Neukirch 3

Jürgen Neukirch
Class Field Theory

-The Bonn lectures-
Newly edited by Alexander Schmidt
Translated from the German by F. Lemmermeyer and W. Snyder

Language Editor: A. Rosenschon
Online Edition 2.0, May 2015
Electronic Version
http://www.mathi.uni-heidelberg.de/~schmidt/Neukirch-en/
Free for private, noncommercial use only. All rights reserved by Springer
Verlag Berlin, Heidelberg, New York. Printed version available at
www.springer.com
Electronic Edition. Free for private, non-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
whether you have the most recent version of the file and whether the mistake
you found is already listed in the errata file.
Foreword
In 1969, Jürgen Neukirch’s book Klassenkörpertheorie was published by Bib-

liographisches Institut Mannheim. The main goal of the book was to grant
the reader, who has acquainted himself with the basics of algebraic number
theory, a quick and immediate access to class field theory.
Although this book has been out of print for many years, it has remained
the favorite introduction to class field theory in Germany. As a student in
the 1980s, I myself studied a copy from the library that showed clear signs of
extensive use. This motivated the idea to make the text available again, as a
printed book as well as a freely accessible file for downloading.
This book should not be confused with Neukirch’s book “Class Field Theory”
(Springer Grundlehren vol. 280, 1986), which has another focus. The text pre-
sented here is essentially identical with the German text based on Neukirch’s
original Bonn lectures; I only corrected mistakes and updated notation.
I would like to thank Rita Neukirch for her generous support for this new
English edition of the work of her late husband. I also thank Rosina Bonn
for her excellent typesetting of the original German text into LATEX and my
friends and colleagues Andreas Rosenschon, Bernd Schober and Malte Witte
for their invaluable help in improving the English edition.
Heidelberg, October 2012 Alexander Schmidt
Dedicated to my teacher
WOLFGANG KRULL
Preface
The present manuscript is an improved edition of a text that first appeared un-
der the same title in Bonner Mathematische Schriften, no. 26, and originated
from a series of lectures given by the author in 1965/66 in W. Krull’s semi-
nar in Bonn. Since the mathematical literature lacked a uniform presentation
of class field theory based on modern cohomological methods, a summarizing
exposition of these lectures seemed to be useful. The main goal was to provide
the reader, who has acquainted himself with the basics of algebraic number
theory, a quick and immediate access to class field theory.
This script consists of three parts, the first of which discusses the cohomology
of finite groups. Nowadays, cohomology has conquered large areas of algebraic
number theory. Nevertheless, the question whether class field theory can be
done without this machinery is a frequent topic of discussion. However, apart
from the possibility of formulating the theory in terms of algebras, which is
closely related to cohomology, we do not dispose of such a theory at this
point, although recent results due to J. Lubin and J. Tate on the explicit
determination of the local norm residue symbol provide some support for this
viewpoint. But one must not overlook the fact that cohomology presents –
in particular for the learner – a wealth of far reaching advantages. In class
field theory, cohomology plays the role of a calculus that allows a clear and
logical development of the theory under a unified viewpoint. Its importance,
however, is by no means only of formal nature. In fact, local class field theory
could originally be developed by defining the norm residue symbol via the
Frobenius automorphism for unramified extensions only. It was cohomology
that gave a vital impetus to the theory, making also the ramified extensions
accessible to class field theoretic methods. This relationship was discovered by
H. Hasse and had an immediate impact also on the global theory. Although
it was formulated in the language of algebras at first, the cohomological prin-
ciples behind it did not remain hidden for long. In addition, beyond class
field theory, the use of cohomological methods in general field theory has led,
via Galois cohomology, to a wealth of far reaching results with a novel allure.
viii Preface
This is another reason why a student may wish to learn the effectiveness of the
cohomological calculus in class field theory, it provides him with a concrete
example of techniques which are used in many other areas of mathematics.
On the other hand, it cannot be denied that some students interested in class
field theory are deterred by cohomology, which at first sight may seem to be
some kind of mysterious formal mechanism that is difficult to understand. For
this reason we only introduce those notions and results from cohomology that
are essential for field theoretic applications, and we have made every endeavor
to present the material in a way that is as elementary as possible by avoiding
the general notions of homological algebra.
The second part discusses local class field theory. We have put Artin’s and
Tate’s theory of class formations at the beginning; it brings out the purely
group theoretical formalism of local and global class field theory based on the
theorem of Tate. For the sake of formal simplicity we have used the notion of a
profinite group; it is, however, not absolutely necessary for an understanding
of what follows, since all the essential theorems only refer to finite groups,
which are the building blocks of profinite groups. In § 7 we have included the
recent results by Lubin and Tate [34] on the explicit determination of the
norm residue symbol, which also will be applied later in the global part of the
proof of Artin’s reciprocity law.
The third part concerns the class field theory of finite algebraic number fields.
For the sake of a development that is as straightforward as possible we have
decided to omit the theory of function fields over finite constant fields. In
order to elaborate to what extend the global theorems can be deduced from
their local counterparts, we have strictly separated considerations of a purely
local character from those possessing a specific global nature. For a clear
presentation it turned out to be appropriate to single out certain cohomology
groups that occur when considering different field extensions simultaneously.
We have exclusively used Chevalley’s notion of idèles for developing the
global theory, and yet have tried to emphasize the importance of the classical
theory going back to Kummer. We have obtained a clear structuring of the
proof of the reciprocity law by strictly separating the treatment of the idèle
group from that of the idèle class group. In the last section we establish the
connection between the modern and the classical purely ideal theoretic version
of class field theory in the sense of Hasse’s Zahlbericht.
I would like to thank my honored teacher, Professor W. Krull, for his active
interest and concern in the genesis of this script. K.-O. Stöhr has acquired
a special merit to the text; I thank him deeply for his first elaboration of
my occasionally rather sketchy accounts on cohomology and local class field
theory, as well as for suggesting many essential improvements.
Bonn, July 1969 Jürgen Neukirch
Contents
Part I: Cohomology of Finite Groups . . . . . . . . . . . . . . . . . . . . . . . 1

§ 1. G-Modules . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
§ 2. The Definition of Cohomology Groups . . . . . . . . . . . . . . . . . . . . . 11
§ 3. The Exact Cohomology Sequence . . . . . . . . . . . . . . . . . . . . . . . . . . 20
§ 4. Inflation, Restriction and Corestriction . . . . . . . . . . . . . . . . . . . . . 32
§ 5. The Cup Product . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 44
§ 6. Cohomology of Cyclic Groups . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 51
§ 7. Tate’s Theorem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 57
Part II: Local Class Field Theory . . . . . . . . . . . . . . . . . . . . . . . . . . . . 61

§ 1. Abstract Class Field Theory . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 63
§ 2. Galois Cohomology . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 77
§ 3. The Multiplicative Group of a p-adic Number Field . . . . . . . . . . 79
§ 4. The Class Formation of Unramified Extensions . . . . . . . . . . . . . . 82
§ 5. The Local Reciprocity Law . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 89
§ 6. The Existence Theorem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 95
§ 7. Explicit Determination of the Norm Residue Symbol . . . . . . . . . 98
Part III: Global Class Field Theory . . . . . . . . . . . . . . . . . . . . . . . . . . 111

§ 1. Number Theoretic Preliminaries . . . . . . . . . . . . . . . . . . . . . . . . . . . 113
§ 2. Idèles and Idèle Classes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 117
§ 3. Cohomology of the Idèle Group . . . . . . . . . . . . . . . . . . . . . . . . . . . 122
§ 4. Cohomology of the Idèle Class Group . . . . . . . . . . . . . . . . . . . . . . 128
§ 5. Idèle Invariants . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 138
§ 6. The Reciprocity Law . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 146
§ 7. The Existence Theorem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 156
§ 8. The Decomposition Law . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 168
§ 9. The Ideal Theoretic Formulation of Class Field Theory . . . . . . . 173
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 181
Index . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 183
Part I
Cohomology of Finite Groups
§ 1. G-Modules 3
§ 1. G-Modules
The cohomology of finite groups deals with a general situation that occurs
frequently in different concrete forms. For example, if L|K is a finite Galois
extension with Galois group G, then G acts on the multiplicative group L×
of the extension field L. In the special case of an extension of finite algebraic
number fields, G acts on the ideal group J of the extension field L. The theory
of group extensions provides us with the following example: If G is an abstract
finite group and A is a normal subgroup, then G acts on A via conjugation. In
representation theory we study matrix groups G that act on a vector space.
The basic notion underlying all these examples is that of a G-module. We will
now present some general considerations about G-modules, some of which the
reader may already know from the theory of modules over general rings.
In the following, G will always denote a multiplicatively written finite group;

its unit element will be denoted by 1.
(1.1) Definition. A G-module A is an (additively written) abelian group A

on which the group G acts in such a way that for all σ, τ ∈ G and all a, b ∈ A
we have
1) 1a = a,
2) σ(a + b) = σa + σb,
3) (στ )a = σ(τ a).
Although in applications we mainly deal with multiplicatively written G-

modules A, we prefer for formal reasons to write these groups in this section
additively.
We can interpret G-modules as modules over rings by introducing the group
ring ZZ[G] of G. This ring consists of all formal sums
X
nσ σ
σ∈G
with integral coefficients nσ ∈ ZZ. In other words, ZZ[G] is the free abelian
group on the elements of G:
nX o
ZZ[G] = nσ σ | nσ ∈ ZZ .
σ∈G
P
Since the sums σ∈G nσ σ can be multiplied, ZZ[G] is a ring. Therefore we may
interpret a G-module A as a module over the ring ZZ[G], where the action of
ZZ[G] on A is defined as
X X
nσ σ a = nσ (σa), a ∈ A.
σ∈G σ∈G
4 Part I. Cohomology of Finite Groups
Of course, ZZ[G] considered just as an additive group is a G-module itself; it

will play a distinguished role in our considerations.
The group ring ZZ[G] contains two distinguished ideals:

nX X o n X o
IG = nσ σ | nσ = 0 and ZZ·NG = n · σ | n ∈ ZZ .
σ∈G σ∈G σ∈G
The ideal IG is called the augmentation ideal of ZZ[G]. It is the kernel of
the homomorphism
X X
ε : ZZ[G] −→ ZZ with ε nσ σ = nσ ,
σ∈G σ∈G
which is called the augmentation of ZZ[G].
P
The element NG = σ∈G σ ∈ ZZ[G] is calledP the norm (or also trace) of
ZZ[G]. For each element τ ∈ G we have τ NG = σ∈G τ σ = NG , which implies
that ZZ·NG is an ideal in ZZ[G]. The map
µ : ZZ −→ ZZ[G] with µ(n) = n·NG
is called the coaugmentation of ZZ[G]. We set JG = ZZ[G]/ZZ·NG , and obtain
the exact sequences1) of rings and ring homomorphisms
ε
0 −→ IG −→ ZZ[G] −→ ZZ −→ 0,
µ
0 −→ ZZ −→ ZZ[G] −→ JG −→ 0.
Viewing these rings only as additive groups, we see immediately that they are
all free abelian groups, and that IG and JG are direct summands of ZZ[G]:
(1.2) Proposition. IG is the free abelian group on the elements σ −1, σ ∈ G,

σ 6= 1, and JG is the free abelian group on the elements σ mod ZZ·NG , σ 6= 1.
We have direct sum decompositions
ZZ[G] = IG ⊕ ZZ·1 ∼
= IG ⊕ ZZ ,
ZZ[G] = ( σ6=1 ZZσ) ⊕ ZZ·NG ∼
L
= JG ⊕ ZZ .
P P
Proof. If σ∈G nσ σ ∈ IG , then σ∈G nσ = 0, from which we obtain
X X
nσ σ = nσ (σ − 1);
σ∈G σ∈G
P
if, in addition, σ∈G, σ6=1 nσ (σ − 1) = 0, then nσ = 0 for all σ ∈ G, σ 6= 1.
1) i j
A sequence · · · → A → B → C → · · · of groups, modules or rings and ho-
momorphisms i, j, . . . is called exact, if the image of each map is equal to the
kernel of the subsequent map. We will often have to deal with short exact se-
i j
quences 0 → A → B → C → 0. Such a sequence encodes the information that
the homomorphism j : B → C is surjective with kernel iA ∼ = A.
§ 1. G-Modules 5
P
Since each element σ∈G nσ σ ∈ ZZ[G] can be written in the following form
X X X
nσ σ = nσ (σ − 1) + nσ · 1,
we have the decomposition as an evidently direct sum ZZ[G] = IG ⊕ ZZ·1.

P
If, on the other hand, σ∈G nσ σ mod ZZ·NG ∈ JG , then we can write
X X X X
nσ σ = (nσ − n1 )σ + n1 · σ≡ (nσ − n1 )σ mod ZZ·NG ,
σ∈G σ6=1 σ∈G σ6=1
P
and σ6=1 nσ σ ∈ Z
Z·NG clearly implies nσ = 0 for all σ 6= 1.
Thus JG is the free abelian group generated by the elements σ mod ZZ·NG
with σ 6= 1. Because of the uniqueness of the representation
X X
nσ σ = (nσ − n1 )σ + n1 ·NG ,
σ∈G σ6=1
L
we also obtain the direct sum decomposition ZZ[G] = σ6=1 ZZσ ⊕ZZ·NG .
The ideals IG and ZZ·NG in ZZ[G] are dual to each other in the following sense.
(1.3) Proposition. IG = Ann ZZ·NG and ZZ·NG = Ann IG .
Proof.
XWe have
X
X X
n σ σ · NG = nσ (σ · NG ) = nσ NG = n σ · NG = 0
σ∈G σ∈G σ∈G σ∈G
if and only if X
nσ = 0.
σ∈G
Thus Ann ZZ·NG = IG . On the other hand, we know from (1.2) that IG is the
free abelian group generated by the elements σ − 1, σ ∈ G. Therefore
X X
nτ τ ∈ Ann IG ⇐⇒ nτ τ (σ − 1) = 0 for all σ ∈ G
τ ∈G τ ∈G
X X
⇐⇒ nτ τ σ = nτ τ for all σ ∈ G
τ ∈G τ ∈G
⇐⇒ nτ = n1 for all τ ∈ G
X
⇐⇒ nτ τ = n1 ·NG ∈ ZZ·NG ,
τ ∈G
so that ZZ·NG = Ann IG , as claimed.
After these remarks on group rings we now return to general G-modules. For
each G-module A we have the following four distinguished submodules:
AG = {a ∈ A | σa = a for all σ ∈ G}, the fixed group of A,

NG A = {NG a = σ∈G σa | a ∈ A}, the norm group of A 2) ,
P
NG A = {a ∈ A | NG a = 0},
P
IG A = { σ∈G nσ (σaσ − aσ ) | aσ ∈ A}.
Since IG is the module generated by the elements σ − 1, σ ∈ G, we obviously
have AG = {a ∈ A | IG a = 0}. On the other hand, IG A is the module
generated by the elements σa − a, a ∈ A, σ ∈ G. Proposition (1.3) provides
us with the inclusions
NG A ⊆ AG and IG A ⊆ NG A,
and we can form the factor groups

AG /NG A and NG A/IG A.
These groups will turn out to be the cohomology groups of the G-module A
of dimension 0 and −1 respectively.
If A is a G-module and g is a subgroup of G, it is clear that A is also a

g-module. Moreover, if g is a normal subgroup of G, the fixed module Ag is
obviously a G/g-module.
In the following we consider the most important functorial properties of G-
modules.
Let A and B denote two G-modules. A homomorphism
f : A −→ B
is called a G-homomorphism if f (σa) = σf (a) for all σ ∈ G. We will
often interpret a G-module A simply as an abelian group; in this case we will
talk about ZZ-modules and ZZ-homomorphisms instead of G-modules and
G-homomorphisms.
Given two G-modules A and B we can construct a third G-module, the module
Hom(A, B)
of all ZZ-homomorphisms f : A → B, on which the elements σ ∈ G act as:
σ(f ) = σ ◦ f ◦ σ −1 , that is, σ(f )(a) = σf (σ −1 a), a ∈ A.
The group HomG (A, B) of all G-homomorphisms from A to B is a subgroup

of Hom(A, B); clearly it is the fixed module of the G-module Hom(A, B):
HomG (A, B) = Hom(A, B)G .
2) P
For the elements σ∈G σa the name trace seems more appropriate. However,
given that our later applications mainly involve multiplicative G-modules, we
have decided use the word norm already here.
§ 1. G-Modules 7
In addition to HomG (A, B) we have another G-module, the tensor product

A ⊗ZZ B.
Roughly speaking,
P the tensor product of A and B consists of all formal sums
of products i ai ·bi , ai ∈ A, bi ∈ B. A precise definition is given by:
(1.4) Definition. Given two abelian groups ( ZZ-modules) A and B, let F be

the free abelian group generated by all pairs (a, b), a ∈ A, b ∈ B, and let R
denote the subgroup of F generated by the elements of the form
(a + a0 , b) − (a, b) − (a0 , b) and (a, b + b0 ) − (a, b) − (a, b0 ).
Then the factor group
F/R = A ⊗ZZ B
is called the tensor product of A and B over ZZ.
Since we only consider tensor products over the ring ZZ, we will write for
simplicity A ⊗ B instead of A ⊗ZZ B. We denote by a ⊗ b the coset
a ⊗ b = (a, b) + R ∈ A ⊗ B.
By definition, the tensor product A ⊗ B consists of all elements of the form
X
ai ⊗ bi , ai ∈ A, bi ∈ B,
i
hence is generated by the elements a ⊗ b.
In the special case A = ZZ we will often regard the tensor product ZZ ⊗ B and
the ZZ-module B as equal by identifying3) n ⊗ b and n · b for n ∈ ZZ, b ∈ B.
If A and B are two abelian groups, we will identify the groups A ⊗ B and
B ⊗ A via the isomorphism
f : A ⊗ B −→ B ⊗ A with f (a ⊗ b) = b ⊗ a .
Similarly, if A, B, C are three abelian groups, we will regard the groups
(A ⊗ B) ⊗ C and A ⊗ (B ⊗ C) as equal, making use of the isomorphism
f : (A ⊗ B) ⊗ C −→ A ⊗ (B ⊗ C) with f ((a ⊗ b) ⊗ c) = a ⊗ (b ⊗ c) .
If A and B are two G-modules, then A ⊗ B becomes a G-module by defining

4)
σ(a ⊗ b) = σa ⊗ σb, a ∈ A, b ∈ B; σ ∈ G.
3)
Starting from this case, we can think of the construction of the general tensor
product as a formal change of the domain of coefficients of B from ZZ to A. For
example, every abelian group B can be extended to a Q-vector space by tensoring
B with Q. This transition is the formal extension of the scalars of the ZZ-module B
from ZZ to Q, and multiplication of b ∈ B by a rational number r ∈ Q corresponds
to forming r ⊗ b ∈ Q ⊗ B.
4)
Since the products a ⊗ b generate the module A ⊗ B, we obtain the action of σ
on the entire module by linear extension.
In general it is not true that AG ⊗ B G is the fixed module of A ⊗ B. In fact

AG ⊗B G might not even be a submodule of A⊗B. We only have the canonical
(but in general neither injective nor surjective) homomorphism
AG ⊗ B G −→ (A ⊗ B)G .
It is easy to verify that the functors HomG and ⊗ are additive:
(1.5) Proposition. Let {Aι | ι ∈ I} be a family of G-modules, and let X be

another G-module. Then we have canonical isomorphisms 5)
Aι ) ∼
M M
X ⊗( (X ⊗ Aι ), =
ι ι
Aι , X) ∼ Aι ) ∼
M Y Y Y
HomG ( = HomG (Aι , X), HomG (X, = HomG (X, Aι ).
ι ι ι ι
Moreover, if X is finitely generated as an abelian group, then
X ⊗ ( Aι ) ∼ Aι ) ∼
Y Y M M
= (X ⊗ Aι ), HomG (X, = HomG (X, Aι ).
ι ι ι ι
Let A and B be G-modules and

h
A −→ A0
a G-homomorphism. Then h induces a G-homomorphism in the ‘opposite’

direction
Hom(A, B) ←− Hom(A0 , B)
given by composition f 7→ f ◦ h (f ∈ Hom(A0 , B)), and a G-homomorphism

A ⊗ B −→ A0 ⊗ B
defined by a ⊗ b 7→ h(a) ⊗ b. On the other hand, given a G-homomorphism
g
B −→ B 0 ,
there are analogously defined G-homomorphisms
Hom(A, B) −→ Hom(A, B 0 )
and
A ⊗ B −→ A ⊗ B 0 .
Because of these properties, Hom is called a contravariant functor in the first
and a covariant functor in the second argument, and the tensor product ⊗ is
called a covariant functor in both arguments.
5) L
The symbol stands for the direct sum, i.e., the group of all families
(. . . , aι , . . .), where
Q only finitely many components aι are different from zero.
In contrast, denotes the direct product, i.e., the group of all families
(. . . , aι , . . .).
§ 1. G-Modules 9
If we have two G-homomorphisms

h g
A0 −→ A and B −→ B 0 ,
we obtain a G-homomorphism
(h, g) : Hom(A, B) −→ Hom(A0 , B 0 )
by defining f 7→ g ◦ f ◦ h (f ∈ Hom(A, B)), and for two G-homomorphisms
h g
A −→ A0 and B −→ B 0
we get a G-homomorphism
h ⊗ g : A ⊗ B −→ A0 ⊗ B 0 ,
by setting h ⊗ g (a ⊗ b) = h(a) ⊗ g(b).
In what follows, G-free G-modules will play an important role. A G-module

A is called G-free or ZZ[G]-free, if it is the direct sum of G-modules which
are isomorphic to ZZ[G]. A first basic property of G-free modules is:
(1.6) Proposition. Let X be a G-free G-module and

h g
0 −→ A −→ B −→ C −→ 0
an exact sequence of G-modules A, B, C and G-homomorphisms h, g. Then
the induced sequence
0 −→ HomG (X, A) −→ HomG (X, B) −→ HomG (X, C) −→ 0
is also exact.
Γι with Γι ∼
L
Write X = ι = ZZ[G]. From (1.5) we have the decomposition
Y
HomG (X, A) = HomG (Γι , A).
ι
∼ HomG (ZZ[G], A) ∼
If we set Aι = HomG (Γι , A) = = A (where the last isomor-
phism is given by the map f ∈ HomG (ZZ[G], A) 7→ f (1) ∈ A), and define Bι
and Cι analogously, we obtain the exact sequence
0 −→ Aι −→ Bι −→ Cι −→ 0,
which implies the proposition.
Remark. Proposition (1.6) is valid more generally for so-called projective

G-modules X, i.e., G-modules which have the property that every diagram
X
f0
f
g
B C 0
with G-modules B, C and G-homomorphisms g, f 0 , g surjective, can be ex-
tended to a commutative diagram by a G-homomorphism f : X → B.
For arbitrary G-modules X it is easy to verify that if one omits the last map
→ 0 in the induced HomG -sequence, the remaining sequence is still exact.
A G-free G-module is, of course, also ZZ-free, and therefore a free abelian
group, since ZZ[G] is the free abelian group generated by the elements of G.
For most questions concerning the exactness of sequences it suffices to consider
only ZZ-modules and ZZ-homomorphisms. In later applications we will need the
following three lemmas.
dq dq+1
(1.7) Lemma. If · · · ←− Xq−1 ←− Xq ←− Xq+1 ←− · · · is an exact
sequence of ZZ-free modules and D is an arbitrary ZZ-module, then the sequence
· · · −→ Hom(Xq−1 , D) −→ Hom(Xq , D) −→ Hom(Xq+1 , D) −→ · · ·
is also exact.
Proof. Let Cq = ker dq = im dq+1 . Since Cq−1 is free as a subgroup of Xq−1 ,

the exact sequence 0 ← Cq−1 ← Xq ← Cq ← 0 is split, i.e., there is a
homomorphism ε : Cq−1 → Xq with dq ◦ ε = id, and Cq is a direct summand
of Xq : Xq = Cq ⊕ Xq0 for all q. Thus if f is in the kernel of Hom(Xq , D) →
Hom(Xq+1 , D), then f vanishes on Cq , and therefore induces a homomorphism
g 0 : Cq−1 → D with f = g 0 ◦ dq . Since Cq−1 is a direct summand of Xq−1 ,
we can extend g 0 to a homomorphism g ∈ Hom(Xq−1 , D), and f is the image
of g under the homomorphism Hom(Xq−1 , D) → Hom(Xq , D). On the other
hand, if f ∈ Hom(Xq , D) is in the image of Hom(Xq−1 , D) → Hom(Xq , D),
f = f 0 ◦ dq with f 0 ∈ Hom(Xq−1 , D), then f ◦ dq+1 = f 0 ◦ dq ◦ dq+1 = 0, i.e.,
f lies in the kernel of Hom(Xq , D) → Hom(Xq+1 , D).
(1.8) Lemma. If 0 → X → Y → Z → 0 is an exact sequence of free

ZZ-modules and A is an arbitrary ZZ-module, then the sequence
0 −→ X ⊗ A −→ Y ⊗ A −→ Z ⊗ A −→ 0
is also exact.
Proof. The exactness of the sequence X ⊗ A → Y ⊗ A → Z ⊗ A → 0

is completely trivial and holds without assuming that the modules are free.
Hence we only need to show that the map X ⊗A → Y ⊗A is injective. Since Z
is free, there is a homomorphism Z → Y whose composite with the given map
Y → Z is the identity map on Z. This implies that the image X 0 of X in Y is a
direct summand, i.e., Y = X 0 ⊕X 00 . Thus we find Y ⊗A = (X 0 ⊗A)⊕(X 00 ⊗A),
which implies the claimed injectivity.
(1.9) Lemma. If 0 → A → B → C → 0 is an exact sequence of ZZ-modules

and X a ZZ-free ZZ-module, then
0 −→ X ⊗ A −→ X ⊗ B −→ X ⊗ C −→ 0
is also an exact sequence.
§ 2. The Definition of Cohomology Groups 11
Proof. Let X = ι Zι with Zι ∼

L
= ZZ. Since the functor ⊗ is additive, we have
the canonical isomorphism
∼ (Zι ⊗ A) ∼ Aι with Aι = Zι ⊗ A ∼
M M
X ⊗A= = = A,
ι ι
and similarly for B and C. Hence the exactness of the sequence
0 −→ Aι −→ Bι −→ Cι −→ 0
immediately implies the exactness of
0 −→ X ⊗ A −→ X ⊗ B −→ X ⊗ C −→ 0.
§ 2. The Definition of Cohomology Groups
It is a characteristic of cohomology theory that in order to give even simple

definitions and theorems, one must introduce an extensive formalism of ho-
momorphisms, functors and sequences. At first this might give the impression
that we are dealing with a particularly difficult and deep mathematical disci-
pline. However, once the reader has become more familiar with these methods,
he will realize that these considerations are particularly simple, and they may
even appear somewhat anemic. Nevertheless, their frequent repeated use leads
to concepts and theorems which could hardly be developed using a more el-
ementary approach. In order to introduce our cohomology groups, we start
with such formal considerations, although the definition of these groups can
be given in a direct and elementary way, as well6) .
Let G be a finite group. By a complete free resolution of the group G, or

also of the G-module ZZ 7) , we mean a complex
· · ·
d−2 d−1 d0 d1 d2 d3
X−2 X−1 X0 X1 X2 ···
µ ε
ZZ
0 0
with the following properties:
6)
Cf. [16], 15.7, p. 236.
7)
We always consider ZZ as a G-module by letting the group G act on ZZ trivially
(i.e., as the identity).
(1) The Xq are free G-modules,

(2) ε, µ, dq are G-homomorphisms,
(3) d0 = µ ◦ ε ,
(4) at every term we have exactness.
Thus a complete free resolution consists in fact of two exact sequences

ε 1 2d d
0 ←− ZZ ←− X0 ←− X1 ←− ···
and µ d−1 d−2

0 −→ ZZ −→ X−1 −−→ X−2 −−→ · · ·
of free G-modules which are spliced together. Originally, cohomology groups

were derived from the first, and homology groups from the second sequence.
But putting these two sequences together is a crucial step which leads to a uni-
fied and, concerning the functorial properties, harmonious fusion of homology
and cohomology.
When defining the cohomology groups of a G-module we could start with an
arbitrary complete resolution, even one in which the Xq need only be projec-
tive G-modules. But then we would need to show the definition is independent
from the chosen resolution. In order to avoid this effort, we start with a par-
ticular complete resolution, the so-called standard resolution, which has its
origins in algebraic topology and arises in the following way.
For q ≥ 1 we consider all q-tuples (σ1 , . . . , σq ), where the σi run through the
group G; we call such a q-tuple a q-cell (with “vertices” σ1 , . . . , σq ). We use
these q-cells as free generators of our G-module, i.e., we set
M
Xq = X−q−1 = ZZ[G](σ1 , . . . , σq ).
For q = 0 we put
X0 = X−1 = ZZ[G],
where we choose the identity element 1 ∈ ZZ[G] as the generating “null cell”.
In particular, the modules
. . . , X−2 , X−1 , X0 , X1 , X2 , . . .
are free G-modules.
Define the G-homomorphisms ε : X0 → ZZ and µ : ZZ → X−1 by (cf. §1, p. 4)
P P
ε( σ∈G nσ σ) = σ∈G nσ (augmentation)
µ(n) = n·NG (coaugmentation)
In order to determine the remaining G-homomorphisms dq it suffices, of

course, to give their values on the free generators (σ1 , . . . , σq ). Here we set
d0 1 = NG for q = 0,
d1 (σ) = σ − 1 for q = 1,
dq (σ1 , . . . , σq ) = σ1 (σ2 , . . . , σq )
Pq−1
+ i=1 (−1)i (σ1 , . . . , σi−1 , σi σi+1 , σi+2 , . . . , σq )
+(−1)q (σ1 , . . . , σq−1 ) for q > 1,
−1
P
d−1 1 = σ∈G [σ (σ) − (σ)] for q = −1,
d−q−1 (σ1 , . . . , σq ) = σ∈G σ −1 (σ, σ1 , . . . , σq )

P
Pq
+ σ∈G i=1 (−1)i (σ1 , . . . , σi−1 , σi σ, σ −1, σi+1 , . . . , σq )
P
+ σ∈G (−1)q+1 (σ1 , . . . , σq , σ)

P
for − q − 1 < −1.
From these definitions we obtain a complex
· · ·
d−2 d−1 d0 d1 d2 d3
X−2 X−1 X0 X1 X2 ···
µ ε
ZZ
0 0,
which we call the standard complex of the group G. We will see that this
complex is a complete free resolution of the group G. The conditions (1)–
(3) are trivially satisfied: By construction the Xq are free G-modules, the
ε, µ, dq are G-homomorphisms; and because µ ◦ ε(1) = µ(1) = NG = d0 1, we
have d0 = µ ◦ ε. Hence it only remains to show that we have exactness at
every term. To prove this fact, we use computations that are based on similar
considerations in algebraic topology. We first show that the sequence
ε 1 2 d 3 d d
(∗) 0 ←− ZZ ←− X0 ←− X1 ←− X2 ←− ···
is exact. To do this, we define the following ZZ-homomorphisms:
E : ZZ −→ X0 with E(1) = 1,
D0 : X0 −→ X1 with D0 (σ) = (σ),
Dq : Xq −→ Xq+1 with Dq (σ0 (σ1 , . . . , σq )) = (σ0 , . . . , σq ) for q ≥ 1.
An elementary calculation shows that
E ◦ ε + d1 ◦ D0 = id and Dq−1 ◦ dq + dq+1 ◦ Dq = id.
These formulas imply that given x ∈ ker ε (resp. x ∈ ker dq ), we have x =

d1 D0 x ∈ im d1 (resp. x = dq+1 Dq x ∈ im dq+1 ), which proves the inclusions
ker ε ⊆ im d1 and ker dq ⊆ im dq+1 for q ≥ 1.
On the other hand, it is easily checked that ε◦d1 = 0, so that ker ε ⊇ im d1 . We

now prove by induction that dq ◦ dq+1 = 0. For this we assume dq−1 ◦ dq = 0.
(For q = 1 we replace d0 by ε and D−1 by E.) Then, on one hand, we have
dq = (Dq−2 ◦ dq−1 + dq ◦ Dq−1 ) ◦ dq = dq ◦ Dq−1 ◦ dq ,

and on the other hand,
dq = dq ◦ (Dq−1 ◦ dq + dq+1 ◦ Dq ) = dq ◦ Dq−1 ◦ dq + dq ◦ dq+1 ◦ Dq .
Subtracting these equations we obtain
dq ◦ dq+1 ◦ Dq = 0.
But since every cell in Xq+1 lies in the image of Dq , we have
dq ◦ dq+1 = 0,
and therefore ker dq ⊇ im dq+1 for q ≥ 1. Thus the sequence (∗) is exact.
The second sequence

µ d−1 d−2 d−3
(∗∗) 0 −→ ZZ −→ X−1 −−→ X−2 −−→ X−3 −−→ · · ·
arises from (∗) by dualizing. In fact, from (∗) we first obtain the sequence
(∗∗∗) 0 −→ Hom(ZZ, ZZ) −→ Hom(X0 , ZZ) −→ Hom(X1 , ZZ) −→ · · · ,
which is exact by (1.7).
Let {xi } be the system of ZZ[G]-free generators of Xq consisting of all q-cells.

Then the “dual basis” {x∗i } of {xi } defined by

 1 for σ = 1 and i = k
x∗i (σxk ) =
 0 otherwise.
is a ZZ[G]-free system of generators of Hom(Xq , ZZ). Thus the G-modules

Hom(Xq , ZZ) and Xq are canonically isomorphic. If we identify xi with x∗i ,
we can write
X−q−1 = Hom(Xq , ZZ) (q ≥ 0) and ZZ = Hom(ZZ, ZZ).
An elementary calculation now shows that under these identifications the
sequence (∗∗∗) is transformed into the sequence (∗∗), hence (∗∗) is exact.
Finally, since µ is injective, ε is surjective, and d0 = µ◦ε, we have ker d0 = ker ε
and im d0 = im µ; therefore
ker d0 = im d1 and im d0 = ker d−1 .
This completes the proof that the standard complex is exact at all terms.
Now we define our cohomology groups using the standard complex. If A is a

G-module, we set
Aq = HomG (Xq , A).
We call the elements of Aq , i.e., the G-homomorphisms x : Xq → A, the
q-cochains of A. From the exact sequence
d−2 d−1 0 1d 2 d 3 d d
· · · ←− X−2 ←− X−1 ←− X0 ←− X1 ←− X2 ←− ···
we obtain the sequence

∂−2 ∂−1 0 ∂ 1 2 ∂3 ∂ ∂
· · · −→ A−2 −→ A−1 −→ A0 −→ A1 −→ A2 −→ ··· ,
in which, because dq ◦ dq+1 = 0, we evidently have ∂q+1 ◦ ∂q = 0; therefore
im ∂q ⊆ ker ∂q+1 .
Contrary to the first sequence, the second sequence is not exact in general,
and the cohomology groups “measure” its deviation from being exact. We set
Zq = ker ∂q+1 , Rq = im ∂q ,
and call the elements in Zq (resp. Rq ) the q-cocyles (resp. q-coboundaries).
We now define the cohomology groups of G with coefficients in A as follows:
(2.1) Definition. The factor group

H q (G, A) = Zq /Rq
is called the cohomology group of dimension q (q ∈ ZZ) of the G-module A;
we also say H q (G, A) is the q-th cohomology group with coefficients in A.
We remark that the cohomology groups H −q−1 (G, A) are the usual homology
groups denoted by Hq (G, A) (q ≥ 1). In algebraic topology, the cohomology
groups (with coefficients in ZZ) were originally introduced as the character
groups of the homology groups. This origin has left traces in the fact that the
left side of the standard complex is obtained from the right side by duality.
We point out again that splicing the two sides into a complete resolution,
which allows interpreting homology groups as cohomology groups of negative
dimension, is a crucial step which yields more than just a formal unification8) .
We now come to the problem of analyzing the concrete meaning of cohomology

groups. The cochain group
Aq = A−q−1 = HomG (Xq , A), q ≥ 1,
consists of all G-homomorphisms x : Xq → A. Since the Xq have the q-cells
(σ1 , . . . , σq ) as free generators, a G-homomorphism x : Xq → A is uniquely
determined by its values on the q-tuples (σ1 , . . . , σq ). Thus we can view every
cochain as a function with arguments in G and values in A, hence as a map
x : G × · · · × G −→ A.
| {z }
q-times
By taking this view, we can identify

Aq = A−q−1 = {x : G × · · · × G −→ A}, q ≥ 1,
| {z }
q-times
8)
This fusion of homology and cohomology is due to J. Tate.
and obviously
A0 = A−1 = HomG (ZZ[G], A) = A.
From the definition of the homomorphisms dq of the standard complex, we
obtain for the maps ∂q in the sequence
∂−2 ∂−1 0 1 ∂ 2 3 ∂ ∂ ∂
· · · −−→ A−2 −−→ A−1 −→ A0 −→ A1 −→ A2 −→ ···
the following formulas
∂0 x = NG x for x ∈ A−1 = A,
(∂1 x)(σ) = σx − x for x ∈ A0 = A,
(∂q x)(σ1 , . . . , σq ) = σ1 x(σ2 , . . . , σq )
Pq−1
+ i=1 (−1)i x(σ1 , . . . , σi σi+1 , . . . , σq )
+(−1)q x(σ1 , . . . , σq−1 ) for x ∈ Aq−1 , q ≥ 1,
−1
P
∂−1 x = σ∈G (σ x(σ) − x(σ)) for x ∈ A−2 ,
(∂−q−1 x)(σ1 , . . . , σq ) = σ∈G σ −1 x(σ, σ1 , . . . , σq )

P
Pq
+ i=1 (−1)i (σ1 , . . . , σi−1 , σi σ, σ −1 , σi+1 , . . . , σq )

+(−1)q+1 x(σ1 , . . . , σq , σ) for x ∈ A−q−2 , q ≥ 0.
The q-cocycles are therefore the maps
x : G × · · · × G −→ A
with ∂q+1 x = 0, and the q-coboundaries among them are those maps for which
there is a y ∈ Aq−1 such that x = ∂q y.
We remark that in algebraic applications only the cohomology groups of low

dimension appear. The reason for this is that only for these we have a con-
crete algebraic interpretation. The cohomological calculus would doubtlessly
acquire considerably more significance if we had a tangible interpretation for
the cohomology groups of higher dimensions as well. For small dimensions,
the cohomology groups are given as follows:
The Group H −1 (G, A). We have

Z−1 = ker ∂0 = NG A ((−1)-cocycles),
R−1 = im ∂−1 = IG A ((−1)-coboundaries).
We thus obtain (cf. §1, p. 6)
H −1 (G, A) = NG A/IG A.
The Group H 0 (G, A). We have

Z0 = ker ∂1 = AG (0-cocycles),
R0 = im ∂0 = NG A (0-coboundaries).
We obtain
H 0 (G, A) = AG /NG A,
the norm residue group of the G-module A; this group is the main object
of interest in class field theory.
The Group H 1 (G, A). The 1-cocycles are the functions x : G → A with
∂2 x = 0, thus satisfying the property
x(στ ) = σx(τ ) + x(σ) for σ, τ ∈ G.
Because this relation is similar to the one of being a homomorphism, the
1-cocycles are often also called crossed homomorphisms.
The 1-coboundaries are obviously the functions
x(σ) = σa − a, σ ∈ G,
with fixed a ∈ A = A0 (i.e., x = ∂1 a).
If the group G acts trivially (i.e., as the identity) on A, then obviously Z1 =
Hom(G, A) and R1 = 0; therefore
H 1 (G, A) = Hom(G, A).
In particular, in case A = Q/ZZ, we obtain the character group of G:
H 1 (G, Q/ZZ) = Hom(G, Q/ZZ) = χ(G).
When studying G-modules and their properties, one is immediately led to the
cohomology group H 1 (G, A) because of the following considerations:
If we start with a short exact sequence of G-modules A, B, C,

i j
0 −→ A −→ B −→ C −→ 0
and pass to the sequence of their fixed modules AG , B G , C G , the sequence
i j
0 −→ AG −→ B G −→ C G
always remains exact, but the homomorphism j is in general no longer sur-
jective. The question why exactness fails at the last term leads to a canonical
δ
homomorphism C G → H 1 (G, A).
j
In fact, let c ∈ C G . Since the homomorphism B → C is surjective, there is
an element b ∈ B with jb = c, but it is not certain that this element b can be
chosen from B G , i.e., in such a way that σb − b = 0 for all σ ∈ G. Since
j(σb − b) = σ(jb) − jb = σc − c = 0 ,

j
we can only say that σb − b lies in the kernel of B → C, and thus in the image
i
of A → B, so that
iaσ = σb − b, aσ ∈ A.
It is easy to verify that aσ defines a 1-cocycle with coefficients in A. The only

freedom in the association c 7→ aσ is the choice of b with jb = c. If we choose
a different element b0 , we obtain a 1-cocycle a0σ that differs from aσ only by
a 1-coboundary. Therefore every c ∈ C G defines a unique cohomology class
aσ ∈ H 1 (G, A), and it is easy to see that c is in the image of B G → C G
if and only if aσ = 0. In other words, we have a canonical homomorphism
δ
C G → H 1 (G, A) with the property that the following sequence is exact
i j δ
0 −→ AG −→ B G −→ C G −→ H 1 (G, A).
In the next section we will encounter these ideas again in a more general
setting.
The Group H 2 (G, A). We have as 2-cocycles all functions x : G × G → A

which satisfy the equation ∂3 x = 0, thus the equation
x(στ, ρ) + x(σ, τ ) = σx(τ, ρ) + x(σ, τ ρ), σ, τ, ρ ∈ G.
Among these we find that the 2-coboundaries are the functions such that
x(σ, τ ) = σy(τ ) − y(στ ) + y(σ)
for an arbitrary 1-cochain y : G → A.
Long before the development of cohomology theory the 2-cocycles were known
in the theory of groups and algebras as so-called factor systems, and it
is fair to say that they historically represent the beginning of cohomological
considerations in algebra. We want to explain briefly how factor systems come
up in the theory of group extensions. This is the following type of problem.
Suppose that we are given an abelian group A, written multiplicatively, and an

arbitrary group G. We want to find all group extensions G
b of A (i.e., all groups
Gb with a subgroup isomorphic to A) such that A is a normal subgroup in G b
∼
and G/A = G. The question now is which data (except A and G) determine
b
the possible solutions of this problem.
Assume first we have a solution G,
b so that A / G b and G/A
b ∼
= G. If we
choose a system of representatives for the factor group G/A
b ∼
= G, i.e., for
each σ ∈ G we choose a preimage uσ ∈ G, then every element in G
b b can be
written uniquely in the form
(1) a · uσ , a ∈ A, σ ∈ G.
In order to obtain the complete group table for the group G, b it obviously
suffices to know how the products uσ · a (σ ∈ G, a ∈ A) and uσ ·uτ (σ, τ ∈ G)
are represented in the form (1).
b uσ · a lies in the same right coset as uσ , i.e.,

Now since A is normal in G,
(2) uσ · a = aσ · uσ
for some aσ ∈ A. This gives the abelian group A a natural structure as a
G-module, because the elements σ ∈ G act on A (independently of the choice
of uσ ) via the rule a 7→ aσ = uσ · a · u−1
σ .
The product uσ · uτ lies in the same coset as uστ , i.e.,

(3) uσ · uτ = x(σ, τ ) · uστ with x(σ, τ ) ∈ A.
In this equation, the factor system x(σ, τ ) appears, which is easily seen to be a
2-cocycle of the G-module A. In fact, since multiplication in G b is associative,
(uσ · uτ ) · uρ = uσ · (uτ · uρ ),
from which we get the equation
(uσ · uτ ) · uρ = x(σ, τ ) · uστ · uρ = x(σ, τ ) · x(στ, ρ) · uστ ρ =
uσ · (uτ · uρ ) = uσ · x(τ ρ) · uτ,ρ = xσ (τ, ρ) · uσ · uτ ρ
= xσ (τ, ρ) · x(σ, τ ρ) · uστ ρ .
It follows that
x(σ, τ ) · x(στ, ρ) = xσ (τ, ρ) · x(σ, τ ρ),
which is precisely the cocycle property.
The above analysis makes use of the choice of a system of representatives uσ .
Given a different system of representatives u0σ , we get from the equation
u0σ · u0τ = x0 (σ, τ ) · u0στ
another factor system x0 (σ, τ ). However, this system differs from x(σ, τ ) only
by a 2-coboundary, namely by the 2-coboundary ∂2 (u0σ · u−1 σ ), as is easily
verified. Since the group G is completely determined by the relations (2) and
b
(3) (and by the relations in the groups A and G), this allows us to conclude:
The solution Gb of the group extension problem is uniquely determined by the
action of the group G on A, and a class of equivalent factor systems x(σ, τ ),
i.e., a cohomology class in H 2 (G, A).
Conversely, if we have made the group A into a G-module in any possible
way9) , and we are given a class c ∈ H 2 (G, A), we always obtain a solution of
the extension problem: For the σ ∈ G choose generators uσ and define G b as
the group generated by the elements of A and the uσ , subject to the relations
aσ = uσ · a · u−1
σ and uσ · uτ = x(σ, τ ) · uστ (x(σ, τ ) a 2-cocycle in c).
Again this can be verified very easily.
9)
More precisely: Recall that for a G-module A every element σ ∈ G provides an
automorphism of the group A. Thus a G-module A is precisely a pair of groups
A and G, together with a homomorphism h : G → Aut(A); the action of σ ∈ G
on A is given by σa = h(σ)a. Thus making the group A into a G-module simply
means choosing a homomorphism from G to the automorphism group Aut(A).
Together with the groups H q (G, A) of dimension q = −1, 0, 1, 2, the cohomol-

ogy group H −2 (G, ZZ) with coefficients in ZZ plays a special role. We will show
later that it is canonically isomorphic to the abelianization Gab = G/G0
(where G0 is the commutator subgroup) of G, a fact which is heavily used in
class field theory. The main theorem of class field theory concerns an isomor-
phism between the abelianization Gab and the norm residue group AG /NG A
of a particular G-module A. It can be formulated in purely cohomological
terms as H −2 (G, ZZ) = H 0 (G, A), and can be proven abstractly given certain
assumptions (cf. (7.3)).
§ 3. The Exact Cohomology Sequence
After having introduced the cohomology groups H q (G, A), we now want to
study how these groups behave in case either the module A or the group G
changes. We will discuss the first case in this section.
If A and B are two G-modules and

f : A −→ B
is a G-homomorphism, then f canonically induces a homomorphism
f¯q : H q (G, A) −→ H q (G, B),
which arises in the following way. From the map
x(σ1 , . . . , σq ) 7−→ f x(σ1 , . . . , σq )
we get a homomorphism
fq : Aq −→ Bq
between the groups of cochains Aq of A and Bq of B with the property that
∂q+1 ◦ fq = fq+1 ◦ ∂q+1 .
Therefore these maps fit into the infinite commutative diagram
· · ·
∂q+1
Aq Aq+1 ···
fq fq+1
∂q+1
··· Bq Bq+1 ···
which means precisely that
x(σ1 , . . . , σq ) 7−→ f x(σ1 , . . . , σq )
takes cocycles to cocycles and coboundaries to coboundaries. It follows that
the homomorphism fq : Aq → Bq induces canonically a homomorphism
f¯q : H q (G, A) −→ H q (G, B).
If c ∈ H q (G, A), the image f¯q c is obtained by choosing a cocycle x from the
class c, and taking the cohomology class of the cocycle f x of the module B.
§ 3. The Exact Cohomology Sequence 21
We thus have a very simple explicit description of the homomorphism f¯q .

