Explicit Brauer Induction: Cambridge Studies in Advanced Mathematics: 40

CAMBRIDGE STUDIES IN
ADVANCED MATHEMATICS : 40
EDITORIAL BOARD
D.J.H. GARLING, T. TOM DIECK, P. WALTERS
Explicit Brauer Induction
Already published
1
W.M.L. Holcombe Algebraic automata theory
K. Petersen Ergodic theory
P.T. Johnstone Stone spaces
W.H. Schikhof Ultrametric calculus
J.-P. Kahane Some random series of functions, 2nd edition
H. Cohn Introduction to the construction of class fields
J. Lambek & P.J. Scott Introduction to higher-order categorical logic
H. Matsumura Commutative ring theory
C.B. Thomas Characteristic classes and the cohomology of finite groups
10
M. Aschbacher Finite group theory
11
J.L. Alperin Local representation theory
12
P. Koosis The logarithmic integral I
13
A. Pietsch Eigenvalues and s-numbers
14
S.J. Patterson An introduction to the theory of the Riemann zeta-function
15
H.J. Baues Algebraic homotopy
16
V.S. Varadarajan Introduction to harmonic analysis on semisimple Lie groups
17
W. Dicks & M. Dunwoody Groups acting on graphs
18
L.J. Corwin & F.P. Greenleaf Representations of nilpotent Lie groups and their
applications
19
R. Fritsch & R. Piccinini Cellular structures in topology
20 H Klingen Introductory lectures on Siegel modular forms

22
M.J. Collins Representations and characters of finite groups
24
H. Kunita Stochastic flows and stochastic differential equations
25
P. Wojtaszczyk Banach spaces for analysts
26
J.E. Gilbert & M.A.M. Murray Clifford algebras and Dirac operators in harmonic
analysis
27
A. Frohlich & M.J. Taylor Algebraic number theory
28
K. Goebel & W.A. Kirk Topics in metric fixed point theory
29
J.F. Humphreys Reflection groups and Coxeter groups
30
D.J. Benson Representations and cohomology I
31
D.J. Benson Representations and cohomology II
32
C. Allday & V. Puppe Cohomological methods in transformation groups
33
C. Soule et al Lectures on Arakelov geometry
34
A. Ambrosetti & G. Prodi A primer of nonlinear analysis
35
J. Palis & F. Takens Hyperbolicity, stability and chaos at homoclinic bifurcation
37
Y. Meyer Wavelets and operators
38
C. Weibel An introduction to homological algebra
39
W. Bruns & J. Herzog Cohen-Macaulay rings
EXPLICIT BRAUER
INDUCTION
with applications to algebra and
number theory
Victor P. Snaith
Britton Professor of Mathematics
McMaster University
AMBRIDGE
UNIVERSITY PRESS
CAMBRIDGE UNIVERSITY PRESS

Cambridge, New York, Melbourne, Madrid, Cape Town, Singapore, Sao Paulo, Delhi
Cambridge University Press

The Edinburgh Building, Cambridge C132 8RU, UK
Published in the United States of America by Cambridge University Press, New York
www.cambridge.org
Information on this title: www.cambridge.org/9780521460156
Cambridge University Press 1994
This publication is in copyright. Subject to statutory exception

and to the provisions of relevant collective licensing agreements,
no reproduction of any part may take place without the written
permission of Cambridge University Press.
First published 1994
A catalogue record for this publication is available from the British Library
ISBN 978-0-521-46015-6 hardback
Transferred to digital printing 2009
Cambridge University Press has no responsibility for the persistence or

accuracy of URLs for external or third-party Internet websites referred to in
this publication, and does not guarantee that any content on such websites is,
or will remain, accurate or appropriate. Information regarding prices, travel
timetables and other factual information given in this work are correct at
the time of first printing but Cambridge University Press does not guarantee
the accuracy of such information thereafter.
Contents
Preface
1
Representations
1.1
Basic definitions
1.2
Complex representations
1.3
GL2Fq
3.2
3.3
Weil representations
Explicit Brauer induction and Shintani descent
Exercises
The class-group of a group-ring

4.1
Adams operations and rationality
4.2
Describing the class-group by representations
4.3
Determinantal congruences
4.4
Detecting elements in the class-group
4.5
Galois properties of local determinants
4.6
Adams operations and determinants
Exercises
A class-group miscellany
5.1
Restricted determinants
5.2
The class-group of Z[Qs]
5.3
Relations between Swan modules
5.4
The class-group of a maximal order
4.7
1
1
5
21
Exercises
Induction theorems
2.1
Induction theorems of Artin and Brauer
2.2
Brauer induction in canonical rational form
Brauer induction in canonical integral form
2.3
2.4
Inductive explicit Brauer induction
2.5
Exercises
3.1
page vii
23
25
32
45
54
67
72
73
89
104
106
107
110
120
131
138
153
166
170
172
176
189
205
Contents
vi
5.5
5.6
5.7
Swan subgroups for nilpotent groups

Cyclic groups
Exercises
Complete discrete valuation fields
6.1
Ramification groups and functions
6.2
Kato's abelian conductor
The non-abelian Swan conductor
6.3
6.4
Exercises
Galois module structure
7.1
Local Chinburg invariants
7.2
The global Chinburg invariant
7.3
The Chinburg invariant modulo D(Z[G])
7.4
Real cyclotomic Galois module structure
7.5
Exercises
219
230
241
245
246
258
282
297
299
300
331
337
365
396
Bibliography
403
Index
407
Preface
`Funny that you should ask. As it happens I have a complete mathematical

vision of power. You could call it a programme. Some pieces are in place, some
theorems are proved, others await proof and revelation. It is based on the theory
of schemes:
Òf finite type?' asked Zhilin, with rapt curiosity.
Àny type, my friend, any type at all.'
`Schemes, eh.' Eli nodded his approval.'Good. Good.'
`Schemes are a little like the crania of topology with little light bulbs of algebra,
called sheaves, stuck all over them.'
`Sheaves, very good. Very agriculdural.'
from The Yukiad (Snaith, 1990c)
This volume began as a one-term advanced graduate course in algebra
which I gave at the beginning of 1990 at McMaster University. As

originally conceived my plan was to give a brief introduction to the
representation theory of finite groups in characteristic zero. This sketch
was to have been succeeded by an outline of the topological construction
of my original Explicit Brauer Induction formula (Snaith, 1988b, 1989b)
followed by a description of the behaviour of Explicit Brauer Induction
with respect to Adams operations 4.1.6 as originally proved in theorem
2.33 of Snaith (1989a). Equipped with 4.1.6 the course was then to have
concluded with a discussion of class-groups of group-rings and a proof
of M.J. Taylor's conjecture concerning determinantal congruences 4.3.10
(see also the stronger congruences of 4.3.37).
However, in 1989, I learnt of the work of Robert Boltje (1989, 1990)
which axiomatised Explicit Brauer Induction formulae and, entirely algebraically, found a different formula. This second formula was easier
to use than my original one - being a homomorphism rather than a

derivation (see 2.3.28) - and its purely algebraic derivation was far better
vii
viii
Preface
suited for exposition in a graduate course on algebra! As it happens, the

formulae of Snaith (1988b) and Boltje (1990) are related by an equation
which is to be found in 2.5.11. Indeed, for p-groups this relation may be
used to derive either formula from the other (see 2.5.16).
By the time I had concluded the course I had collected quite a number
of applications of Explicit Brauer Induction, several of which had figured
vaguely in my original motivation but had taken sufficiently long to

develop that they could not be included in Snaith (1989b). At that point
it seemed to me advantageous to have an algebraic treatment of Explicit
Brauer Induction, together with a selection of typical applications for
the benefit of those who did not like the topological proofs of Snaith
(1989b) which were apt, on occasion, to resort to stable homotopy theory
to derive algebraic results!
Having described the origins of this book, let me move on to sketch
its contents. Each chapter has a more detailed introduction to which the
reader is referred for a more complete account. In general, the topics concern examples of àpplied representation theory' in which algebraic objects
(class-groups, for example) are studied by means of finite-dimensional,
complex representations of finite groups. Under these circumstances Explicit Brauer Induction enables one to use Brauer's induction theorem
either constructively (see 6.3.6, for example) or with greater control (see
4.3.37). As a result, almost all the main results presented here are either
new (for example, 4.3.37, 6.3.6 and 6.3.20) or are proved by a new method
(for example, 4.5.39, 4.6.3 and 7.3.56).
Chapter 1 quickly covers the basic material concerning the finitedimensional representations of a finite group. We specialise almost
immediately to the case of complex representations, emphasising the
properties of induced representations.
Chapter 2 begins with an account of Brauer's canonical version of

Artin's induction theorem. Brauer's induction theorem, in its classical
existential (non-canonical) form, is proved topologically by an extension
of Snaith's (1988b) method. The formalism of Explicit Brauer Induction
is recapitulated from Snaith (1989a,b) and Boltje's axioms are stated and
shown to yield an Explicit Brauer Induction homomorphism with rational
coefficients. The difficult part of the proof is to show that integrality

of the Explicit Brauer Induction map, aG, which is accomplished by
Boltje's (1990) argument. However, for completeness, I have included
a description of the topological construction of aG which is due to
Peter Symonds (1991) and which provides those who have topological
tastes with a more conceptual, alternative point of view. As I have
Preface
ix
remarked previously, natural Explicit Brauer Induction formulae are not

unique, since Snaith's (1988b) and Boltje's (1990) formulae are generally
different. The chapter concludes by deriving a third Explicit Brauer
Induction homomorphism, dG, which is due to Robert Boltje. The map,

dG, has rational coefficients, is natural and commutes with induction.
This remarkable homomorphism was discovered during the course of the
joint work, which appears in Chapter 6, by R. Boltje, G-M. Cram and
myself on conductors in the non-separable residue field case.
Chapter 3 studies the Explicit Brauer Induction formula when applied
to an irreducible representation of the group of invertible, 2 x 2 matrices
with entries in the field with q elements. Each of the irreducible representations of GL2Fg is constructed and the leading terms of the Explicit
Brauer Induction formulae are evaluated. In addition, it is shown how
this `leading term data' enables one to construct the Shintani correspondence between Frobenius-invariant, irreducible representations of one
general linear group and the irreducible representations of the subgroup
of Frobenius-fixed matrices. Our construction, which makes no mention
of character-values or the Shintani norm map, offers an appealingly

intrinsic approach to the 'base-change' which indicates just how useful
Explicit Brauer Induction might prove to be if it were possible to extend
the technique to the case of admissible representations of the general
linear group of a local field (Gerardin & Labesse, 1979).
Chapter 4 introduces the Adams operations in the complex representation ring of a finite group (to be precise, in this special case
these operations were originally due to Burnside). It is shown how
the Explicit Brauer Induction formula enables one to write its image
under any Adams operation as a linear combination of monomial representations. Next the Hom-description is given for the class-group of
finitely generated, projective modules over the integral group-ring of a
finite group. This description presents the class-group as a quotient
of idele-valued functions on the representation ring of the group modulo global-valued functions and other special functions which are called
determinants. Martin Taylor (1978) conjectured that the Adams op-
erations, when applied to determinantal functions, would satisfy some

congruences modulo the residue degree. This determinantal congruence
conjecture is proved, strengthened and reformulated to give some new
homomorphisms whereby to detect class-groups. The chapter concludes
with new proofs of two results concerning determinants. It is shown that
the determinantal subgroup satisfies Galois descent in tamely ramified
extensions, which was one of the three mains steps in M.J. Taylor's proof
Preface
of Frohlich's conjecture (Taylor, 1981). It is also shown that Adams

operations preserve the determinantal subgroup, which was originally
proved in a different manner by Ph. Cassou-Nogues and M.J. Taylor
(Taylor, 1984).
Chapter 5 deals with six topics which are united merely by the fact
that they are related to the class-group of a group-ring. In the previous
chapter new maps out of the class-group were constructed by means of
determinantal congruences. This chapter commences with the construction of restricted determinants, which are new maps into the class-group.
The construction of the restricted determinant homomorphisms is later
refined to yield a technique for the detection of the class-group of a

maximal order in the rational group-ring of a finite group. In Chapter
7 this detection technique, which uses some new types of Hom-groups,
is applied to give a new proof of David Holland's theorem, which states
that the Frohlich-Chinburg conjecture is true in the class-group of a

maximal order. This chapter also contains a calculation of the classgroup of the integral group-ring of the quaternion group of order eight.
Every text on class-groups has its own version of this calculation; mine is
accomplished by means of a little homological algebra and a new type of
reduced norm invariant. Two sections deal with the subgroup, called the
Swan subgroup, which is generated in the class-group by the projective
modules which were so elegantly constructed by R.G. Swan (1960). By
topological techniques using groups actions on spheres new families of
relations between Swan modules are derived. These relations, together
with the algebra of determinantal congruences, are used to calculate the
Swan subgroup of some types of nilpotent groups. These calculations
prompt us to venture a conjecture as to the identity of the Swan subgroup
of any nilpotent group. The final section illustrates the non-triviality of
the class defined in the class-group by the roots of unity which reappear
in the material on real cyclotomic Galois module structure at the end of
Chapter 7.
Chapter 6 deals with the problem, raised in Serre (1960) and Kato
(1989), of constructing a Swan conductor for Galois representations of
complete, discrete valuation fields whose residue fields are inseparable.
The classical theory of the Artin and Swan conductors is recalled, together with Kato's definition of a Swan conductor for a one-dimensional
representation. The Explicit Brauer Induction formula is used to define
the required Swan conductor in general, and many properties are derived for this new conductor, which coincides with the classical Swan
conductor when the residue field extension is separable. Examples are
Preface
xi
given to show that no such generalisation of the Swan conductor can

possess the classical property of ìnductivity in dimension zero'. However,
using the rational-valued Explicit Brauer Induction homomorphism, dG,
of Chapter 2 we show that our Swan conductor is ìnductive in dimension
zero' if it is so on all p-subgroups of a particularly simple type.
Chapter 7 treats the Galois module structure of the unit group in a

ring of algebraic integers. The main conjecture in this topic is due to
A. Frohlich and T. Chinburg and concerns the identity in the class-group
of the group-ring of a Galois group of certain invariants which are

constructed from the theory of class formations. In the tamely ramified
case one of these invariants is simply the projective module furnished by
the additive group of the ring of integers in the tame Galois extension of
number fields. In general, one studies the global Chinburg invariant and
attempts to equate it with an analytically defined class due to A. Frohlich
and Ph. Cassou-Nogues. The chapter begins with the construction of the
local Chinburg invariants. These are described explicitly by means of the
theory of central simple algebras and from this description a new proof
is given of the result, due to T. Chinburg (1985), that the local invariants
vanish in the tamely ramified case. The global Chinburg invariant is
defined and the main conjecture is stated. The chapter contains a new
proof, using the method of Chapter 5, of D. Holland's (1992) result,
which states that the Frohlich-Chinburg conjecture is true in the classgroup of a maximal order. The chapter closes with some new families
of (cyclotomic) extensions for which the Frohlich-Chinburg conjecture
is true in the class-group.
Each chapter is endowed with a selection of exercises, which vary from
relatively straightforward problems, such as the completions of omitted
proofs, to research problems suggested by the topics treated in the text.
As mentioned above, parts of this book have served as the basis for
an advanced graduate course in algebra. In fact the material lends itself
to such a purpose in a number of ways and the reader may find the
following suggestions helpful in designing such courses. Twice I have
given a course which covered Chapters 1 and 2 and then concluded by
applying the results on Adams operations in Chapter 4 to derive the
determinantal congruences. Such a course was particularly successful,
since it assumes only a minimal algebraic background and culminates in
a proof of M.J. Taylor's conjecture concerning determinantal congruences
- all this being accomplished within 40 hours of lectures. A one-term

introductory course on representation theory can be made from Chapter
1, and Chapter 3 and can be given some additional mystique by assuming
xii
Preface
the existence of the Explicit Brauer Induction homomorphism, ac, in

order to analyse the Shintani correspondence in the manner of Chapter
3. Finally, several two-term courses may be designed by combining
the contents of the first two chapters with any one of Chapter 4 (and
possibly Chapter 5), Chapter 6 or Chapter 7. The last three alternatives
are sufficiently advanced that, in the course of the related exercises, they
introduce the student to a number of interesting research problems.
In the course of writing this book I have been helped by conversations,
correspondence and comments from many mathematicians. In particular,
I am very grateful to Robert Boltje for his improvements on my original
Explicit Brauer Induction formula and for his help in our joint work
with Georg-Martin Cram, which forms Chapter 6. Conversations with
Greg Hill, Charles Curtis and my student, Brian McCudden, gave rise
to the treatment of Shintani descent which appears in Chapter 2. I
am also grateful to Ted Chinburg, Ali Frohlich, David Holland and
Martin Taylor for their interest in and advice concerning Galois module
structure. In addition, it was Martin Taylor who started me thinking

about the Swan subgroup of a nilpotent group and about the problem of
Terry Wall, concerning the Swan subgroup of a product, which appears
in Chapter 5.
Victor Snaith,
McMaster University,
September 1993
Representations
Introduction
In this chapter we recall the most elementary facts about the finitedimensional representations of a finite group over a field. In our applications we shall mainly be interested in representations which are realised
over some subfield of the complex numbers.
Therefore, in Section 1, we introduce unitary, orthogonal and sym-
plectic representations in preparation for our topological treatment of

Brauer's Induction Theorem 2.1.20.
In Section 2 we specialise to the case of complex representations

and recapitulate in brief the theory of the character (or trace) of a
representation and the orthogonality relations which are satisfied by
characters of irreducible representations. Induced representations are
introduced and their standard adjointness properties (e.g. Frobenius
reciprocity, the Double Coset formula) are derived.
Section 3 consists of exercises of an introductory nature.
1.1 Basic definitions
Let G be a finite group. Let K be a field and let V be a finitedimensional vector space over K. Let GL(V) denote the group of K-linear
automorphisms of V. A homomorphism
1.1.1
p:G-iGL(V)
gives rise to an action of G on V by means of K-linear automorphisms.

Explicitly, if g e G and v E V the action is given by
1.1.2
g - v = P(g)(v)
1
Representations
A finite-dimensional K -representation of G is the K -isomorphism class
of such an action. That is, the representation given by a G-action on

V1 is equivalent to the representation given by a G-action on V2 if and
V1 - - V2 , such that
only if there is a K-linear isomorphism,
/3(g vi) = g /1(v1) for all g E G, vl E Vi. In terms of homomorphisms,
p, of 1.1.1 two homomorphisms
1.1.3
p1, p2 : G --f GL(V)
give rise to the same representation if and only if there exists B E GL(V)
such that Bpl(g)B-1 = p2(g) for all g E G. Very often we will choose an
isomorphism between V and Kn, where dimK(V) = n is the dimension
of V. In that case a representation will become a conjugacy class of
homomorphisms of the form
1.1.4
p : G - GLH(K) = GL(Kn).
Example 1.1.5
(i) One-dimensional representations Let K* = K 0 = GL1(K) denote the multiplicative group of non-zero elements of K.
A homomorphism, p : G - K*, gives rise to a unique one-dimensional

representation, since K* is abelian.
(ii) Permutation representations Let G act on a finite set, X. Form the

vector space, V, whose K-basis consists of the elements of X. Therefore G
acts on V by permuting the basis vectors. In terms of a homomorphism
into GLn(K), where n = #(X), we obtain a homomorphism, p : G -i

GLn(K), in which p(g) has only one non-zero entry in each row or
column. If we order the basis {xi E X; 1 < i < n} then the (i, j)th entry
in p(g) is 1 if g(xj) = xi and zero otherwise. Such a matrix is called a
permutation matrix.
1.1.6 If V is a G-representation then a K-subspace, W, of V which is
preserved by the action of G is called a subrepresentation. For example,
in the permutation representation of 1.1.5(ii) the subspace given by
W-
{ixi
n
i=1
yields a subrepresentation.
=o}
1.1 Basic definitions
1.1.7 Functorial operations on vector spaces induce corresponding op-
erations on G-representations. The direct sum, tensor product and the

exterior power operations are three fundamental examples.
If V1 and V2 are two vector spaces with K-linear G-actions then so is
the direct sum, V1 V2, if we define
g(vi (D v2) = gvi gv2
(g E G, vi E Vi).
In terms of matrices the direct sum of pi : G -> GL, (K) (i = 1,2) is

given by the homomorphism
P1 P2 : G -+ GLn1+n2(K)
(P1 P2)(g) =
Pi(g)
P2(g)
The tensor product, V1 V2, is the vector space HomK(BiIK(V1 x

V2, K), K) in the finite-dimensional context. Here HomK (V, W) denotes
the space of K-linear maps from V to W and Bi1K(V x W, Z) denotes

the space of K-bilinear maps. The dimension of V1 V2 is equal to
dimK(V1) dimK(V2), the product of the dimensions of V1 and V2.
Similarly we may define the tth exterior power of V, )t(V), by setting
)`(V) = HomK(Alt,(V),K),
where AItK(V) denotes the subspace of multilinear maps,
f: V x V x... x V- K (t factors),
such that
PV 1, v2, ... , v,) = (-1)sign(Q).f (v.(1), V . ( 2 ) ,- . , va(t) )
(vi (=- Vi)
for any permutation, a, of { 1,... ,t}. When char(K) # 2, the dimension of

)`(V) is given by the binomial coefficient, (t), where dimK(V) = n.
In terms of matrix homomorphisms the matrix representing V1 0 V2 is
the Kronecker product of the representing matrices, p1 and P2. When
t = dimK(V) the matrix representing )n(V) is given by det(p) : G -- K'
for p as in 1.1.4.
Theorem 1.1.8 (Maschke's theorem) Suppose that the characteristic of K is
prime to the order of G. Let V be a K-representation of G and let W be

a subrepresentation. Then there exists a subrepresentation, W1, such that
WW1=V.
Representations
Proof There exists a K-linear map, j V -> W, such that j(w) = w

for all w E W. If j(g v) = g (j(v)) for all g E G,v E V then we
could set W1 = ker(j), the kernel of j. However, j may not commute
with the G-action so we replace it by ji : V - W, defined by ji(g) =
(#(G))-i(>gEGg(j(g-i(v)))). Clearly ji(g v) = g (ji(v)) for all g E G,v E
V and if w E W then
j1(w)=(#(G))-1 E g(g-i (v)) = w,

gEG
as required.
A K-representation, V, of G is called indecomposable if an

isomorphism of the form V = W W1, for subrepresentations W and
W1, implies either that W = 0 or W = V. V is called irreducible if V
has no subrepresentations except {0} and V. Theorem 1.1.8 states that
1.1.9
these notions coincide when the order of G is prime to the characteristic

of K, which is the situation which will mainly concern us.
1.1.10 We will often be dealing with representations afforded by vector

spaces over the field, C, of complex numbers. The group, GLn(C) has a
number of compact subgroups which are of special interest. The unitary
group, U(n), is defined to be
1.1.11
U(n) = {X E GLn(C)IXX* = In},
where In is the n x n identity matrix and X* is the matrix whose (i, j)th
entry is Xji, where z denotes the complex conjugate of z E C. A unitary
representation will mean the U(n)-conjugacy class of a homomorphism
of the form
1.1.12
p : G -i U(n).
Clearly, each unitary representation gives rise to a C-representation of G

and this induces a one-one correspondence between U(n)-representations
and n-dimensional C-representations (see 1.3.1). Note that U(n) is the
subgroup of GLn(C) consisting of matrices which preserve the semi-linear
inner product on Cn given by < x, y >= E%1 xi yi, where x = (x1,. .. , xn)
and y = (y1,...,yn)
Similarly, if R is the field of real numbers then the orthogonal group,
O(n), is the subgroup of matrices which preserve the inner product
< x, y >= En i xi yi so that
1.2 Complex representations
0(n) = GL"(R) n U(n).
1.1.13
An orthogonal representation is an 0(n)-conjugacy class of a homomorphism of the form
p:G->0(n).
1.1.14
As in the complex case there is a one-one correspondence between 0(n)representations and n-dimensional R-representations of G (see 1.3.2).
Let H denote the quaternion skew-field. If z = a + ib + jc + kd is a

quaternion then z = a - ib - jc - kd. On H" we have an inner product
given by < x,y >= E 1 x; y; for x = (xl,.,,,x") and y =

The group of (left) H-semilinear automorphisms of H" which preserve
this form is called the symplectic group, Sp(n). As a C-vector space H
is isomorphic to C2 so that Sp(n) is a subgroup of U(2n). A symplectic
representation is a conjugacy class of homomorphisms of the form
p:G->Sp(n).
1.1.15
In this section we shall restrict our attention to C-representations of G.

Given a representation of the form of 1.1.4
p:G-+GL"(C),
1.2.1
we define the character of p, Xp, to be the C-valued function given by

1.2.2
XP(g) = Trace(p(g)) = En ,
Proposition 1.2.3
representation.
P(g)ii
(g E G).
(i) Xp depends only on the class of p in 1.2.1 as a
(ii) Xp(g) is the sum of the eigenvalues of p(g), counted with their multiplicities.
(iii) Xp(1) = dim(p) = n.

(iv) Xp(g-') = XP(g).
Representations
Proof If X E GLn(C) set = XpX-1 then we must verify that xp = Xp.

However, if t is an indeterminate,
det(tIn - 0(g)) = det(X(tIn - P(g))X-1)

= det(tIn - P(g))
= to - Trace(p(g))tn-1 +...,
which proves part (i).
Part (ii) follows by conjugating p(g) into its Jordan canonical form,
which is an upper triangular matrix whose diagonal entries are the

eigenvalues. Each eigenvalue, A, appears on the diagonal mA times, where
m2 is the multiplicity of A.
Part (iii) is clear, since Trace(In) = n and part (iv) follows from part
(ii), since the eigenvalues of p(g-1) =
are the inverses of those for
p(g). However, since p(g) has finite order, these eigenvalues are complex
p(g)-1
numbers of unit norm and therefore the inverse of each eigenvalue is

equal to its complex conjugate.
Proposition 1.2.4 Let pi : G -+ GL(Vi) (i = 1, 2) be two representations
with character functions, Xi = Xp,. Then
(i) xl + X2 is the character of V1 V2 and
(ii) (XI) ' (x2) is the character of V1 V2.
Proof For g E G let pl(g) = X and P2(9) = Y then (PI p2)(g) is given
by the matrix
X0)
C
YJ
whose trace is clearly equal to Trace(X) + Trace(Y).

The tensor product, p1(g) p2(g), is given by the Kronecker product
of X and Y, by 1.3.3. This is the matrix of the form
X *Y11
X'Y1n
X'Y12
XY21 XY22 ... XY2n

X ' Ynl
X ' Yn2
...
X ' Ynn
where dimC(p2) = n. The trace of this matrix is clearly equal to

Trace(X) [,y11 + Y22 +... +Ynn]
Lemma 1.2.5 (Schur's Lemma) Let V1 and V2 be irreducible representations of G. Let HomG(Vl, V2) =
If : V1 ---) V2 linear I f(gvi) = gf(vi); g E G,v1 E V1}.

Then
(i) if V1
V2, then HomG(V1, V2) = 0,
(ii) if V1 = V2,
HomG(V1, V1) ={ If I f (x) = A- x for some 2 E C}.
Proof If f : V1 -) V2 is non-zero then ker(f) and im(f) are subrep0, ker(f) * V1 and
0 so that ker(f) = 0 and im(f) = V2, which means that f is
resentations of V1 and V2 respectively. Since f

im(f)
an isomorphism. This means that V1 = V2 as representations. However,

when V1 =
let 2 be an eigenvalue for f. The subspace
W ={v1 E V1
is a non-zero subrepresentation of V1 so that V1 = W, as required.

Corollary 1.2.6 Let V1,V2 be as in 1.2.5. Let f : V1 - V2 be a linear
map. Define F : V1 - ) V2 by
F(vi)=#(G)'
(g.f(g'.vi)
gEG
Then
(i)F=O
if VI * V2
and
(ii) F(v1) = (dimVi)-' Trace(f) vl
if V1 = V2.
Proof Clearly F(g v1) = g F(v1) for all g E G so that F = 0 unless

V1 = V2. If V1 = V2 then F(vi) = A - v1 for all v1 E V1 and, by taking
traces,
(dimVi) 2 _ #(G)-' E Trace(g f (g-' -)) = Trace(f)

gEG
since Trace(g f (g-1 -)) = Trace(f) for all g E G.

Definition 1.2.7 Suppose that W1, W2 are two representations with characters X1, X2 respectively. The Schur inner product <
W1, W2 > =
Representations
< Xl, X2 > is defined by
(x1()).
< X1, X2 >= #(G)-1 E

gEG
Theorem 1.2.8
In 1.2.7
(i) < X1, X2 >= dimCHomG(W1, W2),

(11) < X1, X2 >=< X2, X1 >,
(iii) < X1, X, >= 1 if and only if W1 is irreducible.
Proof Suppose that &._, As, B1,. .. , B, are irreducible representations

such that
Wi = V,=jAi, W2 =
ijHomG(Ai, Bj). Now let XA;, XB, denote the

characters of these irreducible representations. By 1.2.4, X1 =
1 XA;
and X2 = E'=1 XB, so that
then HomG(Wl, W2)
<X1,X2>=
<XA;,XB,>.
ij
Therefore, in order to prove part (i), we may assume that W1 and W2

are irreducible.
Choose bases for W1 and W2 so that the representation, Wi, corresponds to pi : G --) GL,,; (C). In Corollary 1.2.6 assume that f is
represented by a matrix, X. In this notation 1.2.6 becomes
1.2.9
#(G)-1 Ea,bgEG P2(g)iaXabPl(g)bjl
0
(dim(W1))-1
if i#jor Wi*W2
Trace(X)
if W1 = W2 and i = j.
When W1 # W2 choose f with Xab = 0 except for Xij = 1 so that 1.2.9

implies that
>P2(g)iipl(g-l)jj = 0 for all

gEG
i, j.
Therefore
#(G) < X1, X2 > =
X1(9) ' X2(9)
Ei Ei P1(9)ii ' P2(g-1)ii, by 1.2.3(iv),

=
P1(g-1)iiP2(g)il
= Eii
=0
= dimCHomG(W1, W2).
When W1 = W2 we may again take Xab = 0 except for Xii = 1 to

obtain
10
if i j,
P2(g)ii ' P1(g,i) = t #(G) (dimW1)-1 if i = j.
gEG
1
Therefore
#(G)' < X1, X2 > _ Ei,i EgEG P1(g-1)ii ' P2(g)ii
_ #(G)
and
1 =< Xl, X2 >=< X1, X1 >,
as required.
To prove part (ii) we remark that < X2, X1 > is the complex conjugate
of < X1, X2 >, by definition, but < X1, X2 > is a positive integer and
therefore is fixed by complex conjugation.
Part (iii) follows from part (i), 1.2.5 and the observation that
s
HomG(Ai,Ai) = s.
dimCHomG(W1, W1)
i=1
Corollary 1.2.10 In the notation of 1.2.7, W1 = W2 as representations if

and only if the characters X, and X2 are equal.
Proof Firstly, observe that

dimC W1 = X1(1) = X2(1) = dimCW2,
so that we may prove this by induction on dimension. When dimC W1 = 1

both representations are irreducible and
dimCHomG(W1, W2) = dimCHomG(W1, W1) = 1,
Representations
10
by 1.2.8(i) so that, by 1.2.5, there is a G-isomorphism between W1 and

W2. Now let Al be an irreducible summand of W1 and set
W2 = j=1 Bj,
as in the proof of 1.2.8. Hence

dimCHomG(A1, W2) = dimCHomG(A1, W1) > 1,
by 1.2.8(i). Therefore, for some j, HomG(A1, Bj) # 0. By 1.1.8, W2
A1 WZ and W1 = Al W. The characters of Wi and WW are therefore

equal, by 1.2.4, so that Wi = WW as representations, which completes the
proof.
If W and V are representations of G such that V is

irreducible then < W, V > is equal to the multiplicity of V in W (i.e.
the number of times that V appears in a decomposition of W as a sum of
Corollary 1.2.11
irreducibles).
Proof If W = nV V1 ... Vt with V1 irreducible and not equivalent

to V then
<W,V> =<V,W>
= dimC(HomG(V, W)
= ndimC(HomG(V, V) + E;=1 dimC(HomG(V, Vi))
= n,
by 1.2.5
Definition 1.2.12 The regular representation of G is the permutation representation obtained, by the method of 1.1.5(ii), from the set G together
with the G-action given by left multiplication. The regular representation will be denoted by IndG,,,(I) in recognition of its construction as an
induced representation (see 1.2.31).
Proposition 1.2.13
tion. Then
Let rG denote the character of the regular representa-
if g1,
rG(g)=( 0
l #(G) if g = 1.
Proof By definition (see 1.1.4(ii)) a basis for Ind{1}(1) consists of {x I x E
G}. If g
1 then g x * x so that the trace of multiplication by g is
11
trivial. When g = 1,
rG(l) = dimC(Ind{1}(1)) = #(G).
Let V be an irreducible representation of G with character, Xv, and dimension equal to nv, then the multiplicity of V in rG is
equal to nv.
Corollary 1.2.14
Proof By 1.2.7,
< Xv,rG > = #(G)-1(EgEG Xv(g)rG(g))
= #(G)-1Xv(1)#(G),
by 1.2.13,
= nv,
as required.
Corollary 1.2.15 Let VI, V2,..., Vt denote the distinct irreducible representations of G then (i)
dimC(V,)2,
#(G) _
,=1
and
(ii) if g # 1 then
0=
dimC(V;) Xv,(g)
,=1
Proof By 1.2.14 the regular representation is equal to

8i_1dimC(V;)
V;
so that part (i) follows by taking dimensions. Part (ii) follows by taking
characters at 1 # g E G, using 1.2.13.
Example 1.2.16 Corollary 1.2.15 may be used in finding all the irreducible
representations of G. For example, let D2n denote the dihedral group of
order 2n given by
1.2.17
Den = {X, Y I X" = Y2 = 1, YXY = X-1}.
We may define one-dimensional representations
x1,x2 : Den -> {1} c C'
Representations
12
(x2 will be non-trivial only when n is even) by the formulae

1.2.18
x;(XYI) = { (-
11)6
if i = 1,
if i=2.
We may define a homomorphism

1.2.19
vi
: D2n -) GL2(C)
by
where n = exp(2ni/n).
and otherwise the characters of the {vi} are distinct
Clearly xv, =
so that we obtain two-dimensional representations vi, V2'...' v. where m
is the largest integer smaller than (n/2) - 1. For these values of i, v, is
irreducible, by 1.2.8(iii), since
< vi, vi >
(2n)-1
EgED2. Ixv,(g)12
j=1 ij + Z7'1)2
= 1 + n-1(E`
j=1 n )
1,
as required. These are all the irreducible representations of D2r,. When n =

2t, we have four one-dimensional representations, (t-1) two-dimensional
ones and
When n = 2t + 1 we have two one-dimensional representations, t twodimensional ones and
2.12+1 22=4t+2=2n.
Definition 1.2.20 The (complex) representation ring of G, denoted by
R(G), is defined to be the free abelian group on the irreducible, complex
representations of G. The ring structure on R(G) is induced from the

tensor product in the following manner. In the free abelian group on the
irreducible representations, { V;}, we may identify the formal sum,
13
x = E ai Vi,
with the formal difference of representations
x=W+-WW+ = aipositiveaiVi,
W- = ainegative(-ai)Vi
If y = U+ - U_ then we may define
1.2.21 xy=[(W+U+) (W_U-)l-[(W+ U_)(W_(& U+)J

The product defined by 1.2.21 is well-defined and makes R(G) into a
commutative ring.
1.2.22 Conjugacy class functions
Let
denote the set of complex-valued functions defined on the con-
jugacy classes of G. Hence f E W, if and only if f : G -+ C satisfies

f (hgh-1) = f (g) for all g, h E G. Clearly Ws is a subspace of the complex
vector space of all functions on G. If p : G -> GL(V) is a representation then its character, Xp, is an element of Wg. Hence, by sending an
irreducible representation to its character and extending C-linearly, we
obtain a map of C-vector spaces
1.2.23
CG : R(G) Z C-)
Theorem 1.2.24 (i) In 1.2.23 cG is an isomorphism.

(ii) The number of distinct irreducible, complex representations of G is
equal to the number of conjugacy classes of elements of G.
Proof Part (i) implies part (ii) immediately, since the rank of R(G) equals
the number of distinct irreducible representations of G while a basis for
is given by the characteristic functions of the set of conjugacy classes
of G - that is, the functions which are equal to one on one conjugacy
class and zero on all the others.
To prove part (i) let X1, X2, ... , Xt denote the characters of the distinct
irreducible representations of G. Define a semi-linear inner product on
W g by the formula,
< f1,f2 >= #(G)-' Tf1(g).f2(g)

gEG
Representations
14
By 1.2.5/1.2.8(i) the {xj} are orthonormal with respect to this inner

product and hence they are linearly independent in Wse
To prove that the {xj} span W5 we must verify that, if f E

< f, xi >= 0 for all i then f = 0.
and
Let pi : G --+ GL(V1) be the irreducible representation whose character

is Xi. The map
1.2.25
Fi = 2gEG f(g)Pi(g) : Vi
Vi
satisfies
Fi(hv)
= EgEG f (g)P1(g)P1(h)v
= EgEG f
(g)Pi(h)Pi(h-'gh)v
= Pi(h)(>gEG f
= h(Fi(v))
(h-igh)pi(h-'gh)v)
for all v E Vi and h E H. Hence, by 1.2.5,

1.2.26
Fi(v) = (dimC(Vi))-1(EgEG f (g)xi(g)) . v.
If < f, xi >= 0 for all i then < f, x, >= 0 also, since complex conjugation
induces a permutation of the irreducible characters of G. Hence, for each
i, Fi = 0 since
Fi =
dimC(Vj)-1
< f, X, >,
by 1.2.26. Now, temporarily, write p for the regular representation so

that
p=
dimC(Vi)pj.
Let V denote the vector space whose basis is {g E G}. Define F : V - V

by replacing pi by p in 1.2.25. Clearly
On the other hand, if eg E V is the basis vector corresponding to g E G

then, by definition of the regular presentation,
F(ei) = 1: f(g) . g(el) = >f(g) - eg,

gEG
gEG
so that f (g) = 0 for each g E G, as required.

Example 1.2.27 Let Den denote the dihedral group of 1.2.16.
15
When n = 2t there are t+3 distinct, irreducible representations. Clearly

x` is conjugate to x-i and xiy, x iy and x`+2y are conjugate. Therefore,
by Theorem 1.2.24, the conjugacy classes of Dot are represented by
1, x, x2, ... , x`, y
and
xy.
When n = 2t + 1 there are t + 2 distinct, irreducible representations

and the conjugacy class representatives are
,x,x2 ,...,x and
ly.
Theorem 1.2.28 Let G be a finite group, then G is an abelian group if

and only if all the irreducible representations of G are one-dimensional.
Proof If G is abelian then there are #(G) distinct conjugacy classes in

G. By Theorem 1.2.24 the number of distinct, irreducible representations
is also #(G), of which each must be one-dimensional in order to satisfy
1.2.15(i).
Conversely, if each irreducible representation is one-dimensional then,

by 1.2.15(i), there are #(G) of them and therefore, by 1.2.24, the number
of conjugacy classes is equal to #(G), too. This means that no two distinct
elements of G are conjugate. Thus the set
{ghg-1 I g E G}
consists merely of h and hence hg = gh for all g, h E G.

Corollary 1.2.29 Let H be an abelian subgroup of G. If V is an irreducible
representation of G then dimC(V) < #(G)/#(H).
Proof Consider V as an H-representation, by restricting the action from
G to H. By 1.2.28 there exists an H-subrepresentation, W < V,with

dimC(W) = 1. Hence the subspace
U=1: g(W)=
gEG
g(W)
gEG/H
has dimension less than or equal to #(G)/#(H). However, g(U) 9 U for

all g E G so that U = V, since V is irreducible.
Theorem 1.2.30 (i) Let Vi be an irreducible representation of Gi (i = 1, 2)
then Vl V2 is an irreducible representation of G1 x G2.
(ii) Each irreducible representation of G1 x G2 is of the form given in (i).
Representations
16
Proof Let Xi denote the character of V1 then
<X1x2,x1x2>
#(G1)-'#(G2)-' E(g,,$z)EG,XG2 xl(g1)x2(g2)x1(gl)x2(g2),
by 1.2.4,
= < x1, x1 >< x2, x2 >
= 1, by 1.2.8(iii)
so that V1 V2 is irreducible, by 1.2.8(iii) again. This proves part (i).
By 1.2.24, we need only verify that if V; is another irreducible representation of G then Vj' VV is distinct from V1 V2 unless V1 = Vl and
V2 = V2. However, this follows at once from the relation
<V1 0V2iVl VZ>_<Vl,Vi ><V2,VZ>

together with 1.2.5-1.2.8.
Definition 1.2.31 Induced representations Let p : H --- GL(W) be a

representation of H, where W is a K-vector space. The group ring, K [H],
is defined to be the K-vector space on a basis {h E H}. An arbitrary

element of K [H] has the form x = >hEH khh, where each kh E K. In this
case, K [H] becomes a ring when endowed with the following sum and
product :
1.2.32
(E khh) + (E qhh) = >(kh + qh)h,
khh) . (
qgg)
= >(khgg)hg
g,h
In this case, W may be viewed as a (left) KH-module by means of the

product (w e W)
(> khh) (w) =
khp(h)(w)
Now suppose that H is a subgroup of G. The group-ring, K [G], is a

left K [G]-module and a right K [H]-module so that the tensor product,
K [G] K [H] W, is a left K [G]-module or, equivalently, a representation.
This representation is called the representation of G induced from p and
is denoted by I nd' (p). If w1, ... , wt is a basis for W then K [G] K [H] W
may be constructed by first forming the K-vector space whose basis

consists of symbols {g 0 w;; g E G,1 < i < t} and then dividing out by
the subspace generated by elements of the form

1.2.33
17
gh wi - Ej-1 p(h)jig wj
where h E H, g E G and p(h)(wi) = Ej pjiwj. Equivalently, K [G] K [HI W
may be formed by taking one copy of W - denoted by g 0 W - for

each g E G and dividing 1:gEG(g (& W) by the subspace generated by
elements of the form
1.2.34
gh w - g 0 p(h)(w)
(h E H).
The left G-action on KG K [H] W is given in these terms by the formula

1.2.35
(EgEG kgg)(g' (9 w) = EgEG gg' kgw
Clearly,
dimK(IndH(p)) = [G : H]dimK(p).
Example 1.2.36 (i) Let p = 1 denote the one-dimensional trivial repre-
sentation of the trivial group, H = { 11. The regular representation of

1.2.13 is the induced representation, Ind{1}(l).
(ii) Let D2 denote the dihedral group of 1.2.27:
IX"=1= Y2,YXY =X-1}.

Set H =< X > and define 4 : H - C* by 4)(X) = exp(2iri/n). The
representation
vi : Den
> GL2C
is equal to I ndH "(i). This is seen as follows. Let e E C* then a basis for
I ndH2" (0i) is given by 1e and Y e. The action of X and Y is given by
X(Y(D e)=XYe=Y(YXY)e=Y4(X-1)e=Y0
'e,
Y(1e)=Ye
and
Y(Ye)=Y2e=1e.
This coincides with the formulae of 1.2.27.
Definition 1.2.37 If H < G and P : G -> GL(V) is a representation then

we may consider V as an H-representation. This representation is the
restriction of p to H and will be denoted by ResH(p).
18
Representations
Proposition 1.2.38 Let H be a subgroup of G. Let W, V be K-representations of H, G respectively. There are isomorphisms of the following form:
(i) HomG(IndG(W), V) ?' HomH(W,ResH(V)),
(ii) IndH(Hom(W,K)) = Hom(IndH(W),K) as G-representations and

(iii) HomG(V,IndG(W)) = HomH(Re$(V), W).
P r o o f Let 91, ... , gd be a set of coset representatives for G/H. Therefore,
1(g; (D W ). Suppose that $1 = 1,
as a vector space, I ndH(W) is
then we may assign to a G-map, f : Ed 1(gi W) - V, the H-map,
f : 1 W = W -* V. This is an H-map since
f(h(1 (& w))
This map is injective, since f (gi w) = g,(f (10 w)), so that the map is
an isomorphism, by counting dimensions. This proves part (i).
To define a G-isomorphism in (ii) it suffices, by (i), to define an H-map
F : Hom(W,K) -+ Hom(IndH(W),K).
Define F(f)(gi w) = f (w) if gi = 1 and zero otherwise. Clearly F(f) = 0
if and only if f = 0, so that we obtain an injective G-map
IndGHom(W,K) -> Hom(IndH(W),K),

which must, by dimensions, be an isomorphism.
Let V denote the contragredient of V, given by Hom(V,K), as in 1.3.5.
Part (iii) is deduced from (i)-(ii) in the following manner:
HomG(V,IndG(W)) = HomG(Hom(IndH(W),K), V)
HomG(I ndH(W ), V )
HomH(W,ResH(V))
HomH(V, W).
Theorem 1.2.39 (Frobenius reciprocity) Let H < G and let W, V be Krepresentations of H, G respectively. There is a G-isomorphism
IndH(W 0 ResH(V)) = IndH(W) V.
19
Proof Define
rp
by 4)(gi
since
: IndH(W 0 Re$(V)) -> IndH(W) V
0 v) = (gi 0 w) 0 giv (w E W,v E V). This is well-defined

4)(g1h-1
0 hw 0 hv) = (gih-1 0 hw) (&gih-1hv
=g,0wgiv
= 0(gi 0 w 0 v).
Define
4-Ygi w) 0 v) = gi
gi
These are clearly inverse isomorphisms and 0 is a G-map since

g(4)(gi 0 w 0 v))
= g((gi 0 w) 0 giv)
= (ggi (9 w) 0 ggiv
= 4)(ggi 0 w (D v).
Theorem 1.2.40
Let J, H < G and let W be a representation of H, then

IndjnzHz-l((z-1)*(W))
zEJ\G/H
where (z-1)'(W) is W with the (J fl zHz-1)-action given by
(zhz-1)(w) = h w.
Proof Let $1, ... , gd be coset representatives, as in the proof of 1.2.38.

Hence
In4(W)_ 1(g,W),
the sum of one copy of W for each element of G/H. Since g e G sends
gi W to ggi 0 W, d 1(gi 0 W) breaks up into J-subrepresentations,
one for each J-orbit in G/H. Hence, as a J-representation, IndH(W)

decomposes into a sum of J-subspaces - one for each element of J\G/H.
The subspace corresponding to gi 0 W is given by Ei = >jEJ jgi W.
Now jgi W = gi 0 W if and only if gi 1 jgi E H, in which case

j E J fl g,Hgi 1. If j1,..., j are coset representatives of J/(J n g,Hg, 1)
then
u
Ei = (jsgi (D W).
S=1
Representations
20
We have to verify that the J-action on E; corresponds to

IndJng.Hg-i((g1 T(W))
However, this follows since, for j E J, j(jsgi (D w) = j jsgi w and if

j E J n giHgi 1 then
jsgi (gi ljgi)(w) = jsgigi
ljgi
w = jsjgi 0 w.
Corollary 1.2.41 (Mackey's irreducibility criterion) Let H < G and let W

be a C-representation of H. Then IndH(W) is irreducible if and only if
(i) W is irreducible and
(ii) for each g e G - H,
< W,(g-1)*ResHngHg_1(W) >= 0.
Proof By 1.2.8(iii), IndH(W) is irreducible if and only if

= 1. However, by 1.2.8(i) and 1.2.40
< IndH(W),IndH(W) > = dimCHomG(IndH(W),In4(W))

= dimCHomH(W,ResHIndH(W))
LzEH\G/H < W, Res H
HnzHz-i (W) >
- '`
One of the terms in this sum is < W, W >, which is at least one. Hence
IndH(W) is irreducible if and only if < W, W >= 1 and all the other
terms are zero.
Proposition 1.2.42
If J < H < G then

I ndHI ndH (U) = I ndf (U).
Proof An isomorphism is given by sending g h u to gh 0 u (g E G, h E
H, u E U). This map is evidently a G-map. It is injective, as is seen by

expressing it explicitly in terms of chosen coset representatives for G/H
and H/J, but both representations have the same dimension.
Theorem 1.2.43 Let p : H -> GL(V) be a complex representation with
character, XP. If H < G then the character of W = IndH(p) is given by
XP(Ygy_1).
X'(g)
yEG,ygy-' EH
1.3 Exercises
21
P r o o f W e may write I ndH (V) as 1(gi (D V) where gl, ... , gs is a set of
coset representatives for G/H. Since multiplication by g E G sends gi V

to ggi V the only summands on which the trace of multiplication by g
can contribute a non-zero amount are those for which ggi V = gi 0 V.
This can only happen when gI 7 1ggi E H. In this case, if v E V,
g(gi v) = gi P(gi 1ggi)v
so that we obtain
XW(g) = T,
Xv(gi 1ggi)
i,9; 'ggtEH
X(h_1g` 'ggih)
= X(g1 lggi) if h E H so that we may allow the

sum to run over all y E G such that ygy-1 E H provided that we divide
by #(H). This completes the proof of Theorem 1.2.43.
However,
1.3 Exercises
(i) Let G be a finite group and let p : G --> GL (C) be a homomorphism. Show that there exists X E GL (C) such that X p(G)X-' < U(n).
(ii) Given homomorphisms, pi : G -> U(n) (i = 1, 2 ), which are
conjugate in GL (C), show that they are conjugate in U(n). (Hint: For
(i), by averaging, construct an inner product on C" which is preserved by
p(g) for each g E G. For (ii), let Z' denote the conjugate transpose of
Z. If Xp1X-1 = P2 then XX* is Hermitian. Find a suitable Z such that
Z2 = X"X and set U = XZ-1.)
1.3.1
1.3.2 Prove the analogous result in which GL (C) and U(n) are replaced
by GL (R) and O(n).
1.3.3 Show that the tensor product, V1 V2, may be represented in terms
of matrices by the Kronecker product (see 1.2.4).
Let V be a C-representation of G. Let Sym2(V) be the representation afforded by the vector subspace of V 0 V consisting of elements
which are fixed by the involution that interchanges the two factors (cf.
1.3.4
1.1.6).
(i) Show that 22(V) is given by the (-1)-eigenspace of the involution.

(ii) Show that V V = .12(V) Sym2(V).
Representations
22
(iii) If X is the character of V, show that

XSyM2(v)(g) = 1/2[(X(g))2 + X(g2)]
and that
Xa2(v)(g) = 1/2[(X(g))2 - X(g2)]
Let p : G -> GL(V) be a representation. The contragredient of p

is the representation, P, afforded by W = HomK (V, K), where the action
is given by the formula
1.3.5
P(g)(f)(v) = f (P(g-1)v),
where gEG,vE V,f E W.

(i) Find the matrix of P in terms of that of p.
(ii) If K = C and p is unitary, show that Xp = xp.
Let 1 denote the one-dimensional representation given by C with
the trivial G-action. Show that the multiplicity of 1 in p is equal to
1.3.6
#(G)-1 E XP(g)
gEG
where Xp is the character of p.
Let {V1;1 < i < s} be a set of distinct irreducible representations

of G and set dimC(V,) = n;.
1.3.7
(i) If Ei=1 n' = #(G), prove that {V,} consists of all the distinct
irreducible representations of G.
(ii) Apply part (i) to find all the irreducible representations of the
generalised quaternion group
Q4,, = {X, Y I X" = Y 2, Y 4 = 1, Y X Y -1 = X-11.
1.3.8
Verify, in 1.2.20, that R(G) is a ring.
For complex representations, give proofs of Frobenius reciprocity

and the Double Coset Formula of 1.2.40 by means of characters.
1.3.9
Induction theorems
Introduction
The chapter concerns induction theorems; that is, theorems which express
arbitrary representations as linear combinations of induced representations within the representation ring, R(G), tensored with a suitable ring
of coefficients.
We begin the chapter with a proof of Brauer's canonical form for

Artin's Induction Theorem. Artin's result, in this form, gives a canonical rational form for the identity representation within R(G) Q, the
rationalised representation ring, in terms of representations which are
induced from cyclic groups. By Frobenius reciprocity Artin's theorem
yields a similar rational canonical form for any representation. Brauer's
Induction Theorem, in its original (non-canonical) form states that any
representation can be expressed as an integral linear combination of
representations which are obtained by induction from one-dimensional
representations of elementary subgroups. This celebrated result has been
proved by a variety of methods, to which we add a new topological
proof here. This proof involves replacing the given representation of
G by the unique, equivalent, unitary representation and using the latter
to perform some elementary algebraic topology on the resulting action
of G on the compact manifold given by the unitary group modulo the
normaliser of its maximal torus. This topological proof of the existence
part of Brauer's theorem was the basis of the first derivation of a canonical form for Brauer's theorem, which appeared in Snaith (1988b). This
topologically derived canonical form possessed two important properties:
namely, of naturality and of being the identity on one-dimensional representations. These two properties, together with additivity, were taken
23
24
Induction theorems
as axioms in Robert Boltje's algebraic construction of a canonical form

for Brauer's theorem.
In Section 2 we introduce the free abelian group, R+(G), on conjugacy
classes of one-dimensional subhomomorphisms. This is the group within
which the canonical form lies. There is a natural map, bG, to the representation ring, and an Explicit Brauer Induction homomorphism is simply
a section for bG, which satisfies the two axioms. After developing the
formal properties of R+(G) we follow Boltje's analysis of the coefficients
which must appear in an Explicit Brauer Induction homomorphism. We
prove that these coefficients, if they exist, are uniquely determined rational numbers, that many of them are forced to vanish and, finally, that
they must be integral. All this is accomplished by means of the Boltje's
(1989) algebraic arguments.
In Section 3 we show that one may inflict a non-degenerate bilinear
form upon R+(G) in such a manner that the required rational-valued
Explicit Brauer Induction homomorphism is simply the adjoint of the
natural map from R+(G) to R(G). In addition, the algebraic derivation
which we have followed yields Boltje's explicit formula, aG, in terms of
Schur inner products and Mobius functions of the partially ordered set
of one-dimensional subhomomorphisms of G. In the canonical form for
Artin's induction theorem one finds Mobius functions from the partially
ordered set of cyclic subgroups playing a similar role. The section
closes with a number of examples together with an explanation of the
relationship between Boltje's formula and Snaith's (1988b) topological
formula. In fact, as explained in 2.5.16, Snaith's (1988b) construction
can be made to yield aG when G is a p-group. The integrality of the
coefficients in aG is the deepest property in the algebraic analysis and may
be obtained, in general, by a topological method due to Peter Symonds.
Symonds' construction, which uses a group action in a manner similar
to the method of Snaith (1988b) (this time the action is on the projective
space of the representation), is also described in Section 3.
In Section 4 another rational-valued Explicit Brauer Induction homomorphism, dG, is constructed. In the course of the construction of aG
one sees the manner in which an idempotent may be used, together with
Mobius inversion, to construct different types of such homomorphisms.
This one, which was discovered by Robert Boltje, uses an idempotent on
R(G) which was first exploited by Andreas Dress. The homomorphism
dG is shown to be natural with respect to inclusions of subgroups and
to have the remarkable property that it commutes with induction. Many
properties of dG are derived: for example, the fact that dG commutes with
2.1 Induction theorems of Artin and Brauer
25
taking fixed-points and the relationship between dG and ac. Most of these
properties are established for further use in Chapter 6. However, in order
that dG should have the convenient inductive property it is necessary that
its coefficients be non-integral and that it cannot be natural with respect
to the inflation maps which are induced by surjective homomorphisms
of groups.
Section 5 consists of a batch of exercises concerning the Explicit Brauer
Induction homomorphisms.

In this chapter we will be concerned with the development of the induction theorem of Brauer in a canonical form. This was first accomplished
by topological methods in Snaith (1988b). However, in the later sections
of this chapter, we will present an improved approach to Explicit Brauer
Induction, which is due to R. Boltje (1990). I would have liked to incorporate further topological improvements, due to P. Symonds (1991)
which facilitate the verification of the integrality of Boltje's canonical
form. However, I have had to content myself with a few remarks at the
end of the chapter (see 2.3.28), because the introduction of the requisite
topological background would take us too far afield.
In this section, however, we will prove the induction theorems of Artin
and Brauer as preliminary illustrations of the various approaches. Firstly
we will consider Artin's induction theorem and the manner in which it
was given in a canonical form by R. Brauer (1951). Boltje's formula,
involving as it does Mobius functions, is reminiscent of Brauer's formula.
Secondly we will prove Brauer's Induction Theorem by the use of a

minute amount of the algebraic topology which constitutes the basis of
the construction in Snaith (1988b).
Let us begin by recalling the Mobius function.
Definition 2.1.1
(n) _
The Mobius function, (n), is defined by

1
if n=1
(-1)' if n = pi ... Pr, pi distinct primes

0
if n is not square-free.
Recall that the Mobius function is extremely useful for inverting certain
formulae (see 2.5.2). Later we will require the generalisation of this wellknown fact to the Mobius functions of a poset (a partially ordered set).
Induction theorems
26
Lemma 2.1.2 Let p : G -- * GL(V) be a complex representation whose

character, Xp, is rational-valued. Suppose that x, y E G generate the same
cyclic subgroup of G, then Xp(x) = Xp(y).
Proof Let m be the order of the cyclic subgroup generated by x or y

and suppose that xi = y where HCF(j, m) = 1. By 1.2.3(ii), if Al' ... , A.S
are the eigenvalues of p(x) then Xp(x) = E 1 2i and Xp(y) = Es=1 '1;.
However, there exists a Galois automorphism, a, of
which
sends Sm = exp(21ri/m) to gym. Since each 2i is a power of cm we find that
u(Xp(x)) = Xp(y) but Xp(x) is rational-valued so that a(Xp(x)) = Xp(x),

which completes the proof.
Theorem 2.1.3 (Actin induction theorem) Let p : G -* GL(V) be a complex representation with rational-valued character, Xp. In R(G) Q
Xp = >acIndc(1),
where C ranges over cyclic subgroups of G and where
ac = [G :
C]-1 E([B : C])x (b)

C<B
The sum is taken over all cyclic subgroups, C < B and b E B is any
generator.
Proof By 2.1.2 the coefficients, ac, are independent of the choice of

generators, b E B. Let Xc denote the character of IndG(1). By 1.2.42
Xc(g) = #{y E G I
ygy-1
E C} #(C)-1.
Hence, if Xc(g) # 0, there exists ygy-1 E C and Xc(g) = Xc(ygy-1) so

that we may assume that g E C. In this case
Xc(g) = #(NG < g >)#(C)-1= #(NG <
ygy-1
>)#(C)-1,
where NGH denotes the normaliser of H.

Hence
2.1.4
Eccyclic acxc(g) =
#(NG < g
Ccyclic,some ygy 'EC
>)#(G)-1
([B : C])xp(b)
C is
equal to #(G)#(NG < g >)-1. Furthermore, an arbitrary cyclic subgroup
27
can contain at most one subgroup which is isomorphic to < g >.

Therefore we may rewrite 2.1.4 as a sum over cyclic groups which contain
< g >:
>Ccyclic,gEC >CdI [B:<g>] p(d))
= Xp(b),
where =< g >, since Edl, p(d) = 0 unless n = 1 (Hunter, 1964).
By 2.1.2, Xp(b) = Xp(g) so that we have shown that EacInd'(1) has the
same character-value as p on g E G, which completes the proof of 2.1.3.
A complex representation, p, of G is called a monomial

representation if p = I ndH(0) for some 4) : H -+ C*. An M-group is a
group all of whose irreducible complex representations are monomial. A
group, G, is solvable if there exists a chain of subgroups
Definition 2.1.5
with G,_1 < Gi and Gi/Gi_1 abelian. A solvable group is supersolvable if,
in addition, the quotients Gi/Gi_1 are cyclic and the G,_1 are normal in
G, not merely in Gi. A supersolvable group is nilpotent if, in addition,

each Gi/Gi_1 is central in G/Gi_1.
As a representation, p : G -> GL C, is equal to the sum

of monomial representations if and only if there exists X E GL C such
that Xp(G)X-1 < NT" where NT" is the normaliser of the subgroup of
diagonal matrices in the unitary group, U(n).
Lemma 2.1.6
Proof Firstly let us determine the matrix form of a monomial representation, I ndH(0). We note that the direct sum of A E NT" and B E NT'
is A B E NT"+m (cf. 1.1.7).
Let 91, ... , gd be coset representatives for G/H. If g E G write

2.1.7
ggi = ga(g)(j)h(i, g),
where h(i, g) E H and a : G -> Ed is a homomorphism to the symmetric

group. A basis for In4(4)) consists of vectors of the form {gj 1; 1 <
i< d} with G-action given by
2.1.8
g(gi
= go(g)(i) 0(h(i, g)) = 4)(h(i, g))(ga(g)(i) (9 1).
Induction theorems
28
Hence, if we consider a(g) as a permutation matrix, then, with respect to

this basis,
2.1.9
4)(h(1,g))
4)(h(2, g))
P(g) = 6(g)
...
However, NTd < GLdC is generated by the diagonal matrices with nonzero entries of unit length and by permutation matrices. Hence 2.1.9 is a
matrix lying in NTd.
Conversely, suppose that we are given a representation in the matrix
form
p:G-+NT".
Let it : NTn -+ En denote the quotient map whose kernel is the diagonal, maximal torus of U(n). Consider the action of G on { 1, 2, ... , n} via
np. The matrix form of p breaks into the direct sum of homomorphisms,
pi : G --> NT Si, where Ei-1 si = n and {1,2,...,n} consists of t G-orbits

of sizes s1,. .. , s, We may therefore assume that t = 1, which means that
G acts transitively on {1,2,. .. , n}. Let H = {g E G I n p(g)(1) = 11 and
let 0 : H -+ C' denote the homomorphism given by the (1, 1)-entry of
ResG(p). Set W = In4(4)).
We will now verify that the characters of W and p are equal. Denote
these characters by Xw and XP, respectively. Since p(g) E NTn we have
Tracep(g) =
UnP(g)(m)=m p(g)m,m
= Egze=Zx P(z- 1 gz)1,1
where H =stabRP(1)
= XW(g),
since 4)(z-1gz) =
P(z-igz)ii
This completes the proof of 2.1.6.
Let p : G --> U(n) be a unitary representation. Let G

act on the left of U(n)/NT" by g(zNT") = p(g)zNTn. Then p is equal to
Corollary 2.1.10
a sum of monomial matrices if and only if there is a G -fixed point.
Proof If, for all g E G, p(g)zNTn = zNTn then z-lpz maps G to

NT". Therefore z-1 pz, which is equivalent to p , is equal to a sum
of monomial representations, by 2.1.6. Conversely, if p is the sum of
monomial representations it may be represented by p : G -> NT",
which fixes the identity coset.
29
2.1.11 Topological digression
Whenever a finite group, G, acts smoothly upon a compact manifold,

X (as in 2.1.10, for example), it is possible to triangulate X so that
it becomes a simplicial complex in such a manner that the G-action
either fixes a simplex pointwise or translates it by a simplicial map to a
different simplex. Triangulated thus, X is the union of a finite number of
simplices; the boundary of each simplex being the union of simplices of
lower dimension (Maunder, 1970). From this triangulation one obtains
the simplicial chain groups, C1(X), given by the free abelian group on the
i-dimensional simplices of X. As a module, C,(X) is a sum of permutation
modules for the ring Z[G], since G acts simplicially on X. The simplicial
chain complex of X is a sequence of Z[G]-maps
2.1.12
...-->Ci(X)-+C;_1(X)->...-C0(X)->0.
We may tensor this sequence with the complex numbers to obtain a sequence of complex representations each of which is a sum of permutation
representations
2.1.13
Proposition 2.1.14 Set X = U(n)/NT" upon which G acts as in 2.1.10.

Then X may be triangulated so that:
(i) In 2.1.13, Ci(X) C is the sum of permutation representations,

IndH( )(1), where H(a') runs through the stabilisers of orbit representatives of the set of i-dimensional simplices.
(ii) In 2.1.13 the sequence is exact (kerbi = imbi+l) except at Co(X)C
where
Co(X)C
=IER(G),
the trivial, one-dimensional representation.
Proof Choose an i-simplex, a'. In C1(X) C it is clear from the

discussion of 2.1.11 that the C[G]-module generated by a' is isomorphic
to IndH(Q;) (1). This proves part (i), since Ci(X)C is the direct sum of such
C[G]-modules. Part (ii) follows from the fact that the homology with
complex coefficients, H.(U(n)/NT"; C), is isomorphic to the complex
homology of a point (Snaith, 1989b, p. 208, section 2.4).
Induction theorems
30
Proposition 2.1.15
is an M-group.
A p-group is an M-group. In fact, any nilpotent group
Proof Let p be a prime and let G be a p-group. Let p : G -* U(n) be an

irreducible, unitary representation (recall from 1.3.1 that every complex
representation is equivalent to a unique unitary representation). By 2.1.10
we must show that G acts on U(n)/NT" with a fixed point. Suppose not,
then each C;(X) C is the direct sum of permutation representations of
the form IndH(,;)(1) with H(a') * G. Hence p divides each dimension,

dimC(C;(X) 0 C) . However, this contradicts 2.1.14(ii).
If G is an arbitrary nilpotent group then G is the product of its Sylow
subgroups,
G=G1x...xGs.
An irreducible representation of G has the form I ndG, (01)...I ndH/(0s),
by 1.2.29. Each 4, has dimension one and, by 2.5.5, this representation

equals
IndH1X...XH,(P1
An elementary group is a finite group of the form

C x P where P is a p-group and C is a cyclic group of order prime to p.
Definition 2.1.16
Proposition 2.1.17
H < G, such that
Let G be any finite group. There exist M-groups,

1=
nIndG (1) E R(G)
for suitable integers, {n}.
Proof If G is an M-group there is nothing to prove. If G is not an

M-group we may choose an irreducible representation
p : G - U(n),
so that the resulting action of G on U(n)/NT", in 2.1.14, has no fixed
point, by 2.1.10. Hence, in 2.1.14(i),
C,(X)C=IndH(Q;)(1)C,
with H(r') * G. From 2.1.14(ii) and 1.1.8 one easily shows that (see 2.5.6)
1 = j:(-1)`C;(X) C E R(G).
31
Hence there exist integers, mg, and subgroups, Jp # G, such that

2.1.18
1 = p mfInd G(1) E R(G).
By induction, for each Jp which is not an M-group,

2.1.19
1 = y mp,yIndZ (1) E R(Jp),
with Lp,y an M-group. The result follows by substituting 2.1.19 into

2.1.18, where necessary, and using transitivity of induction (1.2.42).
Theorem 2.1.20 (Brauer induction theorem) Let G be a finite group. Given
x e R(G) there exist one-dimensional representations of elementary subgroups, Hi, 4i : Hi
C" and integers, ni, such that
x=
n;I ndG (4i) E R(G).
Proof If we multiply x by the expression in 2.1.17 we obtain, in R(G),

x
naxIn4 (1)
nI n4 (Re$ (x)),
by 1.2.39. Therefore, by 1.2.42, we may assume that G is an M-group.
Let BG denote the Z-linear span in R(G) of representations of the

form 1n4(4) with 0 one-dimensional and H an elementary subgroup.
Since a subgroup of an elementary group is again elementary, Frobenius
reciprocity (1.2.39) shows that BG is an ideal of R(G).
Suppose that p is an irreducible representation of the M-group, G. If
p = I ndG (0) with H # G then, by induction on the order of G, we see
that p E BG. Let p denote the contragredient of p (1.3.5) and suppose
that A is a one-dimensional representation of G. We have
<A,pP>=<Ap,p>
J1, ifA=1,
0,
if not.
This is because p is irreducible and because the character of A p is

(g' --) A(g)xp(g)) which is not equal to (g -- Xp(g)) unless A = 1. Hence,
if dimC(p) >_ 2,
pop= 1+Ep1ER(G),
Induction theorems
32
where dimC(p;) >_ 2 for each i. Since p and the {p;} belong to the ideal,
BG, then so does 1. However, if 1 E BG a R(G) then BG = R(G), as
required.
There remains the case in which all the irreducible representations of
G are one-dimensional, so that G is abelian, by 1.2.28.
Suppose that G = G1 x ... x G, where G, is an abelian group of order
pu' and where pt,... , p, are distinct primes. By 1.2.30 there is a ring
isomorphism
R(G) = R(Gi) ... 0 R(G,).
2.1.21
Also, by the Artin Induction Theorem (2.1.3)
pu E BG; i R(G;)
By 2.5.5, if M = p71 ... pr' = #(G), then each of M/(pu') E BG for

i = 1, 2,..., r. Hence
1 = HCF(M/(pu')) E BG 1 R(G),
and so BG = R(G), as required.
2.2 Brauer induction in canonical rational form

We must begin with some definitions.
Definition 2.2.1
Let G be a finite group. Let R+(G) denote the free
abelian group on G-conjugacy classes of characters, 0 : H --> C", where
H < G. We shall denote this character by (H, 0) and its G-conjugacy

class by (H, O)G E R+(G).
If J < G we define a restriction homomorphism

2.2.2
ResG
j
: R+(G) -+ R+ (J)
by the double coset formula

2.2.3
Resj G ((H, )G) = EZEJ\G/H(J n zHz-1, (z-t)'(O))J,
where (z-1)'(O)(u) = cb(z-tuz) E C" .

If it : J -> G is a surjection then we define

2.2.4
33
ir' : R+(G) -+ R+ (J)
by lr*((H, 0)') = (i-'(H), 0m)'

By means of 2.2.3-2.2.4 we may define f R+(G) -* R+(J) for any
f :J -> G by factorising f as f : J -> im(f) c G and setting
2.2.5
f ` = 1t'Res m(f) : R+(G) -' R+(im(f )) -> R+ (J).
One may also define an induction map, I ndjG : R+(J) -* R+ (G),

and a product which makes R+(G) into a ring-valued functor satisfying
Frobenius reciprocity (see 2.5.7).
Define a homomorphism, which is surjective by 2.1.20,
2.2.6
bG : R+(G) -> R(G)
by
bG((H, 4)G) = I ndH(4)
This is a ring homomorphism for the ring structure of 2.5.7.

Remark 2.2.7 The objective of this section is to construct a natural
homomorphism
aG : R(G) -+ R+ (G)
such that bGaG = 1. This amounts to a canonical form for Brauer

induction.
The first canonical form was given in Snaith (1988b). The method
was based upon the group action of G on U(n)/NT" which was used
in 2.1.6 and 2.1.10. The details, together with several applications, are
given at length in Snaith (1989b) and further elaborated upon in Snaith
(1989a). The topological procedure of Snaith (1988b) automatically gives
a functorial association of an element of R+(G) to a representation.

Furthermore, the simplicity of the group action in the one-dimensional
case ensures that a one-dimensional representation, 0 : G ---+ C*, is
associated with (G, 4)G E R+(G). In this section we will follow the method,
due to R. Boltje (1990), which starts by taking these two properties as

axioms and deduces the formula for aG algebraically from these axioms.
Induction theorems
34
2.2.8 Axioms for aG
(i) For H < G the following diagram commutes:
R(G) a R+(G)
ResH
R(H)
R+(H)
(ii) Let p : G -> GLn(C) be a representation and suppose that

aG(P) _ E a(H,,p)(H, 4)G E R+(G),
then a(G,q5)G =< p, 0 > for each (H, O)G such that H = G.
Let G be any finite group and denote by #G the set of characters

on subgroups, (H, 0), where H < G and 0 : H --> C. JIG is a poset (a
2.2.9
partially ordered set) if we define the partial ordering by

(H, 0) << (H', 0')
2.2.10
if and only if
H < H' and ResH'(0')
In addition, G acts onG by the formula

2.2.11
g(H,4') = (gHg-',(g-')'(4'))
(g E G),
where (g-')'(4')(u) = 4'(g-lug).

We may define a partial order on the orbit space, (JIG)/G, by
(H, 4')G <_ (H', 4'')G
2.2.12
if and only if there exists g E G such that

(H, 0) <_ g(H', 0') =
Define a positive integer
(gH'g-1, (g-')* (4'))
35
Y(H,4),(H',O') E Z+ by the formula

2.2.13
Y(H,O),(H',') = #{g E H\G/H' I (H, 0) < g(H', O')}.
Lemma 2.2.14
(i) y(H,O),(H',O') depends only on the conjugacy classes,
(H, 0)G and (H', c')G.
(ii) Y(H,cb),(H',') > 0 if and only if (H, 4p)G < (H', 0')G.
Proof Part (ii) is clear from the definition. For part (i), suppose that
u, v e G then Yu(H,,fi),v(H',0') =
#{g E uHu-'\G/vH'v-1 I (uHu-',(u-')'(O))

<_
(gvHv-'g-', ((gv)-')'(cb'))}.
However, uHu 1 < gvH'v-1g-1 and 0'((gv)-'agv) = 4(u-1 au) for all
a E uHu ' if and only if H < u 'gvH'v-'g-'u and 0'(v-1 g-'ubu 'gv)
= 4(b) for all b E B. Hence sending g E uHu '\G/vH'v-' to u-1gv E
H\G/H' shows that yu(H,O),v(H'#) = y(H,O),(H'#), as required.
Theorem 2.2.15 There is at most one homomorphism, aG, satisfying the

axioms of 2.2.8. In fact, if
aG(P) = E
a(H,,)Q(P)(H,
(p)G
E R+(G),
(H,)GG_yW1q
then the coefficients, a(H,O)(p) are inductively determined by the equations

2.2.16
a(H,O)H (Reset (p))
The equations of 2.2.16 have at most one solution in the rational numbers.
Proof First consider 2.2.16 and suppose that a(H',,P')6(p) E Q is determined

for all H' > H then all the terms in the right side of 2.2.16 are determined
except for the term
G
Y(H,0),(H,4)a(H,cb) (P).
However, y(H ,) (H.) _ [NG(H, 0) : H], where
NG(H,.O) _ {g E G I (ghg-1) = 0(h) for all h E H}.
Induction theorems
36
Hence y(H) (H O) * 0 (this also follows from 2.2.14(ii)) and, since

a(H,O)H(ResH(p)) =< ResH(p),4) >, we see that a(H,O)G(p) is inductively
uniquely determined by 2.2.16. It remains to derive 2.2.16.
By definition, the (H, 4)H-term in aH(Re$(p)) = Re$(aG(p)) is given
by
a(H',O')G(p)(H fl
(H',4')G
zHz-1, (z-1)'(0'))H
H=HnzH:-1
z EH\G/H',(z-1 )' (45')=j
EH\G/H',(H,c)-z(H'>(k') a(H',c')(p)(H, (P)

Y(H,4),(H',O')11(H',O')G(p)](H, 4)),
which establishes 2.2.16.
Remark 2.2.17
We will construct the homomorphism, aG, by first
constructing a map,
aG : R(G) -, R+(G) Q,
which satisfies 2.2.8(i)-(ii) and then showing that the unique such map
must land in R+(G). Therefore we will temporarily (for the remainder of
this section) assume that we have constructed the map into R(G) Q.
We will require some preparatory results before we can show the

integrality of aG(p) E R+(G) Q.
Suppose that p : G -- GL(V) is an irreducible, complex representation. Let H Q G be a normal subgroup and let W c ResH(V) be
an irreducible H-subrepresentation. For each g E G/H the subspace,

Wg = p(g)(W) c V, is well-defined. Also, Wg is an H-subrepresentation
since, forhEHandwEW,
p(h)p(g)(w) = p(hg)(w) = p(g)(p(S 'hg)(w)) E Wg.
Wg is called a G-conjugate of W.
Lemma 2.2.18
In the situation of 2.2.17

V = 9E4 Wg,
for some subset, A c G/H.
Proof Clearly
reducible, so that
Wg is a G-subrepresentation of V, which is irWg = V. Also, since W is an irreducible

H-representation then so is Wg for each g.
37
Therefore, as an H-representation, V must be the direct sum of a

subset of the { Wg ; g E G/H}, which completes the proof.
Lemma 2.2.19 Let H be a subgroup of G and let : H -+ C` be a
one-dimensional representation. If p is a representation of G then
< 0, Resy(p) >= 0 implies that a(H,4,)G(p) = 0.
Proof When H = G this is guaranteed by the axiom 2.2.8(ii). We will

prove the result by downward induction on #(H). We have
< 0, ResH(p) >
_ {coefficient of (H,(p)H in aH(ReSG(p))}
2.2.20
_ > (HMG<_ie',#'iG Y(H,0),(H',')a(H',cb')G(p),
by 2.2.16.
If H < G then < 4), ResH(p) >= 0 implies that < ', ResH,(p) >= 0
for all (H, 0) < (H', 0') and so, by induction, a(H',,')G(p) = 0 in this case.
Hence 2.2.20 becomes
G (P)
0 =< , ResH(P) >= Y(H,0),(H,q5)a(H,)G(p),
so that a(H,O)G(p) = 0, as required.

Definition 2.2.21 Let p : G -* GL(V) be a complex representation and
let (H, 0) E
2.2.22
G. Define a subspace, F(H, 0) < V, by
F(H, ) = {v E V I p(h)(v) = 4)(h)v for all h E H}.
Conversely, given a subspace, 0
W < V, define P(W) E JIG by
P(W)
2.2.23
=sup{(H,) E .iLIG I p(h)(w) = rp(h)wfor all h E H, w E W}.

Lemma 2.2.24
In 2.2.23, P(W) is well-defined.
Proof We have to show that the supremum exists in 2.2.23.

Firstly we observe that ({1}, 1) satisfies the condition of 2.2.23.
Secondly, if (H, 0) and (K, p) satisfy the condition of 2.2.23, then

ResH
HIK (4)) = ResH( (V), since W * 0. Therefore we can extend 0 and w
uniquely to the group, U =< H,K >, generated by H and K. If we call

this extension p : U -p C' then (U, p) satisfies the condition of 2.2.23.
Induction theorems
38
Hence P(W) exists and is equal to the character on the group generated
by all the H's on the right of 2.2.23 given by the unique extension of all
the 4s on the right of 2.2.23.
Definition 2.2.25 We call (H, 0) E JIG admissible for V (or p) if F(H, 0) *
0 or, equivalently, if < 0,Res$(p) >* 0. Denote by A(V) (or A(p)) the
set of elements of G which are admissible for V. Denote by S(V) the
set of non-zero subspaces of V. Therefore we have maps
2.2.26
F : A(V) - S(V) and P : S(V) --> A(V).
(i) Let (H, ), (K, tp) E A(V) and W, W' E S(V).

(a) If (H', ') < (H, 0) then (H', 0') E A(V).
(b) If g E G then g(H, ) E A(V) where g(H, 0) is as in 2.2.11.
(ii) F(g(H, 0)) = p(g)(F(H, )), P(p(g)W) = g(P(W )) for all g E G.
Proposition 2.2.27
(iii) (K, ip) < (H, ) implies that F(K, ip) >_ F(H, 0) and W < W'
implies that P(W') < P(W).
(iv) (H, 0) < PF(H, 0) and W < FP(W ).
(v) F(H,0) = FPF(H,0) and P(W) = PFP(W).
Proof If K < H and ResK() _
then < 0, ResH(p) >* 0 implies that

< p, ResK (p) ># 0, which proves part (i)(a). Also,
<
Re
> _< (g-')"(O), Re gHg-, ((8-' )'p) >

=< 0, ResH(p) >,
since (g-1)'(p) = p, which proves part (i)(b).

For part (ii) we note that
F(g(H,0)) = {v e V I (g-'zg)(v) = p(z)(v) for all z E gHg-1}
so that v e F(g(H,4)) if and only if p(ghg-1)(v) = 0(h)v, which is

equivalent to p(g)-1(v) E F(H, 4 ). This proves the first part of (ii).
Similarly, (gHg-1, (g-1)'(O)) satisfies p(ghg-')(w) = 0(h)w for all w E W
if and only if (H,0) satisfies p(g-1)(w') _ (P(h)w' for all w' E p(g-1)(W).
The second part of (ii) follows by taking suprema over these sets, as in
2.2.23.
If K < H and ip = ResK (0) then the condition of belonging to F(H, 0)

is stronger than that of belonging to F(K, ip), which proves the first part
of (iii). The second part of (iii) is similar.
Since (H, 0) is one of the objects over which the supremum is taken
when forming PF(H, 0) the first part of (iv) is clear. Also, for every
39
(K, gyp) in the set over which the supremum is taken in 2.2.23, every
w E W satisfies p(k)(w) = ip(k)w for all k E K so that w E FP(W).
Finally, by (iv), (H, O) < PF(H, O) and W < FP(W) so that, by (iii),
F(H,4) >_ FPF(H,4)) and P(W) > PFP(W), which proves part (v),
since, by (iv), F(H,0) < FP{F(H,4)} and P(W) < PF{P(W)}.
Definition 2.2.28 For (H, 0) E A(V) and W E S(V) define their closures,
cl(H,) and cl(W), respectively, by
cl(H, 0) = PF(H, 0) and cl(W) = FP(W).

An object will be called closed if it equals its own closure.
Let (H, 0), (K, p) E A(V) and W, W' E S(V).

(i) cl(H,0) and cl(W) are closed.
(ii) F(cl(H, )) = F(H, ) and P(cl(W )) = P(W ).
Proposition 2.2.29
(iii) If (K, V) < (H, 0) then cl(K, tp) < cl(H, 0) and if W < W' then
cl(W) < cl(W').
(iv) The following are equivalent:
(a) (H,0) (respectively W) is closed.

(b) g(H, ) (respectively p(g)W) is closed.
(c) (H, ) is in the image of P (respectively, W is in the image of F).
(v) F and P are inverse bijections between the closed pairs in A(V) and
the closed subspaces in S(V).
(vi) inf{P(W),P(W')} = P(W+W'). In particular the infimum of closed
pairs is closed.
Proof Parts (i)-(v) follow at once from 2.2.27. To prove part (vi) we
observe that (H, 0) < P(W) and (H, 0) < P(W') if and only if H acts
on both W and W' by 0 and this is equivalent to H acting on W + W'
by 0. Therefore P(W + W') is clearly the largest pair, (H, 0), such that
(H, 0) < P(W) and (H, 4)) < P(W'), which is, by definition, the infimum
of (vi).
Lemma 2.2.30
closed then
In the notation of 2.2.13, if (K, p) < (H, 4) and (H, 0) is

G
Y(K ,w),(H,c1) -
Induction theorems
40
Proof Set cl(K,W) = (U, u) then (U, u) < (H, 0), by 2.2.29(iii) since
(H, 0) is closed. As in the proof of 2.2.15,
ResU(H, O)o = Y(U,),(H,cb)(U, )U + 7(U n
zHz-1 (z-1)'(0))U,
where the sum is taken over z E U\G/H such that

(U n zHz-1, (z-1)"(4))U # (U, )U.
Applying ResK and using ResK = ResKResU we obtain
2.2.31
ResK (H, q5)G
= Y(U,k),(H,#)ResK(U,,u)U +
ResKu (U n
zHz-1, (z-
U.
Since ResK
(U, )U = (K, W)K it is sufficient to verify that, in the last sum
1,
of 2.2.3 no pair (K,W)K appears. For this it suffices to verify for all z
in the sum that

(K,W)U
(U n zHz-1, (z-1)'(4)))'.
Therefore we will prove that

f (K,W)U <_ (U n zHz-1, (z-1)*(4)))U implies
2.2.32
(U n zHz-1, (z-1)'((P))' = (U, )U.
If the condition of 2.2.32 holds then there exists u E U such that

(K, W) < (U n uzHz-lu 1, ((uz)-1)'()).
2.2.33
From 2.2.33 we deduce that
(U,) = cl(K,W) < cl(U n uzHz-lu 1, ((uz)'1)*(O)).
2.2.34
Setting (H', 0') equal to the right side of 2.2.33 we obtain

ResgnuzH(uz)-l (.u) _
which implies that

(K, W) < (U n uzHz-iu 1,((uz)-1)*(4))) <_ (U,) = cl(K,W)
Therefore
2.2.35
(U,,u) = cl(U n uzHz-1 u 1,((uz)-1)'(4))).
41
However,
(U n uzHz-'u ', (uz-')'(4))) = inf ((U, p), (uzH(uz)-1, ((uz)-')'(O)))

and both (U, p) and uz(H, 0) are closed, by 2.2.29(i) and 2.2.27(ii), so that
(U n uzHz-'u ',((uz)-')'(c))) is also closed, by 2.2.29(vi). Hence 2.2.32

follows from 2.2.35, which completes the proof.
Proposition 2.2.36 Let p : G -) GL(V) be a representation and let
(H, ) E .'#G. In 2.2.15, if a(H,O)H(p)
0 then (11,0) is admissible and
closed for V.
In particular, if there exists (H', 4))
(H, 0) in 11G with
< 0', ResH,(p) >=< , ResH(p) >

then a(H,O)G(p) = 0.
Proof By 2.2.19, a(H,O)G(p) # 0 implies that < ,Re$(p) ># 0; hence

(H, ) E A(V). We will show that (H, 0) is closed by descending induction
on the order of H.
If H = G then (G, 0) is closed because it is maximal in A(V).
Now assume that H G and set (U, W) = cl(H, 0). Now consider the
coefficient of (H, 0) in aH(Re$(p)) and of (U, W) in au(ResU(p)). These
are equal to
<4), ResH(P) > = Y(H,0),(H,m)a(H,m)c(P)
2.2.37
+>(H,m)G<(H',A')G
in.,liG/G
and
< W, ResU(p) >

2.2.38
in.,/l/c
G
Y(U,w),(U',W')a(U,,r')H(P)
By induction the sums in 2.2.37 and 2.2.38 may only be taken over
admissible pairs.
If (H, 0) is not closed and cl(H, ) = (U, W) the closed pairs which are
strictly greater than (H, 0) are the closed pairs which are greater than or
equal to (U, W). Therefore, by 2.2.30, the two sums are equal in 2.2.37
Induction theorems
42
and 2.2.38. However, by 2.2.29(ii),
< W, Resv(p) > = dimCF(U,W)

= dimCF(cl(H, ))
= dimCF(H, 0)
_< 0, ResH(p) >,
so that, subtracting 2.2.38 from 2.2.37, we obtain
G
0=
- Y(H,0),(H,O)a(H,O)(p)
and therefore a(H,,#)o(p) = 0, as required.

