0 .U; Gm / ! K ! Z ! H 1 .U; Gm / ! 0;
v2U 0
2 L
0 ! H .U; Gm / ! Br.K/ ! Br.Kv / ! H 3 .U; Gm / ! H 3 .K; Gm / ! 0;
v2U 0
'
H r .U; Gm / ! H r .K; Gm /, r 4:
The first sequence shows that H 0 .U; Gm / and H 1 .U; Gm / have the values claimed
in the statement of the proposition. Global class field theory (Tate 1967a, and
Chapter I, A.7) provides an exact sequence
L
0 ! Br.K/ ! all v Br.Kv / ! Q=Z ! 0:
From this and the second of the above sequences, we see that the kernel-cokernel
sequence of the pair of maps
L L
Br.K/ ! all v Br.Kv / ! v2U 0 Br.Kv /
is the required exact sequence

P
L invv
H 2 .U; Gm / !
0 ! v2S Br.Kv / ! Q=Z !
H 3 .U; Gm / !
H 3 .K; Gm / !
0;
To
L complete the proof in the number field case, we use that H r .K; Gm / '
v real H .Kv ; Gm / for r 3 (see I 4.21), and that H .R; Gm / D 0. In the
r 3
function field case, we know that K has cohomological dimension 2, which

implies that H r .K; Gm / D 0 for r 3: 2
R EMARK 2.2 (a) If S contains at least one nonarchimedean prime, then the map
X L
invv W v2S Br.Kv / ! Q=Z
is surjective, and so in this case there is an exact sequence

L
0 ! H 2 .U; Gm / ! v2S Br.Kv / ! Q=Z ! 0,
and
( L
v ; Gm /; r 4
0 .K
r v real H r even;
H .U; Gm / ' K a number field,
0; r odd; r 3
H r .U; Gm / D 0; r 3; K a function field:
(Recall that, for v real, H 0 .Kv ; / D HT0 .Gv ; /.)

(b) If K has no real primes, then H 2 .X; Gm / D 0, H 3 .X; Gm / D Q=Z, and
H .X; Gm / D 0 for r 4:
r
Cohomology with compact support

We shall define cohomology groups with compact support Hcr .U; F /, r 2 Z, that
take into account the real primes. They will fit into an exact sequence
L
! Hcr .U; F / ! H r .U; F / ! v…U H r .Kv ; Fv / ! HcrC1 .U; F / !
(4)
(sum over all primes of K, including any archimedean primes, not corresponding
' L
to a point of U ). In particular, Hcr .U; F / ! v real H r1 .Kv ; Fv / for r < 0.
The group Hcr .U; F / differs from H r .X; jŠ F / by a group killed by 2, and equals
it except in the case that K is a number field with a real embedding. The reader
who is prepared to ignore the prime 2 can skip this subsection.
Let F be a sheaf on U , and write C .F / for the canonical Čech complex
of F defined in 0. Thus .U; C .F // D C .U; F /, and H r .C .U; F // D
H r .U; F /. For each prime v, there is a canonical map fv C .F / ! C .Fv /

and therefore also a map C .U; F / ! C .Kv ; Fv /. As we noted in (0.8),
C .Kv ; Fv / can be identified with the standard (inhomogeneous) resolution
C .Mv / of the Gv -module Mv associated with Fv . When v is real, we write
S .Mv / (or S .Kv ; Fv /) for a standard complete resolution of Mv , and oth-
erwise we set S .Mv / ' C .Mv /. In either case there is a canonical map
C .Mv / ! S .Mv /. On combining this with the previous maps, we obtain a
canonical morphism of complexes
L
uW C .U; F / ! v…U S .Mv /
(sum over all primes of K not in U /. We define2 Hc .U; F / to be the translate

C .u/Œ1 of the mapping cone of u, and we set Hcr .U; F / D H r .Hc .U; F //.
P ROPOSITION 2.3 (a) For any sheaf F on an open subscheme U X , there is

an exact sequence
L
! Hcr .U; F / ! H r .U; F / ! v…U H r .Kv ; Fv / ! HcrC1 .U; F / ! :
(b) A short exact sequence
0 ! F 0 ! F ! F 00 ! 0
of sheaves on U , gives rise to a long exact sequence of cohomology groups
! Hcr .U; F 0 / ! Hcr .U; F / ! Hcr .U; F 00 / ! .
(c) For any closed immersion iW Z ,! U and sheaf F on Z ,
Hcr .U; i F / ' H r .Z; F /:
(d) For any open immersion j W V ,! U and sheaf F on V ,
Hcr .U; jŠ F / ' Hcr .V; F /:
Therefore, for any sheaf F on U , there is an exact sequence

L
! Hcr .V; F jV / ! Hcr .U; F / ! v2V XU H r .v; iv F / !
(e) For any finite map W U 0 ! U and sheaf F on U 0 , there is a canonical

'
isomorphism Hcr .U; F / ! Hcr .U 0 ; F /:
2 In the original, this was written Hc .
P ROOF. (a) This is obvious from the definition of Hc .U; F / and the properties of
mapping cones (see 0).
(b) From the morphism
0 ! C .U; F 0 / ! C .U; F / ! C .U; F 00 / ! 0

? ? ?
? 0 ?u ? 00
yu y yu
L L L
0 ! S .Kv ; Fv0 / ! S .Kv ; Fv / ! S .Kv ; Fv00 / ! 0
of short exact sequences of complexes, we obtain a distinguished triangle
Hc .U; F 00 /Œ1 ! Hc .U; F 0 / ! Hc .U; F / ! Hc .U; F 00 /:
(see 0.10a). This yields the long exact sequence.

(c) Since the stalk of i F at the generic point is zero, H r .Kv ; .i F /v / D
'
0 for all v. Therefore Hcr .U; i F / ! H r .U; i F /, and the second group is
isomorphic to H r .Z; F /.
Before proving (d), we need a lemma.
L EMMA 2.4 Under the hypotheses of (d), there is a long exact sequence
L
! H r .U; jŠ F / ! H r .V; F / ! v2U XV H r .Kv ; F / ! .
P ROOF. The cohomology sequence of the pair U V is
! HUr V .U; jŠ F / ! H r .U; jŠ F / ! H r .V; F / ! .

L
By excision (Milne 1980, III.1.28), HUr V .U; jŠ F / ' r
v2U XV Hv .Uv ; jŠ F /
where Uv D Spec Ohv , and according to (1.1), Hvr .Uv ; jŠ F / ' H r1 .Kv ; F /.
The lemma is now obvious. 2
P ROOF ( OF (2.3) CONTINUED ) On carrying out the proof of the lemma on the
level of complexes, we find that the mapping cone of
C .U; jŠ F / ! C .V; jŠ F jV / D C .V; F /

L
is quasi-isomorphic to v2U XV C .Kv ; Fv /. The cokernel of
L L
v…U S .Kv ; Fv / ! v…V S .Kv ; Fv /
L
is also v2U XV C .Kv ; F /, and so the mapping cone of Hc .U; jŠ F / ! Hc .V; F /
is quasi-isomorphic to the mapping cone of a map
L L
v2U XV C .Kv ; F / ! v2U XV C .Kv ; F /:
The map is the identity, and therefore Hc .U; jŠ F / ! Hc .V; F / is a quasi-

isomorphism. This completes the proof of the first part of (d), and to deduce
the second part one only has to replace Hcr .U; jŠ F / with Hcr .V; F / in the coho-
mology sequence of (see (1))
0 ! jŠ j F ! F ! i i F ! 0:
Finally, (e) of the proposition follows easily from the existence of isomor-
' '
phisms H r .U; F / ! H r .U 0 ; F / and H r .Kv ; F / ! H r .Kv0 ; F /: 2
P ROPOSITION 2.5 (a) For any sheaves F and F 0 on U X , there is a canonical

pairing
h ; iW ExtrU .F; F 0 / Hcs .U; F / ! HcrCs .U; F 0 /:
(b) For any pairing F F 0 ! F 00 of sheaves on U X , there is a natural
cup-product pairing
Hcr .U; F / Hcs .U; F 0 / ! HcrCs .U; F 00 /:
(c) The following diagram commutes:
H r .U; Hom.F; F 0 // Hcs .U; F / ! HcrCs .U; F 0 / (cup-product)

# k k
ExtrU .F; F 0 / Hcs .U; F / ! HcrCs .U; F 0 / (Ext pairing):
P ROOF. (a) For example, represent an element of ExtrU .F; F 0 / by an r-fold ex-
tension, and take Hcr .U; F / ! HcrCs .U; F 0 / to be the corresponding r-fold
boundary map.
(b) The cup-product pairing on the Čech complexes (Milne 1980, V.1.19)
C .U; F / C .U; F 0 / ! C .U; F 00 /
combined with the cup-product pairing on the standard complexes (Cartan and
Eilenberg 1956, XII)
S .Kv ; Fv / S .Kv ; Fv0 / ! S .Kv ; Fv00 /
gives a natural pairing
Hc .U; F / Hc .U; F 0 / ! Hc .U; F 00 /:
(c) Combine (I 0.14) with (0.7). 2

The cohomology of Gm with compact support

P ROPOSITION 2.6 Let U be an open subscheme of X . Then Hc2 .U; Gm / D 0,
Hc3 .U; Gm / D Q=Z, and Hcr .U; Gm / D 0, r > 3:
P ROOF. Part (a) of (2.3) gives exact sequences

L
0 ! Hc2 .U; Gm / ! H 2 .U; Gm / ! Br.Kv / !
v…U
Hc3 .U; Gm / ! H 3 .U; Gm / ! 0
L
0 ! Hc2r .U; Gm / ! H 2r .U; Gm / ! H 2r .Kv ; Gm / !
v real
Hc2rC1 .U; Gm / ! H 2rC1 .U; Gm / ! 0;
(2r 4). When U ¤ X , these sequences and (2.2a) immediately give the
proposition, but (2.3d) and the next lemma show that Hcr .U; Gm / does not depend
on U if r 2. 2
L EMMA 2.7 For any closed immersion iW Z ,! U whose image i.Z/ ¤ U ,

H r .Z; i Gm / D 0 all r 1:
P ROOF. It suffices to prove this with Z equal to a single point v of U . Then i Gm

corresponds to the gv -module Oun
v , and so H .Z; i Gm / ' H .gv ; O v /.
r r un
The sequence
ord
0 ! Ovun ! Kvun ! Z ! 0
is split as a sequence of gv -modules, and so H r .gv ; Oun
v / is a direct summand of
H .gv ; Kv /. Therefore Hilbert’s theorem 90 shows that H 1 .gv ; Oun
r un
v / D 0,
and we know from (I A.2) that H 2 .gv ; Ounv / D 0. As g v has strict cohomolog-
ical dimension 2, this completes the proof. 2
R EMARK 2.8 (a) Let K be a number field, and let R be the ring of integers in K.
Then there is an exact sequence
L
0 ! Hc0 .X; Gm / ! R ! 2
v real Kv =Kv ! Hc1 .X; Gm / ! Pic.R/ ! 0;
where Pic.R/ is the ideal class group of R. In particular,
Hc0 .X; Gm / ' fa 2 R j sign.av / > 0 all real vg

D groups of totally positive units in K:
Let Id.R/ be the group of fractional ideals in R. Then
Hc1 .X; Gm / ' Id.R/=f.a/ j a 2 K , sign.av / > 0 all real vg

' group of ideal classes of K in the narrow sense
(see Narkiewicz 1974, III, 2, 3).

The cohomology sequence with compact support of
L
0 ! G m ! g G m ! v2X 0 iv Z !0
is
L
Hc0 .X; g Gm / ! v nonarch. Z ! Hc1 .X; Gm / ! Hc1 .X; g Gm /:
The exact sequence given by (2.3a)
L
0 ! Hc0 .X; g Gm / ! K ! 2
v real Kv =Kv ! Hc1 .X; g Gm / ! 0
shows that Hc0 .X; g Gm / is the group of totally positive elements of K and
Hc1 .X; g Gm / D 0.
(b) Unfortunately Hc1 .X; Gm / is not equal to the group of isometry classes
of Hermitian invertible sheaves on X (the “compactified Picard group of R” in
the sense of Arakelov theory; see Szpiro 1985, 1). I do not know if there is a
reasonable definition of the étale cohomology groups of an Arakelov variety. Our
definition of the cohomology groups with compact support has been chosen so as
to lead to good duality theorems.
Cohomology of locally constant sheaves

Let U be an affine open subset of X , and let S be the set of primes of K not
corresponding to a point of U . Since to be affine in the function field case simply
means that U ¤ X , S will satisfy the conditions in the first paragraph of I 4.
With the notations of (I 4), GS D 1 .U; /, and the functor F 7! F defines an
equivalence between the category of locally constant Z-constructible sheaves on
U and the category of finitely generated discrete GS -modules. We write UQ for
the normalization of U in KS ; thus UQ D Spec RS where, as in (I 4), RS is the
integral closure of RK;S in KS .
P ROPOSITION 2.9 Let F be a locally constant Z-constructible sheaf on an open

affine subscheme U of X , and let M D F . Then H r .U; F / is a torsion group
for all r 1, and H r .U; F /.`/ D H r .GS ; M /.`/ for all r if ` is invertible on U
or ` D char.K/:
P ROOF. The Hochschild-Serre spectral sequence for UQ =U is
H r .GS ; H s .UQ ; F jUQ // H) H rCs .U; F /:
As H 0 .UQ ; F / D M , we have to show that H s .UQ ; F jUQ / is torsion for s > 0

and that H s .UQ ; F jUQ /.`/ D 0 if ` is invertible on U or equals char.K/. By
assumption F jUQ is constant, and so there are three cases to considerW F jUQ D
Z=`Z with ` invertible on U , F jUQ D Z=pZ with p D char.K/, and F jUQ D Z:
The first cohomology group can be disposed off immediately, because
H 1 .UQ ; F / ' Hom.1 .UQ ; /; F .UQ //;
and 1 .UQ ; / is zero.

Now let F jUQ D Z=`Z with ` a prime that is invertible in RS . Then Z=`Z
` on UQ , and the remark just made shows that the cohomology sequence of
`
0 ! ` ! G m ! G m ! 0
is
`
0 ! Pic.UQ / ! Pic.UQ / ! H 2 .UQ ; Z=`Z/ ! Br.UQ /` ! 0:
The Picard group of UQ is the direct limit of the Picard groups of the finite étale
coverings U 0 of U and so is torsion. The sequence shows that Pic.UQ /.`/ D 0,
and so H 2 .UQ ; Z=`Z/ injects into Br.UQ /. Let L KS be a finite extension of K
containing the `th roots of 1, and consider the exact sequence (see 2.2a)
L
0 ! Br.RL;S / ! w2SL Br.Lw / ! Q=Z ! 0:
Let L0 be a finite extension of L; it is clear from the sequence and local class field
theory that an element a of Br.RL;S /` maps to zero in Br.RL0 ;S / if ` divides
the local degree of L0 =L at all w in SL . Let H be the Hilbert class field of L.
Then the prime ideal corresponding to w becomes principal in H with generator
1=`
cw say. The field L0 generated over H by the elements cw , w 2 SL , splits
a. As L0 is contained in KS , this argument shows that lim Br.L/` D 0, and
!
therefore that Br.UQ /.`/ D 0. Hence H 2 .UQ ; Z=`Z/ D 0. Finally, (2.2) shows
that H r .UL ; Gm / D 0 for r > 2, where UL D Spec RL;S , because L has no
real primes, and so H r .UQ ; Gm /.`/ D 0 for all r > 2:
In the case F D Z=pZ, p D char.K/, we replace the Kummer sequence
with the Artin-Schreier sequenceW
}
0 ! Z=pZ ! OUQ ! OUQ ! 0; }.a/ D ap a.
As H r .UQ et ; O/ ' H r .UQ Zar ; O/, which is zero for r 1, we see that
H r .UQ ; Z=pZ/ D 0 for r 2:
Finally consider Z. The next lemma shows that H r .UQ ; Z/ is torsion for r >
0, and so from the cohomology sequence of
`
0 ! Z ! Z ! Z=`Z ! 0
and the results in the preceding three paragraphs, we can deduce that
H r .UQ ; Z/.`/ D 0 for r > 0 if ` D char.K/ or ` is invertible on U . 2
L EMMA 2.10 Let Y be a normal Noetherian scheme Y , and let F be a constant

sheaf on Y . For any r > 0, the cohomology group H r .Y; F / is torsion, and so it
is zero if F is uniquely divisible.
P ROOF. We may assume that Y is connected. Let gW ,! Y be its generic

point. Then g g F ' F and the stalks of Rr g .g F /, being Galois coho-
mology groups, are torsion (see Milne 1980, II 3.7, III 1.15). Therefore, if F is
constant and uniquely divisible, then Rr g .g F / D 0 for r > 0, and the Leray
spectral sequence shows that H r .Y; F / ' H r . ; g F / D 0 for r > 0. Now let
F be constant. In proving H r .Y; F / is torsion, we may assume F to be torsion
free. For such a sheaf, the cohomology sequence of
0 ! F ! F ˝ Q ! .F ˝ Q/=F ! 0
shows that H r1 .Y; .F ˝ Q/=F / maps onto H r .Y; F / for r 1 because F ˝ Q
is uniquely divisible. This completes the proof as .F ˝ Q/=F is torsion. 2
C OROLLARY 2.11 Let U be an open subscheme of X , and let S denote the set
of primes of K not corresponding to a point of U .
L
(a) For all r < 0, Hcr .U; Z/ v real H r1 .Kv ; Z/; in particular, Hcr .U; Z/ D
0 if r is even and < 0:
(b) There is an exact sequence
L
0 ! Hc0 .U; Z/ ! Z ! v2S H
0
.Kv ; Z/ ! Hc1 .X; Z/ ! 0:
If S contains at least one nonarchimedean prime, then Hc0 .U; Z/ D 0:

(c) There is an exact sequence
L
0 ! Hc2 .U; Z/ ! H 2 .U; Z/ ! v2S H
2
.Kv ; Z/ !
Hc3 .U; Z/ !H 3 .U; Z/ ! 0:
For all primes ` that are invertible on U or equal the characteristic of K ,

L
0 ! Hc2 .U; Z/.`/ ! H 2 .GS ; Z/.`/ ! v2S H 2 .Kv ; Z/.`/ !
Hc3 .U; Z/.`/ ! H 3 .GS ; Z/.`/ ! 0:
(d) For all r 4 , Hcr .U; Z/ D 0:
P ROOF. All the statements follow from the exact sequence

L
! Hcr .U; Z/ ! H r .U; Z/ ! v2S H r .Kv ; Z/ ! .
For r < 0, H r .U; Z/ D 0, and so the sequence gives isomorphisms

L r1 '
v2S H .Kv ; Z/ ! Hcr .U; Z/:
L L
For r ¤ 0; 1; 2, v2S H r .Kv ; Z/ D v real H r .Kv ; Z/. Since H r .R; Z/ D 0
for odd r, these calculations prove (a).
As H 0 .U; Z/ D Z and H 1 .U; Z/ D Homcts .1 .U; /; Z/ D 0, we have an
exact sequence
L
0 ! Hc0 .U; Z/ ! Z ! v2S H 0 .Kv ; Z/ ! Hc1 .U; Z/ ! 0:
When S contains a nonarchimedean prime, the middle map is injective, and so

this proves (b).
The first part of (c) is obvious from the fact that
H 1 .Kv ; Z/ D 0 D H 3 .Kv ; Z/
for all v, and the proposition allows us to replace H r .U; Z/.`/ with H r .GS ; Z/.`/
for the particular `:
For (d), we begin by showing that Hcr .U; Z/.`/ D 0 for r 4 when ` is a
prime that is invertible on U . Consider the diagram
L
H r1 .GS ; Q=Z/ ! v real H
r1 .K ; Q=Z/
v
? ?
? ?
y y
L
H r .GS ; Z/ ! v real H .Kv ; Z/:
r
The vertical arrows (boundary maps) are isomorphisms for r 2 because Q is

uniquely divisible, and Theorem I 4.10c shows that the top arrow is an isomor-
phism on the `-primary components for r 4 if ` is invertible on U . Therefore
the maps L
H r .GS ; Z/.`/ ! v real H r .Kv ; Z/.`/
are isomorphisms for r 4 and all ` that are invertible on U . As H 3 .R; Z/ D 0,

this proves that Hcr .U; Z/.`/ D 0 for r 4 and such `. Since Hcr .U; F / D
Hcr .U Œ1=`; F / for any r 4, this completes the proof except for the p-primary
component in characteristic p. We may assume that U is affine. The cohomology
sequence of
}
0 ! Z=pZ ! OU ! OU ! 0
shows that H r .U; Z=pZ/ D 0 for r 2. Therefore H r .U; Z/.p/ D 0 for r 3,
and this implies that Hcr .U; Z/.p/ D 0 for r 4. 2
R EMARK 2.12 It has been conjectured that scd` .GS / D L2 for all primes ` that
are invertible in RK;S . This would imply that the map v2S H 2 .Kv ; Z/.`/ !
Hc3 .U; Z/.`/ in (2.11c) is surjective on the `-primary components for such `:
Let F be a constructible sheaf on U such that mF D 0 for some m that is
invertible on U . We shall see that the groups H r .U; F / and Hcr .U; F / are all
finite, and so it makes sense to define
ŒH 0 .U; F /ŒH 2 .U; F / ŒHc0 .U; F /ŒHc2 .U; F /

.U; F / D ; c .U; F / D :
ŒH 1 .U; F /ŒH 3 .U; F / ŒHc1 .U; F /ŒHc3 .U; F /
T HEOREM 2.13 Let F be a constructible sheaf on U such that mF D 0 for some

m that is invertible on U:
(a) The groups H r .U; F / are finite, and
Y ŒF .Kv /
.U; F / D :
ŒH 0 .K v ; F /jŒF .Ks /jv
v arch
(b) The groups Hcr .U; F / are finite, and

Y
c .U; F / D ŒF .Kv /:
v arch
P ROOF. (a) Choose an open affine subscheme V of U such that F jV is locally

constant. Theorem 2.9 shows that H r .V; F / ' H r .GS ; M / for all r, where S is
the set of primes of K not in V and M is the GS module corresponding to F jV .
Therefore Theorem (I 5.1) shows that
Y
.V; F jV /ŒH 3 .V; F jV / D .GS ; M / D ŒH 0 .Gv ; M /=jŒM jv .
v arch
' Q
As H 3 .V; F jV / ! v arch H 3 .Kv ; M / (by I 4.10c), and the groups H r .Kv ; M /
for a fixed archimedean prime v all have the same order (recall that they are Tate
cohomology groups), this proves the result for F jV , and it remains to show that
.U; F / D .V; F jV /. The sequence
L
! v2U XV Hvr .Ovh ; F / ! H r .U; F / ! H r .V; F / !
Q
shows that .U; F / D .V; F jV / v .Ohv ; F /, and the sequence
! Hvr .Ovh ; F / ! H r .Ovh ; F / ! H r .Kv ; F / !
shows that v .Ohv ; F / D .Ohv ; F /.Kv ; F /1 . But F .Kv;s / has order prime to
the residue characteristic of K, and so (I 2.8) shows that .Kv ; F / D 1. Moreover
(see 1.1) H r .Ohv ; F / D H r .gv ; F .Ohv // and F .Ohv / is finite, and so it is obvious
that .Ohv ; F / D 1 (see Serre 1962, XIII1).
(b) The sequence (2.3d)
L
! Hcr .V; F jV / ! Hcr .U; F / ! v2U XV H r .v; i F / !
shows that c .U; F / D c .V; F jV / for any open subscheme V of U , and so we

can assume that U ¤ X and that F is locally constant. There is an exact sequence
Q Q
0 ! v arch H 1 .Kv ; F / ! Hc0 .U; F / ! H 0 .U; F / ! v…U H 0 .Kv ; F / !
Q
! Hc3 .U; F / ! H 3 .U; F / ! v arch H 3 .Kv ; F / ! 0
Q
because (see I 4.10c) H 3 .U; F / ! v arch H 3 .Kv ; F / is surjective (in fact, an
isomorphism). As the groups H r .Kv ; F / for v archimedean all have the same
order,
Q Q
c .U; F / D .U; F / v2XXU .Kv ; F /1 v arch ŒH 0 .Kv ; F /:
According to (I 2.8), .Kv ; F / D jŒF .Ks /jv , and so

Q Q
c .U; F / D v arch ŒF .Kv / v…U jŒF .Ks /j1
v :
Q
But jŒF .Ks /jv D 1 for v 2 U , and so v…U jŒF .Ks /jv D 1 in virtue of the
product formula, and so we obtain the formula. 2
R EMARK 2.14 (a) Let F be a locally constant sheaf on U with mF D 0 for

some m that is invertible on U . In the next section, we shall show that H r .U; F /
is dual to Hc3r .U; F D / for all r. This implies that .U; F /c .U; F D / D 1. If
we let M be the GS -module module corresponding to F , then (2.13) shows that
Q ŒH 0 .Gv ; M /ŒH 0 .Gv ; M D /
.U; F /c .U; F D / D v arch ;
jŒM jv ŒH 0 .Kv ; M /
which (I 2.13c) shows to be 1. Thus our results are consistent.

(b) Assume that U ¤ X and that F is locally constant. Then H 0 .U; F / ,!
Q 1 .U; F / D 0. Therefore
Q 1 '
0
v H .Kv ; F / and, of course, H H .Kv ; F / !
Hc0 .U; F /, and so (2.13b) becomes in this case
ŒHc2 .U; F / Y ŒF .Kv /

1 3
D :
ŒHc .U; F /ŒHc .U; F / ŒH 0 .Kv ; F /
v arch
N OTES So far as I know, K. Kato was the first to suggest defining cohomology groups
“with compact support” fitting into an exact sequence
Q
! Hcr .X; F / ! H r .X; F / ! v real H r .Kv ; Fv / !
(letter to Tate, about 1973). Our definition differs from his, but it gives the same groups.
3 Global results: the main theorem

Statements
We continue with the notations of the last section. From (2.6) (and its proof) we
'
know that there are trace maps Hc3 .U; Gm / ! Q=Z such that
(a) for any V U ,
'
Hc3 .V; Gm / ! Q=Z
?
?
y
'
Hc3 .U; Gm / ! Q=Z
commutes;
(b) for any v … U ,
inv
Br.Kv / ! Q=Z
?
?
y
'
Hc3 .U; Gm / ! Q=Z
commutes.
On combining the pairings
ExtrU .F; Gm / Hc3r .U; F / ! Hc3 .U; Gm /
with this trace map, we obtain maps
˛ r .U; F /W ExtrU .F; Gm / ! Hc3r .U; F / :

3. GLOBAL RESULTS: THE MAIN THEOREM 177
T HEOREM 3.1 Let F be a Z-constructible sheaf on an open subscheme U of X:

(a) For r D 0; 1, ExtrU .F; Gm / is finitely generated and ˛ r .U; F / defines
isomorphisms
ExtrU .F; Gm /^ ! Hc3r .U; F /
where ^ denotes the completion for the topology of subgroups of finite
index.
For r 2, ExtrU .F; Gm / is a torsion group of cofinite-type, and ˛ r .U; F /
is an isomorphism.
(b) If F is constructible, then
ExtrU .F; Gm / Hc3r .U; F / ! Hc3 .U; Gm /
is a nondegenerate pairing of finite groups for all r 2 Z:
Note that (b) implies that H r .U; F / is finite if F is a constructible sheaf on

U without char.K/-torsion because (2.3a) and (I 2.1) show that then H r .U; F /
differs from Hcr .U; F / by a finite group.
Before beginning the proof of the theorem, we list some consequences.
C OROLLARY 3.2 For any constructible sheaf F on an open subscheme j W U !

X of X and prime number `, there is a canonical nondegenerate pairing of finite
groups
ExtrU .F; Gm /.`/H 3r .X; jŠ F /.`/ ! H 3 .X; jŠ Gm /.`/ ' .Q=Z/.`/; r 2 Z;
except when ` D 2 and K is number field with a real prime.
P ROOF. According to (2.3d), Hcr .U; F / ' Hcr .X; jŠ F / for any sheaf F on U
(not necessarily constructible), and it is clear from (2.3a) that Hcr .X; jŠ F / differs
from H r .X; jŠ F / by at most a group killed by 2, and that it differs not at all if K
has no real embedding. Thus, the statement follows immediately from (b) of the
theorem. 2
C OROLLARY 3.3 Let F be a constructible sheaf on U such that mF D 0 for

some m that is invertible on U , and let F D D RHom.F; Gm / (an object of the
derived category of S.Uet /).
(a) If F is locally constant, then H r .F D / D 0 for r > 0; thus in this case F D
can be identified with the sheaf Hom.F; Gm /:
(b) There is a canonical nondegenerate pairing of finite groups
H r .U; F D / Hc3r .U; F / ! H 3 .U; Gm / ' Q=Z; r 2 Z.

P ROOF. Part (a) can be proved by the argument in the proof of Corollary 1.10(a).
Part (b) is obvious from the theorem, because Hr .U; F D / ' ExtrU .F; Gm /. 2
Define3
D r .U; F / D Im.Hcr .U; F / ! H r .U; F //:
C OROLLARY 3.4 Let F be a locally constant constructible sheaf on U such that

mF D F for some m invertible on U . Then there is a nondegenerate pairing of
finite groups
D r .U; F / D 3r .U; F D / ! Q=Z
for all r 2 Z:
P ROOF. From (2.3a), there is an exact sequence

L
0 ! D r .U; F / ! H r .U; F / ! v…U H r .Kv ; F /,
and (3.3) and (I 2.3) show that the dual of this is an exact sequence
L
v…U H
2r
.Kv ; F D / ! Hc3r .U; F D / ! D r .U; F / ! 0:
But this second sequence identifies D r .U; F / with D 3r .U; F D / (apply (2.3a)
again). 2
The proof of the theorem is rather long and intricate. In (3.5 — 3.8) we
show that it suffices to prove the theorem with U replaced by an open subset.
Proposition 3.9 and Corollary 3.10 relate the theorem on U to the theorem on U 0
for some finite covering of U . In Lemma 3.12 it is shown that the groups vanish
for large r when K has no real primes, and hence proves the theorem for such K
and r. Lemma 3.13 allows us to assume that K has no real primes. In (3.14 —
3.17) the theorem is proved by an induction argument for constructible sheaves,
and we then deduce it for all Z-constructible sheaves.
The proof
Throughout the proof, U will be an open subscheme of the scheme X . We
˛ r .U; F / equal to the map ExtrU .F; Gm /^ ! H 3r .U; F / induced by
set b
˛ .U; F / when r D 0; 1, and equal to ˛ r .U; F / otherwise.
r
L EMMA 3.5 Theorem 3.1 is true if F has support on a proper closed subset of
U:
3 The groups D r .U; F / are the intersection cohomology groups of F for the middle perversity.
P ROOF. We can assume that our sheaf is of the form i F where i is the inclusion
of a single closed point v into U . According to (2.3c),
Hcr .U; i F / ' H r .v; F /:

L
0 ! G m ! g G m ! u2U 0 iu Z !0

L
! ExtrU .i F; Gm / ! ExtrU .i F; g Gm / ! u2U 0 ExtrU .i F; iu Z/ ! :
As we observed in the proof of (2.1), Rs g Gm D 0 for s 1, and so4
ExtrU .i F; g Gm / ' Extr .i F j ; Gm /;
which is zero for all r because i F j D 0. Moreover (0.1c) shows that

ExtrU .i F; iu Z/ ' Extru .iu iv F; Z/, which equals 0 unless u D v, in which
case it equals Extrv .F; Z/. Therefore the sequence gives isomorphisms
'
Extr1 r
v .F; Z/ ! ExtU .i F; Gm / (5)
for all r. Let M be the gv -module corresponding to F . Then we have a commu-

tative diagram
ExtrU .i F; Gm / Hc3r .U; i F / ! Hc3 .U; Gm / ' Q=Z

"' k k
Extr1
gv .M; Z/ H 3r .gv ; M / ! H 2 .gv ; Z/ ' Q=Z
and so the theorem follows in this case from (I 1.10).

[There is an alternative proof of (5). One can show (as in 1.7b) that Rr i Š Gm '
Z for r D 1 and is zero otherwise. The spectral sequence (0.1e)
Extrv .F; Rs i Š Gm / H) ExtrCs

U .i F; Gm /
'
now yields isomorphisms Extr1 r
v .F; Z/ ! ExtU .i F; Gm /:] 2
L EMMA 3.6 For any Z-constructible sheaf on U , the groups ExtrU .F; Gm / are
finitely generated for r D 0; 1, torsion of cofinite-type for r D 2; 3, and finite for
r > 3. If F is constructible, all the groups are finite.
4 The functor g has an exact left adjoint g , and so it is left exact and preserves injectives (cf.

(0.1b)).
P ROOF. Note that ExtrU .Z; Gm / ' H r .U; Gm /, and so for F D Z the values of
ExtrU .Z; Gm / can be read off from (2.1). In particular, the lemma is true for Z,
hence for Z=nZ, and so for all constant Z-constructible sheaves F .
Next suppose that F is locally constant and Z-constructible. It then becomes
constant on some finite Galois covering W U 0 ! U with Galois group G, say,
and the spectral sequence (see 0.2)
H r .G; ExtsU 0 .F jU 0 ; Gm // H) ExtrCs

U .F; Gm /;
shows that ExtrU .F; Gm / differs from H 0 .G; ExtrU 0 .F jU 0 ; Gm // by a finite group.
Therefore the lemma for F jU 0 implies it for F .
Finally, let F be an arbitrary Z-constructible sheaf, and let V be an open
subset of U on which F is locally constant. Write j and i for the inclusions of V
and its complement into U . The Ext sequence of
0 ! jŠ j F ! F ! i i F ! 0

! Extr1 r r
U XV .i F; Z/ ! ExtU .F; Gm / ! ExtV .F; Gm / !
(use (5) and (0.1a)). As Extr1

U XV .i F; Z/ is finitely generated for r D 1, finite
for r D 2, torsion of cofinite-type for r D 3, and 0 for all other values of r (see I
1.10), the lemma follows. 2
L EMMA 3.7 Let

0 ! F 0 ! F ! F 00 ! 0
be an exact sequence of Z-constructible sheaves on U . If (3.1) holds for two out
of the three sheaves F 0 , F , and F 00 , then it also holds for the third.
P ROOF. Because Ext1U .F 0 ; Gm / is finitely generated, its image in the torsion

group Ext2U .F 00 ; Gm / is finite. Therefore the sequence
! Extr .F 00 ; Gm / ! Extr .F; Gm / ! Extr .F 0 ; Gm / !
remains exact after the first six terms have been replaced by their completions.
On the other hand, Hom.; Q=Z/ is exact because Q=Z is injective, and so the
lemma follows from the five-lemma. 2
L EMMA 3.8 Let V be a nonempty open subscheme of U , and let F be a Z-

constructible sheaf on U ; the theorem is true for F on U if and only if it is true
for the restriction of F to V .
P ROOF. Write j for the open immersion V ,! U and i for the complementary
closed immersion U X V ,! U . Then (see (0.1a))
ExtrU .jŠ F jV; Gm / ' ExtrV .F jV; Gm /;
and (see (2.3d))

Hcr .U; jŠ F jV / ' Hcr .V; F jV /:
It follows that b
˛ r .U; jŠ F jV / can be identified with b ˛ r .V; F jV /. Therefore the
theorem is true for jŠ .F jV / on U if and only if it is true for F jV on V . Now (3.7,
3.5) and the exact sequence
0 ! jŠ j F ! F ! i i F ! 0
show that the theorem is true for F on U if and only if it is true for jŠ .F jV / on
U: 2
The lemma shows that it suffices to prove Theorem 3.1 for locally constant
sheaves and “small” U .
L EMMA 3.9 Let W U 0 ! U be the normalization of U in a finite Galois exten-

sion K 0 of K .
(a) There is a canonical norm map NmW Gm;U 0 ! Gm;U :
(b) For every Z-constructible sheaf F on U 0 , the composite
Nm
N W ExtrU 0 .F; Gm / ! ExtrU . F; Gm / ! ExtrU . F; Gm /
is an isomorphism.
P ROOF. (a) Let V ! U be étale. Then V is an open subset of the normaliza-

tion of U in some finite separable K-algebra L. By definition .V; Gm / '
.V 0 ; O 0 0 0 0
V 0 / where V Ddf U U V . As V is étale over U , it is normal, and as
it is finite over V , it must be the normalization of V in the finite Galois L-algebra
K 0 ˝K L:
U0 V 0 U 0 U V K 0 ! K 0 ˝K L
? ? x x
?
yfinite
? ? ?
y
0
? ?
étale
U V K ! L:
Consequently, the norm map K 0 ˝K L ! L induces a map .V; Gm / !
.V; Gm /, and for varying V these maps define a map of sheaves NmW Gm;U 0 !
Gm;U .
(b) Let j W V ! U 0 be an open immersion such that V Ddf jV is étale.

Then Milne 1980, V 1.13, shows that the map
ExtrV .j F; Gm / ! ExtrU . jŠ j F; Gm /
defined by the adjunction map V Š V Gm ! Gm is an isomorphism for all r.
The composite of this with
ExtrU .jŠ j F; Gm / ' ExtrV .j F; Gm /
(see 0.1a) is the map in (b). Therefore, (b) is true for the sheaf jŠ j F .
For a sheaf of the form iv F , v 2 U , (b) again follows from (ibid., V 1.13)
because the sequence of maps can be identified with
Tr
Extr1
1 .v/
.F; Z/ ! Extr1 r1
v . F; Z/ ! Extv . F; Z/
(for the trace map Z ! Z, see ibid. V 1.12; for the identification, see (5)).
For the general case, apply the five-lemma to the diagram obtained from the
exact sequence (1)
0 ! jŠ j F ! F ! i i F ! 0: 2
L EMMA 3.10 Let W U 0 ! U be as in Lemma 3.9. For any Z-constructible

sheaf F 0 on U 0 , b
˛ r .U 0 ; F / is an isomorphism if and only if b
˛ r .U; F / is an
isomorphism.
P ROOF. From the norm map (3.8) GmU 0 ! Gm , we obtain a map Nm,
.2:3e/
Hc3 .U 0 ; Gm / Hc3 .U; Gm / ! Hc3 .U; Gm /.
'
For any w 7! v … U , the diagram

Br.Kw0 /
! Hc3 .U 0 ; Gm /
? ?
? ?
yNm yNm
Br.Kv / ! Hc3 .U; Gm /
commutes. The left hand arrow commutes with the invariant maps, and so the
right hand arrow commutes with the trace maps. The diagram
ExtrU 0 .F; Gm / Hc3r .U 0 ; F / ! Hc3 .U 0 ; Gm /

' #N ' ".2:3e/ #Nm
ExtrU . F; Gm / Hc3r .U; F / ! Hc3 .U; Gm /
commutes, and the lemma follows. 2
R EMARK 3.11 In Lemmas 3.9 and 3.10, it is only necessary to assume that K 0
is separable over K.
L EMMA 3.12 (a) If F is constructible, then Hcr .U; F / is zero for r > 3, and if
F is Z-constructible, then it is zero for r > 4.
(b) If F is constructible and K has no real primes, then ExtrU .F; Gm / D 0
for r > 4.
P ROOF. (a) Let F be constructible. According to (2.3d), we can replace U by an

open subset, and hence assume that F is locally constant and, in the number field
case, that mF D 0 for some integer
L m that is invertible on U . We have to show
(see 2.3a) that H r .U; F / ! v real H .Kv ; F / is an isomorphism for r 3.
r
But (2.9) identifies this with the map

L
H r .GS ; M / ! v real H r .Kv ; M /; M D F ; (6)
and (I 4.10c) shows that (6) is an isomorphism for r 3 except possibly when
K is a function field and the order of M is divisible by p. In the last case we can
assume that F is killed by some power of p and have to show that H r .U; F / D 0
for r > 3. From the cohomology sequence of
}
0 ! Z=pZ ! OU ! OU ! 0
we see that H r .U; Z=pZ/ D 0 for r 2 (because H r .Uet ; OU / D H r .UZar ; OU / D

0 for r > 1). In general there will be a finite étale covering W U 0 ! U of de-
gree d prime to p such that F jU 0 has a composition series whose quotients are
isomorphic to the constant sheaf Z=pZ (apply Serre 1962, IX, Thm 3, Thm 2).
Then H r .U 0 ; F jU 0 / D 0 for r 2, and as the composite
trace
H r .U; F / ! H r .U 0 ; F jU 0 / ! H r .U; F /
is multiplication by d (see Milne 1980, V.1.12), this proves that H r .U; F / D 0

for r 2.
Now let F be Z-constructible. Then Ftors is constructible, and so it suffices
to prove the result for F=Ftors : we can assume that F is torsion-free. Then
Hcr1 .U; F=mF / Hcr .U; F /m ; Hcr1 .U; F=mF / D 0 for r > 4;
and so it remains to show that Hcr .U; F / is torsion for r > 4. Again, (2.3d)
allows us to assume F is locally constant. Proposition 2.3a shows that Hcr .U; F /
differs from H r .U; F / by a torsion group for r > 1, and we saw in (2.9) that
H r .U; F / is torsion when r > 0.
(b) Because K has no real primes, H r .U; F / D Hcr .U; F / D 0 for r >
3. If F has support on a closed subscheme Z, the lemma is obvious from the
isomorphism
'
Extr1 r
Z .F; Z/ ! Ext .i F; Gm /
of (3.5). As usual, this allows us to assume that F is locally constant. Then

ExtUr .F; Gm / D 0 for r > 1 (see the proof of 1.10a), and for r D 0; 1, it
is torsion, and is therefore a direct limit of constructible sheaves (0.9). Hence
H r .U; ExtUs .F; Gm // D 0 for r > 3, and so Extr .F; Gm / D 0 for r > 4: 2
L EMMA 3.13 Assume that b ˛ r .X; Z/ is an isomorphism for all r whenever K has
no real primes. Then Theorem 3.1 is true.
P ROOF. When K has no real primes, the assumption implies that Theorem 3.1 is
true for constant sheaves on X , and (3.8) then implies that it is true for constant
sheaves on any open U X .
According to Lemma 3.8, it suffices to prove Theorem 3.1 for pairs .U; F /
with F locally constant and with 2 invertible on U in the number field case. We
prove that b ˛ r .U; F / is an isomorphism in this case by induction on r. Note that
Lemma 3.12a implies that b ˛ r .U; F / is an isomorphism when r < 1 — assume
it to be an isomorphism when r < r0 . For a pair .U; F / as above, there exists a
finite étale covering W U 0 ! U such that U 0 is the normalization of U in field K 0
with no real primes and F becomes constant on U 0 . Let F D F . The trace
map
P (Milne 1980, V.1.12) F ! F is surjective (on stalks it is just summation,
W Fv ! Fv /, and we write F 0 for its kernel. From the commutative diagram
d
ExtrU0 1 .F ; Gm / !
ExtrU0 1 .F 0 ; Gm / !
ExtrU0 .F; Gm / !
ExtrU0 .F ; Gm /
? ? ? ?
? ? ? ?
y y y 3:10y
H 4r0 .U; F / !
H 4r0 .U; F 0 / !
H 3r0 .U; F / !
H 3r0 .U; F /
we see that b ˛ r0 .U; F /W Extr0 .F; Gm / ! H 3r0 .U; F / is injective. (For r D

0; 1, it is necessary to replace the groups on the top row with their completions;
see the proof of (3.7).) Since F 0 is also locally constant, b
˛ r0 .U; F 0 / is also injec-
tive, and the five-lemma implies that b r0
˛ .U; F / is an isomorphism. 2
For a constructible sheaf F , we define ˇ r .U; F /W Hcr .U; F / ! Ext3r

U .F; Gm /
r
to be the dual of ˛ .U; F /:
L EMMA 3.14 For any Z-constructible sheaf F on U , there is a finite surjective
map 1 W U1 ! U , a finite map 2 W U2 ! U with finite image, constant Z-
constructible sheaves Fi on Ui , and an injective map F ! ˚i i Fi :
P ROOF. Let V be an open subset of U such that F jV is locally constant. Then

there is a finite extension K 0 of K such that the normalization W V 0 ! V of V
in K 0 is étale over V and F jV 0 is constant. Let 1 W U1 ! U be the normalization
of U in K 0 , and let F1 be the constant sheaf on U1 corresponding to the group
.V 0 ; F jV 0 /. Then the canonical map F jV ! F jV 0 extends to a map ˛W F !
1 F1 whose kernel has support on U V . Now take U2 to be an étale covering
of U V on which the inverse image of F on V U becomes a constant sheaf,
and take F2 to be the direct image of this constant sheaf. 2
Note that Lemma 3.13 shows that it suffices to prove Theorem 3.1 under the
assumption that K has no real primes. From now until the end of the proof of the
theorem we shall make this assumption.
L EMMA 3.15 (a) Let r0 be an integer 1. If for all K , all constructible sheaves
F on X , and all r < r0 , ˇ r .X; F / is an isomorphism, then ˇ r0 .X; F / is injective.
(b) Assume that for all K , all constructible sheaves F on X , and all r < r0 ,
ˇ .X; F / is an isomorphism; further assume that ˇ r0 .X; Z=mZ/ is an isomor-
r
phism whenever m .K/ D m .Ks /. Then ˛ r0 .X; F / is an isomorphism for all

X and all constructible sheaves F:
P ROOF. (a) Let F be a constructible sheaf on some X , and let c 2 H r0 .X; F /.

