Cohomology of Number Fields

Cohomology
of
Number Fields
by
Jürgen Neukirch
Alexander Schmidt
Kay Wingberg
Second Edition
corrected version 2.3, May 2020
Electronic Edition
www.mathi.uni-heidelberg.de/~schmidt/NSW2e/
Free for private, non-commercial use only.
All rights reserved by Springer-Verlag Berlin, Heidelberg, New York.
Printed version available: ISBN 978-3-540-37888-0. www.springer.com
Electronic Edition. Free for private, not for commercial use.
This free electronic edition will be corrected from time to time. If you have
comments or found a mistake/misprint, please send an email to
schmidt@mathi.uni-heidelberg.de .
Give a short description including page and line number and check beforehand
on
www.mathi.uni-heidelberg.de/~schmidt/NSW2e/
whether you have the most recent version of the file and whether the mistake
you found is already listed in the errata file.

c J.Neukirch, A.Schmidt, K.Wingberg: www.mathi.uni-heidelberg.de/∼schmidt/NSW2e/
Vorwort
Als unser Freund und Lehrer Jürgen Neukirch Anfang 1997 starb, hinter-
ließ er den Entwurf zu einem Buch über die Kohomologie der Zahlkörper,
welches als zweiter Band zu seiner Monographie Algebraische Zahlentheorie
gedacht war. Für die Kohomologie proendlicher Gruppen, sowie für Teile der
Kohomologie lokaler und globaler Körper lag bereits eine Rohfassung vor,
die schon zu einer regen Korrespondenz zwischen Jürgen Neukirch und uns
geführt hatte.
In den letzten zwei Jahren ist, ausgehend von seinem Entwurf, das hier
vorliegende Buch entstanden. Allerdings wussten wir nur teilweise, was Jürgen
Neukirch geplant hatte. So mag es sein, dass wir Themen ausgelassen haben,
welche er berücksichtigen wollte, und anderes, nicht Geplantes, aufgenommen
haben.
Jürgen Neukirchs inspirierte und pointierte Art, Mathematik auf hohem
sprachlichen Niveau darzustellen, ist für uns stets Vorbild gewesen. Leider
erreichen wir nicht seine Meisterschaft, aber wir haben uns alle Mühe gegeben
und hoffen, ein Buch in seinem Sinne und nicht zuletzt auch zum Nutzen seiner
Leser fertig gestellt zu haben.
Heidelberg, im September 1999 Alexander Schmidt

Kay Wingberg


Introduction
Number theory, one of the most beautiful and fascinating areas of mathe-
matics, has made major progress over the last decades, and is still developing
rapidly. In the beginning of the foreword to his book Algebraic Number Theory,
J. Neukirch wrote
" Die Zahlentheorie nimmt unter den mathematischen Disziplinen

eine ähnlich idealisierte Stellung ein wie die Mathematik selbst
unter den anderen Wissenschaften." ∗)
Although the joint authors of the present book wish to reiterate this statement,
we wish to stress also that number theory owes much of its current strong
development to its interaction with almost all other mathematical fields. In
particular, the geometric (and consequent functorial) point of view of arithmetic
geometry uses techniques from, and is inspired by, analysis, geometry, group
theory and algebraic topology. This interaction had already started in the
1950s with the introduction of group cohomology to local and global class
field theory, which led to a substantial simplification and unification of this
area.
The aim of the present volume is to provide a textbook for students, as well
as a reference book for the working mathematician on cohomological topics
in number theory. Its main subject is Galois modules over local and global
fields, objects which are typically associated to arithmetic schemes. In view
of the enormous quantity of material, we were forced to restrict the subject
matter in some way. In order to keep the book at a reasonable length, we
have therefore decided to restrict attention to the case of dimension less than
or equal to one, i.e. to the global fields themselves, and the various subrings
contained in them. Central and frequently used theorems such as the global
duality theorem of G. POITOU and J. TATE, as well as results such as the theorem
of I. R. ŠAFAREVIČ on the realization of solvable groups as Galois groups over
global fields, had been part of algebraic number theory for a long time. But the
proofs of statements like these were spread over many original articles, some
of which contained serious mistakes, and some even remained unpublished. It
was the initial motivation of the authors to fill these gaps and we hope that the
result of our efforts will be useful for the reader.
In the course of the years since the 1950s, the point of view of class field
theory has slightly changed. The classical approach describes the Galois groups
∗) “Number theory, among the mathematical disciplines, occupies a similar idealized position
to that held by mathematics itself among the sciences.”

viii Introduction
of finite extensions using arithmetic invariants of the local or global ground

field. An essential feature of the modern point of view is to consider infinite
Galois groups instead, i.e. one investigates the set of all finite extensions of the
field k at once, via the absolute Galois group Gk . These groups intrinsically
come equipped with a topology, the Krull topology, under which they are
Hausdorff, compact and totally disconnected topological groups. It proves to
be useful to ignore, for the moment, their number theoretical motivation and to
investigate topological groups of this type, the profinite groups, as objects of
interest in their own right. For this reason, an extensive “algebra of profinite
groups” has been developed by number theorists, not as an end in itself, but
always with concrete number theoretical applications in mind. Nevertheless,
many results can be formulated solely in terms of profinite groups and their
modules, without reference to the number theoretical background.
The first part of this book deals with this “profinite algebra”, while the
arithmetic applications are contained in the second part. This division should
not be seen as strict; sometimes, however, it is useful to get an idea of how
much algebra and how much number theory is contained in a given result.
A significant feature of the arithmetic applications is that classical reciprocity
laws are reflected in duality properties of the associated infinite Galois groups.
For example, the reciprocity law for local fields corresponds to Tate’s duality
theorem for local cohomology. This duality property is in fact so strong that it
becomes possible to describe, for an arbitrary prime p, the Galois groups of the
maximal p-extensions of local fields. These are either free groups or groups
with a very special structure, which are now known as Demuškin groups. This
result then became the basis for the description of the full absolute Galois
group of a p-adic local field by U. JANNSEN and the third author.
The global case is rather different. As was already noticed by J. TATE,
the absolute Galois group of a global field is not a duality group. It is the
geometric point of view, which offers an explanation of this phenomenon: the
duality comes from the curve rather than from its generic point. It is therefore
natural to consider the étale fundamental groups π1et (Spec(Ok,S )), where S is
a finite set of places of k. Translated to the language of Galois groups, the
fundamental group of Spec(Ok,S ) is a quotient of the full group Gk , namely,
the Galois group Gk,S of the maximal extension of k which is unramified
outside S. If S contains all places that divide the order of the torsion of a
module M , the central Poitou-Tate duality theorem provides a duality between
the localization kernels in dimensions one and two. In conjunction with Tate
local duality, this can also be expressed in the form of a long 9-term sequence.
The duality theorem of Poitou-Tate remains true for infinite sets of places S
and, using topologically restricted products of local cohomology groups, the
long exact sequence can be generalized to this case. The question of whether

Introduction ix
the group Gk,S is a duality group when S is finite was positively answered by
the second author.
As might already be clear from the above considerations, the basic technique
used in this book is Galois cohomology, which is essential for class field the-
ory. For a more geometric point of view, it would have been desirable to have
also formulated the results throughout in the language of étale cohomology.
However, we decided to leave this to the reader. Firstly, the technique of sheaf
cohomology associated to a Grothendieck topos is sufficiently covered in the
literature (see [5], [139], [228]) and, in any case, it is an easy exercise (at least
in dimension ≤ 1) to translate between the Galois and the étale languages. A
further reason is that results which involve infinite sets of places (necessary
when using Dirichlet density arguments) or infinite extension fields, can be
much better expressed in terms of Galois cohomology than of étale cohomo-
logy of pro-schemes. When the geometric point of view seemed to bring a
better insight or intuition, however, we have added corresponding remarks or
footnotes. A more serious gap, due to the absence of Grothendieck topologies,
is that we cannot use flat cohomology and the global flat duality theorem of
Artin-Mazur. In chapter VIII, we therefore use an ad hoc construction, the
group BS , which measures the size of the localization kernel for the first flat
cohomology group with the roots of unity as coefficients.
Let us now examine the contents of the individual chapters more closely.
The first part covers the algebraic background for the number theoretical ap-
plications. Chapter I contains well-known basic definitions and results, which
may be found in several monographs. This is only partly true for chapter II: the
explicit description of the edge morphisms of the Hochschild-Serre spectral
sequence in §2 is certainly well-known to specialists, but is not to be found in
the literature. In addition, the material of §3 is well-known, but contained only
in original articles.
Chapter III considers abstract duality properties of profinite groups. Among
the existing monographs which also cover large parts of the material, we should
mention the famous Cohomologie Galoisienne by J.-P. SERRE and H. KOCH’s
book Galoissche Theorie der p-Erweiterungen. Many details, however, have
been available until now only in the original articles.
In chapter IV, free products of profinite groups are considered. These are
important for a possible non-abelian decomposition of global Galois groups
into local ones. This happens only in rather rare, degenerate situations for
Galois groups of global fields, but it is quite a frequent phenomenon for
subgroups of infinite index. In order to formulate such statements (like the
arithmetic form of Riemann’s existence theorem in chapter X), we develop the
concept of the free product of a bundle of profinite groups in §3.

x Introduction
Chapter V deals with the algebraic foundations of Iwasawa theory. We will

not prove the structure theorem for Iwasawa modules in the usual way by using
matrix calculations (even though it may be more acceptable to some mathe-
maticians, as it is more concrete), but we will follow mostly the presentation
found in Bourbaki, Commutative Algebra, with a view to more general situa-
tions. Moreover, we present results concerning the structure of these modules
up to isomorphism, which are obtained using the homotopy theory of modules
over group rings, as presented by U. JANNSEN.
The central technical result of the arithmetic part is the famous global duality
theorem of Poitou-Tate. We start, in chapter VI, with general facts about Galois
cohomology. Chapter VII deals with local fields. Its first three sections largely
follow the presentation of J.-P. SERRE in Cohomologie Galoisienne. The
next two sections are devoted to the explicit determination of the structure
of local Galois groups. In chapter VIII, the central chapter of this book, we
give a complete proof of the Poitou-Tate theorem, including its generalization
to finitely generated modules. We begin by collecting basic results on the
topological structure, universal norms and the cohomology of the S-idèle class
group, before moving on to the proof itself, given in sections 4 and 6. In
the proof, we apply the group theoretical theorems of Nakayama-Tate and of
Poitou, proven already in chapter III.
In chapter IX, we reap the rewards of our efforts in the previous chapters. We
prove several classical number theoretical results, such as the Hasse principle
and the Grunwald-Wang theorem. In §5, we consider embedding problems
and we present the theorem of K. IWASAWA to the effect that the maximal
prosolvable factor of the absolute Galois group of Qab is free. In §6, we give
a complete proof of Šafarevič’s theorem on the realization of finite solvable
groups as Galois groups over global fields.
The main concern of chapter X is to consider restricted ramification. Ge-
ometrically speaking, we are considering the curves Spec(Ok,S ), in contrast
to chapter IX, where our main interest was in the point Spec(k). Needless to
say, things now become much harder. Invariants like the S-ideal class group
or the p-adic regulator enter the game and establish new arithmetic obstruc-
tions. Our investigations are guided by the analogy between number fields
and function fields. We know a lot about the latter from algebraic geometry,
and we try to establish analogous results for number fields. For example,
using the group theoretical techniques of chapter IV, we can prove the number
theoretical analogue of Riemann’s existence theorem. The fundamental group
of Spec(Ok ), i.e. the Galois group of the maximal unramified extension of the
number field k, was the subject of the long-standing class field tower problem
in number theory, which was finally answered negatively by E. S. GOLOD and

Introduction xi
I. R. ŠAFAREVIČ. We present their proof, which demonstrates the power of the

group theoretical and cohomological methods, in §8.
Chapter XI deals with Iwasawa theory, which is the consequent conceptual
continuation of the analogy between number fields and function fields. We
concentrate on the algebraic aspects of Iwasawa theory of p-adic local fields
and of number fields, first presenting the classical statements which one can
usually find in the standard literature. Then we prove more far-reaching results
on the structure of certain Iwasawa modules attached to p-adic local fields and
to number fields, using the homotopy theory of Iwasawa modules. The analytic
aspects of Iwasawa theory will merely be described, since this topic is cov-
ered by several monographs, for example, the book [246] of L. WASHINGTON.
Finally, the Main Conjecture of Iwasawa theory will be formulated and dis-
cussed; for a proof, we refer the reader to the original work of B. MAZUR and
A. WILES ([134], [249]).
In the last chapter, we give a survey of so-called anabelian geometry, a
program initiated by A. GROTHENDIECK. Perhaps the first result of this theory,
obtained even before this program existed, is a theorem of J. NEUKIRCH and
K. UCHIDA which asserts that the absolute Galois group of a global field, as a
profinite group, characterizes the field up to isomorphism. We give a proof of
this theorem for number fields in the first two sections. The final section gives
an overview of the conjectures and their current status.
The reader will recognize very quickly that this book is not a basic textbook
in the sense that it is completely self-contained. We use freely basic algebraic,
topological and arithmetic facts which are commonly known and contained in
the standard textbooks. In particular, the reader should be familiar with basic
number theory. While assuming a certain minimal level of knowledge, we
have tried to be as complete and as self-contained as possible at the next stage.
We give full proofs of almost all of the main results, and we have tried not
to use references which are only available in original papers. This makes it
possible for the interested student to use this book as a textbook and to find
large parts of the theory coherently ordered and gently accessible in one place.
On the other hand, this book is intended for the working mathematician as a
reference on cohomology of local and global fields.
Finally, a remark on the exercises at the end of the sections. A few of them
are not so much exercises as additional remarks which did not fit well into the
main text. Most of them, however, are intended to be solved by the interested
reader. However, there might be occasional mistakes in the way they are posed.
If such a case arises, it is an additional task for the reader to give the correct
formulation.

xii Preface to the Second Edition
We would like to thank many friends and colleagues for their mathemat-
ical examination of parts of this book, and particularly, ANTON DEITMAR,
TORSTEN FIMMEL, DAN HARAN, UWE JANNSEN, HIROAKI NAKAMURA and
OTMAR VENJAKOB. We are indebted to Mrs. INGE MEIER who TEXed a large
part of the manuscript, and EVA-MARIA STROBEL receives our special gratitude
for her careful proofreading. Hearty thanks go to FRAZER JARVIS for going
through the entire manuscript, correcting our English.
Heidelberg, September 1999 Alexander Schmidt

Kay Wingberg
Preface to the Second Edition
The present second edition is a corrected and extended version of the first.
We have tried to improve the exposition and reorganize the content to some
extent; furthermore, we have included some new material. As an unfortunate
result, the numbering of the first edition is not compatible with the second.
In the algebraic part you will find new sections on filtered cochain com-
plexes, on the degeneration of spectral sequences and on Tate cohomology of
profinite groups. Amongst other topics, the arithmetic part contains a new
section on duality theorems for unramified and tamely ramified extensions, a
careful analysis of 2-extensions of real number fields and a complete proof of
Neukirch’s theorem on solvable Galois groups with given local conditions.
Since the publication of the first edition, many people have sent us lists
of corrections and suggestions or have contributed in other ways to this edi-
tion. We would like to thank them all. In particular, we would like to thank
JAKOB STIX and DENIS VOGEL for their comments on the new parts of this
second edition and FRAZER JARVIS, who again did a great job correcting our
English.
Regensburg and Heidelberg, November 2007 Alexander Schmidt

Kay Wingberg

Contents
Algebraic Theory 1
Chapter I: Cohomology of Profinite Groups 3

§1. Profinite Spaces and Profinite Groups . . . . . . . . . . . . . . 3
§2. Definition of the Cohomology Groups . . . . . . . . . . . . . . 12
§3. The Exact Cohomology Sequence . . . . . . . . . . . . . . . . 25
§4. The Cup-Product . . . . . . . . . . . . . . . . . . . . . . . . . 36
§5. Change of the Group G . . . . . . . . . . . . . . . . . . . . . . 45
§6. Basic Properties . . . . . . . . . . . . . . . . . . . . . . . . . 60
§7. Cohomology of Cyclic Groups . . . . . . . . . . . . . . . . . . 74
§8. Cohomological Triviality . . . . . . . . . . . . . . . . . . . . . 80
§9. Tate Cohomology of Profinite Groups . . . . . . . . . . . . . . 83
Chapter II: Some Homological Algebra 97

§1. Spectral Sequences . . . . . . . . . . . . . . . . . . . . . . . . 97
§2. Filtered Cochain Complexes . . . . . . . . . . . . . . . . . . . 101
§3. Degeneration of Spectral Sequences . . . . . . . . . . . . . . . 107
§4. The Hochschild-Serre Spectral Sequence . . . . . . . . . . . . . 111
§5. The Tate Spectral Sequence . . . . . . . . . . . . . . . . . . . . 120
§6. Derived Functors . . . . . . . . . . . . . . . . . . . . . . . . . 127
§7. Continuous Cochain Cohomology . . . . . . . . . . . . . . . . 136
Chapter III: Duality Properties of Profinite Groups 147

§1. Duality for Class Formations . . . . . . . . . . . . . . . . . . . 147
§2. An Alternative Description of the Reciprocity Homomorphism . 164
§3. Cohomological Dimension . . . . . . . . . . . . . . . . . . . . 171
§4. Dualizing Modules . . . . . . . . . . . . . . . . . . . . . . . . 181
§5. Projective pro-c-groups . . . . . . . . . . . . . . . . . . . . . . 189
§6. Profinite Groups of scd G = 2 . . . . . . . . . . . . . . . . . . 202
§7. Poincaré Groups . . . . . . . . . . . . . . . . . . . . . . . . . 210
§8. Filtrations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 220
§9. Generators and Relations . . . . . . . . . . . . . . . . . . . . . 224

xiv Contents
Chapter IV: Free Products of Profinite Groups 245

§1. Free Products . . . . . . . . . . . . . . . . . . . . . . . . . . . 245
§2. Subgroups of Free Products . . . . . . . . . . . . . . . . . . . . 252
§3. Generalized Free Products . . . . . . . . . . . . . . . . . . . . 256
Chapter V: Iwasawa Modules 267

§1. Modules up to Pseudo-Isomorphism . . . . . . . . . . . . . . . 268
§2. Complete Group Rings . . . . . . . . . . . . . . . . . . . . . . 273
§3. Iwasawa Modules . . . . . . . . . . . . . . . . . . . . . . . . . 289
§4. Homotopy of Modules . . . . . . . . . . . . . . . . . . . . . . 301
§5. Homotopy Invariants of Iwasawa Modules . . . . . . . . . . . . 312
§6. Differential Modules and Presentations . . . . . . . . . . . . . . 321
Arithmetic Theory 335
Chapter VI: Galois Cohomology 337

§1. Cohomology of the Additive Group . . . . . . . . . . . . . . . 337
§2. Hilbert’s Satz 90 . . . . . . . . . . . . . . . . . . . . . . . . . 343
§3. The Brauer Group . . . . . . . . . . . . . . . . . . . . . . . . 349
§4. The Milnor K-Groups . . . . . . . . . . . . . . . . . . . . . . 356
§5. Dimension of Fields . . . . . . . . . . . . . . . . . . . . . . . 360
Chapter VII: Cohomology of Local Fields 371

§1. Cohomology of the Multiplicative Group . . . . . . . . . . . . . 371
§2. The Local Duality Theorem . . . . . . . . . . . . . . . . . . . 378
§3. The Local Euler-Poincaré Characteristic . . . . . . . . . . . . . 391
§4. Galois Module Structure of the Multiplicative Group . . . . . . . 401
§5. Explicit Determination of Local Galois Groups . . . . . . . . . . 409
Chapter VIII: Cohomology of Global Fields 425

§1. Cohomology of the Idèle Class Group . . . . . . . . . . . . . . 425
§2. The Connected Component of Ck . . . . . . . . . . . . . . . . . 443
§3. Restricted Ramification . . . . . . . . . . . . . . . . . . . . . . 452
§4. The Global Duality Theorem . . . . . . . . . . . . . . . . . . . 466
§5. Local Cohomology of Global Galois Modules . . . . . . . . . . 472
§6. Poitou-Tate Duality . . . . . . . . . . . . . . . . . . . . . . . . 480
§7. The Global Euler-Poincaré Characteristic . . . . . . . . . . . . 503
§8. Duality for Unramified and Tamely Ramified Extensions . . . . . 513

Contents xv
Chapter IX: The Absolute Galois Group of a Global Field 521

§1. The Hasse Principle . . . . . . . . . . . . . . . . . . . . . . . . 522
§2. The Theorem of Grunwald-Wang . . . . . . . . . . . . . . . . . 536
§3. Construction of Cohomology Classes . . . . . . . . . . . . . . . 543
§4. Local Galois Groups in a Global Group . . . . . . . . . . . . . 553
§5. Solvable Groups as Galois Groups . . . . . . . . . . . . . . . . 557
§6. Šafarevič’s Theorem . . . . . . . . . . . . . . . . . . . . . . . 574
Chapter X: Restricted Ramification 599

§1. The Function Field Case . . . . . . . . . . . . . . . . . . . . . 602
§2. First Observations on the Number Field Case . . . . . . . . . . . 618
§3. Leopoldt’s Conjecture . . . . . . . . . . . . . . . . . . . . . . 624
§4. Cohomology of Large Number Fields . . . . . . . . . . . . . . 642
§5. Riemann’s Existence Theorem . . . . . . . . . . . . . . . . . . 647
§6. The Relation between 2 and ∞ . . . . . . . . . . . . . . . . . . 656
§7. Dimension of H i (GST , ZZ/pZZ) . . . . . . . . . . . . . . . . . . 666
§8. The Theorem of Kuz’min . . . . . . . . . . . . . . . . . . . . . 678
§9. Free Product Decomposition of GS (p) . . . . . . . . . . . . . . 686
§10. Class Field Towers . . . . . . . . . . . . . . . . . . . . . . . . 697
§11. The Profinite Group GS . . . . . . . . . . . . . . . . . . . . . . 706
Chapter XI: Iwasawa Theory of Number Fields 721

§1. The Maximal Abelian Unramified p-Extension of k∞ . . . . . . 722
§2. Iwasawa Theory for p-adic Local Fields . . . . . . . . . . . . . 731
§3. The Maximal Abelian p-Extension of k∞ Unramified Outside S . . 735
§4. Iwasawa Theory for Totally Real Fields and CM-Fields . . . . . 751
§5. Positively Ramified Extensions . . . . . . . . . . . . . . . . . . 763
§6. The Main Conjecture . . . . . . . . . . . . . . . . . . . . . . . 771
Chapter XII: Anabelian Geometry 785

§1. Subgroups of Gk . . . . . . . . . . . . . . . . . . . . . . . . . 785
§2. The Neukirch-Uchida Theorem . . . . . . . . . . . . . . . . . . 791
§3. Anabelian Conjectures . . . . . . . . . . . . . . . . . . . . . . 798
Literature 805
Index 821


Algebraic Theory


Chapter I
Cohomology of Profinite Groups
Profinite groups are topological groups which naturally occur in algebraic

number theory as Galois groups of infinite field extensions or more generally as
étale fundamental groups of schemes. Their cohomology groups often contain
important arithmetic information.
In the first chapter we will study profinite groups as objects of interest in
themselves, independently of arithmetic applications, which will be treated in
the second part of this book.
§1. Profinite Spaces and Profinite Groups
The underlying topological spaces of profinite groups are of a very specific

type, which will be described now. To do this, we make use of the concept
of inverse (or projective) limits. We refer the reader to the standard literature
(e.g. [160], [79], [139]) for the definition and basic properties of limits. All
index sets will be assumed to be filtered.
(1.1.1) Lemma. For a Hausdorff topological space T the following conditions

are equivalent.
(i) T is the (topological) inverse limit of finite discrete spaces.
(ii) T is compact and every point of T has a basis of neighbourhoods con-
sisting of subsets which are both closed and open.
(iii) T is compact and totally disconnected.
Proof: In order to show the implication (i) ⇒ (ii), we first recall that the
inverse limit of compact spaces is compact (see [15] chap.I, §9, no.6, prop.8).
Therefore T is compact. By the definition of the inverse limit topology and
by (i), every point of T has a basis of neighbourhoods consisting of sets of the

4 Chapter I. Cohomology of Profinite Groups
form f −1 (W ), where W is a subset of a finite discrete space V and f : T → V

is a continuous map. These sets are both open and closed.
For the implication (ii) ⇒ (iii) we have to show that the connected component
Ct of every point t ∈ T equals {t}. Since T is compact, Ct is the intersection
of all closed and open subsets containing t (see [15] chap.II, §4, no.4, prop.6).
Since T is Hausdorff, we obtain Ct = {t}.
It remains to show the implication (iii) ⇒ (i). Let I be the set of equivalence
relations R ⊆ T × T on T , such that the quotient space T /R is finite and
discrete in the quotient topology. The set I is partially ordered by inclusion
and is directed, because R1 ∩ R2 is in I if R1 and R2 are. We claim that the
canonical map φ : T → lim T /R is a homeomorphism.
←− R∈I
First we see that the map φ is surjective, because for an element {tR }R∈I ∈
lim
←− R∈I
T /R, the sets (pR ◦ φ)−1 (tR ) are nonempty and compact. Since I is
directed, finite intersections of these sets are also nonempty and compactness
then implies that φ−1 ({tR }R∈I ) = R∈I (pR ◦ φ)−1 (tR ) is nonempty.
T
For the injectivity it suffices to show that for t, s ∈ T , t =/ s, there exists an

R ∈ I such that (t, s) ∈/ R. But since s is not in the connected component of t,
there exists a closed and open subset U ⊆ T with t ∈ U , s ∈/ U (see [15] chap.II,
§4, no.4, prop.6). Then the equivalence relation R defined by "(x, y) ∈ R if
x and y are both in U or both not in U " is of the required type. The proof is
completed by the remark that a continuous bijection between compact spaces
is a homeomorphism. 2
In fact one immediately verifies that we could have chosen the inverse system
in (i) in such a way that all transition maps are surjective.
(1.1.2) Definition. A space T is called a profinite space if it satisfies the

equivalent conditions of lemma (1.1.1).
A compactness argument shows that a subset V ⊆ lim←−

Xi of a profinite space
is both closed and open if and only if V is the pre-image under the canonical
projection pi : X → Xi of a (necessarily closed and open) subset in Xi for
some i. Every continuous map between profinite spaces can be realized as
a projective limit of maps between finite discrete spaces. Without giving an
exact definition, we want to note that the category of profinite spaces with
continuous maps is the pro-category of the category of finite discrete spaces.
Recall that a topological group is a group G endowed with the structure of

a topological space, such that the group operations G → G, g 7→ g −1 , and

§1. Profinite Spaces and Profinite Groups 5
G × G → G, (g, h) 7→ gh, are continuous. The reader will immediately

verify that the inverse limit of an inverse system of topological groups is just
the inverse limit of the groups together with the inverse limit topology on the
underlying topological space.
(1.1.3) Proposition. For a Hausdorff topological group G the following con-

ditions are equivalent.
(i) G is the (topological) inverse limit of finite discrete groups.
(ii) G is compact and the unit element has a basis of neighbourhoods consist-
ing of open and closed normal subgroups.
(iii) G is compact and totally disconnected.
Proof: (i) ⇒ (iii): The inverse limit of compact and totally disconnected
spaces is compact and totally disconnected.
(ii) ⇒ (i): Assume that U runs through a system of neighbourhoods of the unit
element e ∈ G, which consists of open normal subgroups. Then the canonical
homomorphism φ : G → lim ←−
G/U is an isomorphism:
U
To begin with, φ is injective, because G is Hausdorff. In order to show the
surjectivity, let x = {xU }U ∈ lim G/U . Denoting the canonical projection
←− U
by φU : G → G/U , we have the equality
φ−1 (x) = φ−1
\
U (xU ).
U
The intersection on the right side is taken over nonempty compact spaces and
finite intersections of these are nonempty. Hence φ−1 (x) is nonempty, and
therefore φ is surjective. Furthermore, φ is open, hence a homeomorphism.
Finally, for every such U , the group G/U is discrete and compact, hence finite.
(iii) ⇒ (ii): By (1.1.1), the underlying topological space of G is profinite,
hence every point has a basis of neighbourhoods consisting of open and closed
subsets. Note that an open subgroup is automatically closed, because it is the
complement of the union of its (open) nontrivial cosets. Let U be an arbitrary
chosen, closed and open neighbourhood of the unit element e ∈ G. Set
V := {v ∈ U | Uv ⊆ U }, H := {h ∈ V | h−1 ∈ V }.
We claim that H ⊆ U is an open (and closed) subgroup in G. We first show
that V is open. Fix a point v ∈ V . Then uv ∈ U for every u ∈ U and therefore
there exist neighbourhoods Uu of u and Vu of v, such that Uu Vu ⊆ U . The open
sets Uu cover the compact space U and therefore there exists a finite subcover,
Uu1 , . . . , Uun , say. Then
Vv := Vu1 ∩ · · · ∩ Vun

is an open neighbourhood of v contained in V . Hence V is open and also

H := V ∩ V −1 , since the inversion map is a homeomorphism. It remains to
show that H is a subgroup. Trivially e ∈ H and H −1 = H by construction.
We now check that xy ∈ H if x, y ∈ H. First we have U xy ⊆ U y ⊆ U , and so
xy ∈ V . In the same way we obtain y −1 x−1 ∈ V , hence xy ∈ H. This proves
that H is an open subgroup of G contained in U . In particular, H has finite
index in G and there are only finitely many different conjugates of H. The
intersection of these finitely many conjugates is an open, closed and normal
subgroup of G contained in U . 2
(1.1.4) Definition. A Hausdorff topological group G satisfying the equivalent

conditions of (1.1.3) is called a profinite group.
Without further mention, homomorphisms between profinite groups are al-

ways assumed to be continuous and subgroups are assumed to be closed. Since
a subgroup is the complement of its nontrivial cosets and by the compactness
of G, we see that open subgroups are closed and a closed subgroup is open if
and only if it has finite index. If H is a (closed) subgroup of the profinite group
G, then the set G/H of coset classes with the quotient topology is a profinite
space. If H is normal, then the quotient G/H is a profinite group in a natural
way.
In principle, all objects and statements of the theory of finite groups have
their topological analogue in the theory of profinite groups. For example, the
profinite analogues of the Sylow theorems are true (see §6). We make the
following
(1.1.5) Definition. A supernatural number is a formal product

pnp ,
Y
p
where p runs through all prime numbers and, for each p, the exponent np is a
non-negative integer or the symbol ∞.
Using the unique decomposition into prime powers, we can view any natural
number as a supernatural number. We multiply supernatural numbers (even
infinitely many of them) by adding the exponents. By convention, the sum of
the exponents is ∞ if infinitely many summands are non-zero or if one of the
summands is ∞. We also have the notions of l.c.m. and g.c.d. of an arbitrary
family of supernatural numbers. In particular, any family of natural numbers
has an l.c.m., which is, in general, a supernatural number.

(1.1.6) Definition. Let G be a profinite group and let A be an abelian torsion

group.
(i) The index of a closed subgroup H in G is the supernatural number
(G : H) = l.c.m.(G/U : H/H ∩ U ),
where U ranges over all open normal subgroups of G.
(ii) The order of G is defined by
#G = (G : 1) = l.c.m. #(G/U ).
U
(iii) The order of A is defined by

#A = l.c.m. #B ,
where B ranges over all finite subgroups of A.
Given closed subgroups N ⊆ H ⊆ G, we have

(G : N ) = (G : H)(H : N ).
Furthermore, the order #A of an abelian torsion group A is just the order of the
profinite group Hom(A, Q/ZZ).
(1.1.7) Definition. Let G be a profinite group. An abstract G-module M is

an abelian group M together with an action
G × M → M, (g, m) 7→ g(m)
such that 1(m) = m, (gh)(m) = g(h(m)) and g(m + n) = g(m) + g(n) for all
g, h ∈ G, m, n ∈ M .
A topological G-module M is an abelian Hausdorff topological group M
which is endowed with the structure of an abstract G-module such that the
action G × M → M is continuous.
For a closed subgroup H ⊆ G we denote the subgroup of H-invariant

elements in M by M H , i.e.
M H = {m ∈ M | h(m) = m for all h ∈ H}.
(1.1.8) Proposition. Let G be a profinite group and let M be an abstract

G-module. Then the following conditions are equivalent:
(i) M is a discrete G-module, i.e. the action G × M → M is continuous for
the discrete topology on M .
(ii) For every m ∈ M the subgroup Gm := {g ∈ G | g(m) = m} is open.
M U , where U runs through the open subgroups of G.
S
(iii) M =

Proof: If we restrict the map G × M → M to G × {m}, then m ∈ M has

pre-image Gm × {m}. This shows (i)⇒(ii). The assertion (ii)⇒(iii) is trivial
because m ∈ M Gm . Finally, assume that (iii) holds. Let (g, m) ∈ G×M . There
exists an open subgroup U such that m ∈ M U . Therefore gU × {m} is an open
neighbourhood of (g, m) ∈ G × M mapping to g(m). This shows (i). 2
In this book we are mainly concerned with discrete modules and so the term
G-module, without the word “topological” or “abstract”, will always mean a
discrete module.
L
If (Ai )i∈I is a family of discrete G-modules, then their direct sum i∈I Ai ,
endowed with the componentwise G-action g((ai )i∈I ) = (g(ai ))i∈I , is again a
discrete G-module, but this is not necessarily true for the product. The tensor
product
A ⊗ B = A ⊗ ZZ B
of two discrete modules endowed with the diagonal action g(a⊗b) = g(a)⊗g(b)
is a discrete module. The set Hom(A, B) = Hom ZZ (A, B) becomes an abstract
G-module by setting g(φ)(a) = g(φ(g −1 (a))). Its subgroup of invariants
HomG (A, B) = Hom(A, B)G
is the set of G-homomorphisms from A to B. Hom(A, B) is a discrete G-
module if G is finite or if A is finitely generated as a ZZ-module.
The groups ZZ, Q, ZZ/nZZ, IFq are always viewed as trivial discrete G-modules,
i.e. G-modules with trivial action of G.
So far we have considered totally disconnected compact groups. If A is any

topological group, then the connected component A0 (of the identity) of A is
a closed subgroup. We have the following general facts for which we refer to
[170] sec. 22, and [15] chap. III, §4.6.
(1.1.9) Proposition. Let A be a locally compact group. Then

(i) A0 is the intersection of all open normal subgroups of A, and A/A0 is the
largest totally disconnected quotient.
(ii) A0 is generated by every open neighbourhood of 1 in A0 .
(iii) If A → B is a continuous surjective homomorphism onto the locally
compact group B, then the closure of the image of A0 is the connected
component of 1 in B.

An essential tool for working with locally compact abelian groups is a duality
theorem due to L. S. PONTRYAGIN. We consider the group IR/ZZ as a topological
group with the quotient topology inherited from IR.
(1.1.10) Definition. Let A be a Hausdorff, abelian and locally compact topo-

logical group. We call the group
A∨ := Homcts (A, IR/ZZ)
the Pontryagin dual of A.
Given locally compact topological spaces X, Y , the set of continuous maps

Mapcts (X, Y ) carries a natural topology, the compact-open topology. A sub-
basis of this topology is given by the sets
UK,U = {f ∈ Mapcts (X, Y ) | f (K) ⊆ U },
where K runs through the compact subsets of X and U runs through the open
subsets of Y . For the proof of the following theorem we refer to [170], th. 5.3
or [146], th. 23, [186], th. 1.7.2.
(1.1.11) Theorem (Pontryagin Duality). If A is a Hausdorff abelian locally

compact topological group, then the same is true for A∨ endowed with the
compact-open topology. The canonical homomorphism
A −→ (A∨ )∨ ,
given by a 7−→ τa : A∨ → IR/ZZ, φ 7→ φ(a), is an isomorphism of topological

groups. Thus ∨ defines an involutory contravariant autofunctor on the cate-
gory of Hausdorff abelian locally compact topological groups which moreover
commutes with limits. Furthermore, ∨ induces equivalences of categories
∨
(abelian compact groups) ⇐⇒ (discrete abelian groups)
∨
(abelian profinite groups) ⇐⇒ (discrete abelian torsion groups).
For an (abstract) abelian group A we use the notation
A∗ = Hom(A, Q/ZZ).
Clearly, if A is a discrete torsion group, then A∨ ∼

= A∗ and we will frequently
∗ ∨
also write A instead of A , at least if we are not interested in the topology
of the dual. If A is abelian and profinite, then it is an easy exercise to see

that every continuous homomorphism φ : A → IR/ZZ has finite image. If,

moreover, A is topologically finitely generated, then every subgroup of finite
index is open in A, and hence also in this case A∨ ∼
= A∗ .
Now assume that we are given a family (Xi )i∈I of Hausdorff, abelian topo-
logical groups and let an open subgroup Yi ⊆ Xi be given for almost all i ∈ I
(i.e. for all but finitely many indices). For consistency of notation, we put
Yi = Xi for the remaining indices.
(1.1.12) Definition. The restricted product

Y
(Xi , Yi )
i∈I
Y
is the subgroup of Xi consisting of all (xi )i∈I such that xi ∈ Yi for almost all i.
The restricted product is a topological group, and a basis of neighbourhoods
of the identity is given by the products
Y Y
Uj × Yi ,
j ∈J i∈I\J
where J runs over the finite subsets of I and Uj runs over a basis of neigh-
bourhoods of the identity of Xj .
Basic examples of restricted products are the product of groups (Yi = Xi for
all i) and the direct sum of discrete groups (Yi = 0 for all i). We will write
Y
Xi
i∈ I
for short if it is clear from the context what the Yi are. The restricted product
is again a Hausdorff, abelian topological group.
(1.1.13) Proposition. If all Xi are locally compact and almost all Yi are
compact, then the restricted product is again an abelian locally compact group.
For the Pontryagin dual of the restricted product, there is a canonical iso-
morphism
(Xi , Yi ) )∨ ∼ (Xi∨ , (Xi /Yi )∨ ).

Y Y
( =
i∈I i∈ I
Proof: The product of compact topological spaces is compact, therefore the

restricted product is locally compact under the given conditions. Furthermore,
sinceYYi is compact and open in Xi , the same is true for (Xi /Yi )∨ in Xi∨ . Hence
also i∈I (Xi∨ , (Xi /Yi )∨ ) exists and is locally compact.

A sufficiently small open neighbourhood of 0 ∈ IR/ZZ containsY no nontrivial

subgroups. Therefore a continuous homomorphism φ : (Xi , Yi ) → IR/ZZ
annihilates Yi for almost all i. In other words, the restriction of φ to Xi lies in
∨ ⊆ ∨
Ysubgroup (Xi /Yi ) Y Xi for almost all i. This yields a bijection between
the
( i∈I (Xi , Yi ) ) and i∈I (Xi∨ , (Xi /Yi )∨ ), which easily can be seen to be a
∨
homeomorphism. 2
Exercise 1. Show that an injective (resp. surjective) continuous map between profinite spaces
may be represented as an inverse limit over a system of injective (resp. surjective) maps
between finite discrete spaces.
Exercise 2. Let X be a profinite space and let X0 ⊆ X be a closed subspace. Show that
every continuous map f : X0 → Y from X0 to a finite discrete space Y has a continuous
extension F : X → Y (i.e. F |X0 = f ) and that any two such extensions coincide on an open
neighbourhood of X0 in X.
Exercise 3. Let G, H be profinite groups. Show that
Hom(G, H) = lim lim Hom(G/U, H/V ),
←− −→
V ⊆H U ⊆G
where the limits are taken over all open normal subgroups V of H and U of G.
Exercise 4. If K ⊆ H are closed subgroups of the profinite group G, then the projection
π : G/K → G/H has a continuous section s : G/H → G/K.
Hint: Let X be the set of pairs (S, s), where S is a closed subgroup such that K ⊆ S ⊆ H and
s is a continuous section s : G/H → G/S. Write (S, s) ≤ (S 0 , s0 ) if S 0 ⊆ S and if s is the
composite of s0 and the projection G/S 0 → G/S. Then X is inductively ordered. By Zorn’s
lemma, there exists a maximal element (S, s) of X. Show that S = K.
Exercise 5. A morphism φ : X → Y in a category C is called a monomorphism if for every
object Z of C and for every pair of morphisms f, g : Z → X the implication "φ ◦ f = φ ◦ g
⇒ f = g" is true. The morphism φ is called an epimorphism if it is a monomorphism in the
opposite category C op (the category obtained from C by reversing all arrows).
(i) Show that the monomorphisms in the category of profinite groups are the injective homo-
morphisms.
(ii) Show that the epimorphisms in the category of profinite groups are the surjective homo-
morphisms.
Hint for (ii): First reduce the problem to the case of finite groups. Assume that there is an
epimorphism φ : G → H of finite groups which is not surjective. Assume that (H : φ(G)) ≥ 3
(otherwise φ(G) is normal in H) and choose two elements a, b ∈ H having different nontrivial
residue classes modulo φ(G). Let S be the (finite) group of set theoretic automorphisms of H.
Let s ∈ S be the map H → H which interchanges the cosets aφ(G) and bφ(G) and which is
the identity on the other left cosets modulo φ(G). Then consider the maps f and g defined by
f (h1 )(h2 ) = h2 h−1 −1
1 and by g(h) = s f (h)s.

§2. Definition of the Cohomology Groups
The cohomology of a profinite group G arises from the diagram

· · · −→
−→
−→ G × G × G −→
−→ −→ G × G −→
−→ −→ G,
the arrows being the projections
di : Gn+1 −→ Gn , i = 0, 1, . . . , n,
given by
di (σ0 , . . . , σn ) = (σ0 , . . . , σ̂i , . . . , σn ),
where by σ̂i we indicate that we have omitted σi from the (n + 1)-tuple
(σ0 , . . . , σn ). G acts on Gn by left multiplication.
From now on, we assume all G-modules to be discrete. For every G-module
A we form the abelian group
X n = X n (G, A) = Map (Gn+1 , A)
of all continuous maps x : Gn+1 −→ A, i.e. of all continuous functions
x(σ0 , . . . , σn ) with values in A. X n is in a natural way a G-module by
(σx)(σ0 , . . . , σn ) = σx(σ −1 σ0 , . . . , σ −1 σn ).
The maps di : Gn+1 −→ Gn induce G-homomorphisms d∗i : X n−1 −→ X n and
we form the alternating sum
n
∂n = (−1)i d∗i : X n−1 −→ X n .
X
i=0
We usually write ∂ in place of ∂ n . Thus for x ∈ X n−1 , ∂x is the function

n
(−1)i x(σ0 , . . . , σ̂i , . . . , σn ).
X
(∗) (∂x)(σ0 , . . . , σn ) =
i=0
Moreover, we have the G-homomorphism ∂ 0 : A → X 0 , which associates to

a ∈ A the constant function x(σ0 ) = a.
(1.2.1) Proposition. The sequence

∂0 ∂1 ∂2
0 −→ A −→ X 0 −→ X 1 −→ X 2 −→ . . .
is exact.
Proof: We first show that it is a complex, i.e. ∂∂ = 0. ∂ 1 ◦ ∂ 0 = 0 is

clear. Let x ∈ X n−1 . Applying ∂ to (∗), we obtain summands of the form
x(σ0 , . . . , σ̂i , . . . , σ̂j , . . . , σn ) with certain signs. Each of these summands
arises twice, once where first σj and then σi is omitted, and again where first

§2. Definition of the Cohomology Groups 13
σi and then σj is omitted. The first time the sign is (−1)i (−1)j and the second
time (−1)i (−1)j−1 . Hence the summands cancel to give zero.
For the exactness, we consider the map D−1 : X 0 → A, D−1 x = x(1), and
for n ≥ 0 the maps
Dn : X n+1 −→ X n , (Dn x)(σ0 , . . . , σn ) = x(1, σ0 , . . . , σn ).
These are homomorphisms of ZZ-modules, not of G-modules. An easy calcu-
lation shows that for n ≥ 0
(∗) Dn ◦ ∂ n+1 + ∂ n ◦ Dn−1 = id.
If x ∈ ker(∂ n+1 ) then x = ∂ n (Dn−1 x), i.e. ker(∂ n+1 ) ⊆ im(∂ n ) and thus
ker(∂ n+1 ) = im(∂ n ) because ∂ n+1 ◦ ∂ n = 0. 2
An exact sequence of G-modules 0 → A → X 0 → X 1 → X 2 → . . . is

called a resolution of A and a family (Dn )n≥−1 as in the proof with the property
(∗) is called a contracting homotopy of it. The above resolution is called the
standard resolution.
We now apply the functor “fixed module”. We set for n ≥ 0
C n (G, A) = X n (G, A)G .
C n (G, A) consists of the continuous functions x : Gn+1 → A such that
x(σσ0 , . . . , σσn ) = σx(σ0 , . . . , σn )
for all σ ∈ G. These functions are called the (homogeneous) n-cochains of
G with coefficients in A. From the standard resolution (1.2.1) we obtain a
sequence
∂1 ∂2
C 0 (G, A) −→ C 1 (G, A) −→ C 2 (G, A) −→ . . . ,
which in general is no longer exact. But it is still a complex, i.e. ∂∂ = 0, and
is called the homogeneous cochain complex of G with coefficients in A.
We now set
∂ n+1
Z n (G, A) = ker (C n (G, A) −→ C n+1 (G, A)),
∂n
B n (G, A) = im (C n−1 (G, A) −→ C n (G, A))
and B 0 (G, A) = 0. The elements of Z n (G, A) and B n (G, A) are called the
(homogeneous) n-cocycles and n-coboundaries respectively. As ∂∂ = 0, we
have B n (G, A) ⊆ Z n (G, A).
(1.2.2) Definition. For n ≥ 0 the factor group

H n (G, A) = Z n (G, A)/B n (G, A)
is called the n-dimensional cohomology group of G with coefficients in A.

For computational purposes, and for many applications, it is convenient to

pass to a modified definition of the cohomology groups, which reduces the
number of variables in the homogeneous cochains x(σ0 , . . . , σn ) by one. Let
C 0 (G, A) = A and C n (G, A), n ≥ 1, be the abelian group of all continuous
functions y : Gn −→ A. We then have the isomorphism
C 0 (G, A) −→ C 0 (G, A), x(σ) 7−→ x(1),
and for n ≥ 1 the isomorphism
C n (G, A) −→ C n (G, A),
x(σ0 , . . . , σn ) 7→ y(σ1 , . . . , σn ) = x(1, σ1 , σ1 σ2 , . . . , σ1 · · · σn ),
whose inverse is given by
y(σ1 , . . . , σn ) 7→ x(σ0 , . . . , σn ) = σ0 y(σ0−1 σ1 , σ1−1 σ2 , . . . , σn−1
−1
σn ).
With these isomorphisms the coboundary operators ∂ n+1 : C n (G, A) −→
C n+1 (G, A) are transformed into the homomorphisms
∂ n+1 : C n (G, A) −→ C n+1 (G, A)
given by
(∂ 1 a)(σ) = σa − a for a ∈ A = C 0 (G, A),
(∂ 2 y)(σ, τ ) = σy(τ ) − y(στ ) + y(σ) for y ∈ C 1 (G, A),
(∂ n+1 y)(σ1 , . . . , σn+1 ) = σ1 y(σ2 , . . . , σn+1 )
n
X
+ (−1)i y(σ1 , . . . , σi−1 , σi σi+1 , σi+2 , . . . , σn+1 )
i=1
+(−1)n+1 y(σ1 , . . . , σn ) for y ∈ C n (G, A).

Setting
∂ n+1
Z n (G, A) = ker (C n (G, A) −→ C n+1 (G, A))
∂n
B n (G, A) = im (C n−1 (G, A) −→ C n (G, A)) ,
∼ C n (G, A) induce isomorphisms
the isomorphisms C n (G, A) −→
H n (G, A) ∼
= Z n (G, A)/B n (G, A).
The functions in C n (G, A), Z n (G, A), B n (G, A) are called the inhomo-
geneous n-cochains, n-cocycles and n-coboundaries. The inhomogeneous
coboundary operators ∂ n+1 are more complicated than the homogeneous ones,
but they have the advantage of dealing with only n variables instead of n + 1.
For n = 0, 1, 2 the groups H n (G, A) admit the following interpretations.

The group H 0 (G, A): We have a natural isomorphism C 0 (G, A) → A,

x 7→ x(1), by which we identify C 0 (G, A) with A. Then, for a ∈ A,
(∂ 1 a)(σ0 , σ1 ) = σ1 a − σ0 a, or (∂ 1 a)(σ) = σa − a in the inhomogeneous setting,
so that
H 0 (G, A) = AG .
The group H 1 (G, A): The inhomogeneous 1-cocycles are the continuous
functions x : G −→ A such that
x(στ ) = x(σ) + σx(τ ) for all σ, τ ∈ G.
They are also called crossed homomorphisms. The inhomogeneous
1-coboundaries are the functions
x(σ) = σa − a
with a fixed a ∈ A. If G acts trivially on A, then
H 1 (G, A) = Homcts (G, A). ∗)
The group H 1 (G, A) occurs in a natural way if we pass from an exact sequence
i j
0 −→ A −→ B −→ C −→ 0
of G-modules to the sequence of fixed modules. Then we lose the exactness
and are left only with the exactness of the sequence
0 −→ AG −→ B G −→ C G .
The group H 1 (G, A) now gives information about the deviation from exactness.
In fact we have a canonical homomorphism
δ : C G −→ H 1 (G, A)
extending the above exact sequence to a longer one. Namely, for c ∈ C G we
may choose an element b ∈ B such that jb = c. For each σ ∈ G there is an
aσ ∈ A such that iaσ = σb − b. The function σ 7→ aσ is a 1-cocycle and
we define δc to be the cohomology class of this 1-cocycle in H 1 (G, A). The
definition is easily seen to be independent of the choice of the element b. If
δc = 0, then aσ = i−1 (σb − b) = σa − a, a ∈ A, so that b0 = b − ia is an element
of B G with jb0 = c. This shows the exactness of the sequence
δ
0 −→ AG −→ B G −→ C G −→ H 1 (G, A).
We shall meet this again in a larger frame in §3.
The group H 1 (G, A) admits a concrete interpretation using the concept of
torsors. Since this concept may be more fully exploited in the framework of
non-abelian groups A, we generalize H 1 (G, A) as follows.
∗) Since G is automatically understood as a topological group, we usually write Hom(G, A)
instead of Homcts (G, A).

A G-group A is a not necessarily abelian group with the discrete topology

on which G acts continuously. We denote the action of σ ∈ G on a ∈ A by
σ
a, so that σ(ab) = σa σb. A cocycle of G with coefficients in A is a continuous
function σ 7→ aσ on G with values in the group A such that
aστ = aσ σaτ .
The set of cocycles is denoted by Z 1 (G, A). Two cocycles a, a0 are said to
be cohomologous if there exists a b ∈ A such that a0σ = b−1 aσ σb. This is an
equivalence relation in Z 1 (G, A) and the quotient set is denoted by H 1 (G, A).
It has a distinguished element given by the cocycle aσ = 1.
A G-set is a discrete topological space X with a continuous action of G. Let
A be a G-group. An A-torsor is a G-set X with a simply transitive right action
X × A −→ X, (x, a) 7→ xa, of A which is compatible with the G-action on
X. This means that for every pair x, y ∈ X there is a unique a ∈ A such that
y = xa, and σ(xa) = σxσa. For example, if
j
1 −→ A −→ B −→ C −→ 1
is an exact sequence of G-groups, then the cosets j −1 (c) for c ∈ C G are typical
A-torsors. It is clear what we mean by an isomorphism of A-torsors. Let
now TORS (A) denote the set of isomorphism classes of A-torsors. It has a
distinguished element given by the A-torsor A, and is thus a pointed set.
(1.2.3) Proposition. We have a canonical bijection of pointed sets

H 1 (G, A) ∼
= TORS (A).
Proof: We define a map

λ : TORS (A) −→ H 1 (G, A)
as follows. Let X be an A-torsor and let x ∈ X. For every σ ∈ G there is a
unique aσ ∈ A such that σx = xaσ . One verifies at once that aσ is a cocycle.
Changing x to xb changes this cocycle to b−1 aσ σb, which is cohomologous.
We define λ(X) to be the class of aσ .
We define an inverse µ : H 1 (G, A) −→ TORS (A) as follows. Let the set X
be the group A. We let G act on X in the twisted form
σ0
x = aσ · σx.
The action of A on X is given by right multiplication. In this way, X becomes
an A-torsor and this defines the map µ. Replacing aσ by b−1 aσ σb, we have
an isomorphism x 7→ b−1 x of A-torsors. One now checks that λ ◦ µ = 1 and
µ ◦ λ = 1. 2

i j
Remark: If 0 −→ A −→ B −→ C −→ 0 is an exact sequence of
G-modules and if we identify in the exact sequence
δ
0 −→ AG −→ B G −→ C G −→ H 1 (G, A)
H 1 (G, A) with TORS (A), then the map δ is given by δc = j −1 (c).
The group H 2 (G, A): We return to the case that A is abelian. The inhomo-
geneous 2-cocycles are the continuous functions x : G × G −→ A such that
∂x = 0, i.e.
x(στ, ρ) + x(σ, τ ) = x(σ, τ ρ) + σx(τ, ρ).
Among these we find the inhomogeneous 2-coboundaries as the functions
x(σ, τ ) = y(σ) − y(στ ) + σy(τ )
with an arbitrary 1-cochain y : G −→ A.
The 2-cocycles had been known before the development of group cohomo-
logy as factor systems and occurred in connection with group extensions. To
explain this, we assume that either A or G is finite, in order to avoid topological
problems (but see (2.7.7)).
The question is: how many groups Ĝ are there, which have the G-module A
as a normal subgroup and G as the factor group (we write A multiplicatively).
To be more precise, we consider all exact sequences
1 −→ A −→ Ĝ −→ G −→ 1
of topological groups (i.e. of profinite groups if A is finite, and of discrete
groups if G is finite), such that the action of G on A is given by
a = σ̂aσ̂ −1 ,
σ
where σ̂ ∈ Ĝ is a pre-image of σ ∈ G. If
1
A Ĝ0 G 1
f
1 A Ĝ G 1
is a commutative diagram of such sequences with a topological isomorphism f ,
then we call these sequences equivalent, and we denote the set of equivalence
classes [Ĝ] by EXT(G, A). This set has a distinguished element given by the
semi-direct product Ĝ = A o G (see ex.1 below).
(1.2.4) Theorem (SCHREIER). We have a canonical bijection of pointed sets

H 2 (G, A) ∼
= EXT(G, A).

Proof: We define a map

λ : EXT(G, A) −→ H 2 (G, A)
as follows. Let the class [Ĝ] ∈ EXT(G, A) be represented by the exact sequence
1 −→ A −→ Ĝ −→ G −→ 1.
We choose a continuous section s : G −→ Ĝ of Ĝ −→ G, and we set σ̂ = s(σ).
Such a section exists (see §1, ex.4). Regarding A as a subgroup of Ĝ, every
γ̂ ∈ Ĝ has a unique representation
γ̂ = aσ̂, a ∈ A, σ ∈ G,
and we have
σ̂a = σ̂aσ̂ −1 σ̂ = σaσ̂.
The elements σ̂ τ̂ and στ
c are both mapped onto στ , i.e.
σ̂ τ̂ = x(σ, τ )στ
c,
with an element x(σ, τ ) ∈ A such that x(σ, 1) = x(1, σ) = 1. Since σ̂ is a

continuous function of σ and A is closed in Ĝ, x(σ, τ ) is a continuous map
x : G × G −→ A. The associativity (σ̂ τ̂ )ρ̂ = σ̂(τ̂ ρ̂) yields that x(σ, τ ) is a
2-cocycle:
(σ̂ τ̂ )ρ̂ = x(σ, τ )στ
c ρ̂ = x(σ, τ )x(στ, ρ)(στ ρ)ˆ,
σ̂(τ̂ ρ̂) = σ̂x(τ, ρ)τcρ = σx(τ, ρ)σ̂ τcρ = σx(τ, ρ) x(σ, τ ρ)(στ ρ)ˆ,
i.e.
x(σ, τ )x(στ, ρ) = σx(τ, ρ) x(σ, τ ρ).
We thus get a cohomology class c = [x(σ, τ )] ∈ H 2 (G, A). This class does not
depend on the choice of the continuous section s : G −→ Ĝ. If s0 : G −→ Ĝ
is another one, and if we set σ̃ = s0 (σ), then σ̃ = y(σ)σ̂, y(σ) ∈ A, and
σ̃ τ̃ = x̃(σ, τ )στ
f . For the 2-cocycle x̃(σ, τ ) we obtain
c = x̃(σ, τ )y(στ )x(σ, τ )−1 σ̂ τ̂
σ̃ τ̃ = x̃(σ, τ )y(στ )στ
= x̃(σ, τ )x(σ, τ )−1 y(στ )y(σ)−1 σ̃y(τ )−1 τ̃
= x̃(σ, τ )x(σ, τ )−1 y(στ )y(σ)−1 σy(τ )−1 σ̃ τ̃ ,
i.e. x̃(σ, τ ) = x(σ, τ )y(σ, τ ) with the 2-coboundary
y(σ, τ ) = y(σ)y(στ )−1 σy(τ ).
The cohomology class c = [x(σ, τ )] also does not depend on the choice of the
representative 1 −→ A −→ Ĝ −→ G −→ 1 in the class [Ĝ]. Namely, if
1 A Ĝ G 1
f
1 A Ĝ0 G 1

is a commutative diagram and σ̂ 0 = f (σ̂), then

σ̂ 0 τ̂ 0 = f (σ̂)f (τ̂ ) = f (σ̂ τ̂ ) = f (x(σ, τ )στ c )0 ,
c ) = x(σ, τ )(στ
i.e. the group extensions Ĝ0 and Ĝ yield the same 2-cocycle x(σ, τ ). We thus
get a well-defined map
λ : EXT(G, A) −→ H 2 (G, A).
In order to prove the bijectivity, we construct an inverse µ : H 2 (G, A) −→
EXT(G, A). Every cohomology class c ∈ H 2 (G, A) contains a normalized
2-cocycle x(σ, τ ), i.e. a cocycle such that
x(σ, 1) = x(1, σ) = 1.
Namely, if x(σ, τ ) is any 2-cocycle in c, then we obtain from the equality
x(στ, ρ)x(σ, τ ) = x(σ, τ ρ) σx(τ, ρ) that
x(σ, 1) = σx(1, 1), x(1, ρ) = x(1, 1).
Setting y(σ) = x(1, 1) for all σ ∈ G, we obtain a 2-coboundary
y(σ, τ ) = y(σ)y(στ )−1 σy(τ ) ,
and the 2-cocycle x0 (σ, τ ) = x(σ, τ )y(σ, τ )−1 has the property that
x0 (σ, 1) = x(σ, 1)(σx(1, 1))−1 = 1, x0 (1, τ ) = x(1, τ )x(1, 1)−1 = 1.
Let now x(σ, τ ) be a normalized 2-cocycle in c. On the set Ĝ = A × G with
the product topology we define the continuous multiplication
(a, σ)(b, τ ) = (x(σ, τ )a σb, στ ).
This product is associative because of the cocycle property:
((a, σ)(b, τ ))(c, ρ) = (x(σ, τ )a σb, στ )(c, ρ)
= (x(στ, ρ)x(σ, τ )a σb στc, στ ρ) = (x(σ, τ ρ) σx(τ, ρ)a σb στc, στ ρ)
= (a, σ)(x(τ, ρ)b τc, τ ρ) = (a, σ)((b, τ ), (c, ρ)).
(1,1) is an identity element:
(a, σ)(1, 1) = (x(σ, 1)a, σ) = (a, σ) = (x(1, σ)a, σ) = (1, 1)(a, σ)
σ −1 −1
and ([ x(σ, σ −1 ) σ a]−1 , σ −1 ) is an inverse of (a, σ) since
−1 −1 −1
(a, σ)([σ x(σ, σ −1 ) σ a]−1 , σ −1 ) = (a σ(σ a)−1 , σσ −1 ) = (1, 1).
In this way Ĝ = A × G becomes a group with the product topology, and the
maps a 7→ (a, 1) and (a, σ) 7→ σ yield an exact sequence
1 −→ A −→ Ĝ −→ G −→ 1.
−1
Setting σ̂ = (1, σ), we have σ̂ −1 = (σ x(σ, σ −1 )−1 , σ −1 ) and
−1
σ̂(a, 1)σ̂ −1 = (x(σ, 1) σa, σ)(σ x(σ, σ −1 )−1 , σ −1 ) = (σa, 1).
We thus obtain an element [Ĝ] in EXT(G, A). This element does not depend
on the choice of the normalized 2-cocycle x(σ, τ ) in c. For, if x0 (σ, τ ) =
x(σ, τ )y(σ, τ )−1 is another one, y(σ, τ ) = y(σ)y(στ )−1 σy(τ ) is a 2-coboundary,

and if Ĝ0 is the group given by the multiplication on A×G via x0 (σ, τ ), then the
map f : (a, σ) 7→ (y(σ)a, σ) is an isomorphism from Ĝ to Ĝ0 and the diagram
1 A Ĝ G 1
f
1 A Ĝ0 G 1
is commutative, noting that y(1) = 1 because 1 = x0 (1, σ) = x(1, σ)y(1)−1
= y(1)−1 . Therefore [Ĝ] = [Ĝ0 ], and we get a well-defined map
µ : H 2 (G, A) −→ EXT(G, A).
This map is inverse to the map λ constructed before. For, if x(σ, τ ) is the
2-cocycle produced by a section G −→ Ĝ, σ 7→ σ̂, of a group extension
1 −→ A −→ Ĝ −→ G −→ 1,
then the map f : (a, σ) 7→ aσ̂ is an isomorphism of the group A × G, endowed
with the multiplication given by x(σ, τ ), onto Ĝ. This proves the theorem.
2
It is a significant feature of cohomology theory that we don’t have concrete

interpretations of the groups H n (G, A) for dimensions n ≥ 3 in general. This
does, however, not at all mean that they are uninteresting. Besides their natural
appearance, the importance of the higher dimensional cohomology groups is
seen in the fact that the theory endows them with an abundance of homomorphic
connections, with which one obtains important isomorphism theorems. These
theorems give concrete results for the interesting lower dimensional groups,
whose proofs, however, have to take the cohomology groups of all dimensions
into account.
Next we show that the cohomology groups H n (G, A) of a profinite group G
with coefficients in a G-module A are built up in a simple way from those of
the finite factor groups of G. Let U, V run through the open normal subgroups
of G. If V ⊆ U , then the projections
Gn+1 −→ (G/V )n+1 −→ (G/U )n+1
induce homomorphisms
C n (G/U, AU ) −→ C n (G/V, AV ) −→ C n (G, A),
which commute with the operators ∂ n+1 . We therefore obtain homomorphisms
H n (G/U, AU ) −→ H n (G/V, AV ) −→ H n (G, A).
The groups H n (G/U, AU ) thus form a direct system and we have a canonical
homomorphism lim H n (G/U, AU ) −→ H n (G, A).
−→ U

(1.2.5) Proposition. The above homomorphism is an isomorphism:

lim H n (G/U, AU ) −→
∼ H n (G, A) .
−→
U
Proof: Already the homomorphism

.
lim C (G/U, AU ) −→ C (G, A) .
−→
U
is an isomorphism of complexes. The injectivity is clear, since the maps
.
C (G/U, AU ) → C (G, A) .
are injective.
Let conversely x : Gn+1 −→ A be an n-cochain of G. Since A is discrete, x
is locally constant. We conclude that there exists an open normal subgroup U0
of G such that x is constant on the cosets of U0n+1 in Gn+1 . It takes values in
AU0 , since for all σ ∈ U0 we have
x(σ0 , . . . , σn ) = x(σσ0 , . . . , σσn ) = σx(σ0 , . . . , σn ).
Hence x is the composite of
xU
Gn+1 −→ (G/U0 )n+1 −→
0
A U0
with an n-cochain xU0 of G/U0 , and is therefore the image of the element in
lim
−→ U
C n (G/U, AU ) defined by xU0 . This shows the surjectivity. Since the
functor lim is exact, we obtain the isomorphisms
−→
lim H n (G/U, AU ) ∼
−→
= H n (lim C (G/U, AU ))
−→
.
U U
∼ .
= H n (C (G, A))
= H n (G, A). 2
Finally we introduce Tate cohomology. We do this for finite groups here,

and will extend the theory to profinite groups in §8. Let G be a finite group for
the remainder of this section.
We consider the norm residue group

Ĥ 0 (G, A) = AG /NG A,
where NG A is the image of the norm map∗)
X
NG : A −→ A, NG a = σa.
σ ∈G
∗) The name “norm” is chosen instead of “trace”, because in Galois cohomology this map
Q
will often be written multiplicatively, i.e. NG a = σa.
σ ∈G

We call the groups

AG /NG A
(
n
for n = 0,
Ĥ (G, A) =
H n (G, A) for n ≥ 1
the modified cohomology groups. We also obtain these groups from a com-
plex. Namely, we extend the standard complex (C n (G, A))n≥0 to
. ∂0 ∂1
Ĉ (G, A) : C −1 (G, A) −→ C 0 (G, A) −→ C 1 (G, A) −→ . . . ,
∂2
where C −1 (G, A) = C 0 (G, A) and ∂ 0 x is the constant function with value

P
σ ∈G x(σ). We then obtain the modified cohomology groups for all n
≥ 0 as
the cohomology groups of this complex,

.
Ĥ n (G, A) = H n (Ĉ (G, A)).
Besides the fixed module AG , we have also a “cofixed module” AG = A/IG A,
where IG A is the subgroup of A generated by all elements of the form
σa − a, a ∈ A, σ ∈ G. AG is the largest quotient of A on which G acts
trivially. We set
H0 (G, A) = AG .
If G is a finite group, then IG A is contained in the group
NG A = {a ∈ A | NG a = 0},
and we set
Ĥ0 (G, A) = NG A/IG A.
The norm NG : A −→ A induces a map NG : H0 (G, A) −→ H 0 (G, A), and
the proof of the following proposition is obvious.
(1.2.6) Proposition. We have an exact sequence

N
G
0 −→ Ĥ0 (G, A) −→ H0 (G, A) −→ H 0 (G, A) −→ Ĥ 0 (G, A) −→ 0.
The group Ĥ0 (G, A) is very often denoted by Ĥ −1 (G, A) for the following
reason. For a finite group G one can define cohomology groups Ĥ n (G, A) for
arbitrary integral dimensions n ∈ ZZ as follows:
For n ≥ 0, let ZZ[Gn+1 ] be the abelian group of all formal ZZ-linear combi-
P
nations a(σ0 ,...,σn ) (σ0 , . . . , σn ), σ0 , . . . , σn ∈ G, with its obvious G-module
structure. We consider the (homological) complete standard resolution of
ZZ, i.e. the sequence of G-modules X = X (G, ZZ) . .
2 ∂ 1 0 ∂ ∂ ∂−1
. . . −→ X2 −→ X1 −→ X0 −→ X−1 −→ X−2 −→ . . .

where Xn = X−1−n = ZZ[Gn+1 ] for n ≥ 0, and the differentials are defined for
n > 0 by
n
(−1)i (σ0 , . . . , σi−1 , σi+1 , . . . , σn )
X
∂n (σ0 , . . . , σn ) =
i=0
n
(−1)i (σ0 , . . . , σi−1 , τ, σi , . . . , σn−1 ) ,
XX
∂−n (σ0 , . . . , σn−1 ) =
τ ∈G i=0
while ∂0 : X0 → X−1 is given by

X
∂0 (σ0 ) = τ.
τ ∈G
The (cohomological) complete standard resolution of A is defined as the

.
sequence of G-modules X = X (G, A) = Hom(X , A) . .
∂ −1 ∂0 ∂1 ∂2
. . . −→ X −2 −→ X −1 −→ X 0 −→ X 1 −→ X 2 −→ . . .
where X −1−n = X n = Hom(Xn , A) = Map(Gn+1 , A) for n ≥ 0 and ∂ n =
.
Hom(∂n , A) for n ∈ ZZ. X is a complex. Using the maps
D−n : X −n+1 −→ X −n
given by
(Dn x)(σ0 , . . . , σn ) = x(1, σ0 , . . . , σn ) for n ≥ 0,
(D−1 x)(σ0 ) = δσ0 ,1 x(1) ∗) for n = 1,
(D−n x)(σ0 , . . . , σn−1 ) = δσ0 ,1 x(σ1 , . . . , σn−1 ) for n ≥ 2,
we get
Dn ◦ ∂ n+1 + ∂ n ◦ Dn−1 = id
for all n ∈ ZZ. From this we conclude that the above complex X is exact. .
For every n ∈ ZZ, we now define the n-th Tate cohomology group Ĥ n (G, A)
as the cohomology group of the complex
.
Ĉ (G, A) = ((X n )G )n∈ ZZ
at the place n:
Ĥ n (G, A) = H n (Ĉ (G, A)). .
Clearly, for n ≥ 0 we get the previous (modified) cohomology groups, and it
is immediate to see that Ĥ −1 (G, A) is our group Ĥ0 (G, A) = NG A/IG A. More
generally, the Tate cohomology in negative dimensions can be identified with
homology (see §8).
∗) i.e. δ = 0 if σ =/ τ and δσ,σ = 1.
σ,τ

Exercise 1. Let G be a profinite group and A a G-group. Assume that either G or A is finite.
The semi-direct product is a group
Ĝ = A o G
containing A and G such that every element of Ĝ has a unique presentation aσ, a ∈ A, σ ∈ G,
and (aσ)(a0 σ 0 ) = a σa0 σσ 0 . We then have a group extension
π
1 → A → Ĝ → G → 1
and the inclusion G ,→ Ĝ is a homomorphic section of π. Two homomorphic sections
s, s0 : G → Ĝ of π are conjugate if there is an a ∈ A such that s0 (σ) = as(σ)a−1 for all σ ∈ G.
π
Let SEC (Ĝ → G) be the set of conjugacy classes of homomorphic sections of Ĝ → G. Then
there is a canonical bijection of pointed sets
H 1 (G, A) ∼
= SEC (Ĝ → G).
Exercise 2. There is the following interpretation of H 3 (G, A). Consider all possible exact
sequences i α π
1 −→ A −→ N −→ Ĝ −→ G −→ 1,
where N is a group with an action σ̂ : ν 7→ σ̂ ν of Ĝ satisfying α(ν) ν 0 = νν 0 ν −1 , ν, ν 0 ∈ N , and
α(σ̂ ν) = σ̂α(ν)σ̂ −1 , ν ∈ N, σ̂ ∈ Ĝ. Impose on the set of all such exact sequences the smallest
equivalence relation such that
1 −→ A −→ N −→ Ĝ −→ G −→ 1
is equivalent to
1 −→ A −→ N 0 −→ Ĝ0 −→ G −→ 1,
whenever there is a commutative diagram
!"#$%&'()* N Ĝ
1 A G 1
N0 Ĝ0
in which the vertical arrows are compatible with the actions of Ĝ and Ĝ0 on N and N 0 (but
need not be bijective). If EXT 2 (G, A) denotes the set of equivalence classes, then we have a
canonical bijection
EXT 2 (G, A) ∼
= H 3 (G, A)
(see [18], chap. IV, th. 5.4).
Exercise 3. Let G be finite and let (Ai )i∈I be a family of G-modules. Show that
Y Y
H r (G, Ai ) = H r (G, Ai )
i∈I i∈I
for all r ≥ 0.
Exercise 4. An inhomogeneous cochain x ∈ C n (G, A), n ≥ 1, is called normalized if
x(σ1 , . . . , σn ) = 0 whenever one of the σi is equal to 1. Show that every class in H n (G, A) is
represented by a normalized cocycle.
Hint: Construct inductively cochains x0 , x1 , . . . , xn ∈ C n (G, A) and y1 , . . . , yn ∈
C n−1 (G, A) such that
x0 = x, xi = xi−1 − ∂yi , i = 1, . . . , n,
yi (σ1 , . . . , σn−1 ) = (−1)i−1 xi−1 (σ1 , . . . , σi−1 , 1, σi , . . . , σn−1 ).
Then xn is normalized and x − xn is a coboundary.

§3. The Exact Cohomology Sequence 25
§3. The Exact Cohomology Sequence
Having introduced the cohomology groups H n (G, A), we now turn to the
question of how they behave if we change the G-module A. If
f : A −→ B
is a homomorphism of G-modules, i.e. a homomorphism such that f (σa)
= σf (a) for a ∈ A, σ ∈ G, then we have the induced homomorphism
f : C n (G, A) → C n (G, B), x(σ0 , . . . , σn ) 7→ f x(σ0 , . . . , σn ),
and the commutative diagram
· ·2+,10-./ · C n (G, A) ∂ n+1
C n+1 (G, A) ···
f f
∂ n+1
··· C n (G, B) C n+1 (G, B) ··· .
In other words, f : A −→ B induces a homomorphism
.
f : C (G, A) −→ C (G, B) .
of complexes. Taking homology groups of these complexes, we obtain homo-
morphisms
f : H n (G, A) −→ H n (G, B).
Besides these homomorphisms there is another homomorphism, the “connect-
ing homomorphism”, which is less obvious, but is of central importance in
cohomology theory. For its definition we make use of the following general
lemma, which should be seen as the crucial point of homological algebra.
(1.3.1) Snake lemma. Let

385674:9; A i
B
j
C 0
α β γ
i 0 j0
0 A0 B0 C0
be a commutative diagram of abelian groups with exact rows. We then have a
canonical exact sequence
i j
ker(i) ker(α) ker(β) ker(γ)
δ
FEBCD?@A><= i0 j0
coker(α) coker(β) coker(γ) coker(j 0 ).

Proof: The existence and exactness of the upper and the lower sequence is
evident. The crucial cohomological phenomenon is the natural, but slightly
hidden, appearance of the homomorphism
δ : ker(γ) −→ coker(α).
It is obtained as follows. Let c ∈ ker(γ). Let b ∈ B and a0 ∈ A0 be elements
such that
jb = c and i0 a0 = βb.
The element b exists since j is surjective and a0 exists (and is uniquely deter-
mined by b) since j 0 βb = γjb = γc = 0. We define
δc := a0 mod α(A).
This definition does not depend on the choice of b, since if b̃ ∈ B is another
element such that j b̃ = c and i0 ã0 = β b̃, ã0 ∈ A0 , then j(b̃ − b) = 0, i.e.
b̃ − b = ia, a ∈ A, so that i0 (ã0 − a0 ) = β(b̃ − b) = βia = i0 αa, and thus
ã0 − a0 = αa, i.e. ã0 ≡ a0 mod α(A).
Exactness at ker(γ): δc = 0 means a0 = αa, a ∈ A, which implies β(b − ia) =

i0 a0 − i0 αa = 0, i.e. b − ia ∈ ker(β) and j(b − ia) = c.
Exactness at coker(α): Let a0 ∈ A0 such that i0 a0 ≡ 0 mod β(B), i.e. i0 a0 =

βb, b ∈ B. Setting c = jb, we have by definition δc = a0 mod α(A). 2
We now show that every exact sequence of G-modules

i j
0 −→ A −→ B −→ C −→ 0
gives rise to a canonical homomorphism
δ : H n (G, C) −→ H n+1 (G, A)
for every n ≥ 0. We consider the commutative diagram
0QGHIJKLMNOP C n (G, A) C n (G, B) C n (G, C) 0

∂A ∂B ∂C
0 C n+1 (G, A) C n+1 (G, B) C n+1 (G, C) 0.
It is exact, which is seen by passing to the inhomogeneous cochains (see also

ex.1). By the snake lemma, we obtain a homomorphism
δ : ker(∂C ) −→ coker(∂A ).

(1.3.2) Theorem. For every exact sequence 0 −→ A → B → C → 0 of

G-modules, the above homomorphism δ induces a homomorphism
δ : H n (G, C) −→ H n+1 (G, A)
and we obtain an exact sequence
δ
0 −→ AG −→ B G −→ C G −→ H 1 (G, A) −→ · · ·
δ
· · · −→ H n (G, A) −→ H n (G, B) −→ H n (G, C) −→ H n+1 (G, A) −→ · · · .
Proof: Setting C n (G, A) = C n (G, A)/B n (G, A) and similarly for B and C
in place of A, we obtain from the above diagram the commutative diagram
VWXYZRSTU C n (G, A) C n (G, B) C n (G, C) 0
∂A ∂B ∂C
0 Z n+1 (G, A) Z n+1 (G, B) Z n+1 (G, C),

which is obviously exact. Noting that
ker(∂A ) = H n (G, A) and coker(∂A ) = H n+1 (G, A),
the snake lemma yields an exact sequence
H n (G, A) H n (G, B) H n (G, C)
δ
àbc[\]^_ H n+1 (G, A) H n+1 (G, B) H n+1 (G, C).

This proves the theorem. 2
The homomorphism δ : H n (G, C) −→ H n+1 (G, A) is called the connecting

homomorphism, or simply the δ-homomorphism, and the exact sequence in
the theorem is called the long exact cohomology sequence.
Remark: If the group G is finite, then, using the unrestricted complex

. . . −→ X n−1 (G, A) −→ X n (G, A) −→ X n+1 (G, A) −→ . . . (n ∈ ZZ)
mentioned in §2, we get by the same argument an unrestricted long exact
cohomology sequence
δ δ
· · · −→ Ĥ n (G, A) −→ Ĥ n (G, B) −→ Ĥ n (G, C) −→ Ĥ n+1 (G, A) −→ · · · .
The connecting homomorphism δ has the following compatibility properties.

(1.3.3) Proposition. If
0knijghdelmf i j
A B C 0
f h g
i 0 j0
0 A0 B0 C0 0
is an exact commutative diagram of G-modules, then the diagrams
orqp C)
H n (G, δ
H n+1 (G, A)
g f
δ
H n (G, C 0 ) H n+1 (G, A0 )
are commutative.
Proof: This follows immediately from the definition of δ. If cn ∈ H n (G, C)

and if bn ∈ C n (G, B) and an+1 ∈ C n+1 (G, A) are such that jbn = cn and ian+1 =
∂ n+1 bn , then δcn = an+1 , and f δcn = f an+1 = f an+1 . On the other hand, setting
c0n = gcn , b0n = hbn , a0n+1 = f an+1 , we have j 0 b0n = c0n , i0 a0n+1 = ∂ n+1 b0n , so
that
δgcn = δc0n = a0n+1 = f an+1 = f δcn . 2
In order to avoid repeated explanations it is convenient to introduce the

notion of δ-functor. Let A and B be abelian categories. An exact δ-functor
from A to B is a family H = {H n }n∈ ZZ of functors H n : A −→ B together
with homomorphisms
δ : H n (C) −→ H n+1 (A)
defined for each short exact sequence 0 −→ A −→ B −→ C −→ 0 in A with
the following properties:
(i) δ is functorial, i.e. if
0stuvwxyz{|} A B C 0
0 A0 B0 C0 0
is a commutative diagram of short exact sequences in A, then
~
H n (C) δ
H n+1 (A)
δ
H n (C 0 ) H n+1 (A0 )
is a commutative diagram in B.

(ii) The sequence

δ
· · · −→ H n (A) −→ H n (B) −→ H n (C) −→ H n+1 (A) −→ · · ·
is exact for every exact sequence 0 → A → B → C → 0 in A.
If a family of functors H n is given only for an interval −∞ ≤ r ≤ n ≤ s ≤ ∞,
then one completes it tacitly by setting H n = 0 for n ∈/ [r, s].
In this sense the family of functors H n (G, −) (completed by H n (G, −) = 0
for n < 0) is a δ-functor from the category of G-modules into the category of
abelian groups. A curious property of δ-functors is their “anticommutativity”.
(1.3.4) Proposition. Let {H n } be an exact δ-functor from A to B. If

0 0 0
0 A0 A A00 0
0 B0 B B 00 0
0 C0 C C 00 0
0 0 0
is a commutative diagram in A with exact rows and columns, then
H n−1 (C 00 ) δ
H n (C 0 )
δ −δ
δ
H n (A00 ) H n+1 (A0 )
is a commutative diagram in B.
Proof: It simplifies the proof if we assume that A is a category whose objects

are abelian groups (together with some extra structure), as then we may prove
statements by “diagram chases” with elements. We may do this, since it can be
shown that every small abelian category may be fully embedded into a category
of modules over an appropriate ring in such a way that exactness relations are
preserved, and in any case we shall apply the proposition only to the category
of G-modules.

Let D be the kernel of the composite map B −→ C 00 , so that the sequence

0 −→ D −→ B −→ C 00 −→ 0
is exact. Let
i : A0 −→ A ⊕ B 0
be the direct sum of the maps A0 −→ A and A0 −→ B 0 and let
j : A ⊕ B 0 −→ B
be the difference d1 − d2 of the maps d1 : A −→ B and d2 : B 0 −→ B. One
checks at once that we get an exact sequence
i j
0 −→ A0 −→ A ⊕ B 0 −→ D −→ 0
and that the diagram
A°±²³®¯¬«¤¥¦§¨©ª£ ¡¢ 0 A A00 B 00 C 00
id pr1 id
A0 A ⊕ B0 D B C 00
−id −pr2 id
A0 B0 C0 C C 00
of solid arrows is commutative. This diagram can be commutatively com-
pleted by homomorphisms D → A00 and D → C 0 , since im(D → B 00 ) ⊆
im(A00 → B 00 ) and A00 → B 00 is injective, and since im(D → C) ⊆ im(C 0 → C)
and C 0 → C is injective. From this we obtain the commutative diagram
H n−1´º¹·¸¶µ¿¾¼½» (C 00 ) δ
H n (A00 ) δ
H n+1 (A0 )
id id
δ δ
H n−1 (C 00 ) H n (D) H n+1 (A0 )
id −id
δ δ
H n−1 (C 00 ) H n (C 0 ) H n+1 (A0 )
and the proposition follows. 2
From the exact cohomology sequence, we often get important isomorphism

theorems. For example, if
0 −→ A −→ B −→ C −→ 0

is an exact sequence of G-modules and if H n (G, B) = H n+1 (G, B) = 0, then

δ : H n (G, C) −→ H n+1 (G, A)
is an isomorphism. For this reason it is very important to know which G-
modules are cohomologically trivial in the following sense.
(1.3.5) Definition. A G-module A is called acyclic if H n (G, A) = 0 for all

n > 0. A is called cohomologically trivial (welk in German, flasque in
French) if
H n (H, A) = 0
for all closed subgroups H of G and all n > 0.
Important examples of cohomologically trivial G-modules are the induced

G-modules given by
IndG (A) = Map (G, A),
where A is any G-module. The elements of IndG (A) are the continuous
functions x : G −→ A (with the discrete topology on A) and the action of
σ ∈ G on x is given by (σx)(τ ) = σx(σ −1 τ ).
If G is a finite group, then we have an isomorphism
IndG (A) ∼
= A ⊗ ZZ[G]
X X
given by x 7→ x(σ) ⊗ σ, where ZZ[G] = { nσ σ | nσ ∈ ZZ} is the group
σ ∈G σ ∈G
ring of G.
(1.3.6) Proposition. (i) The functor A 7→ IndG (A) is exact.

(ii) An induced G-module A is also an induced H-module for every closed
subgroup H of G, and if H is normal, then AH is an induced G/H-module.
(iii) If one of the G-modules A and B is induced, then so is A ⊗ B. If G is
finite, the same holds for Hom(A, B).
(iv) If U runs through the open normal subgroups of G, then
IndG (A) = lim
−→
IndG/U (AU ).
U
We leave the simple proof to the reader (for (ii) use ex.4 of §1 to find a
homeomorphism G ∼ = H × G/H). As mentioned above, the very importance
of the induced G-modules lies in the following fact.

(1.3.7) Proposition. The induced G-modules M = IndG (A) are cohomologi-

cally trivial. If G is finite, we have moreover Ĥ n (G, M ) = 0 for all n ∈ ZZ. ∗)
Proof: We consider the standard resolutions

. .
X (G, A) and X (G, IndG (A))
of A and IndG (A). The map
X n (G, IndG (A))G −→ X n (G, A),
given by x(σ0 , . . . , σn ) 7→ y(σ0 , . . . , σn ) = x(σ0 , . . . , σn )(1) obviously com-
mutes with ∂. Furthermore, it is an isomorphism, since it has the map
y(σ0 , . . . , σn ) 7→ x(σ0 , . . . , σn )(σ) = σy(σ −1 σ0 , . . . , σ −1 σn ) as inverse. We
thus have an isomorphism
.
C (G, IndG (A)) ∼
= X (G, A) .
.
of complexes. But X (G, A) is exact by (1.2.1), so that
.
H n (G, IndG (A)) = H n (C (G, IndG (A))) = H n (X (G, A)) = 0 .
for n ≥ 1. If H is a closed subgroup of G, then by (1.3.6) we may write
IndG (A) = IndH (B) and get H n (H, IndG (A)) = 0. If G is finite, then the same
argument holds for the extended complex (X n )n∈ ZZ , hence Ĥ n (G, IndG (A)) = 0
for all n ∈ ZZ. 2
The above proposition allows us to adopt a technique, called dimension

shifting, by which definitions and proofs concerning the cohomology groups
for all G-modules A and all n, may be reduced to a single dimension n, e.g.
n = 0. Given A, define the G-module A1 by the exact sequence
i
0 −→ A −→ IndG (A) −→ A1 −→ 0,
where ia is the constant function (ia)(σ) = a. This is a sequence of G-modules.
If H is a closed subgroup of G, then H n (H, IndG (A)) = 0 for all n ≥ 1 by
(1.3.7), and the exact cohomology sequence shows that the homomorphism
δ : H n (H, A1 ) −→ H n+1 (H, A)
is surjective for n = 0 and bijective for n > 0. If we define A0 = A and
inductively
Ap = (Ap−1 )1 for p > 0,
then (1.3.7) yields inductively the

∗) We shall see in §8 that Ĥ n (G, A) = 0, n ∈ ZZ, for any cohomologically trivial G-module A.

(1.3.8) Proposition. For all n, p ≥ 0 and all subgroups H ⊆ G, we have a

canonical homomorphism
δ p : H n (H, Ap ) −→ H n+p (H, A),
which is a surjection for n = 0 and an isomorphism for n > 0.
If G is a finite group, then we may also consider the exact sequence

ν
0 −→ A−1 −→ IndG (A) −→ A −→ 0,
P
where ν associates the element σ∈G x(σ) to an element x ∈ IndG (A). We
define
Ap = (Ap+1 )−1 for p < 0 ,
It is easy to see that
Ap ∼ ⊗p
= A ⊗ JG and A−p ∼ ⊗p
= A ⊗ IG
for p ≥ 0, where the G-modules IG and JG are given by the exact sequences
0ÆÇÀÁÂÃÄÅ IG ZZ[G] ε
ZZ 0,
NG
0 ZZ ZZ[G] JG 0.
Here ε is the augmentation map
X X
ε: aσ σ 7→ aσ ,
σ ∈G σ ∈G
P
and NG (1) = σ∈G σ. The G-module IG is called the augmentation ideal
of ZZ[G].
Noting that Ĥ n (H, IndG (A)) = 0, we obtain canonical isomorphisms
Ĥ n (H, A) ∼
= Ĥ n−p (H, Ap )
for all n, p ∈ ZZ. Furthermore, we observe the following rule for the G-module
Hom(A, Q/ZZ): if p ∈ ZZ, then
Hom(A, Q/ZZ)p ∼= Hom(A−p , Q/ZZ) .
Let G again be a profinite group. The G-modules X n = X n (G, A) in the

standard resolution
0 −→ A −→ X 0 −→ X 1 −→ X 2 −→ . . .
are all induced G-modules, since X 0 = IndG (A) and X n = IndG (X n−1 ). They
are thus cohomologically trivial and, in particular, acyclic. We call a resolution
0 −→ A −→ Y 0 −→ Y 1 −→ Y 2 −→ . . .
of A acyclic (resp. resolution by cohomologically trivial G-modules) if the
Y n are acyclic (resp. cohomologically trivial). It is a remarkable fact that the
cohomology groups H n (G, A) can be obtained from any acyclic resolution.

(1.3.9) Proposition. If
∂ ∂ ∂
0 −→ A −→ Y 0 −→ Y 1 −→ Y 2 −→ . . .
is an acyclic resolution of A, then canonically
H n (G, A) ∼
= H n (H 0 (G, Y )). .
∂
Proof: Setting K p = ker(Y p −→ Y p+1 ), we obtain the short exact sequences
0 −→ A −→ Y 0 −→ K 1 −→ 0,
1 1 2
0 −→ K −→ Y −→ K −→ 0,
···
0 −→ K n−2 −→ Y n−2 −→ K n−1 −→ 0,
0 −→ K n−1 −→ Y n−1 −→ K n −→ 0.
Since the Y n are acyclic, the exact cohomology sequence yields for n ≥ 1
isomorphisms
δ δ δ
H 1 (G, K n−1 ) −→ 2
∼ H (G, K
n−2
) −→
∼ · · · −→
n
∼ H (G, A).
On the other hand we have the exact sequence

H 0 (G, Y n−1 ) −→ H 0 (G, K n ) −→ H 1 (G, K n−1 ) −→ H 1 (G, Y n−1 ) = 0
and
H 0 (G, K n ) = ker(H 0 (G, Y n ) −→ H 0 (G, Y n+1 )),
im (H 0 (G, Y n−1 ) −→ H 0 (G, K n )) = im (H 0 (G, Y n−1 ) −→ H 0 (G, Y n )),
which proves that canonically
H n (G, A) ∼
= H 1 (G, K n−1 ) = H n (H 0 (G, Y )). . 2
For example, if H is a closed subgroup of G, then the standard resolution

.
0 → A → X of a G-module A is also an acyclic resolution of the H-
module A, hence H n (H, A) ∼ .
= H n (H 0 (H, X )). This isomorphism is also
. .
obtained from the restriction map X (G, A)H → X (H, A)H , as one may see
by dimension shifting.
Remark: There exists a variant of (1.3.9) for the modified cohomology. If G

is a finite group, then, for all q ∈ ZZ and every G-module A, there are canonical
isomorphisms
Ĥ q (G, A) ∼
= H q (H 0 (G, Y )), .
where Y . is a complete acyclic resolution of A, i.e. a complex

· ·ÉÊÈËÐÑÏÎÍÌ · ∂ −2
Y −2 ∂ −1
Y −1 ∂0
Y0 ∂1
Y1 ∂2
Y2 ∂3
···
ε µ
A
0 0
consisting of cohomologically trivial G-modules Y n , n ∈ ZZ, which is exact
everywhere and ∂ 0 = µ ◦ ε.
Exercise 1. The functor A 7→ C n (G, A) is exact.

Hint: C n (G, A) = X n (G, A)G , and X n (G, A) is induced.
f g
Exercise 2. For any pair of maps A −→ B −→ C of abelian groups, there is an exact sequence
0 ker(f ) ker(g ◦ f ) ker(g)
ÒÓÔÕÖ×ØÙÚÛÜ coker(f ) coker(g ◦ f ) coker(g) 0.
Exercise 3. Let G be a finite group and let

α β γ
0 −→ A −→ B −→ C −→ D −→ 0
be an exact sequence of G-modules. Define a homomorphism
δ 2 : Ĥ n−1 (G, D) −→ Ĥ n+1 (G, A).
Show that the following conditions are equivalent:
(i) δ 2 is an isomorphism for all n ∈ ZZ,
(ii) Ĥ n (G, B) → Ĥ n (G, C) is an isomorphism for all n ∈ ZZ.
Hint: Show this first under the assumption Ĥ n (G, B) = 0 for all n ∈ ZZ. Then apply (1.3.4)
to the exact commutative diagram
òóôîïðñíâãäåæçèéêëìàáßÝÞ 0 0 0
α β
0 A B βB 0
i◦α (i,β)
(id,0) 0+id
0 IndG B IndG B ⊕ C C 0
γ
0 X Y D 0
0 0 0
where X = coker (i ◦ α) and Y = coker (i, β).
Exercise 4. Let A be a G-module and let A0 be the trivial G-module with underlying abelian
group A. Then we have an isomorphism IndG (A) ∼ = IndG (A0 ) of G-modules.

Exercise 5. Let G be a finite group and let 0 → A → I1 → I2 → · · · → Ip → B

→ 0 be an exact sequence of G-modules, where the I1 , . . . , Ip are acyclic G-modules. Then
Ĥ n (G, A) ∼
= Ĥ n−p (G, B) for n ≥ p.
. .
Exercise 6. Let C = (C n , dn )n∈ ZZ and C 0 = (C 0n , d0n )n∈ ZZ be two complexes in an abelian
category and let f = (fn )n∈ ZZ and g = (gn )n∈ ZZ be two morphisms from C to C 0 . A . .
homotopy from f to g is a family h = (hn )n∈ ZZ of morphisms hn : C n+1 → C 0n such that
hn dn+1 + d0n hn−1 = fn − gn .
We say that f and g are homotopic and write f ' g if such a family exists.
.
Show that in this case f and g induce the same homomorphisms H n (C ) → H n (C 0 ) on .
the homology.
Exercise 7. If G is finite and (Ai )i∈I is a projective system of finite G-modules, then
H n (G, lim Ai ) = lim H n (G, Ai ).
←− i ←− i
(Compare exercise 3 in §2.)
Exercise 8. If 1 → A → B → C → 1 is an exact sequence of G-groups, then
1 −→ AG −→ B G −→ C G −→ H 1 (G, A) −→ H 1 (G, B) −→ H 1 (G, C).

is an exact sequence of pointed sets, i.e. the image of a map is equal to the pre-image of the
distinguished element. If A is in the center of B, then the sequence extends exactly by an
δ
arrow → H 2 (G, A), given by cσ 7→ aσ,τ = cσ σcτ c−1
στ .
§4. The Cup-Product
If A and B are two G-modules, then A ⊗ ZZ B is also a G-module (by

σ(a ⊗ b) = σa ⊗ σb), and we obtain for every pair p, q ≥ 0 a bilinear map
∪
(∗) C p (G, A) × C q (G, B) −→ C p+q (G, A ⊗ B)
by
(a ∪ b)(σ0 , . . . , σp+q ) = a(σ0 , . . . , σp ) ⊗ b(σp , . . . , σp+q ).
For this map, we have the following formula.
(1.4.1) Proposition. ∂(a ∪ b) = (∂a) ∪ b + (−1)p (a ∪ ∂b).

§4. The Cup-Product 37
Proof: We have
p+q+1
X
∂(a ∪ b)(σ0 , . . . , σp+q+1 ) = (−1)i (a ∪ b)(σ0 , . . . , σ̂i , . . . , σp+q+1 )
i=0
p
X
= (−1)i a(σ0 , . . . , σ̂i , . . . , σp+1 ) ⊗ b(σp+1 , . . . , σp+q+1 )
i=0
p+q+1
X
+ (−1)i a(σ0 , . . . , σp ) ⊗ b(σp , . . . , σ̂i , . . . , σp+q+1 ).
i=p+1
On the other hand

(∂a ∪ b)(σ0 , . . . , σp+q+1 ) = (∂a)(σ0 , . . . , σp+1 ) ⊗ b(σp+1 , . . . , σp+q+1 )
p+1
X
= (−1)i a(σ0 , . . . , σ̂i , . . . , σp+1 ) ⊗ b(σp+1 , . . . , σp+q+1 )
i=0
and
(a ∪ ∂b)(σ0 , . . . , σp+q+1 ) = a(σ0 , . . . , σp ) ⊗ (∂b)(σp , . . . , σp+q+1 )
q+1
X
= a(σ0 , . . . , σp ) ⊗ (−1)i b(σp , . . . , σ̂p+i , . . . , σp+q+1 ).
i=0
Now, in the second of these seven formula lines, let the index i run from 0 to
p + 1 and in the third from p to p + q + 1. The additional summands appearing
cancel to give the result claimed. 2
From this proposition, it follows that a ∪ b is a cocycle if a and b are cocycles,

and a coboundary if one of the cochains a and b is a coboundary and the other
a cocycle. Therefore the pairing (∗) induces a bilinear map
∪
H p (G, A) × H q (G, B) −→ H p+q (G, A ⊗ B), (α, β) 7−→ α ∪ β.
This map is called the cup-product. For p = q = 0, we obtain the map
AG × B G −→(A ⊗ B)G , (a, b) 7−→ a ⊗ b.
We will see below that the cup-product induces a bilinear map
∪
Ĥ p (G, A) × Ĥ q (G, B) −→ Ĥ p+q (G, A ⊗ B)
on the modified cohomology of a finite group G for all p, q ∈ ZZ.
Whenever a new cohomological map is introduced, we must check its func-

toriality properties and also its compatibility with the cohomological maps
already defined. Directly from the definition follows the

(1.4.2) Proposition. For two homomorphisms A → A0 , B → B 0 of G-

modules, we have the commutative diagram
ö÷øùõ A) × H q (G, B)
H p (G, ∪
H p+q (G, A ⊗ B)
∪
H p (G, A0 ) × H q (G, B 0 ) H p+q (G, A0 ⊗ B 0 ) .
The cup-product is very often used in a slightly more general form. Instead
of the bilinear map A × B → A ⊗ B, we may consider an arbitrary bilinear
pairing of G-modules A × B −→ C, (a, b) 7−→ ab. It factors through A ⊗ B,
and the composite
∪
H p (G, A) × H q (G, B) −→ H p+q (G, A ⊗ B) −→ H p+q (G, C)
is also called the cup-product. The compatibility with the δ-homomorphism is
given in the following proposition.
(1.4.3) Proposition. (i) Let

0 → A0 → A → A00 → 0 and 0 → C 0 → C → C 00 → 0
be exact sequences of G-modules. Let B be another G-module and suppose
we are given a pairing A × B → C which induces pairings A0 × B → C 0 and
A00 × B → C 00 . Then the diagram
þýüûú A00 ) × H q (G, B)
H p (G, ∪
H p+q (G, C 00 )
δ id δ
∪
H p+1 (G, A0 ) × H q (G, B) H p+q+1 (G, C 0 )
is commutative, i.e.
δ(α00 ∪ β) = δα00 ∪ β.
(ii) Let
0 → B 0 → B → B 00 → 0 and 0 → C 0 → C → C 00 → 0
be exact sequences of G-modules and let A × B → C be a pairing which
induces pairings A × B 0 → C 0 and A × B 00 → C 00 . Then the diagram
ÿ A) × H q (G, B 00 )
H p (G, ∪
H p+q (G, C 00 )
id δ (−1)p δ
∪
H p (G, A) × H q+1 (G, B 0 ) H p+q+1 (G, C 0 )
(−1)p δ(α ∪ β 00 ) = α ∪ δβ 00 .

Proof: We show (ii). Let α = ā, β 00 = b̄00 , a ∈ Z p (G, A), b00 ∈ Z q (G, B 00 ).
Let b ∈ C q (G, B) be a pre-image of b00 (b exists by §3, ex.1). Identifying B 0 with
its image in B, δβ 00 is by definition represented by the cocycle ∂b ∈ Z q+1 (G, B 0 )
and δ(α ∪ β 00 ) by the cocycle ∂(a ∪ b) ∈ Z p+q+1 (G, C 0 ). Recalling that ∂a = 0,
we obtain from (1.4.1)
∂(a ∪ b) = (∂a) ∪ b + (−1)p (a ∪ ∂b) = (−1)p (a ∪ ∂b).
Passing to the cohomology classes, we get δ(α ∪ β 00 ) = (−1)p (α ∪ δβ 00 ).
(i) is proven in the same way. 2
As before, we make everywhere the identifications

(A ⊗ B) ⊗ C = A ⊗ (B ⊗ C) and A ⊗ B = B ⊗ A.
(1.4.4) Proposition. The cup-product is associative and skew commutative,

i.e. for α ∈ H p (G, A), β ∈ H q (G, B), γ ∈ H r (G, C) we have
(α ∪ β) ∪ γ = α ∪ (β ∪ γ)
and
α ∪ β = (−1)pq (β ∪ α).
Proof: Let a, b, c be cocycles representing α, β, γ. Then

((a ∪ b) ∪ c)(σ0 , . . . , σp+q+r ) = (a ∪ b)(σ0 , . . . , σp+q ) ⊗ c(σp+q , . . . , σp+q+r )
= a(σ0 , . . . , σp ) ⊗ b(σp , . . . , σp+q ) ⊗ c(σp+q , . . . , σp+q+r )
= a(σ0 , . . . , σp ) ⊗ (b ∪ c)(σp , . . . , σp+q+r ) = (a ∪ (b ∪ c))(σ0 , . . . , σp+q+r ).
Passing to the cohomology classes gives (α ∪ β) ∪ γ = α ∪ (β ∪ γ).
The formula α ∪ β = (−1)pq (β ∪ α) is not so easy to prove at the level of
cocycles. We therefore use the method of dimension shifting introduced in §3.
By (1.3.8), we have the surjections δ n : H 0 (G, An ) −→ H n (G, A). Applying
(1.4.3) (i) p times (resp. (ii) q times), we obtain a commutative diagram

Ap ) × H 0 (G, Bq )
H 0 (G, ∪
H 0 (G, (A ⊗ Bq )p ) = H 0 (G, Ap ⊗ Bq )
δp id δp
∪
H p (G, A) × H 0 (G, Bq ) H p (G, (A ⊗ B)q ) = H p (G, A ⊗ Bq )
id δq (−1)pq δ q
∪
H p (G, A) × H q (G, B) H p+q (G, A ⊗ B).

For p = q = 0 the rule α ∪ β = β ∪ α is clear. Since the vertical arrows are

surjective, α ∪ β = (−1)pq (β ∪ α) follows for p, q ≥ 0. 2
(1.4.5) Proposition. Let

0 −→ A0 −→ A −→ A00 −→ 0 and 0 −→ B 0 −→ B −→ B 00 −→ 0
be exact sequences of G-modules. Suppose we are given a pairing
ϕ : A × B →C
into a G-module C such that ϕ(A0 × B 0 ) = 0, i.e. ϕ induces pairings ϕ0 and ϕ00
such that the diagram
ϕ0
A 0 × B 00 C
i v
ϕ
A × B C
j u
ϕ00
A00 × B 0 C
commutes. Then the diagram
H p (G, A00 ) × H q (G, B 0 ) ∪
H p+q (G, C)
δ δ (−1)p+1
∪
H p+1 (G, A0 ) × H q−1 (G, B 00 ) H p+q (G, C)
(δα) ∪ β + (−1)p (α ∪ δβ) = 0
for α ∈ H p (G, A00 ) and β ∈ H q−1 (G, B 00 ).
Proof: Let a00 ∈ Z p (G, A00 ) and b00 ∈ Z q−1 (G, B 00 ) be cocycles representing α
and β respectively, and let a ∈ C p (G, A) and b ∈ C q−1 (G, B) be pre-images.
Then there exist a0 ∈ C p+1 (G, A0 ) and b 0 ∈ C q (G, B 0 ) such that ia0 = ∂a and
ub 0 = ∂b. Now δα is represented by a0 and δβ by b 0 . It follows that
(δα) ∪ β + (−1)p (α ∪ δβ) = 0
since the left class is represented by
a0 ∪ b00 + (−1)p (a00 ∪ b 0 ) = a0 ∪ vb + (−1)p (ja ∪ b 0 )
= ia0 ∪ b + (−1)p (a ∪ ub 0 )
= ∂a ∪ b + (−1)p (a ∪ ∂b)
= ∂(a ∪ b),
which is a coboundary. 2

For discrete G-modules A and C, let G act on Hom(A, C) by

(gφ)(a) = gφ(g −1 a),
where φ ∈ Hom(A, C), g ∈ G and a ∈ A. If G is finite or if A is finitely
generated as a ZZ-module, then Hom(A, C) is a discrete G-module and the
canonical pairing
Hom(A, C) × A −→ C
induces the cup-product
∪
H p (G, Hom(A, C)) × H q (G, A) −→ H p+q (G, C).
We have the following
(1.4.6) Corollary. Let

0 −→ A0 −→ A −→ A00 −→ 0
be an exact sequence of G-modules and suppose that C is another G-module
such that the sequence
0 −→ Hom(A00 , C) −→ Hom(A, C) −→ Hom(A0 , C) −→ 0
is also exact. Assume further, that G is finite or that A, A0 and A00 are finitely
generated as ZZ-modules. Then the diagram

H p (G, Hom(A0 , C)) × H q (G, A0 ) ∪
H p+q (G, C)
δ δ (−1)p+1
∪
H p+1 (G, Hom(A00 , C)) × H q−1 (G, A00 ) H p+q (G, C)
(δ α̂) ∪ α + (−1)p (α̂ ∪ δα) = 0
for α̂ ∈ H p (G, Hom(A0 , C)) and α ∈ H q−1 (G, A00 ).
Proof: This follows from (1.4.5) applied to the natural pairing

Hom(A, C) × A −→ C. 2
In the next proposition, whose proof is taken from [7], §7, we define the
cup-product in arbitrary integral dimensions if G is a finite group.

(1.4.7) Proposition. Let G be a finite group. Then there exists a unique family
of bilinear maps
∪
Ĥ p (G, A) × Ĥ q (G, B) −→ Ĥ p+q (G, C),
defined for all integers p, q ∈ ZZ and all pairings A × B → C of G-modules,
such that:
(i) These bilinear maps are functorial with respect to the modules.
(ii) For p = q = 0 they are induced by the natural map
AG × B G −→ C G .
(iii) Let
0 → A0 → A → A00 → 0 and 0 → C 0 → C → C 00 → 0
be exact sequences of G-modules. Let B be another G-module and suppose
we are given a pairing A × B → C which induces pairings A0 × B → C 0 and
A00 × B → C 00 . Then the diagram
$"!# A00 ) × Ĥ q (G, B)
Ĥ p (G, ∪
Ĥ p+q (G, C 00 )
δ id δ
∪
Ĥ p+1 (G, A0 ) × Ĥ q (G, B) Ĥ p+q+1 (G, C 0 )
is commutative, i.e. δ(α00 ∪ β) = δα00 ∪ β.

(iv) Let
0 → B 0 → B → B 00 → 0 and 0 → C 0 → C → C 00 → 0
be exact sequences of G-modules and let A × B → C be a pairing which
induces pairings A × B 0 → C 0 and A × B 00 → C 00 . Then the diagram
%)('& A) × Ĥ q (G, B 00 )
Ĥ p (G, ∪
Ĥ p+q (G, C 00 )
id δ (−1)p δ
∪
Ĥ p (G, A) × Ĥ q+1 (G, B 0 ) Ĥ p+q+1 (G, C 0 )
is commutative, i.e. (−1)p δ(α ∪ β 00 ) = α ∪ δβ 00 .
Proof: Without loss of generality, we may assume that C = A ⊗ B. If

. .
X = X (G, ZZ) denotes the complete standard resolution of ZZ, cf. §2, p.22,
then we obtain a homomorphism of complexes
. .
HomG (X , A) ⊗ HomG (X , B) −→ HomG (X ⊗ X , A ⊗ B) . . .
The proof of existence of the cup-product depends on constructing G-module

homomorphisms
ϕp,q : Xp+q −→ Xp ⊗ Xq ,
for all integers p, q, satisfying the following two conditions
(1) ϕp,q ∂ = (∂ ⊗ 1)ϕp+1,q + (−1)p (1 ⊗ ∂)ϕp,q+1 ,
(2) (ε ⊗ ε)ϕ0,0 = ε ,
where ε : X0 → ZZ is defined by ε(σ) = 1 for all σ ∈ G. The induced map of
. . .
complexes ϕ : X −→ X ⊗ X then defines a homomorphism
. .
HomG (X , A) ⊗ HomG (X , B) −→ HomG (X , A ⊗ B) .
(f, g) 7−→ f · g = (f ⊗ g)ϕ .
It follows from (1) that
∂(f · g) = (∂f ) · g + (−1)p f · (∂g) .
Hence if f, g are cocycles, so is f · g, and the cohomology class of f · g depends
only on the classes of f and g. Thus we obtain homomorphisms
∪
Ĥ p (G, A) ⊗ Ĥ q (G, B) −→ Ĥ p+q (G, A ⊗ B) ,
which obviously satisfy (i), and (ii) is a consequence of (2). The properties (iii)
and (iv) are proved as in (1.4.3). This gives us the existence of the cup-product
and the uniqueness is proved by starting with (ii) and shifting dimensions by
(iii) and (iv), as in §3, p.32. Observe that the exact sequences
0 → A → IndG (A) → A1 → 0 and 0 → A−1 → IndG (A) → A → 0
split over ZZ. Thus the result of tensoring these by any G-module B is still
exact and IndG (A) ⊗ B = IndG (A ⊗ B).
It remains to define the maps ϕp,q , which we will do as follows:
If p ≥ 0 and q ≥ 0,
ϕp,q (σ0 , . . . , σp+q ) = (σ0 , . . . , σp ) ⊗ (σp , . . . , σp+q ) .
If p ≥ 1 and q ≥ 1,
ϕ−p,−q (σ1 , . . . , σp+q ) = (σ1 , . . . , σp ) ⊗ (σp+1 , . . . , σp+q ) .
If p ≥ 0 and q ≥ 1,
X
ϕp,−p−q (σ1 , . . . , σq ) = (σ1 , τ1 , . . . , τp ) ⊗ (τp , . . . , τ1 , σ1 , . . . , σq ) ,
X
ϕ−p−q,p (σ1 , . . . , σq ) = (σ1 , . . . , σq , τ1 , . . . , τp ) ⊗ (τp , . . . , τ1 , σq ) ,
X
ϕp+q,−q (σ0 , . . . , σp ) = (σ0 , . . . , σp , τ1 , . . . , τq ) ⊗ (τq , . . . , τ1 ) ,
X
ϕ−q,p+q (σ0 , . . . , σp ) = (τ1 , . . . , τq ) ⊗ (τq , . . . , τ1 , σ0 , . . . , σp ) ,
where the τi on the right-hand side run independently through G. The verifi-
cation that the ϕp,q satisfy the formulae above is straightforward. 2

Remark: Under the assumption that G is finite, the propositions (1.4.4) and
(1.4.5) hold for all integers p, q, r ∈ ZZ.
From the proof of the last proposition, we obtain the following explicit
formulae.
(1.4.8) Proposition. Let G be a finite group. Let x̄ ∈ Ĥ p (G, A), p ≥ −1

and ȳ ∈ H q (G, B), q ≥ 1. Let y ∈ C q (G, B) be an inhomogeneous cocycle
representing ȳ. For p ≥ 0, let x ∈ C p (G, A) be a representing cocycle of x̄.
For p = −1, let x ∈ NG A represent x̄. Then the inhomogeneous cochains
(x ∪ y)(σ1 , . . . , σq ) = x ⊗ y(σ1 , . . . , σq ) for p = 0,
σx ⊗ σy(σ −1 , σ1 , . . . , σq−1 )
X
(x ∪ y)(σ1 , . . . , σq−1 ) = σ ∈G
for p = −1,
(x ∪ y)(σ1 , . . . , σp+q ) = x(σ1 , . . . , σp ) ⊗ σ1 · · · σp y(σp+1 , . . . , σp+q )
for p > 0,
p+q
represent the cup-product x̄ ∪ ȳ ∈ Ĥ (G, A ⊗ B).
Exercise 1. Let R be a G-ring, i.e. a ring with an action of G such that σ(a + b) = σa + σb and
σ(ab) = σaσb. Show that
M
H(G, R) := H n (G, R)
n≥ 0
is a graded ring with respect to the cup-product which is induced by the multiplication
R × R → R.
Exercise 2. Let A be an R-module with a G-operation compatible with the R-module structure.
Show that
M
H(G, A) := H n (G, A)
n≥ 0
is in a natural way an H(G, R)-module.
Exercise 3. Prove the following generalization of (1.4.3): Let
0 → A0 → A → A00 → 0,
0 → B 0 → B → B 00 → 0,
0 → C 0 → C → C 00 → 0
be exact sequences of G-modules. Suppose we are given a pairing ϕ: A × B → C such that
ϕ(A0 × B 0 ) = 0 , ϕ(A × B 0 ) ⊆ C 0 and ϕ(A0 × B) ⊆ C 0 .
Then we get an induced pairing A00 ×B 00 → C 00 , and for α00 ∈ H p (G, A00 ) and β 00 ∈ H q (G, B 00 )
we have
δ(α00 ∪ β 00 ) = (δα00 ) ∪ β 00 + (−1)p α00 ∪ (δβ 00 ).

§5. Change of the Group G 45
§5. Change of the Group G
We now turn to the question of what happens to the cohomology groups

n
H (G, A) if we change the group G. We put ourselves in the most general
situation if we consider two profinite groups G and G0 , a G-module A, a
G0 -module A0 and two homomorphisms
ϕ : G0 −→ G, f : A −→ A0 ,
such that f (ϕ(σ 0 )a) = σ 0 f (a). From such a compatible pair of homomorphisms,
we obtain a homomorphism
C n (G, A) −→ C n (G0 , A0 ), a 7→ f ◦ a ◦ ϕ.
Trivially, this commutes with ∂ and therefore induces a homomorphism
H n (G, A) −→ H n (G0 , A0 ).
If we have two compatible pairs of homomorphisms G00 → G0 → G,
A → A0 → A00 , then the homomorphism
H n (G, A) −→ H n (G00 , A00 ),
induced by the composites G00 → G and A → A00 , is the composite of the
homomorphisms
H n (G, A) −→ H n (G0 , A0 ) and H n (G0 , A0 ) −→ H n (G00 , A00 )
given by G0 → G, A → A0 and G00 → G0 , A0 → A00 . Thus the cohomology
groups H n (G, A) are functorial in G and A simultaneously.
Let (Gi )i∈I be a projective system of profinite groups and let (Ai )i∈I be a
direct system, where each Ai is a Gi -module and the transition maps
Gj → Gi , Ai → Aj
form compatible pairs. Then, with the induced homomorphisms
H n (Gi , Ai ) −→ H n (Gj , Aj ) ,
the cohomology groups H n (Gi , Ai ) form a direct system of abelian groups.
As a generalization of (1.2.5), we have the
(1.5.1) Proposition. If G = lim Gi and A = lim Ai , then

←− i∈I −→ i∈I
H n (G, A) ∼
= lim H n (Gi , Ai ).
−→
i∈I
Proof: For every i ∈ I, the pair G → Gi , Ai → A is compatible. We

therefore have a canonical homomorphism κi : C n (Gi , Ai ) → C n (G, A),
hence a homomorphism
κ : lim
−→
C n (Gi , Ai ) −→ C n (G, A),
i∈I

which obviously commutes with the ∂-homomorphism. It therefore suffices to

show that κ is an isomorphism.
For the surjectivity, let y : Gn → A be the inhomogeneous cochain asso-
ciated to x ∈ C n (G, A). Since Gn is compact, A discrete and y continuous,
y takes only finitely many values and factors through ȳ : (G/U )n → A for a
suitable open normal subgroup U . The finitely many values are represented by
elements of some Ai , i.e. ȳ is the composite of a function ȳi : (G/U )n → Ai
with Ai → A. On the other hand, there exists a j ≥ i such that the projection
G → G/U factors through the canonical map Gj → G/U , i.e. we obtain an
inhomogeneous cochain yj : Gnj → Aj as the composite
ȳ
Gnj −→ (G/U )n −→
i
Ai −→ Aj ,
yj
such that the composite Gn −→ Gnj −→ Aj −→ A is y. If xj ∈ C n (Gj , Aj ) is
the homogeneous cochain associated to yj , then its image in C n (G, A) is x.
This shows the surjectivity of κ.
For the injectivity, let xi ∈ C n (Gi , Ai ) be a cochain which becomes zero in
C n (G, A), i.e. the composite
x
Gn+1 −→ Gn+1
i
i
−→ Ai −→ A
is zero. Since xi has only finitely many values, there exists a j ≥ i such that
the composite
xi
Gn+1
j −→ Gn+1
i −→ Ai −→ Aj
is already zero, i.e. xi becomes zero in C n (Gj , Aj ) and hence represents the
zero class in lim C n (Gi , Ai ). This shows the injectivity of κ. 2
−→
i
We shall have to deal mainly with three special cases of homomorphisms

H (G, A) → H n (G0 , A0 ) coming from compatible pairs G0 → G, A → A0 ,
n
and an additional case, arising in a different way.

1. Conjugation. Let H be a closed subgroup of G, A a G-module and B an
H-submodule of A. For σ, τ ∈ G we write τ σ = σ −1 τ σ and σH = σHσ −1 . The
two compatible homomorphisms
σ
H −→ H, τ 7−→ τ σ , B −→ σB, b 7−→ σb
induce isomorphisms
σ∗ : H n (H, B) −→ H n (σH, σB),
which are called conjugation. We have
1∗ = id and (στ )∗ = σ∗ τ∗ ,
from what we have said above about composition.

2. Inflation. Let H be a normal closed subgroup of G and A a G-module.

Then AH is a G/H-module. The projection and injection
G*+ G/H, AH A
form a compatible pair of homomorphisms, which induces a homomorphism
G/H
inf G : H n (G/H, AH ) −→ H n (G, A),
called inflation. The inflation is transitive, i.e. for two normal closed subgroups
H ⊆ F of G, we have
G/H G/F G/F
inf G ◦ inf G/H = inf G .
3. Restriction. For an arbitrary closed subgroup H of G and a G-module A,

we consider the two homomorphisms
H-, incl G, A id A.
On the cochains they induce the restriction maps and we obtain homomor-
phisms on the cohomology
res G n n
H : H (G, A) −→ H (H, A),
called restriction. Clearly the restriction is transitive, i.e. for two closed
subgroups F ⊆ H, we have
res H G G
F ◦ res H = res F .
4. Corestriction. If H is an open subgroup of G, then besides the restriction,

we have another map in the opposite direction, which is a kind of norm
map and is called the corestriction: it arises from the standard resolution
. .
A → X = X (G, A) of the G-module A, which is also an acyclic resolution
of A as an H-module, i.e.
H n (H, A) = H n ((X )H ) .
(see §3 p.34). For n ≥ 0, we have for the G-module X n the norm map
NG/H : (X n )H →(X n )G . It obviously commutes with ∂, hence we have a
morphism of complexes
.
NG/H : (X )H −→(X )G . .
Taking homology groups of these complexes, we obtain canonical homomor-
phisms
cor H n n
G : H (H, A) −→ H (G, A).
For n = 0, this is the usual norm map
NG/H : AH −→ AG .
For two open subgroups F ⊆ H of G, the equation NG/H ◦ NH/F = NG/F
implies the transitivity
cor H F F
G ◦ cor H = cor G .

On the level of cochains the corestriction is explicitly given as follows. From

every right coset c = Hσ ∈ H\G, we choose a fixed representative c̄ ∈ c and
define the homomorphism
cor : C n (H, A) −→ C n (G, A),
by
c̄ −1 x(c̄σ0 cσ 0−1 , . . . , c̄σn cσ n−1 ).
X
(cor x)(σ0 , . . . , σn ) =
c∈H\G
cor x is again G-linear. In fact, if σ ∈ G, then c̄σ =: τσ cσ for some τσ ∈ H and

−1
σ (cor x)(σσ0 , . . . , σσn )
σ −1 c̄ −1 x(c̄σσ0 cσσ 0−1 , . . .)
X
=
c∈H\G
X −1
= cσ −1 τσ−1 x(τσ cσσ0 (cσ)σ 0 , . . .)
c∈H\G
c̄ −1 x(c̄σ0 cσ 0−1 , . . .).

X
=
c∈H\G
Obviously cor ◦ ∂ = ∂ ◦ cor , so we get a homomorphism

cor : H n (H, A) −→ H n (G, A).
It is functorial in A and commutes with the δ-homomorphism, which we will
see in a moment. By dimension shifting we see that it coincides with the core-
striction cor H
G constructed before: for each n
≥ 0, we have the commutative
diagrams
H 0 (H,120/.3 An ) δ
n
H n (H, A)
cor H
G
cor cor H
G
cor
δn
H 0 (G, An ) H n (G, A) ,
where the horizontal maps are surjective. The vertical arrows cor H
G and cor
are the norm NG/H for n = 0, hence coincide for all n 0.
≥
(1.5.2) Proposition. The maps σ∗ , inf, res, cor are functorial in the G-module
considered, and they commute with the δ-homomorphism.
Proof: The functoriality is seen already on the level of cochains. We show

the commutativity of the corestriction with δ, leaving the other cases to the
reader. Let
07456 i j
A00 A A0 0

be an exact sequence of G-modules. We then obtain a commutative exact

diagram
0 C n (G, A00 ) C n (G, A) C n (G, A0 ) 0
cor cor cor
∂ ∂ ∂
0 C n (H, A00 ) C n (H, A) C n (H, A0 ) 0
∂ ∂ ∂
0 C n+1 (G, A00 ) C n+1 (G, A) C n+1 (G, A0 ) 0
cor cor cor
0 C n+1 (H, A00 ) C n+1 (H, A) C n+1 (H, A0 ) 0
and from this, the commutativity of the diagram

=89:;<JBCDEFGHI>LMNOKA@?RSPQ
VUTW A0 )
H n (H, δ
H n+1 (H, A00 )
cor cor
δ
H n (G, A0 ) H n+1 (G, A00 ).
Namely, let a0 ∈ Z n (H, A0 ) be a cocycle in the class α0 ∈ H n (H, A0 ). If
a ∈ C n (H, A) is a pre-image of a0 , then ∂a is a cocycle in the class δα0 ∈
H n+1 (H, A00 ) and cor ∂a is a cocycle in the class cor δα0 . On the other hand,
cor ∂a = ∂cor a, and cor a is a pre-image of the cocycle cor a0 , which repre-
sents cor α0 ∈ H n (G, A0 ), so that ∂ cor a ∈ Z n+1 (G, A00 ) represents the class
δ cor α0 ∈ H n+1 (G, A00 ). Therefore we have δ ◦ cor = cor ◦ δ.
In the same way one proves δ ◦σ∗ = σ∗ ◦δ, δ ◦res = res ◦δ, δ ◦inf = inf ◦δ,
the latter if the sequence 0 → A00H → AH → A0H → 0 is also exact. 2
(1.5.3) Proposition. The maps σ∗ , inf, res, cor are compatible with the cup-
product as follows.
(i) σ∗ (α ∪ β) = σ∗ α ∪ σ∗ β
for α ∈ H p (H, A), β ∈ H q (H, B) and σ ∈ G.
(ii) inf (α ∪ β) = (inf α) ∪ (inf β)
for α ∈ H p (G/H, AH ), β ∈ H q (G/H, B H ), if H is a normal closed
subgroup of G.
(iii) res (α ∪ β) = (res α) ∪ (res β)
for α ∈ H p (G, A), β ∈ H q (G, B), if H is a closed subgroup of G.
(iv) cor (α ∪ res β) = (cor α) ∪ β
for α ∈ H p (H, A), β ∈ H q (G, B), if H is an open subgroup of G.

Proof. (i), (ii) and (iii) are seen at once on the level of cochains. (iv) is
equivalent to the commutativity of the diagram
\[ZYX A) × H q (H, B)
H p (H, ∪
H p+q (H, A ⊗ B)
cor res cor
∪
H p (G, A) × H q (G, B) H p+q (G, A ⊗ B).
We may by (1.2.5) assume that G is finite: apply lim to the diagram with G, H
−→ U
replaced by G/U, H/U , where U runs through the open normal subgroups U
contained in H. By dimension shifting, we may transform the above diagram
into the diagram
a`_^] Ap ) × Ĥ 0 (H, Bq )
Ĥ 0 (H, ∪
Ĥ 0 (H, Ap ⊗ Bq )
cor res cor
∪
Ĥ 0 (G, Ap ) × Ĥ 0 (G, Bq ) Ĥ 0 (G, Ap ⊗ Bq ),
which comes from the diagram
AbfdecH
p × BqH ⊗
(Ap ⊗ Bq )H
NG/H NG/H
⊗
AG G
p × Bq (Ap ⊗ Bq )G .
But this diagram is commutative:
2
X X
NG/H (a ⊗ b) = σa ⊗ σb = σa ⊗ b = NG/H (a) ⊗ b.
σ ∈G/H σ ∈G/H
The compatibilities of the maps σ∗ , inf, res, cor with each other are described
by the following propositions.
(1.5.4) Proposition. σ∗ commutes with inf, res, cor.
(1.5.5) Proposition. For two closed subgroups V ⊆ U ⊆ G, we have

U/V G/V G/V
(i) inf U ◦ res U/V = res G
U ◦ inf G ,
if V is normal in G,
G/V U/V U/V
(ii) inf G ◦ cor G/V = cor UG ◦ inf U ,
if V is normal and U is open in G.

All this can be seen directly on the level of cochains. Another useful formula
is stated in the following
(1.5.6) Proposition. If U, V are closed subgroups of G, with V open, and if

σ runs through a system of representatives of the (finite) double coset decom-
position
G = . U σV,
[
σ
then we have the double coset formula
−1
res G V
cor UU ∩σV σ ◦ σ∗ ◦ res VV ∩σ−1 U σ .
X
U ◦ cor G =
σ
Proof: Because of (1.2.5), (1.5.4) and (1.5.5), we may assume that G is a

finite group. By dimension shifting we are then reduced to show the formula
only for n = 0, i.e. on H 0 (V, A) = AV . In this case res becomes the inclusion,
cor the norm and σ∗ the map a 7→ σa. Therefore we have to prove the formula
X
NG/V a = NU/U ∩σV σ−1 (σa).
σ
For every σ, we choose a system τσ of left representatives of U/U ∩ σV σ −1 ,

i.e.
U = . τσ (U ∩ σV σ −1 ).
[
τσ
Then
G = . τσ σV,
[
σ,τσ
i.e. τσ σ runs through a system of representatives of G/V , and so

2
XX X
NG/V a = τσ σa = NU/U ∩σV σ−1 (σa).
σ τσ σ
(1.5.7) Corollary. If U is an open subgroup of G, then

cor UG ◦ res G
U = (G : U ).
If U is normal, then
res G U
U ◦ cor G = NG/U .
The first formula is trivial in dimension zero and follows for arbitrary dimen-
sion by dimension shifting. The second formula is the double coset formula
for the case V = U .

(1.5.8) Corollary. If G = U V , then the diagram

jihg A)
H n (V, res
H n (U ∩ V, A)
cor cor
res
H n (G, A) H n (U, A)
is commutative.
For a locally compact abelian group A, let A∨ denote the Pontryagin dual,
see §1, p.9. We have a canonical isomorphism
H 2 (G, ZZ)∨ ∼
= Gab
onto the abelianized group Gab = G/G0 , the quotient of G by the closure G0
of the commutator subgroup. For this, note that H n (G, Q) = 0 for n ≥ 1 (see
(1.6.2)(c)). Therefore the exact sequence 0 → ZZ → Q → Q/ZZ → 0 yields an
isomorphism
H 2 (G, ZZ) ∼
= H 1 (G, Q/ZZ) = Homcts (G, Q/ZZ) = (Gab )∨ ,
and, using Pontryagin duality,
H 2 (G, ZZ)∨ ∼
= (Gab )∨∨ = Gab .
For an open subgroup H of G, we may ask what homomorphism
Gab −→ H ab
is induced by the dual
cor ∨ : H 2 (G, ZZ)∨ −→ H 2 (H, ZZ)∨
of the corestriction. The answer is given by the transfer map (Verlagerung)
of group theory, which is defined as follows.
For each right coset c ∈ H\G, we choose a fixed representative c̄, 1̄ = 1, so
that c = H c̄. Then the transfer is the continuous homomorphism
Ver : Gab −→ H ab , σG0 7−→ c̄σcσ −1 H 0 .
Y
c∈H\G
(1.5.9) Proposition. The map cor ∨ : H 2 (G, ZZ)∨ → H 2 (H, ZZ)∨ induces the
transfer map
Ver : Gab −→ H ab .
If H is normal in G, the composite H ab → Gab → H ab is the norm NG/H .

Proof: On the homogeneous 1-cochains, the corestriction

cor : C 1 (H, Q/ZZ) → C 1 (G, Q/ZZ)
is given by
x(c̄σ0 cσ 0−1 , c̄σ1 cσ 1−1 ).
X
(cor x)(σ0 , σ1 ) =
c∈H\G
On the associated inhomogeneous cochains y(σ) = x(1, σ), it is thus given by

y(c̄σcσ −1 ).
X
(cor y)(σ) = (cor x)(1, σ) =
c∈H\G
Hence on the dual Gab of H 1 (G, Q/ZZ) = Z 1 (G, Q/ZZ) = Hom(Gab , Q/ZZ), it is
given by the transfer map.
Ver
Let H be normal in G. The composite H ab −→ Gab −→ H ab is dual to the
composite of
cor res
H 2 (H, ZZ) −→ H 2 (G, ZZ) −→ H 2 (H, ZZ),
which by (1.5.7) is res ◦ cor = NG/H . 2
If A is a G-module, then for every pair V ⊆ U of open subgroups, we have

two maps
Alk U
incl
AV ,
NU/V
the inclusion and the norm. For the cohomology groups, we have similarly
two maps
nm A) res H n (V, A),
H n (U, cor
and these satisfy the double coset formula (1.5.6). From this observation we
are led to a generalization of G-modules, which gives a conceptual explanation
of the double coset formula. We enlarge the totality of open subgroups U of
G as follows.
We consider the category B(G) of finite G-sets, i.e. of finite sets X with
a continuous action of G. For every open subgroup U of G, the quotient
G/U is a finite G-set by left multiplication. It is a connected, i.e. transitive,
G-set, and every connected finite G-set X is of this form. For, if x ∈ X and
Gx = {σ ∈ G | σx = x}, then G/Gx → X, σGx 7→ σx, is an isomorphism
of G-sets. The category B(G) has the advantage of containing, for every
two finite G-sets X, Y , the disjoint union X q Y (as categorical sum), and
for every pair of morphisms f : X → S, g : Y → S, the fibre product
X ×S Y = {(x, y) ∈ X × Y | f (x) = g(y)}.

(1.5.10) Definition. A G-modulation is a pair of functors

A = (A∗ , A∗ ) : B(G) → Ab
into the category Ab of abelian groups, A∗ covariant, A∗ contravariant, such
that
A∗ (X) = A∗ (X) =: A(X)
for all X ∈ B(G), and that the following two conditions are satisfied:
(i) A(X q Y ) = A(X)⊕A(Y ).
(ii) If the left one of the diagrams
utsv g0 g 0∗
Xrqpo X0 A(X) A(X 0 )
f f0 and f∗ f∗0
g g∗
Y Y0 A(Y ) A(Y 0 )
is cartesian (i.e. X ∼
= X 0 ×Y 0 Y ), then the right one commutes. Here
ϕ∗ = A∗ (ϕ), ϕ∗ = A∗ (ϕ) for a morphism ϕ in B(G). ∗)
In this form the G-modulations were introduced by A. DRESS [43] under the
name “Mackey functors” (they were defined earlier by A. GREEN under still
another name, see [59]).
By condition (i), a G-modulation A is completely determined by its restric-
tion to the full subcategory B0 (G) of the G-sets G/U , where U runs through
the open subgroups of G. One then writes A(U ) in place of A(G/U ). Every
morphism in B0 (G) is in a unique way the composite of a projection
πUV : G/V −→ G/U (V ⊆ U)
and a “conjugation”
c(σ) : G/U −→ G/σU σ −1 , τ U 7−→ τ σ −1 (σU σ −1 ).
We set
res UV = A∗ (πUV ) : A(U ) −→ A(V ),
ind VU = A∗ (πUV ) : A(V ) −→ A(U ),
σ ∗ = A∗ (c(σ)) : A(σU σ −1 ) −→ A(U ),
σ∗ = A∗ (c(σ)) : A(U ) −→ A(σU σ −1 ).
∗) Clearly, this notion of G-modulations extends to G-modulations A : B(G) → A with
values in an arbitrary abelian category A.

(1.5.11) Proposition (Double Coset Formula). Let U, V ⊆ W be open sub-

groups of G and let R be a system of representatives of U \W/V . Then, for
every G-modulation A, we have the formula
−1
res W V
ind UU ∩σV σ ◦ σ∗ ◦ res VV ∩σ−1 U σ .
X
U ◦ ind W =
σ ∈R
Proof: Let X = G/U, Y = G/V, S = G/W . For the fibre product X ×S Y

we have the orbit decomposition
X× Y = . C ,
[
S σ
σ ∈R
U
where Cσ is the G-orbit of the element (U, σV ) ∈ X ×S Y . Let f = πW ,
V
g = πW and let p : X ×S Y → X, q : X ×S Y → Y be the projections and
pσ = p |Cσ , qσ = q |Cσ . Then by the properties (i) and (ii) in (1.5.10)
f ∗ ◦ g∗ = pσ∗ ◦ qσ∗ .
X
(∗)
σ ∈R
The isotropy groups of the elements (U, σV ) and (σ −1 U, V ) of Cσ are U ∩

σV σ −1 and σ −1 U σ ∩ V , and we have the commutative diagram
G/U ∩ yzxw{}| σV σ −1
c(σ)
G/σ −1 U σ ∩ V
χ
πU Cσ πV
pσ qσ
G/U G/V.
V
In the formula (∗) we have f ∗ = res WU , g∗ = ind W , pσ∗ = πU ∗ ◦ c(σ)∗ ◦ χ∗ =
−1
−1
ind UU ∩σV σ ◦ σ∗ ◦ χ∗ and qσ∗ = (πV ◦ χ−1 )∗ = (χ−1 )∗ ◦ πV∗ = χ∗−1 ◦ res VV ∩σ−1 U σ .
This gives the desired result. 2
Recall that a G-modulation A is completely determined by its restriction to

B0 (G), since every finite G-set is the disjoint union of connected G-sets, and
every connected G-set is isomorphic to a set G/U . Conversely, we have the
(1.5.12) Proposition. Let A = (A∗ , A∗ ) : B0 (G) → Ab be a pair of functors,

A∗ covariant, A∗ contravariant, such that A∗ (U ) = A∗ (U ) := A(U ). Assume
that the double coset formula (1.5.11) holds and that moreover σ ∗ ◦ σ∗ = id for
every σ ∈ G. Then A extends uniquely to a G-modulation on B(G).

One obtains the extension of A to B(G) as follows. LetYX ∈ B(G) be

an arbitrary finite G-set. We let G act on the group AX = A(Gx ) by the
x∈X
conjugation σx : A(Gx ) → A(Gσx ), and we set
A(X) = HomG (X, AX ),
i.e. A(X) is the group of all G-equivariant maps X → AX . The proof that this
becomes a G-modulation is left to the reader.
In order to give a G-modulation A, it thus suffices to define an abelian group
A(U ) for every open subgroup U of G and to establish, for every σ ∈ G and
every pair V ⊆ U of open subgroups, the maps
σ∗
~ )
A(U A(σU σ −1 ), A(V )
ind
A(U ),
σ∗ res
which yield the functors A∗ and A∗ , and then to verify the double coset formula.
It is clear what is meant by a morphism A → B of G-modulations. So the
G-modulations form a category, which we denote by Mod(G). We mention
the following examples of G-modulations.
Example 1: G-modules. Let M be a G-module. For every open subgroup U

we set M (U ) = M (G/U ) = M U . For σ ∈ G we define the maps σ ∗ and σ∗ by
a 7→ σ −1 a and a 7→ σa, and for every pair V ⊆ U we define res UV and ind VU
as the inclusion M U ,→ M V and the norm M V → M U , a 7→ NU/V (a) =
Q
σ ∈U/V σa (M is written multiplicatively). In this way, every G-module M
becomes a G-modulation, denoted again by M , and we obtain an embedding
i : Mod(G) −→ Mod(G)
of the category Mod(G) of G-modules. With this embedding, Mod(G) becomes
the full subcategory of Mod(G) of the G-modulations A with Galois descent,
meaning that for every pair V E U of open subgroups the restriction
A(U ) −→ A(V )U/V
is an isomorphism.
We note that, in particular, every abelian group A gives rise to a constant
G-modulation U 7→ A(U ) = A with the maps σ ∗ = σ∗ = id, res UV = id and
ind VU = (U : V ).
The embedding i : Mod(G) → Mod(G) has as left adjoint the functor
Mod(G) → Mod(G), A 7→ Ā = lim A(U ), i.e. for every G-module M we
−→ U
have
HomMod(G) (A, iM ) = HomM od(G) (Ā, M ).

Example 2: Cohomology. For every G-module M and every n ≥ 0, the

cohomology H n yields a G-modulation
A : B0 (G) −→ Ab, G/U 7→ A(U ) = H n (U, M ),
if we choose for res UV and ind VU the cohomological restriction res UV and core-
striction cor VU , and for σ ∗ , σ∗ the cohomological conjugations (σ −1 )∗ , σ∗ .
Example 3: The fundamental G-modulation π ab . We consider the map

B0 (G) −→ Ab, G/U 7−→ π ab (U ) := U ab ,
which associates to every open subgroup U of G the abelianized group U ab =
U/U 0 , where U 0 is the closure of the commutator subgroup of U . For σ ∈ G the
maps σ∗ : U ab → (σU σ −1 )ab and σ ∗ : U ab → (σ −1 U σ)ab are the conjugations
x 7→ σxσ −1 and x 7→ σ −1 xσ, and for a pair V ⊆ U the map ind VU is induced by
the inclusion V ,→ U , whereas res UV is given by the transfer Ver : U ab → V ab .
Example 4: The representation ring. We consider all finite dimensional

complex representations V of the profinite group G, i.e. all continuous homo-
morphisms G → GL(V), where V is any finite dimensional C-vector space.
Such representations have finite images. Let R(G) be the set of isomorphism
classes {V} of such representations V. We define an addition in R(G) by
{V} + {V 0 } := {V ⊕ V 0 }.
Then R(G) becomes a commutative monoid. From R(G) we obtain an additive
group Rep(G) by setting
Rep(G) = (R(G) × R(G))/ ∼,
where the equivalence relation ∼ is defined by
({V}, {V 0 }) ∼ ({W}, {W 0 }) ⇐⇒ {V} + {W 0 } = {W} + {V 0 }.
R(G) becomes a submonoid of Rep(G) by identifying {V} with ({V}, O).
We may even turn Rep(G) into a ring if we define the multiplication by
{V}{W} = {V ⊗ W}. This ring is called the representation ring of G.
Forgetting the ring structure, we obtain a G-modulation
Rep : B0 (G) −→ Ab, G/U 7−→ Rep(U ),
as follows. Let σ ∈ G and let U be an open subgroup of G. If U → GL(V) is
a representation of U , then the composite with the conjugation σU σ −1 → U ,
σuσ −1 7→ u, gives a representation σU σ −1 → GL(V). This assignment
extends to an isomorphism σ∗ : Rep(U ) → Rep(σU σ −1 ), and we set σ ∗ = σ∗−1 .
For a pair V ⊆ U of open subgroups, the map
res UV : Rep(U ) → Rep(V )

is obtained by restricting a representation U → GL(V) to V − this is a

homomorphism (even a ring homomorphism). The map
ind VU : Rep(V ) −→ Rep(U )
is obtained by associating to a representation V of V the induced representation
ind VU (V) of U . The underlying vector space consists of all continuous maps
f : U → V such that f (τ σ) = τ f (σ) for τ ∈ V, σ ∈ U . The action of σ ∈ U on
ind VU (V) is given by f 7→ σf, (σf )(x) = f (xσ). In this way we obtain in fact
a G-modulation, because the double coset formula is in this case a theorem of
I. MACKEY. This brought about the name Mackey functor (see [43]).
For every G-modulation A, the groups A(U ) are in a canonical way topolog-
ical groups: a basis of neighbourhoods of 0 is given by the groups ind VU A(V )
for V ⊆ U . It is easy to see that the maps σ ∗ , σ∗ , res , ind are continuous. A
is called quasi-compact, Hausdorff, compact etc. if all the groups A(U ) have
the corresponding property. A is Hausdorff if V ⊆U ind VU A(V ) = 0 for all U .
T
A is compact if it is quasi-compact and Hausdorff, and this is equivalent to

being profinite, i.e. all A(U ) are profinite groups.
To every G-modulation A there is associated a submodulation NA, called
the modulation of “universal norms”, which is given by
ind VU A(V ).
\
NA(U ) =
V ⊆U
The fact that, for V U , the homomorphisms A(V

⊆ ) ind
A(U ) induce
res
homomorphisms NA(V ) ind
NA(U ) is trivial for ind and follows for res from
res
the double coset formula. The quotient A/NA is a Hausdorff G-modulation
(see exercise 6).
For further results on G-modulations we refer to [147], [161], [235].
Exercise 1. For n ≥ 1 we have lim H n (U, A) = 0, where U runs over the open subgroups
−→ U
of G and the limit is taken over the restriction maps res : H n (U, A) → H n (V, A), V ⊆ U.
Exercise 2. If H is a normal subgroup of G and A a G-module, then res G
H is a homomorphism
res G
H : H n
(G, A) −→ H n
(H, A) G/H
,
and if moreover H is open, then cor HG yields a homomorphism
cor H
G : H n
(H, A)G/H −→ H n (G, A).
Exercise 3. Let H be an open subgroup of G, A a G-module and A → X = X (G, A) the. .

res
standard resolution of A. Then the restriction H n (G, A) −→ H n (H, A) is obtained by taking
the homology of the restriction map
(∗) . res
(X )G −→(X )H . .
Hint: By (1.3.9), we obtain isomorphisms of δ-functors
.
H n (G, A) = H n (H 0 (G, X )), H n (H, A) ∼
= H n (H 0 (H, X )). .
ι
(∗) induces a functorial map H n (G, A) → H n (H, A), which commutes with the δ-homomor-
phism. It coincides with res for n = 0 and for n > 0 by dimension shifting.

Exercise 4. Consider a diagram

1 A Ĝ G 1
f ϕ̂ ϕ
1 A0 Ĝ0 G0 1
with exact rows, where G, G are profinite groups and A (resp. A0 ) is a G-module (resp. a
0
G0 -module). Consider moreover the maps

ϕ∗ f∗
H 2 (G0 , A0 ) −→ H 2 (G, A0 ) ←− H 2 (G, A)
ϕ id id f
given by the pairs (G → G0 , A0 → A0 ) and (G → G, A → A0 ). Let u0 ∈ H 2 (G0 , A0 ), u ∈
H 2 (G, A) be the classes belonging to the upper and lower group extension respectively.
(i) Show that the diagram of solid arrows can be commutatively completed by an arrow
ϕ̂ : Ĝ → Ĝ0 if and only if f is a G-homomorphism (G acting on A0 via ϕ) and
ϕ∗ (u0 ) = f∗ (u).
(ii) Two such arrows ϕ̂1 , ϕ̂2 are called equivalent if there exists an a0 ∈ A0 , such that
ϕ̂2 (σ̂) = a0 ϕ̂1 (σ̂)a0−1 for all σ̂ ∈ Ĝ.
1 0
Show that the set of equivalence classes is an H (G, A ) -torsor.
Exercise 5. Let U, V be open subgroups of G, and let A be a U -module. If V ⊆ U , then
A is also a V -module, which we denote by ResU V A. For σ ∈ G we denote by σA the
σU σ −1 -module, whose underlying abelian group is A and the action of τ ∈ σU σ −1 is given
by a 7→ σ −1 τ σa.
Show that for any two open subgroups U, V and any U -module A, we have an isomorphism
of V -modules
−1
ResG IndV A ∼
M
U = IndU ∩σV σ σResV −1 A,
G U V ∩σ Uσ
σ ∈R
where R is a set of representatives of U \G/V and where the modules Ind A are defined below
on p.61. In particular, if U is a normal open subgroup of G, then
ResG IndU A ∼
M
= UσA. G
σ ∈G/U
Exercise 6. Let A be a G-modulation and let

\
NA(U ) = ind VU A(V ).
V ⊆U
Show that U 7→ NA(U ) is a submodulation of A, and that the modulation A/NA is Hausdorff.
Show that we get also a G-modulation Â, the “completion of A”, by setting
Â(U ) = lim A(U )/NU/V AU .
←−
V ⊆U
Exercise 7. For any two G-modulations A, B we have G-modulations A⊗B and Hom(A, B).
In particular, we have the notion of a “dual” A∗ = Hom(A, Q/ZZ) of a G-modulation A.
Exercise 8. Let X be a finite G-set. A complex vector bundle on X is a continuous
representation G → GL(V) on a finite dimensional C-vector space V such that the projection
X × V → X is a morphism of G-sets. The vector bundle is called a line bundle if dim V = 1.
Define an abelian group Pic(X) of isomorphism classes of line bundles on X. Show that
X 7→ Pic(X) is actually a G-modulation Pic, and show that Pic = Hom(π ab , Q/ZZ).

§6. Basic Properties
We collect in this section some basic properties of cohomology groups,

which will be used repeatedly. If we write Ĥ n (G, A) in the following for
profinite groups G, then for n ≥ 1 this means, as for finite groups, Ĥ n (G, A) =
H n (G, A). From the formula (1.5.7), cor UG ◦ res G
U = (G : U ), follows the
(1.6.1) Proposition. Let G be a profinite group and let U be an open subgroup.

Assume that G is finite or that n ≥ 1. Then for every discrete G-module A
such that Ĥ n (U, A) = 0, we have
(G : U )Ĥ n (G, A) = 0.
In particular, if G is finite, then Ĥ n (G, A) is annihilated by the order of G. If,
moreover, A is finitely generated as a ZZ-module, then Ĥ n (G, A) is finite.
We conclude that for arbitrary profinite groups G the cohomology groups

H n (G, A), n ≥ 1, are torsion groups, since by (1.2.5)
H n (G, A) = lim
−→
H n (G/U, AU ),
U
where U runs through the open normal subgroups of G.
(1.6.2) Proposition. Let G be a finite group and let A be a G-module. Assume

that multiplication by p is an automorphism of A for all prime numbers p | #G.
Then
Ĥ i (G, A) = 0 for all i ∈ ZZ.
If G is profinite, this remains true for i ≥ 1. In particular, A is a cohomologi-
cally trivial G-module in the following cases:
a) A is a torsion group whose supernatural order is prime to #G,
b) A is an abelian profinite group whose supernatural order is prime to #G,
c) A is uniquely divisible.
m
Proof: Let us first assume that G is finite. Putting m = #G, the map A → A
is an automorphism of A by assumption. It therefore induces an isomorphism
m : Ĥ i (G, A) −→
∼ Ĥ i (G, A)
which, by (1.6.1), implies that Ĥ i (G, A) = 0.

Now let G be profinite. We have H i (G, A) = lim H i (G/U, AU ) for i ≥ 1.
−→ U
Let U ⊆ G be a normal open subgroup and m = #(G/U ). By assumption, the
m-multiplication map m : A → A is an isomorphism. Taking U -invariants, we

§6. Basic Properties 61
see that the m-multiplication map m : AU → AU is also an isomorphism. By

the first part of the proof, we obtain H i (G/U, AU ) = 0 for all i ≥ 1, showing
that H i (G, A) = 0 for all i ≥ 1. Finally note that the assumption on the
p-multiplication map for all p | #G is satisfied in the cases a)–c), and that
everything remains true after replacing G by a closed subgroup. This finishes
the proof. 2
Now let H be a closed normal subgroup of G. If A is a discrete G-module,

then the cohomology group H n (H, A) is a discrete G-module, too: each σ ∈ G
acts on it by conjugation σ∗ .
(1.6.3) Proposition. The closed normal subgroup H acts trivially on the

cohomology group H n (H, A), i.e. H n (H, A) is a G/H-module. In particular,
the conjugation σ∗ : H n (G, A) → H n (G, A) is the identity for all σ ∈ G.
Proof: The assertion is trivial for H 0 (H, A) = AH . For n > 0 it follows by

dimension shifting from the commutative diagram (see (1.3.8))
Ĥ 0 (H, An ) δn
H n (H, A)
σ∗ σ∗
δn
Ĥ 0 (H, An ) H n (H, A).
2
Now let H be an arbitrary closed subgroup of G. For every H-module A,

we consider the G-module
M = IndHG (A)
consisting of all continuous maps x : G → A such that x(τ σ) = τ x(σ) for all
τ ∈ H. The action of ρ ∈ G on M is given by x(σ) 7→ (ρx)(σ) = x(σρ). The
module M is said to be obtained by inducing A from H to G. ∗)
∗) The terminology induced module is commonly used, but strictly speaking it is slightly
inaccurate. From a categorical point of view the situation is as follows. Given a pair H ⊆ G
of abstract groups, the forgetful functor Res: G-Mod→H-Mod admits the left adjoint functor
Ind: H-Mod→G-Mod, A 7→ ZZ[G] ⊗ ZZ [H] A, and the right adjoint functor Coind: H-Mod→
G-Mod, A 7→MapH (G, A). If H has finite index in G, both functors are isomorphic. However,
with respect to the groups, the functor Ind is covariant while Coind is contravariant. This
phenomenon can be viewed as the reason for the existence of cor . In the case of profinite
groups, we have the functor Coind on discrete modules, but we write IndH G A for Coind(A).
Furthermore, we have the functor Ind (“compact induction”) on compact modules .

We have a canonical projection

π : IndH
G (A) −→ A, x 7→ x(1).
This is a homomorphism of H-modules. If H is of finite index in G, it maps
the H-submodule A0 = {x : G → A | x(τ ) = 0 for all τ ∈/ H} isomorphically
onto A. We identify A0 with A. When σ1 , . . . , σn is a system of representatives
of G/H, then
n
M
M= σi A .
i=1
If A is a G-module, then IndHG (A) is canonically isomorphic to the G-module

Map(G/H, A) of all continuous functions y : G/H → A, where the action of
ρ ∈ G on y is given by (ρy)(σH) = ρy(ρ−1 σH). The isomorphism
IndH (A) ∼
G= Map(G/H, A)
is given by x(σ) 7→ y(σH) = σx(σ −1 ). In particular, when H = 1 we have a
canonical isomorphism
IndH ∼
G (A) = IndG (A)
with the G-module IndG (A) = Map(G, A) on which G acts by (ρx)(σ) =
ρx(ρ−1 σ) (see §3). We have seen in (1.3.7) that
H n (G, IndG (A)) = 0 for all n > 0.
We generalize this fact with the following proposition, commonly cited as
Shapiro’s lemma.
(1.6.4) Proposition. Let H be a closed subgroup of G and let A be an H-

module. Then, for all n ≥ 0, we have a canonical isomorphism
sh : H n (G, IndH ∼ n
G (A)) −→ H (H, A).
Proof: The groups H n (G, IndH

G (A)) are the homology groups of the complex
.
X (G, IndH
G (A))G
. The canonical homomorphism
π : IndH
G (A) −→ A , x 7→ x(1),
of H-modules induces an isomorphism
H 0 (G, IndH ∼ 0
G (A)) −→ H (H, A) ,
and for each n ≥ 0, a homomorphism
X n (G, IndH G n H
G (A)) −→ X (G, A) .
This is actually an isomorphism, since it has as inverse the map which as-
sociates to a function y(σ0 , . . . , σn ) in X n (G, A) the function x(σ0 , . . . , σn )(σ)
= y(σσ0 , . . . , σσn ). This is readily checked. We thus have
H n (G, IndH (A)) ∼
G
.
= H n (X (G, A)H ).

The complex 0 → A → X 0 (G, A) → X 1 (G, A) → · · · is an acyclic resolution

of the H-module A, since, by the remark following (1.3.8), the X n (G, A) are
induced G-modules. Therefore, by (1.3.9), we obtain a canonical isomorphism
.
H n (X (G, A)H ) ∼
= H n (H, A). 2
Remark: If G is finite, the same argument yields isomorphisms

Ĥ n (G, IndH (A)) ∼
= Ĥ n (H, A)
G
for all n ∈ ZZ.
If A is a G-module, then we have an injective G-homomorphism

i : A −→ IndH
G (A), (ia)(σ) = σa .
If, moreover, H is an open subgroup of G, then we have a G-homomorphism
ν : IndH σx(σ −1 ),
X
G (A) −→ A, ν(x) =
σ ∈G/H
where σ runs through a system of representatives of G/H. ∗) By this and by

the lemma of Shapiro, we get the following interpretation of the restriction and
the corestriction.
(1.6.5) Proposition. We have commutative diagrams

H (A))
H n (G, Ind sh
H n (H, A) H n (G, IndH sh
H n (H, A)
G G (A))
ν∗ cor i∗ res
H n (G, A) H n (G, A), H n (G, A) H n (G, A).
Proof: For the restriction this is obvious, so we prove it only for the corestric-
tion. By dimension shifting, it suffices to consider the case n = 1. For each
class c ∈ H \ G, let c̄ ∈ c be a fixed representative. Let x ∈ Z 1 (G, IndHG (A)).
∗) These homomorphisms are given by the Frobenius reciprocity. On the one hand we
have the isomorphism
HomG (A, IndH ∼ G
G (B)) = HomH (ResH (A), B) ,
where H is a closed subgroup of G, A is a G-module and B is an H-module. Let B = ResG H (A);
then i : A → IndH
G Res G
H (A) is the unit of the adjunction Res a Ind. If, moreover, H is an open
subgroup of G, then we have the isomorphism
HomG (IndH ∼ G
G (B), A) = HomH (B, ResH (A))
(where IndH
L
G (B) is identified with σ ∈G/H σB, see the footnote on p.61). Thus we obtain
H G
ν : IndG ResH (A) → A as the counit of the adjunction Ind a Res.

By definition, sh maps the class of x to the class of the cocycle y = sh(x) ∈

Z 1 (H, A) given by
y(σ0 , σ1 ) = x(σ0 , σ1 )(1).
Noting that the action of σ ∈ G on f ∈ IndH G (A) is given by (σf )(τ ) = f (τ σ),
we obtain
cor sh(x)(σ0 , σ1 ) = c̄−1 x(c̄σ0 cσ 0−1 , c̄σ1 cσ 1−1 )(1)
X
c̄
c̄−1 x(σ0 cσ 0−1 , σ1 cσ 1−1 )(c̄).

X
=
c̄
On the other hand
c̄−1 x(σ0 , σ1 )(c̄).
X
ν∗ (x)(σ0 , σ1 ) =
c̄
Since x is a cocycle, we have
x(σ0 cσ 0−1 , σ1 cσ 1−1 ) = x(σ0 cσ 0−1 , σ0 ) + x(σ0 , σ1 cσ 1−1 )
= x(σ0 cσ 0−1 , σ0 ) + x(σ0 , σ1 ) + x(σ1 , σ1 cσ 1−1 )
= x(σ0 cσ 0−1 , σ0 ) − x(σ1 cσ 1−1 , σ1 ) + x(σ0 , σ1 ).
We therefore have to show that the function
c̄−1 [x(cσ −1 c̄−1 [x(cσ −1
X X
f (σ0 , σ1 ) = 0 , 1)(c̄σ0 )] − 1 , 1)(c̄σ1 )]
c̄ c̄
is a coboundary. Noting that x(cσ −1

i , 1) ∈ IndH
G (A) and c̄σi cσ −1
i ∈ H, we have
c̄−1 [x(cσ −1
X
i , 1)(c̄σi )]
c̄
c̄−1 c̄σi cσ i−1 [x(cσ i−1 , 1)(cσ i σi−1 c̄−1 c̄σi )]

X
=
c̄
σi cσ i−1 [x(cσ i−1 , 1)(cσ i )] = σi

X X
= c̄−1 x(c̄−1 , 1)(c̄)
c̄ c̄
since cσi runs through H\G as c does. This shows that f (σ0 , σ1 ) = σ1 a − σ0 a
where a = − c̄−1 x(c̄−1 , 1)(c̄), so is a coboundary. 2
P
Proposition (1.6.5) (as well as (1.6.4)) will follow without computation from
a general uniqueness theorem for δ-functors, which we will prove in II §6 (see
(2.6.3)). Here, it says that ν∗ and cor ◦ sh are morphisms between the δ-
functors H n (G, IndH n
G (−)) and H (H, −) which coincide for n = 0 and hence
for all n. In the same way one can show the following fact:
If H is a normal closed subgroup of G and A is a G-module, then we have
a further action of G on IndH −1
G (A) given by x 7→ σ∗ x, (σ∗ x)(ρ) = σx(σ ρ).
H
σ∗ is even an automorphism of the G-module IndG (A) (not only of the abelian
group), and is the identity for σ ∈ H. It therefore induces an automorphism on

the cohomology group H n (G, IndH G (A)), which in this way becomes a G/H-
module. Considering the G/H-action on H n (H, A) given by conjugation (see
§5), the map
sh : H n (G, IndH n
G (A)) −→ H (H, A)
becomes an isomorphism of G/H-modules.
Besides the maps inf, res, cor, we sometimes have to consider still another,
more subtle map, whose meaning is best understood in the framework of
spectral sequences (see II §1), but which can explicitly be defined as follows.
(1.6.6) Proposition. Let H be a normal subgroup of G and A a G-module.

Then there is a canonical homomorphism
tg : H 1 (H, A)G/H −→ H 2 (G/H, AH ),
called transgression, which is given as follows. If x : H → A is an inho-
mogeneous 1-cocycle in a class [x] ∈ H 1 (H, A)G/H , there exists a 1-cochain
y : G → A such that y|H = x and that (∂y)(σ1 , σ2 ) is contained in AH and
depends only on the cosets σ1 H, σ2 H, i.e. may be regarded as a cocycle of
G/H. For each such cochain y,
tg[x] = [∂y].
Proof: We construct a cochain y : G → A with the following properties

(i) y|H = x,
(ii) y(στ ) = y(σ) + σy(τ ) for σ ∈ G, τ ∈ H,
(iii) y(τ σ) = y(τ ) + τ y(σ) for σ ∈ G, τ ∈ H.
Let s : G/H → G, γ 7→ sγ, be a continuous section of the projection G →
G/H such that s1 = 1. Such a section exists by ex.4 of §1. Since [x] is
invariant under every γ ∈ G/H, we have
(1) sγx((sγ)−1 τ sγ) − x(τ ) = τ y(sγ) − y(sγ)
for an element y(sγ) ∈ A. We may assume that y(1) = 0 and that γ 7→ y(sγ) is
continuous. In fact, there exists an open normal subgroup U of G such that x(τ )
depends only on the cosets τ (H ∩ U ) and is contained in AU . Therefore the left
side takes the same value for all elements sγ in a coset mod U . So we may
choose for y(sγ) the same value within a coset of G mod U , and this means
that y(sγ) is continuous as a function of γ. For an arbitrary σ = sγτ ∈ G, we
now set
y(σ) = y(sγ) + sγx(τ ).

Then y|H = x. Let τ, τ 0 ∈ H, g = sγ and σ = gτ 0 . Then

y(στ ) = y(gτ 0 τ ) = y(g) + gx(τ 0 τ )
= y(g) + gx(τ 0 ) + gτ 0 x(τ ) = y(σ) + σx(τ ).
This proves (ii). Using (1), we obtain
y(τ g) = y(gτ g ) = y(g) + gx(τ g )
= y(g) + x(τ ) + τ y(g) − y(g) = x(τ ) + τ y(g),
and for arbitrary σ = gτ 0 ,
y(τ σ) = y(τ gτ 0 ) = y(τ g) + τ gx(τ 0 )
= x(τ ) + τ y(g) + τ gx(τ 0 )
= y(τ ) + τ y(σ).
This proves (iii).
The function
∂y(σ1 , σ2 ) = σ1 y(σ2 ) − y(σ1 σ2 ) + y(σ1 )
depends only on the cosets σ1 H, σ2 H, i.e. ∂y(σ1 , σ2 τ ) and ∂y(σ1 τ, σ2 ) are
independent of τ ∈ H. In fact,
∂y(σ1 , σ2 τ ) = σ1 y(σ2 τ ) − y(σ1 σ2 τ ) + y(σ1 )
= σ1 σ2 y(τ ) + σ1 y(σ2 ) − y(σ1 σ2 τ ) + y(σ1 )
= σ1 y(σ2 ) − y(σ1 σ2 ) + y(σ1 ) = ∂y(σ1 , σ2 ),
∂y(σ1 τ, σ2 ) = σ1 τ y(σ2 ) − y(σ1 τ σ2 ) + y(σ1 τ )
= σ1 y(τ σ2 ) − σ1 y(τ ) − y(σ1 τ σ2 ) + y(σ1 τ )
= σ1 y(τ σ2 ) − y(σ1 τ σ2 ) + y(σ1 )
= ∂y(σ1 , τ σ2 ) = ∂y(σ1 , σ2 τ σ2 ) = ∂y(σ1 , σ2 ).
From ∂∂y = 0 and ∂y(τ, σ) = ∂y(1, σ) = 0, we now obtain
τ ∂y(σ1 , σ2 ) = ∂y(τ σ1 , σ2 ) − ∂y(τ, σ1 σ2 ) + ∂y(τ, σ1 ) = ∂y(σ1 , σ2 ),
i.e. ∂y(σ1 , σ2 ) ∈ AH .
Finally, let y 0 : G → A be any cochain such that y 0 |H = x and that ∂y 0 (σ1 , σ2 )
takes values in AH and depends only on the cosets σ1 H, σ2 H. Then, for the
function z = y − y 0 , we have z(τ ) = 0 for τ ∈ H and ∂z(τ, σ) = ∂z(1, τ σ) = 0,
i.e. τ z(σ) − z(σ) = 0, so that z(σ) ∈ AH , and ∂z(1, τ σ) = ∂z(1, σ), so
that z(τ σ) = z(σ). Therefore ∂y and ∂y 0 may be viewed as cocycles in
Z 2 (G/H, AH ) and differ by the coboundary ∂z ∈ B 2 (G/H, AH ), i.e. define
the same cohomology class in H 2 (G/H, AH ). Thus defining tg[x] = [∂y]
gives a well-defined homomorphism. 2

(1.6.7) Proposition (Five Term Exact Sequence). Let H be a closed normal

subgroup of G and let A be a G-module. We then have an exact sequence
inf res
0 −→ H 1 (G/H, AH ) −→ H 1 (G, A) −→ H 1 (H, A)G/H
tg inf
−→ H 2 (G/H, AH ) −→ H 2 (G, A).
Moreover, if H i (H, A) = 0 for i = 1, . . . , n − 1, we have an exact sequence
inf res
0 −→ H n (G/H, AH ) −→ H n (G, A) −→ H n (H, A)G/H
tg inf
−→ H n+1 (G/H, AH ) −→ H n+1 (G, A).
Proof: Consider the first sequence. The image of res is contained in

res
H 1 (H, A)G/H , since the map H 1 (G, A) −→ H 1 (H, A) is a homomorphism
of G-modules and H 1 (G, A)G = H 1 (G, A) by (1.6.3).
Exactness at H 1 (G/H, AH ). For the injectivity of the first map inf , let x :
G/H → AH be an inhomogeneous 1-cocycle such that the composite inf x :
G → G/H → A is a coboundary, (inf x)(σ) = σa − a. For all τ ∈ H, we
have σa − a = στ a − a, hence a ∈ AH . Therefore x(σH) = σHa − a is a
1-coboundary.
Exactness at H 1 (G, A). Let x : G/H → AH be an inhomogeneous 1-cocycle.
Then for τ ∈ H,
(res ◦ inf x)(τ ) = (inf x)(τ ) = x(τ H) = x(H) = x(1) = 0,
i.e. im(inf ) ⊆ ker(res ). Conversely, let x : G → A be an inhomogeneous
1-cocycle such that res x is a coboundary, i.e. x(τ ) = τ a − a for τ ∈ H. The
1-cocycle x0 (σ) = x(σ) − (σa − a) of G defines the same cohomology class as
x and satisfies x0 (τ ) = 0 for τ ∈ H. Hence
x0 (στ ) = x0 (σ) + σx0 (τ ) = x0 (σ),
and also
x0 (τ σ) = x0 (τ ) + τ x0 (σ) = τ x0 (σ).
We now define y : G/H → A by y(σH) = x0 (σ). Then y(σH) ∈ AH , because
y(σH) = y(τ σH) = τ y(σH) for all τ ∈ H, and we obtain a 1-cocycle with
inf y = x0 . This shows ker(res ) ⊆ im(inf ).
Exactness at H 1 (H, A)G/H . If y ∈ Z 1 (G, A) and x = res y represents a class
[x] in H 1 (H, A)G/H , then tg[x] = [∂y] = 0, i.e. im(res ) ⊆ ker(tg). Conversely,
let x ∈ Z 1 (H, A) represent a class [x] ∈ H 1 (H, A)G/H such that tg[x] = 0. Let
y ∈ C 1 (G, A) be a cochain as in (1.6.6). Viewing ∂y as a 2-cocycle of G/H,
then [∂y] = tg[x] = 0, hence ∂y = ∂z, where z ∈ C 1 (G/H, AH ). Viewing y
and z as functions on G, we have y − z ∈ Z 1 (G, A). Since res (y − z) and
res y = x are 1-cocycles of H, so is res z, and since z is constant on H, we

have res z = 0. Thus res (y − z) = res y = x, i.e. [x] = res [y − z]. This proves
ker(tg) ⊆ im(res ).
Exactness at H 2 (G/H, AH ). Let x ∈ Z 1 (H, A) be an inhomogeneous cocycle,
representing a class [x] ∈ H 1 (H, A)G/H . Then, by (1.6.6), there is a cocycle z ∈
Z 2 (G/H, AH ) such that inf z = ∂y and tg[x] = [z]. Thus inf z ∈ B 2 (G, A),
hence inf tg[x] = [inf z] = 0, showing im(tg) ⊆ ker(inf ). Conversely, let
z ∈ Z 2 (G/H, AH ) be a normalized cocycle, i.e. z(1, σ) = z(σ, 1) = 0 (see
p.19), such that inf [z] = [inf z] = 0. Then inf z = ∂y with y ∈ C 1 (G, A).
Setting x = res y we have ∂x = res ∂y = res inf z = 0 and, regarding ∂y as the
2-cocycle z of G/H, tg[x] = [∂y] = [z]. This proves ker(inf ) ⊆ im(tg).
The exact sequence
0 −→ H n (G/H, AH ) −→ H n (G, A) −→ H n (H, A)G/H
tg
−→ H n+1 (G/H, AH ) −→ H n+1 (G, A)
for G-modules A such that H i (H, A) = 0 for i = 1, . . . , n − 1, is obtained by
induction. We have it for n = 1, and we assume it for n ≥ 1. Let A be a
G-module, such that H i (H, A) = 0 for i = 1, . . . , n. Consider the sequence
0 −→ AH −→ ĀH −→ AH
1 −→ 0,
where Ā = IndG (A), which is exact since H 1 (H, A) = 0. We therefore have a

commutative diagram
0¦¥¢¡ «ª©¨§£¤
inf res
H n (G/H, AH
1 ) H n (G, A1 ) H n (H, A1 )G/H
δ δ δ
inf res
0 H n+1 (G/H, AH ) H n+1 (G, A) H n+1 (H, A)G/H
tg inf
H n+1 (G/H, AH
1 ) H n+1 (G, A1 )
δ δ
tg inf
H n+2 (G/H, AH ) H n+2 (G, A),
where the δ’s are isomorphisms, since Ā is G-induced and H-induced, and ĀH
is G/H-induced. The lower map tg is defined by the upper one. Furthermore,
H i (H, A1 ) ∼
= H i+1 (H, A) = 0 for i = 1, . . . , n − 1.
Therefore, by assumption, the upper sequence is exact, hence also the lower
one. 2
Let p be a prime number. A profinite group G is called a pro-p-group if its

supernatural order is a p-power, i.e. for every open normal subgroup U of G,
the finite group G/U is a p-group. Equivalently, a pro-p-group is the projective
limit of finite p-groups.

(1.6.8) Definition. A p-Sylow subgroup of a profinite group G is a closed

subgroup Gp which is a pro-p-group such that the index (G : Gp ) is prime to p.
The Sylow theorems for finite groups hold as well for profinite groups.
(1.6.9) Theorem. Let G be a profinite group and p a prime number.

(i) There exists a p-Sylow subgroup Gp .
(ii) Every pro-p-subgroup is contained in a p-Sylow subgroup.
(iii) The p-Sylow subgroups of G are conjugate.
Proof: Let U run through the open normal subgroups of G and let Σp (U )
denote the finite, nonempty set of all p-Sylow subgroups of G/U . If V ⊆ U are
two open normal subgroups, then the projection G/V → G/U maps p-Sylow
subgroups onto p-Sylow subgroups and induces a surjection Σp (V ) → Σp (U ),
so that the Σp (U ) form a projective system of nonempty finite sets. The
projective limit lim Σp (U ) is a compact, nonempty topological space (see
←−
[160], chap.IV).
(i) Now let (SU ) ∈ lim Σp (U ). For V ⊆ U we have the surjective projection
←−
SV → SU , i.e. the system (SU ) is a projective system of finite p-groups. The
projective limit Gp = lim SU is then a p-Sylow subgroup of G by definition.
←−
(ii) Let H be a pro-p-subgroup of G and let HU be its image under G → G/U .

Let Σ0p (U ) be the set of p-Sylow subgroups of G/U , containing HU . Again,
lim
←−
Σ0p (U ) is nonempty. Let (SU ) ∈ lim
←−
Σ0p (U ). Then the inclusions HU ,→
SU form a morphism of projective systems, and we obtain
H = lim HU ⊆ lim SU = Gp .
←− ←−
(iii) Let Gp and G0p be two p-Sylow subgroups of G and let SU and SU0 be
their images in G/U . Let C(U ) be the set of elements σU ∈ G/U such that
σU SU σU−1 = SU0 . The C(U ) form again a projective system of finite, nonempty
sets. The projective limit lim
←−
C(U ) is nonempty. If σ = (σU ) ∈ lim
←−
C(U ) ⊆ G,
then clearly σGp σ −1 = G0p . 2
The p-primary part A(p) of an abelian torsion group A is the subgroup

L
consisting of all elements of A of p-power order. We have A = p A(p). For
the p-primary part of the torsion group Ĥ n (G, A), we have the

(1.6.10) Proposition. Let A be a discrete G-module and let Gp a p-Sylow

subgroup of G. Assume that G is finite or that n ≥ 1. Then the homomorphism
res : Ĥ n (G, A)(p) −→ Ĥ n (Gp , A)
is injective, and if Gp is open in G, the homomorphism
cor : Ĥ n (Gp , A) −→ Ĥ n (G, A)(p)
is surjective.
Proof: The proposition holds for any closed subgroup H of G of index

(G : H) prime to p. Indeed, if H is open, then by (1.5.7), cor ◦ res = (G : H).
Since (G : H) is prime to p, the map
Ĥ n (G,¬ A)(p) cor ◦ res Ĥ n (G, A)(p)
is an automorphism, so that res must be injective and cor surjective. If H is
not open, the injectivity of res follows from (1.5.1). 2
(1.6.11) Corollary. Assume that G is finite or that n ≥ 1. If Ĥ n (Gp , A) = 0

for all prime numbers p, then
Ĥ n (G, A) = 0.
Because of the above proposition, we are often reduced to the cohomology

of pro-p-groups. Its most frequently used property is the assertion (1.6.13)
which we first prove for finite p-groups.
(1.6.12) Proposition. Let G be a finite p-group and let A be a p-primary

G-module. If H 0 (G, A) = 0 or H0 (G, A) = 0, then A = 0.
Proof: For the proof of H 0 (G, A) = 0 ⇒ A = 0, we may assume that A

is finite, since every element of the p-primary G-module A generates a finite
G-module. A r AG is a disjoint union of G-orbits Ga not consisting of only
one point, hence #Ga ≡ 0 mod p and #A ≡ #AG mod p. If AG = 0, then
#A ≡ 1 mod p, and hence A = 0. If H0 (G, A) = AG = 0, then, setting
A∗ = Hom(A, Qp /ZZp ), we get (A∗ )G ∼ = (AG )∗ = 0, whence A∗ = 0 and A = 0.
2
Recall that a G-module A =/ 0 is said to be simple if it does not contain any

submodule other than 0 and A itself. Since the set {ga | g ∈ G} ⊆ A is finite
for every a ∈ A, a simple module is finitely generated as an abelian group.

p
Therefore there exists a prime number p such that the p-multiplication A −→ A
is not surjective, whence pA = 0. We conclude that a simple module is finite
and there exists a unique prime number p such that pA = 0.
(1.6.13) Corollary. Let G be a pro-p-group. Then every discrete simple

p-primary G-module A is isomorphic to ZZ/pZZ (with trivial G-action). In
particular, if A is a p-primary G-module, then
A = 0 if and only if AG = 0 .
Proof: Let A be a discrete simple p-primary G-module. Then pA = 0.

Further, AU =/ 0 for some open normal subgroup U ⊆ G, and by (1.6.12) we
obtain 0 =/ H 0 (G/U, AU ) = AG ⊆ A. Hence A = AG by simplicity. We
conclude that A is an IFp -vector space with trivial G-action and therefore of
dimension 1. 2
(1.6.14) Proposition.
(i) The maximal closed subgroups of a pro-p-group G are normal of index p.
(ii) A homomorphism G → G0 of pro-p-groups is surjective if and only if the

induced homomorphism H 1 (G0 , ZZ/pZZ) → H 1 (G, ZZ/pZZ) is injective. In
particular,
G = 1 ⇔ H 1 (G, ZZ/pZZ) = 0.
Proof: (i) Let H be a maximal closed subgroup of G. Then there exists

an open normal subgroup U of G such that H U/U =/ G/U , since otherwise
H = G. Clearly, H U/U is a maximal subgroup of the finite group G/U and
is therefore normal of index p. This is a well-known result in group theory
which follows from the first Sylow theorem for finite groups (see [73], chap.4,
(4.2.2)). Since H is maximal, it is the pre-image of H U/U under G → G/U
and is thus also normal of index p.
(ii) If G → G0 is surjective, then obviously H 1 (G0 , ZZ/pZZ) −→ H 1 (G, ZZ/pZZ)
is injective. Conversely, assume the latter. If G → G0 were not surjective, the
image would be contained in a maximal subgroup H of G0 which is normal of
index p. The composition G0 → G0 /H ∼ = ZZ/pZZ would be an element χ =/ 0 in
H 1 (G0 , ZZ/pZZ) which becomes zero in H 1 (G, ZZ/pZZ), a contradiction. 2


φ : G0 −→ G
be a homomorphism of pro-p-groups. Let A be ZZ/pr ZZ, r ∈ IN, or Qp /ZZp
with trivial action. Then φ is an isomorphism if and only if the induced
homomorphism
H i (φ, A) : H i (G, A) −→ H i (G0 , A)
is an isomorphism for i = 1 and injective for i = 2.
Proof: Let us show the nontrivial direction. ZZ/pZZ is a subgroup of A in a

canonical way, and we obtain, for any group H, an inclusion
H 1 (H, ZZ/pZZ) ⊂ H 1 (H, A).
In particular, H 1 (φ, ZZ/pZZ) is injective if H 1 (φ, A) is, and therefore (1.6.14)
implies that φ is surjective. Put K = ker(φ) and consider the 5-term exact
sequence
H 1 (φ) H 2 (φ)
0 → H 1 (G, A) → H 1 (G0 , A) → H 1 (K, A)G → H 2 (G, A) → H 2 (G0 , A).
Our assumptions imply H 1 (K, A)G = 0, thus H 1 (K, A) = 0 by (1.6.13). We
conclude that H 1 (K, ZZ/pZZ) = 0, and K = 0 by (1.6.14)(ii). 2
(1.6.16) Proposition. Let G and G0 be pro-p-groups and assume that Gab

and (G0 )ab are torsion groups. Let pr be a common exponent of the factor
commutator groups and assume that H 2 (G, Qp /ZZp ) = 0 = H 2 (G0 , Qp /ZZp ).
Let φ : G0 → G be a homomorphism. Then the following assertions are
equivalent.
(i) φ is an isomorphism.
(ii) H 1 (φ) : H 1 (G, ZZ/pr ZZ) → H 1 (G0 , ZZ/pr ZZ) is an isomorphism.
(iii) H 2 (φ) : H 2 (G, ZZ/pr ZZ) → H 2 (G0 , ZZ/pr ZZ) is an isomorphism.
Proof: Clearly, (i) implies (ii) and (iii) and, by (1.6.15), (ii) and (iii) to-
gether imply (i). It remains to show that (ii) and (iii) are equivalent. Since
H 2 (G, Qp /ZZp ) = 0, the exact sequence
pr
0 −→ ZZ/pr ZZ −→ Qp /ZZp −→ Qp /ZZp → 0
induces the four term exact sequence
α β
0 → H 1 (G, ZZ/pr ZZ) → H 1 (G, Qp /ZZp ) →
γ
H 1 (G, Qp /ZZp ) → H 2 (G, ZZ/pr ZZ) → 0.

By the assumption on Gab , the map α is an isomorphism. Hence β is zero and

γ is an isomorphism. The same argument also applies to G0 and therefore (ii)
and (iii) are both equivalent to
(iv) H 1 (φ, Qp /ZZp ) : H 1 (G, Qp /ZZp ) → H 1 (G0 , Qp /ZZp ) is an isomorphism.
This concludes the proof. 2
Exercise 1. Let G be a profinite group, H a closed normal subgroup, A a discrete G-module,

and let s : G/H → G be a continuous section of the projection G → G/H. Let x : H → A
be a cocycle representing a class [x] ∈ H 1 (H, A)G/H , and set y(σ) = sγx(τ ) for σ = sγτ ,
γ ∈ G/H, τ ∈ H.
Show that if H acts trivially on A, then y : G → A is a cochain as in (1.6.6), i.e. y|H = x and
∂y(σ1 , σ2 ) depends only on the cosets σ1 H, σ2 H, so that tg[x] = [∂y].
Exercise 2. Let G be finite, H a normal subgroup, A a G-module, and let H 0 be the commutator
subgroup of H. The group extension
1 −→ H ab −→ G/H 0 −→ G/H −→ 1
defines a class u ∈ H 2 (G/H, H ab ).
Assume that H acts trivially on A. Then H 1 (H, A)G/H = H 0 (G/H, Hom (H ab , A)), and
the cup-product
∪
H 2 (G/H, H ab ) × H 0 (G/H, Hom (H ab , A)) −→ H 2 (G/H, A)
yields a homomorphism
u∪ : H 1 (H, A)G/H −→ H 2 (G/H, A).
Show that this homomorphism coincides with −tg. Generalize this result to profinite groups.
Hint: Let x be a cocycle representing a class [x] ∈ H 1 (H, A)G/H . Then xσ = x for all σ ∈ G.
Let y be the function y(σ) = sγx(τ ) (σ = sγτ ) as in ex.1. The class u ∈ H 2 (G/H, H ab )
is represented by the cocycle τ (γ1 , γ2 ) = s̄γ1 s̄γ2 s̄(γ1 γ2 )−1 , where s̄ is the composite of
s
G/H → G → G/H 0 . Show that
∂y(γ1 , γ2 ) = −x(τ (γ1 , γ2 )).
Exercise 3. Let G be a finite group, H a normal subgroup, and let A be a G-module. Consider
the exact sequence
0 −→ A −→ Ā −→ A1 −→ 0
with Ā = IndG (A), the induced G-module. The associated long exact cohomology sequence
yields the exact sequence
δ
0 −→ AH −→ ĀH −→ AH 1
1 −→ H (H, A) −→ 0.
We split this up into the two exact sequences
(1) 0 −→ AH −→ ĀH −→ B −→ 0,
(2) 0 −→ B −→ AH 1
1 −→ H (H, A) −→ 0,
where B denotes the image of ĀH in AH
1 . Since Ā
H
is an induced G/H-module, we get
from (1)
δ2 H
H 1 (G/H, B) −→ 2
∼ H (G/H, A ).

Show that the composite of

δ δ
H 1 (H, A)G/H −→
1
H 1 (G/H, B) −→
2 2 H
∼ H (G/H, A ),
where δ1 is obtained from (2), is the transgression.
Exercise 4. Define the cohomology groups of a pair H ⊆ G and a G-module A by

H 0 (G, H, A) = 0 and H n (G, H, A) = H n−1 (G, Γ(A)) for n ≥ 1,
where Γ(A) is defined by the exact sequence
0 −→ A −→ IndH
G (A) −→ Γ(A) −→ 0.
We then have a “relative exact cohomology sequence” (see [178])

δ
. . . → H n (G, A) → H n (H, A) → H n+1 (G, H, A) → H n+1 (G, A) → H n+1 (H, A) → . . . .
Exercise 5. For two pairs H ⊆ G and L ⊆ K and a homomorphism ϕ : K → G such that

ϕL ⊆ H, we have functorial maps
ϕn : H n (G, H, A) −→ H n (K, L, A).
§7. Cohomology of Cyclic Groups
In this section we will compute the cohomology of cyclic groups and we

will introduce the concept of the Herbrand quotient. Let G be a finite cyclic
group. Recall that for every G-module A we have
Ĥ 0 (G, A) = AG /NG A and Ĥ −1 (G, A) = NG A/IG A.
If σ is a generator of the cyclic group G, then IG A = (σ − 1)A, since for all

i ≥ 1 we have the equality
σ i − 1 = (σ − 1)(σ i−1 + · · · + σ + 1).
(1.7.1) Proposition. Let G be a finite cyclic group. Then the group Ĥ 2 (G, ZZ)
is cyclic of the same order as G. Let χ ∈ Ĥ 2 (G, ZZ) be any generator. Then the
cup-product induces isomorphisms
χ∪ : Ĥ n (G, A) −→
∼ Ĥ n+2 (G, A)
for all n ∈ ZZ and every G-module A. In particular, we have isomorphisms
Ĥ 2n (G, A) ∼
= AG /NG A, Ĥ 2n−1 (G, A) ∼
= NG A/IG A
for all G-modules A and all n ∈ ZZ.

§7. Cohomology of Cyclic Groups 75
Proof: Let σ ∈ G be a generator and let N = #G. Consider the four-term

exact sequence
(∗) ±°¯® 0 ZZ
µ
ZZ[G] σ−1 ZZ[G] ε ZZ 0
where ε is the augmentation map ai σ i 7→ ai and µ(a) = a(1 + σ + · · · +
P P
σ N −1 ). It induces an isomorphism (see the remark following (1.3.8))

δ 2 : Ĥ 0 (G, ZZ) −→
∼ Ĥ 2 (G, Z
Z).
Since
Ĥ 0 (G, ZZ) = ZZ/(1 + σ + · · · + σ N −1 )ZZ = ZZ/N ZZ,
we see that Ĥ 2 (G, ZZ) is cyclic of order N and every generator is of the form
χ = δ 2 (m), m ∈ (ZZ/N ZZ)× .
Now let A be a G-module. Since all objects in (∗) are ZZ-free, it remains
exact when tensored by A. Hence for every n ∈ ZZ, we obtain an isomorphism
∼ Ĥ n+2 (G, A), which fits into the commutative diagram
δ 2 : Ĥ n (G, A) →
µ´²³ A)
Ĥ n (G, Ĥ n (G, A)
m χ∪
δ2
Ĥ n (G, A) Ĥ n+2 (G, A).
It remains to show that multiplication by m induces an automorphism on
Ĥ n (G, A). But this is clear, because by (1.6.1), Ĥ n (G, A) is an abelian group
which is annihilated by N = #G, hence a ZZ/N ZZ-module. 2
(1.7.2) Proposition. Let G be a finite cyclic group. If 0 → A → B → C → 0

is an exact sequence of G-modules, then we have an exact hexagon
¶·¸¹º» Ĥ 0 (G, A) Ĥ 0 (G, B)
Ĥ −1 (G, C) Ĥ 0 (G, C)
Ĥ −1 (G, B) Ĥ −1 (G, A).
Proof: All maps in the hexagon above are the canonical ones, except the ho-
momorphism Ĥ 0 (G, C) → Ĥ −1 (G, A) which is the composite of the connect-
ing homomorphism Ĥ 0 (G, C) → Ĥ 1 (G, A) and the inverse of the isomorphism
Ĥ −1 (G, A) →
∼ Ĥ 1 (G, A) obtained in (1.7.1). In order to prove exactness, we

only have to show that the diagram

¼½¾¿ A)
Ĥ −1 (G, Ĥ −1 (G, B)
Ĥ 1 (G, A) Ĥ 1 (G, B)
commutes, where the vertical maps are the isomorphisms obtained in (1.7.1).
But this is obvious. 2
A very useful concept for calculating indices and orders of abelian groups
is the Herbrand quotient, given by
(1.7.3) Definition. Let A be an abelian group and let f, g be endomorphisms

of A such that f g = gf = 0, i.e. im g ⊆ ker f and im f ⊆ ker g. Then the
Herbrand quotient with respect to f and g is defined to be
(ker f : im g)
qf,g (A) =
(ker g : im f )
provided that both indices are finite.
The following special case is of great importance for the cohomology of

cyclic groups.
(1.7.4) Definition. If G is a finite cyclic group and A a G-module, then the

Herbrand quotient of A is defined to be
#Ĥ 0 (G, A)
h(G, A) = ,
#Ĥ −1 (G, A)
provided that both orders are finite.
If σ is a generator of the cyclic group G of order n, then the endomorphisms

D =σ−1 and NG = 1 + σ + . . . + σ n−1
have the property that D · NG = NG · D = 0. As Ĥ 0 (G, A) = AG /NG A and
Ĥ −1 (G, A) = NG A/IG A, we see that h(G, A) = qD,N (A).
The salient property of the Herbrand quotient is its multiplicativity.

(1.7.5) Proposition. Let G be a finite cyclic group. If 0 → A → B → C → 0

is an exact sequence of G-modules, then
h(G, B) = h(G, A) · h(G, C)
in the sense that, whenever two of these quotients are defined, so is the third
and the identity holds.
Proof: Let f : Ĥ 0 (G, A) → Ĥ 0 (G, B). Using (1.7.2), we obtain an exact

sequence
0 → ker f → Ĥ 0 (G, A) → Ĥ 0 (G, B) → . . . → Ĥ −1 (G, C) → ker f → 0,
and therefore
#Ĥ 0 (G, A)·#Ĥ 0 (G, C)·#Ĥ −1 (G, B) = #Ĥ 0 (G, B)·#Ĥ −1 (G, A)·#Ĥ −1 (G, C).
At the same time, we see that if any two of the quotients are well-defined, then
so is the third. From the last equation we obtain the desired equality. 2
Another special case for the Herbrand quotient is the following: let f = 0
and let g be the multiplication by a natural number n. Then
(A : nA)
q0,n (A) = .
# nA
Considering A as a trivial G-module, where G is a cyclic group of or-
der n, we have q0,n (A) = h(G, A). In particular, for an exact sequence
0 → A → B → C → 0 of abelian groups, we obtain
q0,n (B) = q0,n (A) · q0,n (C),
again in the sense that whenever two of these quotients are defined, so is the
third and the identity holds.
(1.7.6) Proposition. Let A be a finite abelian group and let f, g be endomor-

phisms of A such that f g = gf = 0. Then
qf,g (A) = 1.
In particular, if G is a finite cyclic group and A is a finite G-module, then
h(G, A) = 1.
Proof: Since #A = # ker f · #im f = # ker g · #im g , the result follows. 2
From the last result and from (1.7.5) we obtain h(G, A) = h(G, B) for any
G-submodule B of A of finite index.

A procyclic group is a profinite group G which is topologically generated

by a single element σ, i.e. G is the closure of the subgroup (σ) = {σ n | n ∈ ZZ}.
For example,
ZZp = lim ZZ/pn ZZ and ẐZ = lim ZZ/nZZ
←− ←−
n∈IN n∈IN
are (additive) procyclic groups, and every procyclic group G is a quotient of

ẐZ (see [160], chap.IV, § 2, example 7). As
Y
ẐZ = ZZp ,
p
where p runs through the prime numbers, every procyclic group G is of the
form
G∼ ZZ/pep ZZ ×
Y Y
= ZZp ,
p∈T p∈S
where T and S are disjoint sets of prime numbers and ep , p ∈ T , are natural
numbers.
Q
We assume now that G is torsion-free, i.e. G = p∈S ZZp . Let IN(S) be the
set of natural numbers not divisible by prime numbers p ∈/ S. Then the group
Gn of n-th powers, n ∈ IN(S), are the open subgroups of G and thus
G = lim G/Gn ∼ = lim ZZ/nZZ.
←− ←−
n∈IN(S) n∈IN(S)
We say that an abelian group

[ X is S-divisible if X = nX for all n ∈ IN(S),
and is S-torsion if X = n X, where n X = {x ∈ X | nx = 0}.
n∈IN(S)
Q
(1.7.7) Proposition. Let G = p∈S ZZp be a torsion-free procyclic group and
let A be a discrete G-module.
(i) If A is S-torsion, then H 1 (G, A) ∼
= AG .
(ii) If A is torsion or S-divisible, then H n (G, A) = 0 for n ≥ 2.
σ−1
Proof: (i) Let σ be a topological generator of G. Then AG = ker(A −→ A)
σ−1
and AG = coker(A −→ A). Let Nn = 1 + σ + · · · + σ n−1 . Then
n
H 1 (G, A) = lim H 1 (G/Gn , AG )
−→
n∈IN(S)
∼
= lim NnA
Gn n
/(σ − 1)AG = A0 /(σ − 1)A,
−→
n∈IN(S)
where A0 = {a ∈ A | Nn a = 0 for some n ∈ IN(S)}. The isomorphism

H 1 (G, A) ∼
= A0 /(σ − 1)A

is given by associating to a 1-cocycle x : G → A the value x(σ). If A is

S-torsion, then A0 = A. In fact, for a ∈ A there exist n, m ∈ IN(S), such that
na = 0 and σ m a = a. From this, it follows that
(1 + σ + · · · + σ mn−1 )a = n(1 + σ + · · · + σ m−1 )a = 0,
i.e. a ∈ A0 . This proves (i).
(ii) By (1.7.1), we have
H 2 (G/Gn , AG ) ∼
n n
= AG /Nn AG .
A careful analysis of the definition of this isomorphism shows that the in-
n nm
flation map H 2 (G/Gn , AG ) → H 2 (G/Gnm , AG ) corresponds to the homo-
morphism n nm
AG /Nn AG −→ AG /Nnm AG ,
given as multiplication by m. If A is finite and m is a multiple of the order of
A, then this homomorphism is zero, hence
n
H 2 (G, A) = lim H 2 (G/Gn , AG ) = 0.
−→
If A is torsion, then A = lim Aα , where Aα runs through the finite G-submo-
−→
dules of A, hence H 2 (G, A) = lim
−→
H 2 (G, Aα ) = 0.
Assume now inductively that H n (G, A) = 0, n ≥ 2, for all torsion modules
A. For any torsion module A, consider the exact sequence
0 −→ A −→ IndG (A) −→ A1 −→ 0,
where the left arrow associates to a ∈ A the constant function x(τ ) = a.
Clearly, as A is torsion, so are IndG (A) and A1 (note that every continuous map
x : G → A has finite image). By (1.3.7), IndG (A) is cohomologically trivial,
so that
H n+1 (G, A) ∼
= H n (G, A1 ) = 0.
Now let A be S-divisible and let m ∈ IN(S). From the exact sequence
m
0 −→ m A −→ A −→ A −→ 0 ,
we get the exact cohomology sequence
m
H n (G, m A) −→ H n (G, A) −→ H n (G, A).
Since m A is torsion, H n (G, m A) = 0 for n ≥ 2, i.e. multiplication by m is
injective on H n (G, A) for n ≥ 2. But these groups are S-torsion by (1.6.1),
which finishes the proof. 2
Exercise 1. Let f and g be two commuting endomorphisms of an abelian group A. Show that
q0,gf (A) = q0,g (A) · q0,f (A),
provided all quotients are defined.

Exercise 2. Let G be a cyclic group of prime order p, and let A be a G-module such that
q0,p (A) is defined. Show that
h(G, A)p−1 = q0,p (AG )p /q0,p (A).
Hint: Let σ be a generator of G. Show that in the ring ZZ[G]/ZZNG = ∼ ZZ[ζ], ζ a primitive
p-th root of unity, one has p = (σ − 1)p−1 ε, where ε is a unit in ZZ[G]/ZZNG . Then use the
exact sequence 0 → AG → A → Aσ−1 → 0.
§8. Cohomological Triviality
Let G be a profinite group. For every prime number p, let Gp be a p-Sylow

subgroup of G. We have called a discrete G-module A cohomologically trivial
if H n (H, A) = 0 for all n > 0 and all closed subgroups H of G. We have seen
(cf. (1.3.7)) that induced G-modules are cohomologically trivial. We will give
now further criteria for cohomological triviality.
(1.8.1) Proposition. A discrete G-module A is cohomologically trivial if and

only if for every prime number p it is a cohomologically trivial Gp -module.
Proof: This follows easily from (1.6.11). 2
(1.8.2) Proposition. A discrete G-module A is cohomologically trivial if

and only if for every open normal subgroup U of G, the G/U -module AU is
cohomologically trivial.
Proof: If the AU are cohomologically trivial G/U -modules, then by (1.5.1),

H n (H, A) = lim H n (HU/U, AU ) = 0
−→
U
for n > 0 and every closed subgroup H, i.e. A is a cohomologically trivial

G-module. For a closed subgroup H/U of G/U , the sequence
0 −→ H n (H/U, AU ) −→ H n (H, A) −→ H n (U, A)
is exact if H i (U, A) = 0 for i = 1, . . . , n − 1 (see (1.6.7)). If A is cohomologi-
cally trivial, this is true for all n > 0 and we get H n (H/U, AU ) = 0 for n > 0,
showing that AU is a cohomologically trivial G/U -module. 2

§8. Cohomological Triviality 81
(1.8.3) Proposition. Let G be a finite p-group and let A be a p-primary G-

module. If pA = 0 and if Ĥ q (G, A) = 0 for one q, then A is an induced
G-module and, in particular, cohomologically trivial.
Proof: Let Λ = IFp [G], I a basis of the IFp -vector space AG and V =
L
I Λ.
In the exact sequence of G-modules
0 −→ Hom (A/AG , V ) −→ Hom (A, V ) −→ Hom(AG , V ) −→ 0 ,
B = Hom(A/AG , V ) is an induced G-module by (1.3.6)(iii), so that we have
H 1 (G, B) = 0, and the homomorphism
HomG (A, V ) −→ HomG (AG , V ) = Hom(AG , V G )
is surjective. We have canonically ΛG ∼
= IFp , hence an isomorphism AG ∼
= V G,
which by the above argument extends to a G-homomorphism
j : A −→ V.
By (1.6.12), the equality ker(j|AG ) = ker(j)G = 0 implies that j is injective . If
C is the cokernel of j, then we have an exact sequence
0 −→ AG −→ ∼ V G −→ C G −→ H 1 (G, A) −→ 0.
Hence if H 1 (G, A) = 0, then C G = 0 and thus C = 0, i.e. A ∼
= V is an induced
G-module. If Ĥ q (G, A) = 0 for some q, then H 1 (G, Aq−1 ) ∼
= Ĥ q (G, A) = 0,
i.e., by the argument as above, Aq−1 is an induced G-module. But this implies
H 1 (G, A) = Ĥ 2−q (G, Aq−1 ) = 0, and hence A is induced. 2
(1.8.4) Proposition. Let G be a finite group and let A be a G-module such

that for every prime number p there exists a dimension np ∈ ZZ with
Ĥ np (Gp , A) = Ĥ np +1 (Gp , A) = 0.
Then A is a cohomologically trivial G-module. If A is ZZ-free, it is a direct
summand of a free ZZ[G]-module. Conversely, for every cohomologically
trivial G-module A, we have
Ĥ n (H, A) = 0
for all n ∈ ZZ and all subgroups H of G.
Proof: Let
(1) 0 −→ R −→ F −→ A −→ 0
be an exact sequence with a free ZZ[G]-module F . We claim that, for every
prime number p, R/pR is an induced Gp -module. In fact, since F is an induced
Gp -module and Ĥ np (Gp , A) = Ĥ np +1 (Gp , A) = 0, we obtain
(2) Ĥ np +1 (Gp , R) = Ĥ np +2 (Gp , R) = 0,

and the exact sequence

p
(3) 0 −→ R −→ R −→ R/pR −→ 0
yields Ĥ np +1 (Gp , R/pR) = 0, giving the claim by (1.8.3).
Now we assume that A is ZZ-free. Then from (1) we obtain an exact sequence
0 −→ Hom(A, R) −→ Hom(A, F ) −→ Hom(A, A) −→ 0.
If we knew that H 1 (G, Hom(A, R)) = 0, then the homomorphism
HomG (A, F ) −→ HomG (A, A)
would be surjective, so that the identity map of A could be extended to a
G-embedding A ,→ F , making A a direct summand of F . We thus have to
show H 1 (G, M ) = 0 for M = Hom(A, R). Since A is ZZ-free, we get from (3)
an exact sequence
p
0 −→ M −→ M −→ Hom(A, R/pR) −→ 0,
i.e. M/pM ∼ = Hom(A, R/pR) is an induced Gp -module, since R/pR is an
induced Gp -module. From this it follows that
p
H 1 (Gp , M ) −→ H 1 (Gp , M )
is an isomorphism, whence H 1 (Gp , M ) = 0 for all p. By (1.6.11), we get
H 1 (G, M ) = 0, as required.
Now let A be an arbitrary G-module. Since R is ZZ-free, it follows from
(2) and from what we have just seen, that R is a direct summand of a free
ZZ[G]-module, i.e. of an induced G-module. Therefore R is cohomologically
trivial. As F is cohomologically trivial, so is A.
Conversely, we have seen that for every cohomologically trivial G-module
A and every surjection ε : F A with F free, the module R := ker(ε) is a
direct summand in a free module. Hence
Ĥ n (G, A) ∼
= Ĥ n+1 (G, R) = 0
for all n ∈ ZZ. 2
(1.8.5) Corollary. Let G be a finite group and let A, B be G-modules. If either

A is cohomologically trivial and B is divisible, or if B is cohomologically
trivial and A is ZZ-free, then Hom(A, B) is cohomologically trivial.
Proof: Let A be cohomologically trivial and B divisible. We consider an

exact sequence
0 −→ R −→ F −→ A −→ 0,
where F is a free ZZ[G]-module. Since F and A are cohomologically trivial,
so is R, and since R is ZZ-free, it is moreover a direct summand of a free
ZZ[G]-module, say F 0 . Therefore Hom(R, B) is a direct summand of the

§9. Tate Cohomology of Profinite Groups 83
induced G-module Hom(F 0 , B) and is thus cohomologically trivial. Since B

is divisible, i.e. an injective ZZ-module, the sequence
0 −→ Hom(A, B) −→ Hom(F, B) −→ Hom(R, B) −→ 0
is exact. The last two modules are cohomologically trivial, hence also the
module Hom(A, B).
If B is cohomologically trivial and A is ZZ-free, the same argument applies
if we exchange the roles of A and B. 2
Exercise 1. Let G be a finite group and let A, B be G-modules. Show that A ⊗ B is

cohomologically trivial if A is cohomologically trivial and either A or B is without p-torsion
for all primes p | #G.
Exercise 2. Let f : A → B be a homomorphism of G-modules and let fn : H n (H, A)
→ H n (H, B) be the induced homomorphisms on the cohomology (H ⊆ G). Assume, for
some q ∈ ZZ, that fq−1 is injective, fq is bijective, fq+1 is surjective, for all subgroups H of G.
Then fn is an isomorphism for all n ∈ ZZ and all subgroups H of G (see [124]).
§9. Tate Cohomology of Profinite Groups
In this section we extend the definition of the modified cohomology groups,

defined in §2 for finite groups, to profinite groups. We start with the definition
of homology of finite groups.
Let G be a finite group and let A be a G-module. Recall from §2 the cofixed
module AG = A/IG A, which is the largest quotient of A on which G acts
trivially. We consider the homological complex P• given by
Pn = ZZ[Gn+1 ],
and the differential ∂n : Pn → Pn−1 given by
n
(−1)i (σ0 , . . . , σi−1 , σi+1 , . . . , σn ).
X
∂n (σ0 , . . . , σn ) =
i=0
Moreover, we have the homomorphism ∂0 : P0 → ZZ, which is defined by
aσ σ 7→ aσ . Using the contracting homotopy given by Dn (σ0 , . . . , σn ) =
P P
(1, σ0 , . . . , σn ), one verifies in a similar way as in the proof of (1.2.1) that the
complex
2 ∂ 1 0 ∂ ∂
· · · −→ P2 −→ P1 −→ P0 −→ ZZ −→ 0

is exact. Moreover, this complex consists of torsion-free abelian groups and

therefore remains exact when tensored by any G-module A. Now we apply
the functor “cofixed module”. For n ≥ 0, we set
Cn (G, A) = (Pn ⊗ A)G
and we call the group
Hn (G, A) = Hn (C• (G, A))
the n-dimensional homology group of G with coefficients in A. From the
exact sequence above it follows that
H0 (G, A) = AG
which is consistent with our definition in §2. In contrast to cohomology, which
is contravariant, homology is covariantly functorial in the group. As a functor
in the module, homology is dual to cohomology in the following sense.
(1.9.1) Proposition. Let G be a finite group and let A be a G-module. If

A∗ = Hom(A, Q/ZZ), then we have a natural isomorphism
H i (G, A∗ ) ∼
= Hi (G, A)∗ , i ≥ 0.
Proof: Applying the exact functor Hom(−, Q/ZZ) to the exact sequence
∂ 2 ∂
1 0∂
· · · −→ P1 ⊗ A −→ P0 ⊗ A −→ A −→ 0,
we obtain the exact sequence
0 −→ A∗ −→(P0 ⊗ A)∗ −→(P1 ⊗ A)∗ −→ · · · ,
which, by (1.3.6)(iii), is an induced, hence acyclic, resolution of A∗ . By
(1.3.9), we obtain isomorphisms
H i (G, A∗ ) ∼
= H i (H 0 (G, (P• ⊗ A)∗ )) ∼
= Hi (H0 (G, P• ⊗ A))∗ ∼
= Hi (G, A)∗ .
2
All cohomological notions have their homological counterpart. For example,

for a normal subgroup U of G, we have the coinflation maps
coinf : Hn (G, A) −→ Hn (G/U, AU ),
which are induced by the projections ZZ[Gn+1 ] → ZZ[(G/U )n+1 ] and A → AU .
Recalling the modified homology group Ĥ0 (G, A) = NG A/IG A from §2, we
call the groups
NG A/IG A for n = 0,
(
Ĥn (G, A) =
Hn (G, A) for n ≥ 1
the modified homology groups. In §2 we identified the group Ĥ0 (G, A) with
Ĥ −1 (G, A). The following proposition extends this to arbitrary dimension.

(1.9.2) Proposition. Let G be a finite group and let A be a G-module. Then

we have natural isomorphisms
(
Ĥ n (G, A) for n ≥ 0,
Ĥ (G, A) ∼
n
=
Ĥ−n−1 (G, A) for n ≤ −1.
Proof: We have seen this for n ≥ −1 in §2. In order to deal with the remaining
dimensions, we consider the dual G-modules
(X−1−n )+ = Hom(X−1−n , ZZ)
for n ≥ 0. Let ((σ0 , . . . , σn )∗ )σ0 ,...,σn ∈G , be the dual basis of (X−1−n− )+ , i.e.
(σ0 , . . . , σn )∗ maps (σ0 , . . . , σn ) to 1 and all other basis elements of X−1−n =
ZZ[Gn+1 ] to zero. We consider the homomorphisms
ϕn : Pn −→(X−1−n )+
defined by sending (σ0 , . . . , σn ) to (σ0 , . . . , σn )∗ . A straightforward calculation
shows that the ϕn are G-homomorphisms and that the diagram
PÃÂÁÀ n
ϕn
(X−1−n )+
∂n (∂−n )+
n−1 ϕ
Pn−1 (X−n )+
commutes for n ≥ 1. Therefore, for n ≥ 1, Hn (G, A) is the cohomology of the
complex
((X• )+ ⊗ A)G
in dimension (−1 − n). Since the abelian groups Xi are free of finite rank, the
duality maps Di : (Xi )+ ⊗ A −→ Hom(Xi , A), given by Di (f ⊗ a)(x) = f (x)a,
are isomorphisms. These isomorphisms are easily seen to be G-invariant, and
we obtain an isomorphism of complexes
∼
D : ((X• )+ ⊗ A)G −→ Hom(X• , A)G .
Finally, by (1.3.6)(iii), the modules Hom(Xi , A) are induced. By (1.2.6), the
norm induces an isomorphism of complexes
∼
NG : Hom(X• , A)G −→ HomG (X• , A) = Ĉ • (G, A).
This finishes the proof. 2
Now let G be a profinite group and let A be a discrete G-module. We want

to define Tate cohomology groups Ĥ i (G, A), i ∈ ZZ. For i ≤ 0 this requires the
deflation map
def : Ĥ i (G/V, AV ) −→ Ĥ i (G/U, AU ), i ≤ 0,
for open normal subgroups V ⊆ U , which is defined as follows:

Consider, for j ≥ 1, the composition

coinf N
Hj (G/V, AV ) −→ Hj (G/U, (AV )U ) −→
∗
Hj (G/U, AU ),
where N∗ is the map which is induced on homology by the norm
NU/V : (AV )U −→ AU .
Via the identification Ĥ i = H−i−1 , this defines the deflation in dimension
i ≤ −2. In dimensions i = 0, −1, the map def is induced via the identifications
Ĥ 0 (G/U, AU ) ∼
= AG /NG/U AU and Ĥ −1 (G/U, AU ) ∼ = N AU /IG/U AU G/U
by the identity and the norm map, respectively.
(1.9.3) Definition (Tate cohomology). Let G be a profinite group and let A

be a discrete G-module. For i > 0, we set Ĥ i (G, A) = H i (G, A). For i ≤ 0,
we set
Ĥ i (G, A) = lim Ĥ i (G/U, AU ),
←−
U, def
Our next goal is a cup-product pairing for Tate cohomology. We start by

specifying a map on the chain level on the negative part of the complete standard
resolution which defines the deflation map on cohomology in dimension ≤ −2.
For a finite group G, let X•(G) denote the (homological) complete standard
resolution for G.
(1.9.4) Lemma. Let G be a finite group and let U ⊆ G be a normal subgroup.

(G/U ) (G)
In negative dimension, let the map α−i : X−i −→ X−i be given by
X
(σ0 U, . . . , σi−1 U ) 7−→ (σ0 τ0 , . . . , σi−1 τi−1 ),
τ0 ,...,τi−1 ∈U
where the τ0 , . . . , τi−1 on the right-hand side run independently through U .

Then in dimension ≤ −2 the deflation map is induced by a cocycle map
def : HomG (X•(G) , A) −→ HomG/U (X•(G/U ) , AU ),
which is uniquely defined by the commutative diagram
ÇÅÄÆ (G) , A)G
Hom(X−i
NG (G)
HomG (X−i , A)
(α∗−i ,(NU )∗ ) def
(G/U ) NG/U (G/U )

Hom(X−i , AU )G/U HomG/U (X−i , AU ).

Proof: First note that the map α−i commutes with the differential on the
negative part of the standard complex. Let ((σ0 , . . . , σi−1 )∗ )σ0 ,...,σi−1 ∈G denote
(G) + (G) (G/U )
the dual basis of (X−i ) = Hom(X−i , ZZ). The dual basis of (X−i )+ is
denoted by ((σ0 U, . . . , σi−1 U )∗ )σ0 U,...,σi−1 U ∈G/U . Consider the diagram
(G) +ÓÒÈËÊÉÑÏÎÍÌÐ Di (G) NG (G)

((X−i ) ⊗ A)G Hom(X−i , A)G HomG (X−i , A)
coinf (α∗−i ,pr∗ ) β
(G/U ) + Di (G/U ) NG/U (G/U )

((X−i ) ⊗ AU )G/U Hom(X−i , AU )G/U HomG/U (X−i , AU )
id⊗NU (id∗ ,(NU )∗ ) (id∗ ,(NU )∗ )
(G/U ) + Di (G/U ) NG/U (G/U )

((X−i ) ⊗ AU )G/U Hom(X−i , AU )G/U HomG/U (X−i , AU ).
Here, pr : A → AU is the natural projection, coinf is the map which is induced

(G) + (G/U )
by pr and by the map (X−i ) → (X−i )+ which sends (σ0 , . . . , σi−1 )∗ to
(σ0 U, . . . , σi−1 U )∗ . Via the identification
Hi−1 (G, A) = H −i ((X• )+ ⊗ A)G ,

i ≥ 2,
(see the proof of (1.9.2)) it induces coinflation on homology. The maps D−i are
induced by the canonical duality isomorphisms (X−i )+ ⊗ A ∼ = Hom(X−i , A).
The upper left square commutes and we define β in order to make the upper
right square commute. The lower squares obviously commute. By definition,
def is induced by the composition of the two vertical arrows on the left. This
completes the proof of the lemma. 2
(1.9.5) Proposition. Let i ≥ 2 and q ≥ 1. Let G be a finite group, U a normal

subgroup of G and let A and B be G-modules. Then
x̄ ∪ inf ȳ = inf (def x̄ ∪ ȳ) ∈ H q (G, A ⊗ B),
for x̄ ∈ Ĥ −i (G, A) and ȳ ∈ H q+i (G/U, B U ).
(G) (G/U )
Proof: Let x ∈ HomG (X−i , A) and y ∈ HomG/U (Xq+i , A) be cocycles
representing x̄ and ȳ, respectively. We calculate both sides on the level of
cochains. Let us start with the left-hand side. For (σ0 , . . . , σq ) ∈ Xq(G) we
obtain
X
(x ∪ inf y)(σ0 , . . . , σq ) = x(τ1 , . . . , τi ) ⊗ y(τi U, . . . , τ1 U, σ0 U, . . . , σq U ),
τ1 ,...,τi ∈G

see the definition of the maps ϕp,q in the proof of (1.4.7). In order to compute
(G)
the right-hand side, choose a z ∈ Hom(X−i , A) with x = NG z. Then, using
again the maps ϕp,q of (1.4.7), by (1.9.4) we have

∗
inf (def x ∪ y)(σ0 , . . . , σq ) = (NG/U (α−i , (NU )∗ )z) ∪ y (σ0 U, . . . , σq U ).
The right-hand side is equal to
∗
X
(NG/U (α−i , (NU )∗ )z)(τ1 U, . . . , τi U )⊗y(τi U, . . . , τ1 U, σ0 U, . . . , σq U ).
τ1 U,...,τi U ∈G/U
Using the definition of α−i and of NU this transforms to

X X X X
τ U ∈G/U u∈U τ1 U,...,τi U ∈G/U u1 ,...,ui ∈U
τ u z(τ −1 τ1 u1 , . . . , τ −1 τi ui ) ⊗ y(τi U, . . . , τ1 U, σ0 U, . . . , σq U ),
which coincides with
X
(NG z)(τ1 , . . . , τi ) ⊗ y(τi U, . . . , τ1 U, σ0 U, . . . , σq U ).
τ1 ,...,τi ∈G
This shows the proposition. 2
(1.9.6) Proposition. Let G be a finite group and let U be a normal subgroup

of G. Let A and B be G-modules. Then, for i ≤ 0 and q ≥ 1, the diagram
Ø×ÖÕÔ A)
Ĥ i (G, × H q−i (G, B) ∪
H q (G, A ⊗ B)
(∗) def inf inf
∪
Ĥ i (G/U, AU ) × H q−i (G/U, B U ) H q (G/U, (A ⊗ B)U )
commutes.
Proof: For i ≤ −2 this follows from (1.9.5). Via the isomorphism

NG A/IG A
∼
−→ Ĥ −1 (G, A)
a 7−→ xa : ZZ[G] → A, σ 7→ σ(a),
the cup product is given on the chain level by
X
xa ∪ y(σ0 , . . . , σq ) = σa ⊗ y(σ, σ0 , . . . , σq ).
σ ∈G
The commutativity of (∗) for i = −1 follows immediately, since def is induced
by the norm
U U
NU : NG A/IG A −→ NG/U A /IG/U A .
The case i = 0 is obvious. 2

Now let G be a profinite group and let A be a discrete G-module which

is finitely generated as a ZZ-module. The subgroup of elements of G that act
trivially on A is open, since for all a ∈ A the group Ga = {σ ∈ G | σa = a} is
open. Hence AU = A for U sufficiently small. In this case
Hom(A, B U ) = Hom(A, B)U
for any G-module B. By passing to the limit of the diagram (∗) of (1.9.6), we
obtain the
(1.9.7) Corollary. Let G be a profinite group and let A and B be discrete

G-modules. Then we have a cup-product pairing
Ĥ i (G, A) × H q−i (G, B) −→ H q (G, A ⊗ B)
for i ≤ 0 and q ≥ 1.
For an abelian profinite group A = lim Ai , Ai finite, and a prime number p,

←−
the p-part A(p) of A is defined by
A(p) = lim Ai (p).
←−
i
Let G be a profinite group and let A be a discrete G-module which is finitely

generated as a ZZ-module. By (1.6.1), Ĥ i (G/U, AU ) is finite for all i and all
open normal subgroups U of G. Therefore the abelian groups
Ĥ i (G, A) = lim Ĥ i (G/U, AU ), i ≤ 0,
←−
U, def
are naturally equipped with a profinite topology. For i ≥ 1, we equip the
abelian torsion groups Ĥ i (G, A) with the discrete topology.
(1.9.8) Lemma. Let G be a profinite group and let A be a discrete G-module

which is finitely generated as a ZZ-module. If p is a prime number with p∞ |#G,
then we have a canonical isomorphism
Ĥ 0 (G, A)(p) ∼
= AG ⊗ ZZp .
Proof: It suffices to show that for arbitrarily given n ∈ IN the subgroup

NG/U AU is contained in pn AG for sufficiently small U . Let U be an open
normal subgroup such that AU = A. Then for every normal open subgroup
V ⊆ U with pn |(U : V ) and for every a ∈ AV = AU = A, we have
NG/V a = NG/U NU/V a = (U : V )NG/U a ∈ pn AG .
This shows the lemma. 2


0 −→ A0 −→ A −→ A00 −→ 0
be a short exact sequence of discrete G-modules which are finitely generated
as ZZ-modules. Then there is an associated long exact cohomology sequence
· · · −→ Ĥ −n (G, A0 ) −→ Ĥ −n (G, A) −→ Ĥ −n (G, A00 ) −→ · · ·
ending with · · · → Ĥ 0 (G, A0 ) → Ĥ 0 (G, A) → Ĥ 0 (G, A00 ).
For every prime number p with p∞ |#G, we obtain a long exact sequence
· · · −→ Ĥ i (G, A0 )(p) −→ Ĥ i (G, A)(p) −→ Ĥ i (G, A00 )(p) −→ · · ·
which is unbounded in both directions. The groups are compact for i ≤ 0,
discrete for i > 0 and all homomorphisms are continuous.
Proof: Since AU = A for small U , we obtain for small U a long exact

sequence
· · · −→ Ĥ 0 (G/U, A0U ) −→ Ĥ 0 (G/U, AU ) −→ Ĥ 0 (G/U, A00U )
of finite abelian groups. Passing to the projective limit, we obtain the negative
part of our long exact sequence. Now let p be a prime number with p∞ |#G and
consider for small U the long exact sequence
0 −→ A0G −→ AG −→ A00G −→ Ĥ 1 (G/U, A0U ) −→ · · · .
Tensoring by ZZp and passing to the direct limit over U , we obtain the right-hand
part of our long exact sequence. The left-hand part is obtained by taking the
p-part of the long exact sequence which we obtained above. Both fit together
by (1.9.8). The continuity of the maps is clear from their definitions. 2
Let G be a profinite group and let A be a discrete G-module. For each pair
of open normal subgroups V ⊆ U of G, the exact sequences
(∗) 0 −→ NG/U AU −→ AG −→ Ĥ 0 (G/U, AU ) −→ 0,
(∗∗) 0 −→ IG/U AU −→ NG/U AU −→ Ĥ −1 (G/U, AU ) −→ 0
induce exact sequences
0 −→ NG A −→ AG −→ Ĥ 0 (G, A),
0 −→ IG A −→ NG A −→ Ĥ −1 (G, A),
where we have set \
NG A = lim NG/U AU = NG/U AU ,
←−
U U
U
NG A = lim NG/U A , IG A = lim IG/U AU .
←− ←−
U U

The group NG A is called the group of universal norms. The last exact
sequences are in general not exact on the right, unless we make assumptions
on the module. This motivates the following
(1.9.10) Definition. Let G be a profinite group. A level-compact G-module

is a discrete G-module A which is endowed with the following additional
topological structure: For each open subgroup U ⊆ G the group AU carries
a compact group topology, such that for all open subgroups U and V and all
σ ∈ G with V ⊆ σU σ −1 , the natural map
σ : AU −→ AV , a 7−→ σa,
is continuous.
In particular, for V ⊆ U , the group AU carries the subgroup topology of AV

with respect to the inclusion AU ⊆ AV , and the norm map NU/V : AV → AU
is continuous.
Assume now that the discrete G-module A is level-compact. In this case, the
additional topological structure on A induces a compact group topology on all
groups in the above sequences (∗) and (∗∗). Furthermore, all maps occurring
are continuous. Since the projective limit is exact on compact groups, we
obtain the
(1.9.11) Lemma. Let A be a level-compact G-module. Then

Ĥ 0 (G, A) = AG /NG A and Ĥ −1 (G, A) = NG A/IG A.
If A is level-compact, then the groups

Ĥ i (G, A) = lim
←−
Ĥ i (G/U, AU ), i ≤ 0,
U, def
are abelian profinite groups in a natural way: For each open normal subgroup
U ⊆ G, the group Ĥ i (G/U, AU ) inherits a natural compact topology from AU
via the standard complex. Furthermore, this group is annihilated by (G : U ),
hence Ĥ i (G/U, AU ) is profinite. Since def is continuous, the group Ĥ i (G, A)
is profinite as an inverse limit of abelian profinite groups. For i > 0 we give
Ĥ i (G, A) = H i (G, A) the discrete topology.
We will frequently make use of the following fact: For a level-compact
G-module A, the natural map
lim AU −→ NG A = NG/U AU
\
←−
U,Norm U

is surjective. This follows easily from the fact that a filtered inverse limit of
nonempty compact spaces is nonempty, cf. [160], chap.IV, (2.3). In particular,
the norm map NG/U : NU A → NG A is surjective for all open normal subgroups
U ⊆ G.
The next lemma shows that we have some flexibility when calculating the
Tate cohomology of level-compact modules.
(1.9.12) Lemma. Let A be a level-compact G-module. Suppose that for every

normal open subgroup U ⊆ G a closed (with respect to the additional topology)
G-submodule
A(U ) ⊆ AU
is given in such a way that the following conditions hold
(i) NU A ⊆ A(U ) for all U ,
(ii) for V ⊆ U , NU/V : AV → AU maps A(V ) to A(U ).
Then
Ĥ i (G, A) ∼
= lim Ĥ i (G/U, A(U ))
←−
U
for all i ≤ −1.
Proof: The transition maps in the above inverse system are defined by an
obvious modification of def (possible by condition (ii)). We have seen in (1.9.4)
that the deflation maps in negative dimensions are given by a map on the chain
level. As projective limits are exact on compact groups, we see that for i ≤ −2
the group Ĥ i (G, A) can also be calculated as the quotient of the inverse limit
of the cocycles modulo the inverse limit of the coboundaries. These, however,
take values in the groups of universal norms on the corresponding levels, i.e.,
by condition (i), we may take the limit over the groups A(U ) instead of AU as
well. This shows the lemma for i ≤ −2.
Now we show the assertion for i = −1. We have Ĥ −1 (G, A) = NG A/IG A
with NG A = lim NG/U AU and IG A = lim I G/U AU . Recalling that the
←− U ←− U
functor lim is exact on compact groups, we have to show
←−
NG A = lim NG/U A(U ) and IG A = lim IG/U A(U ).

←− ←−
U U
Obviously, lim NG/U A(U ) ⊆ NG A. Let (aU ) ∈ NG A. Since lim is taken over
←− \ ←−
V U
the norm maps NU/V : A −→ A , we have aU ∈ NU/V AV = NU A ⊆
V ⊆U
A(U ). This proves NG A = lim
←−
NG/U A(U ).

The inclusion A(U ) ⊆ AU yields the injection

lim
←−
IG/U A(U ) −→ lim
←−
IG/U AU = IG A.
U U
For the surjectivity, let (aU ) ∈ lim
←−
IG/U AU . The projective limit is taken over
U
the maps NU/V : IG/V AV −→ IG/U AU given by (σ −1)a 7−→ (σ̄ −1)NU/V (a).
Therefore
IG/U NU/V AV .
\
(∗) aU ∈
V ⊆U
We show aU ∈ IG/U NU A for any fixed U . Let V run through the open normal
subgroups of G contained in U . Let ÃV = σ∈G/U AV and let
Q
fV : ÃV −→ IG/U AU
be the composite of the maps
ÑU/V f
ÃV −→ ÃU −→ U
IG/U AU ,
and fU : (aσ )σ∈G/U 7→ σ∈G/U (σ − 1)aσ . All
Q P
where ÑU/V = σ∈G/U NU/V
groups are compact and fV is continuous. Therefore, using (∗), fV−1 (aU )
is a nonempty, closed and thus compact subset of ÃV . It follows that
lim fV−1 (aU ) =/ ∅.
←− V ⊆U
Let (ãV ) ∈ lim fV−1 (aU ). Then for ãU = (aσ )σ∈G/U we have aσ ∈ NU A,
←− V ⊆U
hence X
aU = fU (ãU ) = (σ − 1)aσ ∈ IG/U NU A.
σ ∈G/U
This proves that (aU ) ∈ lim IG/U NU A ⊆ lim IG/U A(U ), whence IG A =
←− U ←− U
lim
←−
IG/U A(U ). 2
U
In everything that follows, we tacitly assume maps between level-compact

modules to be continuous with respect to the additional topology.

i j
0 −→ A0 −→ A −→ A00
be an exact sequence of level-compact G-modules, such that the induced map
NUÙ A U NU A00
N j
is surjective for a cofinal system of open normal subgroups U of G (and

hence for all open normal subgroups). Then there is an associated long exact
cohomology sequence
· · · −→ Ĥ −n (G, A0 ) −→ Ĥ −n (G, A) −→ Ĥ −n (G, A00 ) −→ · · ·
ending with · · · → Ĥ 0 (G, A0 ) → Ĥ 0 (G, A) → Ĥ 0 (G, A00 ). Moreover, if j is
surjective, we obtain the long exact cohomology sequence unbounded in both
directions, i.e. from −∞ to +∞.

Proof: For every open normal subgroup U in G we consider the kernel

NU j
A0 (U ) := ker(NU A −→ NU A00 ). We have inclusions NU A0 ⊆ A0 (U ) ⊆ A0U ,
and obtain the exact and commutative diagram
0áâãàßÚÛÜÝÞ A0 (U ) NU A NU A00 0
NG/U NG/U NG/U
0 A0G AG A00G .
Consider the long exact cohomology sequence
· · · → Ĥ i (G/U, A0 (U )) → Ĥ i (G/U, NU A) → Ĥ i (G/U, NU A00 ) → · · ·
associated to the upper line. It consists of compact abelian groups, is clearly

exact and all homomorphisms including the boundary maps are continuous
(for example use the snake lemma in the abelian category of compact abelian
groups). Passing to the inverse limit over U , we obtain, using (1.9.12), the
asserted long exact sequence up to dimension −1.
By compactness, the image of NG/U : NU A −→ AG is NG A and hence, by
(1.9.11), the cokernel of this map is Ĥ 0 (G, A). We denote its kernel by
X(A, U ) = NG/U NU A,
which contains Y (A, U ) := IG/U NU A, and the same holds for A00 . The snake
lemma implies an exact commutative diagram
Y (A,éêåæçèä U )
j
Y (A00 , U )
δ
X(A, U ) X(A00 , U ) A0G /NG/U A0 (U ) Ĥ 0 (G, A) Ĥ 0 (G, A00 ).
Observe that
lim
←−
(A0G /NG/U A0 (U )) = Ĥ 0 (G, A0 ).
U
Furthermore, the upper map j : IG/U NU A → IG/U NU A00 is obviously sur-

jective and NG/U : NU A → AG maps IG/U NU A to zero. Therefore δ :
X(A00 , U ) → A0G /NG/U A0 (U ) maps Y (A00 , U ) to zero by the definition of δ.
This means that we may replace in the last diagram the group X(A, U ) by
Ĥ −1 (G/U, NU A) = X(A, U )/Y (A, U ) and X(A00 , U ) by Ĥ −1 (G/U, NU A00 ) =
X(A00 , U )/Y (A00 , U ) and obtain an exact sequence of compact groups and con-
tinuous homomorphisms. Taking projective limits over U , we obtain the exact
sequence
δ
Ĥ −1 (G, A) → Ĥ −1 (G, A00 ) → Ĥ 0 (G, A0 ) → Ĥ 0 (G, A) → Ĥ 0 (G, A00 ).