For the final statement there is nothing to prove if < 0, ResH(p) >= 0,
by 2.2.19. If not then
dimCF(H', ') =< 0', ResH,(P) >

=< 4),ResH(p) >
= dimCF(H, 0),
so that F(H, 4)) = F(H', 4)). Hence cl(H', 0') = cl(H, 0) and therefore
(H, 0) is not closed, since (H, 4))
(H', 4)') < cl(H', 4)').
Let p : G -) GL(V) be a representation. The

centre of p, Z(p), is defined to be the largest subgroup, H, such that
ResH(p) = dimC(p)W for some W : H - C.
Definition 2.2.39
Corollary 2.2.40 (i) Suppose that ResZ(P)(p) = dimC(p)W, then for all
(H, 0) E ./&G, in 2.2.15,
a(H,O)G(p) # 0 implies that (Z(p),W) < (H, 4)).
(ii) For all x E R(G)

a(H,O)G(x) # 0 implies that Z(G) < H.
Proof If a(H,O)G(p) # 0 then (H, 0) is admissible and closed, by 2.2.36.

Also F(Z(p),W) = V >_ F(H,0) so that
(H, 0) = cl(H, 0) = PF(H, 0) >- PF(Z(p), iV) >- (Z (p), iV)
This proves part (i).

For part (ii) we may assume that x is irreducible so that Z(G) < Z(x)
and the result follows from part (i).

Proposition 2.2.41
43
Suppose that the homomorphism of 2.2.15,
aG:R(G)---o R+(G)Q,
satisfies the axioms of 2.2.8. Then, for all X E R(G),
2.2.42
E(H,M)E.d1G/G a(H,cb)G(X)IndH(1) = dimC(X) E R(G) Q.
Proof Let rlG R+(G) --> R+(G) be the natural homomorphism defined
by nG(H, Of = (H,1)G. The left side of 2.2.42 is the image of X under
R(G) ---. R(G) Q.
This map is natural for restrictions to subgroups so that, in order to

evaluate the character of this expression at g r= G, it suffices to verify the
result for the cyclic group, < g >. However, for G cyclic the result is
immediate from axiom 2.2.8(ii).
Theorem 2.2.43
In 2.2.15,
aG(p) E R+(G) c: R+(G) Q for all p E R(G).
Proof Assume for all H G and for all X E R(H) that all(X) E R+(H).
This inductive statement is certainly true for all abelian groups, H. In
order to verify that aG(p) E R+(G) it suffices to show this when p is
irreducible.
Therefore suppose that p is irreducible and that aG(p) V R+(G). Also

suppose that (H, 4) E .fG is maximal such that a(H,O)G(p) V Z. Under
these conditions we will first show that H < G. Let U = NG(H, 0), the
normaliser of (H, 0) in G.
Consider the coefficient of (H, 0)U in Resg(aG(p)) = au(Resg(p)).
This is given by
"(H,4) (Resv(P))
= E(H',/')GC,#G (H'44')(P)
zEUvG/H'
(U n
zH'z-,
(z-')'(4"))U
H=UnzH'z-1,4=(z-' )' (0')
= xa(H,cb)(P) + E (H,O)#(H' O')G (H,)

since only (H', 4") > (H, 4') can contribute a multiple of (H, 4,)U to
the above expression. Also, by maximality of (H, 0), the sum of such
coefficients in the above expression is an integer. On the other hand,
x = #{z E U\G/H I zHz-' = H,(z-1)*(4') = 01 and taking z = 1 we see
Induction theorems
44
that x > 1; but since
zHz-1
= H and (z-1)*(cb) = 0 if and only if z e U
we find that x = 1. Since a(H,,)U(Resv(p)) E Z if U

U = G, because
X ' a(H,cb)G(P) = a(H,O)G(P)
G we find that
Z.
Therefore H < G and (z-1)*() = 0 for all z E G.

Next we will show that H = Z(G), the centre of G.
Since H < G and p is irreducible, we may apply 2.2.18 to assert that
ResH(P) = gEo Wg
for some A c G/H.

<
Since a(H,,P)G(p)
0 we have, by 2.2.19, that
ResH(p) >* 0 and therefore 0 is one of the Wg's. However, if
W = 4) in 2.2.17-2.2.18 then each Wg = also, because is G-invariant.

Hence
Res'(p) = (dimC(p))4) E R(H).
2.2.44
This means that Ker(4)) < Ker(p) and that Ker(b) < G, since 0 is
G-invariant. Therefore there are factorisations of the following form:
p : G -p G/(Ker(4))) --> GL(V)

2.2.45
: H - H/(Ker(4))) -> C.
since Ker(4)) < H.
By 2.5.9,
2.2.46
a(H,cb)G (P) = 0((H1Ke.(0),v)G1Ke.(0)(P)
By minimality of G we must have Ker(4)) = {1} so that 4) : H -> C' is Ginvariant and injective. Hence, if h E H and g E G then 4)(ghg-1) = 4)(h)
and therefore ghg-1 = h so that H < Z(G), as required.
Let Z(p) denote the centre of p, in the sense of 2.2.39 and suppose
0, by
2.2.40(i). On the other hand, (H, 4)) = (Z(G), 4)) < (Z(p), ip). Therefore,
by maximality of (H, 4)), a(H,O)G is the only non-integral coefficient in
aG(p). By 2.2.41, in R(G) Q we have
that Resz(p)(p) = (dimC(p))1p then (Z(p), ip) < (H', (P') if a(H',(P,)G
dimC(P)
2.2.47
=a(H,O)G(P)IndH(1)+
E
(H,cb)G*(H',c')Gin.,l(G/G
06(H'#)GIndH,(1).
2.3 Brauer induction in canonical integral form
45
Since Frobenius reciprocity implies that = 1, we obtain

dimC(p) = a(H,cb)G(P) +
a(H, ,)G E Z,
(H,)G(H',4')Gind1G/G
so that 0((H,0)G E Z, which is a contradiction.
In this section we will complete the construction of the homomorphism

aG : R(G) -> R+ (G),
2.3.1
which satisfies the axioms of 2.2.8(i)-(ii). We will begin by recapitulating

the results of the previous section, which culminates in 2.2.43.
Proposition 2.3.2 Let aG : R(G) --> R+(G) Q be a homomorphism
which satisfies the following two conditions:
(i) For all H < G the following diagram commutes:
R(G)
ac'
--
R+(G) Q
ResH
R(H)
aH
--
R+ (H) Q
(ii) Let p : G -> GLn(C) be a representation and suppose that

aG(P) = E 0t(H,4)G(H, 4)G E R+(G) Q
then a(G,,fi)G =< p, 0 > for each (H, O)G such that H = G.
Then aG(R(G)) c R+(G) and aG satisfies the axioms of 2.2.8(i)-(ii). In
addition, if bG is the homomorphism of 2.2.6 then (see 2.2.7)
bGaG = 1 : R(G) -> R(G).
Proof Only the last statement requires to be proved. However, this is

easy since a natural endomorphism of R(G), such as bGaG, which is the
identity when G is cyclic, must be the identity for all G.
Definition 2.3.3 Let us define a bilinear form
Induction theorems
46
2.3.4
(R+(G) 0 Q) x (R+(G) Q) - Q
by means of the formula

2.3.5
[(H, 4)G, (H', /1)G] = y(H,0),(H'#)'
By 2.2.14(i) this is a well-defined bilinear form, which is not symmetric.
In fact, if we take the poset {(H, 4)G E (.IZG)/G} as a basis then the
matrix representing 2.3.4 with respect to this basis is ùpper triangular'
with respect to the partial ordering. Therefore 2.3.4 is a non-singular
bilinear form.
Let J be a subgroup of G.
If x E R+(J) and y E R+(G) then
Proposition 2.3.6
[I ndG(x), y] = [x, Rest (y)],
where IndjG((U,W)J) = (U,,p)G E R+(G).
Proof It will suffice to take x = (U, ip)J and y = (H, lp)G. In this case
[I nd (x), y]
= [(U, W)G, (H, 4)G]
= #{g E U\G/H I (U, W) < g(H,

On the other hand,
[x, ResjG(y)]
= EzEJvG/H [(U, W )J, (J n
zHz-1 (z-' )* (4))J]
_ >zEJ\G/H #{w E U\J/J n (zHz-1)
(U, w)< w(J n zHz-1, (z-')'(O))}.
To see that these expressions are equal we first observe that w(J n
zHz-1, (z-1)* (4)) = (J n wzH(wz)-1, ((wz)-')*((p)) and that, since U< J,
(JngHg-1,(g-1)'(4))
(U, w) < (gHg-1,(g-1)'(4)) if and only if (U, W) <
Therefore the two expressions which we wish to prove equal are both
sums over the same type of object. The result now follows from the fact
that there is a bijection
{(z, w) E (J\G/H) x (U\J/(J n zHz-1)} F--, {g E U\G/H}

given by sending (z, w) to wz. This bijection is derived from enumerating
the J-orbits and then counting the decomposition of each J-orbit into
is the J-stabiliser of zH in G/H.
U-orbits, using the fact that J n
zHz-'
47
Definition 2.3.7 Since [-, -] is a non-singular bilinear form we may

define a homomorphism
aG : R(G)
as the adjoint of bG : R+(G) -> R(G) of 2.2.6. That is, for all x E
R+(G)Q,pER(G)Q,
[x, aG(P)] =< bG(x), P > .
2.3.8
Theorem 2.3.9
There exists a homomorphism
aG : R(G) -' R+(G)

which satisfies the axioms of 2.2.8(i)-(ii).
Also, if bG is the map of 2.2.6 then bGaG = 1.
Proof It suffices to verify, for aG of 2.3.7, the two conditions of 2.3.2.

For condition (i), let p E R(G) and suppose that H < G is a subgroup.
For all x E R+(H) we have
[x, ResH(aG(P))]
= [I ndH(x), aG(P)],
=< bG(I ndH(x)), p >,
=
=< bH(x),ResH(p) >,
_ [x,aH(Re$(p))],
by 2.3.6
by 2.3.8
by 1.2.8,1.2.38
by 2.3.8 again
so that ResH(aG(p)) = aH(ReSG(p)) by the non-singularity of 2.3.4.
Now let 0 : G --> C' be a one-dimensional representation and let

(H, W)G EG. Therefore, by 1.2.38,
< bG(H' W)G,
=
1
if Rest
>_ { 0
otherwise. )
However,
[(H, W )G' (G, O)G] = #{g E HAG/G I (H, W) < (G, 4)},
so that
[(H, W )G, aG(O)] =< bG((H, W
)G),
4 >= [(H, W)G, (G, O
)G]
for all (H,W)G and therefore aG(4) = (G, O )G. If p is irreducible and not
one-dimensional then the multiplicity of (G, 4))G in aG(p) is equal to
[(G, /)G, aG(P)] =< , P >= 0,
which completes the proof of Theorem 2.3.9.
Induction theorems
48
Definition 2.3.10 (See also 2.5.1)
Let / be a finite partially ordered
set (a poset). The Mobius function of .# is an integer-valued function, ,

on M x . t which is defined in the following manner. A chain of length i
in Ad is a totally ordered subset of elements of td,
MO
2.3.11
Ml#...
Mi.
We define PA,B, for A, B E . &, by

2.3.12
11 A,B
_ J:(-1)i#{chains of length i with MO = A, Mi = B in 2.3.11}.

i
2.3.13 The formula for aG in terms of Mobius functions
Let rO denote the Mobius function for the poset, JIG, of pairs, (H, 0),
of 2.2.9.
We will now give, without proof, the formula for aG in 2.3.7. In order
to do this one considers the formula of 2.2.16:
2.3.14
< 0, ResH(p) >=
1:
v H,0),(H',4')(H',cb')y(P)
(H,0)6 5(H'.O')G
in4'G /G
In order to solve 2.3.14 for the a(H-,O')G(p)s we have to invert the incidence matrix, (y(H
of the posetG/G. The inverse of this
incidence matrix is the Mobius matrix for the poset &G/G. The relationship between the Mobius matrix for MMG, (
and that
for AG/G, consists of the insertion of correction factors, #(H)/#(G).
Further details of the solution of 2.3.14 are to be found in Boltje (1990)
and Boltje (1989 (2.35) et seq.). We will not need the following explicit
formula, except in some illustrative examples.
Theorem 2.3.15
The homomorphism
aG : R(G) --> R+ (G)

of 2.3.9 is given by the formula aG(p)
_ #(G)-'
0', ResH,(P) > (H, O)G.

in..ffG
49
Table 2.1. Character table for Q8

1
4'
42
43
-1
2
3
-1
-1
-2
-1
-1
-1
Table 2.2. Table of (ResH(x))ab for Q8
H
Q8
C4
C4
C4
C2
{1}
1,1,2, 3
/2
/22
402
02
02
02
4)2
41 2
0+43
0
+3
0 + 03
1,4)
2.0
2-1
,I,
1,0 _ 03,4'2
1,,k '. 3, 02
4)3, 02
1, 4)
2.3.16 Example: G = Q8 (cf. 1.3.7)
We denote the three conjugacy classes in Q8 which contain elements of

order four by 41, 42 and 43. Similarly let 2 denote the conjugacy class of
the non-trivial central elements, X2 = Y2, in the notation of 1.3.7. The
character table of Q8 is presented in Table 2.1 (the (i, j)th entry is the
value of the ith irreducible character on the jth conjugacy class).
Now let us tabulate the conjugacy classes of subgroups, H, of Q8

together with the conjugacy classes of one-dimensional characters,
H -> C') E Hom(H, C") = H. We denote the three cyclic subgroups
3
of order four by C4, C42 , C4
and denote the centre by C2. For each group (k
is an injective homomorphism in H. For each irreducible representation,

x, of G we also tabulate the sum of all the one-dimensional representations which appear in Re$(x). We call this sum of one-dimensional
characters the abelian part of Re$(x) and denote it by (Re$(x))ab (See
Table 2.2.)
Now consider the formula for aQ8(v). By 2.2.36-2.2.40 and Table 2.2
this must take the form
2.3.17
aQ8(v) = a(C4, (k)Q8 + b(C4, 4))Q8 + c(C2, 4))Q8 + d(C2, )Q8
Induction theorems
50
Table 2.3. Character table for A5 (al = (1 + 5)/2,a2 = (1 - /5)/2)

2
V3,1
V3,2
V4
V5
51
52
0
0
al
a2
-1
-1
a2
-1
al
-1
-1
However, by maximality of (C4, 0) in tQg we have, by 2.5.10,
a=b=c=<0,ResQ4(v)>/(I NQs(C4i,CaI)=1.
By applying bQ$ : R+(Q8) --> R(Q8) to 2.3.17 and taking dimensions we
find that d = -1 and we obtain

2.3.18
aQ8(v) = Ei 1(C4, b)Q8 - (C2, O )Q1 E R+(Q8)
2.3.19 Example: G = A5
The character table of the alternating group, A5, has the form shown
in Table 2.3, in which the elements 2, 3, 51 and 52 have orders 2,3,5 and
5, respectively.
In Table 2.4, D6 and D10 are the dihedral subgroups of orders six and
ten, respectively, V4 = Z/2 x Z/2 and C denotes a cyclic group of order
n. The characters,
H -> C', can be taken to be any of the obvious
choices.
By 2.2.36-2.2.40, Table 2.4 and 2.5.10
O)A5
O )A,
+ (D6, O)A5 + (C5,

+ (V4, 1)A5
+(C3, 43)A5 + x(C2,1)A5 + y({1},1)A5.
aA5 (V3,1) = (D10,
From 2.2.42
3 = IndD',o(1) +IndD56(1) +IndAcS(1) +IndV4(1)
in R(A5), from which one finds that x = -2 and y = 0 to give

2.3.20
)A,
aA5 (V3,1) = (D10, 4,)A5 + (D6, 4')A5 + (C5, O
+(V4,
in R+(A5).
1,11)A5
+ (C3, 5)A5 - 2(C2,
O )A,
51
Table 2.4. Table of (Res'(X))ab for A5

H
V3,1
V3,2
V4
A5
A4
0
0
0
0
4+02
,1
,A4
1+Q>+0
1+ ,A2A+ ,,11, ,3
0+02+01+04
1 +2+93
At +E12 +u3
1 +111 +92+113
1+20
3.1
D1o
1,
D6
1'.0
C5
V4
C3
C2
{1}
o 0, fp2 ti 03
1, I 2 23
1'.0 _ p2
1,
1+0+02
V5
1+0+02
+ I, + ,1,2
+ ,I,3 + 34
1+1+'f22+W
2.
1+20+2qS2
1+20
2.1+0+4,2
2.1+20
3.1
4.1
5.1
3.1+2,
Induction theorems
52
2.3.21 Topological versions of explicit Brauer induction
I will close this section with some remarks relating Robert Boltje's Explicit
Brauer Induction formula, aG, to my original topological Explicit Brauer
Induction formula (Snaith, 1988b) and to Peter Symonds' topological

construction of aG (Symonds, 1991).
Suppose that v : G --> U(n) is an n-dimensional representation. Define
R+(G, NT") to be the free abelian group on (G - NT")-conjugacy classes
of homomorphisms of the form (4 : H -) NT"), where H < G and

NT" is the normaliser of the diagonal maximal torus, as in 2.1.10. Hence
R+(G, NV) = R+(G).
Let G act, via v, on the left of U(n)/NT" = X, as in 2.1.10. For each
conjugacy class of a subgroup, H < G, let Y(H) denote the subspace of the
orbit space, Y = G\X, consisting of orbits which are isomorphic to G/H
as G-sets. Let {Ya I a e d} denote the set of connected components of the
Y(H)s as H varies. If H,*P,(-; C) denotes compactly supported cohomology
with complex coefficients, we may define an integer, Xf, to be the Euler
characteristic of Ya
2.3.22
X = Ei(-1)' dimCHc'Pt(Ya;C) E Z.
For each a E d choose ga E U(n) which lies above Y. Therefore we

have a homomorphism
got lvga :
(g,-,
1vga)-1(NT) -f NT'
and, if Ya c Y(H), then H is conjugate to Ha = (g-1 vg,)-'(NT'). Hence

we may define
2.3.23
2G(v)
l a
= àEs4
L xa#/H
NT") E R+l(/G NT").
Another formulation of TG(v) may be given by means of a triangulation.

If we triangulate X in such a manner that G acts simplicially, as in 2.1.11,
2.1.14, we may choose an element, g, E U(n), above each i-dimensional
simplex in Y = G\X. Defining HQ as (g;1vg,)-1(NT"), which is the

stabiliser of the simplex above a and containing g, we may express TG(V)
as
2.3.24
TG(v) =
EvinY(-1)dim()(HQ
--p NT") E R+(G,NT").

2.3.25
PG : R+ (G, NT") -) R+ (G)
53
by applying the procedure used in the proof of 2.1.6 which canonically
decomposes a conjugacy class of maps, 0 : H -> NTn, into a sum of

monomial representations, EjIndH.(p ). Define
PG(H
NT n) = >(H.i,Wj)G E R+(G)
This homomorphism is natural if we define
Res : R+ (G, NT") --) R+(J, NT")

by the double coset formula, as for R+(G) in 2.2.1.
Define
2.3.26
TG(V) = PG(ZG(v)) E R+(G)
Also define, in the notation of 2.3.23,

2.3.27
EG(V) = L.IXEd
Xa(Ha,1)G E R+(G)
We collect the properties of these constructions from Snaith (1988b,

1989a, 1989b).
Theorem 2.3.28 (i) TG, TG and EG depend only on the class of the representation, v, in R(G).
(ii) If bn : R+(G,NTn) -+ R(G) is given by
bn(H 0 NTn) = I ndy(4))

then
bn(TG(v)) = v E R(G),
where n = dimC(v).
(iii) If bG = bi : R+(G) -- R(G) is as in 2.2.6 then
bl(EG(v)) = 1 E R(G).
(iv) TG, TG and EG are natural with respect to homomorphisms of G.

(v) If R+(G) is given the ring structure of 2.5.7 then
TG(V ) = TG(v)EG() + EG(v)TG()
and
EG(V P) = CG(V) ' EG(P)-
(vi) (See 2.5.11.)
Induction theorems
54
If aG is as in 2.2.43, then for all representations, v,

TG(v) = aG(v) eG(v) E R+(G).
2.3.29 Symond's description of aG
In Symonds (1991) one finds the following topological construction,

which is similar in flavour to 2.3.24. Given v : G -- p GL(V) we may
let G act, via v, on P(V), the projective space of V. Triangulate P(V)
so that G acts simplicially on P(V). For each simplex, a, of G\P(V)
let H(a) denote the stabiliser of a simplex, 6, chosen above a. The
points of a correspond to lines in V which are preserved by H(a). Let
Q : H(a)
) C* be the resulting one-dimensional representation, given
by H(a) acting on one of these lines.
Symonds (1991) defines
2.3.30
LG(v) =
Ear=G\P(v)(-1)di(o)(H(a), /a)G
E R+(G).
Theorem 2.3.31 LG(v) = aG(v) for all representations, v.
In Symonds (1991) all the properties of aG, which we derived algebraically, are derived topologically for LG.
2.4 Inductive explicit Brauer induction
In 2.3.9 we established the existence of an Explicit Brauer Induction

homomorphism
aG : R(G) -) R+(G)
for any finite group, G. The most essential property of aG is that it

commutes with restriction to subgroups, H < G. Slightly less important
is the property of commuting with inflation maps. However, aG does not
commute with induction. In many applications (see 6.3.8, for example) it
would be desirable to have an Explicit Brauer Induction homomorphism
which commutes with induction.
In this section we shall describe an Explicit Brauer Induction homomorphism with rational coefficients,
dG :R(G)-*R+(G)Q,
which is natural both with respect to induction from and restriction to
subgroups. In order to have this inductive property one must forego
55
something, namely, the integrality of the coefficients and naturality with

respect to inflation.
The discovery of this remarkable homomorphism is due to R. Boltje,
who gives the following characterisation of dG.
Theorem 2.4.1 In the notation of 2.2.1-2.2.8, there is a unique family of
homomorphisms
dG : R(G) -, R+(G) Q,
as G ranges over all finite groups, which satisfies the following properties:
(i) For all H< G the following diagrams commute:

dG
R(G)
dH
R+(H) Q
R(H)
dH
R+(H) Q
R(H)
I ndH
!lnd?J
dG
R+(G) Q
R(G)
(ii) If 0 : G -> C' is a one-dimensional representation and x E R(G)

then
dG((P
x) = (G,
4,)G
- dG(x) E R+(G) Q.
(iii) bGdG = 10 Q : R(G) -+ R(G) Q.

We shall prove Theorem 2.4.1 in a series of steps, culminating in
2.4.13. We begin with some preliminary discussion.
2.4.2
Let G be a cyclic group. Denote by eG the class function on G which is

given at x E G by
Induction theorems
56
2.4.3
if <x>=G,
otherwise.
eG(x) _
When G is non-cyclic we define eG to be identically zero.

Lemma 2.4.4
Let G be a finite cyclic group. In R(G) Q
(i)
eG = #(G)-'
#(K)/K,GIndK(1)
K<G
1 - eG = -(#(G)-')
#(K)uK,GIndK(1),
KEG
where (see 2.5.1) p.K,G is given in terms of the Mobius function, y, by UK,G =
([G : K]).
Proof Part (ii) follows at once from part (i).

If x E G then the character of I ndK (1) at x is equal to [G : K] if x E K
and is equal to zero otherwise. Hence the character value at x of the
right-hand side of (i) is equal to
([G : K]),
<x><K<G
which, by 2.5.2, is equal to 1 if < x >= G and is equal to zero otherwise,

as required.
Corollary 2.4.5
For any finite group, G:
(i)
eG E
fl Ker(ResK) J 4 R(G) Q,
KEG
1- eG E(Im(Ind)
E
KEG
4 R(G) Q,
57
R(G) Q = (nKer(ReSK)) B YIm(IndKG)

K#G
K#G
fl Ker(ResK)) = eG(R(G) Q),

KEG
(v)
_ (1 - eG)(R(G) Q).
I m(I ndK)
K#G
Proof The result is trivial when G is not a cyclic group, since eG = 0 in

this case. Hence we may assume that G is cyclic. In this case eG is clearly
an idempotent of R(G) Q that vanishes on proper subgroups, which
proves (i). Part (ii) follows from 2.4.5(ii). Parts (iii)-(v) follow easily from
the fact that both the summands in (iii) are ideals of R(G) Q.
2.4.6
Let e; denote the multiplicative character group of G, G =
Hom(G, C*). Define a ring homomorphism

2.4.7
7EG
: R+(G) 0 Q -p Q[G]
by the formula
1G((H,
O)G)
=
0
otherwise.
Define a ring homomorphism

2.4.8
pG : R+(G) Q --> (11H<G Q[H])G
by setting the fl-component of PG equal to

G
ItHResH
: R+(G) 0 Q -> R+(H) Q - Q[H],
so that PG = HH 7rHResH. In 2.4.8, G acts on HH<GQ[H] by means of

its (left) conjugation action on the poset of subgroups, H < G.
Induction theorems
58
Proposition 2.4.9
In 2.4.8 PG is a ring isomorphism.
Proof In the proof of 2.2.15 we showed that p E R+(G) is determined

uniquely by its `leading terms' (i.e. those terms of the form (H, )G with
H = G) and by its restriction to R+(H) for all proper subgroups, H < G.
This shows that PG is injective. The fact the PG is onto also follows
from the formulae in the proof of 2.2.15 which were used to show that a
rational solution existed to 2.2.16.
Definition 2.4.10
Define a homomorphism
PG :R(G)Q--.Q[G]
by the formula, for x E R(G),
2.4.11
PG(X) = eGx
where eG is the idempotent of 2.4.2. In 2.4.11 we note that PG(X) is zero

unless G is cyclic and in this case there is a tautological isomorphism
R(G)Q-->Q[G],
so that eGX may be interpreted as an element of Q[G]. Define a homomorphism
G
rG = fl PHResH : R(G) Q -'

H<G
Q[H]
H<G
Define dG by means of the following commutative diagram:

2.4.12
R(G) Q
dG
R+(G)
(IIHSG Q[H])G
59
Hence, by definition, for all H< G and X E R(G),

pH(ResH(x)) = nH(ResH(dG(x))) E Q[H]
2.4.13 Proof of Theorem 2.4.1

For each J < H and X E R(G) we have
nj(ResH(Res' (dG(x))))
= nJ(ResjG(dG(x)))
= pj(ResY(x))
by 2.4.12
= pi(ResH(Res'(x)))
= nJ(ResH(dH(ResH(x))))
by 2.4.12,
so that pH(ResH(dG(x))) = px(dH(ResH(x))). By 2.4.9, PH is injective so

that ResH(dG(x)) = dH(ResH(x)), as required.
Suppose now that x E R(H) for H < G and let J < G be any other
subgroup. In this case we have
nj(Res f (I nd4(dH(x))))
_ EZEJvG/H nJ(Ind.JinzHz- (ResilzHz
((z-')*(dH(x)))))
by 2.5.17
I:zEJ\GIH nj(Ind,Jinzez-' (ReSzHz-'
_
_
zEJvG/H
J<zHz-'
(dzxz ((z-')`(x)))))
nJ(ResJlzHZ-, (dzHz
EJ\GIH
J<zHz-i
In this calculation we have used the fact that

dzHz-,((z-')'(x))
(z-')*(dH(x))
and that
njIndK : R+(K) ---> R+(J)
is trivial when K is a proper subgroup of J.
-' Q[J]
Induction theorems
60
When J fl zHz-1 is not equal to J then

it (dj(I nd.JinzHz-I (ResJnzHz-, ((z-1)`(X)))))
= pj(dj(I ndinzHz 1(ReszjHzHz
((z-1)'(X)))))
= 0,
in the following manner:
so that we may rewrite

zrj(Res,G (I n4(dH(X))))
_ >zEJ\G/H ltJ(dJ(I ndJnzHz-l (ReszH

z-' -i ((z
JnzHz
_ itJ(dJ(ResY (I ndH(X))))
_ mJ(ResY(dG(IndH(X)))),
which implies that pG(IndH(dH(X))) = PG(dG(IndH(X))) and hence

IndHG(dH(X)) = dG(IndH(X)),
as required.
For : G ---> C', X E R(G) and H < G observe that
1rH(ResH(dG(4 'X)))
= pH(ResH() ResH(X))
= ResH(cb) PH(ResH(X)),
since eHResH(4)ResH(X) = ResH(0) (eHRe$(X)) when H is cyclic.

On the other hand,
JVH(ResH((G, 4)G ' dG(X)))
= JVH((H, ResH(q5))H ' ResH(dG(X)))
= ResH(0) ' rH(dH(ResH(X)))
by part (i)
= ResH(0) ' PH(Re$(X)),

which proves part (ii).
For part (iii), by 2.1.3 and the inductivity part of (i), it suffices to
61
assume that G is cyclic. In this case, by 2.4.5(v), it suffices to verify for

x E eGR(G) 0 Q that
bG(dG(X)) = X E R(G) Q.
For such a x suppose that x = Er a;4; where the {4j} are one-dimensional
representations. In this case,
PG(>t ai(G,
0,)G)
= {Ej a;(H, ResH(d1))H}H<G

= {ResH(x)}H<G
otherwise,
while
rG(X)
_ {pH(ResH(x))}H<G
x
if H=G,
otherwise.
Therefore, for such a x, dG(x) = E; a;(G, 4,)G and hence bG(dG(x)) _

Ej a;4, = x, as required.
It remains to prove the uniqueness part of Theorem 2.4.1. However,
as in the proof of part (iii), one is easily reduced to studying the case in
which G is cyclic and x E eGR(G) Q. In this case the argument which
was used to verify part (iii) for our construction of dG also shows that, if
x = Ej a,4 , then we must have dG(x) = E; a,(G, ;)G so that there is at
most one choice for dG, which completes the proof of 2.4.1.
The following result establishes some more useful properties of the

homomorphism, dG, of 2.4.1.
Theorem 2.4.14 Let dG be the homomorphism of 2.4.1, which was constructed in 2.4.12-2.4.13. Then
(i) dG is a ring homomorphism.
Induction theorems
62
(ii) For X E R(G)
dG(X) = #(G)-'
#(K)4uK,H(H,IndK(ResK(X)))G,
KSH<G
K,H cyclic
where (H, IndK(ResK(X)))G = >i e;(H, 4,)G if Ei e,o, = ResK(x) E R(K)
for ei = 1 and O, :H --- C.

(iii) The coefficient of ({1}, 1)G in dG(x) is equal to dim(x)#(G)-i
(iv) Define qG : R+(G) --> R+(G) by 1G((H, 4)G) = (H,1)G (see also
6.3.1), then, for x E R(G),
bG((lG 1)(dG(X))) = dim(X)#(G)-'Ind{i}(1) E R(G).
(v) If dG(x) = Ei ai(Hi, 0i)G E R+(G) Q then

ai = dim(X)#(G)-'
(vi) If aG : R(G) -f R+(G) is the homomorphism of 2.3.2 and 2.3.9 then

dG(X) = aG(X)dG(l)
Proof Observe that PG is a ring homomorphism, being multiplication by

an idempotent, so that rG is a ring homomorphism. Since dG = Pc1rG,
part (i) follows from 2.4.9.
Part (ii) follows by Mobius inversion of the equation pG(dG(x)) = rG(X)
This is similar to the manner in which the explicit formula for aG(x) of
2.3.15 is derived. We shall omit the details (see Boltje, 1990; Boltje, 1989,
(2.35) et seq.; 2.5.18).
Part (iii) follows at once from part (ii).

Next we shall prove part (vi). It suffices to show that
PG(dG(X)) = PG(aG(x)dG(l)) = PG(aG(X))PG(dG(l)).
By construction, pc(dc(X)) = rc(X) = {pH(ResH(X))}H<G, where

pH(ResH(x)) = 0 if H is non-cyclic and is equal to eHResH(x) if H
is cyclic. On the other hand,
PG(aG(X))PG(dG(1))
= {1tH(aH(ResH(X)))}H<G
= {(Ei ai(H, 4i)H)eH}H<G,
{eH}H<G
63
where ai =< 0 ResH(X) > for 4, : H - C*, by 2.3.2(ii). Hence, by

2.4.1(ii),
PG(aG(X)dG(1)) = {eHResH(X)}H<G,
which is {pH(ResH(X))}H<G, as required.
In order to prove part (iv) we shall use part (ii). However, since we
omitted the details of the general case of (ii) we will use only an easy
special case. We shall use only the formula for dG(1) when G is cyclic.
This formula is straightforward to prove directly by induction on #(G).
The homomorphisms, bG, nG and dG, commute with I ndH for H < G.
Suppose that H is a cyclic subgroup of G and 0 : H -p C' is a

one-dimensional representation which satisfies bH((jH 1)(dH(4))) _
#(H)-'Ind{i}(1). Therefore
bG((?1G 1)(dG(IndH(4))))
= I ndH(#(H)-11 nd{i}(1))
= dim(IndH(4))#(G)-'Ind{i}(1),
as required. Therefore, by 2.1.3, we may suppose that G is cyclic. Hence
bG((lG 1)(dG(X)))
= bG(nG(aG(X))) . bG((nG 1)(dG(1)))
= dim(X) . bG((1G (D 1)(dG(l)))
by (vi)
by 2.2.41
= dim(X)#(G)-i EK<H_<G #(K)[H : K]pK,HIndH(1)
= dim(X)#(G)-' LK<H<G #(H)JK,HIndH(1)

= dim(X)#(G)-'I nd{i}(1),
since, as remarked in the proof of 2.4.4, for H fixed,
E IK,H
K<H
vanishes when H is not equal to the trivial subgroup.
by (ii)
Induction theorems
64
Finally, for part (v), we observe that

Ei a;
=< bG(,7G(aG(X))),1 >G
by 1.2.39
= dim(X)#(G)-' < Ind{1}(1),1 >G
by (iv)
= dim(X)#(G)-',
as required. This completes the proof of Theorem 2.4.14.
We will close this section with some results concerning the behaviour
of dG with respect to fixing under normal subgroups. These results will
be used in our study of Swan conductors of Galois representations of
complete, discrete valuation fields in the non-separable residue field case
(see 6.3.20).
Let N a G be a normal subgroup of the finite group,

G. If V is a complex representation of G, define FixN(V) to be equal to
the N-fixed points of V,
Definition 2.4.15
FixN(V)=VN={vEV ln(v)=v for all nEN}.

Since GIN acts on V N' this defines homomorphisms
FixN : R(G) --* R(G/N)

and
FixN : R(G) Q --> R(G/N) Q.

We may also define homomorphisms
Fix+,N : R+(G) -' R+(G/N)

and
Fix+,N : R+(G) Q -* R+(G/N) Q
by the formula
(HN/N, 3)G/N, if ResHfN(4) = 1,
Fix+,N((H, 4)G)
0
where 3(nh) = 4(h) for n E N, h E H.
otherwise,

Proposition 2.4.16
65
The following diagrams commute:

bG
R(G)
R+(G)
Fix+,N
FiXN
bG/N
R+(G/N)
R(G/N)
dG
R(G)
R+(G) Q
Fix+,N
FixN
dG/N
R(G/N)
R+(G/N) Q
Proof If ResHfN(4) = 1 then, by 2.5.14

FixN(bG((H,O)G)) = IndHN(4)
= bG(Fix+,N((H, O)G)),
as required.
When ResHnN(q5) * 1 we must show that FixN(bG((H, 4)G)) = 0. In
this case, by 2.5.13,
FixN(bG((H, ,)G)) = C[G/N] C[Hi V,
where V is the field of complex numbers upon which h E H acts by

multiplication by O(h). If n E H n N and 4(n) * 1 then, in C[G/N] C[Hl
V,
gN C[H] v = gnN C[HJ v = gN C[H] 4(n)v,
so that gN C[H] v = 0 for all g E G and v E V, as required.
Now we shall prove by induction on #(G) that the second diagram

commutes.
First we shall consider
I m(1 ndK )) = (1 - eG)(R(G) (D Q),
xE
K :#G
Induction theorems
66
in which case, by linearity, it suffices to set x = I ndGl(0) for some proper
subgroup, H < G, and for 0 E R(H). Let Res
H/(HnN)HN/N
) denote the
homomorphism induced by the isomorphism, HN/N --> H/(H f1 N).

We will also need to know that there is an isomorphism of complex
G/N-representations of the form, where M is a complex representation
of H,
C[G/N] C[HN/N] ResH/(xnN)(MHnN)
= (IndH(M))".
H
This can easily be seen from 2.5.13, which yields an isomorphism of the
form
C[G/N] C[HI M = (IndH(M))N.

Now we may decompose the H-representation, M, as M = MHON M'
and observe that, by the argument used in the first part of this proof,
C[G/N] C[H] M= 0.
Hence we have
Fix+,N(dG(x))
= Fix+,N(dG(I n4(0)))
= Fix+,N(IndH(dH(0)))
GIN
= I ndHN/N (ReHI(HnsH(HN) (Fix+,HnN(dH(0))))
= dGIN(FixN(I ndH(0)))
= dGIN(FixN(x)),
as required.
Finally we may assume, by 2.4.5, that
xE
eG(R(G) Q)
K*G
and that G is cyclic, since such a x would be trivial otherwise. By the

argument used in the proof of 2.4.1 in 2.4.13 we have dG(x) = (G, x)G, so
that
Fix+,N(dG(x)) = (G/N,
FixN(x))G,N.
2.5 Exercises
67
However, dG/N(FixN(x)) = (G/N, FixN(x))G/N since
FixN(x) E eG/NR(G/N) 0 Q,
by 2.5.20.
2.5 Exercises
2.5.1
Let A < B be cyclic groups such that [B : A] is finite. Define I A,B
by
uA,B =
1:(-1)i#{chains of subgroups of length i from A to B)i2!0
Show that /1A,B = p([B : A]), where p(n) is the Mobius function of 2.1.1.
(See 2.3.10 for the general definition of the Mobius function of a finite
poset.)
2.5.2 Mobius inversion Let f be any arithmetical function defined on positive integers. Show that, if F(n) = Eden f (d) then f (n) = Eden p(n/d)F(d).
Show that every finite p-group is nilpotent. Conversely, show that
every finite nilpotent group is the product of its Sylow subgroups.
2.5.3
2.5.4
Let G be an M-group and let 1 = f1 < f2... < fk be the
distinct dimensions of the irreducible complex representations of G. Let

pi = I ndG (ci) be irreducible with f i = [G : Hi].
(i) Show that ker(pi) contains the ith group, GO), in the derived series
of G (the series obtained by taking successive commutator subgroups).
(ii) Show that an M-group is solvable.
(Hint: Use induction on i. Compare the kernel of pi with the kernel of
IndG (1).)
2.5.5
Let Hi < Gi be finite groups and let pi : Hi -+ GL(Vi) be
K-representations for i = 1, 2. Show that

xG2(P1
Hi XH2
2.5.6
Let
d
d
0->V-->Vn-1-...-*V0->0
P2)
Induction theorems
68
be a sequence of maps of C[G]-modules. Suppose that dd = 0 : V, -+

Vi-2 for each i. Set
H` -
(ker(d) : V1 ---> Vi-1)
(im(d) : V;+1 -) Vi)
and show that E,(-1)'Hi =

2.5.7
>i(-1)'Vi E R(G).
Let J < G and define I nd
R+ (J) --> R+(G) by
I nd j ((H, )J) = (H, 4)G, Also define a product on R+(G) by the formula
((w-'Kw) n H,w*(cb)W)G,
(K,cb)G ' (H,W)G =

wEK\G/H
where w'(4)(z) = 4(wzw-1).

With these definitions, verify that R+(G) is a ring-valued functor. Also
verify that, if A E R+(G) and B E R+(J) then
I nd,Gl (Rest (A) B) = A I nd,GG (B) E R+(G).
Verify that bG, in 2.2.6, is a ring homomorphism which commutes

with the induction and restriction maps of 2.2.1.
2.5.8
2.5.9
In the notation of Theorem 2.2.15 show that, if p : G --p GL(V)
factorises as p : G -+ GIN -+ GL(V) then

(i) a(H,O)G(p)
0 implies that N < H and ResN(4) = 1.
(11) a(H,O)G(P) = a(HIN,O)G/N(P')
Let X E R(G) and suppose that (H, 0) EG is an element which

is maximal among those satisfying < 0, ResGr(x) >* 0. Prove that
2.5.10
a(H,O)G(x) =< 0,ResH(x) > '([NG(H,4) : H])-1.

2.5.11
In the notation of 2.3.28, prove that

aG(v)'EG(v) = TG(v)
for all v : G -* U(n).

2.5.12
Let V E R(G) and let 2 : G ---> C' be a one-dimensional
representation. Show that

aG(A v) = (G,
2)G
' aG(v) E R+(G).
2.5 Exercises
69
That is, if
aG(v) = E ni(He, 4) )G
then
aG(2 (9 v) = E n,(H1, 2. 0,)G.

2.5.13
Let G be a finite group. Let N a G and let V be a finite-
dimensional K-representation of G. Suppose that char(K) is coprime to

#(G). Define a G/N-representation, q(V), by
q(V) = K [G/N]
Show that q(V)andVN={vEV In(v)=v for all nEN} are

isomorphic G/N-representations.
Let 0 : J -> C" be a one-dimensional complex representation

of a subgroup, J < G. Suppose that N a G and that 0 is trivial on N fl J.
2.5.14
Define0 :NJ ---fC' by(nj)=(j) for nEN,j EJ.

In the notation of 2.5.13 show that
q(I nd (4)) = I ndNj().
(Hint: Use 2.5.13 and the discussion of I ndjG (-)N in 5.3.18 (proof) to
obtain a convenient basis.)
Let N a G be finite groups and let V be a finite-dimensional,

complex representation of G. Prove that the character, Xq, of q(V) is
2.5.15
given by
Xq(gN) = #(N)-' T. Xv(gn)

nEN
where Xv is the character of V.
(Hint: Reduce to the case in which V = I nd j (4)), as in 2.5.14. Use

2.5.14 when ResNIV(4)) = 1 and in the remaining case show directly that
both expressions are zero.)
2.5.16 (A topological construction of aG for p-groups) Let G be a finite pgroup. Let e : R+(G) ---> Z be the augmentation defined by e((H, 4))G) _
[G : H] and set IR+(G) = ker(e).
(i) Show that the IR+(G)-adic completion of R+(G) is isomorphic to

Z 9 (IR+(G) 0 Zr), where Z, denotes the p-adic integers. (For the notion
Induction theorems
70
of the completion of a ring with respect to an ideal see Matsumura,

1990.)
(ii) In the notation of 2.5.11, show that TG eG1 gives a well-defined

homomorphism from R(G) to the IR+(G)-adic completion of R+(G)
satisfying the conditions of 2.2.8 save for the fact that the homomorphism
has p-adic coefficients.
(iii) Use the formula of 2.2.16 together with the fact that ZPnZ[1/p] _
Z, the integers, to show that TG eG1 has integral coefficients.
This verifies that aG exists and satisfies the conditions of 2.2.8 and
2.5.11 when G is a p-group.
2.5.17
Let H, J < G be subgroups of a finite group, G. Prove that the
composition
R+ (H) -+ R+(G) -> R+(J)

is given by the double coset formula
Rest (I ndH (w)) _
(ResJHH ,
((z-1)*
(w)))
zEJ\G/H
Prove 2.4.14(ii) directly. (Hint: Prove that the right-hand side of

the formula is inductive and then verify the formula when x (=- eGR(G) 0
2.5.18
Q.)
2.5.19
Let dG be the homomorphism of Theorem 2.4.1 and 2.4.12.
(i) If 0 : Z/p -> C' is a non-trivial character of the cyclic group of

prime order, p, show that
dZ/P(4) = (p - 1)P 1(Z/p, 4)Z/P - P-1
E(Z/P,1V)Z/P
+ P 1(Z/P, 1)Z/P
w#0
(ii) Show that
dZ/4(1) = 2-'(Z/4,1)Z/4
- 2-1(Z/4,x)Z/4 +4-1(Z/2 1)Z/4
-4-1(Z/2, i)Z/4 + 4-1({1},1)Z/4,
where x and i are one-dimensional characters of order two.

(iii) If x is the two-dimensional irreducible representation of the dihedral group of order six show that
dD6(x)
3-1(Z/3,
O)n6
3-1(Z/3,
1)D6
+ 3-1({1}, 1)6.
2.5 Exercises
2.5.20
71
Let N i G be a normal subgroup of the finite group, G. Prove
that, if x E eGR(G) Q, then

FixN(x) E eG/NR(G/N) Q.
2.5.21
Derive the following alternative formula for the homomorphism
of 2.3.15:
aG : R(G) -> R+(G).

For p E R(G),
aG(P) _ (-1)" < 0n, Re$ (P) > (Ho, Oo)G,

where a runs through the set of G-orbits of chains in J/G of the form
(Ho, 00) < (Hl, 01) < ... < (Hn, 0n).
3
GL2Fq
Introduction
In this chapter we evaluate some of the terms in the Explicit Brauer

Induction formula, aGL2Fq, applied to a finite-dimensional, irreducible
complex representation of GL2Fq. I have included these examples because the family of irreducible representations of
(as n and q vary)
exhibit a number of interesting features which one would like to understand from the viewpoint of Explicit Brauer Induction. These features
mirror the conjectural properties of admissible representations of GL
of a local field, which form part of the scheme of conjectures posed by
R.P. Langlands and others (Gerardin & Labesse, 1979). This similarity
is reason enough to wish to examine GLnFq in a new light.
One of the most elegant properties of
is the Shintani correspondence, which assigns to each G(Fgs /Fq )-invariant irreducible representation of GL Fqs an irreducible representation of GL Fq. When n = 1
this correspondence follows at once from Hilbert's Theorem 90. For
higher values of n the correspondence is much more difficult to establish.
However, in Section 2, we describe the Shintani correspondence for GL2
entirely in terms of Explicit Brauer Induction and Hilbert's Theorem 90.
Section 1 is devoted to the construction of the cuspidal representations.
We follow a method, using the Fourier transform, which was used by
Andre Weil to construct cuspidal (admissible) representations of matrix
groups of local fields.
and
Section 2 gives all the other irreducible representations of
evaluates sufficiently many of the terms in the Explicit Brauer Induction
formulae to enable us to describe the Shintani correspondence.
Section 3 consists of a set of exercises and a research problem connected
with the Shintani correspondence for GL Fq for larger values of n.
72
3.1 Weil representations
73
In this chapter we shall calculate some of the maximal terms in the

Explicit Brauer Induction formula, aGL,F, (p), when p is an irreducible
representation of GL2Fq, the group of invertible 2 x 2 matrices with entries
in the finite field, Fq. All such irreducible representations are well-known.
In fact, all the irreducible representations of GL Fq were described in

Green (1955) see also Macdonald (1979, chapter IV) for all values of n
and q. However, for completeness and convenience, we shall recall here
the explicit construction of the irreducible representations of GL2Fq. We
shall begin with the cuspidal or Weil representations which are the most
difficult ones to construct. These representations are originally due to A.
Weil. The construction works in greater generality than we will need. For
example, in Gerardin & Labesse (1979, p. 122) the Weil representation is
described for the case in which Fq is replaced by a local field.
3.1.1
Let F be any field. Define the Boreel subgroup, B < GL2F, to be
B={XEGL2FIX=1 a f)}.
Define the unitriangular subgroup, U < B,\\to be
U={YEBIY=1
/ JJJ
# I}.
Define w E SL2F to be given by

w
O1
The Bruhat decomposition of GL2F takes the form
GL2F = B U BwU.
This is elementary in the case of 2 x 2 matrices and an explicit formula

which gives this decomposition is to be found in 3.3.1.
Proposition 3.1.2 Let F be any field, then GL2F is generated by matrices
of the form (a, 6 E F*, u E F)
0
5 )'
ko
and
1
subject to the following relations:
w =
GL2Fq
74
(i)
10
a 1
6-1
)=(
au8-1
)'
U1
1
)(0 u11u2 )
a 0)w_1=(1 0)0
a
(iv)
(1
)W -(
-u 1
0
-u)(0
)W(0
-u-1
1
and
(v)
Proof See 3.3.2.

We will now discuss the (q -1)-dimensional complex vector space
upon which we are going to inflict the Weil representation of GL2Fq. We
will require some preliminary notation. Let Fqz denote the field of order
q2 so that the Galois group, G(Fg2/Fq), is cyclic of order two generated
by the Frobenius automorphism (see also 4.3.7):
3.1.3
F:Fg2->Fq2
given by F(z) = zq for all z E Fq2.
In order to construct a Weil representation we shall need a character
of the form
O : Fq2 - C
which we shall generally assume to be distinct from its conjugate by the
Frobenius, O * F`(O) where F'(O)(z) = O(F(z)).
Let . denote the following complex vector space:
= If : Fq2 - C I f(t 1x) = O(t)f(x) if N(t) = 1},

where N = NFg2/Fq : FF2 -- Fq is the norm.
75
Let H < B denote the abelian subgroup which consists of matrices

whose diagonal entries are equal. Hence there is an isomorphism of the
form
given by
0
=(z,yz-1).
Define the additive character of Fq to be the homomorphism
T=PF,
3.1.4
by the formula (char(Fq) = p)

PFq(y) = exp(27ti(TraceFq/F,(y))/p).
Hence we may obtain a one-dimensional representation of H, denoted by

O 0 P, by the composition
O'Y=(OOTFq)y :H-) C'.

By induction we obtain a (q - 1)-dimensional induced representation of
B, IndH(O 0 `Y). We may identify the underlying vector space of this as
a mapping space in the following manner. There is an isomorphism
2 : W -- I ndH(O (& `Y),
where
))_()`(P/c)g(X)}
where O denotes the complex conjugatea of O. Explicitly 2 is given by
2(g) = E X 0 g(X) E C[B] C[H] Co`Y,

XEB/H
where Co'Y denotes the complex numbers with the H-action via O T.
The isomorphism, A, is well-defined since
(0
flg(x(0
= X 0 O(a)P(fl/a)g
= X 0 g(X).
(X ( a0
)/
76
GL2Fq
Define an action of B on W by the formula (g e W,X E B)

1
((0
S) g)
=g
(0 S)
(X) = g
\\ O1
Proposition 3.1.5 With this B-action on W
A : W = 1 ndH
(9 'Y)
is an isomorphism of B-representations.
Proof This result will be left as an easy exercise for the reader (3.3.4).
3.1.6
Since matrices of the type
form a set of coset representatives for B/H we may define an isomorphism

of vector spaces
A:.-->W
by the formula, where b E F 2 and NF9,IF,(b) = a 1,
A(h)
3.1.7
((0 0))
O(b)h(b).
Notice that 3.1.7 is well-defined because, if N 2/F9(t) = 1,
O(tb)h(tb) = O(t)O(b)O(t)-lh(b).
Proposition 3.1.8 Define a B-action on
(ot
by the formula
(h) _ {x -
b)
where N = NFy2/F, is the norm and N(2) = ab-1. Then, with this B-action,
A yields an isomorphism of B-representations
A:
-- I ndH(O 0 `Y).
77
Proof First we note that the proposed formula does indeed define an
to other functions in.
action which sends functions in
We shall work in terms of the map A :. -) W. By definition,
) A(h) )
((
vbx
((
?))
6-1
1) (
-N(v62)-1
S-1
v6 00-1
e(6)T(/3(v6)-i)
= [A(h) ((
= O( b)h(b)O(b)`P(f3(vb)-1),
where N(b) = a(vb)-1. Now define g E . by the formula

g(x) = O(b)'Y(fN(x)6-1)h(2x)O(A),
where N(A) = ab-1. We have, if N(c) = v-1,
A(S) ((0 1) )
= O(c)g(c)
= O(c)O(b)T(# N(c)b-1)h(2c)O(.t)
= O(2c)O(6)P(/3(v6)-1)h(.1c)
((0
)A(h))((
?))
since N(2c) = a(vb)-1. This shows that the vector space isomorphism, A,
commutes with the B-actions, as required.
Remark 3.1.9
From 3.1.8 we find that the B-action on
is given by
78
GL2Fq
h x=
() P (uN(x))h(x),
a 0) h) (x) = h(Ax)O(2),
0
where N(2) = a and

C C 10
08 ) h) (x) = h(xF(p)-')O(p),
where N(p) = 8 since, if N(A) = S',

O(8-')O(F(S)-1)h(
h(xF(p)-')19(y).
These formulae coincide with those of Gerardin & Labesse (1979,

p. 122).
In order to manipulate the Fourier transform on -Y we will repeatedly

make use of the following result.
Lemma 3.1.10 As in 3.1.3, let f E
and suppose that O : Fq2 ---p C' is
non-trivial. Then
E f(v)=0.
vEF'2
9
Proof Let xl,... , xq_1 E FF2 comprise a set of coset representatives for
F92/Ker(N) where N = NF92/F9 is the norm. We may rewrite the sum as
EVEF'2 f (V)
Eq-1
i=1 EtEKer(N) f (txi)
_ Eii--11 [EtEKer(N) E)(0-11f (xi)
=0
by 3.3.3.
3.1.11
The Fourier transform on
In the notation of 3.1.1 and 3.1.3 suppose that f E. We define the

Fourier transform, f E , of f by means of the formula
j (z) = -q-' >
3.1.12
yEF,2
q2
79
f (Y)1'F, (YF(z) + zF(y)),
where PFq is the additive character of 3.1.4.

Lemma 3.1.13 The map which sends f E
C-linear endomorphism of order four.
to its Fourier transform is a
Proof If z E Fq2 and t E Ker(N) then, by 3.1.12,
f (tz) = -q-'
f (Y)`V Fq (YF(tz) + tzF(y))

9
= -q-' EyEF'2 f(Y)Ì`Fq(Yt-'F(z) + zF(t-'Y))

9
since F(t) = t-',
= -q-'
f (tv)'PFq (vF(z) + zF(v))
setting v = yt-
_ -q-'
=
0(t)-'f (v)'PFq (vF(z) + zF(v))

9
O(t)-' f(v)
Hence the Fourier transform, which is evidently C-linear, yields an endomorphism of *. To show that the Fourier transform gives an endomorphism of order four we will establish the identity
f(z) =f(-z)
3.1.14
for all z E F92i f E "Y.
This is seen as follows:
f(z) = -q-'
yEF-2 f(Y)PFq(YF(z)
+zF(Y))
= q-2 Ey,VEFg2 f(v)Y'Fq(yF(z) + zF(Y))'I'Fq(YF(v) + vF(y))
q-2
f(v)`PFq(YF(v + z) + (v + z)F(y)).
q
80
GL2Fq
When v + z # 0 we have, by 3.1.10,
Ì'Fq(yF(v + z) + (v + z)F(y)) = -1
yEF'2
q
and when v + z = 0 we have
E TFq(0)=q2-1.
yEFg2
Therefore we obtain
f(z) = q-2{(q2 - 1)f(-z) -
f(v)} = f(-z),
-z*vEF'2
9
by 3.1.10.
Let O : FF2) C* be a non-trivial character, as in
Definition 3.1.15
3.1.3. The following three formulae characterise the Weil representation

associated to O:
r(O) : GL2Fq -> Autc(i) = GLq_1C

(i)
(r(O)(w)f)(x) = f(x)
(f E,x E Fq2),
where w is as in 3.1.1,
(ii)
(r(O)
(0
) f)(x)
= `YFq(uN q2/Fq(x))f(x)
and
(iii)
(r(O) C 0
) f) (x) = O(Q)f (fix),
where a E Fq, f3 E Fq2 and NFqZ/F(Q) = c.
Theorem 3.1.16 The formulae of 3.1.15 characterise a unique, well-defined

(q - 1)-dimensional, irreducible representation, r(O), of GL2Fq.
3.1 Well representations
81
Proof Notice that, by 3.1.2, 3.1.9 and 3.1.13, these formulae do define
a unique automorphism of A' for each element of GL2Fq. Therefore
we must verify that the formulae respect the relations (i)-(v) of 3.1.2.
Relations (i) and (ii) follow from 3.1.8-3.1.9 since they are relations
between elements of the Borel subgroup, B. Also relation (v) follows
from the fact that the Fourier transform is of order four, by 3.1.13.
Relation (iii) is straightforward to verify, using the well-known identity
h(x) = f (xF(/3)-' )
if f E *; x, /3 E Fq2 and h(x) = f (/3x).

In order to verify relation (iv) we will prove the following identity, in
which r() has been abbreviated to r:
3.1.17
(r(w)r
1) r(w)f)
(0
=q
(x)
f (v) Y' Fq (- N (x
- (r (
0 ) r (0
0'
+ v)u ' )
lu ) r(w)r (
1 1 ) f) (x).
We have
1
(r(w)r (0
(r(w)r
(0
_ (r(w){z --> `YF,(uN(z))f(z)))(x)
= -q-i EyEFgz `FF,(uN(Y))f(Y)'`F,(YF(x) +xF(Y))

= q-2 Ey,,.F.2 `FFQ (uYF(Y))`FFq (yF(x) + xF(y))
x Ì'Fq (YF(v) + vF(Y))f (v)
= q-2 Ey,vEF'z'FFq(uyF(Y) + yF(x + v) + (x +v)F(Y))f(v)
GL2Fq
82
If a = x + v and N(z) = u # 0, where N denotes the norm, we observe

that
N(yz + aF(z)-1)
= (yz +
aF(z)-1)(F(y)F(z)
+ F(a)z-1)
= uyF(y) + aF(y) + yF(a) + N(a)u-1.

Therefore we have
EyEF.'1' (uyF(y) + yF(x + v) + (x + v)F(y))

q2
= EyE 92'I'Fg(N(yz + aF(z)-1))'I'Fq(-N(a)u1)
= -1 + EyEFgz'I'F,(N(yz +
aF(z)-'))'I'F9(-N(a)u 1)
= -1 + {EsEFg2 TF9(N(s))}WFq(-N(a)u 1)
= -1 +'PFq(-N(a)u-1){'I'Fg(0) + (q + 1) EZEF9'I'Fq(Z)}
= -1 +'I'F'(-N(a)u 1){1 + (q + 1)(-1)},

by 3.3.3,
= -1 - q`!Fq(-N(a)u 1).
Therefore, substituting into the previous formula,
(r(w)r
1) r(w)f) (x)
f (v)} - q-1 {EvEr, f (v)'VFg (-N(x + v)u 1)}
=
-q-1
q2
f (v)`YFq (-N(x + v)u-1)},
by 3.3.3, which establishes the first half of 3.1.17.
83
On the other hand, by 3.1.9, one easily deduces that
(r(
=
-Ou
(r (0
r(0
) r(w)r
1u )r(w)r(
0
1 )f)(x)
i-1) f) (-xu-1).
Finally,
(r (0 lu) r(w)r (0
='FFq (-uN(-xu-'))f r(w)(z
1 1) f) (-xu 1)
'FFq (-N(z)u-') f (z))}(-xu 1)
= q-1iFFq (-uN(-xu-' )){- >,,:F' fq2 (v)WFq (-N(v)u 1)

'F(v) - F(x)F(u)-1v)}
x
(-u 1(N(v) + N(x) + vF(x) + F(v)x))

_ -q-' >VEF'f(v)'Fq
i
q
-9-1
EvEF'f(v)!Fq(-U 1N(x + v)),

q2
as required to establish the second half of 3.1.17.

It remains to show that r(O) is irreducible. By 3.1.8 it suffices to show
that IndH(O 0'F) is an irreducible representation of B. However,
B
=< O 'F, ResH(I ndH(O (9 'F)) >H
= E.C:F; < O 'F, O 'F(z -) >H,

by 1.2.40, where'F(z -)(a) ='F(za). However, when z # 1,
<'F,'F(z -) >= 0,
by 3.3.3, so that
B= 1,

3.1.18
For future use let us record the conjugacy class information
concerning GL2Fq.
84
GL2Fq
Table 3.1. Conjugacy classes in GL2Fq

Minimal
polynomial
Type
Conjugacy class
representative
(t - a)(t - j3)
a#fEF
q(q + 1)
(t - a)
ac E F9
t2 - (x + F(x))t + xF(x)
F(x) * x E F2
( 0 -xF(x)
(t
II
a)2
aEF
III
IV
Number
in class
2-1
x+ F(x)
q2
-q
The conjugacy class of a matrix, X E GL2Fq, is determined by its

minimal polynomial. The minimal polynomial of X must have degree
one or two. The representatives of each conjugacy class, together with
their minimal polynomial and the number of elements within each class,
are listed in Table 3.1.
The remainder of this section will be devoted to the calculation of the
character values of the Weil representation.
Lemma 3.1.19 Let

Y
(0
x + F(x) ) E GL2Fq
denote the conjugacy class representative of type IV in Table 3.1. Then,
for f E
, (r(O)(1')f)(z)
-q-1 0(-X)
f (y)WFQ ((x + F(x))N(y) + yF(xz) + xzF(y)),

YEF'Z
9
where N = NFgz/Fq is the norm.
Proof We begin with the observation that, by 3.3.1,

3.1.20
Y
0
1
-xF(x)
x+F(x))
where w is as in 3.1.1.
-xF(x)
0
-1 ) w (0
x + F(x)
1
'
3.1 Well representations
85
Write t = x+F(x), d = xF(x) = N(x) and let a E Fq2 satisfy N(a) = -1.
Therefore we have
(r(O)(Y )f)(z)
- (r(
=
(r (-d
0
0
_ -q-1
)f)(z)
01)
r(w){s HP(tN(s))f(s)}) (z)
(r (-d 0
-1 )
{s i
'
f (y)'P(tN(y))'I'(yF(s) + sF(y))}) (z)

9
= -q 1O(ax)O(a){>yEF.2 f(y)T(tN(y)+yF(axzF(a)-')+F(y)axzF(a)-i)}
9
f(yc 2)`P(tN(ya 2) - ya 2F(xz) - F(ya 2)xz)},

9
aF(a) = -1 and N(a-2) = 1,
since
_ -q-10(x)0(-1){>vEF.2 f (v)`P(tN(v) + vF(xz) + F(v)xz)},

q2
by setting v = -ya2, which completes the proof.

Proposition 3.1.21 With the notation of Table 3.1, the character-values of
r(O) on conjugacy classes of type IV are given by
Trace
(r(e) ((0
xF(
x + F(x)
))) _ -{O(x) + O(F(x))}.
Proof In this and subsequent calculations it will be convenient to have

chosen a basis for the representation space,. To this end let us choose,
once and for all, v1,...,vq_1 E 119, so that the set {N(v1);1 < i < q - 1}
is equal to F. Hence the {v;} are a complete set of coset representatives
for Fq2/Ker(N). Let {f; E *< i < q - 1} be the unique functions

which satisfy
1
3.1.22
if ij,
fi(vi)
0
Hence, if Y E GL2Fq and
if i
j.
GL2Fq
86
11
r(O)(Y)(.fl) = I9 ailfl,
3.1.23
then we see at once that

ai.l = (r(O)(Y).fi)(vl)
and
3.1.24
Trace(r(O)(Y)) = E911 r(O)(Y)(fi)(vi)

Therefore, from 3.1.19 and 3.1.24, if we abbreviate by setting x+F(x) _
t then we obtain
Trace (r(O) ((
-xF(x)
x + F(x)
x
(r() ((
x + F(x)
))f)(v)
q-1
-q-10(-x) 1: 1: .fi(y)'1'(tN(y) + yF(xv,) + xviF(y))

i=1 yEF'2
q-1
-q-10(-x) E E .fl(uvi)W(tN(vi)
i=1 uEKer(N)
+uF(x)N(v,) + xF(u)N(v,)),
since
f,(uvj) = 0 if i
j and N(u) = 1,
q-1
_ -q-10(-x) E
O(u)-1W(N(v,)(t + uF(x) + xF(u))).
i=1 uEKer(N)
Next we observe that if t + uF(x) + xF(u)
0 then, by 3.3.3,
q-1
`V(N(v,)(t + uF(x) + xF(u))) = -1.

i=1
This happens for all the values of u E Ker(N) except for u = -1 and
u = -xF(x)-1 for which t + uF(x) + xF(u) = 0. Therefore the terms with
u = -1 contribute
-q-1O(-x)(q
-1)O(-1)-1 = -(q -1)q-10(x)
to the sum, while those for which u = -xF(x)-1 contribute
-q-]O(-x)(q
1)0(-xF(x)-1)-1 = -(q
- 1)q-1O(F(x)).
87
The remaining terms contribute, by 3.3.3,
-q-10(-x){-O(-1)-1
- O(xF(x)-1)-1} = -q-1{O(x) + O(F(x))},
so that adding these contributions together yields the required formula.
0
r(O) on conjugacy classes of types I and III are given by
Trace r(O) ((
(q - 1)O(a) if a = fl,
)))
otherwise.
Proof Let a, b E Fq2 satisfy N(a) = a and N(b) = a where it is understood
that if a = /3 then we shall choose a = b. Choosing the basis for

which was used in the proof of 3.1.21 we find, by 3.1.9, that
Trace I r(O)
(I 0
III =
q-1
O(a)O(b)f;(av;F(b)-1).
When a # /3 then N(aF(b)-1) * 1 and f1(av1F(b)-1) = 0 for all i, which

yields the required formula in the case when a and /3 are distinct. On the
other hand, when a = $
//
Trace (r(e) I
a)) )
= E9 11 0(a)0(a)f,(av,F(a)-1)
= Eq i
O(a)O(a)O(F(a)-1)-1O(a)-1
_ (q - 1)O(a),
as required.
r(O) on conjugacy classes of type II are given by
Trace I
r(O) (( 0
a))) = -O(a).
GL2Fq
88
Table 3.2. Character values of the Weil representation, r(O)

Character
Conjugacy class
representative
Type
value
7a0 0
II
a)
-O(a)
0 a)
III
IV
-xF(x)
x + F(x)
(q - 1)O(a)
-{O(x) + O(F(x))}
Proof As in the proof of 3.1.25, let N(a) = a. Then we have

Trace
(r(e)
= Trace
(( 0
(r(@) ((0 a)) r(O) ((0
Ei_
=
11
)) )
(r()
O(a)2
a1
((
)) f)
(avjF(a)-1)
= E9 it O(a)2W(a-1N(vjaF(a)-1))fj(vjaF(a)-1)
11O(a)T(a 1N(v1)),
since N(aF(a)-1) = 1. This yields the required formula since, by 3.3.3,

q-1
'V(a 1N(v1aF(a)-t)) _ -1.

i=1
For the reader's convenience we tabulate the results of 3.1.21, 3.1.25 and
3.1.26 (see Table 3.2).
3.2 Explicit Brauer induction and Shintani descent
89

In this section we will evaluate some of the maximal terms, (H, q5)GF2Fq
in the expressions for aGL2Fq (p), as p runs through the irreducible rep-
resentations of GL2Fq. In the previous section we constructed the Weil

representations and, before proceeding further, we shall now construct
the remaining irreducible representations of GL2Fq.
Suppose that we are given characters of the form
C
X, X1, X2 : Fq
then we clearly have a one-dimensional representation, L(X), given by

3.2.1
L(X) = X det : GL2Fq
det
Fq
If Xi and X2 are distinct define
InfB(Xi X2):B-IC"
by inflating Xi X2 from the diagonal torus, T, to the Borel subgroup,
B. That is,
InfB(Xi X2)
((0
b ))
=Xi(a)X2(b)
Define a (q + 1)-dimensional representation, R(X1, X2), by

3.2.2
R(Xi, X2) = I ndBr'2Fq (I of B(X1 X2))
When X = X, = X2 we have
InfB(X 0 X)
ResBr'2Fq(L(X)) : B
-> C',
so that there is a canonical surjection of the form

IndB`zFq(InfB(x X))
-> IndGL2Fq(L(X)) = L(x).
Therefore we may define a q-dimensional representation, S(X), by means

of the following short exact sequence of representations (which is split,
by 1.1.8):
3.2.3
0 -f S(X) ---> IndBF2Fq(InfB(X X)) - L(X)
)0.
Theorem 3.2.4 A complete list of all the irreducible representations of

GL2Fq is given by
(i) L(X) of 3.2.1 for X : Fq ---* C',
(ii)S(X)of 3.2.3for X:Fq-->C',
90
GL2Fq
(iii) R(X1, X2) = R(X2, X1) of 3.2.2 for any pair of distinct characters
X1, X2 : FF -' C'
and
(iv) r(O) = r(F'(O)) of 3.1.15 for any character 0 : Fq2 --> C" which
is distinct from its Frobenius conjugate, F'(O).
Proof Assuming that the representations in the list are distinct and
irreducible, the numbers of each type are q - 1, q - 1, (q - 1)(q - 2)/2
and (q2 - q)/2, respectively. Therefore, by 1.2.15, there can be no more
irreducibles since
(q - 1) - 12 + (q - 1) - q2 + ((q - 1)(q - 2)/2)(q + 1)2

+((q2 - q)/2)(q - 1)2
_ (q - 1)(q3 - q)
_ #(GL2Fq)
To see that the representations in the list are distinct and irreducible
we calculate the Schur inner product, using 1.2.40, Frobenius reciprocity
and the Bruhat decomposition (1.2.39 and 3.1.1)
< IndBLZF9(In.fc(XI 0 X2)),IndBLZF9(In.fc(X1 X2)) >GLZFq
=B
=B
= 1+ < InfB(Xi 0 X2),Indc(X2 X1) >B

= 1+ < ResB(I of B(Xi 0 X2)), X2 X1 >T
= 1+ < X1 0 X2, X2 x X1 >T
if X1 = X2,
otherwise.
This calculation, together with 3.1.16, shows that each representation
91
Table 3.3. (Theorem 3.2.5)

L(X)
R(xj, X2)
S(x)
r(O)
x(aI)
xI0)x2($)+X206)xI0)
x(cP)
II
x(a)2
XI(a)X2(a)
0
-O(a)
III
X(a)2
(q + I)XI(a)X20)
IV
X(N(x))
Type
9x(a)2
-X(N(x))
(q - 1)19(a)
-{O(x) + O(F(x))}
in the list is irreducible. A similar calculation shows that

< IndsLZF9(InfT(xi 0 x2)),IndBLZF9(InfB(A, 0 /12)) >GLZFq
=< Inf c(xi X2),I nfe(2i .i2) >B + < Xi X2,'2 AI >T
and this formula, together with dimension, enables one to distinguish

between the listed representation, which completes the proof.
Theorem 3.2.5 With the notation of Table 3.1 and 3.2.4 the character
values of the irreducible representations of GL2Fq are given by Table 3.3,
where N = NQ2/Fe denotes the norm.
Proof The character values of the Weil representation, r(O), were calculated in 3.1.21, 3.1.25 and 3.1.26. The character values of L(X) are trivial
to verify. From 1.2.43 we have the formula
#(H)_i
Trace(IndH(p)(g)) =
Trace(p(ygy-i))
yEG,ygy-' EH
by means of which to calculate the character-values of

I
ndBL2Fq
a
(I of (xl X2))
When X = Xi = X2 we may calculate the character-values of S(X) by

means of the relation
IndBL2F9(InfB(x x)) = S(x) L(x).
The representative for type I can only be conjugated into B by y E B

or y E Bw, the former type of conjugation preserving the diagonal entries
and the latter reversing them. The representative for type II can only be
conjugated into B by y E B, whereas that for type III is central in GL2Fq
and that for type IV cannot be conjugated into B at all, since it has a
92
GL2Fq
characteristic polynomial whose roots do not lie in Fq. The entries in the
table follow easily from these remarks and will be left to the reader.
3.2.6 Maximal pairs in ./GL2F9
As in 2.2.9-2.2.10 let J1IGL2F9 denote the poset of pairs (J, 0) with J <
GL2F9 and 0 : J -) C. From 2.3.15, if V E R(GL2Fq) and (J, 0) is

maximal in J/GL2Fq then the coefficient of (J, (/)GL2F9 in aGL2Fq(v) is given
by
{multiplicity of (J, O)GLZFq in aGL2Fq(v)}
_ [NGL2F9 (J, 4) : J] < 0, Rest L2F9 (V) >
3.2.7
Here
NGL2F9 (J, 0) = {X E GL2F9 I (XJX-1 (X-')'(4)) = (J, 4)}.
In other words, it is easy to calculate the multiplicity of maximal pairs,

(J 0)GL2Fq, in aGL2Fq(v). For this reason we will now introduce four types
of maximal pairs in .I#GL2Fq :
Type A:
(GL2Fq, X det) for x : FF -> C.
Type B:
(H, 2 p) where u : Fq -> C' is non-trivial and 2 : FF
C" is any homomorphism. Here
y:H-{(0
) EGL2Fq}-->Fq xFq
is as in 3.1.3.
(B,Inf'(21 ).2)), in the notation of 3.2.2, where 21,22
Type C:
Fq -* C. are distinct.
Type D:
(Fq21 p) where p and F* (p) are distinct.
Here we consider Fq2 to be the cyclic subgroup generated by the matrix

3.2.8
-xF(x)
x + F(x)
of Table 3.1 where x E Fq2 is a generator. Up to conjugation this subgroup
is independent of the choice of x. On this subgroup the Frobenius map,

F E G(Fg2/Fq), corresponds to conjugation by the matrix
3.2.9
f=
(1
x + F(x) )
-1
93
This is seen as follows. With respect to the Fq-basis, {l,x}, of Fq2

multiplication by x is represented by the matrix of 3.2.8. However
(1
x + F(x)) ( 0 -xF(x) ) (1 x + F(x)

x + F(x)
0
-1
-1
1
-xF(x) + (x + F(x))2)
x + F(x)
-1
-x - F(x)
(1
x + F(x)
-1
x + F(x) xF(x)
-1
'
which represents multiplication by F(x) with respect to this basis.

Proposition 3.2.10 Each of the pairs, (J, 0) E JKGLZFq, listed in types A-D
of 3.2.6, is maximal. In addition, for each of types A-D,
NGLzF9 (J, w) = J.
Proof The result is obvious for type A, (GL2Fq, X det).

From the classification of maximal subgroups of SL2Fq, which is given
in Dickson (1958, p. 286, section 262), it is straightforward to see that any
proper subgroup of GL2Fq which contains H must lie in B = NGL2FgH.
However,
z) (0 6) i
(0
6) (0
az
(0
_
- (0
ay + f3z
6z
a-1
)(
-a 138
6-1
azb-1
z
Hence the action of this matrix on H = Fq x Fq is to multiply the second
coordinate by a8-1. However and y(aS '-) are distinct unless a = 6,

which shows that NGL2F9(H,) ) <- H and proves the result for type B.
The result for type C is immediate from the facts that NGL2FgB = B
and that the conjugate, xvx 1 (x, v E B), has the same diagonal as v.
From the classification of maximal subgroups of SL2Fq (Dickson, 1958,
p. 286, section 262), one readily finds that NGL2FgF92 =< Fqzi f > where
f is as in 3.2.9. However, conjugation by f induces the Frobenius map
94
GL2Fq
of F 2 so that f 0 NGL2Fq (Fqz, P), which easily yields the result for type
D.
Suppose that g E GL2Fq and that (J, 4)) E #GL,F, is

(gJg-1,(g-1)`(4)))
one of the maximal pairs of type A-D, as in 3.2.10. If
= (J, 01) then = 1 in the case of types A and C. For type B, 0 = A Op
Corollary 3.2.11
and 01 = 2 (u -) for some u E FF while, for type D, 01 = or

1 = F'(cb) where F is the Frobenius of G(Fg2/Fq).
Proof We must have g E NGLZFgJ = GL2Fq, B, B or < Fq2, f > for types
A-D, respectively. From this observation the result follows easily from
the computations which were used in the proof of 3.2.10.
Definition 3.2.12
Suppose that v : GL2Fq ---> GL(V) is an irreducible
representation. We will write
3.2.13
GL2Fy
aGL2Fy(v)
+ L,s bs(B, I nJ T (2l,s 22,s))GL2F9
ec( gz,
Pr)GL2F9
+>udu(GL2Fq,Xu
det)GL2F,7
to signify that the multiplicities in aGLZFq(v) of the maximal pairs of types
A-D in 3.2.6 are as shown in 3.2.13 (the ellipsis denoting the sum of all
the terms of other types). In 3.2.13 the sums over r, s, t, u are taken over
all the terms of types A,B,C,D respectively.
Theorem 3.2.14 With the notation of 3.2.4 and 3.2.13
(i)
aGL2F9(L(X)) = (GL2Fq,X . det)GL,F9.
aGL2Fq(R(Xl, X2))
= E1#(H, XiX2
95
N)GL2Fq
+(B,InfB(X1 X2))GL2Fq
+(B, I of B(X2 X1))GL2Fq
F'
z
ResF; (P)=XI X2
(F'2'
p)GL2F,,
9
+
aGL2Fq(S(X)) = E1*(H,X2 (&
p)GL2Fq
+(B, I of B (X (D
+E,
X))GL2Fq
(g2,
F'2
p)GL2Fq
ResF; (p)=X2
+..
aGL2Fq(r())
= EvEFq(H,ResF92(0) 'Fq(v - -))GL2F9
+E
(Fq2, p)GL2Fq
F',
F'2
p${O,F'(O)},ResF;` (p)=ResF; (a)
v
+..
Proof Part (i) follows from 2.3.2(ii).
Parts (ii) and (iii) are similar and therefore we will only prove part (ii).
By 3.2.7 and 3.2.10, the multiplicity of a term of type B from 3.2.6 in
aGL2Fq (R(X1, X2)) is equal to
<2,ResH L2F
I(IndBL2Fq(IofB(X1
N X2))) >
E
ZEH\GL2Fq/B
I ndHnzBz-l (ResHnzBz-. ((z1
n f B(X 1 X2)))) >
96
GL2Fq
=E
z=1,w
I ndHnzB
1(ResHnzBz-1((z-1) (1 of s
(X1 x2)))) >
by the Bruhat decomposition,
=<A,(x1x21)+IndF (X20X1)>
=<A,X1X2IndF (1)>
1
if 2=X1X2,
otherwise.
This accounts for the first part of the formula in part (ii).
By 3.2.7 and 3.2.10, the multiplicity of a term of type C from 3.2.6 in
aG 2Fq(R(X1,X2)) is equal to
12),ResBL,F4(IndBL2F9(InfB(X1
<InfB(j1
xz))) >
=<InfB(Al A2),Infq(xi x2)+IndB(x2 x!) >

_< Al, X1 >< 22,X2 > + < A2, X1 >< 21,x2 >
1
if {21, 22} = {Xl, X21
otherwise.
This accounts for the remaining part of the formula in part (ii).
Clearly there are no terms of type A in aGL2F9 (R(X1, X2)) and the
multiplicity of a term of type D is equal to
L2F
< p, ResF;
92
(R(x1, x2)) >
+(q2 - 1)-1 ExEF;(q + 1)7'(X)X1(X)X2(X)

F'z
=< ResFQ (p), X1X2 >
97
by the character values of 3.2.5. This completes the proof of part (ii).
For part (iv) we observe that, since r(O) is irreducible, there can be no
terms of type A. Also there can be no terms of type C, since
<
A2),IndH(O 'YF,) >
_< 21A2 1, 0 'FFq >
=0.
By 3.2.7 and 3.2.10, the multiplicity of a term of type B from 3.2.6 in
aGLZFq (r((9)) is equal to
< A , Res/L2F9 (r(O)) >
=< A It, ResH(I ndH(0 0'Y)) >

`Y) >
= EzEB/H < A ,
F'2
=EvEFF <A,ResFg (O)PFg(v.-)>,

which accounts for the first part of the formula in part (iv).
Finally, by 3.2.7 and 3.2.10, the multiplicity of a term of type D from
3.2.6 in aGL2Fg(r(O)) is equal to
< p, ResFL2Fg (r(O)) >
9
_ (q2 - l) -I
P(x)(q - 1)0(x)
xEFF
-(q2
1)_i
p(x)(O(x) + O(F(x)))
xEFg2-Fq
by 1.2.7 and 3.2.5
_ (q +
1)(q2
- 1)-i
p(x)O(x)- < p, 0 > - < p, F'(0) >

xEFF
=< ResF.2 (p), ResFt2 (O)
> - < p, O > - < p, F"(O) >,
which accounts for the remaining terms in the formula for part (iv) and
completes the proof.
98
GL2Fq
Corollary 3.2.15 The irreducible representations, v, of GL2Fq are uniquely

characterised by the terms of types A-D, in the terminology of 3.2.6, which
occur in the Explicit Brauer Induction formula
aGLZF,(v) E R+(GL2Fq).
These (maximal) terms are given by the formulae of 3.2.14.
Proof This follows easily by inspection of the formulae of 3.2.14. For

example, the Weil representations are the only ones for which a term
of type D appears. The type B terms in aGLZF9(r(O)) determine the
sum over which the type D terms are taken and the characters O and
F"(O) are characterised by being the only two characters on Fq2 with the
prescribed restriction to Fq which do not appear in the sum. Of course,
r(O) = r(F*(O)).
3.2.16 The Shintani correspondence for GL2Fq
Let E E G(Fgn/Fq) denote the Frobenius transformation. Shintani

(1976) discovered a remarkable one-one correspondence of the following
form :
{irreducible representations, v, of
GLmFgn fixed under E}
3.2.17
I Sh
{irreducible representations, Sh(v), of GLmFq}.
In 3.2.17 the Frobenius, E, acts via its action upon the entries of a
matrix. This correspondence, which was also treated by Shintani for GL2
of a local field, is also sometimes called Shintani descent or lifting (see
Gerardin & Labesse, 1979, for example).
The correspondence of 3.2.17 may be characterised by means of the
Shintani norm. For X E GLmFgn define
3.2.18
N(X) = En-1(X)En-2(X)... E(X)X.
Although N(X) lies in GLmFq., its conjugacy class contains a unique

GLmFq-conjugacy class, which depends only on the conjugacy class of X.
This gives a meaning to the equation

3.2.19
99
Trace(Sh(v)(N(X))) = Trace(v(X)).
The correspondence of 3.2.17 is characterised by the fact 3.2.19 holds for

all X E GL,,,Fgn.
When m = 1 this correspondence is a consequence of Hilbert's Theorem
90, which states that H1(G(L/K); L') = 0. When L/K is an extension of

finite fields we obtain an exact sequence of the form
3.2.20
If V : F9 - + C' satisfies v = E'(v) then, by 3.2.20, there exists a unique
Sh(v) : Fq -* C' such that

3.2.21
Sh(v)(N(x)) = v(x) (x E FF.).
We shall now use Explicit Brauer Induction and 3.2.14-3.2.15

to describe a correspondence of the type of 3.2.17 in the modest circumstances of GL2Fq.. As it happens, our correspondence will coincide with
that of 3.2.16, although no mention of the Shintani norm appears in our
construction. My correspondence will be effected by applying Hilbert's
Theorem 90, in the sense of 3.2.20-3.2.21, to the maximal one-dimensional
3.2.22
characters which appear in the formula for aGLZF I (v).
We begin by observing that, if v is irreducible and E'(v) = v, then

aGLZFq.(Y-*(V)) =Y-*(aGL2Fgn(v)),
where E' : R+(GL2Fgn) - -> R+(GL2Fgn) is given by the formula

E*(J, 4)GL2Fgn
= (E(J),
0(y-1
. _))GL2Fgn
Since E'(v) is also irreducible the maximal terms of types A-D in

aGL2Fg,(E*(v)) will be obtained by applying E' to the maximal terms of
type A-D in
Now let us describe our construction of the correspondence, which will
be denoted by T.
If E'(L(X)) = L(X) then
X and there exists a unique
Fy --->
C' such that x(z) = X(N(z)) where N is the norm. In this case we set
GL2Fq
100
3.2.23
1(L(X)) = L(X).
Next suppose that E'(R(Xi, X2)) = R(X1, X2) then, in 3.2.14(ii),

E(H, X1X2 )GL2Fq = 1:(H, E*(XIX2) (&p)GL2Fq
1#
and
(B, I of B (x1 X2
))GL2Fq
+ (B, I of B (x2 (g X
1))GL2Fq
_ (B,InfB(E*(Xi)
(B,IofB(E*(x2)
E+(X1)))GL2Fq
By 3.2.11, these equations imply that either

(a)
X1 = E'(X1) and X2 = E'(X2)
or
(b)
X1= E"(X2), X2 = E"(x1) and X1X2 = E'(x1x2)
In case (a) there exist unique homomorphisms, Xi : FF --> C' (i = 1,2)

such that Xi(z) = Xi(N(z)) for each i = 1, 2. In this case we set
3.2.24
T(R(X1, X2)) = R(71, X2).
In case (b) we have a surjective homomorphism
Z/n
A:
111
given by 2(g) = (-1)i-1 if g(XI) = Xi. Hence n = 2d, Ker(2) =

and each Xi is fixed by Ker(2). Hence there exists a unique X1 : F92 -+ C
such that X, (z) = X1(N(z)), where N : F9 -> F92 is the norm. Also, if F
generates G(Fg2/Fq) and X2 = F*(X1) then X2(z) = 72(N(z)).
Notice also that, in case (b), E*(XIX2) = X1X2 so that there exists a
unique 71,2 : F9 --> C' such that xi(z)X2(z) = 11,2(N(z)). In fact, we have
Fy
Fy
ResF2 (X 1) = 71,2 = ResF92 (72)q
101
For, if w E F9, v e 19, and r E Fy" satisfy N(v) = w, N(r) = v then

71(w) = X1(vF(v))
= 71(v)Xl(F(v))
= X1(V)X2(v)
= X1(N(r))X2(N(r))
= X1(r)X2(r)
= X1,2(N(r))
= X1,2(w)-
From these characters it is natural to form

(H, 71,2 (D TFq (v . -))GL2Fq
+E
F'2
P)GL2Fq
(Fq2,
P{X1,X2},ResF; (P)=7112
9
+
in R+(GL2Fq).
These are the maximal terms of types A-D in aGL2Fq(r(xl)) and therefore, in case (b), we set
3.2.25
T(R(X1,X2)) = r(X1) = r(72)-
If E'(S(X)) = S(X) then, as in case (a) above, we see that E`(X) = X and
F4 -> C' such that X(z) = X(N(z)). In this
that there exists a unique
case we set
r(S(X)) = S(X).
3.2.26
Finally, suppose that E*(r(O)) = r(O). Hence, by 3.2.14(iv),
EFq" (H,
ResFg2"
(E (O)) 'I`Fq" (v
_ vEFq (H, ResFg2" (O) 'YFq" (v
. _))GL2Fq"
))GL2Fq"
102
GL2Fq
and
(Fg2n1+(P))GL2Fgn
F'2n
F'2n
(p)=ResF9 (O)
I
_
=
F-+
(gzn , P)
GL2Fgn
2'(p)=ResF^"(0)
9
F'
The first equation implies that ResF9 (E'(8)) = Res
F'92"
(O) and so there
exists a unique O : F9 - C+ such that, for all z E F9., O(z) = O(N(z)),

where N : F9n --> FF is the norm. However, E*(O) and O must be distinct
on Fq2,, since F E G(Fg2n/Fqn) acts non-trivially on O, by assumption.
The second equation shows that E+ permutes the set
{p 4 {O, F+(O)}, ResF 2n (p) = ResF:n" (O)}
q"
so that we must have E'(O) = F'(O). Since G(Fg2n/Fq) = Z/2n and

F = In we have
3.2.27
(En-1)'(O) = O.
Since O is not Galois invariant < En-1 > must be a proper subgroup
of < E >. However, HCF(n - 1, 2n) E { 1, 2} so that we must have
HCF(n - 1, 2n) = 2 and therefore n must be odd . This means that
3.2.28
Z /n =< Fn-1 >= G(Fg2n /Fq2 )
and, by 3.2.27, there exists a unique O : Fq2 - C' such that O(w) _
O(N(w)) for all w E F. If z E F9 and S E F9n satisfy N(s) = z then
ResF92 (O)(z)
= ResF92 ()(NF/F(S))
= O(NFg2n/Fg2(S))
= Res
(19)(s)
= O(NFgn/Fq(S))
= O(Z),
so that
ResF Z (O)
103
= O. From these characters it is natural to form

EVEFF(H,6 `YF,(v . -))GL2Fq
F'
(FF21 p)GL2F9
p{O,F'(O)},ResF; (p)=0
v
+ ..
These are the maximal terms of types A-D in aGLZFq(r(O)) and therefore
we set
3.2.29
T(r(E))) = r().
Each of these recipes is reversible and one easily sees that the process
yields a one-one correspondence similar to that of 3.2.17. The discussion
of 3.2.22 may be summarised as follows:
Theorem 3.2.30
The yoga of 3.2.22 yields a one-one correspondence of
the form
{irreducible representations, v, of
GL2Fgn fixed under E}
3.2.31
Ir
{irreducible representations,T(v), of GL2Fq}.