This is an advantage that does not occur very often in cohomology theory. In
fact, for many cohomological maps one knows only of their existence and their
functorial properties, without having an explicit description. Nevertheless, it
is equally significant that in the entire theory one almost always only works
with these functorial properties, and an explicit description of the maps is
required only in a few cases.
A first example of this typical situation is provided by the for the entire
theory fundamental connecting homomorphism δ. Although this map is
still given explicitly, its definition does not exactly leave the impression of
great clarity and immediacy.
(3.1) Proposition. If
i j
0 −→ A −→ B −→ C −→ 0
is an exact sequence of G-modules and G-homomorphisms, then there exists
a canonical homomorphism
δq : H q (G, C) −→ H q+1 (G, A).
The map δq is called the connecting homomorphism or also the δ-
homomorphism.
For the construction of δq consider the following commutative diagram
010/.-+,'()#$%*&"! i j
Aq−1 Bq−1 Cq−1 0
∂ ∂ ∂
i j
0 Aq Bq Cq 0
∂ ∂ ∂
i j
0 Aq+1 Bq+1 Cq+1 0.
(For simplicity we have omitted the indices on the maps i, j, ∂.) The rows in
this diagram result from applying the functor HomG (Xi , ) (i = q−1, q, q+1)
to the exact sequence 0 → A → B → C → 0; since the G-modules Xi are free
(cf. (1.6)), it follows that these rows are exact.
We denote by aq , bq , cq elements of the groups of cochains Aq , Bq , Cq , and

write aq , bq , cq for their images in the cohomology groups H q (G, A), H q (G, B),
H q (G, C).
Assume now cq ∈ H q (G, C), so that ∂cq = 0. We choose a bq such that

cq = jbq .
We have j∂bq = ∂jbq = ∂cq = 0, and thus ∂bq ∈ ker jq+1 . Hence there exists a
aq+1 with ∂bq = iaq+1 . Because i∂aq+1 = ∂iaq+1 = ∂∂bq = 0, i.e., ∂aq+1 = 0,
we see that aq+1 is a (q + 1)-cocycle of A. Now we set
δq cq = aq+1 .
This definition depends, of course, on the choice of the representative cq of
cq and its preimage bq . However, if we choose another representative c0q with
preimage b0q (i.e., jb0q = c0q ) and let a0 q+1 be the resulting class, we have
cq = c0 q ⇒ cq − c0q = ∂cq−1 for a cq−1 ⇒ cq − c0q = ∂jbq−1 for a bq−1

⇒ jbq −jb0q = j∂bq−1 ⇒ bq −b0q −∂bq−1 ∈ ker jq = im iq ⇒ iaq = bq −b0q −∂bq−1
for an aq ⇒ ∂iaq = ∂bq − ∂b0q ⇒ i∂aq = iaq+1 − ia0q+1 ⇒ ∂aq = aq+1 − a0q+1 ⇒
aq+1 = a0 q+1 .
Therefore δq is well defined; it is immediate that δ is a homomorphism.
As already mentioned, it is not necessary to recall the explicit definition of

the connecting homomorphism δ every time this map comes up. Once we
have proved the fundamental property of this map, its explicit definition is
used only occasionally. This fundamental and most important property is
given by the following theorem, which can be considered the main theorem of
cohomology theory.
(3.2) Theorem. Let

i j
0 −→ A −→ B −→ C −→ 0
be an exact sequence of G-modules and G-homomorphisms. Then the induced
infinite sequence
iq jq δq
· · · −→ H q (G, A) −→ H q (G, B) −→ H q (G, C) −→ H q+1 (G, A) −→ · · ·
is also exact. It is called the exact cohomology sequence.
Proof. The homomorphisms iq , j q and δq are respectively induced by:

aq 7→ iaq , bq 7→ jbq , cq 7→ aq+1 ,
where cq = jbq and ∂bq = iaq+1 . It follows that
j q ◦ iq = 0, because aq 7→ iaq 7→ jiaq = 0,
δq ◦ j q = 0, because bq 7→ jbq 7→ aq+1 = 0 (we have iaq+1 = ∂bq = 0),
iq+1 ◦ δq = 0, because cq 7→ aq+1 7→ iaq+1 = ∂bq ∈ ∂Bq .
From this we obtain the inclusions

im iq ⊆ ker j q , im j q ⊆ ker δq , im δq ⊆ ker iq+1 .
Let bq ∈ ker j q so that jbq = ∂cq−1 for a cq−1 . If we choose a bq−1 with
jbq−1 = cq−1 , it follows that j(bq − ∂bq−1 ) = 0. Hence we may assume from
the beginning that the representative bq of bq satisfies jbq = 0. Then there
exists an aq with bq = iaq , and this aq is a cocycle since i∂aq = ∂bq = 0.
Hence bq = iq aq ∈ im iq , which proves the inclusion im iq ⊇ ker j q .
Let cq ∈ ker δq . By definition of δq , there are elements aq+1 and bq such that
δq cq = aq+1 = 0, iaq+1 = ∂bq , and cq = jbq . Because aq+1 = 0 we have
aq+1 = ∂aq , which implies that ∂(bq − iaq ) = 0 and cq = j(bq − iaq ). From
this we obtain cq = j q (bq − iaq ), which shows that im j q ⊇ ker δq .
Let aq+1 ∈ ker iq+1 so that iaq+1 = ∂bq for some bq . If we let cq = jbq , then
∂cq = ∂jbq = j∂bq = jiaq+1 = 0. This shows that cq is a cocycle and implies
aq+1 = δq cq ∈ im δq . It follows that im δq ⊇ ker iq+1 , which completes the
proof of the exactness of the cohomology sequence.
When we introduced the cohomology groups we already mentioned that work-

ing with a complete free resolution of G leads to a unification of homology and
cohomology groups. The essential aspect here is not so much to have a unified
notation but rather the existence of an exact sequence ranging from −∞ to
+∞ that involves both the homology as well as the cohomology groups.
Theorem (3.2) is applied mostly frequently in the following form: If an arbi-
trary term in the exact cohomology sequence
· · · −→ H q (G, A) −→ H q (G, B) −→ H q (G, C) −→ H q+1 (G, A) −→ · · ·
vanishes, then the preceding map is surjective and the subsequent map is
injective. This type of argument will allow us frequently to prove important
isomorphism results. We state this in the following corollary.
(3.3) Corollary. Let

i j
0 −→ A −→ B −→ C −→ 0
be an exact sequence of G-modules. If H q (G, A) = 0 for all q, then j q :
H q (G, B) −→ H q (G, C) is an isomorphism. Similarly, if H q (G, B) = 0
(resp. H q (G, C) = 0) for all q, then δq : H q (G, C) −→ H q+1 (G, A) (resp. iq :
H q (G, A) −→ H q (G, B)) is an isomorphism.
Because of this corollary the G-modules which have only trivial cohomology
groups play a distinguished role.
δ
We continue our discussion of the connecting map AG → H 1 (G, A) from p. 17
and show that the cohomology sequence (3.2) induces an exact sequence which
terminates on the left.
(3.4) Theorem. Let

i j
0 −→ A −→ B −→ C −→ 0
be an exact sequence of G-modules. Then the following sequence is exact:
i j δ i j
0 −→ AG −→ B G −→ C G −→ H 1 (G, A) −→
1
H 1 (G, B) −→
1
··· .
δ
Proof. The homomorphism C G → H 1 (G, A) is defined as the composition
δ
C G −→ C G /NG C = H 0 (G, C) −→
0
H 1 (G, A);
hence it suffices to show exactness at the term C G . To show im j ⊆ ker δ
assume c ∈ im j ⊆ C G . Then c = jb with b ∈ B G , and the claim follows from
δc = δ0 (c + NG C) = δ0 (jb + NG C) = δ0 j 0 (b + NG B) = 0.
For the opposite inclusion ker δ ⊆ im j, consider c ∈ ker δ. Then c ∈ C G and

δc = δ0 (c + NG C) = 0. It follows from (3.2) that we have the identity
c + NG C = j 0 (b + NG B) = jb + NG C,
hence c = jb + NG c0 . If we choose a b0 ∈ B with jb0 = c0 , it follows that
c = jb + NG (jb0 ) = jb + jNG b0 ∈ jB G , which proves our claim.
Because of the exact sequence (3.4) the fixed modules AG , B G , and C G are
often defined as the zeroth cohomology groups; in particular, in case one is
only interested in cohomology groups of positive dimension.
In the following we consider various compatibility properties of the connecting

homomorphism δ.
(3.5) Proposition. If
0<;9:5674238 i j
A B C 0
f g h
0 i0 0 j0
0 A B C0 0
is a commutative diagram of G-modules and G-homomorphisms with exact
rows, then
f¯q+1 ◦ δq = δq ◦ h̄q ;
in other words, the following diagram is commutative:
@?>= C)
H q (G,
δq
H q+1 (G, A)
h̄q f¯q+1
δq
H q (G, C 0 ) H q+1 (G, A0 ).
This follows almost immediately from the definition of δq . Let cq ∈ H q (G, C).
If we choose bq and aq+1 such that cq = jbq and iaq+1 = ∂bq , then δq cq = aq+1 ,
so that (f¯q+1 ◦ δq )cq = f¯q+1 (aq+1 ) = f aq+1 . If we set c0q = hcq , b0q = gbq and
a0q+1 = f aq+1 , it follows that c0q = j 0 b0q and ∂b0q = i0 a0q+1 , from which we obtain
(δq ◦ h̄q )cq = δq c0 q = a0 q+1 = f aq+1 = f¯q+1 ◦ δq )cq , i.e., f¯q+1 ◦ δq = δq ◦ h̄q .
The connecting homomorphism δ is “anticommutative”:
(3.6) Theorem. Assume the diagram of G-modules and G-homomorphisms

ABCDEFGHIJKLMNOPQRSTUVWX 0 0 0
0 A0 A A00 0
0 B0 B B 00 0
0 C0 C C 00 0
0 0 0
is commutative with exact rows and columns. Then the diagram
\[ZY C 00 )
H q−1 (G, δ
H q (G, C 0 )
δ −δ
δ
H q (G, A00 ) H q+1 (G, A0 )
commutes.
Proof. Let D be the kernel of the composite map B → C 00 ; thus the sequence
0 −→ D −→ B −→ C 00 −→ 0
is exact. We define G-homomorphisms
i : A0 → A ⊕ B 0 by i(a0 ) = (a, b0 ), where a (resp. b0 ) is the image of a0 in A

(resp. of a0 in B 0 ),
j : A ⊕ B 0 → D by j(a, b0 ) = d1 − d2 , where d1 (resp. d2 ) is the image of a
(resp. of b0 ) in D ⊆ B.
It is easy to verify that with these definitions the sequence
i j
0 −→ A0 −→ A ⊕ B 0 −→ D −→ 0
is exact, and that the diagram
Apqronklmjcdefghib]^_`a 0 A A00 B 00 C 00
id (id,0) id
i j
A0 A ⊕ B0 D B C 00
−id (0,−id) id
A0 B0 C0 C C 00
commutes. Because im (D → B 00 ) ⊆ im (A00 → B 00 ) and A00 → B 00 is injective,
there is a G-homomorphism D → A00 which extends the above diagram. Sim-
ilarly, since im (D → C) ⊆ im (C 0 → C) and C 0 → C is injective, there is an
analogous G-homomorphism D → C 0 . Since the resulting extended diagram
is commutative, it follows from (3.5) that the following diagram
xwsz{yvut}~| C 00 )
H q−1 (G, δ
H q (G, A00 ) δ
H q+1 (G, A0 )
id id
δ δ
H q−1 (G, C 00 ) H q (G, D) H q+1 (G, A0 )
id −id
δ δ
H q−1 (G, C 00 ) H q (G, C 0 ) H q+1 (G, A0 ),
commutes as well, which immediately implies the theorem.
(3.7) Proposition. Let {Aι | ι ∈ I} be a family of G-modules. Then

Aι ∼
M M
H q G, = H q (G, Aι ).
ι ι
L
Proof. If we set A = Aι , we have from (1.5) a canonical isomorphism
ι
Aq = HomG (Xq , A) ∼
M M
= HomG (Xq , Aι ) = (Aι )q ,
ι ι
and the proposition follows from the infinite commutative diagram
· · · Aq−1 ∂
Aq ···
M ∂
M
··· (Aι )q−1 (Aι )q ··· .
ι ι

Q
The same
L reasoning applies to the direct product ι Aι in place of the direct
sum ι Aι , using the evident isomorphism
Aι ∼
Y Y Y Y
Aι q = HomG Xq , = HomG (Xq , Aι ) = (Aι )q .
ι ι ι ι
Therefore we also have
(3.8) Proposition. H q (G, Aι ) ∼

= ι H q (G, Aι )
Q Q
ι
The G-induced modules. We explained already in (3.3) that the exact co-
homology sequence yields isomorphism theorems in case the underlying exact
sequence contains a G-module with only trivial cohomology groups. A par-
ticular class of such G-modules are the G-induced modules, which we will
make use of in many of the proofs and definitions below.
(3.9) Definition. A G-module A is called G-induced if it can be represented

as a direct sum M
A= σD
σ∈G
with a subgroup D ⊆ A.
L
In particular, the G-module ZZ[G] = σ∈G σ(ZZ·1) is G-induced, and it is clear
that the G-induced modules are represented simply as the tensor products
ZZ[G] ⊗ D
with arbitrary abelian groups D. In fact, if we consider D as a trivial G-
module, we have the G-isomorphism
M M M
ZZ[G] ⊗ D = ZZσ ⊗ D = ZZ(σ ⊗ D) = σ(ZZ ⊗ D).
More generally, we have
(3.10) Proposition. Let X be a G-induced module and A an arbitrary G-

module. Then the module X ⊗ A is also G-induced.
L
By assumption X = σ∈G σD, which implies
σD ⊗ A ∼ (σD) ⊗ (σA) ∼
M M M
X ⊗A= = = σ(D ⊗ A).
(3.11) Proposition. Let A be a G-induced module, and g a subgroup of G.

Then A is a g-induced g-module. If g is normal in G, then Ag is a G/g-induced
G/g-module.

L
Proof. Let A = σD. Then we can write
σ∈G
MM M M
A= στ D = σ( τ D),
σ∈g τ σ∈g τ
where τ ranges over a system of right coset representatives of G with respect

to g. Hence A is g-induced.
Assume g is normal in G. We show the G/g-module Ag has the representation
M
Ag = τ Ng D.
τ ∈G/g
L
Because of the direct sum decomposition A = σ∈G σD, the sum on the
right side of the above identity is obviously direct. Since Ng D ⊆ Ag , it is also
contained in Aq . Conversely,
P suppose a ∈ Ag . The element a has a unique
representation as a = τ ∈G τ dτ with dτ ∈ D. If σ ∈ g, it follows that
X X
a = σa = στ dτ = στ dστ ,
τ ∈G τ ∈G
and by uniqueness that dτ = dστ . From this we obtain the representation
XX X X X
a= τ σdτ σ = τ( σdτ ) = τ Ng (dτ ),
τ σ∈g τ σ∈g τ
where τ ranges over a system of left coset representatives of G/g. This proves
that Ag is G/g-induced.
(3.12) Definition. We say that a G-module A has trivial cohomology if

H q (g, A) = 0
for all q and all subgroups g ⊆ G.
The next theorem is crucial:
(3.13) Theorem. Every G-induced module A has trivial cohomology.
Proof. Because of (3.11) it suffices to prove H q (G, A) = 0, i.e., the sequence

∂
· · · −→ HomG (Xq , A) −→ HomG (Xq+1 , A) −→ · · ·
L
is exact. Let A = σ∈G σD, and let π : A → D is the natural projection of A
onto D. Then the map f 7→ π◦f induces an obviously bijective homomorphism
HomG (Xq , A) −→ Hom(Xq , D),
and we may identify HomG (Xq , A) with Hom(Xq , D). Therefore it suffices to
consider the sequence
· · · −→ Hom(Xq , D) −→ Hom(Xq+1 , D) −→ · · · ,
which is exact by (1.7).
Because G-induced modules have trivial cohomology, we obtain from (3.3)

a very important potential application of G-induced modules, using the fact
that every G-module A can be viewed as a submodule as well as a factor
module of a G-induced module, as we explain next.
As before, we denote by IG the augmentation ideal of ZZ[G], and by JG the
factor module JG = ZZ[G]/ZZ·NG . By (1.2) all terms in the exact sequences
ε
0 −→ IG −→ ZZ[G] −→ ZZ −→ 0 ,
µ
0 −→ ZZ −→ ZZ[G] −→ JG −→ 0 .
consist of free ZZ-modules. From (1.8) we immediately conclude
(3.14) Proposition. For all G-modules A we have the exact sequences

0 −→ IG ⊗ A −→ ZZ[G] ⊗ A −→ A −→ 0 ,
0 −→ A −→ ZZ[G] ⊗ A −→ JG ⊗ A −→ 0 .
Since the G-module ZZ[G] ⊗ A is G-induced by (3.10), these exact sequences

allow us to view A as a submodule as well as a factor module of a G-induced
module.
Because ZZ[G] ⊗ A is cohomologically trivial, applying the exact cohomology

sequence to the exact sequences from Proposition (3.14) yields isomorphisms
δ : H q−1 (g, A1 ) −→ H q (g, A) with A1 = JG ⊗ A,
δ −1 : H q+1 (g, A−1 ) −→ H q (g, A) with A−1 = IG ⊗ A
for every q and every subgroup g ⊆ G; this uses (3.3). We want to iterate this:
For every integer m ∈ ZZ set

Am = JG ⊗ · · · ⊗ JG ⊗A, if m ≥ 0,
| {z }
m-times
Am = IG ⊗ · · · ⊗ IG ⊗A, if m ≤ 0.
| {z }
|m|-times
From the composition of the isomorphism δ (resp. δ −1 ) we obtain maps

H q−m (g, Am ) −→ H q−(m−1) (g, Am−1 ) −→ · · · −→ H q (g, A)
and thus isomorphisms
δ m : H q−m (g, Am ) −→ H q (g, A) (m ∈ ZZ).
This shows:
(3.15) Proposition. Let A be a G-module. There are G-modules

Am = JG ⊗ · · · ⊗ JG ⊗ A (m ≥ 0), resp.
Am = I G ⊗ · · · ⊗ I G ⊗ A (m ≤ 0)
with the property that the m-fold composition of the connecting homomor-
phism δ induces for every q and every subgroup g ⊆ G an isomorphism
δ m : H q−m (g, Am ) −→ H q (g, A) (m ∈ ZZ)
We will frequently use the isomorphism H q (g, A) =∼ H q−m (g, Am ) to deduce

from statements about cohomology groups of dimension q analogous state-
ments for a higher or lower dimension. In particular, this technique will allow
us to reduce many definitions and proofs to the case of zero-dimensional coho-
mology groups, which we understand much better that the higher-dimensional
analogues. This method is called the method of dimension shifting10) .
The following theorem gives a first example of the usefulness of this method:
(3.16) Theorem. Let A be a G-module. Then H q (G, A) is a torsion groups

and the orders of the elements in H q (G, A) divide the order n of the group G
n · H q (G, A) = 0.
Proof. If q = 0, then n · H 0 (G, A) = 0, because H 0 (G, A) = AG /NG A

and na = NG a for all a ∈ AG . The general case follows from this and the
isomorphism H q (G, A) ∼
= H 0 (G, Aq ).
(3.17) Corollary. A uniquely divisible 11) G-module A has trivial cohomol-

ogy.
If A is divisible, the map ‘multiplication by n’, i.e., n · id : A → A, is bijective

for every natural number n, and therefore induces isomorphisms
n · id : H q (g, A) −→ H q (g, A) (g ⊆ G).
If n = |G|, it follows from (3.16) that H (g, A) = n · H q (g, A) = 0.
q
10)
Cohomology theory can be based directly on this principle. In fact, by (3.15)
∼ H 0 (G, Aq ),
H q (G, A) =
where Aq is given canonically by A: Aq = JG ⊗ . . . ⊗ JG ⊗ A for q ≥ 0, resp.
Aq = IG ⊗ . . . ⊗ IG ⊗ A for q ≤ 0. Therefore one may define the cohomology
groups of the G-module A from the beginning as the quotient group
H q (G, A) = (Aq )G /NG Aq .
For cohomology theory developed along these lines, see C. Chevalley [12].
11)
An abelian group A is said to be uniquely divisible if for every a ∈ A and every
natural number n the equation nx = a has a unique solution x ∈ A.
In particular, the G-module Q (on which the group G always acts trivially)
has trivial cohomology. From the cohomology sequence associated with the
exact sequence
0 −→ ZZ −→ Q −→ Q/ZZ −→ 0
we obtain
(3.18) Corollary. There is a canonical isomorphism

H 2 (G, ZZ) ∼
= H 1 (G, Q/ZZ) = Hom(G, Q/ZZ) = χ(G).
The group χ(G) = Hom(G, Q/ZZ) is called the character group of G.
We end this section with the computation of the group H −2 (G, ZZ), which
plays an important role in class field theory. We denote the commutator sub-
group of G by G0 , and its abelianization by Gab = G/G0 .
(3.19) Theorem. There is a canonical isomorphism H −2 (G, ZZ) ∼

= Gab .
Proof. Since ZZ[G] is a G-induced module, it has trivial cohomology, and we

obtain from the exact cohomology sequence associated with
ε
0 −→ IG −→ ZZ[G] −→ ZZ −→ 0
the isomorphism
δ : H −2 (G, ZZ) −→ H −1 (G, IG )
Since H −1 (G, IG ) = IG /IG2
it suffices to produce an isomorphism G/G0 ∼ =
2
IG /IG . (Note that G is written multiplicatively and IG is written additively.)
For this we consider the map
2 2
G −→ IG /IG , σ 7−→ (σ − 1) + IG .
Because σ · τ − 1 = (σ − 1) + (τ − 1) + (σ − 1) · (τ − 1), this map is a homo-
2
morphism. Since IG /IG is abelian, the kernel of this homomorphism contains
the commutator subgroup G0 , which implies that we have a homomorphism
log : G/G0 −→ IG /IG
2
.
In order to show that the map log is bijective, we use that IG is the free
abelian group generated by σ − 1, where σ ∈ G r {1}. Hence setting
σ − 1 7−→ σ · G0
defines an evidently surjective homomorphism from IG to G/G0 . Because
(σ − 1) · (τ − 1) = (στ − 1) − (σ − 1) − (τ − 1) 7−→ στ σ −1 τ −1 G0 = 1̄ ,
2
the elements in IG lie in the kernel, so that we obtain a homomorphism
2
exp : IG /IG −→ G/G0 , (σ − 1) + IG
2
7−→ σG0
with the property that log ◦ exp = id and exp ◦ log = id. Therefore the map
log : G/G0 → IG /IG
2
is an isomorphism.
It is easy to see that for the G-module ZZ we have H −1 (G, ZZ) = NG ZZ/IG ZZ = 0,
H 0 (G, ZZ) = ZZ/nZZ, and H 1 (G, ZZ) = Hom(G, ZZ) = 0. Thus we have deter-
mined the cohomology groups H q (G, ZZ) of dimensions q = −2, −1, 0, 1, 2:
H −2 (G, ZZ) ∼
= Gab , H −1 (G, ZZ) = 0, H 0 (G, ZZ) = ZZ/nZZ,
H 1 (G, ZZ) = 0, H 2 (G, ZZ) = χ(G).
We mention without proof that there is a canonical isomorphism (by duality)

H −q (G, ZZ) ∼
= χ(H q (G, ZZ)) for all q > 0.
§ 4. Inflation, Restriction and Corestriction
In the previous section we studied the dependence of the cohomology groups

H q (G, A) on the module A; now we consider the behavior of these groups in
case the group G varies. This mainly concerns the following questions.
Let A be a G-module and let g be a subgroup of G. Then A is always a g-
module and Ag is a G/g-module, provided g is normal in G. What are the
relations among the cohomology groups
H q (G/g, Ag ), H q (G, A) and H q (g, A) ?
We first restrict our considerations to the case of positive dimension q ≥ 1.
If g is normal in G, we associate with every q-cochain
x : G/g × · · · × G/g −→ Ag
a q-cochain
y : G × · · · × G −→ A
by defining y(σ1 , . . . , σq ) = x(σ1 · g, . . . , σq · g) . We call this y the inflation
of x and denote it by
y = inf x.
It is easy to see that the map x 7→ inf x is compatible with the coboundary
operator ∂, i.e., ∂q+1 ◦ inf = inf ◦ ∂q+1 . Hence cocycles are mapped to cocycles
and coboundaries to coboundaries, and we obtain a map on cohomology:
(4.1) Definition. Let A be a G-module and g a normal subgroup of G. The

homomorphism
inf q : H q (G/g, Ag ) −→ H q (G, A), q ≥ 1,
induced by the homomorphism from the q-th group of cochains of the G/g-
module Ag to the q-th group of cochains of the G-module A is called inflation.
§ 4. Inflation, Restriction and Corestriction 33
Along with inflation we obtain another cohomological map by associating with

every q-cochain
x : G × · · · × G −→ A
its restriction
y : g × · · · × g −→ A
from G × · · · × G to g × · · · × g. We call this q-cochain y the restriction of x
and denote it by
y = res x.
The crucial point here is again that the cochain homomorphism res commutes
with the operator ∂, i.e., ∂q+1 ◦ res = res ◦ ∂q+1 , hence there is an induced
map on cohomology groups:
(4.2) Definition. Let A be a G-module and g a subgroup of G. The homo-

morphism
resq : H q (G, A) −→ H q (g, A), q ≥ 1,
induced by the restriction of the cochains of the G-module A to the group g
is called restriction.
To make sure the cohomological maps defined above fit into the general theory,
we have to verify that they are compatible with the canonical homomorphisms
already given. This is the content of the following propositions.
(4.3) Proposition. Let A and B be two G-modules, g a normal subgroup

of G, and
f : A −→ B
a G-homomorphism. Then the following diagrams
Ag ) f¯ A) f¯
H q (G/g, H q (G/g, B g ) H q (G, H q (G, B)
inf q inf q resq resq
f¯ f¯
H q (G, A) H q (G, B), H q (g, A) H q (g, B)
are commutative. In the second diagram the normality of g in G is not needed.
Note here that the G-homomorphism f : A → B induces a G/g-homomor-

phism f : Ag → B g as well as a g-homomorphism f : A → B .
(4.4) Proposition. Let

0 −→ A −→ B −→ C −→ 0
be an exact sequence of G-modules and G-homomorphisms, and let g be a
normal subgroup of G. If the sequence
0 −→ Ag −→ B g −→ C g −→ 0
is also exact, then the diagram
C g )
H q (G/g, δ
H q+1 (G/g, Ag )
inf q inf q+1
δ
H q (G, C) H q+1 (G, A)
commutes.
(4.5) Proposition. Let

0 −→ A −→ B −→ C −→ 0
be an exact sequence of G-modules and G-homomorphisms, and let g be a
subgroup of G. Then the diagram
C)
H q (G, δ
H q+1 (G, A)
resq resq+1
δ
H q (g, C) H q+1 (g, A)
commutes.
Propositions (4.3), (4.4) and (4.5) are easy to verify. The proof of the last two
statements follows essentially from the fact that the inflation and restriction
maps commute with the operator ∂, together with the definition of δ. We
leave the details to the reader.
If we compose inflation and restriction, we obtain the following relation:
(4.6) Theorem. Let A be a G-module and g a normal subgroup of G. Then

inf res
0 −→ H 1 (G/g, Ag ) −−→ H 1 (G, A) −−→ H 1 (g, A)
is exact.
Proof. To show the inflation map is injective, let x : G/g → Ag be a 1-cocycle

whose inflation inf x is a 1-coboundary of the G-module A. Then
inf x(σ) = x(σ · g) = σa − a, a ∈ A.
Hence we have for all τ ∈ g the equation σa − a = στ a − a, i.e., a = τ a which
implies a ∈ Ag . Therefore x(σg) = σ · ga − a is a 1-coboundary.
In order to prove the exactness at the term H 1 (G, A), consider a 1-cocyle
x : G/g → Ag of Ag . If σ ∈ g, it follows that
res ◦ inf x(σ) = inf x(σ) = x(σg) = x(g) = x(1).
But now x(1) = x(1·1) = x(1)+x(1) = 0 which implies res◦inf = 0. Therefore
im inf ⊆ ker res.
Conversely, let x : G → A be a 1-cocycle of the G-module A whose restriction

to g is a 1-coboundary of the g-module A:
x(τ ) = τ a − a, a ∈ A, for all τ ∈ g.
If we subtract from x the 1-coboundary ρ : G → A, ρ(σ) = σa − a, σ ∈ G,
we obtain a 1-cocycle x0 (σ) = x(σ) − ρ(σ) in the same cohomology class with
x0 (τ ) = 0 for all τ ∈ g. Then
x0 (σ − τ ) = x0 (σ) + σx0 (τ ) = x0 (σ) for all τ ∈ g,
and, on the other hand,
x0 (τ · σ) = x0 (τ ) + τ x0 (σ) = τ x0 (σ) for all τ ∈ g.
0
If we now define y : G/g → A by y(σ ·g) = x (σ), we have y(σ ·g) ∈ Ag because
y(σ · g) = y(τ σ · g) for all τ ∈ g, and y defines a 1-cocycle with inf y = x0 .
This proves that ker res ⊆ im inf.
The analogue of Theorem (4.6) for higher dimensions holds only under certain
conditions:
(4.7) Theorem. Let A be a G-module and g a normal subgroup of G. If

H i (g, A) = 0 for i = 1, . . . , q − 1 and q ≥ 1, then the sequence
inf res
0 −→ H q (G/g, Ag ) −−→ H q (G, A) −−→ H q (g, A)
is exact.
We prove this by induction on the dimension q, using dimension shifting (cf.

§3.) and Theorem (4.6) as the initial induction step. If we set B = ZZ[G] ⊗ A
and C = JG ⊗ A, we have from (3.14) the exact sequence
0 −→ A −→ B −→ C −→ 0.
Because H 1 (g, A) = 0, it follows from Theorem (3.4) that the sequence

0 −→ Ag −→ B g −→ C g −→ 0
is also exact. Hence we have the following commutative diagram
0 H q−1 (G/g, C g ) inf H q−1 (G, C) res H q−1 (g, C)
δ δ δ
inf res
0 H q (G/g, Ag ) H q (G, A) H q (g, A).
Since B is G-induced and g-induced, and B g is G/g-induced (cf. (3.10) and
(3.11)), the connecting maps δ are isomorphisms (cf. (3.3)). It follows that
H i (g, C) ∼
= H i+1 (g, A) = 0 for i = 1, . . . , q − 2.
Hence if we assume by induction that the upper sequence in the above diagram
is exact, then this also holds for the lower sequence.
Of course, one will ask why we have introduced the maps inf and res only in
case of positive dimensions q ≥ 1 instead of defining such maps analogously
using cochains for all negative dimensions as well. However, this is not possible.
In fact, the crucial property of inflation and restriction is that by (4.4) and
(4.5) respectively they are transformed into one another through dimension
shifting by the operator δ, and one would want this property to hold for a
general definition in all dimensions as well. Now for inflation it is necessary
to restrict to q ≥ 1; this follows essentially because if 0 → A → B → C → 0
is an exact sequence of G-modules, then 0 → Ag → B g → C g → 0 (g ⊆ G)
is in general not exact, i.e., the first sequence induces a δ-homomorphism on
cohomology groups, but the second does not.
The situation is different for restriction, however. In fact, this map can be
extended to all dimensions q ≤ 0. For example, if q = 0, one obtains from
a + NG A 7−→ a + Ng A, a ∈ AG ⊆ Ag ,
a homomorphism
res0 : H 0 (G, A) = AG /NG A −→ H 0 (g, A) = Ag /Ng A,
with the property that Proposition (4.5) remains valid for q = 0. We pin this
down in the following lemma:
i j
(4.8) Lemma. Let 0 → A → B → C → 0 be an exact sequence of G-modules,
and let g be a subgroup of G. Then the following diagram commutes
£¢¡ C)
H 0 (G, δ
H 1 (G, A)
res0 res1
δ
H 0 (g, C) H 1 (g, A).
Proof. Let c ∈ C G be a 0-cocycle of the G-module C and c = c + NG C its

cohomology class. Then res0 c = c + Ng C, i.e., c is also a 0-cocycle for the
g-module C. If we choose b ∈ B with jb = c, then j∂b = ∂c = 0 implies that
there exists a 1-cocycle a1 : G → A such that ia1 = ∂b. By definition of δ we
have δc = a1 , and therefore δres0 c = res1 a1 = res1 a1 = res1 δc.
Unfortunately, a similarly elementary definition of the restriction maps resq

in dimensions q < 0 cannot be given. Nevertheless, we will see that if such
a restriction is specified for a single dimension, say q = 0, the compatibility
condition in (4.5) uniquely determines all other restriction maps. This leads
us toward an axiomatic approach to restriction as follows.
(4.9) Definition. Let G be a finite group and g a subgroup of G. Then

restriction is the uniquely determined family of homomorphisms
resq : H q (G, A) −→ H q (g, A), q ∈ ZZ,
with the properties:
(i) If q = 0, then
res0 : H 0 (G, A) −→ H 0 (g, A), a + NG A 7→ a + Ng A (a ∈ AG ).
(ii) For every exact sequence 0 → A → B → C → 0 of G-modules and

G-homomorphisms, the following diagram is commutative
§¦¥¤ C)
H q (G, δ
H q+1 (G, A)
resq resq+1
δ
H q (g, C) H q+1 (g, A)
The restrictions resq are obtained from res0 by dimension shifting as follows:
By (3.15) we have the isomorphisms
δ q : H 0 (G, Aq ) −→ H q (G, A) and δ q : H 0 (g, Aq ) −→ H q (g, A),
given by the q-fold compositions of the connecting homomorphism δ. Condi-
tion (ii) now means that we have to define resq by the commutative diagram
¨©«ª Aq )
H 0 (G, δq
H q (G, A)
res0 resq
q
H 0 (g, Aq ) δ H q (g, A).
This also shows uniqueness of the restriction maps. In particular, the restric-
tions resq for q ≥ 0 defined in this way coincide with those introduced earlier.
It remains to show that the homomorphisms resq satisfy condition (ii). To this
end we consider the following diagram
·¶´³±®¬µ²°¯ C q )
H 0 (G, δ
H 1 (G, Aq )
res res
(−1)q δ q
δq δ
H 0 (g, C q ) H 1 (g, Aq )
δ
H q (G, C) H q+1 (G, A) (−1)q δ q
q
res δ res
δ
H q (g, C) H q+1 (g, A),
where we have used that from the sequence 0 → A → B → C → 0 we obtain

by induction using Proposition (1.2) and Lemma (1.9), the exact sequence
0 −→ Aq −→ B q −→ C q −→ 0.
In the above diagram the upper square commutes by (4.8). The commutativity
of the two side diagrams follows immediately from the definition of the restric-
tion maps by dimension shifting. The back and front diagrams are obtained
by composing q squares of the type (3.6); hence by (3.6) they also commute.
Therefore the commutativity of the upper square implies the commutativity
of the lower square, which completes the proof.
Concerning an explicit description of the homomorphisms resq for q < 0, i.e.,

the question of how the individual cocycles behave under these mappings, we
remark that only by extensive calculations one can achieve some results, which
in turn are hardly useful because they are far too technical. Nevertheless, our
remarks on p. 21 about the general nature of cohomological methods also apply
here. It is essentially the functorial properties of the restriction that come up,
and only in small dimensions, where we have a concrete interpretation of the
cohomology groups, we occasionally have to use an explicit description.
Using the isomorphism from Theorem (3.19), we point out a special case of a
restriction which is important for class field theory:
(4.10) Definition. Let g ⊆ G be a subgroup. The homomorphism

Ver : Gab −→ g ab
induced by the restriction res−2 : H −2 (G, ZZ) −→ H −2 (g, ZZ) is called the
Verlagerung or transfer from G to g.
This canonical homomorphism can also be defined using group theoretic in-
stead of cohomological methods, although this requires some effort and in-
volves quite a bit of formulas. Cf. [16], 14.2.
In addition to restriction, there is another map in the opposite direction

corq : H q (g, A) −→ H q (G, A),
the corestriction. As with restriction, corestriction is completely determined
once it is given for a single dimension. Nevertheless, before giving the general
definition, we explain this map in the two dimensions q = −1 and q = 0:
In case q = −1 we define a homomorphism

cor−1 : H −1 (g, A) −→ H −1 (G, A)
by
a + Ig A 7−→ a + IG A (a ∈ Ng A ⊆ NG A).
In case q = 0, we obtain a homomorphism

cor0 : H 0 (g, A) −→ H 0 (G, A).
by
a + Ng A 7−→ NG/g a + NG A (a ∈ Ag ) .
Here we let NG/g a = σ∈G/g σa ∈ AG for a ∈ Ag , where σ ∈ G/g means
P
that σ ranges over a system of left coset representatives of g in G.
The following lemma is the analogon of (4.8) for the corestriction:
i j
(4.11) Lemma. Let 0 → A → B → C → 0 be an exact sequence of G-
modules. Then the following diagram is commutative
º¹¸» C)
H −1 (g, δ
H 0 (g, A)
cor−1 cor0
δ
H −1 (G, C) H 0 (G, A)
Proof. Let c ∈ Ng C be a (−1)-cocycle for the class c = c + Ig C ∈ H −1 (g, C),

thus c ∈ NG C is a (−1)-cocycle for the class cor−1 c = c + IG C ∈ H −1 (G, C).
If we choose b ∈ B with jb = c, we have j∂b = ∂c = Ng c = 0 which implies
that there exists a 0-cocycle a ∈ Ag with ia = ∂b = Ng b. By definition
δc = a = a + Ng A, therefore cor0 δc = NG/g a + NG A ∈ H 0 (G, A). On the
other hand, δcor−1 c = δ(c + IG C). If we choose the same b ∈ B with jb = c
as above, then ∂b = NG b = NG/g Ng b = NG/g (ia) = i(NG/g a) and we have
δ(c+IG C) = NG/g a+NG A. Therefore cor0 δc = NG/g a+NG A = δ cor−1 c.
Similar to restriction, we define corestriction using an axiomatic approach:
(4.12) Definition. Let G be a finite group, and let g be a subgroup of G.

Then corestriction is the uniquely determined family of homomorphisms
corq : H q (g, A) −→ H q (G, A), q ∈ ZZ,
with the properties:
(i) If q = 0, then
cor0 : H 0 (g, A) −→ H 0 (G, A), a + Ng A 7→ NG/g a + NG A (a ∈ Ag ).
(ii) For every exact sequence 0 −→ A −→ B −→ C −→ 0 of G-modules and

G-homomorphisms, the following diagram is commutative
¿¾½¼ C)
H q (g, δ
H q+1 (g, A)
corq corq+1
δ
H q (G, C) H q+1 (G, A).
Exactly as for the restrictions, the homomorphisms corq arise from the core-
striction cor0 in dimension 0 by dimension shifting:
From (3.15) we have the isomorphisms
δ q : H 0 (G, Aq ) −→ H q (G, A) and δ q : H 0 (g, Aq ) −→ H q (g, A),
and by (ii) the map corq is uniquely determined by the commutative diagram
H 0 (g,ÀÃÁÂ Aq ) δq
H q (g, A)
cor0 corq
q
δ
H 0 (G, Aq ) H q (G, A).
In particular, because of uniqueness and (4.11) we recover the homomorphism

cor−1 introduced on p. 38. The fact that (ii) holds is verified in the same way
as for restriction using the following diagram, together with (4.11) and (3.6),
H −1 (g,ÎÍËÊÉÈÇÆÅÄÌÏ C q+1 ) δ
H 0 (g, Aq+1 )
cor cor
(−1)q+1 δ q+1
δ q+1 δ
H −1 (G, C q+1 ) H 0 (G, Aq+1 )
δ
H q (g, C) H q+1 (g, A) (−1)q+1 δ q+1
q+1
cor δ cor
δ
H q (G, C) H q+1 (G, A).
We remark that one can define the corestrictions for negative dimensions very
easily by a canonical correspondence between cochains, analogously to the
restrictions for positive dimension. However, we will not pursue this further.
In view of (4.10) we now want to prove the following theorem:
(4.13) Theorem. Let g ⊆ G be a subgroup. The homomorphism

κ : g ab −→ Gab
induced by the corestriction cor−2 : H −2 (g, ZZ) −→ H −2 (G, ZZ) coincides with
the canonical homomorphism induced by σg 0 7→ σG0 .
This follows, using the proof of (3.19), from the commutative diagram
ÖÓÒÑÐÕÔ ZZ)
H −2 (g, δ
H −1 (g, Ig ) = Ig /Ig2
log
g ab
cor−2 cor−1 κ
δ log
H −2 (G, ZZ) H −1 (G, IG ) = IG /IG
2
Gab ,
in which, strictly speaking, the map cor−1 : H −1 (g, Ig ) → H −1 (G, IG ) is

the composition of the natural map H −1 (g, Ig ) → H −1 (g, IG ) and cor−1 :
H −1 (g, IG ) → H −1 (G, IG ).
The following relation between restriction and corestriction is important:
(4.14) Theorem. Let g ⊆ G be a subgroup. Then the composition

res cor
H q (G, A) −−→ H q (g, A) −−→ H q (G, A)
is the endomorphism
cor ◦ res = (G : g) · id.
Proof. Consider the case q = 0. If a = a + NG A ∈ H 0 (G, A), a ∈ AG , then

cor0 ◦res0 (a) = cor0 (a+Ng A) = NG/g a+NG A = (G : g)·a+NG A = (G : g)·a.
The general case follows from this by dimension shifting. In fact, the diagram
×ÚÙØ Aq ) cor0 ◦res0 H 0 (G, Aq )
H 0 (G,
δq δq
corq ◦resq
H q (G, A) H q (G, A)
commutes, and since the upper horizontal map is (G : g) · id, it follows that
the same holds for the lower horizontal map, i.e., corq ◦ resq = (G : g) · id.
Because the restriction and corestriction maps res and cor commute with
the connecting homomorphism δ, they also commute with maps induced by
G-homomorphisms:
(4.15) Proposition. If f : A → B is a G-homomorphism of the G-modules

A, B, and g is a subgroup of G, then the following diagram commutes
ßÞÝÜÛà A) f¯
H q (G, H q (G, B)
res cor res cor
f¯
H q (g, A) H q (g, B).
This is clear in case of dimension q = 0, and the general case follows easily
by dimension shifting. In fact, the homomorphism f : A → B induces a
homomorphism f : Aq → B q , and in the following diagram
ãâáæèçåäðïîíìëêé Aq ) f¯
H 0 (G, H 0 (G, B q )
cor cor
res δq res
δq f¯
H 0 (g, Aq ) H 0 (g, B q )
f¯
H q (G, A) H q (G, B) δq
q
cor δ cor
res res
f¯
H q (g, A) H q (g, B)
all vertical squares are commutative. Hence the commutativity of the lower
diagram follows from that of the upper one.
Since the cohomology groups H q (G, A) are abelian torsion groups, they are di-
rect sums of their p-Sylow groups, i.e., the groups H q (G, A)p of all elements
in H q (G, A) of p-power order:
M
H q (G, A) = H q (G, A)p .
p
The group H q (G, A)p is often called the p-primary part of H q (G, A). For
the restriction and corestriction maps on these p-primary parts we have the
following:
(4.16) Theorem. Let A be a G-module, and Gp a p-Sylow subgroup of G.

Then the restriction
res : H q (G, A)p −→ H q (Gp , A)
is injective, and the corestriction
cor : H q (Gp , A) −→ H q (G, A)p
is surjective.
Proof. Since cor ◦ res = (G : Gp ) · id, and since (G : Gp ) and p are relatively
cor ◦ res
prime, the mapping H q (G, A)p −−−−−→ H q (G, A)p is an automorphism. Hence
q
if x ∈ H (G, A)p and res x = 0, it follows immediately from cor ◦ res x = 0
that x = 0, which shows the injectivity of res on H q (G, A)p .
On the other hand, H q (Gp , A) consists of elements whose order is a p-power
(cf. (3.16)), so that cor H q (Gp , A) ⊆ H q (G, A)p . Since cor ◦ res is a bijection
on H q (G, A)p , this inclusion is an equality.
We often encounter the problem that we want to show that certain cohomology
groups vanish. In many of these cases we will use the following consequence
of Theorem (4.16), which reduces this problem to the case of p-groups:
(4.17) Corollary. If for every prime p the group H q (Gp , A) = 0 for a p-Sylow
subgroup Gp of G, then we have H q (G, A) = 0.
Proof. Since res : H q (G, A)p → H q (Gp , A) is injective, the assumption im-
plies that all p-Sylow groups H q (G, A)p are trivial; thus H q (G, A) = 0.
We end this section with a generalization of the notion of a G-induced module;

we will use this type of G-modules in global class field theory.
(4.18) Definition. Let G be a finite group, and let g be a subgroup of G. A

G-module A is called G/g-induced, if it has a representation
M
A= σD,
σ∈G/g
where D ⊆ A is a g-module and σ ranges over a system of left coset represen-

tatives of g in G.
For g = {1} we obviously recover the G-induced modules from (3.9). As a

generalization of the cohomological triviality of G-induced modules, we have
the following result, which is often referred to as Shapiro’s Lemma:
L
(4.19) Theorem. Let A = σD be a G/g-induced G-module. Then
σ∈G/g
q ∼ H q (g, D);
H (G, A) =
this isomorphism is given by the composition
res π̄
H q (G, A) −−→ H q (g, A) −→ H q (g, D),
π
where π̄ is induced by the natural projection A −→ D.
Lm
We give a proof using dimension shifting. Let A = i=1 σi D, where σi ranges
over a system of left coset representatives of G/g, in particular let σ1 = 1. For
q = 0 we define a map in the opposite direction of the homomorphism
res π̄
AG /NG A −−→ Ag /Ng A −→ Dg /Ng D
Pm
by ν : Dg /Ng D → AG /NG A, ν(d + Ng D) = i=1 σi d + NG A. It is easy to
verify that (π̄ ◦ res) ◦ ν = id and ν ◦ (π̄ ◦ res) = id. Therefore π̄ ◦ res is bijective.
In case of arbitrary dimension q we now set
Aq = JG ⊗ · · · ⊗ JG ⊗ A Aq = IG ⊗ · · · ⊗ IG ⊗ A
D∗q = JG ⊗ · · · ⊗ JG ⊗ D resp. D∗q = IG ⊗ · · · ⊗ IG ⊗ D
Dq = Jg ⊗ · · · ⊗ Jg ⊗ D Dq = Ig ⊗ · · · ⊗ Ig ⊗ D
Lm
depending on whether q ≥ 0 or q ≤ 0. Because A = i=1 σi D we have
JG = Jg ⊕ K1 resp. IG = Ig ⊕ K−1
with the g-induced modules

M X m m
M X
K1 = τ ZZ · σ̄i−1 and K−1 = τ ZZ · (σi−1 − 1) .
τ ∈g i=2 τ ∈g i=2
With (1.5) and (3.10) we obtain for all q the canonical g-module decomposition
D∗q = Dq ⊕ C q
for some g-induced g-module C q . Using (3.15), we then obtain the diagram
öõôóø÷òñ Aq )
H 0 (G, res
H 0 (g, Aq )
π̄∗
H 0 (g, D∗q )
ρ̄
H 0 (g, Dq )
δq δq δq
res π̄
H q (G, A) H q (g, A) H q (g, D) ,
in which the map π̄∗ ◦ res in the upper row in dimension 0 is bijective, and the
following map ρ̄ is bijective because of (3.7) and (3.13). Since the composite
π ρ π
Aq →∗ D∗q → Dq is induced by the projection A → D, we see that this
diagram commutes. Thus the bijectivity of the upper map ρ̄ ◦ π̄∗ ◦ res implies
the bijectivity of the lower map π̄ ◦ res.
§ 5. The Cup Product
In the previous section we have seen that the restriction and corestriction maps
are given by canonical data in dimension q = 0, and induce corresponding
maps on cohomology in all dimensions. The same principle applies to the cup
product, which in dimension 0 is just the tensor product.
Let A and B be G-modules. Then A ⊗ B is a G-module, and the map (a, b) 7→
a ⊗ b induces a canonical bilinear mapping
AG × B G −→ (A ⊗ B)G ,
which maps NG A × NG B to NG (A ⊗ B). Hence it induces a bilinear mapping
H 0 (G, A) × H 0 (G, B) −→ H 0 (G, A ⊗ B) by (a, b) 7−→ a ⊗ b 12)
.
We call the element a ⊗ b ∈ H 0 (G, A ⊗ B) the cup product of a ∈ H 0 (G, A)
and b ∈ H 0 (G, B), and denote it by
a ∪ b = a ⊗ b.
This cup product in dimension 0 extends to arbitrary dimensions:
12)
As usual, we denote by a the cohomology class a = a + NG A of the element
a ∈ AG ; similar for b. Likewise, a ⊗ b stands for the cohomology class a ⊗ b =
a ⊗ b + NG (A ⊗ B) of a ⊗ b ∈ (A ⊗ B)G .
§ 5. The Cup Product 45
(5.1) Definition. There exists a uniquely determined family of bilinear map-

pings, the cup product
∪ : H p (G, A) × H q (G, B) −→ H p+q (G, A ⊗ B), p, q ∈ ZZ,
with the following properties:
(i) For p = q = 0 the cup product is given by

(a, b) 7−→ a ∪ b = a ⊗ b, a ∈ H 0 (G, A), b ∈ H 0 (G, B) .
(ii) If the sequences of G-modules

0 −→ A −→ A0 −→ A00 −→ 0
0 −→ A ⊗ B −→ A0 ⊗ B −→ A00 ⊗ B −→ 0
are both exact, then the following diagram commutes

ýüûúù H q (G, B)
H p (G, A00 ) × ∪
H p+q (G, A00 ⊗ B)
δ 1 δ
∪
H p+1 (G, A) × H q (G, B) H p+q+1 (G, A ⊗ B)
so that δ(a00 ∪ b) = δa00 ∪ b, a00 ∈ H p (G, A00 ), b ∈ H q (G, B).
(iii) If the sequences of G-modules

0 −→ B −→ B 0 −→ B 00 −→ 0,
0 −→ A ⊗ B −→ A ⊗ B 0 −→ A ⊗ B 00 −→ 0
are both exact, then the following diagram commutes
H p (G, A) ×ÿþ H q (G, B 00 ) ∪
H p+q (G, A ⊗ B 00 )
1 δ (−1)p δ
∪
H p (G, A) × H q+1 (G, B) H p+q+1 (G, A ⊗ B)
00 00 00
i.e., we have δ(a ∪ b ) = (−1)p (a ∪ δb ), a ∈ H p (G, A), b ∈ H q (G, B 00 ).
The factor (−1)p in the last diagram is necessary and results from the an-
ticommutativity of the connecting homomorphism δ, see below. One cannot
define a reasonable cup product omitting this factor.
As with the general restriction maps, we obtain the general cup product from
the case p = 0, q = 0 by dimension shifting13) .
13)
The readers who are mainly interested in applications of the cohomological cal-
culus won’t lose much by omitting the details of this shifting process. They will
be satisfied with the functorial behavior of the cup product and with its explicit
description in small dimensions (cf. (5.2), (5.6), (5.7) and (5.8)).
We recall that we identify the G-modules A ⊗ B and B ⊗ A, as well as the G-

modules (A⊗B)⊗C and A⊗(B ⊗C) (cf. §1, p. 7). This automatically leads to
a corresponding identification of the cohomology groups of these G-modules.
In particular, we can write (cf. §3, p. 29):
Ap ⊗ B = JG ⊗ · · · ⊗ JG ⊗ A ⊗ B = (A ⊗ B)p and
A ⊗ B q = A ⊗ JG ⊗ · · · ⊗ JG ⊗ B = JG ⊗ · · · ⊗ JG ⊗ A ⊗ B = (A ⊗ B)q
for p, q ≥ 0, and analogously for p, q ≤ 0 with IG in place of JG . We will use

this freely below.
Because of Proposition (3.15) we may start with the case q = 0, p = 0 and

determine the cup product by the following commutative diagram:
H 0 (G, Ap ) × H 0 (G, B q ) ∪

H 0 (G, (A ⊗ B q )p ) = H 0 (G, Ap ⊗ B q )
δp 1 δp
∪
(∗) H p (G, A) × H 0 (G, B q ) H p (G, (A ⊗ B)q ) = H p (G, A ⊗ B q )
1 δq (−1)p·q δ q
∪
H p (G, A) × H q (G, B) H p+q (G, A ⊗ B)
It follows immediately from the conditions (i), (ii) and (iii) that the cup
product is unique. We use this fact to give an explicit description of the
cup product in terms of cocycles in the special cases (p, q) = (0, q) and (p, 0):
(5.2) Proposition. If we denote by ap (resp. bq ) p-cocyles (resp. q-cocycles)

of A (resp. B), and by ap (resp. bq ) their cohomology classes, then
14)
a0 ∪ bq = a0 ⊗ bq and ap ∪ b0 = ap ⊗ b0 .
For the proof note that the products a0 ∪ bq and ap ∪ b0 defined here satisfy
the conditions (i), (ii) and (iii) for (0, q) and (p, 0) respectively. This can
be seen directly from the behaviour of the cocycles under the corresponding
maps. Now if we consider the lower part of the diagram (∗) for p = 0, resp. the
upper part for q = 0, then we see that the product defined by the commutative
diagram (∗) must coincide with the one defined by (5.2).
Thus everything boils down to showing that the product maps defined by (∗)
∪
H p (G, A) × H q (G, B) −→ H p+q (G, A ⊗ B)
14)
Note that if bq (σ1 , . . . , σq ) ∈ B is a q-cocycle, then a0 ⊗ bq (σ1 , . . . , σq ) ∈ A ⊗ B
(a0 ∈ AG ) is also a q-cocycle.
satisfy the conditions (ii) and (iii). To this end, consider the exact sequences
0 −→ A −→ A0 −→ A00 −→ 0,
0 −→ A ⊗ B −→ A0 ⊗ B −→ A00 ⊗ B −→ 0
and
0 −→ B −→ B 0 −→ B 00 −→ 0,
0 −→ A ⊗ B −→ A ⊗ B 0 −→ A ⊗ B 00 −→ 0.
From these we get by (1.9) and (1.2) the exact sequences

q q
0 −→ Aq −→ A0 −→ A00 −→ 0,
0 −→ (A ⊗ B)q −→ (A0 ⊗ B)q −→ (A00 ⊗ B)q −→ 0
and p p
0 −→ B p −→ B 0 −→ B 00 −→ 0,
0 −→ (A ⊗ B)p −→ (A ⊗ B 0 )p −→ (A ⊗ B 00 )p −→ 0 ,
and we have the diagrams
H p (G, A00 ) × H 0 (G, B q ) ∪

H p (G, (A00 ⊗ B)q )
(δ,1) (−1)p·q δ q δ
(1,δ q ) ∪
H p+1 (G, A) × H 0 (G, B q ) H p+1 (G, (A ⊗ B)q )
∪
H p (G, A00 ) × H q (G, B) H p+q (G, A00 ⊗ B) (−1)(p+1)·q δ q
(1,δ q )
(δ,1) δ
∪
H p+1 (G, A) × H q (G, B) H p+q+1 (G, A ⊗ B)
and
H 0 (G, Ap ) ×#"! H q (G, B 00 ) ∪

H q (G, (A ⊗ B 00 )p )
(1,δ) δ
δp
(δ p ,1) ∪
H 0 (G, Ap ) × H q+1 (G, B) H q+1 (G, (A ⊗ B)p )
∪
H p (G, A) × H q (G, B 00 ) H p+q (G, A ⊗ B 00 ) δp
(1,δ) (δ p ,1) (−1)p ·δ
∪
H p (G, A) × H q+1 (G, B) H p+q+1 (G, A ⊗ B).
Here the left sides in both diagrams commute for trivial reasons. The right
sides are composed from q (resp. p) squares as in (3.6), thus they commute as
well. The front and back sides commute by definition (∗) of the cup product,
and the upper squares commute because of (5.2) and the remarks following it.
Since the vertical maps are bijective, the commutativity of the upper squares
implies the commutativity of the lower squares. This completes the proof.
The axiomatic definition of the cup product in (5.1) does not give us an
explicit description of it; i.e., given two cohomology classes in terms of cocyles,
we are for now not in a position to decide which cocyle represents their cup
product in general. Only for the cases (p, q) = (0, q) and (p, 0) we have such
a description by (5.2). The attempt to give an explicit description of the cup
product for general p, q (in particular for p < 0 and q < 0) leads, however, to
major computational problems. Thus we find ourselves in a situation which is
similar to that of the restriction map, which admits a very simple description
in dimensions q ≥ 0, but not for negative dimensions. Nevertheless in both
cases we will need explicit computations only in low dimensions; given these,
one can manage knowing the functorial properties of these maps.
Before giving explicit formulas for small dimensions, we want to convince

ourselves that the cup product is compatible with the usual cohomological
maps defined above.
(5.3) Proposition. Let f : A → A0 and g : B → B 0 be two G-

homomorphisms, and let f ⊗ g : A ⊗ B → A0 ⊗ B 0 be the G-homomorphism
induced by f and g. If a ∈ H p (G, A) and b ∈ H q (G, B), then
f¯(a) ∪ g(b) = f ⊗ g (a ∪ b) ∈ H p+q (G, A0 ⊗ B 0 ).
This is completely trivial for p = q = 0, and follows in general from a simple

dimension shifting argument. We have demonstrated this technique already
frequently enough to leave the details to the reader.
(5.4) Proposition. Let A, B be G-modules, and let g be a subgroup of G. If

a ∈ H p (G, A) and b ∈ H q (G, B), then
res(a ∪ b) = res a ∪ res b ∈ H p+q (g, A ⊗ B),
and
cor(res a ∪ b) = a ∪ cor b ∈ H p+q (G, A ⊗ B).
This follows again from the case p = q = 0 by dimension shifting. In case

p = q = 0 the first formula is immediate. For the second, let a ∈ AG and
b ∈ B g be 0-cocycles representing a and b respectively. By definition (4.12) of
the corestriction in dimension 0, we have
cor(res a ∪ b) = cor(a ⊗ b + Ng (A ⊗ B))

X
= σ(a ⊗ b) + NG (A ⊗ B)
σ∈G/g
X
= a ⊗ σb + NG (A ⊗ B)
σ∈G/g
X
=a⊗( σb) + NG (A ⊗ B)
σ∈G/g
= a ∪ cor b.
We show that the cup product is anticommutative and associative:
(5.5) Theorem. Let a ∈ H p (G, A), b ∈ H q (G, B), and c ∈ H r (G, C). Then
a ∪ b = (−1)p·q (b ∪ a) ∈ H p+q (G, B ⊗ A),
and
(a ∪ b) ∪ c = a ∪ (b ∪ c) ∈ H p+q+r (G, A ⊗ (B ⊗ C))
under the canonical isomorphisms H p+q (G, A ⊗ B) ∼ = H p+q (G, B ⊗ A) and
p+q+r p+q+r
H (G, (A ⊗ B) ⊗ C) = H (G, A ⊗ (B ⊗ C)). 15)
Again, this is trivial for p = q = 0, and follows in general by dimension

shifting.
We now want to compute some explicit formulas for the cup product. For this
we denote by ap (resp. bq ) p-cocycles of A (resp. q-cocycles of B), and write
ap (resp. bq ) for their cohomology classes in H p (G, A) (resp. H q (G, B)).
(5.6) Lemma. We have a1 ∪ b−1 = x̄0 ∈ H 0 (G, A ⊗ B) with

X
x0 = a1 (τ ) ⊗ τ b−1 .
τ ∈G
Proof. By (3.14) we have the G-induced G-module A0 = ZZ[G] ⊗ A and the

exact sequences
0 −→ A −→ A0 −→ A00 −→ 0,
0 −→ A ⊗ B −→ A0 ⊗ B −→ A00 ⊗ B −→ 0 .
We think of A embedded in A0 and A ⊗ B embedded in A0 ⊗ B; to simplify
notation we do not explicitly write out these homomorphisms. Because of the
vanishing H 1 (G, A0 ) = 0, there is a 0-cochain a00 ∈ A0 with a1 = ∂a00 , so that
15)
More precisely, one should say that (−1)p·q (b ∪ a) is the image of a ∪ b under the
canonical isomorphism H p+q (G, A ⊗ B) ∼= H p+q (G, B ⊗ A) induced by A ⊗ B ∼ =
B ⊗ A, and similarly for the second formula. Cf. §1, p. 7.
(∗) a1 (τ ) = τ a00 − a00 for all τ ∈ G.