Now suppose that j is surjective. Consider the commutative and exact diagram
0ö÷ëìíîïðñòóôõøù 0
Ĥ 0 (G, A) Ĥ 0 (G, A00 )
AG A00G H 1 (G, A0 ) H 1 (G, A)

0
NG j 00
NG A NG A
0 0.
The vertical sequences are exact by (1.9.11). We deduce the existence of the
dotted arrow, which glues the already proven long exact sequence in negative
dimension with the long exact sequence
H 1 (G, A0 ) −→ H 1 (G, A) −→ H 1 (G, A00 ) −→ · · · . 2
The homomorphisms in the negative part of the long exact sequence are
easily seen to be continuous. The maps in dimensions at least 1 are continuous
because the groups are discrete. But the boundary map δ 0 : Ĥ 0 (G, A00 ) →
H 1 (G, A0 ) may be discontinuous. Therefore the next proposition is nontrivial.
(1.9.14) Proposition. If p is a prime number, then the long exact sequence of

lemma (1.9.13) also induces a long exact sequence of the p-parts.
Proof: All occurring cohomology groups are either abelian profinite groups
or abelian discrete torsion groups and therefore they naturally decompose into
the direct sum of their p-parts and their prime-to-p-parts. In order to prove the
corollary, it suffices to show that also the differentials decompose into a di-
rect sum of homomorphisms. This is trivially true for continuous differentials,
hence it remains to consider the homomorphism δ 0 : Ĥ 0 (G, A00 ) → H 1 (G, A0 ).
Let x ∈ Ĥ 0 (G, A00 )(p). Choose a pre-image m00 ∈ A00G of x and let m ∈ A
be a pre-image of m00 . Furthermore, let U ⊆ G be an open normal sub-
group in G such that m ∈ AU . The closed subgroup generated by m in
AU maps onto the closed subgroup generated by x which coincides with
ZZp· x ⊆ Ĥ 0 (G, A00 )(p). We conclude that δ 0 (λx) ∈ H 1 (G/U, A0U ) ⊆ H 1 (G, A0 )
for all λ ∈ ZZp . Writing #(G/U ) = N pk with (N, p) = 1, we obtain that
δ 0 (x) = N δ 0 (N −1 x) ∈ N ·H 1 (G/U, A0U ) = H 1 (G/U, A0U )(p) ⊆ H 1 (G, A0 )(p).
A similar argument shows that δ 0 sends the prime-to-p-part of Ĥ 0 (G, A00 ) to
the prime-to-p-part of H 1 (G, A0 ). 2

Applying the above results to the case of a finite module, we obtain the
(1.9.15) Proposition. Let A be a finite G-module. Assume that `∞ | #G for

all prime numbers ` dividing the order of A. Then
Ĥ 0 (G, A) = AG and Ĥ i (G, A) = 0 for all i ≤ −1.
Proof: Let U ⊆ G be an open subgroup with A = AU and let V ⊆ U be an

open normal subgroup. Then the norm map NU/V from A = AV into A = AU
is just the multiplication by (U : V ). As `∞ | #G for all prime numbers `
dividing the order of A, we conclude that NU A = 0 for all open subgroups U
of G. Applying (1.9.13) to the exact sequence
0 −→ 0 −→ 0 −→ A,
i
we obtain Ĥ (G, A) = 0 for all i ≤ −1. The statement on Ĥ 0 (G, A) follows
likewise from (1.9.8) or (1.9.11). 2
Remark: Tate cohomology of profinite groups was first introduced by

G. POITOU in [168]. The presentation of this section essentially follows [196].

Chapter II
Some Homological Algebra
§1. Spectral Sequences
If G is a profinite group and H a closed normal subgroup, then one may

ask whether the cohomology groups H n (G, A) can be computed from the
cohomology groups of the smaller groups H and G/H. We have already seen
a relation of this type, namely the isomorphism
H n (G/H, H 0 (H, A)) ∼
= H n (G, A)
if H n (H, A) = 0 for all n ≥ 1, which follows from (1.6.7). There is a quite
general relation, which is denoted by
H p (G/H, H q (H, A)) ⇒ H n (G, A)
and is called a “spectral sequence”. The situation is slightly involved. It
roughly says that there is a canonical decreasing filtration of H n = H n (G, A),
H n = F 0H n ⊇ F 1H n ⊇ · · · ⊇ F nH n ⊇ F n+1 H n = 0,
such that the quotient F p H n /F p+1 H n is isomorphic not directly to the group
H p (G/H, H n−p (H, A)), but to a certain subquotient of it. The notion of
spectral sequence is very general and of utmost importance in cohomology
theory. The general set-up, which can be generalized in several directions if
the underlying category has inductive limits, is the following.
Let A be an abelian category. A (decreasing) filtration of an object A is a

family (F p A)p∈ ZZ of subobjects F p A of A such that F p A ⊇ F p+1 A for all p.
Write
grp A = F p A/F p+1 A.
By convention, we put F ∞ A = 0 and F −∞ A = A. We say that the filtration
is finite if there exist n, m ∈ ZZ with F m A = 0 and F n A = A. Given filtered
objects A and B in A, a morphism f : A → B is said to be compatible with the
filtration if f (F p A) ⊆ F p B for all p ∈ ZZ.
Let m be a natural number. An Em -spectral sequence∗) in A is a system

E = (Erpq , E n ),
∗) In the applications usually only E - and E -spectral sequences occur.
1 2

98 Chapter II. Some Homological Algebra
consisting of
a) objects Erpq ∈ A for all (p, q) ∈ ZZ × ZZ and any integer r ≥ m,
b) morphisms d = dpq
: Erpq −→ Erp+r,q−r+1 with d ◦ d = 0
r and such that for
each fixed pair (p, q) ∈ ZZ × ZZ the morphisms dpq r and drp−r,q+r−1 vanish
for sufficiently large r,
c) isomorphisms αrpq : ker (dpq p−r,q+r−1 ∼ E pq ,
) −→
r )/im(dr r+1
d) finitely filtered objects E n ∈ A for all n ∈ ZZ,

pq ∼
e) isomorphisms β pq : E∞ −→ grp E p+q .
By b) and c), the objects Erpq are independent of r for r sufficiently large and
pq
are then denoted by E∞ . These are the objects occurring in e).
In other words, for each r ≥ m, the system Erpq is a system of complexes
pq
whose cohomology groups are the objects Er+1 of the next system. A spectral
pq pq pq
sequence is like a book with (infinitely many) pages Em , Em+1 , Em+2 , . . . and
n
a limit page
úûüýþÿ
E at the end. !"#
• • • • • • • • • • • • • •
E1pq • • • • • • • E2pq • • • • • • •
• • • • • • • • • • • • • •
• • • • • • • • • • • • • •
$%&'()*+,-. /01234
• • • • • • • • • • • • • •
E3pq • • • • • • • E4pq • • • • • • •
• • • • • • • • • • • • • •
• • • • • • • • • • • • • •
For an Em -spectral sequence E = (Erpq , E n ), one usually writes

pq
Em ⇒ E p+q
pq
or Em ⇒ E n . The Em pq
are called the initial terms, the E n the limit terms
0
and the dpq
r differentials. By forgetting the first m −m pages, an Em -spectral
sequence induces an Em0 -spectral sequence for all m0 ≥ m in a natural way.
A morphism of Em -spectral sequences
ϕ : E = (Erpq , E n ) −→ E 0 = (Er0pq , E 0n )
in A is a system of morphisms
0pq
ϕpq pq
r : Er −→ Er , ϕn : E n −→ E 0n ,
where the ϕn are compatible with the filtrations of E n and E 0n and the ϕpqr ,ϕ
n
commute with dpq pq

r , αr and β .
pq

§1. Spectral Sequences 99
If Erpq = 0 for p < 0 or q < 0, one speaks of a first quadrant spectral

sequence. In this case we have
Erpq = E∞pq
for r > max(p, q + 1), r ≥ m.
Once a first quadrant spectral sequence is given, we obtain a realm of
homomorphic connections. We restrict to the most important case of an E2 -
spectral sequence. Of basic importance are the two homomorphisms
E2n,0 −→ E n −→ E20,n ,
the so-called edge morphisms . The first one is the composite of the morphisms
E2n,0 −→ E3n,0 −→ . . . −→ E∞ n,0
−→ E n ,
which are well-defined because F n+1 E n = 0 and Ern+r,−r+1 = 0 for r ≥ 2
n,0
(so that Er+1 is a quotient of Ern,0 ). The second one is the composite of the
morphisms
E n −→ E∞ 0,n
−→ . . . −→ E30,n −→ E20,n ,
which are well-defined because F 0 E n = E n and Er−r,n+r−1 = 0 for r ≥ 2 (so
0,n
that Er+1 is embedded in Er0,n ). A direct consequence of the definition of the
edge morphisms is the following
(2.1.1) Proposition. For any first quadrant E2 -spectral sequence the sequence
d
0 −→ E21,0 −→ E 1 −→ E20,1 −→ E22,0 −→ E 2
is exact. It is called the associated five term exact sequence.
We get a generalization of this result under the assumption that E2pq = 0 for
0,n ∼ ∼ E 0,n
0 < q < n. Namely, in this case we have isomorphisms En+1 → ··· → 2
n+1,0 ∼ ∼ E n+1,0 0,n n+1,0
and E2 → ··· → n+1 . Therefore the differential En+1 → En+1 induces
a homomorphism d
E20,n −→ E2n+1,0 .
We obtain the following
(2.1.2) Lemma. Assume that, in a first quadrant E2 -spectral sequence, the

terms E2pq vanish for 0 < q < n and all p. Then
E2m,0 ∼
= Em
for m < n and the sequence
d
0 −→ E2n,0 −→ E n −→ E20,n −→ E2n+1,0 −→ E n+1
is exact.
The proof of this lemma (and also of the results below) is elementary and we
refer to [21], chap.XV, §5. The most frequent application of spectral sequences
are in the following special cases.

(2.1.3) Lemma. Assume that a first quadrant E2 -spectral sequence is given.

(i) If E2pq = 0 for all q > 1 and all p, then we have a long exact sequence
0 E21,0 E1 E20,1
d
E22,0 E2 E21,1
d
BCDEF;<=>?@A56789: E23,0 E3 E22,1 ··· .

(ii) If E2pq = 0 for all p > 1 and all q, then the sequences
0 −→ E21,n−1 −→ E n −→ E20,n −→ 0
are exact for all n ≥ 1.
(2.1.4) Lemma. Assume that, in a first quadrant E2 -spectral sequence, the

term E2pq vanishes for all (p, q) with (p − m)·(q − n) < 0. Then
E mn ∼= E m+n . 2
In particular, if E2pq = 0 for all q > 0, then
E2m,0 ∼
= Em for all m.
Proof. If p > m and q < n or if p < m and q > n, then Erpq = 0 for
all r ≥ 2, since it is a subquotient of E2pq , and hence E∞
pq
= 0. Therefore on
pq mn
the line p + q = m + n, all terms E∞ are zero up to E∞ and consequently
mn ∼
E∞ = E m+n . The maps
d d
Erm−r,n+r−1 −→
r
Erm,n −→
r
Erm+r,n−r+1
are zero for all r ≥ mn
2, hence E2mn = E3mn = · · · = Ermn = E∞ . 2
In practice, the differentials of a spectral sequence are often difficult to

calculate. We are in a rather comfortable situation if they vanish from a certain
point on. In this case one says that the spectral sequence degenerates. The
precise definition is the following.
(2.1.5) Definition. An Em -spectral sequence degenerates at Em0 for some

m0 ≥ m if the differentials
dpq pq p+r,q−r+1
r : Er −→ Er
vanish for all r ≥ m0 and all (p, q) ∈ ZZ × ZZ. Hence, in this case
pq pq pq pq
Em 0 = Em0 +1 = Em0 +2 = · · · = E∞
for all p, q ∈ ZZ × ZZ.

§2. Filtered Cochain Complexes 101
§2. Filtered Cochain Complexes
Many spectral sequences arise from filtered cochain complexes. Let A be an

abelian category. By a (cochain) complex A• in A we understand a sequence
A• = (An , dn )n∈ ZZ of objects and homomorphisms
dn−1 dn dn+1
· · · −→ An−1 −→ An −→ An+1 −→ An+2 −→ · · ·
with dn+1 ◦ dn = 0 for all n ∈ ZZ. The d’s are called differentials. We say that
A• is bounded below (resp. bounded above, resp. bounded) if An = 0 for
n 0 (resp. for n 0, resp. for n 0 and n 0). We set
dn
Z n (A• ) = ker (An −→ An+1 ),
dn−1
B n (A• ) = im (An−1 −→ An ).
The elements of Z n (A• ) and B n (A• ) are called the n-cocycles and n-co-
boundaries, respectively. As d ◦ d = 0, we have B n (A• ) ⊆ Z n (A• ). The factor
group
H n (A• ) = Z n (A• )/B n (A• )
is called the n-dimensional cohomology group of A• .
A homomorphism of complexes f : A• → B • is a sequence f = (f n )n∈ ZZ
of homomorphisms f n : An → B n with f n+1 ◦ dn = dn ◦ f n for all n ∈ ZZ. A
homomorphism f : A• → B • of complexes induces homomorphisms
H n (f ): H n (A• ) −→ H n (B • )
on the cohomology, and we call f a quasi-isomorphism if H n (f ) is an iso-
morphism for all n.
A filtration by subcomplexes of A• is a filtration F • An of An for all n ∈ ZZ
such that for each n, F n A• is a subcomplex of A• . We say that the filtration
F • A• is biregular, if, for each n ∈ ZZ, the filtration F • An is finite.
Examples: For any complex A• we have the following filtrations

1. The trivial filtration tr• A• defined by
(
A• , for p ≤ 0,
trp A• =
0, for p ≥ 1.
We have (
q n • H q (A• ), for n ≤ 0,
H (tr A ) =
0, for n ≥ 1.
2. Consider for p ∈ ZZ the subcomplex τ≤p (A• ) of A• given by
An ,


 for n ≤ p−1,
τ≤p (A• )n = ker(An →d
An+1 ), for n = p,

0, for n ≥ p + 1.


We have
(
q • H q (A• ), for q ≤ p,
H (τ≤p (A )) =
0, for q ≥ p + 1.
The canonical filtration τ • A• on A• is the decreasing filtration defined by
τ p An = τ≤−p (A• )n .
A biregular filtration F • A• induces an E1 -spectral sequence in the following

way.
(2.2.1) Proposition. Let F • A• be a biregularly filtered cochain complex. For

(p, q) ∈ ZZ × ZZ and r ∈ ZZ ∪ {∞}, we put

Zrpq = ker F p Ap+q → Ap+q+1 /F p+r Ap+q+1 ,
Brpq = d(F p−r Ap+q−1 ) ∩ F p Ap+q ,
pq p+1,q−1
Erpq = Zrpq /(Br−1 + Zr−1 ),
F p H p+q (A• ) = im (H p+q (F p A• ) → H p+q (A• )) .
Then the differential d of the complex A• induces homomorphisms

d = dpq pq p+r,q−r+1
r : Er −→ Er
for all r ∈ ZZ in a natural way. There are canonical isomorphisms

pq
αrpq : ker (dpq p−r,q+r−1
r )/im(dr
∼
) −→ Er+1
for all r ∈ ZZ. For fixed (p, q) ∈ ZZ × ZZ, the morphisms dpq p−r,q+r−1
r and dr
vanish for sufficiently large r and we have natural isomorphisms
Erpq −→
∼ E pq ,
∞ r0.
Finally, there exist natural isomorphisms
β pq : E∞ −→ grp H p+q (A• ).
pq ∼
In particular, these data define a spectral sequence

E1pq ⇒ H p+q (A• ).
Remark: We have Erpq = grp Ap+q for all r ≤ 0. For r ≤ −1 the differentials
dpq pq pq
r are zero and αr = id grp Ap+q . The sequence of isomorphisms αr for r
≥ 0
starts with
α0pq : H q (E0p• ) = H q (grp Ap+• ) −→
∼ E pq .
1

Proof of (2.2.1): The fact that d induces homomorphisms dpq r and the exis-
tence of natural isomorphism αrpq can be easily verified using the definition
of the objects occurring (cf. [21], I.3.1.5). As the filtration is biregular, for
each fixed pair (p, q) ∈ ZZ × ZZ the morphisms dpqr and dr
p−r,q+r−1
vanish for
pq pq
sufficiently large r. The natural isomorphisms Er = E∞ for r 0 and
E∞ → grp H p+q (A• ) can be read off directly from the definition, using the
pq ∼
biregularity of the filtration. 2
(2.2.2) Definition. We call the spectral sequence of (2.2.1) the spectral se-
quence associated to the biregularly filtered complex F • A• .
Examples: 1. For the trivial filtration tr• A• , the associated spectral sequence
has the following shape:
(
H q (A• ), for p = 0,
E1pq =
0, for p =/ 0.
The spectral sequence degenerates at E1 , i.e. E1pq = E∞
pq
for all p, q.
2. For the canonical filtration τ • A• , the associated spectral sequence has the
following shape:
(
H −p (A• ), for p + q = −p,
E1pq =
0, for p + q =/ −p.
The spectral sequence degenerates at E1 , i.e. E1pq = E∞
pq
for all p, q.
In both examples, the filtration on the limit terms has only one nontrivial
graded piece, which is not the case in general. Quite often, spectral sequences
arise from double complexes.
(2.2.3) Definition. A double complex A•• is a collection of objects Apq ∈ A,

p, q ∈ ZZ, together with differentials d0pq : Apq → Ap+1,q and d00p,q : Apq → Ap,q+1
such that d0 ◦ d0 = 0 = d00 ◦ d00 and d0 ◦ d00 + d00 ◦ d0 = 0. The associated total
complex
A• = tot (A•• )
is the single complex with An = Apq whose differential d : An → An+1
L
p+q=n
is given by the sum of the maps
d = d0 + d00 : Apq −→ Ap+1,q ⊕ Ap,q+1 , p + q = n.

Example: Let (C • , dC ) and (D• , dD ) be two complexes of abelian groups.

Their tensor product is the double complex
A•• = C • ⊗ D•
with Apq = C p ⊗ Dq , p, q ∈ ZZ, and with the following differentials:
d0pq = dpC ⊗ idDq , d00pq = (−1)p idC p ⊗ dqD .

As there is no danger of confusion, the associated total complex tot(C • ⊗ D• )
is usually also called the tensor product of C • and D• .
Given a double complex A•• , we have the natural filtration F p A• of the total
complex defined by
F p An = Ai,n−i .
M
i≥p
Example: Let D• be any complex of abelian groups and let C • be the complex
given by (
i ZZ, for i = 0,
C =
0, for i =/ 0.
Then tot(C • ⊗D• ) = D• and the induced filtration on D• is the trivial filtration.
Let A•• be a double complex. We will assume in the following that, for
each n, there are only finitely many nonzero Apq on the line p + q = n. Then the
above filtration on A• = tot(A•• ) is biregular and induces a spectral sequence
E1pq ⇒ H p+q (A• )
converging to the cohomology of A• . The initial terms E1pq are obtained by
taking cohomology in direction q:
E1pq = H q (Ap• , d00 ).
The E1 terms give a complex
d0 d0 d0 d0
H q (A•• ) : · · · −→ H q (Ap−1,• ) −→ H q (Ap,• ) −→ H q (Ap+1,• ) −→ · · · ,
whose cohomology yields the E2 -terms:
E2pq = H p (H q (A•• )).
For reasons that will become apparent later, one often forgets the E1 -page and
calls the spectral sequence
E2pq = H p (H q (A•• )) ⇒ H p+q (tot A•• )
the spectral sequence associated to the double complex A•• .

Remark: For a double complex A•• of abelian groups, the differentials

dpq pq
2 : E2 → E2
p+2,q−1
may be described as follows. For each class c ∈ E2pq ,
there are elements x ∈ Apq and y ∈ Ap+1,q−1 with the following properties:
1) d00 x = 0, d00 y = −d0 x,
2) c is represented by x and dpq 0
2 c by d y.
(2.2.4) Lemma. Let A•• be a first quadrant double complex, i.e. Apq = 0
if p < 0 or q < 0. Assume that for each q ≥ 0 the horizontal complex
A0q → A1q → A2q → · · · is exact (i.e. trivial cohomology in dimension ≥ 1).
Then E n = H n (B • ), where B • is the complex ker(A0• → A1• ).
Proof: Setting 0Apq = Aqp , we obtain a double complex 0A•• with the same
total complex A• as A•• , and hence a new spectral sequence 0E2pq ⇒ E n with
the same limit terms E n (with different filtrations, however). Now the vertical
sequences 0Ap0 → 0Ap1 → · · · are exact, so that 0E2pq = 0, i.e. 0E∞
pq
= 0 for q > 0
and all p ≥ 0. This means that E n = 0F n E n ∼ = 0E∞ n,0
= 0E2n,0 = H q (H p (A•• ))
for p = 0, q = n. 2
Our first application of spectral sequences is the following
(2.2.5) Proposition. Let R be a commutative ring with unit. Let f : D• → D0 •

be a quasi-isomorphism of complexes of R-modules and let C • be a complex
consisting of flat R-modules. Assume that one of the following conditions is
fulfilled:
(i) C • is bounded above,
(ii) D• and D0 • are bounded below,
(iii) R is a Dedekind domain.
Then the induced homomorphism
idC • ⊗ f : tot (C • ⊗R D• ) −→ tot (C • ⊗R D0 • )
is a quasi-isomorphism.
Proof: We start with the special case D0 • = 0, i.e. D• is exact, and we have
to show that tot (C • ⊗R D• ) is exact.
Let us first assume that D• is bounded above. Then, if (i) or (ii) holds, the
natural filtration on tot (C • ⊗R D• ) is biregular, and for the associated spectral
sequence

we have
E1pq = H q (C p ⊗ D• ) = 0
as C p is flat for all p. Hence all limit terms H p+q (tot (C • ⊗R D• )) vanish.
If D• is not bounded above, we write D• = lim τ≤n (D• ) and obtain
−→ n
H k (tot (C • ⊗R D• )) = lim
−→
H k (tot (C • ⊗R τ≤n (D• ))) = 0
n
for all k.
Keeping the assumption D0• = 0, we now assume that R is a Dedekind
domain. Then each submodule of a flat R-module is again flat. Hence the
complexes τ≤n (C • ) consist of flat R-modules for all n, and we obtain
H k (tot (C • ⊗R D• )) = lim
−→
H k (tot (τ≤n (C • ) ⊗R D• )) = 0
n
for all k. This settles the case D0• = 0.

In the general case we consider the mapping cone C(f )• of f . It is defined by
C(f )n = Dn+1 ⊕ D0n
with differential d((a, b)) = (−d(a), d0 (b) + f (a)) for a ∈ Dn+1 , b ∈ D0n . The
natural long exact sequence
· · · −→ H n (D• ) −→ H n (D0• ) −→ H n (C(f )• ) −→ · · ·
shows that C(f )• is exact, as f is a quasi-isomorphism. Moreover, the con-
struction of the mapping cone commutes with tensor products, i.e.
C(idC • ⊗ f )• = tot (C • ⊗ C(f )• ).
By the first part of the proof, tot (C • ⊗ C(f )• ) is exact if one of the conditions
(i)–(iii) is satisfied. Hence C(idC • ⊗ f )• is exact and therefore idC • ⊗ f is a
quasi-isomorphism. 2
Exercise 1. Calculate the spectral sequence associated to the stupid filtration

n
• n A , for n ≥ p,
σ≥p (A ) =
0, for n ≤ p − 1.
Does it degenerate?
Exercise 2. Let p be a prime number and R = ZZ/p2 ZZ. Consider the exact and flat complex
of R-modules
p p p
C • = · · · −→ ZZ/p2 ZZ −→ ZZ/p2 ZZ −→ · · · .
Show that H i (tot (C • ⊗R C • )) =/ 0 for all i.

§3. Degeneration of Spectral Sequences 107
§3. Degeneration of Spectral Sequences
In this section we investigate the degeneration of the spectral sequence

attached to a biregularly filtered cochain complex. The first and easiest case is
degeneration at E1 .
(2.3.1) Theorem. For the spectral sequence

E1pq ⇒ H p+q (A• )
associated to a biregularly filtered complex F • A• the following assertions are
equivalent:
(i) The spectral sequence degenerates at E1 .
(ii) For all n, p we have F p An ∩ d(An−1 ) = d(F p An−1 ).
(iii) For all n, p the natural map H n (F p A• ) → H n (A• ) is injective.
If, moreover, the maps in (iii) are split-injections, we obtain a (non-canonical)
splitting
H n (A• ) ∼ E1pq .
M
=
p+q=n
Proof: Without loss of generality, we may work in the category of mod-

ules over a ring. Assume that (i) holds. For sufficiently large p, we have
F p An−1 = 0, hence assertion (ii) holds for p 0. We fix n ∈ ZZ and proceed
by decreasing induction on p. Let x ∈ F p An ∩ d(An−1 ). As the filtration is
biregular, x ∈ d(F m An−1 ) for some m, which we choose as large as possible
(m = ∞ allowed). We assume that m = p−r for some r ≥ 1, and show that this
yields a contradiction. Let y ∈ F p−r An−1 be a pre-image of x. By construction,
y ∈ Zrp−r,n−1−p+r . By assumption, the differential dr : Erp−r,n−1−p+r → Erp,n−p
p,n−p p+1,n−p−1
is the zero map. Hence x = dy ∈ Br−1 + Zr−1 , i.e. we may write x in
the form
x = dy 0 + x0 , y 0 ∈ F p−r+1 An−1 , x0 ∈ F p+1 An .
As x and dy 0 are coboundaries, the same holds for x0 . By our inductive assump-
tion, we have x0 ∈ d(F p+1 An−1 ), which implies x ∈ F p An ∩ d(F p−r+1 An−1 ),
contradicting the maximality of m = p − r. Hence m ≥ p, showing that
x ∈ d(F p An−1 ). This proves (i)⇒(ii). The implication (ii)⇒(i), as well as the
equivalence (ii)⇔(iii) are elementary.
Finally, assume that (i)–(iii) hold. By definition, F p H n (A• ) is the image
of the natural map H n (F p A• ) → H n (A• ), which is a split-injection. As the
filtration on H n (A• ) is finite, we recursively obtain a splitting
H n (A• ) ∼ F p H n (A• )/F p+1 H n (A• )
M
=
p

into a finite direct sum. By (i), we have

E1p,n−p = E∞
p,n−p
= grp H n (A• ) = F p H n (A• )/F p+1 H n (A• ),
showing the last assertion of the proposition. 2
By a formal reindexing procedure, we can displace a spectral sequence in the

pq
following sense: Assume we are given an Em -spectral sequence Em ⇒ E n.
Putting
2p+q,−p
Ẽ n = E n , F p Ẽ n = F p+n E n and Ẽrpq = Er+1 for r ≥ m − 1,
we obtain an Em−1 -spectral sequence converging to the same limit terms, but
with a shifted filtration, which we call the displaced spectral sequence. It is a
remarkable fact that, if the spectral sequence E arises from a biregular filtered
cochain complex, then the displaced spectral sequence Ẽ arises from another
filtration on the same complex, the displaced filtration. This will be useful in
showing that a spectral sequence degenerates at E2 , just by showing that the
displaced spectral sequence Ẽ satisfies the conditions of (2.3.1).
(2.3.2) Definition. Let F • A• be a biregular filtered cochain complex. The

filtration
Dis(F )p An = Z1p+n,−p = {a ∈ F p+n An | da ∈ F p+n+1 An+1 }
is called the displaced filtration. We denote the complex A• , together with

the filtration Dis(F ), by Dis(A• ).
One easily verifies that Dis(A• ) is a filtered complex:
d(Z1p+n,−p ) ⊆ F p+1+n An+1 ∩ ker(d) ⊆ Z∞

p+n+1,−p ⊆ Z1p+n+1,−p ,
and the filtration Dis(F ) is obviously biregular.
Example: Displacing the trivial filtration tr on A• , we obtain the canonical

filtration τ :
Dis(tr)p An = Z1p+n,−p = {a ∈ trp+n An | da ∈ trp+n+1 An+1 }
An ,


 for n ≤ −p − 1,
n n+1
= ker(A → A ), for n = −p,
for n ≥ −p + 1,


0,
= τ p An .

§3. Degeneration of Spectral Sequences 109
(2.3.3) Proposition. For all r ≥ 1, there are natural isomorphism

∼ E 2p+q,−p (A• )
Erpq (Dis(A• )) −→ r+1
commuting with the corresponding differentials. The displaced spectral se-

quence is the spectral sequence associated to the displaced filtration.
Proof: For r ≥ 0 we have

n o
pq
DisF Zr = x ∈ Dis(F )p Ap+q | dx ∈ Dis(F )p+r Ap+q+1
= {x ∈ F 2p+q Ap+q | dx ∈ F 2p+q+r+1 Ap+q+1 }
2p+q,−p
= F Zr+1 .
Analogously,
pq
DisF Br = d(Dis(F )p−r Ap+q−1 ) ∩ Dis(F )p Ap+q
= d(F 2p+q−r−1 Ap+q−1 ) ∩ F 2p+q Ap+q
2p+q,−p
= F Br+1 .
pq p+1,q−1
By definition, Erpq = Zrpq /(Br−1 + Zr−1 ). Hence the natural identifica-
∼ 2p+q,−p
tions above induce isomorphisms Er (Dis(A• )) → Er+1
pq
(A• ) for r ≥ 1.
pq
Therefore, the Er -terms of the displaced spectral sequence and of the spectral
sequence associated to the displaced filtration are canonically isomorphic. The
same holds for the limit terms, which can easily be seen from their definitions.
2
As an application, we obtain the following degeneration result. It should not

be mistaken for the well-known Künneth-formula, which arises from another
filtration on the tensor product (see the exercise below).
(2.3.4) Theorem. Let R be a Dedekind domain and let C • and D• be com-

plexes of R-modules such that the natural filtration on their tensor product
is biregular.∗) If C • consists of flat (i.e. torsion-free) R-modules, then the
spectral sequence of the double complex A•• = C • ⊗R D• degenerates at E2 .
Furthermore, we have a non-canonical splitting
H n (tot (C • ⊗R D• )) ∼ E2pq .
M
=
p+q=n
∗) e.g. both complexes are bounded above, or both complexes are bounded below, or one of
the complexes is bounded.

Proof: Let A• = tot(C • ⊗R D• ), and let F p An = i≥p C i ⊗R Dn−i be the

L
natural filtration. We want to show that the spectral sequence of the double
complex degenerates at E2 . By (2.3.3), it suffices to show that the spectral
sequence associated to the displaced filtration degenerates at E1 . Thus, by
(2.3.1), we have to show that the natural maps
H n (Dis(F )p A• ) −→ H n (A• )
are split-injections for all p and n. Using the flatness of C • , we obtain

Dis(F )p An = ker F p+n An → F p+n An+1 /F p+n+1 An+1 ,
L
= i>p+n C i ⊗R Dn−i ⊕ C p+n ⊗R ker(d−p
D )
n
= tot C • ⊗R τ≤−p (D• ) .
As the filtration on H n (A• ) is finite, it therefore remains to show that for all
n, m ∈ ZZ the natural map

H n tot(C • ⊗R τ≤m (D• )) −→ H n tot(C • ⊗R τ≤m+1 (D• ))
is a split-injection. Let X • be a complex consisting of projective R-modules
together with a quasi-isomorphism X • → τ≤m+1 (D• ). Using the flatness of C • ,
we obtain a commutative diagram
quasi-iso
tot(C • ⊗RHIJG τ≤m (X • )) tot(C • ⊗R τ≤m (D• ))
quasi-iso
tot(C • ⊗R X • ) tot(C • ⊗R τ≤m+1 (D• )) .
By (2.2.5)(iii) the horizontal maps are quasi-isomorphisms. Since X • is a
complex consisting of projective R-modules, the inclusion of complexes
τ≤m (X • ) ,→ X •
has a section for each m ∈ ZZ . Indeed, by our assumption on R, d(X m ) ⊆ X m+1
is a projective R-module, and the short exact sequence
0 −→ ker(dm ) −→ X m −→ d(X m ) −→ 0
splits. Using any section s: X m → ker(dm ) in dimension m, and the obvious
maps in the other dimensions, we obtain a splitting of τ≤m (X • ) ,→ X • .
Therefore the left vertical complex homomorphism in the diagram above
has a section, which finishes the proof. 2
The technique of displacing, as well as the results (2.3.1) and (2.3.3), are
due to P. DELIGNE, cf. [34]. The statement of (2.3.4) is implicitly contained
in the article [98] by U. JANNSEN, and we adopted his idea of proof.

§4. The Hochschild-Serre Spectral Sequence 111
Exercise (Künneth-formula): Let R be a Dedekind domain and let C • and D• be complexes

of R-modules. Assume that C • consists of flat R-modules. Consider the biregular filtration
on A• = tot(C • ⊗R D• ) defined by
 L
i j
 Li+j=n C ⊗R D , for p ≤ −1,


p n i • j
F (A ) = i+j=n Z (C ) ⊗R D , for p = 0,


0, for p ≥ 1,

and show the following assertions regarding the associated spectral sequence:
(i) E2p,q = i+j=q TorR i • j • pq
L
−p (H (C ), H (D )); in particular E2 = 0 for p = / 0, −1.
(ii) For each n we obtain a short exact sequence

M M
0→ H i (C • ) ⊗R H j (D• ) → H n (A• ) → TorR1 (H i (C • ), H j (D• )) → 0.
i+j=n i+j=n+1
Moreover, these sequences split (non-canonically).
§4. The Hochschild-Serre Spectral Sequence
Relevant for the cohomology of profinite groups is the following
(2.4.1) Theorem. Let G be a profinite group, H a closed normal subgroup

and A a G-module. Then there exists a first quadrant spectral sequence
E2pq = H p (G/H, H q (H, A)) ⇒ H p+q (G, A).
It is called the Hochschild-Serre spectral sequence.
.
Proof: To the standard resolution 0 → A → X of the G-module A, we apply
the functor H 0 (H, −), and get the complex
H 0 (H, X 0 ) −→ H 0 (H, X 1 ) −→ H 0 (H, X 2 ) −→ · · ·
of G/H-modules. For each H 0 (H, X q ), we consider the cochain complex
.
H 0 (H, X q )G/H −→ C (G/H, H 0 (H, X q ))
and we put
C pq = C p (G/H, H 0 (H, X q )) = X p (G/H, X q (G, A)H )G/H , p, q ≥ 0.
We make C •• into a (anti-commutative) double complex by using the following
differentials: We let
d0pq : C pq −→ C p+1,q

.
be the differential of the complex X (G/H, X q (G, A)H )G/H at the place p.
Further, we define
d00pq : C pq −→ C p,q+1
.
as (−1)p times the differential of the complex X p (G/H, X (G, A)H )G/H at the
place q. Then (C •• , d0 , d00 ) is a double complex and we define the Hochschild-
Serre spectral sequence as the associated spectral sequence
E2pq ⇒ E n .
We compute the terms E2pq and E n . By definition, E2pq = H p (H q (C •• )). We
.
have H q (H 0 (H, X )) = H q (H, A) (see p.34). The functor C p (G/H, −) is
exact (I §3, ex.1). Therefore
. .
H q (C p ) = H q (C p (G/H, H 0 (H, X ))) = C p (G/H, H q (H 0 (H, X ))) .
= C p (G/H, H q (H, A)),
hence
.
E2pq = H p (C (G/H, H q (H, A))) = H p (G/H, H q (H, A)).
As for the limit terms, we note that for every q ≥ 0 the complexes C q .
.
= C (G/H, H 0 (H, X q )) are exact. In fact, every X q is an induced G-module,
hence H 0 (H, X q ) is an induced, and thus acyclic, G/H-module by (1.3.6)
.
and (1.3.7), i.e. H p (C q ) = H p (G/H, H 0 (H, X q )) = 0 for p > 0. By lemma
. .
(2.2.4), we obtain E n = H n (B ), where B is the complex
. . . .
B = ker(C 0 (G/H, (X )H ) → C 1 (G/H, (X )H )) = ((X )H )G/H = (X )G . .
Therefore
.
E n = H n ((X )G ) = H n (G, A). 2
From (2.1.4) follows the
(2.4.2) Corollary. If H q (H, A) = 0 for q > 0, then

H n (G/H, AH ) ∼= H n (G, A).
Another consequence is the five term exact sequence

0 −→ H 1 (G/H, AH ) −→ H 1 (G, A) −→ H 1 (H, A)G/H
−→ H 2 (G/H, AH ) −→ H 2 (G, A),
which we proved in I §6 in an elementary way, but with some difficulty. It still
requires, however, careful checking to show that the maps are the inflation,
restriction and transgression respectively.
For inf and res, this identification is given in the available literature (e.g.
[128]). For the transgression, we give the proof here.

(2.4.3) Theorem. The differential

d0,1 1
2 : H (H, A)
G/H
−→ H 2 (G/H, AH )
is the transgression tg as defined in (1.6.6).
Proof (TH. MOSER, J. STIX): In order to calculate H 1 (H, A) we use the

.
acyclic resolution A → X (G, A) of the G-module A. Thus an element
z ∈ H 1 (H, A)G/H is represented by an H-invariant 1-cocycle x : G × G → A.
The invariance of z under G/H implies that for ρ, σ ∈ G the cocycles ρx and
σx differ by a 1-coboundary, i.e. there is a map
b : G × G −→ X 0 (G, A)H , (σ, ρ) 7→ bσ,ρ ,
such that
d(bσ,ρ ) = ρx − σx ,
.
where d denotes the differential of X (G, A). We obtain
(∗) bσ,ρ (τ1 ) − bσ,ρ (τ0 ) = (ρx)τ0 ,τ1 − (σx)τ0 ,τ1
for all σ, ρ, τ0 , τ1 ∈ G. Therefore we may assume that b is G-invariant and
b1,1 = 0. Furthermore, since τ x = x for τ ∈ H, we may also assume that b
factors through G/H × G/H. Then, for all σ, ρ, γ ∈ G/H,
bρ,γ (τ ) − bσ,γ (τ ) + bσ,ρ (τ )
is an element of the module AH independent of τ ∈ G, and so the 2-cocycle
∂b: (G/H)3 → X 0 (G, A)H , given by
∂bσ,ρ,γ = bρ,γ − bσ,γ + bσ,ρ ,
is constant with value in AH . By the remark on page 105 it represents the
image of z under d0,1
2 . The associated inhomogeneous 2-cocycle, which also
0,1
represents d2 (z), is
a : G/H × G/H −→ AH , (σ, ρ) 7→ aσ,ρ ,
where
(∗∗) aσ,ρ = ∂b1,σ,σρ (ζ) = bσ,σρ (ζ) − b1,σρ (ζ) + b1,σ (ζ)
with ζ ∈ G arbitrary.
We now represent tg(z) as follows. We restrict x to H × H, and pass to
the associated inhomogeneous 1-cocycle x0 : H → A, x0 (τ ) = x1,τ , which also
represents z. Consider then the function
y : G −→ A, σ 7→ yσ = x1,σ + b1,σ (σ).
We will show that y satisfies the properties (i), (ii), (iii) of the proof of propo-
sition (1.6.6),

(i) y|H = x0 ,
(ii) yστ = yσ + σyτ for σ ∈ G, τ ∈ H,
(iii) yτ σ = yτ + τ yσ for σ ∈ G, τ ∈ H,
hence tg(z) = [∂y]. Since b1,τ = b1,1 = 0 for τ ∈ H, property (i) follows. Let
σ, ρ ∈ G. Then, by (∗) and (∗∗) and since x is a cocycle, we obtain
(∂y)σ,ρ = (dx)1,σ,σρ + (σx)σ,σρ − xσ,σρ
+b1,σ (σ) − b1,σ (σρ) + (∂b)1,σ,σρ (σρ)
= (∂b)1,σ,σρ (σρ) = aσ,ρ .
If σ ∈ H or ρ ∈ H, then the expression above is zero and so (ii) and (iii) follow.
Now, for arbitrary σ, ρ ∈ G, the equality above shows that tg(z) = [∂y] = [a] =
d01
2 . This proves the theorem. 2
A subtle and useful relation of the Hochschild-Serre spectral sequence to

the cup-product is obtained as follows. Let G be a profinite group, H an open
normal subgroup of G and H 0 the closure of the commutator subgroup of H.
Let A be a G-module on which H acts trivially. We then have, for p > 0, two
canonical homomorphisms
(∗) d2 , u∪ : H p−1 (G/H, H 1 (H, A)) −→ H p+1 (G/H, A)
which are defined as follows. The first map d2 is the differential d2p−1,1 of the
Hochschild-Serre spectral sequence
E2pq = H p (G/H, H q (H, A)) ⇒ H p+q (G, A).
On the other hand, the group extension
1 −→ H ab −→ G/H 0 −→ G/H −→ 1
defines a cohomology class u ∈ H 2 (G/H, H ab ). Using the equality
H 1 (H, A) = Hom(H ab , A), we obtain a canonical pairing
H ab × H 1 (H, A) −→ A ,
which induces a cup-product
∪
H 2 (G/H, H ab ) × H p−1 (G/H, H 1 (H, A)) −→ H p+1 (G/H, A).
The second map u∪ is given by x 7→ u ∪ x.
(2.4.4) Theorem. Let A be a G-module and H an open normal subgroup of

G which acts trivially on A. Then, for p > 0, the maps
d2 , u∪ : H p−1 (G/H, H 1 (H, A)) −→ H p+1 (G/H, A)
are the same up to sign, i.e. d2 (x) = −u ∪ x.
In particular, the transgression tg : H 1 (H, A)G/H → H 2 (G/H, A) is given
by tg(x) = −u ∪ x.

Remark: The statement of (2.4.4) remains true for an arbitrary closed normal
subgroup H of G, but then one has to use the continuous cohomology class
2
u ∈ Hcts (G/H, H ab ) representing the group extension 1 → H ab → G/H 0 →
G/H → 1 (cf. (2.7.7)).
Proof: Suppose first that G is a finite group. Then the projection H → H ab

may be regarded as a G/H-invariant 1-cocycle of H, i.e. as an element ε ∈
H 0 (G/H, H 1 (H, H ab )). From the spectral sequence
E2pq = H p (G/H, H q (H, H ab )) ⇒ H p+q (G, H ab ) ,
we obtain the differential
d0,1 1 ab G/H
2 : H (H, H ) −→ H 2 (G/H, H ab ),
which by (2.4.3) is the transgression tg as defined in (1.6.6). We prove
tg(ε) = −u (in the additive notation of H 2 (G/H, H ab )).
Let s : G/H → G be a section of the projection G → G/H, σ 7→ σ̄. Define
the 1-cochain y : G → H ab by
y(σ) = σ(sσ̄)−1 mod H 0 .
Then y|H = ε and
(∂y)(σ1 , σ2 ) ≡ y(σ1 σ2 )−1σ1y(σ2 )y(σ1 )
≡ s(σ1 σ2 )σ2−1 σ1−1 σ1 σ2 (sσ2 )−1 σ1−1 σ1 (sσ1 )−1
≡ [(sσ1 )(sσ2 )s(σ1 σ2 )−1 ]−1 mod H 0 ,
showing that ∂y depends only on the classes σ1 , σ2 ∈ G/H, and hence tg(ε) =
[∂y] by definition of tg. But the function in square brackets is a 2-cocycle
which represents the class u, hence tg(ε) = −u.
The differential
dp,1 p 1
2 : H (G/H, H (H, A)) −→ H
p+2
(G/H, AH )
is obtained from the part
C p (G/H, XLKM 1 (G, A)H ) ∂0

C p+1 (G/H, X 1 (G, A)H )
∂ 00
∂0
C p+1 (G/H, X 0 (G, A)H ) C p+2 (G/H, X 0 (G, A)H )
of the double complex C pq (A) = C p (G/H, X q (G, A)H ) as follows. Let z ∈

H p (G/H, H 1 (H, A)). z is given by an element α ∈ C p,1 (A) such that ∂ 00 α = 0
.
and that the induced function ᾱ : (G/H)p+1 → H 1 (X (G, A)H ) = H 1 (H, A) is
a p-cocycle representing z. This means that there exists a β ∈ C p+1,0 such that
∂ 0 α = ∂ 00 β. The image dp,1 0
2 (z) is then represented by the cocycle ∂ β ∈ C
p+2,0
.

This process may also be interpreted as follows. From the complex

0 → AH → X 0 (G, A)H → X 1 (G, A)H → X 2 (G, A)H , we obtain the ex-
act sequence of G/H-modules
(1) 0 −→ AH −→ X 0 (G, A)H −→ Z −→ H 1 (H, A) −→ 0
.
with Z = Z 1 (X (G, A)H ). Splitting it up into two short exact sequences
(2) 0 −→ I(A) −→ Z −→ H 1 (H, A) −→ 0,
(3) 0 −→ AH −→ X 0 (G, A)H −→ I(A) −→ 0,

we obtain two δ-homomorphisms
δ δ
H p (G/H, H 1 (H, A)) −→ H p+1 (G/H, I(A)) −→ H p+2 (G/H, AH ),
with dp,1
2 = δ ◦ δ. In fact, the element α above is a lifting α : (G/H)p+1
→ Z 1 (G, A)H of the p-cocycle ᾱ representing z, hence δz is represented by
the (p + 1)-cocycle ∂ 0 α. Since ∂ 0 α = ∂ 00 β, β : (G/H)p+2 → X 0 (G, A)H is a
cochain which lifts ∂ 0 α, hence ∂ 0 β represents δ([∂ 0 α]) = δδ(z), showing that
δδ(z) = dp,1
2 (z).
Let us abbreviate the sequences (2) and (3) by
0 −→ S 0 (A) −→ S(A) −→ S 00 (A) −→ 0,
0 −→ T 0 (A) −→ T (A) −→ T 00 (A) −→ 0,
with S 0 = T 00 . Replacing A by the G-module H ab , we obtain exact sequences
of G/H-modules
0 −→ S 0 (H ab ) −→ S(H ab ) −→ S 00 (H ab ) −→ 0,
0 −→ T 0 (H ab ) −→ T (H ab ) −→ T 00 (H ab ) −→ 0.
Setting B = Hom(H ab , A), we have the pairings
S(H ab ) × B −→ S(A), T (H ab ) × B −→ T (A),
which induce pairings S 0 (H ab ) × B → S 0 (A), S 00 (H ab ) × B → S 00 (A), and
similarly for T . We now assume that H acts trivially on A, i.e. Hom(H ab , A) =
H 1 (H, A). We apply proposition (1.4.3) twice and obtain the commutative
diagram
H 0 (G/H,NPQOVTUSR S 00 (H ab )) × H p (G/H, B) ∪
H p (G/H, S 00 (A))
δ δ
∪
H 1 (G/H, S 0 (H ab )) × H p (G/H, B) H p+1 (G/H, S 0 (A))
δ δ
∪
H 2 (G/H, T 0 (H ab )) × H p (G/H, B) H p+2 (G/H, T 0 (A)).

In the upper pairing, we have S 00 (H ab ) = H 1 (H, H ab ), S 00 (A) = H 1 (H, A) and

ε ∪ z = z (recall that z ∈ H p (G/H, Hom(H ab , A)) = H p (G/H, H 1 (H, A))).
Since δδε = d0,1
2 (ε) = tg(ε) = −u, it follows that
dp,1
2 (z) = δδ(z) = δδ(ε ∪ z) = δδε ∪ z = −u ∪ z.
This proves the theorem for a finite group G.

Now let G be an arbitrary profinite group. We denote by G0 the commutator
subgroup of G. We let W run through the open normal subgroups of G which
are contained in H. Setting G = G/W , H = H/W , we have a commutative
exact diagram
1WXYZ[\]^_à H ab G/H 0 G/H 1
1 H ab G/H 0 G/H 1.
The lower group extension defines an element uW ∈ H 2 (G/H,H ab ) which is
π
the image of u under H 2 (G/H, H ab ) → H 2 (G/H,H ab ) (see (3.6.1)). On the
other hand, we have a commutative diagram
edcb 1 (H, A))
H p (G/H, H
d2
H p+2 (G/H, A)
inf
d2
H p (G/H, H 1 (H, A)) H p+2 (G/H, A).
We have shown that d2 xW = −uW ∪ xW for xW ∈ H p (G/H, H 1 (H̄, A)). The

diagram
hgfij
H 2 (G/H,H ab
) × H p (G/H, H 1 (H, A)) ∪
H p+2 (G/H, A)
π inf
∪
H 2 (G/H, H ab ) × H p (G/H, H 1 (H, A)) H p+2 (G/H, A)
is commutative because of the commutativity of the cup-product with the

inflation. So, d2 inf xW = d2 xW = −uW ∪ xW = −πu ∪ xW = −u ∪ inf xW .
Since H 1 (H, A) = lim−→ W
H 1 (H/W, A), each x ∈ H p (G/H, H 1 (H, A)) is of
the form x = inf xW for some W . This completes the reduction to the case of
a finite group G. 2
If 1 → H → G → G/H → 1 is a split group extension and A is a G-

module on which H acts trivially, then (2.4.4) shows that the differentials d∗,1
2
are trivial. However, the Hochschild-Serre spectral sequence does, in general,
not degenerate at E2 , even if A is a trivial G-module, see [46], example on
page 83. But, we have the following

(2.4.5) Proposition. Let 1 → H → G → G/H → 1 be a split exact sequence

of profinite groups, and let A be a discrete G-module on which H acts trivially.
Then
inf : H ∗ (G/H, A) −→ H ∗ (G, A)
is a monomorphism onto a direct summand, and all differentials into the
horizontal edge of the Hochschild-Serre spectral sequence for A vanish, i.e.
dr∗,r−1 = 0 for all r ≥ 2.
Proof: Let ι: G/H ,→ G be a splitting of the projection π: G G/H, i.e.

πι = id. Since AH = A, the edge homomorphisms
H ∗ (G/H, A) = E2∗,0 −→ E∞
∗,0 ⊆
H ∗ (G, A)
are split injections, so E2∗,0 = Er∗,0 = E∞∗,0 ∗,0
for r ≥ 2. Since Er+1 =
∗,0 ∗,r−1
Er /imdr , we obtain dr ∗,r−1
= 0. 2
In case of a direct product, the Hochschild-Serre spectral sequence degen-

erates at E2 . We follow the proof given by U. JANNSEN, see [98].
(2.4.6) Theorem. Let G and H be profinite groups and let A be a discrete

H-module, regarded as a (G × H)-module via trivial action of the group G.
Then the Hochschild-Serre spectral sequence
E2pq = H p (G, H q (H, A)) ⇒ H n (G × H, A)
degenerates at E2 . Furthermore, it splits in the sense that there is a decompo-
sition
H n (G × H, A) ∼ H p (G, H q (H, A)).
M
=
p+q=n
Proof: For a trivial G-module B, we have a natural isomorphism of complexes

C • (G, B) ∼= C • (G, ZZ) ⊗ B.
This is easily verified if G is finite, and the result for general G follows by a
straightforward limit process. Therefore, under the assumptions of the theorem,
the Hochschild-Serre spectral sequence is the spectral sequence associated to
the double complex
C • (G, ZZ) ⊗ X • (G × H, A)H .
As the complex C • (G, ZZ) consists of flat ZZ-modules, the result follows from
(2.3.4). 2
Remark: The decomposition in the theorem is non-canonical: it cannot be

made functorial in A. However, for a fixed H-module A, it can be made
functorial in G. For a proof of these assertions see [98].

Exercise 1. Show that in the Hochschild-Serre spectral sequence, the edge morphisms
H n (G/H, AH ) −→ H n (G, A) −→ H n (H, A)G/H
are the inflation and the restriction.
Exercise 2. From the exact sequence 0 → A → IndG (A) → A1 → 0, one obtains the four term
exact sequence
0 −→ AH −→ IndG (A)H −→ AH 1
1 −→ H (H, A) −→ 0
of G/H-modules, and hence a homomorphism

δ 2 : H p (G/H, H 1 (H, A)) −→ H p+2 (G/H, AH ).
Show that δ 2 is the differential dp,1 p,1 p+2,0
2 : E2 → E2 .
Exercise 3. Let E(G, H, A) denote the Hochschild-Serre spectral sequence

E2pq = H p (G/H, H q (H, A)) ⇒ H n (G, A).
If G0 is an open subgroup of G and H 0 = H ∩ G0 , then we have two morphisms of spectral
sequences E(G,lk H, A) cor E(G0 , H 0 , A).
res
Exercise 4. (i) Assume that H q (H, A) = 0 for q > 1. Then we have an exact sequence
0mnopqrstuv H 1 (G/H, H 0 (H, A)) H 1 (G, A) H 0 (G/H, H 1 (H, A))
H 2 (G/H, H 0 (H, A)) H 2 (G, A) H 1 (G/H, H 1 (H, A))
H 3 (G/H, H 0 (H, A)) H 3 (G, A) H 2 (G/H, H 1 (H, A)) .
(ii) Assume that cd G/H ≤ 1. Then we have exact sequences
0 −→ H (G/H, H n−1 (H, A)) −→ H n (G, A) −→ H n (H, A)G/H −→ 0
1
for all n ≥ 1 and all discrete G-modules A.
Exercise 5. If A×B → C is a pairing of G-modules, then for the terms Erpq of the Hochschild-
Serre spectral sequence, we have a cup-product
0 0 ∪ 0 0
Erpq (A) × Erp q (B) −→ Erp+p ,q+q (C)
such that
dr (α ∪ β) = (dr α) ∪ β + (−1)p+q α ∪ dr (β).
Exercise 6. (Künneth-formula) Let G and H be profinite groups and let A be a discrete

H-module, regarded as a (G × H)-module via trivial action of the group G. Show that there
exist natural short exact sequences for all n
M M
0→ H i (G, ZZ) ⊗ H j (H, A) → H n (G × H, A) → Tor Z1Z (H i (G, ZZ), H j (H, A)) → 0.
i+j=n i+j=n+1
Moreover, these sequences split.

Hint: Consider the double complex C • (G, ZZ) ⊗ X • (G × H, A)H , which, by the proof of
(2.4.6), calculates the cohomology H ∗ (G × H, A). Then use the result of the exercise at the
end of §3.
Exercise 7. Let G and H be profinite groups and let F be a field, considered as a trivial
module. Show that
∼
M
H n (G × H, F ) = H i (G, F ) ⊗F H j (H, F ).
i+j=n

§5. The Tate Spectral Sequence
Another spectral sequence, due to J. TATE, which in a sense is dual to the

Hochschild-Serre spectral sequence, is obtained as follows. Let G be a profinite
group. For an abelian group M we set
M ∗ = Hom(M, Q/ZZ),
which is equipped with the natural G-structure g(φ)(m) = g(φ(g −1 (m))), if M
is a G-module. Let A be a G-module. For two open subgroups V ⊆ U of G,
we have the maps
cor ∗ : H n (U, A)∗ −→ H n (V, A)∗ ,
dual to the corestriction, by which the family (H n (U, A)∗ ) becomes a direct
system of abelian groups.
(2.5.1) Definition. Let G be a profinite group, H a closed subgroup and A a

G-module. Then, for every n ≥ 0, we set
Dn (H, A) = lim H n (U, A)∗ ,
−→
U ⊇H
where U runs through the open subgroups of G containing H.
If H is open, then Dn (H, A) = H n (H, A)∗ . If H is a normal closed subgroup

of G, then it suffices to let U run through the normal open subgroups of G
containing H. Dn (H, A) is then a G/H-module. As the functors X 7→ X ∗
and lim
−→
are exact, we see that the family (Dn (H, −))n≥0 is a contravariant ∗)
δ-functor on Mod(G). We set
Dn (A) = Dn ({1}, A).
(2.5.2) Definition. For a G-module A and an integer n ≥ 0, we write

cd(G, A) ≤ n
if H q (H, A) = 0 for all q > n and all closed subgroups H of G. (The letters
“ cd” stand for “cohomological dimension”.)
We call the following spectral sequence the Tate spectral sequence.

∗) This means that the arrows in the usual exact cohomology sequence are reversed.

§5. The Tate Spectral Sequence 121
(2.5.3) Theorem. If cd(G, A) ≤ n, then for every closed normal subgroup H,

there exists a first quadrant spectral sequence
E2pq = H p (G/H, Dn−q (H, A)) ⇒ H n−(p+q) (G, A)∗ .
In particular, for H = 1, we have a spectral sequence
E2pq = H p (G, Dn−q (A)) ⇒ H n−(p+q) (G, A)∗ .
.
Proof: We consider the standard resolution A → X of the G-module A and
set Z i = ker(X i → X i+1 ). Splitting up the complex
.
X̃ : 0 → X 0 → X 1 → X 2 → · · · → X n−1 → Z n → 0
into short exact sequences and, recalling that the G-modules X i are cohomo-
logically trivial, we obtain for r > 0
H r (H, Z n ) ∼
= H r+1 (H, Z n−1 ) ∼
= ··· ∼
= H r+n (H, A).
In particular, H r (H, Z n ) = 0 for r > 0, since cd(G, A) ≤ n, i.e. Z n is a
cohomologically trivial G-module.
Let U be an open normal subgroup of G. We apply to the complex X̃ .
first the functor H 0 (U, −) and then the functor Hom(−, Q/ZZ), and obtain a
complex
0 −→ Y 0 −→ Y 1 −→ Y 2 −→ · · · −→ Y n −→ 0,
where Y q = H 0 (U, X̃ n−q )∗ for q ≥ 0, and in particular, Y 0 = H 0 (U, Z n )∗ . This
is a complex of G/U -modules, which by (1.8.2) and (1.8.5) are cohomologi-
cally trivial, since X n−q and Z n are cohomologically trivial G-modules. Since
the functor Hom(−, Q/ZZ) is exact, we obtain
. . .
H q (Y ) = H q (H 0 (U, X̃ n− )∗ ) = H q (H 0 (U, X̃ n− ))∗ = H n−q (U, A)∗
for all q.
.
For each Y q , we consider the cochain complex C (G/U, Y q ) and obtain
a double complex C pq = C p (G/U, Y q ), p, q ≥ 0. In order to make it anti-
commutative, we multiply the differentials (p, q) → (p, q + 1) by (−1)p , cf.
the proof of (2.4.1). As described in §2, this double complex yields a spectral
sequence
E2pq ⇒ E p+q .
We compute the initial terms E2pq = H p (H q (C •• )). The functor C p (G/U, −) is
exact (I §3, ex.1), so that
. . .
H q (C •• ) = H q (C (G/U, Y )) = C (G/U, H q (Y )) .
.
= C (G/U, H n−q (U, A)∗ ),
hence
.
E2pq = H p (C (G/U, H n−q (U, A)∗ )) = H p (G/U, H n−q (U, A)∗ ).

As for the limit terms E p+q , we note that for each q ≥ 0 the complex
C (G/U, Y ) −→ C (G/U, Y ) −→ C (G/U, Y q ) −→ · · ·
0 q 1 q 2
is exact; its homology groups are the groups H p (G/U, Y q ), which are zero for
p > 0, since Y q is cohomologically trivial.
. . .
Therefore, setting B = ker(C 0 → C 1 ), we have by (2.2.4)
E p+q = H p+q (B . ) = H (H (G/U, Y . )).
p+q 0
Since Y q is cohomologically trivial, we have Ĥ i (G/U, Y q ) = 0 for all i. By

(1.2.6), we obtain
H 0 (G/U, Y q ) = Hom((X̃ n−q )U , Q/ZZ)G/U = Hom(((X̃ n−q )U )G/U , Q/ZZ)
∗
NG/U
←−
∼ Hom(((X̃ ) ) , Q/ZZ) = H 0 (G, X̃ n−q )∗
n−q U G/U
and consequently
.
E p+q = H p+q (H 0 (G, X̃ n− )∗ ) = H p+q (H 0 (G, X̃ n− ))∗ .
= H n−(p+q) (G, A)∗ .
We thus obtain a spectral sequence
E2pq = H p (G/U, H n−q (U, A)∗ ) ⇒ H n−(p+q) (G, A)∗ .
If we now let U run through the open subgroups of G containing H and take
direct limits, we get the spectral sequence
E2pq = H p (G/H, Dn−q (H, A)) ⇒ H n−(p+q) (G, A)∗ . 2
Remark: Tate gave a proof of this spectral sequence using the cohomology
groups in negative dimensions (see [230]), which we have avoided here.
The Tate spectral sequence

E(G, H, A) : E2pq = H p (G/H, Dn−q (H, A)) ⇒ H n−(p+q) (G, A)∗
is functorial in G and H in the following sense. Let G0 be an open subgroup
of G and H 0 = H ∩ G0 . We then have two morphisms of spectral sequences
cor ∗ res
(∗) xw
E(G, H, A) E(G0 , H 0 , A),
cor res ∗
such that the maps on the initial terms and the limit terms are given as follows.
The E2yzpq E20pq are given as the composites of
H p (G/H, D{~}| n−q (H, A))
res
cor
H p (G0 /H 0 , Dn−q (H, A))
cor ∗
H p (G0 /H 0 , Dn−q (H 0 , A)),
res ∗
where cor ∗ and res ∗ are induced by the direct limit of the maps
cor ∗
A)∗
H n−q (U, H n−q (U 0 , A)∗ ,
res ∗

U running through the open normal subgroups of G containing H and U 0 =

U ∩ G0 . The maps on the limit terms are
cor ∗
H n−(p+q) (G, A)∗ H n−(p+q) (G0 , A)∗ .
res ∗
In particular, for H = {1}, we obtain for the edge morphisms a commutative

diagram
n (A))
H p (G, D H n−p (G, A)∗ H 0 (G, Dn−p (A))
res cor cor ∗ res ∗ incl NG/G0
H p (G0 , Dn (A)) H n−p (G0 , A)∗ H 0 (G0 , Dn−p (A)).
All this results from the following consideration. Assume that H is an open
subgroup of G. Then the spectral sequence E(G, H, A) is obtained from the
double complex
C pq (G, H, A) = C p (G/H, X̃ n−q (G, A)H∗ ),
where X̃ i = X i for i = 0, · · · , n − 1 and X̃ n = ker(X n → X n+1 ) as in the proof
of (2.5.3). We have on the one hand the homomorphisms
(G, A)H∗ )
C p (G/H, X̃ n−q
res
C p (G0 /H 0 , X̃ n−q (G0 , A)H ∗ ).
0
cor
On the other hand, the duals of the maps

X̃ n−q (G, A)
res
X̃ n−q (G0 , A)
cor
yield homomorphisms
cor ∗
X̃ n−q H∗
(G, A) −→
←−
0
X̃ n−q (G0 , A)H ∗ . ∗)
∗
res
After composing, we obtain two morphisms of double complexes

cor ∗ res
C pq (G, H, A) C pq (G0 , H 0 , A),
cor res ∗
and these induce the above morphisms (∗) of spectral sequences. The effects
on the cohomology groups mentioned are obtained in a straightforward manner
by recalling our identifications
E pq ∼
= H p (G/H, H n−q (H, A)∗ ), E p+q ∼
2 = H n (G, A)∗ .
We finish this section on spectral sequences by explicitly determining the

edge morphisms E q → E20,q and E2p,0 → E p of the Tate spectral sequence.
∗) The corestriction cor is defined on all of X̃ n−q (G0 , A) after choosing a sec-
0
tion s: G/G0 → G of G → G/G0 . It induces a homomorphism cor : X̃ n−q (G0 , A)H →
X̃ n−q (G, A)H if this choice is taken in such a way that s(c)σ = τσ s(cσ) for all c ∈ G0 \G,
σ ∈ H and some τσ ∈ H 0 .

(2.5.4) Theorem. The edge morphism

G/H
H n−q (G, A)∗ −→ lim
−→
H n−q (U, A)∗
U ⊇H
in the Tate spectral sequence is the direct limit of the maps cor ∗ , dual to the
corestriction maps
cor : H n−q (U, A) −→ H n−q (G, A).
Proof: We use the notation of the proof of (2.5.3). We may assume that H is
open. The spectral sequence is obtained from the double complex
C pq = C p (G/H, Y q ),
where Y q = H 0 (H, X̃ n−q )∗ and X̃ i = X i for i < n and X̃ n = Z n . Let
. . . .
C = tot (C •• ), B = ker(C 0 → C 1 ) and K = C 0 . Then B is a . . .
. . .
subcomplex of C and H q (B ) = H q (C ) = E q by (2.2.4). Therefore the
edge morphism E q → E20,q is the map
(1) .
H q (B ) −→ H q (K ) .
. π . .
induced by the composite B → C → K , which is the inclusion. Identifying
. . .
C 0 = C 0 (G/H, Y ) with Y , this is the inclusion
(2) .
(Y )G/H ,→ Y . .
The image of (1) is contained in
. ∂ .
E20,q = ker(H q (C 0 ) −→ H q (C 1 )) = H q (Y )G/H , .
and the edge morphism becomes the map
.
edge : H q ((Y )G/H ) −→ H q (Y )G/H , .
induced by the inclusion (2). From what we have seen in the proof of (2.5.3),
this map is identified with a map
H q (G, A)∗ −→ [H q (H, A)∗ ]G/H
as follows. We have the canonical isomorphism
.
Hom((X̃ n− )G , Q/ZZ) ∼= H 0 (G/H, Y ) , .
∗
which is the same as the dual NG/H of
(X̃ n− )H . NG/H .
(X̃ n− )G .
We obtain a commutative diagram

H q (H 0 (G/H, .
Y ))
edge .
H q (Y )G/H
∗
NG/H
.
H q (H 0 (G, X̃ n− )∗ )
∗
NG/H
H q (H 0 (H, X̃ n− )∗ )G/H .
cor ∗
H n−q (G, A)∗ [H n−q (H, A)∗ ]G/H ,

which identifies the edge homomorphism with the dual of the corestriction.
2
We now consider the edge morphism E2p,0 → E p in the Tate spectral se-
quence for the case H = {1}. It is a homomorphism
H p (G, D n (A)) edge
H n−p (G, A)∗ .
In particular, for p = n and A = ZZ, we have a canonical homomorphism, called
the trace map,
tr : H n (G, Dn (ZZ)) −→ Q/ZZ,
provided cd(G, ZZ) ≤ n. On the other hand, for every pair V ⊆ U of open
normal subgroups of G and each i ≥ 0, we have the canonical pairing
H i (V, A)∗ × AU −→ H i (V, ZZ)∗ , (χ, a) 7−→ f (x) = χ(ax).
Taking first the direct limit over V and then over U , we obtain a canonical
bilinear map Di (A) × A −→ Di (ZZ), which gives us a cup-product
∪
H p (G, Di (A)) × H n−p (G, A) −→ H n (G, Di (ZZ)).
For i = n this yields, together with the map tr, a homomorphism
cup
H p (G, Dn (A)) −→ H n−p (G, A)∗ .
(2.5.5) Theorem. Suppose that cd(G, ZZ) ≤ n and let A ∈ Mod(G) be finitely
generated as a ZZ-module. If cd(G, A) ≤ n, then the two maps
edge
n (A))
H p (G, D H n−p (G, A)∗
cup
coincide for all p ∈ ZZ.
Proof: The Tate spectral sequence arises from the double complex
C pq (A) = C p (G/U, H 0 (U, X̃ n−q (G, A))∗ )
and the application of lim . We have a canonical pairing
−→ U
ϕ : (X̃ (G, A) ) × A −→(X̃ n−q (G, ZZ)U )∗ , (χ, a) 7−→ ϕ(χ, a) = f,

n−q U ∗
where f : X̃ n−q (G, ZZ)U −→ Q/ZZ is defined by f (x) = χ(ax).

If z(σ0 , . . . , σp−j ) is a (p − j)-cochain with coefficients in (X̃ n−q (G, A)U )∗
and t(σ0 , . . . , σj ) is an j-cocycle in Z j (G, A), then
(z ∪ t)(σ0 , . . . , σp ) = ϕ(z(σ0 , . . . , σp−j ), t(σp−j , . . . , σp ))
is a p-cochain with coefficients in (X̃ n−q (G, ZZ)U )∗ . Thus we get a map
C p−j,q (A) × Z j (G, A) −→ C p,q (ZZ),
i.e. for fixed t ∈ Z j (G, A), we have a morphism of double complexes
. . ..
∪t : C −j, (A) −→ C , (ZZ)

of degree (j, 0) and hence a transformation of the associated edge morphisms.

Applying lim , we obtain a map
−→ U
∪t : H n−j (G, Dn (A)) −→ H n (G, Dn (ZZ)),

which only depends on the cohomology class of the cocycle t. As the map tr
is defined via the edge morphism, we obtain a commutative diagram
H n−j (G,¢ ¡ Dn (A)) × H j (G, A) ∪
H n (G, Dn (ZZ))
edge tr
H j (G, A)∗ × H j (G, A) Q/ZZ,

where the upper arrow is the cup-product with respect to the pairing
Dn (A) × A −→ Dn (ZZ),
and the lower arrow is the evaluation map (χ, t) 7→ χ(t). From this diagram
we get the commutative diagram
H n−j (G,¥¦¤£ Dn (A))
cup
Hom(H j (G, A), H n (G, Dn (ZZ)))
edge tr
H j (G, A)∗ Hom(H j (G, A), Q/ZZ),

which shows that the edge morphism coincides with the composite tr ◦ cup.
Setting j = n − p, we obtain the assertion of the theorem. 2
We conclude this section with a vanishing criterion for the terms Di (A).
(2.5.6) Lemma. Suppose that Di (ZZ) = 0. Assume that A ∈ Mod(G) is finitely

generated as a ZZ-module and has torsion divisible only by prime numbers `
for which Di+1 (ZZ) is `-divisible. Then Di (A) = 0.
Proof: If A is finitely generated as a ZZ-module, then A = AU for some open

subgroup U of G. Since in the definition of the Di the group G may be replaced
by U , we may assume that A is a trivial G-module. If A is torsion-free, then
A∼ = ZZn as a G-module, hence Di (A) ∼ = Di (ZZ)n = 0. It remains to consider
the case A = ZZ/mZZ. From the exact sequence 0 → ZZ → ZZ → ZZ/mZZ → 0,
m
Di+1 (ZZ) −→ Di+1 (ZZ) −→ Di (ZZ/mZZ) −→ Di (ZZ) = 0,
hence our assumptions imply Di (ZZ/mZZ) = 0. 2

§6. Derived Functors 127
Exercise 1. Compute the Tate spectral sequence for the case G ∼

= ẐZ.
Exercise 2. If 0 < cd (G, A) ≤ n, then, for all open normal subgroups U of G, we have
Dn (A)U = H n (U, A)∗ .
Hint: Use (3.3.11).
Exercise 3. Let G be a profinite group and suppose we are given a direct limit A = lim Aα of
−→
G-modules. Show that there exist natural homomorphisms
Di (G, A) −→ lim Di (Aα )
←−
for all i. What are the images of these maps?
§6. Derived Functors
We have constructed the cohomology groups H n (G, A) in a direct and natural

way from the diagram
· · · −→
−→
−→ G × G × G −→
−→ −→
−→ G × G −→ −→ G.
The advantage of this definition is that it is concrete, elementary and down
to earth. Its disadvantage is that it is difficult to generalize and to get deeper
insights. There is another much more general definition of cohomology which
we describe now.
We have explained in I §3 the notion of δ-functor H = (H n )n≥0 between
abelian categories A and A0 . It is a family of additive functors H n : A → A0 ,
which turns a short exact sequence
0 −→ A −→ B −→ C −→ 0
in A functorially into a long exact sequence
δ
· · · −→ H n (A) −→ H n (B) −→ H n (C) −→ H n+1 (A) −→ · · ·
in A0 . A morphism between two δ-functors H and H 0 from A to A0 is a system
f = (f n )n≥0 of functorial morphisms
f n : H n −→ H 0n ,
which commute with δ. That is, for any exact sequence
0→A→B →C →0
in A, the diagram
ª©¨§
H n (C) δ
H n+1 (A)
f n (C) f n+1 (A)
δ
H 0n (C) H 0n+1 (A)
commutes.

(2.6.1) Definition. A δ-functor H = (H n )n≥0 from A to A0 is called universal

if, for every other δ-functor H 0 = (H 0n )n≥0 from A to A0 , each morphism
0
f 0 : H 0 → H 0 of functors extends uniquely to a morphism f : H → H 0 of
δ-functors.
We have the following criterion for the universality of a δ-functor. An

additive functor F : A → A0 is called effaceable if, for each object A in A,
there is a monomorphism u : A → I in A such that F (u) = 0.
(2.6.2) Theorem. A δ-functor H = (H n )n≥0 from A to A0 is universal if the

functors H n are effaceable for n > 0.
For the proof we refer to [66], chap.I. The idea is the following. Let
H 0 = (H 0n )n≥0 be an arbitrary δ-functor from A to A0 and let f 0 : H 0 → H 00
be a morphism of functors. Assume that we have shown that there exists a
uniquely determined morphism of functors f i : H i → H 0i , i = 1, . . . , n, which
u
commute with δ. Let A ∈ A and let 0 → A → I → J → 0 be an exact sequence
such that H n+1 (u) = 0. Then we obtain a uniquely determined morphism
f n+1 : H n+1 (A) → H 0n+1 (A) using the exact commutative diagram
H n²±°¯«¬® (I) H n (J) δ
H n+1 (A) 0
fn fn f n+1
δ
H 0n (I) H 0n (J) H 0n+1 (A).
It remains to show that f n+1 is functorial and commutes with δ.
If G is a profinite group, then the functors H n (G, −), n > 0, are effaceable,
since every G-module A embeds into the induced G-module IndG (A), which
is acyclic, i.e. has trivial cohomology. We therefore have the
(2.6.3) Theorem. The δ-functor (H n (G, −))n≥0 is universal.
With this theorem many proofs of isomorphism, uniqueness etc. are obtained
automatically.
Example 1. Let H be a closed normal subgroup. We then have the following

proof of Shapiro’s lemma (1.6.4)
sh : H n (G, IndH (A)) ∼
G= H n (H, A).

Noting that IndH H

G is exact and that IndG (IndH (B)) = IndG (B), we see that
n H n
(H (G, IndG (−))) and (H (H, −)) are effaceable δ-functors on Mod(H), and
are thus universal. They are functorially isomorphic in dimension n = 0.
Hence they are isomorphic as δ-functors. By the uniqueness assertion, the
isomorphism is the composite
res π
H n (G, IndH n H ∗ n
G (A)) −→ H (H, IndG (A)) −→ H (H, A),
where π∗ is induced by π : IndH
G (A) → A, f 7→ f (1). In fact, this composite is
a morphism of δ-functors, which in dimension n = 0 coincides with the initial
isomorphism.
Example 2. For every G-module A, we have the commutative diagram
µ´³ H n (G, IndH
G (A))
sh ν∗
cor
H n (H, A) H n (G, A)
as claimed in (1.6.5). In fact, ν∗ and cor ◦ sh are morphisms of universal
δ-functors, which coincide in dimension n = 0 by the definition of sh, cor, ν∗ .
Hence they coincide for all n ≥ 0 by the uniqueness assertion of (2.6.1).
If F : A → A0 is an additive functor, there exists up to a canonical iso-

morphism at most one universal δ-functor H from A to A0 with H 0 = F .
This δ-functor, if it exists, is then called the right derived functor of F and
.
is denoted by R F = (Rn F )n≥0 . Obviously, it is defined up to canonical
isomorphism. The question is, when does it exist?
.
By theorem (2.6.3), the universal δ-functor H (G, −) on Mod(G) is the
right derived functor
.
H (G, −) = R Γ .
of the functor
Γ (−) = H 0 (G, −) : Mod(G) −→ Ab, A 7−→ AG .
Suppose M is a full abelian subcategory of Mod(G) which has the property that
for a discrete G-module M , the induced module IndG (M ) is also in M. Then
the same reasoning as above shows that the restriction of H (G, −) to M is.
.
the right derived functor R Γ of the functor Γ (−) : M −→ Ab, A 7→ AG .
Examples of such subcategories M are the category Modt (G) of discrete
G-modules which are torsion groups, or the category Mod(p) (G) of discrete
G-modules which are p-torsion groups, where p is a prime number.
Recall that an additive functor F : A → A0 is called left exact if for each
exact sequence 0 → A → B → C the sequence
0 −→ F (A) −→ F (B) −→ F (C)
is also exact.

The left exactness of F is clearly a necessary condition for the existence of

.
the right derived functor R F , since H 0 is left exact. This condition is already
sufficient if A has sufficiently many injectives:
An object A in A is injective if for every monomorphism B → C in A the
map Hom(C, A) → Hom(B, A) is surjective. A is said to have sufficiently
many injectives if, for any object A, there exists a monomorphism A → I into
an injective object.
(2.6.4) Theorem. Let A have sufficiently many injectives. Then for each left
exact additive functor F : A → A0 , the right derived functor R F = (Rn F )n≥0.
exists.
For the proof we refer to [21], chap. V, §3, but we explain the idea of it.
Since A has sufficiently many injectives, each object A ∈ A has an injective
resolution, i.e. there is an exact complex
0 −→ A −→ I 0 −→ I 1 −→ I 2 −→ · · ·
with injective objects I n in A. We apply the functor F and get a complex
0 −→ F (A) −→ F (I 0 ) −→ F (I 1 ) −→ F (I 2 ) −→ · · · .
We define
Rn F (A) = H n (F (I )), . n ≥ 0;
in particular, R0 F (A) = ker(F (I 0 ) → F (I 1 )) = F (A).
The independence of this definition from the injective resolution chosen is
. .
seen as follows. If A → I and A0 → I 0 are injective resolutions of A and A0 ,
then, because of the injectivity property of the I n , every morphism u : A → A0
extends to a morphism of complexes
A¸¹¶· I .
u u
A0 I0 , .
and every two such extensions are homotopic (cf. I §3, exercise 6). This means
. .
that the induced maps from F (I ) to F (I 0 ) are homotopic, hence induce
. .
the same homomorphism H n (F (I )) → H n (F (I 0 )) on the homology. In
. . .
particular, if A = A0 , we find extensions u : I → I 0 , v : I 0 → I , such that .
u ◦ v and v ◦ u are homotopic to the identity, hence induce mutually inverse
.
isomorphisms H n (Fº» (I )) .
H n (F (I 0 )). This shows the independence.
The property of being a δ-functor is seen as follows. Any exact sequence
0 → A → B → C → 0 in A may be extended to an exact sequence

0¼½¾¿ÀÁÂÃÄÅÆ A B C 0
0 .
IA .
IB IC. 0
of injective resolutions. Since IAn is injective, all exact sequences
0 −→ IAn −→ IBn −→ ICn −→ 0
split and therefore
0 −→ F (IAn ) −→ F (IBn ) −→ F (ICn ) −→ 0
remains exact. The exact sequence
. . .
0 −→ F (IA ) −→ F (IB ) −→ F (IC ) −→ 0
of complexes yields, in the same way as in I §2, a long exact sequence
δ
· · · −→ Rn F (A) −→ Rn F (B) −→ Rn F (C) −→ Rn+1 F (A) −→ · · · .
.
We have obtained a δ-functor R F = (Rn F )n≥0 . For an injective object I, we
id
have Rn F (I) = 0 for n > 0 since 0 → I → I → 0 is an injective resolution of
I. Since A has sufficiently many injectives, the Rn F , n > 0, are effaceable,
.
hence R F is universal.
(2.6.5) Lemma. If G is a profinite group, then the category Mod(G) of discrete

G-modules has sufficiently many injectives.
Proof: To every abstract G-module M we can associate the submodule

M δ := MU,
[
U ⊆G
where U runs through the open subgroups of G. If we endow M with the
discrete topology, then M δ is the maximal submodule on which G acts con-
tinuously (compare with the remark after (1.1.8)). One easily verifies that
every G-homomorphism from a discrete G-module N to M factors through
M δ . In particular, we see that the discrete module I δ is an injective object
in Mod(G) provided the (abstract) G-module I is injective. The category of
abstract G-modules has sufficiently many injective objects (see [79], chap. IV:
it is canonically equivalent to the category of modules over the group ring
ZZ[G]). Therefore we can embed a given discrete G-module M into an injec-
tive abstract module I and then M is automatically contained in the injective
discrete module I δ . 2
The Hochschild-Serre spectral sequence (2.4.1) becomes a special case of

the following general result.

(2.6.6) Theorem. Let A and A0 be abelian categories with sufficiently many

injectives and let A00 be another abelian category. Let
F F0
A −→ A0 −→ A00
be left exact additive functors. Assume that F maps injective objects from A
to F 0 -acyclic objects, i.e. those annihilated by Rn F 0 for n > 0. Then there is
a cohomological spectral sequence
E2pq = Rp F 0 (Rq F (A)) ⇒ Rp+q (F 0 ◦ F )(A) ,
which is called the Grothendieck spectral sequence.
This spectral sequence is obtained as follows. There exists a homomorphism

.
of the complex F (I ) into a double complex of A0 -injective objects I •• which
induces injective resolutions of all groups F (I q ) and also for all cocycle,
.
coboundary and cohomology groups of the complex F (I ) (a so-called Cartan-
Eilenberg resolution, cf. [21], chap.XVII). Applying to the double complex
I •• = (I pq )p,q≥0 the functor F 0 , we obtain a double complex
(Apq )p,q≥0 = (F 0 (I pq ))p,q≥0 .
The spectral sequence E pq ⇒ E n associated with this double complex is the
maintained spectral sequence
E2pq = Rp F 0 (Rq F (A)) ⇒ Rn (F 0 ◦ F )(A).
For the proof we refer to [66] and [21].
If G is a profinite group and H a closed subgroup, then we have the additive

left exact functors
F = H 0 (H, −) : Mod(G) → Mod(G/H), A 7→ AH ,
F 0 = H 0 (G/H, −) : Mod(G/H) → Ab, B 7→ B G/H ,
F 0 ◦ F = H 0 (G, −) : Mod(G) → Ab, A 7→ AG .
In this case, the Grothendieck spectral sequence
E2pq = Rp F 0 (Rq F (A)) ⇒ Rp+q (F 0 ◦ F )(A)
coincides with the Hochschild-Serre spectral sequence. We easily see that the
E2 -terms and the limit terms are the same, since
Rq F = H q (H, −), Rp F 0 = H p (G/H, −), Rn (F 0 ◦ F ) = H n (G, −).
That both spectral sequences actually coincide follows from the fact that the
functor ‘homogeneous cochain complex’ is a ‘resolving functor’, see [66].
So far we have dealt with the right derivation of a left exact, covariant
functor. In the later applications it will be often useful to work with certain
modifications of this concept.

Assume that we are given abelian categories B and B 0 . We say that a family
H = (Hn )n∈ ZZ of functors Hn : B → B 0 is a homological δ-functor if the
family K = (K n )n∈ ZZ , defined by K n := H−n , is a (cohomological) δ-functor
as defined before. The following notions and statements are dual to those
given before for cohomological δ-functors and we leave their verification to
the reader. We also note that, up to the obvious modifications, one can also
work with contravariant functors.
We say that a homological δ-functor H = (Hn )n≥0 is universal if, for every
other homological δ-functor H 0 = (Hn0 )n≥0 , each morphism f0 : H00 → H0
of functors extends uniquely to a morphism f : H 0 → H of homological
δ-functors. A functor G : B → B 0 is called coeffaceable if, for every object
B ∈ B, there is an epimorphism φ : P → B with G(φ) = 0. A homological δ-
functor H = (Hn )n≥0 : B → B 0 is universal if the functors Hn are coeffaceable
for n > 0. If G : B → B 0 is an additive functor, there exists up to canonical
isomorphism at most one universal homological δ-functor H from B to B 0 with
H0 = G. This δ-functor, if it exists, is then called the left derived functor of
.
G and is denoted by L G = (Ln G)n≥0 .
An object P of B is called projective if for every epimorphism A → B
in B the map Hom(P, A) −→ Hom(P, B) is surjective. We say that B has
sufficiently many projectives if for every object B there exists an epimorphism
P → B with projective P .
The left derived functor of G : B → B 0 exists if G is right exact and B has
sufficiently many projectives.
Now we introduce the homology of profinite groups. The homology groups

are compact abelian groups and they have compact G-modules as coefficients.
In order to prevent confusion, we use the notation D = D(G) for the category of
discrete G-modules, which so far has been denoted by Mod(G). The category
of compact G-modules will be denoted by C = C (G).
(2.6.7) Definition. Let G be a profinite group and let A ∈ C be a compact

G-module. The cofixed module (or module of coinvariants) AG of A is the
largest Hausdorff quotient of A on which G acts trivially, i.e. AG is the quotient
of A by the closed subgroup generated by the elements (ga − a), g ∈ G, a ∈ A.
We denote the category of compact abelian groups by Abc and in order

to stress the difference we write Abd for the category of (discrete) abelian
groups. One easily verifies that (−)G is a right exact functor from C to Abc .
Furthermore, the category C is dual to D by Pontryagin duality and therefore
it has sufficiently many projectives by (2.6.5).

(2.6.8) Definition. For a profinite group G and a compact module A, the

homology groups are defined as the left derivatives of the cofixed-module
functor
Hn (G, A) := Ln (−)G (A).
In particular, we have H0 (G, A) = AG , and if 0 → A → B → C → 0 is an

exact sequence in C , then we get a long exact homology sequence
· · · → Hn+1 (G, C) → Hn (G, A) → Hn (G, B) → Hn (G, C) → · · · .
The homology theory for profinite groups is dual to the cohomology theory:
every cohomological result has its homological analogue. Fortunately we do
not have to prove everything twice because of the following
(2.6.9) Theorem. Let G be a profinite group and A be a compact G-module.

Then there are functorial isomorphisms for all i ≥ 0
Hi (G, A)∨ ∼
= H i (G, A∨ ),
∨
where denotes the Pontryagin dual.
Proof: The theorem is true for i = 0 by the definition of the fixed and the
cofixed module. Now the following diagram of categories and functors is
commutative
CÊÉÈÇ ∨
D
(−)G (−)G
∨
Abc Abd .
Furthermore, Pontryagin duality is an exact, contravariant functor that transfers
C -projectives to D-injectives. The statement of the theorem now follows from
the universal property for the derived functors. 2
We see from the last theorem that, in principle, one can avoid the use
of homology groups, working only with cohomology. Indeed, the decision
whether to work with cohomology or homology, is more or less a question of
personal taste.
We finish this section with a spectral sequence for Ext-groups. Let R be a

ring (with unit). Then the functor HomR (−, −) is a bifunctor from the category
of R-modules to abelian groups, which is contravariant in the first and covariant
in the second variable. Its derivations
ExtiR (−, −)

may likewise be computed using projective resolutions of the first, or using

injective resolutions of the second, variable. (See [79] or any textbook about
homological algebra for this basic fact.) The Hom-acyclic objects in the first
resp. second variable are projective resp. injective R-modules. The following
spectral sequence connects the Ext-groups for modules over a group ring R[G]
with that over R.
(2.6.10) Theorem. Let R be a commutative ring with unit, let G be a finite

group and let M and N be R[G]-modules. Then there exists a natural spectral
sequence
E2pq = H p (G, ExtqR (M, N )) ⇒ Extp+q
R[G] (M, N ) .
Proof: First we observe that

HomR[G] (M, N ) ∼
= HomR (M, N )G ,
thus the left exact functor HomR[G] (M, −) is the composition of the left
exact functors HomR (M, −) and H 0 (G, −). Now assume that N is injec-
tive. Then N is a direct summand of IndG N . By (1.3.6)(iii), the G-module
HomR (M, IndG N ) is induced. Thus HomR (M, N ) is cohomologically trivial
because it is a direct summand of an induced module. Therefore theorem
(2.6.6) gives us the desired spectral sequence. 2
(2.6.11) Corollary. Let G be a finite group whose order is invertible in the

commutative ring R. Then an R[G]-module M is projective if and only if it is
R-projective.
Proof: A free R[G]-module is free as an R-module. If M is a projective

R[G]-module, then it is a direct summand of a free R[G]-module and therefore
also projective as an R-module. In order to show the other implication, assume
that M is an R[G]-module which is R-projective. The cohomology groups
H i (G, M ) are R-modules and annihilated by #G for i ≥ 1 by (1.6.1). Hence
they are trivial for i ≥ 1 and for an arbitrary R[G]-module N , the spectral
sequence (2.6.10) degenerates to a sequence of isomorphisms
ExtiR[G] (M, N ) −→ ExtiR (M, N )G = 0
for i ≥ 1. Hence M is R[G]-projective. 2
We obtain the following result, which is known as Maschke’s Theorem.

(2.6.12) Corollary (MASCHKE). Let G be a finite group and let K be a field

whose characteristic does not divide the order of G. Then the category of
K[G]-modules is semi-simple.
Exercise 1. Define for an abstract group G the homology with values in a G-module as the
left derivation of the cofixed module functor on the category of abstract G-modules. Assume
that G is a finite group and let A be a finite G-module. Then we can view G as an abstract
group and A as an abstract module or we can view G as a profinite group and A as a compact
module. Show that the corresponding homology groups are the same and that both coincide
with the homology groups introduced in I §9.
Exercise 2. Let R be a commutative ring with unit, let G be a finite group and let M and N
be R[G]-modules such that M is R-projective and N is cohomologically trivial. Show that
Ext1R[G] (M, N ) = 0 .
§7. Continuous Cochain Cohomology
In chapter I, we started considering the simplicial diagram

· · · −→
−→ G × G × G −→
−→
−→ −→ G × G −→
−→ −→ G,
.
which gave rise to a standard complex C (G, A), and defined the cohomology
of the profinite group G with values in the discrete module A. In the last
section we learned that this cohomology can be characterized by a universal
property which, in particular, explains its good functorial behaviour. This
functorial approach via universal constructions is extremely useful and clarifies
the principles behind classical homological notions.
Sometimes, however, the simplicial approach seems to reach further. For
instance, the categories of finite, compact, locally compact or of all topological
G-modules lack the existence of sufficiently many “good” objects, and the
existence of derived functors on these categories is not guaranteed. But we
can still define cohomology using the standard complex.
Let G be a profinite group and let A be any topological G-module (see
I §1). We form the continuous homogeneous cochain complex of G with
.
coefficients in A , denoted by Ccts (G, A), by taking the G-invariants of the
.
continuous standard resolution Xcts (G, A), which is defined in exactly the
same way as in I §2 for discrete modules.

§7. Continuous Cochain Cohomology 137
(2.7.1) Definition. We call the n-th cohomology group of the complex

.
Ccts (G, A) the n-th continuous cochain cohomology group of G with co-
n
efficients in A. We denote this group by Hcts (G, A).
n
If A is discrete, then clearly Hcts (G, A) = H n (G, A), but for an arbitrary
topological G-module A, the right-hand side of the equation is not defined.
The same arguments as used in I §3 show the following
(2.7.2) Lemma. Let

α β
0 −→ A −→ B −→ C −→ 0
be a short exact sequence of topological G-modules such that the topology
of A is induced by that of B and such that β has a continuous section (just a
continuous map, not necessarily a homomorphism). Then there exist canonical
boundary homomorphisms
n n+1
δ: Hcts (G, C) −→ Hcts (G, A)
and we obtain an exact sequence
n n n n+1 δ
· · · → Hcts (G, A) → Hcts (G, B) → Hcts (G, C) → Hcts (G, A) → · · ·
δ
which begins 0 → AG → B G → C G → Hcts
1
(G, A).
Note that we can apply this lemma in the particular case that A is an open
submodule of B and C = B/A is the quotient module with the quotient
topology, which is discrete.
Suppose
B: A × B −→ C
is a continuous G-pairing, i.e. a continuous biadditive map such that σ(a · b) =
(σa · σb) for a ∈ A, b ∈ B, σ ∈ G, where a · b denotes B(a, b). Then B induces
biadditive maps
p q p+q
Ccts (G, A) × Ccts (G, B) −→ Ccts (G, C)
via the formula
(f ∪ g)(σ0 , . . . , σp+q ) = f (σ0 , . . . , σp ) · g(σp , . . . , σp+q ).
This cup-product of cochains satisfies the identity
∂(f ∪ g) = (∂f ) ∪ g + (−1)p f ∪ (∂g)
and consequently induces pairings
p q p+q
Hcts (G, A) × Hcts (G, B) −→ Hcts (G, C) ,

which enjoy the same properties as the cup-products considered in I §4.

The maps inf, res, cor with respect to a change of groups are defined and have
the same properties as the maps introduced in I §5. (Note that we exclusively
made calculations there on the chain level!)
In some important special situations, we can still relate the continuous
cochain cohomology to derived functors. For this we need some preparations.
Let A be an abelian category and let AIN be the category of inverse systems
in A indexed by the set IN of natural numbers with the natural order. Thus
objects in AIN are inverse systems
· ·ËÌÍÎÏ ·
d0
An+1 dn An ··· A1 A0
in A and morphisms are commutative diagrams
· ·ÝÜÛÐÑÒÓÔÕÖ×ØÙÚ · An+1 An ··· A1 A0
fn+1 fn f1 f0
··· Bn+1 Bn ··· B1 B0 .

The category A is abelian, with kernels and cokernels taken componentwise.
One can show that AIN has sufficiently many injectives, provided that this is
the case for A. Moreover, an object (An , dn ) in AIN is injective if and only if
all An are injective and all dn are split surjections (see [96], prop. 1.1). A left
exact functor
F : A −→ B
into another abelian category induces a functor F IN : AIN −→ B IN in the obvious
way. If A has sufficiently many injectives, then the right derivatives Ri F IN
exist and we have (loc.cit. prop. 1.2)
Ri F IN = (Ri F )IN for i ≥ 0, i.e. Ri F IN (An , dn ) = (Ri F An , Ri F (dn )).
If inverse limits over IN exist in B, we can define the functor
lim
←−
F : AIN −→ B,
n
which is by definition the composition of the functor F IN with the limit functor
lim
←−
: B IN −→ B.
Assume that A and B have sufficiently many injectives, F is left exact and
IN
F sends injectives to lim
←−
-acyclic objects. Then, by (2.6.6), we have a spectral
sequence
E2pq = lim p (Rq F An ) ⇒ Rp+q (lim F )(An , dn ) ,
←− ←−
n n
.
where lim denotes the right derived functor of lim .
←− ←−
We say that a system (An , dn ) satisfies the Mittag-Leffler property if for
each n the images of the transition maps An+m → An are the same for
all sufficiently large m. Observe that every inverse system of finite abelian

groups automatically has the Mittag-Leffler property. We call a system (An , dn )

ML-zero if for each n there is an m = m(n) such that the transition map
An+m → An is zero.
(2.7.3) Proposition. Let (An , dn ) be a ML-zero system in A. Then

Rp (lim F )(An , dn ) = 0
←−
n
for all p ≥ 0.
Proof: Given a system A = (An , dn ), we define a new system A[1] by

can
A[1] = (An+1 , dn+1 ). There is a canonical map A[1] −→ A given by the
.
composition of the transition maps. If A ,→ I is an injective resolution, then
.
A[1] ,→ I[1] is too, and we obtain a commutative diagram
áÞßà
A[1] I[1] .
can can
A .
I .
∼ lim I m for all m
Since lim I[1]m −→ ≥ 0, we obtain natural isomorphisms
←− ←−
for all p ≥ 0
Rp (lim F )A[1] −→
∼ Rp (lim F )A .
←− ←−
n n
If all transition maps of A are zero, then this canonical isomorphism is the zero
map, and hence Rp (lim F )A = 0 for all p ≥ 0 in this case.
←−
Let (An ) be a ML-zero system and let J = {n1 , n2 , . . .} ⊂ IN be a cofinal
subset such that all transition maps in the inverse system (Anj )j ∈IN are zero.
Now lim F is also the composition of the exact forgetful functor
←−
VJ : AIN −→ AIN , (An ) 7→ (Anj ),

which preserves injectives, with lim F . The spectral sequence implies
←−
p
R (lim
←−
F )(An ) = Rp (lim
←−
F )(Anj ) = 0
n j
for all p ≥ 0. 2
The following well-known proposition was formulated in [182], prop. 1,

with too much generality (see [150] for a counter-example). Therefore we
give a complete proof here in the case that B is the category of modules over a
ring.

(2.7.4) Proposition. Let B be the category of R-modules for some ring R.

Then lim p = 0 for p ≥ 2, and there is a canonical exact sequence
←−
Y id−dn Y
0 −→ lim (Bn ) −→ Bn −−−→ Bn −→ lim 1 (Bn ) −→ 0.
←− ←−
n n n n
1
If B ∈ Ob(B IN ) satisfies the Mittag-Leffler property, then lim
←−
B = 0.
Proof: By the definition of lim

←−
, we have an exact sequence
Y id−dn Y
0 −→ lim
←−
(Bn ) −→ Bn −−−→ Bn ,
n n n
which is also exact on the right, if all transition maps of (Bn ) are surjective (for
any element (bn ) ∈ n Bn a pre-image under (id − dn ) is easily constructed
Q
recursively). Let
0 −→(An ) −→(Bn ) −→(Cn ) −→ 0
be an exact sequence in B IN such that all transition maps of the systems (An )
and (Bn ) are surjective. Then the transition maps of (Cn ) are also surjective
and the snake lemma shows that the sequence of inverse limits
0 −→ lim
←−
(An ) −→ lim
←−
(Bn ) −→ lim
←−
(Cn ) −→ 0
n n n
IN
is exact. Since injective objects in B have surjective transition maps, we
obtain lim p (Bn ) = 0 for all p ≥ 1 and any system (Bn ) with surjective transition
←−
maps.
If (Bn ) is an arbitrary inverse system, we consider the inverse systems (PBn )
and (QPn ) given by PBn = ni=1 Bi and QPn = n−1
Q Q
Qn+1 Qn i=1 Bi , and with the natural
projections pn : i=1 Bi → i=1 Bi as transition maps. We obtain an exact
sequence of inverse systems
0 −→(Bn ) −→(PBn ) −→(QPn ) −→ 0,
given by the commutative exact diagrams
n+1 n
(id,dn ,dn dn−1 ,...) pn −(dn prn+1 ,dn−1 prn ,...)
0èåæçäâãëéêì
Y Y
Bn+1 Bi Bi 0
i=1 i=1
dn pn pn−1
n n−1
(id,dn−1 ,dn−2 dn−1 ,...) Y pn−1 −(dn−1 prn ,dn−2 prn−1 ,...) Y
0 Bn Bi Bi 0,
i=1 i=1
p p
where pri denotes the projection onto Bi . Since lim
←−
(PBn ) = 0 = lim
←−
(QBn )
p
for p 1, we obtain lim (Bn ) = 0 for p 2 and the exact sequence stated in
≥ ≥
←−
the proposition.

Finally, let (Bn ) be a Mittag-Leffler system. We consider the subsystem (Bn0 )

of (Bn ), where, for each n, Bn0 is the image of the transition map Bn+m → Bn ,
for sufficiently large m, which is independent of m. Then the transition maps of
(Bn0 ) are surjective and the cokernel (Bn00 ) of the natural inclusion (Bn0 ) ,→ (Bn )
is ML-zero. Therefore we obtain isomorphisms
0 = lim
←−
p
(Bn0 ) −→
∼ lim p (B )
←−
n
n n
for all p ≥ 1. 2
Suppose we are given a profinite group G and a compact topological G-

module A whose underlying topology is profinite. It is not difficult to see
that such a module is the inverse limit of finite G-modules and we call such
modules profinite modules for short. The continuous cochain cohomology of
profinite modules can also be defined by means of derived functors, namely as
an Ext-group in the category of profinite G-modules.∗) We will not make use
of this in the following, but the reader should notice that this is the conceptual
reason behind the next theorem.
(2.7.5) Theorem. Suppose that the compact G-module A has a presentation

A = lim An
←−
n∈IN
as a countable inverse limit of finite, discrete G-modules. Then there exists a
natural exact sequence
0 −→ lim 1 H i−1 (G, An ) −→ Hcts
i
(G, A) −→ lim H i (G, An ) −→ 0.
←− ←−
n n
Proof: Consider the categories (G-Mod)IN and AbIN of inverse systems of

discrete G-modules and of abelian groups, respectively. Both categories are
abelian and have sufficiently many injective objects. Furthermore, the functor
H 0 (G, −) sends injective G-modules to injective (i.e. divisible) abelian groups:
indeed, this is obvious for modules of the form IndG M for a divisible module
M . If I is injective, then it is divisible and it is a direct summand of IndG I.
Let us consider the functor lim ←−
H 0 (G, −). By (2.7.4), applied to B = Ab,
the spectral sequence for the composition of derived functors degenerates to a
series of short exact sequences
0 → lim 1 H i−1 (G, An ) → Ri (lim H 0 (G, −)) (An ) → lim H i (G, An ) → 0.
←− ←− ←−
n n n
∗) H i (G, A) =
∼ Exti
cts profinite G-modules (ẐZ, A), where G acts trivially on ẐZ. This can easily
be deduced from (5.2.14).

It therefore remains to show that there are isomorphisms for all i:

Ri (lim
←−
H 0 (G, −)) (An ) ∼ i
= Hcts (G, A).
n
To prove this, let us first replace the system (An ) by its subsystem (A0n ) with
A0n =
\
im(An+m )
m
(since the Ai are finite, all intersections are finite too). The system A0n has
surjective transition maps and A ∼ = lim A0n as a topological G-module. The
←−
cokernel of the natural inclusion map (A0n ) ,→ (An ) is ML-zero. By (2.7.3),
we have isomorphisms
Ri (lim H 0 (G, −)) (A0n ) →
∼ Ri (lim H 0 (G, −)) (A )
n
←− ←−
for all i 0. Therefore, replacing (An ) by
≥ (A0n ),
we may assume that all tran-
i
sition maps of the system (An ) are surjective. By definition, Hcts (G, A) is the
i-th cohomology of the continuous homogeneous cochain complex Ccts (G, A), .
.
which can be identified with the inverse limit over n of C (G, An ). The com-
. .
plexes C (G, An ) are the G-invariants of the complexes X (G, An ), which
are acyclic resolutions of An for all n, see I §3. Moreover, these complexes are
functorial, hence we get a H 0 (G, −)IN -acyclic resolution
.
(An ) ,→ (X (G, An )),
i
from which Hcts (G, A) is obtained by applying lim H 0 (G, −) and taking
←−
cohomology. Hence it remains to show that the systems (X (G, An )) are .
lim
←−
H 0 (G, −)-acyclic, which follows easily from the fact that the systems of
abelian groups (C i (G, An )) = (X i (G, An )G ) have surjective transition maps
for all i (cf. I §3, ex.1). 2
(2.7.6) Corollary. Let A be a compact G-module having a presentation

A = lim An
←−
n∈IN
as a countable inverse limit of finite, discrete G-modules. If H i (G, An ) is finite
for all n, then
i+1
Hcts (G, A) = lim
←−
H i+1 (G, An ).
n
If A = lim An , where An is finite, then the corollary applies, for example,

←− n∈IN
1 2
to Hcts (G, A) and, if the group G is finitely generated, also to Hcts (G, A).
As we did in I §2, p.17, in the case of a finite A, we consider for a profinite

G-module A the set EXT(G, A) of equivalence classes of exact sequences of

profinite groups
1 −→ A −→ Ĝ −→ G −→ 1
such that the action of G on A is given by

σ
a = σ̂aσ̂ −1 ,
where σ̂ ∈ Ĝ is a pre-image of σ ∈ G. The same proof as that of (1.2.4) shows
also in this case the
(2.7.7) Theorem. We have a canonical bijection of pointed sets

2
Hcts (G, A) ∼
= EXT(G, A).
Let ` be a prime number and T a topological G-module which, as a topo-

logical group, is a finitely generated ZZ` -module with the natural topology, and
on which G acts ZZ` -linearly.
n
(2.7.8) Proposition. Let Y be a finitely generated ZZ` -submodule of Hcts (G, T ).
n
Then the quotient group Hcts (G, T )/Y contains no nontrivial `-divisible sub-
group.
n
Proof: (cf. [232], prop. 2.1) Suppose xi ∈ Hcts (G, T ), 0 ≤ i < ∞, such that
xi ≡ `xi+1 mod Y for all i. We must show x0 ∈ Y . Let yj , 1 ≤ j ≤ m, be a
finite set of generators for Y . For each i, let fi be an n-cocycle representing
xi , and for each j, let gj be an n-cocycle representing yj . Then there are
(n − 1)-cochains hi and elements aij ∈ ZZ` such that
m
X
fi = `fi+1 + aij gj + ∂hi .
j=1
Hence m
X
f0 = aj gj + ∂h
j=1
with aj = i≥0 ì aij and h = i≥0 ì hi . The use of infinite sums here is
P P
formally justified by the fact that T is the inverse limit of its quotients T /ì T
and consequently
. .
Ccts (G, T ) = lim C (G, T /ì T )
←−
i
is a projective limit of modules, each of which is killed by a fixed power of `.

This completes the proof. 2

n
(2.7.9) Corollary. The ZZ` -module Hcts (G, T ) is finitely generated if and only
n n
if Hcts (G, T )/`Hcts (G, T ) is finite.
Proof: In order to show the nontrivial assertion, assume that y1 , . . . , ym

n
generate Hcts (G, T ) modulo `. Putting Y = hy1 , . . . , ym i, we conclude that the
n
group Hcts (G, T )/Y is `-divisible, hence trivial by the last proposition. 2
(2.7.10) Corollary. Assume that the cohomology groups of G with coefficients

n
in finite `-primary modules are finite. Then Hcts (G, T ) is a finitely generated
ZZ` -module for all n and the canonical map
n
Hcts (G, T ) ⊗ Q` /ZZ` −→ H n (G, T ⊗ Q` /ZZ` )
is an isogeny, i.e. has finite kernel and finite cokernel.
Proof: Replacing, if necessary, T by an open submodule, we may assume

`m
that T is torsion-free. Then the exact sequence 0 → T → T → T /`m → 0
implies the exact sequence
n
0 −→ Hcts (G, T )/`m −→ H n (G, T /`m ) −→ `m Hcts
n
(G, T ) −→ 0 .
Now the statements follow from (2.7.9). 2
Suppose now that T is torsion-free. Tensoring it, over ZZ` , by the exact
sequence 0 → ZZ` → Q` → Q` /ZZ` → 0 gives an exact sequence
(∗) 0 −→ T −→ V −→ W −→ 0 ,
in which V is a finite dimensional Q` -vector space, T is an open compact
subgroup and W is a discrete divisible `-primary torsion group.
(2.7.11) Proposition. There are isomorphisms for all n

n
Hcts (G, V ) ∼ n
= Hcts (G, T ) ⊗ ZZ` Q` .
In the exact cohomology sequence associated with (∗), the kernel of the bound-
ary homomorphism
δ: H n−1 (G, W ) −→ Hcts
n
(G, T )
is the maximal divisible subgroup of H n−1 (G, W ), and its image is the torsion
n
subgroup of Hcts (G, T ).
n
Proof: Since V is a vector space over Q` , so is Hcts (G, V ) for all n. Further-
n
more, H (G, W ) is an `-torsion group for all n. By (2.7.2), we have a long

exact sequence associated to (∗). Tensoring over ZZ` with Q` implies the first
statement. Clearly,
ker(H n−1 (G, W ) → Hcts
n
(G, T )) = im(H n−1 (G, V ) → H n−1 (G, W ))
is `-divisible. On the other hand, by (2.7.8), each divisible subgroup of
H n−1 (G, W ) must be contained in the kernel. Since W is torsion, the group
im(H n−1 (G, W ) → Hcts n
(G, T )) is a torsion group. On the other hand, it is
n n
equal to the kernel of the map Hcts (G, T ) → Hcts (G, V ) and therefore must
n
contain all torsion elements of Hcts (G, T ). 2
(2.7.12) Corollary. Assume that the cohomology groups of G with coefficients

in finite `-primary modules are finite. Then
n n
rank ZZ` Hcts (G, T ) = dimQ` Hcts (G, V )
= corank ZZ` H n (G, T ⊗ Q` /ZZ` )
= rank ZZ` Hn (G, Hom(T, ZZ` )) .
Proof: The first two equalities follow from (2.7.11) and (2.7.10). Since T is
finitely generated, we have
Hom(T ⊗ Q` /ZZ` , Q` /ZZ` ) = Hom(T, Hom(Q` /ZZ` , Q` /ZZ` )) = Hom(T, ZZ` ) ,
thus (2.6.9) implies the third equality. 2
Continuous cochain cohomology was introduced by J. TATE in [232]. In

addition, we have taken several arguments from the paper [96] of U. JANNSEN.


Chapter III
Duality Properties of Profinite Groups
§1. Duality for Class Formations
Let G be a finite group. If A and B are two G-modules, the cup-product

associated with the canonical pairing
Hom(A, B) × A −→ B, (f, a) 7→ f (a),
yields a pairing
∪
Ĥ i (G, Hom(A, B)) × Ĥ n−i (G, A) −→ Ĥ n (G, B) ,
which we call again the cup-product. When B = Q/ZZ, we set
A∗ = Hom(A, Q/ZZ).
(3.1.1) Proposition. Let G be a finite group and let A be a G-module. Then

for all i ∈ ZZ, the pairing
∪
Ĥ i (G, A∗ ) × Ĥ −i−1 (G, A) −→ Ĥ −1 (G, Q/ZZ) = 1
#G
ZZ/ZZ ⊆ Q/ZZ
induces an isomorphism
∼
Ĥ i (G, A∗ ) −→ Ĥ −i−1 (G, A)∗ .
Remark: Via the identification (see (1.9.2)) Hi = Ĥ −i−1 , i > 0, we recover

the isomorphism H i (G, A∗ ) ∼
= Hi (G, A)∗ from (1.9.1).
Proof: First let i = 0. A homomorphism f : A → Q/ZZ is a G-homo-

morphism, i.e. f ∈ H 0 (G, A∗ ), if and only if f (IG A) = 0. Therefore the map
H 0 (G, A∗ ) −→ H0 (G, A)∗ ,
which associates to a G-homomorphism f : A → Q/ZZ the induced ho-
momorphism g : A/IG A → Q/ZZ, is an isomorphism. If f ∈ NG A∗ , i.e.
f = σ∈G σh for h ∈ A∗ , then for a ∈ NG A we have f (a) = (σh)(a)
P P
= h(σ −1 a) = h(NG a) = 0. This shows that we have a well-defined map

P
A∗G /NG A∗ −→ (NG A/IG A)∗ .

148 Chapter III. Duality Properties of Profinite Groups
For the surjectivity, let g : NG A → Q/ZZ be a homomorphism that vanishes on

IG A. Since Q/ZZ is an injective ZZ-module, it can be extended to a homomor-
phism g : A → Q/ZZ, which is a G-homomorphism because g(IG A) = 0. This
shows the surjectivity.
For the injectivity, let f ∈ A∗G be such that f (NG A) = 0. As
NG : A/NG A −→ NG A
is an isomorphism, there exists a g ∈ (NG A)∗ such that f (a) = g(NG a) for
all a ∈ A. We may extend g to a homomorphism g : A → Q/ZZ and then
f = NG g, since
g(σ −1 a) = g(NG a) = f (a).
X
(NG g)(a) =
σ ∈G
This proves the injectivity.

Now let i ∈ ZZ be arbitrary. We use dimension shifting (see I §3, p.32) in
order to reduce to the case i = 0. From (1.4.6) we obtain the commutative
diagram
ïîíñð
Ĥ i (G, Hom(A, Q/ZZ)) × Ĥ −i−1 (G, A) ∪
Ĥ −1 (G, Q/ZZ)
δi δi (−1)i(i+1)/2
∪
Ĥ 0 (G, Hom(A, Q/ZZ)i ) × Ĥ −1 (G, A−i ) Ĥ −1 (G, Q/ZZ).
Since Hom(A, Q/ZZ)i ∼

= Hom(A−i , Q/ZZ), the desired result follows. 2
Next, we consider the case B = ZZ and get the following
(3.1.2) Proposition. If G is a finite group and A is a ZZ-free G-module, then

for all integers i ∈ ZZ, the pairing
∪
Ĥ i (G, Hom(A, ZZ)) × Ĥ −i (G, A) −→ Ĥ 0 (G, ZZ) = ZZ/#GZZ
yields an isomorphism
Ĥ i (G, Hom(A, ZZ)) ∼
= Ĥ −i (G, A)∗ .
Proof: Since A is ZZ-free, the sequence

0 −→ Hom(A, ZZ) −→ Hom(A, Q) −→ Hom(A, Q/ZZ) −→ 0
is exact. Multiplication by #G on Hom(A, Q) is an isomorphism, hence also
on the group Ĥ n (G, Hom(A, Q)), which is therefore zero by (1.6.1). For this
reason, we get a commutative diagram

§1. Duality for Class Formations 149
ôóòöõ ZZ)) × Ĥ −i (G, A)

Ĥ i−1 (G, Hom(A, Q/ ∪
Ĥ −1 (G, Q/ZZ)
δ id δ
∪
Ĥ i (G, Hom(A, ZZ)) × Ĥ −i (G, A) Ĥ 0 (G, ZZ),
in which the vertical arrows are isomorphisms. Hence the proposition follows
from (3.1.1). 2
(3.1.3) Definition. Let G be a finite group. We call a G-module C a class

module if for all subgroups H of G
(i) H 1 (H, C) = 0 and
(ii) H 2 (H, C) is cyclic of order #H.
A generator γ of H 2 (G, C) is called a fundamental class.
Obviously, C is also a class module for every subgroup H of G. Note that

if γ is a generator of H 2 (G, C), then γH = res G 2
H γ is a generator of H (H, C):
since cor ◦ res γ = (G : H)γ, the order of γH is divisible by #H. If U ⊆ G is
normal, then C U is not necessarily a class module for G/U .
If G is cyclic, then ZZ is a class module because of the isomorphisms
(cf. (1.7.1)) Ĥ 1 (H, ZZ) ∼
= Ĥ −1 (H, ZZ) = 0 and H 2 (H, ZZ) ∼
= Ĥ 0 (H, ZZ) =
ZZ/(#H)ZZ. We have seen in (1.7.1) that the cup-product with a fundamental
class is a periodicity operator for the cohomology of G, i.e.
γ∪ : Ĥ n (G, A) −→
∼ Ĥ n+2 (G, A)
for every G-module A and for all n ∈ ZZ.

We relate the property of being a class module to the property of cohomolo-
gical triviality. To each G-module C and each class γ ∈ H 2 (G, C) we associate
L
a G-module C(γ) as follows. Let B = σ=/1 ZZbσ be the free abelian group
with basis bσ , indexed by the elements σ ∈ G, σ =/ 1. We set
C(γ) = C ⊕ B,
and we let G act on C(γ) by means of an inhomogeneous cocycle c(σ, τ )
representing γ as follows: we set b1 = c(1, 1) and define
σbτ = bστ − bσ + c(σ, τ ).
This is really a G-action, i.e. (ρσ)bτ = ρ(σbτ ) and 1bτ = bτ , because of the
cocycle relation ρc(σ, τ ) − c(ρσ, τ ) + c(ρ, στ ) − c(ρ, σ) = 0 ∗) . The G-module
C(γ) is constructed in such a way that the cocycle c(σ, τ ) becomes necessarily
∗) The reader may check that, up to isomorphism, C(γ) depends only on the class γ, not on
the cocycle c(σ, τ ).

in C(γ) a coboundary c(σ, τ ) = σb(τ ) − b(στ ) + b(σ), where the 1-cochain b(σ)
is defined by b(σ) = bσ . Thus H 2 (G, C) → H 2 (G, C(γ)) maps γ to zero. C(γ)
is therefore called the splitting module of γ. ∗)
We obtain a four term exact sequence
i ϕ
0 −→ C −→ C(γ) −→ ZZ[G] −→ ZZ −→ 0
of G-modules if we define ϕ by
ϕ(c) = 0 for c ∈ C and ϕ(bσ ) = σ − 1 for σ =/ 1.
Splitting up this sequence into two short exact sequences, we obtain, for each
n ∈ ZZ and each subgroup H ⊆ G, a homomorphism
δ 2 : Ĥ n (H, ZZ) −→ Ĥ n+2 (H, C).
(3.1.4) Theorem. Let G be a finite group. For each n ∈ ZZ and each subgroup
H ⊆ G, the homomorphism
δ 2 : Ĥ n (H, ZZ) → Ĥ n+2 (H, C)
is given by the cup-product β 7→ γH ∪ β, where γH = res G H (γ). The following
conditions are equivalent.
(i) C(γ) is a cohomologically trivial G-module,
(ii) C is a class module with fundamental class γ,
(iii) δ 2 is an isomorphism for all n ∈ ZZ and all H.
Remark: If C is a class module for G, then, by the above theorem, we have

isomorphisms
(δ 2 )−1 : H 2 (H, C) −→
∼ 1
#H
ZZ/ZZ, γH 7→ #H1
mod ZZ,
where γ ∈ H 2 (G, C) is a chosen fundamental class. These are called invariant
maps and denoted by inv.
Proof: The map δ 2 arises from the two exact sequences

(1) 0 −→ IG −→ ZZ[G] −→ ZZ −→ 0,
(2) 0 −→ C −→ C(γ) −→ IG −→ 0
and is the composite of the maps
δ δ
(3) Ĥ n (H, ZZ) −→1
∼ Ĥ
n+1 2
(H, IG ) −→ Ĥ n+2 (H, C),
where δ1 is always an isomorphism. For n = 0 we have the maps
δ δ
(4) ZZ/#H ZZ = Ĥ 0 (H, ZZ) −→
1
H 1 (H, IG ) −→
2
H 2 (H, C).
∗) The module C(γ) also arises as the extension corresponding to the class γ ∈ H 2 (G, C)
∼ Ext2
via the isomorphisms H 2 (G, C) = ∼ Ext1
Z, C) =
ZZ [G] (Z ZZ [G] (IG , C).

For the generator 1̄ = 1 mod #H of Ĥ 0 (H, ZZ), we have

(5) δ2 δ1 1̄ = γH = res G
H γ.
In fact, a pre-image of the 0-cocycle 1 ∈ Z 0 (H, ZZ) in C 0 (H, ZZ[G]) is 1 ∈ ZZ[G]
and δ1 1̄ ∈ H 1 (H, IG ) is represented by the 1-cocycle (∂1)(σ) = σ − 1. A lift of
∂1 in C 1 (H, C(γ)) is given by x(σ) = bσ , and δ2 δ1 1̄ is represented by
(∂x)(σ, τ ) = σbτ − bστ + bσ = c(σ, τ ) .
This proves (5). Now let β ∈ Ĥ n (H, ZZ), where n is arbitrary. Applying
proposition (1.4.3) to B = ZZ and to the two exact sequences (1) and (2), we
obtain
δ 2 β = δ2 δ1 (1̄ ∪ β) = δ2 (δ1 1̄ ∪ β) = δ2 δ1 1̄ ∪ β = γH ∪ β.
Noting that Ĥ i (H, IG ) = Ĥ i−1 (H, ZZ) = 0 for i = 0 and 2, we obtain from (2)
the exact sequence
0 −→ H 1 (H, C) −→ H 1 (H, C(γ)) −→ H 1 (H, IG )
δ
−→ H 2 (H, C) −→ H 2 (H, C(γ)) −→ 0.
If C(γ) is cohomologically trivial, then H 1 (H, C) = 0 and the composite
δ δ
ZZ/#H ZZ = Ĥ 0 (H, ZZ) −→ 1 2
∼ H (H, IG ) −→ H (H, C)
is an isomorphism which maps 1̄ onto γH . Therefore C is a class module

with fundamental class γ. Conversely, if this is true, then H 1 (H, C) = 0 and
δ : H 1 (H, IG ) → H 2 (H, C) is an isomorphism; hence H i (H, C(γ)) = 0 for
i = 1, 2, and therefore for all i by (1.8.4).
The equivalence (i) ⇔ (iii) follows from (3) and from the exact sequence
. . . −→ Ĥ n (H, C) −→ Ĥ n (H, C(γ)) −→ Ĥ n (H, IG )
δ
−→ Ĥ n+1 (H, C) −→ Ĥ n+1 (H, C(γ)) −→ . . . 2
For ZZ-free G-modules A we can now prove the following duality theorem.
(3.1.5) Theorem (NAKAYAMA-TATE). Let G be a finite group, let C be a class

module for G and let γ ∈ H 2 (G, C) be a fundamental class. Then, for all
integers i ∈ ZZ, the cup-product
∪
Ĥ i (G, Hom(A, C)) × Ĥ 2−i (G, A) −→ H 2 (G, C) ∼
= 1
#G
ZZ/ZZ,
where H 2 (G, C) ∼
= 1
#G
ZZ/ZZ is given by γ 7→ 1
#G
mod ZZ, induces an isomor-
phism
Ĥ i (G, Hom(A, C)) ∼
= Ĥ 2−i (G, A)∗ ,
provided that A is ZZ-free. If, in addition, A is finitely generated, then this is
an isomorphism of finite abelian groups.

Proof: Let 0 −→ C −→ C(γ) −→ IG −→ 0 be an exact sequence as in the

proof of (3.1.4). As A is ZZ-free, the sequences
0 −→ Hom(A, C) −→ Hom(A, C(γ)) −→ Hom(A, IG )−→ 0
0 −→ Hom(A, IG )−→ Hom(A, ZZ[G]) −→ Hom(A, ZZ) −→ 0
are exact and the G-modules in the middle are cohomologically trivial by
(1.8.5), (3.1.4), and (1.3.7). We now have for i ∈ ZZ a commutative diagram
úûùø÷ÿþýü
Ĥ i−2 (G, Hom(A, ZZ)) × Ĥ 2−i (G, A) ∪
Ĥ 0 (G, ZZ)
δ id δ
∪
Ĥ i−1 (G, Hom(A, IG )) × Ĥ 2−i (G, A) Ĥ 1 (G, IG )
δ id δ
∪
Ĥ i (G, Hom(A, C)) × Ĥ 2−i (G, A) Ĥ 2 (G, C),
where the vertical arrows are isomorphisms. Hence the asserted isomorphism
follows from (3.1.2). If A is finitely generated, then the cohomology groups
Ĥ i (G, A) are finite, and so are their duals, by (1.6.1). 2
For a subgroup H of G we have commutative diagrams

,
Ĥ i (H, Hom(A, C)) Ĥ 2−i (H, A)∗
res cor cor ∗ res ∗
Ĥ i (G, Hom(A, C)) Ĥ 2−i (G, A)∗

where cor ∗ and res ∗ are the maps dual to the corestriction and restriction, and
the upper map relies on the choice of the fundamental class γH ∈ Ĥ 2 (H, C).
The commutativity of the diagram follows at once from cor (γ ∪ res β) =
(cor γ) ∪ β (see (1.5.3)(iv)) and inv ◦ cor = inv.
Applying the duality theorem to the case A = ZZ, i = 0, and recalling
H 2 (G, ZZ)∗ ∼
= H 1 (G, Q/ZZ)∗ = Hom(Gab , Q/ZZ)∗ = Gab ,
we obtain
(3.1.6) Theorem. If C is a class module for the finite group G, then we have
an isomorphism ∼
ρ = ρG : Gab −→ C G /NG C,
called the Nakayama map. It depends on the choice of a fundamental class
γ ∈ H 2 (G, C) and satisfies (by definition) the formula
χ(σ) = inv(ρ(σ) ∪ δχ)
δ
for all characters χ ∈ H 1 (G, Q/ZZ) −→ 2
∼ H (G, Z
Z).

The inverse map

∼
ρ−1 : C G /NG C −→ Gab
is often called the reciprocity isomorphism. We call its composite with the
natural projection C G C G /NG C, i.e. the map
rec = rec G : C G Gab ,
the reciprocity homomorphism.

The reciprocity homomorphism is also called the norm residue symbol and,
if α ∈ C G , we also write
(α, G) := rec(α).
The name norm residue symbol reflects the fact that this symbol detects whether
α ∈ C G is the norm of an element in C. Indeed, (α, G) = 0 if and only if
α ∈ NG C.
The isomorphism Gab ∼ = C G /NG C is an abstract version of class field
theory. If L|K is a finite Galois extension of local or global fields, then the
multiplicative group L× is a class module in the local case, and the idèle class
group CL in the global case, as we shall see. Combined with the above theorem
this gives the main theorem of local and global class field theory.
The following theorem gives an explicit description of the Nakayama map.
(3.1.7) Theorem. If C is a class module for the finite group G and γ ∈

H 2 (G, C) is a fundamental class, then the Nakayama map
∼
ρ : Gab −→ C G /NG C
is explicitly given by
X
σ mod [G, G] 7−→ c(τ, σ) mod NG C,
τ
where c is a cocycle representing γ.
Proof: We will prove the theorem by means of the diagram
G
ab × H 1 (G, Q/ZZ) 1
#G
ZZ/ZZ
(∗) ν ρ δ inv
∪
Ĥ 0 (G, C) × H 2 (G, ZZ) H 2 (G, C),

in which the upper horizontal arrow is the evaluation map (σ, χ) 7→ χ(σ),
1
inv(γ) = #G mod ZZ, where ρ is the Nakayama isomorphism and ν is the map
given by
σ̄ 7−→ c(τ, σ) mod N C , ∗)
X
G
τ
where σ̄ = σ mod [G, G]. It suffices to show that the diagram is commutative
with both maps ν and ρ, since then theorem (3.1.5) identifies ν and ρ with the
dual δ ∗ of δ. The commutativity of the diagram with the map ρ follows from
the definition of ρ and is an equivalent version of the formula
χ(σ) = inv(ρ(σ) ∪ δχ) ,
see (3.1.5). It remains to show the commutativity for ν. For this we consider
the larger diagram
G ab × H 1 (G, Q/ZZ) Ĥ −1 (G, Q/ZZ) = 1
#G
ZZ/ZZ
δ1 id δ3
∪
Ĥ −1 (G, IG ) × H 1 (G, Q/ZZ) Ĥ 0 (G, IG ⊗ Q/ZZ)
id δ −δ4
∪
Ĥ −1 (G, IG ) × H 2 (G, ZZ) H 1 (G, IG )
δ2 id δ5
∪
Ĥ 0 (G, C) × H 2 (G, ZZ) H 2 (G, C),
in which δ1 : Gab → Ĥ −1 (G, IG ) = IG /IG2 is defined by σ̄ 7→ σ − 1 mod IG2
and the other δi ’s arise from the exact sequences
(1) 0 −→ IG −→ ZZ[G] −→ ZZ −→ 0,
(2) 0 −→ IG ⊗ Q/ZZ −→ ZZ[G] ⊗ Q/ZZ −→ Q/ZZ −→ 0,
(3) 0 −→ IG −→ IG ⊗ Q −→ IG ⊗ Q/ZZ −→ 0,
(4) 0 −→ C −→ C(γ) −→ IG −→ 0.
We make the homomorphisms δ explicit by choosing in (1) - (4) sections of
the last arrows, by which the cocycles with coefficients in the last groups are
sent to cochains with coefficients in the middle groups, to which we apply the
coboundary operator ∂. For example, the map
M
C(γ) = C ⊕ ZZbσ −→ IG
σ=
/1
∗) Note that the sum is contained in C G because of the cocycle relation ρc(τ, σ) = c(ρτ, σ) +
c(ρ, τ ) − c(ρ, τ σ) and the fact that ρτ and τ σ run through G if τ does.

is defined by C → 0 and bσ 7→ σ − 1. It has the section σ − 1 7→ bσ ,

and b : G → C(γ), σ 7→ bσ , is an inhomogeneous cochain with coboundary
(∂ 2 b)(σ, τ ) = c(σ, τ ), which lifts the 1-cocycle x : G → IG , x(σ) = σ − 1.
We set g = #G, ā = a mod ZZ for a ∈ Q with ga ∈ ZZ. Noting that the elements
in (IG ⊗ Q/ZZ)G are of the form τ (τ − 1) ⊗ ā = NG ⊗ ā, the δ’s are given as
P
follows
δ1 : σ̄ 7−→ σ − 1 mod IG2 ,
X
δ2 : σ − 1 7−→ c(τ, σ) mod NG C,
τ
X
δ3 : ā 7−→ (τ − 1) ⊗ ā,
τ
−δ4 : NG ⊗ ā 7−→ gax,

δ5 : x 7−→ c.
The first assertion holds by definition and the others rely on the relations
X X X X X
∂ 0 bσ = τ bσ = bτ σ − bτ + c(τ, σ) = c(τ, σ),
τ τ τ τ τ
X X
∂ 0 (1 ⊗ ā) = NG (1 ⊗ ā) = τ ⊗ ā = (τ − 1) ⊗ ā,
τ τ
X
∂ 1( (τ − 1) ⊗ a)(ρ) = (ρ − 1)[(NG ⊗ a) − ga1] = −ga(ρ − 1),
τ
(∂ 2 b)(σ, τ ) = σbτ − bστ + bσ = c(σ, τ ).
This result shows that the composite −δ5 δ4 δ3 is inverse to inv, since it maps
1
g
mod ZZ to γ.
The upper partial diagram is commutative, because δ1 (σ) ∪ χ is represented
by
τ (σ − 1) ⊗ χ(τ −1 ) = τ σ ⊗ χ(τ −1 ) − τ ⊗ χ(τ −1 )
X X X
τ τ τ
τ ⊗ χ(στ −1 ) − τ ⊗ χ(τ −1 )
X X
=
τ τ
X X
= τ ⊗ χ(σ) = (τ − 1) ⊗ χ(σ)
τ τ
(see (1.4.8)), hence by the same element as δ3 (χ(σ)). The middle and the lower
partial diagrams are commutative by (1.4.3) and the remark following (1.4.6).
This proves that the diagram (∗) is commutative. 2
From now on let G be a profinite group. We extend the notion of “class

module”, which we needed for the duality theorem (3.1.5), to profinite groups
as follows. We denote open subgroups of G by the letters U, V, W .

(3.1.8) Definition. Let G be a profinite group. A formation module for G is

a discrete G-module C together with a system of isomorphisms
invU/V : H 2 (U/V, C V ) −→
∼ 1
(U :V )
ZZ/ZZ
for every pair V ⊆ U of open subgroups, V normal in U , such that the following
conditions hold.
(i) H 1 (U/V, C V ) = 0.
(ii) For open normal subgroups W ⊆ V of an open subgroup U , the diagram
C V )
H 2 (U/V,
inf
H 2 (U/W, C W ) res
H 2 (V /W, C W )
inv inv inv
1 incl 1 (U :V ) 1
(U :V )
ZZ/ZZ (U :W )
ZZ/ZZ (V :W )
ZZ/ZZ
is commutative.
The pair (G, C) is called a class formation.
Remarks: 1. For finite G the notion of a formation module is stronger than that
of a class module because there is no compatibility condition for the passage
to quotients for the latter.
2. From (ii) it follows that the diagram
$#"! C W )
H 2 (V /W, inv 1
ZZ/ZZ
(V :W )
cor incl
inv 1
H 2 (U/W, C W ) (U :W )
ZZ/ZZ
is commutative because cor VU ◦ res UV = (U : V ).
3. The isomorphisms
inv : H 2 (G/V, C V ) −→
∼ 1
(G:V )
ZZ/ZZ
form a direct system. Passing to the direct limit, we obtain a homomorphism
inv : H 2 (G, C) −→ Q/ZZ,
1
which is called the invariant map. It is injective and its image is #G
ZZ/ZZ =
1
lim (G:V )
ZZ/ZZ.
−→ V
4. In terms of G-modulations (see I §5), a formation module for G is a discrete
G-module C together with an isomorphism
inv : H 2 (C) → Ĥ 0 (ZZ)∗
of G-modulations, which has the additional property that H 1 (C) = 0.

5. For every open normal subgroup U ⊆ G, the module of invariants C U is a

class module for G/U . The elements inv−1 1
Z) ∈ H 2 (G/U, C U )
G/U ( (G:U ) mod Z
form a compatible system of fundamental classes for varying U . We there-
fore obtain reciprocity homomorphisms recG/U : C G (G/U )ab which are
compatible for open normal subgroups V ⊆ U in the sense that recG/U is the
composition of recG/V with the natural projection (G/V )ab (G/U )ab . Pass-
ing to the projective limit, we therefore obtain the reciprocity homomorphism
rec = recG : C G −−−→ Gab ,
which is also called the norm residue symbol. It has dense image and kernel
NG C.
Our aim is to prove a version of theorem (3.1.5) for profinite groups. We

consider a discrete G-module A which is finitely generated as a ZZ-module.
Then, for any discrete G-module B, Hom(A, B) is a discrete G-module and
the pairing
Hom(A, B) × A −→ B, (f, a) 7−→ f (a),
induces a cup-product
Ĥ i (G, Hom(A, B)) × Ĥ 2−i (G, A) −→ H 2 (G, B)
for all i ∈ ZZ: for i = 1 this is the usual cup-product and for i =/ 1 it is the
cup-product constructed in chap. I, §9, see (1.9.7).
We recall the definition of the topology on the Tate cohomology groups of
A from chap. I, §9. By (1.6.1), the groups Ĥ i (G/U, AU ) are finite for all i and
all open normal subgroups U of G. Therefore the groups
Ĥ i (G, A) = lim Ĥ i (G/U, AU ), i ≤ 0,

←−
U, def
are naturally equipped with a profinite topology. For i > 0 we give Ĥ i (G, A)
the discrete topology. Now assume that C is a formation module for G and
that A is free as a ZZ-module. For sufficiently small U we have AU = A, and
hence Hom(A, C U ) = Hom(A, C)U . Therefore the groups
Ĥ i (G, Hom(A, C)) = lim Ĥ i (G/U, Hom(A, C)U ), i ≤ 0,

←−
U, def
are naturally equipped with a profinite topology, as the groups on the right hand
side are finite by (3.1.5). For i ≥ 1, we give Ĥ i (G, Hom(A, C)) the discrete
topology.

(3.1.9) Theorem. Let G be a profinite group and let C be a formation module

for G. Let A be a discrete G-module which is finitely generated and free as a
ZZ-module. Then the cup-product
&% × Ĥ 2−i (G, A)
Ĥ i (G, Hom(A, C)) ∪
H 2 (G, C) 1
ZZ/ZZ ,
inv #G
induces a topological isomorphism

∼
Ĥ i (G, Hom(A, C)) −→ Ĥ 2−i (G, A)∨ .
for all i ∈ ZZ.
Proof: For sufficiently small U we have AU = A and therefore Hom(A, C U ) =

Hom(A, C)U . Thus, for i =/ 1, the theorem follows easily from (3.1.5) and
(1.9.6) by passing to the limit over G/U , where U runs through the open
normal subgroups in G. For i = 1, let U ⊆ G be an open normal subgroup
which acts trivially on A. Since A is ZZ-free, H 1 (U/V, AV ) = 0 for every open
normal subgroup V ⊆ U and therefore
H 1 (G/U, AU ) →
∼ H 1 (G/V, AV ).
Using the fact that C is a formation module for G, we obtain

H 1 (U/V, Hom(A, C)V ) = H 1 (U/V, Hom(A, C V )) ∼ = H 1 (U/V, C V )rank ZZ A = 0.
This implies that
H 1 (G/U, Hom(A, C)U ) →
∼ H 1 (G/V, Hom(A, C)V ).
A stationary limit process shows that (3.1.5) implies the desired result also in
dimension 1. 2
Recall from (1.9.10) that a level-compact G-module is a discrete G-module C

with a natural compact topology on C U for each open subgroup U ⊆ G.
(3.1.10) Corollary. Let G be a profinite group and let C be a level-compact

formation module for G. Then the reciprocity homomorphism
rec : C G −→ Gab
is surjective.
Proof: We apply (3.1.9) for i = 0 and A = ZZ. The reciprocity map rec is the
composite
∼
C G −→ Ĥ 0 (G, C) −→ H 2 (G, ZZ)∨ = Gab .
If C is level-compact, then, by (1.9.11), the first arrow is surjective. 2

The central result of this section is the following duality theorem for level-
compact class formations, which generalizes (3.1.9) to the case when A is not
necessarily free. In what follows the reader should note the following
Topological Remarks: Let A be a discrete G-module which is finitely gener-

ated as a ZZ-module and let C be a level-compact G-module.
For sufficiently small V we have AV = A and therefore
Hom(A, C)V = Hom(A, C V ).
Considering A as a group with the discrete topology, Hom(A, C V ) endowed
with the compact open topology is compact. For an arbitrary open sub-
group U ⊆ G, the group Hom(A, C)U is a closed subgroup of the compact
group Hom(A, C)V , where V is chosen sufficiently small. We conclude
that Hom(A, C) is level-compact in a natural way. In particular, the groups
Ĥ i (G, Hom(A, C)) carry a natural profinite topology for i ≤ 0. If, as happens
in the situation of theorem (3.1.9), the groups Ĥ i (G/U, Hom(A, C)U ) are fi-
nite for all U , then the knowledge of the topology of the groups Hom(A, C)U
is redundant for an understanding of the profinite topology of the groups
Ĥ i (G, Hom(A, C)), i ≤ 0. Finally, recall that Ĥ i (G, Hom(A, C)) carries the
discrete topology for i ≥ 1 by convention.
In addition, assume that C is a formation module. Let U ⊆ G be an open
normal subgroup such that AU = A. Then, for i ∈ ZZ, the cup-product
Ĥ i (G/U, Hom(A, C)U'( ) × Ĥ 2−i (G/U, AU ) ∪ H 2 (G/U, C U ) inv (G:U
1
)
ZZ/ZZ
is continuous with respect to the topologies induced on the cohomology groups
by the complete standard complex. As the group H 2 (G/U, C U ) is compact
and finite, it is discrete. The groups Ĥ 2−i (G/U, AU ) are discrete and finite by
(1.6.1). Thus, for each i ≤ 0, we obtain a continuous homomorphism
Ĥ i (G/U, Hom(A, C)U ) −→ Ĥ 2−i (G/U, AU )∨
from a compact group to a finite discrete group. Passing to the projective limit
over all U , we obtain a continuous homomorphism
Ĥ i (G, Hom(A, C)) −→ H 2−i (G, A)∨
between compact abelian groups. For i ≥ 2 and U as above, the cup-product
defines a homomorphism
H i (G/U, Hom(A, C)U ) −→ Ĥ 2−i (G/U, A)∨ .
Passing to the direct limit, we obtain the (continuous) homomorphism
H i (G, Hom(A, C)) −→ Ĥ 2−i (G, A)∨
between discrete groups. Finally, for i = 1, the usual cup product induces a
(continuous) homomorphism
H 1 (G, Hom(A, C)) −→ H 1 (G, A)∨
from a discrete to a compact group.

(3.1.11) Duality Theorem. Let G be a profinite group and let C be a level-

compact formation module for G such that the group of universal norms NU C
is divisible for all open subgroups U of G. Let A be a discrete G-module
which is finitely generated as a ZZ-module. Consider the cup-product
)* × Ĥ 2−i (G, A) ∪ H 2 (G, C)
Ĥ i (G, Hom(A, C)) 1
Z/ZZ .
inv #G Z
The following assertions are true:
(i) Let i ≤ 0. Then the pairing above induces a topological isomorphism
∼
Ĥ i (G, Hom(A, C)) −→ Ĥ 2−i (G, A)∨ .
This holds true for all i ∈ ZZ if A is ZZ-free.
(ii) Let i ∈ ZZ and let p be a prime number with p∞ |#G. Assume that C is
p-divisible. Then the pairing above induces an isomorphism
∼
Ĥ i (G, Hom(A, C))(p) −→ Ĥ 2−i (G, A)(p)∨ .
Remark: Part (ii) is only interesting if A has p-torsion because otherwise the
statement follows easily from (i). The isomorphism in (ii) is a topological
isomorphism for i =/ 1. For i = 1 it is an algebraic isomorphism from a discrete
group onto a compact group. It is a topological isomorphism if and only if
H 1 (G, A)(p) is finite. This is true in our main application, when G is the Galois
group of the maximal extension of a global field which is unramified outside a
given finite set of places.
Proof: If A is ZZ-free, assertion (i) follows from (3.1.9). For arbitrary A,

there exists an exact sequence
0 −→ R −→ F −→ A −→ 0
of G-modules, where R and F are finitely generated and ZZ-free. Applying the
functor Hom(−, C), we obtain an exact sequence of level-compact modules
0 −→ Hom(A, C) −→ Hom(F, C) −→ Hom(R, C) .
Let U run through the open normal subgroups such that F U = F . Then
NU Hom(F, C) = Hom(F, NU C), NU Hom(R, C) = Hom(R, NU C)
and the map
NU Hom(F, C) −→ NU Hom(R, C)
is surjective since NU C is divisible. For i ≤ 0 we obtain a commutative
diagram in which we write ∼ for Hom(−, C):
+,-./01234567 F̃ )
Ĥ i−1 (G, Ĥ i−1 (G, R̃) Ĥ i (G, Ã) Ĥ i (G, F̃ ) Ĥ i (G, R̃)
H 3−i (G, F )∨ H 3−i (G, R)∨ H 2−i (G, A)∨ H 2−i (G, F )∨ H 2−i (G, R)∨ .

The upper row is exact by (1.9.13) and the lower row is exact, too. The
vertical arrows, except the middle one, are isomorphisms of compact abelian
groups by (3.1.9). Furthermore, using (1.4.6), it is easy to see that all squares
commute up to sign. Hence, by the five lemma (in the abelian category of
compact abelian groups), the middle one is also a topological isomorphism.
This proves assertion (i).
In order to prove (ii), we may, without loss of generality, assume that the
torsion part of A consists only of p-torsion. Since C is p-divisible, we now
obtain the exact sequence
0 −→ Hom(A, C) −→ Hom(F, C) −→ Hom(R, C) −→ 0,
and, using (1.9.14), the long exact cohomology sequence which is unbounded
in both directions
· · · −→ Ĥ i (G, Ã)(p) −→ Ĥ i (G, F̃ )(p) −→ Ĥ i (G, R̃)(p) −→ · · ·
in which we wrote ∼ for Hom(−, C). Furthermore, (1.9.9) gives us a cor-
responding long exact cohomology sequence associated to the short exact
sequence 0 → R → F → A → 0 in which all groups are compact or discrete
and all maps are continuous. Hence this sequence remains exact after taking
Pontryagin duals. Thus the duality map of (3.1.11) defines a map between two
long exact sequences. This map commutes up to signs with the differentials.
For the square
Ĥ 0 (G,89:; R̃)(p) H 1 (G, Ã)(p)
H 2 (G, R)∨ (p) H 1 (G, A)∨ (p)

this follows from the surjectivity of the map H 0 (G, R̃) → Ĥ 0 (G, R̃) by
(1.9.11). For the square
H 1 (G,<=>? R̃)(p) H 2 (G, Ã)(p)
H 1 (G, R)∨ (p) Ĥ 0 (G, A)∨ (p)

it follows from the isomorphism Ĥ 0 (G, A)(p) ∼ = H 0 (G, A) ⊗ ZZp by (1.9.8).
That all other squares commute up to sign is obvious by the definition of
the maps occurring and by the compatibilities proved in I §4. Therefore the
statement (ii) for A follows from that for F and R and from the five lemma.
2
Remark: Theorem (3.1.11)(i), together with the idea of the proof, was formu-
lated by G. POITOU in 1966 (see [168]) in order to prove an important duality
theorem over local and global fields that was announced without proof by
J. TATE in 1962 (see [229]). In 1969, a proof based on the same idea was

independently given by K. UCHIDA (see [236]). A similar proof was presented

by K. HABERLAND in 1978 (see [72]), but this, however, is not without mis-
takes. UCHIDA’s proof is correct, but rather terse, and we have given a detailed
account of his arguments. These papers, however, all failed to prove the cen-
tral fact that the cup-product for the Tate cohomology of profinite groups is
well-defined (see I §9). This gap was filled in [196], where also the extension
of the theorem by assertion (ii) was proven.
We mention a further feature of a class formation (G, C), which plays an

important role in “non-abelian class field theory”. Assume that C has trivial
universal norms, i.e. NU C = V ⊆U NU/V C V = 1 for every open subgroup U
T
of G, and that C is a topological G-module such that C V /C U is compact for

every pair V ⊆ U of open subgroups. For every triple W ⊆ V ⊆ U of open
subgroups, W, V normal in U , we may consider the diagram
1GHIJFE@ABCD CW W(U/W ) U/W 1
(1) NV /W ϕV /W π
1 CV W(U/V ) U/V 1,
where the horizontal sequences are the group extensions defined by the fun-
damental classes uU/W ∈ H 2 (U/W, C W ) and uU/V ∈ H 2 (U/V, C V ). One can
show that the diagram of solid arrows can be commutatively completed by an
arrow ϕV /W . Because of the compactness assumption, one can moreover show
that there exists a transitive family of arrows ϕV /W , i.e. ϕW/W 0 ◦ϕV /W = ϕV /W 0
for U ⊇ V ⊇ W ⊇ W 0 . It is therefore possible to take the (left exact) projective
limit, which yields an exact sequence
f g
(2) 1 −→ C × −→ W −→ G
with C × = lim C V ; the right arrow has a dense image. The group W is
←− V
called the Weil group of the class formation (G, C). W is a topological group
and has the following properties.
(i) Let W(U ) := g −1 (U ) and let W(U )c be the closure of the commutator
subgroup of W(U ). If V is open and normal in U , then
W(U/V ) = W(U )/W(V )c ,
and we have a commutative exact diagram
1QRSTPOKLMN C× W(U ) U
N ϕ π
1 CV W(U/V ) U/V 1,
∼ W(U )ab .
and, in particular, a canonical isomorphism ρU : C U →

(ii) For every pair of open subgroups V ⊆ U and every σ ∈ W, the diagrams
CXWVU V C\[ZY U
ρV ρU
W(V )ab W(U )ab
incl Ver σ∗ cσ
ρU σ ρU σ
CU W(U )ab , CU W(U σ )ab
are commutative, where U σ = σU σ −1 , σ ∗ is the action of g(σ) ∈ G on C and
cσ (x) = σxσ −1 .
The Weil group W is determined by these properties up to isomorphism in the

following sense. If W 0 is another topological group with these properties, then
there is an isomorphism W → W 0 , compatible with the above structures, and
this isomorphism is uniquely determined up to an inner automorphism of W 0
by an element of C × .
The proofs of these assertions are rather deep. Since we will not make use of
the Weil group in this book, we refer for the details to [6], chap.14 and [124],
chap.IX, 3.
Exercise 1. If G is a finite group and C a class module, then H 3 (G, C) = 1 and H 4 (G, C) ∼
=
Hom (G, Q/ZZ).
Exercise 2. Let G be a finite group, C a class module and A a ZZ-free G-module. Show that
every fundamental class γ ∈ H 2 (G, C) defines an isomorphism
γH ∪ : Ĥ n (H, A) → Ĥ n+2 (H, A ⊗ C)
for every n ∈ ZZ and every subgroup H.
Exercise 3. Show that the duality theorem (3.1.11)(i) may be interpreted as an isomorphism
of G-modulations
Ĥ i (Hom(A, C)) ∼= Ĥ 2−i (A)∨ .
Exercise 4. Let G be a finite group, A a G-module and α ∈ H 2 (G, A). Apply to the four term
exact sequence
(1) 0 −→ A −→ A(α) −→ ZZ[G] −→ ZZ −→ 0
the exact functor Hom ( , Q/ZZ) and obtain the exact sequence
(2) 0 −→ Q/ZZ −→ ZZ[G]∗ −→ A(α)∗ −→ A∗ −→ 0.
(i) Show that the homomorphism
δ 2 : Ĥ n (G, A∗ ) −→ Ĥ n+2 (G, Q/ZZ),
arising from this sequence, coincides with the cup-product β 7→ −α ∪ β which is induced by
the pairing A × A∗ → Q/ZZ.
(ii) Show that δ 2 = −α∪ is an isomorphism for all n ∈ ZZ if A is a class module and α is a
fundamental class.
Hint: (i) Apply proposition (1.4.6) to the two pairs of exact sequences
0 −→ IG −→ ZZ[G] −→ ZZ −→ 0, 0 −→ A −→ A(α) −→ IG −→ 0,
0 −→ ZZ∗ −→ ZZ[G]∗ −→ IG
∗
−→ 0, ∗
0 −→ IG −→ A(α)∗ −→ A∗ −→ 0.

Let β ∈ Ĥ n (G, A∗ ) and consider the element 1 ∈ H 0 (G, ZZ). The homomorphism δ 2 :
H 0 (G, ZZ) → H 2 (G, A) arising from (1) maps 1 to α (see the proof of (3.1.5)). Therefore
δ 2 β = δ(δβ) ∪ 1 = (−1)n+2 (δβ ∪ δ1) = (−1)n+2+n+1 (β ∪ δ 2 1)
= −(β ∪ α) = −α ∪ β.
(ii) If A is a class module, then A(α) is cohomologically trivial by (3.1.4), hence also A(α)∗
is by (1.8.5).
§2. An Alternative Description of the

Reciprocity Homomorphism
Let G be a profinite group and let C be a G-module. If C is a formation

module for G, then we have the reciprocity homomorphism
recG : C G −→ Gab ,
which was obtained from the Nakayama-Tate duality theorem (3.1.5) by pass-
ing to the projective limit. If the pair (G, C) satisfies certain conditions de-
scribed below, we get a simple criterion for C being a formation module and
an alternative, non-cohomological, description of the reciprocity homomor-
phism as presented in [160], chap.IV. This method applies, for instance, to the
absolute Galois group of a local or global field and becomes important if one
wants to understand the reciprocity homomorphism more explicitly. However,
we will not use the results below in the following, so the reader might skip this
section on the first reading.
Assume we are given a G-module C and a pair of homomorphisms
d v
(G −→ ẐZ, C G −→ ẐZ) ,
where d is surjective and Z = v(C G ) has the properties
ZZ ⊆ Z and Z/nZ ∼
= ZZ/nZZ for all n ∈ IN ,
in particular, the cokernel ẐZ/Z of v is uniquely divisible. Furthermore, we
assume that, for every open subgroup U of G,
v(NG/U C U ) = fU Z
with fU = (ẐZ : d(U )). Then we have the surjective homomorphisms
dU = 1
fU
d : U −→ ẐZ , vU = 1
fU
v ◦ NG/U : C U −→ Z.

§2. An Alternative Description of the Reciprocity Homomorphism 165
For every pair V ⊆ U of open subgroups, we set fU/V = fV /fU , eU/V =

(U : V )/fU/V , and we get the commutative diagrams
V a`_^]b ab
dV
Chgfedc V
vV
ẐZ ẐZ
i Ver fU/V eU/V NU/V incl fU/V eU/V
d vU
U ab U
ẐZ, CU ẐZ.
The situation is most briefly, and best, formulated in the language of G-modu-
lations (see I §5). We have on the one hand the fundamental G-modulation
π ab : U 7−→ U ab ,
where, for two open subgroups V ⊆ U , the maps
res U
V
U jiab V ab
ind V
U
are the transfer Ver : U ab → V ab and the map induced by the inclusion V ,→ U .
On the other hand, we consider C as the G-modulation
C : U 7−→ C U ,
where res UV and ind VU are the inclusion and the norm. C is endowed with the
submodulation N C : U 7→ NU C. The submodulation N π ab : U 7→ NU U ab of
π ab is trivial, because we have a surjection
1 = lim
←−
V lim
←−
V [U, U ]/[U, U ] = NU U ab .
V ⊆U V ⊆U
Finally, we consider the G-modulation U 7→ ẐZ, where

res UV = eU/V and ind VU = fU/V .
We denote this G-modulation by ẐZ (observe that it depends on d : G → ẐZ
since the numbers fU/V and eU/V do). Then, in the above situation, we have
two morphisms
d : π ab −→ ẐZ, v : C −→ ẐZ
of G-modulations. The main result of abstract class field theory may now be
formulated as follows.
(3.2.1) Theorem. Assume that for every pair V ⊆ U of open subgroups of G,

with V normal in U and U/V cyclic, we have
(U : V ) for i = 0,
(
i V
#Ĥ (U/V, C ) =
1 for i = 1,
the class field axiom. Then there is a unique morphism
r : C −→ π ab
of G-modulations such that v = d ◦ r.

Proof: The theorem is just a reformulation of the results in [160], chap.IV.

For every pair V ⊆ U of open subgroups of G, V normal in U , we have by
[160], chap.IV, (6.3) a canonical isomorphism
r̂U/V : (U/V )ab −→
∼ C U /N V
U/V C .
We briefly recall the definition of r̂U/V (see [160], chap.IV, (5.6), where, in
contrast to our notation, this map was called the reciprocity homomorphism).
Let I = ker d and IU = I ∩ U . The semigroup
Frob(U/IV ) = {σ̃ ∈ U/IV | dU (σ̃) ∈ IN} ,
whose elements we call Frobenius lifts, maps surjectively onto U/V
Frob(U/IV ) → U/V, σ̃ 7→ σ = σ̃ mod V ,
cf. [160], chap.IV, (4.4). If σ ∈ U/V and σ̄ is its image in (U/V )ab , then
r̂U/V (σ̄) is defined by
r̂U/V (σ̄) = NU/S (πS ) mod NU/V C V ,
where S = hσ̃iIV ⊆ U for some Frobenius lift σ̃ of σ and πS ∈ C S is any
element with vS (πS ) = 1. If the class field axiom holds, it was shown in [160],
chap.IV that r̂U/V is a well-defined homomorphism which, in particular, does
not depend on the various choices and which moreover is an isomorphism.
Taking the projective limit over V of the surjections
(r̂U/V )−1
rU/V kl : C U C U /NU/V C V (U/V )ab
yields a family of homomorphisms
rU : C(U ) = C U C U /NU C ,→ lim C U /NU/V C V −→
∼ U ab = π ab (U )
←−
V
with dense image. By [160], chap.IV, (6.4) and (6.5), this family defines a
morphism r : C −→ π ab of G-modulations such that the diagram
Cnmo r
π ab
v d
ẐZ
is commutative.
In order to prove uniqueness, first observe that any morphism r : C → π ab
factors through C/N C, because π ab has trivial universal norms, and the same
is true for v. Let U be an open subgroup of G. For an open normal subgroup
V of U , we consider the group U/IV . Obviously, it is enough to show that in
the commutative diagram
rU/IV
Crpq U (U/IV )ab
vU
dU
ẐZ

the homomorphism rU/IV = lim rU/W (W open and normal in U ) is

←− IV ⊆W ⊆U
uniquely determined by dU and vU , since then we can pass to the projective
limit over V in order to obtain the desired result. Let S be the finite set of
splittings of the surjection dU : U/IV ẐZ. We denote the images by Ws /IV
for s ∈ S, where Ws is an open subgroup of U . Since
hWs /IV , s ∈ Si = U/IV ,
we see that hWs , s ∈ Si = U . By [160], chap.IV, (6.7), we have the equality
0 0
NU |W C W · NU |W 0 C W = NU |W W 0 C W W for two open subgroups W and W 0 of
U . Since S is finite, we obtain
hNU |Ws C Ws , s ∈ Si = C U .
Now the commutative diagram
rU/IV
Cstuvwx U (U/IV )ab
NU |Ws
rWs /IV
C Ws Ws /IV
vWs dWs
ẐZ
shows that rU/IV is uniquely determined by dU and vU since this is the case for
all maps rWs /IV . 2
The morphism r : C → π ab yields, in particular, a homomorphism

rG : C G −→ Gab ,
which we now want to compare with the homomorphism at the beginning of
this section.
U d
For every open subgroup U of G, we again set IU = ker(U −→ ẐZ) and ΓU =
U/IU . We have a commutative diagram
C Iy{|z U
ṽU
Z
v
CU U
Z
where ṽU = lim vU 0 . Recalling that Z/ZZ is uniquely divisible, we get
−→ IU ⊆U 0 ⊆U
the sequence of canonical homomorphisms
(ṽ )
H 2 (ΓU , C IU ) −→
U ∗
H 2 (ΓU , Z) = H 2 (ΓU , ZZ) −→
∼ H 1 (Γ , Q/Z
U
∼ Q/Z
Z) −→ Z,
and we denote its composite by
invU : H 2 (ΓU , C IU ) −→ Q/ZZ.

For a pair V ⊆ U of open subgroups, we have the compatible pair of homomor-

phisms ΓV → ΓU , C IU ,→ C IV , which yields a canonical homomorphism
res UV : H 2 (ΓU , C IU ) −→ H 2 (ΓV , C IV ),
and we have the commutative diagram
H 2 (ΓV}~ , C IV ) H 2 (ΓV , ZZ) δ −1
H 1 (ΓV , Q/ZZ) Q/ZZ
res eU/V ·res eU/V ·res (U :V )
δ −1
H 2 (ΓU , C IU ) H 2 (ΓU , ZZ) H 1 (ΓU , Q/ZZ) Q/ZZ,
noting that eU/V · fU/V = (U : V ) and that the canonical generator ϕV of ΓV is
mapped onto the fU/V -th power of the canonical generator ϕU of ΓU .
(3.2.2) Proposition. If C satisfies the class field axiom (3.2.1), then, for every
open subgroup U of G, the homomorphism
invU : H 2 (ΓU , C IU ) −→ Q/ZZ
is an isomorphism.
U ∗ (ṽ )
Proof: We have to show that H 2 (ΓU , C IU ) −→ H 2 (ΓU , Z) is an isomorphism.
−1 Un
Let Un := dU (nẐZ) and Dn := ker(vUn : C → Z). From [160], chap.IV,
(6.2) it follows that the class field axiom (3.2.1) implies Ĥ i (U/Un , Dn ) = 0 for
i = 0, −1, hence for all i. Since
D := ker(ṽU : C IU → Z) = Dn = lim
[
−→
Dn ,
n n
we get for i ≥ 1
H i (ΓU , D) = lim H i (U/Un , Dn ) = 0 .
−→
n
ṽ
Taking the cohomology of the exact sequence 0 −→ D −→ C IU −→
U
Z −→ 0
of ΓU -modules, we get the exact sequence
(ṽ ) δ
0 = H 2 (ΓU , D) −→ H 2 (ΓU , C IU ) −→
U ∗
H 2 (ΓU , Z) −→ H 3 (ΓU , D) = 0.
Therefore (ṽU )∗ is an isomorphism. 2
(3.2.3) Proposition. Under the above assumptions, the following assertions

are equivalent.
(i) C satisfies the class field axiom (3.2.1).
(ii) H 2 (ΓU , C IU ) = H 2 (U, C) for all open subgroups U of G and C is a
formation module with respect to the isomorphisms
invU : H 2 (U, C) −→
∼
Q/ZZ.

Proof: If C is a formation module, then for every open normal subgroup V

of U ,
ZZ/(U : V )ZZ if i = 2 ,
(
i V ∼
H (U/V, C ) =
0 if i = 1 ,
and H 2 (U/V, C V ) ∼
= Ĥ 0 (U/V, C V ) if U/V is cyclic, so that the class field
axiom holds.
Conversely, assume that C satisfies the class field axiom. We claim that

#H 2 (U/V, C V ) (U : V )

for every pair V / U . In fact, this is true if U/V is cyclic because H 2 ∼

= Ĥ 0 .
If U/V is a p-group, it follows inductively from the exact sequence
0 −→ H 2 (U/W, C W ) −→ H 2 (U/V, C V ) −→ H 2 (W/V, C V ),
where W/V is a normal subgroup of U/V of order p . In the general case, let
(U/V )p be a p-Sylow subgroup of U/V and Up the pre-image of (U/V )p in U .
Since the restriction map
res : H 2 (U/V, C V ) ,→ H 2 ((U/V )p , C Up )

M
is injective by (1.6.10), we obtain

Y Y
#H 2 (U/V, C V ) #H 2 ((U/V )p , C Up )
#(U/V )p = #(U/V ) .
p p
For every open normal subgroup V of U , we have the exact sequence

inf res
0 −→ H 2 (U/V, C V ) −→ H 2 (U, C) −→ H 2 (V, C)
and we identify H 2 (U/V, C V ) with its image in H 2 (U, C). Let n = (U : V )

and let Un = d−1
U (nẐ
Z). Then
H 2 (U/V, C V ) = H 2 (U/Un , C Un ).
In fact, because

#H 2 (U/V, C V ) (U : V ) = (U : Un ) = #H 2 (U/Un , C Un ) ,
it suffices to show the inclusion "⊇". But this follows from the exact commu-
tative diagram

0 H 2 (U/V, C V ) H 2 (U, C) res

H 2 (V, C)
res
H 2 (ΓU , C IU ) H 2 (ΓV , C IV )
invU invV
(U :V )
Q/ZZ Q/ZZ,
in which invU and invV are isomorphisms by (3.2.2). Since H 2 (U/Un , C Un ) ⊆
H 2 (ΓU , C IU ) has order n = (U : V ), it is mapped by the middle arrow res , and
thus by the upper arrow res , to zero, hence
H 2 (U/Un , C Un ) ⊆ H 2 (U/V, C V ).
We therefore obtain
H 2 (U/V, C V ) = H 2 (U/Un , C Un ) = H 2 (ΓU , C IU ).
[ [
H 2 (U, C) =
V n
For V open and normal in U , we have the commutative diagram
C)
H 2 (V,
invV
Q/ZZ
∼
res (U :V )
invU
H 2 (U, C) ∼ Q/ZZ,
and the induced isomorphisms
invU/V : H 2 (U/V, C V ) −→ (U1:V ) ZZ/ZZ
define on C the structure of a formation module in the sense of (3.1.8). 2
Let us assume that C satisfies the class field axiom and thus is, in particular,
a formation module with respect to the isomorphisms
invU : H 2 (U, C) −→
∼ Q/Z
Z.
Noting that H 2 (U, ZZ) ∼
= H 1 (U, Q/ZZ) = Hom(U ab , Q/ZZ), the cup-product
inv
lim
←−
Ĥ 0 (U/V, C V ) × H 2 (U, ZZ) −→ H 2 (U, C) −→
∼ Q/Z
Z
V
yields by (3.1.5) homomorphisms
recU : C U C U /NU C ,→ lim C U /NU/V C V −→
∼ U ab .
←−
V
These homomorphisms commute with conjugation, restriction and corestric-
tion by the rules (1.5.3) and (1.5.7), i.e. they form a morphism
rec : C −→ π ab
of G-modulations.

§3. Cohomological Dimension 171
(3.2.4) Theorem. The morphisms

r, rec : C −→ π ab
of G-modulations coincide.
Proof: Because of the uniqueness assertion of (3.2.1), it suffices to show that

the diagram
C rec
π ab
v d
ẐZ
commutes. Let U be an open subgroup of G. As before, we put IU =
dU
ker(U −→ ẐZ) and ΓU = U/IU . We have the commutative diagram
¡¢ C)
H 0 (U, × H 2 (U, ZZ) ∪
H 2 (U, C) inv
Q/ZZ
inf inf
∪
H 0 (ΓU , C IU ) × H 2 (ΓU , ZZ) H 2 (ΓU , C IU )
vU (dU )∗ (dU )∗ (ṽU )∗
∪ inv
H 0 (ẐZ, Z) × H 2 (ẐZ, ZZ) H 2 (ẐZ, Z) Q/ZZ ,
which induces the commutative diagram
C¤¥¦£ U
recU
U ab
vU dU
Z ẐZ .
This proves the theorem. 2
§3. Cohomological Dimension
Let G be a profinite group, Mod(G) the category of (discrete) G-modules, and

Modt (G), Modp (G), Modf (G) the category of G-modules which are torsion,
p-torsion, finite respectively as abelian groups. A fundamental numerical
invariant of G is the cohomological dimension.

(3.3.1) Definition. The cohomological dimension cd G (resp. strict coho-

mological dimension scd G) of G is the smallest integer n such that
H q (G, A) = 0 for all q>n
and all A ∈ Modt (G) (resp. A ∈ Mod(G)), and is ∞ if no such integer exists.
Let p be a prime number. The cohomological p-dimension cdp G (resp.
strict cohomological p-dimension scdp G) is the smallest integer n such that
the p-primary part
H q (G, A)(p) = 0 for all q > n ∗)
and all A ∈ Modt (G) (resp. A ∈ Mod(G)), and is ∞ if no such integer exists.
The profinite group G is of virtual cohomological dimension vcd G = n
resp. virtual strict cohomological dimension vscd G = n if there exists an
open subgroup U of G such that cd U = n resp. scd U = n. The p-versions
vcdp G and vscdp G are defined analogously.
M every abelian torsion group X decomposes into the direct sum

Since
X= X(p) of its p-primary parts X(p), we have
p
cd G = sup cdp G and scd G = sup scdp G.

p p
If G has an element of order p − in particular, if G is finite and p | #G −

then cdp G = ∞. In fact, if H is a subgroup of order p, then by (1.6.4) and
(1.7.1),
H 2n (G, IndH Z/pZZ)) ∼
G (Z = H 2n (H, ZZ/pZZ) ∼
= Ĥ 0 (H, ZZ/pZZ) = ZZ/pZZ =/ 0
for all n ≥ 0.
(3.3.2) Proposition. The following conditions are equivalent.

(i) cdp G ≤ n,
(ii) H q (G, A) = 0 for all q > n and all A ∈ Modp (G),
(iii) H n+1 (G, A) = 0 for all simple G-modules A with pA = 0.
For a pro-p-group G, we have, in particular,
cd G ≤ n ⇔ H n+1 (G, ZZ/pZZ) = 0.
∗) The p-primary part X(p) of an abelian torsion group X consists of all elements of X
which have p-power order.

M
Proof: (i) ⇔ (ii) follows from A = A(p) if A ∈ Modt (G), and H q (G, A)(p)
p
= H q (G, A(p)). Assume (iii). If A is finite of p-power order, then H n+1 (G, A) =
0 follows by induction on #A by the exact cohomology sequence associated
to the exact sequence 0 → B → A → A/B → 0, where B is a simple non-zero
submodule of A. It then follows for general A ∈ Modp (G) by taking direct
limits. From this, we get (ii) by induction on q, considering the exact se-
quence 0 → A → IndG (A) → A1 → 0, which by (1.3.8) yields the isomorphism
H q+1 (G, A) ∼
= H q (G, A1 ). 2
(3.3.3) Proposition. cdp G ≤ scdp G ≤ cdp G + 1.
Proof: cdp G ≤ scdp G is trivial. Let A ∈ Mod(G) and consider the exact
sequences
p
0 −→ p A −→ A −→ pA −→ 0, 0 −→ pA −→ A −→ A/pA −→ 0.
Let q > cdp G + 1. Then H (G, p A) = H q−1 (G, A/pA) = 0 since p A and
q
A/pA ∈ Modp (G). Therefore

H q (G, A) −→ H q (G, pA) and H q (G, pA) −→ H q (G, A)
are injective. The composite is multiplication by p since the composite of
p
A → pA → A is multiplication by p. It follows that H q (G, A)(p) = 0, showing
scdp G ≤ cdp G + 1. 2
(3.3.4) Corollary. Assume that cdp G = n is finite. Then scdp G = n if and

only if H n+1 (U, ZZ)(p) = 0 for all open subgroups U of G.
Proof: Assume the latter. If the G-module A is finitely generated as a ZZ-

module, then there is an open subgroup U of G acting trivially on A, and A
is a quotient B/C of B = IndUG (ZZm ) for some m. Since scdp G ≤ n + 1, we
have H n+1 (G, A)(p) = 0, and this result extends to an arbitrary G-module A
by passing to the direct limit. 2
Example: Let G = ẐZ = lim ZZ/nZZ. Then

←− n
cdp G = 1 and scdp G = 2.
In fact, if A is a finite G-module of p-power order, then every extension
0 −→ A −→ Ĝ −→ G −→ 1
splits (the closed subgroup (σ) of Ĝ, topologically generated by a pre-image
σ ∈ Ĝ of 1 ∈ G, is mapped isomorphically onto G). By (1.2.4), H 2 (G, A) ∼=

EXT(G, A), hence H 2 (G, A) = 0 for finite A ∈ Modp (G), and by taking
inductive limits, for all A ∈ Modp (G). Noting that H 1 (G, ZZ/pZZ) = ZZ/pZZ =/ 0,
this shows cdp G = 1. On the other hand, H 2 (G, ZZ) ∼ = H 1 (G, Q/ZZ) ∼= Q/ZZ,
so that scdp G = 2.
(3.3.5) Proposition. If H is a closed subgroup of G, then

cdp H ≤ cdp G and scdp H ≤ scdp G.
We have equality in each of the following cases:
(i) (G : H) is prime to p,
(ii) H is open and cdp G < ∞.
Proof: The inequalities follow from Shapiro’s lemma (1.6.4)

H q (G, IndH ∼ q
G (A)) = H (H, A),
noting that if A is torsion, then so is IndH
G (A). On the other hand, in case (i),
res : H (G, A)(p) −→ H q (H, A)(p)
q
is injective (see the proof of (1.6.10)). In case (ii), consider for A ∈ Mod(G)
the exact sequence ν
0 −→ B −→ IndH G (A) −→ A −→ 0
of G-modules, where ν is given by νx = σ∈G/H σx(σ −1 ). We obtain a

P
homomorphism
H n (H, A)(p) = H n (G, IndH n
G (A))(p) −→ H (G, A)(p),
which is surjective if H n+1 (G, B)(p) = 0. This is the case if either n = scdp G
or n = cdp G and A ∈ Modt (G). Thus in either case, (i) or (ii), we have the
implication
H n (H, A)(p) = 0 ⇒ H n (G, A)(p) = 0
and this means scdp H ≥ scdp G or cdp H ≥ cdp G respectively. 2
SERRE has shown that a much weaker condition than cdp G < ∞ guarantees
the equality cdp G = cdp H for an open subgroup H. One requires only that G
contains no element of order p (see [211] and [75]).
(3.3.6) Corollary. If Gp is a p-Sylow subgroup of G, then

cdp G = cdp Gp = cd Gp and scdp G = scdp Gp = scd Gp .
(3.3.7) Corollary. Let G be a profinite group. Then

cdp G = 0 if and only if #G is prime to p.

Proof: By (3.3.6), we may assume that G is a pro-p-group, and then the

assertion follows from (1.6.14)(ii). 2
(3.3.8) Proposition. If H is a closed normal subgroup of G, then

cdp G ≤ cdp G/H + cdp H.
If cdp G/H < ∞ and cdp H < ∞, and if H n (U, ZZ/pZZ) is finite for n = cdp H
and all open subgroups U of H ∗) , then equality holds.
Proof: We may assume that m = cdp G/H and n = cdp H are finite. Consider
for A ∈ Modp (G) the Hochschild-Serre spectral sequence
E2ij = H i (G/H, H j (H, A)) ⇒ H i+j (G, A) = E i+j .
Let q > m + n. If i + j = q, then either i > m or j > n, hence E2ij = 0. As
H q (G, A) has a filtration, whose quotients are subquotients E∞ij
of E2ij , we get
H q (G, A) = 0, so that cdp G ≤ m + n.
Now assume that H n (U, ZZ/pZZ) is finite for all open subgroups U of H. Since
cdp H = n, there exists a finite H-module A with pA = 0 and H n (H, A) =/ 0.
Let G0 be an open subgroup of G such that H 0 = G0 ∩ H acts trivially on A.
The canonical surjection
0 ν
IndH σx(σ −1 ),
X
H (A) → A, x 7→
σ ∈H/H 0
induces a surjection
0 ν
H n (H 0 , A) = H n (H, IndH ∗ n
H (A)) −→ H (H, A),
again because n = cdp H. Therefore H n (H 0 , A) =/ 0, and hence the group

H n (H 0 , ZZ/pZZ) is finite and non-zero.
Now let G00 /H 0 be a p-Sylow subgroup of G0 /H 0 . Then cdp G00 /H 0 =
cdp G0 /H 0 = cdp G/H = m and cdp H 0 = cdp H = n by (3.3.5). Therefore, by
(2.1.4),
H m+n (G00 , ZZ/pZZ) = H m (G00 /H 0 , H n (H 0 , ZZ/pZZ)).
This group is non-zero. Namely, since G00 /H 0 is a pro-p-group, there is
a surjective homomorphism H n (H 0 , ZZ/pZZ) → ZZ/pZZ of G00 /H 0 -modules,
which induces a surjective homomorphism
H m (G00 /H 0 , H n (H 0 , ZZ/pZZ)) −→ H m (G00 /H 0 , ZZ/pZZ) =/ 0.
We therefore obtain cdp G ≥ cdp G00 ≥ m + n. 2
We take the following theorem from [247], see also [45].

∗) If H is a pro-p-group, this is already true if H n (H, ZZ/pZZ) is finite.

(3.3.9) Theorem. Let N be a closed normal subgroup of a profinite group G

and let cdp G = n and cdp N = k. Assume that the group H k (N, ZZ/pZZ) is
finite and nonzero. Then
vcdp G/N = n − k .
Remark: The condition H k (N, ZZ/pZZ) =/ 0 is necessary. If N is a pro-p-

group, then cdp N = k implies H k (N, ZZ/pZZ) =/ 0, see (3.3.2). This is not true
for a general profinite group N .
Proof: In the following we write H ∗ (−) for H ∗ (−, ZZ/pZZ). Replacing

G by the pre-image of a p-Sylow subgroup of G/N under the canonical
projection and using (3.3.6), we may assume that G/N is a pro-p-group. Since
H k (N ) = lim 0 H k (N/N 0 ) is finite, there exists an open subgroup M of N ,
−→ N
which is normal in G, such that the inflation map
inf : H k (N/M ) −→ H k (N )
is surjective.
Now there exists an open subgroup U0 of G containing M such that
(∗) U0 ∩ N = M and U0 N/M ∼ = N/M × U0 /M .
This is a consequence of the following group theoretical
Claim: Let N be a finite normal subgroup of the profinite group G. Then there
exists an open subgroup U of G, such that U is contained in the centralizer of
N in G and U ∩ N = 1. In particular, U N is open in G and U N ∼ = U × N.
Proof of the claim: Let K be the kernel of the canonical homomorphism
G → Aut(N ) which is induced by the conjugation. Then K is open and normal
in G, and K ∩ N is the center Z(N ) of N . Choosing for every element
z ∈ Z(N )\{1} an open normal subgroup Uz of K such that z ∈/ Uz , then the
T
group U = z∈Z(N )\{1} Uz has the desired properties.
Using the claim for the finite subgroup N/M of G/M , we obtain U0 as in
(∗). Since H k (N/M ) is finite, replacing U0 by a suitable open subgroup, we
may further assume that U0 acts trivially on H k (N/M ). Then H k (N ) is a finite
trivial U -module where U = U0 N .
Now we consider the Hochschild-Serre spectral sequences
E2p,q = H p (U/N, H q (N )) ⇒ H p+q (U )
and
E p,q p q
2 = H (U/N, H (N/M )) ⇒ H
p+q
(U/M ) ,
which are associated to the group extensions
1 → N → U → U/N → 1 and 1 → N/M → U/M → U/N → 1 .

By (2.4.6), the second spectral sequence degenerates at E 2 .

Let X • (U, ZZ/pZZ), X • (U/M, ZZ/pZZ) and X • (U/N, ZZ/pZZ) be the standard
resolutions of ZZ/pZZ as a trivial U -module, U/M -module and U/N -module,
respectively. Then the canonical map X • (U/M, ZZ/pZZ) → X • (U, ZZ/pZZ) in-
duces a homomorphism of double complexes
X • (U/N, X • (U/M, ZZ/pZZ)N/M )U/N → X • (U/N, X • (U, ZZ/pZZ)N )U/N ,
and so we obtain a morphism of the associated E2 -spectral sequences

(E pq ¯ pq
r , dr ) −→(Er , dr ) .
Since cdp N = k, and as the second spectral sequence degenerates, we have a

commutative diagram
§¨©ª n+1−k,k
E∞ E2n+1−k,k
E n+1−k,k
∞ E n+1−k,k
2 .
Next we prove that E2n+1−k,k vanishes. By (3.3.5)(ii), we have cdp U = n,

n+1−k,k
which implies E∞ = 0. Therefore it suffices to show that the right-hand
vertical arrow in the commutative diagram above is surjective. This can be
seen as follows: By construction of M , the map
inf : H k (N/M ) −→ H k (N )
is an epimorphism of trivial U/N -modules, which splits, as both groups are
IFp -vector spaces. Hence the map
E 2n+1−k,k = H n+1−k (U/N, H k (N/M )) −→ E2n+1−k,k = H n+1−k (U/N, H k (N ))
is surjective.
We obtain E2n+1−k,k = H n+1−k (U/N, H k (N )) = 0. By assumption, H k (N ) ∼
=
(ZZ/pZZ)m for some m ≥ 1, and by construction of U , H k (N ) is a trivial U/N -
module. Hence
0 = H n+1−k (U/N, H k (N )) ∼
= H n+1−k (U/N )m .
This implies H n+1−k (U/N ) = 0, and so cdp (U/N ) ≤ n − k, as U/N is a pro-
p-group, see (3.3.2). On the other hand, (3.3.8) implies cdp (U/N ) ≥ n − k,
which finishes the proof. 2
(3.3.10) Corollary. Let G be a pro-p-group, cdp G = n < ∞, and let N be a

closed normal subgroup of G such that cdp N = cdp G and H n (N, ZZ/pZZ) is
finite. Then N is open in G.

Proof: By (3.3.9) and the subsequent remark, we obtain vcdp G/N = 0.

Hence G/N is finite by (3.3.7). 2
The cohomological dimension has the following effect on the corestriction.
(3.3.11) Proposition. Let U be an open subgroup of G. If scd G = n (resp.

cdp G = n), then for every G-module A (resp. for every p-torsion G-module),
the corestriction
cor : H n (U, A) −→ H n (G, A)
is surjective. If U is normal in G, then
cor : H n (U, A)G/U −→ H n (G, A)
is an isomorphism.
Proof: The first assertion follows from the second. Indeed, if U0 is an open
normal subgroup of G contained in U , then the composition
U
cor U0 cor U
cor UG0 : H¬« n (U0 , A) H n (U, A) G
H n (G, A)
shows that the surjectivity of cor UG0 implies the surjectivity of cor UG . Therefore
we may assume U to be normal in G.
.
Let A → X be the standard resolution of the G-module A. Then the map
cor : H i (U, A) → H i (G, A) is given by taking cohomology of the map of
. .
complexes N = NG/U : X U → X G . The G-module Y n = ker(X n → X n+1 )
is cohomologically trivial by our assumption on the cohomological dimension.
Therefore
0 −→ A −→ X 0 −→ · · · −→ X n−1 −→ Y n −→ 0
is a finite resolution of A by cohomologically trivial G-modules. By (1.3.9),
we obtain the exact commutative diagram
³´µ²±®¯° )U
(X n−1 (Y n )U H n (U, A) 0
N N cor
(X n−1 )G (Y n )G H n (G, A) 0.
Taking G/U -coinvariants of the upper row, we obtain the exact commutative
diagram
((X n−1¼½¾»º¶·¸¹ )U )G/U ((Y n )U )G/U H n (U, A)G/U 0
N N cor
((X n−1 )U )G/U ((Y n )U )G/U H n (G, A) 0.

By (1.8.2), the G/U -modules (X n−1 )U and Y U are cohomologically trivial.

Therefore, by (1.2.6), the maps N are bijective, and hence also cor , by the five
lemma. 2
We will now consider the Euler-Poincaré characteristic of a pro-p-group.
(3.3.12) Definition. Let G be a pro-p-group of finite cohomological dimension.

Assume that the groups H i (G, IFp ) are finite for all i ≥ 0. Then the Euler-
Poincaré characteristic of G is the alternating sum
(−1)i dimIFp H i (G, IFp ) .

X
χ(G) =
i
If A is a finite p-primary G-module with pA = 0, then we put
(−1)i dimIFp H i (G, A) .

X
χ(G, A) =
i
Let G be as above and let 0 −→ A1 −→ A2 −→ A3 −→ 0 be an exact sequence

of finite IFp [G]-modules. Then obviously
χ(G, A2 ) = χ(G, A1 ) + χ(G, A3 ) ;
in particular, from (1.6.13), it follows for an IFp [G]-module A of order pr that

χ(G, A) = r · χ(G) .
(3.3.13) Proposition. Let G be a pro-p-group of finite cohomological dimen-

sion and assume that the groups H i (G, IFp ) are finite for all i ≥ 0. If U is an
open subgroup of G, then
χ(U ) = (G : U )χ(G) .
Proof: Using (1.6.4), we obtain χ(U ) = χ(G, IndUG IFp ) = (G : U )χ(G).

2
If we drop the assumption that the pro-p-group G is of finite cohomological

dimension (but keep the assumption on the finiteness of H i (G, IFp ) for i ≤ n
for some n), then we define the partial Euler-Poincaré characteristic of G.

(3.3.14) Definition. Let n ≥ 0 and let G be a pro-p-group such that the

groups H i (G, IFp ) are finite for all i with 0 ≤ i ≤ n. Then the n-th partial
Euler-Poincaré characteristic of G is the alternating sum
n
(−1)i dimIFp H i (G, IFp ) .
X
χn (G) =
i=0
For a finite IFp [G]-module A, we define

n
(−1)i dimIFp H i (G, A) .
X
χn (G, A) =
i=0
Induction on the length of a composition series of the IFp [G]-module A

yields the formula:
(3.3.15) Lemma. If G and A are as above, then
(−1)n χn (G, A) ≤ (−1)n dimIFp A · χn (G) .
The following theorem, due to H. KOCH [110], can be considered as a

converse of (3.3.13).
(3.3.16) Theorem. Let G be a pro-p-group such that the groups H i (G, IFp ) are
finite for 0 ≤ i ≤ n. Let U be a cofinal set of open subgroups of G. Then the
following assertions are equivalent:
(i) χn (U ) = (G : U )χn (G) for all U ∈ U.
(ii) cdp G ≤ n.
Proof: The implication (ii) ⇒ (i) is just (3.3.13), so let us assume that (i) holds.
By (3.3.2)(iii), it suffices to show that H n+1 (G, IFp ) = 0. Let ā ∈ H n+1 (G, IFp )
with a ∈ C n+1 (G, IFp ). Then there exists a U ∈ U such that a depends only on
the cosets of G/U . Define the finite IFp [G]-module A by the exact sequence
ϕ
0 −→ IFp −→ IndUG IFp −→ A −→ 0 .
We obtain a commutative diagram
ÁÀ¿ IFp )
H n+1 (G,
ϕ∗
H n+1 (G, IndUG IFp )
res
sh
n+1
H (U, IFp )

§4. Dualizing Modules 181
which, by the choice of U , shows that res ā = 0, and so ϕ∗ ā = 0. Furthermore,

we get from the short exact sequence above the exact cohomology sequence
0 −→ H 0 (G, IFp ) −→ · · · −→ H n (G, A) −→ ker(ϕ∗ ) −→ 0 ,
which implies, using (3.3.15), (1.6.4) and the assumption (i),
dimIFp ker(ϕ∗ ) = (−1)n (χn (G) + χn (G, A) − χn (G, IndUG IFp ))
≤ (−1)n (χn (G) + dimIF A · χn (G) − χn (U ))
p
= (−1)n ((G : U )χn (G) − χn (U )) = 0 .

Thus ker(ϕ∗ ) = 0 and therefore ā = 0, which proves the theorem. 2
Remark: One can prove the following generalization of (3.3.16), cf. [194]: If
there exists a number N such that (−1)n χn (U ) + N ≥ (−1)n (G : U )χn (G) for
all U ∈ U, then G is finite or cd G ≤ n.
Exercise 1. If cdp G = 1, then scdp G = 2.

Exercise 2. Let p =/ 2 and let G be the group of affine transformations x 7→ ax + b with b ∈ ZZp
and a ∈ ZZ×
p . Then cd p G = scd p G = 2.
Exercise 3. If cdp G =/ 0, ∞, then the exponent of p in the order of G is infinite.

Exercise 4. cdp G = cdp G/H +cdp H if H is contained in the center of G and cdp G/H < ∞.
Exercise 5. Let H be a closed normal subgroup of G. If cdp H =/ 0, then scdp G ≤
cdp G/H + scdp H.
Exercise 6. If H n+1 (U, ZZ) = H n+2 (U, ZZ) = 0 for all open subgroups U of G, then scdp G ≤ n.
p
Hint: Using the exact sequence 0 → ZZ → ZZ → ZZ/pZZ → 0, one sees that H n+1 (Gp , ZZ/pZZ)
= 0 for a p-Sylow subgroup Gp , hence cdp G = cd Gp ≤ n, and scdp G ≤ n by (3.3.4).
§4. Dualizing Modules
Let G be a profinite group and let Mod(G) be the category of discrete

G-modules A. In II §5, we introduced the G-modules
Di (A) = lim H i (U, A)∗ ,
−→
U
where U runs through the open normal subgroups of G and the limit is taken
over the maps cor ∗ dual to the corestriction. It comes equipped with the
H i (G, A)∗ −→ Di (A)

and the pairing of G-modules

Di (A) × A −→ Di (ZZ)
(cf. p.125). We obtain a canonical homomorphism
ϕA : H i (G, A)∗ −→ HomG (A, Di (ZZ)).
(3.4.1) Theorem. If n = scd G < ∞, then the map

ϕA : H n (G, A)∗ −→ HomG (A, Dn (ZZ))
is an isomorphism for all A ∈ Mod(G). We call the G-module D = Dn (ZZ) the
dualizing module of G.
Remark: By definition, the dualizing module of an open subgroup U of G is

the dualizing module of G regarded as a U -module.
Proof: By a straightforward limit argument using (1.5.1), we are reduced to

the case where A is a finitely generated ZZ-module. Let U be an open normal
subgroup of G acting trivially on A, i.e. A is a ZZ[G/U ]-module.
Suppose the theorem is proven for free ZZ[G/U ]-modules. Then we may
choose an exact sequence
0 ← A ← F0 ← F1
with free ZZ[G/U ]-modules F0 , F1 and obtain the bijectivity of ϕA from the
exact commutative diagram
0ÈÉÊÇÆÂÃÄÅ H n (G, A)∗ H n (G, F0 )∗ H n (G, F1 )∗
ϕA ϕF0 ϕF1
0 HomG (A, D) HomG (F0 , D) HomG (F1 , D).

So let A = ZZ[G/U ] for an open normal subgroup U of G. Using Shapiro’s
lemma (1.6.4), we obtain the commutative diagram
ϕG, ZZ [G/U ]
H n (G, ZZÍÎÌË [G/U ])∗ HomG (ZZ[G/U ], D)
sh∗
ϕU, ZZ
H n (U, ZZ)∗ HomU (ZZ, D).
Finally, the map ϕU, ZZ is the direct limit over V of the maps
cor ∗ : H n (U, ZZ)∗ −→ HomU (ZZ, H n (V, ZZ)∗ ) = (H n (V, ZZ)U/V )∗ ,
which are isomorphisms by (3.3.11). 2

If U is an open subgroup of G, then, by construction, the diagram
H i (U,ÒÓÔÕÑÐÏ A)∗
ϕA
HomU (A, Di (ZZ)) Hom(A, Di (ZZ))
cor ∗
ϕA
H i (G, A)∗ HomG (A, Di (ZZ)) Hom(A, Di (ZZ))
is commutative. Passing to the direct limit, we obtain the
(3.4.2) Corollary. For A ∈ Mod(G), we have a functorial homomorphism

Di (A) −→ Hom(A, Di (ZZ)),
which is an isomorphism for i = scd G if A is a finitely generated ZZ-module.
We defined the trace map

tr : H n (G, D) −→ Q/ZZ
as the first edge morphism of the Tate spectral sequence H i (G, Dn−j (ZZ))
⇒ H n−(i+j) (G, ZZ)∗ (see also ex.2). The pairing
Hom(A, D) × A −→ D
gives us a cup-product
∪
H i (G, Hom(A, D)) × H n−i (G, A) −→ H n (G, D) ,
which, together with the map tr, yields a homomorphism
H i (G, Hom(A, D)) −→ H n−i (G, A)∗ .
We can now prove the central result of this section.
(3.4.3) Duality Theorem. Let G be a profinite group of strict cohomological

dimension scd G = n < ∞ such that Dk (ZZ) = 0 for k < n. Let A ∈ Mod(G) be
finitely generated over ZZ and assume that the `-primary part of A is nontrivial
only for prime numbers ` such that D = Dn (ZZ) is `-divisible.
Then for all i ∈ ZZ ∗) the cup-product and the trace map
∪ tr
H i (G, Hom(A, D)) × H n−i (G, A) −→ H n (G, D) −→ Q/ZZ
yield an isomorphism
H i (G, Hom(A, D)) ∼
= H n−i (G, A)∗ .
∗) H i = 0 for i < 0 by definition.

Proof: Because of (3.4.2), we may replace Hom(A, D) by Dn (A) and con-

sider the map
(∗) H i (G, Dn (A)) −→ H n−i (G, A)∗
induced by the cup-product with respect to the canonical pairing Dn (A) ×
A → Dn (ZZ). But by (2.5.5) this map is the edge morphism E2i,0 → E i of the
Tate spectral sequence
E2ij = H i (G, Dn−j (A)) ⇒ H n−(i+j) (G, A)∗ = E i+j .
From the assumption in the theorem, it follows that Dk (A) = 0 for k < n by
lemma (2.5.6), hence the spectral sequence degenerates, showing that (∗) is an
isomorphism. 2
Remarks: 1. Setting i = n and A = ZZ, we see that the trace map tr : H n (G, D)
→ Q/ZZ is an isomorphism. It is clear that the duality theorem remains valid if
we replace it by any other isomorphism H n (G, D) ∼ = Q/ZZ. This can be useful
in the applications, where we may have a canonical such isomorphism without
it being clear that it is the edge morphism.
2. The condition D0 = lim −→ U
Q/ZZ = 0 is equivalent to the assertion that every
prime number p divides the order of G infinitely often, in the sense that all
Sylow subgroups Gp are infinite.
We obtain an important variant of the dualizing module and the duality

theorem as follows. Let P be a nonempty set of prime numbers. Let IN(P )
denote the set of all natural numbers which are divisible only by prime numbers
in P . Let ModP (G) denote the category of all P -torsion G-modules, i.e.
modules A consisting of elements a such that na = 0 for some n ∈ IN(P ).
We define the cohomological P -dimension cdP G as the smallest number
n ≥ 0 such that H i (G, A) = 0 for all i > n and all A ∈ ModP (G). If no such
number exists, we set cdP G = ∞. In other words, cdP G = sup{cdp G}.
p∈P
We now set
Di (ZZP ) = lim Di (ZZ/mZZ) = lim lim H i (U, ZZ/mZZ)∗ ,
−→ −→ −→
m∈IN(P ) m∈IN(P ) U
and, in particular,
Di (ẐZ) = lim Di (ZZ/mZZ).
−→
m∈IN
Again we have for A ∈ ModP (G) a functorial homomorphism
ϕA : H i (G, A)∗ −→ HomG (A, Di (ZZP ))
which is obtained as above from the canonical pairing
H i (G, A)∗ × m AU −→ H i (V, ZZ/mZZ)∗ , (χ, a) 7→ χa (x) = χ(cor (ax)),
and by taking direct limits over V and m ∈ IN(P ), and then over U .

(3.4.4) Theorem. If cdP G = n < ∞, then for all A ∈ ModP (G), the map
ϕA : H n (G, A)∗ −→ HomG (A, Dn (ZZP ))
is an isomorphism, and, if A is finite, we obtain an isomorphism of G-modules
Dn (A) ∼
= Hom(A, Dn (ZZP )).
The proof is the same as that of (3.4.1) (resp. (3.4.2)); we have just to replace
ZZ by ZZ/mZZ (m ∈ IN(P )) and lim by lim lim .
−→ −→ −→
U m∈IN(P ) U
The G-module Dn (ZZP ) represents the functor
T : ModP (G) −→ Ab, A 7−→ T (A) = H n (G, A)∗ ,
and is called the dualizing module of G at P (or of ModP (G)). We denote it

briefly by DP . If P consists of a single prime number p, then we write D(p) .
(3.4.5) Corollary (Serre Criterion). Let p be a prime number and let 1 ≤

n = cdp G < ∞. We have scdp G = n + 1 if and only if there exists an open
U
subgroup U of G such that D(p) contains a subgroup isomorphic to Qp /ZZp .
U
Proof: The existence of a subgroup of D(p) isomorphic to Qp /ZZp is equiv-
alent to HomU (Qp /ZZp , D(p) ) =/ 0. Since D(p) is also the dualizing module
of U at p, this means by (3.4.4) that H n (U, Qp /ZZp ) =/ 0. But this last
group is the p-primary component of H n (U, Q/ZZ). The exact sequence
0 → ZZ → Q → Q/ZZ → 0 yields H n (U, Q/ZZ)(p) ∼ = H n+1 (U, ZZ)(p), and the
corollary follows now from (3.3.4) and (3.3.5) (ii). 2
The Tate spectral sequence

H i (G, Dn−j (ZZ/mZZ)) ⇒ H n−(i+j) (G, ZZ/mZZ)∗ ,
m ∈ IN(P ), gives an edge morphism
H n (G, D(ZZ/mZZ)) −→ H 0 (G, ZZ/mZZ)∗ = 1
m
ZZ/ZZ.
Taking limits over m ∈ IN(P ), we obtain a homomorphism
tr : H n (G, DP ) −→ QP /ZZP :=
M
Qp /ZZp ,
p∈P
again called the trace map. We now obtain the following variant of the duality
theorem (3.4.3).

(3.4.6) Duality Theorem. For a profinite group with cdP G = n < ∞, the
following assertions are equivalent.
(i) Di (ZZ/pZZ) = 0 for all p ∈ P and all i < n.
(ii) For all i ∈ ZZ and all finite G-modules A ∈ ModP (G), the cup-product
and the trace map
∪ tr
H i (G, Hom(A, DP )) × H n−i (G, A) −→ H n (G, DP ) −→ QP /ZZP
yield an isomorphism
H i (G, Hom(A, DP )) ∼
= H n−i (G, A)∗ .
In this case G, is called a duality group at P of dimension n.
Proof: The implication (i) ⇒ (ii) follows by the same argument as in the
proof of (3.4.3). So we have only to show the implication (ii) ⇒ (i). For every
pair V ⊆ U of open normal subgroups of G, we have a commutative diagram
of G-modules V
νU
IndVG×ÖØ (A) IndUG (A)
V
νG U
νG
A
X
where νUV (x)(σ) = τ x(τ −1 σ). By the lemma of Shapiro (1.6.4) and by
τ ∈U/V
(1.6.5) and the subsequent remark, we obtain from this a commutative diagram
of G-modules
ÞÝÜàßÛÚÙ V (A))
H i (G, Ind sh
H i (V, A)
G
ν∗ cor
ν∗ H i (G, A) cor
ν∗ cor
sh
H i (G, IndUG (A)) H i (U, A).
In this diagram the maps sh are isomorphisms, and hence define an isomor-
phism of projective systems (H n (G, IndUG (A))) ∼
= (H n (U, A)) of G-modules.
We therefore have a canonical isomorphism
Di (A) ∼
= lim
−→
H i (G, IndUG (A))∗ .
U
Applying (ii) and (1.5.1), we obtain for all i < n and p ∈ P ,

Di (ZZ/pZZ) ∼
= lim H i (G, IndUG (ZZ/pZZ))∗
−→
U
∼
= lim H n−i (G, Hom(IndUG (ZZ/pZZ), DP ))
−→
U
∼
= lim H n−i (G, IndUG (Hom(ZZ/pZZ, DP ))
−→
U
∼
= lim H n−i (U, p DP ) = 0.
−→
U,res
2
(3.4.7) Corollary. If G is a duality group at P of dimension n, then for every

p ∈ P there exists an exact sequence
p
0 −→ Dn (ZZ/pZZ) −→ DP −→ DP −→ 0.
In particular, DP is p-divisible for all p ∈ P .
Proof: Let p ∈ P , m > 1, and consider the exact diagram

0áâãäåæçèéêë ZZ/pm ZZ ZZ/pm+1 ZZ ZZ/pZZ 0
0 ZZ/pm−1 ZZ ZZ/pm ZZ ZZ/pZZ 0.

n ∗
Applying the functor lim
−→ U
H (U, −) and passing to the limit over m, we
obtain the exact sequence
p
0 −→ Dn (ZZ/pZZ) −→ D(p) −→ D(p) −→ 0.
This proves the corollary since D(p) is the p-torsion subgroup of the torsion
group DP . 2
(3.4.8) Corollary. Assume that scd G = n and Dk (ZZ) = 0 for k = 0, . . . , n−1.

Let p be a prime number. Then Dn (ZZ) is p-divisible if and only if G is a duality
group at p of dimension n. In this case
D(p) = Dn (ZZ)(p) .
Proof: Consider the exact sequence

p
0 −→ ZZ −→ ZZ −→ ZZ/pZZ −→ 0
and apply the functor lim
−→
H k (U, −)∗ . We obtain Dk (ZZ/pZZ) = 0 for k ≤ n−2
and an exact sequence
p
0 → Dn (ZZ/pZZ) → Dn (ZZ) → Dn (ZZ) → Dn−1 (ZZ/pZZ) → 0 .

Hence Dn−1 (ZZ/pZZ) = 0, i.e. G is a duality group at p of dimension n by

(3.4.6) if and only if Dn (ZZ) is p-divisible. Further, in this case,
D(p) = lim
−→
H n (U, ZZ/pm ZZ)∗
U,m
= lim H 0 (U, Hom(ZZ/pm ZZ, Dn (ZZ)))
−→
U,m
= Dn (ZZ)(p) .
2
Remark: The duality theorems (3.4.3) and (3.4.6) are due to J. TATE (see
[230]). Tate gave the duality isomorphism as the edge morphism of the Tate
spectral sequence, which by (2.5.5) is induced by the cup-product. Another
slightly different method, due to J. L. VERDIER, is to prove (3.4.3) and (3.4.6)
by analyzing the group of “local homomorphisms” (see [240]). The concept
of duality groups was also inspired by the work of R. BIERI and B. ECKMANN.
in the case of discrete groups (see [11]).
Exercise 1. Let n = scd G. If the homomorphism

H p (G, Dn (A)) −→ H n−p (G, A)∗
is bijective for all p ∈ ZZ and all A ∈ Mod(G) finitely generated over ZZ, then Di = 0 for i < n
and Dn is divisible.
Exercise 2. Applying (3.4.1) to A = D, we obtain a canonical isomorphism

H n (G, D)∗ ∼
= HomG (D, D).
The trace map tr : H n (G, D) −→ Q/ZZ, defined as the first edge morphism of the Tate spectral
sequence, is an element of the left-hand group, which corresponds to id in the right-hand
group, and HomG (D, D) ∼ = ZZ.
Exercise 3. Let G = ẐZ. Show that scd G = 2, cd G = 1. Calculate D and DP for the set P of
all prime numbers. Show that there are isomorphisms
H i (G, Hom(A, Q/ZZ)) ∼
= H 1−i (G, A)∗
for all i and all finite G-modules A.
Exercise 4. Let Fn = hx1 , . . . , xn i be a free pro-p-group on n generators. Show that the

dualizing module D(p) of Fn is given by the exact sequence
0îïìí
diag Ln hxi i
Qp /ZZp i=0 IndFn Qp /ZZp D(p) 0,
−1 −1
where x0 := x−1 n xn−1 · · · x1 and hxi i is the closed procyclic group generated by xi in Fn
(i = 0, . . . , n).
(The characterization of D(p) given in ex.4 on p.V-24 (Cinquième éd. p.76) in [240] is incor-
rect!)

§5. Projective pro-c-groups 189
§5. Projective pro-c-groups
In this section we introduce the concept of embedding problems and con-

sider profinite groups G of cohomological dimension cd G ≤ 1. We start by
considering the abstract embedding problem for a profinite group G.
(3.5.1) Definition. (i) An embedding problem E (G) = E (G, ϕ, α) for a

profinite group G is a diagram
ôñòóð G
ϕ
1 N E α G 1
with an exact sequence of profinite groups and a surjective homomorphism ϕ.
(ii) A solution of the embedding problem E (G) is a homomorphism ψ: G → E
such that α ◦ ψ = ϕ. A solution is called proper if ψ is surjective.
(iii) Two solutions ψ and ψ 0 are called equivalent if
ψ 0 (σ) = a−1 ψ(σ)a
for all σ ∈ G with a fixed element a ∈ N . The set of all solutions of E (G) mod-
ulo equivalence is denoted by SE (G) and is considered as a discrete topological
space.
An embedding problem for a Galois group G corresponds to a field theoret-

ical problem. Let G = G(k̃|k) be the Galois group of a Galois extension k̃ of a
field k and let K|k be a Galois subextension. If Eõ G(K|k) is a group exten-
ö G can G(K|k), α)
sion, then a proper solution of the embedding problem E (G,
defines a Galois extension L ⊇ K ⊇ k with Galois group G(L|k) isomorphic
to E such that α is the canonical projection G(L|k) G(K|k). If the solution
is not proper, then one obtains only a Galois algebra with group E instead of
a field L.
(3.5.2) Definition. (i) A class c of finite groups is called a full class if it is

closed under taking subgroups, homomorphic images and group extensions
(i.e. if 1 → G0 → G → G00 → 1 is an exact sequence of finite groups, then G is
in c if G0 and G00 are).
(ii) If c is a full class of finite groups, a pro-c-group is a projective limit of
groups in c.
(iii) For a full class c of finite groups, S(c) denotes the set of prime numbers p
with ZZ/pZZ ∈ c. For a set S of prime numbers, c(S) denotes the full class of
finite groups whose orders are divisible only by prime numbers in S.

Remark: We have S(c(S)) = S and an inclusion c ⊆ c(S(c)) which is, in

general, strict.
If c is the class of all finite groups, finite p-groups, or finite solvable

groups, we get the profinite groups, the pro-p-groups, or the prosolvable
groups, respectively. The class of pro-p-groups does not contain a nontriv-
ial, proper full subclass, because every finite p-group G has a series of sub-
groups 1 = G0 ⊆ G1 ⊆ · · · ⊆ Gn = G such that Gi−1 is normal in Gi and
Gi /Gi−1 ∼ = ZZ/pZZ for i = 1, . . . , n. In particular, the class of pro-c-groups
contains the class of pro-p-groups if and only if ZZ/pZZ ∈ c.
(3.5.3) Definition. Let c be a full class of finite groups.

(i) For a profinite group G, the group
G(c) = lim
←−
G/U,
G/U ∈c
where U runs through the open normal subgroups of G such that G/U ∈ c, is
called the maximal pro-c-factor group of G.
(ii) For an abelian torsion group A, the subgroup
[
A(c) = B,
B ∈c
where the union is taken over all finite subgroups B ⊆ A with B ∈ c, is called
the c-torsion subgroup of A.
If G is a finite abelian torsion group, then both possible interpretations of

G(c) coincide in the sense that the c-torsion subgroup maps isomorphically
onto the maximal pro-c-factor group.
(3.5.4) Proposition. Let G be a profinite group and let c be a full class of finite
groups.
(i) For the kernel H = ker(G G(c)) we have H(c) = 1.
(ii) Let G0 → G → G00 → 1 be an exact sequence of profinite groups. Then
the sequence
G0 (c) → G(c) → G00 (c) → 1
is exact.
(iii) Let 1 → G0 → G → G00 → 1 be an exact sequence of profinite groups.
If G00 is a pro-c-group, then the sequence
1 → G0 (c) → G(c) → G00 (c) → 1
is exact.

Proof: We start by showing (i). Suppose H has a proper open normal

subgroup N with H/N ∈ c. We let G act on H by conjugation. The subgroup
G0 of all σ ∈ G such that σN σ −1 = N is open in G, since it consists of
all elements σ ∈ G which maps a compact set (namely N ) into an open set
(namely N ). Therefore G/G0 is finite and
σN σ −1
\
Ñ :=
σ ∈G/G0
is open in H and normal in G. The injective homomorphism

H/σN σ −1
Y
H/Ñ −→
σ ∈G/G0
shows that H/Ñ ∈ c. The group extension 1 → H/Ñ → G/Ñ → G/H → 1

shows that G/Ñ is a pro-c-group, contradicting the definition of H.
Now let G0 → G → G00 → 1 be exact. The surjectivity of G(c) → G00 (c) is
immediate. Therefore, in order to show (ii), we may replace G0 by its image
in G and suppose that 1 → G0 → G → G00 → 1 is exact. Let H be the kernel
of G G(c). We have an exact sequence of pro-c-groups
1 → G0 /(G0 ∩ H) → G/H → G/G0 H → 1
and a surjection G0 (c) G0 /(G0 ∩ H). Consider the exact sequence
1 → G0 H/G0 → G/G0 → G/G0 H → 1.
The group G0 H/G0 is a quotient of H and has no nontrivial homomorphisms
to c-groups by (i). We therefore obtain an isomorphism G00 (c) = (G/G0 )(c) →
∼
G/G0 H, showing (ii).

If G00 is a pro-c-group, then H ⊆ G0 and, again using (i), the surjection
G0 → G0 /H induces an isomorphism G0 (c) → ∼ G0 /H. This shows (iii). 2
(3.5.5) Definition. Let c be a full class of finite groups.

(i) A profinite group G is called c-projective if every embedding problem for
G where the kernel is a pro-c-group has a solution.
(ii) A profinite group G is projective, if it is f-projective, where f is the class
of all finite groups.
(iii) If S is a set of prime numbers, then a c(S)-projective group is called
S-projective. If S consists only of one prime number p, i.e. c(S) is the class
of finite p-groups, we call such a group p-projective.
Remark: If G is a pro-c-group which is c-projective in the above sense, then

G is a projective object in the category of pro-c-groups.
Our aim is to prove the following result due to K. W. GRUENBERG [71].

(3.5.6) Theorem. Let c be a full class of finite groups. For a profinite group
G the following assertions are equivalent:
(i) G is c-projective.
(ii) Every group extension 1 → N → E → G → 1 where N is a pro-c-group
splits.
(iii) Every group extension 1 → A → E → G → 1 where A is a finite abelian
p-group with p ∈ S(c) splits.
(iv) cdp G ≤ 1 for all prime numbers p ∈ S(c).
(v) G is S(c)-projective.
The crucial step in the proof of (3.5.6) is the following result.
(3.5.7) Proposition. A profinite group G is c-projective if every embedding

problem ÷øùúû G
(∗)
1 N E G 1
for G is solvable, where E is finite and N ∈ c is a minimal abelian normal
subgroup of E.
Proof: We have to show that every embedding problem (∗) with N a pro-c-
group is solvable. Firstly, we reduce to the case when N is finite. Assume that
all embedding problems (∗) with N a finite c-group are solvable and consider
an embedding problem (∗) with N an arbitrary pro-c-group. Let X be the set
of all pairs (N 0 , ψ 0 ) consisting of a closed subgroup N 0 of N which is normal
in E and a solution ψ 0 : G → E/N 0 of the induced embedding problem
ýþÿü G
ψ0 ϕ
ᾱ
1 N/N 0 E/N 0 G 1.
We write (N 00 , ψ 00 ) ≥ (N 0 , ψ 0 ) if N 00 ⊆ N 0 and if ψ 0 coincides with the composite
ψ 00
G −→ E/N 00 E/N 0 . Then X is inductively ordered and nonempty. By
Zorn’s lemma, there exists a maximal element (N 0 , ψ 0 ). We have to show that
N 0 = 1. Assume the contrary. Then there exists a proper open subgroup U
of N 0 . As the topology of N 0 is induced by that of E, there exists a normal
open subgroup Ẽ ⊆ E such that N 00 = Ẽ ∩ N 0 is contained in U . The group
N 00 is a proper open subgroup in N 0 and normal in E. By assumption, the
embedding problem with finite kernel


G
ψ0
1 N 0 /N 00 E0 im(ψ 0 ) 1
1 N 0 /N 00 E/N 00 E/N 0 1,
where E 0 is the pre-image of im(ψ 0 ) in E/N 00 , has a solution. Therefore we
obtain a homomorphism ψ 00 : G → E/N 00 such that the pair (N 00 , ψ 00 ) is strictly
larger than the pair (N 0 , ψ 0 ). This contradicts the maximality of (N 0 , ψ 0 ).
Next we show that we can also assume E to be finite. Assume all embedding
problems (∗) with E finite and N ∈ c are solvable, and consider an arbitrary
embedding problem with finite kernel N ∈ c. Since N is finite, there exists an
open normal subgroup U of E with U ∩ N = 1. We obtain the commutative
and exact diagram G
ϕ
1 N E G 1
ψ
1 N E/U G/im(U ) 1,
where the broken arrow is obtained by the assumption of this reduction. By
(3.5.8) below, the right hand square is cartesian, i.e. the natural map from E to
the fibre product E/U ×G/im(U ) G is an isomorphism. Therefore the pair (ψ, ϕ)
induces a homomorphism G −→ E, completing the diagram commutatively.
Now we prove by induction on the (finite) order of N ∈ c that every em-
bedding problem (∗) with E finite has a solution. We distinguish between the
following three cases:
(1) N is not a minimal normal subgroup of E. Then we can choose a normal
subgroup M of E such that 1 $ M $ N . By induction,
! G
ψ
1 N/M E/M G 1
has a solution ψ: G → E/M . Again by induction, the embedding problem
#$%&" G
ψ
1 M E E/M 1
has a solution.

(2) Let N be a minimal normal subgroup of E and assume that N is not

contained in the Frattini subgroup Φ(E) of E (which is the intersection of all
maximal subgroups of E). Then there exists a maximal subgroup E1 of E such
/ E1 , hence N E1 = E, i.e. E1 projects onto G. Then N1 = E1 ∩ N is a
that N ⊆
proper subgroup of N and thus, by induction, the embedding problem
'()*+,-./0123 G
1 N1 E1 G 1
1 N E G 1
has a solution.
(3) Let N be a minimal normal subgroup of E and assume that N ⊆ Φ(E).
Since Φ(E) is nilpotent (see [81], III Satz 3.6), the group N is abelian. By
assumption, we get a solution of the embedding problem in this case. 2
(3.5.8) Lemma. Let E be a profinite group and let U and V be closed normal
subgroups of G. Then the diagram of profinite groups
E/U6754 ∩V
pU
E/U
pV
E/V E/U V
is cartesian, i.e. the projections define an isomorphism
E/U8 ∩V
(pU ,pV )
E/U ×E/U V E/V,
where the group on the right hand side is the fibre product, i.e. the subgroup
of elements (a, b) in the product E/U × E/V such that a and b project to the
same element in E/U V .
Proof: Injectivity is trivial. Let a, b ∈ E be elements with the same image

in E/U V and consider the element (pU (a), pV (b)) ∈ E/U ×E/U V E/V . By
assumption, ab−1 = uv for elements u ∈ U , v ∈ V . Then the element
g = u−1 a = vb ∈ E maps to ((pU (a), pV (b)), which shows surjectivity. 2
In order to detect c-projective groups, we are, by (3.5.7), reduced to a special

case of embedding problems with finite abelian kernel. Next we investigate

arbitrary embedding problems with finite abelian kernel A = ker(α). The group
α
G acts on A by conjugation and the group extension 1 → A → E → G → 1 is
determined by its class ε ∈ H 2 (G, A), see (1.2.4).
Let E (G, ϕ, α) be an embedding problem with finite abelian kernel A for
the profinite group G and let H = ker ϕ. Consider the diagram
@<=>?9;: H
ψ0
G
ψ ϕ
1 A E α G 1.
We consider A as a G-module via ϕ. Then H acts trivially on A. A solution ψ
induces a Ḡ-homomorphism ψ0 : H → A. From the Hochschild-Serre spectral
sequence we get the exact sequence (cf. (1.6.7))
res tg inf
0 −→ H 1 (G, A)−→ H 1 (G, A) −→ H 1 (H, A)Ḡ −→ H 2 (G, A) −→ H 2 (G, A).
(3.5.9) Proposition (HOECHSMANN). Let ε ∈ H 2 (G, A) be the cohomology

class corresponding to the group extension of the embedding problem E (G) =
E (G, ϕ, α), where A = ker(α) is a finite Ḡ-module.
Then E (G) has a solution if and only if inf (ε) = 0, i.e. if there exists a
Ḡ-homomorphism ψ0 : H → A with tg(ψ0 ) = ε.
Proof: (see [80], 1.1). Consider the commutative exact diagram

1KABCHIJDEFG
pr2
A E ×G G G 1
pr1 ϕ
1 A E α G 1
where E ×G G = {(e, σ) ∈ E × G | α(e) = ϕ(σ)} is the fibre product of
α and ϕ. The group extension above corresponds to inf (ε) ∈ H 2 (G, A). If
inf (ε) = 0, then this extension splits, i.e. there exists a homomorphism
s: G → E ×G G
such that pr2 ◦ s = id. Obviously, ψ = pr1 ◦ s is then a solution of E (G).
Conversely, a solution ψ defines a section s of pr2 by s(σ) = (ψ(σ), σ), and
therefore inf (ε) = 0. 2

(3.5.10) Corollary. An embedding problem E (G, ϕ, α),

LPMNO G
ϕ
α
1 A E G 1,
where A is a finite abelian p-group, is solvable if and only if the corresponding

p-Sylow embedding problem
URSTQ Gp
ϕp
αp
1 A Ep Gp 1
is solvable. Here the index p indicates the corresponding p-Sylow subgroups,

which are chosen in a compatible way.
Proof: The 2-cocycle εp of the Sylow problem is just the restriction of the
2-cocycle ε of the initial problem. Therefore the result follows from (3.5.9)
and the commutative diagram
[εVWXY p ] ∈ H 2 (Gp , A) H 2 (Gp , A)
[ε] ∈ H 2 (G, A) H 2 (G, A),
where the restriction maps are injective by (1.6.10). 2
(3.5.11) Proposition. Let E (G, ϕ, α) be an embedding problem with finite

abelian kernel A which has a solution. Then SE (G) is a principal homogeneous
space over H 1 (G, A).∗)
The homogeneous subspaces over H 1 (G, A) ⊆ H 1 (G, A) consist of all so-
lutions ψ modulo equivalence whose restrictions to H = ker(ϕ) induce a fixed
G-homomorphism ψ0 : H ab → A.
Proof: Let ψ be a solution of E (G) and let [x] ∈ H 1 (G, A), where x: G → A
is a 1-cocycle. Then
x
ψ: G −→ E , σ 7−→ x(σ) · ψ(σ)
∗) If X is a topological space and Γ a topological group acting continuously and simply

transitively on X, then X is called a principal homogeneous space over Γ .

is a solution of E (G). Indeed, x ψ is a homomorphism since

x(στ ) · ψ(στ ) = x(σ)x(τ )ψ(σ) ψ(σ)ψ(τ ) = x(σ)ψ(σ)x(τ )ψ(τ )
and α ◦ x ψ = ϕ. Another 1-cocycle x0 ∈ [x] induces an equivalent solution of
x
ψ. Thus H 1 (G, A) acts on SE (G) . This action is transitive since for any two
solutions ψ and ψ 0 we get a 1-cocycle x(σ) = ψ 0 (σ)ψ(σ)−1 , hence a class in
H 1 (G, A). Furthermore, the action is simply transitive since two solutions ψ
and ψ 0 are equivalent if and only if x is a 1-coboundary. Finally, from the exact
5-term sequence, it follows that ψ|H = ψ 0 |H if and only if [x] ∈ inf (H 1 (Ḡ, A)).
2
Now we are able to prove the theorem.
Proof of (3.5.6): (i) ⇒ (ii) is trivial and (ii) ⇒ (iii) is also trivial, noting that
the classes c and c(S(c)) contain the same abelian groups.
(iii) ⇒ (iv). Let p ∈ S(c) and let A be a finite G-module in Modp (G). By
assumption, every group extension 1 → A → E → G → 1 splits and therefore,
by (1.2.4),
H 2 (G, A) ∼
= EXT (A, G) = 0.
An arbitrary G-module A ∈ Modp (G) is the union of its finite submodules.
Taking the inductive limit, we see that H 2 (G, A) = 0, and hence cdp G ≤ 1.
(iv) ⇒ (v). Let us assume that cdp G ≤ 1 for all p ∈ S(c). Then H 2 (G, A) = 0
for all finite p-primary G-modules A, p ∈ S(c). By (3.5.9), every embedding
problem (∗) with a minimal abelian normal subgroup A ∈ c(S(c)) of E is
solvable, since such an A is necessarily of the form (ZZ/pZZ)m for some prime
number p ∈ S(c) and some integer m ≥ 1. By (3.5.7), G is S(c)-projective.
Finally, (v) ⇒ (i) is trivial, as c ⊆ c(S(c)). 2
(3.5.12) Corollary. If c ⊆ d are two full classes of finite groups and if G is a

pro-c-group which is c-projective, then it is d-projective.
Proof: We use the equivalence (3.5.6) (i)⇔(iii), thus we have to show that
every group extension
1→A→E →G→1
splits, where A is a finite abelian p-group, p ∈ S(d). By assumption, this is
true for p ∈ S(c). If p ∈ S(d) r S(c), then the orders of the groups G and A are
relatively prime, hence EXT (A, G) ∼ = H 2 (G, A) = 0 by (1.6.2). 2

(3.5.13) Corollary. Denoting by (solv) the full class of finite solvable groups,
the following assertions are equivalent for a profinite group G.
(i) G is projective.
(ii) G is (solv)-projective.
(iii) cd G ≤ 1.
Proof: This follows easily from (3.5.6) (i)⇔(iv). 2
Let p be a prime number. From (3.3.7) we know that the profinite groups
G with cdp G = 0 are exactly the groups of order prime to p. By (3.5.6), the
profinite groups with cdp G ≤ 1 are exactly the p-projective groups. Prototypes
of projective groups are the free pro-c-groups, which are defined as follows.
(3.5.14) Definition. A free pro-c-group over a set X is a pro-c-group F

together with a map i: X → F satisfying the following properties.
(1) Every open subgroup contains almost all elements of i(X) (i.e. all up to
a finite number).
(2) If j : X → G is any other map with the property (1) into a pro-c-group
G, then there exists a unique homomorphism f : F → G with j = f ◦ i.
The free pro-c-group F over a set X always exists and is unique up to unique
isomorphism. One obtains it by starting with the ordinary free group F0 over
X (see [73]) with the inclusion X ,→ F0 . Let U run through all normal
subgroups containing almost all elements x ∈ X and such that F0 /U ∈ c. Then
F = lim F0 /U , together with the induced map i : X → F , is a free pro-c-
←− U
group over X. In fact, if j : X → G is as in (2), then the universal property of
the free group F0 gives a homomorphism f0 : F0 → G such that j = f0 ◦ incl X .
If U runs through all open normal subgroups of G, then we obtain f : F → G
as the composite of the homomorphisms
F −→ lim F0 /f0−1 (U ) −→ lim G/U = G.
←− ←−
U U
f is unique as it is continuous and F is topologically generated by i(X).
Observe that the map i: X → F in (3.5.14) is necessarily injective. The
elements of i(X) ⊆ F are called the free generators of F , the set i(X) is
called a basis of F , and rk(F ) = #X is the rank of the free pro-c-group. We
also write FX (c) for F and Fn (c), resp. Fω (c), if rk(F ) = n ∈ IN, resp. if X has
countable cardinality.

The free pro-c-group of rank 0 is the trivial group. The free pro-c-group of
Q
rank 1 is the product ẐZ(c) = p∈S(c) ZZp .
(3.5.15) Proposition. Let c be a full class of finite groups and let F be the free
pro-c-group over a set X. Then F is projective.
Proof: By (3.5.12), it suffices to show that F is c-projective. Let

π
1→A→E →F →1
be an exact sequence with A a finite abelian p-group for some prime number p ∈
S(c). Then A ∈ c and E is a pro-c-group. We choose a continuous section
F → E of π and obtain a lift s: X → E of i: X → F , which satisfies condition
(1) of (3.5.14). By the universal property for F , s extends to a homomorphism
s: F → E such that π ◦ s is the identity on X, hence on F . This shows that
the extension 1 → A → E → F → 1 splits. By (3.5.6), F is a projective pro-c-
group. 2
(3.5.16) Corollary. Let F be the free pro-c-group over a set X. If the class c
contains the p-groups, then cdp F = 1, otherwise cdp F = 0.
Proof: By (3.3.7), cdp F = 0 if and only if the class c does not contain the
p-groups. As F is projective, (3.5.6) implies cdp F ≤ 1 for all p. 2
For pro-p-groups we have a converse:
(3.5.17) Proposition. A pro-p-group G is free if and only if cd G ≤ 1.
We shall prove this in (3.9.5). See also (4.1.5) for a more general result.
(3.5.18) Definition. The rank rk(G) of a pro-c-group G is the smallest car-

dinal number α such that there exists a set X of cardinality α and a surjection
FX (c) G from a free pro-c-group over X onto G.
The rank of a pro-p-group G is often denoted by d(G). In (3.9.1) we will

see that d(G) = dimIFp H 1 (G, ZZ/pZZ).

In the following, we will prove that a pro-c-group G of at most countable

infinite rank, for which every embedding problem with E a finite c-group has
a proper solution, is isomorphic to the free pro-c-group Fω (c) of countable
infinite rank. We start by showing that Fω (c) itself has this property.
(3.5.19) Lemma. Every embedding problem

^[\]Z Fω (c)
ϕ
α
1 N E G 1
with E a finite c-group has a proper solution.
Proof: Let X be a basis of Fω (c). Then Y = X r ker(ϕ) is finite. For each

y ∈ Y we choose an element ey ∈ E such that α(ey ) = ϕ(y). Furthermore,
we choose a surjective map ψ00 : X ∩ ker(ϕ) → N (observe that X ∩ ker(ϕ)
is infinite) and define ψ0 : X → E by ψ0 (y) = ey for y ∈ Y and by ψ00 on
X ∩ ker(ϕ). Now ψ0 extends to a surjective homomorphism ψ: Fω (c) E
such that α ◦ ψ = ϕ. 2
The following result is due to K. IWASAWA (see [86], th. 4).
(3.5.20) Proposition. Let G be a pro-c-group of at most countable infinite

rank such that every embedding problem
àbc_ G
ϕ
α
1 N E G 1
with E a finite c-group has a proper solution. Then G is isomorphic to Fω (c).
Proof: Let F = Fω (c). Since G and F have at most countable rank, there
exist sequences
F = F0 ⊇ F1 ⊇ F2 ⊇ . . .
G = G0 ⊇ G1 ⊇ G2 ⊇ ...
of open normal subgroups with trivial intersection. We inductively define two

additional sequences
F = F00 ⊇ F10 ⊇ F20 ⊇ ...
G = G00 ⊇ G01 ⊇ G02 ⊇ ...
of open normal subgroups and a sequence of isomorphisms ϕn : G/G0n →
∼
F/Fn0 such that the diagram
G/Ggfed 0n+1
ϕn+1 0
∼ F/Fn+1
can can
ϕn
G/G0n ∼ F/Fn0
commutes for every n. Let n be given and suppose all objects with indices up
0
to n have already been defined. If n is even, then Fn+1 := Fn ∩ Fn0 is open in
0
F and F/Fn+1 ∈ c. The embedding problem
hikj G
ϕ0n+1
can
0 ∼
F/Fn+1 can F/Fn0 (ϕn )−1
G/G0n
has a proper solution ϕ0n+1 by assumption. If G0n+1 = ker(ϕ0n+1 ), then ϕ0n+1

induces an isomorphism ϕn+1 : G/G0n+1 → ∼ F/F 0
n+1 which commutes with ϕn .
If n is odd, exchange the roles of G and F and use (3.5.19) in order to find an
0 ∼ G/G0 . Let ϕ −1 0
isomorphism ψn+1 : F/Fn+1 → n+1 n+1 = ψn+1 . Since Fn+1 ⊆ Fn
0 0 0
and Gn+1 ⊆ Gn , the intersection of all groups Fn (resp. Gn ) is equal to 1. Thus
the isomorphisms ϕn define an isomorphism ϕ: G → ∼ F. 2
Exercise: Consider the diagram
pmnol G
f
π
1 P E Γ 1
with P an abelian group. Two liftings f 0 , f 00 : G → E of f : G → Γ are conjugate if there

exists a σ ∈ P such that f 00 = σ ◦ f 0 ◦ σ −1 . Show that the group H 1 (G, P ) acts simply
transitively on the set of conjugacy classes of liftings, provided some liftings exist. Formulate
and prove this also for non-abelian P .

§6. Profinite Groups of scd G = 2
The profinite groups G of strict cohomological dimension scd G = 2 are

especially important for the theory of number fields, because of their close
connection with class field theory. Trivial examples are all groups of cohomo-
logical dimension 1 (see §3, ex.1).
Let G be any profinite group. We denote open subgroups of G by U, V, W
and we write V / U if V is a normal subgroup of U . In this case V ab is a
U/V -module by conjugation, and we have a group extension
1 −→ V ab −→ U/V 0 −→ U/V −→ 1,
where V 0 denotes the closure of the commutator subgroup [V, V ] of V . By
(1.2.4), the group extension defines a canonical cohomology class
uU/V ∈ H 2 (U/V, V ab ),
which plays an important role in the sequel.
(3.6.1) Lemma. Let W ⊆ V ⊆ U be open subgroups of G.