There exists an embedding F ! I of F into a torsion flabby sheaf I on X . Ac-
cording to (0.9), I is a direct limit of constructible sheaves. As H r0 .X; I / D 0,
and cohomology commutes with direct limits, this implies that there is a con-
structible sheaf F on X and an embedding F ,! F such that c maps to zero in
H r0 .X; F /. Let Q be the cokernel of F ! F . Then Q is constructible, and a
chase in the diagram
H r0 1 .X; F / !
H r0 1 .X; Q/ !
H r0 .X; F / !
H r0 .X; F /
? ? ? ?
? ? ? ?
y y y y
Ext4r
X
0
.F ; Gm / !
Ext4r
X
0
.Q; Gm / !
Ext3r
X
0
.F; Gm / !
Ext3r
X
0
.F ; Gm /
shows that ˇ r0 .c/ ¤ 0. Since the argument works for all c, this shows that
ˇ r0 .X; F / is injective.
(b) Let F be a constructible sheaf on X . For a suitably small open subset U of
X , there will exist a finite Galois extension K 0 of K such that the normalization
U 0 of U in K 0 is étale over U , F jU 0 is constant, and m .K/ D m .Ks / for some
m with mF D 0. In the construction of the preceding lemma, we can take U1
to be the normalization of X in K 0 . Let F D 1 F1 ˚ 2 F2 ; then (3.10) and
(3.5) show respectively that ˇ r0 .X; 1 F / and ˇ r0 .X; 2 F / are isomorphisms.
The map F ! F is injective, and we can construct a diagram similar to the

above, except that now we know that ˇ r0 .X; F / is an isomorphism. Therefore
ˇ r0 .X; F / is injective, and as Q is constructible we have also that ˇ r0 .X; Q/ is
injective. The five-lemma now shows that ˇ r0 .U; F / is an isomorphism. 2
L EMMA 3.16 Theorem 3.1 is true for all constructible sheaves F on X:
P ROOF. We prove ˇ r .X; F / is an isomorphism by induction on r. For r < 0 it

is an isomorphism by (3.12).
To compute the group ExtrX .Z=mZ; Gm /, we use the exact sequence
m
! ExtrX .Z=mZ; Gm / ! H r .X; Gm / ! H r .X; Gm / ! :
By definition H 0 .X; Z=mZ/ D Z=mZ, and it follows from (2.2b) that
Ext3 .Z=mZ; Gm / ' m1 Z=Z:
The pairing is the obvious one, and so ˇ 0 .X; Z=mZ/ is an isomorphism. Now
(3.15b) implies that ˇ 0 .X; F / is an isomorphism for all F:
Lemma 3.15a shows that ˇ 1 .X; F / is always injective. The order of
H 1 .X; Z=mZ/ is equal to the degree of the maximal unramified abelian ex-
tension of K of exponent m. By class field theory, this is also the order of
Pic.X /.m/ ' ŒExt2X .Z=mZ; Gm /, and so ˇ 1 .X; Z=mZ/ is an isomorphism for
all X . It follows that ˇ 1 .X; F / is an isomorphism for all X and F .
Lemma 3.15a again shows that ˇ 2 .X; F / is always injective. To complete the
proof,5 it remains to show (by 3.15) that ˇ r .X; Z=mZ/ is an isomorphism for all
r 2 when m .K/ D m .Ks /.
Fix a K (hence an X ) such that m .K/ D m .Ks / (and, of course, K has
no real primes). Initially assume m is prime to the characteristic of K. Choose a
dense open U X on which m is invertible, and let iW X X U ,! X be its closed
complement. From the exact sequence (1), p140, we obtain a diagram
! Hcr .U; Z=mZ/ ! H r .X; Z=mZ/ ! H r .X; i .Z=mZ// !
? ? ?
? r ? r ? r
yˇ .U;Z=mZ/ yˇ .X;Z=mZ/ yˇ .X;i .Z=mZ//
! ! ! !
From it, we see that ˇ r .U; Z=mZ/
is an isomorphism for r 1 and an injection
for r D 2. For r D 3, ˇ .U; Z=mZ/ arises from a pairing
r
HomU .Z=mZ; Gm / Hc3 .U; Z=mZ/ ! Hc3 .U; Gm /:

5 At this point in the original, I effectively assumed that ˇ 2 .X; F / is an isomorphism (see the
diagram p288). I thank Joël Riou for pointing this out to me.

Any isomorphism Z=mZ ! m Gm induces an isomorphism

Hcr .U; Z=mZ/ ! Hcr .U; m /Hcr .U; Gm /

1
m Z=Z Q=Z
from which it follows that ˇ 3 .U; Z=mZ/ is an isomorphism. As
Hcr .U; F / D 0 D Ext3r

U .F; Gm / for r > 3;
this shows that ˇ r .U; Z=mZ/ is an isomorphism for all r except possibly for
r D 2, when it is injective. Now (2.13) (cf. 2.14a) shows that Hc2 .U; Z=mZ/ and
Ext1U .Z=mZ; Gm / ' H 1 .U; m / have the same order, and so ˇ 2 .U; Z=mZ/ is
an isomorphism for all r.
It remains to treat the sheaf Z=pZ with p ¤ 0 the characteristic of K. From
the cohomology sequence of
}
0 ! Z=pZ ! OU ! OU ! 0
and the Ext sequence of

p
0 ! Z ! Z ! Z=pZ ! 0
we see that
H r .X; Z=pZ/ D 0 D Ext3r

X .Z=pZ; Gm / for r > 2:
Thus, ˇ r .X; Z=pZ/ is an isomorphism for all r except possibly r D 2, when it is

injective. Using the sequences just mentioned, one can show that H 2 .X; Z=pZ/
and Ext1X .Z=pZ; Gm / have the same order,6 which completes the proof. 2
We now complete the proof of Theorem 3.1 by proving that b ˛ r .X; Z/ is an

isomorphism for all r (recall that we are assuming K has no real primes). We are
concerned with the maps
˛ r .X; Z/W H r .X; Gm / ! H 3r .X; Z/ ; r ¤ 0; 1;

˛ r .X; Z/W H r .X; Gm / ! H 3r .X; Z/ ; r D 0; 1:
b
6 Alternatively, one can avoid counting by using the original proof, which actually works in this
case.

H r .X; Gm /^ !
0 ! lim ExtrC1 .Z=nZ; Gm / !
lim H rC1 .X; Gm /n !
0
? n ? n ?
? ? ?
y y y
H 3r .X; Z/ !
H 2r .X; Q=Z/ !
H 2r .X; Q/
For 2 r 1 (that is, for r 1/, H 2r .X; Q/ D 0 D H 3r .X; Q/ (see
2.10), and so H 2r .X; Q=Z/ ! H 3r .X; Z/ is an isomorphism. For r 1,
H rC1 .X; Gm / is finitely generated, and so lim H rC1 .X; Gm /n D 0 (see I 0.19).
r
Therefore it is obvious from the diagram that b ˛ .X; Z/ is an isomorphism for
r 1. For r D 3, the map is the obvious isomorphism Q=Z ! Z , and for all
other values of r, both groups are zero.
R EMARK 3.17 7 For a locally Noetherian scheme Y , let Ysm denote the category
of smooth schemes over Y endowed with the étale topology. Let f W Ysm ! Yet
be the morphism of sites defined by the identity map. Then f is exact and pre-
serves injectives (Milne 1980, III 3.1), and it follows that ExtrYsm .f F; F 0 / '
ExtrYet .F; f F / for all sheaves F on Yet , all sheaves F 0 on Ysm , and all r. There-
fore Uet can be replaced by Usm in the above results provided one defines a Z-
constructible sheaf on Ysm to be the inverse image by f of a Z-constructible sheaf
on Yet :
N OTES Corollary 3.2 in the number field case is the original theorem of Artin and Verdier
(announcement in Artin and Verdier 1964). As far as I know, no complete proof of the
theorem has been published before, but Mazur 1973 contains most of the ingredients.
It and the notes of a 1964 seminar by Mazur were sources for this section. We note
that Theorem 3.1 improves the original theorem in three respects: by taking into account
the archimedean primes, it is able to handle the 2-torsion; it includes the function field
case; and it allows the sheaves to be Z-constructible instead of constructible. To my
knowledge, several people have extended the original theorem to the function field case,
but the only published account is in Deninger 1984. Deninger (1986) showed how to
deduce the Z-constructible case from the constructible case. In Zink 1978 there is an
alternative method of obviating the problem with 2-primary components in the original
theorem.
4 Global results: complements

This section is concerned with various improvements of Theorem 3.1. We also
discuss its relation to the theorems in Chapter I. The notations are the same as in
the preceding two sections.
7 This was Remark 3.18 in the original.
4. GLOBAL RESULTS: COMPLEMENTS 189
Sheaves without sections with finite support

Let F be a sheaf on an open subscheme U of X . For any V étale over U , a
section s 2 .V; F / is said to have finite support if sv D 0 for all but finitely
many v 2 V .
P ROPOSITION 4.1 Let F be a Z-constructible sheaf on an open affine subscheme

U of X . If F has no sections with finite support, then Ext1U .F; Gm / and Hc2 .U; F /
are finite, and ˛ 1 .U; F / is an isomorphism.
P ROOF. Note that HomU .Z; Gm / D O U , which is finitely generated, and that
Ext1U .Z; Gm / ' Pic.U /, which is finite (because, in the function field case, it is
a quotient of Pic0 .X //. It follows immediately that Ext1U .F; Gm / is finite if F
is constant. As we observed in (3.14), there is a finite surjective map W U 0 ! U ,
a constant Z-constructible sheaf F 0 on U 0 , and a morphism F ! F 0 whose
kernel has support on a proper closed subset of U . As F has no sections with
finite support, we see that the map must be injective. Let F 00 be its cokernel. In
the exact sequence
Ext1U . F 0 ; Gm / ! Ext1U .F; Gm / ! Ext2 .F 00 ; Gm /;
Ext1U . F 0 ; Gm / ' Ext1U 0 .F 0 ; Gm / (see 3.9) and so is finite, and Ext2 .F 00 ; Gm /
is torsion, and so Ext1U .F; Gm / (being finitely generated) has finite image in it.
This proves that Ext1U .F; Gm / is finite, and Theorem 3.1 implies that ˛ 1 .X; F /
is an isomorphism. It follows that Hc2 .U; F / is also finite. 2
Nonconstructible sheaves
We say that a sheaf F on U X is countable if F .V / is countable for all V étale
over U . For example, any sheaf defined by a group scheme of finite type over U is
countable. Fix a separable closure K s of K. If F is countable, then there are only
countably many pairs .s; V / with V an open subset of the normalization of U in
a finite subextension of K s and s 2 F .V /. Therefore the construction in (0.9)
expresses F as a countable union of Z-constructible sheaves (of constructible
sheaves if F is torsion).
P ROPOSITION 4.2 Let F be a countable sheaf on an open subscheme U of X ,

and consider the map ˛ r .U; F /W ExtrU .F; Gm / ! H 3r .U; F / .
(a) For r 2, the kernel of ˛ r .U; F / is divisible, and it is uncountable when
nonzero; for r D 0 or r > 4, ˛ r .U; F / is injective.
(b) For r 2, ˛ r .U; F / is surjective.
(c) If F is torsion, then ˛ r .U; F / is an isomorphism for all r .
(d) If U is affine and F has no sections with finite support, then ˛ 2 .U; F / is
an isomorphism and ˛ 1 .U; F / is surjective.
S
P ROOF. Write F as a countable union of Z-constructible subsheaves, F D Fi .
Then (see I 0.21 and I 0.22) there is an exact sequence
0 ! lim .1/ Extr1 r

U .Fi ; Gm / ! ExtU .F; Gm / ! lim ExtrU .Fi ; Gm / ! 0;

and lim.1/ Extr1 .Fi ; Gm / is divisible (and uncountable when nonzero) if each
r1 U
group Ext .Fi ; Gm / is finitely generated, and it is zero if each group
Extr1 .Fi ; Gm / is finite. Theorem 3.1 provides us with a map Extr .Fi ; Gm / !
H r .U; Fi / which is injective for all r and is surjective for r 2; it is an iso-
morphism for any r for which the groups Extr .Fi ; Gm / are finite. On passing to
the inverse limit, we obtain a map lim ExtrU .Fi ; Gm / ! Hc3r .U; F / with the

similar properties. The proposition is now obvious from (3.1) and (4.1). 2
C OROLLARY 4.3 Let G be a separated group scheme of finite type over an open
affine subscheme U of X . Then ˛ 2 .U; G/W Ext2U .G; Gm / ! Hc1 .U; G/ is an
isomorphism. If G defines a torsion sheaf, then ˛ r .U; G/ is an isomorphism for
all r:
P ROOF. If a section s of G over V agrees with the zero section on an open subset
of V , then it agrees on the whole of V (because G is separated over V /. Thus G
(when regarded as a sheaf) has no sections with support on a finite subscheme,
and the corollary results immediately from part (d) of the proposition. 2
E XAMPLE 4.4 In particular,

'
Ext2U .A; Gm / ! Hc1 .U; A/ , A a semi-abelian scheme over U;
'
Ext2U .Gm ; Gm / ! Hc1 .U; Gm / , and
'
ExtrU .Ga ; Gm / ! H 3r .U; Ga / for all r when char.K/ D p ¤ 0:
(In fact the groups ExtrU .G; Gm /, computed for the small étale site, seem to be
rather pathological. For example, if k is a finite field, then Homk .Gm ; Gm /
Gal.k s =k/ D bZ.)
E XERCISE 4.5 (a) Show that there are only countably many Z-constructible sheaves
on U . (HintW Use Hermite’s theorem.)
(b) Show that there are uncountably
L many countable sheaves on Spec Z.
(HintW Consider sheaves of the form p prime ip Fp :/
Tori
We investigate the duality theorem when F is replaced by a torus. By a torus over
a scheme Y , we mean a group scheme that becomes isomorphic to a product of
copies of Gm on a finite étale covering of Y . The sheaf of characters X .T / of
T is the sheaf V 7! HomV .T; Gm / (homomorphisms as group schemes). It is a
locally constant Z-constructible sheaf. In the next theorem, ^ denotes completion
relative to the topology of subgroups of finite index.
T HEOREM 4.6 Let T be a torus on an open subscheme U of X .
(a) The cup-product pairing
H r .U; T / Hc3r .U; X .T // ! Hc3 .U; Gm / ' Q=Z
induces isomorphisms
H r .U; T /^ ! Hc3r .U; X .T // for r D 0; 1;
H r .U; T / ! Hc3r .U; X .T // for r 2:
If U is affine, then H 1 .U; T / is finite.
(b) Assume that K is a number field. The cup-product pairing
H r .U; X .T // Hc3r .U; T / ! Hc3 .U; Gm / ' Q=Z
H r .U; X .T //^ ! Hc3r .U; T / for r D 0
H r .U; X .T // ! Hc3r .U; T / for r 1:
P ROOF. (a) The sheaf X .T / is locally isomorphic to Zdim.T / , and so

ExtUr .X .T /; Gm / is locally isomorphic to the sheaf associated with the presheaf
V 7! H r .V; Gm /dim.T / . It is therefore zero for r > 0. As ExtU0 .X .T /; Gm / D
HomU .X .T /; Gm / D T , the spectral sequence
H r .U; ExtUs .X .T /; Gm / H) ExtrCs
U .X .T /; Gm /

gives isomorphisms H r .U; T / ! ExtrU .X .T /; Gm / for all r. Thus (a) fol-
lows from (3.1) and (4.1).
(b) Consider the diagram
Q r1 Q r
H .Kv ; X .T // !
Hcr .U; X .T // !
H r .U; X .T // !
H .Kv ; X .T //
v…U v…U
? ? ? ?
? ? ? ?
y y y y
Q Q
H 3r .Kv ; T / H 3r .U; T / !
! Hc3r .U; T / !
H 2r .Kv ; T // :
v…U v…U
Replace the groups H 0 and 1

Q Hr c .U; X .T // withQtheir completions. Then (I
2.4) shows that the maps H .Kv ; X .T // ! H 2r .Kv ; T / are isomor-
0
phisms (provided one completes H .Kv ; X .T ///, and (a) shows that the maps
Hcr .U; X .T // ! H 3r .U; T / are isomorphisms (provided one completes
Hcr .U; X .T // for r D 0; 1). Now the five-lemma shows that H r .U; X .T // !
Hc3r .U; T / is an isomorphism for all r (provided one completes H 0 .U; X .T //).2
C OROLLARY 4.7 Assume K is a number field. There are canonical isomor-

phisms
D r .U; X .T // ! D 3r .U; T /
where
D r .U; X .T // D Im.Hcr .U; X .T // ! H r .U; X .T //; r ¤ 0;

0
D .U; X .T // D Im.Hc0 .U; X .T //^ ! H 0 .U; X .T //^ /; r D 0;
D r .U; T / D Im.Hcr .U; T / ! H r .U; T //; r ¤ 0; 1;
D r .U; T / D Im.Hcr .U; T /^ ! H r .U; T /^ /; r D 0; 1:
P ROOF. Part (a) of the theorem and (I 2.4) show that the dual of the sequence
L
0 ! D r .U; X .T // ! H r .U; X .T // ! H r .Kv ; X .T //
(complete the groups for r D 0/ is an exact sequence

L
H 2r .Kv ; T / ! Hc3r .U; T / ! D r .U; X .T // ! 0;
(complete the groups for 3 r D 0; 1/ which identifies D r .U; X .T // with

D 3r .U; T /: 2
Duality for Exts of tori

We wish to interpret (4.6b) in terms of Exts, but for this we shall need to use the
big étale site XEt on X and the flat site Xfl . Recall that for any locally Noetherian
scheme Y , YEt is the category of schemes locally of finite type over Y endowed
with the étale topology, and Yfl is the same category of schemes endowed with
the flat topology. Also f denotes the morphism Yfl ! YEt that is the identity
map on the underlying categories. For the rest of this section, ExtYr fl .F; F 0 / de-
notes the sheaf on YEt associated with the presheaf V 7! ExtrVfl .F; F 0 /. Note
that V 7! HomVfl .F; F 0 / is already a sheaf, and so .V; HomVfl .F; F 0 // D
HomVfl .F; F 0 /:
P ROPOSITION 4.8 For any sheaf F on YEt and smooth group scheme G of finite
type over Y , there is a spectral sequence
H r .YEt ; ExtYs fl .f F; G// H) ExtrCs
YEt .F; G/:
P ROOF. If F 0 is an injective sheaf on Yfl , then f F 0 is also injective (Milne

1980, III 1.20), and so HomYEt .F; f F 0 / is flabby (ibid. III 1.23). Hence
H r .YEt ; HomYEt .F; f F 0 // D 0 for r > 0. But for any V locally of finite
type over Y ,
.V; HomYEt .F; f F 0 // D HomVEt .F; f F 0 /
D HomVfl .f F; F 0 /
D .V; HomYfl .f F; F 0 //:
Therefore H r .YEt ; HomYfl .f F; F 0 // D 0 for r > 0, which means that
HomYfl .f F; F 0 / is acyclic for .YEt ; /. Next note that
.YEt ; HomYfl .f F; F 0 // D HomYfl .f F; F 0 / D HomYEt .F; f F 0 /,
and so there is a spectral sequence
H r .YEt ; ExtYs fl .f F; F 0 // H) RrCs ˛.F 0 /
where ˛ D HomYEt .F; / ı f . There is an obvious spectral sequence
ExtrYEt .F; Rs f F 0 / H) RrCs ˛.F 0 /:
On replacing F 0 with G in this spectral sequence and using that Rs f G D 0
for s > 0 (ibid. III 3.9), we find that RrCs ˛.G/ D ExtrCs
YEt .F; G/. The result
follows. 2
C OROLLARY 4.9 For any sheaf F on Yet and smooth group scheme G on Y ,
there is a spectral sequence
H r .Yet ; ExtYs fl .f F; G// H) ExtrCs
Yet .F; G/
where f now denotes the obvious morphism Yfl ! Yet and ExtYs fl .f F; G/
denotes the sheaf on Yet associated with V 7! ExtsVfl .f F; F 0 /:
P ROOF. Let f 0 W YEt ! Yet be the obvious morphism. For any sheaves F on
Yet and F 0 on YEt , HomYEt .f 0 F; F 0 / D HomYet .F; f0 F 0 /. As f0 is exact and
preserves injectives, this shows that ExtrYEt .f 0 F; F 0 / D ExtrYet .F; f F 0 / for all
r. Moreover the sheaf ExtYs fl .f F; Gm / of the corollary is the restriction to Yet
of the corresponding sheaf in (4.8), and so the result follows from (4.8) because
YEt and Yet yield the same cohomology groups (see Milne 1980, III 3.1). 2
P ROPOSITION 4.10 Let Y be a regular scheme, and let p be a prime such that
pŠ is invertible on Y . Then ExtYr fl .T; Gm / D 0 for 0 < r < 2p 1.
P ROOF. Since this is a local question, we can assume that T D Gm . From Breen
1969, 7, we know that ExtUr fl .Gm ; Gm / is torsion for r 1. Let ` be prime, and
consider the sequence
`
! Ext r1 r r
Yfl .` ; Gm / ! Ext Yfl .Gm ; Gm / ! Ext Yfl .Gm ; Gm / ! :
If Y is connected, then this sequence starts as

` 0 `
0 ! Z ! Z ! Z=`Z ! Ext 1Yfl .Gm ; Gm / ! :
Therefore, ExtY1fl .Gm ; Gm / D 0. We shall complete the proof by showing that

ExtYr fl .` ; Gm / D 0 for all ` if 0 < r < 2p 2.
If ` is invertible on Y , then ` is locally isomorphic to Z=`Z, and so the
sheaf ExtYr fl .` ; Gm / is locally isomorphic to ExtYr fl .Z=`Z; Gm /. There is an
exact sequence
! Ext rYfl .Z=`Z; Gm / ! Ext rYfl .Z; Gm / ! Ext rYfl .Z; Gm / ! :
But ExtYr fl .Z; Gm / is the sheaf (for the étale topology) associated with the presheaf
V 7! H r .Vet ; Gm /; it is therefore zero for r > 0 and equal to Gm for r D 0. The
sequence therefore shows that ExtYr fl .Z=`Z; Gm / D 0 for r > 0:
Next assume that ` is not invertible on Y . Our assumption implies that for all
primes q < p, q` D ` . Therefore the main theorem of Breen 1975 shows that
ExtYr fl .` ; Gm / D 0 for 1 < r < 2p 2, and for r D 1 the sheaf is well-known
to be zero (see Milne 1980, III 4.17). 2
T HEOREM 4.11 Assume that K is a number field. Let T be a torus on an open

subscheme U of X , and assume that 6 is invertible on U .
(a) The group ExtrUEt .T; Gm / is finitely generated for r D 0, finite for r D 1,
and torsion of cofinite type for r D 2; 3.
(b) The map ˛ r .U; F /W ExtrUEt .T; Gm / ! Hc3r .Uet ; T / is an isomorphism
for 0 < r 4, and ˛ 0 .U; T / defines an isomorphism HomU .T; Gm /^ !
Hc3 .U; T / (as usual, ^ denotes completion for the topology of finite sub-
groups).
P ROOF. The lemma shows that the spectral sequence in (4.9) gives isomorphisms

ExtrUEt .T; Gm / ! H r .U; X .T //
for r 4. Therefore the theorem follows from (4.6). 2

R EMARK 4.12 (a) The only reason we did not allow K to be a function field
in (4.6b) and (4.11) is that this case involves additional complications with the
topologies.
(b) It is likely that (4.11) holds with XEt replaced by the smooth site Xsm .
If one knew that the direct image functor f , where f is the obvious morphism
XEt ! Xsm , preserved injectives, then this would be obvious.
(c) It is not clear to the author whether or not pathologies of the type noted
in Breen 1969b should prevent ExtrUEt .T; Gm / being dual to Hc3r .U; T / for all
r 0 (and without restriction on the residue characteristics).
Relations to the theorems in Galois cohomology

In this subsection, U is an open affine subscheme of X and S is the set of primes
of K not corresponding to a point of U . We also make use of the notations
in I 4; for example, GS D 1 .U; /. For a sheaf F on U , we write F D D
Hom.F; Gm /. When M is a GS -module such that mM D 0 for some integer m
that is invertible on U , M D D Hom.M; K s /:
P ROPOSITION 4.13 Let F be a locally constant constructible sheaf on U such

that mF D 0 for some m that is invertible on U , and let M D F and N D M D
be the GS -modules corresponding to F and F D .
(a) The group D r .U; F / D XrS .K; M / and D r .U; F D / D XrS .U; M d /;
consequently, the pairing
D r .U; F / D 3r .U; F D / ! Q=Z
of (3.4) can be identified with a pairing
XrS .K; M / X3r D

S .K; M / ! Q=Z:
(b) The group ExtrU .F; Gm / D H r .GS ; N / and Hcr .U; F / D Extr1
GS .N; CS /;
consequently the pairing
ExtrU .F; Gm / Hc3r .U; F / ! Q=Z;
H r .GS ; N / Ext2r
GS .N; CS / ! Q=Z:
(c) The long exact sequence

M
! Hcr .U; F / ! H r .U; F / ! H r .Kv ; Fv / !
v2S
can be identified with a long exact sequence

M
! H 3r .GS ; N / ! H r .GS ; M / ! H r .Kv ; M / ! :
v2S
P ROOF. (a) Compare the sequences

L
0 ! XrS .K; M / ! H r .GS ; M / ! v2S H r .Kv ; M /
? ? ?
? ? ?
y y y
L
0 ! D r .U; F / ! H r .U; F / ! v2S H r .Kv ; F /
the second of which arises from the sequence in (2.3a) and the definition of
D r .U; F /.
(b) As ExtUr .F; Gm / D 0 for r > 0, ExtrU .F; Gm / D H r .U; F D /, and (2.9)
shows that H r .U; F D / D H r .GS ; N /. The second isomorphism can be read
off from the long exact Extr .F; /-sequence corresponding to the sequence of
sheaves defined by the exact sequence of GS -modules
L
0 ! RS ! Kv ! CS ! 0:
(c) It follows from (2.9) that H r .U; F / D H r .GS ; M /, and it is obvious that
3r .U; F D / ,
H r .Kv ; F / D H .Kv ; M /. According to (3.3), Hc .U; F / D H
r r
and (2.9) again shows that H 3r .U; F / D H

D 3r .U; N /: 2
For a GS -module M , write M d D Hom.M; RS /:
P ROPOSITION 4.14 Let T be a torus on U , and let X .T / be its sheaf of char-

acters. If M D X .T / is the GS -module corresponding to X .T /, then M d D
T . For all ` that are invertible on U and all r 1, D r .U; T /.`/ D XrS .K; M /.`/
and D r .U; X .T //.`/ D XrS .U; M d /.`/; consequently, the pairing
D r .U; T /.`/ D 3r .U; X .T //.`/ ! .Q=Z/.`/
of (4.8) can be identified for r 1 with a pairing
XrS .K; M d /.`/ X3r

S .K; M /.`/ ! .Q=Z/.`/:
P ROOF. In the course of proving (2.9), we showed that H r .UQ ; Gm /.`/ D 0 for
all r > 0. Therefore H r .U; T /.`/ D H r .GS ; T /.`/: 2
5. GLOBAL RESULTS: ABELIAN SCHEMES 197
R EMARK 4.15 Presumably, the maps are the same as those in Chapter I. Once
this has been checked, some of the results of each chapter can be deduced from
the other. It is not surprising that there is an overlap between the two chapters: to
give a constructible sheaf on X is the same as to give a GS -module M for some
finite set of nonarchimedean primes S together with Gal.Kvs =Kv /-modules Mv
for each v 2 S and equivariant maps M ! Mv (see Milne 1980, II 3.16).
Galois cohomology has the advantage of being more elementary than étale
cohomology, and one is not led to impose unnecessary restrictions (for example,
that S is finite) as is sometimes required for the étale topology. Etale cohomology
has the advantage that more machinery is available and the results are closer to
those that algebraic topology would suggest.
N OTES Propositions 4.1 and 4.2 are taken from Deninger 1986.
5 Global results: abelian schemes

The notations are the same as those listed at the start of 2. In particular, U is
always an open subscheme or X . As in (I 6), we fix an integer m that is invertible
on U and write M ^ Ddf lim M=mn M for the m-adic completion of M .

Let A be an abelian scheme over U , and let A be its generic fibre. As A
is proper over U , the valuative criterion of properness (Hartshorne 1977, II 4.7)
shows that every morphism Spec.K/ ! A extends to a morphism U ! A, that
is, A.K/ D A.U /. A similar statement holds for any V étale over U , which
shows that A represents g A on Uet . In fact (see Artin 1986, 1.4) A represents
g A on Usm .
P ROPOSITION 5.1 (a) The group H 0 .U; A/ is finitely generated; for r > 0,
H r .U; A/ is torsion and H r .U; A/.m/ is of cofinite-type; the map
Y
H r .U; A/.m/ ! H r .Kv ; A/.m/
v arch
is surjective for r D 2 and an isomorphism for r > 2.

(b) For r < 0, Y
H r1 .Kv ; A/ ! Hcr .U; A/
v arch
is an isomorphism; Hc0 .U; A/ is finitely generated; Hc1 .U; A/ is an extension of

a torsion group by a subgroup which has a natural compactification; Hc2 .U; A/ is
torsion; and Hc2 .U; A/.m/ is of cofinite-type; for r 3, Hcr .U; A/.m/ D 0:
P ROOF. (a) The group H 0 .U; A/ D A.U / D A.K/, which the Mordell-Weil
theorem states is finitely generated. As Galois cohomology groups are torsion
in degree > 1, the Leray spectral sequence H r .U; Rs g A/ H) H rCs .K; A/
shows that the groups H r .U; g A/ are torsion for r > 0 because Rs g A is tor-
sion for s > 0 and H r .K; A/ is torsion for r > 0. Finally, the finiteness of
H r .U; A/m follows from the cohomology sequence of
m
0 ! Am ! A ! A ! 0
because H r .U; Am / is finite for all r (by 2.13). On replacing m with mn in this
cohomology sequence and passing to the direct limit over n, we obtain an exact
sequence
0 ! H r1 .U; A/ ˝ Qm =Zm ! H r .U; A.m// ! H r .U; A/.m/ ! 0: (5.1.1)
The first term in this sequence is zero for r > 1 because then H r1 .U; A/ is tor-

sion. Hence H r .U; A.m// ! H r .U; A/.m/ for r 2. As Amn is locally con-
stant, H r .U; Amn / D H r .GS ; Amn .KS // (by 2.9). Therefore H r .U; Amn / !
Q
v arch H .Kv ; Amn / is surjective for r D 2 (by I 4.16) and
r
L an isomorphism for
r 3 (by I 4.10c), and it follows that H .U; A/.m/ ! v arch H .Kv ; A/.m/
r r
has the same properties.

(b) All statements follow immediately from (a) and the exact sequence
M
! Hcr .U; A/ ! H r .U; A/ ! H r .Kv ; A/ ! :
v…U 2
The dual abelian scheme At to A is characterised by the fact that it represents

the functor V 7! Ext1V .A; Gm / on Usm (generalized Barsotti-Weil formula, see
Oort 1966, III 18). As Hom.A; Gm / D 0, the local-global spectral sequence for
Exts gives rise to a map H r .U; At / ! ExtrC1
U .A; Gm / all r. On combining this
with the pairing
ExtrU .A; Gm / Hc3r .U; A/ ! Hc3 .U; Gm / ' Q=Z
we get a pairing
H r .U; At / Hc2r .U; A/ ! Hc3 .U; Gm / ' Q=Z:
(For a symmetric definition of this pairing in terms of biextensions, see Chapter
III.) We define
D 1 .U; A/ D Im.Hc1 .U; A/ ! H 1 .U; A//
Y
D Ker.H 1 .U; A/ ! H 1 .Kv ; A//:
v…U
It is a torsion group, and D 1 .U; A/.m/ is of cofinite-type.

T HEOREM 5.2 (a) The group H 0 .U; At /.m/ is finite; the pairing
H 0 .U; At /.m/ Hc2 .U; A/ ! Q=Z
is nondegenerate on the left and its right kernel is the m-divisible subgroup of
Hc2 .U; A/.
(b) The groups H 1 .U; At /.m/ and Hc1 .U; A/.m/ are of cofinite-type, and the
pairing
H 1 .U; At /.m/ Hc1 .U; A/.m/ ! Q=Z
annihilates exactly the divisible subgroups.
(c) If D 1 .U; At /.m/ is finite, then the compact group H 0 .U; At /^ is dual to
the discrete torsion group Hc2 .U; A/.m/.
P ROOF. Because Hcr .U; Amn / is finite for all n and r, passage to the inverse
limit in the sequences
n/
0 ! Hcr1 .U; A/.m ! Hcr .U; Amn / ! Hcr .U; A/mn ! 0
yields an exact sequence
0 ! Hcr1 .U; A/^ ! Hcr .U; Tm A/ ! Tm Hcr .U; A/ ! 0; (5.2.1)
where we have written Hcr .U; Tm A/ for lim Hcr .U; Amn /. Note that Tm Hcr .U; A/

is torsion-free and is nonzero only if the divisible subgroup of Hcr .U; A/.m/ is
nonzero. Corollary 3.3 provides us with nondegenerate pairings of finite groups
H r .U; Atmn / Hc3r .U; Amn / ! Q=Z,
and hence a nondegenerate pairing (of a discrete torsion group with a compact
group)
H r .U; At .m// Hc3r .U; Tm A/ ! Q=Z:
For r D 0, this shows that the finite group
H 0 .U; At /.m/ D H 0 .U; At .m// D A.K/.m/
is dual to Hc3 .U; Tm A/, and (5.2.1) shows that this last group equals
Hc2 .U; A/^ D Hc2 .U; A/=Hc2 .U; A/m-div
because Hc3 .U; A/ D 0. This completes the proof of (a).

From (5.1.1) we obtain an isomorphism

H 1 .U; At .m//=H 1 .U; At .m//div ! H 1 .U; At /.m/=H 1 .U; At /.m/div ,
and the left hand group is dual to Hc2 .U; Tm A/tor (see I 0.20e). The sequence
'
(5.2.1) gives an isomorphism Hc1 .U; A/^
tor ! Hc .U; Tm A/tor , and
2
Hc1 .U; A/^ 1 1

tor D Hc .U; A/.m/=Hc .U; A/.m/div :
This completes the proof of (b):

For (c), consider the diagram
H 0 .U; At /^
0 ! !
H 1 .U; Tm At / !
Tm H 1 .U; At / !
0
? ? ?
? ? ?
y y y
Hc2 .U; A/.m/ !
0 ! Hc2 .U; A.m// !
.Hc1 .U; A/ ˝ Qm =Zm / !
0:
It shows that the map H 0 .U; At /^ ! Hc2 .U; A/.m/ is injective, and that it
is an isomorphism if and only if Tm H 1 .U; At / ! .Hc1 .U; A/ ˝ Qm =Zm / is
injective. On applying Hom.Qm =Zm ; / to the exact sequence
Y
0 ! D 1 .U; At / ! H 1 .U; At / ! H 1 .Kv ; At /;
v…U
we obtain the top row of the following diagram:

Q
Tm D 1 .U; At / !
0 ! Tm H 1 .U; At / !
Tm H 1 .Kv ; At /
? ?
? ?
y y
Q 0
.Hc1 .U; A/ ˝ Qm =Zm // !
.H .Kv ; A/ ˝ Qm =Zm /
Our assumption on D 1 .U; At / implies that Tm D 1 .U; At / D 0, and so the dia-

gram shows that Tm H 1 .U; At / ! .Hc1 .U; A/ ˝ Qm =Zm / is injective. This
completes the proof of (c). 2
C OROLLARY 5.3 The group D 1 .U; A/ is torsion, and D 1 .U; At /.m/ is of cofinite-
type;8 there is a canonical pairing
D 1 .U; At /.m/ D 1 .U; A/.m/ ! Q=Z
whose left and right kernels are the divisible subgroups of the two groups.
8 Recall that throughout this section, we are assuming that m is invertible on U .

P ROOF. The first statement follows directly from the definition of D 1 .U; A/. For
the second statement, we use the commutative diagram
Q
0 ! D 1 .U; At /.m/ ! H 1 .U; At /.m/ ! v…U H .Kv ; A /
1 t
? ?
? ?
y y
Q
0 ! D 1 .U; A/.m/ ! Hc1 .U; A/.m/ ! v…U H 0 .Kv ; A/ :
It demonstrates that there is a map D 1 .U; At /.m/ ! D 1 .U; A/.m/ whose ker-
nel obviously contains the divisible subgroup of D 1 .U; At /.m/. The kernel of
the second vertical map is zero, and that of the first is divisible. A diagram chase
now shows that the kernel D 1 .U; At /.m/ ! D 1 .U; A/ is divisible. Because of
the symmetry of the situation, this implies that the right kernel of the pairing is
also divisible. 2
E XERCISE 5.4 Let 'D W A ! At be the map defined by a divisor on A. Show

that for all a 2 D 1 .U; A/, h'D .a/; ai D 0:
We now show how the above results can be applied to the Tate-Shafarevich
group. Write S for the set of primes of K not corresponding to a point of U:
L EMMA 5.5 The map H 1 .U; A/ ! H 1 .K; A/ induces isomorphisms

H 1 .U; A/ ! H 1 .GS ; A/

D 1 .U; A/ ! X1 .K; A/:
P ROOF. Because A D g A, the Leray spectral sequence for g gives an exact

sequence
0 ! H 1 .U; A/ ! H 1 .K; A/ ! .U; R1 g A/:
This sequence identifies H 1 .U; A/ with the set of principal homogeneous spaces
for A over K that are split by the inverse image of some étale cover of U , or
equivalently, that have a point in Kvun for each v (recall that Kv is the field of
fractions of Ohv /. From the Hochschild-Serre spectral sequence for Kvun over
Kv and (I 3.8), we see that the restriction map H 1 .Kv ; A/ ! H 1 .Kvun ; A/ is
injective. Therefore we have an exact sequence
M
0 ! H 1 .U; A/ ! H 1 .K; A/ ! H 1 .Kv ; A/: (5.5.1)
v2U
We have seen (I 3.10) that the maps H 1 .Kv ; A/ ! H 1 .K b v ; A/ are injective,

and so on comparing this sequence with that in (I 6.5), we see immediately that
H 1 .U; A/.m/ ' H 1 .GS ; A/.m/. On combining (5.5.1) with the exact sequence
M
0 ! D 1 .U; A/ ! H 1 .U; A/ ! H 1 .Kv ; A/;
v…U

M
0 ! D 1 .U; A/ ! H 1 .K; A/ ! H 1 .Kv ; A/;
all v
and this shows that D 1 .U; A/ D X1 .K; A/. 2
T HEOREM 5.6 Let A be an abelian variety over K:

(a) There is a canonical pairing
X1 .K; At / X1 .K; A/ ! Q=Z
whose kernels are the divisible subgroups of each group.

(b) Assume X1 .K; A/.m/ is finite. Then the dual of the exact sequence
M
0 ! X1 .K; A/.m/ ! H 1 .K; A/.m/ ! H 1 .Kv ; A/.m/ ! B ! 0
all v
Y
0 X1 .K; At /.m/ H 1 .K; A/.m/ At .Kv / At .K/^ 0:
all v
P ROOF. (a) Fix a prime ` and choose U so that ` is invertible on U and A has
good reduction at all primes of U . Then A and At extend to abelian schemes on
U , and the lemma shows that the pairing
D 1 .U; At /.`/ D 1 .U; A/.`/ ! Q=Z
X1 .K; At /.`/ X1 .K; A/.`/ ! Q=Z:
(b) Choose U so small that m is invertible on it and A has good reduction at

all primes of U . The assumption implies that X1 .K; At / is also finite. Therefore,
(5.2) shows that the dual of the sequence
L 1
0 ! X1 .K; A/.m/ ! H 1 .U; A/.m/ ! H .Kv ; A/.m/ ! Hc2 .U; A/.m/ !
v…U
Y
0 X1 .K; At /.m/ H 1 .U; A/.m/ At .Kv /^ At .K/^ :
v…U
Now pass to the direct limit in the first sequence and to the inverse limit in the
second overQsmaller and smaller open sets U . According to (I 6.25), the map
At .K/^ ! all v At .Kv /^ is injective, and so the result is obvious. 2
R EMARK 5.7 (a) We have now defined three pairings
X1 .K; At / X1 .K; A/ ! Q=Z:
For the sake of definiteness, we shall refer to the pairing in (5.6) as the Cassels-
Tate pairing.
(b) For any abelian scheme A on a regular scheme Y , Breen’s theorems
(Breen 1969a, 1975) imply that ExtYr .A; Gm / D 0 for r D 0 or 1 < r < 2p 1,
where p is a prime such that pŠ is invertible on Y . Since ExtY1fl .A; Gm / D At , we
see that ExtrUEt .A; Gm / D H r1 .U; At / for r 4, provided 6 is invertible on
U . In particular Ext2UEt .A; Gm / D H 1 .U; At /, which is countable. By way of
contrast, (4.4) implies that Ext2Uet .A; Gm / is countable if and only if the divisible
subgroup of Hc1 .U; A/ is zero. Thus if Ext2UEt .A; Gm / D Ext2Uet .A; Gm /, then
the divisible subgroup of X1 .K; A/ is zero (and there is an effective procedure
for finding the rank of A.K/!).
Finally we show that, by using étale cohomology, it is possible to simplify

the last part of the proof of the compatibility of the conjecture of Birch and
Swinnerton-Dyer with isogenies.
Let f W A ! B be an isogeny of degree prime to char K. The initial easy
calculations in (I 7) showed that to prove the equivalence of the conjecture for A
and B, one must show that
Y
z.f .K// D z.f .Kv // z.f t .K// z.X1 .f //:
v…U
In this formula, for a nonarchimedean prime v, Kv denotes the completion of K

rather than the Henselization, but it is easy to see that this does not change the
value of z.f .Kv //:
Let m be the degree of f , and choose an open scheme U ª X on which m is
invertible and which is such that f extends to an isogeny f W A ! B of abelian
schemes over U . The exact commutative diagram

Q 0
0 ! A.U / ! H .Kv ; A/ ! Hc1 .U; A/ ! X1 .K; A/ ! 0
v…U
? ? ? ?
? ? ? ?
y y y y
Q
0 ! B.U / ! H 0 .Kv ; B/ ! Hc1 .U; B/ ! X1 .K; B/ ! 0
v…U
shows that
Y
z.f .K// z.Hc1 .f // D z.H 0 .Kv ; f // z.X1 .f //:
v…U
To prove the compatibility, it therefore remains to show that

Y
z.Hc1 .f // z.f t .K// D z.H 0 .Kv ; f // z.f .Kv //1 :
v arch
Let F be the kernel of f W A ! B. The exact sequence
0 ! Coker.Hc0 .f // ! Hc1 .U; F / ! Hc1 .U; A/ ! Hc1 .U; B/ ! Hc2 .U; F /
! Hc2 .U; A/ ! Hc2 .U; B/ ! Hc3 .U; F / ! 0

shows that
c .U; F / ŒHc0 .U; F /1 D ŒCoker.Hc0 .f //1 z.Hc1 .f //1 z.Hc2 .f //:
But Y
c .U; F /ŒHc0 .U; F /1 D ŒF .Kv /ŒH 0 .Kv ; F /1
v arch
by 2.14b,
z.Hc2 .f // D z.f t .K//1
by duality, and
Y
Coker Hc0 .f / D Coker H 1 .Kv ; f /
v arch
Q 1 Q 1
because H .Kv ; A/ ! Hc0 .U; A/ and H .Kv ; B/ ! Hc0 .U; B/.
Therefore
Y
z.Hc1 .f // z.f t .K// D ŒF .Kv /1 ŒH 0 .Kv ; F /ŒCoker.H 1 .Kv ; f //;
v arch
6. GLOBAL RESULTS: SINGULAR SCHEMES 205
and it remains to show that for all archimedean primes v
ŒF .Kv /1 z.f .Kv // D ŒH 0 .Kv ; F /1 z.H 0 .Kv ; f //ŒCoker.H 1 .Kv ; f //:
0 ! CokerH 1 .Kv ; f / ! H 0 .Kv ; F / ! Ker H 0 .Kv ; f / ! 0;
we see that this comes down to the obvious fact that
Coker f .Kv / ! Coker H 0 .Kv ; f /
is an isomorphism (cf. I 3.7).
N OTES This section interprets Tate’s theorems on the Galois cohomology of abelian
varieties over number fields (Tate 1962) in terms of étale cohomology.
6 Global results: singular schemes

We now let X be an integral scheme whose normalization is the spectrum of the
ring of integers in K (number field case) or the unique complete smooth curve
with K as its function field (function field case). The definition in 2 of cohomol-
ogy groups with compact support also applies to singular X : for any sheaf F on
an open subscheme U of K, there is an exact sequence
M
! Hcr .U; F / ! H r .U; F / ! H r .Kv ; F / !
v2S
where S is the set of primes of K not corresponding to a point of the normalization

of U . Using (1.12) – (1.15), it is possible to prove an analogue of (2.3).
Let u 2 U , and let h 2 K . Then h can be written h D f =g with f; g 2 Ou ,
and we define
ordu .h/ D length.Ou =.f // length.Ou =.g//.
This determines a homomorphism K ! Z (see Fulton 1984, 1.2). Alternatively,

we could define X
ordu .h/ D Œk.v/W k.u/ ordv .h/
where the sum is over the points of the normalization of U lying over u (ibid.
1.2.3). One can define similar maps for each closed point lying over u on a
scheme V étale over U and so obtain a homomorphism ordu W g Gm ! iu Z.

