Cycloides Field-Serge Lang

Graduate Texts in Mathematics
59
Editorial Board
F. W. Gehring
P.R.Halmos
Managing Editor
C.C.Moore
Serge Lang
Cyclotomic
Fields
Springer-Verlag
New York Heidelberg Berlin
Dr. Serge Lang
Department of Mathematics
Yale University
New Haven, Connecticut 06520
USA
Editorial Board
P. R. Halmos F. W. Gehring C. C. Moore

Managing Editor Department of Mathematics Department of Mathematics
Department of Mathematics University of Michigan University of California
Indiana University Ann Arbor, Michigan 48104 Berkeley, CA 94720
Bloomington, Indiana 47401 USA USA
USA
AMS Subject Classification: 12C20, 12B30, 14G20
Library of Congress Cataloging in Publication Data

Lang, Serge, 1927-
Cyclotomic fields.
(Graduate texts in mathematics: 59)
Bibliography: p.
Includes index.
1. Fields, Algebraic. 2. Cyclotomy. 1. Title.
II. Series.
QA247.L33 512'.3 77-25859
All rights reserved.
No part of this book may be translated or reproduced in any form

without written permission from Springer-Verlag.
© 1978 by Springer-Verlag, New York Inc.

Softcover reprint of the hardcover I st edition 1978
9 8 7 6 5 432 1
ISBN-13: 978-1-4612-9947-9 e-ISBN-13: 978-1-4612-9945-5
DOl: 10.1007/978-1-4612-9945-5
Foreword
Kummer's work on cyclotomic fields paved the way for the development of
algebraic number theory in general by Dedekind, Weber, Hensel, Hilbert,
Takagi, Artin and others. However, the success of this general theory has
tended to obscure special facts proved by Kummer about cyclotomic fields
which lie deeper than the general theory. For a long period in the 20th century
this aspect of Kummer's work seems to have been largely forgotten, except
for a few papers, among which are those by Pollaczek [Po], Artin-Hasse
[A-H] and Vandiver [Va].
In the mid 1950's, the theory of cyclotomic fields was taken up again by
Iwasawa and Leopoldt. Iwasawa viewed cyclotomic fields as being analogues
for number fields of the constant field extensions of algebraic geometry, and
wrote a great sequence of papers investigating towers of cyclotomic fields,
and more generally, Galois extensions of number fields whose Galois group
is isomorphic to the additive group of p-adic integers. Leopoldt concentrated
on a fixed cyclotomic field, and established various p-adic analogues of the
classical complex analytic class number formulas. In particular, this led him
to introduce, with Kubota, p-adic analogues of the complex L-functions
attached to cyclotomic extensions of the rationals. Finally, in the late 1960's,
Iwasawa [Iw 1I] .made the fundamental discovery that there was a close
connection between his work on towers of cyclotomic fields and these
p-adic L-functions of Leopoldt-Kubota.
The classical results of Kummer, Stickelberger, and the Iwasawa-
Leopoldt theories have been complemented by, and received new significance
from the following directions:
1. The analogues for abelian extensions of imaginary quadratic fields in
the context of complex multiplication by Novikov, Robert, and Coates-
Wiles. Especially the latter, leading to a major result in the direction of the
v
Foreword
Birch-Swinnerton-Dyer conjecture, new insight into the explicit reciprocity

laws, and a refinement of the Kummer-Takagi theory of units to all levels.
2. The development by Coates, Coates-Sinnott and Lichtenbaum of an
analogous theory in the context of K-theory.
3. The development by Kubert-Lang of an analogous theory for the units
and cuspidal divisor class group of the modular function field.
4. The introduction of modular forms by Rihet in proving the converse of
Herbrand's theorem.
5. The connection between values of zeta functions at negative integers
and the constant terms of modular forms starting with Klingen and Siegel,
and highly developed to congruence properties of these constant terms by
Serre, for instance, leading to the existence of the p-adic L-function for
arbitrary totally real fields.
6. The construction of p-adic zeta functions in various contexts of elliptic
curves and modular forms by Katz, Manin, Mazur, Vishik.
7.. The connection with rings of endomorphisms of abelian varieties or
curves, involving complex mUltiplication (Shimura-Taniyama) and/or the
Fermat curve (Davenport-Hasse-Weil and more recently Gross-Rohrlich).
There is at present no systematic introduction to the basic cyclotomic
theory. The present book is intended to fill this gap. No connection will be
made here with modular forms, the book is kept essentially purely cyclotomic,
and as elementary as possible, although in a couple of places, we use class
field theory.
Some basic conjectures remain open, notably: Vandiver's conjecture that
h + is prime to p.
The Iwasawa-Leopoldt conjecture that the p-primary part 0f C- is cyclic
over the group ring, and therefore isomorphic to the group ring modulo
the Stickelberger ideal. For prime level, Leopo.!dt and Iwasawa have shown
that this is a consequence of the Vandiver conjecture. Cf. Chapter VI, §4.
Much of the cyclotomic theory extends to totally real number fields, as
theorems or conjecturally. We do not touch on this aspect of the question.
Cf. Coates' survey paper [Co 3], and especially Shintani [Sh].
There seems no doubt at the moment that essential further progress will be
closely linked with the algebraic-geometric considerations, especially via the
Fermat and modular curves.
I am very much indebted to John Coates, Ken Ribet and David Rohrlich
for their careful reading of the manuscript, and for a large number of
suggestions for improvement.
New Haven, Connecticut SERGE LANG