(i) If W / U , then uV /W is the image of uU/W under
res : H 2 (U/W, W ab ) −→ H 2 (V /W, W ab ).
(ii) If W / U and V / U , then (V : W )uU/W is the image of uU/V under the
map
i : H 2 (U/V, V ab ) −→ H 2 (U/W, W ab ),
induced by the pair U/W → U/V, Ver : V ab → W ab .
Proof: Assertion (i) follows from the commutative diagram

1qrstuvwxyz{ W ab V /W 0 V /W 1
1 W ab U/W 0 U/W 1.
For (ii) we consider the commutative diagram
1|}~ W ab U/W 0 U/W 1
1 V ab U/V 0 U/V 1.
From this, we conclude that uU/W and uU/V have the same image u0 in
Ver
H 2 (U/W, V ab ). The composite of W ab −→ V ab −→ W ab is the norm NV /W
by (1.5.9). Therefore we have a commutative diagram

§6. Profinite Groups of scd G = 2 203
V ab )
H 2 (U/V,
inf
H 2 (U/W, V ab ) H 2 (U/W, W ab )
Ver ν
i
H 2 (U/W, W ab ),
where ν is induced by NV /W : W ab → W ab . This yields i(uU/V ) = Ver(u0 ) =

ν(uU/W ). But if g is a finite group, h a normal subgroup, A a g-module, then
the map Ĥ n (g, A) → Ĥ n (g, A) induced by Ng/h : A → A is multiplication by
(g : h). This is trivial for n = 0 and follows for n ≥ 0 by dimension shifting.
Therefore i(uU/V ) = (V : W )uU/W . 2
(3.6.2) Lemma. For every pair V / U of open subgroups of G, we have an


H2 (U/V, ẐZ) H0 (U/V, V ab ) i
U ab (U/V )ab 1
NU/V Ver ρ
1 NU/V V ab (V ab )U/V Ĥ 0 (U/V, V ab ) 1,

where the homomorphism ρ is given by
Y
σ 7−→ u(τ, σ),
τ ∈U/V
u(τ, σ) being a 2-cocycle representing the canonical class uU/V . If U/V is

cyclic, the homomorphism i is injective.
Proof: The Hochschild-Serre spectral sequence

H i (U/V, H j (V, Q/ZZ)) ⇒ H i+j (U, Q/ZZ)
induces the exact sequence
0 → H 1 (U/V, Q/ZZ) → H 1 (U, Q/ZZ) → H 1 (V, Q/ZZ)U/V → H 2 (U/V, Q/ZZ).
Dualizing this sequence, we obtain the upper exact sequence of the diagram.
The lower sequence is trivially exact.
Ver
The composite of V ab −→ U ab −→ V ab is the norm map NU/V by (1.5.9),
which shows that the left partial diagram of (3.6.2) is commutative. Therefore
Ver induces a homomorphism ρ: (U/V )ab → Ĥ 0 (U/V, V ab ). In order to de-
scribe it explicitly, we choose a representative σ̂ ∈ U for each coset σ ∈ U/V ,
i.e. σ = σ̂V = V σ̂. Then the function
u(τ, σ) = τ̂ σ̂ τcσ −1 mod V 0

is a cocycle representing the class uU/V . On the other hand, by definition of

the transfer Ver,
Ver(σ̂ mod U 0 ) = τ̂ σ̂ τcσ −1 mod V 0 =
Y Y
u(τ, σ).
τ ∈U/V τ ∈U/V
Passing from U to U/V and then to (U/V )ab , this map induces the map
ρ : (U/V )ab → Ĥ 0 (U/V, V ab ), given by
Y
σ 7−→ u(τ, σ)
τ ∈U/V
as claimed.
If U/V is cyclic, then
H2 (U/V, ẐZ) ∼
= H 2 (U/V, Q/ZZ)∗ ∼
= Ĥ 0 (U/V, Q/ZZ)∗ = 0,
and therefore the homomorphism i in (3.6.2) is injective. 2
(3.6.3) Lemma. Consider the conditions

(1) H 1 (U/V, V ab ) = 1,
(2) H 2 (U/V, V ab ) is of order #(U/V ) and is generated by uU/V .
If they hold for all pairs V / U such that U/V is cyclic of a prime order p, then
they hold for all pairs V / U .
Proof: Assume the conditions hold whenever U/V ∼ = ZZ/pZZ. We then prove
them for the case #U/V = pn by induction on n. Since the p-group U/V has a
nontrivial center, it sits in an exact sequence
1 −→ W/V −→ U/V −→ U/W −→ 1
with W/V ∼ = ZZ/pZZ and #U/W = pn−1 . We may therefore assume the
conditions (1) and (2) for W/V and U/W . From (3.6.2), with U replaced by
W , it follows that
Ver : W ab −→(V ab )W/V
is an isomorphism, since NW/V is an isomorphism as the kernel of NW/V is
equal to Ĥ −1 (W/V, V ab ) ∼
= H 1 (W/V, V ab ) = 1 and ρ is the Nakayama map
(cf. III §1). Therefore, by (1.6.7), we obtain exact sequences
i res
1 −→ H q (U/W, W ab ) −→ H q (U/V, V ab ) −→ H q (W/V, V ab )
for q = 1 and for q = 2 since H 1 (W/V, V ab ) = 1.
For q = 1 we obtain H 1 (U/V, V ab ) = {1}, and for q = 2
#H 2 (U/V, V ab ) ≤ pn = (U : V ).

Furthermore, uU/V must be of order pn , since otherwise 1 = pn−1 uU/V =

i(pn−2 uU/V ) by (3.6.1). As i is injective, this implies pn−2 uU/V = 1, contra-
dicting the induction hypothesis. This proves condition (2).
We now prove the conditions (1) and (2) for a general pair V / U . For every
prime number p, let Up /V be a p-Sylow subgroup of U/V . By (1.6.10), the
restriction map
res : H q (U/V, V ab ) −→ H q (Up /V, V ab )
M
= 1 yields H 1 (U/V, V ab ) = {1}. When

for q = 1, 2, is injective. The case q M
q = 2, uU/V is mapped to the element uUp /V by (3.6.1), which generates the
p
direct sum. Therefore res is an isomorphism, uU/V generates H 2 (U/V, V ab )
and
#H 2 (U/V, V ab ) = #H 2 (Up /V, V ab ) = 2
Y Y
#Up /V = #U/V.
p p
We are now able to prove the following theorem, which gives a four-fold
characterization of the profinite groups of strict cohomological dimension 2.
(3.6.4) Theorem. For a profinite group G =/ 1, the following conditions are

equivalent:
(i) scd G = 2.
(ii) For every pair V / U , the transfer
Ver : U ab −→(V ab )U/V
is an isomorphism.∗)
(iii) For every pair V / U , the U/V -module V ab is a class module with
fundamental class uU/V , i.e. the conditions (1) and (2) of (3.6.3) hold.
(iv) The G-module D2 = lim U ab is a formation module with D2U = U ab and
−→
with (uU/V ) as a system of fundamental classes.
(v) There exists a level-compact formation module C with trivial universal
norm groups NU C for all open subgroups U ⊆ G.
Proof: (i) ⇒ (ii): This follows from (3.3.11) and (1.5.9).

(ii) ⇒ (iii): Assume (ii) and let V / U be a pair such that U/V is cyclic of
prime order p. Consider the exact commutative diagram (3.6.2), in which the
homomorphism
∗) In other words, the fundamental G-modulation π ab : U 7→ U ab is a G-module (see I §5).

i : H0 (U/V, V ab ) → U ab
is injective and ker(NU/V ) = Ĥ −1 (U/V, V ab ). By the snake lemma, the bijec-
tivity of Ver implies
H 1 (U/V, V ab ) ∼
= Ĥ −1 (U/V, V ab ) = 1
H 2 (U/V, V ab ) ∼
= Ĥ 0 (U/V, V ab ) ∼= U/V ∼
= ZZ/pZZ.
In order to show that uU/V generates H 2 (U/V, V ab ), i.e. uU/V =/ 0, consider

the exact commutative diagram
1 ¡ V ab U/V 0 U/V 1
1 H0 (U/V, V ab ) U ab U/V 1.
If uU/V = 0, then the upper group extension splits, i.e. U/V 0 is a semi-
direct product V ab o Γ with a group Γ which is mapped isomorphically onto
U/V . Its image in U ab is nontrivial. But in the case of a semi-direct product
V ab o Γ , the transfer Ver : V ab o Γ → V ab maps Γ to 1, since for γ ∈ Γ ,
Ver(γ) = β ∈Γ βγ(γβ)−1 = 1. Therefore Ver : U ab →(V ab )U/V is not injective,
Q
a contradiction. This proves (iii) in the case U/V ∼ = ZZ/pZZ, and the general
case follows because of lemma (3.6.3).
(iii) ⇒ (i): If (iii) holds, then by §1, ex.4, the cup-product
uU/V ∪ : H p (U/V, Hom(V ab , Q/ZZ)) −→ H p+2 (U/V, Q/ZZ),
induced by the pairing V ab × Hom(V ab , Q/ZZ) → Q/ZZ is an isomorphism for
p > 0 and a surjection for p = 0. By (2.4.4), this map coincides up to sign with
the differential
d2p,1 : E2p,1 −→ E2p+2,0
of the Hochschild-Serre spectral sequence
E2p,q = H p (U/V, H q (V, Q/ZZ)) ⇒ H p+q (U, Q/ZZ).
The surjectivity implies E3p+2,0 = 0 for p ≥ p,0
0 and thus E∞ = 0 for p ≥ 2.
Therefore the edge morphism
E2p,0 = H p (U/V, Q/ZZ) −→ H p (U, Q/ZZ)
p,0
is the zero map since it factors through E∞ . But this edge morphism is the
inflation map, and we obtain
H p+1 (U, ZZ) ∼
= H p (U, Q/ZZ) = lim H p (U/V, Q/ZZ) = 0
−→
V
for p ≥ 2. From this it follows that scd G = 2 (see §3, ex.6), noting that
scd G = 0 or 1 is impossible.

(ii) ⇒ (iv): The inclusion functor

i : Mod (G) −→ M od (G)
from the category of G-modules into the category of G-modulations identifies
Mod (G) with the full subcategory of G-modulations M with “Galois descent”,
i.e. for which res : M (U ) → M (V )U/V is an isomorphism. The functor i has
as left adjoint the functor
j : M 7−→ lim M U .
−→
U
We have, in particular, D2 = j(π ab ), and the condition (ii) means that the
fundamental G-modulation π ab has Galois descent, i.e. π ab = i(D2 ). Therefore
D2U = (iD2 )(U ) = π ab (U ) = U ab .
Since (ii) is equivalent to (iii), the G-module D2 is a formation module with
respect to the isomorphisms
invU/V : H 2 (U/V, V ab ) −→ (U1:V ) ZZ/ZZ
1
given by uU/V 7→ (U :V )
mod ZZ, i.e. with (uU/V ) as a system of fundamental
classes.
(iv) ⇒ (v): Assuming (iv), D2 is a level-compact formation module, since
D2U = U ab . Therefore it suffices to show that the universal norm groups NU D2
are trivial. For this we consider, for every pair V / U , the diagram
V ©ª¨§¦¥¤¢£ ab U ab
NU/V Ver
i V ab i
NU/V
D2V D2U
i
NU/V
D2V
The top of this diagram is commutative by (1.5.9), the bottom and the left-hand
side diagram are trivially commutative and the right-hand side diagram by the
definition of D2 . Therefore the back diagram is commutative, i.e. we have a
commutative diagram
V®«¬ U
NU/V
D2V D2U
with surjective vertical arrows. Since V ⊆U V = {1}, we obtain
T
NU/V D2V = {1},

\
NU D2 =
V ⊆U
what we wanted to prove.

(v) ⇒ (ii): Let C be a level-compact formation module with trivial universal

norm groups. By class field theory, we have, for every open subgroup U in
G, a canonical continuous homomorphism C U → U ab with dense image. It is
surjective, since C U is compact, and injective, since the kernel NU C is 1. If V
is an open normal subgroup of U , then we have the commutative diagram
C²¯°± U (C V )U/V
Ver
U ab (V ab )U/V ,
which shows that Ver is an isomorphism. 2
Remark: The equivalence (i) ⇔ (iii) in the above theorem is due to J. TATE,
as Y. KAWADA remarks. Kawada gives Tate’s proof in [101] (see ex.4). The
equivalence was independently proven by A. BRUMER (see [19]). Brumer’s
proof, however, is rather involved. Another proof is given in [72]. The
equivalence (ii) ⇔ (iii) was first proven by J.-P. SERRE (see [208], chap. VI),
and another proof of it is found in [72], 2.3. These proofs rely in an essential
way on the use of negative dimensional cohomology.
In the proof presented here, we have also shown that the (level-compact)
dualizing module D2 = lim U ab of a profinite group G of scd G = 2 has trivial
−→
universal norm groups NU D2 . Therefore we may apply POITOU’s duality
theorem (3.1.11)(i) to the G-module D2 and obtain the
(3.6.5) Corollary. Let G be a profinite group of scd G = 2. We then have a

inv : H 2 (G, D2 ) −→
∼ 1
#G
ZZ/ZZ
and the cup-product

∪
Ĥ i (G, Hom(A, D2 )) × Ĥ 2−i (G, A) −→ H 2 (G, D2 ) ∼
= 1
#G
ZZ/ZZ
∼ Ĥ 2−i (G, A)∨
Ĥ i (G, Hom(A, D2 )) −→
for all i ∈ ZZ and every G-module A which is a finitely generated free ZZ-
module. For i ≤ 0 this is true also for every G-module A, finitely generated
over ZZ.

By the equivalence (i) ⇔ (v) in theorem (3.6.4), class field theory seems
to be very strongly restricted to cohomological dimension 2. This, however,
is not really true. The restriction is due to the fact that we insist on the
presence of a G-module rather than a G-modulation with trivial universal
norms and the properties of a class formation. For example, the fundamental
group G = π1 (X, x) of a smooth proper curve X over a finite field IF has
scd G = 3. Thus, by the equivalence (i) ⇔ (ii) of (3.6.4), its fundamental
G-modulation π ab is not a G-module. We have, on the other hand, the G-
modulation P ic : U 7→ Pic (X(U )), which associates to an open subgroup the
Picard group Pic (X(U )) of the unramified covering X(U ) → X determined by
U . Passing to the profinite completion Pdic, we obtain a canonical isomorphism
π ab ∼
= Pdic
of G-modulations (see [208]).
Exercise 1. A profinite group G of scd G ≤ 2 contains no closed abelian subgroups other than
procyclic ones.
Exercise 2. The group G = π1 (X, x), where X is a smooth proper curve over a finite field,
has the property of the groups of ex.1 but has scd G = 3.
Exercise 3. Not every group of scd G = 3 has the property of the groups of ex.1.
Exercise 4. Prove the implication (v) ⇒ (i) in (3.6.4) directly by means of Poitou’s duality
theorem (3.1.11)(i) for i = −1.
Hint: If C is level-compact with trivial universal norms, then by (3.1.11)(i) and (1.9.12), we
have for A = ZZ and A = ZZ/pZZ
H 3 (U, A)∗ ∼
= Ĥ −1 (U, Hom(A, C))
= Ĥ −1 (U, NU Hom(A, C))
= Ĥ −1 (U, Hom(A, NU C)) = 0.
Let Gp be a p-Sylow subgroup of G. Then
H 3 (Gp , ZZ/pZZ) = lim H 3 (U, ZZ/pZZ) = 0,
−→
U ⊇Gp
since
\
Gp = U = lim U,
←−
U ⊇ Gp U ⊇Gp
hence cdp G = cd Gp ≤ 2 by (3.3.2). Since H 3 (U, ZZ) = 0 for all open subgroups U , we obtain
scd G ≤ 2 by (3.3.4), and scd G = 2, since
H 2 (G, ZZ) = H 1 (G, Q/ZZ) =/ 0.

§7. Poincaré Groups
Poincaré groups are groups for which a cohomological duality theorem

of Poincaré type H i × H n−i → H n ,→ Q/ZZ holds with a special dualizing
module. These groups play an important role in topology as well as in number
theory. For example, the pro-p-completion of the fundamental group of a
compact Riemann surface, and the Galois group of the maximal p-extension
of a p-adic local field are typical examples of Poincaré groups. For these and
other reasons they are of special interest. We fix a prime number p for our
investigations.
In II §5 we have introduced, for every G-module A, the G-modules
Di (A) = lim H i (U, A)∗ ,
−→
U
and we have called the G-module
D(p) = lim
−→
Dn (ZZ/pν ZZ)
ν
the dualizing module of G at p if n = cdp G. Its importance lies in the
functorial isomorphism
H n (G, A)∗ ∼
= HomG (A, D(p) )
for all A ∈ Modp (G). For this section we set I = D(p) .
We have called G a duality group at p of dimension n if Di (ZZ/pZZ) = 0
for i < n (see (3.4.6)). In this case, the edge morphism of the Tate spectral
sequence
H i (G, Dn−j (A)) ⇒ H n−(i+j) (G, A)∗
for finite A ∈ Modp (G) is a functorial isomorphism
H i (G, Hom(A, I)) ∼ = H n−i (G, A)∗ ,
which is also obtained from the cup-product
∪ tr
H i (G, Hom(A, I)) × H n−i (G, A) −→ H n (G, I) −→ Qp /ZZp .
(3.7.1) Definition. A duality group G at p of dimension n is called a Poincaré

group at p if I ∼= Qp /ZZp as an abelian group.
Since
Aut(I) = Aut(Qp /ZZp ) = ZZ× p,
the group G acts on I via a continuous character
χ : G −→ ZZ× p,
and I is determined by χ (up to isomorphism). The Serre criterion (3.4.5)
implies that scdp G = n + 1 if χ(G) is finite and scdp G = n if χ(G) is infinite.

§7. Poincaré Groups 211
(3.7.2) Theorem. For a finitely generated profinite group G the following

assertions are equivalent.
(i) G is a Poincaré group at p of dimension 2,
(ii) cdp G = 2 and I ∼ = Qp /ZZp (as an abelian group),
(iii) cdp G = 2 and p I ∼
= ZZ/pZZ (as an abelian group).
Proof: The implications (i) ⇒ (ii) ⇒ (iii) are trivial. Now we show that
(iii) implies Di (ZZ/pZZ) = 0 for i = 0, 1. The p-Sylow subgroups Gp of G
are infinite, since 0 < cdp G = cdp Gp < ∞. This means that for every open
subgroup U , there is an open subgroup V ⊆ U such that p | (U : V ). From
this, it follows that
D0 (ZZ/pZZ) = lim H 0 (U, ZZ/pZZ)∗ = lim ZZ/pZZ = 0,
−→ −→
U U
since the transition maps in the last direct limit are multiplication by (U : V ).
Let A ∈ Modp (G) be a finite G-module with pA = 0 and let us set
0
A = Hom (A, I). Every open subgroup U of G has the same cohomolog-
ical dimension 2 as G (see (3.3.5)) and the same dualizing module I. Noting
00
that A = A, we obtain a canonical isomorphism
(∗) ϕ∗A0 : H 0 (U, A)∗ −→
∼ H 2 (U, A0 ).
Let U ∗ = U p [U, U ], where [U, U ] is the closure of the commutator subgroup

of U . The group U/U ∗ is the largest profinite quotient of U which is abelian
and of exponent p. The restriction
res : H 1 (U, ZZ/pZZ) −→ H 1 (U ∗ , ZZ/pZZ)
is obviously the zero map. Since G is finitely generated, so is U , and thus
U/U ∗ is finite, i.e. U ∗ is open. Now let A be the finite U -module defined by
the exact sequence
∗
0 −→ ZZ/pZZ −→ IndUU (ZZ/pZZ) −→ A −→ 0.
Applying the exact functor Hom (−, I), we obtain another exact sequence
∗
0 −→ A0 −→ IndUU (ZZ/pZZ)0 −→ p I −→ 0.
By (1.6.5), the composite of the maps
∗ ∗ sh
H 1 (U, ZZ/pZZ) −→ H 1 (U, IndUU (ZZ/pZZ)) −→ 1
∼ H (U , Z
Z/pZZ)
is the restriction, hence the zero map. Therefore we obtain from the two exact
sequences the exact commutative diagram
´µ¶·¸¹³ H 1 (U, p I)
αU
H 2 (U, A0 )
∗
H 2 (U, IndUU (ZZ/pZZ)0 )
∗
0 H 1 (U, ZZ/pZZ)∗ H 0 (U, A)∗ H 0 (U, IndUU (ZZ/pZZ))∗

where the vertical arrows are the isomorphisms (∗). From this diagram, we get
on the one hand an injection
D1 (ZZ/pZZ) = lim H 1 (U, ZZ/pZZ)∗ ,→ lim im(αU ) ,
−→ −→
U,cor ∗ U
and on the other hand a surjection

lim
−→
H 1 (U, p I) −→ lim
−→
im(αU ).
U,res U
The left-hand group is zero by (1.5.1), hence D1 (ZZ/pZZ) = 0. Hence G is a

duality group at p of dimension 2. Further, I ∼
= Qp /ZZp or ZZ/pk ZZ, k ≥ 1 (as an
abelian group) since p I ∼
= ZZ/pZZ. But I is p-divisible by (3.4.7) and therefore
G is a Poincaré group. 2
(3.7.3) Corollary. Let G be a finitely generated pro-p-group of cohomological

dimension equal to 2. Then G is a Poincaré group if and only if
dimIFp H 2 (N, ZZ/pZZ) = 1
for every open normal subgroup N of G.
Proof: Since cdp G = 2, we have for open normal subgroups N 0 ⊆ N of the

group G
H 2 (N, ZZ/pm ZZ)/p −→
∼ H 2 (N, Z
Z/pZZ)
and by (3.3.11) the surjectivity of the corestriction map
H 2 (N 0 , ZZ/pZZ) −→ H 2 (N, ZZ/pZZ) .
Thus we obtain for the dualizing module I of G
pI = p ( lim
−→
H 2 (N, ZZ/pm ZZ)∗ ) ∼
= lim
−→
H 2 (N, ZZ/pZZ)∗
m,cor ∗ cor ∗
and p I is isomorphic to ZZ/pZZ if and only if dimIFp H 2 (N, ZZ/pZZ) = 1 for every
open normal subgroup N of G. The result follows from (3.7.2). 2
Later, in §9, theorem (3.9.15), we will see that it is enough to consider only
open subgroups N of G with (G : N ) ≤ p.
The following theorem shows that the class of Poincaré groups is closed
under group extensions. It was first proved by A. PLETCH (see [167]) and later
(independently) by K. WINGBERG (see [253]).

(3.7.4) Theorem. Let

1 −→ H −→ G −→ G/H −→ 1
be an exact sequence of profinite groups such that
a) H i (U, ZZ/pZZ) is finite for all open subgroups U of H and all i ≥ 0,
b) cdp G/H < ∞.
Then if two of the three groups are Poincaré groups at p, so is the third.
Moreover, in this case we have:
(i) cdp G = cdp H + cdp G/H.
(ii) There is a canonical G-isomorphism I(G)∗ ∼
= I(H)∗ ⊗ ZZp I(G/H)∗ .
Remark: Theorem (3.7.4) can be generalized in several directions, cf. [167],

[253]. The most general version known to us is the following: we call a
profinite group virtual duality group at p if some open subgroup is a duality
group at p. Then, assuming a) above, G is a virtual duality group if and only
if H and G/H are. The (virtual) cohomological dimension and the dualizing
module of G can be computed from that of H and of G/H as in (i) and (ii)
(see [SW]).
Proof of (3.7.4) (see [253]): Let d = cdp G, m = cdp H, n = cdp G/H. By

our assumption that two of the groups are Poincaré groups, d, m and n are
finite, recalling that n < ∞, m ≤ d and d ≤ m + n. Because of a) we obtain
from (3.3.8)
d = m + n.
Let g run through the open normal subgroups of G and let h = g ∩ H. Then
g/h = gH/H runs through the open normal subgroups of G/H. For a G-
module A ∈ Modp (G), we consider the Hochschild-Serre spectral sequence
E(g, h, A) : E2ij (g, h, A) = H i (g/h, H j (h, A)) ⇒ H i+j (g, A).
If g 0 ⊆ g is another open normal subgroup of G, then the corestriction yields a
morphism
cor : E(g 0 , h0 , A) −→ E(g, h, A)
of spectral sequences, where h0 = h ∩ g 0 (see II §4, ex.3). The map
E2ij (g 0 , h0 , A) → E2ij (g, h, A)
is the composite of the maps
0 g 0 /h0
h cor g/h
i
H (g 0 0 »º j (h0 , A)) cor h
/h , H i 0 0 j
H (g /h , H (h, A)) H i (g/h, H j (h, A)),
and the map between the limit terms is the corestriction
0
cor gg : H i+j (g 0 , A) −→ H i+j (g, A).

For 2 ≤ r ≤ ∞ we set
r
Dij (G, A) = lim
−→
Erij (g, h, A)∗ .
g
0
If h runs through the open subgroups of H which are normal in G, then the
H j (h0 , A) and thus also lim 0 H j (h0 , A) are G/H-modules. As in the proof of
−→ h
(1.5.1), we see that
(∗) 2
Dij (G, A) = lim lim H i (g/h, H j (h0 , A))∗ ,
−→ −→
h0 g/h
where for both limits the transition maps are (induced by) cor ∗ .
By (3.4.6), G is a duality group at p of dimension d = n + m if and only if
Di+j (G, ZZ/pZZ) = lim H i+j (g, ZZ/pZZ)∗ = 0 for i + j =/ d ∗) ,
−→
g
and this is equivalent to
∞
(∗∗) Dij (G, ZZ/pZZ) = 0 for i + j =/ d.
Assume now that G/H is a Poincaré group at p. Then by (3.4.6)(i) the
open subgroups g/h are also Poincaré groups at p of the same dimension n
and the same dualizing module I(G/H). Furthermore, with condition a) it
follows by a standard argument that the groups H j (h0 , A) are finite for every
finite A ∈ Modp (G). Therefore we obtain by (3.4.6) (ii) (using (1.5.3)(iv) and
(1.5.1))
2
Dij (G, A) = lim
−→
lim
−→
H n−i (g/h, Hom (H j (h0 , A), I(G/H)))
g/h,res h0 ,cor ∗
H j (h0 , A), I(G/H))


 Hom (lim for i = n,
←−
(∗∗∗) = h0
0 otherwise,

 Hom (Dj (H, A)∗ , I(G/H))


for i = n,
=
0 otherwise.
Here Dj (H, A)∗ should be seen in the topological sense, i.e. as a compact
abelian group, and Hom (Dj (H, A)∗ , I(G/H)) are the continuous homomor-
phisms, i.e. the homomorphisms with finite image. From this, we deduce that
the following assertions are equivalent.
(1) H is a duality group at p,
(2) Dj (H, ZZ/pZZ) = 0 for j =/ m,
2
(3) Dij (G, ZZ/pZZ) = 0 for (i, j) =/ (n, m),
∞
(4) Dij (G, ZZ/pZZ) = 0 for (i, j) =/ (n, m),
(5) G is a duality group at p.
∗) D Z/pZZ) has here a different meaning from II §1.
i+j (G, Z

Thus we have proved the following
(3.7.5) Lemma. Suppose that assumptions a) and b) of (3.7.4) are fulfilled

and assume that G/H is a Poincaré group at p of dimension n.
Then H is a duality group at p of dimension m if and only if G is a duality
group at p of dimension m + n. In this case the dualizing module of H is
isomorphic to the dualizing module of G regarded as an H-module.
We proceed with the proof of (3.7.4) and assume now that G and H are
duality groups at p. Since the groups H m−j (h, ZZ/pZZ) are finite, and since
lim 0 H m−j (h0 , ZZ/pZZ) = 0 for j =/ m if h0 runs through the open subgroups h0
−→ h
of h, we find for every h an h0 ⊆ h such that
res : H m−j (h, ZZ/pZZ) −→ H m−j (h0 , ZZ/pZZ)
is the zero map for all j =/ m. Since h and h0 are duality groups at p of
dimension m (as H is), we have a commutative diagram of non-degenerate
pairings
H j (h, pÀ¿¾½¼ I(H)) × H m−j (h, ZZ/pZZ) ∪
ZZ/pZZ
cor res
∪
H j (h0 , p I(H)) × H m−j (h0 , ZZ/pZZ) ZZ/pZZ,
which shows that the left corestriction maps are zero. Therefore we see from
(∗) that
2
Dij (G, A) = 0, for all i and all j =/ m,
and consequently
∞ 2
Dim (G, p I(H)) = Dim (G, p I(H)), for all i.
If we assume, in addition, that H is a Poincaré group, it follows that
p I(H)
∼
= ZZ/pZZ
(as an abelian group), and there exists an open subgroup g of G which acts
trivially on p I(H). Therefore (∗∗) implies
2 2
Dim (G, p I(H)) = Dim (G, ZZ/pZZ) = 0, for all i =/ n.

Together with (∗), this implies

Di (G/H, ZZ/pZZ) = lim H i (g/h, ZZ/pZZ)∗
−→
g/h
= lim
−→
( lim
←−
H i (g/h, H 0 (h0 , ZZ/pZZ)∗ ))∗
g/h,cor ∗ h0 ,res ∗
= lim ( lim H i (g/h, H m (h0 , p I(H))))∗

−→ ←−
g/h,cor ∗ h0 ,cor
= lim lim H i (g/h, H m (h0 , p I(H)))∗

−→ −→
g/h,cor ∗ h0 ,cor ∗
2
= Dim (G, p I(H)) = 0.
We have thus shown that G/H is a duality group at p of dimension n.
The proof of (ii) goes as follows. As before let h0 run through all open
subgroups of H which are normal in G. Since m = cdp H, we have by (2.1.4),
H m+n (g, A) = H n (g/h, H m (h, A)),
and we obtain similarly as in (∗∗∗)
I(G) = lim lim H m+n (g, ZZ/pν ZZ)∗
−→ −→
ν g
= lim lim lim H n (g/h, H m (h0 , ZZ/pν ZZ))∗

−→ −→ −→
ν h0 g/h
= lim lim lim H 0 (g/h, Hom (H m (h0 , ZZ/pν ZZ), I(G/H)))

−→ −→ −→
ν h0 g/h,res
= lim
−→
Hom (lim
←−
H m (h0 , ZZ/pν ZZ), I(G/H))
ν h0
= lim Hom (Dm (H, ZZ/pν ZZ)∗ , I(G/H))

−→
ν
= Hom (I(H)∗ , I(G/H)) = (I(H)∗ ⊗ I(G/H)∗ )∗ .

ZZp
In particular, it follows that

rank ZZp I(G)∗ = rank ZZp I(H)∗ · rank ZZp I(G/H)∗ .
This completes the proof of the theorem. 2
Let us finally consider the case of a pro-p-group G. We set

H i (G) = H i (G, ZZ/pZZ)
and consider these groups as IFp -vector spaces.

(3.7.6) Proposition (SERRE). For an infinite pro-p-group G the following

statements are equivalent:
(i) G is a Poincaré group of dimension n.
(ii) dimIFp H i (G) < ∞ for all i ≤ n, H n (G) ∼
= IFp , and the cup-product yields
a non-degenerate pairing ∗ )
H i (G) × H n−i (G) −→ H n (G) ∼

= IFp for all 0 ≤ i ≤ n.
Proof: (i) ⇒ (ii). Let G be a Poincaré group of dimension n. Since

= Qp /ZZp , we have p I ∼
I ∼ = ZZ/pZZ as abelian groups, and consequently also
as G-modules, since G is a pro-p-group and #Aut(p I) = p − 1 is prime to p.
The G-module Hom(ZZ/pZZ, p I) is also isomorphic to ZZ/pZZ, hence the duality
isomorphism (3.4.6) (ii) yields an isomorphism
H i (G) ∼
= Hom (H n−i (G), IFp ) = H n−i (G)∗ ,
given by the cup-product
∪ tr
H i (G) × H n−i (G) −→ H n (G) −→ IFp .
For i = n, we obtain H n (G) ∼
= IFp and for i ≥ 0,
H i (G)∗∗ ∼
= H n−i (G)∗ ∼
= H i (G),
from which follows that H i (G) is finite. (The canonical homomorphism from
a vector space to its bidual is an isomorphism if and only if the dimension is
finite!)
(ii) ⇒ (i). Let A be any finite G-module such that pA = 0 and let 0 ≤ i ≤ n.
We claim that the pairing
∪
H i (G, A) × H n−i (G, A∗ ) −→ H n (G) ∼
= IFp
induces an isomorphism αiA : H i (G, A∗ ) → ∼ H n−i (G, A)∗ . This is true for
A = ZZ/pZZ by the assumption (ii). We first show that for an arbitrary A (with
pA = 0) αiA is surjective for i = 0, bijective for i = 1, . . . n − 1 and injective
for i = n. We proceed by induction on dimIFp A.
Since ZZ/pZZ is the only simple p-primary G-module, there is an exact
sequence of G-modules
0 −→ A0 −→ A −→ ZZ/pZZ −→ 0.
This, together with the dual sequence
0 −→(ZZ/pZZ)∗ −→ A∗ −→ A∗0 −→ 0 ,
yields an exact diagram for 0 ≤ i ≤ n
∗) A pairing A×B → Q/ZZ is non-degenerate if it induces injections A ,→ B ∗ and B ,→ A∗ .
Clearly, these are isomorphisms if A and B are finite.

H i−1ÍÌËÊÁÂÃÄÅÆÇÈÉ (A∗0 ) H i ((ZZ/pZZ)∗ ) H i (A∗ ) H i (A∗0 ) H i+1 ((ZZ/pZZ)∗ )

A ZZ /p ZZ A0 ZZ /p ZZ
0
αi−1 αi αA
i αi αi+1
H j+1 (A0 )∗ H j (ZZ/pZZ)∗ H j (A)∗ H j (A0 )∗ H j−1 (ZZ/pZZ)∗ ,
where i + j = n and H k (A) means H k (G, A), H k = 0 for negative k and the
map αk is the zero map for k < 0 and k > n. Now the induction step follows
by diagram chasing.
It remains to show duality in dimension i = 0 and i = n and by symmetry it

suffices to deal with the case i = 0. Let A be a finite G-module with pA = 0
and let U be an open subgroup with AU = A. Choose an open subgroup
V $ U (this is where we use that G has infinite order) strictly contained in U .
Consider the exact sequence
φ
0 −→ A−1 −→ Ā −→ A −→ 0,
where Ā := Map(G/U, A), φ is the map f 7→ g∈G/U f (g) and A−1 := ker φ
P
(compare with the remark following (1.3.8)). We claim that the map
H 0 (φ) : H 0 (G, Ā) → H 0 (G, A)
is zero. Indeed, via the identification Ā ∼
= IndUG (A) (see I §6), it corresponds
to the corestriction map cor : H (V, A) → H 0 (G, A), which factors through
0
cor : H 0 (V, A) → H 0 (U, A). Since U acts trivially on A, the last map is
multiplication by (U : V ), hence trivial as pA = 0.
We know that the functor H 0 (G, −) is coeffaceable (see II §2) on G-modules
annihilated by p. Now choose an injection A ,→ B into a G-module B with
pB = 0 such that the map H 0 (G, B ∗ ) → H 0 (G, A∗ ) is the zero map. Then we
get a commutative exact diagram
×ÖÕÏÐÑÒÓÔÎ ∗ )
H 0 (B 0
H 0 (A∗ ) H 1 ((B/A)∗ ) H 1 (B ∗ )
B/A αB
αB
0 αA
0 α1 1
H n (B)∗ H n (A)∗ H n−1 (B/A)∗ H n−1 (B)∗ .
The α1 ’s are bijective, hence α0A is injective. So we have proved duality for
modules annihilated by p and for i = 0, . . . , n.
We apply this result to A = ZZ/pZZ[G/U ], where U runs through the open
normal subgroups of G. By (3.4.1), Shapiro’s lemma and (1.6.5), we obtain
for 0 ≤ i < n

Di (ZZ/pZZ) = lim H i (G, ZZ/pZZ[G/U ])∗

−→
U
= lim
−→
H n−i (G, Hom (ZZ/pZZ[G/U ], ZZ/pZZ))
U
= lim H n−i (G, IndUG (ZZ/pZZ))

−→
U
= lim
−→
H n−i (U, ZZ/pZZ) = 0,
U,res
by (1.5.1).
Next we show cdp G ≤ n. Let x ∈ H n+1 (G, A), A ∈ M odp (G), pA = 0. By
(1.5.1), we have lim
−→
H n+1 (U, A) = 0, i.e. there exists an open subgroup U
U
of G such that x becomes zero in H n+1 (U, A) = H n+1 (G, IndUG (A)). From the
exact sequence
0 −→ A −→ IndUG (A) −→ B −→ 1,
it follows that there is an exact sequence
H n (G, IndUG (A)) −→ H n (G, B) −→ H n+1 (G, A) −→ H n+1 (G, IndUG (A)).
The functor H n (G, −) is right exact on G-modules annihilated by p since it is
dual to the functor H 0 (G, −) which is left exact. Therefore the last arrow is
injective, and x = 0. This proves cdp G ≤ n.
Thus we have shown that G is a duality group at p of dimension n. It remains
to determine the dualizing module I. By (3.4.7), I is a divisible p-torsion group
and therefore it suffices to show p I ∼
= ZZ/pZZ. By (3.4.7), p I ∼
= Dn (ZZ/pZZ) and
the same calculation as above for Di , i < n, shows
Dn (ZZ/pZZ) = lim H 0 (U, ZZ/pZZ) = ZZ/pZZ.
−→
U,res
This proves the proposition. 2
The pro-p-groups G which are Poincaré groups of dimension 2 are called

Demuškin groups. We give an explicit description and classification of them
in §9.
Exercise 1. Let U be an open subgroup of the profinite group G. If cdp G < ∞ then G is a
duality group at p of dimension n if and only if U is.
Exercise 2. If a p-Sylow subgroup Gp of a profinite group G is a duality group at p of
dimension n, then so is G. Is the converse true?
Exercise 3. Let G be a Poincaré group of dimension n > 0.
(i) If H is a proper closed subgroup of G, then the restriction map H n (G, ZZ/pZZ) −→
H n (H, ZZ/pZZ) is 0.
(ii) If H is a closed, but not open subgroup of G, then cd H ≤ n − 1.

§8. Filtrations
In this section G is always a pro-p-group. In the following we introduce the

notion of the q-central series of G.
(3.8.1) Definition. Let G be a pro-p-group and let q be a power of p. Then the

descending q-central series of G is the filtration {Gi }i≥1 recursively defined
by
G1 = G, Gi+1 = (Gi )q [Gi , G],
where [Gi , G] and (Gi )q are the closed subgroups topologically generated by
the commutators
(x, y) = x−1 y −1 xy,
x ∈ Gi , y ∈ G, and by the q-th powers of elements of Gi , respectively. ∗)
If q = 0, then we denote this series by {Gi }i≥1 , i.e.
G1 = G, Gi+1 = [Gi , G] .
It is called descending central series of G.
(3.8.2) Proposition. Let G be a pro-p-group. Then the subgroups Gi of the

q-central series are normal in G and Gi /Gi+1 is contained in the center of
G/Gi+1 . Furthermore,
Gi = 1 .
\
i
If G is finitely generated and q =/ 0, then the subgroups Gi form a fundamental
system of open neighbourhoods of 1.
Proof: The first two statements are obvious. For the third, let U be an open
normal subgroup of G. The projection G → G/U maps the q-central series
(Gi ) of G into a subseries of the q-central series of G/U , which terminates
with {1}. Therefore Gi ⊆ U for i sufficiently large.
Now assume that G is finitely generated and q =/ 0. We will show that
the subgroups Gi are of finite index, hence open, and proceed by induction
on i. For i = 1, this is trivial. Assume that (G : Gi ) is finite. Then Gi is
a finitely generated pro-p-group and Gi /(Gi )q [Gi , Gi ] is a finitely generated
abelian group of exponent q, thus finite, and has Gi /(Gi )q [Gi , G] = Gi /Gi+1
as quotient. This shows that (G : Gi+1 ) is finite, so Gi+1 is open in G. 2
We collect some formulae for commutators and p-powers of the pro-p-group G.

∗) The reader should take care about the different meaning of the superscripts i and q.

§8. Filtrations 221
(3.8.3) Proposition. If x ∈ Gi , y ∈ Gj , a ∈ q r ZZp , then

a
(i) (xy)a ≡ xa y a (y, x)( 2 ) mod Gi+j+max(1,r) ,
a
(ii) (xa , y) ≡ (x, y)a ((x, y), x)( 2 ) mod Gi+j+1+max(1,r) ,
a
(iii) (x, y a ) ≡ (x, y)a ((x, y), y)( 2 ) mod Gi+j+1+max(1,r) .
Proof: Recall the identities

(xy, z) = (x, z)((x, z), y)(y, z), (x, yz) = (x, z)(x, y)((x, y), z) (∗)
and the Hall-Witt identity (where xy = y −1 xy):
((x, y −1 ), z)y ((y, z −1 ), x)z ((z, x−1 ), z)x = 1.
Using these identities, we first observe that [Gi , Gj ] ⊆ Gi+j . For x ∈ Gi , y ∈ Gj
and a ∈ IN, we obtain
a
(xy)a ≡ xa y a (y, x)( 2 ) mod Gi+j+1 .
By passing to the limit, we obtain (i) for a ∈ q r ZZp , r = 0, 1.
We proceed by induction on r. Let r ≥ 2 and a = bq, b ∈ q r−1 ZZp . Then
b

a bq
(xy) = (xy) = (x y (y, x) b b 2 fr−1 )q with fr−1 ∈ Gi+j+r−1 .
By induction and using (∗), we obtain
b

q
(xy)a ≡ xa y a (y, x) 2 mod Gi+j+r .

b a
Since 2
q≡ 2
mod q r+1 , we obtain (i). Using this result, we get
a
y −1 xa y = (y −1 xy)a = (x(x, y))a ≡ xa (x, y)a ((x, y), x)( 2 ) mod Gi+j+1+max(1,r) ,
hence
a
(xa , y) = x−a y −1 xa y ≡ (x, y)a ((x, y), x)( 2 ) mod Gi+j+1+max(1,r) .
This shows (ii) and (iii) follows similarly. 2
(3.8.4) Corollary. For i, j ≥ 1 and r, s ≥ 0, where r + s =/ 0, we have

r s r+s r+s−1
[(Gi )q , (Gj )q ] ⊆ (Gi+j )q · (Gi+j+1 )q · Gi+j+r+s+1 .
The quotient Gi /Gi+1 is an abelian group, which we now write additively.

We denote this additive group by gri (G). The direct sum
∞
M
gr(G) = gri (G)
i=1
has the structure of a Lie algebra over k = ZZp /q ZZp . The Lie bracket [ , ] is
induced by the commutator, that is, if ξ = x̄ ∈ gri (G) and η = ȳ ∈ grj (G), then
[ξ, η] is the image of (x, y) = x−1 y −1 xy in gri+j (G). It is convenient to carry

out commutator calculations additively with respect to the filtration {Gi } in

the Lie algebra gr(G).
Proposition (3.8.3) shows that the map x 7→ xq of Gi into Gi+1 induces a
mapping πi : gri (G) → gri+1 (G). The family (πi ) then induces a map π∗ :
gr(G) → gr(G). Let π be an indeterminate over the ring k = ZZ/q ZZ if q =/ 0
and the zero element of k = ZZp if q = 0. Then there exists a unique mapping
Φ : k[π] × gr(G) −→ gr(G),
which is k-linear in the first variable, such that Φ(π i , ξ) = π∗i (ξ). If we set
α · ξ = Φ(α, ξ), we have π i · (π j · ξ) = π i+j · ξ. Proposition (3.8.3) with a = q
now yields the
(3.8.5) Proposition. Let ξ ∈ gri (G) and η ∈ grj (G). Then

(i) π · (ξ + η) = π · ξ + π · η if i = j > 1,

q
(ii) π · (ξ + η) = π · ξ + π · η + 2
[ξ, η] if i = j = 1,
(iii) π · [ξ, η] = [π · ξ, η] if i =/ 1,
π · [ξ, η] = [ξ, π · η] if j =/ 1,

q
(iv) [π · ξ, η] = π · [ξ, η] + 2
[[ξ, η], ξ] if i = j = 1,

q
(v) [ξ, π · η] = π · [ξ, η] + 2
[[ξ, η], η] if i = j = 1.
Remark:
If G = F is a free pro-p-group and if q is not a power of 2, then
q
2
≡ 0 mod q and gr(F ) is a free Lie algebra over k[π] (see [125]).
For a free pro-p-group F , we obtain the following formula for the intersection
of the p-central series and the central series of F :
(3.8.6) Proposition. Let F be a free pro-p-group and let {F i } be the p-central

series and {Fi } the central series of F . Then for all i ≥ j ≥ 1, there is the
equality
i−j i−j−1
F i ∩ Fj = (Fj )p · (Fj+1 )p · . . . · Fi .
Proof: We proceed by induction on j. For j = 1 we have to prove that

i−1
F i = F p · . . . · Fi .
First, we show that
i−1
F i = F p · . . . · Fi · F i+1 .
This is obviously true for i = 1. So let i > 1 and assume that the assertion is
true for i − 1. Then, using (3.8.4),

§8. Filtrations 223
F i = (F i−1 )p [F i−1 , F ]
i−2 i−2
= (F p · . . . · Fi−1 · F i )p [F p · . . . · Fi−1 · F i , F ]
i−1 p
= F p · . . . · Fi−1 · Fi · F i+1 .
i−1
Assume we have proved that F i = F p · . . . · Fi · F k for k ≥ i + 1. Then
i−1
F i = F p · . . . · Fi · (F k−1 )p [F k−1 , F ]
i−1 k−2 k−2
= F p · . . . · Fi · (F p · . . . · Fk−1 · F k )p [F p · . . . · Fk−1 · F k , F ]
i−1
= F p · . . . · Fi · F k+1 .
Since k F k = 1, we have proved the case j = 1.
T
Now let i ≥ j > 1. Then

i−j+1
F i ∩ Fj = F i ∩ Fj−1 ∩ Fj = (Fj−1 )p · . . . · Fi ∩ Fj
by the induction assumption. Let
pi−j+1 i−j
x = zj−1 y ∈ Fj with zj−1 ∈ Fj−1 , y ∈ (Fj )p · . . . · Fi .
Claim: Fj−1 /Fj is ZZp -torsion-free.
Proof: Using a limit argument we may assume that F is finitely generated.
Since the free pro-p-group F is the completion of a free discrete group, Fj /Fj+1
is the completion of a free (finitely generated) abelian group by [260], Satz 4.
Hence it is a free ZZp -module.
i−j
Using this claim, we see zj−1 ∈ Fj , so that x ∈ (Fj )p · . . . · Fi . This proves
the proposition. 2
We now define a refinement {G(i,j) } of the descending p-central series of G.
(3.8.7) Definition. Let {Gi } and {Gj } be the p-central series and the central
series of the pro-p-group G, respectively. We set, for i, j ≥ 1,
G(i,j) := (Gi ∩ Gj ) Gi+1 .
Obviously we have
G(i,1) = Gi and G(i,j) = Gi+1 for j > i ≥ 1.
We introduce the following notational convention:
The letter ν always stands for a pair (i, j), i ≥ j ≥ 1, and we order these
pairs lexicographically. We say that
ν + 1 = (i, j + 1) if i > j ,
ν + 1 = (i + 1, 1) if ν = (i, i) .
The descending chain {G(ν) } of normal characteristic subgroups is a refinement

of the descending p-central series. In particular, G(ν) /G(ν+1) is an IFp -vector
space for all ν.

(3.8.8) Proposition. For every ν = (i, j), the IFp -vector space homomorphism
ψν : (G/G2 )⊗j −→ G(ν) /G(ν+1)
i−j
x̄1 ⊗ · · · ⊗ x̄j 7−→ ([x1 , [x2 , [· · · , xj ]] · · ·])p mod G(ν+1)
is well-defined and surjective.
Proof: Let S = {xα }α∈I be a minimal system of generators of G. Then their

image S̄ := S mod Gp [G, G] in G/G2 = G/Gp [G, G] is an IFp -basis of G/G2 .
(In fact, if R is any proper subset of S, then R generates a closed subgroup
=/ G, which sits in a maximal, i.e. normal, subgroup M of index p, so that
R̄ ⊆ M/G2 =/ G/G2 cannot generate G/G2 .) Thus the tensors x̄α1 ⊗ · · · ⊗ x̄αj ,
α1 , . . . , αj ∈ I define a basis of (G/G2 )⊗j . Recalling the definition of the (ν)-
filtration, it follows from (3.8.6) (for free groups and then for all pro-p-groups)
that the elements
i−j
([xα1 , [xα2 , [· · · , xαj ]] · · ·])p , α1 , . . . , αj ∈ I
generate G(ν) modulo G(ν+1) . Finally, an inductive application of (3.8.3) shows
that ψν is well-defined. 2
Remark: The same proof shows that we can define ψν as a homomorphism

(G/G2 )⊗j −→ G(ν) /Gi+1 ⊆ Gi /Gi+1

p
if either j > 1 or if p is odd (hence p | 2
).
§9. Generators and Relations
For this entire section G is a pro-p-group. We set

H n (G) = H n (G, ZZ/pZZ)
and regard these groups as IFp -vector spaces. By the dimension, dim H n (G),
we mean the cardinality of a basis. H 1 (G) is the Pontryagin dual of the group
G/G∗ , where G∗ = Gp [G, G], i.e. the closure of the subgroup generated by
commutators and p-th powers.∗) Indeed, the group G/G∗ is the largest profinite
abelian quotient of exponent p of G, so that
H 1 (G) = Hom(G, ZZ/pZZ) = Hom(G/G∗ , Q/ZZ).
A generator system of G is a convergent family S = (gi )i∈I , gi ∈ G, which
generates G as a topological group. By convergent we mean convergent to 1,
∗) G∗ is the Frattini subgroup of G, i.e. the intersection of all maximal closed subgroups.

§9. Generators and Relations 225
i.e. every open subgroup of G contains gi for almost all i. The rank of a pro-p-
group G is the infimum over the cardinalities card(S ) = card(I) of generator
systems S of G, and is denoted by d(G). A generator system is minimal if
no proper subfamily is a generator system.
(3.9.1) Proposition. A convergent family S is a generator system of G if and

only if the family S = (g i )i∈I of residue classes g i = gi mod G∗ generates
G/G∗ . S is minimal if and only if S is. Every minimal system of generators
has cardinality d(G) and we have the equality
d(G) = dim H 1 (G).
In particular, G is finitely generated if and only if H 1 (G) is finite.
The first assertion of the proposition is often called Frattini argument.
Proof: Let H be the closed subgroup of G generated by S . Then we have

the equivalences
the inclusion H → G is surjective

⇔ H 1 (G) → H 1 (H) is injective (by (1.6.14)(ii))
⇔ H/H ∗ → G/G∗ is surjective (by Pontryagin duality).
This shows the statement. 2
Let N be a normal subgroup of G. A generator system of N as a normal

subgroup is a convergent family S such that N is the smallest closed normal
subgroup of G containing all elements of S . In the case N = G, (3.9.1) shows
that S generates G as a normal subgroup (of itself) if and only if S generates
G as a pro-p-group.
(3.9.2) Proposition. Let N be a normal closed subgroup of G, S a convergent

family of elements of N and H the closed subgroup generated by S . Then
the restriction map
res : H 1 (N )G −→ H 1 (H)
is injective if and only if S is a generator system of N as a normal subgroup.
Proof: Let N 0 be the normal closed subgroup of G generated by S = (ni )i∈I

(as a normal subgroup), and consider the commutative diagram

ÙØÚ )G
H 1 (N
res1
H 1 (H)
res2 res3
1 0 G
H (N ) .
As a pro-p-group, N 0 is generated by the family S˜ = (σni σ −1 )i∈I,σ∈G . For
χ ∈ H 1 (N 0 )G , we have
res3 (χ) = 0 ⇒ χ(σni σ −1 ) = χσ (ni ) = χ(ni ) = 0 for all σ and all i
⇒ χ(S˜) = 0 ⇒ χ = 0,
i.e. res3 is injective. Therefore res1 is injective if and only if res2 is in-
jective. But this is equivalent with the injectivity of the homomorphism
0
res N 1 1
N 0 : H (N ) → H (N ) since
ker(res2 ) = ker(res N G N
N 0 ) = 0 ⇐⇒ ker(res N 0 ) = 0
0
by (1.6.12). Finally, the injectivity of res N
N 0 is equivalent with N = N by
(1.6.14). 2
(3.9.3) Corollary. Let N be a normal closed subgroup of G. Then a convergent

family S of elements of N generates N as a normal subgroup if and only if
the family S of residue classes modulo N p [G, N ] generates N/N p [G, N ]. In
particular, S is minimal if and only if S is, and in this case
card(S ) = dim H 1 (N )G .
Proof: Let H be the closed subgroup in N generated by S . Taking into

account that N/N p [G, N ] = (H 1 (N )G )∨ , we the have following equivalences
S generates N as a normal subgroup
⇔ H 1 (N )G → H 1 (H) is injective (by (3.9.2))
∗ p
⇔ H/H → N/N [G, N ] is surjective (by Pontryagin duality).
Let S = (gi )i∈I be a system of generators of the pro-p-group G and let F

be the free pro-p-group with basis (xi )i∈I , see (3.5.14). Sending xi to gi , we
obtain an exact sequence
1 −→ R −→ F −→ G −→ 1.
A relation system, also called a system of defining relations with respect

to S , is a generator system R of R as a normal subgroup of F .

(3.9.4) Proposition. Suppose that the pro-p-group G is finitely generated and

let S be a finite system of generators of G. Then a finite system R of defining
relations with respect to S exists if and only if H 2 (G) is finite. If R is finite
and minimal, we have the equality
card(S ) − card(R) = dim H 1 (G) − dim H 2 (G).
Proof: Applying (3.9.3) to the closed subgroup R of F , we obtain

card(R) = dim H 1 (R)F = dim H 1 (R)G .
Consider the five term exact sequence
0 −→ H 1 (G) −→ H 1 (F ) −→ H 1 (R)G −→ H 2 (G) −→ H 2 (F ).
Since F is free, H 2 (F ) = 0, and we obtain the result by counting IFp -dimen-
sions. 2
In particular, given G, the cardinality of a minimal system of defining

relations only depends on the cardinality of the system S of generators. If S
is a minimal system of generators, we call the cardinality of a minimal relation
system the relation rank of G. We denote it by r(G).
(3.9.5) Proposition. Let G be a pro-p-group. Then the relation rank of G

satisfies the formula
r(G) = dim H 2 (G).
In particular, G is a free pro-p-group if and only if cd G ≤ 1.
Proof: Let S be a minimal system of generators of G. Then the exact se-

quence ∼ H 1 (F ) −→ H 1 (R)G −→ H 2 (G) −→ 0
0 −→ H 1 (G) −→
∼ H 2 (G), and so r(G) = dim H 1 (R)G =
yields the isomorphism H 1 (R)G →
dim H 2 (G). 2
From (3.3.16) we obtain the
(3.9.6) Corollary. Let G be a finitely generated pro-p-group of rank d(G) and

let U be a cofinal set of open subgroups of G. Then G is a free pro-p-group if
and only if
d(U ) − 1 = (G : U )(d(G) − 1) for all U ∈ U.
Remark: If G is a finitely generated pro-p-group, i.e. d(G) < ∞, then the

relation rank r(G) need not be finite. We have the following example occurring

in number theory. Let G = G(k(p)|k) be the Galois group of the maximal p-

extension of a p-adic local field k (cf. VII §5) and let Gt , t ∈ IR, t ≥ 0, be
the higher ramification groups of G (see [130] for the definition). Since G is
finitely generated (cf. (7.5.11)), the quotients G/Gt are also finitely generated.
But N. L. GORDEEV proved in [58] that r(G/Gt ) is infinite if t > 1. This
extended the result of E. MAUS who showed this for t ∈/ ZZ[ p1 ], t > 1, see [131].
The rank and the relation rank of a pro-p-group G played an important role
in the solution of the famous and long standing class field tower problem in
number theory. The question is whether the maximal unramified p-extension
K∅ (p)|K of a number field K need always be of finite degree. It was a
great surprise when, in 1964, the Russian mathematicians E. S. GOLOD and
I. R. ŠAFAREVIČ proved that this is not always the case. The Galois group G
of K∅ (p)|K is a pro-p-group, and the crucial point in the proof (presented in
X §10) was the discovery that for a finite p-group G the relation rank has to be
very large in comparison with the rank. More precisely, we have the
(3.9.7) Theorem. If G is a finite p-group, then
r(G) > 1
4
d(G)2 .
For the proof we need the following
(3.9.8) Lemma. Let G be a finite p-group and Λ = IFp [G]. For every finite
G-module A such that pA = 0, there is a resolution
∂ ∂ ∂
0 −→ A −→ Λb0 −→ Λb1 −→ Λb2 −→ · · · ,
where bn = dim H n (G, A), and ∂((Λbn )G ) = 0.
Proof: We have canonically ΛG ∼ = IFp . Let a1 , . . . , ab0 be a basis of the

IFp -vector space A . Then the isomorphism AG →(Λb0 )G = IFbp0 , ai 7→ ei ,
G
extends to an injective G-homomorphism

j : A −→ Λb0 .
Indeed, the map HomG (A, Λb0 ) → HomG (AG , Λb0 ) is surjective, since in the
exact sequence
0 −→ Hom(A/AG , Λb0 ) −→ Hom(A, Λb0 ) −→ Hom(AG , Λb0 ) −→ 0
of induced G-modules, H 1 of the first term is zero. The extension j is automat-
ically injective, since from ker(j|AG ) = ker(j)G = 0, it follows that ker(j) = 0
by (1.6.12). Since Λb0 is an induced G-module, from the exact sequence
0 −→ A −→ Λb0 −→ B −→ 0 ,

we obtain isomorphisms
H i (G, B) ∼
= H i+1 (G, A)
for i ≥ 1. The same holds for i = 0, since in the exact sequence
AG −→(Λb0 )G −→ B G −→ H 1 (G, A) −→ 0
the first arrow is bijective, i.e. (Λb0 )G is mapped to zero.
Proceeding in the same way with the G-module B in place of A and noting
that dim H 0 (G, B) = dim H 1 (G, A) = b1 , we obtain an exact sequence
0 −→ B −→ Λb1 −→ C −→ 0
such that B G →(Λb1 )G is an isomorphism, i.e. (Λb1 )G is mapped to zero. If we
define ∂ : Λb0 → Λb1 to be the composite map Λb0 → B → Λb1 , then ∂((Λb0 )G ) =
0. Continuing this process, the lemma follows by induction. 2
Proof of theorem (3.9.7): For every finite G-module A such that pA = 0, we

define the “ascending central series”
0 = A0 ⊆ A1 ⊆ A2 ⊆ · · · ⊆ Am = A
by A0 = 0, A1 = AG and An+1 /An = (A/An )G for n ≥ 1. By (1.6.12),
An =/ An+1 , unless An = A. If h : A → B is an injective G-homomorphism,
then we see inductively
An = h−1 (Bn ).
We set cn (A) = dim(An+1 /An ) and form the Poincaré polynomial
cn (A)tn .
X
PA (t) =
n≥0
If 0 < t < 1 is a real variable, then
1
sn (A)tn ,
X
PA (t) =
1 − t n≥0
n
X
where sn (A) = ci (A) = dim(An+1 ).
i=0
We apply the lemma to the G-module A = IFp . Recalling that IFp = ΛG = Λ1 ,
we obtain an exact sequence
∂ ∂
(∗) 0 −→ E −→ D −→ R,
where E = Λ/Λ1 , D = Λd , R = Λr ,
d = dim H 1 (G) = d(G), r = dim H 2 (G) = r(G),
such that ∂(D1 ) = R0 = 0. From this it follows inductively that ∂(Dn ) ⊆ Rn−1 ,
and we obtain the sequences
(∗∗) 0 −→ En −→ Dn −→ Rn−1 , n ≥ 1.

These sequences are again exact, since (∗) is exact and En = ∂ −1 (Dn ). Now
consider the Poincaré polynomials associated to E, D, R. Setting P (t) = PΛ (t),
we obtain
P (t) − 1
PE (t) = , PD (t) = dP (t), PR (t) = rP (t) .
t
From (∗∗) we get the inequalities
sn (D) ≤ sn (E) + sn−1 (R),
(where s−1 (R) = 0), so that
1 1 t
PD (t) ≤ PE (t) + PR (t) .
1−t 1−t 1−t
It follows that
P (t) − 1
dP (t) ≤ + rtP (t) for 0 < t < 1,
t
and so
1 ≤ P (t)(rt2 − dt + 1) if 0 < t < 1.
Since P (t) has positive coefficients, we obtain
0 < rt2 − dt + 1 if 0 < t < 1.
Substituting t 7→ 2rd , we get
r > 41 d2 ,
as asserted. This substitution is valid, since the exact sequence
p
0 → H 1 (G) → H 2 (G, ZZ) → H 2 (G, ZZ) → H 2 (G)
shows that d ≤ r < 2r, i.e. 0 < 2rd < 1. 2
Remarks: 1. GOLOD and ŠAFAREVIČ had originally proved only that r(G) >
1
4
(d(G) − 1)2 (see [57]). The sharper inequality r(G) > 14 d(G)2 was obtained
independently by W. GASCHÜTZ and E. B. VINBERG [241]. The proof given
here is dual to the proof presented by P. ROQUETTE in [183], who uses homology
instead of cohomology. There is another more general proof (based, however,
on the same idea as in [183]) with more far reaching results given by H. KOCH
in [110] and [72], and yet another proof was given by J.-P. SERRE in [210].
2. We want to present the result of KOCH without proof. For that we need
the notion of the Zassenhaus filtration of a pro-p-group. Let G be a finitely
generated pro-p-group and for n ≥ 1 let the ideal I n (G) of IFp [[G]] be the n-th
power of the augmentation ideal I(G). The filtration
G(n) = {g | g − 1 ∈ I n (G)}, n ≥ 1,
is called the Zassenhaus filtration of G. The normal subgroups G(n) form a
full system of neighbourhoods of the identity of G. For the basic properties of
{G(n) }n≥1 , see [110], §7.4. We mention only that
(G(n) )p ⊆ G(np) and [G(n) , G] ⊆ G(n+1) .

Now let
1 −→ R −→ F −→ G −→ 1
be a minimal presentation of the finite p-group G by a free pro-p-group F and
assume that R ⊂ F(m) for some m, where {F(n) }n≥1 is the Zassenhaus filtration
of F . Then
d(G)m
r(G) > m
(m − 1)m−1 .
m
This formula shows that for a finite p-group we get a better bound for r(G) if
we have information on the complexity of the relations.
We are now aiming at the classification of the Demuškin groups, which are
defined as follows.
(3.9.9) Definition. A pro-p-group G is called a Demuškin group if its coho-

mology H i (G) has the following properties:
(i) dimIFp H 1 (G) < ∞,
(ii) dimIFp H 2 (G) = 1,
(iii) the cup-product H 1 (G) × H 1 (G) → H 2 (G) is non-degenerate.
By (3.7.6), an infinite Demuškin group has necessarily cohomological di-

mension 2 and is precisely a Poincaré group of dimension 2. The finite
Demuškin groups are classified by the following
(3.9.10) Proposition. The group G = ZZ/2ZZ is the only finite Demuškin

group.
Proof: The relation rank r = r(G) of a Demuškin group G is equal to 1

by definition. Therefore if G has generator rank d = d(G) ≥ 2, then its
abelianization Gab has a ZZp -rank of at least d − 1 ≥ 1. We conclude that
d = 1 for a finite Demuškin group, i.e. G is cyclic. Hence H 1 (G) is a one
dimensional IFp -vector space and for every x ∈ H 1 (G) we have 2 · (x ∪ x) =
x ∪ x + x ∪ x = 0 by the anti-symmetry of the cup-product. This gives a
contradiction for odd p ∗) and therefore G ∼ = ZZ/2k ZZ for some k ≥ 1. A
∗) In other words: the cup-product is always anti-symmetric, which implies in odd charac-
teristics that it is alternating. A vector space with a non-degenerate alternating bilinear form
is necessarily of even dimension. In characteristic 2 the cup-product may induce a symmetric,
non alternating form.

straightforward computation, which we leave to the reader, shows that the cup-
∪
product H 1 (G) × H 1 (G) −→ H 2 (G) ∼
= IF2 is trivial for k > 1, and nontrivial,
hence non-degenerate, for k = 1. 2
The appearance of G = ZZ/2ZZ as the only finite Demuškin group is interest-

ing from the arithmetic point of view, since G is the absolute Galois group of
the field IR.
Let us consider more generally finitely generated pro-p-groups G with only

one defining relation, i.e.
n = dimIFp H 1 (G) < ∞ and dimIFp H 2 (G) = 1.
Such a group is called a one-relator pro-p-group. We have an exact sequence
involving G,
1 −→ R −→ F −→ G −→ 1,
where F is the free pro-p-group of rank n and R is generated by one element
ρ as normal subgroup of F . We also write R = (ρ). Passing to the abelianized
groups, we obtain Gab as a quotient of F ab ∼ = ZZnp by a subgroup which is
either isomorphic to ZZp or zero. Noting that H 1 (G) = H 1 (Gab ) = (ZZ/pZZ)n ,
we conclude that
Gab ∼
= ZZn p or Gab ∼
= ZZ/pf ZZ × ZZn−1 (f ≥ 1).
p
f
We set q = p in the second case and q = 0 in the first case. The numbers n
and q are invariants of the group G. Since F/F q [F, F ] → G/Gq [G, G] is an
isomorphism by definition of q, we have
R ⊆ F q [F, F ] = F 2 ,
where F i is the q-central series of F .
The central result which we want to prove is the following
(3.9.11) Theorem (DEMUŠKIN). Let G be a one-relator pro-p-group. Suppose

that the invariant q of G is =/ 2. Then G is a Demuškin group if and only if it
is isomorphic to the pro-p-group defined by n generators x1 , . . . , xn subject to
the one relation
xq1 (x1 , x2 )(x3 , x4 ) · · · (xn−1 , xn ) = 1,
where (x, y) is the commutator x−1 y −1 xy. In particular, G is then determined
by the two invariants n and q.
The proof of the theorem is not easy. We follow in essence the presentation
in [117] by J. LABUTE, where the case q = 2 is also treated.

Before considering one-relator pro-p-groups and proving the result stated

above, we consider the case where G is an arbitrary finitely generated pro-
p-group of rank n. If Gab has a nontrivial torsion subgroup, then we set
k = ZZp /q ZZp , where q is the smallest elementary divisor of Gab , i.e. q = pf
is the maximal p-power such that Gab /q is a free ZZp /q ZZp -module. When
Gab ∼ = ZZnp we set k = ZZp and change our usual notation and denote by
i
H (G, ZZp ) the continuous cochain cohomology (see II §7). Since G is finitely
generated, (2.7.6) implies H i (G, ZZp ) = lim H i (G, ZZ/pν ZZ) for i = 0, 1, 2.
←− ν
Let
1 −→ R −→ F −→ G −→ 1
be a minimal presentation of the pro-p-group G. The inflation
inf : H 1 (G, k) −→ H 1 (F, k)
is an isomorphism with which we identify the two groups.
Since F is free, we have H 2 (F, k) = 0, and the five term exact sequence
(1.6.7) shows that also the transgression tg : H 1 (R, k)G → H 2 (G, k) is an
isomorphism. More generally, we obtain isomorphisms
ab
tg : Homcts (RG , ZZ/pn ZZ) = H 1 (R, ZZ/pn ZZ)G −→ Z/pn ZZ)
∼ H 2 (G, Z
for all n ≤ f . Therefore every element ρ ∈ R gives rise to trace maps

tr = trρ : H 2 (G, ZZ/pn ZZ) −→ ZZ/pn ZZ
which are defined by ϕ 7→ (tg−1 ϕ)(ρ). If q = 0, then these maps are defined
for all n ≥ 1 and also with ZZp -coefficients.

1 −→ R −→ F −→ G −→ 1
be a minimal presentation of the finitely generated pro-p-group G.
(i) Every basis x1 , . . . , xn of F defines a k-basis χ1 , . . . , χn of the k-module
H 1 (F, k) = H 1 (G, k) such that χi (xj ) = δij .
(ii) Given an element ρ ∈ R, the bilinear form induced by the cup-product
∪ trρ
H 1 (G, k) × H 1 (G, k) −→ H 2 (G, k) −→ k
is non-degenerate if and only if it is non-degenerate with k replaced by
ZZ/pZZ, and in this case trρ is surjective.
(iii) If G is a one-relator group and if ρ generates R as a normal subgroup,
then trρ is injective.
Proof: (i) By definition of q, a minimal generator system x1 , . . . , xn of F

defines an isomorphism F/F q [F, F ] ∼
= k n , where [F, F ] is the closure of the
commutator subgroup of F . (i) follows from this.

(ii) Consider the commutative diagram

âÛáÜÝÞßà k)
H 1 (G, × H 1 (G, k) ∪
H 2 (G, k)
trρ
k
∪ trρ
H 1 (G, ZZ/pZZ) × H 1 (G, ZZ/pZZ) H 2 (G, ZZ/pZZ) ZZ/pZZ ,
where the vertical arrows are induced by the reduction map k → ZZ/pZZ. If B
is a matrix for the upper bilinear form, then B = B mod p is a matrix for the
lower one. The upper one is non-degenerate if and only if B is invertible over
k, i.e. det (B) ∈ k × , and this is the case if and only if det(B) =/ 0, i.e. if the
lower one is non-degenerate. Clearly trρ must be surjective in this case.
ab
(iii) If ρ generates R as a normal subgroup in G, then it generates RG . Therefore
−1
if tg (ϕ)(ρ) = 0, then ϕ = 0, so that trρ is injective. 2
Let R = {ρi | i ∈ I} be a minimal system of defining relations of the finitely

generated pro-p-group G = F/R.

1 −→ R −→ F −→ G −→ 1
be a minimal presentation of the finitely generated pro-p-group G. Let
x1 , . . . , xn be a basis of F and let χ1 , . . . , χn be the corresponding k-basis
of H 1 (F, k) = Hom(F, k) = Hom(G, k), i.e. χi (xj ) = δij .
(i) Every element ρ ∈ F 2 has a representation
n
qaj
(xk , xl )akl · ρ0 , ρ0
Y Y
ρ= xj · ∈ F 3 , aj , akl ∈ k.
j=1 1≤k<l≤n
The akl are uniquely determined by ρ, and if q =/ 0 so are the aj .

(ii) If R = {ρi | i ∈ I} is a minimal system of defining relations of G and
n
qaij i
(xk , xl )akl · ρ0i , ρ0i F 3 , aij , aikl
Y Y
ρi = xj · ∈ ∈ k, i ∈ I ,
j=1 1≤k<l≤n
then the bilinear form

∪ trρ
H 1 (G, k) × H 1 (G, k) −→ H 2 (G, k) −→
i
k
is given by the matrix Bi = (bikl ) with respect to the basis χ1 , . . . , χn ,
where
−aikl if k < l,


 
bikl = trρi (χk ∪ χl ) =  ailk if k > l,

q
aik

− 2
if k = l.

Proof: The existence of a representation (i) for a finite word in the letters
x1 , . . . , xn is obtained by a simple collecting process (see M. HALL [73], chap.
11.1). For an arbitrary ρ ∈ F 2 it follows by a limit process, noting that the
discrete free group generated by the x1 , . . . , xn is a dense subgroup of F . If
q =/ 0, then k = ZZ/q ZZ and
2
F/F q [F, F ] = (ZZ/q 2 ZZ)n .
The image of ρ in this last group is (qa1 , . . . , qan ), so that the aj are uniquely
determined in k by ρ. The uniqueness of the akl follows from the proof of
(ii) The cohomology class χk ∪ χl ∈ H 2 (G, k) is represented by the inhomo-
geneous 2-cocycle
c0 (σ, τ ) = χk (σ)χl (τ ).
Let c be the inflation of c0 to F . Since H 2 (F, k) = 0, there exists an inhomo-
geneous cochain
u = ukl : F −→ k
such that c = ∂u and, moreover, by subtracting the homomorphism h : F → k,
h(xj ) = u(xj ), we can suppose u(xj ) = 0, j = 1, . . . , n. Then
(1) u(xy) = u(x) + u(y) − χk (x)χl (y), x, y ∈ F.
In particular, u(xy) = u(x) + u(y) whenever x or y is contained in F 2 =
F q [F, F ]. Therefore u is a homomorphism on F 2 which vanishes on F 3 , since
F 3 is topologically generated by the products xq (x, y), x ∈ F 2 , y ∈ F , on
which u behaves multiplicatively.
The restriction v of u to R is an element of H 1 (R, k)G since (y, x) ∈ F 3 for
y ∈ R and x ∈ F , and so
v(x−1 yx) = u(y(y, x)) = u(y) + u((y, x)) = v(y).
By the definition (1.6.6) of the transgression, we now obtain
tg(v) = [∂u] = χk ∪ χl ,
and by definition of the matrix Bi = (bikl ) and of trρi , we get
bikl = trρi (χk ∪ χl ) = v(ρi ) = ukl (ρi ).
Since Bi is anti-symmetric, it therefore remains to show that
−aikl if k < l,
(
(2) u(ρi ) =
− q2 aik if k = l.
We compute the values u(xm j ) and u((xν , xµ )) by means of (1).
−1
m+1
If k =/ l, we have u(xj ) = u(xm m
j ) and u(xj ) = 0, which implies u(xj ) = 0
for any m ∈ ZZ. If k = l, we have
u(xm+1
j ) = u(xm m m
j ) − χk (xj )χl (xj ) = u(xj ) − mδkj .
This implies
m
u(xm
j ) = − 2
δkj for m = 1, 2, 3, . . .

Furthermore, noting that u(x−1 ) + u(x) + χk (x)χl (x) = 0, we have for ν < µ
u((xν , xµ )) = u(x−1 −1
ν ) + u(xµ xν xµ ) + χk (xν )χl (xν )
= −δkν + u(x−1
µ xν xµ ) + δkν δlν
= u(x−1
µ ) + u(xν xµ ) + χk (xµ )χl (xν ) + χk (xµ )χl (xµ )
= −δkν δlµ + u(xν ) + u(xµ ) − χk (xν )χl (xµ ) + δkµ δlν

= δkµ δlν − δkν δlµ ,
i.e. for ν < µ and k < l
−1 if k = ν, l = µ,
(
u((xν , xµ )) =
0 otherwise.
Now applying the homomorphism u : F /F 3 → k to
2
n
qaij i
(xν , xµ )aνµ mod F 3
Y Y
ρi ≡ xj
j=1 1≤ν<µ≤n
yields the desired result (2) and also the uniqueness of the aikl . 2
If q =/ 0 and k = ZZ/q ZZ, we define the Bockstein homomorphism

B : H 1 (G, k) −→ H 2 (G, k)
as the connecting homomorphism in the long exact cohomology sequence
associated to q
0 −→ ZZ/q ZZ −→ ZZ/q 2 ZZ −→ ZZ/q ZZ −→ 0 .
We get a second interpretation of the exponents aij in the expression for the
relations of the group G in (3.9.13) (ii):
(3.9.14) Proposition. With the notation of (3.9.13) and assuming that q =/ 0,

we have
trρi (B(χj )) = −aij .
We leave the proof to the reader, since it follows the same lines as the proof
of (3.9.13).
Before we determine explicitly the defining relation of a Demuškin group, we
first sharpen the result (3.7.3) concerning the characterization of a Demuškin
group by properties of its subgroups.
(3.9.15) Theorem. Let G be a finitely generated one-relator pro-p-group of

rank d(G) > 1. Then the following assertions are equivalent:
(i) G is a Demuškin group.

(ii) cdp G = 2 and the dualizing module of G is isomorphic to Qp /ZZp as an

abelian group.
(iii) cdp G = 2 and dimIFp H 2 (N, ZZ/pZZ) = 1 for every open subgroup N
of G.
(iv) cdp G = 2 and d(N ) − 2 = (G : N )(d(G) − 2) for every open subgroup
N of G.
(v) cdp G = 2 and dimIFp H 2 (N, ZZ/pZZ) = 1 for every open subgroup N of G
with (G : N ) = p.
(vi) d(N ) − 2 = (G : N )(d(G) − 2) for every open subgroup N of G with
(G : N ) = p.
Remark: Observe that in assertion (vi) we do not assume that cdp G = 2. The
equivalences between (i), (v) and (vi) were first proved by I. V. ANDOŽSKII
and later independently by J. DUMMIT and J. LABUTE (without the assertion
cdp G = 2 in (v)) - see [3] and [44].
Proof: The equivalences between (i), (ii) and (iii) follow from (3.7.2) and
(3.7.3). Using (3.3.13), we get the implications (iii) ⇒ (iv) and (v) ⇒ (vi). The
assertions (v) and (vi) are trivial consequences of (iii) and (iv), respectively.
So let us assume that (vi) holds, and try to prove (i).
Let
Gab ∼= ZZp /q ZZp × ZZd(G)−1
p ,
where q is equal 0 if Gab is torsion-free. Furthermore, let
1 −→ R −→ F −→ G −→ 1
be a minimal presentation of G by a free pro-p-group F of rank d = d(G).
Suppose that the cup-product pairing
H 1 (G, ZZ/pZZ) × H 1 (G, ZZ/pZZ) → H 2 (G, ZZ/pZZ) ∼
= ZZ/pZZ
is degenerate, and let χ1 be an element of the radical of this pairing. Extend
this element to a basis χ1 , . . . , χd of H 1 (G, ZZ/pZZ) ∼ = H 1 (F, ZZ/pZZ) and let
x̄1 , . . . , x̄d be the corresponding dual basis of F/F 2 (F i denotes the p-central
series of F ). We define
E = (xp1 , x2 , . . . , xd )F / F
to be the subgroup of F generated by all elements between the brackets and
their conjugates, where xi is an arbitrary lifting of x̄i to F . Then R ⊆ E and
if we put N = E/R ⊆ G and G = G/N = F/E = hx̄1 i ∼ = ZZ/pZZ, we get the

commutative exact ãäåæçèéêëìíîïð diagram

G G
1 R F G 1
1 R E N 1.
We consider the G-module R̄ := RE 2 /E 2 which fits into the exact sequence
(∗) 0 −→ R̄ −→ E/E 2 −→ N/N 2 −→ 0 .
Since R/Rp [R, F ] ∼
= ZZ/pZZ surjects onto R̄G , the G-module R̄ is generated
by one element. Observe that
F2 ⊆ E and F 3 ⊆ [[E, F ], F ] · E 2 .
It follows from (3.9.13)(ii) and the fact that χ1 ∪ χi = 0 for all i ≥ 1, that


 hxp1 i[[E, hx1 i], hx1 i]E 2 /E 2 if q = p =/ 2 ,
R̄ ⊆
[[E, hx1 i], hx1 i]E 2 /E 2 otherwise ,


and so 
hxp1 iE 2 /E 2 · (E/E 2 )(x̄1 −1)
2

 if q = p =/ 2 ,
R̄ ⊆ 2
(E/E 2 )(x̄1 −1) otherwise .


Thus in both cases p−2

R̄(x̄1 −1) =0
−1)p
(since (E/E 2 )(x̄1 = 0) and therefore
dimIFp R̄ ≤ p − 2 .
From the exact sequence (∗), we obtain using (3.9.6)
dimIFp N/N 2 = dimIFp E/E 2 − dimIFp R̄
≥ p(d − 1) + 1 − (p − 2)
> p(d − 2) + 2 .
This contradicts the assumption (vi) and therefore the cup-product pairing has
to be non-degenerate, i.e. G is a Demuškin group. 2
Now we are going to determine a defining relation of a Demuškin group

explicitly. Let G = F/R be a one-relator pro-p-group of rank n where R = (ρ).
First we want to put the relation ρ in the shape
ρ ≡ xq1 (x1 , x2 )(x3 , x4 ) · · · (xn−1 , xn ) mod F 3
and for this, we have to find an appropriate basis of H 1 (F, k).

If q =/ 0, then there is a uniquely determined class σ mod [F, F ] ∈ F ab =

F/[F, F ] such that
ρ ≡ σ q mod [F, F ].
We deduce this from the fact that F ab ∼ = ZZnp and that the image ρ̄ of ρ
in F topologically generates a closed subgroup (ρ̄) such that F ab /(ρ̄) ∼
ab
=
n−1
ZZ/q ZZ × ZZp .
We need a result concerning symplectic bilinear forms in which we write
χ ∪ χ0 for tr(χ ∪ χ0 ).
(3.9.16) Proposition. Assume that q =/ 2 and that the bilinear pairing induced
tr
by the cup-product H 1 (G, k) × H 1 (G, k) → H 2 (G, k) → k is non-degenerate.
Then n is even and there exists a k-basis χ1 , . . . , χn of H 1 (G, k) such that
χ1 ∪ χ2 = χ3 ∪ χ4 = . . . = χn−1 ∪ χn = 1
and χi ∪ χj = 0 for all other i < j, and χi (σ) = δ1i when q =/ 0.
∪
Proof: By (3.9.9), the cup-product H 1 (G) × H 1 (G) → H 2 (G) ∼ = IFp is a
non-degenerate bilinear form over the field IFp . If p = 2, it is alternating,
/
hence n = dimIFp H 1 (G) is even. When q = 2f , f > 1, n is also even, since
the cup-product on H 1 (G, ZZ/2f −1 ZZ) is still alternating and non-degenerate,
which implies that H 1 (G, ZZ/2ZZ) decomposes into a direct sum of hyperbolic
planes. For this, we refer to [106], chap.I, §4.
We start with any k-basis χ1 , . . . , χn of H 1 (G, k) such that χi (σ) = δ1i when
q =/ 0. To find such a basis when q =/ 0, one only has to extend the image
of σ in F/F q [F, F ] to a basis of F/F q [F, F ] and then take the dual basis.
The nondegeneracy of the cup-product means that the matrix B = (χk ∪ χl ) is
invertible over k. Therefore one of the elements χ1 ∪ χi with i > 1 must be a
unit of k. After a permutation, we may assume that χ1 ∪ χ2 is a unit, and after
multiplying χ2 by a unit, we may even assume χ1 ∪ χ2 = 1. If χ1 ∪ χi = ai =/ 0
for some i > 2, replace χi by χi − ai χ2 . Since the condition χi (σ) = δ1i when
q =/ 0 is not altered, we may assume χ1 ∪ χi = 0 for i > 2.
Now if V is the subspace spanned by χ3 , . . . , χn , our cup-product restricted
to V × V is non-degenerate and alternating. Hence we may inductively choose
χ3 , . . . , χn such that
χ3 ∪ χ4 = χ5 ∪ χ6 = . . . = χn−1 ∪ χn = 1
and χi ∪ χj = 0 for all other 2 < i < j. When q =/ 0 the condition χi (σ) = δ1i
is still satisfied, χ1 ∪ χ2 = 1 and χ1 ∪ χi = 0 for i > 2. If we replace χ2 by
χ 2 + a3 χ 3 + · · · + an χ n
with a2i = χ2 ∪ χ2i−1 and a2i−1 = −χ2 ∪ χ2i , we have, in addition, χ2 ∪ χi = 0
for i > 2. This proves the proposition. 2

The following corollary is a first approximation to our main theorem (3.9.11)

on Demuškin groups.
(3.9.17) Corollary. Let G = F/(ρ) be a finitely generated one-relator pro-p-

group with the invariants (n, q). Let q =/ 2.
Then G is a Demuškin group if and only if there exists a basis x1 , . . . , xn of
F such that
ρ ≡ xq1 (x1 , x2 )(x3 , x4 ) · · · (xn−1 , xn ) mod F 3 .
Proof: For any minimal generator system x1 , . . . , xn of F with corresponding

k-basis χ1 , . . . , χn of H 1 (G, k), we have by (3.9.13)(i)
n
qai
(xk , xl )akl mod F 3
Y Y
ρ≡ xi ·
i=1 1≤k<l≤n
with akl ∈ k uniquely determined (and ai ∈ k also, when q =/ 0). Furthermore,

by (3.9.13) (ii), the matrix B = (bkl ) of the cup-product pairing χ ∪ χ0 is given
by
−akl if k < l,


 
bkl = χk ∪ χl = alk if k > l,


q
−

2
ak if k = l.
Assume that
Then b12 = b34 = · · · = bn−1,n = −1 and bkl = 0 for all other k < l, and
bkk = − q2 , i.e. B is the matrix

q
− 1
 
2
 q
−1 −


 2
0 

B=

 ... 

 
 q 

 0 − 2
1 

q
−1 − 2 .
2
It has determinant det(B) = (1 + q2 )n/2 mod q ZZp , which is 1 if p =/ 2, or if
q = 2f =/ 2. Hence the cup-product is non-degenerate and G is thus a Demuškin
group.
Conversely, if G is a Demuškin group, then the cup-product χ ∪ χ0 is
non-degenerate and we may choose a k-basis χ1 , . . . , χn of H 1 (G, k) as in
proposition (3.9.16). Let ξ1 , . . . , ξn be the dual basis of F/F q [F, F ] and
x1 , . . . , xn a lift to F . Then x1 , . . . , xn is a minimal generator system of F

with corresponding basis χ1 , . . . , χn of H 1 (G, k). When q =/ 0, let

n
âi
Y
σ≡ xi mod [F, F ], âi ∈ ZZp .
i=1
In the representation
n
qai
(xk , xl )akl mod F 3 , ai , akl
Y Y
ρ≡ xi · ∈ k,
i=1 1≤k<l≤n
we have
ai ≡ âi mod q ZZp = χi (σ) = δ1i
when q =/ 0, because of the uniqueness of the ai , and
a12 = a34 = · · · = an−1,n = 1
and akl = 0 for all other k < l. We therefore get
ρ ≡ xq1 (x1 , x2 )(x3 , x4 ) · · · (xn−1 , xn ) mod F 3 . 2
Our last task is to make an equation from the congruence

This is achieved by a successive approximation process recalling that by (3.8.2),
the q-central series (F i ) of the finitely generated free pro-p-group F , i.e.
F 1 = F, F i+1 = (F i )q [F i , F ], where q is any nontrivial p-power, form a
fundamental system of open neighbourhoods of 1.
From now on we assume that the rank of F , i.e. the dimension n =

dimIFp H 1 (F ), is even. For every n-tuple y = (y1 , . . . , yn ) ∈ F n , we set
r(y) = y1q (y1 , y2 )(y3 , y4 ) · · · (yn−1 , yn ).
The following lemma is the most subtle part of the proof of theorem (3.9.11).
(3.9.18) Lemma. Let q =/ 2 and let ρ ∈ F 2 . Then for every j ≥ 3 there exists a
basis x = (x1 , . . . , xn ) of F such that
ρ ≡ r(x) mod F j ,
provided this is true for j = 3.
Proof: Let x = (x1 , . . . , xn ) be any minimal generator system of F . Let

t1 , . . . , tn ∈ F j−1 and set yi = xi t−1
i . Then y = (y1 , . . . , yn ) is again a minimal
generator system of F since yi ≡ xi mod F q [F, F ]. We may write
r(x) = r(y)dj−1 (t1 , . . . , tn ),

where dj−1 (t1 , . . . , tn ) is a uniquely determined element of F j . A simple

calculation using proposition (3.8.5) shows that if τi is the image of ti in
grj−1 (F ), then the image of dj−1 (t1 , . . . , tn ) in grj (F ) is

q
(∗) π · τ1 + 2
[τ1 , ξ1 ] + [τ1 , ξ2 ] + [ξ1 , τ2 ] + · · · + [τn−1 , ξn ] + [ξn−1 , τn ],
where ξi is the image of xi in gr1 (F ). Hence dj−1 induces a k-linear homo-
morphism
δj−1 : grj−1 (F )n −→ grj (F ), j ≥ 3.
Claim: The map δj−1 is surjective.

Let Hj = im(δj−1 ). In order to prove that Hj = grj (F ), it suffices to
show π · τ ∈ Hj for every τ ∈ grj−1 (F ). In fact, grj (F ) is generated by
the elements π · τ and [τ, ξi ] with τ ∈ grj−1 (F ); the explicit expression (∗)
for δj−1 (τ1 , . . . , τn ) tells us that we have [τ, ξi ] ∈ im(δj−1 ) for i ≥ 3 and
π · τ + [τ, ξ2 ], [τ, ξ1 ] ∈ im(δj−1 ).
We now proceed by induction. Assume that we have shown that Hj =
im(δj−1 ) for some j ≥ 3. If τ ∈ grj−1 (F ), then
τ = δj−1 (τ1 , . . . , τn )
with τ1 , . . . , τn ∈ grj−1 (F ). But then, using (3.8.5),
π · τ = δj−1 (π · τ1 , . . . , π · τn ),
which implies π · τ ∈ Hj+1 for τ ∈ Hj .
Thus we are reduced to proving the lemma for j = 3, that is, π · τ ∈ H3
for τ ∈ H2 . Moreover, it suffices to take τ in the form π · ξi , [ξi , ξj ], since
these elements generate gr2 (F ) by (3.9.13) (i). The ring k is a local ring with
maximal ideal m = pk. Hence by Nakayama’s lemma, it suffices to prove
π · gr2 (F ) ⊆ H3 + m gr3 (F ), since then we would have gr3 (F ) = H3 + m gr3 (F ).
Set M = m gr3 (F ). Then by proposition (3.8.5), we have
π · [ξi , ξj ] = [π · ξi , ξj ] + m = [ξi , π · ξj ] + m0 ,
where m, m0 ∈ M . Therefore, since [τ, ξi ] ∈ im(δ2 ) if i =/ 2, we have π·[ξi , ξj ] ∈
H3 + M for any i, j. Moreover, as π · τ + [τ, ξ2 ] ∈ H3 for any τ ∈ gr2 (F ), we
have π 2 · ξi ∈ H3 + M for any i. This proves the surjectivity of δj−1 .
Now let j ≥ 3 and assume that there exists a minimal generator system
x = (x1 , . . . , xn ) of F such that ρ ≡ r(x) mod F j , i.e. ρ = r(x)ej , ej ∈ F j .
Let εj be the image of ej in grj (F ). Since δj−1 is surjective, there exists
t1 , . . . , tn ∈ F j−1 such that δj−1 (τ1 , . . . , τn ) = −εj , i.e. dj−1 (t1 , . . . , tj ) = e−1
j
mod F j+1 . Put yi = xi t−1 i . Then y = (y 1 , . . . , y n ) is a minimal generator
system of F such that
ρ = r(x)ej = r(y)dj−1 (t1 , . . . , tn )ej ≡ r(y) mod F j+1 . 2

Proof of theorem (3.9.11): The proof is an immediate consequence of the

corollary (3.9.17) and the above lemma. Let G = F/(ρ) be a one-relator pro-p-
group with invariants (n, q). If there exists a basis x = (x1 , . . . , xn ) of F such
that ρ = r(x), then a fortiori ρ ≡ r(x) mod F 3 , and from (3.9.17) it follows
that G is a Demuškin group.
Conversely, let G be a Demuškin group. Then by (3.9.17) and (3.9.18), there
exists for every j ≥ 3 a minimal generator system x(j) = (x(j) (j)
1 , . . . , xn ) such
that
ρ ≡ r(x(j) ) mod F j .
When q = 0 this remains valid if we replace the central series F j by the
p-central series, i.e. it is always true that
(∗) ρ ≡ r(x(j) ) mod U (j),
where (U (j)) is a series of open normal subgroups which form a basis of
neighbourhoods of 1 by (3.8.2). If X (j) denotes the finite nonempty set of
minimal generator systems x(j) = (x(j) (j)
1 , . . . , xn ) of F/U (j) satisfying (∗), then
(j)
lim X is nonempty and any x = (x1 , . . . , xn ) in this projective limit is a basis
←−
of F such that ρ = r(x). This concludes the proof of theorem (3.9.11). 2
Remark: We have proved a sharper result than the assertion of theorem

(3.9.11). Namely, we have shown that for any defining relation ρ ∈ F of G,
i.e. G = F/(ρ), there exists a basis x1 , . . . , xn of F , such that
ρ = xq1 (x1 , x2 )(x3 , x4 ) · · · (xn−1 , xn ).
Along the same lines, but with more complicated modifications in the indi-
vidual steps, one can also settle the case q = 2. This was originally proved
by J.-P. SERRE in the case when n is odd, and later J. LABUTE obtained the
general case. For this we refer the reader to LABUTE’s article [117]. The result
is the following
(3.9.19) Theorem. Let G = F/(ρ) be a Demuškin group of rank n such that

q = 2.
(i) If n is odd, there exists a basis x1 , . . . , xn of F such that
f
ρ = x21 x22 (x2 , x3 )(x4 , x5 ) · · · (xn−1 , xn )
for some f = 2, 3, . . . , ∞ (f = ∞ means 2f = 0).
(ii) If n is even, there exists a basis x1 , . . . , xn of F such that
f
ρ = x2+α 2
1 (x1 , x2 )x3 (x3 , x4 )(x5 , x6 ) · · · (xn−1 , xn )
for some f = 2, 3, . . . , ∞ and α ∈ 4ZZ2 .

For further results on one-relator pro-p-groups, see [110], [118], [255].
In arithmetic applications one sometimes meets the situation that one is given
a canonical system of generators of a pro-p-group G. Even if the group G is
free, we are interested in a presentation of G in terms of generators and relations,
which is not necessarily minimal, but uses the given system of generators (or
at least generators which are closely related to the given ones). The following
theorem provides a typical example of such a situation. Its proof, for which we
refer the reader to [251], Lemma 2.4, uses an approximation process similar to
that in the proof of theorem (3.9.11).

1 −→ H −→ G −→ D −→ 1
be an exact sequence of pro-p-groups. Assume that G is free and that D is a
Demuškin group of rank n with torsion-free abelianization. Let D be generated
by x̄1 , . . . , x̄n with the one defining relation
(x̄1 , x̄2 ) · · · (x̄n−1 , x̄n ) = 1.
Suppose that the free ZZp -module H/[H, G] has the basis
{ỹj mod [H, G] | j = 1, . . . , s}
and that
s
Y
ỹj ∈ [G, G].
j=1
Then there exist generators x1 , . . . , xn , y1 , . . . , ys of G with

(i) xi mod H = x̄i , i = 1, . . . , n ,
(ii) yj = (ỹj )ατj with α ∈ ZZ×p , τj ∈ G, j = 1, . . . , s ,
subject to the one relation

s
Y
(x1 , x2 ) · · · (xn−1 , xn ) yj = 1.
j=1
Exercise: Let G be a pro-p-group.

\ Consider the filtration G = G1 ⊇ G2 ⊇ . . . defined by
∗
Gn = (Gn−1 ) . Show that Gn = {1} and that the Gn are open if and only if d(G) < ∞.
n≥1

Chapter IV
Free Products of Profinite Groups
§1. Free Products
As in III §5, we let c be a full class of finite groups and we consider the
category of pro-c-groups. A family
κi : Gi −→ G, i ∈ I,
of homomorphisms of pro-c-groups is called convergent (to 1) if every open
subgroup U of G contains the images κi (Gi ) for almost all i, i.e. all but a
finite number. The free products of pro-c-groups are defined by the following
universal property.
(4.1.1) Definition. The free pro-c-product of a family Gi , i ∈ I, of pro-c-groups

is a pro-c-group G together with a convergent family of homomorphisms
κi : Gi −→ G, i ∈ I,
such that for every other convergent family of homomorphisms κ0i : Gi → G0
there is a unique homomorphism of pro-c-groups κ0 : G → G0 such that the
diagrams
Góñò i
κi
G
κ0i κ0
0
G
are commutative. The group G is then denoted by
G= ∗G.
i∈I
i
The free pro-c-product is clearly unique up to isomorphism, if it exists. Its

discr
existence is based on the usual free product G = ∗ Gi of the Gi in the
i∈ I
category of groups (see [73]), which contains the Gi as subgroups. Let BG
denote the family of all normal subgroups N of G such that
(1) G/N ∈ c,
(2) N ⊇ Gi for almost all i ∈ I,
(3) N ∩ Gi is an open subgroup of the pro-c-group Gi .

246 Chapter IV. Free Products of Profinite Groups
Taking BG as a basis of open neighbourhoods of the identity of G defines

a group topology TG on G ∗) , which induces the profinite topology on Gi .
Indeed, if N ∈ BG , then N ∩ Gi is open in Gi , and, conversely, if U is any
open normal subgroup of Gi , we obtain a homomorphism G → Gi /U whose
kernel N is in BG and has the property that N ∩ Gi = U . This homomorphism
is obtained when we map Gj onto 1 ∈ Gi /U for j =/ i and Gi onto Gi /U by
the canonical projection. Since Gi is compact, it is a closed subgroup of G in
the topology TG . We now take the completion
G = lim G/N,
←−
N∈BG
which comes equipped with a canonical continuous homomorphism λ : G → G,

and with the family of composite maps
λ
κi : Gi ,→ G −→ G.
These satisfy the universal condition of a free pro-c-product. In fact, if we have

any convergent family of homomorphisms κ0i : Gi → G0 into a pro-c-group
G0 , then the diagram
Gøù÷õöô i G
κi λ
G
κ0i κ0
κ0
G0
of solid arrows is commutatively completed by the dotted arrows; κ0 exists

because of the universal property of the free product G in the category of
groups, and κ0 exists by the universal property of the completion: because of
the conditions (1), (2), (3) for the normal subgroups N ∈ BG , κ0 is continuous
and thus defines a homomorphism κ0 : G → G0 of pro-c-groups.
Example: Let ẐZ(c) be the free pro-c-group of rank 1 (see III §5). The free
pro-c-group F over a set X is the free pro-c-product
F = ∗ ẐZ(c).
x∈X
This follows at once from the universal properties, as do the following remarks
and propositions.
∗) The topology T is Hausdorff, by remark 1 at the end of §3. But we don’t need this here.
G

§1. Free Products 247
If J is a subset of I, then we have an exact sequence which splits canonically

(∗) ûú
1 −→ N −→ ∗G
i∈I
i
π
s ∗ G −→ 1.
i∈ J
i
The homomorphism π is obtained by choosing the canonical maps κi : Gi →

∗ Gi for i ∈ J and the trivial homomorphism Gi → {1} ⊆ Gi for i ∈ I r J.
i∈ J
∗
i∈ J
The homomorphism s is obtained by the canonical maps κj : Gj → ∗G
i∈ I
i for
j ∈ J. We have π ◦ s = id and consider ∗ G as embedded in ∗ G ,
i∈J
i
i∈I
i
∗G ∗G.
i∈J
i ⊆
i∈ I
i
In particular, we view the Gi as subgroups of G = ∗ G . They generate G i

i∈I
topologically, and the kernel N in (∗) is the smallest closed normal subgroup
of G containing the Gi , i ∈ I r J.
For an arbitrary index set I, we consider the family of finite subsets of I
partially ordered by inclusion.
(4.1.2) Proposition. The free pro-c-product G = ∗G

i∈ I
i is the projective limit
G = lim
←− i∈S
∗G , i
S
where S runs through the finite subsets of I.
Proof: It is sufficient to show that the projective limit on the right-hand side
satisfies the universal property of the free product with respect to homomor-
phisms to finite groups in c. But for a finite group H we have
∗ G , H) = lim Hom( ∗ G , H) = lim
M
Hom(lim i i Hom(Gi , H).
←− i∈S −→ i∈S −→
S ⊆I S ⊆I S ⊆I i∈S
This concludes the proof, because for every convergent family of homomor-
phisms (κi : Gi → H)i∈I , there exists a finite subset S ⊆ I such that κi = 0 for
all i ∈/ S. 2
If c0 ⊆ c is a full subclass of c, then to each pro-c-group G we may associate

the maximal pro-c0 -factor group G(c0 ). This is the projective limit
G(c0 ) = lim G/U
←−
U
over all open normal subgroups U of G such that G/U ∈ c0 . It is characterized

by the universal property that every homomorphism of G into a pro-c0 -group
factors through G(c0 ). From this, we obtain the

(4.1.3) Proposition. Denoting by ∗ the free pro-c -product, we have

0 0
( ∗ G )(c ) = ∗ G (c ).
i
0 0
i
0
i∈I i∈I
We now turn to the cohomology of free pro-c-products. It behaves in a

particularly nice way. Let G = ∗
Gi and let A be a G-module. Then the
i∈ I
restriction maps res i : H (G, A) → H n (Gi , A) define a homomorphism into
n
the direct sum

H n (G, A) −→ H n (Gi , A).
M
i∈I
In fact, if S runs through the finite subsets of I, then by (4.1.2) and (1.5.1) we
obtain
←− i∈S
∗
H n (G, A) = H n (lim Gi , A) = lim H n ( Gi , A)
−→
∗
i∈ S
S S
M M
→ lim
−→
H n (Gi , A) = H n (Gi , A).
S i∈S i∈I
We will now prove the following remarkable isomorphism theorem.
(4.1.4) Theorem. Let G = ∗ G . Then for every torsion G-module A we have

i∈ I
i
a canonical exact sequence

δ res
0 → AG (c) → A(c) → (A/AGi )(c) → H 1 (G, A) →
M M
H 1 (Gi , A) → 0,
i∈ I i∈I
and for all n ≥ 2
H n (G, A) ∼ H n (Gi , A).
M
=
i∈I
Proof: By a direct limit argument we may assume that A is finite. Moreover,

we may assume A ∈ c. Namely, if we decompose A into its p-primary com-
ponents A(p), then the cohomology groups decompose into direct summands.
These are zero if p does not divide the order of G (i.e. the order of a finite
quotient G/U ). Otherwise one sees at once that A ∈ c. Moreover we have
A/AGi = 0 for almost all i ∈ I, since there is an open subgroup U of G which
acts trivially on A.
The exact sequence of the theorem will follow by the snake lemma from the
0ýþÿü A/AG ∂
Z 1 (G, A) H 1 (G, A) 0
res res
∂
Z 1 (Gi , A)
M M M
0 A/AGi H 1 (Gi , A) 0,
i∈I i∈ I i∈I

once we have shown that the restriction map
res : Z 1 (G, A) −→ Z 1 (Gi , A)

M
(∗)
i∈I
on inhomogeneous 1-cocycles is bijective (its image is contained in the direct
sum by the same argument as for H n (G, A)). To this end, consider the exact
commutative diagram
πi
1
A Ĝi si
Gi 1
(∗∗)
π
1 A Ĝ s
G 1,
where Ĝ is the semi-direct product of A by G with the canonical homomorphic

section s, Ĝi = π −1 (Gi ) and si = s|Gi . The map χ 7→ s0 = χs gives a 1-1
correspondence between the 1-cocycles χ ∈ Z 1 (G, A) and the homomorphic
sections s0 : G → Ĝ of π. If res i χ = 1 for all i ∈ I, then s0 |Gi coincides with
si for all i ∈ I, hence s0 = s by the universal property of the pro-c-product, and
so χ = 1. This proves the injectivity of (∗).
Conversely let (χi )i∈I be a family of cocycles χi ∈ Z 1 (Gi , A) such that
χi = 1 for almost all i ∈ I. Then we obtain new sections s0i = χi si : Gi → Ĝi ,
hence a family of homomorphisms s0i : Gi → Ĝ which is convergent since
s0i = si for almost all i ∈ I. This family defines a homomorphism s0 : G → Ĝ
such that s0 |Gi = s0i by the universal property of the free pro-c-product. From
πi ◦ s0i = id it follows that π ◦ s0 = id, i.e. s0 is a new section of π and defines a
1-cocycle χ(σ) = s0 (σ)s(σ)−1 such that χ|Gi = χi for all i ∈ I. This proves the
surjectivity of (∗).
It remains to prove the second assertion of the theorem. By dimension
shifting, we obtain the surjectivity of the map
res : H n (G, A) −→ H n (Gi , A)
M
i∈I
for n ≥ 1. It suffices to prove the injectivity for n = 2, again by dimension
shifting. And again we may assume that A is finite and in c, and moreover
that I is finite, by (4.1.2). Let x ∈ H 2 (G, A) be such that xi = res i x = 0
for all i ∈ I. By (1.2.4), x and the xi define group extensions, and we have
a commutative diagram (∗∗) with homomorphic sections si : Gi → Ĝi ⊆ Ĝ,
since the upper group extensions split. By the universal property of the free
∗
pro-c-product G = Gi , the si define a homomorphic section s : G → Ĝ such
i∈I
that s|Gi = si . Hence the lower group extension splits, i.e. x = 0. This shows
the injectivity. 2

For pro-p-groups G we have a converse result to the above theorem. We set

H n (G) = H n (G, ZZ/pZZ).
(4.1.5) Theorem. Let G be a pro-p-group and let Gi , i ∈ I, be a convergent

subgroup family of G. Then the following conditions are equivalent.
(i) G = ∗G
i∈I
i (free pro-p-product).
(ii) The homomorphism

res : H n (G) −→ H n (Gi )
M
i∈I
is bijective for n = 1 and injective for n = 2.
Proof: The implication (i) ⇒ (ii) follows from (4.1.4) since G acts trivially
on ZZ/pZZ.
Conversely assume that (ii) holds. The inclusions Gi → G are by assumption
a convergent family of homomorphisms, and thus define a homomorphism of
the free pro-p-product
Ĝ = ∗ G −→ G.
i∈ I
i
π
By (4.1.4) we have
H j (Ĝ) = H j (Gi )
M
i∈I
for j = 1, 2, and so, by assumption,

H j (G) −→ H j (Ĝ)
is an isomorphism for j = 1 and injective for j = 2. Now the result follows
from (1.6.15). 2
The concept of free pro-c-product has generalizations in many directions (see

∗
[154]). One can define restricted free pro-c-products (Gi , Hi ) with respect
i∈I
to subgroups Hi ⊆ Gi and one obtains an idèle version. Further, one can
consider free pro-c-groups with amalgamated subgroup G1 ∗ G2 . We mention
H
some facts about the last concept in the exercises.
However, one important generalization will be introduced in §3. While the
groups occurring as factors in the free product have been independent of each
other so far, we will introduce there the free pro-c-product over a family of
pro-c-groups which vary continuously over a topological base space.

Exercise 1. Let i1 : H ,→ G1 , i2 : H ,→ G2 be two injective homomorphisms of pro-c-groups.

The push-out of i1 and i2 is a commutative diagram of pro-c-groups
H
i2
G2
i1 κ2
G1 κ1 G
with the universal property that any two homomorphisms κ01 : G1 → G0 , κ02 : G2 → G0 into
a pro-c-group G0 such that κ01 ◦ i1 = κ02 ◦ i2 determine a unique homomorphism f : G → G0
such that κ1 ◦ f = κ01 and κ2 ◦ f = κ02 . One speaks of an amalgamated free pro-c-product
along H if κ1 and κ2 are injections. One writes in this case
G = G1 ∗ G2 .
H
Show that the push-out always exists. The amalgamated product does not always exist (see
[179] for an example).
Exercise 2. If c is the class of all finite groups, the amalgamated free profinite product exists
in the following cases:
1) H is central in either G1 or G2 ;
2) H is normal in both G1 and G2 and is topologically finitely generated;
3) H is finite.
The amalgamated free pro-c-product G1 ∗ G2 of two pro-p-groups G1 , G2 along a common
H
procyclic subgroup H exists (see [179]).
For the next exercises (see [178]), assume that the amalgamated free pro-c-product
G = G1 ∗ G2 exists and consider the cohomology groups of pairs as defined in I §6, ex.4
H
for a torsion G-module A.
Exercise 3. (Excision) The canonical homomorphism

H n (G, G1 , A) −→ H n (G2 , H, A)
induced by the inclusion (G2 , H) ,→ (G, G1 ) is an isomorphism for all n ≥ 0.
Exercise 4. The inclusions (G1 , H) ,→ (G, H) and (G2 , H) ,→ (G, H) induce an isomorphism
H n (G, H, A) ∼
= H n (G1 , H, A) ⊕ H n (G2 , H, A) (n ≥ 0).
Exercise 5. (Mayer-Vietoris sequence) We have an exact sequence

∆
· · · −→ H n (H, A) −→ H n+1 (G, A) −→ H n+1 (G1 , H, A) ⊕ H n+1 (G2 , H, A)
Ψ
−→ H n+1 (H, A) −→ H n+2 (G, A) −→ · · · ,
where ∆ is the composite of the canonical maps
δ ∼ ι
H n (H, A) −→ H n+1 (G2 , H, A) −→ H n+1 (G, G1 , A) −→ H n+1 (G, A),
and Ψ (x1 ⊕ x2 ) = h1 (x1 ) − h2 (x2 ), where h1 and h2 are induced by the inclusions H ,→ G1
and H ,→ G2 .

§2. Subgroups of Free Products
A very useful result is the following analogue of the well-known Kurosh

subgroup theorem in (discrete) group theory. In the form presented here it is
due to E. BINZ, J. NEUKIRCH and G. WENZEL (see [12]). There are different
proofs and generalizations by several other authors (see [56], [74], [135]).
(4.2.1) Theorem. Let G =

i∈I
∗G i be the free pro-c-product of the Gi and let
H be an open subgroup of G. Then there exist systems Si of representatives
si of the double coset decomposition G = . si ∈Si Hsi Gi for all i and a free
S
pro-c-group F ⊆ G of the finite rank

X
rk(F ) = [(G : H) − #Si ] − (G : H) + 1,
i∈I
such that the natural inclusions induce a free product decomposition
H= ∗ (G
i,si
si
i ∩ H) ∗ F,
where Gsi i (= si Gi s−1

i ) denotes the conjugate subgroup.
Remarks: 1. If N ⊆ H is an open subgroup which is normal in G, then

Gi ⊆ N for almost all i ∈ I, and for such i we have #Si = (G : H). Hence
rk(F ) is finite.
2. It is not true in general that we may choose any systems Si of representatives
of the double coset decompositions. Only if c is the class of p-groups the
theorem is true for any choice of the Si by (4.1.5).
discr
Proof of theorem (4.2.1): Let G = ∗
i∈I
Gi be the (discrete) free product
of the Gi in the category of groups. We give G the topology TG which is
defined by the family BG of open normal subgroups N ⊆ G satisfying the
three conditions
(1) G/N ∈ c,
(2) N ⊇ Gi for almost all i and
(3) N ∩ Gi is an open subgroup of the pro-c-group Gi for all i.
Then G is the completion of G with respect to TG , and the restriction of TG to
Gi is the given profinite topology on Gi (compare the existence proof for free
products in §1). Let κ : G → G be the canonical completion homomorphism.
We denote the isomorphic images of the groups Gi in G and G also by Gi .

§2. Subgroups of Free Products 253
Now let H be an open subgroup in G. We denote the open subgroup

κ−1 (H) ⊆ G by H. In particular, H is of finite index in G and by the Kurosh
subgroup theorem for discrete groups (see [215], chap.I §5.5 th. 14), there
exist systems Si of representatives σi of the double coset decomposition G =
S. Hσ G and a (discrete) free group F ⊆ G of rank
i i
X
rk(F) = [(G : H) − #Si ] − (G : H) + 1,
i∈ I
such that the natural inclusions induce a free product decomposition
discr
H= ∗ (G
i,σi
σi
i
discr
∩ H) ∗ F.
By the argument of remark 1 above, we conclude that the sum on the right side
of the rank equation is finite.
Since TG induces on Gi , and therefore also on Gσi i , the pro-c-topology, and
since H is open, Gσi i ∩ H is an open subgroup of the pro-c-group Gσi i . Now let
TH be the topology on H which is induced by the family BH of open normal
subgroups I ⊆ H satisfying the three conditions
(1) H/I ∈ c,
(2) I ⊇ Gσi i ∩ H for almost all i, σi and
(3) I ∩ Gσi i ∩ H is an open subgroup of the pro-c-group Gσi i ∩ H
for all i, σi .
We first note that TH induces on Gσi i ∩ H the pro-c-topology of Gσi i ∩ H,
and induces on F the topology given by all normal subgroups IF ⊆ F with
F/IF ∈ c. Therefore the completion of H with respect to the topology TH has a
free product decomposition as in the statement of the theorem. Thus we have
to show:
Claim: The restriction of the topology TG to H is equal to TH .
For an N ∈ BG , one easily observes N ∩ H ∈ BH and therefore the topology
TH is finer than TG ∩ H. To show that also TG ∩ H is finer than TH , let I ∈ BH .
We have to find a group N ∈ BG such that N ∩ H ⊆ I. We claim that we can
take for N the intersection Ĩ of all conjugates of I in G. Since I is of finite
index in H and H is of finite index in G, I is of finite index in G. Therefore I
has only finitely many conjugates Iτj in G, i.e. Ĩ is open and of finite index in
G. Denoting by H̃ ∈ BG the intersection of the finitely many conjugates of H
in G, we obtain an exact sequence
1 → H̃/Ĩ → G/Ĩ → G/H̃ → 1.
The group H̃/Ĩ is canonically embedded into the group τj Hτj /Iτj ∈ c, and
Q
the class c is full. Therefore G/Ĩ ∈ c. Next we have to show that Gi ∩ Ĩ is open
in the pro-c-topology of Gi . It is sufficient to show that Gi ∩ Iτj is open in Gi
τ −1 τ −1
or, equivalently, that Gi j ∩ I is open in Gi j . If we write τj−1 = h · σi · gi ,

τ −1
h ∈ H, σi ∈ Gi , we get Gi j ∩ I = (Gσi i ∩ I)h and this group
Si and gi ∈
τj−1 ∗
is open in the pro-c-group (Gσi i )h = Ghσ
i
i gi
= Gi
) . At the same time,
τ −1 τ −1
we see that Gi j ∩ I = (Gσi i ∩ I)h = (Gσi i )h = Gihσi gi = Gi j for almost all
i ∈ I. This proves the claim. The proof of the theorem is then completed by
the observation that the images si ∈ H of the σi under κ form systems Si of
representatives of the double coset decomposition . Hsi Gi in G. 2
S
The next corollary is the profinite analogue of the Nielson-Schreier formula

for discrete groups. If G is a pro-p-group, we have already proved it in III §5
using partial Euler-Poincaré characteristics (see (3.9.6)).
(4.2.2) Corollary. An open subgroup H of a free pro-c-group G is again a free

pro-c-group. The ranks satisfy the equation
rk(H) − 1 = (rk(G) − 1) (G : H).
Proof: A free pro-c-group G is the free pro-c-product G = i∈I Gi , where ∗

Gi ∼ (c) denotes the free pro-c-groups of rank 1 (see III §5). Since (loc.cit.)
= ẐZQ
ẐZ(c) = p∈S ZZp , where S is the set of prime numbers which divide the order
of a group in c, we see that every open subgroup of the free pro-c-group of
rank 1 is isomorphic to the free pro-c-group of rank 1 again. By (4.2.1), there
exists a system Si of representatives si of the double coset decomposition
G = . si ∈Si Hsi Gi and a free pro-c group F ⊆ G of the finite rank rk(F ) =
S
i∈I [(G : H) − #Si ] − (G : H) + 1, such that the natural inclusions induce

P
∗
a free product decomposition H = i,si (Gsi i ∩ H) ∗ F. Therefore H is a free
pro-c-group (as a free pro-c-product of free pro-c-groups) and
rk(Gsi i ∩ H) + rk(F )
X
rk(H) =
i,si
X
= #Si + rk(F )
i∈ I
X
= (G : H) − (G : H) + 1
i∈ I
= (G : H) (rk(G) − 1) + 1. 2
Having proved several results which are quite similar to those in discrete
group theory, the reader should be warned of some curious pathologies in the
profinite case, which are (more or less) due to the existence of infinite words
in the profinite products.
∗) Observe the rule (Ga )b = Gba .

§2. Subgroups of Free Products 255
For example, if a group P is the free product of its subgroups G and H, then
one should expect that for every pair σ, τ ∈ P , the conjugates Gσ and H τ also
form a free product. However, look at the following example due to D. HARAN
(see [74]).
(4.2.3) Proposition. Assume that ZZ/2ZZ, ZZ/3ZZ ∈ c and let G (resp. H) be a

finitely generated pro-c-group having a normal subgroup of index 2 (resp. 3).
Let
P = G ∗ H.
Then there exist elements σ, τ ∈ P such that the subgroups Gσ and H τ do not
form a free pro-c-product in P .
Proof: Let Q be the set of closed subgroups Q ⊆ P such that there exist
σ, τ ∈ P with Gσ , H τ ⊆ Q. For Q ∈ Q the sets {σ ∈ P | Gσ ⊆ Q} and
{τ ∈ P | H τ ⊆ Q} are closed in P . We deduce that Q is closed under
descending chains. By Zorn’s lemma, it has a minimal element, say Q.
Assume that Q = Gσ ∗ H τ for suitable σ, τ ∈ P . Then, by our assumptions on
G and H and by the universal property of the free product, there would exist
a homomorphism φ : Q → S4 (the symmetric group on four elements), such
that φ(Gσ ) = h(12)i and φ(H τ ) = h(134)i. Since S4 is generated by the cycles
(12) and (134), φ is surjective. Choose ρ ∈ Q with φ(ρ) = (1234). Then
φ(hGσ , H ρτ i) = h(12), (134)(1234) i = h(12), (124)i $ S4 .
This shows that hGσ , H ρτ i $ Q is contained in Q. This contradicts the
minimality of Q. 2
Remarks: 1. It follows from the construction of the free pro-c-product that

P = Gσ ∗ H τ whenever σ and τ are finite words in elements of G and H (i.e.
σ and τ are the products of finitely many elements which are in G or H).
2. This kind of pathology does not occur if c is the class of pro-p-groups. Then
P = G ∗ H is the free pro-p-product of Gσ and H τ for every pair σ, τ ∈ P .
This follows from (4.1.5). However, by the result of exercise 3 below, this is
the only good case.
Exercise 1. Assume that we are given pro-c-groups G, H and an open normal subgroup
N ⊆ G. Show that the kernel of the homomorphism
G ∗ H −→ G/N ,
which, by the universal property of the free pro-c-product, is given by the projection G → G/N
and the trivial homomorphism H → {1} ⊆ G/N , is of the form
N ∗ ( ∗ H σ ),
σ ∈R
where R is a system of representatives of the cosets of N in G.

Exercise 2. Let p be a prime number and assume ZZ/pZZ ∈ c. Determine the kernels of the
homomorphisms
ZZ/pZZ ∗ ZZ/pZZ → ZZ/pZZ ,
which, by the universal property of the free pro-c-product, are associated to the pairs (id, id)
and (0, id).
Exercise 3. Assume that the full class c is strictly larger than the class (p-groups) and assume
that G, H are pro-p-groups. Let P = G ∗ H be the free pro-c-product of G and H.
(i) Show that P is not a pro-p-group if G and H are nontrivial.
(ii) Show that there are σ, τ ∈ P such that the closed subgroup U = hGσ , H τ i, generated
by Gσ and H τ in P , is a pro-p-group.
(iii) Show that the closed subgroup U constructed in (ii) is isomorphic to the free pro-p-
product of G and H.
Hint: Use (4.2.1) for (i), the Sylow theorems (1.6.9) for (ii) and (4.1.5) for (iii).
Exercise 4. Assume that the finite group G is the semi-direct product of its subgroup H and
its normal subgroup N . Consider the canonical surjection
ε : N ∗ H −→ G,
which is induced by the inclusions of N and H into G. Show that the kernel of ε is a free
profinite group of rank (#N − 1)(#H − 1) with basis
{hnh−1 n−1
h
∈ N ∗ H | h ∈ H\{1}, n ∈ N \{1}},
where nh := (hnh−1 ) ∈ N , i.e. ker(ε) is the free profinite group on the symbols which
represent the difference between formal and real conjugation of nontrivial elements of N by
nontrivial elements of H.
§3. Generalized Free Products
In this section we define a vast generalization of the free products of §1.

Instead of (discrete) families of profinite groups we now consider families
of profinite groups which vary continuously over a topological base space.
Such generalizations were introduced by D. GILDENHUYS and L. RIBES ([56])
for locally constant families and by D. HARAN ([74]) and (independently) by
O. MELNIKOV ([135]) in its most general form.
As before, let c be a full class of finite groups. Given continuous maps
f : X → Z, g : Y → Z between profinite spaces, we will write
X ×Z Y = {(x, y) ∈ X × Y |f (x) = g(y)}
for the fibre product of X and Y over Z, which is a profinite space.

§3. Generalized Free Products 257
(4.3.1) Definition. Let T be a profinite space. A bundle of pro-c-groups

over T is a group object in the category of profinite spaces over T , such that
the fibre over every point of T is a pro-c-group. In other words, a bundle of
pro-c-groups over T is a tuple (G, p, T, m, e, ι), where G is a profinite space
together with continuous maps
p : G → T (the structure map),
m : G ×T G → G (the multiplication),
e : T → G a section to p (the unit),
ι : G → G (the inversion),
such that the fibre Gt := p−1 (t), together with the induced maps mt : Gt ×Gt →
Gt , ιt : Gt → Gt and the unit element et := e(t), is a pro-c-group for every
point t ∈ T .
Because of the existence of the unit e, the structure map p is necessarily

surjective. Usually we will denote such a bundle by (G, p, T ), or by G if no
confusion is possible.
Example 1. If G is a pro-c-group and T is a profinite space, then we always have

the constant bundle (G × T, prT , T ), where prT is the projection G × T → T
and the maps m, ι, e are those induced by the group operations in G.
Example 2. Assume that (Gi )i∈I is a (discrete) family of pro-c-groups. We
define the associated bundle of pro-c-groups as the bundle over the one point
compactification I¯ := I . {∗} of I which is given as a set by
S
G := . Gi . {∗} ,
[ S
i∈I
and which has the following topology: Gi ⊆ G (together with its profinite
topology) is open in G for all i, and for every open neighbourhood U ⊆ I¯ of
¯ let
∗ ∈ I, [
. G S. {∗}
i
i∈U
be an open neighbourhood of ∗ ∈ G. The space G is profinite and one checks

that the map
¯ Gi 3 gi 7→ i, ∗ 7→ ∗
p : G → I;
is continuous. Viewing {∗} as the group with one element, we see that the
group operations on the Gi ’s induce the structure of a bundle of pro-c-groups
¯
on the triple (G, p, I).
Assume that we are given a pro-c-group G and a profinite space T .

(4.3.2) Definition. A family (Gt )t∈T of subgroups of the profinite group G,

indexed over the profinite space T , is a continuous family of subgroups if for
every open subgroup U ⊆ G the set T (U ) := {t ∈ T | Gt ⊆ U } is open in T .
Given such a continuous family of subgroups, we define a bundle of pro-c-

groups (G, p, T ) over T by putting
G = {(g, t) ∈ G × T | g ∈ Gt }
and defining p to be the restriction to G of the projection of G × T onto T . We
define the maps m, ι, e by restricting the corresponding maps from the constant
bundle.
(4.3.3) Lemma. (G, p, T ) is a bundle of pro-c-groups.
Proof: Once we have proved that G is a profinite space, all properties follow
from the corresponding properties of the constant bundle G × T → T . Hence
it remains to show that G is closed in G × T . Let (h, s) ∈/ G, i.e. h ∈/ Gs .
Then there exist open subgroups V, W ⊆ G with Gs ⊆ W and hV ∩ W = ∅.
Since the family (Gt ) is continuous, T (W ) is open in T . But then (h, s) ∈
hV × T (W ) ⊆ G × T \ G, which finishes the proof. 2
(4.3.4) Definition. A morphism of bundles

φ : (G, pG , T ) → (H, pH , S)
is a pair φG : G → H, φT : T → S of continuous maps such that
(i) the diagram
G
φG
H
pG pH
φT
T S
commutes and
(ii) for every t ∈ T the associated map φt : Gt → HφT (t) is a group homo-
morphism.
We say that φ is surjective if φG (and hence also φT ) is surjective.
We will not distinguish between the pro-c-group G and the bundle (G, p, {∗})
over the one point space {∗}. In particular, a morphism from a bundle (G, p, T )
to a group G is a continuous map φ : G → G such that the induced maps
φt : Gt → G are group homomorphisms for every t ∈ T . One easily verifies

that the family ( φ(Gt ) )t∈T is a continuous family of subgroups of G indexed

by T .
(4.3.5) Definition. The free pro-c-product of a bundle (G, p, T ) of pro-c-groups

is a pro-c-group
G= G ∗
T
together with a morphism ω : G → G, which has the following universal

property: for every morphism f : G → H from G to a pro-c-group H there
exists a unique homomorphism of pro-c-groups φ : G → H with f = φ ◦ ω.
In categorical language this means that the functor “free product over T ” is a
left adjoint to the functor “constant bundle” from (pro-c-groups) to (bundles of
pro-c-groups over T ).
The map ω induces a canonical map ωt : Gt → G for every t ∈ T . We will

see in (4.3.12) that ωt is an injective group homomorphism for every t ∈ T and
that the images of the Gt in G are in some sense independent from each other.
(4.3.6) Proposition. The free pro-c-product ∗ G exists and is unique up to

T
unique isomorphism.
Proof: The uniqueness assertion is clear by the universal property. The

existence follows from the following construction:
Let L be the abstract free product of the family of groups (Gt )t∈T and let
λ : G → L be the map which is given on every Gt as the natural inclusion
∗
of Gt into L. Then define the free product G = T G as the completion of L
with respect to the topology which is given by the family of normal subgroups
N ⊆ L of finite index for which (a) L/N ∈ c and (b) the composition pN ◦ λ of
λ with the natural projection pN : L → L/N is continuous. It is easily verified
that G has the required universal property. (Compare with the construction of
the free product in §1.) 2
(4.3.7) Definition. Assume that (Gt )t∈T is a continuous family of subgroups

of the pro-c-group G indexed over the profinite space T and let G be the
pro-c-group bundle which is associated to the family (Gt )t∈T by (4.3.3). We
say that G is the free pro-c-product over the family (Gt )t∈T if the canonical
∗
homomorphism G → G is an isomorphism.
T

Example 3. Assume that G is a pro-c-group and that T is a profinite space.

Then we denote the free pro-c-product over the constant bundle G × T → T
∗
by T G.
Assume, in addition, that a profinite group Γ acts continuously on T . Then,
by the universal property of the free product, Γ also acts continuously on T G, ∗
∗
i.e. T G becomes a pro-c-Γ operator group, which is defined as follows.∗)
(4.3.8) Definition. Let Γ be a profinite group. A pro-c-Γ operator group

is a pro-c-group G together with a homomorphism
φ : Γ −→ Aut(G),
such that the action: Γ × G → G, (γ, g) 7→ φ(γ)(g), is continuous.

If Γ = T acts on itself by left multiplication and if G = Fr is the free pro-c-
∗
group of rank r, then Fr is the free pro-c-Γ operator group of rank r.
Γ
Free pro-c-Γ operator groups satisfy the obvious universal property in the
category of pro-c-Γ operator groups.
Example 4. Assume that we are given a (discrete) family (Gi )i∈I of pro-c-
groups and let G be the associated bundle over the one point compactification
I¯ of I (see example 2). Then to give a morphism φ : G → G is the same as
to give a convergent (see §1) family of group homomorphisms φi : Gi → G.
Therefore we obtain a canonical isomorphism
∗ G ∼= ∗ G .
I¯ i∈I
i
¯
If I is finite, it is compact and we can omit the point ∗ ∈ I.
More generally, from the universal property follows the
(4.3.9) Proposition. Let (G, p, T ) be a bundle of pro-c-groups and assume that

T = T1 . · · · . Tk is a finite disjoint decomposition of T into closed and open
S S
subsets. Then there is a canonical isomorphism

∗G = ∗G
T T1
1 ∗···∗ ∗G ,
Tk
k
where Gi denotes the restricted bundle p−1 (Ti ) for i = 1, . . . , k.
∗) The notion of a profinite operator group was introduced by H.KOCH

One of the basic facts of the theory of profinite groups is the assertion that
a topological group which is profinite as a topological space is already the
inverse limit of finite groups (see (1.1.3)). The following theorem extends this
fact to continuous families of profinite groups.
We call a bundle of pro-c-groups finite if it is finite as a space, i.e. it is a
finite disjoint union of groups in c.
(4.3.10) Theorem. Every pro-c-group bundle is the inverse limit of a system

of finite bundles with surjective transition morphisms.
Proof: Denote the given bundle by (G, p, T ). As a topological spaces, G and

T are inverse limits of finite discrete spaces. We will make use of the following
claim:
For every pair of disjoint decompositions G = U1 . · · · . Un of G and
S S
T = V . · · · . V of T into open and closed subsets there exists a

S S
1 m
morphism
φ : (G, p, T ) → (H, p0 , S)
to a finite bundle (H, p0 , S), such that the decompositions of G (resp. T )

into the pre-images of the (finitely many) points of H (resp. S) are finer
than the given decomposition.
The proof of this claim will be given below.

Observe that the set of finite quotients of (G, p, T ) is filtered. Indeed, assume
that we are given two finite quotients (Hi , pi , Si ) (i = 1, 2) of (G, p, T ). Then
we first take their pullbacks to a suitable chosen common finite basis S and
then we take the fibre product over S. Now it follows from the claim that the
canonical morphism from (G, p, T ) to the inverse limit of its finite quotients is
an isomorphism.
It remains to show the claim. We construct φ in several steps.
Step 1. Fix a point t ∈ T . Then, by (1.1.3), there exists a finite group
Ht and a homomorphism φt : Gt → Ht , such that the decomposition of Gt
into the pre-images of the elements of Ht is finer than the decomposition
Gt = ni=1 (Ui ∩ Gt ). (In order to be accurate, we should remove the indices i
S
with Ui ∩ Gt = ∅.)
Step 2. We can extend φt : Gt → Ht to a bundle morphism φt : GWt → Ht ,
where Wt is a sufficiently small open and closed neighbourhood of t ∈ T and
GWt denotes the restriction of G to Wt (see exercise 1 below).

Step 3. Making Wt smaller, if necessary, we may assume that the decom-

position of GWt into the pre-images of the elements of Ht is finer than the
decomposition GWt = ni=1 (GWt ∩ Ui ).
S
Step 4. The open and closed sets Wt (t ∈ T ) cover T . Replacing Wt by

Wt ∩ Vj (j = 1, . . . , m), we may assume that every Wt is contained in one of
the closed and open subsets V1 , . . . , Vm . Since T is compact, there exists a
finite subcovering T = Wt1 ∪ · · · ∪ Wtk , which we may assume to be disjoint,
Sj−1
by replacing Wtj by Wtj \ ν=1 Wtν .
Step 5. The bundle H .
S
· · · . H (together with the obvious structure map
S
t1 tk
to the set of k elements) is a finite bundle to which we have a bundle morphism
from (G, p, T ) of the required type. 2
The next proposition is the natural generalization of (4.1.2).

(G, p, T ) = lim (Gi , pi , Ti )
←−
i∈I
be the inverse limit of the pro-c-group bundles Gi . Then
∗ G = lim ∗ G .
T ←− Ti
i
i∈I
Proof: It suffices to show that the inverse limit on the right-hand side satisfies
the universal property of the free product with respect to homomorphisms into
finite groups in c. To begin with, we show the statement in the special case
that all bundles Gi are finite and all transition maps are surjective. In this case,
we have equalities for every finite group H ∈ c :
Hom( ∗ G, H)
T
= Mor((G, p, T ), H) = Mor(lim (Gi , pi , Ti ), H)
←−
i
= lim Mor((Gi , pi , Ti ), H) = lim Hom(
−→ −→
∗ G , H)
Ti
i
i i
= Hom(lim ∗ G , H).
←− Ti
i
i
Returning to the general case, represent every Gi as the inverse limit of its finite
quotients: Gi = lim Gij . Every morphism
←− j
φi1 ,i2 : Gi1 → Gi2

is realized as a limit of morphisms on finite levels. Therefore we can replace Gi
by the system Gij without changing either side in the stated equality. In other
words (changing the index set), we may assume that all Gi are finite. Again,

without changing the projective limit, we can add the images of all transition
maps to the inverse system and since all Gi are finite, the intersections
\
im(Gj ) ⊆ Gi
j ≥i
stabilize and set up a cofinal subsystem. Finally, we arrive at the point that all
bundles in the system are finite and all transition maps are surjective, and this
special case was already solved at the beginning of the proof. 2
Using (4.3.10) and (4.3.11), many results for finite free products carry over
to the case of generalized free products. For example, we have the
(4.3.12) Proposition. Let ω : G → G, G = ∗ G be a free pro-c-product.

T
Then
(i) ω|Gt : Gt → G is an injective group homomorphism for every t ∈ T ,
(ii) if s, t ∈ T and g ∈ G satisfy ω(Gs ) ∩ ω(Gt )g =/ 1, then s = t.
Proof: Since all statements are compatible with inverse limits, we may
assume by (4.3.10), (4.3.11) that G is finite. Then (i) is trivial by the results of
§1. In order to prove (ii), assume s =/ t. It suffices to construct a group H and
a homomorphism f : G → H whose restriction to ω(Gs ) is injective and such
that the images of ω(Gs ) and gω(Gt )g −1 have trivial intersection in H. For
example, we can choose H := Gs × Gt . Let f be the homomorphism which
corresponds via the universal property to the bundle homomorphism which
is given by Gs → H, gs 7→ (gs , 1), Gt → H, gt 7→ (1, gt ), and the trivial
homomorphism on Gu for every other u. Then we have f (ω(Gs )) = Gs × 1
and f (gω(Gt )g −1 ) = f (g)f (ω(Gt ))f (g)−1 = 1 × Gt . 2
Remarks: 1. In fact a much stronger statement than (4.3.12) is true: the

canonical homomorphism from the discrete free product of the Gt to G = G ∗T
is injective. Using (4.3.10) and (4.3.11), this can be deduced from the results
in [56].
2. There are also variants of the Kurosh subgroup theorem for the generalized
free products (see [56], [74], [135], [261]).
Now we calculate the cohomology of free products. We start with the

definition of restricted direct sums.

(4.3.13) Definition. Let (G, T ) be a bundle of pro-c-groups, G = ∗ TG and A

a torsion G-module. For each n ≥ 0 we denote by
M0
H n (Gt , A)
Q t∈T n
the subgroup of elements in t∈T H (G
t , A) which are in the image of some
map
H n (G0t0 , A0 ) −→ H n (Gt , A),
M Y
t0 ∈T 0 t∈T
where (G , T ) is a finite quotient bundle of (G, T ) and A0 is a finite submodule
0 0
∗
of A which is a T 0 G 0 -module.
Remark: If (Gi )i∈I is a discrete family of pro-c-groups and G is the associated

bundle over the one point compactification I¯ = I ∪ {∗} of the discrete set I,
see example 2 page 257, then
M0
H n (Gi , A) = H n (Gi , A).
M
i∈I¯ i∈I
In many arithmetical applications the considered bundle will be the projective
limit of bundles which are associated to discrete families (Gλ,i )i∈Iλ , λ ∈ Λ,
of pro-c-groups, i.e. (G, T ) is the projective limit of the associated bundles
(Gλ , I¯λ ) over the one point compactifications I¯λ = Iλ ∪ {∗} of the discrete sets
Iλ , λ ∈ Λ.
∗
Assume in this situation that the torsion T G-module A is a I¯λ Gλ -module ∗
for all λ ∈ Λ. Since 0t∈T H n (Gt , A) = lim
L L0 n
−→ λ∈Λ i∈I¯λ H (Gλ,i , A), we have in
this case M0
H n (Gt , A) = lim H n (Gλ,i , A).
M
−→
t∈T λ∈Λ i∈Iλ
(4.3.14) Theorem. Let (G, T ) be a bundle of pro-c-groups, G = ∗ TG and A

a torsion G-module. Then we have a canonical exact sequence
M0 M0
0 → AG (c) → A(c) → (A/AGt )(c) → H 1 (G, A) → H 1 (Gt , A) → 0,
t∈T t∈T
and for all n ≥ 2
M0
H n (G, A) ∼
= H n (Gt , A).
t∈T
Proof: By a direct limit argument, we may assume that A is finite. The

composite map G → G → Aut(A) is continuous, hence locally constant.
Using the claim in the proof of (4.3.10), we find a presentation
(G, p, T ) = lim
←−
(Gi , pi , Ti )
i∈ I

of the bundle G as an inverse limit of finite bundles Gi = . s∈Ti Gi,s , i ∈ I,

S
∗ ∗
such that A is a Ti Gi -module for all i. We set Gi = Ti Gi . By (4.3.11),
G = lim Gi , and by (4.1.4), we obtain an exact sequence
←− i∈I
0 → A/AG → lim A/AGi,s → H 1 (G, A) → lim

M M
−→
H 1 (Gi,s , A) → 0
−→
i∈I s∈Ti i∈I s∈Ti
and isomorphisms for all n ≥ 2
H n (G, A) −→ H n (Gi,s , A).
M
∼ lim
−→
i∈I s∈Ti
For n ≥ 1 we consider the natural map

M0
H n (Gi,s , A) −→ H n (Gt , A).
M
φ: lim
−→
i∈I s∈Ti t∈T
Each finite quotient bundle (G 0 , p0 , T 0 ) of (G, p, T ) is a quotient of (Gi , pi , Ti ) for

sufficiently large i ∈ I. Therefore φ is surjective. Let x ∈ s∈Ti0 H n (Gi0 ,s , A)
L
L n
represent a nonzero element in lim
−→ i∈I s∈Ti H (Gi,s , A). Then, for each i ≥ i0 ,
the set Si ⊆ Ti of elements s ∈ Ti such that the s-component of the image of
x in s∈Ti H n (Gi,s , A) is nonzero is nonempty. As the inverse limit of finite
L
nonempty sets is nonempty, φ is injective, hence an isomorphism. Similarly,

the natural map M0
A/AGi,s −→ A/AGt
M
lim
−→
i∈I s∈Ti t∈T
is an isomorphism. This completes the proof of the theorem. 2
(4.3.15) Corollary. Let (G, p, T ) be a bundle of pro-c-groups and let p be a

prime number. Then
∗
cdp G = sup {cdp Gt } and scdp G = sup {scdp Gt }.
T t∈T
∗
T t∈T
Proof: By (4.3.14), we obtain the inequality cdp G ≤ sup t∈T {cdp Gt }. By

(4.3.12), each Gt is a closed subgroup of G, hence, by (3.3.5), cdp G ≥
sup t∈T {cdp Gt }. It remains to show the statement on the strict cohomological
dimension. By the same argument as for cdp , we obtain the inequality scdp G ≥
sup t∈T {scdp Gt } =: n. Assume that n is finite. If all Gt are trivial, so is the
free product. As profinite groups of strict cohomological p-dimension 1 do not
exist, we may assume n ≥ 2. We claim that H m (G, A)(p) = 0 for all discrete
G-modules A and all m ≥ n + 1. By the first part of the proof,
cdp G = sup {cdp Gt } ≤ sup {scdp Gt } = n.
t∈T t∈T
Therefore we may assume that A is torsion-free, hence ZZ-flat. Tensoring the
exact sequence 0 → ZZp → Qp → Qp /ZZp → 0 by A and using the fact that
uniquely divisible modules are cohomologically trivial, we obtain

H m (G, A)(p) ∼
= H m (G, A) ⊗ ZZp ∼
= H m (G, A ⊗ ZZp )
∼
= H m−1 (G, A ⊗ Qp /ZZp )
M0
∼
= H m−1 (Gt , A ⊗ Qp /ZZp ).
t∈T
For each t ∈ T , we have H m−1 (Gt , A ⊗ Qp /ZZp ) ∼
= H m (Gt , A)(p) = 0. Hence
H m (G, A)(p) = 0, and scdp G = n. 2
Exercise 1. Let (G, p, T ) be a bundle of pro-c-groups and assume that for some t ∈ T we
are given a homomorphism f : Gt → H into a finite group H ∈ c. Show that there exists a
closed and open neighbourhood U of t in T and a morphism F : GU → H extending f , i.e.
F |Gt = f .
Hint: By I §1, ex.2 there exists a continuous map F : G → H extending f . The two
maps from G ×T G → H which are given by (g1 , t) × (g2 , t) 7→ F (g1 , t) · F (g2 , t) and by
(g1 , t) × (g2 , t) 7→ F ((g1 , t) · (g2 , t)) coincide on an open neighbourhood of Gt × Gt ⊆ G ×T G.
Exercise 2. Assume that the profinite space T is the topological quotient of the profinite space
T 0 by the action of the profinite group P . Let G be a bundle of profinite groups over T and let
G 0 := G ×T T 0
be the pull-back of G to T . We set G := ∗ G and G0 := ∗ G 0 . Then for every abelian group A,
0
T T0
the cohomology groups H i (G0 , A) are discrete P -modules in a natural way. Show that there
exist isomorphisms for all i
H i (G0 , A) = IndP H i (G, A).
The next exercises deal with profinite operator groups. For more information about operator
groups, we refer the reader to [256] and [136].
Exercise 3. Let Γ be a profinite group, G a pro-c-group and let
φ : Γ → Aut(G)
be a homomorphism. Show that the action: Γ × G → G, (γ, g) 7→ φ(γ)(g), is continuous
if and only if G possesses a system of neighbourhoods of the identity consisting of open
Γ -invariant normal subgroups.
Exercise 4. Prove the assertion that the free pro-c-Γ operator group ∗Γ Fr is equal to the
maximal pro-c-factor group of the kernel of the surjection Fr ∗ Γ Γ , where ∗ denotes the
free product in the category of profinite groups.
Hint: See [94] Satz 3.4.
Exercise 5. Let p be a prime number and let Γ = ZZp be the additive group of p-adic integers.
Let G be a pro-p-Γ operator group. Show that the following conditions are equivalent:
(i) G is a free pro-p-Γ operator group of rank r,
(ii) G is a free pro-p-group, the fixed module (Gab )Γ is trivial and the cofixed module Gab
Γ
is a free ZZp -module of rank r.
Hint: In order to show the difficult implication (ii) ⇒ (i), choose a homomorphism
φ : ∗Γ Fr → G
∼
which induces an isomorphism ZZrp → Gab Γ . Use (1.6.14) and the topological Nakayama
lemma (see 5.2.18) in order to show that φ is surjective. Then apply Γ -homology to the exact
sequence 0 → (ker φ)ab G →Z Zp [[Γ ]]r → Gab → 0 and observe that H1 (Γ, Gab ) = (Gab )Γ = 0.

Chapter V
Iwasawa Modules
The Iwasawa algebra, usually denoted by the Greek letter Λ, is the complete
group algebra ZZp [[Γ ]] of a group Γ , which is noncanonically isomorphic to
∼ Z
ZZp . This means that we will not specify a particular isomorphism φ : Γ → Zp
or, equivalently, we will not fix a topological generator γ of the procyclic
group Γ .
We can view Λ from many different algebraic aspects. By definition, it is

a complete group algebra and we will see that, choosing a generator γ ∈ Γ ,
it can be identified with a power series ring over ZZp . However, Λ is also a
commutative, two-dimensional regular local ring, it is a complete local ring
and it is a compact ZZp -algebra. Using all these properties of Λ, we will develop
the structure theory for Λ-modules, so-called Iwasawa modules. With an eye
on applications to generalized Iwasawa algebras, we will treat each aspect of
Λ separately and in greatest possible generality as long as this does not require
additional work. We have, however, omitted the interpretation of the elements
of Λ as ZZp -valued measures on Γ . For this point of view, which is of particular
importance for the analytic part of Iwasawa theory, we refer the reader to [122]
or [246].
In the first and third sections (after preparing some basic properties of
complete group rings in section 2) we will treat the classification of mod-
ules up to pseudo-isomorphism which was developed by K. IWASAWA and
J.-P. SERRE. To obtain a finer structure theory, we will apply the homotopy
theory of modules, developed in section 4, to the case of Λ-modules. This will
be done in section 5 closely following work of U. JANNSEN. Finally, in section 6
we will investigate some further properties of complete group algebras which
will be needed in the arithmetic applications.
Throughout this chapter, we will assume basic knowledge of commutative

algebra, as can be found in [16] or in many other text books about this subject.

268 Chapter V. Iwasawa Modules
§1. Modules up to Pseudo-Isomorphism
I. In this subsection, let A be a commutative, noetherian and integrally closed

domain with quotient field K. For every prime ideal p in A, one has a canonical
embedding Ap ,→ K and Ap is integrally closed.
Let P (A) be the set of prime ideals of height ht(p) = dim Ap = 1. Since A
is integrally closed, the localization Ap with respect to p ∈ P (A) is a discrete
valuation ring and \
A= Ap
p∈P (A)
(see [16], chap. VII, §1, no.6, th. 4).
(5.1.1) Definition. An A-module M is called reflexive if the canonical map

ϕM : M −→ M ++ = HomA (HomA (M, A), A) ,
m 7−→ ϕM (m) : α 7−→ α(m) ,
of M to its bidual is an isomorphism.
Remark: A reflexive module is torsion-free because the dual M + =

HomA (M, A) of an A-module M is always torsion-free.
If M is a finitely generated and torsion-free A-module, then the localization
Mp of M with respect to a prime ideal p of A is a torsion-free Ap -module and
we have injections
M ,→ Mp ,→ Mp ⊗Ap K = M ⊗A K =: V
M + ,→ (M + )p ,→ (M + )p ⊗Ap K = M + ⊗A K = HomK (V, K) =: V ∧ .
We see that
M+ ∼ = {λ ∈ V ∧ | λ(m) ∈ A for all m ∈ M }
(M + )p ∼
= {λ ∈ V ∧ | λ(m) ∈ Ap for all m ∈ Mp } ∼
= (Mp )+ .
Identifying (M + )p and (Mp )+ , we write Mp+ for this module and consider it
as a submodule of V ∧ .
(5.1.2) Lemma. Let M be a finitely generated torsion-free A-module. Then

\
(i) M + = Mp+ .
p∈P (A)
\
++
(ii) M = Mp ,
p∈P (A)
\
(iii) M = Mp if and only if M is reflexive.
p∈P (A)

§1. Modules up to Pseudo-Isomorphism 269
\
Proof: Let λ ∈ Mp+ . Then for every m ∈ M we get λ(m) ∈ Ap for all
p ∈ P (A), so that λ(m) ∈ A and therefore λ ∈ M + . This proves (i) because the
other inclusion is obvious.
Since Mp is a finitely generated torsion-free module over the discrete valua-
tion ring Ap , p ∈ P (A), Mp is free and Mp → Mp++ is an isomorphism. Via the
identification of V with V ∧∧ , we prove (ii), from which (iii) follows immedi-
ately. 2
(5.1.3) Corollary. If M is finitely generated, then M + is reflexive.
(5.1.4) Definition. A finitely generated A-module M is called pseudo-null if

the following equivalent conditions are fulfilled:
(i) Mp = 0 for all prime ideals p in A of height ht(p) ≤ 1.
(ii) If p is a prime ideal with a = annA (M ) ⊆ p, then ht(p) ≥ 2.
Remarks: 1. We have Mp = 0 if and only if there is an s ∈ A r p such that

sM = 0, hence annA (M ) 6⊂ p. This shows the above equivalence.
2. A pseudo-null module is torsion because M(0) = M ⊗A K = 0.
3. If A is a Dedekind domain, then M is pseudo-null if and only if M = 0.
4. If A is a 2-dimensional, noetherian, integrally closed local domain with
finite residue field, then M is pseudo-null if and only if M is finite. Indeed,
if M is finite, then there exists an r ∈ IN such that mr M = 0, hence supp(M )
⊆ {m},where m denotes the maximal ideal of A. Conversely, if supp(M ) =
{p ∈ Spec A | a ⊆ p} is contained in {m}, then mr ⊆ a for some r ∈ IN and M

is a finitely generated A/mr -module. But A/mr is finite, thus M is finite.
(5.1.5) Definition. A homomorphism f : M → N of finitely generated A-

modules is called a pseudo-isomorphism if ker(f ) and coker(f ) are pseudo-
null or, equivalently, if
∼ N
fp : Mp −→ p
is an isomorphism for all p of height ≤ 1. We write

≈
f : M −→ N.
(5.1.6) Lemma. Let M be a finitely generated A-torsion module and let α ∈ A

be a non-zero element such that supp(A/αA) is disjoint to supp(M ) ∩ P (A).
Then the multiplication on M by α is a pseudo-isomorphism.

Proof: This is clear, since α is a unit in Ap for every p ∈ supp(M ) ∩ P (A)

and Mp = 0 for p = (0) or p ∈ P (A) r supp(M ). 2
For a finitely generated A-module M let

TA (M ) be the torsion submodule and
FA (M ) = M/TA (M ) be the maximal torsion-free quotient of M .
(5.1.7) Proposition. If the A-module M is finitely generated, then

(i) there exists a pseudo-isomorphism
≈
f : M −→ TA (M ) ⊕ FA (M ),
(ii) there exists a finite family (pi )i∈I of prime ideals of height 1 in A, a finite
family (ni )i∈I of natural numbers and a pseudo-isomorphism
≈
A/pni i .
M
g : TA (M ) −→
i∈I
The families (pi ) and (ni ) are uniquely determined by TA (M ) up to

renumbering.
Proof: Let {p1 , . . . , ph } = supp(TA (M )) ∩ P (A). If h = 0, then TA (M ) is

pseudo-null and the maps f : M can FA (M ) and g : TA (M ) −→ 0 are of the
required form. Now let h > 0 and S = hi=1 A r pi = A r hi=1 pi . Then S −1 A
T S
is a semi-local Dedekind domain and therefore a principal ideal domain.

The S −1 A-module S −1 TA (M ) is the torsion module of S −1 M . Using
the structure theorem for modules over principal ideal domains, we see that
S −1 TA (M ) is a direct summand of S −1 M . Since M is finitely generated, we
have
HomS −1 A (S −1 M, S −1 TA (M )) = S −1 HomA (M, TA (M )).
Hence there exists a morphism f0 : M → TA (M ) and s0 ∈ S such that
f0
: S −1 M −→ S −1 TA (M )
s0
is the projector of S −1 M onto its direct summand S −1 TA (M ). Therefore
f0
|
s0 S −1 TA (M )
= idS −1 TA (M ) and thus there exists an s1 ∈ S such that for f1 = s1 f0
f1 |TA (M ) = s1 s0 idTA (M ) .
Now let
f = (f1 , can) : M −→ TA (M ) ⊕ FA (M ) .
The commutative exact diagram

§1. Modules up to Pseudo-Isomorphism 271
0!#$%&'" TA (M ) M FA (M ) 0
f1 |TA (M ) f
0 TA (M ) TA (M ) ⊕ FA (M ) FA (M ) 0
shows ker(f ) = ker(f1 |TA (M ) ) and coker(f ) = coker(f1 |TA (M ) ). But f1 |TA (M ) is
a pseudo-isomorphism by (5.1.6). This proves (i).
In order to prove (ii), let
ri
h M
M n
E := A/pi ij
i=1 j=1
for natural numbers nij such that there exists an isomorphism
g0 : S −1 TA (M ) −→
∼ S −1 E.
We use again the structure theorem for modules over principal ideal domains,
and the fact that S −1 A is a semi-local ring with maximal ideals S −1 pi , i =
1, . . . , h. Using
HomS −1 A (S −1 TA (M ), S −1 E) = S −1 HomA (TA (M ), E),
we obtain a morphism g : TA (M ) → E and an s ∈ S such that g = sg0 . Again

by (5.1.6) we see that g is a pseudo-isomorphism. 2
Remarks: 1. The same argument as in the proof of (ii) shows that for a
pseudo-isomorphism
≈
f : M −→ N
of finitely generated torsion modules there exists a pseudo-isomorphism
≈
g : N −→ M.
Therefore we will use in this case the notion M ≈ N .
For general finitely generated A-modules, the existence of a pseudo-isomor-
≈
phism M −→ N does not imply the existence of a pseudo-isomorphism in the
other direction; see §3, ex.1 for an example.
2. Let
0 −→ M 0 −→ M −→ M 00 −→ 0
be an exact sequence of finitely generated A-torsion modules such that the
associated sets of prime ideals of height 1 of M 0 and M 00 are disjoint. Then
there exists a pseudo-isomorphism
≈
M −→ M 0 ⊕ M 00 .
The proof is similar to the proof of (5.1.7) (ii).

(5.1.8) Proposition. Let M be a finitely generated torsion-free A-module.

Then there exists an injective pseudo-isomorphism of M into a reflexive A-
module M 0 .
Proof: The canonical morphism

ϕM : M −→ M ++
is a pseudo-isomorphism because Mp is a free finitely generated Ap -module,
hence Mp → ∼ M ++ for all p of height ≤ 1. Furthermore, by (5.1.3), M ++ is
p
reflexive and ker(ϕM ) ⊗A K = 0, thus ker(ϕM ) is torsion and therefore zero.
2
II. Now let A be a 2-dimensional regular local ring ∗) . The following

proposition is essential for the structure theory of Iwasawa modules.
(5.1.9) Proposition. Let A be an n-dimensional regular local ring, 2 ≤ n < ∞,

let (p1 , . . . , pn ) be a regular system of parameters generating the maximal ideal
of A and let p0 := 0. For a finitely generated A-module M , the following
(i) For every i = 0, . . . , n − 2, the A/(p0 , . . . , pi )-module M/(p0 , . . . , pi )M
is reflexive.
(ii) M is a free A-module.
In particular, a reflexive A-module M over a 2-dimensional regular local ring
A is free.
Proof (DIEKERT [40]): In order to prove the nontrivial implication we assume

(i). In particular, M is reflexive, hence torsion-free. Therefore multiplication
by p1 is injective on M . If ϕ : ( Ar M is a minimal free presentation of M ,
then we obtain the following commutative exact diagram
0.+,-)*/2301
p1
Ar Ar (A/p1 )r 0
ϕ ϕ ϕ̄
p1
0 M M M/p1 0.
∗) For the definition and properties of regular local rings and regular systems of parameters
see [129].

§2. Complete Group Rings 273
Assume that M/p1 is a free A/p1 -module. Then Nakayama’s lemma im-
plies that ϕ̄ is an isomorphism, hence multiplication by p1 on ker(ϕ) is an
isomorphism. Again by Nakayama’s lemma, we obtain ker(ϕ) = 0.
It remains to show that M/p1 is a free A/p1 -module. We are reduced to the case
n = 2 since for n > 2 we argue by induction applied to the (n − 1)-dimensional
regular local ring A/p1 , the regular system of parameters (p̄2 , . . . , p̄n ), where
p̄i = pi + p1 A, and the module M/p1 .∗)
So let n = 2. Thus A/p1 is regular of dimension 1, i.e. a discrete val-
uation ring, in particular, an integral domain. Therefore the A/p1 -module
HomA (M + , A/p1 ) is torsion-free. Since A is an integral domain, the map
M ++ /p1 = HomA (M + , A) ⊗ A/p1 ,→ HomA (M + , A/p1 )
is injective and, because M is reflexive, we see that M/p1 = M ++ /p1 is a
torsion-free module over the discrete valuation ring A/p1 , so that M/p1 is free.
This finishes the proof of the proposition. 2
From (5.1.7), (5.1.8) and (5.1.9) we obtain the
(5.1.10) Structure Theorem. Let A be a 2-dimensional regular local ring and

let M be a finitely generated A-module. Then there exist finitely many prime
ideals pi , i ∈ I, of height 1, a nonnegative integer r, natural numbers ni ∈ IN
and a pseudo-isomorphism
≈
f : M −→ Ar ⊕ A/pni i .
M
i∈I
The prime ideals pi and the numbers r, ni are uniquely determined by M :

r = dimK M ⊗A K, {pi | i ∈ I} = supp(M ) ∩ P (A).
§2. Complete Group Rings
In this section, let O be a commutative local ring which is complete in its

m-adic topology, where m is the maximal ideal. We further assume that O/mn
is finite for all n; in particular, O is compact. Let p be the characteristic of the
finite residue field k = O/m.∗∗) . Furthermore, let G be a profinite group.
∗) see [129], th. 14.2.
∗∗) In the applications O will always be the ring of integers in a finite extension field K|Q .
p

(5.2.1) Definition. The complete group algebra of G over O is the topolog-

ical inverse limit
O[[G]] := lim O[G/U ],
←−
U
Since O[[G]] is a compact O-algebra, the assignment G → O[[G]] defines

a covariant functor from the category of profinite groups to that of compact
O-algebras. In particular, if N ⊆ G is a closed normal subgroup, then we
4
have an epimorphism O[[G]] O[[G/N ]], whose kernel I(N ) is a two-sided
closed ideal in O[[G]]. It is the closed left (right) ideal generated by the
elements x − 1, x ∈ N . For the particular case N = G, we set
(5.2.2) Definition. The kernel IG := I(G) of the canonical epimorphism, the

augmentation map
5
O[[G]] O,
is called the augmentation ideal of G.∗)
By a (left) O[[G]]-module M , we always understand a separated topological

module, i.e. M carries the structure of a Hausdorff abelian topological group
and the structure of an O[[G]]-module such that the action O[[G]] × M → M
is continuous. In other words, M is a Hausdorff topological O-module with a
continuous G-action. By
MG = M/IG M,
we denote the maximal quotient module of M on which G acts trivially. We
call MG the module of coinvariants of M (cf. II §6). If M is an O[[G]]-module,
then a finitely generated submodule N ⊆ M is the continuous homomorphic
image of the compact module O[[G]]n for some n, and hence closed. Thus the
notions ‘topologically finitely generated’ and ‘finitely generated’ coincide for
O[[G]]-modules, and finitely generated modules are compact.
If we are given a left O[[G]]-module M , we can define a right O[[G]]-
module M 0 keeping the O-module structure and letting g ∈ G act as g −1 .
This establishes an equivalence between the categories of left and right O[[G]]-
modules, and we will sometimes ignore the difference between left and right
modules using this natural equivalence.
The category C of compact O[[G]]-modules and the category D of discrete
O[[G]]-modules will be of particular importance. Both are abelian categories,
∗) The reader should not confuse I ⊆ O[[N ]] with I(N ) ⊆ O[[G]] for a closed normal
N
subgroup N ⊆ G.

and Pontryagin duality defines a contravariant equivalence of categories be-

tween C and D.
(5.2.3) Definition. For a set X we define the free compact O[[G]]-module

with basis X as
F (X) = O[[G]]X =
Y
O[[G]]
x∈X
6
with the product topology. If F (X) M is a continuous surjection onto
a compact O[[G]]-module M , then we call the image of X in M a set of
topological generators.
Remark: The compact O[[G]]-module F (X) has the following universal prop-
erty: for every compact module M ∈ C and every convergent family (mx )x∈X
of elements of M (i.e. for every open neighbourhood U of 0 ∈ M , one has
mx ∈ U for all but finitely many x ∈ X) there exists a unique continuous O[[G]]-
module homomorphism f : F (X) → M such that f (1x ) = mx (compare with
the non-abelian situation in IV §1).
(5.2.4) Proposition. (i) Every compact O[[G]]-module is the projective limit

of finite modules, in particular, it is an abelian pro-p-group. The category C
has sufficiently many projectives and exact inverse limits.
(ii) Every discrete O[[G]]-module is the direct limit of finite modules, in par-
ticular, it is an abelian p-torsion group. The category D has sufficiently many
injectives and exact direct limits.
Proof: Assume that N ∈ D and n ∈ N . Then annO[[G]] (n) is an open ideal

in O[[G]]. Therefore O[[G]] · n ⊆ N is an O/mk [G/U ]-module for some k
and some open normal subgroup U ⊆ G. This shows the first statement of (ii)
and therefore also the first statement of (i) by duality. Hence every M ∈ C
has a convergent set of topological generators. This implies that free compact
modules are projective and that every compact module is the quotient of a
free module. By duality we find that D has sufficiently many injectives. The
statement about the exactness of limits only depends on the underlying abelian
topological groups and is well-known for discrete modules. It follows for
compact modules by duality. 2
(5.2.5) Corollary. A compact O[[G]]-module has a fundamental system of

neighbourhoods of zero consisting of open submodules.

A tensor product for compact O[[G]]-modules is defined by its universal

property. Explicitly, let M be a compact right and N be a compact left
O[[G]]-module. Then the complete tensor product is a compact O-module
M⊗ ˆ O[[G]] N coming along with an O[[G]]-bihomomorphism ∗)
ˆ O[[G]] N
α : M × N −→ M ⊗
with the following property: given any O[[G]]-bihomomorphism f of M × N

into a compact O-module R, there is a unique O-module homomorphism
ˆ → R such that f = g ◦ α.
g : M ⊗N
The complete tensor product is constructed as follows:
ˆ O[[G]] N = lim M/U ⊗O[[G]] N/V,
M⊗
←−
U,V
where U (resp. V ) run through the open O[[G]]-submodules of M (resp. N ).

ˆ is a compact O-module.
Observe that M/U and N/V are finite, so that M ⊗N
The natural bihomomorphisms M × N → M/U ⊗ N/V induce the desired
bihomomorphism α : M × N → M ⊗N ˆ by passing to the limit. The exact
sequence
0 −→ im(M ⊗ V + U ⊗ N ) −→ M ⊗ N −→ M/U ⊗ N/V −→ 0
ˆ is the completion of M ⊗ N in the topology induced by
shows that M ⊗N
taking im(M ⊗ V + U ⊗ N ) as a fundamental system of open neighbourhoods
of 0.
We denote the left derived functors of the (right exact) functor −⊗ ˆ O[[G]] − by
∼
Tor.O[[G]]
(−, −). The canonical isomorphism MG = O⊗ ˆ O[[G]] M induces the
(5.2.6) Proposition. There are canonical isomorphisms for all i ≥ 0 and all
M ∈ C:
Hi (G, M ) ∼
= TorO[[G]]
i (O, M ).
Proof: Since both functors agree for i = 0, it suffices to show that a free
O[[G]]-module F has trivial G-homology and one easily reduces to the case
F = O[[G]]. We have
Hi (G, O[[G]]) = lim
←−
Hi (G/U, O[G/U ]),
U ⊆G
where U runs through the open normal subgroups in G. But the G/U -module
O[G/U ] ∼
= O ⊗ ZZ ZZ[G/U ] is induced, hence homologically trivial. 2
∗) i.e. α is a continuous O-homomorphism such that α(mλ, n) = α(m, λn) for m ∈ M, n ∈
N and λ ∈ O[[G]].

Since C (resp. D) is an abelian category with sufficiently many projectives

(resp. injectives), we have Ext-functors
.
ExtO[[G]] (−, −) : C × C −→ Ab
Ext. O[[G]] (−, −) : D × D −→ Ab
in the usual way. In the following, we will make use of intermediate homo-
morphism groups. Let M ∈ C , N ∈ D and let f : M → N be an O[[G]]-
homomorphism. Then f has finite image which is therefore invariant under an
open ideal in O[[G]]. Thus we have a functor
HomO[[G]] (−, −) : C × D −→ (discrete O-modules)
and we can use either projective resolutions in C or injective resolutions in D
to define the functors
.
ExtO[[G]] (−, −) : C × D −→ (discrete O-modules).
Noting the canonical isomorphism
NG ∼= HomO[[G]] (O, N )
for N ∈ D, we have the
(5.2.7) Proposition. There are canonical isomorphisms

H i (G, N ) ∼
= Exti (O, N ) O[[G]]
for all i ≥ 0 and all N ∈ D.
Proof: For N ∈ D the induced module IndG (N ) is also in D, i.e. carries the
structure of an O[[G]]-module in a natural way. By the arguments of II §6, we
.
therefore see that the functor H (G, −) is universal as a δ-functor on D. The
same is true for ExtiO[[G]] (O, −) and both functors agree in degree 0. 2
(5.2.8) Proposition. If M = lim Mi ∈ C and N = lim Nj ∈ D, then

←− −→
i∈I j ∈J
ExtnO[[G]] (M, N ) = lim ExtnO[[G]] (Mi , Nj )

−→
i,j
for every n ≥ 0.
Proof: It suffices to show that the functor commutes with limits in the first
and in the second variable separately. Let N ∈ D be fixed and represent
every Mi as an inverse limit over its finite quotients: Mi = lim Mi,k . Every
←− k

homomorphism from Mi into N has finite image and therefore factors through
some Mi,k , i.e.
HomO[[G]] (Mi , N ) = lim HomO[[G]] (Mi,k , N ).
−→
k
Thus the statement of the proposition is true for n = 0 in the first variable
if we have a surjective system of finite modules. If i1 ≥ i2 , then for every
7 Mi
finite quotient Mi2 ,k2 of Mi2 , the map Mi1 −→ Mi2 ,k2 factors through
2
some finite quotient Mi1 ,k1 of Mi1 . Therefore we can write M in the form
M = lim←−
Mi,k and we have
i,k
lim
−→
HomO[[G]] (Mi , N ) = lim
−→
HomO[[G]] (Mi,k , N )
i i,k
from what we have already shown. By this consideration, it suffices to consider

the case that all Mi are finite. But then we may change the projective system
Mi to i0 ≥i im(Mi0 → Mi ) again without changing the limits. Thus we have
T
reduced to the case of a system of finite modules with surjective transition maps.
As shown above, in this case Hom(−, −) commutes with inverse limits in the
first argument and using a fixed injective resolution of the second argument
shows the same for Ext(−, −). The proof for limits in the second variable is
formally dual to that for the first variable, and will be omitted. 2
(5.2.9) Corollary. For M ∈ C and N ∈ D, there are canonical isomorphisms

for all i ≥ 0
TorO[[G]]
i (M, N ∨ ) ∼
= ExtiO[[G]] (M, N )∨ ,
where ∨ denotes the Pontryagin dual.
Proof: Let U (resp. V ) run through the open submodules of M (resp. N ∨ ).

Then we obtain by (5.2.8)
∨ ∨
ˆ O[[G]] N ∨
M⊗ = lim M/U ⊗O[[G]] N ∨ /V
←−
U,V
= lim
−→
(M/U ⊗O[[G]] N ∨ /V )∨
U,V
= lim HomO[[G]] (M/U, (N ∨ /V )∨ )

−→
U,V
= HomO[[G]] (M, N ).
This proves the corollary using the methods of homological algebra. 2
From now on we will use the following notational convention:

If a given ring (with unit) A has a natural topology, we tacitly assume that all A-
modules are Hausdorff topological modules. If A is compact, the unspecified
term module will always mean compact A-module.
(5.2.10) Definition. Let A be a ring and let M be an A-module. The projective

dimension pdA M of M is the minimal number n such that there exists a
projective resolution
0 −→ Pn −→ Pn−1 −→ . . . −→ P0 −→ M −→ 0
for M of length n. We set pdA M = ∞ if no such resolution exists. For the
trivial A-module 0, we set pdA 0 = −1.
The projective dimension of the ring A, denoted by pd(A), is defined as
sup{pdA M | M an A-module}.
Observe that simple discrete or compact O[[G]]-modules are finite, i.e. com-
pact and discrete.
(5.2.11) Proposition. For M ∈ C the following assertions are equivalent:

(i) pdO[[G]] M ≤ n,
(ii) Extn+1
O[[G]] (M, N ) = 0 for all simple N ,
(iii) Extn+1
O[[G]] (M, N ) = 0 for all simple N ,
(iv) TorO[[G]]
n+1 (M, N ) = 0 for all simple N .
Proof: Using a projective resolution of M , we have for N simple

Extn+1 n+1
O[[G]] (M, N ) = ExtO[[G]] (M, N ).
Thus (ii) ⇔ (iii). The equivalence (ii) ⇔ (iv) follows from (5.2.9). Us-
ing dimension shifting, the remaining equivalence (i) ⇔ (iii) reduces to the
statement
P ∈ C is projective ⇐⇒ Ext1O[[G]] (P, N ) = 0 for all simple N.
Therefore we have to show that P ∈ C is projective if every exact sequence
0 −→ N −→ N 0 −→ P −→ 0
in C with N simple splits. Assume that 0 → A → B → P → 0 is any exact
sequence in C and consider the collection S of pairs (C, s) consisting of a
closed submodule C ⊆ A and a splitting morphism s : P → B/C such that
8
π ◦ s = idP , where π : B/C P is the canonical projection. Obviously, S

T
is not empty and, since B/ Ci = lim B/Ci , it is an inductively ordered set.
←−
By Zorn’s lemma, there exists a minimal element (C, s) in S. If C =/ 0, we
can find an open submodule C 0 ⊆ C such that C/C 0 is simple. By assumption
there exists a morphism t : P → B/C 0 which makes the diagram
9:;<=> P
t s
0 C/C 0 B/C 0 B/C 0

commutative. The element (C 0 , t) ∈ S is strictly smaller than (C, s). This
contradiction proves that C = 0, thus P is projective. 2
Using (5.2.11) and (5.2.8), we obtain the
(5.2.12) Corollary. For M = lim

←−
Mi ∈ C , one has
i
pdO[[G]] M ≤ sup{pdO[[G]] Mi }.
i
In particular, the inverse limit of projective modules is projective.
(5.2.13) Corollary. pdO[[G]] O = cdp G.
Proof: From (5.2.7) and (3.3.2) we obtain the inequality pdO[[G]] O ≤ cdp G.
The other inequality does not follow directly. The difficulty (which does not
occur for O ∼= ZZp ) lies in the fact that a simple G-module need not be an
O[[G]]-module. We overcome the problem as follows. Let N be a simple
G-module with pN = 0. Then N is a finite dimensional IFp -vector space. By
our assumptions, k = O/m is a finite extension of IFp . Then N ⊗IFp k is an
O[[G]]-module and we obtain for all i
H i (G, N ) ⊗IF k ∼
= H i (G, N ⊗IF k) ∼
p = Exti (O, N ⊗IF k).
p O[[G]] p
i
Hence H (G, N ) = 0 for i > pdO[[G]] O. This finishes the proof. 2
Now we are in the position to identify continuous cochain cohomology of a

compact O[[G]]-module with Ext-groups.
(5.2.14) Proposition. For M ∈ C , we have isomorphisms

i
Hcts (G, M ) ∼
= ExtiO[[G]] (O, M ).
for every i ≥ 0.

Proof: Consider, for every open normal subgroup U ⊆

.
G, the complex P U
of free O[G/U ]-modules given by
PnU = O[(G/U )n+1 ]
with differentials dn : PnU → Pn−1
U
defined for n > 0 by
n
(−1)i (g0 , . . . , gi−1 , gi+1 , . . . , gn ).
X
dn (g0 , . . . , gn ) =
i=0
.
The complex P U is a free resolution of the trivial O[G/U ]-module O. For a
finite O[G/U ]-module A, we have a natural isomorphism of complexes
Hom .
(P U , A) ∼
O[G/U ] .
= C (G/U, A), .
.
where C (G/U, A) is the homogeneous cochain complex of G/U with coef-
ficients in A as defined in I §2. Passing to the inverse limit over U , we obtain
. .
from the complexes P U a complex P which is a resolution of the trivial
O[[G]]-module O by compact O[[G]]-modules. Since every compact O[[G]]-
module is the inverse limit of finite ones, we obtain natural isomorphisms of
complexes for every M ∈ C
Hom .
(P , M ) ∼
O[[G]] .
= C (G, M ), .
cts
.
where Ccts (G, M ) is the continuous homogeneous cochain complex of G with
coefficients in M as defined in II §7. By definition, the cohomology of the
complex on the right is continuous cochain cohomology. Thus it remains to
show that Pn is a projective O[[G]]-module for every n ≥ 0, because then the
cohomology of the left complex is the required Ext-group.
In order to prove that Pn is projective, it suffices to show that Ext1O[[G]] (Pn , N )
vanishes for every finite simple N (see (5.2.11)). Let U ⊆ G be open and
normal. A class x ∈ Ext1O[[G]] (PnU , N ) corresponds to an extension of O[[G]]-
modules
0 −→ N −→ N 0 −→ PnU −→ 0.
Since N is finite, N 0 is an O[G/V ]-module for some open subgroup V ⊆ U
which is normal in G. Consider the pull-back of the above sequence via the
natural surjection
PnV → PnU .
We obtain an exact sequence of O[G/V ]-modules which splits because PnV is
a free O[G/V ]-module. In other words, the image of x in Ext1O[[G]] (PnV , N )
vanishes. Using a projective resolution of the first argument, we see that
for finite N and compact A the groups ExtO[[G]] (A, N ) and ExtO[[G]] (A, N )
coincide. By (5.2.8), we therefore obtain
Ext1 (Pn , N ) ∼
O[[G]] = lim Ext1 (P U , N ) = 0.
−→ O[[G]] n 2
U

The next result is the natural analogue of the universal coefficient theorem
for abstract groups (compare, for example, [79], VI, th.15.1).
(5.2.15) Theorem (Universal Coefficient Theorem). Denoting the homology

group Hn (G, O) by Hn (G), the following holds:
(i) Let N be a discrete O-module considered as an element in D with G
acting trivially on N . Then we have a natural cohomological spectral
sequence
E2pq = ExtpO (Hq (G), N ) ⇒ H p+q (G, N ).
In particular, if O is a discrete valuation ring, then we obtain a natural

exact sequence for all n:
0 −→ Ext1O (Hn−1 (G), N ) −→ H n (G, N ) −→ HomO (Hn (G), N ) −→ 0.
(ii) Let M be a compact O-module considered as an element in C with

G acting trivially on M . Then we have a natural homological spectral
sequence
2
Epq = TorO
p (Hq (G), M ) ⇒ Hp+q (G, M ).
In particular, if O is a discrete valuation ring, then we obtain a natural

exact sequence for all n:
ˆ O M −→ Hn (G, M ) −→ TorO
0 −→ Hn (G)⊗ 1 (Hn−1 (G), M ) −→ 0.
.
Proof: Let P → O be a compact O[[G]]-projective resolution of O (concen-
.
trated in negative degrees) and let N → I be a discrete O-injective resolution
of N . Consider the spectral sequence associated to the double complex
Apq = HomO[[G]] (P −q , I p ) = HomO (PG−q , I p ).
By (2.2.4), its limit term E n is ExtnO[[G]] (O, N ) ∼
= H n (G, N ) (see (5.2.7)). Its
initial terms E2pq are easily computed as ExtpO (Hq (G), N ), which shows the
spectral sequence in (i). If O is a discrete valuation ring, then pd O = 1.
Therefore E2pq = 0 for p ≥ 2, and the spectral sequence induces the asserted
short exact sequences. This proves (i).
The proof of (ii) is similar: one chooses a compact O-projective resolution
.
Q → M and considers the double complex
Apq = P −q ⊗ ˆ O[[G]] Q−p = PG−q ⊗
ˆ O Q−p . 2
Remark: As in the case of abstract groups, one can show that the exact
sequences in (5.2.15) split by an unnatural splitting.

Recall that O/mn is finite for all n. Therefore O[[G]] is an inverse limit of
finite discrete (hence artinian) rings and the ideals
mn O[[G]] + I(U ), n ∈ IN, U ⊆ G open normal,
are a fundamental system of neighbourhoods of 0 ∈ O[[G]]. We denote by
RadG ⊆ O[[G]] the radical of O[[G]], i.e. the inverse limit of the radicals of
O/mn [G/U ]. ∗) Then RadG is a closed two-sided ideal which is the intersection
of all open left (right) maximal ideals. The powers (RadG )n , n ≥ 1, define a
topology on O[[G]], which we will call the R-topology.
(5.2.16) Proposition. (i) The R-topology is finer than the canonical topology
on O[[G]]; in particular, it is Hausdorff.
(ii) The following assertions are equivalent:
a) RadG ⊆ O[[G]] is open.
b) O[[G]] is a semi-local ring.
c) (G : Gp ) < ∞, where Gp is a p-Sylow subgroup in G.
If assertions (a)–(c) hold, then the finitely many left (right) maximal ideals of
O[[G]] are open.
(iii) O[[G]] is a local ring if and only if G is a pro-p-group. In this case, the
maximal ideal of O[[G]] is equal to mO[[G]] + IG .
Proof: For arbitrary n and U , the radical of O/mn [G/U ] is nilpotent, since
this ring is artinian ([17], chap.8, §6, no. 4, th. 3). This shows (i).
In order to prove a) ⇒ b), assume that a) holds. Then O[[G]]/RadG is
finite, hence there are only finitely many open left maximal ideals in O[[G]],
M1 , . . . , Mr say. Let M be any left maximal ideal. If M were not open,
we could find elements x1 , . . . , xr with xi ∈/ Mi , i = 1, . . . , r. The elements
x1 , . . . , xr generate a left ideal, I say, which is necessarily closed being a
homomorphic image of a free (hence compact) module of finite rank. Applying
(5.2.5) to the module O[[G]]/I, we see that I must be contained in an open
left maximal ideal. But this is obviously not possible, hence M must be open.
This shows a) ⇒ b) and the final assertion of (ii), which obviously also holds
for the right maximal ideals. The implication b) ⇒ a) is trivial.
To show a) ⇒ c) assume that RadG is open and choose n and U such that
n
m O[[G]] + I(U ) ⊆ RadG . We conclude that for every open V ⊆ U which is
normal in G and for every u ∈ U the image of u − 1 in O/m[G/V ] is contained
r r
in the radical, hence is nilpotent. This implies (u − 1)p ≡ up − 1 ≡ 0, i.e.
r
up ∈ V , for some r. Hence U is a pro-p-group, showing a) ⇒ c). If O[[G]] is
local, then RadG is open and maximal, hence RadG ⊇ m O[[G]] + IG . Now the
∗) The radical of an abstract ring is the intersection of all left (right) maximal ideals.

argument above shows that G is a pro-p-group. Thus we proved the ‘only if’
part of (iii).
In order to show c) ⇒ a), let (G : Gp ) < ∞. Then the intersection U of all
(finitely many) conjugates of Gp is an open normal subgroup of G. We will
prove that m O[[G]] + I(U ) is contained in RadG . Let M ⊆ O[[G]] be a left open
maximal ideal. Then N = O[[G]]/M is a simple O[[G]]-module; in particular,
N is finite and p-torsion by (5.2.4). We conclude that mn N = 0 for some n and
thus mN = 0 because N is simple. Hence m ⊆ M, and so mO[[G]] ⊆ RadG .
Let V ⊆ U be any open subgroup which is normal in G. We follow
the argument given in the proof of [32], (5.26): the augmentation ideal of
the artinian ring O/m[U/V ] is the only left maximal ideal, hence equal to
the radical and therefore nilpotent. Using the identity g(u − 1)g 0 (u0 − 1) =
0
gg 0 (ug − 1)(u0 − 1), we see that the image of I(U ) in O/m[G/V ] is a nilpotent
ideal, hence contained in Rad(O/m[G/V ]). Varying V and passing to the
limit, we obtain I(U ) ⊆ RadG , showing the implication c) ⇒ a).
If G is itself a pro-p-group, the arguments above show mO[[G]] + IG ⊆ RadG .
But mO[[G]] + IG is an open maximal ideal, so that O[[G]] is a local ring,
showing the remaining assertion of (iii). 2
Now let M be a compact O[[G]]-module. Then, in addition to the given

topology, there are two other topologies on M :
1. The topology given by the sequence of submodules {mn M + I(U )M }n,U ,
where n ∈ IN, and U runs through the open normal subgroups of G. We
call this topology the (m, I)-topology.
2. The R-topology, which is given by the sequence of submodules
{(RadG )n M }n∈IN .
By (5.2.16), the R-topology is finer than the (m, I)-topology.
(5.2.17) Proposition. (i) The (m, I)-topology is finer than the original topology
on M . In particular, the (m, I)- and the R-topology are Hausdorff.
(ii) If M is finitely generated, then the (m, I)-topology coincides with the
original topology on M .
Proof: Assume that N ⊆ M is an open submodule. Then by continuity, for

every x ∈ M there exists a neighbourhood Vx ⊆ M of x, an nx ∈ IN and an
open Ux ⊆ G, such that (mnx O[[G]] + I(Ux ))Vx ⊆ N . Since M is compact, it
is covered by finitely many Vx1 , . . . , Vxr and therefore

mn O[[G]] + I(U ) M ⊆ N

for n = max(nx1 , . . . , nxr ) and U = ri=1 Uxi . This shows (i).

T
If M is finitely generated, then there exists a surjection

? r

O[[G]] M with (m, I)-topology
for some r ∈ IN, which is automatically continuous, as O[[G]] carries the (m, I)-
topology. This shows that M with the (m, I)-topology is quasi-compact and
Hausdorff by (i). Hence the identity map (M with (m, I)-topology) → M is a
continuous bijection between compact spaces and therefore a homeomorphism.
2
The following lemma is called the Topological Nakayama Lemma for

complete group rings.
(5.2.18) Lemma. (i) If M ∈ C and RadG M = M , then M = 0.

(ii) M ∈ C is generated by x1 , . . . , xr if and only if xi + RadG M , i = 1, . . . , r,
generate M/RadG M as an O[[G]]/RadG -module.
Proof: (i) By (5.2.17), the R-topology is Hausdorff. Hence

∞
(RadG )n M = M.
\
0=
n=1
In order to show the nontrivial implication in (ii), assume that we have
x1 , . . . , xr ∈ M such that xi + RadG M , i = 1, . . . , r, generate M/RadG M .
The map r
M id
ϕ: O[[G]] ei −→ (M with (m, I)-topology) −→ M ,
i=1
ei 7−→ xi ,
is continuous by (5.2.17) (i). Since O[[G]] is compact, the image N of ϕ is
closed and therefore M/N ∈ C . But by construction
RadG (M/N ) = (RadG M + N )/N = M/N ,
hence (i) yields M = N , i.e. ϕ is surjective. 2
(5.2.19) Corollary. Assume that the profinite group G is topologically finitely

generated and that (G : Gp ) < ∞, where Gp is a p-Sylow subgroup in G. Then
the R-topology coincides with the canonical topology on the semi-local ring
O[[G]]. For any compact O[[G]]-module M the R- and the (m, I)-topology
coincide. If M is finitely generated, both coincide with the original topology
on M .

Proof: By (5.2.16), the R-topology is finer than the canonical topology, and
O[[G]] is semi-local. Furthermore, RadG is open. In order to show that both
topologies coincide, we have to show that all powers of RadG are open. Choose
n ∈ IN and an open normal subgroup U ⊆ G with
I := mn O[[G]] + I(U ) ⊆ RadG .
It suffices to show that the powers of I are open. As m/m2 ⊆ O/m2 is finite, the
topological Nakayama lemma (5.2.18) (for G = 1) implies that m and hence
also mn is finitely generated. If G is topologically finitely generated, then so
is U and therefore I(U ) is a finitely generated ideal in O[[G]]. We conclude
that I and all its powers are finitely generated; in particular, they are closed
ideals in O[[G]]. It therefore remains to show that I m has finite index in O[[G]]
for all m. To begin with, the ring O[[G]]/I is finite. Furthermore, for all m,
the ideal I m is finitely generated. Hence I m /I m+1 is finitely generated over
O[[G]]/I, and thus finite. Now the assertion follows by induction. By what we
have just seen, the R- and the (m, I)-topology coincide on any O[[G]]-module.
If, finally, M is finitely generated, then, by (5.2.17), the original topology on
M coincides with the (m, I)-topology. 2
(5.2.20) Corollary. Let G be a pro-p-group and let P ∈ C be finitely generated.

Then P is a free O[[G]]-module if and only if P is projective.
Proof: For the nontrivial implication let P be projective and let P/MP ∼
= Kr,
where M is the maximal ideal of the local ring O[[G]] and K = O[[G]]/M.
By (5.2.18) (ii), we get a surjection
@
ϕ : O[[G]] r
P,
so that O[[G]]r ∼
= P ⊕ ker(ϕ) and ker(ϕ)/M ker(ϕ) = 0. Now (5.2.18) (i)
yields ker(ϕ) = 0. 2
(5.2.21) Corollary. Let G be a finite p-group and let P be a finitely generated

O[G]-module which is free as an O-module and cohomologically trivial as a
G-module. Then P is a free O[G]-module.
Proof: From (1.8.4) it follows that P is O[G]-projective (actually in that

proposition we considered ZZ[G]-modules, but the proof is the same for O[G]-
modules). Now the result follows from the previous corollary. 2

We will finish this section with some considerations about abstract O[[G]]-
modules. We saw in (5.2.17) that a given finitely generated compact O[[G]]-
module always carries the (m, I)-topology. However, if we consider an abstract
finitely generated O[[G]]-module M , it is not clear whether it can be endowed
with a topology such that it becomes a compact O[[G]]-module. By (5.2.17),
this is possible if and only if the (m, I)-topology on M is Hausdorff. This
is always true if M is projective, because then it is a direct summand in a
free module of finite rank and the (m, I)-topology on O[[G]] is Hausdorff by
definition. We make the following
(5.2.22) Definition. Let A be a ring (with unit) and let M be an A-module

(abstract, resp. compact if A is a topological ring). Then M is called finitely
presented if there exists an exact sequence (a presentation of M )
P1 −→ P0 −→ M −→ 0
with finitely generated (abstract, resp. compact) projective A-modules P1 and
P0 .
If M is a finitely presented abstract O[[G]]-module and P1 → P0 → M → 0 is

a presentation, then P1 and P0 are naturally endowed with the (m, I)-topology
and they are compact. Furthermore, every O[[G]]-homomorphism is automati-
cally continuous for the (m, I)-topology. Hence the image of P1 in P0 is closed
and we can give M the quotient topology. Then, a posteriori by (5.2.17), this
topology is the (m, I)-topology which is therefore Hausdorff. Thus we have
proved the
(5.2.23) Proposition. The forgetful functor from compact O[[G]]-modules to

abstract O[[G]]-modules defines an equivalence of categories:
   

 finitely presented 
 
 finitely presented 

compact O[[G]]-modules abstract O[[G]]-modules

 
 
 

1−1
←→


 with continuous 




 with abstract 


O[[G]]-homomorphisms O[[G]]-homomorphisms
   
.
If we assume that O[[G]] is noetherian (e.g. O = ZZp and G is a compact

Lie group over Qp , cf. [126], V 2.2.4), then every finitely generated module is
finitely presented and the category of finitely generated modules has sufficiently
many projectives. Therefore we can calculate Tor and Ext in either of the
categories (compare exercise 1), and also the topological projective dimension
coincides with the usual one.

Remark: Most of the material of this section is contained in the article [19] by
A. BRUMER, where the slightly more general notion of pseudocompact algebras
is considered.
Exercise 1. If one of the compact O[[G]]-modules M and N is finitely presented, then

M⊗ ˆ O[[G]] N ∼
= M ⊗O[[G]] N.
Exercise 2. If M = lim Mi and N = lim Nj are in C, then

←− ←−
i∈I j ∈J
TorO[[G]]
n (M, N ) = lim TorO[[G]]
n (Mi , Nj )
←−
i,j
for all n ∈ IN.

Exercise 3. For a compact right O[[G]]-module M and a compact left O[[G]]-module N ,
define a continuous G-action on M ⊗ˆ O N by g(m ⊗ n) = mg −1 ⊗ gn. Show that there is a
natural spectral sequence of homological type
2
= Hi G, TorjO (M, N ) ⇒ TorO[[G]]

Ei,j i+j (M, N ).
Exercise 4. Let H be a closed normal subgroup of G. Show the existence of a spectral

sequence for M ∈ C and N ∈ D of cohomological type
E2i,j = H i G/H, ExtjO[[H]] (M, N ) ⇒ Exti+j

O[[G]] (M, N ).
Exercise 5. Show that
pd O[[G]] = pd O + cdp G,
where p = char(O/m). In particular,
pd O[[G]] = pd O[[Gp ]]
if Gp is a p-Sylow subgroup in G.
Exercise 6. (Topological Nakayama Lemma) Let A be a local ring with maximal ideal m
which is compact in its m-adic topology. Let M be a compact A-module. Then the following
is true.
(i) The m-adic topology of M is finer than the given one. In particular, M is Hausdorff
with respect to the m-adic topology.
(ii) If M is finitely generated, then both topologies coincide.
(iii) If mM = M , then M = 0.
(iv) M is finitely generated if and only if M/mM is a finite dimensional A/m-vector space.
Hint: Imitate the proofs of (5.2.17) and (5.2.18).
Exercise 7. (Generalization of Maschke’s theorem) Let H be a closed, normal subgroup
in G of (not necessarily finite) index prime to p = char(O/m). Then a finitely generated
O[[G]]-module is projective if and only if it is O[[H]]-projective.
Hint: See (2.6.11) and use the spectral sequence in exercise 4.

§3. Iwasawa Modules 289
§3. Iwasawa Modules
As in the last section, we assume that O is a commutative noetherian local

ring with maximal ideal m, finite residue field k = O/m and complete in
its m-adic topology. Let O[[T ]] be the power series ring in one variable
over O. Then O[[T ]] is a local ring with maximal ideal (m, T ), residue field
O[[T ]]/(m, T ) = k, noetherian (see [16], chap.III, §2, no.10) and complete
with respect to its (m, T )-adic topology (loc. cit. no.6).
A well-known and useful technical result is the so-called division lemma.
For the proof, we refer the reader to [16], chap.VII, §3, no.8.
P∞
(5.3.1) Division Lemma. Let f = n=0 an T n ∈ O[[T ]] and let
s := inf{n | an ∈
/ m}
be finite. The number s is called the reduced degree of f . Then every
g ∈ O[[T ]] can be written uniquely as
g = fq + r
with q ∈ O[[T ]] and a polynomial r ∈ O[T ] of degree ≤ s − 1. In particular,
O[[T ]]/(f ) is a free O-module of rank s with basis {T i mod f | i = 0, . . . ,
s − 1}.
(5.3.2) Definition. A polynomial F ∈ O[T ] is called a Weierstraß polyno-

mial if it is of the form
F = T s + as−1 T s−1 + · · · + a1 T + a0
with coefficients a0 , . . . , as−1 contained in m.
(5.3.3) Corollary. Let F be a Weierstraß polynomial. Then the injection

O[T ] ,→ O[[T ]] induces an isomorphism
O[T ]/F O[T ] −→ O[[T ]]/F O[[T ]].
Proof: Let s = deg(F ). Then s is the reduced degree of F . Using (5.3.1),

ABC
O[T ]/(F ) O[[T ]]/(F )
s−1
X
T iO
i=0
gives the result. 2

(5.3.4) Weierstraß Preparation Theorem. Let f ∈ O[[T ]] with finite reduced

degree s. Then there exists a unique decomposition
f =F ·u
into a Weierstraß polynomial F of degree s and a unit u ∈ O[[T ]]. Furthermore,
F is the characteristic polynomial of the endomorphism on the free O-module
O[[T ]]/(f ) given by the multiplication by T .
Proof: We apply the division lemma to f and T s : There exists a unique

v ∈ O[[T ]] and a unique polynomial G = s−1 i
P
i=0 ai T such that
T s = f · v − G.
Since f has reduced degree s and
T s + ās−1 T s−1 + · · · + ā0 = f¯ · v̄
(here ¯ denotes the reduction mod m), it follows that āi = 0 for all i =
0, . . . , s − 1, and the constant term of v is a unit in O. Therefore v ∈ O[[T ]]×
and T s + G is a Weierstraß polynomial. Using corollary (5.3.3), we obtain
O[[T ]]/(f ) = O[[T ]]/(F ) ∼
= O[T ]/(F )
thus proving the last assertion. 2
Now let p = char(O/m) and assume that Γ is a free pro-p-group of rank 1,

i.e. Γ is (noncanonically) isomorphic to the additive group ZZp .
(5.3.5) Proposition. Assume that γ is a topological generator of Γ ∼

= ZZp .
Then the map
∼ O[[Γ ]] ,
O[[T ]] −→
T 7−→ γ − 1 ,
is an isomorphism of topological O-algebras.
Proof: Consider the Weierstraß polynomials

n −1
pX
pn pn −i
!
pn pn
ωn = (T + 1) −1=T + T , n ≥ 0,
i=1 i
and let Γn be the unique subgroup of Γ of index pn . Using corollary (5.3.3),
we see that the map
∼ O[T ]/(ω ) −→ O[Γ/Γ ] ,
O[[T ]]/(ωn ) −→ n n
T mod ωn 7−→ γ − 1 mod Γn ,

is an isomorphism of O-algebras with inverse map γ mod Γn 7→ T +1 mod ωn .

Since n n
ωn+1 = ωn ((T + 1)p (p−1) + · · · + (T + 1)p + 1) ,
DEFG
O[[T ]]/(ω n+1 ) O[Γ/Γn+1 ]
O[[T ]]/(ωn ) O[Γ/Γn ]

and hence an isomorphism
lim
←−
O[[TH ]]/(ωn ) lim
←−
O[Γ/Γn ] = O[[Γ ]].
n n
Finally, the natural homomorphism

I ]]
O[[T lim O[[T ]]/(ωn )
←−
n
is an isomorphism: the compactness of O[[T ]] implies its surjectivity and the
inclusions
ωn O[[T ]] ⊆ (m, T )n+1
ωn O[[T ]] is zero. Thus we obtain the desired result.
T
show that its kernel n
2
Now we specialize to the case O = ZZp .
(5.3.6) Definition. We call the complete group ring Λ = ZZp [[Γ ]] the Iwasawa
algebra and a compact Λ-module an Iwasawa module.
By (5.3.5), the Iwasawa algebra Λ is isomorphic to the power series ring

ZZp [[T ]]. The isomorphism depends on the choice of a topological generator
γ of Γ . In the following we will identify ZZp [[Γ ]] with ZZp [[T ]] using any fixed
generator γ.
(5.3.7) Lemma. The prime ideals of height 1 in Λ are

p = (p) and p = (F ),
where F is an irreducible Weierstraß polynomial over ZZp .
Proof: Since ZZp [[Γ ]] is factorial, the prime ideals of height 1 are of the form
p = (f ) where f is an irreducible element in Λ. Let (f ) =/ (p); then the reduced

element f¯ ∈ ZZ/pZZ[[Γ ]] is not trivial, and using the Weierstraß preparation

theorem (5.3.4) we get
(f ) = (F ), F an irreducible Weierstraß polynomial over ZZp .
But a Weierstraß polynomial is irreducible in ZZp [[T ]] if and only if it is
irreducible in ZZp [T ], see [16], chap.VII, §3.8, cor. of prop. 7. 2
Applying the structure theorem (5.1.10) to Λ and using (5.3.5) and remark 4
after (5.1.4), we obtain the
(5.3.8) Structure Theorem for Iwasawa Modules. Let M be a finitely gen-

erated Iwasawa module. Then there exist irreducible Weierstraß polynomials
Fj , numbers r, mi , nj , and a homomorphism
s t
≈ r
M
mi
M n
M −→ Λ ⊕ Λ/p ⊕ Λ/Fj j
i=1 j=1
with finite kernel and cokernel. The numbers r, mi , nj and the prime ideals
Fj Λ are uniquely determined by M .
(5.3.9) Definition. With the notation of (5.3.8) we call

r(M ) = rankΛ (M ) = r the Λ-rank of M ,
s
X
µ(M ) = mi the Iwasawa µ-invariant of M ,
i=1
t
X
λ(M ) = nj deg(Fj ) the Iwasawa λ-invariant of M ,
j=1
t
Y n
FM,γ = Fj j the characteristic polynomial of M .
j=1
Furthermore, we call a finitely generated Λ-module of the form

s t
n
E = Λr ⊕ Λ/pmi ⊕
M M
Λ/Fj j
i=1 j=1
an elementary Λ-module.
Remarks: 1. The invariants defined above depend on M only up to pseudo-

isomorphism and r(M ), µ(M ) and λ(M ) are independent of the chosen gener-
ator γ, in contrast to the characteristic polynomial. Furthermore, observe that
FM,γ = FTΛ (M ),γ , where TΛ (M ) is the Λ-torsion submodule of M .

2. The invariants µ(M ) and λ(M ) are additive and FM,γ is multiplicative in
short exact sequences of finitely generated Λ-torsion modules. Furthermore, a
finitely generated Λ-torsion module M is finite if and only if λ(M ) = 0 = µ(M ).
3. Let M be a finitely generated Λ-torsion module, then
λ(M ) = dimQp (M ⊗ ZZp Qp )
and FM is the characteristic polynomial of the endomorphism on the Qp -vector
space M ⊗ ZZp Qp given by multiplication by T .
As before, MΓ denotes the module of Γ -coinvariants of M :
MΓ = M/IΓ M ∼
= M/T M,
where IΓ is the augmentation ideal of Λ. The topological Nakayama lemma

(5.2.18) implies the
(5.3.10) Proposition. Let M be an Iwasawa module. Then the following

(i) M is a finitely generated Λ-module.
(ii) MΓ is a finitely generated ZZp -module.
(iii) M/mM is a finite-dimensional IFp -vector space.
Very useful is the following
(5.3.11) Lemma. Let

0 −→ M1 −→ M2 −→ M3 −→ 0
be an exact sequence of Λ-modules. Then there is an exact sequence of
ZZp -modules
0 −→ M1Γ −→ M2Γ −→ M3Γ −→(M1 )Γ −→(M2 )Γ −→(M3 )Γ −→ 0 .
Proof: Since the sequences

γ−1
0 −→ MiΓ −→ Mi −→ Mi −→(Mi )Γ −→ 0
are exact for i = 1, 2, 3, the result follows from the snake lemma. 2
We denote the unique subgroup of Γ of index pn by Γn .

(5.3.12) Definition. Let M be an Iwasawa module. Then M δ denotes the

maximal Λ-submodule of M on which Γ acts discretely:
Mδ = M Γn .
[
n
δ δ
Let M0 = tor ZZp M . If M /M0 =/ 0, then let d = d(M ) be the minimal number
such that Γd acts trivially on M δ /M0 . If this module is zero, we put d = −1.
(5.3.13) Definition. We denote the pn -th cyclotomic polynomial by

ωn
ξn = , n ≥ 0,
ωn−1
where we put ω−1 = 1 and n −1
pX
pn pn −i
!
pn pn
ωn = (T + 1) − 1 = T + T , n ≥ 0.
i=1 i
Hence p−1
k−1
(1 + T )ip
X
ωn = ξ0 · ξ1 · · · ξn and ξ0 = ω0 = T, ξk = for k ≥ 1.
i=0
(5.3.14) Lemma. Let M be a finitely generated Λ-module. Then

(i) M δ is a Λ-torsion module and finitely generated as a ZZp -module, thus
d(M ) < ∞.
(ii) M0 is the maximal finite Λ-submodule of M .
(iii) supp(M δ ) ∩ P (Λ) ⊆ {(ξn ) | n ≥ 0}, i.e. the prime ideals of height 1 in the
support of M δ are principal ideals generated by cyclotomic polynomials.
Moreover, there is a pseudo-isomorphism
Mδ ≈
M
Λ/ξni
i
with ni ≤ d(M ).
(iv) (M/M δ )δ is ZZp -torsion-free.
(v) Let, in addition, M be Λ-torsion. Then MΓn is finite for all n if and only
if d(M ) = −1.
Proof: Since M is a finitely generated Λ-module and Λ is noetherian, M δ

is also finitely generated. Therefore there exists an n ∈ IN such that Γn acts
trivially on M δ , showing (i) and (ii). Since ωd(M ) (M δ /M0 ) = 0, we get (iii).
From the exact sequence
0 −→ M δ −→ M −→ M/M δ −→ 0
we obtain (using (5.3.11) and passing to the limit) the exact sequence
id
0 −→ M δ −→ M δ −→(M/M δ )δ −→ lim
−→
(M δ )Γn .
n

The module lim (M δ )Γn is ZZp -torsion-free since the transition maps
−→ n
ωm
(M δJ )Γn ωn
(M δ )Γm , m ≥ n,
coincide with multiplication by pm−n if n is large enough. Hence (M/M δ )δ is
ZZp -torsion-free. This proves (iv).
Finally, (v) follows from the fact that for a finitely generated Λ-torsion
module M , the group MΓn is finite if and only if M Γn is finite. 2
For a finitely generated Λ-module M , we define the Λ-submodule M cycl

of M in the following way. Let M0 be as above, M1 = M δ and we define
inductively
Mi+1 = ker M −→(M/Mi )/(M/Mi )δ for i ≥ 1 .
From the commutative exact diagram
0KLMNOPQRST Mi+1 M (M/Mi )/(M/Mi )δ 0
0 (M/Mi )δ M/Mi (M/Mi )/(M/Mi )δ 0,

we obtain an exact sequence
0 −→ Mi −→ Mi+1 −→(M/Mi )δ −→ 0 for i ≥ 0 .
The submodules Mi are Λ-torsion modules whose support supp(Mi ) ∩ P (Λ)
is contained in {(ξn ) | n ≥ 0}, and (M/Mi+1 )δ = ((M/Mi )/(M/Mi )δ )δ is ZZp -
torsion-free by (5.3.14)(iv). Therefore, since M is finitely generated, the
sequence of submodules
M0 ⊆ M1 ⊆ · · · ⊆ Mi ⊆ Mi+1 ⊆ ···
stabilizes.
(5.3.15) Definition. Let M be a finitely generated Λ-module. Then

M cycl :=
[
Mi .
i
(5.3.16) Lemma. Let M be a finitely generated Λ-module. Then

(i) M cycl is a Λ-torsion module and finitely generated as a ZZp -module with
M0 ⊆ M δ ⊆ M cycl ,
(ii) supp(M cycl ) ∩ P (Λ) = supp(M δ ) ∩ P (Λ) ⊆ {(ξn ) | 0 ≤ n ≤ d(M )},
supp(M/M cycl ) ∩ P (Λ) is disjoint to {(ξn ) | n ≥ 0},
(iii) (M/M cycl )δ = (M/M cycl )cycl = 0.

Proof: The first assertion is obvious. If

n
M ≈ Λr ⊕ Λ/pmi ⊕ Λ/(ξnk )tk ,
M M M
Λ/Fj j ⊕
i j k
where Fj are irreducible Weierstraß polynomials different to ξn for all n ≥ 0,
then
Λ/(ξnk )min(tk ,i) for i ≥ 0 ,
M
Mi ≈
k
and
M cycl ≈ Λ/(ξnk )tk
M
k
with nk ≤ d(M ). This proves (ii).
As we have seen above, the Λ-module (M/M cycl )cycl is ZZp -torsion-free.
Furthermore, supp(M/M cycl ) ∩ P (Λ) is disjoint to the set of prime ideals
{(ξn ) | n ≥ 0} by (ii), and so (M/M cycl )δ ⊆ (M/M cycl )cycl = 0. 2
One classical feature of Iwasawa theory is the description of the asymptotic

behaviour of #MΓn when n → ∞, provided these orders are finite. This finite-
ness is equivalent to d(M ) = −1 as we saw above. The following proposition
covers the case of nonnegative d.
(5.3.17) Proposition. Let M be a finitely generated Λ-torsion module and let

n0 ≥ d(M ) be a fixed number. Then
n +λn+ν
#(M/ ωωnn0 M ) = pµp
for all n large enough, where µ = µ(M ), λ = λ(M ) and ν is a constant not
depending on n.
Proof: Let n ≥ n0 and put

ωn
νn := ω n0
= ξn0 +1 · · · ξn
n n0 (pn−n0 −1)
= 1 + (1 + T )p 0 + · · · + (1 + T )p .
First, we observe that M/νn M is finite for all n ≥ n0 . Indeed, since νn
νn
is disjoint to supp(M ) ∩ P (Λ) for n ≥ n0 , the homomorphism M −→ M is a
pseudo-isomorphism by (5.1.6). Furthermore, we obtain the
Claim: Multiplication by νn is injective on M/M0 .
Applying the snake lemma to the exact sequence
0 −→ M0 −→ M −→ M/M0 −→ 0,
we obtain the exactness of
0 −→ M0 /νn −→ M/νn −→(M/M0 )/νn −→ 0 ,

n ν
since ker(M/M0 −→ M/M0 ) = 0 by the claim, and so
#M/νn = #(M/M0 )/νn · #M0 /νn .
In order to calculate #(M/M0 )/νn , we put N = M/M0 . Using the structure
theorem, we obtain an exact sequence
0 −→ N −→ E −→ C −→ 0,
where C is finite and E is an elementary Λ-module, i.e.
s t
n
Λ/pmi ⊕
M M
E= Λ/Fj j .
i=1 j=1
Since multiplication by νn is injective on E, we have an exact sequence

0 −→ νn C −→ N/νn −→ E/νn −→ C/νn −→ 0,
n ν
where νn C is defined by the exact sequence 0 → νn C → C → C → C/νn → 0.
It follows that
#N/νn = #E/νn .
n
Let Ei = Λ/pmi . Obviously, #Ei /ωn = pmi p , and the exact sequence
n ν
0 −→ Ei /ωn0 −→ Ei /ωn −→ Ei /νn −→ 0
shows that n −pn0 )
#Ei /νn = pmi (p .
For the other summands of E, we need the
(5.3.18) Lemma. Let M be a finitely generated Λ-torsion module which is

free of rank λ as a ZZp -module. Then
ωn+1 λ(λ − 1)
M = pM for n > .
ωn 2
We proceed with the proof of (5.3.17). If Ej = Λ/Fj (T )nj , then the sequence
ωn+1
ωn
0 −−−→ Ej /νn −−−→ Ej /νn+1 −−−→ Ej /p −−−→ 0
ωn+1
is exact for n 0 by (5.3.18) and the fact that multiplication by ξn+1 = ωn
is
injective on Ej . Therefore
#Ej /νn+1 = pnj deg(Fj ) #Ej /νn ,
showing that
#Ej /νn = pnj deg(Fj )(n−n1 ) #Ej /νn1 ,
where n1 > max(n0 , λ(λ − 1)/2). Putting everything together, we obtain
n
#E/νn = pλ(N )n+µ(N )p · const ,

where the constant does not depend on n if n is large enough. Because

λ(M ) = λ(N ) and µ(M ) = µ(N ), we have finished the proof of the asymptotic
formula. 2
Proof of (5.3.18): The endomorphism on M ⊗ ZZp IFp given by multiplication

by T has the characteristic polynomial X λ . Hence T is nilpotent, or equiva-
lently γ acts unipotently. Choosing a suitable basis of M ⊗ ZZp IFp , the matrix
representing the action of γ is of the form
∗
 
1
..

 . ∈

GL(λ, IFp ),
1
m
hence contained in a p-Sylow subgroup of GL(λ, IFp ). Therefore γ p = 1 on
M ⊗ ZZp IFp for m ≥ λ(λ−1)
2
. Now let n > λ(λ−1)
2
and let A ∈ M (λ × λ, ZZp ) be
the matrix corresponding to the action of γ on M with respect to some basis.
Then n−1
Ap ≡ I mod p
and so n
Ap ≡ I mod p2 ,
where I denotes the unit matrix. It follows that
n (p−1) n
Ap + · · · + Ap + I ≡ pI mod p2 ,
so that
n (p−1) n
Ap + · · · + Ap + I = pU with U ∈ GL(λ, ZZp ) .
Therefore
n (p−1) n
(γ p + · · · + γ p + 1)M = pM . 2
The following proposition gives a criterion for freeness of a Λ-module.
(5.3.19) Proposition. Let M be a finitely generated Λ-module.

(i) The following assertions are equivalent:
a) pdΛ M ≤ 1.
b) M Γn is ZZp -free for some n (resp. every n).
c) M has no finite nontrivial Λ-submodule.
(ii) M is a free Λ-module if and only if M Γ = 0 and MΓ is ZZp -free.
Proof: First we prove (ii). In order to show the nontrivial implication, let
0 −→ K −→ Λd −→ M −→ 0

be a minimal presentation of M by a free module. Using (5.3.11), we obtain

the exact sequence
0 = M Γ −→ KΓ −→ ZZdp −→ MΓ −→ 0.
Since d was chosen minimal and MΓ is ZZp -free, the map on the right is an
isomorphism. Using Nakayama’s lemma, KΓ = 0 implies K = 0.
Now we prove the equivalence of the assertions in (i). Assuming a), there
exists an exact sequence
0 −→ Λd1 −→ Λd0 −→ M −→ 0.
Using (5.3.11), it follows that M Γn ⊆ (Λd1 )Γn showing b). Let M0 be the
maximal finite Λ-module of M . Since M0 = 0 if and only if M0Γn = 0 for
some n, the inclusion M0Γn ⊆ M Γn shows the equivalence between b) and c).
Suppose b) is true and let
0 −→ N −→ Λd0 −→ M −→ 0
be a presentation of M with kernel N . Since (Λd0 )Γ = 0, we obtain by (5.3.11)
the exact sequence
0 −→ M Γ −→ NΓ −→(Λd0 )Γ .
From our assumption it follows that M Γ is ZZp -free, thus NΓ is also ZZp -free.
Hence the Λ-module N is free by (ii) and therefore pdΛ M ≤ 1. 2
For a finitely generated Λ-module M , we define

d0 (M ) = dimIFp H0 (Γ, M )/p,
d1 (M ) = dimIFp p H0 (Γ, M ) + dimIFp H1 (Γ, M )/p,
d2 (M ) = dimIFp p H1 (Γ, M ).
(5.3.20) Proposition. Let M be a finitely generated Λ-module. Then there

exists an exact sequence
0 −→ Λd2 (M ) −→ Λd1 (M ) −→ Λd0 (M ) −→ M −→ 0.
In particular,
rankΛ (M ) = d0 (M ) − d1 (M ) + d2 (M ) = rank ZZp MΓ − rank ZZp M Γ .
Proof: By Nakayama’s lemma (5.2.18), we have a minimal presentation

ϕ : Λd0 (M ) M . Let N = ker ϕ, i.e. we have an exact sequence
0 −→ N −→ Λd0 (M ) −→ M −→ 0 .
Using (5.3.11), we obtain an exact sequence
0 −→ M Γ −→ NΓ −→ ZZdp0 (M ) −→ MΓ −→ 0

which induces the exact sequence

0 −→ M Γ /p −→ NΓ /p −→ p (MΓ ) −→ 0
(note that ZZdp0 (M ) /p →
∼ M /p) and an isomorphism
Γ
Γ ∼
p (M ) −→ p (NΓ ) .
It follows that dimIFp NΓ /p = d1 (M ). Hence, again by Nakayama’s lemma, we

get a minimal presentation ψ : Λd1 (M ) N . Let N 0 be the kernel of ψ. Since
N Γ = 0, we find that NΓ0 is ZZp -free and therefore N 0 is Λ-free by (5.3.19)(ii).
Furthermore,
rankΛ (N 0 ) = dimIFp NΓ0 /p = dimIFp p (NΓ ) = dimIFp p (M Γ ) = d2 (M ).
2
∼ ZZ/pZZ. Show that

Exercise 1. Let M be the kernel of the natural projection Λ −→ Λ(p,Γ ) =
M is a torsion-free Λ-module of rank 1 which is not free, and that no pseudo-isomorphism
≈
Λ −→ M exists.
Exercise 2. Let M be a finitely generated Λ-module. Show the following statements.
n
a) For every n = 0, 1, . . . , the ZZp -module M/ωn M , ωn = (1 + T )p − 1, is finitely
generated and
rank ZZ p M/ωn+1 M ≥ rank ZZ p M/ωn M .
b) M is a Λ-torsion module if and only if rank ZZ p M/ωn M is bounded for n → ∞.
c) Let M be Λ-torsion, then the following assertions are equivalent:
(i) µ(M ) = 0.
(ii) M is a finitely generated ZZp -module.
(iii) dimIFp M ⊗ ZZ p IFp < ∞.
(iv) dimIFp (M/ωn M ⊗ ZZ p IFp ) is bounded for n → ∞.
Exercise 3. Let M be a finitely generated Λ-torsion module. Then the following assertions
are equivalent:
(i) #M Γn < ∞,
(ii) #MΓn < ∞,
n
(iii) FM,γ (ζ − 1) =/ 0 for all ζ with ζ p = 1.
Note that the statement (iii) is independent of the choice of the generator γ of Γ . If the
conditions above are fulfilled, then
#M Γn 1
| FM,γ (ζ − 1)|p ,
Y
= µ(M )pn
#MΓn p pn ζ =1
where | · |p denotes the p-adic valuation on Cp normalized by | p |p = p1 .

§4. Homotopy of Modules 301
Exercise 4. Let M be a finitely generated Λ-module. Show that the following assertions are
equivalent:
(i) M is a free Λ-module.
(ii) There exist two elements a, b ∈ Λ which generate the maximal ideal of Λ, such that
aM =0 and b (M/aM ) = 0,
where for a Λ-module N and an element c ∈ Λ the module c N is defined by {x ∈ N | cx = 0}.
Show that M is Λ-free if and only if M is ZZp -torsion-free and M/pM has no Γ -invariants.
Hint: Use a minimal presentation of M by a free Λ-module and imitate the proof of (5.3.19).
§4. Homotopy of Modules
In this and the following two sections we will follow closely the work of
U. JANNSEN [97]. The reader is strongly advised to consult the original paper
for many more results than are presented here; in particular, classification
theorems of ZZp [[T ]]-modules up to isomorphism.
Let Λ be a ring with unit, not necessarily commutative. Recall that pdΛ M
denotes the projective dimension of a Λ-module M .
(5.4.1) Definition. We denote the full subcategory of Mod(Λ) of modules M

with pdΛ M ≤ 1 by Mod1 (Λ).
(5.4.2) Definition. (i) A homomorphism f : M → N of Λ-modules is homo-

topic to zero (f ' 0) if it factors through a projective module P
f : M −→ P −→ N.
Two homomorphisms f, g : M → N are homotopic (f ' g) if f − g is homo-
topic to zero. We denote the homotopy category of Λ-modules by Ho(Λ), i.e.
the category whose objects are Λ-modules and in which the homomorphism
groups are given by HomΛ (M, N )/{f ' 0}.∗) We denote the full subcategory
of Ho(Λ) whose objects are in Mod1 (Λ) by Ho1 (Λ).
(ii) A homomorphism f : M → N of Λ-modules is a homotopy equivalence if
there exists a homomorphism g : N → M such that f g ' idN and gf ' idM ,
i.e. an isomorphism in Ho(Λ). In this case, we say that M and N are homotopy
equivalent (M ' N ).
∗) Observe that the notion of homotopy is compatible with composition of homomorphisms.

(5.4.3) Proposition. Let f, g : M → N be homomorphisms of Λ-modules.

(i) The following assertions are equivalent.
a) f ' g,
b) f ∗ , g ∗ : ExtiΛ (N, R) −→ ExtiΛ (M, R) are equal for all Λ-modules R
and all i ≥ 1,
c) f ∗ , g ∗ : Ext1Λ (N, R) −→ Ext1Λ (M, R) are equal for all Λ-modules R.
(ii) The following assertions are equivalent.

a) f is a homotopy equivalence,
b) f ∗ : ExtiΛ (N, R) −→ ExtiΛ (M, R) is an isomorphism for all R and
all i ≥ 1,
c) f ∗ : Ext1Λ (N, R) −→ Ext1Λ (M, R) is an isomorphism for all R,
d) There are projective Λ-modules P and Q and an isomorphism σ such
that f factors as
f : VUW M can
M ⊕P σ
N ⊕Q can
N.
Furthermore, if f : M → N is a homotopy equivalence between finitely
generated Λ-modules M and N , then the projective Λ-modules P and Q
in d) may also be chosen finitely generated.
Proof: (i) Obviously, we may assume that g = 0. Since ExtiΛ (P, R) = 0, i ≥ 1,

for a projective module P , f ' 0 implies f ∗ = 0. This proves a) ⇒ b). The
implication b) ⇒ c) is trivial. In order to show c) ⇒ a), let π : P N be a
surjection with P a projective module and let K = ker π. Taking the pull-back
via f , we obtain the commutative exact diagram
0_àbXYZ[\]^ K X M 0
f
π
0 K P N 0.
Since f ∗ : Ext1Λ (N, K) → Ext1Λ (M, K) is zero by assumption c), the upper se-
quence in the diagram splits, thus f factors through the projective Λ-module P .
(ii) Assuming a), there exists g : N → M such that f g ' id and gf ' id.
By (i), it follows that g ∗ f ∗ = id and f ∗ g ∗ = id, and so we obtain b). The
implication b) ⇒ c) is trivial. In order to prove c) ⇒ d), let π : P N be
a surjective Λ-homomorphism with P a projective module and let the module
K be defined by the exactness of the sequence
f +π
(∗) 0 −→ K −→ M ⊕ P −→ N −→ 0 .

Applying ExtΛ (−, K), we obtain the exact sequence

∼ Ext1 (M ⊕ P, K)
HomΛ (M ⊕ P, K) −→ HomΛ (K, K) −→ Ext1Λ (N, K) −→ Λ
where the right-hand map is an isomorphism by assumption c). A pre-image of

id ∈ HomΛ (K, K) gives us a splitting of the exact sequence (∗) and we obtain
an isomorphism
σ : M ⊕ P −→ ∼ N ⊕K
which induces f , using the canonical injection and projection respectively.

Applying ExtΛ (−, R) for an arbitrary Λ-module R and using c) we see that Q :=
K is projective and we obtain d). Obviously, d) implies a), since the injection
M ,→ M ⊕ P and the projection N ⊕ Q N are homotopy equivalences
(having the canonical projection and injection as inverses respectively).
Finally, the proof of the implication c) ⇒ d) shows that we can choose P
finitely generated if N is. The isomorphism σ then implies that Q is also
finitely generated, if M is. 2
(5.4.4) Definition. Let M be a (left) Λ-module. Then
E i (M ) := ExtiΛ (M, Λ) , i ≥ 0,
are (right) Λ-modules by functoriality and the right Λ-module structure of the
bimodule Λ. By convention, we set E i (M ) = 0 for i < 0. The Λ-dual E 0 (M )
will also be denoted by M + .
Remarks: 1. By proposition (5.4.3), the functor E i factors through a functor

Ei
Ho(Λ) −→ Mod (Λ) for i ≥ 1.
2. Clearly, E i can also be viewed as functor from right modules to left modules.
Also several functors defined below will interchange left and right action. In
the case of a group ring, there is a natural equivalence between right and left
modules induced by the involution of the group ring given by passing to the
inverses of the group elements. In general this is not possible, but for the
theory it is not necessary, and in the following we will not specify if we are
talking about left and right Λ-modules or if a functor interchanges left and right
Λ-actions. This would only cause notational complications and it will always
be clear where one has to insert “left” or “right” .
(5.4.5) Definition. We denote the full subcategory of Mod (Λ) of modules M

with M + = 0 by Mod+ (Λ). We denote the full subcategory of Ho(Λ) whose
objects are in Mod+ (Λ) by Ho+ (Λ).

(5.4.6) Lemma. The canonical localization functor ho : Mod(Λ) → Ho(Λ)

induces an equivalence of categories
∼ Ho (Λ).
ho : Mod+ (Λ) −→ +
Proof: Recall that for a projective module P , the canonical homomorphism

ϕP : P → P ++ is injective (see [17], chap.II, §2, no.8). If M ∈ Mod+ (Λ) and
f : M → P is any homomorphism from M to a projective module P , then the
commutative diagram
defgc M
f
P
0 M ++ P ++
shows that f = 0. Therefore a homomorphism in Mod+ (Λ) which is homotopic
to zero is zero. This proves the lemma. 2
Recall that a Λ-module is called finitely presented if there exists an exact

sequence
P1 −→ P0 −→ M −→ 0
with finitely generated projective modules P1 and P0 .
(5.4.7) Definition. We denote the full subcategory of Mod(Λ) whose ob-

jects are finitely presented Λ-modules by Modf p (Λ). The notation Modf1 p (Λ),
Hof p (Λ), Hof1 p (Λ), . . . have their obvious meaning.
Now we will construct a contravariant duality functor

D : Hof p (Λ) −→ Hof p (Λ)
as follows:
For every finitely presented module M , we choose a presentation P1 →
P0 → M → 0 of M by finitely generated projectives. Then we define
DM ∈ Ob(Modf p (Λ)) = Ob(Hof p (Λ)) by the exact sequence
0 −→ M + −→ P0+ −→ P1+ −→ DM −→ 0.
If f : M → N is a homomorphism and Q1 → Q0 → N → 0 is the chosen
presentation of N , then we choose α : P0 → Q0 and β : P1 → Q1 to make the
diagram
Ppohijklmn 1 P0 M 0
β α f
Q1 Q0 N 0

commutative (this is possible since P0 and P1 are projective). We define

Df : DN → DM by the commutative diagram
0~}|qrstuvwxyz{ N+ Q+0 Q+1 DN 0
f+ α+ β+ Df
0 M+ P0+ P1+ DM 0.
(5.4.8) Definition. We call the functor

D : Hof p (Λ) −→ Hof p (Λ)
the transpose.
(i) The functor D is well-defined and (up to canonical functor isomorphism)
independent of the chosen presentations.
(ii) D is a (contravariant) autoduality of Hof p (Λ), i.e. D ◦ D ∼
= id.
(iii) For M ∈ Modf p (Λ), there exists a canonical exact sequence
ϕ
0 −→ E 1 (DM ) −→ M −→
M
M ++ −→ E 2 (DM ) −→ 0 ,
where ϕM is the canonical homomorphism of M to its bidual.
(5.4.10) Definition. For a finitely presented Λ-module M , we set

Ti (M ) := E i (DM ) , i = 1, 2.
Note that, by (5.4.3), the groups E i (DM ) are well-defined in Mod(Λ), while
DM is defined only up to homotopy equivalence. Thus Ti (M ) only depends
on the homotopy equivalence class of M .
Proof of (5.4.9): Let R be a Λ-module. Consider the commutative exact

diagram
P1 ⊗Λ R P0 ⊗Λ R M ⊗Λ R 0
ϕP0 ,R ϕM,R
0 Hom(K, R) Hom(P0+ , R) Hom(M + , R)

where K = ker(P1+ DM ) and ϕM,R is the canonical homomorphism
ϕM ⊗R
M ⊗ Λ R M ++ ⊗Λ R Hom(M + , R).

Observe that ϕP0 ,R is an isomorphism, since P0 is finitely generated and pro-

jective, and that the dotted arrow factors as
ϕP
R
P1 ⊗ 1
Hom(P1+ , R)
,R
Hom(K, R) .
The snake lemma (1.3.1) implies that
= coker (Hom(P + , R) −→ Hom(K, R)) ∼
ker ϕM,R ∼ = Ext1 (DM, R) ,
1 Λ
coker ϕM,R ∼
= coker (Hom(P0+ , R) −→ Hom(M + , R)) ∼
= Ext1Λ (K, R)
∼
= Ext2 (DM, R).
Λ
Therefore we obtain an exact sequence

u ϕM,R v
(∗) 0 → Ext1Λ (DM, R) → M ⊗Λ R −→ Hom(M + , R) → Ext2Λ (DM, R) → 0,
where the homomorphisms u and v a priori depend on the chosen presen-
tation P1 → P0 → M → 0. Now assume that we are given a homomorphism
f : M → N in Modf p (Λ). Let the diagram
P 1 P0 M 0
β α f
Q1 Q0 N 0
be as in the definition of Df , which we denote for the moment by Df(α,β) .
Then for every Λ-module R, we obtain an exact commutative diagram
uP1 ,P0 ϕM,R
0 Ext1Λ (DM, R) M ⊗Λ R Hom(M + , R)
∗
Df(α,β) f ⊗id (f + )∗
uQ1 ,Q0 ϕN,R

0 Ext1Λ (DN, R) N ⊗Λ R Hom(N + , R) .
Therefore Df is well-defined, i.e. independent of α, β (use (5.4.3)(i)). Fur-
thermore, setting M = N and f = id, we see that DM does not depend (up to
canonical homotopy equivalence) on the chosen projective presentation. Thus
D is a functor from Modf p (Λ) to Hof p (Λ).
If f : M → N is homotopic to zero, then by definition it factors through
a projective module, P say. But then choose (compare with the construction
above) α : P0 → Q0 also factorizing through P , and β : P1 → Q1 to be zero:
P ¡¢£¤¥¦§¨©ª«¬ 1 P0 M 0
0 P P 0
Q1 Q0 N 0.