Define G to be the complex of sheaves
P M
ordu
g Gm ! iu Z
u2U 0
on Uet . Note that if U is smooth, then we can identify G with Gm . We shall

frequently make use of the fact that Rs g Gm D 0 for s > 0. This follows from
the similar statement for the normalization UQ of U , and the fact that UQ ! U is
finite.
L EMMA 6.1 For all open subschemes U of X , there is a canonical trace map
'
TrW Hc3 .U; G/ ! Q=Z
such that
(a) whenever U is smooth, Tr is the map defined at the start of 3;
(b) whenever V U , the diagram
Tr
Hc3 .V; G/ ! Q=Z
?
?
y
Tr
Hc3 .U; G/ ! Q=Z
commutes.
P ROOF. The proof is similar to the smooth case. Let S be the set of primes of
K not corresponding to a point in the normalization of U , and assume first that
U ¤ X . From the definition of G; we obtain a cohomology sequence
M
0 ! H 2 .U; G/ ! H 2 .K; Gm / ! Q=Z ! H 3 .U; Gm / ! 0:
u2U 0
P
The middle map sends an element a of Br.K/ to invu .a/ where
P
invu .a/ D Œk.v/W k.u/ invv .a/:
v7!u
The kernel-cokernel exact sequence of the pair of maps

M M
Br.K/ ! Br.Kv / ! Q=Z
all v v2U 0
6. GLOBAL RESULTS: SINGULAR SCHEMES 207
provides us with the top row of the following diagram:

H 2 .U; G/ ! B ! Q=Z ! 0
? ?
? ?
y y
L
0 ! H 2 .U; G/ ! v2S Br.Kv / ! Hc3 .U; G/ ! 0:
L L L
Here B D Ker. all v Br.Kv / ! v2U 0 Q=Z). Write B D B 0 ˚ v2S Br.Kv / .
Then B 0 maps to zero in Q=Z, and as it is the kernel of the middle vertical map,
this shows that the map Q=Z ! Hc3 .U; G/ is an isomorphism.
For U D X , one can remove a smooth point and prove as in (2.6) that
Hc3 .X; G/ D Hc3 .X X fxg; G/: 2
As in 3, the trace map allows us to define maps

˛ r .U; F /W ExtrU .F; G/ ! Hc3r .U; F /
for any sheaf F on U:
T HEOREM 6.2 Let F be a Z-constructible sheaf on an open subset U of X . For
r 2, the groups ExtrU .F; G/ are torsion of cofinite-type, and ˛ r .U; F / is an iso-
morphism. For r D 0; 1, the groups ExtrU .F; G/ are of finite-type, and ˛ r .U; F /
defines isomorphisms
ExtrU .F; Gm /^ ! Hc3r .U; F /
where ^ denotes completion relative to the topology of subgroups of finite index.
If F is constructible, then ˛ r .U; F / is an isomorphism of finite groups for all
r 2 Z:
We begin by proving the theorem when F has support on a finite subset.
L EMMA 6.3 Theorem 6.2 is true if F has support on a proper closed subset of
U:
P ROOF. We can assume that our sheaf is i F where i is the inclusion of a single
closed point v into U . From the analogue of (2.3) for singular schemes, we
see that Hcr .U; i F / D H r .v; F /. As in (1.14), ExtrU .i F; g G/ D 0, and
so Extr1 r
x .F; Z/ ! ExtU .F; Gm / is an isomorphism. We have a commutative
diagram
Ext1U .F; G/ Hc3r .U; F / ! H 3 .U; G/

" # #
Extr1
gx .M; Z/ H 3r .g x; M / ! H 2 .g x ; Z/
and so the theorem follows in this case from (I 1.10). 2

Now let F be a sheaf on U , and let j W V ,! U be a smooth open sub-

scheme of U . For F jV , the theorem becomes (3.1). Since ExtrU .jŠ F jV; G/ D
ExtrV .F; G/ and Hcr .U; jŠ F jV / D Hcr .V; F /, this implies that the theorem is
true for jŠ F jV . The lemma shows that the theorem is true for i i F , and the two
cases can be combined as in (3.7) to prove the general case.
N OTES Theorem 6.2 is proved in Deninger 1986 in the case that U D X and F is
constructible.
7 Global results: higher dimensions

The notations are the same as those in 2. Throughout, W Y ! U will be a
morphism of finite-type, and we define Hcr .Y; F / D Hcr .U; RŠ F /.
P ROPOSITION 7.1 If F is constructible and mF D 0 for some integer m that is

invertible on U , then the groups H r .Y; F / are finite.
P ROOF. For any constructible sheaf F on Y , the sheaves Rr F are constructible

(see Deligne 1977, 1.1). Therefore it suffices to prove the proposition for U itself,
but we have already noted that (3.1) implies the proposition in this case. 2
R EMARK 7.2 In particular, the proposition shows that H 1 .Y; Z=mZ/ is finite
for all m that are invertible on Y , and this implies that 1ab .Y /.m/ is finite. Under
some additional hypotheses, most notably that is smooth, one knows (Katz and
Lang 1981) that the full group 1 .Y /ab is finite except for the part provided by
constant field extensions in the function field case.
Let v be an archimedean prime of K. In the next proposition, we write Yv for

Y U SpecKv and Yv for Y U SpecKvs :
P ROPOSITION 7.3 Let W Y ! U be proper and smooth, and let F be a locally

constant constructible sheaf on Y such that mF D 0 for some m that is invertible
on U . Then
Y .Yv ; F jYv /Gv
.Y; F / D ,
j.Yv ; F jYv /jv
v arch
where
df Y r
.Yv ; F jYv /Gv D ŒH r .Yv ; F jYv /Gv .1/ :
r
7. GLOBAL RESULTS: HIGHER DIMENSIONS 209
P ROOF. The proper-smooth base change theorem (Milne 1980, VI 4.2) shows
that the sheaves Rs F are locally constant and constructible for all r, and more-
over that .Rs F /v D H s .Yv ; F jYv / for all archimedean primes v. Therefore
(2.13) shows that
Y ŒH s .Yv ; F jYv /Gv
.GS ; Rs F / D :
jŒH s .Yv ; F jYv /jv
v arch
On taking the alternating product of these equalities, we obtain the result. 2
R EMARK 7.4 (a) A similar result is true for c .Y; F /:

(b) The last result can be regarded as a formula expressing the trace of the
identity map on Y in terms of the schemes Yv , v archimedean. For a similar
result for other maps, see Deninger 1986b.
Before stating a duality theorem for sheaves on such a Y , we note a slight

improvement of (3.1). Just as in the case of a single sheaf, there is a canonical
pairing of (hyper-) Ext and (hyper-) cohomology groups
ExtrU .F ; Gm / Hc3r .U; F / ! Hc3 .U; Gm /;
for any complex of sheaves F on U . Consequently, there are also maps
˛ r .U; F /W ExtrU .F ; Gm / ! Hc3r .U; F / :
L EMMA 7.5 Let F be a complex of sheaves on U that is bounded below and

such that H r .F / is constructible for all r and zero for r >> 0. Then
˛ r .U; F /W ExtrU .F ; Gm / ! Hc3r .U; F /
is an isomorphism of finite groups.
P ROOF. If F consists of a single sheaf, this is (3.1b). The general case follows
from this case by a standard argument (see, for example, Milne 1980, p280). 2
We write ExtrY;m .F; F 0 / for the Ext group computed in the category of sheaves
of Z=mZ-modules on Yet . Let F be a sheaf killed by m; if F 0 is an m-divisible
and F 0 ! I is an injective resolution of F 0 , then Fm0 ! I is an injective
resolution of Fm0 , and so
ExtrY;m .F; Fm0 / D H r .HomY .F; Im

// D H r .HomY .F; I // D ExtrY .F; F 0 /:
df df
In particular, if F is killed by m and m is invertible on Y , then ExtrY;m .F; m / D

ExtrY .F; Gm /:
Let W Y ! U be smooth and separated with fibres pure of dimension d , and
let m be an integer that is invertible on U . Then there is a canonical trace map
'
R2d Š ˝d
m ! Z=mZ (Milne 1980, p285). On U , m is locally isomorphic
to Z=mZ, and so when the trace map is tensored with m , it becomes an iso-
'
morphism R2d Š ˝dn
C1 ! m . As Rr ˝d
n
C1 D 0 for r > 2d , there is a
canonical trace map

' '
Hc2d C3 .Y; ˝d
m
C1
/ ! Hc3 .Y; m / ! Z=mZ;
and hence a pairing
ExtrY;m .F; ˝d

m
C1
/ Hc2d C3r .Y; F / ! Hc2d C3 .Y; ˝d
m
C1
/ ' Z=mZ:
T HEOREM 7.6 Let Y ! U be a smooth separated morphism with fibres pure of

dimension d , and let F be a constructible sheaf on Y such that mF D 0 for some
m that is invertible on U . Then
ExtrY;m .F; ˝d

m
C1
/ Hc2d C3r .Y; F / ! Hc2d C3 .Y; ˝d
m
C1
/ ' Z=mZ
P ROOF. The duality theorem in Artin, Grothendieck, and Verdier 1972/73, XVIII,
shows that there is a canonical isomorphism
'
RŠ .RHomY;m .F; ˝d
m
C1
Œ2d // ! RHomU;m .RŠ F; m /.
(See also Milne 1980, p285.) On applying R .U; / to the left hand side, we get
a complex of abelian groups whose rth cohomology group is ExtrC2d ˝d C1 /.
Y;m .F; m
On applying the same functor to the right hand side, we get a complex of abelian
groups whose rth cohomology group is ExtrU;m .RŠ F; m /. This is equal to
ExtrU .RŠ F; Gm /, and Lemma 7.5 shows that

ExtrU .RŠ F; Gm / ! Hom.Hc3r .U; RŠ F /; Z=mZ/:
By definition, Hc3r .U; RŠ F / D Hc3r .Y; F /, and so this proves the theorem.2
For any sheaf F on Y such that mF D 0, write F .i/ D F ˝ ˝i m and

F D .i/ D Hom.F; ˝i m /. Note that F D in the old terminology is equal to F D .1/
in the new.
C OROLLARY 7.7 Let W Y ! U be a smooth separated morphism with fibres

pure of dimension d , and let F be a locally constant constructible sheaf on Y
such that mF D 0 for some m that is invertible on U . Then cup-product defines
a nondegenerate pairing of finite groups
H r .Y; F D .d C 1// Hc2d C3r .Y; F / ! Hc2d C3 .Y; ˝d

m
C1
/ D Z=mZ;
for all r:
P ROOF. In this case ExtrY;m .F; ˝d

m
C1 / D H r .Y; F D .d C 1//:
2
As usual, we let Ki be the sheaf on Y defined by the i th Quillen K-functor.
C OROLLARY 7.8 Let F be a constructible sheaf on Y such that `n F D 0 for

some prime ` invertible on U . Assume that Hc2d C2 .Y; K2d C1 / is torsion. Then

there is a trace map Hc2d C3 .Y; K2d C1 /.`/ ! Q` =Z` , and the canonical pairing
ExtrY .F; K2d C1 / Hc2d C3r .Y; F / ! Hc2d C3 .Y; K2d C1 /.`/ Q` =Z`
is a duality of finite groups.
P ROOF. Recall (1.19), that for any m that is invertible on U , there is an exact
sequence of sheaves
m
0 ! ˝iC1
m ! K2iC1 ! K2iC1 ! 0:
Therefore ExtrY;`n .F; ˝d

`n
C1
/ D ExtrY .F; K2d C1 /. Also, there is an exact se-
quence
0 ! Hc2d C2 .Y; K2d C1 /˝Q` =Z` ! Hc2d C3 .Y; ˝d

`1
C1
/ ! Hc2d C3 .Y; K2d C1 /.`/ ! 0:
Since we have assumed Hc2d C2 .Y; K2d C1 / to be torsion, the first term of this
sequence is zero, and so we obtain isomorphisms
Hc2d C3 .Y; K2d C1 /.`/ Hc2d C3 .Y; ˝d

`1
C1
/ Q` =Z` :
The corollary now follows directly from (7.6). 2
R EMARK 7.9 For any regular scheme Y of finite type over a field, it is known
Milne 1986, 7.1, that H r .Y; Ki / is torsion for r > i. (The condition that Y
be of finite type over a field is only required so that Gersten’s conjecture can be
assumed.)
A SIDE 7.10 One would like to weaken the condition that Y is smooth over U
in the above results to the condition that Y is regular. The purity conjecture in
étale cohomology will be relevant for this. It states the following: Let iW Z ! Y
be a closed immersion of regular local Noetherian schemes such that for each
z in Z, the codimension of Z in Y at z is c, and let n be prime to the residue
characteristics; then .Rr i Š /.Z=mZ/ D 0 for r ¤ 2c, and .R2c i Š /.Z=mZ/ D
˝c
n (Artin, Grothendieck, and Verdier 1972/73, XIX).
The author is uncertain as to the exact conditions under which the proof of
the conjecture is complete. See ibid., XIX 2.1, and Thomason 1984.
The strategy for passing from the smooth case to the regular case is as followsW
replace U by its normalization in Y , and note that the theorem will hold on an
open subset V of Y ; now examine the map Y X V ! U . (Compare the proof of
the Poincaré duality theorem VI 11.1 in Milne 1980, especially Step 3.)
In the case that Y D U , Corollary 7.8 is much weaker than Theorem 3.1
because it requires that F be killed by an integer that is invertible on U . We
investigate some conjectures that lead to results that are true generalizations of
(3.1). We first consider the problem of duality for p-torsion sheaves in character-
istic p:
For a smooth variety Y over a field of characteristic p ¤ 0, we let Wn ˝Yi =k be
the sheaf of Witt differential i-forms of length n on Y (Illusie 1979). Define n .i/
to be the subsheaf of Wn ˝Yi =k of locally logarithmic differentials (see Milne
1986a, 1). The pairing
.!; ! 0 / 7! !^ ! 0 W Wn ˝ i Wn ˝ j ! Wn ˝ iCj
induces a pairing n .i/ n .j / ! n .i C j /:
T HEOREM 7.11 Let Y be a smooth complete variety of dimension d over a finite
'
field k . Then there is a canonical trace map H d C1 .Y; n .d // ! Z=p n Z, and
the cup-product pairing
H r .Y; n .i// H d C1r .Y; n .d i// ! H d C1 .Y; n .d // ' Z=p n Z
P ROOF. When dim Y 2 or n D 1, this is proved in Milne 1976. The extension

to the general case can be found in Milne 1986a. 2
C OROLLARY 7.12 The canonical pairing

ExtrY;p n .Z=p n Z; n .d // H d C1r .Y; Z=p n Z/ ! H d C1 .Y; n .d // ' Z=p n Z
P ROOF. One sees easily that ExtrY;p n .Z=p n Z; n .d // ' H r .Y; n .d //, and so
this follows immediately from the theorem. 2
C OROLLARY 7.13 Let Y be a smooth complete variety of dimension d 2

over a finite field k . Then there is a canonical trace map H d C2 .Y; Kd /.p/ !
Qp =Zp , and for any n there are nondegenerate pairings of finite groups
ExtrY .Z=p n Z; Kd // H d C2r .Y; Z=p n Z/ ! H d C2 .Y; Kd /.p/ ' Qp =Zp .
P ROOF. The key point is that there is an exact sequence

pn
0 ! Ki ! Ki ! n .i/ ! 0
for i 2. When i D 0; 1, the exactness of the sequence is obvious; when i D 2,

its exactness at the first and second terms follows from theorems of Suslin (1983a)
and Bloch (Bloch and Kato 1986) respectively. Now the corollary can be derived
in the same manner as (7.8); in particular the trace map is obtained from the maps

H d C1 .Y; 1 .d // ! H d C2 .Y; Kd /.p/ and H d C1 .Y; 1 .d // ! .Q=Z/.p/,
where 1 .i/ D lim n .i/: 2
!
R EMARK 7.14 It should be possible to extend the last three results to noncom-
plete varieties Y by using (in the proof of 7.11) cohomology with compact sup-
port for quasicoherent sheaves (see Deligne 1966 or Hartshorne 1972). However,
one problem in extending them to all constructible sheaves is that the purity the-
orem for the sheaves n .i/ is weaker than its analogue for the sheaves n .i/ (see
Milne 1986, 2). Nevertheless, I conjecture that for any constructible sheaf F of
Z=p n Z-modules on a smooth variety Y of dimension d over a finite field,
ExtrY;p n .F; n .d // Hcd C1r .Y; F / ! Hcd C1 .Y; n .d // ' Z=p n Z

I do not conjecture that the (7.13) holds for varieties of all dimensions.
Let Y be a smooth complete surface over a finite field. Then we have dualitiesW
ExtrY .Z=p n Z; K2 / H 4r .Y; Z=p n Z/ ! H 4 .Y; K2 /.p/ ' Qp =Zp ; p D chark;
ExtrY .Z=`n Z; K3 / H 5r .Y; Z=`n Z/ ! H 5 .Y; K3 /.`/ ' Q` =Z` ; ` ¤ chark:
These are similar, but the numbers do not agree! It appears that in order to obtain
a uniform statement, the sheaves Ki will have to be replaced by the objects Z.i/
conjectured in Lichtenbaum 1984 to exist in the derived category of S.Yet / for
any regular scheme Y . (Beilinson has independently conjectured the existence of

similar objects in S.YZar /.) These are to have the following properties:
(a) Z.0/ D Z, Z.1/ D Gm Œ1:
(b` ) For ` ¤ p, there is a distinguished triangle
`n
˝i
`n
Œ1 ! Z.i/ ! Z.i/ ! ˝i
`n
:
This implies that there is an exact sequence

`n
! H r .Y; Z.i// ! H r .Y; Z.i// ! H r .Y; ˝i
`n
/ ! :
(c) There are canonical pairings Z.i/ Z.j / ! Z.i C j /:

(d) H 2rj .Z.i// D Grr Kj up to small torsion, and H r .Z.i// D 0 for r > i
or r < 0 (also H 0 .Z.i// D 0 except when i D 0):
(e) If Y is a smooth complete variety over a finite field, then H r .Y; Z.i// is
torsion for all r ¤ 2i, and H 2r .Y; Z.r// is finitely generated.
In the present context, it is natural to ask that the complex have the following
additional properties when Y is a variety over a field of characteristic pW
(bp ) There is a distinguished triangle
pn
n .i/Œi 1 ! Z.i/ ! Z.i/ ! n .i/Œi:
This implies that there is an exact sequence

pn
! H r .Y; Z.i// ! H r .Y; Z.i// ! H ri .Y; n .i// ! .
(f) (Purity) If iW Z ,! Y is the inclusion of a smooth closed subscheme of

codimension c into a smooth scheme and j > c, then Ri Š Z.j / D Z.j c/Œ2c:
T HEOREM 7.15 Let W Y ! U be smooth and proper with fibres pure of dimen-
sion d . Let ` be a prime, and assume that either ` is invertible on U or ` DcharK
and Y is complete. Assume that there exist complexes Z.i/ satisfying .b` / and
that Hc2d C3 .Y; Z.d C 1// is torsion. Then there is a canonical isomorphism
'
Hc2d C4 .Y; Z.d C 1//.`/ ! .Q=Z/.`/,
and the cup-product pairing
H r .Y; Z.i//.`/Hc2d C4r .Y; Z.d C1i//.`/ ! Hc2d C4 .Y; Z.d C1//.`/ ' Q=Z.`/
annihilates only the divisible subgroups.

P ROOF. Assume first that ` ¤ char.K/. The same argument as in the proof of
(7.8) shows the existence of an isomorphism

Hc2d C3 .Y; ˝d
`1
C1
/ ! Hc2d C4 .Y; Z.d C 1//.
This proves that a trace map exists. Now the exact sequence
0 ! H r1 .Y; Z.i// ˝ .Q=Z/.`/ ! H r1 .Y; ˝i

`1
/ ! H r .Y; Z.i//.`/ ! 0
shows that H r1 .Y; ˝i

`1
/ modulo its divisible subgroup is isomorphic to
H .Y; Z.i//.`/ modulo its divisible subgroup. Similarly
r
0 ! Hcr .Y; Z.i//^ ! lim Hcr .Y; ˝i`m

/ ! T` HcrC1 .Y; Z.i// ! 0

shows that Hcr .Y; Z.i//.`/ modulo its divisible subgroup is isomorphic to the
torsion subgroup of lim Hcr .Y; ˝i
`m
/. Now the theorem follows from (7.7) using

(I 0.20e).
The proof when ` D p is similar. 2
Consider the following statement:

(*) for any smooth variety Y of dimension d over a finite field and
any constructible sheaf F on Y , there is a duality of finite groups
ExtrY .F; Z.d // Hc2d C2 .Y; F / ! Hc2d C2 .Y; Z.d // ' Q=Z:
When F is killed by some m prime to the characteristic of k and we assume

(b) and that Hc2d C1 .Y; Z.d // is torsion, this can be derived from (7.6) in the
same way as (7.8). When F is a p-primary sheaf, it is necessary to assume the
conjectured statement in (7.14).
T HEOREM 7.16 Let W Y ! U be a smooth proper morphism with fibres of
dimension d . Assume there exist complexes Z.i/ satisfying the conditions (a),
(b), and (f); also assume (*) above, and that Hc2d C3 .Y; Z.d C 1// is torsion, so
'
that there exists a canonical trace map Hc2d C4 .Y; Z.d C 1// ! Q=Z. Then for
any locally constant constructible sheaf F on Y , there is a nondegenerate pairing
of finite groups
ExtrY .F; Z.d C 1// Hc2d C4i .Y; F / ! Hc2d C4 .Y; Z.d C 1// ' Q=Z:
P ROOF. First assume F has support on YZ for some closed subscheme Z of U ,

and write i for the closed immersion YZ ,! Y . The spectral sequence
R HomYZ .F; Ri Š Z.d C 1// D R HomY .i F; Z.d C 1//

shows that
Extr2 r
YZ .F; Z.d 1// D ExtY .i F; Z.i//:
Thus this case of the theorem follows from the induction assumption.
Next suppose that mF D 0 for some m that is invertible on U . In this case
then the theorem can be deduced from (7.6) in the same way as (7.8).
Next suppose that p n F D 0, where p DcharK. In this case the statement
reduces to (*).
The last two paragraphs show that the theorem holds for the restriction of
F to YV for some open subscheme V of U , and this can be combined with the
statement proved in the first paragraph to obtain the full theorem. 2
In (4.11), we have shown that there is a nondegenerate pairing
ExtrU .F ˝ Z.1/; Z.1// Hc4r .U; F ˝ Z.1// ! Hc4 .Z.1// ' Q=Z:
for any torsion-free Z-constructible sheaf F on U . For a finite field k, we also

know that there is a nondegenerate pairing
Extrk .F ˝ Z.0/; Z.0// H 2 .k; F ˝ Z.0// ! H 2 .k; Z.0// ' Q=Z:
This suggests the following conjecture.

C ONJECTURE 7.17 9 For any regular scheme Y of finite-type over Spec Z and
'
of (absolute) dimension d , there is a canonical trace map Hc2d C2 .Y; Z.d // !
Q=Z. For any locally constant Z-constructible sheaf F on Y , the canonical pair-
ing
ExtrY .F ˝L Z.i/; Z.d //Hc2d C2r .Y; F ˝L Z.i// ! Hc2d C2 .Y; Z.d // ' Q=Z

ExtrY .F ˝L Z.i/; Z.d // ! Hc2d C2r .Y; F ˝L Z.i// ; r ¤ 2.d i/;
2i
Ext2d
Y .F ˝L Z.i/; Z.d // ! Hc2C2i .Y; F ˝L Z.i//.
The conjecture has obvious implications for higher class field theory.
Finally, we mention that Lichtenbaum (1986) has suggested a candidate for
Z.2/ and Bloch (1986) has suggested candidates for Z.r/, all r. Also Kato
(1985/6) has generalized (7.11) to a relative theorem, and in the case of a sur-
face Etesse (1986a,b) has generalized it to other sheaves.
N OTES This section owes much to conversations with Lichtenbaum and to his criticisms
of an earlier version.
9 In the original, this was miss-labelled 7.16.

Chapter III
Flat Cohomology
This chapter is concerned with duality theorems for the flat cohomology groups
of finite flat group schemes or Néron models of abelian varieties. In 1 - 4, the
base scheme is the spectrum of the ring of integers in a number field or a local
field of characteristic zero (with perfect residue field of nonzero characteristic).
In the remaining sections, the base scheme is the spectrum of the rings of integers
in a local field of nonzero characteristic or a curve over a finite field (or, more gen-
erally, a perfect field of nonzero characteristic). The appendices discuss various
aspects of the theory of finite group schemes and Néron models.
The prerequisites for this chapter are the same as for the last: a basic knowl-
edge of the theory of sites, as may be obtained from reading Chapters II and III
of Milne, 1980. All schemes are endowed with the flat topology.
The results of the chapter are more tentative than those in the first two chap-
ters. One problem is that we do not yet know what is the correct analogue for the
flat site of the notion of a constructible sheaf. The examples of Shatz 1966 show
that for any nonperfect field K, there exist torsion sheaves F over K such that
H r .Kfl ; F / is nonzero for arbitrarily high values of r. In particular, no duality
theorem can hold for all finite sheaves over such a field. We are thus forced to re-
strict our attention to sheaves that are represented by finite flat group schemes or
are slight generalizations of such sheaves. Another problem is that for a finite flat
group scheme N over an algebraically closed field k, the groups Extrk .N; Gm /
computed in the category of flat sheaves over k need not vanish for r > 0 (see
Breen 1969b); they therefore do not agree with the same groups computed in the
category of commutative algebraic groups over k, which vanish for r > 0.
217
218 CHAPTER III. FLAT COHOMOLOGY
0 Preliminaries
We begin by showing that some of the familiar constructions for the étale site can
also be made for the flat site.
Cohomology with support on closed subscheme

i j
Z ,! X -U
in which i and j are closed and open immersions respectively, and X is the
disjoint union of i.Z/ and j.U /.
L EMMA 0.1 The functor j W S.Xfl / ! S.Ufl / has an exact left adjoint jŠ .
P ROOF. For any presheaf P on U , we can define a presheaf jŠ P on X as follows:

for any morphism V ! X of finite type, set
.V; jŠ P / D ˚P .Vf /
where the sum is over all maps f 2 HomX .V; U / and Vf denotes V regarded as
a scheme over U by means of f . One checks easily that jŠ is left adjoint to the
restriction functor j p W P.Xfl / ! P.Ufl / and that it is exact. Let a be the functor
sending a presheaf on Xfl to its associated sheaf. Then the functor
jŠ a
S.Ufl / ,! P.Ufl / ! P.Xfl / ! S.Xfl /;
is easily seen to be left adjoint to j . It is therefore right exact. But it is also a

composite of left exact functors, which shows that it is exact. 2
L EMMA 0.2 There is a canonical exact sequence
0 ! jŠ j Z ! Z ! i i Z ! 0:
P ROOF. The maps are the adjunction maps. We explicitly compute the two end
terms. Let 'W V ! X be a scheme of finite type over U . When V is connected,
' factors through U in at most one way. Therefore, in this case, the presheaf
jŠ Z takes the value Z on V if '.V / j.U / and takes the value 0 otherwise. It
follows that jŠ Z is the sheaf
V 7! .V; jŠ Z/ D Hom.00 .V /; Z/
where 00 .V / is the subset of 0 .V / of connected components of V whose struc-

ture morphisms factor through U . Since .V; Z/ D Hom.0 .V /; Z/, it is obvi-
ous that jŠ j Z ! Z is injective.
For any V , .V; i Z/ D Hom.0 .' 1 Z/; Z/, which is zero if and only if
'.V / \ Z D ;. It clear from this that the sequence is exact at its middle term,
and that an element of .V; i Z/ lifts to .Vi ; Z/ for each Vi in an appropriate
Zariski open covering of V . This completes the proof. 2
The map F 7! Ker . .X; F / ! .U; F // defines a left exact functor S.Xfl / !
r .X; / for its r th right derived functor.
Ab, and we write HZ
P ROPOSITION 0.3 Let F be a sheaf on Xfl .

(a) For all r , HZr .X; F / D ExtrX .i Z; F /.
(b) For all r , ExtrX .jŠ Z; F / D ExtrU .Z; j F /.
(c) There is a long exact sequence
! HZr .X; F / ! H r .X; F / ! H r .U; F / ! :
P ROOF. (a) On applying Hom.; F / to the exact sequence in (0.2), we get an

exact sequence
0 ! HomX .i Z; F / ! HomX .Z; F / ! HomU .Z; j F /
or,
0 ! HomX .i Z; F / ! .X; F / ! .U; F /:

Therefore HomX .i Z; F / ! HZ0 .X; F /, and on taking the right derived func-
tors we obtain the result.
(b) Note that, because it has an exact left adjoint, j preserves injectives. It
is also exact (Milne 1980, p68). Therefore we may derive the equality
HomX .jŠ Z; F / D HomU .Z; j F /
and obtain an isomorphism
ExtrX .jŠ Z; F / ExtrU .Z; j F /:
(c) The ExtX .; F /-sequence arising from the exact sequence in (0.2) is the
required sequence. 2
Cohomology with compact support

Let X be the spectrum of the ring of integers in a number field or else a complete
smooth curve over a perfect field, and let K be the field of rational functions
on X . For any open subscheme U of X and sheaf F on Ufl , we shall define
cohomology groups with compact support Hcr .Xfl ; F / having properties similar
to their namesakes for the étale topology. In particular, they will be related to the
usual cohomology groups by an exact sequence
L
! Hcr .U; F / ! H r .U; F / ! v2XXU H
r
.Kv ; F / !
Ovh of Ov .
where Kv is the field of fractions of the Henselization S
0
Let Z be the complement of U in X , and let Z D v2XXU SpecKv (disjoint
union). Then Z 0 D lim V X U , where the limit is over the étale neighbourhoods