G
Let a000 ∈ A00 be the image of a00 in A00 . By definition of the connecting
homomorphism δ, we have a1 = δ(a000 ), and we obtain
(5.1) (5.2)
a ∪ b−1 = δ(a000 ) ∪ b−1 = δ(a000 ∪ b−1 ) = δ(a000 ⊗ b−1 ) = ∂(a00 ⊗ b−1 )
X (∗) X
= NG (a00 ⊗ b−1 ) = τ a00 ⊗ τ b−1 = (a1 (τ ) + a00 ) ⊗ τ b−1
τ ∈G τ ∈G
X X
= (a1 (τ ) ⊗ τ b−1 ) + a00 ⊗ NG b−1 = (a1 (τ ) ⊗ τ b−1 )
τ ∈G τ ∈G
because NG b−1 = 0.
In the following we restrict to the case B = ZZ and identify A ⊗ ZZ with A via

a ⊗ n 7→ a · n. Recall that from (3.19) we have the canonical isomorphism
H −2 (G, ZZ) ∼
= Gab .
If σ ∈ G, let σ be the element in H −2 (G, ZZ) corresponding to σ · G0 ∈ Gab .
(5.7) Lemma. a1 ∪ σ = a1 (σ) ∈ H −1 (G, A).
Proof. From the exact sequence

0 −→ A ⊗ IG −→ A ⊗ ZZ[G] −→ A −→ 0
δ
we obtain the isomorphism H −1 (G, A) −→ H 0 (G, A ⊗ IG ). Thus it suffices to
show δ(a1 ∪ σ) = δ(a1 (σ)). Using the definition of δ, we now compute
X
δ(a1 (σ)) = x0 with x0 = τ a1 (σ) ⊗ τ.
τ ∈G
On the other hand, the proof of (3.19) shows that under the isomorphism
δ
H −2 (G, ZZ) → H −1 (G, IG ) the element σ goes to δσ = σ − 1, hence we have
(5.1)
δ(a1 ∪ σ) = −(a1 ∪ δ(σ)) = −a1 ∪ (σ − 1) = y 0 .
For the cocyle y0 we obtain from (5.6)
X X X
y0 = − a1 (τ ) ⊗ τ (σ − 1) = a1 (τ ) ⊗ τ − a1 (τ ) ⊗ τ σ.
τ ∈G τ ∈G τ ∈G
The 1-cocycle a1 (τ ) satisfies a1 (τ ) = a1 (τ σ) − τ a1 (σ). Substituting this into

the last sum, we find X
y0 = τ a1 (σ) ⊗ τ σ.
τ ∈G
P
Therefore y0 − x0 = τ ∈G τ a1 (σ) ⊗ τ (σ − 1) = NG (a1 (σ) ⊗ (σ − 1)), which
shows that x0 = y 0 .
§ 6. Cohomology of Cyclic Groups 51
The following formula (5.8) is of particular interest for us. Note that if we take
an element a2 in the group H 2 (G, A), it provides us with the homomorphism
a2 ∪ : H −2 (G, ZZ) −→ H 0 (G, A),
which maps each σ ∈ H −2 (G, ZZ) to the cup product a2 ∪ σ ∈ H 0 (G, A); we
thus get a canonical mapping from the abelianization Gab to the norm residue
group AG /NG A. In class field theory we will consider a special G-module A for
which this homomorphism will be shown to be bijective; in fact, the resulting
canonical isomorphism Gab ∼ = AG /NG A is the main theorem of class field
theory. For this the following proposition will be important:
a2 (τ, σ) ∈ H 0 (G, A).

P
(5.8) Proposition. We have a2 ∪ σ = τ ∈G
Proof. We consider again the G-module A0 = ZZ[G]⊗A and the exact sequence
0 → A → A0 → A00 → 0 (A00 = JG ⊗ A). Since H 2 (G, A0 ) = 0 there is a 1-
cochain a01 ∈ A01 with a2 = ∂a01 i.e.,
(∗) a2 (τ, σ) = τ a01 (σ) − a01 (τ · σ) + a01 (τ ).
The image a001 of a01 is a 1-cocycle of A00 such that a2 = δ(a001 ). Therefore
(5.1) (5.7) X
a2 ∪ σ = δ(a001 ) ∪ σ = δ(a001 ∪ σ) = δ(a001 (σ)) = ∂(a01 (σ)) = τ a01 (σ)
τ ∈G
(∗) X X X X
= a2 (τ, σ) + a01 (τ · σ) − a01 (τ ) = a2 (τ, σ).
τ ∈G τ ∈G τ ∈G τ ∈G
§ 6. Cohomology of Cyclic Groups
So far we have introduced the basic cohomological maps and have studied
their functorial and compatibility properties. Now we will begin to prove the
central theorems of cohomology theory. We start with G-modules A, where G
is a cyclic group; the cohomology of these G-modules is particularly simple.
Let G be a cyclic group of order n with generator σ. Then we have for the
group ring n−1
M
ZZ[G] = ZZσ i , NG = 1 + σ + · · · + σ n−1 ,
i=0
and because σ −1 = (σ−1)(σ k−1 +. . .+σ+1) (k ≥ 1), the augmentation ideal

k
IG is the principal ideal of ZZ[G] generated by σ − 1, i.e., IG = ZZ[G] · (σ − 1).
(6.1) Theorem. Let G be a cyclic group and let A be a G-module. Then

H q (G, A) ∼
= H q+2 (G, A) for all q ∈ ZZ.
Proof. It suffices to specify an isomorphism H −1 (G, A) ∼ = H 1 (G, A). Given

this, the general case follows from this by dimension shifting (cf. (3.15)), since
H q (G, A) ∼
= H −1 (G, Aq+1 ) ∼
= H 1 (G, Aq+1 ) ∼
= H q+2 (G, A).
The group Z1 of 1-cocycles consists of all the crossed homomorphisms of G
in A; thus if x ∈ Z1 , then
x(σ k ) = σx(σ k−1 ) + x(σ)=σ 2 x(σ k−2 ) + σx(σ) + x(σ)

k−1
P i
= σ x(σ) (k≥1), and
i=0
x(1) = 0 because x(1) = x(1) + x(1).
Pn−1
It follows that NG x(σ) = i=0 σ i x(σ) = x(σ n ) = x(1) = 0, i.e., x(σ) ∈ NG A.
Conversely, it is easy to see that if a ∈ NG A = Z−1 is a (−1)-cocyle, then
k−1
X
x(σ) = a and x(σ k ) = σi a
i=0
defines a 1-cocyle. Therefore the map
x 7−→ x(σ)
is an isomorphism from Z1 to Z−1 = NG A. Under this isomorphism the group
R1 of 1-coboundaries is mapped to the group R−1 of (−1)-coboundaries:
x ∈ R1 ⇐⇒ x(σ k ) = σ k a − a with fixed a ∈ A
⇐⇒ x(σ) = σa − a
⇐⇒ x(σ) ∈ IG A = R−1 .
Thus in the case of a cyclic group G we always have isomorphisms
∼ H 0 (G, A) and H 2q+1 (G, A) ∼
H 2q (G, A) = = H 1 (G, A).
If
0 −→ A −→ B −→ C −→ 0
is an exact sequence of G-modules, we write the corresponding long exact
cohomology sequence in the form of an exact hexagon:
$%&'() H −1 (G, A) H −1 (G, B)
H 0 (G, C) H −1 (G, C)
H 0 (G, B) H 0 (G, A).
For exactness at the term H −1 (G, A), note that the isomorphism H 1 (G, A) ∼
=
H −1 (G, A) from the proof of (6.1) fits into the commutative diagram

*+,- A)
H −1 (G, H −1 (G, B)
H 1 (G, A) H 1 (G, B),

so that the kernel of the map H 1 (G, A) → H 1 (G, B) corresponds to the kernel
of the map H −1 (G, A) → H −1 (G, B).
For many index and order considerations the notion of a Herbrand quotient
is very useful; in particular, it can be used to simplify the computations of
indices in abelian groups. Although it is of particular interest for G-modules
when G is a cyclic group, we want to introduce it in its most general form.
(6.2) Definition. Let A be an abelian group, and let f, g endomorphisms of

A such that f ◦ g = g ◦ f = 0, so that we have inclusions im g ⊆ ker f and
im f ⊆ ker g. Then the Herbrand quotient is defined as
(ker f : im g)
qf,g (A) =
(ker g : im f )
provided both indices are finite.
We are mainly interested in the following special case:

Let A be a G-module with G cyclic of order n. Consider the endomorphisms
f =D =σ−1 and g = N = 1 + σ + · · · + σ n−1 ,
where σ is a generator of G. Obviously we have
D ◦ N = N ◦ D = 0,
and
ker D = AG , im N = NG A ; ker N = NG A , im D = IG A.
0 −1
Hence if both cohomology groups H (G, A) and H (G, A) are finite, then
0 2
|H (G, A)| |H (G, A)|
qD,N (A) = = .
|H −1 (G, A)| |H 1 (G, A)|
If this holds, we call A a Herbrand module. For these special Herbrand
quotients qD,N (A) we want to use the following notation:
(6.3) Definition. Let G be a cyclic group and A a G-module. Then

|H 0 (G, A)| |H 2 (G, A)|
h(A) = = .
|H −1 (G, A)| |H 1 (G, A)|
These special Herbrand quotients h(−) are multiplicative:
(6.4) Theorem. Let G be a cyclic group and

0 −→ A −→ B −→ C −→ 0
an exact sequence of G-modules. Then
h(B) = h(A) · h(C)
in the sense that if two of these quotients are defined, then so is the third,
and equality holds.
Proof. Consider the long exact cohomology sequence, written as the hexagon
3210/. H −1 (G, A)
f1
H −1 (G, B)
f6 f2
H 0 (G, C) H −1 (G, C)
f5 f3
H 0 (G, B) f4
H 0 (G, A)
If we write Fi for the order of the image of fi , then
|H −1 (G, A)| = F6 · F1 , |H −1 (G, B)| = F1 · F2 , |H −1 (G, C)| = F2 · F3 ,
|H 0 (G, A)| = F3 · F4 , |H 0 (G, B)| = F4 · F5 , |H 0 (G, C)| = F5 · F6 ,
and therefore
|H −1 (G, A)| · |H −1 (G, C)| · |H 0 (G, B)|
(∗)
= |H −1 (G, B)| · |H 0 (G, A)| · |H 0 (G, C)| .
Hence whenever two of the three quotients h(A), h(B), h(C) are defined, then
so is the third, and the identity (∗) implies the formula h(B) = h(A)·h(C).
Another special case of a Herbrand quotients occurs when A is an abelian

group and f and g are the endomorphisms f = 0 and g = n (n a positive
integer), i.e., g is the map ‘multiplication by n’ a 7→ n · a ∈ A. Then
(A : nA)
q0,n (A) = (n A = {a ∈ A | n · a = 0}).
|n A|
In fact, this is just a special case of what we considered above:
(6.5) Proposition. If the cyclic group G of order n acts trivially on A, then

h(A) = q0,n (A).
In particular, the Herbrand quotients q0,n are multiplicative16) :

16)
We remark that under certain assumptions one can show multiplicativity for
general Herbrand quotients qf,g
(6.6) Proposition. If 0 → A → B → C → 0 is an exact sequence of abelian

groups, then
q0,n (B) = q0,n (A) · q0,n (C);
this again in the sense that the existence of two of these quotients implies the
existence of the third.
(6.7) Proposition. If A is a finite group, then we always have

qf,g (A) = 1.
Proof. Because of the isomorphisms im f ∼

= A/ ker f and im g ∼
= A/ ker g,
|A| = | ker f | · |im f | = | ker g| · |im g|,
which implies the claim.
In particular, a finite G-module A has Herbrand quotient h(A) = 1. This re-

mark, together with the multiplicativity shown in (6.4), implies the following:
If A is a submodule of finite index in the G-module B, then

h(B) = h(A).
It is in fact this statement that is most useful in applications of the Herbrand

quotient. If the direct computation of the order of the cohomology groups of
a G-module B is not possible, the above fact allows us to consider without
loss an appropriate submodule A, provided it has finite index. This type of
consideration historically motivated the definition of the Herbrand quotient.
In the following we will show how to determine h in case of a cyclic group G

of prime order p from the Herbrand quotients q0,p . For this we need:
(6.8) Lemma. Let g and f be two endomorphisms of an abelian group A.

Then
q0,gf (A) = q0,g (A) · q0,f (A),
where again all three quotients are defined whenever any two of them are.
Proof. Consider the commutative diagram with exact rows

0;<=>789:456
f
g(A) ∩ ker f g(A) f g(A) 0
f
0 ker f A f (A) 0
We obtain the exact sequence

0 −→ ker f /g(A) ∩ ker f −→ A/g(A) −→ f (A)/f g(A) −→ 0,
so that (A : f g(A)) (A : g(A)) · |g(A) ∩ ker f |

= .
(A : f (A)) | ker f |
If we observe that
ker f g/ ker g = g −1 (g(A) ∩ ker f )/g −1 (0) ∼
= g(A) ∩ ker f,
we in fact get
(A : gf (A)) (A : g(A)) (A : f (A))
= · .
| ker gf | | ker g| | ker f |
It is easy to verify that all three quotients are defined, if two of them are.
Now we prove the important
(6.9) Theorem. Let G be a cyclic group of prime order p and A a G-module.

If q0,p (A) is defined, then q0,p (AG ) and h(A) are also defined, and
h(A)p−1 = q0,p (AG )p /q0,p (A).
Proof. Let σ be a generator of G and D = σ − 1. Consider the exact sequence

D
0 −→ AG −→ A −→ IG A −→ 0 .
From the fact that IG A is a subgroup as well as a factor group of A, we
conclude immediately that if q0,p (A) is defined, then q0,p (IG A) is also defined.
Hence as a consequence of (6.6), q0,p (AG ) is also defined, and we have
(∗) q0,p (A) = q0,p (AG ) · q0,p (IG A).
Since G acts trivially on AG , it follows from (6.5) that q0,p (AG ) = h(AG ).
To determine the quotient q0,p (IG A) we use the following interesting trick.
Pp−1
Since the ideal ZZ·NG = ZZ( i=0 σ i ) annihilates the module IG A, we can con-
sider IG A as a ZZ[G]/ZZ·NG -module. Now the ring ZZ[G]/ZZ·NG is isomorphic
to the ring ZZ[X]/(1+X +. . .+X p−1 ) with an indeterminate X. But the latter
is isomorphic to the ring ZZ[ζ] of integral elements of the field Q(ζ) of p-th
roots of unity (ζ a primitive p-th root of unity), and the map σ 7→ ζ induces
an isomorphism ZZ[G]/ZZ·NG ∼ = ZZ[ζ]. In ZZ[ζ] we now have the well-known
decomposition p = (ζ − 1)p−1 · e, e a unit, so that we can write
p = (σ − 1)p−1 · ε, ε unit in ZZ[G]/ZZ·NG .
Since the endomorphism induced by ε is an automorphism on IG A, we find

q0,ε (IG A) = 1. If we now apply Lemma (6.8), we obtain
q0,p (IG A) = q0,Dp−1 (IG A) · q0,ε (IG A) = q0,D (IG A)p−1 = 1/qD,0 (IG A)p−1 .
Since N = NG is the 0-endomorphism on IG A, we also have

q0,p (IG A) = 1/qD,0 (IG A)p−1 = 1/qD,N (IG A)p−1 = 1/h(IG A)p−1 .
In combination with (∗), this implies
§ 7. Tate’s Theorem 57
q0,p (AG ) = h(AG ), q0,p (IG A) = 1/h(IG A)p−1 , q0,p (A) = q0,p (AG )/h(IG A)p−1 .
On the other hand, the sequence 0 → AG → A → IG A → 0 gives the formula

h(A)p−1 = h(AG )p−1 · h(IG A)p−1 ,
and the claim h(A)p−1 = q0,p (AG )p /q0,p (A) follows by substitution.
In global class field theory we will apply this theorem to certain unit groups,
about which we only know that they are finitely generated of known rank. We
show that this alone suffices to compute the Herbrand quotient; namely, from
(6.9) we get the following theorem of C. Chevalley:
(6.10) Theorem. Let A be a finitely generated G-module, where G is a cyclic

group of prime order p. If α (resp. β) denotes the rank of the abelian group
A (resp. AG ), then the Herbrand quotient h(A) is given by the formula
h(A) = p(p·β−α)/(p−1) .
Proof. Since A is finitely generated, we can find a torsion free submod-

ule A1 ⊂ A of finite index (e.g. A1 = nA for suitable chosen n). We have
rank A1 = rank A = α and rank AG G
1 = rank A = β. Thus
h(A)p−1 = h(A1 )p−1 = q0,p (AG p
1 ) /q0,p (A1 ),
where q0,p (AG G G β α
1 ) = (A1 : pA1 ) = p and q0,p (A1 ) = (A1 : pA1 ) = p .
§ 7. Tate’s Theorem
Many theorems in cohomology show that the vanishing of the cohomology

groups in two consecutive dimensions implies the vanishing in all dimensions.
One of the most important results of this type is the following Theorem of
Cohomological Triviality.
(7.1) Theorem. Let A be a G-module. If there is a dimension q0 such that

H q0 (g, A) = H q0 +1 (g, A) = 0
for all subgroups g ⊆ G, then A has trivial cohomology 17) .
We will reduce the general case to that of cyclic groups G, where the result
is an immediate consequence of Theorem (6.1). It is clear that it suffices to
prove the following claim:
17)
This means that H q (g, A) = 0 for all q ∈ ZZ and all subgroups g of G.
If H q0 (g, A) = H q0 +1 (g, A) = 0 for all subgroups g ⊆ G, then H q0 −1 (g, A) = 0

and H q0 +2 (g, A) = 0 for all subgroups g ⊆ G.
Moreover, by dimension shifting, it suffices to consider the case q0 = 1. To see

this, note that if the claim holds for q0 = 1, then the isomorphism from (3.15)
H q−m (g, Am ) ∼
= H q (g, A)
implies H 1 (g, Aq0 −1 ) ∼
= H q0 (g, A) = 0 and H 2 (g, Aq0 −1 ) ∼
= H q0 +1 (g, A) = 0.
q−(q0 −1) q0 −1 ∼ q
Hence H (g, A ) = H (g, A) = 0 for all q.
Assume H 1 (g, A) = H 2 (g, A) = 0 for all subgroups g ⊆ G. We need to show

(∗) H 0 (g, A) = H 3 (g, A) = 0 for all subgroups g ⊆ G.
We prove this by induction on the order |G| of G; the case |G| = 1 is trivial.
Thus we assume that we have proved (∗) for all proper subgroups g of G; it
remains to show H 0 (G, A) = H 3 (G, A) = 0. This is clear if G is not a p-group;
in this case all Sylow subgroups are proper subgroups, and (4.17) shows that
H 0 (G, A) = H 3 (G, A) = 0.
We may therefore assume that G is a p-group. Then there exists a normal
subgroup H ⊂ G such that the quotient G/H is a cyclic group of prime order.
By the induction assumption we have
H 0 (H, A) = H 3 (H, A) = 0 as well as H 1 (H, A) = H 2 (H, A) = 0,
and, using (4.6) and (4.7), we obtain the isomorphisms
inf : H q (G/H, AH ) −→ H q (G, A) for q = 1, 2, 3.
Now H 1 (G, A) = 0 implies H 1 (G/H, AH ) = 0, hence H 3 (G/H, AH ) = 0 by

(6.1), and so H 3 (G, A) = 0.
Next H 2 (G, A) = 0 implies H 2 (G/H, AH ) = 0, hence H 0 (G/H, AH ) = 0 (by
(6.1)), which means AG = NG/H AH = NG/H (NH A) = NG A; here we have
used H 0 (H, A) = 0, i.e., AH = NH A. Thus H 0 (G, A) = 0, which proves the
theorem.
If A and B are G-modules, consider the cup product, i.e., the bilinear map
∪
H p (G, A) × H q (G, B) −→ H p+q (G, A ⊗ B).
For a fixed element a ∈ H p (G, A), the map
a ∪ : H q (G, B) −→ H p+q (G, A ⊗ B), b 7→ a ∪ b (b ∈ H q (G, B))
provides us with a whole family of maps. In the theorems below, we will use
the cup product in this way.
From the Theorem of Cohomological Triviality we deduce the following result:
§ 7. Tate’s Theorem 59
(7.2) Theorem. Let A be a G-module with the following properties: For each
subgroup g ⊆ G we have
I. H −1 (g, A) = 0,
II. H 0 (g, A) is a cyclic group of order |g|.
If a generates the group H 0 (G, A), then the cup product map
a ∪ : H q (G, ZZ) −→ H q (G, A)
is an isomorphism for all q ∈ ZZ.
Proof. The module A itself is not suitable for the proof, since we need to use
the injectivity of the map ZZ → A, n 7→ na0 (a0 + NG A = a), which induces
the cup product above for the case q = 0 (cf. (5.2)). Hence we replace A with
B = A ⊕ ZZ[G]
which we can do without changing the cohomology groups. In fact, if i : A → B
is the canonical injection onto the first component of B, then the induced map
i : H q (g, A) −→ H q (g, B)
is an isomorphism, because ZZ[G] is cohomologically trivial. Now we choose an
a0 ∈ AG such that a = a0 + NG A is a generator of H 0 (G, A). Then the map
f : ZZ −→ B with n 7−→ a0 · n + NG · n.
is injective, because of the second term NG ·n, and induces the homomorphism
f¯ : H q (g, ZZ) −→ H q (g, B).
Using (5.2), we see that the diagram
A@? ZZ) a ∪ H q (G, A)
H q (G,
i
f¯
H q (G, B)
commutes, thus it suffices to show f¯ is bijective. This follows easily from (7.1):
Since the map f : ZZ → B is injective, there is an exact sequence of G-modules
f
(∗) 0 −→ ZZ −→ B −→ C −→ 0 .
Now H −1 (g, B) = H −1 (g, A) = 0 and H 1 (g, ZZ) = 0 for all g ⊆ G, which
implies that the corresponding exact cohomology sequence has the form
f¯
0 −→ H −1 (g, C) −→ H 0 (g, ZZ) −→ H 0 (g, B) −→ H 0 (g, C) −→ 0.
If q = 0, then f¯ is clearly an isomorphism, thus H −1 (g, C) = H 0 (g, C) = 0. By
(7.1) the vanishing of two consecutive cohomology groups implies H q (g, C) =
0 for all q, and it follows from the exact cohomology sequence associated with
(∗) that f¯ : H q (G, ZZ) → H q (G, B) is bijective for all q, as claimed.
From (7.2) we obtain the following, very important result:
(7.3) Tate’s Theorem. Assume that A is a G-module with the following

properties: For each subgroup g ⊆ G we have I. H 1 (g, A) = 0 and II. H 2 (g, A)
is cyclic of order |g|. If a generates the group H 2 (G, A), then the map
a ∪ : H q (G, ZZ) −→ H q+2 (G, A)
is an isomorphism.
Addendum: If a generates the group H 2 (G, A), then res a ∈ H 2 (g, A) gen-
erates the group H 2 (g, A). Thus we also have the isomorphism
res a ∪ : H q (g, ZZ) −→ H q+2 (g, A).
Proof. Consider the isomorphism δ 2 : H q (g, A2 ) → H q+2 (g, A) from (3.15).

The assumptions I. and II. imply that H −1 (g, A2 ) = 0, and that H 0 (g, A2 ) is
cyclic of order |g|. Furthermore, the generator a ∈ H 2 (G, A) is the image of
the generator δ −2 a ∈ H 0 (G, A2 ). It follows from (5.1) that the diagram
BCED ZZ)
H q (G, δ −2 a ∪
H q (G, A2 )
id δ2
a∪
H q (G, ZZ) H q+2 (G, A) ,
commutes. Since δ −2 a ∪ is bijective by (7.2), the map a ∪ is bijective as well.
As for the addendum: Since cor ◦ res a = (G : g) · a, the order of the element
res a ∈ H 2 (g, A) is divisible by |g|, hence res a generates H 2 (g, A) by II.
Tate’s Theorem can be generalized considerably. For example, we may replace

the condition “for all subgroups g ⊆ G” by “for all p-Sylow subgroups”. In
addition, the shifting from q to q + 2 by two dimensions may be extended
(under suitable assumptions) to general dimensions. Finally, the G-module ZZ
may be replaced by more general modules18) . We do not discuss this in detail,
since the form of Tate’s Theorem presented here suffices for most applications.
For class field theory, the case q = −2 is particularly important. In this case,
Tate’s Theorem yields a canonical isomorphism between the abelianization
Gab ∼
= H −2 (G, ZZ) of G and the norm residue group AG /NG A = H 0 (G, A):
Gab −→ AG /NG A.
This canonical isomorphism is the abstract formulation of the main theorem
of class field theory, the so-called “reciprocity law”. For this reason, one can
consider Tate’s Theorem as the foundation for a purely group theoretically
formulated abstract version of class field theory. In the next part, we will
develop this idea in detail.
18)
Cf. [42, IX, §8, Th. 13, p. 156].
Part II
Local Class Field Theory
§ 1. Abstract Class Field Theory 63
§ 1. Abstract Class Field Theory
Local and global class field theory, as well as a series of further theories for
which the name class field theory is similarly justified, have the following
principle in common. All of these theories involve a canonical bijective corre-
spondence between the abelian extensions of a field K and certain subgroups
of a corresponding module AK associated with the field K. This correspon-
dence has the property that if the subgroup I ⊆ AK corresponds to an abelian
field extension L|K (the “class field associated with I”), then there exists a
canonical isomorphism between the Galois group GL|K and the factor group
AK /I. This so-called reciprocity law is the main theorem of class field theory.
This main theorem can be traced back to a common system of axioms for the
concrete theories mentioned above which essentially consists of the assump-
tions in Tate’s Theorem (cf. I, §7); in fact one can view Tate’s Theorem itself
as the abstract version of the main theorem of class field theory. The notion
of a class formation is based on this idea. It separates the purely group
theoretic machinery, which is characteristic of class field theory, from the spe-
cific considerations of field theory, and gives in an easily comprehensible and
elegant way information about the goal and function of the theory.
Let G be a profinite group, i.e., a compact group with the normal-subgroup

topology1) . We may think of G as the Galois group (endowed with the Krull
topology) of an infinite Galois field extension, although the abstract notions
in this section do not use this interpretation. The open subgroups of G are
precisely the closed subgroups of finite index. In fact, the complement of
an open subgroup is the union of (open) cosets, thus open, and since G is
compact, finitely many of these cosets cover the group G, hence the index is
finite. Conversely, a closed subgroup of finite index is open, because it is the
union of finitely many cosets, hence its complement is closed.
Given a profinite group G, we consider the family {GK | K ∈ X} of all open
subgroups of G, i.e., the closed subgroups of finite index. We label each such
subgroup with the index K, and call these indices “ fields”.
The “field” K0 with GK0 = G is called the base field. If GL ⊆ GK , we write
formally K ⊆ L, and define the degree of such an extension L|K as
[L : K] = (GK : GL ).
The extension L|K is called normal if GL ⊆ GK is a normal subgroup of GK .
If L|K is normal, then the Galois group of L|K is defined as the quotient
group
GL|K = GK /GL .
1)
We refer to [9], [28], [41] for the theory of profinite groups.
64 Part II. Local Class Field Theory
An extension L|K is called cyclic, abelian, solvable, etc., if its Galois group
GL|K = GK /GL is cyclic, abelian, solvable, etc. We define the intersection
and the compositum of such fields Ki by setting
n
\
K= Ki , if GK is (topologically) generated by the GKi in G; and
i=1
n
Y Tn
K= Ki , if GK = i=1 GKi .
i=1
If GL0 = σGL σ −1 for σ ∈ G, then we write L0 = σL, and we call two extensions
L|K and L0 |K conjugate in case L0 = σL for some σ ∈ GK . With these
notions, we obtain for each profinite group G a formal Galois theory.
In the following we consider modules A on which a profinite group G acts. In

this context it is important to keep the topological structure on G in mind.
The action of G on A should be in a certain sense continuous. More precisely,
it should satisfy one of the following, equivalent conditions:
(i) The map G × A → A with (σ, a) 7→ σa is continuous2) ,
S a ∈ A the stabilizer {σ ∈ G | σa = a} is open in G,
(ii) For each
(iii) A = U AU , where U runs through all the open subgroups of G.
(1.1) Definition. If G is a profinite group and A is a G-module satisfying the

equivalent conditions (i)-(iii) above, the pair (G, A) is called a formation.
If G is the Galois group of a (infinite) Galois extension N |K, then G acts

on the multiplicative group N × of the field N , and the pair (G, N × ) is a
formation. It is precisely this example that comes into play in local class field
theory, and one may use it as an orientation for what follows.
Let (G, A) be a formation. In the following we think of the module A as

multiplicatively written. Let {GK | K ∈ X} be the family of open subgroups
of G, indexed by the set of fields X. For each field K ∈ X we consider the
fixed module associated with K, i.e.,
AK = AGK = {a ∈ A | σa = a for all σ ∈ GK .}
In the class field theory example mentioned above, we obviously have AK =

K × . If K ⊆ L, then AK ⊆ AL .
If L|K is a normal extension, then AL is a GL|K -module. When we call the
pair (G, A) a formation, we basically mean by this the formation of these
normal extensions L|K, together with the GL|K -modules AL .
2)
Here A is interpreted as a discrete module.
We consider now for each normal extension L|K the cohomology groups of
the GL|K -module AL . For simplicity of notation, we set
H q (L|K) = H q (GL|K , AL ).
If N ⊇ L ⊇ K is a tower of normal extensions of K, we have inclusions

GN ⊆ GL ⊆ GK with GN and GL normal in GK , and cohomology theory
yields the homomorphism
G inf
H q (GL|K , AL ) = H q (GL|K , ANN |L ) −−→ H q (GN |K , AN ),
in other words
inf
→ H q (N |K)
H q (L|K) −−−N for q ≥ 1.
In addition, we also have the restriction and corestriction maps

res cor
H q (GN |K , AN ) −−→ H q (GN |L , AN ) and H q (GN |L , AN ) −−→ H q (GN |K , AN ),
that is, for every integer q homomorphims
res cor
H q (N |K) −−−→
L
H q (N |L) and H q (N |L) −−−K
→ H q (N |K).
Here we need only need to assume that N |K is normal. If both N and L are
normal, then the sequence
inf res
1 −→ H q (L|K) −−−N
→ H q (N |K) −−−→
L
H q (N |L)
is exact for q = 1, and exact for q > 1 if H i (N |L) = 1 for i = 1, . . . , q − 1
(cf. I, (4.7)).
If L|K is normal and σ ∈ G, then

τ GL 7−→ στ σ −1 GσL
defines an isomorphism between GL|K and GσL|σK , and
a 7−→ σa
an isomorphism between AL and AσL . Since (στ σ −1 GσL )σa = σ(τ GL )a,
these isomorphisms are compatible, and we obtain an equivalence between
the GL|K -module AL and the GσL|σK -module AσL . Thus every σ ∈ G yields
an isomorphism
σ∗
H q (L|K) −→ H q (σL|σK).
Using the equivalence of the modules AL and AσL , it is easy to see that the
isomorphism σ ∗ commutes with inflation, restriction and corestriction.
We call a formation (G, A) a field formation when for each normal extension
the first cohomology group vanishes:
H 1 (L|K) = 1.
In a field formation the sequence
inf res
1 −→ H 2 (L|K) −−−N
→ H 2 (N |K) −−−→
L
H 2 (N |L) (N ⊇ L ⊇ K)
is always exact (cf. I, (4.7)). We will soon see that the example mentioned
above, where G is the Galois group of a Galois field extension and A the
multiplicative group of the extension field, represents such a field formation.
If N ⊇ L ⊇ K are normal extensions, then we can always think of the group
H 2 (L|K) as embedded in the group H 2 (N |K), since the inflation map
inf
→ H 2 (N |K)
H 2 (L|K) −−−N
is injective. The presentation of class field theory will become formally espe-
cially simple, if we take this identification one step further. If L ranges over
the normal extensions of K, then the groups H 2 (L|K) form a direct system
of groups with respect to the inflation maps. Taking the direct limit
H 2 ( |K) = lim H 2 (L|K)
−→
L
we obtain a group H ( |K) in which all the groups H 2 (L|K) are embedded
2
via the injective inflation maps. If we identify these groups with their images
under this embedding, then H 2 (L|K) become subgroups of H 2 ( |K), and
[
H 2 ( |K) = H 2 (L|K).
L
In particular, if N ⊇ L ⊇ K is a tower of normal extensions of K, we have so
H 2 (L|K) ⊆ H 2 (N |K) ⊆ H 2 ( |K).
We strongly emphasize that the inflation maps are to be interpreted as inclu-
sions here. An element of H 2 (N |K) is regarded as an element of H 2 (L|K) if
it is the inflation of an element of H 2 (L|K).
Remark. Let GK be a profinite group and let A be a GK -module. Exactly

as for finite groups, we can define cohomology groups H q (GK , A), q ≥ 0 by
taking as cochains the continuous maps GK × . . . × GK → A. Then (cf. [41])
H q (GK , A) ∼
= H q ( |K) = lim H q (L|K).
−→
L
Given any extension K 0 |K of K, we obtain a canonical homomorphism
res 0
H 2 ( |K) −−−
K
→ H 2 ( |K 0 ).
In fact, if c ∈ H 2 ( |K), then there is an extension L ⊇ K 0 ⊇ K, so that c is

contained in the group H 2 (L|K), hence the restriction map
res 0
H 2 (L|K) −−−
K
→ H 2 (L|K 0 )
defines an element
resK 0 c ∈ H 2 (L|K 0 ) ⊆ H 2 ( |K 0 ).
The map c 7→ resK 0 c is independent of the choice of the field L ⊇ K 0 ; this fol-
lows from the trivial fact that restriction commutes with inflation, which is in-
resK 0
terpreted as inclusion. The restriction of the map H 2 ( |K) −−− → H 2 ( |K 0 )
2 0
to the group H (L|K) (L ⊇ K ⊇ K) gives back the usual restriction map
res 0
H 2 (L|K) −−−
K
→ H 2 (L|K 0 ).
From this we immediately obtain
(1.2) Proposition. Let (G, A) be a field formation. If K 0 |K is normal, then

incl res 0
1 −→ H 2 (K 0 |K) −−→ H 2 ( |K) −−−
K
→ H 2 ( |K 0 )
is exact.
The fundamental assertion in both local and global class field theory is the
existence of a canonical isomorphism, the so-called “reciprocity map”
Gab ∼= AK /NL|K AL
L|K
for every normal extension L|K, where Gab

L|K is the abelianization of GL|K and
NL|K AL = NGL|K AL is the norm group of AL . Because of Tate’s Theorem, we
can force the existence of such an isomorphism in abstracto by imposing the
following conditions on our formation (G, A): If L|K is any extension, then
I. H 1 (L|K) = 1 and II. H 2 (L|K) is cyclic of order [L : K].
If this holds, then the cup product with a generator of H 2 (L|K) gives an
isomorphism
Gab ∼
L|K = AK /NL|K AL .
However, there is a certain arbitrariness to this isomorphism, since it depends
on the choice of the generator of H 2 (L|K). In order to get a “canonical”
reciprocity map, we replace II. by the condition that there is an isomorphism
1
between H 2 (L|K) and the cyclic group [L:K] ZZ/ZZ, the so-called “invariant
map”, which uniquely determines the element uL|K ∈ H 2 (L|K) with image
1
[L:K] + Z
Z. The crucial point here is that this element uL|K remains “correct”
when passing to extension fields and subfields, which we ensure by imposing
certain compatibility conditions on the invariant map.
These considerations lead us to the following
(1.3) Definition. A formation (G, A) is be called a class formation if it

satisfies the following axioms:
Axiom I. H 1 (L|K) = 1 for every normal extension L|K (field formation).

Axiom II. For every normal extension L|K there exists an isomorphism
invL|K : H 2 (L|K) −→ 1
[L:K] Z
Z/ZZ,
the invariant map, with the following properties:
a) If N ⊇ L ⊇ K is a tower of normal extensions, then

invL|K = invN |K |H 2 (L|K) .
b) If N ⊇ L ⊇ K is a tower of extensions with N |K normal, then

invN |L ◦ resL = [L : K] · invN |K .
Remark. Formula II b) becomes almost obvious if one replaces it by the

commutative diagram
H 2 (NIHGF |K)
invN |K 1
[N :K] Z
Z/ZZ
resL ·[L:K]
invN |L 1
H 2 (N |L) [N :L] Z
Z/ZZ.
The extension property II a) of the invariant map implies that if H 2 ( |K) =

2
S
L H (L|K), then there is an injective homomorphism
invK : H 2 ( |K) −→ Q/ZZ.
For this map we obtain from formula II b) the following relation: If L|K is an
arbitrary extension of K, then
invL ◦ resL = [L : K] · invK ,
res
where resL is the homomorphism H 2 ( |K) −−−→ L
H 2 ( |L) defined on p. 66.
Conversely, we recover from this equality formula II b), since invN |L (resp.
invN |K ) is the restriction of invL (resp. invK ) to H 2 (N |L) (resp. H 2 (N |K)).
Together with the formulas of Axiom II, we immediately obtain additional
formulas for the corestriction map cor and the conjugate map σ ∗ (cf. p. 65).
(1.4) Proposition. Let N ⊇ L ⊇ K be extensions with N |K normal. Then

a) invN |K c = invL|K c, if L|K is normal
and c ∈ H 2 (L|K) ⊆ H 2 (N |K),
b) invN |L (resL c) = [L : K] · invN |K c, for c ∈ H 2 (N |K),
c) invN |K (corK c) = invN |L c, for c ∈ H 2 (N |L),
d) invσN |σK (σ ∗ c) = invN |K c, for c ∈ H 2 (N |K) and σ ∈ G.
Proof. a) and b) are just restatements of the formulas in Axiom II.

c) The commutative diagram on p. 68 immediately implies that the map
resL
H 2 (N |K) −−−→ H 2 (N |L) is surjective. Hence for every c ∈ H 2 (N |L) we have
c = resL c̃, c̃ ∈ H 2 (N |K), and so corK c = corK (resL c̃) = c̃[L:K] (cf. I, (4.14)).
Thus, by b), invN |K (corK c) = [L : K] · invN |K (c̃) = invN |L (resL c̃) = invN |L c.
d) Let Ne be a normal extension of the base field K0 corresponding to the

group G such that N e contains N . Then σ N e = Ne , i.e., the map a 7→ σa
defines a GNe |K0 -automorphism of the GNe |K0 -module ANe such that
σ ∗ : H 2 (N
e |K0 ) −→ H 2 (N
e |K0 )
2 e ∗
is the identity map on H (N |K0 ). Since σ commutes with inflation (inclusion)
and corestriction (cf. p. 65), we have by a) and c) for every c ∈ H 2 (N |K)
invσN |σK (σ ∗ c) = invNe |σK (σ ∗ c) = invNe |K0 (corK0 (σ ∗ c)) =
invNe |K0 (σ ∗ corK0 c) = invNe |K0 (corK0 c) = invNe |K c = invN |K c.
Now we can distinguish a “canonical” generator in each group H 2 (L|K).
(1.5) Definition. Let L|K be a normal extension. The uniquely determined

element uL|K ∈ H 2 (L|K) such that
1
invL|K (uL|K ) = [L:K] + ZZ
is called the fundamental class of L|K.
From the behavior of the invariant map described in Proposition (1.4), we see
how the fundamental classes of different field extensions are related.
(1.6) Proposition. Let N ⊇ L ⊇ K be extensions with N |K normal. Then

a) uL|K = (uN |K )[N :L] , if L|K is normal,
b) resL (uN |K ) = uN |L ,
c) corK (uN |L ) = (uN |K )[L:K] ,
d) σ ∗ (uN |K ) = uσN |σK for σ ∈ G.
Proof. Since two cohomology classes are equal if they have the same invari-
ants, the proposition follows from
[N :L] 1
a) invN |K ((uN |K )[N :L] ) = [N : L] · invN |K (uN |K ) = [N :K] + ZZ = [L:K] + ZZ
= invL|K (uL|K ) = invN |K (uL|K ),
[L:K] 1
b) invN |L (resL (uN |K )) = [L : K] · invN |K (uN |K ) = [N :K] + ZZ = [N :L] + ZZ
= invN |L (uN |L ),
1 [L:K]
c) invN |K (corK (uN |L )) = invN |L (uN |L ) = [N :L] + ZZ = [N :K] + ZZ
[L:K]
= [L : K] · invN |K (uN |K ) = invN |K ((uN |K ) ),
d) invσN |σK (σ ∗ uN |K ) = invN |K (uN |K ) = 1
[N :K] + ZZ = 1
[σN :σK] + ZZ
= invσN |σK (uσN |σK ).
Now we apply Tate’s Theorem, I, (7.3) to obtain the main theorem of class
formations.
(1.7) Main Theorem. Let L|K be a normal extension. Then the map
uL|K ∪ : H q (GL|K , ZZ) −→ H q+2 (L|K)
given by the cup product with the fundamental class uL|K ∈ H 2 (L|K) is an
isomorphism in all dimensions q.
For q = 1, 2 we immediately obtain
(1.8) Corollary. H 3 (L|K) = 1 and H 4 (L|K) ∼

= χ(GL|K ).
Proof. We have H 3 (L|K) ∼ = H 1 (GL|K , ZZ) = Hom(GL|K , ZZ) = 0, and

∼ ∼
H (L|K) = H (GL|K , ZZ) = H 1 (GL|K , Q/ZZ) = Hom(GL|K , Q/ZZ) = χ(GL|K );
4 2
here the second isomorphism H 2 (GL|K , ZZ) ∼= H 1 (GL|K , Q/ZZ) follows from
the exact cohomology sequence associated with 0 → ZZ → Q → Q/ZZ → 0,
using that the GL|K -module Q is cohomologically trivial (since Q is a uniquely
divisible group).
Since we do not have a concrete interpretation of the groups H q (L|K) in

case q = 3, 4, or generally for all cohomology groups of higher dimension 3) ,
Corollary (1.8) has no immediate concrete application. However, if q = −2,
then we have such an interpretation because of the canonical isomorphisms
Gab ∼L|K= H −2 (GL|K , ZZ) and H 0 (L|K) = AK /NL|K AL .
Thus we obtain the following important general reciprocity law:
(1.9) Theorem. Let L|K be a normal extension. Then the cup product map
uL|K ∪ : H −2 (GL|K , ZZ) −→ H 0 (L|K)
yields a canonical isomorphism
θL|K : Gab
L|K −→ AK /NL|K AL
between the abelianization of the Galois group and the norm residue group
of the module.
The isomorphism θL|K in Theorem (1.9) is called the Nakayama map. Using
I, (5.8) we can give an explicit description of this map as follows:
3)
Recently, however, the case q = −3 has been found to have a beautiful application
in connection with the solution of the “class field tower problem” (cf. [14] and
[41], Ch. I, 4.4).
If u is a 2-cocycle representing the fundamental class uL|K , then we have

h Y i
θL|K (σG0L|K ) = u(τ, σ) · NL|K AL
τ ∈GL|K
for all σG0L|K ∈ Gab

L|K = GL|K /G0L|K .
Despite this description, it turns out that the inverse isomorphism of θL|K ,
AK /NL|K AL −→ Gab
L|K ,
which is also called the reciprocity isomorphism, is often more accessible,
and also more important, in particular for local and global class field theory.
It induces a homomorphism from AK onto Gab L|K with kernel NL|K AL . This
homomorphism is called the norm residue symbol ( , L|K). Hence we
have the exact sequence
( ,L|K)
1 −→ NL|K AL −→ AK −−−−−−→ Gab
L|K −→ 1 ,
and an element a ∈ AK is a norm if and only if (a, L|K) = 1.
The following lemma establishes a relation between the norm residue symbol
( , L|K) and the invariant map invL|K , which will be useful later.
(1.10) Lemma. Let L|K be a normal extension, a ∈ AK , and a =

a·NL|K AL ∈ H 0 (L|K). If χ ∈ χ(Gab 1
L|K ) = H (GL|K , Q/Z Z) is a character,
then 1
χ((a, L|K)) = invL|K (a ∪ δχ) ∈ [L:K] ZZ/ZZ.
In this formula, the symbol δχ denotes the image of χ under the isomorphism
δ
H 1 (GL|K , Q/ZZ) → H 2 (GL|K , ZZ),
which is obtained from the exact sequence
0 −→ ZZ −→ Q −→ Q/ZZ −→ 0.
Note that the above formula provides us with a characterization of the norm
residue symbol (a, L|K) in terms of the invariant map, since an element of
Gab
L|K is uniquely determined by its values under all characters.
Proof. To simplify notation we set

σa = (a, L|K) ∈ Gab ∼ −2 (GL|K , ZZ)
L|K = H
and denote by σ̄a the element in H −2 (GL|K , ZZ) corresponding to σa under

the above isomorphism. By definition of the norm residue symbol, we have
ā = uL|K ∪ σ̄a ∈ H 0 (GL|K , AL ).
Since the cup product is associative and commutes with the δ-map, we obtain
ā ∪ δχ = (uL|K ∪ σ̄a ) ∪ δχ = uL|K ∪ (σ̄a ∪ δχ) = uL|K ∪ δ(σ̄a ∪ χ).
By I, (5.7) we further have

σ̄a ∪ χ = χ(σa ) = r
n + ZZ ∈ 1
nZZ/ZZ = H −1 (GL|K , Q/ZZ),
where n = [L : K]. Hence taking δ : H −1 (GL|K , Q/ZZ) → H 0 (GL|K , ZZ) gives
0
δ(χ(σa )) = n( nr + ZZ) = r + nZZ ∈ H (GL|K , ZZ) = ZZ/nZZ,
and therefore
ā ∪ δχ = uL|K ∪ (r + nZZ) = urL|K .
From this we get
r
invL|K (ā ∪ δχ) = r · invL|K (uL|K ) = n + ZZ = χ(σa ).
The conditions on the behavior of the invariant map under the inflation (inclu-
sion) and restriction maps in Axiom II of the definition of a class formation
already determines how the norm residue symbol behaves when passing to
extension and subfields. We summarize this in the following theorem.
(1.11) Theorem. Let N ⊇ L ⊇ K be a tower of extensions of K with N |K

normal. Then the following diagrams are commutative:
a)
MLKJ
AK
( ,N |K)
Gab
N |K
id π
( ,L|K)
AK Gab
L|K
hence (a, L|K) = π(a, N |K) ∈ GabL|K for a ∈ AK , if L|K is also normal (in
addition to N |K). Here π is the canonical projection of Gab ab
N |K onto GL|K .
b)
QPON
AK
( ,N |K)
Gab
N |K
incl Ver
( ,N |L)
AL Gab
N |L
hence (a, N |L) = Ver(a, N |K) ∈ Gab

N |L for a ∈ AK . Recall that the Ver-
res
lagerung (transfer) is induced from H −2 (GN |K , ZZ) −−→ H −2 (GN |L , ZZ).
c)
AUTSR L
( ,N |L)
Gab
N |L
NL|K κ
( ,N |K)
AK Gab
N |K
hence (NL|K a, N |K) = κ(a, N |L) ∈ Gab