This shows Df ' 0. Hence D is well-defined as a functor

D : Hof p (Λ) −→ Hof p (Λ).
The assertion (ii) is now trivial since we can choose the sequence P0+ → P1+
→ DM → 0 as a projective presentation of DM , showing that D(DM ) ' M .
The exact sequence in (iii) follows from (∗) on setting R = Λ and from
the remark that a posteriori the maps u and v do not depend on the chosen
presentation. 2
(5.4.11) Corollary. The transpose D defines an equivalence of categories

D
Hof1 p® (Λ) Hof+ p (Λ) ,
D
and D restricted to Hof1 p (Λ) coincides with E 1 .
Proof: If M ∈ Hof+ p (Λ), then the exact sequence 0 → P0+ → P1+ → DM → 0

shows that DM ∈ Hof1 p (Λ). Conversely, if M ∈ Hof1 p (Λ) and we choose the
presentation in the form 0 → P1 → P0 → M → 0, then
DM ∼ = coker (P + → P + ) ∼
0 = Ext1 (M, Λ) and
1 Λ
(DM ) ∼
=+
ker (P1++ → P0++ ) ∼
= ker (P1 → P0 ) = 0. 2
Together with (5.4.6), the last result implies the
(5.4.12) Corollary. The functor E 1 defines an equivalence of categories

E1
Hof1 p (Λ) −→ fp
∼ Mod + (Λ).
Now we assume that

Λ = ZZp [[G]],
where G is a profinite group. Recall from the discussion in §2 that every
finitely presented Λ-module carries a natural compact topology and that Λ-
homomorphisms of such modules are continuous. If P is finitely generated
and projective, then so is P + . Suppose that the Λ-module M has a resolution
· · · −→ P2 −→ P1 −→ P0 −→ M −→ 0
by finitely generated projective modules Pi , i = 0, 1, . . . ; in particular, M is
finitely presented. Then the complex of finitely generated projectives
0 −→ P0+ −→ P1+ −→ P2+ −→ · · ·
computes E i (M ). Hence the groups E i (M ) are canonically endowed with a
compact topology. If Λ is noetherian, this applies to every finitely generated

module M . The following theorem relates the Λ-modules E r (M ) to the

discrete G-modules
Dr (M ∨ ) = lim H r (U, M ∨ )∗ , r ≥ 0,
−→
U ⊆G
where
M ∨ = Homcts (M, Qp /ZZp ) = lim Hom (MU , Qp /ZZp ) = lim (MU )∗
−→ −→
U U
is a discrete G-module (see (2.5.1) for the definition of Dr (A) with a discrete
G-module A). We set Dr (M ∨ ) = 0 if r < 0.
(5.4.13) Theorem. Assume that the Λ-module M has a resolution by finitely

generated projective modules.
(i) There exists a functorial exact sequence
0 −→ Dr (M ∨ ) ⊗ ZZp Qp /ZZp −→ E r (M )∨ −→ tor ZZp Dr−1 (M ∨ ) −→ 0
for all r.
If, in addition, tor ZZp M and M/tor ZZp M also have a resolution by finitely
generated projectives, then the following hold for all r:
(ii) E r (M/tor ZZp M )∨ ∼
= lim Dr (pm (M ∨ )),
−→
m
(iii) E r (tor ZZp M )∨ ∼

= lim
−→
Dr−1 (M ∨ /pm ).
m
(iv) There is a long exact sequence

→ E r (M )∨ → lim Dr (pm (M ∨ )) → lim Dr−2 (M ∨ /pm ) → E r−1 (M )∨ →,
−→ −→
m m
functorial in M and in G.
Proof: Assume for the moment that Λ is noetherian. The functor M p

(M + )∨ from the category of finitely generated Λ-modules to abelian groups is
the composition of the right exact functors M p D0 (M ∨ ) (this functor takes
values in ZZp -modules) and N p N ⊗ ZZp Qp /ZZp because
M + = lim HomΛ (M, ZZp [G/U ]) = lim Hom ZZp [G/U ] (MU , ZZp [G/U ])
←− ←−
U ⊆G U ⊆G
= lim
←−
Hom ZZp (MU , ZZp ),
U ⊆G

where the limit is taken over all normal subgroups U ⊆ G with respect to the
duals of the norm maps, so that
(M + )∨ = lim Hom ZZp (MU , ZZp )∨ = lim MU ⊗ ZZp Qp /ZZp .
−→ −→
U ⊆G U ⊆G
The r-th derived functor of M p D0 (M ∨ ) is M p Dr (M ∨ ) and the first
functor in the composition sends projectives to acyclics for the second functor.
We get a Grothendieck spectral sequence of homological type
ZZ
Eij2 = Tori p (Dj (M ∨ ), Qp /ZZp ) ⇒ Ei+j = E i+j (M )∨ .
Since 


N ⊗ Qp /ZZp , i = 0,
ZZ
Tori p (N, Qp /ZZp )
=  tor ZZp N , i = 1,


0, i ≥ 2,
we obtain (i) under the assumption that Λ is noetherian.
If Λ is not noetherian, then the category of modules having a resolution by
finitely generated projectives has no good properties. Therefore we cannot use
the general machinery and have to construct the sequence by hand.
Let · · · → P1 → P0 → M → 0 be a resolution of M by finitely generated
projectives. Then the complex
(∗) · · · −→ D0 (P2∨ ) −→ D0 (P1∨ ) −→ D0 (P0∨ ) −→ 0
.
calculates D (M ∨ ) and consists of torsion-free, hence flat, ZZp -modules.
Choosing a Cartan-Eilenberg resolution of the complex (∗) by a ZZp -projective
double complex, we obtain (i) from the spectral sequence associated to the
double complex tensored by Qp /ZZp .
The exact sequence
(∗∗) 0 −→(M/tor ZZp M )∨ −→ M ∨ −→(tor ZZp M )∨ −→ 0
and the fact that (M/tor ZZp M )∨ /pm = 0 imply that
lim Dr−1 ((tor ZZp M )∨ /pm ) −→
∼ lim D ∨ m
r−1 (M /p ).
−→ −→
m m
This shows that it suffices to show (iii) for the case that M is ZZp -torsion. Next
we show that we may assume M to be ZZp -torsion-free in (ii).
Consider, for every m, the pm -torsion sequence associated to (∗∗)
0 −→ pm (M/tor ZZp M )∨ −→ pm M ∨ −→ pm (tor ZZp M )∨ −→ 0.
Since tor ZZp M is finitely generated, there exists an n such that the pn -multipli-
cation map n
∨ p ∨
pm+n (tor ZZp M ) −→ pm (tor ZZp M )
is the zero map for all m. Hence lim Dr (pm ((tor ZZp M )∨ )) = 0 and thus
−→ m
lim
−→
Dr (pm (M ∨ )) ∼ lim
−→ −→
Dr (pm ((M/tor ZZp M )∨ )).
m m

In order to prove (ii), let

.
P : · · · −→ P1 −→ P0 −→ M −→ 0
be a resolution of the ZZp -torsion-free module M by finitely generated projec-
tives. Then
0 −→(M/pm )∨ −→(P0 /pm )∨ −→(P1 /pm )∨ −→ · · ·
is a resolution of the discrete G-module (M/pm )∨ by cohomologically trivial

G-modules. Therefore
.
E r (M )∨ = H r ((P + )∨ )
r
−→ −→
.
= H (lim lim (((P /pm )∨ )U )∗ )
m U ⊆G
= lim
−→ −→
.
lim H r (((P /pm )∨ )U )∗
m U ⊆G
= lim lim H r (U, (M/pm )∨ )∗
−→ −→
m U ⊆G
= lim Dr (pm (M ∨ )).
−→
m
This shows (ii).

In order to prove (iii), we may assume that M = tor ZZp M . Then M is already
annihilated by some power of p and hence the same is true for Dr−1 (M ∨ ) and
Dr (M ∨ ). Therefore Dr (M ∨ ) ⊗ Qp /ZZp = 0 and (i) implies that
E r (M )∨ ∼
= tor ZZp Dr−1 (M ∨ )
= Dr−1 (M ∨ )
= lim
−→
Dr−1 (M ∨ /pm ).
m
(The last limit becomes stationary.) This shows (iii).

Finally, (iv) follows from (ii),(iii) and the long exact Ext-sequence
· · · → E r (M )∨ → E r (M/tor ZZp M )∨ → E r−1 (tor ZZp M )∨ → · · · . 2
(5.4.14) Corollary. If cdp G = n is finite and if M has a resolution by finitely

generated projectives, then E r (M ) = 0 for r > n + 1.
Proof: Obviously, we have Dr (A) = 0 for r > n and for all p-primary
discrete G-modules A. Using (5.4.13)(i), we obtain the result. 2

(5.4.15) Corollary. Assume that G is a duality group ∗) at p of dimension n

with dualizing module Dn(p) = lim
−→ i
Dn (ZZ/pi ZZ). Then the following hold for
every Λ-module M which has a resolution by finitely generated projectives:
(i) If M is free of finite rank as a ZZp -module, then
 lim Dn ((M/pm )∨ ) ∼


−→
= M ⊗ ZZp Dn(p) if r = n,
r ∨ ∼
E (M ) = m
 0

otherwise.
(ii) If M is a finite p-primary G-module, then


Hom ZZp (M ∨ , Dn(p) ) if r = n + 1,

E r (M )∨ ∼
=
0 otherwise.
Proof: By (5.4.13)(ii), we have in case (i)

E r (M )∨ = lim Dr ((M/pm )∨ ) ,
−→
m
which is zero for r =/ n by (3.4.6), and
E n (M )∨ ∼
= lim lim H 0 (U, Hom ZZp ((M/pm )∨ , Dn(p) ))
−→ −→
m U ⊆G
∼
= lim Hom ZZp ((M/pm )∨ , Dn(p) ) ∼
= M ⊗ ZZp Dn(p) .
−→
m
In order to prove (ii), we use (5.4.13)(iii):
E r (M )∨ ∼
= Dr−1 (M ∨ ) ,
and hence E n+1 (M )∨ ∼
= Dn (M ∨ ) ∼
= Hom ZZp (M ∨ , Dn(p) ) and E r (M ) = 0 for
r =/ n + 1. 2
We finish this section with some remarks concerning the change of the
group. Let H be an open subgroup in G. We consider the forgetful functor
from abstract (resp. compact) ZZp [[G]]-modules to abstract (resp. compact)
ZZp [[H]]-modules.
(5.4.16) Lemma. The forgetful functor sends projectives to projectives, i.e.

an abstract (resp. compact) ZZp [[G]]-module which is projective as an abstract
(resp. compact) ZZp [[G]]-module is also projective as an abstract (resp. compact)
ZZp [[H]]-module. If H is normal and of prime-to-p index in G, then also the
converse statement is true.
∗) see III §4.

Proof: A projective module is a direct summand in a free module. It thus

suffices to show that ZZp [[G]] is a free ZZp [[H]]-module. But this is clear because
ZZp [[G]] ∼
= ZZp [[G]] ⊗ ZZp [[H]] ZZp [[H]] ∼
= IndH
G (Z
Zp [[H]]).
Now suppose that H is normal in G and that p - (G : H). Since #(G/H)−1 ∈
ZZp , the functor M p M G/H is exact on ZZp [G/H]-modules. Therefore the
equality
Hom ZZp [[G]] (M, N ) = Hom ZZp [[H]] (M, N )G/H
implies isomorphisms
ExtiZZp [[G]] (M, N ) ∼
= ExtiZZp [[H]] (M, N )G/H
for all abstract (compact) ZZp [[G]]-modules M, N and all i ≥ 0. This implies
the remaining statement. 2
The following proposition shows that the functors E i commute with the
forgetful functor:
(5.4.17) Proposition. Suppose M is a ZZp [[G]]-module which has a resolution

by finitely generated projectives. Let H be an open subgroup in G. Then we
have natural isomorphisms of (right) ZZp [[H]]-modules
ExtiZZp [[G]] (M, ZZp [[G]]) ∼
= ExtiZZp [[H]] (M, ZZp [[H]])
for all i ≥ 0.
Proof: The isomorphism ZZp [[G]] ∼ = IndHG (Z

Zp [[H]]) (see the proof of (5.4.16)),
together with Frobenius reciprocity, shows that
Hom ZZ [[G]] (M, ZZp [[G]]) ∼
p = Hom ZZ [[H]] (M, ZZp [[H]]).
p
By (5.4.16) and since the index of H in G is finite, the forgetful functor sends
finitely generated projectives to finitely generated projectives. Therefore the
above isomorphism extends to Exti for all i ≥ 0. 2
§5. Homotopy Invariants of Iwasawa Modules
In this section, we specialize again to the case G = Γ ∼ = ZZp , so that

Λ = ZZp [[Γ ]] is the Iwasawa algebra. Then Λ is a commutative 2-dimensional
regular local ring and therefore its projective dimension pd(Λ) is equal to 2

§5. Homotopy Invariants of Iwasawa Modules 313
(see [129], th. 19.2; this also follows from §2, ex.5). Thus for every Λ-module
M the projective dimension pdΛ M is equal or less than 2, i.e. there exists a
projective resolution
0 −→ P2 −→ P1 −→ P0 −→ M −→ 0
of M of length 2 (or smaller). This implies that E i (M ) = 0 for i ≥ 3. By
(5.4.9)(iii), the following definition agrees with the one given in (5.4.10).
(5.5.1) Definition. For a finitely generated Λ-module M we set

T0 (M ) := the maximal finite submodule of M ,
T1 (M ) := ker (ϕM : M → M ++ ) ,
T2 (M ) := coker (ϕM : M → M ++ ) .
(5.5.2) Lemma. For a Λ-module M the following assertions are equivalent.

(i) M is projective,
(ii) M ' 0.
If M is finitely generated, then M is free if these hold.
Proof: Obviously, (i) ⇒ (ii). If M ' 0, then the identity map id : M → M

factors through a projective module, hence M is itself projective, being a direct
summand of a projective. The last assertion follows from (5.2.20). 2
(5.5.3) Proposition. Let M be a finitely generated Λ-module. Then

(i) T1 (M ) is the Λ-torsion submodule of M .
(ii) E 1 (M ) is a Λ-torsion module. If M is Λ-torsion, then E 1 (M ) has no
nontrivial finite submodules, i.e. T0 E 1 (M ) = 0. If M is finite, then
E 1 (M ) = 0.
(iii) T2 (M ) is finite, and T2 (M ) = 0 if and only if M/T1 (M ) is free (hence
M∼ = Λr ⊕ T1 (M ) for some r in this case).
In particular, T2 (M ) = 0 for a Λ-torsion module M .
(iv) E 2 (M ) is finite. One has
E 2 (M ) ∼
= E 2 (T0 (M )) ∼
= T0 (M )∨
and the following assertions are equivalent:
a) E 2 (M )= 0,
b) T0 (M ) = 0,
c) pdΛ M ≤ 1.

Proof: Let K be the field of fractions of Λ. By definition, see (5.5.1), we

have the exact sequence
∼ M ++ ⊗ K −→ T (M ) ⊗ K −→ 0.
0 −→ T1 (M ) ⊗Λ K −→ M ⊗Λ K −→ Λ 2 Λ
Thus T1 (M ) and T2 (M ) are Λ-torsion modules. Since M ++ is torsion-free, we

have proved (i) and the first statement of (ii) because E 1 (M ) ∼
= E 1 (DDM )
∼
= T1 (DM ). Now let M be a Λ-torsion module and let
P1 −→ P0 −→ M −→ 0
be a presentation of M by free Λ-modules. We obtain the exact sequence
0 = M + −→ P0+ −→ P1+ −→ DM −→ 0
showing that pdΛ DM ≤ 1, and so DM has no finite nontrivial Λ-submodules
by (5.3.19)(i) .∗) Thus E 1 (M ) = E 1 (DDM ) = T1 (DM ) has this property.
Finally, if M is finite, then pn and ωn annihilate M for some n large enough
and the same is true for E 1 (M ). Thus E 1 (M ) is finite and therefore zero.
In order to prove (iii), we observe that (M/T1 (M ))p is a free Λp -module of
finite rank by (i). Hence
∼ (M/T (M ))++ = M ++
(M/T1 (M ))p −→ 1 p p
for all prime ideals p of Λ of height ≤ 1. Thus T2 (M ) is pseudo-null, i.e. finite.

Furthermore, M ++ is free and the exact sequence
0 −→ M/T1 (M ) −→ M ++ −→ T2 (M ) −→ 0
shows that (M/T1 (M ))Γ ⊆ (M ++ )Γ = 0 and proves the exactness of
0 −→ T2 (M )Γ −→(M/T1 (M ))Γ −→(M ++ )Γ .
Therefore (M/T1 (M ))Γ is ZZp -free if and only if T2 (M )Γ = 0 or equivalently
T2 (M ) = 0. Now (5.3.19) (ii) implies assertion (iii).
Since E 2 (M ) = T2 (DM ), this module is finite by (iii). From the structure
theorem for Λ-modules, we obtain an exact sequence
f
0 −→ T0 (M ) −→ M −→ E −→ C −→ 0,
where C is finite and
s t
E∼
n
= Λr ⊕ mi
M M
Λ/p ⊕ Λ/Fj j .
i=1 j=1
The long exact Ext-sequence implies the exactness of

E 2 (imf ) −→ E 2 (M ) −→ E 2 (T0 (M )) −→ 0
and that
¯
E 2 (E) E 2 (imf )
∗) In fact, DM is only defined up to homotopy equivalence, but we see that DM has no
nontrivial finite submodules for every choice of DM .

is surjective. Since pdΛ E ≤ 1, we get E 2 (E) = 0, and so E 2 (M ) ∼

= E 2 (T0 (M )).
Now we recall that Γ is a Poincaré group of dimension 1 with dualizing module
Qp /ZZp (trivial action). Therefore (5.4.15)(ii) implies E 2 (T0 (M )) = T0 (M )∨ .
The equivalence between a) and b) is now trivial and the equivalence between
b) and c) is contained in (5.3.19)(i). 2
(5.5.4) Corollary. Let M be a finitely generated Λ-module. Then

∼ E 1 (M ).
(i) E 1 (M/T0 (M )) −→
(ii) E 1 (M ) = 0 if and only if M/T0 (M ) is free
(hence M ∼ = Λr ⊕ T0 (M ) for some r ≥ 0 in this case).
Proof: The exact sequence

0 = E 0 (T0 (M )) −→ E 1 (M/T0 (M )) −→ E 1 (M ) −→ E 1 (T0 (M )),
together with (5.5.3)(ii), implies assertion (i). By (5.5.3)(iv), we have
pdΛ (M/T0 (M )) ≤ 1. Therefore the following assertions are equivalent:
M/T0 (M ) is free ⇐⇒ M/T0 (M ) ' 0 (by lemma (5.5.2))

⇐⇒ E 1 (M/T0 (M )) = 0 (by corollary (5.4.12))
⇐⇒ E 1 (M ) = 0 (by (i)).
2
(5.5.5) Definition. Let M be a finitely generated Λ-torsion module and let

{πn } be a sequence of non-zero elements of Λ such that
π0 ∈ m , πn+1 ∈ πn m ,
and such that the set of prime ideals dividing the principal ideals πn Λ, n ≥ 0,
is disjoint to the set of prime ideals of height 1 in supp(M ). Let
α(M ) := lim
←−
Hom(M/πn M, Qp /ZZp )
n
with respect to the inductive system
M/πn −→ M/πm for m ≥ n ≥ 0 ,
x mod πn 7−→ ππmn x mod πm .
The Λ-module α(M ) is called the Iwasawa-adjoint of M .
(5.5.6) Proposition. For a finitely generated Λ-torsion module M one has a

α(M ) ∼= E 1 (M ).
In particular, α(M ) is independent of the choice of the sequence {πn }.

Proof: From the exact sequence

T0 (M )/πn −→ M/πn −→(M/T0 (M ))/πn −→ 0
and lim
−→ n
T0 (M )/πn = 0, we obtain α(M/T0 (M )) ∼ = α(M ). Thus using
(5.5.4)(i), we may assume T0 (M ) = 0. Then the sequence
n π
0 −→ M −→ M −→ M/πn −→ 0
is exact and M/πn is finite for all n ≥ 0, because the multiplication on M by πn
is a pseudo-isomorphism by (5.1.6). Using (5.5.3)(iv), we get isomorphisms
∼ E 2 (M/π ) ∼ Hom
E 1 (M )/πn −→ n = ZZ (M/πn , Qp /Z
p Zp ).
1
Since πn → 0 in Λ when n → ∞ and E (M ) is finitely generated, it follows
that
E 1 (M ) = lim E 1 (M )/πn ∼
= lim Hom ZZp (M/πn , Qp /ZZp ) = α(M ). 2
←− ←−
n n
(5.5.7) Corollary. Let M be a finitely generated Λ-torsion module. Assume

that µ(M ) = 0. Then
E 1 (M ) ∼
= Hom ZZ (M, ZZp ).p
Proof: Since µ(M ) = 0, we can take πn = pn+1 . From (5.5.6) we obtain

E 1 (M ) ∼
= lim Hom ZZp (M/pn , Qp /ZZp )
←− n
= lim Hom ZZp (M, ZZ/pn ) = Hom ZZp (M, ZZp ). 2
←− n

(i) There exists an exact sequence
0 −→ E 2 (T2 (M )) −→ E 1 (M ) −→ E 1 (T1 (M )) −→ 0
inducing isomorphisms
(ii) E 2 (T2 (M )) ∼
= E 1 (M/T1 (M )) ∼
= T0 (E 1 (M )),
(iii) E 1 (T1 (M )) ∼
= E 1 (M )/T0 (E 1 (M )).
Furthermore, there are canonical isomorphisms
(iv) E 1 (E 1 (M )) ∼
= T1 (M )/T0 (M ),
(v) E 2 (E 1 (M )) ∼
= T2 (M ),
(vi) E 2 (E 2 (M )) ∼
= T0 (M ).


ϕ
M
0 −→ T1 (M ) −→ M −→ M ++ −→ T2 (M ) −→ 0,
we obtain exact sequences
0 = E 1 (M ++ ) → E 1 (im ϕM ) →
∼ E 2 (T (M )) → E 2 (M ++ ) = 0,
2
0 = E 0 (T1 (M )) → E 1 (im ϕM ) → E 1 (M ) → E 1 (T1 (M )) → E 2 (im ϕM ) = 0.
This implies (i) and the first isomorphism of (ii). Using T0 (E 1 (T1 (M ))) = 0 (by
(5.5.3)(ii)) and the finiteness of E 2 (T2 (M )), we obtain the second isomorphism
in (ii) and assertion (iii).
In order to prove (iv), we first observe that by (5.5.4)(i) and (iii) above,
E 1 (E 1 (M )) ∼
= E 1 (E 1 (M )/T0 (E 1 (M )))
∼
= E 1 (E 1 (T1 (M ))) ∼
= E 1 (E 1 (T1 (M )/T0 (M ))).
f
In the proof of (5.4.11) we saw that E 1 coincides with D on Ho1p (Λ). Since
pdΛ (T1 (M )/T0 (M )) ≤ 1 by (5.5.3)(iv), we therefore obtain
E 1 (E 1 (T1 (M )/T0 (M ))) = E 1 (D(T1 (M )/T0 (M )))

= T1 (T1 (M )/T0 (M )) = T1 (M )/T0 (M ).
Finally, from (i) we get the isomorphism
0 = E 2 (E 1 (T1 (M ))) −→ E 2 (E 1 (M )) −→
∼ E 2 (E 2 (T (M ))) −→ 0
2
and using (5.5.3)(iv),
E 2 (E 2 (T2 (M ))) ∼
= T2 (M )∗∗ = T2 (M ) ,
thus proving (v). Again by (5.5.3)(iv) we obtain (vi). 2
(5.5.9) Corollary. We have the following equivalences:
E 1 (M ) is finite ⇐⇒ T1 (M ) is finite ⇐⇒ E 1 (E 1 (M )) = 0.

(i) E 0 (M ) ∼
= lim pm (M
∨ Γn
) is free of the same rank as M ,
←−
n,m
(ii) E 1 (tor ZZp M ) ∼

= lim (M ∨ /pm )Γn ,
←−
n,m
(iii) E 1 (M/tor ZZp M ) ∼

= lim
←−
(pm (M ∨ ))Γn
n,m
(∼
= Hom ZZp (M, ZZp ) if M is a Λ-torsion module),
(iv) E 1 (M δ ) ∼
= lim pm ((M
∨
)Γn ) ∼
= Hom ZZp (M δ , ZZp ),
←−
n,m
(v) E 1 (M/M δ ) ∼
= lim ((M ∨ )Γn )/pm ,
←−
n,m
(vi) E 2 (M ) ∼
= lim (M ∨ )/(pm , Γn ) ,
←−
n,m
[
where the transition maps are the obvious ones. Recall that M δ = M Γn is
n
the maximal Λ-submodule of M on which Γ acts discretely.
Proof: Since H 0 (Γn , A) = AΓn and H 1 (Γn , A) = AΓn for a discrete Γ -module
A, the assertions (i), (ii), (iii) and (vi) follow immediately from (5.4.13)(ii)-(iv).
If M is Λ-torsion, then M/tor ZZp M is a torsion module with trivial µ-invariant
and we obtain from (5.5.7) that
E 1 (M/tor ZZp M ) = Hom ZZp (M/tor ZZp M, ZZp ) = Hom ZZp (M, ZZp ) ,
showing the additional statement in (iii). Using (5.4.13)(i), we obtain an exact

sequence
0 −→ lim
←−
(M ∨ )Γn /pm −→ E 1 (M ) −→ lim
←−
pm ((M
∨
)Γn ) −→ 0 ,
n,m n,m
where the cokernel is isomorphic to Hom ZZp (M δ , ZZp ) while the kernel vanishes
for M = M δ . By (5.5.3)(iv) we have E 2 (M/M δ ) = 0, thus we obtain an exact
sequence
0 −→ E 1 (M/M δ ) −→ E 1 (M ) −→ E 1 (M δ ) −→ 0.
Since the first exact sequence is functorial in M , it must be isomorphic to the

second one, and we get (iv) and (v). 2

For the rest of this section we consider Λ-modules up to pseudo-isomorphism

(denoted by ≈).
(5.5.11) Proposition. Let M and M 0 be finitely generated Λ-torsion modules.

Then
M ≈ M0 if and only if E 1 (M 0 ) ≈ E 1 (M ).
Proof: We may assume T0 (M ) = 0 = T0 (M 0 ). From the exact sequence

0 −→ M −→ M 0 −→ C −→ 0
with a finite Λ-module C, we obtain
0 = E 1 (C) −→ E 1 (M 0 ) −→ E 1 (M ) −→ E 2 (C) ,
and hence E 1 (M 0 ) ≈ E 1 (M ). Conversely, since E 1 (M 0 ) ≈ E 1 (M ), (5.5.8)(iv)
implies the pseudo-isomorphisms M ≈ E 1 (E 1 (M )) ≈ E 1 (E 1 (M 0 )) ≈ M 0 .
2
(5.5.12) Definition. Let M be a Λ-module. We define the Λ-module M ◦ to be

the Λ-module M with the inverse Γ -action:
if x ∈ M ◦ then γ ◦ x := γ −1 x for γ ∈ Γ.
Remark: If F ∈ Λ is a Weierstraß polynomial, then
(Λ/F )◦ ∼
= Hom (Λ/F, ZZp ).
We have (Λ/F )◦ ∼= Λ/F 0 for the Weierstraß polynomial F 0 which is obtained
from F by substituting (1 + T )−1 − 1 for T and multiplying by (1 + T )deg F .
(5.5.13) Proposition. Let M be a finitely generated Λ-module. Then there

exists a pseudo-isomorphism
E 1 (M ) ≈ T1 (M )◦ .

0 −→ tor ZZp M −→ M −→ M/tor ZZp M −→ 0,
0 → E 1 (M/tor ZZp M ) → E 1 (M ) → E 1 (tor ZZp M ) → E 2 (M/tor ZZp M ).

Using (5.5.8)(i) and (5.5.7), we see that

E 1 (M/tor ZZp M ) ≈ E 1 (T1 (M )/tor ZZp M )
∼
= Hom ZZp (T1 (M )/tor ZZp M, ZZp )
∼
= (T1 (M )/tor ZZ M )◦ . p
Furthermore,
s
M
E 1 (tor ZZp M ) ≈ E 1 ( Λ/pmi ) for some mi ≥ 0
i=1
s
∼
M
= E 1 (Λ/pmi )
i=1
s
∼
M
= (Λ/pmi )◦ ≈ (tor ZZp M )◦ ,
i=1
so that
E 1 (M ) ≈ E 1 (M/tor ZZp M ) ⊕ E 1 (tor ZZp M )
≈ (T1 (M )/tor ZZp M )◦ ⊕ (tor ZZp M )◦ ≈ T1 (M )◦ . 2
Exercise: Let M be a finitely generated Λ-torsion module and let Y be defined by the exact
sequence Y
0 −→ T0 (M ) −→ M −→ Mp −→ Y −→ 0.
ht(p)=1
Then there is a canonical isomorphism
α(M ) ∼
= Homcts (Y, Qp /ZZp ).
This was the original definition of the adjoint due to Iwasawa.
Hint: Let {πn } be a sequence of elements of Λ as considered in (5.5.5) and let a be the
annihilator ideal of M . Let S be the multiplicatively closed subset in Λ̄ := Λ/a which is
generated by the images of π0 , π1 , . . .. Prove the isomorphism
A := S −1 Λ̄ ∼ = lim Λ̄(i) ,
−→ i
πj
where the inductive system is given by Λ̄(i) = Λ̄, Λ̄(i) → Λ̄(j) , λ 7→ πi λ. By the assumptions
on π0 , π1 , . . . , the ring A is artinian, and thus
∼
Y
A= Λp /aΛp .
p∈ supp(M )
ht(p)=1
Conclude that
lim M ⊗Λ Λ̄(i) ∼
= M ⊗Λ A ∼
Y
= Mp .
−→
i ht(p)=1
Finally, consider the exact commutative diagram
0½¸¹º»¼³´µ¶·°±²
πi
ker(πi ) M M M/πi M 0
πj πj
πi πi
πj
0 ker(πj ) M M M/πj M 0.

§6. Differential Modules and Presentations 321
§6. Differential Modules and Presentations
R. FOX, in 1953, established a theory of derivations in the group ring of

a free discrete group, the “free differential calculus”. In the following we
consider a topological version of this theory for the completed group ring of a
profinite group.
Throughout this section, let p be a prime number, c a class of finite groups
containing ZZ/pZZ and closed under taking subgroups, homomorphic images
and group extensions. Let G be a pro-c-group and let α : ZZp [[G]] → ZZp denote
the augmentation map (cf. §2).
(5.6.1) Definition. A derivation of ZZp [[G]] into a compact ZZp [[G]]-module A

is a continuous ZZp -linear map
D : ZZp [[G]] −→ A
satisfying
D(f g) = D(f )·α(g) + f ·D(g)
for all f, g ∈ ZZp [[G]].
Remark: A derivation is clearly determined by its restriction to G ⊆ ZZp [[G]]

and this restriction is a continuous inhomogeneous 1-cocycle of G with values
in the compact G-module A (cf. II §7). On the other hand, one easily verifies
that every continuous inhomogeneous 1-cocycle of G with values in A extends
to a derivation of ZZp [[G]] into A. Note that
(i) D(a) = 0 for a ∈ ZZp ,
(ii) D(σ −1 ) = −σ −1 D(σ) for σ ∈ G.
Now we construct a ZZp [[G]]-module Ω ZZp [[G]] as follows: take the free com-
pact ZZp [[G]]-module which is (topologically) generated by the symbols df ,
f ∈ ZZp [[G]], and then take the quotient by the closed submodule generated by
the relations
d(af + bg) − adf − bdg
d(f g) − df ·α(g) − f ·dg
where f, g ∈ ZZp [[G]], a, b ∈ ZZp . (In particular, these relations imply that d1 = 0
for the unit element 1 ∈ G.) The map
d : ZZp [[G]] −→ Ω ZZp [[G]] , f 7−→ df ,
is a derivation and the pair (Ω ZZp [[G]] , d) satisfies the universal property:
For every derivation D : ZZp [[G]] → A, there exists a unique continuous
ZZp [[G]]-homomorphism ϕ : Ω ZZp [[G]] → A with D = ϕ ◦ d.

(5.6.2) Definition. We call Ω ZZp [[G]] the module of (noncommutative) differ-

ential forms of ZZp [[G]].
Recall that the kernel of the augmentation map: α : ZZp [[G]] −→ ZZp is called
the augmentation ideal and denoted by IG ⊆ ZZp [[G]].
(5.6.3) Proposition. There is a canonical isomorphism

∼ I ,
ϕ : Ω ZZp [[G]] −→ G
satisfying ϕ(dσ) = σ − 1 for every σ ∈ G ⊆ ZZp [[G]].
Proof: For σ, τ ∈G, we have the identity

στ − 1 = (σ − 1) + σ(τ − 1)
in IG , so ϕ defines a homomorphism. Furthermore, the inverse map ϕ−1 :
IG −→ Ω ZZp [[G]] , σ − 1 7−→ dσ, is well-defined. 2
(5.6.4) Proposition. If F is a free pro-c-group of rank r on the generators

x1 , . . . , xr , then Ω ZZp [[F ]] is a free ZZp [[F ]]-module of rank r on the generators
dx1 , . . . , dxr .
Proof: Let G be a pro-c-group and let A be a compact ZZp [[G]]-module. We

have a split exact sequence
π
1 −→ A −→ E −→ G −→ 1
where E is the semi-direct product of G by A and the action of G on A is the
natural one.∗) Since A is a pro-p-group, E is a pro-c-group. Therefore we
have a 1−1-correspondence between the continuous homomorphic sections to
π and the derivations from ZZp [[G]] to A.∗∗)
Now let G = F be a free pro-c-group. Then, by the observation above, a
derivation from ZZp [[F ]] to A is uniquely determined by the arbitrarily chosen
images of a set of free generators of F . This proves the proposition. 2
Returning to the general case, assume that we are given an exact sequence
1 −→ H −→ G −→ G −→ 1
of pro-c-groups with the corresponding exact sequence
0 −→ I −→ ZZp [[G ]] −→ ZZp [[G]] −→ 0,
∗) The sequence corresponds to the zero element in H 2 (G, A), cf. I §2.
∗∗) cf. ex.1 in I §2.

where
I = IH ZZp [[G ]].
The following proposition is the profinite analogue of what is called the
second fundamental sequence in commutative algebra.
(5.6.5) Proposition. There exists an exact sequence of ZZp [[G]]-modules

θ
I/I 2 −→ ZZp [[G]] ⊗ ZZp [[G ]] Ω ZZp [[G ]] −→ Ω ZZp [[G]] −→ 0
and the image of θ is canonically isomorphic to H ab (p) = H1 (H , ZZp ).
Proof: Recall that for a compact ZZp [[H ]]-module M ,

MH := M/IH M = ZZp ⊗ ZZp [[H ]] M.
Furthermore, ZZp [[G ]] is a (compactly) induced H -module and therefore ho-
mologically trivial.∗) Now we apply H (H , −) = Tor ZZp [[G ]] (ZZp [[G]], −) to
. .
the exact sequence
aug
0 −→ IG −→ ZZp [[G ]] −→ ZZp −→ 0.
Via the identification ZZp [[G]] = ZZp [[G ]]/I and by (5.6.3), we obtain the com-
mutative diagram with exact rows:
0¾¿ÀÁÂÃÄÅÆÇÈÉÊË IH ZZp [[G ]]/IH IG IG /IH IG
0 H1 (H , ZZp ) ZZp [[G]] ⊗ ZZp [[G ]] Ω ZZp [[G ]]
ZZp [[G ]]/IH ZZp [[G ]] ZZp 0
ZZp [[G]] ZZp 0.

The inclusion (IH ZZp [[G ]])2 ⊆ IH IG then finally implies the existence of
θ : I/I 2 −→ IH ZZp [[G ]]/IH IG . 2
Now assume that G is a finitely generated pro-c-group. Then we have an

exact sequence
1 −→ R −→ Fd −→ G −→ 1
for some d ∈ IN, where Fd is a free pro-c-group of rank d and R is the normal
subgroup of Fd generated by the relations of G (with respect to the chosen
∗) In other words, the Pontryagin dual of ZZ [[G]] is an induced discrete G-module, hence a
p
cohomologically trivial G-module.

generators of G, i.e. the images of a basis of Fd in G). The abelian pro-p-

group Rab (p) is a ZZp [[G]]-module in a natural way and we call it the p-relation
module of G with respect to the given presentation of G. The following
theorem is a profinite analogue of a theorem of Lyndon for discrete groups.
(5.6.6) Theorem. There is a canonical exact sequence

0 −→ Rab (p) −→ ZZp [[G]]d −→ ZZp [[G]] −→ ZZp −→ 0.
In particular, if cdp G ≤ 2, then Rab (p) is a projective ZZp [[G]]-module.
Proof: We apply (5.6.5) to the exact sequence 1 → R → Fd → G → 1 and

observe that Ω ZZp [[G]] = IG by (5.6.3) and that Ω ZZp [[Fd ]] is a free ZZp [[Fd ]]-module
of rank d by (5.6.4). The last assertion follows from (5.2.13). 2
We now consider the following general problem: Given an exact sequence

of pro-c-groups
1 −→ H −→ G −→ G −→ 1,
what can we say about the structure of H ab (p) as a ZZp [[G]]-module?
Theorem (5.6.6) gives us information in the case that G is a free pro-c-

group. In the general case, assume that G is finitely generated and choose a
presentation F G of G by a free pro-c-group F of rank d. Then we obtain
a commutative diagram
ÌÍÎÏÐÑÒÓÔÕÖ×ØÙÚÛÜÝ 1 1
N N
1 R F G 1
1 H G G 1
1 1
where R and N are defined by the exactness of the corresponding sequences.
In addition, we set
X : = H ab (p),
Y : = ZZp [[G]] ⊗ ZZp [[G ]] Ω ZZp [[G ]] = IG /IH IG .

(5.6.7) Proposition. With the notation above, we have a commutative exact

diagram
Þßàáâãäåæçèéêëìíîïðñò 0 0
ab
0 H2 (H , ZZp ) NH (p) Rab (p) X 0
ab
0 H2 (H , ZZp ) NH (p) ZZp [[G]]d Y 0
IG IG
0 0.
In particular, there is an exact sequence
0 −→ Rab (p) −→ X ⊕ ZZp [[G]]d −→ Y −→ 0.
Furthermore, if cdp G ≤ 2, then N ab (p) is a projective ZZp [[G ]]-module and

NHab
(p) is a projective ZZp [[G]]-module. If cdp G ≤ 2 and cdp G ≤ 1, then
H2 (H , ZZp ) is a projective ZZp [[G]]-module.
Proof: The upper horizontal sequence is the homological form of the five
term exact sequence (1.6.7) for the group extension 1 → N → R → H → 1
and the module ZZp . The zero on the left follows because cdp R ≤ cdp F = 1.
We obtain the lower horizontal sequence by taking H -homology of the exact
sequence
(∗) 0 −→ N ab (p) −→ ZZp [[G ]]d −→ IG −→ 0
(apply (5.6.6) to 1 → N → F → G → 1), using the homological H -triviality

of ZZp [[G ]] and the resulting isomorphism H1 (H , IG ) ∼
= H2 (H , ZZp ).
The right-hand vertical sequence is the H -homology sequence associated to
0 → IG → ZZp [[G ]] → ZZp → 0 (using the same argument as above). Finally, the
left-hand vertical sequence is the sequence from theorem (5.6.6). The diagram
commutes since all arrows are the natural ones and we use the compatible
identifications
ZZp [[G]]d = ZZp [[G]] ⊗ ZZp [[F ]] Ω ZZp [[F ]]
= ZZp [[G]] ⊗ ZZp [[G ]] ZZp [[G ]] ⊗ ZZp [[F ]] Ω ZZp [[F ]] = H0 (H , ZZp [[G ]]d ).

Considering the pull back extension X 0 of

0óôõö÷øù Rab (p) ZZp [[G]]d IG 0
X0 Y,
0 ∼ d
we see that X = ZZp [[G]] ⊕ X, which gives us the exact sequence asserted in
the proposition.
Finally, we assume cdp G ≤ 2. Then by (5.2.13) and the exact sequence (∗),
we conclude that N ab (p) is a projective ZZp [[G ]]-module, so that NH ab
(p) is a
projective ZZp [[G]] -module. If, in addition, cdp G ≤ 1, then pd ZZp [[G]] ≤ 2 (see
§2 ex.5 and (5.2.13)), and so pd ZZp [[G]] H2 (H , ZZp ) = pd ZZp [[G]] X − 2 = 0. 2
(5.6.8) Corollary. With the notation above, we assume in addition that G is a

finitely generated pro-p-group with finitely many defining relations. Let
1 −→ N −→ F −→ G −→ 1
be a minimal presentation of G by a free pro-p-group F . Then NH ab
is a finitely
generated free ZZp [[G]]-module and
ab
rank ZZp [[G]] NH = dimIFp H 2 (G , IFp ).

0 −→ H 1 (G , IFp ) −→
∼
H 1 (F, IFp ) −→ H 1 (N, IFp )G −→ H 2 (G , IFp ) −→ 0
it follows that
dimIFp (NH ab
)/G = dimIFp H 1 (N, IFp )G = dimIFp H 2 (G , IFp ),
ab
and so NH is finitely generated as a ZZp [[G]]-module. By (5.6.7) the ZZp [[G]]-
ab
module NH (p) is projective, and so it follows from (5.2.20) that it is free.
2
Now we give a description of the ZZp [[G]]-module Y = IG /IH IG up to homo-

topy equivalence in terms of the p-dualizing module D2(p) = lim
−→ m
D2 (ZZ/pm ZZ)
of G .
(5.6.9) Proposition. Assume that cdp G = 2 and that N ab (p) is a finitely

generated ZZp [[G ]]-module. Let Z = (D2(p)H )∨ . Then
Y ' DZ,
so Y is determined by Z up to projective summands. If, in addition, the group
H2 (H , ZZp ) vanishes, then
E 1 (Y ) ∼
=Z.

Proof: Applying (5.6.6) to the exact sequence 1 → N → F → G → 1, we

see that the finitely generated ZZp [[G ]]-module N ab (p) is projective. Thus ZZp
possesses a resolution by finitely generated projective ZZp [[G ]]-modules and we
get an exact sequence
(ZZp [[G ]]d )+ −→ N ab (p)+ −→ E 1 (IG ) −→ 0.
Applying (5.4.13)(ii) to ZZp , we obtain
E 1 (IG ) ∼
= E 2 (ZZp ) = (lim
−→
D2 (ZZ/pm ZZ))∨ = (D2(p) )∨ .
m
Taking H -coinvariants of the sequence above, yields the exact sequence
(ZZp [[G]]d )+ −→ NH
ab
(p)+ −→ Z −→ 0
(here we use Hom ZZp [[G ]] (M, ZZp [[G ]])H = Hom ZZp [[G]] (MH , ZZp [[G]]) for a
finitely generated projective ZZp [[G ]]-module M ). From the exact sequence
ab
NH (p) −→ ZZp [[G]]d −→ Y −→ 0
ab
and the fact that NH (p) is projective, we now obtain DY ' Z and Y ' DZ.
If H2 (H , ZZp ) = 0, then we have the exact sequence
ab
0 −→ NH (p) −→ ZZp [[G]]d −→ Y −→ 0
inducing the exact sequence
(ZZp [[G]]d )+ −→ NH
ab
(p)+ −→ E 1 (Y ) −→ 0 .
Thus E 1 (Y ) ∼
= Z. 2
In the following, we will make use of some well-known facts about group
algebras of finite groups, which we briefly recall.
(5.6.10) Theorem. Let R be a complete discrete valuation ring with quotient

field K of characteristic zero and residue field k of positive characteristic,
and let G be a finite group. Furthermore, let L, M, N be finitely generated
R[G]-modules. Then the following hold:
(i) If M ⊕ L ∼
= N ⊕ L, then M ∼ = N.
(ii) If M and N are projective and M ⊗ K ∼
= N ⊗ K as K[G]-modules, then
∼
M = N.
(iii) If M and N are projective and M ⊗ k ∼
= N ⊗ k as k[G]-modules, then
∼
M = N.
Assertion (i) is a consequence of the Krull-Schmidt-Azuyama theorem, see

[32], §6, cor.6.15. For (ii) we refer the reader to [214], chap.16.1, cor. 2 of

thm. 34. Finally, (iii) follows in a straightforward manner from Nakayama’s

lemma.
Returning to the profinite situation, one often wants to determine the ZZp [[G]]-
module structure of Y = IG /IH IG not only up to homotopy equivalence but
up to isomorphism. The following proposition is useful in the applications.
(5.6.11) Proposition. Let G be a profinite group. Let M and N be finitely

generated ZZp [[G]]-modules such that
(i) M ' N ,
(ii) M ⊗ Qp ∼= (N ⊕ ZZp [[G]]m ) ⊗ Qp for some m ∈ IN.
Then
M∼
= N ⊕ ZZp [[G]]m .
In particular, a finitely generated projective ZZp [[G]]-module P is free if and

only if P ⊗ Qp is (ZZp [[G]] ⊗ Qp )-free. Furthermore, instead of (ii), it suffices
to assume the weaker condition
(ii)0 MU ⊗ Qp ∼
= NU ⊗ Qp ⊕ Qp [G/U ]m
for all open normal subgroups U of G.
Proof: Since M ' N ' N ⊕ ZZp [[G]]m , it follows from (5.4.3)(ii) that there
are finitely generated projective ZZp [[G]]-modules P1 and P2 such that
M ⊕ P1 ∼ = N ⊕ ZZp [[G]]m ⊕ P2 .
From the second assumption (ii) (or from (ii)0 ) and by (5.6.10)(i), we get
(P1 )U ⊗ Qp ∼ = (P2 )U ⊗ Qp
for all open normal subgroups U of G. Hence the projective ZZp [G/U ]-modules
(P1 )U and (P2 )U are isomorphic by (5.6.10)(ii). Again by (5.6.10)(i), we obtain
MU ∼= NU ⊕ ZZp [G/U ]m .
In particular, we have, for every n ∈ IN and every open normal subgroup
U ⊆ G, an isomorphism of finite ZZ/pn ZZ[G/U ]-modules
(M/pn )U ∼= (N/pn )U ⊕ ZZ/pn ZZ[G/U ]m .
Passing to the projective limit, the result follows by the usual compactness
argument. 2


1 −→ H −→ G −→ ZZp −→ 1
be an exact sequence of a profinite groups such that H is of order prime to p.
Then the augmentation ideal IG of ZZp [[G]] is a free ZZp [[G]]-module of rank 1,
i.e.
IG ∼= ZZp [[G]] .
Proof: Since cdp G = 1, the ZZp [[G]]-module IG is projective by (5.2.13). Let

U ⊆ G be an open normal subgroup. From the exact sequence
0 −→ H1 (U, ZZp ) −→(IG )U −→ ZZp [G/U ] −→ ZZp −→ 0 ,
using Maschke’s theorem (2.6.12) and observing that H1 (U, ZZp ) = U ab (p) is
isomorphic to ZZp , we obtain
Qp ⊕ (IG )U ⊗ Qp ∼
= Qp [G/U ] ⊕ Qp ,
and by (5.6.10)(i), we get (IG )U ⊗ Qp ∼
= Qp [G/U ]. The result now follows
from (5.6.11). 2

1 −→ H −→ G −→ G −→ 1
be an exact sequence of profinite groups, where G is finitely generated of
cdp G ≤ 2 and G is a pro-p-group of cdp G ≤ 2 with finite relation rank
r(G) = dimIFp H 2 (G, ZZ/pZZ) < ∞. Furthermore, assume that
H2 (G , ZZp ) = 0 = H2 (H , ZZp ).
In particular, this is fulfilled if scdp G ≤ 2.
Then X = H ab (p) is a finitely generated ZZp [[G]]-module with pd ZZp [[G]] X ≤ 1
(i) X is free as a ZZp [[G]]-module.
(ii) XG is free as a ZZp -module.
(iii) The map p G ab −→ p Gab is injective.
Proof: Since cdp G ≤ 2, we have an exact sequence

p
0 −→ H2 (G, ZZp ) −→ H2 (G, ZZp ) −→ H2 (G, ZZ/pZZ) ,
and from our assumption, we know that dimIFp H2 (G, ZZ/pZZ) = r(G) is finite.
Thus H2 (G, ZZp ) is free of finite rank as a ZZp -module. The Hochschild-Serre
sequence
0 −→ H2 (G, ZZp ) −→ XG −→ G ab (p) −→ Gab −→ 0

(using H2 (G , ZZp ) = 0) shows that XG is finitely generated as a ZZp -module and

that XG is ZZp -free if and only if ker (G ab (p) → Gab ) is ZZp -torsion-free. This
proves the equivalence (ii) ⇐⇒ (iii).
Since H2 (H , ZZp ) = 0 and noting that cdp G , cdp G ≤ 2, proposition (5.6.7)
shows that
ab
0 −→ NH (p) −→ Rab (p) −→ X −→ 0
is a projective resolution of X of length 1, i.e. pd ZZp [[G]] X ≤ 1.
Recall that ZZp [[G]] is a local ring since G is a pro-p-group by (5.2.16)(iii),
and therefore finitely generated projective ZZp [[G]]-modules are free by (5.2.20).
Since XG is finitely generated as a ZZp -module, Nakayama’s lemma (5.2.18)
implies that X is a finitely generated ZZp [[G]]-module.
Now assume that XG is ZZp -free and let
0 −→ P −→ ZZp [[G]]r −→ X −→ 0
be a minimal resolution of X, i.e. (ZZp [[G]]r )G = ZZrp →
∼ X . We obtain an
G
∼
isomorphism H1 (G, X) = PG . Since cdp G ≤ 2 and H2 (G , ZZp ) = 0, the
Hochschild-Serre spectral sequence shows that
2 ∞
H1 (G, H1 (H , ZZp )) = E1,1 = E1,1 = 0.
Thus PG = 0 and therefore the compact module P is trivial by Nakayama’s
lemma . This proves the nontrivial implication (ii) ⇒ (i). 2
(5.6.14) Corollary. Assertions (i) – (iii) of (5.6.13) are true if cdp G ≤ 1, and
in this case
rank ZZp [[G]] X = dimIFp H 1 (G , IFp ) − d(G) + r(G) ,
where d(G) = dimIFp H 1 (G, IFp ) and r(G) = dimIFp H 2 (G, IFp ).
Proof: Since cdp G ≤ 1, we have scdp H , scdp G ≤ 2. Furthermore,

pG = (H 2 (G , IFp ))∨ = 0
ab
showing that (iii) is true. The Hochschild-Serre sequence

0 −→ H 1 (G, IFp ) −→ H 1 (G , IFp ) −→ H 1 (H , IFp )G −→ H 2 (G, IFp ) −→ 0
implies that
dimIFp XG /p = dimIFp H 1 (H , IFp )G
= dimIFp H 1 (G , IFp ) − d(G) + r(G).
Since X is ZZp [[G]]-free, we have dimIFp XG /p = rank ZZp [[G]] X. This finishes
the proof. 2

We now consider the following situation, which is related to Iwasawa theory.

Let
1 −→ H −→ G −→ Γ −→ 1
be an exact sequence of profinite groups, where G is finitely generated and

Γ ∼= ZZp . Let Λ = ZZp [[Γ ]]. Then X = H ab (p) = H1 (H , ZZp ) is a finitely
generated Iwasawa module, since
dimIFp (X/p)Γ = dimIFp H 1 (G , ZZ/pZZ) − 1
is finite by (5.3.10). Furthermore, we assume that the dimensions
hi = dimIFp H i (G , ZZ/pZZ) , i ≤ 2 ,
are finite, and we set
2
(−1)i hi .
X
χ2 (G ) =
i=0
(5.6.15) Lemma.
(i) pdΛ X ≤ 1 if and only if H 1 (Γ, H 1 (H , Qp /ZZp )) is p-divisible.
(ii) If cdp G ≤ 2, then H2 (H , ZZp ) = H 2 (H , Qp /ZZp )∨ is a free Λ-module of
finite rank.
Proof: From (5.3.19)(i), we know that pdΛ X ≤ 1 is equivalent to the state-

ment that X Γ is ZZp -free, or dually that H 1 (Γ, H 1 (H , Qp /ZZp )) is p-divisible.
This proves (i).
The surjection H 2 (G , Qp /ZZp ) H 2 (H , Qp /ZZp )Γ (which is obtained from
the Hochschild-Serre spectral sequence noting that cdp Γ = 1) shows that the
ZZp -module H2 (H , ZZp )Γ is finitely generated, and so H2 (H , ZZp ) is a finitely
generated Λ-module by (5.3.10). It follows from (5.6.7) that H2 (H , ZZp ) is
projective, hence Λ-free by (5.2.20).
2
(5.6.16) Proposition. Suppose pdΛ X ≤ 1. Then there exists an exact sequence
0 → Λh2 −t → Λh1 −1 → X → 0 ,
where t = dimIFp (H2 (H , ZZp )/p)Γ . In particular,
rankΛ (X) = −χ2 (G ) + t.
If cdp G ≤ 2, then t = rankΛ H2 (H , ZZp ).

Proof: This follows from (5.3.20). Indeed, we have

d0 (X) = dimIFp (X/p)Γ = h1 − 1,
d1 (X) = dimIFp p (XΓ ) + dimIFp X Γ /p
= dimIFp H 1 (H , Qp /ZZp )Γ /p + dimIFp p H 1 (Γ, H 1 (H , Qp /ZZp ))
= dimIFp H 1 (G , Qp /ZZp )/p + dimIFp p H 1 (Γ, H 1 (H , Qp /ZZp )) ,
d2 (X) = dimIFp p X Γ = 0 ,
where the last equality follows from pdΛ X ≤ 1 and (5.3.19)(i). The exact
sequence
0 → H 1 (Γ, H 1 (H , Qp /ZZp )) → H 2 (G , Qp /ZZp ) → H 2 (H , Qp /ZZp )Γ → 0 ,
which is induced by the Hochschild-Serre spectral sequence noting that

cdp Γ = 1, and the fact that H 1 (Γ, H 1 (H , Qp /ZZp ))/p = 0 by (5.6.15), shows
that
d1 (X) = dimIFp H 1 (G , Qp /ZZp )/p + dimIFp p H 2 (G , Qp /ZZp )
−dimIFp p H 2 (H , Qp /ZZp )Γ .
The exact cohomology sequence
0 → H 1 (G , Qp /ZZp )/p → H 2 (G , ZZ/pZZ) → p H 2 (G , Qp /ZZp ) → 0

p
obtained from 0 → ZZ/pZZ → Qp /ZZp → Qp /ZZp → 0 now implies that
d1 (X) = dimIFp H 2 (G , ZZ/pZZ) − dimIFp p H 2 (H , Qp /ZZp ))Γ

= h2 − t .
Finally, t is equal to rankΛ H2 (H , ZZp ) if cdp G ≤ 2 by (5.6.15). 2
(5.6.17) Theorem. With the notation above, the following assertions are
equivalent:
(i) X contains no finite nontrivial Λ-submodule and rankΛ X = −χ2 (G ).
(ii) H 2 (H , Qp /ZZp ) = 0 and H 2 (G , Qp /ZZp ) is p-divisible.
Proof: By (5.3.19)(i) and (5.6.16), assertion (i) is equivalent to pdΛ X ≤ 1

and t = 0, hence to the fact that H 1 (Γ, H 1 (H , Qp /ZZp )) is p-divisible and that
H 2 (H , Qp /ZZp ) = 0. But this is precisely statement (ii), which can be seen
using the exact sequence
0 → H 1 (Γ, H 1 (H , Qp /ZZp )) → H 2 (G , Qp /ZZp ) → H 2 (H , Qp /ZZp )Γ → 0 .
2

(5.6.18) Theorem. Assume that G (and hence also H ) is a pro-p-group.

Then the following assertions are equivalent:
(i) H is a free pro-p-group.
(ii) H 2 (G , Qp /ZZp ) is p-divisible, H 2 (H , Qp /ZZp ) = 0 and the µ-invariant of
X is zero.
Proof: A pro-p-group G is free if and only if H 2 (G, ZZ/pZZ) = 0 (see (3.5.17)

and (3.3.2)(ii)), thus if and only if Gab is ZZp -torsion-free and H 2 (G, Qp /ZZp )
= 0. Hence (i) is equivalent to the statement that X is ZZp -torsion-free and
H 2 (H , Qp /ZZp ) = 0. The latter occurs precisely when µ(X) = 0 and X Γ =
H 1 (Γ, H 1 (H , Qp /ZZp ))∨ is ZZp -torsion-free. Again using the exact sequence
in the proof of (5.6.17), we obtain the result. 2
(5.6.19) Corollary. In the situation of theorem (5.6.18), assume, in addition,

that cdp G ≤ 2. Let U be an open subgroup of G , V = H ∩ U and
Γ 0 = U /V ⊆ Γ . Then Y = V ab is a finitely generated Λ0 = ZZp [[Γ 0 ]]-module
(i) µ(X) = 0 and H 2 (H , Qp /ZZp ) = 0.
(ii) µ(Y ) = 0 and H 2 ( V , Qp /ZZp ) = 0.
Proof: Since cdp U = cdp G ≤ 2, the cohomology groups H 2 (G , Qp /ZZp ) and

H 2 (U , Qp /ZZp ) are p-divisible. Thus (i) (resp. (ii)) is equivalent to the freeness
of H (resp. V ) by (5.6.18). Since V is open in H and cdp H < ∞, we have
cdp V = cdp H by (3.3.5)(ii). Now (3.5.17) implies the result. 2


Arithmetic Theory


Chapter VI
Galois Cohomology
§1. Cohomology of the Additive Group
The Galois groups G = G(L|K) of Galois extensions L|K are profinite

groups and we may such use cohomology theory which we have developed in
the preceding chapters. We are particularly interested in the meaning of the
cohomology groups H n (G, A) for extensions of local and global fields, but we
first study their properties for general Galois extensions L|K.
Of particular importance is the absolute Galois group GK = G(K̄|K) of a
field K. It depends on the choice of a separable closure K|K, and is therefore
unique only up to inner automorphisms (we will return to this point in chapter
XII). By (1.6.3), however, its cohomology is independent of the choice and we
write
H n (K, A) := H n (GK , A).
In the following we will assume that all extension fields are contained in a
fixed separable closure.
The first G(L|K)-module which comes to mind is the additive group of a

Galois extension L of K. It is cohomologically trivial. In order to show this,
we may assume that G = G(L|K) is finite, see (1.2.5). Then the assertion is
trivial if char(K) = 0, because in this case L is uniquely divisible and the result
follows from (1.6.2)(c). In general, one argues as follows: the G-module L is
induced, because of the existence of a normal basis, i.e. of an element θ ∈ L
such that
M
L= Kσθ.
σ ∈G
Because of (1.3.7) and (1.2.5), we obtain the
(6.1.1) Proposition. If L|K is an arbitrary Galois extension with Galois

group G, then
H q (G, L) = 0 for all q > 0.

338 Chapter VI. Galois Cohomology
We call a field L p-closed if it has no Galois extensions of degree p. For

example, the separable closure K|K is p-closed and also the maximal p-
extension K(p)|K, i.e. the composite of all Galois extensions of p-power
order.
Let us assume now that the characteristic p = char(K) is positive. If the field
L is p-closed then the homomorphism
℘ : L −→ L, ℘(x) = xp − x,
is surjective, which can be seen as follows. Consider, for a ∈ L, the separable
polynomial f (x) = xp − x − a. If α is a root of f , then α + 1, . . . , α + p − 1 are
the other roots. Therefore, if a were not in the image of ℘, then the splitting
field of this polynomial would be a cyclic extension of L of degree p. But we
assumed L to be p-closed. We thus obtain the exact sequence
℘
0 −→ ZZ/pZZ −→ L −→ L −→ 0,
and the associated exact cohomology sequence yields, together with (6.1.1),
the
(6.1.2) Corollary. Let p = char(K) > 0. If L|K is a p-closed extension, then


 K/℘K for n = 1,
H n (G(L|K), ZZ/pZZ) =
 0 for n ≥ 2.
Applying the last corollary to the fixed field of L with respect to a p-Sylow
group of G(L|K), (3.3.6) implies the
(6.1.3) Corollary. Let p = char(K) > 0. If L|K is a p-closed extension, then

cdp G(L|K) ≤ 1 .
From the above corollaries and (3.9.1), (3.9.5), we obtain the following
theorem, which in a sense gives a complete survey of the Galois extensions of
K of p-power degree.
(6.1.4) Theorem. If p = char(K) > 0, then the Galois group G of the maximal
p-extension K(p)|K is a free pro-p-group of rank
rk(G) = dimIFp K/℘K.
(6.1.5) Corollary. If K is a countable field of characteristic p > 0 then

G(K(p)|K) is a free pro-p-group of finite or countable infinite rank.

§1. Cohomology of the Additive Group 339
Next we determine the Galois groups of the maximal p-extensions of local

and global fields of characteristic p.
(6.1.6) Lemma. Let K = k((t)) be the field of Laurent series over a perfect
field k with p = char(k) > 0. Then tk[[t]] ⊆ ℘K, in particular, ℘K is
open in the topology of K as a discrete valuation field. If R is a system of
representatives of k modulo ℘k, then the set L of Laurent polynomials,
ai ti + a0 , ai
X
L={ ∈ k for i < 0 and a0 ∈ R}
(i,p)=1
i<0
is a system of representatives of K modulo ℘K.
P∞
Proof: An easy computation shows that all power series of the form i=1 ai ti
are in the image of ℘, and for a = bp and i < 0 we have
atpi − bti = ℘(bti ).
With these observations in mind, one immediately verifies that the set L is a
system of representatives of K modulo ℘K. 2
We obtain the following
(6.1.7) Proposition. Let K = k((t)) be the field of Laurent series over a field k
with p = char(k) > 0. If k is finite or countable, then G(K(p)|K) is a free
pro-p-group of countable infinite rank. If k is uncountable, then G(K(p)|K) is
free of uncountable rank.
Proof: In order to prove the statement, we may assume that k is perfect:

otherwise we replace k by its perfect hull without changing the absolute Galois
group and the cardinalities in question. Now the result follows from (6.1.6)
and (6.1.4). 2
In the global case we obtain the
(6.1.8) Proposition. Let K be a global field of positive characteristic p > 0.

Then G(K(p)|K) is a free pro-p-group of countable infinite rank.
Proof: By (6.1.5), the group in question is a free pro-p-group of at most

countable rank. In order to prove that the rank is infinite, we may restrict

to the case K = IF(t), where IF is a finite field. All the infinitely many
representatives of IF((t)) modulo ℘IF((t)) given in (6.1.6) are in K and obviously
also independent modulo ℘K. Therefore, by (6.1.4), the rank of G(K(p)|K)
is infinite. 2
Let Kp |K denote the maximal abelian extension of exponent p = char(K),

i.e. the composite of all cyclic extensions of degree p. Its Galois group is
G(Kp |K) = G/Gp [G, G],
where [G, G] is the closure of the commutator subgroup of G = G(K(p)|K).
For the Pontryagin dual of this Galois group we obtain the isomorphism
G(Kp |K)∨ ∼ = H 1 (G, ZZ/pZZ) ∼
= K/℘K
The isomorphism associates to a ∈ K the character χa : G(Kp |K) → IFp , given
by χa (σ) = σα − α, where α is a root of the equation xp − x = a. Dually, we
obtain the isomorphism of Artin-Schreier theory (cf. [160], chap.IV, §3)
G(Kp |K) ∼ = Hom(K/℘K, Q/ZZ).
This explicit description of Kp |K has the following generalization to the
maximal abelian extension Kpn |K of exponent pn , whose Galois group is
n
G(Kpn |K) = G/Gp [G, G].
For every n ≥ 1, there exists a unique functor Wn from the category of rings
to itself, such that for every commutative ring R with unit the underlying set
of Wn (R) is Rn and the map
gh : Wn (R) −→ Rn ,
gh(a0 , . . . , an−1 ) =
n−1 n−2 n−3
(a0 , ap0 + pa1 , . . . , a0p + pa1p + p2 ap2 + · · · + pn−1 an−1 ),
is a homomorphism of rings (see [212], chap.II, §6 or [16], chap.IX, §1). The
elements (a0 , . . . , an−1 ) ∈ Wn (R) are called the Witt vectors of length n and
n−1
the elements a0 , ap0 + pa1 , . . . , a0p + · · · + pn−1 an−1 the ghost components
of (a0 , . . . , an−1 ). For R = IFp , we have a canonical isomorphism
∼ Z
Wn (IFp ) −→ Z/pn ZZ,
n−1 n−2
(a0 , . . . , an−1 ) 7−→ ã0p + ã1p p + · · · + ãn−1 pn−1 mod pn
where ãi is some lift of ai to ZZ (see [16], chap.IX, §1, no.7 ex.3). The Witt
vectors with respect to a general ring R should be seen as an abstract version
of this p-adic expansions.
For an arbitrary ring R we have W1 (R) = R, and for each n ≥ 1 an exact
sequence of abelian groups
V
0 −→ Wn (R) −→ Wn+1 (R) −→ R −→ 0,

§1. Cohomology of the Additive Group 341
where V is given by (a0 , . . . , an−1 ) 7→ (0, a0 , . . . , an−1 ) and is called the

Verschiebung. If R is an integral domain of characteristic p, i.e. pR = 0, then
we have an exact sequence
℘
0 −→ Wn (IFp ) −→ Wn (R) −→ Wn (R),
where ℘ = F − id, F ((a0 . . . , an−1 )) = (ap0 , . . . , apn−1 ), so that ker(℘) = Wn (IFp )
∼
= ZZ/pn ZZ. We obtain the following generalization of Artin-Schreier theory.
(6.1.9) Theorem (ARTIN-SCHREIER-WITT). Let K be a field of characteristic

p > 0. Then for the Galois group of the maximal abelian extension Kpn |K of
exponent pn , we have a canonical isomorphism
G(Kpn |K) ∼= Hom(Wn (K)/℘Wn (K), Q/ZZ).
Proof: Consider the separable closure K|K and the exact sequence of GK -
modules
℘
(∗) 0 −→ ZZ/pn ZZ −→ Wn (K) −→ Wn (K̄) −→ 0 ;
here the surjectivity of ℘ is obvious for n = 1, and for n > 1 it follows
recursively from the commutative and exact diagrams
0úûüýþÿ Wn (K) V
Wn+1 (K) W1 (K) 0
℘ ℘ ℘
V
0 Wn (K) Wn+1 (K) W1 (K) 0.
The exact sequence
V
0 −→ Wn (K) −→ Wn+1 (K) −→ K̄ −→ 0
shows by induction on n that the GK -module Wn (K) is cohomologically trivial,
since K is cohomologically trivial by (6.1.1). The cohomology sequence
associated to (∗) is therefore an exact sequence
℘
0 −→ ZZ/pn ZZ −→ Wn (K) −→ Wn (K) −→ H 1 (GK , ZZ/pn ZZ) −→ 0
which yields the isomorphism
H 1 (GK , ZZ/pn ZZ) = Hom(G(Kpn |K), Q/ZZ) ∼
= Wn (K)/℘Wn (K),
and, dually, the isomorphism
G(Kpn |K) ∼
= Hom(Wn (K)/℘Wn (K), Q/ZZ). 2
If K|k is a Galois extension of local or global fields, it is natural to consider

also the ring of integers OK as a Galois module. We formulate the next results
in such generality that they apply to the local and the global case as well.

Recall that the normalization B of a Dedekind domain A in a finite Galois

extension K of its quotient field k is called tamely ramified if for each prime
ideal p of A the unique prime factorization of pB in B has the form
pB = (P1 · · · Pg )e ,
with e prime to the characteristic of A/p, and such that the residue field
extensions (B/Pi ) | (A/p), i = 1, . . . , g, are separable. If K|k is an infinite
Galois extension, we say that B|A is tamely ramified if the normalization of A
in every finite normal subextension of K|k is tamely ramified.
(6.1.10) Theorem. Let A be a Dedekind domain with quotient field k, K|k

a Galois extension and B the integral closure of A in K. Then the following
conditions are equivalent
(i) B is a cohomologically trivial G(K|k)-module.
(ii) B|A is tamely ramified.
Proof: By (1.8.2), we may assume that K|k is finite. We consider the trace
map
X
TrB|A : B −→ A, x 7→ σx .
σ ∈G(K|k)
Claim: TrB|A is surjective if and only if B|A is tamely ramified.

Proof of the claim: We start with the case that A, and hence also B, is a
complete discrete valuation ring and denote the residue fields of A and B
by κ and κ0 , respectively. By Nakayama’s Lemma, the homomorphism of
finitely generated A-modules TrB|A is surjective if and only the induced map
Tr: κ0 → κ is surjective. As the inertia group acts trivially on κ0 , we have
Tr = e · Trκ0 |κ , where e is the ramification index. As Trκ0 |κ is surjective if κ0 |κ is
a separable field extension and zero otherwise, we see that TrB|A is surjective
if and only if κ0 |κ is separable and p - e, where p = char(κ).
In the general situation, TrB|A is surjective if and only if for every prime
ideal p of A the induced map
TrB|A ⊗ Âp : B ⊗A Âp −→ Âp
is surjective. As
Y
B ⊗A Âp = B̂P ,
P|p
the general case reduces to the case of complete discrete valuation rings. This
proves the claim.
We put G = G(K|k) and start by showing the implication (i)⇒(ii). If B
is cohomologically trivial, then Ĥ 0 (G, B) = 0, hence TrB|A is surjective. By

§2. Hilbert’s Satz 90 343
the claim, B|A is tamely ramified. It remains to show (ii)⇒(i). As the trace
is surjective, we have Ĥ 0 (G, B) = 0. Next we prove that H 1 (G, B) = 0. Let
a(σ) ∈ B be a 1-cocycle and let x ∈ B be such that TrB|A (x) = 1. Setting
X
b := a(σ)σx ,
σ ∈G
we obtain for τ ∈ G,
X X
τb = τ a(σ)(τ σx) = (a(τ σ) − a(τ ))(τ σx) = b − a(τ )TrB|A (x) .
σ ∈G σ ∈G
Therefore a(τ ) = (1 − τ )b , hence a(τ ) is a 1-coboundary.

Since the arguments above hold in the same way for all subgroups of G(K|k),
the result follows from (1.8.4). 2
In the case of a finite, cyclic Galois group, it is easy to calculate the Herbrand
index.
(6.1.11) Proposition. Let A be a Dedekind domain with quotient field k, K|k

a finite, cyclic Galois extension and B the integral closure of A in K. If A/m
is finite for all maximal ideals m ⊆ A, then the Herbrand index h(G(K|k), B)
is trivial, i.e.
#Ĥ −1 (G(K|k), B) = #Ĥ 0 (G(K|k), B),
and both groups are finite.
Proof: Let G(K|k) = {σ1 , . . . , σn }. We consider a normal basis of K|k, i.e.

we choose an α ∈ K such that σ1 α, . . . , σn α is a k-basis of K. For suitably
chosen a ∈ k × , we have an inclusion
B ⊆ Aaσ1 α + . . . + Aaσn α
of B into a free A[G(K|k)]-module. The quotient has finite A-length and is
therefore finite. By (1.7.5) and (1.7.6), we obtain h(G(K|k), B) = 1. 2
§2. Hilbert’s Satz 90
Again let L|K be a Galois extension and G its Galois group. The multi-
plicative group L× is also a G-module. It is only cohomologically trivial in
exceptional cases, but we always have

(6.2.1) Theorem (Hilbert’s Satz 90). H 1 (G, L× ) = 1.
Proof: By (1.2.5), we may assume that L|K is finite. Let a : G → L× be an

inhomogeneous 1-cocycle. For c ∈ L× we put
X
b= a(σ)σc.
σ ∈G
Since the automorphisms σ ∈ G are linearly independent (see [17], chap.5, §7,
No.5), we may choose c ∈ L× in such a way that b is non-zero. For τ in G we
obtain
a(τ )−1 a(τ σ)τ σc = a(τ )−1 b.
X X
τ (b) = τ (a(σ))τ σc =
σ ∈G
Thus a(τ ) = bτ (b)−1
for all τ ∈ G, i.e. a is a coboundary. 2
In the preceding section we obtained from the equality H 1 (G, L) = 0 the

theorem of Artin-Schreier-Witt for abelian extensions of exponent pn , where
p = char(K). From the vanishing of H 1 (G, L× ) we obtain a similar result,
called Kummer theory.
Let n be a natural number, not divisible by the characteristic of K. We
denote the group of n-th (resp. of all) roots of unity in the separable closure K
of K by µn (resp. µ). As H 1 (GK ,K × ) = 1, we obtain from the exact sequence
n
1 −→ µn −→K × −→K × −→ 1
the exact sequence
n δ
K × −→ K × −→ H 1 (GK , µn ) −→ 0,
H 1 (GK , µn ) ∼
= K × /K ×n .
Let Kn |K be the maximal abelian extension of exponent n, i.e. the composite
of all finite abelian extensions L|K in K of degree dividing n. If µn ⊆ K, we
obtain isomorphisms
Homcts (G(Kn |K), µn ) ∼= H 1 (GK , µn ) ∼= K × /K ×n .
Dually, we obtain the
(6.2.2) Theorem. If n ≥ 1 is prime to the characteristic of K and µn ⊆ K,

then
G(Kn |K) ∼= Hom(K × /K ×n , µn ).
Hilbert’s Satz 90 may be extended from the multiplicative group L× of a

Galois extension L|K to the G-group GLn (L) of all invertible n × n-matrices
over L in terms of non-abelian cohomology (see I §2).

(6.2.3) Theorem. For every Galois extension L|K with Galois group
G = G(L|K) we have
H 1 (G, GLn (L)) = 1.
Proof: By proposition (1.2.5) (which also holds by the same arguments for
non-abelian cohomology), we may assume that L|K is finite. Let a be an
1-cocycle of G with values in GLn (L) and consider for a vector x ∈ Ln the
vector X
b(x) = a(σ)σx.
σ ∈G
The set {b(x), x ∈ Ln } generates the L-vector space Ln . In fact, if f is a linear
form on Ln which vanishes on the b(x), then for all λ ∈ L,
X X
0 = f (b(λx)) = f (a(σ)σλσx) = f (a(σ)σx)σλ,
σ ∈G σ ∈G
i.e. we have a linear relation between the σλ. But the automorphisms σ are
linearly independent, so that f (a(σ)σx) = 0, and since the matrices a(σ) are
invertible, we find f = 0.
From this observation, we can find vectors x1 , . . . , xn such that the yi = b(xi )
are linearly independent. If c is the matrix with columns x1 , . . . , xn , then
b := b(c) = (b(x1 ), . . . , b(xn )) = (y1 , . . . , yn ) is an invertible matrix and
X
b= a(σ)σc.
σ ∈G
As in the case n = 1, we conclude that a(σ) = b(σb)−1 , i.e. a is a coboundary.
2
We denote the G-group of all invertible n × n-matrices over L with deter-

minant equal to 1 by SLn (L).
(6.2.4) Corollary. H 1 (G, SLn (L)) = 1.

det
1 −→ SLn (L) −→ GLn (L) −→ L× −→ 1,
we obtain the exact sequence of pointed sets
det
GLn (K) −→ K × −→ H 1 (G, SLn (L)) −→ H 1 (G, GLn (L)) = 1
(see I §3 ex.8), in which det is surjective. 2
The non-abelian cohomology groups H 1 have various interesting meanings,

all of which arise from the bijection
H 1 (G, A) ∼
= TORS (A)

(see (1.2.3)). We mention the following application to the theory of quadratic

forms.
Let V be a vector space over K equipped with a quadratic form ϕ. Denote
by ϕL the canonical extension of ϕ to the L-vector space VL = V ⊗K L. The
orthogonal group O(ϕL ) of ϕL is the group of automorphisms of the pair
(VL , ϕL ), i.e. the L-automorphisms f : VL → VL with ϕL (f (v)) = ϕL (v) for all
v ∈ VL . The Galois group G = G(L|K) acts on VL by σ(v ⊗ λ) = v ⊗ σλ and
on O(ϕL ) by σ(f ) = σ ◦ f ◦ σ −1 for σ ∈ G and f : VL → VL in O(ϕL ). Thus
O(ϕL ) is a G-group.
We say that (V, ϕ) and (V 0 , ϕ0 ) are isomorphic over L if the pairs (VL , ϕL )
and (VL0 , ϕ0L ) are isomorphic. We denote by Eϕ (L|K) the set of isomorphism
classes of pairs (V 0 , ϕ0 ) which become isomorphic to (V, ϕ) over L.
(6.2.5) Proposition. We have a canonical bijection of pointed sets

Eϕ (L|K) ∼
= H 1 (G, O(ϕL )).
Proof: Let (V 0 , ϕ0 ) be a representative of a class in Eϕ (L|K) and let X(V 0 , ϕ0 )

be the nonempty set of isomorphisms f : (VL , ϕL ) →(VL0 , ϕ0L ). This is a G-set
by σ(f ) = σ ◦ f ◦ σ −1 and is equipped with a simply transitive right action of
the G-group O(ϕL ) compatible with the G-action. In other words, X(V 0 , ϕ0 )
is an O(ϕL )-torsor in the sense of I §2 and we obtain a map
(∗) Eϕ (L|K) −→ TORS (O(ϕL )).
We prove the bijectivity of this map by constructing an inverse as follows.
Let X be an O(ϕL )-torsor and let x ∈ X. Then for every σ ∈ G we have a
unique Aσ ∈ O(ϕL ) such that σx = xAσ . The function σ 7→ Aσ is a 1-cocycle.
Since O(ϕL ) ⊆ GL(VL ) and H 1 (G, GL(VL )) = {1} by (6.2.3), there exists an
L-automorphism f of VL such that
Aσ = f −1 ◦ σ(f ) for all σ ∈ G.
For the quadratic form ϕ0 = f (ϕ) we have
σ(ϕ0 ) = σ(f )(σ(ϕ)) = σ(f )(ϕ) = (f ◦ Aσ )(ϕ) = f (ϕ) = ϕ0 .
Hence ϕ0 is rational over K, i.e. a quadratic form on V . Associating to X the
isomorphism class of the pair (V, ϕ0 ), we get a map TORS (O(ϕL )) → Eϕ (L|K).
A straightforward check, which we leave as an exercise for the reader, shows
that this map is inverse to (∗). The bijection
Eϕ (L|K) ∼ = H 1 (G, O(ϕL ))
is now the composite of (∗) with the bijection TORS (O(ϕL )) ∼
= H 1 (G, O(ϕL ))
of (1.2.4). 2

Remark: The proof shows that the bijection Eϕ (L|K) ∼ = H 1 (G, O(ϕL )) is
0 0
explicitly given by associating to a pair (V , ϕ ) the class of the 1-cocycle Aσ
given by σ(f ) = f Aσ , where f is an isomorphism (VL , ϕL ) →(VL0 , ϕ0L ).
Another application of non-abelian cohomology is to the G-group P GLn (L)

which may be defined by the exact sequence
1 −→ L× −→ GLn (L) −→ P GLn (L) −→ 1.
Let IPn−1
K be the (n − 1)-dimensional projective space over K, considered as a
K-variety. The group P GLn (L) is the automorphism group
P GLn (L) = AutL (IPLn−1 ),
where IPn−1
L = IPn−1
K ⊗K L. The Galois group G = G(L|K) acts on IPLn−1 via
the action on L and the action on P GLn (L) is induced by the action on IPLn−1 :
for σ ∈ G and A ∈ P GLn (L), we have σA = σ ◦ A ◦ σ −1 .
A Brauer-Severi variety over K of dimension n−1 is a K-variety X which
becomes isomorphic to IPn−1 over a Galois extension L|K. This means that
X ⊗K L and IPn−1
L are isomorphic as L-varieties. We then say that X splits
over L. Let us denote by BSn (L|K) the pointed set of isomorphism classes
[X] of Brauer-Severi varieties over K of dimension n − 1 which split over L.
We define a canonical map
BSn (L|K) −→ H 1 (G, P GLn (L))
as follows. Let X be an (n − 1)-dimensional Brauer-Severi variety over K
which splits over L. Then the set T (X) of all isomorphisms of L-schemes
f : IPLn−1 −→ X ⊗K L
is a P GLn (L)-torsor. In fact, on the one hand G acts on X ⊗K L via the second
factor, and hence on T (X) by σ(f ) = σ ◦ f ◦ σ −1 , and on the other hand T (X)
is equipped with a simply transitive right action of P GLn (L), compatible with
the G-action:
σ(f ◦ A) = σ ◦ f ◦ A ◦ σ −1 = σf σ −1 σAσ −1 = σ(f ) σA.
Therefore we obtain a canonical map
(1) BSn (L|K) −→ TORS (P GLn (L)), X 7−→ T (X).
The composite with the bijection (1.2.4)
TORS (P GLn (L)) ∼ = H 1 (G, P GLn (L))
gives a map
(2) BSn (L|K) −→ H 1 (G, P GLn (L)).
It is explicitly given by associating to a Brauer-Severi variety X the class of
the cocycle Aσ = f −1 ◦ σ(f ), where f is an element of T (X).

(6.2.6) Theorem. H 1 (G, P GLn (L)) ∼

= BSn (L|K).
Proof: We construct an inverse of the map (1) as follows. Let T be any

P GLn (L)-torsor and let f ∈ T . For every σ ∈ G there is a unique Aσ ∈
P GLn (L) such that σ(f ) = f Aσ . Aσ is a 1-cocycle of G with values in
P GLn (L). We change the action of σ ∈ G on IPLn−1 by σ ∗ = Aσ ◦ σ (note
that (στ )∗ = Aστ στ = Aσ σAτ στ = Aσ σAτ σ −1 στ = Aσ σAτ τ = σ ∗ τ ∗ ). In
order to indicate the changed G-action on IPLn−1 , we write Y (T ) instead. The
action of G on Y (T ) is a so-called “descent datum” and by descent theory
of algebraic geometry, there exists a K-variety X(T ) = Y (T )/G, unique up
to isomorphism, such that we have an isomorphism of L-varieties X(T ) ⊗K
L ∼= Y (T ) compatible with the G-action (see [67], chap.VIII, 7.8). This
isomorphism induces an isomorphism of P GLn (L)-torsors
T (X(T )) = IsomL (IPn−1 ∼
L , Y (T )) = T,
which is given by A 7→ f A. If we start with another f ∈ T , then we obtain
a Brauer-Severi variety over K isomorphic to X(T ). If, conversely, we start
with a Brauer-Severi variety X which splits over L, then X(T (X)) ∼ = X, since
we have an isomorphism
X(T (X)) ⊗K L ∼ = X ⊗K L
of L-varieties compatible with the G-action, and so X(T (X)) and X are both
solutions of the same descent problem. 2
Exercise 1. Let M |K be a Galois extension and L|K a finite subextension. Denote by c 7→ c̄

a section G(M |L)\G(M |K) → G(M |K) of G(M |K) → G(M |L)\G(M |K), i.e. a system of
right representatives. Associate to every inhomogeneous 1-cocycle x : G(M |L) → P GLn (L)
the function cor x : G(M |K) → P GLn (M ) given by
Y
(cor x)(σ) = c̄ −1 x(c̄σcσ −1 ).
c
Show that cor x is a 1-cocycle and that we obtain a canonical map
cor L 1 1
K : H (G(M |L), P GLn (M )) −→ H (G(M |K), P GLn (M )), [x] 7−→ [cor x],
which does not depend on the choice of the section c 7→ c̄. P GLn may be replaced by any
other algebraic group over K.
Exercise 2. Let M ⊇ L ⊇ K be as above and let 1 → A → B → C → 1 be an exact sequence

of algebraic groups over K with A in the center of B. Show that we have a commutative
diagram
C(M )) δ H 2 (G(M |L), A(M ))
H 1 (G(M |L),
cor cor
δ
H 1 (G(M |K), C(M )) H 2 (G(M |K), A(M )).

§3. The Brauer Group 349
§3. The Brauer Group
Again let L|K be a Galois extension with group G = G(L|K). We now

study the group H 2 (G, L× ). It is linked with a classical theory which began
with the discovery of the quaternion algebra
H = IR + IRi + IRj + IRk
by the Irish mathematician W. R. HAMILTON (1805–1865). The multiplication
in H is given by i2 = j 2 = k 2 = −1, ij = −ji = k (implyingnjk =−kj = i and
o
ki = −ik = j). H is also given as the matrix algebra H = −zū uz̄ | z, u ∈ C .
H has center IR and is a skew field, i.e. has no nontrivial two-sided ideal,
and it is in essence the only IR-algebra with these properties. This example
raised the question of central simple algebras over arbitrary fields K, i.e.
finite dimensional K-algebras with center K and without nontrivial two-sided
ideals. For these algebras we have the following equivalent characterizations.
(6.3.1) Proposition. For a finite dimensional K-algebra A the following

conditions are equivalent:
(i) A is a central simple algebra.
(ii) If K|K is the separable closure of K, then the K-algebra AK = A ⊗K K
is isomorphic to a full matrix algebra Mn (K).
(iii) There exists a finite Galois extension L|K such that AL = A ⊗K L is a
full matrix algebra Mn (L).
(iv) A ∼ = Mm (D), where D is a skew field over K of finite degree.
For the proof of this proposition and for the basic properties of central
simple algebras, we refer to [102], [42] and [14], chap.VIII, §5, §10. We say
that the central simple K-algebra A splits over the extension L|K, or that L is
a splitting field for A, if A ⊗K L ∼ = Mn (L) for some√ n. A is called a cyclic
algebra if it has a cyclic splitting field L|K, of degree dimK A.
Two central simple K-algebras A and B are called similar if
A ⊗K Mr (K) ∼ = B ⊗K Ms (K)
for some r, s and we write A ∼ B. This is the same to say that the skew fields
associated with A and B are isomorphic. The tensor product A ⊗K B of two
central simple K-algebras is again central simple. This leads to the
(6.3.2) Definition. The Brauer group Br(K) of a field K is the set of

all similarity classes [A] of central simple K-algebras A, endowed with the
multiplication
[A][B] = [A ⊗K B].

The product is well-defined since

[A ⊗K Mr (K) ⊗K B ⊗K Ms (K)] = [A ⊗K B ⊗K Mrs (K)] = [A ⊗K B].
The identity of Br(K) is 1 = [K] and the inverse of [A] is [A]−1 = [Aop ],
where Aop is A as a K-vector space and the multiplication of Aop is given by
Aop × Aop → Aop , (a, b) 7→ ba.
If L|K is an extension of fields, then we have the homomorphism

res K
L : Br(K) −→ Br(L), A 7−→ [A ⊗K L],
and we denote the kernel by Br(L|K). This is the group of central simple K-
algebras which split over L. If L|K runs through the finite Galois subextensions
of K|K, then by (6.3.1)(iii)
[
Br(K) = Br(L|K).
L
In the finite Galois case the classes of Br(L|K) are explicitly represented by the
following K-algebras. Let G = G(L|K) and n = [L : K]. Let x : G×G → L×
be a normalized (i.e. x(σ, 1) = x(1, σ) = 1) inhomogeneous 2-cocycle. On the
n2 -dimensional K-vector space
M
A= Leσ (eσ formal symbols)
σ ∈G
we define a multiplication by
X X X
( xσ eσ )( yτ eτ ) = xσ σyτ x(σ, τ )eστ .
σ τ σ,τ
This multiplication is associative because of the cocycle relation

x(σ, τ )x(στ, ρ) = σx(τ, ρ)x(σ, τ ρ),
and makes A a K-algebra with the identity 1 = e1 . This K-algebra is called
the crossed product of L and G by x and is denoted by
A = C(L, G, x).
The crossed products have the following properties
(i) C(L, G, x) is a central simple K-algebra which splits over L.
(ii) The normalized cocycles x and y are cohomologous if and only if
C(L, G, x) ∼
= C(L, G, y).
(iii) C(L, G, xy) ∼ C(L, G, x) ⊗K C(L, G, y).
(iv) Every simple central K-algebra which splits over L is similar to a crossed
product.

For the proof of this proposition we refer to [102]. Associating now to a

cohomology class [x] ∈ H 2 (G, L× ) the class [C(L, G, x)] (x normalized), we
obtain a map
H 2 (G, L× ) −→ Br(L|K),
which is well-defined and injective by (ii), multiplicative by (iii) and surjective
by (iv), hence an isomorphism of groups. If M ⊇ L ⊇ K are two finite Galois
extensions, then it is evident that the diagram

H 2 (G(M |K), M ×) Br(M |K)

inf incl
H 2 (G(L|K), L× ) Br(L|K)
is commutative. Taking direct limits, we obtain the
(6.3.4) Theorem. For every Galois extension L|K we have a canonical iso-
morphism
H 2 (G(L|K), L× ) ∼
= Br(L|K).
In particular,
H 2 (K,K × ) ∼
= Br(K),
showing that Br(K) is a torsion group.
In general, Brauer groups are difficult to calculate. We will determine them

for local and global fields in later chapters. For finite fields, we have the
following result.
(6.3.5) Proposition. Let K be a finite field. Then Br(K) = 0.
Proof: Let L|K be a finite extension. Then G(L|K) is a cyclic group, and
h(G(L|K), L× ) = 1 by (1.7.6). Therefore
#Br(L|K) = #H 2 (G(L|K), L× ) = #H 1 (G(L|K), L× ) = 1
by Hilbert’s Satz 90. Passing to the limit over all L, we obtain Br(K) = 0.
2
For a natural number n prime to the characteristic of K, we consider the

exact sequence n
1 −→ µn −→K × −→K × −→ 1
of GK -modules. The associated exact cohomology sequence, Hilbert’s Satz 90
and the above theorem yield an isomorphism
H 2 (K, µn ) ∼
= n Br(K).

0
Taking direct limits over n and denoting by Br(K)(p ) the subgroup of elements
of order prime to p, we obtain the
(6.3.6) Corollary. We have

H 2 (K, µ) ∼ or H 2 (K, µ) ∼
0
= Br(K) = Br(K)(p )
according to whether char(K) = 0 or char(K) = p > 0.
The Brauer group Br(K) is functorial in K in a twofold sense. If ρ : K →

K is a homomorphism of fields, then K 0 may also be viewed as a K-algebra
0
and we obtain for every K-algebra A, a K-algebra

Aρ = A ⊗K,ρ K 0
together with a homomorphism ρ : A → Aρ , a 7→ a⊗10 . Aρ is also a K 0 -algebra
and we obtain a homomorphism
ρ∗ : Br(K) −→ Br(K 0 ), [A] 7−→ [Aρ ].
If ρ is an inclusion K ⊆ L, then ρ∗ is the restriction map
res K
L : Br(K) −→ Br(L).
If L|K is a finite separable extension, we have on the other hand also a

corestriction homomorphism
cor LK : Br(L) −→ Br(K),
which is obtained as follows. Let ρ : L →K run through the K-embeddings of
L into the separable closure K of K. If A is any L-algebra, then we form the
tensor product of the L-algebras Aρ ,
O
T (A) = Aρ .
ρ
For every σ ∈ GK we have the K-isomorphism σ : Aσ−1 ρ → Aρ and we get a

GK -action on T (A), viewed as a K-algebra, given by
σ( ⊗ρ aρ ) = ⊗ρ bρ with bρ = σaσ−1 ρ .
We now take the fixed ring and obtain a K-algebra
cor LK (A) = T (A)GK .
If A is a central simple L-algebra, then cor LK (A) is a central simple K-algebra
(see [102]). It is obvious that cor LK (A ⊗L B) ∼ = cor LK (A) ⊗K cor LK (B) and
cor LK (Mn (L)) = Mnd (K) with d = [L : K]. Therefore we obtain a canonical
homomorphism
cor LK : Br(L) −→ Br(K), [A] 7−→ [cor LK (A)],
the corestriction for Brauer groups.

(6.3.7) Proposition. Let L|K be a finite subextension of the separable closure

K̄|K and let σ ∈ GK = G(K|K). We then have the commutative diagrams

H 2 (L,K ×
) ∼
Br(L)
H 2 (L,K ×
) ∼
Br(L)
res cor res cor σ∗ σ∗
∼ ∼
H 2 (K,K × ) Br(K), H 2 (σL,K × ) Br(σL).
One has to show that, for a finite Galois extension M |K containing L, the
diagrams
H 2 (G(M |L), M × ) Br(M |L) H 2 (G(M |L), M × ) Br(M |L)
res cor res cor σ∗ σ∗
H 2 (G(M |K), M × ) Br(M |K), H 2 (G(M |σL), M × ) Br(M |σL)

are commutative, where the horizontal maps are given by forming crossed
products. Let x be a normalized 2-cocycle of G(M |K), resp. of G(M |L), with
values in M × . Then by a straightforward argument
C(M, G(M |K), x) ⊗K L = C(M, G(M |L), res x),
resp.
C(M, G(M |L), x) ⊗K,σ σL = C(M, G(M |σL), σ∗ x),
giving the commutativity for the maps res and σ∗ . For cor the proof is more
involved and needs a more comprehensive study of non-abelian cohomology.
We refer to [181], th. 11 or [91], th. 3.13.20.
In I §5 we have seen that the cohomology functor

H 2 (K × ) : L 7−→ H 2 (L,K × )
may be interpreted as a G-modulation for the absolute Galois group G = GK
with respect to the maps res , cor , σ∗ in the sense of (1.5.12). The isomorphism
(6.3.6) and the above proposition (6.3.7) show that also the Brauer group may
be interpreted as a G-modulation and that we have the
(6.3.8) Theorem. The functor

Br : L 7−→ Br(L)
is a G-modulation and the family of isomorphisms H 2 (L,K × ) ∼
= Br(L) is an
isomorphism
H 2 (K × ) −→
∼ Br
of G-modulations.

The main assertion of this theorem is the double coset formula (1.5.11) for
the classes [A] of central simple algebras, i.e. the similarity
K
◦ cor LK (A) ∼ cor M σL L
Y
res M M ◦ σ∗ ◦ res Lσ −1 M (A),
σ ∈R
for two finite separable extensions M, L ⊇ K and a central simple L-algebra
A. Here R is a system of representatives of GM \GK /GL . When L = M one
may replace by and the similarity ∼ by an isomorphism
Q N
cor LK (A) ⊗K L ∼
O O
= σ∗ (A) = (A ⊗K,σ σL).
σ σ
For a K-algebra A, we have
cor LK (A ⊗K L) ∼
= A⊗d , d = [L : K].
We finish this section by giving the following geometric interpretation for the
Brauer group. In the preceding section we defined the Brauer-Severi varieties
over K as the K-varieties X which become isomorphic to IPn−1 for some n
over some Galois extension L, i.e.
X ⊗K L ∼ = IPn−1 L
as L-varieties. We denote by BS(K) the set of isomorphism classes of all

Brauer-Severi varieties over K, by BS(L|K) the subset of classes which
split over L and by BSn (L|K) the subset of BS(L|K), consisting of the
isomorphism classes of K-varieties which become isomorphic to IPn−1 over
L. Then
[
BS(L|K) = BSn (L|K).
n∈IN
(6.3.9) Theorem. For every Galois extension L|K and every n ≥ 1 there exists
a natural injective map
BSn (L|K) ,→ Br(L|K)
whose image is the set of classes in Br(L|K) which are represented by a central
simple algebra of dimension n2 over K. If [L : K] is finite and divides n, then
the above map is bijective.
Proof: The exact sequence of G(L|K)-groups

1 −→ L× −→ GLn (L) −→ P GLn (L) −→ 1
induces a map (see I §3 ex.8)
δ : BSn (L|K) ∼= H 1 (G(L|K), P GLn (L)) −→ H 2 (G(L|K), L× ).

A K-algebra A is central simple of dimension n2 and splits over L if and

only if A ⊗K L ∼ = Mn (L). Since AutL (Mn (L)) = P GLn (L), the same argu-
ments as in the case of Brauer-Severi varieties show that the set CSAn (L|K)
of K-isomorphism classes of such algebras is naturally isomorphic to
H 1 (G(L|K), P GLn (L)). Thus we obtain a diagram
H 1 (G(L|K),"#!$ P GLn (L)) δ

H 2 (G(L|K), L× )
can
CSAn (L|K) Br(L|K)
where the map can sends a central simple algebra to its similarity class in
Br(L|K). Two similar central simple algebras of the same dimension are
isomorphic and therefore can is injective. The statement of the theorem now
follows from the fact (see [102] V §30) that the above diagram commutes.
Finally, if n = m · [L : K], then the class in H 2 (G(L|K), L× ) of a 2-cocycle α
is represented by the algebra C(L, G(L|K), α) ⊗K Mm (K). 2
Exercise 1. If M ⊇ L ⊇ K are two finite separable extensions, then the corestriction of Brauer
groups obeys the rule cor L M M
K ◦ cor L = cor K .
Exercise 2. Let L|K be a cyclic extension of degree n, σ a generator of the Galois group G
and a ∈ K × . Show that the ring
n−1
M
(a, L|K, σ) := L ei
i=0
n
with multiplication e = a, eλ = (σλ)e (λ ∈ L), is a cyclic central simple algebra.
Exercise 3. Under the assumptions of ex.2 show that (a, L|K, σ) ∼ = (b, L|K, σ) provided
a/b ∈ NL|K (L× ), and
∼ Mn ((ab, L|K, σ)).
(a, L|K, σ) ⊗K (b, L|K, σ) =
Exercise 4. Keeping the assumptions of ex.2, show that the assignment a 7→ (a, L|K, σ)
θσ : K × /NL|K L× −→∼
Br(L|K).
Exercise 5. Deduce from ex.4 that the quaternion algebra H is the only central skew field
over IR different from IR (Theorem of Frobenius).
Exercise 6. (Theorem of Albert-Hochschild). If K|k is a totally inseparable extension, then

the restriction map Br(k) → Br(K) is surjective.
Exercise 7. If K is a field of characteristic p > 0, then the Brauer group Br(K) is p-divisible.

§4. The Milnor K -Groups
The Hilbert symbol is the main ingredient in the formulation of the general
reciprocity law of n-th power residues (see [160], chap.V, §3). It is defined for
a p-adic local field Kp , and is a map
,
: Kp× × Kp× −→ µn
p
which is multiplicative in both arguments a and b and satisfies the relation
( a,1−a
p
) = 1 for all a =/ 0, 1.∗) This example leads naturally to a general notion
of a symbol for any field F as a multi-multiplicative map
F × × ·{z
|
· · × F ×} −→ A, (a1 , . . . , an ) 7−→ [a1 , . . . , an ],
n times
into a (multiplicatively written) abelian group A such that
[a1 , . . . , an ] = 1 whenever ai + aj = 1 for some i =/ j.
Every such symbol factors through a group KnM (F ), which is the universal
target of symbols and is defined as follows.
(6.4.1) Definition. The n-th Milnor K-group of a field F is the quotient

KnM (F ) = (F × ⊗ ZZ · · · ⊗ ZZ F × )/In ,
where In is the subgroup of F × ⊗ ZZ · · · ⊗ ZZ F × generated by the elements
a1 ⊗ · · · ⊗ an such that ai + aj = 1 for some i =/ j.
We have the canonical symbol

F × × · · · × F × −→ KnM (F ), (a1 , . . . , an ) 7−→ {a1 , . . . , an },
where {a1 , . . . , an } = a1 ⊗ · · · ⊗ an mod In . Every other symbol of n ar-
guments is obtained from this by composition with a homomorphism of the
group KnM (F ).
It is convenient to put K0M (F ) = ZZ. The multiplication in KnM (F ) will be
written additively, i.e.
{. . . , ai bi , . . .} = {. . . , ai , . . .} + {. . . , bi , . . .},
although for = F × the multiplicative notation will also be used. For
K1M (F )
×
n, m ∈ IN the images of In ⊗ F|
· · ⊗ F ×} and F
⊗ ·{z |
×
· · ⊗ F ×} ⊗ Im in
⊗ ·{z
m times n times
∗) The reciprocity law is the product formula a,b

= 1 for two numbers a, b ∈ K × in a
Q
p p
number field K with the completions Kp (infinite primes included).

§4. The Milnor K-Groups 357
F × ⊗ ·{z
|
· · ⊗ F ×} belong to In+m . Therefore we have a homomorphism
n+m times
KnM (F ) × Km
M M
(F ) −→ Kn+m (F ),
({a1 , . . . , an }, {b1 , . . . , bm }) 7−→ {a1 , . . . , an , b1 , . . . , bm },
and we obtain a graded ring

∞
M
KnM (F ).
M
K (F ) =
n=0
One knows that KnM (IF) = 0 for a finite field IF and n ≥ 2, and that
×
K2M (Q) = µ2 ⊕ IF× × ×
3 ⊕ IF5 ⊕ IF7 ⊕ IF11 ⊕ . . .
and KnM (Q) ∼ = µ2 for n ≥ 3. If F is algebraically closed, then it is not difficult

to show that the groups KnM (F ) are uniquely divisible for n ≥ 2. In general it is
a difficult and usually most cumbersome problem to compute the K-groups or
investigate their properties. On the other hand K-theory has nowadays attained
a most prominent position in algebra, number theory and arithmetic geometry.
But it is beyond the scope of this book to go into the K-groups more closely
(we refer to [142] and [232] for further literature). We mention them here
because of their close relation to Galois cohomology. This connection arises
as follows.
Let N be a natural number prime to the characteristic of F . From the exact
sequence
N
1 −→ µN −→ F̄ × −→ F̄ × −→ 1,
we obtain a surjective homomorphism
δF : F × −→ H 1 (F, µN )
with kernel F ×N . On the other hand, we have for every n ≥ 1 the cup-product
∪
H 1 (F, µN ) × · · · × H 1 (F, µN ) −→ H n (F, µ⊗n ∗)
N ),
hence a map
(∗) |
· · × F ×} −→ H n (F, µ⊗n
F × × ·{z N ).
n times
Let us denote the image of (a1 , . . . , an ) ∈ F × × · · · × F × by

(a1 , . . . , an )F := δF a1 ∪ · · · ∪ δF an .
∗) Note that µ⊗n is isomorphic to µ as an abelian group, but as a G -module it is different:
N N F
if χ : GF →(ZZ/N ZZ)× is the character that gives the action of GF on µN , σ : ζ 7→ ζ χ(σ) , than
χn is the character for µ⊗n
N .

(6.4.2) Theorem (TATE). The map (∗) induces a homomorphism
hF : KnM (F ) −→ H n (F, µ⊗n

N ),
called the Galois symbol.
Proof: The multiplicativity in each argument is clear from the definition

and we only have to show that (a1 , . . . , an )F = 1 if ai + aj = 1 for i =/ j. It
suffices to consider the case n = 2: if n > 2 and, say i = 1, j = 2, then
(a1 , . . . , an )F = (a1 , a2 )F ∪ (a3 , . . . , an )F .
So let n = 2 and a ∈ F × , a =/ 1. Let X N − a = i fi (X) with fi (X) monic
Q
and irreducible in F [X]. For each i let ai be a root of fi (X) and let Fi = F (ai ).
Then
Y Y
1−a= fi (1) = NFi |F (1 − ai ).
i i
Hence
Y Y
(1 − a, a)F = ( NFi |F (1 − ai ), a)F = (NFi |F (1 − ai ), a)F .
i i
Because of the formula cor (α ∪ res β) = (cor α) ∪ β (see (1.5.3) (iv)), and
because cor is the norm on H 0 and commutes with δ, we have
(NFi |F (1 − ai ), a)F = cor FFi (1 − ai , a)Fi = cor FFi (1 − ai , aN
i )Fi
Fi N
= cor F (1 − ai , ai )Fi = 1,
hence (1 − a, a)F = 1. 2
For the Galois symbol, we have the fundamental
(6.4.3) Theorem (VOEVODSKY-ROST). For every field F and every N ∈ IN

prime to the characteristic of F , the Galois symbol yields an isomorphism
∼ H n (F, µ⊗n ).
hF : KnM (F )/N KnM (F ) −→ N
For N = 2 this was first conjectured by J. MILNOR in [142] (in the language
of quadratic forms) and the general form was stated by S. BLOCH and K. KATO
in [13]. For many years it was known as the Bloch-Kato conjecture.
Theorem (6.4.3) is trivial for n = 0 and arbitrary N ∈ IN, as well as for

N = 1 and arbitrary n. For n = 1 it is just a reformulation of Kummer theory.

§4. The Milnor K-Groups 359
For a proof of (6.4.3), which lies far beyond the scope of this book, we
refer to V. VOEVODSKY’s paper [Vo]. The special case n = 2 had been settled
already in 1982 by A. S. MERKUR’EV and A. A. SUSLIN, see [138], [239] or
[137]. The next steps towards the general conjecture was the case n = 3 and
N a power of 2 which was proved by Merkur’ev and Suslin (see [MS]) and,
independently, by M. ROST (see [Ro]) in 1986. In 2001, Voevodsky proved
the case with arbitrary n and N a power of 2, see [242], [243]. Finally, using
results of Rost, Voevodsky proved the general statement in 2010. There also
exists a version of (6.4.3) for N a power of the characteristic of F , which had
been proven already by Bloch and Kato themselves, see [13].
For local and global fields and n = 2 the Bloch-Kato Conjecture was proven
earlier by J. TATE (see [231] or [232]), who also constructed an `-adic variant
of the Galois symbol. Let ZZ` (n) = lim µ⊗n as a compact GF -module. Then
←− m `m
there exists an `-adic Galois symbol
hF : KnM (F ) −→ Hcts
n
(F, ZZ` (n)).
Tate showed that the following theorem follows from (6.4.3) (for n = 2).
(6.4.4) Theorem. For every prime number `, the `-primary part of K2M (F ) is
the direct sum of its maximal divisible subgroup, which is killed by hF , and a
subgroup which is mapped isomorphically by hF onto the torsion subgroup of
2
Hcts (F, ZZ` (2)).
If F is a global field, then the group K2M (F ) is a torsion group with no

non-zero divisible subgroup; so, in this case, its `-primary part is mapped
2
isomorphically onto the torsion subgroup of Hcts (F, ZZ` (2)).
Exercise 1. Let K be a field with a normalized discrete valuation v. Let O be its valuation
ring, p the maximal ideal, π a prime element and κ(v) = O/p the residue field. Define a map
∂v : K × × · · · × K × −→ Kn−1
M
(κ(v))
| {z }
n times
as follows. Let α1 , . . . , αn ∈ K , αi = π ai εi , ai = v(αi ), εi ∈ O× , and put ε̄i = εi

×
mod p ∈ κ(v).
For n = 1, put ∂(α1 ) = a1 . For n > 1 and k1 , . . . , km with 1 ≤ k1 < . . . < km ≤ n, m ≤ n,
put
∂ k1 ,...,km (α1 , . . . , αn ) = ak1 . . . akm xy,

M
where x is the symbol {ε̄1 , . . . , ε̄n } ∈ Kn−m (κ(v)) with omitted elements at the k1 , . . . , km -th
M
places if m < n, and equal to 1 if m = n; y is equal to {−1, . . . , −1} ∈ Km−1 (κ(v))

if m > 1, and equal to 1 otherwise. Now put

X
∂v (α1 , . . . , αn ) = (−1)n−k1 −···−km ∂ k1 ,...,km (α1 , . . . , αn ),
k1 ,...,km
1≤m≤n
and show that this function induces a homomorphism

∂v : KnM (K) −→ Kn−1
M
(κ(v)),
which is called the tame symbol.
Hint: [50], chap. IV, 2.
Exercise 2. Keeping the assumptions of ex.1, let {α, β} ∈ K2M (K). Then
∂v (α, β) = (−1)v(α)v(β) αv(β) β −v(α) mod p.
Exercise 3. Let K = F (X) be a rational function field in one variable and let v run through the
normalized discrete valuations of K which are trivial on F and which are different from the
valuation v∞ given by v∞ (f (X)/g(X)) = deg (g(X)) − deg (f (X)) for f (X), g(X) ∈ F [X].
The sequence
0 −→ KnM (F )% −→ KnM (K)
⊕∂v
M
M
Kn−1 (κ(v)) −→ 0
v=
/ v∞
is exact and splits (Theorem of Tate-Milnor, cf. [142]).

Hint: Study the proof in [142] or in [50] chap. IX, 2.4.
§5. Dimension of Fields
An important invariant of a field k is the cohomological dimension cd Gk of

its absolute Galois group. In order to study its properties we have to make use
of a remarkable principle, which roughly says that a homogeneous polynomial
equation f (x1 , . . . , xn ) = 0 over a field k has automatically a non-zero solution
in k if the number of variables is large compared to its degree. This principle
gives rise to another notion of dimension, which is directly attached to a field
and which we are now going to introduce.
By an n-form in k we mean a homogeneous polynomial f ∈ k[x1 , . . . , xn ].
A nontrivial zero in k is an n-tuple (α1 , . . . , αn ) ∈ k n , different from (0, . . . , 0),
such that f (α1 , . . . , αn ) = 0.
We obtain special n-forms as follows. Let K|k be a finite extension of de-
gree n and let x1 , . . . , xn be indeterminates. Then the extension K(x1 , . . . , xn )
of k(x1 , . . . , xn ) is also finite and we may consider the norm N of this extension;

§5. Dimension of Fields 361
N (ξ) is the determinant of the transformation a 7→ ξa of the k(x)-vector space

K(x). Choosing a basis ω1 , . . . , ωn of K|k,
n
X
N (x1 , . . . , xn ) := N xi ωi
i=1
defines an n-form in k of degree n which induces the norm map from K to k:

NK|k : K −→ k , (a1 , . . . , an ) 7→ N (a1 , . . . , an ) .
These forms are called norm forms. They have the special property of having
only the trivial zero in k (since NK|k (α) =/ 0 for α =/ 0). More generally, we
call an n-form f of degree d a normic form of order i if n = di and if f has
only the trivial zero in k.
(6.5.1) Lemma. If k is not algebraically closed, then k admits normic forms

of arbitrarily large degree.
Proof: Since k is not algebraically closed, there exists a finite extension K|k
of degree n > 1 for some n ∈ IN, thus a normic n-form of order 1, as we saw
above.
Let f, g be forms in n1 , n2 variables of degrees d1 , d2 respectively. We denote
by f (g | . . . | g) the form which is obtained by inserting g for each variable
and using new variables after each occurrence of |. We obtain in this way
an n1 n2 -form of degree d1 d2 , which is normic of order i if both f and g are
normic of order i. From this observation we obtain for each normic n-form f
of degree d and order i and each ν ∈ IN a normic nν -form f (ν) of degree dν and
order i by setting
f (1) = f and f (ν+1) = f (ν) (f | . . . | f ).

2
(6.5.2) Definition. The diophantine dimension dd(k) of a field k is the

smallest integer r ≥ 0 such that any n-form f of degree d ≥ 1 has a nontrivial
zero in k, whenever n > dr . We set dd(k) = ∞ if no such number r exists.
Clearly, the fields with dd(k) = 0 are the algebraically closed fields. Fields
with dd(k) ≤ 1 are called quasi-algebraically closed and fields with dd(k) ≤ i
are called Ci -fields. Thus a field is a Ci -field if every n-form of degree d with
n > di has a nontrivial zero.

(6.5.3) Theorem (ARTIN, LANG, NAGATA). Let dd(k) = r and let f1 , . . . , fs be

n-forms of degree d. If n > sdr , then these forms have a common nontrivial
zero in k.
Proof:∗) If k is algebraically closed, i.e. r = 0, then the equations f1 =

0, . . . , fs = 0 define a projective variety in the projective space IPn−1 (k) of
dimension greater than or equal to (n − 1) − s ≥ 0. The result follows from
this.
Assume then that k is not algebraically closed. Let Φ be a normic form of
order 1 and of degree e ≥ s, which exists by (6.5.1) . Consider the sequence of
forms
Φ(1) = Φ(1) (f ) = Φ(f1 , . . . , fs | f1 , . . . , fs | . . . | f1 , . . . , fs | 0, . . . , 0)

Φ(2) = Φ(2) (f ) = Φ(1) (f1 , . . . , fs | f1 , . . . , fs | . . . | f1 , . . . , fs | 0, . . . , 0)
etc., where after each vertical line we use new variables, and we insert as many
complete sets of f ’s as possible. We put Φ(0) = Φ.
Thus Φ(1) has n[ es ] variables and has degree de ≤ ds([ es ] + 1). (If x is a real
number, the symbol [x] will always mean the largest integer less than or equal
to x.) If r = 1, we want
n[ es ] > ds([ es ] + 1) or (n − ds)[ es ] > ds.
Since n − ds > 0, this can be arranged by taking e large. Then Φ(1) has a
nontrivial zero, which, since Φ is normic, is a common zero of all the fi .
If r > 1, we have to use the higher Φ(m) . Now Φ(m) has degree Dm = dm e,
and if Nm is the number of variables in Φ(m) , then N0 = e and
Nm+1 = n[ Nsm ] .
We want to choose m so large that Nm > (Dm )r . Now [ Nsm ] = Nm

s
− tm
s
, where
0 ≤ tm < s. Hence
Nm+1 n[ Nsm ] n Nm n tm
= = −
(Dm+1 )r dr (Dm )r sdr (Dm )r sdr er (dr )m
n Nm n s
≥
r r
− r r r m.
sd (Dm ) sd e (d )
Using the same inequality for m, m − 1, . . . , 2, 1, we get

∗) The proof is taken from [60].

2 ! !
Nm+1 n Nm−1 s n s

≥ − r r m−1 −
(Dm+1 )r sdr (Dm−1 ) r e (d ) sd r e (dr )m
r
.. .. ..
. . .
m+1 m m−1 !
n N0 s n 1 n n

≥ − + + ··· + 1
sdr (D0 )r er s (dr )m+1 s s
m
n

n
m+1
e s n 1 −1
= − r s .
sdr er e s (d )m+1
r n
−1
s
s2
m+1
n 1

≥ (e − ).
sdr er n−s

n
Since e can be chosen large enough that (n − s)e − s2 > 0, and sdr
> 1, we
see that (DNmm)r → ∞ as m → ∞ and we are done. 2
Note: Lang [121] generalized this theorem to the case where the fi have
different degrees d1 , . . . , ds and n > di1 + · · · + dis , but only under the extra
hypothesis that k has a normic form of order i of every degree. It would be
interesting to remove this hypothesis if possible.
We are now able to prove the following theorem, which in this generality is
due to S. LANG and in the essential case to C. TSEN.
(6.5.4) Theorem. If K|k is an extension of finite transcendence degree n, then

dd(K) ≤ dd(k) + n.
Proof: Suppose first that K|k is algebraic, i.e. n = 0. Let dd(k) = r < ∞
and let f (x1 , . . . , xm ) be an m-form in K of degree d with m > dr . We have
to show that f has a nontrivial zero in K. Since the coefficients of f lie in a
finite extension of k, we may assume that K|k is finite. Let ω1 , . . . , ωs be a
basis of K|k. Introduce new variables yij with
xi = ω1 yi1 + · · · + ωs yis ,
i = 1, . . . , m. Then f (x) = f1 (y)ω1 + · · · + fs (y)ωs , where f1 , . . . , fs are sm-
forms of degree d in k. Finding a nontrivial zero of f is equivalent to finding a
common nontrivial zero of f1 , . . . , fs in k. But this can be done by the previous
theorem, since sm > sdr .
Now let n > 0. Then K is an algebraic extension of a purely transcendental
extension k(t1 , . . . , tn ). By the above proof and by induction, we are reduced

to the case K = k(t) with an indeterminate t. By homogeneity, it suffices to

consider forms with coefficients in the polynomial ring k[t].
Suppose f (x1 , . . . , xm ) is an m-form of degree d with m > dr+1 and with
coefficients in k[t]. We have to show that it has a nontrivial zero in K.
Introduce new variables yij with
xi = yi0 + yi1 t + yi2 t2 + · · · + yis ts ,
i = 1, . . . , m, where s will be specified later. If ` is the highest degree of the
coefficients of f , we get
f (x) = f0 (y) + f1 (y)t + · · · + fds+` (y)tds+` ,
where f0 , . . . , fds+` are m(s + 1)-forms of degree d in k. We can apply theorem
(6.5.3) to these forms provided that
m(s + 1) > (ds + ` + 1)dr
or
(m − dr+1 )s > (` + 1)dr − m.
This can be satisfied by taking s large. The common nontrivial zero of the fµ ’s
in k gives a nontrivial zero of f in K = k(t). 2
(6.5.5) Corollary (TSEN). If K is a function field in one variable over an

algebraically closed field k, then dd(K) = 1.
In fact, K|k is of transcendence degree 1, hence dd(K) ≤ dd(k) + 1 = 1, and

thus dd(K) = 1, since K is not algebraically closed.
We mention the following two further results without giving the proofs.
Both may be found in [60], 6.25 and 2.3, and the second one can also be found
in [213], chap.1, §2, th.3.
(6.5.6) Theorem (LANG). A field K which is complete with respect to a discrete

valuation with algebraically closed residue field has dd(K) = 1.
(6.5.7) Theorem (CHEVALLEY - WARNING). A finite field IF is of diophantine

dimension one, i.e. dd(IF) = 1.
For the cohomological applications the fields K with dd(K) = 1 play a

particularly important role.

(6.5.8) Proposition. Let K be a field with dd(K) = 1. Then the Brauer group
Br(K) is zero, and for every finite Galois extension L|K the norm map
NL|K : L× −→ K ×
is surjective.
Proof: Let L|K be a Galois extension of degree n. PLet ω1 ,

. . . , ωn be a basis
× n
and α ∈ K . The norm form N (x1 , . . . , xn ) = N i=1 xi ωi is an n-form of
degree n, and
f (x1 , . . . , xn , x) = N (x1 , . . . , xn ) − αxn
is an (n+1)-form of degree n. It has therefore a nontrivial zero (a1 , . . . , an , a) ∈
K n . We have a =/ 0, since otherwise (a1 , . . . , an ) would be a nontrivial zero of
the norm form. Setting bi = ai /a we obtain
N (b1 ω1 + · · · + bn ωn ) = N (b1 , . . . , bn ) = α.
This shows that the norm is surjective.
We now have Ĥ n (G(L|K), L× ) = 1 for n = 0, and for n = 1 this is true by
Hilbert’s Satz 90. Since every intermediate field K 0 of L|K has dd(K 0 ) ≤ 1,
this holds also for the extension L|K 0 . From this it follows by (1.8.4) that L×
is a cohomologically trivial G-module, i.e. H 2 (G(L|K), L× ) = 0. Therefore
Br(K) ∼ = H 2 (K,K × ) = lim H 2 (G(L|K), L× ) = 0. 2
−→
L
We now turn to the cohomological dimension of fields.
(6.5.9) Definition. The cohomological dimensions cdp (k), cd(k), scdp (k) and
scd(k) of a field k are defined as the cohomological dimensions cdp G, cd G,
scdp G, scd G of its absolute Galois group G = Gk (see (3.3.1)).
The cohomological dimension cdp (k) for a field of char(k) = p > 0 plays an
exceptional role:
(6.5.10) Proposition. If k is a field of characteristic p > 0, then cdp (k) ≤ 1.
Proof: This is a special case of (6.1.3). 2
We now consider a prime number p different from the characteristic of the

field in question.

(6.5.11) Proposition. For a field k such that char(k) =/ p, and for a natural
number n ∈ IN, the following conditions are equivalent.
(i) cdp (k) ≤ n,
(ii) H n+1 (K,K × )(p) = 0 and H n (K,K × ) is p-divisible for every algebraic
extension K|k,
(iii) H n+1 (K,K × )(p) = 0 and H n (K,K × ) is p-divisible for every finite sepa-
rable extension K|k.
Proof: Let K|k be algebraic. The exact cohomology sequence associated to

the exact Kummer sequence
1 −→ µ −→K × −→K × −→ 1 ∗)
p
p
says that the condition

H n+1 (K,K × )(p) = 0 and H n (K,K × ) is p-divisible
is equivalent to H n+1 (K, µp ) = 0. Assume cdp (k) ≤ n. Since GK is iso-
morphic to a closed subgroup of Gk , we have by (3.3.5) cdp GK ≤ n, hence
H n+1 (K, µp ) = 0. This yields (i) ⇒ (ii). The implication (ii) ⇒ (iii) is trivial.
Assume that (iii) holds. Let K be the fixed field of a p-Sylow subgroup
Gp . The degree [K(µp ) : K] divides p − 1 as well as p. This means that
µp ⊆ K, i.e. µp ∼ = ZZ/pZZ as a GK -module. Writing K as a union of finite
separable extension of k which contain µp , we obtain H n+1 (K, ZZ/pZZ) = 0,
hence cdp k = cdp K ≤ n by (3.3.6) and (3.3.2)(iii) . 2
(6.5.12) Corollary. If dd(K) ≤ 2 and cd(K) = 2, then dd(K) = 2.
Proof: If dd(K) ≤ 1, then dd(K 0 ) ≤ 1 for every finite extension by (6.5.4),

hence Br(K 0 ) = 0 by (6.5.8) and thus cd(K) ≤ 1 by (6.5.11). 2
If K|k is an algebraic field extension, then by (3.3.5)

cdp (K) ≤ cdp (k),
since GK is isomorphic to a closed subgroup of Gk . If K|k is a finite extension,
then the inequality cdp (K) =/ cdp (k) is an exceptional case, as in the example
0 = cd2 (C) < cd2 (IR) = ∞.
Namely, from (3.3.5) we obtain the
∗) As before, K denotes always the separable closure of K.

(6.5.13) Proposition. If K|k is a finite extension such that cdp (k) < ∞ or
p - [K : k], then cdp (K) = cdp (k) and also scdp (K) = scdp (k).
The main result on cohomological dimension of general fields is the follow-

ing
(6.5.14) Theorem. Let k be a field such that cdp (k) < ∞ for a prime number
p =/ char(k). If K|k is a finitely generated extension of transcendence degree n,
then
cdp (K) = cdp (k) + n.
Proof:∗) The field K is a finite extension of a purely transcendental extension

k(t1 , . . . , tn ) of k. By (6.5.13) and induction on n, we are reduced to the case
K = k(t). Consider the diagram of fields
+*&')( k(t)
Gk̄(t)
k(t)
Gk(t)
k Gk
Gk k(t)
k
where k (resp. k(t)) denotes the separable closure of k (resp. k(t)). We have
G(k(t)|k(t)) = G(k|k) = Gk and we obtain a group extension
(∗) 1 −→ Gk(t) −→ Gk(t) −→ Gk −→ 1.
From (3.3.8) it follows that
cdp (k(t)) ≤ cdp (k) + cdp (k(t)).
Every finite extension K of k(t) has dd(K) = 1 by (6.5.5) and thus has a trivial
Brauer group Br(K) ∼ = H 2 (K,K × ) = 0. Since H 1 (K,K × ) = 0, we obtain
from (6.5.11) cdp (k(t)) = 1, hence
cdp (k(t)) ≤ cdp (k) + 1.
For the proof of the equality we replace k by the fixed field of a p-Sylow
subgroup of Gk . The dimensions cdp (k) and cdp (k(t)) do not change by (3.3.6).
Hence we may assume that Gk is a pro-p-group. It follows that the Gk -module
µp of p-th roots of unity is contained in k, and is hence isomorphic to ZZ/pZZ.
∗) The proof is taken from [210], chap.II, § 4.

Let d = cdp (k). We have to show H d+1 (k(t), µp ) =/ 0. To this end we consider
the Hochschild-Serre spectral sequence
E2ij = H i (Gk , H j (H, µp )) ⇒ H i+j (Gk0 , µp ),
associated to the exact sequence (∗), where k 0 = k(t) and H = G(k(t) | k(t)).
We have proved already that cdp H = 1, i.e. H j (H, µp ) = 0 for j > 1.
Therefore, by (2.1.4),
H d (Gk , H 1 (H, µp )) ∼
= H d+1 (Gk0 , µp ).
p
From the exact sequence 1 → µp → k(t) × → k(t) × → 1 we obtain an isomor-
phism of Gk -modules H 1 (H, µp ) = k(t)× /k(t)×p , and hence an isomorphism
H d+1 (Gk0 , µp ) ∼
= H d (Gk , k(t)× /k(t)×p ).
Now let vp : k(t)× → ZZ be the p-adic valuation associated to the prime
ideal p = (t) of k[t]. The valuation vp induces a surjective homomorphism
k(t)× /k(t)×p → ZZ/pZZ of Gk -modules, hence a homomorphism
H d (Gk , k(t)× /k(t)×p ) −→ H d (Gk , ZZ/pZZ),
which is again surjective because cdp Gk = d. Since Gk is a nontrivial pro-p-
group, H d (Gk , ZZ/pZZ) =/ 0 by (3.3.2)(iii). This implies H d (Gk , k(t)× /k(t)×p )
=/ 0, and so H d+1 (Gk , µp ) =/ 0. 2
Remark. The formula cdp (k(t1 , . . . , tn )) = cdp (k) + n holds true if cdp (k) = ∞
(see [8]).
Using Lang’s theorem (6.5.6), we obtain in a similar way the
(6.5.15) Theorem. Let K be a field, complete with respect to a discrete

valuation with perfect residue field k. If p =/ char(K) is a prime number and
cdp (k) < ∞, then
cdp (K) = cdp (k) + 1.
The proof is completely analogous to the above proof. One uses the exact
sequence 1 → GK̃ → GK → Gk → 1, where K̃|K is the maximal unramified
extension with Galois group G(K̃|K) ∼
= Gk .
From the last two theorems, we obtain C(X, Y ), IFq (X), Qp as examples
of fields K with cd(K) = 2. The first two fields have also diophantine
dimension 2, which follows from (6.5.8) and (6.5.12). E. ARTIN conjectured
that dd(Qp ) = 2 also. But this is not true as was shown by G. TERJANIAN
[234]. However, Qp comes close to the C2 -property in the following sense. A
field k is said to have property Ci (d) if every form of degree d in more than

di variables has a nontrivial zero in k. J. AX and S. KOCHEN [9] proved the

following result:
For every positive integer d there exists a finite set of primes, A(d),
such that Qp has property C2 (d) for all p ∈/ A(d).
One knows that Qp has the properties C2 (2) and C2 (3) for all p. One also knows
that for given p and d there exists an integer i ≥ 2 such that Qp has property
Ci (d). But Qp is not Ci for any i (see [AK]). In contrast, notice that the fields
Fp ((X)) are C2 , see [60], cor. 4.9. For general fields k it is conjectured by
J.-P. SERRE, cf. [210], 5th edition, chap. II §4.5, that the inequality
cd(k) ≤ dd(k)
should hold. In this direction Serre showed (in an exercise) that cd2 (k) ≤ dd(k)
if the Bloch-Kato conjecture holds for N = 2. As mentioned before, this
conjecture has been proven in the meantime (cf. (6.4.3)). Furthermore, we
have the following result.
(6.5.16) Theorem. If k is a C2 -field, then cd(k) ≤ dd(k).
If dd(k) ≤ 1, then dd(K) ≤ 1 for every finite extension K|k by (6.5.4), hence
Br(K) = 0 by (6.5.8), and therefore cd(k) ≤ 1 by (6.5.11). The implication
dd(k) ≤ 2 ⇒ cd(k) ≤ 2 is a result of A. S. MERKUR’EV and A. A. SUSLIN, see
[223], cor. 24.9.
Exercise: Assume k is a C2 -field, i.e. dd(k) ≤ 2.

(i) Every quadratic form of 5 variables has a nontrivial zero.
(ii) If D is a skew field with center k and finite over k, then the reduced norm Nrd : D× → k ×
is surjective.
(The reduced norm Nrd is the composite of D −→ D ⊗k K ∼
det
= Mn (K) −→ K × , where
K|k is a splitting field of D and n2 = dimk D.)


Chapter VII
Cohomology of Local Fields
§1. Cohomology of the Multiplicative Group
We now begin the development of cohomology in number theory. As a

ground field we take a nonarchimedean local field k, i.e. a field which is
complete with respect to a discrete valuation and has a finite residue field.
This covers two cases, namely p-adic local fields, i.e. finite extensions of Qp
for some prime number p, and fields of formal Laurent series IF((t)) in one
variable over a finite field. For the basic properties of local fields we refer to
[160], chapters II and V. As always, k̄|k denotes a separable closure of k and
K|k the subextensions of k̄|k. vk denotes the valuation of k, normalized by
vk (k × ) = ZZ, and κ the residue field. For every Galois extension K|k we set
H i (K|k) := H i (G(K|k), K × ), i ≥ 0.
If K|k is finite, we also set Ĥ i (K|k) = Ĥ i (G(K|k), K × ) for i ∈ ZZ. The basis
for the results in this chapter is the following theorem.
(7.1.1) Theorem. (Class Field Axiom). For a finite cyclic extension K|k we
have 
 [K : k] for i = 0,
#Ĥ i (K|k) =
 1 for i = 1.
Proof: We first show that for any finite Galois extension K|k there exists
a submodule V1 ⊆ UK1 which has finite index and is cohomologically trivial.
Let G = G(K|k) = {σ1 , . . . , σn }. We consider a normal basis of K|k, i.e. we
choose an α ∈ K such that σ1 α, . . . , σn α is a k-basis of K. For suitably chosen
a ∈ k × , we find an inclusion of G-modules
M := Ok aσ1 α + . . . + Ok aσn α ⊆ OK .
We have an isomorphism M ∼ = Ok [G] of G-modules. Furthermore, M has
m
finite index in OK , hence π OK ⊆ M for some m ≥ 1, where π ∈ Ok
is a uniformizer. This implies (π m+i M ) · (π m+i M ) ⊆ π 2m+2i OK ⊆ π m+2i M .
Therefore the subsets

372 Chapter VII. Cohomology of Local Fields
Vi := 1 + π m+i M ⊆ UK1
are G-submodules of finite index and we have isomorphisms of G-modules
∼ M/πM,
Vi /Vi+1 −→ vi 7−→ π −m−i (vi − 1) mod πM
for all i ≥ 1. Since Vi /Vi+1 ∼= (Ok /π)[G] is cohomologically trivial, the finite
G-modules V1 /Vi are cohomologically trivial for all i ≥ 1, and so is their
inverse limit V1 ∼= lim
←− i
V1 /Vi .
Now assume that G is cyclic. For the Herbrand index of UK , we obtain
h(G, UK ) = h(G, V1 ) · h(G, UK /V1 ) = 1
by (1.7.5) and (1.7.6). The exact sequence 0 → UK → K × → ZZ → 0 implies
h(G, K × ) = h(G, ZZ) = #G. Finally, H 1 (G, K × ) = 0 by Hilbert’s Satz 90, and
so #Ĥ 0 (G, K × ) = #G = [K : k]. 2
Next we show the vanishing of the cohomology of the unit group in unram-
ified extensions.
(7.1.2) Proposition. Let K|k be a Galois extension. Then the following holds.
(i) If K|k is unramified, then the group of units UK and the group of principal
units UK1 are cohomologically trivial G(K|k)-modules.
(ii) If K|k is tamely ramified, then the group of principal units UK1 is a
cohomologically trivial G(K|k)-module.
Proof: By a direct limit argument, we may assume that K|k is finite. Let
K|k be unramified and let λ|κ be the associated residue extension. Then
G(K|k) ∼
= G(λ|κ), and
UKi /UKi+1 ∼
=λ
is a cohomologically trivial G(K|k)-module for all i ≥ 1 by (6.1.1). It follows
by induction that the finite G-module UK1 /UKi is cohomologically trivial for all
i ≥ 1. Since the cohomology of a finite group with values in an inverse limit of
finite modules is the inverse limit of the cohomology groups, the isomorphism
UK1 ∼= lim
←−
UK1 /UKi implies that
i
H (G(K|k), UK1 ) = lim H n (G(K|k), UK1 /UKi ) = 0 for all n ≥ 1.
n
←− i
Hence UK1 is cohomologically trivial. Finally, by Hilbert’s Satz 90 and the

vanishing of the Brauer group of a finite field (6.3.5), λ× is cohomologically
trivial, and the exact sequence
0 −→ UK1 −→ UK −→ λ× −→ 0
shows that UK is also cohomologically trivial.

§1. Cohomology of the Multiplicative Group 373
If K|k is tamely ramified, let K 0 be the maximal unramified extension of k

inside K. Since UK1 is a ZZp -module and the order of G(K|K 0 ) is prime to p,
we have H j (K|K 0 , UK1 ) = 0 for j > 0, and the Hochschild-Serre spectral
sequence
E2ij = H i (K 0 |k, H j (K|K 0 , UK1 )) ⇒ H i+j (K|k, UK1 )
induces natural isomorphisms
H i (K 0 |k, UK1 0 ) →
∼ H i (K|k, U 1 )
K
for all i ≥ 0. Therefore (i) implies (ii). 2
We now consider the maximal unramified extension k̃|k. Its Galois group
Γk = G(k̃|k) is topologically generated by the Frobenius automorphism ϕk and
is canonically isomorphic to ẐZ; ϕk corresponds to 1 ∈ ẐZ. From the two exact
sequences
vk̃
0 −→ Uk̃ −→ k̃ × −→ ZZ −→ 0,
0 −→ ZZ −→ Q −→ Q/ZZ −→ 0,
in which Uk̃ and Q are cohomologically trivial Γk -modules, we obtain the
isomorphisms
v δ −1 ϕ
H 2 (k̃|k) −→k̃ 2
Z) −→
∼ H (Γk , Z
1
Z) −→
∼ H (Γk , Q/Z Z,
∼ Q/Z
where ϕ is given by ϕ(χ) = χ(ϕk ). We denote the composition by

invk̃|k : H 2 (k̃|k) −→
∼ Q/Z
Z.
For a finite separable extension K|k we have a commutative diagram
/.-,
H 2 (K̃|K) inv
Q/ZZ
res [K:k]
inv
H 2 (k̃|k) Q/ZZ,
where the map res is induced by the compatible pair ΓK → Γk , k̃ × ,→ K̃ × .
Indeed, this follows directly from the definition of the map inv and the diagram
6978423501
H 2 (K̃|K) H 2 (ΓK , ZZ) δ −1
H 1 (ΓK , Q/ZZ) Q/ZZ
res eK|k ·res eK|k ·res eK|k ·fK|k
δ −1
H 2 (k̃|k) H 2 (Γk , ZZ) H 1 (Γk , Q/ZZ) Q/ZZ,
which is commutative since the Frobenius automorphism ϕK ∈ ΓK is mapped
onto the fK|k -th power of the Frobenius automorphism ϕk ∈ Γk . Further

observe that eK|k · fK|k = [K : k]. This, and what follows, was already seen
in III §2 in a general setting.
If K|k is a Galois extension, then, because H 1 (K|k) = 0, the sequence
inf res
(∗) 0 −→ H 2 (K|k) −→ H 2 (k̄|k) −→ H 2 (k̄|K)
is exact.∗) We identify H 2 (K|k) with its image in H 2 (k̄|k). Of crucial
importance is the following
(7.1.3) Theorem. H 2 (k̄|k) = H 2 (k̃|k).
Proof: First, we claim that

#H 2 (K|k) [K : k]

for a finite Galois extension K of k. In fact, this is true if K|k is cyclic because
of H 2 ∼ = Ĥ 0 . If G(K|k) is a p-group, it follows inductively from the exact
sequence
0 −→ H 2 (L|k) −→ H 2 (K|k) −→ H 2 (K|L),
where G(K|L) is a normal subgroup of G(K|k) of order p . In the general
case, let Σp be a p-Sylow subfield of K|k. Since the restriction map
M
res : H 2 (K|k) ,→ H 2 (K|Σp )
p
is injective by (1.6.10), we obtain
Y Y
#H 2 (K|k) #H 2 (K|Σp )
[K : Σp ] = [K : k] .
p p
Let n = [K : k] and let kn be the unramified extension of k of degree n.
Then
H 2 (K|k) = H 2 (kn |k) ,
where we identify H 2 (K|k) and H 2 (kn |k) with their images in H 2 (k̄|k). In
fact, because
#H 2 (K|k) [K : k] = [kn : k] = #H 2 (kn |k),
it suffices to show the inclusion "⊇". But this follows from the exact commu-
tative diagram
0@:;<=>?AB H 2 (K|k) H 2 (k̄|k) res H 2 (k̄|K)
res
H 2 (k̃|k) H 2 (K̃|K)
invk̃|k invK̃|K
[K:k]
Q/ZZ Q/ZZ,
∗) The group H 2 (k̄|k) is often called the “Brauer group”, because it is isomorphic to the
group Br(k) of central simple algebras by (6.3.4).

in which invk̃|k and invK̃|K are isomorphisms as shown above. Since H 2 (kn |k) ⊆
H 2 (k̃|k) has order n = [K : k], it is mapped by the middle arrow res , and thus
by the upper arrow res , to zero, hence
H 2 (kn |k) ⊆ H 2 (K|k).
We therefore obtain
[ [
H 2 (k̄|k) = H 2 (K|k) = H 2 (kn |k) = H 2 (k̃|k).
K n
2
(7.1.4) Corollary. We have a canonical isomorphism
invk : H 2 (k̄|k) −→
∼ Q/Z
Z,
called the invariant map. For every finite separable extension K|k we have
the commutative diagrams
HGFEDC
H 2 (k̄|K) inv
Q/ZZ
res cor [K:k] id
inv
H 2 (k̄|k) Q/ZZ.
The commutativity for cor follows from cor ◦ res = [K : k]. If K|k is Galois,
then the exact sequence (∗) shows that invk induces an isomorphism
invK|k : H 2 (K|k) −→
∼ 1
[K:k]
ZZ/ZZ.
In other words (see (3.1.8)):
(7.1.5) Corollary. The pair (Gk , k̄ × ) is a class formation.
As a first application, we obtain (by (6.3.4)) the
(7.1.6) Corollary. For the Brauer group Br(k) of central simple k-algebras
we have a canonical isomorphism
∼ Q/Z
invk : Br(k) −→ Z,
and Br(L)(p) = 0 for any extension L|k of degree divisible by p∞ .

The last assertion follows from (7.1.4): if Lα |k runs through the finite
subextensions of L|k, then
H 2 (k̄|L)(p) = lim H 2 (k̄|Lα )(p) ∼

= lim (Q/ZZ)(p) = 0.
−→ −→
res [Lα :k]
From (3.1.4) and III §1 ex.4, we obtain the
(7.1.7) Corollary. Let K|k be a finite Galois extension with Galois group G.
The cup-product with the fundamental class uK|k ∈ H 2 (K|k) yields isomor-
phisms for i ≥ 0 :
∼ Ĥ i+2 (G, K × ),
uK|k ∪ : Ĥ i (G, ZZ) −→
uK|k ∪ : Ĥ i (G, Hom(K × , Q/ZZ)) −→

∼ Ĥ i+2 (G, Q/Z
Z).
(7.1.8) Theorem. The following assertions hold:
(i) The cohomological dimension of Gk is

(
2, if p =/ char(k),
cdp (k) =
1, if p = char(k),
and cdp (L) ≤ 1 for every extension L|k of degree divisible by p∞ .
(ii) Suppose the characteristic of k does not divide n ∈ IN. Then we have
 × ×n
 k /k


for i = 1,
i 1
H (k, µn ) = n
ZZ/ZZ for i = 2,



0 for i ≥ 3.
(iii) If A is a finite Gk -module of order prime to char(k), then the groups

H i (k, A) are finite for all i ≥ 0.
(iv) Let A be a Gk -module which is finitely generated as a ZZ-module and

assume that the order of tor(A) is prime to char(k). Then H 1 (k, A) is
finite.
Proof: (i) Since H 1 (L, k̄ × ) = 0 and H 2 (L, k̄ × )(p) = 0 as we have just seen, we
obtain from (6.5.11) that cdp (L) ≤ 1. We apply this to the maximal unramified
extension k̃|k. Let Γ = G(k̃|k) and let A be a p-torsion Gk -module. We have
cd (k̃) ≤ 1, hence H j (k̃, A) = 0 for j > 1. Therefore the Hochschild-Serre

spectral sequence
E2ij = H i (Γ, H j (k̃, A)) ⇒ H i+j (k, A)
together with cdp Γ = 1 yields by (2.1.4)
H i+1 (k, A) ∼
= H i (Γ, H 1 (k̃, A)) = 0 for i ≥ 2,
hence cdp (k) ≤ 2. If p =/ char(k), the equality cdp (k) = 2 follows from
H 2 (k, µp ) ∼
= ZZ/pZZ =/ 0, see below. If p = char(k), then cdp (k) = 1 by (6.1.3).
(ii) Recalling that H 0 (k, k̄ × ) = k × , H 1 (k, k̄ × ) = 1, H 2 (k, k̄ × ) ∼
= Q/ZZ, the
Kummer sequence (for n prime to char(k))
n
1 −→ µn −→ k̄ × −→ k̄ × −→ 1
yields H 1 (k, µn ) ∼
= k × /k ×n and H 2 (k, µn ) ∼
= n1 ZZ/ZZ. H i (k, µn ) = 0 for i > 2
follows from (i).
(iii) Let A be a finite Gk -module of order m prime to char(k). Then H i (k, A) =

0 for i > 2 by (i). Choose a finite Galois extension K|k in k̄ over which A
and µm become trivial Galois modules. As a GK -module, A is isomorphic to
a finite direct sum of GK -modules of type µn , n|m. Since H j (K, µn ) is finite
for all j ≥ 0, so is H j (K, A), and the spectral sequence
E2ij = H i (G(K|k), H j (K, A)) ⇒ H n (k, A)
shows that also H n (k, A) is finite as a group with a finite filtration, whose
quotients are subquotients of the finite groups on the left-hand side.
(iv) Let tor(A) denote the torsion subgroup of A. The cohomology sequence
associated to 0 → tor(A) → A → A/tor(A) → 0 and (iii) show that we may
assume A to be ZZ-free. Let K|k be a finite Galois extension over which
A is trivial, i.e. A ∼
= ZZn as a GK -module for some n. Then H 1 (K, A) =
n
H (K, ZZ) = 0 and H 1 (k, A) = H 1 (G(K|k), A) is finite.
1
2
Using (6.3.6), we obtain the following corollary.
(7.1.9) Corollary. If k is a p-adic local field, then canonically

H 2 (k, µ) ∼
= Q/ZZ,
and, if p = char(k) > 0, then
[ M
H 2 (k, µ) = 1
n
ZZ/ZZ = Q` /ZZ` .
p-n `=
/p

§2. The Local Duality Theorem
Local fields are topological fields and topological questions enter into the
game. These questions become even more important in the global theory
because of the appearance of the idèle and idèle class groups. For this reason we
premise the discussion of local and global theory with the following preparatory
Topological Remarks: Assume we are given an abelian topological group M

on which a profinite group G acts in such a way that the action is continuous
with respect to the given topology as well as with respect to the discrete
topology. Then let
A = H i (G, M )
be the i-th cohomology group of M as a discrete G-module. If i ≥ 1, we
consider these groups always as discrete topological groups. Hence the dual
H i (G, M )∨ for i ≥ 1 is always a profinite group. But for i = 0 the group
H 0 (G, M ) inherits the initial topology of M . Examples of this situation are
M = k × , where k is a local field, and the level-compact modules considered
in I §9. Suppose that
(,)
A × B −−−→ C
is a continuous bilinear pairing of topological groups. We then obtain contin-
uous homomorphisms
α
A −→ Hom(B, C), a 7−→ fa : b 7−→ (a, b),
β
B −→ Hom(A, C), b 7−→ gb : a 7−→ (a, b),
inducing bijective continuous homomorphisms
A/ ker(α) −→ im(α) and B/ ker(β) −→ im(β),
where the left groups have the quotient topology and the right ones the induced
topology. We say that the pairing A × B → C is non-degenerate if both
maps α and β are injective. Furthermore, we want to recall the notation
A∗ = Hom(A, Q/ZZ) for an abelian group A. If A is a discrete abelian torsion
group, then A∗ coincides with the Pontryagin dual A∨ , but we consider A∗ as
a discrete group, while A∨ is equipped with a natural compact topology. We
always have A∨∨ = A but the equality A∗∗ = A holds if and only if A is finite.
Corollary (7.1.5) says that for every finite Galois extension K|k of a local
field k the multiplicative group K × is a class module for the Galois group
G = G(K|k) in the sense of (3.1.3). By (7.1.4), we have even a canonical
fundamental class γ ∈ H 2 (G, K × ). Therefore, by the theorem of Nakayama-
Tate (3.1.5), we obtain the

§2. The Local Duality Theorem 379
(7.2.1) Theorem. Let K|k be a finite extension of local fields with Galois group
G. Let A be a finitely generated ZZ-free G-module and A0 = Hom(A, K × ).
Then for all i ∈ ZZ the cup-product
∪
Ĥ i (G, A0 ) × Ĥ 2−i (G, A) −→ H 2 (G, K × ) = 1
#G
ZZ/ZZ
induces an isomorphism of finite abelian groups
Ĥ i (G, A0 ) ∼
= Ĥ 2−i (G, A)∗ .
The case A = ZZ and i = 3 yields the
(7.2.2) Corollary. H 3 (G, K × ) = 0.
In the case i = 0 and A = ZZ, we have H 2 (G, ZZ)∗ ∼ = H 1 (G, Q/ZZ)∗ =

(Gab )∗∗ = Gab , and we obtain the main theorem of local class field theory, the
“local reciprocity law” (for a proof of the second statement see [212], XIV, §6):
(7.2.3) Theorem. Let k be a local field and let K|k be a finite Galois extension.
Then there is a canonical isomorphism
k × /NK|k K × ∼
= G(K|k)ab .
The norm groups NK|k K × for finite Galois extensions K|k are precisely the
open subgroups of finite index in k × .
It was this isomorphism which initiated the development of cohomology

theory in number theory. Before this development the isomorphism was ob-
tained only in a complicated and obscure way via a detour to the theory of
global fields. It was J. TATE who put the reciprocity law on a conceptual basis
in the above cohomological setting. The law may also be proved in a direct
group theoretical way (see [160], chap. IV and V).
However, theorem (7.2.1) is still too rigid for our intended applications as it
holds only for ZZ-free Galois modules and applies only to finite Galois groups.
Our aim is to prove a duality theorem of the above type for the absolute Galois
group Gk of a local field and for Gk -modules which are finitely generated
as ZZ-modules. The essential step on this path is the explicit determination
of the dualizing module D0 = D2 (ẐZ) of the category Modt (Gk ) of discrete
Gk -modules which are torsion as abelian groups (see III §4). It is defined by
D0 = lim H 2 (K, ZZ/nZZ)∗ ,
−→
K,n
where K|k runs through the finite subextensions of k̄|k, and it is characterized
by a functorial isomorphism H 2 (K, A)∗ ∼
= HomGK (A, D0 ) for A ∈ Modt (Gk ).

(7.2.4) Theorem. The Gk -module D0 is canonically isomorphic to the Gk -

module µ of all roots of unity in k̄.
Proof: Let n ∈ IN, and suppose (n, p) = 1 if p = char(k) > 0. Let n I =

n
ker(D0 → D0 ). Since D0 is also the dualizing module of every open subgroup
V of G, we obtain canonically for the Gk -module µn
HomV (µn , n I) = HomV (µn , D0 ) ∼= H 2 (V, µn )∗ ∼
= ZZ/nZZ
by (7.1.8). This shows that HomV (µn , n I) is independent of V , and we obtain
a canonical isomorphism of Gk -modules
Hom(µn , n I) ∼= ZZ/nZZ.
Let fn : µn → n I be the element corresponding to 1 mod nZZ. The other
elements of Hom(µn , n I) are fni (ζ) = fn (ζ i ), i = 0, . . . , n − 1. fn is a Gk -
homomorphism. It is injective, since it has order precisely n. It is also
surjective, since otherwise we would have an x ∈ n I r im(fn ), and hence a
homomorphism µn →(x) ⊆ n I, different from the fni . If k has characteristic
zero, we obtain a Gk -isomorphism
= D0 .
[ [
f :µ= ∼
µn −→ nI
n∈IN n∈IN
If p = char(k) > 0, this remains true, since the p-primary components of µ and
D0 are trivial, the latter because cdp Gk = 1. 2
(7.2.5) Corollary. For every prime number p we have scdp (k) = 2.
In fact, if p =/ char(k), this is true by Serre’s criterion (3.4.5), and if p =

char(k) > 0, then cdp (k) = 1 by (6.5.10), and hence scdp (k) = 2 (cf. III §3
ex.1).
As the main result of the cohomology theory of local fields we obtain the
(7.2.6) Theorem (Tate Duality). Let k be a p-adic local field . Let A be a

finite Gk -module and set A0 = Hom(A, µ). Then the cup-product
∪
H i (k, A0 ) × H 2−i (k, A) −→ H 2 (k, µ) ∼
= Q/ZZ
induces for 0 ≤ i ≤ 2 an isomorphism of finite abelian groups
H i (k, A0 ) −→
∼ H 2−i (k, A)∗ .
If k is a local field of characteristic p > 0, the same holds[true for the finite
Gk -modules A of order prime to p, except that H 2 (k, µ) ∼
= 1
n
ZZ/ZZ.
p-n

Proof: This theorem is a special case of the abstract duality theorem (3.4.6).
We have only to show that for p =/ char(k)
Di (ZZ/pZZ) = lim
−→
H i (K, ZZ/pZZ)∗ = 0
K|k
for i = 0, 1. Since every subextension K|k of k̄|k admits a subextension K 0 |K

of a degree divisible by p, we have D0 (ZZ/pZZ) = lim ZZ/pZZ = 0. Let K|k run
−→ K
through the subextensions of k̄|k. Then from the main theorem of local class
field theory (7.2.3) we obtain a surjection (which is in fact an isomorphism)
I ×p
K × /K Gab
K /p.
Passing to the direct limit, we obtain a surjection from lim K × /K ×p =
−→ K
k̄ × /k̄ ×p = 1 onto lim Gab
K /p = lim H 1 (K, ZZ/pZZ)∗ = D1 (ZZ/pZZ), and so
−→ K −→ K
D1 (ZZ/pZZ) = 0. 2
The duality theorem (7.2.6) deals with finite Gk -modules. It is desirable to

extend it to Gk -modules A which are finitely generated as ZZ-modules. The
main reason is that it may then be applied to algebraic tori: a Gk -module A
which is finitely generated and free as a ZZ-module is the character group of
the algebraic torus T = Hom(A, Gm ) over k, and A0 = Hom(A, k̄ × ) = T (k̄) is
the Gk -module of k̄-rational points of T . This generalization is accomplished
by the following computation of the dualizing module
D2 (ZZ) = lim H 2 (U, ZZ)∗
−→
U
of Gk , i.e. the dualizing object for the category Mod(Gk ) of all Gk -modules.
Let K|k be a finite separable extension and let UK denote its group of units.
UK is compact and we consider the profinite completion of K × ,
K̂ × = lim K × /N,
←−
N
where N runs through the open subgroups of finite index. We then have an
0JKLTMNOPQRS
vK
UK K× ZZ 0
v̂K
0 UK K̂ × ẐZ 0,
hence K̂ × /K × ∼
= ẐZ/ZZ, which is a uniquely divisible group. The inclusion
k ,→ K is continuous and induces an injection k̂ × ,→ K̂ × . We will consider
× ×

the injective limit lim K̂ × , where K runs through the finite subextensions of
−→
k̄|k. If e is the ramification index of K|k, then vK|k = evk and we have a
commutative diagram of topological isomorphisms
XVWU ×
K̂ × /K
v̂K
ẐZ/ZZ
∼
v̂k
k̂ × /k × ∼ ẐZ/ZZ,
noting that ẐZ/ZZ is uniquely divisible. Applying lim to the exact sequence
−→ K
1 −→ K × −→ K̂ × −→ ẐZ/ZZ −→ 1,
we obtain an exact sequence of Gk -modules
1 −→ k̄ × −→ lim K̂ × −→ ẐZ/ZZ −→ 1,
−→
K
since lim ẐZ/ZZ ∼ = ẐZ/ZZ. Since vK (x) = vK (σx) for all σ ∈ G(K|k), the
−→ e
latter group is a trivial Gk -module and, since it is uniquely divisible, it is a
cohomologically trivial Gk -module. The exact cohomology sequence therefore
yields the
H 0 (k, lim K̂ × ) = k̂ × ,
−→
K
H i (k, lim K̂ × ) = H i (k, k̄ × ) for i > 0.
−→
K
(7.2.8) Theorem. The dualizing module D2 = D2 (ZZ) of Gk is canonically

isomorphic to the Gk -module lim K̂ × .
−→
K
Proof: For every finite Galois extension K|k we have the norm residue
symbol
( , K|k) : k × /NK|k K × −→ G(K|k)ab ,
which is an isomorphism, as above. By (7.2.3), the norm groups NK|k K ×
are precisely the open subgroups of finite index in k × . Hence, passing to the
projective limit, we obtain a canonical isomorphism
( , k) : k̂ × −→ Gab ∼
= H 2 (k, ZZ)∨ .
k
For a finite Galois extension K|k, we obtain the commutative diagram

K̂_YZ[\]^ × Gab
K H 2 (K, ZZ)∨
Ver cor ∨
k̂ × Gab
k H 2 (k, ZZ)∨
(cf. [160], chap.IV, (6.4) and (1.5.9)). Taking direct limits, we get an isomor-
phism
( , k̄) : lim
−→
K̂ × −→
∼ D (Z
2 Z). 2
K
Let A be a Gk -module which is finitely generated as a ZZ-module. By tor(A)

we denote the torsion subgroup of A, which is a Gk -module. Furthermore, we
will use the following notation for the rest of this chapter:
A0 = Hom(A, k̄ × ),
AD = Hom(A, lim
−→
K̂ × ) = Hom(A, D2 ).
K
As a generalization of Tate’s local duality theorem (7.2.6) we have the
(7.2.9) Theorem. Let A be a Gk -module which is finitely generated as a

ZZ-module and assume that #tor(A) is prime to char(k). Then, for 0 ≤ i ≤ 2,
the cup-product and the map inv
∪ inv
H i (k, AD ) × H 2−i (k, A) −→ H 2 (k, k̄ × ) −→ Q/ZZ
induce an isomorphism
∼ H 2−i (k, A)∗ .
∆i : H i (k, AD ) −→
For i = 1 these groups are finite.
Proof: We know by (7.2.5) that scd(k) = 2. We have D0 (ZZ) = 0 by the

remark following (3.4.3), and, trivially, D1 (ZZ) = lim H 1 (K, ZZ)∗ = 0. If
−→
char(k) = 0, then k̄ × is divisible and so is D2 (ZZ). If p = char(k) > 0, then
D2 (ZZ) is `-divisible for all prime numbers ` =/ char(k). Therefore we obtain
the isomorphisms by Tate’s duality theorem (3.4.3) and the subsequent remark.
The finiteness of H 1 (k, A) is part (iv) of (7.1.8). 2
The Gk -module D2 = lim

−→
K̂ × looks rather awkward and one may ask
K
whether it can be replaced by k̄ × in the theorem. This is clear for a finite
module A, since Hom(A, D2 ) = Hom(A, k̄ × ) = Hom(A, µ). Furthermore, the

groups AD = Hom(A, D2 ) and A0 = Hom(A, k̄ × ) inherit the natural topologies

of D2 ∼= lim
−→
K̂ × ∗) and of k̄ × . Therefore the groups H 0 (k, A0 ) and H 0 (k, AD )
are also topological groups in a natural way.
Since H 2 (k, A) is a discrete torsion group, we may replace H 2 (k, A)∗ by the
compact topological group H 2 (k, A)∨ (which has the same underlying abstract
group) and obtain an abstract isomorphism H 0 (k, AD ) −→ ∼ H 2 (k, A)∨ between
abelian topological groups.
(7.2.10) Proposition. The duality isomorphism

∼ H 2 (k, A)∨
H 0 (k, AD ) −→
is a homeomorphism. The natural injection
H 0 (k, A0 ) ,→ H 0 (k, AD )
is continuous and H 0 (k, AD ) is the profinite completion of H 0 (k, A0 ) with
respect to the open subgroups of finite index. We have isomorphisms
H i (k, A0 ) −→
∼ H i (k, AD ) for i > 0
and the groups are finite for i = 1.
Proof: The exact cohomology sequence associated to the sequence

0 −→ tor(A) −→ A −→ A/tor −→ 0,
the finiteness of the cohomology of tor(A), and the identity tor(A)0 = tor(A)D ,
allow us to assume A to be ZZ-free. Let K|k be a finite Galois extension over
which A becomes a trivial Galois module, and let G = G(K|k) be its Galois
group. We have an isomorphism of GK -modules A ∼ = ZZN for some N ∈ IN.
Now consider the commutative diagram
H 0 (K, Aàbcdef 0 )G(K|k) H 0 (K, AD )G(K|k) (H 2 (K, A)∨ )G(K|k)
H 0 (k, A0 ) H 0 (k, AD ) H 2 (k, A)∨ .

The vertical arrow on the right is an isomorphism by (3.3.11) and the right
upper horizontal isomorphism is a homeomorphism by the definition of the
topology on D2 and because A is a trivial GK -module. The right commutative
square now gives the first statement.
By definition of K̂ × , the group H 0 (K, AD ) = Hom(A, K̂ × ) ∼
= (K̂ × )N is the
0 0 × ∼ × N
completion of H (K, A ) = Hom(A, K ) = (K ) with respect to the open
∗) The topology depends on the choice of this isomorphism. We use the natural isomorphism
constructed in (7.2.8).

subgroups of finite index, and this remains valid for the fixed modules under
G(K|k). This shows the second statement.
We now prove the maintained isomorphisms for i > 0. From the exact
sequence
0 −→ k̄ × −→ D2 −→ ẐZ/ZZ −→ 0,
0 −→ A0 −→ AD −→ Hom(A, ẐZ/ZZ) −→ 0.
Since ẐZ/ZZ is uniquely divisible, so is the Gk -module Hom(A, ẐZ/ZZ), and
hence it is cohomologically trivial. From the exact cohomology sequence it
follows that
(∗) H i (k, A0 ) ∼
= H i (k, AD )
for i > 1. For i = 1 we need a little additional argument: for any open
subgroup H ⊆ Gk which acts trivially on A, the sequence
0 −→(A0 )H −→(AD )H −→ Hom(A, ẐZ/ZZ) −→ 0
is exact, since H 1 (H, A0 ) ∼
= H 1 (H, Hom(ZZN , k̄ × )) = H 1 (H, k̄ × )N = 0. Be-
i
cause Ĥ (Gk /H, Hom(A, ẐZ/ZZ)) = 0 for i ≥ 0, we obtain
H 1 (Gk /H, (A0 )H ) = H 1 (Gk /H, (AD )H ),
and from this follows (∗), by taking direct limits. Finally, the finiteness of
H 1 (k, A0 ) ∼
= H 1 (k, AD ) follows from (7.2.9). 2
The above duality theorem contains local class field theory as a special case.
Namely, taking i = 0 and A = ZZ, we get an isomorphism
∨ ∼
k̂ × = H 0 (k, D2 ) −→
∼ H 2 (G , Z 1
k Z) = H (Gk , Q/Z Z)∨ = Gab , k
and from this we get the following theorem.
(7.2.11) Theorem. Let k be a local field. Then there is an exact sequence
0ijgh
( ,k)
k× Gab
k ẐZ/ZZ 0,
where ( , k) is the norm residue symbol of local class field theory.
The homomorphism ( , k) is by definition (cf. (3.1.6)) characterized by the
(7.2.12) Proposition. For every χ ∈ H 1 (Gk , Q/ZZ) the norm residue symbol
satisfies the formula
χ((a, k)) = inv(a ∪ δχ),
∼ H 2 (G , Z
where δ : H 1 (Gk , Q/ZZ) → k Z).

We have seen that there are two ways to construct abelian extensions of
exponent m of the ground field k, where m is prime to char(k). First, using
Kummer theory (if µm is contained in k), we obtain an isomorphism
δ
k × /k ×m −→ 1
∼ H (k, µm ).
Secondly, class field theory yields

rec
k × /k ×m −→ ab
∼ Gk /m.
We want to compare these maps. Using (1.4.6) for the exact sequence
m
0 −→ ZZ −→ ZZ −→ ZZ/mZZ −→ 0
m
and the Gk -module k̄ × (observe that the sequence 0 → µm → k̄ × → k̄ × → 0
is also exact), we get the commutative diagram
onmlk k̄ × ) ×
H 0 (G, H 2 (G, ZZ) ∪
H 2 (G, k̄ × )
δ δ −id
∪
H 1 (G, µm ) × H 1 (G, ZZ/mZZ) H 2 (G, k̄ × ).
From (7.2.12) it follows that
χ((a, k)) = inv(a ∪ δχ) = inv(−δ(a) ∪ χ)
for a ∈ H 0 (G, k̄ × ) and χ ∈ H 1 (G, ZZ/mZZ) ⊆ H 1 (G, Q/ZZ). Thus we obtain
(7.2.13) Corollary. Let m ∈ IN be not divisible by the characteristic of k.

Then the diagram
Gabustrqp × Hom(Gk , ZZ/mZZ) ∪

k /m ZZ/mZZ
rec
k × /k ×m inv
−δ
∪
H 1 (k, µm ) × H 1 (k, ZZ/mZZ) H 2 (k, µm )
is commutative. The cup-product on top is given by applying a character

χ ∈ Hom(Gk , ZZ/mZZ) to an element of Gab
k /m.
An important addendum to the local duality theorem arises from the presence
of the maximal unramified extension k̃|k. Let Γ = G(k̃|k) be its Galois group
and T = G(k̄|k̃) the inertia group. Let Õ be the valuation ring of k̃ and observe
that Õ× is a Gk -submodule of D2 . A Gk -module M is called unramified if
MT = M.

Let A be a Gk -module which is finitely generated as a ZZ-module. We

assume that the order of the ZZ-torsion subgroup tor(A) of A is prime to the
characteristic of the residue field of k. If A is unramified, we write
Ad = Hom(A, Õ× ),
which is a submodule of AD = Hom(A, D2 ). Obviously, Ad is also unramified.
If A is finite and unramified, then
AD = A0 = Ad = Hom(A, µ) .
(7.2.14) Definition. For a Gk -module A, we set

i
Hnr (k, A) = im(H i (Γ, AT ) −→ H i (Gk , A))
and these groups are called the unramified cohomology groups . ∗)
If A is unramified, then we have

0
Hnr (k, A) = H 0 (k, A), 0
Hnr (k, Ad ) ⊆ H 0 (k, A0 ) ,
and if in addition A is finite, then
2
Hnr 2
(k, A) = Hnr (k, Ad ) = 0
as cd Γ = 1.
(7.2.15) Theorem. Let A be an unramified Gk -module which is finitely

generated as a ZZ-module and assume that #tor(A) is prime to the characteristic
i
of the residue field of k. Then the groups Hnr (k, Ad ) and Hnr
2−i
(k, A) annihilate
each other in the pairing
∪
H i (k, AD ) × H 2−i (k, A) −→ H 2 (k, k̄ × ) ,→ Q/ZZ .
They are mutually orthogonal complements for i = 1, and for 0 ≤ i ≤ 2 if A
is finite.
Proof: The Γ -module Õ× is cohomologically trivial by (7.1.2)(i) and the

commutative diagram
H i (Γ,wxyzv Ad ) × H 2−i (Γ, A) ∪
H 2 (Γ, Õ× ) = 0
∪
H i (k, AD ) × H 2−i (k, A) H 2 (k̄, D2 ) ,→ Q/ZZ
∗) The letters nr stand for “non ramifié”.

i
shows that the images Hnr (k, Ad ) and Hnr
2−i
(k, A) of the upper groups are
orthogonal. If A is finite, then
0
Hnr (k, Ad ) = H 0 (k, AD ) and Hnr
2
(k, A) = 0
recalling that cd Γ = 1,
2
Hnr (k, Ad ) = 0 and Hnr
0
(k, A) = H 0 (k, A).
We have therefore to investigate only the case i = 1 for a Gk -module A which
is finitely generated as a ZZ-module. Using the exact sequence
0 −→ tor(A) −→ A −→ A/tor −→ 0,
{|}~
H 1 (k, tor(A)) H 1 (T, tor(A))Γ
H 1 (k, A) H 1 (T, A)Γ ,

where the upper map is surjective, since cd Γ = 1, and the map on the right-
hand side is an isomorphism, since A is a trivial T -module and therefore
H 1 (T, A/tor) = Hom(T, A/tor) = 0. Thus in the commutative diagram
0 H 1 (Γ, A) H 1 (k, A) H 1 (T, A)Γ 0
ϕA
H 1 (k, AD )∗ H 1 (Γ, Ad )∗ 0
the upper sequence is exact. The map on the bottom is surjective, since
inf
H 1 (Γ, Ad ) ,→ H 1 (Γ, (AD )T ) −→ H 1 (k, AD )
is injective. Since we already saw that H 1 (Γ, A) and H 1 (Γ, Ad ) annihilate each
other, we obtain the dotted map ϕA , which is necessarily surjective. We have
to show that ϕA is bijective.
In the commutative diagram

H 1 (T, tor(A)) Γ
H 1 (T, A)Γ
ϕtor(A) ϕA
H 1 (Γ, (tor(A))d )∗ H 1 (Γ, Ad )∗ ,

the lower map is injective, which one sees as follows: let Γ 0 be an open
subgroup of Γ acting trivially on A/tor. Because scd Γ = 2, we have a
surjection
cor
H 2 (Γ 0 , Hom(A/tor, Õ× )) −→ H 2 (Γ, Hom(A/tor, Õ× )) −→ 0

and the group on the left is zero, since Õ× is cohomologically trivial. Thus
H 2 (Γ, (A/tor)d ) = 0 and we are reduced to the case of a finite module: if ϕtor(A)
is injective, then ϕA is also injective.
Since the order of tor(A) is prime to the residue characteristic p, we can
replace T by its tame part T0 . Using Kummer theory, we get a natural isomor-
phism
T0 ∼= lim
←−
µn
(n,p)=1
(cf. the remarks at the beginning of §5). As an abelian group, T0 is isomorphic

Q
to `=/p ZZ` and is therefore a duality group of dimension 1 for torsion modules
of order prime to p with dualizing module I ∼
Q
= `=/p Q` /ZZ` . As a Gk -module,
I is naturally isomorphic to the prime-to-p part of the module µ of all roots of
unity. Therefore we obtain
H 1 (T0 , tor(A)) ∼
= H 0 (T0 , Hom(tor(A), µ))∗ = Hom(tor(A), µ)∗ ,
so that, using H 1 (T, tor(A)) = H 1 (T0 , tor(A)), we get
H 1 (T, tor(A))Γ ∼
= (Hom(tor(A), µ)Γ )∗ = H 1 (Γ, (tor(A))d )∗ .
This finishes the proof of the theorem. 2
If A is an unramified Gk -module which is finitely generated and free as a

ZZ-module, then we can consider Ad as a torus over Ok with character group
A. Using the technique of smooth base change, the next corollary can also be
derived (at least for the part prime to the residue characteristic) from a theorem
of S. LANG which asserts that a connected algebraic group over a finite field is
cohomologically trivial.
(7.2.16) Corollary. Let A be an unramified Gk -module which is finitely

generated and free as a ZZ-module. Then
H i (Γ, Ad ) = 0 for all i ≥ 1 .
Proof: This is clear for i ≥ 3 and we saw the assertion for i = 2 in the proof
of (7.2.15). Finally,
H 1 (Γ, Ad ) ∼
= (H 1 (T, A)Γ )∗ = HomΓ (T, A)∗ = 0. 2
For the cohomology theory of global fields, we will need also the following
duality theorem for the extension C|IR.

(7.2.17) Theorem. Let G = G(C|IR). Let A be a finitely generated G-module

and set A0 = Hom(A, C× ). Then H 2 (G, C× ) ∼
= 12 ZZ/ZZ, and the cup-product
∪
Ĥ i (G, A0 ) × Ĥ 2−i (G, A) −→ H 2 (G, C× )
induces for all i ∈ ZZ an isomorphism
Ĥ i (G, A0 ) −→
∼ Ĥ 2−i (G, A)∗ .
Proof: The exact sequence 1 → µ2 → C× → C× → 1 yields H 2 (G, C× ) ∼

2
=
H 3 (G, µ2 ) ∼
= H 1 (G, µ2 ) ∼
= 12 ZZ/ZZ. The theorem is true for the G-modules
(∗) A = ZZ/2ZZ, ZZ, ZZ(1),
where ZZ(1) = ZZ as abelian group, and the generator σ of G acts as multiplica-
tion by −1 (A0 is then C× with the action σz = z̄ −1 ).
In fact, for A = ZZ the cohomology groups Ĥ i (G, A) and Ĥ i (G, A0 ) are
trivial for odd i and the same holds for A = ZZ(1) with even i. The cohomology
groups that occur in all other cases have order 2. If x and y are the nontrivial
elements of Ĥ i (G, A0 ) and Ĥ 2−i (G, A), then a direct elementary computation
(and then using the periodicity for the cohomology of cyclic groups) shows
that x ∪ y is the nontrivial element of H 2 (G, C× ). This proves the theorem for
the modules (∗).
Suppose we are given an exact sequence 0 → B → A → C → 0 of finitely
generated G-modules. Setting Ĥ i (M ) = Ĥ i (G, M ), we obtain a diagram
Ĥ i−1 (B 0 ) Ĥ i (C 0 ) Ĥ i (A0 ) Ĥ i (B 0 ) Ĥ i+1 (C 0 )
α β γ δ ε
Ĥ 2−i+1 (B)∗ Ĥ 2−i (C)∗ Ĥ 2−i (A)∗ Ĥ 2−i (B)∗ Ĥ 2−i−1 (C)∗
with exact rows. The partial diagrams are commutative or anti-commutative.
The five-lemma now implies that the assertion is true for A if it holds for B
and C.
If we want to prove the theorem for an arbitrary finitely generated G-module
A, we may assume that its torsion submodule is 2-primary. Since the theorem
is true for ZZ/2ZZ, which is the only finite simple 2-primary G-module, we
therefore obtain the result for finite G-modules.
Now consider the general case. First of all, we may assume the module A
to be torsion-free. Let 0 =/ a ∈ A be arbitrary. If a + σa =/ 0, it generates a
submodule isomorphic to ZZ and otherwise a = −σa generates a submodule
isomorphic to ZZ(1). Therefore the general case follows by induction on the
ZZ-rank of the module A. 2

§3. The Local Euler-Poincaré Characteristic 391
We finish this section by fixing conventions in the archimedean case. Firstly,

we consider the extension C|IR as ramified, hence the inertia group is the full
group:
T (C|R) = G(C|R).
i
We therefore have Hnr (K, A) = 0 for all i ≥ 0, any archimedean local field K
and any G(K|K)-module A. Furthermore, we put UK = K × and
IR×
(
if K = IR,
UK1 = +
C× if K = C.
Exercise 1. Let k be a local field and let k̃|k be the maximal unramified extension. Let d :
Gk → ẐZ be the surjective homomorphism coming from the canonical isomorphism G(k̃|k) ∼ =
ẐZ.
Show that the Weil group Wk of the class formation (Gk , k̄ × ) (see III §1) is the pre-image
Wk = d−1 (ZZ).
Hint: For every finite Galois extension K|k we have a commutative exact diagram
1 ¡¢ K× W (K|k) G(K|k) 1
ρ g
1 G(K̃|K) G(K̃|k) G(K|k) 1,

where ρ(a) = ϕvKK (a) . Show that the image of g is the pre-image d−1 (ZZ) under d : G(K̃|k) → ẐZ,
and then pass to the limit.
Exercise 2. Let k be a local field, let k̃ be the maximal unramified extension of k and let
T = G(k̄|k̃) be the inertia subgroup of Gk . Let A be a finite Gk -module of order prime to the
residue characteristic of k (not necessarily unramified). Show that
1
Hnr (k, A)⊥ = Hnr
1
(k, A0 )
with respect to the Tate-pairing.
§3. The Local Euler-Poincaré Characteristic
Let k be a local field and let p be its residue characteristic. If A is a

finite Gk -module of order prime to char(k) (in the case char(k) > 0), then
the cohomology groups H i (k, A) are finite groups by (7.1.8)(iii). We set
hi (k, A) = #H i (k, A) and define the Euler-Poincaré characteristic of A by
i h0 (k, A)h2 (k, A)
hi (k, A)(−1) =
Y
χ(k, A) = .
i h1 (k, A)

If 0 → A → B → C → 0 is an exact sequence of finite Gk -modules of order

prime to char(k), then the alternating product of the orders of the groups in the
exact cohomology sequence
. . . −→ H i (k, A) −→ H i (k, B) −→ H i (k, C) −→ H i+1 (k, A) −→ . . .
is 1, and from this it follows that
χ(k, B) = χ(k, A)χ(k, C).
Our aim is to prove the
(7.3.1) Theorem (TATE). For every finite Gk -module A of order a prime to

char(k) we have
χ(k, A) = ||a||k ,
where || ||k is the normalized absolute value of k.∗)
The formula is simple, but the proof is surprisingly difficult. It mirrors an

explicit description of the multiplicative group K × of a finite Galois extension
K|k as a G(K|k)-module.
Let G = G(K|k) and let ` be a prime number =/ char(k). We consider
the Grothendieck group K00 (IF` [G]) of finite IF` [G]-modules (cf. [32], chap.II,
§16b):
Let F (G) be the free abelian group generated by the set of isomorphism
classes of finite IF` [G]-modules. Denoting the isomorphism class of A by
{A}, let R(G) be the subgroup in F (G) generated by all elements
{B} − {A} − {C}
arising from short exact sequences 0 → A → B → C → 0. Then K00 (IF` [G])
is defined as the quotient F (G)/R(G). We denote by [A] the class of {A}
in K00 (IF` [G]). K00 (IF` [G]) becomes a ring by linear extension of the product
[A][B] = [A ⊗IF` B].
If A is a finite IF` [G]-module, then we can view it as a Gk -module which
becomes a trivial Galois module over K, and the cohomology groups H i (K, A)
are also finite IF` [G]-modules. We define
2
(−1)i [H i (K, A)],
X
h(K, A) =
i=0
which is an element in K00 (IF` [G]).

Theorem (7.3.1) will be a consequence of
the following theorem, which is a sharpening of it.
∗) That is, ||x|| = q −v(x) , where q is the cardinality of the residue field κ and v is the additive
k
valuation of k with value group ZZ.

(7.3.2) Theorem (SERRE). Let A be a finite IF` [G]-module. If ` is not equal to

the residue characteristic p of k, then h(K, A) = 0. If ` = p and char(k) = 0,
then h i
h(K, A) = − dimIFp (A)[k : Qp ] IFp [G] .
For the proof of Serre’s theorem we need the following
(7.3.3) Lemma.
(i) Let A be a finite G-module. Then
[` A] = [A` ] . ∗)
(ii) Let V be a G-module such that V` and ` V are finite. If W ⊆ V is a
submodule of finite index, then
[V` ] − [` V ] = [W` ] − [` W ] .
Proof: We prove (ii) first. Using a Jordan-Hölder series, we may assume that
V /W is a finite simple G-module. In particular, ` (V /W ) ∼
= (V /W )` . Consider
the diagram
0ª«¬©¨£¤¥¦§ W V V /W 0
` ` `
0 W V V /W 0.
The snake lemma gives the exact sequence
0 −→ ` W −→ ` V −→ ` (V /W ) −→ W` −→ V` −→(V /W )` −→ 0,
and hence the result. Assertion (i) follows by applying (ii) to V = A, W = 0.
2
Proof of (7.3.2): First suppose that A = µ` . Using the Kummer sequence,

we get
H 0 (K, µ` ) = µ` (K),
H 1 (K, µ` ) = K × /K ×` ,
H 2 (K, µ` ) = H 0 (K, ZZ/`ZZ)∗ = ZZ/`ZZ.
Therefore
h(K, µ` ) = [µ` (K)] − [K × /K ×` ] + [ZZ/`ZZ].
∗) For a ZZ -module V , the modules V and V are defined by the exact sequence
` ` `
`
0 → ` V → V → V → V` → 0.

Let U denote the group of units of the valuation ring O of K. The exact
sequence
0 −→ U/U ` −→ K × /K ×` −→ ZZ/`ZZ −→ 0
gives us [K × /K ×` ] = [ZZ/`ZZ] + [U/U ` ], hence
(1) h(K, µ` ) = [µ` (K)] − [U/U ` ] = [` U ] − [U` ].
Let p be the maximal ideal of O and V = 1 + p the group of principal units. V
is a finitely generated ZZp [G]-module, and we have the exact sequence
1 −→ V −→ U −→ κ× −→ 1,
with κ× the multiplicative group of the residue field of K. Since κ× is finite
we obtain from (1) and (7.3.3)
(2) h(K, µ` ) = [` U ] − [U` ] = [` V ] − [V` ].
Now if ` =/ p, then ` V = V` = {1} since V is a pro-p-group, hence h(K, µ` ) = 0.
So let ` = p and char(k) = 0. Let W = 1 + pn be the group of n-th principal
units. It is a ZZp [G]-submodule of finite index in V , hence by lemma (7.3.3)
again, we obtain from (2)
h(K, µ` ) = [p W ] − [Wp ].
If n is sufficiently large, then the logarithm log : W → pn is an isomorphism
(see [160], chap.II, (5.5)) and, since pn is a subgroup of finite index in O, we
get
(3) h(K, µ` ) = [p O] − [O/pO].
M
The extension K|k has a normal basis, K = kσθ, and we may choose θ in O.
σ ∈G M
If R denotes the valuation ring of k, then M = Rσθ is a ZZp [G]-submodule
σ ∈G
of O of finite index and p M = 0 and M/pM ∼
= R/pR[G] ∼
= IFp [G][k:Qp ] . We
now obtain from (3) via (7.3.3)
h(K, µ` ) = −[M/pM ] = −[k : Qp ][IFp [G]].
Now let A be an arbitrary finite IF` [G]-module. The cup-product on the
cochain groups C i (K, ZZ/`ZZ) ⊗ A → C i (K, A) is obviously an isomorphism,
hence we have an isomorphism of G-modules H i (K, ZZ/`ZZ) ⊗ A ∼ = H i (K, A),
and so
h(K, A) = h(K, ZZ/`ZZ) · [A].
The functor M 7→ M ∗ = Hom(M, IF` ) is exact on finite IF` [G]-modules and
thus defines an endomorphism ξ 7→ ξ ∗ of the group K00 (IF` [G]). By the duality
theorem (7.2.6), we have
h(K, ZZ/`ZZ)∗ = h(K, µ` ),
and therefore, since [IF` [G]] = [IF` [G]∗ ],

(
−[k : Qp ][IFp [G]] · [A], if p = ` ,
h(K, A) =
0, if p =/ ` .
Furthermore, if A0 denotes the trivial G-module with underlying group A, then
IF` [G] ⊗ A0 → IF` [G] ⊗ A, σ ⊗ a 7→ σ ⊗ σa, is an isomorphism of IF` [G]-
modules (this is just the dual statement to IndG A ∼
{1}
= IndG A, see I §6, p.62),
so that
[IF` [G]] · [A] = [IF` [G]] · [A0 ] = dimIF` (A)[IF` [G]].
This finishes the proof of theorem (7.3.2). 2
For the deduction of Tate’s formula

(∗) χ(A) = ||a||k ,
we need from the representation theory of finite groups another
(7.3.4) Lemma. The group K00 (IF` [G]) ⊗ Q is generated by the images of
K00 (IF` [H]) ⊗ Q under IndH
G , where H runs through all cyclic subgroups of G
of order prime to `.
Proof: A theorem of E. Artin (see [214], 12.5 th. 26) asserts that the map
K00 (Q` [H]) ⊗ Q −→ K00 (Q` [G]) ⊗ Q
M
Ind ⊗ Q :
H ∈T
is surjective where T is the set of all cyclic subgroups of G. (This holds with
any field of characteristic zero replacing Q` .) By [214], 16.1, th. 33, there
exists a surjective homomorphism K00 (Q` [G]) ⊗ Q K00 (IF` [G]) ⊗ Q which is
natural with respect to the group G. Therefore the above map Ind ⊗ Q remains
0
surjective if we replace Q` by IF` . Let H = H (` ) × H` be a cyclic subgroup of
G where H` is the `-Sylow subgroup of H. Let M be a simple IF` [H]-module.
By (1.6.13), M H` =/ 0. Since M H` is also an H-module, we obtain M = M H` ,
(`0 )
hence an isomorphism of IF` [H]-modules IndH H ∼
H ResH (`0 ) M = M ⊗ ZZ IF` [H` ].
(`0 )
Since ZZ/`ZZ is the only simple IF` [H` ]-module, the class of IndH H
H ResH (`0 ) M
0 n
in K0 (IF` [H]) is equal to n[M ], where #H` = ` . Therefore the images of
0
K00 (IF` [H]) ⊗ Q and K00 (IF` [H (` ) ]) ⊗ Q under Ind ⊗ Q in K00 (IF` [G]) ⊗ Q are
the same. 2
Proof of theorem (7.3.1): Let A be a finite Gk -module of order a. We may

assume À = 0 for some prime number ` =/ char(k). The general case follows
from this via the exact sequence 0 → ` A → A → A/` A → 0 by induction on
the order of A, since χ and || ||k are multiplicative with respect to short exact

sequences. Every finite Gk -module becomes a trivial Galois module over some
finite Galois extension K|k. It thus suffices to prove the formula (∗) for finite
IF` [G]-modules with G = G(K|k). The functions χ(k, A) and ϕ(A) = ||a||k are
additive, i.e. they define homomorphisms
χ, ϕ : K00 (IF` [G]) −→ Q×
+,
and we have to show χ = ϕ. Using lemma (7.3.4) and observing that Q× + is

torsion-free, it suffices to check this equality on elements of the form B =
IndHG (A), where A is a finite IF` [H]-module and H is a cyclic subgroup of G
of order prime to `. Let k 0 be the fixed field of H. Shapiro’s lemma yields
χ(k, B) = χ(k 0 , A)
and for b = #B we have
0
||a||k0 = ||a||[k
k
:k]
= ||b||k .
This reduces our problem to the case where G is a cyclic group of order prime
to `. Then the Hochschild-Serre spectral sequence for the group extension
1 → GK → Gk → G → 1
and the G-module A degenerates. Thus H i (k, A) = H 0 (G, H i (K, A)). There-
fore if
d : K00 (IF` [G]) → ZZ
is the homomorphism given by
d([M ]) = dimIF` (H 0 (G, M )),
then
χ(k, A) = `d(h(K,A)) .
If ` =/ p, then by (7.3.2) h(K, A) = 0, so that χ(k, A) = 1, and if ` = p, then
h(K, A) = − dim(A)[k : Qp ][IFp [G]], and since d([IFp [G]]) = dim(IFp [G]G ) =
dim(IFp ) = 1, we obtain
χ(k, A) = p−[k:Qp ] dim A = ||a||k . 2
For the fields IR and C we have a similar statement.
(7.3.5) Theorem. If k = IR or C, then for any finite Gk -module A of order a

we have
h0 (k, A)h0 (k, A0 )
= ||a||k ,
h1 (k, A)
where ||x||IR = |x| and ||x||C = |x|2 .

Proof: If k = C, then H 0 (k, A) = A and H 0 (k, A0 ) = A0 both have order a,

H 1 (k, A) = 0 and ||a||C = a2 . Let k = IR and G = G(C|IR), and let σ be the
generator of G. For x ∈ A and f ∈ A0 , we have
((1 − σ)f )(x) = f (x)/σ(f (σx)) = f (x)(f (σx)) = f ((1 + σ)x),
noting that σζ = ζ −1 for a root of unity in C. Therefore 1 − σ : A0 → A0 is

adjoint to 1 + σ : A → A, and so, in the pairing A0 × A → C× , (A0 )G and NG A
are exact annihilators. Therefore
||a||IR = #A = #(A0 )G #NG A = #H 0 (G, A0 )#(H 0 (G, A)/Ĥ 0 (G, A)).
Since A is finite, we have by [160], chap.IV, (7.3) and (1.7.1)

#Ĥ 0 (G, A) = #Ĥ −1 (G, A) = #Ĥ 1 (G, A) ,
and the theorem follows. 2
We saw in VII §2 that the Galois module µ plays an important role in the
case of local fields. Now we introduce the following general terminology.
Assume that k is any field. The subgroup µ of roots of unity contained in
the separable closure k̄ of k is a Gk -module in a natural way. There exists a
ZZ×
Y
∼
h : Aut(µ) −→ ` ,
`=
/ char(k)
given by ϕ(ζ) = ζ h(ϕ) . The right side of the equality is defined as follows: if
ζ n = 1, n ∈ IN, then ζ α := ζ a for any a ∈ ZZ with a ≡ α mod n. The action of
Gk on µ is given by a character
ZZ×
Y
χcycl : Gk −→ ` .
`=
/ char(k)
(7.3.6) Definition. The character χcycl is called the cyclotomic character.

Let A be a finite Gk -module whose order is prime to char(k). For i ∈ ZZ we
denote by A(i) the Gk -module which is equal to A as an abelian group and
which is endowed with the (twisted) action
σ(a) := χcycl (σ)i · σa,
where the action on the right-hand side is the original action of Gk on A.
We call A(i) the i-th Tate twist of A. We apply the same definition to a
discrete resp. compact G-module which is a direct resp. projective limit of
finite modules of order prime to char(k).

In particular, A = A(0) and for i, j ∈ ZZ we have the rule

A(i + j) = A(i)(j) = A(j)(i).
Note that the above definition can only be applied to modules A such that the
multiplication by χcycl (σ) is a well-defined automorphism. In particular, the
module ZZ(i) does not exist (unless we consider archimedean local fields).
For (n, char(k)) = 1 and i ≥ 1 we have
ZZ/nZZ (i) ∼
= µ⊗i
n and ZZ/nZZ (−i) ∼
= Hom(µ⊗i
n ,Z
Z/nZZ).
For a prime number ` and i ∈ ZZ we have
ZZ` (i) = lim ZZ/`m ZZ(i) and Q` /ZZ` (i) = lim ZZ/`m ZZ(i) .
←− −→
m m
If A is finite with nA = 0, then A(i) = A ⊗ ZZ ZZ/nZZ (i) for all i ∈ ZZ.
Returning to the case of a local field, we have the
(7.3.7) Theorem. Let k be a local field with residue characteristic p and let
` =/ char(k) be a prime number. Then for all j ∈ ZZ
2
(
−[k : Qp ] if ` = p ,
(−1)i dimIF` H i (Gk , ZZ/`ZZ (j)) =
X
i=0
0 otherwise.
Proof: This follows directly from (7.3.1), since

(
p−[k:Qp ] if ` = p ,
||`||k =
1 otherwise.
2
(7.3.8) Corollary. With the notation as in (7.3.7), the following equalities hold
for all j ∈ ZZ:
2
(
X
i −[k : Qp ] if ` = p ,
(−1) rank ZZ` Hi (Gk , ZZ` (j)) =
i=0
0 otherwise,
and
2
(
X
i i −[k : Qp ] if ` = p ,
(−1) rank ZZ` Hcts (Gk , ZZ` (j)) =
i=0
0 otherwise.
Proof: The second equality follows from the first and (2.7.12). In order to
prove the first statement, we observe that
rank ZZ` Hi (Gk , ZZ` (j)) = dimIF` Hi (Gk , ZZ` (j))/` − dimIF` ` Hi (Gk , ZZ` (j))

and
dimIF` Hi (Gk , ZZ/`ZZ(j)) = dimIF` H i (Gk , ZZ/`ZZ(−j)) .
by (2.6.9). The exact sequence
`
0 −→ ZZ` (j) −→ ZZ` (j) −→ ZZ/`ZZ(j) −→ 0
shows that
dimIF` H0 (Gk , ZZ/`ZZ(j)) = dimIF` H0 (Gk , ZZ` (j))/` ,
dimIF` Hi (Gk , ZZ/`ZZ(j)) = dimIF` Hi (Gk , ZZ` (j))/`
+ dimIF` ` Hi−1 (Gk , ZZ` (j))

for i ≥ 1. Since cd` Gk = 2, we obtain
2 ∞
i
(−1)i rank ZZ` Hi (Gk , ZZ` (j))
X X
(−1) rank ZZ` Hi (Gk , ZZ` (j)) =
i=0 i=0
∞
(−1)i dimIF` H i (Gk , ZZ/`ZZ(−j))
X
=
i=0
2
(−1)i dimIF` H i (Gk , ZZ/`ZZ(−j)).
X
=
i=0
Now the corollary follows from (7.3.7). 2
(7.3.9) Corollary. With the notation as in (7.3.7), we have


 1 + δ + [k : Qp ] if ` = p ,
dimIF` H 1 (Gk , ZZ/`ZZ) =
 1+δ otherwise,
where δ = 1 or 0 according to whether the `-th roots of unity are contained in
k or not.
Proof: By duality, we have dimIF` H 2 (Gk , ZZ/`ZZ) = dimIF` H 0 (Gk , µ` ) = δ.

Since dimIF` H 0 (Gk , ZZ/`ZZ) = 1, the assertion follows from (7.3.7) with j = 0.
2
For a prime number ` =/ char(k) and i ∈ ZZ, i =/ 0, we introduce the numbers

n o
wì := max `n exp G(k(µ`n )|k) divides i .
Here exp(G) denotes the exponent of a finite group G. In particular, we have
w`1 = #µ`∞ (k) and wì = w`−i .
Remark: If k(µ`∞ )|k is pro-cyclic (this can only fail if k is a dyadic number
field, i.e. a finite extension of Q2 , and ` = 2), then obviously
n o
wì = max `n [k(µ`n ) : k] | i .

(7.3.10) Proposition. Assume that k is a finite extension of degree d of Qp

and let ` be a prime number. Then

 Q` /ZZ` for i = 0,
(i) H 0 (Gk , Q` /ZZ` (i)) ∼
=  ZZ/w i ZZ for i =/ 0.
`



 ZZ/w`1 ZZ ⊕ (Q` /ZZ` )d+1 for i = 0, ` = p,

(Q` /ZZ` )d+1 for i = 1, ` = p,





 ZZ/w 1−i ZZ ⊕ (Q /ZZ )d

for i =/ 0, 1, ` = p,

(ii) H 1 (Gk , Q` /ZZ` (i)) ∼ ` ` `
= 



ZZ/w`1 ZZ ⊕ (Q` /ZZ` ) for i = 0, ` =/ p,




 Q` /ZZ` for i = 1, ` =/ p,

 ZZ/w 1−i ZZ

for i =/ 0, 1, ` =/ p.
`

 H 2 (Gk , k̄ × )(`) ∼
= Q` /ZZ` for i = 1,
(iii) H 2 (Gk , Q` /ZZ` (i)) ∼
=  0 for i =/ 1.
Remark: From the proposition above, the continuous cochain cohomology

groups with values in ZZp (i) can be easily calculated by the rule
j
Hcts = H 2−j (Gk , Q` /ZZ` (1 − i))∨ ∼
(Gk , ZZ` (i)) ∼ = H2−j (Gk , ZZ` (i − 1)) .
Proof of (7.3.10): If i =/ 0, then

i
H 0 (Gk , Q` /ZZ` (i)) = {ζ ∈ µ`∞ | ζ χcycl (σ) = ζ for all σ ∈ Gk }
= {ζ ∈ µ`∞ | σ i ζ = ζ for all σ ∈ G(k(µ`∞ )|k)},
and therefore n o
#H 0 (Gk , Q` /ZZ` (i)) = max `n σ i = 1 for all σ ∈ G(k(µ`n )|k) = wì .
This proves (i). The assertion for H 2 follows from the local duality theorem
(7.2.6): (
2 ∼ 0 ∨ ∼ Q` /ZZ` if i = 1,
H (Gk , Q` /ZZ` (i)) = Hcts (Gk , ZZ` (1 − i)) =
0 otherwise.
0 2
Furthermore, from (7.3.8) and the statements for H and H , we obtain


 d + 1 for i = 0, 1 ` = p,
 d

for i =/ 0, 1, ` = p,
corank ZZ` H 1 (Gk , Q` /ZZ` (i)) =


 1 for i = 0, 1 ` =/ p,
0 for i =/ 0, 1, ` =/ p.

It remains to calculate the cotorsion of the group H 1 (Gk , Q` /ZZ` (i)). Let
m ≥ 1. From the exact sequence 0 → ZZ/`m ZZ(i) → Q` /ZZ` (i) → Q` /ZZ` (i) → 0
follows the exact sequence
H 1 (Gk , Q` /ZZ` (i))/`m ,→ H 2 (Gk , ZZ/`m ZZ (i)) `m H 2 (Gk , Q` /ZZ` (i)) .

§4. Galois Module Structure of the Multiplicative Group 401
Applying the projective limit, this yields for i =/ 1

lim H 1 (Gk , Q` /ZZ` (i))/`m ∼= lim H 2 (Gk , ZZ/`m ZZ (i))
←− ←−
m m
∼
= (lim H 0 (Gk , ZZ/`m ZZ (1 − i)))∗
−→
m
∼
= ZZ/w`1−i ZZ ,
since H 2 (Gk , Q` /ZZ` (i)) = 0. If i = 1, then the group H 1 (Gk , Q` /ZZ` (1)) ∼
=
×
k ⊗ Q` /ZZ` is `-divisible. Therefore
(

1 ∗

∼ ZZ/w`1−i ZZ for i =/ 1,
tor H (Gk , Q` /ZZ` (i)) =
0 for i = 1.
This finishes the proof of (7.3.10). 2
§4. Galois Module Structure of the

Multiplicative Group
In this section we combine some of the results of chapter V §6 with those of

the last sections in order to determine the structure of the p-adic completion
of the multiplicative group of a local field and of relation modules of certain
extensions of local fields.
(7.4.1) Theorem. Let k be a p-adic local field and let n = [k : Qp ]. Then the
absolute Galois group G = G(k̄|k) of k is generated by n + 2 elements. If
1 −→ N −→ Fn+2 −→ G −→ 1
is a presentation of G = G(k̄|k) by a free profinite group Fn+2 of rank n + 2,
then
N ab (p) ∼
= ZZp [[G ]]
as ZZp [[G ]]-modules.
If K|k is a Galois extension of p-adic local fields with Galois group G =

G(K|k), then let H = G(k̄|K). We want to determine the structure of the
ZZp [[G]]-module
X = H ab (p) ∼
m
= lim A(L) = lim L× /L×p ,
←− ←−
L L,m

where L runs through all finite subextensions of K|k and the projective limit
m
is taken with respect to the norm maps. A(L) = lim L× /L×p is the p-
←− m
completion of the multiplicative group of the local field L, which is (via the
reciprocity map) isomorphic to G(k̄|L)ab (p). We denote the group of roots of
unity of p-power order in L by µp∞ (L).
(7.4.2) Theorem. Let k be a p-adic local field and let n = [k : Qp ].

(i) Let K|k be a Galois extension with G = G(K|k) and let
1 −→ Rn+2 −→ Fn+2 −→ G −→ 1
be a presentation of G. If X denotes the ZZp [[G]]-module G(k̄|K)ab (p),
then there exists an exact sequence
ab
0 −→ ZZp [[G]] −→ Rn+2 (p) −→ X −→ 0 .
/ K, then G is generated by n + 1 elements and there is an isomor-
If µp ⊆
phism
ab
Rn+1 (p) ∼
=X.
(ii) Let σ1 , . . . , σn+2 be topological generators of G = G(k̄|k). Let ai ∈ ZZp

with σi (ζ) = ζ ai for all ζ ∈ µp∞ (k̄) and let σ̄i be the image of σi in
G = G(K|k), i = 1, . . . , n + 2. Then there exists an exact sequence
0 −→ ZZp [[G]] −→ ZZp [[G]]n+2 −→ Y −→ 0,
1 7−→ (σ̄i − ai )i ,
where Y = IG /IH IG , as in V §6. If µp∞ (K) = 1, then Y is a free
ZZp [[G]]-module of rank n + 1.
Before we prove the two theorems, which are taken from [97], we provide
two lemmas.
(7.4.3) Lemma. Let k be a local field of characteristic p and let K|k be a

finite tamely ramified Galois extension. Then UK1 /(UK1 )p is a cohomologically
trivial G(K|k)-module, and the norm map induces an isomorphism

UK1 /(UK1 )p ∼ U 1 /(U 1 )p .
−→ k k
G(K|k)
Proof: Since the group of principal units UK1 is ZZp -torsion-free, we have the
exact sequence
p
0 −→ UK1 −→ UK1 −→ UK1 /(UK1 )p −→ 0.

By (7.1.2)(ii), UK1 is a cohomologically trivial G(K|k)-module. It follows that

UK1 /(UK1 )p is also cohomologically trivial, and the commutative diagram
NG(K|k) G(K|k)
UK1 /(UK1®°¯ )p

UK1 /(UK1 )p
G(K|k)
NK|k
Uk1 /(Uk1 )p
gives the desired result. 2
(7.4.4) Lemma. Let k be a local field of residue characteristic p and let K|k
be a finite Galois extension with Galois group G = G(K|k).
(i) If k is a p-adic field of degree n over Qp , then there are isomorphisms of
Qp [G]-modules
A(K) ⊗ Q ∼ = Qp [G]n ⊕ Qp ,
UK1 ⊗ Q ∼ = Qp [G]n .
(ii) If k is the field of Laurent series over a finite field and K|k is tamely
ramified, then there are isomorphisms of ZZp [G]-modules
A(K) ∼
= ZZp [G]IN ⊕ ZZp ,
UK1 ∼
= ZZp [G]IN .
Proof: (i) Let UKm be the group of principal units of level m of K. For m
large enough, the p-adic logarithm induces a G-invariant isomorphism
log : UKm −→
∼ pm ⊆ OK ,
see [160], chap.II, (5.5). Tensoring by Q and noting that UKm has finite index
in UK1 , the existence of a normal basis for K|k gives us a G-isomorphism
U1 ⊗ Q ∼ = Qp [G]n .
K
Applying p-completion to the exact sequence

0 −→ UK −→ K × −→ ZZ −→ 0,
0 −→ UK1 −→ A(K) −→ ZZp −→ 0.
Using the semi-simplicity of the category of finitely generated Qp [G]-modules
(2.6.12), we obtain
A(K) ⊗ Q ∼ = U 1 ⊗ Q ⊕ Qp , K
hence the result.

(ii) Let k = IFq ((t)), where q = pf0 , and let K|k be a finite tamely ramified
Galois extension. By the properties of tamely ramified extensions, one knows
that there
√ exists an unramified extension K 0 of K which is of the form K 0 =
k(ζ, t), where ζ is a primitive e-th root of unity and e = q f − 1, f ≥ 1. Using
e
(7.4.3), we may assume that K = K 0 . Then the Galois group G = G(K|k) has
the following structure:
G = hσ, τ | στ σ −1 = τ q , σ f = 1, τ e = 1i,
where √ √
σ(ζ) = ζ q and σ(√ e
t) = e√t ,
τ (ζ) = ζ and τ ( e t) = ζ e t .
If H = hτ i, then G/H = G(k(ζ)|k) ∼ = G(λ|κ), where λ ∼ = IFqf is the residue
∼
field of K and κ = IFq is the residue field of k. The existence of a normal basis
σ̄ i (α), i = 0, . . . , f − 1,
of λ gives us an isomorphism λ ∼
= κ[G/H] of IFq [G/H]-modules.
For 0 r e − 1, let λ(r) = εr λ ⊆ λ[H] be the 1-dimensional eigenspace
≤ ≤
corresponding to the idempotent

1 e−1
ζ −rj τ j ,
X
εr =
e j=0
i.e. τ (β) = ζ r β for β ∈ λ(r). Obviously, λ(r) is an IFq [H]-module. However,
it is even an IFq [G]-module, as σεr = εr σ in IFq [G], i.e. σ(β) = β q for β ∈ λ(r).
Therefore the map
e−1
M
λ(r) −→ κ[G],
r=0
defined by σ i (α) 7→ σ i εr for σ i (α) ∈ λ(r), i = 0, . . . , f − 1, r = 0, . . . , e − 1,

is an isomorphism of IFq [G]-modules. In particular, the modules λ(r) are
IFq [G]-projective.
Let i1 < i2 < · · · < ie(p−1) be natural numbers such that 1 ≤ iν < ep and
p - iν , and let ie(p−1)+1 = ep. For each 0 ≤ r ≤ e − 1, there are exactly p − 1
numbers iν such that iν ≡ r mod e. Therefore we obtain an isomorphism of
IFq [G]-modules
e(p−1)
∼ κ[G]p−1 .
M
λ(iν ) −→
ν=1
For each j ≥ 0, we have an IFq [G]-isomorphism
√
UKepj+iν (UK1 )p /UKepj+iν+1 (UK1 )p [1 + α ( t)epj+iν ] 7→ α .
∼ λ(i ), e
−→ ν
Since the modules λ(iν ) are IFq [G]-projective, it follows that

UKepj (UK1 )p /UKep(j+1) (UK1 )p ∼
= κ[G](p−1) ,
and so we get IFp [G]-isomorphisms

(∗) UK1 /UKepj (UK1 )p ∼

= IFp [G]f0 (p−1)j
for all j ≥ 1. Passing to the projective limit, we get UK1 /(UK1 )p ∼
= IFp [G]IN .
IN
Since ZZp [G] is the free compact ZZp [G]-module (on countably many ge-
nerators), the diagram
±²³´ IN
ZZp [G]
φ
UK1
IFp [G]IN UK1 /(UK1 )p

can be commutatively completed by a continuous homomorphism φ, which
is surjective by the topological Nakayama lemma (5.2.18)(i) (for ZZp ). Using
the fact that UK1 is ZZp -torsion-free, a second application of the topological
Nakayama lemma implies that ker(φ) = 0, hence φ is an isomorphism.
In order to show the second isomorphism, we consider the exact sequence
0 → UK1 → A(K) → ZZp −→ 0. Using (5.2.14) and (2.7.6), we have
Ext1ZZp [G] (ZZp , UK1 ) ∼ 1
= Hcts (G, ZZp [G]IN ) = 0,
and so an isomorphism A(K) ∼
= UK1 ⊕ ZZp , which is the desired result. 2
Proof of (7.4.1): First let K|k be a finite tamely ramified Galois extension.
Then G = G(K|k) is generated by two elements, σ and τ say, acting on µp∞ (K)
by ζ σ = ζ a and ζ τ = ζ b , a, b ∈ ZZp . We obtain an exact sequence
ϕ
ZZp [G]2 −→ ZZp [G] −→ µp∞ (K)∨ −→ 0,
where ker(ϕ) = (σ − a, τ − b), which induces the exact sequence
(∗) 0 −→ ZZp [G] −→ ZZp [G]2 −→ M0 −→ 0,
with M0 ' D(µp∞ (K)∨ ) since (µp∞ (K)∨ )+ = 0. Setting Y = IG /IH IG ,
where H = G(k̄|K), we get from (5.6.9) and (7.2.4) the homotopy equivalence
M0 ' Y,
and for X = G(k̄|K)ab (p) we have the exact sequence
0 −→ X −→ Y −→ IG −→ 0
by (5.6.5). By Maschke’s theorem, finitely generated Qp [G]-modules are
projective, so that
Y ⊗Q∼ = X ⊗ Q ⊕ IG ⊗ Q ,
and given (7.4.4)(i) and the exact sequence (∗), it follows that
Y ⊗Q∼
= Qp [G]n ⊕ Qp ⊕ IG ⊗ Q ∼
= Qp [G]n+1 ∼
= M0 ⊗ Q ⊕ Qp [G]n .
From (5.6.11), we get the isomorphism
(∗∗) Y ∼
= M0 ⊕ ZZp [G]n .

If
1 −→ Rn+2 −→ Fn+2 −→ G −→ 1
is a presentation of the finite group G by a free profinite group Fn+2 , then
ab
we obtain from (5.6.6) an exact sequence 0 → Rn+2 (p) → ZZp [G]n+2 → IG → 0.
Since ZZp [G]n+2 is projective, we get, using the isomorphism (∗∗) and the exact
sequence (∗), an commutative exact diagram
0µ¶·¸¹º»¼½¾¿ ab
Rn+2 (p) ZZp [G]n+2 IG 0
α
0 X Y IG 0,
where the kernel of α is isomorphic to ZZp [G]. Thus we obtain a G-invariant
ab
surjection β : Rn+2 (p) X, whose kernel is isomorphic to ZZp [G], and an
isomorphism
β∗
(ZZ/#G ZZ)(p) ∼ ab
= H 2 (G, Rn+2 (p)) −→ 2
∼ H (G, X) .
Now the exact diagram

1ÉÊÀÁÂÃÄÅÆÇÈ ab
Rn+2 (p) Fn+2 /[Rn+2 , Rn+2 ]R(p) G 1
β
1 X G /[H , H ]H (p) G 1,
where R(p) = ker(Rn+2 Rn+2 (p)) and H (p) = ker(H H (p)), can
be completed to a commutative diagram, since the corresponding 2-cocycles
of the group extensions are mapped by β∗ onto each other (after possibly
multiplying with a unit in ZZp ); use scd Fn+2 = scd G = 2 and (3.6.4)(iii) and
I §5 ex.4.
Let ktr be the maximal tamely ramified extension of k. Passing to the
projective limit over all finite Galois extensions K|k inside ktr , we obtain by
the usual compactness argument a surjection
Fn+2 G /[G(k̄|ktr ), G(k̄|ktr )] .
Thus the profinite group G /[G(k̄|ktr ), G(k̄|ktr )] is generated by n+2 elements.
Since G(k̄|ktr ) is a pro-p-group, the Frattini argument (3.9.1) implies that G
itself is generated by n + 2 elements.
In order to prove the second assertion, we first observe that N ab (p) is ZZp [[G ]]-
ab
projective by (5.6.7), so that NH (p) is ZZp [G]-projective for all open normal
subgroups H of G , where G = G /H . From the exact sequence
ab ab
(∗) 0 −→ NH (p) −→ Rn+2 (p) −→ X −→ 0
for X = H ab (p) (recalling that scd G = 2), we obtain
Rab (p) ⊗ Q ∼ = X ⊗ Q ⊕ N ab (p) ⊗ Q ,
n+2 H

and hence, using again (7.4.4)(i),
Qp ⊕ Qp [G]n+1 ∼
= Qp ⊕ Qp [G]n ⊕ NH
ab
(p) ⊗ Q
so that
ab
NH (p) ⊗ Q ∼
= Qp [G] .
It follows that NHab
(p) ∼
= ZZp [G] using (5.6.15). We complete the proof of
(7.4.1) by passing to the limit over all finite quotients G of G , 2
Proof of (7.4.2): Let G = G /H for some closed normal subgroup H of

G (not necessarily of finite index). Again we obtain the exact sequence (∗)
ab
used in the proof of (7.4.1) and, since NH (p) ∼
= ZZp [[G]], we obtain the first
statement of (i).
/ K, then the ZZp [[G]]-module Y is free of rank n + 1, which can
If µp ⊆
be shown in the same manner as in the proof of (7.4.1). Thus we get an
isomorphism Rn+1 ab ∼ X. As above, this implies that G can be generated
(p) −→
by n + 1 elements.
Now we prove assertion (ii). Since ai ∈ ZZ×
p , we get the exact sequence
ZZp [[G ]]n+2 −→ ZZp [[G ]] −→ ZZp −→ 0 ,

−1
ei 7−→ ai (σ̄i − 1)
where {ei | i = 1, . . . , n + 2} is a basis of ZZp [[G ]]n+2 , see (5.6.6). Tensoring by
ZZp (−1), using the isomorphism
∼
ZZp [[G ]] ⊗ ZZp ZZp (−1) −→ ZZp [[G ]]
g⊗1 7−→ χcycl (g) · g ,
and taking G(k̄|K)-coinvariants, we obtain the exact sequence
ZZp [[G]]n+2 −→ ZZp [[G]] −→ µp∞ (K)∨ −→ 0 .
ei 7−→ (σ̄i−1 − ai )
Since (µp∞ (K)∨ )U is finite for every open normal subgroup U of G, we have
(µp∞ (K)∨ )+ = 0. Thus the last sequence implies an exact sequence
0 −→ ZZp [[G]] −→ ZZp [[G]]n+2 −→ M0 −→ 0
1 7−→ (σ̄i − ai )i ,
with M0 ' D(µp∞ (K)∨ ). As in the proof of (7.4.1), we obtain isomorphisms
YU ∼
= (M0 )U ⊕ ZZp [G/U ]n
for any open normal subgroup U of G. By (5.6.11), this implies the existence
of an isomorphism Y ∼= M0 ⊕ ZZp [[G]]n , showing the assertion. 2

For a local field k the absolute Galois group, the inertia group and the
ramification group is denoted by Gk , Tk and Vk , respectively. Furthermore,
Gk = Tk /Vk is the Galois group of the maximal tamely ramified extension ktr
of k. The next theorem is taken from [94].
(7.4.5) Theorem. Let k be a local field with residue characteristic p.

(i) If k is a p-adic field with n = [k : Qp ], then there is an exact sequence of
ZZp [[Gk ]]-modules
0 −→ ZZp [[Gk ]] −→ ZZp [[Gk ]]n+1 −→ Vkab −→ 0 .
(ii) If k is a local field of characteristic p, then there is an isomorphism of
ZZp [[Gk ]]-modules
Vkab ∼
= ZZp [[Gk ]]IN ,
i.e. Vkab is a free compact ZZp [[Gk ]]-module of countably infinite rank.
Proof: (i) By (7.4.2), we have an exact sequence of ZZp [[Gk ]]-modules

0 −→ ZZp [[Gk ]] −→ ZZp [[Gk ]]n+2 −→ Y −→ 0 ,
where Y = IGk /IGktr IGk , and with X = G(k̄|ktr )ab = Vkab we have the exact
sequence
0 −→ X −→ Y −→ IGk −→ 0 .
By (5.6.12), the augmentation ideal IGk is a free ZZp [[Gk ]]-module of rank 1,
hence
Y ∼
= X ⊕ ZZp [[Gk ]] .
This gives the desired result.
(ii) Let K|k be a finite tamely ramified Galois extension. Then, by (7.4.3),
the norm map
NG(K|k) : UK1 /(UK1 )p Uk1 /(Uk1 )p
is surjective, and furthermore we have NG(K|k) UKeK pj = Ukek pj ; see [212] chap.V.
Therefore
NG(K|k) : UK1 /UKeK pj (UK1 )p Uk1 /Ukek pj (Uk1 )p
is well-defined and surjective. In the proof of (7.4.4)(ii) we have shown the
assertion (∗): For j ≥ 1 there is an isomorphism
UK1 /UKeK pj (UK1 )p −→
∼ IF [G(K|k)]f0 (p−1)j
p
of IFp [G(K|k)]-modules. Passing to the projective limit over all finite tamely
ramified Galois extension K|k, we obtain

§5. Explicit Determination of Local Galois Groups 409
Vkab /p ∼
= lim UK1 /(UK1 )p
←−
K|k
∼
= lim lim UK1 /UKeK pj (UK1 )p
←− ←−
K|k j
∼
= lim lim IFp [G(K|k)]f0 (p−1)j
←− ←−
K|k j
∼
= lim IFp [[Gk ]]f0 (p−1)j ∼
= IFp [[Gk ]]IN
←−
j
Since Vkab is ZZp -torsion-free, we may apply the topological Nakayama lemma
in a similar way as in the proof of (7.4.4)(ii), to obtain the desired isomorphism
Vkab ∼= ZZp [[Gk ]]IN . 2
§5. Explicit Determination of Local Galois Groups
The absolute Galois group Gk of a local field is prosolvable. It is the

projective limit of the groups G(K|k) of the finite Galois extensions K|k which
contain the inertia group T (K|k) and the ramification group V (K|k) as normal
subgroups. V (K|k) is a p-group, and T (K|k)/V (K|k) and G(K|k)/T (K|k)
are cyclic, hence G(K|k) is a solvable group. Furthermore, the inertia group
Tk = lim T (K|k) and the ramification group Vk = lim V (K|k) are normal
←− K ←− K
subgroups of Gk . From (7.1.8)(i) and (7.4.5) we obtain:
(7.5.1) Proposition. The ramification group Vk of the absolute Galois group

Gk of a local field k with residue characteristic p is a free pro-p-group of
countably infinite rank.
The group Gk has several interesting quotients. The simplest is the quotient
by the inertia group Tk . This is the Galois group Γ = G(k̃|k) of the maximal
unramified extension k̃|k. It is canonically isomorphic to the absolute Galois
group Gκ of the finite residue field κ, hence to ẐZ, and has the Frobenius
automorphism σk as a canonical topological generator.
We next consider the quotient Gk of Gk by the ramification group Vk . This
is the Galois group
Gk = G(ktr |k)

of the maximal tamely ramified extension ktr |k. Gk is a group extension of

Gk̃ = G(ktr |k̃) by Γ , i.e. we have an exact sequence
1 −→ Gk̃ −→ Gk −→ Γ −→ 1,
which splits after choosing of a pre-image σ of σk ∈ Γ . In other words, Gk is
the semi-direct product of Gk̃ , which is an abelian group, and Γ . We obtain a
complete description of Gk by determining explicitly Gk̃ as a Γ -module. This
is easily achieved:
0
Let p be the residue characteristic of k and let µ(p ) be the group of roots
0
of unity in k̄ of order prime to p. µ(p ) is a Γ -module, which is canonically
isomorphic to κ̄× , via the reduction map. The value group [of the normalized
1
valuation of k is ZZ, and that of its extension to ktr is ∆ = n
ZZ, hence
p-n
(p0 )
M
∆/ZZ = Q` /ZZ` =: (Q/ZZ) .
`=
/p
By [160], chap.II,(9.15), we have canonically

= Hom(∆/ZZ, κ̄× ). ∗)
Gk̃ ∼
The Frobenius automorphism σk ∈ Γ acts on both groups by x 7→ xq (q = #κ),
since this is its action on κ̄× . Thus Gk̃ is canonically isomorphic to the additive
(p0 ) (p0 ) Y
Γ -module ẐZ (1), which is isomorphic to ẐZ = ZZ` as an abelian group,
`=
/p
and on which σk acts as multiplication by q. We have thus obtained the
(7.5.2) Proposition. If Gk denotes the Galois group of the maximal tamely

ramified extension ktr of a local field k, then we have a split group extension
(p0 )
1 −→ ẐZ (1) −→ Gk −→ Γ −→ 1.
We may reformulate this result as follows (cf. [87]).
(7.5.3) Theorem (IWASAWA). The Galois group Gk of the maximal tamely

ramified extension of a local field k is isomorphic to the profinite group G
generated by two elements σ, τ with the only relation
στ σ −1 = τ q .
∗) This isomorphism comes from Kummer theory via the following observation: let K|k̃
be a finite extension of degree e with (e, p) = 1 and let Π, π be uniformizers of K, k̃. Then
Πe = π · u, where u ∈ K is a unit. Note that u1/e ∈ K and (Π/u1/e )e = π. Hence
K = k̃(π 1/e ).

Proof: Let F be the free profinite group generated by two elements σ and
τ . Let N be the normal closed subgroup of F generated by στ σ −1 τ −q , and
set G = F/N . G is the profinite group generated by the images σ̄, τ̄ of σ, τ
with the defining relation σ̄ τ̄ σ̄ −1 = τ̄ q . The homomorphism F → Γ , given by
σ 7→ σk , τ 7→ 1, induces a surjection G → Γ . The kernel is the closed normal
subgroup Z topologically generated by τ̄ . In fact, Z is normal in G because
σ̄ τ̄ σ̄ −1 = τ̄ q , and G/Z is generated by the image of σ̄, i.e. is procyclic with a
surjection G/Z → Γ which must be an isomorphism.
Writing Z additively, the action of σk on Z becomes multiplication by q.
Since it is an automorphism of Z, the p-Sylow subgroup of Z must be trivial.
(p0 ) Y
In other words, Z is a quotient of ẐZ = ZZ` .
`=
/p
Now consider the Galois group Gk = G(ktr |k). Let τ 0 be a topological

(p0 )
generator of Gk̃ = G(ktr |k̃) ∼ = ẐZ (1) and σ 0 a pre-image of σk ∈ Γ . Then
σ 0 τ 0 σ 0−1 = τ 0q , and the surjective homomorphism F → Gk , given by σ 7→ σ 0 ,
τ 7→ τ 0 , factors through G. We obtain an exact commutative diagram
1ÐËÌÍÎÏÑÒÓÔÕ Z G Γ 1
α β
(p0 )
1 ẐZ (1) Gk Γ 1,
(p0 )
where α is surjective. But α is necessarily an isomorphism, since ẐZ (1) is
(p0 ) Y (p0 )
the procyclic group ẐZ = `=/p ZZ` , and Z is a quotient of ẐZ . Therefore β
is an isomorphism, and the theorem is proved. 2
We want to take a closer look at the group Gk . For this we introduce the
following notation.
If G is a profinite group and g an element of G, then we define the α-power g α
of g for α ∈ ẐZ as follows: Consider the homomorphism
ϕ : ẐZ hgi ⊆ G,
which is given by 1 7→ g. Then we define g α = ϕ(α). Observe that for α ∈ ZZ
we obtain the usual powers of g.
For a prime number ` consider the projectors π` , ∆` ∈ ẐZ ∼
Y
= ZZr given by r
( (
1 for r = ` 0 for r = `
(π` )r = (∆` )r =
0 for r =/ ` , 1 for r =/ ` .
Then
(`0 )
π` ẐZ = ZZ` and ∆` ẐZ = ẐZ ,

(`0 )
where ZZ` and ẐZ are embedded as direct factors into ẐZ. Further, ∆` + π` = 1,
∆` π` = 0 and for every α ∈ ẐZ:
π`α = π` and ∆α` = ∆` .
If G is a profinite abelian group, which (written additively) is a ZZr -module,
then raising an element to the π` -power is the zero map if r =/ ` and is the
identity if r = `.
For a prime number ` =/ p and a prime number r|(` − 1) we define an (` − 1)-th
root of unity of r-power order
e(`, r) ∈ µ`−1 (r) ⊆ ZZ`
by
Y
p≡ e(`, r) mod `
r|`−1
(e(`, r) depends on p although we do not indicate this in the notation). We

make the convention that e(`, r) = 1 if r - (` − 1) and we put
Y
e(`) = e(`, r) ∈ ZZ` ,
r|`−1
i.e. π` p = e(`) · u, where u ∈ ZZ` is a principal unit.
(7.5.4) Lemma. (i) Let r be a prime number. Then for every n ∈ IN, there
exists a prime number ` =/ p such that the r-power root of unity e(`, r) is at
least of order rn .
(ii) For q = pf , 1 ≤ f ∈ IN, the homomorphism
0
ψ : ẐZ −→ Aut(ZZ(p ) ) α 7−→ (x 7→ q α x)
is injective.
√
Proof: (i) Consider the fields k0 = Q(µrn ) and k = k0 ( r p). The set of prime
numbers ` which are completely decomposed in k is
`−1
{` | ` ≡ 1 mod rn , p r ≡ 1 mod `} ,
since a prime number ` splits completely in k if and only if ` - p r, ` ≡ 1
√
mod rn and r p ∈ ZZ` . But
√ √
Z` ⇔ r e(`) ∈ ZZ`
r p ∈ Z
`−1
⇔ e(`) r = 1
`−1
⇔ p r ≡ 1 mod ` .
Therefore the density of the set
`−1
S = {` | ` ≡ 1 mod rn , p r 6≡ 1 mod `}

is [k0 : Q]−1 − [k : Q]−1 = 1/rn and, in particular, S is not empty. Thus for
` ∈ S, the root of unity e(`, r) has order rm with m ≥ n. This proves (i).
In order to prove (ii), assume that α is an element in the kernel of ψ, i.e. for
all ` =/ p the equality q α π` = π` holds. Then, in particular,
e(`, r)f α = (e(`)f )α = 1 ∈ ZZ`
Y
r|`−1
and hence also e(`, r)f α = 1 for all ` =/ p, r|(` − 1).

For an arbitrary prime number r and an arbitrary n ∈ IN, we can use (i) in order
to find a prime number ` =/ p such that
n−1
e(`, r)r =/ 1.
Hence f α ∈ rn ẐZ and since r and n were arbitrary, we conclude that f α = 0.
Finally, since 1 ≤ f ∈ IN and since ẐZ is torsion-free (as an abelian group), we
obtain α = 0.
An alternative possibility to see assertion (ii) is the following: the homo-
morphism ψ (after replacing q by q −1 ) describes the action of G(κ̄|κ) on the
dual Hom(κ̄× , Q/ZZ) of the multiplicative group κ̄× . Obviously, this action is
faithful. 2
For a prime number r (possibly equal to p), we set
σ r = σ πr , τr = τ πr
where σ and τ are generators of the group Gk = hσ, τ | τ σ = τ q i, q = pf .
(7.5.5) Proposition. With the notation as above, we have for arbitrary prime
numbers r, and ` =/ p
 −f



 τ`qe(`) −1 if r = ` ,
f
[σr , τ` ] =  τè(`,r) −1 if r|` − 1 ,

 1 if r =/ ` , r - ` − 1 .

Proof: For r =/ ` we have

π` q πr = π` pπr f = (π` p)πr f = e(`, r)f ,
since all other components in the decomposition of π` p are annihilated (i.e.
sent to 1) when raised to the πr -th power. Similarly,
π` q π` = π` pπ` f = (π` p)π` f = π` qe(`)−f .

It follows that
 −f


τ`qe(`) −1 if r = ` ,
πr −1 πr −1)
 f
[σr , τ` ] = τ`q = τ`π` (q = τè(`,r) −1 if r|` − 1 ,


 1 if r =/ ` , r - ` − 1 .

(7.5.6) Corollary. For every prime number r there exists a prime number
` =/ p such that σr and τ` do not commute. In particular, σp does not commute
with τ .
Proof: By (7.5.5), we have the equality

f −1
[σr , τ` ] = τè(`,r)
for every prime number r with r|` − 1, and from (7.5.4)(i) it follows that there
are prime numbers ` such that the order of e(`, r) is bigger than the r-part of
the fixed number f . 2
(7.5.7) Corollary.
(i) The ramification group Vk of the absolute Galois group Gk of a local
field k is the maximal normal pro-p subgroup of Gk .
(ii) The subgroup Tk /Vk = hτ i of Gk = Gk /Vk is the unique maximal abelian
normal subgroup of Gk (which exists in Gk !).
Proof: (i) Since Vk is a normal pro-p subgroup of Gk and hVk , σp i is a pro-p

Sylow subgroup of Gk , the result follows from the previous corollary.
(ii) Every abelian normal subgroup of Gk is contained in hτ i by (7.5.5) and
(7.5.4)(i). Thus the result follows. 2
We next study the maximal pro-p-factor group Gk (p) of Gk , where p is the

residue characteristic of the local field k. This is the Galois group G(k(p)|k)
of the maximal p-extension k(p)|k, i.e. of the composite of all finite Galois
extensions of p-power degree.
First, we have to compare the cohomology for a Gk (p)-module A with
respect to Gk (p) and Gk . This will be done in the next proposition in a slightly
more general form. Recall the notion G(c) for the maximal pro-c-quotient of a
profinite group G with respect to a full class c of finite groups.

(7.5.8) Proposition. Let c be a full class of finite groups, ` a prime number

such that ZZ/`ZZ ∈ c and A a Gk (c)-module. Then for every i ≥ 0 the inflation
map
H i (Gk (c), A)(`) −→ H i (Gk , A)(`)
is an isomorphism.
Proof: The degree of k(c)|k is infinitely divisible by `, since the maximal

unramified `-extension of k is contained in k(c) and has Galois group ZZ` .
Using (7.1.8)(i) and (6.1.3), we obtain cd` (k(c)) ≤ 1. Let H = G(k̄|k(c)). Then
H i (H, A)(`) = 0 for i ≥ 1. Indeed, since cohomology commutes with direct
limits and since A is a trivial H-module, one reduces to the cases A = ZZ, ZZ/`ZZ
and, using the exact sequence 0 → ZZ → Q → Q/ZZ → 0, we finally reduce to
the case A = ZZ/`ZZ. By cd` H ≤ 1, the assertion is obvious for i ≥ 2.
Since ZZ/`ZZ ∈ c, the group H has no nontrivial homomorphism to an `-group,
showing the case i = 1.
Now the spectral sequence
H i (Gk (c), H j (H, A))(`) ⇒ H i+j (Gk , A)(`)
gives isomorphisms
H i (Gk (c), A)(`) ∼
= H i (Gk , A)(`) for all i ≥ 0 . 2
We can now explicitly determine the structure of the pro-`-group Gk (`). A

consequence of (7.5.3) is the
(7.5.9) Proposition. Let k be a local field with residue characteristic p and let
` be a prime number different to p.
/ k, then Gk (`) ∼
If µ` ⊆ = ZZ` ; if µ` ⊆ k, then Gk (`) is a Poincaré group of
dimension 2 (i.e. a Demuškin group) of rank 2 with dualizing module µ(`), the
group of all `-power roots of unity in k(`). Furthermore,
Gk (`) ab ∼
= ZZ/`s ZZ ⊕ ZZ` ,
where `s = #µ(k)(`).
From (6.1.7), we obtain the
(7.5.10) Theorem. Let k be a local field with char(k) = p. Then Gk (p) is a

free pro-p-group of countably infinite rank.

(7.5.11) Theorem. Let k be a p-adic local field.

(i) If µp is not contained in k, then Gk (p) is a free pro-p-group of rank N + 1
with N = [k : Qp ].
(ii) If µp ⊆ k, then Gk (p) is a Poincaré group of dimension 2 (i.e. a Demuškin
group) of rank N + 2 . The dualizing module of Gk (p) is the group µ(p)
of all p-power roots of unity in k(p).
Proof: By the (7.5.8) and by the duality theorem (7.2.6), we obtain

(1) H 1 (Gk (p), ZZ/pZZ) = H 1 (Gk , ZZ/pZZ) ∼
= H 1 (Gk , µp )∗ ∼
= (k × /k ×p )∗ ,
(2) H 2 (Gk (p), ZZ/pZZ) = H 2 (Gk , ZZ/pZZ) ∼
= H 0 (Gk , µp )∗ .
From [160], chap.II, (5.7)(i), we get k × /k ×p ∼
= (ZZ/pZZ)r with r = N + 1 if
/ k, or N + 2 if µp ⊆ k. Therefore, by (3.9.1), the pro-p-group Gk (p) has
µp ⊆
rank
rk(Gk (p)) = N + 1 or N + 2,
according to whether µp ⊆ / k or µp ⊆ k.
If µp / k, then from (2) it follows that H 2 (Gk (p), ZZ/pZZ) = 0, and Gk (p) is
⊆
a free pro-p-group by (3.9.5). This proves (i).