In fact, T coincides with the Shintani correspondence as described in
Shintani (1976, p. 410, section 4).
Proof The fact that T satisfies the characterisation of Sh which is given

by 3.2.21 is easily verified by means of the table of character-values in
3.2.5.
It would be very interesting to develop for GLFq

a yoga similar to that which is given in 3.2.22 for GL2Fq. In such an
enterprise one would have to determine suitable generalisations of types
A-D of 3.2.6. In this example the types were arrived at by considering
Remark 3.2.32
first the maximal abelian pairs and then, should they prove not to be selfnormalising, their normalisers. In the case of GL2Fq what we have given
is merely a calculation and in general one would wish for a more intrinsic
104
GL2Fq
proof; preferably one which, in the presence of a suitable Explicit Brauer
Induction technique, would extend to the case of GL F where F is a

local field.
3.3 Exercises
3.3.1
Let F be any field. Suppose that
A=
('
d )EGL2F.
Verify that
A
-(0
)w
u
1
where
b = -c, u = d/c, (3 = -a and a = -det(A)/c.

3.3.2
Use 3.3.1 to prove 3.1.2.
3.3.3
Suppose that A is a finite, abelian group and that x : A -> C'
is a non-trivial homomorphism. Show that
E x(a) = 0.
aEA
3.3.4
Verify 3.1.5.
Verify directly from the formulae of 3.1.15 that they are compatible
with those of 3.1.9 in the sense that
3.3.5
(r() (
6f (x) = O(f)f (xF (fl)-i),
where NFgz/Fq(/3) = b. ( Use the identity
h(x) = f(xF(f)-')
if f E; x, /3 E Fq2 and h(x) = f (/3x). )
Establish the irreducibility of the Weil representation, r(O), by
means of its character function and the Schur inner product on GL2Fq.
3.3.6
3.3 Exercises
105
For each partition . # = (mi, ... , m,) of m = >; m; let U t denote

the subgroup of GL,,,Fq consisting of upper triangular matrices which
have the identity matrices,
block-wise down the diagonal. A
representation
3.3.7
p:GL,,,Fq-->GL(V)
is called cuspidal if, for each partition,
Prove that the Weil representation r(O), of 3.1.15 is cuspidal.

3.3.8
Complete the proof of Theorem 3.2.5.
3.3.9
Let r(O) : GL2Fq --> U(q - 1) denote a Weil representation, as
in 3.1.15.
Show that the fixed-point set of Reser2F9(r(O)),
(U(q - I)INTq-i)B
consists of a single point. (See 2.1.10.)
Give a description of a correspondence of the

Shintani type for GL,,,Fgn using the maximal terms in the Explicit Brauer
Induction formula together with Hilbert's Theorem 90, generalising the
method of 3.2.22.
3.3.10 (Research problem)
4
Introduction
Section 1 shows how Adams operations, 1 k, behave with respect to Explicit Brauer Induction. In particular it is proved that one may express
W
k( V)
as an integral linear combination of monomial representations
(i.e. induced from one-dimensional characters of subgroups) by applying

k
to each one-dimensional subhomomorphism in the Explicit Brauer
1P
Induction formula for V and then mapping the result to the representation ring, R(G). This result holds for all the Explicit Brauer Induction
formulae, since it depends mainly on the naturality property. The effect
of this result is to give one a form of Brauer's theorem which `commutes
with Adams operations'. This result, which was first proved in Snaith
(1989a) using the results of Snaith (1988b), is very convenient and rather
unexpected and the remainder of this chapter consists of implications of
this result.
In Section 2 we describe the adelic Hom-description of Frohlich, which
gives the class-group of an integral group-ring of a finite group in terms
of groups of Galois-equivariant functions from R(G) to the ideles of

a suitably large number field. Those who are familiar with algebraic
K-theory will recognise the Hom-description as being equivalent to the
exact K-theory sequence (at dimension zero) which was first obtained by
C.T.C. Wall by applying algebraic K-theory to the canonical adelic fibre
square of group-rings. As an example, the Swan modules are introduced
and their classes in the class-group are expressed in terms of the Homdescription and from this we prove the well-known result that for cyclic
and dihedral groups the class of a Swan module is trivial.
Section 3 is devoted to a proof of a conjecture of Martin Taylor, which
was posed in (Taylor, 1978). The Hom-description gives the class-group
106
4.1 Adams operations and rationality
107
as a complicated quotient of the weak product of the groups of Galoisequivariant maps from R(G) to all the local completions of a large number
field. The quotient involves factoring out by two subgroups of which one
is the group of determinantal functions. For the group functions into the
units of an 1-adic local field one may construct endomorphisms, denoted
by (FW'/1), by means of the Adams operations. Taylor conjectured that
for 1-groups the result of this endomorphism on determinants would be
congruent to one modulo 1. Using the result of Section 1 it is easy to

deduce these congruences for any group, G. From the basic congruences
we proceed to deduce higher order congruences, for all G.
In Section 4 the higher order determinantal congruences are used to
construct new maps out of the class-group of an integral group-ring. In
the case of 1-groups, M.J. Taylor (1978) had previously manufactured
such maps by means of some rather weaker congruences and we give his
determination of the Swan subgroup of an 1-group. This calculation uses
the determinantal congruence maps together with the fact that the Artin
exponent of G annihilates the Swan subgroup, T(G).
One of the three main steps in M.J. Taylor's proof of the tame Frohlich
conjecture (Frohlich, 1983; Taylor, 1981) is the tame Galois descent for
the determinantal functions. The original proof involved the difficult
construction of the group-ring logarithm (independently discovered by
R. Oliver). In Section 5 we give the simplified proof which is available
once one has the determinantal congruences, for then the mysterious,
non-commutative group-ring logarithm becomes merely log(FW'/1) (see
4.5.30), where log is the classical l-adic logarithm.
In Section 6 we prove that Adams operations preserve the subgroup
of determinantal homomorphisms. Once again the determinantal congruences enable one to simplify the original proof, which is due to Ph.
Cassou-Nogues and M.J. Taylor. Our proof is similar to the original in
that we reduce to special types of groups, but it is significantly different in
that nowhere do we mention the decomposition homomorphism, which
is used in an essential manner in the proof which is to be found in Taylor
(1984).
Section 7 consists of a collection of exercises which are centred about

the class-group of a group-ring.

Definition 4.1.1
Let p : G -> GL(V) be a complex representation of
108
a finite group with character, Xp. Let k >_ 0 be an integer. The Adams
operation, Wk, is defined by the character formula
4.1.2
for all g E G.
Xwk(p)(g) = Xp(gk)
Lemma 4.1.3 (i) The Adams operation defines a natural ring homomorphism
Wk : R(G) ---> R(G).
(ii) For p as in 4.1.1

Wk(p) = Nk(AI(p),...,Adim(p)(p))
where Nk is the kth Newton polynomial and 2'(p) is the ith exterior power
(cf. 1.1.7).
(iii) If dimC(p) = 1 then Wk(p) = pk
Proof Recall that the Newton polynomial is defined in the following

manner. Let t1,.. . , tm denote indeterminates and let a i(tj, ... , tm) = CA)
be the ith elementary symmetric function, which is equal to the sum of
all products of i-tuples of distinct indeterminates. In Z[tl,..., tm] we have
the identity
4.1.4
Em 1 tk = Nk(o (t),...,6k(t))
In order to prove part (ii) it suffices to calculate the character of

Nk(21(p)
2dim(p)(p))
at g e G. However, since < g > is cyclic

G
Res<g>(p)
= LI ... a) L,,
where the {Li} are one-dimensional and n = dimC(p). Also,

A' (L1 ... Ln) =
Ln) E R(< g >),
so that 4.1.4 implies, in R(< g >), that

Adim(p)(P)))
= Li ... Ln,
where L;` = Li ... Li (k copies). Therefore the value of the character

of Nk(A1(p),,Adim(p)(p))
l /n at g E G is equal to
i=1 XLk(g)
= Ei=1
n XLi(S) k
(gk)
Ein=1
XL;
= Wk(g),
109
Once we know, from part (ii), that 1 k(p) E R(G) then parts (i) and (iii)
follow immediately by easy character calculations.
O
Remark 4.1.5 (cf. 4.7.1) In general, Wk(I ndG (p)) and I ndG (Wk(p)) are
not equal. For example, take G = Z/p, H = {1{, p = 1 and k = p. In this
case Ind,,, (WP(1)) = Ind2}P(1) is the regular representation. However,
since gP = 1 for all g E Z/p , p"(Ind2}P(1)) = p E R(Z/p).

However, the proof of the following result, which was originally proved
in Theorem 2.33 of Snaith (1989a), shows that ipk behaves well with
respect to Explicit Brauer Induction.
Theorem 4.1.6 Let p : G -> GL(V) be a representation of a finite group
over any algebraically closed field, K, of characteristic zero. Then, for
all k > 0, there exist integers, ni, and one-dimensional representations of
subgroups, 4; : H; -o K', such that
n;IndGHi (0k)
i E RK(G),
wk(P) _
where RK(G) is the ring of K-representations of G.
Proof In this situation RK(G) = R(G) and therefore we may assume that
K = C, the complex numbers. Define a homomorphism
upk
: R+(G) -* R+(G)
by `Yk((H, )G) = (H,

The homomorphism, Y' k, is a natural map
with respect to the restriction to subgroups. Therefore the composition
q5k)G.
R(G)
R+(G)
R+ (G)
bG)
R(G)
is a natural homomorphism. Thus, in order to evaluate this composite, it

suffices to evaluate the character of bG(Tk(aG(p))) at g E G. By naturality
this is the same as the character-value of
b<g>(`Vk(a<g>(Res<g>(P))))
at g. Suppose that Res<g>(p) = E'=, Lj where dimC(Li) = 1, then the

character-value is given by
j=i b<g>(`yk(a<g>(Lj)))(g) =
b<g>(``k(< g >,L>)<g>)(g)
= Ei=1 L;(g)
= ipk(Res<g>(P))(g)
110
Therefore, by naturality of 1 k, bo(`l'k(ao(p))) = Wk(p) E R(G), which

proves the result if ao(p) = E; ni(H,, ,)o E R+(G).
4.1.7 Adams operations and rationality
Let K be a subfield of the complex numbers. Denote by RK(G) the

representation ring of G which is generated, in R(G), by representations
of the form
p : G ---p GL (K ).
4.1.8
RK(G) is a subring of R(G).

We say that a representation of G is K-rational if its class in R(G) lies in
RK (G). Let N be the least common multiple of the orders of the elements
of G. Hence, if bN = exp(2ni/N), any one-dimensional representation,

4): H --> C', is WO-rational and therefore so is IndH(4)).
Theorem 4.1.9 (i) With the notation introduced above
RQ(N)(G) = R(G).
(ii) If n = HCF(k, N) and m = N/n then

Wk(R(G))
Proof Part (i) follows from part (ii) by setting k = 1. Part (ii) is a
corollary of Theorem 4.1.6. For, if 4); : Hi --> C` is a one-dimensional
character, then the image of k lies in the roots of unity of Q(ckv).

Therefore 4);(g)k is a power of m and therefore each term, In4.(4 ) in
Theorem 4.1.6, is Q(lm)-rational, which completes the proof.
4.2 Describing the class-group by representations

4.2.1 Adeles and ideles
Let K be an algebraic number field; that is, a field extension of finite

degree, K /Q. Denote by (9K the ring of algebraic integers of K (Lang,
1970). By a place of K we will mean an embedding of K as a dense
subfield into a complete discrete valuation field, K. For example, k might
be isomorphic to R or C, the real or complex numbers, respectively.

Embeddings into such fields are called infinite or Archimedean places of
K. The finite places of K are comprised of all the non-Archimedean
111
ones and they all come about by choosing a prime ideal, P a (K, and
completing K in the P-adic topology. This is the topology in which
{x+P';n >_ 0} is a base of neighbourhoods of the element, x, in (9K. We
denote this field by Kp.
Example 4.2.2 If p is a rational prime then the p-adic completion of Z,
the integers, is given by
ZP = lira Z/(pn)
n
and is called the ring of p-adic integers. Its field of fractions is given by
QP = z,[1/p]
4.2.3
The completed fields introduced in 4.2.1 are called local fields. If
P a (9K is a prime such that the integral ideal, P n Z, equals pmZ for
a rational prime, p, then Kp/Qp is a finite extension and we say that P
divides p or that P lies over p. This is equivalent to the existence of a
commutative square of the following form:
Q - K
QP
Kp
It is customary to refer to `places' as `primes' and vice versa. In which

case, to say that one Archimedean prime divides another means that
there is a corresponding commutative square
in which the vertical maps are infinite places and L/K is an extension of
number fields. If P is a finite prime then the algebraic integers of Kp,
(9Kp, are given by the P-adic completion of (9K. By convention, if k is
Archimedean we set (9K equal to K.
112
The adele ring of K is defined to be the ring given by the restricted

product
J(K) = 11PprimeKP
4.2.4
In 4.2.4 fl' signifies that we take those elements of the topological ring,
flP Kp, for whom almost all entries lie in (9KP. The group of ideles is the
group of units in J(K),
4.2.5 J' (K) = {(xp) E J(K) I xp # 0 and almost everywhere xp E CKP},
where (9KP denotes the multiplicative group of units in (9KP. The unit
ideles is the subgroup
4.2.6
U((9K) = 11Pprime (9KP.
Now let G be a finite group. We may extend the adeles and ideles to
the group-rings, OK [G] and K [G]. Define
4.2.7
J(K [GI)
J*(K [G])
= [T' prime Kp [G],
U(&K [G])
11Pprime OKp
_ {(ap) E J(K [G]) I aP E (9KP [G]' for

almost all P and ap E Kp [G]` otherwise},
[G]*.
Now suppose that E/K is a finite Galois extension with Galois group
G(E/K). In this case G(E/K) acts on the set of primes of E and
hence acts upon the groups J'(E), U(&E),J*(E[G]) and UWE[G]). If Q
is a prime of E which divides the prime, P, of K then G(Eg/KP) is a

subgroup of G(E/K) which is called a decomposition group for P and
depends only on P, up to conjugation in G(E/K).
If E is large enough to contain all #(G)th roots of unity we call E a
splitting field for G. In this case, by 4.1.9(i), there is an isomorphism of
the form
4.2.8
RE(G) = R(G)
and G(E/K) acts upon R(G) by the entry-by-entry action on a representation

4.2.9
T : G ----> GL (E).
Therefore we may consider the group of G(E/K)-equivariant maps

4.2.10
113
HomG(E/K)(R(G),J*(E)) = {f : R(G) -* J' (E) I f(g(z))
= g(f(z)) for all g e G(E/K),z c R(G)}.

More generally, if L/K is a Galois extension which contains E/K then
G(L/K) acts on R(G) and on J"(E), since E/K is Galois and G(E/K)
is a quotient of G(L/K). Therefore we may pass to the absolute Galois
group, f1K, which is the topological group defined by
4.2.11
OK = invlim K.L1K, G(L/K),

L/K Gal ois
where K`/K is a chosen (algebraic) algebraic closure of K. In this case

we have
4.2.12
HomoK(R(G),J'(E)) = HomG(E/K)(R(G),J`(E)).
4.2.13 (9K [G] -modules and determinants
A more extensive reference for the material of this section is Curtis &
Reiner (1987, p. 334 et seq.).
Let M be an (9K [G] -module of rank one which is locally free. This
means that M cK (9K, is a free (9K, [G] -module on one generator, xp,
for each prime, P, of K and that M oK K is a free K [G] -module on one
generator, x0. Since K [G] and (9K, [G] are subrings of Kp [G] this means
that there is a unit, 2p E Kp [G] * which is defined by
4.2.14
In fact, ),p will almost always lie in OK, [G]' so that we obtain an idele
4.2.15
(2p) E J*(K[G]).
Now suppose that T is a representation, as in 4.2.9. We may apply T

to each 2p to obtain
4.2.16
T(2p) E
M (A) denotes the n x n matrices with entries in A. There is a ring

isomorphism of the form
4.2.17
Kp K E = fl
Q,P
QprimeofE
Therefore we obtain an element
EQ.
114
det(T(2p)) E fj
4.2.18
,,P
QprimeofE
E.
Since (k.) E J*(K[G]) we obtain an OK-equivariant map, given by

4.2.18 at the primes of E which divide P,
Det((2p)) E HomoK(R(G),J'(E)).
4.2.19
Now let us consider the dependence of 4.2.19 upon the choices of xo

and xp in 4.2.14. If we replace xp by another generator, xP, these choices
will be related by an equation
x' = upxp for some up E (9Kp [G]*
4.2.20
so that we obtain a unit idele

u = (up) E U((9K [G])
4.2.21
and 4.2.19 will be altered by multiplication by

4.2.22
Det(u) E Det(U((9K[G])) c Homo,(R(G),J*(E)).
Also there is a diagonal embedding of E* into J*(E) which induces an

inclusion
4.2.23
HomnK (R(G), E') c HomnK (R(G), J' (E )).
By a similar argument, changing xo to xo will change Det((2p)) by

a function which lies in the subgroup of 4.2.23. Therefore we have
associated to each locally free (9K [G] -module of rank one, M, a welldefined element
4.2.24
Det[M] E
HomnK
[G]))'
Now let us recall the definition of the class-group, ''.((9K [G]). This
is defined to be the Grothendieck group of finitely generated, locally free
(9K [G]-modules. To be precise, consider the set of isomorphism classes
of finitely generated, locally free (9K[G]-modules, Mod.l.f.((9K[G]). This
set is a monoid if we endow it with an addition operation defined by
4.2.25
[M] + [N] = [M N],
where [-] denotes an isomorphism class. Define an equivalence relation

on Mod.l.f.((9K[G]) which is generated by two types of relations
115
4.2.26
(i) [M] - [N] if for some m, n

M((K[G])m =N ED WK [G])"
and
(ii) [A] + [C] - [B] if there exists an exact sequence of (9K [G] -modules
With this notation

W2'(&K[G]) = {Mod.l.f.((9K[G])}/ -.
4.2.27
It is known that 1((9K [G]) is generated by locally free OK [G] -modules

of rank one; the connection with 4.2.24 is the following Hom-description,
due to Frohlich.
Theorem 4.2.28 (Curtis & Reiner, 1987, p. 334; Frohlich, 1983) With the
notation introduced above there is an isomorphism
Det :
Homo,, (R(G),P(E))
OK[G]) -> HomnK (R(G), E') Det(U((9K [G] ))
which sends a locally free module of rank one, M, to Det[M] of 4.2.24.
We will not prove this theorem and for our purposes the reader may
take the group of 4.2.24 as the definition of the class-group.
4.2.29 The kernel group, D((9K [G])
An (9K-order, A in K [G], is a subring containing (9K, which is a finitely

generated, projective (9K-module such that K aK A = K [G]. Suppose
that A is a maximal (9K-order of K [G], then we may define the kernel
group
4.2.30
D(&K [G]) = ker(,W'((9K [G]) ---> `RY(A)),
where '2'(A) is the class-group of A, defined by a Grothendieck group

construction analogous to that of 4.2.27. The group, D((9K [G]), defined
in this manner, is independent of the choice of A. There is also a

Horn-description of this group.
To describe this we need to introduce a subgroup
4.2.31
Hom,K (R(G), (9E) c HomnK (R(G), (9E ),
116
An irreducible representation, T : G -> GL (E), as in 4.2.19 gives rise to

a complex representation, T : G -* GL (C), by choosing an embedding
of E into C. Let H denote the skewfield given by the quaternions. A

(left) H-vector space, V, of dimension m over H, may be considered
as a 2m-dimensional complex vector space. If GLm(H) is the group of
invertible H-linear maps from V to itself then we obtain a map
4.2.32
c : GLm(H)
> GL2m(C).
The representation, T, is called quaternionic or symplectic if T : G -i

GL2m(C) factors through the map, c, of 4.2.32. On the other hand, if T is
symplectic then, by 4.7.4, complex conjugation fixes T in R(G) so that if
f e HomnK (R(G), E*) and if K is a subfield of the real numbers, R, then
f (T) lies in R for every Archimedean prime of E which divides K c R.
Therefore it makes sense to define HomnK (R(G), E*) to be
t f E HomnK (R(G), E*) I f (T) is positive, T symplectic},
4.2.33
where, in 4.2.33, positive means that f (T) is positive under every Archimedean place of E which lies over a real place of K.
Similarly we may define
4.2.34
HomoK(R(G),(9E) _ {f E Hom+K(R(G), E *) I im(f)
(9E}
Theorem 4.2.35 (Curtis & Reiner, 1987, p. 334 et seq.; Frdhlich, 1983)
The isomorphism of 4.2.28 induces an isomorphism
Det : D(C9K [GI)
Remark 4.2.36
HomnK (R(G), U((OE))

)
HomnK (R(G), (9E) Det(U((9x [G]))
By 4.7.3, (9EQ [G] is a maximal (9E,,, -order in EQ [G] for all
primes, Q, which do not divide the order of G. Hence one may rewrite
Theorem 4.2.35 as an isomorphism of the following form:
4.2.37
Det : D((9K [G]) ->
HomcK(R(G), [IQI#(G) (9EQ)

HP1#(G) (9K, [G]*)
Hom+
R( G), (9*)
Det(111
E
OK(
Next we will work towards the construction of new maps out of

D(Z[G]) by means of 4.2.37. In preparation, let us consider the action of
4.2.38
117
52K and G(E/K) on U((9E) = UQprime (9EQ The action is induced by the
action of g E G(E/K) on KK E by the formula
4.2.39
g(a b) = a g(b)
(a E KP, b E E).
Via the isomorphism of 4.2.17 the action of G(E/K) permutes the

primes, Q, of E which divide P and G(EQ/KP) is the subgroup of G(E/K)
which preserves the factor, EQ, in 4.2.17. From this discussion we see
that, if f E HomG(EIK)(R(G), UQlP EQ), then taking the G(EQ/Kp)-map
given by the EQ-component gives an isomorphism
4.2.40
Homc(E/K)(R(G), fQIP EQ) =+ HomnKP (R(G), EQ).
On the level of ideles we have an isomorphism of rings (Curtis &

Reiner, 1987, p. 331)
4.2.41
(9KP OK C9E = fIQIP (9EQ
and a corresponding isomorphism

4.2.42
HomcK (R(G), IIQIP (P ) - Homo,,, (R(G), (9EQ ).
We will make use of 4.2.42 in the next section.

Example 4.2.43 Swan modules
Let G be a finite group of order n. Let k be any integer which is coprime

to n and denote by S(k) the Z[G]-module which is given by the ideal
4.2.44
S(k) = ideal {k, a } a Z[G],
where Q = EgEG g. The Z[G]-module of 4.2.44 will be called a Swan

module. Clearly S(k) is free when k = 1. In general, S(k) is a locally
free (and hence projective) Z[G]-module which is often non-trivial in
W2'(Z [G] ).
To see that S(k) is locally free we give another description of S(k). Let
4.2.45
x = (1 - (6/n)) + k (a/n) E Q[G]*.
In 4.2.45 x is a central unit of Q[G] since a is central and
(1 + fl a)(1 - (a/n) + k(a/n))
= 1 +a(-1/n+k/n+/ik)
=1,if/3k=(1-k)/n.
118
Therefore, within Q[G], we have the following isomorphisms of Z[G]modules:
{k, a} = (x/k)
{k, a}
{x,(xa)/k}
{x, a}
since (xa)/k = a2/n = a.
Within Qp[G] the Z[G]-module, {x,a}, is equal to Zp[G] < 1 > if

p X#(G), since in that case x E ZP[G]'. However, if pl#(G) then {x,a} is
equal to Zp[G] < x >, since (xa)/k = a2/n = a. Similarly, at Archimedean primes {x, a} is free with generator given by 1. Therefore S(k)
is locally free of rank one and the idele associated to such a module in
4.2.15 is equal, in J'(E), to
4.2.46
2Q =
{ 1 at Archimedean primes or at Q X#(G),
lx
at finite primes Q I #(G).
Now let T : G --> GL (E) be an irreducible representation. If T # 1

and Q I #(G) then
det(T(2Q)) = det(I + ((k - 1)/n)
T(g)).
gEG
However, by Schur's lemma (1.2.5), EgEG T (g) =,u I, and
n = Trace(>2gEG T(g))
= >gEG T race(T (g))
0,
so that det(T(2Q)) = 1 if T # 1. If T = 1 then det(T(AQ)) = k so that

we have proved the following result:
Proposition 4.2.47 In terms of the Hom-description of 4.2.28, S(k) is represented by the element of HomnQ(R(G),J'(E)) given by
1
S(k)(T) =
k
1
if T is irreducible, T # 1, at any prime,

if T = 1, at finite primes, Q I #(G),
if T = 1 otherwise.
In particular, by 4.2.35, S(k) E D(Z[G]).

Proposition 4.2.48 Let G = C,,, Den be the cyclic group of order n or the
119
dihedral group of order 2n, respectively. If HCF(k, #(G)) = 1 then S(k) is
trivial in '2(Z[G]).
Proof We will prove this by means of the Hom-description of D(Z[G])
of 4.2.35 in the form 4.2.37.
Consider first the case when G = C,,. By 4.7.9, 1+x+...+xk-1 E
for each prime, p, dividing n. Let
exp(2ni/n) and let y : C -> C.
be given by y(x) = n. Hence
Det(1+x+...+xk-1)(y')=
if y'
1E
if y`1.
Therefore the function, f, given by
f(y) _
(1 - n')/(l - n) if y` 1,
1
ify1=1.
lies in HomQQ(R(Cn),Z[n]'). However, this group equals

HomnQ (R(Cn ), Z
since a cyclic group has no irreducible symplectic characters. Hence, by

4.2.47, S(k) = Det(1 + x +... + xk-1) . f-l at all primes dividing n and is
therefore in the indeterminacy of 4.2.37.
Next we consider the case when G = Den. In this case, also, Hom,Q is
equal to HomnQ.
Let Uk E Zp[D2n]" denote the unit of 4.7.10, when p is a prime which
divides 2n. If X : Den --) C' is a one-dimensional representation then

x(D2n) _ {1} and an easy calculation yields
1
Det(uk)(x) =
Now let p, = Indl"(y`) with y`
2k + 1
if x * 1,
if x = 1.
1. If il = ;,, one finds that
Det(uk)(p;)
l+rl+...+nk
-det1 l +11-1
.i.....+fl-(k-1)
\\\
l+il+...+nk-1
+q-1 +...+q-k
= [(ilk+1 - 1)(n-(k+l) - 1) - (ilk - 1)(11-k - 1)](11 -
= (p2k
1)('1-(k+1)
1)-1(>7-1
1)-'
/
l)-1(n-1 - 1)-1
E Z[ n]'.
120
Define h E Hom,+ Q (R(D20, Z[ ] *) by
h(T) =
if dimC(T) = 1,
Det(uk)(Pi)
if T = pi, y`
1,
then S(k) = Det(uk) h-1 at all primes dividing 2n so that it is in the

indeterminacy of 4.2.37.
4.3 Determinantal congruences
Throughout this section let I be a prime and let M be an unramified

extension of the l-adic field, Qi. Let N/Qi be a large, finite Galois
extension such that N contains M and all the nth roots of unity. Let G
be a finite group of order n. We will consider the group of ilm-equivariant
functions
4.3.1
Homn,, (R(G), (9N )
where R(G) is identified with RN(G) and is therefore generated by representations of the form
T : G -?'
4.3.2
We have a determinantal homomorphism

4.3.3
Det : (9M[G]*
Homn,, (R(G), (9N)
given by the formula of 4.2.18

4.3.4
Det(>, 2y)(T) = det(>y A, T(y)) E 6N.
Since M/Qi is unramified, I is a prime in (9M and the residue field of

M is given by
4.3.5
m=(9,M/l.
The residue field of Q, is Fi, the field of order 1. In addition, there is

an isomorphism of Galois groups
4.3.6
G(M/Qi) = G(m/F,).
The right-hand group of 4.3.6 is generated by the Frobenius map, F,

which is defined by
F(z) = z1
4.3.7
121
(z E m),
so that we may lift F uniquely to F E G(M/Q1), which satisfies
F(w) - wi
4.3.8
(mod l
(9M)
for allwE(9M.
If uY Ayy E (9M [G] we may therefore define
4.3.9
F(E7 AYy) = E, F(2Y)y E (M [G]
We are now in a position to state our first result concerning determinantal congruences. The following result proves a conjecture of M.J.
Taylor (1978) which is also mentioned in Frohlich (1983, p. 79).
Theorem 4.3.10 Let z E (9M [G]'. Then, for all T E R(G),
Det(F(z))(y)'(T))/(Det(z)(T))i E 1+l(9N.
Here Wi denotes the Adams operation of 4.1.1.
Proof By Theorem 4.1.5 there exist integers, {nil, and one-dimensional

representations, {i : Hi -+ N'}, such that
4.3.11
r T = >i nil ndH.(0i) E R(G) and

l ip'(T) _ iniIndH;(4;) E R(G).
By multiplicativity
Det(F(z))(IndH1(4i))n`
Det(F(z))('(T)) =
E (9 N
11
(Det(z)(1 ))i
(Det(z)(Indc' (4.)))!nj
4.3 .12
so that we are reduced to the modulo 1 comparison of the expressions

Det
(A) (In4(0i))i and Det

Y
(F()
Y
in (9N, where z = UY Ayy) E (9M[G]'.

Let us abbreviate (Hi, i) to (H, ).
Choose coset representatives, x1, ... , xd E G, for G/H. There is a

homomorphism, 6 : G -> Ed, such that for g E G
gxi = x,(g)(j)h(i, g),
(h(i, g) E H).
122
With this notation we find that

Det(>y A
det(>y 1
(y) diag[4(h(1,7)),...,4(h(d,y))])
= det(X), say,
where diag [ul,... , ud] is the diagonal matrix whose (i, i)th entry is equal
to u;. Also we have
Det(>y F(Ay)y)(Ind4(4'))
= det(>y F(2y)a(y) diag [r1(h(1, y)),. .. , 4'(h(d, y))] )
= det(Y ), say.
If X;j is the (i, j)th entry of X then

det(X) = E sign(hF')X1,f(1)...Xd,fl(d) E (9N
#EEd
and, by 4.3.8 and the binomial theorem,

det(X)1
= EPEEd sign(i3)Xi,g(1) ... Xdl,a(d)

EEd sign(#)Yt,)3(1) ... Yd,fl(d)
det(Y)
(mod l(9N)
(mod l(9N)
(mod IN)-
Since det(X)' and det(Y) both lie in (9N we see that det(Y)det(X)-1 E
1 + (9N, as required.
When l = 2 it is very convenient to work modulo four. For this reason

I will record the following determinantal congruence, which is obtained
by squaring the result of Theorem 4.3.10.
Corollary 4.3.13 When I = 2 in 4.3.10, if z E (9M[G]*, then
Det(F(z2))(p2(T ))
Det(z2)(T )2
E 1 + 4 (9N.
4.3.14 log and exp
The congruences of 4.3.10 and 4.3.13 will permit us to take the l-adic
logarithms of
Det(F(z))(W'(T))
(Det(z)(T))'
in 4.3.10. By analysing this construction we shall naturally be led to the

group-ring logarithm of Curtis & Reiner (1987, p. 359) which was originally discovered, independently, by R. Oliver and M.J. Taylor. However,
123
our route to the logarithm will remove some of the agony and all of the
mystery.
We will begin by recalling some well-known facts from Curtis & Reiner
(1987, p. 356).
Let N/Q, be a finite extension of local fields. Let it E ON be a

uniformising element, so that < it > '40N is the maximal ideal. Let
v : N -+ Q be the normalised exponential valuation which counts the ndivisibility of elements of N. This function is normalised so that v(l) = 1.
The ramification index of N/Qi is the integer, e, such that
<T,e>=I'(9N 1 (9N
Hence v(n) = e-' is the least positive value taken by v on N. Define two
formal series by
E(-1)n-lxn
log(1 +x) _
/n,
4.3.15
exp(x) _ Elo xn/(n!).

Proposition 4.3.16 (Curtis & Reiner, 1987, section 54.2) (i) The series,
log(1 +x), 1-adically converges to an element of N for all x E< TC > -4(9NIf v(x) >- 1 then log(1 + x) E 1(9N-
(ii) The series, exp(x), converges to an element of 1+ < it >g (9N

whenever v(x) > (1 - 1)-1. If l * 2 and v(x) >- 1 then exp(x) E 1 + ION(iii) If 1 2 and v(x) >- 1 then
exp(log(1 + x)) = 1 + x,
log(exp(x)) = x.
(iv) If 1 = 2 and v(x) > 2 then

exp(log(1 + x)) = 1 + x,
log(exp(x)) = x.
Corollary 4.3.17 (i) If w E ON is an l-primary root of unity then

(o-1 E< it > and log(o)) = log(1+(w-1)) converges. In fact, log((O) = 0.
(ii) Let x E (9N. Suppose that w E ON is an I-primary root of unity such
that, for some k > 0
x-w
(mod lk+i(9N) and
x = 1 (mod ION),
then we may choose
if 1#2,
t +1 if 1=2.
(
124
Proof Part (i) is well-known (for example, Curtis & Reiner, 1987, p. 357).
For part (ii) assume first that 1 2. In this case there is nothing to prove
if k = 0. Therefore we shall suppose that k > 1. Let us recall some facts
from Curtis & Reiner (1987, pp. 356-7) concerning the valuation, v.
Let X E (9N and let n >- 1 be an integer then, if v(x) >- 1,
v(x"/n) >_ 1 and v(x"/(n!)) > n(v(x) -1/(l - 1)).

Suppose that x/w = 1 + lk+1u then, by part (i), log(x) is convergent and
= log(x) - log((O)
= log(x/(0)
log(x)
= C `0'(-1)n-1(lk+1u)n/n,
which lies in lk+1(9N, by 4.3.16. When l * 2 then exp(log(x)) E 1+lk+1(9N,
using 4.3.16 again, and therefore
x = exp(log(x)) = 1
(mod lk+1(9N)
as required.
When l = 2 we may write
(x/w)2 = 1+2 k+2U
from which we find that log(x2) E 2k+2(9N and exp(log(x2)) E I+ 2k+2 0N,
by 4.3.16. Hence
x2 = 1
(mod 2k+2&N)
x = +1
(mod 2k+1(9N)
and therefore
since such elements, x2, have precisely two square roots, which differ by
a sign and lie in 1 + 2k+1(9N
4.3.18 In Theorem 4.3.10 we produced a mod 1 determinantal congruence
which holds for all units, z E (9m[G]*, and for all representations, T.
Theorem 4.3.10 proved the conjecture of M.J. Taylor which was posed in
Frohlich (1983, p. 76) and Taylor (1978). However, when T is congruent
to zero mutually higher powers of 1, in the sense of Definition 4.3.31,
we can improve the determinantal congruence of 4.3.10 considerably.
We shall work towards such an improvement and in the course of this
process we will recover the group-ring logarithm, which was originally
discovered by M.J. Taylor (1980) and R. Oliver (1980), independently.
125
By Theorem 4.3.10 and 4.3.16(i) we may form the following composite

homomorphism:
log(Fip'/l(Det)) : OM[G]* --+ HomnM(R(G),1 + ION)
4.3.19
) HomnM(R(G),ION)
given by (z E (9M [G] *, T E R(G))
log(Ft1/l(Det(z)))(T) = log
DetF(z)(Wl(T))
E ION( Det(z)(T)1 )
We would like to rewrite the logarithm in 4.3.19 as a difference of

logarithms which both lie in ION. The obvious expansion of the logarithm
of a quotient as a difference of the logarithms of the numerator and

denominator cannot be applied to 4.3.19 for all z E OM[G]*. Therefore
we will study, pro tem, the case in which z is congruent to one mutually
the Jacobson radical, J.
The Jacobson radical, J 4 OM[G] (Lang, 1984, p. 636), is the left ideal
which is equal to the intersection of all the maximal left ideals of (9M [G].
In fact, J, is a two-sided ideal and (9M [G] /J is semi-simple. Hence, by
Wedderburn's theorem (Lang, 1984, p. 629) COM[G]/J is a product of
matrix rings over division rings. However, in this finitely generated, local
situation some power of J lies in lCOM [G] (Curtis & Reiner, 1981, section
5.22, p. 112). Hence the division algebras have characteristic 1. The
division algebras whose centre is the field, K, are measured by the Brauer
group, Br(K), and for a finite field the Brauer group vanishes (Serre,
1979, p. 161) so that there is an isomorphism of the form
4.3.20
One[G]/J - f'=, Mn,(Fla)
Lemma 4.3.21 Let T E R(G) and let r E J, the Jacobson radical of

OM[G] in 4.3.10. Then 1 - r E OM[G]* and both log(Det(1 - r)(T)) and
log(Det(1 - F(r))(ip'(T))) are 1-adically convergent in ON. In addition,
log(Det(1 -F(r))(p1(T))/[Det(1 -r)(T)]1)

= log(Det(1 - F(r))( 1(T ))) - llog(Det(1 - r)(T)).
Proof Since r" E 10M [G] for some n (Curtis & Reiner, 1981, section
5.22, p. 112) the series for (1 - r)-1 converges l-adically and 1 - r E
OM[G]*. Since 4.3.19 is 1-adically convergent it suffices to show that
126
log(Det(1 - r)(T)) is l-adically convergent and, by additivity, we may

assume that T is a representation of the form of 4.3.2.
Since r" E 1GUM [G] we find, for large m, that
I - T(r")
T(1 - r)1"
(mod 1)
(mod 1).
On the other hand, T(1 -r) E (9N whose residue field, (9N/ < it >, has no
/-primary roots of unity so that T (l - r) = 1 - T (r) = 1 (mod < it >).
Therefore det(T(1 - r)) - 1(mod < it >) and log(Det(1 - r)(T)) is

1-adically convergent in (9N.
4.3.22
4.3.23
If r E J as in 4.3.21, define the logarithm, L0(1 - r), by
Lo(1 - r) = 1 E' i r"/n
- E' i F(r")/n E (9M[G],
where FCEc' M _ EyEG F(2y)yi, F being the lifted Frobenius of 4.3.8.

If T E R(G) has characteristic function, XT, so that
XT(g) = Trace(T(g))
for all g e G, then we may define a homomorphism

gT :(9M[G]--p(9N
4.3.24
by the formula
E 'lyY
= E'lyXT(Y)
yEG
yEG
Proposition 4.3.25 Let r e J and T E R(G) be as in 4.3.10-4.3.21. Then,

if T is as in 4.3.24,
i
log
(Det(1 -(I (DetF(rr)(T))T)))
_ T(Lo(1 -r)) E (9N.
Proof By additivity we may suppose that T is a representation, as in 4.3.2.

Let 2i, ... , A,, denote the eigenvalues of T (r). In the course of proving
Lemma 4.3.21 we saw that each 2i lies in the maximal ideal, < it > < (9N,
if N is large enough. Therefore the following series converges:
log(Det(1 - T(r))) = log(fl 1(1 - Ai))

u
ao
aimlm
- i=1
m=1
Eco=1
u=1 Amilm
Li
_ - Em i Trace(T(rm))/m.
127
Similarly, since the eigenvalues of p'(T)(r) are {2;}, we find that

00
log(Det(1 - F(r))(ip'(T))) = - E Trace(T(F(rm)))/m,

m=1

Definition 4.3.26
Let
AG = (9M[G]l E (9M(xY -Yx)

x,YEG
Thus AG is an (9M-module and we may define, for r E J a (9M [G],
L(1 - r) = n(Lo(1 - r)) E AG,
4.3.27
where it :(9m[G] --> AG is the canonical quotient map.
Proposition 4.3.28 Let G be any finite group. If r E J then
L(1 - r) = n(Lo(1 - r)) E IAG.

Proof Consider the series
L(1 - r) = I Y, ir(rm/m) - E 1C(F(rm)/m) E AG.
m=1
M=1
If l does not divide m then lrm/m E l(9M[G] and the n-image of such a
term is 1-divisible in AG. Now consider the remaining terms
4.3.29
n(F(r`")/m)
m=1
=
Em_1
(rlm - F (rm))l m.
q with HCF(q, l) = 1 then we may set t = rq so
Suppose that m =
that
(rim
- E(rm))/m = (tis - F(t'
Therefore we must show that, if r E J,

4.3.30
it(r1s
- F(ris-' )) E ISAG
' ))l m.
128
for alls>-1.
Suppose that r = t aigi (summed over all gi E G). Therefore
ris
= 1: aj, ... ads gJl ... $JI:
where (j1,. .. , jts) ranges over all possible Is-tuples, which we shall think
its), and consider
of as `permutations'. Fix a permutation, i =
all the terms, j = (jl, ... , jts ), which are obtained from i by a cyclic
permutation. The products gi ... g,, and $2 ... g,,gl are conjugate in G so
that each term in the subsum of js will have the same n-image in AG.
Let H denote the cyclic subgroup of order is which cyclically permutes
i = (i1,...,its) and suppose that the stabiliser of i in H has order 1". In
this case there are ls-u terms in the subsum of j's and, in AG, we obtain
is-" times their common it-image.
When u = 0 the contribution of these terms to 4.3.30 will lie in IsAG.
Now suppose that u > 1 so that we may write
and we may consider the terms in rls-` which are of the form
aj, ... aj,,_, gj, ... g j,s_, ,
where j = (ii.. . , jts-1) is a cyclic permutation of
The
stabiliser of in the cyclic group of order is-1 is of order lu-1 so that

there are is-" terms in this subsum of F(rts-1) which map to the same
image in AG.
Now, in AG, let us collect together the terms from these two troublesome subsums when u > 1. From the subsum of js we obtain a
contribution of the form ls-ual1t(gl) while from the subsum of js we
obtain -is-uF(at 1)11(g') where g is the monomial, (gi,giz...). However,
by induction starting with 4.3.8, we find that
atu
- F(atu-1) E l"(9M
so that the difference of the two subsums contributes an element of

ls-u(IuAG), which establishes 4.3.30 and completes the proof.
Definition 4.3.31
Let T E R(G). We shall say that T = 0 (mod lk) if,
for all g E G, XT(g) = 0 (mod lkON) where XT is the character of T.

Theorem 4.3.32 Let G be a finite group and let J denote the Jacobson
radical of (9M [G], in the notation of 4.3.10-4.3.21. Let r E J and suppose
that
T E R(G) satisfies
129
T - 0 (mod lk) then
Det(1 - F(r))(p'(T ))
E + 1'+'(9N
(Det(1 - r)(T))i
where e = 1 if l is odd and e = 1 if l = 2.

Proof There is nothing to prove if k = 0 so we may assume that k > 1.
The map, T of 4.3.24, factorises as
ST : (9M[G]
") AG
) (9N,
where iT(1r(E 2,Y)) = E tyXT (Y)
By 4.3.28, L(1-r) = it(Lo(1-r)) E lAG so that T(Lo(1-r)) E 1'+'(9N

Therefore, by 4.3.25,
4.3.33
to g
Det(1 - F(r))(W'(T ))
( (Det(1 - r)(T))'
lk+l(9N
From 4.3.33 we may now finish by means of the argument which was
used to prove Corollary 4.3.17(ii).
Suppose first that l * 2, then we know from 4.3.10 and 4.3.16 that
4.3.34
-F(r))(W'(T))1)
exp I log f Det(1
\\\
(Det(1 - r)(T))'
Det(1 - F(r))(W'(T))
(Det(1 - r)(T ))l
On the other hand we know, from 4.3.17(ii)(proof) that
exp(1k+1&N) c 1 + lk+'ON,
which completes the proof for odd primes, 1.

When l = 2 we may, as in 4.3.17, apply the previous argument to the
square of 4.3.34 and then extract the square root in 1 + 2k+2ON.
4.3.35 We will close this section with a strengthening of Theorem 4.3.32.
Firstly let us define some subgroups, WG(k) c 1 + l&N, in the notation
of 4.3.10-4.3.21.
Set
130
4.3.36
if 1
1 + lk+1(9N
2, for all G or
if 1=2 and k = 0,
WG(k) =
2, G has no quotient
isomorphic to D6,
if 1
{+1 }J + 2k+1 ON
{1, } + 2k ON if l = 2 and G has a

quotient isomorphic to D6.
Notice that N is a `large' field so that

E N.
With this preparatory notation we are now ready to state and prove
an extension of Theorem 4.3.32 to apply to all units of (9M [G].
Theorem 4.3.37 Let G be a finite group and, in the notation of 4.3.10-
4.3.21, let z E (9M[G]'. Suppose that T E R(G) satisfies T - 0 (mod lk)

then
DetFz
4.3.38
1(T
(Det(z)(T))'
E yyG(k) cl+lON.
Proof By 4.3.10 we may assume that k >_ 1. Fixing T E R(G), the

expression of 4.3.38 defines a homomorphism on z E (9M [G]' which, by
4.3.10-4.3.32, factorises to give a homomorphism
(9M[G]
l+ j
1+l(9N
1+ lk+10N
when l# 2
and
T2
(9M [G]'
1+20N
1+J
{1} + 2k+1 ON
when 1= 2.
The map, T1, sends the coset, z(1 +J) to 4.3.38. In each case the range of
'F1 is an abelian 1-group, as is seen by means of the logarithm of 4.3.15
and 4.3.16.
However, by 4.3.20, we have an isomorphism of the form
4.3.39
(9M[G]*/(1 +J) =
We will use 4.3.39 to show that 'F, is trivial unless l = 2 and G has a
quotient which is isomorphic to the dihedral group, D6. In the remaining
case we will show that 'F2 is the trivial map. When G has no D6 quotient
this will obviously show that 4.3.38 lies in WG(k) of 4.3.36. When l = 2
and G/H = D6 for some H 1 G then the square of 4.3.38 will lie in
({1} + 2k+1(9N) c N' and the square root of any such element exists
4.4 Detecting elements in the class-group
131
and belongs to {+1, +,[--11 + 2' (9N, which equals WG(k) in this case,
also.
By 4.3.39 it will suffice to show that Hom(GL (Fsd), Z/l) is zero when
l * 2 or when l = 2 and d > 1. For the remaining case GL (F,d) =

GL2(F2) = D6 and the only homomorphisms from D6 to a finite abelian
2-group are of order two.
When n = 1, GLI(Fld) = Fed has no /-torsion and so Hom(Fid, Z/l) = 0.
More generally,
Z/l) =
Z/l),
which vanishes if d(l - 1) > 1, by Quillen (1972, theorem 6, p. 578).

Alternatively, when ld > 4 or n >- 3
is simple (Lang, 1984, pp.
472-480) so that Hom(SL (Fld), Z/l) = 0 and therefore
Z/l) = Hom(F,d, Z/1) = 0,
too.
These remarks rule out all cases except GL2(F2) = D6, which completes
the proof of Theorem 4.3.37.

4.4.1
Suppose now that K /Q is a finite Galois extension (Q is the field
of rationals) and that 1 is a prime which is unramified in K/Q. Let

P i &K be a prime which divides 1. Therefore
G(KP /Qi) c G(K/Q).
Let Fp denote the Frobenius of G(Kp/Qj), which is characterised by the

congruence of 4.3.8
Fp (x) - x1
(mod P) (x E CO-
Now let E/Q be a `large' finite Galois extension, containing K, as in

4.2.28. Let us suppose that there exists
4.4.2
Pp E G(E/Q) such that FP = Fp

on Kand Fp acts like the identity
on T(g) for all g E G, T E R(G).
In 4.4.2 it is sufficient that FP acts trivially on the matrix entries of T(g)
where T is a representation of the form 4.2.9 and T runs through a
132
set of generators for R(G) = RE(G). The condition of 4.4.2 is vacuously
fulfilled when K = Q for then Pp is the identity. It may also be

fulfilled, for example, when #(G) = le for then we may take E =
K Q(exp(2rzi/le)). In that case K/Q, being unramified at 1, is linearly
disjoint from Q(exp(2ni/le))/Q.
4.4.3 The detection homomorphisms
Let T E R(G) and suppose that K/Q satisfies the conditions of section
4.4.1, so that Pp E G(E/Q) exists in 4.4.2. Consider the map
4.4.4
HomnK (R(G), (Kp K E)")
(Frw p
HomaK(R(G),(Kp K E)')
given by
(FpiV')(f)(T) = Pp (f (W'(T)))I(f (T))'.

Notice that if CO E S2K then zp = FP 1wFp E S2K so that
w(Fp(f(ip'(T))))
= Fp(zp(f(ip'(T))))
= Fpf(zp(ip'(T)))
= Fpf(FPIw(w'(T))),
= Ppf(FP '(ip'(w(T ))))
= Fpf(ip'((o(T))),
by 4.4.2,
by 4.4.2,
so that 4.4.4 takes values in S2K-equivariant homomorphisms.

For each T E R(G) we also have an evaluation homomorphism
4.4.5
evalT : HomnK(R(G),(Kp K E)*) ---> (Kp
By Galois invariance 4.4.5 must take values in (Kp K K(XT))', where
K(XT) is the smallest Galois extension of K which contains all the

character values of T, {xT(g); g E G}.
Finally, by means of the isomorphism of 4.2.17,
4.4.6
(Kp K K(XT))* = [IQIp K(XT)Q,
where Q runs through the primes of K(XT) which divide P. We therefore

have a projection map
4.4.7
IEQ
: (Kp K K(XT))* - K(XT)Q.
Combining 4.4.4-4.4.7 we obtain a homomorphism

4.4.8
133
ST,Q = mQ(evalT(FPlp1 /l)) :

Homo,, (R(G),f p'Ip EP,)
) K(XT)g.
Now consider the effect of ST,Q on a function of the form
Det(u) : R(G) -f fi Ep,,

P'IP
where u E (9Kp [G]'. If a =:yEG , y then, by the assumptions of 4.4.2,

(FPiV'11)Det(u)(T) =
Det(EFP(,y)y)(tp'(T))
(Det(E A.yy)(T ))'
Therefore, if T - 0 (mod lk) then Theorem 4.3.37 implies that

4.4.9
ST,Q(Det(u)) E WG(k) rl K(xT)g = V(T)p, say.
Also note that HomoK(R(G),E') and Homo K(R(G),(9E) are mapped by

ST,Q into K(XT)* and (9K(XT) respectively.
We may now apply the maps {ST,Q} to obtain homomorphisms out of

'M'(Z[G]) and D(Z[G]).
For T E R(G), define
4.4.10
V(T) = flp V(T)p
U((9Q(xT)),
where V (T )p is as in 4.4.9. With this notation 4.4.9 together with the

Hom-descriptions of 4.2.28, 4.2.35 and 4.2.37 yields the following result:
Theorem 4.4.11 Let G be a finite group.
(i) With the preceding notation there is a well-defined homomorphism

(T E R(G))
ST = {STQ} : WY(Z[G])
where Hom denotes HomiQ(R(G),E') and
Jf,,,(Q(xT))
V(T)' ST(Hom)
denotes the adelic group
obtained by deleting the Archimedean places from J'(E).

(ii) There is a well-defined homomorphism
IIPEs/ (9
ST = {ST,Q} : D(Z[G])
Q(xT)P
S Hom+ )
flPE.a V(T)P'T(
where Hom+ denotes Hom+Q(R(G), (9E) and d is the set of primes of

Q(XT) which divide #(G).
134
Example 4.4.12 Let l be a prime and suppose that #(G) = lk. Take T
to be the regular representation, T = Ind{i}(1) E R(G). Hence
(0
xT(g) = Sl
if g 1,
#(G) if g = 1,
so that T - 0 (modulo lk). Since the character values of T are rational,

we obtain in this case
ST : D(Z[G]) -
Z;
{1} (1 + lk+1Z1)
from Theorem 4.4.11 (ii), since D6 is not a quotient of G. The {1} in the
denominator of the target group arises because ST(Hom+) c Z'
-1}.
In terms of the Hom-description ST is given here by
ST(f) = f(W'(Ind{i}(1)) - llndg}(1)),
where f E
(R(G), (9EP) and P is any prime of E which divides 1.
4.4.13 The Swan subgroup
The Swan subgroup of G is defined to be the subgroup of D(Z[G])

which is generated by the classes of the Swan modules, S(k) (with
HCF(k, #(G)) = 1), which were introduced in 4.2.43. The Swan subgroup
will be denoted by T(G). By 4.7.6-4.7.8 we may define a surjective

homomorphism
4.4.14
S : (Z/#(G))'/{1} --> T(G)
by
S(k) = [ideal{k,a}]
of 4.2.44.
We will use the map, ST, of Example 4.4.12 to determine T(G) when
#(G) = lk. The proof is originally due to M.J. Taylor (1978), who did not
have available all the maps of Theorem 4.4.11 but did have the particular
case of 4.4.12.
4.4.15 The Artin exponent
The Artin exponent of G, which we will denote by A(G), is defined (Curtis
& Reiner, 1987, p. 782) to be the smallest integer such that there is an
equation in the rational representation ring, RQ(G), of the form

4.4.16
135
A(G) 1 = Ei1 ndC,(4;) E RQ(G),
where C, is a cyclic subgroup of G and O+ E RQ(C;).

By 2.1.3, A(G) divides the order of G.
Lemma 4.4.17 The Artin exponent annihilates T(G). In particular, #(G)
T(G)=0.
Proof Let C be a subgroup of G. We have maps
Resc : R(G) - R(C)

and
I ndG : R(C) -- R(G).

These maps induce homomorphisms, via the Hom-description of 4.2.35,
4.4.18
IndG D(Z[C]) --> D(Z[G]) and

ResC D(Z[G]) --> D(Z[C]) respectively.
In terms of modules, I ndG [M] _ [Z[G] Z[cl M] and Rest [M] = [M]
(Curtis & Reiner, 1987, pp. 339-341; Frohlich, 1983).
Now suppose that f : R(G) -) U((9E) is an QQ-equivariant homomorphism whose class in 4.2.35 represents S(k) E T(G). In this case
A(G)S(k) = A(G)[f].
From Curtis & Reiner (1987, section 76.8, p. 784) there exists an
equation, which apparently improves upon 4.4.16, of the form
A(G) 1 =
ncI ndG(1) E R(G),

c
where the sum is taken over cyclic subgroups, C, and the {nc} are
integers.
Hence, if T E R(G), we have

4.4.19
f(A(G)T) =f(Ecnc(T IndG(1)))

_ FIc(f(IndG(ResG(T))))"c
by 1.2.39.
However, Resc of 4.4.18 maps T(G) to T(C) (Curtis & Reiner, 1987,
p. 345) so that [f IndG] E T(C). However, T(C) is the trivial group, by
4.2.48. Hence [f(A(G) -)] is trivial in D(Z[G]), by 4.4.19.
136

Table 4.1. (Theorem 4.4.20)
G = {1-group of order Ik}
#(T(G))
cyclic
non-cyclic , I # 2
lk-1
generalised quaternion of order 21, m > 3

dihedral of order 2", m >_ 2
semi-dihedral of order 21, m >_ 4
all other non-cyclic groups of orders 2'
2
1
2
2k-2
Theorem 4.4.20 Let l be a prime and let #(G) = 1k. The Swan subgroup,
T(G), is a cyclic 1-group whose order is given by Table 4.1.
Proof The l-primary torsion in (Z/lk)'/{1} is cyclic so that T(G) is a

cyclic l-group, by 4.4.17. It remains to determine #(T(G)).
The cyclic and dihedral Swan subgroups are both trivial, by 4.2.48.
The Swan subgroup of the generalised quaternion group, Q8, is of order
two, by 4.7.11. Let Q2" denote the generalised quaternion group of order
2". By 4.7.12, the restriction map, T (Q2") -> T(Q8) is onto, so that
it suffices to show that 2 T(Q2") = 0. This will follow from Theorem
5.3.3, since Q2" has a cyclic subgroup of index two, C2"-1, and there is no
element of order two in Q2" - C2"-1. The case of the semi-dihedral group
is dealt with similarly.
The semi-dihedral group, SD", has order 2n+2 and is given by
SD" = {a, b I a2n+` = 1 = b2, bab = a2"-1 }.
SD" has a subgroup which is isomorphic to Q8, namely the subgroup

generated by {a2"-`,bal. The restriction map, T(SD") --> T(Q8) is onto.
Therefore it suffices to show that 2 T(SD") = 0. If we apply Theorem
5.3.3 to SD" and its subgroup of index two, J = {a}, we obtain a relation
of the form, where k is any odd integer,
0 = 2 S(k) - TIndC(g)(S(k)) E D(Z[SD"]).

This sum ranges over some elements of order two in SD" - J and C(g)
denotes the centraliser of g. Such elements of order two have the form
a2'b and have a copy of the dihedral group, D8, for a centraliser. Since
0 = S(k) E T(D8) the relation shows that S(k) is annihilated by two, as
required.
137
Now let us suppose that G belongs to one of the remaining cases; therefore G is not a cyclic, dihedral, semi-dihedral or generalised quaternion
group. For these remaining cases
(Z/#(G))'/{1} = (Z/lk)'/{1}
has an /-primary part which is cyclic of order Ik-1 when l 2 and 21-2
when I = 2 so that it will suffice to detect classes, S(k), which are of at
least this order. To do this we will evaluate
ST(S(1 + l)) E
Zj
{1}(1 + lk+1Z1)
where ST is as in 4.4.12. By definition this is equal to

S(1 + 1)(1p'(Ind{1}(1)) - l Ind(11 (1)),
4.4.21
where S(1 + 1) is the idelic-valued function given in 4.2.47. Therefore

4.4.21 is equal to
(1 + l)e E
Z
{1}(1 + lk+1Zt)'
where
e =< 1,ip'(Ind{1}(1)) - l Ind{}(1) >

=< l,ip'(Ind{1}(1)) > -l
[#(G)'
EgEG XInd{}(1)(91)) -1,
by 4.1.2 and 1.2.7,
=#{gEGIg'=1}-1.
However, by a theorem of Kulakoff (when l # 2) and of Alperin, Feit
and Thompson (when l = 2) (see Huppert, 1967, p. 314; Isaacs, 1976, p.
52)
#{gEGIg'=1}-0(mod12).
Therefore e = lq with HCF(l, q) = 1 so that
ST( S ( l
+ l )) = ( 1 + l )
19
Zl
{1}(1 + lk+1 Z,) '
which has order 1k-1 when I * 2 and 2k-2 when l = 2. This completes
the proof of Theorem 4.4.20.
Corollary 4.4.22 Let I be an odd prime.

If G is a finite 1-group then the Artin exponent, A(G), is given by
A(G) =
#(G) l-1
if G is cyclic,
if G is non-cyclic.
138
Proof The result is trivial when G is cyclic and otherwise it is clear from
2.1.3 that A(G) divides #(G) 1-1. Therefore, by 4.4.20, A(G) = #(G) I-1
since A(G) T(G) = 0, by 4.4.17.
(i) When #(G) = 2k, A(G) = 2k-2 if G is not cyclic,

dihedral, semi-dihedral or generalised quaternionic and is 1,2,4,2, reRemark 4.4.23
spectively, in the exceptional cases (Curtis & Reiner, 1987, p. 365).

(ii) In the course of proving Theorem 4.4.20 we used Kulakoff's theorem
that, if l # 2 and G is an 1-group, then
#{gEGIg'=1}- 0(modl2).
In fact the proof shows that for all k satisfying HCF(k, 1) = 1
ST(S(k)) = kA_' E
Z*
{1}(1 + lk+1ZI) '
where A=ale=#{gEGIg'=1}.
However, if k has order (1-1)/2 in (Z/lk)*/{1} then ST(S(k)) = 1 so
that we find that A - l is a multiple of (l - 1)/2.
This is equivalent to the following:
Proposition 4.4.24 Let #(G) = lk with l an odd prime. Then
#{g E G I g' = 1} = (1 + t(l - 1)/2)12

for some positive integer, t.
4.5 Galois properties of local determinants

4.5.1
Let M/QI be a finite Galois extension and let G be a finite
group. If N/QI is a 'large' Galois extension, in the sense of 4.2.10, which

contains M/QI then we have the determinantal homomorphism
4.5.2
Det : (9M[G]* -+ HomnM(R(G), (9N),
whose image is Det(0M[G]*).

In Frohlich (1983, p. 84) we find the following result of M.J. Taylor:
Theorem 4.5.3 (Frohlich, 1983, Theorem 10A, p. 84; Taylor, 1980, p. 93)
Let M/QI be a tame, finite Galois extension with Galois group,

H = G(M/Q1). Then
(Det((9M [G] *))H
Det(ZI [G] * ).
139
Theorem 4.5.3 is one of the three main steps in Taylor's proof of the
Frohlich conjecture, which is the subject of the book (Frohlich, 1983).
We will not go into the details of the Frohlich conjecture here (see
Chapter 7). Suffice it to say that the ring of algebraic integers in a
tame Galois extension of number fields is a projective module over the
integral group-ring of the Galois group and Frohlich's conjecture gives
an analytic description of the class of this module in the class-group.
Incidentally, in Theorem 4.5.3, M/Q1 is tame if and only if (9m is a
projective Z1 [H]-module.
As an application of the material of Section 3 we will prove Theorem

4.5.3 when G is an /-group and M/Q1 is unramified (see Theorem 4.5.39)
- a case to which 4.5.3 is readily reducible (see Frohlich, 1983, pp.

85-89; see also 4.6.4). Our proof will follow the basic lines of that given
in Frohlich (1983) but is much more natural, being simplified by the fact
that we start with the new determinantal congruences of 4.3.10. En route
we shall encounter several maps and isomorphisms which are of interest
in their own right.
Definition 4.5.4
Denote by M{G} the M-vector space whose basis
consists of the conjugacy classes of elements of G. Let
4.5.5
c:M[G]- )M{G}
denote the M-linear map which sends g E G to its conjugacy class. M{G}
may be identified with the centre of M[G] (see Lang, 1984, section 4.1,
p. 647), but we will not use that description here.
Let N/Q1 be a finite Galois extension which contains M and all the
n-th roots of unity (#(G) = n). Hence N is `large' in the sense of Section
3. Suppose that yl,... , ye denote the distinct conjugacy classes of G and
let yl, ... , Ye be the corresponding characteristic functions given by
y'(Yj)
if i=j,
if i
j.
Finally, let T1,..., Te denote the distinct,irreducible N-representations

of G so that Tj E RN(G) = R(G). Let xj denote the character function of
Tj.
Lemma 4.5.6 Let n = #(G) and let n = exp(2iti/n). Let `'G denote the
space of N-valued class functions on G (cf. 1.2.22). There exists a matrix
A = (aij) E
c GLe(N),
140
such that
e
Yi = E IXijxj E WG.
j=1
Proof By 1.2.23 we know that the matrix, A, exists with entries in some
large field. We must show that the entries lie in
By 2.1.3 there exist cyclic subgroups, {C5}, and one-dimensional representations, {his : CS ---> N*}, such that (for suitable as E Z)
n = #(G)
as(I ndG (cbs)) E R(G).

s
Therefore, by Frobenius reciprocity for class functions (1.2.39)

nYi
= Es asY1 I nd's(4s)
_ Es a,I ndG (4s 0 Rescs(Yi))
Therefore we are reduced to proving that any class function on a cyclic

group is a
combination of characters of representations.
Let C be a cyclic group of order t with generator, x. Let y : C -> N*
be given by y(x) = i. Define an invertible matrix in
X =(Xij)=
4.5.7
so that, for 1 < i, j < t,

Xij =
4.5.8
yi-l(xj-t)
= 0-0(j-1)
From 4.5.8
Ei(X-'
)s,iyi-1(XU)
= Ei(X-')s,iXi,u+i
I
ifs=u,
if s=u.
The result follows, since X-i E GLt(Q(ci)).

4.5.9
0 : Homo,,,(R(G), N)
141
N{G}
by the formula
4.5.10
4(f) = Es f (Ys)Ys = Es=i Ee=I asi.f (Xi )Ys,
where (aid) is the matrix of 4.5.6.
The map of 4.5.10 is the composition of two maps of the following

form :
4.5.11
HomoM (R(G), N) ---> Homo,,, (R(G) 0 N, N)
f i- + mult(f 0 1)
(where `mult' denotes multiplication in N) and
4.5.12
HomcM(R(G) (D N, N) --> N{G}

e
h'-'
h(Ys)Ys
s=t
If co c f1m then
wMf)) =Esw(f(Ys))Ys
= Es f (w(Ys))Ys
Es f (ys)Ys
= O(f)
so that the image of 4.5.10 is contained within N{G}lM = M{G}.
On the other hand, we may define another map
4.5.13
W: M{G} --+ HomoM(R(G), N)
by
(mvY) (T) _ I: mYTraceT(y).

(I:
Y
Proposition 4.5.14
The homomorphisms of 4.5.10-4.5.13 define (inverse)
isomorphisms
:Homo,, (R(G),N) -> M{G}.
142
and
W : M{G} -.+ HomnM(R(G),N).
Proof Consider the composition

O(W(Ey myy)) = (T i --)Ey m7Trace(T(y)))
= Es Ey my T race(Ys)(Y)Y8
= E7, mysYs
=Eymyy,
so that W = 1.
On the other hand,
e
W Mf ))
lp( E E sjf
s=1 j=1
e
(xj )Ys)
e
(xta --) T, Eaqf(xj)xt(Ys))

s=1 j=1
(Xt
F> E asjf
(xj)(A-1)ts)
s=1 j=1
where
A=((x,j)is as in 4.5.6,
Qt
f Qt))
= f,
so that W4 = 1 also.
Remark 4.5.15 If one identifies M{G} with the centre of M[G] (see
4.5.4) then Proposition 4.5.14 may be proved by means of the reduced
norm map, nr : M[G]' -> M{G} = Centre(M[G]), (Curtis & Reiner,

1987, section 52.9, p. 332).
4.5.16 Explicit Brauer Induction and M{G}
Recall from 2.2.15 the explicit Brauer induction homomorphism
aG : R(G) -> R+(G)
143
given by
a(H,cb)13(P)(H,4>)G E R+ (G),
aG(P) =
(H,O)cE_&j14
where (H, )G denotes the G-conjugacy class of a one-dimensional representation, : H --) N*.
The absolute Galois group, S2M, acts on R+(G) by the formula
4.5.17
w((H, 4)G) = (H, (0(q5))G
((0 E SZM).
Since aG and w(aG((O-'(-))) satisfy the axioms of 2.2.8 they must be

equal, so that aG is 1 M-equivariant
4.5.18
aG E HomnM(R(G),R+(G)).
If H is a subgroup of G let NGH denote the normaliser of H in G and

set WGH = NGH/H, the Weyl group of H in G. Let : H ---> N' be a
homomorphism so that we may uniquely factorise 0 as
0:H
-H>Hab->N',
where Hab denotes the abelianisation of H. When there is no risk of

confusion we will also denote the resulting map, Hab -p N', by 0. The
Weyl group, WGH, acts upon the set of characters, 4> : H - N', by
conjugation. Since (H, 4>)G is the G-conjugacy class of 0 : H --p N'
then, once we have chosen a conjugacy class representative for H, 0 is
determined up to the action of WGH.
Therefore aG induces a homomorphism
4.5.19
AG : R(G)
(H)R(Hab)WH,
where (H) runs through the conjugacy classes of subgroups of G and

XWGH = X/{z(x) - x I X E X,z E WGH} denotes the coinvariants of
the WGH-action on X. The homomorphism, AG, is given by the same
formula as aG. Moreover, the canonical map gives an isomorphism
4.5.20
HomnM(R(Hab)WcH,N') ---> Homo, (R(Hab) N*)WGH
since WGH permutes the basis of the free abelian group, R(Hab), given
by the one-dimensional representations.
144
In addition, 4.5.19 is a split surjection whose inverse, by 4.2.7, is given

by the map
dH
R(Hab)WGH
R(G).
rzH > R(H)NGH
Thus, by 4.5.14 and the foregoing discussion, the dual of 4.5.19 yields
a split surjection of the form
4.5.21
(M[Hab]WGH)
(H)
H<G
-> M{G},
where the sum in 4.5.21 is taken over the conjugacy classes, (H), of
subgroups of G.
Now let us derive the naturality property of 4.5.21, which results from
the naturality of AG in 4.5.19, in the sense of 2.2.8.
Suppose that J < G, then we may define an M-linear map
4.5.22
I NDJ
I -<J
(M[I
(M[Hab]WGH)
ab]WJI)
H<G
by the formula (for y E M[Iab]WJI)

e
I ND (y)(H)
= 1: 1:
YS(z-'yZ)Y,
E M[Hab]WGH.
S=I
I czHz-(
In 4.5.22 I NDJ W(H) denotes the (H)-component of I NDJ (y). Notice

also that in 4.5.22 the {y,} run through the conjugacy class representatives
for H, not for Hab. The facts that 4.5.22 is well-defined and that it does
indeed lie within the WGH-invariants may either be proved directly or
follows from (the proof of) the following result.
Theorem 4.5.23 Let M be any 1-adic local field or any number field. Let
G be a finite group. There is a split-surjective, M-linear map, defined by
the construction of 4.5.19-4.5.21,
AG : (H) M[Hab]WGH
H<G
M{G}.
145
Let i : J --> G be the inclusion of a subgroup, then the following diagram

commutes:
A"J
M{G}
Proof The preceding discussion, which applies equally well to number

fields as to local fields, establishes the existence of AG. It remains to
prove the commutativity of the diagram.
From 2.2.8(i) we know that there is a commutative diagram
R(G)
AG
(H)R(Hab)WH
Resi
Res
R(J)
AJ
O(I)R(I ab)WJI
where
4.5.24
lab
N*) = E,EJ\G/I((J n
zlz-1)ab (z
A) N').
Therefore we have only to show the commutativity of the follow} are induced by the maps
of 4.5.13 and 4.5.10 when G is set equal to H and I, respectively:
ing diagram,in which the {(Hob} and {ip

M[Iabi WJI
'4tab
p(I)Homu,,(R(I ab)wJI, N)
(Restt )"
HM[Hab]WGH
H.b
(H)HomnM(R(Hab)WGH,N)
146
Table 4.2. Character table of Q8

1
L1
L2
L1L2
x2
-2
x
y
-1
-1
-1
-1
0
0
-1
-1
xy
If y C M[Iab]WJI we have
((H)4H-b)(Resj')*(WI-b(Y))
= ((D(H)4Hab)(Resj)'((A :jab
=(
(H)Y'Hab) 1 L.
_ ((H)OHab)
zEJ\G/H
JnzHz-l=1
CzHz 'zI
N*)'-' )(Y))
((p' : Hab
=EJ\G/H ((p :
Hab
N*) H
Ii(Z-1yz)))
II
- N") H p(Z-1yz)))Jl
so that the (H)-component of this map is given by

OH.b C zEJ\G/H
zHz-' >I
_ E zEJ\G/H
zHz 1
((p : Hab
N*) H (Z-1YZ)))
fs(Z-1YZ)YS
ES
= I NDi (Y)(H),
as required.
Let G = Q8 = {x, y I x2 = y2, y4 = 1, yxy-1 = x 1 } be

the quaternion group of order eight. We may choose one-dimensional
representations, L1 and L2 of Q8, together with an irreducible twodimensional representation, v, so that the character values of Q8 are
Example 4.5.25
given by Table 4.2.
147
From Table 4.2 one finds that
I = (1 + Li + L2 + L1L2 + 2v)/8,
x2=(1+L1+L2+L1L2-2v)/8,
(1-L1+L2-L1L2)/4,
y=(1+L1-L2-L1L2)/4,
and
zy = (1 - L1 - L2 +L1L2)/4.
The homomorphism, aQ8 : R(Q8) -* R+(Q8), is given by
AQ8(L) = (Q8, L)Q8
for L = 1, L1, L2 or L1 L2 and

aQ8(v) = ({x}, A1)Q8 + ({y}, A2)Q8 + ({xy}
A3)Q8
- ({x2}, A4)Q'
where Al, A2, A3, A4 are all injective homomorphisms into the fourth roots
of unity in N. From this, one may calculate that AQ8, which is defined
on the Weyl group invariant elements of M{Q$b} M{x} M{y}
M{xy} M{1} and maps to M{Q8}, is given by the following formula.
On M{Q8b}, if x,y are the images of x and y then
AQ8(l) = (1 + x2)/4,AQ8(x) = x,AQ8(y) = y and AQ8(z y) = xy.
On M{x}, if z = (1 - x2)/4, then

AQ8(1) = z,AQ8(x) = iz,AQ8(x2) = -z and AQ8(x3) = -iz.
On M{y}, AQ8(y`) is equal to the image of x' under the map, AQ8, on
M{x} and similarly for AQ8 on M{xy}. On M{x2}, AQ8(1) = 2z and
AQ8 (x2) = -2z. Finally, AQ8 is trivial on M{ 11.
Whenever one is given a functor of G which has a

Hom-description in terms of R(G) then one can make a map, induced
Remark 4.5.26
by AG, which is analogous to 4.5.2 1. For (C2'((9K[G]) and D(OK[G]) this

analogue is called a restricted determinant and will be described in 5.1.4.
Definition 4.5.27 For the remainder of this section we shall suppose
that M/Q1 is an unramified extension of local fields and that G is a finite
1-group.
Define 4m(G) to be equal to the kernel of the natural map from

OM [G] to OM [Gab],
148
.CM(G) = ker((9M[G] -) (9M[Gal])
4.5.28
Therefore CM(G) c J, the Jacobson radical of (9M[G], which was

introduced in 4.3.19. We are going to study the subgroup Det(1+dM(G))
of HomOM(R(G), (9N). We recall a result of C.T.C. Wall (1979).
Proposition 4.5.29 Det(1 + .QlM(G)) is torsion free.
We may define a natural map, aG, by means of the following

commutative diagram of G(M/Q,)-maps:
4.5.30
Det(1 + dM(G))
HomnM(R(G),1 + l(9N)
M{G}
HomoM(R(G), N)
HomcM(R(G), ION)
The lower map in this diagram is the injection induced by the inclusion
of l(9N into N. Here F is the Frobenius of 4.3.7 and Fip'/l is the map
given by 4.3.10. The map ip is 4.5.13, whose inverse is 0 of 4.5.10.
Lemma 4.5.31 If w E 1 + ION then log((o) = 0 if and only if co is an
1-primary root of unity.
Proof Firstly we note that if w is an l-primary root of unity then log(co)

is defined and is zero by 4.3.17. Conversely, there is an integer, m, such
that cor E 1 + 12(9N so that
w
exp(log((oi'"))
= exp(lmlog((o))
= exp(0)
=1.
149
Proposition 4.5.32 The G(M/Q,)-map of 4.5.30
aG : Det(1 + dM(G)) -+ M{G}

is injective.
Proof When G is abelian there is nothing to prove; therefore we will

assume that G is non-abelian.
Let U E 1 + 4M(G) and suppose that aG(Det(u)) = 0. This occurs if
and only if the homomorphism
T H to g (
Det(F(u))W'(T)
(Det(u)(T))'
is zero for all T E R(G). Since R(G) is finitely generated, 4.5.31 implies
that there exists an integer, m, such that Det(u)" = Det(ur) is in the

kernel of (FW'/l). However, since Det(1 + CM(G)) is torsion free, it will
suffice to show that the kernel of (FW'/l) is trivial. Assume, therefore,
that (FW'/l)(Det(u)) = 1 and, furthermore, since aG is a G(M/Q1)-map,
we also have
(FWD/l)(Det(Fj(u))) = 1
for all j >_ 0. This means that we have equations of the form
Det(F(u))(W'(T)) = Det(u)(T)',
Det(F2(u))(W!2(T))
=Det(F2(u))(W'(W'(T)))
= (Det(F(u))(W'(T)))1
= (Det(u)(T))'2,
and, in general, that for all T E R(G)
Det(F"(u))(W"(T)) = (Det(u)(T))".
Now suppose that #(G) = l' so that
W'"(T) = dim(T) E R(G).
This means that, if e : (9M [G] -> (9M [{ 1 }] _ (9M is the augmentation
map, then
Det(u" )(T) = Det(e(F" (u)))di,(T )
=1
since e(1 + 4M(G)) _ {1}. Therefore Det(u)1' = 1 and so Det(u) = 1, by
4.5.29, which completes the proof.
150
Recall from 4.5.5 that there is a G(M/Qj)-map, c : M[G] ->

M{G}, which sends a group element to its conjugacy class. Therefore
4.5.33
c(dM(G)) is an (9M-submodule of M{ G}. In particular, I-c(dM(G)) is also

an (9M-submodule of M{G} which is preserved by the G(M/Q1)-action.
Theorem 4.5.34 The map, aG, of 4.5.30 induces an isomorphism
aG : Det(1 + dM(G)). -> I c(dM(G))
Proof We begin by proving that Im(aG) lies in
c(dM(G)).
Let
u E 1 + sa7M(G) then we may define, as in 4.3.23,

Lo(u) E (9M [G],
which, by naturality, lies in sim(G) so that we have

Lo(u) E Wm(G).
By 4.3.28 the image of Lo(u) in M{G} lies in I- (9M{G}, where (9M{G}

denotes the free (9M-module on the conjugacy classes of G. The map, gyp,
of 4.5.13 is given by
'p
(>myy) (T) = Em,Trace(T(Y)) = ST X myY)
is the map of 4.3.24. Therefore Proposition 4.3.25 translates

into the statement that
where
bT
aG(Det(u)) = c(Lo(u)) E I. (9M{G}.
Hence
aG(Det(u)) E I - c((9M[G]) n c(dM(G)) = I - c(,qtM(G)),
which completes the first part of the proof.

Now we must prove that aG maps onto l c(slM(G)). We shall do this
by induction on the order of G.
Choose a central element z E G which is of order I and which is a

commutator. We shall prove next that
4.5.35
c(Lo(1 - (1 - z)(9M[G])) = l c((1 - z)(9M[G]).
If x E (9M[G] then, since z is central, F((1 - z)x) = 0 because z1 = 1.

Therefore
Lo(1 - (1 - z)x) = l(E' 1(1 - z)"x"/n)

l(1 - z)x - (1 - z)1x'
(modl(1 - 2)2x2)
151
but (1- Z)' E l(1- z)(9M[G] so that
Lo(1-(1-z)x)E 1(1-z)(9M[G]
and therefore c(Lo(1 - (1 - z))(9M[G]) c I c((1 - z))(9M[G]).
If X E J then one sees that
lc((1 - z)x) E c(Lo(1- (1- z)(9M[G]))
by means of a standard approximation argument ( see 4.7.13 ).

Hence we claim that
c((1 - z)(9M[G]) c c((1 - z)J),
4.5.36
which will complete the proof of 4.5.35. Write z =alb-lab for a, b E G

and let v = E myy E (9M [G]. We have to show that
c((1 - z)v) E c((1 - z)J).
4.5.37
Rewrite v as v =
my(y - a) + E mya so that
c((1 - z)v) = c((1 - z)(E my(y - a) + > mya))

= c((1 - z) E my (y - a)),
since c(a-za) = c(a-b-lab) = 0. However, y-a E Ker((9M[G] -> (9M),

the kernel of the augmentation, and since G is an I-group the kernel of
the augmentation lies inside J.
Finally, we prove the surjectivity of aG by induction on the order of
G. Set H = G/{z} and assume that aH is surjective.
The group, {z}, acts (by multiplication in G) on (9M[G], CM{G}, MEG]
and M{G}, since z is central. It is easily verified that
Ker((9M{G} - (9M{H}) = (1- z)(9M{G}
and that
Ker(M{G} " M{H}) = (1 - z)M{G}.