N |K for a ∈ AL , where κ is the
ab
canonical homomorphism from GN |L into GabN |K .
d)
YXWV
AK
( ,N |K)
Gab
N |K
σ σ∗
( ,σN |σK)
AσK Gab
σN |σK
hence (σa, σN |σK) = σ(a, N |K)σ −1 for a ∈ AK , where for σ ∈ G, the

σ σ∗ −1
maps AK → AσK and Gab ab
N |K −→ GσN |σK are a 7→ σa and τ 7→ στ σ .
All statements follow essentially from the formulas for the behavior of funda-
mental classes under extension given in Proposition (1.6). More precisely:
a) Let χ ∈ χ(GL|K ) = H 1 (GL|K , Q/ZZ), inf χ ∈ H 1 (GN |K , Q/ZZ). Then
χ(π(a, N |K)) = inf χ((a, N |K)) = invN |K (a ∪ δ(inf χ)) = invN |K (a ∪ inf(δχ))
= invN |K (inf(a ∪ (δχ))) = invL|K (a ∪ δχ) = χ(a, L|K)
by (1.10). Since this identity holds for all characters χ ∈ χ(GL|K ), we obtain
π(a, N |K) = (a, L|K).
For the proof of b) and c) we need to convince ourselves that the diagrams
uN |K ∪
cba_`\]^Z[
AK H 0 (N |K) H −2 (GN |K , ZZ) Gab
N |K
incl res res Ver

uN |L ∪
AL H 0 (N |L) H −2 (GN |L , ZZ) Gab
N |L ,
uN |L ∪
Ademlkfghij L H 0 (N |L) H −2 (GN |L , ZZ) Gab
N |L
NL|K cor cor κ
uN |K ∪
AK H 0 (N |K) H −2 (GN |K , ZZ) Gab
N |K .
commute. This follows for the left squares from I, (4.9) and from the def-
inition given in I, p. 38. For the right squares we refer to I, (4.10) and I,
(4.13). The middle squares commute since for z ∈ H −2 (GN |K , ZZ), resp.
z 0 ∈ H −2 (GN |L , ZZ), we have by (1.6) and I, (5.4) the following identities
res(uN |K ∪ z) = res(uN |K ) ∪ (res z) = uN |L ∪ (res z), resp.
0 0 0
cor(uN |L ∪ z ) = cor(res(uN |K ) ∪ z ) = uN |K ∪ (cor z ).
The proof of d) is analogous and left to the reader.
Note that we could not argue similarly in the proof of (a), since inflation is
only defined for positive dimensions; here Formula (1.10) has proved useful.
The essential statements of class field theory are the reciprocity law (1.9) and
the properties of the norm residue symbol. In a concrete case, for example, for
local or global class field theory4) , the development of the theory will be made
considerably more concise by the abstract presentation given in this section,
that is, by anticipating the purely group-theoretic part. It leaves us the task
to verify Axioms I and II of class formations; however, we admit that there
are only very few cases where this is easy.
We want to extract some consequences of the theorems we have proved so far.
If L|K is normal, then the abelianization Gab L|K is the Galois group of the
ab
maximal abelian extension L |K contained in L, and the reciprocity law
gives an isomorphism between the Galois group of this extension and the norm
residue group AK /NL|K AL . We will now show that these abelian extensions
are uniquely determined by their norm groups, in fact, that the entire structure
of these abelian extensions of K is uniquely reflected in the group AK of the
given base field K.
A subgroup I of AK is called a norm group if there is a normal field extension
L|K such that I = NL|K AL . The following theorem shows that every norm
group I = NL|K AL is indeed the norm group of an abelian extension of K,
namely, the norm group of the maximal abelian field Lab in L.
(1.12) Theorem. Let L|K be a normal extension and Lab the maximal
abelian extension of K contained in L. Then
NL|K AL = NLab |K ALab ⊆ AK .
4)
In addition to these, there are other interesting examples of class formations.
For example, the theory of Kummer fields (cf. Part III, §1). One can even show
that for each profinite group G there is a G-module A such that (G, A) is a class
formation.
Proof. The inclusion NL|K AL ⊆ NLab |K ALab follows from the multiplicativ-
ity of the norm. The reciprocity law gives isomorphisms
AK /NL|K AL ∼= Gab = GLab |K ∼
L|K = AK /NLab |K ALab ,
and (AK : NL|K AL ) = (AK : NLab |K ALab ) < ∞ implies that we have the
equality NL|K AL = NLab |K ALab .
(1.13) Corollary. The index (AK : NL|K AL ) divides the degree [L : K] with
equality if and only if L|K is abelian.
In fact, (AK : NL|K AL ) = (AK : NLab |K ALab ) = [Lab : K] is a divisor of

[L : K], and is equal to [L : K] if and only if L = Lab , i.e., if and only if L|K
is abelian.
(1.14) Theorem. The norm groups I of AK form a lattice. The map

L 7−→ IL = NL|K AL
gives an inclusion reversing isomorphism between the lattice of abelian exten-
sions L of K and the lattice of norm groups I of AK . Hence we have
IL1 ⊇ IL2 ⇐⇒ L1 ⊆ L2 ; IL1 ·L2 = IL1 ∩ IL2 ; IL1 ∩L2 = IL1 · IL2 ,
if L1 and L2 are abelian extension fields.
Moreover, every group I ⊆ AK containing a norm group is itself a norm group.
Proof. If L1 and L2 are two abelian extensions of K, it follows from the

multiplicativity of the norm that IL1 ·L2 ⊆ IL1 ∩ IL2 . If a ∈ IL1 ∩ IL2 , then the
element (a, L1 ·L2 |K) has trivial projections (a, L1 |K) = 1 and (a, L2 |K) = 1
in GL1 |K and GL2 |K , i.e., (a, L1 ·L2 |K) = 1, so that a ∈ IL1 ·L2 . This proves
IL1 ·L2 = IL1 ∩ IL2 . Given this, we obtain further
IL1 ⊇ IL2 ⇔ IL1 ∩ IL2 = IL2 = IL1 ·L2 ⇔ [L1 ·L2 : K] = [L2 : K] ⇔ L1 ⊆ L2 .
Since every norm group is already the norm group of an abelian extension
((1.12)), we conclude that the map L 7→ IL is a inclusion reversing bijection
between the set of abelian extensions L|K and the set of norm groups. The
remaining statements are obvious consequences of this correspondence.
The above theorem shows that to understand the abelian extensions it is im-
portant to give a characterization of the norm groups using only intrinsic
properties of the group AK of the underlying base field K. For the concrete
class formations we are interested in such a characterization is possible, be-
cause in these cases there is a canonical topology on the group AK and the
norm groups turn out to be the closed subgroups of finite index. This result
is also called the Existence Theorem, since it shows the existence of an
abelian extension L which has a given closed subgroup I of finite index in
AK as its norm group. This (uniquely determined) field is called the class
field associated with the group I. One can derive such an existence theorem
in the theory of abstract class formations by adding certain existence axioms
to Axioms I and II. However, since we do not need this for our applications,
we only refer the interested reader to [42].
To complete this section we consider the norm residue symbol from a universal
viewpoint. If we start with a field K, then the groups Gab
L|K form a projective
system of groups, namely, the projective system of Galois groups of all abelian
extensions of K. We denote the projective limit of this system by
Gab lim Gab
K =← − L|K
In the case where we are dealing with actual field extensions L|K, Gab
K is the
Galois group of the maximal abelian extension over K. We can also write
Gab
K =← lim GL|K ,
−
where L ranges over all abelian extensions of K. For every a ∈ AK the elements
(a, L|K) ∈ GabL|K form by (1.11a) a compatible system of elements in the
projective system of Gab
L|K . Taking the limit, we obtain a unique element
(a, K) = lim(a, L|K) ∈ Gab

K
←−
which is called the universal norm residue symbol of K. If
πL : Gab ab
K −→ GL|K
are the individual projections from Gab ab

K onto the Galois group GL|K , then the
ab
element (a, K) ∈ GK is uniquely determined by the identities
πL (a, K) = (a, L|K).
The universal norm residue symbol yields a homomorphism AK → Gab
K whose
kernel and image we describe in the following theorem:
(1.15) Theorem. The kernel of the homomorphism

( ,K)
AK −−−−→ Gab
K
is equal to the intersection of all norm groups

\
DK = NL|K AL ,
L
and its image is dense in Gab

K (with respect to the normal-subgroup topology).
Proof. We have (a, K) = lim (a, L|K) = 1 if and only if (a, L|K) T = 1 for
←−
all normal extensions L|K, therefore if and only if a ∈ DK = NL|K AL .
The density of the image follows equally easily: If σ ∈ Gab
K , then the sets σ·H
form a fundamental system of neighborhoods of σ, where H runs through all
the open subgroups of GabK . But if H is an open subgroup, then GK /H =
ab
§ 2. Galois Cohomology 77
GL|K is the Galois group of an abelian extension L|K, and since the norm
residue symbol ( , L|K) : AK → GL|K is surjective, we find an a ∈ AK with
πL (a, K) = (a, L|K) = πL σ, i.e., (a, K) ∈ σ·H.
§ 2. Galois Cohomology
Let L|K be a finite Galois field extension and G = GL|K its Galois group.
Given such a field extension L, we immediately have two natural G-modules,
namely, the additive L+ and the multiplicative group L× of L. The additive
group is cohomologically uninteresting, because of the following result:
(2.1) Theorem. H q (G, L+ ) = 0 for all q.
This follows from the existence of a normal basis of L|K. In fact, if c ∈ L

is
Lchosen+in suchLa way that {σc | σ ∈ G} is a basis of L|K, then L+ =
+ +
σ∈G K ·σc = σ∈G σ(K ·c), which means that L is a G-induced module.
Therefore, by I, (3.13) all of its cohomology groups are trivial.
On the other hand, for the multiplicative group L× we have the following,
very important theorem:
(2.2) Theorem (Hilbert-Noether). H 1 (G, L× ) = 1.
Proof. Let aσ ∈ L× be a 1-cocycle of the G-module L× . If c ∈ L× , consider

X
b= aσ · σc.
σ∈G
Since the automorphisms σ are linearly independent (cf. [7], Ch. V, §7, n◦ 5),
there is an element c ∈ L× such that b 6= 0. Therefore
X X
τ (b) = τ aσ (τ σc) = a−1 −1
τ · aτ σ (τ σc) = aτ · b,
σ∈G σ∈G
−1 −1
i.e., aτ = τ (b )/b . Hence aτ is a 1-coboundary.
Theorem (2.2) is a generalization of the well-known “Hilbert’s Theorem 90”:
(2.3) Theorem (Hilbert). Let L|K be a cyclic extension, and let σ be a

generator of G. If x ∈ L× with NL|K x = 1, then there is a c ∈ L× such that
σc
x= .
c
×
This theorem is just a reformulation of NL|K L = (L× )σ−1 , and therefore of
H −1 (G, L× ) ∼
= H 1 (G, L× ) = 1.
Theorem (2.2) says that the finite Galois extensions L|K of K constitute a
field formation in the sense of §1. In such a formation we can think of the
cohomology groups H 2 (L|K) as the elements of the union
[
Br (K) = H 2 ( |K) = H 2 (L|K)
L
by viewing the inflation maps (which are injective because H 1 (L|K) = 1)

as the inclusions. Br (K) is also called the Brauer group of the field K. It
is an abstract variant of the well-known Brauer group in the theory of alge-
bras, which arises as follows: Consider all central simple algebras over K. By
Wedderburn’s Theorem every such algebra A is isomorphic to a full matrix
algebra Mn (D) over a division ring D|K, which is up to isomorphism uniquely
determined by A. Two algebras are in the same class if the corresponding di-
vision rings are isomorphic. The tensor product of two central simple algebras
is again a central simple algebra, and induces a multiplication on the set of
algebra classes, which makes this set a group, namely the Brauer group. We
briefly describe how this group is obtained from Br (K) = H 2 ( |K):
Let c̄ ∈ Br (K) = H 2 ( |K), say c̄ ∈ H 2 (L|K), and let c be a 2-cocycle of the
class c̄. With each σL ∈ GL|K we associate a basis element uσ and form the
K-vector space A = σ∈GL|K L · uσ . In this vector space, the formulas
uσ · λ = (σλ) · uσ (λ ∈ L), uσ · uτ = c(σ, τ ) · uστ ,
define a multiplication that makes A into a central simple algebra over K.
Another 2-cocycle of c̄ yields an equivalent algebra, and this construction
gives an isomorphism between the group H 2 ( |K) and the Brauer group of
algebras (cf. [1]).
Before the introduction of cohomology theory, algebras were used to describe
local class field theory (cf., e.g., [38]); we remark that the use of cohomology
has led to considerable simplifications.
For finite fields we can derive the following consequence from Theorem (2.2):
(2.4) Corollary. If L|K is an extension of a finite field, then

H q (GL|K , L× ) = 1 for all q.
Proof. The group GL|K is cyclic. Since L× is a finite GL|K -module, we have
for the Herbrand quotient h(L× ) = |H 0 (GL|K , L× )|/|H 1 (GL|K , L× )| = 1.
Hence H q (GL|K , L× ) = 1 for q = 0, 1, and therefore for all q.
§ 3. The Multiplicative Group of a p-adic Number Field 79
§ 3. The Multiplicative Group of a p-adic

Number Field
Let K be a p-adic number field, that is, a complete discrete valuation field of
characteristic 0 with finite residue field5) . We introduce the following notation.
Let v be a discrete valuation of K, which we always think of as normalized so
that its smallest positive value is 1,
O = {x ∈ K | v(x) ≥ 0} the valuation ring,
p = {x ∈ K | v(x) > 0} the maximal ideal,
K = O/p the residue field of K, p the characteristic of K,
U = O r p the unit group,
U 1 = 1 + p the group of principal units, and
U n = 1 + pn the higher unit groups.
We denote by q the number of elements in the residue field K, thus q = (O : p).

If f is the degree of K over the prime field of p elements, then q = pf .
Aside from v we also consider the normalized multiplicative absolute value
| |p , which arises as follows: If x ∈ O, consider the absolute norm of x,
N(x) = (O : xO) = (O : p)v(x) = pf ·v(x) ,
and set
|x|p = N(x)−1 .
If x ∈ K r O, then x−1 ∈ O, and we define |x|p = |x−1 |−1

p = N(x
−1
).
(3.1) Proposition. The group K × has the direct decomposition

K × = U × (π),
where π is a prime element of p and (π) = {π k }k∈ZZ is the infinite cyclic
subgroup of K × generated by π.
This is clear since with respect to a fixed prime element π, every x ∈ K × has
a unique decomposition x = u · π k , u ∈ U . Thus the short exact sequence
v
1 −→ U −→ K × −→ ZZ −→ 0
5)
We do not use the completeness of the field K. We only need that the valuation
is absolutely indecomposable. The local class field theory developed here is there-
fore valid without restrictions and verbatim for fields of characteristic 0 with a
henselian valuation, i.e., an absolutely indecomposable discrete valuation v with
finite residue field. Only in §7 one needs some modifications.
splits, and has the group (π) ∼

= ZZ as its group of representatives. We remark
that since the choice of the prime element π ∈ p is arbitrary, there is no
distinguished group of representatives.
In the unit group U consider the decreasing chain of higher unit groups U n :
U ⊃ U1 ⊃ U2 ⊃ U3 ⊃ · · · .
The following result shows that the factor groups of this chain are finite:
= K × and U n /U n+1 ∼
(3.2) Proposition. U/U 1 ∼ = K + for n ≥ 1.
Proof. The map that takes each u ∈ U to its residue class u mod p ∈ K ×
defines a homomorphism from U onto K × with the kernel U 1 . To show
U n /U n+1 ∼
= K + , we choose a prime element π. It is easy to see that the
map 1 + a · π n 7→ a mod p defines a homomorphism from U n onto the additive
group K + with kernel U n+1 .
(3.3) Proposition. The unit group U is an open and closed compact sub-
group of K × with respect to the valuation topology 6) .
Proof. If u ∈ U , then {x ∈ K × | v(x − u) > 0} = u + p is evidently a

neighborhood of u, which is entirely contained in U . Hence U is open in K × .
The complement of U is the union of (open) cosets of U , therefore U is also
closed.
Let S be a system of open subgroups of K × which covers U . Assume that U
cannot be covered by finitely many sets from S. Then the same is true for a
coset u1 ·U 1 ⊆ U , since the index (U : U 1 ) is finite. In u1 ·U 1 there exists again
among the finitely many cosets u2 · U 2 ⊆ u1 ·U 1 one which cannot be covered
by finitely many sets from S. Continuing in this way, we obtain a chain
u1 ·U 1 ⊇ u2 ·U 2 ⊇ u3 ·U 3 ⊇ · · · ,
and since U is complete as a closed subgroup of K × , there is a unit u ∈ U
such that u·U n = un ·U n for n = 1, 2, . . . The u·U n = u + pn form an open
nested sequence of neighborhoods of u, and if S is a set in S containing u,
there is an n with u·U n = un ·U n ⊆ S, which is a contradiction. Hence U is
compact7) .
(3.4) Corollary. The group K × is locally compact 6) .

6)
For a henselian field in the sense of 5) we have to substitute for compact, resp. in
(3.4) locally compact, relative compact resp. relative locally compact.
7)
One can also argue as follows: the groups U n = 1 + pn form a fundamental
system of neighborhoods of the unit element 1 ∈ U . Therefore U = lim U n /U n+1
←−
(U 0 = U ), where the inverse limit lim U n /U n+1 is compact as a profinite group.
←−
§ 3. The Multiplicative Group of a p-adic Number Field 81
If x ∈ K × , then x · U is an open and compact neighborhood of x by (3.3).
(3.5) Lemma. If m is a positive integer, then the map x 7→ xm yields for

sufficiently large n an isomorphism
U n −→ U n+v(m) .
Proof. If π is a prime element of p and x = 1 + a · π n ∈ U n , then

m 2 2n
xm = 1 + m · a · π n + a · π + · · · ≡ 1 mod pn+v(m) ,
2
and therefore xm ∈ U n+v(m) for sufficiently large n.
To prove this map is surjective we have to show that for every a ∈ O there
exists an element x ∈ O such that
1 + a · π n+v(m) = (1 + x · π n )m ,
i.e., 1 + a · π n+v(m) = 1 + m · π n · x + π 2n · f (x), where f (x) is an integral
polynomial in x. Obviously m = u · π v(m) , u ∈ U , and we obtain the equation
−a + u · x + π n−v(m) · f (x) = 0.
If n > v(m), then Hensel’s Lemma clearly gives a solution x ∈ O.
If additionally n is chosen so large that U n contains no m-th root of unity,
then our map is also injective.
(3.6) Corollary. If m is a positive integer, then the group of m-th powers

(K × )m is an open subgroup of K × . Furthermore,
∞
\
(K × )m = 1.
m=1
Proof. If xm ∈ (K × )m , then for a sufficiently large positive integer n the set

xm · U n+v(m) = (x · U n )m ⊆ (K × )m
T∞
is an open neighborhood
T∞ of xm . If a ∈ m=1 (K × )m , then trivially a ∈ U ,
and therefore a ∈ m=1 (U )m , i.e., a = um m , um ∈ U , for all m. Now if n is an
arbitrary
T∞ positive integer and m = (U : U n ), then a = um n
m ∈ U , and hence
a ∈ n=1 U n = {1}.
We denote by µm (K) the group of m-th roots of unity in K and prove the
(3.7) Proposition. The group (K × )m has finite index in K × ; more precisely,

(K × : (K × )m ) = m · q v(m) · |µm (K)| = m · |m|−1
p · |µm (K)|,
where q is the number of elements in K and |µm (K)| is the number of elements
in µm (K).
For the proof we make use of the Herbrand quotient with respect to the two
endomorphisms 0 and m (cf. I, p. 54). Then we have
(K × : (K × )m ) = q0,m (K × ) · |µm (K)|.
Since q0,m is multiplicative, we can decompose the first factor on the right as
q0,m (K × ) = q0,m (K × /U ) · q0,m (U/U n ) · q0,m (U n ).
Here q0,m (K × /U ) = q0,m (ZZ) = m because of (3.1), q0,m (U/U n ) = 1 since

U/U n is finite by (3.2); and q0,m (U n ) = (U n : U n+v(m) ) = q v(m) for suffi-
ciently large n by (3.5), using the fact that (U i : U i+1 ) = q.
Putting all this together proves the above formula. At the same time we get
(3.8) Corollary. (U : (U )m ) = |m|−1

p · |µm (K)|
8)
.
§ 4. The Class Formation of Unramified

Extensions
A relatively simple example of a class formation is given by the unramified

extensions of a p-adic number field K. Although this class formation is a
special case of the more general formations we consider in the next section,
we need to look at it separately first. This, because its reciprocity law is
remarkably simple, and because the results obtained for this special case will
be applied in the proof of the general local reciprocity law.
In what follows we consider finite extensions L|K of p-adic number fields and
append to the notation v, O, p, etc. introduced in §3 the relevant field as
an index, thus writing vK , OK , pK ; vL , OL , pL , etc. The valuation vK has a
unique extension to L, namely the valuation 1e · vL , where e is the ramification
index of L|K.
The extension L|K is unramified when e = 1, i.e., if a prime element π ∈ K
for pK is also a prime element for pL . This is equivalent to the statement that
the degree of the field extension [L : K] is the same as the degree [L̄ : K] of
the residue field extension L̄|K.
An unramified extension L|K is normal, and there is a canonical isomorphism
GL|K ∼
=G L̄|K
between the Galois group GL|K of the extension L|K and the Galois group
GL̄|K of the residue field extension L̄|K. In fact, if σ ∈ GL|K , we obtain from
σ(x + pL ) = σx + pL , x ∈ OL ,
a K-automorphism σ of L̄.
8)
We denote by (U )m the group of m-th powers of U (in contrast to U m ).
§ 4. The Class Formation of Unramified Extensions 83
The group GL̄|K is cyclic as the Galois group of a finite field L̄. As one can
immediately verify, it has the distinguished generating automorphism
x̄ 7−→ x̄qK , x̄ ∈ L̄,
where qK is the number of elements contained in K. Because GL|K ∼
= GL̄|K ,
we also obtain a canonical K-automorphism of L which generates GL|K .
(4.1) Definition. The automorphism ϕL|K ∈ GL|K which is induced by the

automorphism
x̄ 7−→ x̄qK , x̄ ∈ L̄,
of the residue field L̄ is called the Frobenius automorphism of L|K.
(4.2) Proposition. Let N ⊇ L ⊇ K be unramified extensions of K. Then

[L:K]
ϕL|K = ϕN |K |L = ϕN |K GN |L ∈ GL|K and ϕN |L = ϕN |K .
Proof. This follows easily from the fact that for all x ∈ OL we have
(ϕL|K x) mod pL = xqK mod pL = xqK mod pN = (ϕN |K x) mod pN ,
and for all x ∈ ON ,
[L:K]
[L:K]
(ϕN |L x) mod pN = xqL mod pN = xqK mod pN = (ϕN |K x) mod pN .
Because of its canonical nature and its good properties with respect to re-
strictions and extensions given by (4.2), the Frobenius automorphism plays a
significant special role in class field theory.
The following theorem is particularly important in both local and global class
field theory.
(4.3) Theorem. Let L|K be an unramified extension. Then

H q (GL|K , UL ) = 1 for all q.
Proof 9) . If we identify the group GL̄|K with the group GL|K , then
1 −→ UL1 −→ UL −→ L̄× −→ 1
is an exact sequence of GL|K -modules. Since H q (GL|K , L̄× ) = 1 (cf. (2.4)), it
follows that H q (GL|K , UL ) ∼
= H q (GL|K , UL1 ).
A prime element π ∈ K for pK is also a prime element of pL . Thus the map
ULn−1 → L̄+ , 1 + a · π n−1 7−→ a mod pL , a ∈ OL ,
9)
For the proof of this theorem one uses usually the completeness of the field L
(cf. [42]). We avoid this, so that our entire exposition also holds verbatim for
henselian fields in the sense of 5) .
defines a GL|K -homomorphism, and from the exact sequence of GL|K -modules
1 −→ ULn −→ ULn−1 −→ L̄+ −→ 0
we obtain, using that H q (GL|K , L̄+ ) = 0 for all q by (2.1), the isomorphism
H q (GL|K , U n ) ∼
= H q (GL|K , U n−1 ).
L L
n
Thus it follows that the injection UL → UL induces an isomorphism
H q (GL|K , ULn ) −→ H q (GL|K , UL ).
m
If m is a positive integer, the map x 7→ xm defines a homomorphism UL → UL ,
n+v(m)
and by (3.5) an isomorphism ULn → UL , provided n is sufficiently large.
m
Hence we have a homomorphism H (GL|K , UL ) −→ H q (GL|K , UL ), and an
q
m n+v(m)
isomorphism H q (GL|K , ULn ) −→ H q (GL|K , UL ). Consider the diagram
qnop , U n )
H q (GL|K H q (GL|K , UL )
L
m m
n+v(m)
H q (GL|K , UL ) H q (GL|K , UL )
This diagram obviously commutes, and all maps except for the right vertical
map are known to be bijections. Hence it follows that the homomorphism
m
H q (GL|K , UL ) −→ H q (GL|K , UL ),
that sends every cohomology class c to its m-th power cm is also a bijection
for all m. But the elements of H q (GL|K , UL ) have finite order (cf. I, (3.16)),
so that we must have H q (GL|K , UL ) = 1.
For q = 0 we obtain the
(4.4) Corollary. If L|K is unramified, then

UK = NL|K UL .
Hence in the unramified case every unit is also a norm.
We now show that the unramified extensions L|K form a class formation
with respect to the multiplicative group L× . To do this, we have to specify an
invariant map satisfying Axiom II (cf. §1, (1.3)). We proceed as follows. From
the long exact cohomology sequence associated with the sequence
v
1 −→ UL −→ L× −−→
L
ZZ −→ 0,
we obtain, using that H q (GL|K , UL ) = 1, the isomorphism
v̄
H 2 (GL|K , L× ) −→ H 2 (GL|K , ZZ).
Moreover, the exact sequence 0 → ZZ → Q → Q/ZZ → 0, together with the
fact that Q is cohomologically trivial, implies that the connecting map
δ −1
H 2 (GL|K , ZZ) −−→ H 1 (GL|K , Q/ZZ) = Hom(GL|K , Q/ZZ) = χ(GL|K )
1
is an isomorphism. If χ ∈ χ(GL|K ), then χ(ϕL|K ) ∈ [L:K] ZZ/ZZ ⊆ Q/ZZ, and
since the Frobenius automorphism ϕL|K generates the group GL|K , the map
ϕ
H 1 (GL|K , Q/ZZ) = χ(GL|K ) −→ 1
[L:K] Z
Z/ZZ
is also an isomorphism. Taking the composition of these three isomorphisms,
v̄ δ −1 ϕ
H 2 (GL|K , L× ) −→ H 2 (GL|K , ZZ) −−→ H 1 (GL|K , Q/ZZ) −→ 1
[L:K] Z
Z/ZZ,
we obtain the desired map
(4.5) Definition. If L|K is an unramified extension, define

invL|K : H 2 (GL|K , L× ) −→ 1
[L:K] Z
Z/ZZ
to be the isomorphism invL|K = ϕ ◦ δ −1 ◦ v̄.
For simplicity, we now set

H q (L|K) = H q (GL|K , L× ).
Let K0 be a fixed p-adic number field, and let T be the maximal unramified
field extension of K0 , thus the union of all unramified extensions L|K0 ; the
field T is also called the inertia field over K0 . We denote by GT |K0 the Galois
group of T |K0 .
(4.6) Theorem. The formation (GT |K0 , T × ) is a class formation with respect
to the invariant map defined in (4.5).
Proof. Axiom I is always satisfied by (2.2): H 1 (L|K) = 1. For the proof of

Axiom II a) and b) we need to prove that the following two diagrams commute
srt{zyxvuw
H 2 (L|K) v̄
H 2 (GL|K , ZZ) δ −1
H 1 (GL|K , Q/ZZ)
ϕ 1
[L:K] Z
Z/ZZ
incl inf inf incl
v̄ δ −1 ϕ 1
H 2 (N |K) H 2 (GN |K , ZZ) H 1 (GN |K , Q/ZZ) [N :K] Z
Z/ZZ,
H 2 (N}|~ |K) δ −1
v̄ ϕ 1
H 2 (GN |K , ZZ) H 1 (GN |K , Q/ZZ) [N :K] Z
Z/ZZ
res res res [L:K]
−1 ϕ
v̄ δ 1
H 2 (N |L) H 2 (GN |L , ZZ) H 1 (GN |L , Q/ZZ) [N :L] Z
Z/ZZ.
where N ⊇ L ⊇ K are two unramified extensions of K. That the left squares

commute follows immediately from the behavior of 2-cocycles under the maps
v̄, inf, and res. The middle squares commute because the inflation and restric-
tion maps are compatible with the connecting homomorphism δ (cf. I, (4.4)
and I, (4.5)).
To prove the commutativity of the right squares, let χ ∈ H 1 (GL|K , Q/ZZ) and
χ ∈ H 1 (GN |K , Q/ZZ) respectively. From (4.2) we have the formulas
inf χ(ϕN |K ) = χ(ϕN |K GN |L ) = χ(ϕL|K ), and
[L:K]
res χ(ϕN |L ) = χ(ϕN |L ) = χ(ϕN |K ) = [L : K] · χ(ϕN |K ),
which complete the proof.
We could now apply the entire theory developed in §1 to this special class
formation. However, we will not pursue this here, since we will do this for
more general, not necessarily unramified, extensions in the next section.
If T is the maximal unramified field extension of K0 , and therefore the maxi-

mal unramified extension of every unramified extension K|K0 , then we set
[
H 2 (T |K) = H 2 (L|K),
L
where L|K runs through all (finite) unramified extensions of K. Here we view
the inflation maps as inclusions (cf. §1, p. 66), so that for two normal exten-
sions N ⊇ L ⊇ K we have H 2 (L|K) ⊆ H 2 (N |K). Because of the extension
property of the invariant map II a), we obtain an injective homomorphism
invK : H 2 (T |K) −→ Q/ZZ
S∞
(cf. §1, p. 68). This homomorphism is even bijective, since Q/ZZ = n=1 n1 ZZ/ZZ,
and since for every positive integer n there exists (exactly) one unramified
extension L|K of degree n = [L : K], for which we have the bijective homo-
morphism invL|K : H 2 (L|K) → n1 ZZ/ZZ. Thus we have shown:
(4.7) Proposition. H 2 (T |K) ∼

= Q/ZZ.
If L|K is an unramified extension, the Galois group GL|K is cyclic and thus co-
incides with its abelianization Gab
L|K . Hence the norm residue symbol ( , L|K)
yields the exact sequence
( ,L|K)
1 −→ NL|K L× −→ K × −−−−−−→ GL|K −→ 1 .
What is special about the reciprocity law in the unramified case is that the
norm residue symbol has a very simple, explicit description.
(4.8) Theorem. Let L|K be unramified, and a ∈ K × . Then

K v (a)
(a, L|K) = ϕL|K .
Proof. If χ ∈ χ(GL|K ), δχ ∈ H 2 (GL|K , ZZ) and a = a · NL|K L× ∈ H 0 (L|K),

χ(a, L|K) = invL|K (a ∪ δχ)
by (1.10). This formula, together with (4.5) implies that
χ(a, L|K) = invL|K (a ∪ δχ) = ϕ ◦ δ −1 ◦ v̄(a ∪ δχ) = ϕ ◦ δ −1 (vK (a) · δχ)
K v (a)
= ϕ(vK (a) · χ) = vK (a) · χ(ϕL|K ) = χ(ϕL|K ).
K v (a)
Since this holds for all χ ∈ χ(GL|K ), it follows that (a, L|K) = ϕL|K .
Theorem (4.8) raises the question whether one can obtain the reciprocity
law without the cohomological calculus and the notion of class formations.
It appears that one could get the reciprocity law in a much more natural
way by simply defining the norm residue symbol explicitly by the formula
vK (a)
(a, L|K) = ϕL|K , and then directly verifying its essential properties. In the
unramified case, this is in fact possible. After closer inspection, the use of
the invariant map for the present case seems to be unnecessarily complicated,
in fact might appear as an attempt to actually force this idea artificially
into a cohomological framework. The reason to follow this approach lies in
the problem of a class field theoretical treatment of ramified field extensions.
Historically, it was precisely at this point, where cohomology (in the theory of
algebras) entered class field theory. In fact, for ramified extensions one cannot
readily give an explicit definition of the norm residue symbol, whereas this can
be done for an invariant map, which extends canonically the one constructed
here to the domain of arbitrary normal extensions. We will see this in the next
section.
In the class formation of unramified extensions, the extension fields L of K

correspond by (1.14) precisely to the norm groups in K × . Because of (4.8),
these norm groups can be given explicitly.
(4.9) Theorem. Let K be a p-adic number field and π a prime element. Then
UK × (π f ) 10)
is the norm group of the unramified extension L|K of degree f .
Proof. Since ϕL|K generates the group GL|K , the degree f = [L : K] is also
the order of ϕL|K in GL|K . Hence an element a ∈ K × lies in NL|K L× if and
v (a)
K
only if (a, L|K) = ϕL|K = 1, thus if and only if vK (a) ≡ 0 mod f , i.e.,
a = u · π k·f , k ∈ ZZ, u ∈ UK .
10)
We denote by (π f ) the infinite cyclic group {π k·f }k∈ZZ generated by the ele-
ment π f .
We end this section with some remarks about the universal norm residue sym-
bol of our class formation (cf. §1, p. 76). Let T |K be the maximal unramified
extension of K. If L|K ranges over all finite unramified extensions, then the
projective limit
GT |K = lim
←−
GL|K
L
is the Galois group of T |K.
For a ∈ K × we get the universal norm residue symbol (a, T |K) ∈ GT |K by

(a, T |K) = lim
←−
(a, L|K).
L
This gives a homomorphism
( ,T |K)
K × −−−−−−→ GT |K .
If πL : GT |K → GL|K is the canonical projection of GT |K onto GL|K , then
K v (a)
πL (a, T |K) = (a, L|K) = ϕL|K ∈ GL|K .
Because of (4.2), the elements ϕL|K ∈ GL|K form a compatible system of
elements in the projective system of GL|K , and we call the element
ϕK = lim
←−
ϕL|K ∈ GT |K
L
the “universal” Frobenius automorphism of K. It has infinite order, since from
ϕnK = 1 it would immediately follow that πL (ϕnK ) = ϕnL|K = 1 for all ϕL|K ,
which is obviously impossible.
For this symbol ( , T |K), we now have the
v (a)
(4.10) Theorem. If a ∈ K × , then (a, T |K) = ϕKK . The kernel of the
homomorphism
( ,T |K)
K × −−−−−−→ GT |K
is the unit group UK .
Proof. If L|K is an unramified extension, and πL : GT |K → GL|K is the

canonical projection from GT |K onto GL|K , then
K v (a) v (a)
πL (a, T |K) = (a, L|K) = ϕL|K = πL (ϕKK ).
v (a) v (a)
This shows that (a, T |K) = ϕKK . Hence (a, T |K) = ϕKK = 1 if and only
if vK (a) = 0 (ϕK has infinite order), therefore if and only if a ∈ UK .
The class formation of unramified extensions provides an example that shows

that the universal norm residue symbol is not surjective in general. In fact, its
image is the infinite cyclic group generated by ϕK , which is a dense subgroup
of GT |K isomorphic to ZZ. Since this is not a profinite group, it obviously
cannot coincide with GT |K ; only its completion in GT |K coincides with GT |K .
§ 5. The Local Reciprocity Law 89
§ 5. The Local Reciprocity Law
We fix a p-adic number field K0 and let Ω denote its algebraic closure. For
every normal extension L|K with L finite over K0 we again set (cf. §2, p. 78)
H q (L|K) = H q (GL|K , L× ),
Br (K) = H 2 ( |K) = L|K H 2 (L|K)
S
(Brauer group of K).
If G = GΩ|K0 is the Galois group of Ω|K0 , the formation (G, Ω × ) is a field

formation, because H 1 (L|K) = 1 (cf. (2.2)). We will show in this section that
it is even a class formation. For this we have to extend the invariant map
introduced in §4 to ramified extensions L|K. The following lemma, which
is also called the “second fundamental inequality”, provides the key to this
generalization:
(5.1) Lemma. Let L|K be a normal extension. Then the order |H 2 (L|K)| of
H 2 (L|K) is a divisor of the degree [L : K]:
|H 2 (L|K)| [L : K].
Proof. Assume first L|K is cyclic of prime degree p = [L : K]. We show the
Herbrand quotient h(L× ) = |H 2 (L|K)|/|H 1 (L|K)| = |H 2 (L|K)| = p. If q0,p
is the Herbrand quotient with respect to the endomorphisms 0 and p, then
h(L× )p−1 = q0,p (K × )p /q0,p (L× )
by I, (6.9). Using (3.7), we have for the Herbrand quotients on the right side
v (p)
q0,p (K × ) = (K × : (K × )p )/|Kp | = p · qKK ,
× × × p v (p)
q0,p (L ) = (L : (L ) )/|Lp | = p · qLL .
If f = [L̄ : K] is the inertia degree and e the ramification index, then p = e · f ,
f
qL = qK , and vL (p) = e · vK (p). Substitution into the above formulae yields
p·vK (p) e·f ·vK (p)
h(L× )p−1 = pp · qK /p · qK = pp−1 , i.e., h(L× ) = p.
The general case follows from this by purely cohomologically methods. Since
the Galois group GL|K is solvable, there exists a cyclic intermediate field K 0
over K of prime degree, K ⊆ K 0 ⊆ L. Because H 1 (L|K 0 ) = 1, the sequence
res
1 −→ H 2 (K 0 |K) −→ H 2 (L|K) −−→ H 2 (L|K 0 )
is exact. This shows that
|H 2 (L|K)| |H 2 (L|K 0 )|·|H 2 (K 0 |K)| .
We have already shown that |H 2 (K 0 |K)| = [K 0 : K], and when we assume by
induction on the field degree that |H 2 (L|K 0 )| [L : K 0 ], then it follows that
|H 2 (L|K)| [L : K 0 ] · [K 0 : K] = [L : K] .
The above proof makes use of the solvability of the Galois group GL|K , which
follows immediately from the fact that we have a cyclic inertia field between
K and L, and above it the cyclic ramification field over which L has prime
power degree. One can get around this by using I, (4.16) to reduce to the case
of an extension of prime power degree, and then proceeding as above.
That we can extend the invariant map to the case of ramified extensions is
obvious, once we have proved the following theorem:
(5.2) Theorem. If L|K is a normal extension and L0 |K is the unramified

extension of the same degree [L0 : K] = [L : K], then
H 2 (L|K) = H 2 (L0 |K) ⊆ H 2 ( |K).
Proof. It suffices to show the inclusion H 2 (L0 |K) ⊆ H 2 (L|K); if this holds,
the inclusion must have an equality because |H 2 (L0 |K)| = [L : K] (cf. (4.5)),
and |H 2 (L|K)| [L : K] by (5.1).
If N = L · L0 , and L0 |K is unramified, then N |L is also unramified, 11)
. Let
c ∈ H 2 (L0 |K) ⊆ H 2 (N |K). It follows from the exact sequence
res
1 −→ H 2 (L|K) −→ H 2 (N |K) −−−→
L
H 2 (N |L),
that c lies in H 2 (L|K) if and only if resL c = 1. Since resL c = 1 if and only if
invN |L (resL c) = 0 (cf. (4.6)), our theorem follows once we have shown that
invN |L (resL c) = [L : K] · invL0 |K c,
1
since invL0 |K c ∈ [L:K] Z
Z/ZZ. The last equality is a special case of the following
lemma.
(5.3) Lemma. Let M |K be a normal extension containing the two extensions

L|K and L0 |K with L0 |K unramified. Then N = L·L0 |L is also unramified 11) .
If c ∈ H 2 (L0 |K) ⊆ H 2 (M |K), then resL c ∈ H 2 (N |L) ⊆ H 2 (M |L), and
invN |L (resL c) = [L : K] · invL0 |K c.
Proof. The fact that the 2-cocycles of the class resL c have their values in N ×
(cf. §1, p. 66) implies that resL c ∈ H 2 (N |L).
Let f be the inertia degree and e the ramification index of the (not necessarily
normal) extension L|K. We think of the valuations vK and vL as extended
to M . Then we have vL = e · vK . By definition, the invariant map is the
composite of the three isomorphisms v̄, δ −1 , ϕ (cf. (4.5)); hence to prove the
above formula it suffices to check that the following diagram commutes
11)
Note: If T is the maximal unramified extension of K, then T · L is the maximal
unramified extension of L.
H 2 (L 0 |K) δ −1
v̄K ϕ 1
H 2 (GL0 |K , ZZ) H 1 (GL0 |K , Q/ZZ) [L0 :K] Z
Z/ZZ
incl inf inf incl
1
H 2 (M |K) H 2 (GM |K , ZZ) H 1 (GM |K , Q/ZZ) [M :K] Z
Z/ZZ
resL e·res e·res ·[L:K]
v̄L −1 ϕ
δ 1
H 2 (N |L) H 2 (GN |L , ZZ) H 1 (GN |L , Q/ZZ) [N :L] Z
Z/ZZ.
In this diagram it is understood that the lower vertical maps only map the
images of the upper vertical maps to the cohomology groups in the bottom
row. That the left square commutes follows from the behavior of the 2-cocyles
under the maps in question. The middle square commutes because the inflation
and restriction maps commute with the homomorphism δ (cf. I, (4.4) and I,
(4.5)). To see that the right square commutes, we have to consider the equation
ϕN |L L0
= ϕfL0 |K
which is a generalization of (4.2). But this is easy to see that if a ∈ L0 , then
f f
ϕN |L (a) ≡ aqL mod pN = aqK mod pN = aqK mod pL0 = ϕfL0 |K (a).
Now if χ ∈ H 1 (GL0 |K , Q/ZZ), then
[L : K] · χ(ϕL0 |K ) = e · f · χ(ϕL0 |K ) = e · χ(ϕfL0 |K ) = e · χ(ϕN |L L0
)
= e · infχ(ϕN |L ) = e · (res ◦ inf)χ(ϕN |L ).
hence the right diagram commutes, which completes the proof of the lemma.
From Theorem (5.2) we have the equality

[
Br (K) = H 2 ( |K) = H 2 (T |K) = H 2 (L|K),
L|K
unramified
and because of the invariant map we obtain using (4.7) the
(5.4) Theorem. The Brauer group Br (K) of a p-adic number field K is

canonically isomorphic to Q/ZZ:
Br (K) ∼
= Q/ZZ.
(5.5) Definition. Let L|K be a normal extension and L0 |K be the unramified

extension of the same degree [L0 : K] = [L : K], so that H 2 (L|K) = H 2 (L0 |K).
Define
invL|K : H 2 (L|K) −→ [L:K]
1
ZZ/ZZ
to be the isomorphism invL|K c = invL0 |K c (c ∈ H 2 (L|K) = H 2 (L0 |K)).
With the definition of this invariant map we have reached our goal. If K0 is
a p-adic number field, Ω its algebraic closure and GK0 = GΩ|K0 is the Galois
group of Ω|K0 , then
(5.6) Theorem. The formation (GK0 , Ω × ) is a class formation with respect

to the invariant map defined in (5.5).
Proof. Axiom I is satisfied by (2.2): H 1 (L|K) = 1.

If N ⊇ L ⊇ K are normal extensions over a fixed p-adic base field K0 and
N 0 |K (resp. L0 |K) is the unramified extensions of degree [N 0 : K] = [N : K],
(resp. [L0 : K] = [L : K]) (N 0 ⊇ L0 ⊇ K), then for every c ∈ H 2 (L|K) =
H 2 (L0 |K) we have, using (4.6),
invN |K c = invN 0 |K c = invL0 |K c = invL|K c.
This proves Axiom II a). For the proof of Axiom II b), let L|K be an arbitrary
extension with L finite over K0 . If resL is the homomorphism
res
H 2 ( |K) −−−→
L
H 2 ( |L)
(cf. §1, p. 66), then we have to show that
invL ◦ resL = [L : K] · invK
(cf. the remarks made at the end of (1.3)). If c ∈ H 2 ( |K), we can assume by
(5.2) that c ∈ H 2 (L0 |K), where L0 |K is unramified. Then N = L·L0 |L is also
unramified, and resL c ∈ H 2 (N |L) ⊆ H 2 ( |L). From Lemma (5.3) we obtain
invL (resL c) = [L : K] · invK c,
which proves our claim.
Theorem (5.6.) allows us to apply the theory of abstract class formations

developed in §1; we consider the general results established there once more
in their special form for the case at hand.
For every normal extension L|K we have the fundamental class
uL|K ∈ H 2 (L|K) with the invariant invL|K uL|K = 1
[L:K] + ZZ.
The Main Theorem of Local Class Field Theory is the statement
(5.7) Theorem. Let L|K be a normal extension. Then the homomorphism

uL|K ∪ : H q (GL|K , ZZ) −→ H q+2 (L|K)
is bijective for every integer q.
For q = 1, 2 it follows that (cf. (1.8))
(5.8) Corollary. H 3 (L|K) = 1 and H 4 (L|K) = χ(GL|K ).
For q = −2 we get the local reciprocity law:
(5.9) Theorem. For every normal extension L|K we have the isomorphism
uL|K ∪
Gab ∼ −2 (GL|K , ZZ) −−−−→ H 0 (L|K) = K × /NL|K L× .
L|K = H
The inverse isomorphism induces an exact sequence

( ,L|K)
1 −→ NL|K L× −→ K × −−−−−−→ Gab
L|K −→ 1,
where ( , L|K) is the norm residue symbol. When passing to subfields,

extension fields and conjugate fields, the norm residue symbol behaves as
follows:
(5.10) Proposition. Let N ⊇ L ⊇ K be extensions of K with N |K normal.

Then the following diagrams are commutative (cf. (1.11)):
K × K ×
( ,N |K) ( ,N |K)
a) Gab
N |K b) Gab
N |K
id π incl Ver
( ,L|K) ( ,N |L)
K× Gab
L|K , L× Gab
N |L ,
L × K£¢¡ ×
( ,N |L) ( ,N |K)
c) Gab
N |L d) Gab
N |K
NL|K σ
κ σ∗
( ,N |K) ( ,σN |σK)
K× Gab
N |K , σK × Gab
σN |σK .
Here in diagram a) the extension L|K is also assumed to be normal, and in

diagram d) σ denotes an element in GK0 .
By (1.10) the norm residue symbol and the invariant map are related:
(5.11) Lemma. Let L|K be a normal extension, a ∈ K × , and ā =

a · NL|K L× ∈ H 0 (L|K). If χ ∈ χ(Gab 1
L|K ) = χ(GL|K ) = H (GL|K , Q/Z
Z),
then 1
χ(a, L|K) = invL|K (ā ∪ δχ) ∈ [L:K] ZZ/ZZ,
δ
where δχ is the image of χ under H 1 (GL|K , Q/ZZ) → H 2 (GL|K , ZZ).
For unramified extensions L|K we gave in (4.8)) an explicit description of the

the norm residue symbol ( , L|K) in terms of the Frobenius automorphism:
K v (a)
(a, L|K) = ϕL|K .
It is very important to have such an explicit representation in case of ramified

extensions as well. Concerning this question, a fairly general result has been
obtained in recent years by J. Lubin and J. Tate. We will consider this
representation in §7.
We extract the following consequence of (4.8):
(5.12) Theorem. If L|K is an abelian extension, then the norm residue

symbol ( , L|K) maps the unit group UK onto the inertia group of GL|K
1
and the principal unit group UK onto the ramification group.
Proof. Let Lτ be the inertia field between K and L, f = [Lτ : K] and Gτ the
inertia group of GL|K , thus the fixed group of Lτ . If u ∈ UK , then by (5.10 a)
v (u)
we have π(u, L|K) = (u, L|K) · Gτ = (u, Lτ |K) = ϕLKτ |K = 1, and therefore
(u, L|K) ∈ Gτ . Conversely, let τ ∈ Gτ and a ∈ K × with (a, L|K) = τ . Then
v (a)
π(a, L|K) = (a, L|K) · Gτ = 1, therefore (a, Lτ |K) = ϕLKτ |K = 1,
i.e., vK (a) ≡ 0 mod f . If we choose a b ∈ L× with vL (b) = f1 vK (a), then
vL (NL|K b) = e · vK (NL|K b) = [L : K] · vL (b) = e · vK (a),
therefore vK (a) = vK (NL|K b), a = u · NL|K b with u ∈ UK . From this we have
(a, L|K) = (u, L|K) = τ , i.e., UK is mapped onto the entire inertia group.
n
Observing that (UK , L|K) = 1 for sufficiently large n, we conclude that the
ramification group Gv , which is the only p-Sylow subgroup of Gτ , is the image
1 n n
of the p-Sylow subgroup UK /UK of UK /UK (cf. (3.2)).
It is possible to strengthen Theorem (5.12) by showing that the higher prin-

n
cipal unit groups UK , when suitably numbered, are mapped onto the higher
ramification groups of GL|K . For this we refer to [42], XV, §2, Cor. 3.
To end, we briefly discuss the universal norm residue symbol of our class
formation (cf. §1, p. 76). For each abelian extension L|K we have the map
( ,L|K)
K × −−−−−−→ GL|K .
By forming the projective limit
Gab
K =← lim GL|K (L|K abelian)
−
×
we get, for every a ∈ K , the element
(a, K) = lim(a, L|K) ∈ Gab
K
←−
in the Galois group Gab
K of the maximal abelian extension of K.
§ 6. The Existence Theorem 95
(5.13) Theorem. The universal residue symbol defines an injective homo-

morphism
( ,K)
K × −−−−→ Gab K.
Proof. By (1.15) the intersection DK = L NL|K L× is the kernel of ( , K).

T
By (3.7) the groups of m-th powers (K × )m are of finite index, and therefore are
norm groups
T∞ by Theorem (6.3), which is proved in the next section. Therefore
DK ⊆ m=1 (K × )m = 1.
If a ∈ K × , then the restriction of (a, K) ∈ Gab

K to the inertia field T |K gives
the universal norm residue symbol of the class formation of the unramified
extensions of K discussed in §4 (cf. §4, p. 87). By (4.10) we thus have
v (a)
(a, K) T
= (a, T |K) = ϕKK ∈ GT |K ,
where ϕK = limL|K ϕ
unram. L|K
∈ GT |K denotes the universal Frobenius au-
←−
tomorphism.
In global class field theory we have to consider besides the p-adic number
fields also the field IR of real numbers. There is also a reciprocity law over
the reals, which is so simple, however, that we can explain it with only a few
words.
The field IR has only one algebraic extension, namely the field C of complex
numbers. The pair (GC|IR , C× ) constitutes a class formation in a trivial way:
The group
H 2 (GC|IR , C× ) ∼
= H 0 (GC|IR , C× ) = IR× /NC|IR C×
is cyclic of order 2, since an element a ∈ IR× is a norm from C if and only if
a > 0. The invariant map
invC|IR : H 2 (GC|IR , C× ) −→ 21 ZZ/ZZ
is defined in the obvious way, and the norm residue symbol ( , C|IR) is char-
acterized by the equation
√ √
(a, C|IR)( −1) = ( −1 )sgn(a) ,
since (a, C|IR) is either the identity or the conjugation map, depending on
whether a is a norm or not, i.e., whether a > 0 or a < 0.
§ 6. The Existence Theorem
From the abstract class field theory of §1 we see (cf. (1.14)) that the norm
groups in a p-adic number field K correspond bijectively to the abelian ex-
tensions of K:
(6.1) Theorem. Let K be a p-adic number field. Then the correspondence

L 7−→ IL = NL|K L× ⊆ K ×
gives an inclusion reversing isomorphism between the lattice of abelian exten-
sions L|K and the lattice of norm groups in K × . Every group containing a
norm group is again a norm group.
The last theorem shows that the structure of the abelian extensions of K is
reflected in the multiplicative group K × , and naturally leads to the question
of how the norm groups in K × can be characterized by intrinsic properties of
K × . This is done by the following so-called Existence Theorem:
(6.2) Theorem. The norm groups of K × are precisely the open (and thus
closed) subgroups of finite index.
Proof. By the reciprocity law (5.9) every norm group IL ⊆ K × has finite
index in K × . If m is this index, then clearly (K × )m ⊆ IL . By (3.6) (K × )m is
open, hence IL is open as the union of (open) cosets of (K × )m in IL .
Conversely, let I ⊆ K × be an open subgroup of finite index m in K × . Then
(K × )m ⊆ I and by (6.1) I is a norm group if (K × )m is a norm group. We show
this first in case that K contains
√ the m-th roots of unity. For each a ∈ K ×
we form the field La = K( a), and set
m
[
L= La .
a∈K ×
Then L|K is a finite abelian extension, because K × /(K × )m is finite (cf. (3.7)),
and therefore there are only finitely many distinct fields among the La . We
now claim that \
(K × )m = IL = ILa 12)
.
a∈K ×
√
The degree [La : K] = [K( a ) : K] = d is obviously a divisor of m, hence the
m
inclusion (K × )d ⊆ ILa implies (K × )m ⊆ ILa for all a. Therefore (K × )m ⊆ IL .