Assume µp ⊆ k. Then (2) implies that H 2 (Gk (p), ZZ/pZZ) ∼ = ZZ/pZZ and
i
H (Gk (p), ZZ/pZZ) = 0 for i > 2 by (7.5.8) and (7.1.8), so that cdp Gk (p) = 2.
By the duality theorem (7.2.6), the cup-product yields a non-degenerate pairing
H 1 (Gk (p), ZZ/pZZ) × H 1 (Gk (p), ZZ/pZZ) → H 2 (Gk (p), ZZ/pZZ) ∼
= ZZ/pZZ.
Since µp ⊆ k, the p-part µ(p) of the dualizing module of µ of Gk (see (7.2.4))
is a Gk (p)-module. Therefore (7.5.8) shows that µ(p) is the dualizing module
of Gk (p) for the category of p-torsion Gk -modules. From (3.7.2) it follows
now that Gk (p) is a Poincaré group. 2
When µp ⊆ k we get an explicit description of Gk (p) by applying the results

of (3.9.11) and (3.9.19).
(7.5.12) Theorem (DEMUŠKIN). Let k be a p-adic local field of degree N =

[k : Qp ] and let ps be the largest p-power such that µps ⊆ k. If ps > 2, then
Gk (p) is the pro-p-group defined by N + 2 generators x1 , . . . , xN +2 subject to
the one relation
s
xp1 (x1 , x2 )(x3 , x4 ) · · · (xN +1 , xN +2 ) = 1.

Proof: By class field theory the abelianized group Gk (p)ab is isomorphic

to the pro-p-completion k̂ × of the multiplicative group k × which, by [160],
× ∼
chap.II, (5.7), is isomorphic to ZZ ⊕ ZZ/(pf − 1)ZZ ⊕ ZZ/ps ZZ ⊕ ZZN
p , i.e. k̂ =
ZZ/ps ZZ ⊕ ZZN
p
+1
. Therefore ps is the number which we have denoted by q in
(3.9.11). Moreover, we have
H 1 (Gk (p), ZZ/pZZ) ∼
= Hom(Gk (p), ZZ/pZZ) ∼
= (ZZ/pZZ)N +2 ,
so that rk(Gk (p)) = N + 2. The theorem follows now from (3.9.11). 2
The structure of Gk (2) when ps = 2 has been determined by J.-P. SERRE if

N is odd and in general by J. LABUTE (cf. (3.9.19) and [209], [117]). Gk (2)
is again generated by N + 2 generators x1 , . . . , xN +2 with one defining relation
ρ, but the shape of this relation depends on further conditions:
If N is odd, then
ρ = x21 x42 (x2 , x3 )(x4 , x5 ) · · · (xN +1 , xN +2 ).
In particular, if k = Q2 , then Gk (2) is generated by three elements with the

defining relation x2 y 4 (y, z) = 1.
If N is even, then we have to take the 2-part of the cyclotomic character
χ : Gk → ZZ× 2 into account, which is obtained from the action of Gk on the
group µ2 of all roots of unity of 2-power order. We have End(µ2∞ ) = ZZ2
∞
and the action of Gk gives the homomorphism χ : Gk → Aut(µ2∞ ) = ZZ× 2 . The

structure of Gk (2) depends now on the image of χ:
If im(χ) is the closed subgroup of ZZ× f
2 generated by −1 + 2 (f
≥ 2), then
f
ρ = x2+2
1 (x1 , x2 )(x3 , x4 ) · · · (xN +1 , xN +2 ).
If im(χ) is generated by −1 and 1 + 2f (f ≥ 2), then

f
ρ = x22 (x1 , x2 )x23 (x3 , x4 ) · · · (xN +1 , xN +2 ).
The cohomological methods leading to these results were extended to give

an explicit description of the entire absolute Galois group Gk of a local field. In
the case that k is the field of Laurent series over a finite field, this was done by
H. KOCH [108]. If k is a p-adic local field, the structure of Gk was determined
by U. JANNSEN and K. WINGBERG [99] for p =/ 2; and by V. DIEKERT [39] for
p = 2 under the condition that k(µ4 )|k is unramified. The structure of GQ2 is
not known.
The result for local fields of positive characteristic is the following.

(7.5.13) Theorem (KOCH). Let k be a local field of characteristic p with residue

field κ of cardinality q. Then Gk is isomorphic to the semi-direct product
Gk ∼
= ∗F Gk
ω o Gk ,
where Gk = G(ktr |k) is the Galois group of the maximal tamely ramified
∗
extension of k, and Gk Fω is the free pro-p-Gk operator group of countably
infinite rank (cf. (4.3.8)).
In other words, the group Gk has generators σ, τ and xi , i ∈ IN, subject to
the following defining conditions resp. relations:
A) The closed normal subgroup topologically generated by xi , i ∈ IN, is a

pro-p-group.∗)
B) The elements σ, τ satisfy the relation στ σ −1 = τ q .
Proof: By (7.5.1) and (7.4.5)(ii), the ramification group Vk is a free pro-p-

group, and we have an isomorphism of compact ZZp [[Gk ]]-modules
Vkab ∼
= ZZp [[Gk ]]IN ,
i.e. Vkab is a free ZZp [[Gk ]]-module of countably infinite rank. Since the order of
G(ktr |k̃) is prime to p, we have cdp Gk = cdp G(k̃|k) = 1. Therefore the group
Gk is p-projective by (3.5.6), and the exact sequence
1 −→ Vk −→ Gk −→ Gk −→ 1
splits. Fixing a splitting Gk ∼
= Vk o Gk , the ramification group Vk becomes
a pro-p-Gk operator group. Therefore we find a surjective homomorphism of
pro-p-Gk operator groups
ψ: ∗F
Gk
ω Vk ,
∗
where Gk Fω is the free pro-p-Gk operator group of countably infinite rank and
ψ is defined by mapping the generators of Fω to lifts of the ZZp [[Gk ]]-generators
of Vkab . Since Vk is a free pro-p-group and the map
ψ̄ : ( ∗F
Gk
ω )ab = ZZp [[Gk ]]IN −→
∼ V ab
k
induced by ψ is an isomorphism, ψ is bijective by (1.6.15). Therefore we

obtain an isomorphism
(ψ, id) : ∗F
Gk
ω
∼ V o G ∼ G .
o Gk −→ k k = k 2
∗) This topological condition could be replaced by an infinite set of algebraic relations.

Now let k be a p-adic field. We describe the structure of Gk in the case

p =/ 2. It depends on the degree N = [k : Qp ], the cardinality q = #κ of the
residue field κ, the order ps of the group µps of all p-power roots of unity in the
maximal tamely ramified extension ktr |k and on two further numbers g, h ∈ ZZp
which are defined as follows. By (7.5.3), the Galois group Gk = G(ktr |k) is
generated by two elements σ, τ with defining relation στ σ −1 = τ q . The actions
of σ and τ on µps are given by two numbers g, h ∈ ZZp such that
ζ σ = ζ g , ζ τ = ζ h for ζ ∈ µtr .
In the following we denote the commutator xyx−1 y −1 by [x, y] in contrast

to the commutator (x, y) = x−1 y −1 xy used in III §8. If one had taken there
the commutator [x, y] instead of (x, y), the shape of the Demuškin relation
would not change because it is the same modulo F 3 (the third filtration step of
the descending p-central series), and then the iteration process works as well.
Which commutator to use is just a matter of personal taste.
(7.5.14) Theorem (JANNSEN-WINGBERG). The group Gk is isomorphic to the

profinite group generated by N + 3 generators σ, τ, x0 , . . . , xN , subject to the
following defining conditions resp. relations.
A) The closed normal subgroup, topologically generated by x0 , . . . , xN is a
pro-p-group.
B) The elements σ, τ satisfy the “tame” relation
στ σ −1 = τ q .
C) In addition, the generators satisfy one further relation:

(i) for even N
s
xσ0 = hx0 , τ ig xp1 [x1 , x2 ][x3 , x4 ] · · · [xN −1 , xN ],
(ii) for odd N

s
xσ0 = hx0 , τ ig xp1 [x1 , y1 ][x2 , x3 ][x4 , x5 ] · · · [xN −1 , xN ],
where
p−1 p−2
π
hx0 , τ i = (xh0 τ xh0 τ · · · xh0 τ ) p−1
(π = πp being the element of ẐZ with π ẐZ = ZZp ), and where y1 is a certain
element in the subgroup generated by x1 , σ, τ , described below.

The definition of the element y1 is a little bit subtle. Let α : Gk →(ZZ/ps ZZ)×
be the character describing the action of Gk = G(ktr |k) on µps , and let β :
Gk → ZZ×p be a lift of α (not necessarily homomorphic). For ρ ∈ hσ, τ i
⊆ Gk
and x Gk set
∈
π
p−2
{x, ρ} := (xβ(1) ρ2 xβ(ρ) ρ2 · · · xβ(ρ ) ρ2 ) p−1 .
Writing τ2 = τ π2 and σ2 = σ π2 , where π2 is the element of ẐZ with π2 ẐZ = ZZ2 ,
y1 is given by p+1
τ p+1 a b 2
y1 = x12 {x1 , τ2p+1 }σ2 τ2 {{x1 , τ2p+1 }, σ2 τ2a }σ2 τ2 +τ2 .
Here a, b ∈ ZZ are chosen in such a way that
−α(στ a ) mod p ∈ (IF× 2 b
p ) and − α(στ ) mod p ∈
× 2
/ (IFp ) .
For the proof we refer to [99] and [39]. It is based on a theory of H. KOCH
[111] which axiomizes the fact that for every finite, tamely ramified extension
K|k the group GK (p) is a Demuškin group. We would like to mention the
following special cases.
For p > 2, the group GQp has four generators σ, τ, x0 , x1 satisfying the
relations
τσ = τp ,
p−1 p−1 p+1 p+1
τ p+1
xσ0 = hx0 , τ i xp1 [x1 , x12 {x1 , τ2p+1 }σ2 τ2 {{x1 , τ2p+1 }, σ2 τ2 2 }σ2 τ2
2 2 +τ 2
2 ],
and for the group GQp (ζp ) there are generators σ, τ, x0 , . . . , xp−1 satisfying
τσ = τp ,
xσ0 = (x0 τ )π xp1 [x1 , x2 ] · · · [xp−2 , xp−1 ] .
Remarks: 1. Although we know by (7.4.1) that Gk can be generated by N + 2

elements, it is more convenient to use N +3 generators in order to obtain “nice”
relations.
In the case that µp ⊆ k, there is also a satisfactory description with N + 2
generators, cf. [99], §1.4(d).
2. Generators and relations for Gk were also assigned by A. V. JAKOVLEV in
[92]. However, some mistakes required a comprehensive correction. This has
been sketched only for the case of even N , and produced three relations, one
of them being a somewhat complicated limit (cf. [93]).
We finish this section by showing that the absolute Galois group Gk of a

p-adic local field k, p =/ 2, possesses nontrivial outer automorphisms. This is

easy to see if N = [k : Qp ] > 1. Let σ, τ, x0 , x1 , . . . , xN be the generators of

Gk described in (7.5.14), satisfying the tame relation στ σ −1 = τ q and the wild
relation s
xσ0 = hx0 , τ ig xp1 [x1 , −] · · · [xN −1 , xN ] .
If we define ψ : Gk −→ Gk by
ψ(y) = y for y = σ, τ, x0 , . . . , xN −1 and ψ(xN ) = xN · xN −1 ,
then ψ is an automorphism of Gk . Indeed, since
[xN −1 , xN ] = [xN −1 , xN xN −1 ] ,
the generators σ, τ, x0 , . . . , xN −1 , xN · xN −1 satisfy both relations if N > 1.
Now suppose that ψ is an inner automorphism, i.e. there is an element ρ ∈ Gk
such that ψ(z) = z ρ for all z ∈ Gk . Recall that Vki denotes the i-th term
of the p-central series of the ramification group Vk with Vk1 = Vk . Since
xρN −1 = ψ(xN −1 ) = xN −1 and since xN −1 Vk2 generates a free IFp [[Gk ]]-module in
Vk /Vk2 , by [99], §2, we obtain that ρ ∈ Vk . It follows that xN Vk2 = xN · xN −1 Vk2
which is a contradiction. Thus, if N = [k : Qp ] > 1, we have constructed a
nontrivial outer automorphism of Gk .
The case k = Qp is more difficult. We use the following result from [250].

ψ3 : Gk /Vk3 −→
∼ G /V 3
k k
be an automorphism of Gk /Vk3 which induces the identity on the factor group

Gk = Gk /Vk . Then there exists an automorphism ψ of Gk which coincides
with ψ3 modulo Vk2 .
Remark: This result was proven in [250], Satz 2, but was stated there in an
incorrect manner. The result above is exactly what was needed for all results
in the two papers [99] and [250], except for the statement in [99], §5.1. There
an automorphism of GQp was defined and claimed to be a nontrivial outer
automorphism. But the argument used the incorrect formulation of [250],
Satz 2 and therefore the constructed automorphism might be inner.
In order to proceed, we need some more notation. The homomorphism

α : Gk →(ZZ/ps ZZ)× induces an involution ∗ on IFp [[Gk ]] which is defined by
ρ∗ = α(ρ)ρ−1 for ρ ∈ Gk . We define the element E of IFp [[Gk ]] by
1 e−1
τ 2i α(τ )−i ,
X
E = lim
K e
i=0

where K runs through all finite tamely ramified Galois extensions of k and e
denotes the ramification index of K|k. Then E ∗ = E and E is an idempotent
which is central in IFp [[Gk ]], because
1 e−1
τ 2iq α(τ )−i σ = E q σ = Eσ.
X
σE = lim
K e i=0
From now on, let k = Qp . Then q = p, α(σ) = 1 and α(τ ) is a primitive

(p − 1)-th root of unity. For
ε = 1 − 2E
we have ε∗ = ε and ε2 = 1. In particular, ε is a central unit in IFp [[GQp ]]. We
write G, G and V for GQp , GQp and VQp , respectively.
Since cdp G = 1, the group G is p-projective by (3.5.6), and the exact
sequence
1 −→ V −→ G −→ G −→ 1
splits. Fixing a splitting G ∼= V o G, the ramification group V becomes
a pro-p-G operator group. Therefore we find a surjective homomorphism of
pro-p-G operator groups
f: ∗F
G
2 V,
∗
where G F2 is the free pro-p-G operator group of rank 2 with basis {x0 , x1 }
and f is defined by mapping the generators x0 and x1 to lifts of the ZZp [[G]]-
generators of V ab . We obtain a surjection
∗
F := F2 o G V o G ∼
G
= G.
Let ψ : F → F be defined by
ψ(y) = y for y = σ, τ, x0 and ψ(x1 ) = xε1
(where the “sum” ε is chosen in some ordering). Then ψ is an automorphism
of F since it is an automorphism modulo F 2 (ε is a unit in IFp [[G]]).
We will show that for every finite tamely ramified Galois extension K|k the
automorphism ψ induces an automorphism
ψK : GK (p)/GK (p)3 −→
∼ G (p)/G (p)3 .
K K
In order to prove this, let κK and λK be elements of IFp [[G]] defined as

−1
fX
κK = σ i α(σ)−i ,
i=0
1 e−1
τ i α(τ )−i ,
X
λK =
e i=0

where e and f denote the ramification index and the residue degree of K
respectively. Let FK be the pre-image of GK in F. By [99], §2 (observe that
the index of the p-central series differs there from our notation by −1), the
relation rK in GK (p) satisfies
p
rK ≡ (x−σ
0 hx0 , τ ix1 [x1 , y1 ])
κK λK
mod FK (p)3 .
Let δ ∈ IFp [[G]] be such that y1 = xδ1 mod F 2 . Then we get
κK λK pεκK λK
ψ(rK ) ≡ (x−σ
0 hx0 , τ i) x1 [xε1 , xεδ
1 ]
κK λK
κK λK pκK λK ∗
≡ (x−σ
0 hx0 , τ i) x1 [x1 , xεδε
1 ]κK λK
κK λK pκK λK
≡ (x−σ
0 hx0 , τ i) x1 [x1 , xδ1 ]κK λK
≡ rK mod FK (p)3 ,
since ελK = λK and ε is central in IFp [[G]], so that εδε∗ = δεε∗ = δε2 = δ. This
gives us the automorphism ψK .
Now, in the limit over all K, ψ induces an automorphism of V /V 3 which is
G-invariant because ψ|G = id. Thus we obtain an automorphism of G/V 3 ∼ =
3
V /V o G, which is also denoted by ψ. By (7.5.15), it extends to an automor-
phism ϕ of G which coincides with ψ modulo V 2 .
Suppose that ϕ is an inner automorphism, i.e. there exists an element ρ ∈ G
such that ϕ(z) = z ρ for all z ∈ G. Then xρ1 ≡ ϕ(x1 ) ≡ ψ(x1 ) ≡ xε1 mod V 2 .
Since x1 generates a free IFp [[G]]-module in V /V 2 , by [99], §2, we obtain ε = ρ
mod V ∈ G, which is a contradiction. Thus we have shown that GQp possesses
nontrivial outer automorphisms.


Chapter VIII
Cohomology of Global Fields
§1. Cohomology of the Idèle Class Group
Having established the cohomology theory for local fields, we now begin its
development for global fields, i.e. algebraic number fields and function fields
in one variable over a finite field. The cohomology theory treats both types of
fields equally.
The role that the multiplicative group of fields played in the local theory is
now taken over for a global field k by the idèle class group. Let k be a global
field, k̄ a separable closure of k and Gk = G(k̄|k) its absolute Galois group.
The idèle group Ik of k is defined as the restricted product
kp× ,
Y
Ik =
p
where p runs through all primes of k including the archimedean ones if k is
a number field, kp is the completion of k at p and the restricted product is
taken with respect to the unit groups Up in kp× . The multiplicative group k ×
of k injects diagonally into Ik and we define the idèle class group of k as the
quotient
Ck := Ik /k × .
We refer the reader to [160], chap.VI, §1 for basic properties of the idèle and
the idèle class group.
If K is a finite separable extension of k, then Ik naturally injects into IK and
we call the direct limit
I = lim
−→
IK
K|k
the idèle group of k̄. I is a discrete Gk -module and one easily observes that
IK = I GK
for every finite separable extension K|k. The quotient
C := I/k̄ ×
is the idèle class group of k̄. We have
C = lim
−→
CK
K|k

426 Chapter VIII. Cohomology of Global Fields
and a straightforward application of Hilbert’s Satz 90 implies that for every

finite separable extension K|k,
CK = C GK .
For every (possibly infinite) separable extension K of k we put IK = I GK ,
CK = C GK , and if K|k is Galois, we set
H i (K|k) = H i (G(K|k), CK ),
and we also write Ĥ i (K|k) for Ĥ i (G(K|k), CK ).
The basis of our results in this chapter is the following theorem, called the
class field axiom. It is an immediate consequence of the so-called first and
second fundamental inequalities for global fields. For a proof of these results,
we refer the reader to [6], chap.5,6, or, for the number field case, to [160],
chap.VI, (4.4).
(8.1.1) Theorem (Class Field Axiom). For a finite cyclic extension K|k we
have 
 [K : k] for i = 0,

i
#Ĥ (K|k) =
 1 for i = 1.

Since H 1 (K|k) ∼
= Ĥ −1 (K|k) for K|k cyclic, the class field axiom is a statement
about the kernel and cokernel of the norm map
NK|k
Ö
CK /IG(K|k) CK Ck
(here IG(K|k) is the augmentation ideal in ZZ[G(K|k)]), and can therefore be
considered as a noncohomological assertion. Starting with this input, we will
calculate the cohomology groups of CK for arbitrary Galois extensions.
Let K|k be a finite separable extension. Setting

×
Y
IK (p) = KP ,
P|p
we obtain a decomposition
Y
IK = IK (p),
p
where p runs through all primes of k and the restricted product is taken with
respect to the subgroups
Y
UK (p) := UK,P ,
P|p
×
where UK,P denotes the group of units in KP .

§1. Cohomology of the Idèle Class Group 427
Passing to various extension fields, we frequently have to choose compatible

prolongations of primes. Therefore we make the following fixed choice. For
every prime p of k, we choose a k-embedding ip : k̄ ,→ k̄p , in particular a
prime p of k̄ above p, see [160], chap.II, (8.1). We denote the subextensions of
k̄|k by K|k and for each such extension the prime of K lying under p by the
pointed letter P. . Thus every separable extension K|k comes equipped with a
distinguished prime P. above p. By abuse of notation we write
Kp := ip (K)kp
for the extension KP. of the local field kp which corresponds to the prime P.
of K (and which is the completion of K at P. if K|k is finite). We adopt the
×
same convention for the unit groups and write UK,p for group of units in KP..
Furthermore, we write Gp (K|k) (or simply Gp if K is clear from the context)
for the decomposition group of P. in G(K|k), i.e. Gp (K|k) = G(Kp |kp ).
Now let K|k be a finite Galois extension and let G = G(K|k) be its Galois
group. The idèle group IK is then a G-module and
Y
IK = IK (p)
p
is a decomposition into G-modules. Then
Y
×
Y
× p × G
IK (p) = KσP. = σKP. = IndG (Kp ),
σ ∈G/Gp σ ∈G/Gp
G
and similarly UK (p) = IndGp (UP. ). By Shapiro’s lemma, we obtain isomor-
phisms
Ĥ i (G, IK (p)) ∼
= Ĥ i (Gp , Kp× )
for all i ∈ ZZ. Furthermore, if p is unramified in K, we have for all i ∈ ZZ
Ĥ i (G, UK (p)) ∼
= Ĥ i (Gp , UP. ) = 0
by (7.1.2)(i). From this follows the
(8.1.2) Proposition. For a finite Galois extension K|k, we have

Ĥ i (G, IK ) ∼ Ĥ i (Gp , K × )
M
= p
p
for all i ∈ ZZ, where p runs through all primes of k.
Proof: Let S run through the finite sets of primes of k containing the primes
p ramified in K and the infinite primes if k is a number field. Setting
S
Y Y
IK = IK (p) × UK (p),
p∈S /S
p∈
we have IK = lim
−→ K
I S and
S

Ĥ i (G, IK ) = lim
−→
Ĥ i (G, IK
S
)
S
Y Y
= lim Ĥ i (G, IK (p)) × Ĥ i (G, UK (p))
−→
S p∈S /S
p∈
Y Y
i i
= lim
−→
Ĥ (G, IK (p)) × Ĥ (G, UK (p))
S p∈S /S
p∈
∼
Y
Ĥ i (Gp , Kp× ). 2
M
= lim Ĥ i (Gp , Kp× ) =
−→
S p∈S p
By Hilbert’s Satz 90 and by (7.2.2), we obtain the
(8.1.3) Corollary. H 1 (G, IK ) = H 3 (G, IK ) = 0.
By proposition (8.1.2), we may associate to every cohomology class c ∈

Ĥ i (G, IK ) its local components cp ∈ Ĥ i (Gp , Kp× ): cp is the image of c under
the composite of the maps
resp πp
Ĥ i (G, IK ) −→ Ĥ i (Gp , IK ) −→ Ĥ i (Gp , Kp× ),
× ×
where πp is induced by the projection IK → KP . = Kp . Since these maps
commute with inf , res, cor, we obtain in a straightforward way the
(8.1.4) Proposition. Let L ⊇ K ⊇ k be finite Galois subextensions of k̄|k.

Then for i ≥ 1
K|k Kp |kp
(i) inf L|k (c)p = inf Lp |kp (cp ) for c ∈ H i (G(K|k), IK ),
k
(ii) res kK (c)P = res KpP (cp ) for c ∈ H i (G(L|k), IL ),
X K .
(iii) cor K
k (c)p = σ∗−1 corkpσP (cσP. ) for c ∈ H i (G(L|K), IL ),
σ ∈G/Gp
× ×
where σ∗ is the map H i (G(KP. |kp ), KP i
. ) −→ H (G(KσP. |kp ), KσP. ) induced
by σ : KP. → σKP. = KσP. . For the last two formulae it suffices to require
only that L|k is Galois.
Although the idèle group I = lim IK of k̄ is not the restricted product

−→ K|k
k̄p× , we obtain the following direct decomposition for its cohomology by
Q
p

applying lim with respect to the inflation maps to the formula (8.1.2):
−→ K|k
H i (k, I) ∼ H i (Gp (k̄|k), (k̄)×

M
= p) for all i ≥ 1.
p
Now the question naturally occurs of whether the canonical surjection

G(k̄p |kp ) Gp (k̄|k) is an isomorphism, or equivalently, whether (k̄)p = (kp ).
(8.1.5) Proposition. Let p be a prime of the global field k. Then

(k̄)p = (kp ).
For the proof we need
(8.1.6) Krasner’s Lemma. Let κ be a complete field with respect to a nonar-

chimedean valuation and let Ω be an algebraic closure of κ. Let α ∈ Ω be
separable over κ and let α = α1 , . . . , αn be the conjugates of α over κ. Suppose
that for β ∈ Ω we have
|α − β| < |α − αi | for i = 2, . . . , n,
where | | denotes the unique extension of the valuation to Ω. Then κ(α) ⊆ κ(β).
Proof: Consider the extension κ(α, β)|κ(β) and let K|κ(β) be its Galois
closure. Let σ ∈ G(K|κ(β)). Then σ(β − α) = β − σ(α). Since |σ(x)| = |x|
for all x (by the uniqueness of the extension of the absolute value), we have
|β − σ(α)| = |β − α| < |αi − α|
for i = 2, . . . , n. Therefore
|α − σ(α)| < max{|α − β|, |β − σ(α)|} < |α − αi |
for i = 2, . . . , n. It follows that σ(α) = α, and so α ∈ κ(β). 2
Proof of (8.1.5): We have the natural inclusion (k̄)p ⊆ (kp ). The proposition
is trivial if p is archimedean, so assume that p is finite. Let α be in (kp ) and
let f ∈ kp [X] be its minimal polynomial. Since k is dense in kp , we can
choose a polynomial g ∈ k[X] near to f . Then |g(α)| = |g(α) − f (α)| is small.
Writing g(X) = (X − βj ) with βj ∈ k̄ ⊆ (k̄)p , we see that |α − β| is small
Q
for some root β of g(X). In particular, we can choose g(X) and then β such
that |β − α| < |αi − α| for all conjugates αi ∈ (kp ) of α, αi =/ α. By Krasner’s
lemma, we obtain α ∈ kp (β) = (k(β))p ⊆ (k̄)p . 2

Passing to the limit over all finite extensions K of k inside k̄, (8.1.2) and
(8.1.5) immediately imply
H i (k, I) ∼ H i (kp , k̄p× ) for all i ≥ 1.

M
(8.1.7) Proposition. =
p
Since H 1 (k, I) = 0, the inflation maps

H 2 (G(K|k), IK ) → H 2 (k, I)
are injective. We identify the first group with its image. Then the equality
H 2 (k, I) = lim H 2 (G(K|k), IK ) becomes
−→
K|k
[
H 2 (k, I) = H 2 (G(K|k), IK ).
K|k
In the local case we have seen this with the multiplicative groups in place
of I. Like the unramified extensions in the local case, a decisive role is played
here by the cyclic extensions K|k because of their periodic cohomological
behaviour. For this reason the following analogue (8.1.9) of the local situation
is of crucial importance. But first we observe the
(8.1.8) Proposition. Let L|k be a Galois extension and let p be a prime number.
Suppose in the number field case that L is totally imaginary if p = 2. If p∞
divides the local degrees [Lp : kp ] for all finite primes p of k, then
H 2 (L, I)(p) = 0
and
H 2 (G(L|k), IL )(p) ∼
= H 2 (k, I)(p).
Proof: Let K|k run through the finite subextensions of L|k. Then H 1 (L, I) =
lim
−→
H 1 (K, I) = 0, hence the sequence
K,res
0 −→ H 2 (G(L|k), IL ) −→ H 2 (k, I) −→ H 2 (L, I)
is exact. Therefore it suffices to prove H 2 (L, I)(p) = 0. Using (8.1.7) and
passing to the direct limit, we obtain
G (L|k)
H 2 (Lp , L̄×
M
H 2 (L, I)(p) = p
IndG(L|k) p )(p).
p
For finite primes we have H 2 (Lp , L̄×

p )(p) = 0 by (7.1.6), and for the archimedean
primes the group H 2 (Lp , L̄×
p )(p) is trivially zero for p =/ 2 and by assumption
also for p = 2. 2

Let k(µ) be the extension of k obtained by adjoining all roots of unity in

k̄ to k. This extension has an abelian Galois group. In the function field
case, k(µ) is the extension obtained by passing to the algebraic closure of the
constant field and G(k(µ)|k) ∼ = ẐZ. We use the notation k̃ = k(µ) in this case.
If k is a number field, let T be the torsion subgroup of G(k(µ)|k). Then the
field k̃ := k(µ)T is an extension of k with Galois group ẐZ which we call the
cyclotomic ẐZ-extension of k ∗) . The decomposition group of a finite prime
Gp (k̃|k) ⊆ G(k̃|k) is open and, in particular, isomorphic to ẐZ. Since ẐZ is
torsion-free, archimedean primes are unramified in k̃|k.
[
(8.1.9) Proposition. H 2 (k, I) = H 2 (G(K|k), IK ) .
K|k
cyclic
Moreover, it suffices to take the union over all cyclic extensions of k inside k̃
if k is a function field or a totally imaginary number field. If k is a number
field having a real place, it suffices to take the union over all cyclic extensions
of k inside k̃1 , where k1 |k is an arbitrarily chosen totally imaginary quadratic
extension of k.
Proof: Let x ∈ H 2 (k, I). Using (8.1.7), we decompose x into an archimedean

and a nonarchimedean part: x = xa + xn . By (8.1.8), we find a finite exten-
sion K of k inside k̃ such that res kK xn ∈ H 2 (K, I) vanishes. This finishes the
proof if k is a function field or a totally imaginary number field. If k has a
real place, let k1 be an arbitrarily chosen totally imaginary quadratic extension
of k. Let K1 |k be the uniquely defined cyclic extension of k inside k̃ of degree
2 · [K : k]; in particular, K1 is a quadratic extension of K. The extensions
k1 |k and K1 |k are linearly disjoint because the first is totally ramified and
the second is unramified at all real places. The extension k1 K1 |K has Galois
group ZZ/2ZZ ⊕ ZZ/2ZZ and thus contains a unique subextension L|K of degree 2
distinct from K1 and k1 K. We have the following diagram of fields.
×ØÙÚÛÜÝÞß k1 K1
k1 K L K1
k1 K
k
∗) The cyclotomic ẐZ-extension is the composite of the cyclotomic ZZ -extensions for all
p
prime numbers p. We will recall the definition of the cyclotomic ZZp -extension in XI §1.

Since L is totally imaginary and cyclic over k, we conclude that res kL x ∈

H 2 (L, I) is zero. 2
Again let K|k be a finite Galois extension with Galois group G and decom-
position groups Gp = G(Kp |kp ). For every prime p, we have by VII §1 the
invariant map
invKp |kp : H 2 (Gp , Kp× ) −→
∼ 1
[Kp :kp ]
ZZ/ZZ.
From the decomposition
H 2 (G, IK ) ∼ H 2 (Gp , Kp× ),
M
=
p
we obtain a canonical homomorphism

invK|k : H 2 (G, IK ) −→ [K:k]
1
ZZ/ZZ,
given by X
invK|k (c) = invKp |kp (cp ).
p
This invariant map is compatible with inf , res, cor as the following proposition
shows.
(8.1.10) Proposition. If L ⊇ K ⊇ k are finite separable extensions with L|k

Galois, then we have the commutative diagrams
invL|K
àåäãâá
H 2 (G(L|K), IL ) 1
ZZ/ZZ
[L:K]
res cor [K:k] incl
invL|k 1
H 2 (G(L|k), IL ) [L:k]
ZZ/ZZ .
Moreover, if K|k is Galois, then invL|k is an extension of invK|k .
Proof: The proposition is an immediate consequence of (8.1.3) and of the

analogous properties of the local inv’s as invariant maps of a class formation
(see (3.1.8)). If
c ∈ H 2 (G(K|k), IK ) ⊆ H 2 (G(L|k), IL ),
then
X X
invL|k (c) = invLp |kp (cp ) = invKp |kp (cp ) = invK|k (c).
p p
Letting P run through the primes of K and choosing always a prime P0 |P of

L, we obtain for c ∈ H 2 (G(L|k), IL )

X X k
invL|K (res c) = invLP0 |KP (reskK (c)P ) = invLP0 |KP (resKpP (cp ))
P P
X
= [KP : kp ] invLP0 |kp (cp )
P
XX
= [KP : kp ] invLP0 |kp (cp )
p P|p
X
= [K : k] invLP0 |kp (cp )
p
= [K : k] invL|k (c).
Finally, for c ∈ H 2 (G(L|K), IL ) we obtain
X XX K
invL|k (cor c) = invLP0 |kp (corK
k (c)p ) = invLP0 |kp (corkpP (cP ))
p p P|p
2
XX
= invLP0 |KP (cP ) = invL|K (c).
p P|p
By the compatibility of inv with the inflation, we also obtain invariant maps
for infinite Galois subextensions K|k of k̄|k,
invK|k : H 2 (G(K|k), IK ) −→ [K:k]
1
ZZ/ZZ,
by passing to the direct limit over the finite Galois subextensions Kα |k of K|k
and setting [
1 1
[K:k]
ZZ/ZZ = [Kα :k]
ZZ/ZZ.
α
In particular, we have an invariant map
invk : H 2 (k, I) −→ Q/ZZ
for the idèle group I of the separable closure k̄.
For every finite Galois extension K|k with Galois group G, local class field
theory provides us with the norm residue symbol
( , K|k) : Ik −→ Gab
given by
Y
(α, K|k) = (αp , Kp |kp ).
p
The product on the right is finite, i.e. well-defined, because (αp , Kp |kp ) = 1 for
all primes p such that αp ∈ Up and p is unramified in K|k. This norm residue
symbol is linked with the invariant map invK|k as follows. For every character
χ ∈ H 1 (G, Q/ZZ), the exact sequence
0 −→ ZZ −→ Q −→ Q/ZZ −→ 0

yields an element δχ ∈ H 2 (G, ZZ). Moreover, we have for every idèle α ∈ Ik =

H 0 (G, IK ), the cup-product α ∪ δχ ∈ H 2 (G(K|k), IK ).
(8.1.11) Proposition. For every χ ∈ H 1 (G, Q/ZZ) and α ∈ Ik , we have
χ((α, K|k)) = invK|k (α ∪ δχ).
Proof: The result follows from its local analogue (7.2.12). If χp is the
restriction of χ to Gp = G(Kp |kp ), then αp ∪ δχp ∈ H 2 (Gp , Kp× ) is obviously
the local component of α ∪ δχ, so that
X
invK|k (α ∪ δχ) = invKp |kp (αp ∪ δχp )
p
2
X
= χp ((αp , Kp |kp )) = χ((α, K|k)).
p
The next proposition is a first step towards theorem (8.1.22) below, which
claims that the pair (Gk , C) is a class formation.
(8.1.12) Proposition. For every finite Galois extension K|k
H 1 (G(K|k), CK ) = 0.
Proof: If K|k is cyclic, this is part of the class field axiom (8.1.1). If
[K : k] = pn is a prime power, then there exists a cyclic subextension L|k of
K|k of degree p, and we can proceed by induction on [K : k] using the exact
sequence
0 −→ H 1 (G(L|k), CL ) −→ H 1 (G(K|k), CK ) −→ H 1 (G(K|L), CK ).
If K|k is arbitrary, we consider the fixed fields Σp of the p-Sylow subgroups
of G(K|k). Then, by (1.6.10), the restriction map
Y
res : H 1 (G(K|k), CK ) ,→ H 1 (G(K|Σp ), CK )
p
is injective and we are done. 2
(8.1.13) Corollary. H 1 (k, C) = 0.
By (6.3.4), the group H 2 (k, k̄ × ) is canonically isomorphic to the Brauer

group Br(k) of central simple algebras over the global field k.

(8.1.14) Proposition. Let k be a global field. Then

(i) [
Br(k) = Br(K|k).
K|k
cyclic
(ii) Let K|k be an infinite Galois extension such that p∞ divides the local
degrees [Kp : kp ] for all finite primes p of K and assume in the number
field case that K is totally imaginary if p = 2. Then
Br(K)(p) = 0.
Remark: We could have sharpened assertion (i) in a similar manner to (8.1.9).
Proof: For every finite Galois extension K|k, (8.1.13) and H 1 (K, I) = 0 =
H 1 (K, k̄ × ) give the exact commutative diagram
0æçèéêëìíîïðñ H 2 (K, k̄ × ) H 2 (K, I)
0 H 2 (k, k̄ × ) H 2 (k, I)
0 H 2 (G(K|k), K × ) H 2 (G(K|k), IK )
0 0.
Considering this diagram for cyclic extensions K|k, we obtain assertion (i)
from (8.1.9). Since Br(K)(p) injects into H 2 (K, I)(p), the second statement
follows from (8.1.8). 2
(8.1.15) Proposition. Let K|k be a cyclic extension with Galois group G.

Then the sequence
invK|k
0 −→ H 2 (G, K ×ò ) −→ H 2 (G, IK ) 1
[K:k]
ZZ/ZZ −→ 0
is exact.
1
Proof: We start by showing that invK|k : H 2 (G, IK ) → [K:k] ZZ/ZZ is surjec-
tive. We first assume that [K : k] is a prime power, say pn . Let K 0 be the
unique extension of degree p inside K|k. If all primes of k would split in K 0 ,

then the norm map NK 0 |k : IK 0 → Ik would be surjective, and so would be the

norm map NK 0 |k : CK 0 → Ck . Since this contradicts #Ĥ 0 (G(K 0 |k), CK 0 ) = p,
we find a prime p of k which is inert in K 0 , and hence also in K. Therefore
G = Gp = G(Kp |kp ), and we know that
invKp |kp : H 2 (Gp , Kp× ) −→ 1

[Kp :kp ]
ZZ/ZZ
is bijective. We conclude that
invK|k : H 2 (G, IK ) → [K:k]

1
ZZ/ZZ
is surjective if [K : k] is a prime power. The general case now easily follows

from (8.1.10).
Since G is cyclic, H 3 (G, K × ) = H 1 (G, K × ) = 0. Moreover, H 1 (G, CK ) = 0,
and so the exact sequence
0 → K × → IK → CK → 0
yields the exact sequence
0 −→ H 2 (G, K × ) −→ H 2 (G, IK ) −→ H 2 (G, CK ) −→ 0;
in particular, the map H 2 (G, K × ) −→ H 2 (G, IK ) is injective and, by (8.1.1),

its cokernel has order
#H 2 (G, CK ) = #Ĥ 0 (G, CK ) = [K : k].
It remains to show that inv is trivial on the image of H 2 (G, K × ). We prove

this without the assumption of K|k being cyclic. So let α ∈ Br(k) be arbitrary.
If k is a function field, then, by (8.1.14) and the following remark, we may
assume that α ∈ Br(K|k), where K = k(ζn ) for some n prime to char(k). If
k is number field, let K be a finite extension of k which is Galois over Q and
such that α ∈ Br(K|k). Then
invk (α) = invK|k (α) = invK|Q (corkQ (α)).
Hence we may assume that k = Q and, by the same argument as in the function
field case, α ∈ Br(K|Q) for a cyclic subextension K of Q(ζn )|Q for some n.
In both cases, let G = G(K|k) and let χ be a generator of H 1 (G, Q/ZZ). Then
δχ is a generator of H 2 (G, ZZ), and the cup-product
δχ∪ : Ĥ 0 (G, K × ) −→ H 2 (G, K × )

is the periodicity isomorphism (see (1.7.1)). Therefore every element of

H 2 (G, K × ) is of the form ā ∪ δχ with a ∈ k × . By (8.1.11), we have
invK|k (ā ∪ δχ) = χ((a, K|k)).
It therefore remains to show that (a, K|k) = 1 for a ∈ k × . Hence it suffices to

show that (a, k(ζn )|k)ζn = ζn , where k is a function field and (n, char(k)) = 1
or k = Q and n arbitrary. Let k be a function field. Then, for any place p of k,
we have
vp (a)
(a, kp )ζn = ζn#k(p) .
The ‘product formula’ yields
#k(p)vp (a) = 1,
Y
hence
Q
p
#k(p)vp (a)
(a, k)ζn = ζn = ζn .
The argument in the case k = Q is similar, but the computation of the local
norm residue symbols at the primes p | n is much more involved, see, e.g.,
[160], chap.VI, (5.3) and chap.V, (2.4). This proves the proposition. 2
Using (8.1.14)(i) and (8.1.9), we obtain from (8.1.15) the following
(8.1.16) Corollary. We have an exact sequence
0 −→ H 2 (k, k̄ ×ó ) −→ H 2 (k, I)
invk
Q/ZZ −→ 0 .
By (8.1.7), we have
H 2 (k, I) ∼ H 2 (kp , k̄p× ),
M
=
p
and the summands are the Brauer groups Br(kp ) of the local fields kp . Hence
corollary (8.1.16) gives us the famous Hasse principle for central simple
algebras:
(8.1.17) Theorem. We have an exact sequence
0÷ôõö
M invk
Br(k) Br(kp ) Q/ZZ 0,
p
where invk is the sum of the local invariant maps invkp : Br(kp ) → Q/ZZ.

(8.1.18) Corollary. Let p be a prime number and let K|k be an infinite sepa-
rable extension of the global field k which we assume to be totally imaginary
if k is a number field and p = 2. If p∞ divides the local degrees [Kp : kp ] for
all nonarchimedean primes p of k, then
cdp GK ≤ 1.
Proof: If p = char(K), this is always true by (6.5.10). Otherwise we have

to show that Br(L)(p) = 0 for every finite separable extension L|K (see
(6.5.11)). Passing to inductive limits, (8.1.17) implies that Br(L)(p) injects
into p Br(Lp )(p). Furthermore, Br(Lp )(p) ∼ = H 2 (Lp , µp∞ ) = 0 by (7.1.8)(i)
Q
or by assumption if p = 2 and p|∞. 2
We now carry over the results on the cohomology of the idèles to the coho-
mology of the idèle class group. Recall the notation
H i (K|k) = H i (G(K|k), CK ).
(8.1.19) Lemma. For every finite Galois extension K|k, we have

#H 2 (K|k) | [K : k].
Proof: If K|k is cyclic, the assertion follows from the class field axiom
(8.1.1). If K|k is a p-extension and L|k a cyclic subextension of degree p, then
the exact sequence (observe that H 1 (K|L) = 0)
0 −→ H 2 (L|k) −→ H 2 (K|k) −→ H 2 (K|L)
shows, using induction on [K : k],
#H 2 (K|k) | #H 2 (K|L) · #H 2 (L|k) | [K : L] · [L : k] = [K : k].
In the general case, let Σp be a p-Sylow field of K|k. Since the restriction map
M
res : H 2 (K|k) ,→ H 2 (K|Σp )
p
is injective, we obtain
#H 2 (K|k) | #H 2 (K|Σp ) | 2
Y Y
[K : Σp ] = [K : k].
p p
If N ⊇ K ⊇ k are two finite Galois extensions in k̄|k, then the sequence

inf res
0 −→ H 2 (K|k) −→ H 2 (N |k) −→ H 2 (N |K)
is exact by (1.6.7), since H 1 (N |K) = 0. As for the idèles, we identify H 2 (K|k)
with its image in H 2 (k, C) so that

[
H 2 (k, C) = H 2 (K|k),
K|k
where K|k varies over the finite Galois subextensions of k̄|k.

We would like to deduce from the idèlic invariant maps
invK|k : H 2 (G(K|k), IK ) → [K:k]
1
ZZ/ZZ
invariant maps for the groups H 2 (K|k). This is not possible in a direct way,
since the map H 2 (G(K|k), IK ) → H 2 (G(K|k), CK ) is in general not surjective.
But it becomes possible if we first pass to the direct limit.
As before let k̃|k be the cyclotomic ẐZ-extension in the number field case and
the ẐZ-extension obtained by passing to the algebraic closure of the constant
field in the function field case. Since scd ẐZ = 2, we have H 3 (k̃|k, k̃ × ) = 0, and
therefore the exact sequence
0 −→ k̃ × −→ Ik̃ −→ Ck̃ −→ 0
induces the exact cohomology sequence
0 −→ H 2 (k̃|k, k̃ × ) −→ H 2 (k̃|k, Ik̃ ) −→ H 2 (k̃|k, Ck̃ ) −→ 0 .
Passing to the limit over all finite subextensions of k̃|k, (8.1.15) induces the
exact sequence
0 −→ H 2 (k̃|k, k̃ × ) −→ H 2 (k̃|k, Ik̃ ) −→ Q/ZZ −→ 0 .

Therefore we obtain an isomorphism
invk̃|k : H 2 (k̃|k, Ck̃ ) −→

∼ Q/Z
Z.
For a finite separable extension K|k we have a commutative diagram by

(8.1.10)
invK̃|K
øùüûú C )
H 2 (K̃|K, Q/ZZ
K̃
res
H 2 (K̃|k, CK̃ ) [K:k]
inf
invk̃|k
H 2 (k̃|k, Ck̃ ) Q/ZZ .
(8.1.20) Proposition. The sequence
0 −→ H 2 (k, k̄ × ) −→ H 2 (k, I) −→ H 2 (k, C) −→ 0
is exact.

Proof: Let K|k be an arbitrary finite Galois extension of degree n = [K : k]

and let kn be the unique extension of k in k̃ of degree n. We claim that
H 2 (K|k) = H 2 (kn |k) ,
where we identify H 2 (K|k) and H 2 (kn |k) with their images in H 2 (k, C) under
the inflation maps. In fact, (8.1.19) and the class field axiom (8.1.1) imply that

#H 2 (K|k) [K : k] = [kn : k] = #H 2 (kn |k),

and it therefore suffices to show the inclusion "⊇". But this follows from the
0ýþÿ H 2 (K|k) H 2 (k, C) res
H 2 (K, C)
res
H 2 (k̃|k, Ck̃ ) H 2 (K̃|K, CK̃ )
invk̃|k invK̃|K
[K:k]
Q/ZZ Q/ZZ,
in which invk̃|k and invK̃|K are isomorphisms as shown above. Since H 2 (kn |k) ⊆
H 2 (k̃|k) has order n = [K : k], it is mapped by the middle arrow res, and thus
by the upper arrow res , to zero, so that
H 2 (kn |k) ⊆ H 2 (K|k),
which shows the claim. Hence we obtain
[ [
H 2 (k, C) = H 2 (K|k) = H 2 (kn |k) = H 2 (k̃|k, Ck̃ ).
K n
In particular, this implies

[
H 2 (k, C) = H 2 (K|k) .
K|k
cyclic
The same assertion holds if we replace C by I or k̄ × , from (8.1.9) or (8.1.14)(i),

respectively. Now for K|k cyclic, we have the exact sequence
0 −→ H 2 (K|k, K × ) −→ H 2 (K|k, IK ) −→ H 2 (K|k, CK ) −→ 0,
because H 1 (K|k, CK ) = 0 and H 3 (K|k, K × ) ∼
= H 1 (K|k, K × ) = 0. From this
and from the above observation now follows the statement of the proposition.
2
(8.1.21) Corollary. We have a canonical isomorphism
invk : H 2 (k, C) −→
∼
Q/ZZ.

From this corollary we obtain for every finite Galois extension K|k a canon-
ical invariant map
invK|k : H 2 (K|k) −→
∼ 1
[K:k]
ZZ/ZZ
using the commutative diagram
0
H 2 (K|k) H 2 (k, C) res
H 2 (K, C) 0
invK|k invk invK
1 [K:k]
0 [K:k]
ZZ/ZZ Q/ZZ Q/ZZ 0.
The results above and the compatibility of inv with inf and res , which follows
from (8.1.10), therefore imply the
(8.1.22) Theorem. The G-module C is a formation module with respect to

the invariant maps invK|k (see (3.1.8)).
We now obtain the main theorem of global class field theory from (3.1.6):
the “global reciprocity law”.
(8.1.23) Theorem. Let K|k be a finite Galois extension of global fields with
Galois group G(K|k). Then there is a canonical isomorphism
Ck /NK|k CK ∼
= G(K|k)ab .
By the results of III §1 (see the remark 5 on p.157), we obtain a reciprocity

homomorphism
rec : Ck −→ Gab
k ,
which has a dense image, and whose kernel is the group of universal norms,
i.e. the intersection
\
NK|k CK = NGk C .
K
As before we also call rec the norm residue symbol and we write it in the
form rec(α) = (α, k).
Essential is the following theorem, which is called the existence theorem for
global class field theory. For a proof, see [6], chap. 8, th. 1. For number fields,
see also [123], chap.XI §2 th. 1 or [160], chap.VI th. 6.1.

(8.1.24) Theorem. The norm groups NK|k CK , where K|k is a finite separable
extension, are exactly the open subgroups of finite index in Ck .
The idèle group Ik = p kp× is a locally compact topological group, k × is a

Q
discrete, closed subgroup (see [160], chap.VI, (1.5)), and so Ck is also a locally
compact topological group. We have a canonical homomorphism, called the
“absolute value”,
| | : Ik −→ IR×
Y
+ , |α| = |αp |p ,
p
which is trivial on k × , and hence induces a homomorphism

| | : Ck −→ IR×
+.
The kernel Ck0 = {x ∈ Ck | |x| = 1} is a compact group (cf. [22], chap.II, §16,
Theorem, or [160], chap.VI, (1.6) for the number field case).
If k is a number field, then the homomorphism | | is surjective: Since IR× +

has no nontrivial finite quotient, Ck and Ck0 have the same image under the
reciprocity map, which is all of Gab 0
k since Ck is compact and (Ck , k) is dense
in Gabk . This yields the
(8.1.25) Proposition. In the number field case we have an exact sequence of

topological groups
0
( ,k)
NGk C Ck Gab
k 0.
If k is a function field, with constant field κ of cardinality q, we have the

following situation. The group Ik , and hence Ck , is totally disconnected by
(1.1.9)(iii); in particular, Ck0 is an abelian profinite group. For α ∈ Ck we have
|α| = q − deg(α) , where
deg(α mod k × ) =
X
deg : Ck → ZZ, vp (αp ) · [κ(p) : κ] ,
p
is the degree map. Let π : Gab ∼ ∼ Z be the canonical

k G(k κ̄|k) = Gκ = Ẑ
projection. Then, for all α ∈ Ck , we have π((α, k)) = Frobdeg(α) , where
Frob denotes the Frobenius automorphism x 7→ xq on κ̄. Therefore deg is
surjective∗) and we obtain an exact sequence of topological groups
deg
0 −→ Ck0 −→ Ck −→ ZZ −→ 0.
Since Ck0 is profinite the intersection of all open subgroups of finite index in
Ck is zero. The degree map extends to a map from C k := lim Ck /NK|k CK
←− K|k
∗) This can also be seen directly: one can construct an injection coker(deg) ,→ Br(κ) = 0.

§2. The Connected Component of Ck 443
onto ẐZ. We therefore obtain a commutative exact diagram

0
deg
Ck0 Ck ZZ 0
deg
0 Ck0 Ck ẐZ 0,
which yields, using (8.1.23), the
(8.1.26) Proposition. In the function field case we have an exact sequence of

topological groups
0"#!
( ,k)
Ck Gab
k ẐZ/ZZ 0.
§2. The Connected Component of Ck
We now study the topological properties of the idèle class group Ck and,
in particular, the connected component Dk of 1 in Ck . The results are due to
J. TATE (see [6], chap.9) and are quoted in [160], chap.VI, §1, ex.1–10.
We denote the connected component (of 1) of the idèle class group Ck by
Dk . If k is a function field, then Ik , and hence Ck , is totally disconnected, so
that Dk = 1.
Therefore we assume for the rest of this section that k is a number field.
We introduce the following notation:
S∞ = S∞ (k) the set of archimedean primes of k ,
SIR = SIR (k) the set of real primes of k ,
SC = SC (k) the set of complex primes of k ,
r1 = r1 (k) the number of real primes of k ,
r2 = r2 (k) the number of complex primes of k ,
r = r(k) = r1 (k) + r2 (k) .
The continuous surjection
| | : Ck −→ IR×
+
has a continuous section s : IR×
+ → Ck : for any infinite prime p, the restriction
×
of | | to the subgroup IR+ of kp× ⊆ Ck is an isomorphism, and its inverse gives
a section of the homomorphism | |. Since IR× + is connected, the image of s is
contained in Dk and we obtain compatible isomorphisms of topological groups
Ck ∼= C 0 × IR× , Dk ∼k = D0 × IR× ,
+ k +

where Dk0 = Ck0 ∩ Dk is the connected component of Ck0 . The group Dk has
the following characterizations.
(8.2.1) Theorem.
Y
(i) Dk is the closure of the image of Ũ = Up1 under the projection
p|∞
Ik → Ck , where Up1 = IR× 1 ×
+ if p is real and Up = C if p is complex.
(ii) Dk is the intersection of all open subgroups of Ck of finite index.

\
(iii) Dk is the group of universal norms Nk̄|k C = NK|k CK and Dk0 is the
K|k
\
0 0
group of the universal norms Nk̄|k C = NK|k CK .
K|k
(iv) For any finite extension K|k the norm maps NK|k : DK → Dk and
0
NK|k : DK → Dk0 are surjective.
(v) Dk is the group of all divisible elements of Ck , i.e. elements x which are
n-th powers, x = y n , for every n ∈ IN.
(vi) D is divisible, i.e. it is the maximal divisible subgroup in C .∗)
k k
Remark: All open subgroups in Ck have finite index. We will prove this fact
in the context of restricted ramification in (8.3.14).
Proof: (i) Since Ũ is the connected component of 1 in I, the assertion follows

from (1.1.9)(iii).
(ii) Since Ck /Dk is totally disconnected, the intersection of all its open sub-
groups Ūi is trivial by (1.1.9)(i). But Ck /Dk = Ck0 /Dk0 is compact, i.e. the Ūi
are of finite index. The pre-images Ui in Ck are open subgroups of finite index
and Dk is their intersection. On the other hand, any open subgroup contains
Dk . This proves (ii).
(iii) By the existence theorem (8.1.24), the open subgroups of finite index in Ck
are just the norm groups NK|k CK of the finite Galois extensions K|k, so that
(ii) implies that Dk = Nk̄|k C. As IR×+ is cohomologically trivial, we have an
0 0 ∼
isomorphism Ck /NK|k CK = Ck /NK|k CK for every finite Galois extension
K|k. This implies
Nk̄|k C 0 = Ck0 ∩ Nk̄|k C = Ck0 ∩ Dk = Dk0 .
∗) If A is an abelian group, then the subgroup of divisible elements contains the maximal
divisible subgroup of A, but in general is not divisible itself.

(iv) By assertion (iii), Dk0 is the group of universal norms of the level-compact
module C 0 , hence NK|k DK 0
= Dk0 for every finite Galois extension K|k. The
analogous assertion for D follows from the commutative diagram
0-$%&'()*+,. 0
DK DK IR×
+ 0
NK|k NK|k NK|k
0 Dk0 Dk IR×
+ 0
noting that the norm map NK|k : IR× ×
+ → IR+ is just the map x 7→ x
[K:k]
and is
therefore surjective.
(v) Ck /Dk = Ck0 /Dk0 is a compact, totally disconnected Hausdorff group, i.e. a
profinite group. If x ∈ Ck is divisible, then its image in Ck /Dk is contained in
every open subgroup, i.e. the image is 1 and hence x ∈ Dk . Hence (v) follows
from (vi).
(vi) Since IR× 0
+ is divisible, it suffices to show that Dk is divisible. Let ` be a
prime number. If the finite extension K of k contains the `-th roots of unity,
then for sufficiently large S containing all primes above ` and all archimedean
primes,
q the group (CK )` · ŪKS is the norm group of the Kummer extension
K( ` OS× )|K. Here ŪKS = UKS K × /K × and UKS = p∈/S Up × p∈S 1 (see [123],
Q Q
chap.XI §2 th. 1 or the proof of [160], chap.VI th. 6.1). Hence, for any such S,
DK ⊆ (CK )` Ū S .
As the absolute value | | is trivial on Ū S , we obtain
0 ` S 0 `
\
0
DK ⊆ (CK ) Ū = (CK ).
S
Using (iv) this implies
0 `
(∗) Dk0 = NK|k DK
0 ⊆ (NK|k CK ).
For each a ∈ Dk0 , let a1/` denote the set of all elements of Ck0 whose `-th power
0
is a. From (∗) we see that the sets XK = (NK|k CK ) ∩ (a1/` ) are nonempty.
`
The kernel of the map Ck0 → Ck0 , x 7→ x` , is compact and NK|k CK 0
is closed,
T
and so XK is compact. Therefore the intersection K|k XK is not empty. By
(iii), an element of this intersection is an element of Dk0 whose `-th power is a.
This proves (vi). 2
By (8.2.1), we obtain from (8.1.25) the
(8.2.2) Corollary. We have an exact sequence of topological groups
02/01
( ,k)
Dk Ck Gab
k 0.

In order to give an explicit description of the connected component Dk ∼

=
Dk0 × IR×+ , we have first to consider the group Z
Z /ZZ , where
Y
ZZ = IR × ZZp = IR × ẐZ
p
is the group of integral elements in the adèle ring AQ of Q.
(8.2.3) Proposition. The topological group ZZ/ZZ is compact, connected and

uniquely divisible. It is called the solenoid.∗)
Proof: The canonical projection ZZ = ẐZ × IR ẐZ induces an exact sequence

0 −→ IR −→ ZZ/ZZ −→ ẐZ/ZZ −→ 0 .
This shows that ZZ/ZZ is uniquely divisible, since IR and ẐZ/ZZ have this property.
Furthermore, IR is dense in ZZ/ZZ: let λ = (z, x) mod ZZ be an arbitrary element
in (ẐZ × IR)/ZZ, let m ∈ ZZ be given (describing a neighbourhood of ZZ) and
let a ∈ ZZ such that a ≡ z mod m. Then λ = (z − a, x − a) mod ZZ and
(z − a, x − a) is contained in the set (mẐZ, 0) + (0, x − a) ⊆ mẐZ × IR, which is
mapped into a neighbourhood of x−a ∈ IR ⊆ ZZ/ZZ. Now, since IR is connected
and the closure of a connected set is connected, we conclude that the solenoid
is connected. (Another possible method to show the connectedness of ZZ/ZZ
is to use the exact sequence at the beginning of the proof and to show the
connectedness of ẐZ/ZZ .)
Since ZZ is closed in ZZ, the quotient ZZ/ZZ is a Hausdorff topological group.
Finally, the compact set ẐZ×[0, 1] ⊆ ZZ is mapped under the canonical projection
ZZ ZZ/ZZ onto ZZ/ZZ. Hence the solenoid is compact. 2
We consider the group

Uk = Ū × Ũ ,
where
Y Y
Ū = Up and Ũ = Up1 ,
p-∞ p|∞
and we write every idèle α ∈ Uk as a product α = ᾱα̃ with ᾱ ∈ Ū , α̃ ∈ Ũ .

The abelian profinite group Ū is a (multiplicative) ẐZ-module in a natural
way: The exponentiation
IN × Ū → Ū , (n, α) 7→ αn ,
extends continuously to an exponentiation
ẐZ × Ū −→ Ū , λ 7−→ αλ .
∗) For another description of the solenoid, see ex.1.

(8.2.4) Lemma. If ε1 , . . . , εr−1 are ZZ-independent units of k, r = r1 + r2 , then

Q
the corresponding idèles ε̄1 , . . . , ε̄r−1 ∈ Ū = p-∞ Up are ẐZ-independent.
Proof: By Dirichlet’s unit theorem, the group generated by ε1 , . . . , εr−1 (with

ordinary integers as exponents) has finite index d in Ok× . Consider a relation
z
(∗) ε̄z11 · · · ε̄r−1
r−1
= 1 , zi ∈ ẐZ.
We have to show z1 = · · · = zr−1 = 0. Let n ∈ IN. For each zi we find an
νr−1
integer νi ∈ ZZ such that νi ≡ zi mod 2dn. Then ε = εν11 · · · εr−1 is an element
of k and we may write
ν ν
ε = ε̄1ν1 · · · ε̄r−1
r−1 ν1 r−1
ε̃1 · · · ε̃r−1 .
Dividing the right-hand side by the left-hand side of (∗), we obtain
ν −zr−1 ν1 ν
ε = ε̄ν11 −z1 · · · ε̄r−1
r−1
ε̃1 r−1
· · · ε̃r−1 .
At each finite prime each idèle ε̃νi i has component 1; the remaining factors have
ẐZ-exponents divisible by 2dn. This means that ε is a 2dn-th power in kp× for
each finite prime p. But from this it follows that ε is a dn-th power in k × , i.e.
ε = η dn , η ∈ k × , see (9.1.11)(ii). This η must be a unit and consequently η d is
µr−1
contained in the subgroup generated by the ε1 , . . . , εr−1 , i.e. η d = εµ1 1 · · · εr−1 ,
µi ∈ ZZ. We now get
ν µ n
ε = εν11 · · · εr−1
r−1
= εµ1 1 n · · · εr−1
r−1
.
Since the εi are ZZ-independent, we must have νi = µi n. Because νi ≡ zi
mod 2dn, this shows that each zi is divisible by n, i.e. zi ∈ nẐZ. But n was
arbitrary, hence zi = 0. 2
We shall now determine the structure of Dk0 . It contains a torus Tk which is

defined as follows: for each complex prime p we have the embedding kp× ,→ Ck
and we obtain an embedding
kp× ,→ Ck .
Y
p∈SC
Each factor kp× = C× contains the unit circle Sp1 = {z ∈ kp× | |z| = 1} and we
put Y
Tk = Sp1 .
p∈SC
It is a compact connected subgroup of Dk0 , also described by the topological

isomorphism
Y
exp : IR/ZZ −→ Tk , θ = (θp ) 7−→ (e2πiθp )p∈SC .
p∈SC
We may regard Tk as a subgroup of the idèle group Ũ ⊆ Uk .

For p ∈ S∞ , we have an exponentiation IR × Up → Up , (x, αp ) 7→ αpx =

e , where log is the natural logarithm of IR×
x log αp
+ if p is real, and is any fixed
×
branch of the logarithm of C if p is complex. For fixed α ∈ Uk , the map
IR → Ũ , x 7→ α̃x , is continuous, and so is the map
Z̄Z = ẐZ × IR −→ Uk , λ = (z, x) 7−→ ᾱz α̃x .
Let O+ (k) be the subgroup of Ok× of totally positive units of k, i.e. ε ∈ O+ (k)
if and only if εp > 0 for every real prime p. Since O+ (k) has finite index in Ok× ,
Dirichlet’s unit theorem implies that it is a finitely generated group of rank
rk(O+ (k)) = #S∞ (k) − 1 = r − 1,
and we choose a fixed r − 1-tuple
ε = (ε1 , . . . , εr−1 )
of ZZ-independent units. In addition to Tk , we consider the subgroup
λ
I(ε) = {ελ1 1 · · · εr−1
r−1
| λ1 , . . . , λr−1 ∈ ZZ}
of Uk together with the continuous surjective homomorphism
λ
expε : ZZr−1 −→ I(ε), (λ1 , . . . , λr−1 ) 7−→ ελ1 1 · · · εr−1
r−1
.
Note that Tk is a subgroup of Uk0 = {x ∈ Uk | |x| = 1} and the same holds
for I(ε), since εzi i ∈ Up for every finite prime p, so that |ελi i | = |ε̃xi i | = |ε̃i |xi =
|εi |xi = 1 (where λi = (zi , xi ) ∈ ẐZ × IR = ZZ). We get a continuous homomor-
phism
(∗) exp : (IR/ZZ)r2 × ZZr−1 −→ Uk0
with image Tk I(ε).
(8.2.5) Theorem (TATE). The homomorphism (∗) induces a topological iso-

morphism
exp : (IR/ZZ)r2 × (ZZ/ZZ)r−1 −→
∼ D0 .
k
For the connected component Dk of Ck , this yields a topological isomorphism
Dk ∼
= (IR/ZZ)r2 × (ZZ/ZZ)r−1 × IR .
Proof: The second statement follows from the first since Dk ∼ = Dk0 × IR×+ . In
+ 0
order to prove the first, observe that O (k) is a subgroup of Uk , and we obtain
a homomorphism
ZZr−1 −→ Uk0 /O+ (k) .
Since ελi i ∈ O+ (k) when λi = (n, n) ∈ ZZ ⊆ ZZ, this homomorphism factors
through (ZZ/ZZ)r−1 . We obtain a continuous homomorphism
expε : (ZZ/ZZ)r−1 −→ Ck0 ,

whose image is contained in the connected component Dk0 since ZZ/ZZ is con-
nected by (8.2.3).
λ
I. Injectivity of exp: Let α ∈ Tk and β = expε (λ) = ελ1 1 · · · εr−1 r−1
, where λ =
(λ1 , . . . , λr−1 ) ∈ ZZr−1 , λi = (zi , xi ). For the injectivity of exp we have to show:
suppose that αβ ∈ Uk0 is a principal idèle,
αβ = a ∈ k × ,
then λ ∈ ZZr−1 . Looking only at the components at the finite primes, we have
z
(1) ε̄z11 · · · ε̄r−1
r−1
= a.
The group generated by ε1 , . . . , εr−1 has finite index in O+ (k) so that ad =
µr−1
εµ1 1 · · · εr−1 with µi ∈ ZZ. Raising (1) into the d-th power, we obtain
dz −µr−1
ε̄1dz1 −µ1 · · · ε̄r−1
r−1
=1
and by lemma (8.2.4), dzi − µi = 0 in ẐZ. From this it follows that the zi are
contained in ZZ. Now
zr−1
ε = εz11 · · · εr−1
is an element in k which coincides with a in kp for p - ∞, hence ε = a. Looking
at the components at the infinite primes, we obtain from αβ = ε
x z
α̃ε̃x1 1 · · · ε̃r−1
r−1
= εz11 · · · εr−1
r−1
.
Taking absolute values | |p for every p ∈ S∞ , we get
|ε1 |px1 −z1 · · · |εr−1 |xp r−1 −zr−1 = 1.
Taking the logarithms of these equations, we conclude that xi −zi = 0, since the
vectors (log |εi |p )p∈S∞ , i = 1, . . . , r − 1, are linearly independent by Dirichlet’s
unit theorem. This proves λi = (zi , xi ) ∈ ZZ, i.e. λ ∈ ZZr−1 , and thus the
injectivity, as desired.
II. Surjectivity of exp: Let D0 ⊆ Dk0 be the image of exp and let a ∈ Dk0 . For
the proof that a ∈ D0 we use the fact that a is divisible by (8.2.1)(v). So we
may write a = b2hm where h is the class number of k and m a highly divisible
integer. The ideal class group Clk is the quotient Ik /Vk k × with
kp× ,
Y Y
Vk = Up ×
p-∞ p|∞
so the class bh can be represented by an idèle in Vk , hence b2h by an idèle

β ∈ Uk . Since a ∈ Dk0 , we have β ∈ Uk0 . The element a is therefore represented
by the idèle β m = β̄ m β̃ m . Choosing a suitable highly divisible m, the idèle β̄ m
is contained in a neighbourhood of 1 which is as small as we like. We shall
prove that the idèle class c of β̃ m belongs to D0 . From this it follows that a is in
the closure of D0 . But D0 is compact, hence closed and consequently a ∈ D0 .
Set α̃ = β̃ m ∈ Uk0 . Let pν , ν = 1, . . . , r, range over the archimedean
primes. From the independence of the units ε1 , . . . , εr−1 it follows that the

vectors ai = (log |εi |p1 , . . . , log |εi |pr−1 ), i = 1, . . . , r − 1, in IRr−1 are linearly
independent. Setting a = (log |α̃|p1 , . . . , log |α̃|pr−1 ), we find x1 , . . . , xr−1 ∈ IR
such that
x1 a1 + · · · + xr−1 ar−1 = a,
i.e.
|α̃|pν = |ε̃1 |xpν1 · · · |ε̃r−1 |xpνr−1 for ν = 1, . . . , r − 1.
Since |α̃| = 1, we also have
|α̃|pr = |ε̃1 |xpr1 · · · |ε̃r−1 |xprr−1 .
Let λi = (0, xi ) ∈ ZZ. Since α̃ is totally positive, we have
x
α̃p = (ε̃x1 1 · · · ε̃r−1
r−1
)p
for every real prime p. For complex p, both sides differ by an element of
value 1. Hence we can write
x λ
α̃ = ε̃x1 1 · · · ε̃r−1
r−1
γ̃ = ελ1 1 · · · εr−1
r−1
γ̃
with γ̃ ∈ Tk . This proves that the class c of α̃ is contained in the image D0 of
exp as contended. 2
It is not unimportant to know also the cohomology of the connected compo-

nent DK of a finite Galois extension K|k. It is given by the
(8.2.6) Corollary. Let K|k be a finite Galois extension with Galois group
G = G(K|k) and let m be the number of real primes of k that become complex
in K. Then
(ZZ/2ZZ)m if i is even,
(
i ∼
Ĥ (G, DK ) =
0 if i is odd,
and H 0 (G, DK ) ∼
= Dk ⊕ (ZZ/2ZZ)m .
× Q
Proof: We may consider the product P∈SC (K) KP as a G-submodule of the
idèle group IK as well as of DK ⊆ CK . In this product we have the canonical
G-submodule Y
1
TK = SP ,
P∈SC (K)
×
where 1
SPis the unit circle in = C× . By the above theorem and by (8.2.3),
KP
the quotient
DK 0
/TK ∼
= (ZZ/ZZ)r(K)−1
is uniquely divisible, hence cohomologically trivial. Therefore the exact co-
homology sequence yields
Ĥ i (G, DK ) = Ĥ i (G, DK
0
) = Ĥ i (G, TK ).

Let M be the set of primes p of k lying under the complex primes of K. For
every p ∈ M we choose a fixed prime P0 |p of K and consider the decom-
position group GP0 . We regard the group IR/ZZ as a GP0 -module by letting
σ ∈ GP0 act by σθ = θ if σ = 1, and σθ = −θ if σ =/ 1. Then
×
IR/ZZ −→ KP 0
, θ 7−→ e2πiθ ,
1
is an injective GP0 -homomorphism with image SP 0
, and we obtain an injective
G-homomorphism
GP × GP Y
×
IndG 0 (IR/ZZ) −→ IndG 0 (KP0
)= KP = IK (p)
P|p
and hence an injective G-homomorphism
M GP Y
IndG 0 (IR/ZZ) −→ IK (p)
p∈M p∈M
with image TK . Shapiro’s lemma now yields
GP
Ĥ i (G, IndG 0 (IR/ZZ)) ∼
M M
Ĥ i (G, TK ) = = Ĥ i (GP0 , IR/ZZ)
p∈M p∈M
M
= Ĥ i (GP0 , IR/ZZ) ,
p∈N
where N is the set of real primes of k becoming complex in K. Let p ∈ N;
then GP0 is generated by an element σ of order 2. If i is even, then
Ĥ i (GP0 , IR/ZZ) ∼
= Ĥ 0 (GP0 , IR/ZZ) = (IR/ZZ)GP0 /NGP0 (IR/ZZ).
The fixed module consists of the elements θ ∈ IR/ZZ with σθ = −θ = θ, i.e. is
1
2
ZZ/ZZ, and the norm group consists of the elements NGP0 θ = θ+σθ = θ−θ = 0.
This proves Ĥ i (GP0 , IR/ZZ) = ZZ/2ZZ.
If i is odd, then
Ĥ i (GP0 , IR/ZZ) ∼
= Ĥ −1 (GP0 , IR/ZZ) = NGP (IR/ZZ)/(σ − 1)(IR/ZZ).
0
For θ ∈ IR/ZZ, we have (σ − 1)θ = σθ − θ = −2θ, so that (σ − 1)(IR/ZZ) =

2(IR/ZZ) = IR/ZZ, and thus Ĥ i (GP0 , IR/ZZ) = 0.
Finally, note that NK|k DK = Dk by (8.2.1)(iv) and that Dk is divisible
by (8.2.1)(vi). Hence the calculation of H 0 (G, DK ) follows from that of
Ĥ 0 (G, DK ). 2
Exercise 1. Show that the solenoid ZZ/ZZ is the Pontryagin dual Homcts (Q, IR/ZZ) of Q. It may
n
also be identified with the projective limit lim S 1 over the maps S 1 → S 1 , z 7→ z n (n ∈ IN),
←− n
of the unit circle S 1 = {z ∈ C | |z| = 1}.
Exercise 2. The Pontryagin dual Hom(Dk , IR/ZZ) of the connected component Dk of Ck is

topologically isomorphic to ZZr2 × Qr−1 × IR, where r2 is the number of complex primes and
r is the number of all infinite primes.

§3. Restricted Ramification
In the foreground of the Galois cohomological considerations in the previous

chapters has been the absolute Galois group Gk of the fields. In the case of a
global field k, cohomology theory gives much more subtle and deeper lying
arithmetic laws if we study not just the absolute Galois groups, but also Galois
groups “with restricted ramification”. This theory is of great importance, so
that we pay particular attention to its development.
Let k be a global field, which we regard as a fixed ground field. Let S be a
nonempty set of primes of k containing the set S∞ of infinite (i.e. archimedean)
primes if k is a number field. In some of the following formulas for a global
field k the expressions S∞ , SIR and SC will occur. If k is a function field, then
these terms should be redundant.
In place of the separable closure k̄|k, we now consider the maximal subex-
tension kS |k which is unramified outside S. We denote its Galois group by
GS = G(kS |k).
If K|k is any finite subextension of kS |k, we set
GS (K) = G(kS |K).
We also let the symbol S stand for the set of primes of K which lie above the
primes in S. Thus P ∈ S means that P is a prime of K lying above a prime
p ∈ S of k. The ring of S-integers of K is defined by
OK,S = {a ∈ K | vP (a) ≥ 0 for all P ∈/ S}.
×
Its group of units OK,S and its ideal class group ClS (K) play a particularly
important role. The last group is the quotient of the usual ideal class group
ClK of K by the subgroup generated by the classes of all prime ideals in S. It
is finite (in the function field case note that S is nonempty) and is called the
S-ideal class group. The ring of integers of kS is denoted by
[
OS = OK,S ,
K|k
where the union is taken over all finite extensions K of k inside kS . Finally,
we set
×
IN(S) = {n ∈ IN | n ∈ Ok,S }.
If k is a number field, then these are the natural numbers such that vp (n) = 0 for
all p ∈/ S. If k is a function field, then they are the numbers prime to char(k).
The following proposition is a special case of (6.1.10).

§3. Restricted Ramification 453
(8.3.1) Proposition. Let K|k be a Galois extension with K ⊆ kS . Then

i
H (K|k, OK,S ) = 0 for all i > 0.
(8.3.2) Corollary. If p = char(k) > 0 and L|k is a p-closed extension inside kS ,

then
 Ok,S /℘Ok,S for i = 1,

i
H (L|k, ZZ/pZZ) =

0 for i > 1,
p
where ℘(x) = x − x.
Proof: Let K|k be a finite subextension of L|k and let a ∈ OK,S . If a is not of
the form αp − α for some α ∈ OK,S , consider the cyclic p-extension K(α)|K
given by f (α) = 0, f = X p − X − a. We have
(
p(p−1)/2 0 −1 if p ≡ 1 mod 4,
d(α) = disc(f ) = (−1) Res(f, f ) =
+ 1 else.
Obviously, α lies in OK(α),S , i.e. α is an integer in K(α)P for P|p and p ∈/ S.
For these primes the discriminant of the field extension K(α)P |Kp divides
d(α) = ±1, whence the extension is unramified. Thus K(α)|K is unramified
outside S. Since L|k is a p-closed extension inside kS , we conclude that
℘ : OL,S → OL,S is surjective and obtain the exact sequence
℘
0 −→ ZZ/pZZ −→ OL,S −→ OL,S −→ 0 .
Taking cohomology, the result follows from (8.3.1). 2
Since constant field extensions are unramified, p∞ divides the order of

G(L|k). Applying the last corollary to the fixed field of L with respect to a
p-Sylow group of G(L|k), (3.3.6) implies the
(8.3.3) Corollary. If p = char(k) > 0 and if L|k is a p-closed extension inside

kS , then
cdp G(L|k) = 1.
For the group of units of OS , we have the exact Kummer sequence:
(8.3.4) Proposition. If n ∈ IN(S), then µn ⊆ kS and the group of units OS× of

n
OS is n-divisible, i.e. the map OS× −→ OS× , ε 7−→ εn , is surjective. In other
words, we have an exact sequence of GS -modules
n
0 −→ µn −→ OS× −→ OS× −→ 0.

Proof: If p ∈/ S, then vp (n) = 0, hence kp (µn )|kp is unramified. Therefore

k(µn )|k is unramified outside S, i.e. µn ⊆ kS . Let ε ∈ OS× and let K =√
k(µn , ε).
For every finite prime P ∈/ S, ε and n are√units in KP , hence KP ( ε) | KP
n
is unramified. Therefore the extension K( n ε)|K is unramified outside S, i.e.

√
n
ε ∈ OS× . 2
In the sequel we denote the finite subextensions of kS |k by K|k. As in §1

we select a fixed embedding ip : kS ,→ k̄p for every prime p of k, in particular
a prime p̄ of kS , and denote by P. the prime of K lying under p̄. We set
kS,p = ip (kS )kp and
Kp = KP. = ip (K)kp .
We consider the S-idèle group
×
Y
IK,S := KP .
P∈S
×
It contains the group of S-units OK,S as a discrete subgroup and we set
×
CK,S = IK,S /OK,S .
In spite of the analogy with the formation of the idèle class group CK = IK /K × ,
it is not this group which takes the role of CK in the “S-theory”. The reason
is the failure of Galois descent, i.e. if K|k is Galois, then Ck,S is not always
G(K|k)
the fixed module CK,S . It will become clear in a moment that we have to
consider the group
CS (K) = IK /K × UK,S
instead of CK,S , where UK,S is the compact subgroup
Y Y
UK,S = {1} × UP
P∈S /S
P∈
of the full idèle group IK . Since K × ∩ UK,S = 1, we may regard UK,S as a

subgroup of CK = IK /K × and may also write
CS (K) = CK /UK,S .
This group is called the S-idèle class group. Note that if K|k is Galois,
then UK,S is a cohomologically trivial G(K|k)-module. Namely, by the same
argument as for (8.1.2), we have
Ĥ i (G(K|k), UK,S ) = Ĥ i (G(KP. |kp ), UP. ),
M
/S
p∈
and since KP. |kp is unramified for p ∈/ S, Ĥ i (G(KP. |kp ), UP. ) = 1 by

(7.1.2)(i). The difference of the groups CK,S and CS (K) is given by the

(8.3.5) Proposition. CK,S is an open subgroup of CS (K) and there is an exact

sequence
π
0 −→ CK,S −→ CS (K) −→ ClS (K) −→ 0.
In particular, CK,S = CS (K) if S omits only finitely many primes.
Proof: Consider the canonical injection IK,S → IK , α 7→ α̃, and the induced
homomorphism
j : IK,S −→ IK /K × UK,S = CS (K).
The image of j is closed. If α ∈ IK,S is an idèle such that j(α) = 1, then α̃ = aũ
with a ∈ K × , ũ ∈ UK,S . This means that
αP = α̃P = a for P ∈ S and 1 = ãP = auP for P ∈/ S,
× ×
hence α = a and a ∈ OK,S .
Therefore ker(j) = OK,S
and j induces an injection
CK,S ,→ CS (K) with closed image. Its cokernel is
×
IK /IK,S UK,S K × = ( /UP ))/im(K × ) = ( ZZ)/im(K × ),
M M
(KP
/S
P∈ /S
P∈
which can be identified with the finite ideal class group ClS (K). In particular,
CK,S has finite index in CS (K) and is thus open.
If S omits only finitely many prime ideals, then OK,S is a Dedekind ring with
only finitely many prime ideals and is hence a principal ideal domain by [16],
chap.VII, §2, prop.1. Therefore in this case ClS (K) = 0 and CK,S = CS (K).
2
The proposition shows that the groups CK,S and CS (K) are in general
different. But they become equal in the limit over all K ⊆ kS . We set
IS = lim
−→
IK,S , US = lim
−→
UK,S ,
K|k K|k
CS = lim CK,S , C(kS ) = lim CK ,
−→ −→
K|k K|k
where K|k runs through all finite subextensions of kS |k. These are GS -
modules and US is a cohomologically trivial GS -module, since UK,S is a
cohomologically trivial G(K|k)-module if K|k is Galois. Taking the fixed
module of CS under GS (K) = G(kS |K), we do not recover CK,S but CS (K).
(8.3.6) Proposition. The GS -module CS has the following properties:

(i) CS = lim
−→
CS (K) = C(kS )/US ,
K|k
(ii) CSGS (K) = CS (K).

Proof: (i) We have lim ClS (K) = 0, since every ideal in OK,S becomes
−→ K|k
a principal ideal in a suitable finite unramified Galois extension L|K; if K
is a number field, we take the maximal subfield of the Hilbert class field of
K which is completely split at the primes in S (the principal ideal theorem,
see [160], chap.VI, (7.5)). The same argument holds in the function field case
since ClS (K) is finite for S =/ ∅. Therefore (i) follows from (8.3.5).
(ii) The exact sequence 0 → US → C(kS ) → CS → 0 yields the exact cohomo-
logy sequence
0 −→ UK,S −→ CK −→ CSGS (K) −→ H 1 (GS (K), US ) = 0,
hence CSGS (K) = CK /UK,S = CS (K). 2
If K|k is Galois, then we have the exact sequence of G(K|k)-modules

0 −→ UK,S −→ CK −→ CS (K) −→ 0.
Since UK,S is cohomologically trivial, we obtain the
(8.3.7) Proposition. For every finite Galois subextension K|k of kS |k and

every i ∈ ZZ,
Ĥ i (G(K|k), CS (K)) ∼
= Ĥ i (G(K|k), CK ).
We now compute the cohomology of the three GS -modules in the exact

sequence
0 −→ OS× −→ IS −→ CS −→ 0.
(8.3.8) Proposition (Cohomology of IS ).

(i) H 0 (GS , IS ) = Ik,S .
(ii) H 1 (GS , IS ) = 0.
M
(iii) H 2 (GS , IS )(p) = Br(kp )(p) for every prime number p ∈ IN(S).
p∈S
(iv) H 3 (GS , IS ) = 0.
Proof: (i) is a consequence of (8.1.2). For i ≥ 1 and K|k Galois, we obtain

just as in §1
×
H i (G(K|k), IK,S ) = H i (G(KP |kp ), KP
M
)
p∈S

and furthermore
H i (GS , IS ) = lim
−→
H i (G(K|k), IK,S ).
K|k
From this (ii) follows because of Hilbert’s Satz 90. The assertion (iv) is a
× ∼
consequence of H 3 (G(KP |kp ), KP ) = H 1 (G(KP |kp ), ZZ) = 0 (see (7.1.7)).
For (iii) we have the invariant map of local class field theory
×
(∗) invp : H 2 (G(KP |kp ), KP ∼
) −→ 1
[KP :kp ]
ZZ/ZZ
for nonarchimedean primes p. If k is a function field, then we have constant
extensions of every degree, i.e. lim of (∗) gives Q/ZZ. If k is a number field
−→ K|k
and p ∈ IN(S), then k(µp∞ )|k is an extension inside kS |k of degree divisible by
p∞ , hence
×
lim H 2 (G(KP |kp ), KP )(p) = Qp /ZZp
−→
K|k
for nonarchimedean primes p. For p ∈ S∞ the group G(KP |kp ) is of order √ 1 or

2 depending on whether KP = kp or KP =/ kp . If 2 ∈ IN(S), then K = k( −1)
×
is contained in kS , and so H 2 (G(KP |kp ), KP ) = 12 ZZ/ZZ if p is real. This proves
the proposition. 2
(8.3.9) Proposition (Cohomology of CS ). The pair (GS , CS ) is a class forma-

tion, and we have
(i) H 0 (GS , CS ) = CS (k),
(ii) H 1 (GS , CS ) = 0,
(iii) H 2 (GS , CS ) ∼
= 1
#GS
ZZ/ZZ,
(iv) H 3 (GS , CS ) = 0.
Proof: (i) is the assertion (ii) of (8.3.6). The other equalities follow from
H i (G(K|k), CS (K)) = H i (G(K|k), CK ) , i ≥ 1,
(see (8.3.7)) and from the fact that the Gk -module C = lim CK , where K|k
−→ K|k
runs through all finite Galois subextensions of the separable closure k̄|k, is a
formation module by (8.1.22). 2
As a corollary of the proposition above and (3.1.6), we obtain the global

reciprocity law for restricted ramification.

(8.3.10) Theorem Let K|k be a finite Galois extension inside kS with Galois
group G(K|k). Then there is a canonical isomorphism
CS (k)/NK|k CS (K) ∼ = G(K|k)ab .
(8.3.11) Proposition (Cohomology of OS× ).

(i) H 0 (GS , OS× ) = Ok,S
×
.
(ii) H 1 (GS , O× ) ∼
S = ClS (k).
(iii) For every prime number p ∈ IN(S)
M
H 2 (GS , OS× )(p) = ker H 2 (kp , µp∞ ) −→ H 2 (GS , CS )(p)
p∈S
∼
M
Σ
= ker Qp /ZZp −→ Qp /ZZp or
p∈S\S∞
∼
M
M
1 Σ
= ker Q2 /ZZ2 ⊕ 2
ZZ/ZZ −→ Q2 /ZZ2
p∈S\S∞ p∈SIR
according to whether p =/ 2 or p = 2 (if k is a function field, then the

expressions S∞ and SIR are redundant).
(iv) H 3 (GS , OS× )(p) = 0 for every prime number p ∈ IN(S).
Proof: (i) is trivial and the other statements follow from the exact cohomology
sequence
δ
−→ H i−1 (GS , CS ) −→ H i (GS , OS× ) −→ H i (GS , IS ) −→ H i (GS , CS ) −→ .
For i = 1 we obtain by (8.3.8) and (8.3.9) the exact sequence
δ
Ik,S −→ CS (k) −→ H 1 (GS , OS× ) −→ 0.
The cokernel of the left arrow is ClS (k) by (8.3.5), proving (ii).
If k is a function field or p =/ 2, then (iii) follows from the commutative diagram
0783456 H 2 (GS , OS× )(p) H 2 (GS , IS )(p) H 2 (GS , CS )(p)
L
invp
Σ
M
Qp /ZZp Qp /ZZp ,
p∈S\S∞
where the lower horizontal map is the summation over the components and
the zero on the left in the upper row is a consequence of H 1 (GS , CS ) = 0. If
p = 2 and k is a number field, then the same argument gives the result using
H 2 (GS , IS )(2) ∼
= p∈S\S∞ Q2 /ZZ2 ⊕ p∈SIR 12 ZZ/ZZ.
L L
Finally, (iv) follows from the surjectivity of the upper right-hand arrow and
from the equality H 3 (GS , IS ) = 0. 2

(8.3.12) Corollary. Let k be a global field and let p ∈ S be a prime number.

(i) If k is a function field or p =/ 2, then H 2 (GS , OS× ) is p-divisible.
(ii) If k is a number field and p = 2, then the canonical map
H 2 (GS , OS× )/2 −→ H 2 (kp , k̄p× )
M
∼
p∈SIR
is an isomorphism. In particular, H (GS , OS× ) is 2-divisible if and only

2
if k is totally imaginary.
(iii) If k is a number field and p = 2, then the restriction map
M
H 3 (GS , ZZ/2ZZ) −→
∼ H 3 (kp , ZZ/2ZZ)
p∈SIR
is an isomorphism.
Proof: The first two assertions follow immediately from the fact that the
group H 2 (GS , OS× )(p) is isomorphic to the kernel of the summation over the
local components.
Using (8.3.11)(iv), it follows from the exact sequence
2
0 −→ µ2 −→ OS× −→ OS× −→ 0
= H 3 (GS , ZZ/2ZZ). Since H 2 (kp , k̄p× ) ∼
that H 2 (GS , OS× )/2 ∼ = H 3 (kp , ZZ/2ZZ)
for p ∈ SIR , we obtain the last assertion from (ii). 2
The groups CS (k) are locally compact topological groups exactly like Ck .
Concerning its connected component we have the
(8.3.13) Theorem. Assume that k is a number field. The connected component

DS (k) of CS (k) = Ck /Uk,S is given by
DS (k) = Dk Uk,S /Uk,S .
It is divisible, and there is an exact sequence of topological groups
( ,kS |k)
09:;< DS (k) CS (k) Gab
S 0.
In particular, DS (k) is the group of universal norms NGS CS .
Proof: Consider the continuous projection

(∗) Ck −→ CS (k) = Ck /Uk,S .
For general topological reasons, the connected component DS (k) of CS (k) is
the closure of the image Dk Uk,S /Uk,S of the connected component Dk of Ck .

But this image is already closed, since Uk,S is compact and hence (∗) is a proper
map. This proves the first assertion. The divisibility follows from that of Dk .
For the second statement we use the exact sequence (8.2.2):
0@=>?
( ,k)
Dk Ck Gab 0.
For each nonarchimedean prime p the norm residue symbol ( , k) maps the
subgroup kp× ⊆ Ck into the decomposition group of Gab k with respect to p and
the group of units Up ⊆ kp× onto the inertia group (see [160], chap.VI, (5.6) and
chap.V, (6.2)). Therefore the compact group Uk,S is mapped onto the subgroup
H of Gab ab
k = G(k |k) which is generated by the inertia groups for the primes
not in S. The fixed field of H is therefore the maximal subextension kSab |k
of k ab |k which is unramified outside S. This gives us an exact commutative
diagram
ABCDEFGH Uk,S G(k ab |kSab ) 0
0 Dk Ck G(k ab |k) 0
of topological groups. The snake lemma yields an exact sequence of topolog-
ical groups
0 −→ Dk Uk,S /Uk,S −→ Ck /Uk,S −→ G(kSab |k) −→ 0,
and since G(kSab |k) = G(kS |k)ab = Gab
S , the theorem is proved. 2
(8.3.14) Corollary. Let k be a number field. For a subgroup N of CS (k) the

following conditions are equivalent:
(i) N is the norm group NK|k CS (K) of a finite Galois extension K|k inside
kS |k.
(ii) N is an open subgroup of CS (k).
Each open subgroup N of CS (k) contains DS (k) and has finite index.
Proof: By (8.3.9), the pair (GS , CS ) is a class formation. The isomorphism

(1) ∼ Gab
CS (k)/DS (k) −→ S
is the projective limit of the isomorphisms of finite groups
(2) CS (k)/NK|k CS (K) ∼
= G(K|k)ab ,
where K|k runs through the finite Galois subextensions of kS |k. Therefore the
norm groups are precisely the open subgroups of finite index in CS (k) which
contain DS (k). It remains to show that each open subgroup in CS (K) has finite
index and contains DS (k). As DS (k) is divisible, it suffices to show that each
open subgroup in CS (k) has finite index.

As CS (k) is a topological quotient of Ck , we may assume that S is the set

of all primes. Let W ⊆ Ck be an open subgroup. The pre-image of W in Ik is
open and therefore contains a subgroup of the form
Y
V = Vp ,
p
where Vp ⊆ Up is an open subgroup for all p, Vp = Up for almost all p and

Vp ⊆ Up1 for archimedean p. For nonarchimedean p, an open subgroup in
Up has finite index and for archimedean p the group Up1 has no proper open
subgroup. Hence V has finite index in the group
Y Y
Uk = Up × Up1 ⊆ Ik ,
p-∞ p|∞
and it suffices to show that the image of Uk in Ck has finite index. We have
Ik /Uk ∼
M M
= ZZ ⊕ ZZ/2ZZ,
/S∞
p∈ p∈SIR
which gives an isomorphism

Ik /k × Uk ∼
= Cl0 (k)
where Cl0 (k) is the ideal class group of k in the narrow sense, which is known
to be finite. This concludes the proof. 2
For a function field k we have a continuous injection

CI k
( ,k)
Gab
k .
It is not surjective, since in Ck we have the group Ck0 = {x ∈ Ck | |x| = 1}, and
the quotient Ck /Ck0 ∼ = ZZ is not profinite. With the same arguments as above
in this situation, we obtain the
(8.3.15) Theorem. If k is a function field, then we have an exact sequence of

topological groups
( ,kS |k)
0LMJK CS (k) Gab
S ẐZ/ZZ 0.
(8.3.16) Corollary. Let k be a function field. For a subgroup N of CS (k) the

following conditions are equivalent:
(i) N is the norm group NK|k CS (K) of a finite Galois subextension K|k of
kS |k.
(ii) N is an open subgroup of finite index.
(iii) N is an open subgroup which is not contained in the group
CS0 (k) = {x ∈ CS (k) | |x| = 1}.

As before let k be a global field, S a nonempty set of primes of k containing

the set S∞ of infinite primes if k is a number field, and let kS |k be the maximal
extension unramified outside S. Now we study the cohomological dimension
of GS . In the function field case we have the following result (for the case
S = ∅, which is not considered here, we refer to X §1).
(8.3.17) Theorem. If k is a function field, then for all prime numbers p

scdp GS = 2.
Proof: Since the constant field extensions are contained in kS , every prime
number p divides the order of GS , hence cdp GS > 0 and therefore scdp GS ≥ 2,
because the strict cohomological dimension is never 1. It therefore suffices to
prove scd GS ≤ 2. By (3.6.4), this is equivalent to the existence of a level-
compact formation module for GS with trivial universal norms. We construct
such a formation module from the GS -module CS = C(kS )/US , which is a
formation module by (8.3.9).
Let L|K be a finite normal intermediate extension in kS |k. The norm group
N = NL|K CL is an open subgroup of finite index in CK . The property of being
unramified at a prime p is equivalent to Up ⊆ NLP |Kp L× ×
P = N ∩ Kp by class
field theory (see [160], chap.VI, (5.8)). This means that the norm groups of
finite Galois subextensions L|K of kS |K are open subgroups N of finite index
in CK containing UK,S = UK,S K × /K × . By the existence theorem of class field
theory, they are precisely all such groups (8.3.16), and hence UK,S K × /K × is
their intersection. Therefore CS (K) = CK /UK,S becomes a dense subgroup
of the compact group
C S (K) = lim
←−
CK /N.
N
The canonical surjective map
X
deg : CS (K) → ZZ, deg(α) = vp (αp ) deg(p) ,
p
extends to C S (K) and yields an exact commutative diagram
0QRSTNOPUVWX
deg
CS0 (K) CS (K) ZZ 0
0,
deg
0 CS0 (K) C S (K) ẐZ
which shows that the quotient C S (K)/CS (K) is isomorphic to ẐZ/ZZ and is
consequently uniquely divisible. Because of this fact, for a normal interme-
diate extension L|K with group G, the G-module C S (L)/CS (L) has trivial

cohomology, so that
H i (G, C S (L)) = H i (G, CS (L)) for i ≥ 1,
and from CS (L)G = CS (K) it follows that C S (L)G = C S (K). This shows that
the G-module C S = lim C̄S (K) is a formation module since CS is.∗) It
−→ K|k
is level-compact because C̄SGS (K) = C S (K), and it has trivial universal norms
because
NL|K CL = UK,S K × /K × ,
\ \
i.e. NL|K C S (L) = 0.
K ⊆L⊆kS K ⊆L⊆kS
C S is the desired formation module and the theorem is proved. 2
Now we assume that k is a number field. In this case the existence of

the infinite primes causes severe difficulties for the determination of the strict
cohomological dimension scdp GS . However, the cohomological dimension
cdp GS is more easy to handle.
For a prime number p we put
Sp = Sp (k) = {p a prime of k dividing p} .
(8.3.18) Proposition. Let Sp ∪ S∞ ⊆ S and assume that the number field k is

totally imaginary if p = 2. Then
cdp GS ≤ 2.
Proof: kS contains the group µp of p-th roots of unity since Q(µp )|Q, and
hence also k(µp )|k, is unramified outside p. The group OS× of S-units of kS
contains µp and is p-divisible by (8.3.4), i.e. the sequence
p
(∗) 0 −→ µp −→ OS× −→ OS× −→ 0
is exact. Let K|k be any finite subextension of kS |k. Proposition (8.3.11)
shows that H 3 (GS (K), OS× )(p) = 0, and by (8.3.12) the group H 2 (GS (K), OS× )
is p-divisible. Therefore the exact cohomology sequence associated to (∗)
yields H 3 (GS (K), µp ) = 0. Now let Gp be a p-Sylow subgroup of GS . Let
Σp be its fixed field and let K|k run through the finite subextensions of Σp |k.
Noting that µp ⊆ Σp , we get
H 3 (Gp , ZZ/pZZ) ∼
= H 3 (Gp , µp )(−1) = lim
−→
H 3 (GS (K), µp )(−1) = 0,
K
hence cdp GS = cdp Gp ≤ 2. 2
∗) If S = ∅, then C is no longer a formation module.
S

(8.3.19) Proposition. Let S2 ∪ S∞ ⊆ S and let K be an (possibly infinite)

extension of the number field k inside kS . Then cd2 GS (K) = ∞ if and only if
K has a real place, and cd2 GS (K) ≤ 2 otherwise.
√
Proof: As S2 ∪ S∞ ⊆ S the field kS contains K( −1). Therefore, if K has a
real place, then GS (K) contains its decomposition group which is isomorphic
to ZZ/2ZZ. Hence cd2 GS (K) = ∞ by (3.3.5).
If cd2 GS (K) = ∞, then H n ((GS (K))2 , ZZ/2ZZ) =/ 0 for all n and every 2-
Sylow subgroup (GS (K))2 of GS (K) by (3.3.6) and (3.3.2)(iii). We therefore
find a finite extension K 0 |K inside kS such that H 3 (GS (K 0 ), ZZ/2ZZ) =/ 0. From
(8.3.12)(iii) we obtain
0 =/ H 3 (GS (K 0 ), ZZ/2ZZ) ∼
M
= lim H 3 (Lp , ZZ/2ZZ) ,
−→
L p∈SIR (L)
where L runs through the finite extensions of k in K 0 . We conclude that K 0 ,

and hence also K, has at least one real place.
assume that cd2 GS (K) is finite.
Finally, √ √ Then cd2 GS (K) is equal to
cd2 GS (K( −1)) by (3.3.5)(ii). √But K( −1) is the union of totally imaginary
number fields, hence cd2 GS (K( −1)) ≤ 2 by (8.3.18). 2
Remark: The assumptions of (8.3.18) may be weakened to Sp ⊆ S and, if

p = 2, S contains no real primes. The statement of (8.3.19) may be sharpened
in the following way: Assume that S2 ⊆ S; then cd2 (GS (K)) = ∞ if and only
if S contains a real prime, and cd2 (GS (K)) ≤ 2 otherwise, see (10.6.1).
Together with (8.3.11), we obtain from (8.3.18) the following finiteness

theorem.
(8.3.20) Theorem. Let A be a GS -module which is finitely generated as a

ZZ-module and whose torsion submodule tor(A) is of an order which is a unit
in Ok,S . Then the following hold:
(i) If S and A are finite, then the cohomology groups H n (GS , A) are finite
for all n ≥ 0.
(ii) If S is finite or A is torsion-free, then H 1 (GS , A) is finite.
Proof: (i) Let K|k be a finite Galois subextension of kS |k over which A

becomes a trivial Galois module and which contains the n-th roots of unity,
where n = #A, and which is totally imaginary in the number field case. Then
A is isomorphic over K to a direct sum of modules µpν , p|n. If the assertion

holds for A = µpν and base field K, the general case follows using the spectral
sequence
H i (G(K|k), H j (GS (K), A)) ⇒ H i+j (GS , A).
Therefore let A = µpν . By induction, using the exact sequence 0 → µp → µpν
→ µpν−1 → 0 and its associated exact cohomology sequence, we may assume
A = µp with p ∈ IN(S). Therefore, as in the proof of (8.3.18), we have the
exact sequence p
0 −→ µp −→ OS× −→ OS× −→ 0,
and the associated exact cohomology sequence
p
H i−1 (GS , OS× ) −→ H i−1 (GS , OS× ) −→ H i (GS , µp ) −→ . . .
p
. . . −→ H i (GS , OS× ) −→ H i (GS , OS× ).
× × p
For i = 1 we obtain the finiteness of H 1 (GS , µp ) since Ok,S /(Ok,S ) and
1 × ∼
H (GS , OS ) = ClS (k) are finite. For i = 2 we use the finiteness of the set
p
S and see by (8.3.11)(iii) that the kernel of H 2 (GS , OS× ) → H 2 (GS , OS× ) is
finite, and thus also H 2 (GS , µp ), again using H 1 (GS , OS× ) ∼= ClS (k). Since
n
H (GS , µp ) = 0 by (8.3.18) for n 3, part (i) of the theorem is proved.
≥
(ii) Let A be torsion-free and let K|k be a finite Galois subextension of kS |k

over which A becomes a trivial Galois module, i.e. A = ZZr for some r ≥ 0.
Then the exact sequence
0 −→ H 1 (G(K|k), A) −→ H 1 (GS , A) −→ H 1 (GS (K), ZZr ) = 0
gives the desired result. If S is finite and is A arbitrary, then the exact sequence
H 1 (GS , tor(A)) −→ H 1 (GS , A) −→ H 1 (GS , A/tor(A))
gives the result using (i) and the result for torsion-free modules. 2
Let N run through the open subgroups of finite index in Ik,S and put
I¯k,S = lim Ik,S /N,
←−
N
i.e. I¯k,S is the profinite completion of Ik,S . We can collect many of the results
obtained in the form of two commutative exact diagrams.
(8.3.21) Corollary.
(i) Let k be a function field. Then we have the commutative exact diagram
0fdYZ[\]^_àbce ×
Ok,S Ik,S CS (k) ClS (k) 0
dense dense dense
×
0 Ok,S I¯k,S GS (k)ab ClS (k) 0.

(ii) Let k be a number field. Then we have the commutative exact diagram
vwxughijklmnopqrst
M
NGp k̄p× DS (k)
p∈S∞
×
0 Ok,S Ik,S CS (k) ClS (k) 0
dense dense
×
0 Ok,S I¯k,S GS (k)ab ClS (k) 0.
× ×
Here Ok,S denotes the closure of the image of Ok,S in I¯k,S . The left-hand
vertical arrow in (ii) is injective if S contains at least one nonarchimedean
prime.
Remark: If S is finite, local class field theory induces an isomorphism I¯k,S →

∼
Q ab ¯
p∈S Gkp . For infinite S, we obtain an isomorphism Ik,S →
∼ ˆ p∈S Gk , where
Q ab
p
the latter group is the free corestricted proabelian product of the groups Gabkp
with respect to their inertia subgroups; see [154].
Exercise 1. Show the isomorphism H 2 (GS , OS× ) ∼

= Br(kS |k).
Exercise 2. Let k be a number field. Let K|k be a finite Galois subextension of kS |k with
Galois group G, and let DS (K) = DK ∩ UK,S . If k is totally imaginary, then
(i) H 0 (G, DS (K)) = DS (k),
(ii) H 0 (G, DS (K)) = DS (k) ⇐⇒ H 1 (G, DS (K)) = 0,
(iii) Ĥ i (G, DS (K)) ∼
= Ĥ i (G, DS (K)) for all i ∈ ZZ.
Hint: DK is a cohomologically trivial G-module.
×
Exercise 3. Show that the group Ck,S = Ik,S /Ok,S contains the connected component Dk
of Ck .
§4. The Global Duality Theorem
In this section we consider the profinite group

GS = G(kS |k)
of the maximal extension kS |k which is unramified outside the given nonempty
set of primes S (where S∞ is contained in S if k is a number field). The abstract
duality theorem (3.1.9) applies at once to any Galois extension K|k inside kS
in view of (8.3.9).

§4. The Global Duality Theorem 467
(8.4.1) Theorem. Let K|k be a Galois extension which is unramified outside S,

let G be its Galois group and let A be a finitely generated and ZZ-free G-module.
Then for all i ∈ ZZ the cup-product
∪ inv
Ĥ i (G, Hom(A, CS (K))) × Ĥ 2−i (G, A) −→ H 2 (G, CS (K)) −→ Q/ZZ
yields a topological isomorphism
Ĥ i (G, Hom(A, CS (K))) ∼

= Ĥ 2−i (G, A)∨ .
If K|k is finite, these are isomorphisms of finite abelian groups.
Remark: Recall that if K|k is infinite, the topologies on the occurring co-
homology groups are defined as the direct resp. projective limit topologies of
the groups on the finite levels, which are finite and endowed with the discrete
topology. The topology of the S-idèle class group does not play any role here.
The theorem (8.4.1), concerning only ZZ-free G-modules, is too rigid for
the applications that we have in mind. In the local case we have proved a
cohomological duality for finite Galois modules A and A0 = Hom(A, k̄ × ). We
would like to prove an analogous theorem for the global Galois group GS ,
where the idèle class group CS offers itself as an analogue of the multiplicative
group k̄ × . However, for several reasons CS cannot take over the role of k̄ × :
we cannot apply the duality theorem (3.4.3) of Tate, since on the one hand it is
not known whether scd GS ≤ 2 if S is finite and k is a number field, and on the
other hand for large sets S (e.g. S has density equal to 1) the module CS is not
divisible. The duality theorem (3.4.6) seems to be more convenient, since one
only needs the cohomological dimension of GS , but again there is a problem
with the divisibility of the dualizing module of GS . We will see in X §9 that
in fact CS is p-divisible if p ∈ S and S is finite. But this will be a consequence
of the results of this and the following sections.
We have still the abstract duality theorem (3.1.11)(i), which does not require
the divisibility of CS . Instead it requires a level-compact formation module
and divisible universal norm groups. CS satisfies the last condition, not the
first one. But we may pass to a modified formation module satisfying both
conditions without changing the cohomology. This was the idea of G. POITOU
and K. UCHIDA. In what follows the reader should recall the topological remarks
before (3.1.11).
As before, K|k denotes the finite subextensions of kS |k. We consider the

GS -module CS = C(kS )/US with the fixed modules
CS (K) = CSGS (K) = CK /UK,S .

For the following discussion we have to distinguish between the case of number
fields and of function fields. First let k be a number field. UK,S is contained in
the compact group
||
CK0
= ker(CK −→ IR× + ),
and we replace CS (K) by

CS0 (K) = CK
0
/UK,S .
The quotient CS (K)/CS0 (K) ∼
= IR×
+ is uniquely divisible. We set
CS0 = lim
−→
CS0 (K) = C 0 (kS )/US
K|k
where K|k runs through the finite subextensions of kS |k.
(8.4.2) Proposition. Let k be a number field. Then the following is true.

(i) The GS -module CS0 is a level-compact formation module with divisible
universal norm group NkS |k CS0 .
(ii) If A is a GS -module which is finitely generated as a ZZ-module, then for all
i ∈ ZZ the natural map CS0 → CS induces an isomorphism
Ĥ i (GS , Hom(A, CS0 )) −→
∼ Ĥ i (G , Hom(A, C )).
S S
Proof: The GS -module US is cohomologically trivial as mentioned on sev-

eral previous occasions. Therefore, applying H 0 (G(kS |K), −) to the exact
sequence
0 −→ US −→ C 0 (kS ) −→ CS0 −→ 0,
0
0 −→ UK,S −→ CK −→(CS0 )G(kS |K) −→ 0,
and so
(CS0 )G(kS |K) = CS0 (K).
Thus CS0 is a level-compact GS -module. By (8.3.13), NkS |K CS = DS (K) is
||
divisible, hence also DS0 (K) = ker(DS (K) −→ IR× + ). Since |NK 0 |K (x)| = 1 if
0 0
and only if |x| = 1 for x ∈ CS (K ), where K |K is a finite subextension of
kS |K, we get NkS |K CS0 = DS0 (K), and hence NkS |K CS0 is divisible.
Now let A be a GS -module which is finitely generated as a ZZ-module and
let K|k be a finite Galois extension inside kS over which A becomes trivial.
Consider the exact sequence of G(K|k)-modules
0 −→ CS0 (K) −→ CS (K) −→ Q(K) −→ 0

in which Q(K) = CS (K)/CS0 (K) ∼ = IR is uniquely divisible. It follows that

Hom(A, Q(K)) is also uniquely divisible and the sequence
0 −→ Hom(A, CS0 (K)) −→ Hom(A, CS (K)) −→ Hom(A, Q(K)) −→ 0
is exact. We therefore obtain natural isomorphisms
Ĥ i (G(K|k), Hom(A, CS0 (K))) −→
∼ Ĥ i (G(K|k), Hom(A, C (K)))
S
for all i ∈ ZZ. 2
For function fields there is the following similar result. For each finite
subextension K|k of kS |k consider the compact group
C S (K) = lim
←−
CK /N,
N
where N runs through the open subgroups of finite index in CK which contain
UK,S . As before let
C S = lim
−→
C S (K).
K|k
(8.4.3) Proposition. Let k be a function field. Then the following is true.

(i) The GS -module C S is a level-compact formation module with trivial uni-
versal norm group NkS |k C S = 0.
(ii) If A is a GS -module which is finitely generated as a ZZ-module, then for all
i ∈ ZZ the natural homomorphism CS → C S induces an isomorphism
Ĥ i (GS , Hom(A, CS )) −→
∼ Ĥ i (G , Hom(A, C )).
S S
Proof: The first assertion was already shown in the proof of (8.3.17) and the
second follows in the same way as in (8.4.2) using the exact sequence
0 −→ CS (K) −→ C S (K) −→ ẐZ/ZZ −→ 0,
where K|k is a finite Galois extension inside kS trivializing A. Since ẐZ/ZZ is
uniquely divisible, we obtain isomorphisms
Ĥ i (G(K|k), Hom(A, CS (K))) −→
∼ Ĥ i (G(K|k), Hom(A, C (K)))
S
for all i ∈ ZZ. 2
Applying the duality theorem (3.1.11)(i) to an arbitrary global field, we

obtain the

(8.4.4) Theorem. Let A be a discrete GS -module which is finitely generated

as a ZZ-module. If k is a number field, then the pairing
∪
Ĥ i (GS , Hom(A, CS0 )) × Ĥ 2−i (GS , A) −→ H 2 (GS , CS0 ) ∼
= 1
#GS
ZZ/ZZ
induces topological isomorphisms
Ĥ i (GS , Hom(A, CS0 )) ∼

= Ĥ 2−i (GS , A)∨
for all i ≤ 0. This assertion holds for all i ∈ ZZ if A is assumed to be ZZ-free.

If k is a function field, then the statements above hold with CS0 replaced by C S .
Remark: In order to apply also part (ii) of (3.1.11) to (GS , CS0 ) (resp. (GS , C S ))
one has to investigate the question of finding those primes p for which the group
CS is p-divisible. We will do this in chapter X, see (10.11.8).
We formulated the theorem with CS0 (resp. C S ) instead of CS to obtain the

additional information that the duality isomorphisms are topological. If one
wants to define a topology on the cohomology groups of CS (A) directly, one
runs into the problem that the category of locally compact abelian groups is not
abelian. While it is possible to overcome this technical difficulty, we restrict
our considerations here to the case of cohomology in dimension zero:
As before, let A be a discrete GS -module which is finitely generated as a
ZZ-module. For U ⊆ GS open, we endow the groups CS (A)U with a natural
(locally compact) topology: For sufficiently small V we have AV = A and
therefore
CS (A)V = Hom(A, CSV ) .
Considering A as a group with the discrete topology, Hom(A, CSV ) endowed
with the compact open topology is locally compact. For an arbitrary open
subgroup U ⊆ GS , the group Hom(A, CS )U is a closed subgroup of the locally
compact group Hom(A, CS )V , where V is chosen sufficiently small. For
U ⊆ GS the norm map NGS /U : CS (A)U → CS (A)GS is continuous. If U is
normal in GS , we endow Ĥ 0 (GS /U, CS (A)U ) with the quotient topology from
H 0 (GS /U, CS (A)U ) = CS (A)GS . Finally, we endow
Ĥ 0 (GS , CS (A)) = lim Ĥ 0 (GS /U, CS (A)U )

←−
U
with the projective limit topology. Note that the natural homomorphism
H 0 (GS , CS (A)) −→ Ĥ 0 (GS , CS (A))
is continuous and has dense image.

(8.4.5) Proposition. The algebraic isomorphisms

∼ Ĥ 0 (G , C (A)),
Ĥ 0 (GS , CS0 (A)) −→ S S
given by (8.4.2) if k is a number field, and

Ĥ 0 (GS , CS (A)) −→
∼ Ĥ 0 (G , C (A)),
S S
given by (8.4.3) if k is a function field, are homeomorphisms. The composite

∼ H 2 (G , A)∨
CS (A)GS −→ Ĥ 0 (GS , CS (A)) −→ S
is continuous and has dense image. It is surjective if k is a number field or if

A is finite.
Proof: Let K = kSU be a finite Galois extension of k inside kS such that

AU = A. We first show that the norm map NGS /U : CS (A)U → CS (A)GS has
closed image. From this follows that Ĥ 0 (GS , CS (A)) is Hausdorff.
We start with the number field case. Let v be an archimedean prime of k and
let Γk be the image of the subgroup of positive real numbers under the injection
kv× ,→ CS (k). Let ΓK be the image of Γk in CS (K). Then we have an algebraic
decomposition of CS (A)U into a product of closed G(K|k)-submodules
CS (A)U ∼= C 0 (A)U × Hom(A, ΓK ) .
S
As CS0 (A)U is compact and the norm is continuous, NGS /U CS0 (A)U is a
compact subgroup of CS (A)GS . The norm map NGS /U : Hom(A, ΓK ) →
HomGS (A, ΓK ) is a homomorphism of finite dimensional IR-vector spaces and
has closed image. Therefore
NGS /U CS (A)U = NGS /U CS0 (A)U × NGS /U Hom(A, ΓK )
is a closed subgroup of CS (A)GS .
In the function field case, we proceed in a similar way. Let CS0 ⊆ CS be the
subgroup of idèle classes of degree zero. CS0 is a level-compact GS -module.
Let x ∈ CS (k) be a pre-image of 1 under the degree map deg : CS (k) ZZ,
and let Γk ∼= ZZ be the discrete subgroup in CS (k) generated by x. Denote by
ΓK the image of Γk in CS (K). Then
CS0 (A)U × Hom(A, ΓK )
is an open subgroup of finite index in CS (A)U . The norm group NGS /U CS0 (A)U
is compact and NGS /U Hom(A, ΓK ) is discrete as a subgroup of the discrete
group HomGS (A, Γk ). We conclude that
NGS /U CS0 (A)U × NGS /U Hom(A, ΓK ) ⊆ Hom(A, CS )GS
is closed, as it is the product of a compact subgroup and a discrete subgroup.
Therefore NGS /U CS (A)U , containing a closed subgroup of finite index, is
closed.
In the number field case, we conclude that the isomorphism

(∗) Ĥ 0 (G(K|k), Hom(A, CS0 (K))) −→

∼ Ĥ 0 (G(K|k), Hom(A, C (K)))
S
obtained in the proof of (8.4.2) is a continuous bijection from a compact group

onto a Hausdorff group, and therefore a homeomorphism.
Now let k be a function field. In the proof of (8.4.3) we obtained the
continuous bijection
(∗∗) Ĥ 0 (G(K|k), Hom(A, CS (K))) −→
∼ Ĥ 0 (G(K|k), Hom(A,
C S (K))).
The group CS (K)/CS0 (K) is a trivial G(K|k)-module isomorphic to ZZ. There-
fore the image of the natural homomorphism
Ĥ 0 (G(K|k), Hom(A, CS0 (K))) −→ Ĥ 0 (G(K|k), Hom(A, CS (K)))
is a compact subgroup of finite index. Hence Ĥ 0 (G(K|k), Hom(A, CS (K))) is
compact and (∗∗) is a homeomorphism.
Passing to the projective limit over all K in (∗) or (∗∗), respectively, we obtain
the first result, and the composite CS (A)GS → Ĥ 0 (GS , CS (A)) → ∼ H 2 (G , A)∨
S
is continuous and has dense image.
If k is number field, then the module Hom(A, CS0 ) is level-compact, and
so CS0 (A)GS → Ĥ 0 (GS , Hom(A, CS0 )) is surjective by (1.9.11). It follows that
CS (A)GS → Ĥ 0 (GS , CS (A)) is surjective. If k is a function field and A is
finite, then the module Hom(A, CS ) = Hom(A, C S ) is level-compact, and
CS (A)GS → Ĥ 0 (GS , CS (A)) is also surjective in this case. 2
§5. Local Cohomology of Global Galois Modules
As a bridge from local to global cohomology, we shall now introduce an

idèlic cohomology for GS -modules, GS again being the Galois group G(kS |k).
We start with some general observations concerning the passage from global
to local Galois modules.
Let k be a global field, p a prime of k, K|k a (possibly infinite) Galois

extension and P a prime of K above p. The field
kp0 0 ,
[
KP =
k0
where k runs through the finite extensions of k inside K and p0 = P ∩ k 0 , is
0
a (possibly infinite) Galois extension of the local field kp . The Galois group
G(KP |kp ) is naturally isomorphic to the decomposition group GP (K|k) of P
in G(K|k). We fix a separable closure K P of KP (which is also a separable
closure of kp ).

§5. Local Cohomology of Global Galois Modules 473
Now let A be a discrete G(K|k)-module. We consider A as a G(K P |kp )-

module via the natural projection
G(K P |kp ) GP (K|k).
As any two extensions P1 and P2 of p to K are conjugate, and as in-
ner automorphisms of groups act trivially on group cohomology, the groups
H i (G(K P1 |kp ), A) and H i (G(K P2 |kp ), A) are canonically isomorphic for all i.
We denote them by
H i (kp , A),
hiding all choices from the notation. We also write the inclusion of K to K P
in the form K ,→ k̄p . There are canonical homomorphisms
H i (G(K|k), A) −→ H i (kp , A)
for all i ≥ 0. If A is trivialized by a finite extension k 0 of k, which happens
for example if A is finitely generated as a ZZ-module, then A is an unramified
Gkp -module for all primes p which are unramified in k 0 |k. In particular, A is
an unramified Gp -module for almost all primes p of k.
We will apply the above considerations to the case K = kS . Recall the
notation
×
IN(S) = {n ∈ IN | n ∈ Ok,S }.
(8.5.1) Definition. We denote by ModS (GS ) the category of discrete GS -

modules A which are finitely generated as ZZ-modules and whose torsion has
order #tor(A) ∈ IN(S).
For a finite A ∈ ModS (GS ), we define the dual GS -module of A by

A0 = Hom(A, OS× ) = Hom(A, µ),
where OS× is the group of S-units of kS and µ is the group of roots of unity in
kS . As n = #A is in IN(S) by assumption, kS contains all n-th roots of unity.
Therefore A0 is finite of the same order as A and contained in ModS (GS ).
Furthermore, we have a canonical identification A00 = A.
We set
P i (GS , A) = H i (kp , A) ,
Y
i ≥ 0,
p∈S
i
where the restricted product is taken with respect to the subgroups Hnr (kp , A) =
i i i
im(H (k̃p |kp , A) → H (kp , A)) of H (kp , A), where k̃p is the maximal unrami-
fied extension of kp , see (7.2.14). In view of the duality theorem (7.2.17) for
the groups Ĥ i (kp , A) at the archimedean primes p, we make the following

Convention: In the case i = 0 we agree that at archimedean primes the groups

H 0 (kp , A) denote the modified cohomology groups Ĥ 0 (kp , A).
Since the groups H i (kp , A) are all finite, we see that

Y Y
P 0 (GS , A) = H 0 (kp , A) × Ĥ 0 (kp , A) is compact,
S\S∞
p∈Y p∈S∞
P 1 (GS , A) = H 1 (kp , A) is locally compact,
p ∈S
M
P 2 (GS , A) = H 2 (kp , A) is discrete,
p∈ S
M
P i (GS , A) = H i (kp , A) is finite for i ≥ 3.
p∈S∞
The direct sums in the last two lines occur because cd G(k̃p |kp ) = cd ẐZ = 1 and
cd Gkp = 2. If S is finite, all groups above are finite and the restricted product
is just the product.
The local duality theorem (7.2.9) has the following effect on the groups P i .
Let A ∈ ModS (GS ) be finite. Then for p - ∞, the pairing
∪ invp
H i (kp , A) × H 2−i (kp , A0 ) −→ H 2 (kp , k̄p× ) −→ Q/ZZ,
0 ≤ i ≤ 2, yields an isomorphism
(∗) Ξip : H i (kp , A) −→ H 2−i (kp , A0 )∨ .
For p | ∞ we have to replace H 0 by Ĥ 0 . If A is unramified at p - ∞ and

2−i
p - #A, then, by (7.2.15), the groups Hnr (kp , A0 ) and Hnr
i
(kp , A) are exact
orthogonal complements under the pairing, i.e. Hnr (kp , A) is mapped by Ξip
i
isomorphically onto the subgroup

(H 2−i (kp , A0 )/Hnr
2−i
(kp , A0 ))∨ ⊆ H 2−i (kp , A0 )∨ .
For the restricted product of the right-hand groups with respect to the left-hand
subgroups we have the equality (cf. (1.1.13))
H 2−i (kp , A0 ))∨ .
Y Y
H 2−i (kp , A0 )∨ = (
p∈S p∈S
Since A is unramified for almost all p, we obtain the
(8.5.2) Proposition. If A ∈ ModS (GS ) is finite, the maps Ξip define homeo-
morphic isomorphisms
Ξi : P i (GS , A) −→ P 2−i (GS , A0 )∨
for 0 ≤ i ≤ 2.

Now let A ∈ ModS (GS ) be arbitrary. We again define the dual GS -module
of A by
A0 = Hom(A, OS× ).
However, this is not completely satisfactory if A is not finite. Firstly, A0 is not
necessarily in ModS (GS ), and in particular, the situation is asymmetric; also
an equality like A00 = A is no longer true. Looking at this situation from the
point of view of sheaves, we are interested in the dual sheaf, which is no longer
locally constant, i.e. is not represented by a Galois module. That is why we
make the following notational convention:
For every prime p we write
def
H i (kp , A0 ) = H i (kp , Hom(A, k̄p× )),
i.e. we consider A as a local module first, and then take the dual in the local
sense. Note that we have an injection A0 ,→ Hom(A, k̄p× ) of G(k̄p |kp )-modules,
which induces maps for all i
H i (GS , A0 ) −→ H i (kp , A0 ).
If A (and hence also A0 ) is finite, both possible interpretations of H i (kp , A0 )
coincide. Now assume that A is unramified at the finite prime p. Then we
define
def
Hnri
(kp , A0 ) = Hnr
i
(kp , Ad ),
where Ad = Hom(A, Õp× ) (see VII §2). Again, no confusion occurs if A is
finite. If M is A or A0 , we set, as in the case of finite modules,
P i (GS , M ) = H i (kp , M ) ,
Y
i ≥ 0,
p∈S
i
where the restricted product is taken with respect to the subgroups Hnr (kp , M )
i 0
of H (kp , M ), and we keep our convention concerning Ĥ at archimedean
primes.
However, we have to be careful about the topologies: the cohomology
groups of A are considered as topological groups with the discrete topology
and the same holds for H i (kp , A0 ), i > 0. But for i = 0 the group H 0 (kp , A0 )
inherits the topology of Hom(A, k̄p× ) and is not discrete. Now for the topology
of P i (GS , M ), where M = A or A0 , a basis of neighbourhoods of the identity
is given by the subgroups
i
Y Y
Xp × Hnr (kp , M ),
p∈T p∈S\T
where T varies over the finite subsets of S and Xp varies over a basis of
neighbourhoods of the identity in H i (kp , M ); so if i > 0 or M = A, we may
take Xp = 1.

Therefore if S and A are infinite, among the groups P i (GS , A) only

P (GS , A) is locally compact because H 1 (kp , A) is finite by (7.2.9). Con-
1
sidering the cohomology of A0 , we see that Hnr 0

(kp , A0 ) = H 0 (kp , Ad ) is an
open compact subgroup of the locally compact group H 0 (kp , A0 ). By (1.1.13),
we conclude that P 0 (GS , A0 ) is locally compact. Since H 1 (kp , A0 ) is finite,
P 1 (GS , A0 ) is also locally compact.
Let us finally mention the special case when A is ZZ-free. Then, for all p
such that A is an unramified Gkp -module, we have
1
Hnr (kp , A) = H 1 (kp , A) and i
Hnr (kp , A0 ) = 0, i ≥ 1,
see (7.2.16). Therefore, if A is ZZ-free, P 1 (GS , A) is a product of finite groups,
hence compact, and P i (GS , A0 ), i ≥ 1, is discrete, being a direct sum of discrete
groups.
Again, let A ∈ ModS (GS ) be arbitrary. For each p, the local duality theorem
(7.2.9) together with (7.2.10) induces a continuous composite homomorphism
for i = 1, 2
H 2−i (kp , A0 ) −→ H 2−i (kp , AD ) −→ H i (kp , A)∨ .
This is an isomorphism of finite groups for i = 1, and for i = 2 it is an injection
with dense image of a locally compact group into a compact group. Dualizing,
we obtain continuous homomorphisms for i = 1, 2
H i (kp , A) −→ H 2−i (kp , A0 )∨ .
This is an isomorphism of finite groups for i = 1. For i = 2 it is an injection
with dense image of a discrete group into a locally compact group. If A
is unramified at p, then, by (7.2.15), this map induces a homomorphism of
subgroups
i
Hnr (kp , A) −→(H 2−i (kp , A0 )/Hnr
2−i
(kp , A0 ))∨ ,
which is an isomorphism for i = 1. By the compatibility of Pontryagin
duality with restricted products, cf. (1.1.13), we therefore obtain continuous
homomorphisms for i = 1, 2.
Ξi : P i (GS , A) −→ P 2−i (GS , A0 )∨ .
From the above considerations, we obtain the following
(8.5.3) Proposition. Let A ∈ ModS (GS ). Then for i = 1, 2 the homomorphism

Ξi is continuous and
(i) Ξ1 is an isomorphism of locally compact groups,
(ii) Ξ2 is a continuous injection with dense image.

We will now compare the groups P i (GS , A0 ) for A ∈ ModS (GS ) with the
cohomology groups H i (GS , IS (A)) of the GS -module
IS (A) = Hom(A, IS ),
where IS is the S-idèle group of kS .
Let Gp = G(kS,p |k) be the decomposition group of p in G(kS |k) with respect
to some prolongation p̄ of p to kS , see the discussion at the beginning of
this section. If K|k is a finite Galois subextension of kS with Galois group
G = G(K|k), then let GP = G(KP |kp ) where P = p̄|K . Finally, let
×
IP (A) = Hom(A, KP ) and UP (A) = Hom(A, UK,P ),
where UK,P is the group of units in KP . In the limit we use the notation
×
Ip (A) = Hom(A, kS,p ) and Up (A) = Hom(A, UkS ,p ).
Now let K|k be large enough so that G(kS |K) acts trivially on A. Then
IS (A)G(kS |K) = Hom(A, IS )G(kS |K) = Hom(A, IK,S )
and Y G ×
IK,S = IndḠP (KP ).
p∈S
The functor Hom(A, −) commutes with , since it commutes with and with
Q Q
lim
−→
, recalling that A is finitely generated, and hence with restricted products.
Thus we obtain a bijective homomorphism
×
ΨK : H i (G, Hom(A, IK,S )) −→ H i (GP , Hom(A, KP
Y
∼ )),
p∈S
where the restricted product is taken with respect to the subgroups

im H i (GP , UP (A)) −→ H i (GP , IP (A)) .
Note that ΨK is continuous. For i = 0 it is homeomorphic, but not for i > 0, as
H i (G, Hom(A, IK,S )) carries the discrete topology by definition. Taking the
direct limit over K, we obtain a commutative diagram
y{|z
lim H i (G, Hom(A, ∼ ×
Y
IK,S )) lim H i (GP , Hom(A, KP ))
−→ −→
K K p∈S
o inf (infp )p∈S
Y
H i (GS , IS (A)) H i (kp , A0 ) = P i (GS , A0 ) .
p∈S
As any finite extension K|k is unramified at almost all primes, the restricted
×
product p∈S H i (GP , Hom(A, KP
Q
)) remains unchanged if we take it with
respect to the subgroups

im H i (GP /H P , UP (A)H P ) −→ H i (GP , IP (A)) ,
where H P is the inertia subgroup of GP . Therefore the right vertical arrow
exists and is continuous. As the left vertical map is a topological isomorphism,
we obtain the dotted continuous map from H i (GS , IS (A)) into P i (GS , A0 ).

(8.5.4) Definition. The continuous homomorphisms

shi : H i (GS , IS (A)) −→ P i (GS , A0 ) , i ≥ 0,
are called Shapiro maps.
(8.5.5) Proposition. Let A ∈ ModS (GS ). Then the following hold:

(i) sh0 is a bijection for function fields. If k is a number field, then sh0 is
surjective with kernel
Nk̄p |kp Hom(A, C× ) .
Y
ker sh0 =
p∈S∞ (k)
(ii) sh1 is injective, and bijective if A is ZZ-free.

(iii) If A is ZZ-free, then sh2 is injective and for each ` ∈ IN(S) the induced
map on `-torsion subgroups
H 2 (GS , IS (A))(`) −→ P 2 (GS , A0 )(`)
is an isomorphism.
Remarks: 1. For ZZ-free A, the bijection sh1 of (ii) is a homeomorphism, as

both groups are discrete. The same is true for the isomorphism stated in (iii).
2. In (10.11.9) we will show the following additional facts for an arbitrary
A ∈ ModS (GS ):
• sh2 is injective and induces an isomorphism on the `-torsion part for all
` ∈ IN(S),
• if S is finite, then sh1 is bijective.
This will follow with the same proof as below using the additional fact that the
local fields (kS )p are `-closed for ` ∈ IN(S) and p ∈ S(kS ); see (10.11.6).
Proof: We start with the case i = 0. Let K|k be a finite Galois subextension
of kS such that G(kS |K) acts trivially on A. Putting G = G(K|k), we obtain
H 0 (GS , IS (A)) = Hom(A, IK,S )G
Y G
×
= IndG
G Hom(A, KP )
P
p∈S
× GP
Y
= Hom(A, KP )
p∈S
Y Y
= Ip (A)GP = H 0 (Gp , Ip (A)) .
p∈S p∈S
Let Np = G(k̄p |kS,p ), i.e. Gp = Gkp /Np . Then

H 0 (kp , A0 ) = (Hom(A, k̄p× )Np )Gp = Hom(A, kS,p
× Gp
) = H 0 (Gp , Ip (A)).

If A is unramified at p - ∞, i.e. the inertia group with respect to p acts trivially

on A, then we get Hnr 0
(kp , A0 ) = Hnr
0
(Gp , Ip (A)) in the same way. Therefore
sh is surjective, and recalling the difference between H 0 and Ĥ 0 for p ∈ S∞ ,
0
we obtain the description of ker sh0 .

Since
lim H i (G(K|k), Hom(A, IK,S )) = H i (GS , IS (A))
−→
K
and (inf p )p∈S is injective for i = 1, we immediately obtain that sh1 is injective,
i.e. the first assertion of (ii).
Now let A be ZZ-free. Since A ∈ ModS (GS ), we have an isomorphism
A∼ = ZZr for some r, of trivial G(k̄p |(kS )p )-modules. Using Hilbert’s Satz 90,
we obtain
H 1 (G(k̄p |(kS )p ), Hom(A, k̄ × )) ∼
p = H 1 (G(k̄p |(kS )p ), k̄ × )r = 0.
p
The Hochschild-Serre spectral sequence induces an isomorphism

H 1 (G((kS )p |kp ), Hom(A, (kS )× ∼ 1 ×
p )) → H (G(k̄p |kp ), Hom(A, k̄p ))
and an injection
H 2 (G((kS )p |kp ), Hom(A, k̄p× )) ,→ H 2 (G(k̄p |kp ), Hom(A, k̄p× )).
Using the commutative diagram
H 2 (G((kS )p |kp}~ ), Hom(A, (kS )× H 2 (G(k̄p |kp ), Hom(A, k̄p× ))
M M
p ))
p∈S p∈S
sh 2
H 2 (GS , IS (A)) P 2 (GS , A0 ),
we obtain the first assertion of (iii). Now let ` ∈ IN(S). Then `∞ | G((kS )p |kp )
for all nonarchimedean primes p ∈ S. As kS is totally imaginary, we obtain
H i (k̄p |(kS )p , A0 )(`) = 0 for all p ∈ S and i ≥ 1. Hence the horizontal injection
in the last diagram is an isomorphism on the `-torsion subgroups. This finishes
the proof of (iii).
Finally, we show (ii). For an arbitrary A ∈ ModS (GS ) we consider the
commutative diagram
H 1 (G((kS )p |kp ), Hom(A, (kS )× H 1 (G(k̄p |kp ), Hom(A, k̄p× ))
Y Y
p ))
p∈S p∈S
sh 1
H 1 (GS , IS (A)) P 1 (GS , A0 ),
which shows that sh1 is injective. If A is ZZ-free, then P 1 (GS , A0 ) maps iso-
morphically to the direct sum p∈S H 1 (G(k̄p |kp ), Hom(A, k̄p× )) and the same
L
is true for H 1 (GS , IS (A)). Hence sh1 is an isomorphism if A is ZZ-free. This

finishes the proof of (ii). 2

§6. Poitou-Tate Duality
We proved in (7.2.9) a cohomological duality theorem for local fields and

in (8.4.4) an analogous duality theorem for global fields. The latter, however,
does not play the same important role for global fields as its local prototype
for local fields. But both together lead to the main result of arithmetic Galois
cohomology: a duality principle for a local-global statement. The present
section is devoted to the formulation and the proof of this main theorem.
Recall that GS = G(kS |k) is the Galois group of the maximal extension kS |k
which is unramified outside the given nonempty set of primes S, where S∞ is
contained in S if k is a number field. ModS (GS ) is the category of discrete
GS -modules which are finitely generated as ZZ-modules and whose torsion has
order #tor(A) ∈ IN(S), and
A0 = Hom(A, OS× )
is the dual GS -module of A ∈ ModS (GS ), where OS× is the group of S-units
of kS . Further recall the convention that the cohomology groups H ∗ (kp , A0 )
are defined by first considering A as a local Galois module and then taking the
dual in the local sense, i.e. applying Hom(−, k̄p× ).
(8.6.1) Proposition. Let A ∈ ModS (GS ) and let M be equal to A or A0 . Then

the homomorphism
res i : H i (GS , M ) → H i (kp , M )

Y
p∈S
Y
maps H i (GS , M ) into P i (GS , M ) = H i (kp , M ).
p∈S
Proof: First let M = A. Since A is finitely generated, it becomes a trivial Ga-

lois module over a finite Galois subextension LA |k of kS |k. If x ∈ H i (GS , A),
then there exists a finite Galois subextension K|k of kS |k such that x is the
image of an element y ∈ H i (K|k, AGK ) under the inflation map. It follows
that for all nonarchimedean primes p of S which are unramified in KLA |k, the
i
class xp = resp x is contained in the subgroup Hnr (kp , A) of H i (kp , A).
Now let M = A0 and x ∈ H i (GS , A0 ). As above we have x = inf y with
y ∈ H i (K|k, (A0 )GK ) for a finite Galois subextension K|k of kS |k. For all
nonarchimedean primes p of S which are unramified in KLA |k, the exten-
sion Kp |kp is a subextension of the maximal unramified extension k̃p of kp .

§6. Poitou-Tate Duality 481
Therefore the class xp = resp x is contained in the image of

H i (k̃p |kp , Hom(A, OS× )Gk̃p ) = H i (k̃p |kp , Hom(A, (OS× )Gk̃p ))
in H i (kp , A0 ), i.e. in Hnr
i
(kp , A0 ) = Hnr
i
(kp , Ad ) = Hnr
i
(kp , Hom(A, Õp× )). This
proves the proposition. 2
(8.6.2) Definition. The homomorphisms

λi : H i (GS , M ) −→ P i (GS , M )
for M = A and M = A0 , which exist by (8.6.1), are called the localization
maps. We set
Xi (GS , M ) = ker(λi ) . ∗)
It is evident that for M = A0 , the map λi is the composition

shi
H i (GS , A0 ) −→ H i (GS , IS (A)) −→ P i (GS , A0 ),
where the first arrow is induced by A0 = Hom(A, OS× ) → Hom(A, IS ) = IS (A),
and the second is the Shapiro map.
The groups Xi (GS , M ) are of great interest. Their vanishing would imply a
strict “local-global principle”. As a rule, the groups Xi (GS , M ) are non-zero,
but we shall see that they are always finite.
(8.6.3) Lemma. For m ∈ IN(S), we have natural isomorphisms

X1 (GS , ZZ/mZZ) ∼
= Hom(ClS (k), ZZ/mZZ) ,
X1 (GS , ZZ0 ) ∼
= ClS (k) .
Proof: By class field theory, ClS (k) is naturally isomorphic to the Galois
group of the maximal abelian extension of k inside kS in which all primes of S
split completely. Hence

X1 (GS , ZZ/mZZ) = ker H 1 (GS , ZZ/mZZ) −→
Y
H 1 (kp , ZZ/mZZ)
p∈S
∼
= Hom(ClS (k), ZZ/mZZ) .
Furthermore, X1 (GS , ZZ0 ) = H 1 (GS , OS× ) ∼
= ClS (k) by Hilbert’s Satz 90 and
(8.3.11)(ii). 2
∗) If A is the G -module A(k ) of k -rational points of an abelian variety A over k, then
S S S
this kernel is classically called the Šafarevič-Tate group. This explains the Russian letter X
(sha).


0 −→ OS× −→ IS −→ CS −→ 0,
we obtain a sequence of GS -modules
0 −→ A0 −→ IS (A) −→ CS (A) −→ 0 ,
which is exact as Ext1ZZ (A, OS× ) = 0 because OS× is #tor(A)-divisible by (8.3.4).
(8.6.4) Theorem. Let A ∈ ModS (GS ). Then the boundary map

δ
CS (A)GS −→ H 1 (GS , A0 )
is continuous and has finite image. It maps a norm group NG/U CS (A)U to zero
for some open normal subgroup U ⊆ GS . The group X1 (GS , A0 ) is finite.
Proof: First note that im(δ) = X1 (GS , A0 ), since

sh1 : H 1 (GS , IS (A)) −→ P 1 (GS , A0 )
is injective by (8.5.5)(ii). Let K = kSH be a finite Galois extension of k
inside kS which trivializes A and µn , where n = #tor(A). Furthermore, let
K1 = kSH1 be a finite Galois extension of k inside kS which contains the
maximal unramified abelian extension of K in which all primes p ∈ S(K) are
completely decomposed. As an H-module, A is isomorphic to a direct sum of
modules of the form ZZ or µm with m|n. We have X1 (H, ZZ0 ) = ClS (K) and
X1 (H, (µm )0 ) = Hom(ClS (K), ZZ/mZZ)
by (8.6.3), and we also have the corresponding statement for the group H1 . By
class field theory and the principal ideal theorem, we conclude that the maps
0 0 H1 0 0
res H 1 1 1 1
H1 : X (H, A ) → X (H1 , A ), cor H : X (H1 , A ) → X (H, A )
are zero. The commutative diagram

δH1
H1
IS (A) CS (A)H1 X1 (H1 , A0 ) 0
NH/H1 NH/H1 0
δH
IS (A)H CS (A)H X1 (H, A0 ) 0
NGS /H NGS /H cor H
GS
δ
IS (A)GS CS (A)GS X1 (GS , A0 ) 0
shows that δ annihilates the subgroup NGS /H1 CS (A)H1 ⊆ CS (A)GS . Further-
more, the diagram

δH 1
¡¢ H1
IS (A) CS (A)H1 X1 (H1 , A0 ) 0
res H
H1 res H
H1 0
δH
IS (A)H CS (A)H X1 (H, A0 ) 0
G G G
res HS res HS res HS
δ
IS (A)GS CS (A)GS X1 (GS , A0 ) 0
shows that CS (A)GS is a subgroup of the group
B := im(IS (A)H1 −→ CS (A)H1 ).
By (8.3.5), the image CK1 ,S of the homomorphism IK1 ,S → CS (K1 ) is open of
×
finite index. Its kernel OK 1 ,S
is a discrete subgroup of IK1 ,S . For m ∈ IN(S),
Hom(µm , IK1 ,S ) is compact. As X1 (H1 , ZZ/mZZ) = Hom(ClS (K1 ), ZZ/mZZ) is
finite, the image of the homomorphism
Hom(µm , IK1 ,S ) −→ Hom(µm , CS (K1 ))
is a closed subgroup of finite index, hence open. The kernel of this homomor-
×
phism is the finite group Hom(µm , OK 1 ,S
) = ZZ/mZZ. Therefore B ⊆ CS (A)H1
is open of finite index and
H1
0 −→ A0 −→ IS (A)H1 −→ B −→ 0
is an exact sequence of locally compact groups with continuous GS /H1 -action.
Furthermore, there exists a GS /H1 -invariant open neighbourhood U of 1 in
IS (A)H1 which maps homeomorphically onto its image U 0 in B. Let V ⊆ U
be a G/H1 -invariant open neighbourhood of 1 with V · V −1 ⊆ U and let V 0
be the homeomorphic image of V in B. The subset V 0 ∩ CS (A)GS is an open
neighbourhood of 1 in CS (A)GS .
Let x ∈ V 0 ∩ CS (A)GS . There exists a unique pre-image y ∈ V of x. The
inner derivation GS /H1 → IS (A)H1 , g 7→ y −1 gy, takes values in A0 H1 and
induces a derivation cy : GS /H1 → A0 H1 . The class c̄y ∈ H 1 (GS /H1 , A0 H1 )
of cy maps to δ(x) ∈ H 1 (GS , A0 ) under inflation. For g ∈ GS /H1 , we have
y −1 gy ∈ U ∩ A0 H1 = {1}. Hence δ(x) = 0. This shows that the subgroup
ker(δ) contains an open neighbourhood of 1 in CS (A)GS and is therefore
open. Consequently, δ is locally constant. Furthermore, δ factors through the
compact quotient CS (A)GS /NGS /H1 CS (A)H1 . Therefore im(δ) = X1 (GS , A0 )
is finite. 2
In the following we want to construct a canonical non-degenerate pairing

between the groups X1 (GS , A0 ) and X2 (GS , A) for each finitely generated
GS -module A ∈ ModS (GS ). We need two lemmas. As before, write
A0 = Hom(A, OS× ) , IS (A) = Hom(A, IS ) , CS (A) = Hom(A, CS ).

(8.6.5) Lemma. Let A ∈ ModS (GS ) and i = 1 or i = 2. Then the homomor-

phism
λi Ξi
ξ i : H i (GS , A) −→ P i (GS , A) −→ P 2−i (GS , A0 )∨ ,
where Ξi was defined in (8.5.2), has kernel ker ξ i = Xi (GS , A).
Proof: By (8.5.3), the homomorphisms Ξi , i = 1, 2, are injective. 2
Recall that P 2−i (GS , A0 )∨ is a locally compact group for i = 1, 2. Therefore

we can dualize the composite map ξ i = Ξi ◦ λi for i = 1, 2 (but not Ξ2 , since
“P 2 (GS , A)∨ ” does not exist at least if S and A are infinite).
(8.6.6) Lemma. For i = 0 and i = 1, we have a commutative diagram of

topological groups
H i (GS ,¦¥¤£ IS (A)) can
H i (GS , CS (A))
shi ∆i
(ξ 2−i )∨
P i (GS , A0 ) H 2−i (GS , A)∨ ,
where ∆i is induced by the cup-product pairing

∪ inv
H i (GS , CS (A)) × H 2−i (GS , A) −→ H 2 (GS , CS ) −→ Q/ZZ.
The map (ξ 2−i )∨ on the bottom is the dual to the map ξ 2−i obtained in (8.6.5).
Proof: Let p ∈ S be a prime. We make the convention that H 0 (kp , −) denotes

Ĥ 0 (kp , −) if p is archimedean. Using the notation of the preceding section, we
have the commutative diagram
±°²¯®¬«ª©§¨
H i (kp , Hom(A, k̄p× )) × H 2−i (kp , A) ∪
H 2 (kp , k̄p× )
invp
Q/ZZ
inf inf
× ∪ ×
H i (Gp , Hom(A, kS,p )) × H 2−i (Gp , A) H 2 (Gp , kS,p )
resp resp resp
∪ inv
H i (GS , Hom(A, IS )) × H 2−i (GS , A) H 2 (GS , IS ) Q/ZZ .
For x ∈ H i (GS , IS (A)), we therefore obtain a commutative diagram

º·¶¸¹³´µ S , A) H i (kp , A0 )∨
Y Y
H 2−i (G H 2−i (kp , A)
p∈S p∈S
∪x ∪shi (x)
⊕invp M
H 2 (GS , IS ) Q/ZZ
p∈S
Σ
inv
H 2 (GS , CS ) Q/ZZ .
The map from H 2−i (GS , A) to Q/ZZ via the upper right corner is exactly
(ξ 2−i )∨ ◦ shi (x) ∈ H 2−i (GS , A)∨ , while the map via the lower left corner is
∆i ◦ can(x). This shows that the diagram of the lemma commutes. 2
Now we are ready to prove one of the main results of this section.
(8.6.7) Theorem (Poitou-Tate Duality). Let A be a finitely generated GS -

module with #tor(A) ∈ IN(S). Then there is a perfect pairing ∗)
X1 (GS , A0 ) × X2 (GS , A) −→ Q/ZZ
of finite groups, which is induced by the cup-product, i.e. the diagram
H 0 (GS ,»¼½¾¿ CS (A)) × H 2 (GS , A) ∪

H 2 (GS , CS ) Q/ZZ
X1 (GS , A0 ) × X2 (GS , A)
commutes.
Proof: Consider the diagram with exact upper row

ÉÈÂÇÅÆÄÃÁÀ GS
IS (A)
γ
CS (A)GS δ
H 1 (GS , A0 ) i
H 1 (GS , IS (A))
(1) sh0 ∆0 sh1
(ξ 2 )∨ ε λ1
P 0 (GS , A0 ) H 2 (GS , A)∨ H 1 (GS , A0 ) P 1 (GS , A0 ) .
∗) A pairing A × B → Q/ZZ is perfect if it induces isomorphisms A →

∼ ∼
B ∨ and B → A∨ .

The map ∆0 is the composition of the global duality map (8.4.4), (8.4.5),
∼ H 2 (G , A)∨ ,
∆ˆ 0 : Ĥ 0 (GS , CS (A)) −→ S
with the canonical homomorphism H 0 (GS , CS (A)) → Ĥ 0 (GS , CS (A)) (which

has dense image). The left-hand square is commutative by (8.6.6) and the right-
hand one by the remark following (8.6.2). By (8.6.4), the homomorphism δ
maps a norm group NGS /U CS (A)U with U sufficiently small to zero, and we
obtain a commutative diagram
ÏÎÌÍÊË GS
CS (A) Ĥ 0 (GS , CS (A)) ∆ˆ 0
H 2 (GS , A)∨
∼
π ε
δ
CS (A)GS /NGS /U CS (A)U H 1 (GS , A0 )
where ε = δ ◦ π ◦ (∆ˆ 0 )−1 . This map ε completes diagram (1). We have

im(ε) = im(δ) = ker(i) = ker(λ1 ) = X1 (GS , A0 ) (observe that sh1 is injective
by (8.5.5)(ii)). By (8.5.5)(i), the map sh0 is surjective. Since
ε(ξ 2 )∨ sh0 = δπ(∆ˆ 0 )−1 (ξ 2 )∨ sh0 = δπ(∆ˆ 0 )−1 ∆0 γ = δγ = 0 ,
we obtain ε(ξ 2 )∨ = 0, i.e. we have an inclusion im((ξ 2 )∨ ) ⊆ ker(ε). Since
ε is continuous and has finite image by (8.6.4) and (8.4.4), ker(ε) is an open
subgroup in H 2 (GS , A)∨ . In particular, ker(ε) contains the closure of im((ξ 2 )∨ )
(note that im((ξ 2 )∨ ) is closed if A is finite or if k is a number field).
Let x ∈ ker(ε) and let V ⊆ ker(ε) be an arbitrarily chosen open sub-
group. As ∆0 has dense image by (8.4.5), there exists y ∈ CS (A)GS with
∆0 (y) ∈ x + V . Hence δ(y) = ε∆0 (y) = 0. Thus there exists an element
z ∈ IS (A)GS such that γ(z) = y. It follows that ∆0 (y) = (ξ 2 )∨ (sh0 (z)), and so
x + V contains an element of im((ξ 2 )∨ ). As H 2 (GS , A)∨ is profinite, it has a
basis of neighbourhoods of zero consisting of subgroups. Therefore im((ξ 2 )∨ )
is dense in ker(ε) and the induced map ker(ε)∨ → P 0 (GS , A0 )∨ is injective.
We conclude that ξ 2 factors as
ξ 2 : H 2ÐÑ (GS , A) ker(ε)∨ P 0 (GS , A0 )∨ .
This implies that there is an isomorphism
X1 (GS , A0 )∨ = im(ε)∨ ∼
= ker(ξ 2 ) = X2 (GS , A),
which obviously is induced from the diagram
ÕÓÔÒ H 0 (GS , CS (A)) × H 2 (GS , A) ∪
Q/ZZ
δ
X1 (GS , A0 ) × X2 (GS , A) . 2

For finite modules we can describe the pairing obtained in (8.6.7) explicitly.
Let A ∈ ModS (GS ) be finite and let us consider the “new” pairing
t
X1 (GS , A0 ) × X2 (GS , A) −→ Q/ZZ
which is defined as follows: let x and x0 be cocycles representing classes
[x] ∈ X2 (GS , A) and [x0 ] ∈ X1 (GS , A0 ). As H 3 (GS , OS× )(p) = 0 for every
prime number p ∈ IN(S) (see (8.3.11)), there is a cochain z ∈ C 2 (GS , OS× ) such
that
x0 ∪ x = ∂z.
Moreover, for every p there are cochains yp ∈ C 1 (kp , A) and yp0 ∈ C 0 (kp , A0 )
such that for the components xp = λp (x), x0p = λp (x0 ) we have
xp = ∂yp and x0p = ∂yp0 .
Then yp0 ∪ xp − zp and x0p ∪ yp − zp are 2-cocycles with values in k̄p× which
differ by a coboundary. In fact,
∂(yp0 ∪ xp ) = ∂yp0 ∪ xp = x0p ∪ xp = ∂zp
and analogously ∂(x0p ∪ yp ) = ∂zp , and moreover
∂(yp0 ∪ yp ) = x0p ∪ yp − yp0 ∪ xp .
The “new” pairing is now defined by
[x0 ] t [x] = invp [yp0 ∪ xp − zp ] = invp [x0p ∪ yp − zp ].
X X
p∈S p∈ S
One checks in a straightforward manner that this definition does not depend on
the choice of the representing cocycles x and x0 and the choice of the cochains
z, yp and yp0 .
(8.6.8) Proposition. The pairing defined above coincides with the pairing
obtained in (8.6.7).
Proof: Let [x0 ] ∈ X1 (GS , A0 ) and [x] ∈ X2 (GS , A). The set of 0-
cochains yp0 ∈ H 0 (kp , A0 ) with ∂yp0 = x0p , p ∈ S, can be interpreted as
a 0-cochain in H 0 (GS , IS (A)). Under the projection IS → CS the element
(yp0 ) ∈ H 0 (GS , IS (A)) becomes a 0-cochain y 0 ∈ H 0 (GS , CS (A)) such that
δy 0 = [x0 ] and the value of the pairing in (8.6.7) is inv[y 0 ∪ x]. Consider the
commutative diagram (8.6.6)
H 0 (GS ,Ö×ØÙ IS (A)) H 0 (GS , CS (A))
P 0 (GS , A0 ) H 2 (GS , A)∨ .

With the notation of the “new” pairing, we obtain

[x0 ] t [x] = invp [yp0 ∪ xp − zp ] = inv[y 0 ∪ x] ,

X
p∈S
because the image of the 2-cochain (zp ) ∈ C 2 (GS , OS× ) in H 2 (GS , CS ) is zero.
2
For A ∈ ModS (GS ), we have constructed a natural perfect pairing between

X1 (GS , A0 ) and X2 (GS , A). If A is finite, we can interchange the roles of
A and A0 to obtain a perfect pairing between X1 (GS , A) and X2 (GS , A0 ) as
well. If A is infinite, the situation is no longer symmetric. If A is ZZ-free, i.e.
A0 is a Spec(Ok,S )-torus, we have the following
(8.6.9) Theorem. Let A ∈ ModS (GS ) be torsion-free. Then there is a perfect

pairing of finite groups
X1 (GS , A) × X2 (GS , A0 ) −→ Q/ZZ ,
which is induced by the cup-product, i.e. the diagram
H 1 (GÛÜÝÞÚ S , A) × H 1 (GS , CS (A)) ∪

H 2 (GS , CS ) Q/ZZ
X1 (GS , A) × X2 (GS , A0 )
commutes.
Proof: Since H 1 (GS , A) is a finite group by (8.3.20)(ii), its subgroup

X1 (GS , A) is also finite. By (8.6.6), the diagram
H 1 (GS ,èçæäåãáâßà IS (A)) H 1 (GS , CS (A)) δ

H 2 (GS , A0 ) H 2 (GS , IS (A))
sh1 ∆1 sh2
(ξ 1 )∨ ϕ λ2
P 1 (GS , A0 ) H 1 (GS , A)∨ H 2 (GS , A0 ) P 2 (GS , A0 )
commutes. The map ∆1 is an isomorphism by (8.4.4), and ϕ is defined so that
the square in the middle commutes. Obviously, the upper sequence is exact.
By (8.5.5)(ii), (iii) the map sh1 is bijective and sh2 is injective. Therefore
the lower sequence in the commutative diagram above is also exact. Since
coker (ξ 1 )∨ = X1 (GS , A)∨ by (8.6.5), we obtain X1 (GS , A)∨ ∼
= X2 (GS , A0 ).
Finally, the associated pairing to Q/ZZ is induced by the cup-product, as stated.
This finishes the proof of the theorem. 2

Remarks: 1. If S is a finite set containing the set Sp for some prime number p
and A ∈ ModS (GS ) is arbitrary, then one can prove the duality between
X1 (GS , A) and X2 (GS , A0 ) at least for the p-components of these groups;
see (10.11.10).
2. The duality between Xi (GS , A) and X3−i (GS , A0 ), i = 1, 2, proven above
for torsion-free A (resp. for a torus), has a vast generalization if S is the set of
all primes: it holds for 1-motives; see [76].
3. If S is the set of all places, then (8.6.9) holds for an arbitrary finitely gener-
ated Gk -module A. This can be seen as follows (statement and proof are due
to J.-L. COLLIOT-THÉLÈNE): There exists an exact sequence 0 → C → B →
A → 0 with B torsionfree and C ∼ = ni=1 IndUGik ZZ, where Ui ⊆ Gk are open sub-
L
groups; see [CS], Lemma 0.6. As the groups H 1 (k, C), H 1 (k, C 0 ), X2 (k, C)
and X2 (k, C 0 ) vanish, we obtain isomorphisms X1 (k, B) ∼ = X1 (k, A) and
X2 (k, B 0 ) ∼
= X2 (k, A0 ), and so the statement for general A follows from that
in the torsionfree case. This argument extends to the case when S has Dirichlet
density 1.
We proved a duality theorem for global fields in (8.4.4) which, together

with its analogue (7.2.6) for local fields, leads to the main result of arithmetic
Galois cohomology: a 9-term exact sequence connecting the local and global
cohomology groups.
(8.6.10) Long Exact Sequence of Poitou-Tate. Let S be a nonempty set of

primes of a global field k and assume that S ⊇ S∞ if k is a number field. Let
A be a finite GS -module of order #A ∈ IN(S).
(i) There is a canonical exact sequence of topological groups
0éêëìíîïðñòóô H 0 (GS , A) P 0 (GS , A) H 2 (GS , A0 )∨ H 1 (GS , A)

(finite) (compact) (compact) (discrete)
(locally compact) P 1 (GS , A)
0 H 0 (GS , A0 )∨ P 2 (GS , A) H 2 (GS , A) H 1 (GS , A0 )∨

(finite) (discrete) (discrete) (compact)
(ii) For i ≥ 3, the restriction map

H i (GS , A) −→ H i (kp , A)
M
∼
p∈SIR
is an isomorphism.

Remark: Some people consider it more suggestive to arrange the terms of the
exact 9-term sequence in the following form.
0õö÷øùúûüýþÿ H 0 (GS , A) P 0 (GS , A) H 2 (GS , A0 )∨
H 1 (GS , A) P 1 (GS , A) H 1 (GS , A0 )∨
H 2 (GS , A) P 2 (GS , A) H 0 (GS , A0 )∨ 0.
Proof of (8.6.10)(ii): For function fields the assertion is obviously true by

(8.3.17), so we may assume that k is a number field. If A = 0 the statement is
trivial. Therefore we may assume that S ⊇ Sp for at least one prime number p.
In particular, kS is totally imaginary.
As `∞ | #GS for all ` ∈ IN(S), (1.9.15) implies Ĥ −1 (GS , A0 ) = 0. From the
exact sequence
0 −→ A0 −→ IS (A) −→ CS (A) −→ 0
of level-compact GS -modules (observe that CS (A) = Hom(A, CS0 )) we obtain
the exact sequence
0 → Ĥ −1 (GS , IS (A)) → Ĥ −1 (GS , CS (A)) → Ĥ 0 (GS , A0 ) → Ĥ 0 (GS , IS (A))
by (1.9.13). Again by (1.9.15), we have Ĥ 0 (GS , A0 ) = A0GS . For p ∈ S put
Gp = G(kS,p |kp ). For p ∈ S r S∞ and ` ∈ IN(S) we have `∞ | #Gp and (1.9.15)
implies Ĥ 0 (Gp , A0 ) = A0Gp . We obtain
Ĥ 0 (GS , IS (A)) ∼ A0Gp × Ĥ 0 (Gp , A0 ).
Y Y
=
p∈S\S∞ p∈S∞
Therefore the map Ĥ 0 (GS , A0 ) → Ĥ 0 (GS , IS (A)) is injective, yielding the iso-
morphism
Ĥ −1 (GS , IS (A)) −→ ∼ Ĥ −1 (G , C (A)).
S S
By (1.9.15), we have Ĥ −1 (Gp , A0 ) = 0 for all p ∈ S r S∞ and therefore a

Ĥ −1 (GS , IS (A)) ∼ Ĥ −1 (Gp , A0 ) ∼ Ĥ −1 (kp , A0 ).
Y M
= =
p∈S p∈SIR
Using the duality theorem for archimedean primes (7.2.17)
Ĥ −1 (kp , A0 ) ∼
= H 3 (kp , A)∨ ,
we obtain the commutative diagram
Ĥ −1 (GS , IS (A)) Ĥ −1 (GS , CS (A))
(λ3 )∨
H 3 (kp , A)∨
M
H 3 (GS , A)∨
p∈SIR

where the vertical isomorphism on the right-hand side is given by (8.4.4).

Hence M
λ3 : H 3 (GS , A) −→
∼ H 3 (kp , A).
p∈SIR
is an isomorphism.
In order to prove the assertion for all i ≥ 3 we use the exact sequence
0 −→ A −→ IndG A −→ B −→ 0,
where G = G(K|k), for K a finite totally imaginary extension of k inside
kS such that A is a trivial G(kS |K)-module, and the finite module B is de-
fined as the quotient of IndG A by A. For i ≥ 3 we have H i (GS , IndG A) =
H i (G(kS |K), A) = 0 and
H i (kp , IndG A) = H i (KP , A) = 0, p ∈ SIR .
M
P|p
Therefore the horizontal maps in the commutative diagram
H i (G S , B) H i+1 (GS , A)
λi λi+1
H i (kp , B) H i+1 (kp , A)

M M
p∈SIR p∈SIR
are isomorphisms for i ≥ 3. Now the result follows by induction. 2
In order to prove part (i) of theorem (8.6.10), we start with the following
observations.
(8.6.11) Proposition. For A ∈ ModS (GS ), the localization map

λ1 : H 1 (GS , A) −→ P 1 (GS , A)
is proper.∗)
Proof: Since A is finitely generated, A is a GT -module for all sufficiently

large finite subsets T ⊆ S. The groups
Y
PT = P 1 (GT , A) × 1
Hnr (kp , A)
p∈S\T
are compact, and each compact neighbourhood of 1 in P 1 (GS , A) is con-
tained in some PT for T ⊆ S finite. Therefore we have to show that
(λ1 )−1 (PT ) ⊆ H 1 (GS , A) is compact and, since H 1 (GS , A) is discrete, this
comes down to showing that (λ1 )−1 (PT ) is finite. Consider the canonical
injections
∗) A map of topological spaces is called proper if the pre-images of compact subsets are
compact.

M M diag Y
H 1 (kp , A)/Hnr
1
(kp , A) ,→ H 1 (Tp , A) ,→ H 1 (TP , A),
p∈S\T (k) p∈S\T (k) P∈S\T (kT )
where Tp denotes the inertia group of the local group G(k̄p |kp ) for a prime p.
We obtain the commutative and exact diagram
H 1 (G(kS
|kT ), A)
Y
H 1 (TP , A)
P∈S\T (kT )
H 1 (GS , A) P 1 (GS , A)
Y
H 1 (GT , A) P 1 (GT , A) × 1
Hnr (kp , A) = PT .
p∈S\T (k)
Since A is a trivial G(kS |kT )-module and since the images of the inertia groups
TP , P ∈ S r T (kT ), generate G(kS |kT ) as a normal subgroup, the upper
horizontal map is injective. Thus
(λ1 )−1 (PT ) ⊆ H 1 (GT , A) .
We will show that H 1 (GT , A) is a finite group. Recalling (8.3.20), we may
assume that A is torsion-free. Then the exact sequence
0 −→ H 1 (GT /H, A) −→ H 1 (GT , A) −→ H 1 (H, A) = 0,
where H is an open normal subgroup of GT acting trivially on A, shows that
H 1 (GT , A) is finite. 2
(8.6.12) Lemma. Let A ∈ ModS (GS ) be finite. Then there is a commutative

exact diagram
C× )
ϕ
Y
Nk̄p |kp Hom(A, NGS CS (A)
p∈S∞
δ
H 0 (GS , IS (A)) H 0 (GS , CS (A)) H 1 (GS , A0 ) H 1 (GS , IS (A))
ε
where the map ϕ is an isomorphism. We have a natural exact sequence
0 → H 0 (GS , A0 ) → P 0 (GS , A0 ) → H 2 (GS , A)∨ → X1 (GS , A0 ) → 0 .

Proof: As we can represent kS as the union of a countable tower of finite

subextensions, we can apply the results of II §7 to projective systems indexed
by the set of finite extensions K of k inside kS . Consider the projective system
which attaches to each K the group H 0 (G(kS |K), A0 ) with the norm maps as
transition maps. If K trivializes A0 and L|K is a finite extension, then the
norm
NL|K : A0 = H 0 (G(kS |L), A0 ) −→ A0 = H 0 (G(kS |K), A0 )
is just multiplication by [L : K]. For all prime numbers ` | #A0 ∈ IN(S),
`∞ divides the order of GS (observe that k(µ`∞ ) ⊆ kS ). Therefore the above
projective system is a ML-zero system, and (2.7.3) implies
Ri (lim
←−
)H 0 (G(kS |K), A0 ) = 0, i≥0.
We let K run through the extensions of k inside kS which trivialize A0 and
consider the exact sequences
0 −→ A0 −→ Hom(A, IK,S ) −→ Hom(A, IK,S )/A0 −→ 0.
Passing to the projective limit over K and using the ML-zero property, we
∼ lim Hom(A, I 0
lim
←−
Hom(A, IK,S ) −→ ←−
K,S )/A .
K K
1
By (8.5.5) the map sh is injective, hence

X1 (G(kS |K), A0 ) = ker H 1 (G(kS |K), A0 ) −→ H 1 (G(kS |K), IS (A) .
Therefore the exact sequence
0 −→ A0 −→ IS (A) −→ CS (A) −→ 0
implies the exact sequence
0 −→ Hom(A, IK,S )/A0 −→ Hom(A, CS (K)) −→ X1 (G(kS |K), A0 ) −→ 0 .
Since
∨
1 0 2
lim X (G(kS |K), A ) = lim X (G(kS |K), A) = 0,
←− −→
K K
we obtain the isomorphism lim ∼ lim
Hom(A, IK,S ) → Hom(A, CK,S ). Con-
←− K ←− K
sider the commutative diagram
lim
Hom(A, IK,S ) lim Hom(A, CK,S )
←− ←−
K K
NGS IS (A) NGS CS (A).

As the GS -module CS (A) = CS0 (A) is level-compact, the right vertical map is
surjective, and hence the natural map
NGS IS (A) −→ NGS CS (A)

is also surjective. Let K be an extension of k inside kS trivializing A0 and

let p ∈ S r S∞ (K). For a sufficiently large finite extension L of K inside kS
(adjoin roots of unity) the norm map
Hom(A, L× ×
Y
NL|K : P ) −→ Hom(A, Kp )
P|p
is the zero map. Therefore the universal norms of IS (A) have trivial compo-
nents at the nonarchimedean primes and we obtain an isomorphism
Nk̄p |kp Hom(A, C× ) −→
Y
∼ N
GS IS (A) .
p∈S∞
We define ϕ as the composite of this isomorphism with the natural map

NGS IS (A) −→ NGS CS (A), which is surjective, as shown above.
Let us explain the other objects and maps in the diagram of the lemma.
The second row of the diagram is part of the long exact cohomology sequence
associated to the short exact sequence A0 ,→ IS (A) CS (A). The first arrow
in the third row is the dual of the localization map λ2 (A), while the third
arrow is the localization map λ1 (A0 ). For the commutativity of the left- and
right-hand lower squares, see (8.6.6) and the definition of the localization map
(8.6.2). The map ε is given by the perfect pairing between X1 (GS , A0 ) and
X2 (GS , A); see (8.6.7). In particular, the lower row is exact. The square in the
middle commutes by the construction of ε. The exactness of the two columns
follows from (8.5.5)(i) and (8.4.4).
Consider the commutative diagram
()*+,!"#$%&' Nk̄p |kp Hom(A, C× )
ϕ
Y
NGS CS (A)
p∈S∞
0 H 0 (GS , A0 ) H 0 (GS , IS (A)) H 0 (GS , CS (A))
0 H 0 (GS , A0 ) P 0 (GS , A0 ) H 2 (GS , A)∨

The third row is a complex; the middle row and the columns are exact. A
diagram chase shows that ϕ is an isomorphism and that the lower row is exact.
In order to prove the exactness of the 9-term sequence, we apply (8.6.12) to

A and A0 and get the two horizontal exact sequences in the diagram (where we
write G for GS )

0567843-./012 H 0 (G, A) P 0 (G, A) H 2 (G, A0 )∨ H 1 (G, A) P 1 (G, A)
(∗) ψ Ξ1
λ∨
0 H 0 (G, A0 )∨ P 2 (G, A) H 2 (G, A) H 1 (G, A0 )∨ P 1 (G, A0 )∨ .
We define the map ψ as the composition λ∨ ◦ Ξ1 . The pairings

H 1 (G;:9A?=><@ S , A) × H 1 (GS , A0 ) ∪
H 2 (GS , OS× )
λ λ0
∪ ⊕ invp M
P 1 (GS , A) × P 1 (GS , A0 ) H 2 (GS , IS ) Q/ZZ
p∈S
ι Σ
inv
H 2 (GS , CS ) Q/ZZ
show that im(λ) and im(λ0 ) annihilate each other. Therefore the map ψ induces
a 9-term complex which we denote by
P T (k, S, A),
and we know that it is exact at all places but the middle one, which remains
to be shown. If the set S is finite, then all groups occurring are finite and
the complex is exact if and only if the alternating product of the orders of the
groups in the complex is equal to 1. For a finite GS -module A, we define the
second partial Euler-Poincaré characteristic by
#H 0 (GS , A) · #H 2 (GS , A)
χ2 (GS , A) = ,
#H 1 (GS , A)
and put
Y #H 0 (kp , A)
ϕ(GS , A) = χ2 (GS , A)/ ,
p∈S∞ ||#A||p
where for function fields the set S∞ is empty by definition.
(8.6.13) Lemma. Let S be a finite nonempty set of primes of the global field k
containing S∞ if k is a number field. Let A be a finite GS -module of order
#A ∈ IN(S). Then
(i) ϕ(GS , A) · ϕ(GS , A0 ) = 1 if and only if the complex P T (k, S, A) is exact.
(ii) ϕ(GS , −) is multiplicative on short exact sequences of finite modules in
ModS (GS ).

Proof: Using the product formula (7.3.5) and (7.3.1), we obtain

Y #H 0 (kp , A) · #H 0 (kp , A0 ) Y #Ĥ 0 (kp , A)
=
p∈S∞ (||#A||p )2 p∈S∞ ||#A||p
Y Y
= ||#A||p #Ĥ 0 (kp , A)
p∈S\S∞ p∈S∞
Y Y #Ĥ 0 (kp , A) · #Ĥ 2 (kp , A)

= ||#A||p
p∈S\S∞ p∈S∞ #Ĥ 1 (kp , A)
#P 0 (GS , A) · #P 2 (GS , A)
=
#P 1 (GS , A)
It follows that
#P 0 (GS , A) · #P 2 (GS , A)
ϕ(GS , A) · ϕ(GS , A0 ) = χ2 (GS , A) · χ2 (GS , A0 )/ ,
#P 1 (GS , A)
thus showing (i). In order to prove (ii), let
0 −→ A1 −→ A2 −→ A3 −→ 0
be an exact sequence of finite modules in ModS (GS ). From the exact cohomo-
logy sequence and (8.6.10)(ii) we obtain
χ2 (GS , A1 )χ2 (GS , A3 ) #P 3 (GS , A2 )
=
χ2 (GS , A2 ) #P 3 (GS , A1 ) · #P 3 (GS , A3 )
#P 4 (GS , A1 ) · #P 4 (GS , A3 )
· · #δP 4 (GS , A3 )
#P 4 (GS , A2 )
= #δP 4 (GS , A3 )
because the Herbrand index of finite modules is equal to 1 by (1.7.6), and so
#P 3 (GS , A) = #P 4 (GS , A). For p ∈ S∞ the sequence
0 −→ H 0 (kp , A1 ) −→ H 0 (kp , A2 ) −→ H 0 (kp , A3 ) −→ δH 0 (kp , A3 ) −→ 0
is exact and
#δH 0 (kp , A3 ) = #δ Ĥ 0 (kp , A3 ) = #δH 4 (kp , A3 ).
Therefore
χ2 (GS , A1 ) · χ2 (GS , A3 ) Y #H 0 (kp , A1 ) · #H 0 (kp , A3 )
= ,
χ2 (GS , A2 ) p∈S∞ (k) #H 0 (kp , A2 )
and using ||#A2 ||p = ||#A1 ||p · ||#A3 ||p , this proves assertion (ii). 2

Proof of (8.6.10)(i): We proceed with several reduction steps.

Step 1. P T (k, S, A) is exact for A = ZZ/pZZ and A = µp , p ∈ IN(S).
Proof of step 1: As the Poitou-Tate sequence is self-dual, it suffices to deal
with the case A = µp . Using (8.3.4), we find the isomorphism
∼ C /p
IS /p −→ S
and the exact sequence

0 −→ µp −→ p IS −→ p CS −→ 0 .
Taking cohomology, we obtain the exact sequence
(1) H 1 (GS , µp ) −→ H 1 (GS , p IS ) −→ H 1 (GS , p CS ).
Since H 1 (GS , CS ) = 0, we get from the exact sequences
0 → p CS → CS →(CS )p → 0 and 0 →(CS )p → CS → CS /p → 0
the commutative and exact diagram
H 0 (GSDECBFGHIJK , CS )
p δ
H 0 (GS , (CS )p ) H 1 (GS , p CS ) 0
p
H 0 (GS , CS ) H 0 (GS , CS ) CS (k)/CS (k)p 0
H 0 (GS , CS /p)
and therefore an exact sequence
ϑ
(2) 0 −→ H 1 (GS , p CS ) −→ CS (k)/p −→ H 0 (GS , CS /p).
Replacing CS by IS , we obtain an analogous sequence
ϑ
(3) 0 −→ H 1 (GS , p IS ) −→ IS (k)/p −→ H 0 (GS , IS /p),
and all three sequences (1)–(3) fit into the commutative and exact diagram
H 1 (GRSTQLMNOP S , µp ) H 1 (GS , p IS ) H 1 (GS , p CS )
ϑ ϑ
ϑ̃
IS (k)/p CS (k)/p
H 0 (GS , IS /p) H 0 (GS , CS /p).

We get an exact sequence
ϑ̃
H 1 (GS , µp ) −→ IS (k)/p −→ CS (k)/p .
By (7.2.13), the diagram

H 1 (GYXUWV S , µp ) ϑ̃
IS (k)/p
λ1 rec
Ξ1
Gab
Y
P 1 (GS , µp ) P 1 (GS , ZZ/pZZ)∨ kp /p .
p∈S
is anti-commutative. As DS (k) is divisible by (8.3.13), the reciprocity map

∼ Gab /p .
rec : CS (k)/p −→ S
We obtain the commutative diagram
H 1 (Gab_`^[\]Z S , µp ) ϑ̃
IS (k)/p CS (k)/p
rec rec
−Ξ1 ·λ1
Gab
Y
kp /p Gab
S /p
p∈S
(λ1 )∨
H 1 (GS , ZZ/pZZ)∨ .
The sequence with the diagonal arrows is up to sign the central part of
P T (k, S, µp ), which is therefore exact.
Step 2. P T (k, S, A) is exact if P T (k, T, A) is exact for all finite subsets T ⊆ S.

Proof of step 2: We have seen that im λ0S ⊆ (im λS )⊥ , where
λS : H 1 (GS , A) −→ P 1 (GS , A) and λ0S : H 1 (GS , A0 ) −→ P 1 (GS , A0 ),
and the orthogonal complement ⊥ is taken with respect to the pairing
∪ Σ
P 1 (GS , A) × P 1 (GS , A0 ) −→ H 2 (GS , IS ) −→
M
Q/ZZ −→ Q/ZZ.
p∈S
Assume that we have proved
im λ0T = (im λT )⊥
for all finite subsets T of S with #A ∈ IN(T ) (and S∞ ⊆ T in the number
field case) for the corresponding localization maps λ0T and λT . Let πT :
P 1 (GS , A0 ) P 1 (GT , A0 ) be the canonical projection. Then the commutative
diagram
H 1 (Gdcfe T , A0 )
inf
H 1 (GS , A0 )
λ0T λ0S
πT
P 1 (GT , A0 ) P 1 (GS , A0 )
shows that
im λ0T = πT (im(λ0S ◦ inf )).

In order to prove the inclusion (im λS )⊥ ⊆ im λ0S , let x ∈ (im λS )⊥ . Then for
all sufficiently large finite subsets T ⊆ S, we have πT (x) ∈ (im λT )⊥ = im λ0T ,
so that
πT−1 (im λ0T ) = πT−1 πT (im λ0S ) = im λ0S .
\ \
x∈
T T
Since λS is proper by (8.6.11), im λS is closed in P 1 (GS , A) and analogously

im λ0S is closed in P 1 (GS , A0 ), and so x ∈ im λ0S = im λ0S . We obtain
im λ0S = (im λS )⊥ .
Hence, in order to prove the exactness of P T (k, S, A), we may assume that S
is finite.
Step 3. Let 0 → A → B → C → 0 be an exact sequence of finite modules

in ModS (GS ). If P T (k, S, A) and P T (k, S, C) are exact, then P T (k, S, B) is
also exact.
Proof of step 3: Since we may assume that S is finite, this follows directly
from (8.6.13).
Step 4: P T (k, S, A) is exact if pA = 0 and A is trivialized by a finite p-extension

of k inside kS , where p is any prime number.
Proof of step 4: By (1.6.13), A has a composition series whose graded pieces
are isomorphic to ZZ/pZZ. Therefore the exactness of P T (k, S, A) follows from
steps 1 and 3.
Step 5: P T (k, S, A) is exact for any finite A ∈ ModS (GS ).

Proof of step 5: By step 3, we may assume that A is a simple module; in
particular, pA = 0 for some prime number p. Let K|k be a finite Galois
subextension of kS |k which trivializes A. Let Gp be a p-Sylow subgroup
of G = G(K|k) and Kp ⊆ K the subfield associated to Gp . By step 4,
P T (Kp , S, A) is exact. The composition of cor and res between the complexes
P T (k, S, A) and P T (Kp , S, A) is multiplication by [Kp : k], which is prime
to p. Hence P T (k, S, A) can be identified with a direct summand of the exact
complex P T (Kp , S, A) and is therefore exact. This finishes the proof. 2
There is an alternative way to prove the exactness of the Poitou-Tate sequence

which avoids the reduction step to finite S and counting the orders of all
cohomology groups occurring. The following proof of step 3 uses some
calculations with double complexes and is taken from [169] (dévissage). We
start with the following

Claim: Let (A•• , d0 , d00 ) be a (commutative or anti-commutative) double com-

plex of abelian groups. For (i, j) ∈ ZZ × ZZ we put
Z ij = ker(d0 : Aij → Ai+1,j ) ∩ ker(d00 : Aij → Ai,j+1 ),
B ij = im(d00 ◦ d0 : Ai−1,j−1 → Aij ) and H ij = Z ij /B ij .

If for (i, j) ∈ ZZ × ZZ the groups H i (A•,j ) and H i+1 (A•,j ) vanish, then we have
an exact sequence
h
H ij −→ H j (Ai,• ) −→ H i+1,j −→ H i,j+1 −→ H j+1 (Ai,• ) .
Proof: The map H j (Ai,• ) → H i+1,j is induced by d0 . The map h is defined

as follows. Let x ∈ H i+1,j and let a ∈ Ai+1,j be a representing element. We
choose b ∈ Aij with d0 (b) = a and define h(x) ∈ H i,j+1 as the class of d00 (b). Our
assumptions imply that h is well-defined. The verification of the exactness of
the sequence is a straightforward diagram chase.
Now we build a double complex A•• in the following way. In column i =

−1 + 3k, k ∈ ZZ, we put P T (k, S, A) shifted by 3k positions downwards. In
column i = 3k, k ∈ ZZ, we put P T (k, S, B) shifted by 3k positions downwards
and in column i = 1 + 3k, k ∈ ZZ, we put P T (k, S, C) shifted by 3k positions
downwards. Let us look at a part of this unbounded double complex.
ghijklmnopqrstuvwxyz{|}~
H 1 (A0 )∨ H 1 (B 0 )∨ H 1 (C 0 )∨ H 0 (A0 )∨
P 1 (A) P 1 (B) P 1 (C) P 2 (A)
H 1 (A) H 1 (B) H 1 (C) H 2 (A)
H 2 (A0 )∨ H 2 (B 0 )∨ H 2 (C 0 )∨ H 1 (A0 )∨
.
1
The object P (B) is situated at position (0, 0) and the double complex is
periodic with period (3, −3). The rows are the long exact sequences (truncated
if k is not totally imaginary) attached to the short exact sequence 0 → A →
B → C → 0. The columns are exact, except the i = 0 column at j = 0, where
we want to show the exactness.

Applying the claim, we obtain an exact sequence

h h0
H 1,−1 −→ H 0,0 −→ H 0 (A0,• ) −→ H 1,0 −→ H 0,1
and homomorphisms
h h0
H 2,−2 → ∼ H −1,1 ,
∼ H 1,−1 ,→ H 0,0 → H 2,−1 → ∼ H −1,2 .
∼ H 1,0 H 0,1 →
Since H 2,−1 = H −1,2 is a subquotient of X2 (GS , A), it is finite. Hence

h0 is an isomorphism. As H 1,−1 is a subquotient of X1 (GS , C), the group
H 2,−2 = H −1,1 is finite, and so h is an isomorphism. The exact sequence above
shows that H 0 (A0,• ) = 0, hence P T (k, S, B) is exact.
For finitely generated modules we have the following statement which is

weaker than (8.6.10)(i).
(8.6.14) Theorem. Let S be a nonempty set of primes of a global field k and

assume that S ⊇ S∞ if k is a number field. Let A ∈ ModS (GS ). Then we have
an exact sequence
H 1 (GS , A) → P 1 (GS , A) → H 1 (GS , A0 )∨ → H 2 (GS , A) → P 2 (GS , A) .
Proof: The existence and exactness of the partial sequence

P 1 (GS , A) → H 1 (GS , A0 )∨ → H 2 (GS , A) → P 2 (GS , A)
follows from (8.6.7). The existence of the complex
H 1 (GS , A) → P 1 (GS , A) → H 1 (GS , A0 )∨
follows in exactly the same way as for finite A, cf. the proof of (8.6.10)(i). We
denote by AI the cokernel of the map sh1 : H 1 (GS , IS (A)) → P 1 (GS , A0 ).
Recall that sh1 is injective by (8.5.5). Denote by AC the cokernel of the
map α : H 1 (GS , CS (A)) → H 1 (GS , A)∨ induced by the cup-product and the
invariant map H 2 (GS , CS ) → Q/ZZ. By (8.4.4), α is an isomorphism if A is
torsion-free. The commutative diagram
H 1 (GS , IS (A)) P 1 (GS , A0 )
H 1 (GS , CS (A)) H 1 (GS , A)∨

induces a homomorphism AI → AC , and our next aim is to show that it is
injective.
Let T ⊆ A be the torsion submodule of A and let F = A/T . The statement of
the theorem is true for A = T by (8.6.10). Therefore we have the commutative
exact diagram

H 1 (G S , T 0 ) H 1 (GS , IS (T )) H 1 (GS , CS (T )) H 2 (GS , T 0 )
sh1 αT
H 1 (GS , T 0 ) P 1 (GS , T 0 ) H 1 (GS , T )∨ H 2 (GS , T 0 )
TI TC
We conclude that the maps αT : H 1 (GS , CS (T )) → H 1 (GS , T )∨ and TI → TC

are injective. We proceed by considering the commutative and exact diagram
H 1 (GS ,§¨©ª«¬¥¤ ¡¢£¦ IS (F )) H 1 (GS , IS (A)) H 1 (GS , IS (T )) H 2 (GS , IS (F ))
sh1 sh1 sh1 sh2
P 1 (GS , F 0 ) P 1 (GS , A0 ) P 1 (GS , T 0 ) P 2 (GS , F 0 )
AI TI
By (8.5.5) the map sh2 on the right-hand side induces an isomorphism on the
`-torsion part for all prime numbers ` | #T . Therefore the map AI → TI is
injective and the commutative diagram
A®¯° I TI
AC TC
shows that the map AI → AC is also injective. Next we consider the commu-
tative and exact diagram
H 0 (GS ,±²³´µ¶·¸¹º CS (T )) H 1 (GS , CS (F )) H 1 (GS , CS (A)) H 1 (GS , CS (T ))
αA αT
H 2 (GS , T )∨ H 1 (GS , F )∨ H 1 (GS , A)∨ H 1 (GS , T )∨ .
The left-hand vertical arrow is surjective by (8.4.5). We conclude that the map
αA is injective. Finally, consider the diagram

§7. The Global Euler-Poincaré Characteristic 503
H 1 (G»¼½¾¿ÀÁÂÃÄ S , A0 ) H 1 (GS , IS (A)) H 1 (GS , CS (A))
H 1 (GS , A0 ) P 1 (GS , A0 ) H 1 (GS , A)∨
AI AC .
All sequences but the middle row are exact, hence the middle row is also exact.
As Ξ1 is an isomorphism by (8.5.3), we obtain the statement of the theorem by
dualizing the middle row. 2
Theorem (8.6.10) was announced by J. TATE in [229], but he never published

a proof. G. POITOU gave a proof in the case that S is the set of all places in
[169], and we have used several of his arguments here. The extension to finitely
generated modules in (8.6.14) is due to T. TAKAHASHI [226].
§7. The Global Euler-Poincaré Characteristic
In this section we calculate the Euler-Poincaré characteristic of a finite

module A ∈ ModS (GS ), where S is finite. We start by proving an assertion
×
concerning the Galois module structure of OK,S for a finite Galois extension
K|k of number fields.
(8.7.1) Lemma. Let G be a finite group and E|F an extension of fields. Let
M1 and M2 be finite dimensional F [G]-modules such that M1 ⊗ E and M2 ⊗ E
are isomorphic as E[G]-modules. Then M1 and M2 are F [G]-isomorphic.
Proof: ∗) For finite dimensional F [G]-modules M1 and M2 consider the

isomorphism
HomF [G] (M1 , M2 ) ⊗ E ∼
= HomE[G] (M1 ⊗ E, M2 ⊗ E) ,
∗) The essential part of the proof is taken from [7], chap.IV, §8, lemma.

which is induced from the canonical isomorphism HomF (M1 , M2 ) ⊗ E ∼ =

HomE (M1 ⊗E, M2 ⊗E), cf. [17], chap.II, §5, prop.7(ii), by taking G-invariants.
If M1 ⊗ E ∼ = M2 ⊗ E, then, in particular, dimF M1 = dimF M2 and we can
speak of the determinant of a homomorphism between M1 and M2 (by choosing
bases of M1 and M2 ). Let ηi , i = 1, . . . , m, be an F -basis of HomF [G] (M1 , M2 ).
Then {ηi } is also an E-basis of the vector space HomE[G] (M1 ⊗ E, M2 ⊗ E).
Since M1 ⊗ E and M2 ⊗ E are isomorphic by assumption, there exist ai ∈ E
P
such that det( ai ηi ) =/ 0. Let
X
f (t) = det( ti ηi ) ∈ F [t1 , . . . , tm ].
f (t) is not the zero polynomial because f (a1 , . . . , am ) =/ 0. Then there exists
a finite extension F 0 of F and an element b = (b1 , . . . , bm ) ∈ F 0 m such that
f (b) =/ 0 (if F is infinite, then we can take F 0 = F ). It follows that bi ηi is
P
an F 0 [G]-isomorphism of M1 ⊗ F 0 onto M2 ⊗ F 0 . Considered as an F [G]-

homomorphism, this is an isomorphism M1n → ∼ M n , where n = [F 0 : F ]. By
2
the Krull-Schmidt theorem, see [32], §6, thm. 6.12, we obtain M1 ∼ = M2 as
F [G]-modules. 2
(8.7.2) Proposition. Let K|k be a finite Galois extension of number fields

with Galois group G = G(K|k). Let r1 and r2 be the number of real and
complex primes of k, respectively, and let r10 ≤ r1 be the cardinality of the set
0
S∞ of real primes of k which become complex in K. Finally, let S ⊇ S∞ be a
finite set of primes of k and let S f = S r S∞ . Then there are isomorphisms of
Q[G]-modules
×
⊗Q ∼
M
Q ⊕ OK,S = Q
P∈S(K)
∼ G
M
= Ind GP Q
p∈S(k)
∼ 0 G
M
= Q[G]r2 +r1 −r1 ⊕ Ind GP Q .
0 ∪S f )(k)
p∈(S∞
Proof: Consider the G-invariant map

×
M
Lg = (∆, log) : ZZ ⊕ OK,S −→ IR
P∈S(K)
where ∆ is the diagonal embedding of the trivial G-module ZZ and log(a) =
(log |a|P )P∈S(K) . Then by Dirichlet’s unit theorem ker Lg = µ(K) and the
L
image of Lg is a G-lattice Γ of rank s = #S(K) in P∈S(K) IR, cf. [160],
chap.VI, (1.1). It follows that
×
⊗ IR ∼
M
IR ⊕ OK,S = IR
P∈S(K)
as IR[G]-modules, hence by (8.7.1) for F = Q and E = IR, we obtain the
desired result. 2

(8.7.3) Corollary. Let K|k be a finite extension of number fields with Galois
group G = G(K|k) and let p be a prime number not dividing the order of G.
Then there exists a ZZp [G]-isomorphism
×
⊗ ZZp ∼
M
ZZp ⊕ OK,S = µp∞ (K) ⊕ ZZp
P∈S(K)
where S ⊇ S∞ is a finite set of primes and G acts on the sum on the right via
its action on S(K).
Proof: Consider the ZZp [G]-lattices

×
M
Γ1 = ZZp ⊕ (OK,S ⊗ ZZp )/µp∞ (K) and Γ2 = ZZp .
P∈S(K)
Since the order of G is prime to p, every finitely generated, ZZp -free ZZp [G]-
module is projective by (2.6.11). From the Qp [G]-isomorphism Γ1 ⊗ Qp ∼ =
Γ2 ⊗ Qp obtained in (8.7.2), it follows from representation theory (5.6.10)(ii)
that Γ1 ∼
= Γ2 . Therefore we obtain an exact G-invariant sequence
×
M
0 −→ µp∞ (K) −→ ZZp ⊕ OK,S ⊗ ZZp −→ ZZp −→ 0
P∈S(K)
which splits, since Γ2 is ZZp [G]-projective. 2
Recall that the second partial Euler-Poincaré characteristic of a finite GS -

module A is defined by
#H 0 (GS , A) · #H 2 (GS , A)
χ2 (GS , A) = .
#H 1 (GS , A)
(8.7.4) Global Euler-Poincaré Characteristic Formula. Let S be a finite

nonempty set of primes of the global field k containing S∞ if k is a number
field. Let A be a finite GS -module of order #A ∈ IN(S). Then
Y #H 0 (kp , A) Y #Ĥ 0 (kp , A0 )
χ2 (GS , A) = = 0
.
p∈S∞ ||#A||p 0
p∈S∞ #H (kp , A )
If k is a function field, this simply says χ2 (GS , A) = 1.
Remarks: 1. Recall that ||#A||p is #A if p is real and (#A)2 if p is complex.

2. The second equality in the statement (8.7.4) follows from (7.3.5).
3. If k is a function field or a totally imaginary number field and A a finite
module in ModS (GS ), then χ(GS , A) = χ2 (GS , A) by (8.3.17) and (8.3.18).

Proof: First we prove the formula for the Euler-Poincaré characteristic for
the GS -module A = µp , where p ∈ IN(S). Let K = k(µp ) and let G = G(K|k).
We consider the Grothendieck group K00 (IFp [G]) of finitely generated IFp [G]-
modules M and we denote the corresponding class of M by [M ]. We obtain
(i) [H 0 (kS |K, µp )] = [µp ],
×
(ii) [H 1 (kS |K, µp )] = [OK,S /p] + [p ClS (K)],
(iii) [H 2 (kS |K, µp )] = [ClS (K)/p] − [IFp ]
M M
+[ IFp ] + [ Ĥ 0 (Gp , ZZ/pZZ)],
p∈S\S∞ (K) p∈S∞ (K)
where S∞ is redundant in the function field case. (i) is obvious and (ii) follows
from the exact Kummer sequence (8.3.4), which induces the exact sequence
×
0 −→ OK,S /p −→ H 1 (kS |K, µp ) −→ p ClS (K) −→ 0
where we used (8.3.11)(ii). From (8.6.12) we obtain the exact sequence
0 → H 0 (kS |K, ZZ/pZZ) → P 0 (kS |K, ZZ/pZZ) → H 2 (kS |K, µp )∨
→ X1 (G(kS |K), ZZ/pZZ) → 0 .
We have
X1 (G(kS |K), ZZ/pZZ)∨ = ClS (K)/p
by (8.6.3), and since
M M
P 0 (kS |K, ZZ/pZZ) = IFp ⊕ Ĥ 0 (Gp , ZZ/pZZ) ,
p∈S\S∞ (K) p∈S∞ (K)
we obtain (iii).
If k is a function field, then
×
M
[OK,S /p ⊕ IFp ] = [ IFp ] + [µp ]
p∈S(K)
and we obtain 2
(−1)i [H i (GS , µp )] = 0
X
i=0
since [p ClS (K)] = [ClS (K)/p]. This proves the formula in the function field
case for A = µp .
We proceed with the proof for number fields in the case that A = µp . From
(8.7.3) we obtain
×
M
[OK,S /p] = [ IFp ] + [µp ] − [IFp ].
p∈S(K)
Combining this with (i)−(iii), we obtain
2
(−1)i [H i (kS |K, µp )] = [
X M M
Ĥ 0 (Gp , ZZ/pZZ)] − [ IFp ].
i=0 p∈S∞ (K) p∈S∞ (K)

Now we consider the homomorphism

Θ : K00 (IFp [G]) −→ ZZ , [M ] 7−→ dimIFp M G
(observe that #G is equal to 1 for p = 2 and is prime to p for p =/ 2, so that
M 7→ M G is an exact functor). Obviously,
2
P
Θ (−1)i [H i (G(kS |K),M )]
χ2 (GS , M ) = p i=0 .
Therefore
χ2 (GS , µp ) = 1
for function fields and
Y #Ĥ 0 (kp , IFp )
χ2 (GS , µp ) = 0
p∈S∞ (k) #H (kp , IFp )
for number fields. This finishes the proof in the case that A = µp .
Now we will prove the formula for the Euler-Poincaré characteristic for an
arbitrary finite GS -module whose order is in IN(S). Recall that the function
ϕ(GS , −) defined by
Q #H 0 (kp ,A)
ϕ(GS , A) = χ2 (GS , A)/ p∈S∞ ||#A||p
Q #Ĥ 0 (kp ,A0 )

= χ2 (GS , A)/ p∈S∞ #H 0 (kp ,A0 ) .
is multiplicative on short exact sequences, see (8.6.13)(ii). A second property
of ϕ is
ϕ(GS , IndUGS A) = ϕ(U, A) for U ⊆ GS open .
For the χ2 -part of ϕ, this follows from Shapiro’s lemma. Furthermore,
||#IndUGS A||p =
Y
||#A||P ,
P|p
because #IndUGS A = (#A)(GS :U ) , and

#H 0 (kp , IndUGS A) =
Y
#H 0 (KP , A)
P|p
for p ∈ S∞ (k), where K denotes the fixed field of U . The last equality is trivial
for complex primes, and if p is real, we have by I §5 ex. 5
H 0 (Gp , IndUGS A) = H 0 (GσP0 , Aσ ).
Y
σ ∈Gp \GS /U
Thus the desired property of ϕ for induced modules is proved.

The reduction from the general statement (8.7.4) to the case A = µp now
goes as follows. Let K|k be a finite, totally imaginary subextension of kS |k
trivializing A and A0 with Galois group G = G(K|k). Since ϕ is multiplicative,

we may assume that pA = 0 for some prime number p ∈ IN(S), and we consider
ϕ as a homomorphism from the Grothendieck group K00 (IFp [G]) of finitely
generated IFp [Ḡ]-modules into Q×+:
ϕ : K0 (IFp [G]) −→ Q×
0
+ .
We want to prove that ϕ is identical to 1. The group Q× + is torsion-free. Thus,

C
using lemma (7.3.4), we have to show that ϕ(GS , IndḠ N ) = ϕ(C, N ) = 1,
where C is a cyclic subgroup of Ḡ of order prime to p, C is the open subgroup
of GS given by C = C/G(kS |K) and N is a finitely generated IFp [C̄]-module.
Hence we may assume that K|k is a cyclic extension of degree prime to p
trivializing A and A0 . We have
H i (kS |k, A) ∼
= H i (kS |K, A)G for all i ≥ 0.
Let
2
χ0 : K00 (IFp [G]) −→ K00 (IFp [G]) , [M ] 7−→ (−1)i [H i (kS |K, M )] ,
X
i=0
and
Θ : K00 (IFp [G]) −→ ZZ , [M ] 7−→ dimIFp M G ;
then
0
χ2 (GS , M ) = pΘχ ([M ])
7 M G is an exact functor). ∗)
(observe that M →
Claim: For a finite IFp [G]-module M , we have
χ0 ([M ]) = [M 0∨ ] χ0 ([µp ]) .
Proof: The pairing
µp × Hom(M 0 , IFp ) −→ Hom(M 0 , µp ) = M , (ζ, f ) 7→ (x 7→ ζ f (x) ) ,
defines G-isomorphisms via the cup-product
H i (kS |K, µp ) ⊗ Hom(M 0 , IFp ) −→
∼ H i (k |K, M )
S
(recall that µp and M are trivial G(kS |K)-modules). This proves the claim.
Since χ0 ([µp ]) = 0 for function fields, we obtain χ2 (GS , A) = 1 in this case.
For the rest of the proof, let k be a number field. For p ∈ S∞ (K) we define
ψp0 : K00 (IFp [G]) −→ K00 (IFp [G]) , [M ] 7−→ [Ĥ 0 (KP , M )]−[H 0 (KP , M 0 )] ,
X
P|p
and
ψ 0 : K00 (IFp [G]) −→ K00 (IFp [G]) , [M ] 7−→ ψp0 ([M ]) .
X
p∈S∞ (k)
∗) Θ is not a ring-homomorphism with respect to the ring structure of K 0 (IF [G]) given by
0 p
the tensor product.

Obviously,
ψp0 ([M ]) = [M 0 ] ψp0 ([µp ]).
Using the claim and the assertion χ0 ([µp ]) = ψ 0 ([µp ]) proven at the beginning
of this part, we find for the finite GS -module A that
0 0
ϕ(GS , A) = pΘχ ([A])−Θψ ([A])
Y 0∨ ]ψ 0 ([µ ])−[A0 ]ψ 0 ([µ ]))
= pΘ([A p p p p
.
p∈S∞ (k)
It remains to prove that for p ∈ S∞ (k) and for an IFp [G]-module A,

Θ([A∨ ]ψp0 ([µp ])) = Θ([A]ψp0 ([µp ])) .
It is easy to see that
0, p = 2, p is real and remains real in K,
(
ψp0 ([µp ]) =
−[IndG
G IFp ] ,
P
otherwise.
This proves the equality above if p = 2 and p is real and remains real in K.
Therefore we assume that we are in one of the other cases. Then it follows
from IndḠP IFp ⊗ A ∼
G
= IndGG A and
P
H 0 (G, IndG 0
G A) = H (GP , A)
P
that
Θ([A]ψp0 ([µp ])) = Θ(−[IndG G A])
p
= − dimIFp (IndG
G A)
p G
= − dimIFp AGP .
Since
dimIFp AGP = dimIFp AGP = dimIFp (A∨ )GP ,
the equality holds in all cases. This finishes the proof. 2
(8.7.5) Corollary. Let p be a prime number and let S be a finite set of primes
of the number field k containing S∞ and all primes above p. If p =/ 2, then for
all j ∈ ZZ,
2
(
X
i i −r1 (k) − r2 (k) for j odd,
(−1) dimIFp H (GS , ZZ/pZZ(j)) =
i=0
−r2 (k) for j even,
and if p = 2, then
2
(−1)i dimIF2 H i (GS , ZZ/2ZZ) = −r2 (k) .
X
i=0
Here r1 (k) and r2 (k) denote the number of real and complex places of the
number field k. In the function field case, this alternating sum of dimensions
is zero.

Proof: If p is odd, then for a number field k and an archimedean prime p,

the valuation ||p||p is equal to p resp. p2 if p is real resp. complex, and the
order of H 0 (kp , ZZ/pZZ(j)) is 1 for p real, j odd, and p in all other cases. In
the case p = 2, observe that all twists are equal, and we obtain the result as
above. 2
(8.7.6) Corollary. With the notation of (8.7.5), the following equalities hold
for p ≥ 2 and all j ∈ ZZ:
2
(
X
i −r1 (k) − r2 (k) for j odd,
(−1) rank ZZp Hi (GS , ZZp (j)) =
i=0
−r2 (k) for j even,
and
2
(
X
i i −r1 (k) − r2 (k) for j odd,
(−1) rank ZZp Hcts (GS , ZZp (j)) =
i=0
−r2 (k) for j even.
Proof: The second equality follows from the first using (2.7.12), and with
exactly the same arguments as in the proof of (7.3.8) (replacing Gk by GS ),
we obtain
2 2
i
(−1)i dimIFp H i (GS , ZZ/pZZ(−j))
X X
(−1) rank ZZp Hi (GS , ZZp (j)) =
i=0 i=0
−dimIFp p H2 (GS , ZZp (j)) .
Now the corollary follows from (8.7.5) and the following lemma. 2
(8.7.7) Lemma. Let k be a global field, let p =/ char(k) be a prime number

and let S be a finite nonempty set of primes of k containing S∞ and all primes
above p if k is a number field.
If p is odd, then for all j ∈ ZZ, the homology group H2 (GS , ZZp (j)) is ZZp -
torsion-free, and if p = 2, then
(
(ZZ/2ZZ)r1 (k) for j odd,
tor(H2 (GS , ZZ2 (j))) =
0 for j even.
Proof: We prove the dual statement, i.e. H 2 (GS , Qp /ZZp (−j)) is p-divisible
if p is odd or j is even, and its co-torsion is equal to (ZZ/2ZZ)r1 (k) otherwise.
This is clear for function fields by (8.3.17) and in the number field case for
p =/ 2 since cdp GS ≤ 2 in these cases. We have the commutative diagram

ÅÆÇÈ m ZZ(−j))
H 3 (GS , ZZ/2 2m H
3
(GS , Q2 /ZZ2 (−j))
H 3 (kp , ZZ/2m ZZ(−j))

Y Y
3
2m H (kp , Q2 /ZZ2 (−j))
p∈S∞ (k) p∈S∞ (k)
where the vertical maps are isomorphisms by (8.6.10)(ii). Thus

H 2 (GS , Q2 /ZZ2 (−j))/2m −→ H 2 (kp , Q2 /ZZ2 (−j))/2m .
Y
∼
p∈S∞ (k)
Since we have
(
ZZ/2ZZ for j odd,
H 2 (kp , Q2 /ZZ2 (−j)) ∼
= Ĥ 0 (kp , Q2 /ZZ2 (−j)) ∼
=
0 for j even,
for a real prime p ∈ S∞ (k), this yields the lemma. 2
A very useful application of the Poitou-Tate duality theorem and the global
Euler characteristic formula is a duality statement for certain subgroups of
H 1 (Gk , A) which are defined by local conditions. In particular, this is important
for the theory of elliptic curves where one considers the so-called Selmer group
of an elliptic curve.
(8.7.8) Definition. Let k be a global field and A a finite Gk -module of order

prime to the characteristic of k . A collection L of local conditions for A is a
family Lp of subgroups of H 1 (Gp , A) such that
Lp = H 1 (Gp /Tp , ATp ) = ker(H 1 (Gp , A) → H 1 (Tp , A))
for almost all p, where Tp denotes the inertia subgroup of Gp .
If L is a collection of local conditions for A, then LD is a collection of local
conditions for the Gk -module A0 = Hom(A, µ) defined by
LD
p := the orthogonal complement of Lp with respect to the pairing
∪
H 1 (Gp , A) × H 1 (Gp , A0 ) −→ Q/ZZ .
The corresponding global groups are defined by

Y
HL1 (k, A) := ker H 1 (Gk , A) −→ H 1 (Gp , A)/Lp ,
p

HL1 D (k, A0 ) := ker H 1 (Gk , A0 ) −→ H 1 (Gp , A0 )/LD
Y
p .
p

If L is a collection of local conditions for the finite Gk -module A, then we

choose a finite nonempty set S of primes of k containing all primes p such
that ATp =/ A or Lp =/ H 1 (Gp /Tp , ATp ), and S∞ and all divisors of #A if k is
number field. Hence A and A0 are GS -modules and we have
Y
HL1 (k, A) = ker H 1 (GS , A) −→ H 1 (Gp , A)/Lp ,
p∈S

HL1 D (k, A0 ) = ker H 1 (GS , A0 ) −→ H 1 (Gp , A0 )/LD
Y
p .
p∈S
The following theorem is due to A. WILES.
(8.7.9) Theorem. Let k be a global field and A a finite Gk -module of order

prime to char(k), and let L be a collection of local conditions for A. Then
#HL1 (k, A) #H 0 (k, A) Y #Lp
1 0
= 0 0
· 0
.
#HLD (k, A ) #H (k, A ) p #H (kp , A)
Proof: Let S be a finite set of primes of k chosen as above. Recall, by

convention we have S∞ = ∅ in the function field case. From the commutative
exact diagram
ÉÊËÌÍÎÏÐÑÒÓ
Y Y
H 1 (kp , A)/Lp H 1 (kp , A)/Lp HL1 D (k, A0 )∨
p∈S p∈S
X1 (GS , A)
Y
H 1 (GS , A) H 1 (kp , A) H 1 (GS , A0 )∨
p∈S
(H 1 (kp , A0 )/LD ∨
Y Y
HL1 (k, A) Lp p ) ,
p∈S p∈S
0 → X1 (GS , A) → HL1 (k, A) → Lp → H 1 (GS , A0 )∨ → HL1 D (k, A0 )∨ → 0,

Y
p∈S
and so the exact sequence

0 → H 0 (GS , A) → P 0 (GS , A) → H 2 (GS , A0 )∨
Lp → H 1 (GS , A0 )∨ → HL1 D (k, A0 )∨ → 0.
Y
→ HL1 (k, A) →
p∈S

§8. Duality for Unramified and Tamely Ramified Extensions 513
Recall that in the number field case, for p ∈ S∞ , the group H 0 (Gp , A) is
replaced by Ĥ 0 (Gp , A) in the term P 0 (GS , A). Using (8.7.4), we get
#HL1 (k, A) #H 0 (GS , A) #Lp #Lp

·χ2 (GS , A0 )·
Y Y
1 0
= 0 0
· 0
#HLD (k, A ) #H (GS , A ) p∈S∞ #Ĥ (kp , A) p∈S\S∞ #H (kp , A)
0
#H 0 (k, A) Y #Lp Y #Lp

= 0 0
· 0
· 0
.
#H (k, A ) p∈S∞ #H (kp , A) p∈S\S∞ #H (kp , A)
But this is the desired result, since for p ∈/ S the group Lp is equal to
H 1 (Gp /Tp , ATp ) and for all p we have #H 1 (Gp /Tp , ATp ) = #H 0 (Gp , A) be-
cause of the exact sequence
0. 2
1−Frobp
0ÔÕÖ×Ø H 0 (Gp , A) ATp ATp H 1 (Gp /Tp , ATp )
§8. Duality for Unramified and Tamely Ramified

Extensions
In this section we prove duality theorems for unramified and tamely ramified
extensions of a global field.
Let us fix some notation. Let k be a global field and let S and T be sets of
primes of k. Further, let c be a full class of finite groups. We set
kST the maximal extension of k unramified outside S in which all

primes of T are completely decomposed,
kST (c) the maximal c-extension of k inside kST .
From now on we assume that T is nonempty and contains S∞ if k is a number

field. Furthermore, we assume that S ∩ T = ∅; in particular, S contains only
nonarchimedean primes. We set
kST,tr the maximal extension of k unramified outside S in which all

primes of T are completely decomposed and all primes of S
are at most tamely ramified,
kST,tr (c) the maximal c-extension of k inside kST,tr .

T,tr
Remarks: 1. We have k∅T = k∅ ∅
and the field k∅ is the maximal unramified
∅ S∞
extension of k. If k is a number field, we have k∅ = k∅ .
2. Let k be a number field and let S be a set of finite primes of k such that
S ∩ Sp = ∅ for all prime numbers p with ZZ/pZZ ∈ c. Then a c-extension is
automatically tamely ramified at the primes in S, and we have kSS∞ ,tr (c) = kS (c).
For any extension K of k inside kST,tr we put

S ×
Y
1
EK,T = ker OK,T −→ UK,p /UK,p ,
p∈S(K)
S
i.e. EK,T is the group of T -units which are principal units at all places of S. If
S S
T = S∞ , we write EK = EK,S ∞
.
The main result of this section is the following theorem (8.8.1). It generalizes
a duality theorem for unramified extensions contained in [257] and a duality
theorem for tamely ramified extensions obtained by G. WIESE in [248].
(8.8.1) Theorem. Let k be a global field and let T be a nonempty set of primes
of k which contains S∞ if k is a number field. Let S be a finite set of primes
disjoint from T and let c be any full class of finite groups. Let A be a discrete
G(kST,tr (c)|k)-module which is finitely generated and free as a ZZ-module. Then
there are canonical topological isomorphisms
Ĥ i (G(kST,tr (c)|k), Hom(A, EkST,tr (c),T )) ∼ T,tr

= Ĥ 3−i (G(kS (c)|k), A)∨
S
for all i ∈ ZZ.
Before proving the theorem, we deduce some corollaries.
(8.8.2) Corollary. Let k be a number field and let T be a set of primes of k

containing S∞ . For the maximal unramified extension k∅T in which all primes
of T are completely decomposed there are natural isomorphisms
Ĥ i (G(k∅T |k), Ok×T ,T ) ∼

= Ĥ 3−i (G(k∅T |k), ZZ)∨
∅
for all i ∈ ZZ. In particular, H 2 (G(k∅T |k), Ok×T ,T ) = 0.

∅
Proof: This follows from (8.8.1) by setting c to be the class of all finite
groups, S = ∅ and A = ZZ. 2

(8.8.3) Corollary. Let k be a number field and let S be a finite set of nonar-
chimedean primes of k. Let p be a prime number such that S ∩ Sp = ∅. Then
there are natural isomorphisms for all i ∈ ZZ
Ĥ i (G(kS (p)|k), EkSS (p) ) ∼

= Ĥ 3−i (G(kS (p)|k), ZZ)∨ .
In particular, H 2 (G(kS (p)|k), EkSS (p) ) = 0.
Proof: By remark 2 above, we have kSS∞ ,tr (p) = kS (p). Therefore, applying
(8.8.1) to the class of finite p-groups with T = S∞ and A = ZZ, we obtain the
result. 2
In order to prove (8.8.1), we introduce some further objects. Let K be a

finite extension of k inside kST,tr . We consider the group
S
Kp× ×
Y Y Y
1
UK,T = UK,p × UK,p ,
p∈
/(S∪T ) p∈T p∈S
which is a subgroup of the idèle group IK . Let ClTS (K) be the cokernel of
S
the composite map UK,T −→ IK −→ CK . The notation is motivated by the
∅
identification ClT (K) = ClT (K). The reciprocity homomorphism of global
class field theory induces an isomorphism of finite abelian groups
∼ G(k T,tr |K)ab .
ClTS (K) −→ S
For an infinite extension K of k in kST,tr we define the above groups as the

direct limit over the finite layers. For any K, we obtain an exact sequence
S S
0 −→ EK,T −→ UK,T −→ CK −→ ClTS (K) −→ 0.
We have ClTS (kST,tr (c))(c) = 0 by the principal ideal theorem.
In the following we use the notation Ĥ i (K|k, −) for Ĥ i (G(K|k), −). We

start with a lemma.
(8.8.4) Lemma. We have a natural isomorphism
H 1 (kST,tr (c)|k, EkST,tr (c),T ) ∼

= ClTS (k)(c).
S
Proof: Let ZZ(c) be the subring of Q obtained from ZZ by inverting all primes p
with ZZ/pZZ ∈/ c. For an abelian torsion group A, we have a natural isomorphism
A(c) →∼ A⊗ Z Z(c) . Since the extension kST,tr (c)|k is unramified outside S, tamely
ramified at S and completely decomposed at T , the G(kST,tr (c)|k)-module

UkST,tr (c),T is cohomologically trivial; see (7.1.2). Since ClTS (kST,tr (c))(c) = 0,
S
we have an exact sequence
0 → EkST,tr (c),T ⊗ ZZ(c) → UkST,tr (c),T ⊗ ZZ(c) → CkT,tr (c) ⊗ ZZ(c) → 0 ,
S S S
which induces a natural isomorphism

Z(c) ∼

H 1 (kST,tr (c)|k, EkST,tr (c),T ) −→
∼ coker U S → C
k,T k ⊗Z = ClTS (k)(c).
S
Proof of (8.8.1): Applying (8.4.1) to K = kST,tr (c), we obtain natural isomor-

phisms
Ĥ i (k T,tr (c)|k, Hom(A, C T,tr )) ∼
S kS (c)
T,tr
= Ĥ 2−i (k (c)|k, A)∨ , i ∈ ZZ.
S
Let K|k be a finite Galois extension inside kST,tr (c) which trivializes A. We
consider the exact sequence
S S
(∗) 0 −→ EK,T −→ UK,T −→ CK −→ ClTS (K) −→ 0.
The association K 7→ ClTS (K) is merely a G(kST,tr (c)|k)-modulation but not
a module, as we are lacking Galois descent; see (1.5.10) and page 56. How-
ever, for each fixed K, the group G(K|k) acts on ClTS (K), and we will con-
sider ClTS (K) as a G(K|k)-module. Thus (∗) is an exact sequence of dis-
S
crete G(K|k)-modules. The group UK,T is a cohomologically trivial G(K|k)-
module. Applying Hom(A, −) and taking cohomology, we therefore obtain
isomorphisms
Ĥ i+1 (K|k, Hom(A, E S )) ∼
= Ĥ i (K|k, Hom(A, X(K))),
K,T
where X(K) denotes the kernel of the surjection CK ClTS (K), and a long
exact sequence
· · · −→ Ĥ i (K|k, Hom(A, X(K))) −→ Ĥ i (K|k, Hom(A, CK ))
−→ Ĥ i (K|k, Hom(A, ClTS (K))) −→ · · · .

As the cohomology groups of Hom(A, CK ) and of Hom(A, ClTS (K)) are finite,
the same is true for the cohomology groups of Hom(A, X(K)).
Now let K 0 be another finite Galois extension of k inside kST,tr (c) which
contains K. We consider ClTS (K) as a G(K 0 |k)-module via the projec-
can
tion G(K 0 |k) G(K|k). Then the natural maps ClTS (K) −→ ClTS (K 0 ) and
norm
ClTS (K 0 ) −→ ClTS (K) are homomorphisms of G(K 0 |k)-modules and we obtain
associated homomorphisms
inf : Ĥ i (K|k, Hom(A, ClTS (K))) → Ĥ i (K 0 |k, Hom(A, ClTS (K 0 ))), i ≥ 1,
def : Ĥ i (K 0 |k, Hom(A, ClTS (K 0 ))) → Ĥ i (K|k, Hom(A, ClTS (K))), i ≤ 0.

Similarly, we obtain associated maps between the cohomology groups of

Hom(A, X(K 0 )) and Hom(A, X(K)).
Let K 0 be the maximal abelian extension of K in kST,tr (c). By class field
theory, K 0 is a finite extension of K and we have a natural isomorphism
∼ G(K 0 |K). The commutative diagram
rec : ClTS (K)(c) →
ÜÛÚÙ 0 )(c)
ClTS (K norm
ClTS (K)(c)
rec rec
G(kST,tr (c)|K 0 )ab can G(kST,tr (c)|K)ab

shows, since can is the zero map, that
Ý 0 )(c) norm ClS (K)(c)
ClTS (K T
is the zero map. Noting that
Ĥ i (K|k, Hom(A, ClTS (K))) = Ĥ i (K|k, Hom(A, ClTS (K)(c))),
we obtain
lim Ĥ i (K|k, Hom(A, ClTS (K))) = 0 for i ≤ 0.
←−
K
Furthermore,
lim Ĥ i (K|k, Hom(A, ClTS (K))) = 0 for i ≥ 1,
−→
K
since
Þ
ClTS (K)(c) can
ClTS (K 0 )(c)
is the zero map by the principal ideal theorem. Passing to the inverse, resp.
direct, limit, we obtain isomorphisms
lim Ĥ i (K|k, Hom(A, X(K))) ∼
←−
T,tr
= Ĥ i (k (c)|k, Hom(A, C T,tr )), i ≤ 0,
S kS (c)
K, def
lim Ĥ i (K|k, Hom(A, X(K))) ∼ T,tr

= Ĥ i (kS (c)|k, Hom(A, CkT,tr (c) )), i ≥ 1.
−→ S
K, inf
Therefore we obtain canonical isomorphisms for i ≥ 2:
Ĥ i (kST,tr (c)|k, Hom(A, ETS )) ∼
= lim H i (K|k, Hom(A, EK,T
S
))
−→ K
∼
= lim Ĥ i−1 (K|k, Hom(A, X(K)))
−→ K
∼ T,tr
= Ĥ i−1 (kS (c)|k, Hom(A, CkT,tr (c) ))
S
∼
= Ĥ 3−i T,tr
(k (c)|k, A)∨
,
S
and for i ≤ 0:
Ĥ i (kST,tr (c)|k, Hom(A, ETS )) ∼
= lim H i (K|k, Hom(A, EK,T
S
))
←− K
∼
= lim Ĥ i−1 (K|k, Hom(A, X(K)))
←− K
∼ T,tr
= Ĥ i−1 (k (c)|k, Hom(A, C T,tr ))
S kS (c)
∼ T,tr
= Ĥ 3−i (kS (c)|k, A)∨ ,
where ETS = EkST,tr (c),T .
S

For i = 1, we need an additional argument. Let K0 |k be a finite Galois

extension inside kST,tr (c) which trivializes A and let K be a finite Galois ex-
tension of k inside kST,tr (c) containing the maximal abelian extension K0ab of
K0 in kST,tr (c). Since H 1 (kST,tr (c)|K0 , ETS ) ∼
= ClTS (K0 )(c) by (8.8.4), we obtain
from the Hochschild-Serre spectral sequence for the group extension
1 −→ G(kST,tr (c)|K) −→ G(kST,tr (c)|K0 ) −→ G(K|K0 ) −→ 1
the exact sequence
S
0 −→ H 1 (K|K0 , EK,T ) −→ ClTS (K0 )(c) −→ ClTS (K)(c).
The map on the right is zero as K contains the field K0ab , and we get an
isomorphism
H 1 (K|Kß 0 , EK,T H 1 (kST,tr (c)|K0 , ETS ) ∼
S inf
) = ClTS (K0 )(c).
Therefore we obtain an exact sequence
S S
0 −→ H 1 (K0 |k, Hom(A, EK 0 ,T
)) −→ H 1 (K|k, Hom(A, EK,T )) −→
H 0 (K0 |k, Hom(A, ClTS (K0 )(c))) −→ H 2 (K0 |k, Hom(A, EK

S
0 ,T
)),
which shows, by comparing it with the same sequences for all extensions of K
inside kST,tr (c), that
à
H 1 (K|k, Hom(A, S
EK,T ))
inf
H 1 (kST,tr (c)|k, Hom(A, ETS ))
is an isomorphism. In exactly the same way one proves that the injective map
inf
H 2 (K|k, A) → H 2 (kST,tr (c)|k, A) induces an isomorphism
á
H 2 (K|k, A)
inf
H 2 (kST,tr (c)|k, A).
S
Furthermore, since the groups Hom(A, UK,T )G(K|k) , Hom(A, EK,T
S
)G(K|k) and
1 S ab
H (K|k, Hom(A, EK,T )) do not depend on K ⊇ K0 , the exact sequence
S
0 −→ Hom(A, EK,T )G(K|k) −→ Hom(A, UK,T
S
)G(K|k) −→
Hom(A, X(K))G(K|k) −→ H 1 (K|k, Hom(A, EK,T

S
)) −→ 0
shows that the natural map
0
∼ Hom(A, X(K 0 ))G(K |k)
Hom(A, X(K))G(K|k) −→
is an isomorphism for any finite Galois extension K 0 inside kST,tr (c) containing
K. If K 0 is the maximal abelian extension of K in kST,tr (c), then in the
commutative and exact diagram
âãäåæçè Hom(A, X(K 0 ))G(K |k)
0
Hom(A, CK 0 )G(K |k)
0 0
Hom(A, ClTS (K 0 ))G(K |k)
ϕ
Hom(A, X(K))G(K|k) Hom(A, CK )G(K|k) Hom(A, ClTS (K))G(K|k)

the canonical map ϕ is zero on the c-part. This implies an isomorphism
Hom(A, X(K))G(K|k) ⊗ ZZ(c) −→
∼ Hom(A, C )G(K|k) ⊗ Z
K Z(c) ,

where ZZ(c) is the subring of Q obtained from ZZ by inverting all primes p with
ZZ/pZZ ∈/ c. Now we consider the commutative and exact diagram
NK|k Hom(A, éêëìíîïX(K)) ⊗ ZZ(c) NK|k Hom(A, CK ) ⊗ ZZ(c)
Hom(A, X(K))G(K|k) ⊗ ZZ(c) Hom(A, CK )G(K|k) ⊗ ZZ(c)
Ĥ 0 (K|k, Hom(A, X(K))) Ĥ 0 (K|k, Hom(A, CK )).

Replacing k by K0 (i.e. A ∼
= ZZrankA ), the lower map is an isomorphism since
Ĥ 0 (K|K0 , X(K)) ∼ S
= Ĥ 1 (K|K0 , EK,T )
∼ T,tr S
= H 1 (kS (c)|K0 , EK,T )
∼ 2 T,tr
= Ĥ (k (c)|K0 , ZZ) ∨
S
∼
= Ĥ 2 (K|K0 , ZZ)∨
∼
= G(K|K0 )ab
∼
= Ĥ 0 (K|K0 , CK ).
Therefore
∼ N
NK|K0 Hom(A, X(K)) ⊗ ZZ(c) −→ K|K0 Hom(A, CK ) ⊗ Z
Z(c)
is an isomorphism. As NK|k = NK0 |k ◦ NK|K0 , we obtain an isomorphism
∼ N
NK|k Hom(A, X(K)) ⊗ ZZ(c) → K|k Hom(A, CK ) ⊗ Z
Z(c) .
This shows that
Ĥ 0 (K|k, Hom(A, X(K))) →
∼ Ĥ 0 (K|k, Hom(A, C ))
K
is an isomorphism and therefore

S
H 1 (K|k, Hom(A, EK,T )) ∼
= Ĥ 0 (K|k, Hom(A, X(K)))
∼
= Ĥ 0 (K|k, Hom(A, CK ))
∼
= Ĥ 2 (K|k, A)∨ .
As the inflation maps
inf
S
H 1 (K|k, Hom(A, EK,T )) −→ H 1 (kST,tr (c)|k, Hom(A, ETS ))
inf
and H 2 (K|k, A) −→ H 2 (kST,tr (c)|k, A) are isomorphisms, this shows the exis-
tence of the required isomorphism for i = 1.
It remains to show that this isomorphism is natural, i.e. independent of
the choice of K. For typesetting reasons, we prove this below in the case
c = (all finite groups), leaving the necessary modification in the general case
to the reader. We consider the perfect pairing of finite groups

S
(∗) H 1 (K|k, Hom(A, EK,T )) × H 2 (K|k, A) → Q/ZZ
associated to the isomorphism H 1 (K|k, Hom(A, EK,TS
)) →∼ H 2 (K|k, A)∨ ob-
tained above. By construction, it is induced via the vertical isomorphisms in

ñðòôõó
Ĥ 0 (K|k, Hom(A, X(K))) × H 2 (K|k, A) ∪
H 2 (K|k, X(K)) inv0
Q/ZZ
o δ δ
S ∪ S
H 1 (K|k, Hom(A, EK,T )) × H 2 (K|k, A) H 3 (K|k, EK,T ),
followed by the pairing in the upper line, where inv0 denotes the composite
map
inv
H 2 (K|k, X(K)) −→ H 2 (K|k, CK ) −→ Q/ZZ.
÷öýûüùúøþÿ E S )
H 3 (K|k, δ
H 2 (K|k, X(K)) H 2 (K|k, CK ) inv
Q/ZZ
K,T
inf inf inf
H 3 (kST,tr |k, ETS ) δ

∼ H 2 (kST,tr |k, X(kST,tr )) H 2 (kST,tr |k, CkT,tr ) inv
Q/ZZ.
S
Denoting the map H 3 (kST,tr |k, ETS ) → Q/ZZ induced by the lower line by inv00 ,
we conclude that the perfect pairing (∗) is the cup-product pairing
S S ∪
H 1 (K|k, Hom(A, EK,T )) × H 2 (K|k, A) −→ H 3 (K|k, EK,T )
inf inv00
composed with the composite H 3 (K|k, EK,T S
) −→ H 3 (kST,tr |k, ETS ) −→ Q/ZZ.
It is therefore compatible via the inflation maps with the analogous diagram
for an extension K 0 of K in kST,tr . This concludes the proof. 2

Chapter IX
The Absolute Galois Group

of a Global Field
In this chapter we will deal with two fundamental questions in algebraic

number theory. The first of these questions is the determination of all extensions
of a fixed base field k (where the most important case is k = Q), which means
exploring how these extensions are built up over each other, how they are
related and how they can be classified. In other words, we want to study the
structure of the absolute Galois group Gk of k as a profinite group. But in
contrast to the local Galois groups we are far from a complete understanding of
the global situation and there are many conjectures but only a few conceptual
results. For example, there is a famous conjecture due to I. R. ŠAFAREVIČ
which asserts that the subgroup Gk(µ) of Gk is a free profinite group, where
k(µ) is the field obtained from k by adjoining all roots of unity. This was
proved by F. POP [171] for function fields, but the conjecture is open in the
number field case.
However, this first question reflects only one side of the situation which we
consider in algebraic number theory. Namely, it is of purely algebraic nature
and can be asked for any field. But number fields are arithmetic fields, endowed
with primes, completions etc. Therefore, the second fundamental problem is
the question of the decomposition of prime ideals in an extension, and it can
be interpreted as the problem of determining the local groups Gkp with respect
to the global group Gk , i.e. how they lie inside Gk and how they interact.
As we are far from being able to answer these two questions concerning
the “algebraic structure” and the “arithmetic structure” of Gk in complete
generality, we will only give partial answers to these questions. In the first
section we deal with the so-called Hasse principle, also called the local-global
principle. This is the question for the kernel of the natural maps
res i : H i (Gk , A) −→ H i (Gkp , A) ,
Y
i = 1, 2,
p
where A is a Gk -module and res i is given by the restriction maps res ip . The
kernel is just the obstruction to a local-global principle (i.e. that the global
properties of H i (Gk , A) are determined by the corresponding local ones). One

522 Chapter IX. The Absolute Galois Group of a Global Field
also wants to know this in the case of restricted ramification, so that one wants
to determine the kernel

Xi (kS , T, A) = ker H i (GS (k), A) −→ H i (Gkp , A)
Y
p∈T
where T ⊆ S are sets of primes of k.