Therefore, if we are given y E 1 c(dM(G)) then we know that there exists
v E 1 + dM(H) such that

4.5.38
iv(y) = aH(Det(v)) = c(Lo(v)) E 1 c(dM(H)).
Since (9M [G] is local we may lift v to v' E 1 + .CM(G), using the fact
that G -+ Gab factors through H. Thus
ao(Det(v')) - y E Ker(ir :
c(&M[G]) -> I c((M[H]))

= 1 Ker(xc c((PM[G]) -' c((M[H]))
l
152
so that
aG(Det(v')) - y E l c((1 - z)(9M[G])
and, by 4.5.35, there exists v" E (9M [G] with
aG(Det(v')) - y = c(Lo(1 - (1 - z)v"))

= aG(Det(1 - (1 - z)v")).
Since 1 - (1 - z)v" E 1 +dM(G) we have shown that y E Im(aG), which

completes the proof of Theorem 4.5.34.
We shall now close this section by proving the following weak form of
Theorem 4.5.3, from which one may easily deduce Theorem 4.5.3 by the
argument of Frohlich (1983, pp. 85-89).
Theorem 4.5.39 Let M/Q1 be an unramified Galois extension of local
fields. Let G be a finite l-group. Then
Det((9M[G]*)G(M/Q') - Det(Zj[G]').
Proof Consider the following diagram, whose rows are easily seen to be
short exact (that is, the left map is injective and the right one is surjective,
while the kernel equals the image in the middle):
1 + dM(G) - (9M [G]'

Det
Det(1 +dM(G))
{1}
Det
, (9M [Gab].
= Det
Det((9M[G]") - Det((9M[Gad]*)
{1}
{1}
in which the vertical maps are induced by the determinant, which is an

isomorphism for abelian groups.
Let H = G(M/Ql) then we may compare the bottom row for Qj with
the H-invariants of the bottom row for M.
4.6 Adams operations and determinants
Det(1 +dQ,(G)) - Det(ZI[G]*)
153
- Det(ZI[Gab]*)
/32
/33
Det(1+s1M(G))H -- Det((9M[G]*)H - Det(l9M[Gab]*)H
The map, N3, is an isomorphism because

(LM[Gab]*)H ti ZI[Gab]*
and therefore the lower sequence is short exact. However, by 4.5.34, /31
may be identified with the isomorphism
1- c(,4Q,(G)) = I . (c(dM(G)))H,
so that /32 is an isomorphism, by the five-lemma, which completes the

proof.
As in the previous section let M/QI be a finite Galois extension

and suppose, as in 4.5.39, that this extension is unramified. Let G be
a finite group and let N/QI be a `large' Galois extension, as in 4.3.1,
which contains M/Qj. For any integer, h, we have the Adams operation,
Wh : R(G) -+ R(G), of 4.1.1 and an induced homomorphism
4.6.1
4.6.2
Wh : HomoM(R(G),N*) --+ HomcM(R(G),N*)
given by
Wh(f)(X) =f(Wh(X))
for all X E R(G).
In this section we shall be concerned with the proof of the following

result, which is originally due to Ph. Cassou-Nogues and M.J. Taylor and
is described in chapter 9 of Taylor (1980).
154
Theorem 4.6.3 Let Det((9M [G] *) c HomfM (R(G), N`) denote the image of
the determinant homomorphism of 4.5.2 then, for all 0 < h E Z in 4.6.2
Wh(Det((9M[G]')) c Det((9M[G]').
Remark 4.6.4 (a) Theorem 4.6.3 will be proved in a series of steps, of

which (as was the case with Theorem 4.5.3) the most difficult is the case
in which G is an /-group. In the previous section we proved only the

1-group case of Theorem 4.5.3, claiming that reduction to that case is not
difficult. In this section we shall carry out the analogous reduction to the
1-group case. The assiduous and independently minded reader is invited
to use this pattern of reduction to derive Theorem 4.5.3 from Theorem
4.5.39.
Here and there our proof of Theorem 4.6.3 will differ considerably
from that of chapter 9 of Taylor (1980) by virtue of the fact that we
have available better determinantal congruences; in particular, we do not
require the use of the decomposition homomorphism (cf. Taylor, 1980,
p. 106).
(b) The main application of Theorem 4.6.3 is to show that the Adams
operation, Wh, induces an endomorphism of the class group
Wh : WY((9K [G]) -> WY((9K [G])
These endomorphisms exist when K is a number field which is unramified
at the prime divisors of #(G) and when, in addition, h is odd or G has

no irreducible symplectic characters. This last condition is required to
take care of the infinite places in the Hom-description of 4.2.28. It is
necessary in order that Wh should preserve the subgroup of R(G) which
is generated by symplectic characters, which in turn ensures that Wh
preserves the Hom-group, Homy (R(G), E') of 4.2.33. For further details
the reader is referred to Taylor (1980, p. 102).
Let us first deal with a very trivial case of 4.6.3.
Lemma 4.6.5 Theorem 4.6.3 is true when HCF(l, #(G)) = 1.
Proof In this case (9M[G] is a maximal order in M[G] (see 4.7.3) and
therefore, by Frohlich (p. 23),
Det((9M [G]') = Homo,, (R(G), (9N ),
which is clearly mapped to itself by Wh
155
Lemma 4.6.6 Let p be any prime (not necessarily equal to the residue
characteristic, 1) and let G be a finite p-group. If
aG : Det((9M[G]') -+ M{G}
is the homomorphism defined in 4.5.30 then
aG(Det((9M[G1`)) c l(9M{G}.
Proof Let J 1 (9M [G] denote the Jacobson radical. If r E J then,

by the argument which was employed in the proof of Theorem 4.5.34,
aG(1 + r) E l(9M{G}. Hence aG induces a map
4.6.7
G]'
a : GM
(i+r)
MG
10m (G)
The image of & must be a finite, abelian l-group and, since when l = 2
the group G has no homomorphic image equal to D6, & must be trivial,
by the argument which was used in the proof of Theorem 4.3.37.
Lemma 4.6.8 Let G be a finite l-group. Suppose that
f E Ker(HomoM(R(G),N') -+ HomcM(R(Gad),N"))
satisfies
Pop, (f)) = f`,

where, as in 4.4.2, P extends the Frobenius in G(M/Q1) and is trivial on
the l-primary roots of unity. Then, for some m >_ 1,
f"
= 1.
Proof We use an argument from the proof of 4.5.32. By that argument

there exists an integer, t, such that
f(dim(T)) = Pt(f(dim(T))) = f(T)"

for all T E R(G). However, since f lies in the kernel of the abelianisation
map, f(dim(T)) = 1, which completes the proof.
Lemma 4.6.9 Let 1..(N) denote the subgroup of N* consisting of I-primary
roots of unity. If M/Q1 is unramified then for any group, G,
Homo,, (R(G),l_(N)) = HomoQ (R(G), pi-(N)).
156
Proof Since f1m is a subgroup of % we have an inclusion of HomQQ

into HomoM. However, since M/Qi is unramified, f1m and F generate
%,. Since F acts trivially on ul.(N) it also acts trivially on R(G) and
therefore any QM-equivariant map, h : R(G) -> pl.(N), is automatically
an QQ -map.
Definition 4.6.10 Let G be any finite 1-group. Suppose that M1 > M > Q1
is a chain of finite, unramified Galois extensions. Define the norm

homomorphism
NM,/M : Homo,,, (R(G), N') - -> HomoM(R(G), N")
4.6.11
by
NM,/M(f) =
fi
g(f),
gEG(Mi /M)
where we identify G(M1 /M) with S M/cM, .

Lemma 4.6.12 Under the circumstances of 4.6.10
NMl/M(Det(PMI [G]*)) = Det((9M[G]')
for any 1-group, G.
Proof As in the proof of Theorem 4.5.39 we have two horizontal short

exact sequences in the following commutative diagram:
4.6.13
1. (9M{G}
TrM,/M
Det((9M, [G]') - Ml [Ga! ]*
NM1 /M
(9M,{G} - Det((9M[G]')
NMI lm
M[Gb]'
In 4.6.13 the trace map on the left is surjective, because M1 /M is

unramified (Serre, 1979), and so is the right-hand norm map (from the
157
case when G is abelian; see 4.7.16). A diagram chase completes the proof.
Proposition 4.6.14 Let G be a finite l-group. For any integer, h,

Wh(Det((9M[G]*)) c Det((9M[G]*).
Proof Let X = E ayy E (9M[G]* and let Y = E ayyh. Since the Jacobson
radical is preserved by the function induced by sending y (y E G) to yh
and since ((9M[G])/J is finite we see that Y E (9M[G]*. Therefore we may
consider
4.6.15
Wh
f=
" Det x
Det(Y)
E HomnM(R(G), N*).
Notice that f lies in the kernel of the abelianisation map of 4.6.8, since
on an abelian group is induced by the hth power map.
Now let us introduce the temporary notation
4.6.16
B = (FWl/l)-'(HomoM(R(G),1 + 1' (9N)),
where F is the extended Frobenius of 4.4.2 and 4.6.8 and (FW'/l) is

the endomorphism of HomnM(R(G),N*) which was introduced in 4.4.3.
Therefore, as in 4.5.30, we may define a homomorphism
4.6.17
an = W-'(log(FW'/l)) : B -> M{G},
which extends the map of 4.5.30

aG : Det(1 +dM(G)) ---+ M{G}.
Since Wt and Wh commute we find, from the determinantal congruences

of 4.3.10, that
4.6.18
f E B n ker(HomnM(R(G), N*) -> HomOM(R(Gb), N*)).
By 4.6.6, aG(Det(X)) and aG(Det(Y)) both lie in l (9M{G}. Hence we

may write
4.6.19
ao(Det(X)) = 1[y] /3[y] [y] E M{G}
where [y] runs through the conjugacy class of G and //[y] E l

Therefore, for T E R(G),
W(aG(Det(X)))(Wh(T)) _
Q[y]Trace(T(yh))
[y]
(9M.
158
and
1p(aG(Det(X)))(Wh(T)) = log(FW'/l(Det(X)))(Wh(T))
= log(FW'/l(Wh(Det(X))))(T),
= W (aG(Wh(Det(X))))(T ),
since Wh commutes with FW1.
Therefore
aG(Wh(Det(X))) =
a[Y] [yh] E l (M{G}

[Y]
and we find that

aG(f) E I (9M{G} nker(M{G} --> M{Gab})
and, by Theorem 4.5.34, there exists
u E Det(1 + dM(G))
4.6.20
which satisfies
aG(f l u) = 1.
By the properties of log (see 4.3.16) we may apply 4.6.10 and 4.6.11 to
f /u E HomnM (R(G), N') (or to (f /u)2 when l = 2) to conclude that
4.6.21
f/u E HomnM(R(G),pt-(N)) = HomoQ (R(G),j_(N)).
Consider the operation which assigns to Det(X) the l-primary torsion

element, flu, of 4.6.2 1. If both
Wh(Det(X)) W"(Det(X))
E HomnM(R(G),j-(N))
Det(a)
Det(fl)
for a, /3 E (9M [G]" then
Det(a//3) E Det((9M[G]*)
HomnM(R(G),jc(N))
which is trivial, by a result of Wall (1979) see also Taylor (1980, p. 45).
Hence we obtain a homomorphism
4.6.22
I'M : Det((9M[G]*)
HomnM(R(G), pj-(N)) given by
I'M(Det(X)) = f/u.
We must show that f /u = 1 to complete the proof. Suppose that

Homo,, (R(G),l-(N)) has exponent le. Let M1/M denote the unique
unramified extension of degree le. Since the norm maps commute with
Wh we obtain a commutative diagram in which the vertical maps are
norms.
159
4.6.23
FM,
Det((9M, [G]`)
Homo,,,, (R(G), to(N))
FM
Det((9M[G]')
HomnM(R(G), p1_(N))
In 4.6.23 the left-hand norm is surjective while the right-hand norm is

the leth power map, by 4.6.9. Hence there exists V E Det((9M, [G]') such
that
f/u = I'M(Det(X))
= I^M(NM,/M(V ))
= NM1/M(FMi(V))
= (FM, (V )f
= 1,
which completes the proof of Proposition 4.6.14.

Definition 4.6.24
Let p be a prime. A finite group, G, is called Q-
p-elementary (or p-hyperelementary) if it is a semi-direct product of the

form, H c C, where H is a p-group and C is a cyclic group whose order
is prime to p. Thus G contains C as a normal subgroup and is generated
by C and H with G/C = H.

Let m be a divisor of #(C) and suppose that x : C --> N" is a character
of order m. Set
4.6.25
Hm = ker(H -) Aut(x(C)))
so that Hm is the kernel of the conjugation action by H on C and thence

on x(C). Also set
4.6.26
A. = H/Hm,
Gm = HmC .4 G.
Lemma 4.6.27 Let h be any positive integer.

Let G be a Q-p-elementary group, as in 4.6.24. Let a E (9M [G]' then
there exists b E (9M[G]' such that, for all a e R(C) and On E R(Hm) (where
m divides #(C)),
Det(b)(IndgM(aOm)) = Det(a)(IndGGM
160
Here a6 is the representation which is additive in a and when a is onedimensional is given by the extension of 0 : Hm -> GL(V) to G = HmC
by demanding that c E C acts via scalar multiplication by a(c).
Proof For x E C and y E H we may define a homomorphism, p : G -->

G, by p(xy) = xh y. Hence p induces a ring endomorphism of (9M [G] and
we may set b = p(a). It is sufficient to verify the relation when dim(a) = 1.
Since G/Gm = H/Hm we may choose coset representatives of G/Gm of
the form yl,... , yt ( yi E H). If g = xy (x E C, Y E H) then there is a

homomorphism
a:G->G/C->Et
such that
gyj = YQ(g)(i)h(i, g) = YQ(v)h(i, g),
where h(i, g) E Hm. In addition,

h(i, g) = h(i, Y)(y, 'yxyy ),
so that, by the formula of 2.1.9,

I ndGm (ip"(a)9)(g)
a(Y1 lY 1xyy1)h8(h(1,Y))
0
0
a(Y2 lY-1xYY2)h0(h(2, y))

0
= o(g)
= I ndQ (c 0)(xhy),
which establishes the required formula.
Let G = H a C be a Q-1-elementary group. For each

divisor, m, of #(C) let m = exp(27ci/m) and let A1, ... , 2k(m) : C -> N'
be a complete set of 52M-orbits of abelian characters of order m. The
group, Am, of 4.6.26 acts by conjugation to permute the members of the
Definition 4.6.28
set {21,...,Ak(m)}.
Define a map
161
4.6.29
k(m)
HomnM(R(G), (9N) -' f fl HomnM( )(R(Hm), (9 )

ml#(C)
i=1
by am,i(f)(6m) = f(IndGm(2iOm)) for all em E R(Hm).
From Taylor (1980, p. 108 and p. 70, section 3.9) we have the following
result.
Proposition 4.6.30 In the notation of 4.6.28 the map, a, of 4.6.29 induces
an isomorphism of the form
k(m)
a : Det((9M[G]*) - f
ml#(C)
i=fl1
Proposition 4.6.31 Let G be a finite Q-l-elementary group. For any integer,

h,
Wh(Det(C9,y[G]*))
Det((9M[G]*).
Proof Let a, b E (9M [G] * be as in 4.6.27. We are going to show that, for
each m I #(C) and for each 1 < i < k(m), in 4.6.29
am,i(Wh(Det(a))) = Wh(am,i(Det(b)))
By 4.6.14 and 4.6.30, this will establish Proposition 4.6.31.

Suppose that Om E R(Hm) and that, as in 2.3.9,
aH,"(em) _ E nJ(VJ,
A,J)H"
E R+(Hm)
Therefore, by 4.1.6, for all t,

4.6.32
WI(Bm) = J:J nJIndy"(4J) E R(Hm).
Also, if 2m,i : C -+ N* is a character of order m then )1m,i0m E R(Gm)
is a one-dimensional character such that, by a slight extension of 2.5.12,

E nj(CVJ, ).m,,Oj)G' E R+(Gm)
J
Therefore, for all t,

4.6.33
Wt(tm,iOm) _ EJ njIndcv;(A;,,,,O`J) E R(Gm).
162
Hence we have
am>i(W'(Det(a))(em))
by 4.6.33 and 4.6.29,
= rj;[Det(b)(IndGm(Ah Ind5"(0h)))]";
by 4.6.32,
= Wh(am,i(Det(b)))(0m),
as required.
Proposition 4.6.34 Let h be any positive integer. Let p be a prime, p * 1,

and let G be a Q-p-elementary group. There exists an integer, m, such that
{zim
Z E Wh(Det((9M[G]"))} c Det((9m[G]').
Proof By definition (cf. 4.6.28) we may write G = H oc (C x C'), where

H is a p-group, C is a cyclic 1-group and C' is a cyclic group whose order
is prime to lp. We will prove the proposition by induction on the order
of C. When C is trivial we may take m = 0, by 4.6.5. Now let us assume
that #(C) # 1.
Every irreducible character of G is one of the form (cf. Serre, 1977,

section 8.2) V = I ndjG(a), where C x C' c J and dim(s) = 1. By 4.7.17,
W'(V) = Wi(I nd j (a)) = I ndj(ai). Let C" c C' denote the subgroup
of order 1. Hence C" a G and al is trivial when restricted to C".
Therefore ip'(V) is a representation of G which is trivial on C". Let

iv : G --p G/C" be the quotient map, so that we have established that
ii(R(G)) is contained in the image of n' : R(G/C") - R(G).
Let z E (9M [G]' so that, by Theorem 4.3.10,
Wi(Det(F(z)))
E HomnM(R(G),1 + l(9N).
(Det(z))'
Since Wh commutes with Wi (and with the Frobenius, F),

Wi(Wh(Det(F(z))))
E HomnM(R(G),1 + 1(9N).
(Wh(Det(z)))i
163
If V belongs to R(G) and W'(V) = x`(V') with V' E R(G/C") then, by

induction, there exists an integer, t, and z' E (9M [G/C"] ` such that
{WI(Wh(Det(F(z))))}It(V )
_ {Wh(Det(z))}tt(ir`(V))
= Det(z')(V').
Note that, by induction, t may be chosen so as to work for all z E (9M[G]
We may choose z" E &M[G]` such that n(z") = z' so that
{Wl(Wh(Det(F(z))))}!t(V )
= Det(it(z"))(V')
= Det(z')(ic`(V'))
= W'(Det(z"))(V )
Using Theorem 4.3.10 once more on

W`(Det(z"))
(Det(F-i(z" )))'
we find that, for all z E (9M[G]`,

(Wh(Det(z)))l`+`
E HomcM(R(G),1 + 1(9N)-
Since 1 + ON is a pro-l-group, so is HomnM(R(G),1 + l(9N), so that any

finite quotient group is a finite 1-group. The group
4.6.35
[Det((9M[G]`)HomcM(R(G),1 +l(9N)]/(Det((9M[G]`))
Wh(z)I'+'+'
is finite, so that there exists an integer, s, such that

is trivial in
4.6.35 for all z, which completes the proof of Proposition 4.6.34.
4.6.36 Determinants and KI((9M[G])
We are almost ready for the final step in the proof of Theorem 4.6.3.
In order to complete that step we shall require a small digression into
algebraic K-theory. Further details on this material is to be found in
Curtis & Reiner (1987, p. 61 et seq.).
164
Let A be a ring with a unit. We may embed the group,

invertible matrices in M (A) into
of
by sending X to
x
C0
O
1
In this manner we may form the infinite general linear group,

GL(A) = (J, GL,(A)
4.6.37
and thereby define an abelian group, K1(A), by the formula

4.6.38
K1(A) = GL(A)ab = GL(A)/[GL(A), GL(A)].
It is a result of J.H.C. Whitehead that the commutator group,

[GL(A), GL(A)],
is equal to the group generated by the elementary matrices. In particular,

since (9M[G]' is equal to GL1((9M[G]), the determinantal construction of
4.3.3 extends to give a commutative diagram of the following form:
4.6.39
(QM [G]' )
Det
HomQM(R(G), (9; )
Det
K1((VM[G])
In addition, K1((9M[G]) admits restriction and induction maps which

make the following diagrams commutative, for all H < G.
4.6.40
Det
HomOM(R(H), (9N)
Det
K1(&M[G])
HomnM(R(G),(9N)
165
4.6.41
Det
K1(OM [G])
HomnM, (R(G), ON)
(I nd' )'
ResH
Det
HomcM(R(H), ON)
K1((OM [Hl)
In 4.6.40 and 4.6.41 f ` is the map induced by the map, f , of representation rings.
Proposition 4.6.42 Let Q, < M < N be as in 4.6.1 and let 0 : G
GL (M) be a representation. If z E (9M [G]" then the map, (X F-+ Det(z)(0
X) E Homo,, (R(G),(ON) lies in the subgroup, Det(OM[G]").
>
Proof The key to this result is the fact that (Curtis & Reiner, 1987;
Wall, 1966) Det(Ki((9M[G])) = Det((OM[G]*). If Z = Eayy then 0(z) =
E a,O(y) E GL (OM [G]) and therefore gives rise to a class, 0.(z) E
Kl(OM[G]). It is clear that the function, (X '--> Det(z)(0 X)), is equal
to (X'--> Det(0.(z))(X)), which lies in Det(KI(OM[G])). This completes the
proof of 4.6.42.
This proof uses the partial results concerning Q-p-elementary groups

(4.6.31-4.6.34) to derive the general case.
Let p be a prime and let G be an arbitrary finite group. By Serre

(1977, section 12.6) and Curtis & Reiner (1981) there exists an integer,
n(p), which is prime to p and virtual Q-representations
{0, E RQ(J,); J; < G, J; Q - p - elementary},

such that
4.6.44
n(p) = >; I ndG (0,) E R(G).
If z E OM [G]" and T E R(G) we obtain, by Frobenius reciprocity

(1.2.39),
n(p)T = > I nd (O,Res (T )) E R(G)
166
and
(W"(Det(z))(T))n(P)
Det(z)(Ind (OjRes (W"(T))))
_ fi Det(Res% (z))(OiW"(Res (T))),
by 4.6.41, where
E K1((9M[Ji]).
u; E (9M[Ji]*, such that, for each i,
By 4.6.42 there exist units,
Det(Resj'(z))(OiW"(Res'(T))) = W"(Det(ui))(Res'(T)).
If p = I then, by 4.6.31 applied to Ji, there exists vi E (9M[Ji]*, which

satisfies
W"(Det(ui))(ResjG(T ))
= Det(vi)(ResJGi (T))
= Det(Ind5 (vi))(T),
so that (W"(Det(z)))"(') E Det((9M[G]*).
However, if G is a Q-p-elementary group with p * l then, by 4.6.34,

there exists an m such that (W"(Det(z)))'" E Det((9M[G]*). Since
HCF(lm,n(l)) = 1 we find that (W"(Det(z))) E Det((9M[G]*) when G
is Q-p-elementary with p * 1.
Finally, let us return to the general case and suppose that n(l) _
pit ... p"r where pl,... , pr are distinct primes. Repeating the first step with
Q-l-elementary groups replaced successively by Q-pi-elementary ones

we find a set of integers, n(pi), which satisfy HCF(n(pi), pi) = 1 and
(W"(Det(z)))n(P1) E Det((9M[G]*). Therefore HCF(n(l),n(pi),...,n(pr)) =
1 and so (W"(Det(z))) E Det((9M[G]*), which completes the proof of
Theorem 4.6.3.
4.7 Exercises
4.7.1
If HCF(k, #(G)) = 1 show that

Wk(I ndG (p))
4.7.2
= I ndH(Wk(p))
Let T be the binary tetrahedral group
T = {a, b, c I a2 = b2 = c3, c6 = 1, bab-1 =a-', aca 1 = bc, ac = cb}.
4.7 Exercises
167
T consists of the 24 elements
T ={ah'bick;0<h<4,0< j <2,0<k<3}.
Also T/{a2} = A4. Define a two-dimensional, faithful representation,
v:T --) SU(2), by

v(a) _
-i
v(b) _
and
v(c)
1(
e7
f7
e5
where e = (1 + i)/.J2, i2 = -1.

Verify the following formulae in R(T):
(i) W2(v) = I nd ,>(1) - I nd<c>(1) -I nd<a>(1),
(ii)
4(v) = w 2(v) - I nd<az>(1) + 21 nd<a>(1)
and
(iii) tp6(v) = I nd>(1) - I nd<a>(1).
4.7.3
Let K be a local field whose ring of integers is (9K. Let G be a
finite group with #(G) = n. Prove that (9K [G] is a maximal order in K [G]
if n is invertible in (9K (cf. Curtis & Reiner, 1981, p. 582).
Prove, under the action of 52Q on R(G), that complex conjugation

fixes every symplectic representation of G.
4.7.4
Explain, to your own satisfaction, how Theorem 4.2.35 is derived

from Theorem 4.2.28.
4.7.5
Let #(G) = n and suppose that k = 1 + mn for some integer, m.

By considering 1 +am E Zp[G], when pI#(G), show by means of 4.2.28
4.7.6
that S(k) = 0 E '2'(Z[G]).

4.7.7
By means of the Hom-description show that
S(kk') = S(k) + S(k') E'e.(Z[G]).
168
4.7.8
Prove 4.7.6 and 4.7.7 directly in terms of Z[G]-modules. For
example, find a Z[G]-module isomorphism of the form

Z[G] S(kk') = S(k) S(k').
Let C denote the cyclic group of order n with generator, x. Show

that, if HCF(k, n) = 1, then 1 + x +... + xk-1 E Z1 [Cn]" for all primes, 1,
which divide n.
4.7.9
4.7.10
Let Den denote the dihedral group of order 2n,

D2 = {a, b I an = 1 = b2, bab = a 1 }.
Let uk = 1 + Ek 1 a` + jk o a`b E Z[D2 ].

Assume that 2k + 1 is coprime to n and let l be a prime dividing 2n.
(i) If k = to show that uk E Z1 [D2 ]
(ii) Suppose that (1 +v)uk = uk+r(2k+1) for some v E Z[D2n]. Prove that
.
(1 + v + aJ b + a1+1)Uk = uk+(r+1)(2k+1)
for j = 2k + r(2k + 1).
(iii) Prove that Uk E Z1 [D2n]"
4.7.11 (See Martinet, 1977a, p. 527; see 5.2.13.) Let G = Qs = {x, y I

x2 = y2, y4 = 1, yxy-1 = X-11 be the quaternion group of order eight.
Let M be a projective Z[G]-module of rank one (i.e. Q (&Z M is a free

Q[G]-module on one generator). Set M+ = {m E M I x2(m) = m} and
M_ = {m E M I x2(m) = -m}.
Hence M+ is a projective module over the ring R = Z[G/{x2}] _

Z[Z/2 x Z/2] and hence is free of rank one. Also M_ is a free, rank one
module over S = Z [i, j, k], the integral quaternions.
Let M+ have an R-basis, 4), and M_ have an S-basis, T. Prove that
either (a)
(D _ `P (mutually 2M)
(D _ (x + y + xy)W (mutually 2M).
or (b)
Prove that M is free if and only if (a) holds. Hence prove that
WY(Z[G]) = Z/2 = T(G) generated by the Swan module, S(3).
4.7.12
If H < G show that ResH maps T (G) to T (H) and that

ResH(S(k)) = S(k) E T(H).
(Hint: See Curtis & Reiner, 1987, p. 345.)
4.7 Exercises
169
Complete the details of the approximation argument to show, in

the proof of Theorem 4.5.34, that
4.7.13
c(Lo(1 - (1 - z)(9M[G])) >- lc((1 - z)J),

where J is the Jacobson radical of (9,y [G].
4.7.14
Let p be a prime and let G = (Fp)", where Fp (^= Z/p) is the
additive group of the field of p elements. Let xl,... , x, denote generators

for the individual Fp-factors of G. Suppose that HCF(k,p) = 1. Consider
U= 11
(1 +0(z)+...+0k-l(z)) E Zp[G]
[z)EP(FP)
where [z] = [zi,... , zn] runs through representatives of the projective

space of G and 0(z) = xi' ... xn^.
Use U to show that p"-i T (G) = 0 if p * 2 and that
2n-2T(G)
= 0 if
p=2.
4.7.15
Generalise the calculation of 4.7.14 to the group
G=(Fp, x...xFp,)n,
where {pi} are distinct primes.
4.7.16
Let G be a finite abelian 1-group and let Q, s M 9 Mi be a
chain of finite, unramified Galois extensions. Show that the norm
N : Mi [G]' -'
M[G].
(Mi
[G]*)G(M,IM),
given by N(x) = rlgEG(M,IM) g(x), is surective.

4.7.17
Let I be a prime and let J < G be finite groups. Assume that
the set of all l-primary elements of G forms a normal subgroup which is

contained in J. If a E R(J) show that
ip'(I nd (a)) = I nd (tp'(a)) E R(G).
5
Introduction
The determinantal functions play an important role in the Hom-description of the class-group of a group-ring. By applying the determinantal
construction to the components of the Explicit Brauer Induction formula
for a representation one may construct new types of homomorphisms,
which I have christened restricted determinants, defined on the representation ring, R(G), and giving rise to elements in the class-group. In Section
1 we describe these new maps and show that their images generate the
entire class-group.
In Section 2 we use some homological algebra to derive a new invariant,
which is a type of `reduced norm' and which detects elements in the
class-group of the quaternion group of order eight, W2'(Z[Q8]). This
class-group is a very popular place in which to test conjectures concerning
Galois module structure and I have found this invariant to be a little

more convenient than that of Martinet (1977a); see also 4.7.11) since it
does not require one to rewrite a projective module as an ideal plus a
free module in order to use it. As an example we give a new derivation,
in 5.2.33, of the formula for the element of the class-group which is given
by a finite abelian group of odd order upon which Q8 acts.
In Section 3 we derive some new relations between Swan modules, using
the geometry of some maps, originally due to Frank Adams, between

unit spheres in representations. These relations enable us, in some simple
cases, to obtain new upper bounds for the Swan subgroup of G. The
motivation for this section is probably more interesting than the results.
In the study of the stable homotopy groups of spheres one of the major
achievements was the evaluation of the subgroup which is called `the
image of the J-homomorphism'. All that we require to know about this
170
Introduction
171
image is that is it a computable subgroup of a very complicated group

and in that way is analogous to the Swan subgroup of the class-group
of a group-ring. In fact, the analogy goes further. In a famous series
of papers Frank Adams (1963, 1965a,b, 1966) showed how to reduce
this stable homotopy problem to a conjecture about Adams operations
and sphere-bundles. The estimation of lower bounds for the image of
the J-homomorphism by the use of Adams operations resembles very
closely M.J. Taylor's later method for the determination of the Swan
subgroup of a p-group (see 4.4.12). Pursuing this analogy leads one to
adapt to the case of the Swan subgroup Adams' method for obtaining
upper bounds for the image of the J-homomorphism by means of the
algebraic topology of some maps between spheres.
The class-group of a maximal order in the rational group-ring of a finite
group also has a Hom-description. In Section 4 we use a modification

of the restricted determinants of Section 1 to design a means whereby
to detect this class-group in some new Hom-groups which are defined
in terms of the cyclic subquotients of G. In Chapter 7, Section 3 these
detection homomorphisms are used to give a new proof of a result of
David Holland which states that the Frohlich-Chinburg conjecture is true
in the class-group of a maximal order. In this Hom description, as with
the Horn description of the `kernel group', one of the quotients consists
of a subgroup of adelic-valued functions on R(G) whose values satisfy
a total positivity condition on symplectic representations. The main

new result, which is necessary to make these detection homomorphisms
work, is the analysis of the Explicit Brauer Induction homomorphism,
aG, on symplectic representations (see 5.4.42 and 5.4.43). In the proof of
this result it was very convenient to use the more geometric version of
Explicit Brauer Induction, which was given in Snaith (1988b), to analyse
the behaviour of aG on symplectic lines.
After the calculation of the Swan subgroup of a p-group the next case
is the calculation of the Swan subgroup of a nilpotent group. Since

a nilpotent group is a product of p-groups this calculation is related
to a question of C.T.C. Wall (5.5.3) concerning the Swan subgroup of
the product of two groups of coprime order. For nilpotent groups this
problem can be reduced to the study of two primes at a time and
we determine the answer in the elementary abelian case by the use
of determinantal congruences (or, equivalently, the detection maps of
Chapter 4). We conjecture that the elementary abelian case suffices to
detect the Swan group of an arbitrary nilpotent group and verify this
conjecture for several examples which involve small primes. In these
172
examples we make use of some of the relations which were derived in

Section 3 in order to reduce the study of the Swan group of a product
with a semi-dihedral 2-group to the case of a product with the dihedral
group of order eight.
Section 6 studies the elements in the class-group of a cyclic group
which are given by p-power roots of unity acted upon by the cyclotomic
Galois group. These finite Galois modules are cohomologically trivial
and therefore define elements in the class-group of the integral groupring. These modules not only provide many examples of non-trivial
class-group elements but they also figure centrally in the computation,
carried out in Chapter 7, of the global Chinburg invariant of a totally real
cyclotomic Galois extension of p-power conductor. For these reasons,
a discussion of these elements seemed appropriate for inclusion in this
miscellany.
Section 7 consists of a collection of exercises, together with research

problems which are suggested by our results on the Swan subgroup.
5.1 Restricted determinants
In 4.5.16 we showed how to combine the Explicit Brauer Induction

homomorphism of 2.3.15 with the Hom-description of M{G} in order to
construct a split surjection of the form (see Theorem 4.5.23)
sl* : (H) M[HallWGH -f M{G}.
5.1.1
H<G
In this section we shall briefly describe an analogous construction of

maps onto the groups, W'((9K [G]) and D((9K [G]) of 4.2.28 and 4.2.29,
where K is a number field.
We maintain the notation of 4.5.16. In particular, if H < G then

WGH = NGH/H, where NGH is the normaliser of H in G.
Let E/K be a large Galois extension of number fields, as in 4.2.8.

Recall from 2.3.15 the Explicit Brauer Induction homomorphism
aG : R(G) -+ R+ (G)
given by the formula aG(p) =
#(G)-'
0, ResH,(p) > (H, 4))G,

in-#G
173
where (H, 4)G denotes the G-conjugacy class of a one-dimensional rep-
resentation, : H -- E. For each H1 < G we may define a homomorphism

5.1.2
detH, : J*(K[Hib])WwH'
HomQK(R(G),J'(E))
by the formula (a E J'(K[Hab])WGH, P E R(G))

5.1.3
detH,(a)(P) = li(H,,O)GE.,Ic/G det(4(cc))a(H1O)G E J*(E),
where, in 5.1.3, the product is taken over those (H, O)G in

aG(P) = E a(H,O)c(H, 4)G,
for which H = H1 (up to conjugation in G).

Note that 5.1.3 is well-defined because a is a WGHI-invariant idele
and 0 : H1 ---> E' factors through the abelianisation map, H1 Hl b. Passing to the quotient in 4.2.28 yields a restricted determinantal
homomorphism
5.1.4
DetH, : J'(K [Hab])WH'
Cey(OK [G])
for each subgroup, H1 < G.

Now let A(G), (9K [G] < A(G) < K [G], denote a maximal order. Let
J`(A(G)) c J*(K[G]) denote the subgroup whose elements have all their
coordinates lying in the completion, A(G)p, as P varies through the

primes of K. We remark that A(G) is well-defined up to conjugation by
a unit of K [G]. Also, when W is abelian then K [ W] is isomorphic to a
product of number fields by means of an isomorphism which is induced
by characters of the form, 0 : W -* E. Since the ring of algebraic

integers is a maximal order in a number field this means that when W
is abelian 0 induces a homomorphism, Det(O) : A(W)' -p (9E, and an
idelic homomorphism of the form
5.1.5
Detp : J*(A(W)) -> U((9E).
Therefore the homomorphism of 5.1.2 gives, by means of 4.2.35, a

restricted determinantal homomorphism
5.1.6
DetH : J'(A(Hab))WcH _, D(&K[G]).
Remark 5.1.7 (i) The restricted determinants were first introduced in

Snaith (1990b); see also Snaith (1990a). These homomorphisms enjoy
174
naturality properties similar to those of s in Theorem 4.5.23, which are

induced from the naturality property of aG (see 2.3.8(i)). We leave the
derivation of this property to the interested reader and refer to Snaith

(1990b) for further details.
(ii) As in the case for Det in 4.6.39, the restricted determinants

factor through the adelic algebraic K-groups, K1(J(K[H1]))WGH and
K1(J(A(Hb)))WGH where J(K[Hb]) is the adele ring of 4.2.7.
Theorem 5.1.8 Let G be a finite group and let K be a number field. Then
E DetH(J'(K[Hb])n'GH) = `Wy((9K[G])
(1)
HSG
and
DetH(J*(A(Hb))WH) = D(GrK [G])
(11)
H<G
Proof We shall prove only (ii), since part (i) is similar but easier.
Let ' denote a set of conjugacy class representatives of subgroups of
G. If X E R(G) and aG(X) = >j a;(H;, cb;)G then, for H E ', define
aG,H(X) = E a,(H,, 0t)G.
H;=H
We may consider aG,H(X) as a linear combination of one-dimensional

characters of H and we may form IndH(aG,H(X)) E R(G). Suppose that
z E J'(A(G)) then we may consider the function
5.1.9
(X -) Det(z)Ind4(aG,H(X))) E HomnK(R(G), U(OE)),
where, as usual, E is a large splitting field for G.
Let P be a prime of K and let Q be a prime of E over P. Let us

consider the EQ-component of 5.1.9 which is given by
(X -) Det(zp )I nd4 (aG,H(X))) E HomoK, (R(G), (9EQ )
The image of zp under the composition

5.1.10
A(G)p -i K1(Kp[G]) 1_ Ki(Kp[H])1H
-*
lies in
diagram:
K1(Kp[Hb])WH =
(A(Hb)p)WcH.
(Kp [H ab]*)WGH
To see this consider the following commutative
175
5.1.11
K1(A(G)p) 0 R(H)
A(G)p 0 R(Hab)
Ind
K1(Kp[G]) R(Hab)
IInd1
K1(Kp[H])WGH R(Hab)
(Kp [Hab]*)WGH
If we start with z
R(Hab)
Det
E*Q
W in the top left corner of 5.1.11 we see that, for
all w E R(Hab),
Det(wp 0 tp) E (9EQ,
where wp is the image of zp under 5.1.10. This means that wp lies in

the intersection of (Kp[Hab]*)WGH with a maximal (9EQ-order of EQ[Hab].
This is because the characters, lp, of Hab induce an isomorphism of

E-algebras
fi Detw : E [Hab] --0 fl EQ,

which carries a maximal order in EQ [Hab] onto fJ,, (0EQ (see 5.1.4).
Therefore Wp E (A(Hab)P)WGH
Now, by the fundamental property of Explicit Brauer Induction,

X=
IndH(aG,H(X)) E R(G)
HERB
so that
5.1.12
Det(z)(X) = IIHE, Det(z){IndH(aG,H(X))}.
As z E T(A(G)) varies, the functions, Det(z), generate D((9K [G]) so

that the functions
{Det(z){IndH(aG,H(_))};H E ',z E J*(A(G))}
176
also generate D((9K[G]). However, if z = (zp) then w = (Wp) E

J*(Hab)wcH, by the previous discussion, and
Det(z){IndH(aG,H(X))} = DetH(w)(x),
by definition. This completes the proof of Theorem 5.1.8.
Let Q8 denote the quaternion group of order eight. In

this case the restricted determinant, DetQB, maps J'(A(Qgb)) (and even
Example 5.1.13
U(Z[Qgb]), see 5.7.1) onto D(Z[Q8]).
5.2 The class-group of Z[Q8]
In this section we shall examine the class-group, '2'(Z[Q8]), of the

quaternion group of order eight
5.2.1
Q8 = {x, Y I x2 = Y2, Y4 = 1, xyx 1 = y-1 I-
We shall give an explicit isomorphism

5.2.2
n : '2'(Z[Q8]) -' Z/2
(cf. 4.7.11) whose description differs from those to be found in Martinet

(1977a) and Taylor (1984, p. 88, proposition 2.21). The map, n, is a type
of reduced norm map. It does not seem to have appeared previously in
the literature. As an application of 5.2.2 we shall derive a result (5.2.33)
which was first obtained in Chinburg (1989, p. 44) from Taylor (1984,
p. 88).
Incidentally, our invariant which gives the isomorphism of 5.2.2 is

similar in spirit to that of Martinet (1977a), except that 5.2.2 has the
advantage of being directly applicable to any finitely generated, projective
Z[Q8]-module, M, whereas to apply the criterion of Martinet (1977a);

(see 4.7.11) one must first find an isomorphism of the form
5.2.3
M(Z[Q8])'M1,
where M1 is an ideal of Z[Q8]-
Let M be a finitely generated, projective Z[Q8]-module.

Define M+ and M_ by the formulae
Definition 5.2.4
177
M+= {mEMI x2(m)=m}

and
M_= {mEMIx2(m)=-m}.
Note that M+ is naturally a module over the ring Z[Q8]/(x2 - 1)
Z[Z/2 x Z/2], while M_ is a module over Z[Q8]/(x2 + 1) = HZ, where
HZ denotes the ring of integral quaternions,
5.2.5
HZ=Z[i,j,k]/(i2=j2=k2=-l,ij=k,jk=i,ki=j).
Proposition 5.2.6 Let N be a finitely generated, projective R-module where
R = Z[Z/2 x Z/2] or HZ, then N is a free R-module.

Proof See Martinet (1977a).
Proposition 5.2.7 Let M be a finitely generated, projective Z[Q8]-module.
Then there is an isomorphism of Z[Z/2 x Z/2]-modules
'AVM
: 11'1 /M-
M+
given by lpM(m + M_) = (1 + x2)m.
Proof Clearly ''PM is injective. In addition, AVM is evidently an isomor-
phism when M is free. If M is a finitely generated, projective module

then there exists a complementary module, N, such that M N is free.
Since WMoN = 'WM EDWN is an isomorphism one sees that WM is surjective,

Alternatively, one can show that 'PM is surjective by observing that the
cohomology group
5.2.8
H2({x2}; M) = M+/((1 + x2)M)
vanishes, because M is projective over Z[Z/2] = Z[{x2}].

We are now ready to define the invariant which yields the isomorphism
of 5.2.2.
178
Definition 5.2.9 The invariant, n([M])
Let M be a finitely generated, projective Z[Qg]-module. Hence M+ (and

M_) is a finitely generated, projective module over Z[Z/2 x Z/2] (and
HZ, respectively) and hence is free.
Therefore we may take a free HZ-basis
a = {al,...,ak} E M_
and we may choose, by 5.2.7,
a={al,...,AIEM
such that
{al + M, ... , ak + M-)

is a free Z[Z/2 x Z/2]-basis for M/M_ = M+.
Now form
(1 - x2)a = {(1 - x2)al,...,(1 - x2)ak} c M_.

There is a k x k matrix, X, with entries in HZ, such that
(1 - x2)al
al
=X
5.2.10
(l - x2)ak
(=Mk.
ak
Hence X E Mk(HZ), the k x k integral quaternionic matrices. Let

5.2.11
c : Mk(HZ) -+ MMk(Z[i])
denote the complexification map (i2 = -1; Z[i], the Gaussian integers)
and set
5.2.12
n([M]) = det(c(X)) (mod
4) E Z/4.
Note that 5.2.12 makes sense since det(c(X)) is a positive integer (see
5.7.2).
Theorem 5.2.13 The construction of Section 5.2.9 yields an invariant,
n([M]) E (Z/4)" 2_`Z/2,
which depends only on [M], the class of M in'1f(Z[Q8]), and defines an

isomorphism
n : 12'(Z[Q8]) -> Z/2.
5.2 The class-group of Z[Qs]
179
The proof of Theorem 5.2.13 will be accomplished in the following

series of subsidiary results (5.2.14-5.2.18).
Lemma 5.2.14 If X E Mk(HZ) then det(c(X)) (mod 4) E Z/4 depends

only upon X (mod 2Mk(HZ)).
Proof If HZ is as in 5.2.5 we may write X = Y + W j, where Y, W E

Mk(Z[i]) so that
C(X) =
5.2.15
Y -W 1
W
'
where W denotes the complex conjugate of W.
Let Est denote the k x k matrix having a 1 in the (s, t)th entry and
zeros elsewhere. Let us examine the effect of changing X by adding 2x
in the (s, t)th entry, where x E Z[i].
Write M(Y)St for the cofactor formed by taking the determinant of
the complex matrix obtained from Y by deleting the sth row and the tth
column.
Thus
det(c(X + 2xEst)) = det ( Y + 2xEst

W
-W
Y + 2xEst /J
= det(c(X)) + 2x(-1)s+tM(c(X))st
+2x(-1)k+s+k+t M(c(X))k+s,k+t
(mod 4),
det(c(X)) + (-1)s+t [2xM(c(X))st

+2xM(A)s,t]
(mod 4)
where
-W Y
Since we are working modulo 4 the signs within the square brackets
are irrelevant and therefore we obtain
det(c(X + 2xEst))
det(c(X)) + (-1)s+t2[xM(c(X))St + xM(c(X))St]
det(c(X))
(mod 4).
180
A similar calculation applies to analyse the case when X = Y + W j is

replaced by Y + ((W + 2xEsc)j)
Proposition 5.2.16 The invariant of 5.2.9 yields a well-defined homomorphism
n : Wf(Z[Qs]) -' (Z/4)".

Proof In the notation of 5.2.9, if the HZ-basis, a is changed to
a' = Va where V E GLk(HZ), then X is changed to XY-1. However,
det(c(Y)) E Z[i]' is a positive integer, by 5.7.2. Therefore det(c(Y)) = 1
and det(c(XY-1)) =
= det(c(X)).
Given a Z[Z/2 x Z/21-basis,
det(c(X))det(c(Y))-1
!+M_ _ {a1 +M..... A +M_},

the vector, a = (c 1, ... , ak) E Mk is well-defined only up to the addition
of # =
/3k) where /3; E M_. However,
(1-x2)(a+/3) _ (1-x2)a+2fl
=Xa+2X'a
for some X' E Mk(HZ). Since, by 5.2.14, det(c(X)) __ det(c(X + 2X'))
(mod 4) we see that n([M]) is independent of the lifting of a + M_ to
a E Mk.
Now suppose that the Z[Z/2 x Z/2]-basis, a + M_ = {a1 + M_, ... , ak +
M_}, is changed to Ua + M_ for some U E GLk(Z[Z/2 x Z/2]). Hence

there exists U' E Mk(HZ) such that Ua + M_ lifts to Ua and
(1 - x2)Ua =(1 - x2)U'a + M_

= U'(1 - x2)a
= U'Xa.
However, H Z/2 = Z/2[Z/2 x Z/2] and therefore the reductions
modulo 2 of U and U' coincide. Hence U' E Mk(HZ) is a unit (modulo
2) and therefore there exists V' E Mk(HZ) such that 1 =- U'V' ( mod
2Mk(HZ)). Therefore, by 5.2.14,
det(c(U'X)) = det(c(U'))det(c(X))
det(c(X))
(mod 4).
Hence we have shown that, given M, the invariant n([M]) E Z/4 is

independent of the choices of bases a, a + M_ and of the lifting a.
181
To complete the proof of 5.2.16 it remains to show that

n([M M']) = n([M])n([M']),
since n([Z[Q8]]) = 1. However, (MM')+ = M+M+ and (M M')_ _

M_ AL. Therefore bases and liftings for (M M')+ and (M M')_
may be obtained by combining bases and liftings for M+, M_, M+ and
ML. In this case, if M and M' give rise, respectively, to matrices X and
X', in the construction of 5.2.9, then (M (D M') will yield the matrix,
X X'. Hence
n([M M']) = det(c(X (D X'))
det(c(X))det(c(X'))
n([M])n([M'] ),

Proposition 5.2.17 In 5.2.16, the homomorphism, n, is surjective.
Proof Let M =< 3, a >, the Swan module of 4.2.43. Here a = Eg,Qg g E
Z[Q8]. In this case
M_ =< 3, a > n(1 - x2)Z[Q8] = 3(1 - x2)Z[Q8]

and a basis for this HZ-module is given by 3(1 - x2).
On the other hand, a-3(x+y+xy) E< 3, a > is a basis for the quotient
< 3, a > /M_. To see this embed < 3, a > /M_ into Z [Z/2 x Z/2] via
the abelianisation map. This induces an isomorphism
<3,a> /M_ _) < 3, 2T >=< 3, T > a Z [Z/2 x Z/2],
where T= 1 +x+y+xy E Z[Z/2 x Z/2].

Since T(3-T) = 3T-4T = -T we see that < 3,T > is a free Z[Z/2xZ/2]module with basis
3 - T = 3(1 - T) + 2T,
so that < 3,a > /M_ is a free Z[Z/2 x Z/2]-module with basis, a =
a - 3(x + y + xy) E Z[Q8].
Hence (1 - x2)a = 3(x2 - 1)(x + y + xy) and since
det(c(i + j + ij)) = det

is equal to three we find that n(< 3, or >)
3 (mod 4).
5.2.18 Conclusion of the proof of Theorem 5.2.13
182
By virtue of 5.2.3, 5.2.16 and 5.2.17 it will suffice to prove that there
are precisely two isomorphism classes of projective Z[Q$]-modules, M,
of rank one. We shall classify these by means of some elementary

homological algebra for which the background may be found in Hilton
& Stammbach (1971, chapter III) and Snaith (1989b, chapter 2).
Let 1 and r denote, respectively, the trivial and the non-trivial homomorphisms in Hom({x2}, Z'). Hence we have isomorphisms of Z[Q8]modules
M_ = IndQBZ)(i) , M+ =' Ind? }(1).
5.2.19
Consider the short exact sequence of Z[Qs]-modules
M_->M->M/M_=M+.
5.2.20
The group
Ext2[Qe] (I ndQ 2) (1), IndQBZ}(i))
classifies, by 5.2.19 and chapter III of Hilton & Stammbach (1971),

equivalence classes of sequences such as 5.2.20 under the equivalence
relation generated by the relation that (M_ -) A -+ M+) is equivalent
to (M_ --) B -+ M+) if there is a commutative diagram of the form
5.2.21
M_ --. A , M+
1
M_
.B
M+
In addition there are `change of rings' isomorphisms of the form

5.2.22
Ext2
[Q-]
(I ndQXZ } (1), I ndQBZ } (t))
ExtZ[{XZ}]
(Z,
IndQez)(T))
H1({x2};IndQBZ) (i))
H1({x2}; IT)
Z/2[Z/2 x Z/2]
HZ 0 Z/2.
183
Notice that
5.2.23
AutQs(IndQ82}(i)) = HZ
and
AutQ$(Ind?2}(1)) = Z[Z/2 x Z/2]'
5.2.24
and that, in terms of 5.2.22, the natural actions of the left sides of 5.2.23
and 5.2.24 correspond to multiplication (reduced modulo 2) by HZ and
Z[Z/2 x Z/2] respectively.
We do not wish to classify M up to the equivalence relation generated
by 5.2.21 but rather up to isomorphism. Therefore we wish to evaluate
Ext'Z[Q8] (I ndQ8
(1) '14Q8
n (XZ) O )/
(XZ}
5.2.25
(AutQB(IndQX2}(i)) x AutQB(IndQ2}(1)))
Z/2[Z/2 x Z/2]/(Z/2 x Z/2),
where Z/2 x Z/2 acts by multiplication. The orbits of this action are
readily seen to be
{0},{1,x,y,xy},
5.2.26
{1+x+y,xy+x+y,x+1+xy,y+xy+1},
{1+x,y+xy},{1+y,x+xy},
{1 +xy,x+y} and {1 +x+y+xy}.
The proof will be completed if we can show that only the two orbits
of order four represent projective modules. In fact, we shall show that
only these orbits represent cohomologically trivial modules. For this we
need to recall some cohomological facts.
The `change of rings' isomorphisms in 5.2.22 arise from the pullback
diagram obtained from 5.2.20 (see Hilton & Stammbach, 1971; Snaith,
1989b) :
184
5.2.27
M_-. V , Z
1
M-, M
--
M+ = IndQZ}(1)
in which
p(m) = 1 0 m E Z[Q8] Z[{x2}] Z
and
V = 7r-1(p(Z))
r%12IR
Also, 5.2.27 is a diagram of Z[{x2}]-module homomorphisms. There

results a commutative diagram of coboundary maps and `change of rings'
isomorphisms.
5.2.29
b 1,M
Hi+1({x2};M-)
H'( {x2}; Z)
gEZ/2xZ/2Hi+1({x2}; ,r),
06
SM
H'(Q8; M+)
The homomorphism, /3a, in 5.2.29 is simply the map which adds

together the four components.
If M is cohomologically trivial then the coboundary
aM : H`(Q8 ; M+) -f H'+1(Q8; M-)
must be an isomorphism for all i > 1 and surjective for i = 0. Identifying

this map with 13aS1,M in 5.2.29 it suffices, by the cohomological periodicity
185
of {x2} (Snaith, 1989b, p. 16), to analyse when the homomorphism
flab1,M : Z = H({x2};Z) - -> H1({x2};i) = Z/2
is onto. However, if one unravels the isomorphisms of 5.2.22, one finds

that in
gEZ/2xZ/2H1({x2};T)g
ED
9EZ/2xZ/2 Z/2,
61,M(1) = E(1)g, when M is the module corresponding to the orbit of

>i gi (gi E Z/2 x Z/2).
Hence, in the list of 5.2.26, fab1,M(1) is zero except for the orbit
{1,x,y,xy} for which
fab1,M(1) = $a((1)x) = 1 E Z/2
and for the orbit {1 + x + y, xy + x + y.... } for which

Qab1,M(l) _ fla((1)x + (1)y + (1),xy) = 3 E Z/2.
This completes the proof of Theorem 5.2.13.

5.2.30 Cohomologically trivial modules of finite type
Let G be a finite group and let M be a Z[G]-module of finite type which

is cohomologically trivial. The Grothendieck group (the class-group) of
such modules is isomorphic to W2'(Z[G]) (see Tate, 1984, Lemma 8.1,
p. 65). Hence the module, M, gives rise to a class, [M] E '2(Z[G]). This
class is represented by [B] - [A] E '.(Z[G]), where
0->A-pB-*M-)0
is any short exact sequence of Z[G] -modules in which A and B are finitely
generated and projective.
In particular, any M which is a finite abelian group with a G-action

and whose order is prime to #(G) is cohomologically trivial and gives
rise to a class in W2'(Z[G]).
Suppose now that G = Q8, the quaternion group of order eight. We
shall examine the class of a finite Z[Q8]-module of odd order.
5.2.31
x+ : (Z/8)` -> {1} = (Z/4)'
by x+(-1) = 1, x+(3) = -1.
186
Lemma 5.2.32 Let M be a finite abelian group of odd order upon which
Q8 acts. Define M_, as in 5.2.4, to be the (-1)-eigengroup of x2. Then

there is an integer, r_ > 0, such that
#(M-) = (r-)2.
Proof We may decompose M into its p-Sylow subgroups, where p is an
odd prime, and then filter each p-Sylow subgroup in such a manner that
the associated graded Q8-module consists of Fp[Q8]-modules, where Fp
is the field with p elements. Hence it suffices to assume that M is an
Fp-vector space. Therefore we must show that M_ is even-dimensional.
Let Fp denote the algebraic closure of Fp. It suffices to show that

M_ OF, Fp is even-dimensional over Fp. However, by Mashke's theorem,
Fp [Qs] is semi-simple. In fact,
Fp[Qs]
M2(Fp) (Fp)4
and the only irreducible component upon which x2 acts like (-1) is
M2(Fp). Since the irreducible M2(Fp)-modules are even-dimensional it
follows that so also is M_.
An alternative proof is to be found in the argument preceding 5.2.38.
Theorem 5.2.33 (Chinburg, 1989, section 4.3.6, p. 44) Let M be a finite
abelian group of odd order upon which Q8 acts. Then, in the notation of
5.2.30 and 5.2.32,
[M] = X+(#(M+))r- E (Z/4)' = WY(Z[Qs])Here, as in 5.2.4, M+ is the (+1)-eigengroup of x2.
Proof Choose projective Z[Qs]-modules, A and B with rank(A) =

rank(B) = k, as in 5.2.30 and consider the following commutative diagram, whose rows are exact:
187
5.2.34
A-
0 A/A_
B/B_ -. M/M_
Choose HZ-bases a = {al,... , ak } and b = {b1,. .. , bk } for A_ and B_

respectively. Choose sets of elements a = {al,..., ak} and f =
Q+B_ are Z[Z/2 x Z/2]-bases
so that, in the notation of 5.2.9, a+A_ and
for A/A_ and B/B_ respectively. Hence there are matrices
X, Y, U, U', V E Mk(HZ)
and
V E Mk(Z[Z/2 x Z/2])
such that V and V' coincide modulo 2 and
(1 - x2)a
= Xa
(1 - x2)f = Yb
5.2.35
f_(a)
= Ub
and
f (a)
= VII + U'b.
The commutative diagram 5.2.34 gives rise to the following calculation:
XUb = f_(Xa)
= f_((1 - x2)a)
= (1 - x2)f(a)
= (1- x2)(Vf3 + U'b)
= V((1 - x2)p) + U'((l - x2)b)
= V'Yb+2U'b.
Therefore
188
XU = V'Y + 2U'E Mk(HZ).
5.2.36
Taking complex reduced norms (that is, applying det(c(-))) modulo 4

we obtain
det(c(X))det(c(U)) = det(c(V))det(c(Y)) E Z/4.
5.2.37
However, if R is the realisation map
R : Mk(HZ) -* Mak(Z),
then
#(M_) = det(R(U)) = det(c(U))2
since the unreduced norm is the square of the reduced norm in this case
(cf. Curtis & Reiner, 1987, section 45, p. 138; Serre, 1977, p. 92). By
5.7.2, det(c(U)) is a non-negative integer so that
det(c(U)) = r_.
5.2.38
Therefore, from 5.2.37, we see that det(c(V')) is an odd integer and in

IWY(Z[Q8]) = (Z/4)' we have
[M] = det(c(Y))det(c(X))-' = r_(det(c(V'))) E (Z/4)'.
5.2.39
From 5.2.39 det(c(V')) is an invariant which lies in the class-group and

depends only upon M+. To compute this invariant and thereby complete
the proof we shall use induction on #(M+).
We shall examine the case in which rank(A) = rank(B) = 1. In this

case we may suppose that
V = p + qx + ry + sxy E Z[Z/2 x Z/2]
and that
V' = p+qi+rj+sk E HZ.

We have
p
5.2.40
#(M+)
det
q
s
=(P+q+r+s)(P+q-r-s)(P+r-q-s)(P+s-q-r)
= [(p + q)2 - (r + s)2] [(P - q)2 - (r - s)2].
5.3 Relations between Swan modules
189
Hence p + q + r + s is odd and either one or three of p, q, r, s are even.

One easily verifies in all the cases that
2
x+(#(M+)) = p2 + q2 +r + s2 = det(c(V')) E (Z/4)'.
For example, if p is odd and q, r, s even, then p2 + q2 + r2 + s2
(mod 4) and
#(M+) (1 - (r + s)2)(1 - (r - s)2) (mod 8)
1 - 2r2 -2s 2
1
(mod 8)
(mod 8)
as claimed.
For the induction step let M be a finite Z[Q8]-module of odd order

with M+ # {0}. Choose 0 z E M+ and define n : Z[Qs] -) M by
n(1) = z. Since #(im(n)) is odd, ker(m) will be cohomologically trivial
and torsion-free and therefore projective. In the class-group
[M] = [im(7r)] + [M/im(7[)]
and, by induction on the order of the (+1)-eigengroup, we may assume

that the theorem has been established for im(rz) and for M/im(x) and
therefore holds for M.
Remark 5.2.41 Theorem 5.2.33 determines the class in '2'(Z[Qg]) of a
finite module of odd order. To give a formula which determines the class
corresponding to a cohomologically trivial, finite module whose order

is a power of two seems to be much more difficult. However there are
many modules of this type. Here is an example, which I learnt from
P. Symonds, of such a module representing the non-trivial class in the
class-group.
Define a finite Z[Qs]-module by means of the following exact sequence:
0-><3,a>-Z[Qs] -) M-) 0
where h(x) = x(2 - a)3-1 for all x E< 3, a >.
Throughout this section we shall assume that G is a finite group in an

extension (not necessarily split) of the form
190
5.3.1
>{1}
{1}
p is prime.
We shall use a geometrical technique, which to my knowledge is new

but which is very much in the geometrical spirit which motivated R.G.
Swan's original introduction of the Swan modules, S(k) (see 4.2.43; Swan,
1960, p. 278).
Theorem 5.3.3 Let G be as in 5.3.1 and 5.3.2 and let HCF(k, #(G)) = 1.
Then
S(kP) - DndNc<g>(S(k)) = 0 E D(Z[G]),

g
where g E G - J runs over generators for the distinct conjugacy classes

of subgroups, < g >, of order p and NG < g > denotes the normaliser of
<g>inG.
This result will be proved in 5.3.29-5.3.42. Before proceeding to the
proof let us pause to examine some applications.
Example 5.3.4 Let p be an odd prime and let Z/p act on Z/pn (n > 2)
by 2(z) = z(1+P)Pn 2, where z E Z/pn and i generates Z/p. From Curtis &
Reiner (1987, p. 365; 4.4.20):
T(Z/p oC Z/pn) = Z/p".

If we set G = Z/p oc Z/pn and J = Z/pn we may apply Theorem 5.3.3
to any integer, k, which is not divisible by p. Hence
0 = S(kP) - > I ndNG<g>(S(k)),
g
where the sum runs over some of the g E G - J of order p. However,

for each of these No < g >= Z/p oc Z/pn-i so that if we set k = tPo-1
then each IndNQ<g>(S(k)) vanishes, by induction on the order of G, and
we find that
S(t'') = 0 E T(Z/p oC Z/pn),

as expected.
Example 5.3.5
order 8n,
191
Let G = Q(8n), the generalised quaternion group of
Q(8n) = {X, Y I Y2 = X2", Y4 = 1, YXY-1 = X-1}.
Setting J =< X >, p = 2 and observing that there are no elements of

order two in G - J we find that, for all n,
2T(Q(8n)) = 0.
5.3.6
In Bentzen & Madsen (1983) one finds some very difficult and technical
calculations of Swan subgroups which are related to 5.3.6. En route to

the main theorem of Bentzen & Madsen (1983), one finds a calculation of
T(Q(8p)), when p is a prime (see Bentzen & Madsen, 1983, theorem 3.5).
The preceding application of Theorem 5.3.3 gives a new manner in which
to derive an upper bound for T(Q(8n)), which is less computational than
the method of Bentzen & Madsen (1983).
Notice that the Artin exponent of Q(8p) is equal to four (Bentzen &
Madsen, 1983, p. 462), so that 5.3.6 improves upon the estimate given by
Theorem 4.4.17.
Let us recall part of Theorem 3.5 of Bentzen & Madsen (1983).
Theorem 5.3.7 Let p be an odd prime such that either p - 3 (mod 8) or

p - 1 (mod 8) and ordp(2) is even, where ordp(t) is the order of t in Fp.
Then
T(Q(8p)) - Z/2 Z/2.

Corollary 5.3.8 Let Q(8n) denote the generalised quaternion group of order
8n. Suppose that n = 2api' ... p', with m, > 1, a > 0 and {pi;1 < i < r} a
set of distinct primes such that either pi - 3 (mod 8) or pi - 1 (mod 8)

and ordp,(2) is even, where ordp(t) is the order oft in F;. Then
T(Q(8n)) = (Z/2)r+1
Proof By 5.3.6, T(Q(8n)) is a quotient of
fjr 1(Z/P"'')*/{1}) Z/2

Z/2 X fr 1
((Z/2a+3)*
Fp,l(Fp,)2
(Z/2)r+1.
However, inspection of Theorem 3.5 of Bentzen & Madsen (1983)

shows that the natural maps to the Swan groups of the subquotients,
192
{Q(8p;);1 Z/r denote the reduction (modulo r) of
the augmentation map. If HCF(r, #(G)) = 1 then
[Ker(e : Z[G] -) Z/r)] = -S(r) e WY(Z[G]).

Proof By the construction of S(r) given in Curtis & Reiner (1987, section
53, p. 343; 4.2.44) S(r) = {r, a} i Z[G], where a = EgEG g. Therefore we
have exact sequences of the form
0
*Ker(e)-Z[G] -*Z/r )0
and
0--*(rZ[G])-+
where ir(r) = 0 and ir(a) = 1. Since r(Z[G]) = Z[G], Schanuel's lemma
(Swan, 1960, p. 270) implies that
Ker(e) S(r) = Z[G] Z[G],

from which the result follows.
Remark 5.3.10 When G is a p-group the calculation of T(G) is given
in 4.4.20 (see also Curtis & Reiner, 1987, section 54). Most of the cases
are due to M.J. Taylor, using determinantal congruences and a particular
case of the maps, ST, of 4.4.11. This manner of detection, using ip'/l
to give a lower bound for T(G), is reminiscent of J.F. Adams' approach
to the detection of the image of the J-homomorphism in the stable
homotopy groups of spheres (Adams, 1963, 1965a,b, 1966). The more
difficult part of the determination of the image of the J-homomorphism
was the construction of an upper bound. This led to the celebrated
Adams conjecture (Adams, 1963, 1965a,b, 1966) and its ultimate proof
by D. Sullivan (1974) and D.G. Quillen (1971). The Adams conjecture
concerns the maps between spheres of vector bundles. This motivated
the method which we shall now employ to prove Theorem 5.3.3. We shall
study maps between unit spheres of representation spaces. This approach
will involve a considerable digression into the topology of group actions,
for which I must ask the reader's indulgence, before we return to the
193
Swan subgroup. As to the prerequisites from algebraic topology, the

proof will require only the elementary properties of homology theory
such as may be found in Greenberg & Harper (1971).
The proof of Theorem 5.3.3 will establish a slightly more general result
(see 5.7.5).
5.3.11
Let X1,..., Xp be topological spaces. The p-fold join, X1 * ... * XP,
is defined to be the space which is formed in the following manner. If

I = [0, 1] is the unit interval consider the set, Z, of 2p-tuples
(t1,x1,t2ix2,...,tp,xp) E I X X1 X I X X2 X ... X I X Xp,
which satisfy Ep 1 tZ = 1. Impose on Z the relation that
(tl,x1,...,0,xi,...)
whenever tj = 0. The quotient space, Z /
is X1 *
* X p.
When X1 = X2 = ... = Xp = Sn, the n-dimensional sphere, then

* XP is homeomorphic to the sphere, SP(n+1)-1. For example,
such a homeomorphism is given by sending (t1, x1, t2, x2i ... , tP, XP) to
X1 *
(t1x1, t2x2,..., tpxp) E P 1Rn+1
Now let J > S' c C" be a one-dimensional representation, which

we will later assume to be a faithful embedding of the cyclic group, J, into
the unit circle. Let Sl = S(4 ) denote the three-dimensional sphere,
S3, upon which J acts via j(wl, w2) _ (W(j)wi, 4'(j)w2), where w1, w2 E C
and I w1 12 + w2 12= 1. If (WI, W2) = x we shall write j(x) = 4'(j)(x) for
this action on Si S.
Let z1, z2, .... zP be a set of coset representatives for G/J. Therefore
there is a homomorphism into the symmetric group

5.3.12
EP
such that, for g E G,

5.3.13
gzi = za(g)(i)j(i, g)
(j(i, g) E J).
Let G act on Sl * ... * Sl (p copies) by the formula
g(t1,xl,...,tP,xP)
5.3.14
_ a(g)(ti, 3(j(1, g))(xl), ... , ti,, O(j(p, g))(xp)),
194
where a(g) permutes the p-tuple of pairs {(t;, 3(j(i, g))(xi))}. Clearly
Sl * ... * Sl with this action is just the unit sphere in the representation
space afforded by I ndJ (, 0).
Let k be an integer which is coprime to the order of G. Let

k
: Si --> S3( 4,k) = S2, say, denote the map which is the iden-
tity on the first circle coordinate and is the kth power map on the second
circle coordinate. Define a G-map
5.3.15
Fk :S1 *...*Sl -SS *...*SZ
by
Fk=k*k*...*k
(p copies). In view of the previous discussion we have
5.3.16
Fk : S(I nd (, 4,)) -- S(I nd (c 0k)),
where S(V) denotes the unit sphere in the vector space, V.

Remark 5.3.17 Recall that a continuous map, f : S" ---> S" has a degree
which is equal to the endomorphism induced by f on the homology group,
H (S"; Z) = Z. The kth power map on S' has degree k and the degree
of the join of a number of such maps is equal to the product of their
degrees.
Lemma 5.3.18 For H < G let XH denote the H -fixed points of X. For all
H < G, Fk restricts to a map of spheres
Fk : S(1 nd (4 4)))H - S(I ndjG ( ok))H
whose degree is a power of k. In particular the degree of Fk in 5.3.16 is

k1G.Jl. Note that S(V)H = S(VH) for any G-representation, V.
Proof Since Fk is a G-map it induces maps on H-fixed point sets. Also

the degree of Fk is equal to k[G.JI, by 5.3.17.
Now let us examine the degree on the H-fixed point sets. By the double
coset formula 1.2.40
I ndHrgJg ((g-')*(
ResHI nd,Gi (4 0) _
gEH\G/J
where (g-1)*(4))(u) = 4)(g-lug)
195
Also
((g-1)`( 4)))H) =
dim(IndH
HngJg
if Resg_,HgIV((p)
otherwise.
Therefore, since k is coprime to #(G),
dim((I nd (4 0))H) = dim((I nd ( ok))H)
In addition, there is an injective map

Ag : IndHngJg-'((8
5.3.19
4)))
Ind'(4 (D 0)
given by
2g(h HngJg ' v) = hg J v.
5.3.20
The sum of all the )g of 5.3.19 for which the H-fixed points are non-trivial
yields an isomorphism
5.3.21
g1 ndHng Jg, ((g-1)`(o 0))H -=> I ndG ( O)H.
The same is true when the second copy of 0 is replaced by 4k and

then Fk between the H-fixed spheres on the right of 5.3.21 corresponds
to the join of the induced maps between H-fixed spheres on the left of
5.3.21. However, a non-zero vector in I ndHng jg_, ((g-1)`(4 0))H is of
the form (v = (z1, z2) E S; )
h v
hEH/(HngHg' )
and the map induced by Fk simply raises the second coordinate of v to

the kth power. Hence the degree of Fk on the H-fixed points is a power
of k, as required.
Define D(G, J, 0, k) to be the mapping cone of Fk in 5.3.16. Hence
D(G, J, q5, k) = [S (I nd (4 k )) U (S (I ndi (4 4))) x I)]/
where - is given by
(z,1) - (z',1)
(z, z' E S(I ndf (4 4))))
and
(z,0)
Fk(z)
D(G, J, 0, k) is a finite CW complex with a cellular action induced in

the obvious manner from the action on the representation spheres.
196
Corollary 5.3.22 The reduced homology of the fixed point set

H.(D(G, J, 0, k)H ; Z(P))
is trivial for all H 5 G and for all localisations, Z(P), of the integers, Z, at
a prime dividing #(G).
Proof In Z(P) the integer, k, is invertible. However
Hm(S(I nd j (4 0,0))H; Z(P))
= Hm(S(I nd j (4, E) 0'))H; Z(P))
is zero except when m = 4pdim(I ndG(4,)H) - 1, in which case it is Z(P).

From the exact sequence
Hm(S(I nd G (O 4,))H ; Z(P)) (F)') FIm(S(I ndG(4 4,k))H ; Z(P))
Hm(D(G, J, 0, k)H; Z(P)) -' .. .
we see, by 5.3.18, that (Fk). is invertible (with Z(p) coefficients) and

therefore H.(D(G, J, 0, k)H; Z(P)) = 0, as required.
5.3.23 Chain complexes and class-group elements
We now describe a method, which appears in Oliver (1978), whereby to

associate an element of the class-group, '.(Z [G] ), to a chain complex
such as C. (D(G, J, , k); Z), the cellular chains of D(G, J, 0, k).
If S is a finite G-set let Z(S) denote the free Z-module with S as a

basis and with the induced action of G. Let v(G) denote the class of all
chain complexes of Z[G]-modules of the form
5.3.24
0-*Z(Sn) ">Z(Sn-1)
S-')
Z(S0) -i0
such that for all H < G, and for all i, 8, maps Z(SH) into Z(SH1). An
example of such a chain complex is provided by the cellular chains of
a finite G-CW-complex, X, with a cellular action of G. In this example
the subcomplex (Z(SH), 6.) is provided by the chains of XH. All chain
complexes 5.3.24 of v(G) can be realised geometrically in this manner by
some simply connected G-space. Let v"(G) denote the complexes of v(G)
for which (Z(SH), 5.) Z(P) is exact for all primes, p, dividing #(G) and
for all p-groups, H < G. A resolution for 5.3.24 is a complex in v'(G) of
the form
5.3.25
0 --i Z(Sn)
Z(Sn_1) '") ...
61
) Z(Sp) -* 0,
197
such that Si > Si for each i and S; extends Si and which satisfies the
following conditions:
(i) G acts freely on Si - Si for all i,
(ii) 5.3.25 is exact except at Z(Sn) and
is a projective module.
In Proposition 4 of Oliver (1978) it is shown that every complex 5.3.24
in v'(G) admits at least one resolution, as in 5.3.25, and that the class
[ker(8n)] E '2'(Z[G])
5.3.26
is a well-defined element, depending only on the chain homotopy class of

5.3.24 in v'(G). In particular, if X is a connected, finite G-CW-complex,
all of whose H-fixed point sets have vanishing reduced Z( )-homology, we
may, by taking 5.3.24 to be the reduced chain complex of X, associate to
it in this manner a well-defined Euler characteristic which will be denoted
by
M(X) E WY(Z[G]).
5.3.27
Geometrically M(X) is realised by embedding X into a finite G-CW-
complex, W, of the same dimension as X such that G acts freely on

W _X.
We may apply this construction to D(G, J, , k).
Lemma 5.3.28 In the notation of 5.3.22 and 5.3.23
M(D(G,J,
k)) = [Ker(Z[G] -- Z/(k[G`J]))] E 16Y(Z[G])
where e(>g ).gg) _ > g 1g (mod k[G:J])

By 5.3.9, [Ker(e)] = -S(k[G:Jl) E WY(Z[G]).
Proof Consider the reduced chain complex of D(G,J,4,k):

6,,
6n-
61
where q is the augmentation which sends every zero-cell to 1 E Z. Since

Hi(D(G, J, 0, k); Z)
if i
Z/(k[G:Jl)
n - 1,
if i=n-1,
we may build up a resolution in the following manner. We leave C1

unaltered for j < n - 1. In Ker(Sn_1) we may find an element, xn-1,
whose image generates
Hn-1(D(G,J,0,k);Z) =
Z/(k[GJ])
198
By examining the map Fk explicitly, one can see that xn_1 may be taken
to be G-fixed. Now we alter Cn to Cn = C, Z[G] < xn > and set

6,(xn) = xn_1. From this description one readily sees that
Ker(8n : Cn -> Cn_1)
[Ker(Z[G] -)
as required.
Z/(k[G.J]))7
JJ
Now we shall give the proof of Theorem 5.3.3. While conceptually

simple, the proof is technically quite involved and for this reason we
shall break the proof into the non-split and the split cases in order that
the reader may see some of the details in the (simpler) non-split case
first.
5.3.29 Proof of Theorem 5.3.3 in the non-split case
In this case we shall assume that 5.3.1 is not a split extension, which
means that there are no elements g c G - J of order p. Under these
circumstances we must show that S(k[G:J]) = pS(k) = 0 in'2'(Z[G]). To
do this we will take 4 : J -p C' to be injective in 5.3.11 and we shall
show that
M(D(G, J, 0, k)) = 0 E Wf(Z[G]).
The result will then follow from 5.3.28. By the discussion of 5.3.23, this
relation will follow if we can find a G-embedding of D(G, J, 0, k) into a
connected space W which is acyclic (i.e. R. (W ; Z) = 0) and where G acts
freely on W - D(G, J, 0, k).
Write X = Si , Y = SZ and let Z = D(J, J, 0, k) be the mapping cone
of the map k : X -> Y which was introduced earlier. The projective

module associated to Z is equal to
M(Z) = -S(k) E T(J) c WY(Z[J]),

which is trivial if J is cyclic, by 4.2.48. Hence we can find a J-embedding
into a simply connected space, D,
i:Z--)D,
where H. (D; Z) = 0 and J acts freely on D - i(Z). This geometrical
realisation is guaranteed by a theorem of C.T.C. Wall (1966) since the
space, Z, and the chain complex which D must realise are both simply
connected.
Let * E i(Z) be a J-fixed point (which must be the cone-point since J

acts freely on X and Y). Form the following quotient of the p-fold join
of D:

5.3.30
199
E = D * D * ... * D/{(t1, *, t2i *,..., tp, *) = *},
on which G acts by the formula of 5.3.14. We may define a G-embedding
It :D(G,J,0,k)-*.E
5.3.31
in the following manner. If C(X * ... * X) is the cone on X * ... * X then
D(G,J,0,k)=[C(X*...*X)u(Y*...*Y)ll
Define It of 5.3.31 by
lt(t1,Y1,...,tP,YP) _ (ti,i(Y1),...,tp,i(YP))
and
7((t1, x1, ... , tp, x,), s) _ (t1, i(s, xl ), ... , tp, 1(s, xP)),
where yj E Y, x1 E X and 0 < ti, s < 1. Notice that H. (E; Z) = 0 because

D * ... * D is acyclic and E is obtained from this connected, acyclic space
by collapsing a (contractible) simplex to a point. Such a collapsing map
induces an isomorphism on homology groups.
It remains to verify that if 1 * g E G and w = (t1) wl, .... tp, wp) 0 im(it)
then g(w) w.
We shall divide the possibilities into the following cases:
caseA :
caseB :
caseC :
caseD :
g E J and w E (i(Z) * ... * i(Z))/ -,

g 0 J and w E (i(Z) * ... * i(Z))/ -,
g J and w = (tl, wl, ...) with Wi E D - i(Z), and
g E J and w = (t1, wl, ...) with w1 E D - i(Z).
In case A, because g E J a G, we must examine

= (tl, j(l, g)(wl ), ... , tp, j(p, g)(wp))
g(w)
=w
= (ti, wl,..., tp,Wp).
This implies that j(i, g)(wi) = wi, for each 1 < i < p such that t, * 0.
However, because 0 is injective, the action of J on Z is free except for
the cone-point, * , which is J-fixed. Therefore w = (t1, *, ... , tp, *) E im(n).
In case B, since g J and p is prime, a(g) must be a p-cycle. Let us
suppose that o(g) = (1, 2,..., p). Hence we must examine
g(w) _ (tP,j(p,g)(wp),...,tP-1,j(p-I, OWPA)

W
(ti, wi,...,tp,Wp).
200
This equation implies either that each wi E i(S2) and w E im(7C) or that
each wi lies in the image of the cone on S. In the latter case each w,
must be at the same level in the cone and therefore, again, w E im(n).
In case C let us once more suppose that o(g) = (1, 2, ... , p). We may
also assume that tl * 0 and that w1 E D - i(Z). In this case
g(t1,w1,...,tp,wp) = o(g)(tl,j(1,g)(wl),...,tp,j(p,g)(wp))
=(tl,wl,...,tp,Wp),
then
ti = t2 = ... = tp
and
w1
= j(p, g)(Wp),
W2 = j(1,g)(W1),
wp = j(p - 1, g)(wp-1)
Substituting each of these equations into the previous one we obtain, in
the notation of 5.3.13,
w1
=j(p,g)j(p- 1,g)... j(1,g)(wl)

-1
zi(WI)
=
_1
za(g)(p)gzpza(g)(p-1)gzp-1...
_ (zi lgpzl)(w1)
Since wl E D - i(Z) and gP E J, this implies that 1 = zl lgpzl and

therefore that 1 = gP, which is the contradiction that completes the proof
in case C.
Finally, in case D,
g(w) = (ti, g(w1), ...) = (ti, w1, ...) = w
so that g = 1 because wl E D - i(Z) and J acts freely on D - i(Z). This

completes the proof of Theorem 5.3.3 in the non-split case.
5.3.32 Proof of Theorem 5.3.3 in the split case
We retain the notation of 5.3.29. Also, we shall assume that G = Z/p oc J
where Z/p =< t >. Let {z = ii-1 1 < i < p} serve as a set of coset
I
representatives of G/J. Suppose that

w=(t1,w1,...,tp,Wp)
201
(0 < t; < 1, E; t2 = 1, wi e D) denotes a point of E - im(n) which is fixed
by g E G. By the discussion of 5.3.29, if g = hr for some h E J then
tl=t2=...=tp=P P-1/2 and

wl = j(p, g)(wp),
W2 = j(1,g)(wl),
wp = j(p - 1, g)(wp-i ),
where j(i, g) = ii+1gzj and g = hT must be of order p. Hence, if w is fixed
byg=hr(hEJ)then
5.3.33
w=
(P-,/2, wl,p1/2,(T21gT1)(w1)'p 1/2,(T31g2T1)(wl),...),
where w1 E D - i(Z). If g = hTs with 2 < s < p - 1 then w has a form

similar to 5.3.33 which is determined by its first entry, w1 E D - i(Z).
For g = hT (h e J) we obtain an embedding
5.3.34
pg :DDg =g(D)cE
given by g(wi) = w, where w is given by the formula of 5.3.33. In
general, for g = hTs (h E J, 2 < s < p - 1), we obtain a mapping

of the form 5.3.34 which embeds D into E. Specifically, phts(wl) is
given by hts(wl) = (p 1/2, 11, ... , p1/2, (Xp) where acs+1 = T-`(hTs)`(wl )
= i `sg`(wl ). Here g = hr. must be of order p.

Lemma 5.3.35 Let * E Z denote the (J fixed) cone-point and let xo
(p 1/z, *
p 1/2 *) E E. Suppose that g',g E G - J and that gP = 1
(g')P. Then
(i)xoEDg=pg(D),
(ii) either Dg fl Dg, = xo or Dg = Dgand
(iii) Dg = Dg' if and only if < g >=< g' >.

In case (iii), ug(wi) = g'(w1) for all w1 E D.
Proof Suppose that g = hTs with h E J then the (ts + 1)th coordinate is
equal to T`S(hTS)`(*) = *, since * is J-fixed. This proves part (i).
For part (ii), take g = hT and g = h'TS (1 < s < p - 1). If w1, wi E
D - {*} suppose that g(w1) = pg,(wi). By comparison of the first
coordinates we see that wl = wi. Comparison of the second coordinates
yields (T-1hT)(w1) = T-SS'(h'TS)S'(wl), where ss' - 1 (mod p).
202
Since J acts freely on wl, hT = (h'Ts)s' or < g >=< g' >. However, in
this case, the (ts + 1)th entry of pg(wl) equals
ti ts(hT)ts =
ti-ts(hfts)tss' =
ti tsthrTs)t,
which is the (ts + 1)th entry of ug,(wl). Hence Dg = Dg' and g = g,.
This proves parts (ii) and (iii).
Now define
W = E - it(D(G, J, , k)) U (UgDg),
5.3.36
where, in 5.3.36, g runs through representatives of subgroups, < g >, of

order p in G - J.
Lemma 5.3.37 (i) Let z E J, then the following diagram commutes:
g
Dg
zgz t
Dzgz i
(ii) The action of G is free on W in 5.3.36.

(iii) Let W be as in 5.3.36, then for all H < G and for all localisations,
Z(q), of Z at a prime, q, dividing #(G)
ft. ((E - W )H; Z(q)) = 0.
(iv) If z E NG < g >, the normaliser of < g >, then z(Dg) = Dg.
Proof For part (i), the (ts+l)th coordinate of ht=(wl) (h E J, 1 < s < p-i
and his of order p) is equal to Tts(hTs)t(wl). Hence the (ts+1)th coordinate
of z(hts(wl)) is equal to
T-tszTtsti ts(hTS)t(wl) = ti tsz(hTs)t(wl)
However, the (ts + 1)th coordinate of pzhtsz-l(z(wl)) is equal to

T-tsz(hTs)tz lz(wl)
= titsz(hTs)t(wl),

Part (ii) follows from the discussion of 5.3.29 as recapitulated in 5.3.32.
203
If Zg = pg(Z) then I.(-; Z(q)) vanishes on
Zg , Dg and 7r(D(G, J, 0, k))H.