1978
vi
Contents
Foreword v
CHAPTER 1
Character Sums 1
1. Character Sums Over Finite Fields 1
2. Stickelberger's Theorem 6
3. Relations in the Ideal Classes 14
4. Jacobi Sums as Hecke Characters 16
5. Gauss Sums Over Extension Fields 20
6. Application to the Fermat Curve 22
CHAPTER 2
Stickelberger Ideals and Bernoulli Distributions 26
1. The Index of the First Stickel berger Ideal 27
2. Bernoulli Numbers 32
3. Integral Stickelberger Ideals 43
4. General Comments on Indices 48
5. The Index for k Even 49
6. The Index for k Odd 50
7. Twistings and Stickelberger Ideals 51
8. Stickel berger Elements as Distributions 53
9. Universal Distributions 57
10. The Davenport-Hasse Distribution 61
CHAPTER 3
Complex Analytic Class Number Formulas 69
1. Gauss Sums on Z/mZ 69
2. Primitive L-series 72
vii
Contents
3. Decomposition of L-series 75
4. The (± l)-eigenspaces 81
5. Cyclotomic Units 84
6. The Dedekind Determinant 89
7. Bounds for Class Numbers 91
CHAPTER 4
The p-adic L-function 94
1. Measures and Power Series 95
2. Operations on Measures and Power Series 101
3. The Mellin Transform and p-adic L-function 105
4. The p-adic Regulator 112
5. The Formal Leopoldt Transform 115
6. The p-adic Leopoldt Transform 117
CHAPTER 5
Iwasawa Theory and Ideal Class Groups 123
1. The Iwasawa Algebra 124
2. Weierstrass Preparation Theorem 129
3. Modules over Z,,[[X]] 131
4. Z,,-extensions and Ideal Class Groups 137
5. The Maximal p-abelian p-ramified Extension 143
6. The Galois Group as Module over the Iwasawa Algebra 145
CHAPTER 6
Kummer Theory over Cyclotomic Zp-extensions 148
1. The Cyclotomic Z,,-extension 148
2. The Maximal p-abelian p-ramified Extension of the Cyclotomic
Z,,-extension 152
3. Cyclotomic Units as a Universal Distribution 157
4. The Leopoldt-Iwasawa Theorem and the Vandiver Conjecture 160
CHAPTER 7
Iwasawa Theory of Local Units 166
1. The Kummer-Takagi Exponents 166
2. Projective Limit of the Unit Groups 175
3. A Basis for U(x) over A 179
4. The Coates-Wiles Homomorphism 182
5. The Closure of the Cyclotomic Units 186
CHAPTER 8
Lubin-Tate Theory 190
1. Lubin-Tate Groups 190
2. Formal p-adic Multiplication 196
viii
Contents
3. Changing the Prime 200

4. The Reciprocity Law 203
5. The Kummer Pairing 204
6. The Logarithm 211
7. Application of the Logarithm to the Local Symbol 217
CHAPTER 9
Explicit Reciprocity Laws 220
1. Statement of the Reciprocity Laws 221
2. The Logarithmic Derivative 224
3. A Local Pairing with the Logarithmic Derivative 229
4. The Main Lemma for Highly Divisible x and IX = Xn 232
5. The Main Theorem for the Symbol <x, xn>n 236
6. The Main Theorem for Divisible x and IX = unit 239
7. End of the Proof of the Main Theorems 242
Bibliography 244
Index 251
ix
Notation
ZeN) = integers mod N = Zj NZ.