In the second section we deal with the cokernels of these maps. For a finite
set T and a “large” set S, the surjectivity is (under certain assumptions) the
famous theorem of GRUNWALD-WANG. As a corollary, one obtains that given
finitely many abelian local extensions Kp |kp , there exists an abelian global
extension K|k realizing the given local ones (except for one very special case).
In §3 we extend the surjectivity results of §2 by showing the existence of
global cohomology classes which realize finitely many given local classes and
are of a very special kind. This will be used in an essential way in the proofs
of the theorems of Neukirch and Šafarevič later in this chapter.
In §4 we deal with the question of whether a finite number of local groups Gkp
can form a free product inside Gk (actually we only consider the corresponding
pro-p-factor groups for a prime number p). Here the theorem of Grunwald-
Wang will play an important role.
A partial answer to the conjecture of Šafarevič is given by K. IWASAWA:
the maximal prosolvable quotient of Gk(µ) is a free prosolvable group. The
same is true for the Galois group of the maximal prosolvable extension k̃
of the maximal abelian extension k ab of k. This will be proved in §5 by
considering embedding problems. In this section we also prove the theorem of
Neukirch which gives a sufficient condition under which finitely many given
local extensions can be simultaneously realized by a solvable global extension.
We conclude the chapter with a famous theorem of Šafarevič which asserts
that, for a given global field k, every finite solvable group G occurs as a Galois
group of a Galois extension K of k, i.e. G(K|k) = G.
§1. The Hasse Principle
The name Hasse principle has its origin in the classical theorem of Hasse-
Minkowski, which asserts that a quadric has a rational point over a number
field k if and only if it has a rational point over all completions of k. Another
example of a local-global principle is the theorem of Hasse on the Brauer group
of a global field (8.1.17).

§1. The Hasse Principle 523
Let k be a global field and S a set of primes of k. We do not assume

that S is nonempty nor that S contains S∞ if k is a number field. We recall
the convention that an archimedean prime is unramified if and only if it is
completely decomposed, i.e. we consider the extension C|IR of archimedean
local fields as ramified. Given any separable closure k̄ of k, we denote the
maximal extension of k inside k̄ which is unramified outside S by kS and we
set GS (k) = G(kS |k). The ring of S-integers of k is defined by
Ok,S = {a ∈ k | vp (a) ≥ 0 for all p ∈/ S ∪ S∞ }.
If k is a function field and S = ∅, then Ok,S is the algebraic closure of the
prime field in k. Recall the notation
×
IN(S) = {n ∈ IN | n ∈ Ok,S }.
For a finite GS (k)-module A with #A ∈ IN(S) we set A0 = Hom(A, OS× ), where
OS× = Ok×S ,S .
We are interested in the kernel X1 (kS , T, A) given by the exact sequence
0 −→ X1 (kS , T, A) −→ H 1 (kS |k, A) −→

Y
H 1 (kp , A) ,
p∈T
where T ⊆ S and A is a finite GS (k)-module. If T = S, we also use the

notation X1 (kS , A) := X1 (kS , S, A) = X1 (GS , A).
(9.1.1) Definition. We say the Hasse principle holds (in dimension 1) for the
GS -module A and the set of primes T ⊆ S if
X1 (kS , T, A) = 0 .
In this generality the Hasse principle is obviously not true even if T = S. To

give an example, let A = ZZ/pZZ and let S ⊇ Sp ∪ S∞ be a finite set of primes
of the number field k. Then X1 (kS , S, A) is equal to the dual of the p-primary
part of the S-ideal class group ClS (k) of k which is nontrivial in general.
One might think that the Hasse principle would hold for a GS -module A
if S = T is the set of all primes of k. But for a reducible Gk -module A there
are counterexamples; see [72], 7.3, and [95]. In this section we will deduce
several criteria on S, T and A that imply the vanishing of X1 (kS , T, A).
If k(A) denotes the minimal trivializing extension of k for the module A,
i.e. if Gk(A) is the kernel of the homomorphism Gk → Aut(A) given by the
action of Gk on A, then we define, for a Galois extension K|k inside kS and
containing k(A), the group X1 (K|k, T, A) by the exactness of
res
0 −→ X1 (K|k, T, A) −→ H 1 (K|k, A) −→
Y
H 1 (Kp |kp , A).
p∈T

As above, we put X1 (K|k, A) := X1 (K|k, S, A) if T = S. We need some

further notation.
(9.1.2) Definition. Let k be a global field and let S be a set of primes of k.

The Dirichlet-density δ(S) of S is the limit
P −s
p∈S\S∞ N (p)
δ(S) = δk (S) = lim+ P −s
s→1 all finite p N (p)
if it exists. We have 0 ≤ δ(S) ≤ 1. For sets S1 and S2 of primes, we use the
notation
S1 ⊂
∼ S2 :⇐⇒ δ(S1 r S2 ) = 0,
i.e. S1 is contained in S2 up to a set of primes of density zero. If Ω|k is a finite

separable extension of k, then we denote by
δΩ (S) = δΩ (S(Ω))
the Dirichlet density of the set S(Ω) of primes of Ω given by all extensions of
S = S(k). Furthermore, we set
cs(Ω|k) := {p a prime of k | p splits completely in Ω|k},
Ram(Ω|k) := {p a prime of k | p ramifies in Ω|k}
for a finite Galois extension Ω|k.
Let Ω|k be a finite Galois extension and let p vary over the primes of k which
are unramified in Ω|k. Then the set Frobp of the Frobenius automorphisms
with respect to the prolongations of p to Ω is a conjugacy class in G(Ω|k). For
a conjugacy class C in G(Ω|k) we put
PΩ|k (C) = {p a prime of k | Frobp = C}.
For a proof of the following theorem refer to [160], chap.VII, (13.6), and [53],
chap. 5, §4 in the function field case.
(9.1.3) Čebotarev’s Density Theorem. Let Ω|k be a finite Galois extension

with group G and let C be a conjugacy class in G. Then
#C
δ(PΩ|k (C)) = .
#G
For a set S of primes of k and a finite Galois extension Ω|k we mention the
relation
δΩ (S) = δk S ∩ cs(Ω|k) · [Ω : k].
We start with some calculations.

(9.1.4) Lemma. Let p be a prime number, m ≥ 1 a natural number and let

G ⊆ (ZZ/pm ZZ)× = Aut(ZZ/pm ZZ) be a subgroup. Let A be the G-module
which is isomorphic to ZZ/pm ZZ as an abelian group and on which G acts in the
canonical way. Then
Ĥ i (G, A) = 0 for all i ∈ ZZ,
unless p = 2, m ≥ 2 and −1 ∈ G, in which case
Ĥ i (G, A) ∼
= ZZ/2ZZ for all i ∈ ZZ.
Proof: Let α = 1 + ps u ∈ U 1 , s ≥ 1, u ∈ ZZ× Zp . If p is

p , be a principal unit of Z
odd or if s ≥ 2, then vp (αr − 1) = s + vp (r) for all r ∈ IN.
Let us start with the case p odd. Then
G1 = ker( G → (ZZ/pZZ)× ) = G ∩ im(U 1 →(ZZ/pm ZZ)× )
is a p-Sylow subgroup of G and in order to show the first statement, we may
replace G by the cyclic group G1 . Let pm−s , s ≥ 1 be the order of G. We
use the same letter for an element α ∈ ZZ× Z/pm ZZ)× . Fix a
p and its image in (Z
generator α ∈ U s r U s+1 of G. Then AG = ker(α − 1: A → A) = pm−s A. As
pm−s
X−1
m−s
αp
i −1
α = = pm−s v,
i=0 α−1
with v Z×
∈ Z p, we have NG (A) = pm−s A = AG . Since finite cyclic groups have
trivial Herbrand index by (1.7.6), we obtain Ĥ i (G, A) = 0 for all i ∈ ZZ.
Now we assume p = 2 and we leave the trivial case m = 1 aside. Then there
are three possibilities for G:
a) G ∼= {±1} × hαi with α ∈ U 2 ,
b) G ∼= hαi with α ∈ U 2 .
c) G ∼= hαi with −α ∈ U s r U s+1 , 2 ≤ s ≤ m − 1.
Note that −1 ∈ G only in case a). In case b) the same argument as for
odd p shows Ĥ i (G, A) = 0 for all i ∈ ZZ. In case c) we have the equality
AG = ker(α − 1: A → A) = 2m−1 A. The order of G is equal to 2m−s ≥ 2. As
2m−s
X−1
m−s m−s
i α2 − 1 (−α)2 −1
α = = = 2m−1 v,
i=0 α−1 α−1
with v Z×
∈ Z 2,we have NG (A) = 2m−1 A. In the same way as in case b) we
i
conclude Ĥ (G, A) = 0 for all i ∈ ZZ. Let us finally consider case a) and let
2m−s be the order of the subgroup hαi ⊆ G. Since
2m−s
X−1 2m−s
X−1
i
α + −αi = 0,
i=0 i=0
the norm map NG : A → A is the zero map. We have AG = 2m−1 A and

IG A = 2A, hence Ĥ i (G, A) ∼
= ZZ/2ZZ for i = −1, 0. By case b), we have

Ĥ i (hαi, A) = 0 for all i. For i ≥ 1 we obtain

#H i (G, A) = #H i ({±1}, Ahαi ) = #Ĥ 0 ({±1}, 2m−s A) = 2.
In the same way, using homology, we obtain #Ĥ i (G, A) = 2 for i ≤ −2. 2
We will use the lemma above in the following situation: Let G be a subgroup
of the Galois group G(k(µpm )|k), where k is any prime field with char(k) =/ p.
Then G acts on µpm via the cyclotomic character χcycl : G ,→ (ZZ/pm ZZ)× . In
the case when k = Q and p = 2 the group G corresponds to a field in the
diagram

Q(ζ2m )
Q(ζ2m−1 )
Q(ζ8 ) Q(i · η2m )
Q(ζ4 ) Q(η2m )
√
Q( −2) Q(η2m−1 )
√
Q( 2)
Q
where η2m = ζ2m + ζ2−1 m

m and ζ2m is a primitive 2 -th root of unity.
(9.1.5) Definition. Let k be a field and m = 2r m0 , m0 odd, a natural number

prime to char(k). We say that we are in the special case (k, m) if r ≥ 2 and −1
is in the image of the cyclotomic character χcycl : G(k(µ2r )|k) → (ZZ/2r ZZ)× .
(9.1.6) Proposition. Let p be a prime number, r ∈ IN, and let k be any field
with char(k) =/ p. Then
Ĥ i (G(k(µpr )|k), µpr ) = 0 for all i ∈ ZZ,
except when p = 2, r ≥ 2 and we are in the special case (k, 2r ). In this case
Ĥ i (G(k(µ2r )|k), µ2r ) ∼

= ZZ/2ZZ for all i ∈ ZZ.
Let p = 2 and r ≥ 2. Then the special case (k, 2r ) occurs if and only if
char(k) = 0 and Q(µ2r ) ∩ k is real,
or char(k) = ` ≡ −1 mod 2r and IF` (µ2r ) ∩ k = IF` .

Remark: The special case in positive characteristic occurs if and only if

k(µ2r+1 ) = k(µ2r ) and this field is a quadratic extension of k (cf. ex. 3 below).
Proof: As the group G = G(k(µpr )|k) acts on µpr via the cyclotomic character
χcycl : G ,→ (ZZ/pr ZZ)× , we may apply (9.1.4). It therefore suffices to identify
the case −1 ∈ im(χcycl ) if p = 2, r ≥ 2. If char(k) = 0, then the cyclotomic
character factors as
∼ G(Q(µ r )|k ∩ Q(µ r )) ,→ (Z
χcycl : G −→ 2 2 Z/2r ZZ)×
and, since −1 ∈ (ZZ/2r ZZ)× ∼ = G(Q(µ2r )|Q) is the complex conjugation, we
have −1 ∈ im(χcycl ) if and only if k ∩ Q(µ2r ) is real.
Now let char(k) = ` ∈/ {0, 2}. Then the cyclotomic character factors as
χcycl : G −→ ` 2 ` 2 Z/2r ZZ)× .
∼ G(IF (µ r )|k ∩ IF (µ r )) ,→ (Z
The group G(IF` (µ2r )|k ∩IF` (µ2r )) is cyclic, generated by the relative Frobenius
automorphism σ and
χcycl (σ) = `[IF` (µ2r )∩k:IF` ] .
Hence −1 ∈ im(χcycl ) if and only if `[IF` (µ2r )∩k:IF` ]·n ≡ −1 mod 2r for some
n ∈ IN. The last condition is equivalent to ` ≡ −1 mod 2r and IF` (µ2r )∩k = IF` .
2
(9.1.7) Definition. Let k be a global field, m = 2r m0 , m0 odd, a natural

number prime to char(k) and T a set of places of k. We say that we are in
the special case (k, m, T ) if we are in the special case (k, m) and all primes
p ∈ T decompose in k(µ2r )|k.
(9.1.8) Lemma. Let k be a global field. Assume that char(k) =/ 2, k(µ2r )|k is
cyclic and δ(T ) > 1/2. Then we are not in the special case (k, 2r , T ).
Proof: If we are in the special case (k, 2r ) and k(µ2r )|k is cyclic, then we
have [k(µ2r ) : k] = 2. By Čebotarev’s density theorem, the set of primes which
do not decompose in k(µ2r )|k has density 1/2 and therefore has a nontrivial
intersection with T . 2
Remarks: 1. The extension k(µ2r )|k can be non-cyclic only if k is a number

field, r ≥ 3 and k ∩ Q(µ2r ) is real.
2. If k(µ2r )|k is not cyclic, then all primes p - 2 are decomposed in k(µ2r )|k.
Indeed, these primes have a cyclic decomposition group since all p ∈/ S2 ∪ S∞
are unramified.

3. If δ(T ) > 1/2, then, by the first two remarks and (9.1.8), the special case
(k, m, T ) is equivalent to the statement
k is a number field, m = 2r m0 , m0 odd, r ≥ 3,
k(µ2r )|k is not cyclic, and
all primes p dividing 2 in T decompose in k(µ2r )|k.
Now we can formulate some first cases where the Hasse principle holds.
(9.1.9) Theorem. Let k be a global field and let S be a set of primes of k.

The Hasse principle holds for a finite GS (k)-module A and the sets of primes
T ⊆ S of k, i.e.
X1 (kS , T, A) = 0,
(i) A is a trivial GS (k)-module and δ(T ) > 1/p, where p is the smallest
prime divisor of #A.
(ii) A = µm with m = pr11 · · · prnn , where pi are pairwise different prime

numbers in IN(S), and
1
δ(cs(k(µpri i )|k) ∩ T ) >
pi · [k(µpri i ) : k]
for all i = 1, . . . , n, except we are in the special case (k, m, T ), where
X1 (kS , T, µm ) ∼ = ZZ/2ZZ.
∼ T and #G(k(A)|k) = lcm{#G(k(A)p |kp ) | p ∈ T }.

(iii) cs(k(A)|k) ⊂
Proof: (i) We may assume that A = ZZ/pr ZZ and want to show the injectivity
of the homomorphism
H 1 (kS |k, ZZ/pr ZZ) −→ H 1 (kp , ZZ/pr ZZ).
Y
p∈T
If ϕ : GS (k) → ZZ/pr ZZ is in the kernel, then ker(ϕ) corresponds to a cyclic

Galois extension L|k of degree ps , 0 ≤ s ≤ r, which is unramified outside S
and completely decomposed at T . Since δ(T ) > 1/p, Čebotarev’s density
theorem (9.1.3) implies s = 0, hence ϕ = 0.
(ii) We may assume that m = pr and therefore K := k(A) = k(µpr ). The



Y
H 1 (kS |K, A) H 1 (Kp , A)
T (K)
X1 (kS , T, A)
Y
(∗) 0 H 1 (kS |k, A) H 1 (kp , A)
T (k)
X1 (K|k, T, A)
Y
0 H 1 (K|k, A) H 1 (Kp |kp , A)
T (k)
shows, using
1
δK (T ) = δk T ∩ cs(K|k) · [K : k] >
p
and (i), that
X1 (kS , T, A) = X1 (K|k, T, A).
By (9.1.6) we have H 1 (K|k, µpr ) = 0 and therefore the group X1 (kS , T, A) is
trivial, unless we are in the special case (k, 2r ), where
H 1 (k(µ2r )|k, µ2r ) ∼
= ZZ/2ZZ.
If there exists a prime p ∈ T such that p does not decompose in k(µ2r ), then
G(k(µ2r )|k) = G(kp (µ2r )|kp ) and therefore X1 (k(µ2r )|k, T, µ2r ) = 0.
On the other hand, the homomorphism
ZZ/2ZZ ∼
= H 1 (k(µ 2r )|k, µ2r )
res p
H 1 (kp (µ2r )|kp , µ2r )
is the zero map if the prime p decomposes in k(µ2r ). Indeed, in this case
G(kp (µ2r )|kp ) is a proper subgroup of G(k(µ2r )|k). Let k 0 be its fixed field.
Then
G(k(µ r )|k0 )
ker res p = H 1 (k 0 |k, µ2r 2 ),
G(k(µ r )|k0 )
which is isomorphic to ZZ/2ZZ since µ2r 2 = µ2 or k 0 = k(µ2t ) for some
t ≥ 2, in which case we apply (9.1.6) again. This shows assertion (ii).
(iii) Let, for a prime number p, ps be the maximal p-power dividing the order
of G = G(K|k), where K = k(A). By assumption there exists at least one
p ∈ T such that ps | #Gp , hence Gp = G(Kp |kp ) contains a Sylow group Gp
of G. Since
Y
H 1 (G, A) −→ H 1 (Gp , A)
p
1
is injective by (1.6.10), the kernel X (K|k, T, A) is zero. Using the diagram
(∗) of the proof of part (ii), we obtain the result. 2
Using Poitou-Tate duality, we obtain injectivity results for the second coho-
mology group.

(9.1.10) Corollary. Let k be a global field and let S be a nonempty set of

primes such that S∞ ⊆ S if k is a number field. Let A be a finite GS (k)-module
with #A ∈ IN(S). Then the canonical homomorphism
M
H 2 (kS |k, A) −→ H 2 (kp , A)
S
is injective in the following cases:
(i) A = ZZ/mZZ with m = pr11 · · · prnn , where pi are pairwise different prime
1
δ(cs(k(µpri i )|k) ∩ S) >
for all i = 1, . . . , n, except we are in the special case (k, m, S), where
X2 (kS , ZZ/mZZ) ∼ = ZZ/2ZZ.
(ii) A0 is a trivial GS (k)-module and δ(S) > 1/p, where p is the smallest
(iii) cs(k(A0 )|k) ⊂ 0 0
∼ S and #G(k(A )|k) = lcm{#G(k(A )p |kp ) | p ∈ S}.
Proof: From the Poitou-Tate duality theorem (8.6.7), we know that

X2 (kS , A) ∼
= X1 (kS , A0 )∨ .
Now everything follows from (9.1.9) with T = S considering A0 instead of A.
2
A first application is a local-global principle for m-th powers.
(9.1.11) Theorem. Let k be a global field, m a natural number prime to char(k)

and T a set of primes of k of density δ(T ) = 1. Then the following holds.
(i) The localization homomorphism
k × /k ×m −→ kp× /kp×m
Y
p∈T
is injective, except in the following case:

k is a number field, m = 2r m0 , m0 odd, r ≥ 3,
k(µ2r )|k is not cyclic, and
all primes p dividing 2 in T decompose in k(µ2r )|k,
where the kernel is cyclic of order 2.
(ii) If α ∈ k × is a 2m-th power in kp× for all p ∈ T , then α is an m-th power
in k × .

Proof: In view of k × /k ×m ∼ = H 1 (k, µm ) assertion (i) follows from (9.1.9)(ii),

where S is the set of all primes of k together with (9.1.8) and the subsequent
Remark 3.
r
In order to prove (ii), we may reduce to the case √ that m = 2 , r 3 and we
≥
are in the special case (k, m, T ). In particular, −1 k. Let K = k(µ2m ).

∈
/
Then, by (9.1.9)(i), the localization homomorphism
K × /K ×2m −→ Kp× /Kp×2m
Y
p∈T (K)
is injective. Let α ∈ k × be a 2m-th power in kp× for all p ∈ T . Then α ∈ K ×2m .

The commutative and exact diagram
×2m
K × /K H 1 (GK , µ2m )
k × /k ×2m H 1 (Gk , µ2m )
H 1 (K|k, µ2m )
together with (9.1.6) shows that α2 = β 2m for some β ∈ k × . Hence α = ±β m .
Suppose α = −β m . As α in an m-th power in kp× for all p ∈ T , the same holds
√
for −1 ∈ k × . Hence the quadratic extension k( −1)|k is decomposed at a set
of primes of density 1 which contradicts the Čebotarev density theorem. We
conclude α = β m , showing (ii). 2
Now let S be an arbitrary set of primes of k. For a natural number m prime

to char(k), we consider the Kummer group∗)

VS (k, m) = ker k × /k ×m −→ kp× /kp×m × kp× /Up kp×m .
Y Y
p∈S /S
p∈
(Recall the convention Up = kp× if p is archimedean.) We denote the dual of

this group by the letter B (the Russian “B”), i.e.
BS (k, m) := VS (k, m)∨ .
∗) The group V (k, m) is a localization kernel for flat cohomology. We have

S
VS (k, m) = ker Hfl1 (Spec(Ok,S ), µm ) → p∈S H 1 (kp , µm ) .

Q

(9.1.12) Proposition. Let k be a global field, m a natural number prime to

char(k) and S a set of primes. Then the following holds:
(i) The group VS (k, m) is finite.
(ii) If S has density δ(S) = 1, then VS (k, m) = 0 if we are not in the special
case (k, m, S).
(iii) Assume that m ∈ IN(S) and S ⊇ S∞ if k is a number field. Then
VS (k, m) ∼
= X1 (GS , µm ).
Proof: The isomorphism VS (k, m) ∼ = X1 (GS , µm ) stated in (iii) follows

immediately from Kummer theory. Hence, by (8.6.4), VS (k, m) is finite if S is
nonempty, m ∈ IN(S) and S ⊇ S∞ if k is a number field. In the general case,
let T be a nonempty finite set containing all primes dividing m∞. Then we
have an exact sequence
Up /Upm ,
M
0 −→ VS∪T (k, m) −→ VS (k, m) −→
p∈T \S
showing (i). Finally, (ii) follows from (9.1.11)(i) and the definition of VS (k, m).
2
Next we are going to discuss the Hasse principle for the cohomology of
simple modules.
(9.1.13) Lemma. Let G be a profinite group and let A be a finite simple

G-module. Assume there exists a normal subgroup N of G of order prime to
#A which acts nontrivially on A. Then
H i (G, A) = 0 for all i ≥ 0.
Proof: Since A is simple and N is normal in G acting nontrivially on A,

we have AN = 0. As N is of order prime to #A, we have H i (N, A) = 0 for
i ≥ 1. Now the result follows from the Hochschild-Serre spectral sequence
E2st = H s (G/N, H t (N, A)) ⇒ H s+t (G, A). 2
The next proposition is due to W. GASCHÜTZ, see [220], lemma 1.
(9.1.14) Proposition. Let A be a simple Gk -module, let K = k(A) and suppose

that G = G(K|k) =/ 1 is solvable. Then
H i (G, A) = 0 for all i ≥ 0.

Proof: Since A is simple, it is an IFp -vector space for some prime number p.
The solvable group G has a nontrivial abelian normal subgroup N . We claim
that its p-Sylow subgroup Np is trivial. Indeed, Np is a normal subgroup in G
and if Np =/ 1, we obtain ANp = 0 or ANp = A since A is simple. In the first
case it would follow that A = 0 by (1.6.12) and in the second that k(A) was
not the minimal A-trivializing extension of k.
Thus G has a nontrivial normal subgroup N of order prime to p which acts
nontrivially on A (as k(A) is the minimal A-trivializing extension of k). The
result follows from (9.1.13). 2
Now we can prove that the Hasse principle holds for simple modules under
certain conditions.
(9.1.15) Theorem. Let k be a global field, S a set of primes of k and let A be

a finite simple GS (k)-module. Let p be the prime number with pA = 0 and let
T ⊆ S be a subset. Then
X1 (kS , T, A) = 0,
(i) G(k(A)|k) is solvable and cs(k(A)|k) ⊂

∼ T.
(ii) p ∈ IN(S), G(k(A0 )|k) is solvable and cs(k(A0 )|k) ⊂

∼ T.
(iii) p ∈ IN(S) and there are finite Galois extensions K ⊆ Ω of k inside kS

such that p - [Ω : K], and
/ K and k(A0 ) ⊆ K,
µp ⊆
µp ⊆ Ω and cs(Ω|k) ⊂
∼ T.
Proof: (i) Put K = k(A). If K =/ k, then H 1 (K|k, A) = 0, by (9.1.14). If

K = k, this is trivially true. In the commutative diagram
!"#
Y
H 1 (kS |K, A) H 1 (Kp , A)
T (K)
Y
H 1 (kS |k, A) H 1 (kp , A)
T
the upper map is injective by (9.1.9)(i), since cs(K|k) ⊂

∼ T and hence δK (T ) = 1.
This shows assertion (i).

(ii) Note that A0 = Hom(A, µp ) is simple, since A is simple. Let K = k(A) and
K 0 = k(A0 ). Then K 0 = k(A0 ) ⊆ kS and
K 0 (µp ) = K(µp ) .
Therefore the group H 1 (K 0 (µp )|K, A) = H 1 (K(µp )|K, A) is zero. By (9.1.14),
we have H 1 (K|k, A) = 0, since A is simple and G(K|k) is solvable as
G(K(µp )|k) = G(K 0 (µp )|k) is. As cs(K 0 |k) ⊂ ∼ T we have δK 0 (T ) = 1 and
δK(µp ) (T ) = 1. This explains the injectivity of the two left-hand side vertical
and the upper horizontal arrow in the commutative diagram
$%&'()* p ), A)
Y
H 1 (kS |K(µ H 1 (K(µp )p , A)
T (K(µp ))
Y
H 1 (kS |K, A) H 1 (Kp , A)
T (K)
Y
H 1 (kS |k, A) H 1 (kp , A) ,
T (k)
which therefore implies the desired result.
(iii) Since k(A0 ) ⊆ K ⊆ Ω and µp ⊆ Ω, we have k(A) ⊆ Ω. In addition,

cs(Ω|k) ⊂∼ T implies δΩ (T ) = 1. Therefore in the commutative and exact
diagram
H 1 (kS +,-./|Ω, A)
Y
H 1 (Ωp , A)
T (Ω)
Y
H 1 (kS |k, A) H 1 (kp , A)
T
H 1 (Ω|k, A)
the upper horizontal map is injective by (9.1.9)(i). Consider the normal sub-
group N = G(Ω|K) ⊆ G(Ω|k). By assumption #N is prime to p. As
k(A0 ) ⊆ K and µp ⊆ / K, N acts nontrivially on the simple module A. This
implies H 1 (Ω|k, A) = 0 by (9.1.13) and shows (iii). 2
Using Poitou-Tate duality, we obtain the

(9.1.16) Corollary. Let k be a global field and let S be a nonempty set of

primes such that S∞ ⊆ S if k is a number field. Let A be a finite simple
GS (k)-module with #A ∈ IN(S) and let p be the prime number with pA = 0.
Then the canonical homomorphism
M
H 2 (kS |k, A) −→ H 2 (kp , A)
S
is injective in the following cases:
(i) G(k(A)|k) is solvable and cs(k(A)|k) ⊂
∼ S.
(ii) G(k(A0 )|k) is solvable and cs(k(A0 )|k) ⊂

∼ S.
(iii) There are finite Galois extensions K ⊆ Ω of k inside kS such that

p - [Ω : K],
/ K and k(A) ⊆ K,
µp ⊆
∼ S.
Proof: Using the Poitou-Tate duality theorem (8.6.7),

X2 (kS , A) ∼
= X1 (kS , A0 )∨ ,
the assertion follows from (9.1.15) with T = S by considering A0 instead of A.
2
Exercise 1. Let k be a number field, m = 2r m0 , m0 odd, and let T be a set of primes of

k of density δ(T ) > 1/2. Let ζs be a primitive 2s -th root of unity and let ηs = ζs + ζs−1 .
Assume ηs ∈ k but ηs+1 ∈/ k. Show that the special case (k, m, T ) is equivalent to the following
properties:
1) −1, 2 + ηs and − (2 + ηs ) are not squares in k,
2) r > s,
3) {p | p divides 2 and − 1, 2 + ηs , −(2 + ηs ) are not squares in kp } ∩ T = ∅ .
Hint: [6], chap.10.
√
Exercise 2.√ Show that (Q( 7), 23 , P), where P is the set of all places, is a special case, i.e.
for k = Q( 7) the map
Y
k × /k ×8 −→ kp× /kp×8
p
is not injective.
Exercise 3. Let m ≥ 2 and let k be a field of positive characteristic ` =/ 2. Show that the
following conditions are equivalent.
(i) ` ≡ −1 mod 2m and IF` (µ2m ) ∩ k = IF` .
(ii) k(µ2m+1 ) = k(µ2m ) is a quadratic extension of k.

§2. The Theorem of Grunwald-Wang
Let, as in the last section, k be a global field and S a set of primes of k,

T ⊆ S a subset and A a finite GS (k)-module.
In this section we are interested in the cokernel of the restriction map res i =
res i (S, T ) = (res ip )p∈T
res i
H i (kS |k, A) −→ H i (kp , A) coker i (kS , T, A)
Y
for i = 1, 2. If i = 2 then the restricted product is just a direct sum and the
group coker 2 (kS , T, A) is often trivial by the following
(9.2.1) Proposition. Let k be a global field and let S be a nonempty set

of primes of k such that S∞ ⊆ S if k is a number field. Let A be a finite
GS (k)-module with #A ∈ IN(S) and let T ⊆ S be a subset. If S r T contains
a nonarchimedean prime, then the map
M
res 2 (S, T, A) : H 2 (kS |k, A) −→ H 2 (kp , A)
T
is surjective.
Proof: The long exact sequence of Poitou-Tate induces the commutative

exact diagram
H 2 (kS012345678 |k, A)
M
H 2 (kp , A) H 0 (kS |k, A0 )∨ 0
S
M
H 2 (kS |k, A) H 2 (kp , A) coker 2 (kS , T, A) 0.
T
We obtain the exact sequence
0 −→ coker 2 (kS , T, A)∨ −→ H 0 (kS |k, A0 ) −→ H 2 (kp , A)∨ .

M
S rT
If p ∈ S r T is nonarchimedean, then H (kS |k, A ) injects into H 0 (kp , A0 ) ∼

0 0
=
H 2 (kp , A)∨ . This shows the statement. 2
Now we investigate res 1 assuming that T is finite. If S is nonempty, contains

S∞ if k is a number field and #A ∈ IN(S), then the local and global duality

§2. The Theorem of Grunwald-Wang 537
theorems induce the following commutative exact diagram:

9:;<=>?@ABCDEFG 0
X1 (kS , S r T, A0 )
Y
H 1 (kS |k, A0 ) H 1 (kp , A0 )
S rT
X1 (kS , A0 )
Y
H 1 (kS |k, A0 ) H 1 (kp , A0 ) H 1 (kS |k, A)∨
S
Y Y
H 1 (kp , A0 ) H 1 (kp , A)∨
T T
0 coker 1 (kS , T, A)∨
0.
This diagram implies the
(9.2.2) Lemma. Assume that S is nonempty and contains S∞ if k is a number

field. If T is finite and #A ∈ IN(S), then there is a canonical exact sequence
0HIJK X1 (kS , A0 ) X1 (kS , S r T, A0 ) coker 1 (kS , T, A)∨ 0.
Using the Hasse principles of the last section, we obtain the following result.
(9.2.3) Theorem. Let k be a global field, and let T ⊆ S be sets of primes of k

where T is finite and S contains S∞ if k is a number field. Let A be a finite
GS (k)-module with #A ∈ IN(S). Then the canonical homomorphism
M
T
is surjective in the following cases:
(i) A0 is a trivial GS (k)-module and δk (S) > 1/p, where p is the smallest
(ii) A = ZZ/mZZ with m = pr11 · · · prnn , where pi are pairwise different prime
1
δ(cs(k(µpri i )|k) ∩ S) >
for i = 1, . . . , n, except we are in the special case (k, m, S r T ). In this
case the cokernel of res 1 (S, T, ZZ/mZZ) is of order 1 or 2.
(iii) cs(k(A0 )|k) ⊂ 0 0
∼ S and #G(k(A )|k) = lcm{#G(k(A )p |kp ) | p ∈ S r T }.

(iv) A is simple, G(k(A0 )|k) is solvable and cs(k(A0 )|k) ⊂

∼ S.
(v) A is simple, G(k(A)|k) is solvable and cs(k(A)|k) ⊂

∼ S.
(vi) A is simple, pA = 0 for a prime number p ∈ IN(S), and there are finite
Galois extensions K ⊆ Ω of k inside kS such that p - [Ω : K],
/ K and k(A) ⊆ K,
µp ⊆
∼ S.
(vii) δk (S) = 1 and G(k(A0 )p |kp ) is cyclic for all p ∈ T .
Proof: By (9.2.2), we know that

coker 1 (kS , T, A) ⊆ X1 (kS , S r T, A0 )∨ .
Thus assertions (i)-(vi) follow from the corresponding statements in (9.1.9)
and (9.1.15), respectively. In order to prove (vii), we first note that, since
G0p = G(k(A0 )p |kp ) is cyclic for all p ∈ T and since δk (S) = 1, by Čebotarev’s
density theorem there exists a prime p̃ ∈ S r T for each p ∈ T such that
G0p = G0p̃ ⊆ G0 = G(k(A0 )|k). If x ∈ X1 (k(A0 )|k, S r T, A0 ), then res 1p̃ x = 0
by definition, hence res 1p x = 0 and therefore x ∈ X1 (k(A0 )|k, S, A0 ). Thus the
injection X1 (k(A0 )|k, S, A0 ) ,→ X1 (k(A0 )|k, S r T, A0 ) is an isomorphism.
Since δk(A0 ) (S r T ) = 1, we have
X1 (kS |k(A0 ), A0 ) = 0 = X1 (kS |k(A0 ), S r T, A0 )

and the commutative exact diagram
X1 (kSLMNOP |k, A0 ) X1 (kS |k, S r T, A0 ) coker 1 (kS , T, A)∨
X1 (k(A0 )|k, S, A0 ) X1 (k(A0 )|k, S r T, A0 )

implies (vii). This finishes the proof of the theorem. 2
(9.2.4) Corollary. Let k be a global field, and let T ⊆ S be sets of primes

of k where T is finite and S contains S∞ if k is a number field. Let K be a
finite Galois extension of k inside kS with Galois group G = G(K|k) and let
A = IFp [G]n , where p ∈ IN(S) is a prime number. If δ(S) = 1, then the map
M
p∈T
is surjective.

Proof: By (9.2.3)(ii), the lower restriction map in the commutative diagram

H 1 (kSSTRQ |k, A) res
M
H 1 (kp , A)
p∈T
H 1 (kS |K, IFpn ) res

H 1 (KP , IFpn )
M
P∈T (K)
is surjective, hence the same is true for the upper restriction map. 2
The statement of (9.2.4) also holds if p = char(k). In fact, we have the

following stronger result.
(9.2.5) Theorem. Let k be a global field of characteristic p > 0 and let T ⊆ S

be sets of primes of k where T is finite and S r T =/ ∅. Let A be a p-primary
GS (k)-module. Then the canonical homomorphism
M
T
is surjective.
Proof: We start with the case A = IFp . By (8.3.2), we have

H 1 (GS (k), IFp ) = Ok,S /℘Ok,S ,
where ℘(x) = xp −x. For any prime p of k, (6.1.2) yields H 1 (kp , IFp ) = kp /℘kp .
Furthermore, ℘kp is open in kp by (6.1.6). Since S r T =/ ∅, the strong
approximation theorem (cf. [22], th. 15.1) shows that the natural map
M
Ok,S −→ kp /℘kp
p∈T
is surjective. This proves the assertion for A = IFp . Using the same argument as
in the proof of (9.2.4), we obtain the result for modules of the form A = IFp [G]n ,
where G is any finite quotient of GS (k).
Now let 0 → A0 → A → A00 → 0 be an exact sequence of p-primary
GS (k)-modules. Since cdp GS (k) = 1 by (8.3.3), the exact and commutative
diagram
H 1 (kS UVWXYZ[\]|k, A0 ) H 1 (kS |k, A) H 1 (kS |k, A00 ) 0
H 1 (kp , A0 ) H 1 (kp , A00 )

M M M
H 1 (kp , A) 0
T T T

shows that the surjectivity statement for A implies that for A00 , and that sur-
jectivity for A0 and A00 shows surjectivity for A. Any finite GS (k)-module A
with pA = 0 is a quotient of a module of the form IFp [G]n . Hence the result is
true for such modules. Any finite p-primary GS (k)-module has a composition
series with graded pieces annihilated by p. Hence the result is true for all finite
p-primary GS (k)-modules. Finally, the general case follows by passing to the
inductive limit over all finite submodules. 2
(9.2.6) Corollary. Let k be a global field of characteristic p > 0 and let T

be a finite set of primes of k. If k T is the maximal extension of k which is
completely decomposed at the primes of T , then
cdp G(k T |k) = 1 .
Proof: Since the Galois group GkT is generated by the decomposition groups
GkP , P|p and p ∈ T , the map res 2 in the commutative diagram
c_àb^ A) res 1 M
H 1 (k, H 1 (kp , A)
p∈ T
res 2
T
H 1 (k, A) H 1 (k T , A)G(k |k) H 2 (k T |k, A) 0
T
is injective; here A is a p-primary G(k |k)-module, also considered as a Gk -
module. The zero in the lower exact sequence is a consequence of (6.5.10).
Since res 1 is surjective by the theorem above, it follows that res 2 is an isomor-
phism, and therefore H 2 (k T |k, A) = 0. Hence cdp G(k T |k) ≤ 1 by (3.3.2). As
the p-Sylow subgroups of G(k T |k) are nontrivial, we obtain the result. 2
For the trivial module A = ZZ/pr ZZ we are going to prove a more far reaching
surjectivity result.
(9.2.7) Theorem. Let k be a global field, p a prime number and S a set of

primes of k with δ(S) = 1. Let T0 and T be disjoint subsets of S such that T0
is finite and δ(T ) = 0.
If k is a function field or if k is a number field and we are not in the special
case (k, pr , S r(T0 ∪ T )), then the canonical homomorphism
H 1 (GS (k), ZZ/pr ZZ) −→ H 1 (kp , ZZ/pr ZZ) ⊕ H 1 (Tp , ZZ/pr ZZ)Gp
M M
p∈T0 p∈T
is surjective, where Gp = G(k̄p |kp ) and Tp ⊆ Gp is the inertia group.

If k is a number field and we are in the special case (k, pr , S r(T0 ∪ T )),
then the cokernel of this map is of order 1 or 2.

Proof: By convention, Tp = Gp , hence H 1 (Tp , ZZ/pr ZZ)Gp = H 1 (kp , ZZ/pr ZZ),

if p is archimedean. Therefore we may move all archimedean primes of T to
T0 and assume T ∩ S∞ = ∅. We first assume that S ⊇ Sp ∪ S∞ if k is a number
field. In the following we write H i (−) for H i (−, ZZ/pr ZZ).
Let T1 ⊆ T be a finite subset. Noting that H 1 (kp ) → H 1 (Tp )Gp is surjective,
(9.2.3)(ii) and (9.2.5) imply the result for the natural map
H 1 (Tp )Gp .
M M
H 1 (G(S r T )∪T1 (k)) −→ H 1 (kp ) ⊕
p∈T0 p∈T1
For T1 ⊆ T2 ⊆ T , we obtain the commutative diagram

H 1 (G(S defg r T )∪T2 ) H 1 (Tp )Gp
M M
H 1 (kp ) ⊕
p∈T0 p∈T2
H 1 (Tp )Gp .
M M
H 1 (G(S r T )∪T1 ) H 1 (kp ) ⊕
p∈T0 p∈T1
Passing to the limit over all finite subsets of T , we obtain the desired result.
It remains to consider the case when k is a number field and S does not
contain Sp ∪ S∞ . Let V = (Sp ∪ S∞ ) r S. By what we just proved, the map
H 1 (Tp )Gp ,
M M
H 1 (GS∪V (k)) −→ H 1 (kp ) ⊕
p∈T0 ∪V p∈T
is surjective, unless we are in the special case (k, pr , S r(T0 ∪ T )) where the
cokernel is of order 1 or 2. Since H 1 (kS∪V |kS )GS (k) ,→ p∈V H 1 (Tp )Gp is
L
injective, the same statement as above holds true also for the map
H 1 (Tp )Gp .
M M M
H 1 (GS (k)) −→ H 1 (kp ) ⊕ 1
Hnr (Gp ) ⊕
p∈T0 p∈V p∈T
This implies the desired result. 2
The following assertion is known as the theorem of GRUNWALD-WANG,

who, however, only considered the case of a cyclic group A.
(9.2.8) Theorem. Let S be a finite set of primes of a global field k and let
A be a finite abelian group. Let for all p ∈ S finite abelian extensions Kp |kp
be given such that G(Kp |kp ) may be embedded into A. If k is a number field
assume we are not in the special case (k, exp(A), P(k) r S), where P(k) is the
set of all places of k and exp(A) denotes the exponent of A.
Then there exists a global abelian extension K|k with Galois group A such
that K has the given completions Kp for all p ∈ S.

Proof: We show the following stronger statement: the map

M
Epi(Gk , A) −→ Hom(Gkp , A)
S
is surjective, where Epi(Gk , A) denotes the set of surjective homomorphisms
from Gk onto A.
Let q1 , . . . , qr be primes not in S (and not dividing 2 in the number field
case), and let
ϕqi : Gkqi −→ A
be homomorphisms such that the images of the ϕqi generate the group A. For
each prime p ∈ S, let ϕp : G(Kp |kp ) ,→ A be an embedding of the local group
into A. Denote by S 0 the union S ∪ {q1 , . . . , qr } and let T and T 0 be the
complements of S and S 0 in P(k).
Since δ(T ) = δ(T 0 ) = 1, (9.1.8) implies that the special cases (k, 2r , T ) and
(k, 2r , T 0 ) can only occur if k is a number field and k(µ2r )|k is not cyclic.
Then all primes p - 2 decompose in k(µ2r )|k and therefore (k, 2r , T ) and
(k, 2r , T 0 ) are equivalent. Furthermore (k, 2r , T ) implies (k, 2r+1 , T ). Decom-
posing A into a product of cyclic groups, (9.2.3) (ii) and (9.2.5) imply the
surjectivity of the map H 1 (k, A) −→ S 0 H 1 (kp , A) if we are not in the spe-
L
cial case (k, exp(A), T 0 ) = (k, exp(A), T ). Now a pre-image (ϕ : Gk → A) ∈

Hom(Gk , A) = H 1 (k, A) of
M M
(ϕq1 , . . . , ϕqr , ϕp , p ∈ S) ∈ Hom(Gkp , A) = H 1 (kp , A)
S0 S0
realizes the local extensions Kp |kp and it is surjective by the choice of the
homomorphisms ϕqi . 2
Remark: The conclusion of (9.2.8) may hold even if we are in the special case
(k, exp(A), P(k) r S). By (9.2.2) this is exactly the case when
X1 (k̄, P(k), A0 ) ∼
= X1 (k̄, P(k) r S, A0 ).
Finally, we give an application to embedding problems with induced G-

modules as kernel.
(9.2.9) Proposition. Let K|k be a finite Galois extension of global fields with
Galois group G = G(K|k) and let A = IFp [G]n , where p is a prime number.
Then the embedding problem
hijkl Gk
1 A E G 1
is properly solvable.

§3. Construction of Cohomology Classes 543
Proof: Since H 2 (G, A) vanishes, the embedding problem has a solution

ψ0 : Gk −→ E, see (3.5.9). Let p1 , . . . , pr be primes of k which split com-
pletely in K and let ϕi : Gkpi −→ A be homomorphisms such that their images
generate A. By (9.2.4) and (9.2.5) the natural map
r
M
H 1 (k, A) −→ H 1 (kpi , A)
i=1
is surjective. Let [x] ∈ H 1 (k, A) be a 1-class such that res i [x] = ϕi − ψ0 |Gkp
i
for i = 1, . . . , r. Then ψ = x·ψ0 : Gk −→ E is a new solution of the embedding
problem; this is proper, since ψ|Gkp = ϕi for i = 1, . . . , r, and so ψ(GK ) = A.
i
2
Exercise 1. Let k be a global field. Prove that every finite abelian group A occurs as a Galois
group of a finite abelian extension of k.
Exercise 2. Prove the following generalization of the theorem of GRUNWALD-WANG.
Let S be a finite set of primes of a global field k and let A be a finite abelian group. Let
for all p ∈ S finite abelian extensions Kp |kp be given such that G(Kp |kp ) may be embedded
into A. Let T be a set of primes of k of density zero which is disjoint to S. Assume we are
not in the special case (k, exp(A), P(k) r(S ∪ T )).
Then there exists a global abelian extension K|k with Galois group A such that K has the
given completions Kp for all p ∈ S and K|k is unramified at all p ∈ T .
§3. Construction of Cohomology Classes
The aim of this section is to establish the existence of global cohomo-

logy classes with given local behaviour. More precisely, given classes xp ∈
H 1 (kp , A) for all p in a finite set of primes T , we are looking for a global class
x ∈ H 1 (k, A) which maps to xp for all p ∈ T and satisfies additional conditions
at all other primes. We start by introducing some notation.
(9.3.1) Definition. Let kp be a local field and let A be a Gkp -module.

(i) We call a class xp ∈ H 1 (kp , A) cyclic if it is split by a cyclic extension
of kp , i.e. if there exists a cyclic extension Kp |kp such that xp lies in the
kernel of the restriction map H 1 (kp , A) → H 1 (Kp , A).
(ii) We call xp unramified if it is split by an unramified extension, i.e. if it is
1
contained in the unramified part Hnr (kp , A) of H 1 (kp , A). In particular,
if xp is unramified, then it is cyclic.

For A = µp the following theorem is equivalent, via Kummer theory, to

Šafarevič’s theorem about the existence of certain algebraic numbers ([188])
and we shall make use of the ideas of Šafarevič’s proof.
(9.3.2) Theorem. Let k be a global field and let p be a prime number different
to char(k). Let K ⊆ Ω be finite Galois extensions of k and assume that K
contains the group µp of p-th roots of unity. Let S0 be a nonempty finite set of
primes of k containing Sp ∪ S∞ if k is a number field. Let T be a finite set of
primes of k containing Ram(Ω|k) ∪ S0 and let S = cs(Ω|k) ∪ T .
Let A be a finite IFp [G(K|k)]-module and assume that we are given a class y
in H 1 (kS |K, A) such that
yP is unramified for P ∈ T (K) and
yP = 0 for P ∩ k ∈ Ram(K|k) ∪ S0 .
Then there exists an element x ∈ H 1 (kS |k, A) such that
xp = (corKk y)p for p ∈ T and xp is cyclic for all p ∈/ T.
Proof: Setting x = corK 1

k z, it suffices to construct z ∈ H (kS |K, A) with
(a) zP = yP for P ∈ T (K),
1
(b) if P ∈/ T (K) and zP is ramified (i.e. not contained in Hnr (KP , A)), then
zP is cyclic and zσP = 0 for every σ ∈ G(K|k) r{1}.
We first prove the existence of z in the case A = µp , when the cyclicity
condition for zP is trivially satisfied. Assume first that p is odd. We will apply
the method of [188] in order to obtain the element z which we are looking for,
as a sum of two members of a sequence
z1 , z2 , z3 , . . . ∈ H 1 (kS |K, µp )
which will be constructed having the following properties.
(1) There exists a prime Pi ∈ S r T (K) such that
1
· (zi )Pi ≡ FrobPi modulo Hnr (KPi , µp ),
1
· (zi )P ∈ Hnr (KP , µp ) for all P =/ Pi ,
(2) (zi )P = 21 yP for P ∈ T (K),
(3) (zn+1 )σPi = −(zi )σPi for i ≤ n and all σ ∈ G(K|k) r{1}.
In (1) we view FrobP as an element of H 1 (KP , µp )/Hnr
1
(KP , µp ) via
G(K nr |KP ) H 1 (KP , ZZ/pZZ)∨ ∼
P nr = H 1 (KP , µp )/H 1 (KP , µp ),
nr
where the isomorphism is induced by the local duality theorem; see (7.2.6)
and (7.2.15). Assume that we have already constructed z1 , . . . , zn (n ≥ 0) and
set Tn = T ∪ {(P1 ∩ k), . . . , (Pn ∩ k)}, i.e. Tn (K) consists of T (K) and of
all G(K|k)-conjugates of P1 , . . . , Pn . Observe that Tn ⊆ S and consider the
commutative diagram

mnopqrstuv µp )
Y
H 1 (kS |K, H 1 (KP , µp ) H 1 (kS |K, ZZ/pZZ)∨ .
S
Y M
H 1 (kS |K, µp ) H 1 (KP , µp ) × H 1 (KP , µp )/Hnr
1
H 1 (kTn |K, ZZ/pZZ)∨
Tn S\Tn
Y
H 1 (KP , µp ) H 1 (Ω|K, ZZ/pZZ)∨
T
The upper row is part of the long exact sequence of Poitou-Tate, and the
exactness of the middle row follows in a straightforward manner. By definition
of S, all P ∈ S r T (K) are completely decomposed in Ω|K. Therefore, for
P ∈ S r T (K), the homomorphism
H 1 (KP , µp ) −→
∼ H 1 (K , Z
P Z/pZZ)∨ −→ H 1 (Ω|K, ZZ/pZZ)∨
is zero, showing that the lower square in the diagram commutes. Consider the
element Y M
ξ∈ H 1 (KP , µp ) × H 1 (KP , µp )/Hnr
1
Tn S\Tn
1
given by ξP = for P ∈ T (K), ξσPi = −(zi )σPi for i ≤ n, σ ∈ G(K|k) r{1}
y
2 P
and ξP = 0 for all other P. Then ξ has the same image in H 1 (Ω|K, ZZ/pZZ)∨
as 12 y ∈ H 1 (kS |K, µp ). By the exactness of the rows, this image is trivial.
Let KTeln be the maximal elementary abelian p-extension of K inside kTn . We
consider the diagram of fields
wxyz{|} kTn
KTeln Ω
KTeln Ω
KTeln ∩ Ω
k.
Let η be the image of ξ in H (kTn |K, ZZ/pZZ)∨ = G(KTeln |K). As the image
1
of ξ in H 1 (Ω|K, ZZ/pZZ)∨ is zero, η lies in the subgroup G(KTeln |KTeln ∩ Ω). By

Čebotarev’s density theorem applied to KTeln Ω|k, we can choose a prime P•
in KTeln Ω outside Tn (KTeln Ω) such that FrobP• |KTel = −η and FrobP• |Ω = 1.
n
Hence P• ∈ S r Tn (KTeln Ω). We put Pn+1 = P• |K ∈ S r Tn (K). Then the
image of −ξ in H 1 (kTn |K, ZZ/pZZ)∨ is equal to FrobPn+1 . By the exactness of
the middle row, we find a class zn+1 ∈ H 1 (kS |K, µp ) with properties (1), (2), (3).

Let G(K|k) r{1} = {σ1 , . . . , σr } and consider the maps

r
Y
ψn : {z1 , . . . , zn } −→ µp ,
1
for n = 1, 2, . . . given by

ψn (zi ) = (zi )σ1 Pi (Frobσ1 Pi ), . . . , (zi )σr Pi (Frobσr Pi ) .
1
Observe that by construction zi ∈ Hnr (Kσj Pi , µp ) for j = 1, . . . , r. By the
pigeonhole principle, there exists an N with ψN (zN ) = ψN (zi ) for some i < N .
We claim that
z = zi + zN ∈ H 1 (kS |K, µp )
satisfies conditions (a) and (b) above. Indeed, (a) is trivial by condition (2). It
therefore remains to show that if zP is ramified for some P, then P ∈ S r T (K)
and zσP = 0 for σ ∈ G(K|k) r{1}. By construction, z is only ramified at Pi
and PN , and by condition (3), we know that zσPi = 0 for σ =/ 1. In order
to show the corresponding statement at σPN (σ =/ 1), recall that for arbitrary
classes a, b ∈ H 1 (kS |K, µp ), we have the product formula
Y
(a, b)P = 1
P∈S(K)
for the Hilbert symbol. (The symbol (a, b)P is defined as the image of a ∪ b
under the trace homomorphism H 2 (KP , µ⊗2 ∼
p ) −→ µp .) Since zi and zN are
unramified at σPN , it suffices to show that their values on FrobσPN are mutually
inverse in µp . We have
zN (FrobσPN ) = zi (FrobσPi ) because ψn (zi ) = ψn (zN ),
= (zi , σzi )σPi by condition (1) for zi ,
= (zi , σzi )−1
Pi by the product formula and (1), (2),
= (zi , σzN )Pi by condition (3),
= (zi , σzN )−1
σPN by the product formula and (1), (2),
= zi (FrobσPN )−1 by condition (1) for zN .
This finishes the case A = µp , for p odd.
In the case p = 2 we have to modify the construction, and we will obtain z
as a sum of three other elements. We use the combinatorial method of [85],
chap.5, §3.
Let {G1 , G2 , G3 } be a partition of the set G(K|k) r{1} such that G1 consists
of all elements of order 2 and G2 = G−1 3 .
We construct recursively a sequence z1 , z2 , . . . of elements in H 1 (kS |K, µ2 )
satisfying the following properties:
(1) There exists a prime Pi ∈ S r T (K) such that
1
· (zi )Pi ≡ FrobPi modulo Hnr (KPi , µ2 ),
1
· (zi )P Hnr (KP , µ2 ) for all P =/ Pi ,
∈
(2) (zi )P = yP for P ∈ T (K),

(3) (zn+1 )σPi = 0 for i ≤ n and all σ ∈ G1 .

(4) If ψn (zi ) =/ ψn (zj ) for all j with
( n j > i, then
≥
0 if σ ∈ G2 ,
(zn+1 )σPi =
(σ 2 zi )σPi if σ ∈ G3 ,
(
(zi )σPi if σ ∈ G2 ,
and otherwise (zn+1 )σPi =
0 if σ ∈ G3 .
The existence of the sequence of classes z1 , z2 , . . . is proved similarly to the
case of odd p. In addition, we obtain
Claim: (zi )σPi = 0 for σ ∈ G1 .
Proof of the claim: Let z̃i ∈ K × be a representative of zi ∈ H 1 (kS |K, µ2 )
⊆ H 1 (K, µ2 ) ∼ = K × /K ×2 . By condition (1), we have z̃i OK,S0 = Pi A2 for
some fractional ideal A of the Dedekind ring OK,S0 . By Čebotarev’s density
theorem, there exists a prime ideal Q of K with Q ∈/ (T (K) ∪ {Pi , σPi })
and Q =/ σQ such that Q = A · xOK,S0 with x ∈ K × . Hence we have
z̃i x2 OK,S0 = Pi Q2 in OK,S0 . Replacing z̃i by z̃i x2 and using condition (2) we
obtain
×2
z̃i ∈ OK,S0 , z̃i OK,S0 = Pi Q2 and (z̃i )P ∈ KP for P ∈ S0 .
In addition, choose δ ∈ OK,S0 such that K = K σ (δ) and δ 2 ∈ OK σ ,S0 , where
K σ is the fixed field of K with respect to hσi. Then there are a, b ∈ K σ with
z̃i = a + bδ, σ z̃i = a − bδ. Since S10 ⊆ S0 , we have a ∈ OK σ ,S0 and bδ ∈ OK,S0 .
Note that a =/ 0 and b =/ 0, because otherwise σPi would divide z̃i = ±σ z̃i .
Let Σ be the set of prime divisors of bδ in OK,S0 which are not in Ram(K|k).
Since vσPi (z̃i ) = 0 and vσPi (σ z̃i ) = 1, we conclude vσPi (2bδ) = 0, in particular,
σPi ∈/ Σ. Furthermore,
(∗) z̃i−1 (2bδ) = 1 − z̃i−1 σ z̃i ∈ UσP
1
i
.
×2
For P ∈/ (Σ ∪Ram(K|k)∪S0 ), P =/ σPi , we have 2bδ ∈ UP and σ z̃i ∈ UP KP .
×2
Hence (2bδ, σ z̃i )P = 1. For P ∈ Ram(K|k) ∪ S0 we have σ z̃i ∈ KP , hence
(2bδ, σ z̃i )P = 1 also for these primes. Therefore the product formula for the
Hilbert symbol implies the equality
Y
(∗∗) (2bδ, σ z̃i )σPi = (2bδ, σ z̃i )P .
P∈Σ
Since z̃i and σ z̃i are coprime in OK,S0 , we have vP (a) = 0 for P ∈ Σ and
therefore
(∗∗∗) a−1 σ z̃i = 1 − a−1 bδ ∈ UP1 for P ∈ Σ .
We obtain
zi (FrobσPi ) = (z̃i , σ z̃i )σPi by definition of z̃i ,
= (2bδ, σ z̃i )σPi by (∗),
Q
= P∈Σ (2bδ, σ z̃i )P by (∗∗),
Q
= P∈Σ (2bδ, a)P by (∗∗∗).

The last product is easily seen to be unity: For P ∈ Σ, the element a ∈ K σ

is a unit in the completion with respect to P ∩ K σ . If P is inert in K|K σ ,
σ
then a is a square in the unramified quadratic extension KP of KP∩K σ , so that
σ
(2bδ, a)P = 1. If P Σ splits in K|K , then σP Σ and
∈ ∈
(2bδ, a)P · (2bδ, a)σP = (2bδ, a)P · (−2bδ, a)P = (−1, a)P = 1.
This proves the claim.
Now choose N minimal such that
ψN (zi ) = ψN (zj ) = ψN (zN )
for numbers i < j < N . We claim that z = zi + zj + zN satisfies conditions
(a) and (b). Indeed, (a) follows immediately from (2). It therefore remains to
show that if zP is ramified for some P, then P ∈ S r T (K) and zσP = 0 for
σ ∈ G(K|k) r{1}. By construction, z is only ramified at Pi , Pj and PN .
For σ ∈ G1 and N ≥ s, t ≥ 1, we have
zs (FrobσPt ) = 1,
which is seen for s > t by condition (3), for s = t by the claim and follows for
s < t by (1)-(3) and the product formula
zs (FrobσPt ) = (zs , σzt )σPt = (σzs , zt )Pt = (σzs , zt )σPs = zt (FrobσPs ) = 1.
Summing up, we obtain zσPi = zσPj = zσPN = 0 for σ ∈ G1 .
If σ G2 , then by condition (4)
∈
zσPi = (zi )σPi + (zj )σPi + (zN )σPi = (zi )σPi + 0 + (zi )σPi = 0.
Furthermore, since (zN )σPj = 0 by condition (4) and
zi (FrobσPj ) = (zi , σzj )σPj by condition (1),
= (zi , σzj )Pi by the product formula,
−1
= (σ zi , zj )σ−1 Pi using Galois invariance,
= zj (Frobσ−1 Pi ) by condition (1),
−2
= σ zi (Frobσ−1 Pi ) by condition (4),
= zi (FrobσPi ) using Galois invariance,
= zj (FrobσPj ) because ψN (zi ) = ψN (zj ) ,
we obtain
zσPj = (zi )σPj + (zj )σPj + (zN )σPj = (zi )σPj + (zj )σPj + 0 = 0 .
Finally,
zσPN = (zi )σPN + (zj )σPN + (zN )σPN = 0,
since
zi (FrobσPN ) = (zi , σzN )σPN by condition (1),
= (zi , σzN )Pi by the product formula,
−1
= (σ zi , zN )σ−1 Pi using Galois invariance,
= zN (Frobσ−1 Pi ) by condition (1),
= 0 by condition (4), σ −1 ∈ G3 ,

and
zj (FrobσPN ) = zN (Frobσ−1 Pj )
= σ −2 zj (Frobσ−1 Pj ) by condition (4), σ −1 ∈ G3 ,
= zj (FrobσPj ) using Galois invariance,
= zN (FrobσPN ) because ψN (zN ) = ψN (zj ).
In the same way one verifies that the local classes zσPi , zσPj and zσPN also
vanish if σ ∈ G3 . This finishes the proof for A = µp .
The general case will be proven by induction on dimIFp A. Let A = A0 ⊕ µp

(as an abelian group). For each z ∈ H 1 (KS |K, A), let z = z 0 + z 00 be the decom-
position of z into the components z 0 ∈ H 1 (KS |K, A0 ) and z 00 ∈ H 1 (KS |K, µp ),
0 00
and similarly zP = zP + zP for zP ∈ H 1 (KP , A). By induction we find an
0 ∈ 1 0
element z H (KS |K, A ) such that
(a0 ) zP
0
= yP0
for P ∈ T (K),
0 0 0 0
(b ) if P T (K) and zP
∈
/ is ramified, then zP is cyclic and zσP = 0 for every
σ ∈ G(K|k) r{1}.
Let K 0 |K be the extension defined by the homomorphism z 0 , so that
z 0 : G(kS |K) G(K 0 |K) ⊆ A0 ,
and let K̃ be its Galois closure over k. By construction, K 0 |K, and hence also
K̃|K, is unramified at all primes in T (K) and it only ramifies at primes in
cs(Ω|k)(K).
Set T 0 = T ∪ Ram(K̃|k), Ω 0 = K̃Ω, S 0 = cs(Ω 0 |k) ∪ T 0 and suppose we have
found a class ỹ ∈ H 1 (KS 0 |K, µp ) with
00
· ỹP = yP for P ∈ T (K),
· ỹP = 0 for P ∈ T 0 r T (K).
Then we can apply the induction hypothesis to the extensions Ω 0 |K|k, the sets
T 0 , S 0 and the module A = µp , in order to find a class z 00 in H 1 (KS 0 |K, µp )
with
(a00 ) zP 00
= ỹP for P ∈ T 0 (K),
(b00 ) if P ∈/ T 0 (K) and zP 00 00
is ramified, then zP 00
is cyclic and zσP = 0 for every
σ ∈ G(K|k) r{1}.
Noting that T 0 r T ⊆ cs(Ω|k), so that S 0 ⊆ S, it is now easily verified that the
class z = z 0 + inf z 00 ∈ H 1 (KS |K, A) satisfies conditions (a) and (b). Indeed,
for P ∈ T (K) we have
0 00 0 00
zP = zP + zP = yP + yP = yP .
For P ∈ T 0 r T we get
0 00 0
zP = zP + zP = zP .
0
Therefore zP is cyclic and if zP = zP is ramified, then the underlying prime
0
p of P splits completely in Ω and zσP = zσP = 0 for σ ∈ G(K|k) r{1}.

Finally suppose P ∈/ T 0 . If zP is unramified, then it is cyclic. Suppose zP is

0 00 0 00
ramified. Since zP = zP + zP and zP is unramified, zP must be ramified. Thus
0 0 0
p splits completely in Ω , and hence in K . From this we obtain zσP = 0 for
0 1 0 0
all σ G(K|k), since the element z becomes zero in H (K , A ) by definition
∈
of K 0 and thus zσP
0
is zero in H 1 (KQ 0
, A0 ) = H 1 (KσP , A0 ), where Q is a prime
of K 0 above σP. Therefore zP = zP 00 00
, i.e. zP is cyclic and zσP = zσP = 0 for
σ ∈ G(K|k) r{1}.
It therefore remains to construct a class ỹ with the above properties. Consider
the commutative exact diagram
∨
~
Y Y
H 1 (KP , µp ) H 1 (KP , ZZ/pZZ)
S 0 \T 0 S 0 \T 0
Y
H 1 (kS 0 |K, µp ) H 1 (KP , µp ) H 1 (kS 0 |K, ZZ/pZZ)∨
S0
α
Y
H 1 (KP , µp ) H 1 (Ω 0 |K, ZZ/pZZ)∨ .
T0
If we can show that α annihilates the element ξ = (ξP )P∈T 0 (K) of the group
Q 1 00
T 0 H (KP , µp ) given by ξP = yP for P ∈ T (K) and ξP = 0 for P ∈
0
T r T (K), then the existence of ỹ follows by diagram chasing. We use the
injection
H 1 (Ω 0 |K, ZZ/pZZ)∨ ,→ H 1 (Ω|K, ZZ/pZZ)∨ ⊕ H 1 (K̃|K, ZZ/pZZ)∨
in order to write the image of ξ in the form α(ξ) = (α1 (ξ), α2 (ξ)). Since K̃|K
is unramified at all P ∈ T (K), α2 factors through the quotient
Y Y Y
H 1 (KP , µp )/( Hnr1
(KP , µp ) × {0}).
T0 T T 0 \T
Hence α2 (ξ) = 0. Finally, the diagram
α
Y
H 1 (KP , µp ) H 1 (Ω 0 |K, ZZ/pZZ)∨
T0
Y
H 1 (KP , µp ) H 1 (Ω|K, ZZ/pZZ)∨
T
Y
H 1 (kS |K, µp ) H 1 (KP , µp ) H 1 (kS |K, ZZ/pZZ)∨
S
shows that α1 (ξ) is equal to the image of y 00 ∈ H 1 (kS |K, µp ) in the group
H 1 (Ω|K, ZZ/pZZ)∨ , hence is trivial by the exactness of the lower row. This
finishes the proof. 2

The proof of the next theorem uses ideas of [158].
(9.3.3) Theorem. Let k be a global field and let A be a simple Gk -module

annihilated by the prime number p =/ char(k). Let S be a set of primes of k
containing Sp ∪ S∞ if k is a number field. Assume there exist finite Galois
extensions K ⊆ Ω of k such that p - [Ω : K],
(a) µp ⊆/ K and k(A) ⊆ K,
(b) µp ⊆ Ω, Ram(Ω|k) ⊆ S and cs(Ω|k) ⊂ ∼ S.
Let T ⊆ S be a finite subset and let local classes yp ∈ H 1 (kp , A) for p ∈ T be

given. Then there exists an element x ∈ H 1 (kS |k, A) such that
(i) xp = yp for p ∈ T ,
(ii) if p ∈/ T then xp is cyclic, and if xp is ramified then p splits completely
in Ω.
Proof: By condition (b), S is nonempty. Let S0 ⊆ S be a finite nonempty

subset which contains Sp ∪ S∞ if k is a number field.
If the assertion of the theorem holds for a finite set T 0 ⊆ S with T ⊆ T 0 , then it
also holds for T (impose the local conditions yp = 0 for p ∈ T 0 r T ). Therefore
we may assume Ram(Ω|k) ∪ S0 ⊆ T ; in particular, A is a GT (k)-module. Let
S1 = (S ∩ cs(Ω|k)) ∪ T . By (9.2.3)(vi), there exists an element
z ∈ H 1 (kS1 |k, A) ⊆ H 1 (kS |k, A)
such that zp = yp for p ∈ T . Let
V = T ∪ {p ∈ S1 | zp is ramified }.
Note that all p ∈ V r T split completely in Ω. Thus it remains to show that
we can choose the element z in such a way that the ramified local classes are
cyclic.
We introduce two additional fields. Let z 0 be the image of z under the
restriction map H 1 (kS1 |k, A) → H 1 (kS1 |K, A)G(K|k) . Then z 0 is a G(K|k)-
homomorphism z 0 : GS1 (K) → A and defines a finite abelian p-extension K 0 |K
which is Galois over k. We set Ω 0 = K 0 Ω. It follows that Ω 0 |k is Galois,
µp ⊆/ K 0 and p - [Ω 0 : K 0 ]. Furthermore, Ram(Ω 0 |k) ⊆ V , since the set
Ram(Ω|k) is contained in T and all primes p with zp ramified are in V . Let
S10 = (S ∩ cs(Ω 0 |k)) ∪ V ⊆ S1 .
0 ⊂ 0
Since cs(Ω|k) ⊂ ∼ S, we have cs(Ω |k) ∼ S1 . Suppose we have constructed a
1 1
class ξ ∈ H (kS10 |k, A) ⊆ H (kS |k, A) such that
(1) ξp = 0 for p ∈ T ,
(2) zp + ξp is cyclic for p ∈ V r T ,
(3) ξp is cyclic for p ∈/ V .

We verify that x = z + ξ ∈ H 1 (kS |k, A) satisfies conditions (i) and (ii) of the
theorem. For p ∈ T we have xp = zp = yp . If p ∈ V r T , then xp = zp + ξp
is cyclic and p splits completely in Ω. Let p ∈/ V . Then zp is unramified.
If xp is unramified, it is cyclic. So assume that xp is ramified. Then also
ξp = xp − zp is ramified, and so p ∈ S10 r V . Hence p splits completely in Ω 0 |k,
in particular, it splits completely in K 0 |k. By construction of K 0 , the image of
z in H 1 (kS1 |K 0 , A) is zero. If P is any prime of K 0 above p, we obtain that zp
0
is zero in H 1 (KP , A) = H 1 (kp , A). Thus we have xp = ξp , and xp is cyclic.
Therefore it remains to construct a class ξ ∈ H 1 (kS10 |k, A) satisfying condi-
tions (1)–(3) above.
1
Claim. For p ∈ V r T there exists a local class wp ∈ Hnr (kp , A) such that
zp + wp is cyclic.
Proof of the claim: The prime p splits completely in Ω, hence µp ⊆ kp and A
is a trivial Gkp -module. Since p ∈/ S0 , the IFp -vector space H 1 (kp , ZZ/pZZ) is
2-dimensional. Therefore, if κ|kp is a cyclic ramified extension of degree p, the
natural map H 1 (κ|kp , A) ⊕ Hnr
1
(kp , A) → H 1 (kp , A) is an isomorphism. This
shows the claim.

H 1 (kS10 |K(µ p ), A)
res cor
ε
Y
H 1 (kS10 |K, A) H 1 (KP , A)
P∈V (K)
cor cor
Y
H 1 (kS10 |k, A) H 1 (kp , A) .
p∈V
0 0 ⊂ 0 0 0 ⊂ 0
Since cs(Ω |K) ⊂∼ cs(Ω |k)(K) and cs(Ω |k) ∼ S1 , we have cs(Ω |K) ∼ S1 (K).
As A is a trivial GS10 (K)-module, (9.2.3)(vi), applied to the simple module

IFp and the Galois extensions K 0 ⊆ Ω 0 of k, implies that the natural map ε is
surjective.
We consider the element
Y
1
(ηp )p∈V ∈ Hnr (kp , A), with ηp = 0 for p ∈ T and ηp = wp for p ∈ V r T,
p∈V
1
where the elements wp ∈ Hnr (kp , A) are chosen as in the claim. Since all
p ∈ V r T split completely in K|k, we have a pre-image
Y Y
1
(η̃P )P∈V (K) ∈ Hnr (KP , A) ⊆ H 1 (KP , A)
P∈V (K) P∈V (K)

§4. Local Galois Groups in a Global Group 553
of (ηp )p∈V under the corestriction map. Let η̃ ∈ H 1 (kS10 |K, A) be a pre-image
of (η̃P )P∈V (K) under ε and let d = [K(µp ) : K]. Note that (d, p) = 1 and
consider the class
y = res K 1
K(µp ) (η̃/d) ∈ H (kS10 |K(µp ), A).
1
We have yP = 0 for P ∈ T (K(µp )) and yP ∈ Hnr (kS10 |K(µp ), A) for P ∈
1
V r T (K(µp )). By (9.3.2), we find a class ξ ∈ H (kS10 |k, A) with
K(µp )
ξp = (cor k y)p = (cor K
k η̃)p = ηp for p ∈ V
such that ξp is cyclic for all p ∈

/ V . The class ξ satisfies conditions (1)–(3).
§4. Local Galois Groups in a Global Group
With the notation of the preceding sections we now consider the maximal
pro-c-factor group Gk (c) of Gk , where c is any full class of finite groups.
(9.4.1) Definition. Let k be a field, k̄ a separable closure of k and c a full class

of groups. We denote by k(c) the composite of all finite Galois extensions K
of k inside k̄ with G(K|k) ∈ c.
If k is a global field and S is a set of places of k, we denote by kS (c) the
composite of all finite Galois extensions K of k inside kS with G(K|k) ∈ c.
The Galois group G(k(c)|k) is nothing else but the maximal pro-c-factor
group of Gk . Similarly, G(kS (c)|k) is the maximal pro-c-factor group of
GS (k), cf. III §5.
(9.4.2) Lemma.
(i) The field k(c) admits no finite Galois extension K with G(K|k(c)) ∈ c.
(ii) Let k = k0 ⊆ k1 ⊆ · · · ⊆ kn be finite extensions of k in k̄ such that
ki+1 |ki is Galois and G(ki+1 |ki ) ∈ c for i = 0, . . . , n − 1. Then kn ⊆ k(c);
in particular, G(k̃n |k) ∈ c, where k̃n is the Galois closure of kn over k.

Proof: Assertion (i) follows from (3.5.4)(i), and (ii) follows from (i). 2
If k is a global field, we are interested in the question of whether for a given

prime p of k the local Galois group Gkp (c) is a subgroup of Gk (c). This is
equivalent to the question of whether the equality
(k(c))p = kp (c)
holds. If c is the class of all finite groups, i.e. k(c) is the separable closure k̄ of
k, then we have seen this in (8.1.5). If c is arbitrary, we will prove a slightly
more general statement using the theorem of Grunwald-Wang.
(9.4.3) Theorem. Let k be a global field and let c be a full class of finite
groups. Let S be a set of primes of k of density δ(S) = 1. Then, for the
maximal c-extension kS (c) of k unramified outside S and a prime p ∈ S, we
have
(kS (c))p = kp (c),
or equivalently, the canonical map
Gkp (c) Gp (kS (c)|k) ⊆ GS (k)(c)
is injective.
Proof: We have to show that for every finite Galois extension Kp |kp with
G(Kp |kp ) ∈ c there exists a global Galois extension L|k unramified outside S
with G(L|k) ∈ c such that Kp ⊆ Lp . Since Gkp is a prosolvable group and by
(9.4.2)(ii), we may assume that G(Kp |kp ) ∼ = ZZ/pZZ for a prime number p with
ZZ/pZZ ∈ c. Choose a class αp ∈ H 1 (kp , ZZ/pZZ), α =/ 0, which is split by Kp ,
i.e. res αp ∈ H 1 (Kp , ZZ/pZZ) is trivial. By (9.2.7), the homomorphism
H 1 (GS (k), ZZ/pZZ) −→ H 1 (kp , ZZ/pZZ)
is surjective. Let α ∈ H 1 (GS (k), ZZ/pZZ) be a pre-image of αp . Then α is split
by a cyclic extension K|k of degree p which realizes Kp . 2
In particular, if c is the class of p-groups and p a prime number, then Gkp (p)
injects into Gk (p) by the theorem above. Now we ask the much deeper question
for the interaction of finitely many local subgroups GkP (p) in the global group
Gk (p) (where we index these groups by primes P of k(p)).
A first question is whether the intersection of GkP (p) and GkP0 (p) for dif-
ferent primes P and P0 is trivial. We will prove that even more is true: under
a mild assumption, finitely many local groups GkP (p) are as independent

§4. Local Galois Groups in a Global Group 555
as possible, i.e. the subgroup which they generate inside Gk (p) is their free
pro-p-product.
Of course, one wants to know this not only for the maximal pro-p-factors,
but also for general classes c; in particular, if c is the class of all finite groups.
This statement is also true in a “measure theoretical sense”: If P1 , . . . , Pr
are primes of k̄, then for “almost all” r-tuples (σ1 , . . . , σr ) ∈ (Gk )r , the sub-
group hGkσ1 P1 , . . . , Gkσr Pr i ⊆ Gk is the free profinite product of the groups
Gkσi Pi , i = 1, . . . , r. This result is due to W.-D. GEYER; see [55] for a precise
statement. In view of (4.2.3), this assertion cannot be true for all r-tuples
(σ1 , . . . , σr ) ∈ (Gk )r .
For the case of pro-p-groups we now come to the statement mentioned above.
(9.4.4) Theorem. Let k be a global field, S a set of primes of k of density

δ(S) = 1 and kS (p) the maximal p-extension of k unramified outside S. Let
T ⊆ S(kS (p)) be a finite subset,
H := hGP (kS (p)|k), P ∈ T i ⊆ GS (k)(p)
the closed subgroup generated by the decomposition groups of the primes in T
and K = kS (p)H the fixed field of H.
(i) Let Q ∈ S(kS (p)) be a prime with Q|K =/ P|K for all P ∈ T . Then
KQ = kQ (p). In particular, Q splits completely in kS (p)|K.
(ii) Let T0 ⊆ T be a maximal subset such that the primes in T0 have pairwise
different restrictions to K. Then the natural map
∗
P∈T0
GkP (p) → GS (k)(p)
is injective, or equivalently, the subgroup H is the free pro-p-product

H= ∗
P∈T0
GkP (p).
Proof: By (9.4.3), for any prime P in kS (p), the decomposition group

GP (kS (p)|k) is the full local group GkP (p). Let
H0 := hGP (kS (p)|k), P ∈ T0 i ⊆ H .
Since the conjugates hH0 h−1 , h ∈ H, generate H, we obtain H0 = H, as H is
a pro-p-group, see (3.9.1). Therefore we may replace T by T0 .
Now let T 0 ⊆ S(kS (p)) be a finite set of primes having pairwise different
restrictions to K = kS (p)H . For a subfield k 0 ⊆ kS (p) we denote by T 0 (k 0 ) the
(finite) set of restrictions of the primes in T 0 to k 0 . Replacing k by a finite

extension inside K if necessary, we may assume that all primes of T 0 have

pairwise different restrictions to k. From (9.2.7) we obtain a surjection
H 1 (kS |k, ZZ/pZZ)
Y
H 1 (kp , ZZ/pZZ),
T 0 (k)
and passing to the direct limit we get a surjection

H 1 (kS |K, ZZ/pZZ)
Y
(∗) H 1 (kP (p)|KP , ZZ/pZZ).
T 0 (K)
0
If T ⊆ T , then the map (∗) is injective because the groups GkP (p), P ∈ T ,
generate H. Now let Q be a prime of kS (p) with Q|K ∈/ T (K). Using the
isomorphism (∗) for T 0 = T and T 0 = T ∪ {Q}, we obtain
H 1 (kQ (p)|KQ , ZZ/pZZ) = 0,
and so
G(kQ (p)|KQ ) = 1.
This shows (i).
In the extension kS (p)|K, it are exactly the decomposition groups GP (kS (p)|K)
= GkP (p) with P ∈ T (K) which are nontrivial. In the commutative diagram
H 2 (kS |K, ZZ/pZZ)

M
H 2 (KP , ZZ/pZZ)
S(K)
inf inf
M M
H 2 (H, ZZ/pZZ) H 2 (kP (p)|KP , ZZ/pZZ) = H 2 (GkP (p), ZZ/pZZ)
S(K) T (K)
the inflation map on the left is injective, since

H 1 (kS |kS (p), ZZ/pZZ) = Hom(G(kS |kS (p)), ZZ/pZZ) = 0,
i.e. kS (p) has no p-extension inside kS by definition. (The inflation map
on the right is an isomorphism by (7.5.8), but we do not need this.) From
(9.1.10)(i) and passing to the inductive limit, we obtain the injectivity of the
upper horizontal map in the diagram, hence the lower map is also injective.
Now (4.1.5) gives the result. 2
Remark: If T consists of two primes, then T0 = T , but if T contains at least

three primes, then T0 may be a proper subset of T .
Let us consider the case T = {P1 , P2 }, P1 =/ P2 , first. Let K1 and K2
be the decomposition fields of P1 and P2 in kS (p) and K = K1 ∩ K2 . If
P1 ∩ K =/ P2 ∩ K, then H is the free product of G1 and G2 by (9.4.4). Now
suppose that the restrictions to K are equal. Then G1 and G2 are conjugate
in H and therefore (say) G1 generates H as a normal subgroup. Since H
is a pro-p-group, we conclude that H = G1 = G2 and K = K1 = K2 . But

§5. Solvable Groups as Galois Groups 557
P1 ∩ K1 = P2 ∩ K2 has exactly one extension to kS (p), which shows P1 = P2

in contrast to our assumption.
To construct an example with a set of three primes, choose any element
σ H = G1 ∗ G2 not contained in G1 and put P3 = σP1 . Then G3 = σG1 σ −1
∈
is contained in H. Hence K = K1 ∩ K2 ∩ K3 and P1 |K = P3 |K . This gives
the counterexample.
Of greater arithmetical importance, but also much deeper lying, is the ques-
tion of analogous results in the case where we restrict the ramification to a
given finite set of primes. Is it true that (kS (p))p = kp (p), and do finitely many
decomposition groups GP (kS (p)|k) ∼ = Gkp (p) and their conjugates form a free
pro-p-product inside GS (k)(p), where S is a finite set of primes? This would
mean on the one hand that a local p-extension with respect to a prime p ∈ S
can be realized by a global p-extension which is unramified outside S and on
the other hand that finitely many local groups are as independent as possible
in the global group GS (k)(p). Most difficult is the situation in the number field
case for primes dividing p. We will consider this problem in X §5-§7.
§5. Solvable Groups as Galois Groups
Class field theory provides us with a complete solution for a great number of
the abstract and arithmetic problems concerning abelian extensions of global
fields. The next natural step forward to the general case is to consider solvable
extensions. The reason for this is that the solvable extensions are built up
by abelian extensions and one is necessarily led to the so-called embedding
problem in number theory: a given Galois extension K|k has to be embedded
in a larger extension L|k in such a way that the Galois groups of L|k and K|k
realize a given group extension. For a global field such an embedding problem
has an arithmetic structure. Each embedding problem defined over a global
field k canonically induces local embedding problems over the completions
kp . Hence the local-global question naturally arises: if all local problems are
solvable, is then the global one also solvable? This question is connected with
the Hasse principle studied in §1. A refinement of the above question is: does
there exist a global solution of an embedding problem which induces given
local solutions? Here the theorem of Grunwald-Wang comes into play.

One main result of this section is a theorem of K. IWASAWA [86] concerning

the structure of the Galois group of the maximal prosolvable extension k̃ over
some large field (e.g. over the maximal abelian extension of k). But first we
need the notion of a powerful global field.
(9.5.1) Definition. A (not necessarily finite) separable extension K of a global

field k is called powerful if it has the following property:
Let E be a finite group and let K 0 |K be a finite separable extension of K.
Then there exist primes P1 , . . . , Ps of K which decompose completely in K 0
and group homomorphisms
ϕi : G(K Pi |KPi ) −→ E , i = 1, . . . , s ,
such that the images of ϕi generate E.
Remark: It is easy to find infinite extensions of a global field k which are not
powerful. An obvious example is a Sylow-field K of k, i.e. GK is a p-Sylow
subgroup of Gk for some prime number p.
(9.5.2) Proposition. Let k be a global field and K|k a (not necessarily finite)
abelian extension. Then K is powerful. In particular, each global field is
powerful.
Proof: Let K 0 |K be a finite separable extension, and let ` be a prime number.

We will show that there are infinitely many primes P of K which are completely
decomposed in K 0 |K and having the property that the local Galois groups
G(K P |KP ) have a surjection onto ZZ` . Then the desired result follows.
Let K0 |k be a finite extension inside K|k and K00 |K0 a finite separable
extension such K 0 = KK00 . Let ζ be a primitive `-th root of unity if ` is
different to the characteristic of k and equal to 1 otherwise.
By Čebotarev’s density theorem (9.1.3), there are infinitely many nonar-
chimedean primes of K0 which are completely decomposed in K00 (ζ)|K0 .
Considering their prolongations to K, we get infinitely many nonarchimedean
primes of K which are completely decomposed in K 0 |K, and the completions
of K with respect to these primes contain ζ.
Let P be one them, and let p = P ∩ k be the underlying prime of k and p the
characteristic of the residue field of kp . We identify the decomposition group
GP (K) with the local group GKP = G(K P |KP ), see (8.1.5). If the maximal
unramified `-extension of kp inside KP is finite, then we have a surjection
GKP (`) GKP (`)/TKP (`) → ∼ Z Z` .

Assume now that the maximal unramified `-extension inside KP |kp is

infinite, thus KP realizes the maximal unramified `-extension k̃p (`) of kp .
First we consider the case ` =/ p (and so ` =/ char k). As ζ ∈ KP , we
have TKP (`) → ∼ G
KP (`). Since KP |kp is abelian, it follows from (7.5.9)
that T (KP |kp )(`) is finite, and since Tkp (`) ∼ = ZZ` , see (7.5.2), we have
∼
TKP (`) = ZZ` . Thus we get the desired surjection.
Now let ` = p and let K̃ be the maximal p-extension of k inside K|k. Since
K̃P |kp is an abelian infinite p-extension, the group GK̃P (p) is a nontrivial free
pro-p-group (in fact of infinite rank), see (7.5.10) for p = char k, and (7.1.8)
and (7.5.11) for p-adic fields. Therefore the same holds for the group GKP (p),
and we get a surjection of GKP onto ZZp . 2
(9.5.3) Theorem (IWASAWA). Let K be an infinite powerful extension of a

global field with cd GK ≤ 1 and let K̃ be the maximal solvable extension of K.
Then
G(K̃|K) ∼ = Fω (solv),
where (solv) denotes the class of all solvable finite groups.
Proof: Since the set of finite separable extensions of K (inside a fixed

separable closure) is countable, the group G = G(K̃|K) has rank at most ℵ0 .
We will show that every embedding problem for G with finite solvable E has
a proper solution. Then, using (3.5.20), the result follows. So let
GK
can
G
ϕ
1 A E G 1
be an embedding problem for G with E finite and solvable. In order to find a
proper solution, we may proceed inductively and assume that A is abelian and
even more that A is a simple G-module. Since H 2 (GK , A) = 0 by assumption,
it follows from (3.5.9) that we have a solution ψ: GK → E for the embedding
problem for GK . Let N = ker(ϕ ◦ can). The image of ψ0 = ψ|N in A is a
G-submodule of A, hence equal to A or trivial. In the first case we found a
proper solution, which factors through G because E is solvable, and we are
done. So let us assume that im ψ0 = 0 (so that the group extension splits).
Let G = G(K 0 |K), K 0 ⊆ K̃, then A is a trivial G(K̃|K 0 )-module. Since
G(K̃|K) = lim←− k
G(K̃|k) where k runs through all finite subextensions of K,
we can find a global field k0 ⊆ K and a Galois extension k00 |k0 such that

K 0 = Kk00 and G(K 0 |K) ∼

= G(k00 |k0 ). Obviously this holds for all intermediate
fields k0 ⊆ k ⊆ K:
¡¢£¤¥¦§¨ K̃
K0
K
k0 kk00
k
k00
k0
Via the canonical homomorphism Gk G(k 0 |k) ∼= G(K 0 |K), we consider A

as a Gk -module. Since K is powerful, there exist primes P1 , . . . , Ps of K
such that
(i) P1 , . . . , Ps split completely in K 0 ,
(ii) there are group homomorphisms ϕi : G(K Pi |KPi )) −→ A such that the
images ϕi (G(K Pi |KPi )) generate A.
Enlarging k0 , we may assume that the underlying primes pi = Pi ∩ k0 , i =
1, . . . , s, of k0 split completely in k00 . Let p be the prime number with pA = 0.
If p =/ char(k), (9.2.3)(v) (recalling that G(k 0 |k) is solvable) implies that for all
k between k0 and K the map
s
© A)
Y
H 1 (k, H 1 (kPi , A)
i=1
is surjective. If p = char(k), then the same assertion follows from (9.2.5).

Passing to the direct limit, we see that
s
ª A)
Y
H 1 (K, H 1 (KPi , A)
i=1
is surjective. Therefore we obtain a class [x] ∈ H 1 (K, A) mapping to

res i [x] = ϕi ∈ Hom(G(K Pi |KPi ), A) = H 1 (KPi , A); observe that A is
0
a trivial GKPi -module since G(KP i
|KPi ) = 1 by condition (i). Now let
0 x
ψ = ψ Hom(GK , E) be the solution of the given embedding problem for
∈
GK obtained by multiplying the old solution ψ by the 1-cocycle x. Then
ψ00 = ψ 0 |N = x · ψ0 = x: N → A
is surjective, since G(K Pi |KPi ) ⊆ N and the images of ψ 0 |G(K Pi |KPi ) = ϕi
generate A by condition (ii). Since E is solvable, ψ 0 factors through G. This
finishes the proof of the theorem. 2

(9.5.4) Corollary. Let K be an infinite separable extension of a global field.

Then the Galois group of the maximal solvable extension K̃ over K is a free
prosolvable group of countable rank,
G(K̃|K) ∼ = Fω (solv) ,
(i) K is a ẐZ-extension ∗) of a global field k such that p∞ | [Kp : kp ] for all
prime numbers p and all nonarchimedean primes p of k and k is totally
imaginary in the number field case.
(ii) K is the Kronecker field of a global field k, i.e. K = k(µ) where µ is the
group of all roots of unity in k̄.
(iii) K = k ab is the maximal abelian extension of a global field k.
Proof: By (8.1.18) and (9.5.2), we know in all these cases that cd GK ≤ 1

and that the field K is powerful. Thus (9.5.3) implies the corollary. 2
In §2 we dealt with the problem of whether a given finite family of abelian

local Galois extensions of a global field can be simultaneously realized by an
abelian global extension. By the theorem of Grunwald-Wang, this is always
possible, unless we are in a very special situation.
As a next step of investigation it would be natural to ask whether a similar
statement is true for a given finite family of non-abelian local extensions. Since
local Galois groups are automatically solvable, it is natural to ask whether there
exists a global solvable extension which simultaneously realizes the given local
extension.
As the reader may have expected, this question is extremely difficult to
answer. It is, however, surprising that even the much weaker question of
whether we can find any global extension realizing a given solvable group as
a Galois group (i.e. without any local conditions at all) is highly nontrivial. A
positive answer to this problem has been given by I. R. ŠAFAREVIČ.
Theorem. Let k be a global field and let G be a finite solvable group. Then
there exists a Galois extension K|k with G(K|k) ∼
= G.
We will prove this theorem in §6. Let us first make some general comments.
By definition, a solvable group is built up by successive extensions of abelian
groups. Constructing the required global extension recursively by abelian
steps, the first step is given by the theorem of Grunwald-Wang. In the second,
and every subsequent step, we have to solve embedding problems with abelian
∗) i.e. a Galois extension with Galois group isomorphic to ẐZ.

kernel. These are not always solvable; in fact we can reach a deadlock very
soon as the following example might indicate (for a proof see [216], th. 1.2.4).
Suppose that√ k is a field of characteristic not equal to 2. Then the quadratic
extension k( a)|k can be embedded into a cyclic extension of degree 4 if and
only if a is a sum of two squares in k.
We learn from the above example that although we did not impose local
conditions, there might be a global obstruction to the existence of a solution
of our embedding problem. Therefore it is not very promising to solve the
embedding problems of every induction step separately.
The case when all groups occurring are of order prime to #µ(k), the order
of the group of roots of unity contained in the global field k, can be tackled
by a method which had been developed in 1936 independently by A. SCHOLZ
[206] and H. REICHARDT [177]. Their idea was to choose the solutions of
the inductively given embedding problems in a very special way, in order to
avoid deadlocks. Scholz and Reichardt used this method in order to show that
every finite nilpotent group of odd order occurs as a Galois group over Q. The
reader can find a proof of this statement in the spirit of Scholz and Reichardt
in Serre’s book [216].
The most far reaching result exploiting the Scholz-Reichardt method, is the
following theorem of Neukirch [158], which we are going to explain next.
f
Let Γ be a fixed profinite group. The (continuous) homomorphisms G → Γ
of arbitrary profinite groups into Γ are the objects of a category if one de-
f f0 ψ
fines as morphisms from G → Γ to G0 → Γ all homomorphisms G → G0 with
f 0 ◦ ψ = f . We call two such morphisms ψ and ψ 0 equivalent if there exists an
element a ∈ ker f 0 such that
ψ 0 (σ) = a−1 ψ(σ)a for all σ ∈ G,
and we denote the set of all equivalence classes [ψ] by HomΓ (G, G0 ) and the
subset of all [ψ] with surjective ψ: G G0 by
HomΓ (G, G0 )epi .
Let ϕ
Gk −→ Γ
be a homomorphism of the absolute Galois group of a global field k to the
profinite group Γ and let
ϕp
Gkp −→ Γ

be its restriction to the decomposition group Gkp of Gk with respect to a

prime p. Then if f : G → Γ is a homomorphism of an arbitrary profinite group
G into Γ with kernel H, we obtain diagrams
¯®«¬ Gk
ψ ϕ
f
1 H G Γ
and
´³°±² Gkp
ψp
ϕp
f
1 H G Γ.
Furthermore, we have a canonical restriction map
HomΓ (Gk , G) −→ HomΓ (Gkp , G) .

Y
The following theorem was proved by J. NEUKIRCH in [158] (for number

fields).
(9.5.5) Theorem. Let k be a global field and let ϕ : Gk Γ be a surjective

homomorphism onto the finite group Γ , i.e. Γ = G(K|k) for a finite Galois
extension K of k.
Let f : G Γ be a surjective homomorphism whose kernel is separable∗) ,
prosolvable and of finite exponent which is prime to the order of the group
µ(K) of roots of unity of K. If
HomΓ (Gkp , G) =/ ∅ ,
Y
then the map

HomΓ (Gk , G)epi −→ HomΓ (Gkp , G)
Y
p∈S
is surjective for every finite set S of primes of k.
We shall deduce (9.5.5) from the special case, where the kernel of the homo-
morphism f : G Γ is a finite abelian group A, i.e. we have an embedding
∗) A profinite group is separable, if it has a countable basis of neighbourhoods of the identity.

problem E (Gk , ϕ, f )
¹¶·¸µ Gk
ϕ
f
1 A G Γ 1
with finite abelian kernel A on which Γ acts by conjugation and A becomes a
Gk -module via ϕ. The space SE (Gk ) of solutions of this embedding problem
is nothing else but the set HomΓ (Gk , G). Therefore HomΓ (Gk , G) is either
empty or a principal homogeneous space over H 1 (Gk , A) by (3.5.11).
For a prime p we obtain the associated diagram
¾»¼½º Gkp
ϕp
f
1 A G Γ 1.
Let Γp ⊆ Γ be the image of ϕp and we set Gp = f −1 (Γp ). Then
HomΓ (Gkp , G) = HomΓp (Gkp , Gp )
and the latter group is the set of solutions SE (Gkp ) of the local embedding
problem E (Gkp , ϕp : Gkp Γp , fp : Gp Γp ). Therefore HomΓ (Gkp , G) is
either empty or a principal homogeneous space over H 1 (Gkp , A).
Thus the canonical map
HomΓ (Gk , G) −→ HomΓ (Gkp , G)
Y
has to be understood as the map which assigns to a solution of the global em-
bedding problem E (Gk , ϕ, f ) the associated solutions of the local embedding
problems E (Gkp , ϕp , fp ) for all p.
An element [ψ] ∈ HomΓ (Gkp , G) is called unramified if the inertia group
Tkp of Gkp lies in the kernel of ψ. As before, we set Tkp = Gkp if p is an infinite
prime. By HomΓ (Gkp , G)nr we denote the subset of unramified elements in
HomΓ (Gkp , G). If A is an unramified Gkp -module, then HomΓ (Gkp , G)nr is
1
either empty or a principal homogenous space over Hnr (Gkp , A). Observe that
∅ =/ HomΓ (Gkp , G)nr ⊆ HomΓ (Gkp , G),

ϕp
if Gkp → Γ is unramified. Indeed, since ϕp factors through Gkp /Tkp → Γ , it
can be lifted to a homomorphisms Gkp Gkp /Tkp → G as Gkp /Tkp ∼ = ẐZ is
free.
Next we establish a local-global principle.

(9.5.6) Lemma. If the kernel A of the homomorphism f : G Γ = G(K|k)

is a simple Γ -module such that pA = 0 and p - #µ(K), then
HomΓ (Gk , G) =/ ∅ HomΓ (Gkp , G) =/ ∅ .
Y
⇐⇒
p
Proof: Let us consider the commutative diagram

H 2 (GÁ¿À k , A)
M
H 2 (Gkp , A)
p
ϕ∗
(ϕ∗p )p
2
H (Γ, A).
If p =/ char(k), then the horizontal map is injective by (9.1.16)(iii) applied with
S being the set of all primes and Ω = K(µp ). If p = char(k), then the horizontal
map is trivially injective as H 2 (Gk , A) = 0 by (8.3.3). Let x ∈ H 2 (Γ, A) be the
class belonging to the group extension 1 → A → G → Γ → 1. Using (3.5.9),
we obtain the equivalences
HomΓ (Gk , G) =/ ∅ ⇐⇒ ϕ∗ (x) = 0
⇐⇒ ϕ∗p (x) = 0 for all p
HomΓ (Gkp , G) =/ ∅. 2
Y
⇐⇒
p
Besides the group extension

1 −→ A −→ G −→ Γ −→ 1
with the finite group Γ , the simple Γ -module A (pA = 0) and the homomor-
phism
ϕ: Gk Γ = G(K|k)
we now consider a further surjective homomorphism
EÂ
g
G
with a prosolvable kernel of finite exponent e. Let n be a multiple of ep.
(9.5.7) Main Lemma. Let T be an arbitrary finite set of primes of the global
field k and assume that (n, #µ(K)) = 1. If p HomΓ (Gkp , G) =/ ∅, then there
Q
exists an element [ψ] ∈ HomΓ (Gk , G)epi with the following properties:
(i) [ψ] induces given elements [ψp ] ∈ HomΓ (Gkp , G) at the primes p ∈ T .
(ii) If p ∈/ T is unramified in K|k, then HomG (Gkp , E) =/ ∅, i.e. ψp can be
lifted to a G-homomorphism Gkp → E.
(iii) For the field N defined by ψ: Gk G we have (n, #µ(N )) = 1.

Proof: If T 0 ⊇ T is a larger finite set of primes, then the main lemma holds for
T if it holds for T 0 . In order to see this, choose for every p ∈ T 0 r T an element
[ψp ] in the nonempty set HomΓ (Gkp , G). If p ∈ T 0 r T is unramified, then
we choose [ψp ] ∈ HomΓ (Gkp , G)nr =/ ∅. If [ψ] ∈ HomΓ (Gk , G)epi satisfies
conditions (i)–(iii) for T 0 , then so it does for the set T . Indeed, (i) and (iii)
are trivially satisfied and (ii) follows from the fact that HomG (Gkp , E)nr =/ ∅
for unramified primes p ∈ T 0 r T , since we have chosen unramified elements
[ψp ]. Therefore we may assume that
Ram(K|k) ∪ Sn ∪ S∞ ⊆ T,
where Sn is the set of all primes dividing n if k is a number field, and empty
if k is a function field.
We next show that it suffices to prove the existence of an element [ψ] in
HomΓ (Gk , G) with (i)–(iii), i.e. we can drop the surjectivity of [ψ]. Let q ∈/ T
be a prime which splits completely in K|k and let a ∈ A, a =/ 0. Let further
ψq : Gkq Gkq /Tkq hai be an unramified surjective homomorphism
onto the cyclic group hai ⊆ G. Since q splits completely in K, we have
ϕq (Gkq ) = {1}, hence [ψq ] ∈ HomΓ (Gkq , G). Let T 0 = T ∪ {q} and let [ψ]
in HomΓ (Gk , G) satisfying the conditions (i)–(iii) for T 0 . Then it satisfies
conditions (i)–(iii) also for T , as is seen in the same way as above. Moreover,
ψ is surjective. Namely, ψ(Gkq ) ⊆ G is conjugate to ψq (Gkq ) = hai under an
element of A, and so a ∈ ψ(Gkq ) ⊆ ψ(Gk ). Therefore A ∩ ψ(Gk ) is a nontrivial
Γ -submodule of A. Hence A ⊆ ψ(Gk ) since A is simple; and ψ is surjective.
A last reduction step is the following: it suffices to prove the existence of an
element [ψ] ∈ HomΓ (Gk , G) satisfying only conditions (i) and (ii).
Let n0 = n if k is a number field and let n0 be the maximal divisor of n prime
to char(k) if k is a function field. Let ζn0 be a primitive n0 -th root of unity and
let σ1 , . . . , σr be generators of G(K(ζn0 )|K). Let q1 , . . . , qr ∈/ T be different
primes such that qi splits completely in K and
FrobQi = σi , i = 1, . . . , r ,
where Qi is a prime of K above qi . Then ϕqi is the trivial homomor-
phism and the trivial homomorphism ψqi : Gkqi → G yields an element in
HomΓ (Gkqi , G), i = 1, . . . , r. Let T 0 = T ∪ {q1 , . . . , qr } and let [ψ] in
HomΓ (Gk , G) satisfying the conditions (i) and (ii) for T 0 , and so for T .
Moreover it satisfies condition (iii): the restriction of ψ to Gkqi is the trivial
homomorphism, i.e. q1 , . . . , qr split completely in the field N |k, which is de-
fined by ψ: Gk → G. Therefore the images of the automorphisms σi = FrobQi
under the surjection G(K(ζn0 )|K) G(N ∩ K(ζn0 )|K) are trivial. It follows
that N ∩ K(ζn0 ) = K. Let d = (n0 , #µ(N )) and let ζd be a primitive d-th root
of unity. Then ζd ∈ N ∩ K(ζn0 ) = K, i.e. d|(n0 , #µ(K)). Since n/n0 is prime
to #µ(K) if K is a function field, we have (n0 , #µ(K)) = (n, #µ(K)) = 1, and
so d = 1.

We now come to the essential step in the proof of the lemma. By the
assumption p HomΓ (Gkp , G) =/ ∅ and by (9.5.6), there is an element
Q
[ψ0 ] ∈ HomΓ (Gk , G) .

We will change this element by a cohomology class x ∈ H 1 (Gk , A) into an
element ψ which satisfies the conditions of the lemma. Let N0 |k be the
field defined by ψ0 : Gk → G. The extension N0 |K is abelian. Let [ψ1p ] ∈
HomΓ (Gkp , G) be the elements which are given in advance for p ∈ T . Let
p1 , . . . , pr be the primes p ∈/ T for which ψ0p is ramified. Since Ram(K|k) ⊆ T ,
the primes pi are unramified in K|k, i.e. the homomorphisms ϕpi are unramified
and can therefore be lifted to unramified Γ -homomorphisms ψ1pi : Gkpi → G.
Let T ∗ = T ∪ {p1 , . . . , pr } and for each p ∈ T ∗ let yp ∈ H 1 (Gkp , A) be the
cohomology class which sends [ψ0p ] to [ψ1p ], i.e.
[ψ0p ]yp = [ψ1p ] .
We will deal with the case p = char(k) later and assume p =/ char(k) for the
moment. Let n0 be defined as above and let Ω = N0 (ζn0 ), ζn0 a primitive n0 -th
root of unity. Let K 0 be the subfield of Ω which corresponds to the prime-to-p
part G(Ω|K)(p0 ) of the abelian group G(Ω|K). Then K 0 |K is a p-extension,
hence µp ⊆ / K 0 because µp ⊆ / K. Therefore we can apply (9.3.3) with S being
the set of all primes to find an element x ∈ H 1 (Gk , A) such that
(a) xp = yp for p ∈ T ∗ ,
(b) if p ∈/ T ∗ , then xp is cyclic and if xp is ramified, then p splits completely
in Ω.
We now prove that the element
[ψ] = [ψ0 ]x ∈ HomΓ (Gk , G)
satisfies conditions (i) and (ii) of the lemma. If p ∈ T ∗ , then
[ψp ] = [ψ0p ]xp = [ψ0p ]yp = [ψ1p ] ,
thus condition (i) is satisfied. Let p ∈/ T be unramified in K|k. If p ∈ T ∗ r T ,
then [ψp ] = [ψ1p ] is unramified, and so ψp can be lifted to a G-homomorphism
Gkp → E, i.e. HomG (Gkp , E) =/ ∅.
Let p ∈/ T ∗ . If [ψp ] is unramified, then HomG (Gkp , E) =/ ∅ for the same
reason. Let [ψp ] = [ψ0p ]xp be ramified. Since [ψ0p ] is unramified for p ∈/ T ∗ ,
the cohomology class xp must be ramified. Thus the prime p splits completely
in Ω|k by (b), and in particular it splits in N0 |k and in k(ζn0 )|k. Therefore
ψ0p : Gkp → G is the trivial homomorphism, i.e. N0p = kp , and ζn0 ∈ kp .
Because Kp = kp the Galois module A is a trivial Gkp -module, i.e.
xp : Gkp −→ A ⊆ G
is a homomorphism which represents the class [ψp ] = [ψ0p ]xp . It remains to
show that xp can be lifted to a G-homomorphism Gkp → E.

Since xp is cyclic, it defines a cyclic ramified extension N√p |kp of degree p.

Since p - p, this extension is tamely ramified. Thus Np = kp ( p π) with a prime
element π. Let σ̄ be a generator of
G(Np |kp ) = xp (Gkp ) ⊆ G
and let σ be an element of E which is mapped onto σ̄ under g : E G.
As σ̄ has order p, we may choose σ in a p-Sylow subgroup Ep of E. Since
ker(E G) has exponent e = n/p, the order m of σ is a p-power which
divides n0 . Since ζn0√∈ kp , the m-th roots of unity are in kp . Therefore the
extension Mp = kp ( m π)|kp is cyclic of degree m. Mapping a generator τ
of G(Mp |kp ) with image σ̄ in G(Np |kp ) onto σ, we obtain the commutative
diagram
ÈÉÃÄÅÆÇ Gkp
G(Mp |kp ) G(Np |kp ) xp =ψp
g
E G.
Hence ψp can be lifted to a G-homomorphism Gkp → E. This finishes the

proof of the lemma in the case p =/ char(k).
Now let p = char(k). Choose any prime q ∈/ T ∗ of k which is completely
decomposed in K|k. By (9.2.5), we find an element x ∈ H 1 (Gk , A) such that
(a) xp = yp for p ∈ T ∗ ,
(b) x is unramified outside T ∗ ∪ {q}.
We now prove that the element
[ψ] = [ψ0 ]x ∈ HomΓ (Gk , G)
satisfies conditions (i) and (ii) of the lemma. If p ∈ T ∗ , then
[ψp ] = [ψ0p ]xp = [ψ0p ]yp = [ψ1p ] ,
thus condition (i) is satisfied. Let p ∈/ T be unramified in K|k. If p ∈ T ∗ r T ,
then [ψp ] = [ψ1p ] is unramified, and so ψp can be lifted to a G-homomorphism
Gkp → E, i.e. HomG (Gkp , E) =/ ∅.
Let p ∈/ T ∗ . If [ψp ] is unramified, then HomG (Gkp , E) =/ ∅ for the same
reason. Let [ψp ] = [ψ0p ]xp be ramified. Since [ψ0p ] is unramified for p ∈/ T ∗ , the
cohomology class xp must be ramified. Thus p = q. By the choice of q, it splits

completely in Ω|k. Therefore ψ0q : Gkq → G is the trivial homomorphism, i.e.

N0q = kq . Because Kq = kq the Galois module A is a trivial Gkq -module, i.e.
xq : Gkq −→ A ⊆ G
is a homomorphism which represents the class [ψq ] = [ψ0q ]xq . It remains
to show that xq can be lifted to a G-homomorphism Gkq → E. However, xq
factors through the maximal pro-p-factor group Gkq (p), which is a free pro-
p-group by (6.1.4), and therefore a projective profinite group by (3.5.13). We
obtain a commutative diagram
ÏÍÎÊËÌ Gkq
Gkq (p) xq =ψq
g
E G.
Hence also in the case p = char(k) the homomorphism ψq can be lifted to a
G-homomorphism Gkq → E. This finishes the proof of the lemma. 2
Proof of (9.5.5): We consider the embedding problem

ÑÒÓÔÐ Gk
ϕ
f
1 H G Γ 1,
where Γ = G(K|k) and the kernel H is a separable, prosolvable group of finite
exponent n and (n, #µ(K)) = 1. We can find a chain
H = H0 ⊇ H1 ⊇ H2 ⊇ H3 ⊇ ...
of open subgroups of H, normal in G, such that the quotients Hi−1 /Hi are
abelian and i Hi = {1}. Then Hi−1 /Hi is a G/Hi−1 -module and we find a
T
finite chain
Hi−1 ⊇ Hi−1,1 ⊇ Hi−1,2 ⊇ · · · Hi−1,r(i−1) = Hi
of subgroups of Hi−1 containing Hi such that the quotients Hi−1,j−1 /Hi−1,j
are simple G/Hi−1 -modules. After renumbering, we therefore can find a chain
H = H0 ⊇ H1 ⊇ H2 ⊇ H3 ⊇ . . . of open subgroups of H, normal in G, such
that i Hi = {1} and the quotients Ai = Hi−1 /Hi are simple G/Hi−1 -modules.
T
For every p ∈ T we prescribe an element [ψp ] ∈ HomΓ (Gkp , G) and select

a Γ -homomorphism ψp : Gkp → G in [ψp ]. We will construct an element
[ψ] ∈ HomΓ (Gk , G)epi such that [ψ|Gkp ] = [ψp ] for p ∈ T . Let Gi = G/Hi
and consider the commutative diagrams with exact rows

æçèéêëäåÞßàáâãÜÝÚÛÙØ×ÖÕ Gk Gkp
ϕ ψp
ϕp
f
H G Γ G Γ
ψpi
fi
H/Hi Gi Γ Gi Γ
πi πi
fi−1
H/Hi−1 Gi−1 Γ Gi−1 Γ, p ∈ T.
Let pi Ai = 0 and ni = exp(Hi ). We denote the composite of ψp and G Gi

by ψpi .
Claim. There exists a sequence of surjective Γ -homomorphisms ψi : Gk Gi
with the following properties:
1) ψi−1 = πi ◦ ψi , i = 1, 2, . . . .
2) [ψi |Gkp ] = [ψpi ] in HomΓ (Gkp , Gi ) for p ∈ T .
3) If the field Ki is defined by ψi , then we have (n, #µ(Ki )) = 1.
HomGi (Gkp , G) =/ ∅.
Y
4)
p
Proof of the claim: If i = 0, then we can identify G0 with Γ and choose ψ0 = ϕ.

The first condition is empty, the second is true since ϕ|Gkp = ϕp , and the third
and fourth are assumptions of the theorem.
Let us assume that the homomorphism ψi−1 : Gk Gi−1 is already con-
structed. Let Ti−1 = T ∪ Ram(Ki−1 |k), where the field Ki−1 is defined by ψi−1 .
For each p ∈ Ti−1 r T we can choose a Gi−1 -homomorphism ψp : Gkp → G
since the set HomGi−1 (Gkp , G) is not empty by condition 4. We consider the
diagrams
íîìðï Gk Gkp
ψpi
ψi−1 ψi−1 |Gk
p
πi
Gi Gi−1 Gi Gi−1 , p ∈ Ti−1 .
Since [ψpi ] ∈ HomGi−1 (Gkp , Gi ), the right diagrams are commutative. The
kernel of πi is the simple Gi−1 -module Ai . Its exponent pi divides n. Since

the exponent ni of Hi divides n, it follows that npi is a multiple of ni pi

and (npi , #µ(Ki−1 )) = (n, #µ(Ki−1 )) = 1 by condition 3. Furthermore, by
condition 4 we have
HomGi−1 (Gkp , Gi ) =/ ∅ ,
Y
and so we can apply (9.5.7) to the situation replacing T by Ti−1 and E G

by G Gi and ϕ by ψi−1 . We find an element [ψi ] in HomGi−1 (Gk , Gi )epi
with the properties:
(i) [ψi |Gkp ] = [ψpi ] in HomGi−1 (Gkp , Gi ) for p ∈ Ti−1 .

(ii) If p ∈/ Ti−1 , then HomGi (Gkp , G) =/ ∅.
(iii) For the field Ki |k defined by ψi we have (n, #µ(Ki )) = 1.
Let ψi : Gk Gi be a representative of [ψi ]. Then ψi is a surjective Γ -
homomorphism, and we show that it satisfies conditions 1 − 4 of the claim.
Condition 1 is satisfied because of [ψi ] ∈ HomGi−1 (Gk , Gi ). By (i) the
Γ -homomorphisms ψi |Gkp and ψpi , p ∈ Ti−1 , are conjugate by an element of
Ai = Hi−1 /Hi ⊆ H/Hi = ker(Gi Γ ).
Therefore [ψi |Gkp ] = [ψpi ] in HomΓ (Gkp , Gi ), showing condition 2. Condi-

tion 3 is satisfied by (iii) and, finally, condition 4 follows from (ii) and from
[ψp ] ∈ HomGi (Gkp , G) for p ∈ Ti−1 . This finishes the proof of the claim.
Gi , since i Hi = {1}. Therefore the surjective Γ -
T
We have G = lim ←−
homomorphisms ψi : Gk Gi define a surjective Γ -homomorphism
ψ: Gk G, i.e. an element in [ψ] ∈ HomΓ (Gk , G)epi . For each p ∈ T
we have [ψi |Gkp ] = [ψpi ] in HomΓ (Gkp , Gi ), and so there exist elements
ãpi ∈ ker(Gi Γ ) = H/Hi such that
ψi |Gkp = ã−1
pi ψpi ãpi , i = 1, 2, . . . .
Using a compactness argument, we find an element ap ∈ H such that
ψi |Gkp = a−1
pi ψpi api , i = 1, 2, . . . ,
where the elements api are the images of ap in H/Hi . We obtain
ψ|Gkp = a−1
p ψp ap ,
i.e. [ψ|Gkp ] = [ψp ] in HomΓ (Gkp , G). This finishes the proof of (9.5.5). 2
In the language of embedding problems, (9.5.5) implies the following local-

global principle.

(9.5.8) Corollary. Let E (Gk , ϕ, f ) be an embedding problem of the global

field k,
òóôñõ Gk
ϕ
f
1 H G Γ 1,
and assume that Γ = G(K|k) is finite and H is a separable, prosolvable group
of finite exponent prime to #µ(K). Then the following holds.
(i) E (Gk , ϕ, f ) has a proper solution if and only if all local embedding
problems E (Gkp , ϕp , fp ) are solvable.
(ii) E (Gk , ϕ, f ) has a proper solution in the following cases:
(a) K|k is unramified.
(b) E (Gk , ϕ, f ) has a solution.
(c) The group extension splits.
Remark: Assertion (ii)(b) was proven by Ikeda [83] in the case that k is a
number field and H a finite abelian group.
Proof: The solvability of all local embedding problems is equivalent with

p HomΓ (Gkp , G) =
Q
/ ∅. Thus (i) follows from (9.5.5).
Since the local embedding problem E (Gkp , ϕp , fp ) is always solvable if the
extension Kp |kp is unramified, we obtain (ii)(a).
The assertion HomΓ (Gk , G) =/ ∅ implies p HomΓ (Gkp , G) =/ ∅ and
Q
so HomΓ (Gk , G)epi =/ ∅. This proves (ii)(b) and assertion (c) is a trivial
consequence of (b). 2
Furthermore, we obtain the following existence theorem of Grunwald/Wang

type.
(9.5.9) Theorem. Let k be a global field and let T be a finite set of primes of k.
Let G be a separable, prosolvable group of finite exponent prime to #µ(k), and
for p ∈ T let Kp |kp be Galois extensions whose Galois groups G(Kp |kp ) are
embedable into G.
Then there exists a Galois extension K|k with Galois group isomorphic to G,
which has the given extensions Kp |kp as completions for the primes p ∈ T .

Proof: Apply (9.5.5) to the case Γ = {1} and the homomorphisms
ψp : Gkp G(Kp |kp ) ,→ G, p ∈ T. 2
Finally, we obtain existence results without local conditions.
(9.5.10) Corollary. Let k be a number field and let G be a separable, prosolv-

able group of finite odd exponent.
Then there exists a Galois extension K|k with Galois group isomorphic to
the group G.
Remarks: 1. As already mentioned, the theorem of Šafarevič, which we

prove in the next section, shows that the assertions above holds for every finite
solvable group.
2. The finiteness of the exponent of G is in fact a necessary condition. For
example, the group ZZp × ZZp is not the Galois group of an extension of Q.
Proof: We apply (9.5.9) to the field Q, the set T = Ram(k|Q) and the trivial
extensions Kv = Qv for v ∈ T . Then trivially the groups G(Kv |Qv ) are
embedable into G and the exponent of G is prime to #µ(Q) = 2. Therefore
there exists a surjective homomorphism ψ: GQ G such that the restrictions
ψv : GQv → G are trivial for v ∈ T . Let K 0 be the fixed field of ker(ψ). Since
ψ(GQv ) = 1 for all v ∈ T , the primes in T split completely in K 0 , and so
K 0 ∩ k|Q is an unramified extension. It follows that K 0 ∩ k = Q. Setting
K = K 0 k, we obtain G(K|k) ∼ = G(K 0 |Q) ∼
= G. 2
The function field analogue of (9.5.10) is the following
(9.5.11) Corollary. Let k be a global field of characteristic p > 0 and let G be

a separable, prosolvable group of finite exponent e which is prime to p − 1.
Then there exists a Galois extension K|k with Galois group isomorphic to
the group G.
Remark: If char(k) = 2, then the corollary is sharper than the statement of

Šafarevič’s theorem.

Proof: We choose an embedding IFp (t) ⊆ k, such that k|IFp (t) is finite and
separable. We apply (9.5.9) to IFp (t), T = ∅ and the n-fold self-product Gn
of G, where n is some natural number. Since e is prime to #µ(IFp (t)) = p − 1,
we obtain a Galois extension K|IFp (t) with Galois group Gn . The projections
pri : Gn G, i = 1, . . . , n, yield n linearly disjoint extensions Ki |IFp (t) with
Galois group G. If n was chosen sufficiently large, at least one of the fields Ki
is linearly disjoint to k, and Ki k|k is an extension with Galois group G. 2
Finally, we would like to mention a result of G. HENNIART in [78], where

only one given local Galois extension at a nonarchimedean prime is considered,
but no restriction on the order of the Galois group is made:
Let k be a totally real number field and p a prime of k dividing the odd prime
number p. Let Kp |kp be a finite Galois extension. Then there exists a totally
real Galois extension K|k such that [K : k] = [Kp : kp ] and the completions
with respect to the primes P above p are isomorphic to Kp .
§6. Šafarevič’s Theorem
In this section we will prove the celebrated theorem of I. R. ŠAFAREVIČ

concerning the existence of an extension of a global field realizing a given
solvable group as a Galois group.
(9.6.1) Theorem (ŠAFAREVIČ). Let k be a global field and let G be a finite

solvable group. Then there exists a Galois extension K|k with G(K|k) ∼
= G.
Since 1954, when Šafarevič proved this result (see [187], [188], [189], [190]),
several mathematicians have made suggestions how to reprove it using the sub-
sequent developments in number theory and, in particular, the duality theorem
of Poitou-Tate. Unfortunately, however, there is no such proof of theorem
(9.6.1) accessible in the literature. Moreover, the original article [187] con-
tains a mistake relative to the prime 2 ∗) . We have therefore decided to include
a complete proof of Šafarevič’s theorem in this section. Of course, we use the
original ideas of Šafarevič, in particular, the remarkable technique of shrinking
∗) Šafarevič explains how to correct this in [192].

§6. Šafarevič’s Theorem 575
obstructions, which is highly instructive. The authors do not know any other
argument in number theory which utilizes a similar technique.
As already explained in §5 the idea of Scholz and Reichardt was to choose
the solutions of the inductively given embedding problems in a very special
way, in order to avoid deadlocks when constructing the required global solvable
extension inductively by abelian steps. This works if the order of G is prime
to the order of µ(k). Šafarevič uses the same special kind of solutions of the
inductively given embedding problems (“Scholz solutions”), but in the general
situation one can run into a deadlock. The key idea of Šafarevič’s approach is
to modify the solutions of the first i − 1 induction steps already found, in order
to leave a deadlock within the i-th step. This happens in a rather complicated
way within a shrinking procedure. This method works without any restriction
on the group, but unfortunately it seems to be impossible to realize given local
extensions. Therefore with Šafarevič’s approach, one only can say that every
solvable group occurs as a Galois group over k such that the associated local
extensions are of a particular type.
We now explain Šafarevič’s method in detail, and give a proof of theorem

(9.6.1). To begin with, let us explain the main ideas. In the first reduction step,
one shows that the result follows from the assertion that every split embedding
problem of finite groups with nilpotent kernel N
ö÷øùúû Gk
1 N E G 1
has a proper solution. In order to solve these split embedding problems, we
may assume that N is a p-group and it suffices to consider the generic kernel,
i.e. N = F(n)/F(n)(ν) , where F(n) is a free pro-p-G operator group of rank n
and F(n)(ν) denotes the ν-th term of a filtration of F(n), which we defined in
III §8 and which refines the descending p-central series.∗)
We proceed by induction on ν whereas n is arbitrary. The case ν = 1 is
trivial. The problems which have to be solved within every induction step are
of the form
üýþÿ Gk
1 F(n)(ν)/F(n)(ν+1) F(n)/F(n)(ν+1) o G F(n)/F(n)(ν) o G 1.

∗) This refinement, which was proposed by Šafarevič in his correction note, is necessary in
order to deal with the case p = 2.

This induction step is proved in four substeps. In the first step, one shows
that this problem is locally solvable everywhere, i.e. for every prime p of k the
induced local problem
Gkp
1 F(n)(ν) /F(n)(ν+1) Ep (F(n)/F(n)(ν) o G)p 1
has a solution. This can be done if the old solution Nνn |k with
G(Nνn |k) = F(n)/F(n)(ν) o G
is locally of certain type, namely a so-called “Scholz solution”.
In the second step one uses the local-global principle in order to show that a
global solution exists.
In step three and step four we modify the global solution in order to get a
proper “Scholz solution”, in such a way that the new local problems for the
next induction step ν + 1 will be solvable.
Within the induction steps, obstructions to the existence of solutions of the
given embedding problems occur. These obstructions really exist and are
nontrivial. Šafarevič’s idea how to overcome this problem is the following:
We revise the solution found in the (ν − 1)-th induction step. Instead we use
the (ν − 1)-th induction step, not for n but for a very large m > n. The solution
of that problem (which exists by the induction hypothesis) induces a new
solution for our original problem via any surjective G-invariant homomorphism
ψ : F(m) F(n).
If m is sufficiently large, the shrinking process, which we will explain in a
moment, gives a homomorphism ψ in such a way that all obstructions for the
embedding problem in the ν-th induction step for F(n) vanish. In this way one
proves the induction step from ν − 1 to ν and for every n. Having the result
for all n, we can use the shrinking procedure in the next induction step again.
The next proposition is the technical skeleton of the shrinking process. A

similar statement was already contained in the original paper [190].
(9.6.2) Proposition. Let G be a finite group. Suppose that M and N are finitely
generated IFp [G]-modules and let s, t ∈ IN. Then for r ∈ IN sufficiently large∗)
the following holds:
∗) i.e. for all r ≥ r0

Given elements
M )⊗s ⊗ N ,
M
z1 , . . . , zt ∈ (
r
there exists a = (ai )i=1,...,r ∈ IFrp such that
r
ϕa :
M X
M M, (xi )i=1,...,r 7−→ ai x i ,
r i=1
is a surjective IFp [G]-homomorphism (i.e. not all ai are zero) and such that the
induced homomorphism
ψa = (ϕ⊗s M )⊗s ⊗ N −→ M ⊗s ⊗ N
M
a ) ⊗ id : (
r
maps all zi , i = 1, . . . , t, to zero.
Proof: ∗) Let r be arbitrary. Then ϕa is IFp [G]-invariant and surjective if

a =/ 0. Let n = t · dimIFp (M ⊗s ⊗ N ) and suppose that r > s n. Consider the set
V = {a ∈ IFrp | ψa (z1 ) = . . . = ψa (zt ) = 0} .
Then V is the set of common zeros of n polynomials of degree s. It contains
the trivial element and by the theorem of Chevalley-Warning (see (6.5.7)), it
follows that it must also contain a nontrivial point a. Then ϕa has all desired
properties. 2
Let us recall some filtrations that we defined in III §8. First, we have the
descending p-central series {P i }i≥1 of a pro-p-group P which is given by
p
P 1 = P, P i+1 = (P i ) [P i , P ], i ≥ 1.
In addition, we use the descending central series {Pi }i≥1 of P which is defined
by
P1 = P, Pi+1 = [Pi , P ], i ≥ 1.
Setting
P (i,j) := (P i ∩ Pj )P i+1 ,
we have
P (i,1) = P i and P (i,j) = P i+1 for j > i ≥ 1.
Recall the notational convention:
The letter ν always stands for a pair (i, j), i ≥ j ≥ 1, and we order these pairs
lexicographically. We say that ν + 1 = (i, j + 1) if i > j and ν + 1 = (i + 1, 1)
if ν = (i, i).
∗) This proof is based on an idea of J. SONN. We also want to thank A. DEITMAR for a
further simplification.

The descending chain {P (ν) } of normal characteristic subgroups is a refinement

of the descending p-central series. In particular, P (ν) /P (ν+1) is an IFp -vector
space for all ν, and by (3.8.8) we have surjective IFp -vector space homomor-
phisms ∗)
ψ(i,j) : (P/P 2 )⊗j −→ P (i,j) /P ((i,j)+1) ,

i−j
x̄1 ⊗ · · · ⊗ x̄j 7−→ ([x1 , [x2 , [···[xj−1 , xj ]]···])p mod P ((i,j)+1) .
Recall the definition of a free pro-p-G operator group: if G is a finite group

and Fd a free pro-p-group of rank d, then we set
F(d) = ∗F
G
d ,
a free pro-p-G operator group of rank d. The group F(d) is a free object in the
category of pro-p-groups which are endowed with a continuous action of the
group G. As a pro-p-group, F(d) is free of rank #G · d and one can choose a
basis {xi,g | i = 1, . . . , d, g ∈ G} of F(d) such that G acts by
g 0 xi,g = xi,gg0 , g 0 ∈ G,
compare IV §3, example 3.
Now we will apply the shrinking process to cohomology groups with respect
to G and F(d) in order to annihilate given cohomology classes.
(9.6.3) Proposition. Let G be a finite group and let F(d) be the free pro-p-G
operator group of rank d. Let n, t ∈ IN, k ∈ ZZ, a finitely generated IFp [G]-
module T and ν = (i, j) be given. Then for m ≥ n large enough∗∗) the
following holds:
Given elements
x1 , . . . , x t ∈ Ĥ k (G, F(m)(ν) /F(m)(ν+1) ⊗ T ) ,
there exists a surjective pro-p-G operator homomorphism

ψ : F(m) F(n)
such that the induced homomorphism
ψ∗ : Ĥ k (G, F(m)(ν) /F(m)(ν+1) ⊗ T ) −→ Ĥ k (G, F(n)(ν) /F(n)(ν+1) ⊗ T )
maps x1 , . . . , xt to zero.
∗) ψ can belifted to a map (P/P 2 )⊗j → P (i,j) /P i+1 ⊆ P i /P i+1 if either j > 1 or if p is
(i,j)
odd (hence p | p2 ). This is the reason why we could also work with the descending p-central
series if p is odd and why this is not possible in the case p = 2.
∗∗) i.e. for all m ≥ m ≥ n
0

Proof: We set E(n, ν) = F(n)(ν) /F(n)(ν+1) . Using dimension shifting (cf.

I §3) we have an isomorphism
∼ Ĥ −1 (G, E(n, ν) ⊗ T ⊗ A ) ,
Ĥ k (G, E(n, ν) ⊗ T ) −→ k
where Ak = JG⊗(k+1) for k > −1, A−1 = IFp and Ak = IG⊗−(k+1) for k < −1, and
IG and JG are given by the exact sequences
0 −→ IG −→ IFp [G] −→ IFp −→ 0,
0 −→ IFp −→ IFp [G] −→ JG −→ 0.
Since T was arbitrary, we may restrict to the case k = −1.

Observe that for every d ≥ 1 the canonical surjective IFp -homomorphism
defined in (3.8.8)

θν (d) : (F(d)/F(d)2 ⊗j
) F(d)(ν) /F(d)(ν+1)
is obviously G-invariant, and given a pro-p-G operator homomorphism ψ :
F(d0 ) → F(d) we have the compatibility θν (d0 )ψ∗ = ψ∗ θν (d).
Now let m = rn, r sufficiently large, so that we can apply (9.6.2) with the
G-module T . Given elements
z1 , . . . , zt ∈ (F(m)/F(m)2 )⊗j ⊗ T = (IFp [G]rn )⊗j ⊗ T,
there exists a surjective IFp [G]-homomorphism

ψ̄ : F(m)/F(m)2
F(n)/F(n)2 ,
such that (ψ̄ ⊗j ⊗ id)(zα ) = 0, α = 1, . . . , t. By the universal property of free
pro-p-G operator groups, ψ̄ extends to a pro-p-G operator homomorphism

ψ : F(m) F(n),
which is necessarily surjective (by the Frattini argument).
Now we consider the commutative diagram
2 )⊗j ⊗ T
(F(m)/F(m) (E(m, ν) ⊗ T )G Ĥ −1 (G, E(m, ν) ⊗ T )
ψ̄ ⊗j ⊗id ψ∗ ⊗id ψ∗
(F(n)/F(n)2 )⊗j ⊗ T (E(n, ν) ⊗ T )G Ĥ −1 (G, E(n, ν) ⊗ T )

and choose zα as a pre-image of the image of xα in the group (E(m, ν) ⊗ T )G .
Choosing an appropriate ψ, the diagram shows that ψ∗ (xα ) = 0 for α = 1, . . . , t.
We will apply (9.6.3) only for k = 2 and k = −2, and in the latter dimension
we also need the following variant, which goes back to an idea of V. V. IŠHANOV.

(9.6.4) Proposition. Let G be a finite group and let F(d) be the free pro-p-G
operator group of rank d. Let n, t ∈ IN, a finitely generated IFp [G]-module T
and ν = (i, j) be given. Then for sufficiently large m ≥ n the following hold:
(i) Given elements
x1 , . . . , x t ∈ H −2 (F(m)/F(m)(ν) o G, F(m)(ν) /F(m)(ν+1) ⊗ T ) ,

ψ : F(m) F(n) ,
ψ∗ : H −2 (F(m)/F(m)(ν) o G, F(m)(ν) /F(m)(ν+1) ⊗ T ) −→
H −2 (F(n)/F(n)(ν) o G, F(n)(ν) /F(n)(ν+1) ⊗ T )
(ii) Given elements
x1 , . . . , x t ∈ H −2 (F(m)/F(m)(ν) , F(m)(ν) /F(m)(ν+1) ⊗ T ) ,

ψ : F(m) F(n)
ψ∗ : H −2 (F(m)/F(m)(ν) , F(m)(ν) /F(m)(ν+1) ⊗ T ) −→
H −2 (F(n)/F(n)(ν) , F(n)(ν) /F(n)(ν+1) ⊗ T )
Proof: We keep the notation

F(n)/ν = F(n)/F(n)(ν) and E(n, ν) = F(n)(ν) /F(n)(ν+1) .
If ν = 1, then the statement to prove is just a special case of (9.6.3). So we
may assume that ν = (i, j) ≥ (2, 1). Recall that H −2 = H1 and consider the
exact sequence
H1 (F(n)/ν, E(n, ν) ⊗ T ) −→ H1 (F(n)/ν o G, E(n, ν) ⊗ T ) −→
H1 (G, E(n, ν) ⊗ T ) −→ 0,
which is induced by the homological Hochschild-Serre sequence. Since E(n, ν)
and T are trivial F(n)/ν-module, we obtain
H1 (F(n)/ν, E(n, ν) ⊗ T ) ∼
= F(n)/F(n)2 ⊗ E(n, ν) ⊗ T

and (3.8.8) implies the existence of a G-invariant surjection

(F(n)/F(n)2 )⊗(j+1) ⊗ T H1 (F(n)/ν, E(n, ν) ⊗ T ),
where ν = (i, j). This is obviously true for arbitrary n, and the maps
are compatible. Let r ≥ n have the property that t arbitrary elements in
(F(r)/F(r)2 )⊗(j+1) ⊗ T are annihilated by the homomorphism induced by a
suitably chosen G-invariant surjection F(r) F(n) (and which exists by
(9.6.2)). The above surjection shows that r has the same property with respect
to t given arbitrary elements in
H1 (F(r)/ν, E(r, ν) ⊗ T ).
This proves (ii). In order to show (i), let m ≥ r have the property that t arbitrary
elements in
H1 (G, E(m, ν) ⊗ T )
are annihilated by the homomorphism induced by a suitably chosen G-invariant
surjection F(m) F(r) (and which exists by (9.6.3)). We obtain the com-
mutative exact diagram
H1 (F(m)/ν, E(m, ν)⊗T ) H1 (F(m)/ν o G, E(m, ν)⊗T ) α

H1 (G, E(m, ν)⊗T )
π
β
H1 (F(r)/ν, E(r, ν)⊗T ) H1 (F(r)/ν o G, E(r, ν)⊗T ) H1 (G, E(r, ν)⊗T )
ε
H1 (F(n)/ν, E(n, ν)⊗T ) H1 (F(n)/ν o G, E(n, ν)⊗T ) H1 (G, E(n, ν)⊗T ) ,
in which the vertical maps are induced by G-invariant surjections

F(m) F(r) F(n),
which we choose in the following way. Let arbitrary elements x1 , . . . , xt ∈
H1 (F(m)/ν o G, E(m, ν) ⊗ T ) be given. Choose F(m) F(r) such that
the induced homomorphism annihilates the elements α(x1 ), . . . , α(xt ). Hence
π(x1 ), . . . , π(xt ) are contained in the image of β and we choose the surjec-
tion F(r) F(n) such that ε annihilates arbitrarily chosen β-pre-images of
π(x1 ), . . . , π(xt ). The composite F(m) F(n) has the desired property.
2
For a prime number p =/ char(k) we denote, as before, the set of primes of

k with residue characteristic p by Sp = Sp (k). The set Sp (k) is finite and it
is empty if k is a function field. In the number field case we denote the set
of archimedean places of k by S∞ = S∞ (k). In the function field case we
choose (only in this section) any finite, nonempty set of primes of k and call it
S∞ = S∞ (k). For every extension field K|k, we denote by S∞ (K) the set of
primes of K which lie over S∞ (k).

In the next technical lemma, which will be needed later, we freely use the
notation introduced in §§1,2; in particular, for a Gk -module A the minimal
trivializing extension of k is denoted by k(A).
(9.6.5) Lemma. Let k be a global field, p =/ char(k) a prime number, and

assume that we are given sets of primes of k
S0 ⊇ S ⊇ T ⊇ Sp ∪ S∞ ,
where T is finite. Let A be a finite GS -module which is annihilated by p. In
addition, suppose that we are given a finite subextension N ⊆ kS , with
a) k(A) ⊆ N ,
b) S 0 r T ⊆ cs(N |k),
c) µp ⊆ N .
Consider the diagram with solid arrows (the rows are not exact)
coker(kS!"#&$% , T, A)
η
X1 (kS , S\T, A0 )∨ H 1 (N |k, A0 )∨
φ
η0
coker(kS 0 , T, A) X1 (kS 0 , S 0 \T, A0 )∨ H 1 (N |k, A0 )∨
in which A0 := A∨ (1) = Hom(A, µp ). The horizontal maps on the left are
induced by (9.2.2) and those on the right come from the Hochschild-Serre
sequence and from conditions a), b), c).
Then in the above situation a natural dotted arrow φ exists which makes the
diagram commutative. If in addition
0
cs(N |k) ⊂
∼ S r T,
then the surjection η 0 is an isomorphism.
Proof: First observe that the homomorphism η 0 is obtained from the following
commutative exact diagram
()*+,-'. ι
H 1 (NP , A0 )
Y
H 1 (NS 0 , A0 )
S 0 \T
X1 (kS 0 , S 0 \T, A0 ) H 1 (kp , A0 )

Y
H 1 (kS 0 , A0 )
S 0 \T
(η 0 )∗ 0
1 0
H (N |k, A )
and in a similar way we get the homomorphism η.

Now consider the commutative exact diagram with natural homomorphisms
89:;<=7645/0123
X1 (kS , S\T, H 1 (kp , A0 )
Y
A0 ) H 1 (GS , A0 )
S\T
κ
α
H 1 (kp , A0 ) × (kp , A0 )
Y Y
1
ker(α) H 1 (GS , A0 ) Hnr
S\T S 0 \S
ε
X1 (kS 0 , S 0 \T, A0 ) H 1 (kp , A0 ) × H 1 (kp , A0 )

Y Y
H 1 (GS 0 , A0 )
S\T S 0 \S
β
H 1 (Tp , A0 )Gkp ,
Y
H 1 (kS 0 |kS , A0 )GS
S 0 \S
in which Tp ⊆ Gkp denotes the inertia group. Observe that β is injective, since
A0 is a trivial G(kS 0 |kS )-module and since kS has no extensions in kS 0 which are
unramified at all places in S 0 r S by definition. Diagram chasing then shows
that ε is an isomorphism. We can now define φ as the dual homomorphism to
κ ◦ (ε−1 ). Conditions a), b), c) imply that H 1 (N |k, A0 ) is canonically contained
in the groups X1 (kS , S\T, A0 ), ker(α) and X1 (kS 0 , S 0 \T, A0 ). Thus we have
constructed φ and we see that the right-hand part of the diagram is commutative.
But that the left-hand part of the diagram is also commutative can be seen from
the diagram preceding lemma (9.2.2) which defines the maps occurring.
0 0
Now assume that cs(N |k) ⊂ ∼ S r T . Then δN (S \T ) = 1, and by (9.1.9)(i),
the homomorphism ι in the exact commutative diagram at the beginning of the

proof is injective. Hence the inclusion (η 0 )∗ is an isomorphism. 2
The essential step in the proof of Šafarevič’s theorem (9.6.1) is the following
(9.6.6) Theorem. Let K|k be a finite Galois extension of the global field k
and let ϕ : Gk G(K|k) = G. Then every split embedding problem
>?@ABC Gk
1 H HoG G 1
with finite nilpotent kernel H has a proper solution.

Since a finite nilpotent group is the direct product of its p-Sylow subgroups
and since every finite G-operator p-group is a quotient of F(n)/F(n)(ν) for
some n and some ν, it suffices to show the following assertion.
For every prime number p, all n ∈ IN and all ν = (i, j), the split embedding
problem
DEFGHI Gk
1 F(n)/F(n)(ν) F(n)/F(n)(ν) o G G 1
has a proper solution Nνn |k.

Let us first assume that p =/ char(k). We will proceed by induction on ν
whereas n will be arbitrary. If ν = (1, 1), there is nothing to show. Now we
assume that we have already found a solution ϕn,ν : Gk F(n)/F(n)(ν) o G
and we consider the embedding problem
KLJ Gk
(∗) ϕn,ν
F(n)(ν) /F(n)(ν+1) F(n)/F(n)(ν+1) o G F(n)/F(n)(ν) o G.

This embedding problem is in general not solvable, but we are going to solve
it after replacing ϕn,ν by another solution ϕ̃n,ν for induction step ν on level n.
This new solution is induced by a solution
ϕm,ν : Gk → F(m)/F(m)(ν) o G
for some large m ≥ n via a suitably chosen G-invariant surjection ψ : F(m)
F(n). Let us consider the associated commutative exact diagram
SNOPQRTM Gk
(∗∗) ϕm,ν
F(m)(ν) /F(m)(ν+1) F(m)/F(m)(ν+1) o G F(m)/F(m)(ν) o G

ψ̄ ψν+1 ψν

To shorten notation we again set
F(n)/ν = F(n)/F(n)(ν) , E(n, ν) = F(n)(ν) /F(n)(ν+1) .
Since E(n, ν) is contained in the center of F(n)/ν+1, the action of F(n)/ν o G
on E(n, ν) factors through the canonical projection
F(n)/ν o G G.

In particular, GK ⊆ Gk acts trivially on E(n, ν).

Let αm and αn denote the 2-classes corresponding to the group extensions
in (∗∗) and consider the commutative exact diagram
ϕ∗m,ν
XY[ZWVU G, E(m, ν))
H 2 (F(m)/ν o H 2 (k, E(m, ν))
ψ̄∗ ψ̄∗
ϕ∗m,ν
(∗∗∗) H 2 (F(m)/ν o G, E(n, ν)) H 2 (k, E(n, ν))
ψν∗
ϕ∗n,ν
H 2 (F(n)/ν o G, E(n, ν)) H 2 (k, E(n, ν)) .
By ex.4 in I §5, we have ψν∗ (αn ) = ψ̄∗ (αm ), and by (3.5.9), the embedding prob-
lem on level n is solvable if and only if ϕ∗n,ν (αn ) = 0. We are searching for an
m ≥ n, a solution ϕm,ν on level m and a suitable surjective G-homomorphism
ψ : F(m) F(n) such that
ψ̄∗ (ϕ∗m,ν (αm )) = 0 ∈ H 2 (k, E(n, ν)).
As we will show below, this can be achieved for m large enough if the solution
ϕm,ν is of a special type. If we can guarantee that the new solution is also
of this special type, then the induction process works for a modified, stronger
statement. In fact we are going to prove the sharper
(9.6.7) Theorem. Let K|k be a finite Galois extension of the global field k and
let ϕ : Gk G(K|k) = G. In addition, let a natural number e be given. Then
for every prime number p, all n ∈ IN and all ν = (i, j), the split embedding
problem
]^_à\ Gk
ϕ
1 F(n)/F(n)(ν) F(n)/F(n)(ν) o G G 1
has a proper solution Nνn |k. If p =/ char(k), we can choose the solution in such
a way that the following conditions are satisfied:
(i) All p ∈ Ram(K|k) ∪ Sp ∪ S∞ are completely decomposed in Nνn |K.
(ii) If p is ramified in Nνn |K, then p splits completely in K(µpe )|k and Nν,p
n
|kp
is a (cyclic) totally ramified extension of local fields.
Proof: We prove the theorem by induction on ν, whereas e, n and G are

arbitrary. We defer the case char(k) = p and assume that char(k) =/ p. Then we
may restrict to the case µp ⊆ K: For this, we lift the embedding problem via
Gk G(K(µp )|k) G(K|k),

thus enlarging G. Note that this does not affect conditions (i) and (ii).
If ν = (1, 1), there is nothing to show. For the induction step ν 7→ ν + 1,
we may assume that pe is greater than the exponent of the group F(n)/F(n)(ν)
(which does depend on ν but not on n and G). We prove the induction step,
i.e. we solve the embedding problem defined by the diagram (∗) above in four
substeps.
First Step: After changing ϕn,ν the problem (∗) induces local split embedding
problems at all p ∈ Ram(K|k)∪Sp ∪S∞ and is locally solvable (not necessarily
properly) at every prime p.
a) If p ∈ Ram(K|k) ∪ Sp ∪ S∞ , then Gp (Nνn |k) = (F(n)/F(n)(ν) o G)p ∼ =
Gp (K|k) by (i). We show that, after changing ϕn,ν , the local group extensions
corresponding to these primes are split extensions. In particular, the associated
local embedding problems are solvable in a trivial way.
Let αn (p) be the 2-class in H 2 ((F(n)/F(n)(ν) o G)p , F(n)(ν) /F(n)(ν+1) ) which
corresponds to the group extension given by the upper row of the diagram
bcdefgh F(n)(ν) /F(n)(ν+1) Ep (F(n)/F(n)(ν) o G)p
F(n)(ν) /F(n)(ν+1) F(n)/F(n)(ν+1) o G F(n)/F(n)(ν) o G .

Apply the induction hypothesis to the corresponding embedding problem on
some large level m. The number m and a surjective G-invariant homomor-
phism ψ : F(m) F(n) will be chosen below. By (i), we know that p is
completely decomposed in Nνm |K and we have a commutative diagram for the
associated local groups (writing Gp for Gp (K|k))
(F(m)/νijkl o G)p Gp
(F(n)/ν o G)p Gp .
Therefore we obtain the diagram
H 2 ((F(m)/ν oqponmsr G)p , E(m, ν))

inf
H 2 (Gp , E(m, ν))
ψ̄∗ ψ̄∗
inf
H 2 ((F(m)/ν o G)p , E(n, ν)) H 2 (Gp , E(n, ν))
ψν∗
inf
H 2 ((F(n)/ν o G)p , E(n, ν)) H 2 (Gp , E(n, ν)) .

G
Using (9.6.3) with G, k = 2 and T = IndGp IFp , we find an m ≥ n and a surjective
pro-p-G homomorphism ψ : F(m) F(n) such that the homomorphism
t
H 2 (Gp , E(m, ν)) = H 2 (G, E(m, ν) ⊗ T )
ψ̄∗
H 2 (Gp , E(n, ν)) = H 2 (G, E(n, ν) ⊗ T )

−1
maps inf (αm (p)) to 0, so that the above diagram implies that αn (p) = 0.
Now we can execute the above procedure for all the finitely many primes
p ∈ Ram(K|k) ∪ Sp ∪ S∞ , making m bigger each time. Note that we do not
destroy the success already achieved for the primes p1 , . . . , pr for which the
local embedding problems already split. Indeed, the property of inducing a
split embedding problem at a prime p survives the shrinking process from m
to n if p is completely decomposed in Nνm |K (and we suppose this holds for
p ∈ Ram(K|k) ∪ Sp ∪ S∞ ). Therefore we can perform a shrinking process
for the prime pr+1 , inducing a solution from F(m), which already has the
desired property for p1 , . . . , pr (and which we could have produced by another
shrinking F(m0 ) F(m)).
An alternative way to proceed at this point is to replace the module T in
G
the above argument by the direct sum of IndGp IFp , where p runs through
Ram(K|k) ∪ Sp ∪ S∞ . In this way we deal with all these primes within one
shrinking process.
Finally, we see that the embedding problem
vuwx Gk
ϕm,ν
F(m)/F(m)(ν) o G
ψν
F(n)(ν) /F(n)(ν+1) F(n)/F(n)(ν+1) o G F(n)/F(n)(ν) o G
induces split group extensions for all p ∈ Ram(K|k) ∪ Sp ∪ S∞ .

We have just changed ϕn,ν to ψ̄ ◦ ϕm,ν , and thus Nνn to some Ñνn . But by
assumption conditions (i) and (ii) are also satisfied for Nνm , hence also for the
new field Ñνn . We will not change the notation.
b) If p is unramified in Nνn |k, then the homomorphism
ϕp
ẐZ ∼ y
= Gp (k)/T p (k) Gp (N n |k) ν
obviously extends since ẐZ is free.

c) Let p ∈ Ram(Nνn |K). Then p splits completely in the extension K(µpe )|k and
Gp (Nνn |K) ∼
= ZZ/pa ZZ with a ≤ e − 1 by the induction assumption (ii) and our
n
condition on e. Since Nν,p |Kp is totally ramified by assumption, there exists a
n √
prime element πp of Kp such that Nν,p = Kp ( pa πp ). An arbitrarily chosen pre-
image of a generator of the cyclic group Gp (Nνn |K) in F(n)/F(n)(ν+1) o G
has order pa+ε , where 0 ≤ ε ≤ 1 We can solve our embedding problem by
taking a pa+ε -th root of πp , since µpa+ε ⊆ µpe ⊆ Kp .
Second Step: The problem (∗) induces local split embedding problems at all
p ∈ Ram(K|k) ∪ Sp ∪ S∞ and is globally solvable (not necessarily properly)
after changing ϕn,ν .
As above we consider the problem for different numbers m ≥ n:
{|}~z Gk
ϕm,ν
F(m)(ν) /F(m)(ν+1) F(m)/F(m)(ν+1) o G F(m)/F(m)(ν) o G

ψ̄ ψν+1 ψν

Let αm and αn denote the 2-classes corresponding to the above group exten-
sions. A surjective pro-p-G homomorphism ψ : F(m) F(n) (inducing ψν ,
ψν+1 and ψ̄) will be defined below. Both problems (for m with ϕm,ν and for n
with ϕn,ν = ψν ◦ ϕm,ν ) are locally solvable by step 1.
In order to show the existence of a solution ϕn,ν+1 , we have to prove that the
2-class αn maps to zero under the inflation map ϕ∗n,ν , cf. (3.5.9):

Y
H 2 (kp , E(n, ν))
p
ϕ∗n,ν
H 2 (F(n)/F(n)(ν) o G, E(n, ν)) H 2 (k, E(n, ν))
X2 (k, E(n, ν)).

Here as before we have set E(n, ν) = F(n)(ν) /F(n)(ν+1) . By the first step we
can assume that
ϕ∗n,ν (αn ) ∈ X2 (k, E(n, ν)) , ϕ∗m,ν (αm ) ∈ X2 (k, E(m, ν)) .
As we already observed at the start,
ψν∗ (αn ) = ψ̄∗ (αm ) ,
which gives us
ϕ∗n,ν (αn ) = (ψν ◦ ϕm,ν )∗ (αn ) = (ϕ∗m,ν ◦ ψν∗ )(αn )
= (ϕ∗m,ν ◦ ψ̄∗ )(αm ) = (ψ̄∗ ◦ ϕ∗m,ν )(αm )

(see diagram (∗∗∗)). In order to shrink the obstruction, we look for a surjective
homomorphism onto X2 (k, E(n, ν)), which has a “shrinkable” source, i.e. to
which (9.6.3) applies.
Claim: We have a commutative diagram

ν)(−1))
H −2 (G, E(m, X2 (k, E(m, ν))
ψ̄∗ ψ̄∗
H −2 (G, E(n, ν)(−1)) X2 (k, E(n, ν))

with surjective horizontal maps, where (−1) denotes the (−1)-Tate twist.
Using this claim and (9.6.3) with k = −2, the IFp [G]-module T =
Hom(µp , ZZ/pZZ) and an element x1 which is a pre-image of ϕ∗m,ν (αm ) in
H −2 (G, E(m, ν)(−1)), we obtain a surjective pro-p-G operator homomorphism
ψ : F(m) F(n) such that ψ̄∗ ϕ∗m,ν (αm ) = 0, and so ϕ∗n,ν (αn ) = 0. Thus the
embedding problem is solvable. Furthermore, as explained in the first step, the
local condition at the primes in Ram(K|k) ∪ Sp ∪ S∞ remains unaffected by
the shrinking process. In order to finish step 2, it remains to give the
Proof of the claim: Let E(n, ν)0 = Hom(E(n, ν), µp ). By the Poitou-Tate
duality theorem, we know that
X2 (k, E(n, ν)) ∼
= X1 (k, E(n, ν)0 )∗ .
Using the Hasse principle and the fact that E(n, ν)0 is a trivial GK -module
(µp ⊆ K), we see that the homomorphism ι in the commutative exact diagram
ι
Y
H 1 (K, E(n, ν)0 ) H 1 (KP , E(n, ν)0 )
P
X1 (k, E(n, ν)0 )

Y
0 H 1 (k, E(n, ν)0 ) H 1 (kp , E(n, ν)0 )
p
H 1 (K|k, E(n, ν)0 )
0
is injective. Hence we get an injective homomorphism X1 (k, E(n, ν)0 ) ,→
H 1 (K|k, E(n, ν)0 ), and noting that the dual of cohomology is homology (hence
is cohomology in negative dimensions for finite groups), we obtain a canonical
surjection
H 1 (G, E(n, ν)0 )∗ X1 (k, E(n, ν)0 )∗
H −2 (G, E(n, ν)(−1)) X2 (k, E(n, ν)) .

This proves the claim.

Third step: After changing ϕn,ν , the problem (∗) has a proper global solution
n
which satisfies condition (i) and all primes p ∈ Ram(Nν+1 |Nνn ) r Ram(Nνn |K)
are completely decomposed in Nνn |k. Furthermore, the local extension
n
Nν+1,p |Kp is (cyclic) totally ramified for p ∈ Ram(Nνn |K).
We achieve this with the following procedure. Consider a solution ϕn,ν+1 of
the embedding problem (∗) which we obtained in step 2. Its equivalence class
[ϕn,ν+1 ] is an element of the space S(∗) of solutions of (∗) modulo equivalence.
Conditions (i), (ii), properness and the other conditions that we want to achieve
in this third step, only depend on the equivalence class of a solution. The space
S(∗) is a principal homogeneous space over H 1 (Gk , E(n, ν)); see (3.5.11).
Recall that the action is defined as follows: choose a representing cocycle
Gk → E(n, ν) and multiply a solution Gk → F(n)/ν +1 o G of the embedding
problem with the cocycle. This yields a map Gk → F(n)/ν + 1 o G which is a
homomorphism, and the equivalence class of this new solution is independent
of the choices made.
Now we look for a suitable cohomology class ε ∈ H 1 (Gk , E(n, ν)) such that
the new solution
ϕ̃n,ν+1 = ε ϕn,ν+1
has the required properties. We assume that ϕm,ν+1 is obtained from step 2 and
that ϕm,ν satisfies (i) and (ii). Note that the properness of the solution is only
a problem for the first step (1, 1) → (2, 1), since in all higher induction steps,
the properness follows automatically from the induction hypothesis and from
the Frattini argument.
Let us consider how the local behaviour of ϕ̃n,ν+1 = ε ϕn,ν+1 is connected
to that of ϕn,ν+1 . By this we mean that we want to compare the ramification
n
and decomposition of primes in the associated field extensions Ñν+1 |Nνn and
n n
Nν+1 |Nν . (Since we do not know whether the solutions are proper, one or
both of these field extensions might be trivial.) Let p be a prime in Nνn . The
n
behaviour of p in Nν+1 |Nνn is characterized by the homomorphism
ϕn,ν+1 |G(Nνn )p ∈ Hom(G(Nνn )p , E(n, ν)).
Since G(Nνn )p acts trivially on E(n, ν), we can interpret ϕn,ν+1 |G(Nνn )p as an
element in
H 1 ((Nνn )p , E(n, ν))Gkp .
Consider the exact sequence
α β
0 −→ H 1 ((Nνn )p |kp ) −→ H 1 (kp ) −→ H 1 ((Nνn )p )Gkp −→ H 2 ((Nνn )p |kp ) ,
which is obtained from the Hochschild-Serre sequence for the tower of fields
k̄p | (Nνn )p | kp and in which E(n, ν) are the coefficients (not written) of the
cohomology groups. We see that ϕ̃n,ν+1 |G(Nνn )p is given by
ϕn,ν+1 |G(Nνn )p + α(ε) ∈ H 1 ((Nνn )p , E(n, ν))Gkp .

Now we choose a finite set T 0 of primes in cs(Nνn |k) and homomorphisms

xp : Gkp /Tkp = G(Nνn )p /T(Nνn )p −→ E(n, ν)
for p ∈ T 0 such that their images generate E(n, ν). (The set T 0 will be
responsible for the properness of the new solution.)
Set
T1 = Ram(K|k) ∪ Sp ∪ S∞ ,
T2 = Ram(Nνn |K) ,
n
T3 = Ram(Nν+1 |K) r(Ram(Nνn |k) ∪ Sp ∪ S∞ ) ,
T = T 0 ∪ T 1 ∪ T 2 ∪ T 3 and
S = cs(Nνn (µpe )|k) ∪ T .
Let p ∈ T 0 . Then, since p splits completely in Nνn |k, there exists ξp ∈

H 1 (kp , E(n, ν)) with
α(ξp ) = xp : G(Nνn )p −→ E(n, ν) .
Let p ∈ T 1 . Then, by the induction hypothesis, the extension (Nνn )p |Kp is

trivial and the group extension in the diagram
Gkp
ϕn,ν+1 |Gk
p
ϕn,ν |Gk
p
1 E(n, ν) Ep Gp (Nνn |k) 1

splits. Hence β(ϕn,ν+1 |G(Nνn )p ) = 0 and we can therefore find an element
ξp ∈ H 1 (kp , E(n, ν)) such that
α(ξp ) = −ϕn,ν+1 |G(Nνn )p : G(Nνn )p −→ E(n, ν) .
If p ∈ T 2 , then by the induction hypothesis p ∈/ T 1 , Kp |kp is trivial and (Nνn )p |Kp

is a (cyclic) totally ramified extension. We consider the commutative exact
diagram
1
Hnr (kp ) Hnr 1
((Nνn )p )Gkp
α
H 1 ((Nνn )p |kp ) H 1 (kp ) H 1 ((Nνn )p )Gkp .
Since p ∈ T 2 , the dotted arrow in the diagram above is an isomorphism. Thus
there is a ξp ∈ H 1 (kp , E(n, ν)) such that
ϕn,ν+1 |G(Nνn )p + α(ξp ) : G(Nνn )p −→ E(n, ν)
is either trivial (if ϕn,ν+1 |G(Nνn )p is unramified) or induces a totally ramified
extension of degree p of (Nνn )p .
If p ∈ T 3 , then we have a commutative exact diagram

H 1 ((Nν¢£ ¡ n )p |kp ) H 1 (kp ) α

H 1 ((Nνn )p )Gkp
res
H 1 (Tkp )Gkp H 1 (T(Nνn )p )Gkp

where now the lower dotted arrow is an isomorphism, since (Nνn )p |kp is un-
ramified. Let ξp ∈ H 1 (kp , E(n, ν)) be such that
ϕn,ν+1 |G(Nνn )p + α(ξp ) : G(Nνn )p −→ E(n, ν)
is unramified.
In order to complete step 3, it is therefore sufficient to show the existence of
an element ε ∈ H 1 (GS , E(n, ν)) ⊆ H 1 (Gk , E(n, ν)) with εp = ξp for all p ∈ T .
The exact sequence
H 1 (GS , ¤¥ E(n, ν)) πn
Y
H 1 (kp , E(n, ν)) coker(kS , T, E(n, ν))
T
shows that the obstruction to the existence of such an ε is the vanishing of
πn (ξn ) with
Y Y
ξn = ξp ∈ H 1 (kp , E(n, ν)).
p∈Tn p∈Tn
(In the following we denote the sets T i and T by Tni and Tn , respectively, in
order to indicate at which level the embedding problem is considered.) By
(9.2.2), we have a canonical injection
(1) coker(kSn , Tn , E(n, ν)) ,→ X1 (kSn , Sn r Tn , E(n, ν)0 )∨ ,
where E(n, ν)0 = Hom(E(n, ν), µp ).
Recall that H −2 = H1 and that by the induction hypothesis, the solution ϕn,ν
is proper. Thus we obtain
H1 (F(n)/ν o G, E(n, ν)(−1)) ∼
= H 1 (F(n)/ν o G, E(n, ν)∨ (1))∨
∼
= H 1 (N n |k, E(n, ν)0 )∨ . ν
Therefore there exists a canonical isomorphism

(2) H −2 (F(n)/ν o G, E(n, ν)(−1)) −→
∼ H 1 (N n |k, E(n, ν)0 )∨ .
ν
Now we are going to shrink the obstruction to the existence of a 1-class ε

as above. If ϕn,ν is induced by a ϕm,ν for m ≥ n via a G-invariant surjection
F(m) F(n), then the inclusion (1) is obviously also true in the form
(1)0 coker(kSm , Tm , E(n, ν)) ,→ X1 (kSm , Sm r Tm , E(n, ν)0 )∨ ,
where Sm and Tm are chosen as above but at the level Nνm . Using (9.6.4)
we choose m ≥ n so that an arbitrarily chosen element in H −2 (F(m)/ν o
G, E(m, ν)(−1)) is annihilated by the map which is induced by a suitably

chosen surjection F(m) F(n). Then we consider the diagrams, in which

we write c for coker, Ed for E(d, ν) (d = m, n) and S̃n = cs(Nνn |k) ∪ Tm :
H 1 (kSm¦§°±²³´µ¶®¯¬ª«¨© |k, Em )
Y
π
H 1 (kp , Em ) c(kSm , Tm , Em ) X1 (kSm , Sm \Tm , Em
0 ∨
)
Tm
ψ
Y
π
H 1 (kSm |k, En ) H 1 (kp , En ) c(kSm , Tm , En ) X1 (kSm , Sm \Tm , En0 )∨
Tm
inf
Y
π
H 1 (kS̃n |k, En ) H 1 (kp , En ) c(kS̃n , Tm , En ) X1 (kS̃n , S̃n \Tm , En0 )∨ ,
Tm
X1 (kSm , Sm·¸¹º»¼½¾¿À \Tm , Em

0 ∨
) 0 ∨
H 1 (Nνm |k, Em ) H −2 (F(m)/ν o G, Em (−1))
X1 (kSm , Sm \Tm , En0 )∨ H 1 (Nνn |k, En0 )∨
X1 (kS̃n , S̃n \Tm , En0 )∨ H 1 (Nνn |k, En0 )∨ H −2 (F(n)/ν o G, En (−1)).
The existence of all maps and the fact that the diagrams are commutative follow
from the arguments above and from (9.6.5).
Now let
Y Y
ξm = ξp ∈ H 1 (kp , E(n, ν))
p∈Tm p∈Tm
be arbitrary. By theorem (9.6.4)(i), we can choose a G-invariant surjection
ψ : F(m) F(n) such that πn ψ∗ (ξm ) = 0.
Observe that we have not yet got precisely what we want, because ε has the
required property with respect to the sets of primes Tmi and S̃n . Nevertheless,
one easily verifies that if we modify the solution, which we have obtained
after shrinking, by the cocycle ε (which now exists), then we obtain a solution
satisfying all required properties. This finishes step 3.
Fourth Step: After changing ϕn,ν again, there exists a proper solution ϕn,ν+1
of (∗) which satisfies properties (i) and (ii).
The solution ϕn,ν+1 which we obtained in step 3, has almost all properties
n
we need, except that for p ∈ Ram(Nν+1 |K) r Ram(Nνn |K) the local extension
n
(Nν+1 )p |kp might not be (cyclic) totally ramified. But we know that for such a
prime p the extension (Nνn )p |kp is trivial (in fact, Nνn (µpe )p |kp is trivial). In order
to get a totally ramified cyclic extension, we have to remove the unramified

n
part of the extension (Nν+1 )p |(Nνn )p and to make sure that at places where
new ramification occurs by this procedure, we have cyclic local extensions
(these are automatically totally ramified, since the decomposition group is an
elementary abelian p-group).
In order to retain the properness of the solution obtained in step 3, we
choose a finite set of primes T 0 ⊆ cs(Nνn |k) r(Ram(Nν+1n
|k) ∪ Sp ∪ S∞ ) such
n n 0 n n
that Gp (Nν+1 |Nν ), p T , generate G(Nν+1 |Nν ).
∈
We want to alter the solution found in step 3 once again using a class x in
1
H (kS |K, E(n, ν)), where
· S = cs(Nνn (µpe )|k) ∪ T with T = Ram(Nν+1 n
|k) ∪ Sp ∪ S∞ ∪ T 0 ,
· for p ∈ Ram(Nνn |k) ∪ Sp ∪ S∞ ∪ T 0 , we have xp = 0 ,
n
· if the prolongations of p to K are in Ram(Nν+1 |K) r Ram(Nνn |K), then
xp ∈ Hnr1
(kp , E(n, ν)) = Hnr 1
((Nνn )p , E(n, ν)) has the property that
ϕn,ν+1 |G(Nνn )p + xp ∈ H 1 ((Nνn )p , E(n, ν))
is cyclic,
· xp is cyclic for all p ∈/ T .
n
For every p ∈ S(k) such that prolongations of p to K are in Ram(Nν+1 |K) but
n
not in Ram(Nν |K), we fix a prolongation p0 S(K) of p to K (note that p
∈
splits completely in K|k). Let
Y
η ∈ H 1 (KP , E(n, ν))
T
be such that
· ηP = 0 if P ∩ k ∈ Ram(Nνn |k) ∪ Sp ∪ S∞ ∪ T 0 ,
n
· if P ∈ Ram(Nν+1 |K) r Ram(Nνn |K) and P =/ (P ∩ k)0 , then ηP = 0,
n
· if P ∈ Ram(Nν+1 |K) r Ram(Nνn |K) and P = (P ∩ k)0 , then
1
ηP ∈ Hnr (KP , E(n, ν)) = Hnr 1
((Nνn )P , E(n, ν)) has the property that
ϕn,ν+1 |G(Nνn )P + ηP ∈ H 1 ((Nνn )P , E(n, ν))
is cyclic.
Applying theorem (9.3.2) in the situation where Ω is the field Nνn (µpe ) and
A = E(n, ν), we see that in order to finish the proof, it suffices to construct an
element y in H 1 (kS |K, E(n, ν)) with yP = ηP for all P ∈ T . Indeed, using this
procedure, we get new ramification only at places which are completely decom-
posed in Nνn (µpe )|k. Hence their decomposition groups are cyclic (by (9.3.2))
and contained in the p-elementary abelian group G(Nν+1 n
|Nνn ) ∼
= E(n, ν). Thus
the local extensions associated to these new ramification primes are cyclic of
order p, and, in particular, are totally ramified. Furthermore, by the choice of
T 0 , the new solution remains proper.
Similarly to the situation with the class ε in step 3, the exact sequence

H 1 (KS |K,ÁÂ E(n, ν)) πn

Y
H 1 (Kp , E(n, ν)) coker(KS , T, E(n, ν))
T
shows that the obstruction to the existence of such a y is πn (η) = 0.
Now we apply the shrinking procedure as in step 3, but the commutative
diagrams used there have to be modified as follows: replace k by K in the first
diagram and instead of the second consider the following diagram
ÃÄÅÆÇÈÉÊËÌ \Tm , Em
X1 (KSm , Sm 0 ∨
) 0 ∨
H 1 (Nνm |K, Em ) H −2 (F(m)/ν, Em (−1))
X1 (KSm , Sm \Tm , En0 )∨ H 1 (Nνn |K, En0 )∨
X1 (KS̃n , S̃n \Tm , En0 )∨ H 1 (Nνn |K, En0 )∨ H −2 (F(n)/ν, En (−1)).
Then we use part (ii) of theorem (9.6.4) instead of part (i).

Therefore, after a further shrinking, we get a class y with the properties
above, and theorem (9.3.2) then induces the existence of the desired class
x ∈ H 1 (kS |k, E(n, ν)).
The new solution ϕ̃n,ν+1 = x ϕn,ν+1 fulfills conditions (i) and (ii), hence step
4 and the proof in the case p =/ char(k) of theorem (9.6.7) are complete.
The proof in the case p = char(k) is comparatively easy. Again we proceed

by induction on ν, where n is arbitrary. The case ν = (1, 1) is trivial. In the
case ν = (2, 1) we get an embedding problem with abelian kernel isomorphic to
IFp [G]n , which is properly solvable by (9.2.9). In the next induction steps we do
not care about the properness of the solutions, because they are automatically
proper by the Frattini argument. By (6.1.4), we have cdp Gk = 1, so that
Gk is p-projective by (3.5.6) and we can solve the embedding problems in
all induction steps. Therefore the proof of (9.6.7) and also that of (9.6.6) is
complete. 2
In order to deduce the theorem of Šafarevič, we use an argument which goes

back to O. ORE [165]. We need two facts from group theory and we recall the
following notation. If G is a finite nontrivial group, then
Φ(G) is the intersection of all maximal subgroups of G and is called the
Frattini subgroup of G,
F (G) is the composite of all nilpotent normal subgroups of G and is
called the Fitting subgroup of G.

The group Φ(G) is a characteristic subgroup of G and is contained in F (G).

The group F (G) is a normal nilpotent subgroup of G. We cite the following
two facts, see [81], Kap.III, Satz 3.2 (b) and Satz 4.2 (c).
(9.6.8) Proposition. Let N be a normal subgroup of the finite group G such

that N ⊆/ Φ(G). Then there exists a partial complement U of N in G, i.e.
U = G and G = N · U .
/
(9.6.9) Proposition. Let G be a nontrivial finite solvable group. Then Φ(G) is

a proper subgroup of F (G).
Proof of Šafarevič’s theorem: Let F (G) be the Fitting subgroup of G =/ {1}.

By the two propositions above, F (G) has a (solvable) partial complement
U $ G, so there exists a surjection
F (G)Í o U G.
Assuming inductively (on the order of G) that U is the Galois group of a finite
normal extension of k, we obtain the result using theorem (9.6.6). 2
Finally, we would like to mention the following corollary to (9.6.7).
(9.6.10) Theorem. Let K|k be a finite Galois extension of the global field k and
let ϕ : Gk G(K|k) = G be the natural projection. Then every embedding
problem
ÏÐÑÒÎ Gk
ϕ
1 N E G 1
with finite nilpotent kernel N which has a solution, can also be solved properly.
Proof: Let ψ : Gk → E be a solution of the embedding problem. Then we

obtain a commutative exact diagram
ÓÔÕÖ×ØÙÚÛÜÝÞ Gk
1 N N o im(ψ) im(ψ) 1
1 N E G 1,

where the action of im(ψ) ⊆ E on N is induced by the action of E on N , and

the surjection N o im(ψ) E is given by the inclusions of N and im(ψ)
into E. By (9.6.6), the split embedding problem given by the upper row of the
diagram above has a proper solution, hence the initial problem is also properly
solvable. 2
Exercise: (a) Go through the proof of Šafarevič’s theorem and show that one can choose K
in such a way that finitely many given primes of k split completely in K (if char(k) = p > 0,
use (9.2.6)).
(b) If k is a number field, prove that the extension K|k which realizes the solvable group G
can be chosen to be tamely ramified.


Chapter X
Restricted Ramification
In this chapter we will investigate Galois groups of global fields with re-
stricted ramification. Let K be a global field and S be any set of primes of K.
We are interested in properties of the Galois group
GS = GS (K) = G(KS |K) ,
where KS is the maximal extension of K which is unramified outside the
primes in S.
The case S = {all places of K}, i.e. GS = GK , was extensively studied in
chapter IX. The case when S is finite and especially the case S = ∅, where GS
is the Galois group of the maximal unramified extension of K, is of particular
arithmetic interest.
One soon observes that the function field case is comparatively easy to deal
with. The reason is that if S is finite, then we can view GS as the étale
fundamental group of a smooth algebraic curve over a finite field. The closed
points of this curve X correspond to the places of K which are not in S. Then
we can translate results from algebraic geometry into properties of GS .
The reason why such a method works is the fact that curves (except proper,
rational curves) are K(π, 1)-spaces in the sense of (étale) homotopy. This
means that the universal covering space X̃ of a curve X has trivial homo-
topy groups (and is contractible, in a rather vague sense). In particular, the
Hochschild-Serre spectral sequence for the covering X̃|X is degenerate, i.e.
identifies the cohomology of the fundamental group with the cohomology of
the space itself.
Having this geometric background in mind, it is not surprising that there
are qualitatively different results in the cases S = ∅ (projective curve) and
S =/ ∅ (affine curve), and that the case of a rational curve causes an additional
exception.
We will treat the function field case in §1, referring there to some well-known
results of algebraic geometry.
The rest of this chapter will be devoted to the much more complicated
number field case, where we are mainly interested in the case when S is finite.
The reasons that the number field case is more difficult are at least twofold.

600 Chapter X. Restricted Ramification
In the first instance, the affine scheme Spec(OK,S ) is in general not a K(π, 1).
×
However, for the finitely many p ∈ IN(S) = IN ∩ OK,S do the p-parts of the
et
higher homotopy groups πi (Spec(OK,S )) vanish. ) ∗
The second reason is the occurrence of wild ramification. In contrast to the

function field case, there is not only one but infinitely many different residue
characteristics in OK,S . In order to get information about the p-part of the
cohomology of GS we must, by the reasons mentioned before (or just because
of the necessary conditions of the Poitou-Tate theorem), enlarge the set S to
S ∪ Sp . But then we are led to consider the rather difficult higher ramification
groups at the places dividing p. For example, we will see in §3 that the question
of the strict cohomological p-dimension of GS is independent of the finite set
S ⊇ Sp ∪ S∞ and is related to the Leopoldt conjecture.
As a consequence of these difficulties we will restrict to easier classes of
Galois groups, i.e. we consider the maximal pro-c-factor of GS , where c is a
full class of finite groups, which will be in most cases the class of p-groups for
some prime number p.
In order to do this, our first task is to calculate the cohomology groups of
GS (c). In §4 we will investigate the group GS (L) for large number fields, i.e.
for number fields of infinite degree over Q. Given a prime number p, we will
give several criteria on L and S for the vanishing of H i (GS (L), ZZ/pZZ) for
i ≥ 2. As a corollary we obtain a result due to O. NEUMANN, which asserts that
the inflation maps
H i (GS (p), A)(p) −→ H i (GS , A)(p)
are isomorphisms for every discrete GS (p)-module A and all i, provided that
p ∈ IN(S).
The next problem we are confronted with is the question of whether, for a
given prime p ∈ S(K), the canonical homomorphism
G(Kp (c)|Kp ) Gp (KS (c)|K) ⊆ G(KS (c)|K)
is injective. This is a local-global problem and equivalent to the question of
which local c-extensions can be globally realized inside KS (c). In chapter IX
we have given a positive answer for every full class c if S is of density 1. The
similar question for abelian extensions (the class of abelian groups is not a full
class) is answered by the theorem of Grunwald-Wang.
However, if S is finite, then there are arithmetic obstructions to the global
realization of local extensions. For instance, the group GS might be finite or
∗) We will neither explain nor use étale homotopy in the following. We refer the reader to [54]
for the definition of étale homotopy groups. See [193] for a proof that πiet (Spec(OK,S ))(p) = 0
for i ≥ 2 and p ∈ IN(S).

601
even trivial. Therefore we restrict to the case when c is the class of p-groups
where p is a prime number such that S ⊇ Sp . In §5 we show the number
theoretical analogue of Riemann’s existence theorem which asserts that the
Galois group G(k(p)|kS (p)) decomposes into a free pro-p-product of inertia
groups.
The prime number p = 2 frequently causes technical problems and requires

additional efforts. In Galois cohomology these problems are essentially due
to the fact that the decomposition groups of the real places are 2-groups. This
was the reason that, so far, we have always assumed k to be a totally imaginary
number field if we considered the case p = 2. In section §6 we consider the
group GS (2) when not all real places are in S.
In §7 we calculate the dimensions of the cohomology groups H i (GTS , ZZ/pZZ)

for i = 1, 2, where GTS = G(KST |K) is the Galois group of the maximal
extension of K which is unramified outside S and completely decomposed at
the primes of the set T .
In §8 we present the following result due to L. V. KUZ’MIN : if S ⊇ Sp ∪ S∞

for a prime number p with µp ⊆ k, then in the generic case
(KS (p))p = Kp (p)
for all primes p in K.
In §9 we derive a criterion for the group GS (p) to be a free pro-p-product

of local groups. If GS (p) does not decompose in this way, then it is a pro-p
duality group of dimension 2 in many cases.
The question whether for a given prime number p the group GS (p) is infinite
or not, is trivial if Sp ⊆ S. In the case S = ∅ it became prominent as the
problem of “p-class field towers”. We will consider that problem in §10; in
particular, we derive a sharpened form of the classical result of E. S. GOLOD
and I. R. ŠAFAREVIČ, who constructed the first infinite p-class field towers in
1964.
In the final section §11, we investigate the profinite group GS rather than its
pro-p-factors. We will prove a duality theorem for this group and a second form
of the Poitou-Tate duality for finitely generated GS -modules. Furthermore,
using analytic results, we will present a theorem of Y. IHARA which asserts that
for an infinite unramified Galois extension L of a number field K, the set of
primes of K which split completely in L cannot be “too big”.

§1. The Function Field Case
Assume that K is a global field of char(K) = p > 0 and let k ⊆ K be the

finite constant field with algebraic closure k̄. Let S be a finite set of places
of K and let GS = G(KS |K). Up to this point we have always assumed
S to be nonempty but from now on the case S = ∅ will be included. We
denote the complete curve over k which is associated to K by X. Then the
extensions of K inside KS correspond to the unramified covers of the curve
X r S, which is obtained by removing the finitely many points from X that
correspond to places in S. We denote the (geometric) genus of the curve X by
g = g(X) = g(K). Now consider the exact sequence
1 −→ GS (K k̄) −→ GS −→ G(k̄|k) −→ 1.
The group G(k̄|k) is a free profinite group of rank 1, and the Frobenius auto-
morphism Frob ∈ G(k̄|k) is a canonical topological generator. We will denote
the unique extension of k inside k̄ of degree n ∈ IN by kn .
The group GS (K k̄) is related to the unramified covers of Xk̄ r S, where
Xk̄ is the base change of X to k̄. Recall that the group ClS (K), resp.
ClS (K k̄), may be geometrically interpreted as the Picard group Pic(X r S),
resp. Pic(Xk̄ r S). When S = ∅, there exists a degree map
deg : Cl(K) → ZZ
which assigns to the class [p] of a prime divisor the degree [k(p) : k]. The
homomorphism deg is surjective and we denote its kernel by Cl0 (K). Passing
from k to k̄, we obtain an exact sequence
deg
0 −→ Cl0 (K k̄) −→ Cl(K k̄) −→ ZZ −→ 0.
The following result is not very deep from the point of view of algebraic
geometry, but it will be of crucial importance for our investigation of GS .
A similar result in the number field case would be desirable; compare the
discussion of “µ = 0” in chapter XI.
(10.1.1) Proposition. The group Cl0 (K k̄) is divisible. More precisely,

Cl0 (K k̄) ∼ (Q` /ZZ` )2g ⊕ (Qp /ZZp )h ,
M
=
`=
/p
where g is the genus of X and 0 ≤ h ≤ g. If S is nonempty, then the group

ClS (K k̄) is also divisible.
Remark: The number h is the height of the p-divisible group attached to the
Jacobian variety of X. It is also called the p-rank or the Hasse-Witt invariant
of X.

§1. The Function Field Case 603
Proof: Proposition (10.1.1) is a consequence of the existence of the Jacobian

variety of the curve X. In fact, the group Cl0 (K k̄) can be canonically identified
with the group of k̄-rational points of the abelian variety Jac(X) (see [141]).
The abelian variety Jac(X) is self-dual and its dimension is equal to the genus
g of X (loc.cit.). Therefore the torsion subgroup of Cl0 (K k̄) has the desired
structure by the general theory of abelian varieties (see [140]). On the other
hand, the group Cl0 (Kkn ) is finite for every n ∈ IN, because it is a subgroup
of the group of kn -rational points of Jac(X) (see [141], Remark 1.5)∗) . But
the latter is a finite group. Hence Cl0 (Kkn ) is finite. (This is a classical result,
which can be found in various places in the literature as the function field
analogue of the finiteness of the ideal class group of number fields.) Therefore
Cl0 (K k̄) = lim
−→
Cl0 (Kkn ) is the direct limit of finite groups, hence torsion.
n
Finally, if S is nonempty, then ClS (K k̄) is a quotient of Cl0 (K k̄). 2
We denote the class of finite groups of order prime to p = char(K) by (p0 )

0
and we write A(p ) for the (p0 )-torsion subgroup of an abelian group A, i.e.
the subgroup of elements of finite prime-to-p order. The maximal pro-(p0 )-
0
quotient of a profinite group G will be denoted by G(p ) . Let n be the number
of geometric points in S, i.e. n = #S(K k̄).
(10.1.2) Theorem. (i) The group GS (K k̄) has the following properties.
a) If S = ∅ and g = 0, then GS (K k̄) = {1}.
b) If S = ∅ and g > 0, then GS (K k̄) is a Poincaré group of dimension 2 at
the class (p0 ) and scd(p0 ) GS (K k̄) = 3.
c) If S =/ ∅, then cd` GS (K k̄) = 1 for every ` ∈ (p0 ).
(ii) For every prime number ` ∈ (p0 ) there exists a presentation of the maximal
pro-`-factor group of GS (K k̄) as a pro-`-group by 2g + n generators and one
relation of the form
GS (K k̄)(`) ∼
= hx1 , . . . , x2g , y1 , . . . , yn | (x1 , x2 )· · ·(x2g−1 , x2g ) y1 · · · yn = 1i.
In particular, if S =/ ∅, then GS (K k̄)(`) is a free pro-`-group of rank 2g + n − 1.

Furthermore, the elements y1 , . . . , yn may be chosen as generators of the
procyclic inertia groups of prolongations of primes in S(K k̄) to KS (`).
∗) In fact, this canonical inclusion is an isomorphism in our situation, since Br(k ) = 0 (see
n
[141], Remark 1.6.).

(10.1.3) Corollary. The group GS = GS (K) has the following properties.

(i) If S = ∅ and g = 0, then GS = G(k̄|k) ∼ = ẐZ.
(ii) If S = ∅ and g > 0, then GS is a Poincaré group at (p0 ) of dimension 3
with dualizing module µ, i.e. the module of all roots of unity is contained
in KS . Furthermore, in this case scd(p0 ) GS = 3.
(iii) If S =/ ∅, then GS is a duality group at (p0 ) of dimension 2. Its (p0 )-
dualizing module D2 = D2 (ZZ(p0 ) ) is canonically isomorphic to the (p0 )-
torsion part of the S-idèle class group of KS . In particular, there exists
an exact sequence
M G
0 −→ µ −→ IndGpS µ −→ D2 → 0,
p∈ S
where Gp is the decomposition group in GS for a fixed prolongation of

the prime p ∈ S to KS .
Proof of theorem (10.1.2) and corollary (10.1.3): If g = 0, we have Xk̄ ∼ =

IP1k̄ and statement (i) a) follows from the Hurwitz genus formula, which shows
that there must be ramification in every separable cover of IP1k̄ (see [77],
chap.IV, §2). This also shows assertion (i) of the corollary.
In order to prove c), observe that (K k̄)p = Kpnr , where Kpnr denotes the
maximal unramified extension of the local field Kp . Since cd(p0 ) G(K p |Kpnr )
≤ 1 by (7.1.8), we can calculate for ` ∈ (p0 )
H 2 (GS (K k̄), µ` ) = lim

−→
X2 (GS (Kkn ), µ` )
n
= lim
−→
X1 (GS (Kkn ), ZZ/`ZZ)∨
n
= ClS (K k̄)/` = 0,
by (10.1.1). Since the cyclotomic character is trivial on GS (K k̄) and since
the same arguments also hold for every finite extension of K inside KS , we
conclude that cd` GS (K k̄) ≤ 1 for every ` ∈ (p0 ). Furthermore, the Kum-
mer sequence 0 → µ` → OS× → OS× → 0, together with the equality
H 1 (GS (K k̄), OS× ) = ClS (K k̄) (see (8.3.11)), implies that there is an exact
sequence
×
0 → OK k̄,S
/` → H 1 (GS (K k̄), µ` ) → ` ClS (K k̄) → 0.
In particular, H 1 (GS (K k̄), µ` ) is finite and nontrivial unless g = 0 and n = 1.
In the latter case, Xk̄ r S ∼= A1k̄ = Spec(k̄[t]) and there exist no cyclic unram-
ified covers of degree prime to p. But in this case there exist (many) cyclic
covers of degree p which have positive genus (see ex.1). Hence GS has open
subgroups which correspond to curves of positive genus. We conclude that
cd` GS (K k̄) = 1 for every ` ∈ (p0 ) if S =/ ∅, which shows c). Furthermore, the

assumptions of lemma (3.7.5) are fulfilled and therefore GS (K) is a duality

group of dimension 2 at (p0 ) in this case. Finally, from (8.4.3)(i) it follows
that NGS (m CS ) = 0, thus by the duality theorem (8.4.4) we have a canonical
isomorphism
∼ 2
m CS (L) = H (GS (L), ZZ/mZZ)∨
for (m, p) = 1 and for every finite extension L of K in KS . Passing to the limit
over all L and m, we deduce the statement about the dualizing module. The
exact sequence for D2 then follows from the exact sequence
0 −→ OS× −→ IS −→ CS −→ 0,
noting that OS× is `-divisible for all ` ∈ (p0 ).
Now assume that S = ∅ and g > 0. In order to prove b), it is sufficient
to show that for every ` ∈ (p0 ) and every open subgroup U ⊆ G∅ (K k̄),
the maximal pro-`-quotient U (`) is an (infinite) Demuškin group. Since the
assumptions carry over to every finite separable extension of K, we may restrict
to the case U = G∅ (K k̄). The Kummer sequence implies that
G∅ (K k̄)ab (p ) ∼
0 0
= H 1 (G∅ (K k̄), Q/ZZ(p ) )∨
∼
= H 1 (G∅ (K k̄), µ)∨ (1)
∼ 0
= (Cl0 (K k̄)(p ) )∨ (1)
and by (10.1.1) we know the structure of Cl0 (K k̄) as an abelian group. In
particular, G∅ (K k̄)ab (`) is nontrivial and torsion-free for every ` ∈ (p0 ), and
for every open subgroup V ⊆ G∅ (K k̄)(`), the Hurwitz formula implies the
rank equality
d(V ) = (G∅ (K k̄)(`) : V )(d(G∅ (K k̄)(`)) − 2) + 2,
where d(V ) = dimIF` H 1 (V, IF` ) (cf. III §9). Therefore G∅ (K k̄)(`) is not free
(see (4.2.2)).
Since H 2 (G∅ (K k̄)(`), µ` ) is a subgroup of H 2 (G∅ (K k̄), µ` ) (use the Hoch-
schild-Serre spectral sequence for the group extension 1 → R → G∅ (K k̄) →
G∅ (K k̄)(`) → 1 and note that H 1 (R, µ` ) = 0), we conclude that
dimIF` H 2 (G∅ (K k̄), µ` ) ≥ 1.
Now let T be a finite and nonempty set of places of K, and suppose that the
primes in T are of degree 1, i.e. they do not decompose in K k̄|K. (We may
replace K by Kkn for a suitable n.) If H := G(KT |K∅ ), we have an exact
sequence
(∗) 1 −→ H −→ GT (K k̄) −→ G∅ (K k̄) −→ 1.
The inertia groups Tp (KT |K∅ ) = Gp (KT |K∅ ) for p ∈ T (fix any prolongation
to KT ) generate H as a normal subgroup in GT (K k̄). Therefore the canonical
homomorphism

β : H 1 (H, ZZ/`ZZ)G∅ (K k̄) −→ H 1 (Kpnr , ZZ/`ZZ)

M
p∈T
is injective.
Denote the `-dualizing module of GT (K) by I. It is also the `-dualizing
module for GT (Kkn ), n ≥ 1, and by (3.7.5) I is also the `-dualizing module
for the group GT (K k̄).
Now consider the following commutative exact diagram
çßàáâãäåæ 0 H 1 (G∅ (K k̄), µ` ) H 1 (GT (K k̄), µ` )

α
0 H 0 (GT (K k̄), µ` ) H 0 (GT (K k̄), ` I) H 1 (GT (K k̄), µ` )
ïèéêëìíî H 0 (G∅ (K k̄), H 1 (H, µ` )) H 2 (G∅ (K k̄), µ` ) 0

β γ
H 1 (Kpnr , µ` )
M
H 1 (GT (K k̄), ` I) 0.
p∈T
The upper sequence is the Hochschild-Serre spectral sequence for the group
extension (∗) and the (trivial) module µ` . By c) we know that cd` H ≤
cd` GT (K k̄) = 1. The lower exact sequence is obtained by passing to the
limit over n and over the Poitou-Tate sequences for GT (Kkn ) and the module
µ` , where we use in addition the duality group property of GT (Kkn ).
Since β is injective, α exists and is surjective, and diagram chasing shows
that γ is injective. But H 1 (GT (K k̄), ` I) ∼
= ZZ/`ZZ, hence γ is an isomorphism,
2
because we already know that H (G∅ (K k̄), µ` ) is nontrivial. (Thus β is also
an isomorphism.)
We conclude that G∅ (K k̄)(`) is a one-relator pro-`-group and the generator
ranks of open subgroups can be calculated by the Hurwitz genus formula. This
implies already that G∅ (K k̄)(`) is a Demuškin group by (3.9.15). However,
we will show directly that the pairing
∪
H 1 (G∅ (K k̄), ZZ/`ZZ) × H 1 (G∅ (K k̄), µ` ) −→ H 2 (G∅ (K k̄), µ` ) ∼
= ZZ/`ZZ
is non-degenerate. Since the module µ` has trivial G∅ (K k̄)-action and since
the associated cohomology groups of G∅ (K k̄)(`) are canonically the same for
µ` and ZZ/`ZZ (which is trivial for H 1 and was seen above for H 2 ), it then
follows from (3.7.6) that G∅ (K k̄)(`) is a Demuškin group, hence showing
assertion (i) b) of (10.1.2).
It remains to show that the cup-product pairing is non-degenerate. Consider

H 1 (G∅ (Kôóðòñ k̄), ZZ/`ZZ) × H 1 (G∅ (K k̄), µ` ) ∪

H 2 (G∅ (K k̄), µ` )
inf α γ
∪
H 1 (GT (K k̄), ZZ/`ZZ) × H 0 (GT (K k̄), ` I) H 1 (GT (K k̄), ` I).
Since the lower pairing is non-degenerate, it follows that the upper pairing
is non-degenerate from the left. But µ` is a trivial G∅ (K k̄)-module and
H 1 (G∅ (K k̄), µ` ) = H 1 (G∅ (K k̄), ZZ/`ZZ)(1) is finite. Thus the upper pairing is
a perfect pairing of finite dimensional IF` -vector spaces. As explained above,
this shows b), and statement (ii) of the corollary follows from (3.7.4) and from
the Serre criterion (3.4.5).
Finally, observe that G∅ (K k̄) is either trivial (if g = 0) or a Demuškin group
of rank 2g with torsion-free abelianization. Therefore the remaining statement
(ii) of the theorem follows from theorem (3.9.11) and from the group theoretical
lemma (3.9.20) applied to the exact sequence (∗) above. 2
Remark: Using étale cohomology, a straightforward and natural way to prove the above
results about GS in the function field case is the following: if S is empty and g = 0, then
GS (K k̄) = 0 since there are no connected étale covers of the projective line. In all other cases,
show that for all i > 0 and every prime number ` =/ p = char(K)
i
lim Het (Y, ZZ/`ZZ) = 0,
−→
where Y runs through the connected étale covers of X r S. Conclude that
H i (GS (K k̄), ZZ/`ZZ) ∼
= H i (X r S, ZZ/`ZZ)
et k̄
for all i ≥ 0 and all ` =/ p. Then extend this isomorphism to arbitrary prime-to-p torsion GS -
modules resp. locally constant sheaves on Xk̄ r S. Finally, calculate the étale cohomology
groups on the right and apply the étale Poincaré duality theorem.
(10.1.4) Corollary. Let S be a nonempty (possibly infinite) set of places of

K and let c be a full class of finite groups. Then, for every discrete torsion
GS (K)(c)-module A and every i ≥ 0, the inflation map
H i (GS (K)(c), A) −→ H i (GS (K), A)(c)
is an isomorphism.
Proof: We may assume that A is `-torsion, where ` is a prime number

such that ZZ/`ZZ ∈ c. By (10.1.2)(c) (if ` =/ p) and by (8.3.3) (if ` = p),
we know that cd` G(KS |KS (c)k̄) ≤ 1. As the index of G(KS |KS (c)k̄) in
G(KS |KS (c)) is prime to `, we obtain cd` G(KS |KS (c)) ≤ 1. Noting that
H 1 (G(KS |KS (c)), A) = 0, the Hochschild-Serre spectral sequence
H i (GS (K)(c), H j (G(KS |KS (c)), A)) ⇒ H i+j (GS (K), A)
implies the result. 2

Unfortunately, using the methods above, we can only determine the structure
of pro-`-factor groups of GS . In fact, it seems that algebraic methods are not
very effective in determining the structure (in terms of generators and relations)
of profinite groups. One can show that a profinite group of countable rank is
free by solving embedding problems (see (3.5.20)). Sometimes global Galois
groups can be shown to be the free product of local Galois groups. And, as
the most far reaching result in this direction, one can determine the structure
of the absolute Galois group GKp of a local field Kp (cf. VII §5). This is rather
deep, and the proof extensively exploits the fact that GKp contains a normal
subgroup, the ramification group, which is a pro-p-group. In other words GKp
is “not too profinite”, which makes it accessible to algebraic methods.
But there seems to be no idea how one should determine the structure of
GS using algebraic methods. In the function field case, however, one can
obtain very deep results by exploiting the connections with topology. This is
possible by using the specialization map, defined by A. GROTHENDIECK. With
this method, one can show that the quotient GStame (K k̄) (see below) of GS (K k̄)
is topologically finitely generated and one can determine the maximal prime-
to-p-factor group of GS (K k̄) by relating it to the (well-known) topological
fundamental group of a Riemann surface. We will briefly explain this beautiful
and strong method below; however, to number theorists’ sorrow, there seems
to be no way to exploit similar techniques in the number field case.
In order to explain Grothendieck’s approach, let us change the notation for
a moment. Assume that k is an algebraically closed field of char(k) = p > 0.
Assume that X is a smooth, proper curve of genus g over k and let {P1 , . . . , Pn }
be a finite (possibly empty if n = 0) set of points in X(k). The points in
X(k) correspond to primes (valuations) of the function field k(X), and the
étale fundamental group π1et (X r{P1 , . . . , Pn }) (we omit the base point) is
isomorphic to the Galois group of the maximal extension of k(X) unramified
outside the primes which are associated to P1 , . . . , Pn . This group is not finitely
generated for n ≥ 1, because there are many covers of X r{P1 , . . . , Pn } with
wild ramification along P1 , . . . , Pn .
Let us consider the tame fundamental group π1tame (X r{P1 , . . . , Pn }),
which classifies étale covers with at most tame ramification along the divisor
P1 + · · · + Pn . It is isomorphic to the quotient of π1et (X r{P1 , . . . , Pn }) by the
normal subgroup generated by the ramification groups of P1 , . . . , Pn . Galois
covers of prime-to-p degree are tamely ramified, so we have an isomorphism
π et (X r{P1 , . . . , Pn })(p ) ∼
0 0
1 = π tame (X r{P1 , . . . , Pn })(p )
1
of the maximal pro-(p0 )-factor groups.
Now let A be a complete discrete valuation ring with residue field k and quo-
tient field K of characteristic zero (e.g., the ring of Witt vectors over k). Then

one can show that there exists a “lift to characteristic 0” of (X, {P1 , . . . , Pn }),
i.e. a connected, smooth and proper scheme X defined over Spec(A), and sec-
tions si : Spec(A) → X, i = 1, . . . , n, such that X = Xk can be identified with
the special fibre Xk of X in such a way that the sections s1 , . . . , sn specialize
to the points P1 , . . . , Pn . It may be seen as a consequence of Hensel’s lemma
that (omitting the suitably chosen base points) the canonical homomorphism
ϕ : π1tame (Xk r{P1 , . . . , Pn }) −→ π1tame (X r{s1 , . . . , sn }),
which is induced by the inclusion of the special fibre, is an isomorphism.
Let XK be the generic fibre of X and denote the specializations of s1 , . . . , sn
to XK by P1 , . . . , Pn (no confusion should arise by using the same letters as
for the points in Xk ). Let K be an algebraic closure of K. Then the inclusion
of the generic fibre composed with the inverse of ϕ induces a homomorphism
of profinite groups
sp : π1et (XK r{P1 , . . . , Pn }) −→ π1tame (Xk r{P1 , . . . , Pn }),
which is called the specialization map. For this map we have the
(10.1.5) Theorem (GROTHENDIECK). The specialization map sp is surjective.

It defines an isomorphism
0 0 ∼ 0
sp(p ) : π1et (XK r{P1 , . . . , Pn })(p ) −→ π1et (Xk r{P1 , . . . , Pn })(p )
on the maximal pro-(p0 )-factor groups.
For a proof of this theorem and of the facts mentioned before we refer the
reader to [67], exp. XIII.
Having “lifted” our problem, there remains the easier problem of determining
the structure of the algebraic fundamental group of a smooth curve over an
algebraically closed field K of characteristic 0. By a general principle we may
assume that K = C, so that Y = XK r{P1 , . . . , Pn } is an algebraic Riemann
surface. Every topological covering of Y (C) can be uniquely endowed with
a holomorphic structure and, by the famous Riemann existence theorem,
these coverings are in fact algebraic curves. Since topological coverings are
classified by the subgroups of the topological (i.e. path-) fundamental group,
this implies that there is a canonical isomorphism
co : π top (Y (C))∧ ∼
1 = π et (Y )
1
from the profinite completion of the topological fundamental group to the

algebraic fundamental group of Y .
Now the structure of the topological fundamental group is well-known.
Denoting the genus of X by g, π1top (X r{P1 , . . . , Pn }) is the group

Πg,n = hx1 , . . . , x2g , y1 , . . . , yn | (x1 , x2 ) · · · (x2g−1 , x2g ) y1 · · · yn = 1i,
where x1 , . . . , x2g generate the fundamental group of the complete curve X and
y1 , . . . , yn are loops around the removed points P1 , . . . , Pn ∈ X. In particular,
co maps the loop yi to a generator of the procyclic inertia group of the prime
associated to Pi ∗) for i = 1, . . . , n.
Returning to our original notation, we deduce the following theorem, which

generalizes the statement of (10.1.2) (ii) from the maximal pro-`-quotients
(` ∈ (p0 )) to the full maximal pro-(p0 )-factor group. Let GStame (K) denote
the Galois group of the maximal extension of K in KS with at most tame
ramification at the primes in S.
(10.1.6) Theorem. The group GStame (K k̄) is topologically finitely generated.

0
The group GS (K k̄)(p ) has as a pro-(p0 )-group a presentation by 2g+n generators
and one relation of the form
{x1 , . . . , x2g , y1 , . . . , yn | (x1 , x2 ) · · · (x2g−1 , x2g ) y1 · · · yn = 1}.
0
In particular, if S =/ ∅, then GS (K k̄)(p ) is a free pro-(p0 )-group of rank
2g + n − 1. Furthermore, the elements y1 , . . . , yn may be chosen as generators
0
of the procyclic inertia groups of prolongations of primes in S(K k̄) to KS(p ) .
(10.1.7) Corollary. The group GStame (K) is topologically finitely generated.
In the topological situation, consider the (topological) universal cover X̃

of X(C). The set of points lying over P1 , . . . , Pn is a discrete subset in X̃.
Therefore the normal subgroup in Πg,n generated by y1 , . . . , yn is the free
(discrete) group over the set of loops around the pre-images of P1 , . . . , Pn in
X̃. In order to formulate an algebraic version of this result, we have to make
some preparations.
Let K be a global (number or function) field and let M |K be a (possibly
infinite) Galois extension of K. We denote the one-point compactification of
the discrete set of places of K by Sp(K). The compactifying point, which
will be denoted by ηK , should be thought as the generic point of Sp(K) in the
sense of algebraic geometry or as the trivial valuation from the valuative point
of view. We set
Sp(M ) := lim
←−
Sp(L),
L⊆M
∗) In fact the inertia group is associated to a prolongation of P which is determined by the

i
choice of base points that we omitted from our notation.

where the limit is taken over all finite extensions L of K inside M . In particular,
Sp(M ) is a profinite space.
As is the case for Sp(K), the set Sp(M ) consists of the places of M plus one
generic point ηM . Topologically, however, Sp(M ) might not be the one-point
compactification of Sp(M ) r{ηM }. Its topology reflects the fact that M is an
inductive limit of global fields.
One can also give an intrinsic definition of the topology of Sp(M ). In order to do this, we
remind the reader of the definition of the constructible topology (see [70], chap.I, §7, (7.2.11)).
For a (commutative) ring A, the subsets f ∗ (Spec(B)) ⊂ Spec(A), where f : A → B is a
ring homomorphism, satisfy the axioms for closed sets in a topological space. The associated
topology is the constructible topology. Spec(A) with the constructible topology is a Hausdorff,
compact and totally disconnected topological space, i.e. a profinite space (see I §1). This
definition extends to arbitrary schemes by gluing.
In the function field case, consider the integral closure XM of the curve XK in M . This
one-dimensional scheme can be constructed from the field M in the same way as XK from K;
in particular, it depends only on M but not on K. Then one verifies that there is a canonical
isomorphism (of sets)
Sp(M ) ∼
= XM ,
which becomes a homeomorphism if we endow the (not necessarily noetherian) scheme XM
with the constructible topology.
If we exclude the archimedean primes, a similar statement is true in the number field case:
Sp(M ) r S∞ (M ) ∼ = Spec(OM )constr.top. .
If S is a finite set of primes in K, then the set S(M ) ⊆ Sp(M ) of places

in M lying over S is closed in the constructible topology. If S consists of
a single prime p ∈ Sp(K), then the set S(M ) is topologically isomorphic to
the compact set of coset classes G(M |K)/GP (M |K), where GP (M |K) is
the decomposition group in M |K of an arbitrary chosen prolongation P of p
to M .
Assume that we are given a closed subset S ⊆ Sp(M ). Then the set of inertia
groups ( TP (M |K) )P∈S is a continuous family of subgroups∗) of G(M |K).
This follows from the fact that only finitely many primes ramify in a finite
separable extension of global fields.
(10.1.8) Definition. The free product over the bundle of profinite groups
associated to the continuous family ( TP (M |K) )P∈S by (4.3.3), is called the
free product of the inertia groups in S.
∗) See IV §3. We set T (M |K) = {1} here, which is justified because separable extensions
η
of K are unramified at η in the sense of algebraic geometry.

There exists a canonical continuous homomorphism

∗T
P∈S
P (M |K) −→ G(M |K).
If ηM ∈/ S, then the restriction of S to every finite extension of K inside M

is finite. In this case the family ( GP (M |K) )P∈S of decomposition groups is
also continuous and we can form the free product of decomposition groups in
the same way.
Suppose that we are given an intermediate field L ⊆ M and a closed subset
S ⊆ Sp(L).
(10.1.9) Definition. We say that G(M |L) is the free product of the inertia
groups of the primes in S if there exists a continuous section s : S → Sp(M )
over S to the projection π : Sp(M ) → Sp(L) such that the canonical homo-
morphism
∗ TP (M |L) −→ G(M |L)
P∈s(S)
is an isomorphism. In this case, we write (by abuse of notation)

∗ T (M |L) ∼= G(M |L) ,
p∈ S
p
and we will omit the generic point, if it is contained in S, from the notation.
Remark: There always exist (many) sections s : S → Sp(M ); see [74],

lemma 8.1 or [135], lemma 4.7, and also ex.5 below. However, it is clear from
the discussion in chapter IV §2 that we cannot expect the homomorphism
∗
P∈s(S)
TP (M |L) → G(M |L)
to be an isomorphism for every choice of a section s : S → Sp(M ), unless we

pass to the maximal pro-p-factor groups.
We denote the inertia group of G(L̄p |Lp ) by Tp . If, in the above situation,
the canonical surjection Tp Tp (M |L) is an isomorphism for every p ∈ S,
then we write
Tp ∼∗
= G(M |L)
p∈S
and we say that G(M |L) is the free product of full local inertia groups.
If M |K is a c-extension for a full class of finite groups c, then the above
concepts generalize to pro-c-products in a straightforward manner.
Now we are in position to formulate the algebraic analogue of Riemann’s
existence theorem. For finite sets S it follows from the topological results
above, and we can pass to infinite S by (4.3.11).

(10.1.10) Riemann’s Existence Theorem (Algebraic Form). Assume that K

is a global function field of char(K) = p > 0 with finite constant field k. Let
S be any nonempty set of places of K and assume that g(K) > 0. Then there
exists a canonical exact sequence of pro-(p0 )-groups
1→ ∗
p∈S(K∅ (p0 ))
0 0 0
Tp(p ) → GS (K k̄)(p ) → G∅ (K k̄)(p ) → 1,
∗
where denotes the free pro-(p0 )-product and Tp(p ) ∼
0 0
= ẐZ(1)(p ) is the full prime-
to-p part of the local inertia group at p. If g = 0 (so that G∅ (K k̄) = {1}), we
∗
p∈S(K k̄)
0
Tp(p ) −→
0
∼ G (K k̄)(p ) ,
S
p/
= p0
where p0 is an arbitrarily chosen prime in S(K k̄).
Remark: It is not difficult to deduce a pro-` version of Riemann’s existence

theorem in a purely algebraic way using similar arguments to the proof of
theorem (10.1.2).
This section would be incomplete without mentioning Frobenius weights.

Choose an ` ∈ (p0 ) and consider (if g > 0) the Q` -vector spaces
IHi := Hi (G∅ (K k̄), ZZ` ) ⊗ ZZ` Q`
for i = 0, 1, 2. Trivially IH0 = Q` and by (10.1.2) we have IH2 ∼
= Q` (1) and
IH1 ∼
= (lim H 1 (G∅ (K k̄), µ`n )) ⊗ ZZ` Q` ∼
= T` (JacX) ⊗ ZZ` Q` ,
←−
n
where T` (JacX) is the `-adic Tate module of the Jacobian variety associated
to X.
The general theory of abelian varieties over finite fields (see [140]) shows
that the characteristic polynomial of the Frobenius automorphism acting on
T` (JacX) is contained in ZZ[T ] (rather than in ZZ` [T ]) and is independent
of ` ∈ (p0 ). The Frobenius eigenvalues are therefore algebraic integers and
(loc.cit.) they have absolute value q 1/2 in every complex embedding, where
q = pf = #k. We conclude that the Frobenius eigenvalues on IHi have absolute
value q i/2 for i = 0, 1, 2. This result can be reformulated into a statement about
zeros and poles of the zeta function Z(s, Xk ) of Xk (cf. XI §6). In particular,
Re(s) = 1/2 for every zero of Z(s, Xk ). This result is therefore called the
1-dimensional Riemann hypothesis in positive characteristic.
Now consider the exact sequence
0 → H :→ H1 (GS (K k̄), ZZ` ) ⊗ ZZ` Q` → IH1 → 0.

The kernel H is generated by the images of H1 (Tp , ZZ` ) ∼ = ZZ` (1) for the
places p ∈ S. Therefore the above exact sequence provides a filtration (weight
filtration) of H1 (GS (K k̄), ZZ` ) ⊗ ZZ` Q` with respect to the absolute values (the
weights) of the Frobenius eigenvalues.
The generalization of the above results to varieties of higher dimensions,

the “Weil conjecture”, proved by P. DELIGNE [35], [36], is one of the major
achievements of modern algebraic geometry. The use of étale cohomology is
essential here and in fact étale cohomology was introduced with this application
in mind.
We finish this section with the treatment of the cohomological properties

of GS with respect to p = char(K). The following theorem is a somewhat
weakened formulation of a result of M. RAYNAUD (see [176]). Recall the
definition of the p-rank h = h(X) from the remark after (10.1.1).
(10.1.11) Theorem. Assume that X is a smooth, projective curve of genus

greater than or equal to 2 over an algebraically closed field of characteristic
p > 0. Then there exist cyclic covers of degree prime to p with arbitrary large
p-rank h.
For a proof see [176], théorème 4.3.1.
(10.1.12) Theorem. If K is a global function field of char(K) = p > 0, then

scdp GS = 2. Furthermore, the following hold:
(i) If g = 0, then G∅ (K) ∼
= G(k̄|k) ∼
= ẐZ.
(ii) If g = 1, then G∅ (K k̄) is abelian and we distinguish the following two
cases:
(a) If h = 0, then G∅ (K k̄)(p) = 0; in particular, cdp G∅ (K) = 1.
(b) If h = 1, then G∅ (K k̄)(p) ∼= ZZp and the group G∅ (K) is a
Poincaré group at p of dimension 2.
(iii) If g ≥ 2, then cdp G∅ (K k̄) = 1. The group G∅ (K) is a duality group at
p of dimension 2 with dualizing module I = Cl(K∅ )(p).
(iv) If S =/ ∅, then
cdp GS (K k̄) = cdp GS (K) = 1.

Proof: If S is nonempty, then scdp GS = 2 by (8.3.17). We will slightly

modify the proof of (8.3.17) for the case S = ∅. As in chapter VIII, we use
the notation
Y
UK := Up ,
p
where p runs through all primes of K and Up is the unit group of the completion
of K at p. Then global class field theory gives the exact sequence
0 → k × → UK → C(K) → Cl(K) → 0.
Since finite fields are perfect, k × is uniquely p-divisible. Further, recall that
UL is a cohomologically trivial G(L|K)-module if L|K is unramified. Thus
we obtain
Cl(L)G(L|K) = Cl(K),
H i (G(L|K), C(L)) ∼ = H i (G(L|K), Cl(L))
for i ≥ 1 and every unramified p-extension L|K. This shows that the pair
(G(K∅ (p)|K), Cl(K∅ (p)) is a class formation. Furthermore, we have an
isomorphism
G∅ (p)ab ∼
= Cl(K) ⊗ ZZ ZZp
by global class field theory. Therefore we obtain that scdp G∅ (K)(p) ≤ 2 by

(3.6.4). The same argument applies to every finite extension of K inside K∅ .
By a limit process, we conclude that a (every) p-Sylow subgroup of G∅ (K)
has scdp ≤ 2, which shows the assertion.
Statement (i) is a repetition of (10.1.3)(i). Now assume that g = 1. Fix any
point O in X(k̄) in order to make it into an elliptic curve E = (Xk̄ , O). Then it
is well-known (see [140]) that every étale covering of E is an isogeny. Hence
π1 (E) is abelian and (loc.cit.) isomorphic to
lim n Cl(X).
←−
n
Hence G∅ (K k̄)(p) ∼= 0 or G∅ (K k̄)(p) ∼= ZZp depending on the p-rank h. The

remaining statement in (b) then follows from (3.7.4).
In order to prove assertion (iii), we set G = G∅ (K). We investigate the
terms
Di (ZZ) = lim
−→
H i (U, ZZ)∗ ,
where U runs through the open subgroups in G, cf. III §4. We will show that
D0 (ZZ) = 0 = D1 (ZZ) and that D2 (ZZ) is p-divisible. By (3.4.8) this shows that
G is a duality group of dimension 2 at p with p-dualizing module D2 (ZZ)(p).
First we observe that D0 (ZZ) = 0, since K k̄ ⊆ K∅ and hence every prime
number divides #G infinitely often. Further, D1 (ZZ) = 0 for trivial reasons.

Finally, we show that D2 (ZZ) is p-divisible. Recalling that H 2 (U, ZZ) ∼

=
(U ab )∗ , global class field theory induces the exact sequence
0 → Cl(KU ) → H 2 (U, ZZ)∗ → ẐZ/ZZ → 0,
where KU denotes the subextension in K∅ which is associated to U . Passing
to the limit over all open subgroups U ⊆ G, we obtain
D2 (ZZ)(p) ∼
= Cl(K∅ )(p) and D2 (ZZ)/p ∼
= Cl(K∅ )/p.
From the observations at the beginning of the proof, we know that
Cl(KU )(p) = Cl(K∅ )(p)U
for every open subgroup U G. By (10.1.1), we have
⊆
Cl(K∅ )/p ∼

= Cl(K∅ )/Cl0 (K∅ ) /p.
Now let L be a finite extension of K in K∅ . Then we have the commutative
diagram
0õö÷øùúûüýþÿ Cl0 (Lk̄) Cl(Lk̄) ZZ 0
[Lk̄:K k̄]
0 Cl0 (K k̄) Cl(K k̄) ZZ 0.

By (10.1.11), we know that the order of G∅ (K k̄) is divisible by p∞ . Hence
D2 (ZZ)/p = 0. It remains to show cdp G∅ (K k̄) ≤ 1. We have
H 2 (G∅ (K k̄), ZZ/pZZ) = lim
−→
H 2 (G∅ (Kkn ), ZZ/pZZ)
n
= (lim p G∅ (K k̄)ab )∗
←−
n
= (lim p Cl(Kkn ))∗ .
←−
n
But p Cl(Kkn ) = p Cl(Kkm ) = p Cl(K k̄) for m ≥ n >> 0, so that the last
projective limit vanishes (observe that the transition maps are induced by the
norm). Since the same argument applies to every finite separable extension of
K, this shows (iii).
Now assume S =/ ∅. Then H i (GS , ZZ/pZZ) = 0 for i ≥ 2 by (8.3.2) and since
the same is true for every open subgroup, we see that
cdp GS (K k̄) ≤ cdp GS (K) ≤ 1.
It remains to show that GS (K k̄) has a nontrivial p-Sylow group. This may be
achieved by a direct computation of
H 1 (GS (K k̄), ZZ/pZZ) = OS (K k̄)/℘,
where ℘ is the map x 7→ xp − x. But we will give a slightly different argument.
By the result of ex.1 below, we know that there are many étale covers of degree
p of the affine line A1k̄ . Now choose any point p0 ∈ S(K k̄) and construct a

morphism f : Xk̄ → IP1k̄ such that f −1 (∞) = p0 (such an f exists by the

Riemann-Roch Theorem). Then we obtain many unramified covers of Xk̄ r S
by base change from A1k̄ . 2
Exercise 1. Let k be an algebraically closed field of characteristic p > 0. Consider the

coverings Ym of A1k = Spec(k[X]) which are given by the Artin-Schreier equations Y p − Y =
X m for m ∈ IN. Show that these coverings are cyclic and unramified (i.e. they are ramified
only over the infinite point). Assume that m = m0 pe with (m0 , p) = 1. Show that
1
g(Ym ) = (m0 − 1)(p − 1).
2
Exercise 2. Let K be a global field of char(K) = p > 0 and let ` =/ p be a prime number.
Show that


 a Poincaré group of dimension 3 if Cl(K)(`) =/ 0 and µ` ⊂ K,

a duality group of dimension 2 with

G∅ (K)(`) ∼
=

 dualizing module Cl(K∅ (`))(`) if Cl(K)(`) =/ 0 and µ` 6⊂ K,

ZZ` if Cl(K)(`) = 0.

Exercise 3. Let K be a global field of char(K) = p > 0. Show that


 a duality group of dimension 2 with
G∅ (K)(p) ∼
= dualizing module Cl(K∅ (p))(p) if Cl(K)(p) =/ 0,

ZZp if Cl(K)(p) = 0.
Exercise 4. Let K be a global field of char(K) = p > 0 and let S be a set of places of K.
Show that the maximal pro-p-factor group of GS (K) is finitely generated if and only if S = ∅.
Exercise 5. Let K be a global field and M |K be a Galois extension. Let L ⊆ M be an

intermediate field which is Galois over K. Show that there exists a continuous section
s : Sp(L) → Sp(M )
to the canonical projection π : Sp(M ) → Sp(L).
Hint: First construct the section over the sets Sp (L) of primes in L which lie over a fixed
prime p of K. Observe that, fixing a prolongation P of p to M , there are isomorphisms
Sp (M ) ∼= G(M |K)/GP (M |K),
Sp (L) ∼ = G(L|K)/GP∩L (L|K) ∼ = G(M |K)/G(M |L)GP (M |K).
Then use the result of ex.4 in I §1.
Exercise 6. Let K be a global field of char(K) = p > 0. Let T ⊆ S be sets of primes of K

where T is finite and S r T =/ ∅, and let KST (p) be the maximal p-extension of K inside KS
in which all primes of T are completely decomposed. Show that the canonical homomorphism
∗ G(KP (p)|KP ) −→ G(KS (p)|KST (p))

P∈T (KST (p))
is an isomorphism.
Hint: Both groups are free pro-p-groups. Use (9.2.5) and (1.6.15).

§2. First Observations on the Number Field Case
In this section we start our investigations into the number field case. Let us
fix some notation. Suppose we are given
K a number field,
S a set of places of K,
c a full class of finite groups,
KS (c) the maximal c-extension of K which is unramified outside
the primes in S,
GS (K)(c) = G(KS (c)|K).
We will omit c if it is the class of all finite groups and we will write GS instead
of GS (K) if K is clear from the context.
In contrast to chapter IX, where our main interest was devoted to the case
S = {all places of K} (i.e. GS = GK ), in the present chapter we are mainly
interested in the case when S is finite.
Let us first assume that c is the class of all finite groups and that S ⊇ S∞ .
×
Then, recalling the notation IN(S) = IN ∩ OK,S of VIII §3, we know from
(8.3.18), (8.3.20) that
• cdp GS ≤ 2 if p ∈ IN(S), provided that SIR (K) = ∅ if p = 2,

• if S is finite and if A is a finite GS -module such that #A ∈ IN(S), then
H i (GS , A) is finite for i = 0, 1, 2.
Several questions naturally arise.

(1) How big is the decomposition group of a prime p of K in GS ? Do we
attain the theoretical maximum (KS )p = (Kp ) (resp. (KS )p = Kpnr ) for
p ∈ S (resp. p ∈/ S) ?
(2) Is cdp GS finite for p ∈/ IN(S) ?
(3) Is cdp GS equal to 1 or equal to 2 for p ∈ IN(S) ?
(4) Is scdp GS equal to 2 for p ∈ IN(S) ?
(5) If S is finite, is H i (GS , A) finite for i ≥ 0, if also #A ∈/ IN(S) ?
(6) For which S is the group GS topologically finitely generated?
1. If S contains all but finitely many primes, the results of IX §4 imply that
the maximal possible local extensions are indeed realized by the global field
KS . But if S is smaller, for instance if S is finite, the situation is much
more complicated. In general GS might be finite or even trivial (for instance
for K = Q and S = S∞ ), and so we find examples of primes with trivial
decomposition group in GS . But we will prove the following (see §§8,11):

§2. First Observations on the Number Field Case 619
If p is a prime number and if S ⊇ Sp ∪ S∞ , then the local field (KS )p

is closed under p-extensions (closed under unramified p-extensions) if p is
contained (not contained) in S.
2. Question (2) is also rather difficult. Even the much weaker question of
determining the set of prime numbers p ∈/ IN(S) which divide the order of GS
is hard to answer. We will see in §11 that if S ⊇ S` for at least one prime
number `, then there exist infinitely many different prime numbers p dividing
the order of GS . But the p-Sylow subgroup for such a prime number could
also be finite. There exist examples with cd GS (p) = 2 for a prime number
p ∈/ IN(S), see [119], [198].
3. The answer to the third question is that cdp GS = 2 always for p ∈ IN(S),
provided that SIR (K) = ∅ if p = 2. However, it might happen that the maximal
pro-p-factor group GS (p) is of cohomological p-dimension equal to 1. We will
deduce a criterion for this to occur in §9. In §11 we will show that subgroups U
with cdp U (p) = 2 are cofinal among the open subgroups of GS , which shows
that always cdp GS = 2.
In most cases however, we can read off the cohomological p-dimension
directly from the module I = D2 (ZZp ). Assume that K is totally imaginary if
p = 2 and S ⊇ Sp ∪ S∞ . Since cdp GS ≤ 2, two alternatives exist: either I
is nontrivial, then cdp GS = 2 and I is the p-dualizing module of GS , or I is
trivial, in which case cdp GS = 1. Let us denote the subset of finite primes
in S by S f . From the calculation below we can see that I is nontrivial (and
hence cdp GS = 2) if there is more than one prime in S f (KS ). This covers
most cases, for example if #S f (K) > 1. However, if S = Sp ∪ S∞ and if there
is exactly one prime dividing p in K, then we cannot easily decide whether
there exists an extension of K inside KS splitting this prime.
(10.2.1) Proposition. Let K be a number field, let p be a prime number and

assume that S ⊇ Sp ∪ S∞ is a set of primes of K. Then the GS -module
I = D2 (ZZp ) is characterized by the exact sequence
Y
0 −→ µp∞ (KS ) −→ lim µpn (Lp ) −→ I −→ 0,
−→
L,n p∈S f (L)
where L runs through the finite extensions of K inside KS . In particular, if S

is finite, then there exists an exact sequence
M G
0 −→ µp∞ −→ IndGpS µp∞ −→ I −→ 0,
p∈S f (K)
where Gp is the decomposition group of a prolongation of p to KS .

Proof: Recall (III §4) that

I = lim
−→
H 2 (U, ZZ/pn ZZ)∗ ,
n,U
where U runs through the open subgroups of GS and the transition maps with
respect to the subgroups are the duals of the corestriction maps. The duality
theorem of Poitou-Tate yields
lim X2 (U, ZZ/pn ZZ)∗ = lim X1 (U, µpn ) = 0,
−→ −→
n,U n,U
since the transition maps with respect to U on the right-hand side are the
restriction maps. Therefore the result follows from the second part of the
Poitou-Tate sequence for the module ZZ/pn ZZ, by passing to the limit over all
open subgroups U ⊆ GS . 2
We can also interpret the last result (10.2.1) in the following way. The
p-dualizing module I of GS is the quotient of the p-torsion subgroup of CS , by
the subgroup of those classes which are represented by an idèle with support
in the archimedean places. Recall that the S-idèle class group CS (K) was
defined as the quotient of CK by the subgroup UK,S = P∈S {1} × P∈/S UP .
Q Q
×
Using the convention that UP = KP for an archimedean prime P, we define
Y Y
UK,S f = {1} × UP .
P∈S f /S f
P∈
If S is strictly larger than S∞ , which we assume here, we may regard UK,S f as

a subgroup of CK . We set
CS f (K) = CK /UK,S f
and we call it the S f -idèle class group of K. It is not difficult to see that we
have an exact sequence
× ×
Y
0 → OK,S → KP → CS f (K) → Cl0 (K) → 0,
P∈S f
where Cl0 (K) is the ideal class group in the narrow sense of K, i.e. the
quotient of the ideal group by the subgroup of those principal ideals which are
generated by a nontrivial totally positive element∗) . If K is totally imaginary,
then Cl0 (K) = Cl(K) and the pair (GS , CS f (KS )) is a class formation which
is a slight modification of the usual formation (GS , CS (KS )). The difference is
that it has a smaller group of universal norms, because the infinite idèles have
been divided out. However, we will not use this fact in the following. The next
corollary is a reformulation of (10.2.1) using the terminology just introduced.
∗) We call an element totally positive if its image in every real embedding of K is positive;
in particular, if K is totally imaginary, then every element is totally positive.