By 5.3.35, UgDg is a wedge of the Dgs which intersects 7c(D(G, J, 0, k)) in
the wedge of the Zgs. Hence part (iii) follows from the Mayer-Vietoris
homology exact sequence.
Finally, if z E NG < g > nJ then part (iv) follows from part (i). If
z = h'rs E NG < g > we may, by 5.3.37, assume that g = hrs. Then g
acts trivially on Dg so that
z(Dhts) = h'h-i(Dg) =
Dh'h-'gh(h')-l = Dzgz-, = Dg,
which completes the proof of part (iv).

Corollary 5.3.38 In the notation of 5.3.27 and 5.3.36
0 = M(E - W) E W2'(Z[G]).
Proof E - W is embedded in the acyclic space, E, with a free G-action

on the complement, W, by 5.3.37 (ii).
Lemma 5.3.39 In 5.3.38

Hi(E - W ; Z) _
Z/kP
gZ/k
0
if i = 4p - 1,
if i = 4,
otherwise
where g runs through an indexing set for the distinct Dgs.
Proof This follows from the Mayer-Vietoris sequence for

E - W = it(D(G,JO,k)) U (UgDg)
since
7r(D(G, JO, k)) n (UgDg) = UgZg,
H.(Zg;Z) is equal to Z/k in dimension three, FI.(7z(D(G,J(P,k));Z) is

equal to Z/kP in dimension 4p - 1 and each Dg is acyclic. Note that, by
5.3.35, UgDg and UgZg are wedge sums with * as the base-point.
5.3.40 Conclusion of the proof of Theorem 5.3.3
In order to obtain the relation of 5.3.3 in the class-group we will evaluate

M(E-W) of 5.3.38. To do this we may assume that E-W is triangulated
204
so that the G-action is simplicial. We must build a resolution (in the

sense of 5.3.23) of the augmented chain complex
...)
CO(E - W)) Z>0,
following the recipe of 5.3.23. The process begins in C4(E - W) where

we must kill the generators of the Z/ks which are indexed by the Dg,s,
as in 5.3.39.
Fix g E G - J such that gP = 1 and consider Dg. By the homology

calculation of 5.3.39 we may find a 4-chain, a4, which is a cycle representing the Z/k in H4(E - W) which corresponds to Dg and whose
cells all lie within Dg. We know from 5.3.35-5.3.37 that NG < g > acts
trivially on the homology class of ag while G/NG < g > acts freely on it,
permuting it among the generators of the other Z/k's which correspond
to the Dzgz t's.
In dimension five we insert free modules, Z[G] < 8g >, where bs(5) _
a4, one for each conjugacy class of g E G - J such that gP = 1.
Suppose that b5(Ez 2zZ(8g)) is in 65(C5(E - W)). Passing to homology
we find, for each z E G - NG < g >, that
5.3.41
EvENG<g> All = 0
(mod k),
since, in homology,
zEG/NG<g> vENG<g> Az"]
Conversely, if 5.3.41 holds for each z E G - NG < g > there is a unique

a E C5(E - W)/Ker(65) such that d5(E2zZ(Sg) - a) = 0. Hence, since
ker(S5) = im(b6), we obtain an exact sequence
5.3.42
0 --> im(S6) -p ker(bs) --> gPg
)0,
where
Pg = IndN,,<g>(ker(Z[NG < g >]
Z/k)) = -In4 <g>(S(k))
and the direct sum is taken over representative generators, g, for the
conjugacy classes of subgroups of order p which are not contained in J.
The sequence 5.3.42 is exact because the submodule consisting of

elements satisfying 5.3.41 for all z E G - NG < g > is just Z[G] Z[NG<g>
(Ker(e)) = -I ndNG<g>(S(k)), by 5.3.9. The sequence of 5.3.42 splits, since
5.4 The class-group of a maximal order
205
Pg is projective, so that
ker(85) = im(C56) (gPg).
It is now simple to complete the resolution. To kill each of the Pgs

in dimension five we insert a free module, Fg, in dimension six with
56(Fg) = Pg. This produces a contribution of g(-Pg) in ker(b6), where
Fg = Pg (-Pg). Continuing in this manner we obtain a contribution of
g(-Pg) in the kernel in dimension 4p, the top dimension. Also, as we get
to dimension 4p we pick up -S(kP) from the Z/kP = H4p_1(E - W ; Z),
as in 5.3.28, and the top-dimensional kernel of the resolution is equal to
M(E - W) = -S(kP) ((Dg(-Pg))

_ -S(kP) ((DgIn4 <g>)(S(k)),
as required.

In this section we shall apply the Explicit Brauer Induction map, aG, in a
manner similar to that used in the construction of restricted determinants
(5.1.4), to study the class-group of a maximal order in the rational groupring, Q [G].
Let A(G) denote a maximal order of Q[G], as in 4.2.29. Therefore we

have inclusions
5.4.1
Z[G] c A(G) c Q[G].
The class-group, '2(A(G)), admits a Hom-description in terms of the

representation ring, R(G), and the group of fractional ideals of a large
Galois extension, E/Q, which is a splitting field for G, as in 4.2.8. A
fractional ideal of E (Lang, 1970, p. 18) is an (9E-submodule of E of the
form c -I where 0 *- c E (9E and I d (9E. The set of non-zero fractional
ideals forms a group under multiplication and this group is isomorphic
to
5.4.2
5(E) = J'(E)/U((9E),
where J"(E) and U(&E) are as in 4.2.5 and 4.2.6. A non-zero fractional
ideal corresponds to the class of the idele given by the generators of its
completions.
The absolute Galois group of the rational field, OQ, acts on E and
206
hence on 5(E) so that we may consider the group of f2Q-equivariant

maps
HomuQ (R(G), 5(E)).
5.4.3
Notice that if E c F is an inclusion, 5(E) is a subgroup of 5(F) in

the obvious manner.
Definition 5.4.4
Let Fac(G) denote the subgroup of 5.4.3 given by
Fac(G) =
{g E HomnQ(R(G),5(E)) I g(x) E 5(Q(x)) for all x E R(G)},
where Q(x) is the smallest Galois extension of Q, the rational numbers,

which contains all the character values, {Trace(x(z)),z E G} of X.
The Fac-notation is used in Holland (1992) and is derived from its

connection with A. Frohlich's notion of factorisable functions. We will
not need this notion here, but for further details the interested reader
may consult Frohlich (1988) and Holland (1992).
Definition 5.4.5 A fractional ideal, c1I, of E is principal if I = a(9E
for some a E t9E. Define a subgroup,
5.4.6
1 E
PF+(G) c Fac(G)
to consist of all g E Fac(G) such that

(i) g(x) is a principal fractional ideal for all x E R(G)
and
(ii) for all irreducible, symplectic representations, x, the fractional ideal,
g(x), has a generator which is positive under all Archimedean places of
E.
Notice that, since x has a real-valued character, g(x)p will be generated
by a real number for all infinite primes, 9, so that it makes sense to

ask that the generator be totally positive, in the above sense. Note also
that, on symplectic representations whose complexification is of the form
p + p (p is the complex conjugate of p), g is automatically totally positive
(Frohlich, 1983, pp. 22-23). Hence (ii) is equivalent to g being totally
positive on all symplectic representations of G.
207
Let A(G) be a maximal order in Q[G]. A A(G)-module,

M, is said to be locally freely presented if there is an exact sequence of
A(G)-modules
Definition 5.4.7
0-->P-->N-*M-)0
in which P and N are locally free A(G)-modules of the same (finite) rank.
Hence M is finite.
Let KOT(A(G)) denote the Grothendieck group of locally freely presented A(G)-modules, taken with respect to exact sequences. Hence if
0
)MI
)M
)M2
)0
is a short exact sequence pf such modules, then

[M] = [Ml] + [M2] E K0T(A(G)).
Theorem 5.4.8 (Frdhlich, 1984; Holland, 1992; Taylor, 1984) (i) There is
a natural isomorphism
G : KO T(A(G)) = Fac(G).
(ii) There is a commutative diagram in which the horizontal sequence is

exact and in which cG is the canonical Cartan map:
PF+(G) - Fac(G) - W2'(A(G)) 1 0
K0T(A(G))
Denote by (H) the G-conjugacy class of a subgroup, H < G. We

have a homomorphism
5.4.9
5.4.10
BG :
R(H)WGH
H.) R(Had)WGH
(H)
defined by
BG(4 : Hab
E*) = IndH(InfHab(0))
We also have a homomorphism
R(G)
208
AG : R(G) --->
5.4.11
(H)
R(Hab)WGH,
H<G
defined by setting the R(Hab)wH-component of AG(X) equal to the sum

of all the terms of the form
(H,)(Hab _-f
which appear in the formula (see 2.3.15)

aG(X) _ F(J) a(J,w)a(J, T)G E R+(G).
5.4.12
Note that, once one has chosen a representative for (J), each (J,ip)G
determines a homomorphism J -i Jab - E' which is well-defined up
to conjugation by elements of WGJ.
By Theorem 2.3.9
BGAG = 1 : R(G) - + R(G).
5.4.13
Since aG is QQ-equivariant (see 4.5.18) we immediately obtain the following result:

Proposition 5.4.14 The homomorphisms of 5.4.10 and 5.4.11 induce maps
HomoQ (R(G), 5(E))
BG)
E Homo, (R(Hab), 5(E))WGH

(H)
H<G
and
E Homo, (R(Hab), 5(E))WaH
HomnQ (R(G), 5(E))
(H)
H<G
which satisfy
1.
As in 5.4.9, let H < G be finite groups and let Hab

denote the abelianisation of H. Define FacG(Hab) to be the following
Definition 5.4.15
subgroup of HomnQ(R(Hab), 5(E)):

FacG(Hab) = {f I f(0 : Hab -) E') E 5(Q(IndHG(InfH,((P))))}.
Note that FacG(Hab) is a group of SZQ-invariant functions which

depends on G, the ambient group, and on H < G, not just on Hab.

Theorem 5.4.16
209
The homomorphisms of 5.4.14 restrict to give homomor-
phisms
Fac(G)
E(FacG(Hab))WaH
(H)
H<G
and
J:(FacG(Hab))WGH
Fac(G)
(H)
H<G
which satisfy
1.
Proof Fix a conjugacy class representative, H < G, and consider the

(H)-component of
(f E Fac(G)). By definition,
Bc(f)(4 : Hab -' E`) =f(IndH(InfHab(4))),

which belongs to J(Q(IndH(InfHab(0)))). Hence the (H)-component of
BG(f) lies in FacG(Hab), which completes the verification for B.
It remains to show that 4({g(H)}) lies in Fac(G) if
g(H) E FaCG(Hab)WGH
Choose a set of representatives, E, for the conjugacy classes of subgroups of G. Suppose that x E R(G) and that
5.4.17
aG(x) = EHEE aG,H(x) E R+(G),
where aG,H(x) _ Ei ni(H, i)G. Since aG(X) is fixed by f Q(x) the subhomomorphisms, {(H, 0,)G}, which appear in aG,H(X), are permuted by the
(abelian) Galois group G(Q(4,)/Q(x)) and the corresponding coefficients,
{ni}, are constant along orbits. Let
5.4.18
stab((H, 4,)G) < G(Q(4,)/Q(x))
denote the stabiliser of (H, 4,)G. Also let

5.4.19
Z(H, q5,) = {g E WGH I g'(4i) = Qii : H --> E*}
denote the stabiliser of (Oi : H -) E') in WGH. If a E stab((H,i)G)

then there exists g(a) E WGH such that
5.4.20
a(oi) = g(a)*(Oi) : H - E'.
Therefore there is an injection
210
5.4.21
yi
: stab((H, cii)G)
WGH/Z(H, 4i),
given by
yi(6G(Q(oi)/Q(X))) = g(o)Z(H, Oi)
Now consider the subsum of aG,H(X) given by the sum of the terms in
the orbit of (H, 4,)G. This subsum equals
5.4.22
ni (UtEG(Q(Oi)/Q(X))/stab((H,o+)G)(H, T(4 )G))
To complete the proof it will suffice to show that applying g(H) to

5.4.22 yields a fractional ideal in ,>f(Q(X)).
However, if w E G then
IridH(0i)(w) =
5.4.23
KEG/H
Oi(x lwx),
X ' WXEH
by 1.2.43. Since
H<Z(H,q5i)<WGH<G,
we may rewrite 5.4.23 as
5.4.24
IndH(4i)(w) = [Z(H, ci) : H] J:xEG/z(H,mil 4i(x iwx).

x-' wxEH
By means of the injection 5.4.21 we may write the coset representatives
of G/Z(H, 4i) in the form

where 1 < u < uo and a runs through the cosets of stab((H, i)G). Hence,
in 5.4.24,
5.4.25
1 nd'(cbi)(w) = [Z (H, 4i) : H] E
xu
x_
Clearly, 5.4.25 is in the image of the trace from Q(4i) to Ki where
Q(X) < Ki < Q(Oi) is the fixed field of stab((H, 4i)G) = G(Q(oi)/Ki).
Hence
5.4.26
g(H)(H, 4i) E .(K,),
so that
5.4.27
g(H)(ni(ETEG(Q(oi)IQ(X))lstab((H,Oi)G)(H, T(4i)G)))
11
tEG(K;/Q(X))
T(g(H)((H, 0,)G)))"I.
211
Combining 5.4.26 and 5.4.27 for each i, we see that

5.4.28
g(H)(aG,H(X)) E .(Q(X)),
which implies that
E Fac(G), as required.
Next we must work towards the detection of PF+(G) of 5.4.6.

This will require an analysis of the composite
5.4.29
RSp(G) ` 0 R(G) - R+(G),

where RSp(G) is the Grothendieck group of symplectic representations
and c is the (injective) homomorphism which sends a symplectic representation to its underlying complex representation (cf. 4.2.32).
We begin by examining the image of a symplectic representation given
by a symplectic line.
Proposition 5.4.30 Let v : G -> Sp(1) be a one-dimensional symplectic
representation then
aG(c(v)) E R+(G)
an integral linear combination of elements of the form

{(H, O)G +(H, )G}, where is the complex conjugate of (k, or of the form
(J, )G where I ndj%) is the complexification of a symplectic representation
of G.
is
Proof The representation, c(v), is two-dimensional. For a two-dimensional representation it is easy to show (see Boltje, 1990; Boltje, Snaith
& Symonds, 1992) that aG coincides with the original Explicit Brauer
Induction formula of Snaith (1988b), which is described in 2.3.21. Hence
5.4.31
aG(c(v)) = pG(XG(c(v))),
where PG and TG are as in 2.3.23-2.3.25.

Suppose that, as in 2.3.23,
5.4.32
TG(c(v)) =
Xa (H
s. `mss'
NT z) E R+ (G, NT 2),
then each g,, lc(v)ga, when described as a complex representation, is the

complexification of a symplectic representation of Ha, since it is conjugate
to the restriction of c(v) to H.
Now consider the homomorphism of 2.3.25:
212
5.4.33
PG :
R+(G,NT2) - R+ (G).
If ga-1c(v)ga(HH) < T2 < NT2 then the image is a cyclic group and there
is a homomorphism, A H,, --> S1, such that
A(h)
$a
c(v)$a(h) =
E U(2)
,(h)
for all h E H. In this case, by definition of PG in 2.3.25,

PG($a lc(v)$a) = (Ha, 2)G + (Ha, L)G.
If got 1c(v)g,,(Ha)
T2 then set
J = Ker(H., - )NT2 --> NT2/T2 = {1})

and let the restriction of ga lc(v)ga to J have the form
g lc(v)$(J) =
for some
C 0J)
4(J) /
J -> S1, then

PG($a lc(v)$) = (J, 4)G.
However, in this second case (Snaith, 1989b, lemma 3.35(v), p. 234),

I ndH"(4) = $ lc(v)ga E R(H,),
which is symplectic. By 5.4.31, this completes the proof of the proposition.
5.4.34 Symplectic explicit Brauer induction
Let G be a finite group and let RSp(G) denote the Grothendieck group of
finite-dimensional, symplectic representations of G. Let RSp+(G) denote
the free abelian group on G-conjugacy classes of subhomomorphisms of
the form (with H < G)
5.4.35
:H-->Sp(1)=53.
Denote by (H, )G the class of 5.4.35 in RSp+(G). Define homomorphisms
Res RSp+(G) -> RSp+(J),

IndjG : RSp+(J) - RSp+(G)
213
and
bspG : RSp+(G) --> RSp(G)
in a manner which is analogous to the complex case.

One may define a symplectic Explicit Brauer Induction homomorphism
by the method of Symonds (1991, section 6),
5.4.36
aspG : RSp(G)
) RSp+(G).
Explicitly, suppose that
p: G
5.4.37
.Sp(n)
is a symplectic representation of G. Let G act on the symplectic projective
space, P(H"), via p. Triangulate P(H") in such a manner that G acts

simplicially. If a is a simplex of G\P(H") and Q is a choice of a simplex
above a in P(H") then the stabiliser, stab(Q), acts via
(v) : stab(&) --> Sp(1)
on the symplectic line given by any point in the interior of 8. Set

5.4.38
aspG(P) _
EQEG\P(Hf)(-1)di,"(Q)(stab(a),
E RSp+(G)
Remark 5.4.39 The method of Snaith (1988b) also yields a symplectic

Explicit Brauer Induction formula
TSpG(p) E RSp+(G),
which is a derivation in the sense of 2.3.28(v). The expressions, TSpG(p)

and aspG(p), are related by a formula
TSPG(P) = aspG(P) ESPG(P),
which is analogous to the formula of 2.5.11 Boltje, Snaith & Symonds

(1992).
5.4.40
c+,G : RSp+(G) -' R+ (G)
by the formula
214
5.4.41
c+,G((H, 4)G) = Indy(c(q5) : H -+ Sp(1) -> U(2)) E R+(G).

Theorem 5.4.42
With the notation introduced above
c+,G aspG = aG c : RSp(G) ---> R+ (G).
Proof Clearly the homomorphism, aG c, is natural with respect to

inclusions of subgroups, J < G,
ResJG
We shall show that c+,G aspG is also natural with respect to ResG
J.
Since 5.4.36 commutes with ResGj we have only to verify that the same is
true for c+,G
If H, J < G then there is a double coset formula for the composition

ResfIndH : R+(H) -
R+(J),
whose proof is the same as that of 2.5.7,

ResJ(Ind4((K,y,)H)) _
IndjjnwHw-1(w*(ResHnW
ijw((K,W)H)))
wEJ\G/H
Hence, if (H, O)G E RSp+(G) with 4 : H -+ Sp(1), then

Res(c+,G((H, cb)G))
= ResY(IndH(aH(c(O) : H -> U(2))))

= X:wCJ\G/H I ndJnwHw-1(w* (ResHnw-lJw(aH(c(O)))))
_ >wEJ\G/H I ndjJnwHw 1 (w*(aJnwHw l (c(ResHnw- Jw(O)))))

EwEJ\G/H I ndjJnwHw 1(aJnwHw 1(c(ResJ wHW ((w-' )* (4 )))))
= c+,r(ResjG((H, 4)G)),
as required.
To complete the proof, as in the proof that aG is unique if it satisfies
2.2.8, we need only verify that each of the natural transformations,

c+,G aspG and aG c, have the same `leading terms'. That is, we must
evaluate the coefficients of the terms of the form (G, A)G which appear
215
in c+,G aspG(p) and ac, c(p), where p is the symplectic representation of
5.4.37, and for this purpose we may suppose that p is irreducible as a

symplectic representation. If p is a symplectic line then
aspG(p) = (G,p : G -, Sp(l))G
and the coefficient of (G, 1)G in c+,G(p) is the coefficient of (G, 2)G in
aG(c(p)), which equals the Schur inner product, < 2,c(p) >, as required.
If p is irreducible but not a symplectic line then there are no terms of the
form (G, O)G in aspG(p) and no terms of the form (G, A)G in c+,G(aspG(p)).
On the other hand, in this case, if 2 : G -> U(1) were a one-dimensional

representation for which < A, c(p) > was non-zero then 2 1 would be a
summand of c(p) (even if A = 7) and 2 would be the complexification
of a symplectic summand of p, which is impossible since p is irreducible.
Hence there are no such terms in aG(c(p)) in this case and again the
leading terms are equal.
Corollary 5.4.43 Let p be a symplectic representation of G, as in 5.4.37.
Then
aG(c(p)) E R+(G)
is an integral linear combination of elements of the form {(H, 4)G+(H, )G},
where 0 is the complex conjugate of , and of the form (J, )G where

I ndjG() is the complexification of a symplectic representation.
Proof By Theorem 5.4.42 we have

aG(c(P)) =
niI nd4.(aH,(c(cti) : Hi -' Sp(1))),
where
aspG(p) _
ni(Hi, 4i)G E R+(G).
By 5.4.30 each aH;(c((ki)) is an integral linear combination of terms of

the form {(K, ipi)Hi + (K, iV,)H, } or of the form (J,.u)Hi where I ndH'() is
symplectic. In the latter case I ndjG() = I ndH. (I ndH ()) is also symplectic,
In 5.4.16 we detected Fac(G) by means of the groups,

FacG(Hb). Now we define corresponding groups, PFG (Hn), by means
of which we can detect PF+(G) of 5.4.6.
Definition 5.4.44
216
For each subgroup, H < G, define a subgroup (cf. Definition 5.4.5)

PFG (Hab) C FacG(Hab)
5.4.45
to be the subgroup consisting of all f E FacG(Hab) such that

(i) f (O : Hab --> E') is a principal fractional ideal for all 0 : Hab
E' and
(ii) for all 0 in (i) such that IndH(InfHab(4)) is symplectic the fractional
ideal
f(O : Hab -) E`) E J(Q(IndHG(InfHpb(4))))
has a generator which is positive under all Archimedean places of E.

Theorem 5.4.46 The homomorphisms of 5.4.16 induce maps
PF+(G)
PFG(Hab)waH
(H)
H<G
and
A'
(H)
PFF+ (Hah)wH
H<G
PF+(G).
Proof For f E PF+(G), the (H)-component of

BG(f)(O : Hab
Thus it is clear that
is given by
E")(H) = f(IndH(InfHab(cb)))
takes its values in principal fractional ideals
and is totally positive on 0 whenever I ndH (Inf Hab (O)) is symplectic, which
proves the assertion concerning
Let us examine
where g(H) E PFG
(Hab)WGH. In the notation
of 5.4.17
Ac({g(H)})(X) = fig(H)(aG,H(X))
This is a principal fractional ideal and it is totally positive, in the sense

of 5.4.4, when X is symplectic, by virtue of 5.4.43. This is because
g(H)(0)g(H)(0) automatically has a totally positive generator and g(J)()
has, by assumption, a totally positive generator when I ndJ () is symplectic.
Definition 5.4.47
For H <_ G define'2'G(Hab) to be the quotient

rPYG(Hab)
FacG(H ab)
PFG+(Hab)
217
Hence W'G(Hab) depends upon G and upon H, up to conjugacy, not

merely Hab. Notice that there is an isomorphism of WGH-invariants:
'2' (Hah)WGH
5.4.48
Faca Hab WGH

PFG (Haa)WGH
This is seen in the following manner: WGH acts on R(Hab) by permuting

the basis of one-dimensional representations, 0j, preserving the character
fields, Q(IndH(InfHab(4;))), and the property that IndH(InfHa,(0;)) is

symplectic. Therefore PFG (Hab) is a direct sum of WGH-groups of
the form (Z[WGH] Z[U, Z) X, where X is torsion free and WGH
acts trivially on X. Such WGH-modules have trivial one-dimensional
cohomology since (see Snaith, 1989b)
H'(WGH; (Z[WGH]
(&ZIUI
H'(WGH;Z[WGH]
Z) X)
Z[u] Z) X
H'(U; Z) X
Hom(U, Z) (9 X,
which is trivial, since U is finite. Therefore the short exact sequence of

WGH-modules
PFG (Hab)
) FacG(Hab)
) WYG(Hab)
yields a long exact cohomology sequence (Snaith, 1989b, p. 9) which

reduces at the left-hand end to the short exact sequence
PFG(Hah)WGH
) FaCG(Hah)WGH
)'WYG(Hah)WGH
l
,
since the next term is H'(WGH;PFG(Hab)), which is zero.

Combining 5.4.16 and 5.4.46 we obtain the main result of this section:
%2(A(G)) BG
(H) c
2G(Hah)WGH
H<G
and
(D (H)
H<G
which satisfy
1.
WYG(Hah)WGH A;) WY(A(G))
218
5.4.50
For each H < G and for each J < Hab we may define the groups
FacG(Hab/J), PFG+(Hb/J) and c2G(Hab/J)
in a manner analogous to that of 5.4.15, 5.4.44 and 5.4.47. In addition,

if J < J' < Hab, there are inflation maps
Inf : FacG(Hab/J) --> FacG(Hab/J')
and
Inf : PFG(Hab/J)
) PFG(Hab/J')
We may assemble these maps into a homomorphism
liminf : W'G(Ha')
(2G(Hal/J).
lim
Hb/Jcyclic
Lemma 5.4.51 The homomorphism of 5.4.50 is an isomorphism and induces

an isomorphism
liminf : W2G(Hab)WH
WYG(Hab/J)
lim
Hab/Jcyclic
Proof The homomorphism of 5.4.50 is clearly WGH-equivariant. Hence

it suffices to show that it is an isomorphism.
Define a map
p:
lim
oYG(H''/J)
WYG(Hab)
Hb/Jcyclic
on representing functions in FacG(H'/J) by

iV(tgHa6/J})(0
Hab
E*) = gHb/Ker(O)(4 : Hab - + E`) E 5(E).
This is clearly well-defined and
(lim.- I nf)(p({gHae/J}))(2 : Hab/J
Hab/J
V'({gHab/J})(Hab
= gHa6/Ker(xrz)(2Tr :
Hab/Ker(An)
but Ker(1,r) >_ J so that

gHa6/J(2) =
so that (limesb/J I of )tp = 1.
A-'
E*)
E-)
E`)
5.5 Swan subgroups for nilpotent groups
219
Conversely,
W(lim_Inf)(g)(0 :
Hab
-) E')
= (lim_ 1 n f (g))(0 : Hab/Ker(4) -) E')

= g(c1 : Hab
) Hab/Ker(4) -* E*)
so that W(limHab/J I nf) = 1 also.
11
Combining 5.4.49 and 5.4.51 we obtain the following result:

WGH
16'(A(G))
(H)
lim
c2'G(Hab/d)
Hb/Jcyclic
and
(H)
lim
'eyo(Hab/J)
WY(A(G))
Hb/Jcyclic
which satisfy AGBG = 1.
In this section we shall study the Swan subgroup, T(G), in the case when
G is a nilpotent group. A finite nilpotent group is of the form
5.5.1
G = G(pl) x ... x G(p,),
where G(pi) is a pi-group and pl,...,p, are distinct primes. The Swan
groups, T(G(pi)), are known by 4.4.20 so that the determination of T(G)
in 5.5.1 constitutes the next case in the hierarchy of difficulty.
By Curtis & Reiner (1987, section 53.13, p. 347) the restriction maps
may be assembled to yield a surjection
5.5.2
ResG(pi) : T (G) -' i i T (G(pi))
Therefore our problem is to determine the kernel of 5.5.2. This is a

particular case of the following question, which was posed to me by M.J.
Taylor and was first raised by C.T.C. Wall (1979, p. 546) :
220
Question 5.5.3
Let G1 and G2 be finite groups of coprime order. What
is the kernel of
ResGI x ResG2 : T(G) --> T(G1) T(G2)
when G = G1 x G2?
5.5.4
In 5.5.1 suppose that #(G(pi)) = p"'. Recall that, if p is odd,

(Z/P")* = FP
When p = 2 and n
r 1 PZ
1+ p Z
Z/(p - 1) X
Z/p"-1
2 we also have
(Z/2n)* ={1} x
+4Z)
1+2Z
C I
= {1} x Z/2 n-2.
Write A(i) for F. and B(i) for Z/p"'-1 when pi # 2 and A(i) = {1},
B(i) = Z/2ni-2 when pi = 2. Let A(G(pi)) denote the Artin exponent of
G(pi) (cf. 4.4.15) so that the map, S, of 4.4.14 factors to give a surjection
5.5.5
S : {(fl 1 A(i) x B(i))/{1}} (Z/(fl;=1 A(G(pj))) -* T(G),
where G is as in 5.5.1. In some cases, for trivial reasons, 5.5.2 and 5.5.5
combine to show that 5.5.2 is an isomorphism. However, when some pi
divides a (pj - 1) the image of the factor A(j) under S may yield some
èxotic torsion' of pi-primary order. Exotic torsion of this type will lie in
the kernel of 5.5.2.
The most basic nilpotent groups whose Swan subgroup might admit
exotic q-primary torsion are of the form
Z/p x (Z/q)n,
where p, q are primes and q divides (p - 1). As we shall see, the exotic qprimary torsion in T(Z/p x (Z/q)') is the image of the q-primary torsion
in FP and, typically, the order of the cyclic group of exotic q-primary
torsion in such a group increases with n to a limit. For example, the

exotic 2-primary torsion in T(Z/3 x (Z/2)n) is trivial for n = 1 and of
order two for n > 2. In T(Z/5 x (Z/2)n) the exotic 2-primary torsion is
trivial when n = 1, is of order two when n = 2 and is of order four for
n>-3.
In this section we will determine the exotic q-primary torsion in these
examples. Our motivation for studying these elementary examples lies in
the following conjecture.
221
Conjecture 5.5.6 The exotic q-primary torsion (i.e. q-primary torsion in

the kernel of 5.5.2) is detected by the family of maps
r
T(G) = T
G(pi)
T(Z/p x (Z/q)n),
i=1
induced by passing to the Swan subgroups of subquotients of G of the

form Z/p x (Z/q)n where p and q are primes and q divides (p - 1).
5.5.7
Let p and q be primes such that q divides (p - 1). Set bm =
exp(2ni/m) and let
H(p, q; n) = Z/p x (Z/q)n.

Let k be a positive integer such that
k-1
5.5.8
(mod qn)
and
kq`
(mod p)
for some 1 < c E Z.

We wish to analyse when
5.5.9
0 = S(k) E T(H(p, q; n)).
Let us establish some notation for the one-dimensional representations
of H(p, q; n). For Z/p =< x >, say, let 0 : Z/p -> Q(cp)' be given
by (x) = gyp. The one-dimensional representations of Z/p are then
1, 0, 02'...' Op-1. For (Z/q)' let yi generate the ith copy of Z/q and
: (Z/q)n -+ Q(q)" to be given by
define Xi
if i = j,
Xi(Yj) =
if i * j.
Write X(j) = Xi` ... XJ," so that the irreducible representations of
H(p,q;n) are all of the form X(j) S for some j = (j1,...,jn) and
0<s<p-1.
Using these characters the p-adic group-ring, Zp [H(p, q; n)], may be

decomposed in the following manner. Let b be a set of multi-indices,
j, such that the set {X(j) j e .9'} is a complete collection of representatives for the
of the X(j)s. Define a ring
I
homomorphism
222
5.5.10
[ljE." X(J) : ZP[(Z/q)n] -p ZP X
11o,,jEy ZP
where X(j)(E mgg) = > m X(j)(g). The homomorphism of 5.5.10 is

an isomorphism which commutes with the action of Fq as G(Q(gq)/Q)
on the right and diagonally on each Z/q-factor as Aut(Z/q) on the left.
Since q E ZP when q divides p - 1 we have an isomorphism
q-1
ZP
given by sending x
to
x 9, ... , x 9-1). Combining this isomorphism with 5.5.10 yields an isomorphism
5.5.11
fj X(J) :
ZP[(Zlq)n]
-- fj ZP.
To see that 5.5.10 and 5.5.11 are isomorphisms we recall that the corresponding map
II X(I) : QP[(Z/q)n] -+ QP X fi Q Z[Sq]

jEq'
fQP
0* jE9'
is well-known to be an isomorphism (Lang, 1984, pp. 641-645). The

isomorphisms of 5.5.10 and 5.5.11 are given by the restriction of this
isomorphism to the maximal Zr-orders of each side (Curtis & Reiner,
1981, section 27.1, p. 582). Therefore, taking group-rings for Z/p, we
obtain an isomorphism of the form
5.5.12
HjE.? X(J) : ZP[H(P, q; n)]'
ZP [Z/P]' X 11 Zp
0* jE.9'
[Z/P]' fl ZP [Z/P]',
where ZP[l:q] denotes Z, 0 Z[l;gJ.

Similarly, interchanging the roles of p and q, we obtain an isomorphism
of the form
5.5.13
(1, 0) : Zq [H(P, q; n)]' --> Zq [(Z/q)n]' X Zq
Recall from 4.2.37 that there is an isomorphism of the form
[(Z/q)n]
223
D(Z[H(p, q; n)])
5.5.14
Homnn (R(H(P,q;n)),(Z, [dry])' X (Z,
HomnQ(R(H(P,gt)),(
[pq
XZq
v[H(
Proposition 5.5.15 In the notation of 5.5.7 and 5.5.8 suppose that there
exists a unit, u E Z[lp]' such that

Zg [gyp];
1=uE
(1 + gnZq [cp])
and k = u E Fp, then

0 = S(k) E T(H(p, q; n)).
Proof We define local units, a E Zp[H(p, q; n)]' and a E Zq [H(p, q; n)]',

by means of the isomorphisms of 5.5.12 and 5.5.13.
Define P to be equal to
f = (1 + (k - 1)q-nv,1 + (u - 1)q-n6),
where a = >gE(Z/q)" g.
Now consider the pullback square
ZP [Z/pl
ZP'
ZP [bP]
F;P
in which e denotes the augmentation. Choose ul E Zp[Z/p]' such that

4(u1) = u and e(ul) = k and seta equal to the unit whose 1-component
is ul and whose j-component is trivial when x(j) = 1.
Finally, define f E
by f(1 0 0) = u,
f (x(j) (& ) = 1 when x(j) # 1 and f (x(j) 0 1) = 1.
Hence
(Det(a), Det(fl)) f -1 is trivial on all x(j) S T 1 and (Det(a), Det(fl))

f-1(1) = (k,k). By 4.2.47 this is the function that represents S(k).
Now we come to the main result of this section.
224
Theorem 5.5.16 Let k be an integer as in 5.5.8. Then, in the notation of

5.5.7,
0 = S(k) E T (H(p, q; n))
if and only if there exists u E Z[ip]* such that

1=UE
Zq [`gyp]
(1 +gnZq[ p])
andk=uEF;.
Proof We must prove the converse of 5.5.15. We shall prove this result
(when q * 2) by means of the determinantal congruences of 4.3.37, which
amounts to the analysis of the maps, ST, of 4.4.11 when T = E, X(j)
and T = Ej X(.1) 0.
Let us begin by supposing that q * 2.
If S(k) = 0 then there exist local units a E Z, [H(p, q; n)]' and

Zq [H(p, q; n)]" together with a global function
fE
such that
Det(a)(X(J) 0) = f(X(J) 0 0) = u(J) E Zp[bpgl %

Det(fi)(X(J) 0) = f (X(J) (D 0) = u(j) E Zq [cpgl *,
5.5.17
Det(a)(X(J) 0 1) = f(X(j) 0 1) = a(j) E ZpNJ *

if X(j) 1,
Det(l3)(X(J) 0 1) = f(X(j) (9 1) = a(j) E
for all X(j),

and Det(a)(1) = kf(1) = ka(0) E
This is because there is a q-adic unit, namely y = 1 + (k - 1)q-nr E

Zq[(Z/q)n]' where a = EgE(Z/q)" g, such that X(j)(y) = 1 if X(j) # 1 and
X(o)(y) = k.
Now consider the representations of H(p, q; n) given by
Pi =
X(.1)
1
and P2 =
X(j) 4
1
We have pi = 0 (mod qn) for i = 1, 2. Hence, for i = 1, 2,

5.5.18
Det(f)(Ipq(p,) - qpi) E 1 + qn+1Zq [pql,
225
by 4.3.37. However, lpq(p1) = qn and lpq(p2) = q"4 so that

-q
-q
u(O)gn-q
l
fi u(j)
and
a(O)q"-q
fJ a(j)
Oj
(2*j
both lie in 1+qn+1Zq[Spq] However, both these elements are qth powers
of
and therefore lie in the qth powers of the subgroup
Zq [p] *, which contains no non-trivial qth roots. Hence both
u(Q)q"-1-1 (flu(i))
a(O)q'-'-1 (TT a(1)
and
O*j
9::jj
n
On the other hand, at the prime p,
lie in (1 +
Det(a)(p"(x(j) 0) - px(j) 0 0) and Det(a)(ipP(X(j)) - px(j))
both lie in 1 +
Also, W"(X(J) 0 0) = x(J) so that, if j
0,
u(j) - u(j)p (mod p)

a(j) (mod p)
and
u(0) = u(0)" (mod p)

ka(0) (mod p).
Therefore
5.5.19
u=
lies in (1 + gnZq
[a(Q)q'-'- I
llo#j
u(j)][U(O)q"-'- I
11o
j a(j)]-1
n Z[ip] and
u - kq"-1-1
(mod p).
For q # 2 the result now follows since k has q-primary order (modulo p).
When q = 2 a similar proof would result, by 4.3.37, only in the existence
of u E Z[] ' such that u2 E (1 + 2n+1 Z2 [gyp]) n Z [ p]' and k = u E F;.
Therefore we resort to another proof in this case.
We will use an explicit description of generators of Z2 [H(p, 2; n)]' in
order to show that the element, u, of 5.5.19 lies in Z
n (1 +2Z2[]).
2n Z2
226
By 5.5.13 we may examine Z2 [l; p] [(Z/2)"]' and Z2 [(Z/2)"]' separately.

By 5.7.10 generators for Z2 [2;p] [(Z/2)"] * may be taken to be either torsion
units of the form
where co E Z2 [l; p]* is a root of unity and g E (Z/2)",
or to be torsion-free units of the form

5.5.20
1 + (v -1)2-S(1 + L1)(1 + L2) ... (1 + LS),
where v E 1+2SZ2 [gyp] and, without loss of generality, we may assume that
L1,.. . , LS are distinct characters from the set {Xl, ... , X"}. If
(A1,132)
in 5.5.13 with $2 having the form of 5.5.20 then the factor

5.5.21
u(0)2"-'-1(l
to j
u(J))-1
v2"-`-2"-r
in 5.5.19 is equal to
which lies in 1 + 2"Z2 [gyp]. On the other
hand, if #2 is a torsion unit, w g, then the product in 5.5.21 is equal
to jig X(j)(g), which is equal to 1 since g2 = 1 and n >_ 2. A similar
discussion applies to #1 E Z2 [(Z/2)"]' and shows that the factor
5.5.22
a(0)2"-`-1(Ho, j a(J))-1
lies in 1 + 2"Z2. By 5.5.17, u, in 5.5.19, is a product of units and therefore

lies in Z [lp]' n (1 + 2"Z2 [l; p] ), as required.
We shall now use Theorem 5.5.16 to verify Conjecture

5.5.6 in some cases of the form G = G(p) x G(q), where p, q are primes
and q divides p - 1. We will begin with the case when p = 3, q = 2.
Suppose that G = G(3) x G(2) with #(G(2)) = 2S and #(G(3)) = 3`
with 1 < s, t. In this case the exotic 2-primary torsion can be at most a
copy of F3, which is of order two. By 5.5.16, this copy of F3 is detected
Example 5.5.23
in T(Z/3 x (Z/2)1) for any n >- 2. This is because Z[3]' is cyclic of

order six, generated by (-6) and 1, b3, 2;3 are distinct in the finite field,
F4 = Z2 [g3] Z/2, while +1 are distinguished by reduction modulo four.
Note that every non-abelian 2-group has either a normal subgroup or a
quotient of the form (Z/2)" with n >- 2 (Gorenstein, 1968, theorem 4.10,
p. 199). Table 5.1 summarises the immediate implications of 5.5.16 for
the Swan subgroup, T(G(3) x G(2)). In all cases the F3 denotes exotic
torsion of order two.
Much of Table 5.1 is justified by means of the following observations:
(i) Artin exponent considerations (4.4.15, 4.4.17): When G(3) (resp. G(2))
is cyclic the Artin exponent of G = G(3) x G(2) is a power of 2 (resp. 3).

See Curtis & Reiner (1987, p. 365) for the Artin exponents of a p-group.
227
Table 5.1. Table of T(G(3) x G(2)) (#(G(2)) = 2s and #(G(3)) = 3t)

G(3)cyclic
G(2) (s23)
F3 xZ/2
quaternionic
semi-dihedral
Z/3t-1
{1}
G(2)cyclic
G(2) (s2.4)
G(3)non-cyclic
F3 x Z/2 x
Z/3t-1
F3 x (Z/2 or Z/4)
F3 x (Z/2 or Z/4) x Z/3t-1
F3 x({O} or Z/2)
F3 x({O} or Z/2) x
G(2) (s22)
dihedral
Z/3t-1
G(2) any other
2-group
F3 x Z/2s-2
F3 x Z/21-2 x Z/3t-1
(ii) When G(2) is non-cyclic then G has a quotient of the form Z/3 x
(Z/2)2.
The following result concerns the ambiguous cases.
Proposition 5.5.24 Let SD, denote the semi-dihedral group of order 2n+2
(n >- 2) and let D8 denote the dihedral group of order eight. If T(Z/3t x
D8) = F3 then
T(Z/3t x SDn) = F3 x Z/2.

Proof The semi-dihedral group (Curtis & Reiner, 1987, p. 349; 4.4.20) is
given by
= 1 = b2, bab = a2"-1 }.
SDn = {a, b I
a2"+t
We may apply Theorem 5.3.3 to the case when
G = Z/3t x SDn and J = Z/3t x< a >= Z/2n+13'
The elements of order two in G - J are of the form {a2Jb} whose

normaliser, NG < a2ib >, is isomorphic to Z/3t x D8. Let k be an integer
which is coprime to #(G) and is congruent to 1 modulo Y. By Theorem
5.3.3
0 = 2S(k) - I: IndZ/3'xo8(S(k)) E T(Z/3t x SDn).

s
However, by hypothesis, 0 = S(k) E T(Z/3t x D8) so that 2S(k) = 0.

Hence the surjection, S of 4.4.14 becomes
S : Z/2 x F3 --) T(Z/3t x SDn).
The Z/2-factor is detected by T(SDn) = Z/2, while the F3-factor is

detected by Theorem 5.5.16.
228
Consider Theorem 5.5.16 in the case when p = 5 and

q = 2. The units, Z[i;s]', are generated by (-N and (1 + N. Also
Example 5.5.25
(1 + 5)ls 1-
3) (mod 8).
From Theorem 5.5.16 ones sees easily that S(17) has order two in T(Z/5 x
(Z/2)2) and has order four in T(Z/5 x (Z/2)3). In fact, 5.5.5 induces a
surjection
S : (FS x {l})/{1} -- ) T(Z/5 x (Z/2)2),

which shows that S(FS) generates T(Z/5 x (Z/2)2) so that, by 5.5.16,
S : F/(F)2 -> T(Z/5 x (Z/2)2).

Similarly, there is an isomorphism
S : FS x Z/2 -> T(Z/5 x (Z/2)3).

By Theorem 1.1 of Gorenstein (1968, p. 123), if G is a 2-group all
of whose elementary abelian subquotients, (Z/2)k, have rank at most
two (k < 2), then G has at most two generators. For example, when
#(G) = 16 the non-abelian groups of this type are D16, Q16, SD4 and
the semi-direct product of the form Z/4 oc Z/4 (Thomas & Wood, 1980,
Tables 16/12, 16/14, 16/13 and 16/10 respectively). It is not difficult, by
induction on the order, to classify all such non-abelian 2-groups. One
method for such a classification is to study central extensions of the type
{1}>Z/2oG>H){1}.
In this extension H is again of the same type and the isomorphism
classes of possible Gs are in one-one correspondence with the Aut(H)orbits of the cohomology group, H2(H; Z/2). The cohomology groups,
H' (D2. ; Z/2) and H' (Q2- ; Z/2), are described in Snaith (1989b, p. 24
and p. 34 respectively). For example, when n >_ 3, H2(D2n;Z/2) is an
F2-vector space with basis w, xi and x2 in the notation of Snaith (1989b,
p. 24). There is an outer automorphism,
: D2n - D2n such that
V(w) = w, V (x2) = x1 and '(x2) = x1 + x2. Hence H2(D2'; Z/2) has
six Aut(D2.)-orbits which correspond to Z/2 x D2n, D2n+', Q2'+,,
Z/4 oc Z/2n-1 and Z/2 oc (Z/2i-1 x Z/2) (cf. Thomas & Wood, 1980,
tables 16/6, 16/12, 16/14, 16/13, 16/10 and 16/8), when n = 3. Hence G
must be one of D2n+i, Q2-+t, SDn_1 or Z/4 oc Z/2r-1
Now let G(2) and G(5) be a finite 2-group and 5-group respectively.
The following table summarises the immediate implications resulting from
229
Table 5.2. Table of T(G(5) x G(2)) (#(G(2)) = 2s and #(G(5)) = 5i)

upper bound
lower bound
T(G(5))
T(G(5))
FS/(FS)2 x Z/2 x T(G(5))
FS /(FS )2 x Z/2 x T(G(5))
FS x Z/4 x T(G(5))
FS/(Fs)2 x Z/2 x T(G(5))
FS /(FS )2 x Z/2 x T(G(5))
FS/(Fs')' X T(G(5))
FS x Z/2s-2 x T(G(5))
Fs /(FF )2 x Z/2 -2 x T(G(5))
FS x Z/2 -2 x T(G(5))
FS x Z/2 -2 x T(G(5))
G(2) cyclic
G(2) (s>_3)
quaternionic
G(2) (s24)
semi-dihedral
G(2) (s22)
dihedral
others with rank of
subquotients < 2
G(2) any other
2-group
Theorem 5.5.16 and elementary considerations such as Artin exponents.

In the table the analogue of 5.5.24 relates the ambiguous dihedral and
semi-dihedral cases.
Lemma 5.5.26 Let p and q be primes such that q properly divides p - 1.

Let f denote the order of q in the multiplicative group, FP. Hence f >- 3.
Then
q-1
(1 -P)qf -1 = 1 - E(q + q/2 +... + ql
f-t
(mod q2).
i=1
Proof If HCF(s, q) = 1 then (qf - sqe)/(sqe) = of-e/s - 1 is a q-adic

unit which is congruent to (-1) (modulo q2) provided that e < f - 2.
Applying this observation to the factors in the binomial coefficients
of -I
(qf - 1)(qf - 2)...(qf - i)
shows that the coefficient ofp in (1- P)qf -1 is congruent to one modulo
q2 except when i = sef-1 with 1 < s < q - 1. In the latter case the
coefficient is easily seen to be equal to
(1 - q)(1 - q/2) ... (1 - q/s).

Since of - 1 is divisible by p we have 1 + P + ... + qf-1 = 1 and the
result follows immediately.
Example 5.5.27
Consider Theorem 5.5.16 in the case when p = 7 and
q = 3. The units, Z[7}, are generated by
a = (1 - )/(1 - 7)
230
Table 5.3. Table of T(G(7) x G(3))(#(G(3)) = 3S and #(G(7)) = 7`)
G(7) cyclic
G(7) non-cyclic
G(3) cyclic
G(3)
{1}
Z/3 x
Z/3 x
Z/7t-1
non-cyclic
Z/3s-1
Z/7t-1
Z/3s-1
and
(1 - l;7 )/(l - 7). When considering 3-primary torsion we may
ignore the roots of unity. Also, by 5.5.26 (with f = 6),
(a#b)tl =_ 1
- 3a1;7 -
(3a +
(mod 9),
so that as#b is congruent to 1 (modulo 9) only when a and b are divisible
by 3. In this case the reduction of aafn (modulo 7) is equal to (-1)1/3

Therefore, since S(-1) is always trivial in the Swan subgroup, 5.5.16
implies that T(Z/7 x (Z/3)2) contains exotic torsion of order three.
It is straightforward to verify the contents of Table 5.3.
5.6 Cyclic groups
As explained in 7.1.29 a finite abelian group, X, together with a G-action

which is cohomologically trivial, yields a class
[X] E WY(Z[G]).
For such a Z[G]-module there exists an exact sequence of Z[G]-module

homomorphisms of the form
0-fP1->Po--) X-k0,
in which PO and P1 are finitely generated, projective Z[G]-modules. The
class, [X], is defined by
[X] = [Po] - [Pi] E WY(Z[G]).
In this section, by way of illustration, we shall recall some elementary facts about several examples of cohomologically trivial modules
(and their Hom-descriptions) over cyclic groups in which the underlying
module is also a cyclic group. These modules are closely related to
the modules of roots of unity which will enter into the computation, in
Chapter 7, of the global Chinburg invariant in the real cyclotomic case.
Let p be a prime. Let G denote a cyclic p-group of order pS with
5.6 Cyclic groups
231
Table 5.4. H'(G; Ma,t)

Generator, even i
Generator, odd i
pt-a-t
pt-s
pt-a-1
2t-a-2
2t-s
Z/2a+2
2t-a-2
Z/2
2t-'
H`(G; Ma,t)
p odd,
Z/pa+t+s-t
s<t
ts
p odd,
Z/pa+l
p = 2, s < t,
1a
Z/2a+2+s-t
p = 2,t < s,
1<a
p=2,s=t-2,
a=0
p=2,t-1<s,
a=0
generator, g E G ^_' Zips. Let Ma,t denote the cyclic group, Z/pt with a
non-trivial G-action given, for x E Ma,t, by the formulae
5.6.1
g(x) =
(1+p)Px
if
p'2,t-1-a<s,
32' x
if
p=2,t-2-a<s.
Recall that the Tate cohomology groups, H`(G; Ma,t), coincide with the
ordinary cohomology groups in positive dimensions and are given by
if
0 (mod 2),
Ker(N)/((Ma,t)G) if
1 (mod 2),
(Ma,t)G/N(Ma,t)
5.6.2
H`(G; Ma,t) _
where N = 1 +g +... +gPs-1

The Herbrand difference, #(H(G;Ma,t))-#(H'(G;Ma,t)), vanishes for
cyclic groups and cyclic modules so that the groups of 5.6.2 are cyclic
groups whose order is independent of i. From these observations one
sees that the Tate cohomology groups, H'(G; Ma,t), are given by Table
5.4 (in which the generators lie in Z/pt = Ma,t).
From Table 5.4 we see that Zipt = Ma,t is a cohomologically

trivial Z [Z/ps]-module precisely when p is odd and a = t - s- I or when
p = 2 and a = t - s - 2. In these cases we obtain a class
5.6.3
[Ma,t] E '2(Z [Zips] ).
232
5.6.4 The Hom-representative of Ma,t
Now let us determine the Hom-description, in the sense of 4.2.13, of the

class of Ma,t.
Let q = (1+p)P' with a+s+ 1 = t if p is an odd prime and a+s+2 = t

if p = 2. If < g >= Z/ps then g(x) = qx for x E Mat = Z/pt. Consider
the surjection, q : Z[Z/ps]
Ma,t, given by q(1) = 1. Hence Ker(q)
contains pt and g - q. If rl(Ei aigi) = 0 then Ei aigi 0 (pt) so that
aig` = E i(g` - q`) + mpt
i
for some m and therefore this element belongs to < g - q, pt > 1Z[Z/ps].
If 1 is a prime different from p then we have
Zi [Z/ps] < 1 >= Zi Z < g - q, pt >
At the prime p, in ZpZ < g - q, pt > we have the identity
(g -
q)(gP`-1
+ ggPs-2 +... +
However, for all primes p, 1 - q
ps
qp,-1)
= 1 - q".
= ptv where HCF(p, v) = 1. Therefore
Zp [Z/ps] < g - q >= Z p Z < g - q, pt > .
Therefore the idele, jail E J`(Q[Z/ps]) corresponding to this choice of

bases is given by
g-(1+p)" if l=p,
5.6.5
al =
1
otherwise.
The following result is immediate from the previous discussion.

Proposition 5.6.6 Under the conditions of 5.6.4 the cohomologically trivial
Z[Z/ps]-module, Ma,t, is represented in the class-group by the function
(x(g) - (I +
p)p)-1
if l = p,
xH
1
otherwise.
5.6.7 Some more Hom-representatives
Let p be any prime. Let u, t be positive integers such that t >- u + 1 if p is

odd and t >- u + 2 if p = 2. Let Mp denote the Z[Z/pu]-module given by
5.6 Cyclic groups
233
Z/2[Z/2] if p = 2, u = 1,
Mp =
5.6.8
M1_i_2,1
if p = 2, u > 1,
Mt_u_1,,
otherwise.
Here MQ,t is the module of 5.6.1. Hence Mp is a finite, cohomologically

trivial Z[Z/pu]-module.
Suppose now that n is any positive integer prime to p so that, if Z/n

acts trivially on Mp, then Mp is a finite, cohomologically trivial Z[Z/npu]-
module. We shall `now generalise the results of 5.6.4-5.6.6 to give the

Hom-representative of the element
[Mp] E 712(Z[Z/npu]).
5.6.9
Theorem 5.6.10 Let x, y be generators of Z/n and Z/pu respectively. The

class of Mp in 5.6.9 is represented in terms of the Hom-description by the
function which is trivial except at the p-adic places where it is given by
if p = 2,u = 1,
Det(x + 1)-1
Det(xy -
5.6.11
321-a-2
if p = 2,u > 1,
)-1
Det(xy - (1 + p)P'
')-1
otherwise.
Proof Suppose that p = 2 and u = 1. The kernel of the map

W : Z[Z/n x Z/2] ---> Z/2[Z/2]
given by p(x) = 1 and ip(y) = y is equal to the ideal, < 2,x - 1 >.
However, in Z2 OZ < 2, x - 1 >, we have
(x +
1)(xn-1 - xn-2 +...
+ (-1)n-1) = xn + (-1)n-1 = 2
so that Z2Z < 2, x - 1 >= Z2 [Z/n x Z/2] < x + 1 >, as required.

In the remaining two cases set a = (1 + p)p where a = t - s - 1 if p is
odd and a = t - s - 2 when p = 2. The kernel of the map
W : Z[Z/n x Z/p'] --) Mp,

given by p(x) = 1 and W(y) = a, is equal to the ideal < p`, y - a, x - 1 >.
It is clear that
ZpZ<p`,y-a,x-1>=ZpZ<y-a,x-1>.
234
However, arguing in the same manner as before, we obtain the identities

a)((xy)n-1 + a(xy),,-2.+
(xy -
... +
n-1)
= y" - a"
and
(xy -
a)((xy)pu-1
+ a(xy)pu-2 + ... +
aPu-1)
= xpu
apu
= xpu - 1 + vpt
for some integer, v, such that HCF(v,p) = 1, which shows that
ZpZ<p`,y-a,x-1>=ZpZ<xy-a>,
as required.
5.6.12 The cyclic group of order p2
Let p be an odd prime and let n denote a primitive nth root of unity,
as usual. Let g be a generator of the cyclic group of order p2. For the
remainder of this section we shall be concerned with calculations in the
kernel subgroup
D(Z[Z/p2]) c P'(Z[Z/p2]).
We shall show that the class of 5.6.4 ( with t = 3, s = 2) and 5.6.7:
[Mo,3] = [MP] = -[< p3, g - 1 _P>] E D(Z[Z/p2])

can sometimes be non-zero.
Set
Z[ip] = Z/p 0 Z[AP]
Z/p[iri]/(ni-'),
where x = 1- gyp. From Curtis & Reiner (1987, p. 285) there is an exact
sequence of the form
5.6.13
Z[lp]' ---> X[pp] -L D(Z[Z/p2])
--)0,
in which the left-hand map is given by reduction modulo p. This exact

sequence is derived from the exact K-theory Mayer-Vietoris sequence of
a pullback (fibre square) of the following form:
5.6 Cyclic groups
235
Z[41
In this diagram the ring, A, is given by

A = ZE J/((Dp(g)(Dp2(g)),
where ap(t) = 1 + t +... + tp-1 and 12(t) =I + tp +... + OP-1) are

cyclotomic polynomials. The unlabelled homomorphisms in the diagram
are given by sending g to p or [;p2 and by reduction modulo p. If
72 = 1- pz then we have an isomorphism of the form

Z/p
Z/p[7r2]/(7L2
The map, 0, is given by composing reduction modulo p with the map

sending 7[2 to 711.
Lemma 5.6.14 If w denotes the reduction of w modulo p then the homomorphism
0:
Z/p
is given by
0(z) =
where NLIK denotes the norm for the field extension, L/K.
Proof The map, 0, is the ring homomorphism which is characterised by

O(1) = 1 and 4(7<2) = 7t1. Therefore it suffices to verify that the map, p,
defined by
p(z)
is a ring homomorphism. Since the norm is multiplicative, so is W.
236
such that
However, given z, z1 it is clear that there exists u E
,p(z + zi) = w(z) + W(zi) +

where TL/K denotes the trace map for the field extension, L/K. However
1P ,2 if i=pj,
TQ(p2)/Q(p)(p2) _
0
Hence
otherwise.
0.
Now consider the projective Z[Z/p2]-module given by the ideal

5.6.15
< p3, g - 1 - p >= 14 Z[Z/p2].
Clearly, one has a canonical isomorphism of the form

A Z[Z/n2] I ='< p3, g- 1- -p > -4A.
This is seen by considering the exact sequence of cohomologically trivial
Z[Z/p2]-modules
0 --> I --> Z[Z/p2] --> Z/p3
)0.
Upon tensoring with A we obtain an exact sequence of the form

TorZ[Z/p2](A,Z/p3) --- A Z[Z/p2] I -
A.
However, Tor2[Z/P'](A,Z/p3) is annihilated by p3, while A Z[Z/p2] I is

torsion free, since it is a summand of
A Z[Z/p2] (Z[Z/P2])" = Ar.
Hence the inclusion of I into Z[Z/p2] induces an injective homomorphism
A Z[Z/p2] I -> A Z[Z/p2] Z[Z/P2] = A.
The image of this injection is clearly the ideal, < p3, g - 1 - p >.
Next observe that there are integers u, v prime to p which satisfy
(1+p)p2-1=p3v
5.6.16
(1+p)p-1=p2u.
5.6 Cyclic groups
237
Therefore, in A, we have the following identities

p3v
= (l + p)p2 - 1
p2-1
1)p2-1-j)(P
(1: gj(P +
+ 1 - g)
j=0
p2-1
1)p2-1-j
(1: gj(P +
_ g1)(P + 1 - g)
j=0
p2-1
1)P2-1-j
P(E 9j[(P +
- l]/P)(P + 1 - g),
j=0
so that, since A is torsion-free,

5.6.17
1)P2-1-j
P2v = Q 02 gJ [(P +
- 1] /P)(P + 1 - g).
Since HCF(p2v, p3) = p2 we see that

A Z[Z/p2] I
=<
P2, g- 1- p> 4 A.
A similar argument shows that there is a canonical isomorphism of

the form
p2,g-1-p>=<p2,'p-1-p>
Z[Sp]
and that, in Z[ip],

p-2
p [(p + 1)P-1-j - I] /P
Pu =
(p +
j=o
where u is as in 5.6.16. Hence
Z[Sp]n <P2,g-1-p>=<p,"p-1-p>=<in
>.1
From 5.6.17 we have, in Z[2],

7P2-2
P2v = E p2 [(P +
1)P2-1-
_ l]/P (P + 1 - p2).
j=o
In this equation the expression in the first factor is divisible by p in

see this, expand (p + 1)P2- 1-j by the binomial theorem and
238
observe that every term in the resulting double sum is p-divisible with
the exception of the terms
P2-2
- E PZ(j + l).
i=o
However, in this sum each term with j = wp - 1 is clearly p-divisible,

while the remaining terms may be grouped into subsums of the form
I+...
(j+1)
(j + 1)Sp2(1 + bP + 2 + ... + p1)
(mod p)
(mod p).
This calculation shows that pv E< p2, p + 1 - p2 >
and we
obtain a canonical isomorphism of the form

Z[I;P2]OA
Now let us recall how the Mayer-Vietoris boundary homomorphism, 6

in 5.6.13, is defined. We begin by considering the fibre square of canonical
maps.
A Z[Z/P2] I
Z[`'P2] Z[Z/P2] I
Z[AP] Z[Z/P2] I
Z[AP]
Z[Z/P2] I
The previous discussion identifies this fibre square with a diagram
5.6 Cyclic groups
<p3,g-1-p>
239
<n1>
<1t2>
Z/p < 7C] >
in which the upper horizontal map and both vertical maps are the natural
isomorphisms
ones. We also have a Z[ip]- and
< ni >= Z[ip],

< 7C2 >=
given by dividing by it1 or it. Hence we may identify the fibre square
with one of the form
< p3, g - 1 - p >
Z[ip]
l1
Z[ip]
Z[ p2]
in which the unlabelled maps are the natural ones. However, there exists
a E Z[Fp]+ = Z/p[ni]/(n
-1)
such that, for all z E

5.6.18
(z) = a(z),
where 0 is as in 5.6.14. The map, 6 of 5.6.13, satisfies
240
8(a) = [I] E D(Z[Z/p2]).
5.6.19
Lemma 5.6.20 6(1 + rzi-2) = [< p3,g - 1- p >] E D(Z[Z/p2]).
Proof We will find a in 5.6.19 by evaluating the image of p + 1 - g in

Z[ip]. Via the clockwise route p + 1- g is mapped to p + nl E< 71 > or
to 1 + (p/iil) E Z[ip] and thence to
E Z/p[nl]/(ir 1).
However, 1 - P = jnl + ziii so that 5.6.21 becomes
1+
5.6.21
p-1
1 + fJ(.liti) = 1 + 7ri-2 E Z/P[itll/(ni-1).

i=2
On the other hand the image of p + 1- g in < n2 > is equal to p + ice,

which becomes 1 + (p/it2) in Z [gp2]. However,
N(1 + (P/72)) = (((P + 1)p - 1) + xi)nri
= 1 +Pu(P/ii),
so that 4(1 + (p/72)) = 1 and, from 5.6.18,

7ri-2
a = 1 +
E Z/P[iril/(ii-'),
as required.
Corollary 5.6.22 If p >- 5 is an odd prime then the element
[< p3, g - 1 - p >] E D(Z[Z/p2])

is non-trivial.
Proof The argument to follow was shown to me by Al Weiss.

By 5.6.13-5.6.20, we must show that
a = 1 + 7r
p-2
E Z/P[ni]/(ii-
is not the image of a unit from Z[ip]'. However, by Kummer's lemma

(Curtis & Reiner, 1987, p. 286), we may write any such unit in the form
p(3 for some integer, a, and such that (3 = fi, the complex conjugate of
/3. Suppose that
Lt =cpf3.
Therefore,
a/a = ( pQ)l (gyp a!') = b2a
5.7 Exercises
241
However, Tt1 = -gyp lnl, which implies that

CC = 1 -
PnP-2
=1-
1rP-2
E Z/P[x!]/(xP-1).
Therefore,
a/OC = (1 +
1GP-2)2
2TG1-2 E Z/P[7rll/(1Gp
=1+
In particular, since p >_ 5, a/a is congruent to one modulo 7T1 while Pa

is congruent to 1which implies that a is divisible by p. However,
this is impossible since it would imply that
1 =
Pa
= a/a = 1 + 2icP-2 E
Z/P[itll/(iti-1).
This contradiction completes the proof.
5.7 Exercises
5.7.1
Let Q8 denote the quaternion group of order eight:

Q8 = {x, Y I x2 = Y2, Y4 = 1, xyx-1 = Y-1
Let u2 = x + xy + y E Z2 [Qgb] `. Show, by using 4.2.4, that
DetQB(u) E '22(Z[Q8]) = T(Q8) = Z/2
is a generator, where u E U(Z[Qgb]) is equal to u2 at the prime two and

is equal to 1 at all other primes.
Let HZ denote the ring of integral quaternions. Suppose that X

is a k x k matrix with entries in HZ. Show that, in the notation of 5.2.11,
5.7.2
det(c(X)) is a positive integer.

(Hint: See Taylor (1984, p. 9)). To see that det(c(X)) is real, and hence
an integer, simply write X = Y + W j where Y, W E Mk(Z[i]) so that

c(X) is equal to
and observe that this is conjugated to its complex conjugate by the signed
permutation matrix
CO
1
-1
0
in M2k(Z[i]). To see the positivity observe that det(c(X)) is the norm of
242
X when X is a quaternion and use elementary row or column operations

to reduce the k x k case to the 1 x 1 case.)
5.7.3
In the proof of 5.2.33 prove directly that
X+(#(M+)) = det(c(V')) (mod 4).
Construct some more Z[Qs]-modules, similar to that given in

5.2.41, which are cohomologically trivial and whose order is a power of
two. Determine their classes in the class-group.
5.7.4
Generalise Theorem 5.3.3 in the following manner. Suppose that

there exists an extension of the form
5.7.5
{1}-) J->G-iZ/p-->{1},
where p is a prime. Suppose that J acts freely on spheres, S1 and S2, of
dimension n >- 1. Suppose, finally, that there exists a J-map, f : S1 -> S2
of degree k with HCF(k, #(G)) = 1.
Prove that, in the notation of Theorem 5.3.3,
S(kP) - >IndNG<g,(S(k)) = 0 E D(Z[G]).
g
(Oliver, 1978, Proposition 4) Prove that every chain complex in

5.3.24 has a resolution in the sense of 5.3.25.
5.7.6
Let p and q be odd, positive integers such that HCF(p, q) = 1.

Let AP and Aq denote elements of order p and q, respectively, in Z/pq.
5.7.7
Define Q(8, p, q) to be the semi-direct product

Q(8, p, q) = Q(8) oc Z/pq,
where
XAPX-1
YAPY-1
= Ap,
= AP
XAgX-1
1,
YAqY-1
= Aq 1,
= Aq.
Show that 4T (Q(8, p, q)) = 0. (When p and q are prime it is shown in

Bentzen & Madsen (1983) that 2T(Q(8,p,q)) = 0.)
5.7.8 (Research problem) Generalise the result of 5.3.3 to the case of a

group, G, which is given by an extension of the form
{1}->Z/n-'G->(Z/p)m
{1}.
5.7 Exercises
243
One would expect the relations to take the form of an alternating sum involving m+1 terms, rather like an Euler characteristic. Some modification
of the geometric proof of 5.3.3 might well yield such a generalisation.
5.7.9 (Research problem) Generalise the result of 5.3.3 to the case of a

group, G, which is given by an extension of the form
More precisely, attempt to prove that the relation
S(kP) - DndNG<g>(S(k)) = 0 E D(Z[G]).

g
holds for any integer k for which HCF(k, #(G)) = 1 and 0 = S(k) E
D(Z[H]).
It should be admitted that the conjectured relation is suggested merely
by the evidence of the metacyclic case (Theorem 5.3.3), the p-group case
(Theorem 4.4.20) and its aesthetically pleasant form.
5.7.10
Let R = Z2 or Z2 [l;P] with p an odd prime. Show that the units
of R[(Z/2)n] are generated by torsion units of the form w g with co

equal to a root of unity in R and g E Z/2 and by torsion-free units of
the form
1 + (v - 1)2-S(1 + Ll)(1 + L2)... (1 + LS),
where L1,. .. , LS are distinct characters of (Z/2)" and v E 1 + 25R. (Hint:
Let A be a ring containing no 2-torsion. Consider the injection
: A[Z/2] --) A A
given by u(a + by) = (a + b, a - b), where y generates Z/2, and use

induction on n. See also Sehgal (1978, p. 54).)
5.7.11
Verify the contents of Table 5.1.
5.7.12 (Research problem) Resolve the ambiguities in Table 5.1. (See also
5.5.24.)
Let p be an odd prime whose class number, hP = #(W27(Z[ P]))

is odd. Using a generalisation of 5.5.26 show that the exotic torsion in
T(Z/p x (Z/2)") is cyclic of order HCF(2n-1, p - 1).
5.7.13
244
Verify, from the definition of 5.6.2, the identities of the cohomology groups which are tabulated in Table 5.4.
5.7.14
Derive the exact sequence of 5.6.13, obtaining a description of

the homomorphism, S, from which to derive the formula of 5.6.19.
5.7.15
5.7.16
Use 5.6.22 to discover some more non-trivial classes among the

[Ma,,] E W2'(Z[Z/ps])
of 5.6.3.
In the proof of 5.3.29 the vanishing of the Swan group of a cyclic

group (see 4.2.48) was used to obtain an embedding of a mapping cone
(denoted by Z in 5.3.29) into an acyclic space with a free action on the
5.7.17
complement.
Prove the vanishing results of 4.2.48 geometrically, in the spirit of

Theorems 5.3.3 and 5.3.29.
6
Introduction
This chapter concerns complete, discrete valuation fields and the conductor invariants of their Galois representations.
In Section 1 we recall the ramification groups, ramification functions,
the Artin-Swan conductors and the Artin-Swan representations of a

complete, discrete valuation field. The Artin and Swan conductors are
integral invariants of Galois representations of complete, discrete valuation fields which are defined when the residue field extension is separable.
Their important properties are listed in this section - including invariance under inflation, Galois invariance and inductivity in dimension zero.
Each of these conductors is given by the Schur inner product with a corresponding Galois representation whose character is defined in terms of
the ramification functions.
In addition, if G(L/K) denotes the Galois group and g E G(L/K )

the ramification functions at gP-' and at gP' were conjectured by A.
Grothendieck to be congruent modulo p". In 6.1.34 we give a proof of
these conjectures which is due to S. Sen. In 6.4.2 one sees that these
congruences, together with the determinantal congruences of Chapter 4,
restrict the possible identities of the ramification subgroups of G(L/K).
Section 1 concludes with the derivation of a formula relating the Artin
conductor of V to that of pP(V) when p is the residue characteristic and
V is uniformly ramified.
J-P. Serre (1960) asked whether it is possible to extend the theory
of the Artin conductor (or, equivalently, the Swan conductor) to the

case when the residue field extension is not necessarily separable. This
question has been reiterated by several authors, including K. Kato, who
gave a definition for a suitable Swan conductor in the case of a one245
246
dimensional representation. In Section 2 we introduce Kato's abelian

conductor and establish the formulae which relate its value on a onedimensional character to the value on the restriction of the character
to the inertia group or to the first wild ramification group. We give a
number of calculations of the Kato-Swan conductor when the Galois
group is isomorphic to Z/p x Z/p and p is the residue characteristic.
These examples suffice to show that one cannot construct a non-abelian
conductor which generalises all the others, is invariant under inflation
and is inductive in dimension zero.
In Section 3 we use Explicit Brauer Induction to construct an integervalued conductor function which agrees with those of Swan and Kato,
when the latter are defined. In addition we show that the obstruction
to inductivity in dimension zero for our generalised Swan conductor lies
in the cases where the Galois group is Z/p" x Z/p or Z/p" and p is
the residue characteristic. This result 6.3.20 is proved by comparing our
conductor with the rational-valued, inductive conductor which one may
define by applying our construction to the Explicit Brauer Induction
homomorphism, dG, of Chapter 2, Section 4.
Section 4 consists of a collection of exercises concerning conductors.
6.1 Ramification groups and functions
Let L/K be a finite Galois extension of complete, discrete valuation

fields. Let (9K, mK and (9K = (9K - mK denote, respectively, the valuation
ring of K, the maximal ideal of (9K and the multiplicative group of units
of (9K. Let G(L/K) denote the Galois group and let xL E (9L generate
mL 4 (9L. Let L = (9L/ML and K = (9K/mK denote the residue fields of
L and K respectively.
We will refer to the case when L/K is separable as the `classical case'.
For the moment we will make no such separability assumption.
Let vL denote the L-adic valuation, vL : L* ---) Z. Hence vL(xL) = 1.
Following Serre (1979, chapters IV and V), we may define normal

subgroups
6.1.1
...,< Gi 4 ...,< Go ,4 G_1 = G(L/K)
by the condition that

6.1.2
g E Gi (
) vL(g(a) - a) > i + 1 for all a E (9L.
Following Serre (1979, p. 62) and Kato (1989, p. 121) define ramifica-
247
tion functions iG(L/K) and SG(L/K) on G(L/K) by the formulae (note that
our SG(L/K) is -SG(L/K) of Kato, 1989).
6.1.3
iG(L/K)(g)
= infaEOL vL(a - g(a))
if g
1,
if g
1,
g(x)x-1)
sG(L/K)(g) =
iG(L/K)(1)
= - E1ggEG(L/K)1G(L/K)(g),
SG(L/K)(1)
= - E1gEG(L/K)SG(L/K)(g)
In the classical case (when L/K is separable)

6.1.4
iG(L/K)(g) - 1
if 1 * g E Go,
SG(L/K)(g) =
I 1G(L/K)(g)
if 1 * g V Go.
In fact, in the classical case, 9L is generated as an (9K-algebra by one

element, x, and the infima in 6.1.3 are attained at this element (Serre,
1979, chapter II, section 6, proposition 12 and p. 61). In the general case,
if Go = G(L/M) 1 G(L/K) then
G(M/K) = G(L/K)/Go = G(M/K),

where M/K is the maximal separable subextension of L/K and
6.1.5
where
f7IK = IM : K] = [G(L/K) : Go],

denotes the separable residue degree of L/K.
6.1.6 The Artin and Swan conductors
We shall recall briefly the definitions and properties of the classical Artin
and Swan representations and their related conductor homomorphisms.
For the remainder of this section we shall assume that L/K is a separable
extension (the `classical' case). Under these hypotheses one has complex
representations (not merely virtual representations)
6.1.7
AG(L/K), S WG(L/K) E R(G(L/K )),
248
where R(G) denotes the complex representation ring of G (Serre, 1977,

p. 68).
If V E R(G) let Xv denote its character function. The characters of

AG(L/K) and SWG(L/K) are given, for g E G(L/K), by the formulae
XAG(L,K)(g)
= -fL/K
iG(L/K)(9)
and
6.1.8
XSWw(L/K)(g) = -fL/K SG(L/K)(g),
where fL/K denotes the residue degree of L/K, fL/K = [L : K].

The representations, AG and S WG, are called the Artin representation
and the Swan representation, respectively, and, by 6.1.4, they are related
by the formula (Serre, 1977, p. 160; Kato, 1989, p. 121)
SWG =AG+IndGo(1)-Ind{1}(1),
6.1.9
where I ndG (V) denotes the induced representation of V and 1 E R(G) is

the class of the one-dimensional trivial representation.
If Gi i G(L/K) (Serre, 1979, p. 62; 6.1.2) is the ith ramification group
(with G_1 = G(L/K)) then (Serre, 1979, p. 100)
Eo[Go : Gi]-hIndG(L/K)(Ind{i}(
6.1.10
E
R(G(L/K)).
Notice that Gi = {1} for large i so that 6.1.10 is only a finite sum.
From Serre (1979, p. 101), if G(L/K') < G(L/K) and SKI/K is the discriminant of K'/K then the restriction to G(L/K') is given, in R(G(L/K')),
by
6.1.11
Res
c(L/xG(L/K)
(AG(L/K)) = VK(bK,/K)Ind{l}
{1}
(1) + fK'/KAG(L/K'),
where VK is the valuation on K* and fK'/K denotes the residue degree,

[K' : K].
Remark 6.1.12 Note that, under our current assumptions of separability,
fK /x is also equal to the separable residue degree, fK IK = [K' : K]S

(Lang, 1984, p. 282). In generalising the conductor homomorphism to the
case of non-separable residue field extensions (see 6.3.3) it will become
necessary to choose between fx /x and fZ "P'IK' We shall choose the latter,
where necessary, based on examples (see 6.2.17 and 6.3.20).
249
Suppose that N < G. Define

6.1.13
: R(G) -* R(G/N)
by
q(V) = C[G/N] C[G] V = VN,
the N-fixed points of V. If K < K' < L and K'/K is Galois then (Serre,
1979, p. 101)
6.1.14
q(AG(L/K)) = AG(K'/K)
E R(G(K'/K)).
The Artin conductor is the homomorphism
fK : R(G(L/K)) --> Z,
6.1.15
given by fK(X) =< AG(L/K),X >G(L/K), where < V, W >G is the Schur
inner-product of 1.2.7 (V, W E R(G)).
The Swan conductor
SWK : R(G(L/K)) --> Z
6.1.16
is defined by swK(X) =< SWG(L/K), X >G(L/K) and, by 6.1.9,

6.1.17
SWK(X) = fK(X)+ < 1, X >Go -dim(X).
If N < G then the inflation map

6.1.18
Info/N : R(G/N) ---> R(G)
is adjoint to a of 6.1.13 in the sense that
< V,Infc/N(W) >G=< q(V), W >G/N
for V E R(G) and W E R(G/N). Therefore, if K < K' < L with K'/K
Galois then, by 6.1.14,
6.1.19
fK(I nfc(K'/K)(X)) = fK(X) E Z
for all X E R(G(K'/K)).
Since IndGI is adjoint to Res' we find for K < K' < L and X E
R(G(L/K')), by 6.1.11,
6.1.20
G
)(X))
fK(IndG(L/K'L/K)
= vK(bK'/K)dim(X) +fK'/KfK'(X)
250
By 6.1.17 we easily find, when X is one-dimensional, that

swK(In
6.1.21
fG(G(K/K'IK) (X)) = swK,(X) E Z
and that
6.1.22
if dim(X) = 1 and X = 1 on Go
swK (x)
and
fK(X) =
1
swK(X) + 1,
if dim(X) = 1, otherwise.
Hence, in particular, the trivial character has trivial Artin and Swan
conductors.
If K < K' < L then

fK(IndG(L/xG(L/K)) (1)) = VK(aK'lK),
6.1.23
as a special case of 6.1.20. Also 6.1.20 and 6.1.23 combine to yield

6.1.24
fK(I nd
(X - dim(X))) = (fK'/K) fK'(X - dim(X))
and (see 6.4.3)

SWK(IndG(L/K)
G(L/K')(X
- dim(x))) _ (fK'/K) swK,(X - dim(X)).
For any x E R(G(L/K))

fK(X) =
6.1.25
SWK(X) =
x))
where Go = G(L/M). This follows from Frobenius reciprocity (1.2.39)

since, by 6.1.10,
6.1.26
AG(L/K) = IndGoL/K)(AG(L/mr))
Suppose that the Galois group, G = G(L/K ), is given. The

disposition of the ramification groups, {Gi}, cannot be assigned arbitrarily. Similarly there are restrictions on the possible representations in
6.1.27
251
R(G) that can occur as an Artin representation for some choice of L/K
with G = G(L/K ). For example, in the classical case some restrictions
are given by congruences between values of the ramification functions.
These congruences were conjectured by A. Grothendieck and proved by
S. Sen (1969, Theorem 1.1). The majority of the rest of this section
will be devoted to the study of these congruences. Note that, by 6.1.26,
there would be no loss of generality in studying the totally ramified case
(G(L/K) = Go) if we were interested in restrictions upon AG(L/K)
Let L be a local field with residue field, L, of characteristic p > 0.

Hence L is either a finite extension of the p-adic field, Q,, or a onedimensional function field of power series over L. Hence, in the function
field case the constant functions yield a multiplicative map
f :L->(9L,
6.1.28
such that the reduction of f (modulo mL) yields the identity map on
L. Similarly, when L is of characteristic zero and L is the field with q
elements we have the Teichmuller map (Iwasawa, 1986, lemma 2.2, p. 19)
which is defined by
f(x mod mL) = lim(x' ).

n- oc
Lemma 6.1.29 Let L/K be a Galois extension of local fields. There exists
a subset, C < (9L, such that
(i) C is a complete set of representatives for (9L/mL = L
and
(ii) if c E C then g(c) = c for all g E Go i G(L/K).
Proof Take C = f (L) where f is an in 6.1.28.

Lemma 6.1.30 Let L/K be as in 6.1.27 and let g E Go i G(L/K ).
m = pes with HCF(p,s) = 1 then
If
vL(gm(nL) - ltL) = vL(gP'(xL) - ltL)
(Note that, in the notation of 6.1.3, vL(gm(1L) - IEL) _ iG(L/K)(gm) = 1 +

sG(L/K)(gm) when gm * 1.)
Proof Set h = gP` so that
gm -1 = (h - 1)(1 + h +... + hs-1)
252
Since h E Go, h`(7tL)
7rL (mod mL) and, for some a E (9L,

vL(7rL + h(7IL) + ... + hs-1(nL))
= VL(S7rL + a7rL)
= VL(ItL)
since HCF(p, s) = 1. Both 7tL and S7rL +a7tL are choices for a uniformiser
(the prime element) of (9L so that the infimum, iG(LIK)(-) of 6.1.3 is

attained at either of these elements. Therefore
vL((gm - 1)7 L)
= vL((h - 1)(s7<L + a7tL))
= vL((h - 1)70.
Lemma 6.1.31 Let g E Go i G(L/K ). For each integer, m, there exists

xm c L such that
(i) vL(xm) = m and
(ii) vL(g(xm) - xm) = m + vL(gm(1L)7TL 1 -1).
Proof When m = 0 choose xo to be any non-zero element of C in 6.1.29.

(x_m)-1 since
If m < 0, we may set xm =
vL(xm) = vL((x_m)-') = -vL(x_m) = -(-m) = m
and
vL(g(xm) - xm) = vL((x-m -
g(x-m))/x-mg(x-m))
= vL(x-m - g(x-m)) - 2VL(x-m)
= -m + vL(g-m(7rL)7rj' - 1) + 2m
= m + vL(gm(7rL)7rL- 1 - 1).
When m > 0, set

m-1
X. = 11 g`(7rL)
1=0
Clearly, vL(xm) _ >I1 vL(g`(lrL)) = m and g(xm)xml = gm(7rL)7rL' so

that vL(g(xm)xm' - 1) = vL(gtl1(7TJ7EL' - 1), as required.
253
If x E L then
Corollary 6.1.32
00
x=
wm
m=ve(x)
where wm satisfies 6.1.31(i) and (ii).
Proof We may write x in the form of a convergent series

00
X=
Cmxm,
m=vL(x)
where cj E C in 6.1.29 and xi is as in 6.1.31. However, since cj E (9L and

is fixed by g, wi = c;xi also satisfies 6.1.31(i) and (ii).
Proposition 6.1.33 In 6.1.27, let g E Go Q G(L/K) satisfy gp-` * 1. Hence,
in the notation of 6.1.3, sG(L/K)(g) = vL(g(nL)lri i - 1). Suppose that, for
all0<j<n,
sG(L/K)(g) = SG(L/K)(g") (mod p'),
then the integers

,
{m + sG(L/K)(gm) I V (m) < n} U {sG(L/K)(gp" )}
are all distinct. Here vp denotes the p-adic valuation, as in 6.1.30.
Proof Suppose that m + sG(L/K)(gm) = j + SG(L/K)(9i) If vp(m) = vp(j)

then, by 6.1.30, sG(L/K)(gm) = sG(L/K)(g') and therefore m = j. On the
other hand, vp(m-j) = min(vp(m),vp(j)) while the congruences imply that
vp(sG(L/K)(g3) - sG(L/K)(gm)) > min(vp(m),vp(j))
so that m - j
sG(L/K)(gJ) - sG(L/K)(gm)
Finally, if vp(m) < n then

i)
vp(SG(L/K)(g"
so that
- SG(L/K)(gm)) > vp(m),
sG(L/K)(gpn
) * m + sG(L/K)(gm).
We are now ready to establish the congruences of theorem 1.1 of Sen

(1969).
Let L/K be a Galois extension of local fields with

char(t) = p. Let g c Go 1 G(L/K) satisfy gP"
1, then, in the notaTheorem 6.1.34
tion of 6.1.3,
SG(L/K)(gp'
)=
(mod p").
254
Proof By induction on n (replacing g by gP) we may assume that

sG(L/K)(gpn ')
= SG(L/K)(gPn) (mod pn-1).
Set sj = sG(L/K)(gP') then we have to show that vp(sn - s,_1) > n. If this is
not so then vp(Sn - Sn_1) = n -1. Applying 6.1.31 to gP with m = Sn-1- Sn
we obtain z E L such that vL(z) = Sn_1 - sn and
VL(gP(Z) - Z)
= Sn_1 - Sn + SG(L/K)(gpm)
= Sn_1 - Sn + SG(L/K)(gP"),
by 6.1.30. Hence vL(gP(z) - z) = Sn-1 - Sn + Sn = Sn_1.
Set x = (1 + g +... + gP-1)z. Since 1 + g +... + gP-1 - (g -1)P-1 (mod

p) we see that
VL(X) ? vL((g - 1)P-1z) > vL(z) = Sn_1 - Sn
and
6.1.35
vL((g - 1)x) = vL((gP - 1)z) = Sn-1
Apply 6.1.32 to obtain an expansion x = .mVL(X) wm and set u =

(g - 1)x = >m-VL(X) um, where um = (g - 1)wm.
If vp(m) >- n then
VL(um) > VL(Wm) + Sn, by 6.1.30. Also, vL(x) < vL(wm) so that
VL(um) > VL(X) + Sn > VL(Z) + Sn = Sn-1 = VL(u),
by 6.1.35. Hence
VL(U) = VL(>vp(m)<n Um + Evp(m)>n um)
= VL(>vp(m)<n um)
However, vp(m) < n implies, by 6.1.33 and the induction hypothesis,

that the integers vL(um) = m + SG(L/K)(gm) are distinct from one another
and from sn_1 = vL(u). Therefore
vL(u) = minvp(m)<n(VL(um))
vL(u)
This contradiction shows that vp(sn-Sn_1) > n and completes the proof
of Theorem 6.1.34.
Example 6.1.36
Suppose that char(L) = 2 and G(L/K) = Q16, the
generalised quaternion group of order 16 (see 1.3.7). Hence Q16 = {x, y I
x4 = y2 y4 = 1, xyx = y}. The congruences of Theorem 6.1.34 show that,
255
for example, if x2 E G1 then x4 E G3, while if x E Go then x2 E G2 and

x4 E G6.
In Serre (1960, section 4, p. 413) (see also Snaith, 1989b, pp. 260
and 261) a 2-adic example is given for which G(L/Q2) = Q16 and
iG(L/Q2)(x2) = 2, iG(L/Q2)(x4) = 4 which shows that x
Go in this example.
6.1.37 Conductors and Adams operations
The congruences of Theorem 6.1.34 suggest congruences between the

character values of AG(L/K) and WP(AG(L/K)), where WP is the Adams
operation of 4.1.2. However, the anomalous definition of iG(L/K)(1) seems
to obscure the behaviour of such congruences. In 6.4.2 an example is

given of the sort of determinantal congruence restrictions which may be
derived from 6.1.34.
For similar reasons one cannot expect very good properties for the
Artin conductor of WP(X). Nevertheless, one may sometimes evaluate
fK(WP(X)) under conditions which are related to the Explicit Brauer

Induction formula, aG(L/K)(X) In order to discuss this topic we shall need
to recapitulate the upper numbering system for ramification groups.
Let L/K be a finite Galois extension of local fields (i.e. the classical
case). If u E R is a real number and u > -1 set G = G1, where i is the
smallest integer such that i >- u. Thus, by 6.1.2,
g E G 4=t> iG(L/K)(g) ? U + 1.
Set
OL/K(u) = b(u) = LU
[G_1
[Go G]When
-1 < t < 0 we adopt the convention that [Go : Gj equals

: Go]-' for t = -1 and equals 1 = [Go : Go]-' for -1 < t < 0.
Hence, for -1 < u < 0, 0(u) = u.