On the other hand, the theory of Kummer extensions (cf. Part III, §1, p. 115,
(1.3)) gives an isomorphism between the factor group K × /(K × )m and the
character group of the Galois group GL|K , so that by (5.9)
(K × : (K × )m ) = |GL|K | = (K × : IL ).
Thus (K × )m = IL , and therefore (K × )m is a norm group. If K does not
contain the m-th roots of unity, let K1 be the extension obtained by adjoining
the m-th roots to K. From the above we know that (K1× )m is the norm group
of an extension L|K1 : (K1× )m = NL|K1 L× . Let L e be the smallest normal
extension of K containing L. Then
12) S
The right equation follows from (6.1) because L = a∈K × La .
e× e× × × m
NL|K e 1 L ) ⊆ NK1 |K (NL|K1 L ) = NK1 |K ((K1 ) )
e L = NK1 |K (NL|K
= (NK1 |K K1× )m ⊆ (K × )m .
Hence (K × )m is a group containing the norm group NL|K e×
e L , thus by (6.1)
(K × )m is a norm group itself, and our Existence Theorem is proved.
Theorem (6.2) is called the Existence Theorem because its crucial assertion is
that given an open subgroup I of finite index in K × , there exists an abelian
extension L|K whose norm group NL|K L× = I. This field L is uniquely
determined and is called the class field associated with I.
It is clear that the open subgroups of finite index in K × are also closed of
finite index, and vice versa, since the complement of a subgroup of finite index
in K × consists of its finitely many cosets. More generally, we have
(6.3) Theorem. If I is a subgroup of K × , then the following conditions are

equivalent:
(i) I is a norm group,
(ii) I is open of finite index,
(iii) I is is closed of finite index,
(iv) I has finite index.
Proof. The conditions (i), (ii), (iii) are equivalent by (6.2) and our remark
above. Furthermore, (iv) is equivalent to (ii), since a subgroup I of finite index
m contains the open group (K × )m , and is therefore open.
Apart from the topological characterization of norm groups given by Theorem

(6.3), we also have the following description of these groups, which is of an
arithmetic nature (cf. also (4.9)).
(6.4) Theorem. The norm groups of K × are precisely the groups containing
n
UK × (π f ), n = 0, 1, 2, . . . , f = 1, 2, . . .
Here = UK , π is a prime element of K, and (π f ) is the subgroup of K ×
0
UK
generated by π f .
Proof. Every group UK n

× (π f ) has finite index in K × = UK
0
× (π), and is
therefore a norm group by (6.3), hence has only norm groups containing it.
n
Conversely, if I is a norm group, it is open. Since the UK form a fundamental
× n n
system of neighborhoods of 1 ∈ K , there is a UK with UK ⊆ I. If π is a prime
element and f the index (K × : I), then π f ∈ I, and thus UK n
×(π f ) ⊆ I.
We will give a more detailed account of norm groups in the next section.
§ 7. Explicit Determination of the Norm

Residue Symbol 13)
In Theorem (4.8) we have given an explicit description of the norm residue

symbol for unramified extensions via the Frobenius automorphism. In this
section we derive such an explicit formula for the norm residue symbol in
certain special totally ramified extensions. Since these extensions together
with the unramified extensions generate the maximal abelian field, this allows
us to explicitly determine the universal norm residue symbol.
Let K be a p-adic number field, O the ring of integers of K, π a prime element

and q = (O : π O) the number of elements in the residue field K.
We consider the set ξπ of all power series f (Z) ∈ O[[Z]] such that
f (Z) ≡ π · Z mod degree 2 and f (Z) ≡ Z q mod π.
Two power series are called congruent mod degree n (resp. mod π) if their
terms of degree less than n coincide (resp. when their coefficients are congruent
mod π). The simplest example of a power series in ξπ is the polynomial f (Z) =
π · Z + Z q , which one may regard as a standard model. Let
f n (Z) = f (f ( · · · f (Z) · · · )) ∈ O[[Z]]
be the power series obtained by n-fold substitution in f (Z), where f 0 (Z) = Z.
Let Λf,n be the set of those elements λ of positive valuation in the algebraic
closure Ω of K with f n (λ) = 0. We consider the fields
Lf,n = K(Λf,n ), n = 1, 2, . . .
Because
f n (Z) = f (f n−1 (Z)) = f n−1 (Z) · φn (Z), φn (Z) ∈ O[[Z]],
it is immediately clear that Λf,n−1 ⊆ Λf,n , and therefore that
Lf,n−1 ⊆ Lf,n ,n = 1, 2, . . .
S∞ S∞
We set Λf = n=1 Λf,n and Lf = K(Λf ) = n=1 Lf,n .
We will show that the extensions Lf,n |K are abelian and totally ramified,
n
and that they are associated with the norm groups UK × (π) (cf. (6.4)). The
essential idea here is to use certain power series to make the zero set Λf,n an
O -module in such a way that multiplication of Λf,n by a unit u ∈ O produces
a permutation of Λf,n , which induces a K-automorphism of Lf,n |K, namely
the automorphism (u−1 , Lf,n |K).
13)
In this section we follow [34]. For Part III only Theorem (7.16) is used.
§ 7. Explicit Determination of the Norm Residue Symbol 99

Pn
(7.1) Lemma. Let f (Z), g(Z) ∈ ξπ and L(X1 , . . . , Xn ) = i=1 ai Xi be a
linear form with coefficients ai ∈ O. Then there is a uniquely determined
power series F (X1 , . . . , Xn ) with coefficients in O with the properties
F (X1 , . . . , Xn ) ≡ L(X1 , . . . , Xn ) mod degree 2,
f (F (X1 , . . . , Xn )) = F (g(X1 ), . . . , g(Xn )) .
Proof. We set X = (X1 , . . . , Xn ) and g(X) = (g(X1 ), . . . , g(Xn )). One im-
mediately verifies that if F (X) is a power series and Fr (X) ∈ O[X] are its
truncations consisting of all terms of F (X) of degree at most r, then F (X) is
a solution of the above problem if and only if F (X) ≡ L(X) mod degree 2,
thus F1 (X) = L(X), and for every r the following congruence holds
(∗) f (Fr (X)) ≡ Fr (g(X)) mod degree (r + 1).
For r = 1, i.e., for F1 (X) = L(X) this is true. If we have found a unique
Fr (X) ∈ O[X] satisfying condition (∗), and we set Fr+1 (X) = Fr (X) +
∆r+1 (X) with a homogeneous form ∆r+1 of degree r + 1, it follows from
f (Fr+1 (X)) ≡ f (Fr (X)) + π · ∆r+1 (X) mod degree (r + 2),
r+1
Fr+1 (g(X)) ≡ Fr (g(X)) + π · ∆r+1 (X) mod degree (r + 2),
that for ∆r+1 the congruence
f (Fr (X)) − Fr (g(X))
∆r+1 (X) ≡ mod degree (r + 2)
π r+1 − π
must be satisfied. Hence we obtain ∆r+1 is a unique way as the first trunca-
tion, i.e., as the homogeneous
form of (r + 1)-th degree of the power series
f (Fr (X)) − Fr (g(X)) / π r+1 − π . Because
f (Fr (X)) − Fr (g(X)) ≡ (Fr (X))q − Fr (X q ) ≡ 0 mod π
the form ∆r+1 has integral coefficients, and hence so has Fr+1 = Fr + ∆r+1 .
This shows existence and uniqueness of the series F (X) = limr→∞ Fr (X).
Remark. The proof actually shows that F is the only power series in every
field containing O which satisfies the equations of the lemma.
For us, the cases L(X, Y ) = X + Y and L(Z) = aZ, a ∈ O, are important. If
f ∈ ξπ , let Ff (X, Y ) be the uniquely determined solution of the equations
Ff (X, Y ) = X + Y mod degree 2,
f (Ff (X, Y )) = Ff (f (X), f (Y )) .
Moreover, for every a ∈ O and f, g ∈ ξπ let the series af,g (Z) ∈ O[[Z]] be the
uniquely determined solution of
af,g (Z) ≡ aZ mod degree 2,
f (af,g (Z)) = af,g (g(Z)).
For simplicity, we write af for af,f . The following proposition shows that
Ff (X, Y ) in a certain sense plays the role of “addition,” while af corresponds
to “multiplication.”
(7.2) Proposition. Let f, g, h ∈ ξπ and a, b ∈ O. Then
(1) Ff (X, Y ) = Ff (Y, X),

(2) Ff (Ff (X, Y ), Z) = Ff (X, Ff (Y, Z)),
(3) af,g (Fg (X, Y )) = Ff (af,g (X), af,g (Y )),
(4) af,g (bg,h (Z)) = (a · b)f,h (Z),
(5) (a + b)f,g (Z) = Ff (af,g (Z), bf,g (Z)),
(6) (π n )f (Z) = f n (Z) , n = 0, 1, 2, . . .
To prove these formulas, one shows that the left and right side of each equation
are both solution of a problem as in (7.1), hence by uniqueness of the solution
they are equal. We leave the details to the reader.
For f = g = h we obtain from (1)–(6) the formal rules of an O-module.

Therefore we call Ff a formal Lie O-module. From such a formal Lie O-
module we obtain an ordinary O-module by letting the variables X, Y, Z attain
values from a domain in which the power series converge. If L is an arbitrary
algebraic extension of K, then the prime ideal pL of the elements of positive
valuation in L represents such a domain. In fact, if x1 , . . . , xn ∈ pL and
G(X1 , . . . , Xn ) ∈ O[[X1 , . . . , Xn ]], then the series G(x1 , . . . , xn ) converges and
gives an element in pL if the constant term of G is zero14) . We have so
(7.3) Proposition. Let f ∈ ξπ , and let L be an algebraic extension of K.

Then the set pL is an O-module with addition and multiplication defined by
x + y = Ff (x, y) and a · x = af (x), x, y ∈ pL , a ∈ O;
(f )
we write pL for this O-module.
Obviously, the additive inverse of x is (−1)f (x). One needs to be careful not
(f )
confuse the operations pL with the ordinary operations on the O-module.
(f )
(7.4) Proposition. The set of zeros Λf,n of f n (x) is a submodule of pLf,n .
Proof. The set of zeros Λf,n is the annihilator of the element π n ∈ O, since
(f )
Λf,n = {λ ∈ pLf,n | f n (λ) = (π n )f (λ) = 0} = {λ ∈ pLf,n | π n · λ = 0}.
14)
G(x1 , . . . , xn ) converges in the finite (and thus complete) extension of K generated
by x1 , . . . , xn .
(7.5) Proposition. Let f, g ∈ ξπ and a ∈ O. Then the map

λ 7−→ ag,f (λ)
yields a homomorphism from Λf,n to Λg,n . This homomorphism is an isomor-
phism if a is a unit in O.
Proof. This follows immediately from the formulas in (7.2). By (3) and (4)
this map is a homomorphism. If a is a unit, then (4) and (6) imply that
(a−1 )f,g (ag,f (λ)) = 1f (λ) = λ for λ ∈ Λf,n ,
and in reverse order
ag,f ((a−1 )f,g (λ)) = 1g (λ) = λ for λ ∈ Λg,n .
Hence the map ag,f : Λf,n → Λg,n is a bijection with inverse map (a−1 )f,g .
(7.6) Corollary. Let f ∈ ξπ . Then we have an isomorphism of O-modules

Λf,n ∼
= O/π n ·O.
Proof. If f, g ∈ ξπ , then the map 1g,f : Λf,n → Λg,n is an isomorphism by

(7.5). Thus it suffices to consider the module Λf,n with f (Z) = πZ + Z q ∈ ξπ .
The O-module Λf,1 consists of the zeros of the equation f (Z) = πZ + Z q = 0,
thus has q elements, and is therefore a one-dimensional vector space over the
field O/π·O. For n = 1 the result follows from the isomorphism Λf,1 ∼
= O/π O.
Assume that Λf,n ∼= O/π n ·O. By (7.5) the element π defines the homomor-
phism πf : Λf,n+1 → Λf,n , from which we obtain the exact sequence
πf
0 −→ Λf,1 −→ Λf,n+1 −−→ Λf,n −→ 0.
To see this, note first that πf (λ) ∈ Λf,n for λ ∈ Λf,n+1 , because f n (πf (λ)) =
f n (f (λ)) = f n+1 (λ) = 0. If λ ∈ Λf,n and λ∗ (∈ Ω) is a root of the equa-
tion f (Z) − λ = Z q + πZ − λ = 0, then λ∗ ∈ Λf,n+1 because f n+1 (λ∗ ) =
f n (f (λ∗ )) = f n (λ) = 0, thus πf (λ∗ ) = f (λ∗ ) = λ, and πf is surjective.
The kernel of πf consists of the elements λ with πf (λ) = f (λ) = 0, i.e., the
elements of the module Λf,1 ⊆ Λf,n+1 .
Since Λf,1 ∼
= O/π·O and Λf,n ∼ = O/π n ·O, the order of Λf,n+1 is equal to q n+1 . If
λ ∈ Λf,n+1 but λ 6∈ Λf,n , then π n+1 ·O is clearly the annihilator of λ. Hence the
map a 7→ a · λ gives an isomorphism between O/π n+1 ·O and the O-submodule
of Λf,n+1 generated by λ, which must coincide with Λf,n+1 , since O/π n+1 ·O
and Λf,n+1 both have order q n+1 . Therefore Λf,n+1 ∼ = O/π n+1 ·O.
(7.7) Corollary. Every automorphism of the O-module Λf,n is of the form

uf : Λf,n → Λf,n with a unit u ∈ UK . The map uf is the identity on Λf,n if
n n
and only if u ∈ UK . Thus the group UK /UK represents the full automorphism
group of Λf,n .
The proof follows easily from the isomorphism Λf,n ∼

= O/π n ·O and is left to
the reader.
(7.8) Proposition. The field Lf,n depends only on the prime element π, but
not on the choice of the power series f ∈ ξπ .
Proof. If f, g ∈ ξπ and λ ∈ Λf,n , then along with λ we have 1g,f (λ) ∈ Lf,n .
Since 1g,f : Λf,n → Λg,n is surjective, Λg,n ⊆ Lf,n , i.e., Lg,n ⊆ Lf,n . By
symmetry, Lf,n ⊆ Lg,n , and therefore Lf,n = Lg,n .
S∞
Because of this result we write Lπ,n for the field Lf,n , and set Lπ = n=1 Lπ,n .
We can always think of Lπ,n as generated by the roots of the polynomials
f n (Z) with f (Z) = πZ + Z q ∈ ξπ , hence Lπ,n |K is a normal extension.
We denote its Galois group by Gπ,n . The projective limit Gπ = lim Gπ,n is
←−
the Galois group of the extension Lπ |K. Each element σ ∈ Gπ,n yields an
automorphism of the O-module Λf,n by the usual operation of Gπ,n on the set
Λf,n ⊆ Lπ,n . This is due to the fact that σ acts continuously on Lπ,n , and that
the operations of the O-module Λf,n are defined by convergent power series
whose coefficients lie in the base field K, and therefore are fixed by σ. On the
n n
other hand, by (7.7) every class u · UK ∈ UK /UK yields the automorphism
uf : Λf,n → Λf,n . We show
(7.9) Theorem. For every σ ∈ Gπ,n there is a uniquely determined class

n n n
u·UK ∈ UK /UK such that σ(λ) = uf (λ), λ ∈ Λf,n . The map σ 7→ u · UK
yields an isomorphism
Gπ,n ∼
= UK /UKn
.
Proof. Each σ ∈ Gπ,n induces an automorphism of the O-module Λf,n . Since

n
by (7.7) UK /UK represents the full automorphism group of Λf,n , there is a
n n
(obviously unique) class u · UK ∈ UK /UK with σ(λ) = uf (λ) for all λ ∈ Λf,n .
n
The map σ 7→ u · UK is injective, since the set Λf,n generates the field Lπ,n ;
thus it follows immediately from σ(λ) = uf (λ) = λ for all λ ∈ Λf,n that σ = 1.
n
To prove σ 7→ u · UK is surjective, we show that the order of Gπ,n is not less
than the order q n−1 (q − 1) of UK /UK n
(cf. (3.2)). We have
f n (Z) = f (f n−1 (Z)) = f n−1 (Z) · φn (Z) with
n−1 q−1
φn (Z) = (f (Z)) + π ∈ O[Z].
n−1
Since all coefficients of the polynomial f n−1 (Z) = Z q + · · · + π n−1 Z have
positive valuation, φn (Z) is an Eisenstein polynomial and as such irreducible
over K. If λ is a root of φn (Z), and therefore a root of f n (Z), then K(λ) is a
totally ramified extension of Lπ,n . Its degree [K(λ) : K] is equal to the degree
q n − q n−1 = q n−1 (q − 1) of the polynomial φn (Z) = f n (Z)/f n−1 (Z), thus
equal to the order of Gπ,n , which is at least q n−1 (q − 1) = |UK /UK

n
|. There-
∼ n
fore Gπ,n = UK /UK and Lπ,n = K(λ), where λ is a root of the Eisenstein
polynomial φn (Z) = (f n−1 (Z))q−1 + π. This completes the proof.
The proof of (7.9) also shows
(7.10) Theorem. The extension Lπ,n |K is abelian and totally ramified of

degree q n−1 (q − 1), and is generated by a root of the Eisenstein equation
φn (Z) = (f n−1 (Z))q−1 + π = 0.
The last statement shows that the prime element π is a norm for every
extension Lπ,n |K. In fact, if λ is a root of φn (Z), then Lπ,n = K(λ) and
π = NLπ,n |K (−λ).
So far we have always fixed an arbitrarily chosen prime element π of K.

Now we have to investigate what happens when we pass from π to another
prime element π 0 . For this we use a lemma about the completion Tb of the
maximal unramified extension T over K. We again denote by ϕ the (universal)
Frobenius automorphism of T |K, whose restriction to an unramified finite
extension L|K yields the Frobenius automorphism ϕL|K (cf. §4, p. 88). If we
think of ϕ as continuously extended to the completion Tb, then we have the
Sublemma. UTϕ−1
b = UTb and (ϕ − 1)OTb = OTb .
Proof. It follows immediately from the definition of the Frobenius automor-

phism ϕ that the automorphism ϕ̄ of the algebraic closure of the residue field
Tb = T induced by ϕ satisfies the following equations
× × + +
(∗) (Tb )ϕ̄−1 = Tb and (ϕ̄ − 1)(Tb ) = Tb ,
× +
where Tb and Tb are the multiplicative and additive group of Tb. Furthermore
× + +
(∗∗) UTb /UT1b ∼
= Tb , UTnb /UTn+1
b
∼
= Tb and O Tb /pTb ∼
= pnTb /pn+1
Tb
∼
= Tb .
× +
Now if x ∈ UTb , resp. x ∈ OTb , then x̄ = ϕ̄ȳ1 /ȳ1 ∈ Tb , resp. x̄ = ϕ̄ ȳ1 − ȳ1 ∈ Tb ,
so that
ϕy1
x= a1 , y1 ∈ UTb , a1 ∈ UT1b , resp. x = ϕy1 − y1 + a1 , y1 ∈ OTb , a1 ∈ pTb .
y1
Because of (∗) and (∗∗) we obtain

ϕy2
a1 = a2 , y2 ∈ UT1b , a2 ∈ UT2b , resp. a1 = ϕy2 − y2 + a2 , y2 ∈ pTb , a2 ∈ p2Tb ,
y2
and therefore
ϕ(y1 · y2 )
x= · a2 , resp. x = ϕ(y1 + y2 ) − (y1 + y2 ) + a2 .
y1 · y2
Continuing in this way, we have
ϕ(y1 · · · yn )
x= · an , yn ∈ UTn−1 , an ∈ UTnb , resp.
y1 · · · yn b
x = ϕ(y1 + · · · + yn ) − (y1 + · · · + yn ) + an , yn ∈ pn−1

Tb
, an ∈ pnTb ,
and passing to the limit

∞ ∞
ϕy Y X
x= , y= yn ∈ UTb , resp. x = ϕy − y, y = yn ∈ OTb .
y n=1 n=1
Using this sublemma, we now prove a result similar to (7.1).
(7.11) Lemma. Let π and π 0 = u · π (u ∈ UK ) be two prime elements of

K, and let f ∈ ξπ , f 0 ∈ ξπ0 . Then there is a power series
θ(Z) ≡ εZ mod degree 2, ε a unit,
with coefficients in the ring O Tb of integers of Tb with the following properties:
(1) θϕ (Z) = θ(uf (Z)), 15)
(2) θ(Ff (X, Y )) = Ff 0 (θ(X), θ(Y )),

(3) θ(af (Z)) = af 0 (θ(Z)) for all a ∈ O.
Proof. By the sublemma we have u = ϕε/ε, ε ∈ UTb , and we set θ1 (Z) = εZ.
We assume that we have constructed a polynomial θr (Z) of degree r such that
θrϕ (Z) ≡ θr (uf (Z)) mod degree (r + 1),
and look for a polynomial θr+1 (Z) = θr (Z) + bZ r+1 , which satisfies the same
congruence with r + 1 instead of r. If we set b = a · εr+1 , we obtain for a the
condition a − ϕa = c/(ϕε)r+1 , where c is the coefficient of Z r+1 in the series
θrϕ (Z) − θr (uf (Z)). Because of the sublemma there always exists such an a;
thus we obtain θr+1 and therefore the series θ(Z) = limr→∞ θr (Z) satisfying
the condition θϕ (Z) = θ(uf (Z)).
15)
θϕ is the power series obtained from θ by applying ϕ to the coefficients of θ.
In order to obtain (2) and (3), we have to modify the θ we just constructed.
Consider the series
h = θϕ ◦ f ◦ θ−1 = θ ◦ uf ◦ f ◦ θ−1 = θ ◦ πf0 ◦ θ−1 ,
where the symbol ◦ stands for evaluation. Its coefficients lie in OTb , and because
hϕ = θϕ ◦ πf0ϕ ◦ θ−ϕ = θϕ ◦ f ◦ uf ◦ θ−ϕ = h, they even lie in O, since an element
in Tb which is fixed by ϕ lies in K, as one can easily verify. From this we get
h(Z) ≡ ε · π 0 · ε−1 Z = π 0 Z mod degree 2 and
−1 −1 −ϕ
ϕ
h(Z) = θ (f (θ ϕ
(Z))) ≡ θ (θ q ϕ
(Z) ) ≡ θ (θ (Z )) ≡ Z q
q
mod π 0 ,
so that h ∈ ξπ0 . Now we replace θ by 1f 0 ,h ◦ θ; then (1) still holds for this
modified θ, and we have f 0 = θϕ ◦ f ◦ θ−1 = θ ◦ πf0 ◦ θ−1 .
For the proof of (2), we show that the series
F (X, Y ) = θ(Ff (θ−1 (X), θ−1 (Y )))
satisfies the conditions of (7.1), which characterize the series Ff 0 (X, Y ). It
is clear that F (X, Y ) ≡ X + Y mod degree 2, a trivial calculation using the
formula f 0 = θ ◦ πf0 ◦ θ−1 shows that F (f 0 (X), f 0 (Y )) = f 0 (F (X, Y )), and by
the remark at the end of (7.1) the coefficients of F (X, Y ) lie in O.
Now (3) follows similarly, one shows that the series θ ◦ af ◦ θ−1 satisfies the
conditions of (7.1), which characterize the series af 0 .
(7.12) Corollary. Let π and π 0 = u · π be two prime elements in K, and let

f ∈ ξπ , f 0 ∈ ξπ0 . Then λ 7→ θ(λ) yields an isomorphism of O-modules
Λf,n ∼
= Λf 0 ,n .
Proof. Note first that if λ ∈ Λf,n , then θ(λ) ∈ Λf 0 ,n , because

n n
f 0 (θ(λ)) = (π 0 )f 0 (θ(λ)) = θ((un · π n )f (λ)) = θ(0) = 0.
That the map λ 7→ θ(λ) is a homomorphism follows immediately from the
formulas (2) and (3) of (7.11). If θ(λ) = 0, then we necessarily have λ = 0,
since otherwise 0 = ε + a1 λ + · · · , which is not possible because ε is a unit.
Therefore the map is injective. It is also surjective, since by (7.6) both Λf,n
n
and Λf 0 ,n are isomorphic to O/π n ·O = O/π 0 ·O, and therefore have the same
order.
For distinct prime elements π and π 0 of K, the fields Lπ,n and Lπ0 ,n may very
well be distinct. However, the previous corollary implies the following:
(7.13) Proposition. T ·Lπ,n = T ·Lπ0 ,n .
Proof. By (7.12) Λf 0 ,n = θ(Λf,n ) ⊆ Td ·Lπ,n (completion of T ·Lπ,n ). Since

·Lπ0 ,n ⊆ Td
Λf 0 ,n generates the field Lπ0 ,n , we have Td ·Lπ,n , and thus by symme-
try T ·Lπ0 ,n = T ·Lπ,n . Therefore T ·Lπ0 ,n = T ·Lπ,n , since both fields represent
d d
the algebraic closure of K in Td ·Lπ,n .
Since T is unramified and Lπ,n is totally ramified over K, T ∩ Lπ,n = K.

Therefore the Galois group GT ·Lπ,n |K of T ·Lπ,n |K is the direct product
GT ·Lπ,n |K = GT |K × Gπ,n .
We now define a homomorphism
ωπ : K × −→ GT ·Lπ,n |K
as follows: If a = u · π m ∈ K × , u ∈ UK , then let
ωπ (a) T
= ϕm ∈ GT |K ,
ωπ (a) Lπ,n
= σu ∈ Gπ,n ,
where σu is the automorphism on Lπ,n that corresponds by (7.9) to the class
u−1 ·UK
n
∈ UK /UK n
; in other words, the restriction ωπ (a) Lπ,n is determined
by ωπ (a)λ = (u−1 )f (λ), λ ∈ Λf,n .
We now come to the main goal of this section by showing that the homo-
morphism ωπ coincides with the one induced by the universal norm residue
symbol ( , K):
(7.14) Proposition. For every a ∈ K × , we have

ωπ (a) = (a, K)|T ·Lπ,n .
Proof. Since the prime elements of K obviously generate K × , it suffices to

prove the theorem for prime elements a. First let a = π. Then
ωπ (π)|T = ϕ = (π, T |K) = (π, K)|T (cf. (4.10))
and, since by (7.10) π is a norm from Lπ,n , we have
ωπ (π)|Lπ,n = σ1 = idLπ,n = (π, Lπ,n |K) = (π, K)|Lπ,n .
Therefore
ωπ (π) = (π, K)|T ·Lπ,n .
0
If π = u · π, u ∈ UK , is another prime element of K, then T · Lπ,n = T · Lπ0 ,n
by (7.13) and, we have again (cf. (4.10))
ωπ (π 0 )|T = ϕ = (π 0 , T |K) = (π 0 , K)|T .
Thus it remains to show that
ωπ (π 0 )|Lπ0 ,n = (π 0 , K)|Lπ0 ,n .
Now (π 0 , K)|Lπ0 ,n = (π 0 , Lπ0 ,n |K) = idLπ0 ,n , since by (7.10) π 0 is a norm of
Lπ0 ,n . This means that we have to verify that
ωπ (π 0 )|Lπ0 ,n = idLπ0 ,n ,
in other words, we have to show that ωπ (π 0 )λ0 = λ0 for λ0 ∈ Λf 0 ,n , where
f 0 ∈ ξπ0 . By (7.12) Λf 0 ,n = θ(Λf,n ), and our claim will follow if we can show
ωπ (π 0 )θ(λ) = θ(λ) for λ ∈ Λf,n .
For this, we consider
ωπ (π 0 ) = ωπ (u · π) = ωπ (u) ◦ ωπ (π).
From the above we have the following identities
ωπ (π)λ = λ (λ ∈ Λf,n ), ωπ (π)|T = ϕ, ωπ (u)|T = idT .
If we think of the last two automorphisms as continuously extended to Tb,

then we obtain from (7.11) the formula
ωπ (π 0 )θ(λ) = (ωπ (u) ◦ ωπ (π))θ(λ) = ωπ (u)θϕ (λ) = θϕ (ωπ (u)λ)
= θϕ ((u−1 )f (λ)) = θ(λ),
which completes the proof.
We can describe the norm residue symbol ( , Lπ,n |K) of the abelian and
totally ramified extension Lπ,n |K as follows:
(7.15) Theorem. If a = u · π m ∈ K × , u ∈ UK , then

(a, Lπ,n |K)λ = (u−1 )f (λ) for all λ ∈ Λf,n ⊆ Lπ,n .
n
The norm group of the extension Lπ,n |K is the group UK × (π).
Proof. It follows from (7.14) that (a, K)|T ·Lπ,n = ωπ (a). Therefore (a, K)λ =
(a, Lπ,n |K)λ = ωπ (a)λ = (u−1 )f (λ). Thus an element a = u · π m ∈ K × ,
u ∈ UK , is a norm of the extension Lπ,n |K if and only if (a, Lπ,n |K)λ =
(u−1 )f (λ) = λ for all λ ∈ Λf,n . By (7.7) this is equivalent to u ∈ UK
n
, i.e., to
n
a ∈ UK × (π).
As an application of (7.15) we discuss an example which can be considered

the starting point of the material presented in this section; this example is
also important for global class field theory (cf. the proof of (5.5) in Part III).
Let K = Qp be the field of p-adic numbers. Then p is a prime element in K,

and we choose for f ∈ ξp the polynomial

p 2
f (Z) = (1 + Z)p − 1 = p Z + Z + · · · + Z p,
2
so that n
f n (Z) = (1 + Z)p − 1.
The zero set Λf,n consists of the elements λ = ζ − 1, where ζ runs through the
pn -th roots of unity. Thus the field Lp,n is precisely the field of pn -th roots of
unity over Qp . Hence we have:
(7.16) Theorem. Let a = u · pm ∈ Q×

p , u a unit, and let ζ be a primitive
pn -th root of unity. Then
(a, Qp (ζ)|Qp )ζ = ζ r ,
where r is a positive integer which is mod pn determined by the congruence
r ≡ u−1 mod pn .
Proof. Set λ = ζ − 1 ∈ Λf,n . Then r · u ≡ 1 mod pn , and by (7.15) and (7.7)

(a, Qp (ζ)|Qp )λ = (u−1 )f (λ) = rf (λ).
On the other hand,
rf (Z) = (1 + Z)r − 1,
since this polynomial obviously satisfies the conditions of the definition
rf (Z) ≡ r Z mod degree 2 and f (rf (Z)) = rf (f (Z)).
Therefore
(a, Qp (ζ)|Qp )ζ = rf (λ) + 1 = rf (ζ − 1) + 1 = ζ r .
After this example we return to the general case. In (4.9) we have shown that
the norm groups of the unramified extensions L|K are the groups UK × (π f ).
We now characterize the norm groups in the totally ramified case.
(7.17) Theorem. The norm groups of the totally ramified (abelian) exten-
sions L|K are precisely the groups which contain the groups
n
UK × (π) (π prime element).
n
Proof. By (7.15), a group containing UK × (π) belongs to a subfield of Lπ,n ,
and thus to a totally ramified extension L|K. On the other hand, a totally
ramified extension L|K will be generated by a root λ of an Eisenstein equation
X e + · · · + π = 0,
where the prime element π is the norm of the element ±λ. Therefore we have
(π) ⊆ NL|K L× . Since NL|K L× is open in K × , we further have UK
n
⊆ NL|K L×
for an appropriate n, hence NL|K L× contains the group UK n
× (π).
(7.18) Corollary. Every totally ramified abelian field L|K is contained in

some Lπ,n .
In view of Theorem (6.4) we also note the following fact:
(7.19) Theorem. The group UK n

×(π f ) is the norm group of the field K 0 ·Lπ,n ,
0
where K |K is the unramified extension of degree f .
Obviously UK n
×(π f ) = (UK ×(π f ))∩(UK
n
×(π)) = NK 0 |K K 0× ∩NLπ,n |K L×
π,n =
0 ×
NK 0 ·Lπ,n |K (K · Lπ,n ) .
(7.20) Definition. Let L|K be an abelian extension, and let n be the smallest
n
integer ≥ 0 such that UK ⊆ NL|K L× . Then the ideal
f = pnK
is called the conductor of L|K 16) .
The conductor of the extension Lπ,n |K is the ideal f = pnK . For the unramified
extensions L|K, we have the
(7.21) Theorem. An abelian extension L|K is unramified if and only if it

has conductor f = 1.
The follows immediately from (4.9), L|K is unramified if and only if the norm
group NL|K L× has the form UK 0
× (π f ), thus if and only if f = p0K = 1.
The notion of a conductor is closely related to the discriminant and plays a

role in global class field theory17) .
We end this section with a brief discussion of the universal norm residue
symbol ( , K), which is characterized by the following theorem.
S∞
(7.22)STheorem. Let π be a prime element of K, f ∈ ξπ , Λf = n=1 Λf,n ,
∞
Lπ = n=1 Lπ,n = K(Λf ), and Gπ = GLπ |K .
The field T ·Lπ is (independent of π) the maximal abelian field over K. Thus
Gab
K = GT |K × Gπ .
If a = u · π m ∈ K × , u ∈ UK , then the norm residue symbol (a, K) is given by
(a, K)|T = ϕm , (a, K)λ = (u−1 )f (λ) for λ ∈ Λf .
Proof. By (6.4) the norm group of an abelian field L|K is a group containing
the group UKn
× (π f ). By (7.19) L is a subfield of a field K 0 ·Lπ,n , K 0 ⊆ T ,
hence a subfield of T ·Lπ , the maximal abelian extension of K.
16) 0
Here we set UK = UK .
17)
Cf. [3], [20].
Because Gab
K = GT |K × Gπ , the norm residue symbol (a, K) is determined by
the equations
(a, K)|T = ϕm , (a, K)λ = (u−1 )f (λ) for λ ∈ Λf ,
which hold by (4.10) and (7.15).
Part III
Global Class Field Theory
§ 1. Number Theoretic Preliminaries 113
§ 1. Number Theoretic Preliminaries
We assume the reader is familiar with the basic concepts and theorems of
algebraic number theory for which we refer to the standard text books, for
example, [6], [21], [30]. Nevertheless, in this section we briefly summarize the
for us most important facts.
If K is a finite algebraic number field, we mean by the primes p of K the classes
of equivalent valuations of K, where we distinguish between the finite and
infinite primes p. The finite primes p are associated with the nonarchimedian
valuations of K and correspond bijectively to the prime ideals of the field
K, for which we use the same symbol p. For the infinite primes we have
to distinguish further between the real and complex infinite primes. The real
primes correspond bijectively to the different embeddings of K into the field IR
of real numbers, while the complex primes correspond bijectively to the pairs
of complex conjugate embeddings of K into the field C of complex numbers;
we observe that two conjugate embeddings of K into C produce the same
valuation of K. We write p - ∞ (resp. p | ∞) if p is finite (resp. infinite).
If p is a finite prime, we denote by vp the exponential valuation of K associ-
ated with p, normalized with smallest positive value 1. We get an additional
normalized valuation if we associate with each prime p its p-absolute value
| |p . This is done in the following way:
1) If p is finite and p is the rational prime lying under p, then for a ∈ K,

a 6= 0, let |a|p = N(p)−vp (a) = p−fp ·vp (a) . Here N(p) denotes the absolute
norm of the ideal p, thus the number pfp of elements in the residue field
of p; fp is the inertia degree, i.e., the degree of the residue field of p over
its prime field (cf. II, §3, p. 79).
2) If p is real infinite and ι is the embedding of K into the field IR associated
with p, then we set |a|p = |ιa|, a ∈ K.
3) If p is complex infinite and ι is one of the pair of conjugate embeddings
of K into the field C of complex numbers associated with p, then we let
|a|p = |ιa|2 , a ∈ K.
With this normalization of the (multiplicative) valuations of K we have

|a|p = 1 (a ∈ K r {0}) for almost all primes p, and the fundamental product
formula
Y
|a|p = 1, a ∈ K × . 1)
1)
Cf. [21], III, §20, p. 314. As always, K × denotes the multiplicative group of the
field K.
114 Part III. Global Class Field Theory
If S is a finite set of primes of the field K containing all infinite primes, then
K S = {a ∈ K × | vp (a) = 0 (i.e., |a|p = 1) for all p 6∈ S}
is called the group of S-units of K. In particular, if S = S∞ is the set of all
infinite primes of K, then K S∞ is the usual unit group of K. We have the
following generalization of Dirichlet’s
(1.1) Unit Theorem2) . The group K S is finitely generated, and its rank is
equal to |S| − 1, where |S| denotes the number of primes in S.
We write JK for the group of ideals of K, and PK ⊆ JK for the group of

principal ideals. The factor group JK /PK is called the ideal class group
of K. It satisfies the
(1.2) Theorem3) . The ideal class group JK /PK is finite; its order h is called
the class number of the field K.
For every prime p of K we have the completion Kp of K with respect to the

valuation associated with p. If p is finite, then Kp is a p-adic number field. The
prime p is real (resp. complex) infinite if and only if Kp = IR, (resp. Kp = C).
We set (
unit group of the field Kp , if p is finite;
Up =
Kp× , if p is infinite.
It is convenient to define the unit groups Up for infinite primes as well, since we
do not want to always have to distinguish between finite and infinite primes.
If L|K is a finite extension of a number field K, we denote the primes in the

extension field L by P. If P is a prime of L lying above the prime p of K, then
we write P | p for short. In this case the completion LP of L by P contains
the field Kp , since the restriction of the valuation associated with P from L
to K yields the valuation of the field K associated with p.
¤¥¦§ LP We have illustrated this situation in the
adjacent diagram. The transition from the
“global” extension L|K to the “local” ex-
L Kp tensions LP |Kp at the individual primes is
the fundamental principle behind class field
theory.
K
e0
If p is a prime ideal of K and p = Pe · · · P0 is the prime decomposition of p
in the extension field L, then b
p=P b e , where b
p (resp. P),
b denotes the prime
2)
Cf. [21], III, §28, p. 528.
3)
Cf. [21], III. §29, p. 542.
§ 1. Number Theoretic Preliminaries 115
ideal of the field Kp (resp. of the extension field LP ). Moreover, P

b has the
same degree over b p as P over p. If P runs through all the primes of L lying
over p, then we have the fundamental equation of number theory
X
[LP : Kp ] = [L : K].
P|p
Let L|K be a finite normal field extension with Galois group G = GL|K . If
σ ∈ G, then along with P | p we also have σP | p, where σP is the prime of
L conjugate to P with respect to σ.
¨©ª«¬ LP LσP If we complete L with respect to P and

σP, then Kp is contained in LP as well
L Kp
as in LσP , since p lies under P as well as
under σP.
K
There is a canonical Kp -isomorphism
σ
LP −→ LσP
between LP and LσP which we also denote by σ. In fact, if α ∈ LP , thus

α = P–lim αi for some sequence αi ∈ L, then the sequence σαi ∈ L converges
in LσP with respect to σP, and the canonical isomorphism is obtained from
α = P– lim αi ∈ LP 7−→ σα = σP– lim σαi ∈ LσP .
Under this isomorphism the field Kp is obviously fixed elementwise. In par-
ticular, in case P = σP we obtain a Kp -automorphism
σ
LP −→ LP ,
and therefore an element of the Galois group GLP |Kp of LP |Kp . This auto-
morphism is simply the continuous extension of the automorphism σ of L to
the completion LP . If we observe that P = σP if and only if σ is an element
of the decomposition group GP ⊆ G of P over K, we see that for each σ ∈ GP
there is a corresponding element in GLP |Kp . Conversely, every automorphism
of GLP |Kp yields an automorphism in GP by restriction to the field L. This
yields a canonical isomorphism between the Galois group GLP |Kp of the lo-
cal extension LP |Kp and the decomposition group GP , and we can identify
GLP |Kp and GP ; therefore we can consider GLP |Kp as a subgroup of G, i.e.,
GLP |Kp = GP ⊆ G. In what follows we will identify these groups without
further mentioning.
The Theory of Kummer Extensions. Later we will apply the following

observations to algebraic number fields; however, since they hold for arbitrary
fields, we state them in full generality.
Let K be a field containing the n-th roots of unity but whose characteristic
does not divide n. By a general Kummer extension of K we mean a (finite
or infinite) Galois extension L of K whose Galois group GL|K is abelian and
has exponent n, i.e., has the property that σ n = 1 for all σ ∈ GL|K . It is easy
to see that the compositum of two Kummer extensions L1 and L2 of K is
again a Kummer extension. Therefore the union N of all abelian extensions
of K with a Galois group of exponent n is the largest Kummer extension of
K. The following theorem shows that one can read off the structure of the
Kummer extensions of K from the multiplicative group K × of K.
(1.3) Theorem. There exists an inclusion preserving isomorphism between

the lattice of Kummer extensions L of K and the lattice of certain subgroups
∆ of K × . Under this isomorphism the group ∆ = (L× )n ∩ K √
×
⊇ (K × )n cor-
n
responds to the Kummer extension L, and the field L = K( ∆) corresponds
to the group ∆, (K × )n ⊆ ∆ ⊆ K × :
L 7−→ ∆ = (L× )n ∩ K × ,
√
∆ 7−→ L = K( n ∆).
The factor group ∆/(K × )n is isomorphic to the character group χ(GL|K ) of

the Galois group GL|K .
Proof. Let L be a Kummer extension of K and G = GL|K its Galois group.

Then G has exponent n and acts trivially on the group µn of n-th roots of
unity which lies in K. Hence χ(G) = Hom(G, µn ) = H 1 (G, µn ). If we associate
with every x ∈ L× its n-th power xn ∈ (L× )n , we obtain the exact sequence
n
1 −→ µn −→ L× −→ (L× )n −→ 1
and from the resulting cohomology sequence the connecting homomorphism
G n δ
L× = K × → ((L× )n )G = (L× )n ∩ K × → H 1 (G, µn ) → H 1 (G, L× ) = 1 4)
,
which yields the isomorphism
(L× )n ∩ K × /(K × )n ∼
= H 1 (G, µn ) = χ(G).

It is easy to check that this isomorphism takes a class a · (K × )√

n
∈ ((L
√
× n
) ∩
× × n 1
K )/(K ) to the character χa ∈ H (G, µn ) with χa (σ) = σ( a)/ a. n n
In particular, if we consider the maximal Kummer extension N over K with

the Galois group G = GN |K , then (N × )n ∩ K × = K √
×
, since otherwise there
would exist an element a ∈ K whose n-th root a would not lie in N × ,
× n
and therefore would generate over N a still larger Kummer extension over K.
4)
Note that H 1 (G, L× ) = 1 by II, (2.2)). Of course, the exactness of this sequence
can be shown without reference to cohomology, and one can develop the theory
of Kummer extensions without using cohomological methods.
§ 2. Idèles and Idèle Classes 117
Consequently we have an isomorphism

K × /(K × )n ∼
= H 1 (G, µn ) = χ(G).
By Pontryagin Duality, there is an inclusion reversing lattice isomorphism
between the lattice of closed subgroups of G and the lattice of subgroups
of χ(G). Since χ(G) ∼ = K × /(K × )n , we have by Galois theory an inclusion
preserving isomorphism between the lattice of Kummer extensions and the
lattice of subgroups of K × which contain (K × )n . If the field L corresponds
to the group ∆ in this sense, then
1. ∆ = (L× )n ∩ K × : If
√
×
a ∈ K√ , then a ∈ ∆ ⇔ χa√ (σ) = 1 (χa ∈ χ(GN |K ))
for σ ∈ GN |L ⇔ σ( a) = a for σ ∈ GN |L ⇔ n a ∈ L ⇔ a ∈ (L× )n .
n n
√
2. L = K( n ∆): If σ ∈ G = GN |K , then σ ∈ GN |L ⇔ χa (σ) = 1 for a ∈
√ √
∆ ⇔ σ( n a) = n a for a ∈ ∆ ⇔ σ K( √ n
∆)
= idK( √
n √
∆) ⇔ σ ∈ GN |K( ∆) .
n
§ 2. Idèles and Idèle Classes
In the following we will consider idèles5) which were first introduced by

C. Chevalley. The notion of idèles is a slight modification of the notion
of ideals, or, more precisely, of divisors. Its significance lies in the fact that it
permits a transition between global and local number theory, and therefore
represents a suitable mean for applying the local-global principle, which is a
method to obtain theorems and definitions in global class field theory from
local class field theory. The development of the global theory using idèles to-
gether with cohomological methods is particularly transparent, and has led
to a plethora of far-reaching results. As a result, the analytic methods, i.e.,
Dirichlet series and their generalizations, which were necessary in the classical
ideal theoretic treatment of class field theory, have disappeared6) .
Let K be an algebraic number field. An idèle a of K is a family a = (ap ) of

elements ap ∈ Kp× such that p ranges over all primes of K, but ap is a unit in
Kp for almost all primes p. We also obtain these idèles by the following
(2.1) Definition. Let S be a finite set of primes of K. The group

Y Y Y
S
IK = Kp× × Up ⊆ Kp×
p∈S p6∈S p
5)
The idèles were first called ideal elements. Then this was abbreviated to id. el.
from which the name idel (in French idèle) evolved.
6)
Yet even today these methods represent an essential counterpart to the approach
we follow here, and have not lost their importance.
is called the group of S-idèles of K. The union

[ Y
IK = S
IK ⊆ Kp× ,
S p
where S runs through all finite sets of primes of K, is the idèle group of K.
If a = (ap ) ∈ IK , ap ∈ Kp× , then the ap are the local components of the
idèle a; an ap ∈ Kp× is an essential component of a if ap is not a unit.
In particular, an idèle has at most finitely many essential components. The

S-idèles are precisely those idèles which have essential components at most at
the primes in the set S.
The reason why we allow units as “nonessential” components at almost all
primes of an idèle is that this allows us to canonically embed the multiplicative
group K × of the field K into the idèle group IK of K:
If x ∈ K × , then we let (x) ∈ IK be the idèle whose components are (x)p =
x ∈ Kp× . Observe that x is a unit in Kp for almost all primes p. In this way
we always consider K × as embedded in IK , thus view K × as a subgroup of
IK . The idèles in K × are called the principal idèles of K.
If S is a finite set of primes of K, then we denote by
K S = K × ∩ IK
S S
⊆ IK
the group of S-principal idèles. The elements in K S are also called the S-units
of K, since they are units for all primes p 6∈ S (cf. §1, p. 114). In particular, if
S = S∞ is the set of all infinite primes of K, then K S∞ is the ordinary unit
group of the field K.
(2.2) Definition. The factor group

CK = IK /K ×
is called the idèle class group of the field K.
In our development of class field theory the group CK is the main object of
interest. The connection between the idèles and the ideals of a field K is given
in the following proposition.
(2.3) Proposition. Let S∞ be the set of all infinite primes of the field K,
S∞
and let IK be the group of idèles which have units as components at all finite
primes. Then we have a canonical isomorphism
IK /I S∞ ∼
K = JK , IK /I S∞ ·K × ∼
K = JK /PK ,
where JK and PK denote the group of ideals and principal ideals respectively.
Proof. If p is a finite prime of K, we let vp be the valuation of Kp normalized

with minimal positive value 1. If a ∈ IK is an idèle of K, then we have ap ∈ Up
for almost all finite primes, and therefore vp ap = 0. We obtain from the map
Y
a 7−→ pvp ap ,
p-∞
where p ranges over all finite primes of K, a canonical homomorphism from

IK onto JK . Here a is mapped to the unit ideal if and only if vp ap = 0, i.e.,
S∞
ap ∈ Up for all p - ∞, thus if and only if a ∈ IK . Hence the kernel of this
S∞
homomorphism is IK . On the other hand, we have the homomorphism
Y
a 7−→ pvp ap · PK
p-∞
from IK onto JK /PK , and a is in its kernel if and only if p-∞ pvp ap ∈ PK , i.e.,
Q
vp ap
= (x) = p-∞ pvp x with x ∈ K × , thus if and only if vp ap = vp x,
Q Q
p-∞ p
S∞
vp (ap ·x−1 ) = 0 for all p - ∞. This again is the case if and only if a · x−1 ∈ IK ,
S∞ S∞ ×
thus a ∈ x·IK , i.e., if and only if a ∈ IK ·K .
S∞
The group IK /IK is none other than the well-known group of fractional
ideals of K. It is easy to see that when passing to ideals, the S-idèles become
precisely those ideals that are generated only by prime ideals in S.
Unlike the ideal class group JK /PK , the idèle class group CK = IK /K × is
not finite. However, the finiteness of the ideal class group is reflected in the
S
fact that all idèle classes in CK can be represented by S-idèles a ∈ IK for a
fixed finite set S of primes. This is the assertion of the following proposition.
(2.4) Proposition. Let S be a sufficiently large finite set of primes. Then

S
IK = IK ·K ×, and therefore CK = IK
S
·K × /K ×.
Proof. The ideal class group JK /PK is finite (cf. (1.2)). Hence we can choose a
finite set of ideals A1 , . . . , An which represent the classes in JK /PK . The ideals
A1 , . . . , An are further made up of only finitely many prime ideals p1 , . . . , ps .
Now if S is any finite set of primes containing the primes p1 , . . . , ps and all
the infinite primes of K, then in fact
S
IK = IK · K ×.
S∞ ∼
In order to see this, consider the isomorphismQ IKvp/I K = JK (cf. (2.3)). If
ap
a ∈ IK , then the corresponding ideal A = p-∞ p lies in a class Ai · PK ,
i.e., A = Ai · (x), where (x) ∈ PK denotes the principal ideal given by x ∈ K × .
0
The idèle aQ = a · x−10 is mapped under the homomorphism IK → JK onto the
ideal A0 = p-∞ pvp ap = Ai . Since the prime ideal components of Ai lie in the
set S, we have vp a0p = 0, i.e., a0p ∈ Up for all p 6∈ S; thus a0 = a · x−1 ∈ IK

S
,
S ×
a ∈ IK · K .
We now investigate what happens when we pass to an extension field.
Let L|K be a finite extension of algebraic number fields. If p is a prime of K

and P a prime of L lying over p, we write simply P | p. The idèle group IK
of K is embedded in the idèle group IL of L as follows: If we map an idèle
a ∈ IK to the idèle a0 ∈ IL with the components
a0P = ap ∈ Kp ⊆ LP for P | p ,
then we obtain an injective homomorphism
IK −→ IL .
This homomorphism allows us to think of IK as being embedded in IL , and
to regard IK as a subgroup of IL . With this identification, an idèle a ∈ IL is
in the group IK if and only if its components aP lie in Kp (where P | p), and
moreover any two primes P and P0 lying over the same prime p of K have
equal components aP = aP0 ∈ Kp .
If L|K is normal and G = GL|K denotes its Galois group, IL is canonically

a G-module: An element σ ∈ G defines a canonical isomorphism from Lσ−1 P
onto LP , which we also denote by σ (cf. §1, p. 115). Here we associate with
an idèle a ∈ IL with components aP ∈ L×
P the idèle σa ∈ IL with components
(σa)P = σaσ−1 P ∈ LP .
Note that aσ−1 P ∈ Lσ−1 P is the σ −1 P-component of a, which is mapped by σ
into LP . If we take into account that the P-component (σa)P of σa is essential
if and only if the σ −1 P-component aσ−1 P of a is essential, we immediately see
that when passing to ideals, the map induced by a 7→ σa is just the ordinary
conjugation map on the ideal group JL .
(2.5) Proposition. Let L|K be normal with Galois group G = GL|K . Then
ILG = IK .
Proof. The inclusion IK ⊆ ILG is easy. If σ ∈ G, then the isomorphism

σ
Lσ−1 P → LP is a Kp -isomorphism (P | p), and if a ∈ IK is considered as an
idèle of IL , then (σa)P = σaσ−1 P = σaP = aP ∈ Kp , i.e., σa = a.
For the inclusion ILG ⊆ IK , consider a ∈ IL with σa = a for all σ ∈ G.
Then (σa)P = σaσ−1 P = aP for all primes P of L. By §1, p. 115 we can
consider the decomposition group GP of P over K as the Galois group of
the extension LP |Kp . For every σ ∈ GP we have σ −1 P = P, and because
aP = σaσ−1 P = σaP , we obtain aP ∈ Kp (P | p). Hence if σ is an arbitrary
element of G, then (σa)P = aP = σaσ−1 P = aσ−1 P ∈ Kp , i.e., two primes
P and σ −1 P lying above the same prime p of K have the same components
aP = aσ−1 P ∈ Kp , so that a ∈ IK .
It is well known that an ideal of a field K can very well become a principal
ideal in an extension field L without being a principal ideal in the base field K.
The following proposition shows that the idèles behave differently.
(2.6) Proposition. If L|K is an arbitrary finite extension, then

L× ∩ IK = K × .
In particular, if a ∈ IK is an idèle of K that becomes a principal idèle in the

extension L, i.e., a ∈ L× , then a is already principal in K.
Proof. The inclusion K × ⊆ L× ∩ IK is trivial. Let L̃ be a finite normal

extension of K containing L, and let G̃ = GL̃|K be its Galois group. Then IK
and IL are subgroups of IL̃ . If a ∈ L̃× ∩ IK , then (2.5) shows that a ∈ IL̃G̃ , i.e.,
σa = a for all σ ∈ G̃, and because a ∈ L̃× , we even have a ∈ (L̃× )G̃ = K × .
Therefore L̃× ∩ IK = K × , which implies L× ∩ IK ⊆ L̃× ∩ IK = K × .
By (2.6) we can embed the idèle class group CK of a field K into the idèle
class group CL of a finite extension field L using the canonical homomorphism
ι : CK −→ CL , a · K × 7→ a · L× (a ∈ IK ⊆ IL ).
To see that ι is injective note that if the class a · K × ∈ CK is mapped to the
unit class L× ∈ CL , hence a · L× = L× , a ∈ L× , then we know by (2.6) that
a ∈ L× ∩ IK = K × , i.e., a · K × = K × is the unit class of CK .
In the following we view CK as embedded in CL via this canonical map,
hence as a subgroup of CL . An element a · L× ∈ CL (a ∈ IL ) lies in CK if and
only if the class a · L× contains a representative a0 from IK (⊆ IL ) such that
a0 · L× = a · L× .
(2.7) Theorem. Let L|K be with Galois group G = GL|K . Then CL is

canonically a G-module, and
CLG = CK .
Proof. If a · L× ∈ CL (a ∈ IL ), we set σ(a·L× ) = σa·L× . This definition is

clearly independent of the choice of a ∈ IL , and makes CL a G-module.
From the exact sequence of G-module 1 → L× → IL → CL → 1 we obtain

the exact cohomology sequence (cf. I, (3.4))
1 −→ (L× )G −→ ILG −→ CLG −→ H 1 (G, L× ),
where (L× )G = K × , ILG = IK , and H 1 (G, L× ) = 1; hence CLG = CK .
We briefly summarize the most important results from this section.