V of Z in X W
o V X U
?V
~~
~~~ étale
~
~~
Z /X o U
Let i 0 be the canonical map i 0 W Z 0 ! U , and let F ! I .F / be an injective
resolution of F on Ufl . Then i 0 is exact and preserves injectives, and so F jZ 0 !
I .F /jZ 0 is an injective resolution of F jZ 0 . There is an obvious restriction map
uW .U; I .F // ! .Z 0 ; I .F /jZ 0 /;
and we define1 Hc .U; F / to be the translate C .u/Œ1 of its mapping cone.

Finally, we set Hcr .U; F / D H r .Hc .U; F //.
P ROPOSITION 0.4 (a) For any sheaf F on an open subscheme U X , there is

an exact sequence
L
! Hcr .U; F / ! H r .U; F / ! H r .Kv ; Fv / ! :
v2XXU
(b) For any short exact sequence
0 ! F 0 ! F ! F 00 ! 0
of sheaves on U , there is a long exact sequence of cohomology groups
! Hcr .U; F 0 / ! Hcr .U; F / ! Hcr .U; F 00 / ! :

1 In the original, this was denoted Hc .
(c) For any sheaf F on U and open subscheme V of U , there is an exact sequence
L
! Hcr .V; F jV / ! Hcr .U; F / ! H r .Ohv ; F / ! :
v2U XV
(d) If F is the inverse image of a sheaf F0 on Uet , or if F is represented by a

smooth algebraic space, then Hcr .Ufl ; F / D H r .Xet ; jŠ F /.
(e) For any sheaves F and G on U , there are canonical pairings
ExtrU .F; G/ Hcs .U; F / ! HcrCs .U; G/:
P ROOF. (a) Directly from the definition of Hc .U; F /, we see that there is a dis-
tinguished triangle
Hc .U; F / ! .U; I .F // ! .Z 0 ; I .F /jZ 0 / ! Hc .U; F /Œ1:
As
H r . .U; I .F /// D H r .U; F /
and
L
H r . .Z 0 ; I .F /jZ 0 // D H r .Z 0 ; F jZ 0 / D H r .Kv ; F /;
v2XXU
we see that the required sequence is simply the cohomology sequence of this
triangle.
(b) From the morphism
0 ! .U; I .F 0 // ! .U; I .F // ! .U; I .F 00 // ! 0

? ? ?
? 0 ?u ? 00
yu y yu
0 ! .Z 0 ; I .F 0 /jZ 0 / ! .Z 0 ; I .F /jZ 0 / ! .Z 0 ; I .F 0 /jZ 0 / ! 0
of short exact sequences of complexes, we may deduce (II 0.10a) the existence of
a distinguished triangle
Hc .U; F 00 /Œ1 ! Hc .U; F 0 / ! Hc .U; F / ! Hc .U; F 00 /:
This yields the required exact sequence.

(c) Let F ! I .F / be an injective resolution of F on U , and consider the
maps
a L b L L
.V; I .F // ! .Kv ; I .F // ! .Kv ; I .F //˚ v .Ovh ; I .F //Œ1:
v…V v…U v2U XV
The map b is such that H r .b/ is the sum of the identity maps
H r .Kv ; F / ! H r .Kv ; F / .v 2 X X U /
and the maps in the complex
H r .Ovh ; F / ! H r .Kv ; F / ! HvrC1 .Ovh ; F /; v 2 U X V:
From these maps, we get a distinguished triangle (II 0.10c)
C .b/Œ1 ! C .a/ ! C .bıa/ ! C .b/:
Clearly C .a/ D Hc .V; F /Œ1. We shallL

show that there exist isomorphisms
C .bıa/ Hc .U; F /Œ1 and C .b/ .Ovh ; I .F //Œ1 (in the derived
v2U XV
category). Thus the cohomology sequence of this triangle is the required se-
quence.
From
bıa L L c L
.V; I .F // ! .Kv ; I .F // ˚ v .Ovh ; I .F //Œ1 ! v .Ovh ; I .F //Œ1
v…U v2U XV v2U XV
we get a distinguished triangle
C .c/Œ1 ! C .b ı a/ ! C .c ı b ı a/ ! C .c/:
L
But C .c ı b ı a/ .U; I .F //Œ1 and C .c/ .Kv ; I .F //Œ1, which
v…U
shows that C .b ı a/ Hc .U; F /Œ1.
Finally, the statement about C .b/ is obvious from the distinguished triangles
(for v 2 U X V )
.Ovh ; I .F // ! .Kv ; I .F // ! v .Ovh ; I .F //Œ1 ! .Ovh ; I .F //Œ1:

(d) Since H r .Uet ; F / ! H r .Ufl ; F / and

H r ..SpecKv /et ; F / ! H r ..SpecKv /fl ; F /
(Milne 1980, III 3), this follows from a comparison of the sequence in (0.4a) and
with the corresponding sequence for the étale topology.
(e) Let c 0 2 Hcr .U; F /, and regard it as a homotopy class of maps of degree r
c 0 W Z ! Hc .U; F /Œr:
Let c 2 ExtrU .F; F 0 /, and regard it as a homotopy class of maps of degree s;
cW I .F / ! I .F 0 /Œs:
On restricting the maps in this class, we get a similar class of maps
cjZ 0 W I .F /jZ 0 ! .I .F 0 /jZ 0 /Œs:
The last two maps combine to give a morphism Hc .U; F / ! Hc .U; F 0 /Œs, and
we define hc; c 0 i to be the composite of this morphism with c. 2
R EMARK 0.5 (a) Let iW Z,!Y be a closed immersion, and let F be a sheaf on
Z. The proof in Milne 1980, II 3.6, of the exactness of i for the étale topology
(hence the equality H r .Uet ; i F / D H r .Zet ; F //, fails for the flat topology.
(b) Note that the sequence in (c) has the same form as (II 2.3d) except that in
the latter sequence it has been possible to replace H r .Ovh ; F / with H r .v; i F /.
In the case of the flat topology, this is also possible if F is represented by a smooth
algebraic space (Milne 1980, III 3.11).
R EMARK 0.6 (a) In the case that X is the spectrum of the ring of integers in a
number field, it is natural to replace Hc .U; F / with the mapping cone of
L
S .Kv ; Fv / ! Hc .U; F /:
v arch
Then the cohomology groups with compact support fit into an exact sequence
L
! Hcr .U; F / ! H r .U; F / ! v H r .Kv ; F / !
where the sum is now over all primes of K, including the archimedean primes,
not in U , and for archimedean v,
H r .Kv ; F / D HTr .Gal.Kvs =Kv /; F .Kvs //:
(b) In the definition of Hcr .U; F / it is possible to replace Kv with its comple-
tion. Then the sequence in (0.4a) will be exact with Kv the completion of K at v,
and the sequence in (0.4c) will be exact with Ovh replaced with O b v . This approach
has the disadvantage that the groups no longer agree with the étale groups (that
is, (0.4d) will no longer hold in general).
The definition given here of cohomology groups with compact support is sim-
ple and leads quickly to the results we want. I do not know whether there is a more
natural definition nor in what generality it is possible to define such groups.
Topological duality for vector spaces

Let k be a finite field, and let V be a locally compact topological vector space
over k. Write V _ for the topological linear dual of V , V _ D Homk;cts .V; k/.
T HEOREM 0.7 The pairing
V _ V ! C ; .f; v/ 7! exp. 2 i
p Trk=Fp f .v//
identifies V _ with the Pontryagin dual of V .
P ROOF. Let V be the Pontryagin dual of V . The pairing identifies V _ with a

subspace of V . Clearly the elements of V _ separate points in V , and so V _ is
dense in V . But V _ is locally compact, and so it is an open subset of its closure
in V ; hence it is open in V . As it is a subgroup, this implies that it is also
closed in V and so equals V . 2
Let R be a complete discrete valuation ring of characteristic p ¤ 0 having

a finite residue field k, and let K be the field of fractions of R. The choice

of a uniformizing parameter t for R determines an isomorphism K ! P k..t//
carrying R onto kŒŒt. Define a residue map resW K ! k by setting res. ai t i / D
a1 .
C OROLLARY 0.8 Let V be a free R-module of finite rank, and let V _ be its
R-linear dual. Then the pairing
V _ .V ˝ K/=V ! C ; .f; v/ 7! exp. 2 i

p Trk=Fp .res.f .v///
identifies V _ with the Pontryagin dual of .V ˝ K/=V .
P ROOF. Consider first the case that V D R. Then R is isomorphic (as a topo-
logical k-vector space) to the direct product of countably many copies of k, and
K=R is isomorphic to the direct sum of countably many copies of k. The pairing
.a; b/ 7! res.ab/W R K=R ! k
identifies R with the k-linear topological dual of K=R, and so (0.7) shows that

R ! .K=R/ . In the general case, the pairing
V _ .V ˝ K/=V ! k; .f; v/ 7! res.f .v//
similarly identifies V _ with k-linear topological dual of .V ˝K/=V , and so again

the result follows from the proposition. 2
The Frobenius morphism

For any scheme S of characteristic p ¤ 0, the absolute Frobenius map Fabs W S !
S is defined to be the identity map on the underlying topological space and a 7!
ap on OS . It is functorial in the sense that for any morphism W X ! S, the
diagram
Fabs
X X
? ?
? ?
y y
Fabs
S S
commutes, but it does not commute with base change. The relative Frobenius
map FX=S is defined by the diagram
Fabs
to
X .p/ o
df
X x
X X .p/ D X S;Fabs S:
xx
FX=S
xxx
.p/ xx
xx

x
xx
Fabs {x
x
So S
For any morphism T ! S, FX=S T id D FXS T =T .
A scheme S is said to be perfect if Fabs W S ! S is an isomorphism. For
example, an affine scheme SpecR is perfect if the p th power map a 7! ap W R !
R is an isomorphism. In the case that S is perfect, it is possible to identify
1 ı W X ! S and F
.p/ W X .p/ ! S with Fabs X=S with Fabs .
h WN !
A finite group scheme N over a scheme S is said to have height h if FN=S
h/
N .p h1 is not zero. For any flat group scheme N , there is a canon-
is zero but FN=S
ical morphism V D VN W N .p/ ! N , called the Verschiebung (see Demazure and
Gabriel 1970, IV, 3, 4).
The Oort-Tate classification of group schemes of order p

Let D ZŒ
; .p.p 1//1 \ Zp , where
is a primitive pth root of 1, and
the intersection is taken inside Qp . We consider only schemes X such that the
unique morphism X ! SpecZ factors through Spec. For example, X can be
any scheme of characteristic p because has Fp as a residue field. The following
statements classify the finite flat group schemes of order p over X (see Oort and
Tate 1970, or Shatz 1986, 4).
L of order p over
(0.9a) It is possible to associate a finite flat group scheme Na;b
X with each triple .L; a; b/ comprising an invertible sheaf L over X , an element
a 2 .X; L˝p1 /, and an element b 2 .X; L˝1p / such that a ˝ b D wp for

a certain universal element wp .
(0.9b) Every finite flat group scheme of order p over X is isomorphic to Na;b L
for some triple .L; a; b/.

L ! 0
(0.9c) There exists an isomorphism Na;b NaL0 ;b 0 if and only if there is an

isomorphism L ! L0 carrying a to a0 and b to b 0 .
(0.9d) For all X -schemes Y ,
L
Na;b .Y / D fy 2 .Y; L ˝ OY /j y ˝p D a ˝ yg:
_
L is N L .
(0.9e) The Cartier dual of Na;b b;a
(0.9f) When X has characteristic p, wp D 0. If a D 0 in this case, then
L is L with the p-power map
N has height one, and the p-Lie algebra of Na;b
f 7! f .p/ D b ˝ f .
Duality for unipotent perfect group schemes

When the ground field is finite, our duality theorems will be for the cohomology
groups endowed with the structure of a topological group. When the ground
field is not finite, it will be necessary to endow the cohomology groups with
a stronger structure, namely the structure of a perfect pro-algebraic group, and
replace Pontryagin duality with Breen-Serre duality. We now describe this last
duality.
Let S be a perfect scheme of characteristic p ¤ 0. The perfection X pf of an
S-scheme X is the projective limit of the system
F .p 1 / F F .p n / F
Xred Xred Xred
It is a perfect scheme, and has the universal property that
HomS .X; Y / D HomS .X pf ; Y /
for any perfect S-scheme Y .

Let S be the spectrum a perfect field k. A perfect S-scheme X is said to
be algebraic if it is the perfection of a scheme of finite type over S. From the
corresponding fact for the algebraic group schemes over S, one sees easily that
the perfect algebraic group schemes over X form an abelian category. Define
the perfect site Spf to be that whose underlying category consists of all algebraic
perfect S-schemes and whose covering families are the surjective families of étale
morphisms. For any algebraic group scheme G over S, the sheaf on Spf defined
by G is represented by G pf . One checks easily, that any sheaf on Spf that is an

extension of perfect algebraic group schemes is itself represented by a perfect
algebraic group scheme.
We write S.p n / for the category of sheaves on Spf killed by p n .
pf pf
T HEOREM 0.10 For all r > 0, ExtrS.p/ .Ga ; Ga / D 0.
P ROOF. See Breen 1981, where the result is proved with the base scheme the
spectrum of any perfect ring of characteristic p ¤ 0. 2
L EMMA 0.11 Let f W Sfl ! Spf be the morphism of sites defined by the iden-
tity map. For any affine commutative algebraic group scheme G on S , f G is
represented by G pf and Rr f G D 0 for r > 0.
P ROOF. We have already observed that f G is represented by G pf . If G is

smooth, then Rr f G D 0 for r > 0 because of the coincidence of flat and
étale cohomology groups of smooth group schemes (Milne 1980, III 3.9). We
calculate Rr f G for G equal to ˛p or p by using the exact sequences
F
0 ! p ! Gm ! Gm ! 0
F
0 ! ˛p ! Ga ! Ga ! 0:
pf pf
Since F is an automorphism of Gm and Ga , we have Rr f p D 0 D Rr f ˛p
for all r. The general case follows from these case because, locally for the étale
topology, any G has a composition series whose quotients are Gm , Ga , p , ˛p ,
or an étale group scheme. 2
By a p-primary group scheme, we mean a group scheme killed by a power of

p.
L EMMA 0.12 Let G be a perfect p -primary affine algebraic group scheme on S ;

let U be its identity component, and let D D G=U . Then D is étale and U has a
pf
composition series whose quotients are all isomorphic to Ga .
P ROOF. In view of the exactness of f , this is a consequence of the structure

theorem for affine commutative algebraic group schemes on S. 2
Let G.p n / be the category S

of perfect affine algebraic group schemes on S
killed by p n , and let G.p 1 / D G.p n /. Note that (0.12) shows that G.p 1 / can
also be described as the category of perfect unipotent group schemes on S.
L EMMA 0.13 Let G be a perfect unipotent group scheme on S ; let U be its

identity component, and let D D G=U .
(a) If G is killed by p n , then there exists a canonical isomorphism
'
RHomS.p n / .G; Z=p n Z/ ! RHomS .G; Qp =Zp /:
'
(b) HomS .G; Qp =Zp / ! HomS .D; Qp =Zp /, which equals D , the Pon-
tryagin dual of D .
'
(c) ExtS1 .G; Qp =Zp / ! ExtS1 .U; Qp =Zp /, which is represented by a con-
nected unipotent perfect group scheme; if U has the structure of a Wn .k/-
module, then there is a canonical isomorphism of Wn .k/-modules
'
ExtS1 .G; Qp =Zp / ! HomWn .k/ .U; Wn .OS //:
(d) ExtSr .G; Qp =Zp / D 0 for r > 1.
P ROOF. (a) Choose an injective resolution I of Qp =Zp . The usual argument in

the case of abelian groups shows that an injective sheaf is divisible. Therefore the
kernel Ipn of p n W I ! I is a resolution of Z=p n Z, and it is obvious that it is
an injective resolution in S.p n /. Consequently
RHomS.p n / .G; Z=p n Z/ D HomS.p n / .G; Ipn /

D HomS .G; I /
D RHomS .G; Qp =Zp /:
(b,c,d) When D D Z=pZ, the sequence

p
0 ! HomS .Z=pZ; Q=Z/ ! HomS .Z; Q=Z/ !HomS .Z; Q=Z/ !
ExtS1 .Z=pZ; Q=Z/ ! 0
shows that
RHomS .D; Q=Z/ D HomS .D; Q=Z/ D D :
A general étale group D is locally (for the étale topology) an extension of copies
of Z=pZ, and so the same equalities holds for it.
pf
If U D Ga , then (0.10) shows that the exact sequence
pf pf 1F pf pf pf
0 ! HomS.p/.Ga ; Ga / ! HomS.p/.Ga ; Ga / ! ExtS1.p/ .Ga ; Z=pZ/ !
yields an isomorphism
pf
RHomS.p/.Ga ; Z=pZ/ GŒ1
where G is the cokernel of 1 F . It is well-known (see, for example, Serre 1960)

pf pf pf
that G D Ga . Therefore ExtS1 .Ga ; Z=pZ/ D ExtS1.p/ .Ga ; Q=Z/ is connected,
pf
and ExtSr .Ga ; Q=Z/ D 0 for r ¤ 1. A general connected unipotent group U is
pf
an extension of copies of Ga , and so ExtSr .U; Q=Z/ is connected for r D 1 and
zero for r ¤ 1.
Suppose that U has the structure of a Wn .OS /-module. The Artin-Schreier
sequence
F 1
0 ! Z=p n Z ! Wn .OS / ! Wn .OS / ! 0
gives a morphism Wn .OS / ! .Z=p n Z/Œ1 in the derived category of S.p n /, and
hence a canonical homomorphism
HomWn .OS / .U; Wn .OS // ! RHomS.p n/ .U; Z=p n Z/;
which is Wn .OS /-linear for the given structure on U . One checks easily that
1
ExtW n .OS /
.U; Wn .OS // D 0:
as U has a filtration by sub-Wn .OS /-modules such that the quotients are linearly
pf
isomorphic to Ga , it suffices to show that this homomorphism is an isomorphism
pf
for U D Ga , which is assured by (0.10). It follows that the homomorphism is an
isomorphism.
The assertions (b), (c), and (d) can now be proved in the general case by
making use of the exact sequence
0 ! U ! G ! D ! 0: 2
For any perfect connected unipotent group U , we write U _ for Ext 1S .U; Qp =Zp /.
Let D b .G.p 1 // be the full subcategory of the derived category of S.S pf / con-
sisting of those bounded complexes whose cohomology lies in G.p 1 /. For any
G in D b .G.p 1 //, define G t D RHomS .G ; Qp =Zp /.
T HEOREM 0.14 For any G in D b .G.p 1 //, G t also lies D b .G.p 1 //, and
'
there is a canonical isomorphism G ! G t t . There exist canonical exact
sequences
0 ! U rC1 .G /_ ! H r .G t / ! D r .G / ! 0;
where U r .G / is the identity component of H r .G / and
D r .G / D H r .G /=U r .G /:
P ROOF. The cohomology sheaves of RHomS .G ; Qp =Zp / are the abutment of

the spectral sequence
E2r;s D Ext rS .H s .G /; Qp =Zp / H) Ext rCs

S .G ; Qp =Zp /:
After (0.13), E2r;s D 0 for r ¤ 0; 1, so that the spectral sequence reduces to a

family of short exact sequences
0 ! ExtS1 .H rC1 .G /; Qp =Zp / ! ExtSr .G ; Qp =Zp / ! HomS .H r .G /; Qp =Zp / ! 0;
which are the required exact sequences. They imply moreover that G t is in
D b .G.p 1 // because G.p 1 / is stable under extension. Finally one shows that
the homomorphism of biduality G ! G t t is an isomorphism by reducing the
question to the cases of Z=pZ and Ga , which both follow directly from (0.13). 2
R EMARK 0.15 Denote by Extrk .G; H / the Ext group computed in the category
of affine perfect group schemes over k. Each such Ext group can be given a
canonical structure as a perfect group scheme, and (0.10) implies that, when G
and H are unipotent, Extrk .G; H / agrees with ExtSr .G; H /.
PAIRINGS IN THE DERIVED CATEGORY

We review some of the basic definitions concerning pairings in the derived cate-
gory. For more details, see Gamst and Hoechsmann 1970 or Hartshorne 1966.
Fix a scheme X , endow it with a Grothendieck topology, and write S.X /
for the resulting category of sheaves. Write C.X / for the category of complexes
in S.X /, and K.X / for category with the same objects but whose morphisms
are homotopy classes of maps in C.X /. The derived category D.X / is obtained
from K.X / by formally inverting quasi-isomorphisms. Thus, for example, a map

A ! B and a quasi-isomorphism B C define a morphism A ! C
in D.X /. As usual, C C .X /, C .X /, and C b .X / denote respectively the cate-
gories of complexes bounded below, bounded above, and bounded in both direc-
tions. We use similar notations for the homotopy and derived categories. Since
S.X / has enough injectives, for every A in C C .X /, there is a quasi-isomorphism

A ! I.A / with I.A / a complex of injectives, and there is a canonical equiv-
alence of categories I C .X / ! D C .X /, where I C .X / is the full subcategory of
K C .X / whose objects are complexes of injective objects.
Recall that a sheaf P is flat if ˝ P W S.X / ! S.X / is exact. For any

bounded-above complex A , there is a quasi-isomorphism P .A / ! A with
P .A / a complex of flat sheaves. If B is a second bounded-above complex, then
P .A / ˝ B is a well-defined object of D.X /, which is denoted by A ˝L B .

Despite appearances, there is a canonical isomorphism A ˝L B ' B ˝L A .
Let M and N be flat sheaves on X , and let A and B be objects of C b .X /.
There is a canonical pairing
ExtrX .M; A / ExtsX .N; B / ! ExtrCs L
X .M ˝ N; A ˝ B /
that can be defined as follows: represent elements f 2 ExtrX .M; A / and g 2
ExtsX .N; B / as homotopy classes of maps f W M ! I.A /Œr and gW N !
I.B /Œs; then f ˝ g is represented by

M ˝ N ! I.A /Œr ˝ I.B /Œs P .I.A// ˝ I.B /Œr C s:
The pairing is natural, bi-additive, associative, and symmetric (up to the usual
signs). It also behaves well with respect to boundary maps (Gamst and Hoechs-
mann, ibid.). A similar discussion applies when N is not flat — replace it with a
flat resolution.
Consider a map A ˝L B ! G . The above discussion gives a pairing
H r .X; A / ExtsX .M; B / ! ExtrCs
X .M; G /;
There is also the usual (obvious) pairing
ExtrX .B ; G / ExtsX .M; B / ! ExtrCs
X .M; G /:
The map A ˝L B ! G defines a map A ! Hom.B ; C / and hence edge
morphisms H r .X; A / ! ExtrX .B ; G /.
T HEOREM 0.16 The following diagram commutes:
H r .X; A / ExtsX .M; B / ! ExtrCs
X .M; G /
# k #
ExtrX .B ; G / ExtsX .M; B / ! ExtrCs
X .M; G /:
P ROOF. See Gamst and Hoechsmann 1970. 2
N OTES The definition of cohomology groups with compact support for the flat topology
is new, and will play an important role in this chapter.
The duality for unipotent perfect group schemes has its origins in a remark of Serre
(1960, p55) that Ext’s in the category of unipotent perfect group schemes over an alge-
braically closed field can be used to define an autoduality of the category. For a detailed
exposition in this context, see Bégueri 1980, 1; Serre in fact worked with the equivalent
category of quasi-algebraic groups. The replacement of Ext’s in the category of perfect
group schemes with Ext’s in the category of sheaves, which is essential for the appli-
cations we have in mind, is easy once one has Breen’s vanishing theorem (0.10). Our
exposition of the autoduality is based on Berthelot 1981, II, which, in turn, is based on
Milne 1976.
Most of the rest of the material is standard.
1 Local results: mixed characteristic, finite group schemes

Throughout this section, R will be a Henselian discrete valuation ring with finite
residue field k and field of fractions K of characteristic zero. In particular, R is
excellent. We use the same notations as in (II 1): for example, X D SpecR and i
and j are the inclusions of the closed point x and the open point u of X into X .
The characteristic of k will be denoted by p and the maximal ideal of R by m.
L EMMA 1.1 Let N be a finite flat group scheme over R.

(a) The map N.R/ ! N.K/ is a bijective, and H 1 .X; N / ! H 1 .K; NK / is
injective; for r 2, H r .X; N / D 0.
(b) The boundary map H r .K; N / ! HxrC1 .X; N / defines isomorphisms
'
H 1 .K; N /=H 1 .R; N / ! Hx2 .X; N /
'
H 2 .K; N / ! Hx3 .X; N /I
for r ¤ 2; 3,
Hxr .X; N / D 0:
P ROOF. (a) As N is finite, it is the spectrum of a finite R-algebra A. The image

of any R-homomorphism A ! K is finite over R and is therefore contained in R.
This shows that N.R/ D N.K/. An element c of H 1 .X; N / is represented by a
principal homogeneous space P over X (Milne 1980, III 4.3), and c D 0 if and
only if P .R/ is nonempty. Again P is the spectrum of a finite R-algebra, and so
if P has a point in K, then it already has a point in R.
From Appendix A, we know that there is an exact sequence
0 ! N ! G ! G0 ! 0
in which G and G 0 are smooth group schemes of finite type over X . According
to Milne 1980, III 3.11, H r .X; G/ D H r .k; G0 / and H r .X; G 0 / D H r .k; G00 /
for r > 0, where G0 and G00 are the closed fibres of G and G 0 over X . The five

lemma therefore shows that H r .X; N / ! H r .k; N0 / for r > 1. We now use
that there is an exact sequence
0 ! N0 ! G1 ! G10 ! 0
with G1 and G10 smooth connected group schemes over k (for example, abelian
varieties). By Lang’s lemma, H r .k; G1 / D 0 D H r .k; G10 / for r > 0, and it
follows that H r .k; N0 / D 0 for r > 1.
1. LOCAL: MIXED CHARACTERISTIC, FINITE GROUP SCHEMES 233
(b) This follows from the first statement, because of the exact sequence
! Hxr .X; N / ! H r .X; N / ! H r .K; N / ! :
and the fact that H r .K; N / D 0 for r > 2 (K has cohomological dimension 2).2
R EMARK 1.2 The proof of the lemma does not use that K has characteristic zero.
The same argument as in the proof of (a) shows that H r .X; N / D 0 for r 2
if N is a finite flat group scheme over any Noetherian Henselian local ring with
finite residue field.
Let F be a sheaf on X . The pairing
ExtrX .F; Gm / ExtsX .i Z; F / ! ExtrCs

X .i Z; Gm /
can be identified with a pairing
ExtrX .F; Gm / Hxs .X; F / ! HxrCs .X; Gm /I
see (0.3a). Since Gm is a smooth group scheme, the natural map Hxr .Xet ; Gm / !
Hxr .Xfl ; Gm / is an isomorphism for all r, and so (see II 1) there is a canonical
trace map
'
Hx3 .Xfl ; Gm / ! Q=Z:
Let N be a finite group scheme over X . The sheaf defined by the Cartier dual
N D of N can be identified with Hom.N; Gm /, and the pairing N D N ! Gm
defines a pairing
H r .X; N D / Hx3r .X; N / ! Hx3 .X; Gm / ' Q=Z:
This can also be defined using the edge morphisms H s .X; N D / ! ExtsX .N; Gm /
and the Ext-pairing (0.16).
T HEOREM 1.3 For any finite flat group scheme N on X ,
H r .X; N D / Hx3r .X; N / ! Hx3 .X; Gm / ' Q=Z
is a nondegenerate pairing of finite groups, all r .
We shall give two proofs, but first we list some corollaries.

C OROLLARY 1.4 For any finite flat group scheme N over X , H 1 .X; N D / is the
exact annihilator of H 1 .X; N / in the pairing
H 1 .K; N D / H 1 .K; N / ! H 2 .K; Gm / ' Q=Z
of (I 2.3).
P ROOF. The diagram
H 1 .X; N D / H 1 .X; N / ! H 2 .X; Gm / D 0

# # #
H 1 .K; N D / H 1 .K; N / ! H 2 .K; Gm / ' Q=Z
shows that H 1 .X; N D / and H 1 .X; N / annihilate each other in the pairing. For
r D 1, (1.1) allows us to identify the pairing in the theorem with
H 1 .X; N D / H 1 .K; N /=H 1 .X; N / ! H 2 .K; Gm /:
Thus we see that the nondegeneracy of the pairing in this case is equivalent to the
statement of the corollary. 2
C OROLLARY 1.5 Let N be a finite flat group scheme on X . For all r < 2p 2,
ExtrX .N; Gm / Hx3r .X; N / ! Hx3 .X; Gm / ' Q=Z
P ROOF. Let N.p/ be the p-primary component of N . According to Breen 1975,

r
ExtX .N.p/; Gm / D 0 for 1 r < 2p 2;
and as we explained in the proof of (II 4.10), this implies that ExtX
r
.N; Gm / D
0 for 1 < r < 2p 2 (ExtX .N; Gm / D 0 by Milne 1980, III 4.17, and
1
ExtX r .N.`/; G / D 0 for r > 1 and ` ¤ p because N.`/ is locally constant

m
for the étale topology). Hence H r .X; N D / D Extr .N; Gm / for r < 2p 2. 2
Write f W Xfl ! Xet for the morphism defined by the identity map.
C OROLLARY 1.6 Let N be a quasi-finite flat group scheme over X whose p -

primary component N.p/ is finite over X . Let N D be the complex of sheaves
such that (
D HomXfl .N.`/; Gm / if ` D p
N .`/ D
f RHomXet .N.`/; Gm / if ` ¤ p .
Then
P ROOF. For each ` ¤ p, Hxs .X; N.`// ' Hxs .Xet ; N.`// and
H r .X; N D .`// ' H r .Xet ; RHomXet .N.`/; Gm // ' ExtrXet .N.`/; Gm /:
Therefore, for the prime-to-p components of the groups, the corollary follows
from (II 1.8). For the p component it follows immediately from the theorem. 2
Q UESTION 1.7 Does there exist a single statement that fully generalizes both
(1.3) and (II 1.8b)?
T HE FIRST PROOF OF T HEOREM 1.3

The first proof is very short, but makes use of (A.6). We begin by proving a
duality result for abelian schemes.
P ROPOSITION 1.8 Let A be an abelian scheme over X , and let At be its dual.
Then the pairing
H r .X; At / Hx2r .X; A/ ! Hx3 .X; Gm / ' Q=Z
defined by the canonical biextension At ˝L A ! Gm (see Appendix C) induces

an isomorphism H 0 .X; At /^ ! Hx2 .X; A/ (^ denotes the completion for the
profinite topology); for r ¤ 0, both groups are zero.
P ROOF. Let A and A0 be the open and closed fibres respectively of A=X . Then
H r .X; A/ D H r .x; A0 / for r > 0 (see Milne 1980, I 3.11), and H r .X; A0 / D 0
for r > 0 by Lang’s lemma. Moreover, A.X / D A.K/ because A is proper over
X . Therefore Hxr .X; A/ is zero for r 1 and equals H r1 .K; A/ for r > 1.
Hence Hx2 .X; A/ D H 1 .K; A/, and Hxr .X; A/ D 0 for all other values of r.
Consequently, when r D 0, the pairing becomes
H 0 .K; At / H 2 .K; A/ ! H 2 .K; Gm / ' Q=Z;
and both groups are zero for all other values of r. The proposition now follows
from (I 3.4). 2
P ROOF ( OF T HEOREM 1.3) We now prove (1.3). Note that the Lemma 1.1 im-
plies that H 0 .X; N / and H 1 .X; N / are finite (NK is a finite étale group scheme).
According to (A.6) and (A.7), N can be embedded into an exact sequence
0!N !A!B!0
with A and B abelian schemes over X . This leads to an exact cohomology se-
quence
0 ! Hx2 .X; N / ! Hx2 .X; A/ ! Hx2 .X; B/ ! Hx3 .X; N / ! 0;
which we regard as a sequence of discrete groups.

There is a dual exact sequence
0 ! N D ! B t ! At ! 0
(see Appendix C), which leads to a cohomology sequence
0 ! H 0 .X; N D / ! H 0 .X; B t / ! H 0 .X; At / ! H 1 .X; N D / ! 0
The two middle terms of the sequence have natural topologies, and the two end
terms inherit the discrete topology. Therefore the sequence remains exact after
the middle two terms have been completed. The theorem now follows from the
diagram
H 0 .X; N D / !
0 ! H 0 .X; B t /^ !
H 0 .X; At /^ !
H 1 .X; N D / !
0
? ? ? ?
? ? ? ?
y y y y
Hx3 .X; N / !
0 ! Hx2 .X; B/ !
Hx2 .X; A/ !
Hx2 .X; N D / !
0
2
T HE SECOND PROOF OF T HEOREM 1.3

The second proof will use p-divisible groups, for whose basic theory we refer the
reader to Tate 1967b or Shatz 1986. In order to simplify the argument, we shall
assume throughout that R is complete.
Let H D .H
; i
/
1 be a p-divisible group over X . Let L be a finite ex-

tension of K, and let RL be the integral closure of R in L. Then the group of
points H.RL / of H with values in RL is defined to be lim H.RL =miL /, where
S i
H.RL =miL / Ddf lim H
.RL =miL /. Let MH D H.RL / where L runs over
!v
the finite extensions of K contained in K s . Then MH becomes a discrete module
under the obvious action of Gal.K s =K/.
L EMMA 1.9 (a) The group H.R/ is compact if and only if its torsion subgroup
is finite.
(b) The group of elements of MH fixed by GK is H.R/.
(c) The sequence of Gal.K s =K/-modules

p
0 ! H
.K s / ! MH ! MH ! 0
is exact.
P ROOF. (a) Let H ı be the identity component of H . Then H ı .R/ is an open

subgroup of H.R/, and H.R/=H ı .R/ is torsion. Since H ı .R/ is compact and
its torsion subgroup is finite (H.R/ is isomorphic to Rdim.R/ ), the assertion is
obvious.
(b) It suffices to show that H.RL /G D H.R/ for L a finite Galois extension
of K with Galois group G. When we write H D SpfA, H.RL / is the set of
continuous homomorphisms A ! RL , and so the assertion is obvious.
(c) The sequences
p
0 ! H
.RL =mi RL / ! H.RL =mi RL /!H.RL =mi RL /
are exact, and so on passing to the inverse limit, we obtain an exact sequence
p
0 ! H
.RL / ! H.RL / ! H.RL /.
The term H
.RL / has its usual meaning, and we have observed in (1.1) that
H
.RL / D H
.L/. Therefore on passing to the direct limit we obtain an exact
sequence
p
0 ! H
.Ks / ! MH ! MH :
It remains to show that pW MH ! MH is surjective. If H is étale, then
MH D H.ks /, which is obviously divisible by p. If H is connected, say H D
SpfA, then the map pW H ! H turns A into a free A-module of finite rank, and
so the divisibility is again obvious. The general case now follows from the fact
that
0 ! H ı .RL / ! H.RL / ! H et .RL / ! 0
is exact for all L (see Tate 1967b, p168). 2
Let H t be the p-divisible group dual to H (ibid. 2.3), so that .H t /

D H
D .
P ROPOSITION 1.10 Assume that the torsion subgroups of H.R/ and H t .R/ are
both finite. Then there is a canonical pairing
H 1 .K; MH t / H.R/ ! Q=Z
which identifies the discrete group H 1 .K; MH t / with the dual of the compact
group H.R/.
P ROOF. From the cohomology sequence of the sequence in (1.9c), we get an

exact sequence
/
0 ! H.R/.p ! H 1 .K; H
/ ! H 1 .K; MH /p ! 0:
I claim that the first map in the sequence factors through H 1 .X; H
/,!H 1 .K; H
/.
Note first that it is possible to define H.R0 / for any finite flat R-algebra R0 . Let
P 2 H.R/; then the inverse image P under p
W H ! H (regarded as map of
functors of finite flat R-algebras) is a principal homogeneous space for H
over
X whose generic fibre represents the image of P in H 1 .K; H
/. This proves the
claim.
As we observed in the proof of (1.4), the images of H 1 .X; H
D / and H 1 .X; H
/
annihilate each other in the nondegenerate pairing
H 1 .K; H
D / H 1 .K; H
/ ! H 2 .K; Gm / ' Q=Z:
/
Therefore, the images of H t .R/.p / and H.R/.p annihilate each other under
the same pairing, and so the diagram
/
0 ! H.R/.p ! H 1 .K; H
/
?
?'
y
/
0 ! .H 1 .K; MH t /p / ! H 1 .K; H
t / ! H t .R/.p

shows that the pairing induces an injection H.R/.p / ,!.H 1 .K; MH t /p / . In
the limit this becomes an injection
/
lim H.R/.p ,! .lim.H i .K; MH t /p / D H 1 .K; MH t / ;
!

and because of our assumption on H.R/, lim H.R/.p / D H.R/.

We therefore have an injection H.R/ ! H 1 .K; MH t / , and to prove that it
is surjective, it suffices to show that ŒH.R/.p/ D ŒH 1 .K; MH t /p . This we do
using an argument similar to that in the proof of (I 3.2). From (I 2.8), we know
that
.K; H1 / D .RW pR/h D .K; H1t /
where h is the common height of H and H t . The logarithm map (Tate 1967b,
2.3), and our assumptions on H.R/ and H t .R/ show that H.R/ and H t .R/
0
contain subgroups of finite index isomorphic to Rd and Rd respectively where
d and d 0 are the dimensions of H and H t . Therefore
ŒH.R/.p/ =ŒH.R/p D .RW pR/d ;
0
ŒH t .R/.p/ =ŒH t .R/p D .RW pR/d :
From the cohomology sequence of

p
0 ! H1t .Ks / ! MH t ! MH t ! 0
we see that
ŒH t .R/p ŒH 2 .K; H1t /
.K; H1t / D
ŒH t .R/.p/ ŒH 1 .K; MH t /p
or
1 1 ŒH 0 .K; H1 /
D 0 :
.RW pR/h .RW pR/d ŒH 1 .K; MH t /p
But d C d 0 D h (Tate 1967b, Pptn 3), and so this shows that
ŒH 1 .K; MH t /p D .RW pR/d ŒH.R/p D ŒH.R/.p/;
as required. 2
R EMARK 1.11 The proposition is false without the condition that the torsion
subgroups of H.R/ and H t .R/ are finite. For example, if H D .Z=p
Z/
1 ,
then H.R/ (D Qp =Zp ) and H 1 .K; MH t / are both infinite and discrete, and so
cannot be dual. If H D .p /
1 , then
H.R/ D fa 2 R j a 1 mod mg
and H 1 .K; MH t / D Hom.Gal.Ks =K/; Qp =Zp /, which are not (quite) dual.
We now complete the second proof of (1.3). For r D 0, the pairing can be
identified with the pairing
H 0 .K; N D / H 2 .K; N / ! H 2 .K; Gm / ' Q=Z
of (I 2.3), and for r ¤ 0; 1 both groups are zero. This leaves the case r D 1,
and we saw in the proof of (1.4) that this case is equivalent to the statement
that H 1 .X; N D / and H 1 .X; N / are exact annihilators in the duality between
H 1 .K; N D / and H 1 .K; N /. We know that the groups in question do annihilate
each other, and so
ŒH 1 .K; N / ŒH 1 .X; N /ŒH 1 .X; N D /;
and to show that they are exact annihilators it suffices to prove that equality holds.
According to (A.4) and (A.7), there is an exact sequence
'
0 ! N ! H ! H 0 ! 0
with H and H 0 both p-divisible groups. Moreover from the construction of the
sequence, it is clear that H , H 0 , and their duals satisfy the hypotheses of (1.10).
Write H r .X; H / for lim H r .X; H
/, and let
!
H 0 .R/.'/ D Coker.'W H.R/ ! H 0 .R//;

H 1 .X; H /' D Ker.'W H 1 .X; H / ! H 1 .X; H 0 //:
The cohomology sequence of the above sequence and its dual
't
0 ! N D ! H 0t ! H t ! 0
show that
ŒH 1 .X; N / D ŒH 0 .R/.'/ ŒH 1 .X; H /' ŒH 0 .R/.'/ and

t t
ŒH 1 .X; N D / D ŒH t .R/.' / ŒH 1 .X; H 0t /' t ŒH t .R/.' / :
On combining the three inequalities, we find that

t
ŒH 1 .K; N / ŒH 1 .X; N /ŒH 1.X; N D / ŒH 0 .R/.'/ ŒH t .R/.' / :
't '
It follows from (1.10) that H 0t .R/ ! H t .R/ is dual to H 1 .K; MH / !
t
H 1 .K; MH 0 /, and so ŒH t .R/.' / D ŒH 1 .K; MH /' . But
ŒH 1 .K; N / D ŒH 0 .R/.'/ ŒH 1 .K; MH /'
, from which it follows that all of the above inequalities are equalities. This
completes the second proof of (1.3).
R EMARK 1.12 The above argument shows that for any isogeny 'W H ! H 0 of
'
p-divisible groups over X , H 0 .R/.'/ ! H 1 .X; Ker.'// and H 1 .X; H /' D 0,
provided the torsion subgroups of H.R/ and H t .R/ are finite.
A duality theorem for p-divisible groups

If A is an abelian variety over K, then (I 3.4) shows that there is an exact sequence
0 ! A.K/ ˝ Qp =Zp ! H 1 .K; A.p// ! At .K/ .p/ ! 0:
Our next result is the analogue of this for p-divisible groups. Recall that H r .K; H / Ddf
lim H r .K; H
/.
!
P ROPOSITION 1.13 Assume that R is complete, and let H be a p -divisible group

over X such that the torsion subgroups of H.R/ and H t .R/ are finite. Then
H 1 .X; H / D 0, and there is an exact sequence
0 ! H.R/ ˝ Q=Z ! H 1 .K; H / ! H t .R/ ! 0:
P ROOF. On applying Remark 1.12 to the isogeny p
W H ! H , we find that

H.R/.p / ! H 1 .X; H
/ and H 1 .X; H /p D 0. As H 1 .X; H / is p-primary,
the equality shows that it is zero. From (1.4) we know there is an exact sequence
0 ! H 1 .X; H
/ ! H 1 .K; H
/ ! H 1 .X; H
t / ! 0:
On using the isomorphism to replace the first and third terms in this sequence, we
obtain an exact sequence
/ /
0 ! H.R/.p ! H 1 .K; H
/ ! H t .R/.p ! 0:
Now one has only to pass to the direct limit to obtain the result. 2
If N is a finite flat group scheme over X , then the groups H r .X; N / are finite for
all r and zero for r > 1. We define .X; N / D ŒH 0 .X; N /=ŒH 1 .X; N /. Let
N D SpecB. Recall that the order of N is defined to be the rank of B over R,
and the discriminant ideal of N is the discriminant ideal of B over R.
T HEOREM 1.14 Let N be a finite flat group scheme over X , and let n be its order
and d its discriminant ideal. Then .RW d/ is an nth power and
.X; N / D .RW d/1=n :
When N is étale, H r .X; N / D H r .g; N.Run //, and so both sides of the
equation are 1. This allows us to assume that N is local. We can also assume
that R is complete because passing to the completion does not change either side.
Consider an exact sequence
'
0 ! N ! H ! H 0 ! 0
with H and H 0 connected p-divisible groups. As H 1 .X; H / D 0;
df
.X; N / D z.'.R// D ŒKer '.R/=ŒCoker '.R/:
Write H D SpfA and H 0 D SpfA0 . Then A and A0 are power series rings in d
variables over R, where d is the common dimension of H and H 0 . The map '
corresponds to a homomorphism ' a W A0 ! A making A into a free A0 -module of
rank n. It also defines a map d' a W ˝A
1
0 =R ! ˝A=R .
1
L EMMA 1.15 Choose bases for ˝A 1 1 0

0 =R and ˝A=R , and let and be the cor-
V V
responding basis elements for d ˝A0 =R and d ˝A=R over A0 and A respec-
tively. If
Vd a Vd V
d' W ˝A0 =R ! d ˝A=R
maps 0 to a , a 2 A, then NA=A0 a generates the discriminant ideal of A over
A0 .
Vd Vd
P ROOF. This follows from the existence of a trace map TrW ˝A=R ! ˝A0 =R ;
see Tate 1967b, p165. 2
Let !A=R and !A0 =R be the R-modules of invariant differentials on H and

H0 respectively. The inclusion !A=R ! ˝A=R
1 induces isomorphisms !A=R ˝R

A ! ˝A=R 1
and !A=R ! ˝A=R 1
˝A R. Let and 0 be basis elements for
Vd Vd
!A=R and !A0 =R . On taking and 0 to be ˝ 1 and 0 ˝ 1 in the
lemma, we find that d' a . 0 / D a , a 2 R, and that the discriminant ideal of A
over A0 is generated by an . Since A ˝A0 R D B, where B D .N; ON /, this
shows that the discriminant ideal d of N is generated by an . It remains to show
that .X; N / D .RW aR/.
Let T .H / and T .H 0 / be the tangent spaces to H and H 0 at zero. They are
dual to !A=R and !A0 =R , and so aR is equal to the determinant ideal of the map
of R-modules d'W T .H / ! T .H 0 /. Recall Tate 1967b, 2.4, that there exists a
logarithm map logW H.R/ ! T .H / ˝R K, and that if we choose an isomorphism
A RŒŒX1 ; :::; Xd , then for any c with c p1 < jpj, log gives an isomorphism
between
H.R/c D fx 2 H.R/j jxi j c all ig
and
T .H /c D f 2 T .H /j j.Xi /j c all ig D p c T .H /:
From the commutative diagram
'c
H.R/c ! H 0 .R/c
? ?
?
ylog
?
ylog
.d'/c
T .H /c ! T .H 0 /c
we see that .RW det.d'// D .RW det.d'/c / D ŒCoker.'c /. But
ŒCoker.'c / D z.'/.H.R/W H.R/c /.H 0 .R/W H 0 .R/c /1

and .H.R/W H.R/c / D p d.c1/ D .H 0 .R/W H 0 .R/c /. Therefore
.RW aR/ D .RW det.d'// D z.'/ D .X; N /:
This completes the proof of Theorem 1.14.

For a finite flat group scheme N over X , define2
ŒHx2 .X; N /
x .X; N / D :
ŒHx3 .X; N /
C OROLLARY 1.16 Let N be a finite flat group scheme over X , and let n be its
order and d its discriminant ideal. Then
x .X; N / D .RW nR/.RW d/1=n :
P ROOF. From the cohomology sequence of X u, we find that
x .X; N / D .X; N /.K; N /1 D .RW dR/1=n .RW nR/
(see (1.14) and (I 2.8).) 2
As H r .X; N / is dual to Hx3r .X; N D /, .X; N /x .X; N D / D 1. Therefore

(1.14) and (1.16) imply that d.N /d.N D / D .nn /. This formula can also be
directly deduced from the formulas
NmB=R D.N / D d.N /; D.N /D.N D / D .n/;
where D.N / and D.N D / are the differents of N and N D and B D .N; ON /
(see Raynaud 1974, Pptn 9, and Mazur and Roberts 1970, A.2).
Let N be a quasi-finite, flat, separated group scheme over X . Because X is
Henselian, there is a finite flat group scheme N f N having the same closed
fibre as N ; moreover, N=N f is étale and it is the extension by zero of its generic
fibre (cf. Milne 1980, I 4.2c). In this case we write nf for the order of N f and
df for its discriminant ideal.
C OROLLARY 1.17 Let N be as above.

f
(a) .X; N / D .RW df /1=n ;
f
(b) x .X; N / D .RW nR/.RW df /1=n .
2 The original had Hx1 in the denominator, which is clearly wrong (1.1).
P ROOF. (a) From the cohomology sequence of
0 ! N f ! N ! N=N f ! 0
we find that .X; N / D .X; N f /.X; N=N f /. Therefore it suffices to prove

the formula in the cases that N D N f or N f D 0. In the first case, it becomes
the formula in (1.14), and in the second N D jŠ NK , and so both sides are 1.
(b) Again, it suffices to prove the formula in the cases N D N f or N f D 0.

In the first case the formula becomes that in (1.16). In the second, H r .K; NK / !
HxrC1 .U; N /, and so
x .X; N / D .K; NK /1 D .RW nR/
by (I 2.8). 2
C OROLLARY 1.18 Let H D .H

/
1 be a p -divisible group over X ; then

.X; H
/ D .RW p d
R/, where d is the dimension of H .
P ROOF. According to Tate 1967b, Pptn 2, the discriminant ideal of H

is gener-
h h h
ated by p d
p . Therefore .X; H
/ D ..RW p d
R/p /p D .RW p d
R/. 2
Extensions of morphisms
For each finite group scheme N over X , define h.N / to be the pair .NK ; H 1 .X; N //.
A morphism h.N / ! h.N 0 / is a K-morphism 'K W NK ! NK 0 such that
0 0
H 1 .'K /W H 1 .K; NK / ! H 1 .K; NK /
maps H 1 .X; N / into H 1 .X; N 0 /.

T HEOREM 1.19 The functor N 7! h.N / of finite group schemes over X is fully
faithful; that is, a homomorphism 'K W NK ! NK0 extends to a homomorphism
'W N ! N 0 if and only if H 1 .'K / maps H 1 .X; N / into H 1 .X; N 0 /, and the
extension is unique when it exists.
P ROOF. This is the main theorem of Mazur 1970b. 2
E XAMPLES
Assume that K contains the pth roots of 1. For each a; b 2 R with ab D p, there
is a well-defined finite flat group scheme Na;b over R given by the classification
of Oort and Tate (cf. 0.9). It is a finite group scheme of order p and discriminant
ap . Therefore .X; Na;b / D .RW aR/.
2. LOCAL: MIXED CHARACTERISTIC, ABELIAN VARIETIES 245
Because K contains a pth root of 1, m Ddf ord.p/=.p 1/ is an integer. We

say that Na;b splits generically if its generic fibre is isomorphic to Z=pZ. This
is equivalent to a being a nonzero .p 1/st power in R. Choose a uniformizing
parameter in R. Then the generically split group schemes of order p over R
correspond to the pairs .a; b/ with a D .p1/i , 0 i m, and b D p=a.
For example, if a D 1, then N D Z=pZ, and if a D .p1/m D .unit/p, then
N D p . Let U D R and U .i/ D fa 2 U j ord.1 a/ ig.
P ROPOSITION 1.20 Let N be a finite flat generically split group scheme of order
p over X . Then there is a nonzero map 'W N ! p , and for any choice of ' , the
map H.'/W H 1 .X; N / ! H 1 .X; p / identifies H 1 .X; N / with the subgroup
U .i/ U p =U p of H 1 .X; p / D U=U p , where i D pm ord.discN /=.p 1/.
P ROOF. Let N D Na;b , and suppose a D .p1/i . Then, for any X -scheme Y ,
N.Y / D fy 2 .Y; OY /j y p D ayg, and so y 7! y .mi/ defines a morphism
of functors N.Y / ! p .Y / and hence a nonzero map N ! p . For the proof
of the proposition, one first shows that the image of H 1 .X; N / is contained in
U .i/ U p =U p and then uses (1.14) to show that it equals this group. See Roberts
1973. 2
More explicitly, if N D Na;b with a D .p1/i , then
H 1 .X; N / D U .j / U p =U p ; where j D p ord.p/=.p 1/ pi:
For example, if a D 1 so that N D Z=pZ, then H 1 .X; N / D U .pm/ U p =U p ,

and if N D .p1/m so that N D p , then H 1 .X; N / D U .0/ U p =U p .
R EMARK 1.21 In the examples in (1.20), the map H 1 .X; N / ! H 1 .Xi ; N / is
an isomorphism for i >> 0. This is true for any finite group scheme N , as can