If A is an abelian group, we usually denoted by AN the elements x E A
such that Nx = O. Thus for a prime p, we denote by Ap the elements of order
p. However, we also use p in this position for indexing purposes, so we rely
to some extent on the context to make the intent clear. In his book, Shimura
uses A[p] fot the kernel of p, and more generally, if A is a module
over a ring, uses A[a] for the kernel of an ideal a in A. The brackets are
used also in other contexts, like operators, as in Lubin-Tate theory. There is
a dearth of symbols and positions, so some duplication is hard to avoid.
We let A(N) = AjNA. We let A(p) be the subgroup of A consisting of all
elements annihilated by a power of p.
xi
1
Character Sums
Character sums occur all over the place in many different roles. In this
chapter they will be used at once to represent certain principal ideals, thus
giving rise to annihilators in a group ring for ideal classes in cyclotomic fields.
They also occur as endomorphisms of abelian varieties, especially Jacob-
ians, but we essentially do not consider this, except very briefly in §6. They
occur in the computation of the cuspidal divisor class group on modular
curves in [KL 6]. The interplay between the algebraic geometry and the
theory of cyclotomic fields is one of the more fruitful activities at the moment
in number theory.
§1. Character Sums Over Finite Fields

We shall use the following notation.
F = Fq = finite field with q elements, q = pn.
ZeN) = ZjNZ.
e = primitive pth root of unity in characteristic O. Over the complex
numbers, e = e21tilp.
Tr = trace from F to Fp.
JlN = group of Nth roots of unity.
2: F -i>- Jlp the character of F given by
2(x) = eTr(X).
X: F* -i>- Jlq -1 denotes a character of the multiplicative group.

We extend X to F by defining X(O) = O.
The field Q(JlN) has an automorphism (1' -1 such that
(1'_1:(1-+(-1.
1. Character Sums
If CI. E Q(JiN) then the conjugate ex denotes (J -lCl.. Over the complex numbers,
this is the complex conjugate.
The Galois group of Q(JiN) over Q is isomorphic to Z(N)*, under the map
where
Let/, g be functions on Fwith values in a fixed algebraically closed field of

characteristic O. We define
S(f, g) = L f(x)g(x).
XEF
We define the Fourier transform Tfby
Tf(y) = L f(x)Je( -xy) = Lf(x)e-Tr<XY).

XEF
Then Tf is again a function on F, identified with its character group by Je,

and T is a linear map.
Theorem 1.1. Let f- be the function such that f-(x) = f( -x). Then
T2f = qf-, that is
T2f(z) = qf(-z).
Proof We have
T2f(z) = L Lf(x)Je( - yx)Je( -zy)

Y x
= Lf(x - z) LJe(-yx).
x 1I
If x =f. 0 then y 1---7 Je(yx) is a non-trivial character, and the sum of the
character over F is O. Hence this last expression is
= qf( -z)
as desired.
We define the convolutionf * g between functions by the formula
(f * g)(y) = L f(x)g(y - x).

x
A change of variables shows that
2
Theorem 1.2. For functions J, g on F we have
T(f * g) = (Tf)(Tg)
1
T(fg) = q- Tf* Tg.
.
Proof For the first formula we have
T(f*g)(z) = L(f*g)(Y)A(-ZY) = LLf(x)g(y - X)A(-ZY).

11 11 x
We change the order of summation, let t = Y - x, Y = x + t, and find
= LfCx)A( -zx) Lg(t)A( -zt)

x t
= (Tf)(Tg )(z),
thereby proving the first formula.
The second formula follows from the first because T is an isomorphism

on the space of functions on F, so that we can write f = Tit and g = Tg l
for some functions It, gl' We then combine the first formula with Theorem
1.1 to get the second.
We shall be concerned with the Gauss sums (Lagrange resolvant)
Sex, A) = sex) = L XCU)A(U)

u
where the sum is taken over U E F*. We ceuld also take the sum over x in F
since we defined X(O) = O. Since A is fixed, we usually omit the reference to A
in the notation. The Gauss sums have the following properties.
GS O. Let Xl be the trivial character 1 on F*. Then
This is obvious from our conventions. It illustrates right at the beginning the
pervasive fact, significant many times later, that the natural object to con-
sider is -Sex) rather than Sex) itself. We shall also write
SCI) = SCI, A),
but the convention remains in force that even for the trivial character, its
value at 0 is O.
GS 1. For any character X '=f. 1, we have TX = X( -I)S(x)X- l •
3
1. Character Sums
Proof We have
TX(Y) = L.. x(x),.l( - yx) .