Let K be an algebraic number field. Then
S
Kp× ×
Q Q
IK = p∈S p6∈S Up the group of S-idèles of K
(S a finite set of primes of K),
S
S
IK = S IK the idèle group of K,
×
K ⊆ IK the group of principal idèles,
×
CK = IK /K the idèle class group of K,
S ×
IK = IK ·K for a sufficiently large finite set S
of primes.
If L|K is a finite extension of algebraic number fields, we have embeddings

IK ⊆ IL idèle groups,
× ×
K ⊆L groups of principal idèles,
CK ⊆ CL idèle class groups.
If L|K is a finite normal extension with Galois group G = GL|K , then L× ,
IL and CL are G-modules whose G-invariants are given by
G
L× = K ×, ILG = IK , CLG = CK .
§ 3. Cohomology of the Idèle Group
Let L|K be a finite normal extension with Galois group G = GL|K . We

consider the cohomology groups H q (G, IL ) of the G-module IL . These co-
homology groups reveal a particular advantage of working with idèles since
they can in a certain sense be completely “localized ” , i.e., decomposed into
a direct product of cohomology groups over the local fields Kp . The goal of
this section is to explain this natural localization process.
Let S be a finite set of primes of the base field K and S̄ the finite set of primes
of the extension field L above the primes in S. To simplify things, we denote
the group of S̄-idèles ILS̄ of L also by ILS , and speak of the S-idèles of the field
L; we will use the same convention in later sections as well. Thus we have
§ 3. Cohomology of the Idèle Group 123

Y Y YY YY
ILS = L×
P× UP = L×
P × UP .
P|p∈S P|p6∈S p∈S P|p p6∈S P|p
We consider the products ILp = P|p L× p

Q Q
P and UL = P|p UP as subgroups
of IL , where we think of the elements in IL (resp. in ULp ), as those idèles
S p
which have the component 1 at all the primes of L not lying above p (resp.
and in addition have only units as components at the primes of L lying above
p). Since the automorphisms σ ∈ G only permute the primes P above p, the
groups ILp and ULp are G-modules. Thus we have decomposed ILS into a direct
product of G-modules:
Y p Y p
ILS = IL × UL .
p∈S p6∈S
About the G-modules ILp and ULp we have the following
(3.1) Proposition. Let P is a prime of L lying over p. Then

H q (G, I p ) ∼
= H q (GP , L× ),
L P
where GP is the decomposition group of P over K, considered also as the
Galois group of LP |Kp . If p is a finite unramified prime in L, then for all q
H q (G, ULp ) = 1.
Addendum. The above isomorphism above is given by the composition

res π̄
H q (G, ILp ) −−→ H q (GP , ILp ) −→ H q (GP , L×
P) ,
π
where π̄ is induced by the canonical projection ILp −→ L×
P that takes each
idèle in ILp to its P-component.
Proof. If σ ∈ G runs through a system of representatives of the cosets G/GP ,

we write for simplicity σ ∈ G/GP , then σP runs through all distinct primes
of L above p. Hence
Y Y Y Y
ILp = L×σP = σL×P and UL =
p
UσP = σUP ,
σ∈G/GP σ∈G/GP σ∈G/GP σ∈G/GP
which shows that and ILp ULp

are G/GP -induced G-modules. Applying Shapiro’s
Lemma I, (4.19) yields
H q (G, I p ) ∼
L = H q (GP , L× ) and H q (G, U ) ∼
P
p
= H q (GP , UP ),
L
where the isomorphism H q

(G, ILp )
→H q ×
(GP , LP ) is
the composition of the
homomorphisms res and π̄ described in the addendum. If p is unramified in L,
then the extension LP |Kp is unramified, and we can refer to local class field
theory (cf. II, (4.3)) to obtain the result H q (G, ULp ) ∼
= H q (GP , UP ) = 1.
Because of Proposition (3.1) and the decomposition ILS = p∈S ILp × p6∈S ULp
Q Q
the cohomology groups of the idèle groups ILS and IL are easy to compute. By
I, (3.8) we have
H q (G, ILS ) ∼
Y Y
= H q (G, ILp ) × H q (G, ULp ).
p∈S p6∈S
If the finite set S contains all (finite) primes of K which are ramified in L, then
by (3.1) H q (G, ILp ) = H q (GP , L× q p
P ) (P any prime above p), and H (G, UL ) = 1
for each p 6∈ S. Therefore
H q (G, I S ) ∼
Y
= L H q (GP , L× ), P
p∈S
where P denotes any prime above p.

Because IL = S ILS , we also have
S
H q (G, IL ) ∼ H q (G, ILS ) ∼ )∼

Y M
= lim = lim H q (GP , L× = H q (GP , L×
P)
7)
,
−→ −→ P
S S p∈S p
where S runs through all finite sets of primes of K which contain all ramified
primes 8). Thus we have proved the following theorem:
(3.2) Theorem. Let S be a finite set of primes of K which contains all primes
ramified in L. Then
H q (G, ILS ) ∼
Y
= H q (GP , L×
P) ,
p∈S
H q (G, IL ) ∼
M
= H q (GP , L×
P ).
7)
Here P denotes any prime above p.
From the proof and the Addendum (3.1) we obtain further the
Addendum. The isomorphism H q (G, IL ) = ∼ L H q (GP , L× ) is given by the

p P
projections H q (G, IL ) −→ H q (GP , L×
P ), i.e., the composition of the maps
res π̄
H q (G, IL ) −−→ H q (GP , IL ) −→ H q (GP , L×
P) ,
π
where π̄ is induced by the canonical projection IL −→ L×
P which takes each
idèle a to its P-component aP .
7) L
By the symbol we mean the direct sum, i.e., the (here multiplicative) group
of families (. . . , cp ,Q. . . ), in which only finitely components cp not equal to 1 ap-
pear. By contrast means the direct product, i.e., the group of all families
(. . . , cp , . . . ).
8)
One can prove H q (G, IL ) ∼ q ×
L
= p H (GP , LP ) directly without forming the
somewhat mysterious limit: View an element of H q (G, IL ) as an element of
H q S
Q (G, ILq ) for an appropriate S, and Q mapq it by the isomorphism H q (G, ILS ) ∼
=
× ×
p∈S H (GP , LP ) into the group p∈S H (GP , LP ), which also can be viewed
as a subgroup of p H q (GP , L×
L
P ). It is easy to verify that this is an isomorphism.
The above projections map each element c ∈ H q (G, IL ) to its p-components

cp ∈ H q (GP , L× P ) (as always, P is a fixed prime above p). The theorem
says that each c is uniquely L determined by its local components cp , which
because of the direct sum are almost all equal to 1. In dimensions q > 0 the
map c 7→ cp can be described in the following simple way. Given a cohomology
class c ∈ H q (G, IL ), choose a cocycle a(σ1 , . . . , σq ) representing c. This is
a function on the group G with values in the idèle group IL . Restrict this
function to the group GP , and take the P-components aP (σ1 , . . . , σq ) of the
idèle a(σ1 , . . . , σq ). The resulting function from GP to L× P is a cocyle, and its
cohomology class cp ∈ H q (GP , L× P ) is the p-component of c.
The following proposition shows how changing the field affects taking local
components.
(3.3) Proposition. Let N ⊇ L ⊇ K be normal extensions of K, and let

P0 | P | p be primes of N , L, and K respectively. Then
(inf N c)p = inf NP0 (cp ), c ∈ H q (GL|K , IL ), q ≥ 1,
(resL c)P = resLP (cp ), c ∈ H q (GN |K , IN ),
c ∈ H q (GN |L , IN ).
P
(corK c)p = P|p corKp (cP ),
For the last two formulas it suffices to assume that only N |K is normal.
For the third formula note that for each P | p we choose a prime P0 of N lying
above P, thus the corestrictions corKp (cP ) lie in (a priori) distinct cohomology
×
groups H q (GNP0 |Kp , NP 0 ). However, we can identify these as follows: Given
two primes of N lying over p, there is an automorphism σ ∈ GN |K that

× σ ×
interchanges these primes; given this, the isomorphism NP 0 → NσP0 induces
q × ∼ q ×
a canonical isomorphism H (GNP0 |Kp , NP0 ) = H (GNσP0 |Kp , NσP0 ) (cf. also
(3.1)). Hence we may view corKp (cP ) for each P | p as an element of the
× 0
group H q (GNP0 |Kp , NP 0 ) for a fixed choice of P | p, and form the sum in this
group.
The proof of Proposition (3.3) uses the general and purely cohomological fact
that the restriction map which occurs when passing to the local components
commutes with the maps inf, res, and cor. This is easy to see at the cocycle
level for inf and res if q ≥ 1, and for cor if q = −1, 0. The general case follows
from this by dimension shifting. The details are left to the reader.
With Theorem (3.2) we have achieved a complete “localization” of the co-

homology of the idèle groups, i.e., instead of the groups H q (G, IL ) we may
consider the local cohomology groups H q (GP , L×P ), a principle, which we will
frequently use below. We make the important remark that Theorem (3.2)
is not a deep number-theoretic result. Apart from purely cohomological re-

sults, the only number-theoretical facts that are being used in its proof is
the finiteness of the number of ramified primes in the extension L|K and the
cohomological triviality of the unit group UP in an unramified local extension
LP |Kp . Nevertheless, Theorem (3.2) is fundamental for the idèle-theoretic de-
velopment of global class field theory. In dimension q = 0 it allows us to prove
the following corollary, which we refer to as the Norm Theorem for Idèles:
(3.4) Corollary. An idèle a ∈ IK is the norm of an idèle b of IL if and only

if each component ap ∈ Kp× is the norm of an element bP ∈ L× P (P | p), i.e.,
if and only if it is a local norm everywhere.
Proof. Note that H 0 (G, IL ) = ILG /NG IL = IK /NG IL and H 0 (GP , L× P) =

Kp× /NGP L× . Thus by Theorem (3.2) we have I /N I ∼ L
K ×
/N L×
P K G L = p p GP P.
If a ∈ IK , then this isomorphism takes the 0-cohomology class a · NG IL = a
to its components ap , which by the Addendum (3.2) can be computed as
ap = ap · NGP L× P . Now, since we have an isomorphism, a = 1 if and only if
ap = 1, i.e., a ∈ NG IL if and only if for every component ap ∈ NGP L× P.
The Norm Theorem for Idèles is an analogue of the “Hasse Norm Theorem”,
which states that if L|K is a cyclic extension, an element x ∈ K × is the norm
of an element y ∈ L× if and only if it is everywhere a local norm, i.e., is a norm
for every extension LP |Kp (cf. §4, (4.8)). Contrary to the Norm Theorem for
idèles, the Hasse Norm Theorem is a very deep number-theoretic result, for
which up to date we do not have a direct proof. Corollary (3.4) only says
that the element x ∈ K × , considered as a principal idèle, is the norm of an
idèle b of L; it leaves open the question whether this idèle can be chosen as a
principal idèle y ∈ L× .
(3.5) Corollary. H 1 (G, IL ) = H 3 (G, IL ) = 1.
This follows from (3.2), since H 1 (GP , L× 1 ×

P ) = H (GP , LP ) = 1 for all P (cf.
II, (2.2) and II, (5.8)).
The fact that H 1 (GL|K , IL ) = 1 implies that the extensions L|K form with
respect to the idèle groups IL a field formation in the sense of II, §1. This
allows us to think of the cohomology groups H 2 (GL|K , IL ) as the elements of
[
H 2 (GΩ|K , IΩ ) = H 2 (GL|K , IL ), 9)
L
9)
Here Ω denotes the field of all algebraic numbers; however H 2 (GΩ|K , IΩ ) is used
only as notation for the union on the right. If we consider IΩ as the union of all
IL , or more precisely IΩ = lim IL , then IΩ is a GΩ|K -module, and we can define
−→
H 2 (GΩ|K , IΩ ) directly also for infinite Galois groups GΩ|K (cf. [41]).
where the inclusions are given by the injective (because H 1 (GL|K , IL ) = 1)

inflation maps. We will use this interpretation in all further considerations
below. In particular, if N ⊇ L ⊇ K are two normal extensions of K, then
H 2 (GL|K , IL ) ⊆ H 2 (GN |K , IN ) ⊆ H 2 (GΩ|K , IΩ ).
In
S local class field theory we have seen that the Brauer group Br (K) =
2 ×
L H (GL|K , L ) of a p-adic number field K is the union of the cohomol-
ogy groups H (GL|K , L× ) of the unramified extensions L|K, for which it is
2
relatively easy to prove the reciprocity law. The role of the unramified exten-
sions in the local theory is played in the global case by the cyclic cyclotomic
field extensions, i.e., cyclic extensions which are contained in a field which
is formed by adjoining roots of unity. We show already at this point the
(3.6) Theorem. Let K be a finite algebraic number field. Then

[ [
Br (K) = H 2 (GL|K , L× ) and H 2 (GΩ|K , IΩ ) = H 2 (GL|K , IL ),
L|K cyclic L|K cyclic
where L|K ranges over all cyclic cyclotomic extensions.
For the proof we use the following
(3.7) Lemma. Let K be a finite algebraic number field, S a finite set of

primes of K, and m a natural number. Then there exists a cyclic cyclotomic
field L|K with the property that
• m | [LP : Kp ] for all finite p ∈ S,

• [LP : Kp ] = 2 for all real-infinite p ∈ S.
Proof. It suffices to prove the lemma for K = Q, the general case follows
from this by taking compositia. More precisely, if N |Q is a totally imaginary
cyclic cyclotomic field such that for every prime number p above which there
is a prime of K in S the degree [NP : Qp ] is divisible by m · [K : Q], then
L = K · N has the desired property.
Let ln be a prime power and let ζ be a primitive ln -th root of unity. If l 6= 2,
then the extension Q(ζ)|Q is cyclic of degree ln−1 · (l − 1), and we denote the
cyclic subfield of degree ln−1 by L(ln ).
If l = 2, then the Galois group of Q(ζ)|Q is the direct product of a cyclic
group of order 2 and a cyclic group of order 2n−2 . In this case we consider the
field L(2n ) = Q(ξ) with ξ = ζ − ζ −1 . The automorphisms of Q(ζ) are defined
n−1
by σν : ζ 7→ ζ ν , ν odd, and we have σν (ξ) = ζ ν − ζ −ν . Because ζ 2 = −1,
n−1
σν (ξ) = σ−ν+2n−1 (ξ), and since either ν or −ν + 2 ≡ 1 mod (4), the
automorphisms of L(2n ) = Q(ξ) are induced by those σν with ν ≡ 1 mod (4).
Now an elementary calculation shows that the Galois group of L(2n )|Q is
cyclic of order 2n−2 . Moreover, because σ−1 ξ = −ξ, the field L(ln ) is totally
imaginary for large n.
If p is a prime number, then as n increases the local degree [L(ln )P : Qp ]
becomes an arbitrarily large l-th power, since in any case [Qp (ζ) : Qp ] becomes
arbitrarily large, and we have [Qp (ζ) : L(ln )P ] ≤ l − 1, resp. ≤ 2 in case l = 2.
Now if m = l1r1 · · · lsrs , then the field
L = L(l1n1 ) · · · L(lsns ) · L(2t )
has the desired properties, provided the ni , t are chosen sufficiently large. In
fact, for the finitely many prime numbers p ∈ S the local degrees [LP : Qp ] are
divisible by every power liri , and therefore by m; L is totally complex because
of the factor L(2t ), and L is cyclic over Q, since the L(ln ) are cyclic over Q
with relatively prime degrees.
Proof of (3.6): We only give the proof for H 2 (GΩ|K , IΩ ); the assertion on
the Brauer group follows from that on H 2 (GΩ|K , IΩ ) as soon as we know that
Br(K) → H 2 (GΩ|K , IΩ ) is injective. This will follow from H 1 (GΩ|K , CΩ ) = 0
(Theorem III (4.7)).
Let c ∈ H 2 (GΩ|K , IΩ ), say c ∈ H 2 (GL0 |K , IL0 ), let m be the order of c and
let S be the (finite) set of primes p of K, for which the local components cp
of c are not equal to 1. By the previous lemma there is a cyclic cyclotomic
field L|K with m | [LP : Kp ] for the finite p ∈ S and [LP : Kp ] = 2 for the
real-infinite p ∈ S. If we form the compositum N = L0 · L, then we have
H 2 (GL0 |K , IL0 ) and H 2 (GL|K , IL ) ⊆ H 2 (GN |K , IN ),
and we will show that c lies in the group H 2 (GL|K , IL ). Since the sequence
res
1 −→ H 2 (GL|K , IL ) −→ H 2 (GN |K , IN ) −−−→
L
H 2 (GN |L , IN )
is exact, it suffices to show that resL c = 1. But by local class field theory,
together with (3.2) and (3.3), we have resL c = 1 ⇔ (resL c)P = resLP cp = 1
for all primes P of L ⇔ invNP0 |LP (resLP cp ) = [LP : Kp ] · invNP0 |Kp cp =
[LP :Kp ] [LP :Kp ]
invNP0 |Kp cp = 0 for all primes p of K ⇔ cp = 1 for all p ∈ S.
Now the last equality holds, because cm p = 1 and m | [LP : Kp ] for the finite
primes, and [LP : Kp ] = 2 for the real-infinite p ∈ S.
§ 4. Cohomology of the Idèle Class Group
The role of the multiplicative group of a field in the local theory is taken
by the idèle class group in global class field theory. Thus our aim is to show
§ 4. Cohomology of the Idèle Class Group 129
that there is a canonical reciprocity isomorphism between the abelianization

of the Galois group G = GL|K of a normal extension L|K of finite algebraic
number fields and the norm residue group CK /NG CL ; in other words: that
the finite normal extensions L|K of an algebraic number field K constitute a
class formation in the sense of II, §1 with respect to the idèle class groups CL .
In particular, we will have to prove that H 1 (G, CL ) = 1 and that H 2 (G, CL )
is cyclic of order [L : K]. This will follow from the so-called first and second
fundamental inequalities, which we prove below. The ideal-theoretic versions
of these inequalities were already used in the proof of the main theorems of
class field theory by Takagi (cf. [17], Teil I). Their proofs required, however, a
considerable effort; in particular, the second inequality is proved using mainly
Dirichlet series, thus analytic methods, which are redundant when working
with idèles.
In what follows we fix a normal extension L|K with a cyclic Galois group G =
GL|K of prime order p. The first fundamental inequality is the relation
(CK : NG CL ) ≥ p.
It follows immediately from the following
(4.1) Theorem. The idèle class group CL is a Herbrand module with Her-
brand quotient
|H 0 (G, CL )|
h(CL ) = = p 10) .
|H 1 (G, CL )|
From this we obtain as a
(4.2) Corollary.
|H 0 (G, CL )| = (CK : NG CL ) = |H 2 (G, CL )| = p · |H 1 (G, CL )| ≥ p.
Remark. If we knew that H 1 (G, CL ) = 1, then Corollary (4.2) would imme-

diately imply |H 2 (G, CL )| = p, thus that H 2 (G, CL ) is cyclic of order [L : K].
However, contrary to the cases where instead of CL one considers the idèle
group IL (cf. (3.5)), or the multiplicative group L× (Hilbert-Noether Theo-
rem), this is not easy to show. That indeed H 1 (G, CL ) = 1 will follow only
from second fundamental inequality (CK : NG CL ) = |H 0 (G, CL )| ≤ p proved
in the next section. Because of the isomorphism H 1 (G, CL ) ∼ = H −1 (G, CL ),
1
it is easy to see that the statement H (G, CL ) = 1 is actually equivalent to
Hasse’s Norm Theorem (cf. (4.8)) mentioned in the previous section. This
is the reason why a direct proof of Hasse’s Norm Theorem would be very
desirable; however, to date we do not have such a proof.
10)
See I, §6 for more on Herbrand quotients.
Proof of Theorem (4.1). Let S be a finite set of primes of K such that

1. S contains all infinite primes and all primes ramified in L,
2. IL = ILS ·L× ,
3. IK = IKS
·K × .
Note that by (2.4) such a set S certainly exists. Then we have
CL = I S ·L× /L× ∼
L = I S /LS ,
L
where LS = L× ∩ILS is the group of S-units, i.e., the group of all those elements
in L× which are units for all the primes P of L which do not lie above the
primes in S (cf. §2, p. 118 and §1, p. 114). From I, (6.4) we obtain
h(CL ) = h(ILS ) · h(LS )−1 ,
in the sense that when two of these Herbrand quotients are defined, then so
is the third, and we have equality.
The proof splits into two parts, i.e., the computations of h(ILS ) and h(LS ).
Because of Theorem (3.2), the computation of h(ILS ) is a local question. Let
n the number of primes in S,
N the number of primes of L, which lie above S, and
n1 the number of primes in S, which are inert in L.
Since [L : K] has prime degree, a prime of K that is not inert splits completely,
i.e., decomposes into exactly p primes of L; thus N = n1 + p · (n − n1 ).
To compute the quotient h(ILS ) = |H 0 (G, ILS )| / |H 1 (G, ILS )|, we have to deter-
mine |H 0 (G, IQS 1 S
L )| and |H (G, IL )|. We do this making use of the isomorphism
H q (G, ILS ) ∼
= p∈S H q
(GP , L×
P ) from Theorem (3.2).
1 S
If q = 1, the above isomorphism immediately yields Q H (G,0 IL ) = ×
1, because
1 × 0 S ∼
H (GP , LP ) = 1. If q = 0, then H (G, IL ) = p∈S H (GP , LP ), and it
remains to determine the order of H 0 (GP , L× P ), which is done using local
class field theory. In fact, we have H 0 (GP , L× ∼
P = GP (cf. II, (5.9)), so that
)
(
1, if the prime p lying under P splits (because GP = 1),
|H 0 (GP , L×P )| =
p, if p is inert (because GP = G).
Hence |H 0 (G, ILS )| = pn1 ; since H 1 (G, ILS ) = 1, we have so h(ILS ) = pn1 .
For the computation of h(LS ) we use the formula for the Herbrand quotient
from Theorem I, (6.10). By (1.1), the group LS = L× ∩ ILS of S-units of L is
finitely generated of rank N − 1, and its fixed group (LS )G = K S = K × ∩ LS
is the group of S-units of K and finitely generated of rank n − 1. Theorem I,
(6.10) yields
h(LS ) = p(p(n−1)−N +1)/(p−1) = pn1 −1 .
Since both Herbrand quotients h(ILS ) and h(LS ) are defined, h(CL ) is also
defined, and the above formulas imply h(CL ) = h(ILS ) · h(LS )−1 = p.
Theorem (4.1) has the following
(4.3) Corollary. Let L|K be a cyclic extension of prime power degree. Then
K has infinitely many primes which are inert in L.
Proof. First let the degree [L : K] = p be a prime number. We assume

that the set U of the primes of K remaining inert in L is finite. We show
that under this assumption CK = NG CL (G = GL|K ), contradicting the first
fundamental inequality (4.2). Let a ∈ CK and let a ∈ IK be a representative
idèle of a with local components ap ∈ Kp× . The group of p-th powers (Kp× )p is
open in Kp× with respect to the valuation topology by II, (3.6). Therefore for
each p ∈ U, ap ·(Kp× )p is an open neighborhood of the element ap , and since
the field K is dense in its completion Kp , we can find an xp ∈ K × which lies in
this neighborhood: xp ∈ ap ·(Kp× )p . The Approximation Theorem in valuation
theory implies further that there exists an x ∈ K, which approximates xp
arbitrarily closely with respect to the prime p for all p ∈ U. In particular, we
may assume that along with xp we also have x ∈ ap ·(Kp× )p , and therefore
ap ·x−1 ∈ (Kp× )p for all p ∈ U. We now claim that the idèle a0 = a·x−1 is
the norm of an idèle b in IL . By (3.4) this is the case if and only if every
component a0p ∈ Kp× is the norm of an element bP ∈ L× P (P | p). For p ∈ U
this is true, because [LP : Kp ] = p and ap = ap ·x ∈ (Kp× )p , and for p 6∈ U
0 −1
this is trivially the case, since p splits completely because of the prime degree,
and therefore LP = Kp . Thus we have a0 = a·x−1 = NG b, b ∈ IL , from which
we get a = a · K × = a0 · K × = NG b · K × = NG (b · L× ). Hence CK = NG CL .
Now let L|K be cyclic of degree pr . We assume that almost all primes of K
split in L. This means that the decomposition fields ZP are proper extensions
of K for almost all primes P of L, and therefore in each case contain a field L0
between K and L of degree p. But in the cyclic extension L|K there is only one
field L0 of degree p, which is therefore contained in almost all decomposition
fields ZP . This implies that almost all primes p of K decompose in the cyclic
extension L0 of degree p, which contradicts the first part of the proof.
We now prove the second fundamental inequality (CK : NG CL ) ≤ p

for cyclic extensions L|K of prime degree, making the additional assumption
that K contains the p-th roots of unity. In this case L is a Kummer extension:
√
L = K( p x0 ), x0 ∈ K × . We start with the following lemma:
√
(4.4) Lemma. Let N = K( p x), x ∈ K × , be any Kummer extension over
K, and let p be a finite prime of K not lying over the prime number p. Then
p is unramified in N if and only if x ∈ Up · (Kp× )p , and p splits completely in
N if and only if x ∈ (Kp× )p .
√
Proof. Let P be a prime of N over√ p. Then
√ NP = Kp ( p x). If x = u · y p ,
u ∈ Up , y ∈ Kp× , then NP = Kp ( p x) = Kp ( p u). If the equation X p − u = 0
is irreducible over the residue field of Kp , then it is also irreducible over Kp ,

and NP |Kp is an unramified extension of degree p. If X p − u = 0 is reducible
over the residue field of Kp , then it splits into p distinct linear factors there,
since p is distinct from the characteristic of the residue field, and by Hensel’s
Lemma X p − u = 0 also splits into linear factors over Kp , so that NP = Kp .
In both cases NP |Kp is unramified, i.e., p is unramified in N .
√
Conversely, if p is√unramified in N , then NP = Kp ( p x) is unramified over
k
Kp , and we have p x = u · π , where u ∈ UP and π ∈ Kp is a prime element
(of smallest value 1). Thus we have x = up · π k·p , and therefore up ∈ Up ,
π k·p ∈ (Kp× )p , i.e., x ∈ Up · (Kp× )p .
√
The prime p decomposes in N if and only if NP = Kp ( p x) = Kp , hence if
and only if x ∈ (Kp× )p .
(4.5) Theorem. Let L|K be a cyclic extension of prime degree p. Assume

the field K contains the p-th roots of unity. Then
|H 0 (G, CL )| = (CK : NG CL ) ≤ p.
The difficulty here is that that we cannot a priori decide which idèle classes
in CK are represented by a norm idèle, and therefore lie in NG CL . This is
completely different from the case of idèle groups, where by the Norm Theorem
for idèle groups a ∈ IK is a norm if and only if it is a local norm everywhere
(cf. (3.4)). We work around this by considering instead of NG CL an auxiliary
group F which is constructed such that its elements are represented by norm
idèles, hence F ⊆ NG CL , and which has the property that its index (CK : F )
can actually be shown to be equal to p. Using this F , we obtain the inequality
(CK : NG CL ) ≤ (CK : F ) = p.
√
Let L = K( p x0 ), x0 ∈ K × . Let S be a finite set of primes of K such that
1. S contains all the primes above p and all infinite primes of K,
S
2. IK = IK · K ×,
S
3. x0 ∈ K = IK S
∩ K × (i.e., x0 is an S-unit).
Here 2. can be satisfied by (2.4), and 3. because x0 is a unit for almost all
primes.
Together with S we choose m additional primes q1 , . . . , qm 6∈ S that split
completely in L; set S ∗ = S ∪ {q1 , . . . , qm }. To construct F , we have to specify
an idèle group F ⊆ IK whose elements represent the idèle classes of F . It must
consist of nothing but norm idèles so that F ⊆ NG CL , it must be sufficiently
large to ensure that the index (CK : F ) is finite, and it must be simple enough
so that it is possible to compute this index. These properties are satisfied by
the idèle group
Y m
Y Y
F = (Kp× )p × Kq×i × Up 11)
.
p∈S i=1 p6∈S ∗
To see that F ⊆ NG IL , it suffices by the Norm Theorem for idèles to convince

ourselves that the components ap of each idèle a ∈ F are norms from the
extension LP |Kp (P | p).
This is true for p ∈ S, because ap ∈ (Kp× )p ⊆ NLP |Kp L× P (regardless of
[LP : Kp ] = p or = 1); this is trivially true for p = qi , because qi splits
completely so that LP = Kp ; and it is true for p 6∈ S ∗ , because x0 ∈ Up by 3.
√
and therefore by Lemma (4.4) each p 6∈ S ∗ is unramified in L = K( p x0 ), so
that ap ∈ Up ⊆ NLP |Kp L× × ×
P by II, (4.4). If we now set F = F ·K /K , then
F ⊆ NG CL , since each idèle class a is represented by a norm idèle a ∈ F . To
compute the index (CK : F ), we consider the following decomposition:
∗ ∗
S
(CK : F ) = (IK ·K × /K × : F ·K × /K × ) = (IKS
·K × : F ·K × ) =
∗ ∗
∩ K × ) : (F ∩ K × ) 12) .
S S

(IK : F )/ (IK
It allows us to split the computation of (CK : F ) into two parts, the compu-
S∗
tation of (IK : F ), which is of a purely local nature, and the computation of
∗
S
((IK ∩ K × ) : (F ∩ K × )), which uses global considerations.
S∗
: F ) = p∈S (Kp× : (Kp× )p ); since S ⊆ S × , the map
Q
I. We have (IK
S∗
Y Y
IK −→ Kp× /(Kp× )p with a 7−→ ap · (Kp× )p
p∈S p∈S
∗
S
is trivially surjective, and its kernel consists precisely of those idèles a ∈ IK
× p
for which ap ∈ (Kp ) for p ∈ S lie in the kernel; i.e., the idèles in F . By the
local theory (cf. II, (3.7)) we have
(Kp× : (Kp× )p ) = p2 · |p|−1
p ,
∗
S
: F ) = p2n · p∈S |p|−1
Q
so that (IK p , where n is the number of primes in S.
Since the primes p 6∈ S do notQlie above theQprime number p, |p|p = 1 for p 6∈ S,
S∗
and by the product formula p∈S |p|p = p |p|p = 1, hence (IK : F ) = p2n .
11)
That we consider precisely this idèle group is motivated as follows: If we start
heuristically with the reciprocity√ law, which has not been proved yet, we see that
p
the Kummer extension T = K( Q K S ) has as norm
Q group the idèle class group
E = E ·K × /K × formed from E = p∈S (Kp× )p × p6∈S Up (cf. (7.7)). By inserting
additional factors Kq×i , i.e., choosing suitable primes qi , we try to enlarge E to a
√
group F such that F becomes the norm group of the field L = K( p x0 ) ⊆ T .
12)
The last of these equations results from a general elementary group-theoretical
fact: If B ⊆ A, C are subgroups of an abelian group, then the canonical surjective
homomorphism A/B → A·C/B·C has kernel A ∩ B·C/B ∼ = A ∩ C/B ∩ C.
II. An elementary calculation shows that

∗ ∗
S
((IK ∩ K × ) : (F ∩ K × )) = (K S : (F ∩ K × ))
∗ ∗ ∗
= (K S : (K S )p )/((F ∩ K × ) : (K S )p ),
∗
where K S is the group of S ∗ -units. By (1.1) this group is finitely generated of
∗
rank n + m − 1 (n + m is the number of primes in S ∗ ). Moreover, K S contains
∗ ∗
the p-th roots of unity, and it easily follows that (K S : (K S )p ) = pn+m .
Altogether we therefore have
∗
(CK : NG CL ) ≤ (CK : F ) = pn−m · ((F ∩ K × ) : (K S )p ),
and the second fundamental inequality is proved, provided that we can choose
the primes q1 , . . . , qm splitting in L in such a way that m = n − 1, and
m
K × ∩ F = K × ∩ ( (Kp× )p × Kq×i ×
Q Q Q
Up ) =
p∈S i=1 p6∈S ∗
m
= K× ∩ (Kp× )p ∩ Kq×i ∩
T T T
Up =
p∈S i=1 p6∈S ∗
S∗ p
= K× ∩ (Kp× )p ∩
T T
Up = (K ) ;
p∈S p6∈S ∗
using Lemma (4.4), we formulate this as follows:
Sublemma. There exist n − 1 primes of K, q1 , . . . , qn−1 6∈ S that split

completely in L and satisfy the following condition:
√
If N = K( p x) is a Kummer extension over K in which all p√∈ S split
completely and all p 6= q1 , . . . , qn−1 are unramified, then N = K( p x) = K.
In fact, the desired equality

\ \ ∗
K× ∩ (Kp× )p ∩ Up = (K S )p
p∈S p6∈S ∗
follows immediately from this. The inclusion

√ ⊇ is trivial. Let x ∈ K × ∩
T × p T
p∈S (Kp ) ∩ p6∈S ∗ Up and N = K( x). By (4.4) every p ∈ S splits com-
p
pletely in N , since x ∈ (Kp× )p . For p 6∈ S ∗ we have x ∈ Up ⊆ Up · (Kp× )p , so

∗
√ p 6∈ S is unramified in N by (4.4). Hence the sublemma yields
that every
N = K( p x) = K, so that x ∈ (K × )p , and because x ∈ Up for p 6∈ S ∗ , x lies
∗ ∗
in (K × )p ∩ K S = (K S )p .
The statement about the prime decomposition in Kummer extensions given

by the sublemma represents the global part of the proof of the √ second funda-
p
mental inequality. To prove it, we consider the field T = K( K S ) obtained
by adjoining the p-th roots of all the elements in the group K S . By (1.3)
χ(GT |K ) ∼
= K S ·(K × )p /(K × )p ∼
= K S /(K S )p .
Since K S is finitely generated of rank n − 1 = |S| − 1 (cf. (1.1)) and contains

the p-th roots of unity, we have K S ∼ = (ζ)×ZZ×· · ·×ZZ; also K S /(K S )p ∼
= (ζ)×
ZZ/pZZ × · · · × ZZ/pZZ (ζ a primitive p-th root of unity). Thus the Galois group
GT |K is the direct product of cyclic groups Zi of order p, GT |K = Z1 ×· · ·×Zn ,
and we have for the field degree [T : K] = (K S : (K S )p ) = pn .
√
The field L = K( p x0 ) lies in T because x0 ∈ K S , and it is easy to see that
we can assume GT |L = Z1 × · · · × Zn−1 . Let Ti ⊆ T be the fixed field of
Zi ⊆ GT |K , thus GT |Ti = Zi .
Because Zi ⊆ GT |L we have L ⊆ Ti for
®¯°±² T i = 1, . . . , n − 1, and the given fields form an
arrangement as in the diagram on the left.
Now we choose for each i = 1, . . . , n a prime
T1 Tn−1 Tn
Qi of the field Ti , such that the Qi are inert
in T , and such that the primes q1 , . . . , qn ly-
L ing under the Qi are all distinct and are not
in S. This is possible by (4.3). We claim that
K the q1 , . . . , qn−1 6∈ S satisfy the conditions
of the lemma.
To prove that q1 , . . . , qn−1 split completely in L, we observe that Ti is the
decomposition field of the unique extension Q0i of Qi to T over K, i = 1, . . . , n.
This decomposition field Zi is contained in Ti , since Q√i is inert in T . On the
other hand, by (4.4), qi is unramified in every field K( p x), x ∈ K S , thus also
in T , so that the Galois group GT |Zi of T |Zi is isomorphic to the Galois group
of the residue field extension of T |Zi , and therefore is cyclic. But a generator
of GT |Zi has order p as an element of GT |K , hence [T : Zi ] = p and Zi = Ti .
Since L is contained in the decomposition fields Ti for i = 1, . . . , n − 1, it
follows that the primes q1 , . . . , qn−1 split completely in L.
Set Ui = Uqi , i = 1, . . . , n. We show next that the homomorphism
Y n n
Y
K S /(K S )p −→ Ui /(Ui )p , x · (K S )p 7→ x · (Ui )p (x ∈ K S ),
i=1 i=1
is bijective. For injectivity note that if x ∈ (Ui )p ⊆ (Kq×i )p , then it √follows
from (4.4)
√ that the primes q1 , . . . , qn split completely in the field K( p x), so
that√K( x)p
Tnis contained in the decomposition fields Ti , i = 1, . . . , n. Hence
K( p x) ⊆ i=1 Ti = K, and x ∈ (K × )p ∩ K S = (K S )p .
We show surjectivity by comparing orders. We know that (K S : (K S )p ) = pn ;
on the other hand, using II, (3.8) we find (Ui : (Ui )p ) = p · |p|−1
qi = p, i.e.,
Qn
(K S : (K S )p ) and i=1 Ui /(Ui )p have the same order pn .
√
Now let N = K( p x), x ∈ K × be a Kummer extension in which the p ∈ S split
completely and the p 6= q1 , . . . , qn−1 are unramified. To prove that N = K,
it suffices by (4.2) to show that CK = NN |K CN . Let a ∈ CK = IK /K × =
IKS
·K × /K × , and let a ∈ IKS
be a representative of the class a. If we set
ai = aqi · (Ui )p (aqi ∈ Ui ), i = 1, . . . , n, then the fact that K S /(K S )p −→

Q n p S p
i=1 Ui /(Ui ) is surjective implies that there is a y ∈ K with y · (Ui ) = ai ,
p 0 −1
so that aqi = y · ui , ui ∈ Ui for i = 1, . . . , n. The idèle a = a · y belongs to
the same idèle class as a, and it follows easily from from the Norm Theorem
for idèles that moreover a0 ∈ NN |K IN : If p ∈ S, then a0p is a norm, since
p splits completely in N , and therefore NP = Kp (P | p). For the primes
qi , i = 1, . . . , n − 1, a0qi = upi as the p-th power of a norm, and for p 6∈ S,
p 6= q1 , . . . , qn−1 , the norm property of a0p follows by II, (4.4) from the fact
that a0p ∈ Up is a unit and by assumption p is in N , i.e., NP |Kp (P | p) is
unramified. Therefore we obtain a = a0 · K × ∈ NN |K CN , i.e., N = K. This
proves the sublemma and consequently completes the proof of Theorem (4.5).
Theorems (4.1) and (4.5) together imply
(4.6) Corollary. If L|K is a cyclic extension of prime degree p with Galois

group G = GL|K , and K contains the p-th roots of unity, then
H 0 (G, CL ) ∼
= H 2 (G, CL ) ∼
= G and H 1 (G, CL ) = 1.
From what we have shown for Kummer extensions, it is easy to prove the
following, more general result.
(4.7) Theorem. If L|K is a normal extension with Galois group G = GL|K ,

then we have H 1 (G, CL ) = 1.
We prove this by induction on the order n of the group of G. The case n = 1 is

trivial. Let us assume that H 1 (G, CL ) = 1 for every extension L|K of degree
< n. If the order n = |G| is not a p-power, then each p-Sylow subgroup Gp of G
has order smaller than n, so that by the induction hypothesis H 1 (Gp , CL ) = 1,
and therefore H 1 (G, CL ) = 1 by I, (4.17).
Thus is suffices to prove this for a p-group G. In this case, let g ⊆ G be a
normal subgroup of index p; g is the Galois group of an intermediate field M ,
K ⊆ M ⊆ L, g = GL|M . Now if p < n, then by assumption H 1 (G/g, CM ) =
H 1 (g, CL ) = 1, and from the exact sequence (cf. I, (4.6))
inf res
1 −→ H 1 (G/g, CM ) −−→ H 1 (G, CL ) −−→ H 1 (g, CL )
we see that H 1 (G, CL ) = 1.
³´µ¶ L0 Assume p = n. In order to be able to apply Corollary

(4.6), we replace K by the extension K 0 obtained by
L adjoining a primitive p-th root of unity to K, and set
L0 = L · K 0 . Obviously [K 0 : K] ≤ p − 1 < p = n and
K0 [L0 : K 0 ] = p. Because [K 0 : K] < n, resp. (4.6), we
have H 1 (GK 0 |K , CK 0 ) = H 1 (GL0 |K 0 , CL0 ) = 1, and
K from the exact sequence
inf res
1 −→ H 1 (GK 0 |K , CK 0 ) −−→ H 1 (GL0 |K , CL0 ) −−→ H 1 (GL0 |K 0 , CL0 )
we obtain H 1 (GL0 |K , CL0 ) = 1. On the other hand, because the sequence

inf
1 −→ H 1 (G, CL ) −−→ H 1 (GL0 |K , CL0 ) = 1
is also exact, we see that H 1 (GL0 |K , CL0 ) = 1 implies H 1 (G, CL ) = 1.
For cyclic extensions Theorem (4.7) is just another form of the Hasse Norm
Theorem mentioned earlier:
(4.8) Corollary. If the extension L|K is cyclic, then an element x ∈ K × is

a norm if and only if it is locally a norm everywhere.
Proof. The sequence of G-modules 1 → L× → IL → CL → 1 yields the exact

cohomology sequence
H −1 (G, CL ) −→ H 0 (G, L× ) −→ H 0 (G, IL ) ∼
M
= H 0 (GP , L× ). P
p
Since G is cyclic, H −1 (G, CL ) ∼

= H 1 (G, CL ) = 1 by Theorem (4.7), which
implies that the canonical homomorphism
M
K × /NL|K L× −→ Kp× /NLP |Kp L×
P
p
is injective; this is precisely the assertion of the Hasse Norm Theorem.
(4.9) Theorem. Let L|K be a normal extension with Galois group G = GL|K .
Then the order of H 2 (G, CL ) is a divisor of the degree [L : K].
We prove this again by induction on the order n of the group G. For n = 1

the theorem is trivial, and we assume it holds for all normal extensions of
degree less than n. If the order of |G| = n is not a prime power, every p-
Sylow subgroup Gp of G has a smaller order that n, and by assumption the
order |H 2 (Gp , CL )| divides the order np of Gp , i.e., the maximal p-power
dividing n. If Hp2 (G, CL ) denotes the p-Sylow subgroup of H 2 (G, CL ) 13) ,
then the restriction map
res
Hp2 (G, CL ) −−→ H 2 (Gp , CL )
is injective by I, (4.16). Hence the order |Hp2 (G, CL )| divides the maximal
p-power np which divides n, and since H 2 (G, CL ) is a direct product of its
p-Sylow groups, |H 2 (G, CL )| is a divisor of n.
13)
Hp2 (G, CL ) consists precisely of those elements of H 2 (G, CL ) which have p-power
order; Hp2 (G, CL ) is often referred to as the p-primary component of H 2 (G, CL ) .
Thus we can again assume G is a p-group. Choose a normal subgroup g ⊆ G

of index p. Because |g| = n/p < n, the order of H 2 (g, CL ) is a divisor of n/p,
and taking into account that H 1 (g, CL ) = 1, I, (4.7) shows that the sequence
inf res
1 −→ H 2 (G/g, CLg ) −−→ H 2 (G, CL ) −−→ H 2 (g, CL )
is exact. But G/g is the Galois group of a cyclic extension of prime de-
gree L0 |K, K ⊆ L0 ⊆ L, i.e., G/g = GL0 |K , CLg = CL0 , and by (4.6)
we have |H 2 (G/g, CLg )| = p. It follows from the above exact sequence that
|H 2 (G, CL )|/p divides n/p, and therefore |H 2 (G, CL )| divides n, as claimed.
With Theorem (4.9) we have not yet reached our goal to show that H 2 (G, CL )
is cyclic of the same order as [L : K]. To show this, we will associate with the
group H 2 (G, CL ) an invariant homomorphism, as required by Axiom II for
class formations.
§ 5. Idèle Invariants
Recall that we want to show that the extensions L|K form a class formation
in the sense of II, §1 with respect to the idèle class group CL . With Theorem
(4.7) we have shown that Axiom I is satisfied. Hence it remains to show that
for every normal extension L|K there is an invariant isomorphism
H 2 (GL|K , CL ) −→ 1
[L:K] Z
Z/ZZ
which satisfies the compatibility properties required by Axiom II. It is, of
course, essential that we construct the invariant isomorphism in a canoni-
cal way to also obtain a canonical reciprocity law, the Artin reciprocity
law. In a certain sense we will retrieve this invariant map, and with it the
reciprocity law, from the local theory, by relating the group H 2 (GL|K , CL )
to the group H 2 (GL|K , IL ) formed with the idèle group IL as the un-
derlying module. L For the latter group we obtain from the decomposition
H 2 (GL|K , IL ) ∼
= 2 ×
p H (GLP |Kp , LP ) immediately an invariant map from
local class field theory by taking the sum of the canonical invariant isomor-
phisms of the local extensions LP |Kp . We will show that the invariants of the
elements in H 2 (GL|K , IL ) yield invariants for the elements of H 2 (GL|K , CL ).
Let L|K be a normal extension of finite algebraic number fields, and let GL|K
be its Galois group. By (3.2) we have the decomposition
H 2 (GL|K , IL ) ∼
M
= H 2 (GLP |Kp , L×
P ),
p
L
where again denotes the direct sum. For every prime p of K we have from
local class field theory the isomorphism (cf. II, (5.5))
§ 5. Idèle Invariants 139
invLP |Kp : H 2 (GLP |Kp , L×

P ) −→
1
[LP :Kp ] Z
Z/ZZ ⊆ 1
[L:K] Z
Z/ZZ (P | p) 14)
The local invariant isomorphism invLP |Kp is the composition of three homo-
morphisms; however, for now it is not necessary for us to know this map
explicitly, at this point it is only important that we know that it satisfies the
compatibility conditions from Axiom II of class formations (cf. II, (5.6)).
(5.1) Definition. If cp ∈ H 2 (GLP |Kp , L×

P ) ( P an arbitrarily chosen prime
over p) are the local components of c ∈ H 2 (GL|K , IL ), then we set
X
1
invL|K c = invLP |Kp cp ∈ [L:K] ZZ/ZZ.
p
Note that here almost all cp = 1, so that the sum contains only finitely many
non-zero summands. In particular, we obtain an invariant homomorphism
invL|K : H 2 (GL|K , IL ) −→ 1
[L:K] Z
Z/ZZ.
(5.2) Proposition. If N ⊇ L ⊇ K are normal extensions of the field K, then

invN |K c = invL|K c, c ∈ H 2 (GL|K , IL ) ⊆ H 2 (GN |K , IN ),
invN |L (resL c) = [L : K] · invN |K c, c ∈ H 2 (GN |K , IN ),
invN |K (corK c) = invN |L c, c ∈ H 2 (GN |L , IN ).
The last two formulas require only that N |K be normal.
Here we use the convention made after (3.5) to interpret the inflation map
H 2 (GL|K , IL ) −→ H 2 (GN |K , IN ) (N ⊇ L ⊇ K)
as an inclusion, so that H 2 (GL|K , IL ) ⊆ H 2 (GN |K , IN ).
Proof. The proposition follows from the behavior of the local invariants with
respect to the maps incl, res and cor. If c ∈ H 2 (GL|K , IL ), then by (3.3)
X X
invN |K c = invNP0 |Kp cp = invLP |Kp cp = invL|K c
p p
Here P0 is an arbitrary prime of N over p and P is the prime of L lying under

P0 . If c ∈ H 2 (GN |K , IN ), and P runs through the primes of L, then
14)
This invariant isomorphism is independent of the choice of P | p in the follow-
ing sense: If P0 | p is another prime of L, then the canonical Kp -isomorphism
LP → LP0 yields a canonical isomorphism between H 2 (GLP |Kp , L× P ) and
H 2 (GLP0 |Kp , L×
P0 ), which trivially preserves the invariant map.