be easily deduced from the exact sequence
H.R/ ! H 0 .R/ ! H 1 .X; N / ! 0
arising from a resolution of N by p-divisible groups.
N OTES Theorems 1.3 and 1.14 are due to Mazur and Roberts (Mazur and Roberts 1970;
Mazur 1970a). The second proof of (1.3) and the proof of (1.14) are taken from Milne
1973. The first proof of (1.3) is new. Theorem 1.19 is due to Mazur (1970b).
2 Local results: mixed characteristic, abelian varieties

The notations are the same as in 1. Except in the last two results, X will be
endowed with its smooth topology.
Let A be an abelian variety over K, and let A be its Néron model over X . As
in Appendix C, we write Aı for the open subgroup scheme of A whose closed
fibre Aıx is connected. There is an exact sequence of sheaves on Xsm
0 ! Aı ! A ! i ˚ ! 0:
We often regard ˚ as a Gal.k s =k/-module. Recall that for any submodule of

˚ , A denotes the inverse image of in A. There is an exact sequence
0 ! Aı ! A ! i ! 0 (2.0.1)
P ROPOSITION 2.1 The map A .X / ! .x/ arising from (2.0.1) is surjec-

tive, and H r .X; A / ! H r .x; / is an isomorphism for r 1; therefore,
H r .X; A / D 0 for r 2.
P ROOF. According to Milne 1980, III 3.11, H r .X; A / D H r .x; A

x / for r >
0, and Lang’s lemma implies that H r .x; Aıx / D 0 for r > 0. Therefore the
cohomology sequence of (2.0.1) leads immediately to the result. 2
L EMMA 2.2 For any , there is an exact sequence
˚.x/ ! .˚ = /.x/ ! H 1 .X; A / ! H 1 .K; A/;
in which the last map is the restriction map; in particular, if Gal.k s =k/ acts triv-
ially on ˚ , then H 1 .X; A / ! H 1 .K; A/ is injective.
P ROOF. We first consider the case that D ˚ . Then A D A, and as A D j A,

the Leray spectral sequence for j shows immediately that the map H 1 .X; A/ !
H 1 .K; A/ is injective. In the general case, the lemma can be deduced from the
diagram
H 1 .X; A / ! H 1 .X; A/
? ?
? ?
y' y'
˚.x/ ! .˚ = /.x/ ! H 1 .x; / ! H 1 .x; ˚ /: 2
L EMMA 2.3 We have

(
0; r ¤ 1; 2
Hxr .X; A / D
.˚ = /.x/; r D 1

0 ! .x/ ! ˚.x/ ! .˚ = /.x/ ! H 1 .X; A / ! H 1 .K; A/ ! Hx2 .X; A / ! 0:
P ROOF. Consider the exact sequence

0 ! Hx0 .X; A / ! H 0 .X; A / ! H 0 .K; A/ ! Hx1 .X; A / ! H 1 .X; A / ! :
Obviously A .X / ! A.K/ is injective, which shows that Hx0 .X; A / D 0. As
H r .X; A / and H r .K; A / are both zero for r > 1, the sequence shows that
Hxr .X; A / D 0 for r > 2.
Take D ˚ , so that A D A; then A.X / ! A.K/ is an isomorphism and
(2.2) shows that H 1 .X; A/ ! H 1 .K; A/ is injective. Therefore the sequence
shows that Hx1 .X; A/ D 0. In the general case the exact sequence
0 ! Hx0 .X; ˚ = / ! Hx1 .X; A / ! Hx1 .X; A/

gives an isomorphism .˚ = /.x/ ! Hx1 .X; A /. The existence of the required
exact sequence follows from
Hx1 .X; A / !
H 1 .X; A / !
H 1 .K; A/ !
Hx2 .X; A / !
H 2 .X; A / !
0
?
?'
y
˚.x/ ! H 1 .X; A / !
.˚ = /.x/ ! H 1 .K; A/:
2
We now consider an abelian variety A over K, its dual abelian variety B, and
a Poincaré biextension W of .B; A/ by Gm . Recall (C.12) that W extends to a
0
biextension of .B ; A / by Gm if and only if 0 and annihilate each other in
the canonical pairing ˚ 0 ˚ ! Q=Z.
L EMMA 2.4 If 0 and are subgroups of ˚ 0 and ˚ that annihilate each other,
then the following diagrams commuteW
H 1 .K; B/ H 0 .K; A/ ! H 2 .K; Gm /

" # #'
0
H 1 .X; B / Hx1 .X; A / ! Hx3 .X; Gm /;
H 1 .K; B/ H 0 .K; A/ ! H 2 .K; Gm /

# " #'
0
Hx2 .X; B / H 0 .X; A / ! Hx3 .X; Gm /:
P ROOF. In the first diagram, the first vertical arrow is the restriction map, and the
second and third arrows are boundary maps H r .u; / ! HxrC1 .X; /. Since
the top pairing is defined by the restriction to u of the biextension defining the
bottom pairing, the commutativity is obvious. The proof that the second diagram
commutes is similar. 2
T HEOREM 2.5 The canonical pairing ˚ 0 ˚ ! Q=Z is nondegenerate; that is,

Conjecture C.13 holds in this case.
P ROOF. The groups and the pairing are unchanged when we replace K with its
completion. After making an unramified extension of R, we can assume that
Gal.k s =k/ acts trivially on ˚ and ˚ 0 . By symmetry, it suffices to show that the
pairing ˚ 0 ˚ ! Gm is left nondegenerate, and for this, it suffices to show that
the pairing
H 1 .x; ˚ 0 / H 0 .x; ˚ / ! H 1 .x; Q=Z/ ' Q=Z
is left nondegenerate.
The canonical pairing of ˚ 0 with ˚ is so defined that
B ! 1 .A; j G /
ExtX m
? ?
? ?
y y
i ˚ 0 ! HomX .i ˚; i Q=Z/
commutes (see C.11). Alternatively, we can regard it as being the unique homo-
morphism ˚ 0 ! Extx1 .˚; Z/ making
B ! ExtX
1
.A; j Gm /
? ?
? ?
y y
i ˚ 0 ! ExtX
1 .i ˚; i Z/

commute. From this we get a commutative diagram

H 1 .X; B/ ! Ext2X .A; j Gm /
? ?
? ?
y y
H 1 .X; i ˚ 0 / ! Ext2X .i ˚; i Z/:
These maps are used to define the two lower pairings in the following diagram
H 1 .K; B/ H 0 .K; A/ ! H 2 .K; Gm / ' Q=Z

"inj "' "'
H 1 .X; B/ H 0 .X; A/ ! H 2 .X; j Gm /
#' #surj #'
H 1 .X; i ˚ 0 / H 0 .X; ˚ / ! H 2 .X; i Z/;
and so the the diagram commutes (the upper arrows are all restriction maps).
The top pairing is nondegenerate (I 3.4), and so the lower two pairings are left
nondegenerate. This proves the theorem. 2
C OROLLARY 2.6 Suppose that 0 and are exact annihilators under the canon-
ical pairing of ˚ 0 and ˚ . Then the map
0
B ! ExtX
1
sm
.A ; Gm /
defined by the extension of W is an isomorphism (of sheaves on Xsm /.
P ROOF. See (C.14). 2
T HEOREM 2.7 Assume that 0 and are exact annihilators. Then the pairing
0
H r .X; B / Hx2r .X; A / ! Hx3 .X; Gm / ' Q=Z
0
defined by the canonical biextension of .B ; A / by Gm induces an isomor-
phism
' 0
Hx2 .X; A / ! B .X /
of discrete groups for r D 0 and an isomorphism of finite groups
0 '
H 1 .X; B / ! A .X /
for r D 1. For r ¤ 0; 1, both groups are zero.

0 0
˚ 0 .x/ .˚ 0 = /0 .x/ !
! H 1 .X; B / !
H 1 .K; B/ !
Hx2 .X; B / !
0
? ? ? ? ?
?' ?' ?a ?' ?
y y y y yb
H 1 .x; ˚ / !
H 1 .x; / !
Hx1 .x; A / !
H 0 .K; A/ !
H 0 .X; A / !
0:
The top row is the exact sequence in (2.3), and the bottom row is the dual of the
cohomology sequence of the pair X u. That the last two squares commute is
proved in (2.4). The first two vertical maps are the isomorphisms induced by the
canonical pairing between ˚ 0 and ˚ . Thus the first square obviously commutes,
and second was essentially shown to commute in the course of the proof of (2.5).
It follows from the diagram that a is injective and b is surjective. But the two
0
groups H 1 .X; B / and Hx1 .X; A / have the same order (see (2.1) and (2.3)),
and so a is an isomorphism. This in turn shows that b is an isomorphism. 2
R EMARK 2.8 (a) Once (2.7) is acquired, it is easy to return and prove (2.5): the
map H 1 .x; ˚ 0 / ! H 0 .x; ˚ / can be identified with the isomorphism
H 1 .X; B 0 / ! Hx1 .X; Aı / given by the (2.6).
b D SpecR.
(b) Let X b Then it follows from (I 3.10) that the maps Hxr .X; A / !
b ; A / are isomorphisms for all r, and that H r .X; A / ! H r .X
Hxr .X b ; A / is
an isomorphism for all r > 0. The map A.X / ! A.X b / is injective and its image
b b
includes the torsion subgroup of A.X/; A.X/ is the completion of A.X / for the
topology of subgroups of finite index.
0
(c) When R is complete, B .X / is compact. Therefore in this case the pair-
ing induces dualities between:
0
the compact group B .X / and the discrete group Hx2 .X; A /I
0
the finite group H 1 .X; B / and the finite group A .X /.
n 0 n
Write Bfng for the complex B ! B n˚ and Afng for the complex A˚n !
0
Aı . The pairings B ˝L Aı ! Gm Œ1 and B n˚ ˝L A˚n ! Gm Œ1 defined by
a Poincaré biextension induce a pairing Bfng ˝L Afng ! Gm in the derived
category of sheaves on Xsm (see Grothendieck 1972, VIII 2).
P ROPOSITION 2.9 The map Bfng ˝L Afng ! Gm defines nondegenerate pair-

ings
H r .X; Bfng/ Hx3r .X; Afng/ ! Hx3 .x; Gm / ' Q=Z
of finite groups for all r .
P ROOF. From the exact sequences of complexes

0
0 ! B n˚ Œ1 ! Bfng ! B ! 0
0 ! Aı Œ1 ! Afng ! A˚n ! 0

we get the rows of the diagram
0
! H r1 .X; B n˚ / ! H r .X; B.n// ! H r .X; B/ !
? ? ?
?' ? ?'
y y y
! Hx3r .X; A˚n / ! Hx3r .X; A.n// ! Hx2r .X; Aı / ! :
Since the diagram obviously commutes, the theorem follows from (2.7). 2
C OROLLARY 2.10 Assume that n is prime to the characteristic of k or that A has

semistable reduction. Then for all r , there is a canonical nondegenerate pairing
of finite groups
Hxr .Xfl ; Bn / H 3r .Xfl ; An / ! Hx3 .Xfl ; Gm / ' Q=Z:

n 0 n
P ROOF. The hypothesis implies that B ! B n˚ and A˚n ! Aı are surjective
when regarded as a maps of sheaves for the flat topology (see C.9). Hence Bn
Bfng and An Afng, and so
H r .Xfl ; Bn / H r .Xfl ; Bfng/ H r .Xsm ; Bfng/
and
Hxr .Xfl ; An / Hxr .Xfl ; Afng/ Hxr .Xsm ; Afng/: 2
Curves over X
By exploiting the autoduality of the Jacobian, it is possible to use (2.9) to prove a
duality theorem for a curve over X .
T HEOREM 2.11 Let W Y ! X be a proper flat map whose fibres are pure of
dimension one. Assume that the generic fibre YK is smooth and connected, that
the special fibre Yx is connected, and that there is a section to . Assume further
that PicY =X D J , where J is the Néron model of the Jacobian of YK . Then
there is a canonical duality of finite groups
H r .Y; n / HY5r
x
.Y; n / ! Hx3 .X; Gm / ' Q=Z:
P ROOF. We use the Leray spectral sequence of . Under the hypotheses,
R0 n ' n ;
n
R1 n ' Ker.J ! J /;
R2 n ' Z=nZ;
Rr n D 0; for r > 2:
Moreover J D J ı . On taking A D J D B in (2.9), we find that H r .X; R1 n /

is dual to Hx3r .X; R1 n / for all r. The result can be obtained by combining
this duality with the duality of H r .X; R0 n / and Hx3r .X; R2 n /. 2
For conditions on Y =X ensuring that the hypotheses of the theorem hold, see
the last few paragraphs of Appendix C. Our hypotheses are surely too stringent.
Because of this, we make the following definition. Let X be the spectrum of an
excellent Henselian discrete valuation ring (not necessarily of characteristic zero)
with finite residue field, and let W Y ! X be a proper flat morphism whose
generic fibre is a smooth curve. If there is a canonical pairing
R n R n ! Gm Œ2:

extending that on the generic fibre and such that the resulting pairing
H r .Y; n / HY5r
x
.Y; n / ! H 3 .X; Gm / ' Q=Z
is nondegenerate, then we shall say that the local duality theorem holds for Y =X
and n.
N OTES This section is based on McCallum 1986.
3 Global results: number field case

Throughout this section, X will be the spectrum of the ring of integers OK in a
number field K. For an open subscheme U of X , U Œ1=n denotes
Spec .U; OX /Œ1=n, and Hcr .U; / denotes the flat cohomology group with
compact support as defined in (0.6a) (thus, it takes account of the infinite primes).
Finite sheaves
Let U be an open subscheme of X . As Gm is smooth,
Hcr .Ufl ; Gm / D Hcr .Uet ; Gm /;

'
and so (see II 3) there is a canonical trace map Hc3 .U; Gm / ! Q=Z. Therefore,
for any sheaf F on U , there is a canonical pairing
ExtrU .F; Gm / Hc3r .U; F / ! Hc3 .U; Gm / ' Q=Z
(see 0.4e).
Let f W Ufl ! Uet be the morphism of sites defined by the identity map. Recall
Milne 1980, V 1, that the constructible sheaves on Uet are precisely those sheaves
that are representable by étale algebraic spaces of finite-type over U ; moreover,
if FQ represents F on Uet , then it represents f F on Ufl .
T HEOREM 3.1 Let U be an open subscheme of X , and let F be a sheaf on Ufl

such that nF D 0 for some integer n. Assume
(i) the restriction of F to U Œ1=nfl is represented by an étale algebraic space
of finite-type over U Œ1=n;
(ii) for each v 2 U XU Œ1=n, the restriction of F to .SpecOv /fl is represented
by a finite flat group scheme.
Let F D be the sheaf on U such that F D jU Œ1=n D f R HomU Œ1=net .F; Gm /
and F D jV D HomVfl .F; Gm / for any open subscheme V of U where F jV
is represented by a finite flat group scheme. Then there are canonical maps
3. GLOBAL RESULTS: NUMBER FIELD CASE 253
F D ! RHomU .F; Gm /, hence H r .U; F D / ! ExtrU .F; Gm /, and the result-

ing pairing
H r .U; F D / Hc3r .U; F / ! Hc3r .U; Gm / ' Q=Z
P ROOF. We first note that on V Œ1=n D V \ U Œ1=n, F is represented by a

finite flat étale group scheme whose order is prime to the residue characteristics.
Therefore
RHomV Œ1=net .F; Gm / D HomV Œ1=net .F; Gm /;
and so the requirements on F D coincide on V Œ1=n, which shows that F D exists.
For all r,
Hcr .U Œ1=net ; F / D Hcr .U Œ1=nfl ; F /;

H r .U Œ1=nfl ; F D / D H r .U Œ1=net ; F D /:
Therefore, for the restriction of F to U Œ1=n, the theorem becomes (II 3.3). To
pass from U Œ1=n to the whole of U , one uses the diagram
L
H r .U; F D / !
! H r .U Œ1=n; F D / ! HvrC1 .Ohv ; F D / !

v2U XU Œ1=n
? ? ?
? ?' ?'
y y y
L
Hc3r .U; F / !
! Hc3r .U Œ1=n; F / !
H 2r .Ohv ; F / !

v2U XU Œ1=n
and (1.3). 2
C OROLLARY 3.2 Let N be a finite flat group scheme over U , and let N D be its
Cartier dual. Then
H r .U; N D / Hc3r .U; N / ! Hc3 .U; Gm / ' Q=Z
is a nondegenerate pairing of finite groups for all r .
P ROOF. When the sheaf F in (3.1) is taken to be that defined by N , then F D is

the sheaf defined by N D . 2
C OROLLARY 3.3 Let N be a quasi-finite flat separated group scheme over U ,

and let nN D 0. Assume that there exists an open subscheme V of U such that
(i) V contains all points v of U whose residue characteristic divides nI
(ii) N jV is finite;
(iii) if j denotes the inclusion of V Œ1=n into U Œ1=n, then the canonical map
N jU Œ1=net ! j j .N jV Œ1=net / is an isomorphism.
Let N D D HomUfl .N; Gm /. Then the canonical pairing
P ROOF. Because N jV is finite, N D jV is the Cartier dual of N jV . Therefore the

theorem shows that H r .V; N D / is finite and dual to Hc3r .V; N /. The corollary
therefore follows from
L
H r .U; F D / !
! H r .V; F D / ! HvrC1 .Ohv ; F D / !

v2U XV
? ? ?
? ? ?'
y y y
L
Hc3r .U; F / !
! Hc3r .V; F / ! H 2r .Ohv ; F / ! :
v2U XV
and (II 1.10b). 2
Let A be an abelian variety over K, and let A and B be the Néron minimal
models over U of A and its dual B. Let n be an integer such that A has semistable
reduction at all v dividing n. There are exact sequences
0
0 ! Bn ! B ! B n˚ ! 0
0 ! An ! A˚n ! Aı ! 0:
The Poincaré biextension of .B; A/ by Gm extends uniquely to biextensions of

0
.B; Aı / by Gm and of .B n˚ ; A˚n / by Gm . Therefore (cf. 1), we get a canonical
pairing
Bn A n ! Gm :
C OROLLARY 3.4 Let B n and An be as above. Then
H r .U; Bn / Hc3r .U; An / ! Hc3 .U; Gm / ' Q=Z
is a nondegenerate pairing of finite groups for all r .
P ROOF. Over the open subset V where A has good reduction, An is a finite flat
group scheme with Cartier dual Bn , and so over V , the corollary is a special case
of (3.2). To pass from V to U , use (2.10). 2
3. GLOBAL RESULTS: NUMBER FIELD CASE 255
We extend (II 2.13) to the flat site. Let N be a quasi-finite flat separated group
f
scheme over U . For each closed point v 2 U , let nv be the order of the maximal
f f
finite subgroup scheme Nv of N U Spec.Ovh /, and let dv be the discriminant
f
of Nv over Ovh . Also, we set
ŒH 0 .U; N /ŒH 2 .U; N /

.U; N / D ;
ŒH 1 .U; N /ŒH 3 .U; N /
ŒHc0 .U; N /ŒHc2 .U; N /
c .U; N / D :
ŒHc1 .U; N /ŒHc3 .U; N /
T HEOREM 3.5 Let N be quasi-finite, flat, and separated over U . Then

Q Q f Q ŒN.Kv /
.U; N / D jŒN.Ks /jv .Ovh W dfv /1=nv
v2XXU v2U v arch ŒH 0 .K ; N /
v
and Q Q
f
c .U; N / D .RW dfv /1=n ŒN.Kv /:
v2U v arch
P ROOF. Let V be an open subset of U such that N jV is finite and has order prime
to the residue characteristics of V , so that, in particular, N jV is étale. The exact
sequence
Q
! Hvr .U; N / ! H r .U; N / ! H r .V; N / !
v2U XV
Q
shows that .U; N / D .V; N / h
v2U XV v .Ov ; N /, and (II 2.13) and (1.17b)
show respectively that
Y ŒN.Kv /
.V; N / D
ŒH 0 .K v ; N /jŒN.Ks /jv
v arch
f f
and v .Ovh ; N / D jŒN.Ks /j1
v .RW dv /
1=nv . The formula in (a) follows imme-
diately.
The exact sequence
L
! Hcr .V; N / ! Hcr .U; N / ! v2U XV H r .Ovh ; N / !
Q
shows that c .U; N / D c .V; N / v .Ovh ; N / and (II 2.13) and (1.17a) show
Q f f
respectively that c .V; N / D v arch ŒN.Kv / and .Ovh ; N / D .RW dv /1=n . 2
Néron models
Let A be an abelian
L variety over K, and let A be its Néron model. Then
A=A Ddf ˚ D v iv ˚v (finite sum) where ˚v D iv .Av =A0v /.
ı
P ROPOSITION 3.6 Let be a subgroup of ˚ , and let A be the corresponding

subscheme of A.
(a) The group H 0 .U; A / is finitely generated; for
L r > 0,r H .U; A / is
r
torsion and of cofinite-type; the map H .U; A / ! v arch H .Kv ; A/ is sur-

r
jective for r D 2 andL an isomorphism for r > 2.

v arch H .Kv ; A / ! Hc .U; A / is an isomorphism;
(b) For r < 0, r r
Hc .U; A / is finitely generated; Hc .U; A / is an extension of a torsion group

0 1
by a subgroup which has a natural compactification; Hc2 .U; A / is torsion and

of cofinite-type; for r 3, Hcr .U; A / D 0.
P ROOF. Fix an integer m, and let V be an open subscheme of U such that m

is invertible on U and A is an abelian scheme over V . Then all statements are
proved in (II 5.1) for AjV and m. The general case follows by writing down the
usual exact sequences. 2
Let B be the dual

L abelian variety to A, and let B be its
L Néron model. 0 Let
B=B Ddf ˚ D 0 0
v iv ˚v . For any subgroups D v iv v and D
0
L 0 of ˚ and ˚ 0 , the Poincaré biextension over K extends to a biextension
i
v v v
over U if and only if each v annihilates each v0 in the canonical pairing. In this
case we get a map
0
B ˝L A ! Gm Œ1:
T HEOREM 3.7 Suppose that v and v0 are exact annihilators at each closed
point v .
0
(a) The group H 0 .U; B /tors is finite; the pairing
0
H 0 .U; B / Hc2 .U; A / ! Q=Z
is nondegenerate on the left and its right kernel is the divisible subgroup of
Hc2 .U; A ).
0
(b) The groups H 1 .U; B ) and Hc1 .U; A /tors are of cofinite-type, and the
pairing
0
H 1 .U; B / Hc1 .U; A /t ors ! Q=Z
annihilates exactly the divisible groups.
(c) If the divisible subgroup of X1 .K; A/ is zero, then the compact group
0
H 0 .U; B /^ (completion for the topology of subgroups of finite index) is
dual to the discrete torsion group Hc2 .U; A /.
4. LOCAL: MIXED CHARACTERISTIC, PERFECT RESIDUE FIELD 257
P ROOF. Fix an integer m, and choose an open subscheme V of U on which m is

invertible and A and B have good reduction. Theorem II 5.2 proves the result over
V for the m-components of the groups. To pass from there to the m-components
of the groups over U , use (2.7). As m is arbitrary, this completes the proof. 2
Curves over U
For a proper map W Y ! U and sheaf F on Yfl , we define Hcr .Y; F / to be
Hcr .U; R F /.
T HEOREM 3.8 Let W Y ! U be a proper flat map whose fibres are pure of di-
mension one and whose generic fibre is a smooth geometrically connected curve.
Assume that for all v 2 U , Y U SpecOhv ! SpecOhv satisfies the local duality
theorem for n (see 2). Then there is a canonical nondegenerate pairing of finite
groups
H r .Y; n / Hc5r .Y; n / ! Hc3 .U; Gm / ' Q=Z:
P ROOF. Choose an open subscheme V of U such that n is invertible on V and
j 1 .V / is smooth. For j 1 .V / the statement becomes that proved in The-
orem II 7.7. Let Z D Y X YV . To pass from V to U , use the exact sequences
! HZr .Y; n / ! H r .Y; n / ! H r .YV ; n / !
and
L
! Hcr .YV ; n / ! Hcr .Y; n / ! H r .Y U Spec.Ovh /; n / !
v2U XV
L
and note that HZr .Y; n / D v2U XV Hvr .Y U Spec.Ovh /; n /. 2
N OTES Theorem 3.1 was proved by the author in 1978. Earlier Artin and Mazur had
announced the proof of a flat duality theorem over X (neither the statement of the theorem
nor its proof have been published, but two corollaries are stated3 in Mazur 1972, 7.2, 7.3;
I believe that their original theorem is the special case of (3.3) in which U D X and n is
odd).
4 Local results: mixed characteristic, perfect residue field

In this section we summarize the results of Bégeuri 1980. Throughout, X will
be the spectrum of a complete discrete valuation ring R whose field of fractions
K is of characteristic zero, and whose residue field k is perfect of characteristic
p ¤ 0. (Essentially the same results should hold if R is only Henselian.) We let
m be the maximal ideal of R, and we let Xi D SpecR=miC1 .
3 And made essential use of.
Some cohomological properties of K

P ROPOSITION 4.1 If k is algebraically closed, then for any torus T over K ,
H r .K; T / D 0 all r 0.
P ROOF. Let L be a finite Galois extension of K with Galois group G. Then

H 1 .G; L / D 0 by Hilbert’s theorem 90, and H 2 .G; L / D 0 because the
Brauer group of K is trivial .K is quasi-algebraically closed, Shatz 1972, p116).
These two facts show that L is a cohomologically trivial G-module (Serre 1962,
IX 5, Thm 8). Choose L to split T . Then Hom.X .T /; L / D T .L/, and
(ibid. Thm 9) shows that Hom.X .T /; L / is also cohomologically trivial be-
cause Ext1 .X .T /; L / D 0. 2
C OROLLARY 4.2 Assume that k is algebraically closed, and let N be a finite

group scheme over K .
(a) For all r 2, H r .K; N / D 0.
(b) Let K 0 be a finite Galois extension of K , and let G D Gal.K 0 =K/. Then
HTr .G; H 1 .K 0 ; N // is finite for all r 2 Z , and is isomorphic to
HTrC2 .G; N.K 0 //. The canonical homomorphism H0 .G; H 1 .K 0 ; N // !
H 1 .K; N / deduced from the corestriction map is an isomorphism.
P ROOF. (a) Resolve N by tori,
0 ! N ! T0 ! T1 ! 0;
and apply the proposition.

(b) From the above resolution, we get an exact sequence
0 ! N.K 0 / ! T0 .K 0 / ! T1 .K 0 / ! H 1 .K 0 ; N / ! 0:
Since the middle two G-modules are cohomologically trivial, the iterated cobound-
ary map is an isomorphism HTr .G; H 1 .K 0 ; N // ! HTrC2 .G; N.K 0 //. The last
statement is proved similarly (see Bégeuri 1980, p34). 2
The algebraic structure on H r .X; N /

For any k-algebra , let Wi ./ be the ring of Witt vectors over of length i,
and let W ./ be the full Witt ring. For any scheme Y over W .k/ and any i, the
Greenberg realization of level i, Greeni .Y /, of Y is the scheme over over k such
that
Greeni .Y /./ D Y.Wi .//
for all k-algebras (see Greenberg 1961). Note that R has a canonical structure
as a W .k/-algebra, and so for any scheme Y over X , we can define Gi .Y / to be the
Greenberg realization of level i of the restriction of scalars of Y , ResX=Spec.W .k// Y .
Then Gi .Y / is characterized by the following condition: for any k-algebra ,
Gi .Y /./ D Y.R ˝W .k/ Wi .//:
In particular, Gi .Y /.k/ D Y.R=p i R/ D Y.Xi1 /. Note that G1 .Y / D Y ˝R k D

Yk . For varying i, the Gi .Y / form a projective system G.Y / D .Gi .Y //. The
perfect group scheme associated with Gi .Y / will be denoted by4 Gi .Y /. Thus
Gi .Y /./ D Y.R ˝W .k/ Wi .//
for any perfect k-algebra and Gi .Y /.k/ D Y.R=p i R/. We let G.Y / be the
perfect pro-group scheme .Gi .Y //.
When G is a smooth group scheme over X , we let5 V .!G / be the vector
group associated with the R-module !G of invariant differentials on G.
P ROPOSITION 4.3 Let G be a smooth group scheme over R. For all i 1,

Gi .G/ is a smooth group scheme over k , and for all i 0 i , there is an exact
sequence of k -groups
0 ! Gi .V .!G // ! GiCi 0 .G/ ! Gi 0 .G/ ! 0:
In particular, GiC1 .G/ ! Gi .G/ is surjective with kernel !G ˝R k , and Gi .G/

is an extension of Gk by a smooth connected unipotent group. The group scheme
Gi .G/ is connected if and only if its special fibre is connected. The dimension of
Gi .G/ is ei dim.Gk / where e is the absolute ramification index of R.
P ROOF. We may assume that k is algebraically closed and apply Bégueri 1980,
4.1.1. 2
L EMMA 4.4 Let

'
0 ! N ! G0 ! G1 ! 0
be an exact sequence of R-groups with G0 and G1 smooth and connected and

N finite. For all i 1, the k -group Coker.Gi .'// is smooth, and when k is
algebraically closed its group of k -points is H 1 .Xi1 ; N /.
4 In the original, this is denoted Gi .Y /.
5 In the original, this is denoted V.!G /.
P ROOF. The first statement follows from the fact that Gi .'/ is a homomorphism
of smooth group schemes over k. For the second, note that H r .Xi1 ; G/ D
H r .k; Gk / D 0 for r > 0, and so we have a diagram
Gi .G0 /.k/ ! Gi .G1 /.k/ ! Coker.Gi .'/.k// ! 0
?
?
y
G0 .Xi1 / ! Gi .Xi1 / ! H 1 .Xi1 ; N / ! 0 2
Define HQ 1 .Xi ; N / to be the sheaf on Spec.k/qf associated with the presheaf

7! H 1 .Xi ˝W .k/ W ./; N /. Then the lemma realizes HQ 1 .Xi ; N / as an
algebraic group, and the next lemma shows that this realization is essentially in-
dependent of the choice of the resolution.
L EMMA 4.5 Let
'0
0 ! N ! G00 ! G10 ! 0
be a second resolution of N by smooth algebraic groups. Then there is a canonical
'
isomorphism Coker.Gi .'// ! Coker.Gi .' 0 //.
P ROOF. It is easy to construct a diagram

0 ! G00 ! G00 ! 0
? ? ?
? ? ?
y y y
' 00
0 ! N ! G0 ˚ G00 ! G ! 0
? ?
? ?
y y
'
0 ! N ! G0 ! G1 ! 0
with G a smooth algebraic group. When we apply Gi , the resulting diagram gives
an isomorphism

Coker.Gi .' 00 // ! Coker.Gi .'//;
and a similar construction gives an isomorphism

Coker.Gi .' 00 // ! Coker.Gi .' 0 //: 2
We now regard HQ 1 .Xi ; N / as an algebraic group, and we write HQ 1 .X; N /

for the pro-algebraic group .HQ 1 .Xi ; N //i
0 .
For the definition of the absolute different D of a finite group N scheme over
R, we refer the reader to Raynaud 1974, Appendice. It is an ideal in R.
T HEOREM 4.6 Let N be a finite flat group scheme of order a power of p over X .
For all i 0, the smooth algebraic k -group HQ 1 .Xi ; N / is affine, connected, and
unipotent. There exists an integer i0 such that HQ 1 .X; N / ! HQ 1 .Xi ; N / is an
isomorphism for all i i0 . The group scheme HQ 1 .X; N / has dimension ord.D/
where D is the different of N .
P ROOF. We may assume that the residue field is algebraically closed and apply
Bégeuri 1980, 4.2.2. 2
P ROPOSITION 4.7 A short exact sequence
0 ! N 0 ! N ! N 00 ! 0
of finite flat p -primary group schemes gives rise to an exact sequence of algebraic
groups
0 ! G.N 0 / ! G.N / ! G.N 00 / ! HQ 1 .X; N 0 / ! HQ 1 .X; N / ! HQ 1 .X; N 00/ ! 0:
Bégeuri 1980, 4.2.3. 2
We write H1 .Xi ; N / and H1 .X; N / for the perfect algebraic groups asso-
ciated with HQ 1 .Xi ; N / and HQ 1 .X; N /. Suppose that k is algebraically closed.
If
0 ! N ! G0 ! G1 ! 0
is a smooth resolution of N and i is so large that N.R/ \ p i G0 .R/ D 0, then the
kernel and cokernel of the map
i i/
G.G0 /.R/.p / ! G.G1 /.R/.p
are N.R/ and H1 .X; N /.k/ respectively (ibid. p44–45).
The algebraic structure on H 1 .K; N /

Let T be a torus over K. According to Raynaud 1966, T admits a Néron model
over X W this is a smooth group scheme T over X (not necessarily of finite type)
such that T .Y / D T .YK / for all smooth X -schemes Y . Write G.T / for G.T /.
It is a perfect pro-algebraic group over k whose set of connected components
0 .G.T // is a finitely generated abelian group, equal to the set of connected
components of the special fibre of T .
L EMMA 4.8 Let N be a finite group scheme over K , and let

'
0 ! N ! T0 ! T1 ! 0
be a resolution of N by tori. The cokernel of G.'/W G.T0 / ! G.T1 / is a

pro-algebraic perfect group scheme, and when k is algebraically closed it has
H 1 .X; N / as its group of k -points.
P ROOF. Bégeuri 1980, 4.3.1. 2
We write H1 .K; N / for the cokernel of G.'/. It is a perfect pro-algebraic

group scheme with group of points H 1 .K; N / when k is algebraically closed.
An argument as in the proof of (4.5) shows that H1 .K; N / is independent of the
resolution. The identity component of H1 .K; N / is unipotent.
P ROPOSITION 4.9 Let N be a finite flat p -primary group scheme over X . Then
the standard resolution defines a closed immersion
H1 .X; N / ! H1 .K; N /:
Bégeuri 1980, 4.4.4. 2
T HEOREM 4.10 For any finite K -group N , the perfect group scheme H1 .K; N /
is affine and algebraic. Its dimension is ord.ŒN /, where ŒN is the order of N .
P ROOF. The basic strategy of the proof is the same as that of the proof of (I 2.8);
see Bégeuri 1980, 4.3.3. 2
The reciprocity isomorphism

Assume first that k is algebraically closed. For any finite extension K 0 =K, let
UK 0 D G.Gm;R0 / where R0 is the ring of integers in K 0 . Then the norm map
NR0 =R W ResR0 =R Gm;R0 ! Gm;R induces a surjective map UK 0 ! UK of affine
k-groups. Let VK 0 be the kernel of this map, and let VKı 0 be the identity compo-
nent of VK 0 . Then we have an exact sequence
0 ! 0 .VK 0 / ! UK 0 =VKı 0 ! UK ! 0:
Assume that K 0 is Galois over K, and let t 0 be a uniformizing parameter in K 0 .

The homomorphism
Gal.K 0 =K/ab D H 2 .Gal.K 0 =K/; Z/ ! .UK 0 =VKo 0 /.k/;

sending 2 Gal.K 0 =K/ to the class of .t 0 /=t 0 in UK 0 allows us to identify the

preceding exact sequence with an exact sequence
N
0 ! H 2 .Gal.K 0 =K/; Z/ ! UK 0 =VKı 0 ! UK ! 0:
On passing to the inverse limit over the fields K 0 , we get an exact sequence
0 ! Gal.K ab =K/ ! lim UK 0 =VKı 0 ! UK ! 0:

As UK is connected, this sequence defines a continuous homomorphism
recK W 1 .UK / ! Gal.K ab =K/
and the main result of Serre 1961 is that this map is an isomorphism.
It is also possible to show that 1 .U / Ddf lim 1 .UK 0 / is a class formation,
!
and so define recK as in (I 1).
Recall (Serre 1960, 5.4) that for any perfect algebraic group G and finite
perfect group N , there is an exact sequence
0 ! Ext1k .0 .G/; N / ! Ext1k .G; N / ! Homk .1 .G/; N / ! 0:

'
In particular, when G is connected Ext1k .G; N / ! Hom.1 .G/; N /. (We are
still assuming that k is algebraically closed.) Therefore, recK gives rise to an
isomorphism
'
Hom.Gal.K ab =K/; Z=p n Z/ ! Hom.1 .UK /; Z=p n Z/

H 1 .K; Z=p n Z/ Ext1K .UK ; Z=p n Z/:
If we assume that K contains the p n th roots of 1, and we replace Z=p n Z with
p n .K/, then the isomorphism becomes
'
˚n W H 1 .K; p n / ! Ext1K .UK ; p n /:
Both groups have canonical structures of perfect algebraic groups.
P ROPOSITION 4.11 The map ˚n is a morphism of perfect algebraic groups.
P ROOF. See Bégeuri 1980, 5.3.2. 2
When we drop the assumption that k is algebraically closed, we obtain an

isomorphism
recK W .UK / ! Gal.K ab =K/
where .UK / is the maximal constant quotient of 1 .UK /. See Hazewinkel 1969.
Duality for finite group schemes over K

T HEOREM 4.12 Let N be a finite group scheme over K ; then there is a canonical
isomorphism of connected perfect unipotent groups H1 .K; N /ı ! .H1 .K; N D /ı /_ .
P ROOF. We can assume that k is algebraically closed and apply Bégeuri 1980,
6.1.6. 2
This result can be improved by making use of derived categories (ibid. 6.2).
Assume that k is algebraically closed, and let
Mn D category of finite group schemes over K killed by p n ;
Qn D category of perfect pro-algebraic groups over k killed by p n ;
Sn D category of sheaves on .Spec k/pf killed by p n .
Let C W D b .Mn / ! D b .Mn /, SW D b .Qn / ! D b .Qn /, and BW D b .Sn / ! D b .Sn /
be the functors defined respectively by Cartier duality, Serre duality, and Breen-
Serre duality (see 0; here D b .
/ denotes the derived category obtained from the
category K b .
/ of bounded complexes and homotopy classes of maps). Then
H1 W Mn ! Qn admits a left derived functor, and we have a commutative diagram
(up to an isomorphism of functors):
LH1 can
D b .Mn / ! D b .Mn / ! D b .Mn /
? ? ?
? ? ?
yC yS yB (4.12.1)
LH1 can
D b .Mn / ! D b .Mn / ! D b .Mn /I

moreover, .canıLH1 /.N / ! RH0 .N /Œ1. See Bégeuri 1980, 6.2.4.
Duality for finite group schemes over R

T HEOREM 4.13 For any finite flat p -primary group scheme N over X , there is
a canonical isomorphism of k -groups
'
H1 .X; N / ! .H1 .K; N D /ı =H1 .X; N D //t :
P ROOF. Ibid. 6.3.2.. 2
Duality for tori

Let T be a torus over K, and let T be its Néron model over X .
T HEOREM 4.14 The pairing H 0 .K; X .T //T .K/ ! Z defines isomorphisms

'
H 0 .K; X .T // ! Hom.0 .Tk /; Z/
'
H 1 .K; X .T // ! Ext1k .0 .Tk /; Z/ (finite groups)
'
H 2 .K; X .T // ! Homcts .1 .T .K//; Q=Z/:
P ROOF. Ibid. 7.2. 2
Duality for abelian varieties

Let A be an abelian variety over K, and let A be its Néron model over X . We
write G.A/ for G.A/ and i .A/ for i .G.A//.
T HEOREM 4.15 Let A be an abelian variety over K .

(a) The pairing 0 .Ak / 0 .Atk / ! Q=Z defined in (C.11) is nondegenerate.
(b) There is a canonical isomorphism

H 1 .K; At / ! Ext1k .G.A/; Q=Z/:
P ROOF. (a) We can assume that k is algebraically closed, and in this case the
result is proved in Bégeuri 1980, 8.3.3.
(b) From (ibid. 8.3.6) we know that the result holds if k is algebraically
closed; to deduce the result in the general case, apply the Hochschild-Serre spec-
tral sequence to the left hand side and the spectral sequence (I 0.17) to the right
hand side. 2
C OROLLARY 4.16 Assume that Ak is connected. Then there is a nondegenerate

pairing of Gal.K un =K/-modules
H 1 .K un ; At / 1 .A/ ! Q=Z:
P ROOF. Again we can assume that k is algebraically closed. As we noted above,

for any connected perfect group scheme G over k and finite perfect group scheme
N , Ext1k .G; N / D Homk .1 .G/; N /. This shows that Ext1k .G.A/; Q=Z/ D
Homk .1 .G.A/; Q=Z/, and so the result follows from the theorem. 2
N OTES The results in this section are due to Bégeuri 1980. Partial results in the same
direction were obtained earlier by Vvedens’kii (see Vvedens’kii 1973, 1976 and earlier
papers).
5 Two exact sequences

We write down two canonical short resolutions that are of great value in the proof
of duality theorems in characteristic p. Throughout, X will be a scheme of char-
acteristic p ¤ 0.
The first exact sequence

The first sequence generalizes the sequences
1F
0 ! Z=pZ ! Ga ! Ga ! 0
F
0 ! ˛p ! Ga ! Ga ! 0
to any group scheme N D that is the Cartier dual of a finite group scheme of height
one. Note that in each sequence, Ga is the cotangent space to N .
Let N be a finite flat group scheme over X of height 1, and let eW X ! N
be the zero section. Let I ON be the ideal defining the closed immersion e
(so that .ON =I/je.X / D OX ), and let InfX1 .N / Ddf Spec.ON =I 2 / be the
first order infinitesimal neighbourhood of the zero section. Then I=I 2 is the
cotangent space !N of N over X . Locally on X , there is an isomorphism of
pointed schemes
N Spec.OX ŒT1 ; :::; Tm =.T1p ; :::; Tmp //;
and therefore
I=I 2 .T1 ; :::; Tm /=.T12 ; :::; Tm2 /
(see Messing 1972, II 2.1.2). In particular, !N is a locally free OX -module of
finite rank, and hence it defines a vector group V .!N / over X . We shall almost al-
ways write !N for V .!N /. This vector group represents MorX-ptd .InfX1 .N /; Gm /
viewed as a functor of schemes over X . (The notation X -ptd means that the mor-
phisms are required to respect the canonical X -valued points of the two schemes.)
The Cartier dual N D of N represents HomX .N; Gm /, and so the inclusion
InfX1 .N /,!N defines a canonical homomorphism .N /W N D ! !N .
Recall from 0 that the Verschiebung is a map V W N .p/ ! N . It induces
a map !N ! !N .p/ , and on combining this with the canonical isomorphism
.p/ .p/
!N .p/ ' !N , we obtain a homomorphism '0 W !N ! !N . The relative
Frobenius morphism for the vector group !A over X is also a homomorphism
.p/
'1 W !N ! !N , and we define ' D '0 '1 .
5. TWO EXACT SEQUENCES 267
T HEOREM 5.1 For any finite flat group scheme N of height one over X , the
sequence
' .p/
0 ! N D ! !N ! !N ! 0
is exact.
For N D p and N D ˛p the sequence becomes one of those listed above.

In the case that X is an algebraically closed field, every N has a composition
series whose quotients are isomorphic to p or to ˛p , and the theorem can be
proved in this case by induction on the length of N (see Artin and Milne 1976,
p115).
L EMMA 5.2 The sequence is a complex, that is, ' ı D 0.
P ROOF. This can be proved by direct calculation (ibid., p114). 2
The next lemma shows that, when X in Noetherian, the theorem follows from
the case that X is an algebraically closed field.
L EMMA 5.3 Let

'
0 ! G 0 ! G ! G 00 ! 0
be a complex of flat group schemes of finite type over a Noetherian scheme X .
Assume that for all geometric points x of X , the sequence of fibres
0 ! Gx0 ! Gx ! Gx00 ! 0
is exact. Then the original sequence is exact.
P ROOF. The faithful flatness of the 'x , combined with the local criterion for
flatness Grothendieck 1971, IV 5.9, implies that ' is faithfully flat. Thus Ker.'/
is flat and of finite type, and by assumption factors through it. Now the same
argument shows that W N ! Ker.'/ is faithfully flat. Finally, the kernel of is
a group scheme over X whose geometric fibres are all zero, and hence is itself
zero. 2
We now complete the proof of (5.1). It suffices to check the exactness of the
sequence locally on X , and so we can assume that X is quasi-compact. Then
there will exist a Noetherian scheme X0 , a finite flat group scheme N0 over X0 ,
and a map X ! X0 such that N D N0 X0 X . We know that the sequence for N0
is exact, but since the construction of the sequence commutes with base change,
this proves that the sequence for N is exact.
E XAMPLE 5.4 Suppose N has order p. Then it can be written N D N0;a L in the
Oort-Tate classification (0.9) with L an invertible sheaf on X and a 2 L˝1p . Its

L_
dual N D D Na;0 . The cotangent sheaf !N is equal to L_ , and when we identify
L_.p/ with L_˝p , the map ' in the sequence in the theorem becomes
z 7! z ˝p a ˝ zW L_ ! L_˝p :
In this case the exactness of the sequence is obvious from the description Oort
and Tate give of the points of N D (see 0.9d).
R EMARK 5.5 The theorem shows that every finite flat group scheme N over X
whose dual has height one gives rise to a locally free OX -module V of finite rank
and to a linear map '0 W V ! V .p/ . To recover N from the pair .V; '0 /, simply
form the kernel of '0 '1 where '1 is the relative Frobenius of V (regarded as
a vector group). These remarks lead to a classification of finite group schemes of
this type that is similar, but dual, to the classification of finite flat group schemes
of height one by their p-Lie algebras (see Demazure and Gabriel 1970, II, 7).
The second exact sequence

We now let W X ! S be a smooth map of schemes of characteristic p, and we as-
sume that S is perfect. Write X 0 for X regarded as an S-scheme by means of 0 D
Fabs ı. Because S is perfect, we can identify .X 0 ; 0 / with .X .1=p/; .1=p/ / (see
0), and when we do this, the relative Frobenius map FX .1=p/ =S W X .1=p/ ! X
becomes identified with the absolute Frobenius map F D Fabs . For exam-
ple, if S D SpecR and X D SpecA for some R-algebra iW R ! A, then
X 0 D SpecA, with A regarded as an R-algebra by means of a 7! i.a/p , and
the Frobenius map X X 0 corresponds to a 7! ap W A ! A. We write
1
˝X=S;cl for the sheaf of closed differential forms on X relative to S, that is,
1
˝X=S;cl D Ker.d W ˝X=S1 ! ˝X=S1 /. We regard ˝X=S1 1
and ˝X=S;cl as sheaves
on Xet .
Again N is a finite flat group scheme of height one on X , and we let n be the
Lie algebra of N (equal to the tangent sheaf of N over X /.
T HEOREM 5.6 Let f W Xfl ! Xet be the morphism of sites defined by the identity
map. Then Rr f N D 0 for r ¤ 1, and there is an exact sequence
0 ! R1 f N ! n ˝OX ˝X
1 1
0 =S;cl ! n ˝OX ˝X=S ! 0:
We first show that Rr f N D 0 for r ¤ 1. According to (A.5), there is an

exact sequence
0 ! N ! G0 ! G1 ! 0
with G0 and G1 smooth. As Rr f G D 0 for r > 0 if G is a smooth group

scheme (see Milne 1980, III 3.9), it is clear that Rr f N D 0 for r > 1 (this part
of the argument works for a finite flat group scheme over any scheme). The sheaf
f N is the sheaf defined by N on Xet , which is zero because N is infinitesimal
and all connected schemes étale over X are integral.
As far as the sequence is concerned, we confine ourselves to defining the
maps; for the proof of the exactness, see Artin and Milne 1976, 2. First we need
a lemma.
L EMMA 5.7 The map H 1 .X; N / ! H 1 .X 0 ; N / is zero.
P ROOF. Since N has height one, FN=X W N ! N .p/ factors through e .p/ .X /
where eW X ! N is the zero section. Therefore, Fabs W N ! N factors through
e.X /, which means that the image of a 7! ap W ON ! ON is contained in OX
ON . By descent theory, this last statement holds for any principal homogeneous
space P of N over X W there is a map W OP ! OX whose composite with the
id
inclusion OX ,!OP is the p th power map. The composite OP ! OX ! OX 0
is an OX -morphism, and therefore defines an X -morphism X 0 ! P . This shows
that P becomes trivial over X 0 . Since all elements of H 1 .X; N / are represented
by principal homogeneous spaces, this proves the lemma. 2
Let P be a principal homogeneous space for N over X . Then the lemma

shows that there is a trivialization ' 0 W X 0 ! P , and the fact that N is purely
infinitesimal implies that ' 0 is unique. It gives rise to two maps X 0 X X 0 D
X 00 ! P , namely, ' 0 ıp1 and ' 0 ıp2 . Their difference is an element ˛ 00 of N.X 00 /
that is zero if and only if ' 0 is arises from N.X /. Now if N denotes the nilradical
of OX 00 , then OX 00 =N OX 0 and N =N 2 ˝X 1 00 0
0 =S . Since ˛ is trivial on X ,
the restriction of ˛ 00 to Spec.OX 00 =N 2 / defines a map !N ! ˝X 1

0 =S , whose
1
image can be shown to lie in ˝X 0 =S;cl . This map can be identified with an element
of n ˝ ˝X 1
0 =S .
The same construction works for any U étale over X ; for such a U , we get
a map H 1 .Ufl ; N / ! n ˝OU 0 ˝U 1 1
0 =S;cl . As R f N is the sheaf associated with
U 7! H 1 .Ufl ; N /, this defines a map
R1 f N ! n ˝OU 0 ˝U
1
0 =S;cl ;
which we take to be the first map in the sequence.
L EMMA 5.8 There exists a unique map C W ˝X=S;cl

1 ! ˝X=S
1 with the proper-
ties:
(i) C.f p !/ D f C.!/, f 2 OX , ! 2 ˝X=S;cl

1 I
(ii) C.!/ D 0 if and only if ! is exact;
(iii) C.f p1 df / D df .
P ROOF. Every closed differential 1-form is locally a sum of exact differentials

and differentials of the form f p1 df , and so (ii) and (iii) completely describe C .
For the proof that C exists, see Milne 1976, 1.1, and Katz 1970, 7.2. 2
Note that (ii) says that C is p 1 -linear. According to our conventions, ˝X

1
0 =S D
1 0
˝X=S as sheaves of abelian groups on X D X , but f 2 OX acts as f on p
0 =S . Therefore, when regarded as a map ˝X 0 =S;cl ! ˝X=S , C is OX -linear.

˝X1 1 1
Define 0 W n ˝ ˝X 1
0 =S;cl ! n ˝ ˝X=S to be 1 ˝ C .
1
Recall (Demazure and Gabriel 1970, II, 7), that n has the structure of a p-
Lie algebra, that is, there is a map n 7! n.p/ W n ! n such that .f x/.p/ D
0
f p x .p/. Also we have a canonical inclusion ˝X 0 =S;cl ! ˝X=S (because X D
1 1
1W n ˝ ˝X 0 =S;cl ! n ˝ ˝X=S to be n ˝ ! 7! n
X /. Define 1 1 .p/ ˝ !, and to
be 0 1; thus .n ˝ !/ D n ˝ C ! n.p/ ˝ !.
E XAMPLE 5.9 Let N D N0;b L , b 2 .X; L˝1p / (in the Oort-Tate classifica-
tion (0.9)). Then we can describe explicitly. It is the map
1 1
L ˝ ˝X 0 =S;cl ! L ˝ ˝X=S , x ˝ ! 7! x ˝ C ! .b ˝ x/ ˝ !:
If L D OX , then the OX -structure on ˝X 1

0 =S;cl is irrelevant, and so we can
1
identify it with ˝X=S;cl . The map then becomes
1 1
.! 7! C ! b!/W ˝X=S;cl ! ˝X=S :
p
For example, if b D 1, then N D p and R1 f p D OX =OX ; the sequence is
1 p dlog 1 C 1 df
0 ! OX =OX ! ˝X=S;cl ! ˝X=S ! 0; dlog.f / D :
f
p
If b D 0, then N D ˛p and R1 f ˛p D OX =OX ; the sequence is
p d1 1 C
0 ! OX =OX ! ˝X=S;cl ! ˝X=S ! 0:
The canonical pairing of the complexes

We continue with the notations of the last subsection. Set
U .N / D .n ˝ ˝X
1 1
0 =S;cl ! n ˝ ˝X=S /;
'
V .N D / D .!N ! !N
.p/
/:
The terms on the right are complexes supported in degrees zero and one.
P ROPOSITION 5.10 There is a canonical pairing of complexes
V .N D / U .N / ! U .p /:
P ROOF. In order to define a pairing of complexes, we have to define pairings

. ; /0;0 W V 0 .N D / U 0 .N / ! U 0 .p /
. ; /1;0 W V 1 .N D / U 0 .N / ! U 1 .p /