If y = 0 then TX(Y) = 0 (summing the multiplicative character over the
multiplicative group). If Y"# 0, we make a change of variables x = -ty-I,
and we find precisely the desired value
X( -1)S(x)X(y-l).
GS 2. We have S(i) = X( -1)S(x) andfor X "# 1, S(x)S(i) = X( -1)q, so

S(x)S(x) = q, for X "# 1.
Proof Note that T2X = T(X( -1)S(x)X-l) = S(x)S(X-l)X. But we also
know that T2X = qX-. This proves GS 2, as the other statements are obvious.
Over the complex numbers, we obtain the absolute value
We define the Jacobi sum
J(xI> X2) = - L.. Xl(X)X2(1 - x) .
Observe the minus sign, a most useful convention. We have
J(1, 1) = -(q - 2).
GS 3. If XIX2 "# 1 then
J(x X) = - S(1)S(x2) .
1, 2 S(XIX2)
In particular, J(I, X2) = JU1, 1) = 1. If XIX2 = 1 but not both Xl, X2
are trivial, then
JUI> X2) = Xl( -1).

Proof We compute from the definitions:
S(1)S(x2) = L L Xl(X)X2(y),.l(X + y)
.. II
= L L Xl(X)X2(y - x)A(y)
.. II
= L: L Xl(X)X2(U -
.. u¢o
x),.l(u) + L Xl(X)X2( -x) .
..
4
If XlX2 =F 1, the last sum on the right is equal to O. In the other sum, we inter-
change the order of summation, replace x by ux, and find
2: XIX2(U)A.(U) 2: Xl(x)X2(1 -
u x
x),
thus proving the first assertion of GS 3. If XlX2 = 1, then the last sum on the
right is equal to Xl( -l)(q - 1), and the second assertion follows from
GS2.
Next we give formulas showing how the Gauss sums transform under
Galois automorphisms.
GS 4.
Proof Raising to the pth power is an automorphism of F, and therefore
Tr(xP) = Tr(x).
Thus S(xP) is obtained from S(X) by permuting the elements of F under

x 1-7 x p • The property is then obvious.
Let m be a positive integer dividing q - 1, and suppose that X has order m,

meaning that
Then the values of X are in Q(l1m) and
For any integer c prime to m we have an automorphism (le,l of Q(l1m, I1p)

such that
(l e,l: ( 1-7 (e and (l e,l is identity on I1p.
For any integer v prime to p, we have an automorphism (ll,v such that
We can select v in a given residue class mod p such that v is also prime to m.
In the sequel we usually assume tacitly that v has been so chosen, in particular
in the next property.
GS 5.
Proof The first is obvious from the definitions, and the second comes by
making a change of variable in the Gauss sum,
5
1. Character Sums
Observe that
0"1,){X) = eVTr(x) = eTr(Vx) = A(VX).
The second property then drops out.
The diagram of fields is as follows.
From the action of the Galois group, we can see that the Gauss sum
(Lagrange resolvant) satisfies a Kummer equation.
Theorem 1.3. Assume that X has order m.

(i) S(X)m lies in Q(/lm).
(ii) Let b be an integer prime to m, and let O"b = O"b,1' Then S(x)b-"b lies in
Q(/lm)·
Proof In each case we operate on the given expression by an automorphism
0"1,v with an integer v prime to pm. Using GS 5, it is then obvious that the
given expression is fixed under such an automorphism, and hence lies in
Q(/lm)·
§2. Stickelberger's Theorem

In the first section, we determined the absolute value of the Gauss sum.
Here, we determine the prime factorization. We shall first express a character
in terms of a canonical character determined by a prime.
Let V be a prime ideal in Q(/lq-1), lying above the prime number p. The
residue class field of V is identified with F = Fq. We keep the same notation
as in §l. The equation X q - 1 - 1 = 0 has distinct roots modp, and hence
reduction mod V induces an isomorphism
~ F* -- F*
/lq - 1 ---l>- q.
Phrased another way, this means that there exists a unique character w of
F* such that
w(u) mod V = u.
This character will be called the Teichmuller character. This last equation
will also be written in the more usual form
w(u) == u (mod V).
6
§2. Stickel berger's Theorem
The Teichmuller character generates the character group of F*, so any

character X is an integral power of w.
We let
n=e-l.
Let ~ be a prime ideal lying above p in Q(jlq-l, jlp). We use the symbol
A ,...., B to mean that AlB is a unit, or the unit ideal, depending whether A, B
are algebraic numbers or (fractional) ideals. We then have
p ,...., ~P-l
because elementary algebraic number theory shows that p is totally ramified

in Q(e), and p is totally ramified in Q(jlq-l, jlp).
Let k be an integer, and assume first that 0 ~ k < q - 1. Write the
p-adic expansion
with 0 ~ kj ~ p - 1. We define
I s(k) = ko + kl + ... + k n - 1 •
For an arbitrary integer k, we define s(k) to be periodic mod q - 1, and