(10.2.2) Corollary. Assume that S ⊇ Sp ∪ S∞ . Then the p-dualizing module

I = D2 (ZZp ) of GS is canonically isomorphic to the p-torsion subgroup of the
S f -idèle class group of KS .
4. We clearly have to assume that K is totally imaginary if p = 2. If the set S

is finite, then we will see in the next section that for p ∈ IN(S) the vanishing
of H 2 (GS , Qp /ZZp ) is equivalent to the Leopoldt conjecture. Although this
conjecture is verified only in special cases (see §3), we strongly believe that
the answer to question (4) “must” always be yes, but we cannot prove this for
any number field at all.
However, if the set S is of density 1, we have the following positive answer
to question (4).
(10.2.3) Theorem. Let K be a number field. Suppose that the set of primes S
contains S∞ and has Dirichlet density 1. Then
scdp GS = 2
for every p ∈ IN(S), provided that K is totally imaginary if p = 2.
Proof: Since cdp GS ≤ 2 by (8.3.18), it suffices to show that the cohomology

group H 2 (U, Qp /ZZp ) vanishes for every p ∈ IN(S) and every open subgroup
U ⊆ GS (see (3.3.4)). Since the assumptions carry over to every finite extension
of K inside KS , we can restrict to the case U = GS . The set S has Dirichlet
density 1 and therefore we obtain (using the notation of IX §1)
cs(K(µpr )|K) ⊂
∼ S
for every prime number p and every r ∈ IN. Using (9.1.10) we obtain injections
H 2 (GS , ZZ/pr ZZ) ,→ H 2 (Kp , ZZ/pr ZZ)
M
p∈S
for every r and every p ∈ IN(S), provided we are not in the special case. Then,
passing to the limit over r and observing that local fields have scd = 2, cf.
(7.2.5), we deduce the result. √
If we are in the special case, cf. (9.1.9), then p = 2 and i = −1 ∈/ K. In
particular, we are not in the special situation if we replace K by K(i), and so
we obtain by the above argument that H 2 (GS (K(i)), Q2 /ZZ2 ) = 0. By (3.3.11)
the corestriction map
0 = H 2 (GS (K(i)), Q2 /ZZ2 ) −→ H 2 (GS (K), Q2 /ZZ2 )
is surjective. This finishes the proof. 2

5. The answer to question (5) is “yes” for i ≤ 2, by the more general theorem
below. When c is the class of p-groups, it is due to H. KOCH (see [110]), whose
original proof did not use the duality theorem of Poitou-Tate.
(10.2.4) Theorem. Let c be a full class of finite groups and let S be a finite set
of primes of the number field K. Then the cohomology groups
H i (GS (c), M )
are finite for i = 0, 1, 2 and every finite GS (c)-module M .
Proof: Passing to a trivializing open subgroup of GS (c) and using the

Hochschild-Serre spectral sequence, we may assume that M is a trivial module
and we immediately reduce to the case M = ZZ/pZZ for some prime number p.
Next we see that the theorem is true for S if it is true for a finite set of primes
T ⊇ S. This follows from the Hochschild-Serre spectral sequence associated
to the group extension
1 → G(KT (c)|KS (c)) → GT (c) → GS (c) → 1
and the fact that G(KT (c)|KS (c)), being a normal subgroup in GT (c), is gener-
ated by the finitely many inertia groups Tp (KT (c)|KS (c)), p ∈ T r S (choose
prolongations to KT (c)).
We denote the full local group G(K p |Kp ) by Gp and its inertia group by Tp .
Then there are surjections
Gp Gp (KT (c)|K) and Tp Tp (KT (c)|KS (c)).
We obtain an exact sequence
0 → H 1 (GS (c), M ) → H 1 (GT (c), M ) → H 1 (G(KT (c)|KS (c)), M )GS (c) →
→ H 2 (GS (c), M ) → H 2 (GT (c), M )

and canonical injections
H 1 (Tp (KT (c)|KS (c)), M )Gp (KT (c)|K)
M
H 1 (G(KT (c)|KS (c)), M )GS (c) ,→
p∈T \S
H 1 (Tp , M )Gp .
M
,→
p∈T \S
The last group is finite by the results of chapter VII, which shows that the
middle term in the five term exact sequence above is finite. Hence we may
enlarge S; in particular, we may assume that S ⊇ Sp ∪ S∞ . If c is the class
of all finite groups, we are done by the result of (8.3.20). If c is a class which
does not contain ZZ/pZZ, then the cohomology groups in question are trivial
for i ≥ 1. Finally, if c is not the class of all finite groups and ZZ/pZZ ∈ c, then
the result follows from (10.4.8) (and we do not use the result for such c before
then). 2

6. If S is the set of all primes, then GS is not finitely generated (see chapter
IX). Surprisingly, the answer to question (6) is unknown for finite S. It is not
difficult to see that Gab
S is finitely generated if S is finite, because, on the one
hand, the decomposition groups for the finitely many primes in S are finitely
generated, and, on the other hand, Gab ∅ is of finite order. This, however, only
implies that the maximal pro-p-factor group of GS is finitely generated for
every prime number p.
In contrast to many other questions about GS , where we “know” what the
right answer should be, but are not able to give a real proof, it is even not
clear what one should believe to be the right answer to question (6). Many
mathematicians (including the authors) tend to think of GS as being not finitely
generated. In order to actually prove such a statement, one should construct
many extensions which are unramified outside S. One method of constructing
such extensions is the adjunction of points of geometric objects (e.g. moduli
spaces) which have good reduction outside S. One of the major achievements
of the number theoretical research of the last decades is some insight into
the moduli of elliptic curves. (This was crucial for the proof of the Main
Conjecture of Iwasawa Theory by B. MAZUR and A. WILES, and also for the
proof of Fermat’s Last Theorem by A. WILES.) But it is not clear whether for
the purpose explained above it suffices to use only the moduli spaces of elliptic
curves. Since our knowledge of other moduli spaces is even smaller, it seems
to be very hard to achieve progress in this direction. However, we have the
following weaker result.
(10.2.5) Theorem. Assume that K is a number field and that S is a finite set
of primes in K. Then the group GS = GS (K) is (topologically) generated by
the conjugacy classes of finitely many elements.
In §11 we will deduce theorem (10.2.5) as a corollary of a result of Y. IHARA

on the decomposition of primes in infinite unramified extensions of number
fields. It can also be deduced from the fact that Gab
S is finitely generated and
from the following purely group theoretical result.
(10.2.6) Theorem (GURALNICK, WEISS). Let G be a profinite group with

abelianization of rank r ∗) . Then G can be topologically generated by r (or 1
if r = 0) conjugacy classes.
We omit the proof of (10.2.6), referring the reader to the exercises below.
∗) i.e. Gab can be topologically generated by r elements and r is minimal.

Exercise 1. Let G be a group. We define the Frattini subgroup Φ(G) of G as the intersection
of all maximal subgroups of G (see also IX §6). Show that Φ(G) is a normal subgroup of G
and that an element g ∈ G is contained in Φ(G) if and only if it is a “non-generator”, i.e. if it
can be removed from every generating family of elements of G.
Exercise 2. Assume that N / G is a minimal normal subgroup, i.e. generated by the conjugates
of one element n ∈ G. Suppose that N is not contained in Φ(G). Show that the canonical
projection
G −→ G/N
has a section, i.e. there exists a decomposition of G as a semi-direct product G = N o H for
some subgroup H ⊆ G.
Exercise 3. Let G be a finite group and let N / G be a minimal normal subgroup. Suppose
that the abelianizations of G and of G/N have the same rank r ≥ 1 and assume that G/N is
generated by the r conjugacy classes C 1 , . . . , C r . Show that these conjugacy classes can be
lifted to conjugacy classes C1 , . . . , Cr in G, such that
G = hC1 , . . . , Cr i.
Hint: Use ex.1 and 2 in order to reduce to the case that G = N × H and N is simple. Choose
elements c1 , . . . , cr ∈ H with ci mod N ∈ C i for i = 1, . . . , r and choose any nontrivial
element n ∈ N . Then the conjugacy classes of n · c1 and of c2 , . . . , cr generate G.
Exercise 4. Use induction on #G and the exercises above in order to prove theorem (10.2.6)
in the case that G is finite.
Exercise 5. Extend the result of ex.4 to an arbitrary profinite group G.
§3. Leopoldt’s Conjecture
It is one of the fundamental principles in number theory that none of the

places of a number field should be privileged above the others, i.e. they all play
comparable roles. In particular, this includes the archimedean places, but in
practice we are often confronted with the situation that statements which are
well-known in the archimedean case turn out to be difficult or even unsolved
questions in their p-adic formulations.
However, the truth of these assertions (maybe up to minor modifications)
is predicted by the analogy and many conjectures in number theory originate
from this source. One prominent example for this phenomenon is the p-adic
version of Minkowski theory, which will be described in this section. Several
arguments below are taken from P. SCHNEIDER’s thesis [202].
Let K be a number field and let r1 = #SIR (K), resp. r2 = #SC (K), be the
number of real, resp. complex, places of K. In this chapter we will also use
×
the notation EK for the group of units OK of K. Recall (cf. [160], chap.I,

§3. Leopoldt’s Conjecture 625
§§5,7) that (the multiplicative form of) Minkowski theory investigates the
homomorphism
Q
σ
log(| |)
j∞ : K × −→ C× −−−−−−−→
Y Y
IR
σ ∈Hom(K,C) σ ∈Hom(K,C)
which is given by j∞ (x)σ := log(|σ(x)|). Dirichlet’s unit theorem asserts that

j∞ (EK ) is a lattice of rank r1 + r2 − 1 in σ IR (and the IR-subspace spanned by
Q
j∞ (EK ) is the space of elements of trace 0 which are invariant under complex
conjugation).
Now we are going to investigate the p-adic analogue of the above situation.
Let p be a prime number, which will be fixed for the rest of this section. Fix an
algebraic closure Q̄p of the field Qp . The field Q̄p is not complete with respect
to the p-adic absolute value and we denote its completion by Cp .
(10.3.1) Definition. We call Cp the field of p-adic complex numbers.
The p-adic absolute value naturally extends to Cp and Q̄p is dense in Cp .
(10.3.2) Proposition. The field Cp is algebraically closed.
Proof: Let α ∈ C̄p and let f ∈ Cp [X] be the minimal polynomial of α over
Cp . Note that f is separable because char(Cp ) = 0. Since Q̄p is dense in Cp , we
can choose a polynomial g ∈ Q̄p [X] near to f . Then |g(α)| = |g(α) − f (α)| is
small. Writing g(X) = (X − βj ), with βj ∈ Q̄p , we see that |α − β| is small
Q
for some root β of g(X). In particular, we can choose g(X) and then β such
that |β − α| < |αi − α| for all conjugates αi =/ α of α over Cp . By Krasner’s
lemma (8.1.6), we obtain α ∈ Cp (β) = Cp . 2
Remark: The fields C and Cp are isomorphic as abstract fields, because

they are algebraically closed fields of the same transcendence degree over Q.
However, they are not topologically isomorphic.
Recall (cf. [160], chap.II, (5.4)) that for every local field k|Qp we have a
uniquely defined p-adic logarithm
logp : k × −→ k.
It satisfies logp (p) = 0 and for a principal unit (1 + x) ∈ Uk1 it is given by the

(convergent) series
x2 x3
logp (1 + x) = x −
+ − ···.
2 3
This p-adic logarithm naturally extends to Q̄p and, by continuity, also to a
function: logp : C×
p → Cp . For a given number field K consider the homo-
morphism Q
σ
logp ( )
jp : K × −→ C×
Y Y
p −−−−−
−−→ Cp
σ ∈Hom(K,Cp ) σ ∈Hom(K,Cp )
which is given by jp (x)σ := logp (σ(x)). Let ε1 , . . . , εr1 +r2 −1 be a basis of EK

modulo its torsion subgroup. We list the elements of Hom(K, Cp ) as σ1 , . . . , σd
with d := [K : Q].
(10.3.3) Definition. We define the regulator matrix

 
logp σ1 (ε1 ) ··· logp σd (ε1 )

Rp (ε1 , . . . , εr1 +r2 −1 ) :=  .. ... .. 
 . . 

logp σ1 (εr1 +r2 −1 ) · · · logp σd (εr1 +r2 −1 )
and set
rrp (K) := rank Rp (ε1 , . . . , εr1 +r2 −1 )
volp (K) := max {| det R|p | R is a (r1 + r2 − 1) × (r1 + r2 − 1)-
minor of Rp (ε1 , . . . , εr1 +r2 −1 )}.
We call rrp (K) the p-adic regulator rank of K.
Remarks: 1. It is easy to see that rrp (K) and volp (K) are independent of the
choice of the system ε1 , . . . , εr1 +r2 −1 and of the chosen ordering σ1 , . . . , σd of
the elements of Hom(K, Cp ), i.e. they are arithmetic invariants of K.
2. We think of volp (K) as the p-adic covolume of the unit lattice of K, because
of the analogy with the archimedean case, where the corresponding value is (up
√
to the factor r1 + r2 and a power of 2) the volume of a fundamental domain
of the unit lattice in the Minkowski space (cf. [160], chap.I, (7.5)).
Assume for a moment that the number field K is totally real, so that r1 +
r2 − 1 = d − 1. Then the regulator matrix is a (d − 1) × d-matrix and the sum
of all columns equals zero, since
X
logp σ(εi ) = logp (NK|Q εi ) = logp (±1) = 0.
σ
In this case the determinant of a (d − 1) × (d − 1)-minor is (up to sign)
independent of the choice of the minor. Also, changing the basis of the unit
group modulo torsion or changing the ordering σ1 , . . . , σd might introduce a
factor (±1).

(10.3.4) Definition. If K is totally real, we call

 
logp σ1 (ε1 ) · · · logp σd−1 (ε1 )

Rp (K) := det  .. .. .. 
 . . . 

logp σ1 (εd−1 ) · · · logp σd−1 (εd−1 )
the p-adic regulator of K. It is well-defined up to sign.
Returning to the case of an arbitrary number field we obviously have the

inequality rrp (K) ≤ r1 +r2 −1 and the analogy to the archimedean case predicts
the following conjecture.
(10.3.5) Leopoldt’s Conjecture. For every number field K and every prime
number p, the p-adic regulator rank rrp (K) is equal to r1 + r2 − 1.
For totally real number fields the Leopoldt conjecture is equivalent to the
non-vanishing of the p-adic regulator Rp . It is also equivalent to the non-
vanishing of certain p-adic L-functions at s = 1 (see [246]) in this case.
If ε is a unit which is not a root of unity, then logp (ε) =/ 0. Therefore
rrp (K) = r1 + r2 − 1 provided that r1 + r2 ≤ 2. In other words, the Leopoldt
conjecture is true for K and every prime number p if K is a quadratic number
field or if K is an extension of degree 2 of an imaginary quadratic number field
or if K is a cubic field which is not totally real. M. WALDSCHMIDT has shown
that rrp (K) ≥ 12 (r1 + r2 − 1) for every number field K and every prime number
p (see [244]). Later on in this section, we will see that the Leopoldt conjecture
is true for abelian extensions K|k where k = Q or k is imaginary quadratic.
There exist a large number of equivalent formulations of the Leopoldt con-
jecture, some of which we will describe below. Let us fix some notation.
As before we let the prime number p be fixed and we denote the p-adic
completion of an abelian group A by Â, i.e.
Â := lim
←−
A/pn A.
n
Observe that Â is a ZZp -module in a natural way. Further, Â = A ⊗ ZZ ZZp if
A is a finitely generated ZZ-module, and Â = A if A is a finitely generated
ZZp -module. If A is p-torsion-free, then we have an exact sequence
0 −→ Â −→ Â ⊗ ZZp Qp −→ A ⊗ ZZ Qp /ZZp −→ 0.
For a finite prime p of K we denote the group of units of the local field Kp by
Up and we set Up := Kp× for an archimedean p. Then Ûp is finite if p ∈/ Sp (K).
If p ∈ Sp (K), then the inclusion of the principal units Up1 ⊆ Up induces an
isomorphism Up1 → ∼ Û .
p

Assume that S ⊇ Sp ∪ S∞ and T ⊆ S are finite sets of places of K. We

denote by S f the subset of finite (i.e. nonarchimedean) primes in S. Consider
the diagonal embedding∗)
Kp× ,
Y Y
∆ : EK,T ,→ Up × e 7−→ (e, . . . , e).
p∈S\T p∈ T
It induces a homomorphism ∆0 : EK,T → p∈S\T Ûp × p∈T K̂p× . The kernel

Q Q
of ∆0 is the group µp0 (K) of roots of unity of order prime to p contained in K.

Indeed, we clearly have the inclusion µp0 ⊆ ker ∆0 . On the other hand, assume
e ∈ ker ∆0 and let p ∈ Sp be arbitrary. Then e = ζe0 in Kp , with ζ ∈ µp0 (Kp )
and e0 ∈ Up1 . The image of ζ in Ûp is trivial. Since Up1 maps isomorphically
onto Ûp , we conclude that e0 = 1. Therefore e = ζ in Kp and hence also in K.
We denote the topological closure of the image of EK,T under ∆0 in the
group p∈S\T Ûp × p∈T K̂p× by E (S)
Q Q
K,T . It is equal to the image of the induced
homomorphism
ˆ : EK,T ⊗ ZZ ZZp → K̂p× .
Y Y
∆ Ûp ×
p∈S\T p∈T
In particular, E (S)
is a finitely generated ZZp -module. If T = ∅, so that
K,T
EK,T = EK , we denote E (S) (S)
K,T by E K .
(10.3.6) Theorem. Let K be a number field, p be a prime number and assume

that S is a finite set of places of K containing Sp ∪ S∞ . Then the following
(i) Leopoldt’s conjecture is true for K and p.
(ii) rank ZZpE (S)
K = r1 + r2 − 1.
(iii) The canonical homomorphism
Y
EK ⊗ ZZ ZZp −→ Ûp
p∈S
is injective.
(iv) The kernel of the canonical homomorphism
Y
EK ⊗ ZZ Qp /ZZp −→ Up ⊗ ZZ Qp /ZZp
p∈S
is finite.
(v) rank ZZp H1 (GS (K), ZZp ) = r2 + 1.
(vi) H2 (GS (K), ZZp ) = 0.
(vii) The p-torsion subgroup of the universal norm group DS (K) of the class
formation (GS (K), CS (KS )) is isomorphic to
Y
µp (Kv ).
v ∈SC (K)
∗) Archimedean primes are ignored in T .

(10.3.7) Corollary. The following numbers agree and are independent of the
finite set S ⊇ Sp ∪ S∞ :
. rank ZZp ker(EK ⊗ ZZ ZZp −→
Q
p∈S Ûp ),
. corank ZZp ker(EK ⊗ ZZ Qp /ZZp −→
Q
p∈S Up ⊗ ZZ Qp /ZZp ),
. (S)
r1 + r2 − 1 − rank ZZp ĒK ,
. r1 + r2 − 1 − rrp (K),
. rank ZZp H1 (GS (K), ZZp ) − r2 − 1,
. rank ZZp H2 (GS (K), ZZp ).
We denote this number by dp (K) and call it the Leopoldt defect. One has
0 ≤ dp (K) ≤ r1 + r2 − 1,
and the Leopoldt conjecture for K and p is true if and only if dp (K) = 0.
(10.3.8) Corollary. Assume that the Leopoldt conjecture is true for K and p.
If p is unramified in K|Q, then

Up ⊗ ZZ Qp /ZZp = pr1 +r2 −1 · volp (K)−1 .
Y
# ker EK ⊗ ZZ Qp /ZZp →
p∈Sp
(10.3.9) Corollary. Suppose that p =/ 2 or that K is totally imaginary. If

S ⊇ Sp ∪ S∞ , then scdp GS (K) = 2 if and only if the Leopoldt conjecture holds
for p and every finite extension L of K inside KS .
We can also formulate the Leopoldt conjecture in terms of continuous

cochain cohomology.
(10.3.10) Corollary. We have

1
dp (K) = rank ZZp Hcts (GS (K), ZZp ) − r2 − 1
2
= rank ZZp Hcts (GS (K), ZZp ) .
2
In particular, the Leopoldt conjecture holds if and only if Hcts (GS (K), ZZp ) is
finite.
(10.3.11) Corollary. Let K|k be a finite extension. Then the following is true:
(i) If the Leopoldt conjecture holds for K and p, then it holds for k and p.
(ii) If K is a CM-field and k = K + is its maximal totally real subfield, then
the Leopoldt conjecture holds for K and p if and only if it holds for K +
and p.

For the proof of theorem (10.3.6) and its corollaries we need two lemmas.
(10.3.12) Lemma. Let S ⊇ T be sets of primes of K. Then there is a

canonical exact sequence
α
Kp× /Up −→ ClT (K) → ClS (K) → 0,
M
0 → EK,T → EK,S →
p∈S\T
where α is the map sending ⊕p xp mod Up to the class of pvp (xp ) .

Q
p
Proof: This is clear from the definition of the objects and maps occurring.
2
(10.3.13) Lemma. For finite S ⊇ Sp ∪ S∞ and T ⊆ S there is a canonical

H2 (GS
, ZZp ) K̂p×
Y Y
EK,T ⊗ ZZp Ûp × Gab
S (p) ClT (K)(p)
p∈S\T p∈T
K̂p×
Y
H2 (GS , ZZp ) EK,S ⊗ ZZp Gab
S (p) ClS (K)(p).
p∈S
(Our notational convention implies for an archimedean prime p that Ûp = K̂p× =
µ2 if p is real and p = 2, and Ûp = K̂p× = 0 otherwise.) In particular, there is
an exact sequence
0 −→ H2 (GS , ZZp ) −→ EK,T ⊗ ZZ ZZp −→E (S)
K,T −→ 0 .
Proof: We first show the exactness of the lower sequence. Dualizing the
Poitou-Tate sequence for the module ZZ/pn ZZ, we obtain an exact sequence
0 → X2 (GS , ZZ/pn ZZ)∨ → H 1 (GS , µpn ) →

M
H 1 (Kp , µpn )
p∈ S
→ H1 (GS , ZZ/pn ZZ) → X1 (GS , ZZ/pn ZZ)∨ → 0.
Local Kummer theory implies

lim H 1 (Kp , µpn ) = K̂p× ,
←−
n
and since H 2 (Kp , Qp /ZZp ) = 0 (see (7.2.5)), we obtain

X2 (GS , Qp /ZZp ) = H 2 (GS , Qp /ZZp ).

The global Kummer sequence, together with the finiteness of ClS (K), implies
that there is an isomorphism
EK,S ⊗ ZZ ZZp = lim H 1 (GS , µpn ).
←−
n
Further, observe that
X1 (GS , ZZ/pn ZZ)∨ = ClS (K)/pn .
Therefore we obtain the lower exact sequence of the lemma by passing to the
inverse limit over n in the dualized Poitou-Tate sequence above.
In order to obtain the right-hand part of the upper exact sequence, observe
that the cokernel of the homomorphism
K̂p× → K̂p× → H1 (GS , ZZp )
Y Y Y
Ûp ×
p∈S\T p∈T p∈S
is canonically isomorphic to ClT (K)(p) by global class field theory. In order

to complete the diagram, we use the exact sequence
K̂p× /Ûp −→ ClT (K)(p) ClS (K)(p),
M
EK,T ⊗ ZZp ,→ EK,S ⊗ ZZp →
p∈S\T
which is obtained from (10.3.12) by tensoring by the flat ZZ-module ZZp . 2
Proof of (10.3.6) and its corollaries: We set t = r1 + r2 − 1 and we omit the

suffix K during the proof.
(i) ⇔ (ii): Since Ûp is finite for primes p ∈/ Sp , we have
rank ZZpE (S) = rank ZZpE (Sp ) .
Thus we may work with S = Sp instead. The p-adic logarithms induce a
continuous homomorphism
Q
Y logp Y
Log : Ûp −−−−−−−→ Kp
p∈Sp p∈Sp
with finite kernel. Therefore we can equivalently calculate the rank of the
image of E (Sp ) under Log.
For p ∈ Sp , let dp := [Kp : Qp ] and let ap1 , . . . , apdp be a ZZp -basis of Op . Further,
we fix a basis ε1 , . . . , εt of E modulo torsion, and suppose that
dp
p p
X
logp (εi ) = ξij aj
j=1
for p ∈ Sp and 1 ≤ i ≤ t with ξijp ∈ Qp . Now let X be the (t × d)-matrix

p p1 p p
ξ111 ··· ξ1dp1 ξ112 · · · ξ1d2 p2 · · ·
 

p .. 

ξ211 . 
X :=
 

.. 
.
 
 
p
ξt11 · · · ,

where pv runs through Sp . Further, let

φp1 , . . . , φpdp : Kp ,→ Cp
be the dp embeddings of Kp into Cp . We order the d = [K : Q] embeddings
K ,→ Cp in the following way
p p p
φ1 , . . . , φd = φ11 |K , . . . , φd1p1 |K , φ12 |K , . . . .
Finally, let Dp (resp. D) be the (dp × dp ) (resp. (d × d)) matrices
φp1 (ap1 ) φpdp (ap1 )
 
...

Dp :=  .. .. .. 
 . . . 

p p
φ1 (adp ) ... φpdp (apdp )
for p ∈ Sp and
 
Dp1 0

 Dp2 

D := 
.. 

 . 

0 .
A simple calculation yields
Rp (ε1 , . . . , εt ) = X · D.
By the discriminant-product formula (see [160], chap.III, (2.11)), we have
| det D|2p = |discK|Q |p =/ 0.
Summarizing, we obtain
rank ZZpE (S) = rank ZZp Log(E (Sp ) )
= rank X
= rank Rp (ε1 , . . . , εt )
= rrp (K).
This shows the equivalence (i) ⇔ (ii). The equivalence (ii) ⇔ (iii) ⇔ (vi)
follows from (10.3.13) together with the fact that H2 (GS , ZZp ) is torsion-free
(see (8.7.7). Furthermore, the equivalence (v) ⇔ (vi) follows similarly by
counting ZZp -ranks in (10.3.13) or from the global Euler-Poincaré characteristic
formula (8.7.6).
We denote the torsion subgroup of an abelian group A by tor(A) and the
maximal torsion-free quotient by A/tor. Consider the commutative exact
diagram
ZZ ZZp
E/tor ⊗ E/tor ⊗ ZZ Qp E ⊗ ZZ Qp /ZZp
ϕ ϕ⊗Qp ψ
Y Y Y
Ûp /tor (Ûp /tor) ⊗ ZZp Qp Ûp ⊗ ZZp Qp /ZZp .
p∈S p∈ S p∈S

This diagram shows

rank ZZp ker ϕ = corank ZZp ker ψ.
Since E is a finitely generated abelian group this implies the equivalence (iii)
⇔ (iv).
Now assume that (i)–(vi) are true. Then the snake lemma shows that
Y
ker ψ = tor(( Ûp /tor)/imϕ).
p∈S
If Kp |Qp is unramified at every p ∈ Sp , then the p-adic logarithm logp : Ûp →

Kp has image exactly pOp and its kernel is equal to µ2 if p = 2 and trivial
otherwise (see [160], chap.II, §5). Hence
Y
#tor(( Ûp /tor)/imϕ)
p∈Sp
= pt · |product of the elementary divisors of X|−1

p
p | Y is (t × t)-minor of X}
= pt · min{| det Y |−1
p | R is (t × t)-minor of Rp (ε1 , . . . , εt )}
= pt · min{| det R|−1
= pt · volp (K)−1 .
This shows corollary (10.3.8).
It remains to show the equivalence of (vii) to the other conditions. Observe
that p Gab
S = p H1 (GS , Z
Zp ) and consider the diagram
H2 (GS , ZZp )/p
Y
0 µp H 0 (GS , CS (ZZ/pZZ)) H2 (GS , ZZ/pZZ) 0
v ∈SC (K)
rec ab
0 p DS (K) p CS (K) p GS 0.
The exactness of the upper row follows from (8.6.12), and the exactness of
the lower row follows from the divisibility of DS (K). The exactness of the
column follows from the short exact sequence ZZp ,→ ZZp ZZ/pZZ and the
dotted arrow is the induced one. Hence the snake lemma gives the exact
sequence
Y
0→ µp → p DS (K) → H2 (GS , ZZp )/p → 0.
v ∈SC (K)
By Nakayama’s lemma, the vanishing of H2 (GS , ZZp )/p is equivalent to the

vanishing of H2 (GS , ZZp ) itself and therefore we have proved the equivalence
(vi) ⇔ (vii).

Corollary (10.3.7) follows from the above proof, and (10.3.9) follows from
the equivalence (i) ⇔ (vi) and from corollary (3.3.4). Using (2.7.12), one
obtains statement (10.3.10) for continuous cochain cohomology. Corollary
(10.3.11)(i) follows from (10.3.7) and from the commutative diagram
EK ⊗!"# ZZ ZZp
Y
ÛP (K)
P∈Sp (K)
Y
Ek ⊗ ZZ ZZp Ûp (k).
p∈Sp (k)
Finally, (10.3.11)(ii) follows from the same diagram, noting that rank ZZ EK =
rank ZZ Ek if k is the maximal totally real subfield of the CM-field K. 2
Remark: From the equivalence (i) ⇔ (ii) in (10.3.6) and from the exact
sequence in (10.3.13) it follows that the Leopoldt conjecture is true for K
and p if and only if the canonical surjective homomorphism
EK,T ⊗ ZZ ZZp −→ E (S)

K,T
is an isomorphism for one (and hence every) pair S ⊇ T of finite sets of primes
of K with S ⊇ Sp ∪ S∞ .
Having established a number of equivalent formulations of the Leopoldt

conjecture, the question arises as to which of them might be the best one to
actually prove the conjecture. We will see later in §6 that the cohomological
condition (vi) can be verified for pairs (K, p) provided that a certain arithmetic
invariant vanishes. In this situation one can calculate H2 (GS , ZZp ) in terms of
local cohomology groups and then one uses (7.2.5). This method, however,
only covers cases which one should view as degenerate.
The authors do not know an algebraic proof of the Leopoldt conjecture in
any generic case.∗)
If the number field K is a finite abelian extension of a number field k with

r1 (k) + r2 (k) = 1 (i.e. k = Q or k imaginary quadratic), then the Leopoldt
conjecture can be deduced from the following deep result from transcendence
theory. It was proved by A. BRUMER and generalizes an archimedean result of
A. BAKER to the p-adic case.
∗) Attempts to give an algebraic proof often run into the problem that projective and inductive
limits do not commute.

(10.3.14) Theorem (BRUMER). Let α1 , . . . , αn ∈ C×

p be algebraic over Q and
assume that log (α ), . . . , log (α ) ∈ C
p 1 p n p
are linearly independent over Q. Then these logarithms are also linearly
independent over the algebraic closure of Q in Cp .
For the proof we refer the reader to [20]. In the following we will make use
of Dedekind’s determinant relation:
(10.3.15) Lemma. Let G be a finite group and let f : G → Cp be a map. Then

rank(f (σ −1 τ )σ,τ ∈G ) = #{χ ∈ Hom(G, C×
p) |
X
χ(σ)f (σ) =/ 0}.
σ ∈G
Proof: The Cp -vector space of all maps of G to Cp has two natural bases,
namely
(1) the characters χ ∈ Hom(G, C×
p ) and (
1 for σ = τ ,
(2) the characteristic maps dσ , σ ∈ G, with dσ (τ ) =
0 for σ =/ τ .
Consider the linear operator T with
X
T g(τ ) := f (σ)g(στ ) for g : G → Cp
σ ∈G
on this vector space. Then one easily calculates that T is represented by the
P
diagonal matrix diag( σ∈G χ(σ)f (σ))χ∈Hom(G,C×p ) with respect to basis (1) and
is represented by the matrix (f (σ −1 τ )σ,τ ∈G ) with respect to basis (2). 2
(10.3.16) Theorem. Assume that the number field K is an abelian extension

of Q or of an imaginary quadratic number field. Then the Leopoldt conjecture
holds for K and every prime number p.
Proof: If K is an abelian extension of Q, we can replace K by its maximal

real subfield if necessary, because of the equivalence (i) ⇔ (iii) of (10.3.6).
Hence we may assume that K is an abelian extension of a subfield k such that
Ek is finite and no archimedean prime splits in K|k. Then by (8.7.2) there
exists an isomorphism
EK ⊗ Q ∼ = IG ⊗ Q
as Q[G]-modules, where G = G(K|k) and IG is the augmentation ideal of
ZZ[G]. Therefore there exists a unit ε ∈ EK ∗) such that {σ(ε)}1=/σ∈G is an
independent system of units of K.
∗) If K|Q is abelian, ε is called a Minkowski unit.

Now fix any embedding φ : K ,→ Cp and consider the map

f : G → Cp ,
σ 7→ logp φ(σε) .
Clearly {φ ◦ σ −1 }σ∈G are r1 + r2 = #G different embeddings K ,→ Cp and

X X Y
f (σ) = logp φ(σε) = logp φ( σε) = logp ±1 = 0.
σ ∈G σ ∈G σ ∈G
P
Now assume that σ ∈G χ(σ) logp φ(σε) = 0 for a nontrivial character χ ∈
Hom(G, C×
p ). Then
X
(1 − χ(σ)) · logp φ(σε) = 0,
/ σ ∈G
1=
and the elements (1 − χ(σ)) are algebraic over Q and not all are zero. By
(10.3.14), there exist nσ ∈ ZZ, not all zero, with
(σε)nσ
X Y
nσ · logp φ(σε) = 0, i.e. ∈ µ(K).
/ σ ∈G
1= / σ ∈G
1=
This, however, contradicts the choice of the unit ε. With the help of (10.3.15),
we obtain
r1 + r2 − 1 = #G − 1 = rank(logp φ(σ −1 τ ε)σ,τ ∈G ) ≤ rrp (K).
This yields the nontrivial inequality, and hence the theorem is proved. 2
The Leopoldt conjecture is closely related to the existence of certain infinite

Galois extensions of K, the so-called ZZp -extensions.
(10.3.17) Definition. Let L|K be a Galois extension of fields. We call L a

ZZp -extension of K if G(L|K) is a free pro-p-group of rank 1, i.e. (noncanon-
ically) isomorphic to the additive group ZZp .
The closed subgroups of ZZp are exactly the groups pn ZZp for n = 0, 1,
2, . . . , ∞, where by convention p∞ ZZp = 0. Hence we can list the extensions
of K inside L in the form
K = K0 $ K1 $ K2 $ · · · $ K∞ = L,
where G(Kn |Km ) ∼ = ZZp /pn−m ZZp for n ≥ m. In particular, L = K∞ is also a

ZZp -extension of Kn for every 0 ≤ n < ∞.

Assume that p =/ char(K) and consider the extension

K(µp∞ )|K,
which is obtained by adjoining all roots of unity of p-power order to K. This is
a Galois extension and G(K(µp∞ )|K) is canonically isomorphic to the image
of the p-part of the cyclotomic character
κp : GK → ZZ×
p,
which is given by g(ζ) = ζ κp (g) for ζ ∈ µp∞ , g ∈ GK , cf. (7.3.6). Since

ZZ× ∼ Z/(p − 1)ZZ ⊕ ZZp for p =/ 2 and ZZ× ∼ Z/2ZZ ⊕ ZZ2 , the image of κp is
p = Z 2 = Z
either finite or has a unique quotient isomorphic to ZZp .
(10.3.18) Definition. If K(µp∞ )|K is infinite (which is the case if K is a local

or global field of characteristic =/ p), then it contains a unique ZZp -extension
K∞ , which we call the cyclotomic ZZp -extension of K.
If ζ2p ∈ K, then clearly K∞ = K(µp∞ ). A finite field has a unique ZZp -

extension for every p, which is the cyclotomic ZZp -extension if char(K) =/ p .
In the following we will always assume that p =/ char(K) if K is a local or
global field.
Let K be a local field with residue field k. Then K has a unique unramified
ZZp -extension, which is the cyclotomic one if p =/ char(k). If p = char(k), then
the cyclotomic ZZp -extension is ramified, so it is different to the unramified
ZZp -extension (however, their first n steps may coincide for some n < ∞).
The composite of all ZZp -extensions of K is equal to the maximal p-extension
of K with a torsion-free abelian Galois group. Hence
rank ZZp H1 (GK , ZZp )
is the (possibly infinite) number of independent ZZp -extensions of K. Therefore
the following lemma is an easy consequence of (7.5.3) and (7.3.10).
(10.3.19) Lemma. (i) If K is a local field with p =/ char(K) > 0, then K

has exactly one (namely the unramified) ZZp -extension which comes by base
change from the unique ZZp -extension of the finite constant field.
(ii) If K is a finite extension of Q` , then there exists exactly one (namely
the unramified) ZZp -extension of K if p =/ `, and when p = `, the number of
independent ZZp -extensions of K is equal to [K : Qp ] + 1.

Now assume that K is a global field (of characteristic =/ p). Then by (10.3.19)
a ZZp -extension is unramified at all primes of residue characteristic =/ p. Since
Cl(K) (resp. Cl0 (K) in the function field case) is finite, we observe by (10.3.7)
the
(10.3.20) Proposition. (i) A global function field K of characteristic =/ p

has exactly one ZZp -extension which comes by base change from the unique
ZZp -extension of the finite constant field; in particular, it is unramified.
(ii) If K is a number field, then every ZZp -extension is unramified outside p
and is ramified at least at one prime dividing p. The number of independent
ZZp -extensions of K is equal to r2 + 1 + dp , where r2 is the number of complex
places of K and dp is the Leopoldt defect (see (10.3.7)); in particular,
rank ZZp GS (K)ab (p) = r2 + 1 + dp ≤ [K : Q].
Hence another formulation of the Leopoldt conjecture is that there are at

most r2 + 1 independent ZZp -extensions of K. If K is totally real, this means
that the cyclotomic ZZp -extension is the unique ZZp -extension of K.
From now on we will restrict to the number field case. Let K be a number
field and let K∞ |K be any ZZp -extension.
(10.3.21) Definition. We say that the weak Leopoldt conjecture holds for
K∞ |K if the numbers dp (Kn ) are bounded independently from n ∈ [0, ∞).
We will see in a moment that the (strong) Leopoldt conjecture for K and
p implies the weak Leopoldt conjecture for every ZZp -extension of K, which
justifies the name.
In order to give a couple of equivalent formulations of the weak Leopoldt
conjecture, let us fix some notation. Let
K∞ |K be a fixed ZZp -extension,
Γ = G(K∞ |K) ∼ = ZZp ,
Γn = G(K∞ |Kn ) ⊆ Γ ,
S ⊇ Sp ∪ S∞ be a finite set of primes of K,
GS = G(KS |K),
HS = G(KS |K∞ ) ⊆ GS ,
XS = H1 (HS , ZZp ).
Then XS is a Λ = ZZp [[Γ ]]-module in a natural way.

(10.3.22) Theorem. The following assertions are equivalent:

(i) The weak Leopoldt conjecture is true for K∞ |K.
(ii) H2 (HS , ZZp ) = 0.
(iii) XS has no finite nontrivial submodules and rankΛ XS = r2 .
Observe that (i) is independent of the choice of the finite set of primes
S ⊇S∞ ∪ Sp . Hence (ii) and (iii) are true for all S if they are true for one S.
Proof: By (8.3.20), we can apply the results at the end of V §6 (p.331 ff)
to the triple HS , GS , Γ . By (8.7.7), H2 (GS , ZZp ) is p-torsion-free, and by
(8.7.5), χ2 (GS , ZZ/pZZ) = −r2 . Hence the equivalence (ii) ⇔ (iii) follows from
(5.6.17).
Now assume H2 (HS , ZZp ) = 0. Then the Hochschild-Serre spectral sequence
implies that
H2 (GS (Kn ), ZZp ) ∼
= H1 (Γn , H1 (HS , ZZp )) = XSΓn .
Therefore dp (Kn ) is globally bounded by rank ZZp XSδ , where XSδ is the maximal
discrete submodule of XS (see (5.3.12)). This shows (ii) ⇒ (i).
Suppose that (i) holds. Then by (10.3.7), the group
H2 (HS , ZZp ) = lim
←−
H2 (GS (Kn ), ZZp )
n
is a finitely generated ZZp -module. Assume for a moment that cdp GS ≤ 2
(i.e. p =/ 2 or SIR (K) = ∅). Then H2 (HS , ZZp ) is also a free Λ-module by
(5.6.15)(ii). This shows (ii).
If p = 2 and SIR (K) =/ ∅, we have to modify the above argument. Since K∞ |K
is unramified at all infinite places, it is disjoint from K(i)|K (or any totally
imaginary quadratic extension of K). We identify Γ with G(K∞ (i)|K(i)) and
we denote the open normal subgroup GS (K∞ (i)) ⊆ HS by HS0 . Then, as above,
we conclude that H2 (HS0 , ZZ2 ) is a free Λ-module. It is therefore sufficient to
find a Λ-invariant injection H2 (HS , ZZ2 ) ,→ H2 (HS0 , ZZ2 ) in order to finish
the proof. Recall (see (8.7.7)) that H2 (GS (Kn ), ZZ2 ) is torsion-free for all n.
Therefore H 2 (HS , Q2 /ZZ2 ) is divisible, and the usual restriction-corestriction
argument (use (1.5.7)) implies that
cor : H 2 (HS0 , Q2 /ZZ2 ) → H 2 (HS , Q2 /ZZ2 )
is surjective. This implies the required injection on the homology. 2
(10.3.23) Corollary. If the (strong) Leopoldt conjecture is true for K and p,

then the weak Leopoldt conjecture is true for every ZZp -extension K∞ of K.

Proof: This follows from the injectivity of the map

H2 (HS , ZZp )Γ ,→ H2 (GS , ZZp )
and from Nakayama’s lemma. 2
(10.3.24) Corollary. The weak Leopoldt conjecture is true for a ZZp -extension
K∞ |K if and only if the canonical homomorphism
lim
←−
EKn ,T ⊗ ZZ ZZp −→ lim
←−
E (S)
Kn ,T
n n
is an isomorphism for one (and hence every) pair S ⊇ T, S ⊇ Sp ∪ S∞ , of finite

sets of primes of K.
Proof: By (10.3.13), we have canonical exact sequences of compact abelian

groups for all n
0 −→ H2 (GKn ,S , ZZp ) −→ EKn ,T ⊗ ZZ ZZp −→E (S)
Kn ,T −→ 0 .
Passing to the projective limit over n, the result follows from (10.3.22). 2
In chapter XI we will investigate how the (strong) Leopoldt conjecture is

encoded in the Iwasawa module structure of XS = H1 (HS , ZZp ). We finish this
section with the
(10.3.25) Theorem. Let K be a number field and p be a prime number. Then

the weak Leopoldt conjecture is true for the cyclotomic ZZp -extension of K.
Proof: We verify condition (ii) of (10.3.22). Since the weak Leopoldt

conjecture descends, we may replace K by any finite extension. Therefore
we may assume µ2p ⊆ K, and so µp∞ ⊆ K∞ , i.e. the p-part of the cyclotomic
character is trivial on HS . We obtain
H2 (HS , ZZp )(−1)∨ = X2 (HS , Qp /ZZp )(1) = X2 (HS , µp∞ ),
where the first equality follows from (7.2.5). Then Poitou-Tate duality implies
X2 (HS , µp∞ ) = lim
−→
X2 (GS (Kn ), µpm )
n,m
= lim
−→
X1 (GS (Kn ), ZZ/pm ZZ)∨
n,m
= lim ClS (Kn )/pm
−→
n,m
= lim
−→
ClS (Kn ) ⊗ ZZ Qp /ZZp = 0 ,
n
since ClS (Kn ) is finite for every n. 2

Remark: Instead of using Poitou-Tate duality, we could have considered the

exact sequence
0 −→ µp∞ −→ EKS ,S −→ EKS ,S /µp∞ −→ 0.
Since ES /µp∞ is uniquely p-divisible, the result then follows from proposition
(8.3.11).
(10.3.26) Corollary. Let p be a prime number and let K be a number field

which is totally imaginary if p = 2. Let K∞ be the cyclotomic ZZp -extension
of K. Then for every set S of primes of K containing Sp ∪ S∞ we have
scdp G(KS |K∞ ) ≤ 2 .
Proof: If U is an open subgroup of G(KS |K∞ ), then there exists a finite

extension L|K inside KS such that U = G(KS |L∞ ). Thus by (10.3.25) we have
H 2 (U, Qp /ZZp ) = 0 and therefore H 3 (U, ZZ)(p) = 0. Since cdp G(KS |K∞ ) ≤
cdp G(KS |K) ≤ 2, the result follows from (3.3.4). 2
Remark: We will see in (10.11.3) that under the above assumptions the field
KS is always strictly larger than K∞ , hence scdp G(KS |K∞ ) = 2.
We have seen in this section that the Leopoldt conjecture is equivalent to the
vanishing of H 2 (GS , Qp /ZZp ), where S is any finite set of primes containing
Sp ∪S∞ . It is natural to consider also other Tate twists of Qp /ZZp . For instance it
is not difficult to show that X2 (GS , µp∞ ) = 0, and so we have an isomorphism
H 2 (GS , Qp /ZZp (1)) ∼

f
= (Qp /ZZp )#S −1 .
P. SCHNEIDER [201] has conjectured that the twist by +1 is the only Tate twist
of Qp /ZZp having a nontrivial second cohomology group. For positive twists
this has been proved by C. SOULÉ [219] by relating Galois cohomology to the
higher K-groups of Ok , which were defined and shown to be finitely generated
by D. QUILLEN :
(10.3.27) Theorem. Let k be a number field, let p be an odd prime number

and assume that S is a finite set of primes of k containing Sp ∪ S∞ . Then
H 2 (GS , Qp /ZZp (i)) = 0
for i ≥ 2.

Exercise: Show that the Leopoldt conjecture holds for k and p if and only if for all finite sets
T ⊇ S ⊇ Sp ∪ S∞ , the natural map
M
T (kpab (p)|kp ) −→ G(kTab (p)|k)
p∈T \S
ab
is injective. Here T (kp (p)|kp ) denotes the inertia group of the Galois group of the maximal
abelian p-extension of the local field kp (which is isomorphic to µ(kp )(p)).
Hint: Use the Hochschild-Serre sequence and (10.5.3) to obtain the exact sequence
M
H2 (GT (k), ZZp ) → H2 (GS (k), ZZp ) → T (kpab (p)|kp ) → GT (k)(p)ab → GS (k)(p)ab → 0.
p∈(T \S)(k)
Then pass to the limit over all T and use H 2 (Gk , Qp /ZZp ) = 0.
§4. Cohomology of Large Number Fields
In this section we investigate algebraic extensions K|Q of not necessarily

finite degree. We call (by abuse of language, and only in this and the next
section) such a field K a number field and we will say that K is finite if it is of
finite degree over Q, i.e. if it is a number field in the strict sense.
Suppose that we are given sets of primes S and T of the number field K.
Without further mention we will tacitly assume that S = S 0 (K), respectively
T = T 0 (K), where S 0 and T 0 are sets of primes of a finite number field k ⊆ K.
In other words, we assume that S and T are closed and open in Sp(K) (see §1).
As before, S f denotes the subset of finite primes in S. Finally, recall the
notation
K T is the maximal extension of K which is completely decomposed

at T ,
KST is the maximal extension of K which is unramified outside S
and completely decomposed at T .
(10.4.1) Definition. Let p be a prime number. The field K is called p-(S, T )-

closed if KST (p) = K, i.e. if there is no nontrivial Galois p-extension of K
inside KST . We call p-(S, ∅)-closed fields p-S-closed for short.
The following theorem generalizes a result of O. NEUMANN [162] about

Galois groups of extensions of p-S-closed number fields.

§4. Cohomology of Large Number Fields 643
(10.4.2) Theorem. Let L|K be a Galois extension of number fields and let p
and S ⊇ Sp ∪ S∞ be as above. Suppose that
(i) L is p-S-closed,
(ii) K is p-(Sp ∪ S∞ )-closed, or
(ii)0 p∞ divides the absolute degree of Kp for all p ∈ S f , SIR (K) = ∅ if p = 2,
K is p-(S∞ , T )-closed for some T ⊆ S and µp ⊆ KST∞ .
Then H 2 (G(L|K), ZZ/pZZ) = 0. (In particular, cdp G(L|K) ≤ 1 if L|K is a

p-extension.) If in addition L ⊆ KS , then
H i (G(L|K), ZZ/pZZ) = 0 for all i ≥ 2.
Remark: Theorem (10.4.2) has the following function field analogue. Assume
that S =/ ∅ and p =/ char(K). Then the statement of (10.4.2) remains true with
the following modifications of the second assumption:
- replace (ii) by: K is p-∅-closed,
- replace (ii)0 by: p∞ divides the absolute degree of Kp for all p ∈ S,
T
K is p-(∅, T )-closed for some T ⊆ S and µp ⊆ K∅ .
All corollaries have their obvious function field analogue.
(10.4.3) Corollary (NEUMANN). Let L|K be a Galois extension of p-S-closed

number fields with S ⊇ Sp ∪ S∞ and L ⊆ KS . Then
H i (G(L|K), ZZ/pZZ) = 0 for all i ≥ 1.
(10.4.4) Corollary. Let k be a finite number field and let S ⊇ T ⊇ Sp ∪ S∞ be

sets of primes of k. Then
cdp G(kS (p)|kT (p)) ≤ 1.
(10.4.5) Corollary. Let k be a totally imaginary finite number field, S ⊇

S
Sp ∪ S∞ be a set of places of k and assume that µp ⊆ k. Let L0 = (k∞ )Sp∞ (p)
be the maximal unramified p-extension of the cyclotomic ZZp -extension k∞ of
k which is completely decomposed at all primes in Sp . Then
cdp G(kS (p)|L0 ) ≤ 1.
For the proof of (10.4.2) we need the

(10.4.6) Lemma. Let L|K be a Galois extension of number fields and let
S ⊇ Sp ∪ S∞ be a set of primes in K. Then the inflation maps
×
H i (G(KS |K), OK S ,S
)(p) −→ H i (G(LS |K), OL×S ,S )(p)
are isomorphisms for all i ≥ 0.

×
→ k× →
M
0 → Ok,S ZZ → ClS (k) → 0,
/S(k)
p∈
where k is any finite number field (containing the finite number field over
which we assume the set S to be defined). Passing to the limit over all k, and
since Cl(K) = 0, we obtain the exact sequence
×
→ K × → lim
M
0 → OK,S ZZ → 0.
−→
/S(k)
k⊆K p∈
The transition maps in the limit on the right are induced by multiplication
with the ramification index. Therefore the limit on the right is a Q-vector
space; in particular, it is cohomologically trivial as a G(K|K)-module. Taking
G(K|KS )-cohomology and recalling that ClS (KS ) = 0 (take the direct limit
over ClS (kS ) = 0, k ⊆ K finite), we obtain isomorphisms for all i ≥ 1
× ∼ H i (G(K|K ),K × ).
H i (G(K|KS ), OK,S ) −→ S
×
Hence H i (G(K|KS ), OK,S )(p) is zero for all i ≥ 1: for i = 1 by Hilbert’s
Satz 90, for i = 2 by (8.1.14) (ii) since the absolute local degree of every
p ∈ S f is divisible by p∞ (observe that Q(µp∞ ) ⊆ KS ), and for i ≥ 3 since
scdp G(K|KS ) ≤ 2 by (10.2.3). Therefore the Hochschild-Serre spectral se-
quence implies isomorphisms for all i
× ∼ H i (G(K|K), O × )(p).
)(p) −→ K,S
We also can replace KS by LS in the above argument. Then the diagram

$%& O × )(p)
H i (G(LS |K), ×
H i (G(K|K), OK,S )(p)
LS ,S
×
)(p)
shows the statement of the lemma. 2
Proof of (10.4.2): First observe that in case (ii), K contains the cyclotomic
ZZp -extension of Q. Hence the local field Kp has absolute degree divisible by
p∞ for every prime p ∈ S f . In addition SIR (K) = ∅ if p = 2 in case (ii). Let us
write H i (L|K) for H i (G(L|K), ZZ/pZZ) in the following.

§4. Cohomology of Large Number Fields 645
We first assume that L = LS . Let K 0 := K(µp ). From (8.3.11)(iii) and

lemma (10.4.6) we obtain a inclusion
Y
pH
2
(G(L|K 0 ), OL,S
×
) ,→ p H 2 (Kp0 , µp∞ )
p∈S
showing that
pH
2
(G(L|K 0 ), OL,S
×
)=0
by (7.1.8)(i). Therefore the Kummer sequence implies the isomorphism
∼
H 1 (G(L|K 0 ), OL,S
×
)/p → H 2 (G(L|K 0 ), µp ).
Twisting by (−1) and taking G = G(K 0 |K)-invariants, we obtain isomor-
phisms (observe that G is of order prime to p)
(∗) H 2 (L|K) ∼
= H 2 (L|K 0 )G ∼
= (H 1 (G(L|K 0 ), O× )/p)(−1)G . L,S
Now let 

 KSp ∪S∞ in case (ii)
K :=

 KST∞ in case (ii)0 .
Then K 0 ⊆ K and K is closed under p-extensions, which are (everywhere)

unramified and completely decomposed at the primes in S. Therefore (8.3.11),
(10.4.6) and the principal ideal theorem, imply
H 1 (G(L|K), O× )(p) ∼ = lim ClS (k)(p) = 0.
L,S −→
k⊆K
The Hochschild-Serre sequence therefore yields an isomorphism
(∗∗) H 1 (G(K|K 0 ), O× )(p) ∼
= H 1 (G(L|K 0 ), O× )(p).
K,S L,S
By definition of K and by our assumptions on K, we have
H 1 (G(K|K 0 ), µp )(−1)G ∼
= H 1 (K|K) = 0.
Using the Kummer sequence, the last equality implies
G
pH
1
(G(K|K 0 ), OK,S
×
)(−1) =0
and hence by the isomorphism (∗∗) that
G
(∗∗∗) (G(L|K 0 ), OL,S
pH
1 ×
)(−1) = 0.
×
Since H 1 (G(L|K 0 ), OL,S )(p) ∼
= ClS (K 0 )(p) and ClS (K 0 ) is a torsion group,
we conclude that
(H 1 (G(L|K 0 ), OL,S
×
)/p)(−1)G = 0.
Finally, the isomorphism (∗) shows that H 2 (L|K) = 0.
Now drop the assumption that L = LS . The group H 1 (LS |L) vanishes by
condition (i), so that
H 2 (L|K) ,→ H 2 (LS |K) ,
which proves H 2 (L|K) = 0; in particular, G(L|K) is free if L|K is a p-
extension.

If L ⊆ KS , then consider the Hochschild-Serre spectral sequence for the

extensions KS |L|K. By the above arguments, we know that H 2 (KS |L) = 0,
and since cdp G(KS |L) ≤ cdp GS (K) ≤ 2, we obtain isomorphisms
H i (L|K) ∼
= H i (KS |K) = 0
for all i ≥ 3. 2
(10.4.7) Corollary (NEUMANN). Let K be a p-(Sp ∪ S∞ )-closed number field.

Then for every set of primes S ⊇ Sp ∪ S∞ ,
ClS (K(µp ))(p)(j)G(K(µp )|K) = 0
for the Tate twists j = 0, −1.
Proof: The case j = −1 follows from equation (∗∗∗) in the last proof. The
case j = 0 is a direct consequence of the principal ideal theorem. 2
The following corollary is a result of O. NEUMANN in the case that c is the

class of p-groups and that M is a finite p-torsion module.
(10.4.8) Corollary. Let c be a full class of finite groups with ZZ/pZZ ∈ c. Let k
be a finite number field and S ⊇ Sp ∪ S∞ be a set of primes of k. Then the
inflation maps
H i (G(kS (c)|k), M )(p) −→ H i (G(kS |k), M )(p)
are isomorphisms for all i and every G(kS (c)|k)-module M .
Proof: Using the Hochschild-Serre spectral sequence it suffices to show that

H i (G(kS |kS (c)), M ) = 0
for all i ≥ 1 and every trivial G(kS |kS (c))-module M . Since cohomology
commutes with direct limits, one easily reduces to the cases M = ZZ, ZZ/pZZ.
Using the exact sequence 0 → ZZ → Q → Q/ZZ → 0, we finally reduce to the
case M = ZZ/pZZ. Now the corollary follows from (10.4.2), since conditions
(i) and (ii) are clearly satisfied. 2
Remark: In §6 we will prove results corresponding to (10.4.7) and (10.4.8)

in the case where p = 2 and S ⊇ S2 does not necessarily contain all infinite
primes, see (10.6.6) and (10.6.7).

§5. Riemann’s Existence Theorem 647
(10.4.9) Corollary. Let p be a prime number, k a number field and S a set of

primes of k containing Sp ∪ S∞ . Then
(i) cdp G(kS (p)|k) ≤ cdp G(kS |k),
(ii) scdp G(kS (p)|k) ≤ scdp G(kS |k).
Assume, in addition, that k is totally imaginary if p = 2 and let k∞ be the
cyclotomic ZZp -extension of k. Then
(iii) scdp G(kS (p)|k∞ ) ≤ 2.
Proof: This follows from (10.4.8) and (10.3.26). 2
§5. Riemann’s Existence Theorem
Now we are prepared to prove the number field analogue of Riemann’s

existence theorem. In the special case k = Q, it is due to J. NEUKIRCH (see
[155]). A first attempt to generalize Neukirch’s theorem to arbitrary number
fields was made by O. NEUMANN [163]; however, he was still lacking the
notion of generalized free products of profinite groups over a topological base.
In the special case k = Q, the product occurring is a product over a discrete
base, hence the use of generalized products may be dispensed with. In the
general case its use is crucial. In the form presented below the theorem
was proved in [251]∗) . In this section we keep the convention of calling an
algebraic extension of k of Q a number field and to call k a finite number field
if [k : Q] < ∞.
(10.5.1) Riemann’s Existence Theorem (Number Theoretical Analogue). Let

k be a finite number field, p a prime number and T ⊇ S ⊇ Sp ∪ S∞ sets of
primes of k. Then the canonical homomorphism
φT,S : ∗
p∈T \S(kS (p))
T (kp (p)|kp ) −→ G(kT (p)|kS (p))
is an isomorphism. Here T (kp (p)|kp ) ⊆ G(kp (p)|kp ) is the inertia group and ∗
denotes the free pro-p-product.
∗) In [251] one has to replace the definition of generalized free products by that given in
chap. IV §3.

Remarks: 1. The index set T r S(kS (p)), resp. its closure in Sp(kS (p)), is a
profinite index space (cf. the discussion in §1). The inertia groups T (kp (p)|kp )
form a bundle of profinite groups over this base space. The bundle structure can
either be defined directly, or we can view it as the bundle which is associated
to the continuous family T (kp (p)|kp ) = Tp (k(p)|k) of subgroups in G(k(p)|k)
(cf. (9.4.3)).
As before, we do not specify the choice of a continuous section of the
projection T r S(kT (p)) → T r S(kS (p)). Since we work in the category of
pro-p-groups, it is clear from the discussion in chapter IV that the particular
isomorphism claimed in the theorem depends on the choice of such a section;
but whether the canonical homomorphism is an isomorphism or not is inde-
pendent of the chosen section. The straightforward extension of (4.1.5) to the
case of generalized free products provides us with a convenient cohomological
criterion. Finally, note that T (kp (p)|kp ) = 1 for primes p ∈ T r S(k) for which
N (p) 6≡ 1 mod p, i.e. if there is no primitive p-th root of unity contained in the
local field kp , and T (kp (p)|kp ) ∼
= ZZp (1) otherwise (see 7.5.2).
2. The assumption S ⊇ S∞ (only interesting if p = 2 and k has real places) can
be removed from the theorem, see (10.6.1).
Before proving theorem (10.5.1), let us deduce some corollaries.
(10.5.2) Corollary. Let K be a p-S-closed number field and assume that

T ⊇ S ⊇ Sp ∪ S∞ . Then the canonical homomorphism
∗
p∈T \S(K)
G(Kp (p)|Kp ) −→ G(KT (p)|K)
is an isomorphism.
Proof: Let k ⊆ K be a finite number field over which S and T are defined.
Since K is p-S-closed, it contains the cyclotomic ZZp -extension k∞ of k.
Hence T (Kp (p)|Kp ) = G(Kp (p)|Kp ) for every prime p ∈ T r S(K). Now the
corollary follows from theorem (10.5.1) by passing to the limit over all finite
subextensions of k in K. 2
Specializing to the case K = kS , where S∞ ∪ Sp ⊆ S, and recalling that

µp ⊆ kS in this case, we obtain the
(10.5.3) Corollary. If S∞ ∪Sp ⊆ S ⊆ T , then there is a canonical isomorphism

∗
p∈T \S(kS )
Tp (p) −→ G(kT |kS )(p),
where Tp is the inertia group in the full local group Gp = Gkp .

Proof of (10.5.1): By (10.4.4), the pro-p-group G(kT (p)|kS (p)) is free and the
same is true for the free pro-p-product on the left. Therefore it suffices to show
that the induced homomorphism on the abelianizations is an isomorphism.
Using (4.3.14) and local class field theory, the abelianization of the free product
can be calculated as
∗
ab
∼

(T (kp0 (p)|kp0 ))ab
Y
T (kp (p)|kp ) = lim
p∈T \S(kS (p)) ←−
k0 ⊆kS (p) p∈T \S(k0 )
∼
Y
= lim Ûp ,
←−
k0 ⊆k S (p) p∈T \S(k0 )
where k 0 runs through the finite subextensions of k in kS (p) and Ûp is the pro-p
completion of the unit group of the local field kp0 .
Passing to the inverse limit over the upper exact sequences of (10.3.13) for
T = ∅ and all finite subsets of S(k 0 ) containing Sp ∪ S∞ , and doing the same
for T instead of S, we obtain the commutative exact diagram (writing Ek0 for
Ok×0 , GS for GS (k 0 ) and GT for GT (k 0 ))
H2 (GT'()*+,-./0123 , ZZp )
Y
Ek0 ⊗ ZZp Ûp H1 (GT , ZZp ) Cl(k 0 )(p)
p∈T (k0 )
Y
H2 (GS , ZZp ) Ek0 ⊗ ZZp Ûp H1 (GS , ZZp ) Cl(k 0 )(p).
p∈S(k0 )
Now we pass to the limit over all k 0 ⊆ kS (p). Observe that

H2 (GT (kS (p)), ZZp ) = 0 by (10.4.2),
H2 (GS (kS (p)), ZZp ) = 0 by (10.4.2),
H1 (GS (kS (p)), ZZp ) = 0 by definition,
lim Cl(k 0 )(p) = 0 since kS (p) has no unramified p-extensions.
←−
k0 ⊆kS (p)
Therefore we obtain the commutative exact diagram
E45678 k0 ⊗ ZZp
Y
lim
←−
lim
←−
Ûp G(kT (p)|kS (p))ab
k0 ⊆kS (p) k0 ⊆kS (p) p∈T (k0 )
Y
lim
←−
Ek0 ⊗ ZZp lim
←−
Ûp
0
k0 ⊆kS (p) k0 ⊆kS (p) p∈S(k )
in the limit. This finishes the proof of theorem (10.5.1). 2
The next corollary follows from (10.5.2) and (4.3.14).

(10.5.4) Corollary. Let k be a finite number field, let p be a prime number

and let T ⊇ S ⊇ Sp ∪ S∞ be sets of primes of k. Assume that K is a p-S-
closed Galois extension of k. Furthermore, let A be a p-primary abelian group
(considered as a trivial module). Then
H i (G(KT |K), A) = 0 for i ≥ 2 ,
and there exists an isomorphism of G(K|k)-modules
M0
H 1 (G(KT |K), A) ∼
= H 1 (Kp , A).
p∈T \S(K)
Remark: Recall that

M0
H 1 (kp0 , A),
M
H 1 (Kp , A) = lim
−→
p∈T \S(K) k0 p∈T \S(k0 )
0
where k runs through the finite extensions of k inside K; see the remark after
(4.3.13).
(10.5.5) Corollary. Let A be a finite p-primary, trivial G(kS |k∞ )-module and
let T ⊇ S ⊇ Sp ∪ S∞ be sets of primes of the finite number field k. If k∞ is the
cyclotomic ZZp -extension of k, then the canonical inflation map
H 2 (G(kS |k∞ ), A) −→ H 2 (G(kT |k∞ ), A)
is surjective.
Proof: Consider the Hochschild-Serre spectral sequence for the extensions

kT |kS |k∞ :
E2ij = H i (G(kS |k∞ ), H j (G(kT |kS ), A)) ⇒ H i+j (G(kT |k∞ ), A).
By (10.5.3), the spectral term E21,1 vanishes and by (10.4.2) E20,2 = 0 also.
Since E22,0 E∞2,0
, this gives the statement of the corollary. 2
Let us consider the special case k = Q and assume that p is odd. Then
QSp ∪S∞ (p) = Q∞ .
Indeed, by the theorem of Kronecker-Weber, the maximal abelian extension
of Q is obtained by adjoining all roots of unity. The ramification behaviour of
these extensions is well-known, and
QSp ∪S∞ (p)ab = Q∞ .

Therefore dimIFp H 1 (GSp ∪S∞ , IFp ) = 1, i.e. GSp ∪S∞ (p) is procyclic, which
shows the desired identity.
Applying Riemann’s existence theorem in this special case yields the fol-
lowing result of NEUKIRCH [155]:
(10.5.6) Theorem (NEUKIRCH). Let S ⊇ Sp ∪ S∞ be a finite set of primes of

Q and assume that p is odd. Then the canonical homomorphisms define an
isomorphism
∗
p∈S\(Sp ∪S∞ )(Q∞ )
T (Qp (p)|Qp ) ∼
= G(QS (p)|Q∞ ).
(10.5.7) Corollary. Let S ⊇ Sp ∪ S∞ be a finite set of primes of Q. Assume

that p is odd and that S contains at least one prime number ≡ 1 mod p. Then
G(QS (p)|Q) is a pro-p duality group of dimension 2. If S does not contain a
prime number ≡ 1 mod p, then
QS (p) = Q∞ .
Proof: Consider the group extension

(∗) 1 −→ G(QS (p)|Q∞ ) −→ G(QS (p)|Q) −→ G(Q∞ |Q) −→ 1.
By (10.5.6), the group on the left is a free product of inertia groups. These
are trivial if N (p) 6≡ 1 mod p and isomorphic to ZZp otherwise. Therefore the
group on the left is either trivial or a free pro-p-group of finite rank. In the
latter case it follows from (3.7.5) that G(QS (p)|Q) is a duality group. 2
Corollary (10.5.7) provides the first example where the maximal pro-p-
factor of GS is a duality group. We will investigate this phenomenon in more
generality in §9.
The above exact sequence (∗) provides a canonical filtration of the group
G(QS (p)|Q). Since Γ := G(Q∞ |Q) is a free pro-p-group, the sequence splits.
After choosing of a section s : Γ → G(QS (p)|Q), the group Γ acts on the
free product. If there is only one nontrivial factor in the free product (so that
G(QS (p)|Q∞ ) is abelian), then this action is easily understood. If there is more
than one nontrivial factor in the free product, the question arises, whether it is
possible to understand the non-abelian action of Γ (maybe for a specific choice
of s). This interesting problem seems to be unsolved. A solution would give
us a description of G(QS (p)|Q) as a pro-p-group in terms of generators and
relations.

If the set of primes S has Dirichlet density δ(S) = 1, then we have a

more general form of Riemann’s Existence Theorem, which also involves
decomposition groups. It was first proved in [258]. Recall the following
notation for sets of primes R and S:
k R (p) is the maximal p-extension of k which is completely decomposed
at R,
kSR (p) is the maximal p-extension of k which is unramified outside S
and completely decomposed at R.
(10.5.8) Theorem. Let k be a finite number field, p a prime number and

T ⊇ S ⊇ R sets of primes of k. If R is finite and δ(S) = 1, then the natural
homomorphism
φR
T,S : ∗R (p))
p∈R(kS
G(kp (p)|kp ) ∗ ∗
R (p))
p∈T \S(kS
∼ G(k (p)|k R (p))
T (kp (p)|kp ) −→ T S
is an isomorphism.
Remarks: 1. We have made no assumption on the set Sp ∪ S∞ . In particular,

the cyclotomic ZZp -extension of k is not necessarily contained in kSR (p). Further
note that the obvious analogs of (10.5.2) and (10.5.3) hold.
2. The situation for c-extensions is much harder, where c is a class of finite
groups which is strictly larger than the class of p-groups for some prime
number p. Recall that whether or not the homomorphism
φRT,S (c) : ∗ G(kp (c)|kp ) ∗
R (c))
p∈R(kS
∗
T (kp (c)|kp ) −→ G(kT (c)|kSR (c))
R (c))
p∈T \S(kS
is an isomorphism depends on the choice of a continuous section s: R(k R (c)) →

R(k(c)).
If c is the class of all finite groups, T = S is the set of all primes and R is
finite, then there exists a continuous section s : R(k R ) → R(k̄) such that φR is
an isomorphism of profinite groups. This was proven in the special case that
R = S∞ is the set of archimedean primes by FRIED-HARAN-VÖLKLEIN in [52];
the general case was proven by POP in [174].
In the proof of (10.5.8) we will use the following result.
(10.5.9) Proposition. Let R ⊆ S be sets of primes of k. If R is finite and

δ(S) = 1, then 
 kp (p), if p ∈ S r R,


R nr
(kS (p))p = kp (p), if p ∈/ S,



kp , if p ∈ R,

where kpnr (p) denotes the maximal unramified p-extension of kp .
Proof: The assertion is trivial for primes p ∈ R. In order to show the statement
for primes p ∈/ R, it suffices to show that, for each finite sets of primes T1 ,
T2 with T1 ⊆ S r R and T2 ∩ S = ∅, and each finite extension k 0 of k inside
kSR (p), the restriction homomorphism
(1) H 1 (kSR (p)|k 0 , ZZ/pZZ) −→ H 1 (kp0 , ZZ/pZZ) ⊕ (kp0 , ZZ/pZZ)
M M
1
Hnr
p∈T1 (k0 ) p∈T2 (k0 )
is surjective. Indeed, let p ∈ S r R. Putting T1 = {p}, we see that the field

(kSR (p))p has no elementary abelian p-extension and is therefore equal to kp (p).
The argument for p ∈/ S is similar.
For T1 and T2 as above, the restriction homomorphism
H 1 (kS∪T2 |k 0 , ZZ/pZZ) −→ H 1 (kp0 , ZZ/pZZ)
M
(2)
p∈T1 ∪T2 ∪R(k0 )
is surjective by (9.2.7). Therefore, each α ∈ p∈T1 ∪T2 ∪R(k0 ) H 1 (kp0 , ZZ/pZZ)

L
with αp ∈ Hnr 1
(kp0 , ZZ/pZZ) for p ∈ T2 and αp = 0 for p ∈ R has a pre-image in
H 1 (kS∪T2 |k 0 , ZZ/pZZ), which obviously lies in H 1 (kSR (p)|k 0 , ZZ/pZZ). Hence the
restriction homomorphism in (1) is surjective. This finishes the proof. 2
Proof of (10.5.8): It suffices to show the theorem in the case T = {all places},
because the statement for general T ⊇ S follows by dividing out the normal
subgroup generated by the inertia groups of the primes p ∈/ T (kSR (p)).
We write H ∗ (−) for H ∗ (−, ZZ/pZZ). By (1.6.15) it suffices to show that the
homomorphism

φi : H Gi
k(p)|kSR (p) →H i
∗ G(kp (p)|kp ) ∗
R (p))
p∈R(kS
∗ T (kp (p)|kp )
/S(k R (p))
p∈ S
is an isomorphism for i = 1 and injective for i = 2 (we omit the coefficients

ZZ/pZZ from the notation). By (4.3.14), the right hand side is isomorphic to
M0 M0
H i (G(kp (p)|kp )) ⊕ H i (T (kp (p)|kp )).
R (p))
p∈R(kS R (p))
p∈/S(kS
The map φ1 is injective since G(kT (p)|kSR (p)) is generated by the inertia groups
of the primes in T r S and the decomposition groups of the primes in R. Next
we show that φ1 is surjective.
Let k 0 be a finite extension of k inside kSR (p). By (10.4.8) and (7.5.8), the
maps H i (G(k(p)|k 0 ) → H i (G(k̄|k 0 )) and H i (G(kp (p)|kp0 )) → H i (G(k̄p |kp0 )) are
isomorphisms for all i.
As the complement of S has density zero, the restriction homomorphism
0
H 1 (G(k̄|k 0 )) −→ H 1 (G(k̄p |kp0 ) ⊕ H 1 (T (k̄p |kp0 ))G(k̄p |kp )
M M
(∗)
p∈R(k0 ) /S(k 0 )
p∈

is surjective by (9.2.7). For p ∈/ S, we have isomorphisms

G(k̄p |kpnr (p)) G(k̄p |kpnr (p))
∼

H 1 T (k̄p |kp ) = H 1 G(k̄p |kpnr )
∼

= H 1 G(k̄p |k nr (p))p
∼

= H G(kp (p)|kpnr (p))
1
∼

= H 1 T (kp (p)|kp ) .
By (10.5.9), kp0 approaches kpnr (p) when k 0 runs through the finite extensions of
k inside kSR (p). Therefore, passing to the limit in (∗), we obtain the surjectivity
of φ1 .
It remains to show that φ2 is injective. For any finite extension k 0 of k in
kSR (p), the Hasse principle (9.1.10)(i) yields an injection
H 2 (k(p)|k 0 ) ,→ H 2 (kp (p)|kp0 ).

M
Passing to the limit over all k 0 , (10.5.9) implies that only the entries for p ∈ R
survive on the right hand side. Consequently, φ2 is injective. 2
Using (10.5.8), we are now able to show the following theorem on cohomo-
logical dimensions.
(10.5.10) Theorem. Let k be a finite number field, p a prime number and

S ⊇ R sets of primes of k. Assume that k is totally imaginary if p = 2. If R is
finite and δ(S) = 1, then
cd G(kSR (p)|k) = scd G(kSR (p)|k) = 2.
Proof: For a nonarchimedean prime p, the groups G(kp (p)|kpnr (p)) and
G(kpnr (p)|kp ) are free. Therefore the Hochschild-Serre spectral sequence pro-
vides an isomorphism
H 1 (kpnr (p)|kp , H 1 (kp (p)|kpnr (p))) −→
∼ H 2 (k (p)|k ),
p p
where we omit the coefficients ZZ/pZZ from the notation. Next we consider the
Hochschild-Serre spectral sequence for the tower k(p)|kS (p)|k. By (10.4.9)
and (8.3.18), we have cd G(k(p)|k) ≤ 2, hence H 3 (k(p)|k) = 0. By (10.5.8) for
T = {all places} and R = ∅, the group G(k(p)|kS (p)) is free (hence E2ij = 0
for j ≥ 2) and we have an isomorphism
H 1 (k(p)|kS (p)) ∼
M G (k (p)|k)
= p S
IndG(kS (p)|k)
H 1 (T (kp (p)|kp )).
/S
p∈
Together with (10.5.9) this implies for E21,1 = H 1 (kS (p)|k, H 1 (k(p)|kS (p))):

E21,1 ∼
M G (k (p)|k)
= H 1 (kS (p)|k, IndG(k
p S
S (p)|k)
H 1 (T (kp (p)|kp )))
p S
∈
/
∼ H 1 (kpnr (p)|kp , H 1 (T (kp (p)|kp )))
M
=
p S
∈
/
∼
M
= H 2 (kp (p)|kp ).
p∈/S
Therefore the Hochschild-Serre spectral sequence, the long exact sequence of

Poitou-Tate, the Hasse principle (9.1.10)(i) and (10.4.8) imply a commutative
exact diagram
9:;<=>?@
M
H 2 (kp (p)|kp ))
p∈S
M
H 2 (k(p)|k) H 2 (kp (p)|kp ) H 0 (k̄|k, µp )∨
p
M
H 1 (kS (p)|k, H 1 (k(p)|kS (p))) H 2 (kp (p)|kp )
/S
p∈
H 3 (kS (p)|k),
which shows that H 3 (kS (p)|k) = 0. Next we consider the Hochschild-Serre

spectral sequence for the tower kS (p)|kSR (p)|k. By (10.5.8) for T = S, the
groups
H j (kS (p)|kSR (p))
are induced G(kSR (p)|k)-modules for j ≥ 1, hence E2ij = 0 for i, j ≥ 1. We
obtain an injection
H 3 (kSR (p)|k) = E22,0 ,→ H 3 (kS (p)|k) = 0,
implying that cd G(kSR (p)|k) ≤ 2. It remains to show equality. Since δ(S) = 1
and R is finite, we can find a prime p ∈ S r(Sp ∪ S∞ ∪ R) with ζp ∈ kp .
By (10.5.9), the decomposition group of p in G(kSR (p)|k) is isomorphic to
G(kp (p)|kp ), which is a group of cohomological dimension 2 by (7.5.9). This
implies that cd G(kSR (p)|k) = 2.
In order to show the assertion concerning the strict cohomological dimension,
recall that H 2 (G(k(p)|k), Qp /ZZp ) = 0 by (10.2.3) and (10.4.8). Thus we have
the exact sequence (we omit the coefficients Qp /ZZp from the notation)
BA
H 1 (k(p)|k) res R
H 1 (k(p)|kSR (p))G(kS (p)|k) H 2 (kSR (p)|k).

Using (10.5.8), we obtain a canonical isomorphism

R
H 1 (k(p)|kSR (p))G(kS (p)|k) −→ H 1 (Tp )Gp .
M M
∼ H 1 (Gp ) ⊕
p∈R(k) /S(k)
p∈
By (9.2.7), the restriction map
H 1 (k(p)|k, ZZ/pm ZZ) −→ H 1 (Gp , ZZ/pm ZZ) ⊕ H 1 (Tp , ZZ/pm ZZ)Gp

M M
p∈R(k) /S(k)
p∈
is surjective for every m ≥ 1, except we are in the special case, where the
cokernel is annihilated by 2. Passing to the limit over all m, we see that
coker(res) ∼= H 2 (kSR (p)|k, Qp /ZZp ) is annihilated by 2. Since cd G(kSR (p)|k) = 2,
the group H 2 (kSR (p)|k, Qp /ZZp ) is divisible, hence trivial. This remains true for
every finite extension k 0 of k inside kSR (p), and therefore scd G(kSR (p)|k) = 2
by (3.3.4). 2
By considering p-Sylow subgroups, we obtain the
(10.5.11) Corollary. Let k be a finite number field, p a prime number and

S ⊇ R sets of primes of k. Assume that k is totally imaginary if p = 2. If R is
finite and δ(S) = 1, then
cdp G(kSR |k) = scdp G(kSR |k) = 2.
§6. The Relation between 2 and ∞
Let k be a number field, p a prime number, S a set of primes of k containing

the set Sp of primes above p, and GS (p) = G(kS (p)|k) the Galois group of the
maximal p-extension kS (p) of k which is unramified outside S. If p is odd or
if k is totally imaginary, then GS (p) = GS∪S∞ (p), and the latter group has been
studied in the previous sections. In this section we consider the exceptional
case where p = 2 and k has at least one real place. If S contains all real places,
then these places become complex in kS (2) and therefore GS (2), containing
involutions, has infinite cohomological dimension. Furthermore, the virtual
cohomological dimension vcd GS (2) is less than or equal to two in this case,
i.e. GS (2) has an open subgroup U with cd U ≤ 2. The case when not all real
places are in S will be considered below. The results are taken from [197].

§6. The Relation between 2 and ∞ 657
(10.6.1) Theorem. Let k be a number field and let S be a set of primes

of k which contains all primes dividing 2. If no real prime is in S, then
cd GS (2) ≤ 2. If S contains real primes, then they become complex in kS (2)
and cd GS (2) = ∞, vcd GS (2) ≤ 2.
If S is finite, then H i (GS (2), ZZ/2ZZ) is finite for all i and
χ2 (GS (2)) = −r2 ,
where χ2 (GS (2)) = 2i=0 (−1)i dimF2 H i (GS (2), ZZ/2ZZ) is the second partial
P
Euler characteristic and r2 is the number of complex places of k.
The key to the proof of (10.6.1) is the following theorem (10.6.2) in the case
p = 2 and T = S ∪ SR , where SR is the set of real places of k. Theorem (10.6.2)
is the number theoretical analogue of Riemann’s existence theorem which we
proved in §5 under the assumption that p is odd or that S contains SR . Again
we often write H ∗ (−) for H ∗ (−, ZZ/2ZZ).
(10.6.2) Riemann’s Existence Theorem. Let k be a number field, p a prime

number and T ⊇ S ⊇ Sp sets of primes of k. Then the canonical homomorphism
∗
p ∈ T \S(kS (p))
T (kp (p)|kp ) −→ G(kT (p)|kS (p))
is an isomorphism. Here T (kp (p)|kp ) is the inertia subgroup of G(kp (p)|kp ).
In the proof of (10.6.2) we will use the following
(10.6.3) Lemma. Let p be a prime number, k a number field, and

Sp ⊆ S ⊆ S0 ⊆ S 00
sets of primes in k. Let K be a p-S-closed extension of k. Then the following
assertions are equivalent:
(i) The natural homomorphism
φS 00 ,S : ∗
p∈S 00 \S(K)
T (Kp (p)|Kp ) → G(KS 00 (p)|K)
is an isomorphism.
(ii) The natural homomorphisms
φS 0 ,S : ∗
p∈S 0 \S(K)
T (Kp (p)|Kp ) → G(KS 0 (p)|K)
and
φS 00 ,S 0 : ∗
p∈S 00 \S 0 (KS 0 (p))
T (Kp (p)|Kp ) → G(KS 00 (p)|KS 0 (p))
are isomorphisms.

Proof: The homomorphisms φS 00 ,S , φS 00 ,S 0 and φS 0 ,S are obviously surjec-

tive. Furthermore, all primes p ∈ S 00 r S 0 (K) are completely decomposed in
KS 0 (p)|K because these primes are unramified and K contains the cyclotomic
ZZp -extension of k, i.e. Kp has no nontrivial unramified p-extensions for p ∈/ Sp .
Let us assume that φS 00 ,S is an isomorphism. By dividing out the inertia
groups of the primes in S 00 r S 0 , we conclude that φS 0 ,S is also an isomorphism.
Using the subgroup theorem for free products (4.2.1), one verifies that for each
finite extension K 0 of K inside KS (p) the homomorphism
∗ T (Kp (p)|Kp ) −→ G(KS 00 (p)|K 0 )
p∈S 00 \S 0 (K 0 )
is injective. Passing to the limit over all K 0 , we conclude that φS 00 ,S 0 is injective,
hence an isomorphism. This shows that (i) implies (ii).
Now assume that (ii) holds, i.e. that φS 00 ,S 0 and φS 0 ,S are isomorphisms. By
(1.6.15) and (4.3.14), we have to show that the induced homomorphism
M0
H i (φS 00 ,S ) : H i (G(KS 00 (p)|K)) −→ H i (T (Kp (p)|Kp ))
p∈S 00 \S(K)
is bijective for i = 1 and injective for i = 2. As φS 0 ,S is an isomorphism and
all primes p ∈ S 00 r S 0 (K) are completely decomposed in KS 0 (p)|K, the short
exact sequence of pro-p-groups
1 → G(KS 00 (p)|KS 0 (p)) → G(KS 00 (p)|K) → G(KS 0 (p)|K) → 1
splits. Therefore, in the associated Hochschild-Serre spectral sequence, the
differential
d0,1 0,1 1
2 : E2 = H (G(KS 00 (p)|KS 0 (p)))
G(KS 0 (p)|K)
→ E22,0 = H 2 (G(KS 0 (p)|K))
is zero by (2.4.4).
Since φS 00 ,S 0 is an isomorphism and all primes p ∈ S 00 r S 0 (K) are completely
decomposed in KS 0 (p)|K, we have
M0
H i (G(KS 00 (p)|KS 0 (p))) ∼
= IndG(K 0 (p)|K)
S
H i (T (Kp (p)|Kp ))
p∈S 00 \S 0 (K)
for i ≥ 1. This implies that E2ij = 0 for i, j ≥ 1. For i ≥ 1, we obtain the

HIGFEDC
H i (φS 0 ,S ) M0
H i (G(KS 0 (p)|K)) H i (T (Kp (p)|Kp ))
p∈S 0 \S(K)
αi
H i (φS 00 ,S ) M0
H i (G(KS 00 (p)|K)) H i (T (Kp (p)|Kp ))
p∈S 00 \S(K)
βi
H i (φS 00 ,S 0 ) M0
H i (G(KS 00 (p)|KS 0 (p)))G(KS0 (p)|K) H i (T (Kp (p)|Kp )).
p∈S 00 \S 0 (K)

The map α1 is injective, and since d0,1 2 = 0, the map β1 is surjective and α2
is injective. Hence H (φS 00 ,S ) is an isomorphism and H 2 (φS 00 ,S ) is injective.
1
Therefore φS 00 ,S is an isomorphism. 2
Proof of (10.6.2): If p is odd or if p = 2 and S ⊇ SR , then (10.6.2) is just

theorem (10.5.1). So we assume that p = 2 and S /⊇ SR .
We show (10.6.2) first in the special case that T = S2 ∪ SR , S = S2 . The
∗
groups p∈SR (kS2 (2)) T (kp (2)|kp ) and G(kS2 ∪SR (2)|kS2 (2)) are both generated by
involutions and so their maximal abelian factor groups are annihilated by 2.
Since H 2 (T (kp (2)|kp ), Q2 /ZZ2 ) = 0 for every p ∈ SR (kS2 (2)), we have
H 2( ∗
p∈SR (kS2 (2))
T (kp (2)|kp ), Q2 /ZZ2 ) = 0.
By (10.4.8), the inflation map

H 2 (G(kS2 ∪SR (2)|kS2 (2)), Q2 /ZZ2 ) −→ H 2 (G(kS2 ∪SR |kS2 (2)), Q2 /ZZ2 )
is an isomorphism and, since kS2 (2) contains the cyclotomic ZZ2 -extension
k∞ (2) of k, the validity of the weak Leopoldt conjecture for the cyclotomic
ZZp -extension (see (10.3.25)) implies (by (10.3.22)) that
H 2 (G(kS2 ∪SR (2)|kS2 (2)), Q2 /ZZ2 ) = 0.
By (1.6.16) and the calculation of the cohomology of free products (see
(4.3.14)), it therefore suffices to show that the natural map
M0
H 2 (φ) : H 2 (G(kS2 ∪SR (2)|kS2 (2)) → H 2 (T (kp (2)|kp ))
p∈SR (kS2 (2))
is an isomorphism.
Now let K be a finite extension of k inside kS (2). The long exact sequence
of Poitou-Tate induces the exact sequence
0 → X2 (KS2 ∪SR , ZZ/2ZZ) → H 2 (G(kS2 ∪SR |K)) →
H 2 (G(k̄p |Kp )) → H 0 (G(kS2 ∪SR |K), µ2 )∨ → 0 .
M
p∈S2 ∪SR (K)
Furthermore, we have
X2 (KS2 ∪SR , ZZ/2ZZ) ∼
= X1 (KS2 ∪SR , µ2 )∨ = ClS2 (K)/2.
For a finite, nontrivial extension K 0 of K inside kS2 (2), the corresponding
homomorphism
H 0 (G(kS2 ∪SR |K), µ2 )∨ → H 0 (G(kS2 ∪SR |K 0 ), µ2 )∨
is the dual of the norm map, hence trivial. Furthermore, by (7.1.8)(i),
H 2 (G(k̄p |kS2 (2)p )) = 0 for p ∈ S2 (kS2 (2)).

Therefore we obtain the following exact sequence in the limit over all finite
subextensions K|k inside kS2 (2)|k:
M0
0 → ClS2 (kS2 (2))/2 → H 2 (G(kS2 ∪SR |kS2 (2))) → H 2 (G(k̄p |kp )) → 0.
p∈SR (kS2 (2))
The principal ideal theorem implies that ClS2 (kS2 (2))/2 = 0. Furthermore,
G(k̄p |kp ) = T (kp (2)|kp ) for p ∈ SR (kS2 (2)), and the inflation map
H 2 (G(kS2 ∪SR (2)|kS2 (2))) −→ H 2 (G(kS2 ∪SR |kS2 (2)))
is an isomorphism, see (10.4.8). This concludes the proof in the case when
T = S2 ∪ SR , S = S2 . Furthermore, the assertion holds for T = {all primes}
and S = S2 ∪ S∞ by (10.5.1). Applying (10.6.3), we obtain the assertion in
the ‘extremal’ case T = {all primes}, S = S2 . Applying (10.6.3) again, we
obtain the case T = {all primes} and S arbitrary, and then the general case.
This concludes the proof of (10.6.2). 2
A straightforward limit process shows the following variant of (10.6.2).
(10.6.4) Corollary. Let k be a number field, p a prime number and T ⊇ S ⊇ Sp

sets of primes of k. Let K be a p-S-closed extension field of k. Then the
∗
p ∈ T \S(K)
T (Kp (p)|Kp )−→G(KT (p)|K)
is an isomorphism.
Proof of (10.6.1): We may assume that S /⊇ SR and we investigate the

Hochschild-Serre spectral sequence
E2ij = H i (GS (2), H j (G(kS∪SR (2)|kS (2))) ⇒ H i+j (GS∪SR (2)),
where we omit the coefficients ZZ/2ZZ = µ2 from the notation. By (10.5.1), we
have complete control over the GS (2)-modules H j (G(kS∪SR (2)|kS (2)), which
are isomorphic to
H j (G(C|R))
M
IndGS (2)
p∈SR \S(k)
for j ≥ 1. In particular, E2ij

= 0 if i, j ≥ 1. Therefore the spectral sequence
induces an exact sequence
M
0 → H 1 (GS (2)) → H 1 (GS∪SR (2)) → H 1 (G(C|R)) →
p∈SR \S(k)
(∗)
M
H 2 (GS (2)) → H 2 (GS∪SR (2)) → H 2 (G(C|R)) → 0
p∈SR \S(k)

and exact sequences

0 → H i (GS (2)) → H i (GS∪SR (2)) → H i (G(C|R)) → 0.
M
(∗∗)
p∈SR \S(k)
for i ≥ 3. If S is finite, this shows the finiteness statement for the cohomology
of GS (2) and that
χ2 (GS (2)) = χ2 (GS∪SR (2)).
But χ2 (GS∪SR (2)) = χ2 (GS∪SR ) = −r2 by (8.7.5) and (10.4.8).
For arbitrary S and i ≥ 3, the restriction map
H i (GS∪SR (2)) → H i (G(C|R))
M
p∈SR (k)
is an isomorphism, see (8.6.10)(ii) and (10.4.8). This, together with the exact
sequence (∗∗), shows that the natural homomorphism
H i (GS (2)) → H i (G(C|R))
M
p∈S∩SR (k)
is an isomorphism for i ≥ 3. Therefore cd GS (2) ≤ 2 if S ∩ SR = ∅.

In order to conclude the proof of (10.6.1), it remains to show that every real
prime in S ramifies in kS (2). Let S f be the subset of nonarchimedean primes
in S. Then (10.5.1) yields an isomorphism
∗ T (kp (2)|kp ) ∼
p∈SR (kS f (2))
= G(kS (2)|kS f (2))
which shows the required assertion. This finishes the proof of (10.6.1). 2
For later use we formulate the result just obtained in the proof of (10.6.1) as
a proposition.
(10.6.5) Proposition. Let k be a number field and let S ⊇ S2 be a set of primes.

Then the natural homomorphism
H i (GS (2), ZZ/2ZZ) → H i (G(C|R), ZZ/2ZZ)
M
p∈S∩SR (k)
is an isomorphism for i ≥ 3.
To fix conventions, we recall the following notation: For a set S of primes

×
of k, the group Ok,S of S-units is defined as the subgroup of k × whose
elements are units at every finite prime not in S and positive at every real prime
not in S. The S-ideal class group Cl0,S (k) in the narrow sense of k is the
quotient of the group of fractional ideals of k by the subgroup generated by
the nonarchimedean primes in S and the principal ideals (a) with a positive at
every real place of k not contained in S. In particular, Cl0,∅ (k) = Cl0 (k) is

the ideal class group in the narrow sense and Cl0,S∪SR (k) = ClS (k) is the usual
S-ideal class group. By class field theory, Cl0,S (k) is isomorphic to the Galois
group of the maximal abelian extension of k which is unramified outside SR
and in which every prime in S splits completely.
Let S be a set of finite primes of k and Σ = S ∪ SR . If S is finite, then the
image of the group of Σ-units of k under the logarithm map
×
M
Log : Ok,Σ −→ R, a 7→ (log |a|v )v∈S
v ∈Σ
is a lattice of rank equal to #S + r1 + r2 − 1 by Dirichlet’s unit theorem.
Complementary to this map is the signature map (which is also defined for
infinite S) M ×
×
Signk,S : Ok,Σ −→ R /R×2 .
v ∈SR
More or less by definition, there exists a five-term exact sequence
× ×
R× /R×2 → Cl0,S (k) → Cl0,Σ (k) → 0,
M
0 → Ok,S → Ok,Σ →
v ∈SR (k)
and so the cokernel of Signk,S measures the difference between the usual S-
ideal class group ClS (k) = Cl0,Σ (k) and that in the narrow sense. Of course
this discussion is void if k is totally imaginary. If K is an infinite extension
of k, we define the signature map
×
R× /R×2
M
SignK,S : OK,Σ −→ lim
−→ 0
k0 v ∈SR (k )
as the limit over the signature maps Signk0 ,S , where k 0 runs through all finite
subextensions k 0 |k of K|k.
Now we consider p-S-closed extensions. If p is odd and K is p-S-closed,
then the group
ClS (K(µp ))(p)(j)G(K(µp )|K)
is trivial for j = 0, −1 by (10.4.7). The corresponding result for p = 2 is the
following
(10.6.6) Theorem. Let k be a number field, S ⊇ S2 a set of primes of k and

K a 2-S-closed extension of k. Then the following hold.
(i) ClS (K(µ4 ))(2) = 0.
(ii) Cl0,S (K)(2) = 0.
Remarks: 1. The triviality of Cl(K)(2), and hence also that of ClS (K)(2),
follows easily from the principal ideal theorem; assertions (i) and (ii) do not.
2. In (i) one can replace K(µ4 ) by any totally imaginary extension of degree 2
of K in KS (2).

Proof: If K is 2-S-closed, then ClS (K)(2) = 0 and so statement (ii) of

(10.6.6) is equivalent to the statement that the signature map SignK is surjective.
By (10.6.1) all real places in S become complex in K. By the principal
ideal theorem, Cl(K)(2) = 0 and so statements (i) and (ii) are trivial if K is
totally imaginary (note that K = K(µ4 ) in this case). So we may assume that
SR (K) =/ ∅ and, by (10.6.1), we may suppose that S ∩ SR = ∅.
Let K 0 = K(µ4 ). Then K 0 is totally imaginary and G = G(K 0 |K) is cyclic
of order 2. Let Σ = S ∪ SR and let KΣ be the maximal (not just the pro-2)
extension of K which is unramified outside Σ. Inspecting the Hochschild-
Serre spectral sequence associated to KΣ |KΣ (2)|K and using (10.4.8), we see
that
H 1 (G(KΣ (2)|K), O× KΣ (2),Σ) = H 1 (G(KΣ |K), O× )(2) ∼ = ClS (K)(2) = 0,
KΣ ,Σ
and the same argument shows that
H 1 (G(KΣ (2)|K 0 ), OK
×
Σ (2),Σ
)∼
= ClS (K 0 )(2).
Next we consider the Hochschild-Serre spectral sequence for the extension
×
KΣ (2)|K 0 |K and the module OK Σ (2),Σ
. Using the two isomorphisms obtained
above, we obtain an exact sequence
× φ
0 → ClS (K 0 )(2)G → H 2 (G, OK 2 ×
0 ,Σ ) → H (G(KΣ (2)|K), OK (2),Σ ).
Σ
In order to prove assertion (i), it suffices to show that φ is injective. Let c

be a generator of the cyclic group H 2 (G, ZZ). For each prime p ∈ SR (K)
(respectively for the chosen prolongation of p to KΣ (2)), the composition
Tp (KΣ (2)|K) → G(KΣ (2)|K) → G is an isomorphism, and we denote the
image of c in H 2 (Tp (KΣ (2)|K), ZZ) by cp . The cup-product with c induces
× ∼ ×
an isomorphism Ĥ 0 (G, OK 2
0 ,Σ ) → H (G, OK 0 ,Σ ), and the analogous statement
holds for each cp , p ∈ SR (K).

×
The quotient OK Σ (2),Σ
/µ2∞ is uniquely 2-divisible, and so we obtain a natural
isomorphism
∼ ×
H 2 (G(KΣ (2)|K), µ2∞ ) −→ H 2 (G(KΣ (2)|K), OK Σ (2),Σ
).
Furthermore, for each p ∈ SR r S, we obtain an isomorphism
∼ ×
H 2 (Tp (KΣ (2)|K), µ2∞ ) → H 2 (Tp (KΣ (2)|K), OK Σ (2),Σ
)
∼
= H 2 (G(K p |Kp ),K ×
p ).
Therefore the calculation of the cohomology in dimension i ≥ 2 of free products

with values in torsion modules, see (4.1.4), and (10.5.1) for the pair (Σ, S)
show that we have a natural isomorphism
∼ M0
×
H 2 (G(KΣ (2)|K), OK Σ (2),Σ
) −→ H 2 (G(K̄p |Kp ),K ×
p ).
p∈SR (K)
(Alternatively, we could have obtained this isomorphism from the calculation
of the cohomology of the Σ-units, cf. (8.3.11)(iii), by passing to the limit over

all finite extensions of k inside K.) We obtain the following commutative

diagram
M0
Ĥ 0 (G, NLMKJOK
× ψ
0 ,Σ ) Ĥ 0 (G(K p |Kp ),K ×
p)
p∈SR (K)
L0
∪c ∪ cp
φ M0
× ×
H 2 (G, OK 0 ,Σ ) H 2 (G(KΣ (2)|K), OK Σ (2),Σ
) H 2 (G(K p |Kp ),K ×
p ).
p∈SR (K)
Hence ker(φ) ∼ = ker(ψ) and coker(φ) ∼ = coker(ψ). Since Ĥ 0 (G, OK ×

0 ,Σ ) =
× ×
OK,Σ /NK 0 |K (OK 0 ,Σ ), each element in ker(ψ) is represented by an S-unit in K,
0 ×
√ to show that all these are norms of Σ-units in K . Let e OK,S .
and we have ∈
Then K( e )|K is a 2-extension which is unramified outside S, hence trivial.
×
Therefore e is a square in K and if f 2 = e, then f ∈ OK,Σ and e = NK 0 |K (f ).
This concludes the proof of assertion (i).
To show assertion (ii), it remains to show that coker(SignK,S ) = coker(ψ) ∼ =
coker(φ) is trivial. Using the same spectral sequence as before, in order to see
that coker(φ) = 0, it suffices to show that the spectral terms
×
- E202 = H 0 (G, H 2 (G(KΣ (2)|K 0 ), OK Σ (2),Σ
)) and
- E211 = H 1 (G, ClS (K 0 )(2))
are trivial. The first assertion is easy, because K 0 is totally imaginary and
contains k∞ (2) and so
H 2 (G(KΣ (2)|K 0 ), OK
×
Σ (2),Σ
) = 0.
That the second spectral term is trivial follows from (i). This completes the
proof of (10.6.6). 2
Finally, we consider the full extension kS , i.e. the maximal extension of k

which is unramified outside S, and its Galois group GS = G(kS |k).
(10.6.7) Theorem. Let k be a number field and S a set of primes of k

containing all primes dividing 2. Then vcd2 GS ≤ 2 and cd2 GS ≤ 2 if and only
if S contains no real primes. For every discrete GS (2)-module A, the inflation
maps
inf : H i (GS (2), A) −→ H i (GS , A)(2)
are isomorphisms for all i ≥ 1.

Remark: If cd GS (K)(2) = 2 (e.g. if K contains at least two primes dividing 2,

see (10.9.2) for some finite extension K of k inside kS , then vcd2 GS = 2. This
is always the case if S ⊇ SR because the class numbers of the cyclotomic fields
Q(µ2n ) are nontrivial for n 0. But, for example, we do not know whether
cd2 G(QS2 |Q) equals 1 or 2. The answer would be ‘2’ if at least one of the real
cyclotomic fields Q(µ2n )+ , n = 2, 3, . . ., had a nontrivial class number. But this
is unknown.
Proof: The statement on cd2 GS and vcd2 GS follows by choosing a 2-Sylow

subgroup H ⊆ GS and applying theorem (10.6.1) to all finite extensions of k
inside (kS )H . It remains to show the statement about the inflation map. It is
equivalent to the statement that
inf ⊗ ZZ(2) : H i (GS (2), A) ⊗ ZZ(2) −→ H i (GS , A) ⊗ ZZ(2)
is an isomorphism for every discrete GS (2)-module A and all i ≥ 0, where ZZ(2)
denotes the localization of ZZ at the prime ideal (2).
Since cohomology commutes with inductive limits, we may assume that A
is finitely generated as a ZZ-module. Using the exact sequences
0 −→ tor(A) −→ A −→ A/tor(A) −→ 0,
0 −→ A/tor(A) −→ (A/tor(A)) ⊗ Q −→ (A/tor(A)) ⊗ Q/ZZ −→ 0,
and using the limit argument for (A/tor(A)) ⊗ Q/ZZ again, we are reduced to
the case that A is finite. Every finite GS (2)-module is the direct sum of its
2-part and its prime-to-2-part. The statement is obvious for the prime-to-2-
part and every finite 2-primary GS (2)-module has a composition series whose
quotients are isomorphic to ZZ/2ZZ. Therefore we are reduced to showing the
statement about the inflation map for A = ZZ/2ZZ. But it is more convenient to
work with A = Q2 /ZZ2 (with trivial action), which is possible because of the
exact sequence
0 −→ ZZ/2ZZ −→ Q2 /ZZ2 −→ Q2 /ZZ2 −→ 0.
Using the Hochschild-Serre spectral sequence for the extensions kS |kS (2)|k,
we thus have to show that
H i (G(kS |kS (2)), Q2 /ZZ2 ) = 0
for i ≥ 1. The case i = 1 is obvious by the definition of the field kS (2). By
theorem (10.6.1), every real prime in S becomes complex in kS (2) and therefore
cd2 G(kS |kS (2)) ≤ 2. It remains to show that H 2 (G(kS |kS (2)), Q2 /ZZ2 ) = 0.
Therefore the next proposition implies the remaining statement of (10.6.7).
(10.6.8) Proposition. Let k be a number field, S ⊇ S2 a set of primes in k and

K ⊇ k∞ (2) an extension of K inside kS . Then
H 2 (G(kS |K), Q2 /ZZ2 ) = 0.

Proof: Let H be a 2-Sylow subgroup of G(kS |K) and L = (kS )H . Then the
restriction map
H 2 (G(kS |K), Q2 /ZZ2 ) −→ H 2 (G(kS |L), Q2 /ZZ2 )
is injective and so, replacing K by L, we may suppose that kS = KS (2).
Applying (10.6.4) to the 2-S-closed field KS (2), we obtain an isomorphism
G(KS∪S (2)|KS (2)) ∼
R = ∗
Tp (Kp (2)|Kp ).
p∈SR (KS (2))
Hence we have complete control over the Hochschild-Serre spectral sequence
associated to KS∪SR (2)|KS (2)|K. Furthermore, the weak Leopoldt conjecture
holds for the cyclotomic ZZ2 -extension and K ⊇ k∞ (2), which implies that
H 2 (G(KS∪SR (2)|K), Q2 /ZZ2 ) = 0. The exact sequence (∗) in the proof of
(10.6.1), applied to all finite subextensions k 0 |k of K|k, yields a surjection
M0
H 1 (T (Kp (2)|Kp ), Q2 /ZZ2 ) −
H 2 (G(KS (2)|K), Q2 /ZZ2 )
p∈SR \S(K)
and therefore, in order to prove the proposition, it suffices to show that the group
H 2 (G(KS (2)|K), Q2 /ZZ2 ) is 2-divisible. This is trivial if S ∩ SR (K) = ∅ be-
cause then cd G(KS (2)|K) ≤ 2. Otherwise, this follows from the commutative
diagram
OPQR
H 2 (G(KS (2)|K), Q2 /ZZ2 )/2 H 3 (G(KS (2)|K), ZZ/2ZZ)
M0 M0
H 2 (T (Kp (2)|Kp ), Q2 /ZZ2 )/2 H 3 (T (Kp (2)|Kp ), ZZ/2ZZ).
p∈S∩SR (K) p∈S∩SR (K)
The right hand vertical arrow is an isomorphism by (10.6.5). But the group
H 2 (T (Kp (2)|Kp ), Q2 /ZZ2 ) is trivial for all p ∈ S ∩ SR (K) and therefore the
object in the lower left corner is zero. 2
§7. Dimension of H i(GST , ZZ/pZZ)
Throughout this section k denotes a global field and p =/ char k is a fixed

prime number. Let S and T be sets of primes of k, which are allowed to be
empty. We want to calculate the dimensions
hi (GTS ) = dimIFp H i (GTS , ZZ/pZZ) and hi (GTS (p)) = dimIFp H i (GTS (p), ZZ/pZZ)
for i = 1, 2, where GTS = GTS (k) is the Galois group G(kST |k) and
kST is the maximal extension of k which is unramified outside S
and completely decomposed at T .

§7. Dimension of H i (GST , ZZ/pZZ) 667
Let us recall some conventions. If k is a number field, then

Sp = Sp (k) = {p a prime of k dividing p} .
We denote the set of real primes and of complex primes of k by SIR and SC ,
respectively, and we put r1 = #SIR and r2 = #SC .
If k is a function field, Cl(k) denotes the group of divisor classes of k and
Ok = Ok,∅ is the finite field of constants of k. The sets Sp , S∞ , SIR and SC are
empty and r1 = r2 = 0.
We introduce the following notation:
(10.7.1) Definition. We set

( (
1, µp ⊆ k, 1, µp ⊆ kp ,
δ= and δp =
/ k,
0, µp ⊆ / kp ,
0, µp ⊆
where p is a prime of k.
Recall the group

VS (k) = VS (k, p) = ker k × /k ×p −→ kp× /kp×p × kp× /Up kp×p ,
Y Y
p∈S /S
p∈
which was defined in IX §1 and shown to be finite in (9.1.12). For S = ∅ we
have the
(10.7.2) Proposition. For p =/ char k there is an exact sequence

0 −→ Ok× /p −→ V∅ (k) −→ p Cl(k) −→ 0 .
In particular, the dimension of B∅ (k) is given by
dimIFp B∅ (k) = dimIFp p Cl(k) + dimIFp Ok× /p .
Proof: The homomorphism

{a ∈ k × |a ∈ Up kp×p for all p} −→ p Cl(k) , a 7−→ a with (a) = ap ,
induces a surjection from V∅ (k) onto p Cl(k) whose kernel is isomorphic to
Ok× /p. 2
Remark: For arbitrary S, the group VS (k) is a subgroup of V∅ (k), and we

obtain dimIFp BS (k) ≤ dimIFp p Cl(k) + dimIFp Ok× /p.
In the following we denote H i (−, ZZ/pZZ) by H i (−) and put hi (−) =

dimIFp H i (−). For a prime p of k, we denote by Gp the absolute Galois group
of field kp and by Tp the inertia group in Gp .

Using the Poitou-Tate theorem, we now determine hi (GS ) for a nonempty

set S which contains the archimedean primes and the primes above p if k is a
number field.
(10.7.3) Theorem. Let k be a global field, p =/ char k a prime number and S a

nonempty set of primes of k containing Sp ∪ S∞ if k is a number field. Then
X
h1 (GS ) = 1 + δp − δ + dimIFp BS (k) ,
p∈ S
X
h2 (GS ) = δp − δ + dimIFp BS (k) .
p∈S\SC
In particular, h1 (GS ) and h2 (GS ) are finite if and only if S contains only a
finite number of primes p with δp = 1. If µp ⊆ k we have dimIFp BS (k) =
dimIFp ClS (k)/p.
Proof: The formula for h2 follows from the exact sequence
X2 (GS , ZZ/pZZ) ,→ H 2 (GS , ZZ/pZZ) → H 2 (kp , ZZ/pZZ) H 0 (GS , µp )∨

M
p∈S
and the isomorphism

X2 (GS , ZZ/pZZ) ∼
= X1 (GS , µp )∨ = BS (k) .
The formula for h1 is then a consequence of the Euler-Poincaré characteristic
formula (8.7.5). For the last statement, observe that when µp ⊆ k
BS (k) = ker(H 1 (GS , µp ) → H 1 (kp , µp ))∨
Q
p∈S
= (ClS (k)/p)(−1) . 2
In order to proceed, we need some exact sequences.
(10.7.4) Lemma. Let k be a global field, p =/ char k a prime number and S ⊇ T

sets of primes of k. Then there are exact sequences
H 1 (Tp )Gp → BT (k) → BS (k) → 0.

M
(i) 0 → H 1 (GT ) → H 1 (GS ) →
p∈S\T
(Gp ) → H 1 (Gk , µp )∨ BS (k).

Y Y
(ii) H 1 (GS ) ,→ H 1 (Gp ) × 1
Hnr
p∈S /S
p∈
Proof: We consider the commutative exact diagram

H 1 (GSTUVWXYZ[\ kS )GS H 1 (Tp )Gp

M
p∈/S
Y
H 1 (Gk ) H 1 (Gp ) H 1 (Gk , µp )∨
p
Y Y
H 1 (GS ) H 1 (Gp ) × 1
Hnr (Gp ) H 1 (Gk , µp )∨ BS (k) .
p∈S /S
p∈
The upper horizontal map is injective since GkS = G(k̄|kS ) is generated by

the inertia groups of the primes outside S in Gk . The exactness of the mid-
dle row follows from the long exact sequence of Poitou-Tate together with
X1 (Gk , ZZ/pZZ) = 0 = X1 (Gk , µp ), see (9.1.9). The exactness of the bottom
row is just the definition of BS (k) = VS (k)∨ as Hnr
1
(Gp )∨ = kp× /Up kp×p . By dia-
gram chasing, we obtain exact sequence (ii). Thus the rows in the commutative
diagram
]^_àbcdefg H 1 (Tp )Gp
M
p∈S\T
Y Y
H 1 (GS ) H 1 (Gp ) × 1
Hnr (Gp ) H 1 (Gk , µp )∨ BS (k)
p∈S /S
p∈
Y Y
H 1 (GT ) H 1 (Gp ) × 1
Hnr (Gp ) H 1 (Gk , µp )∨ BT (k)
p∈T /T
p∈
are exact. Using the snake lemma, we obtain exact sequence (i). 2
In the function field case, we have the following result in the case S = ∅.
(10.7.5) Theorem. Let k be a global field of positive characteristic and let

p =/ char k be a prime number. Let g = g(k) be the genus. Then hi (G∅ ) = 0
for i ≥ 4 and
h1 (G∅ ) = 1 + dimIFp p Cl(k)
(
0, if g = 0,
h2 (G∅ ) =
dimIFp p Cl(k) + δ, if g ≥ 1,
(
0, if g = 0,
h3 (G∅ ) =
δ, if g ≥ 1.

Proof: If g = 0, then p Cl(k) = 0 and G∅ ∼ = ẐZ by (10.1.3)(i), and the

assertions made are obvious. From now on we assume g ≥ 1. Let S be an
arbitrary nonempty finite set of primes of k. Using (10.7.3) and the exact
sequence (10.7.4)(i), we obtain h1 (G∅ ) = 1 − δ + dimIFp B∅ (k) . Therefore
(10.7.2) shows the statement for h1 . By (10.1.3)(ii), we obtain hi (G∅ ) = 0
for i ≥ 4 and h3 (G∅ ) = h0 (G∅ , µp ) = δ. It remains to show the statement
about h2 .
Let κ be the finite field of constants of k and Γ = G(κ̄|κ) ∼
= ẐZ. We consider
the Hochschild-Serre spectral sequence
E2ij = H i (Γ, H j (G∅ (k κ̄))) ⇒ H i+j (G∅ ) ,
which degenerates at E2 since cd ẐZ = 1. Consider the exact sequence
Frob−1
H j (G∅ (k κ̄))Γ ,→ H j (G∅ (k κ̄)) −→ H j (G∅ (k κ̄)) H j (G∅ (k κ̄))Γ .
Since the groups H j (G∅ (k κ̄)) are finite and
H 1 (Γ, H j (G∅ (k κ̄))) ∼
= H j (G∅ (k κ̄))Γ
by (1.7.7), we obtain
dimIFp H 0 (Γ, H j (G∅ (k κ̄))) = dimIFp H 1 (Γ, H j (G∅ (k κ̄)))
for j ≥ 0. The spectral sequence implies
3
hi (G∅ ) = 0,
X
i=0
and the result for h2 follows from the calculation of hi for i =/ 2. 2
It remains to consider the group GS in the number field case with an arbitrary
set S. The following results are due to I. R. ŠAFAREVIČ and H. KOCH, see
[191] and [110].
For an arbitrary set S of primes of k we define the Šafarevič-Tate groups
Xi (GS ) = Xi (GS , ZZ/pZZ) by the exactness of the sequences
0 −→ Xi (GS ) −→ H i (GS ) −→ H i (Gp ) .

Y
p∈S
If p =/ char k, and S nonempty containing Sp ∪ S∞ in the number field case,

then we have isomorphisms of finite abelian groups
X2 (GS ) ∼
= X1 (GS , µp )∨ ∼
= BS (k) ,
by (8.6.7) and (9.1.12)(iii). Also, for arbitrary sets S we can relate X2 (GS ) to
the group BS (k).

(10.7.6) Theorem (H. KOCH). Let k be a global field and let p =/ char k be
a prime number. Then, for any set S of primes of k, there exists a natural
injection
h S)
X2 (G BS (k).
In particular, X2 (GS ) is finite.
Proof: Consider the commutative diagram

0ijklmnop H 1 (GS ) H 1 (Gk ) H 1 (G(k̄|kS ))GS H 2 (GS ) H 2 (Gk )
M M
H 2 (Gp ) H 2 (Gp ).
p∈S p
The row is the 5-term sequence associated to the group extension

1 −→ G(k̄|kS ) −→ Gk −→ GS −→ 1,
hence exact. The right-hand vertical map is injective by (9.1.10)(i). Thus the
sequence
0 → H 1 (GS ) → H 1 (Gk ) → H 1 (G(k̄|kS ))GS → H 2 (GS ) →
M
H 2 (Gp )
p∈S
is exact. Using (10.7.4)(i) (where S := {all primes} and T := S) and the

vanishing of B{all primes} (k) (see (9.1.12)(ii)), we obtain a commutative and
exact diagram
qrstuvwxy k ) H 1 (Tp (k))Gp (k)
M
H 1 (G BS (k) 0
/S
p∈
H 1 (Gk ) H 1 (G(k̄|kS ))GS X2 (GS ) 0.

Thus the desired result follows. 2
Remark: If S = ∅, then (10.7.6) can be deduced in a convenient way from

duality theorems for arithmetic curves: If k is a function field, let X be the
regular projective curve with function field k. Étale Poincaré duality shows
2
Het (X, ZZ/pZZ) ∼
= Het 1
(X, µp )∨ ∼
= B∅ (k).
If k is a number field (and p =/ 2), let X = Spec(Ok ). The flat duality theorem
of Artin-Mazur shows that
H 2 (Xet , ZZ/pZZ) ∼
= H 2 (Xfl , ZZ/pZZ) ∼
= H 1 (Xfl , µp )∨ ∼
= B∅ (k).
In both cases, the Hochschild-Serre spectral sequence yields an inclusion
X2 (G∅ ) = H 2 (G∅ , ZZ/pZZ) ,→ H 2 (Xet , ZZ/pZZ) .

(10.7.7) Corollary. Let S be a set of primes of the global field k and let
p =/ char k be a prime number. Then there is an equality
X X
h1 (GS ) = 1 + δp − δ + dimIFp BS (k) + np − r ,
p∈S\SC p∈S∩Sp
where np = [kp : Qp ] is the local degree with respect to p ∈ Sp and r = r1 + r2

denotes the number of archimedean primes. We have an inequality
X
h2 (GS ) ≤ δp − δ + dimIFp BS (k) + θ ,
p∈S\SC
where θ = θ(k, S) is equal to 1 if δ = 1 and S ⊆ S∞ if p =/ 2 (resp. S ⊆ SC if

p = 2), and zero in all other cases.
Proof: Let us first assume that δ = 1. Recall that X2 (Gk ) = 0, see (9.1.10).
H 2 (Gp ) −→ H 0 (Gk , µp )∨ −→ 0,
M
0 −→ H 2 (Gk ) −→
p
we see that the map
M
H 2 (Gk ) −→ H 2 (Gp )
p=
/ p0
remains injective, where p0 is an arbitrary nonarchimedean prime or a real

prime if p = 2 (if one exists). Therefore if S is not contained in S∞ if p =/ 2
resp. SC if p = 2, i.e. S = S 0 ∪{p0 } for some prime p0 as above, the commutative
diagram
}{|z S )
H 2 (G
inf
H 2 (Gk )
ϕS 0
M M
H 2 (Gp ) H 2 (Gp )
p∈S 0 p=
/ p0
shows that X2 (GS ) ∼= ker inf ∼

= ker ϕS 0 .
Now let δ be arbitrary. From the definition of X2 (GS ) and the consideration
above, it follows that
δp − δ + dimIFp X2 (GS ) + θ .
X
h2 (GS ) ≤
p∈S\SC
Using (10.7.6), we obtain the result for h2 . The assertion for h1 follows
from the exact sequence (10.7.4)(i) with T = ∅ and (10.7.2) together with
h1 (G∅ ) = dimIFp Cl(k), and the local computations (7.5.9) and (7.5.10). 2
Next we extend our calculations and consider the group GTS , where T and S
are sets of primes. First, we introduce further notation.

(10.7.8) Definition. Let p =/ char k and let S0 ⊆ S be sets of primes of k. Then
VSS0 (k) = {a ∈ k × | a ∈ kv×p for v ∈ S0 and a ∈ Uv kv×p for v ∈

/ S}/k ×p ,
where Uv is the unit group of the local field kv (by convention Uv = kv× if v is
archimedean). The dual group is denoted by BSS0 (k) = VSS0 (k)∨ .
We have canonical inclusions

VS (k) = VSS (k) ⊆ VSS00 (k) ⊆ VSS0 (k).
If S ⊇ Sp ∪ S∞ , then Kummer theory implies

VSS0 (k) = X1 (kS , S0 , µp ) = ker H 1 (GS , µp ) −→
M
H 1 (kv , µp ) .
v ∈S0
(10.7.9) Lemma. Let k be a global field, p =/ char k a prime number and S ⊇ T

sets of primes of k. Then there is an exact sequence
0 → H 1 (GTS ) → H 1 (GS ) → H 1 (Gp ) → BSS\T (k) → BS (k) → 0.

Y
p∈T
Proof: Consider the commutative and exact diagram

~ T )GS
T Y
H 1 (kS |k S H 1 (Gp )
p∈T
Y Y
H 1 (GS ) H 1 (Gp ) × 1
Hnr (Gp ) H 1 (Gk , µp )∨ BS (k)
p∈S p∈/S
BSS\T (k).
Y Y
H 1 (GTS ) H 1 (Gp ) × Hnr
1
(Gp ) H 1 (Gk , µp )∨
p∈S\T /S
p∈
The upper horizontal map is injective since G(kS |kST ) is generated by the de-
composition groups of the primes in T . The middle row is exact by (10.7.4)(ii).
The exactness of the bottom row is just the definition of BSS\T (k) = VS\T
S
(k)∨ ,
1
as Hnr (Gp )∨ = kp× /Up kp×p and H 1 (Gp )∨ = kp× /kp×p . By diagram chasing, we
obtain the claimed exact sequence. 2
Using (10.7.9), our results on GS extend to GTS . We obtain the

(10.7.10) Theorem. Let S and T be finite sets of primes of the global field k.
Then we have an equality
h1 (GTS ) = 1 + δp − δ + dimIFp BS∪T

X X
S\T (k) + np − r − # T \S∞ .
p∈S\(T ∪SC ) p∈(S\T )∩Sp
Furthermore, we have an inequality

h2 (GTS ) ≤ δp − δ + dimIFp BS∪T
X
S\T (k) + θ ,
p∈S\(T ∪SC )

Proof: Since GTS = GTS∪T , we may assume that T ⊆ S. Now the formula for
h1 (GS ) given in (10.7.7) together with (10.7.9) imply the equality for h1 (GTS ).
Since GTS = GTS\T , we may assume that S ∩ T = ∅ when proving the inequality
for h2 (GTS ). We consider the exact sequence
res T
0 → H 1 (GTS ) → H 1 (GS ) → H 1 (G(kS |kST ))GS → H 2 (GTS ) → H 2 (GS ).
Using (10.7.7), the result for h1 (GTS ) and the injectivity of the map
T
H 1 (G(kS |kST ))GS −→
Y
1
Hnr (Gp ),
p∈T
we obtain
T
h2 (GTS ) ≤ h2 (GS ) − h1 (GS ) + h1 (GTS ) + dimIFp H 1 (G(kS |kST ))GS
h2 (GS ) − h1 (GS ) + h1 (GTS ) + 1
P
≤ p∈T dimIFp Hnr (Gp )
− δ + dimIFp BS∪T 2
P
p∈S\(T ∪SC ) δp S\T (k) + θ .
≤
For the partial Euler-Poincaré characteristic

2
χ2 (GTS ) (−1)i hi (GTS ) ,
X
=
i=0
we obtain the
(10.7.11) Corollary. Let S and T be finite sets of primes of the global field k
and let p =/ char k. Then the groups H i (GTS , ZZ/pZZ), 0 ≤ i ≤ 2, are finite and
χ2 (GTS ) ≤ θ −
X
np + r + # T \S∞ ,
p∈(S\T )∩Sp


Next we consider the maximal pro-p-factor group GTS (p) of GTS . We have
the following equality for the generator rank h1 (GTS (p)) and the inequality for
the relation rank h2 (GTS (p)) of GTS (p).
(10.7.12) Theorem. Let S and T be finite sets of primes of a global field k

and let p =/ char k be a prime number. Then we have an equality
h1 (GTS (p)) = 1 + δp − δ + dimIFp BS∪T

X X
S\T (k) + [kp : Qp ] − r − # T \S∞ ,
p∈S\(T ∪SC ) p∈(S\T )∩Sp
and an inequality
h2 (GTS (p)) ≤ δp − δ + dimIFp BS∪T
X
S\T (k) + θ .
p∈S\(T ∪SC )
Here r = r1 + r2 is the number of archimedean primes and θ = θ(k, S) is equal

to 1 if δ = 1 and S ⊆ S∞ if p =/ 2 (resp. S ⊆ SC if p = 2), and zero in all other
cases. Furthermore,
χ2 (GTS (p)) ≤ θ −
X
np + r + # T \S∞ .
p∈(S\T )∩Sp
Proof: The homomorphism

H i (GTS (p), ZZ/pZZ) −→ H i (GTS , ZZ/pZZ)
is bijective for i = 1 and injective for i = 2, since H 1 (kST |kST (p), ZZ/pZZ) = 0.
Together with (10.7.10) and (10.7.11), this gives the result. 2
We get an equality for the relation rank in the following situation.
(10.7.13) Theorem. Let k be a global field and p =/ char k a prime number.

Let S be a nonempty set of primes of k containing Sp ∪ S∞ if k is a number
field. Then X
h1 (GS (p)) = 1 + δp − δ + dimIFp BS (k) ,
p∈S
X
2
h (GS (p)) = δp − δ + dimIFp BS (k) .
p∈S\SC
In particular, we have χ2 (GS (p)) = −r2 .
Proof: By (10.1.4) and (10.4.8), the inflation maps

H i (GS (p), ZZ/pZZ) → H i (GS , ZZ/pZZ)
are isomorphisms for all i ≥ 0. Therefore we obtain the desired equalities from
(10.7.3). 2

(10.7.14) Corollary. Let k be a global field and p =/ char k a prime number.

Let S be a nonempty set of primes of k containing Sp ∪ S∞ if k is a number
field. Then the pro-p-group GS (p) is free if and only if
X
δp = δ and BS (k) = 0 .
p∈S\SC
If in addition µp ⊆ k, then GS (p) is free if and only if
#S = 1, if k is a function field,

p ClS (k) = 0 and
#(S\SC ) = #Sp = 1, if k is a number field.
Proof: This follows from (3.9.5), (10.7.13) and that BS (k) = (ClS (k)/p)(−1)
if µp ⊆ k. 2
Remarks: 1. If k is a number field and S is a finite set of primes disjoint

from Sp , then GS (k)(p)ab is finite by (10.3.20)(ii). Therefore GS (k)(p) is either
trivial or not free in this case.
2. Let k be a number field, p = 2 and S ⊇ S2 . If S contains a real place, then
GS (2) has infinite cohomological dimension by (10.6.1). If S ∩ SIR = ∅, then
(10.9.2) will give a criterion for the freeness of GS (2) .
3. If k is a function field, then G∅ (k)(p) is free if and only if Cl(k)(p) = 0 (cf.
ex. 2 and 3 in §1).
Finally, we calculate the numbers hi (GS (p)) for the number field k = Q.
(10.7.15) Theorem. Let p be a prime number. Let S be a finite set of places of

Q and GS (p) = G(QS (p)|Q). If p =/ 2 assume that S is minimal in the sense that
it does not contain ∞ and contains only prime numbers ` with ` ≡ 0, 1 mod p.
If S = ∅, then GS (p) = 1, and if S =/ ∅, then the following holds.
(i) If p =/ 2, then
h1 (GS (p)) = #S,
(
#S − 1, if p ∈ S,
h2 (GS (p)) =
#S, if p ∈/ S,
i
h (GS (p)) = 0, if p ∈ S, i ≥ 3.
(ii) If p = 2, then
(
#S, if 2 ∈ S,
h1 (GS (2)) =
#S + α − 1, if 2 ∈/ S,
2
h (GS (2)) = ( #S + α − 1,
1, if ∞ ∈ S and 2 ∈ S,
hi (GS (2)) = i ≥ 3,
0, if ∞ ∈/ S and 2 ∈ S,
where α = 0 if 2 ∈ S, ∞ ∈ S or ` ∈ S for some ` ≡ 3 mod 4, and α = 1 in all
other cases.

Remark: The theorem omits the calculation of hi (GS (p)), i ≥ 3, if p ∈/ S. This

is difficult. There exist (so-called “circular”) finite sets S of prime numbers,
not containing p, such that cd GS (p) = 2. These sets are cofinal among all
finite sets of primes S with p ∈/ S, see [119], [198] (see [199] for the case of a
general number field).
Proof: If S = ∅, then GS (p) is the trivial group as there are no unramified

extensions of Q. We assume that S =/ ∅ and start with the case p =/ 2. Then
BS (Q) = 0 and δ = 0 = δp . Therefore the statement for h1 follows from
(10.7.12). If p ∈ S, then the assertion for h2 follows from (10.7.13). If p ∈/ S,
then (10.7.12) yields the inequality h2 (GS (p)) ≤ #S. From the exact sequence
p
0 → ZZ/pZZ → Q/ZZ → Q/ZZ → 0
we obtain the inclusion
H 1 (GS (p), Q/ZZ)/p ,→ H 2 (GS (p), ZZ/pZZ).
By class field theory, GS (p)ab is finite if p ∈/ S. It follows that
h1 (GS (p)) = dimIFp p H 1 (GS (p), Q/ZZ)

= dimIFp H 1 (GS (p), Q/ZZ)/p
≤ h2 (GS (p)),
hence the above inequality is an equality. Finally, cd GS (p) ≤ 2 if p ∈ S. Thus

the proof of (i) is complete.
Now let p = 2. Then dimIF2 BS (Q) = α (we leave the verification to the
reader, see ex.1). Therefore (10.7.12) implies the statement for h1 . If 2 ∈ S,
then χ2 (GS (2)) = 0 by (10.6.1), showing the statement for h2 . Furthermore, if
2 ∈ S, then (10.6.5) implies the statement for hi , i ≥ 3. There remains the case
2 ∈/ S. By (10.7.12), we obtain the inequality h2 (GS (p)) ≤ #S + α − 1, and the
same argument as in the case p =/ 2 shows that this inequality is an equality.
Exercise 1. Let k = Q. Show that BS (Q, p) = 0 if p =/ 2, and BS (Q, 2) = 0 if and only if S

contains 2, ∞ or a prime number ` ≡ 3 mod 4.
Exercise 2. Let IF be a finite field, k = IF(t) and p =/ char IF a prime number. Show that
BS (k) = 0 for every S if µp is not contained in k, and if #µ(k)(p) = ps , s ≥ 1, then BS (k) = 0
if and only if S contains a prime p0 such that N (p0 ) 6≡ 1 mod ps+1 .

§8. The Theorem of Kuz’min
This section is devoted to the question of which local p-extensions are

globally realized. More exactly, assume k is a number field, p is a prime
number and let S ⊇ Sp ∪ S∞ be a finite set of primes in k. We will exclusively
consider p-extensions in this section, and so we make the following notational
convention:
Unless the contrary is explicitly stated, every field extension is tacitly as-
sumed to be a p-extension. We write kS for kS (p), GS for GS (p), Gk for
Gk (p) and so on. Furthermore, we write ClS (k) for ClS (k)(p) and k∞ for the
cyclotomic ZZp -extension of k.
Recall that every proper subgroup of a pro-p-group is contained in a proper,
normal subgroup, cf. (1.6.14)(i). Therefore a subgroup whose normal closure
is the full group, is the full group itself. We conclude that the vanishing of
ClS (k) implies that GS is generated by the decomposition groups of (arbitrary
chosen prolongations of) the primes in S.
Suppose that we are given a finite prime p ∈ S and a p-extension N |kp . Our
central question is:
“Does there exist an extension K of k inside kS with Kp = N ?”
The answer is in general “no”, as the following examples show.
Example 1. k = Q, p odd, S = Sp ∪ S∞ . Then
GS = G(Q∞ |Q) = ZZp ,
where Q∞ denotes the cyclotomic ZZp -extension of Q. (See the discussion
before (10.5.6)).
On the other hand, we know by (7.5.11) that GQp is a free pro-p-group of
rank 2. Hence not every given local p-extension at the prime p can be globally
realized.
Example 2. k = Q(µp ) with a regular, odd prime number p, S = Sp ∪ S∞ .
In this case there is exactly one prime p ∈ Sp (k) and in the long Poitou-Tate
sequence (recalling (7.5.11)) there occurs an isomorphism
∼ H 0 (G , µ )∨ .
H 2 (kp , ZZ/pZZ) = P 2 (GS , ZZ/pZZ) −→ S p
Using (10.4.8), we obtain

H 2 (GS , ZZ/pZZ) ∼
= X1 (GS , µp )∨ ∼
= (ClS (k)/p) (−1) = 0.
Hence GS is a free pro-p-group of rank (p + 1)/2 (use the Euler-Poincaré
formula (8.7.5)). The group Gkp is a Demuškin group of rank p + 1 by (7.5.11).

§8. The Theorem of Kuz’min 679
Furthermore, since ClS (k) = 0, the canonical homomorphism

Gkp −→ GS
is surjective (see above). From our calculations this map is not an isomorphism,
so also in this example not every local extension is globally realized. As a
by-product, we conclude (see (10.3.6)) that the (strong) Leopoldt conjecture
holds for p and every subextension K ⊆ kS .
We denote the decomposition group of a prime v in GS by Gv and the local

Galois group Gkv by Gv . We write Tv for the inertia subgroup in Gv .
(10.8.1) Theorem. For v ∈ S r Sp , the canonical homomorphism

Gv −→ GS
is injective. If k is not totally real, then the image is not open, i.e. of infinite
index in GS .
Remark: The example k = Q, p = 3, S = {3, 7, ∞} shows that the assumption

that k is not totally real is essential. Indeed, by (10.5.6), we see that GS = G7
in this example.
Proof: If v is archimedean, then the assertion is obvious, therefore we may

assume that v is finite. Since the cyclotomic ZZp -extension of k globally realizes
the maximal unramified extension of kv , the kernel of the homomorphism
Gv → GS must be contained in Tv . Thus this kernel is trivial since Tv maps
injectively to GS by Riemann’s existence theorem (10.5.1) and (4.3.12).
It therefore remains to show the statement about the infinite index. By the
trivial Leopoldt inequality (see (10.3.7)) GS has a ZZp -free abelian quotient of
rank r2 + 1 which is unramified outside p. A decomposition group of a prime
not dividing p therefore has an image of rank 1 in this group. Hence such a
decomposition group cannot be open in GS if r2 > 0. 2
(10.8.2) Theorem. Let k be totally imaginary and let v ∈ S f . Suppose that

Gv is open in GS . Then v ∈ Sp and either Gv = GS or p = 2, (GS : Gv ) = 2,
#S f (k) = 1 and S f (kS ) = 2.
Proof: By theorem (10.8.1), we know that v ∈ Sp . Let k v be the decomposi-

tion field of v in kS , i.e. k v = kSGv . Assume that k v is of finite degree over k.
The trivial Leopoldt inequality (see (10.3.7)) shows that
r2 (k v ) + 1 ≤ rank ZZp GS (k v )ab .

Since k, and hence k v , is totally imaginary, we obtain

[k v : Q]/2 + 1 ≤ rank ZZp GS (k v )ab
= rank ZZp Gab
v
≤ rank ZZ G ab = [kv : Q ] + 1
p v v
≤ [k : Q] + 1.
Hence [k v : k] ≤ 2. Thus it remains to consider the case (GS : Gv ) = 2; in

particular, p = 2. Observe that dimIF2 H 2 (Kp , IF2 ) = 1 for every number field
K and every p ∈ S f (K). If K is totally imaginary, we therefore obtain by
counting dimensions in the long exact Poitou-Tate sequence
dimIF2 H 2 (GS (K), IF2 ) ≥ #S f (K) − 1.
The Euler-Poincaré characteristic formula (8.7.5) yields
(∗) dimIF2 H 1 (GS (K), IF2 ) ≥ r2 (K) + #S f (K).
Applying (∗) to every K ⊆ kS containing k v , we obtain
[Kv : Q2 ] + 2 = dimIF2 H 1 (GKv , IF2 )
≥ dimIF H 1 (Gv (kS |K), IF2 )
2
= dimIF2 H 1 (GS (K), IF2 )
≥ [K : Q]/2 + #S f (K).
By our assumption, v splits in k v |k, so that [Kv : Q2 ] ≤ [K : Q]/2 and

#S f (K) ≥ 2. Comparing with the above inequality, we deduce that #S f (K) = 2
and finally #S f (k) = 1. 2
Next we investigate the decomposition groups of primes dividing p. In order

to simplify notation, we write H i (G) for H i (G, IFp ) if G is a pro-p-group. We
denote the extension of degree pm in the cyclotomic ZZp -extension of a field k
by km . If v ∈ S f (k), then Gv,m := G((kS )v |kv,m ) = G(kS |(k v )m ).
(10.8.3) Proposition. Assume SIR (k) = ∅ if p = 2 and let v ∈ Sp . Then

(i) (a) cdp Gv ≤ 2,
(b) dimIFp H i (Gv ) < ∞ for all i.
Furthermore, if µp ⊆ k, then the following is true.
(ii) dimIFp H 2 (Gv,m ) is bounded for m → ∞.
(iii) scdp Gv = 2.
(iv) If p∞ |(GS : Gv ), then
res M
H 2 (Gv ) −→
∼ H 2 (Gkwv ),
w∈S f (kv )
where kwv is the local field associated to the prime w of the decomposition
v
field k of v.

Proof: By (8.3.18) and (10.4.8), cdp GS ≤ 2, so that cdp Gv ≤ 2, showing

(a). Being a quotient of Gv , the group Gv is finitely generated, so in order to
show (b) we can restrict to the case i = 2. Denote the Galois group of the local
cyclotomic ZZp -extension by Γv = G(kv,∞ |kv ) and consider the exact sequence
1 −→ Gv,∞ −→ Gv −→ Γv −→ 1.
By the Hochschild-Serre sequence, the group H 1 (Gv,∞ )Γv is finite, and so Gab
v,∞
is a finitely generated ZZp [[Γv ]]-module by the topological Nakayama lemma
(5.2.18). Since H 2 (Gv,∞ , Qp /ZZp ) = 0 by (10.3.26), we have
H 1 (Γv , H 1 (Gv,∞ , Qp /ZZp )) ∼
= H 2 (Gv , Qp /ZZp ) .
∗
0 −→(p Gab 2 2
v ) −→ H (Gv ) −→ p H (Gv , Qp /Z
Zp ) −→ 0 ,
it follows that
dimIFp H 2 (Gv ) = dimIFp p Gab 1 1
v + dimIFp p H (Γv , H (Gv,∞ , Qp /Z
Zp ))
= dimIFp p Gab ab Γv
v + dimIFp (Gv,∞ ) /p
≤ dimIFp H 1 (Gv ) + dimIFp (Gab δ

v,∞ ) /p < ∞ ,
where X δ denotes the maximal discrete submodule of an Iwasawa module X

(see (5.3.12)). This shows (b).
From now on we assume that µp ⊆ k. Since (k v )∞ contains the maximal
unramified extension of k∞ which is completely decomposed at all primes in
S, we conclude from (10.4.5) that
cdp Gv,∞ ≤ 1.
This yields
H 2 (Gv,m ) = H 1 (Γv,m , H 1 (Gv,∞ )),
and therefore
dimIFp H 2 (Gv,m ) ≤ dimIFp (Gab δ
v,∞ /p) < ∞.
This shows (ii).
Since µp ⊆ k, we have k v (µpm ) ⊆ kS and these fields are closed under
extensions which are unramified outside S and completely decomposed at v.
The principal ideal theorem implies
ClS (k v (µpm )) = 0.
(Recall our notational convention ClS = ClS (p)!) Now consider the commu-
tative exact diagrams
H 1 (K(µ pm )|K) H 1 (GS (K)) H 1 (GS (K(µpm )))
α β γ
M M M
H 1 (Kw (µpm )|Kw ) H 1 (Kw ) H 1 (Kw (µpm )),
w∈S(K) w∈S(K) w∈S(K(µpm ))

where K is a finite extension of k inside k v and the coefficients of the coho-

mology groups are µpm .
For sufficiently large K ⊆ k v , the map α is injective since v occurs as one
of the places in S(K) (α is the map 0 → 0 in almost all cases, cf. (9.1.6)). We
conclude that for sufficiently large K ⊆ k v
ker β = X1 (GS (K), µpm ) ,→ ker γ = (ClS (K(µpm ))/pm )∨ (1).
Dualizing and passing to the limit over all finite K ⊆ k v , we obtain a surjection
0 = (ClS (k v (µpm ))/pm )(−1) X2 (GS (k v ), ZZ/pm ZZ).
The Poitou-Tate sequence therefore implies the exactness of
H 2 (Gv , ZZ/pm ZZ) ,→ lim H 2 (Kw , ZZ/pm ZZ) → lim H 0 (GS (K), µpm )∨ .
M
−→ −→
K w∈S(K) K
Passing to the limit over all K ⊆ k v , the limit of the right-hand term is finite
and vanishes if p∞ |[k v : k].
Furthermore, H 2 (Gv , ZZ/pm ZZ) is finite by (i), and so
H 2 (kwv , ZZ/pm ZZ) = 0
for all but finitely many w ∈ S(k v ). This shows (iv) (setting m = 1). Finally,
passing to the limit over m ≥ 1 and since the strict cohomological dimension
of local fields is 2, we obtain
H 2 (Gv , Qp /ZZp ) = 0.
The same argument applies to every open subgroup of Gv , hence showing
assertion (iii). 2
The following theorem was first proved by L. KUZ’MIN [115]. The proof
presented below is a slight modification of that given in [254], Appendix.
(10.8.4) Theorem (KUZ’MIN). Let k be a totally imaginary number field with

µp ⊆ k and let S ⊇ Sp ∪ S∞ be a finite set of primes in k. Suppose that for a
prime v ∈ S(k) the group Gv is not open in GS . Then the canonical map
Gv −→ Gv
is an isomorphism, i.e. every p-extension of the local field kv is realized by a
p-extension of the global field k which is unramified outside S.
Remark: When v ∈
/ Sp (k), this result is already contained in (10.8.1).

Let us explain the strategy of the proof, which we will give below. The
prime v clearly must be nonarchimedean. By (10.8.3), both groups are of strict
cohomological dimension 2, therefore it is sufficient to show that the associated
homomorphism of abelianized groups is an isomorphism. This comes down
showing that every abelian extension of kv is realized by a (not necessarily
abelian) extension of k inside kS . By local class field theory, we have to show
that the group of universal norms of (kS )v |kv is contained in the subgroup of
p-divisible elements in kv× (which is the group of roots of unity of order prime
to p). This will be achieved in two steps. In the first step we show that every
element of this norm group is a p-th power in kv× . We do this by constructing
a suitable cyclic extension of degree p of k inside kS . The construction uses
Kummer theory; this is where we need µp ⊆ k.
In the second step we go up the cyclotomic tower in order to show that if
Gv were not the full local group, then a suitable extension k 0 of k inside kS
(for which all assumptions remain valid) would produce a contradiction to the
result of step 1.
However, the reader will find that the ideas explained above really lie behind
the proof. In fact, no universal norm groups will explicitly occur. Instead our
main technical tool is a careful analysis of the ZZp [[Gv ]]-module structure of the
abelianized kernel of Gv → Gv . Here we use the results of chapter V (for the
group ring of a non-abelian group!) in an essential way.
Proof of (10.8.4): For w ∈ S f (k), consider the exact sequence

1 −→ Rw −→ Gw −→ Gw −→ 1.
Here we tacitly have chosen a prolongation of w to kS and Rw is defined by
the exactness of the sequence. We set
Nw := ker(Gwab /p → Gab p p
w /p) = Rw [Gw , Gw ]Gw /[Gw , Gw ]Gw .
Global class field theory implies the exactness of the sequence

× ψ rec
Ok,S /p IS (k)/p G(kSab |L0 )/p,
where L0 is defined as the maximal extension of k inside kSab in which all
primes of S are completely decomposed. Local class field theory gives an
isomorphism
IS (k)/p ∼ Gwab /p
M
=
w∈S
Gab
L
and the map rec clearly factors through w∈S w /p. Therefore we conclude
that
M
Nw ⊆ ker(rec) = im(ψ).
w∈S

Now assume that there exists an xv ∈ Nv , xv =/ 0. Then there exists a global

×
S-unit e ∈ Ok,S with
M
ψ(e) = (0, . . . , xv , . . . , 0) ∈ Nw .
w∈S
Consider the extension
√
k 0 := k( p e).
The field k 0 is a cyclic extension of k inside kS , v does not decompose in k 0
and all w ∈ S(k) with w =/ v split in k 0 |k. We denote the cyclic Galois group
by G = G(k 0 |k), the decomposition group of v in kS |k 0 by Hv , the associated
full local group by Hv and we set
Nv0 = ker(Hvab /p → Hvab /p) = Rv [Hv , Hv ]Hvp /[Hv , Hv ]Hvp .
Then local class field theory induces a commutative diagram
N v
NG
Nv0
Ver
Gvab /p Hvab /p
rec rec
"incl"
kv× /p kv0 × /p.
The xv ∈ Nv chosen initially corresponds to the class of the global S-unit e in
kv× /p. This shows that
NG (xv ) = 0.
So far the prime v ∈ S f has been arbitrary. From now on we assume that Gv
is not open in GS . We will show that the map
NG : Nv −→ Nv0
is then injective, hence showing Nv = 0.
For this we first observe that by (10.8.3) we have isomorphisms
H 2 (Gv ) ∼
M
= H 2 (Gw ),
w∈S(kv )
H 2 (Hv ) ∼
M
= H 2 (Hw ).
w∈S(k0 v )
Since all w =/ v split in k 0 |k, we obtain an IFp [G]-module isomorphism

H 2 (Hv ) ∼ = H 2 (Hv ) ⊕ IFp [G]t
for some finite t (recall dimIFp H 2 (Hv ) < ∞). Furthermore, the corestriction
cor : H 2 (Gv ) −→
∼ H 2 (H ).
v

Dualizing, we conclude that H2 (Hv )/H2 (Hv ) is a free IFp [G]-module of finite
rank and the norm induces an isomorphism
G N G
H2 (Gv )/H2 (Gv ) −→
∼ (H2 (Hv )/H2 (Hv )) .
Keeping the assumption that Gv is not open in GS , we next investigate the

ZZp [[Gv ]]-module structure of Rvab . Consider the commutative diagram
v )
H2 (G H2 (Gv )
ab ab
p Gv p Gv .
The vertical maps are isomorphisms because scdp Gv = scdp Gv = 2 and the
upper horizontal arrow is injective by (10.8.3)(iv). Hence the lower horizontal
arrow is injective and it therefore follows from (5.6.13) that Rvab is a free
ZZp [[Gv ]]-module of finite rank.
We consider the cyclic extension k 0 |k constructed above and we compare
the homological Hochschild-Serre sequences for the group extensions Rv ,→
Gv Gv and Rv ,→ Hv Hv . Using the notation above, we obtain an exact
diagram
0§¦ ¡¢£¤¥ (H2 (Hv )/H2 (Hv ))G ((Rvab /p)Hv )G (Nv0 )G
NG NG NG
0 H2 (Gv )/H2 (Gv ) (Rvab /p)Gv Nv 0.

The left-hand and middle vertical arrows are isomorphisms, by the module
structure of Rvab and of H2 (Hv )/H2 (Hv ). Therefore the right-hand vertical
arrow is injective.
Since we have already constructed a nontrivial element in the kernel of
NG : Nv → Nv0 , we obtain a contradiction. Hence Nv = 0 and, since all
assumptions remain valid, the same is true for every finite extension of k
inside kS .
Setting r = rank ZZp [[Gv ]] Rvab , we obtain that
X := (Rvab )Gv,∞
is a free ZZp [[Γv ]]-module of rank r, where Γv = G(kv,∞ |kv ). The vanishing of
Nv (on every level) yields surjections
H2 (G¨ v,m ) (X/p)Γ ∼ = IFp [Γv /Γv,m ]r
v,m
for every m ≥ 1. Since the IFp -dimension of H2 (Gv,m ) is bounded as m → ∞

by (10.8.3)(ii), we conclude that r = 0, and hence Rv = 0. 2

It is much harder to show results corresponding to (10.8.4) for the full group
GS instead of its maximal pro-p-factor group, which we considered in this
section. In 2007, G. CHENEVIER (cf. [25]) succeeded in showing that the full
local group Gp is realized by GS under certain conditions. His proof uses `-adic
representations attached to automorphic forms. Afterwards he and L. CLOZEL
(see [CC]) refined the method and obtained the following theorem, in which
GS and Gkp denote the full Galois groups, i.e. the notational convention from
the beginning of this section does not apply to (10.8.5).
(10.8.5) Theorem (CHENEVIER-CLOZEL). Let k be a totally real number field,

p a finite prime of k and ` a rational prime number such that p - `. Then the
map
Gkp −→ GS` ∪S∞ ∪{p} (k)
is injective. In particular, any natural number divides the supernatural order of
the group GS` ∪S∞ ∪{p} (k).
Exercise (see [148]): Assume that k is a CM-field, i.e. k is a totally imaginary extension of
degree 2 of a totally real subfield k + . Let p be an odd prime number and assume that all primes
dividing p split in the extension k|k + . Show that the canonical homomorphism
G© p Gp ⊆ GS (k)
of pro-p-groups is an isomorphism for every finite set of primes S ⊇ Sp ∪ S∞ and for every
prime p dividing p. (Note that we did not assume µp ⊆ k!)
Hint: In this special situation every local abelian p-extension can be realized by a global
abelian p-extension. In order to prove this, examine the upper exact sequence of lemma
(10.3.13). If scdp Gp were equal to 2, this would suffice to prove the statement. Now change
to the cyclotomic ZZp -extension k∞ of k and apply (10.4.9)(iii).
§9. Free Product Decomposition of GS (p)
In this section we investigate how the decomposition groups of the primes

in S lie inside the group GS = GS (p), where p is a fixed prime number with
Sp ⊆ S.
We derive a criterion for the group GS to be a free product of local groups
and we call this the degenerate case. If we are in the generic (i.e. not in the
degenerate) case and if µp ⊆ k, then we show that GS is a pro-p duality group
of dimension 2 in which all decomposition groups are of infinite index (the
case p = 2 requires some modifications).

§9. Free Product Decomposition of GS (p) 687
The degenerate case is the easier one. Owing to the free product decom-
position, we obtain complete control over all extensions of k inside kS . In
particular, it can be easily deduced that the (strong) Leopoldt conjecture holds
in the degenerate case. Most of the results below are taken from [254].
Let k be a number field and let p be a prime number. We keep the notational
convention of the last section, i.e. unless the contrary is explicitly stated, we
tacitly assume extensions to be p-extensions and we always assume that the
finite set of primes S contains Sp ∪ S∞ . At those few places where we need
the full Galois group of the maximal extension KS of k which is unramified
outside S, we will denote this group by GS = G(KS |k), while GS = G(kS |k)
will always denote its maximal pro-p-factor. We define the group IS by the
exact sequence
1 −→ IS −→ GS −→ GS −→ 1.
In addition we use the following notation:
Sf the set of finite primes in S,
Gv the decomposition group of the prime v in GS ,
Tv the inertia subgroup of v in Gv ,
Gv = G(kv (p)|kv ), the full local group,
Tv the inertia subgroup in Gv ,
CS the S-idèle class group,
CS f the S f -idèle class group (see §2),
torp (A) the p-torsion subgroup of the abelian group A.
Recall the definitions of δ, δv and BSS0 = (VSS0 )∨ from (10.7.1) and (10.7.8). We
often write H i (−) for H i (−, ZZ/pZZ).
(10.9.1) Theorem. Let S0 be a subset of S f . Then the following assertions

are equivalent.
(i) There exists a finite set of primes T ⊇ S such that the canonical homo-
morphism
∗
Gv ∗
v ∈S\S0
∗
Gv /Tv −→ GS
v ∈T \S
is an isomorphism.
(ii) BSS0 = 0 and
X
δv = δ.
v ∈S0
If µp ⊆ k, then (i) and (ii) are equivalent to

0
(ii) S0 = {v0 } with v0 |p and Gv0 = GS .
Furthermore, if (i) and (ii) hold, then
X
#(T r S) = 1 + [kv : Qp ] − #(S r S0 ).
v ∈S0 ∩Sp

Remarks: 1. If (i) and (ii) hold, then GS0 is a free pro-p-group of rank
rk(GS0 ) = #(T r S) + #(S r S0 )f =
X
[kv : Qp ] − r1 − r2 + 1.
v ∈Sp ∩S0
Indeed, dividing out the inertia groups Tv , v ∈ S r S0 , we obtain from (i) the
isomorphism
∗
Gv /Tv ∗
v ∈S\S0
Gv /Tv −→∗
∼ G .
v ∈T \S
S0
2. The set T in (i) is not unique.
Proof of (10.9.1): Suppose that GS has a free product decomposition as

in (i). Since H 2 (Gv /Tv , Qp /ZZp ) = 0 = H 2 (Gv , Qp /ZZp ) by (7.2.5), it follows
from (4.1.4) that
H 2 (GS , Qp /ZZp ) = 0,
in other words (see (10.3.6), (i) ⇔ (vi)), the Leopoldt conjecture is true for k
and p. Hence (loc.cit.)
r2 + 1 = rank ZZp Gab
S
rank ZZp Gvab +

X X
= rank ZZp Gv /Tv
v ∈(S\S0 )f v ∈T \S
[kv : Qp ] + #(T r S0 )f .
X
=
v ∈Sp ∩(S\S0 )
By (10.7.3), we have
δv − δ + dimIFp BSS
X
dimIFp H 1 (GS ) = 1 +
v ∈S
and by (i), (4.1.4) and (7.5.11), this must be equal to
X X
dimIFp H 1 (Gv ) + dimIFp H 1 (Gv /Tv )
v ∈S\S0 v ∈T \S
δv + #(S r S0 )f + #(T r S).

X X
= [kv : Qp ] +
v ∈Sp ∩(S\S0 ) v ∈S\(S0 ∪SC )
Combining these equalities, we obtain
dimIFp BSS +
X
δv − δ = 0,
v ∈S0
and hence
BSS = 0
X
and δv = δ.
v ∈S0
Consider the exact sequence (9.2.2)
0 −→ coker(kS , S r S0 , ZZ/pZZ) −→ BSS0 −→ BSS −→ 0.
We have just seen that the term on the right-hand side vanishes and condition
(i), in conjunction with (4.1.4), implies the vanishing of the left-hand term.
Hence BSS0 = 0 and the proof of the implication (i) ⇒ (ii) is complete.

Now assume that (ii) holds. Then the exact sequence above implies that
coker(kS , S r S0 , ZZ/pZZ) = 0 = BSS .
By Čebotarev’s density theorem, we can find a finite set of primes T ⊇ S such
that the canonical restriction homomorphism
X1 (kS , S r S0 , ZZ/pZZ) −→
Y
1
Hnr (kv )
v ∈T \S
is an isomorphism. For such a set T , the middle horizontal arrow α in the

diagram
X1 (kS , S r¬®¯°ª« S0 , ZZ/pZZ)
Y
1
Hnr (kv )
v ∈T \S
α
Y Y
H 1 (GS ) H 1 (kv ) × 1
Hnr (kv )
v ∈S\S0 v ∈T \S
Y
H 1 (GS ) H 1 (kv )
v ∈S\S0
is an isomorphism. In order to show that the canonical homomorphism

∗
v ∈S\S0
Gv ∗ ∗
v ∈T \S
Gv /Tv −→ GS
is an isomorphism of pro-p-groups, it therefore (see (4.1.5)) remains to show

that X2 (kS , S r S0 , ZZ/pZZ) = 0.
Consider the following diagram, in which the exact row in the middle is part
of the long exact sequence of Poitou-Tate (use also (10.4.8)).
±²³´µ¶·¸¹
Y
H 2 (kv )
v ∈S0
X2 (kS , S, ZZ/pZZ)
Y
H 2 (GS ) H 2 (kv ) H 0 (GS , µp )∗
v ∈S
X2 (kS , S r S0 , ZZ/pZZ)
Y
H 2 (GS ) H 2 (kv ).
v ∈S\S0
Since X2 (kS , S, ZZ/pZZ) ∼

= BSS = 0, the snake lemma implies the exact se-
quence
0 −→ X2 (kS , S r S0 , ZZ/pZZ) −→ H 0 (kv , µp )∗ −→ H 0 (GS , µp )∗ −→ 0,
Y
v ∈S0
where we used the local duality isomorphism H 2 (kv ) ∼

= H 0 (kv , µp )∗ . Hence

dimIFp X2 (kS , S r S0 , ZZ/pZZ) =

X
δv − δ = 0.
v ∈S0
This shows the implication (ii) ⇒ (i).
P
If (ii) holds and µp ⊆ k, then the condition v∈S0 δv = δ implies that S0
consists of exactly one finite prime, say v0 . By Kummer theory, the group BSS0
corresponds to the maximal elementary abelian extension of k inside kS in
which v0 is completely decomposed. As BSS0 = 0, we obtain GS = Gv0 by
the Frattini argument. On the other hand, if (ii)0 holds, then the equation
P S
v ∈S0 δv = δ holds for trivial reasons and Kummer theory implies BS0 = 0. It
remains to show that v0 |p. Assume the contrary and choose a prime v1 ∈ Sp ⊆ S.
Using the isomorphism in (i), we obtain the contradiction
1 = rk ZZp Gab ab ≥
v0 = rk ZZp GS rk ZZp Gvab1 = [kv1 : Qp ] + 1.
Hence v0 |p. Finally, we show the statement about the rank. We have
X
#(T r S) = dimIFp H 1 (GS ) − dimIFp H 1 (kv )
v ∈S\S0
δv − δ + dimIFp BSS
X
= 1+
v ∈S
[kv : Qp ] − #(S r S0 )f −
X X X
− δv − δv
v ∈Sp ∩(S\S0 ) v ∈(S\S0 )f v ∈SIR
X
= 1 + r1 + 2r2 − [kv : Qp ] − #(S r S0 )
v ∈Sp ∩(S\S0 )
X
= 1+ [kv : Qp ] − #(S r S0 ).
v ∈Sp ∩S0
As a first application of the theorem above, we can calculate the cohomolog-

ical dimension of the group GS (p) in the difficult case where p = 2, S ⊇ S2 and
S ∩ SIR = ∅. By (10.6.1), we know that cd GS (2) ≤ 2. Since the cyclotomic
ZZ2 -extension k∞ (2) of k is contained in kS2 (2), the group GS2 (2) is infinite; in
particular, it is nontrivial. Hence, for S ⊇ S2 and S ∩ SR = ∅, the group GS (2)
is of cohomological dimension 1 or 2, and the next theorem gives a criterion
for which case occurs. Recall that Cl0,S (k) denotes the S-ideal class group in
the narrow sense.
(10.9.2) Theorem. Assume that S ⊇ S2 and S ∩ SR = ∅. Then cd GS (2) = 1

if and only if the following conditions (1)–(3) hold.
(1) S2 = {p0 }, i.e. there exists exactly one prime dividing 2 in k.
(2) S = {p0 } ∪ {complex places}.
(3) Cl0,S (k)(2) = 0.
In this case, GS (2) is a free pro-2-group of rank r2 + 1 and p0 does not split in
kS∪SR (2). In particular, if k is totally real and GS (2) is free, then kS (2) = k∞ (2).

Proof: Assume that GS (2) is free. By (10.6.2), we have an isomorphism

∗
p∈S\S2 (kS2 (2))
∼
T (kp (2)|kp ) −→ G(kS (2)|kS2 (2)) .
For nonarchimedean primes p ∈/ S2 , the maximal unramified 2-extension of

kp is realized by k∞ (2) ⊆ kS2 (2). Therefore, for p ∈ S r S2 , the maximal
2-extension of the local field kp is realized by kS (2) or, in other words, the
natural homomorphism
G(kp (2)|kp ) −→ GS (2)
is injective. But for these primes we have cd G(kp (2)|kp ) ≥ 2, showing that
S r S2 = ∅, i.e. S = S2 .
By Čebotarev’s density theorem, we find a finite set of nonarchimedean
primes T ⊇ S2 such that the natural homomorphism
M
H 1 (GS2 (2)) −→ H 1 (Gp /Tp )
p∈T \S
is an isomorphism. Then, by (1.6.15), the natural homomorphism

∗
p∈T \S2
Gp /Tp −→ GS2 (2)
is an isomorphism. By (10.6.2) with T = S2 ∪ SR and S = S2 and by (10.6.3),

the natural homomorphism
∗
p∈T \S2
Gp /Tp ∗ ∗
p∈SR
Gp −→ GS2 ∪SR (2)
is an isomorphism. Then, by (10.9.1), we obtain the conditions (1)–(3) and

that the unique prime p0 dividing 2 in k does not split in kS2 ∪SR .
If, on the other hand, conditions (1)–(3) of (10.9.2) are satisfied, then we
obtain the above isomorphism and deduce that GS2 (2) is free. The statement
about the rank of GS2 (2) follows as χ2 (GS2 (2)) = −r2 . If k is totally real, then
the homomorphism
GS2 (2) −→ G(k∞ (2)|k)
is a surjection of free pro-2-groups of rank 1 and hence an isomorphism. This
concludes the proof of (10.9.2). 2
Theorem (10.9.1) motivates the following
(10.9.3) Definition. We say that GS is degenerate if BSS0 = 0 for a subset

S0 ⊆ S f which satisfies the property
X
(+) δv = δ.
v ∈S0

Remarks: 1. For S00 ⊆ S0 we have a surjection BSS00 BSS0 . Therefore,

to decide whether GS is degenerate, it suffices to consider maximal subsets
S0 ⊆ S with the property (+).
2. Assume that GS is degenerate and µp ⊆ k. Then, by (10.9.1), there exists
a prime v0 |p such that Gv0 = GS . By (10.7.3) and (7.3.9), we obtain the
inequality
#S f + r1 + r2 ≤ dimIFp H 1 (GS ) = dimIFp H 1 (Gv0 )
≤ dimIFp H 1 (Gv0 ) = 2 + [kv0 : Qp ]
≤ 2 + 2r2 + r1 ,
and thus #S f ≤ r2 + 2. Therefore GS is not degenerate if µp ⊆ k and S contains
more than r2 + 2 finite primes.
3. Assume that GS is degenerate and µp ⊆ / k. By definition, there exists a
subset S0 of S such that µp / kp for all p ∈ S0 and
⊆
∗
v ∈S\S0
Gv ∗ ∗
v ∈T \S
∼ G .
Gv /Tv −→ S

#(S r(S0 ∪ S∞ )) ≤ rank ZZp GSab
X
δp ≤ ≤ [k : Q].
p∈S\S∞
/ k and S contains more than [k : Q]

Therefore GS is not degenerate if µp ⊆
finite primes p with δp = 1.
The following are examples of a degenerate GS (cf. §7).

• K = Q, p arbitrary and S = Sp ∪ S∞ .
• K = Q(µp ), p regular and S = Sp ∪ S∞ .
(10.9.4) Corollary. If GS is degenerate, then it decomposes into the free

product of local groups.
(10.9.5) Corollary. If GS is degenerate, then the Leopoldt conjecture holds

for p and for every finite extension of k inside kS .
Proof: The free product decomposition together with (4.1.4) and (7.2.5)
shows that GS has an open subgroup of strict cohomological dimension 2
(there might occur factors of the form G(C|IR) if p = 2). Now the corollary
follows from (10.3.9) and (10.3.11). 2

The group GS has an essentially different behaviour, depending on whether

or not there exists a prime v ∈ S such that Gv is open in GS . As we have seen
in §6, if k is not totally real, then this is possible only for primes v dividing
p, and if k is totally imaginary, then (GS : Gv ) ≤ 2 by (10.8.2). If k is totally
real, the decomposition groups of primes not dividing p can also be open.
(10.9.6) Definition. The group GS is called of local type if there exists a

prime v ∈ S such that Gv = GS .
Moreover, we say that GS is potentially of local type if an open subgroup
of GS is of local type, and otherwise GS is called of global type.
Mainly for the case p = 2, we introduce the following terminology.
(10.9.7) Definition. We say that a profinite group has virtually property (P )

if property (P ) is satisfied for all sufficiently small open subgroups.
For example, if S ⊇ S2 ∪S∞ , then GS is always virtually of cohomological 2-

dimension less than or equal to two, since every subgroup which corresponds
to a totally imaginary field has this property. In our applications the word
“ virtually ” will only occur if p = 2, and then it will always mean that the prop-
erty holds for all subgroups which correspond to totally imaginary extensions
of the base field.
(10.9.8) Theorem. If p =/ 2 and µp ⊆ k, then the group GS has one of the

following forms.
(i) If BS{v} =/ 0 for all primes v ∈ S f , then GS is of global type and it is a

duality group of dimension 2.
(ii) If BS{v0 } = 0 for a prime v0 ∈ S f , then GS is of local type and there exists
a finite set of primes T ⊇ S such that the canonical homomorphism
∗
v ∈S\{v0 }
Gv ∗ ∗
v ∈T \S
Gv /Tv −→ GS
is an isomorphism.
In (ii) the prime v0 is unique but the set T is not. For p = 2 we have the
following variant of (10.9.8).

(10.9.9) Theorem. If p = 2, then GS has one of the following forms.

S
(i) If B{v} =/ 0 for all v ∈ S f and #S f (kS ) > 2, then GS is of global type
and it is a virtual duality group of dimension 2.
S
(ii) If B{v} =/ 0 for all v ∈ S f and #S f (kS ) = 2, then GS is potentially of
local type and it is a virtual Poincaré group of dimension 2.
(iii) If BS{v0 } = 0 for a prime v0 ∈ S f , then GS is of local type and there exists
a finite set of primes T ⊇ S such that the canonical homomorphism
∗
v ∈S\{v0 }
Gv ∗ ∗
v ∈T \S
Gv /Tv −→ GS
is an isomorphism.
Before we prove (10.9.8) and (10.9.9), we first calculate the module I =

D2 (ZZp ). It is nontrivial if and only if GS is virtually of cohomological dimen-
sion 2, and if cdp GS = 2, it is the dualizing module.
(10.9.10) Lemma. I∼
= torp (CS f (kS )).
Proof: By (10.2.1), the dualizing module of GS is canonically isomorphic to

torp (CS f (KS )).
Therefore the statement of the lemma follows from (10.4.8). 2
(10.9.11) Corollary. Assume that µp ⊆ k. Then we have the following

equivalences:
GS is virtually free ⇐⇒ #S f (kS ) = 1,
GS is virtually a Demuškin group ⇐⇒ #S f (kS ) = 2.
Proof: If µp ⊆ k, then I = torp (CS f (kS )) fits into the exact sequence
IndG
M
0 −→ µp∞ −→ GS µp∞ −→ I −→ 0
v
v ∈S f
(cf. (10.2.1)). The basic properties of I show that GS is virtually free, i.e.
virtually of cd = 1 if and only if I = 0. By (3.7.2), we conclude that GS is a
virtual Demuškin group if and only if I ∼= Qp /ZZp as an abelian group. This
shows the asserted equivalences. 2
In order to decide whether a given pro-p-group is a duality group of dimen-

sion 2, one has to calculate the term D1 (see III §4).

(10.9.12) Lemma. D1 (GS , ZZ/pZZ) ∼

= CS (kS )/p.
Proof: By global class field theory (and, in particular, since GS is a finitely

generated pro-p-group), we obtain the following equalities, in which the field K
runs through all finite extensions of k inside kS :
D1 (GS , ZZ/pZZ) = lim

−→
H 1 (G(kS |K), ZZ/pZZ)∗
= lim H 1 (G(kS |K), ZZ/pZZ)∨
−→
= lim
−→
G(kS |K)ab /p
= lim CS (K)/p = CS (kS )/p.
−→
2
(10.9.13) Proposition. The group GS is a duality group of dimension 2

with dualizing module torp (CS f (kS )) if and only if cdp GS = 2 and CS (kS ) is
p-divisible.
Proof: This follows from (3.4.6), (10.9.12) and (10.9.10). 2
(10.9.14) Proposition. If µp ⊆ k and if GS is of global type, then GS is a

(virtual, if p = 2 and SIR (k) =/ ∅) duality group of dimension 2.
0 −→ EkS ,S −→ IS (kS ) −→ CS (kS ) −→ ClS (kS ) −→ 0.
By the principal ideal theorem, we have ClS (kS ) = 0 ∗) , and EkS ,S is p-divisible
as µp ⊆ k. Thus we have an isomorphism
IS (kS )/p ∼
= CS (kS )/p.
By definition,
Kv× ∼ IndG ×
M M
IS (kS ) = lim
−→
= GS kS v .
v
K ⊆kS v ∈S(K) v ∈S(k)
Therefore CS (kS )/p = 0 if and only if (kS )v = kv (p) for all v ∈ S. By the
theorem of Kuz’min (10.8.4), this is the case if all decomposition groups are
of infinite index in GS , in other words, if GS is of global type. 2
∗) Recall our notational conventions!

S
Proof of (10.9.8): By Kummer theory, B{v} is non-zero if and only if there
S
exists an extension of k inside kS in which v splits. If B{v} = 0 for one v, then
we can apply (10.9.1) in order to see that we are in case (ii). Otherwise GS is
of global type by (10.8.2), and by (10.9.14) we are in case (i). 2
Proof of (10.9.9): If BS{v0 } = 0 for a prime v0 ∈ S f , then we can apply

(10.9.1), which gives (iii). Now assume that BS{v} =/ 0 for every prime v ∈ S f
which satisfies property (+) of (10.9.3). Then we see, using Kummer theory,
that Gv $ GS for all v. If GS is global, then we are in case (i) by (10.9.14).
So let us assume that there exists a prime v such that Gv is open in GS . Let
H ⊆ GS be the subgroup (of index 1 or 2) of GS which corresponds to the
subextension k(i) ⊆ kS . Then Gv ∩ H is open in H and we deduce from
(10.8.2) that v ∈ S2 and one of the following cases occurs:
First case: H ⊆ Gv .
Since Gv =/ GS , we have H = Gv and i ∈/ k in this case. Furthermore, v
decomposes in k(i)|k, so v = v1 v2 say. Let us consider the situation at the
level of k(i). Since v1 does not split in kS |k(i), we obtain BS{v1 } = 0. Applying
(10.9.9)(iii) we get a free product decomposition
Gv1 = H ∼
= Gv2 ∗ (other terms).
But
rk(Gv2 ) = rk(Gv1 ) ≥ rk(Gv1 ).
Hence there are no other terms and H = Gv1 = Gv2 . In particular, H is a
Demuškin group, by (7.5.11), and of local type. Furthermore, #S f (kS ) = 2 by
(10.9.11). Hence we are in case (ii).
Second case: (H : Gv ∩ H) = 2.
Let ṽ be a prime of k(i) above v (the extension k(i)|k can be trivial). If
Gv ∩ H = GS (k 0 ), then ṽ decomposes in k 0 |k(i), so ṽ = v1 v2 say. As in the
first case, it follows that GS (k 0 ) = Gv1 (k 0 ) = Gv2 (k 0 ), and so #S f (kS ) = 2 by
(10.8.2) (thus v does not decompose in the extension k(i)|k). Furthermore, GS
is potentially of local type and virtually a Demuškin group and we are again in
case (ii). 2
It would be interesting to obtain more information about GS in the generic

/ k. ∗) We have the following proposition, at least, which generalizes
case if µp ⊆
(10.5.7). (See XI for the definition of the Iwasawa µ-invariant.)
∗) The sequel to [254], which appeared in J. reine u. angew. Math. 416 (1991), contains a
mistake.

§10. Class Field Towers 697
(10.9.15) Proposition. Let k be totally real and p =/ 2. Assume that GS is non-

degenerate and that the Iwasawa µ-invariant of the cyclotomic ZZp -extension
k∞ of k vanishes. (This is known for abelian number fields.) Then GS is a
pro-p duality group of dimension 2 with dualizing module torp (CS f (kS )).
Proof: Since GS is non-degenerate, it is not free, and so G(kS |k∞ ) =/ 1.

Since µ = 0, it follows from (11.3.7) that G(kS |k∞ ) is a free pro-p-group.
Furthermore, it is of finite rank since k is totally real. Hence the result follows
from (3.7.5) and (10.9.10). 2
Remark: Another situation where we know that GS is a duality group, is the

following. Let k be a CM-field with totally real subfield k + and let S ⊇ Sp ∪S∞
be a finite set of primes of k + . Suppose that
(1) µp ⊆ kp+ for all p ∈ Sp (k + ) and
(2) all primes in Sp (k + ) split in k|k + .
Then the result of the exercise of §8 shows that (kS )p = kp (p) for all primes
p ∈ S(k). Removing all primes p with N (p) 6≡ 0, 1 mod p from S(k) (this does
not alter GS ), we may assume that kp contains a primitive p-th root of unity for
all primes p ∈ S(k), so that IS (kS ) and therefore also CS (kS ), is p-divisible.
§10. Class Field Towers
We define a sequence of extensions of an algebraic number field k as follows.

Let k0 = k and for n ≥ 0, let kn+1 be the Hilbert class field of kn :
k = k0 ⊆ k1 ⊆ k2 ⊆ ··· .
This sequence of fields is called the class field tower of k. Obviously, the field
Lsolv = n kn is the maximal unramified extension of k with prosolvable Galois
S
group. The class field tower is called finite if the extension Lsolv |k is finite
and infinite otherwise. For a prime number p, let L = L(p) be the maximal
p-subextension of Lsolv |k. Hence L|k corresponds to the p-class field tower
of k, which is a sequence of fields
k = k0 ⊆ k1(p) ⊆ k2(p) ⊆ ··· ,
(p)
where kn+1 is the p-Hilbert class field of kn(p) , i.e. the maximal abelian unramified
p-extension of kn(p) .

In 1964 E. S. GOLOD and I. R. ŠAFAREVIČ proved, cf. [57], that there exist
algebraic number fields which possess infinite class field towers. The existence
of such fields is a consequence of a theorem that they proved about finite p-
groups. We presented a proof of its sharpened form, due to W. GASCHÜTZ and
E. B. VINBERG, in III §9. The example given by Golod and Šafarevič is the
imaginary quadratic number field
√
k = Q( −2 · 3 · 5 · 7 · 11 · 13 · 17 · 19 ) ,
which has an infinite 2-class field tower.

More generally, in this section we consider p-class field towers in which
primes of a given set T are completely decomposed, i.e. we ask whether the
T
maximal unramified p-extension k∅ of k which is completely split at T is
infinite or not. We will see that we can always find a finite Galois p-extension
T
K|k such that K∅ |K is infinite. This happens if the p-rank of the T -ideal
class group of k is large enough, and this will be the case if sufficiently many
(depending on T ) primes ramify in K|k.
Another method of obtaining infinite p-class field towers can be applied in
the case of CM-fields when p is odd. Using the action of complex conjugation,
it is possible to find a smaller bound for the p-rank of the ideal class group.
Let us fix k and p and recall the notation δ and δp from (10.7.1). We first
consider the case of a nonempty set S. Of course, if Sp ⊆ S, then GS is infinite,
since then kS (p) contains the cyclotomic ZZp -extension.
The following primes cannot ramify in a p-extension, and are therefore
redundant in S:
1. Complex primes.
2. Real primes if p =/ 2.
3. Primes p - p with N (p) 6≡ 1 mod p (see (7.5.2)).
Removing all these redundant places from S, we obtain a subset Smin ⊆ S
which has the property that
GS (p) = GSmin (p).
(10.10.1) Theorem. Let p be a prime number, k an algebraic number field and

S an arbitrary set of primes of k. If
√
#Smin ≥ 1 + r1 + r2 + 2 r1 + r2 + δ,
then the group GS (k)(p) is infinite.
Proof: From (10.7.12), we obtain

h1 (GS (k)(p)) ≥ 1 + #Smin − δ − (r1 + r2 ) ,

h2 (GS (k)(p)) − h1 (GS (k)(p)) ≤ r1 + r2 − 1 .
If GS (k)(p) is finite, it follows from the theorem of Golod and Šafarevič that
q
h1 (GS (k)(p)) < 2 + 2 h2 (GS (k)(p)) − h1 (GS (k)(p)) + 1 ,
and hence
√
1 + #Smin − δ − (r1 + r2 ) < 2 + 2 r1 + r2 . 2
(10.10.2) Corollary. Let p be a prime number and let S be a set of places of Q.

For odd p, the group GS (Q)(p) is infinite if
#Smin ≥ 4 . ∗)
The group GS (Q)(2) is infinite, provided that #Smin ≥ 5.
Remark: If #Smin ≤ 3 + δ, then the group GS (Q)(p) can be finite or infinite,

see [109] and [107].
Now we will consider the case when S is empty. Recalling the notation
Ram(K|k) for the set of primes of k which ramify in K|k, we start with the
(10.10.3) Proposition. Let K|k be a cyclic p-extension and let T ⊇ S∞ be a

finite set of primes of k contained in S := Ram(K|k)∪S∞ . Then the inequality
dimIFp ClT (K)/p ≥ #S r T (k) − r1 (k) − r2 (k) − δ(k) + r10 (k)
holds, where r10 (k) is the number of real places of k which become complex
in K.
Remark: We see that for fixed T we can find extensions K|k of degree p
with arbitrary large dimIFp ClT (K)/p. Indeed, by the above proposition, the
extension K|k has to be ramified at all finite primes in T ∗∗) and at sufficiently
many other primes. Such extensions exist by the theorem of Grunwald-Wang
(9.2.8).
∗) The set S depends on p.
min
∗∗) It follows from the proof that it suffices that the places in T f do not split in K|k.

Proof: For a finite abelian group A, we set dp (A) = dimIFp A/pA = dimIFp p A.
Note that dp (B) ≤ dp (A) for every subquotient B of A. Since G is cyclic, we
have Ĥ i (G, M ) ∼
= Ĥ i+2 (G, M ) for all i and every G-module M . Consider the
exact sequence
× ×
Y Y
0 −→ OK,T −→ KP × UP −→ CS (K) −→ ClT (K) −→ 0.
P∈T (K) P∈S\T (K)
Q × Q
Setting G = G(K|k), IT = P∈T (K) KP and US\T = P∈S\T (K) UP , we obtain
the inequality
× ×
dp (Ĥ 0 (G, (IT ×US\T )/OK,T )) ≥ dp (Ĥ 0 (G, IT ×US\T )) − dp (Ĥ 0 (G, OK,T )).
Further,
dp (Ĥ 0 (G, CS (K))) = dp (G(K|k)) = 1
and
×
dp (Ĥ 0 (G, OK,T )) ≤ #T (k) − 1 + δ.
Therefore
dimIFp ClT (K)/p ≥ dp (Ĥ −1 (G, ClT (K)))

×
≥ dp (Ĥ 0 (G, (IT ×US\T )/OK,T ))−dp (Ĥ 0 (G, CS (K)))
×
≥ dp (Ĥ 0 (G, IT ×US\T ))−dp (Ĥ 0 (G, OK,T ))−dp (Gab )
×
X X
≥ dp (Ĥ 0 (KP |kp , KP )) + dp (Ĥ 0 (KP |kp , UP ))
p∈T (k) p∈S\T (k)
−#T (k) + 1 − δ − 1.
By assumption, all primes in S f are ramified in K|k, hence we are summing

up over terms which are equal to at least 1 for finite primes and which are 0
or 1 for infinite primes depending on whether or not these primes ramify in
K|k (which can happen only if p = 2). This finishes the proof. 2
(10.10.4) Corollary. Let K|Q be a quadratic number field and let S =

Ram(K|Q) ∪ S∞ . Then the following inequalities are true:

 #S r S∞ (Q) − 1 , if K is imaginary,
dimIF2 Cl(K)/2 ≥ 
#S r S∞ (Q) − 2 , if K is real.
Proof: This follows from (10.10.3) with T = S∞ . 2

For an arbitrary set T of primes of k we now consider the Galois group

T T
G(k∅ (p)|k), where k∅ (p) is the maximal unramified p-extension of k which is
completely decomposed at every prime of T . For brevity we set
T T
k∅ = k∅ (p) and T
GT∅ = G(k∅ (p)|k) ,
and we denote the maximal unramified p-extension k∅ (p) of k by L. By class
field theory, we have
h1 (GT∅ ) = dimIFp H 1 (GT∅ , ZZ/pZZ) = dimIFp ClT (k)/p .
(10.10.5) Theorem. Let T be a finite set of primes of the number field k. Then
T
the extension k∅ |k is infinite if
q
dimIFp ClT (k)/p ≥ 2 + 2 r1 + r2 + δ + # T r S∞ .
In particular, the maximal unramified p-extension L of k is infinite if
q
dimIFp Cl(k)/p ≥ 2 + 2 r1 + r2 + δ .
Proof: Assume that GT∅ is finite. Then the theorem of Golod and Šafarevič
(3.9.7) implies the inequality
1
4
h1 (GT∅ )2 < h2 (GT∅ ),
and hence
(h1 (GT∅ ) − 2)2 < 4 (h2 (GT∅ ) − h1 (GT∅ ) + 1).
q
dimIFp ClT (k)/p = h1 (GT∅ ) < 2 + 2 χ2 (GT∅ )
q
≤ 2+ r1 + r2 + δ + # T r S∞ .
This contradicts our assumption, and therefore GT∅ is infinite. 2
(10.10.6) Corollary. Let k be a number field and let p be a prime number.

Then given an arbitrary finite set of primes T , there exists a Galois extension
T
K|k of degree p such that K∅ |K is infinite.
Proof: By (10.10.3) and the remark after it, we can find cyclic extensions
K|k of degree p with dimIFp ClT (K)/p arbitrary large. The corollary follows
from (10.10.5) applied to these fields K, since the number r1 +r2 +δ +# T r S∞
is bounded independently of the choice of K. 2

(10.10.7) Corollary. Let K|Q be a quadratic number field such that at least 8
(resp. 6 ) prime numbers are ramified if K is real (resp. imaginary). Then K
has an infinite 2-class field tower.
Proof: Let S = Ram(K|Q) ∪ S∞ . Then by (10.10.4),

√
dimIF2 Cl(K)/2 ≥ #S\S∞ (Q) − 2 ≥ 6 ≥ 2 + 2 2 + 1,
if K is real, and
√
dimIF2 Cl(K)/2 ≥ #S\S∞ (Q) − 1 ≥ 5 ≥ 2 + 2 1 + 1,
if K is imaginary. Therefore the result follows from (10.10.5). 2
(10.10.8) Corollary. Let K|Q be a quadratic number field such that




 4, if K is complex and p is odd,

5, if K is complex and p = 2,


dimIFp Cl(K)/p ≥




5, if K is real and p is odd,


6, if K is real and p = 2.
Then K has an infinite p-class field tower.
√
Proof: Since δ = 0 for p odd except for p = 3 and the field K = Q( −3)
(which has class number 1), the result follows directly from (10.10.5). 2
In the case where p is odd, it is possible to find a smaller bound for the
p-rank of the ideal class group of a quadratic field having an infinite p-class
field tower. This will follow from the results below where more general fields
are considered which are quadratic extensions of subextensions having certain
properties. In particular, this can be applied to fields of CM-type. The method
was developed in [112], [104] and [256].
Let k|k0 be a quadratic extension with Galois group ∆ = G(k|k0 ) ∼ = ZZ/2ZZ
and let p be an odd prime number. Let T be an arbitrary finite set of primes of k
closed under the action of ∆. Then ∆ acts on H i (GT∅ ) = H i (G(k∅ T
|k), ZZ/pZZ)
and we have the following theorem, where the (+) and (−) sign denote the
eigenspaces with respect to the action of ∆.
(10.10.9) Theorem. In the above situation assume that H 1 (GT∅ )+ = 0 =

H 2 (GT∅ )+ and suppose we have
q
dimIFp ClT (k)/p ≥ 3 · max(1, 3
u− + 2 #T (k) ) ,
with u− = dimIFp (Ok× /p)− . Then G(k∅
T
|k) is infinite.

Remark: If k is a CM-field with maximal totally real subfield k + = k0 , then

we have u− = δ in the above theorem.
(10.10.10) Corollary. Let k|k0 be a quadratic extension, p an odd prime

number and L the maximal unramified p-extension of k. Assume that
H 1 (G(L|k))+ = 0 = H 2 (G(L|k))+ . Then the extension L|k is infinite if
√3
dimIFp Cl(k)/p ≥ 3 · max(1, u− ).
Proof of (10.10.9): Let

T
1 −→ R −→ F −→ G(k∅ |k) −→ 1
T
be a minimal presentation of G(k∅ |k) where F is a free pro-p-group. The
cup-product
∪
H 1 (GT∅ ) × H 1 (GT∅ ) −→ H 2 (GT∅ )
is a ∆-equivariant symplectic form. Since
H 1 (GT∅ ) = H 1 (GT∅ )− ,
its image is contained in H 2 (GT∅ )+ = 0, and so the pairing is trivial. It follows

T
from (3.9.13)(ii) that the defining relations of G(k∅ |k) are contained in (F )p F 3 ,
i.e.
R ⊆ (F )p F 3 ,
where F 3 is the third term of the descending p-central series of F . Thus
R ⊆ F(3) ,
where F(n) denotes the Zassenhaus filtration of F , cf. the remark (2) before
(3.9.9). From the sharpened form of the Golod-Šafarevič theorem due to Koch
(loc.cit.), it follows that
h2 (GT∅ ) > 4
27
h1 (GT∅ )3 ,
provided that GT∅ is finite. Taking the minus part of the exact sequence
T G∅ T
H 1 (k∅ |k∅ ) −→ H 2 (GT∅ ) −→ H 2 (G∅ ) ,
we get the inequality
dimIFp H 2 (GT∅ )− ≤ dimIFp H 2 (G∅ )− + #T (k) .
Furthermore, the exact sequence (10.7.2) induces the exact sequence
0 −→(Ok× /p)− −→ B∅ (k)∗ − −→ p Cl(k)− −→ 0 .

Using (10.7.6) we now obtain
dimIFp H 2 (GT∅ ) = dimIFp H 2 (GT∅ )−

≤ dimIFp X2 (G∅ )− + #T (k)
≤ dimIFp B∅ (k)∗ − + #T (k)
= dimIFp p Cl(k)− + u− + #T (k)
≤ dimIFp p ClT (k) + u− + 2 #T (k) ,
so that
h2 (GT∅ ) − h1 (GT∅ ) ≤ u− + 2 #T (k) ,
and therefore
4 1
27
h (GT∅ )3 − h1 (GT∅ ) ≤ u− + 2 #T (k)
if GT∅ is finite. This gives us the desired result. 2
(10.10.11) Corollary. Let p be an odd prime number and let K|Q be a quadratic
number field such that
dimIFp Cl(K)/p ≥ 3.
Then K has an infinite p-class field tower.
Proof: This follows from (10.10.9), since

(Cl(K)/p)+ = Cl(Q)/p = 0,
u− ≤ 1 and H 2 (G(L|K), ZZ/pZZ)+ = 0, which we see as follows:
By (10.7.6), the group H 2 (G(L|K), ZZ/pZZ)+ = X2 (K)+ is contained in
B∅ (K)+ . Furthermore, the exact sequence (10.7.2)
×
0 −→(OK /p)+ −→ B∅ (K)∗ + −→ p Cl(K)+ −→ 0
shows that the latter group is trivial. 2
Examples: One can find the following examples of the corollary above. Let
√
k1 = Q( −3321607)
√
k2 = Q( 39345017)
√
k3 = Q( −222637549223)
and let p = 3 in the first two cases and p = 5 in the last one. Then
dimIFp Cl(k)/p = 3, and so for these fields the maximal unramified p-extension

is infinite. See [207] for references and other examples of fields with infinite
class field towers.
Remark: As we have seen in this section, the class field tower of a number
field can be infinite, and moreover this is the typical case. However, there
is a conjecture due to J.-M. FONTAINE and B. MAZUR which claims that an
infinite unramified Galois extension of a number field never has the structure
of a p-adic Lie group. This means that every unramified finite dimensional
p-adic representation of GK has a finite image. More generally, the precise
conjecture is the following:
(10.10.12) Fontaine-Mazur Conjecture. Let k be a number field and let S be

a finite set of places of k. If S ∩ Sp = ∅ and if n is any natural number, then
every continuous representation
ρ : GS (k) −→ Gln (Qp )
factors through a finite quotient of GS (k).
The above conjecture follows from a more general principle conjectured

by Fontaine and Mazur, which says that Galois representations which “look
geometric” indeed arise from algebraic geometry in a well-defined sense. We
refer the reader to the original paper [51]; see also [105].
Regarding the Fontaine-Mazur conjecture from a technical point of view,
we make the following observation: the Golod-Šafarevič inequality (3.9.7)
holds in a slightly modified way also for p-adic analytic groups (see [41], th.
6.29 or [127], prop. 1.3 for the stronger form using the Zassenhaus filtration).
All methods to produce infinite class field towers which we have at hand
at the moment, use the Golod-Šafarevič inequality. Therefore we do not
have a method which could produce counterexamples to the Fontaine-Mazur
conjecture.
Since a pro-p-group whose open subgroups have globally bounded rank is
analytic, see [41], cor. 9.35, we obtain the
(10.10.13) Corollary. Assume that the Fontaine-Mazur conjecture is true. If

K is an infinite unramified p-extension of the number field k, then the p-rank
dimIFp Cl(k 0 )/p
becomes arbitrarily large as k 0 varies over the finite extensions of k in K.

§11. The Profinite Group GS
In this section we will prove a duality theorem for the profinite Galois group
GS = G(kS |k), where kS is the maximal extension of the number field k which
is unramified outside the finite set S of primes of k. We assume that S contains
all archimedean places S∞ . Again we use the notation S f for the set of finite
primes in S.
Let p be a prime number. As before we denote the set of primes of k dividing

p by Sp (k). The following theorem was first proved in [193].
(10.11.1) Theorem. Let S ⊇ Sp ∪ S∞ be finite and assume that k is totally

imaginary if p = 2. Then GS is a duality group at p of dimension 2 with p-
dualizing module I = torp (CS f (kS )), i.e. for every finite p-primary GS -module
M and all i, the cup-product
∪
H i (GS , M ) × H 2−i (GS , Hom(M, I)) −→ H 2 (GS , I) −→
∼ Q /Z
p Zp
defines a perfect pairing of finite groups.
Before we are going to prove this theorem we will collect some facts about
the asymptotic behaviour of the class number of the cyclotomic field Q(ζpn )
as n tends to infinity. The proof of the following proposition can be found in
[246], th. 4.20.
(10.11.2) Proposition. Let h−

n be the minus part of the class number of Q(ζpn ).
Then
1
log h−
n ∼ (p − 1)p
(n−1)
n log p as n → ∞ . ∗)
4
As an easy consequence of the proposition above, we obtain the
(10.11.3) Proposition. Let S ⊇ Sp ∪ S∞ be a set of primes of the number

field k and assume that k is totally imaginary if p = 2. Then
cdp GS = 2 .
∗) a ∼ b means a/b → 1.

§11. The Profinite Group GS 707
Proof: We know from (8.3.18) that cdp GS ≤ 2. Now let n be large enough
so that the class number hn of Q(ζpn ) is greater than 1. By class field theory,
there exists a cyclic unramified extension F |Q(ζpn ) of degree `, where ` is a
prime number dividing hn . Since the only prime of Q(ζpn ) which divides p is
principal, it splits completely in F . Thus there are ` different primes dividing
p in F , and if K = F k, then #Sp (K) > 1. From the exact sequence
H 2 (Kp , ZZ/pZZ) −→ H 0 (kS |K, µp )∗ −→ 0
M
H 2 (kS |K, ZZ/pZZ) −→
p∈S(K)
obtained from (8.6.10) it follows that
dimIFp H 2 (kS |K, ZZ/pZZ) ≥ #S f (K) − 1 ≥ #Sp (K) − 1 ≥ 1 .
Therefore the cohomological p-dimension of GS is equal to 2. 2
The following result due to L. C. WASHINGTON, see [245], cor.3, is of great

importance for the proof of (10.11.1).
(10.11.4) Theorem. Let p be a prime number and let k be an imaginary abelian

number field containing the group µ2p . Then the set
H = {` | ` a prime number dividing h−
n for some n}
is infinite. Here h−n denotes the minus part of the class number of the n-th
layer of the cyclotomic ZZp -extension k∞ |k.
Proof: We may assume that k = Q(ζ2p ). We use the following result, also
due to Washington, see [246], th. 16.12: ∗)
If ` is a prime number different to p, then the `-part of hn is bounded
independently of n.
From Iwasawa theory, we know that the asymptotic behaviour of the p-part
of h−n is given by the formula (11.1.6)
e n = λ− n + µ− pn + ν − ,
where pen is the exact power of p dividing h− − − −
n and λ , µ and ν are constants
∗∗ −e −
independent of n. ) Thus, using (10.11.2), we see that p hn tends to infinity
n
as n → ∞ and therefore, recalling the result on the `-parts of hn , it follows that

there must always be new prime numbers dividing h− n if n → ∞. 2
∗) Actually, in the original proof of (10.11.4) which appeared before this result was proved,
Washington used a much weaker statement, which is sufficient for this purpose: the `-part of
pn
h−
n is bounded asymptotically by ` . Further, the assumption that µ2p is contained in k is
not necessary, see [245].
∗∗) In fact, µ− = 0 by the theorem of Ferrero and Washington, see [246], th. 7.15, but we will
not use this result.

In order to prove theorem (10.11.1), we use (3.4.6). We have to verify the

vanishing of the limit
Di (ZZ/pZZ) = lim H i (U, ZZ/pZZ)∗
−→
U ⊆GS
cor ∗
for i = 0, 1, since we already proved in (10.2.2) that the p-dualizing module of

GS ,
I = lim
−→
D2 (ZZ/pm ZZ) = lim lim H 2 (U, ZZ/pm ZZ)∗ ,
−→ −→
m m U ⊆GS
is isomorphic to torp (CS f (kS )). Here the limits run through the open subgroups
U of GS and the transition maps are the duals of the corestriction maps and the
canonical projections from ZZ/pm ZZ onto ZZ/pn ZZ, respectively.
Obviously, D0 (ZZ/pZZ) = 0 since p∞ |#GS . In order to prove the vanishing
of D1 (ZZ/pZZ), the following theorem is crucial.
(10.11.5) Theorem. If S ⊇ Sp ∪ S∞ is finite, then CS (kS ) is p-divisible.
(10.11.6) Corollary. Let S ⊇ Sp ∪ S∞ be a set of primes of the number field k.

Then (kS )p is a p-closed local field for all p ∈ S(kS ).
Remark: We already knew that (kS )p is p-closed in the following cases:

• p ∈ S r Sp , by (10.8.1),
• δ(S) = 1 and p ∈ S arbitrary, by (10.5.9).
In §9 we investigated a similar pro-p version of theorem (10.11.5) and we

will need this result now. We use the following notation:
kS (p) the maximal p-extension of k inside kS ,
GS (p) the Galois group of kS (p)|k,
kp (p) the maximal p-extension of kp , p a prime of k,
Gp (p) the Galois group of kp (p)|kp ,
Gp (p) the decomposition group of p in GS (p).
From the results of §9, we obtain the
(10.11.7) Corollary. Assume µp ⊆ k and let S ⊇ Sp ∪ S∞ be finite. Suppose

that CS (kS (p)) is not p-divisible. Then there exists a prime p ∈ Sp (k) such that
the following inequality holds:
X
[kp : Qp ] ≥ [kp0 : Qp ].
p0 ∈Sp (k)
p0 =
/p

Proof: Since µp ⊆ k, we have by (7.3.9) the (in)equality

rank Gp0 (p) ≤ rank Gp0 (p) = [kp0 : Qp ] + 2
for primes p0 ∈ Sp (k). Now assume that #Sp (k) > 1 (otherwise the statement
is trivial) and assume that CS (kS (p)) is not p-divisible, so that GS (p) is not
a (virtual) duality group of dimension 2 by (10.9.12). From (10.9.8) resp.
(10.9.9) and (10.9.1)(ii)0 , it follows that there exists a prime p ∈ Sp (k) such
that GS (p) = Gp (p), so that
[kp : Qp ] ≥ rank Gp (p) − 2 = rank GS (p) − 2 ,
and the free product decomposition of GS (p) obtained there yields the inequal-
ities X
[kp : Qp ] ≥ rank Gp0 (p) − 2
p0 ∈Sp (k)
p0 =
/p
X
= ([kp0 : Qp ] + 2) − 2
p0 ∈Sp (k)
p0 =
/p
X
≥ [kp0 : Qp ] . 2
p0 ∈Sp (k)
p0 =
/p
Proof of (10.11.5) and (10.11.6): It suffices to show that CS (KS (p)) is p-

divisible for a cofinal set of finite extensions K of k in kS . Since µ2p ⊆ kS ,
we assume without loss of generality that µ2p ⊆ k; in particular, k is totally
imaginary, containing the imaginary abelian field Q(ζ2p ). Assume that K is a
finite extension of k inside kS and that CS (KS (p)) is not p-divisible.
Claim: There exists a finite extension K 0 |K in kS such that CS (KS0 (p)) is
p-divisible.
Proof of the claim: Using (10.11.4), we choose a prime number ` > p and a
number n such that
(i) ` | h(Q(ζpn )) ,
(ii) (`, [K : Q]) = 1.
It follows that there exists an unramified extension F |Q(ζpn ) of degree `, in
which necessarily the only prime of Q(ζpn ) above p splits completely. By
condition (ii), we see that F and K(ζpn ) are linearly disjoint over Q(ζpn ).
Thus every prime p of K(ζpn ) dividing p splits into ` different primes in KF .
Therefore the field KF has the property that for every prime dividing p there are
at least `−1 other primes having the same absolute local degree. It follows that
there exists no prime p ∈ Sp (KF ) satisfying the inequality of (10.11.7). Thus
the group CS ((KF )S (p)) is p-divisible, which proves the claim and therefore
the theorem.

It remains to show (10.11.6). A straightforward limit argument shows that

we may assume that S ⊇ Sp ∪ S∞ is finite. Hence CS (kS ) is p-divisible by
(10.11.5). Consider the exact sequence
0 −→ OS× (kS ) −→ IS (kS ) −→ CS (kS ) −→ 0.
As the group OS× (kS ) is p-divisible, we conclude that IS (kS ) is also p-divisible.
Hence the multiplicative group (kS )× p is p-divisible for all p ∈ S(kS ). Since
µp ⊆ kS ⊆ (kS )p , Kummer theory shows that the local fields (kS )p are p-closed.
2
Proof of (10.11.1): By (10.11.3), we have cdp GS = 2, and using (10.11.5),

we can prove the vanishing of D1 (ZZ/pZZ):
D1 (ZZ/pZZ) = lim H 1 (kS |K, ZZ/pZZ)∗
−→
K
= lim
−→
G(kS |K)ab /p
K
= lim
−→
CS (K)/p
K
= CS (kS )/p = 0 .
Since the p-dualizing module I of GS is isomorphic to torp (CS f (kS )), the proof
of (10.11.1) is complete. 2
Using (10.11.5) we obtain the following general duality theorem which was
already mentioned in VIII (see (8.4.4)).
(10.11.8) Theorem. Let p be a prime number and let S be a finite set of primes
of the number field k such that S ⊇ S∞ . Assume either that Sp is contained
in S or that S ∩ Sp = ∅ and p∞ |#GS . Then the cup product
∪ inv
Ĥ i (GS , Hom(A, CS )) × Ĥ 2−i (GS , A) −→ H 2 (GS , CS ) −→
∼
1
#GS
Z/Z
induces isomorphisms
∼
Ĥ i (GS , Hom(A, CS ))(p) −→ Ĥ 2−i (GS , A)(p)∨
for all i ∈ ZZ and every discrete GS -module A which is finitely generated as a
ZZ-module.
Proof: Let us first assume that S ⊇ Sp . Then p∞ |#GS , for instance because kS
contains the cyclotomic ZZp -extension of k. By (10.11.5), CS and hence also
CS0 are p-divisible. Furthermore CS0 is a level-compact formation module with

divisible universal norm groups. Since Hom(A, CS0 ) has the same cohomology
as Hom(A, CS ) by (8.4.2), the theorem follows in this case from (3.1.11)(ii).
By the same argument, it remains to show the p-divisibility of CS in the case
S ∩ Sp = ∅. This follows from the principal ideal theorem. Indeed, under
the given assumptions, for every finite extension K of k inside kS , the group
GS (K)ab (p) is finitely generated (since S is finite) and torsion (a ZZp -extension
is ramified at least at one prime dividing p by (10.3.20)), hence finite. By the
group theoretic form of the principal ideal theorem, see [160] chap.VI (7.6),
we conclude that
lim
−→
GS (K)ab (p) = 0.
K ⊆kS
Since DS (K) is divisible, the exact sequences
DS (K) −→ CS (K) −→ GS (K)ab −→ 0
induce an isomorphism
∼
CS /p −→ lim
−→
GS (K)ab /p = 0
K ⊆kS
in the limit. Hence CS and thus also CS0 are p-divisible. 2
If A is a finitely generated GS -module with #tor(A) ∈ IN(S), then we have

a perfect pairing between the finite groups X1 (GS , A0 ) and X2 (GS , A), see
(8.6.7). If S is finite, we will prove that this holds true if we interchange the
roles of A and A0 , at least for the `-parts, where ` ∈ IN(S).
We start by investigating the Shapiro maps

shi : H i (GS , IS (A)) −→ P i (GS , A0 ),
which were defined in VIII §5. We described sh0 and obtained partial results
for sh1 and sh2 in (8.5.5). Using (10.11.6), we are able to sharpen these results.
(10.11.9) Proposition. Let S ⊇ S∞ be a finite set of primes of the number

field k and let A ∈ ModS (GS ). Then the following hold.
(i) sh0 is surjective with kernel
Nk̄p |kp Hom(A, C× ) .
Y
ker sh0 =
p∈S∞ (k)
(ii) sh1 is injective, and bijective if A is ZZ-free or if S is finite.

(iii) sh2 is injective and for each ` ∈ IN(S), the induced map on `-torsion
subgroups
H 2 (GS , IS (A))(`) −→ P 2 (GS , A0 )(`)
is an isomorphism.

Proof: In (8.5.5) we already proved assertion (i), the injectivity of sh1 , and
assertions (ii) and (iii) in the case that A is ZZ-free. We go through the proof of
this proposition again to show the additional assertions.
Since A ∈ ModS (GS ), we have an isomorphism A ∼ = ZZr ⊗ A0 of trivial
G(k̄|kS )-modules, where r ≥ 0 and A0 is a finite abelian group with #A0 ∈
IN(S). As A00 is a trivial G(k̄|kS )-module and (kS )p is `-closed for p ∈ S and
` ∈ IN(S) by (10.11.6), we obtain
H 1 (G(k̄p |(kS )p ), A0 ) ∼
= H 1 (G(k̄p |(kS )p ), k̄p× )r ⊕ H 1 (G(k̄p |(kS )p ), A00 ) = 0
for p ∈ S using Hilbert’s Satz 90. The Hochschild-Serre spectral sequence
H 1 (G((kS )p |kp ), Hom(A, (kS )× ∼ 1 ×

p )) → H (G(k̄p |kp ), Hom(A, k̄p ))
and an injection
H 2 (G((kS )p |kp ), Hom(A, k̄p× )) ,→ H 2 (G(k̄p |kp ), Hom(A, k̄p× )).
Using the commutative diagram
H 2 (G((kS )p |kpº»¼½ ), Hom(A, (kS )× H 2 (G(k̄p |kp ), Hom(A, k̄p× ))

M M
p ))
p∈S p∈S
sh2
H 2 (GS , IS (A)) P 2 (GS , A0 )
we obtain the first assertion of (iii). Now let ` ∈ IN(S). Then `∞ | G((kS )p |kp )
for all nonarchimedean primes p ∈ S. As kS is totally imaginary, we obtain
H i (k̄p |(kS )p , A0 )(`) = 0
for all p ∈ S and i ≥ 1. Hence the horizontal injection in the last diagram is an
isomorphism on the `-torsion subgroups. This finishes the proof of (iii).
Finally, assume that S is finite. Then the commutative diagram
H 1 (G((kS )p |kp¾¿ÀÁ ), Hom(A, (kS )× H 1 (G(k̄p |kp ), Hom(A, k̄p× ))

Y Y
p ))
p∈S p∈S
sh1
H 1 (GS , IS (A)) P 1 (GS , A0 ),
shows the remaining assertion in (ii). 2

(10.11.10) Theorem (Poitou-Tate Duality, second form). Let S ⊇ S∞ be a

finite set of primes of the number field k and let A be a finitely generated
GS -module with #tor(A) ∈ IN(S). Then for all ` ∈ IN(S) there is a perfect
pairing
X1 (GS , A)(`) × X2 (GS , A0 )(`) −→ Q/ZZ
of finite groups, which is induced by the cup-product, i.e. the diagram
H 1 (GSÃÄÅÆÂ , A)(`) × H 1 (GS , CS (A))(`) ∪

H 2 (GS , CS ) inv
Q/ZZ
X1 (GS , A)(`) × X2 (GS , A0 )(`)
commutes.
Proof: Since H 1 (GS , A) is a finite group by (8.3.20)(ii), its subgroup

X1 (GS , A) is also finite. By (8.6.6), the diagram
H 1 (GS ,ÏÎÌÍËÉÊÇÈ IS (A)) H 1 (GS , CS (A)) δ

H 2 (GS , A0 ) H 2 (GS , IS (A))
sh1 ∆1 sh2
(ξ 1 )∨ λ2
commutes. By (10.11.9)(ii), sh1 is an isomorphism. For ` ∈ IN(S), the maps

∆1 and sh2 induce isomorphisms on the `-primary parts by (10.11.8) and
(10.11.9)(iii), respectively. As the upper sequence is exact, we get an exact
sequence for the `-primary parts of the groups in the lower line. We obtain the
ÓÔÑÐÒ H 1 (GS , CS (A))(`) δ
X2 (GS , A0 )(`)
∆1
(ξ 1 )∨
P 1 (GS , A)(`)∨ H 1 (GS , A)(`)∨ X2 (GS , A)(`)∨ .
Since coker (ξ 1 )∨ = X1 (GS , A)∨ by (8.6.5), this proves the theorem. 2
Remark: For a finite module A ∈ ModS (GS ), the pairing constructed in

(10.11.10) coincides with that constructed in (8.6.7) for A0 up to sign. Indeed,
denoting these pairings by t1 and t2 respectively, the diagram

H 0 (GS , ×àáÞßÙÚÛÜÝØÖÕ CS (A0 )) × H 2 (GS , A0 ) ∪

Q/ZZ
δ
t2
X1 (GS , A00 ) × X2 (GS , A0 ) Q/ZZ
(−1)ψ∗
t1
X1 (GS , A) × X2 (GS , A0 ) Q/ZZ
δ
H 1 (GS , A) × H 1 (GS , CS (A)) ∪ Q/ZZ

commutes by (1.4.5); here ψ is the canonical isomorphism A00 →∼ A and the
maps δ are the connecting homomorphisms with respect to the exact sequences
0 → M 0 → I(M ) → C(M ) → 0, where M = A0 and M = A, respectively.
In order to obtain further information about the decomposition behaviour

of primes in infinite algebraic extensions, we need some facts from analytic
number theory.
We use from now on the following notation:
k is an algebraic number field,
M |k is an infinite unramified Galois extension of k,
p runs through all finite primes of k,
f (p) is the residue extension degree of p in M |k (1 ≤ f (p) ≤ ∞),
T = T (M |k) = {p | f (p) < ∞}.
Furthermore, let dk be the discriminant of k and

 1 (log 4π + γ), if p ∈ S∞ is real,


2
βp =


 log 2π + γ, if p ∈ S∞ is complex,
where γ = lim (1 + 1/2 + · · · + 1/n − log n) = 0.577 · · · is the Euler constant.
n→∞
Finally, we set
1 X
α(k) = log |dk | − βp .
2 p∈S∞
We will prove the following theorem, due to Y. IHARA [82]. It says that only a
small proportion of primes is almost completely decomposed in M |k.
(10.11.11) Theorem.
X log N (p)
≤ α(k) .
p∈T (M |k)
N (p)f (p) − 1
In particular, the expression on the left is convergent.

Remark: If the Riemann hypothesis were to be valid for the Dedekind zeta
function ζK (s) for all finite algebraic number fields K with k ⊆ K ⊆ M , then
a much stronger result would hold:
X log N (p) 1 X
≤ log |dk | − αp ,
p∈T (M |k)
N (p)(1/2)f (p) − 1 2 p∈S∞
where 
 1 (log 8π + γ + π2 ), if p ∈ S∞ is real,


2
αp =


 log 8π + γ, if p ∈ S∞ is complex.
For this see the original paper of Ihara [82] where the following analogue for
function fields is also proved:
Let IFq be the constant field of k and let g be its genus. We denote the degree
over IFq of a prime p of k by deg(p). Then
X deg(p)
≤ max(g − 1, 0) .
p∈T (M |k)
N (p)(1/2)f (p) − 1
The presence of the factor 1/2 in the exponent of N (p) is due to the theorem
of Weil for curves.
Let ζK (s) be the Dedekind zeta function of K,

(1 − N (P)−s )−1 ,
Y
ζK (s) = Re(s) > 1,
P
and let 0
ζK (s)
ZK (s) = − .
ζK (s)
The Dirichlet series corresponding to ZK (s) is given by
0
ζK (s) X X log N (P)
− = ms
.
ζK (s) P m≥1 N (P)
For a real number x > 1, let ψK (x) be the Čebyšev function obtained as the
partial sum of coefficients of the Dirichlet series of ZK (x), i.e.
X
ψK (x) = log N (P) .
m≥1
N (P)m <x
We will need the following well-known identity for ZK (s):
(10.11.12) Lemma. For Re(s) > 1

0 Z ∞
ζK (s)
− = s−1 ψK (x)x−s−1 dx .
ζK (s) 1

Proof: We will use Abel’s summation formula. Let (an ) be a sequence of

complex numbers and let ϕ(x) be a complex-valued function on (0, ∞) having
a continuous derivative. Let A(x) = n≤x a(n) and assume that A(x)ϕ(x) → 0
P
as x → ∞. Then
∞ Z ∞
A(t)ϕ0 (t)dt ,
X
a(n)ϕ(n) = −
n=1 1
provided that either side is convergent, cf. [24], chap.VII, th. 6.

Now we set 
log N (P), if n is a power N (P)m , m > 0,



a(n) =


 0, otherwise,
and ϕ(x) = x−σ , where σ is real and σ > 1. Then A(x) = ψK (x) and
A(x)ϕ(x) → 0 as x → ∞, since ψK (x) < x log x, so that
A(x)ϕ(x) = O(x1−σ log x) = o(1) .
Thus we can apply the summation formula and obtain the desired result for
real s > 1, and for Re(s) > 1 by analytic continuation. 2
For an algebraic number field K we set

A(K) = π −r1 (K)/2 (2π)−r2 (K) |dK |1/2 ,
where as usual r1 (K) and r2 (K) denote the number of real and complex primes
of K respectively. For the proof of (10.11.11) we also need the following
lemma due to E. LANDAU, cf. [120], Satz 180.
(10.11.13) Lemma. We have the partial fraction decomposition of ZK (s)

r1 (K) Γ 0 s Γ0 1 1 X0 1
ZK (s) = log A(K) + ( ) + r2 (K) (s) + + − ,
2 Γ 2 Γ s s − 1 ρ∈Z(K) s − ρ
where Z(K) is the set of all nontrivial zeros of ζK (s) and the sum ρ0 is taken
P
with multiplicity, where the terms for ρ and ρ̄ should be summed together.
Proof: Let
s
ξK (s) = s(s − 1)A(K)s Γ ( )r1 (K) Γ r2 (K) ζK (s) .
2
Then ξK (s) is an entire function and ξK (1 − s) = ξK (s), cf. [160], chap.VII,
(5.10). Since ξK (s) is of order 1, Hadamard’s factorization theorem yields the
expression
Y s
ξK (s) = ea+bs 1 − es/ρ
ρ ρ

for some complex numbers a and b, where ρ runs through all the zeros of ξK (s),
which are exactly the nontrivial zeros of ζK (s). ∗)
Taking the logarithmic derivative of this product, we obtain
0
ξK X 1 1
(s) = b + + ,
ξK ρ s−ρ ρ
where the sum converges absolutely. From
0
ξK ξ0
(s) = − K (1 − s)
ξK ξK
it follows that
X 1 1 X 1 1
b+ + = −b − + .
ρ s−ρ ρ ρ (1 − ρ) − s ρ
Since 1 − ρ is a zero whenever ρ is, we obtain
X0 1
b=− ,
ρ ρ
where we now have to sum the ρ and ρ̄ terms together. Thus the formula above
becomes
0
ξK X0 1
(s) = .
ξK ρ s−ρ
2
Proof of (10.11.11): Let K run through the finite Galois extensions of k

inside M , so K|k is unramified. Then
0
ζK (s) X X log N (P)
ZK (s) = − = ms
ζK (s) P m≥1 N (P)
XX log N (p)
= [K : k] · f (P|p)ms
p m≥1 N (p)
X log N (p)
= [K : k] · f (P|p)s − 1
, Re(s) > 1,
p N (p)
where f (P|p) is the residue extension degree of P in K|k, and for the real
number x > 1, we have for the Čebyšev function
X X
ψK (x) = log N (P) = [K : k] · log N (p) .
m≥1 m≥1
N (P)m <x N (p)f (P|p)m <x
Now we set
XX log N (p) X log N (p)
ZM (s) := f (p)ms
= f (p)s − 1
,
p∈T m≥1 N (p) p∈T N (p)
∗) For the definition of the order of an entire function and for the product theorem, see [166],
appendix 5.

which is a convergent Dirichlet series holomorphic on Re(s) > 1, and

X
ψM (x) := log N (p) .
m≥1,p∈T
N (p)f (p)m <x
From the definitions above it follows that

ψM (x) ≤ [K : k]−1 ψK (x) ,
and by (10.11.12), which similarly holds for ZM (s), we obtain the inequality
for Re(s) > 1
ZM (s) ≤ [K : k]−1 ZK (s) .
Now let s = σ > 1 be real. Using (10.11.13) and the fact that
1 1
+ > 1 for σ > 1,
σ − ρ σ − ρ̄
we obtain by letting K → M
X log N (p) r1 Γ 0 σ Γ0
ZM (σ) = f (p)σ − 1
≤ log A(k) + ( ) + r2 (σ) (σ > 1) .
p∈T N (p) 2 Γ 2 Γ
(Observe that log |dK | = [K : k] log |dk |, since K|k is unramified.) Obviously,
for any finite subset T 0 of T we have the inequality above with T 0 in place of
T . Letting σ → 1 then gives
X log N (p) r1 Γ 0 1 Γ0
≤ log A(k) + ( ) + r2 (1) .
p∈T 0
N (p)f (p) − 1 2 Γ 2 Γ
Since T 0 is an arbitrary finite subset of T , this last inequality is valid for T in
place of T 0 , and since
Γ0 Γ0 1
(1) = −γ, ( ) = − log 4 − γ
Γ Γ 2
and
r1 1
log A(k) = − log π − r2 log 2π + log |dk |,
2 2
we have proved the theorem. 2
Recall that given a number field k and a set of nonarchimedean primes T of k,

we denote the maximal unramified extension of k in which all primes of T are
T
completely decomposed by k∅ . The following corollary, which immediately
follows from (10.11.11), asserts that the set T cannot be too big if the extension
T
k∅ |k is infinite.
T
(10.11.14) Corollary. Assume that the extension k∅ |k is infinite. Then
X log N (p)
≤ α(k) .
p∈T N (p) − 1

Another consequence of the theorem of Ihara is the following result, which

was already mentioned in X §2 and which can also be obtained by a pure
group-theoretical method; see (10.2.6) and the exercises in X §2.
(10.11.15) Corollary. Let S be a finite set of primes of the algebraic number

field k. Then the Galois group G(kS |k) can be topologically generated by a
finite number of conjugacy classes.
Proof: The decomposition groups of G(kS |k) with respect to the primes
p ∈ S are finitely generated as homomorphic images of the absolute local
Galois groups Gkp . So we are reduced to the case S = ∅, i.e. kS is the
maximal unramified extension k nr of k.
If k nr |k is finite, we are done, so let us assume that k nr |k is infinite. Since
the sum of log N (p)/(N (p) − 1) over all nonarchimedean primes p of k is
divergent, we can find primes p1 , . . . , pn of k such that
n
X log N (pi )
> α(k) ,
i=1 N (pi ) − 1
where α(k) was defined in (10.11.11). Now let M 0 |k be the maximal unramified
extension of k which is completely decomposed at all primes pi , i = 1, . . . , n.
The extension M 0 |k cannot be infinite by (10.11.11). Thus the normal subgroup
(Gp1 , . . . , Gpn ) generated by the decomposition groups Gpi = hFrobpi i, i =
1, . . . , n, has finite index in G(k nr |k). Therefore the latter group is generated
by finitely many conjugacy classes. 2


Chapter XI
Iwasawa Theory of Number Fields
As shown in the previous chapters, there is a remarkable analogy between

the theory of algebraic number fields and the theory of function fields in one
variable over a finite field. This analogy should also extend to the theory
of ζ-functions and L-functions of global fields. If, for a function field k,
one considers the corresponding smooth and proper curve C/IF, where IF is
the field of constants of k, then the ζ-function of the curve C is a rational
function. More precisely, if ϕ is the arithmetic Frobenius automorphism, i.e.
the natural generator of the Galois group G(IF|IF) ∼ = ẐZ, then ϕ acts on the
Tate module of the Jacobian variety J of C, which is defined as the projective
limit of the groups of pn -torsion points of J(IF), where p =/ char(IF) is a prime
number. Tensoring with Qp , one obtains a Qp -vector space of dimension 2g,
where g is the genus of C. The characteristic polynomial with respect to the
endomorphism ϕ−1 is the essential part of the ζ-function of the curve C.
In order to obtain an analogous result for a number field k, the idea of

K. IWASAWA [88] was to consider ZZp -extensions of k, and in particular, the
cyclotomic ZZp -extension k∞ |k (obtained by adjoining all roots of unity of
p-power order if k contains µp ). The analogy asserts that over k∞ one is in
a geometric situation: the group Γ = G(k∞ |k) ∼ = ZZp acts on various Galois
groups of abelian extensions of k∞ . The main conjecture for number fields
states that the characteristic polynomials of these actions are related to p-adic
L-functions. This was first proved by B. MAZUR and A. WILES [134] under the
assumption that the base field is abelian over Q, and later by A. WILES [249]
for general totally real fields.
In the first section we start by proving the beautiful theorem of Iwasawa

which describes the behaviour of the p-part of the class number in a ZZp -
extension. Furthermore, we study the ZZp [[Γ ]]-module structure of the Iwasawa
modules Xnr = G(H|k∞ ) and Xcs = G(H 0 |k∞ ), where H is the maximal
abelian unramified p-extension of k∞ and H 0 is its maximal subextension
which is completely decomposed over k∞ at the primes above p.

722 Chapter XI. Iwasawa Theory of Number Fields
After establishing the Iwasawa theory over local fields in §2, we consider the
Λ-module X1 = G(kΣ (p)|k∞ )ab in §3, where kΣ (p) is the maximal p-extension
of k unramified outside the set Σ = Sp ∪ S∞ of primes above p and ∞. Most
of the results will follow from the general homotopy theory of Λ-modules
developed in chapter V.
In §4 we consider the case of Iwasawa modules over CM-fields. Using
complex conjugation we get further insight into the structure of the relevant
Iwasawa modules. In §5 we present the (non-abelian) concept of positively
ramified extensions of number fields.
Finally in §6, we give an overview of the main conjecture of Iwasawa theory
and its applications.
Although not always mentioned explicitly, we owe a lot to a paper of
U. JANNSEN [97]. Further, we have used the original article of K. IWASAWA
[88] and the paper [252]. Though we do not refer to the beautiful book of
L. C. WASHINGTON [246] in this chapter, the reader is strongly advised to com-
pare our approach with the presentation there and to look at the many other
important aspects which are not treated here.
§1. The Maximal Abelian Unramified

p-Extension of k∞
We recall a definition given in X §3. A ZZp -extension of a number field k is

a Galois extension k∞ |k with Galois group G(k∞ |k) ∼ = ZZp , the additive group
of p-adic integers. One can regard such a ZZp -extension as a tower of fields
[
k = k0 ⊆ k1 ⊆ . . . ⊆ k∞ = kn
n
with G(kn |k) ∼= ZZ/pn ZZ, since the nontrivial closed subgroups of ZZp are of
n
the form p ZZp for some n. The fields kn ⊆ k∞ are uniquely determined by the
property [kn : k] = pn .
Let k(µp∞ ) be the extension of k obtained by adjoining all roots of unity of
p-power order. Then
G(k(µp∞ )|k) = Γ × ∆
where Γ is isomorphic to ZZp and ∆ ⊆ ZZ/(p − 1)ZZ if p is odd or ∆ ⊆ ZZ/2ZZ if
p = 2. By Galois theory, there exists precisely one ZZp -extension k∞ |k inside
k(µp∞ ). This extension is called the cyclotomic ZZp -extension of k.

§1. The Maximal Abelian Unramified p-Extension of k∞ 723
The next proposition shows that every ZZp -extension is unramified outside
the primes above p.
(i) Let k∞ |k be a ZZp -extension. Then every (possibly archimedean) prime
p of k not dividing p is unramified in k∞ |k and at least one prime p
(necessarily above p) ramifies in k∞ |k.
(ii) The cyclotomic ZZp -extension is ramified at every prime p above p.
(iii) The cyclotomic ZZp -extension Q∞ of Q is totally ramified at p.
Proof: (i) Let Tp be the inertia group of G(k∞ |k) with respect to p - p. Then
Tp = 0 or Tp ∼= pn ZZp for some n. In the first case we are done, and in the
second case we see that p is nonarchimedean. By class field theory, Tp is the
image of the homomorphism
Up (k)(p) −→â G(k̄|k)ab (p) G(k∞ |k).
But Up (k)(p) = µ(kp )(p) is finite, and so Tp = 0. Observing that a ZZp -extension
cannot be unramified everywhere because of the finiteness of the ideal class
group of k, we have therefore proved (i).
The assertion (iii) is known from the theory of cyclotomic fields, cf. [160],
chap.I, (10.1). Now statement (ii) follows easily for an arbitrary number field
k because k∞ = kQ∞ . 2
In X §3 we were dealing with the question of how many independent ZZp -

extensions of a number field k exist, and we saw that this problem is related to
the Leopoldt conjecture. We reformulate (10.3.20)(ii) as follows:
(11.1.2) Theorem. Let k̃ be the composite of all ZZp -extensions of k. Then

G(k̃|k) ∼
= ZZrp2 +1+dp ,
where r2 is the number of complex places of k and dp is the Leopoldt defect,
i.e. dp = rank ZZ Ok× − rank ZZp Ok× .
Now we consider unramified p-extensions of number fields. For n ≥ 0 let

Ln be the maximal unramified p-extension of kn and let L0n be the maximal
unramified p-extension of kn which is completely decomposed at all primes
above p. The maximal abelian extension Hn = Labn inside Ln is the p-Hilbert
0 0ab
class field of kn . Let Hn = Ln and let

L = lim Ln and L0 = lim L0n

−→ −→
n n
be the corresponding extensions over k∞ (in X §4 the field L0 was denoted by
S
(k∞ )Sp∞ (p)). Setting
H = Lab = lim
−→
Hn and H 0 = L0ab = lim
−→
Hn0 ,
n n
we have the following diagram of fields:
ãäåæçèéê H
H0
k∞ Hn
Hn0
kn
k
and the Galois groups
Xnr := G(H|k∞ ) = G(L|k∞ )ab ,
Xcs := G(H 0 |k∞ ) = G(L0 |k∞ )ab

are Iwasawa modules for Λ = ZZp [[Γ ]].
We introduce some further notation.
(11.1.3) Definition. Let k∞ |k be a ZZp -extension and let Tp = Tp (L|k) be the

inertia subgroup of G(L|k) with respect to a prime p. Then Tp maps onto an
open subgroup Γnp in Γ if Tp =/ {1}. We set
n0 = n0 (k∞ |k) = max{np | p ramifies in k∞ |k},
sn = s(kn ) = #{p a prime of kn | p ramifies in k∞ |kn },
s∞ = s(k∞ ) = #{P a prime of k∞ | P ∩ k ramifies in k∞ |k}.
By (11.1.1)(i), the numbers s∞ ≥ sn are finite, since ramified primes lie

above p and have open decomposition groups in Γ .
(11.1.4) Proposition. For every n ≥ 0

rank ZZp G(H 0 |k∞ )Γn ≤ rank ZZp G(H|k∞ )Γn ≤ sn − 1 ≤ s∞ − 1.
In particular, Xnr = G(H|k∞ ) and Xcs = G(H 0 |k∞ ) are finitely generated
Λ-torsion modules.

Proof: The last assertion follows from the first and (5.3.10). The exact
sequence
1 −→ G(H|k∞ ) −→ G(H|kn ) −→ Γn −→ 1
shows that
rank ZZp G(H|k∞ )Γn = rank ZZp G(H|kn )ab − 1.
Let Tn be the normal subgroup of G(H|kn ) generated by the inertia groups
TPi (H|kn ) ∼ = Tpi (k∞ |kn ) ⊆ Γn with respect to the ramified primes pi , i =
1, . . . , sn , in k∞ |kn . By definition of Hn we obtain the exact sequence
Tn /[Tn , G(H|kn )] −→ G(H|kn )ab −→ G(Hn |kn ) −→ 1,
and so, because G(Hn |kn ) ∼
= Cl(kn )(p) is finite,
rank ZZp G(H|kn )ab ≤ rank ZZp Tn /[Tn , G(H|kn )] ≤ sn .
Since G(H 0 |k∞ ) is a quotient group of G(H|k∞ ), the other inequality follows.
2
ë n)
For n ≥ n0 (k∞ |k), the canonical surjection G(H|k G(k∞ |kn ) induces
isomorphisms
∼ Γ
TPi (H|kn ) −→ for i = 1, . . . , s∞ .
n
In particular, for every i the group G(H|kn ) is the semi-direct product of

G(H|k∞ ) and TPi (H|kn ) and therefore there exist elements gi ∈ G(H|k∞ )
such that
τi = gi τ1 , i = 2, . . . , s∞ ,
n
where τi ∈ TPi (H|kn ), i = 1, . . . , s∞ , lifts the generator γ p ∈ Γn , where γ is a
chosen generator of Γ .
Recall that for n ≥ 0 the Weierstraß polynomials ωn ∈ ZZp [[T ]] ∼ = Λ are
defined by n −1
pX n n
pn pn p
ωn = (T + 1) − 1 = T + i
T p −i .
i=1
(11.1.5) Lemma.
(i) For n ≥ n0 (k∞ |k), the extensions k∞ and Ln are linearly disjoint. In
particular, G(Hn |kn ) ∼
= G(Hn k∞ |k∞ ).
(ii) For n ≥ n0 (k∞ |k),
G(H|Hn k∞ ) = hωn G(H|k∞ ), g2 , . . . , gs∞ i,
ωn+1
G(H|Hn+1 k∞ ) = ωn
G(H|Hn k∞ ),
Cl(kn )(p) ∼
= G(Hn |kn )
∼
= G(H|k∞ )/ ωωnn G(H|Hn0 k∞ ).
0

(iii) Suppose that s∞ = 1. Then for n ≥ n0 (k∞ |k),

∼ G(H |k ).
G(H|k∞ )Γn −→ n n
All statements hold analogously for H 0 and Hn0 if we replace Cl(kn )(p) by
ClSp (kn )(p), n0 (k∞ |k) by the finite number
m0 (k∞ |k) = max{mp | p ∈ Spf d (k)} ,
where mp is defined by Gp (k∞ |k) = Γmp and
Spf d (k) = {p ∈ Sp (k) | p is finitely decomposed in k∞ |k} ,
and s∞ by t∞ = #Spf d (k∞ ), where
Spf d (k∞ ) = {P ∈ Sp (k∞ ) | P ∩ k is finitely decomposed in k∞ |k} .
Proof: (i) follows from the fact that for n ≥ n0 the extension k∞ |kn is totally
ramified for some prime p|p, and (iii) is a direct consequence of (i) and the first
assertion of (ii):
G(H|k∞ )Γn = G(H|k∞ )/ωn G(H|k∞ )
= G(H|k∞ )/G(H|Hn k∞ )
∼
= G(Hn |kn ).
In order to prove (ii), observe that by (i) and the fact that Hn is the maximal
abelian unramified extension of kn inside H, we have
G(H|k∞ )/G(H|Hn k∞ ) ∼
= G(Hn |kn )
= G(H|kn )/h[G(H|kn ), G(H|kn )], TP1 , . . . , TPs∞ i
= G(H|kn )/hωn G(H|k∞ ), g2 , . . . , gs∞ , τ1 i
∼
= G(H|k∞ )/hωn G(H|k∞ ), g2 , . . . , gs i. ∞
Thus we have proved the first assertion of (ii). Similarly, we obtain

G(H|Hn+1 k∞ ) = hωn+1 G(H|k∞ ), g20 , . . . , gs0 ∞ i ,
where gi0 ∈ G(H|k∞ ) such that τip = gi0 τ1p , i ≥ 2. It follows that
1+τ1 +···+τ1p−1 p
τip = (gi τ1 )p = gi τ1
and therefore
τip = (γn gi )τ1p ,
where
n n2 n (p−1) ωn+1
γn = 1 + γ p + γ p + · · · + γp = .
ωn

We obtain
G(H|Hn+1 k∞ ) = hωn+1 G(H|k∞ ), ωωn+1
n
g2 , . . . , ωωn+1
n
gs∞ i
ωn+1
= ωn
hωn G(H|k∞ ), g2 , . . . , gs∞ i
ωn+1
= ωn
G(H|Hn k∞ ).
Finally, by class field theory, we have

Cl(kn )(p) ∼
= G(Hn |kn )
∼
= G(Hn k∞ |k∞ )
∼
= G(H|k∞ )/G(H|Hn k∞ )
= G(H|k∞ )/ ωωnn G(H|Hn0 k∞ ).
0
The arguments for the extensions H and Hn0 are exactly the same except that
0
one has to consider the decomposition groups

GP (H 0 |kn ) ∼
i = Gp (k∞ |kn ) i
instead of TPi for Pi ∈ Spf d (k∞ ). 2
Now we consider the p-parts of the ideal class groups Cl(kn ) of kn . The
theorem of Iwasawa gives an asymptotic formula for
pen := #Cl(kn )(p).
(11.1.6) Theorem (IWASAWA). Let k∞ |k be a ZZp -extension. Then there exist

integers λ = λ(k∞ |k) ≥ 0, µ = µ(k∞ |k) ≥ 0 and ν = ν(k∞ |k), all independent
of n, such that
en = λn + µpn + ν for all n large enough.
Proof: By (11.1.5), we have
Cl(kn )(p) ∼ ω
= G(H|k∞ )/ ωnn G(H|Hn0 k∞ )
0
for n ≥ n0 , so that
#Cl(kn )(p)
= #(G(H|k∞ )/G(H|Hn0 k∞ )) · #(G(H|Hn0 k∞ )/ ωωnn G(H|Hn0 k∞ ))
0
= #G(Hn0 |kn0 ) · #(G(H|Hn0 k∞ )/ ωωnn G(H|Hn0 k∞ )).