. ; /0;1 W V 0 .N D / U 1 .N / ! U 1 .p /
such that
.v; u/0;0 D .'v; u/1;0 C .v; u/0;1
for all .v; u/ 2 V 0 .N D / U 0 .N /. If we set
.˛; n ˝ ! 0 /0;0 D ˛.n/! 0
.ˇ; n ˝ ! 0 /1;0 D ˇ.n/! 0
.˛; n ˝ !/0;1 D ˛.n/!;
then it is routine matter to verify that these pairings satisfy the conditions (ibid.,
3). 2

Note that Theorems 5.1 and 5.6 give us quasi-isomorphisms RfN D !

V .N D / and RfN ! U .N /Œ1. Also, that the pairing N D N ! p
gives a pairing Rf N D ˝L Rf N ! Rf p .
P ROPOSITION 5.11 The following diagram commutes (in the derived category)
Rf N D ˝L RfN ! Rfp
? ?
? ?
y y
V .N D / ˝L U .N /Œ1 ! U .p /Œ1:
P ROOF. This is a restatement of Artin and Milne 1976, 4.6. 2
N OTES This section summarizes Artin and Milne 1976.

6 Local fields of characteristic p

Throughout this section, K will be a local field of characteristic p ¤ 0 with finite
residue field k. Let R be the ring of integers in K. A choice of a uniformizing

parameter t for R determines isomorphisms R ! kŒŒt and K ! k..t//.
We shall frequently use that any group scheme G of finite-type over K has
a composition series with quotients of the following types: a smooth connected
group scheme; a finite étale group scheme; a finite group scheme that is local with
étale Cartier dual; a finite group scheme that is local with local Cartier dual. We
shall refer to the last two group schemes as being local-étale and local-local re-
spectively. A finite local group scheme has a composition series whose quotients
are all of height one, and a finite local-local group scheme has a composition se-
ries whose quotients are all isomorphic to ˛p (Demazure and Gabriel 1970, IV,
3,5).
Čech cohomology
Fix an algebraic closure of K a of K. For any sheaf F on (Spec K/fl and finite
extension L of K, we write HL r .L=K; F / for the r th cohomology group of the
complex of abelian groups
0 ! F .L/ ! F .L ˝K L/ ! ! F .˝rK L/ ! F .˝rC1

K L/ ! (6.0.1)
In the case that L is Galois over K with Galois group G, this complex can be
identified with the complex of inhomogeneous cochains of the G-module F .L/
(see Shatz 1972, p207, or Milne 1980, III 2.6). We define HL r .K; F / to be
lim HL r .L=K; F / where L runs over the finite field extensions of K contained
! a
in K .
P ROPOSITION 6.1 For any group scheme G of finite-type over K and any r 0,
the canonical map HL r .K; G/ ! H r .K; G/ is an isomorphism.
The proof uses only that K is a field of characteristic p. The first step is to
show that a short exact sequence of group schemes leads to a long exact sequence
of Čech cohomology groups. This is an immediate consequence of the following
lemma.
L EMMA 6.2 For any short exact sequence
0 ! G 0 ! G ! G 00 ! 0
6. LOCAL FIELDS OF CHARACTERISTIC P 273
of group schemes of finite type over K and any r , the sequence
0 ! G 0 .˝rK K a / ! G.˝rK K a / ! G 00 .˝rK K a / ! 0
is exact.
P ROOF. We show that H 1 .˝rK K a ; G/.D lim H 1 .˝rK L; G// is zero. For any
!
finite extension L of K, ˝rK L is an Artin ring. It is therefore a finite product of
Q 1 fields Li are finite extensions of L. If G is smooth,
local rings whose residue
H 1 .˝rK L; G/ D HL .Li ; G/ (see Milne 1980, III 3.11), and so obviously
lim H .˝K L; G/ D 0. It remains to treat the case of a finite group scheme N of
1 r
!
height one. Denote Spec ˝rK L by X . Then (5.7) shows that the restriction map
1 1 1 1
H 1 .X; N / ! H 1 .X .p / ; N / is zero. Since X .p / D Lp ˝K ˝K Lp ,
we again see that lim H 1 .˝rK L; G/ D 0. 2
!
We next need to know that HL r .K; G/ and H r .K; G/ are effaceable in the
category of group schemes of finite-type over K.
L EMMA 6.3 Let G be a group scheme of finite-type over K .

(a) For any c 2 HL r .K; G/, there exists an embedding G ,! G 0 of G into a
group scheme G 0 of finite-type over K such that c maps to zero in HL r .K; G 0 /.
(b) Same statement with HL r .K; G/ replaced by H r .K; G/.
P ROOF. In both cases, there exists a finite extension L of K, L K a , such that c

maps to zero in HL r .L; G/ (or H r .L; G//. Take G 0 to be ResL=K G and the map
to be the canonical inclusion G,! ResL=K G. In the case of Čech cohomology, a
simple direct calculation shows that c maps to zero in HL r .K; G 0 /, and in the case
of derived-functor cohomology, H r .K; G 0 / D H r .L; G/. 2
We now prove the proposition by induction on r. For r D 0 it is obvious, and

so assume that it holds for all r less than some r0 . For any embedding G,!G 0 ,
G 0 =G is again a group scheme of finite type over K, and we have a commutative
diagram
! HL r0 1 .K; G 0 =G/ !

HL r0 1 .K; G 0 / ! HL r0 .K; G/ !
HL r0 .K; G 0 /
? ? ? ?
?' ?' ? ?
y y y y
H r0 1 .K; G 0 / !
! H r0 1 .K; G 0 =G/ ! H r0 .K; G 0 /:
H r0 .K; G/ !
Let c be a nonzero element of HL r0 .K; G/, and choose G,!G 0 to be the em-
bedding given by (6.3a); then a diagram chase shows that the image of c in
H r0 .K; G/ is nonzero. Let c 0 2 H r0 .K; G/, and choose G,!G 0 to be the em-
bedding given by (6.3b); then a diagram chase shows that c 0 is in the image of
HL r0 .K; G/ ! H r0 .K; G/. As c and c 0 are arbitrary elements, this shows that
HL r0 .X; G/ ! H r0 .X; G/ is an isomorphism and so completes the proof.
First calculations
Note that for any finite group scheme N over K, H 0 .K; N / is finite.
P ROPOSITION 6.4 Let N be a finite group scheme over K .
(a) If N is étale-local, then H r .K; N / D 0 for r ¤ 0; 1.
(b) If N is local-étale, then H r .K; N / D 0 for r ¤ 1; 2.
(c) If N is local-local, then H r .K; N / D 0 for r ¤ 1.
P ROOF. (a) Since N is étale, H 1 .K; N / D H 1 .Gal.K s =K/; N.K s //, and be-
cause its Cartier dual is local, N must have p-power order. Therefore the assertion
follows from the fact that K has Galois p-cohomological dimension 1.
(b) We can assume that N has height one, and then (5.6) shows that Rr f N D
0 for r ¤ 1. Therefore H r .K; N / D H r1 .Ket ; R1 f N /, and R1 f N is a p-
torsion sheaf.
(c) We can assume that N D ˛p . In this case the statement follows directly
from the cohomology sequence of
F
0 ! ˛p ! Ga ! Ga ! 0: 2
The topology on the cohomology groups

The ring ˝rK L has a natural topology. If G is an affine group scheme of fi-
nite type over K, then G.˝rK L/ also has a natural topology: immerse G into
some affine space An and give G.˝rK L/ An .˝rK L/ the subspace topology.
One checks easily that the topology is independent of the immersion chosen.
Therefore, for any group scheme G of finite type over K, G.˝rK L/ has a nat-
ural topology, and the boundary maps in (6.0.1) are continuous because they are
given by polynomials. Endow Z r .L=K; G/ C r .L=K; G/.D G.˝rC1 K L//
L
with the subspace topology, and H .L=K; G/ D Z .L=K; G/=B .L=K; G/
r r r
with the quotient topology. We can then give H r .K; G/ the direct limit topology:
a map H r .K; G/ ! T is continuous if and only if it defines continuous maps on
HL r .L=K; G/ for all L.
L EMMA 6.5 Let G be a group scheme of finite type over K , and let L K a be
a finite extension of K .
(a) The group H r .X; G/ is Hausdorff, locally compact, and -compact (that
is, a countable union of compact subspaces).
(b) The maps in the cohomology sequence arising from a short exact sequence
of group schemes are continuous.
(c) The restriction maps H r .K; G/ ! H r .L; G/ are continuous.
(d) When G is finite, the inflation map InfW HL 1 .L=K; G/ ! H 1 .K; G/ has
closed image and defines a homeomorphism of HL 1 .L=K; G/ onto its im-
age.
(e) Cup-product is continuous.
P ROOF. (a) The groups C r .L=K; G/ are Hausdorff, -compact, and locally com-
pact. As Z r .L=K; G/ is a closed subspace of C r .L=K; G/, it has the same
properties. Also the image B r .L=K; G/ of C r1 .L=K; G/ is a locally compact
subgroup of a Hausdorff group, and so is closed. Hence H r .L=K; G/ is the quo-
tient of a Hausdorff, -compact, locally compact space by a closed subspace, and
it therefore inherits the same properties.
(b) Obvious.
(c) It suffices to note that, for any L0 L, the maps C r .L0 =K; G/ !
C r .L0 =L; G/ are continuous.
(d) It suffices to show that for any L0 L, the inflation map HL 1 .L=K; G/ !
HL 1 .L0 =K; G/ is closed. The map G.L ˝K L/ ! G.L0 ˝K L0 / is closed, and
therefore its restriction to Z 1 .L=K; G/ ! Z 1 .L0 =K; G/ is also closed. Since
B 1 .L0 =K; G/ is compact (it is finite), the map Z 1 .L0 =K; G/ ! HL 1 .L0 =K; G/
is closed, and the assertion follows.
(e) Obvious. 2
R EMARK 6.6 (a) Let

0 ! N ! G ! G0 ! 0
be an exact sequence, and suppose that H 1 .K; G/ D 0. Then H 0 .K; N / has the
subspace topology induced from H 0 .K; N / ,! G.K/ and H 1 .K; N / has the
quotient topology induced from G 0 .K/ H 1 .K; N /.
(b) Our definition of the topology on H r .K; G/ differs from, but is equivalent
to, that of Shatz 1972, VI.
E XAMPLE 6.7 (a) The cohomology sequence of

}
0 ! Z=pZ ! Ga ! Ga ! 0
shows that H 1 .K; Z=pZ/ D K=}K. This group is infinite, and it has the discrete
topology because R=}R is a finite open subgroup of K=}K.
(b) From the Kummer sequence we find that H 1 .K; p / D K =K p . This

group is compact because R =Rp is a compact subgroup of finite index.
(c) The group H 1 .K; ˛p / D K=K p . The subgroup R=Rp of K=K p is
compact and open, and the quotient K=RK p is an infinite discrete group.
(d) The group H 0 .K; Gm / equals K with its locally compact topology. The
group H 2 .K; Gm / equals Q=Z with the discrete topology because H 2 .K; Gm / D
S
H 2 .L=K; Gm /, and H 2 .L=K; Gm / is finite and Hausdorff.
G H 0 .K; G/ H 1 .K; G/ H 2 .K; G/

étale-local finite, discrete torsion, discrete 0
Table 6.8. local-étale 0 compact finite, discrete
local-local 0 locally compact 0
torus locally compact finite, discrete discrete
To verify the statements in the table, first note that, because the topologies on
L
H .L=K; G/ and H r .K; G/ are Hausdorff, they are discrete when the groups
r
are finite. Next note that HL r .L=K; G/ contains HL r .RL =RK ; G/ as an open
subgroup. Moreover, if G is Z=pZ, p , ˛p , or Gm , each assertion follows from
(6.7). It is not difficult now to deduce that they are true in the general case.
Duality for tori

Let M be a finitely generated torsion-free module for Gal.K s =K/. As M be-
comes a module with trivial action over some finite separable extension L of K,
it is represented by an étale group scheme locally of finite-type over K. The
method used above for group schemes of finite-type defines the discrete topology
on the groups H r .K; M /.
T HEOREM 6.9 Let T be a torus over K and let X .T / be its group of characters.
Then the cup-product pairing
H r .K; T / H 2r .K; X .T // ! H 2 .K; Gm / ' Q=Z
defines dualities between:

the compact group H 0 .K; T /^ (completion relative to the topology of open
subgroups of finite index) and the discrete group H 2 .K; X .T //I
the finite groups H 1 .K; T / and H 1 .K; X .T //I
the discrete group H 2 .K; T / and the compact group H 0 .K; X .T //^ (com-
pletion relative to the topology of subgroups of finite index).
P ROOF. As T is smooth, H r .K; T / D H r .Gal.K s =K/; T .K s //, and the group

H r .K; X .T // D H r .Gal.K s =K/; X .T //. All the groups have the discrete
topology except H 0 .K; T / D T .K/, which has the topology induced by that on
K. The theorem therefore simply restates (I 2.4). 2
Finite group schemes

We now let N be a finite group scheme over K.
T HEOREM 6.10 For any finite group scheme N over K , the cup-product pairing
H r .K; N D / H 2r .K; N / ! H 2 .K; Gm / ' Q=Z
identifies each group with the Pontryagin dual of the other.
P ROOF. After (I 2.3) we may assume that N is a p-primary group scheme.

Suppose first that N is étale. Then there is a short exact sequence of discrete
Gal.Ks =K/-modules
0 ! M1 ! M0 ! N.K s / ! 0
with M0 and M1 finitely generated and torsion-free (as abelian groups). Dually
0 ! ND ! T0 ! T1 ! 0
with T 0 and T 1 tori. Because the cohomology groups of the modules in the first
sequence are all discrete, the dual of its cohomology sequence is exact. Therefore
we get an exact commutative diagram
! H r .K; N D / ! H r .K; T 0 / ! H r .K; T 1 / !

? ? ?
? ? ?
y y y
! H r .K; N / ! H r .K; M0 / ! H r .K; M1 / ! :
For r 1, the second two vertical arrows in the diagram are isomorphisms,
'
and this shows that H r .K; N D / ! H r .K; N / for r 2. The diagram also
shows that the image of H 1 .K; N D / in H 1 .K; N / is dense. As H 1 .K; N D /
is compact, its image is closed and so equals H 1 .K; N / . If H 1 .K; N D / !
H 1 .K; N / were not injective, then there would exist an element b 2 T 1 .K/
that is in the image of T 0 .K/^ ! T 1 .K/^ , but which is not in the image of
T 0 .K/ ! T 1 .K/. I claim that the image of T 0 .K/ is closed in T 1 .K/. Let L
be a splitting field for T 0 and T 1 , and consider the diagram

0 ! Hom.X .T 0 /; RL / ! T 0 .K/ ! Hom.X .T 0 /; Z/ !
? ? ?
? ? ?
y y y
0 ! Hom.X .T 1 /; RL
/
! T 1 .K/ ! Hom.X .T 1 /; Z/ !
(Hom’s as G-modules). Let b 2 T 1 .K/. If the image of b in Hom.X .T 1 /; Z/ is

not in the image from T 0 , then b Hom.X .T /; RL / is an open neighbourhood
of b that is disjoint from the image of T 0 .K/. Therefore we may assume b 2

Hom.X .T 1 /; RL /; but Hom.X .T 0 /; R / is compact and so its complement
L
1
in Hom.X .T /; RL / is an open neighbourhood of b. This proves the claim and
completes the proof of the theorem in the case that N or its dual is étale.
Next assume that N D ˛p . Here one shows that the pairing
H 1 .K; ˛p / H 1 .K; ˛p / ! Q=Z
can be identified with the pairing
K=K p K=K p ! p 1 Z=Z Q=Z; .f; g/ 7! p 1 Trk=Fp .res.f dg//;
(see Shatz 1972, p240-243); it also follows immediately from the elementary case
of (5.11) in which N D ˛p ), and this last pairing is a duality.
The next lemma now completes the proof. 2
L EMMA 6.11 Let

0 ! N 0 ! N ! N 00 ! 0
be an exact sequence of finite group schemes over K . If the theorem is true for
N 0 and N 00 , then it is true for N .
P ROOF. Proposition I 0.22 and the discussion preceding it show that the bottom
row of the following diagram is exact, and so this follows from the five-lemma:
! H r .K; N 00D / ! H r .K; N D / ! H r .K; N 0D / !

? ? ?
?' ? ?'
y y y
! H 2r .K; N 00 / ! H 2r .K; N / ! H 2r .K; N 0 / ! :
2
R EMARK 6.12 (a) It is possible to give an alternative proof of Theorem 6.10

using the sequences in 5. After (6.11) (and by symmetry), it suffices to prove
the theorem for a group N of height one. Then the exact sequences in 5 yield
cohomology sequences
0 ! H 0 .K; N D / ! V 0 .N D / ! V 1 .N D / ! H 1 .K; N D / ! 0
0 ! H 1 .K; N / ! U 0 .N / ! U 1 .N / ! H 2 .K; N / ! 0:
Here V 0 .N D / is the K-vector space !N and U 1 .N / is the K-vector space n ˝

1
˝K=k . The pairing .˛; n ˝ !/ 7! res.˛.n/!/ identifies !N with the k-linear
dual of n ˝ ˝K=k
1 , and so the pairing
.˛; n ˝ !/ 7! p 1 Trk=Fp res.˛.n/!/
1
identifies !N with the Pontryagin dual of n˝˝K=k (see 0.7). Similarly V 1 .N D /
is the Pontryagin dual of U 0 .N /. Therefore the pairing of complexes in (5.10)
shows that the dual of the first of the above sequences can be identified with an
exact sequence
0 ! H 1 .K; N D / ! U 0 .N / ! U 1 .N / ! H 0 .K; N D / ! 0:
'
Thus there are canonical isomorphisms H 1 .K; N / ! H 1 .K; N D / and
'
H 2 .K; N / ! H 0 .K; N D / , and (5.11) shows that these are the maps given
by cup-product.
(b) It is also possible to deduce a major part of (6.10) from (I 2.1). Let N
be étale over K. Then H r .Kfl ; N / D H r .Ket ; N /, and so we have to show that
ExtrKet .N; Gm / D H r .Kfl ; N D /. Note (Milne 1980, II 3.1d) that f .N jKet / D
N , where f W .SpecK/fl ! .SpecK/et is defined by the identity map. From
the spectral sequence ExtrKet .N; Rs f Gm / H) ExtrCs
Kfl .f N; Gm / and the
vanishing of the higher direct images of Gm , we see that ExtrKet .N; Gm / D
ExtrKfl .N; Gm / for all r. But N is locally constant on .SpecK/fl , and the ex-
act sequence
n
! H r .X; Gm / ! H r .X; Gm / ! ExtrKfl .Z=nZ; Gm / !
and the divisibility of Gm on the flat site show that ExtKr .Z=nZ; G / D 0 for
fl
m
r > 0. Therefore ExtKfl .N; Gm / D 0 for r > 0, and the local-global spectral
r
sequence for Exts shows that ExtrKet .N; Gm / D H r .Kfl ; N D /.

R EMARK 6.13 Much of the above discussion continues to hold if K is the field of
fractions of an excellent Henselian discrete valuation ring with finite residue field.
b N / for all r because K and
For example, if N is étale, then H r .K; N / D H r .K;
b have the same absolute Galois group, and N.K s / D N.K
K b s /; if N is local-
b N / and H 2 .K; N / D H 2 .K;
étale, then H 1 .K; N / is dense in H 1 .K; b N /; if N
1 1
is local-local, then H .K; N / is dense in H .K; b N /. The map H .K; N D / !
r
H 2r .K; N / given by cup-product is an isomorphism for r ¤ 1, and is injective

with dense image for r D 1.
N OTES The main results in this section are taken from Shatz 1964; see also Shatz 1972.
Theorem 6.10 was the first duality theorem to be proved for the flat topology and so can
be regarded as the forerunner of the rest of the results in this chapter.
7 Local results: equicharacteristic, finite residue field

Throughout this section, R will be a complete discrete valuation ring of character-
istic p ¤ 0 with finite residue field k. As usual, we use the notations X D SpecR
and
i j
Spec k D x ,! X - u D SpecK:

Let N be a finite flat group scheme over X . As in (1.1), we find that
H 0 .X; N / D N.X / D N.K/ D H 0 .K; NK /;

H 1 .X; N /,!H 1 .K; N /;
H r .X; N / D 0; r > 1;
and
Hx2 .X; N / D H 1 .K; N /=H 1 .X; N /;

Hx3 .X; N / D H 2 .K; N /;
Hxr .X; N / D 0; r ¤ 2; 3:
In the last section we defined topologies on the groups H r .K; N /. We en-

dow H r .X; N / with its topology as a subspace of H r .K; N /, and we endow
Hxr .X; N / with its topology as a quotient of H r .K; N /. With respect to these
topologies H 0 .X; N / is discrete (and finite), and H 1 .X; N / is compact and
Hausdorff; Hx2 .X; N / and Hx3 .X; N / are both discrete (and Hx3 .X; N / is finite).
7. LOCAL: EQUICHARACTERISTIC, FINITE RESIDUE FIELD 281
T HEOREM 7.1 For any finite flat group scheme N over X , the canonical pairings
define dualities between:

the finite groups H 0 .K; N D / and Hx3 .X; N /I
the compact group H 1 .X; N D / and the discrete torsion group Hx2 .X; N /.
Before giving the proof, we list some corollaries.
C OROLLARY 7.2 For any finite flat group scheme N over X , H 1 .X; N D / is the
exact annihilator of H 1 .X; N / in the pairing
H 1 .K; N D / H 1 .K; N / ! H 2 .K; Gm / ' Q=Z
of (6.10).
P ROOF. As in the proof of (1.4), one sees easily that the corollary is equivalent
to the case r D 1 of the theorem. 2
C OROLLARY 7.3 Let N be a finite flat group scheme over X . For all r < 2p 2;
ExtrX .N; Gm / Hx3r .X; N / ! Hx3 .X; Gm / ' Q=Z
P ROOF. The proof is the same as that of (1.5). 2
Write f W Xfl ! Xet for the morphism of sites defined by the identity map.
C OROLLARY 7.4 Let N be a quasi-finite flat group scheme over X whose p -

primary component N.p/ is finite over X . Let N D be the complex of sheaves
such that (
D HomXfl .N.`/; Gm / if ` D p
N .`/ D
f RHomXet .N.`/; Gm / if ` ¤ p .
Then
P ROOF. For the prime-to-p components of the groups, the corollary follows from
(II 1.8); for the p-primary component, it follows immediately from the theorem.2
P ROOF ( OF 7.1) Assume first that N has height one. Then the first exact se-
quence in 5 leads to a cohomology sequence
.p/
0 ! H 0 .X; N D / ! H 0 .X; !N / ! H 0 .X; !N / ! H 1 .X; N D / ! 0;
and the second leads to a cohomology sequence
0 ! Hx2 .X; N / ! Hx1 .X; n ˝ ˝X

1 1 1 3
0 / ! Hx .X; n ˝ ˝X / ! Hx .X; N / ! 0:
The pairing
1 1
.˛; n ˝ w/ 7! ˛.n/wW !N n ˝ ˝X ! ˝X . OX /
realizes H 0 .X; !N / as the R-linear dual of H 0 .X; n ˝ ˝X 1

/; therefore (0.8)
0
shows that the compact group H .X; !N / is the Pontryagin dual of the discrete
group Hx1 .X; n ˝ ˝X 1 / D H 0 .X; n ˝ ˝ 1 / ˝ K=H 0 .X; n ˝ ˝ 1 /. Similarly the
X X
.p/
compact group H 0 .X; !N / is the Pontryagin dual of Hx1 .X; n ˝ ˝X
1 /, and so
0
the pairing in (5.10) gives a commutative diagram

.p/
H 0 .X; N D / !
0! H 0 .X; !N / !
H 0 .X; !N / !
H 1 .X; N D / !
0
? ?
? ?
y y
Hx3 .X; N / !
0! 1
Hx1 .X; n ˝ ˝X / ! 1
Hx1 .X; n ˝ ˝X 0/ Hx2 .X; N / !
! 0:

The diagram provides isomorphisms H 0 .X; N D / ! Hx3 .X; N / and

H 1 .X; N D / ! H 2 .X; N /, which we must show are those given by the pairing
in the theorem. For this we retreat to the derived category. In (5.11) we saw that
there is a commutative diagram:
RfN D ˝L RfN ! Rf p

? ?
? ?
y y
V .N D / ˝L U .N /Œ1 ! U .p /Œ1:
From this we get a commutative diagram
H r .X; N D / Hx3r .X; N / ! Hx3 .X; p /

# # #
H r .X; V .N D // Hx2r .X; U .N // ! Hx2 .X; U .p //;
which exactly says that the two pairs of maps agree. This completes the proof of
the theorem when N has height one.
Next we note that for r D 0 the theorem follows from (6.10), and that for
r D 1 it is equivalent to (7.2). Since this last statement is symmetric between N
and N D , we see that the theorem is also true if N is the dual of a finite group
scheme of height one. Every finite group scheme over X has a composition series
each of whose quotients is of height one or is the dual of a height one group
(this is obvious over K, and one can apply (B.1) to obtain it over X /, and so the
theorem follows from the obvious fact that it is true for any extension of groups
for which it is true. 2
R EMARK 7.5 The original proof of the Theorem 7.1 (Milne 1970/72, 1973) was
more explicit. We include a sketch. It clearly suffices to prove (7.2). Write dK
. K=K p / and dR . R=Rp / for the images of the maps d W K ! ˝K=k 1 and
d W R ! ˝R=k .
1
Assume first that N D ˛p . Then the diagram
H 1 .X; N D / H 1 .X; N /
# # (7.5.1)
H 1 .K; N D / H 1 .K; N / ! H 2 .K; Gm / ' Q=Z
R=Rp dR
# #
K=K p dK ! p 1 Z=Z Q=Z;
where the bottom pairing is .f; !/ 7! p 1 Trk=Fp res.f !/. It is obvious that the
upper groups are exact annihilators in the lower pairing.
Let N D Na;0 in the Oort-Tate classification (0.9) with a D t .p1/c . Then
the diagram (7.5.1) can be identified with
t cp R=..t cp R/ \ }R/ dlog.K / \ t cp Rdt

# #
K=}K dlog.K / ! p 1 Z=pZ Q=Z
where the lower pairing is .f; !/ 7! Trk=Fp .res.f !//. It is easy to check that
the upper groups are exact annihilators in the lower pairing.
These calculations prove the theorem whenever N D N0;0 (that is, N D ˛p /,
N D Na;0 where a D t c.p1/ , or N D N0;b where b D t c.p1/ . The general
case can be reduced to these special cases by means of the following statements
(Milne 1973, p84-85):
(i) for any finite group scheme of p-power order N over R, there exists a finite
extension K 0 of K of degree prime to p such that over R0 , N has a composition
series whose quotients have the above form;
(ii) the theorem is true over K if it is true over some finite extension K 0 of K
of degree prime to p.
R EMARK 7.6 Let N be a finite flat group scheme over X . We define
ŒH 0 .X; N /
.X; N / D
ŒH 1 .X; N /
when both groups are finite. Then the same proof as in (1.14) shows that when
NK is étale,
.X; N / D .RW d/1=n ;
where n is the order of N and d is the discriminant ideal of N over R. Note that
H 1 .X; N / is infinite ” NK is not étale ” d D 0;
and so, if we interpret 1=1 as 0, then the assertion continues to hold when NK is
not étale.
It is possible to prove a weak form of (1.19).
E XERCISE 7.7 (a) Let N and N 0 be finite flat group schemes over R; then a
homomorphism 'W NK ! NK 0 extends to a homomorphism N ! N 0 if and only
if, for all finite field extensions L of K, H 1 .'L /W H 1 .L; NL / ! H 1 .L; NL /

maps H 1 .RL ; NL / into H 1 .RL ; NL0 /.
(b)6 7 Use (a) to prove the characteristic p analogue of the main theorem of
Tate 1967b: if G and G 0 are p-divisible groups over R, then every homomor-
phism 'W GK ! GK 0 extends to a unique homomorphism G ! G 0 .
6 (In original.) The author does not pretend to be able to do part (b) of the problem; the question
is still open in general.

7 Tate’s theorem was extended to characteristic p by de Jong (Homomorphisms of Barsotti-Tate
groups and crystals in positive characteristic. Invent. Math. 134 (1998), no. 2, 301–333; erratum,
ibid. 138 (1999), no. 1, 225).
Abelian varieties and Néron models

We extend (I 3.4) and the results of 2 to characteristic p. Note that in the main
results, the coefficient groups are smooth, and so the cohomology groups can
be computed using the étale topology (or even using Galois cohomology). The
proofs however necessarily involve the flat site.
T HEOREM 7.8 Let A be an abelian variety over K , and let B be its dual. The
pairings
H r .K; B/ H 1r .K; A/ ! H 2 .K; Gm / ' Q=Z
induced by the Poincaré biextension W of .B; A/ by Gm define dualities between:
the compact group B.K/ and the discrete group H 1 .K; A/I
the discrete group H 1 .K; B/ and the compact group A.K/.
For r ¤ 0; 1, all the groups are zero.
We note first that the pairing is symmetric in the sense that the pairing defined
by W and by its transpose W t are the same (up to sign) (see C.4). Therefore it
suffices to show that the map ˛ 1 .K; A/W H 1 .K; B/ ! A.K/ is an isomorphism
and that H r .K; B/ D 0 for r 2. Consider the diagram
0 ! B.K/.n/ ! H 1 .K; Bn / ! H 1 .K; B/n ! 0

? ? ?
? ? ?
y y y
0 ! H 1 .K; A/n ! H 1 .K; An / ! A.K/.n/ ! 0
Theorem 6.10 shows that the middle vertical arrow is an isomorphism, and it
follows that ˛ 1 .K; A/n is surjective for all n. As A.K/ is a profinite group, its
dual A.K/ is torsion, and so this proves that ˛ 1 .K; A/ is surjective. To show
that it is injective, it suffices to prove that H 1 .K; B/` ! A.K/.`/ is injective
for all primes. For ` ¤ p, we saw in (I 3) that this can be done by a counting
argument. Unfortunately, for ` D p the groups involved are not finite (nor even
compact), and so we must work more directly with the cohomology groups of
finite group schemes.
We dispose of the statement that H 2 .K; B/ D 0 (note that H r .K; B/ D 0
for r > 2 because K has strict (Galois) cohomological dimension 2). Consider
the diagram
H 1 .K; B/ ! H 2 .K; Bn / ! H 2 .K; B/n ! 0

? ? ?
? ? ?
y y y
A.K/ ! An .K/ ! 0
which is just a continuation to the right of the previous diagram. We have seen that
the first vertical arrow is surjective, and the second is an isomorphism by (6.10).
A diagram chase now shows that H 2 .K; B/n D 0 for all n, and so H 2 .K; B/ is
zero.
The next lemma will allow us to replace K by a larger field.
L EMMA 7.9 If for some finite Galois extension L of K , ˛ 1 .L; AL / is injective,
then ˛ 1 .K; A/ is injective.
P ROOF. Since K is local, the Galois group G of L over K is solvable, and so we

may assume it to be cyclic. There is an exact commutative diagram
H 1 .G; B.L// !
0 ! H 1 .K; B/ !
H 1 .L; B/G !
HT0 .G; B.L// !
H 2 .K; B/
? ? ? ?
? ?surj ? ?
y y y y
HT0 .G; A.L// !
0 ! A.K/ A.L/G
! H 1 .G; A.L// !
! 0
in which the top row is part of the sequence coming from the Hochschild-Serre
spectral sequence (except that we have replaced HT2 .G; B.L// with HT0 .G; B.L///,
and the bottom row is the dual of the sequence that explicitly describes HT0 and
HT1 for a cyclic group. The second and third vertical arrows are ˛ 1 .K; A/ and
˛ 1 .L; A/, and the first and fourth are induced by ˛ 1 .K; A/ and by the dual of
˛ 1 .K; B/ respectively. From the right hand end of the diagram we see that
HT0 .G; B.L// ! H 1 .G; A.L//
is an isomorphism, and by interchanging A and B we see that
HT0 .G; A.L// ! H 1 .G; B.L//
is an isomorphism. Thus all vertical maps but the second are isomorphisms, and
the five-lemma shows that it also is an isomorphism. 2
To proceed further, we need to consider the Néron models A and B of A and

B. Let i ˚ D A=Aı , and write A for the subscheme of A corresponding to a
subgroup of ˚ .
L EMMA 7.10 (a) The map A .X / ! .x/ is surjective, and the map
H r .X; A / ! H r .x; / is an isomorphism for all r 1; therefore H r .X; A / D
0 for r 2.
(b) There is an exact sequence
˚.x/ ! .˚ = /.x/ ! H 1 .X; A / ! H 1 .K; A/:

(c) We have
(
0 if r ¤ 1; 2
Hxr .X; A / D
.˚ = /.x/ if r D 1,
0 ! .x/ ! ˚.x/ ! .˚ = /.x/ ! H 1 .X; A / ! H 1 .K; A/ ! Hx2 .X; A / ! 0:
P ROOF. The proofs of (2.1–2.3) apply also in characteristic p. 2
After we make a finite separable field extension, A (and B/ will have semi-
stable reduction and Ap and Bp will extend to finite group schemes over X . The
group ˚.k/p then has order p where is the dimension of the toroidal part
of the reduction Aı0 of Aı (equal to the dimension of the toroidal part of the re-
duction of B ı ). The extension of the Poincaré biextension to .B ı ; Aı / defines a
pairing
Hx2 .X; B ı / H 0 .X; Aı / ! Hx3 .X; Gm / ' Q=Z;
which the proposition allows us to identify with a map H 1 .K; B/ ! Aı .X / .
Clearly this map is the composite of ˛ 1 .K; A/W H 1 .K; B/ A.K/ with A.K/
Aı .X / . In order to complete the proof of the theorem, it suffices therefore to
show that the kernel of H 1 .K; B/p ! Aı .X /.p/ has order ŒA.K/.p/ =Aı .X /.p/ D
Œ˚.k/.p/ .
From the diagram
0 ! ˚ 0 .k/.p/ ! Hx2 .X; B ıp / ! Hx2 .X; B ı /p ! 0
? ? ?
? ? ?
y y y
0 ! H 0 .X; Aıp / ! Aıp .X /.p/ ! 0;
we see that it suffices to show that the kernel of Hx2 .X; Bpı / ! H 0 .X; Aıp / has
order Œ˚ 0 .k/.p/ Œ˚.k/.p/.
The map Hx2 .X; Bpı / ! H 0 .X; Aıp / is the composite of the maps
Hx2 .X; Bpı / ! Hx2 .X; Bp / ! H 0 .X; Ap / ! H 0 .X; Aıp / :

The middle map is an isomorphism (6.1) and the remaining two maps are surjec-
tive with kernels respectively Hx1 .X; ˚p0 / D H 1 .x; ˚p0 / and H 0 .x; ˚p / . The
shows that the kernel of Hx2 .X; Bpı / ! H 0 .X; Aıp / has the required order. (See
also Milne 1970/72:)
Once (7.8) is acquired, the proofs of (2.5) to (2.10) apply when the base ring
has characteristic p. We merely list the results.
T HEOREM 7.11 The canonical pairing ˚ 0 ˚ ! Q=Z is nondegenerate; that

is, Conjecture C.13 holds in this case.
C OROLLARY 7.12 Suppose that 0 and are exact annihilators under the canon-
ical pairing on ˚ 0 and ˚ . Then the map
0
B ! ExtX
1
sm
.A ; Gm /
defined by the extension of the Poincaré biextension is an isomorphism (of sheaves
on Xsm /.
T HEOREM 7.13 Assume that 0 and are exact annihilators. Then the pairing
0
H r .X; B / Hx2r .X; A / ! Hx3 .X; Gm / ' Q=Z
0
defined by the canonical biextension of .B ; A / by Gm induces an isomor-
phism
' 0
Hx2 .X; A / ! A .X /
of discrete groups for r D 0 and an isomorphism of finite groups
0 '
H 1 .X; B / ! A .X /
for r D 1. For r ¤ 0; 1, both groups are zero.
R EMARK 7.14 Assume that R is an excellent Henselian discrete valuation ring,
b D SpecR.
and let X b Then it follows from (I 3.10) that the maps Hxr .X; A / !
r b ; A / are isomorphisms for all r, and H r .X; A / ! H r .X;
Hx . X b A / is an
isomorphism for all r > 0. The map A.X / ! A.X/ b is injective and maps onto
b b
the torsion subgroup of A.X/; in fact, A.X/ is the completion of A.X / for the
topology of open subgroups of finite index.
n 0 n
Write Bfng for the complex B ! B n˚ and Afng for the complex A˚n !
0
ı ı
A . The pairings B ˝ A ! Gm Œ1 and B n˚ ˝L A˚n ! Gm Œ1 defined by
L
the Poincaré biextension induce a pairing Bfng ˝L Afng ! Gm in the derived

category of sheaves on Xsm .
T HEOREM 7.15 The map Bfng ˝L Afng ! Gm defines nondegenerate pairings
H r .X; Bfng/ Hx3r .X; Afng/ ! Hx3 .x; Gm / ' Q=Z
for all r .
T HEOREM 7.16 Assume that n is prime to p or that A has semistable reduction.
Then for all r , there is a canonical nondegenerate pairing
Hxr .Xfl ; Bn / H 3r .Xfl ; An / ! Hx3 .Xfl ; Gm / ' Q=Z:
8. GLOBAL: CURVES OVER FINITE FIELDS, FINITE SHEAVES 289
Curves over X
It is possible to prove an analogue of Theorem 2.11. Note that the methods in
Artin and Milne 1976, 5, can be used to prove a more general result.
N OTES Theorem 7.1 was proved independently by the author (Milne 1970/72, 1973) and
by Artin and Mazur (unpublished). The above proof is new. Theorem 7.8 was also proved
by the author (Milne 1970/72) (Shatz 1967 contains a proof for elliptic curves with Tate
parametrizations). The stronger forms of it are due to McCallum (1986).
8 Global results: curves over finite fields, finite sheaves

Throughout this section, X will be a complete smooth curve over a finite field k.
The function field of X is denoted by K, and p is the characteristic of k. For a
sheaf F on an open subscheme U of X , Hcr .U; F / denotes the cohomology group
with compact support as defined in (0.6b). Thus, there exist exact sequences
M
! Hcr .U; F / ! H r .U; F / ! H r .Kv ; F / !
v2XXU
M
! Hcr .V; F / ! Hcr .U; F / ! bv ; F / !
H r .O
v2U XV
with Kv and O b v the completions of K and Ov at v. With this definition, a short

exact sequence of sheaves gives rise to a long exact cohomology sequence, and
there is a pairing between Ext groups and cohomology groups with compact sup-
port (see 0.4b and 0.4e), but in general the flat cohomology groups with compact
support will not agree with the étale groups even for a sheaf arising from an étale
sheaf or a smooth group scheme (contrast 0.4d).
The duality theorem

When N is a quasi-finite flat group scheme on U , we endow H r .U; N / with the
discrete topology.
L EMMA 8.1 For any quasi-finite étale group scheme N on an open subscheme
U of X , Hcr .U; N / D Hcr .Uet ; N / all r .
P ROOF. Let KQ v be the field of fractions of Ovh . Then, as we observed in (6.13),

H r .KQ v ; N / ! H r .Kv ; N / for all r, and as H r .KQ v;et ; N / ! H r .KQ v ; N /,
the lemma follows from comparing the two sequences

L
! Hcr .Uet ; N / ! H r .Uet ; N / ! H r .KQ v;et ; N / !
v2XXU
? ? ?
? ? ?
y y y
L
! Hcr .U; N / ! H r .U; N / ! H r .Kv ; N / ! :
v2XXU
2
T HEOREM 8.2 Let N be a finite flat group scheme over an open subscheme U
of X . For all r , the canonical pairing

defines isomorphisms Hc3r .U; N / ! H r .U; N D / .
After Lemma 8.1 and Theorem II 3.1, it suffices to prove the theorem for a
group scheme killed by a power of p. We first need some lemmas.
L EMMA 8.3 Let

0 ! N 0 ! N ! N 00 ! 0
be an exact sequence of finite flat group schemes on U . If the theorem is true for
N 0 and N 00 , then it is true for N .
P ROOF. Because the groups are discrete, the Pontryagin dual of
! H r .U; N 00D / ! H r .U; N D / ! H r .U; N 0D / !
is exact. Therefore one can apply the five-lemma to the obvious diagram. 2
L EMMA 8.4 Let V be an open subscheme of U . The theorem is true for N on U

if and only if it is true for N jV on V .
P ROOF. This follows from (7.1) and the diagram

L
! Hc3r .V; N /
! Hc3r .U; N /
! bv ; N /
H 3r .O !
v2U XV
? ? ?
? ? ?
y y y
L
! H r .V; N D /
! H r .U; N D /
! b v ; N D /
Hvr .O ! :
v2U XV
2
L EMMA 8.5 The theorem is true if U D X and N or its dual have height one;
moreover, the groups involved are finite.
P ROOF. Assume first that N has height one. The first exact sequence in 5 yields
a cohomology sequence
.p/
! H r .X; N D / ! H r .X; !N / ! H r .X; !N / !
and the second a sequence
! H rC1 .X; N / ! H r .X; n ˝ ˝X
1 r 1
0 / ! H .X; n ˝ ˝X / ! :
The canonical pairing of complexes (5.10) together with the usual duality theorem
for coherent sheaves on a curve show that the finite-dimensional k-vector spaces
.p/
H r .X; !N / and H r .X; !N / are the k-linear (hence Pontryagin) duals of the
k-vector spaces H 1r .X; n ˝ ˝X1 / and H 1r .X; n ˝ ˝ 1 /, and moreover that
X
there is a commutative diagram
H 3r .X; N / !
! H 2r .X; n ˝ ˝X
1 / !
0 H 2r .X; n ˝ ˝X
1/ !

? ? ?
? ? ?
y y y
.p/
H r .X; N D / !
! H r1 .X; !N / !
H r1 .X; !N / !
:

The diagram gives an isomorphism H 3r .X; N / ! H r .X; N D / for all r,
and (5.11) shows that this is the map in the statement of the theorem.
In this case the groups H r .X; N / and H r .X; N D / are finite, and the state-
ment of the theorem is symmetric between N and N D . Therefore, the theorem is
proved also if the dual of N has height one. 2
We now prove the theorem. Let N be a finite group scheme over U . Lemma
8.4 allows us to replace U by a smaller open subset, and so we can assume that
N has a composition series all of whose quotients have height 1 or are the Cartier
duals of groups of height 1. Now Lemma 8.3 allows us to assume that N (or its
dual) has height 1. According to Proposition B.4, NK extends to a finite flat group
scheme N on X which is of height one (or has a dual of height one). After again
replacing U by a smaller open set, we can assume that N jU D N . According to
(8.5), the theorem is true for N on X , and (8.4) shows that this implies the same
result for N jU D N .
C OROLLARY 8.6 Let N be a finite flat group scheme on U . For all r < 2p 2,
the pairing
ExtrU .N; Gm / Hc3r .U; N / ! Hc3 .U; Gm / ' Q=Z
defines isomorphisms Hc3r .U; N / ! ExtrU .N; Gm / .
P ROOF. Under the hypotheses, H r .U; N D / D ExtrU .N; Gm / (see the proof of
(1.5)), and so this follows immediately from the theorem. 2
C OROLLARY 8.7 Let N be a quasi-finite flat group scheme over U whose p -

primary component N.p/ is finite over U . Let N D be the complex of sheaves
such that (
HomUfl .N.`/; Gm / `Dp
N D .`/ D

f RHomUet .N.`/; Gm / ` ¤ p:
Then the pairing

defines isomorphisms Hc3r .U; N / ! H r .U; N D / for all r .
P ROOF. For the p-primary component of N , this follows directly from the theo-
rem; for the `-primary component, ` ¤ p, Lemma 8.1 shows that it follows from
(II 1.11b). 2
P ROBLEM 8.8 The group H r .U; N D / is torsion, and so H r .U; N D / is a com-

pact topological group. The isomorphism in the theorem therefore gives Hcr .U; N /
a natural topology as a compact group. Find a direct description of this topology.
When U ¤ X , the groups H r .U; N / will usually be infinite, even when N is a
finite étale group scheme over U (for example, H 1 .A1 ; Z=pZ/ D kŒT =}kŒT ,
which is infinite). This restricts us to considering the case U D X .
L EMMA 8.9 For any finite flat group scheme N over X , the groups H r .X; N /
are finite.
P ROOF. When N or its dual have height one, we saw that the groups are finite
in (8.5). In the general case, NK will have a filtration all of whose quotients are
of height one or have duals that are of height one, and by taking the closures of
the groups in the filtration, we get a similar filtration for N (cf. B.1). The lemma
now follows by induction on the length of the filtration. 2
When N is a finite flat group scheme on X , we define
ŒH 0 .X; N /ŒH 2 .X; N /

.X; N / D :
ŒH 1 .X; N /ŒH 3 .X; N /
P ROBLEM 8.10 Find a formula for .X; N /.
Let X D X ˝k ks , and let D Gal.ks =k/. If the groups H r .X ; N / are

finite, then it follows immediately from the exact sequences
0 ! H r1 .X; N / ! H r .X; N / ! H r .X; N / ! 0
given by the Hochschild-Serre spectral sequence for X over X , that .X; N / D 1.

When N is étale, the finiteness theorem in étale cohomology (Milne 1980, VI
2.1) shows that H r .X; N / is finite, and a duality theorem (see 11) shows that
the same is true when N is the dual of an étale group. Otherwise the groups are
often infinite. For example,
H 1 .X; ˛p / D Ker.F W H 1 .X; OX / ! H 1 .X; OX //;
which is finite if and only if the curve X has an invertible Hasse-Witt matrix.
Nevertheless, .X; ˛p / D 1.
Let N D Na;0 L in the Oort-Tate classification. Then (cf. 5.4), we have an
exact sequence
'
0 ! N ! L ! L˝p ! 0
˝p /
with '.z/ D z ˝p a ˝ z. Therefore .X; N / D q .L/ .L where q is the
order of k. But the Riemann-Roch theorem shows that
.L/ D deg.L/ C 1 g
.L˝p / D p deg.L/ C 1 g;
and so
.X; N / D p .p1/ deg.L/ :
It is easy to construct N for which deg.L/ ¤ 0: take L0 to be any invertible sheaf
of degree > 0; then for some r > 0, .X; L˝r.p1/
0 / ¤ 0, and so we can take
N D Na;0 L with L D L˝r and a any element of .X; L˝.p1/ /.
0
P ROBLEM 8.11 As we mentioned above, the groups Hcr .U; N / have canonical
compact topologies. Is it possible extend the above discussion to c .U; N / by
using Haar measures?
R EMARK 8.12 We show that, for any scheme Y proper and smooth over a finite
field k of characteristic p, the groups H r .Y; p / are finite for all r and
df r
.Y; p / D ŒH r .Y; p /.1/ D 1:
From the exact sequence (special case of (5.6))

C 1
0 ! R1 f p ! ˝Y1 =k;cl ! ˝Y1 =k ! 0
we see that it suffices to show that the groups H r .Y; ˝Y1 =k;cl / and H r .Y; ˝Y1 =k /
are all finite and that .Y; ˝Y1 =k;cl / D .Y; ˝Y1 =k /. Consider the exact sequences
of sheaves on Yet ;
F
0 ! OY ! OY ! d OY ! 0
and
C
0 ! d OY ! ˝Y1 =k;cl ! ˝Y1 =k ! 0:
From the cohomology sequence of the first sequence, we find that H r .Yet ; d OY /
is finite for all r and is zero for r > dim.Y /; moreover .Y; d OY / D 1. From
the cohomology sequence of the second sequence, we find that H r .Yet ; ˝Y1 =k;cl /
is also finite for all r and zero for r > dim.Y /, and that .Y; ˝Y1 =k;cl / D
.Y; ˝Y1 =k /, which is what we had to prove.
R EMARK 8.13 It has been conjectured that for any scheme Y proper over Spec
Z, the cohomological Brauer group H 2 .Y; Gm / is finite. The last remark shows
that when the image of the structure map of Y is a single point .p/ in Spec Z,
then H r .Y; Gm /p is finite for all r.
N OTES In the very special case that U D X and N is constant without local-local fac-
tors, Theorem 8.2 can be found in Milne 1977, Thm A 2, with a similar proof. The case
U D X and a general N is implicitly contained in Artin and Milne 1975.
9 Global results: curves over finite fields, Néron models

The notations are the same as those in the last section. In particular, X is a
complete smooth curve over a finite field, U is an open subscheme of X , and
Hcr .U; F / is defined so that the sequence
M
! Hcr .U; F / ! H r .U; F / ! H r .Kv ; F / !
v…U
is exact with Kv the completion of K.

Let A be an abelian variety over K, and let A and B be the Néron models
over U of A and its dual B. If either A has semistable reduction or n is prime to
p, there are exact sequences
0
0 ! Bn ! B ! B n˚ ! 0
0 ! An ! A˚n ! Aı ! 0:
9. GLOBAL: CURVES OVER FINITE FIELDS, NÉRON MODELS 295
The Poincaré biextension of .B; A/ by Gm extends uniquely to biextensions of

0
.B; Aı / by Gm and of .B n˚ ; A˚n / by Gm . Therefore (cf. 1), we get a canonical
pairing
Bn A n ! Gm
in this case.
P ROPOSITION 9.1 Assume that n is prime to p or that A has semistable reduc-

tion at all primes of K . Then the pairing
H r .U; Bn / Hc3r .U; An / ! Hc3 .U; Gm / ' Q=Z
induces an isomorphism Hc3r .U; An / ! H r .U; B n / for all r .
P ROOF. Let V be an open subset where A (hence also B/ has good reduction.
Over V , An is a finite flat group scheme with Cartier dual Bn , and so the propo-
sition is a special case of (8.2). To pass from V to U , we use the diagram
L
! Hc3r .V; An /
! Hc3r .U; An /
! b v ; An / !
H 3r .O
v2U XV
? ? ?
? ? ?
y y y
L
! H r .V; B n /
! H r .U; B n /
! Hvr .Ohv ; B n /
! :
v2U XV

b v ; Bn /, and Corollary 7.16 shows that
Obviously Hvr .Ovh ; Bn / ! Hvr .O
b v ; An / ! Hvr .O
H 3r .O b v ; B n /
is an isomorphism. Therefore the proposition follows from the five-lemma. 2
As usual, we write
A=Aı D ˚ D ˚iv˚v (finite sum).
P ROPOSITION 9.2 There are exact sequences

M
0 ! Aı .X / ! A.K/ ! ˚v .k.v// ! H 1 .X; Aı / ! X.K; A/ ! 0
and M
0 ! X.K; A/ ! H 1 .X; A/ ! H 1 .v; ˚v /:
P ROOF. Let U be an open subscheme of X such that AjU is an abelian scheme;

in particular, AjU D Aı jU . As in (II 5.5), we have an exact sequence
Y
0 ! H 1 .U; A/ ! H 1 .K; A/ ! H 1 .Kv ; A/.
v2U
There is an exact sequence

L L
Aı .X / !
0 ! A.U / !
Hv1 .Ohv ; Aı / !
H 1 .X; Aı / !
H 1 .U; A/ !
Hv2 .Ohv ; Aı /
v2XXU
v2XXU

L L
A.K/ ˚v .k.v// H 1 .KQ v ; A/I
v2XXU v2XXU
(for the second two equalities, see (7.10)). According to (I 3.10), the field of