defined by the above sum in the range first assumed. For convenience, we also
define
to be the product of the k i ! in the first range, and then also define y(k) by
(q - I)-periodicity for arbitrary integers k. If the dependence on q is
desired, one could write
sik) and Ylk).
Theorem 2.1. For any integer k, we have the congruence
S(w- k , eTr) _ -1
(e - I),(k) = y(k) (mod ~).
In particular,
ord$ S(w- k ) = s(k).
Remark. Once more, we see how much more natural the negative of the
Gauss sum turns out to be, for we have
-S(w-k, 2) _ 1
ns(k) = y(k) (mod ~)
with 1 instead of -Ion the right-hand side.
7
1. Character Sums
Proof of Theorem 2.1. If k = 0 then the relation of Theorem 2.1 is clear

because both sides of the congruence to be proved are equal to -1. We
assume 1 ::; k < q - 1, and prove the theorem by induction. Suppose first
that k = 1. Then
S(w- k ) =L w(u)-leTr<u)
u
= L w(u)-I(1 + n)Tr<u)
= L w(u)-I(1 + (Tr u)n + O(n2»
(interpreting Tr u as an integer in the given residue class mod p). But
W(U)-I Tr(u) == u-I(u + uP + ... + upn - 1 ) mod ~

== 1 + up - I + ... + U pn - L I •
Each u H- uPJ - I is a non-trivial character of F*. Hence
L W(U)-I Tr(u) == q - 1 == -1 (mod ~)

and therefore
S(W-I)
- - == -1 (mod~)
n
thus proving the theorem for k = 1.
Assume now the result proved for k - 1, and write
for 1 < k < q - 1. We distinguish two cases.
Case 1. plk, so we can write k = pk' with 1 ::; k' < q - 1. Then trivially
s(k) = s(k') and y(k) = y(k')

because k has the same coefficients k i as k', shifted only by one index. Let
Up = U p ,1, so Up leaves e fixed. Since
we find that applying Up to the inductive congruence
S(w- k ') -1
nS(k') == y(k') (mod~)
yields a proof for the present case, because Up is in the decomposition group
of~, whence up~ = ~.
8
Case 2. p t k. Then 1 ::::;; k o• Furthermore,
s(k) = s(k - 1) + 1 and y(k - 1) = (k o - I)! k 1 !··· k n - 1 !.
Then
S(w- k) S(W- 1W-(k-l») _ S(w- 1) S(W-(k-l» -I
~ ns(k) = - n - ns(k-l) J(w 1, w (k 1»
= -1. -1 -1 ( d ~)
- y(k - 1) J(w 1, W (k 1» mo .
To conclude the proof, it will suffices to get the right congruence for J. We
use GS 3 from §I, to get:
-J(w-I, W-(k-l» == L: u- 1(1 - U)-(k-l)+Q-l (mod ~),
and the sum is at first taken for u =f 0, 1, but with the additional positive
exponent q - I which does not change anything, we may then suppose that
the sum is taken for u =f 0 in F. Hence we get further
If j =f 1 then L ui -1 = 0, so we get the further congruence
-J(w-I, W-(k-l» == (-I)(q - k)(q - ) == -ko (mod ~),

thereby proving the theorem.
Having obtained the order of the Gauss sum at one prime above p, we also
want the full factorization. Suppose that m is an integer > 1 and that p t m.
Let .p be a prime ideal above p in Q(flm) and let
N.p = q = pn.
Let k be an integer such that
~1 has order m in Q/Z.

q-
Let <t> denote the smallest real number ~ 0 in the residue class mod Z of a
real number t. Let
G = Gal(Q(flm)/Q).
Define the Stickelberger element in the rational group ring
e(k,.p) = L:
ce2:(m)'
< q
~ 1 )lJ'c -1 E Q[G].
9
1. Character Sums
Let ~ be the prime ideal in Q(fJ.m, fJ.p) lying above .p. Let w as before be the
Teichmuller character on Fr
We let U e = Ue,l'
Theorem 2.2. We have the factorization

S(w-k) '" ~(P-l)O(k,P) ""' .p0(k,P).
Proof We have
orda;lp S(w- k) = ord!j3 UeS(W-k)

= ord!j3 S(w- kC )
= s(kc)
by Theorem 2.1. On the other hand, the isotropy group of .p in the Galois
group G consists of the powers
{Upl} for i = 0, ... , n - 1.