0
Now for n large enough, (5.3.17) implies the result, since the first factor in the
equality above is finite of an order independent of n. 2

Remarks: 1. With exactly the same arguments one can also prove such an
asymptotic formula for the p-parts of the Sp -ideal class groups ClSp (kn ), since
ClSp (kn )(p) ∼
= G(Hn0 |kn ).
2. For the invariants λ and µ we obviously have
µ(k∞ |kn ) = pn µ(k∞ |k),
λ(k∞ |kn ) = λ(k∞ |k).
3. If one considers the analogous situation for a function field and the ZZp -
extension given by enlarging the field of constants, then it is known that µ = 0,
i.e. G(L|k∞ )ab is a finitely generated ZZp -module. The same is conjectured for
the cyclotomic ZZp -extension of number fields, but is (up to now) only proven
for abelian extensions k over Q. This is the famous result of B. FERRERO and
L. C. WASHINGTON [49]. Another proof, using p-adic L-functions, was given
later by W. SINNOTT [218]. For an arbitrary ZZp -extension the assertion µ = 0
is not true; it can become arbitrary large [89].
If k is totally real, there is a generalization of the Vandiver conjecture (i.e. p
does not divide the class number of Q(ζp )+ ) but it is in some sense weaker: for
the cyclotomic ZZp -extension k∞ of k, the Iwasawa invariants are
λ=0 and µ = 0,
which means that the Λ-module G(H|k∞ ) is finite. This conjecture is due to
R. GREENBERG, cf. [62], and is widely believed to be true. However, as far
as we know, it is only verified in special cases. For p = 3 and small degrees
of k|Q, several mathematicians (e.g. [114], [233]) have carried out extensive
computations verifying the Greenberg conjecture for many fields.
4. The invariants for the (cyclotomic) ZZp -extension Q∞ of Q are
µ(Q∞ |Q) = 0 = λ(Q∞ |Q)
since s(Q∞ |Q) = 1 and n0 (Q∞ |Q) = 0, and so by (11.1.5) (iii) we have
G(H|Q∞ )Γ ∼ = G(H0 |Q) = 0 ;
thus we even get H = Q∞ .
The following proposition is a generalization of a classical result due to

H. WEBER and PH. FURTWÄNGLER for the field k = Q(µp ).
(11.1.7) Proposition. Let k∞ |k be a ZZp -extension in which exactly one prime

is ramified. Assume this prime is totally ramified; then
e0 = 0 implies en = 0 for all n ≥ 0 .

Proof: Using (11.1.5) (iii) and the assumptions s∞ = 1 and n0 (k∞ |k) = 0,
we get
G(H|k∞ )Γ ∼ = G(H0 |k).
Therefore e0 = 0 implies that H = k∞ and consequently
G(Hn |kn ) ∼
= G(H|k∞ )Γ = 0. n 2
There is a remarkable duality property between the inductive limit and the
projective limit of Cl(kn ) (and an analogous result holds for ClSp (kn )). In
order to state the next theorem, recall that M ◦ denotes the Λ-module M with
the inverse action of Γ , cf. (5.5.12).
(11.1.8) Theorem. Let k∞ |k be a ZZp -extension. Then there are isomorphisms

and pseudo-isomorphisms of Λ-torsion modules
(i) Hom(Cl(k∞ ), Qp /ZZp ) ∼
= E 1 (G(H|Hn0 k∞ )) ≈ Xnr ◦
,
(ii) Hom(ClS (k∞ ), Qp /ZZp ) ∼
p = E 1 (G(H 0 |H 0 k∞ )) ≈ X ◦ .
m0 cs
Proof: We will only prove the first statement since (ii) follows in exactly the
same way. By (11.1.5)(ii), we have
Cl(kn )(p) ∼
= G(Hn |kn ) ∼
= G(H|k∞ )/νn,n0 G(H|Hn0 k∞ )
where νn,n0 = ωn . Since G(H|k∞ )/G(H|Hn0 k∞ ) ∼
ωn
= G(Hn0 |kn0 ) is finite, it
0
follows that
Cl(k∞ )(p) = lim Cl(kn )(p) ∼
= lim G(H|k∞ )/νn,n G(H|Hn k∞ )0 0
−→ −→
n n
∼
= lim G(H|Hn0 k∞ )/νn,n0 G(H|Hn0 k∞ ) .
−→
n
Since G(H|Hn0 k∞ )/νn,n0 G(H|Hn0 k∞ ) is finite, the principal ideals (νn,n0 ),

n ≥ n0 , are disjoint to the prime ideals of height 1 in supp(G(H|Hn0 k∞ )), and
so we obtain a canonical isomorphism
Hom(Cl(k∞ ), Qp /ZZp ) ∼= α(G(H|Hn0 k∞ )) ,
where α denotes the Iwasawa-adjoint. Using (5.5.6), it follows that
Hom(Cl(k∞ ), Qp /ZZp ) ∼= E 1 (G(H|Hn0 k∞ )) ,
and by (5.5.13) we get a pseudo-isomorphism
E 1 (G(H|Hn0 k∞ )) ≈ G(H|Hn0 k∞ )◦ ≈ Xnr ◦
. 2
From the Hochschild-Serre spectral sequence for n ≤ m ≤ ∞,

H i (Γn /Γm , H j (G(kS |km ), OS× )) ⇒ H i+j (G(kS |kn ), OS× ) ,


0 −→ H 1 (Γn /Γm , Ok×m ,S ) −→ ClS (kn ) −→ ClS (km )Γn .
Here S denotes a finite set of primes of k containing Σ = Sp ∪ S∞ . We are
interested in the behaviour of the kernel H 1 (Γn /Γm , Ok×m ,S ) in the tower k∞ |k.
This kernel is the obstruction to the capitulation of prime ideals of Okn ,S in
the ring Okm ,S .
(11.1.9) Proposition. Let k∞ |k be a ZZp -extension and let S ⊇ Σ be a finite set

of primes of k. Then
(i) the order of ker (ClS (kn )(p) −→ ClS (km )(p)) = H 1 (Γn /Γm , Ok×m ,S ) is
bounded independently of n ≤ m ≤ ∞,
(ii) the order of ker (Cl(kn )(p) −→ Cl(km )(p)) is bounded independently
of n ≤ m ≤ ∞.
Proof: We only show (ii) since the proof of (i) is similar. We may assume
that m ≥ n ≥ n1 for some fixed n1 ≥ n0 (k∞ |k). Let Y = G(H|Hn0 k∞ ). By
(11.1.5)(ii), we have a commutative diagram
νm,n
Xnr /νìíîï n,n0 Y Xnr /νm,n0 Y
Cl(kn )(p) Cl(km )(p)

where νm,n = ωωmn and m < ∞. Since Xnr ≈ Y ≈ Y /T0 (Y ) =: Ȳ , where
T0 (Y ) is the maximal finite Λ-submodule of Y , it is enough to show that the
map
ð Ȳ νm,n Ȳ /νm,n Ȳ
Ȳ /νn,n1 1
has kernel whose order is bounded independently of m ≥ n ≥ n1 . But νm,n is

even injective if n1 is large enough. Indeed, as we saw in the claim contained
in the proof of (5.3.17), νm,n
Ȳñ Ȳ
is injective for ∞ > m n n1 . So the commutative and exact diagram
≥ ≥
νn,n1
0øûüùúôõö÷òó Ȳ Ȳ Ȳ /νn,n1 Ȳ 0
νm,n νm,n
νm,n1
0 Ȳ Ȳ Ȳ /νm,n1 Ȳ 0
implies the result for finite m. Passing to the limit, we obtain the assertion for
m = ∞. 2

§2. Iwasawa Theory for p-adic Local Fields 731
§2. Iwasawa Theory for p-adic Local Fields
Let k be a finite extension of Q` and let k∞ |k be a ZZp -extension with

Galois group Γ = G(k∞ |k), where p and ` may be equal. In this section we
want to determine the Λ-module structure of the projective limit A(k∞ ) of the
p-completion A(kn ) of the multiplicative group of kn :
m
A(k∞ ) = lim A(kn ) = lim kn× /kn×p
←− ←−
n n,m
where the limit lim n is taken with respect to the norm maps. By local class
←−
field theory, we have an isomorphism
∼
A(k∞ ) −→ G(k(p)|k∞ )ab .
In particular, A(k∞ ) is a finitely generated Λ-module, since the ZZp -module

G(k(p)|k∞ )ab
Γ
⊆ G(k(p)|k)ab is finitely generated, cf. (5.3.10).
We want to consider a slightly more general situation. Let K|k be a finite

Galois extension of degree prime to p and let K∞ = Kk∞ . Let
G = G(K∞ |k) = Γ × ∆ ,
where
Γ = G(K∞ |K) and ∆ = G(K∞ |k∞ ) .
m
Then A(K∞ ) = lim
←−
L× /L×p , where L runs through all finite subexten-
L,m
sions of K∞ |k, is a finitely generated ZZp [[G]]-module.
Let G = G(k̄|k) and H = G(k̄|K∞ ), so that
1 −→ H −→ G −→ Γ × ∆ −→ 1
is exact and
X = H ab (p) ∼
= A(K∞ ).
(11.2.1) Proposition. pd ZZp [[G]] A(K∞ ) ≤ 1.
Proof: From (7.2.5) we know that scdp G = 2 and therefore the result follows
from (5.6.13). 2
(11.2.2) Lemma. There exists a canonical ZZp [[G]]-isomorphism
= (D2 (G )H )∨ = µ(K∞ )(p)∨ .

E 1 (A(K∞ )) ∼ (p)

Proof: Since IG is projective by (5.2.13), we have (using the notation of

V §6) an isomorphism Y ∼
= X ⊕ IG and, in particular,
X ' D((D2(p) (G )H )∨ ) = D(µ(K∞ )(p)∨ )
by (7.2.4) and (5.6.9) (observe that cdp G = 2 by (7.1.8)(i) and N ab (p) is a
finitely generated ZZp [[G ]]-module by (7.4.1)). Applying the functor E 1 gives
us E 1 (X) ∼
= E 1 (D(µ(K∞ )(p)∨ )) = µ(K∞ )(p)∨ . 2
(11.2.3) Theorem. Let k be a finite extension of Q` of degree n = [k : Q` ]

and let k∞ |k be a ZZp -extension. Let K|k be a finite Galois extension of degree
prime to p, K∞ = Kk∞ and G = Γ × ∆, where Γ = G(K∞ |K) ∼ = ZZp and
∆ = G(K∞ |k∞ ). Let µ(K∞ )(p) be the group of roots of unity of p-power order
in K∞ .
(i) Let ` = p.
(1) If µ(K∞ )(p) is infinite, so that K∞ |K is the cyclotomic ZZp -extension,
then
A(K∞ ) ∼ = ZZp [[G]]n ⊕ ZZp (1) .
(2) If µ(K∞ )(p) is finite, then there exists an exact sequence of ZZp [[G]]-
modules
0 −→ A(K∞ ) −→ ZZp [[G]]n −→ µ(K∞ )(p) −→ 0.
(ii) Let ` =/ p. Then

ZZp (1) , if µp ⊂ K ,
(
A(K∞ ) ∼
=
0, otherwise .
Proof: As we saw in the proof of (11.2.2), we have X ' D(µ(K∞ )(p)∨ ). Let
us assume that ` = p. It follows from (5.4.9)(ii), (5.4.15) and the fact that the
dualizing module at p of the group G is D1(p) = Qp /ZZp (since it is equal to the
dualizing module of its open subgroup Γ ), that
ZZp (1) , if µ(K∞ )(p) is infinite,
(
E (DX) = E (µ(K∞ )(p) ) ∼
1 1
= ∨
0, otherwise,
0, if µ(K∞ )(p) is infinite,
(
E (DX) = E (µ(K∞ )(p) ) ∼
2 2
= ∨
µ(K∞ )(p) , otherwise.
++
The ZZp [[G]]-module X is projective. Indeed, this follows from (5.4.16) and
the fact that X ++ is free (hence projective) as a ZZp [[Γ ]]-module by (5.1.9).
From the exact sequence (5.4.9)(iii)
0 −→ E 1 (DX) −→ X −→ X ++ −→ E 2 (DX) −→ 0 ,
we obtain

§2. Iwasawa Theory for p-adic Local Fields 733
XU++ ⊗ Qp ∼
= XU ⊗ Q p
for an open normal subgroup U of G (observe that E 1 (DX)U is finite). If
G/U = G(k 0 |k), then the exact sequence
0 ab
0 −→ G(k̄|K∞ )ab ab
U (p) −→ G(k̄|k ) (p) −→ U (p) −→ 0
implies that
Qp ⊕ X U ⊗ Qp ∼
= Qp [G/U ]n ⊕ Qp ,
where we use (7.4.4)(i). It follows that
X U ⊗ Qp ∼ = Qp [G/U ]n ,
and so
X ++ ∼
= ZZp [[G]]n
by (5.6.11). This already gives us the assertion if µ(K∞ )(p) is finite, and if not
it suffices to observe that X ++ is projective.
The case ` =/ p is trivial, since then K∞ is the maximal unramified p-
extension of K and G(K̄|K∞ )(p) = G(Ktr |K∞ )(p) where Ktr is the maximal
tamely ramified extension of K. The latter Galois group is isomorphic to ZZp (1)
if µp ⊆ K and is zero otherwise; see VII §5. 2
When ` = p, we are also interested in the ZZp [[G]]-structure of the projective

limit of the principal units
U 1 (K∞ ) = lim U 1 (L) ,
←−
L
where L runs through all finite subextensions of K∞ |k.
(11.2.4) Theorem. With the notation and assumptions of (11.2.3), suppose

that ` = p. Then the following holds:
(i) If µ(K∞ )(p) is infinite, then
U 1 (K∞ ) ∼
= ZZp [[G]]n ⊕ ZZp (1).
(ii) If µ(K∞ )(p) is finite, then U 1 (K∞ ) is a submodule of ZZp [[G]]n of finite
index equal to or less than #µ(K∞ )(p).
(iii) Let K∞ |K be unramified, then
U 1 (K∞ ) ∼
= A(K∞ ).
Proof: Consider the inverse system of exact sequences over the finite exten-
sions Kn |K
0 −→ U 1 (Kn )/pm −→ A(Kn )/pm −→ ZZ/pm −→ 0 ,

where the transition maps in the middle and on the left are the norm maps and
are multiplication by the residue degree on the value groups. Therefore
lim ZZ/pm = lim (lim ZZ/pm ) = 0
←− ←− ←−
m,n m n
if K∞ |K is unramified, and we have proved assertion (iii).

Assume now that K∞ |K is ramified. Then we get an exact sequence
v
0 −→ U 1 (K∞ ) −→ A(K∞ ) −→ ZZp −→ 0
where v denotes the valuation. Since pd ZZp [[G]] A(K∞ ) ≤ 1 and pd ZZp [[G]] ZZp ≤ 1,
we see that pd ZZp [[G]] U 1 (K∞ ) ≤ 1.
If µ(K∞ )(p) is infinite, then by (11.2.3) A(K∞ ) ∼ = ZZp [[G]]n ⊕ ZZp (1). Since
ZZp is a trivial ZZp [[G]]-module, the factor ZZp (1) of A(K∞ ) has to be in the kernel
U 1 (K∞ ) of v . Changing the basis of ZZp [[G]]n , we obtain a commutative exact
diagram
0ýþÿ U 1 (K∞ ) A(K∞ ) ZZp 0
0 IG ⊕ ZZp [[G]]n−1 ⊕ ZZp (1) ZZp [[G]]n ⊕ ZZp (1) ZZp 0,

U 1 (K∞ ) ∼
= IG ⊕ ZZp [[G]]n−1 ⊕ ZZp (1) ∼
= ZZp [[G]]n ⊕ ZZp (1) ,
where we use (5.6.12). This proves (i).
Now we assume that µ(K∞ )(p) is finite. Since pd ZZp [[G]] U 1 (K∞ ) ≤ 1, we
have
D(U 1 (K∞ )) ' E 1 (U 1 (K∞ ))
by (5.4.11). It follows from (11.2.3), using (5.5.8)(iv) and (5.4.17), that
E 1 (D(U 1 (K∞ ))) = E 1 (E 1 (U 1 (K∞ ))) = T1 (U 1 (K∞ )) ⊆ T1 (A(K∞ )) = 0 .
Further, E 1 (A(K∞ )), which is isomorphic to µ(K∞ )(p)∨ by (11.2.2), surjects
onto E 1 (U 1 (K∞ )). Thus by (5.4.15)(ii) we get
E 2 (D(U 1 (K∞ ))) = E 2 (E 1 (U 1 (K∞ ))) = E 1 (U 1 (K∞ ))∨ ⊆ µ(K∞ )(p) .
As in the proof of theorem (11.2.3), U 1 (K∞ )++ is ZZp [[G]]-projective,
U 1 (K∞ )++ ∼ 1
U ⊗ Qp = U (K∞ )U ⊗ Qp for U
E G open, and from the exact
sequence
0 −→ U ab (p) −→ U 1 (K∞ )U −→ A(K∞ )U −→ ZZp −→ 0
it follows that
Qp ⊕ U 1 (K∞ )U ⊗ Qp ∼
= A(K∞ )U ⊗ Qp ⊕ Qp
∼
= Qp [G/U ]n ⊕ Qp .

§3. The Maximal Abelian p-Extension of k∞ Unramified Outside S 735
Thus
U 1 (K∞ )U ⊗ Qp ∼
= Qp [G/U ]n ,
and it follows by (5.6.11) that
U 1 (K∞ )++ ∼
= ZZp [[G]]n .
Putting everything together, the exact sequence
0 −→ E 1 (D(U 1 (K∞ ))) −→ U 1 (K∞ ) −→ U 1 (K∞ )++ −→ E 2 (D(U 1 (K∞ )))
implies (ii). This finishes the proof of the theorem. 2
Exercise: Calculate the quotient ZZp [[G]]n /U 1 (K∞ ) in theorem (11.2.4)(ii).
§3. The Maximal Abelian p-Extension of k∞

Unramified Outside S
Let k∞ be a ZZp -extension of the number field k and let S be a finite set of
primes of k containing Σ = Sp ∪ S∞ . We assume throughout this section that
k is totally imaginary if p = 2. In §1 we considered the Λ-modules
Xnr = G(L|k∞ )ab ,
Xcs = G(L0 |k∞ )ab ,
where L is the maximal unramified p-extension of k∞ and L0 |k∞ its maximal
subextension which is completely decomposed at p. Now we are interested in
the Λ-module
XS = G(kS (p)|k∞ )ab .
For S = Σ we set
X = G(kΣ (p)|k∞ )ab .
(11.3.1) Proposition. Let S ⊇ Σ be a finite set of primes of k. Then the

Λ-module XS is finitely generated.
Proof: This follows from (5.3.10) and (8.3.20) since

dimIFp (XS /p)Γ = dimIFp H 1 (G(kS |k), ZZ/pZZ) − 1. 2

The Iwasawa invariants of XS are denoted by

µS = µ(XS ) and λS = λ(XS )
and we set
µ = µ(XΣ ) and λ = λ(XΣ ) .
Since we have already denoted the Iwasawa invariants of Xnr by µ and λ in
§1, we will use the notation
µnr = µ(Xnr ) , λnr = λ(Xnr ) ,
µcs = µ(Xcs ) , λcs = λ(Xcs ) ,

for the rest of this chapter.
In the following, the weak Leopoldt conjecture, i.e. the vanishing of the
group H 2 (G(kΣ |k∞ ), Qp /ZZp ), will play an important role. The next theorem
collects the results on this conjecture which we already know from (10.3.22),
(10.3.25) and (10.3.23).
(11.3.2) Theorem.
(i) H 2 (G(kS |k∞ ), Qp /ZZp ) = 0 if and only if H 2 (G(kΣ |k∞ ), Qp /ZZp ) = 0.
(ii) H 2 (G(kΣ |k∞ ), Qp /ZZp ) = 0 if k∞ |k is the cyclotomic ZZp -extension.
(iii) If the (strong) Leopoldt conjecture holds, i.e. H 2 (G(kΣ |k), Qp /ZZp ) = 0,
then the weak conjecture is also true.
(iv) The weak Leopoldt conjecture is true if and only if pdΛ XS ≤ 1 and
rankΛ XS = r2 , where r2 denotes the number of complex places of k. In
particular, in this case XS is a Λ-torsion module if k is totally real.
The dependence between the weak and strong form of this conjecture is
contained in the following
S
(11.3.3) Proposition. Let k∞ = kn be a ZZp -extension of k.
(i) The (strong) Leopoldt conjecture holds for k if and only if the weak form
is satisfied and X Γ = 0.
(ii) The following assertions are equivalent:
(1) The Leopoldt conjecture is true for all layers kn .
(2) There exists some n1 ≥ λ such that the Leopoldt conjecture
holds for kn1 .

Proof: (i) follows from the exact sequence (which is obtained from the
Hochschild-Serre spectral sequence)
0 → H 1 (Γn , X ∨ ) → H 2 (G(kΣ |kn ), Qp /ZZp ) → H 2 (G(kΣ |k∞ ), Qp /ZZp )Γn → 0
for n = 0. In order to prove the nontrivial implication (2) ⇒ (1) of (ii), we
first observe that the weak Leopoldt conjecture is true and X Γn1 = 0 by (i).
Thus the set of prime divisors of ωn1 Λ is disjoint from the set of prime ideals
of height 1 in supp(torΛ X). Since
ωn = ξ0 · ξ1 · · · ξn
where p−1
k−1
(1 + T )ip
X
ξ0 = ω0 = T , ξk = for k ≥ 1,
i=0
are the irreducible cyclotomic polynomials, it follows that (ξn ) ∈
/ P (X) if
n ≤ n1 . But the same holds for n > n1 because
deg(ξn ) = (p − 1)pn−1 ≥ n > n1 ≥ λ = deg(FX ) ,
where FX is the characteristic polynomial of TΛ (X). Thus X Γn is finite, hence
zero, for all n (observe that X Γn is ZZp -free by (5.3.19)(i)). Since the weak
Leopoldt conjecture holds, the result follows. 2
Recall the notation S f for the subset of finite primes in S. In order to

consider the relation between XS and X we introduce the following
(11.3.4) Definition. Let k∞ |k be a ZZp -extension and let S be a set of primes

of k. Then
S f d (k) = {p ∈ S f (k) | p is finitely decomposed in k∞ |k},
S cd (k) = {p ∈ S f (k) | p is completely decomposed in k∞ |k}.
Furthermore, in this section we will use the following

Notation: If G is a profinite group, H a closed subgroup of G and M a
compact ZZp [[H]]-module, then
IndH ˆ
G M := M ⊗ ZZp [[H]] Z
Zp [[G]]
denotes the compact induction of M from H to G. The adjunction between
⊗ˆ and Hom implies the following compatibility formula, relating the compact
induction to the discrete (co)induction defined in I §6
∨ ∨
(IndH H
G M ) = IndG (M ).
This notation may lead to confusion if M is finite, i.e. compact and discrete,
at least if H is of infinite index in G. However, it should always be clear

from the context which induction is meant. In this section we will only use
the compact induction which, unraveling the definition of the complete tensor
product, satisfies the formula
HN/N
IndH
G M = lim
←−
IndG/N M/IN ∩H M ,
N
where N runs through all open normal subgroups of G.
(11.3.5) Theorem. Assume that the weak Leopoldt conjecture holds for the
ZZp -extension k∞ |k and let S ⊇ Σ be finite. Then there exists a canonical exact
sequence of Λ-modules
M Γ
0 −→ IndΓp (T (kp (p)|kp )Gk∞,p ) −→ XS −→ X −→ 0.
p∈S\Σ(k)
In particular, there is an exact sequence of Λ-torsion modules

M Γ
0 −→ IndΓp (T (kp (p)|kp )Gk∞,p ) −→ T1 (XS ) −→ T1 (X) −→ 0.
p∈S\Σ(k)
Proof: Since H 2 (G(kΣ (p)|k∞ ), Qp /ZZp ) = 0, we have an exact sequence

0 −→ H 1 (kΣ (p)|k∞ ) −→ H 1 (kS (p)|k∞ ) −→ H 1 (kS (p)|kΣ (p))GΣ (k∞ ) −→ 0
with coefficients in Qp /ZZp or, dually, using (10.5.4)
Y
0 −→ lim T (kp (p)|kp )Gkn,p −→ XS −→ X −→ 0.
←−
n p∈S\Σ(kn )
Suppose µp ⊆ kp for p ∈ S\Σ (otherwise T (kp (p)|kp ) = 0). Then T (kp (p)|kp ) =
T (kp (p)|kp )Gk∞,p is isomorphic to ZZp (1) if p is finitely decomposed in k∞ |k,
and otherwise T (kp (p)|kp )Gk∞,p = µ(k∞,p )(p) = µ(kp )(p) is finite. Therefore
T (kp (p)|kp )Gkn,p ∼ Λ/ptp
Y M M
lim
←−
= Λ/ωrp (1) ⊕
n p∈S\Σ(kn ) p∈(S\Σ)f d (k) p∈(S\Σ)cd (k)
µp ⊆kp
where prp = [Γ : Γp ] and p tp = #µ(kp )(p) (cf. (11.3.4) for the definition of
(S\Σ)f d and (S\Σ)cd ). In particular, this module is Λ-torsion and therefore
the second statement follows from the first. 2
(11.3.6) Corollary. Let k∞ |k be a ZZp -extension for which the weak Leopoldt
conjecture is true and let S ⊇ Σ be finite. Then
X
(i) µS = µ + tp , where p tp = #µ(kp )(p),
p∈(S\Σ)cd (k)
(ii) λS = λ + #{p ∈ (S\Σ)f d (k∞ ) | µp ⊆ kp }.

The µ-invariant not only influences the abelian pro-p-group XS but its van-
ishing implies an important property of the whole Galois group G(kS (p)|k∞ ).
(11.3.7) Theorem. Assume the weak Leopoldt conjecture holds for the ZZp -
extension k∞ |k. Then G(kS (p)|k∞ ) is a free pro-p-group if and only if µS = 0.
In particular, if k∞ is the cyclotomic ZZp -extension of k, then G(kS (p)|k∞ )
is free if and only if µS = 0.
Proof: Using the assumption and (10.4.8), we have

H 2 (kS (p)|k∞ , Qp /ZZp ) = H 2 (kS |k∞ , Qp /ZZp ) = 0 .
Furthermore, since cdp G(kS (p)|k) ≤ 2 (recall that k is totally imaginary if
p = 2), the group H 2 (kS (p)|k, Qp /ZZp ) is p-divisible. Now the desired result
We are interested in how properties of XS are affected by change of base

field. This is no longer an abelian question. The next theorem shows that the
validity of the weak Leopoldt conjecture and the vanishing of the µ-invariant
are properties which have to be considered simultaneously. From (5.6.19)
follows the
(11.3.8) Theorem. Let K|k be a finite Galois p-extension inside kS , k∞ |k be

a ZZp -extension and K∞ = Kk∞ . Then
( ) ( )
µS (K∞ |K) = 0 and µS (k∞ |k) = 0 and
⇐⇒
H 2 (G(KS |K∞ ), Qp /ZZp ) = 0 H 2 (G(kS |k∞ ), Qp /ZZp ) = 0 .
In particular, if k∞ |k is the cyclotomic ZZp -extension, then
µ(K∞ |K) = 0 ⇐⇒ µ(k∞ |k) = 0.
Proof: The last assertion follows from the first, since the weak Leopoldt con-
jecture holds true for the cyclotomic ZZp -extension and µS = µ by (11.3.6)(i).
2
Now we combine the results of the local and global theory. Let us recall
some notation: let K|k be a finite Galois extension of degree prime to p, k∞ |k
a ZZp -extension and K∞ = Kk∞ . Let G = G(K∞ |k) and let S ⊇ Σ be a finite
set of primes of k large enough so that K ⊆ kS . Let GS = G(KS (p)|k) and
HS = G(KS (p)|K∞ ), so that there is an exact sequence

1 −→ HS −→ GS −→ G −→ 1.
Furthermore, let
1 −→ N −→ Fd −→ GS −→ 1
be a presentation of GS with a free profinite group Fd of finite rank d. We
obtain a commutative exact diagram

N N
(∗) 1 R Fd G 1
1 HS GS G 1.
Finally, let
Y
× ×p m
AS = AS (K∞ ) = lim Ap (Kn ) , Ap (Kn )= lim Kn,p /Kn,p ,
←− ←−
n p∈S f (Kn ) m
m
UKn,p /UKp n,p ,
Y
US = US (K∞ ) = lim
←−
Up (Kn ) , Up (Kn )= lim
←−
n p∈S f (Kn ) m
×
ES = EK∞ ,S = lim
←−
(OK n ,S
⊗ ZZp ) ,
n
×
E = EK∞ = lim
←−
(OK n
⊗ ZZp ) ,
n
where the projective limits are taken with respect to the norm maps and the
canonical projections. Analogously to Xcs we define
S
Xcs = G(LS |K∞ )ab ,
where LS is the maximal unramified p-extension of K∞ which is completely
decomposed at every prime above S (so LS = (K∞ )S∞ ,S (p) with the notation
of (10.4.1)).
(11.3.9) Lemma. pd ZZp [[G]] AS ≤ 1 and pd ZZp [[G]] US ≤ 1.
Proof: Since AS is the direct product of AS f d and AS cd , where S f d and S cd are

the subsets of finite primes of S which are finitely and completely decomposed
in k∞ |k respectively, we consider these factors separately. The assertion for
AS f d follows from (11.2.1) and (5.4.17). The second factor
Y G
AS cd = IndGP AP
p∈S cd

decomposes as a Λ-module into the direct sum of a free Λ-module and a

“µ-part”, i.e. an elementary Λ-module which is ZZp -torsion. In particular,
AS cd has projective Λ-dimension less than or equal to 1 by (5.3.19)(i). Hence
the same holds as a ZZp [[G]]-module by (5.4.17). The result for US follows
analogously using (11.2.4). 2
The following theorem generalizes results from [97].
(11.3.10) Theorem. With the notation as above the following holds:

(i) There is a commutative exact diagram of ZZp [[G]]-modules
0!"#$%& H2 (HS , ZZp ) E US XS Xnr 0
S
0 H2 (HS , ZZp ) ES AS XS Xcs 0.
(ii) There is a canonical exact sequence

M G S
0 −→ E −→ ES −→ IndGp ZZp −→ Xnr −→ Xcs −→ 0 ,
p∈S cd ∪S r
where S cd = {p ∈ S f (k) | p is completely decomposed in k∞ |k} and

S r = {p ∈ S f (k) | p is ramified in k∞ |k}.
(iii) N ab (p) is a finitely generated projective ZZp [[GS ]]-module and
NHabS (p) ∼
G 0
IndGp ZZp ⊕ ZZp [[G]]d−r2 −r1 −1 ,
M
=
0
p∈S∞
0
where S∞ is the set of all real places of k becoming complex in K∞ ,
r1 = #S∞ , and d is chosen greater than or equal to r2 + r10 + 1.
0 0
(iv) Let ZS := (D2(p) (GS )HS )∨ . Then

XS ' DZS .
If H2 (HS , ZZp ) = 0, then there exists a canonical ZZp [[G]]-isomorphism
E 1 (XS ) ∼
= ZS .
Proof: (i) follows from (10.3.13) with T = ∅ by passing to the projective

limit, and (ii) then follows by the snake lemma.

The ZZp [[GS ]]-module N ab (p) is projective by (5.6.7). Considering N ab (p)

as a ZZp [[G(KS (p)|K)]]-module, we see by (5.6.13) that this module is finitely
generated, since H 2 (G(KS (p)|K), ZZ/pZZ) is finite, cf. (8.3.20) and (10.4.8).
Thus N ab (p) is also finitely generated as a ZZp [[GS ]]-module. The structure of
NHabS (p) as a ZZp [[G]]-module can be seen from (5.6.11): First observe that this
module is homotopic to the right-hand side of the isomorphism stated in (iii)
(both modules are projective). In order to prove assumption (ii)0 of (5.6.11), let
Ḡ = G/Γn = G(Kn |k). Then from the diagram (∗) we obtain the commutative
exact diagram
'()*+,-./01234 N N
1 M (n) Fd G 1
1 GS (Kn ) GS G 1
where M (n) is a profinite group defined by the exactness of the upper row. In
the Grothendieck group K00 (Qp [G]), which we denote here by K0 (Qp [G]) since
every finitely generated Qp [G]-module is projective, we obtain using (5.6.6),
[NGabS (Kn ) (p) ⊗ Qp ] = [M (n)ab (p) ⊗ Qp ] + [H2 (GS (Kn ), ZZp ) ⊗ Qp ]

−[GS (Kn )ab (p) ⊗ Qp ]
= [Qp [G]d−1 ] + [Qp ] + [H2 (GS (Kn ), ZZp ) ⊗ Qp ]

−[GS (Kn )ab (p) ⊗ Qp ] .
From (10.3.13), Dirichlet’s unit theorem (8.7.3) and (7.4.4)(i), we then get
[NGabS (Kn ) (p) ⊗ Qp ] = [Qp [G]d−1 ] + [Qp ] − [UKn ,S ⊗ Qp ] +[EKn ⊗ Qp ]
= [Qp [G]d−1 ] + [Qp ] − [Qp [G]2r2 +r1 ]
0
IndG
M
r2 +r1 −r1
+[ G Qp ] + [Qp [G]
p
] − [Qp ]
0
p∈S∞
0
IndG
M
= [Qp [G]d−r2 −r1 −1 ] + [ G Qp ].
p
0
p∈S∞
The proof of (iv) is analogous to (11.2.2). Since cdp G ≤ 1, the augmentation

ideal IG is projective and therefore Y ∼= XS ⊕ IG , so that
XS ' DZS
by (5.6.9) (noting (iii)). Since H2 (HS , ZZp ) = 0, it follows that E 1 (XS ) ∼
= ZS ,
again by (5.6.9). 2

(11.3.11) Theorem. With the notation as above, suppose H 2 (HS , Qp /ZZp ) = 0.

Then the following are true:
(i) pd ZZp [[G]] XS ≤ 1 and
E 0 (XS ) ∼
G
= ZZp [[G]]r2 ⊕ Ind p ZZ− ,
M
G p
0
p∈S∞
where ZZ−p is the Gp -module ZZp on which the generator of Gp ∼
= ZZ/2ZZ
acts as multiplication by −1.
(ii) pd ZZp [[G]] ES ≤ 1 and there is an exact sequence of ZZp [[G]]-modules
05678
G 0
IndGp ZZp ⊕ ZZp [[G]]r2 +r1 −r1
M
ES µp∞ (K∞ ) 0
0
p∈S cd ∪S∞
if µp∞ (K∞ ) is finite, and an isomorphism
ES ∼
G 0
IndGp ZZp ⊕ ZZp [[G]]r2 +r1 −r1 ⊕ ZZp (1)
M
=
0
p∈S∞
if µp∞ (K∞ ) is infinite.
(iii) pd ZZp [[G]] E ≤ 1 and there is an exact sequence of ZZp [[G]]-modules
G 0
IndGp ZZp ⊕ ZZp [[G]]r2 +r1 −r1 −→ µp∞ (K∞ )
M
0 −→ E −→
0
p∈S∞
if µp∞ (K∞ ) is finite, and an isomorphism
E∼
G 0
IndGp ZZp ⊕ ZZp [[G]]r2 +r1 −r1 ⊕ ZZp (1)
M
=
0
p∈S∞
if µp∞ (K∞ ) is infinite.
Proof: (i) By (5.6.12), the augmentation ideal IG is a free ZZp [[G]]-module of

rank 1. Therefore, using (5.6.7), we obtain an exact sequence
0 −→ NHabS (p) −→ ZZp [[G]]d −→ XS ⊕ IG −→ 0 ,
which shows that pd ZZp [[G]] XS ≤ 1 since NHabS (p) is projective. In order to prove
the assertion concerning E 0 (XS ), we proceed as in the proof of part (iii) of
theorem (11.3.10). First observe that E 0 (XS ) is ZZp [[G]]-projective by (5.4.16)
and the fact that it is ZZp [[Γ ]]-free by (5.5.10)(i). From the exact sequence
above, we obtain the exact sequence
0 −→ E 0 (XS ) ⊕ (IG )+ −→(ZZp [[G]]d )+ −→(NHabS (p))+ −→ E 1 (XS ) −→ 0.
For Γn E G let Ḡ = G/Γn = G(Kn |k). Since E 1 (XS ) is a Γ -torsion module
and IG ∼
= ZZp [[G]],
[E 0 (XS )Γn ⊗ Qp ] = [Qp [G]d−1 ] − [(NHabS (p))+Γn ⊗ Qp ]
0
IndG
M
= [Qp [G]r2 +r1 ] − [ G Qp ]
p
0
p∈S∞
−
IndG
M
= [Qp [G]r2 ] + [ G Qp ]
p
0
p∈S∞

in the Grothendieck group K0 (Qp [G]), where we have used (11.3.10)(iii) and
G G G
ZZp [[G]] = IndGp ZZp [Gp ] = IndGp ZZp ⊕ IndGp ZZ− 0
p for p ∈ S∞ . Now we get the
result from (5.6.11).
(ii) Since pd ZZp [[G]] AS ≤ 1 and pd ZZp [[G]] US ≤ 1 by (11.3.9), this also holds for
ES and E (which is easily seen using (5.3.19)(i) and (5.4.17)).
Now we split the 4-term exact sequence (11.3.10)(i)
9:;<=>? B
S
0 ES AS XS Xcs 0
into two short exact sequences and obtain the commutative exact diagram
DEF@ABC S )
E 1 (X E 1 (XS ) E 1 (B) S
E 2 (Xcs )
cs
ϕ
E 1 (XS ) E 1 (AS )
E 1 (ES )
using the fact that E 2 (B) vanishes, being a quotient of E 2 (XS ). We obtain the
commutative diagram with exact rows
GHIJKLMNOPQRST coker(ϕ)
E 1 (XS ) E 1 (AS )
E 1 (ES )
G
(D2(p) (GS )HS )∨ IndGP (D2(p) (Gkp )GK∞,P )∨
M
µp∞ (K∞ )∨
p∈S(k)
M
(lim H 2 (KS |K 0 , ZZ/pm ))HS (lim H 2 (Kp0 , ZZ/pm ))HS µp∞ (K∞ )∨ ,
←− ←− 0
K 0 ,m K 0 ,mS(K )
where K 0 runs through all finite subextensions of KS (p)|k. The lower exact
sequence is obtained by taking the limit and taking HS -coinvariants of the
right-hand part of the Poitou-Tate long exact sequence (8.6.10). The left-hand
vertical isomorphism is assertion (11.3.10)(iv) and the vertical isomorphism in
the middle is obtained as follows. Let p ∈ S be a prime such that µ(K∞,P ) is
infinite for P|p. Then, by (11.2.3), (5.4.15)(i) and observing that
D1(p) (G) = Qp /ZZp ,

we obtain
APn (Kn )) = E 1 (IndGP ZZp (1)) ∼
G G
= IndGP (µ(K∞,P )(p))∨ .
Y
E 1 (lim
←−
n Pn |p
If p ∈ S is finitely decomposed in k∞ |k and µ(K∞,P ) is finite for P|p, then by

(11.2.3) and (5.4.15)(ii) we obtain
APn (Kn )) = E 2 (IndGP (µ(K∞,P )(p))) ∼
G G
= IndGP (µ(K∞,P )(p))∨ .
Y
E 1 (lim
←−
n Pn |p
For a prime p ∈ S which splits completely in k∞ |k, we have by (5.4.13)(iii)

Y G
E 1 (lim APn (Kn )) = E 1 (IndGP (µ(KP )(p)))
←−
n Pn |p
∼ G
= D0 ((IndGP (µ(KP )(p)))∨ )∨
G
= IndGP (µ(KP )(p))∨
(observe that in this case µ(KP )(p) = µ(K∞,P )(p)). Finally, we recall that
µ(K∞,P )(p) = D2(p) (Gkp )GK∞,P by (7.2.4).
From the diagram above, it now follows that
E 1 (ES ) ∼
= µp∞ (K∞ )∨ .
Because pd ZZp [[G]] ES ≤ 1, the last isomorphism implies that DES ' µp∞ (K∞ )∨
by (5.4.11), and therefore
ZZp (1) , if µp∞ (K∞ ) is infinite,
(
T1 (ES ) = E (DES ) ∼
1
=
0, otherwise,
0, if µp∞ (K∞ ) is infinite,
(
T2 (ES ) = E 2 (DES ) ∼
=
µp∞ (K∞ ) , otherwise.
In order to calculate ES++ , we proceed as in (iii) of (11.3.10): ES++ is ZZp [[G]]-
projective (because ES++ is Λ-projective) and
[(ES++ )Γn ⊗ Qp ] = [(ES )Γn ⊗ Qp ]
in K0 (Qp [G]) for G = G/Γn . We have exact sequences
UVWXYZ[\]^_` 0 B Γn (ES )Γn (AS )Γn BΓn 0
0 B Γn X S Γn S Γn
(Xcs ) BΓn (XS )Γn S
(Xcs )Γn 0
(observe that (AS )Γn = 0 by (11.2.3)), and from the exact sequence in the proof
of (i), we get an exact sequence
0 −→ XSΓn −→ (NHabS (p))Γn −→ ZZp [G]d −→(XS )Γn ⊕ (IG )Γn −→ 0 .

S
Thus the facts that Xcs is Λ-torsion, IG ∼
= ZZp [[G]], (11.3.10)(iii) and (11.2.3),
imply that
[(ES )Γn ⊗ Qp ] = [(AS )Γn ⊗ Qp ] + [B Γn ⊗ Qp ] − [BΓn ⊗ Qp ]
= [(AS )Γn ⊗ Qp ] + [XSΓn ⊗ Qp ] − [(XS )Γn ⊗ Qp ]
= [(AS )Γn ⊗ Qp ] + [(NHabS (p))Γn ⊗ Qp ] − [Qp [G]d ]
+ [(IG )Γn ⊗ Qp ]
0
IndG
M
r2 +r1
= [(AS )Γn ⊗ Qp ] + [ G Qp ] − [Qp [G]
p
]
0
p∈S∞
0
IndG IndG
M M
= [Qp [G]r2 +r1 −r1 ] + [ G Qp ] + [
p
G Qp ] .
p
0
p∈S cd p∈S∞
It follows that
ES++ ∼
G 0
IndGp ZZp ⊕ ZZp [[G]]r2 +r1 −r1
M
=
0
p∈S cd ∪S∞
and from (5.4.9) (iii), we get the result for ES .
Assertion (iii) for E follows the same lines as that for ES by considering US
instead of AS and noting that
[(US )Γn ⊗ Qp ] = [Qp [G]2r2 +r1 ]
and
T1 (E) = T1 (ES ) , a S)
E 1 (E E 1 (E)
(which follows from (11.3.10)(ii)). 2
Remark: Let S ⊇ Σ be a finite set of primes and

E S (K∞ ) = lim
←−
E (S)
Kn ,S and E(K∞ ) = lim
←−
E (S)
Kn ,
n n
(S) (S) ×
where E Kn ,S (resp. E Kn ) is the closure of the image of OK in AS (Kn ) (resp.
n ,S
×
of OKn in US (Kn )) with respect to the idèle topology, cf. p.628. We have
proved that E S (K∞ ) = EK∞ ,S (resp. Ē(K∞ ) = EK∞ ) if and only if the weak
Leopoldt conjecture holds, i.e. H 2 (HS , Qp /ZZp ) = 0, cf. (10.3.24). Therefore
we may replace the ZZp [[G]]-modules ES and E by E S and E in (11.3.11).
(11.3.12) Corollary. Let K∞ |K be a ZZp -extension for which the weak

Leopoldt conjecture holds and let S ⊇ Σ be a finite set of primes of K.
Then
rankΛE S = r2 + r1 + dS and rankΛE = r2 + r1 ,
where dS is the cardinality of the set S cd of finite primes in S which completely
decompose in K∞ |K.

(11.3.13) Corollary. Let K = k(µp ) and ∆ = G(K|k), k∞ |k the cyclotomic

ZZp -extension and K∞ = Kk∞ = k(µp∞ ) with Galois group G(K∞ |k) = Γ ×∆.
Then there is an exact sequence of Λ[∆]-modules
0bcdefghijklmnopqr Xcs (−1) E 1 (XS ) E 1 (AS ) E 1 (ES ) 0
M (Γ ×∆)
0 Xcs (−1) ZS Ind Γ ×∆ p ZZp (−1) ZZp (−1) 0.
p∈S f
Proof: We merely have to calculate the kernel of E 1 (XS ) → E 1 (AS ). From

the diagram in the proof of (11.3.11), we see that this is
lim
←−
X2 (Kn , ZZ/pm ) ∼
= lim
←−
X1 (Kn , µpm )∗
n,m n,m
= lim lim ClS (Kn )/pm (−1)

←− ←−
m n
= lim lim ClΣ (Kn )/pm (−1)

←− ←−
m n
= Xcs (−1) . 2
(11.3.14) Corollary. With the assumptions of (11.3.13), there is an exact

sequence of Λ[∆]-modules
0 −→ T1 (ES ) −→ T1 (AS ) −→ T1 (XS ) −→ E 1 (Xcs (−1)) −→ 0.
Proof: Splitting the exact sequence in (11.3.13) into two short exact se-
quences
0stuvwxyz Xcs (−1) E 1 (XS ) C 0,
0 C E 1 (AS ) E 1 (ES ) 0,
and applying the functor E 1 , we obtain the exact sequence
0 → E 1 E 1 (ES ) → E 1 E 1 (AS ) → E 1 E 1 (XS ) → E 1 (Xcs (−1)) → E 2 (C)
because E 2 E 1 (ES ) = 0 by (11.3.11)(ii), (5.5.3)(iv) and (5.4.17). But the
projective dimensions of ES , AS and XS are less than or equal to 1, so we can
replace E 1 E 1 by E 1 D = T1 (5.4.11). Furthermore, E 2 (C) = 0, since
pdΛ[∆] C ≤ pdΛ[∆] E 1 (AS ) ≤ 1.
2

(11.3.15) Corollary. With the assumptions of (11.3.13) the following holds:

(i) T0 (XS ) = 0, T1 (XS ) = E 1 (ZS ), T2 (XS ) = T0 (Xcs (−1))∨ .
(ii) T0 (Xcs ) ∼
= lim H 1 (G(K∞ |Kn ), O× ) ,
←− K∞ ,S
n
where the limit is taken with respect to the corestriction maps.
(iii) There is an exact sequence of Λ[∆]-modules
∆
0 → T1 (XS ) → XS → Λ[∆]r2 ⊕ Ind∆p Λ− → T0 (Xcs (−1))∨ → 0 ,
M
0
p∈S∞
where Λ− is the Λ[∆p ]-module Λ with ∆p ∼ = ZZ/2ZZ acting by −1. In

particular, FΛ (XS ) = XS /T1 (XS ) is a free Λ-module if and only if Xcs
contains no finite nontrivial Λ-submodule.
Proof: (i) It remains only to show the last assertion, which is obtained as
follows:
T2 (XS ) = E 2 (DXS ) = E 2 (E 1 (XS )) = E 2 (T0 (E 1 (XS )))
= T0 (E 1 (XS ))∨ = T0 (Xcs (−1))∨
where we have used (5.4.11), (5.5.3)(iv) (together with (5.4.17) and the remark
that for a finitely generated ZZp [[G]]-module M , the ZZp [[Γ ]]-module T0 (M ) is
also a ZZp [[G]]-module), (5.4.15)(ii) and (11.3.13).
(ii) follows from the commutative exact diagram for m ≥ n
0{|}~ ×
H 1 (K∞ |Km , OK ∞ ,S
) ClS (Km ) ClS (K∞ )Γm
cor N N
×
0 H 1 (K∞ |Kn , OK ∞ ,S
) ClS (Kn ) ClS (K∞ )Γn
which induces the exact sequence
×
0 −→ lim H 1 (K∞ |Kn , OK ∞ ,S
) −→ Xcs −→ lim ClS (K∞ )(p)Γn .
←− ←−
n n
Since
Γ
lim ClS (K∞ )(p)Γn = lim ClS (K∞ )(p)Γ
←− ←−
n n
is uniquely p-divisible (the inverse limit on the right-hand side is taken via
multiplication by p), we see that lim ClS (K∞ )(p)Γ has no nontrivial finite
←− n
Γ -submodule. Thus the same is true for lim ClS (K∞ )(p)Γn . Since the
←− n
×
order of H 1 (K∞ |Kn , OK ∞ ,S
) is bounded independently of n by (11.1.9)(i),
× ×
lim H (K∞ |Kn , OK∞ ,S ) is finite, and so lim H 1 (K∞ |Kn , OK
1
∞ ,S
) = T0 (Xcs ) .
←− ←−
n n
Now (iii) follows from (5.4.9)(iii) and (11.3.11)(i) since XS++ = E 0 (XS )+
and HomΛ[∆] ( Ind∆p Λ− , Λ[∆]) ∼
∆ ∆
= Ind∆p Λ− . 2

We easily obtain from the preceding results the following corollaries for the
Iwasawa invariants of X, Xnr and Xcs .
(11.3.16) Corollary. Let k∞ |k be a ZZp -extension such that every prime above p
is finitely decomposed in k∞ . Then
µnr = µcs ,
λcs ≤ λnr ≤ λcs + s(k∞ |k) ,
where s(k∞ |k) is the number of primes in k∞ ramifying in k∞ |k. Furthermore,
there is a pseudo-isomorphism
G(H|H 0 ) ≈
M
Λ/ξni
i
deg(ξni ) ≤ s(k∞ |k).
P
with some cyclotomic polynomials ξni satisfying i
p−1
ωn n−1
(1 + T )k p
X
Proof: Recall that ξn = = is the (irreducible) pn -th
ωn−1 k=0
cyclotomic polynomial in the variable 1 + T . By (11.3.10)(ii) with S = Σ =
Sp ∪ S∞ , we get the exact sequence
M
Λ/ωnp −→ Xnr −→ Xcs −→ 0,
p∈S r (k)
X
where pnp = [Γ : Γnp ] and deg(ωnp ) = #S r (k∞ ) = s(k∞ |k), which
p∈S r (k)
implies the result. 2
(11.3.17) Corollary. Assume that µp ⊆ k and let k∞ |k be the cyclotomic

ZZp -extension. Then
µ = µcs and λ = λcs + #Sp (k∞ ) − 1 .
Proof: This follows from (11.3.14) since

µ(T1 (EΣ )) = 0 , µ(T1 (AΣ )) = 0 ,
λ(T1 (EΣ )) = 1 , λ(T1 (AΣ )) = #Sp (k∞ ) ,
by (11.3.11)(ii) and (11.2.3), and the λ- and µ-invariants of E 1 (Xcs (−1)) and
Xcs coincide by (5.5.13). 2
We finish this section by listing all results of this section for the cyclotomic
ZZp -extension. The set Sp ∪ S∞ is again denoted by Σ.

Let K = k(µp ), K∞ = k(µp∞ ) the cyclotomic ZZp -extension of K and ∆ =

G(K∞ |k∞ ) ∼
= G(K|k). We denote the p-part of the cyclotomic character
by
κ : G(K∞ |k) −→ Aut(µp∞ ) ∼= ZZ× p ,
i.e. ζ σ = ζ κ(σ) for all ζ ∈ µp∞ and σ ∈ G(K∞ |k). For j ∈ ZZ and p|p we define
( (
1 , κj |∆p = 1 , 1 , κj |∆ = 1 ,
δj,p = δj =
0 , otherwise , 0 , otherwise .
The Λ[∆]-module X (and analogously Xnr , Xcs ) has a decomposition coming
from the action of ∆: M
X= ei X ,
i mod d
where d = #∆ and {ei = 1P

κ−i (σ)σ
}i mod d are the idempotents of ZZp [∆];
d σ ∈∆
thus ei X is the i-th eigenspace of the ∆-module X. We set
µ(i) = µ(ei X) and λ(i) = λ(ei X) ,
and denote the invariants of ei Xnr and ei Xcs analogously.
(11.3.18) Theorem. Let K|k be a finite Galois extension of degree prime to p

with Galois group ∆ and assume that k is totally imaginary if p = 2. Let K∞
be the cyclotomic ZZp -extension of K. Then
. the weak Leopoldt conjecture holds for K |K, ∞
. E ∼= Λ[∆]
Σ ⊕ (r2 +r1 −r10
M
Ind Λ) ⊕ ZZ (1),
∆p
∆ p
0
p∈S∞
.E ∼ 0
= Λ[∆]r2 +r1 −r1 ⊕ (
M ∆
Ind∆p Λ) ⊕ ZZp (1),
0
p∈S∞
.A Σ
∼
= Λ[∆]r1 +2r2 ⊕
M
ZZp (1),
p∈Sp (k∞ )
.U Σ
∼
= Λ[∆]r1 +2r2 ⊕
M
ZZp (1).
p∈Sp (k∞ )
Assume in addition that K = k(µp ). Then
.X Σcontains no finite nontrivial Λ-submodules, and there is an exact
sequence of Λ[∆]-modules
∆
0 −→ XΣ /T1 (XΣ ) −→ Λ[∆]r2 ⊕ Ind∆p Λ− −→ T0 (Xcs (−1))∨ −→ 0,
M
0
p∈S∞
.µ (i)
= µ(1−i)
nr = µ(1−i)
cs for all i ∈ ZZ; in particular, µ = µnr = µcs ,
.λ (i)
= λ(1−i)
cs +
X
δ1−i,p − δ1−i for all i ∈ ZZ; in particular,
p∈Sp (K∞ )
λ = λcs + #Sp (K∞ ) − 1.

§4. Iwasawa Theory for Totally Real Fields and CM-Fields 751
0
Recall that S∞ is the set of all real primes of k which become complex in
K, r1 = #S∞ and Λ[∆p ] = Λ ⊕ Λ− for p ∈ S∞
0 0 0
, where ∆p acts on Λ− by −1
and trivially on Λ. The assertions concerning the µ- and λ-invariants follow
from (11.3.16) and (11.3.13), noting that ei E 1 (M ) = E 1 (e−i M ) if M denotes
a Λ[∆]-module.
§4. Iwasawa Theory for Totally Real Fields and

CM-Fields
If k is a number field of CM-type, i.e. k is a totally imaginary quadratic

extension of its maximal totally real subfield k + , then we obtain more informa-
tion on the Iwasawa modules considered in the previous sections by using the
involution ρ which generates the Galois group
G(k|k + ) ∼
= ZZ/2ZZ.
We assume in this section that p is odd. Thus, if k contains the group µp of
p-th roots of unity, then k = k + (µp ). If A is a G(k|k + )-module, then we denote
the (+) resp. (−)-eigenspace of A with respect to the action of G(k|k + ) by
A± = (1 ± ρ)A.
Obviously, A = A+ ⊕ A− , if multiplication by 2 is an isomorphism on A.
In this section we first consider the following situation:
p is an odd prime number,

k0 is a totally real number field,
k is the CM-field k0 (µp ) with Galois group ∆ = G(k|k0 ),
k∞ |k is the cyclotomic ZZp -extension, Γ = G(k∞ |k) and Λ = ZZp [[Γ ]],
kΣ is the maximal extension of k which is unramified outside Σ, where
Σ = Sp ∪ S∞ ,
L is the maximal unramified p-extension of k∞ ,
L0 is the maximal unramified p-extension of k∞ which is completely
decomposed everywhere.
We consider the Kummer pairing

×
h , i : Gab Zp ) −→ µp∞
k∞ (p) × (k∞ ⊗ Qp /Z
which is defined by
√
n √
n
hσ, xi = σ( p α)/ p α

−n ×
if σ ∈ Gab
k∞ (p) and x = α ⊗ p mod ZZp with α ∈ k∞ (the definition obviously
n
does not depend on the chosen p -th root of α). It is a non-degenerate, (Γ ×∆)-
invariant pairing of the compact abelian group Gab ab
k∞(p) = G(k(p) |k∞ ) with the
×
discrete abelian group k∞ ⊗ Qp /ZZp . If H is a closed subgroup of G = Gab k∞ (p),
⊥ ×
then H denotes the annihilator of H in k∞ ⊗ Qp /ZZp . From the theory of
Pontryagin duality we get non-degenerate pairings
×
H × (k∞ ⊗ Qp /ZZp )/H ⊥ −→ µp∞ ,
G/H × H⊥ −→ µp∞ .
(11.4.1) Proposition. With the notation as above, consider the following

×
submodules of k∞ ⊗ Qp /ZZp
n
M = {α ⊗ p−n | n ≥ 0, α ∈ U∞,p · (k∞,p
×
)p if p ∈/ Σ},
n n
L0 = {α ⊗ p−n | n ≥ 0, α ∈ U∞,p · (k∞,p
× ×
)p if p ∈/ Σ, α ∈ (k∞,p )p if p ∈ Σ},
where U∞,p is the inductive limit over the group of local units Ukn,p of kn,p .
Then we have the equalities
G(k(p)ab |k(p)ab ∩ kΣ )⊥ = M and G(k(p)ab |k(p)ab ∩ L0 )⊥ = L0 .
Thus there are non-degenerate pairings
X× M −→ µp∞ ,
Xcs × L0 −→ µp∞
involving the Iwasawa modules X = G(kΣ |k∞ )ab (p) and Xcs = G(L0 |k∞ )ab .
Proof: This follows directly from the definition of kΣ and L0 . 2
(11.4.2) Proposition. With the notation as above, there are canonical (Γ × ∆)-
invariant exact sequences
ϕ
(i) 0 −→ Ok×∞ ⊗ Qp /ZZp −→ M −→
1
Cl(k∞ )(p) −→ 0 ,
ϕ
(ii) 0 −→ Ok×∞ ,Σ ⊗ Qp /ZZp −→ M −→
2
ClΣ (k∞ )(p) −→ 0 .
Proof: The second sequence follows from the exact sequence (8.3.4)
pm
0 −→ µpm −→ OΣ× −→ OΣ× −→ 0
by applying cohomology:
0 −→ Ok×∞ ,Σ /pm −→ H 1 (kΣ |k∞ , µpm ) −→ pm H 1 (kΣ |k∞ , OΣ× ) −→ 0 ,
passing to the direct limit and noting that
H 1 (kΣ |k∞ , OΣ× )(p) = ClΣ (k∞ )(p) ,

H 1 (kΣ |k∞ , µp∞ ) ∼ = G(kΣ |k∞ )ab (p)(−1)∨ = M .

It is easy to see that the map ϕ2 is given explicitly by
ϕ2 : M −→ ClΣ (k∞ )(p), α ⊗ p−n 7−→ [a1 ] ∈ ClΣ (km )(p),
n
where α ∈ km ×
and αOkm = ap1 a2 with an ideal a2 having only prime divisors in
Sp (km ). Since all primes dividing p have an infinite ramification index in k∞ |k,
n
we see that, for sufficiently large s, αOkm+s = ap for some ideal a in km+s . We
define ϕ1 by sending α ⊗ p−n to the class [ a ] ∈ Cl(k∞ )(p). The map ϕ2 is the
composition of ϕ1 with the natural projection Cl(k∞ )(p) to ClΣ (k∞ )(p).
Restricting ϕ1 to Ok×∞ ,Σ ⊗ Qp /ZZp and denoting this map by ϕ̃1 , we get an
exact sequence
Ok×∞ ⊗ Qp /ZZp
ϕ̃
Ok×∞ ,Σ ⊗ Qp /ZZp 1 Cl(k∞ )(p) ClΣ (k∞ )(p) .
Now the commutative exact diagram
Ok×∞ ,Σ ⊗ Qp /ZZp M M/Ok×∞ ,Σ ⊗ Qp /ZZp
ϕ1 ϕ̄2
im(ϕ̃1 ) Cl(k∞ )(p) ClΣ (k∞ )(p)

×
shows that ϕ1 is surjective with kernel Ok∞ ⊗ Qp /ZZp . 2
Let
κ : G(k∞ |k0 ) −→ Aut(µp∞ ) = ZZ×
p
be the p-part of the cyclotomic character, d = #∆, and let
1 X −i
ei = κ (σ)σ ∈ ZZp [∆] , i ∈ ZZ/dZZ ,
d σ∈∆
be the corresponding idempotents.
(11.4.3) Theorem. The Kummer pairing induces a non-degenerate pairing

ei X × e1−i Cl(k∞ )(p) −→ µp∞
for all even i, hence
ei X ∼
= Hom(e1−i Cl(k∞ ), µp∞ ).
Proof: The subgroup hOk×n+ , µ(kn )i of Ok×n has index 1 or 2 (see [246], th.
4.12), and hence, since p is odd,
ei (Ok×∞ ⊗ Qp /ZZp ) = 0 for i odd.
For i even it follows from (11.4.2)(i) that
e1−i M ∼
= e1−i Cl(k∞ )(p)
and from (11.4.1) that
Hom(e1−i M, µp∞ ) ∼ = ei X. 2

As in §3 we set
µ(i)
• = µ(ei X• ), λ(i)
• = λ(ei X• )
for i ∈ ZZ, and define

X
µ± ± ±
• := µ• (k) := µ(X• ) = µ(i)
• ,
i even
odd
X
λ± ± ±
• := λ• (k) := λ(X• ) = λ(i)
• ,
i even
odd
where the index • is nothing, “nr”, “cs” or a set S ⊇ Σ. Furthermore, recall

that M ◦ denotes the Λ-module M with the inverse action of Γ , cf. (5.5.12).
(11.4.4) Corollary. There is an isomorphism and a pseudo-isomorphism of

Λ-modules for each even i
(ei X)(−1) ∼
= E 1 (e1−i G(H|Hn k∞ )) ≈ (e1−i Xnr )◦ .
0
In particular,
µ(i) = µnr
(1−i)
, λ(i) = λ(1−i)
nr ,
and
µ± = µ∓
nr , λ± = λ∓
nr .
Proof: This follows from the last theorem and (11.1.8). 2
Regarding the Λ-ranks of the different eigenspaces of EΣ = lim (Ok×n ,Σ ⊗ZZp ),

←− n
E = lim (Ok×n ⊗ ZZp ) and X and the λ-invariants of X ± and Xcs
±
, we obtain the
←− n
(11.4.5) Proposition. For each i ∈ ZZ we have

[k0 : Q] , i odd,
(
(i) rankΛ ei X =
0, i even,
e Λ[k0 :Q] , i even,


 i


(ii) ei EΣ ∼
=  ZZp (1) , i ≡ 1 mod #∆,


0, otherwise,
and the same for ei E,
(iii) λ+ = λ− +
cs + #Sp (k∞ ) − #Sp (k∞ ),
λ− = λ+cs + #Sp (k∞

+
) − 1.

0
Proof: All assertions follow from (11.3.18) (observe that S∞ = S∞ , so that
0
r2 = 0 and r1 = r1 = [k0 : Q]). 2
Now we want to consider the “difference” between Xnr and Xcs . Recall the
notion of the Weierstraß polynomials ωn (5.3.13), and the numbers δi,p defined
by
(ZZ/pZZ(i))∆p ∼
= (ZZ/pZZ)δi,p ,
where ∆p is the decomposition group of ∆ = G(k|k0 ) with respect to the
prime p.
(11.4.6) Proposition. For odd i ∈ ZZ, there is an exact sequence of Λ[∆]-

modules
Λ/ωnδi,p
M
0 −→ p
−→ ei Xnr −→ ei Xcs −→ 0;
p∈Sp (k)
consequently,
X
λ(i) (i)
nr = λcs + δi,p , µ(i) (i)
nr = µcs ,
p∈Sp (k∞ )
and
λ− − +
nr − λcs = #Sp (k∞ ) − #Sp (k∞ ).
Proof: The exact sequence follows from (11.3.10)(ii) and (11.4.5)(ii). The as-
sertions concerning the Iwasawa invariants may also be obtained from (11.4.4)
and (11.3.18). 2
It is more difficult to obtain results for the even eigenspaces. According to

+ +
the conjecture of Greenberg, Xnr and Xcs should be finite (nevertheless they
can be non-zero; see [62] for examples).
(11.4.7) Proposition. Assume that the (strong) Leopoldt conjecture holds for
some layer kn+ of k∞
+
|k + with n ≥ λ+ . Then for i even
λ(i) (i)
nr = λcs
and, in particular,
λ+nr = λ+cs .
Proof: From (11.3.3) we know that under our assumption, the Leopoldt
+
conjecture holds for all finite layers of the tower k∞ |k + , and so
(ei X)Γn = 0 for i even and all n.

Consequently, the characteristic polynomial of the eigenspace ei X, i even, is

prime to ωn for all n and this also holds for its quotient ei Xnr , since ei X is a
Λ-torsion module by fact (4) of the beginning of §3. Now the exact sequence
(11.3.10)(ii) shows that the kernel of ei Xnr ei Xcs is finite. 2
In the following, we do not assume that µp is contained in our CM-field. So

let k|k + be a CM-field with Galois group ∆ = G(k|k + ) ∼
= ZZ/2ZZ and let k∞ be
the cyclotomic ZZp -extension of k, for p > 2.
(11.4.8) Theorem. We obtain for the projective dimensions of the Λ-modules

− −
Xnr and Xcs that
− −
pdΛ Xnr ≤ 1 and pdΛ Xcs ≤ 1.
− −
Consequently, Xnr and Xcs do not contain a finite nontrivial Λ-submodule.
Proof: First let us assume that µp ⊆ k. Then, by (11.3.15)(iii), there is an

exact sequence
Λ− −→ T0 (Xcs (−1))∨ −→ 0 .
M
0 −→ X/T1 (X) −→
p∈S∞ (k+ )
Thus (T0 (Xcs (−1))∨ )+ = 0, and so T0 (Xcs

− −
) = 0, which gives the result for Xcs .
If µp ⊆/ k, then let K = k(µp ), which is again a CM-field with maximal
totally real subfield K + such that K + ∩ k = k + . It follows from the above that
X(K)− −
cs contains no finite nontrivial Λ-submodule. Since X(k)cs is a direct
summand of X(K)− cs , the same holds for this module, and the result follows.
−
The assertion for Xnr is a consequence of the next lemma, which holds for
arbitrary CM-fields k|k + (without the assumption that µp is contained in k).
2
(11.4.9) Lemma. Let k|k + be a CM-field. Then there is a canonical exact

sequence of Λ[G(k|k + )]-modules
Λ/ωnp )− −→ Xnr
− −
M
0 −→( −→ Xcs −→ 0 .
Sp (k)
Proof: This follows from the exact sequence (11.3.10)(ii) and the isomor-
∼ E− ( ∼ Z
phism E − → δ
Σ = Zp (1) , where δ = 1 if µp
⊆ k, and zero otherwise).

We now come to the question of how the λ-invariants behave under a change
of the ground field. Although nothing is known if the base change is given
by an extension K|k where [K : k] is not a power of p, there is a remarkable
analogy to the Riemann-Hurwitz formula in algebraic geometry if K|k is a
Galois p-extension of CM-fields. This was first observed by Y. KIDA [103]
and later other authors proved this by different methods. We will proceed as
in [90].
Let k|k + be a CM-field, K + |k + be a finite Galois p-extension and K = K + k.
We denote the Galois group of the extension K∞ |k∞ by G,
∼ G(K + |k + ) ,
G = G(K∞ |k∞ ) −→ ∞ ∞
where k∞ is the cyclotomic ZZp -extension and K∞ = Kk∞ . Since X(K + ) is a

Λ-torsion module by (11.3.2)(iv) and (ii), the Qp [G]-module X(K + ) ⊗ Qp is
finitely generated. Regarding its structure, we obtain the
(11.4.10) Theorem. Suppose that the Iwasawa invariant µ+ = µ(X(k + )) is

zero. Then there is an isomorphism of Qp [G]-modules
X(K + ) ⊗ Qp ∼
+ G
= Qp [G]λ (k)−1 ⊕ Qp ⊕
M
IndGP IGP
+
p prime of k∞
p-p
where IGP is the augmentation ideal in Qp [GP ] and P denotes an arbitrary

+ +
extension of the prime p of k∞ to K∞ .
Proof: First observe that the sum on the right is finite because GP =/ 1, and
+ +
so IGP =/ 0 only for primes P which are ramified in K∞ |k∞ . Since the weak
Leopoldt conjecture holds for the cyclotomic ZZp -extension and since µ+ = 0,
it follows from (11.3.7) and (11.3.6)(i) that G(kS+ |k∞
+
)(p) is a finitely generated
free pro-p-group for every finite S ⊇ Σ. Let S be finite and large enough so
that K + ⊆ kS+ (p). Then, by (5.6.6) applied to the presentation
1 −→ G(kS+ (p)|K∞
+
) −→ G(kS+ (p)|k∞
+
) −→ G −→ 1 ,
we obtain (using (11.3.6)(ii))
XS (K + ) ⊗ Qp ∼
+
= Qp [G]λS (k)−1 ⊕ Qp
∼ + X
= Qp [G]λ (k)−1 ⊕ Qp ⊕ Qp [G]d , d= δp
+ )
p∈S\Σ(k∞
∼ + G
M
= Qp [G]λ (k)−1 ⊕ Qp ⊕ IndGP Qp [GP ]δp .
+ )
p∈S\Σ(k∞

+
From the G-invariant sequence (11.3.5) (for the ZZp -extension K∞ |K + ), we see
XS (K + ) ⊗ Qp ∼
G
IndGP Qδpp .
M
= X(K + ) ⊗ Qp ⊕
+ )
p∈S\Σ(k∞
Putting both formulae together and recalling (5.6.10)(i), we obtain the result.
2
(11.4.11) Corollary (Riemann-Hurwitz Formula). Let K + |k + be a finite Galois

p-extension of totally real fields and assume that µ(k + ) = 0 for the cyclotomic
ZZp -extension of k + , p > 2. Then
X
λ+ (K) − 1 = [K∞
+ +
: k∞ ](λ+ (k) − 1) + (eP − 1) ,
P-p
+ +
where eP denotes the ramification index of K∞ |k∞ with respect to a prime P
+
of K∞ .
Proof: Take the Qp -dimensions of both sides of the isomorphism (11.4.10)

and recall that dimQp IGP = #GP − 1 = eP − 1. The formula follows. 2
− −
Using duality, we obtain similar results for the modules Xnr and Xcs in the
case of CM-fields.
(11.4.12) Theorem. Let p > 2 and let K|k be a finite Galois p-extension of
CM-fields. Assume that µ(k) = 0 for the cyclotomic ZZp -extension k∞ |k and
let G = G(K∞ |k∞ ). Then there are isomorphisms of Qp [G]-modules
−
Xnr (K)− ⊗ Qp ∼
G
= Qp [G]λnr (k)−δ ⊕ Qδp ⊕
M
IndGP IGP ,
p prime of k∞ +
p-p, p splits in k∞
−
Xcs (K)− ⊗ Qp ∼
G
= Qp [G]λcs (k)−δ ⊕ Qδp ⊕
M
IndGP IGP ,
p∈S(k∞ + )
p splits in k∞
+ +
where δ = 1 if µp ⊆ k and zero otherwise, and S(k∞ ) is the union of Sp (k∞ )
+ +
and all ramified primes of the extension K∞ |k∞ .
Proof: If µp ⊆ k, then
Xnr (K)− ⊗ Qp ∼
= X(K + ) ⊗ Qp
as Qp [G]-modules by (11.4.4) and λ+ = λ− nr . If IGP =
/ 0 for a prime p - p of
+ + +
k∞ , then p ramifies in K∞ . Hence µp ⊆ k∞,p and p splits in k∞ . Therefore, if
µp ⊆ k, the first formula follows from (11.4.10).

Now suppose µp ⊆ / k. Using the exact sequence (11.4.2)(i) for the field
0 0
K = K(µp ) and taking G(K∞ |K∞ )-invariants, we obtain the exact sequence
×
0 −→ OK ∞
⊗ Qp /ZZp −→ H 1 (KΣ |K∞ , µp∞ ) −→ Cl(K∞ )(p) −→ 0 ,
so that
H 1 (KΣ |K∞ , µp∞ )− ∼
= Cl(K∞ )(p)− .
Again let S ⊇ Σ be finite and sufficiently large that K ⊆ kS (p). The Hochschild-
Serre spectral sequence (2.4.1)
E2ij = H i (G, H j (kS |K∞ , µp∞ )) ⇒ H i+j (kS |k∞ , µp∞ )
/ k∞ and G(kS |k∞ ) is free) that
implies (since µp ⊆
H 0 (G, H 1 (kS |K∞ , µp∞ )) = H 1 (kS |k∞ , µp∞ ) ,

(∗)
H i (G, H 1 (kS |K∞ , µp∞ )) = 0 for i ≥ 1.
Thus H 1 (kS |K∞ , µp∞ )∨ is a cohomologically trivial ZZp [G]-module and the
same is true for the minus-part. Since
(H 1 (KΣ |K∞ , µp∞ )− )∨ ∼

= Hom(Cl(K∞ ), Qp /ZZp )−
∼
= E 1 (G(H|Hn0 K∞ )− )
≈ (Xnr (K∞ )− )◦
by (11.1.8) and µnr (K∞ |K) = µ(K∞ |K) = 0 by (11.3.8), we see that
(H 1 (KΣ |K∞ , µp∞ )− )∨ = G(KΣ (p)|K∞ )ab + (−1) is ZZp -free and finitely gen-
erated as a ZZp [G]-module. Using (11.3.5), it follows that the same holds for
(H 1 (kS |K∞ , µp∞ )− )∨ . Thus by (5.2.21) and (∗),
(H 1 (kS |K∞ , µp∞ )− )∨ ∼
= ZZp [G]r
is a free ZZp [G]-module.
By (11.3.6)(ii) and (11.4.4) its rank is equal to λ−
nr (k) + d, where d is the sum
+
of the numbers δp for the primes p ∈ S r Σ(k∞ ) which split in k∞ . It follows
that
−
Xnr (K)− ⊗ Qp ∼ = Qp [G]λnr (k) ⊕
M
IG . P
p prime of k∞ +
The assertion for Xcs (K)− is now an easy consequence of the above, using the
isomorphism
Xnr (K)− ⊗ Qp ∼ = Xcs (K)− ⊗ Qp ⊕
M M
Qp
p∈Sp (k∞ + ) P|p
p splits in k∞

and the equality λ− −

nr (k) = λcs (k) + #{p ∈ +
Sp (k∞ ) | p splits in k∞ }, which
Considering Qp -dimensions on both sides of the isomorphisms (11.4.12),

we obtain the
(11.4.13) Corollary. Let p > 2 and let K|k be a finite Galois p-extension of
CM-fields. Assume that µ(k) = 0 for the cyclotomic ZZp -extension k∞ |k. Then
λ− −
X X
nr (K) − δ = [K∞ : k∞ ](λnr (k) − δ) + (eP − 1)
p prime of +
k∞ P|p
and
λ− −
X X
cs (K) − δ = [K∞ : k∞ ](λcs (k) − δ) + (nP − 1) ,
p∈S(k∞ + ) P|p
p splits in k∞
where δ = 1 if µp ⊆ k and zero otherwise, eP denotes the ramification index
+ + +
and nP the local degree of K∞ |k∞ with respect to a prime P of K∞ . The set
+ + + +
S(k∞ ) is the union of Sp (k∞ ) and the set of all primes ramifying in K∞ |k∞ .
(11.4.14) Proposition. Let p =/ 2 and let k = k + (µp ) be a CM-field such

+
/ k∞,p
that µp ⊆ for every prime p above p. Then there exists a canonical
isomorphism
− ∼ E 1 (X + )
Xcs (−1) −→
for the cyclotomic ZZp -extension.
Proof: This follows directly from (11.3.13). 2
We conclude this section by showing that an isomorphism as above can

also be obtained on the level of finite coefficients, provided that µ = 0. The
isomorphism of (11.4.14) can be recovered from this result by passage to the
limit over n, but we will neither prove nor use this fact. We introduce the
following notation. In the situation where
p is an odd prime number,
k = k + (µp ) is a CM-field with Galois group ∆ = G(k|k + ) ∼
= ZZ/2ZZ ,
let
kΣCM = kΣ+ (p)(µp )
be the maximal CM-field of k inside kΣ (p). We denote the maximal p-
extension of kΣCM which is completely decomposed everywhere by L̃0 .

/ kp+ for all primes

(11.4.15) Proposition. In the above situation assume that µp ⊆
above p and that the Iwasawa µ-invariant of the cyclotomic ZZp -extension k∞ |k
is zero. Then the cup-product induces a perfect pairing of finite groups
CM
H 0 (kΣ |k∞ , H 1 (L̃0 |kΣ
CM
)(1))− × H 1 (kΣ
CM
|k∞ ) ∪ CM
H 1 (kΣ |k∞ , H 1 (L̃0 |kΣ
CM
)(1))∆
−
(Xcs /pm )(−1)∗ (X + /pm )∗ ZZ/pm ZZ,
where the coefficients of the cohomology groups are ZZ/pm ZZ.
Proof: We may assume that λ+ = λ− cs is non-zero since otherwise all groups

under consideration are trivial. From the exact sequence
0 −→ OΣ× −→ IΣ −→ CΣ −→ 0
we obtain the exact cohomology sequence
CM ) CM ) δ
Hom(ZZ/pm , IΣ )G(kΣ |kΣ → Hom(ZZ/pm , CΣ )G(kΣ |kΣ → H 1 (kΣ |kΣCM , µpm )
H 1 ((kΣCM )p , µpm ) ,
Y
→
CM )
p∈Σ(kΣ
recalling that OΣ× is p-divisible. We get an isomorphism

CM δ
(Hom(ZZ/pm ZZ, CΣ )G(kΣ |kΣ ) )∆ −→ 1 0 CM
Z/pm ZZ)− (1)
∼ H (L̃ |kΣ , Z
/ kp× for all p|p. Furthermore, the following diagram is commutative

since µp ⊆
(assuming n large enough):
¡ ZZ/pm ZZ, CΣ )) × H 2 (kΣ |kn )

H 0 (kΣ |kn , Hom( ∪
H 2 (kΣ |kn , pm CΣ )
∼ ZZ/pm ZZ
=
Γn ∪
H 0 (kΣ |k∞ , pm C Σ ) ∆ × H 1 (kΣ |k∞ )∆
Γn H 1 (kΣ |k∞ , pm CΣ )Γn ×∆
inf
CM CM CM ∪ CM CM
H 0 (kΣ |k∞ , pm CΣ (kΣ ))∆ × H 1 (kΣ |k∞ ) H 1 (kΣ |k∞ , pm CΣ (kΣ ))∆
H 0 (δ) H 1 (δ)
∪
CM
H 0 (kΣ CM −
|k∞ , H 1 (L̃0 |kΣ CM
) (1)) × H 1 (kΣ |k∞ ) CM
H 1 (kΣ |k∞ , H 1 (L̃0 |kΣ
CM −
) (1)),
where the missing coefficients of the cohomology groups are ZZ/pm ZZ (for the
compatibility of the upper two pairings observe that H 1 (−)Γn = H 1 (Γ, H 1 (−))
and that the Hochschild-Serre spectral sequence is functorial with respect to the
cup-product). The isomorphism H 2 (kΣ |kn , pm CΣ ) ∼ = ZZ/pm ZZ follows from
(10.11.5) and (8.3.9)(ii),(iii). The map inf on the right is an isomorphism as
one sees as follows:

If r ≥ n, then the diagram

H 1 (kΣ |k∞¢£¤¥¦§¨ , pm CΣ )Γn H 2 (kΣ |kn , p m CΣ ) ZZ/pm ZZ
cor
H 1 (kΣ |k∞ , pm CΣ )Γr H 2 (kΣ |kr , p m CΣ ) ZZ/pm ZZ

commutes, showing that H 1 (kΣ |k∞ , pm CΣ ) is fixed by Γn for some n. Thus
inf is injective. Now let K|k∞ be a finite Galois extension inside kΣ (p). Then
H 1 (kΣ |K, pm CΣ ) = lim H 1 (kΣ |K, pm CΣ )Γn ∼
= µpm (K)(−1)∨ ∼
= ZZ/pm ZZ
←−
n
and therefore
H 1 (kΣ |kΣCM , pm CΣ ) = lim
−→
H 1 (kΣ |K, pm CΣ ) ∼
= ( lim
←−
µpm (K)(−1))∨ = 0,
CM
K ⊆kΣ CM
K ⊆kΣ
since the degree of [kΣCM : k∞ ] is divisible by p∞ and the projective limit is

taken via p-multiplication. This shows that inf is surjective.
Since the upper pairing in the diagram above is non-degenerate on the right
by the global duality theorem (8.4.4), the same holds for the lower pairing. In
order to prove that the pairing is perfect, we will show that the orders of both
groups coincide. Obviously,
+
#H 1 (kΣCM |k∞ , ZZ/pm ZZ) = #(X + /pm ) = pm·λ
and
− +
−
#(Xcs /pm ) = pmλcs = pmλ ,
by (11.4.8) and the assumption that µ = 0. It remains to show that there is an
isomorphism
)− (1) ∼
CM |k )
(H 1 (L̃0 |kΣCM , ZZ/pm ZZ)G(kΣ ∞
= H 1 (L0 |k∞ , ZZ/pm ZZ)− (1)
∼
= (X − /pm )(−1)∨ .
cs
But this follows from the commutative exact diagram with coefficients ZZ/pm ZZ
CM©ª«¬®¯ G(kΣ
CM CM
Y CM
CM Gp (kΣ |k∞ )
H 1 (L̃0 |kΣ ) |k∞ ) CM G(kΣ
H 1 (kΣ |kΣ ) |k∞ )
H 1 (kΣ,p )
p∈Σ(k∞ )
Y
H 1 (L0 |k∞ ) H 1 (kΣ |k∞ ) H 1 (k∞,p )
p∈Σ(k∞ )
+
/ k∞,p
if one takes the minus-parts, and recalling again that µp ⊆ for p|p. 2

§5. Positively Ramified Extensions 763
Exercise: (Leopoldt’s Spiegelungssatz)

(a) Let k be a number field, p an odd prime number and K|k a finite abelian extension
containing µp . Assume that the Galois group ∆ = G(K|k) has exponent p − 1. For each
character χ : ∆ → ZZ× p and let M
(χ)
= {x ∈ M | gx = χ(g)x for all g ∈ ∆} be the χ-
eigenspace of a ZZp [∆]-module M . Prove the following inequality
−1
dimIFp (ClSp (K)/p(−1))(χ ) ≤
dimIFp (ClSp (K)/p)(χ) + dimIFp (ESp (K)/p)(χ) .
Hint: Consider the exact sequence

0 −→ ESp (K)/p −→ H 1 (G(KSp |K), µp ) −→ p ClSp (K) −→ 0
and the inclusion Hom(ClSp (K), µp ) ,→ H 1 (G(KSp |K), µp ).
(b) Let p be an odd prime number, K = Q(ζp ) and ω : G(K|Q) = ∼ (ZZ/pZZ)× → ZZ× the
p
Teichmüller character, i.e. the character with ω(ã) ≡ a mod p, where ã the automorphism
ã(ζp ) = ζpa . Let i be an even and j an odd integer such that i + j ≡ 1 mod p − 1. Prove the
following inequalities:
i j i
dimIFp (ClSp (K)/p)(ω ) ≤
dimIFp (ClSp (K)/p)(ω ) ≤
dimIFp (ClSp (K)/p)(ω ) + 1 .
Hint: Use (a) and

(ω r ) 1, if r is even,
dimIFp (ESp (K)/p) = for 1 < r < p − 1.
0, if r is odd,
§5. Positively Ramified Extensions
In 1969, in his foreword to H. KOCH’s book [110], I. R. SAFAREVIČ recalled

that already HILBERT, when laying the foundations of class field theory, used
the analogy between algebraic number fields and function fields as his starting
point. He concluded:
“From this point of view, a noncommutative generalization of

class field theory must correspond to the investigation of the fun-
damental group of a Riemann surface, which, as is well-known,
is noncommutative.”
The present book confirms this analogy with function fields to a large extent.
As explained in X §1, the analogy to Riemann surfaces comes about after a
base change to the algebraic closure of the finite ground field.
The natural analogue to this base change for number fields is the passage
to the cyclotomic, say, ZZp -extension for a prime number p. But G(kS (p)|k∞ )
is a free pro-p-group (provided that µ = 0), and thus is analogous to the
fundamental group of an open, i.e. noncompact, Riemann surface. This is not
surprising since GS is the fundamental group of the affine curve Spec(OS ).

It is a remarkable observation that we can overcome this problem by intro-

ducing further restrictions on the wild ramification (which doesn’t occur in
complex geometry). We will see in this section that dualities between certain
Iwasawa modules which were shown in the last section, can be viewed as a
perfect cup-product pairing of a (non-abelian) Demuškin group. This group,
the Galois group of the maximal positively ramified extension, is analogous to
the fundamental group of a compact Riemann surface. However, despite the
clear analogy, we are far from a deeper understanding of this phenomenon.
Let p be an odd prime number and let k|k + be a CM-field containing the
group µp of the p-th roots of unity. Thus k = k + (µp ). Let k∞ be the cyclotomic
ZZp -extension of k, Γ = G(k∞ |k) and Λ = ZZp [[Γ ]] be the Iwasawa algebra. We
assume that
(1) the Iwasawa µ-invariant of k∞ |k is zero,
(2) / kp+
µp ⊆ for all primes p of k above p.
We define a natural Galois p-extension k̃Σ inside kS (p) as the maximal

p-extension of k unramified outside p and positively ramified at p.
(11.5.1) Definition. Let k = k + (µp ) be a CM-field satisfying condition (2). A

finite Galois p-extension K|k is called positively ramified at p if
Kp ⊆ kp+ (p)(µp )
for all primes p dividing p.
Remarks: 1. The composite of positively ramified p-extensions is again

positively ramified. Hence the maximal positively ramified p-extension k̃ of
k exists. We set
k̃Σ = k̃ ∩ kΣ (p) ,
where Σ = Sp ∪ S∞ . Obviously, k∞ ⊆ k̃Σ and
kΣCM = kΣ+ (p)(µp ) ⊆ k̃Σ ,
where kΣCM is the maximal CM-field of k inside kΣ .
2. We also can give a description of k̃ by its Galois group. Setting
∆ = G(k|k + ) ∼= ZZ/2ZZ, we have ∆ = ∆p for all p|p by assumption (2). Let
Tp = T (kp (p)|kp ) be the inertia subgroup of the local Galois group G(kp (p)|kp )
for a nonarchimedean prime p of k. The sequence
1 −→ G(kp (p)|kp+ (p)(µp )) −→ G(kp (p)|kp ) −→ G(kp+ (p)|kp+ ) −→ 1

is exact and we set

Tp− := G(kp (p)|kp+ (p)(µp )).
By definition of k̃, we obtain
G(k̃|k) = G(k(p)|k)/(Tp− , p|p)
and
G := G(k̃Σ |k∞ ) = G(k(p)|k∞ )/(Tp− , p|p ; Tp , p - p) ,
(recall that for closed subgroups Hi , i ∈ I, of a profinite group H, the subgroup
(Hi , i ∈ I) denotes the normal closure of the subgroup which is generated by
the groups Hi , i ∈ I). Observe that by (9.4.3) the groups G(kp (p)|kp ) can be
considered as subgroups of G(k(p)|k). If
H := G(k̃Σ |kΣCM ) ,
then
G/H = G(kΣCM |k∞ ) ∼
= G(kΣ+ (p)|k∞
+
).
3. We have the following diagram of fields
ÀÁ¿½¾¼»º´µ¶·¸¹±²³° k(p)
k̃
∆
k + (p) k CM kΣ (p)
k̃Σ
H
∆
kΣ+ (p) kΣCM G
L0
+ ∆
k∞ k∞
Γ Γ
∆
k+ k = k + (µp ) .
(11.5.2) Proposition. Under the above assumptions (1) and (2), the following
are true:
(i) There is a canonical isomorphism of Λ[∆]-modules
∼ X+ ⊕ X− ,
G(k̃Σ |k∞ )ab −→ cs
where X = G(kΣ (p)|k∞ )ab and Xcs = G(L0 |k∞ )ab .

(ii) If λ+ =/ 0, then
H 2 (G(k̃Σ |k∞ ), µp ) ∼
= ZZ/pZZ .

Proof: As above, let G = G(k̃Σ |k∞ ) and H = G(k̃Σ |kΣCM ). Then G/H ∼ =
CM ∼ + + +
G(kΣ |k∞ ) = G(kΣ (p)|k∞ ) is a free pro-p-group of rank λ by (11.3.7).
Considering the Hochschild-Serre spectral sequence for the group extension
1 −→ G(kΣ (p)|k̃Σ ) −→ G(kΣ (p)|k∞ ) −→ G −→ 1
and noting that G(kΣ (p)|k∞ ) is free (since µ = 0, cf. (11.3.7)), we obtain a
ÑÒÐÂÃÄÅÆÇÈÉÊËÌÍÎÏ X+ X+
0 H2 (G, ZZp ) G(kΣ (p)|k̃Σ )ab

G X G ab 0
ϕ1 ϕ2 ϕ3
0 EΣ− A−
Σ X− −
Xcs 0,
where the canonical map
ϕ2 : A− ∼ − rec (Tp− )ab ab

Y
Σ = UΣ −→ Gk∞,p −→ G(kΣ (p)|k̃Σ ) G
p∈Sp (k∞ )
is surjective by definition of k̃Σ . It follows that ϕ3 is injective and ϕ1 is

−
surjective. This proves (i) since the exact sequence 0 → Xcs → G ab → X + → 0
−
splits (using the action of ∆). The Λ-torsion modules X + and Xcs have
projective dimension less than or equal to one by (11.3.2) (ii),(iv) and (11.4.8).
Since µ = 0 by assumption, G ab is ZZp -free and of ZZp -rank λ+ + λ− cs , which
+
equals 2λ by (11.4.14). From the exact sequence
0 −→ H −→ G −→ G/H −→ 0,
0 −→ H/[H, G] −→ G ab −→(G/H)ab −→ 0 ,
since G/H is free, and so
−
H/[H, G] = Xcs .
Furthermore, we get from the diagram above a surjection
(ZZ/pZZ)(1)Ó ∼
= EΣ− /p (p H 2 (G, Qp /ZZp ))∨ = H 2 (G, ZZ/pZZ)∨ .
It remains to show that H 2 (G, ZZ/pZZ) =/ 0. First observe that
G(k̃Σ |kΣCM )ab ∼

= G(L̃0 |kΣCM )ab − ,
where L̃0 is the maximal p-extension of kΣCM which is completely decomposed

everywhere. Indeed, for the (−)-part this follows from the commutative exact
diagram

0ÔÕÖ×ØÙÚÛÜ CM −
Y
H 1 (L̃0 |kΣCM )− H 1 (kΣ |kΣCM )− H 1 (kΣ,p )
CM )
p∈Σ(kΣ
CM −
Y
0 H 1 (k̃Σ |kΣCM )− H 1 (kΣ |kΣCM )− H 1 (kΣ,p ) ,
CM )
p∈Σ(kΣ
where the coefficients are Qp /ZZp , and the (+)-part of Hab = G(k̃Σ |kΣCM )ab is
trivial, since this holds for the group HGab as
CM |k )
H 1 (k̃Σ |kΣCM , Qp /ZZp )G(kΣ ∞
= (H/[H, G])∨ = (Xcs
− ∨
) .
Now we consider the diagram of pairings
CM àáâãäåßÞÝ CM
|k∞ ) ∪
H 1 (k̃Σ |kΣ , µp )G(kΣ CM
× H 1 (kΣ |k∞ , ZZ/pZZ) CM
H 1 (kΣ CM
|k∞ , H 1 (k̃Σ |kΣ , µp ))∆
∪
H 1 (G, µp )∆ × H 1 (G, ZZ/pZZ)∆ H 2 (G, µp )∆
∪
H 1 (G, µp ) × H 1 (G, ZZ/pZZ) H 2 (G, µp ),
which commutes since the Hochschild-Serre spectral sequence is functorial

with respect to the cup-product. We know from (11.4.15) that
H 1 (kΣCM |k∞ , H 1 (k̃Σ |kΣCM , µp ))∆ ∼
= H 1 (kΣCM |k∞ , H 1 (L̃0 |kΣCM , µp ))∆ ∼
= ZZ/pZZ
and the upper row is a perfect pairing of finite groups. It follows that
H 2 (G, µp ) ∼
= ZZ/pZZ.
2
Now we determine the structure of the group G(k̃Σ |k∞ ), which is exactly
analogous to the structure of the fundamental group of a Riemann surface
(cf. p.610 and (10.1.2)(ii)).
(11.5.3) Theorem. Let k = k + (µp ), p =/ 2, be a CM-field such that µp ⊆ / kp+ for

all primes p above p. Let k∞ |k be the cyclotomic ZZp -extension and assume
that the Iwasawa µ-invariant of k∞ |k is zero.
Then G(k̃Σ |k∞ ) is trivial or a Demuškin group of rank 2g where g = λ+ = λ− cs .
Moreover, there are 2g generators x1 , y1 , . . . , xg , yg of G = G(k̃Σ |k∞ ) and one
defining relation g Y
[xi , yi ] = 1 .
i=1

Proof: We know that G ab ∼

+
= ZZ2λ
p is ZZp -torsion-free. If G is nontrivial,
+ / 2 ∼
i.e. λ = 0, then H (G, ZZ/pZZ) = ZZ/pZZ by (11.5.2)(ii). Furthermore, the
cup-product pairing
∪
H 1 (G, ZZ/pZZ) × H 1 (G, ZZ/pZZ) −→ H 2 (G, ZZ/pZZ)
is non-degenerate, as shown in the proof of the previous proposition. The
result follows by the theory of Demuškin groups (3.9.11). 2
(11.5.4) Corollary. With the assumptions as above, the Galois group G(k̃Σ |k)
is isomorphic to ZZp or is a Poincaré group of dimension 3 of rank 2g + 1.
Proof: This follows from (11.5.3) and (3.7.4). 2
Having the function field case (cf. X §1) and the corollary above in mind, one
is led to ask whether for an arbitrary number field k the absolute Galois group
Gk possesses a pro-p-factor group for every prime number p which is either a
Poincaré group of dimension 3 or a duality group of dimension 2 according to
whether ζp ∈ k or not (supposing we are not in the “genus 0” case). And indeed
this is the case, at least up to a finite number of “bad” primes, for every number
field. One can define an arithmetic site over the rings of integers of algebraic
number fields such that the desired factor groups occur as fundamental groups.
To be precise, it is very likely that the groups constructed have the expected
properties, but we can show this only in special situations. The appendix below
gives more details.
Appendix: An Arithmetic Site

In the following we try to give a survey of the generalized theory of positive
ramification, and we will freely use the notion of Grothendieck topologies and
their associated cohomology theories. The definitions and theorems below are
taken from [195].
We start with the definition of orientable p-adic local fields. First we define,
for an odd prime number p, a Galois extension Qpre p which is called the maximal
pre-orientable extension of Qp . If L(odd) denotes the maximal Galois extension
of a number field L whose (supernatural) degree is odd, and if Lnr denotes the
maximal unramified extension of L, then we set
Qpre
p
+
:= Qp (ζp + ζp−1 )nr (odd) ,
pre
Qp := Qpre
p
+
(ζp ) ,
where ζp is a primitive p-th root of unity.

pre + ∼
If p is odd, then the subgroup hρi = G(Qpre p |Qp ) = ZZ/2ZZ is a nor-
pre
mal subgroup of G(Qp |Qp ). Therefore the automorphism ρ acts on every
local field contained in Qpre p . The involution ρ extends the automorphism
−1
Qp (ζp ) → Qp (ζp ), ζp 7→ ζp , and we think of it as a local analogue of complex
conjugation. The field Qpre p is the maximal extension of Qp naturally having
such an involution. If a p-adic local field k is contained in Qpre
p , we denote its
subfield of ρ-invariant elements by k + . The extension k|k + is of degree 1 or 2.
(11.5.5) Definition.
(i) Let p be an odd prime number. A finite extension field k of Qp is called
orientable if
• k ⊆ Qpre
p and
• either k = k + or ζp is contained in k nr .
(ii) (Ad hoc definition) No 2-adic local field is orientable.
Note that if p is odd, every abelian extension of Qp which contains µp is ori-

entable. Unfortunately, there is still no reasonable definition of an “orientation”
on a 2-adic local field.
(11.5.6) Definition. An extension L|K of number fields is called positively

ramified at a prime p|p of L if
(i) Lp ⊆ Qpre
p Kp (⊆ Q̄2 if p = 2), and if
(ii) the Galois closure L̂p |Kp has at most pure wild ramification.
We remark that
. the cyclotomic ẐZ-extension of a number field,
. the maximal p-extension of Q(ζp + ζp−1 ) unramified outside p, and
. unramified extensions
are everywhere positively ramified.
This leads to a covering type (the positive coverings) over the rings of
integers of algebraic number fields and defines a Grothendieck topology which
is finer than the étale topology, i.e. étale coverings are positively ramified.
If K|Q is a number field, then there are only finitely many primes p of K
such that the completion Kp is not orientable. These can be thought as primes
where Spec(OK ) has “bad reduction”, since the local duality pairing associated
to Spec(OKp )pos is degenerate at these primes. Suppose d(K) is the smallest
positive integer such that Kp is orientable at all primes p not dividing d(K).
We always have d(K) | 2dK , where dK is the absolute discriminant of K. But
usually d(K) is much smaller; for example, d(K) = 2 if K is a cyclotomic
field.

The cohomology groups of sheaves of abelian groups over the positive

∗
topology are denoted by Hpos . They satisfy the following global duality
theorem, which is the exact analogue to the étale Poincaré duality theorem for
complete curves over finite fields.
(11.5.7) Theorem. Let K be a number field and let X = Spec(OK ). For every
integer n with (n, d(K)) = 1, there is a canonical trace map
3 ∼ Z
tr : Hpos (X, µn ) −→ Z/nZZ
and for every locally constant constructible sheaf F of ZZ/nZZ-modules on
Xpos , the cup-product
i
Hpos æ F ) × H 3−i (X, Hom(F, µn ))
(X, ∪ 3
Hpos ∼ Z
(X, µn ) −→ Z/nZZ
pos
induces a perfect pairing of finite groups for all i.
The fundamental group π1pos (X) of X = Spec(OK ) exists with respect to

the positive topology, since the necessary axioms are fulfilled. Let p be a
prime number. We consider the maximal pro-p-factor group π1pos (X)(p) of
π1pos (X). From the Hochschild-Serre spectral sequence, we get canonical
homomorphisms
φM,i : H i (π1pos (X)(p), M ) −→ Hpos
i
(X, M )
pos
for all i and every discrete π1 (X)(p)-module M (also considered as a locally
constant sheaf on Xpos ). We say that
X is a K(π, 1) for p
if φM,i is an isomorphism for all i and every finite p-primary π1pos (X)(p)-
module M . The analogy to function fields predicts that this should be “generi-
cally” true. One can verify this condition in some cases and then one can obtain
information about π1pos (X)(p) from the duality theorem above. In particular,
we naturally obtain corollary (11.5.4) in the following situation.
(11.5.8) Theorem. Let p be an odd prime number and let K be an abelian

extension of Q. Assume that all primes above p ramify in K|K + . Then the
following is true:
(i) If K = K + , then
(
pos free pro-p-group of finite rank, if BSp (K) = 0,
π1 (X)(p) =
duality group of dimension 2, otherwise.
(ii) If [K : K + ] = 2, then either π pos (X)(p) ∼
1 = ZZp (the genus 0 case) or
(
Poincaré group of dimension 3, if ζp ∈ K,
π1pos (X)(p) =
duality group of dimension 2, if ζp ∈
/ K.

§6. The Main Conjecture 771
§6. The Main Conjecture
In this section we want to explain how certain Iwasawa modules are con-
nected to other arithmetic objects such as p-adic L-functions, Euler character-
istics and K-groups. A full presentation of the subject is beyond the scope of
this book, so we will explain the situation, present the main results and provide
the most important references. Again the function field case is the easier one
and was a guide for the conjectures (now theorems) in the number field case,
which are known under the name “Main Conjecture” of Iwasawa theory.
1. Function Fields
Assume that k is a function field in one variable over a finite field IF = IFq of
characteristic p. ∗) Let us consider the function field analogue of the Dedekind
zeta function X 1 Y 1
ζ k (s) = s
= −s
.
a N (a) p 1 − N (p)
If r is a natural number, we denote the field kIFqr by kr and we set

br := number of primes in k of degree r,
Nr := number of primes in kr of degree 1.
Geometrically speaking, Nr is the number of IFqr -rational points of the pro-
jective curve attached to k. In particular, Nr and br are finite and linked by
P
the formula Nn = r|n rbr . We can therefore transform the product expansion
into
1 X q −rs def
= Z(q −s ) .
Y
ζ k (s) = −rs )br
= exp Nr
r (1 − q r r
The following theorem was proven by H. HASSE for elliptic curves and by
A. WEIL in the general case.
(11.6.1) Theorem.
(i) The function Z(t) is a rational function in t. In particular, ζ k (s) = Z(q −s )
can be uniquely defined as a meromorphic function on the complex plane.
(ii) We have the functional equation
1
ζ k (1 − s) = (q 2−g ) 2 −s ζ k (s) ,
where g is the genus of k.
(iii) ζ k (s) has simple poles at s = 0 and s = 1 and all zeros lie on the line
Re(s) = 12 .
∗) We assume IF to be algebraically closed in k.

The crucial step in the proof of theorem (11.6.1) is the following
(11.6.2) Theorem. Let X be the smooth projective curve associated to k and

let α1 , · · · , α2g be the zeros of the characteristic polynomial of the geometric
Frobenius automorphism F acting on the Jacobian variety Jac X. Then
(1 − α1 t) · · · (1 − α2g t)
Z(t) = .
(1 − t)(1 − qt)
Choose a prime number ` different to the characteristic p of k. By [140],

12.1, the characteristic polynomial of F acting on Jac X coincides with the
characteristic polynomial of F acting on the 2g-dimensional Q` -vector space
T` (Jac X) ⊗ ZZ` Q` ∼ ¯ ZZ` ) ⊗ Q`
= IHi := Hi (G∅ (k IF),
¯
(cf. X §1, p.613). These groups are naturally locally compact G(IF|IF)-modules.
¯
Let γ ∈ G(IF|IF) be the arithmetic Frobenius automorphism. It can be shown
that the actions of F and γ are inverse to each other. Hence we can reformulate
the result in the form
2
. ∗)
Y i+1
Z(t) = det(1 − γt | IHi )(−1)
i=1
In particular, we see that the above expression is independent of the prime

number ` =/ p.
Result: In the function field case, the zeta function is a rational function in
q −s . It therefore can also be viewed as a function on Q` (or C` ) and then
it is essentially the characteristic polynomial of the Frobenius automorphism
acting on an associated finite dimensional Q` -vector space.
The above result has vast generalizations. Instead of smooth projective
curves, one can consider arbitrary varieties over finite fields and L-functions
associated to constructible sheaves of Q` -vector spaces.
2. p-adic Zeta Functions

Now assume that k is a number field. The Dedekind zeta function ζ k (s) is
a meromorphic function on the complex plane having a simple pole at s = 1.
It is, however, not a rational function. Adding a certain Euler factor L∞ (s)
for the infinite places (see [160], chap.VII, §§4,5), we obtain a modified zeta
function with simple poles at s = 0, 1, which satisfies a functional equation
like (11.6.1)(ii) with the term (q 2−g ) replaced by |dk |−1 (loc.cit.). The question
∗) Working with `-adic cohomology IHi , one has to replace γ by γ −1 .

1
whether all zeros are on the line Re (s) = 2
is the famous (and unproven)
Riemann hypothesis. We have the
(11.6.3) Theorem (SIEGEL-KLINGEN). Let n ≥ 1 be an integer. Then

ζ k (1 − n)
is a rational number. These values vanish for n > 1 either if n is odd or if k
has a complex place. If k is totally real and n is even, then these values are
non-zero.
These special zeta-values satisfy several congruence relations ∗) which can

be reformulated as the existence of a continuous function on the p-adic num-
bers. More precisely, let k be a totally real number field, p a prime number and
d = [k(µ2p ) : k]. Then we have the following result, cf. [37]:
(11.6.4) Theorem. There exists a unique continuous function
ζ k,p : ZZp r{1} −→ Qp

satisfying
Y
ζ k,p (1 − n) = ζ k (1 − n) 1 − N (p)n−1
p|p
for all n > 1 with d | n. The function ζ k,p (s) is p-adic analytic, having at most
a simple pole at s = 1.
Observe that the second factor on the right-hand side of the defining equation
is just the Euler factor at p which has to be removed before finding a p-adic
interpolating function.
The existence of such a p-adic interpolating function has been verified for
abelian number fields by T. KUBOTA and H. W. LEOPOLDT. More generally,
they showed the existence of p-adic L-functions attached to Dirichlet characters
(see below). Many mathematicians made contributions to extend the result of
Kubota-Leopoldt to arbitrary totally real number fields. Amongst others, we
mention the names J.-P. SERRE, N. KATZ, D. BARSKY, P. CASSOU-NOGUÈS,
P. DELIGNE and K. RIBET. The general idea of the construction is to interpret
the elements of Λ as p-adic measures and then to obtain the required L-function
as a Mellin transform with respect to p-adic integration. The interested reader
should consult [122], chap. 4, or [246], chap. 12 and the references given there.
∗) They are called Bernoulli congruences if k = Q.

There exists the following p-adic analogue of the analytic class number
formula in order to compute the residue of ζ k,p at s = 1 (see [246], th. 5.24 for
abelian number fields and [30] for the general case).
(11.6.5) Theorem (p-adic Analytic Class Number Formula).
2n Rp h Y
lim (s − 1)ζ k,p (s) = q 1 − N (p)−1 ,
s→1
w |dk | p|p
where Rp is the p-adic regulator (see X §3), h is the class number and dk is
the discriminant of k; w (= 2) is the number of roots of unity contained in the
(totally real) number field k and n = [k : Q].
q
Remark: In the above formula both Rp and |dk | are determined only up to
sign but it is possible to give their ratio a well-defined sign, see [2].
(11.6.6) Corollary. ζ k,p has a (simple) pole at s = 1 if and only if the Leopoldt
conjecture is true for k and p.
In particular, this is the case if k is an abelian number field, by (10.3.16).

This connection between the residue at s = 1 of the p-adic zeta function and
the Leopoldt conjecture can be easily deduced from the main conjecture (see
below); however, it is used in the proof.
Let us now formulate the main conjecture for the p-adic zeta function. Since
the case p = 2 introduces additional, rather subtle difficulties, we will assume
that p is odd in what follows.
We denote the cyclotomic ZZp -extension of k by k∞ and we fix a topological
generator γ of Γ = G(k∞ |k). Note that we do not have a canonical generator
like the Frobenius automorphism in the function field case. Therefore we
choose any generator and all statements and results should be independent of
the choice made. Let
κ : Γ = G(k∞ |k) ∼
= G(k(µp∞ )|k(µp )) −→ ZZ×
p
be the p-part of the cyclotomic character. We set

q := κ(γ) ,
and we think of q as a substitute for the order of the ground field in the function

field case. Since q is a principal unit, we have a well-defined q-exponentiation

map
ZZp −→ ZZ× p, s 7−→ q s .
In contrast to the function field case, the p-adic zeta function has no pole at
s = 0 and it does not satisfy a functional equation. It is therefore natural to
look for an expansion around the pole at s = 1. The following result is proved
together with the existence of the p-adic zeta function.
(11.6.7) Theorem. There exists a unique power series Gk,p ∈ ZZp [[T ]] such
that
ζ k,p (s) = Gk,p (q 1−s − 1)/(q 1−s − 1) .
The main conjecture of Iwasawa theory claims that Gk,p is essentially the
characteristic polynomial of the Iwasawa module X = XΣ , i.e. of the Galois
group of the maximal abelian extension of k∞ which is unramified outside
Σ = Sp ∪ S∞ . This was first proved by B. MAZUR and A. WILES [134] under
the assumption that the base field is abelian over Q, and later by A. WILES
[249] for general totally real fields.
(11.6.8) Theorem (Main Conjecture). Let FX (t) and µ(X) be the characteristic
polynomial and the Iwasawa µ-invariant of X = XΣ . Then we have the
following equality of ideals in ZZp [[T ]]:

Gk,p (T ) = pµ(X) · FX (T ) .
In other words, the functions on both sides differ by an invertible power series.
It is not difficult to see that the above result is invariant under a change of
the generator γ ∈ Γ (changing γ also alters q = κ(γ)). Conjecturally we have
µ(X) = 0 and this is proven for abelian number fields. In general, the theorem
says that the analytic and algebraic µ-invariants coincide and that FX (T ) is the
Weierstraß polynomial associated to p−µ(X) Gk,p .
Remark: Consider the field k(µp ). By (11.4.4), there is a pseudo-isomorphism
of Λ-modules
X = e0 X(k(µp )) ≈ (e1 Xnr (k(µp )))◦ ,
where ei , i = 0, 1, are certain idempotents in ZZp [G(k(µp )|k)] (see §4). There-
fore the main conjecture can be reformulated in terms of the Iwasawa module
e1 Xnr (k(µp )).

3. Applications
The main conjecture has several important applications. Most of the theo-
rems below were first proved assuming the main conjecture before it was itself
proved.
As a first application we will consider the group K2 of rings of integers in
totally real number fields. Consider the Steinberg group St(R) of a ring R.
It is generated by symbols sij (α), i, j ∈ IN, i =/ j, α ∈ R, modulo the relations
(
1 for i =/ l, j =/ k,
sij (α)sij (β) = sij (α + β), [sij (α), skl (β)] =
sil (αβ) for j = k, i =/ l.
The group K2 (R) is defined as the center of the Steinberg group St(R). If
R is a field, then by Matsumoto’s theorem, K2 (R) coincides with the Milnor
K-group defined in VI §4. Its arithmetic importance lies in the following
exact sequence, which is part of the long localization sequence for the higher
K-theory of Quillen:
d
k(v)× −→ 0.
M
0 −→ K2 (Ok ) −→ K2 (k) −→
v finite
Here k is a number field and d is defined by the tame symbols

dv : K2 (k) → k(v)×
for all finite primes v (see VI §4, ex.1). It can be shown that K2 (Ok ) is finite,
and using Tate’s result (6.4.4), J. COATES [26] showed the existence of an
isomorphism
G(k(µp∞ )|k)
K2 (Ok )(p) ∼

= Cl(k(µp∞ ))(p) ⊗ ZZp (1) .
This allows us to compute the odd part of #K2 (Ok ) via the main conjecture.
Additional results of M. KOLSTER and A. WILES for the 2-primary part (see
[113], [249]) then imply the
(11.6.9) Theorem (Birch-Tate Conjecture). Let k be a totally real number

field and let w2 (k) be the largest positive integer N such that G(k(µN )|k) has
exponent 2. Then
#K2 (Ok ) = w2 (k) · |ζ k (−1)|.
Another application of the main conjecture (see [10]) is the calculation of

special values of zeta functions as Euler characteristics, as was conjectured by
S. LICHTENBAUM.

(11.6.10) Theorem. Let k be a totally real number field, p an odd prime

number and n be an even positive integer. Then
Y (−1)i+1
|ζ k (1 − n)|p = #H i (GSp , Qp /ZZp (n))
i=0,1
Y (−1)i
i
= #Hcts (GSp , ZZp (n)) .
i=1,2
The cohomology groups on the right-hand side have finite order by Soulé ’s
theorem (10.3.27) in conjunction with (8.7.6).
As another application, let us consider the group of cyclotomic units. It is

defined for abelian number fields in the following way (see [217]): if k is an
abelian number field, we denote its unit group by Ek . Let n be an integer > 1,
and let a be any integer not divisible by n. The number
NQ(µn )| k∩Q(µn ) (1 − ζna )
lies in k × ; we define the cyclotomic numbers Dk of k to be the group generated
in k × by −1 and all such elements NQ(µn )|k∩Q(µn ) (1 − ζna ). The cyclotomic units
are then defined by
Ck = Ek ∩ Dk .
We have (loc.cit.) the following
(11.6.11) Theorem. If k is a real abelian number field, then the group of

cyclotomic units is of finite index in the full unit group and
[Ek : Ck ] = hk · ck ,
where hk is the class number and ck is a rational number which has a nontrivial
p-part only for primes p dividing 2·[k : Q].
We conclude that the ZZp [G(k|Q)]-modules

Cl(k) ⊗ ZZp and (Ek /Ck ) ⊗ ZZp
have the same order for p - 2[k : Q]. In general, however, they are not
isomorphic. For example √(see the remark after th. 8.2. in [246]), for the
real quadratic field k = Q( 62501) the 3-part of Ek /Ck is cyclic of order 9
but Cl(k)(3) ∼= ZZ/3ZZ × ZZ/3ZZ. Nevertheless, G. GRAS conjectured that,
if p - 2[k : Q], both modules have the same image in K00 (ZZp [G(k|Q)]).
R. GREENBERG [63] showed that this conjecture is indeed a consequence of
the main conjecture.

(11.6.12) Theorem (Conjecture of Gras). Let k be a real abelian number field

and assume that the odd prime number p does not divide [k : Q]. Then the
ZZp [G(k|Q)]-modules (Ek /Ck ) ⊗ ZZp and Cl(k) ⊗ ZZp have isomorphic Jordan-
Hölder series.
The conjecture of Gras is an analogue of an earlier conjecture of Iwasawa and

Leopoldt concerning the odd eigenspace of ideal class groups of cyclotomic
fields, relating it to the quotient of ZZp [G(k|Q)] by the Stickelberger ideal. This
conjecture was proven by Mazur and Wiles [134], thereby extending earlier
results of Greenberg [61].
4. Characters
All ideas, conjectures and theorems stated so far can be refined to statements
about characters. Moreover, the main conjecture was formulated for characters
from the beginning, and was also proved in this greater generality.
We keep the assumption that k is a totally real number field and that p is an
odd prime number. Let χ be a one-dimensional even character of Gk , i.e. we
are given a continuous homomorphism
×
χ : Gk −→ Q̄
which sends every complex conjugation to 1. The image of χ is necessarily
finite and let kχ be the extension attached to χ, i.e. χ defines a faithful represen-
tation of G(kχ |k). Since χ is even, the field kχ is again totally real. Choosing
embeddings Q̄ ⊆ C and Q̄ ⊆ Cp , we will likewise view χ as a complex or as
a p-adic character. We denote the ring of integers in the field Qp (χ), obtained
from Qp by adjoining all values of χ, by Oχ = ZZp [χ]. Furthermore, let
ω : G(k(µp )|k) −→ µp−1 ⊆ ZZ×
p
be the Teichmüller character, i.e. the tame part of the cyclotomic character.
Deligne and Ribet have shown that there exists a continuous p-adic L-function
Lp (s, χ) on ZZp r{1}, and even at s = 1 if χ is not trivial, which satisfies the
interpolation property
Y
(1) Lp (1 − n, χ) = L(1 − n, χω −n ) 1 − χω −n (p)N (p)n−1
p|p
for every integer n ≥ 1. Here, L(1 − n, χω −n ) is the value of the classical

complex L-function. To make sense of this, we recall the formula
X
(2) L(1 − n, ψ) = ψ(σ)ζ k (σ, 1 − n)
σ ∈G(kψ |k)
for a complex character ψ. The value ζ k (σ, 1 − n) of the partial zeta func-
tion is rational by the theorem of Siegel-Klingen (cf. [160], chap.VII, (9.9)).

We therefore can use (2) in order to interpret L(1−n, χω −n ) as a p-adic number

in (1). If χ is the trivial character, then obviously Lp (s, χ) = ζ p,k (s).
Following Greenberg, we say that χ is of type S if kχ ∩ k∞ = k and of
type W if kχ ⊆ k∞ . Let Hχ (T ) be defined as (χ(γ)(1 + T ) − 1) if χ is of type
W and 1 otherwise. Then there exists a power series Gχ (T ) ∈ Oχ [[T ]] such
that
Lp (s, χ) = Gχ (q 1−s − 1)/Hχ (q 1−s − 1).
If ψ is of type W , then
Gχψ (T ) = Gχ (ψ(γ)(1 + T ) − 1).
The main conjecture compares Gχ with an algebraically defined polynomial

fχ for characters χ of type S. The polynomial fχ will be of the form
fχ = π µχ · fχ∗ ,
where π is a uniformizer of Oχ and fχ∗ ∈ Oχ [[T ]] is a Weierstraß polynomial.

Let us define fχ∗ first. Consider the ZZp [[Γ ]]-module
X = XSp (kχ ).
It carries a canonical G(kχ |k)-module structure. After tensoring with Q̄p , we
can decompose it into eigenspaces with respect to idempotents. We set
V = X ⊗ ZZp Q̄p ,
and we denote the χ-eigenspace of V by V (χ) ∗) .
(11.6.13) Definition. We define fχ∗ as the characteristic polynomial for the

action of γ − 1 on V (χ) .
If the order of χ is prime to p, then the definition of µχ is straightforward.
(11.6.14) Definition. If the order of χ is prime to p, then we define µχ as the

µ-invariant of the Oχ [[T ]]-module
(X ⊗ Oχ )(χ) = {x ∈ X ⊗ Oχ | gx = χ(g)x, ∀g ∈ G(kχ |k)}.
One can also define a µ-invariant for characters of order divisible by p. By

the theorem of Ferrero-Washington, these µ-invariants are equal to zero for
Dirichlet characters and in view of the conjecture that µ = 0 the following
discussion is presumably empty for all characters. Therefore the reader who
∗) i.e. V (χ) = {v ∈ V | gv = χ(g)v for all g ∈ G(kχ |k)}.

is only interested in the zeros of the p-adic L-functions may skip the following
considerations.
(We wish to thank R. Greenberg for his help with this point.) Assume that
we are given a character whose order is divisible by p. We can write it uniquely
in the form χ = ϕ · ψ, where ϕ is of order prime to p and ψ is of p-power
order, say of order pn . We have inclusions k ⊆ kψ ⊆ kχ = kϕ kψ . The idea of
the definition of µχ is the following observation from the analytic side:
Consider the p-adic L-function of ϕ as a character of the field kψ and as a
character of (the subfield of kψ of degree p of) kψp . Their ratio is the product of
the p-adic L-functions over k of the characters ϕ · ψ 0 , where ψ 0 varies over all
Q-conjugates of ψ. All these L-functions have the same analytic µ-invariant.
This motivates the following definition.
Let X (ϕ) be defined as
X (ϕ) = (XSp (kχ ) ⊗ ZZp Oϕ )(ϕ) .
Then X (ϕ) is a Oψ [H][[T ]]-module, where H := G(kψ |k). The decomposi-
n
tion of (tp − 1) into irreducible cyclotomic polynomials, together with the
evaluation of ψ, induces an isomorphism
Oϕ [H] ⊗ ZZ Q ∼
= Qp (ϕ) ⊕ Qp (ϕ)(µp ) ⊕ · · · ⊕ Qp (ϕ)(µpn ).
The projection to the highest component therefore induces a homomorphism
α : Oϕ [H] −→ Oχ ⊆ Qp (ϕ)(µpn ) = Qp (χ).
Let h be a generator of the cyclic p-group H. Then ker α is generated by

Φpn (h), where Φpn is the pn -th cyclotomic polynomial. The action of Oϕ [H]
on the submodule
n−1
(hp − 1)X (ϕ) ⊆ X (ϕ)
factors through α.
(11.6.15) Definition. For a general character χ, we define µχ as the µ-invariant

n−1
of (hp − 1)X (ϕ) , viewed as an Oχ [[T ]]-module via α.
Remark: The Weierstraß polynomial fχ∗ defined above coincides with the
n−1
characteristic polynomial of the Iwasawa module (hp − 1)X (ϕ) .
Now we are able to state the main conjecture in its general form (but only
for odd p).

(11.6.16) Main Conjecture of Iwasawa Theory. Assume that χ is even and of

type S. Then with the above notation we have an equality of ideals in Oχ [[T ]]:

Gχ (T ) = π µχ · fχ∗ (T ) .
Wiles [249] only defined the µ-invariant for characters of order prime to p
and therefore he stated his main theorem concerning the µ-invariants only for
those characters. Using the above definition of µ-invariants for characters of
arbitrary order, (11.6.16) is easily deduced from the main results th. 1.3 and
1.4 of [249].
There exists yet another generalization; namely, the existence of p-adic L-

functions can be extended to characters of arbitrary degree (provided that kχ is
totally real). This has been proved by R. GREENBERG [64], who also showed
that there exists an analogous power series expansion
Lp (s, χ) = Gχ (q 1−s − 1)/Hχ (q 1−s − 1).
However, Gχ is a priori only in the quotient field of ZZp [χ][[T ]]. The following
is the p-adic version of the famous Artin conjecture (cf. [160], chap.VII, §10).
(11.6.17) Theorem. If p is odd, then Gχ ∈ ZZp [χ][[T ]] ⊗ Qp .
Greenberg (loc.cit.) showed that (11.6.17) follows from the main conjecture.
5. Motivation
Let us finally say a few words about the history, the motivation and the proof
of the main conjecture. To make things easier, we restrict to the maximal real
abelian subfield of Q(ζp + ζp−1 ) and we assume that p is odd. The analytic part
of the arguments sketched below can be found in Washington’s book [246].
Recall that the prime number p is called regular if the class number of
k = Q(µp ) is not divisible by p. It was already known to E. KUMMER that this
is equivalent to the fact that none of the (numerators of the) Bernoulli numbers
B2 , B4 , . . . , Bp−3 is divisible by p.
This arises as follows. Using (11.4.3) and (11.1.7), it is not difficult to see
that for even i we have the equivalence
ei GΣ (k)ab /p = 0 ⇐⇒ e1−i Cl(k)(p) = 0,
with an obvious modification in the case i = 0. This equivalence is known
under the name Leopoldt’s Spiegelungssatz , see §4 ex. In particular, p is

regular if and only if p does not divide the order h− k of the (−)-part of Cl(k)(p)
with respect to the action of the complex conjugation. It follows from the
analytic class number formula that p - h− k if and only if p does not divide the
i
L-values L(0, ω ) for i = 1, 3, . . . , p − 4. Finally, the congruence
Bi+1
L(0, ω i ) ≡ − mod p
i+1
implies the statement.
It is then natural to look for a finer correspondence between the non-triviality
of ei Cl(k) and the p-valuation of L(1, ω i ) for every odd i separately. An
investigation of the Stickelberger ideal shows that L(1, ω i ) indeed annihilates
the group ei Cl(k) for i = 3, 5, . . . , p−2. We deduce (the Theorem of Herbrand)
that for these i,
ei Cl(k)(p) =/ 0 ⇒ p|Bp−i .
Iwasawa showed that the p-adic L-function can alternatively be constructed
using Stickelberger elements in the group ring ZZp [G(Q(µpn )|Q)]. This ap-
proach made it possible for him to derive a connection between the p-adic
L-functions and the quotient of the local units modulo the cyclotomic units.
More precisely, using the notation of §3, we consider the Iwasawa module of
local units U = USp (k∞ ) and its submodule
C = lim
←−
(Ckn ⊗ ZZp ),
n
where Ckn is the group of cyclotomic units in the field kn = Q(µpn ). Since
the weak Leopoldt conjecture is true for the cyclotomic ZZp -extension, cf.
(10.3.25), C is a subgroup of U .
(11.6.18) Theorem (IWASAWA). Let i 6≡ 0 mod p − 1 be even. Then
ei (U/C) ≈ Λ/fi (T ), where fi (q 1−s − 1) = Lp (s, ω i ).
This important theorem relates an analytic object, namely the p-adic L-

function, with a purely algebraic defined object, the Iwasawa module ei (U/C).
Now recall that the order of the p-part of Ekn /Ckn is essentially equal to
h+kn = #Cl(Q(ζpn + ζp−1
n ))(p) ,
and consider the exact sequence (11.3.10)

0 −→ E −→ U −→ XΣ −→ Xnr −→ 0.
We see that if h+kn is bounded independently of n, i.e. if the Greenberg
conjecture (see p. 728) holds, then we have a pseudo-isomorphism for even
i 6≡ 0 mod p − 1,
ei (U/C) ≈ ei (U/E) ≈ ei XΣ .

Therefore the Greenberg conjecture implies the main conjecture. In particu-

lar (see (11.1.7)), this proves the main conjecture for the powers of the (mod p)-
Teichmüller character if p is an odd prime number such that p - h(Q(ζp + ζp−1 ))
(i.e. if Vandiver’s conjecture holds for p).
The above results are at least an indication that one possible method to prove
the main conjecture is to compare the asymptotic orders of ei (Ekn /Ckn ) and
of ei Cl(kn )(p). Since the product over all eigenspaces is the same for both
terms by the analytic class number formula, it suffices to show an inequality
in one direction.
Indeed, there now exists a proof along these lines (cf. [184] or the new
edition of [246]) making use of the technique of Euler systems introduced by
V. A. KOLYVAGIN and F. THAINE. However, this technique can only be applied
to abelian number fields because we are lacking cyclotomic units in the more
general situation.
Mazur and Wiles applied another more geometric method. A similar class
number formula argument as above shows that it suffices to prove that one of
the ideals in the main conjecture divides the other one. Using the variant with
the Iwasawa module Xnr , this comes down to constructing many unramified
abelian extensions in the (−)-eigenspaces. Here, “many” depends on the
p-valuation of certain values of L-functions. A first important step in this
direction was achieved by K. RIBET, who proved the converse to Herbrand’s
theorem.
(11.6.19) Theorem (RIBET). For i = 3, 5, . . . , p − 2,
p|Bp−i ⇒ ei Cl(k)(p) =/ 0.
Ribet’s method of constructing the required unramified extension of Q(µp )

uses arithmetic geometry. The extension comes from a 2-dimensional repre-
sentation of GQ which is attached to a certain quotient of the modular variety
J1 (p). The quotient is associated to a normalized weight 2 cusp form whose
coefficients satisfy certain congruences and which exists if the corresponding
Bernoulli number is divisible by p (see [180]).
The proof of the main conjecture given by Mazur and Wiles for abelian
number fields, and by Wiles for arbitrary totally real number fields, strongly
refines this technique. We refer the reader to the original papers [134], [249].
The main conjecture as presented above is only the starting point of a much
more general theory of “Iwasawa theory of motives”, which is, however, to a

large extent conjectural on both the analytic and the algebraic side. As a first
step in this direction, J. COATES and A. WILES investigated p-adic L-functions
associated to elliptic curves over Q with complex multiplication by the ring
of integers of an imaginary quadratic number field. They established results
in the direction of a main conjecture relating these p-adic L-functions to the
Iwasawa module structure of the Tate module of the elliptic curve (see [29],
[33]). Using techniques developed by V. A. KOLYVAGIN, K. RUBIN proved this
main conjecture, see [185]. For related work on abelian varieties see [132],
[133], [203].
For a survey of the general ideas of motivic Iwasawa theory, we refer the
reader to the articles [204], [27], [65], [100].
Exercise 1. (Invariance of the main conjecture under liftings.) Show that it is not necessary
to work with the minimal field kχ , i.e. suppose that we are given a finite totally real extension
k 0 |kχ . Then we may consider the (not faithful) character χ0 of G(k 0 |k) which is the composite
of χ with the restriction of G(k 0 |k) to G(kχ |k). Define fχ0 in an analogous manner from X(k 0 )
and prove that fχ0 (T ) = fχ (T ) (see [64], prop.1).
Exercise 2. Define the algebraic p-adic L-function associated to a character χ by
Lalg
p (s, χ) = fχ (q
1−s
− 1)/Hχ (q 1−s − 1).
Show that this definition is independent of the choice of the generator γ ∈ Γ . Furthermore,
assume that we are given a finite abelian extension K|k of totally real number fields with
K ∩ k∞ = k. Show that
Y
ζ alg alg
p,K (s) = ζ p,k (s) Lalg
p (s, χ),
χ=
/1
where χ runs through all nontrivial characters of G(K|k).

Exercise 3. Deduce the following generalization of the main conjecture: let S ⊇ Sp be a finite
set of finite primes of k, and let χ be an even character of Gk as before. Then there exists a
continuous p-adic L-function Lp,S (s, χ) satisfying the interpolation property
Y
Lp,S (1 − n, χ) = L(1 − n, χω −n ) 1 − χω −n (p)N (p)n−1
p ∈S
for every integer n ≥ 1. Show that a variant of the main conjecture holds with the algebraic
p-adic L-function which is defined in the same way as before but with the Iwasawa module
XS instead of X = XSp .

Chapter XII
Anabelian Geometry
In this last chapter we want to give some idea of the “anabelian” program
of A. GROTHENDIECK. The term anabelian should be read as “far from being
abelian” and as we understand the matter, a group is far enough away from
being abelian if all of its subgroups of finite index have a trivial center. The
principal idea is the following: in topology, a space X of type K(π, 1) is
determined by its fundamental group π up to weak homotopy equivalence.
If we require that X is a CW-complex, then X is already determined up to
strong homotopy equivalence. The “anabelian” idea is that something similar
should also be true for schemes, i.e. a scheme X which is an étale K(π, 1)
should be essentially reconstructible from its étale fundamental group. This is
obviously not correct in general, but it should be true under certain conditions;
for example, X should be absolutely finitely generated and π1et (X) is supposed
to be “anabelian”. The smallest constituents of this anabelian world are points,
i.e. spectra of fields which are finitely generated over their prime fields. Here
the étale fundamental group is just the absolute Galois group. Finite fields
have an abelian absolute Galois group (in fact all these fields have the same
Galois group ẐZ), and so the first objects of interest are global fields. In §1,2
we will present some results on “anabelian properties” of global fields, which
already existed before Grothendieck formulated his program. We will explain
the general conjectures in §3.
§1. Subgroups of Gk
In the previous chapters we were mainly interested in arithmetically relevant
quotients of the absolute Galois group Gk of a global field k; for example,
we studied the group GS of the maximal extension of k unramified outside a
set of primes S. This gave the insight that a global field has more structure
than just the structure of an abstract field. It comes equipped with a set of
valuations which provide a family of local structures fitting together globally
in an arithmetically relevant way. On the level of Galois groups we therefore
should consider not only Gk , but Gk together with the distinguished family of

786 Chapter XII. Anabelian Geometry
closed subgroups Gp , p prime of k. It is a remarkable and surprising fact that

this distinguished family is already encoded in Gk , i.e. we can see whether a
closed subgroup in Gk is the decomposition group for a prime in a purely group
theoretical way. In other words, we can reconstruct the arithmetic structure
of Gk from its algebraic structure.
We need the following application of Krasner’s lemma (8.1.6).
(12.1.1) Lemma. Let k be a field, complete with respect to a valuation | | ∗)

and let f1 = a0,1 + a1,1 X + · · · + ad,1 X d ∈ k[X] be a separable polynomial.
Then every polynomial f2 = a0,2 + a1,2 X +· · · +ad,2 X d ∈ k[X] with sufficiently
small distance
def
|f1 − f2 | = max |aj,1 − aj,2 |
j=0,...,d
has the same splitting field as f1 .
Proof: If k = C, there is nothing to prove. Let k = IR. If f1 has splitting field

IR, i.e. if f1 has d different real zeros, then the same is obviously true for every
f2 sufficiently near to f1 . If f1 has a zero in C r IR, then every f2 near to f1
also has such a zero. Thus by the theorem of Ostrowski, [160], chap.II, (4.2),
we may assume that the valuation is nonarchimedean. Assume that f2 is
near to f1 and that α is a root of f1 in the separable closure k̄ of k. Then
|f2 (α)| = |(f2 − f1 )(α)| is small. Writing f2 = c · j (X − βj ), we see that
Q
|α − β| is small for some root β of f2 . In particular, for f2 sufficiently near to

f1 , we can choose a root β of f2 such that |β − α| < |αi − α| for all other roots
αi =/ α of f1 . This set contains all conjugates of α over k, hence, by (8.1.6),
we obtain that α ∈ k(β). Therefore the splitting field of f1 is contained in that
of f2 if f2 is sufficiently near to f1 . Now perform the above procedure for all
the zeros α1 , . . . , αd of f1 , finding for every i a root βj(i) of f2 which is near to
αi . Since α1 , . . . , αd are pairwise different, we obtain for f2 sufficiently near
to f1 a bijection αi ↔ βj(i) between the zero-sets of both polynomials, in such
a way that
|βj(i) − αi | < |βj(i) − βj |
for all i and all j =/ j(i). Another application of Krasner’s lemma shows that
the splitting field of f2 is contained in that of f1 , provided that f2 is sufficiently
near to f1 . 2
The following proposition is due to F. K. SCHMIDT [200].

∗) We only consider rank 1 valuations here, i.e. the value group is contained in IR× .
>0

§1. Subgroups of Gk 787
(12.1.2) Proposition. A proper subfield K of the separable closure k̄ of a

global field k possesses at most one prime which is indecomposable in k̄.
Proof: Suppose p1 and p2 are two different primes of K indecomposable in k̄.

Let f1 , f2 be any separable polynomials of the same degree d over K. By the
approximation theorem, for every ε > 0 there exists a polynomial f ∈ K[X]
such that |f − f1 |p1 < ε and |f − f2 |p2 < ε. For ε sufficiently small, (12.1.1)
implies that the splitting fields of f and f1 over Kp1 coincide and the same
holds for f and f2 with respect to p2 . But by assumption, p1 and p2 do not
split in k̄|K, and so the splitting fields of f1 and f2 over K are the same. In
particular, we can apply this in the case where f1 is any separable, irreducible
polynomial and f2 = di=1 (X − xi ), where xi are pairwise different elements
Q
in K. Hence K = k̄. 2
(12.1.3) Corollary. Let P1 and P2 be two distinct primes of the separable

closure k̄ of a global field k. Then
GP1 ∩ GP2 = 1 .
(12.1.4) Corollary. Let P be a prime of k̄. Then the decomposition group

GP is its own normalizer in Gk .
Proof: Let g ∈ Gk such that GgP = GP . Then GgP = GP , hence by the

previous corollary, gP = P and therefore g ∈ GP . 2
(12.1.5) Proposition. Let K|k be a finite Galois extension of global fields.

Then the canonical homomorphism
Gk −→ Aut(GK ),
which sends σ ∈ Gk to the automorphism g 7→ σgσ −1 of GK , is injective.
Proof: Assume σ lies in the kernel. Then σ fixes the open subgroup GP ∩GK
of every decomposition group GP of Gk . Hence GσP ∩ GP =/ {1}, which
implies that P = σP, i.e. σ ∈ GP for every prime P. But two different
decomposition groups have a trivial intersection. 2
In the special case when K = k, this implies the
(12.1.6) Corollary. The absolute Galois group of a global field has a trivial
center.

Since Gk has finite cohomological dimension if k is a function field, it

cannot contain finite subgroups. The same is true in the number field case
if k is totally imaginary. If k has real places, their decomposition groups
are isomorphic to ZZ/2ZZ, and it follows from the above results that all these
subgroups are different, and further, no two of them commute.
The following famous result, due to E. ARTIN [4], provides the converse
statement.
(12.1.7) Theorem. Let K be a proper subfield of the algebraic√ closure k̄ of a

number field k with [k̄ : K] < ∞. Then [k̄ : K] = 2, k̄ = K( −1) and K has
a real place. In other words, every finite subgroup in Gk is of order 2 and is
the decomposition group for a real prime of k.
√
Proof: The field K( −1) is a union of totally imaginary number fields, so
cd (GK(√−1) ) ≤ 2 by (8.3.18). On the other hand, the group GK(√−1) is finite,
√
hence it is trivial, i.e. k̄ = K( −1). Since K is a proper subfield of k̄, we
conclude that GK ∼ = ZZ/2ZZ and K has a real place by (8.3.19). (K has exactly
one real place by (12.1.2).) 2
Such a converse statement is also true for finite primes. This was first
observed by J. NEUKIRCH [152]. We say that a local field κ is of type (p0 , p1 )
if it is of characteristic p0 and has residue characteristic p1 . By convention,
IR and C are of type (0, 0). Hence the pair (p0 , p1 ) is of one of the following
forms: (0, 0), (0, p), (p, p), where p is a prime number. Note that we can read
off the type of κ from Gκ . Indeed,
- If Gκ is finite, then κ is of type (0, 0).
- If dimIFp H 1 (Gκ , IFp ) = ∞ for a prime number p, then κ is of type (p, p).
This follows from (6.1.2) and (7.1.8)(iii).
- If H 1 (Gκ , IF` ) is finite for all prime numbers `, then there exists exactly
one prime number p such that Gκ has a closed subgroup which is a
pro-p-group of rank > 2 and then κ is of type (0, p).
It might happen, however, that Gκ contains Gκ0 as a closed subgroup of
infinite index and κ is of type (0, p), while κ0 is of type (p, p). Moreover, this
happens quite often as J.-P. WINTENBERGER has shown:
Let κ|Qp be a finite extension and let κ∞ be a ramified ZZp -

extension of κ. Then there exists a local field κ0 of characteristic p
such that Gκ∞ ∼= Gκ0 .

§1. Subgroups of Gk 789
For a proof we refer the reader to the original article [259]. The field κ0
is called the field of norms, since we have an isomorphism lim κ× ∼
= (κ0 )× .
←− n
Moreover, such a field κ0 exists for a larger class of infinite extensions of Qp : it
suffices that κ∞ |κ is a so-called arithmetically profinite extension (loc.cit.).
Let κ be a finite extension of Qp . Then, in addition to the prime number p,
we can also reconstruct the ramification group Vκ and the inertia group Tκ from
Gκ as follows:
Vκ is the uniquely determined maximal element among the normal,
closed subgroups in Gκ which are pro-p-groups, by (7.5.7)(i).
Tκ /Vκ is the uniquely determined maximal element among the abelian,
normal, closed subgroups in Gκ /Vκ , by (7.5.7)(ii). Hence Tκ is
also uniquely determined by Gκ .
The order q of the residue field of κ can also be reconstructed from Gκ .
Indeed, by (7.3.10), we obtain for ` =/ p the number #µ`∞ (κ) = w`1 as the
order of the quotient of H 1 (Gκ , Q` /ZZ` ) by its maximal divisible subgroup and
q − 1 = `=/p w`1 .
Q
As usual, we call an element F ∈ Gκ a Frobenius lift if its image in

Gκ /Tκ ∼= G(κnr |κ) is equal to the arithmetic Frobenius automorphism.
(12.1.8) Lemma. An element F ∈ Gκ is a Frobenius lift if and only if for

every t ∈ Tκ
F tF −1 ≡ tq mod Vκ .
In particular, the set of Frobenius lifts is determined by Gκ .
Proof: By (7.5.4)(ii), the map G(κnr |κ) −→ Aut(Tκ /Vκ ) is injective. There-
fore F is a Frobenius lift if and only if conjugation by F defines the same
automorphism of Tκ /Vκ as the conjugation with a Frobenius lift. 2
Returning to our problem of detecting local groups in a global Galois group,

we have the following
(12.1.9) Theorem (NEUKIRCH). Let k be a global field, κ a nonarchimedean

local field, and assume that Gk has a closed subgroup H ∼ = Gκ . Then there
exists a unique prime p in k and a unique extension P of p to k̄ such that
H ⊆ GP .
If κ is a finite extension of Qp for some prime number p, then k is a number
field and (GP : H) < ∞. Furthermore, in this case p|p and [κ : Qp ] ≥
[kp : Qp ].

For the proof we need the
(12.1.10) Lemma. Let k be a global field, P a prime of k̄ and H an infinite

closed subgroup in Gk such that
(H : H ∩ GP ) < ∞.
Then H ⊆ GP .
Proof: Let K = k̄ H and let L = k̄ U for some open subgroup U ⊆ H ∩ GP

which is normal in H. Then [L : K] < ∞ and L is henselian with respect to
P ∩ L, i.e. P ∩ L is indecomposable in k̄|L. All extensions of P ∩ K to L are
conjugate, and so L is also henselian with respect to these other primes. Since
H was infinite, L is not separably closed, hence by (12.1.2), P ∩ L must be
the only extension of P ∩ K to L. Therefore H ⊆ GP . 2
Proof of (12.1.9): The uniqueness of p and P follows from (12.1.3). In

order to prove their existence, we can use (12.1.10) to replace H and Gk
by suitable open subgroups. Thus, fixing an arbitrary odd prime number `
different to char(κ) and char(k), we may assume that µ` ⊆ k and µ` ⊆ κ. Then
H 2 (U, IF` ) ∼
= IF` for every open subgroup of H by (7.1.8) (ii). Setting K = k̄ H ,
consider the injective map
Y
H 2 (GK , IF` ) ,→ H 2 (GKP , IF` )
P
which is obtained from (9.1.10) by passing to the limit. It follows that

H 2 (GKP , IF` ) =/ 0
for at least one prime P of K, which is nonarchimedean because ` =/ 2.
We claim that P does not decompose in k̄|K. Indeed, every finite separable
extension L of K corresponds to an open subgroup of H and (9.2.1) implies a
surjection
IF` ∼
Y
= H 2 (GL , IF` ) H 2 (GL 0 , IF` ).
P
P0 |P
Recall that H 2 (GKP , IF` ) =/ 0 implies that H 2 (V, IF` ) =/ 0 for every open
subgroup V of GKP , by (7.1.8) (i),(ii). Hence P does not decompose in L|K
and L was arbitrary, so that H = GKP .
We denote the unique extension of P to k̄ also by P and put p = P ∩ k.
Then we have the inclusion GKP ⊆ GP , where GP is the decomposition group
of P in Gk .
Now assume that κ is a finite extension of Qp . Then cd` H = 2 for all prime
numbers `. In particular, k must be a number field, cf. (6.1.3). Furthermore,

§2. The Neukirch-Uchida Theorem 791
GP ⊇ H ∼ = Gκ contains closed subgroups which are pro-p-groups of rank

greater than 2, so p|p. An inspection of H 2 (−, IFp ) implies that µp ⊆ κ if
and only if µp ⊆ KP and p∞ - [KP : Qp ]. Thus the p-part of (GP : H) is
finite. Suppose (GP : H) is infinite. For open subgroups V in GP containing
res
H with p - (V : H), the map H 1 (V, IFp ) −→ H 1 (H, IFp ) is injective. But
H 1 (H, IFp ) is finite and #H 1 (V, IFp ) becomes arbitrarily large as (GP : V )
tends to infinity. Hence (GP : H) < ∞, and (7.3.9) shows that [κ : Qp ] =
[KP : Qp ] ≥ [kp : Qp ]. 2
(12.1.11) Corollary. Let k be a global field. A closed subgroup H ⊆ Gk is the

decomposition group of a prime if and only if H is maximal among the closed
subgroups which are isomorphic to the absolute Galois group of a local field.
Exercise 1. Let k be a global field and let H be a pro-abelian subgroup of Gk . Then either
H∼= ZZ/2ZZ or
H∼
Y a
= ZZ` `
`
where a` ∈ {0, 1}. The first case can only occur if k is a number field which is not totally
imaginary.
Hint: Use scd` (ZZ` × ZZ` ) = 3.
Exercise 2. Let k be a global field. Assume that the closed subgroup H ⊆ Gk is a pro-`
Demuškin group. Show that H is contained in the decomposition group GP of a uniquely
determined prime P. If H is infinite (i.e. not cyclic of order 2), then P is finite and of residue
characteristic different to `; in particular, char(k) =/ `. Furthermore, H ∼ = ZZ` o ZZ` in this
case.
§2. The Neukirch-Uchida Theorem
Now we come to the question of what extent a global field is characterized

by its absolute Galois group. Recall that the absolute Galois group Gk is not
an invariant of k: it depends on the choice of a separable closure k̄ of k, and
therefore Gk is determined by k only up to inner automorphisms. This detail is
unimportant as long as one considers abelian class field theory, and by (1.6.3)
it has also no effect on cohomological considerations. In the language of étale
fundamental groups (and by analogy to topology) the choice of k̄ is the choice
of a base point on which the fundamental group depends.

We will restrict to the number field case in this section. The main result is the-
orem (12.2.1), which is due to J. NEUKIRCH and K. UCHIDA [152], [153], [237].
Based on results of Neukirch, this theorem was also proven by M. IKEDA [84]
and K. IWASAWA (unpublished).
Let k1 and k2 be two number fields and let k̄1 and k̄2 be fixed separable
∼ k and let
closures. Let Iso(k2 , k1 ) be the set of field isomorphisms k2 → 1
Iso(k̄2 |k2 , k̄1 |k1 ) denote the set
n o
∼ k̄ | α(k ) = k
Iso(k̄2 |k2 , k̄1 |k1 ) = α : k̄2 −→ 1 2 1 .
We consider the Galois groups G(k̄1 |k1 ) and G(k̄2 |k2 ) ∗) . G(k̄2 |k2 ) acts on
Iso(k̄2 |k2 , k̄1 |k1 ) by the rule σ(φ) = φ ◦ σ −1 and we have an isomorphism
Iso(k̄2 |k2 , k̄1 |k1 )/G(k̄2 |k2 ) ∼

= Iso(k2 , k1 ).
An element α ∈ Iso(k̄2 |k2 , k̄1 |k1 ) induces an isomorphism
α∗ : G(k̄1 |k1 ) −→
∼ G(k̄ |k )
2 2
by α∗ (g1 )(x2 ) = α−1 (g1 (α(x2 ))), g1 ∈ G(k̄1 |k1 ), x2 ∈ k̄2 .
(12.2.1) Theorem. Let k1 and k2 be number fields and let

∼ G(k̄ |k )
σ : G(k̄1 |k1 ) −→ 2 2
be an isomorphism of profinite groups. Then there exists a unique element

α ∈ Iso(k̄2 |k2 , k̄1 |k1 ) inducing σ, i.e. σ = α∗ .
In order to reformulate this theorem in terms of absolute Galois groups

without fixing a separable closure, we introduce the following notation. Let
G1 , G2 be profinite groups. We denote the set of isomorphisms of profinite
groups G1 → ∼ G by Iso(G , G ) and the group of inner automorphisms of G
2 1 2 i
by Inn(Gi ), i = 1, 2.
The group Inn(G2 ) acts on Iso(G1 , G2 ) by the rule σ(φ) = σ ◦ φ and we call
OutIso(G1 , G2 ) := Iso(G1 , G2 )/Inn(G2 )
the set of outer isomorphisms from G1 to G2 . This notation has the advantage
that for fields k1 , k2 the set
OutIso(Gk1 , Gk2 )
∗) In geometric language, these are the groups π et (Spec k , Spec k̄ ).
1 i i

∼ k
does not depend on the choice of separable closures. An isomorphism k2 → 1
∼
can be extended to an isomorphism k̄2 → k̄1 of arbitrarily chosen separable
closures and therefore induces a well-defined element in OutIso(Gk1 , Gk2 ).
Theorem (12.2.1) can now be reformulated in the following way.
(12.2.2) Corollary. Let k1 and k2 be number fields. Then the natural map
Iso(k2 , k1 ) −→ OutIso(Gk1 , Gk2 )
is an isomorphism.
In particular, for a number field k there is a canonical isomorphism
∼ Aut(G )/Inn(G ) =: Out(G ).
Aut(k) −→ k k k
(12.2.3) Corollary. All automorphisms of GQ are inner. Moreover, the

GQ −→ Aut(GQ )
which sends g ∈ GQ to the automorphism h 7→ ghg −1 is an isomorphism.
Let us deduce the corollaries first.
Proof of (12.2.2) and (12.2.3): By theorem (12.2.1), we obtain an isomor-

phism of sets
Iso(k̄2 |k2 , k̄1 |k1 ) −→ Iso(G(k̄1 |k1 ), G(k̄2 |k2 ))
α 7−→ α∗
which is easily seen to be G(k̄2 |k2 )-invariant if we let G(k̄2 |k2 ) act by inner
automorphisms on the right-hand side. Factoring out by the G(k̄2 |k2 )-action,
we obtain the required isomorphism
Iso(k2 , k1 ) −→∼ OutIso(G , G ).
k1 k2
This shows (12.2.2). Applying this result to the case k = Q, we obtain
Aut(GQ ) = Inn(GQ ), because Aut(Q) = 1. Finally, the homomorphism GQ →
Inn(GQ ) is injective and hence an isomorphism, because GQ has a trivial center
by (12.1.6). 2
The first step towards a proof of (12.2.1) is to establish a local correspon-

dence. Recall the definition of Sp(K) from X §1. Sp(K) is a totally discon-
nected Hausdorff topological space whose underlying set is the set of primes
of K plus one generic point. Suppose that we are given an isomorphism
∼ G(k̄ |k ).
σ : G(k̄1 |k1 ) −→ 2 2

Then (12.1.9) shows that for every prime P1 of k̄1 , the group σ(GP1 ) is
the decomposition group of a uniquely determined prime P2 of k̄2 . Send-
ing the generic point of Sp(k̄1 ) to that of Sp(k̄2 ), this induces a bijection
∼ Sp(k̄ ).
σ∗ : Sp(k̄1 ) −→ 2
(12.2.4) Local Correspondence. The bijection

∼ Sp(k̄ )
σ∗ : Sp(k̄1 ) −→ 2
is a homeomorphism. If the extension field K1 , k1 ⊆ K1 ⊆ k̄1 , corresponds via

σ to the field K2 , k2 ⊆ K2 ⊆ k̄2 , then σ∗ induces a homeomorphism
σ∗,K1 ,K2 :èéç Sp(K1 ) Sp(K2 )
Sp(Q)
which commutes with the canonical projection to Sp(Q). If p ∈ Sp (K1 ) for a

prime number p, then [K1,p : Qp ] = [K2,σ∗ (p) : Qp ].
Proof: We have to show that if two primes P and Q in Sp(k̄1 ) restrict to the
same prime in K1 , then σ∗ (P), σ∗ (Q) ∈ Sp(k̄2 ) restrict to the same prime in K2 .
But P and Q restrict to the same prime in K1 if and only if GP ∩ G(k̄1 |K1 ) and
GQ ∩ G(k̄1 |K1 ) are conjugate subgroups in G(k̄1 |K1 ) and this easily translates
to the other side of the correspondence. Hence
∼ Sp(K )
σ∗,K1 ,K2 : Sp(K1 ) −→ 2
is a well-defined bijection which commutes with the natural projection to

Sp(Q), since the type of KP is encoded in GP ; see the discussion before
(12.1.8). If K1 is finite over k1 (and hence also [K2 : k2 ] < ∞), then
σ∗,K1 ,K2 is automatically a homeomorphism because Sp(Ki ) is the one-point-
compactification of the discrete set of primes of Ki , i = 1, 2. If K1 , K2 are
of infinite degree, then σ∗,K1 ,K2 can be identified with the inverse limit of the
corresponding maps on the finite levels. Further, the topology on Sp(Ki ) is the
inverse limit topology obtained from the finite levels. Hence σ∗,K1 ,K2 is also
continuous in the general case. In particular, this applies to the case K1 = k̄1 ,
K2 = k̄2 . Finally, the remaining equality of degrees over Qp follows from
(12.1.9). 2
In order to deduce (12.2.1), we recall the following application of Čebotarev’s

density theorem, cf. [160], chap.VII, (13.9).

(12.2.5) Theorem. Let K be a finite normal extension of Q and let L|Q be

finite. If every prime number p which has a prime factor of degree 1 in L splits
completely in K|Q, then K ⊆ L.
Proof: Consider the finite normal extension KL|L. Every prime of degree 1
in L splits completely in KL by assumption. But this set of primes has Dirichlet
density 1, so that KL = L by Čebotarev’s density theorem. 2
(12.2.6) Corollary. Let k be a number field and assume that every prime
number p which has a prime factor of degree 1 in k splits completely in k|Q.
Then k|Q is normal.
Proof: Apply (12.2.5) in the situation when L = k and K is the normal

closure of k over Q. 2
Proof of theorem (12.2.1): (We follow [152] and [237] with a simplification
due to H. W. LENSTRA at the end.) Let A be the algebraic closure of Q in
C (i.e. A is a fixed model for Q̄). Identifying k̄1 ∼
= A and k̄2 ∼ = A via any
isomorphisms, we consider k1 and k2 as subfields of A and we have to show:
Every isomorphism σ : G(A|k1 ) → ∼ G(A|k ) is induced by a uniquely
2
determined automorphism α ∈ Aut(A) such that α(k2 ) = k1 and σ = α∗ .
In order to simplify notation, we make the following convention: if K1 is
an extension of k1 in A, then we denote by K2 the extension of k2 in A which
corresponds to K1 via the isomorphism σ, and vice versa.
The local correspondence (12.2.4) now shows that the set of prime numbers
which have a prime factor of degree 1 in K1 coincides with the set of prime
numbers which have a prime factor of degree 1 in K2 . The same is true for the
set of prime numbers which split completely in K1 resp. K2 . Hence if K1 is
finite and normal over Q, then K2 |Q is normal by (12.2.6), and (12.2.5) implies
that K1 = K2 .
From now on the letter N will always denote a subfield of A which con-
tains the composite k1 k2 and which is finite and normal over Q. The above
observations show that σ(N ) = N , and we obtain an induced homomorphism
σN : G(N |k1 ) → G(N |k2 ). Furthermore, for every α ∈ Aut(A), we have
α∗ (N ) = N and we obtain an induced homomorphism αN ∗
: G(N |Q) →
G(N |Q).
We first show the uniqueness of α. Assume that we have α1 , α2 ∈ Aut(A)
such that αi (k2 ) = k1 for i = 1, 2 and α1∗ = α2∗ : G(A|k1 ) →
∼ G(A|k ). Then
2
for every field N which satisfies the above convention, we have αi (N ) = N

and α1∗ |G(A|N ) = α2∗ |G(A|N ) . Therefore α1 and α2 have the same image under
the canonical map Aut(A) = G(A|Q) → Aut(G(A|N )). Hence α1 = α2 by
(12.1.5).
In order to prove the existence of α, it suffices to show that for every N ,
there exists an αN ∈ Aut(A) with αN (k2 ) = k1 and such that the induced
isomorphism
(αN )∗N : G(N |k1 ) −→
∼ G(N |k )
2
∼ G(N |k ) induced by σ.
is equal to σN , the isomorphism G(N |k1 ) → 2
Indeed, having shown this, the compact sets
AN = {αN ∈ Aut(A) | (αN )∗N = σN }
are nonempty and define a projective system for which A = lim ←− N
AN is
nonempty.
Now we fix N and we write G = G(N |Q), Gi = G(N |ki ), i = 1, 2. Let
us first assume that N |k2 is cyclic and let F be a generator of G(N |k2 ). By
Čebotarev’s density theorem, there exists a prime P such that GP ⊆ G(A|k2 ),
P ∩ k2 is unramified in the extension N |k2 and F ≡ FrobP mod G(A|N ).
Then σ −1 GP = GP̃ for some prime P̃|p, so there exists an αN ∈ Aut(A)
such that P̃ = αN P. It follows that αN (k2 ) = k1 because
αN (k2 ) = k1 ⇐⇒ G(A|k2 ) = (αN )∗ G(A|k1 )
⇐⇒ G(A|N )GP = (αN )∗ (G(A|N )GP̃ )
⇐⇒ GP = (αN )−1 GP̃ αN
and the last equality is true since P̃ = αN P.
Fixing any Frobenius lift FrobP ∈ GP , we know from lemma (12.1.8)
that σ −1 (FrobP ) ∈ GαN P is a Frobenius lift and the same is trivially true for
αN FrobP (αN )−1 . Since αN P is unramified and does not split in the cyclic
extension N |k1 , we observe that the two homomorphisms
(αN )∗N , σN : G(N |k1 ) → G(N |k2 )
coincide on a generator of G(N |k1 ) and hence are equal.
In order to deal with the general situation, let n = #G and let p be any prime
number greater than n. The split embedding problem
êëìíî G(k̄1 |Q)
1 IFp [G] E G(N |Q) 1

has a proper solution M |Q by (9.2.9), i.e. E = G(M |Q) and IFp [G] = G(M |N ).
Then σ induces isomorphisms
∼ G(M |k )
σM : G(M |k1 ) −→ 2
and

N ∼ IF [G].
σM := σM |G(M |N ) : IFp [G] −→ p
N
We are going to describe σM more explicitly. Let λ ∈ IFp [G] be any non-zero
element of the group ring and let L1 and L2 be the fixed field of the subgroup
hλi ⊆ IFp [G] ⊆ E = G(M |Q) and of σM (hλi) ⊆ IFp [G] ⊆ G(M |Q), so that
∼ G(A|L ).
σ : G(A|L1 ) −→ 2
We have the following diagram of fields

ñòóôõö÷øùïð A
M
hλi
L1 IFp [G] L2
k1 G k2
Q.
By the above considerations in the cyclic case, we see the existence of an
α = αM ∈ Aut(A) such that
∗
σM = αM : G(M |L1 ) → G(M |L2 ).
∗
The automorphism α : G(A|Q) → G(A|Q) induces an automorphism α∗ :
G → G, which is just conjugation with h = α−1 mod G(A|N ) ∈ G. Then
∗
αM : G(M |N ) → G(M |N )
is left multiplication by h on the G-module G(M |N ) ∼ = IFp [G]. In particular,
N ∗
σM (λ) = αM (λ) = hλ ∈ IFp [G].
For each h ∈ G consider
N
Uh = {λ ∈ IFp [G] | σM (λ) = hλ} ⊆ IFp [G].
The union of the subgroups Uh , h ∈ G, is all of IFp [G]. Since IFp [G] is a
p-group and p > n = #G, a counting argument shows that one of the Uh must
be equal to IFp [G]. Therefore there exists an h0 ∈ G with
N
(∗) σM (λ) = h0 λ for all λ ∈ IFp [G].
Now let g1 ∈ G(N |k1 ) be arbitrary. Then we have
N N
σM (g1 ) = σN (g1 )σM (1)
by the definition of the occurring maps. Applying (∗) to λ = g1 , we obtain
N N
h0 g1 = σM (g1 ) = σN (g1 )σM (1) = σN (g1 )h0 .
−1 ∗
Hence σN (g1 ) = h0 g1 h0 = α (g1 ), which shows that αN := α satisfies the
required condition. 2

Closing remark: K. UCHIDA [238] proved theorem (12.2.1) (with a small

modification necessary in positive characteristic) also in the function field
case. Furthermore, the result remains true after replacing the absolute Galois
groups by their maximal prosolvable quotients. F. POP [172], [173], [175]
showed that (12.2.1) is true in the much more general situation that k1 , k2 are
infinite fields which are finitely generated over their prime fields; see also
[224].
Finite fields are obvious examples of fields which are not anabelian. Another
example is p-adic local fields, which can be seen as follows: as is well-known,
every automorphism of a finite extension k|Qp is continuous∗) ; in particular,
Aut(Qp ) = 1. If (12.2.1) (and hence also (12.2.2)) were true for p-adic local
fields, then taking k1 = k2 = Qp we would obtain an isomorphism
∼ Out(G ) = Aut(G )/Inn(G ).
1 = Aut(Qp ) −→ Qp Qp Qp
But we constructed a nontrivial outer automorphism of GQp in VII §5, so p-

adic local fields do not deserve to be called anabelian. This can also be seen
using (7.5.3): it is not difficult to construct local fields which are not isomorphic
but have the same absolute Galois group.
However, there is the following variant due to S. MOCHIZUKI [144].
(12.2.7) Theorem. Let k1 and k2 be finite extensions of Qp . Let IsoQp (k2 , k1 )

denote the set of Qp -isomorphisms from k2 to k1 and let OutIsoF ilt (Gk1 , Gk2 )
denote the set of outer isomorphisms of filtered groups between the absolute
Galois groups of k1 and k2 equipped with the filtrations defined by the higher
ramification groups in the upper numbering. Then the natural map
IsoQp (k2 , k1 ) −→ OutIsoF ilt (Gk1 , Gk2 )
is an isomorphism.
§3. Anabelian Conjectures
In this last section we present the anabelian conjecture(s) in detail, thereby

making free use of the language of schemes and their étale fundamental groups.
∗) Indeed, choose a prime number ` =/ p such that µ (k) = 1. Then the unit group U
` k
can particularly be characterized as the subgroup of `-divisible elements in k × . Thus every
automorphism φ ∈ Aut(k) fixes Uk . An element a ∈ Ok is either a unit or of positive valuation
and the latter is characterized by the property that a + u ∈ Uk for every u ∈ Uk . Hence
φ(Ok ) = Ok , which implies that φ is continuous.

§3. Anabelian Conjectures 799
Let k be a field with separable closure k̄ and let X be a scheme of finite

type over k which is geometrically connected. Fixing a geometric point x̄ :
Spec k̄ → X, we have the following exact sequence of profinite groups
X p
1 −→ π1 (X̄, x̄) −→ π1 (X, x̄) −→ G(k̄|k) −→ 1,
where X̄ is the base change from X to k̄ and π1 denotes the étale fundamental
group.
When k is finitely generated over Q, A. GROTHENDIECK posed the conjec-
ture that
if X is “anabelian”, then it is functorially determined by pX .
In order to make this more precise, we introduce some notation. If G1 , G2
denote profinite groups, let
Homop (G1 , G2 )
be the set of continuous homomorphisms φ : G1 → G2 with open image. If
G1 , G2 are augmented, i.e. if we are given homomorphisms pi : Gi → G,
i = 1, 2, to another profinite group G, then Homop G (G1 , G2 ) is the subset of
op
Hom (G1 , G2 ) consisting of homomorphisms φ with φ ◦ p2 = p1 . If H2 ⊆ G2
is a closed subgroup with trivial center, then Inn(H2 ) acts on Homop (G1 , G2 )
and it acts on HomopG (G1 , G2 ) if H2 is contained in the kernel of p2 .
For schemes X1 , X2 , we denote the set of dominant morphisms from X1 to
X2 by
Mordom (X1 , X2 ).
Following Grothendieck’s philosophy, there should exist a full subcategory
Ank of the category of schemes of finite type over k such that (in a base point
free version as in §2) the following holds:
(12.3.1) Anabelian Bijections.

(i) (For isomorphisms): If X1 , X2 ∈ Ank , then the canonical map
IsoSch|k (X1 , X2 ) −→ IsoGk (π1 (X1 ), π1 (X2 ))/Inn(π1 (X̄2 ))
is a bijection.
(ii) (For dominant morphisms): If X1 , X2 ∈ Ank , then the canonical map
MorSch|k,dom (X1 , X2 ) −→ Homop
Gk (π1 (X1 ), π1 (X2 ))/Inn(π1 (X̄2 ))
is a bijection.
(iii) (For sections): For every proper curve X ∈ Ank and for every finite
separable extension K|k, the canonical map
MorSch|k (Spec K, X) −→ HomGk (GK , π1 (X))/Inn(π1 (X̄))
is a bijection.

Clearly (i) follows from (ii). Note that the triviality of the center of π1 (X̄2 )
is an implicit assumption. It is part of the philosophy that the property of being
anabelian should be geometric, i.e. whether X is in Ank should only depend
on X̄. But which schemes are anabelian?
Following Grothendieck, the category Ank should contain all hyperbolic
curves and successive fibrations of such curves over each other. Here a smooth,
geometrically connected curve X over k is called hyperbolic if it satisfies:
def
χ(X) = 2 − 2g − n < 0, where g is the genus of the smooth compactification
C of X and n is the cardinality of C(k̄) r X(k̄). In particular, all curves of
genus ≥ 2 are hyperbolic and (using “Artin neighbourhoods”) every point on
a smooth variety admits a fundamental system of anabelian neighbourhoods.
Furthermore, Grothendieck conjectured that the moduli spaces Mg,k of curves
of a given genus g over k are also anabelian.
Consider the following variant of (12.3.1)(i).
(12.3.2) Absolute Anabelian Bijection of Isomorphisms.

If X1 , X2 ∈ Ank , then the canonical map
IsoSchemes (X1 , X2 ) −→ Iso(π1 (X1 ), π1 (X2 ))/Inn(π1 (X2 ))
is a bijection.
The generalization of theorem (12.2.1) to fields which are finitely generated

over Q, proved by F. POP (cf. §2), should be seen as a birational version of
(12.3.2). A birational version of (12.3.1)(ii) was proved by S. MOCHIZUKI
[145]. One can ask whether an absolute birational version of (12.3.1)(ii) is
also true. Since homomorphisms of fields are always injective, the following
question naturally arises:
(12.3.3) Question. Assume that k1 , k2 are finitely generated over Q. Is every

open homomorphism Gk1 → Gk2 injective?
We do not know the answer to question (12.3.3), even in the case when k1
and k2 are number fields.
If k|Q is finitely generated, then the subgroup π1 (X̄, x̄) ⊆ π1 (X, x̄) can be
detected in a purely group theoretical way as follows: since X is of finite type,
π1 (X̄, x̄) is finitely generated. On the other hand, the field k is Hilbertian, so
G(k̄|k) does not contain a nontrivial finitely generated normal subgroup, by

[53], th. 15.10. Hence π1 (X, x̄) possesses a unique maximal finitely generated
normal subgroup which is just its geometric part π1 (X̄, x̄). Using Pop’s re-
sults about absolute Galois groups, we see that (12.3.1)(ii) is equivalent to its
absolute version (12.3.2) if k|Q is finitely generated.
Suppose we are given a proper curve of genus greater than or equal to 2 over
a number field. If we are interested in the isomorphism class of the curve, then,
just by knowing that it is defined over a number field, we know that there are
at most a countable number of possibilities. If we give ourselves, in addition,
the Tate module of the curve (and this can be easily determined from the étale
fundamental group), then we know the primes where the Jacobian of the curve
has bad reduction. Hence by the Šafarevič conjecture, proved by G. FALTINGS
[47], and even before we discuss (12.3.1)(i), the curves are already determined
up to a finite number of possibilities. It is therefore surprising how difficult
the problem is, even in the case of curves. For a period of more than ten years
(Grothendieck formulated his conjecture in a letter to Faltings in 1983 [68])
only a few partial results (e.g. [149]) had been known.
This rapidly changed in 1995, when A. TAMAGAWA made the observation
that (12.3.2) is true for affine hyperbolic curves over finite fields [227]. This
came rather unexpectedly since finite fields themselves are not anabelian.∗)
Let us briefly explain the ideas of Tamagawa’s proof:
In the first step a local correspondence is established. Recall that in the case
of global fields (see §1), the places could be characterized by their decomposi-
tion groups and the latter could be detected via their second cohomology. Let
us denote the function field of the curve X by K and let S be the finite set of
places of K which do not lie on X. The field K is a global field of positive
characteristic and π1 (X) = GS (K). Every prime of K which lies on X has
a decomposition group isomorphic to ẐZ in GS (K). Thus one is confronted
with the (seemingly impenetrable) task of characterizing the decomposition
groups among the huge set of subgroups of GS (K) which are isomorphic to ẐZ.
Tamagawa solved this problem in an elegant way: assume for simplicity that
x ∈ X(k) is a prime of K that corresponds to a k-rational point of X and let x̃
be a prolongation of x to KS . Let L be a finite extension of K in KS and let XL
be the normalization of X in L (which is étale over X). Then the restriction
of x̃ to L defines a k-rational point on XL if GS (L) contains the decomposition
group Gx̃ (KS |K). In particular, we have #XL (k) > 0 for those L. Via the
Lefschetz trace formula, the last equality can be reformulated into a cohomo-
logical statement which can (for large enough L) be solely expressed in terms
of GS (L). This makes it possible to characterize the decomposition groups by
the way in which they lie inside GS (K).
∗) Note that (12.3.1)(i) is false over finite fields.

The global part of the proof generalizes that of Uchida [238] for the func-
tion field analogue of (12.2.1) to a large extent (at this point, however, the
assumption of X being affine enters).
Then (12.3.1)(i) for fields which are finitely generated over Q can be derived
from (12.3.2) for finite fields [227]. In this process the characterization of good
reduction of curves via outer pro-` Galois representations (due to T. ODA, cf.
[164], [227], th. 5.3) is applied. This showed (12.3.1)(i) for affine hyperbolic
curves over fields which are finitely generated over Q. Finally, the restriction
to affine curves was removed by S. MOCHIZUKI using methods of logarithmic
algebraic geometry [143].∗∗)
Only a short time afterwards, Mochizuki [145] (see also [48]) achieved fur-
ther progress by changing the point of view. He claimed that the Grothendieck
Conjecture for hyperbolic curves is best understood not as a global, num-
ber theoretical result, but rather as a p-adic result. He succeeded in proving
(12.3.1)(ii) for hyperbolic curves over sub-p-adic local fields, i.e. subfields of
finitely generated field extensions of Qp . This particularly includes all fields
which are finitely generated over Q. The proof is completely independent
of Tamagawa’s approach and uses p-adic Hodge theory, a theory which in-
vestigates properties of p-adic Galois representations arising from arithmetic
geometry. Mochizuki obtains even stronger results. Among other generaliza-
tions, he also shows, cf. [145], th. D, that (12.3.1)(i) holds for hyperbolically
fibred surfaces over a sub-p-adic field.
Concerning (12.3.1)(iii), which is also referred to as the anabelian sec-

tion conjecture, only the injectivity of the canonical map is known so far
([145], th. C). A positive answer to the section conjecture would be of special
interest because of the following vague hope: let X be a proper curve of
genus ≥ 2 over a number field. The set HomGk (Gk , π1 (X))/Inn(π1 (X̄)) should
belong in some sense to a “compact” world, while X(k) is “discrete”. If the
section conjecture holds, this fact could possibly be used in order to show the
finiteness of #X(k). This would yield a new proof of the famous Mordell
conjecture, proved by G. FALTINGS.
The analogy between number fields and function fields, together with Tama-
gawa’s results, raises the following question.
(12.3.4) Question. Are a number field k and a set S of places of k functorially

determined by GS (k) if S is large enough?
∗∗) J. STIX has shown variants of these results for curves over absolutely finitely generated
fields of positive characteristic, see [221], [222].

This might be a little bit too optimistic because, in contrast to the function
field case, we cannot go up the cyclotomic ẐZ-extension of k inside kS . Writing
S
k(µ) = n k(µn ), we conclude this chapter with the following
(12.3.5) Conjecture. Let k1 , k2 be number fields and let Si be sufficiently

large finite sets of primes of ki , i = 1, 2. Then the canonical map
Iso(Ok2 ,S2 , Ok1 ,S1 ) −→ OutIso(G(k1 (µ)S1 |k1 ), G(k2 (µ)S2 |k2 ))
is a bijection.
Finally, we should mention that these anabelian ideas are only part of a larger
program, initiated by A. GROTHENDIECK [69], which aims at a description of
the absolute Galois group of Q in geometric terms.


Literature
ALBERT, A.A.
[1] Structure of algebras. Am. Math. Soc. Providence 1964
AMICE, Y., FRESNEL, J.

[2] Fonctions zêta p-adiques des corps des nombres algébriques abéliens reéls. Acta Arith.
Warzawa 20 (1972) 353–384
ANDOŽSKII, I.V.
[3] Demuškin groups (in Russian). Mat. Zametki 14 (1973) No.1 121–126, English trans-
lation in Math. Notes 14 (1974) 626–628
ARTIN, E.
[4] Kennzeichnung des Körpers der reellen algebraischen Zahlen. Hamb. Abh. 3 (1924)
319–323
ARTIN, M., GROTHENDIECK, A., VERDIER, J.L.

[5] Théorie des Topos et Cohomologie Etale des Schémas (SGA 4), tome 1,2,3. Lecture
Notes Math. 269, 270, 305, Springer 1972–1973
ARTIN, E., TATE, J.

[6] Class Field Theory. Benjamin New York, Amsterdam 1967
ATIYAH, M.F., WALL, C.T.C.

[7] Cohomology of groups. Chapter IV in [23], 94–115
AX, J.
[8] Proof of some conjectures on cohomological dimension. Proc. Amer. Math. Soc. 16
(1965) 1214–1221
AX, J., KOCHEN, S.

[9] Diophantine problems over local fields I + II. Am. J. of Math. 87 (1965), III, Ann. of
Math. 83 (1966) 437–456
BAYER, P., NEUKIRCH, J.

[10] On values of zeta functions and `-adic Euler characteristics. Invent. Math. 50 (1978)
35–64
BIERI, B., ECKMANN, B.

[11] Groups with homological duality generalizing Poincaré duality. Invent. Math. 20 (1973)
103–124
BINZ, E., NEUKIRCH, J., WENZEL, G.

[12] A subgroup theorem for free products of profinite groups. J. of Algebra 19 (1971)
104–109

806 Literature
BLOCH, S., KATO, K.

[13] p-adic étale cohomology. Publ. Math. IHES 63 (1986) 107–152
BOURBAKI, N.
[14] Algèbre Chap.8. Hermann, Paris 1958
[15] General Topology Part 1,2. Hermann, Paris 1966
[16] Commutative Algebra. Hermann, Paris 1972
[17] Algebra I. Hermann, Paris 1974
BROWN, K.
[18] Cohomology of Groups. Springer 1982
BRUMER, A.
[19] Pseudocompact algebras, profinite groups and class formations. J. of Algebra 4 (1966)
442–470
[20] On the units of algebraic number fields. Mathematika 14 (1967) 121–124
CARTAN, E., EILENBERG, S.

[21] Homological Algebra. Princeton Math. Ser. 19, Princeton 1956
CASSELS, J.W.S.
[22] Global fields. Chapter II in [23], 42–84
CASSELS, J.W.S., FRÖHLICH, A. (ed.)

[23] Algebraic Number Theory. Academic Press, London New York 1967
CHANDRASEKHARAN, K.
[24] Introduction to Analytic Number Theory. Springer 1968
CHENEVIER, G.
[25] On Number Fields with given Ramification. Comp. Math. 143 (2007) 1359–1373
COATES, J.
[26] K-theory and Iwasawa’s analogue of the Jacobian. In Algebraic K-theory II, Lecture
Notes in Math. 342 Springer 1973, pp. 502–520
[27] p-adic L-functions for motives. In [28], pp. 141–172
COATES, J., TAYLOR, M.J. (ed.)

[28] L-Functions and Arithmetic. Proceedings of the Durham Symposium, July 1989, Cam-
bridge University Press 1991
COATES, J., WILES, A.

[29] On the conjecture of Birch and Swinnerton-Dyer. Invent. Math. 39 (1977) 223–251
COLMEZ, P.
[30] Résidu en s = 1 des fonctions zêta p-adiques. Invent. Math. 91 (1988) 371–389
CORNELL, G., SILVERMAN, J.H. (ed.)

[31] Arithmetic Geometry. Springer 1986

Literature 807
CURTIS, C.W., REINER, I.

[32] Methods of Representation Theory Vol I. Wiley-Interscience Publication New York,
Chichester, Brisbane, Toronto 1981
DE SHALIT, EHUD
[33] Iwasawa Theory of Elliptic Curves with Complex Multiplication. Perspectives in Math.
Vol.3 Academic Press Boston 1987
DELIGNE, P.
[34] Théorie de Hodge II. Publ. Math. IHES 40 (1971) 5–57
[35] La conjecture de Weil I. Publ. Math. IHES 43 (1974) 273–308
[36] La conjecture de Weil II. Publ. Math. IHES 52 (1981) 313–428
DELIGNE, P., RIBET, K.

[37] Values of abelian L-functions at negative integers over totally real fields. Invent. Math.
59 (1980) 227–286
DEMUŠKIN, S.P.
[38] The group of the maximal p-extension of a local field (in Russian). Izv. Akad. Nauk.
USSR Math. Ser. 25 (1961) 329–346
DIEKERT, V.
[39] Über die absolute Galoisgruppe dyadischer Zahlkörper. J. reine u. angew. Math. 350
(1984) 152–172
[40] Eine Bemerkung zu freien Moduln über regulären lokalen Ringen. J. of Algebra 101
(1986) 188–189
DIXON, J.D., DU SAUTOY, M. P. F., MANN, A., SEGAL, D.

[41] Analytic pro-p Groups. London Math. Soc. Series 157 Cambridge University Press
1991
DRAXL, P.K.
[42] Skew fields, London Math. Soc. Series 81, Cambridge University Press 1983
DRESS, A.
[43] Contributions to the theory of induced representations. Lect. Notes Math. 342, Springer
1973
DUMMIT, J., LABUTE, J.P.

[44] On a new characterization of Demuškin groups. Invent. Math. 73 (1983) 413–418
ENGLER, A., HARAN, D., KOCHLOUKOVA, D., ZALESSKII, P.A.

[45] Normal subgroups of profinite groups of finite cohomological dimension. J. Lond.
Math. Soc. 69 (2004) 317–332
EVENS, L.
[46] The Cohomology of Groups. Clarendon Press, Oxford 1991

808 Literature
FALTINGS, G.
[47] Endlichkeitssätze für abelsche Varietäten über Zahlkörpern. Invent. Math. 73 (1983)
349–366, Erratum 75 (1984) 381. English translation in [31], 9–27
[48] Curves and their fundamental groups (following Grothendieck, Tamagawa and
Mochizuki). Séminaire Bourbaki 50ème année, 1997–98, No. 840
FERRERO, B., WASHINGTON, L.

[49] The Iwasawa invariant µp vanishes for abelian number fields. Ann. of Math. 109 (1979)
377–395
FESENKO, I.B., VOSTOKOV, S.V.

[50] Local Fields and their Extensions. A Constructive Approach. AMS Providence, Rhode
Island 1993
FONTAINE, J-M., MAZUR, B.

[51] Geometric Galois representations. In: Elliptic Curves, Modular Forms, & Fermat’s Last
Theorem, edited by J. Coates, and S.T. Yau. International Press, Boston 1995
FRIED, M.D., HARAN,D.,VÖLKLEIN,H.

[52] Absolute Galois group of the totally real numbers. C.R. Acad. Sci., Paris 317 (1993)
995–999
FRIED, M.D., JARDEN, M.

[53] Field Arithmetic. Springer 1996
FRIEDLANDER, E.M.
[54] Etale Homotopy of Simplicial Schemes. Princeton University Press 1982
GEYER, W.-D.
[55] Galois groups of intersections of local fields. Israel J. of Math. 30 (1978) 382–396
GILDENHUYS, D., RIBES, L.

[56] A Kurosh subgroup theorem for free pro-c-products of pro-c-groups. Trans. Amer.
Math. Soc. 186 (1973) 309–329
GOLOD, E.S., ŠAFAREVIČ, I.R.

[57] On class field towers (in Russian). Izv. Akad. Nauk SSSR 28 (1964) 261–272. English
translation in Am. Math. Soc. Transl. (2) 48 (1965) 91–102
GORDEEV, N.L.
[58] Infinitude of the number of relations in the Galois group of the maximal p-extension
of a local field with restricted ramification. Izv. Akad. Nauk SSSR 45 (1981) 592–607
English translation in Math. USSR, Izv. 18 (1982) 513–524
GREEN, J.A.
[59] Axiomatic representation theory of finite groups. J. of Pure and Applied Algebra 1
(1971) 41–77
GREENBERG, M.J.
[60] Lectures on Forms in Many Variables. Benjamin 1969

Literature 809
GREENBERG, R.
[61] On a certain `-adic representation. Invent. Math. 21 (1973) 117–124
[62] On the Iwasawa invariants of totally real number fields. Amer. J. Math. 93 (1976)
263–284
[63] On p-adic L-functions and cyclotomic fields II. Nagoya Math. Journ. 67 (1977) 139–158
[64] On p-adic Artin L-functions. Nagoya Math. Journ. 89 (1983) 77–87
[65] Iwasawa theory for motives. In [28], pp. 211–233
GROTHENDIECK, A.
[66] Sur quelques points d’algèbre homologique. Tôhoku Math. J. 9 (1957) 119–221
[67] Revêtements Étales et Groupe de Fondamental (SGA1). Lecture Notes in Math. 224,
Springer 1971
[68] Brief an G. Faltings (1983). Reprinted in [205] 49–58. English translation: loc. cit.
285–293
[69] Esquisse d’un Programme. In [205], 5–48. English translation: loc. cit. 243–284
GROTHENDIECK, A., DIEUDONNÉ, J.A.

[70] Eléments de Géométrie Algébrique I (EGA 1). Springer Grundlehren Bd. 166, 1971
GRUENBERG, K.W.
[71] Projective profinite groups. J. London Math. Soc. 42 (1967) 155–165
HABERLAND, K.
[72] Galois Cohomology of Algebraic Number Fields. Deutscher Verlag der Wiss., Berlin
1978
HALL, M.
[73] The Theory of Groups. Macmillan Company, New York 1968
HARAN, D.
[74] On closed subgroups of free products of profinite groups. Proc. London Math. Soc. (3)
55 (1987) 266–298
[75] A proof of Serre’s theorem. J. of Indian Math. Soc. 55 (1990) 213–234
HARARI, D., SZAMUELY, T.

[76] Arithmetic duality theorems for 1-motives. J. reine u. angew. Math. 578 (2005) 93–128
HARTSHORNE, R.
[77] Algebraic Geometry. Springer 1977
HENNIART, G.
[78] Relèvement global d’extensions locales: quelques problèmes de plongement. Math.
Ann. 319 (2001) 75–87
HILTON, P.J., STAMMBACH, U.

[79] A Course in Homological Algebra. Springer 1971
HOECHSMANN, K.
[80] Zum Einbettungsproblem. J. reine u. angew. Math. 229 (1968) 81–106

810 Literature
HUPPERT, B.
[81] Endliche Gruppen I. Springer 1967
IHARA, Y.
[82] How many primes decompose completely in an infinite unramified Galois extension of
a global field? J. Math. Soc. Japan 35 (1983) 693–709
IKEDA, M.
[83] Zur Existenz eigentlicher galoisscher Körper beim Einbettungsproblem. Hamb. Abh.
24 (1960) 126–131
[84] Completeness of the absolute Galois group of the rational number field. J. reine u.
angew. Math. 291 (1977) 1–22
IŠHANOV, V.V., LUR’E, B.B., FADDEEV, D.K.

[85] The Embedding Problem in Galois Theory. Trans. of Math. Monographs 165 AMS
Providence 1997
IWASAWA, K.
[86] On solvable extensions of algebraic number fields. Ann. of Math. 58 (1953) 548–572
[87] On Galois groups of local fields. Trans. AMS 80 (1955) 448–469
[88] On ZZ` -extensions of algebraic number fields. Ann. of Math. 98 (1973) 246–326
[89] On the µ-invariant of ZZ` -extensions. Conf. on number theory, algebraic geometry, and
commutative algebra in honor of Y. Akizuki, Tokyo 1977, 1–11
[90] Riemann-Hurwitz formula and p-adic Galois representations for number fields. Tohoku
Math. J. 33 (1981) 263–288
JACOBSON, N.
[91] Finite-dimensional Division Algebras over Fields. Springer 1996
JAKOVLEV, A.S.
[92] The Galois group of the algebraic closure of a local field. Math. USSR-Izv. 2 (1968)
1231–1269
[93] Remarks on my paper “The Galois group of the algebraic closure of a local field". Math.
USSR-Izv. 12 (1978) 205–206
JANNSEN, U.
[94] Über Galoisgruppen lokaler Körper. Invent. Math. 70 (1982) 53–69
[95] Galoismoduln mit Hasse-Prinzip. J. reine u. angew. Math. 337 (1982) 154–158
[96] Continuous étale cohomology. Math. Ann. 280 (1988) 207–245
[97] Iwasawa modules up to isomorphism. Advanced Studies in Pure Mathematics 17 (1989)
171–207
[98] The splitting of the Hochschild-Serre spectral sequence for a product of groups. Canad.
Math. Bull. 33 (1990) 181–183
JANNSEN, U., WINGBERG, K.

[99] Die Struktur der absoluten Galoisgruppe p-adischer Zahlkörper. Invent. Math. 70 (1982)
71–98
KATO, K.
[100] Iwasawa theory and p-adic Hodge theory. Kodai Math. J. 16 No.1 (1993) 1–31

Literature 811
KAWADA, Y.
[101] Class formations. Proc. Symp. Pure Math. Vol. XX, AMS, Providence (1969) 96–114
KERSTEN, I.
[102] Brauergruppen von Körpern. Vieweg 1990
KIDA, Y.
[103] `-extensions of CM-fields and cyclotomic invariants. J. of Number Theory 12 (1980)
519–528
KISILEVSKY, H., LABUTE, J.P.

[104] On a sufficient condition for the p-class tower of a CM-field to be infinite. In: Proceed-
ings of the Int. Number Conf. Laval 1987, Berlin 1989
KISIN, M., WORTMANN, S.

[105] A note on Artin motives. Math. Res. Lett. 10 (2003) 375–389
KNUS, M.-A.
[106] Quadratic and Hermitian Forms over Rings. Springer 1980
KOCH, H.
[107] `-Erweiterungen mit vorgegebenen Verzweigungsstellen. J. reine und angew. Math. 219
(1965) 30–61
[108] Über die Galoissche Gruppe der algebraischen Abschließung eines Potenzreihenkörpers
mit endlichem Konstantenkörper. Math. Nachr. 35 (1967) 323–327
[109] Zum Satz von Golod-Schafarewitsch. Math. Nachr. 42 (1969) 321–333
[110] Galoissche Theorie der p-Erweiterungen. Deutscher Verlag der Wiss., 1970, Springer
1970 (Russian translation Moscow 1973)
[111] The Galois group of a p-closed extension of a local field. Soviet. Math. Dokl. 19 (1978)
10–13
KOCH, H., VENKOV, B.B.

[112] Über den p-Klassenkörperturm eines imaginär-quadratischen Zahlkörpers. Soc. Math.
France, Astèrisque 24–25 (1975) 57–67
KOLSTER, M.
[113] A relation between the 2-primary parts of the main conjecture and the Birch-Tate
conjecture. Can. Math. Bull. 32 No.2 (1989) 248–251
KRAFT, J.S., SCHOOF, R.

[114] Computing Iwasawa modules of real quadratic number fields. Comp. Math. 97 (1995)
135–155
KUZ’MIN, L.V.
[115] Local extensions associated with `-extensions with given ramification. Izv. Akad. Nauk
SSSR 39 (1975) No. 4. English translation in Math. USSR Izv. 9 (1975) No. 4 693–726

812 Literature
LABUTE, J.P.
[116] Classification des groupes de Demuškin. C.R. Acad. Sci. Paris 260 (1965) 1043–1046
[117] Classification of Demuškin groups. Can. J. Math. 19 (1967) 106–132
[118] Algèbres de Lie et pro-p-groupes definis par une seule relation. Invent. Math. 4 (1967)
142–158
[119] Mild pro-p-groups and Galois groups of p-extensions of Q. J. reine u. angew. Math.
596 (2006) 155–182
LANDAU, E.
[120] Einführung in die elementare und analytische Theorie der algebraischen Zahlen und
der Ideale. Chelsea Publ. New York 1949
LANG, S.
[121] On quasi-algebraic closure. Ann. of Math. 55 (1952) 373–390
[122] Cyclotomic Fields I and II. Springer 1990
[123] Algebraic Number Theory. Springer 1994
[124] Rapport sur la cohomologie des groupes. Benjamin, New York-Amsterdam 1966, Eng-
lish translation: Topics in Cohomology of Groups. Springer LNM 1625, 1996
LAZARD, M.
[125] Sur les groupes nilpotents et les anneaux de Lie. Ann. Ec. Norm. Sup. 71 (1954)
101–190
[126] Groupes analytiques p-adiques. Publ. Math. I.H.E.S. 26 (1965) 389–603
LUBOTZKY, A.
[127] Group representations, p-adic analytic groups and lattices in SL2 C. Ann. of Math. 118
(1983) 115–130
MACLANE, S.
[128] Homology. Springer 1967
MATSUMURA, H.
[129] Commutative Ring Theory. Cambridge University Press 1986
MAUS, E.
[130] Über die Verteilung der Grundverzweigungszahlen von wild verzweigten Erweiterun-
gen p-adischer Zahlkörper. J. reine u. angew. Math. 257 (1972) 47–79
[131] Relationen in Verzweigungsgruppen. J. reine u. angew. Math. 258 (1973) 23–50
MAZUR, B.
[132] Rational points of abelian varieties with values in towers of number fields. Invent. Math.
18 (1972) 183–266
MAZUR, B., TATE, J., TEITELBAUM, J.

[133] On p-adic analogues of the conjectures of Birch and Swinnerton-Dyer. Invent. Math.
84 (1986) 1–48
MAZUR, B., WILES, A.

[134] Class fields of abelian extensions of Q. Invent. Math. 76 (1984) 179–330

Literature 813
MELNIKOV, O.V.
[135] Subgroups and homology of free products of profinite groups (in Russian). Izv. Akad.
Nauk SSSR 53 (1989) No. 1. English translation in Math. USSR Izv. 34 (1990) No. 1
97–119
MELNIKOV, O.V., SHAROMET, A.A.

[136] The Galois groups of a multidimensional local field of positive characteristic (in Rus-
sian). Math. Sborn. 180 (1989), No. 8 English translation in Math. USSR Sborn. 67
(1990), No. 2 595–610
MERKUR’EV, A.S.
[137] K2 of fields and the Brauer group. Contemporary Math. 55 Part 2, AMS 1986
MERKUR’EV, A.S., SUSLIN, A.A.

[138] K-cohomology of Severi-Brauer varieties and the norm residue homomorphism. Izv.
Akad. Nauk. SSSR 46 (1982) 1011–1046. English translation in Math. USSR Izv. 21
(1983) 307–340
MILNE, J.S.
[139] Etale Cohomology. Princeton Univ. Press 1980
[140] Abelian varieties. Chapter V in [31], 103–150
[141] Jacobian varieties. Chapter VII in [31], 167–212
MILNOR, J.
[142] Algebraic K-theory and quadratic forms. Invent. Math. 9 (1969/70) 318–344
MOCHIZUKI, S.
[143] The profinite Grothendieck conjecture for closed hyperbolic curves over number fields.
J. Math. Sci. Tokyo 3 (1996) 571–627
[144] A version of the Grothendieck conjecture for p-adic local fields. Int. J. Math. 8 (1997)
499–506
[145] The Local Pro-p Anabelian Geometry of Curves. Invent. Math. 138 (1999) 319–423
MORRIS, S.A.
[146] Pontryagin Duality and the Structure of Locally Compact Abelian Groups. LMS Lecture
Notes Series 29, Cambridge University Press 1977
MOSER, TH.
[147] G-Modulationen und Reziprozität. Diplomarbeit, Regensburg 1992
MUKHAMEDOV, V.G.
[148] Local extensions associated with the `-extensions of number fields with restricted
ramification (in Russian). Mat. Zametki 35 (1984) No.4 481–490, English translation
in Math. Notes 35 No.3–4 253–258
NAKAMURA, H.
[149] Rigidity of arithmetic fundamental group of a punctured projective line. J. reine u.
angew. Math. 405 (1990) 117–130

814 Literature
NEEMAN, A.
[150] A counterexample to a 1961 “theorem” in homological algebra. Invent. Math. 148
(2002) 397–420
NEUKIRCH, J.
[151] Klassenkörpertheorie. Bibliographisches Institut, Mannheim 1969
[152] Kennzeichnung der p-adischen und der algebraischen Zahlkörper. Invent. Math. 6
(1969) 296–314
[153] Kennzeichnung der endlich-algebraischen Zahlkörper durch die Galoisgruppe der maxi-
malen auflösbaren Erweiterungen. J. reine u. angew. Math. 238 (1970) 135–147
[154] Freie Produkte pro-endlicher Gruppen und ihre Kohomologie. Archiv der Math. 22
(1971) 337–357
[155] Einbettungsprobleme mit lokaler Vorgabe und freie Produkte lokaler Galoisgruppen. J.
reine u. angew. Math. 259 (1973) 1–47
[156] Über das Einbettungsproblem in der algebraischen Zahlentheorie. Invent. Math. 21
(1973) 59–116
[157] Über die absoluten Galoisgruppen algebraischer Zahlkörper. Soc.Math. de France,
Astérisque 41–42 (1977) 67–79
[158] On solvable number fields. Invent. Math. 53 (1979) 135–164
[159] Class Field Theory. Springer 1986
[160] Algebraische Zahlentheorie. Springer 1992, English translation: Algebraic Number
Theory. Springer 1999
[161] Micro primes. Math. Ann. 298 (1994) 629–666
NEUMANN, O.
[162] On p-closed algebraic number fields with restricted ramification. Math. USSR Izv. 9
(1975) No.2 243–254
[163] On p-closed number fields and an analogue of Riemann’s existence theorem. In:
A.Fröhlich (ed.): Algebraic Number Fields. Academic Press London 1977 625–647
ODA, T.
[164] A note on ramification of the Galois representation on the fundamental group of an
algebraic curve II. J. Number Theory 53 (1995) 342–355
ORE, O.
[165] Contributions to the theory of groups of finite order. Duke Math. J. 5 (1939) 431–460
PATTERSON, S.J.
[166] An introduction to the theory of the Riemann Zeta-Function. Cambridge University
Press 1989
PLETCH, A.
[167] Profinite duality groups I and II. J. Pure Applied Algebra 16 (1980) 55–74 and 285–297
POITOU, G.
[168] Remarques sur l’homologie des groupes profinis. Colloques Int. CNRS 143 (1966)
201–213
[169] Cohomologie galoisienne des modules finis. Dunod, Paris 1967

Literature 815
PONTRYAGIN, L.S.
[170] Topological Groups. Gordon and Breach, New York, London, Paris 1966
POP, F.
[171] Étale Galois covers of affine smooth curves. Invent. Math. 120 (1995) 555–578
[172] On Grothendieck’s conjecture of birational anabelian geometry. Ann. of Math. 138
(1994) 145–182
[173] On Grothendieck’s conjecture of birational anabelian geometry II. Preprint 1994
[174] Embedding problems over large fields. Ann. of Math. 144 (1996) 1–34
[175] Alterations and birational anabelian geometry. Resolution of singularities (Obergurgl,
1997), Progr. Math. 181, Birkhäuser Basel 2000 519–532
RAYNAUD, M.
[176] Section des fibrés vectoriels sur une courbe. Bull. Soc. Math. France 110 (1982)
103–125
REICHARDT, H.
[177] Konstruktion von Zahlkörpern mit gegebener Galoisgruppe von Primzahlpotenzord-
nung. J. reine u. angew. Math. 177 (1937) 1–5
RIBES, L.
[178] On a cohomology theory of pairs of groups. Proc. AMS 21 (1969) 230–234
[179] Amalgamated products of profinite groups. Math. Zeitschrift 123 (1971) 357–364
RIBET, K.
[180] A modular construction of unramified p-extensions of Q(µp ). Invent. Math. 34 (1976)
151–162
RIEHM, C.
[181] The corestriction of algebraic structures. Invent. Math. 11 (1970) 73–198
ROOS, J.-E.
[182] Sur les foncteurs dérivés de lim , Applications. C. R. Acad. Sci. Ser. I 252 (1961)
←−
3702–3704
ROQUETTE, P.
[183] On class field towers. Chapter IX in [23], 231–249
RUBIN, K.
[184] The main conjecture. Appendix to [122]
[185] The "main conjectures" of Iwasawa theory for imaginary quadratic fields. Invent. Math.
103 (1991) 25–68
RUDIN, W.
[186] Fourier Analysis on Groups. Interscience Publishers 1967

816 Literature
ŠAFAREVIČ, I.R.
[187] On the construction of fields with a given Galois group of order `α (in Russian). Izv.
Akad. Nauk SSSR 18 (1954) 261–296, English translation in Amer. Math. Soc. Transl.
4 (1956) 107–142
[188] On an existence theorem in the theory of algebraic numbers (in Russian). Izv. Akad.
Nauk. SSSR 18 (1954) 327–334, English translation in Amer. Math. Soc. Transl. 4
(1956) 143–150
[189] On the problem of imbedding fields (in Russian). Izv. Akad. Nauk SSSR 18 (1954)
389–418, English translation in Amer. Math. Soc. Transl. 4 (1956) 151–183
[190] Construction of fields of algebraic numbers with given solvable groups (in Russian).
Izv. Akad. Nauk. SSSR 18 (1954) 525–578, English translation in Amer. Math. Soc.
Transl. 4 (1956) 185–237
[191] Extensions with given ramification points (in Russian). Publ. Math. IHES 18 (1963)
295–319, English translation in Amer. Math. Soc. Transl. 59 (1966) 128–149
[192] Factors of a decreasing central series (in Russian). Mat. Zametki 45 (1989) 114–117,
English translation in Math. Notes 45 (1989) 262–264
SCHMIDT, A.
[193] Extensions with restricted ramification and duality for arithmetic schemes. Comp. Math.
100 (1996) 233–245
[194] Bounded defect in partial Euler characteristics. Bull. London Math. Soc. 28 (1996)
463–464
[195] An arithmetic site for the ring of integers of algebraic number fields. Invent. Math. 123
(1996) 575–610
[196] On Poitou’s duality theorem. J. reine u. angew. Math. 517 (1999) 145–160
[197] On the relation between 2 and ∞ in Galois cohomology of number fields. Comp. Math.
133 (2002) 267–288
[198] Circular sets of prime numbers and p-extensions of the rationals. J. reine u. angew.
Math. 596 (2006) 115–130
[199] Rings of integers of type K(π, 1). Doc. Math. 12 (2007) 441–471
SCHMIDT, F.K.
[200] Mehrfach perfekte Körper. Math. Ann. 108 (1933) 1–25
SCHNEIDER, P.
[201] Über gewisse Galoiscohomologiegruppen. Math. Z. 168 (1979) 181–205
[202] Die Galoiscohomologie p-adischer Darstellungen über Zahlkörpern. Dissertation Re-
gensburg 1980
[203] p-adic height pairings I and II. Invent. Math. 69 (1982) 401–409 and Invent. Math. 79
(1985) 329–374
[204] Motivic Iwasawa theory. Advanced Studies in Pure Mathematics 17 (1989) 421–456
SCHNEPS, L., LOCHAK, P. (ed.)

[205] Geometric Galois Actions I. London Math. Soc. Lect. Notes 242 Cambridge 1997
SCHOLZ, A.
[206] Konstruktion algebraischer Zahlkörper mit beliebiger Gruppe von Primzahlpotenzord-
nung. Math. Z. 42 (1936) 161–188

Literature 817
SCHOOF, R.
[207] Infinite Class Field Towers of Quadratic Fields. J. reine u. angew. Math. 372 (1986)
209–220
SERRE, J.-P.
[208] Groupes algébriques et corps de classes. Hermann, Paris 1959
[209] Structure de certains pro-p-groupes. Séminaire Bourbaki (1962/63) Exp. 252
[210] Cohomologie Galoisienne. Lecture Notes in Mathematics 5, Springer 1964 (Cinquième
édition 1994)
[211] Sur la dimension cohomologique des groupes profinis. Topology 3 (1965) 413–420
[212] Corps Locaux. Hermann, Paris 1968. English translation: Local Fields. Springer 1995
[213] A Course in Arithmetic. Springer 1973
[214] Linear Representations of Finite Groups. Springer 1977
[215] Trees, Springer 1980
[216] Topics in Galois Theory. Jones and Bartlett Publ. Boston 1992
SINNOTT, W.
[217] On the Stickelberger ideal and the circular units of abelian number fields. Invent. Math.
62 (1980) 181–234
[218] On the µ-invariant of the Γ -transform of a rational function. Invent. Math. 75 (1984)
273–282
SOULÉ, C.
[219] K-théorie des anneaux d’entiers de corps de nombres et cohomologie étale. Invent.
Math. 55 (1979) 251–295
STAMMBACH, U.
[220] Cohomological characterisations of finite solvable and nilpotent groups. J. Pure and
Appl. Alg. 11 (1977) 293-301
STIX, J.
[221] Affine anabelian curves in positive characteristic. Compositio Math. 134 (2002) 75–85
[222] Projective anabelian curves in positive characteristic and descent theory for log-étale
covers. Dissertation, Bonner Mathematische Schriften 354 (2002)
SUSLIN, A.A.
[223] Algebraic K-theory and the norm residue homomorphism. J. Soviet. Math. 30 (1985)
2556–2611
SZAMUELY, T.
[224] Groupes de Galois de corps de type fini (d’après Pop). Astérisque No. 294 (2004), ix,
403–431
TAKAHASHI, T.
[225] Galois cohomology of unramified extensions of algebraic function fields. Tôhoku Math.
J. 24 (1972) 33–39
[226] Galois cohomology of finitely generated modules. Tôhoku Math. J. 21 (1969) 102–111
TAMAGAWA, A.
[227] The Grothendieck conjecture for affine curves. Comp. Math. 109 (1997) 135–194

818 Literature
TAMME, G.
[228] Introduction to Étale Cohomology. Springer 1994
TATE, J.
[229] Duality theorems in Galois cohomology over number fields. Proc. Int. Congress Stock-
holm 1962
[230] Letter to Serre. Annexe to Chap.I in [210]
[231] Symbols in arithmetic. Proc. Int. Congress Nice 1970, Vol.1, 201–211
[232] Relations between K2 and Galois cohomology. Invent. Math. 36 (1976) 257–274
TAYA, H.
[233] Computation of ZZ3 -invariants of real quadratic fields. Math. Comp. 65 (1996) 779–784
TERJANIAN, G.
[234] Un contre-exemple à une conjecture d’Artin. C. R. Acad. Sci. Paris 262 (1966) p. 612
THÉVENAZ, J. and WEBB, P.

[235] The structure of Mackey functors. Trans. AMS. 347 (1995) 1865–1961
UCHIDA, K.
[236] On Tate’s duality theorems in Galois cohomology. Tôhoku Math. J. 21, No. 4 (1969)
92–101
[237] Isomorphisms of Galois groups. J. Math. Soc. Japan 28 (1976) 617–620
[238] Isomorphisms of Galois groups of algebraic function fields. Ann. of Math. 106 (1977)
589–598
VAN DER KALLEN, W.

[239] The Merkurjev-Suslin theorem. In I. Reiner, K. W. Roggenkamp (ed.): Orders and
their Applications. Lecture Notes in Math. 1142, Springer 1985, pp. 157–168
VERDIER, J.L.
[240] Dualité dans la cohomologie des groupes profinis. In [210].
VINBERG, E.B.
[241] On the dimension theorem for associative algebras (in Russian) Izv. Akad. Nauk. SSSR
29 (1965) 209–214
VOEVODSKY, V.
[242] Motivic cohomology with ZZ/2-coefficients. Publ. Math. Inst. Hautes Études Sci. 98
(2003) 59–104
[243] Reduced power operations in motivic cohomology. Publ. Math. Inst. Hautes Études Sci.
98 (2003) 1–57
WALDSCHMIDT, M.
[244] Transcendance et exponentielles en plusieurs variables. Invent. Math. 63 (1981) 97–127
WASHINGTON, L.C.
[245] Class numbers and ZZp -extensions. Math. Ann. 214 (1975) 177–193
[246] Introduction to Cyclotomic Fields. Springer 1982 (Second ed. 1997)

Literature 819
WEIGEL, TH.,ZALESSKII, P.A.

[247] Profinite groups of finite cohomological dimension. C. R. Acad. Sci. Paris 338 (2004)
353–358
WIESE, G.
[248] Zur Fontaine-Mazur-Vermutung für Erweiterungen mit beschränkter Verzweigung von
CM-Körpern. Diplomarbeit, Heidelberg 2002
WILES, A.
[249] The Iwasawa conjecture for totally real fields. Ann. of Math. 131 (1990) 493–540
WINGBERG, K.
[250] Der Eindeutigkeitssatz für Demuškin-Formationen. Invent. Math. 70 (1982) 19–113
[251] Ein Analogon zur Fundamentalgruppe einer Riemannschen Fläche im Zahlkörperfall.
Invent. Math. 77 (1984) 557–584
[252] Duality theorems for Γ -extensions of algebraic number fields. Comp. Math. 55 (1985)
333–381
[253] On Poincaré groups. J. London Math. Soc. 33 (1986) 271–278
[254] On Galois groups of p-closed algebraic number fields with restricted ramification. J.
reine u. angew. Math. 400 (1989) 185–202
[255] On Demuškin groups with involution. Ann. Sci. Éc. Norm. Sup. 22 (1989) 555–567
[256] On the maximal unramified extension of an algebraic number field. J. reine u. angew.
Math. 440 (1993) 129–156
[257] On the Fontaine-Mazur conjecture for CM-fields. Comp. Math. 131 (2002) 341–354
[258] Free Product Decomposition of Galois Groups of Number Fields. Publ. RIMS, Kyoto
Univ. 41 (2005) 295–306
WINTENBERGER, J.-P.
[259] Le corps des normes de certaines extensions infinies de corps locaux; applications.
Ann. Sci. Éc. Norm. Sup. 16 (1983) 59–89
WITT, E.
[260] Treue Darstellung Liescher Ringe. J. reine u. angew. Math. 177 (1937) 152–160
ZALESSKII, P.A.
[261] Open subgroups of free profinite products. Contemp. Math. 131 (1992) 473–492
Additional References
ARKHIPOV, G.I., KARATSUBA, A.A.

[AK] Local representation of zero by a form. (Russian) Izv. Akad. Nauk SSSR Ser. Mat. 45
(1981) 948–961. English translation in Math. USSR, Izv. 19 (1982) 231–240
CHENEVIER, G., CLOZEL, L.
[CC] Corps de nombres peu ramifiés et formes automorphes autoduales. J. Amer. Math. Soc.
22 (2009), no. 2, 467–519

820 Literature
COLLIOT-THÉLÈNE, J.-L., SANSUC, J.-J.

[CS] Principal homogeneous spaces under flasque tori: applications. J. Algebra 106 (1987),
no. 1, 148–205
MERKUR’EV, A.S., SUSLIN, A.A.

[MS] Norm residue homomorphism of degree three. Izv. Akad. Nauk SSSR Ser. Mat. 54
(1990), no. 2, 339–356; translation in Math. USSR Izv. 36 (1991), no. 2, 349–367
ROST, M.
[Ro] Hilbert 90 for K3 for degree-two extensions. Preprint 1986 (available on M. Rost’s
homepage)
SCHMIDT, A., WINGBERG, K.

[SW] Extensions of profinite duality groups. arXiv:0809.4371 (2008)
VOEVODSKY, V.
[Vo] On motivic cohomology with Z/l-coeffcients. Ann. of Math. 174 (2011), 401–438

Index
abelian torsion group — inhomogeneous 14

— order of 7 — normalized 24
acyclic 31, 33 cochain 13
— F -acyclic 132 — complex 13
— resolution 33 — homogeneous 13
affine transformation 181 — inhomogeneous 14
algebra — normalized 24
— central simple 349 cochain complex 101
— crossed product 350 cocycle 13
— cyclic 349 — homogeneous 13
— splitting field of 349 — inhomogeneous 14
algebraic tori 381 — non-abelian 15
amalgamated free pro-c-product 251 — normalized 24
arithmetically profinite extension 789 coeffaceable 133
Artin-Schreier theory 340 cofixed module 22, 83, 133
ascending central series 229 cohomological dimension 120, 171, 365
augmentation — P -dimension 184
— ideal 33, 274 — p-dimension 172
— map 33, 75, 274 — strict 172
cohomologically trivial 31, 80, 150
BS (k, m) 531
— resolution 33
basis of a free pro-c-group 198
cohomology
Bloch-Kato conjecture 358
— of the S-idèle class group 457
Bockstein homomorphism 236
— of the S-idèles 456
Brauer group 349
— of the S-units 458
— corestriction 352
cohomology group 13, 57, 101
Brauer-Severi variety 347
— modified 22
bundle of pro-c-groups 257
— non-abelian 15
c-torsion subgroup 190 — of a pair 74
Ci -field 361 cohomology sequence, exact 25
cs(Ω | k) 524 — relative 74
canonical class 202 coinflation 84
capitulation of ideals 730 coinvariants 133
Čebotarev’s density theorem 524 collection of local conditions 511
class field axiom 165 compact induction 61, 737
— global 426 compact module 134
— local 371 — homology of 134
class field theory 152, 153, 162 compact-open topology 9
class field tower 697 complete acyclic resolution 34
class field tower problem 227 complete group algebra 274
class formation 156 complete standard resolution 22
class module 149 complete tensor product 276
CM-field 629, 751 complex 12, 101
coboundary 13 — bounded 101
— homogeneous 13 — cochain complex 13

822 Index
— double complex 103 — tensor product 104

— total complex 103 — total complex 103
complex vector bundle 59 double coset formula 51, 55, 59
conjugation 46, 61 dual G-modulation 59
connected component 8 duality
connecting homomorphism 27 — Poitou 160
constructible topology 611 — Poitou-Tate 485, 713
continuous cochain cohomology 137, 280 — Tate 380, 383
contracting homotopy 13 duality group 186, 210
corestriction 47, 63, 178 dualizing module 182
crossed homomorphism 15 — at P 185
cup-product 37 — at p 185, 210
cyclic groups, cohomology of 74
cyclotomic edge morphism 99
— ZZp -extension 637, 722 — of the Hochschild-Serre spectral se-
— ẐZ-extension 431 quence 119
cyclotomic character 397 — of the Tate spectral sequence 124, 125
— p-part 750 effaceable 128
embedding problem 189
cyclotomic polynomial 294
— equivalent solutions 189
cyclotomic units 777
— proper solution 189
— solution 189
δ-functor 28
Euler-Poincaré characteristic 179
— exact 28
— global 495, 505
— homological 133
— local 391
— universal 128
— partial 180
δ-homomorphism 27
exact δ-functor 28
defining relations 226
exact hexagon 75
deflation map 85
exact sequence
degenerate 691
— five term 67
Demuškin group 231
extension 17
derivation 321
— group extension 17
derived functor 127
— left 133 factor system 17
— right 129 field
descending q-central series 220 — Ci -field 361
descending central series 220 — of formal Laurent series 371
descent datum 348 — of norms 789
descent theory 348 — quasi-algebraically closed 361
differential 98, 101, 105 filtration 97
dimension shifting 32 — biregular 101
dimension, cohomological 120, 171, 365 — by subcomplexes 101
— P -dimension 184 — canonical 102
— p-dimension 172 — displaced 108
— strict 172 — finite 97
diophantine dimension 361 — stupid 106
Dirichlet density 524 — trivial 101
displaced filtration 108 finite number field 642
division lemma 289 first quadrant spectral sequence 99
double complex 103 Fitting subgroup 595
— natural filtration 104 five term exact sequence 67

Index 823
flasque 31 Greenberg conjecture 728, 755, 782

Fontaine-Mazur conjecture 705 group extension 17
formation module 156 group ring 31
Frattini argument 225 Grunwald-Wang theorem 541
Frattini subgroup 194, 224, 595, 624
free Hasse principle 437, 523
— pro-p-group 199, 226 Hasse-Witt invariant 602
— pro-c-Γ operator group 260 Herbrand quotient 76
— pro-c-group 198 Herbrand’s theorem 782
basis of 198 hexagon 75
free generators 198 Hilbert symbol 356
rank of 198 Hilbert’s Satz 90 344
free differential calculus 321 Hochschild-Serre spectral sequence 111,
free product of 114, 119, 131, 132
— full inertia groups 612 homogeneous
— inertia groups 611 — coboundary 13
— pro-c-groups 245, 259 — cochain 13
— pro-p-groups 250 — cocycle 13
homogeneous cochain complex
Frobenius lift 789
— continuous 136
Frobenius reciprocity 63
homology group 134, 136
Frobenius weight 613
— modified 84
full class 189
homology of finite group 84
fundamental G-modulation 57, 165, 205
homotopic 301
fundamental class 149, 202, 204
— to zero 301
homotopy 13, 36
G-group 16
— category of Λ-modules 301
G-modulation 54
— contracting 13
— completion 59
— equivalence 301
— dual 59
— equivalent 301
— fundamental 57, 165, 205
— representation ring 57 idèle group 425
G-module 12 idèle class group 425
— abstract 7 ideal class group
— acyclic 31 — narrow sense 620, 661
— cohomologically trivial 31 index 7
— compact 61, 134 induced module 31, 61
— dualizing 182, 185 — compact induction 61
— flasque 31 inflation 47
— induced 31, 61 inhomogeneous
— simple 70 — coboundary 14
— topological 7 — cochain 14
— trivial 8 — cocycle 14
— welk 31 initial terms of a spectral sequence 98
G-set 16 injectives, sufficiently many 130
GS -module 473 invariant map 150, 156
— dual 473, 475 — local 375
Galois symbol 358 Iwasawa algebra 291
generator system 224 Iwasawa module 291
Golod-Šafarevič inequality 228 — λ-invariant 292
Gras conjecture 778 — µ-invariant 292

824 Index
— Λ-rank 292 modified cohomology group 22

— adjoint 315 modified homology group 84
— characteristic polynomial 292 modulation 54
— elementary 292 module
— finitely presented 287
Jacobian variety 602 — free compact 275
— of coinvariants 274
Krasner’s lemma 429, 786 — of differential forms 322
Kronecker field 561 — profinite 141
Krull-Schmidt theorem 327 — pseudo-null 269
Kummer pairing 751 — reflexive 268
Kummer sequence 453 — unramified 386
Kummer theory 344
Kurosh subgroup theorem 252 n-form 360
Künneth-formula 111 IN(S) 78
Nakayama lemma
left derived functor 133 — for complete group rings 285
left exact functor 129 — topological 288
Leopoldt conjecture 627 Nakayama map 152
— defect 629 Nakayama-Tate, theorem of 151
— weak 638 norm form 361
Leopoldt’s Spiegelungssatz 763, 781 norm residue group 21
level-compact 91, 160 norm residue symbol 153, 157, 385, 441
Lie algebra of a pro-p-group 221 normalized cochains 24
limit terms of a spectral sequence 98 normic form 361
local field 371 number field of CM-type 629, 751
— orientable 769
localization map 481 one-relator pro-p-group 231
long exact cohomology sequence 27 operator group
Lyndon, theorem of 324 — free pro-c-Γ operator group 260
— pro-c-Γ operator group 260
Mackey functor 54, 58 order 7
main conjecture 721 orthogonal group 346
Maschke’s theorem 136 outer isomorphism 792
— generalized 288
maximal p-extension 338 p-adic complex numbers 625
— of a global field 554 p-adic local field 356, 371, 380, 401, 416
— of a local field 414 p-class field tower 697
maximal CM-field 760, 764 p-closed field 338
maximal pro-c-factor group 190 p-divisible group 602
maximal pro-p-factor group 190 — height 602
maximal tamely ramified extension 410 p-primary part 69
maximal unramified extension p-projective 191
— global 697, 723 p-rank of a curve 602
— local 373, 386, 409 p-(S, T )-closed 642
maximal c-extension 553 p-Sylow embedding problem 196
Mayer-Vietoris sequence 251 p-Sylow group 69
Milnor K-group 356 pairing
Milnor conjecture 358 — non-degenerate 217
Minkowski unit 635 — perfect 485
Mittag-Leffler property 138 periodicity for cyclic groups 74

Index 825
Poincaré group 231 reciprocity law

Poincaré group at p 210 — global 441, 457
Poincaré polynomial 229 — local 379
Poitou duality theorem 160 reduced degree 289
Poitou-Tate duality 485, 713 reduced norm 369
Poitou-Tate theorem 489 regulator matrix 626
Pontryagin dual 9 regulator, p-adic 627
Pontryagin duality 9 relation module 324
positively ramified relation rank 227
— at p 764 relation system 226
— extension 769 relative cohomology sequence 74
— maximal p-extension 764 representation ring 57
powerful global field 558 resolution 13
presentation of a module 287 — acyclic 33
principal homogeneous space 196 — cohomologically trivial 33
pro-p-group 68, 190 — complete 34
— free 199 — complete standard 22
— free pro-p-product 250 — injective 130
pro-c-Γ operator group 260 — standard resolution 13
pro-c-group 189 restricted product 10
— (solv)-projective 198 restriction 47, 58
— Γ -operator group 260 Riemann-Hurwitz formula 758
— c-projective 191 right derived functor 129
— amalgamated product 251
— bundle of 257 S-divisible 78
— free 198 S-idèle class group 454
— free pro-c-product 245, 259 S-idèle group 454
— rank of 199 S-ideal class group 452
procyclic group 78 S-integers 452, 523
— cohomology 78 S-projective 191
profinite group 12, 190 S-torsion 78
— order of 7 S-units 452
projective 191 S f -idèle class group 620
— S-projective 191 S cd 737
— p-projective 191 S f d 737
projective dimension Šafarevič-Tate group 481
— of a module 279 Schreier’s theorem 17
— of a ring 279 semi-direct product 24
projectives, sufficiently many 133 Serre criterion 185
prosolvable group 190 set of topological generators 275
pseudo-isomorphism 269 Shapiro map 478
Shapiro’s lemma 62
quasi-isomorphism 101 signature map 662
snake lemma 25
Ram(Ω | k) 524
solenoid 446
radical 283
special case 526, 527
rank of a free pro-c-group 198
specialization map 609
rank of a pro-p-group 224
spectral sequence 97, 131
— relation rank 227
— cup-product 114
reciprocity homomorphism 153, 157, 441
— degenerating 100
reciprocity isomorphism 153

826 Index
— differentials 98 topology
— displaced 108 — R-topology 283
— edge morphism 99 — (m,I)-topology 284
— first quadrant 99 torsor 16
— Grothendieck 132 total complex 103
— Hochschild-Serre 111, 132 trace map 125, 183
— initial terms 98 transfer 205
— limit terms 98 transfer map 52
— morphism 98 transgression 65, 74, 112, 114
— Tate 120 transpose functor 305
— transgression 112, 114 trivial G-module 8
splitting module 150 trivial cohomology 31
standard resolution 13 trivializing extension 523
Steinberg group 776
structure theorem universal δ-functor 128
— 2-dimensional regular local rings 273 universal coefficient theorem 282
— Iwasawa modules 292 universal norms 91, 440
stupid filtration 106 unramified cohomology 387
subgroup
— index of 7 VS (k, m) 531
supernatural number 6 vector bundle 59
Sylow subgroup 69 — line bundle 59
Sylow theorems 69 Verlagerung 52, 205
symbol 356 Verschiebung 341
symbol, norm residue 153, 157 virtual cohomological dimension 172
system of defining relations 226 virtually 693
Weierstraß polynomial 289

tame fundamental group 608 Weierstraß preparation theorem 290
tame symbol 360 weight 613
tamely ramified 342 Weil group 162
Tate cohomology 23, 86 welk 31
Tate duality, local 380, 383 Witt vectors 340
Tate module 613
Tate spectral sequence 120, 188, 210 ZZp -extension 636, 722
Tate twist 397 — cyclotomic 637, 722
Teichmüller character 763 Zassenhaus filtration 230


Cohomology of Number Fields

Hochgeladen von

Dokumentinformationen

Copyright

Verfügbare Formate

Dieses Dokument teilen

Dokument teilen oder einbetten

Freigabeoptionen

Stufen Sie dieses Dokument als nützlich ein?

Sind diese Inhalte unangemessen?

Copyright:

Verfügbare Formate

Cohomology of Number Fields

Hochgeladen von

Copyright:

Verfügbare Formate

Cohomology

Heidelberg, im September 1999 Alexander Schmidt

" Die Zahlentheorie nimmt unter den mathematischen Disziplinen

of finite extensions using arithmetic invariants of the local or global ground

Chapter V deals with the algebraic foundations of Iwasawa theory. We will

I. R. ŠAFAREVIČ. We present their proof, which demonstrates the power of the

xii Preface to the Second Edition

Heidelberg, September 1999 Alexander Schmidt

Preface to the Second Edition

Regensburg and Heidelberg, November 2007 Alexander Schmidt

Chapter I: Cohomology of Profinite Groups 3

Chapter II: Some Homological Algebra 97

Chapter III: Duality Properties of Profinite Groups 147

Chapter IV: Free Products of Profinite Groups 245

Chapter V: Iwasawa Modules 267

Arithmetic Theory 335

Chapter VI: Galois Cohomology 337

Chapter VII: Cohomology of Local Fields 371

Chapter VIII: Cohomology of Global Fields 425

Chapter IX: The Absolute Galois Group of a Global Field 521

Chapter X: Restricted Ramification 599

Chapter XI: Iwasawa Theory of Number Fields 721

Chapter XII: Anabelian Geometry 785

Profinite groups are topological groups which naturally occur in algebraic

§1. Profinite Spaces and Profinite Groups

The underlying topological spaces of profinite groups are of a very specific

(1.1.1) Lemma. For a Hausdorff topological space T the following conditions

4 Chapter I. Cohomology of Profinite Groups

form f −1 (W ), where W is a subset of a finite discrete space V and f : T → V

For the injectivity it suffices to show that for t, s ∈ T , t =/ s, there exists an

(1.1.2) Definition. A space T is called a profinite space if it satisfies the

A compactness argument shows that a subset V ⊆ lim←−

Recall that a topological group is a group G endowed with the structure of

§1. Profinite Spaces and Profinite Groups 5

G × G → G, (g, h) 7→ gh, are continuous. The reader will immediately

(1.1.3) Proposition. For a Hausdorff topological group G the following con-

6 Chapter I. Cohomology of Profinite Groups

is an open neighbourhood of v contained in V . Hence V is open and also

(1.1.4) Definition. A Hausdorff topological group G satisfying the equivalent

Without further mention, homomorphisms between profinite groups are al-

(1.1.5) Definition. A supernatural number is a formal product

§1. Profinite Spaces and Profinite Groups 7

(1.1.6) Definition. Let G be a profinite group and let A be an abelian torsion

(iii) The order of A is defined by

Given closed subgroups N ⊆ H ⊆ G, we have

(1.1.7) Definition. Let G be a profinite group. An abstract G-module M is

For a closed subgroup H ⊆ G we denote the subgroup of H-invariant

(1.1.8) Proposition. Let G be a profinite group and let M be an abstract

8 Chapter I. Cohomology of Profinite Groups

Proof: If we restrict the map G × M → M to G × {m}, then m ∈ M has

So far we have considered totally disconnected compact groups. If A is any

(1.1.9) Proposition. Let A be a locally compact group. Then

§1. Profinite Spaces and Profinite Groups 9

(1.1.10) Definition. Let A be a Hausdorff, abelian and locally compact topo-

A∨ := Homcts (A, IR/ZZ)

the Pontryagin dual of A.

Given locally compact topological spaces X, Y , the set of continuous maps

UK,U = {f ∈ Mapcts (X, Y ) | f (K) ⊆ U },

(1.1.11) Theorem (Pontryagin Duality). If A is a Hausdorff abelian locally

given by a 7−→ τa : A∨ → IR/ZZ, φ 7→ φ(a), is an isomorphism of topological

For an (abstract) abelian group A we use the notation

Clearly, if A is a discrete torsion group, then A∨ ∼

10 Chapter I. Cohomology of Profinite Groups

that every continuous homomorphism φ : A → IR/ZZ has finite image. If,