fractions KQ v of Ovh can be replaced by Kv in H 1 .KQ v ; A/. The kernel-cokernel
exact sequence of
Y Y
H 1 .K; A/ ! H 1 .Kv ; A/ ! H 1 .Kv ; A/
v2X v2U
Y
0 ! X.K; A/ ! H 1 .U; A/ ! H 1 .Kv ; A/;
v2XXU
and it follows from this and the six-term sequence that X.K; A/ is the image
of H 1 .X; Aı / in H 1 .U; A/. The first exact sequence can now be obtained by
truncating the six-term exact sequence, and the second sequence can be obtained
by comparing the last sequence above with
Y
0 ! H 1 .X; A/ ! H 1 .U; A/ ! Hv2 .Ovh ; A/:
v2XXU 2
C OROLLARY 9.3 For any ˚ , H 1 .X; A / is torsion and of cofinite type.
P ROOF. It suffices to prove this with D ;. Then the group equals X.K; A/,
which is obviously torsion. It remains to show that X.K; A/p is finite. There is
an elementary proof of this in Milne 1970b. It can also be proved by using (8.12)
in the case of a surface to show that X.K; A/p is finite when A is a Jacobian
variety, and then embedding an arbitrary abelian variety into a Jacobian to deduce
the general case. 2
Let B=B 0 Ddf ˚ 0 D ˚iv˚v0 . For any subgroups D ˚iv v and 0 D

˚iv v0 of ˚ and ˚ 0 , the Poincaré biextension over K extends to a biextension
over U if and only if each v annihilates each v0 . In this case we get a map
0
B ˝L A ! Gm Œ1:
T HEOREM 9.4 Suppose that v and v0 are exact annihilators at each closed
point v .
(a) The the kernels of the pairing

0
H 1 .U; B / Hc1 .U; A /tors ! Q=Z
are exactly the divisible groups.
0
(b) If X1 .K; A/ is finite, then H 0 .U; A /^ is dual to Hc2 .U; B /.
P ROOF. If A has good reduction on U , this can be proved by the same argument
as in (II 5.2) (using 8.2). This remark shows that the theorem is true for some
V U , and to pass from V to U one uses (7.13). 2
C OROLLARY 9.5 The Cassels-Tate pairing (II 5.7a)

X1 .K; B/ X1 .K; A/ ! Q=Z
annihilates only the divisible subgroups.
P ROOF. This follows from (9.4) and the diagram
L 0
˚v .k.v// ! H 1 .X; B ı / ! X1 .K; B/ ! 0
? ? ?
? ? ?
y y y
L 1
H .v; ˚v / ! H 1 .X; A/ ! X1 .K; A/ ! 0
because (7.11) shows that the first vertical map is an isomorphism. 2
Application to the conjecture (B-S/D) for Jacobians

Recall that the index of a curve C over a field F is the greatest common divisor
of the degrees of the fields F 0 over F such that C has a rational point in F 0 .
Equivalently, it is the least positive degree of a divisor on C .
In this subsection, we let Y be a regular connected surface over k, and we let
W Y ! X be a proper morphism such that
(i) the generic fibre is a smooth geometrically connected curve over KI
(ii) for all v 2 X , the curve YKv has index one.
We write A for the Jacobian of the generic fibre of .
P ROPOSITION 9.6 (a) The orders of the Brauer group of Y and the Tate-Shafarevich
group of A are related by
ı 2 ŒBr.Y / D ŒX.K; A/
where ı is the index of YK .

(b) The conjecture of Artin and Tate (Tate 1965/66, Conjecture C) holds for
Y if and only if the conjecture of Birch and Swinnerton-Dyer (I 7, B-S/D) holds
for A.
P ROOF. (a) Once (9.5) is acquired, the proof in Milne 1981 applies.
(b) It is proved in Gordon 1979 that (a) implies (b). 2
C OROLLARY 9.7 Let A be a Jacobian variety over K arising as above. The

following statements are equivalent:
(i) for some prime ` (` D p is allowed), the `-primary component of X.K; A/
is finite;
(ii) the L-series L.s; A/ of A has a zero at s D 1 of order equal to the rank of
A.K/;
(iii) the Tate-Shafarevich group X.K; A/ is finite, and the conjecture of Birch
and Swinnerton-Dyer is true for A.
P ROOF. After part (b) of the theorem, the equivalence of the three statements (i),
(ii), and (iii) for A follows from the equivalence of the corresponding statements
for Y (Milne 1975).8 2
The behaviour of conjecture (B-S/D) with respect to p-isogenies.

We partially extend Theorem I 7.3 to the case of p-isogenies.
T HEOREM 9.8 Let f W A ! B be an isogeny of abelian varieties over K , and

let N be the kernel of its extension f W A ! B to the Néron models of A and B
over X . Assume that either the degree of f is prime to p or that A and B have
semistable reduction at all points of X and H r .X ˝ks ; N / is finite for all r . Then
the conjecture of Birch and Swinnerton-Dyer is true for A if and only if it is true
for B .
8 In Milne 1975, it is assumed that p is odd, but this condition is used only in the proof of
Theorem 2.1 of the paper. The proof of that theorem uses my flat duality theorem for a surface,
which in turn uses Bloch’s paper (listed as a preprint), which assumes p odd. Illusie (Ann. Sci.
École Norm. Sup. (4) 12 (1979), no. 4, 501–661) does not require that p be odd, so if you replace
the reference to Bloch by a reference to Illusie you can drop the condition from my duality paper
(Ann. Sci. Ecole Norm. Sup. 9 (1976), 171-202), and hence from my 1975 paper.
P ROOF. After (I 7.3), we may assume that the degree of f is a power of p. The
initial calculations in (I 7) show that in order to prove the equivalence, one must
show that
!
Y v .A; !A /
z.f .K// D z.f t .K// z.X1 .f //:
v .B; !B /
v2X
(Note that we cannot replace the local terms with z.f .Kv // because the cokernel
of f .Kv / need not be finite.) Let N ı D Ker.f ı W Aı ! B ı /. From the exact
sequence
0 ! N ı ! Aı ! B ı ! 0
we get an exact sequence
0 ! N ı.X / ! Aı .X / ! B ı .X / ! H 1 .X; N ı/ ! H 1 .X; Aı / ! H 1 .X; B ı/

! H 2 .X; N ı/ ! H 2 .X; Aı / ! H 2 .X; B ı / ! H 3 .X; N ı / ! 0:
The sequence shows that
z.H 0 .f ı // D z.H 1 .f ı // z.H 2 .f ı //1 .X; N ı /:
But .X; N 0 / D 1 (because we have assumed that the groups H r .X ˝ks ; N / are
finite; cf. the discussion following (8.10)), and (9.3) shows that z.H 2 .f ı //1 D
z.H 0 .f t //. Therefore it remains to show that
!
z.f .K// z.X.f // Y v .A; !A /
D :
z.H 0 .f ı // z.H 1 .f ı // v .B; !B /
v2X
From (9.2), we get a diagram

L
Aı .X / !
0 ! A.K/ !
H 1 .X; Aı / !
˚v .k.v// ! X.K; A/ !
0
? ? ? ? ?
? ? ? ? ?
y y y y y
L
B ı .X / !
0 ! B.K/ !
˚v0 .k.v// !
H 1 .X; B ı / !
X.K; B/ !
0
which shows that
z.f .K// z.X.f // Y Œ˚v .k.v//
D :
z.H 0 .f ı // z.H 1 .f ı // v Œ˚v0 .k.v//
It remains to show that

v .A; !A / Œ˚v .k.v//
D 0 ;
v .B; !B / Œ˚v .k.v//
but this follows from the formula
v .A; !v / D Œ˚v .k.v//=Lv .1; A/;
!v a Néron differential on A ˝ Kv , for which we have no reference to offer the

reader. 2
R EMARK 9.9 It was first pointed out in Milne 1970, p296, that, because the
group H 1 .X ˝ ks ; N / may be infinite, X1 .K; A/p may be infinite when K
is a function field with algebraically closed field of constants. This phenomenon
has been studied in the papers Vvedens’kii 1979a, 1979b, 1980/81, which give
criteria for the finiteness Xp (and hence of the groups H r .X ˝ ks ; N / in the
above theorem).
P ROBLEM 9.10 Prove the above result for every isogeny f W A ! B.
There seems to be some hope that the method used in (II 5) may be effective
in the general case; the groups are no longer finite, but they are compact.
Duality for surfaces

It is possible to prove a similar result to (3.8) (see also Artin and Milne 1976).
10 Local results: equicharacteristic, perfect residue field

Throughout this section, X D SpecR where R is a complete discrete valuation
ring with algebraically closed residue field k. We let Xi D SpecRi , where Ri D
R=miC1 .

In the equicharacteristic case, the Greenberg construction becomes a special case
of Weil restriction of scalars: for each i 0, the k-algebra structure on Ri
defines a map ˛i W Xi ! Spec k, and for any group scheme G over X , Gi .G/ D
ResXi =k G. We write G.G/ for the pro-algebraic group .Gi .G//i
0 on X .
L EMMA 10.1 (a) The functor G 7! G.G/ from smooth group schemes on X to
pro-algebraic groups on Spec.k/ is exact.
(b) When G is smooth and has connected fibres, Gi .G/ is smooth and con-
nected for all i .
(c) When N is a finite flat group scheme of height one, Gi .N / is connected
for all i .
10. LOCAL: EQUICHARACTERISTIC, PERFECT RESIDUE FIELD 301
P ROOF. See Bester 1978, 1.1, 1.2. 2
Let 0 be the functor sending a pro-algebraic group scheme over k to its

maximal étale quotient, and let r be the rth left derived functor of 0 . Write
r .G/ for r .G.G//. When N is a finite flat group scheme X , we choose a
resolution of N by smooth connected formal groups
0 ! Ni ! Gi ! Hi ! 0; i 0;
and define Fi .N / D Coker.1 .Gi / ! 1 .Hi //. It is an pro-étale group scheme

on X , and we let F.N / be the pro-étale group scheme .Fi .N //i
0.
Write ˛i for the morphism Xi ! Speck, and Rr b ˛ for the functor F 7!
lim Rr ˛i; .F jXi /. If G is a group scheme of finite-type over X , then b ˛ G D

G.G/, and if N is a finite flat group scheme over X , then R b
1 ˛ N is representable
by a pro-algebraic group scheme over k.
L EMMA 10.2 Let N be a finite flat group scheme over X .

(a) The group scheme F.N / is independent of the choice of the resolution.
(b) A short exact sequence
0 ! N 0 ! N ! N 00 ! 0
of finite group schemes defines an exact sequence
0 ! 1 .N 0/ ! 1 .N / ! 1 .N 00 / ! F.N 0 / ! F.N / ! F.N 00 / ! 0:
(c) There is an exact sequence
0 ! 0 .N / ! F.N / ! 1 .R1b
˛ N / ! 0
of pro-sheaves on X .
P ROOF. See Bester 1978, 3.2, 3.7, 3.9. 2
Write 1 for the direct system of finite group schemes .p n /.

'
L EMMA 10.3 There is a canonical isomorphism Hx2 .X; F.1 // ! Qp =Zp .
P ROOF. From the Kummer sequence
pn
0 ! p n ! Gm ! Gm ! 0

pn
0 ! 1 .Gm;X / ! 1 .Gm;X / ! F.p n / ! 0:
This yields a cohomology sequence
0 ! HZ2 .X; 1 .Gm;X // ! HZ2 .X; 1 .Gm;X // ! HZ2 .F.p n // ! 0:
But the higher cohomology groups of the universal covering group of G.Gm;X /
are zero, and so
Hx2 .X; 1 .Gm;X // D Hx1 .X; Gm;X / D K =R ' ZI
thus HZ2 .F.p n // ' Z=p n Z. 2
Assume that N is killed by p n . Then the pairing
N D N ! p n
induces a pairing
N D F.N / ! F.p n /;
and hence a pairing
Hx2 .X; N D / F.N / ! Hx2 .X; F.p n // ' Z=p n Z:
T HEOREM 10.4 The above pairing defines an isomorphism
Hx2 .X; N D / ! Homk .F.N /; Z=p n Z/:
P ROOF. If N (or its dual) has height one, this can be proved using the exact
sequence in 5. The general case follows by induction on the length of NK . See
Bester 1978, 2.6. 2
As in 4, we can endow H 1 .X; N / and H 1 .K; N / with the structures of per-
fect pro-algebraic group schemes over k. We write H1 .X; N / and H1 .K; N / for
these group schemes. Note that H1 .X; N / is the perfect group scheme associated
with R1b ˛ N . For any finite group scheme N over X , the map H1 .X; N / ,!
H1 .K; N / is a closed immersion, and we write H2x .X; N / for the quotient group.
T HEOREM 10.5 For any finite flat p -primary group scheme over X , there is a
canonical isomorphism
'
H1 .X; N / ! .H2x .X; N D /ı /t :
10. LOCAL: EQUICHARACTERISTIC, PERFECT RESIDUE FIELD 303
P ROOF. This follows from the commutative diagram (see 10.2c)
0 !
Homk .1 .H 1 .X; N //; Q=Z/ !
Homk .F.N /; Q=Z/ !
Homk .0 .N /; Q=Z/ !
0
? ? ?
? ? ?
y y y
0 !
H2x .X; N D /ı !
H2x .X; N D / !
0 .H2x .X; N D // !
0;
and the isomorphism

'
Homk .1 .H1 .X; N //; Qp =Zp // ! Ext1k .H1 .X; N /; Qp =Zp /: 2
R EMARK 10.6 The above proof shows that the dual of the continuous part
H2x .X; N /ı of H2x .X; N / is H1 .X; N / and the dual of its finite part 0 .H2x .X; N //
is the finite part 0 .N / of H0 .X; N /. Write Hx .X; N / and H.X; N / for the
canonical objects in the derived category such that H r .Hx .X; N // D Hrx .X; N /
and H r .H.X; N // D Hr .X; N /. Then the correct way to state the above results
is that there is a canonical isomorphism
'
Hx .X; N D / ! H.X; N /t Œ2
where the t denotes Breen-Serre dual (0.14).
Abelian varieties
Let A be an abelian variety over K, and let A be its Néron model over X . We
write r .A/ for r .A/ and G.A/ for G.A/.
C ONJECTURE 10.7 There is a canonical isomorphism
H 1 .K; At / ! Ext1k .G.A/; Q=Z/:

'
In particular, if Ak is connected, then H 1 .K; At / ! Homk .1 .A/; Q=Z/.
The second part of the statement follows from the first, as in (4.16). For the
components of the groups prime to p, the conjecture is proved in Ogg 1962 and
Shafarevich 1962.
n
In the case that A has good reduction, F.Ap n / D 1 .A/.p / and
Hx2 .X; At /p n Hx2 .X; Atp n / and so the conjecture can be obtained by pass-
ing to the limit in (10.4). See Bester 1978, 7.1.
One can also show by a similar argument to that in (7.9) that it suffices to
prove the result after passing to a finite separable extension of K.
Finally, one can show that the result is true if A is an elliptic curve with a Tate
parametrization (cf. Shatz 1967). In this case there is an exact sequence
n7!q n
0 ! Z ! L ! A.L/ ! 0
for all fields L finite over K. Therefore

H 1 .K; A/ ! H 2 .K; Z/ ! Homcts .Gal.K s =K/; Q=Z/:
On the other hand, A.K/ D R .Z= ord.q//, and so

Ext1k .G.A/; Qp =Zp / ! Ext1k .G.Gm;R /; Qp =Zp /:
As Ext1k .G.Gm;R /; Qp =Zp / D Homk .1 .Gm;R /; Qp =Zp / (see Serre 1960,
5.4), the duality in this case follows from the class field theory of Serre 1961.
It is to be hoped that the general case can be proved by the methods of 7.
R EMARK 10.8 The discussion in this section holds with only minor changes
when the residue field k is an arbitrary perfect field.
N OTES This section is based on Bester 1978. Some partial results in the same direction
were obtained earlier by Vvedens’kii (1973, 1976, and earlier papers).9
11 Global results: curves over perfect fields

Throughout this section, S D Spec k with k a perfect field of characteristic p ¤
0, and W X ! S is a complete smooth curve over S. Again we define Hcr .U; F /
so that the sequence
L
! Hcr .U; F / ! H r .U; F / ! v…U H
r
.Kv ; F / !
is exact with Kv the completion of K at v.

Let N be a finite flat group scheme over U X , and write Rr N and
R Š N for the sheaves on the perfect site Spf associated with S 0 7! H r .US 0 ; N /
r
and S 0 7! Hcr .US 0 ; N /.
T HEOREM 11.1 (a) The sheaves Rr N and Rr Š N are representable by per-

fect group schemes on S .
9 Most of the open questions in this section are answered in: Bertapelle, Alessandra, Local flat
duality of abelian varieties. Manuscripta Math. 111 (2003), no. 2, 141–161.

11. GLOBAL RESULTS: CURVES OVER PERFECT FIELDS 305
(b) The canonical pairing
R N D RŠ N ! RŠ Gm ' Q=ZŒ2
induces an isomorphism
R N D ! RHomS .RŠ N; Q=ZŒ2/:
We begin the proof with the case that X D U .
L EMMA 11.2 The theorem is true if U D X and N or its dual have height one.
P ROOF. Assume first that N has height one. The first exact sequence in 5 yields
an exact sequence
.p/
! Rr N D ! Rr !N ! Rr !N !
and the second a sequence
! RrC1 N ! Rr .n ˝ ˝X
1 r 1
0 / ! R .n ˝ ˝X / ! :
Since two out of three terms in these sequences are vector groups, it is clear that
Rr N and Rr N D are represented by perfect algebraic groups. The usual
duality theorem for coherent sheaves on a curve show that the k-vector spaces
.p/
H r .X; !N / and H r .X; !N / are the k-linear duals (hence Breen-Serre duals) of
the k-vector spaces H 1r .X; n˝˝X 1 / and H 1r .X; n˝˝ 1 /. The pairing (5.10)
X0

induces an isomorphism R V .N D / ! R U .N /t Œ1. Since R N D !

R V .N D / and R N ! R U .N /Œ1, this (together with (5.11)) shows

that R N D ! .R N /t Œ2, as required. (For more details, see Artin and
Milne 1976.) 2
L EMMA 11.3 Let

0 ! N 0 ! N ! N 00 ! 0
be an exact sequence of finite flat group schemes on U . If the theorem is true for
N 0 and N 00, then it is true for N .
P ROOF. This is obvious from (0.14). 2
L EMMA 11.4 Let V be an open subscheme of U . The theorem is true for N on

U if and only if it is true for N jV on V .
P ROOF. This follows from the distinguished triangles

M M
b v ; N D / ! R .N D / ! R .N jV /D !
Hv .O b v ; N D /Œ1
Hv .O
v2U XV v2U XV
M
RŠ .N jV / ! RŠ N ! b v ; N / ! RŠ .N jV /Œ1;
H.O
v2U XV
and (10.6). 2
We now prove the theorem. Let N be a finite group scheme over U . After
replacing U by a smaller open subset we can assume that N has a composition
series all of whose quotients have height 1 or are the Cartier duals of groups of
height 1. Now Lemma 11.3 shows that we can assume that N (or its dual) has
height 1. According to Appendix B, NK extends to a finite flat group scheme N
on X which is of height one (or has a dual of height one). After again replacing
U by a smaller open set, we can assume that N jU D N . According to (11.2), the
theorem is true for N on X , and (11.4) shows that this implies the same result for
N jU D N .
Néron models
We now assume the ground field k to be algebraically closed. Let A be an abelian
variety over K, and let A be its Néron model over X .
L EMMA 11.5 The restriction map H 1 .X; A/ ! H 1 .K; A/ identifies H 1 .X; A/
with X1 .K; A/.
P ROOF. The argument in the proof of Proposition (9.2) shows again that there is
an exact sequence
M
0 ! X1 .K; A/ ! H 1 .X; A/ ! H 1 .v; ˚v /;
but in the present case, the final term is zero. 2
In the proof of the next theorem, we shall use without proof that R2 A
has no connected part. The argument that the tangent space to R2 A should
equal R2 .tangent sheaf to A/, which is zero because X is a curve, makes this
plausible. This assumption is not needed if A has good reduction everywhere.
T HEOREM 11.6 There is an exact sequence
M
0 ! H 1 .X; A/ ! H 1 .K; A/ ! H 1 .Kv ; A/ ! .Tp At .K//
all v
APPENDIX A: EMBEDDING FINITE GROUP SCHEMES 307
with At the dual abelian variety to A; in particular, if A L

has no constant part
(that is, the K=k -trace of A is zero), then H .K; A/ !
1 1
all v H .Kv ; A/ is
surjective.
P ROOF. Let U be an open subscheme of X . The cohomology sequence of the

pair X U
Q 2
0 ! H 1 .X; A/ ! H 1 .U; A/ ! Hv .Ov ; A/ ! H 2 .X; A/ !
v…U
can be rewritten as
Q
0 ! X.K; A/ ! H 1 .U; A/ ! H 1 .Kv ; A/ ! H 2 .X; A/ ! :
v…U
Because the residue fields at closed points are algebraically closed, for any open

V X , Hc2 .V; A/ ! H 2 .X; A/. Choose a V such that AjV is an abelian
scheme. There is an exact sequence
0 ! Hc1 .V; A/ ˝ Qp =Zp ! Hc2 .V; A.p// ! Hc2 .V; A/.p/ ! 0:
Interprete Hcr .V; / as Rr .jV /Š . Then Theorem 11.1 shows that 0 .Hc2 .U; A.p///
is dual to H 0 .U; Tp At / D Tp .At .K// (and Hc2 .U; A.p//ı is dual to H 1 .U; Tp A/ı /.
From our assumption, the map Hc2 .V; A.p// ! H 2 .X; A/.p/ factors through
0 .Hc2 .U; A.p///, and so we can replace H 2 .X; A/.p/ in the sequence with
Tp .At .K// . Now pass to the direct limit over smaller open sets U .
If A has no constant part, then At .K/ is finitely generated by the generalized
Mordell-Weil theorem Lang 1983, Chapter 6, and so Tp .At .K// D 0. 2
R EMARK 11.7 The last theorem is useful in the classification of elliptic surfaces
with given generic fibre. See Cossec and Dolgachev 1986, Chapter 5.
P ROBLEM 11.8 Extend as many as possible of the results in Raynaud 1964/5, II,
to the p-part.
N OTES In the case U D X , Theorem 11.1 is in Artin and Milne 1976.
Appendix A: Embedding finite group schemes

An embedding of one group scheme into a second is a map that is both a homo-
morphism and a closed immersion.
T HEOREM A.1 Let R be a local Noetherian ring with perfect residue field k ,
and let N be a finite flat group scheme over SpecR. Write m for the maximal
ideal of R, Ri for R=miC1 , and Ni for N ˝R Ri . Then there exists a family
of embeddings 'i W Ni ,! Ai such that Ai is an abelian scheme over Ri and
'iC1 ˝ Ri D 'i for all i . Consequently there is an embedding of the formal
completion N b.
b of N into a formal abelian scheme A over Spf.R/
P ROOF. To deduce the last sentence from the preceding statement, note that the
family .Ai / defines a formal scheme A over Spf.R/ b and that the 'i define an
b
embedding of N into A (see Grothendieck and Dieudonné 1971, 10.6).
We first prove the theorem in the case that R D k is an algebraically closed
field of characteristic exponent p. The only simple finite group schemes over k
are Z=`Z .` ¤ p/, Z=pZ, p , and ˛p . The first three of these can be embedded
in any nonsupersingular elliptic curve over k, and the last can be embedded in
any supersingular elliptic curve. We proceed by induction on the order of N .
Consider an exact sequence
0 ! N 0 ! N ! N 00 ! 0
in which N 0 and N 00 can be embedded into abelian varieties A0 and A00 . Let e
be the class of this extension in Ext1k .N 00 ; N 0 /, and let e 0 be the image of e in
Ext1k .N 00 ; A0 /. As Ext2k .A00 =N 00 ; A0 / D 0 (see Milne 1970a, Thm 2), e lifts
to an element eQ of Ext1k .A00 ; A0 /, and N embeds into the middle term of any
representative of e: Q
0 ! N 0 ! N ! N 00 ! 0 .D e/

? ?
? ?
y y
0 ! A0 ! X ! N 00 ! 0 .D e 0 /
? ?
? ?
y y
0 ! A0 ! A ! A00 ! 0 Q
.D e/:
We next consider the case that R D k is a perfect field. The first step implies
that N embeds into an abelian variety over a finite extension k 0 of k, Nk0 ,! A.
Now we can form the restriction of scalars (Demazure and Gabriel 1970, I, 1,
6.6), of this map and obtain an embedding N ,! Resk0 =k Nk0 ,! Resk0 =k A. The
fact that k 0 =k is separable implies that Resk0 =k A is again an abelian variety,
because
Œk 0 Wk
.Resk0 =k A/ ˝k k s D Resk0 ˝k ks =ks A D Aks :
To complete the proof, we prove the following statement: let R be a local

Artin ring with perfect residue field k D R=m; let I be an ideal in R such that
mI D 0, and let R D R=I ; let N be a finite flat group scheme over R, and let
'W N ˝R R ,! A be an embedding of N ˝R R into an abelian scheme over R;
then ' lifts to a similar embedding ' over R. 2
L EMMA A.2 Let X be a smooth scheme over R, and let L.X / be the set of
isomorphism classes of pairs .Y; / where Y is a smooth scheme over R and is

an isomorphism Y ˝R ! X . Let T X0 be the tangent sheaf on X0 Ddf X ˝R k .
(a) The obstruction to lifting X to R is an element ˛ 2 H 2 .X0 ; T X0 / ˝k I .
(b) When nonempty, L.X / is a principal homogeneous space for H 1 .X0 ; T X0 /˝k
I.
(c) If X is an abelian scheme over R, then ˛ D 0.
P ROOF. See Grothendieck 1971, III 6.3, and Oort 1971. 2
The lemma shows that there is an .A; / 2 L.A/. As A is smooth, the

zero section of A over R lifts to a section of A over R. Now the rigidity of
abelian schemes (Mumford 1965, 6.15) implies that the group structure on A lifts
to a group structure on A, and it follows that A is an abelian scheme. (See also
Messing 1972, IV 2.8.1.) It remains to show that ' lifts to an embedding ' of N
into A (after possibly changing the choice of A/.
L EMMA A.3 There is an exact sequence
HomR .N; A/ ! HomR .N ; A/ ! Ext1k .N0 ; T0 .A0 / ˝k I /
where T0 .A0 / is the tangent space at zero to A0 and we have used the same
notation for the vector space T0 .A0 / ˝k I and the vector group it defines.
P ROOF. In disagreement with the rest of the book, we shall write YFl for the big
flat site on Y (category of all schemes locally of finite-type over Y with the flat
topology) and Yfl for the small flat site (category of all schemes locally of finite-
type and flat over Y with the flat topology). There is a well known short exact
sequence
0 ! T0 .A0 / ˝k I ! A.R/ ! A.R/ ! 0:
A similar sequence exists with R replaced by any flat R-algebra, and so, if we
write i and { for the closed immersions Spec k ,! SpecR and SpecR ,! SpecR,
then there is an exact sequence
0 ! i .T0 .A0 / ˝k I / ! A ! { A ! 0
of sheaves on .SpecR/fl . This yields an exact sequence

HomR;fl .N; A/ ! HomR;fl .N; { A/ ! Ext1R;fl .N; i .T0 .A0 / ˝k I //
where the groups are computed in .SpecR/fl . Note that HomR;fl .N; { A/ D
HomR;fl .{ N; A/ and that (because N , N , A and A are all in the underlying
categories of the sites)
HomR;fl .N; A/ D HomR .N; A/; HomR ;fl .{ N; A/ D HomR .N ; A/
(the right hand groups are the groups of homomorphisms in the category of
group schemes). Let f W .SpecR/Fl ! .SpecR/fl be the morphism of sites de-
fined by the identity map. Then f is exact and preserves injectives, and so
ExtrR;Fl .f F; F 0 / D ExtrR;fl .F; f F 0 / for any sheaves F on .SpecR/fl and F 0
on .SpecR/F l . In our case f N D N (see Milne 1980, II 3.1d), and so we
can replace Ext1R;fl .N; i .T0 .A0 / ˝k I // in the above sequence with the same
group computed in the big flat site on R. Next Rr i .T0 .A0 / ˝k I / D 0 for
r > 0, because T0 .A0 / ˝k I is the sheaf defined by a coherent module and so
H r .VFl ; T0 .A0 / ˝k I / D H r .VZar ; T0 .A0 / ˝k I / D 0 for r > 0 when V is an
affine k-scheme. Therefore
ExtrR;Fl .N; i .T0 .A0 / ˝k I // D Extrk;Fl .i N; T0 .A0 / ˝k I /
for all r. Finally Ext1k;Fl .N0 ; T0 .A0 / ˝k I / D Ext1k .N0 ; T0 .A0 / ˝k I /.

We know that TA0 is the free sheaf T0 .A0 /˝k OA0 . On tensoring the isomor-

phism Ext1k .A0 ; Ga / ! H 1 .A0 ; OA0 / (Serre 1959, VII 17) with T0 .A0 / ˝k I
we get an isomorphism,

Ext1k .A0 ; T0 .A0 / ˝k I / ! H 1 .A0 ; TA0 / ˝k I
The inclusion N0 ,! A0 defines a map
Ext1k .A0 ; T0 .A0 / ˝k I / ! Ext1k .N0 ; T0 .A0 / ˝k I /;
which is surjective because Ext2k .; Ga / D 0; see Oort 1966, p.II 14-2, and (I
0.17). Consider
H 1 .A0 ; T A0 / ˝k I
?
?surj
y
HomR .N; A/ ! HomR .N ; A/ ! Ext1k .N0 ; T0 .A0 / ˝k I /:
It is clear from this diagram and Lemma A.2b that if 'W N ! A does not lift to a
map ' from N to A, then a different choice of A can be made so that ' does lift.
The lifted map ' is automatically an embedding. 2
C OROLLARY A.4 In addition to the hypotheses of the theorem, assume that R

is complete and that N has order a power of p . Then N can be embedded in a
p -divisible group scheme H over R.
P ROOF. Take the p-divisible group scheme associated with the formal abelian
scheme A. 2
We next consider the problem of resolving a finite flat group scheme by

smooth group schemes. Let N be a finite flat group scheme over a Noetherian
scheme S. Then the functor MorS .N; Gm / is representable by ResN=S Gm ,
which is a smooth affine group scheme of finite type over S. Note that N D
is (in an obvious way) a closed subgroup of MorS .N; Gm /. Write N i for
N S ::: S N , and let MorS .N 2 ; Gm /sym be the kernel of the map
MorS .N 2 ; Gm / ! MorS .N 2 ; Gm /
sending f to the function fsym , where fsym .x; y/ D f .y; x/f .x; y/1. Finally,
let Z 2 .N; Gm /sym be the kernel of the boundary map
d W MorS .N 2 ; Gm /sym ! MorS .N 3 ; Gm /;

df .x; y; z/ D f .y; z/f .xy; z/1f .x; yz/f .x; y/1:
The image of the boundary map
d W MorS .N; Gm / ! MorS .N 2 ; Gm /;

df .x; y/ D f .xy/f .x/1f .y/1 ;
is contained in Z 2 .N; Gm /sym .

T HEOREM A.5 The sequence
d
0 ! N D ! MorS .N; Gm / ! ZS2 .N; Gm /sym ! 0
is an exact sequence of affine group schemes on S . The final two terms are smooth
over S .
P ROOF. See Bégeuri 1980, 2.2.1. 2
The exact sequence in the theorem is called the canonical smooth resolution
of N D .
T HEOREM A.6 Let N be a finite flat group scheme over a Noetherian scheme
S . Locally for the Zariski topology on S , there is a projective abelian scheme A
over S and an embedding N ,! A.
P ROOF. The idea of the proof is to construct (locally) a smooth curve W X ! S

over S and a principal homogeneous space Y for N D over X . Such a Y defines
an element of R1 N D , and cup-product with this element defines a map N D
HomS .N D ; Gm / ! R1 Gm D PicX=S whose image is in the abelian scheme
Pic0X=S PicX=S . One shows that the map is a closed immersion. For the
details, see Raynaud 1979 and Berthelot, Breen, and Messing 1982, 3.1.1. 2
R EMARK A.7 It is possible to construct quotients by finite flat group schemes

(see Dieudonné 1965, p114). Therefore from (A.1), we get exact sequences
0 ! N i ! A i ! Bi ! 0
with Bi an abelian scheme, and from (A.4), we get an exact sequence
0 ! N ! H ! H0 ! 0
with H 0 a p-divisible group over R. Finally, (A.6) shows that (locally) N fits into
an exact sequence
0!N !A!B!0
with A and B projective abelian schemes.
R EMARK A.8 Let S be the spectrum of a discrete valuation ring R with field of
fractions K, and let N be a quasi-finite flat separated group scheme over S. If the
normalization NQ of N in NK is flat over R, then N is a subgroup of an abelian
scheme A (because it is an open subgroup of NQ by Zariski’s main theorem, and
NQ is a closed subgroup of an abelian scheme). Conversely, if N is a subgroup of
an abelian scheme A over S, then its normalization is flat (because NK .AK /n
for some n, and the closure of NK in An is flat, see Appendix B).
The quotient A=N is represented by an algebraic space (Artin 1969, 7.3),
but it is not an abelian scheme unless N is finite because it is not separated (the
closure of the zero section is NQ =N /.
N OTES Theorem A.1 is due to Oort (1967), and Theorem A.6 is due to Raynaud. Lemma
A.3 is an unpublished result of Tate; the above proof of it was suggested to me by Mess-
ing.
Appendix B: Extending finite group schemes

Let R be a discrete valuation ring with field of fractions K. In Raynaud 1974,
p271, it is asserted that, when K has characteristic p, every finite group scheme
APPENDIX B: EXTENDING FINITE GROUP SCHEMES 313
over K killed by a power of p extends to a finite group scheme over R (the

statement is credited to Artin and Mazur). Our first proposition provides a coun-
terexample to this assertion. Then we investigate some cases where the group
does extend. First we recall a well known lemma.
L EMMA B.1 Let R be a discrete valuation ring with field of fractions K , and let
0 ! N 0 ! N ! N 00 ! 0
be an exact sequence of finite group schemes over R. Assume that N extends to

a finite flat group scheme N over R. Then there exists a unique exact sequence
0 ! N 0 ! N ! N 00 ! 0
of finite flat group schemes over R having the original sequence as its generic
fibre.
P ROOF. The group N 0 is the closure of N 0 in N , and N 00 is the quotient of N

by N 0 . (Alternatively, N 0 is such that .N 0 ; ON 0 / is the image of .N ; ON / in
.N 0 ; ON 0 /, and N 00 is such that .N 00 ; ON 00 / D .N ; ON /\ .N 00 ; ON 00 /:/2
Now let R be a discrete valuation ring of characteristic p ¤ 0. Consider an

extension of Z=pZ by p over KW
0 ! p ! N ! Z=pZ ! 0:
Such extensions are classified by Ext1K .Z=pZ; p /, and the following diagram
shows that this group is isomorphic to K =K p :
HomK .Z; Gm / K
?
?p
y
HomK .Z; Gm / K
?
?
y

HomK .Z=pZ; Gm / ! Ext1K .Z=pZ; p / ! Ext1K .Z=pZ; Gm / ! 0
?
?
y
0 Ext1K .Z; Gm / 0
Let ˛.N / be the class of the extension in K =K p , and let a.N / be ord.˛.N //
regarded as an element of Z=pZ.
P ROPOSITION B.2 The finite group scheme N extends to a finite flat group scheme
over R if and only if a.N / D 0. Therefore, there exists a finite group scheme
over K killed by p 2 that does not extend to a finite flat group scheme over R.
P ROOF. Assume that N extends to a finite flat group scheme N over R, and let
0 ! N 0 ! N ! N =N 0 ! 0;

be the extension of the sequence given by Lemma B.1. The isomorphism Z=pZ !
.N =N 0 /K extends to a map Z=pZ ! N =N 0 over RW simply map each section
of Z=pZ over R to the closure of its image under the first map. Similarly, the

Cartier dual of the isomorphism NK 0 ! p extends to a map Z=pZ ! .NK 0 /D ,
and the dual of this map is a map N ! p whose generic fibre is the original
0
isomorphism. Now after pulling back by Z=pZ ! N =N 0 and pushing out by

N 0 ! p , we obtain an extension of Z=pZ by p over R whose generic fibre is
exactly the original extension. Thus we see that N extends over R if and only if
the original extension of Z=pZ by p extends over R. The result is now obvious
from the commutative diagram,
' ' '
Ext1K .Z=pZ; p / ! Ext1K .Z=pZ; Gm / ! H 1 .K; p / ! K =K p
x x x x
? ? ? ?
? ? ? ?
' ' '
Ext1R .Z=pZ; p / ! Ext1R .Z=pZ; Gm / ! H 1 .R; p / ! R =Rp
ord
because R =Rp is the kernel of K =K p ! Z=pZ. 2
R EMARK B.3 The same argument shows that an extension N of Z=pZ by Z=pZ
need not extend to an extension of Z=pZ by Z=pZ over R, because
Ext1K .Z=pZ; Z=pZ/ D H 1 .K; Z=pZ/ ' K=}K;
Ext1R .Z=pZ; Z=pZ/ D H 1 .R; Z=pZ/ ' R=}R:
However, N does extend to an extension of Z=pZ by some finite flat group
scheme over R. Indeed, in (7.5) we note that if N 0 Na;0 in the Oort-Tate
classification (0.9) with a D t c.p1/ , then H 1 .R; N 0 / is the image of t cp R
in K=}K, and so if c is chosen sufficiently large, the class of the extension in
Ext1K .Z=pZ; Z=pZ/ will lie in Ext1R .N 0 ; Z=pZ/.
P ROPOSITION B.4 Let X be a regular quasi-projective scheme of characteristic

p ¤ 0 over a ring R, and let K be the field of rational functions on X . Any finite
flat group scheme N of height one over K extends to a finite flat group scheme
of height one over X .
APPENDIX B: EXTENDING FINITE GROUP SCHEMES 315
P ROOF. We first prove this in the case that N has order p. Then N D N0;b for
some b 2 K. We have to show that there exists an invertible sheaf L on X , a

trivialization LK ! K, and a global section ˇ of L˝1p corresponding to b
under the trivialization. Let D be a Weil divisor such that D 0 and .b/ D,
and let L D O.D/. Then under the usual identification of O.D/K with K,
.X; L˝1p / D .X; O..1 p/D/ D fg 2 K j.g/ .p 1/Dg:
Clearly b 2 .X; L˝1p /.

We now consider the general case. Recall (Demazure and Gabriel (1970), II,
7) that a (commutative) p-Lie algebra V on a scheme Y of characteristic p ¤ 0
is a coherent sheaf of OY -modules together with a map 'W V ! V such that
'.x C y/ D '.x/ C '.y/ and '.ax/ D ap '.x/. With each locally free p-Lie
algebra V there is a canonically associated finite flat group scheme N D G.V/
of height 1. Moreover, when Y is the spectrum of a field, every finite group
scheme N is of the form G.V/ for some p-Lie algebra V. Note that to give ' is
the same as to give an OY -linear map V ! V .p/ .
Thus let .V; '/ be the p-Lie algebra associated with N over K. We have to
show that .V; '/ extends to a p-Lie algebra over X . Extend V in some trivial way
to a locally free sheaf V on X , and regard ' as a linear map V ! V .p/ . Then '
is an element of HomK .V; V .p/ / and we would like to extend it to a section of
HomOX .V; V .p/ /. This will not be possible in general, unless we first twist by an
ample invertible sheaf. Let L be such a sheaf on X , and write V.r/ for V ˝ L˝r .
Then
HomOX .V.r/; V.r/.p/ / D HomOX .V.r/; V .p/ .pr//

D HomOX .V; V .p/ /.pr p/;
and so for a sufficiently high r, HomOX .V.r/; V.r/.p/ / will be generated

Pby its
global sections (Hartshorne 1977, II 7). Therefore, we can write ' D ˛i 'i
with ˛i 2 K and 'i 2 HomOX .V.r/; V.r/.p/ /. Now choose a divisor D such
that .˛i / D for all i. Then ' is a global section of
HomOX .V.r/ ˝ OX .D/; V.r/˝OX .D/.p/ / D

HomOX .V.r/; V.r/.p/ / ˝ OX ..p 1/D/:2
C OROLLARY B.5 Let X be a regular quasi-projective scheme of characteristic

p ¤ 0 over a ring R, and let K be the field of rational functions on X . Any finite
flat group scheme N whose Cartier dual is of height one over K extends to a finite
flat group scheme over X .
P ROOF. Apply the proposition to N D , and take the Cartier dual of the resulting
finite flat group scheme. 2
R EMARK B.6 It is also possible to prove (B.5) directly from (5.5).

N OTES The counterexample (B.2) was found by the author in 1977.
Appendix C: Biextensions and Néron models

Throughout, X will be a locally Noetherian scheme endowed with either the
smooth or the flat topology.
Biextensions
Let A, B, and G be group schemes over X (commutative and of finite type as
always). A biextension of .B; A/ by G is a scheme W together with a surjective
morphism W W ! B X A endowed with the following structure:
(a) an action W BA GBA ! W of GBA on W making W into a GBA -
torsor;
(b) a B-morphism mB W W B W ! W and a section eB of W over B making
W into a commutative group scheme over B; and
(c) an A-morphism mA W W A W ! W and a section eA of W over A making
W into a commutative group scheme over A.
These structures are to satisfy the following conditions:
(i) if GB ! W is the map g 7! eB g, then

0 ! GB ! W ! AB ! 0
is an exact sequence of group schemes over BI
(ii) if GA ! W is the map g 7! eA g, then

0 ! GA ! W ! BA ! 0
is an exact sequence of group schemes over A;
(iii) the following diagram commutes
mA mA
.W A W / BB .W A W / / W B W
PPP
PPPmB
PPP
PPP
P'
n7 W
nnnnn
n
nnn
mB mB nnn mA
.W B W / AA .W B W / / W A W
APPENDIX C: BIEXTENSIONS AND NÉRON MODELS 317
(The “equality” at left is .w1 ; w2 I w3 ; w4 / $ .w1 ; w3 I w2 ; w4 /). See Grothendieck

1972, VII.
E XAMPLE C.1 Let A and B be abelian varieties of the same dimension over a
field. We call an invertible sheaf P on B A a Poincaré sheaf if its restrictions to
f0g A and B f0g are both trivial and if .B A; P/ D ˙1. It is known (Mum-
ford 1970, 13, p131), that then the map of functors b 7! .b 1/ PW B.T / !
Pic.AT / identifies B with the dual abelian variety At of A. Moreover, for any
abelian variety A over a field, there exists an essentially unique pair .B; P/ with
P a Poincaré sheaf on B A (ibid. 8, 10-12).
Let P be a Poincaré sheaf on B A. With P, we can associate a Gm -torsor
W D IsomBA .OBA ; P/ (less formally, W is the line bundle associated with
P with the zero section removed). For each point b 2 B, Pb is a line bundle on
A, and Wb has a canonical structure of a group scheme over A such that
0 ! Gm ! Wb ! A ! 0
is an exact sequence of algebraic groups (Serre 1959, VII, 3). This construction
can be carried out universally (on B), and gives a group structure to W regarded
as an A-scheme which is such that
0 ! GmB ! W ! AB ! 0
is an exact sequence of group schemes over B. By symmetry, we get a group

structure on W regarded as a group scheme over A, and these two structures
form a biextension of .B; A/ by Gm . Any biextension arising in this way from a
Poincaré sheaf will be called a Poincaré biextension.
When A, B, and G are sheaves on Xfl (or Xsm /, it is possible to modify the
above definition in an obvious fashion to obtain the notion of a biextension of
.B; A/ by G: it is a sheaf of sets W with a surjective morphism W ! B A
having the structure of a GBA -torsor and partial group structures satisfying the
conditions (i), (ii), and (iii). When A, B, and G are group schemes, we write
BiextX .B; AI G/ for the set of biextensions of .B; A/ by G, and when A, B, and
G are sheaves, we write BiextXfl .B; AI G/ (or BiextXsm .B; AI G// for the similar
set of sheaves. Clearly, there is a map
BiextX .B; AI G/ ! BiextXfl .B; AI G/
and also BiextX .B; AI G/ ! BiextXsm .B; AI G/ when G is smooth over X /.

P ROPOSITION C.2 Let A, B , and G be group schemes over X . If G is flat and

affine over X , then the map
BiextX .B; AI G/ ! BiextXfl B; AI G/
is bijective, and if G is smooth and affine over X , then
BiextX .B; AI G/ ! BiextXsm .B; AI G/
is bijective.
P ROOF. The essential point is that, in each case, torsors in the category of sheaves
are representable, and torsors in the category of schemes are locally trivial for the
respective topologies (see Milne 1980, III 4.2 and 4.3). 2
Consider a biextension (of schemes) W of .B; A/ by G. Given an X -scheme

T and a T -valued point t of B, we can pull-back the sequence in (i) and so obtain
an extension
0 ! GT ! W .t/ ! AT ! 0:
This gives us a map B.T / ! Ext1T .AT ; GT /, which can be shown to be a group
homomorphism. In this manner, a biextension of .B; A/ by G defines homomor-
phisms of sheaves B ! ExtX 1 .A; G/ and A ! Ext 1 .B; G/. A biextension
fl Xfl
of sheaves determines similar maps. This has a pleasant restatement in terms of
derived categories.
P ROPOSITION C.3 There is a canonical isomorphism

BiextXfl .B; AI G/ ! HomXfl .B ˝L A; GŒ1/
(and similarly for the smooth topology).
P ROOF. See Grothendieck 1972, VII 3.6.5. 2
Given a biextension of .B; A/ by G, we can define pairings
H r .X; B/ H s .X; A/ ! H rCsC1 .X; G/
in three different ways: directly from the map B ˝L A ! GŒ1, by using the map
B ! Ext 1 .A; G/, or by using the map A ! Ext 1 .B; G/.
P ROPOSITION C.4 The three pairings are equal (up to sign).
BiextX .B; AI G/ ! BiextXsm .B; AI G/

P ROOF. See Theorem 0.15. 2
Let W be a biextension of .B; A/ by G. For any integer n, the map B ˝L A !

GŒ1 defines in a canonical way a map Bn ˝ An ! G (Grothendieck 1972, VIII
2). Up to sign, the following diagram commutes
n
0 ! Bn ! B ! B
? ? ?
? ? ?
y y y
n
Hom.An ; G/ ! Ext 1 .A; G/ ! Ext 1 .A; G/
(and similarly with A and B interchanged). The pairing Bn An ! G defines
pairings of cohomology groups.
C OROLLARY C.5 The diagram
H r .X; B/ H s .X; A/ ! H rCsC1 .X; G/

# " k
H rC1 .X; Bn / H s .X; An / ! H rCsC1 .X; G/
commutes.
P ROOF. This is obvious from the definitions. 2
When W is a Poincaré biextension on B A, the pairing Bn An ! Gm

identifies each group with the Cartier dual of the other. For n prime to the char-
acteristic, it agrees with Weil’s en -pairing.
Néron models
From now on X is a Noetherian normal integral scheme of dimension one with
perfect residue fields. The fundamental theorem of Néron (1964) on the existence
of canonical models can be stated as follows.
T HEOREM C.6 Let gW ! X be the inclusion of the generic point of X into X .

For any abelian variety A over , g A is represented on Xsm by a smooth group
scheme A.
P ROOF. For a modern account of the proof, see Artin 1986.10 2
10 See also: Bosch, Siegfried; Lütkebohmert, Werner; Raynaud, Michel. Néron models.
Springer-Verlag, Berlin, 1990.
The group scheme A is called the Néron model of A. It is separated and of

finite type, and A D A. It is obviously uniquely determined up to a unique iso-
morphism. Its formation commutes with étale maps X 0 ! X and with Henseliza-
tion and completion. Let Aı be the open subscheme of A having connected fibres.
Then there is an exact sequence of sheaves on Xsm
0 ! Aı ! A ! ˚ ! 0
L
in which ˚ is a finite sum iv ˚v with a ˚v finite sheaf on v.
Assume now that X is the spectrum of a discrete valuation ring R, and write
K and k for the field of fractions of R and its (perfect) residue field. As usual, i
and j denote the inclusions of the closed and open points of X into X .
If the identity component Aı0 of the closed fibre A0 of A is an extension of
an abelian variety by a torus, then A is said to have semistable reduction. In this
case, the formation of Aı commutes with all finite field extensions K ! L.
T HEOREM C.7 There exists a finite separable extension L of K such that AL has
semistable reduction.
P ROOF. There are several proofs; see for example Grothendieck 1972, IX 3, or
Artin and Winters 1971. 2
Let 'W A ! B be an isogeny of abelian varieties over K. From the definition

of B, we see that ' extends uniquely to an homomorphism 'W A ! B. Write ' ı
for the restriction of ' to Aı , and let N D Ker.' 0 /; it is group scheme over X .
P ROPOSITION C.8 Let 'W A ! B be the map defined by an isogeny A ! B .

The following conditions are equivalent:
(a) 'W A ! B is flat;
(b) N is flat over X ;
(c) N is quasi-finite over X ;
(d) .' 0 / ˝R k is surjective.
When these conditions are realized, the following sequence is exact on Xfl W
'ı
0 ! N ! Aı ! B ı ! 0:
P ROOF. The same arguments as those in Grothendieck 1972, IX 2.2.1, suffice to

prove this result. 2
C OROLLARY C.9 Let A be an abelian variety over K , and let n be an integer.

The following conditions are equivalent:
(a) nW A ! A is flat;
(b) nW Aı ! Aı is surjective;
(c) Aın is quasi-finite;
(d) n is prime to the characteristic of k or A has semistable reduction.
P ROOF. It is easy to see from the structure of Aı ˝R k that condition (d) of the
corollary is equivalent to condition (d) of the proposition. The corollary therefore
follows directly from the proposition. 2
Biextensions of Néron models

From now on, we endow X with the smooth topology. Also we continue to
assume that X is the spectrum of a discrete valuation ring, and we write x for its
closed point. Let W be a Poincaré biextension on B A, and write i ˚ 0 and i ˚