If m is a positive integer and m < u < m + 1, then
O(u) = (90)-' (91 + 92 + ... + (u - m)gm+1),
where $, = #(G1). Hence ca(m) + 1 = (go)_'(Em 0$1). Since 0 is continuous, increasing and piece-wise linear we may define ip to be its inverse
function. The left and right derivatives of 0 and W are easily computed
(Serre, 1979, p. 73). Define
G' = G,(ll)
or, equivalently, GO(u) = G.
Hence G-1 = G(L/K), G = Go and G = {1} for v >> 0. When
256
H i G(L/K) then G(L/K)/H = G(F/K), F = LH and (G(L/K)/H) _

G(L/K )H/H.
Suppose now that L/K is abelian and that
X : G(L/K) -* C*
is an injective, one-dimensional representation. By local class field theory
we have the Artin reciprocity map
K"
) G(K`/K)ab -> G(L/K).
Here, as in 4.2.11, K` is an algebraic closure of K. This map annihilates

norms from L* and factors to yield
w : K*/NL/K(L`) ---> G(L/K)
and (Serre, 1979, corollary 3, p. 228) transforms the filtration
{UKINL/K(ULv(n))}
into the filtration, {G"(L/K)}, where
UL={uEGLJu==-1 (mod mL)}
(Serre, 1979, p. 66). The largest integer, c, such that GJL/K) * {1} is
related to fK(X) by Serre (1979, corollary 2, p. 228)
fK(X) = O(c) + 1.
6.1.38
In other words, (fK(X) - 1) is the least integer such that X is non-trivial

on Ge = GO(B) = w(UK(e)), or fK(X) is the least integer such that X
is trivial on UKK(x)), when considered (via reciprocity) as the character
Xw:K'-->C'.
Let eL/K be the ramification index of L/K so that [L:K] = fL/KeL/K.
Now suppose that K is a local field over Q, and that fK(X)-1-eK/QP >
eK/QP/(p - 1). In this case the pth power map induces isomorphisms
,fK
(x)-eK /QP
UK
,fK (x)-eK/(b, -1
UK
LL
N
UKK(X)-1
257
by Serre (1979, proposition 9, p. 212). This means that p"(), which is

fK(x)-e 11 -1
just the p-th power of x, is non-trivial on UK
and trivial on
fK (x)-eKIQ1
UK
Therefore we have proved that
Lemma 6.1.39 If fK(x) > 1 + eK/Q,(p + 1)/p, then

fK(WP(x)) = fK(X) - eK/Qn.
Let G(L/K) be the Galois group of a finite extension

of p-adic local fields. Let v : G(L/K) - GL,,,(C) be a representation.
Then v will be called uniformly ramified if
Definition 6.1.40
aG(LIK)(v) _ Y, aa(G(L/Fa), xa)G(LIK) E R+(G(L/K))

a
and, for each a, fF,(xa) > 1 + eF/Qp(p + 1)/p.

Theorem 6.1.41 If v is uniformly ramified, in the sense of 6.1.40, then
fK(1P"(v)) = fK(v) - eK/QP . dim(v).
Proof By 4.1.6, in R(G(L/K)), we have

V=
G(L/K)
E a I ndG(L/F,)(xa)
a
and
WPM _
aaI nd
G(L/F,G(L/K))(xa)
Hence, by 6.1.20 and Lemma 6.1.39,
fK(ipP(v)) = Ea aa(vK(5FIK) +fF,/KfF,(xa))

= Ea aa(vK(5F,/K) +fFa/K(fF,(Xa) - eF,IQ,))
= fK(v) - Ea as fF/K - eF,IQP.

Now [Fa : Q,] = [Fa : K] [K : Q,] and
fF,IQ,, =[P
Fp]=[Fa K] [K :FP]=fF,/KfK/QP'
while eL/KfL/K = [L : K], so that

fF,/K eF,IQ, = fF,IK , eF,1K eK/Q, = [Fa : K]eK/Qn.
258
Therefore
fK(v)-fK(wp(v)) =eK/QP(Eaa,[F.:K])
eK /QP
. dim(v),
since, by 2.2.42,
dim(v)
XOC))
_Eaaa[F., :K].
Now let us turn to the case of 1pq when HCF(p, q) = 1. Let v be

as in 6.1.40 and set
6.1.42
d _ {a I xa(Go(L/FF)) * I,X,(Gi(L/FF)) =1= xa(Go(L/Fa))}.

We will now calculate fK(p (v)). By Serre (1979, proposition 1, p. 100; see
also 6.1.25) we may assume that L/K is totally ramified. The following
result is left to the reader as exercise 6.4.1.
Theorem 6.1.43 Assume that L/K is totally ramified. With the notation
of 6.1.42
fK(tp"(v)) = fK(v) - T. a,
aE&V
6.2 Kato's abelian conductor
In this section we shall recall a generalisation, due to K. Kato (1989) of

the Swan conductor. In order to state this definition, which is given in
terms of Galois cohomology, we must recall some Galois modules.
6.2.1
Let n be a non-zero integer and let r E Z. Let K be a complete,
discrete valuation field and suppose that KC is an algebraic closure of K.
Let Z/n denote the fIK-module given by the nth roots of unity in K`,
where f1K = Gal(K`/K) is the absolute Galois group of K. Denote by
Z/n(r) the rth Tate twist of Z/n. Hence Z/n(0) is the module with the
trivial action and for r > 0, Z/n(r) = (Z/n)'. If char(K) = p > 0 we
shall also need the module, WSS1K,tog [-r], given by the logarithmic part
of the de Rham-Witt complex, WflK (Illusie, 1979, 1 section 5.7, p. 596).
259
Define an S2K-module, Z/n(r), in the following manner.

6.2.2
Z/n(r) WSS2K,,,,g [-r],
if char(K) = p, n = psm,
s >- 0 and HCF(m,p) = 1,
Z/n(r)
otherwise.
Z/n(r) =
We will write Hn (K) for the Galois cohomology group, Hq(K ; Z/n(q 1)) and Hq(K) for limy Hn(K). For example,
6.2.3
H1(K) = Hom,ontspK,Q/Z)
H2(K) = Br(K),
where Br(K) denotes the Brauer group of K.
Suppose that L/K is a finite

6.2.4
x : G(L/K) ---> C. is a one-dimensional
Galois extension and that
representation of the Galois

group, G(L/K). Identifying the roots of unity with Q/Z we may interpret x as a continuous homomorphism, x : f1K -* Q/Z, and hence
obtain
6.2.5
x E H1(K).
If M/K is any field extension (not necessarily finite or Galois) denote

by
6.2.6
XM E H'(M)
the image of 6.2.5 under the natural map.

If it is the chosen uniformiser in C9K and T is an indeterminate then,
for n >- 0,
6.2.7
1 + ,n+ 'T G (WK [T] (.))(h))-,
the unit group of the henselianisation, with respect to (n), of the localisation, (9K [T](,,). If M denotes the field of fractions of C9K [T] (h) this implies
(Kato, 1989, p. 103) that 1 + nn+1 T represents a class in the cohomology

group,
He (SpeC((9K [T](nj); Z/s(1))
and hence we obtain
260
6.2.8
1 + itn+' T E lim_ H'(M; Z/s(l)).

S
We may form the cup-product

{XM, 1 + ,n+1 T } E H2(M).
6.2.9
Let K be a complete, discrete valuation field and let

) C* denote a one-dimensional Galois representation, as
in 6.2.4. The Kato-Swan conductor of x,
Definition 6.2.10
x : G(L/K)
6.2.11
SWK(X) E Z,
is defined to be the least integer, n >- 0, such that
0={xM,1+7Cn+'T}
in 6.2.9.
An important indication of the `correctness' of this abelian generalisation of the Swan conductor is the following formula of Kato, which
often gives swK in terms of the ramification function, SG(L/K)
Theorem 6.2.12 Let L/K be a finite Galois extension of complete, discrete

valuation fields. Let X : G(L/K) -* C` be a one-dimensional representation and let 5G(L/K) denote the ramification function of 6.1.3.
If either L/K is separable or eL/K = 1 and L is generated over K by

one element, then
swK(X) _ -(eL/K)-i
sG(L/K)(g)X(g)gEG(L/K)
Proof See Kato (1989, proposition 6.8, p. 12).

Corollary 6.2.13 In the classical case the abelian conductor of 6.2.10 coincides with the Swan conductor of 6.1.16.
Proof Let x be a one-dimensional representation of G(L/K), as in 6.2.10.

In the classical case the residue field extension is separable so that 6.2.12
yields
swK(X) = -(eL/K)' EgEG(L/K) SG(L/K)(g)X(g)

= 7r(G(L/K))-' >geG(L/K)(-fL/KSG(L/K)(g))X(g)
=< SWG(L/K), X >,
261
which is the classical definition of the Swan conductor given in 6.1.16.

For completeness I will sketch a cohomological proof of 6.2.13 which
proceeds directly from the definition of 6.2.10.
Proposition 6.2.14 Let K be a classical local field and suppose that
dim(x) = 1, then the conductors, swK (x), of 6.1.16 and 6.2.10 coincide.
Proof We shall treat only the case when x is non-trivial on Go, since the
remaining case is simpler. In this case, by 6.1.22, the Artin and Swan
conductors of x differ by one.
If {xM,1 + ir"+1T} = 0 in 6.2.9 then, by Theorem 6.3 of Kato (1989),
{x,1 +7r"+l&K} < H2(K;Q/Z(1)) - Q/Z

vanishes.
Now suppose that {x, l + IE"+1 OK } = 0 and that {XM,1 + 7rm+1 T } = 0
with m > n. We shall show that {XM,1 + nmT} = 0 and hence, by

induction, that {XM, 1 + 7T"+1 T j = 0.
Since m > 0, we know that {XM, 1 + irtT} is p-torsion, by Kato

(1989, section 5.5(i)). Now let K denote the residue field of K. There
is a subgroup, VP ((9K [T]), of HP(M) which is defined in Kato (1989,
definition 1.7.1). In addition there is an injective homomorphism, A, and
a commutative diagram of the following form (Kato, 1989, section 5.6.1).
6.2.15
Hp(K[T])BHp(K[T])
HQ(K) Hp(K)
rz
Vp((9K[T])
HP(M)
7r
V ((9K)
Hp (K)
In 6.2.15 HP(F[T]) is the qth etale cohomology group of Spec(F[T])

with coefficients in Z/p(q - 1). The vertical maps are given by the
262
specialisation sending T to 1. Hence a is an isomorphism. The horizontal

maps, An, An and j are injective (Kato, 1989, sections 5.6 and 6.3). In
addition, by Kato (1989, section 5.1), there exists (a, b) such that
2n(a, b) = {XM,1 + it'T }.
However,
4n(a, b)
= I3({XM, l + irmT})
_ {X,1 + xcm}
= 0,
so that (a, b) = 0 and {XM, l + it' T } = 0, as required.

Finally, from abelian class field theory, we may identify the pairing
H'(K;Q/Z(0)) H1(K) -> H2(K) = Q/Z

with the evaluation pairing
Homconts(K;Q/Z) limK*/(K")"
Q/Z.
Therefore the least n for which {X,1 + it"+1(9x } = 0 is the least n for
C' vanishes on UK 1 = 1 + icn+1 OK. By Serre (1979,
which X : K*
p. 102, proposition 5) the Artin conductor is also equal to the least

integer, n + 1, such that X vanishes on UK 1. This completes the proof of
6.2.14.
Remark 6.2.16 In the next section we shall construct a generalisation

of the Swan conductor by applying Explicit Brauer Induction to extend
the abelian conductor of 6.2.10. For our generalised Swan conductor we
will be able establish a number of properties which are possessed by the
classical Swan conductor (see 6.3.6 and 6.3.20).
However, we will find that the inductivity in dimension zero (see

6.1.24) is a property which cannot hold in general for any generalisation
of the Swan conductor which coincides with 6.2.10 in the abelian case.
In order to see this we shall pause to examine several examples in

which G(L/K) = Z/p x Z/p. Local Galois extensions whose group is
elementary abelian may all be constructed by the Artin-Schreier theory
of Serre (1979, p. 154). Further details on the non-existence of suitable
generalisations of the Swan conductor are to be found in Boltje, Cram
& Snaith (1993, section 5).
For these examples, which were shown to me by G-M. Cram, we shall
calculate the behaviour of any Swan conductor which is invariant under
263
inflation (as in 6.1.21) and coincides with the Swan conductor of 6.2.10 in
the abelian case. In examples 6.2.18-6.2.20 we shall see that such a Swan
conductor cannot be inductive in dimension zero, in general. Examples
6.2.19 and 6.2.20 have a subextension in common (Np/K in 6.2.19 is

Rp/K in 6.2.20). This feature makes 6.2.19 and 6.2.20 into particularly
good illustrations of the sort of relations between values of the Swan
conductor which would be implied by inductivity in dimension zero.
Example 6.2.17
Let Fp denote the finite field of order p.
Let K = Fp((T))((it)), (9K = Fp((T))[[rc]], where it is the prime element. To construct Galois extensions of K with group Z/p we use a
modification of Artin-Schreier theory.
Consider the polynomial
fa(X) =Xp-TCp-1X -a E K[X].
If A is a root of fa(X) then the other roots are A + n where
E Fp.
Hence either MX) splits completely in K or K (A) is a Galois extension

of degree p. Using Galois cohomology one can show that every Galois
extension of degree p over K may be obtained in this manner (see, for
example, Serre, 1979, Chapter X, section 3, p. 154). Set Mp = K(H)
Since f,,(X) is an Eisenstein polynomial
where n is a root of
(Lang, 1984, p. 200) Mp/K is totally ramified - that is, eMP/K = p.
Hence (9M,, = OK[H], by Serre (1979, chapter I, section 6, proposition
17). Set Mo = K(A) where A is a root of fa(X) and a E (9K is a unit such
that a (modulo n) is not a pth power (modulo H). For example, one
may take a = T. Hence fa(X) - XP - a (mod 7C), which is an irreducible
polynomial in the residue field, Fp((T )). Hence (9M0 = (9K [A] (see Serre,
1979, chapter I, section 6, proposition 15) and Mo/K is unramified with
purely inseparable residue field extension, Mo = Fp(T,A), where A is the
image of A. For example, if a = T then Mo = Fp(J' T).
Set L = MoMp. This is a Z/p x Z/p-extension of K, since f,(X)

remains irreducible over Mo because it is a prime in Mo. We have
(9L = (9,no [II] = (9K [A, H]. The Galois group, G(L/K ), has generators
co, a1 with
CO(II) = rI,
ao(A) = A + 7r,
a1(H) = IZ + 1C,
a1(A) = A.
Denote the fixed field of U; =< opal > by M, for i = 0,1, ... , p - 1 so
264
that we have the following diagram of fields:
Mo
M2
Mi
...
Mp
One finds that Mi = K(A1) with Ai = A - ill, since Ai is a root of
fa, =Xp -
np-iX
- (a-iTt).
Set Up =< co >. Since ai = a - i7t - a ( mod 7t) we see that Mo = Mi
Mp-i
Now we calculate the functions, iu; and su,, of 6.1.3. We have Uj =
G(L/Mi) and (9L = CM, [II] for i = 0,1, ... , p - 1 so that, for 1 * a E Ui,
iu,(a) = infxEOL(vL(x - Q(x)))
= v(H - v(n))
= VL(Tt)
=p
and
su,(u)
= infxEL'(vL(1 - Q(x)x I))

= VL(1 - 7(H)II-1)
=p-1.
265
Table 6.1. Field extensions (Example 6.2.17)

sep
E/F
eE/F
fE/F
f E/F
p
p
L/M,
i=0,...,p-1
M,/K
i=0'...,p-1
L/Mp
Mp/K
Also, Up = G(L/Mp) and (9L = (9m, [A] so that, for 1 * a E Up,

iuu(a)
= v(A - a(A))
=p
and sup(a) = p.
Next we ha ve G(L/K)/Ui = G(Mi/K) and

i = 0,1, ... , p - 1 so that, when a
iG(Mi/K)(a)
(9 M; = (9K [Ai] for
Ui,
= VM,(AI - a(Ai))
= VM;(7t)
=1
and SG(M;/K)(a) = 1.
Finally, for G(Mp/K), (9Mp = (9K [II] and, when a 0 Up,

iG(MM/K)(a) = vMQ(n - a(n))
= VMP(Tt)
=p
and SG(MP/K)(a) = p - 1.
The table of ramification indices, residue degrees and separable residue

degrees of these field extensions are given in Table 6.1
Let 0 : G(L/Mo)
C* be given by 0(al) = p = exp(2iri/p). Lift
0 to 0 : G(L/K) -+ C' given by 6(aoal) = 0(al) = P. We have

G(L/K) = Uo x Up, where Up =< co >. Define ip : G(L/K) --> C. by
266
ip(o cr) _ p so that

P-1
G(L/K)
IndG(L/ti1o)(B - 1) =
(Btp
tV )
i=o
For i = 1, ... , p - 1 we have Ker(lp`) = Uo and

Ker(Oip1)
= {Qpob I 1 = r;p+ia}
= Uj,
where ij = -1 (mod p). Also Ker(b) = Up. Since the ramification

indices, residue degrees and separable residue degrees are the same for
U0,..., Up_,, we find that
G(L/MG(L/K)o)(0
swK(Ind
-1)) = SWK(b).
Also, since fMp/K = 1 and eMp/K = p,
C')
SWK(b : G(Mp/K)
= -P 1 EgEG(Mp/K) SG(MM/K)(g)e(g)
= P 1 {-(P - 1)(Eg&1 8(g)) + (p - 1)2}
= p {(P - 1) + (P - 1)2}
=p-1.
Also, since f L/Mo = 1 and eL/Mo = p,
SWMO(O : G(L/Mo) -' C')
= pl {-(P - 1)(>g#10(g)) + (P - 1)2}
=PTherefore, since fMo/K =

swK(Indc
1)) = fMp/KSw,yo(B - 1).
267
As a second example we take the same field extensions

)
and let 01 : G(L/Mp)
C* be given by 01(ao) = gp. Lift 01 to
Example 6.2.18
Bl : G(L/K) ---> C' given by el(U0 1) = P. Define W1 : G(L/K) -+ C'

by W1(6061 _ gyp. Hence, for i = 1,...,p - 1 we have Ker(ip) = Up and
Ker(O1Wi) = {f0
1 = P+ib}
= {(60a1)b}
= U;.
We have
1))
_ EP o SWK(81W1) - I? o SWK(Wi)
= PSWK (e1) - (p - 1)swK (W1)
Since eMu/K = 1,
) C*)
SWK(B1 : G(Mo/K)
_ {(-1)(Egl el(g)) + (p -1)}

= p.
Since eMP/K = p and f M/K = 1,
sWK(W1 : G(Mp/K)
C-)
=P 1f(-l)(>g*l(P- 1)1P1(g))+(p-
=p-1.
Hence
1))
=p2-(p_1)2
=2p-1.
268
However, since eL/Mp =
1,
swMp(O1
: G(L/Mp)
C*)
= {(-p)(Eg*i Bi(g)) + p(p - 1)}
=p+p(p-1)
= p2.
Hence
swK(Indc L//Mp)(01 - 1))
fMp/KS'Mp(01 - 1).
Example 6.2.19 We will now consider a parametrised version of 6.2.17.

Let K = Fp((T))((7r)) as in 6.2.17. Choose a e C9K = FP((T))[[7r]]* such
that
XP - a E FP((T))[X] = K[X]
is irreducible. Choose two integers, 1 < s < t, and assume that s is

coprime to p. Define polynomials
fa,s(X) = XP
-7ts(P-l)X
-a
fa,t(X) = XP -
7Ct(P-')X
-a
and
with roots a,,, and aa,t, respectively.

Set L = NoNP, where No = K(aa,t) and NP = K(aa,s).
Hence L/K is a Galois extension with group G(L/K) = Z/p x Z/p

whose generators, tYa,s and aa,t, are defined by
Qas aas = aas + ITS,
bas a.,[ = aa,t,
Qa,t(aa,s) = aa,s,
aa,t(aa,t) = aa,t +7Ct.
Since fa,s(X) = fa,t(X) = XP - a E K[X] is irreducible, the residue

fields of No and NP are isomorphic and are purely inseparable with
[No:K]=[NP:K]=p.
The intermediate fields, K < Ni < L, are the fixed fields of Ua,saa t for
0 < i < p - 1 and of rra,t. The fixed field of aa,t is NP = L = K(aa,s)

while Ni = La
269
= K(aa,t - i7rt-S06a,s) for 0 < i < p -1 since
6a,s6a t(aa,t - hT
t-s
aas)
= Ga 0a,t1(aa,t +71t - i7rtaes
= Qas(aa,t + i7lt = aa,t + i7Ct -
= aa,t -
l7rt-s0Ca,s)
l7rt-s(0Ca,s + 7ls)
illt-saas
The minimum polynomial of aa,t -
i7rt-S0(a,s is
7ctP-1)X
f.-;nP(`-s)a,t(X) = XP -
- (a - i7<P(t-s)a)
since
fa-i,I('-s)a t(aa t - in'-sass)
= as t - i7tp(t-s)aa s +i7rt(P-1)+t-s0Ca,s
= fa,t(agt) -
7rt(P-1)aa
- a + i7r(t-S)Pa
l7CP(t-s) faas
7rtP-s-Pt+Psaa,s
7rP(t-s)-P(t-s)a1
= (-i7rP(t-s))fa,s(aa,s)
=0.
We will now show that eLIN1 = p. To accomplish this we will find a
prime element, 7rL, for L and find its irreducible polynomial over N.
270
Set /3 = aa,t - aa,s Clearly L = K(aa,s, /3) = Np(/3). Also

f3P -
1rt(P-1)Q
= ttQ t - as s -
71t(P-1)pta
t+
7rt(P-1)aas
= a - aPa s + nt(P-1)0(a s
= 1tt(P-1)aa,t
=
lies in
N.
70P-1)1%as
1rs(P-1)aas(1r(t-s)(P-1)
- 1)
Also, since 7r(t-s)(P-1) -1 E (9K,

=VNp(7Ts(P-1)Otas)
VNN(fP-7rt(P-1)N)
= VNp(1s(P-1)),
= s(p - 1),
since aa,s E (9;Np and 7r is prime in N. On the other hand, vL(f;P 7rt(P-1)$) = eL/Mps(p - 1). However, (3 E (9L so that vL(fl) >- 0 and t > s
implies that
vNp(fP - nt(P-1)f) = eL/Nps(P - 1)
= vL(IP)
= PvL(l3)
Since HCF(s, p) = 1 we must have eL/Np = p and vL(fl) = s(p - 1). Now
choose integers, r1 and r2, such that r1s(p - 1) + rep = 1 and set
7CL = $1i2,
so that vL(7rL) = r1s(p - 1) + rep = 1 and 7rL is a prime element for L.

Now let us record the table of ramification indices, residue degrees and
separable residue degrees of these field extensions (see Table 6.2.)
Since Ni = NP and eL/Np = p, we must have 1 = fL/Np = fL/N; for all
i. Hence f N;/K = p for all i and therefore eN,/K = 1 for all i.
Next we compute
vL(ra,sca,t(1rL) - 70 = 1G(L/N;)(aa,s7a,)
271
Table 6.2. Field extensions (Example 6.2.19)

E/F
fsep
eE/F
fE/F
Nj/K
i=0....,p
L/N;
i0,...,p
for
and
VL(6a,t(ltL) - ltL) = iG(L/Np)(6a,t)
Temporarily write 6i for our chosen generator of G(L/Ni) for 0 < i < p.
Ifi=0,l,...,p-l thenQQ
6as6a,t(aa,t - aa,s)
= 6a,s(a,t + i7Ct - OCa s)
= /3 +
7CS(i7tt-s
- 1),
while
= 6a,t(aa,t - a,s)
6p(Y)
Since 7CL = #'I 7r'2 and 7r E K,
/art )
= r2VL(lr) + VL(6i(Nrt) -
VL(6i(tL) - JtL)
= pr2 +VL(a,(#rt) - Nr!).
When i = 0,1, ... , p - 1, then

VL(6i(QQI'r') - fire )
=llL(>' j
Crl
Rj,,s(rt-j)(i t-s
=
= (rl - 1)s(p - 1) + sp
=rls(p-1)+s,
1))
1)r!-j)
272
Table 6.3. Ramification function (Example 6.2.19)

E/F
SG(E/P)(g) (g * 1)
L/N;
i = 0,...,p- 1
(t-s)p+s
L/N,
so that, in these cases,

VL(Qi(7CL) - 7tL) = pre + rls(p - 1) + s = 1 + s.
When i = p,
VL(6p(7CL) - 70
= pr2 + VL
rl
U)
p.i t(r,-J)
= pre + vL(rlIr,-i71r)
= prz + (ri - 1)s(p - 1) + tp
=(t-s)p+s+1.
For 1 * g E G(L/N1),
SG(L/N,)(g) = vL(6P(707tL ' - 1), = iG(LINi)(g) - 1,
so that we obtain the values given in Table 6.3.

Hence if 0 : G(L/Ni) ---> C' is non-trivial then, by 6.2.12 since L = Vi,
SWN,(O) = p){-Y-i
sG(L/Ni)(g)(0(g) -1)}
=p 1SG(LINj)(Qi)(p - 1 + 1)
= SG(L/Ni)(a )
s,
if i=O,...,p-1,
1 (t-s)p+s if i=p.
273
Now consider G(Nt/K) = G(L/K)/G(L/N;), which is generated by 0a,s

when i = 0,1, ... , p - 1. Since N;/K is unramified,
SG(Nr/K)(6a,s)
= iG(Nj/K)(6a,s)
= llIVi(6as(IXa,t -
i7Ct-saas))
= VNj(aa t - 17rt-s((Xa,s + 70) - Cta,t + l7Ct-sCCas)
= VN;(int)
= t.
When i = p, Qa,s still generates but
SG(Np/K)(Qa,s)
= CG(Nv/K)(6a,s)
= VNp(6a,s(aa,s) - aa,s)
= VN5(7Cs)
= S.
Hence if 0 : G(N;/K) -> C' is non-trivial then, by 6.2.12,

SWK(0)
= {- El*, sG(Ni/K)(g)(O(g) - 1)}

= SG(N;/K)(aa,s)(P - 1 + 1)
ps,
1 pt,
if i = p,
if i=O,...,p-1.
We record the values of the Swan conductors on non-trivial one-dimensional representations, 0, in Table 6.4
Let 0, : G(L/N;) -* C' be a non-trivial one-dimensional represen-
tation and also denote by Oi an extension of this homomorphism to

G(L/K). Define
ipi
: G(L/K) --> G(L/K)/G(L/Nt)
G(N,/K) -> C*
274
Table 6.4. Swan conductors (Example 6.2.19)
E/F
eE/F
fE/F = fE/F
SWF(O)
L/N;
i=O,...,p-1
p
p
L/Np
p(t - s) + s
N;/K
i=O,...,p-1
Np/K
Pt
Ps
by the formulae
lpp(ua,t) = 1
lpp(aa,s) _ p,
or
if i=0,...,p-1.
pi(aa,s) _ p i, wi(a ,t) = Sp

Hence
p-1
I ndG(LIN) )(Oi - 1) _ P-1
Bitpi - E tp .
j=o
j=o
Consider the case when i = p. Let us choose Op(aa,s) = 1 and 9p(aa,t) _
p so that
j
8pvp(aa,tQa,,) -
+v'
and Ker(9ptpp) =< Qa,,aa,s >= G(L/N_j) for j = 0,...,p - 1 E Fp.

Similarly, Ker(ippi) = G(L/Np), so that
1))
= p(pt) - (p - 1)ps
= p2(t - s) + ps.
Therefore
-1))
f Q,KSwNp(9p - 1) = At - s) + s.
Now consider simultaneously the cases given by i = 0, 1, ... , p - 1.

Choose Oi(aa,s) = p and Oj(Qa,t) = 1 so that
ej u
iWi
zzv-ijv+uj
Sp
Thus Ker(Oitpj) runs through the set {G(L/N,,) I 0< w < p} for 0< j<
p, while Ker(tp,) runs through the set {G(L/N,v) 1 0 < w < p - 1} for
275
1 < j < p. Hence

swK (I ndG(L/N) (Oi - 1))
= swK (Op : G(N /K)
= Ps.
Therefore, for 0 < i < p - 1,

SWK(I ndG(LINI)(ei - 1))
N/KSWNI(0i - 1) = s.
Example 6.2.20 In this example we shall construct an extension, S/K,

by applying the process of 6.2.19 to
fa,s(X) = XP -
7Cs(P-1)X
-a
and
fbs(X) = XP - 7ts(P-1)X - b,
where a,b e (9K with vK(a - b) = 1. Now set Ro = K(ab,s), RP = NP =
K(aa,s) and S = K(ab,s,aa,s) = RoRp so that G(S/K) = Z/p x Z/p. In

this case the intermediate fields are K(ab,s - iaa,s) = Ri for i = 0,..., p -1
and Rp = K(aa,s). In addition,
fb-ia,s(X) = XP - 7Cs(P-1)X - b + is
has ab,s-iaa,s as a root. Among the fb_ia,s(X) there is a special polynomial,

namely fb_a,s(X). Since b - a is a prime element this polynomial is
Eisenstein. Hence eR,/K = p and we may choose 7<s = ab_a,s = ab,s - aa,s
for the prime element of S. It follows easily that (9K [aa,s, 7 TS] = (9s.
Hence Ro, R2, ... , Rp_ 1, Rp are generated by a root of the polynomial
fb_ia,s(X) for i * 1 and the determination of the first two columns of Table
6.5 are identical to the calculations Mo, M1,. .. , Mp_1 in 6.2.17. In the final
column of Table 6.5 0 is any non-trivial one-dimensional representation
of G(E/F). In the second column it is always true that fE/F =
and
in the third column it is understood that 1 * g E G(E/F).

The verification of the remaining entries in Table 6.5 is left to the
reader as an exercise.
Define pi : G(S/Ri) -- C' by pi(aa,sa,s) = lp if i = 0,1,...,p -1 and

PP(ab,s) = gyp. Define
Ai
: G(S/K) -+ G(S/K)/G(S/Ri) = G(Ri/K) --> C'
276
Table 6.5. Ramification functions and conductors (Example 6.2.20)
E/F
eE/F
JE/F
SG(E/F)(g)
iG(E/F)(g)
SWF(8)
ps-1
ps
ps-1
p
p
ps
Ps
pzs
ps - 1
ps
ps
SIR;
i=0,2,...,p
R; /K
i=0,2,...,p
SIR,
R, /K
ps - 1
by
).p(eas) =
Ai(Qas) =
bpi
Sp'
) p(6b,s) = 1
or
M(0b,s) = by
if
i = 0, ..,p-1.
Extend pi to G(S/K) by the formulae

Pp(ob,s) = p,
Pp(Qa,s) = 1
or
Pi(Qa,s) = bp,
Pi(Qb,s) = 1
if i = 0,...,p- 1.
Therefore
P-1
G(s/K)
R) (Pi
- 1) = T(PiAA -- 2;
2; ).
j=0
Hence, if i * 1,
PS - 1 = SwK(IndG(s/Ri)(Pi - 1)) = fR/KS'R;(Pi - 1)

and, in the remaining case,
ps+p- 1 =
1))
fR /KswR,(P1 - 1) = pas.
There are a number of non-trivial properties possessed by swK which

attest to its `rightness' as a generalisation of the classical Swan conductor
in the abelian case. The following result is one such example:
Proposition 6.2.21 Let M be a henselian discrete valuation field over K
such that (9K c (9M and mM = (9iymK. Assume that the residue field, M,
is separable over K. Let L/K be a finite Galois extension of complete,

discrete valuation fields such that K < M < L, then
SWK(X) = swM(ResG(G(L/K)/M)(X))
for all one-dimensional representations, X : G(L/K) -i C.
Proof See Kato (1989, section 6.2, p. 119).
277
11
Corollary 6.2.22 Let L/K be a finite Galois extension of complete, discrete

valuation fields with Galois group, G(L/K), and inertia group, Go(L/K) =
G(L/M), as in 6.1.1. Let X : G(L/K) --> C' be a one-dimensional representation, then

/K)(X))
swK(X) = swm(ResG(LIK)
(L
Proof In the classical case this result is 6.1.25.
By 6.1.5, M/K is unramified (so that mm = (9MmK) and M/K is

the maximal separable subextension of L/K. Also complete, discrete
valuation rings are henselian (Milne, 1980, p. 32) so that 6.2.21 applies
to yield the result.
Let L/K be a finite Galois extension of complete, discrete

valuation fields. Let T be an indeterminate so that we may form the
6.2.23
localisation with respect to KL, (9L[T](nL), of the polynomial ring, (9L[T].
The field of fractions of the henselianisation of (9K [T](,K) was used in

the definition of the Swan conductor in 6.2.10. We shall now study the
effect of that henselianisation process upon the extension, L/K. We will
see that the role of the henselianisations in 6.2.10 is to provide a Galois
extension with the same Galois group but with more cohomological
structure than the original extension.
Let L denote the field of fractions of (9L[T](,,L). An element of the

completed ring is clearly of the form
00
z=
aiT'
(ai E (9L; vL(ai)
co as i - oo).
i=o
If ao E (9 = (9L - mL then z is invertible. The field L consists of Laurent

series of the form
00
w=Ea1T' (aieL; vL(ai)->ooas i->oo).

For this ring is clearly a field in which the inverse of w is given by
(a_n T 0) w-1 =
+ E00-n+1 ai(a_n)-1 T')-1. Hence L =
Tn(a-n)_'(1
(9L [T ] (RL) [(1GL T)-1 ]
278
Proposition 6.2.24 Let Autk(L) denote the group of K-linear field automorphisms of L. Every E Autk (L) is of the form
(fair)
=
c(ai)Ti
i=-n
i=-n
for some 0 E G(L/K).
Proof Let : L -- L be a k-algebra isomorphism. Since T E k we

have 4 (T) = T and therefore i is determined by c : L ---> L. For
a E L' suppose that
O(a) = T-i(a)(4o(a) + 01(a)T +... + 4m(a)T' +...),
where 4i(a) E L and 4o(a) # 0. Hence we have a map, j : L' - + Z such

that j(ab) = j(a) + j(b) and j(K = 0. Now, clearly, j is trivial on the
roots of unity in L" and on U1 = 1 + nL(9L since the former consists of
torsion and the latter is divisible by integers which are prime to the residue
characteristic. Finally, j(nL) = 0 since 7, iK E K'/(K' n U)) c L*/UL so
that eL/K j(nL) = 0. Hence
q5(a) = Oo(a) + 1(a)T +... + 4m(a)Tm +...,
with 4o E G(L/K).
We must show that if 1, 2,

s = 0, too.
4 s_1 are all identically zero then
Assume that char(L) = 0 and take x E L' such that 0,(x) # 0. Let
f(T)=tr+altr-1+...+ao
EK[T]
be the irreducible polynomial of x. Hence f (x) = 0 and

0 = Os(f (x)) = 0s(xr) + 05(al tr-1) + ... + cs(ao)
However, since 4S(ab) = 0o(a)4 (b) + 4S(a)4o(b) and O jai) = 0, we find

that
0 = 0,(X) T (00(x))
Therefore fo(x) is a repeated root of f(T), which is impossible.
Now suppose that char(L) = p, then L = LP, since L/K is Galois.

Therefore the derivation property of (ps implies that it vanishes on LP,
279
Corollary 6.2.25 If K is a complete, discrete valuation field let M(K)

denote the field of fractions of the henselianisation, (OK [T](nK))("), which
was introduced in 6.2.7. If L/K is a finite Galois extension of complete,
discrete valuations fields then M(L)/M(K) is a finite Galois extension and
restriction yields a natural isomorphism of Galois groups
G(L/K)
G(M(L)/M(K)).
Proof By the construction of the henselianisation of a local ring (Milne,

1980, p. 37), the field, M(L), is constructible as the limit of finite Galois
extensions of L of 6.2.24 for which the residue field extension is trivial.
By 6.2.24, we have a split extension of Galois groups
H -i G(M(L)/K) -> G(L/K) = G(L/K).

Finally, one observes that the G(L/K)-fixed points of the integral closure
of OL in M(L) is a henselianisation of OK from which the result follows,
since M(K) is the field of fractions of this henselianisation.
Theorem 6.2.26 Let L/K be a finite Galois extension of complete, discrete
valuation fields and suppose that G(L/K) = Go(L/K) in 6.1.1. Suppose

that G(L/M) 4 G(L/K) is a normal subgroup whose index is prime to the
residue characteristic, 0 # p = char(K).
Let X : G(L/K)
C" be a one-dimensional representation, then
[M . K]swK(x) = swM(ResG(L/M)(x)),
where swK is the Swan conductor of 6.2.10.
Proof Set e = [M : K] = [G(L/K) : G(L/M)]. Since #(G(L/K)) _

#(Go(L/K)) = eL/K f'ns and HCF(e,p) = 1 we must have fLnl = fijM
and e = eM/K. Therefore there exists u E (9; such that TrK = 7Ceyu E OM.
Therefore, if we apply the norm, N, to 1 +iMa E 1 +xxMO y [T] we obtain
N(1 + 7rMa)
FIgEG(M/K)(1 +g(irKU na))
= 11gEG(M/K)(1 + i4 g(u na)),

which lies in 1 + 7EK OK [T].
Hence we have the inclusion, i, and the norm map which induce
homomorphisms
6.2.27
(i+nKarc[T])
(1+nK'aK[T])
i+nyoy[T]
(i+nM oM(TI)
G(M/K) N
(1+nK aK[T])'
280
The groups in 6.2.27 are annihilated by p so that the formulae, i(N(z)) _

ze and N(i(w)) = we, imply that i and N are isomorphisms in 6.2.27. In
6.2.27 the left-hand group is isomorphic to K [T].
Since f M/K = fylK f nr/K = 1 we have K = M. Setting
1 +7rne+i(9M[T]
B`
1+
(9
[T]
we see that Bo has a composition series of the form
0 = Be c Be-1 = ... c Bo,
with B;/B1+1 = K [T] for 0 < i < e - 1. Let z E B1 be represented by

1
nM+la E 1 +7rM+1(9M[T]. Thus

N(z) = IIgEG(MIK)(1 +7CKg(1rMu na))
= 1 + 7tK Q,
where /3 E (9K [T] c (9M[T] vanishes in M[T]. Since K = M we see that

/3 vanishes in K [T] and N(z) E 1 + 7CK 1OK [T]. Hence, in 6.2.27, the
norm vanishes on elements whose representatives lie in B1.
Now take X : G(L/K) -) C. In the notation of 6.2.25, suppose
that {X,1 + 7rK T} * 0 and that {X,1 + 7rn 1 T} = 0. Temporarily let
torsp(A) denote the p-torsion subgroup of A. By Kato (1989, section

5.5(i)), {X,1 + 7CK T j E torsp(H 2 (M(K))). Since [M : K] is prime to p
i
: torsp(H2(M(K))) -* torsp(H2(M(M)))
is injective and therefore

0 # i({X,1 + 7rK T}) = {ResG(LIM)(X),1 + 7CMUnT} E H2(M(M)).
Therefore
ne = swK(X)[M : K].
Now we must show that 0 =
Taking the cup-product with

of the form
Bo - 1 + 7cM(9M T
1 +?CM+e(9M[T]
7rM+1T}.
induces a homomorphism
H2(M(M))
By the previous discussion 1 + 7<M+1 T is in the kernel of the norm,

N : Bo -> Bo. Since HCF(#(G(L/K)), p) = 1, H1(G(L/K); Bo) =0 and
281
therefore there exists a class, w E B0, such that

1 + nM+l T = 7(w)w-1
where y E G(L/K) is a generator. Therefore, by bimultiplicativity of the

cup-product,
{ResG(LIK)
G(L/M) (X) 1 + x1 T }
{ResG(L/K)
G(L/M) (X),Y(w)}
G(LIK)
_ {Y( Res (
- {ResG(L/K) (X),w}
(L/M)
G(L/K)(X)
(X)),YO}
w
- { ResG(L/M)
,w}
= y({ResG(LIM)(X), w})
w}.
Next we observe that, since the henselianisation, (K H M(K)), preserves the triviality of residue field extensions and preserves Galois groups
(by 6.2.25) we have M(K) = M(M). Therefore, for each n, there is a

commutative diagram of natural homomorphisms (cf. 6.2.15) which are
defined in Kato (1987, section 3.5.5):
X
H,2,(M(K)) H,,(M(K))
H,2,(M(K))
H, (M(K)) Hn(M(K))
H, (M(M))
One may see that the left vertical map may be taken to be (1, e) by
means of the formula of Kato (1987, section 3.5.5) :
1(x, Y) = i2(x) + {il(Y), nM(K)},
where iq : H9(M(K)) -> H9(M(K)) is the natural map.
Now let n = p for some (large) integer, a > e. The extension

M(M)/M(K) falls under case I of theorem 3.6 of Kato (1987) and the
282
formula given there shows that

c(L/K) (X)
{Resc(L/nT) '
w} E im(2') and hence
w} E im(i). Since the image of i is G(M/K)-invariant

w}) =
W11
6.3 The non-abelian Swan conductor

Throughout this section let L/K be a finite, Galois extension of complete,
discrete valuation fields, as in 6.2.4. In particular, we do not assume that
the residue field extension is separable. Let G(L/K) denote the Galois
group. Under these circumstances we shall construct a non-abelian Swan
conductor which generalises the conductors of 6.1.16 and 6.2.11.
Let G be a finite group. Let R+(G) denote, as in 2.2.1,

the free abelian group on G-conjugacy classes of characters, 0 : H -> C.
where H< G. Define a natural ring homomorphism
Definition 6.3.1
nc : R+(G) -- R+(G)
by the formula riG((H, 4)G) = (H, 1)G. The image of nc is called the
Burnside ring of G and the homomorphism, rlG, retracts R+(G) onto the
Burnside ring. Let e : R(G) -) Z denote the homomorphism which
sends a representation to its dimension and set
I R(G) = Ker(e).
Theorem 6.3.2 There is a natural homomorphism
aG : IR(G) -> Ker(IG)

defined by
aG(v - dim(v)) = aG(v) -11G(ac(v)) E R+(G)

In addition,
bG(aG(v - dim(v))) = v - dim(v) E R(G).
Proof This is an immediate consequence of 2.2.42 and 2.3.2.
Definition 6.3.3 (Definition of swK) We may define an additive homomorphism

SWK : Ker(qG(L/K)) --) Z
283
by (K < K' < L)

6.3.4
sWK((G(L/K'), O)o(r'IK)/- (G(L/K'),1)G(L/K))
= fK /K ' SWK'(()
where swK,(4) is as in 6.2.10 and fK IK is the separable residue degree of
K'/K. Define
6.3.5
SWK : R(G(L/K)) -> Z
to be the homomorphism which is trivial on the trivial representations

and on IR(G(L/K)) is given by the composition
swK :IR(G(L/K))
-+ Ker(gG(c/K)) s+ Z.
The following is the main result of this section:

Theorem 6.3.6 Let L/K be a finite, Galois extension of complete, discrete
valuation fields with group, G(L/K). The conductor homomorphism, swK,
defined in 6.3.3, possesses the following properties:
(i) When X : G(L/K) -+ C' is a one-dimensional representation then
swK(x) coincides with the Kato conductor of 6.2.10.
(ii) When L/K is separable (i.e. in the classical case) then swK coincides
with the classical Swan conductor of 6.1.16.
(iii) If K < F < L is a chain of Galois extensions and A : G(L/K)
G(F/K) is the canonical surjection then
R(G(F/K))
SWK
Z
SWK
R(G(L/K ))
commutes, where A =Inf.

(iv) Let nQ denote the absolute Galois group of the rationals. Hence
QQ acts on R(G(L/K)) by means of its action on character values. If
w E nQ and x E R(G(L/K)) then
SWK(w(x)) = SWK(').
284
(v) If G(L/M) = Go 4 G(L/K) is as in 6.1.1 then

swK(X)
swM(ResGOL/K)(x))
for all y E R(G(L/K)).
(vi) If G(L/K) = Go(L/K) and G(L/M) = G(L/K) then, for all y E

R(G(L/K)),
[M : K]swK(y) = swM(ResG(L/K)
G(L/M) (y)).
Proof If y is one-dimensional then SWK(X - 1)

= SWK(aG(L/K)(X - 1))
= SWK((G(L/K), y)G(L/K) - (G(L/K),1)G(L/K)),
by 2.3.2(ii),
= SWK(y),

In the classical case, by 6.1.24 and 6.2.13,
SWK((G(L/K'), O)G(L/K)
- (G(L/K'),1)G(L/K))
= SWK(IndG(L/K')(4 - 1)),
where SWK is the classical Swan conductor. Hence
SWK = SWKbG(L/K) : Ker(riG(L/K)) - Z,
which, by 2.3.2, proves part (ii).
Suppose that 2 : G(L/K) ---> G(F/K) is the canonical surjection and
that : G(F/K') -+ C' is a one-dimensional representation. Then

swK-(4) = swK-(24 ), since the conductor of 6.2.10, swK,(4), is defined by
means of Galois cohomology groups which depend only on K'. Also, in
R+(G(L/K)),
I nfG(F1K) ((G(F/K'), 4,)G(F/K)) = (G(L/K'), 4 2)G(L/K),
so that
sWK(1 nfG(F/K)((G(F/K'), 4)G(F/K) - (G(F/K'),1)G(F/K)))
= SWK((G(L/K'),42)G(L/K) - (G(L/K'),1)G(L/K)).
Part (iii) follows from the naturality (with respect to inflation maps) of
aG in Theorem 6.3.2.
285
Clearly, the nQ-action does not alter the Swan conductor of 6.2.10,
SWK (O).
Part (iv) follows from the fact that, in Theorem 2.3.2, aG
commutes with the QQ-action. This is seen by appealing to the fact
that aG is uniquely characterised by the two properties of 2.3.2. Hence
aG(-) = co-i(aG(o)(-))) for co E 91Q, since both homomorphisms fulfil
the characterisation.
To prove (v) we first note that, by naturality with respect to restriction
(in the sense of 2.3.2(i)) of aG, aG and qG, it suffices to show that
SWK((G(L/F), 4)G(L/K) - (G(L/F),1)G(L/K))
= swM(ResGoLIK)((G(L/F), c)G(LIK) - (G(L/F),1)G(L/K))),

where Go = G(L/M). By 2.2.3
ResGOL/K)((G(L/F), 4)G(L/K) - (G(L/F),1)G(LIK))
= EZEGo\G(L/K)/G(L/F)(Go n
zG(L/F)z-' (z-')*(4))Go
-(Go n zG(L/F)z-1, 1)Go,
where (z-1)'(4) = O(z-1 - z). By definition 6.3.3,

SWM((Go n
zG(L/F)z-1
(z-1)*(O))Go - (Go n zG(L/F)z-1, 1)Go)
= SWM((G(L/Mz(F)), (z-1)-(4,))Go - (Go n G(L/Mz(F)),1)G0)

J Mz(F)/M ' SWMz(F)(ResG(L/Mz(F))((z
=
since it is clear from the definition of swK (x) that
SWMi(F)(ReSG(L/MZ(F))((z-1)*(',)))
= SWMF(ResG(L/MF.)(4.))
Now, by Kato (1989, lemma 6.2); (see also 6.2.22), since M/K is
separable,
SWF(4
: G(L/F) -> C*) = SwMF(ResG(L/F)

G(L/F)nGo (4))
G(L/ F)(4 ))
= SWMF(ResG(L/M
286
Hence
sWK((G(L/F), 4)o(L/K) - (G(L/F),1)c(L/K))
SWF(cb : G(L/F) -- C')
= J FMK
Finally,
#(Go\G(L/K)/G(L/F)) = [G(L/K) : G(L/MF)]

= [G(L/K) : Go]([G(L/MF) : Go])-'
= fseP ([G(L/F) : Go n G(L/F)])-1
fsep
-1
sep
L/K(fL/F)
(by 6.1.5)
sep
fF K
so that, by the preceding discussion,
swM(ResGOL1K)((G(L/F), 4)G(L/K)
- (G(L/F),
1)G(L/K)))
= fF/K(SWMF(ResG(L/MF)(4))),
which completes the proof of part (v).
As in part (v), in order to prove (vi) it suffices by naturality to show

that
SWK((G(L/F), O)G(L/K) - (G(L/F),1)G(L/K))[M : K]
= swM(ResG(L/K)((G(L/F), 4)G(L/K)
- (G(L/F),1)o(L/K)))
287
By a computation similar to that used in the proof of part (v), the latter
expression is equal to
#(Gj\G(L1K)1G(L1F))SWMF(ResG(LIF)
Gi(G/F)(4))
_ [Go(L/K) : Gi(L/K)]([Go(L/MF) : Gi(L/MF)])-i'

swMF(ResG(WF)
Gi(WF) (c))
= [M : K]swF()
by 6.2.26
= [M K] fFIKSWF(0)
= [M : K]swK((G(L/F), 4,) (L/K) - (G(L/F),1)G(L/K)),
as required. This completes the proof of Theorem 6.3.6.

Definition 6.3.7
Let L/K be a finite Galois extension of complete,
discrete valuation fields.
We shall say that the Swan conductor, SwK, of 6.3.3 is inductive in
dimension zero on R(G(L/K)) if, for all G(L/F) < G(L/K) and X E
R(G(L/F)),
SwK (I
(x - dim(x))) = f;IK swF(x - dim(x)).
In the examples of 6.2.18, 6.2.19 and 6.2.20 we saw that it is not

possible in general to construct a conductor function which extends the
abelian conductor of 6.2.10, is inflative in the sense of 6.3.6(iii) and which
is inductive in dimension zero, in the sense of 6.3.7. Nevertheless, the
remainder of this section will be devoted to proving (Theorem 6.3.20)
that the obstruction to inductivity in dimension zero for SWK lies with
p-groups which are of the form Z/p" or Z/p" x Z/p, where p is the residue
characteristic. That is, we shall show that swK is inductive in dimension
zero on R(G(L/K)) if and only if SWF is inductive in dimension zero on
R(G(L/F)) for each p-subgroup, G(L/F), of the above form.
We will begin by reducing the question to the case of p-groups.
6.3.8
Proposition 6.3.9 Let L/K be a finite Galois extension of complete, discrete
valuation fields and suppose that Go = G(L/M) < G(L/K). If SwM is

inductive in dimension zero on R(G(L/M)), in the sense of 6.3.7, then SWK
is inductive in dimension zero on R(G(L/K)).
288
Proof If x E I R(G(L/F)) then (if Go = G(L/M) = Go(L/M))

SWK (I ndG(L/F)(x))
G(L/F)
G(L/M
swM(ResG(L/
IndG(L/)(x))
by 6.3.6(v)
swM(IndGanG(L/z(F))((z-l)'(ResG(L/MF)(x))))
= LzEGo\G(L/K)/G(L/F)
= LzEGo\G(L/K)/G(L/F) sWMz(F)((Z
/F)(x))
[G(L/K) : GOG(L1F)]SWMF(ResG(LIF)
(L
= fFlK ' SWF(X),
by hypothesis, since the equality [G(L/K) : GOG(L/F)] = J FeK was

established in the course of the proof of 6.3.6(v).
Let L/K be a finite Galois extension of complete,

discrete valuation fields and suppose that Gl = G(L/M) < G(L/K) =
Go(L/K). If swM is inductive in dimension zero on R(G(L/M)), in the
Proposition 6.3.10
sense of 6.3.7, then swK is inductive in dimension zero on R(G(L/K)).
Proof If x E I R(G(L/F)) then (if Gi = G(L/M) )

[M : K]SWK (I
(x))
swM(ResG(L/K)IndG(L/K)(x))
G(L/M)
G(L/F)
by 6.3.6(vi)
= EzEGiAG(L/K)/G(L/F) swM(I
SWMz(F)((z-i)*(ResG,(L/F)(x)))
= [G(L/K) : G1G(L/F)]swMF(ResG(L/F) (x))

[M: K][Go(L/MF) : Gi(L/MF)]-iswMF(ResG (L/F)(x))
= [M K]swF(x)
by 6.3.6(vi)
= [M : K] fFIK ' 5WF(x),
289
by hypothesis, since the equality
[G(L/K) : G1G(L/F)] = [M : K][Go(L/MF) : G1(L/MF)]-1

was established in the course of the proof of 6.3.6(vi).
In order, as described in 6.3.8, to reduce the question of the

inductivity of swK to abelian p-groups of small rank we shall construct
an inductive, rational-valued conductor
6.3.11
6.3.12
SWG(LIK) : R(G(L/K)) -) Q
using the homomorphism, dG(L/K), of 2.4.1-2.4.12 to replace aG(L/K) in

the construction of swK in 6.3.5.
It is important to notice that SWG(L/K), constructed in this manner,
will automatically be inductive in degree zero, in the sense of 6.3.7, but
will not necessarily be inflative in the sense of 6.3.6(iii).
Lemma 6.3.13 Let dG denote the homomorphism of 2.4.1-2.4.12
dG : R(G) -> R+(G) Q.

Then, in the notation of 6.3.1,
dG(IR(G)) s Ker(rlG) Q.
Proof This follows from the formula of 2.4.14(ii) since, if dim(x) = 0 and
K < H are cyclic, I ndK (Res4 (x)) may be expressed as Ei(oi - Wi) where
4;, Wi : H -+ C" are one-dimensional representations.
Definition 6.3.14 Definition of SWG(L/K) Let L/K be a finite, Galois
extension of complete, discrete valuation fields. Define a rational-valued
homomorphism
SWG(LIK) : R(G(L/K)) -> Q
to be the homomorphism which annihilates trivial representations and

on IR(G(L/K)) is equal to the composition
dc(WK)
SINK 1
R(G(L/K)) -> Ker(rjG(L1K)) Q ---> Q,

where s1NK is the homomorphism of 6.3.4. This definition makes sense
by virtue of 6.3.13.
290
Proposition 6.3.15 In 6.3.14, S WG(L/K) possesses the following properties:
(i) Let fIQ denote the absolute Galois group of the rationals. Hence %
acts on R(G(L/K)) by means of its action on character values. If w E KIQ
and X E R(G(L/K )), then
SWG(L/K)((O(x)) = SWG(L/K)(x)
(ii) If G(L/M) = G Q G(L/K) is as in 6.1.1, then

SWG(L/K)(x) = SWG(L/M)(ResGO`lK)(x))
for all x E R(G(L/K)).

(iii) If G(L/K) = Go(L/K) and G(L/M) = Gl(L/K), then, for all x E
R(G(L/K)),
[M : K]SWG(L/K)(x) =
S WG(L/K) is inductive in dimension zero, in the sense of 6.3.7.
Proof The proofs of (i)-(iii) are the same as those for the corresponding
results ((iv)-(vi) respectively) of 6.3.6. For part (iv) it suffices, by 2.4.1(i),
to observe that for G(L/F) G(L/K)
/
fFMKSWF((G(L/N), 4')G(`/F) - (G(L/N),1)G(`/F))
= fF/KfN/FS'N(4')
= fN/KSWNW )
= SWK((G(L/N), 4)G(L/K) - (G(L/N), 1)G(L/K))
= 5WK(I ndc L/F ((G(L/N), O)G(L/F) - (G(L/N),1)G(`/F))).
6.3.16 Let P1,... , pr be distinct primes and let
C=Z/pi' x...xZ/P;'
be a cyclic group. Write Co < C for the subgroup
C=
Z/ply-1
x ... X
Z/pr'-1,
with the convention that {110 = {11. Recall that the Euler totient
function, (n), is the multiplicative function on the positive integers
291
which is characterised by the formula (b(pm) = pm-i(p - 1) on prime

powers (Hunter, 1964). Hence
fi(n) = #{m I 1 < m < n,HCF(n,m) = 1}.
Lemma 6.3.17 Let G be any finite group. With the notation of 6.3.16
(i) dG(1)
_ #(G)-'
#(C)4(#(Ker(W)/C))4Ker(w),c(C,W)G
(C.W)
Ccycl ic,Res'o (pp)=1
and
(ii)for all G-->C`

dG(u) _ #(G)-'E
#(C)4(#(Ker(W)/C))
(CAW)
Ccycl ic,Resco (w)=1
/lKer(w),C(C, 1pResc())G.
Proof Clearly part (ii) follows from part (i), by 2.4.1(ii). Part (i) follows
from the formula of 2.4.14(ii) upon noticing that 12K,H = 0 unless [H : K]
is square-free when one collects the remaining terms, using the familiar
identity (Hunter, 1964)
#(K/H0)I1K,Ker(ip) = 0(#(Ker(W)1H0))
H<K <Ker(w)
In the following result it is essential that G be a p-group. It is for

this reason that we have reduced the study of swK to the case when
G(L/K) = G1(L/K) by means of 6.3.6(v) and (vi).
Proposition 6.3.18 Let G be a finite p-group. Let N i G be a normal
subgroup and let u : G ---> C' be a one-dimensional representation which
factors through G/N. Then dG( - 1) is a Q-linear combination of pairs,

(H,W)G with ReSHnN(W) = 1.
Proof By the formula of 6.3.17 we have dG(u - 1) =
#(G) '
#(C)o(#(Ker(W)/C))/ Ker(N),CX(C,W),
(C.a0
Ccycl ic,Resco (W)=1
where X(C,W) =
(C,W)G]
292
Now consider the pair (C, V )G. If C n N = C then W = W on C, since
p is trivial on N. If C n N# C then C n N< C, since C is a cyclic

p-group, and hence W and W are both trivial on C n N. This completes
the proof of 6.3.18.
Proposition 6.3.19 Let L/K be a finite, Galois extension of complete, discrete valuation fields. Suppose that, for all one-dimensional representations
of subgroups W : G(L/F) -* C'
SWG(L/F)(W - 1) = SWF(1P - 1),
then
SWG(L/K) = SWK : R(G(L/K)) -* Z C Q.
Proof Let X E R(G(L/K)) and suppose that

ai(G(L/Fi), 4)G(L/K)
aG(LIK)(X) =
i
Therefore
SWK (x - dim(x))
aiswK((G(L/Fi), 0i)G(L/K) - (G(L/Fi), 1)G(L/K))

aifF /K swF,(4i)
E aifF/KSWF (oi - 1)
sep
sep
aifF/KSWG(L/F )(Oi - 1)
aIS WG(L/K)(I ndG(L,F)(Oi - 1))
by 6.3.15(iv)
S WG(L/K)(bG(L/K)(aG(L/K)(x - dim(x))))
by 6.3.2
= SWG(L/K)(x - dim(x)),
as required.
293
Theorem 6.3.20 Let L/K be a finite, Galois extension of complete, discrete

valuation fields. Let p = char(K) and suppose that G(L/K) is a p-group.
Suppose that we have subgroups, G(L/M) i G(L/F) < G(L/K), which

satisfy the following conditions:
(i) If G(M/F) = G(L/F)/G(L/M) = Z/p" then

1)) = fNI'FSWN(O - 1),
where Z/p"' = G(M/N) = G(M/F) < G(M/F) and 6 : G(M/N) -->

C' is a faithful one-dimensional representation.
(ii) If G(M/F) = G(L/F)/G(L/M) = Z/p" x Z/p then

1)) = f N/FSWN(9 - 1),
SWF(I
where Z/p" = G(M/N) < G(M/F) and 0 : G(M/N) -+ C' is a faithful

one-dimensional representation.
Then, for all G(L/R) < G(L/K) and A.: G(L/R) -> C',
S WG(R/R)(A - 1) = SWR(A - 1).
Corollary 6.3.21 Under the hypotheses of 6.3.20, 5WK is inductive in dimension zero on R(G(L/K)) in the sense of 6.3.7.
Proof This follows from 6.3.19 and 6.3.20.

Suppose that G(L/F) Q G(L/K) is a cyclic, normal subgroup of order p.

C* is trivial on G(L/F). We will begin
Suppose also t h a t ) . : G(L/K)
by showing that
6.3.23
SWG(L/K)(A - 1) = SWG(FIK)( - 1),
where, in 6.3.23, 2 : G(F/K) = G(L/K)/G(L/F) -+ C' is induced by A.

By 6.3.14 and 6.3.17,
294
6.3.24
#(G(L/K))SWG(L/K)(2 - 1)
_ EG(L/T) Ew #(G(L/T))O(#(Ker(W)/G(L/T)))
/Ker(lp),G(L/T)fT1KSWT(A1P -W)
, [(P
_ EG(L/T) #(G(L/T)).fsse
TIK
1))
- Ey, SWT(21j) - W)A
where, in 6.3.24, the first sum is taken over cyclic subgroups, G(L/T), such
that {1} * G(L/T) < G(L/K) and over ip : G(L/T)/G(L/T) -) C',

and the second sum is taken over cyclic subgroups, G(L/T ), and faithful
representations, yp
Similarly,
: G(L/T)/G(L/T) = Z/p --> C.
6.3.25
#(G(L/K))SWG(F/K)(A-1)
#(G(F/S))fs`1K [(P - 1)sws(ResG(is))(A -1))
=P
G(F/s)
sws(2W
- W)],
where the sum is taken over G(L/F) < G(L/S) < G(L/K) such that
G(F/S) is non-trivial, cyclic and ip : G(L/S)/G(L/S) ---> C' is a faithful
one-dimensional representation.
Hence, with the summation as in 6.3.24 and 6.3.25, we must prove that
6.3.26
EG(L/T) #(G(L/T))fTjKswT((ResG(L,T (.1- 1))(P -I nd'(LI
)O(1)))
_ EG(F/S) P#(G(F/S))fs1Ksws((ResG(L,s))(2 - 1))(P - I
In order to verify the identity of 6.3.26, consider the following map

between the indexing sets of the two sums (note that the term with T = F
295
vanishes in 6.3.24):
{{1} * G(L/T) < G(L/K) I G(L/T) cyclic, T * F}
{G(L/F) < G(L/S) < G(L/K) I F
S,G(F/S) cyclic},
given by the formula

6.3.27
F(G(L/T)) = G(L/T)G(L/F) = G(L/T n F).
We wish to show that F is surjective. We claim that
6.3.28
either (a)
or
(b)
G(L/S) is cyclic
G(L/S) = G(L/F) x C
for some cyclic subgroup, C. For if G(L/S) is not cyclic then the cyclic
group, G(L/S)/G(L/F) = G(F/S) acts trivially on G(L/F), being a pgroup, so that G(L/S) is abelian and must be isomorphic to Z/p x Z/p"
for some n. An examination of the lattice of subgroups of Z/p x Z/p"
shows that each subgroup of order p has a complement and therefore
G(L/S) G(L/F) x C, as claimed. This means that F is surjective.
Now we will prove that (noting that p#(G(F/S)) = #(G(L/S))p 1)
(A - 1)(p -
6.3.29
I
r(G(L/T))=G(L/S)
where G(L/U) is the pre-image of G(F/S) under G(L/S) --- G(F/S).

Note that S = F n T under the map of 6.3.27 and, as we shall explain,
#(G(L/T)) is constant for all T such that F(G(L/T)) = G(L/S), so that
dividing by #(G(L/T)) makes sense in 6.3.29.
In case (a) of 6.3.28 G(L/S) is cyclic and F-1(G(L/S)) consists only of
G(L/S) so that T = S, #(G(L/S)) = p#(G(L/T)), G(L/U) = G(L/T)

and both sides of 6.3.29 coincide.
In case (b) of 6.3.28, F '(G(L/S)) consists of the p complements of

G(L/F) in G(L/S), G(L/E1),...,G(L/EP), say. Hence G(L/E;) = Z/p" for
1 < i< p. Therefore, by part (ii) of the hypothesis, the sum over G(L/T)
of 6.3.29 becomes (where G(L/U) n G(L/E;) = G(L/UE;) is independent
296
of i and
1 = pl-npn+lP l
(#(G(L/Ei)))-1#(G(L/S))P
=p
for all i)
2p
fE,/KSWE,((ResG(L/E (A - 1))(P - IndG(LIUE jl )))
1
SeP
P
- 1 ))(pI ndG(LIS)
= i=1 fS/K sw S(( ResG(L/K)
G(L/S) (
G(L/E;)( 1) I ndG(L/s)
G(L/UE;)( )))
=pfs /
G(L/s)
1))(>pi=1IndG(L/s)
1 -IndG(L/s)
G(L/E;)O
G(L/UE,)(1 )))
Now G(L/S)/G(L/UEi) = Z/p x Z/p and

P
ip E R(G(L/S)),
p}
i=1
where p : G(L/S)/G(L/UEi) = G(UEi/S) --p C' runs over all characters

whose kernels belong to the set {G(L/Ei)}. On the other hand,
(1)
IndG(L/s)
G(L/UE,)
p E R(G(L/S)),
P
where p runs through all p : G(UE,/S) -- C. Hence the difference

between these two expressions equals
p-I
E R(G(L/S))
and we find that the G(L/T)-sum in 6.3.29 reduces to

pfs1K sws (ResG(L1S) )(A - 1)(p - I ndG(LIU) (1))),
as required to verify 6.3.29 and 6.3.26, and thereby to establish 6.3.23.

Since swK is invariant under inflation, in the sense of 6.3.23, we may
filter the kernel of ) : G(L/K) -+ C" by subgroups
G(L/Fo) .1 G(L/F1) i ... a G(L/Fi) = Ker(A) .i G(L/K)

in such a manner that [Fi+l : F1] = p for all i. By induction on the order
of G(L/K) we may therefore reduce to showing that
6.3.30
S WG(L/K)(t - 1) = SWK(A - 1)
in the case when G(L/K) is a cyclic p-group. If #(G(L/K)) is equal to 1

or p then 6.3.30 is immediate from the definition. Also we may suppose
that A : G(L/K) -- C' is faithful. Suppose that G(L/K) = Z/pn with
6.4 Exercises
n > 2 and let). : G(L/F) =
297
C* be a faithful representation.
Z/pn-1
Hence
P
I ndG(L/K)(
-1) =
G(L/F)
wi(g)
i=1
i=1
where 0)1,...,w, E CIQ and 1p1,...,WP : G(L/K) - C' are trivial

G(L/F). Therefore, by 6.3.6(iv), 6.3.15(i), 6.3.20(i) and induction,
on
PSWK(A - 1) - EP SWG(L/K)(Wi - 1)
1
= PSWK() - 1) - EP 1 SWK(Wi - 1)
(A - 1))
= SWK (I
= fF KSWF(2 - 1)
= fFIKSWG(L/F)(A- 1)
(A - 1))
= SWG(L/K)(I
= pSWG(L/K)(A - 1) - Ep 1 SWG(L/K)(Wi - 1),
which proves 6.3.30 and completes the proof of Theorem 6.3.20.
6.4 Exercises
6.4.1
Prove the formula of Theorem 6.1.43.
6.4.2
Let Q2m (m> 3) denote the generalised quaternion group of order
2m:
x2m2
Q2m = {x,Y I
= Y2, Y4 = 1, xYx = Y}.
Suppose that L/K is a finite Galois extension of 2-adic local fields with
group G(L/K) = Go(L/K) = Q. Let AG(L/K) E R(Q2m) denote the Artin
representation of 6.1.7.
(i) Prove that, in the sense of 4.3.31,
AG(L/K) - WZ(AG(L/K)) = 0
(mod 2).
(ii) For z E Z2[Q2m]* prove that, in the sense of 4.3.38,

Det(z)(3W2(AG(L/K))
- W4(AG(L/K)) - 2AG(L/K)) E {1} + 40N,
298
where N/K is a finite Galois extension which is large enough to contain

all the 2ri-1 th roots of unity.
In 6.1.24 prove that the Artin conductor, fK, is inductive in

dimension zero if and only if the same is true of the Swan conductor,
6.4.3
SWK. That is, show that
fK(IndG(L/K'G(L/K>) (X - dim(X))) = (A'10 .fK'(X - dim(X))

holds for all X E R(G(L/K')) if and only if
G(L/K'G(L/K))
SWK (I nd
6.4.4
(X - dim(X))) = (fK'/K) SWK'(X - dim(X))
Let SG(L/K) and iG(L/K) be the ramification functions of 6.1.3.
(i) In the classical case, find an element at which these infima are
attained.
(ii) Verify the formula of 6.1.4.
Complete the verification of the table of ramification functions

and conductors in 6.2.20 (Table 6.5) and verify the other conductor
calculations which are omitted in that example.
6.4.5
6.4.6 (Research problem) Let L/K be a finite, Galois extension of complete, discrete valuation fields. If K < F < L is a subextension, find a
formula for
G(LK
swK (I ndG(LIF) (1)) E Z,
where SWK is as in 6.3.3.
6.4.7 (Research problem) Let p be a prime. Let L/K be a finite, Galois

extension of complete, discrete valuation fields in characteristic p. Is the
conductor, swK, of 6.3.3 inductive in dimension zero modulo (p - 1)?
A weaker question would be the following: is swK inductive in dimension zero modulo (p - 1) when K is a higher dimensional local field, in
the sense of Kato (1987)?
7
Introduction
In Section 1 we consider a finite Galois extension of local fields, L/K. The

theory of local class formations asserts the existence of a fundamental class
which gives a canonical generator for H2(G(L/K); L'). To each generator
of this cyclic group and to each cohomologically trivial subgroup, U < L*,
such that L"/U is a finitely generated Z[G(L/K)]-module we show how

to assign an Euler characteristic in the class-group. When applied to the
canonical generator this gives the local Chinburg invariant, KI(L/K, U),
which lies in the class-group, 16'(Z[G(L/K)]). We develop sufficient
cohomological technology to be able to describe S2(L/K, U) canonically.
This is accomplished by means of a description of the fundamental class,
which is due to J-P. Serre, together with a homological construction
which is given in terms of a K-division algebra containing L as a

maximal subfield. We use this canonical description to give a new proof
of the theorem of Chinburg which states that Q(L/K, UL) is trivial when
L/K is tamely ramified.
In Section 2 we define the global Chinburg invariant, c2(L/K, 2) which
lies in '21(Z[G(L/K)]), where L/K is a Galois extension of number

fields. This was originally described as another type of Euler characteristic
construction. Here we describe KI(L/K, 2) in terms of local Chinburg

invariants at the wildly ramified primes in a manner which is due to
S. Kim. In addition we show how to use the Artin root numbers
of the symplectic representations of G(L/K) to construct a 2-torsion
element, WL/K E '2'(Z[G(L/K)]). In the tamely ramified case the
Chinburg invariant becomes simply the projective module given by the
ring of integers of L, which was conjectured by Frohlich to be equal to
the analytic Cassou-Nogues-Frohlich class, WLIK. This conjecture was
299
300
proved by M.J. Taylor. The Frohlich-Chinburg conjecture asserts that

S2(L/K, 2) and WLIK should be equal in general.
To date, the best general result concerning the Frohlich-Chinburg

conjecture is David Holland's result, which shows that the conjecture
holds when mapped to the class-group of a maximal order of the rational
group-ring. In Section 3 Holland's result is proved in a new manner by
means of the detection machinery which was developed in Chapter 5,
Section 4.
Section 4 contains a derivation of the Hom-description for the global
Chinburg invariant,
2), of the totally real cyclotomic ex-
tension of p-power conductor when p is an odd, regular prime. These

Chinburg invariants had not been considered previously and the appearance of the p-adic L-function in their Hom-description is particularly
satisfying and probably significant. Using naturality under passage to
quotient groups the Chinburg invariant, 12(FS+1 /Q, 2), is shown to vanish, where FS+i /Q is the intermediate extension having Galois group of
order ps. This provides further examples in which the Frohlich-Chinburg
conjecture of 7.2.14 is true. Unlike the material of Section 3, these calculations are not derived by the use of Explicit Brauer Induction, but they
seem particularly appropriate for inclusion into this chapter.
Section Five consists of a collection of exercises concerning class

formations, cohomology and the Frohlich-Chinburg conjecture.
7.1 Local Chinburg invariants

Let G be a finite group and suppose that A,B are Z[G]-modules. We will
begin by recalling some homological algebra. Suppose that
...-*P, -*...P1 -*Po-fA- 0
7.1.1
is a projective Z[G]-resolution of A. Hence each Pi is a projective

Z[G]-module and the Z[G]-homomorphisms of 7.1.1 satisfy
dd=0, ed=0
and 7.1.1 is exact (i.e. the kernel of each map equals the image of its
predecessor). The group, Ext'lGl(A,B), is defined to be the ith homology
group of the chain complex
7.1.2
... 4- HomZ[GJ(P,,,B) 4-- ... <

d*
d*
HomZfoJ(Po,B) +- 0.
301
Hence
7.1.3
Ext'[G)(A, B) =
ker(d' :HomZtGl (Pi,B)-+HomZ G, (Pi+I,B))
where we adopt the convention that Pi = 0 if i < 0. Up to a canonical

isomorphism, this definition is independent of the choice of resolution in
7.1.1.
The ith cohomology group, H'(G; B), is defined (Snaith, 1989b, p.2) to
be given by
H'(G; B) = ExtZ[G)(Z, B),
7.1.4
where G acts trivially on Z, the integers.

A canonical projective resolution of Z is given by the bar resolution
(Snaith, 1989b, p. 3)
...d2
7.1.5
B2G-BIG o) BOG' ) 0.
In 7.1.5
is the free left Z[G]-module on G. If (91,...,gn) E Gn we
write [gl 192 I . . . gn] for the corresponding Z[G]-basis element of B.
We write [ ] for the basis element of BOG. The Z[G]-homomorphisms of
7.1.5 are given by
I
e(g1 []) = 1,
dn([g1 I g2 I
and
I gn+1])
. . .
= 91 [g2
I gn+1]+
1(_ I)i [gl

I
(_ 1)n+1 [gl
I g2 I
I gigi+l
...
..
I gn+11+
I gn].
One may show that 7.1.5 is exact by constructing a contracting homotopy
0-Z- oBOG -pB1G.... .. . ...

given by the formulae
n(1)=[],
sn(g1 [g2
...
and
I gn])
= [91 192 1
... I gn] for n >_ 0.
302
One readily verifies the following identities

1 = eh,
1 = ne + doso
and
1 =
for n >- 1.
In this section we will mainly be interested in H2(G; B). For future reference we will record the description of H2(G; B) which the bar resolution
provides.
Proposition 7.1.6
notation). Then
Let B be a left Z[G]-module (written in additive

H2(G; B)
Z2/B2,
where
Z2 =
f: G x G --> B I glf(g2,g3) +f(g1,9293)

=f(g192,93)+f(91,g2)
for all 91,92,93 E G
and
B2 =
f : G x G --+ B I f(g1, g2) = gih(g2) + h(gl) - h(gig2)

for all 91,92 E G and
for some It : G
)B
(Z2 and B2 are called the groups of 2-cocycles and 2-coboundaries,

respectively.)
Now suppose that L/K is a finite Galois extension of local fields with
group G(L/K). The multiplicative group, L*, is a Z[G(LIK)] -module
and there is a canonical isomorphism (Serre, 1979, pp. 166 and 195):
7.1.7
inv : H2(G(L/K); L*) -- Z/[L : K] c Q/Z.
An explicit description of the generator, inv-1(-[L : K]-1) may be given

by means of the fact (cf. 5.2.20 and 5.2.21) that H2(G(L/K); L*) classifies
extensions of the form
0-L* -+A-B-+Z-*0
and generators of H2(G(L/K);L*) classify those extensions for which A
and B are cohomologically trivial Z[G]-modules (i.e. H`(H; M) = 0 for
all i > 0 and H < G when M = A, B).
303
The following construction is taken from Serre (1979, p. 202).

Let W/K be the maximal unramified subextension of L/K. Let Knr/K
denote the maximal unramified extension so that Go(Knr /K) = {11 and
there are isomorphisms
G(Knr/K)
G(Knr/K) = G(Fq/Fq),
where Fq is the finite field of order q and F9 is its algebraic closure. Let
F E G(Knr/K) correspond to the Frobenius element (see 4.3.7) given by
the qth power map on F. Set Lo = KnrL. Consider the sequence
7.1.8
0L'`>(LKKnr)*(l F)il(LKKnr)*-">Z_*0.
In 7.1.8, if aEL,/EKnr,
(1F)(a0 /3)=aF(/3).
7.1.9
If d = [W : K], there is an isomorphism of K-algebras

7.1.10
d 1 Lo
A : L OK Knr

a2
.1(a N)
_ (Fd-1(Q)a,F
The map, w in 7.1.8, is given by

7.1.11
1L'o.Z.
w=(Ed v u )
Now let us consider the Galois action in terms of 7.1.10. We have a

map, given by restriction of the action of g to W (denoted by (g I W))
h1
: G(Knr/K) x G(L/K) --> G(W/K)
7.1.12
hi(F`, z) = (F I W)(z I W)-1.
If (F`,z) E Ker(hl) then we may define an element of G(Lo/K) which is

equal to F` on Knr and to z on L. This induces an isomorphism
7.1.13
Define an element
Ker(hl)
G(Lo/K).
304
Fo E G(Lo/K)
7.1.14
by (Fo I Knr) = Fd and (Fo I L) = 1.

Lemma 7.1.15 Define
(1F):d1Lo -->d1Lo
by (1 0 F)(x1,...,xd) = (Fo(xd),x1,...,xd_1), then the following diagram
commutes:
(1 F)
L OK Knr
L OK Knr
(1 F)
d 1Lo
d1Lo
Proof If a E L and # E Knr then

2(1 (9 F)(a (D /3)
= A(a 0 F(/3))
= (Fd(Q)a, Fd-1(l3)a, ..., F(l3)a),
while
(1 0 F)(2(a /3)) = (
= (Fo(fla), Fd-1(l3)a, ... , F(l3)a)
and the result follows since, by definition,

Fo(/a) = Fo(/3)a = Fd($)a.
Lemma 7.1.16 The sequence of 7.1.8 is exact.
Proof Let nL E CL denote the uniformiser of L. Hence there exists

u E (L OK Knr)* such that 2(u) = (nL,1,1,...,1) and therefore w(u) =
vLo(7rL) = 1, by 7.1.11, so that w is onto. By means of 7.1.15, (1 F)/1
may be identified with the map
(10 F)/1 : d 1Lo -- * d 1Lo
305
(1 F)/1(xl,...,xd) =
7.1.17
,x1x2 1,x2x3 1 ,...,xd-1xd
(FO(Xd)
1).
Hence the kernel of (1(D F)/ 1 consists of d-tuples of the form (y, y,... , y),
where y E Lo and Fo(y) = y or, equivalently, y E L. Since 2(i(y)) =

t(y 1) = (y,y,...,y) we see that Ker((1 F)/1) = Im(i).
Finally, suppose that a c (L K Knr)* satisfies w(a) = 0. Let )(a) =
(x1,..., xd) with x; E Lo. By 7.1.11, Ed vL,(x;) = 0 so that x = jjd 1 x; E
1
(9L, the units of (9k. We claim that there exists u E (94 * such that
= x. Let us assume this fact for the moment. Consider the
Fo(u)u-1
equation
(Fo(Yd)Yi 1,Y1Y21,.. .Yd-1Yd1)
= (1 F)/1(yi,...,Yd)
by 7.1.17
= (X1, ... , xd)
We may solve this by choosing Yd = u and then selecting
to satisfy y;y;+1 = x;+1 for 1 < i < d - 1. With these choices the first
coordinate becomes
Fo(Yd)Yi 1
= Fo(u)Yj
= u(l
xi)Y1-1
= Ydx1YIY2 1Y2Y3 1 . . .Yd-lYd
1Y1
= xl,
as required. Therefore, to complete the proof, we must construct u E (9L.
There is a finite intermediate Galois extension, E/K, such that K r

E c Lo with x r= (9E. Let E denote the residue field of E and
let z E EM denote the image of x. We have W = L = Fq, Lo = Knr = Fq
and FO induces the Frobenius on E/Fq. Since qd - 1 is prime to q we
may take a (qd - 1)th root of x in E, by making E sufficiently large.
WcL
Therefore choose u1 E E* such that

d-1
u1
FO(u1)(u1)-1
= z.
306
We may lift W, to ul E (99E to obtain an equation of the form

Fo(ul)ul 1 = x + 7Gx1
where x1 E (9E and it = 7[L, = lrL denotes the prime element. Now
suppose that we have found xm E (9E and um c (9E such that, for m
1,
FO(um)um x + 7[mxm = x(1 + 7CmxrX 1).
Consider the congruences (v E (9E)

Fo(1 + 7cmv)(1 + 7Crv)-l
- (1 + 7!'Fo(v))(1 - 7[mv) (mod 7[m+1(9E)

1 + 7Cm(F(v) - v)
(mod 7Cm+l(9E)-
By 7.5.2, enlarging E if necessary, we can find v E (9E such that
F(v) - v - xm.x 1 (mod 7[(9E).

Replacing um by um+1 = um(l + 7Cmv) E (9E we obtain xm+1 E (9E such
that
Fo(um+1)um+l = x + 7Cm+lxm+1
The sequence, {um}, converges 7C-adically to u E (9L such that
Fo(u)u 1 = x,
which completes the proof of 7.1.16.
7.1.18
Given a short exact sequence of Z[G]-modules
0->A-- BP) C--)0

there is a natural coboundary map (Snaith, 1989b, p. 9)
6 : H`(G; C) -+ H`+1(G; A)
which fits into a long exact sequence of the form
H`(G; A) - H`(G; B)
H(G; C) - H'+'(G; A) -+ ... .
Therefore, when B is cohomologically trivial, 6 is an isomorphism for
i > 1. In terms of the bar resolution 6 is describable in the following

manner. Let [g] E H`(G; C) be represented by a Z[G]-homomorphism
g:B;G-)C
307
such that gd; = 0. Since B,G is a free Z[G]-module we may choose a

Z[G]-homomorphism
h :B;G->B
such that $h = g. The homomorphism
hd1 :B,+1G-->B,G--->B
becomes trivial when composed with /3, since fJhd, = gd, = 0. Hence we
obtain a Z [G] -homomorphism
H :B;+1G-->A=Ker(f3)
such that aH = hd;. Also H is an (i+1)-cocycle, since aHdi+1 = hdjd;+1 =
0 and a is injective. We define 6[g] by the formula
8[g] = [H] E H`+1(G;A).
7.1.19
In particular, when we have an exact sequence of the form of 7.1.8, we

may derive two coboundary maps
S : H(G;Z) --) H'(G;Ker(w)) = H1(G;Im((1 (D F)/1))

and
S : H'(G;Im((l F)/1)) -f H2(G;L').