P P
invN |L (resL c) = invNP0 |LP (resL c)P = invNP0 |LP (resLP cp )
P P
P PP
= [LP : Kp ] · invNP0 |Kp cp = [LP : Kp ] · invNP0 |Kp cp ,
P p P|p
where P0 represents any prime of N over P and p the prime of K lying

under P. If we note that the invariants invNP0 |Kp cp are independent of the
choice of the primes P0 of N lying above p (cf. the footnote 14) on p. 138), and
that by the fundamental equation of number theory (cf. §1, p. 115) we have
X
[LP : Kp ] = [L : K],
P| p
then (P0 a fixed prime of N over p):

XX
invN |L (resL c) = [LP : Kp ] · invNP0 |Kp cp
p P| p
X
= [L : K] · invNP0 |Kp cp
p
= [L : K] · invN |K c.
Finally, for c ∈ H 2 (GN |L , IN ) it follows from the formulas in (3.3) that
X
invN |K (corK c) = invNP0 |Kp (corK c)p
p
XX
= invNP0 |Kp (corKp cP )
p P| p
XX
= invNP0 |LP cP = invN |L (c).
p P|p
Since H 1 (GL|K , IL ) = 1, it follows that the extensions L|K satisfy with re-
spect to the idèle group IL and the idèle homomorphism invL|K the con-
ditions for a class formation, except for that the homomorphism invL|K :
1
H 2 (GL|K , IL ) → [L:K] ZZ/ZZ is not an isomorphism. To makes this an isomor-
phism, we have to pass from the idèle group IL to the idèle class group CL .
Before explaining this in detail, we consider abelian extensions and introduce
an invariant homomorphism invL|K , as well as the following symbol.
(5.3) Definition. Let L|K be an abelian extension. If a ∈ IK with local

components ap ∈ Kp× , then we set
Y
(a, L|K) = (ap , LP |Kp ) ∈ GL|K .
p
In this definition we have used the following facts: For each prime p, the symbol
(ap , LP |Kp ) defines an element of the local abelian Galois group GLP |Kp ,
which we always consider as a subgroup of GL|K . Hence
(ap , LP |Kp ) ∈ GLP |Kp ⊆ GL|K .
Since ap is a unit for almost all primes p and since LP |Kp is unramified for
Qp, we have (ap , LP |Kp ) = 1 for almost all p by II, (4.4). Thus the
almost all
product p (ap , LP |Kp ) ∈ GL|K is well defined, and it is also independent of
the order of the factors, since GL|K is abelian. The symbol ( , L|K) and the
invariant mapping invL|K are related as follows.
(5.4) Lemma15) . Let L|K be an abelian extension, a ∈ IK and (a) =

a · NL|K IL ∈ H 0 (GL|K , IL ). If χ ∈ χ(GL|K ) = H 1 (GL|K , Q/ZZ), then
1
χ(a, L|K) = invL|K ((a) ∪ δχ) ∈ [L:K] Z
Z/ZZ.
This is a consequence of the analogous formula which relates the local norm
residue symbol ( , LP |Kp ) with the local invariant map invLP |Kp (cf. II,
(5.11)). If we denote by χp the restriction of χ to GLP |Kp , and by (ap ) =
ap · NLP |Kp L×P , then
X X
χ(a, L|K) = χp (ap , LP |Kp ) = invLP |Kp ((ap ) ∪ δχp ).
p p
The remark after (3.2) shows that the classes

((ap ) ∪ δχp ) ∈ H 2 (GLP |Kp , L×
P)
are the local components of (a) ∪ δχ ∈ H 2 (GL|K , IL ); we only need to note

that ap · δχp (σ, τ ) (resp. a · δχ(σ, τ )) is a 2-cocycle of the class ((ap ) ∪ δχp )
(resp. ((a)∪δχ)) (cf. II, (5.11)). Thus χ(a, L|K) = invL|K ((a)∪δχ) as claimed.
When changing from the idèle invariants to the idèle class invariants, the fol-
lowing theorem is of central importance. From the exact cohomology sequence
associated with the exact sequence
1 −→ L× −→ IL −→ CL −→ 1
we see, using H 1 (GL|K , CL ) = 1, that the induced homomorphism
H 2 (GL|K , L× ) −→ H 2 (GL|K , IL )
is injective.
We use this injection to think of H 2 (GL|K , L× ) as a subgroup of H 2 (GL|K , IL ),
i.e., we view the elements of H 2 (GL|K , L× ) as the idèle cohomology classes
that are represented by cocycles with values in the principal idèle group L× .
(5.5) Theorem. If c ∈ H 2 (GL|K , L× ), then invL|K c = 0.

15)
Cf. also II, (1.10).
We will see that apart from purely technical considerations the proof of this
theorem is based on the following two facts: the explicit description of the
local norm residue symbol and the product formula for algebraic numbers.
Proof. We start with the simple observation that it suffices to consider the
case when K = Q and L is a cyclic cyclotomic extension of Q. In fact, if
c ∈ H 2 (GL|K , L× ), and N is a normal extension of Q containing L, then
c ∈ H 2 (GL|K , L× ) ⊆ H 2 (GN |K , N × ) ⊆ H 2 (GN |K , IN ),
corQ c ∈ H 2 (GN |Q , N × ), and invL|K c = invN |K c = invN |Q (corQ c) by (5.2).
Hence to show invL|K c = 0 it suffices to consider the case K = Q. Since by
(3.6) there exists a cyclic cyclotomic extension L0 |Q with c ∈ H 2 (GL0 |Q , L×
0 ),
we can even assume that L|Q itself is a cyclic cyclotomic extension.
Let χ be a generator of the cyclic character group χ(GL|Q ) = H 1 (GL|Q , Q/ZZ).
Then δχ is a generator of H 2 (GL|Q , ZZ), and Tate’s Theorem I, (7.3) implies
δχ ∪ : H 0 (GL|Q , L× ) −→ H 2 (GL|Q , L× ) 16)
is bijective. Thus each element c ∈ H 2 (GL|Q , L× ) has the form c = (a) ∪ δχ

with (a) = a · NL|Q L× ∈ H 0 (GL|Q , L× ), a ∈ Q× . From (5.4) we obtain
invL|Q c = invL|Q ((a) ∪ δχ) = χ(a, L|Q),
Q
and we need to show that (a, L|Q) = p (a, LP |Qp ) = 1. Now L is a cyclo-
tomic extension, i.e., L ⊆ Q(ζ) for some root of unity ζ. The automorphism
(a, L|Q) is precisely the restriction of (a, Q(ζ)|Q) to L; this follows easily
from the behavior of the local norm residue symbol (a, Qp (ζ)|Qp ) when pass-
ing to the extension LP |Qp (cf. II, (5.10a)). It therefore suffices to show that
(a, Q(ζ)|Q) = 1 for a ∈ Q× . Now Q(ζ) is generated by roots of unity of prime
power order and it suffices to show the vanishing of (a, Q(ζ)|Q) for these gen-
erators, hence we may assume that ζ is a primitive ln -th root of unity (l a
prime number). With this reduction we come to the actual core of the proof.
Let ζ be a primitive ln -th root of unity; if l = 2, we assume n ≥ 2. If p
ranges over the prime numbers and the infinite prime over p = p∞ of Q, then
the Qp (ζ)|Qp are the local extensions associated with Q(ζ)|Q. The extension
Qp (ζ)|Qp is unramified for p 6= l and totally ramified for p = l (cf. [21], §5,
3.); if p = p∞ , then Qp (ζ)|Qp means the extension C|IR. We have to show:
Y
For each a ∈ Q× , (a, Q(ζ)|Q) = (a, Qp (ζ)|Qp ) = 1.
p
Here it obviously suffices to assume that a is integral. We consider the effect

of the local norm residue symbol (a, Qp (ζ)|Qp ) on the ln -th roots of unity ζ.
1. For p 6= l, p 6= p∞ , we have by II, (4.8)
(a, Qp (ζ)|Qp )ζ = ϕvp (a) ζ,
16)
This can also be shown in an elementary way.
where vp is the valuation on Qp and ϕ is the Frobenius automorphism on

Qp (ζ)|Qp . Since the residue field of Qp has p elements, clearly ϕζ = ζ p , thus
vp (a)
(a, Qp (ζ)|Qp )ζ = ζ p .
2. For p = l, we obtain from II, (7.16), writing a = u · pm = u · pvp (a) , u a unit:

(a, Qp (ζ)|Qp )ζ = ζ r ,
where r is a natural number which is determined mod pn by the congruence
r ≡ u−1 ≡ a−1 · pvp (a) mod pn .
3. For p = p∞ the automorphism (a, C|IR) is either the identity or complex
conjugation, depending on whether a > 0 or a < 0 (cf. II, §5, p. 95). Thus
(a, Qp (ζ)|Qp )ζ = ζ sgn a .
Combining all of the above, we obtain
Y Q vp (a)
·r
(a, Q(ζ)|Q)ζ = (a, Qp (ζ)|Qp )ζ = ζ sgn a· p6=l p .
p
But by the product formula we have for the power appearing on the right side
Y Y 1
sgn a · pvp (a) · r ≡ sgn a · pvp (a) lv` (a) a−1 = Q = 1 mod ln ,
p6=l p6=l p |a|p
therefore (a, Q(ζ)|Q)ζ = ζ, i.e., we have in fact (a, Q(ζ)|Q) = 1.
Theorem (5.5) shows that the group H 2 (GL|K , L× ) lies in the kernel of the
1
homomorphism invL|K : H 2 (GL|K , IL ) → [L:K] ZZ/ZZ. We have to ask further
whether or not it is precisely the kernel, and in addition whether or not invL|K
is a surjective homomorphism. For the cyclic case we have:
(5.6) Proposition. If L|K is a cyclic extension, then the sequence

invL|K
1 −→ H 2 (GL|K , L× ) −→ H 2 (GL|K , IL ) −−−−→ 1
[L:K] Z
Z/ZZ −→ 0
is exact.
Proof. a) To show invL|K is surjective, we assume first [L : K] is a prime

1 1
power pr . Because [L:K] +ZZ generates [L:K] ZZ/ZZ, it suffices to find an element
2 1
c ∈ H (GL|K , IL ) with invL|K c = [L:K] + ZZ. We use the decomposition
H 2 (GL|K , IL ) ∼
M
= H 2 (GLP |Kp , L× P)
p
and determine c by its local components cp ∈ H 2 (GLP |Kp , L× P ). Since L|K is

cyclic of prime power degree, it follows from (4.3) that K contains a prime p0
which is inert in L. Since p0 is inert, we have [LP0 : Kp0 ] = [L : K], (P0 |p0 ),
and local class field theory yields an element cp0 ∈ H 2 (GLP0 |Kp0 , L×P0 ) with
1 1
invLP0 |Kp0 cp0 = [LP0 :Kp0 ] + ZZ = [L:K] + ZZ. Now if c is the element in
2
H (GL|K , IL ) that is determined by the local components
. . . , 1, 1, 1, cp0 , 1, 1, 1, . . . ,
then X
1
invL|K c = invLP |Kp cp = invLP0 |Kp0 cp0 = [L:K] + ZZ.
p
That invL|K is also surjective in the general case [L : K] = n = pr11 · · · prss

follows easily from this. For every i = 1, . . . , s there obviously exists a cyclic
intermediate field Li of degree [Li : K] = pri i . Consider the decomposition
1 n1 ns
= r1 + · · · + rs
n p1 ps
into partial fraction. By the previous case there is a ci ∈ H 2 (GLi |K , ILi ) with
ni
invLi |K ci = invL|K ci = ri + ZZ.
pi
Thus if we set
c = c1 · · · cs ∈ H 2 (GL|K , IL ),
then s s
X X ni 1
invL|K c = invL|K ci = Z = + ZZ,
ri + Z
i=1
p
i=1 i
n
which shows that invL|K is surjective for any cyclic extension.
b) By (5.5) H 2 (GL|K , L× ) is contained in the kernel of invL|K . To show that
the group H 2 (GL|K , L× ) in fact equals the kernel of invL|K we use a sim-
ple argument involving the orders of these groups. Since the map invL|K is
surjective, we only need to show that the order of the factor group
H 2 (GL|K , IL )/H 2 (GL|K , L× )
1
is at most the order of [L:K] Z
Z/ZZ, i.e., the degree of [L : K]. Using the sequence
1 −→ L× −→ IL −→ CL −→ 1
we obtain, using that H 1 (GL|K , CL ) = 1, the exact cohomology sequence
1 −→ H 2 (GL|K , L× ) −→ H 2 (GL|K , IL ) −→ H 2 (GL|K , CL ).
Therefore the order of H 2 (GL|K , IL )/H 2 (GL|K , L× ) divides the order of
H 2 (GL|K , CL ). By (4.9) H 2 (GL|K , CL ) divides [L : K], and we are done.
For the following it would be very convenient if we could show that invL|K is
a surjective homomorphism in general. Unfortunately this is false. In order for
1
every element of [L:K] ZZ/ZZ to be in the image of the invariant map, it is neces-
sary to enlarge the field L by forming the compositum with a cyclic extension.
For technical reasons it is best to let L range over all normal extensions of K
and to consider the union
[
H 2 (GΩ|K , IΩ ) = H 2 (GL|K , IL )
L
(cf. remark after (3.5)). If N ⊇ L ⊇ K are two normal extensions of K, then

H 2 (GL|K , IL ) ⊆ H 2 (GN |K , IN ),
and since by (5.2) the invariant map can be extended from H 2 (GL|K , IL ) to
H 2 (GN |K , IN ), we obtain a homomorphism
invK : H 2 (GΩ|K , IΩ ) −→ Q/ZZ ,
whose restriction to H 2 (GL|K , IL ) ⊆ H 2 (GΩ|K , IΩ ) coincides with the initial
invariant map invL|K . If we take into account that for each positive integer
m there is a cyclic extension L|K with m | [L : K] (for example, by (3.7)), we
1
see that Q/ZZ is already covered by the groups [L:K] ZZ/ZZ coming from cyclic
extension L|K. Now the map invL|K is surjective in the cyclic case, thus we
obtain for the invariant map invK defined above the following
(5.7) Theorem. The homomorphism

invK : H 2 (GΩ|K , IΩ ) −→ Q/ZZ
is surjective.
The results obtained so far can be summarized to give a theorem about the
Brauer group of an algebraic number field K and its completion Kp . Recall
that we defined the Brauer group Br (K) of a field K in II, §2 as the union
(more precisely the direct limit)
[
Br (K) = H 2 (GL|K , L× ),
L
where L runs through all finite Galois extensions of K. If K is an algebraic
number field, then we choose over each prime p of K a fixed valuation of the
algebraic closure Ω; each such valuation in turn determines in every finite
extension L|K a prime P above p. Then we have
[
Br (Kp ) = H 2 (GLP |Kp , L×
P) .
L
From the homomorphisms
invL|K
H 2 (GL|K , L× ) −→ H 2 (GL|K , IL ) ∼
M
= H 2 (GLP |Kp , L×
P ) −−−−→
1
[L:K] Z
Z/ZZ
p
we obtain by passing to the direct limit (i.e., in this case, the union) the
canonical homomorphism
inv
Br (K) −→ H 2 (GΩ|K , IΩ ) ∼
M
= Br (Kp ) −−−K → Q/ZZ ,
p
where invK is the sum of the local invariant maps invKp : Br (Kp ) −→ Q/ZZ
(cf. II, §1, p. 68 and II, (5.4)).
We now have Hasse’s Main Theorem on the Theory of Algebras:
(5.8) Theorem. For every finite algebraic number field K one has the canon-
ical exact sequence
inv
M
1 −→ Br (K) −→ Br (Kp ) −−−K → Q/ZZ −→ 0 .
p
Proof. The groups Br (K),

L ∼ 2
p Br (Kp ) (= H (GΩ|K , IΩ )), and Q/Z Z are,
by (3.6) and the remark after (5.7), the union of the groups H 2 (GL|K , L× ),
L 2 × ∼ 2 1
p H (GLP |Kp , LP ) (= H (GL|K , IL ), and [L:K] Z
Z/ZZ, respectively, provided
L|K runs through all the cyclic extensions. But for such a cyclic extension
L|K we have by (5.6) the exact sequence
invL|K
1 −→ H 2 (GL|K , L× ) −→ H 2 (GL|K , IL ) −−−−→ 1
[L:K] Z
Z/ZZ −→ 0,
from which the theorem follows immediately.
§ 6. The Reciprocity Law
Having studied the idèle invariants in the previous section, we now want to
construct invariants for the elements of the groups H 2 (GL|K , CL ). We start
with the following observations:
If L|K is a normal extension, then we obtain from the exact sequence
1 −→ L× −→ IL −→ CL −→ 1,
using H 1 (GL|K , CL ) = 1 = H 3 (GL|K , IL ) = 1, the exact cohomology sequence
j
1 −→ H 2 (GL|K , L× ) −→ H 2 (GL|K , IL ) −→ H 2 (GL|K , CL )
δ
−→ H 3 (GL|K , L× ) −→ 1.
If c ∈ H 2 (GL|K , CL ) and c ∈ H 2 (GL|K , IL ) is such that c = jc, then we set
1
invL|K c = invL|K c ∈ [L:K] Z
Z/ZZ.
This definition is independent of the choice of the preimage c ∈ H 2 (GL|K , IL ),
because two such preimages differ only by an element in H 2 (GL|K , L× ),
which by (5.5) has invariant 0. Of course, this only works if the element
c ∈ H 2 (GL|K , CL ) lies in the image of the homomorphism j. But the map j
is in general not surjective; in fact, j being surjective would be equivalent to
the group H 3 (GL|K , L× ) being trivial, however, this group is not equal to 1
in general (cf. [2], Ch. 7, Th. 12). Nevertheless, we can show that at least in
the cyclic case the map j is surjective.
§ 6. The Reciprocity Law 147
(6.1) Proposition. If L|K is a cyclic extension, then the homomorphism

j
H 2 (GL|K , IL ) −→ H 2 (GL|K , CL )
is surjective.
Proof. In L|K is cyclic, then H 3 (GL|K , L× ) ∼

= H 1 (GL|K , L× ) = 1 (cf. II,
(2.2)).
In order to define an invariant map for arbitrary normal extensions L|K, we

proceed in a similar way as for the groups H 2 (GL|K , IL ) with respect to the
invariant map for idèles at the end of the previous section.
First we note that the homomorphism
j
H 2 (GL|K , IL ) −→ H 2 (GL|K , CL )
commutes with the maps inf and res; i.e., if N ⊇ L ⊇ K are two normal
extensions of K, then we have
j ◦ inf N = inf N ◦ j and j ◦ resL = resL ◦ j ,
where in the last formula we only need to assume that N |K is normal. For
simplicity we set
(6.2) Definition. H q (L|K) = H q (GL|K , CL ).
Because H 1 (L|K) = 1 (cf. (4.7)), the extensions L|K form a field formation
in the sense of II, §1 with respect to the idèle group CL as the module. To
simplify things we will, as for the idèle cohomology groups in §5 and generally
for every field formation, interpret the injective inflation maps
inf
H 2 (L|K) −−→ H 2 (N |K) (N ⊇ L ⊇ K)
as inclusions. More precisely, this means that we form the direct limit
H 2 (Ω|K) = lim
−→
H 2 (L|K) 17)
L
where L ranges over all finite normal extensions of K. We view the groups
H 2 (L|K) as being embedded in H 2 (Ω|K) via the inflation maps. Thinking of
the H 2 (L|K) as subgroups of H 2 (Ω|K), we have
[
H 2 (Ω|K) = H 2 (L|K).
L
Hence if N ⊇ L ⊇ K are two normal extensions, then we have inclusions
H 2 (L|K) ⊆ H 2 (N |K) ⊆ H 2 (Ω|K),
17)
Here Ω denotes again the field of all algebraic numbers, and the same convention
as in footnote 9) for the group H 2 (GΩ|K , IΩ ) applies to the group H 2 (Ω|K).
where, according to this interpretation, an element c ∈ H 2 (N |K) comes from

an element of H 2 (L|K) if and only if it is the inflation of an element of
H 2 (L|K).
Now the crucial input is the following theorem which plays a role similar to
one of Theorem (II, 5.2) in the local theory.
(6.3) Theorem. If L|K is a normal extension and L0 |K a cyclic extension

of equal degree [L0 : K] = [L : K], then
H 2 (L0 |K) = H 2 (L|K) ⊆ H 2 (Ω|K).
Since for every positive integer m there is a cyclic extension L|K of degree m
(see (3.7), for instance), this theorem has the following
[
(6.4) Corollary. H 2 (Ω|K) = H 2 (L|K).
L|K cyclic
Proof of (6.3). We first show that H 2 (L0 |K) ⊆ H 2 (L|K). If N = L·L0 is the
compositum of L and L0 , then a simple group theoretic argument shows that
if the extension L0 |K is cyclic, then the extension N |L is also cyclic. Now let
c ∈ H 2 (L0 |K) ⊆ H 2 (N |K). Because of the exact sequence
res
1 −→ H 2 (L|K) −→ H 2 (N |K) −−−→
L
H 2 (N |L),
an element c ∈ H 2 (N |K) is an element of H 2 (L|K) if and only if resL c = 1.
To show this, we use the idèle invariants. By (6.1) the homomorphism
j
H 2 (GL0 |K , IL0 ) −→ H 2 (L0 |K)
is surjective, so that c = jc, c ∈ H 2 (GL0 |K , IL0 ) ⊆ H 2 (GN |K , IN ). From
the remarks made above, we know that the map j commutes with inflation
(interpreted here as inclusion) and with restriction, hence we have the formulas
resL c = resL (jc) = jresL c.
Thus resL c = 1 if and only if resL c lies in the kernel of j, and therefore
in H 2 (GN |L , N × ). Since N |L is cyclic, this holds by (5.6) if and only if
invN |L (resL c) = 0, and this last statement now follows from
invN |L (resL c) = [L : K] · invN |K c = [L0 : K] · invL0 |K c = 0.
Therefore H 2 (L0 |K) ⊆ H 2 (L|K).
To show that the above inequality is in fact an equality we consider orders.
Because H 1 (L0 |K) = 1 and H 3 (GL0 |K , L0× ) ∼
= H 1 (GL0 |K , L0× ) = 1, we obtain
the exact cohomology sequence
1 −→ H 2 (GL0 |K , L0× ) −→ H 2 (GL0 |K , IL0 ) −→ H 2 (L0 |K) −→ 1,
where |H 2 (L0 |K)| = [L0 : K] = [L : K] by (5.6). On the other hand, |H 2 (L|K)|

divides the degree [L : K] by (4.9), hence H 2 (L0 |K) = H 2 (L|K).
Let N ⊇ L ⊇ K be two normal extensions. Because the map

j
H 2 (GL|K , IL ) −→ H 2 (L|K)
is compatible with inflation, it can be extended to a canonical homomorphism
j
H 2 (GN |K , IN ) −→ H 2 (N |K).
Thus we obtain a homomorphism
j
H 2 (GΩ|K , IΩ ) −→ H 2 (Ω|K)
whose restriction to the groups H 2 (GL|K , IL ) are the initial homomorphisms
H 2 (GL|K , IL ) → H 2 (L|K). If these are not surjective, then we still have the
(6.5) Theorem. The homomorphism

j
H 2 (GΩ|K , IΩ ) −→ H 2 (Ω|K)
is surjective.
Proof. If c ∈ H 2 (Ω|K), then it follows from (6.4) that there is a cyclic

extension L|K such that c ∈ H 2 (L|K). Since for a cyclic extension the map
j
H 2 (GL|K , IL ) −→ H 2 (L|K)
is surjective by (6.1), c = jc for some c ∈ H 2 (GL|K , IL ) ⊆ H 2 (GΩ|K , IΩ ).
Given this theorem, it is easy to obtain class invariants for the elements of
H 2 (Ω|K) = L H 2 (L|K) from the invariant map of the idèle cohomology
S
classes. From the homomorphism
invK : H 2 (GΩ|K , IΩ ) −→ Q/ZZ,
which is surjective by (5.7), we in fact come to the following
(6.6) Definition. If c ∈ H 2 (Ω|K) and c = jc, c ∈ H 2 (GΩ|K , IΩ ), then define

invK c = invK c ∈ Q/ZZ.
Of course we have to convince ourselves that this definition is independent of

the choice of the preimage c ∈ H 2 (GΩ|K , IΩ ) of c. But if c0 is another element
in H 2 (GΩ|K , IΩ ) with c = jc0 , then c, c0 ∈ H 2 (GL|K , IL ) ⊆ H 2 (GΩ|K , IΩ )
for a sufficiently large normal extension L|K, where we may assume that this
extension is so large that c ∈ H 2 (L|K). Because c = jc = jc0 , c and c0 differ
only by an element in the kernel of the mapping j : H 2 (GL|K , IL )−→H 2 (L|K),
and thus by an element of H 2 (GL|K , L× ), which has invariant 0 by (5.5).
We remark that in the above definition Theorem (5.5) plays an essential role,
in fact one may consider this result an important step in the direction of the
reciprocity law. From (6.6) we obtain a homomorphism
invK : H 2 (Ω|K) −→ Q/ZZ.
The restriction of invK to the group H 2 (L|K) coming from a finite normal
extension L|K yields a homomorphism
invL|K : H 2 (L|K) −→ 1
[L:K] Z
Z/ZZ,
because the orders of the elements in H 2 (L|K) divide the degree [L : K] (cf.
1
I, (3.16)), and consequently are mapped to the only subgroup [L:K] ZZ/ZZ of
Q/ZZ of order [L : K].
1
We briefly recall the construction of the map invL|K : H 2 (L|K) → [L:K] Z
Z/ZZ:
If c ∈ H 2 (L|K), then we obtain the invariant invL|K c by choosing a cyclic

extension L0 |K of equal degree [L0 : K] = [L : K], so that by (6.3) H 2 (L0 |K) =
H 2 (L|K) ((6.3)); in particular, c ∈ H 2 (L0 |K). In this cyclic case we have by
(6.1) an idèle cohomology class c ∈ H 2 (GL0 |K , IL0 ) with c = jc, and obtain so
X 1
invL|K c = invL0 |K c = invL0 |K c = invL0P |Kp cp ∈ ZZ/ZZ.
p
[L : K]
The detour using cyclic extensions, which we have described by introducing

the groups H 2 (GΩ|K , IΩ ) and H 2 (Ω|K) (and interpreting inflation as inclu-
sion) is necessary, because in general the map
j
H 2 (GL|K , IL ) −→ H 2 (L|K)
is not surjective. However, for the elements in the image of j we immediately
obtain from Definition (6.6) the simple
(6.7) Proposition. If c = jc, c ∈ H 2 (L|K), c ∈ H 2 (GL|K , IL ), then

invL|K c = invL|K c.
(6.8) Theorem. The invariant maps

invK : H 2 (Ω|K) −→ Q/ZZ
and
invL|K : H 2 (L|K) −→ 1
[L:K] Z
Z/ZZ
are isomorphisms.
Proof. It suffices to verify that invL|K is bijective. Let L0 |K be a cyclic

extension of degree [L0 : K] = [L : K], so that H 2 (L0 |K) = H 2 (L|K). If α ∈
1
Z/ZZ, then by (5.6) there is a c ∈ H 2 (GL0 |K , IL0 ) with invL0 |K c = α. Set
[L:K] Z
c = jc ∈ H 2 (L0 |K) = H 2 (L|K). Then invL|K c = invL0 |K c = invL0 |K c = α,

i.e., invL|K is surjective.
That invL|K is bijective follows now easily from the fact that the order of
H 2 (L|K) is a divisor of the degree [L : K] (cf. (4.9)), and therefore a divisor
1
of the order of [L:K] ZZ/ZZ.
We now come to the main theorem of class field theory. Let K0 be a fixed
algebraic number field, Ω the field of all algebraic numbers,
S and G = GΩ|K0
the Galois group of Ω|K0 . We form the union CΩ = K CK , where K runs
through all finite extensions of K0 18) . Then CΩ is canonically a G-module: If
c ∈ CΩ , say c ∈ CL for an appropriate finite normal extension L|K0 , we set
σc = σ L c ∈ CL ⊆ CΩ (σ ∈ G).
The pair (G, CΩ ) is obviously a formation in the sense of II, §1, and the
fundamental result of all our constructions is the following
(6.9) Theorem. The formation (G, CΩ ) is a class formation with respect to

the invariant map introduced in (6.6).
For the proof we have to verify the axioms in II, §1, (1.3).
Axiom I: H 1 (L|K) = 1 for every normal extension L|K of each finite extension
field of K0 (cf. (4.7)).
Axiom II: For every normal extension L|K of each finite extension field of K0
we have by (6.8) the isomorphism
invL|K : H 2 (L|K) −→ 1
[L:K] Z
Z/ZZ.
a) If N ⊇ L ⊇ K are two normal extensions and c ∈ H 2 (L|K), then c ∈

H 2 (N |K), and
invN |K c = invL|K c,
since invN |K , and invL|K are defined as the restrictions of invK to H 2 (N |K)
and H 2 (L|K) ⊆ H 2 (N |K) respectively (cf. p. 150).
b) Let N ⊇ L ⊇ K be two extensions fields of K with N |K normal. If
c ∈ H 2 (N |K), then resL c ∈ H 2 (N |L). For the proof of the formula
invN |L (resL c) = [L : K] · invN |K c
we use the analogous formula for the idèle invariants (cf. (5.2)). By (6.5)
there is a c ∈ H 2 (GΩ|K , IΩ ) with jc = c, where we can assume that there
is a normal extension M |K containing N , M ⊇ N ⊇ L ⊇ K, such that
18)
More precisely, we should write CΩ = lim CK . Nevertheless, we think of CK as
−→
embedded in this direct limit and interpret CΩ as the union of the CK .
c ∈ H 2 (GM |K , IM ). From the formulas in (5.2), and using the convention

that the inflation maps are to be interpreted as inclusions, we have by (6.7)
invN |L (resL c) = invM |L (resL jc) = invM |L (jresL c) = invM |L (resL c)
= [L : K] · invM |K c = [L : K] · invM |K jc = [L : K] · invN |K c.
Because of this theorem we can now apply the entire abstract theory of class
formations to the case of algebraic number fields. If we again denote by
uL|K ∈ H 2 (L|K)
the fundamental class of the normal extension L|K, which is uniquely de-
1
termined by the formula invL|K uL|K = [L:K] + ZZ, then we have the general
(6.10) Theorem. The homomorphism cup product with the fundamental

class
uL|K ∪ : H q (GL|K , ZZ) −→ H q+2 (L|K)
is bijective.
From this, together with II, (1.8) we immediately obtain the
(6.11) Corollary. H 3 (L|K) = 1 and H 4 (L|K) ∼

= χ(GL|K ).
For the case q = −2 Theorem (6.10) yields Artin’s Reciprocity Law:
(6.12) Theorem. The map cup product with the fundamental class
uL|K ∪
Gab ∼ −2 (GL|K , ZZ) −−−−→ H 0 (L|K) = CK /NL|K CL
L|K = H
yields a canonical isomorphism, i.e., the reciprocity map between the

abelianization Gab
L|K of the Galois group GL|K and the norm residue group
CK /NL|K CL of the idèle group CK
θL|K : Gab
L|K −→ CK /NL|K CL .
The inverse of the reciprocity map θL|K is induced from the homomorphism
( , L|K) : CK → Gab L|K with kernel NL|K CL , the norm residue symbol.
(6.13) Theorem. The following sequence is exact

( ,L|K)
1 −→ NL|K CL −→ CK −−−−−−→ Gab
L|K −→ 1.
Because the invariant map is compatible with inflation (inclusion) and restric-
tion, we see that the norm residue symbol behaves with respect to varying
field extensions as follows (cf. II, (1.11)):
Let N ⊇ L ⊇ K be two extensions with N |K is normal. Then the following

diagrams are commutative:
a)
º¹¸·
CK
( ,N |K)
Gab
N |K
id π
( ,L|K)
CK Gab
L|K
Hence (a, L|K) = π(a, N |K) ∈ Gab L|K , for a ∈ CK , if L|K is also normal (in
addition to N |K). Here π is the canonical projection of Gab ab
N |K onto GL|K .
b)
¾½¼»
CK
( ,N |K)
Gab
N |K
incl Ver
( ,N |L)
CL Gab
N |L
Hence (a, N |L) = Ver(a, N |K) ∈ Gab N |L for a ∈ CK . Recall that the Ver-
lagerung (transfer) Ver is induced by restriction
res
∼ −2 (GN |K , ZZ) −
Gab
N |K = H −→ H −2 (GN |L , ZZ) ∼
= Gab
N |L .
c)
CÂÁÀ¿ L
( ,N |L)
Gab
N |L
NL|K κ
( ,N |K)
CK Gab
N |K
Hence (NL|K a, N |K) = κ(a, N |L) ∈ Gab

N |K for a ∈ CL , where κ is the canoni-
cal homomorphism from Gab N |L into Gab
N |K .
d)
ÃÆÅÄ
CK
( ,N |K)
Gab
N |K
σ σ∗
( ,σN |σK)
CσK Gab
σN |σK
Hence (σa, σN |σK) = σ(a, N |K)σ −1 for a ∈ CK , where for σ ∈ G the maps
σ σ∗ −1
CK → CσK and Gab ab
N |K → GσN |σK are induced by a 7→ σa and τ 7→ στ σ .
In case N |K is abelian, the homomorphisms between the Galois groups in

these diagrams are, with the exception of the Verlagerung, the obvious ones.
If N |K is abelian, i.e., if the Galois groups coincide with their abelianiza-

tions, then we have in a) the quotient map, in c) the inclusion, and in d) the
isomorphism induced by conjugation.
A subgroup I of the idèle class group CK of a number field K is called a norm

group if there is a normal extension L|K with I = NL|K CL . By II, (1.14) we
have
(6.14) Theorem. The map

L 7−→ IL = NL|K CL ⊆ CK
yields an inclusion reversing isomorphism between the abelian extensions L|K
and the lattice of norm groups IL of CK . Therefore
IL1 ·L2 = IL1 ∩ IL2 and IL1 ∩L2 = IL1 · IL2 .
Every subgroup which contains a norm group is again a norm group. If L and
I correspond to each other, then L is called the class field associated with I.
This theorem says that the structure of the abelian extensions of K can al-
ready be read off of the idèle class group of the base field K. Of course we have
to ask if the norm groups can be characterized only by intrinsic information
given by the group CK and independent of the field extensions, similarly to
the norm groups in local class field theory which are determined as the closed
subgroups of finite index in the multiplicative group of the base field. Such a
characterization is presented in the next section.
The reciprocity map θL|K from (6.12) and its inverse, the norm residue symbol
( , L|K), are defined canonically; however, their explicit description is still
too complicated and abstract. We are therefore interested in finding ways to
explicitly compute the norm residue symbol.
This is accomplished by the following beautiful theorem which is essentially
due to H. Hasse. It connects in a simple way global and local class field
theory.
(6.15) Theorem. Let L|K be an abelian extension and a ∈ CK , a = a · K × ,

a ∈ IK . Then we have
Y
(a, L|K) = (ap , LP |Kp ) ∈ GL|K .
p
Note here that (ap , LP |Kp ) ∈ GLP |Kp ⊆ GL|K , and that the components ap
of the idèle a representing a are units for almost all p; in particular, since the
extensions LP |Kp are almost all unramified, the local norm residue symbols
are almost all equal to 1.
Since ( , L|K) is the norm residue symbol of a class formation, we can use
Lemma II, (1.10) to prove the theorem: If we denote by (a) = a · NL|K CL ∈
H 0 (L|K), then for every character χ ∈ χ(GL|K ) = H 1 (GL|K , Q/ZZ) we have
χ(a, L|K) = invL|K ((a) ∪ δχ).
On
Q the other hand, we have alreadyQ introduced in (5.3) the notation (a, L|K) =
(a
p p , LP |Kp ) for the product p (ap , LP |Kp ), and have shown in (5.4) that
χ(a, L|K) = invL|K ((a) ∪ δχ),
where (a) = a · NL|K IL ∈ H 0 (GL|K , IL ). The homomorphism
j
H q (GL|K , IL ) −→ H q (GL|K , CL )
maps (a) ∈ H 0 (GL|K , IL ) to (a) ∈ H 0 (GL|K , CL ), and therefore maps (a) ∪
δχ ∈ H 2 (GL|K , IL ) to (a) ∪ δχ ∈ H 2 (GL|K , CL ) = H 2 (L|K), hence
j((a) ∪ δχ) = (a) ∪ δχ .
With (6.7) we obtain
χ(a, L|K) = invL|K ((a) ∪ δχ) = invL|K ((a) ∪ δχ) = χ(a, L|K),
and since this holds for all characters χ ∈ χ(GL|K ), it follows that
Y
(a, L|K) = (a, L|K) = (ap , LP |Kp ).
p
Hence the global norm residue symbol ( , L|K) is determined by the local
norm residue symbols ( , LP |Kp ). We will use this observation later when
we analyze the reciprocity law for the prime ideal decomposition in abelian
extensions.
We close this section with a remark about the universal norm residue
symbol ( , K) (cf. II, §1, p. 76). For every abelian extension we have the
homomorphism
( ,L|K)
CK −−−−−−→ GL|K .
The projective limit

Gab
K =←
lim GL|K
−
of the Galois groups GL|K of all (finite) abelian extensions L of K is the Galois
group of the maximal abelian extension field AK of K. For every a ∈ CK we
obtain the element
(a, K) = lim(a, L|K) ∈ Gab K
←−
as the compatible system formed by the elements (a, L|K) ∈ Gab L|K . This yields
a homomorphism
( ,K)
CK −−−−→ Gab K
whose kernel is the intersection of all norm groups (cf. II, (1.15))

\
DK = NL|K CL ,
L
and whose image is a dense subgroup of GabK.

Finally, we remark without giving details, that the product formula (6.15)
yields an analogous product formula for the universal norm residue symbol
Y
(a, K) = (ap , Kp ),
p
where ( , Kp ) denotes the universal norm residue symbol of local class field
theory, which can be embedded into the group GabK (cf. [2], Ch. 7, Cor. 2).
§ 7. The Existence Theorem
By Theorem (6.14) the abelian extensions of a number field K correspond

bijectively to the norm groups of CK . In this section we will characterize these
norm groups, similarly to those in local class field theory, as the subgroups of
finite index in CK that are closed with respect to a canonical topology.
The idèle
Q group IK ofQ an algebraic number field K is the union of the groups
S
IK = p∈S Kp× × p6∈S Up , where S ranges over all finite sets of primes
of K. The factors Kp× and Up are equipped with their respective valuation
topologies. These induce the Tychonoff topology on the direct product
Y Y
S
IK = Kp× × Up ,
p∈S p6∈S
0
S
so that IK becomes a topological group19). If S ⊆ S 0 , then IK S
⊆ IKS
, and the
S0 S
Tychonoff topology on IK induces the Tychonoff S topology on IK . Thus we
S
have a canonical topology on the idèle group IK = S IK , the so-called idèle
topology. If we want to define the idèle topology directly, we only need to
specify a fundamental system of neighborhoods of the identity of IK , and it is
completely obvious that such a fundamental system of neighborhoods is given
by the subsets Y Y
Wp × Up ⊆ IK ,
p∈S p6∈S
where the Wp ⊆ Kp× range over a fundamental system of neighborhoods of

the identity of Kp× , and S over all finite sets of primes of K.
In the following we always consider IK equipped with this canonical topology.

Roughly speaking, two idèles are close to each other when they are close
19)
For the theory of topological groups we refer to [9], Ch. III.
componentwise for many primes. The idèle topology is Hausdorff, since the
valuation topologies on Kp× , and therefore the Tychonoff topology on IK
S
are
Hausdorff.
(7.1) Proposition. The idèle group IK is locally compact.
×
Proof. If Wp is a compact neighborhood of the identity of Q Kp for the finite
primes p, say Wp = Up , then the direct Tychonoff product p Wp is a compact
neighborhood of the identity of IK . This shows that IK is locally compact.
(7.2) Proposition. K × is a discrete and therefore closed subgroup of IK .
Proof. It obviously suffices to show that the identity 1 ∈ IK has an open

neighborhood which aside from 1 contains no other principal idèle. Such a
neighborhood is given by the set
U = {a ∈ IK |ap − 1|p < 1 for p ∈ S, |ap |p = 1 for p 6∈ S},
where S denotes a finite set of primes containing all the infinite primes. If
there were a principal idèle x ∈ U, x 6= 1, then we would have
Y Y Y Y Y
|x − 1|p = |x − 1|p · |x − 1|p < |x − 1|p ≤ max{|x|p , 1} = 1,
p p∈S p6∈S p6∈S p6∈S
which is a contradiction to the product formula.
Since K × is closed in IK , the quotient CK = IK /K × is also a Hausdorff topo-

logical group; of course, since IK is locally compact, so is CK . The canonical
homomorphism IK → CK is continuous and takes open sets to open sets.
For each prime p, we consider the homomorphism

np : Kp× −→ IK ,
that maps x ∈ Kp× to the idèle np (x) ∈ IK which at the prime p has the
component x ∈ Kp× , and at all other primes has component 1. If we further
map x ∈ Kp× to the idèle class np (x) = np (x) · K × ∈ CK represented by np (x),
then we obtain a homomorphism
np : Kp× −→ CK ,
for which we have the following result:
(7.3) Proposition. The homomorphism

np : Kp× −→ CK
is a topological embedding of Kp× into CK .
In fact, from np (x) = 1, i.e., from np (x) ∈ K × , it follows immediately that

x = 1, so that we are dealing with an injection which is trivially topological.
We associate with each idèle a ∈ IK its absolute value

Y
|a| = |ap |p ∈ IR×
+,
p
and obtain a (clearly continuous) homomorphism of IK to the group IR× + of

positive real numbers. Of course this homomorphism is surjective; indeed, the
group IR× + is already exhausted by the absolute value of the idèles of the form
np (ap ) = (. . . , 1, 1, 1, ap , 1, 1, 1, . . . ), ap ∈ Kp× , where p is an infinite prime.
0
We denote by IK the (closed) kernel of this homomorphism, i.e., the group of
idèles of absolute value 1. By the product formula (cf. §1, p. 113) it contains
the principal idèle group K × . Thus the absolute value yields a (continuous)
homomorphism from the idèle class group CK onto IR× + with (closed) kernel
CK0
= IK0
/K × . The group CK 0
plays a very similar role in CK as the unit
group Up = {x ∈ Kp× | |x|p = 1} in the multiplicative group of a local field
Kp . The decomposition Kp× ∼ = Up × ZZ corresponds to the following
0
(7.4) Proposition. CK = CK × ΓK with ΓK ∼
= IR×
+.
Proof. We have to find a splitting for the group extension

||
0
1 −→ CK −→ CK −→ IR×
+ −→ 1,
i.e., we have to find an injection IR×
+ → CK which when composed with the
absolute value map | | : CK → IR× ×
+ yields the identity on IR+ . To do this, we
choose an infinite prime p and consider the embedding
np : Kp× −→ CK .
Now Kp× contains IR× ×
+ as a subgroup. If p is real, then np : IR+ → CK is an
×
injection of the desired form, since |np (x)| = |x|p = x ∈ IR+ . If p √
is complex,
then |np (x)| = |x|p = x2 , and one has to choose the map x 7→ np ( 2 x).
Let us mention that there is no distinguished subgroup of representatives in

0 ∼
CK for the factor group CK /CK = IR× +.
We will now show that, similar to the the unit group in the local case, the
0
group CK is compact. For this we need the following lemma.
(7.5) Lemma. For every prime p of the field K let an αp ∈ |Kp× |p (the value
group of | |p ) be given such that
1) αp = 1 forpalmost all p,
Q
2) p αp ≥ |∆|, where ∆ is the discriminant of K.
Then there exists an x ∈ K × with |x|p ≤ αp for all p.
e
Proof. We set αp = |πpp |p for p - ∞, where πp ∈ Kp is a prime element
for p.QBecause of 1) ep = 0 for almost all p, and we can consider the ideal
A = p-∞ pep , which, because αp = N(pep )−1 , has absolute norm
Y −1
N(A) = αp .
p-∞
Let a1 , . . . , an be an integral basis of A, γ1 , . . . , γn the embeddings of K into

the field C of complex numbers, and pi the infinite prime of K attached to γi .
We consider the linear forms
n
X
Li (x1 , . . . , xn ) = xk · γi (ak ) , i = 1, . . . , n.
k=1
If γi is real, we set L0i = Li , βi = αpi ; in the other √ case when γi and γj ,
i < j, are complex conjugates, we set Li = L0i + −1 · L0j , thus Lj = L̄i =
√
L0i − −1 · L0j , and βi = βj = αpi /2. If r is the number of complex primes,
p
it follows that
1 r 1 p
| det(L01 , . . . , L0n )| = √ det(L1 , . . . , Ln ) = r N(A) · |∆|
−2 −1 2
n
1 Y −1 p 1 Y Y
= r αp · |∆| ≤ r αp = βi .
2 2 i=1
p-∞ p|∞
The well-known Minkowski Theorem about Linear Forms (cf. e.g. [22], §17)
now yields an integral vector Z = (m1 , . . . , mn ) ∈ ZZn , such that |L0i (Z)| ≤ βi ,
√
i = 1, . . . , n, hence |Li (Z)| ≤ αpi (resp. ≤ αpi ) if pi is real (resp. complex).
Set x = m1 a1 + · · · + mn an ∈ A. Then |x|p ≤ αp for the finite primes by
construction of A, and for the infinite primes because |γi (x)| = |Li (Z)|.
0
(7.6) Theorem. The group CK is compact.
Proof. To simplify notation we set I = IK , I 0 = IK

0
, C = CK , and C 0 = CK
0
.
We consider the set
Y p p
K= Kp , with Kp := {a ∈ Kp× | 1/ |∆| ≤ |a|p ≤ |∆|},
all p
which is compact as the direct Tychonoff product ofpcompact spaces. We have

Kp = Up for all non-archimedian p with N (p) > |∆|, and since these are
all but finitely many, K is a subset of I. Since I 0 is closed in I, K0 := K ∩ I 0
is thus a compact subset of I 0 . To prove C 0 is compact it now suffices to

show that K0 is mapped by the (continuous) transformation I 0 → C 0 onto
the entire group C 0 = I 0 /K × , i.e., that for each idèle a ∈ I 0 there exists an
x ∈ K × with xa−1 ∈ K0 . For this we choose a fixed infinite prime q and set
p
αq = |∆| · |aq |q and αp = |ap |p for all p 6= q.
p
Because a ∈ I 0 , p αp = |∆|, and by (7.5) there is an x ∈ K × with
Q
|x|p ≤ αp , so that |x · a−1

p p
p |p ≤ 1 (≤ |∆| ), p 6= q, resp. |x · a−1
q |q ≤ |∆|. By
the product formula we obtain further
Y Y
1= |x · a−1 −1
p |p = |x · aq |q · |x · a−1
p |p ,
p p6=q
= ( p6=q |x · a−1
p
a−1 −1
Q
therefore |x · q |q p |p ) ≥ 1 ≥ 1/ |∆|; for p 6= q it follows
that |x · a−1 −1
Q −1
p
p |p ≥ p0 6=q |x · ap0 |p = 1/|x · aq |q ≥ 1/ |∆|.
0
Altogether we therefore have

p p
1/ |∆| ≤ |x · a−1
p |p ≤ |∆| for all p,
i.e., x · a−1 ∈ K0 . Thus all is proved.∗)
S
If S is a finite set of primes of K, then we let UK be the idèle group
S S
UK = {a ∈ IK |ap = 1 for p ∈ S; ap ∈ Up for p 6∈ S} ⊆ IK
and
U SK = UK
S
· K × /K × ⊆ CK .
In each neighborhood of the identity of CK there is a group U SK 20)

; this can
be seen from the following observations:
• The groups Y Y
Wp × Up ⊆ IK
p∈S p6∈S
form a fundamental system of neighborhoods of the identity of IK , where

Wp is a fundamental system of neighborhoods of the identity of Kp× and
S runs through all the finite primes (cf. p. 156),
S
• they always contain one of the groups UK ,
• passing from IK to CK , a fundamental system of neighborhoods is trans-
S
formed into a fundamental system of neighborhoods, and UK into U SK .
∗) Remark of the editor: The proof does not use the finiteness of the ideal class group.
But this finiteness follows from (7.6): The composition of the homomorphisms
0
CK ,→ CK JK /PK is surjective and continuous with respect to the discrete
topology on JK /PK . Consequently JK /PK is discrete as well as compact, and
thus finite.
20)
Note that the U SK themselves are not open. But since the groups Up are closed
in Kp× , the U S
K are closed.
The following theorem on Kummer extensions is the key ingredient for the
proof of the fundamental Existence Theorem announced above.
(7.7) Theorem. Let K be a field which contains the n-th roots of unity. If S
is a finite set of primes of K such that
1) S contains all the infinite primes and all the primes lying above the prime
numbers dividing n,
S
2) IK = IK · K ×,
√
n
n
then CK · U SK is the norm group of the Kummer extension T = K( K S )|K.
n
Addendum. If K does not contain the n-th roots of unity, CK · U SK is still
a norm group.
Proof. By (1.4) χ(GT |K ) ∼ = K S ·(K × )n /(K × )n ∼

= K S /(K S )n . Since K S is
finitely generated of rank N − 1 = |S| − 1 (cf. (1.1)) and contains the n-th
roots of unity, K S /(K S )n is the direct product of N cyclic groups of order n.
Therefore GT |K is also the direct product of N cyclic groups of order n.
It follows from this that for each an ∈ CK
n
we have (an , T |K) = (a, T |K)n = 1,
n n S
i.e., a ∈ NT |K CT , so that CK ⊆ NT |K CT . Furthermore, every idèle a ∈ UK
is a norm idèle of the extension T |K. To see this we must
√ convince ourselves
n
by (3.4) that ap is a norm of the local extension Kp ( K S )|Kp for every p.
For p ∈ S this is trivial because ap = 1.√If p 6∈ S, then ap ∈ Up , and ap is
n
by II, (4.4) a norm if the extension Kp ( K S )|Kp is unramified. But this is
immediately clear, since every a ∈ K S is a unit for p 6∈ S, and since n is
relatively prime to the characteristic char K p of the residue field if p 6∈√S, the
equation X n −a = 0 is separable over the residue field K p , so that Kp ( n a)|Kp
is unramified. It follows that U SK ⊆ NT |K CT and therefore
n
CK · U SK ⊆ NT |K CT .
We show this inclusion is an equality using an index argument. We have
(CK : NT |K CT ) = |GT |K | = (K S : (K S )n ) = nN , N = |S|
by the Reciprocity law. On the other hand,
n
(∗) (CK : CK ·U SK ) = (IK
S
·K × : (IK
S n S
) ·UK ·K × )
S
= (IK S n
: (IK S
) ·UK S
)/((IK ∩ K × ) : ((IK
S n S
) ·UK ∩ K × )) 21)
.
Here the index in the numerator is
Y
S
(IK S n
: (IK S
) ·UK )= (Kp× : (Kp× )n ) ,
p∈S
21) 12)
For this cf. the footnote on p. 133.
S
→ p∈S Kp× /(Kp× )n , a 7→ p∈S ap ·(Kp× )n is obviously
Q Q
since the map IK
S
surjective, and its kernel consists of those idèles a ∈ IK , for which ap ∈ (Kp× )n
S n S
for p ∈ S; but these are precisely the idèles in (IK ) ·UK .
By II, (3.7) we obtain, using that |n|p = 1 for p ∈
/ S, the identity
Y Y Y
S S n S × × n
(IK : (IK ) ·UK ) = (Kp : (Kp ) ) = n ·|n|p = n2N ·
2
|n|p = n2N .
p∈S p∈S p
For the index in the denominator of the equation (∗) we have IK S

∩ K× = KS
S n S × S n
and (IK ) ·UK ∩ K = (K ) . The first statement is clear. For the second, it
is clear that we have in any case the inclusion (K S )n ⊆ (IK
S n S
) · UK ∩ K ×.
S n S ×
To show the opposite inclusion, let x√∈ (IK ) ·UK ∩ K√ , thus x = an · u,
S S S
a ∈ IK , u ∈ UK . We form the field K(
√
n
x) and show K( x) = K. If b ∈ IK
n
,
×
then b is always a norm-idèle
√ of K( n
x)|K. In fact, if p ∈ S, then b p ∈ Kp
is a norm because Kp ( n x) = Kp ( n anp ) = Kp . If p 6∈ S the same holds, since
p
√ √
bp ∈ Up and Kp ( n x) = Kp ( n up )|Kp because p - n is unramified (cf. II,
(4.4)). If we keep in mind that IK = IK S
·K × , we have thus shown that
NK( √ √
n x)|K CK( n x) = CK ,
√
which
√ implies by the Reciprocity Law that K = K( n x). Hence we have
n
x = y ∈ K × , x = y n ∈ (K × )n ∩ K S = (K S )n .
From the equation (∗) we now obtain
n
(CK : CK · U SK ) = n2N /(K S : (K S )n ) = n2N /nN = nN = (CK : NT |K CT ),
n
and therefore NT |K CT = CK · U SK .
n
If we drop the assumption on the roots of unity, then CK ·U SK nevertheless
turns out to be a norm group, as claimed in the addendum. In fact, if the
field K does not contain the n-th roots of unity, then we adjoin these and let
K 0 |K be the resulting extension. If S 0 is a finite set of primes of K 0 which
contain all the primes above the primes in S and in addition is sufficiently
S0 0× n S0
large so that IK 0 = IK 0 · K , then by what we just proved CK 0 ·U K 0 is the
norm group of a normal extension L0 |K 0 . If L is the smallest normal extension

of K containing L0 , then we have
NL|K CL = NK 0 |K (NL0 |K 0 (NL|L0 CL )) ⊆ NK 0 |K (NL0 |K 0 CL0 ) =
0 0
n S n S n S
NK 0 |K (CK 0 ·U K 0 ) = (NK 0 |K CK 0 ) ·NK 0 |K U K 0 ⊆ CK ·U K .
n
Thus CK ·U SK , as a group which contains the norm group NL|K CL , is itself a
norm group (cf. (6.14)).
(7.8) Existence Theorem. The norm groups of CK are precisely the closed
subgroups of finite index.
Proof. Let NL = NL|K CL ⊆ CK be the norm group of a normal extension

L|K. By the Reciprocity Law the index (CK : NL ) = |Gab L|K | is finite. The
0
norm map NL|K : CL → CK is clearly continuous. We have CK = CK × ΓK ,
0 ∼ × ×
CL = CL × ΓL , with ΓK , ΓL = IR+ . The injection IR+ → CK from the proof
of (7.4) obviously also yields a group of representatives for CL /CL0 ; therefore
we can assume that ΓK = ΓL ⊆ CL . It follows that
NL|K CL = NL|K CL0 × NL|K ΓK = NL|K CL0 × ΓK
n
= NL|K CL0 × ΓK .
The image of the compact group CL0 in CK is compact, therefore closed, and
since ΓK ⊆ CK is also closed, NL|K CL is in fact closed.
Conversely, let N ⊆ CK be a closed subgroup of finite index (CK : N ) = n.
n
Then in any case CK ⊆ N.
Moreover, N is (as the complement of its finitely many closed cosets) also
open, and therefore contains one of the groups U SK (cf. the remark on p. 160).
n
But CK ·U SK is by the addendum to (7.7) a norm group for sufficiently large S,
and by (6.14) the same holds for the group N as it contains a norm group.
We have proved using (7.7) that for every closed subgroup N ⊆ CK of finite
index there exists a normal extension L|K with norm group NL|K CL = N ,
which is precisely the fundamental existence statement.
From the identification of the norm groups given by the Existence Theorem
we obtain without difficulty a further characterization of these groups which
is of a predominantly arithmetic nature. It is the idèle theoretic version of the
formulation of the Existence Theorem in the classical theory using ideals. 22) .
By a modulus m we mean a formal product
Y
m= pnp
p
of prime powers, such that np ≥ 0 and np = 0 for almost all primes p; for the
infinite primes p we allow only the exponents np = 0 and 1.
For a prime p of K let



 the np -th unit group of Kp , Up0 = Up , when p - ∞,

 IR× ⊆ K × , when p is real and n = 1,

n + p p
Up p = × ×
 IR = Kp , when p is real and np = 0,


 ×
C = Kp× , when p is complex.