for B=B ı and A=Aı respectively. According to Grothendieck 1972, VIII, there
is a canonical pairing of
Gal.k s =k/-modules
˚ 0 ˚ ! Q=Z
which represents the obstruction to extending W to a biextension of .B; A/ by
Gm . We review this theory, but first we need a lemma.
If is a submodule of ˚ , then we write A for the inverse image of i in
A. Thus A has the same generic fibre as A and Ax =Aıx D .
L EMMA C.10 For any submodule ˚ , there is a canonical isomorphism

1 1
j ExtK sm
.A; Gm / ' ExtX sm
.A ; j GmK /I
therefore B ' ExtX

1 .A ; j G
sm
mK /.
P ROOF. We first show that R1 j Gm (computed for the smooth topology) is zero.
For each Y smooth over X , R1 j Gm jYet is the sheaf (for the étale topology)
associated with the presheaf U 7! Pic.UK /. We shall show that in fact the sheaf
associated with U 7! Pic.UK / for the Zariski topology is zero.
Let y 2 Y , and let be a uniformizing parameter for R. We have to show
that Pic.OY;y Œ 1 / D 0. Let Y 0 D SpecOY;y and write i 0 and j 0 for the
inclusions Z 0 ,!Y 0 and U 0 ,!Y 0 corresponding to the maps OY;y OY;y =./
and OY;y ,!OY;y Œ 1 . If Z 0 D ;, then OY;y Œ 1 D OY;y , and the assertion
is obvious. In the contrary case, Z 0 is a prime divisor on the regular scheme Y 0
(because Y is smooth over X /, and so there is an exact sequence
0 ! Gm ! j0 Gm ! i0 Z ! 0:
The cohomology sequence of this is
! Pic.Y 0 / ! Pic.U 0 / ! H 1 .Y 0 ; i0 Z/ ! :
But Pic.Y 0 / D 0 because Y 0 is the spectrum of a local ring, and H 1 .Y 0 ; i0 Z/ D

H 1 .Z 0 ; Z/, which is zero because Z 0 is normal. Therefore Pic.U 0 / D 0.
As j A D A, there is a canonical isomorphism of functors
j HomK .A; / ' HomX .A ; j /
(see Milne 1980, II 3.22). We form the first right derived functor of each side and
evaluate it at Gm . Because HomK .A; Gm / D 0, on the left we get j ExtK 1
.A; Gm /,
and because R j Gm D 0, on the right we get ExtX .A ; j Gm /, which proves
1 1
the lemma. 2

Let B ! ExtK
1 .A; G / be the map defined by W . On applying j , we get
m

a map B ! j ExtK
1 .A; G / ! Ext 1 .A; j G
m X mK /. From the exact sequence
0 ! Gm ! j Gm ! i Z ! 0
we get an exact sequence

1 1 1
HomX .A; i Z/ ! ExtX .A; Gm / ! ExtX .A; j Gm / ! ExtX .A; i Z/:
But ExtXr .A; i Z/ D i Ext r .i A; Z/, and i A D A because A is in the under-

x k
lying category of Xsm (see Milne 1980, II 3.1d). Therefore HomX .A; i Z/ D
i Homx .Ax ; Z/ D 0 and
1
ExtX .A; i Z/ D i Extx1 .Ax ; Z/
D i Homx .Ax ; Q=Z/
D i Homx .˚; Q=Z/:
This gives us the lower row of the diagram below.
L EMMA C.11 There is a unique map ˚ 0 ! Homx .˚; Q=Z/ making
0 ! Bı ! B ! i ˚ 0 ! 0

? ? ?
? ? ?
y y y
0 ! ExtX
1
.A; Gm / ! ExtX
1
.A; j Gm / ! i Homx .˚; Q=Z/ ! 0
commute.
P ROOF. Obviously the composite of the maps

1
B ! ExtX .A; j Gm / ! i Homx .˚; Q=Z/
factors through B=B ı D i ˚ 0 . 2
To give a map of sheaves ˚ 0 ! Homx .˚; Q=Z/ is the same as to give a map
of Gal.k s =k/-modules ˚ 0 ! Hom.˚; Q=Z/, or to give an equivariant pairing
˚ 0 ˚ ! Q=Z. We shall refer to the pairing defined by the map in the lemma
as the canonical pairing.
P ROPOSITION C.12 The biextension W of .B; A/ by Gm extends to a biexten-
0
sion of .B ; A / by Gm if and only if 0 and annihilate each other in the
canonical pairing between ˚ 0 and ˚ . The extension, if it exists, is unique.
P ROOF. For a detailed proof, see Grothendieck 1972, VIII 7.1b. We merely note
that it is obvious from the following diagram (extracted from C.11)
0
0 ! Bı ! B ! i 0 ! 0
? ? ?
? ?inj ?
y y y
1 .A ; G /
0 ! ExtX m ! ExtX
1 .A ; j G /
m ! i Homx .; Q=Z/ ! 0
0
that B ! ExtX 1
.A ; j Gm / factors through ExtX
1
.A ; Gm / if and only if 0
maps to zero in Hom.; Q=Z/. 2
C ONJECTURE C.13 The canonical pairing ˚ 0 ˚ ! Q=Z is nondegenerate.

The conjecture is due to Grothendieck (ibid., IX 1.3). We shall see in the
main body of the chapter that it is a consequence of various duality theorems for
abelian varieties.11
P ROPOSITION C.14 If 0 annihilates , then there is a commutative diagram
0
0 ! B ! B ! i .˚ 0 = 0 / ! 0
? ? ?
? ? ?
y y y
1 .A ; G /
0 ! ExtX m ! ExtX
1 .A ; j G /
m ! i Homx .; Q=Z/ ! 0
11 This assertion is misleading since (at best) this is true for finite residue fields. According to
a lecture of Siegfried Bosch (20.10.04), the status of the conjecture over a discrete valuation ring
R is as follows. When the residue field k is perfect, it is known when R has mixed characteristic
.0; p/ (Bégeuri), k is finite (McCallum), A is potentially totally degenerate (i.e., after an extension
of the field its reduction is a torus) (Bosch), or A is a Jacobian (Bosch and Lorenzini); it is still
open when K is of equicharacteristic p > 0 and the residue field is infinite. For k nonperfect, the
conjecture fails. The first examples were found by Bertapelle and Bosch, and Bosch and Lorenzini
found many examples among Jacobians.
If conjecture (C.13) holds and 0 is the exact annihilator of , or if D 0 and

0 D ˚ 0 , then all the vertical maps are isomorphisms.
P ROOF. The diagram can be constructed the same way as the diagram in (C.11).
The final statement is obvious. 2
The Raynaud group

Assume now that A has semistable reduction, and write ˚0 for Ax =Aıx . Over k
0 ! T ! Aıx ! B ! 0
with B an abelian variety over k. The next theorem shows that this sequence has
a canonical lifting to R.
For any group scheme G over X , we write G b for the formal completion of G
along the closed point x of X (Hartshorne 1977, II 9).
T HEOREM C.15 There is a smooth group scheme A# over R and canonical iso-
morphisms Ab !
bı !
.A# /^ and A

.A#ı /^ .
(a) There is an exact sequence over R
0 ! T ! A#ı ! B ! 0
with T a torus and B an abelian scheme; the reduction of this sequence

modulo the maximal ideal of R is the above sequence.
(b) Let ˚ D A# =A#ı ; then ˚ is a finite étale group scheme over R whose
special fibre is ˚0 :
(c) Let N D .A#ı /p ; then N is the maximal finite flat subgroup scheme of
the quasi-finite flat group scheme Aıp , and there is a filtration
Ap D .Aıp /K N Tp 0
with N D NK , Tp D .Tp /K , and N=Tp D .Bp /K .

(d) Let At be the dual abelian variety to A, and denote the objects correspond-
ing to it with a prime. The nondegenerate pairing of finite group schemes
over K
A0p Ap ! Gm
induces nondegenerate pairings
N 0 =Tp0 N=Tp ! Gm
A0p =Tp0 N ! Gm :
(e) Assume R is Henselian. If A.K/p D A.K a /p , then ˚p has order p ,

where is the dimension of the maximal subtorus of Ax .
P ROOF. For (a), (b), and (c) see Grothendieck 1972, IX.7.
(d) The restriction of the pairing on A0p Ap to N 0 N extends to a pairing
N 0 N ! Gm induced by the biextension of .At0 ; Aı / by Gm . This pairing is
trivial on Tp0 and Tp , and the quotient pairing on Bp0 Bp is that defined by the
canonical extension of a Poincaré biextension of .BK 0 ; B / by G . This shows
K m
that Tp0 and Tp are the left and right kernels in the pairing N 0 N ! Gm .
The pairing A0p =Tp0 N ! Gm is obviously right nondegenerate. But A0p =Tp0
has order p 2d where d is the common dimension of A and At and is the
common dimension of T and T 0 , and N has order p C2˛ where ˛ is the common
dimensions of B and B 0 . As d D C ˛, this proves that the pairing is also left
nondegenerate.
(e) From the diagram
0 ! Aı .R/ ! A.K/ ! ˚.k/ ! 0

? ? ?
?p ?p ?p
y y y
0 ! Aı .R/ ! A.K/ ! ˚.k/ ! 0
0 ! Aı .R/p ! A.K/p ! ˚.k/p ! Aı .R/.p/ :
Let a 2 ˚.k/p . There will exist a finite flat local extension R0 of R such that a
maps to zero in Aı .R0 /.p/ (because pW Aı ! Aı is a finite flat map), and so the

image of a in ˚.k 0 / lifts to A.K 0 /p . By assumption, A.K/p ! A.K 0 /p , and
so a lifts to A.K/p . This shows that
0 ! Aı .R/p ! A.K/p ! ˚.k/p ! 0
is exact, and the result follows by counting. 2
The group A# is called the Raynaud group scheme.
Néron models and Jacobians

Let X again be any Noetherian normal integral scheme of dimension one, and
let W Y ! X be a flat proper morphism of finite-type. Recall that PicY =S is
defined to be the sheaf on XEt associated with the presheaf X 0 7! Pic.Y X X 0 /.
Write P D PicY =X . When P is representable by an algebraic space, then P D
PicY =X is defined to be the subsheaf of P such that, for all X -schemes X 0 , P .X 0 /

consists of the sections whose image in .Px =Pxı /.X 0 / is torsion for all x 2 X .
Assume now that Y is regular, that the fibres of Y over X are all pure of
dimension 1, and that Y is a smooth and geometrically connected. Moreover, as-

sume that OX ! OY universally (that is, as sheaves on Xfl /. For each closed
point x of X , define dx to be the greatest common divisor of the multiplicities
of the irreducible components of Yx (the multiplicity of Yi Yx is the length of
OYi ;yi where yi is the generic point of Yi /.
T HEOREM C.16 (a) The functor P is representable by an algebraic space locally

of finite type over X .
(b) If dx D 1 for all x , then P is representable by a separated group scheme
over X .
(c) Assume that the residue fields of X are perfect. Under the hypothesis of
(b), P is the Néron model of the Jacobian of Y .

P ROOF. (a) Our assumption that OS ! OX universally says that is coho-
mologically flat in dimension zero. Therefore the statement is a special case of a
theorem of Artin (1969b).
(b) This is a special case of Raynaud 1970, 6.4.5.
(c) This is a special case of Raynaud 1970, 8.1.4. 2
R EMARK C.17 The hypotheses in the theorem are probably too stringent.
The autoduality of the Jacobian

Let C be a smooth complete curve over a field k. Then there is a canonical biex-
tension of .J; J / by Gm , and the two maps J ! Extk1 .J; Gm / are isomorphisms
(and differ only by a minus sign) (see Milne 1986c, 6, or Moret-Bailly 1985). It
is this biextension which we wish to extend to certain families of curves.
Let W Y ! X be a flat projective morphism with fibres pure of dimension
one and with smooth generic fibre; assume that Y is regular and that Y has a sec-
tion s over X . Endow both Y and X with the smooth topology. Then R1 Gm is
representable by a smooth group scheme PicY =X over X . Write P 0 for the ker-
nel of the degree map on the generic fibre; thus P 0 .X / is the set of isomorphism
classes of invertible sheaves on Y whose restriction to s.X / is trivial and whose
restriction to Y has degree zero. Note that PicY =X D P 0 ˚ Z.
T HEOREM C.18 There exists a biextension of .P 0 ; P 0 / by Gm whose restriction

to the generic fibre is the canonical biextension.
P ROOF. In the case that P has connected fibres, this follows from the result
Grothendieck 1972, VIII 7.1b that for any two group schemes B and A over
X with connected fibres, and any nonempty open subset U of X , the restriction
functor
BiextX .A; BI Gm / ! BiextU .A; BI Gm /
is a bijection. See also Moret-Bailly 1985, 2.8.2. For the general case, we refer
the reader to Artin 1967. 2
C ONJECTURE C.19 Assume that Y is the minimal model of its generic fibre.
Then the maps P 0 ! ExtX 1 .P 0 ; G / (of sheaves for the smooth topology) in-
m
duced by the biextension in (C.18) are isomorphisms.
A proof of this conjecture has been announced by Artin and Mazur (Artin
1967), at least in some cases. We shall refer to this as the autoduality hypothesis.
N OTES The concept of a biextension was introduced by Mumford (1969), and was de-
veloped by Grothendieck (1972). Apart from Néron’s Theorem C.6 and the theorems of
Artin and Raynaud (C.16), most of the results are due to Grothendieck. The exposition is
partly based on McCallum 1986.
... and so there ain’t nothing more to write about, and I am rotten glad of it,
because if I’d knowed what a trouble it was to make a book I wouldn’t a tackled
it and ain’t agoing to no more. But I reckon I got to light out for the Territory
ahead of the rest, because Aunt Sally she’s going to sivilize me and I can’t stand
it. I been there before.
H. Finn
328 Bibliography
Bibliography
The date is the date of publication, except for a few seminar talks. Recent preprints have
been assigned the date 1986.12 .
Artin, E. and Tate, J. 1961, Class Field Theory, Harvard University. (Reprinted by Ben-
jamin, New York, 1968.)
Artin, M. 1967, Autoduality of the Jacobian (Notes by B. Wyman), Bowdoin College
(mimeographed notes).
Artin, M. 1969a, The implicit function theorem in algebraic geometry, In Algebraic
Geometry, (Internat. Coll. Bombay, 1968), Oxford U. Press, London, 21–71.
Artin, M. 1969b, Algebraization of formal moduli I, In Global Analysis, Tokyo, 21–71.
Artin, M. 1986, Néron models, In Arithmetic Geometry (Cornell and Silverman, Ed.),
Springer, Heidelberg.
Artin, M., Grothendieck, A., and Verdier, J.-L. 1972/73 Théorie des Topos et Cohomolo-
gie Etale des Schémas, Lecture Notes in Math. 269, 270, 305, Springer, Heidelberg,
1972–73.
Artin, M. and Milne, J. 1976, Duality in the flat cohomology of curves, Invent. Math. 35
111–129.
Artin, M. and Verdier, J.-L. 1964, Seminar on étale cohomology of number fields, AMS
Woods Hole Summer Institute 1964, notes produced in a limited edition only, 5pp.
(A translation has been published in Uspecki Mat. Nauk. 20 (1965), 13-18.)
Artin, M. and Winters, G. 1971, Degenerate fibres and stable reduction of curves, Topol-
ogy 10 373–383.
Atiyah, M. and MacDonald, I. 1969, Introduction to Commutative Algebra, Addison-
Wesley, Reading.
Bashmakov, M. 1972, The cohomology of abelian varieties over a number field, Russian
Math. Surveys 27 no.6, 25–71.
Bégeuri, L. 1980, Dualité sur un corps local à corps résiduel algébriquement clos, Mem.
Soc. Math. Fr., 108, fasc. 4, 121pp.
Berthelot, P. 1981, Le théorème de dualité plate pour les surfaces (d’apres J. S. Milne).
In Surfaces Algébriques (Orsay, 1976–78) Lecture Notes in Math., 868, Springer,
Heidelberg 1981, pp. 203–237.
Berthelot, P., Breen, L, and Messing, W. 1982, Théorie de Dieudonné Cristalline II,
Lecture Notes in Math., 930, Springer, Heidelberg.
Bester, M. 1978, Local flat duality of abelian varieties, Math. Ann. 235, 149–174.
Bloch, S. 1986, Algebraic cycles and higher K-theory, Adv. in Math. 61 (1986), no. 3,
267–304.
Bloch, S. and Kato, K. 1986, p-adic etale cohomology, Inst. Hautes Études Sci. Publ.
Math. No. 63, (1986), 107–152.
Bogomolov, F. 1981, Points of finite order on an abelian variety, Math. USSR Izv. 17,
55–72.
12 Complete references have been added for preprints (the citing date may not match the actual
date of publication)
Bibliography 329
Borel, A. 1979, Automorphic L-functions, In Automorphic Forms, Representations, and

L-functions. Proceedings of Symposia in Pure Math. 33, part 2, 27–61.
Borel, A. and Harder, G. 1978, Existence of discrete cocompact subgroups of reductive
groups over local fields, J. reine angew. Math. 298, 53–64.
Bourbaki, N. Alg., Algèbre, Eléments de Math. 4,6,7,11,14,23,24, Hermann, Paris, 1947–
59.
Bourbaki, N. Tpgy, General Topology, Part 1, Addison-Wesley, Reading, 1966.
Breen, L. 1969a, Extensions of abelian sheaves and Eilenberg-MacLane algebras, Inven-
tiones math. 9, 15–44.
Breen, L. 1969b, On a nontrivial higher extension of representable abelian sheaves, Bull.
A.M.S. 75, 1249–1253.
Breen, L. 1975, Un théorème d’annulation pour certains Exti de faisceaux abeliens. Ann.
Sci. Ecole Norm. Sup. (4), 8, 339–352.
Breen, L. 1981, Extensions du groupe additif sur le site parfait. In Surfaces Algébriques
(Orsay, 1976–78), Lecture Notes in Math. 868, Springer, Heidelberg 1981, 238–262,
Browder, W. 1977, Algebraic K-theory with coefficients Z=p. In Geometric Applications
of Homotopy Theory I, Lecture Notes in Math. 657, Springer, Heidelberg, pp 40-84.
Brumer, A. 1966, Pseudocompact algebras, profinite groups and class formations, J .
Algebra 4, 442–470.
Cartan, H. and Eilenberg, S. 1956, Homological Algebra, Princeton U. Press, Princeton,
1956.
Cassels, J. 1962, Proof of the Hauptvermutung, J. Reine Angew. Math. 211, 95–112.
Cassels, J. 1964, The dual exact sequence, J. Reine Angew. Math. 216, 150-158.
Cassels, J. 1965, On conjectures of Birch and Swinnerton-Dyer, J. Reine Angew. Math.
217, 180–199.
Cossec, F., and Dolgachev, I., 1989, Enriques surfaces. I. Progress in Mathematics, 76.
Birkhäuser Boston, Inc., Boston, MA, 1989. x+397 pp.
Deligne, P. 1966, Cohomologie à support propre et construction du foncteur f Š , 404–421;
in Hartshorne 1966.
Deligne, P. 1977, Théorèmes de finitude en cohomologie `-adique. In Cohomologie Etale
(SGA 4 12 /. Lecture Notes in Math. 569. Springer, Heidleberg, 233-261.
Demazure, M. and Gabriel, P. 1970, Groupes Algébriques, Tome 1, Géométrie algébriques
— généralités — groupes commutatifs, Masson, Paris.
Deninger, C. 1984, On Artin-Verdier duality for function fields, Math. Z. 188, 91–100.
Deninger, C. 1986a, An extension of Artin-Verdier duality to non-torsion sheaves, J.
Reine Angew. Math. 366, 18–31.
Deninger, C. 1986b, `-adic Lefschetz numbers of arithmetic schemes, J. Reine Angew.
Math., 375/376 (1987), 326–345.
Deninger, C. 1986c, Duality in the etale cohomology of one-dimensional proper schemes
and generalizations, Math. Ann. 277 (1987), no. 3, 529–541.
Deninger, C. and Wingberg, K. 1986, Artin-Verdier duality for n-dimensional local fields
involving higher dimensional algebraic K-sheaves, J. Pure Appl. Algebra, 43 (1986),
no. 3, 243–255.
Dieudonné, J. 1964, Fondements de la Géométrie Algébriques Moderne, La Presses de
l’Université de Montréal.
330 Bibliography
Etesse, J.-Y. 1982a, Cohomologie du complexe de de Rham-Witt a coefficients dans un

F -cristal unite et dualite plate pour les surfaces. C. R. Acad. Sci. Paris Ser. I Math.,
294, no. 18, 621–624.
Etesse, J.-Y. 1982b, Complexe de de Rham-Witt a coefficients dans un cristal. C. R.
Acad. Sci. Paris Ser. I Math., 294, no. 17, 581–584.
Fulton, W. 1984, Intersection Theory, Springer, Heidelberg.
Gabber, O. 1983, K-theory of Henselian local rings and Henselian pairs, Handwritten
notes, 1983.
Gamst, J. and Hoechsmann, K. 1970, Products in sheaf cohomology, Tohoku Math. J.
22, 143–162.
Gordon, W. 1979, Linking the conjectures of Artin-Tate and Birch-Swinnerton-Dyer,
Compos. Math. 38, 163–199.
Greenberg, M. 1961, Schemata over local rings, Annals of Math. 73, 624–648.
Greenberg, M. 1966, Rational points in Henselian discrete valuation rings, Publ. Math.
I.H.E.S, 31, 59–64.
Grothendieck, A. 1971, Revêtements Etales et Groupe Fondamental. Séminaire de Géométrie
Algébrique du Bois Marie 1960-1961 (SGA 1). Lecture Notes in Math. 224, Springer,
Heidelberg 1971.
Grothendieck, A. 1972, Groupes de Monodromie en Géométrie Algébrique. I. Séminaire
de Géométrie Algébrique du Bois-Marie 1967-1969 (SGA 7 I). Lecture Notes in
Math. 288. Springer, Heidelberg 1972.
Grothendieck, A. and Dieudonné, J. 1971, Eléments de Géométrie Algébrique, I, Springer,
Heidelberg.
Haberland, K. 1978, Galois Cohomology of Algebraic Number Fields, VEB Deutscher
Verlag der Wissenschaften, Berlin.
Harder, G. 1965/66, Über die Galoiskohomologie halbeinfacher Matrizengruppen, I, Math.
Zeit. 90 (1965), 404-428, II, Ibid., 92 (1966), 396-415.
Harder, G. 1967/68, Bericht über neuere Resultate der Galoiskohomologie halbeinfacher
Matrizengruppen, Jahresbericht d. D.M.V. 70 (1967/68), 182–216.
Harris, M. 1979, p-adic representations arising from descent on abelian varieties, Com-
posit. Math. 39, 177–245, (correction, ibid. 121 (2000), 105–108).
Hartshorne, R. 1966, Residues and Duality, Lecture Notes in Math, 20, Springer, Heidel-
berg.
Hartshorne, R. 1972, Cohomology with compact supports for coherent sheaves on an
algebraic variety. Math. Ann., 195, 199–207.
Hartshorne, R. 1977, Algebraic Geometry. Graduate Texts in Mathematics, No. 52.
Hazewinkel, M. 1969, Abelian extensions of local fields, Thesis, Amsterdam.
Herstein, I. 1968, Noncommutative Rings, Carus Math. Monograph 15, MAA.
Hewitt, E. and Ross, K. 1963, Abstract Harmonic Analysis I, Springer, Heidelberg.
Hilton, P. and Stammbach, U. 1970, A Course in Homological Algebra, Springer, Hei-
delberg.
Illusie, L. 1979, Complexe de de Rham-Witt et cohomologie cristalline, Ann. Sci. Ecole
Norm. Sup. (4), 12, 501–661.
Bibliography 331
Jannsen, U. 1982, Galoismoduln mit Hasse-Prinzip, J. Reine Angew. Math. 337, 154–
158.
Jarden, M. 1969, Rational points on algebraic varieties over large number fields, Bull.
AMS, 75, 603–606.
Jarden, M. 1972, Elementary statements over large algebraic fields, Trans. AMS, 164,
67–91.
Jensen, C. 1972, Les Foncteurs Dérivés de lim et leurs Applications en Théorie des Mod-

ules, Lecture Notes in Math., 254, Springer, Heidelberg.
Kato, Katzuya 1979, A generalization of local class field theory by using K-groups, I, J.
Fac. Science, Univ. Tokyo, Sec IA, 26, 303–376.
Kato, Kazuya 1985/86, Duality theories for the p-primary etale cohomology, I, Algebraic
and topological theories (Kinosaki, 1984), 127–148, Kinokuniya, Tokyo, 1986; II,
Algebraic and Topological Theories, 127–148; II, Compositio Math. 63 (1987), no.
2, 259–270.
Katz, N. 1970, Nilpotent connections and the monodromy theorem: Applications of a
result of Turritin. Inst. Hautes Etudes Sci. Publ. Math., 39, 175–232.
Katz, N. and Lang, S. 1981, Finiteness theorems in geometric classfield theory. (With an
appendix by Kenneth A. Ribet.) Enseign. Math. (2), 27, 285–319.
Kawada, Y. 1960, Class formations. VI. Restriction to a subfamily, J. Fac. Sci. Univ.
Tokyo Sect. I 8, 229–262.
Kawada, Y., 1971, Class formations, Proc. Symp. Pure Math., Vol XX, 96-114, AMS,
Providence.
Kawada, Y. and Tate, J. 1955, On the Galois cohomology of unramified extensions of
function fields in one variable, Amer. J. Math. 77, 197–217.
Kazarnanovski, B. 1972, Proof of a theorem of Tate, Russ. Math. Surveys 27, 56–68.
Klingen, N. 1983, Einbettungsprobleme mit proendlichen Kern und ihre Lösbarkeit bei
Beschrankung der Verzweigung, J. No. Theory 16, 106–118.
Kneser, M. 1965, Galois-Kohomologie halbeinfacher algebraischer Gruppen über p-
adischen Körpern, I. Math. Zeit. 88 (1965), 40-47; II. ibid. 89 (1965), 250–272.
Kneser, M. 1966, The Hasse principle for H 2 of simply connected groups, Proc. Sympos.
Pure Math., A.M.S. 9, 159–163.
Kneser, M. 1969, Lectures on Galois Cohomology of Classical Groups, Lectures on
Math. 47, Tata, Bombay.
Kottwitz, R. 1984, Stable trace formula: cuspidal tempered terms, Duke Math. J. 51,
611–650.
Kramer, K. 1977, Two-descent for elliptic curves in characteristic two, Trans. Amer.
Math. Soc. 232, 279–295.
Labesse, J.P. 1984, Cohomologie, L-groups et fonctorialité, Compositio Math. 55, 163–
184.
Lang, S. 1958, Introduction to Algebraic Geometry, Interscience, New York.
Lang, S. 1959, Abelian Varieties, Interscience, New York.
Lang, S. 1966, Rapport sur la Cohomologie des Groupes, Benjamin, New York.
Lang, S. 1970, Algebraic Number Theory, Addison-Wesley, Reading.
Lang, S. 1983, Fundamentals of Diophantine Geometry, Springer, Heidelberg.
332 Bibliography
Lang, S. and Tate, J. 1958, Principal homogeneous spaces over abelian varieties, Amer.
J. Math. 80, 659–684.
Langlands, R. 1968, Representations of algebraic groups, Mimeographed notes, Yale,
31pp (Pacific J. Math. 1997, Special Issue, 231–250).
Langlands, R. 1983, Les Débuts d’une Formule des Traces Stable, Publ. Math. Univ.
Paris VII, 13.
Lichtenbaum, S. 1984, Values of zeta-functions at non-negative values, In Number The-
ory, Noordwijkerhout 1983, Lecture Notes in Math. 1068, Springer, Heidelberg,
1984.
Lichtenbaum, S. 1986, The construction of weight two arithmetic cohomology, Invent.
Math. 88 (1987), no. 1, 183–215.
Litvak, G. 1980, Nondegeneracy of the Tate pairing in Galois cohomology of finite mod-
ules over general local fields of zero characteristic, Ukrain. Math. J. 32, 392-394.
McCallum, W. 1986, Duality theorems for Néron models, Duke Math. J. 53 (1986), no.
4, 1093–1124.
Manin, Ju. 1971, Cyclotomic fields and modular curves, Russ. Math. Surveys, 26, no. 6,
7–78.
Mattuck, A. 1955, Abelian varieties over p-adic ground fields, Ann. of Math. 62, 92–
119.
Mazur, B. 1970a, Local flat duality, Amer. J. Math. 92, 343–361.
Mazur, B. 1970b, Finite flat structures, Proc. Symp. Pure Math. 17, A.M.S., Providence,
219–225.
Mazur, B. 1972, Rational points of abelian varieties with values in towers of number
fields, Invent. Math. 18, 183–266.
Mazur, B. 1973, Notes on étale cohomology of number fields, Ann. Sci. Ecole Norm.
Sup. 6, 521–552.
Mazur, B. and Roberts, L. 1970, Local Euler characteristics, Invent. Math. 9, 201–234.
Merkur’ev, A. and Suslin, A. 1982, K-cohomology of Severi-Brauer varieties and the
norm residue homomorphism, Izv. Akad. Nauk. SSSR Ser. Mat., 46, 1011–1046,
1135–1136.
Messing, W. 1972, The Crystals Associated to Barsotti-Tate Groups: with Applications
to Abelian Schemes, Lecture Notes in Math. 264, Springer, Heidelberg.
Milne, J. 1967, The conjectures of Birch and Swinnerton-Dyer for constant abelian vari-
eties over function fields, Thesis, Harvard University. (Available at
www.jmilne.org/math/.)
Milne, J. 1968, The Tate-Safarevic group of a constant abelian variety, Invent. Math. 5,
63–84.
Milne, J. 1970a, The homological dimension of commutative group schemes over a per-
fect field, J. of Algebra, 16, 436–441.
Milne, J. 1970b, Elements of order p in the Tate-Safarevic group, Bull. London Math.
Soc. 2, 293–296.
Milne, J. 1970/72, Weil-Chatelet groups over local fields, Ann. Sci. Ecole Norm. Sup. 3
(1970), 273–284; addendum, ibid., 5 (1972), 261–264.
Milne, J. 1972, On the arithmetic of abelian varieties, Invent. math. 17, 177–190.
Milne, J. 1973, On a theorem of Mazur and Roberts, Amer. J. Math. 95, 80–86.
Bibliography 333
Milne, J. 1975, On a conjecture of Artin and Tate, Annals of Math. 102, 517–533.
Milne, J. 1976, Duality in the flat cohomology of a surface, Ann. scient. ENS, 9, 171–
202.
Milne, J. 1980, Etale Cohomology, Princeton Univ. Press, Princeton.
Milne, J. 1981, Comparison of the Brauer group with the Tate-Safarevic group, J. Fac.
Science, Univ. Tokyo, Sec. IA, 28, 735–743.
Milne, J. 1986a, Values of zeta functions of varieties over finite fields, Amer. J. Math.,
108, 297–360.
Milne, J. 1986b, Abelian varieties, In Arithmetic Geometry (Cornell, G. and Silverman,
J. Eds.), Springer, Heidelberg.
Milne, J. 1986c, Jacobian varieties, Arithmetic Geometry (Cornell and Silverman, Ed.),
Mitchell, B. 1965, Theory of Categories, Academic Press, New York.
Moret-Bailly, L. 1985, Métriques permises, In Séminaire sur les Pinceaux Arithmétiques:
La Conjecture de Mordell, (Szpiro ed.), Astérisque 127, 29–87.
Mumford, D. 1965, Geometric Invariant Theory, Springer, Heidelberg.
Mumford, D. 1969, Biextensions of formal groups, In Algebraic Geometry, (Internat.
Coll. Bombay, 1968), Oxford University Press, Oxford, 1969.
Mumford, D. 1970, Abelian Varieties, Oxford University Press, Oxford.
Narkiewicz, W. 1974, Elementary and Analytic Theory of Algebraic Numbers, PWN –
Polish Scientific Pub., Warsaw.
Néron, A. 1964, Modèles minimaux des variétés abéliennes sur les corps locaux et globaux,
Publ. Math. I.H.E.S. 21, 1–128.
Neumann, O. 1977, Proper solutions of the imbedding problem with restricted ramifica-
tion, Acta Arith. 33, 49–52.
Oersterlé, J. 1982/83, Compatibilité de la suite exacte de Poitou-Tate aux mesures de
Haar, Séminaire de Théorie des Nombres de Bordeaux, 1982/83, Ex 19.
Oesterlé, J. 1984, Nombres de Tamagawa et groupes unipotents en caractristique p, In-
ventiones math., 78, 13–88.
Ogg, A. 1962, Cohomology of abelian varieties over function fields, Ann. of Math., 76,
284–292.
Ono, Takahashi 1961, Arithmetic of algebraic tori, Annals of Math., 74, 101–139.
Ono, Takahashi 1963, On the Tamagawa number of algebraic tori, Annals of Math., 78,
47–73.
Oort, F. 1966, Commutative Group Schemes, Lecture Notes in Math. 15, Springer, Hei-
delberg.
Oort, F. 1967, Embeddings of finite group schemes into abelian schemes Mimeographed
notes, Bowdoin College, 9pp..
Oort, F. 1971, Finite group schemes, local moduli for abelian varieties, and lifting prob-
lems. Compositio Math., 23, 265–296.
Oort, F. and Tate, J. 1970, Group schemes of prime order, Ann. scient. E.N.S. 3, 1–21.
Poitou, G. 1966, Remarques sur l’homologie des groupes profini, In Les Tendances
Géometrie en Algèbre et Théorie des Nombres, CNRS, Paris, 201–213.
Poitou, G. 1967, Cohomologie Galoisienne des Modules Finis, Dunod, Paris.
334 Bibliography
Raynaud, M. 1964/65, Caractéristique d’Euler-Poincaré d’un faisceau et cohomologie

des variétés abéliennes (d’après Ogg-Shafarévitch et Grothendieck), Séminaire Bour-
baki 1964/65, 286. (Published in: Dix Exposés sur la Cohomologie des Schémas,
North-Holland, Amsterdam, 1968, 12–30.)
Raynaud, M. 1966, Modèles d’Néron, C.R. Acad. Sci. Paris 262, 413–416.
Raynaud, M. 1970, Anneaux Locaux Henséliens, Lecture Notes in Math. 169.
Raynaud, M. 1971, Specialization du foncteur de Picard, Publ. Math. I.H.E.S. 38, 27–76.
Raynaud, M. 1974, Schémas en groupes de type .p; :::; p/. Bull. Soc. Math. France,
102, 241–280.
Raynaud, M. 1979, “p-torsion” du schema de Picard. In Journées de Géométrie Algébrique
de Rennes (Rennes, 1978), Vol. II, Astérisque 64 (1979), pp. 87–148.
Rim, D. and Whaples, G. 1966, Global norm residue symbol map over quasi-finite field,
Nagoya Math J. 27, 323–329.
Roberts, L. 1973, The flat cohomology of group schemes of rank p, Amer. J. Math. 95,
688–702.
Roos, J.-E. 1961, Sur les foncteurs dérivés de lim. Applications. Compte Rendu Acad.

Sci. Paris 252, 3702–3704. Erratum ibid. 254 p1722.
Rubin, K. 1985, Elliptic curves and Zp -extensions, Compos. Math. 56, 237–250.
Sansuc, J.-J. 1981, Groupe de Brauer et arithmétique des groupes algébriques linéaires
sur un corps de nombres, J. reine angew. Math. 327, 12–80.
Serre, J.-P. 1959, Groupes Algébriques et Corps de Classes, Hermann, Paris.
Serre, J.-P. 1960, Groupes Proalgébriques, Publ. Math. IHES, 7, 341–403.
Serre, J.-P. 1961, Sur les corps locaux à corps résiduel algébriquement clos, Bull. Soc.
Math. France 89, 105–154.
Serre, J.-P. 1962, Corps Locaux, Hermann, Paris.
Serre, J.-P. 1964, Cohomologie Galoisienne, Lecture Notes in Math. 5, Springer, Heidel-
berg.
Serre, J-P. 1964/71, Sur les groupes de congruence des variétés abéliennes, I. Izv. Akad.
Nauk SSSR, Ser. Mat.28 (1964), 3-20; II. ibid., 35 (1971), 731–737.
Serre, J.-P. 1967, Local class field theory, In Algebraic Number Theory, (Ed. Cassels J.
and Fröhlich, A.), Academic Press, London and New York, 129–161.
Serre, J.-P. 1967b, Représentations Linéaires des Groupes Finis, (Second edition) Her-
mann, Paris.
Serre, J.-P. 1977, Modular forms of weight one and Galois representations, In Algebraic
Number Fields: L-Functions and Galois Properties, Academic Press, London, 193-
268.
Shafarevich, I. 1954, Construction of fields of algebraic numbers with given solvable
Galois group, Izv. Akad. Nauk SSSR, Ser. Mat. 18, 525–578.
Shafarevich, I. 1962, Principal homogeneous spaces defined over a function field, A.M.S.
Transl., Series 2, 36, 85–114.
Shatz, S. 1964, Cohomology of artinian group schemes over local fields, Ann. of Math.,
79, 411–449.
Shatz, S. 1966, The cohomological dimension of certain Grothendieck topologies, Ann.
of Math., 83, 572–595.
Bibliography 335
Shatz, S. 1967, The cohomology of certain elliptic curves over local and quasi local fields,
Illinois J. Math. 11, 234–241.
Shatz, S. 1972, Profinite groups, Arithmetic, and Geometry, Annals of Math. Studies 67,
Princeton U. P., Princeton.
Shatz, S. 1986, Group schemes, formal groups, and p-divisible groups, In Arithmetic
Geometry, (Cornell and Silverman, Ed.) Springer, Heidelberg.
Shelstad, D. 1986, Twisted endoscopic groups in the abelian case, preprint.
Suslin, A. 1983a Torsion in K2 of fields, K-Theory 1 (1987), no. 1, 5–29.
Suslin, A. 1983b, On the K-theory of algebraically closed fields, Invent. math., 73, 241–
245.
Suslin, A. 1984, On the K-theory of local fields, J. Pure Appl. Algebra, 34, 301–318.
Swan, R. 1960, Induced representations and projective modules, Ann of Math., 77. 267–
291.
Szpiro, L. 1985, Degrés, intersections, hauteurs, In Séminaire sur les Pinceaux Arithmétiques:
La Conjecture de Mordell, (Szpiro ed.), Astérisque 127, 11–28.
Takahashi, Toyofumi 1969, Galois cohomology of finitely generated modules, Tohoku
Math. J. 11, 102–111.
Tate, J. 1957/58, WC-groups over p-adic fields, Séminaire Bourbaki, Exposé 156, 13pp.
Tate, J. 1962, Duality theorems in Galois cohomology over number fields, Proc. Intern.
Congress Math., Stockholm, pp234–241.
Tate, J. 1965/66, On the conjectures of Birch and Swinnerton-Dyer and a geometric ana-
log, Séminaire Bourbaki, 306. (Reprinted in: Dix Exposés sur la Cohomologie des
Schémas, North-Holland, Amsterdam, 1968).
Tate, J. 1966, Letter to T. Springer, dated 13 Jan 1966, 6pp.
Tate, J. 1967a Global class field theory, In Algebraic Number Theory (Cassels, J. and
Frohlich, A., ed.) Academic Press, London and New York, 163–203.
Tate, J. 1967b, p-divisible groups, Proc. of a Conf. on Local Fields (at Driebergen),
Springer, Heidelberg, 1967, 158–183.
Tate, J. 1979, Number theoretic background. In Automorphic Forms, Representations,
and L-functions, Part 2, Proc. Symp. Pure Math. 33, AMS, 3–26.
Thomason, R. 1984a, Absolute cohomological purity. Bull. Soc. Math. France, 112,
397–406.
Uchida, K. 1969, On Tate duality theorems on Galois cohomology, Tohoku Math. J. 21,
92–101.
Vvedens’kii, O. 1973, On local “class fields” of elliptic curves, Math. USSR Izvestija, 7,
19–84.
Vvedens’kii, O. 1976, On quasi-local “class fields” of elliptic curves, I, Math. USSR
Izvestija, 10, 913–936.
Vvedens’kii, O. 1979a, The frequency of the appearance of the Artin-Milne effect in
elliptic curves, Dokl. Akad. Nauk. SSSR 245, 780–781.
Vvedens’kii, O. 1979b, The Artin effect in elliptic curves, I, Izv. Akad. Nauk. SSSR,
Ser. mat. 43, 1042-1053, 1197.
Vvedens’kii, O. 1980/81, The Artin effect in abelian varieties, I , Dokl. Akad. Nauk.
SSSR 252 (1980), 20; II, Izv. Akad. Nauk SSSR 45 (1981), 23–46, 239.
336 Bibliography
Vvedens’kii, O. and Krupjak, I. 1976, A remark on the nondegeneracy of the Tate pairing
in the Galois cohomology of finite modules over a general local field, Ukrain. Mat.
Z. 28, 526–528.
Wake, R. 1986, Phantom points on abelian varieties, preprint.
Weil, A. 1961, Adeles and Algebraic Groups, Mimeographed notes, IAS, 1961. (Reprinted
as Progress in Math. 23, Birkhauser, Boston, 1982.)
Weil, A. 1967, Basic Number Theory, Springer, Heidelberg.
Weiss, E. 1969, Cohomology of Groups, Academic Press, New York.
Whaples, G. 1952/54, Generalized local class field theory, I, Duke Math. J. 19 (1952),
505-517; II, ibid. 21 (1954), 247-255; III, ibid. 21 (1954), 575-581; IV, ibid. 21
(1954), 583-586.
Zink, Th. 1978, Etale cohomology and duality in number fields, Appendix to Haberland
1978, p127-145.
Index
absolute different, 260 dual

admissible homomorphism, 114 Cartier, vii
augmented cup-product, 8 Pontryagin, vii
autoduality hypothesis, 327 dual torus, 113
duality
Barsotti-Weil formula, 40, 198 Breen-Serre, 226
biextension, 316
Brauer group, 135 embedding, 307
Breen-Serre duality, 226, 264 embedding problem, 125
excellent, 130
canonical pairing, 323 extensions
canonical smooth resolution, 311 of algebraic groups, 11
Cartier dual, vii of modules, 4
Cassels-Tate pairing, 203, 297 of sheaves, 142
Čech complex, 145
class formation, 17 field
P, 25 d-local, 37
cofinite-type, vii global, vii
cohomology group local, vii
Tate, 2 quasi-finite, 127
cohomology groups finite support, 189
with compact support, 165, 220 Frobenius map, 225
cohomology sequence of a pair, 141
complete resolution, 2 global field, vii
conjecture Greenberg realization, 258
duality, 216 group
Neron components, 323 Brauer, 135
of Birch and Swinnerton-Dyer, 95 Selmer, 75
constructible, 146 Tate-Shafarevich, 74
Z, 146 Weil, 102
countable sheaf, 189 group of multiplicative type, 28
cup-product group scheme
augmented, 8 p-primary, 227
different of a finite group scheme, 243 Hasse principle, 117, 119, 120, 136
discriminant ideal, 241 height, 95
divisible subgroup, vii of a finite group scheme, 225
337
338 INDEX
Henselian, 127 Raynaud group scheme, 325

Henselization, 127 reciprocity law, 134
homomorphism reciprocity map, 18
compatible of modules, 11 reduced Galois group, 134
resolution
idèles, 132 complete, 2
induced module, 8 standard, 3
invariant map, 17
inverse limits, 15 semistable reduction, 320
sheaf
kernel-cokernel exact sequence, 16 constructible, 146
flat, 230
local duality theorem holds, 252 -compact, 13
local field, vii site
d, 37 big étale, viii
big flat, viii
map étale, viii
corestriction, 106 perfect, 226
Frobenius, 225 small fpqf, viii
invariant, 17 smooth, viii
reciprocity, 18 solvable, 117
transfer, 50 spectral sequence
Verlagerung, 50 for Exts, 4
mapping cone, 147 Hochschild-Serre, 7
module splits generically, 245
induced, 8 standard complete resolution, 3
morphism standard resolution, 3
strict, 13 strict morphism, 13
subgroup
Néron model, 320 divisible, vii
norm group, 129
Tamagawa number, 124
order Tate cohomology group, 2
of a finite group scheme, 241 theorem
abelian class field theory, 106, 114, 115
pairing algebraic structure of cohomology group,
Cassels-Tate, 203 262
pairings an exact sequence, 267, 268
compatibility of, 9 behaviour of B-S/D for isogenies, 298
in the derived category, 230 cohomology of finite flat group schemes,
perfect scheme, 225 261
algebraic, 226 compatibility of B-S/D, 97
perfection compatibility of pairings, 231
of a scheme, 226 duality
Poincaré biextension, 317 for a class formation, 25
Poincaré sheaf, 317 for a d-local field, 38, 159
Pontryagin dual, vii for a global field, 52, 56
for a local field, 26
quasi-finite field, 127 for a torus, 191, 194
339
for abelian schemes, 288, 297, 306 Verlagerung map, 50

for abelian varieties, 82, 265, 285 Verschiebung, 225
for abelian variety, 41
for an abelian scheme, 199 Weil group, 102
for archimedean local field, 35
for class formation, 21
for constructible sheaves, 153, 158,
177
for finite components, 288
for finite flat group schemes, 233
for finite group schemes, 264, 277,
281, 290, 302, 304
for flat sheaves, 252
for Henselian local field, 36
for higher dimensional schemes, 160,
210, 212, 214, 215
for Neron components, 248
for Neron models, 249, 251
for perfect group schemes, 229
for Tate-Shafarevich groups, 202
for tori, 113, 265, 276
embedding finite group schemes, 308,
311
Euler-Poincare characteristic, 241
Euler-Poincare characteristics, 174, 255
existence of Neron models, 319
global class field theory, 131
global Euler-Poincare characteristic, 67,
71
Hasse principle, 117, 121, 123
local Euler-Poincare characteristic, 31
local reciprocity law, 129
semistable reduction, 320
spectral sequence for Exts, 5
spectral sequence of exts, 142
Tamagawa number, 124
Tate-Nakayama, 3
topological duality for vector spaces,
224
unramified duality, 30
weak Mordell-Weil, 74
torus, viii, 23
dual, 113
over a scheme, 191
unramified character, 115

unramified module
module, 30

(J. S. Milne) Arithmetic Duality Theorems (B-Ok - Xyz)

Uploaded by

Copyright:

Available Formats

(J. S. Milne) Arithmetic Duality Theorems (B-Ok - Xyz)

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

(J. S. Milne) Arithmetic Duality Theorems (B-Ok - Xyz)

Uploaded by

Copyright:

Available Formats

Arithmetic Duality Theorems

A paperback version of this work is available from booksellers worldwide and

II Etale Cohomology 139

III Flat Cohomology 217

3 Global results: number field case . . . . . . . . . . . . . . . . . . 252

Preface to the first edition.

1 Inevitably, the sentence preceding this in the original contained a solecism

Preface to the second edition.

20.02.2004. First version on web.

Notations and Conventions

variety) of an abelian variety is denoted by At . For a vector space M , M _ denotes

3 And sometimes, in this edition, '.

abelian varieties (Milne 1986b).

Tate (modified) cohomology groups

HTr .G; M / D H r .G; M /; r > 0;

of finitely generated free ZŒG-modules, together with an element e 2 LG

with .x/e D d0 .x/ and .m/ D me. If we let

H r .G; M / D H r .HomG .LC

By the standard resolution LC C

there is a family of cup-product pairings

.x; y/ 7! x Y yW HTr .G; M /  HTs .G; N / ! HTrCs .G; P /

with the following properties:

T HEOREM 0.2 (TATE -NAKAYAMA ) Let G be a finite group, C a G -module, and

x 7! x Y uW HTr .G; M / ! HTrC2 .G; M ˝ C /

for all integers r .

P ROOF. Serre 1962, IX 8. 2

.f; g/ 7! f  gW ExtrG .N; P /  ExtsG .M; N / ! ExtrCs

such that f  g is obtained from f W N ! P and gW M ! N by composition

.f; g/ 7! f ı gW HomG .N; P /  HomG .M; N / ! HomG .M; P /:

ExtrG .M; N /  H s .G; M / ! H rCs .G; N /:

An r-fold extension of M by N defines in a natural way a class in ExtrG .M; N /

A spectral sequence for Exts

Then HomH .M; N / is a discrete G=H -module, and we define ExtH

be the r th right derived functor of the left exact functor

N 7! HomH .M; N /W ModG ! ModG=H :

In the case that H D f1g, we drop it from the notation; in particular,

T HEOREM 0.3 Let H be a closed normal subgroup of G , and let N and P be

ExtrG=H .M; ExtH

This will be shown to be the spectral sequence of a composite of functors, but

P ROOF. There is a standard isomorphism

HomG .M; HomH .N; P //

L EMMA 0.5 If I is an injective G -module and N is a torsion-free G -module,

P ROOF. We have to check that

HomG=H .; HomH .N; I //W ModG=H ! Ab

P ROOF. We use a resolution of N

by torsion-free G-modules to compute TorZr .M; N /. Thus TorZ1 .M; N / and

0 ! TorZ1 .M; N / ! M ˝Z N1 ! M ˝Z N0 ! TorZ0 .M; N / ! 0;

and TorZr .M; N / D 0 for r  2. For each open subgroup U of G containing H ,

The direct limit of these sequences is an injective resolution

0 ! HomH .N; I / ! HomH .N0 ; I / ! HomH .N1 ; I / ! 0

When M is finitely generated, this is simply a long exact sequence

E XAMPLE 0.9 In the case that N D Z, the spectral sequence becomes

The map ExtrG=H .M; P H / ! ExtrG .M; P / is obviously an isomorphism for

ExtrG .M; N / D lim ExtrGi .M; N /.

P ROPOSITION 0.12 Let

be exact sequences of G -modules. Then a pair of pairings

coinciding on M 0  N 0 defines a canonical family of (augmented cup-product)

P ROOF. See Lang 1966, Chapter V. 2

H r .G; M /  H s .G; N / ! H rCs .G; P / (cup-product)

with .x/e D d0 .x/ and .m/ D me. If we let

.x; y/ 7! x Y yW HTr .G; M / HTs .G; N / ! HTrCs .G; P /

.f; g/ 7! f gW ExtrG .N; P / ExtsG .M; N / ! ExtrCs

such that f g is obtained from f W N ! P and gW M ! N by composition

.f; g/ 7! f ı gW HomG .N; P / HomG .M; N / ! HomG .M; P /:

ExtrG .M; N / H s .G; M / ! H rCs .G; N /:

and TorZr .M; N / D 0 for r 2. For each open subgroup U of G containing H ,

coinciding on M 0 N 0 defines a canonical family of (augmented cup-product)

H r .G; M / H s .G; N / ! H rCs .G; P / (cup-product)

coinciding on M 0 N 0 . These data give rise to canonical maps H r .G; M 00 / !

H r .G; M 00 / H s .G; N 00 / ! H rCsC1 .G; P / (augmented cup-product)

Extrk .A; B/ H s .k; A/ ! H rCs .k; B/:

ExtrG .A; B/ H s .k; A/ ! H rCs .k; A/

0 ! lim An ! lim Bn ! ! lim 1 Cn ! 0.

h ; iW C G Homcts .G; Q=Z/ ! Q=Z

hc; i D .recG .c// all c 2 C G , 2 Homcts .G ab ; Q=Z/.

for all open subgroups V U with V normal in U . It is equivalent to the usual

1 .U / D lim 1 .UL /; K L K s; ŒLW K < 1:

L EMMA 1.9 For r 4, ExtrG .M; C / D 0; when M is torsion-free, Ext3G .M; C /

is an isomorphism for all r 3. The diagram

H r .G; M / H 1r .G; M / ! H 1 .G; Q=Z/ ' Q=Z:

ExtrG .X .T /; C / H 2r .G; X .T // ! H 2 .G; C / ' Q=Z