Hence in the ideal .p0(k) the prime occurs with multiplicity
<
U;l.p
2: -
n-l
1=0q - 1
.
kcpl )
Hence to prove Theorem 2.2 it will suffice to prove:
Lemma 1. For any integer k we have
s(k) = (p - 1).z
n-l<
1=0
--?-
q
k
1
. I )
Proof We may assume that 1 :::; k < q - 1 since both sides are (q - 1)-
periodic in k, and the relation is obvious for k = O. Since pn == 1 (mod q - 1)
we find:
k = ko + k1P + ... + kn_1pn-l
pk == k n - 1 + kop + ... + k n_2pn-l (mod q - 1)
p 2k == k n- 2 + kn-1P + ... + k n _3pn-l (mod q - 1)
Hence
< kl ) =
q-1
right-hand side of ith equation.
q-1
<
Summing yields
n~l kpi ) = s(k)(1 + p + ... + pn-l) = s(k) _1_,

I~ q - 1 q - 1 p - 1
thereby proving the lemma.
10
§2 Stickelberger's Theorem
In Theorem 2.2 we note that the Gauss sum is not necessarily an element
of Q(flm), and the equivalence of ideals is true only in the appropriate ex-
tension field. Similarly, the Stickel berger element has rational coefficients.
By the same procedure, we can both obtain an element in Q(flm) and a corre-
sponding element in the integral group ring, as follows.
For any integers a, b E Z and any real number t, we have
b<t) - <bt) E Z and <at) + <bt) - «a + b)t) E Z.

The proof is obvious. Let us define R = Z[G], and
I = ideal of R generated by all elements Ub - b with b prime to m.
Then the above remark shows that
I() c R = Z[G].
Although we won't need it, we may prove the converse for general insight.
The matter is analyzed further in Chapter 2, §3.
Lemma 2. We have I() = R() n R.

Proof Note that mEl because
m = -(U1+m - (1 + m)).
Suppose that an element of R() lies in R, that is
with z(b) E Z. Then
whence
"'2 z(b)b == 0 (mod m),
and L: z(b)b is in 1. But then
"'2 Z(b)Ub = "'2 Z(b)(Ub - b) + "'2 z(b)b
is in I, thus proving the lemma.
It will be convenient to formulate the results in terms of the powers of one

character, depending on the integer m. Thus we let
< XI' = wp(NI'-l)/m
11
1. Character Sums
where Wp is the Teichmuller character. We define the Stickelberger element

of level m by
Oem) = 2: <~)(J;;1.
CEZ(m)'m
As a special case of Theorem 2.2, we then obtain the factorization
FAC 1.
Therefore, if b is an integer prime to m, and (Jb = (Jb,b then
FAC 2.
In FAC 2 the algebraic number on the left lies in Q(flm), and the group ring
element O(m)(b - (Jb) lies in Z[G], namely
Thus we have the ideal factorization of the (b - (Jb)-power of the Gauss

sum in terms of powers of conjugates of the prime ~ in Q(flm).
We return later to the application of this factorization to the study of the
ideal classes in the cyclotomic field, but it is worth while here to mention the
simplest consequence. In every ideal class there exists an ideal prime to m.
Since the ideal
is principal for every prime ~ t m, we find:

Theorem 2.3. Let ((j be the ideal class group of Q(Pm). Then for all b prime
tom,
annihilates ((j.
For each integer r let
We are now allowing r to have common factors with m. Let:
vi! = module generated over Z by all elements Or with r E Z, called the

Stickelberger module,
y = vi! n R, called the Stickelberger ideal.
Observe that vi! is also an R-module.
12
Theorem 2.4. The Stickelberger ideal annihilates the ideal class group of
Q(Pm).
Proof. Let
IX = '2 z(r)Or(m)
r
E R
be an element of the Stickelberger ideal, with z(r) E Z, and the sum taken with
only a finite number of coefficients t= O. Then
'2 z(r)r == 0 mod m.

r
By Theorem 2.2 we have the factorization
and it is immediately verified that the left-hand side lies in Q(flm) by using
GS 5 of the preceding section. This proves the theorem.
Next we look at the Jacobi sums. If d is an integer, then d operates in a

natural way on RjZ by multiplication. We denote this operation by [d].
Thus on representatives, we let
[d]<t) = <dt), t E R.
It is convenient to let
Recall the Jacobi sum for XIX2 t= 1:
Let ah a2 be integers, al + a2;jiE 0 mod m. Then from FAC 1 we get:
FAC3.
where
and LI[ah a2]8(m) E Z[G] lies in the integral group ring. We know that the
Jacobi sum lies in Q(flm), so again we have an ideal factorization of an element
of Q(flm).
13
i. Character Sums
It will be convenient to introduce an abbreviation. Let
denote a pair of integers. We let
In several applications, e.g., in the next section, the level m is fixed, and
consequently we omit m from the notation, and write simply
8(m)[a] = 8[a].
If d is an integer prime to m then trivially
The next two sections are logically independent and can be read in any order.
They pursue two different topics begun in §2.
§3. Relations in the Ideal Classes