There is an isomorphism H(G; Z) = Z in which n E Z is represented
by the homomorphism, gn : BOG -> Z, given by
n. Therefore
7.1.8 gives rise to a class
6(6(1)) = 6(6([g1])) E H2(G(L/K); L`) = Z/#(G).

Theorem 7.1.20 (Serre, 1979, p. 202) In the exact sequence of 7.1.8, the
module (L K
is cohomologically trivial and the equivalence class of
the extension of 7.1.8 corresponds to 6(6(1)) E H2(G(L/K);L"). Furthermore, in 7.1.7,
b(b(l)) = -inv-1([L : K]-').
Proof By the theory of the classification of 2-extensions by means

of cohomology 7.1.8 corresponds to 8(5(1)) (see Hilton & Stammbach,
1971).
The remainder of the proof is left to the reader as exercise 7.5.1.
308
Theorem 7.1.20 permits us to give an explicit description of the

2-cocycle in 7.1.6 which represents
7.1.21
inv-1(-[L : K]-') E H2(G(L/K);L').

Following 7.1.18 we begin with g, : BG(L/K) --p Z defined by g,([ ]) _
1 and lift it to the homomorphism

It
: BG(L/K) - d1Lo
given by h([ ]) = (ltL,1,1,...,1). Now define
i = hd : B1G(L/K) -- d 1L*
so that, for z E G(L/K),
i([z])
= z(h([ ]))(h([ ]))-1
7.1.22
=z(nL,1,1,...,1)(rci,1,1,...,1).
Now choose j : B1G(L/K) -- d 1Lo such that
(1 0 F)(j)lj = i B,G(L/K) -- d1L*

and set
f : G(L/K) x G(L/K)
) L"
to be any 2-cocycle such that, for z1,z2 E G(L/K),

f(Zl,Z2) = j(dl[Z1
I Z2])
7.1.23
= Z1(j([Z2]))j([z1Z2])-1j([Z1]) E L',
where L' is embedded diagonally into d 1Lo.

There is a second manner in which to describe the extensions which are
classified by H2(G(L/K); L'). This uses the fact that H2(G(L/K); L') also
classifies central simple K-algebras of index n = [L : K] (Reiner, 1975,
pp. 242-3). Before describing this we will need a simple cohomological
result.
Lemma 7.1.24 Let G be a finite group and let IG 4 Z[G] denote the
augmentation ideal. The coboundary
6 : Z = H(G;Z) -. H1(G;IG)
309
associated to the short exact sequence
0)IG)Z[G]Z)0
is given by
S(1) = [w],
where w : B1G -+ IG is given by w([g]) = g - 1 for g E G.
Proof Lift g1 : BOG -> Z to g1 : BoG - Z[G] given by h([ ]) = 1 E

Z[G]. Hence hdo([g]) = h(g[ ] - [ ]) = g - 1 is a representative for b(1),
by 7.1.19.
7.1.25 Central simple algebras
Let L/K be a finite Galois extension of local fields with Galois group,
G(L/K), and n = [L : K]. Given a 2-cocycle
f : G(L/K) x G(L/K) - ) L'

we may construct a central simple K -algebra of index n from f (Reiner,
1975, p. 242). Let Vf denote the L-vector space on a basis {ug 1 g E
G(L/K)} where u1 = 1. Endow Vf with the K-algebra structure defined
by the relations
ugxugl = g(x)
(x E L")
and
f (g1, g2)ugjg2 = ug1ug2
(91,92 E G(L/K ))
With this structure Vf becomes a central simple K-algebra which depends,

up to isomorphism, only on the cohomology class,
[f] E H2(G(L/K);L')
in 7.1.6.
In addition, the invariant
inv([f]) E Z/#(G(L/K))
is called the Hasse invariant of Vf. The Vf which are division algebras are
precisely those whose Hasse invariants lie in (Z/#(G(L/K)))" (Reiner,

1975, sections 14.6 and 31.1).
Suppose now that D = Vf is such a division algebra. By Reiner

(1975, p. 240, section 28.10) there is an embedding, L c D, as a maximal
subfield. This embedding is unique up to inner automorphisms of D.
310
Let NDL' denote the normaliser of L' in D' = D - {0}. Conjugation

by z E NDL* induces a K-automorphism of L and hence we obtain a
homomorphism
0 : NDL' -+ G(L/K),
which is surjective, by the Skolem-Noether theorem (Reiner, 1975,
p. 103). The kernel of contains V. However, if z E Ker(4)) then
L(z) is a subfield of D. Since L is a maximal subfield of D we must have
Ker(4)) = L' and therefore
L' -) NDL'
7.1.26
G(L/K)
is exact. In fact, NDL' =< L';ug,g E G(L/K) > and d(ug) = g.

By Hilton & Stammbach (1971, p. 198) 7.1.26 yields a short exact
sequence of (left) Z[G]-modules of the form
7.1.27
L'
X-
Z[G]
INDL' -')IG(L/K)
given by
X(z) = 1(z -1)
(z E L*)
v(g (m - 1)) = g(m) - g (g E G(L/K), m E NDL').
Now let us calculate the image of [w] E H1(G(L/K);IG(L/K)) of

7.1.24 under the coboundary map
6 : H1(G(L/K);IG(L/K)) -+ H2(G(L/K);L')
associated to 7.1.27. We begin by lifting w to the homomorphism
v E HomG(L/K)(BIG(L/K), Z[G]
INDL*)

1). Hence vd1 : B2G(L/K)
Z[G] Z[NDL.] INDL*, by
given by v([g]) = 10 (ug
vdl([gi I g2]) = gi
311
L' is given, in
2-1)-10(ug,g2-1)+10 (ug,-1)
=1ug,(Ug2-1)-10(Ug,g2- 1)+1(Ug1 -1)

(Ug,Ug2 - Ug,g2)
= 10 Wg1, 92) - 1)Ug,g2

= 9192 0 (Ug1g2)-1(f(g1,92) - 1)Ug,g2
= 9192 0 (g1g2)-1(f(g1,g2) - 1)
= 10 W91, 92) -1),
which is the image of P91, $2) E V. Hence
b([w]) _ [f] E H2(G(L/K);L`).
Since inv([1]) E (Z/[L : K])* the theory of local class formations

implies that the homomorphisms
H`(G(L/K);Z) -L H+1(G(L/K);IG(L/K)) -> Hi+2(G(L/K);L')

are isomorphisms for all i >- 1 and therefore that
H`(G(L/K);Z[G] Z[NDL.) INDL*) = 0
for all i >- 1. Furthermore, this discussion together with that of 7.1.18
yields the following result:
Proposition 7.1.28 Let L/K be a finite Galois extension of local fields
with n = [L : K]. Let D denote the division algebra of index n over K and
with Hasse invariant, s E (Z/n)'. Then
(i) Z[G(L/K)] Z[NDL.] INDL' is cohomologically trivial and
(ii) the 2-extension
L' -- Z[G(L/K)] Z[NDL.i INDL" ---> Z[G(L/K)] E> Z,

obtained by splicing together the sequences of 7.1.24 and 7.1.27, is represented by inv-1(s) E H2(G(L/K);L').
312
7.1.29 c2(L/K, U)
We are now in a position to describe the local Chinburg invariants

(cf. Chinburg, 1985).
Suppose that
L*--->A-->B-*Z
is a 2-extension of Z[G(L/K)]-modules in which A and B are cohomologically trivial and L/K is a finite extension of p-adic local fields, as
in 7.1.7. The module, A, cannot be chosen to be projective since it will
contain torsion elements and, furthermore, it is not finitely generated.
The module, B, may be chosen to be torsion-free and finitely generated
and, being cohomologically trivial, will therefore be projective. In 7.1.28
B = Z [G(L/K )], for example.
Suppose also that this extension is classified by
inv-1([L : K]-1) E H2(G(L/K);L`).
If U c L' is a Z[G(L/K)]-submodule which is cohomologically trivial

then the induced map
H2(G(L/K);L') -> H2(G(L/K);L'/U)

is an isomorphism and consequently the sequence
L' / U --> AI U --+ B --) Z

corresponds to inv-1([L : K]-') E H2(G(L/K); L'/U) = Z/n.
Suppose, in addition, that L"/U is a finitely generated Z[G(L/K)]module, then we may assume, by 7.1.28, that A/U and B are finitely
generated (and hence projective, if torsion free) Z[G(L/K)]-modules. In
any case, a finitely generated, cohomologically trivial Z[G(L/K)]-module,
X, has a finitely generated projective resolution of the form
0 ---). P1 -> Po
)X
)0
and therefore defines a class
[X] = [Po] - [P,] E W2'(Z[G(L/K)])
Therefore we have classes [A/U] and [B] in W2(Z[G(L/K)]), and we

may define a class
7.1.30
KI(L/K, U) = [A/U] - [B] E c$(Z[G(L/K)]).
313
Proposition 7.1.31 S2(L/K, U) in 7.1.30 is independent of the choice of A

and B in the 2-extension
L*
> A - B -f Z.
Proof Let
L`-_A'-+B'--)Z
be another 2-extension, with B' finitely generated and A', B' cohomologically trivial, which represents inv-'([L : K]-I) E H2(G(L/K);L').
Therefore there is a commutative diagram resulting from the equivalence
of these two 2-extensions:
L'/U 1
L'/U
A/U - B
a
Z
1
13
A'/U '- B'
We may form a chain complex of Z[G(L/K)]-homomorphisms
0>A/U i BBA'/U- B'->0.

It suffices to show that this sequence is exact, since exactness implies
the relation
[A/ U] + [B'] = [A/ U B'] = [B A'/ U] = [B] + [A' / U] E W2(Z[G(L/K )] ).
However, at the left, if (a, a)(z) = (0, 0) then z E L' / U c Al U and then
0 = a(z) = z. Also, at the right, if b' E B' we may choose b E B such that
e(b) = e'(b') and then b'- fl(b) E ker(e') = im(a'). Finally, if b E B and
x E A'/U satisfy /3(b) = a'(x) then e(b) = F (a'(x)) = 0 so that b E im(a)
and b = a(y). Therefore
a'(x - a(y)) = /3(b) - a'(a(y))

= /3(b) - fl(a(y))
=0
314
so that x = a(y) + w for some w E L' / U and

(a, a)(y + w) = (a(y) + a(w), a(y) + a(w))
= (a(y), a(y) + w)
= (b, x),
which establishes exactness in the middle.
Proposition 7.1.32 Let k be an integer such that
HCF(k,#(G(L/K))) = 1.
Suppose that there is a commutative diagram of 2-extensions, as in 7.1.29,
N A/U
L'/U
iB
Z
k
L'/U
A'/U
+ B' -
where the right-hand map is multiplication by k. Then
[A/U] - [B] = [A'/U] - [B'] + S(k) E c9s(Z[G(L/K)]),

where S(k) is the Swan module of 4.2.44.
Proof In this case the diagram chase in the proof of 7.1.31 yields an
exact sequence
0-->A/UBA'/UB'-pZ/k--->0.
Each of these modules is cohomologically trivial so that
[A/ U] - [A'/ U] - [B] + [B'] - [Z/k]

vanishes in the class-group. However, by 5.3.9,
S(k) = [Z[G(L/K)]] + [Z/k] = [Z/k] E c'2'(Z[G(L/K)])

and the result follows.
315
Alternatively, if the reader prefers to work only with projective modules

the result may be proved by the use of Schanuel's lemma (Swan, 1960, p.
270) to splice the above sequence together with
0
kZ[G(L/K)] --> S(k) -> Z/k
)0
to eliminate the Z/k's.
Corollary 7.1.33 If L' -* A' -+ B' -> Z is a 2-extension representing

inv '(k/[L : K]) E H2(G(L/K);L') and U c L* is as in 7.1.29 then
[A'/U] - [B'] = S1(L/K, U) - S(k) E cS(Z[G(L/K)]).
Proof By the classification of 2-extensions (Hilton & Stammbach, 1971,
p. 206) one knows that there exists a diagram, as in 7.1.32, in which the
upper row represents the generator, inv-1([L : K]-').
Corollary 7.1.34 Let f : G(L/K) x G(L/K) --* L' be a 2-cocycle representing the 2-extension of 7.1.8. Let D denote the division algebra, Vf, of
7.1.25, then
S2(L/K, U) = [(Z[G(L/K)] Z[NDL i I NDL*)/U] E WY(Z[G(L/K)])
Proof By 7.1.20 and 7.1.33 the module in question represents
12(L/K, U) - S(-1)
but S(-1) - Z[G(L/K)] and therefore represents the trivial class in

WY(Z[G(L/K)]).
Proposition 7.1.35 Let L/K be a tamely ramified, finite Galois extension
of p-adic local fields. Then, for i >- 1,
H`(G(L/K); UL) = 0,
where Ui = 1 + lr7(9L c (9i denotes the group of units of level n.
Proof By 7.5.2,
10 ifi>1,
K
if i=0.
316
However, if Go 1 G(L/K) denotes the inertia group then Go is cyclic of

order prime to p and there is an extension of the form
Go -* G(L/K) -) G(L/K).
There is also an isomorphism of G(L/K)-modules of the form
ULn /Ui+1 = L,
where G(L/K) acts via G(L/K). There is a spectral sequence (Snaith,

1989b, p. 23) of the form
Es't = HS(G(L/K); Ht(Go; L)
However,
Hs+t(G(L/K);
L)
since L consists of p-torsion and HCF(#(Go), p) = 1,
HS(Go; L) = 0 ifs > 0, by Snaith (1989b, p. 12) and Hilton & Stammbach
(p. 228).
Therefore EZ't = 0 except when (s, t) = (0, 0).
This implies that
H3(G(L/K);L) = 0 if j > 0. To see this we need to know very little about spectral sequences. In fact we need only that Hi(G(L/K);L)
has a (finite) filtration whose sth composition factor is E and that
the latter is computed from EZ't = 0 by successive operations of taking
homology subquotients. Hence all the composition factors are clearly
zero when j > 0.
From the long exact cohomology sequences of the short exact sequences
0 --> Ui/ Ui+l -' UL/ Un+1 -' Ui/ Ui
)0
we see, by induction, that Hj(G(L/K); UL/ UL) = 0 whenever j > 0 and

m > 1. Finally, if j > 1,
H3(G(L/K); UL) = limHj(G(L/K); UL/ UL) = 0.
Proposition 7.1.36 Suppose in 7.1.35 that L/K is unrami led. Thus G(L/K)
(= G(L/K)) is cyclic of order d, generated by the Frobenius, F. There is
an exact sequence of Z[G(L/K)]-modules of the form
0 -+ Z[G(L/K)] ) Z[G(L/K)] -" >

where K = Fq.
)0'
317
Proof Let i; be a generator for the cyclic group, V. Define q by 1(1) =

so that q is clearly surjective. Now suppose that Ed I a,F` E ker(q) so
that
rd
KE,_, aiq'
=1
in L*. Hence
d
a;q` = 0
(mod (qd - 1)).
For each i > 0,

F`
- q' _ (F -
q)(Fi-1 + qFi-2 +
... + qi-') E (F - q)Z[G(L/K)]
and in particular (1 - qd) E (F - q)Z[G(L/K)]. Hence
Ed
1 a+F`
Ed
1
ai(F` - q') (mod (F - q)Z[G(L/K)])
(mod (F - q)Z[G(L/K)]).
Corollary 7.1.37 In 7.1.36
0 = [L*] E W'(Z[G(L/K)]).
Proposition 7.1.38 Let L/K be a finite, unramiled Galois extension of
p-adic local fields then
0 =1(L/K, UL) E cf(Z[G(L/K)]).

Proof By 4.2.48 the Swan modules, S(k), are trivial in '2'(Z[G(L/K)]).
Hence, by 7.1.33, we may choose any 2-cocycle,
f : G(L/K) x G(L/K) -> L',

whose class in H2(G(L/K);L') represents a division algebra, D, and then
12(L/K, UL') = [(Z[G(L/K)]
INDL')/Uf] E c'18(Z[G(L/K)])
As a Z[G(L/K)]-module
L*/Uf = L x Z,
where G(L/K) acts trivially on Z. From 7.1.27 there is a short exact
sequence of the form
0 -*L xZ--pZ[G(L/K)]Z[NI IN--->IG(L/K) )0,
318
where N = NDL' and the middle module represents SZ(L/K, Ui) in the
class-group. By 7.1.37, we may divide out by L* to obtain
0 -> Z -+ Z[G(L/K)] Z[N,] IN1 --> IG(L/K) -i 0,

where N1 = N/L'. We must evaluate the middle module of this exact
sequence.
However, this sequence is obtained from the extension

7.1.39
Z->N1-->G(L/K)^='Z/d
in the same manner as was used to obtain 7.1.27 from 7.1.26. We also
have cohomology isomorphisms
H2(G(L/K);L")
H2(G(L/K);L'/UL)
H2(G(L/K);L* x Z)
H2(G(L/K);L) H2(G(L/K); Z)
H2(G(L/K); Z)
Z/d
so that the class of the central extension of 7.1.39 coincides with the class
of [f] in H2(G(L/K);L*) and therefore is equal to an element of (Z/d)'.

This means that N1 = Z and that 7.1.39 is equal to
Zd i Z -- Z/d.
In this case it will suffice to construct an isomorphism of the form
W : Z[Z/d] - Z[Z/d] Z[Z] IZ.

Let z generate N1. Define W by the formula
W((X)=a(z-1) (aEZ[Z/d]).
Therefore the composition
Z[Z/d] -- Z[Z/d] Z[Z] IZ -+ IZ/d
319
is given by sending a to a(z - 1). The kernel of this surjective map is
< 1 + F +... + Fd-1 >. However,
W(l+F+...+Fd-1) =(1+F+...+Fd-1)(z- 1)
=1(1+z+...+zd-1)(z-1)
=1(zd-1),
which is the generator of the kernel of Z[Z/d] Z[Z] IZ -p IZ/d,
thereby showing that W is an isomorphism.
7.1.40 The tamely ramified case
The majority of the remainder of this section will be devoted to showing

that fl(L/K, UL) vanishes when L/K is tamely ramified. This result
(Theorem 7.1.56) was first proved by Chinburg (1985, section VI). Our
proof differs from that of Chinburg (1985) and is based upon 7.1.34. Of
course, Theorem 7.1.56 generalises 7.1.38, which I have included because
it seemed worth while to start with a simple example.
Let us begin by setting up a presentation for the Galois group of a

tame local extension. Let a E Go be a generator and let b E G(L/K)
be an element whose image in G(L/K) is the inverse Frobenius, F-1.
Assume that K = Fq and that L = Fqd. Then
7.1.41
G(L/K) = {a, b I ae = 1, bd = a`, b-lab = a9}.
Let f : G(L/K) x G(L/K) --> LR be a 2-cocycle whose class in

H2(G(L/K);LR) represents the division algebra, D, with Hasse invariant
equal to 1/de = [L : K]-1. The calculation of 7.1.25 amounts to showing,
in the classification of group extensions by H2(G(L/K);LR), that [f] also
corresponds to the extension of 7.1.26:
LR -> NDLR -p G(L/K).

-R
For the remainder of this section set N = NDLR/Ui. Let l; generate
L and let Tr denote the image of 1[L in
L*/Ui=LR xZ=< ,7C>.

Note that this module is not a sum of L and Z, as Z[G(L/K)]-modules,
unless L/K is unramified.
We have a short exact sequence of Z[G(L/K)]-modules
320
7.1.42
0 -* L* x Z -) Z[G(L/K)] ZEN] IN
z. I G(L/K)
)0
and we must evaluate the class of the middle module in WY(Z[G(L/K)]).

For this purpose define a homomorphism
W : Z[G(L/K)]zl Z[G(L/K)]z2 --f Z[G(L/K)]
Z[N]
IN
by the formulae:
W(zl)=1(ua-1)
7.1.43
I W(z2) = 1 (ub - l)
in the notation of 7.1.25.

Define elements R, S and T by the formulae:
R = (1 + a +... + ae-1)zi,
7.1.44
S = (b(1 + a +... + a9-1) - 1)zl - (a - 1)z2,
T=(1+b+...+bd-1)z2(1+a+...+ac-1)zl.
Lemma 7.1.45 The elements R,S and T lie in ker(AW).
Proof We have
.1W(R) = t((1+a+...+ae-')0 (ua-1))
=(1+a+...+ae-1)(a- 1)
=ae
= 0,
321
),W(S) = 2((b(1 + a +... + a9-1) - 1) (ua - 1)
-(a - 1) (ub - 1))
_ (b(1 + a +... + a9-1) - 1)(a - 1) - (a - 1) - (a - 1)(b - 1)
=b(a9-1)-a+l-ab+b+a-1
=baq - b - ab+b
= 0
and
AW(T) = .1((1 + b +... + bd-1) (ub -1)
-(1 + a +... + ac-1) (ua - 1))

=(I + b + ... + bd-1)(b - 1) - (1 + a +... + ac-1)(a - 1)
=bd-1-(ac-1)
= 0,
as required.
Lemma 7.1.46 If z E ker(ip) then
z=fR+yS+mT
for some /3,y E Z[G(L/K)] and m E Z.
Proof Write z = /31z1 + y1z2 and write yl in the form
d-1
Y1 = EYI,jbj
j=0
with yl,j E Z[Go]. The image of 2,W(z) in IG(L/K) is equal to

d-1
0 = E e(Y1,j)bj(b -1)
j=0
322
where e : Z[Go] -- Z is the augmentation homomorphism. Hence there

exists an integer, m, such that
e(Y1,j) = m
for all j = 0,. .. , d. Subtracting mT from z, we may suppose that

e(y1j) = 0 for all j = 0,..., d. Therefore
Y1 jbi = byi,1(a - 1)
and we may add

d-1
E b'Y11j
to to z to give an element of the form fjz1. Clearly, for an element of this

form to be in ker(W), N1 must be equal to a multiple of (1+a+...+ae-1),
Next we must evaluate W(R), W(S) and W(T) in terms of the 2-cocycle,
) L* x Z.
f : G(L/K) x G(L/K)
Lemma 7.1.47
e-1
W(R)_11 f(a3,a)EL* xZ.

i=o
Proof By definition
W(R) =(1+a+...+ae-')
= 1:i=o 1 (ua; ua - uaj )
= E j_I 10 (f (a,, a)uaj+t - 1) - 10 (ua/ - 1).

However, as in 7.1.25, if y E L x Z and g E G(L/K) then
1 0 (yug - ug)
= g ug1(y - 1)ug
= g 0 (g-1(y) - 1)
=10(y-1)
323
so that
W(R)
=>j=01(f(aJ,a)-1)+10 (uaj+t-1)-10 (ua;-1)

_ yi=o 1 (f (al, a) - 1)
which is the image of jl;-1 f (aJ, a) E L` x Z.

Lemma 7.1.48
q-1
W(S)=
fJ f(bai, a) f (a, b)-'.

=o
Proof By definition
W(S)
_ Ej=o bai
(ua - 1) - 10 (Ua - 1) - (a - 1) (ub - 1)
_ Ei=a 10 (f(ba',a)uba;+, - Ubaj) - 10 (Ua - 1)

-1 (f(a,b)Uab - 1) + 10 (ub - 1) + 10 (Ua - 1)
Eq=0
10 (f (baf, a)Uba.i+I - Ubai+l )
rq-1
+ u1=0 10 (Ubaj+l - Ubaj)

-10 (f(a,b)Uab - Uab) -
(Uab - 1) + 1 (Ub - 1)
=1((jl j=o f (ba', a)) - 1) + 10 (Ubaq - Ub)
-1 0 (f(a,b) - 1) - 1 (Uab - Ub)
= 1 0 ((rjj=0 f(baf,a))f(a,b)-1 - 1),

as required.
Lemma 7.1.49
W(T)
d-1
c-1
11 f (b1, b)
fJ f (as, a)
i=a
s=o
)-1
E L x Z.
324
Proof By definition
p(T) =Ed=obi(ub-1)-Es=oas(ua-1)
_ d=0 1 (f(bJ, b)ubj+l - ubj )
2s =o 10 (f(as, a)uas+i - uas )
Ed=o 1 (f (bJ, b)ubj+l - ubj+, )

+Ed=0 1 (ubj+1 - Ubj)
Esc=-o 1 (f (as, a)uas+i - uas+i )
- .s=o 10 (U&+1 - uas )
= Ed=0 10 (f(bi,b) -1) + 10 (Ubd -1)

EC-1
(f (as, a) - 1) - 10 (uac - 1)
s=o
and bd = ac, which yields the required formula.

7.1.50
7.1.21.
Now we must evaluate f (z1, z2) E Lt x Z using the method of
Suppose that
a(7r)
=uEL cL xZ.
Thus u is a primitive eth root of unity in L* and a(u) = u since a E Go.

There exists y E K,,, such that
u = y9d-1
Fo(Y) E L'
Y
and
Fo(YS)
Ys
_ = as(ir)
it
325
Hence, in the notation of 7.1.22, ifs >- 0

i(as)
= (as(s)n-1,1,...,1) E dLo
= (Fo(Ys)Y-s, YSY-s, ... , ysy-s)
= (1 (9 F)/ 1(ys,ys,...,ys).
Here we have used the fact that a E Ker(hl) in 7.1.13 so that a acts
component by component on i L
Therefore
(ys ys
, YS)
if s *0
I([as)) _
if s =0.
1,..., 1)
(1,
Therefore
)as
f (as, a) = as
((a +
as
j(as+l )
if she-1,
if s=e-1.
(ye,ye'...'ye)
From 7.1.47 we obtain the following result:
Lemma 7.1.51 W(R)=yeEL cL xZ.

Proof The embedding of L` into (L K K,,,)* = iLo sends ye to
(ye, ye
. , ye) since (ye)gd-1 = ue 1 implies that y is a root of unity which
lies in W c L.
7.1.52
Now suppose that

b(n)
It
=u1 EL
then, as in 7.1.50, there exists yl E K,r such that

u1 = yld-1 = Fo(yi)
Y1
326
In the notation of 7.1.12 and 7.1.13, b = (F,1) (F-1, b) and (F-1, b) E

Ker(hl). Hence
b(n,1,1,...,1) _ (F,1)(F-',b)(n,1,..,,1)
_ (F,1)(b(n),1,1,...,1)
_ (1,b(n),1,...,1),
by 7.1.15. Therefore
i(b) = (n-l, b(n),1,1, ...,1),

while
((10 F)/1)(b(n)yi) Y1,...,Y1,Y1)
= (Fo(yi),b(n)Y1,...,Y1)(b(n)-lyi 1,...,y1 1)
= (ulb(n)-1, b(n),1,1,...,1)
= (n-1,b(n),1,...,1),
so that
j(b) = (b(n)yi,Y1,...,Y1)
Similarly,
j(ab) =
and
7.1.53
f (a, b) = [a(b(n)Y1, Y1, ... , Y1)] (ab(n)vvi,vYi)...,vvi) = 1
since a acts trivially on K,,,.

Lemma 7.1.54
W(S) = 1 E L X Z.
Proof Combining 7.1.48 and 7.1.53 we find that
W(S) =
n,-o f(baJ,a)
.
11rjq-1 bal
.1=O
' a ' baJ

j(ba+ )
_ (Ilj-o b(j(a)))j(ba9)-lj(b),
since a acts trivially on j(a) = (y, y, ... , y) E iK,r.
327
However,
j(b) = (b(iv)Yj,yI,...,Yl) and j(baq) = j(ab) = (ab(it)YYi,YY1,...,YYi) by

the preceding calculation. Also
b(j(a))q
= b(yq,Yq,...,Yq)
(F,1)(F-i,b)(Yq,Yq,...,Yq)
_ (F, 1)(y, y,..., y)

_ (Fo(Y),Y,Y,...,Y)
_ (uY,Y,Y,...,Y)
Therefore
IV(S)
(uy,y,y,...,),)(b(it)y1,y,,...,yj)
(ab(n)yy1,yyi.....yy, )
= (ub(ir)ab(ir)-i 1,1,...,1)
= 1,
since ab(rz) = a(uiir) = ului = ub(it).
Lemma 7.1.55 For some a,
ip(T)=(
t)EL*xZ.
-t
Proof We will write x - y if x, y E L x Z have the same second

coordinate.
By 7.1.49,
w(T) - jl jd_o f(b', b)
= (lj=o
bJ(b(it)Yi,Yi,...,Y1))j(bd)-1
(llj=o bJ(b(ir),1,...,1))j(a`)-1
(f-o(F1,1)(F-1,bJ)(b(n),1,...,1)).
328
However
(F',1)(F-i,bi)(b(it),1,...,1)
(Fi, l)(bi+1(7r),1,...,1)
-(1,1,...,
where it is in the jth coordinate. Therefore
W(T)-(n,it,...,n),
which is the image of TC E L' in i Lo, as required.
Theorem 7.1.56 Let L/K be a tamely ramified, finite Galois extension of

p-adic local fields. Then
0 = S2(L/K, UL) E c'(Z[G(L/K)]).

Proof Consider the map, W, of 7.1.43. The composition of W with the
projection, A, onto I G(L/K) is clearly surjective, because a -1 and b -1
generate I G(L/K ). Let us examine the kernel of W, which we will prove
to be a free Z[G(L/K)]-module. By 7.1.46 the kernel consists of elements
of the form z = PR + 6S + mT. By 7.1.51, 7.1.54 and 7.1.55
1p(z) = Mye, 0) + (Kam, M) E L* X Z.
yqd-1
Since u is a primitive eth root of unity and u =

then ye is a generator
of L*, which implies that W is surjective. In addition, since Z[G] < R >=
Z[G(L/K)] < R >, 7.1.36 implies that /3R lies in Z[G(L/K)](b-1 - q) <
R >. However
(-b)(b-1 - q)R =
(-b)(b-1-
q)(1 + a +... + ae-1)zl
= (qb - 1)(l + a +... + ae-1)zt
=(1+a+...+ae-1)S.
Hence we see that the kernel of W is isomorphic to Z[G(L/K)] < S >
which is a free module of rank one. Hence we have a short exact sequence
of the form
0 ---> Z[G(L/K)] --p 2Z[G(L/K)] -- Z[G(L/K)]
Z[N]IN->0,
which implies that the class of Z[G(L/K)] Z[N] IN is trivial in

W2(Z [G(L/K )] ).
329
We conclude this section with a result concerning the local Chinburg

invariant, S2(L/K, U) of 7.1.29, which is related to the exercise 7.5.1.
Theorem 7.1.57 Let L/K be a finite Galois extension of local fields and
let d = [L : K] denote the residue degree. Let
d Lo(11 d Lo-0Z->0
7.1.58
denote the 2-extension of 7.1.8. Suppose that J = G(L/M) i G(L/K) is a

normal subgroup. Then taking J -fixed points of 7.1.58 yields a 2-extension
of Z[G(M/K)]-modules which is equivalent to
0-M
(1F)/1
> I Mo -+ i Mo -> Z -
0,
(r = [M : K]) the 2-extension of 7.1.8 associated to M/K.
Proof Firstly we remark that taking the J-fixed points of 7.1.58 yields
an exact sequence, because dLo is cohomologically trivial.
Since the inertia group, I = Go(L/M) a G(L/M), is the kernel of

G(L/M) --> G(L/M) -+ G(L/K) it is a characteristic subgroup and
hence is normal in G(L/K ). Therefore we may compute the J-invariants
of a Z[G(L/K)]-module, X, by means of the isomorphism
V = (XI )JII
Therefore we may reduce the computation to the following two cases:
Case A:
J=I,K=Fq,M=L=Fqd or
Case B :
I = {11, K = Fq, M = Fqr, L = Fqd and d = rs.
In case A, J = I c G(Lo/K) and the J-invariants of 7.1.58 are given

by
7.1.59
0 - > M* = (LJ)
d(Lo)'
i(Lo)* -''-> Z
)0.
However, 7.1.59 is clearly the 2-extension of 7.1.8 for M/K, since Lo =

(KnrL)' = Knr(Lj) = KnrM = MOIn case B it suffices (Hilton & Stammbach, 1971, p. 148) to construct
a commutative diagram of G(L/K)/J-extensions of the form
330
7.1.60
(10 F)II
M.
iMo
iMo
v
-y
M" = (Lj)* - (E)1)i
(10 F)II
(dL;)J
Let b E J map to F-r E G(L/M) then b acts on i Lo as b =

(Fr,1)(F-r,b) where x = (F-r,b) E G(Lo/K) c G(Knr/K) x G(L/K).
Hence, if z1 E Lo then
b(zl,...,Zd) =
(Fr, 1)(x(Zl),...,x(Zd))
= (FO(x(Zd-r+1)),...,FO(x(Zd)),x(Zi),...,x(Zd-r))
Therefore b(zl, ... , zd) = (z1, ... , zd) if and only if

Z1 = FO(x(Zd-r+l))
Z2 = FO(x(Zd-r+2))
7.1.61
Zr = FO(x(Zd))
Zr+1 = x(Z1)
l Zd = x(Zd-r)
Therefore the coordinates, Zd-r+1, , Zd, determine all the other {z,} and
we obtain a diagram of the form of 7.1.60 by choosing
iMo
to have the form
' i L;
7.2 The global Chinburg invariant
331
cb(Zd-r+1, ... , Zd) = (.... Zd-r+1, ... , Zd).
7.1.62
The valuation of Lo induces that of MM in such a way as to make the

right-hand square commute. It is also evident that the left-hand square
commutes once we have verified that the central square is commutative.
However, by 7.1.61,
(1 (9 F)(Zd-r+l, ... , Zd) _ (... , Zd_r, Zd-r+l, .... Zd-1)
and
Zd-r
= x(Zd-2r)
= FO(xs-l(Zd)),
which ensures that (1F)4 _ 0(1F) on (Zd_r+1, ... , Zd) E ; Mo, since
Foxs-1 is equal to Fd-r(s-1) = Fr on Kr and is equal to b-1 = 1 on M.
By definition 7.1.14, this is the manner in which FO acts on Mo.
Throughout this section let L/K be a finite Galois extension of number

fields. In Chinburg (1985) three class-group invariants, )(L/K,i) E
'f(Z[G(L/K)]), were constructed in a cohomological manner which is
analogous to the construction of the local invariant of 7.1.29. In this
section we shall examine c2(L/K, 2) and we shall give a description of
this class, which is originally due to S. Kim (1991); (see also Kim, 1992)
and is also used in S.M.J. Wilson (1990).
For each finite prime, P .1 (9K, choose a prime lying over it, Q q (9L.
Hence G(L/K) contains a subgroup, called the decomposition group of
Q, which is isomorphic to G(LQ/Kp). We shall say that P is tame if

LQ/Kp is tamely ramified (i.e. Gl(LQ/Kp) = 1) and that P is wild
otherwise. By a theorem of E. Noether, if P is tame then (9LQ is a free
(9Kp [G(LQ/Kp)]-module of rank one. Therefore we may choose an adele
(ap)=aefP(9LQ
7.2.1
(the product being taken over finite primes of K with Q being the chosen
prime over P) such that
7.2.2
1 (i) ap E (9LQ and Kp[G(LQ/Kp)]ap = LQ for each P

1
(ii) (9Kp [G(LQ/Kp)]ap = (9LQ for each tame P.
332
By 4.2.17 there are isomorphisms
R7i LR = I
L K Kp
RIP
and
(9L aK (9Kp = fi (9L,, = I nd'
(LQ/xP)((9LQ )
RIP
so that
7.2.3
(i) Kp [G(L/K)]ap = 11RIP LR for each P

(ii) (9Kp [G(L/K)]aP = 11RIP (9LR for each tame P.
We will abbreviate (9Kp [G(LQ/KP)]ap to XQ and set
X = (9x [G(L/K)]a.
This is to be interpreted as meaning that X is the intersection of L

with the product of its P-completions, Xp, where both are considered as
subgroups of the adeles. Hence X is a locally free (9K[G(L/K)]-module
whose P-completion is
7.2.4
XP = (9KP [G(L/K)]ap =
In addition we shall assume henceforth (by replacing X by mX
for a suitable integer, m E Z, if necessary) that the Q-adic exponential

defines an isomorphism
exp:XQ-01+XQc(9iQ
for all wild LQ/KP.
Definition 7.2.6
Since X is locally free, it is cohomologically trivial and
so also is XQ for each Q. Hence X defines a class in '2'(Z[G(L/K)])

and we may define
c (L/K, 2) = [X] +
XQ))
Pwild
in W.(Z[G(L/K)]). Here 52(LQ/KP,1 + XQ) is the local Chinburg invariant of 7.1.29. The assumption of 7.2.5 ensures that 1 + XQ is cohomologically trivial.
333
Proposition 7.2.7 In 7.2.6 fl(L/K, 2) is independent of the choice of X,

satisfying the conditions of 7.2.2(i),(ii) and 7.2.5.
Proof Let P be a wild prime of K and suppose that XQ and YQ are two
choices for the lattice in 7.2.2 such that XQ c YQ. Hence we have, in the
notation of 7.1.34, a commutative diagram of the following form:
INDL')/(1 +XQ) - IG(LQIKP))
LQ/(1 +XQ) - (Z[G(LQIKP)]
IMP
I NDL*)/(1 + YQ)
Lg/(1 + YQ) - (Z[G(LQIKP)]
IG(LQ/Kp)
The vertical maps are surjective and the horizontal sequences are short
exact so that there is an isomorphism
7.2.8
Ker(f3p) = (1 + YQ)/(1 +XQ) = YQ/XQ.
The Z[G(LQ/KP)]-modules of 7.2.8 are finite and cohomologically trivial,

by 7.2.2(ii).
Similarly Y/X is a finite Z[G(L/K)]-module, which is therefore the

sum of its P-completions (Y/X)p = Yp/XP (taken only over the wild
P, since XQ = YQ when LQ/KP is tame, by 7.2.2(ii)). Hence, in
WY(Z[G(L/K)]), we have
[Y] - [X] = [Y/X]

>Pwild [YP /XP]
EPwild[YP] - [XP]
EPwild I ndZ
by 7.2.8
YQ)
EPwild
-S2(LQ/KP,1 + XQ)).
334
This proves the result when X c Y. Given two general lattices, X and
Y, we may find a third lattice, W, which lies in X f1 Y and may apply
the preceding argument to compare the effect of choosing X or Y with
that of choosing W.
Remark 7.2.9
Observe that
When LQ/KF is tame for all P we say that L/K is tame.
O(L/K, 2) = [h] E c2'(Z[G(L/K)])

when L/K is tame.
7.2.10 The analytic class, WL/K
Let L/K be a finite, Galois extension of number fields and let

p : G(L/K) ---> GL(V) be a finite-dimensional, complex representation.
To this data is attached the extended Artin L -function
AK (S, P)
(cf. Snaith, 1989b, p. 253), which is a meromorphic function of the

complex variable, s, satisfying the following properties:
(i) AK(s,PI P2) = AK(s,Pl)AK(s,P2)
(ii)
If K
L c N is a chain of finite Galois extensions and
G(N/K) -> G(L/K) is the canonical map then

AK(s,InfG(L/K))(p)) = AK(s,P)
(iii) If F is an intermediate field of L/K and W : G(L/F) --p GL(W)

is a representation then
AK (s, I
(W)) = AF(s, W)
(iv) If p denotes the complex conjugation of p then

AK(1 - s, p) = WK (p)AK (S, T),
where WK(p) is a complex number of modulus one.

The invariant, WK (p), is called the Artin root number of p. By properties
(i)-(iv)
WK : R(G(L/K)) - C"
is a homomorphism into the unit circle. Furthermore, if p = p then
WK (p) E {1} and if p is an orthogonal representation then WK (p) = 1
(Tate, 1977, p. 130; Snaith, 1989b, p. 289, (3.9) (proof)). Therefore, if
335
p is an irreducible complex representation derived from the underlying

complex vector space of a symplectic representation then WK(p) E {1}.
In terms of these invariants of analytic origins we may define a class
WL/K E c2'(Z[G(L/K)])
7.2.11
in the following manner. We shall use the Hom-description of 4.2.28. Let

E/Q be a large Galois extension of the rational numbers, as in 4.2.8. The
absolute Galois group, nQ = G(Q'/Q), acts transitively on the infinite

places of E. Let v., denote one of these places. Define a homomorphism
WL/K E HomnQ(R(G(L/K)),J"(E))
by the following formula for the with coordinate of WL/K(p), where p is

any irreducible representation:
1
if v is finite
if p is not symplectic
WL/K(p)v =
I a 1(WK(a(p))) if p symplectic, v = a
aEQQ.
I
By construction, WL/K(p) is S1Q-equivariant and therefore represents a
class, WL/K, in 7.2.11. However, we must show that the construction
of WL/K is independent of the choice of vim. In the tamely ramified
case it is possible to define WL/K without making a choice of v., and
the construction of WL/K was introduced in this form by Ph. CassouNogues (1978) and by A. Frohlich. The generalisation of 7.2.11 is due to
Chinburg (1989, p. 18).
Proposition 7.2.12 The class, WL/K, of 7.2.11 is independent of the choice
of vim.
Proof If p is a representation of G(L/K) then, for each prime P d (9K

with Q a (9L above it, we may form the conductor ideal (Martinet, 1977b,
p. 14)
JP(p) = PfKp(PP) 1 (9Kp,
where pp = ResG(L/K) (p) and fKp is the Artin conductor of 6.1.15. The
( 4/KP)
Artin conductor ideal, f (p), is the product of these ideals and its absolute
norm, Nf(p), is the integer given by the product
336
7.2.13
Write
Since
Nf(p) = jjp #((9K,IIP(P))

for the positive square-root of 7.2.13.
is a positive real number the elements
a 1(,/(Nf(a(P)))) = a 1(,,l(Nf(P))) E E-'(vim)
are positive reals at all infinite places
Hence we may define
h E Homo,(R(G(L/K)),J"(E))
by replacing a-1(WK(a(p))) with a

in the definition of
WLIK(p)v. The resulting class is equal to one at all finite places and is
totally positive on symplectic representations at all infinite places. By
Taylor (1984, p. 9); (see also Chinburg, 1989, p. 19, (2.6.1) (proof)) It
is a determinant and represents the trivial element in '18(Z[G(L/K)]).
Hence WL/K may be represented by WLIK h-1. However, when p is
irreducible and symplectic,
a 1(WK(a(P)))
a-1(,/(Nf (a(P))))
WK(P)
,
I(Nf (P))
(Chinburg, 1983, p. 327) so that the a 1(v,,,)-component of this representative for WL/K is independent of a and hence does not depend on the
choice of vim.
7.2.14 The Frohlich-Chinburg Conjecture
Let L/K be a finite Galois extension of number fields. In this case

Chinburg (Cassou-Nogues et al., 1991; Chinburg, 1989; Chinburg, 1985)
has conjectured that
f (L/K, 2) = WL/K E c'1t(Z[G(L/K)])

This conjecture has a considerable amount of supporting evidence. In
the first place it generalises, by 7.2.9, a conjecture of Frohlich which
states, if L/K is tamely ramified, that
[(9L] = WL/K E (g22(Z[G(L/K)])
This case was proved by M.J. Taylor (1981) and is also described in
Frohlich (1983). Furthermore, several wildly ramified examples have
been proved (Chinburg, 1989; Kim, 1991). Also, M. Rogers has proved
that S2(L/K, 2) - WL/K is in the kernel of the map from '2'(Z [G(L/K )] )
to Go(Z[G(L/K)]), the Grothendieck group of all finitely-generated

Z[G(L/K)]-modules. In fact (Chinburg, 1983, p. 327), D(Z[G(L/K)])
7.3 The Chinburg invariant modulo D(Z[G])
337
is contained in this kernel and D. Holland (1992) has shown the stronger
result that
12(L/K, 2) - WLIK E D(Z[G(L/K)]).
This amounts to the evaluation of fl(L/K, 2) in W'(A(G(L/K))), the

class-group of a maximal order. Using the detection mechanism which
was developed in 5.4.52 we will give a different proof of Holland's

theorem in the next section.
The mixture of the algebraic and the analytic aspects of L/K in the
Frohlich-Chinburg conjecture is especially attractive and number theory
owes a great debt to Ali Frohlich for his discovery of this deep connection
and for his considerable contributions to its successful progress.
In this section we will prove a result of D. Holland (1992) which states

that the Frohlich-Chinburg conjecture is true modulo D(Z[G(L/K)]).
This result was mentioned in 7.2.14. As in the previous section, let L/K
be a finite Galois extension of number fields with group G(L/K). Let
E/Q be a large Galois extension of the rationals, as in 4.2.8, and let

A(G(L/K)) denote a maximal order of Q[G(L/K)], as in 5.4.1. Extension
of scalars induces a homomorphism of class-groups
7.3.1
'2'(Z[G(L/K)])
) WY(A(G(L/K))),
whose kernel is D(Z[G(L/K)]). Since we wish to evaluate the image

7.3.2
S2(L/K,2) E WY(A(G(L/K)))
we must discuss means whereby we can detect elements in '2(A(G)),

where G is any finite group. In our calculations we will be interested
in related computations which take place in the Grothendieck group of
locally freely presented A(G)-modules, K0 T(A(G)), of 5.4.7. Recall from
5.4.8 that there is a natural isomorphism
7.3.3
G : Fac(G)
+ K0T(A(G))
and that the kernel of the natural surjection

CG : K0T(A(G))
IKY(A(G))
338
is isomorphic to PF+(G) of 5.4.6. Therefore, with this type of application

in mind, we shall begin by studying invariants of functions which lie in
the group, HomoQ(R(G),5(E)) or in the subgroup, Fac(G), of 5.4.4.
7.3.4
We begin by observing that a function

g E HomnQ (R(G), 5(E))
is determined by its local components in the following manner. For each
integral prime, p, choose a prime, P a OE, which divides p. A nonzero fractional ideal is determined by the class of the idele given by the
generators of its completions. The action of OQ permutes the primes,
P over p, with stabiliser CIQP. Hence we see that g is determined by its
local components
gp E HomoQP (R(G), 5(EP )).
Henceforth we shall usually work in terms of these local components.
Let G be a finite group and let p be an integral prime.

Define SP(G) to be the set of all pairs
Definition 7.3.5
(C, O)o E R+(G)
where C is cyclic and HCF(p, #(im(4))) = 1. Hence, if Cp is the Sylow

p-subgroup of C, we may write C = Cp x C' and 0 factorises through
the projection onto C',
Each such 0 determines an irreducible idempotent, ej, of Z, [C'] (cf.

5.5.12) since representatives of the Galois orbits of homomorphisms,
w : C' -> E', induce an isomorphism of the form
7.3.6
flip : Zp [C] -' IIy, CQP(w).
In 7.3.6 the product is taken over representatives of the c2Q-orbits of the
M.
In terms of characters the idempotent, e4,, induces an irreducible
idempotent of Qp [C] whose character as a Qp-representation is equal to

X(eo) =
COO),
CO
339
where co runs through c2 /stab(4)). That is, X(e#) is the sum of all the
distinct c2QP -conjugates of 4).
Define SP(G) to be the set of pairs (C, e) where C < G is a cyclic subgroup and e is an irreducible idempotent of Zp [C']. All such idempotents
are of the form, e = eo, for some 0. We will often abbreviate X(eo) to
X(e) in the remainder of this section.
Let f P : SP(G) -* f(Qp) be any map. We say that

g E HomOQP (R(G), f(Ep ))
7.3.7
is the canonical factorisation of fP if, for all (C,e) E SP(G),

7.3.8
g(I ndc(I nfc,(X(e))) = fp((C, e)).
If f * is a family of maps of the form
f" _ {fP : SP(G) - JI(Qp)}
7.3.9
we say that
g E HomnQ(R(G),f(E))
is the canonical factorisation of f " in 7.3.9 if, for each p, the p-component
of g is the canonical factorisation of f P in the sense of 7.3.8.
Now we shall introduce the functions, f " as in 7.3.9,

which are associated to Z[G]-modules. Our treatment follows Holland
(1992, section 2) but the reader should also compare that of Burns (1991)
and Frohlich (1988). See also Cassou-Nogues et al. (1991, p. 95).
Definition 7.3.10
Let M, and M2 be Z[G] -modules, which need not be finitely generated.

Let
i:Ml-+M2
be an injective Z[G]-homomorphism with finite cokernel, coker(i). Define
f, = {f"; p prime) in the following manner. If (C, e) E SP(G) we may

form the Zp[C']-modules (i = 1,2)
MPP=(M;Zp)'
and
MPPe
where e is the idempotent of 7.3.5. Define f P(C, e) to be the order ideal

given by the Zr-order
7.3.11
fr(C,e) = #(coker(ic')p e) = #((MzP . e)/(MI" . e)) E J(Qp).
340
(In general, if M is a finitely generated, torsion (9E-module its order ideal

in 5(E) is given by the unique ideal, J i (9E, such that (9E/J and M have
the same (9E-composition factors.)
Following Holland (1992) we shall alternatively write f, as &1,m2

when there is no ambiguity about the map, i : M1 - M2. In addition, if
M1 = 0 and M2 = N is finite we will abbreviate AN N = f, to f N. Should
it be necessary to indicate the dependence upon G we will denote f, by
fG,i and fN by fc,N
Theorem 7.3.12 Let G be a finite group then the family of maps, f, =
{f ; p prime} of 7.3.11 possesses the following properties:
(i) Let j : M1 ---> M2 and i : M2 --> M3 be injective Z[G]-homomorphisms with finite cokernels, then
f, = fi fj = {ff;p prime}.
(ii) For H < G, let NGH denote the normaliser of H in G. If, for each
H < G, there is an Z[NGH]-isomorphism of the form
coker(iH) = coker(jH),
then
fi =fj.
(iii) Suppose that, for each H <_ G, there is a Z[NGH]-isomorphism of
the form
coker(iH) = (coker(i))H,
then
fi = fcoker(i)
In particular, this is true for the injection i : M2 -+ M3 if H1(H; M2) = 0
for allH<_G.
(iv) Suppose that there is a commutative diagram of Z[G]-modules
M1 ' M2
M3
N1
N2
N3
341
and that, for each H < G, the induced sequence
0 --> coker(iH) -+ coker(kH)
) coker(jH)
)0
is an exact sequence of Z[NGH]-modules. Then
fk = fi f;.
Proof We begin by observing that, if (C, e) E SP(G), then C < NGCP and
hence e E Zp[NGCp]. Also (- Z Z,,) is an exact functor.
Therefore, for part (i), we have inclusions M P
Mi P and M2 P -->
M3 P. Consequently there are inclusions M1 Pe --> M1 Pe and M2 e
M3Pe, from which the result follows since
f j(C, e)
= #(M3Pe/Mj,Pe)
= #(M3Pe/M2,Pe) - #(M2Pe1MiPe)
= f'(C,e)f?(C,e).
Assume that i : M --) N and j : M' - N' satisfy

(MP)cl(N'')c = (Mp)c'/(NP)cp
for all (C, e) E SP(G). Since taking idempotents is an exact functor,

applying e to this isomorphism yields an isomorphism of the form
(M'')cPe/(NP)cPe = (Mp)c1el(Np)c1e,

If i : M2 -> M3 is an injective Z[G]-homomorphism then feoker(i) is
given by
f oker(i)(C, e) = #((coker(i)p)e),
which, under the conditions of part (iii), is equal to

#(M3PelM2Pe)
= f'(C,e).
In particular, if H1 (H; M2) = 0 then the exact cohomology sequence

of
M2 - M3 -) coker(i)
starts with
0 -f MZ -> M H
coker(i)H --) H1(H; M2) = 0
342
and we obtain an isomorphism between coker(i)H and coker(iH), as

required.
Applying e to the cokernel exact sequence in part (iv) yields an exact

sequence of finite abelian p-groups
0 -+ coker(iCP)Pe
coker(kCP)pe
coker(joo)pe
)0
for each (C, e) E SP(G). The order of the central group, fk (C, e), is
therefore equal to the product of the orders of the other two groups,
which is f,(C, e) f (C, e), as required.
Theorem 73.13 Let G be a finite group and let f _ {fP : SP(G) -.f(QP)} be as in 7.3.9. Then, if the canonical factorisation of f* exists in
7.3.5, it is unique.
Proof Choose a prime, p, and consider the canonical factorisation of f P :

SP(G) --> J(QP). The difference between two canonical factorisations of
f P will be a canonical factorisation of the constant function whose value
is the ideal, Z. We must show that any canonical factorisation
g E HomnQP (R(G), f(Ep ))
of the constant map is also equal to the constant function whose value
is the ideal, ZP.
Since HomnQ (R(G), f(Ep )) is torsion free, Artin's induction theorem
P
(see 2.1.3) implies that the natural maps, induced by IndG with C cyclic,
yield an injective map
HomoQP (R(G), J(Ep )) -s ED , Hom0QP (R(C), f(Ep ))
cyclic
Therefore we must show that
Z, = g(1ndc(1)) E f(Ep)
for all A : C --> E', where C is cyclic. If C = C. x C', as in 7.3.5, we
7.3.14
may write 2 as the product of 0 : C' -- E* and W : CP -+ E. By

7.3.8, we know that 7.3.14 is true when W is trivial. This is because, if
e = eo,X(e) = X(eo) and 0 are as in 7.3.5, then
ZP = g(Indg(Infc,(X(e))))
g(Indc(I of ,((o(q5))))
= g(I ndc(I nfc(0)))`,
343
where t is the number of distinct homomorphisms in the QQ-orbit of r.
Fix (C, ) E SP(G), as in 7.3.5, and let H < Cp be a subgroup. Since

H x C' is cyclic we have
Zp = g(I ndHxc (I nfc xC
g(IndCG,.C IndCP(1) 0 ))
_ fa g(Indc' xc,(ipa 0 0)),

where Indc (1) _
W for one-dimensional representations,
Cp-'C,,/H--*E'.
By induction on the order of the image of Wa we may suppose that
Zp = g(I ndc,c,(W# (&4)))
for all W# except those {W,} for which #(im(W)) is maximal. However,
these Wa comprise a Galois orbit consisting of
say. Hence
Zp
(,
xC,(lpa, 0
= f1m
S= , g(I ndG
CP
4)))
= g(I ndCpxc (Wa, ))m
so that Z. = g(IndG C,(W. 4))) for all the W's. This completes the
induction step, which starts with W = 1, to show that Zp = g(IndG(W(D 4)))
for all C, 0 and W

Let KoT(Z[G]) denote the Grothendieck group of finite, locally
freely presented Z[G]-modules, whose definition is analogous to that of
KoT(A(G)) in 5.4.7. As in 5.4.8 there is a Cartan map
7.3.15
CG : KoT(Z[G]) --- WY(Z[G]).
In addition, we have the Hom-description of 4.2.28

HomnQ (R(G), J* (E))
Det : (Y(Z[G])
HomnQ (R(G), E') Det(U(Z [G] ))
There is a similar Hom-description (Taylor, 1984, p. 10)

Homo Q(R(G), J'(E))
KoT(Z[G]) =
Det(U(Z[G]))
344
and cG may be identified with the canonical quotient map. Therefore a

finite, cohomologically trivial Z[G]-module, T, defines a class
[T] E K0T(Z[G]),
which is represented by a homomorphism
7.3.16
hT E HomnQ(R(G),J*(E)).
7.3.17
In addition, there is a natural ideal map
1 : J"(E) -> 5(E),
7.3.18
whose kernel is U(E).
The following result relates canonical factorisations with representatives of the image of [T].
Theorem 7.3.19 Let [T] E K0T(Z[G]) be the class of a finite, cohomologically trivial module, which is represented by hT in 7.3.17. Let
f'T = {fT : SP(G) -> 5(QP)} be the family of functions defined in 7.3.10.
Then
I hT E HomnQ (R(G), 5(E))

lies in Fac(G) of 5.4.4 and is the canonical factorisation of f'T.
Proof As in 5.4.7, K0T(Z[G]) is generated by the classes, [T], modulo

relations which come from short exact sequences of such {T}. However,
a short exact sequence of finite, cohomologically trivial Z[G]-modules
remains exact over Z[NGH] upon taking H-fixed points. Hence, by
7.3.12(iv), (T -4 f*.) factors through K0T(Z[G]). Since K0T(Z[G]) is
generated by classes of the form [Z[G]/J] where J 4 Z [G] is a locally
free ideal, we may assume that T = Z[G]/J. In this case J = Z[G]a, for
some idele a E J'(Q[G]) and, by 4.2.28, hT = Det(a). By naturality of the
isomorphism of 5.4.8(i), g = I Det(a) E Fac(G). Furthermore, if p is a
prime, the local components of g are given by
gP = I Det(aP)
by Frohlich (1983, II, lemma 2.1). Hence we may restrict our attention
to the local components
gp E HomnQP (R(G), 5(Ep )).
For (C, e) E SP(G) we must show that

7.3.20
345
Det(c*)(Ind'(Infc,(X(e)))) = #(TQ'e).
Since Rest = Det(a)(IndG(-)) represents ResG([T]) E KOT(Z[G])

(cf. 4.4.18) we may assume that G = C. Furthermore, by 7.5.6,
Det((x)(Infc,(-)) represents [TCP] E KOT(Z[C']) so that we may assume that HCF(p,#(C)) = 1. In this case ZP[C]e c QP[C]e which is the
QP-representation whose QP-character is
X(e) = X(eo) = E (0(q5),

Sty
orbits
as in 7.3.5. Hence the left side of 7.3.20 becomes

Det(a)(X(e)),
which is the order of the finite cokernel of the map induced by a:

a : ZP[C]e - ) ZP[C]e
(cf. 5.2.32(proof)). Finally, the order of this cokernel is

#(ZP[C]e/aZP[C]e) = #(TP ye),
as required.
Proposition 7.3.21 Let N 4 G be a normal subgroup and let i : V -+ W

be an injective homomorphism of Z[G/N]-modules such that W/i(V) is
finite. If fG/Ni has canonical factorisation
{gp} = g E HomnQ(R(G/N),f(E))
then (Z H g(ZN)) is the canonical factorisation of fGi.
Proof Once more we will work with the local components.
Let (C, e) E SP(G) then we must show that
7.3.22
gp((Indg(Infc'(X(e))))N) = fc,i(C,e),
where X(e) is as in 7.3.5. Since
Infc,(X(e)) = 1 X(e) : C = C, x C' -* GLt(QP)

we have, by 2.5.14,
=IndcjcfN((1
X)cnN).
346
Hence
gP((I ndC(I nfc,(X(e))))N)
GIN
= gP(I ndC/CnN((1 X)CnN))

ZP,
if ker(X(e)) 5t C' fl N,
fG/N,i((C/C n N, e)),
otherwise,
where e E Z,,[C/C n N] is the image of the idempotent, e.

However, if X = W or V, then XP = Xp /cPnN since N acts trivially
on X. Therefore
XPC"e =
Xc/cnN_
P
which is trivial unless 4) (or equivalently, X(e) = X(eo)) vanishes on C'f1N.

These observations easily imply 7.3.22.
D
Proposition 7.3.23 Let H < G be a subgroup and let i : V -> W be an

injective Z[H]-module homomorphism. If fH, has canonical factorisation
{gp} = g E HomaQ(R(H),5(E))
then (Z H g(ResH(Z)) is the canonical factorisation of f; J , where j is

the tensor product map
j = Z[G] Z[H3 i : Z[G] Z[H] V --> Z[G] Z[H] W.
Proof Let (C, e) E SP(G) then we must show that

nfc,(X(e)))) = fc,i(C, e).
By the double cosec formula of 1.2.40, the left side of 7.3.24 is equal to
7.3.24
gP(ResH(I
HZEH\G/cgP(IndHnZcZ ,(z-')*(ResZ HZnc(Infc,(X(e)))))

= UZEHVG/c gP(1 ndHnZcZ (I of HnZc z ((z-1)' (X(e)))))
ZP[zCz-1] and let eZ denote the unique indecomLet z(e) =
E
posable idempotent of Z,, [(H n (zCz-1))'] such that
zez-1
eZ(z(e)) = z(e).
Here, as in 7.3.5,
H n (zCz-1) = (H n (zCz-1))P x (H n (zCz-1))'.
347
Let X(ez) denote the character of the Qp-representation,
Qp[(H n (zCz-'))']ez.
Let mz denote the degree of the àbstract' field extension
Qp [(zCz-')']z(e)lQp [(H n (zCz-'))']ez.
In other words,
dimQP (Qp [(zCz-' )']z(e))
mz
dimQP(Qp[(H n (zCz-'))']ez)'
the ratio of the Qp-dimensions of the local fields which correspond to

the factors in the idempotent decompositions of 7.3.6.
Hence
RestinCz-t(1 (z-')'(X(e))) = mz(1 X(ez))

and we find that the left side of 7.3.24 becomes
11zEH\c/c
X(ez)))'"z
= 11zEH\G/C(fH,i(H n zCz-', ez))"'=.
Similarly, if X = V or W, there is a Zp-isomorphism of the form

(I nd?Gi(Xp))CPe
(XpHnzCz i)P)ez
zEH\G/CZp[C']z(e) ZP[(Hr1zCz-')']e,
from which we deduce that
fc,i(C, e) = 11 (fH,,(H n zCz-', ez))mz,

zEH\G/C
as required.
Example 7.3.25 Let us consider the case in which L/K is a wildly
ramified Galois extension of p-adic local fields with cyclic Galois group,
G(L/K), generated by x of order n.
348
Consider the following commutative diagram:

7.3.26
0- Z
Z[G(L/K)]
fl-
IG(L/K)
0 -. L'/(1 +X) - Z[G(L/K)] Z[N] IN
IG(L/K) - 0
In 7.3.26 /3(z) = (x - 1)z and i(z) = z 0 (ux - 1). The lower sequence of
7.3.26 is that of 7.2.7 (see also 7.1.28).
The map, y, may be assumed to be injective. In fact, we may arrange

that
y(1) = nK E K'/(I +XG(LIK)) - (L'/(1 +X))G(LIK).
7.3.27
We may establish 7.3.27 in the following manner. By the calculation of

7.1.47, a generator for ker((3) of 7.3.26 is given by E"o x' O (ux - 1) and
n-1
n-1
=o
i=o
y(1)x'0(ux-1)=[Jf(x',x)E L'/(1+X),
where f : G(L/K) x G(L/K)
) L* is the 2-cocycle of 7.1.34. We may

alter f by any boundary map, d(h), where h : G(L/K)
) L" is any map.
If h(x) = nL then changing f to f d(h) changes y(l) by a factor which is
equal to
n-1
x'(h(x))h(x')
11
i=o
where NLIK
hx
h(xi+1)
= NLIK(IrL),
: L' - o K' denotes the norm map. Up to units in (9K,

nKÌK
where fL/K is the residue degree. Therefore,

since KK is of infinite order in L' /(1 + X), we may change f so as to
ensure that y(1) has infinite order. In addition, if fLIK = 1, we may
NL/K(irL) is equal to
ensure that y(l) is represented by ltK, in 7.3.27.

Now suppose that f L/K * 1. Local class field theory yields an isomorphism
7.3.28
G(L/K) = K'/(NLIK(L'))
349
By Serre (1979, lemma 4, p. 178), f"_o f (x', x) E L' lies in K* and

corresponds to x E G(L/K) under the isomorphism of 7.3.28. The valuation of K induces a surjective homomorphism, with kernel isomorphic
to G0(L/K),
vK : K*/(NL/K(L*)) - Z/fL/K
Since vK is surjective, we see that y(l) may also be taken to be represented
by nK in the case when f L/K * 1.
Now set T = coker(i) in 7.3.26. Hence T is a finite Z[G(L/K)]-module
which is cohomologically trivial. Since [T] E K0T(Z[G(L/K)]) maps to
S2(L/K,1 + X) E W2'(Z[G(L/K)]) under the Cartan map, we wish to

calculate f'T. By 7.3.12
7.3.29
fT =fl* = fY.
Consider the following commutative diagram.

7.3.30
Z
eL/K
(9L/(1 +X)
vL
L'/(1 +X)
Note that 1 + X c (9i since L/K is wild. Set

7.3.31
M=(q/(1+X)
and let (C, e) E SP(G(L/K )). Therefore

7.3.32
fP(C,e) = #([(L'/1 +X)pe]/Zpe)

#(Mpe)
if x(e)
#(MPe) eLc/K
if x(e) = 1.
1,
L'/(9L=Z
350
The first case of 7.3.32 follows from the fact that e # 1 annihilates trivial
modules such as ZP. When e = 1 we have We = Wp and

VL((L'/I +X)p) = VL(((LC)'/l +XC)p) = eLILCZ,
which yields the second case of 7.3.32 since eLc/KeLILc = eL/K.
Definition 7.3.33 Let L/K be any finite, wild Galois extension of p-adic
local fields. Define a family of functions (cf. 7.3.9)
(f' _ {UP : SP(G(L/K)) --o Jf (E)}
by the formula
eLc/K
if x(e) = 1,
if x(e) # 1.
li P(C, e) =
Hence, with this notation, we have proved the following result:

Lemma 7.3.34 When L/K is cyclic in 7.3.33 then, in 7.3.25,
fT=fM.ff'.
Proposition 7.3.35 Let L/K be any finite, wild Galois extension of p-adic
local fields and let M be as Fin 7.3.31. Set A = (9L/X then
fM = f;f j (fL)-1.
Proof Let RL = ideal {?CL} 4 OL and let m : 1 + X - Oi, a : X --) OL
denote the inclusions. By the hypothesis of 7.2.5 we may assume that
exp induces an isomorphism, if X s RL,
7.3.36
RL/X =(1+RL)/(1+X).
For this value of t, define inclusions ao : RL -- OL, aS : RL --> RL

for s = 1,. .. , t - 1 and at+1 : X --> R. Similarly define inclusions
1
mo
:1+RR1 --> 1+RL fors=1....,t-l and
mt+1 : 1 +X --+ 1 + Rt.

By 7.3.12(i)
t
fA=fa =fjfa,
s=0
351
and
fM = fm =
fm,,
s=0
For each H < G,
= (90,
(H-)* = (L*)H,
((QL)H = (OLx
LH = (L)H
so that, by 7.3.12(iii), for 0< s< t - 1

fa, = fcoker(a,)' fm, = fcoker(m,)
Since X is cohomologically trivial

(RL/X)H - (RL')H/XH = (RLH)/XH
and
(1 + RL/1 +X)H = (1 + Rix)/(1 +XH).

Hence, by 7.3.12(iii),
fa, = foker(a,), fm, = fcoker(m,)
By 7.3.36 and the exponential isomorphisms

Ri/RLs 1 = (1 + RLs)/(1 + RLs 1)
for 1 < s, we find that

t
fA = Hs=ofcoker(a,)
= Icoker(ao)(lls=1 fcoker(m,))
= fl,fM(foker(mo))-1
= fLfM(fi
as required.
Suppose that L/K is as in 7.3.35 with L = Fq , K = Fq. Let p

be a prime and (C,e,) E SP(G(Fq./F9)). Here 0 : C' -* E* has image
of order prime to p. As in 7.3.5, let x(eo) : C' -* GL (ZP) (where
7.3.37
v = dimQP(X(eo))) be the QP-representation associated to .
352
Suppose that C = G(Fq./Fqd) and that n = p"td with HCF(p,t) = 1.

Hence
Cp = G(Fgn/Fqd,), C' = G(Fq-/Fgdp-)
and, by 7.1.36, we have a short exact sequence of C/Cp = G(Fgdt/Fqd)modules

7.3.38
q-F
0 -- ZP[G(Fq^/Fq)]cP
0 Zp[G(Fq.lFq)]ct
(Fg )PCP (Fgdr)p
)0-
The Frobenius, F, generates the cyclic group, G(Fq./Fq), so that there is

an isomorphism
2 : Zp[G(FgdtlFq)] => Zp[G(Fq^/Fq)]cP
given by
dt-1
dt-1
E aiF)
i=0
aiF`
(1 + Fdt + F2dt
+... +
F(p"-1)dt)
i=0
Hence 7.3.38 may be identified with the sequence

7.3.39
Zp[G(Fgdt/Fq)] 9 Zp[G(Fgdt/Fq)] --> (FQd,)p
of Zp[C/Cp]-modules. The generator of C/Cp c G(Fgdt/Fq) is Fd. Hence,

as a Zp[C/Cp]-module,
7.3.40
Zp [G(Fgdt /Fq)]
j=oZp[CICp] < F> >.
Therefore, applying the idempotent, eo, to 7.3.40 yields the d-fold sum
However,
of the Zp[C/Cp]-module given by X(e4,) : C/Cp --i
q - F cyclically permutes the summands, Zp[C/Cp] < F> >. In other

words the injection
Zp[G(Fgdt/Fq)]e, ) Zp[G(Fgdt/Fq)]e.
353
may be identified with

G(
yd,IFq)
IndG(FgdtIF9d)(Zp)
q-F
G(FgdllFq)
) 1ndG(Fgde/Fgd)(Zp
where C/Cp = G(Fgdt/Fq) acts on Zp via x(eo). As in 5.2.33(proof) the

order of the cokernel of q - F is given by the determinant of q - F on
this induced module. In other words,
f
In order to compute f- E*
may proceed in a similar manner, using the
sequence
7.3.41
(Zp[G(Fgn/Fq)])fK/QP P >
(Zp[G(FgnlFq)])f"7Q'
-) L
)0.
This sequence is exact because, by the Normal Basis Theorem, L is

isomorphic to K[G(Fgn/Fq)]. In 7.3.41, fK/QP is the residue degree so
that q = pf K/QP . Hence
f*(C,eO) =
and
7.3.42
7.3.43
f (f)-'(C,eo) = Det(1 -
q-'F)(IndG(LlK)(Infc,(x(eO))))
Now let us consider the general case (i.e. not necessarily cyclic).
Let L/K be a finite, wildly ramified Galois extension of p-adic local

fields with Galois group, G(L/K). We wish to evaluate the image in
W2'(A(G(L/K))) of the local Chinburg invariant, S2(L/K,1 + X). For
this purpose we shall use the following two commutative diagrams:
7.3.44
KoT(Z[G(L/K)])
KoT(A(G(L/K)))
WY(Z[G(L/K)])
WY(A(G(L/K)))
Fac(G(L/K))
354
7.3.45
K0T(Z[G(L/K)])
Fac(G(L/K))
Res
Res
KOT(Z[G(L/F)]) -
K0T(Z[G(M/F)])
Fac(G(M/F))
Fac(G(L/F))
U
FacG(L/K)(G(M/F))
In 7.3.45, we have G(L/F) < G(L/K) and G(L/M) a G(L/F) with cyclic
quotient (via the canonical isomorphism)
G(L/F)/G(L/M) = G(M/F).
By 7.3.19, the vertical maps in 7.3.45 are described in terms of the

canonical factorisations of 7.3.9.
In order to calculate the image of S2(L/K,1 + X) in the group,

FacG(L/K)(G(M/F)) of 5.4.15, we will lift KI(L/K,1 + X) to a finite,
cohomologically trivial module, T, which possesses a class in
Ko T (Z [G(L/K )] )
We will then chase [T] anti-clockwise around 7.3.45 and evaluate the
result in Fac(G(M/F)). Upon doing so we shall find that it lies in the
subgroup, FacG(L/K)(G(M/F)). Unfortunately, during this process we
will express the images of [T] in terms of modules which do not define
elements of '2'(Z[G(L/F)]) or '2'(Z[G(M/F)]) so that we cannot chase
[T] around 7.3.44.
We will begin by considering the following commutative diagram,

which is similar to that studied in the cyclic case in 7.3.26. In particular,
the horizontal rows of this diagram are short exact:
355
7.3.46
a(L/ K) Z[G(L/K)]
fi
(n - a(L/K))Z[G(L/K)]
Ji
1j2
L'/(1 + X) - Z[G(L/K)] Z[N] IN
IG(L/K)
In 7.3.46, a(L/K) = EgEG(L/K) g E Z[G(L/K)] and n = [L : K]. The

upper left-hand map is given by multiplication by a(L/K ). The map, j2i
is the inclusion and jl(z) = >gEG(L/x) z (u1- ug). Hence the map, A, is
determined by jl and, as a consequence of 7.3.27, we shall see in 7.3.49
that 2 (and hence j1) may be chosen to be injective.
Let G(L/F) < G(L/K) be a subgroup of index d = [L : F]. Choose

coset representatives xl,...,xd for G(L/K)/G(L/F). As a Z[G(L/F)]module Z[G(L/K)] is free on the {xi} while IG(L/K) is isomorphic to
IG(L/F)(d 2Z[G(L/F)]). In fact, as a diagram of Z[G(L/F)]-modules
7.3.46 becomes the following diagram (see 7.5.8).
7.3.47
Z-
(r - a(L/F))Z[G(L/F)]
, 1Z[G(L/F)]
(Z[G(L/F)]d-1)
h1
hi
L'/(1 +X)
Z[G(L/K)] Z[M IN
IG(L/F)
where a = (a(L/F), 0,..., 0) , dr = n and

73.48
01(b1,0,...,0)
01(0, ... , o, bi, 0, ... , 0)
= (d 2b1,db1,...,db1)
= (d(r - a(L/F))bi, -a(L/F)bi, ... ,

(n - a(L/F))bi, -a(L/F)bi,... ,
-a(L/F)bi).
(Z[G(L/F)]d-1)
356
In 7.3.48 the term (n - o (L/F))b, (2 < i) is in the ith coordinate.

Lemma 7.3.49
In 7.3.46 and 7.3.47 the map, A, may be assumed to be
injective.
Proof It suffices to show that 2 is injective in 7.3.47 for one choice of

L/F.
Let G(L/F) < G(L/K) be a non-trivial cyclic subgroup with generator,
x, of prime order, r. By 7.5.4 the restriction map
H2(G(L/K);L') ^_' H2(G(L/K);L*/(1 + X))
> H2(G(L/F); L` /(1 + X)) = Z/r
is onto. Hence, if we glue the lower sequence of 7.3.47 to
IG(L/F) F "') Z[G(L/F)] (D F (22)) Z,

where F = d 2Z[G(L/F)] and p is the inclusion, we obtain a 2-extension
L'/(1 +X) --> Z[G(L/K)]
Z[N]
IN - ) Z[G(L/F)] F -) Z,
which represents a generator of H2(G(L/F); L" /(1 + X)). Choose a split

injective Z[G(L/F)]-homomorphism, s : F - Z[G(L/K)] Z[N] IN,
which splits F from the lower sequence in 7.3.47. Dividing out by F
yields a 2-extension
L*/(1 + X) ---> (Z[G(L/K)] Z(NI IN)/s(F) --> Z[G(L/F)] --> Z,
which also represents a generator of H2(G(L/F);L'/(1 + X)).

We have a commutative diagram
(o(L/F), 0,..., 0)
Z
At
IA
L'/(1 +X) -
(Z[G(L/K)]
Z[N]
IN)/s(F)
-. I G(L/F)
in which
r-1
0,0,...,0)) = r - a(L/F) = T(1 - xv) E IG(L/F).

V=1
357
W e may ch oose the representi ng 2-cocyc l e, f : G(L/K) x G ( L/K)
L' ,
so that on G(L/F) it makes the map, y, injective in 7.3.26, as in 7.3.25.

With this choice one sees that
(1) _ v=1 vY(1)

= (r(r - 1)/2)y(1),
which is an element of infinite order in L'/(1 + X), by 7.3.30.
Suppose that G(L/M).i G(L/F) is a normal subgroup such that

G(L/F)/G(L/M) = G(M/F) is cyclic of order m = [M : F]. Hence, in
the notation of 7.3.47, m[L : M] = r = [L : F]. Set c = [L : M] so that
7.3.50
me = r.
We wish to consider the diagram of Z[G(M/F)]-modules which is

obtained by taking the G(L/M)-invariants of 7.3.47. Notice that the
upper sequence of 7.3.47 remains short exact upon applying (-)c(L/M) =
H(G(L/M); -) since H1(G(L/M); Z) = Hom(G(L/M), Z) = 0. Also

H1(G(L/M);L') = 0, by Hilbert's Theorem 90 (Serre, 1979, p. 150),
while
H2(G(L/M);1 + X) = H2(G(L/M);X) = 0.
Therefore H1(G(L/M);L*/(1 + X)) is trivial and the G(L/M)-invariants
of the lower row of 7.3.47 also form a short exact sequence. By 7.5.9, the
G(L/M)-invariants of 7.3.47 form the following commutative diagram in
which the horizontal rows are short exact while the vertical maps are
injective with finite cokernels.
7.3.51
c(m - a(M/F))Z[G(M/F)]
d 1Z[G(M/F)] -
Z
A
(Z[G(M/F)]d-1)
02
!1P2
(1+x(M)) - (Z[G(L/K)] Z[N] 1N)c(L1M) _
IG(M/F)
(Z[G(M/F)]d-1)
In 7.3.51, a' = (a(M/F), 0,..., 0), S = (c(m - a(M/F)),1, ... ,1), X(M) _
XG(L1M) and
358
7.3.52
= (d 2zi,dzl,...,dz1)
4)2(zi,0,...,0)
02(0.... , o, zi, 0, ... , 0) = c(d(m - a(M/F))zi, -a(L/F)zi, ... ,
(dm - a(M/F))zi, -a(L/F)zi,... ,

-a(L/F)zi).
In 7.3.52, (dm - a(M/F))zi lies in the ith coordinate.
The diagram of 7.3.51 may be modified by an isomorphism which
leaves A and 2 unchanged and puts the lower short exact sequence into
a more amenable form.
There are isomorphisms
Ext2[G(M/F)](I G(M/F) (d 2Z[G(M/F)]), M* /(1 + X(M)))
ExtZ[G(M/F)l (I G(M/F), M` /(1 + X (M)))
H2(G(M/F); M'/(1 + X(M)))

Z/ [M : F],
where m = [M : F], in the notation of 7.3.50. By 7.1.57, the lower

2-extension corresponds to the generator, inv-1([M : F]-1). Hence, by
the theory of the cohomological classification of 2-extensions (Hilton &
Stammbach, 1971), we may alter W2 to obtain a new diagram which is
isomorphic to 7.3.51 and has the following form. The details are left to
the interested reader as exercise 7.5.10.
7.3.53
d 1Z[G(M/F)]
c(m - a(M/F))Z[G(M/F)]
(Z[G(M/F)]d-1)
02
1V2
(1+Xmj
W (Z[G(M/F)]d-1)
IG(M/F)
(Z[G(M/F)]d-1)
359
where W = Z[G(M/F)] Z[NI) INI. In 7.3.53 the 2-extension
M`/(1 +X(M)) --+ Z[G(M/F)] Z[NI) INI
IG(M/F)
is obtained from the lower short exact sequence of 7.3.26 upon replacing
L,K,N by M,F,N1.
Before embarking upon the first main result of this section, the evaluation of C (L/K,1 + X) in W'(A(G(L/K))), we require two more straightforward technical results.
Lemma 7.3.54 Let G(M/F) be a cyclic subquotient of G(L/K), as in
7.3.49 and 7.3.50. Define a function
r : SP(G(M/F)) --) 5(E)

by
f1
if 1
x(e),
r(G(M/E), e) =
1 (p - part of [F : K]) if 1 = x(e)

for (G(M/E),e) E SP(G(M/F)).
(i) Then r has a canonical factorisation
r1 E HomnQ(R(G(M/F)),5(E))
given by
rl(V) = ideal < [F : K]dim(V`'(MIF)) >

(ii) In (i)
r1 E PFG+(LIK)(G(M/F)) c Fac(G(M/F)).
Proof Part (i) follows from the fact that, if C = G(M/E),

dim((IndG(M/F)(Infc,(V)))G(MIF)) _< V,1 >C,
When V = X(e), an irreducible QP-representation, this dimension is zero

except when x(e) = 1.
Part (ii) follows from the fact that r1 takes values in . f(Q).
Lemma 7.3.55 Let G(L/M) i G(L/F) < G(L/K) be as in 7.3.49 and
7.3.50. Consider the following short exact sequence of Z[G(M/F)]-modules:
0 -) IG(M/F) J) Z[G(M/F)]/(u(M/F)) - Z/(fMIF)
)0-
360
(i) For (G(M/E) E SP(G(M/F))

fE/F
if
if 1# x(e)
1 = X(e),
fG(N/F),.(G(M/E),e) _
(ii) The canonical factorisation of f G(M/F),.% exists and lies in

PFG(L/K)(G(M/F)).
Proof If 1 * x(e) then e annihilates trivial modules and I G(M/F)Pe

ZP[G(M/F)]e so that fc(M/F)j(G(M/E),e) = 1 in this case.
If 1 = x(e) then f c(M/F) (G(M/E), e) is equal to the p-part of the order
of the cokernel of
I G(M/F)G(M/E) - I G(E/F)
(Z[G(M/F)]/(a(M/F)))G(M/E) = Z[G(E/F)U(o(E/F)),
which establishes part (i).
For part (ii) we first note that, by 7.3.21, f G(M/F),j has the canonical
factorisation
(V,_, s(VGo(M/F))),
where s is as in 7.5.13.
Suppose that V is an irreducible representation of G(M/F), then

either V = VG0(M/F) or 0 = VG0(M/F). In the second case s(V) = 1, which
is certainly a principal fractional ideal of Q(I ndG(L/F) (I nfG(M/F)(V )))
satisfying the total positivity condition of 5.4.45(ii). If V = VGo(M/F)
then V is a one-dimensional representation which factorises through

G(M/F). Hence IndG(L/K)
G(L/F)(InfG(L/F)(M/F) (V)) factorises through G(L/K) and,
since G(L/K) is cyclic,

Q(I ndG(L/F) (I nfG(M/F)(V ))) = Q(V )
Therefore, in both cases, s(V) is a principal fractional
ideal of
dG(L/K)(I nfG(M/F)(V)))
G(L/F)
Q(I nG(L/F)
It remains to verify the total positivity condition of 5.4.45(ii). Dividing
into cases as before we see that we need only consider the case when
V = VG0(M/F) is one-dimensional. However, in this case
I ridG(L/K)
G(L/F) (I rifG(L/F)
G(MIF) (V)) = I ri f G(L/K) (I ridG(L/K) (V))
G(L/K)
G(M/F)
361
is a sum of one-dimensional representations and is symplectic if and only

if I ndG(L/x)
(V) is symplectic.
G(M/F)
We may assume that V is a faithful representation of G(M/F).
In general, if {11 # Z/a < Z/ab and 0 : Z/a --> C` is a faithful,

one-dimensional representation, we may extend 0 to 0 : Z/ab and then
I
0(1 + oa +... + Dab-a).
C'
For this to be symplectic we must have b = 2$ and

b-1
b-1
0-ia-1 E R(Z/2a$).
0ia+1
ii=o
i=o
This condition can only happen if a = 2 but then

cannot be
symplectic because it is the sum of symplectic lines plus 1 + 01fl, which
is not symplectic. This discussion shows that, when V factorises through
G(M/F), Indg(i/F)(InfG(L F) (V)) can only be symplectic if V = W W
for some W E R(G(M/F)) and in this case the total positivity of s(V) is
automatic (cf. Theorem 5.4.46(proof)).
Theorem 7.3.56 Let L/K be a finite, wildly ramified Galois extension of
p-adic local fields with Galois group, G(L/K). In the notation of 7.3.43
and 7.3.44, the image of the local Chinburg invariant, S2(L/K,1 + X), in
the class-group of the maximal order, A(G(L/K)), is given by
O(L/K,1 + X) = [(QL/X] E WY(A(G(L/K))),
where [(9L/X] denotes the class represented by the canonical factorisation
of
f;,/x.
Proof We begin with [T] E Ko T(Z [G(L/K )]) where T = coker(ji) in

7.3.46. By 7.3.19, f T has a canonical factorisation which gives the image
of [T] in Fac(G(L/K)). By 5.4.52, it suffices to find the image of [T] in
FacG(L/K)(G(M/F))
PFG(L/K)(G(M/F))
for each K < F < M < L as in 7.3.50. By the naturality in G of 7.3.19,

the image of [T] in Fac(G(M/F)) is represented by coker(ip2) of 7.3.51.
In other words, we must evaluate the canonical factorisation of fc(M/F),wz
in Fac(G(M/F)). However, by 7.3.12(iv) applied to 7.3.51,
362
7.3.57
f'2 = f2f42-
Suppose now that the composition
z--+L"/(1 +X)-+Z
is multiplication by veL/K. One finds easily, by 7.3.12(i), that ff
where f y" is as in 7.3.26. Therefore, by 7.3.34 and 7.3.35
7.3.58
fz = f*f;.1(X(M))f * (f *)-19' E Fac(G(M/F)).
By 7.5.11, ff has a canonical factorisation which lies in

PFG(L/K)(G(M/F)),
since g(x) E .1(Q) in 7.5.11.
Now let us turn our attention to f;2tr.