22)
Cf. [17], Teil I, §4, Satz 1.
If ap ∈ Kp× , then let

n
ap ≡ 1 mod pnp ⇐⇒ ap ∈ Up p .
For a finite prime p and np ≥ 1 (resp. np = 0) this means ordinary congruence

(resp. ap ∈ Up ), for a real prime p with exponent np = 1 this is the positivity
condition ap > 0, and for the remaining cases where p is real and np = 0 or p
is complex, there is no restriction.
pnp is a modulus, then for an idèle a ∈ IK we set

Q
If m = p
a ≡ 1 mod m ⇐⇒ ap ≡ 1 mod pnp for all p,
and consider the group
n
Y
m
IK = {a ∈ IK | a ≡ 1 mod m} = Up p ⊆ IK .
p
If in particular m = 1, then obviously

Y Y
1
IK S∞
= IK = Kp× × Up ,
p|∞ p-∞
where S∞ is the set of infinite primes of K.

We call the idèle class group
m
CK m
= IK ·K × /K × ⊆ CK
m
the congruence subgroup mod m of CK . The factor group CK /CK is also
called the ray class group mod m. If particular, if m = 1, then we have
∼ IK /I S∞ ·K × .
CK /C 1 = IK /K × /I 1 · K × /K × =
K K K
Hence by (2.3) the ray class group mod 1 is isomorphic to the ideal class
group JK /PK , and its order is equal to the class number h of K.
(7.9) Theorem. The norm groups of CK are precisely the groups containing
m
the congruence subgroups CK .
m 1 1 m 1 m
Proof. The index (CK : CK ) = (CK : CK ) · (C
QK : CK ) =nph · (CK : CK ) =
1 × m × 1 m
h · (IK · K : IK · K ) ≤ h · (IK : IK ) = h · p (Up : Up ) is finite. Since
m
Q n m
IK = p Up p is open in IK , the image CK is also open and therefore closed
in CK . Thus by (7.8) the congruence subgroups, and all groups containing
them, are norm groups.
Assume conversely that N is a norm group of CK , so that by (7.8) N is a
closed subgroup of finite index. Then N is also open and has an open preimage
I in IK . This open preimage I contains an open subgroup W of the form
Y Y
W = Wp × Up ,
p∈S p6∈S
where S is a finite set of primes S and each Wp is an open neighborhood

of the identity of Kp× ; in fact, these groups form a fundamental system of
neighborhoods of the identity of IK (cf. p. 156). If p is finite, then we can
n n
assume Wp = Up p , since the unit groups Up p ⊆ Kp× form a basis. If p is
infinite, then the open sets Wp generate the entire group Kp× or the group
IR× m
+ in the real case. The group generated by W is thus a group IK ⊆ I for an
appropriate modulus m, and N is a group containing CK = IK ·K × /K × .
m m
m
The (abelian) class field L|K associated with the norm group CK is called the
ray class field mod m. By the Reciprocity Law its Galois group GL|K is
m
isomorphic to the ray class group CK /CK . In case K = Q, we have:
(7.10) Theorem. Let m be a positive integer, p∞ the infinite prime of Q,

and m the modulus m = m · p∞ . The ray class field mod m is precisely the
field Q(ζm ) of m-th roots of unity.
We point out that this theorem implies that the Existence Theorem for the
field Q is precisely the famous Theorem of Kronecker:
(7.11) Theorem. Every abelian extension of the field Q of rational numbers

is a cyclotomic field, i.e., a subfield of a field of roots of unity Q(ζ).
Proof of (7.10). Let ζ be a primitive m-th root of unity and m = p pnp .

Q
n np
m
= p6=p∞ Up p × IR× ⊆ Q×
Q
Then IQ + . The group Up p consists of norms in
the extension Qp (ζ)|Qp . In fact, the field Qp (ζ) is the compositum of the field
of pnp -th roots of unity Qp (ζpnp ) and the unramified extension over Qp of
the m0p -th roots of unity Qp (ζm0p ) (m0p = m/pnp ). The first field contains the
n
elements in Up p as norms because of II, (7.15), the second because of II, (4.4).
Since the norm group of the compositum Qp (ζ) = Qp (ζpnp )Qp (ζm0p ) is the
n
intersection of the two norm groups of Qp (ζpnp ) and Qp (ζm0p ), the group Up p
m
is contained in the norm group of Qp (ζ)|Qp . It follows from (3.4) that IQ
m
consists of norm idèles of the extension Q(ζ)|Q, i.e., CQ ⊆ NQ(ζ)|Q CQ(ζ) .
On the other hand,
m
(CQ : CQ 1
) = (CQ : CQ 1
) · (CQ m
: CQ 1
) = (IQ · Q× : I Q
m
· Q× )
1
= (IQ m
: IQ 1
)/((IQ ∩ Q× ) : (IQ
m
∩ Q× )).

n
1
= p6=p∞ Up × IR× and IQm
= p6=p∞ Up p × IR× 1 ×
Q Q
Since IQ + , we have IQ ∩ Q =
m ×
{1, −1} and IQ ∩ Q = {1}. We thus obtain for the above index the formula
1 Y
m
(CQ : CQ )= · (Up : Upnp ) · (IR× : IR×+)
2
p6=p∞
Y Y
= (Up : Upnp ) = pnp −1 · (p − 1) = ϕ(m).
p|m p|m
But this is the degree ϕ(m) = [Q(ζ) : Q] of the field extension Q(ζ)|Q, and it
m
follows that (CQ : CQ ) = (CQ : NQ(ζ)|Q CQ(ζ) ), i.e., the congruence subgroup
m
CQ is in fact the norm group of the cyclotomic field Q(ζ).
Because of (7.10) we can think of the ray class fields over a number field K as
the fields corresponding to the fields of roots of unity in case K = Q. In this
context Kronecker’s Theorem (7.11) corresponds to the deep generalization
that every abelian extension of K is a subfield of a ray class field. In this
sense class field theory appears as a generalization of the theory of cyclotomic
fields; in fact, the historical development of class field theory has been guided
to a large extend by the example of cyclotomic fields.
With the introduction of ray class fields we obtain a good overview over the
lattice of all abelian extensions of a base field K. The ray class fields them-
selves correspond to the different moduli m of K, where the larger modulus
corresponds to the smaller congruence subgroup, thus the larger ray class field.
More precisely, if m and m0 are two moduli of K, then m | m0 implies that
the ray class field mod m is contained in the ray class field mod m0 . Among
all ray class fields over K, there is one which plays a special role but only
appears when we get away from the base field K = Q. This is the ray class
1
field mod 1, i.e., the class field L|K associated with the congruence group CK
with modulus m = 1. It is called the Hilbert or also the absolute class field
1
over K. Its Galois group is canonically isomorphic to CK /CK , and therefore
to the ideal class group JK /PK (cf. p. 164). Its degree [L : K] is equal to the
ideal class number h of K. We will discuss the Hilbert class field in the next
section.
0 n
We use the compactness of the group CK and the fact that the groups CK · U SK
are norm groups (cf. (7.7) and Addendum) to prove the following theorem
about the universal norm residue symbol ( , K).
(7.12) Theorem. The universal norm residue symbol

( ,K)
CK −−−−→ Gab
K
from CK to the topological Galois group Gab K of the maximal abelian exten-
sion A K |K is continuous, surjective homomorphism, and its kernel DK =
23)
T
L NL|K CL is the group of all infinitely divisible elements of CK , i.e.,
\∞
n
DK = CK .
n=1
T∞ n
Proof. We first prove the last statement. If a ∈ n=1 CK , and if NL =
n
NL|K
T∞ C L is any norm group, then a = b ∈ N L if (C K : N L ) = n. Therefore
n
T
C
n=1 K ⊆ D K = L N L|K C L .
n
On the other hand, the groups CK · U SK are norm T groups for sufficiently large
n S ∞ n
S, i.e., DKT⊆ CK · U K . For the inclusion D K ⊆ n=1 CK it thus suffices to
n
· U SK = CKn n
· U SK ; then for every S we have
T
show that S CK . Let a ∈ S CK
n
a = bS · ūS , bS ∈ CK , ūS ∈ U SK . Because S UK S
T
the representation = 1 we
S
T
also have S U K = 1, and this means that the sequence ūS for increasing S
converges to 1, i.e., the sequence a· ū−1 n
S ∈ CK converges to a: a = limS a· ūS .
−1
n
Consider now CK = CK0
× ΓK (ΓK ∼ = IR× n
+ ). Since CK −→ CK is continuous,
0
CK is compact and ΓK is closed, thus we see that the group
n 0 n n 0 n
CK = (CK ) × ΓK = (CK ) × ΓK
−1 n
T n S n
is closed, so that a = limS a · ūT
S ∈ CK . Hence DK ⊆ K · U K = CK for
S CT
∞ n ∞ n
all n, which shows that DK ⊆ n=1 CK , and proves DK = n=1 CK .
In particular, since CK = CK 0
× ΓK and ΓK ∼ = IR×+ consists of infinitely
divisible elements, this group is contained in the kernel DK of ( , K), i.e.,
0
(CK , K) = (CK , K) ⊆ Gab
K.
( ,K)
Now if the homomorphism CK −−−−→ Gab K is continuous, then the compact-
0 0
ness of CK implies that the image (CK , K) = (CK , K) is closed in Gab
K , and
because of denseness we have (CK , K) = GabK , which will prove surjectivity.
But to show the map ( , K) is continuous is almost trivial. If H is an open
subgroup of GabK , thus a closed subgroup of finite index, and L is the fixed
field of H, then the norm group NL = NL|K CL ⊆ CK is open, and because
(NL , L|K) = (NL , K) · H = 1 it is mapped by ( , K) into H.
In addition to the above characterization, the group DK has a purely topo-

logical description as the connected component of the identity of CK . For the
proof of this non-trivial fact we refer the interested reader to [2], Ch. 9.
23)
An element a ∈ CK is infinitely divisible, if for every natural number n there
n
exists a b ∈ CK with a = b .
§ 8. The Decomposition Law
Many of the deepest statements of number theory find their common expres-
sion in Artin’s Reciprocity Law. For example, and without giving details,
one can regard Gauss’s Reciprocity Law for quadratic residues as a special
case24), and more generally, the theory of higher power residues is dominated
by Artin’s Reciprocity Law. Another important application concerns the ques-
tion of which ideals of a base field K become principal in an extension field L,
which we will return to later. The most important consequence, however, is
the answer to the question of how the prime ideals p of a basis field K split in
an abelian extension. For this, we consider instead of the prime ideal p of K
an associated “prime idèle” by choosing a prime element π ∈ Kp and forming
the idèle np (π) = (. . . , 1, 1, π, 1, 1, . . . ). If we first disregard the finitely many
ramified primes, then the decomposition of the prime ideal p in the abelian
extension L|K can be immediately read off from a relation between the prime
idèle np (π) in the norm group NL = NL|K CL ⊆ CK which determines L,
namely, simply from the order of the idèle class np (π) modulo NL . This is the
content of the following theorem:
(8.1) Theorem. Let L|K be an abelian extension of degree n and p an

unramified prime ideal of K.
If π ∈ Kp is a prime element, np (π) ∈ CK is the idèle class represented by
the idèle np (π) = (. . . , 1, 1, 1, π, 1, 1, 1, . . . ), and f is the smallest number such
that
np (π)f ∈ NL|K CL ,
then the prime ideal p factors in the extension L into r = n/f distinct prime
ideals P1 , . . . , Pr of degree f .
Hence if one knows the norm group NL|K CL , then one can simple read off
from the idèle class group CK of the base field K how p decomposes in L.
Proof. Since the prime ideal p is unramified, it factors in L into dis-

tinct prime ideals of equal degree: p = P1 · · · Pr . By the reciprocity law
CK /NL|K CL ∼ = GL|K , and np (π) mod NL|K CL in CK /NL|K CL has the same
order f as (cf. (6.15))
(np (π), L|K) = (π, LP |Kp ) ∈ GLP |Kp ⊆ GL|K (P | p).
But by II, (4.8) (π, LP |Kp ) = ϕp is the Frobenius automorphism of the un-
ramified extension LP |Kp . It generates the group GLP |Kp , thus the order f
24)
The terminology “reciprocity law” also comes from this; this might appear a little
strange, since at first glance Artin’s Reciprocity Law seems to have nothing in
common with Gauss’s law.
§ 8. The Decomposition Law 169
coincides with the degree [LP : Kp ], and therefore with the degree of P. There-
fore the number r of distinct prime ideals P1 , . . . , Pr over p can be computed
from this by the fundamental equation of number theory: n = r · f .
Later, when we derive the classical, ideal-theoretic theorems of class field

theory from our idèle-theoretic theorems, we will encounter Theorem (8.1)
again in another, purely ideal theoretic formulation.
Theorem (8.1) only applies to unramified prime ideals. To decide whether a

prime ideal p is unramified, the following observations are useful.
If L|K is an abelian extension of algebraic number fields, and if LP |Kp are
the associated local extensions, then L is uniquely determined by the norm
group NL|K CL ⊆ CK and LP by the norm group NLP |Kp L× ×
P ⊆ Kp . There
is a simple relation between these two norm groups. To see this, we use (7.3)
and consider the group Kp× as embedded in CK via the homomorphism
np : Kp× −→ CK ,
where we identify xp ∈ Kp× with the idèle class represented by the idèle
np (xp ) = (. . . , 1, 1, 1, xp , 1, 1, 1, . . . ). Using the abbreviations N = NL|K and
NP = NLP |Kp we then obtain the
(8.2) Proposition. If L|K is an abelian extension, then

N CL ∩ Kp× = NP L×
P.
Proof. If xp ∈ NP L×
P , then the idèle np (xp ) = (. . . , 1, 1, 1, xp , 1, 1, 1, . . . ) has
only norm components, and is therefore a norm idèle of L by (3.4). Thus
NP L× ×
P ⊆ N CL ∩ Kp .
Conversely, let a ∈ N CL ∩ Kp× . Then a is represented on one hand by the

norm idèle a = N b, b ∈ IL , and on the other hand by an idèle np (xp ) =
(. . . , 1, 1, 1, xp , 1, 1, 1, . . . ), xp ∈ Kp× , so that
np (xp ) · a = N b with a ∈ K × .
Passing to the components, we see that a is a norm for all q 6= p. Therefore it
follows from the product formula (6.15) that a is also a norm at the prime p, so
that xp ∈ NP L× ×
P , which proves the other inclusion N CL ∩ K ⊆ NP LP .
×
The following theorem now shows how the ramification of a prime p of K in

an abelian extension L|K is reflected in the norm group N CL ⊆ CK . We call
an infinite prime p unramified if it splits completely, i.e., if LP = Kp .
(8.3) Theorem. Let L|K be an abelian extension, N = N CL ⊆ CK its norm

group and p a prime of K. Then we have the following equivalences:
p is unramified in L ⇐⇒ Up ⊆ N ,
p splits completely in L ⇐⇒ Kp× ⊆ N .
Proof. The prime p is unramified ⇔ LP |Kp is unramified (i.e., LP = Kp for

p|∞) ⇔ (by II, (4.9) and (8.2)) Up ⊆ NP L× ×
P = N ∩ Kp ⇔ Up ⊆ N . Similarly,
×
p splits completely ⇔ LP = Kp ⇔ Kp = NP LP = N ∩ Kp× ⇔ Kp× ⊆ N .
×
As in the local case, we have in global class field theory the notion of a
conductor. For a local abelian extension LP |Kp the conductor fp was defined
as the smallest p-power pn such that Upn ⊆ NP L× P . In the global case we
have to replace the p-powers pn by the moduli m (cf. §7, p. 163) and the
groups Upn = {xp ∈ Kp× | xp ≡ 1 mod pn } by the congruence subgroups
m
CK = {a ∈ CK | a ≡ 1 mod m}.
If we keep in mind that by (7.9) every norm group N ⊆ CK contains a

m0
congruence subgroup CK m
, and that m | m0 implies the inclusion CK m
⊆ CK
(and not conversely!), then we come to the following definition.
(8.4) Definition. Let L|K be an abelian extension with norm group N =

N CL . By the conductor f of N , or also of L|K, we mean the g.c.d. of all
m
moduli m such that CK ⊆ N.
f m
Thus CK is the largest congruence subgroup contained in N , and CK ⊆N
m
if and only if f | m. Note that N = CK , i.e. N is the norm group of the ray
class field mod m, does not imply in general that m is the conductor of N
1 p∞
(for example, CQ = CQ by (7.10)).
For the conductor f of an abelian extension L|K one has a localization theorem
analogous to that for the discriminant. If we define for an infinite prime the
conductor fp by p or 1, depending on whether LP 6= Kp or LP = Kp , we have
(8.5) Theorem. If f is the conductor of an abelian extension L|K and fp is

the conductor of the local extension LP |Kp , then
Y
f= fp .
p
Proof. We have to show: If N = N CL and m = p pnp is a modulus of K,

Q
Y
CKm
⊆ N ⇐⇒ fp | m ⇐⇒ fp | pnp for all p.
p
§ 8. The Decomposition Law 171
Using the notation from §7, p. 163 and (8.2), this follows from the equivalences
m
CK ⊆ N ⇐⇒ (a ≡ 1 mod m =⇒ a ∈ N ) for a ∈ IK
⇐⇒ (ap ≡ 1 mod pnp =⇒ np (ap ) ∈ N ∩ Kp× = NP L×
P)
n
⇐⇒ (ap ∈ Up p =⇒ ap ∈ NP L×
P)
n
⇐⇒ Up p ⊆ NP L×
P
⇐⇒ fp | pnp .
We call an infinite prime p of K ramified in L if LP 6= Kp . With this definition,

we obtain from II, (7.21) the following result.
(8.6) Theorem. A prime p of K is ramified in L if and only if it appears in

the conductor f of L|K.
If we further call L|K unramified if all finite as well as all infinite primes are
unramified, this implies
(8.7) Corollary. An abelian extension L|K is unramified if and only if its

conductor f = 1.
In §7, p. 166 we called the ray class field mod 1, i.e., the class field associated
1
with the norm group CK , the Hilbert class field over K. We can characterize
this field as follows:
(8.8) Theorem. The Hilbert class field over K is the maximal unramified
extension of K.
1 ∼
Because of the isomorphism CK /CK = JK /PK , the degree of the Hilbert class
field over K is the ideal class number h = (JK : PK ) of K (cf. §7, p. 166).
Therefore if h = 1, which occurs, for example, if K = Q, then every abelian
extension of K is ramified, and the Hilbert class field coincides with K.
The following famous theorem has been conjectured by Hilbert but remained
unproved for a long time.
(8.9) Principal Ideal Theorem. In the Hilbert class field over K every
ideal a of K becomes a principal ideal.
E. Artin used the reciprocity law to reduce the proof of this theorem to a
purely group theoretic problem whose solution was given shortly thereafter
by Ph. Furtwängler. In the following we explain Artin’s reduction.
If we start with K and take its Hilbert’s class field K1 , then the Hilbert class
field K2 of K1 , and we continue in this way, we obtain a chain of class fields
K = K0 ⊆ K1 ⊆ K2 ⊆ . . . ,
the so-called class field tower. For this tower of fields we show first:
(8.10) Proposition. The i-th class field Ki is normal over K, and K1 is the
largest abelian subfield of K2 ; in other words:
The Galois group GK2 |K1 is the commutator subgroup of GK2 |K .
Proof. We assume as induction hypothesis Ki , i ≥ 1, is normal. Let σ be

an isomorphism of Ki+1 |K. Then σKi = Ki , and since Ki+1 |Ki abelian and
unramified, the same holds for σKi+1 |σKi , i.e., σKi+1 |Ki . Then by (8.8)
σKi+1 ⊆ Ki+1 , thus Ki+1 |K is normal.
Let K 0 be the maximal abelian extension of K contained in K2 . Then we have
K1 ⊆ K 0 , and K 0 ⊆ K1 , since K 0 |K is unramified; thus in fact K 0 = K1 .
For the proof of the Principal Ideal Theorem we now have to translate the
statement that every ideal of K becomes a principal ideal in the class field K1
into the language of idèles. This is obviously equivalent to the canonical map
JK /PK −→ JK1 /PK1 ,
that takes the ideal a ∈ JK to the ideal in the field K1 generated by a being
trivial. On the other hand, by §7, p. 164 we have the canonical isomorphisms
CK /C 1 ∼
K = JK /PK and CK /C 1 ∼ 1 K1 = JK /PK ,
1 1
and therefore the commutative diagram

CK /CÊÇÈÉ 1 JK /PK
K
1
CK1 /CK 1
JK1 /PK1 ,
where the homomorphism i is induced from the canonical embedding CK →
CK1 . Thus the Principal Ideal Theorem says precisely that the map i is trivial,
1
which is equivalent to the statement that we have an inclusion CK ⊆ CK 1
.
1
Since CK 1
is the norm group of the extension K2 |K1 , this means that we
simply have to show, making use of the norm residue symbol, that
1 = (CK , K2 |K1 ) = Ver(CK , K2 |K).
§ 9. The Ideal Theoretic Formulation of Class Field Theory 173
Now if we take into account that

(CK , K2 |K) = Gab
K2 |K = GK2 |K /GK2 |K1 = GK1 |K ,
and that GK2 |K1 is the commutator subgroup of GK2 |K , then we see that the
Principal Ideal Theorem reduces to the following theorem:
(8.11) Theorem. If G is a metabelian finite group, i.e., a finite group with

abelian commutator subgroup G0 , then the transfer of G to G0 25)
ab
Ver : Gab −→ G0 = G0
is the trivial map.
This theorem is purely group theoretic but by no means of a simple nature. It

was first proved (1930) by Ph. Furtwängler, a simpler proof was given by
S. Iyanaga (1934). It would exceed the scope of these lectures to give this
proof, instead we refer to [23].
A problem posed by Furtwängler, which is closely related to the Principal

Ideal Theorem, is the famous class field tower problem. It is the question
of whether the class field tower
K = K0 ⊆ K1 ⊆ K2 ⊆ . . .
(Ki+1 the Hilbert class field over Ki ) terminates after finitely many steps.
A positive answer to this question would have the following interesting con-
sequence: If Ki+1 = Ki for sufficiently large i, then Ki has class number 1.
One would therefore obtain for every algebraic number field K a canonically
given solvable extension, in which not only the ideals of K but all the ideals
become principal ideals. This problem was open for a long time until in the
year 1964 is was decided in the negative by E.S. Golod and I.R. Šafarevič
who showed that there are in fact infinite class field towers. It is interesting to
note that similar to the proof of the Principal Ideal Theorem, this was done
using a reduction to a purely group theoretic conjecture which was solved
shortly afterwards. We refer the interested reader to [11], IX, and [14].
§ 9. The Ideal Theoretic Formulation of Class

Field Theory
The results of global class field theory obtained so far are almost exclusively
formulated in terms of idèles. We have seen that the idèle-theoretic language
25)
We recall that the transfer or Verlagerung of a group G to a subgroup g is defined
res
as the restriction Gab ∼
= H −2 (G, ZZ) −−→ H −2 (g, ZZ) ∼
= g ab .
has remarkable technical advantages which justify its central role in our dis-
cussions. Now that we have reached some form of conclusion, we should also
derive from the results obtained so far the classical purely ideal-theoretic the-
orems of class field theory as stated, for example, in Hasse’s Zahlbericht [17].
In the idèle-theoretic formulation of the reciprocity law, the abelian extensions

L|K correspond uniquely to the norm groups NL|K CL ⊆ CK . In the ideal-
theoretic form there is a similar correspondence, however, not quite as simple.
Here the abelian extensions also correspond to certain norm groups in the
ideal group JK of K. The norm residue symbol ( , L|K) : CK → GL|K
is replaced by a symbol which maps the ideals a ∈ JK to elements of the
Galois group GL|K . It is a characteristic of the ideal-theoretic version that
such a symbol cannot be defined for all ideals, in contrast to the norm residue
symbol, which is defined for all idèles without any restriction. More precisely,
the ideals that have to be omitted are the ones which are divisible by ramified
primes. This is done by choosing a (sufficiently large) so-called “modulus of
definition m”, in which all the ramified prime ideals appear, and to which
the ideals being considered are to be relatively prime. Field theoretically this
choice corresponds to the embedding of the abelian extension L|K into the
ray class field mod m, in which by §8 only prime divisors of m are ramified.
This process, namely the choice of a sufficiently large modulus of definition m
for every given abelian extension L|K, the transition to the ideals relatively
prime to m, and the embedding of L|K into the ray class field mod m, are
the crucial inputs to the ideal-theoretic formulation of the reciprocity law. We
describe this more precisely below.
We take an algebraic number field K as our base field. By JK (resp. PK ), we

denote again the ideal (resp. principal ideal) group of K; however, since the
base field K is always fixed, we omit the index K, just as for the idèle group
IK and the idèle class group CK , and we write J, P , I, C etc.
If m = p pnp is a modulus of K, then we let
Q
J m be the group of all ideals relatively prime to m,

26)
P m the group of all principal ideals (a) ∈ P with a ≡ 1 mod m .
P m is called the ray mod m, and every group between P m and J m is in the
classical terminology called an ideal group defined mod m. For the ideal
groups defined mod m we use the notation H m .
The factor group J m /P m is called the ray class group mod m. Obviously,
if m = 1, then J m = J and P m = P , i.e., we obtain the full ideal class group
J/P as the ray class group mod 1.
In place of the idèle class group C we now have a whole family of ray class
groups J m /P m . Whereas the idèle class group C was the source group for all
26)
This congruence is again to be understood in the sense of §7, p. 164.
abelian extensions of the base field K, the ray class groups J m /P m are only
responsible for the extensions contained in the ray class field mod m.
If L|K is any extension, then we call a modulus m a modulus of definition

for L |K if L lies in the ray class field mod m (i.e., C m ⊆ NL|K CL ). If m | m0
and if m is a modulus of definition for L|K, then so is m0 ; by definition, the
conductor of L|K is the g.c.d. of all the moduli of definition (cf. (8.4)).
After choosing a modulus of definition m, we now associate with every abelian

extension L|K the following ideal group defined mod m:
(9.1) Definition. If L|K is an abelian extension and m is a modulus of

definition for L|K, then
H m = NL|K JLm · P m
is called the ideal group defined mod m associated with L|K. Here JLm
denotes the group of all ideals of L relatively prime to m.
The map from the norm group H m /P m into the ray class group J m /P m is
the ideal-theoretic analogue of the map from the norm group NL|K CL into
the idèle class group CK .
In place of the norm residue symbol we now define a homomorphism

L|K
m
J −−−−−−→ GL|K ,
by mapping each ideal a of K relatively prime to m to an automorphism L|K

a
in GL|K , which is called the Artin symbol. By multiplicativity, it suffices to
consider the prime ideals p of K which do not divide m. For these we set

L|K
= ϕp ∈ GL|K ,
p
where ϕp is the Frobenius automorphism of L|K associated with p. We
briefly recall its definition: If P is a prime of L lying over p, then ϕp is an
element of the decomposition group GP = GLP |Kp ⊆ GL|K of P over K and
is as such (because there is no ramification) uniquely determined by
ϕp a ≡ aq mod P for all integers a ∈ L.
Here q is the number of elements of the residue field of p; ϕp does not depend
on the choice of the prime ideals P but only on p, since GL|K is abelian and
an ideal conjugate to P yields a Frobenius automorphism conjugate to ϕp .
The Artin Reciprocity Law in its classical ideal-theoretic form now reads
as follows:
(9.2) Theorem. If L|K is an abelian extension and m is a modulus of defi-

nition for L|K, then the Artin symbol yields the exact sequence

L|K
m m m m
1 −→ H /P −→ J /P −−−−−−→ GL|K −→ 1,
where J m is the group of all ideals of K relatively prime to m and H m is the

ideal group defined mod m associated with L (cf. (9.1)).
Remark. Note that this implies in particular that the Artin symbol does
not depend on the ideals themselves but only on the ideal classes mod P m ;
therefore it induces a homomorphism on the class group J m /P m −→ GL|K .
We will prove the exactness of the sequence

L|K
1 −→ H m /P m −→ J m /P m −−−−−−→ GL|K −→ 1
by comparing it with the analogous idèle-theoretic sequence
( ,L|K)
1 −→ NL|K CL −→ CK −−−−−→ GL|K −→ 1
which is exact. More precisely, we will compare it with the exact sequence
( ,L|K)
1 −→ NL|K CL /C m −→ C/C m −−−−−→ GL|K −→ 1,
where we pass from idèles to ideals as in (2.3), using the homomorphism
Y
κ : I −→ J, a 7−→ pvp (ap ) .
p-∞
For this homomorphism κ we now have the
(9.3) Proposition. The map κ induces a canonical isomorphism of ray class

groups
κm : C/C m −→ J m /P m ,
whose restriction to NL|K CL /C m yields an isomorphism
κm : NL|K CL /C m −→ H m /P m .
Proof. To prove κm is an isomorphism, we start with the isomorphism

C/C m ∼
= I/I m ·K × .
Thus we need to find for every idèle class a · I m ·K × a representative idèle a,
which is mapped under κ to the group J m of ideals relatively prime to m. We
find these representatives in the group
I hmi = {a ∈ I | ap = 1 for all p | m}.
In fact, if a ∈ I, then by the Approximation TheoremQthere is an element

a ∈ K × such that ap · a ≡ 1 mod pnp for all p | m = p pnp . Therefore we
can write a · a = a0 · b, where a0 , resp. b, is defined by the components
a0p = 1 for p | m, a0p = ap · a for p - m, resp.
bp = ap · a ≡ 1 mod pnp for p | m, bp = 1 for p - m.
Since a0 ∈ I hmi and b ∈ I m , we have a ∈ I hmi ·I m ·K × . The isomorphism κm

results now from the fact that the homomorphism
I hmi −→ J m /P m , a 7−→ κ(a) · P m
is surjective and has kernel I m · K × ∩ I hmi , which is easy to verify. Hence
∼ I hmi ·I m ·K × /I m ·K × ∼
C/C m = = I hmi /I m ·K × ∩ I hmi .
For the second claim we have to show further that

κm (NL|K CL /C m ) = NL|K JLm · P m /P m .
hmi
If we set IL = {a ∈ IL | aP = 1 for P | m} ⊆ IL , then we have from the
hmi
above IL = IL · ILm · L× , and therefore
hmi
NL|K CL /C m = NL|K IL ·K × /I m ·K × = NL|K IL · I m ·K × /I m ·K × .
From the definition of κm it follows that
hmi hmi
κm (NL|K CL /C m ) = κ(NL|K IL )·P m /P m = NL|K (κ(IL ))·P m /P m ,
hmi
and because κ(IL ) = JLm , we obtain the desired result.
The isomorphism κm : C/C m −→ J m /P m yields a surjective homomorphism

κm : C −→ J m /P m
with kernel C m . Since J m is generated by the prime ideals p - m, we can
describe the map κm as follows.
Let p - m be a prime ideal of K, π ∈ Kp a prime element in Kp , and np (π) =
(. . . , 1, 1, π, 1, 1, . . . ) the “prime idèle” associated with p in the idèle class
np (π) = np (π) · K × ∈ C. Then the map κm takes the idèle class np (π) to the
class p · P m , since κ(np (π)) = p.
Theorem (9.2) is now an immediate consequence of the following theorem

which establishes a relation between the idèle-theoretic and the ideal-theoretic
reciprocity law.
(9.4) Theorem. Let L|K be an abelian extension, and let m be a modulus

of definition for L|K. Then the diagram
1ÕÔÒÓÎÏÐÑËÌÍ
( ,L|K)
NL|K CL CK GL|K 1
κm κm id
L|K
( )
1 H m /P m J m /P m GL|K 1
m
commutes, and the surjective homomorphisms κm both have kernel CK .
Addendum: If p - m is a prime ideal of K, π ∈ Kp is a prime element, and
np (π) = (. . . , 1, 1, π, 1, 1, . . . ) is the prime idèle associated with p in the idèle
class np (π) ∈ CK , then the Artin and the norm residue symbols satisfy

L|K
= np (π), L|K .
p
Proof. The equality follows easily from the product formula for the norm
residue symbol (6.15) and II, (4.8):

L|K
= ϕp = (π, LP |Kp ) = (np (π), L|K).
p
Because κm (np (π)) = p, it follows that the diagram commutes.
In this section we always considered only unramified prime ideals and excluded
the ramified primes by choosing a modulus of definition m, thus embedding
the field in question into a suitable ray class field.
It is easy to see that this
restriction is necessary. The Artin symbol L|K p , which for an unramified
prime ideal p is defined by the above Addendum as the norm residue symbol
(np (π), L|K) = (π, LP |Kp ), does not permit such a definition in the ramified
case, because the norm residue symbol still depends on the choice of the prime
element π. The inclusion of the ramified primes into class field theoretical
required the step from ideals to idèles, which allows to reduce the global to the
local theory, where one can deal with ramified extensions using cohomological
methods.
To end, we want to formulate the decomposition law for the unramified prime
ideals in an abelian extension L|K in terms of the corresponding ideal group
defined mod m H m , which determines the field L as a class field. This gives
the ideal-theoretic formulation of Theorem (8.1).
(9.5) Theorem. Let L|K be an abelian extension and p an unramified prime

ideal of K. Let further m be a modulus of definition for L|K not divisible by
p (say the conductor) and let H m be the corresponding ideal group. If f is
the order of p mod H m in the class group J m /H m , and therefore the smallest
number such that
pf ∈ H m ,
then p factors in the extension field L into exactly r = [L : K]/f distinct
prime ideals P1 , . . . , Pr of equal degree f over p.
Proof. Let p = P1 · · · Pr is the prime decomposition of p in L. Because p

is unramified, the P1 , . . . , Pr are distinct and of equal degree f over p. This
degree coincides with the order of the decomposition group of Pi over K, and
therefore with the order of the Frobenius automorphism ϕp ∈ GL|K , which
generates the decomposition group. Under the isomorphism
J m /H m ∼
= GL|K

induced by the Artin symbol the element ϕp = L|K p corresponds to the
class p · H m ∈ J m /H m , hence this class has order f . q.e.d.
References
1. Albert, A. Structures of Algebras. A.M.S. Publ., Providence (1961).

2. Artin, E. & Tate, J. Class Field Theory. Harvard (1961).
3. Artin, E. Die gruppentheoretische Struktur der Diskriminanten algebra-
ischer Zahlkörper. J. reine und angew. Math. 164, 1 – 11, (1931).
4. Artin, E. Kohomologie endlicher Gruppen. Universität Hamburg (1957).
5. Artin, E. Theory of Algebraic Numbers. Lecture notes, Göttingen (1959).
6. Borevič, S.I. & Šafarevič, I.R. Zahlentheorie. Birkhäuser Verlag, Basel-
Stuttgart (1966).
7. Bourbaki, N. Algèbre. Ch. V. Hermann, Paris (1950).
8. Bourbaki, N. Algèbre commutative. Hermann, Paris (1964).
9. Bourbaki, N. Topologie générale, Ch. I, II, III. Hermann, Paris (1961).
10. Cartan, H. & Eilenberg, S. Homological Algebra. Princeton Math. Ser.
No 19, (1956).
11. Cassels, J.W.S. & Fröhlich, A. Algebraic Number Theory. Thompson
Book Comp. Inc. Washington D.C. (1967).
12. Chevalley, C. Class Field Theory. Universität Nagoya (1954).
13. Endler, O. Bewertungstheorie. Bonner Math. Schr. 15, Bd. 1, 2, (1963).
14. Golod, E.S. & Šafarevič, I.R. Über Klassenkörpertürme (russisch). Izv.
Akad. Nauk. SSSR, 28, 261 – 272, (1964). Engl. Übersetzung in A.M.S.
Transl. (2) 48, 91 – 102.
15. Grothendieck, A. Sur quelques points d’algèbre homologique. Tohoku
Math. J. 9, 119 – 221, (1957).
16. Hall, M. (Jr.) The Theory of Groups. MacMillan, New York (1959).
17. Hasse, H. Bericht über neuere Untersuchungen und Probleme aus der The-
orie der algebraischen Zahlkörper. Jahresber. der D. Math. Ver. 35, (1926),
36, (1927); 39, (1930).
18. Hasse, H. Die Normresttheorie relativ-abelscher Zahlkörper als Klassen-
körpertheorie im Kleinen. J. reine u. angew. Math. 162, 145 – 154, (1930).
19. Hasse, H. Die Struktur der R. Brauerschen Algebrenklassengruppe über
einem algebraischen Zahlkörper. Math. Annalen, 107, 248 – 252, (1933).
20. Hasse, H. Führer, Diskriminante und Verzweigungskörper abelscher Zahl-
körper. J. reine und angew. Math. 162, 169 – 184, (1930).
21. Hasse, H. Zahlentheorie. Akademie Verlag Berlin (1963).
182 References
22. Hilbert, D. Die Theorie der algebraischen Zahlkörper. Jahresber. der D.

Math. Ver. 4, (1897).
23. Iyanaga, S. Zum Beweis des Hauptidealsatzes. Hamb. Abh. 10, 349 – 357,
(1934).
24. Kawada, Y. Class Formations. Duke math. J. 22 (1955).
25. Kawada, Y. Class Formations II. (With Satake, Ichiro). J. Fac. Sci. Univ.
Tokyo, Sect. I, 7, (1956).
26. Kawada, Y. Class Formations III, IV. J. Math. Soc. Japan 7, (1955), 9,
(1957).
27. Krull, W. Galoissche Theorie der unendlichen algebraischen Erweiterun-
gen. Math. Annalen 100, 687 – 698, (1928).
28. Krull, W. Zur Theorie der Gruppen mit Untergruppentopologie. Hamb.
Abh. 28, 50 – 97, (1965).
29. Lang, S. Algebra. Addison-Wesley, Reading, Mass. (1965).
30. Lang, S. Algebraic Numbers. Addison-Wesley, Reading, Mass. (1964).
31. Lang, S. Rapport sur la Cohomologie des Groupes. Benjamin, New York
(1966).
32. Lazard, M. Sur les groupes de Lie formel à un paramètre. Bull. Soc. Math.
France 83, 251 – 274, (1955).
33. Lubin, J. One parameter formal Lie groups over p-adic integer rings. Annals
of Math. 80, 464 – 484, (1964).
34. Lubin, J. & Tate, J. Formal Complex Multiplication in Local Fields. An-
nals of Math. 81, 380 – 387, (1965).
35. MacLane, S. Homology. Springer-Verlag, Berlin-Göttingen-Heidelberg
(1963).
36. Poitou, G. Cohomologie Galoisienne des Modules finis. Dunod, Paris
(1967).
37. Pontryagin, L. Topologische Gruppen. Teile 1, 2. Teubner, Leipzig (1957–
58).
38. Schilling, O.F.G. The Theory of Valuations. Math. Surveys IV, New York
(1950).
39. Shimura, G. & Taniyama, Y. Complex Multiplication of Abelian Varieties.
Publ. Math. Soc. Japan 6, (1961).
40. Serre, J.-P. Abelian `-adic Representations and Elliptic Curves. Benjamin,
New York (1968).
41. Serre, J.-P. Cohomologie Galoisienne. Lecture Notes in Math. 5, Springer-
Verlag, Berlin-Heidelberg-New York (1964).
42. Serre, J.-P. Corps locaux. Hermann, Paris (1962).
43. Serre, J.-P. Groupes algebriques et corps de classes. Hermann, Paris
(1959).
44. Stöhr, K.-O. Homotopietheorie pro-algebraischer Gruppen und lokale
Klassenkörpertheorie. Universität Bonn (1967).
45. Tate, J. The Higher Dimensional Cohomology Groups of Class Field The-
ory. Ann. of Math. 56, 294 – 297, (1952).
46. v.d. Waerden, B.L. Algebra. Springer-Verlag, 1. Teil (4. Auflage) Berlin
(1955), 2. Teil (4. Auflage) Berlin (1959).
47. Weil, A. Basic Number Theory. Springer-Verlag, New York (1967).
48. Weiss, E. Algebraic Number Theory. McGraw-Hill, New York (1963).
49. Zariski, O. & Samuel, P. Commutative Algebra, I – II. Van Nostrand,
New York (1958 – 60).
Index
abelianization, 20 Existence Theorem, 75, 96, 162

absolute norm, 79
absolute value, 79, 158 factor systems, 18
anticommutativity, 25, 49 field formation, 65, 126, 147
Artin symbol, 175, 178 fixed group, 6
augmentation, 4 formation, 64
augmentation ideal, 4 Frobenius automorphism, 83, 88, 175
fundamental class, 69, 92, 152
Brauer group, 78
G-free, 9
G-homomorphism, 6
character group, 17, 31
G-induced modules, 27
class field, 76, 97, 154, 172
G-module, 3
class field tower, 172
G/g-induced, 43
class field tower problem, 173
group extensions, 18
class field, absolute, 166
group ring, 3
class field, Hilbert, 166, 171
class formation, 63, 67, 70, 85, 92, 151 Herbrand module, 53, 129
class number, 114 Herbrand quotient, 53, 129
coaugmentation, 4 higher unit group, 79
cohomology group, 15 Hilbert’s Theorem 90, 77
cohomology sequence homology groups, 15
exact, 22 homomorphisms, crossed, 17
cohomology, trivial, 28
components, local, 125 ideal class group, 114
conductor, 109, 170 ideal group, 174
congruence subgroup, 164 idèle class group, 118
connecting homomorphism, 21 idèle class invariant, 141
corestriction, 38, 39 idèle invariant, 138, 141
cup product, 44 idèle topology, 156
cyclotomic field, 127, 166 idèles, 117
inequality, first fundamental, 129
decomposition law, 168, 178 inequality, second fundamental, 89, 131
dimension shifting, 30 inertia field, 85
184 Index
inflation, 32 q-cocycles, 15
invariant map, 67, 85, 90, 138, 149
ray, 174
Kummer extension, 115, 131, 134, 161 ray class field, 165, 166, 174
ray class group, 164, 174, 176
Lie O -module, formal, 100 reciprocity isomorphism, 71
reciprocity law, 70, 138, 152, 177
Main Theorem on the Theory of
reciprocity law, Artin, 175
Algebras, Hasse’s, 145
reciprocity law, local, 93
modulus, 163, 174
resolution, free, 11
modulus of definition, 174, 175, 178
restriction, 33, 37
Nakayama map, 70
norm, 4, 6, 126 S-idèles, 118
norm group, 6, 74, 87, 96, 97, 154, 162 S-units, 114, 118
norm residue group, 17 sequence, exact, 4
norm residue symbol, 71, 86, 93, 95, 98, standard complex, 13
107, 152, 155, 178 standard resolution, 12
norm residue symbol, universal, 76, 88,
109, 155, 167 tensor product, 7, 44
Norm Theorem for Idèles, 126 trace, 4, 6
Norm Theorem, Hasse, 126, 129, 137 transfer, 38, 173
p-absolute value, 113 unit group, 80, 114

prime idèle, 168 unramified, 82, 171
primes, 113
principal idèles, 118 Verlagerung, 38, 153
Principal Ideal Theorem, 171
product formula, 113, 142, 143 ZZ-free, 10
ZZ-homomorphism, 6
q-coboundaries, 15 ZZ-module, 6
q-cochains, 14 ZZ[G]-free, 9

Neukirch 3

Uploaded by

Copyright:

Available Formats

Neukirch 3

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Neukirch 3

Uploaded by

Copyright:

Available Formats

Jürgen Neukirch

Class Field Theory

Newly edited by Alexander Schmidt

Translated from the German by F. Lemmermeyer and W. Snyder

In 1969, Jürgen Neukirch’s book Klassenkörpertheorie was published by Bib-

Heidelberg, October 2012 Alexander Schmidt

Bonn, July 1969 Jürgen Neukirch

Part I: Cohomology of Finite Groups . . . . . . . . . . . . . . . . . . . . . . . 1

Part II: Local Class Field Theory . . . . . . . . . . . . . . . . . . . . . . . . . . . . 61

Part III: Global Class Field Theory . . . . . . . . . . . . . . . . . . . . . . . . . . 111

Cohomology of Finite Groups

In the following, G will always denote a multiplicatively written finite group;

(1.1) Definition. A G-module A is an (additively written) abelian group A

Although in applications we mainly deal with multiplicatively written G-

4 Part I. Cohomology of Finite Groups

Of course, ZZ[G] considered just as an additive group is a G-module itself; it

The group ring ZZ[G] contains two distinguished ideals:

(1.2) Proposition. IG is the free abelian group on the elements σ −1, σ ∈ G,

we have the decomposition as an evidently direct sum ZZ[G] = IG ⊕ ZZ·1.

(1.3) Proposition. IG = Ann ZZ·NG and ZZ·NG = Ann IG .

6 Part I. Cohomology of Finite Groups

AG = {a ∈ A | σa = a for all σ ∈ G}, the fixed group of A,

and we can form the factor groups

If A is a G-module and g is a subgroup of G, it is clear that A is also a

The group HomG (A, B) of all G-homomorphisms from A to B is a subgroup

In addition to HomG (A, B) we have another G-module, the tensor product

(1.4) Definition. Given two abelian groups ( ZZ-modules) A and B, let F be

If A and B are two G-modules, then A ⊗ B becomes a G-module by defining

8 Part I. Cohomology of Finite Groups

In general it is not true that AG ⊗ B G is the fixed module of A ⊗ B. In fact

It is easy to verify that the functors HomG and ⊗ are additive:

(1.5) Proposition. Let {Aι | ι ∈ I} be a family of G-modules, and let X be

Moreover, if X is finitely generated as an abelian group, then

Let A and B be G-modules and

a G-homomorphism. Then h induces a G-homomorphism in the ‘opposite’

given by composition f 7→ f ◦ h (f ∈ Hom(A0 , B)), and a G-homomorphism

If we have two G-homomorphisms

In what follows, G-free G-modules will play an important role. A G-module

(1.6) Proposition. Let X be a G-free G-module and

Remark. Proposition (1.6) is valid more generally for so-called projective

10 Part I. Cohomology of Finite Groups

Proof. Let Cq = ker dq = im dq+1 . Since Cq−1 is free as a subgroup of Xq−1 ,

(1.8) Lemma. If 0 → X → Y → Z → 0 is an exact sequence of free

Proof. The exactness of the sequence X ⊗ A → Y ⊗ A → Z ⊗ A → 0

(1.9) Lemma. If 0 → A → B → C → 0 is an exact sequence of ZZ-modules

§ 2. The Definition of Cohomology Groups 11

Proof. Let X = ι Zι with Zι ∼

§ 2. The Definition of Cohomology Groups

It is a characteristic of cohomology theory that in order to give even simple

Let G be a finite group. By a complete free resolution of the group G, or

12 Part I. Cohomology of Finite Groups

(1) The Xq are free G-modules,

Thus a complete free resolution consists in fact of two exact sequences

and µ d−1 d−2

of free G-modules which are spliced together. Originally, cohomology groups

In order to determine the remaining G-homomorphisms dq it suffices, of

§ 2. The Definition of Cohomology Groups 13

d−q−1 (σ1 , . . . , σq ) = σ∈G σ −1 (σ, σ1 , . . . , σq )

+ σ∈G (−1)q+1 (σ1 , . . . , σq , σ)