Let G = Gal(Q(Jlm)/Q), so that elements of G can be written in the form a e ,
with c E Z(m)*. We recall the Stickelberger element
from formulas FAC 1 and FAC 2. Let
I = ideal of Z[G] generated by all elements b - ab , with integers b prime

to m.
Let p be prime number prime to the Euler function ¢(m). For instance, if
m = p itself, the prime p does not divide p - 1. The character group on G
takes its values in ¢(m)th roots of unity. We let q = pn be a power of p such
that ¢(m) divides q - 1. We let Oq be the ring of p-adic integers in the un-
ramified extension of Zp of degree n, so that Oq/pOq = oip) is the finite field
with pn = q elements. Then Oq contains the ¢(m)th roots of unity. If m = p
then we take q = p and Oq = Zp.
Let'(j' be the ideal class group of Q(Jlm), and '(j'(p) its p-primary component.
We have an isomorphism
The elementary divisors of '(j'(p) over Zp are the same as the elementary
divisors of
Oq ® '(j'(p) over Oq.
14
§3. Relations in the Ideal Classes
If A is an oq-ideal, on which G operates, we let AU) be the x-eigenspace.

We let
Ix = oq-ideal generated by all elements b - X(b) with integers b prime to m.
By abuse of notation, we write often X(b) instead of X(tTb)' The important

special case we shall consider is when m = p, in which case it is easy to
determine IX" We assume p :2: 3.
Lemma 1. (i) If X = w is the Teichmuller character, then Ix = (P).

(ii) If X is non-trivial and not equal to the Teichmuller character, then
Ix = (1).
Proof For (i), we can take an integer b of the form
b=(+pu
where u is ap:-adic unit, and ( = web) is a (p - l)th root of unity. This makes
(i) clear, and (ii) is obvious, from the definitions.
In the next sections we shall deal with Bernoulli numbers systematically.

For the moment, we need only a special case, so we define ad hoc the first
Bernoulli polynomial
and the first Bernoulli number BI = --t, its constant term. For any function
fan Z(m) we define
BI,t = :2 f(X)BI«~»)'
xeZ(m) m
In particular,
BI,x = :2 «~)
m
ceZ(m)'
- 2!)X(C).
If X is non-trivial, then L X(c) = 0, and hence in this case,
Then in the present terminology, Theorem 2.3 can be reformulated as

follows.
Theorem 3.1. For non-trivial X, the ideal Bl,'j/x annihilates ~(P)U).
15
1. Character Sums
Corollary 1. Assume that m p is prime ;:::: 3. If X is not equal to the

=
Teichmuller character and is non-trivial, then
ord B1,xIx = ord B q .

Proof Immediate from the lemma and the theorem.
Corollary 2. If X is equal to the Teichmuller character then B1,xIx = (1),

and ct'(Pl(x) = 0.
Proof Mod Zp, we have the congruence
B1,w- 1 = - L
Pc=l
CW(C)-l == - L 1 == p - -
I P- 1 IP-l-l
Pc=l P
(mod Zp).
Hence B1,x has a pole of order 1 at p. Lemma l(i) concludes the proof.
Corollary 3 (Herbrand's theorem). Assume again that m = p. Let X = w1-k,

with 2 :0; k :0; P - 2. Ifct'(Pl(x) i= 0, then plBk' where Bk is the kth Bernoulli
number.
Proof In the next chapter Theorem 2.5, we shall prove the congruence
1
- B n,w k-n
n == k-1 Bk (modp)
for k in the given range, and any positive integer n. By Corollary 1, we know
that B1,x annihilates ct'(Pl(x), and
If p does not divide B k, it follows that Bq is a p-unit, whence ct'(Pl(x) = 0, thus

proving Herbrand's theorem.
The converse of Herbrand's theorem has been proved by Ribet [Ri].

For analogues on the modular curves, see the [KL] series, especially [KL 6].
The reader interested in pursuing the ideas of this section may skip the
rest of this chapter, read the first section of Chapter 3, and then go to
Chapter 5.
§4. Jacobi Sums as Hecke Characters

Let, throughout this section be a fixed primitive mth root of unity. We con-
sider the additive group
Z(m)<21 = Z(m) x Z(m),
of order m2 • Its elements will be denoted by
16
The dot product is the usual one, a·b = alb l + a2b2' For any functionfon
Z(m)'2) we have its Fourier transform], and the inversion formulas:
(*) f(a) = L j(bW'a

b
(**)
whose verifications are simple exercises.