If (G(M/E), e) E SP(G(M/F)) and X(e) = 1 then
f012(G(M/E),e)9P(G(M/E),e) is equal to the p-part of
I G(L/K)G(L/E)
(([L:KI-a(L/K))
= [L : E]([L :
[G(L/K)])G(L/E)
eE/F
K][E:Kl-2)eE/F,
by 7.3.10 and 7.3.12(ii). If X(e) * 1 then 8P(G(M/E), e) = 1 and

fop2(G(M/E), e)6P(G(M/E), e)
-#
IG(L/K) e
([L:K]-a(L/K)) p[G(L/K)]e
ZZ [G(L/K)le
_ # (`[L:KI
[G(L/K)]e)
vG
since e annihilates trivial modules.

Consider the multiplication map (see 7.5.12)
[L : K] : Z[G(L/K)] --> Z[G(L/K)].

The above calculation shows that, in Fac(G(M/F)),
{f02'P(f
[L:K))-1}(G(M/E), e)
363
is trivial if 1 * x(e) and otherwise is given by the p-part of

eE/F[L : E]([L : K] [E:K]-2)( [L : K] -[E:K])
K]-2fE/FfE)F
= eE/F[L : E] [L :
_[LE][E
_ [F :
:F][F:K][F:K]-'fJF[L:K]-2
K]-1 [L
K]-1
By 7.3.54, 7.3.55 and 7.5.11, this function has a canonical factorisation

which lies in PFG(L/K)(G(M/F)). Therefore, by 7.3.57 and 7.3.58, we have
shown that [T] is represented in

Fac(G(M/F))/PFG(L/K)(G(M/F))
by the canonical factorisation of

fVM1(XM)1M (f M)-1'
However, G. VV' has a canonical factorisation which lies in

PFG(L/K)(G(M/F)).
This last fact is seen by observing that 7.3.21 and 7.3.42 imply that the
canonical factorisation of fM (fM)-1 is equal to the image of (the p-part
of) the function on R(G(L/K)) given by sending V to
Det
1-
(F I VGo(LIK))
#(K )
which is the value at s = 1 of an Euler factor (above p) from the Artin Lfunction (Martinet, 1977b). Since this function depends only on VG0(L/K)
- as a representation of the cyclic quotient, G(L/K) - one sees that

this function is in PF(G(L/K)) by the argument which was used to
establish 7.3.55(ii). By 5.4.46 the image, fM (fj)-1 E Fac(G(M/F)), lies
in PFG(L/K)(G(M/F)), as required. This completes the proof of Theorem
7.3.56.
7.3.59 We shall now state a result which relates the analytic class, WL/K,
to the canonical factorisation of modules. This result is proved using the
theory of norm resolvents to represent classes in Hom-descriptions and
it is given in Holland (1992, section 4); (see also Wilson, 1990). We shall
not give a proof of this result here, since our Explicit Brauer Induction
techniques have nothing to add to this step in the programme.
364
Let L/K be a finite Galois extension of number fields with Galois

group G(L/K) and let X = 6K [G(L/K)]a be as in 7.2.3. Let b E X be a
free generator for L as a K [G(L/K )]-module.
Theorem 7.3.60 (i) The function, f o,/baK [G(L/K)]' has a canonical factorisation which represents a class
[(' L/b(9K [G(L/K)]] E c2'(A(G(L/K)))

(ii) If WL/K E 11f(Z[G(L/K)]) is the class defined in 7.2.11 then
WL/K = [(OLIb(9K [G(L/K)]] E'2'(A(G(L/K)))
Theorem 7.3.61 Let L/K be a finite Galois extension of number fields
with Galois group, G(L/K). Let
f2(L/K, 2),WL/K E W1(Z[G(L/K)])

be as in 7.2.6 and 7.2.11, respectively. Then
KI(L/K, 2) = WL/K E 12'(A(G(L/K))).
Proof Let X be as in 7.2.3 and 7.2.4. By definition

S2(L/K, 2)
IK)
= IX I + >P wild I ndG(LQ/KP)(n(LQ/KP,1 + XQ))

= 1X I + >P wild I ndG(LQ/KP) [QLQ /((9KP [G(LQ /KP )]aP )]
in'2'(A(G(L/K))). (In 7.2.3 (9K, [G(LQ/KP)]ap was denoted by XQ c

(9LQ, while in 7.3.51 it was denoted by X(LQ).)
Now let b E X be a free generator for L as a K [G(L/K)]-module so

that
7.3.62
[X] = [X/b(9K[G(L/K)]] E 1eY(A(G(L1K))).
By 7.2.2-7.2.4, for each wild prime P,

7.3.63
Ind G(L/K)
G(LQ/KP) [(9LQ l(&Kp[G(LQIKP)]ap)] -((9LIX)aK (9Kp.
7.4 Real cyclotomic Galois module structure
365
Since (9L/X is trivial at tame primes we may combine our formula for
S2(L/K, 2) together with 7.3.62 and 7.3.63 to obtain
c (L/K, 2)
= [X/b(9K[G(L/K)]] + [(9LIX]
= [(OLI b(9x [G(L/K )]]
in '2'(A(G(L/K))), which completes the proof of Theorem 7.3.61.

7.4.1 Based exact sequences
We shall begin this section by establishing some elementary properties of

`torsion elements' (reminiscent of Reidermeister torsion and Whitehead
torsion) which lie in K1(Qp[G]) and are associated to exact sequences
of locally free, based modules. This passage from exact sequences of

based modules to K-theory will be applied below to analyse the Homdescription of the global Chinburg invariant.
Let A be a ring and suppose that
1I:
{0}
B") C-->{0}
is a short exact sequence of A-modules with preferred bases

2CA(*) = {xA(1),...,xA(t)},
xB(*) = {xB(1),...,xB(s)}
and
xc(*) = {xc(l),...,xc(s-t)}
for A, B and C, respectively.
This information yields an element, [II] E K1(A), in the following

manner. Choose a A-homomorphism, 0, from the free A-module on
{yc(i);1 ---), B
such that 4'(yc(i)) = xc(i) for each 1 < i< s - t. Hence we have
366
Lemma 7.4.2
The homomorphism
2+O:AA<yc(*)>) B
is an isomorphism.
Therefore there exists an invertible matrix, U E GLS(A), such that

1
XA(1)
XA(t)
/ XB(1)
=U
(2+0)
Yc(l)
\ xB(S) l
Yc(s - t) I
Definition 7.4.3 Define [II] = [U] E K1(A) = GL(A)ab.

Proposition 7.4.4 [U] does not depend on the choice of .
Proof If 0' is a second choice of homomorphism then we have a

commutative diagram of the following form, in which p = (A + ')-1(A +
4)
A
AA<yc(*)>
AA<yc(*)>
Therefore the difference in the two values of [U], which result from
the two choices, is represented by the matrix of p with respect to the
basis {, A(*), yc(*)}. This matrix has the form
I,
IS_t
which represents zero in K1(A), as required.
Now suppose that A = Z,, [G], a p-adic group-ring. Suppose that V

is a finitely generated, free Z,[G]-module and that X is a projective
367
Z[G]-module together with an embedding, k : Xp = X Z Zp - V. In

addition, let us suppose that V/Xp is a finite p-group. Suppose also that
we have a finitely generated, projective Z[G]-resolution of the form
{0}-->P1 -4PO
Let x1,p(*) = {x1,p(1),...,x1,p(t)} be a basis for Pi Z Zp over Zp[G]
and let xo,p(*) = {xo,p(l),...,xo,p(t)} be a basis for Po Z Zp. Also let
xl,o(*) = {x1,o(1),...,x1,o(t)} be a basis for Pi Z Q over the rational
group-ring, Q[G], and let xo,o(*) = i(xl,o(*)) so that
{xo,0(1),...,xo,o(t))
is a Q[G]-basis for Po Z Q. Let

V1, Vo E GLI(Qp[G])
be the matrices given by (i = 0, 1)

xi,o(1)
x,,p(1)
Vi
xi,o(t)
xi,p(t)
i(xl,p(*))
= i(V1x1,o(*))
Therefore
= Vii(xi,o(*))
= Vlxo,o(*)
= V1Vo'xo,p(*)
so that the matrix of i : Pi Z Zp --> Po Z Zp is equal to

V1V6-' E GL,(Qp[G])
Notice that V1 V6-1 has entries which lie in Z p [G] and that the class,
[Vi] E K1(Qp[G]) is the p-component of the K1-idele which is associated
to the based module, Pi, in the Hom-description construction of 4.2.13
and 4.6.36.
From the projective Z[G]-resolution of V/Xp we may construct an

exact sequence of Zp[G]-modules of the form
{0} -- Pi Z Zp
O'D
(Po Z Zp) Xp
(
(ft,-k)
{0}.
368
Here j fits into a commutative diagram of the following form:

i
PI Oz Zp
PO ZZp
k
XP
The following diagram also commutes:

V
Po Z Zp
V/Xp
Now suppose that X Z Q has a basis, uo(*) = {uo(1),...,uo(r)} over

Q[G] and that Xp has a Zp[G]-basis, up(*) = {up(1),...,up(r)}. Therefore
there is a matrix, W E GLr(Qp[G]) such that
WUO(*) = up(*).
Let V ZP Qp be endowed with the Qp[G]-basis

vo(*) = k(uo(*))
so that vo(a) = k(uo(a)) for 1 < a < r.

Finally, let vp(*) = {vp(1),...,vp(r)} be a Zp[G]-basis of V.
Hence we have an exact sequence
7.4.5
II:
Xp(ft)k)V
PlZZp
to which to apply the construction of [H] in 7.4.4.

Suppose that the matrix with entries in Zp [G] which corresponds to
(i, j) is
(V1Vo 1, J)
so that J is a t x r matrix. Also let k correspond to the r x r matrix with

entries in Zp [G]
K E GLr(Qp[G])
369
so that k(up(*)) = Kvp(*). Let ft correspond to the t x r matrix, P, with

entries in Z,[G].
Now we must choose 0 in the following commutative diagram:
0
There are matrices, whose entries lie in Zp [G], (D, E) such that
O(vp(*)) = (Dxo,p(*),Eup(*))
where D is t x r and E E GLr(Qp[G]). Hence the matrix of

(P1 Z Zp)
(i,.i)+0 :(P1 Oz Zp)ED V

XP
has the form

V1Vo 1
In addition we have
Ir = PD - KE
and
PV1Vo1=KJ.
370
Consider the following matrix computation:

It
It
Ir
It
D \ I ViVo 1
-1r / \
V1Vo1
Ir
P PD - Ir
I,
V1 Vo 1
Ir
P KE
Vi V1 1
0
Jr
PV1Vo-1 KE
Therefore the image of [II] E K1(Zp [G]) in K1 (Qp [G]) is given by the
following result.
Proposition 7.4.6 In K1(Qp[G]) the image of 7.4.5 is given by
[II] = rank(V) < -1 > +[V1] - [Vo] - [K]

where < -1 >E K1(Zp) = ZP denotes the class corresponding to -1 E ZP.
The relation between [K] and [W] is given by the equation

Wuo(*) = up(*),
which implies, upon applying k, that

K-1 Wvo(*) = vp(*).
Therefore 7.4.6 may be rewritten as
371
Proposition 7.4.7 In K, (Q , [G]) the image of 7.4.5 is given by
[H] = rank(V) < -1 > +[VI]
- [Vol + [K-1 W] - [W],
where < -1 >E K,(ZP) ^_-' Zp denotes the class corresponding to -1 E Z.

Remark 7.4.8 Note that the classes [VI], [Vol, [W] E K,(QP[G]) are
precisely the elements which are obtained by comparing the Z,,[G]-basis
to the Q[G]-basis for the modules Pl, Po, X, while [K-1 W] is the class
obtained by comparing the Z p [G] -basis to the Q,, [G] -basis for the module,
V, in the manner of 4.2.13 and 4.6.36.
Now we shall examine the Chinburg invariant of 7.2.6 in the case

of an extension, E/F, of number fields such that, for each wild prime
P a OF, the decomposition group G(EQ/FF) is cyclic. We shall refer to
such an extension as the locally cyclic case. The main result concerning
locally cyclic extensions (Theorem 7.4.35) is rather technical. It gives the
Hom-description of S2(E/F, 2) in locally cyclic cases which satisfy a mild
cohomological condition.
Suppose that L/K is a Galois extension of local fields with cyclic

Galois group, G(L/K), with a chosen generator, g, of order n.
7.4.9
In the 2-extension of 7.1.7 we may choose B = Z[G(L/K)] mapping

to Z by the augmentation map, e, given by e(g`) = 1. Hence we have a
short exact sequence of the form
7.4.10
{1} -> L' -> A
IG(L/K) -p {0},
where IG(L/K) is the kernel of the augmentation map,
e : Z[G(L/K)] -- Z.
Let nL denote the prime of L and let vL : L' --> Z denote the
valuation, normalised so that vL(xL) = 1. Let eLIK denote the ramification
degree of L/K.
Proposition 7.4.11
With the notation of 7.4.9 there exists an injective
Z [G(L/K )]-homomorphism
y : Z[G(L/K)] -) A
such that
(i) 2(y(z)) = (g - 1)z for all z E Z[G(L/K)].
372
(ii) Let o = >"o g' E Z[G(L/K)]. There exists an element, x E K`,

which satisfies VL(X) = eL/K, whose image generates K`/(NL/K(L`)) = Z/n
and such that y can be chosen so that y(a) = x.

(iii) For y, x as in (i) and (ii) there is an isomorphism of the form
L`/ < x >= A/(y(Z[G(L/K)])),

which is induced by i of 7.4.10.
(iv) For any x E K` with VL(x) = eL/K, whose image generates

K*/(NL/K(L`)), L`/ < x > is cohomologically trivial. Furthermore, if
U is as in 7.1.29 and lies in (9i then
S2(L/K, U) _ [L`/ < x, U >] E WY(Z[G(L/K)]).
(v) For any x E K as in (iv), there is a short exact sequence of

Z[G(L/K)]-modules of the form
{1}
> (9L
L`/ < x >) Z/eL/K --> {0},
where v is induced by VL.
Proof This is an elaboration of 7.3.27 and may be proved in the same

manner.
Example 7.4.12
Let p be an odd prime and let K = Qp, the p-adic
rationals. If n denotes a primitive nth root of unity, let L =

for
some 0 < s. In this case a suitable choice for x in 7.4.11 (iv) is given by
x = (1 + P)Pp-1
We shall return to this example later in this section.
7.4.13
Let p be a rational prime. Suppose that E/F is a locally cyclic
Galois extension of number fields and that P a OF is a wild prime, which

lies over p. Let Q < (9E be the chosen prime over P in 7.2.2. Hence we
have XQ c (9EQ and exp(XQ) = 1 + XQ c (9EQ. In the construction of
S2(EQ/Fp, I + XQ) we have the fundamental 2-extension
{1} -+ EQ --+ A --> Z[G(EQ/Fp)] - Z ---> {0},

which represents inv-1(1 (mod [EQ : Fp])). Since
1 + XQ C UEQ
we may apply 7.4.11(iv). Therefore we may choose any uniformiser
373
up E F;
7.4.14
whose image generates FP/(NEQ1FP(EQ)) and then

7.4.15
i2(EQ/Fp,1 +XQ) = [EQ/ < Up,1 +XQ >] E'.(Z[G(EQ/FP)]).
Let Tors(A) denote the torsion subgroup of A.

Lemma 7.4.16 If Wp = Tors(EQ/ < up >) is a cohomologically trivial
Z[G(EQ/FP )]-module then
VP = (EQ/ < up >)/{Tors(EQ/ < up >)}

is a free Z,[G(EQ/FP)]-module whose rank is equal to [Fp : Qp].
Proof If Wp is cohomologically trivial then, by 7.4.11(iv) and the long

exact cohomology sequence Vp is cohomologically trivial and torsionfree over Z,. Hence Vp is a free module, as can easily be deduced from
Serre (1979, theorem 7, p. 144).
Remark 7.4.17 From Washington (1982, lemma 13.27, p. 287) one

can deduce that the cohomological triviality condition is satisfied for the
finite subextensions of the cyclotomic Zp-extension over F(op), where F
is any totally real number field (cf. 7.4.39).
Under the circumstances of 7.4.13-7.4.16, choose a Zp [G(EQ /FP )]-basis
for Vp (np = [Fp : Qp])
7.4.18
yp(1),...,yp(np) E VP.
By 7.4.11(v), we have an exact sequence of the form

{1} --> (9EQ ---) EQ/ < up >-> Z/eEQIFP
If u(EQ) denotes the roots of unity in EQ then (9EQ = (EQ) UE where
UEQ = 1 + ?CEQ (EQ, and EQ = Z < REQ > (9EQ so that the induced
Zp [G(EQ/FP )]-homomorphism
UEQ/(Tors(UEQ)) -> Vp
induces an isomorphism of Qp [G(EQ/FP )]-modules,
{UEQ/(Tors(UEQ))} ZP Qp --p Vp OZ,
374
In addition, we have a commutative diagram

1 +XQ
XQ
exp
exp
U1EQ
nEQ (9EQ
in which the vertical maps are injections with finite cokernels. Therefore
the exponential is an injective map of Z,[G(E/Fp)]-modules
exp : XQ -+ VP,
which becomes an isomorphism when tensored with Q,,.

Suppose that we have a Z[G(EQ/Fp)]-resolution
{O}-*T1->To") Vp/(1+XQ){0}.
As in 7.4.5, we may construct an exact sequence of free Z,,[G(EQ/Fp)]modules of the form
{0} -- Tl Z ZP + (To Z Zp) XQ
Or
)-P) VP
{01
where n' is any lifting of it. As in 7.2.4, identify Xp = X OF (9Fp with

and set P; =
Therefore we obtain a short
exact sequence of free Z,[G(E/F)]-modules
7.4.19
Pi Z
ZP (
Z XP
(P Z P)
(ft, Ixp))
IndG(EQ/FF)
G(E/F) (Vp).
Also, we have a short exact sequence

7.4.20
fl: 11P1ZZP ->

PIP
"Po Z ZpJ) I f Xp I - +f IndG(EQ/Fp)(VP).

PIP
`PIP
PIP
If X is the locally free (9F[G(E/F)]-module of 7.2.3 then, considered

as a Z[G(E/F)]-module,
7.4.21
X Z Zp = (X (90, (OF) Z Zp = X OF
11 OFp
PIP
flXp.
PIP
375
By comparing a choice of Zp[G(E/F)] -basis for 7.4.20 with a choice of

Qp[G(E/F)]-basis for the complex obtained by tensoring 7.4.20 with Qp
one obtains an element (see 7.4.4)
7.4.22
[H] E Ki(Qp[G(E/F)])
In particular, suppose that we use a Q[G(E/F)]-basis as the Qp[G(E/F)]basis for (Pi (&Z Qp) and its image under i" as the Qp[G(E/F)]-basis for
(Po Z Qp). Also, suppose that we use a Q[G(E/F)]-basis for X 0 Q

as a Qp[G(E/F)]-basis for X Qp and its image under exp 0 1 as a
Qp[G(E/F)]-basis for rjpIpIndG(EQ1FP)(Vp). In this case the comparison
of the bases for (rjpIp Pi (&Z Zp) for i = 1, 2 yields elements
[Vi] E K1(Qp[G(E/F)]),
which represent the p-component of the idele in Kj(J*(Q[G(E/F)]))

which corresponds to the class of [Pi] E 2'(Z[G(E/F)]) in the Homdescription of 4.2.13 and 4.6.36. Also, with these choices, the comparison
of bases for X Z Qp gives a representative of the p-component of

the idele corresponding to NF/Q([X]) E'2'(Z[G(E/F)]), where [X] E
'.(&F[G(E/F)]) is the class of 7.2.6.
The class rjpIP wild([Pi] - [Po]) E''2(Z[G(E/F)]) is equal to
Indg(E'
)(f(EQ/FP,1 + XQ) - [WP]),
Pip wild
where Wp is as in 7.4.16.
If 1 is a prime then f2(EQ/Fp,1 + XQ) - [WF] is represented by a
cohomologically trivial 1-group, if P 11, and therefore has trivial p-adic
idelic representative at primes p different from 1. Therefore, from 7.4.6,
we have the following result (since [II] E K1(Zp[G(E/F)]) its determinant
lies in the Hom-description indeterminacy).
Proposition 7.4.23 Let vo(*) be the choice of Qp[G(E/F)]-basis for

G(E4/F
(flIndP)VP)
z Qp
PIP
described above and let vp(*) be a Zp[G(E/F)]-basis for

G(E/F)
H I ndG(EQ/FP)(VP ).
PIP
376
If F is the invertible matrix with entries in Qp[G(E/F)] given by

moo(*) = vp(*),
then
[I] E K1(Qp[G(E/F)])
is a representative of the p-adic idelic component, in the sense of the Homdescription of 4.2.13 and 4.6.36, corresponding to the class of 7.2.6,
S2(E/F, 2) - E IndG(EQIFF)([WP]) E WY(Z[G(E/F)]),

PIp wild
where Wp = Tors(EQ/ < up >) is as in 7.4.16.

7.4.24 Representing fl(E/F, 2)
From 7.4.23 and the discussion of 7.4.17-7.4.22 it is possible to describe
an explicit representative for 0(E/F,2) E W2(Z[G(E/F)]) in terms of

the Hom-description of 4.2.13, in the locally cyclic case when all the
Z[G(EQ/FP)]-modules, Wp = Tors(EQ/ < up >) in 7.4.16, are cohomologically trivial. This representation is obtained merely by amalgamating
7.4.23 with the definitions of 7.2.1-7.2.6 and 4.2.13-4.2.28.
We have to produce a function lying within the group
HomoQ (R(G(E/F)), J* (N))
where N/Q is a Galois extension of the rationals containing E and all

the [E : F]th roots of unity.
Let a E E* be a normal basis element for E/F such that
7.4.25
aEXfl(0E c(XeFF)=E.
Hence E = F[G(E/F)]a is the free module generated by a. Even if the

identity of X is unknown in 7.2.3, one may multiply the element, a, by
a large power of [E : F] in order to ensure that 7.4.25 holds, without
changing the class which we are going to represent in '2'(Z[G(E/F)]).
For each tame prime, P i (9F, and for the chosen Q 1 OE above P we
have ap E (9EQ such that
7.4.26
(9FP [G(E/F)]ap = IIRIP OE,,-
Also
7.4.27
Fp [G(E/F)]a = E OF FP = IIRIP ER
377
so that there exists Yp E Fp [G(E/F)]* such that

Ypa = ap E RRIp ER-
7.4.28
Hence, since ER E N=11sIPNs,

7.4.29
det(Yp) E Homo, (R(G(E/F)), r1s,p Ni).
Write J,*,,me(N) for the restricted idelic product involving only those
completions, Ns, for S a ON which lie above tame primes of F. Hence
Jtame(N) is an f2Q-module. We have a norm map (Frohlich, 1976; Curtis
& Reiner (1987, section 52.25, p. 341)
NF/Q : HomnF(R(G(E/F)), fsIp Ns)

-+ Hom0Q(R(G(E/F)),
Ns)
given by NF/Q(f) = H E0Q/nF g(ff

Therefore we obtain
7.4.30
(g-'
. -))
NF/Q({det(Yp)}p tame) E HomoQ(R(G(E/F)),Jtame(N)).
Let a E X be the normal basis element of 7.4.25 and let z(1),...,z(t) E

OF (t = [F : Q]) be a Q-basis for F. Hence z(*)a = {z(1)a,...,z(t)a} is a
Q-basis for X Z Q. From 7.4.21 we have an isomorphism
7.4.31
AP :X ZQp(11P,pXP)Z,QP
whose P-component is characterised by

(,.p(z(i)a))p = a z(i) E X oF OF, = Xp.
From 7.4.20 we also have a map of Zp[G(E/F)]-modules

7.4.32
rlP,pIndG(EQ/FP)(exp) : (flp,pXp)
- > [J I ndG(EQG(EQ/FP)(Vp ).
PIP
If yp(1),...,yp(np) E Vp is the Zp[G(EQ/Fp)]-basis then yp(i) 0 1 E

I ndG(EIF)
(VP) and
G(EQ/FP)
(yp(*);P I P) _ {yp(1) 1,...,yp(np);P I p}
378
Note that
is a Zp[G(EQ/FP)]-basis for
E rankZP(VP) = E [FP : QP] = [F : Q] = t,

PIP
PIP
as expected. Up to multiplication by a permutation matrix, we obtain a

matrix
BP E GLt(QP[G(EQ/FP)])
which satisfies
BP
(HIndG(EQ/FP)(exP(AP(z(*)afl) _ (yp(*);P I P).

PIP
By the construction of 4.2.13-4.2.28 we obtain a homomorphism

7.4.33
Det(BP) E HomcQ(R(G(E/F)), rjsIP NN).
If BP is varied by multiplication by a permutation matrix, a, then Det(BP)

is changed by multiplication by the function,
TF-+
sign(a)dim(T),
which lies in the subgroup Det(ZP[G(E/F)]*).

Let JWild(N) denote the product of the Ns for primes, S c ON lying
over wild primes of F. Hence
(ri Ns
JW ild(N) =
P
SIP
where p is a rational prime lying beneath a wild prime P 1 OP. Similarly,

write JA(N) for the product of the Archimedean components of J*(N)
so that
J*(N) = JWild(N) X Jtame(N) X JA(N).
With this notation 7.4.33 yields a homomorphism

7.4.34
Det(BP) E Homo, (R(G(E/F)),JW;Id(N)).
Theorem 7.4.35 Let E/F be a locally cyclic Galois extension of number

fields for which, in 7.4.16,
Wp = Tors(EQ/ < up >)
379
is a cohomologically trivial Z[G(EQ/FP)]-module for each wild prime, P a

(9Fp
Define
sE/F E HomnQ (R(G(E /F)), J' (N))
to be the function whose coordinates are trivial in JA(N) and are given by
7.4.30 in Jl*ame(N) and by 7.4.34 in
If [sE/F] is the class given by
the Hom-description of 4.2.13 then
40id(N).
IndG(EQ/FP)([WP])
[SE/F] = S2(E/F, 2) -
/
7
E c'2'(Z[G(E/F)])
P wild
7.4.36 Cyclotomic examples
Let p be an odd prime and let ,, denote a primitive nth root of unity.
and let x = (1 +
For each s >_ 0 let Ls+1 =
E Q. Note
that x is a uniformiser for Qp with the property that LS+1 / < x > is
cohomologically trivial, as in 7.4.11. Identify G(Ls+1/Qp) with (Z/ps+l)'
in the customary manner. Set
7.4.37
Wp(s+ 1) = Tors(Ls+1/ <x >) and
J
1 Vp(s + 1) _ (LS+1/ <x >)/Wp(s + 1).
In this section we shall show that Wp(s+ 1) is a cohomologically trivial
Z[(Z/ps+l)']-module and that Vp(s + 1) is a free

of rank one. We shall also be concerned with the analogous modules for
the totally real subfield
W1 (s + 1) = Tors((L+1)'l < x >) and

Vp (s + 1) = ((L+1)'/ < x >)/Wp (s + 1).
These Z[G(L+1/Qp)]-modules are equal to the a_1-fixed points of 7.4.37,

where
= ,+, for any a E (Z/ps+1) Once again Wp (s+ 1) is
and we shall show
a cohomologically trivial
that Vp (s + 1) is a free
whose generator
we shall specify (Theorem 7.4.46). We shall accomplish part of this by
induction on s, using the following elementary result.
Lemma 7.4.38 Let p be a prime and let m be a positive integer. Let M
be a cohomologically trivial, torsion free Zp[Z/n]-module, where n = pm.
380
If a E MZ/P is a free generator for MZ/p as a Zp[Z/m]-module of rank

one then M is a free Zp [Z/n] -module of rank one. In fact, a generator is
given by any fi E M such that
+gm+g2m+...+gm(p-1))/3
(1
= a,
where g generates Z/n.
Proof Let N denote multiplication by 1+gm+g2m+...+gm(P1) E Zp[Z/n].
Since H2(Z/n; M) = 0, MZ/p = N(M) so that there exists fi such that

N($) = a. Define W Zp[Z/n] --> M by W(z) = z/i then, by hypothesis, W
is an isomorphism on Z/p-invariants. Since M is a free Zp[Z/n]-module
of rank one, it suffices to show that W is surjective. For in this case

Ker(W) would be torsion free and cohomologically trivial and therefore
free and necessarily trivial.
Given x E M there exists u1 E Zp[Z/n] such that N(x) = W(N(u1)) and
therefore
x - W(u1) E Ker(N) = (gm - 1)M.
By induction there exist U2, U3.... E Zp [Z/n] such that
x = W (gm - 1)Ù)
i=1
This element is in the image of W since (gm - 1)p E pZp[Z/n], which

ensures that the series is convergent.
Proposition 7.4.39 (i) In 7.4.37, Wp(s + 1) is a cohomologically trivial
[(Z/ps+1)']-module.
Z
(ii) More generally, let F be a totally real number field and let L1 =
F(op). Let L,+1/L1 be the completion at some p-adic prime of the subextension of the cyclotomic Zp-extension of L1 such that G(LS+1/L1) = Z/pl.
Lo denote the corresponding completion of F. Suppose that x E Lo is
chosen as in 7.4.11 then
Tors(Ls+il < x >)

is a cohomologically trivial Z[G(LS+1/Lo)]-module.
Proof The proof of (ii) is similar to that of (i), once one knows that the
p-primary roots of unity are cohomologically trivial (Washington, 1982,
lemma 13.27, p. 287). In the case of (i) this fact is clear from Table 5.4
and 5.6.3 (see also 5.7.14).
381
From 7.4.11(v) there is a short exact sequence of the following form
LS+1 / < x >> Z/ps x FP -> { 1 }

since eL,+l/Q, = ps(p - 1). The p-primary roots of unity, which lie in
7.4.40
{ I) -* FP x UL,+1
ULs+I, are given by the Z[(Z/ps+')']-module, Mo,s+1, of 5.6.1, which is

cohomologically trivial, by Table 5.4.
Temporarily set Y = UL,+I /Mo,s+1 so that Y is a torsion-free ZPmodule and, as abelian groups, UL, = Mo,s+1 Y.
Identify the cyclic group, (Z/ps+' )' with Z/ps x F. Consider the
(p - 1)-power map, 0, on LS+1/ < x >. Since 0 is an isomorphism on

Z/ps and ULs+, and annihilates FP, 7.4.40 implies that there is a short
exact sequence of the form
{1} -+ U1 s+1 --> Im(q 2) --> Z/ps -> {1}.
Therefore H'(FP;Im(2)) = 0. Hence H'(FP;Ker(2)) = 0. However,

there is a short exact sequence of the form
{1}->Fp) Ker(02)->R-+{1},
where R c FP and FP acts trivially on F. The coboundary must be an
isomorphism
6 :H`(F'P '
P'
so that we must have R = F. Therefore Ker(2) consists of all the torsion

of LS+1/ < x > of order prime to p and, being a cohomologically trivial
Z[F;]-module, it is also a cohomologically trivial Z[(Z/ps+')']-module.
When s = 0 it is clear from 7.4.40 that the cohomologically trivial

module, Mo,1 Ker((b2) is equal to Tors(L/ < x >) = WP(1), which
proves the result in this case. When 1 < s, if A = Mo,s+l Ker(42), we
obtain from 7.4.40 a short exact sequence of the form
{1} - ULs+i/Mo,s+1 -' (Ls+1/ < x >)/A
Z/ps
{1}.
In addition the prime element, ns+i =

E Ls+i, maps to a generator
of the right-hand group. To see that the cohomologically trivial module,
(Ls+1/ < x >)/A, is torsion-free we argue in the following manner. Let
N denote the torsion subgroup. Thus N is a submodule and the action

upon it must be trivial since we may embed it into the trivial module,
Z/ps, by means of the quotient map from the exact sequence. This means
that the induced map
H2i((Z/ps+1);N) = N
-> H2`((Z/PS+1)*;Z/ps) - Z/ps
382
factors through the trivial cohomology group,

HZ`((Z/ps+1)*; (Ls+1/
< x >)/A) = 0.
The only manner in which this can happen, since the cohomology map
is the inclusion of N into Z/ps, is for N to be the trivial group.
Therefore we have shown that (Ls+1/ < x >)/A is torsion free and the
cohomologically trivial module, Mo,s+1 Ker(42), is equal to
Tors(Ls+1/ < x >) = Wp(s+ 1),

In the course of the proof of 7.4.39 we also established the following

result.
Corollary 7.4.41 Let Vp(s + 1) be the torsion free, cohomologically triv-
ial Zp[(Z/ps+l)*]-module of 7.4.37. If 1 < s, then there is a short exact

sequence of Zp[(Z/ps+1)*]-modules of the form
{1} - ULs+i/Tors(ULs+1) -* Vp(s+ 1) - Z/ps ----> {1}.

The image of the prime, ns+l = 1 - laps+i is a generator of the right-hand
group.
Now we begin to find the Zp[(Z/ps+1)*/{l}]-basis for the free modules, Vp (s + 1). We will need the following result.
Proposition 7.4.42 In the following commutative diagram, whose upper and
lower rows are isomorphisms,
(UL,)/(Z/p)
p
ULl
logy
2
7r1 OL1
Up+l
Li
logy
poi (9 Li
the left-hand logarithm induces a Galois isomorphism of the form
loge : UL1/Tors(UL1) - p nidL,.
383
on nl = 1-ip
In addition, 7E1OL, and ni&L, are free

and 7r, respectively.
Proof In the diagram, the surjectivity of the upper pth power map follows
from Hasse (1980, p. 236) since the pth power map induces isomorphisms
of the form
Up+l+d
U2+d
L,
U3+d
L,
L,
- Up+2+d
L,
for all d >- 0. The right-hand logarithm is an isomorphism whose inverse

is exp. Hence
logy : UL,/Tors(UL1) _! 1v1OL,
is an isomorphism of Galois modules.

From the exact sequence
0-+nl(9L,
)OL, --Z/p--) 0
a basis for n1OL, is given by {1-gyp, l-c,...,1-1;P-2, p}. If Qj e G(L1/Lo)

is given by cr (gyp) = l;P then this basis is
{61(x1),72(x1),...,Qp-2(xi),(Ql +...+Q'y-1)(ivi)}.
Hencen1OL, = Zp[G(L1/Lo)] < x1 >.
The argument for ni h, is similar. It suffices to show that the natural

map, Zp[(Z/p)'] < ni >-4 n?OL is onto, since both are free modules
of rank one. By p-adic approximation it suffices to show that the image
of Zy[(Z/p)*] < 7r, > generates
E2
xi (9L, /(PI
= niOL, /(ni+l(9L, )
However, this is clear since
oa(ii) = a`zvi - a`(a - 1)/2xi+1
(mod n,+2)
for l < a< p- l and 2< i< p- l and (ai + ... + Qp_1)(xci) =p.
E]
7.4.43 Let log denote the usual p-adic logarithm (Curtis & Reiner, 1987,
p. 356; see also 4.3.14) defined on the one-units, ULS+,. Let logy denote
the p-adic logarithm of Washington (1982, p. 50). Hence logy is defined
on all of LS+1 and coincides with log on UL, and is normalised so that
logp(p) = 0.
384

Xogp : VP(s + 1)
LR
LS+1 =
which is an inverse to exp 1. In fact, define Xogp = log = loge on
1) c (Ls+1/ < x >)/Tors(Ls+1/ < x >)
Uis+1 /Tors(U1J
and define
7.4.44
Xogp(is+1) = logy
PU+P>
/PS(P - 1)
= logp(ir) - Ps(p -
1)-llogp(1 + P).
This formula makes sense since n +(i-1)/p(1 + p) lies in UL, and x =

p(1 +
is trivial in Vp(s + 1).
7.4.45
Let L+1 denote the fixed field of `complex conjugation' (which
sends y+1 to +,
acting on Ls+l. Hence (L+1)'/ < x > is
isomorphic to the subgroup consisting of the elements of LS+1/ < x >
which are fixed by conjugation. In addition,
(Z/Ps+1)/{1}.
G(Ls+1/Qp) -
Let Vp (s + 1) denote the subgroup of Vp(s+ 1) consisting of the elements

fixed by conjugation. Since TFs+l = 1- Ze+1 = -gps+1(1- ,+1) we see that
fs+1 = its+l E Vp(S + 1)
so that ns+l E VP (s + 1).
Theorem 7.4.46 Let p be an odd, regular prime. Let L+ 1/Q, and Vp (s+l)
be as in 7.4.45. Then, for 0 < s, Vp (s + 1) is the free Zp[G(L+1/Qp)]module generated by the image of irs+1 = 1 - >>s+1.
Proof By 7.4.38 it will suffice to establish this result in the case of VP (1).
Also, by 7.4.42, we have an isomorphism of the form
Xogp : VP (1)
(lrl(9L1)+.
385
The only non-trivial linear relations over Qp between

j < (p - 1)/2} are (Washington, 1982, p. 74) multiples of
1:(p11)/2
logp(1 - gyp)
= (1/2) Ep=i logp(1 - gyp)
= (1/2)logp(p)
=0.
Therefore, if
(p-1)/2
aj(logp(l - P) - logp(1 + p)) = 0
j=1
we may apply U 1 ,-- . , a(p-1)/2 and add the results to obtain E7 aj = 0.

Hence a1 = ... = a(p-1)/2 = 0. This means that Xog(it1) is a free generator
of Vp (1) Qp as a Qp[(Z/p)'/{1}]-module.
Now let us examine the circumstances under which Xogp(n,) is a free
generator of V,1(1) as a Zp [(Z/p)' {11] -module. Consider the following
(p - 1)/2 x (p - 1)/2 matrix.

7.4.47
logp(1-gyp)-(p-l)-llogp(l+p)
The (i,j)th entry in 7.4.47 is equal to a;(aj(Xogp(n1))). The determinant,

Rp, of this matrix is non-zero, since Xogp(n1) is a free Qp[(Z/p)'/{1}]
generator for (n1 (9L, )+ Q p = Li . We would like to know the p-adic
valuation of Rp E Q. By elementary row and column operations (adding
all the lower rows to the top row and then subtracting the first column
from each of the others) we see that
Rp((-1/2)logp(l
+p))-1
386
is equal to
1
det
logP(1-l
p)-(p-1)-1logP(l+p)
which, in turn, is equal to

1
Q2(XOgp(7t1))
Q2(l09p((1 - p)/(1 - P)))
a3(X09pOC1))
a3(logP((1 - p)/(1 - P)))
det
Hence
RP = (Y-1/2)logp(1 + p)det(a;,j),
where aij = aj+l(logp((l - by 1)/(1 - SP))) for 1 < i,j < (p - 3)/2.
The cyclotomic units
{(1 - P)/(1 - gyp); 2 < j < (p - 1)/2}

are a basis for the cyclotomic units of Z[lp]+ modulo torsion. Hence Rp
is related to the p-adic regulator, Rp(Q(cp)+), of Washington (1982, p. 70)
by the equation
7.4.48
Rp = (-1/2)logp(1 +
for some p-adic unit, u.

In order to calculate the p-adic ideal generated by Rp, recall the p-adic
class number formula (Washington, 1982, section 5.24, p. 71)
2(P-3)l2h(Q(bp)+)RP(Q(bp)+)
_ 11 Lp(1, X),
X
where the product is taken over non-trivial one-dimensional characters

of
When p is a regular prime the right-hand side of the
p-adic class number formula is a p-adic unit (Washington, 1982, section
387
is a p-adic unit if p is a regular prime
5.23, p. 70). Also
so that
zz
RP(Q(Sp)+)
d(
(Sp)+/Q)
is a p-adic unit.
Therefore, by 7.4.48, the discriminant ideal of the Zp-llattice,
X1 = Zp[(Z/p)*/{1}] < XogP(it1) >C Zp[Sp]+
in
is given by
bx, = RP = p2RP(Q( P)+)2 = PZSZP[
P]+.
However, X1 c X2 = (7r 1(9L, )+ so that, by Serre (1979, pp. 48-49),

X1 = X2 if and only if 8x, = 6x2. On the other hand, since
WL, /(r2))+ Z/p,

the quotient of Z,
by X2 is equal to Z/p so that, by Serre (1979,
proposition 5, p. 49)
6x2
= P26ZP[ P1+ =
bx
which shows that it is a free generator of VP (1), as required.

7.4.49
Let E =
Q so that G(E/F) = (Z/ps+1)* We
shall now consider the wild prime, p, and apply the procedure of 7.4.317.4.33 for obtaining the p-adic component of the Hom-description of the
second Chinburg invariant for
/Q).
In the notation of 7.4.31-7.4.33, EQ =

Qp
so that G(EQ/FP) = G(E/F) in this example. In Z[1/p][G(Q(lps+,/Q))]
define idempotents by the formulae
ei
= p j(es+1
EG L(p L-
g)
(1
< j < s)
7.4.50
and
=Ps
1)-1(
gEG(L
F:+, /QP ) g)'

The integers, (9E = Z[pr+,], have a Z-basis given by
{ jP',
0<r<s,HCF(j,p)=1,1<j<ps+1-r_1
and (j,r)_(1,s+1)}.
Write Qj E G(E/F) for the map given by a ( ps+,) =
t = Ypt + Kpr-, +... +p
Hence, if
388
then
7.4.51
eu(as+l) = as+l-u
and
61(1 - er+l)erer_l ... el(as+1) = SpS+i

7.4.52
and
P)-1
es+l...el(as+l) _ (1 -
An arbitrary element, z E (9E[1/p], is of the form

Z=
j,r Pjf+'i + 1,s+1,
j,r
where the sum is taken over
{O<r<s,HCF(j,p)= 1,1 < j
<ps+l-r-1}
and clearly z =
j,r7j(1 - er+1)erer-l ... el(cls+1) - 1,s+1(P - 1)es+l ... el(as+l)
j,r
so that as+l will serve as the normal basis element which is denoted by
a E (9E in 7.4.31-7.4.33. If necessary we may replace a by a suitable
multiple, pC(p - 1)da, in order to guarantee that a E X. This change

does not alter the element represented in the Hom-description of the
class-group, since it changes the representative by a function which is
clearly in the image of the induction map from the class-group of the
trivial group. Therefore we shall assume that a = as+l
Observe that if 1 is a prime different from p and Q a (9E is a prime
over l then as+1 will also serve as a Zl[G(EQ/Ql]-basis element for (9EQ
or, equivalently, as a
element for
Zt [fps+,] = Zl (9E = fl (9EQ.
QI1
We are now ready to combine Theorem 7.4.46 with the method of

Theorem 7.4.35 to evaluate
2). The formula will involve
the value of the p-adic L-function at s = 1.
389
7.4.53 p-adic L -Junctions
Let p be an odd prime. Choose a generator, g E (Z/ps+1) and let

n = exp(2iit/n). Let
Xi :
(Z/Ps+l) -, C
denote the Dirichlet character given by

Xi(g) _ ,pS(p-1)
if j = g" E (Z/ps+l)' Denote by fx the conductor of x (Washington,

1982, p. 19) so that fx; = pE is the least positive integer such that Xj
factors through (Z/ps+1) , (Z/pE)

Now let us recall Leopoldt's formula (7.4.55), which connects the padic logarithms and p-adic L-functions (Washington, 1982, p. 63). Let
x : (Z/ps+1)* _, C' be a Dirichlet character with conductor, fx = ps+1
Write T(x) for the Gauss sum
7.4.54
T(X) = IaE(Z/ps+') X(a) p,+'
If x = 1, the trivial character, then set T(1) = 1.

In this situation the p-adic L-function of x at s = 1, Lp(l, x), satisfies
7.4.55
Lp(l, X) = -T(X) KaE(Z/p'+') X(a)p(s+1)logp(1
- a+')
We would like to write Lp(l,xi) in the form of 7.4.55 for all Xi. In
Washington (1982, p. 63) 7.4.55 is proved for non-trivial characters, x,
with x(-1) = 1 (x is even) and fx = ps+1 However, if x(-1) = -1 (x is
odd), then both sides of 7.4.55 are zero. If x = 1 then the right-hand side
of 7.4.55 equals
log,(l - gyp,+') = logP(P) = 0,
aE(Z/ps+' )'
so that we shall define Lp(1,1) to be zero. On the other hand, if x is even

and fx = p` > 1 then Lp(l,X)T(X)-1
_ -EaE(Z/p')' 7(a)p`logp(1 - l;pS
+,)
= ->aE(Z/p')' 7(a)p` EbE(Zlpl+')*,b=- I (pl) logP(1 - P1+

EaE(Z/p''+i) X(a)p`logp(1 - Sps+' )
Thus 7.4.55 holds in general and we may write
390
7.4.56
LP(1,Xi)
EQE(Z/v+') Xi(a)fX;'T(Xi)log,(1 -fps+.)
for all j E (Z/ps+1)*

7.4.57
+i)/Q, 2)
Following the recipe of 7.4.35, there exists

QPI(Z/PS+1)/{1}]"
Bp,s+1 E
such that
7.4.58
B+
p,s+1(a+s+1) = Xo $P(ns+1) = loge
(;;;) /PS(P- 1)
where a = Q_1(a) is the complex conjugate of a and a+1 = as+1 + as+1

For Q(gps+i)+/Q we have decomposed the Chinburg invariant
2)
into the sum of a finite module, W, (s+1) = WP(s+1)`-1' (the invariants

of WP(s + 1) under complex conjugation), and a module whose Homdescription representative (for tame primes, see Frohlich, 1976, theorem 1,
p. 388) is
det(BP +1)(X)
at p
(a+1 I X)-1(at I X)
at other l
XH
where (a,) is as in 7.2.1 and 7.2.2 and a+1 is a normal basis element for
Q(I;P,+t)+/Q. Here (z I X) denotes a resolvent defined by the formula
Eg(z)X(g-1).
(z I X) _
s
In 7.4.50-7.4.52 we observe that as+1 would serve as a Zl normal basis

element if 1 is different from p. Hence al may be taken to equal as+1 (and
similarly a+1 = al) so that

(a+
I X)-1(ai I X) = 1
at primes, 1, different from p in 7.4.57.
Next we remark that

BP +1(a +1) = Xog'(ns+1)
so that for all X (Frohlich, 1976, proposition 1.3, p. 386)

(XogP(irs+1) I x) = det(BP+,)(X)(a+1 1 x)
391
In addition, if X is one-dimensional (Frohlich, 1976, (1.4),
in
p. 385),
(XogP(ns+i) I X) _ EY(XogP(its+i))X(Y)
Y
(summed over (Z/ps+')*/{1})

_ 1: [Y(logP(1
p-S(p - 1)-'logP(1 +P)]X(Y)
(from 7.4.44)
_ (1/2) E [logp(1
p S(p-
1)-ilogP(1
+P)]X(a)
aE(Z/ps+i).
= -(fx/2T(X))LP(1, X)
if X is non-trivial, by 7.4.56. Also,
(XogP(ns+i) 11) = -(1/2)logp(1 +p).

Lemma 7.4.59 Let X be a one-dimensional complex representation of
G(Q(lp,+i)/Q) with conductor 1 < fx = pi+i Then
(i) (as+i I X) = ps+'fX'T(X),
(11) (a +1 I X) _ (s+i I X) = ps+'fx'T(X), if X(-1) = 1,
(iii) (t+1 11) _ (as+1 11) _ -PS
In the above formula the Gauss sum is the purely local Gauss sum of 7.4.54.
Proof The proof is left as an exercise (see 7.5.20).

This discussion, together with 7.4.35, yields the following result.
Theorem 7.4.60 Let p be a regular, odd prime. For Q(c,+i)+/Q the
Chinburg invariant
Q(Q(S,+i)+/Q,2) E W2(Z[(Z/Ps+i)1{1}])
392
decomposes into the sum of the finite module, Wp (s + 1), and a module
whose Hom-description representative is
X --*
fxLp(1,X)
at p,X non-trivial,
-plogp(1 + p)
at p, X = 1,
at other 1.
)+/Q, 2) is the sum of Wp (s + 1) and an
Proof By 7.4.35,
element whose Hom-description is trivial at primes different from p. At

p a one-dimensional character, X, is sent to
(Xogp(ns+1) I X)(a +1 I
x)-1
-(fx/2i(x))Lp(1, X)(Ps+'fX'T(x))-'
x non-trivial,
-(1/2)logp(1 +p)(-Ps)-'
X = 1,
-(1/2)fxp s-1Lp(1,X)
(1/2)Plogp(1 +p)P
since T(x)2(X) =
1-1
X non-trivial,
X = 1,
X(6-1)fx = fx.
However, the function sending X to (-1/2)dim(x) equals Det(-1/2),

which lies in Det(Zp[((Z/ps+1)')/{1}]'). Similarly the function sending x
to (ps-1)dim(x) lies in
for all primes, 1, different
from p. Hence, dividing by the global function
(X'' (P s-1)dim(x)) E HomnQ(R(((Z/Ps+1)')/{1}), Q')

and multiplying by these l-adic determinants alters the Hom-description
to the required form.
7.4.61 fl(Fs+1/Q,2)
Let Fs+i/Q be the subextension of
given by Fs+1 =
Hence G(Fs+1 /Q) - Z/ps with a generator, y, say, which

may be taken to be the image of 1 + p E (Z/ps+1)=
In this situation we shall prove the following result.
393
Theorem 7.4.62 Let p be an odd, regular prime. Then, as predicted by the

conjecture of 7.2.14,
S2(FS+l /Q,2) = 0 c D(Z[Z/ps])

This result will be proved in the course of the subsequent discussion.
Lemma 7.4.63 In W2)(Z[Z/ps]) the function given on one-dimensional
representations, x, by
at p, x non-trivial
at p,x=1,
at other l
represents the trivial element.
Proof In HomnQ(R(Z/ps),Q') we have the global function given by
xH
fx
X non-trivial
1 p2 x=1.
Dividing by this we may change the Hom-representative of this element
so that it is represented by the function which is trivial on all characters
at p and on all non-trivial characters away from p. At all other primes
the function sends the trivial one-dimensional character to p. However,
each component of this function is a local determinant since, by Frohlich
(1983, proposition 2.2, pp. 23-24),
Det(R[Z/ps]') = HomlQ(R(Z/PS), (Q(cs) Q R)*)
and (if l is different from p)
Det(Zi[Z/ps]*) =
Z Zt)*)
This completes the proof, since these functions represent the trivial
element in the class-group.
Next, we observe that the contribution to S2(FF+i /Q, 2) from Wp(s+1)
is trivial since this is represented by WP(s + 1)(Z/P)*. By 7.5.21, this has
trivial action and no elements of order p. Therefore, by 5.3.9, this module
represents a Swan module, which is trivial in the class-group in the case
394
of a cyclic group (4.2.48; Curtis & Reiner, 1987, p. 344). Now let us
examine the other part of the Chinburg invariant.
From Washington (1982, pp. 118-123), if W and 0 are complex (onedimensional) characters of Z/ps and (Z/p)', respectively, then x = OW is
a character of
(Z/p)* x Z/ps = (Z/ps+1)* ~
Furthermore, if x(-1) = 1 then

LP(1, x) = MP + 1)T(1 + p) - 1, 0),
7.4.64
where
f(T, 0) = g(T, 0)(T - p)-1(1 + p)
and g(T,0) E Zp[[T]]. Since we chose the generator, y, to correspond to

1 + p E (Z/ps+1)' we see that
f((P + 1)i (1 + p) - 1, 0) = W(f((P + 1)Y - 1, 0))
since y = 1 +p corresponds to 1 + T under the isomorphism (Washington,

1982, p. 113)
Zp[[T]] = limZp[Z/ps].
S
Note that this isomorphism ensures that g((p + 1)y -1, 0) makes sense in
ZP [Z/PS]
The character, x, factors through G(FS+1 /Q) '" Z/ps if and only if
0 = 1 and W = X. In this case
7.4.65
LP(1, X) = X(g((P + 1)y - 1,1))(X(Y
- 1))-1.
However, g(T, 1) E Zp[[T]]' (Washington, 1982, p. 125, lemma 7.12)

so that
g((p+ 1)y - 1,1) E Zp[Z/ps]'

and the function
7.4.66
x'-' X(g((P + 1)y - 1, 1)) = Det(g((p + 1)y-1 - 1,1))(x)
lies in Det(Zp[Z/ps]') c Homo Q(R(Z/ps),Zp[l p,]*). Therefore the element of

given by the function of 7.4.60

fXLp(1,x)
7.4.67
X2 -->
395
at p,x non-trivial,
-plogp(1 + p) at p, x = 1,
at other 1.
represents the same class in '2'(Z[Z/ps]) as does
(x(y) - 1)-'
x'--->
at p, x non-trivial,
-p 'logp(1 + p)(g(O, 1))-' at p,x = 1,

at other 1.
Lemma 7.4.68 The function on one-dimensional representations, x, given

by
7.4.69
x' -->
at p, X non-trivial,
-p'logp(1 + p)(g(0,1))-'
at p, X = 1,
at other 1
represents the trivial element in D(Z[Z/ps]).
Proof Observe that -p ' logp(1 + p)(g(0,1))-' is a p-adic unit, which is

congruent to 1 modulo p. We may write this unit as uNVN where vN is an
integer prime to p and uN is a unit which is congruent to 1 modulo p^'.

Since uN is a pN-'th power, the element of D(Z[Z/ps]) represented by
1
UN
at p, X = 1,
at other 1
is divisible by p^'-' while the element represented by

1
vN
atp,x=1,
at other 1
396
is trivial, being a Swan module. Hence the class represented by 7.4.69 is

divisible by p"' for all m in the (finite) class-group and hence is zero.
Finally, therefore, we see that 7.4.67 represents the same class as
x1>
Q(y) - 1)-1
atp,x=1,
at other 1.
However, the p-adic values of this representative lie in Z1 [ s] * for all

primes, 1, different from p. Therefore the argument which was used in
the proof of 7.4.63 shows that this formula gives a representative for the
trivial element, as required.
The Frohlich-Chinburg conjecture of 7.2.14 (sometimes referred to as
the Second Chinburg Conjecture (Cassou-Nogues et al., 1991, section
4.2; Chinburg, 1985, section 3.1; Holland, 1992, section 1.1) anticipates
that the invariant, f (E/F, 2), will be equal to a class of order two
which has a Hom-description in terms of the Artin root numbers of
the irreducible symplectic representations of G(E/F). In particular, this
conjecture would predict that f2(E/F, 2) would vanish when G(E/F) were
abelian. Theorem 7.4.62 adds some new, wildly ramified examples to the
list of cases in which the conjecture is true.
7.5 Exercises
7.5.1 (Serre, 1979, p. 202) Prove Theorem 7.1.20.
(Hint: Show that the cohomology classes of 7.1.20 are fundamental

classes for a local class formation in the sense of Serre (1979, chapter XI,
section 3) and use this fact to reduce the calculation to the unramified
case.)
Let Fq denote the finite field with q elements and let Fq denote
its algebraic closure. Let FO denote the Frobenius map given by the gdth
7.5.2
power.
(i) Prove that H1(G(Fgds/Fgd); Fqds) = 0. (That is, prove that
1+Fo+...+Fo
Fo-1
Fqds
is exact in the middle.)
Fqds
Fgas
7.5 Exercises
397
(ii) Prove, for any y E Fqd,, that there exists w E Fq such that
Fo(w) - w = y.
Verify that the 2-cocycle, f : G(L/K) x G(L/K)

is normalised. (That is, show that
7.5.3
L`, of 7.1.23
f(x, l) =f 0, X) = 1.)
From the description of the cohomology generators in terms

of 2-extensions of 7.1.8 and 7.1.20, prove that the following diagram
7.5.4
commutes:
H2(G(L/K);L*)
my
Q/Z
(- [F : K])
H2(G(L/F);L')
7.5.5
inv
Q/Z
Let F be an intermediate extension of L/K. Show that

Resc(L K (Q(L/K, 2)) = S2(L/F, 2) E WY(Z [G(L/F)] ).
Let N .i G be finite groups. In terms of the Hom-description for

KOT(Z[G]), verify that InfGIN : R(G/N) -- R(G) induces the homomorphism which sends the module, T, to its N-fixed points, TN.
7.5.6
7.5.7
Let L/K be a Galois extension of number fields and let
G(E/N) a G(E/K)
be a normal subgroup. Let
infE/N : f,2(Z[G(E/K)])
`WY(Z[G(NlK)])
denote the canonical map (induced by inflation on representations or by

taking G(E/N)-fixed points on modules).
Use 7.1.57 to show that
infEIN(KI(E/K,2)) = S2(N/K,2) E WY(Z[G(N/K)])
7.5.8
Let G(L/F) < G(L/K) be as in 7.3.43 with coset representatives
X1, ... , xd.
398
(i) Verify that the homomorphism
: d 1Z[G(L/F)] -p Z[G(L/K)]
u1
given by u1(b1,...,bd) =
Ed 1 bixi is an isomorphism of left Z[G(L/F)]-
modules.
(ii) Verify that
u1(o(L/F),... , o(L/F)) = o(L/K).

(iii) Define
U2 : d 1Z[G(L/F)] -' d 1Z[G(L/F)]
by u2(b1,...,bd) = (bl,b2+bl,...,bd+b1). Verify that

u2(o(L/F), 0,..., 0) = (r(L/F),... , o(L/F)).
(iv) Verify that the homomorphism
v :I G(LIK) --> IG(L/F) (d 2Z[G(L/F)])

given by (ai E Z[G(L/F)])
v
i=1
i=1
(aixi) =
ai, a2, ... , ad
is an isomorphism of Z[G(L/F)]-modules.
(v) With the identifications of (i)-(iv), verify that 7.3.46 may be identified with 7.3.47 as a diagram of Z[G(L/F)]-modules.
Let G(L/M) a G(L/F) be as in 7.3.50.

(i) Prove that there are Z[G(M/F)]-module isomorphisms
7.5.9
L'
(ii) Let y1, ... ,
M*
G(L/M)
(1 + X)
(I + X)G(L/M)
1 + X G(L/M)
be a set of coset representatives for
G(L/F)/G(L/M)
and let o(L/M) be as in 7.3.46. Show that the map which sends y,G(L/M)
to yso(L/M) induces Z[G(M/F)]-module isomorphisms of the form
a : Z[G(M/F)] - (Z[G(L/F)])G(LIM)
and
a :I G(M/F) -- (I
G(L/F))G(L/M).
7.5 Exercises
399
(iii) Show that the homomorphism, a, of (ii) induces an isomorphism

c(m - o(M/F))Z[G(M/F)] ^_-' ((r - o.(L/F))Z[G(L/F)])G(LIM).
(iv) With the identifications of (i)-(iii) verify that the G(L/M)-invariants of 7.3.47 may be identified with 7.3.51.
7.5.10 Prove that the G(L/M)-invariants of 7.3.47 may be identified
with 7.3.53.
Let G be a finite group. Let v be a non-zero integer and also

multiplication by v, v : Z ---p Z. In the notation of 7.3.11, show that f
has a canonical factorisation g E HomnQ (R(G), 5(E)) where
7.5.11
fv if 1 = x,
g(x) =
if 1
x, irreducible.
Let G be a finite group. Let w be a non-zero integer and

also let multiplication by w, be denoted by w : Z[G] -+ Z[G]. In the
7.5.12
notation of 7.3.11, show that fwZ[G],Z[G] has a canonical factorisation

h e HomcQ(R(G),5(E)) where
h(x) = w
dimC(X)
for all x E R(G).
Let G be a finite, cyclic group. Let j be an injective Z[G]-module

homomorphism in a short exact sequence of the following form:
7.5.13
0-->IG-Z[G]/(fGZ[G])-) Z/#(G)-i0.
Show that f,* has a canonical factorisation s e HomnQ(R(G),5(E)),
where
if 1=x,
ideal < x(x) - 1) > if
x, irreducible.
Here x is a generator of G, aG = EgEG g and ideal < x(x) - 1) > lies in

J(Q(x)).
400
Let H < G be groups and suppose that #(G) = n. Let oG be as

in 7.5.12. Let 9 be a set of coset representatives for G/H and assume
that 1 E R.
(i) Show that
7.5.14
A = (I GH)/((n - OG)Z[G])H)
is isomorphic to the abelian group on generators
{2gIgE-4-{l}}
subject to the relations
nA.g =0 for g E .GP - { 1 }
and
#(H) E 2g
= 0.
(9E9?-{1}
(ii) Prove that
#(A) =
#(H)n(IG:HD-2).
(Hint: Calculate a determinant, as in Theorem 5.2.33(proof).)

7.5.15
Let L/K be a Galois extension of number fields with group,
G(L/K). Let S be a finite, G(L/K)-stable set of places of L which includes

the infinite places, the ramified places and such that, for K < F < L, the
S-class number of F is one. Set
JS,L = fl Lll x 11 (91,,,

VES
vS
and
US,L={xEL'IxE(9i,ifvVS}.
We have a diagonal embedding of US,L into JS,L and a resulting short
exact sequence of abelian groups
0-)
Global class field theory shows that {H2(G(L/K);Cs,L)} is a class
formation (Serre, 1979, pp. 166 and 221). Consequently there are universal
cohomology classes which represent the Weil group extension
CS,L -a WS,L --> G(L/K ).
7.5 Exercises
401
(i) Use the Weil group extension to construct a class
f)(L/K, S) E c P(Z[G(L/K)])
(cf. 7.1.34).
(ii) How does fl(L/K, S) vary with S ?

(iii) Let G(L,,/K&< G(L/K) be a decomposition group. What is the
relationship, if any, between Q(L/K, S) and the local Chinburg invariant
of 7.1.29.
Embed Q2(1/(-5)) as a maximal subfield into a 2-adic division

algebra whose Hasse invariant is equal to 1/2. (See Reiner 1975, p. 148,
for example, for models of division algebras and consider the matrix
7.5.16
3 -3
x =
-2
Z3 - 3
where 3 = exp(2ni/3) and z denotes the complex conjugate of z.)

7.5.17
Embed Q2(J(-2), ./(-5)) as a maximal subfield into a 2-adic
division algebra, D, of index four, whose Hasse invariant is equal to 1/4.

(If a = 3 - Z3 and n2 = -2 in D consider z = a(l + 1r2 + 7t3) + 2n in the
model for D given in Reiner (1975, p. 148).)
7.5.18
If 13,13 E QZ then the Hilbert symbol
(a, fl) E {1}
takes the value one if there exist x, y E Q2 such that a + 13x2 = y2 and is
equal to minus one otherwise.
(i) Suppose that u, v E QZ are non-squares. Show that there exists

a quadratic extension L/Q2(.,/(u), V(v)) such that L/Q2 is Galois with
G(L/Q2) = Qg if and only if
(u, v)(-1, u)(-1, v) = 1.
(Cf. Snaith, 1989b, p. 102, Example (2.31).)
(ii) Construct a Q8-extension containing Q(,/(-2), J(-5)).

7.5.19 Complete the details of the proof of 7.4.11, following the argument
which was used to establish 7.3.27.
7.5.20
Use 3.3.3 to prove the formulae of 7.4.59.
402
7.5.21
Let Wp(s + 1) and WP +(s + 1) be as in 7.4.37, where p is an odd
prime.
(i) Determine the Galois action of (Z/ps+i)' on Wp(s + 1).

(ii) Show that WP (s+ 1) is a finite group of order prime to p on which
the p-Sylow subgroup of
acts trivially.
Bibliography
Adams, J.F. (1963), On the groups J(X) I, Topology, 2, 181-195.

Adams, J.F. (1965a), On the groups J(X) 11, Topology, 3, 137-171.
Adams, J.F. (1965b), On the groups J(X) III, Topology, 3, 193-222.
Adams, J.F. (1966), On the groups J(X) IV, Topology, 5, 21-71.
Bentzen, S. & Madsen, I. (1983), On the Swan subgroup of certain periodic
groups, Math. Ann., 264, 447-474.
Boltje, R. (1989), Canonical and explicit Brauer induction in the character ring
of a finite group and a generalisation for Mackey functors, Augsburg Univ.
thesis.
Boltje, R. (1990), A canonical Brauer induction formula, Representations

lineaires des groupes finis, Asterisque #181-182, Societe Mathematique de
France, 31-59.
Boltje, R., Cram, G-M. & Snaith, V.P. (1993) Conductors in the non-separable
residue field case, NATO ASI Series Vol. 407 Proc. Algebraic Topology and
Algebraic K-theory, Lake Louise, Kluwer, 1-34.
Boltje, R., Snaith, V. & Symonds, P. (1992), Algebraicisation of Explicit Brauer
Induction, J. Alg., 148, 504-527.
Brauer, R. (1951), Bezeihungen zwischen Klassenzahlen von Teilkorpern eines
Galoischen Korpers, Math. Nachr., 4, 158-174.
Brylinski, J-L. (1983), Theorie du corps de classes de Kato et revetements
abeliens de surfaces, Ann. Inst. Fourier, 33, 23-38.
Burns, D.J. (1991), Canonical factorisability and a variant of Martinet's
conjecture, J. London Math. Soc., 142, 24-46.
Cassou-Nogues, Ph. (1978), Quelques theoremes de base normal d'entiers, Ann.
Inst. Fourier, 28, 1-33.
Cassou-Nogues, Ph., Chinburg, T., Frohlich, A. & Taylor, M.J. (1991),
L-functions and Galois modules, London Math. Soc. 1989 Durham
Symposium, L -functions and Arithmetic, Cambridge University Press,
75-139.
Chinburg, T. (1983), On the Galois structure of algebraic integers and S-units,
Inventiones Math., 74, 321-349.
Chinburg, T. (1985), Exact sequences and Galois module structure, Annals of
Math., 121, 351-376.
Chinburg, T. (1989), The Analytic Theory of Multiplicative Galois Structure,
Mem. A.M. Soc. No. 395.
403
Bibliography
404
Coombes, K. (1982), Algebraic K-theory and the abelianised fundamental

groups of curves, University of Chicago thesis.
Coombes, K. (1986), Local class field theory for curves, Contemp. Math., 55,
117-134.
Curtis, C.W. & Reiner, I. (1981, 1987), Methods of Representation Theory, vols. I
& II, Wiley.
Dickson, L.E. (1958), Linear Groups, Dover.
Frohlich, A. (1976), Arithmetic and Galois module structure for tame
extensions, J. Reine. Angew. Math., 286/7, 380-440.
Frohlich, A., ed., (1977), Algebraic Number Fields, London Math. Soc. 1975
Durham Symposium, Academic Press.
Frohlich, A. (1983), Galois Module Structure of Algebraic Integers, Ergeb. Math.
3 Folge-Band No. 1, Springer-Verlag.
Frohlich, A. (1984), Classgroups and Hermitian Modules, Progress in Math. No.
48, Birkhauser.
Frohlich, A. (1988), Module defect and factorisability, Ill. J. Math., 32, 407-421.
Gerardin, P. & Labesse, J-P. (1979), The solution of a base change problem for
GL(2) (following Langlands, Saito, Shintani), Amer. Math. Soc. Proc. Symp.
Pure Math., Automorphic Forms, Representations and L-Functions, 115-134.
Gorenstein, D. (1968), Finite Groups, Harper and Row.
Green, J.A. (1955), The characters of the finite general linear groups, Trans.
Amer. Math. Soc., 80, 402-447.
Greenberg, M. & Harper, J. (1971), Lectures on Algebraic Topology, W.A.
Benjamin.
Hasse, H. (1980), Number Theory, Springer-Verlag Grund. Math. Wiss. No. 229.
Hilton, P.J. & Stammbach, U. (1971) A Course in Homological Algebra; Grad.
Texts in Math. No. 4, Springer-Verlag.
Holland, D. (1992), Additive Galois module structure and Chinburg's invariant,
J. reine angew. Math, 425, 193-218.
Holland, D. & Wilson, S.M.J. (1992), Frohlich's and Chinburg's conjectures in
the factorisability defect class-group, preprint.
Hunter, J. (1964), Number Theory; Oliver and Boyd.
Huppert, B. (1967), Endliche Gruppen I. Springer-Verlag.
Hyodo, O. (1987), Wild ramification in the imperfect residue field case, in Galois
Representations and Arithmetic Algebraic Geometry (ed. Y. Ihara), Adv.
Studies in Pure Maths. No. 12, Kinikuniya Co., 287-314.
Illusie, L. (1979), Complexe de de Rham-Witt et cohomologie cristalline, Ann.
Sci. Ec. Norm. Sup., 12, 501-661.
Isaacs, I.M. (1976), Character Theory of Finite Groups Academic Press.
Iwasawa, K. (1986), Local Class Field Theory, Oxford Math. Monographs,
Oxford University Press.
Kato, K. (1987), Swan conductors with differential values, Advanced Studies in
Pure Math., 12, 315-342.
Kato, K. (1989), Swan conductors for characters of degree one in the imperfect
residue field case, Contemp. Math., 83, 101-131.
Kim, S. (1991), A generalization of Frohlich's theorem to wildly ramified
quaternion extensions of Q, Ill. J. Math., 35, 158-189.
Kim, S. (1992), The root number class and Chinburg's second invariant, J. Alg.,
153, 133-202.
Lang, S. (1970), Algebraic Number Theory, Addison-Wesley.
Lang, S. (1984) Algebra, 2nd ed., Addison-Wesley.
Bibliography
405
Macdonald, I.G. (1979), Symmetric Functions and Hall Polynomials, Oxford

Math. Monographs, Oxford University Press.
Martinet, J. (1977a), H8, London Math. Soc. 1975 Durham Symposium,
Algebraic Number Fields, ed. A. Frolich, Academic Press, 525-538.
Martinet, J. (1977b), Character theory and Artin L-functions, London Math.
Soc. 1975 Durham Symposium, Algebraic Number Fields, ed. A. Frolich,
Academic Press, 1-87.
Matsumura, H. (1990), Commutative ring theory, Cambridge Studies in
Advanced Mathematics No. 8, Cambridge University Press.
Maunder, C.R.F. (1970), Algebraic Topology, Van Nostrand Reinhold.
Milne, J.S. (1980), Etale cohomology, Princeton Mathematical Series No. 33,
Princeton University Press.
Oliver, O. (1978), Subgroups generating D(Z[G]), J. Algebra, 55, 43-57.
Oliver, O. (1980), SKI for finite group rings II; Math. Scand., 47, 195-231.
Quillen, D.G. (1971), The Adams conjecture, Topology, 10, 67-80.
Quillen, D.G. (1972), On the cohomology and K-theory of the general linear
groups over a finite field, Annals of Math., 96, 552-586.
Rota, G-C. (1964), On the foundations of combinatorial theory I. Theory of
Mobius functions, Z. Wahrsch verw. Gebiete, 2, 340-368.
Reiner, I. (1975), Maximal Orders, L.M.Soc. Monographs No. 5, Academic
Press.
Rogers, M (1990), On a multiplicative-additive Galois invariant and wildly

ramified extensions, Columbia University thesis.
Sehgal, S.K. (1978), Topics in Group Rings, Mono. in Pure and Applied Math.
No. 50, Marcel Dekker.
Sen, S. (1969), On automorphisms of local fields, Annals of Math., 90, 33-46.
Serre, J-P. (1960), Sur la rationalite des representations d'Artin, Annals of Math.,
72, 405-420.
Serre, J-P. (1977), Linear Representations of Finite Groups, Grad. Texts in Math.
No. 42, Springer-Verlag.
Serre, J-P. (1979), Local Fields, Grad. Texts in Math. No. 67, Springer-Verlag.
Shintani, T. (1976), Two remarks on irreducible characters of finite general
linear groups, J. Math. Soc. Japan, 28, 396-414.
Snaith, V.P. (1988a), A construction of the Deligne-Langlands local root
numbers of orthogonal Galois representations, Topology, 27, 119-127.
Snaith, V.P. (1988b), Explicit Brauer Induction, Inventiones Math., 94, 455-478.
Snaith, V.P. (1989a), Invariants of representations, NATO ASI Series No. 279,
Algebraic K-theory: Connections with Geometry and Topology, Kluwer,
445-508.
Snaith, V.P. (1989b), Topological Methods in Galois Representation Theory,
C.M.Soc Monographs, Wiley.
Snaith, V.P. (1990a), On the class group and the Swan subgroup of an integral
group-ring, McMaster University preprint No. 3.
Snaith, V.P. (1990b), Restricted determinantal homomorphisms and locally free
class groups, Can. J. Math., XLII, 646-658.
Snaith, V.P. (1990c), The Yukiad, a novel, The Book Guild (ISBN 0-86332-478-9).
Sullivan, D. (1974), Genetics of homotopy theory and the Adams conjecture,
Annals of Math., 100, 1-79.
Swan, R.G. (1960), Periodic resolutions for finite groups, Annals of Math, 72,
267-291.
406
Bibliography
Symonds, P. (1991), A splitting principle for group representations, Comm. Math.

Helv., 66, 169-184.
Tate, J.T. (1977), Local constants, London Math. Soc. 1975 Durham Symposium,
Algebraic Number Fields (ed. A. Frohlich), Academic Press, 89-131.
Tate, J.T. (1984), Les conjectures de Stark sur les fonctions L d'Artin en s = 0,
Progress in Mathematics No. 47, Birkhauser.
Taylor, M.J. (1978), Locally free classgroups of groups of prime power order, J.
Algebra, 50, 463-487.
Taylor, M.J. (1980), A logarithmic approach to class groups of integral group
rings, J. Algebra, 66, 321-353.
Taylor, M.J. (1981), On Frohlich's conjecture for rings of integers of tame
extensions, Inventiones Math., 63, 41-79.
Taylor, M.J. (1984), Class Groups of Group Rings, L.M.Soc. Lecture Notes Series
No. 91, Cambridge University Press.
Thomas, A.D. & Wood, G.V. (1980), Group Tables, Shiva Math. Series 2, Shiva
Publishing.
Wall, C.T.C. (1966), Finiteness conditions for CW complexes, Proc. Royal Soc.
A., 295, 129-139.
Wall, C.T.C. (1974), Norms of units in group rings, Proc. London Math. Soc. (3),
29, 593-632.
Wall, C.T.C. (1979), Periodic projective resolutions, Proc. London Math. Soc. (3),
39,509-553.
Washington, L. (1982), Introduction to Cyclotomic Fields, Springer-Verlag Grad.
Texts in Math. No. 83.
Wilson, S.M.J. (1990), A projective invariant comparing integers in wildly
ramified extensions J. reine angew. Math., 412, 35-47.
Index
absolute norm 335

Adams operation 108, 110, 121, 153, 257
conjecture 192
adele 110, 331
of a group-ring 112
additive character 75
admissible subhomomorphism 38
algebraic closure 113
Alperin, Feit, Thompson Theorem 137
analytic class WL/K 334
Artin
conductor 247, 335
ideal 335
exponent 134, 191
Induction Theorem 26, 342
L-function 334
reciprocity 256
representation 248
root number 334
Artin-Schreier theory 263
bar resolution 301
binomial theorem 122
Brauer
Group 125
Induction Theorem 31
Bruhat decomposition 73
Burnside ring 282
canonical factorisation 339, 342, 344
Cassou-Nogues-Frohlich class 335
cellular chain 196
central extension 228
classification of 319
central simple algebra 309
Chinburg invariant
local 312, 313, 315, 332
global 332
class-group
Hom-description of 115, 116, 147, 343
of maximal order 207, 361
class field theory 262, 348
class formation 300, 400
class number 243, 386
closed subhomomorphism 39
closure 39
coboundary 306, 310
cocycle 308, 309, 317
normalised 397
compact manifold 29
compactly supported cohomology 52
complexification 178
composition factor 316
conductor 247, 389
ideal 335
cone point 198
conjugacy class function 13
contracting homotopy 301
degree 194
de Rham-Witt complex 258
derived series 67
determinant 114, 178
determinantal congruence 121, 129, 130
direct sum 3
discriminant 248
division algebra 309, 311, 315
maximal subfield of 309
division ring 125, 309
Double Coset Formula 19, 22, 194, 214
centre 42, 142
chain complex 196

change of rings 183, 184
character 5, 9
Dirichlet 389
even/odd 389
Eisenstein polynomial 263

Euler
characteristic 52, 197, 243
factor 363
Euler's totient function 290
407
408
Index
exact functor 341

exact sequence 156, 192, 204, 304, 306, 310,
342
based 365
cohomology 306, 341
exponential 122, 332, 383
exterior power 3, 21, 108
factorisable functions 206

field
algebraic number 110

complete discrete valuation 110, 246,
255, 260, 282, 283
cyclotomic 379
finite 74, 263

local 72, 111, 120, 246, 302
residue 126, 246
separable residue 246, 248, 277
splitting 112, 205
totally real 373
unramified extension of 120, 131, 152,
316
maximal 303
tame extension of 138, 319, 328, 331, 334
totally ramified extension 263
wild extension 331, 332, 336, 361
fixed points 28, 105, 194, 196, 249, 397
Fourier transform 78
fractional ideal 205, 338
principal 206
Frobenius
automorphism 74, 93, 120, 303
Reciprocity 18, 22
Frohlich-Chinburg
conjecture 336
Frohlich's conjecture 107, 336

Hom-description 115
fundamental class 299
Galois cohomology 258
Gauss sum 389, 391
group 1
abelian 15
absolute Galois 113, 205
Borel 73
cyclic 26, 56, 62, 118
dihedral 11, 14, 118, 228
decomposition 112, 331
elementary 30
Galois 112, 152, 246, 302
generalised quaternion 49, 136, 180, 191,

228
Grothendieck 114, 212, 337

M-group 27, 67
nilpotent 27, 67, 219
p-elementary 159, 161
p-group 67, 69, 219, 234
ramification 246
semi-dihedral 136, 228

semi-direct product 159
solvable 27, 67
supersolvable 27
Sylow 67, 186
symmetric 28
Weyl 143, 172
Hasse invariant 309, 311

henselianisation 259, 279
Herbrand difference 231
Hermitian 21
Hilbert's `Theorem 90' 99
homology 29, 197
reduced 196
idel 110
unit 112
of a group-ring 112
idempotent 56, 338
indecomposable 4
inductivity 55
in dimension zero 287
integers
algebraic 110
Gaussian 178
unit group of 112
p-adic 111
Jacobson radical 125

join 198
Kt-idele 367, 375
Kato-Swan conductor 260
kernel group 115, 337
Kronecker product 3, 6, 21
Kulakoff's theorem 137
Kummer's lemma 240
Leopoldt's formula 389

localisation 196
locally cyclic 371, 376
logarithm
group-ring 126
1-adic 122
Mackey's irreducibility criterion 20

Mashke's theorem 3
maximal pair 92
normaliser of 92, 93
maximal order 115, 173, 205
maximal torus 27
Mayer-Vietoris sequence 203
minimal polynomial 84, 263
Mobius function 25, 48, 67
module
cohomologically trivial 183, 302, 307
finite rank 207
Index
locally free 113
locally freely presented 207

projective 176, 205
rank one 113
Swan 117, 134, 189, 315
monoid 114
Newton polynomial 108

norm 100, 102
reduced 142, 170
normalised exponential valuation 123
normaliser 27, 190
orbit 28, 52
orthogonal group 4, 21
p-adic class number formula 386

p-adic L-function 389
p-adic regulator 386
P-adic topology 111
P-completion 111, 332
permutation matrix 2, 28
place
Archimedean 111
finite 111
infinite 111
poset 25, 34, 48, 67
prime 111
Archimedean 111
decomposition group of a 112

element 246, 259, 263
finite 111
regular 391
wild 331
projective resolution 300
pullback 234
409
regular 10
restriction of 17
ring 12
symplectic 5, 116, 206, 212, 335
function positive on 116
unit sphere of a 170, 194
unitary 4
Weil 73, 80
character-values of 88
residue degree 248, 266

resolvent 390
restricted determinant 172, 205

restriction homomorphism 17, 32
Schanuel's Lemma 192, 315
Schur inner product 7, 249
Schur's Lemma 7, 118
semi-simple 125
separable residue degree 247, 248, 266
Shintani
correspondence 98, 103
descent 98
lifting 98
norm 98
simplex 29, 52, 54
simplicial
chain group 29
complex 29
Sk6lem-Noether theorem 310
spectral sequence 316
stabiliser 46
Swan
conductor 247, 283
non-abelian 282, 283
representation 248
Swan subgroup 134

of a p-group 136
ramification
function 247, 253
group 246, 248
index 123, 256, 263
representation 2
character of 5, 20, 22, 128
centre of 42
complex 5, 13
contragredient of a 22
cuspidal 72, 80, 105
equivalent 2
faithful 294
induced 16
irreducible 4, 20, 89
K-rational 110
monomial 27
one-dimensional 2
orthogonal 5
permutation 2, 29
quaternionic 5, 116
rationality of 110
of a quaternion group 168, 178

of nilpotent group 219
symmetric power 21
sub-representation 2
Tate cohomology 231
Teichmuller map 251
tensor product 3, 21
totally positive 206, 216
trace 5
triangulation 29, 52, 54
uniformly ramified 257
uniformising element 259
unitary group 4, 21
unitriangular subgroup 73
valuation 123, 246

Wedderburn's Theorem 125
wedge 203
Whitehead lemma 164

Explicit Brauer Induction: Cambridge Studies in Advanced Mathematics: 40

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

Explicit Brauer Induction: Cambridge Studies in Advanced Mathematics: 40

Uploaded by

Copyright:

Available Formats

CAMBRIDGE STUDIES IN

Explicit Brauer Induction

W.M.L. Holcombe Algebraic automata theory

K. Petersen Ergodic theory

P.T. Johnstone Stone spaces

W.H. Schikhof Ultrametric calculus

J.-P. Kahane Some random series of functions, 2nd edition

H. Cohn Introduction to the construction of class fields

J. Lambek & P.J. Scott Introduction to higher-order categorical logic

H. Matsumura Commutative ring theory

C.B. Thomas Characteristic classes and the cohomology of finite groups

M. Aschbacher Finite group theory

J.L. Alperin Local representation theory

P. Koosis The logarithmic integral I

A. Pietsch Eigenvalues and s-numbers

S.J. Patterson An introduction to the theory of the Riemann zeta-function

H.J. Baues Algebraic homotopy

V.S. Varadarajan Introduction to harmonic analysis on semisimple Lie groups

W. Dicks & M. Dunwoody Groups acting on graphs

R. Fritsch & R. Piccinini Cellular structures in topology

20 H Klingen Introductory lectures on Siegel modular forms

M.J. Collins Representations and characters of finite groups

H. Kunita Stochastic flows and stochastic differential equations

P. Wojtaszczyk Banach spaces for analysts

A. Frohlich & M.J. Taylor Algebraic number theory

K. Goebel & W.A. Kirk Topics in metric fixed point theory

J.F. Humphreys Reflection groups and Coxeter groups

D.J. Benson Representations and cohomology I

D.J. Benson Representations and cohomology II

C. Allday & V. Puppe Cohomological methods in transformation groups

C. Soule et al Lectures on Arakelov geometry

A. Ambrosetti & G. Prodi A primer of nonlinear analysis

J. Palis & F. Takens Hyperbolicity, stability and chaos at homoclinic bifurcation

Y. Meyer Wavelets and operators

C. Weibel An introduction to homological algebra

W. Bruns & J. Herzog Cohen-Macaulay rings

CAMBRIDGE UNIVERSITY PRESS

Cambridge University Press

Cambridge University Press 1994

This publication is in copyright. Subject to statutory exception

Cambridge University Press has no responsibility for the persistence or

The class-group of a group-ring

Swan subgroups for nilpotent groups

`Funny that you should ask. As it happens I have a complete mathematical

This volume began as a one-term advanced graduate course in algebra

which I gave at the beginning of 1990 at McMaster University. As

to use than my original one - being a homomorphism rather than a

suited for exposition in a graduate course on algebra! As it happens, the

vaguely in my original motivation but had taken sufficiently long to

Chapter 2 begins with an account of Brauer's canonical version of

coefficients. The difficult part of the proof is to show that integrality

remarked previously, natural Explicit Brauer Induction formulae are not

Induction homomorphism, dG, which is due to Robert Boltje. The map,

of character-values or the Shintani norm map, offers an appealingly

erations, when applied to determinantal functions, would satisfy some

of Frohlich's conjecture (Taylor, 1981). It is also shown that Adams

refined to yield a technique for the detection of the class-group of a

that the Frohlich-Chinburg conjecture is true in the class-group of a

given to show that no such generalisation of the Swan conductor can

Chapter 7 treats the Galois module structure of the unit group in a

of the group-ring of a Galois group of certain invariants which are

- all this being accomplished within 40 hours of lectures. A one-term

the existence of the Explicit Brauer Induction homomorphism, ac, in