For any prime ideal ~ in Q(Jl.m) not dividing m, and a E Z(m)<2) we define
We extend the definition to fractional ideals of Q(Jl.m) prime to m by multi-

plicativity, thus defining lea, a) for all a prime to m. We have:
J O. J(O,~) = -(N~ - 2).
We get J(O, a) by multiplicativity. We also need the congruence
J 1. J(O, a)Na == 1 mod m 2 •
By multiplicativity it suffices to prove it for prime ideals. In that case it is

immediate, since m divides N~ - 1, and by J 0,
°
If all or a2, or al + a2 == mod m, then we shall say that a is special.
Otherwise we say that a is non-special. The absolute value of the Gauss sum
determined in GS 2 immediately implies a corresponding result for the Jacobi
sum, namely:
J2. J(a, a)l(a, a) = Na if a is non-special.
If a is special, a =I 0, note that J(a, a) = 1 or -l. In all cases, we have
J 3. lea, ~) =- L XlJ
u
a1 (U)XlJ a2 (1 - U) = LJ(b, ~Kb.a
b
where the Fourier coefficient -J(b,~) is the number of solutions u of the

equations
17
1. Character Sums
By multiplicativity, it follows that the Fourier coefficients J(b, a) are integers

for arbitrary a, that is
J(b, a) E Z.
For the rest of this section, it will be convenient to assume that all number
fields are contained in the complex numbers.
We have seen that 8[a] is in the integral group ring Z[G]. For any non-zero
element a E Q(Jlm), we let
w(a, a) = J(a, (a»a- 8[al if a is non-special,

w(a, a) = J(a, (a» if a is special, a =1= °
w(O, a) = 1.
As usual, (a) is the principal (fractional) ideal generated by a.

If d is an integer prime to m, then trivially from GS 5,
(J'dJ(a, a) = J(da, a) and (J'dw(a, a) = w(da, a).
Theorem 4.1. The algebraic number w(a, a) is a root of unity.

Proof As (a) ranges over all principal fractional ideals, the numbers
w(a, a) form a group. It will therefore suffice to prove that these numbers
have absolute value 1, for then their conjugates also have absolute value 1,
and these numbers form a finite group. In case a is special the theorem is
true by definition. Otherwise we can use J 2, so that
J(a, (a»J(a, (a» = Na:.
°
On the other hand, the product of a8[al and its conjugate is equal to Na
under the hypothesis that a1 + a2 t= mod m. Indeed, we have
If t is a real number and not an integer, then
<t>+(-t)=I,
and
'"L (J'c-1
ceZ(m)'
18
operates multiplicatively like the absolute norm. The desired relation for the
product of 1X9[a] and its conjugate follows at once. The theorem follows by
using J 2, the analogous relation for the Jacobi sums.
The next theorem was proved originally by Eisenstein for prime level, and
by Weil [We 2] in the general case, which we follow.
Theorem 4.2. If IX is an algebraic integer in Q(f.1m) , and IX == 1 (mod m 2)

then/or all a we have w(a, IX) = 1, that is,
J(a, (IX)) = 1X6[a].
Proof We fix IX and view J, was functions of a, omitting IX from the nota-
tion. In the Fourier inversion relation, we know that the Fourier coefficients
J(b) are integers. But IX == 1 (mod m 2) implies that
w(a) == J(a) (mod m 2).

This is obvious from the definition if a =I 0, and follows at once from J 1
if a = O. Hence web) is an algebraic integer for all b. Furthermore, for d
prime to m,
= web).
It follows that web) E Z for all b. Now by the Plancherel formula,
Since we know that Iw(a)12 = 1, and web) is an integer for all b, it follows that
web) =I 0 for a single value of b, and is 0 for all other values of b. In particular,
for this special b,
w(a) = w(b)(b.a.
But w(O) = 1, so web) = l. Putting a = (1,0) and a = (0, 1) we get:
w(1,O) = J(l, 0) = 1 and w(l,O) = (b1
w(O, 1) = J(O, 1) = 1 and w(O, 1) = (b2.
It follows that
w(a) =1
for all a, thus proving the theorem.
19
1. Character Sums
§5. Gauss Sums Over Extension Fields

We prove in this section a theorem of Davenport-Hasse [D-H].
Theorem 5.1. Let F = Fq be the finite field with q elements, and let E be a
finite extension. Let
be the trace and norm from E to F. Let
Then
- SEUE, AE) = (- SU, A))lE:Fl.
Proof Let m = [E: F]. For any polynomial
f(X) = xn + C1X n- 1 + ... + Co
with coefficients in F, define
Then
tjJ: Monic polynomials of degree ~ lover F -+ F
is a homomorphism, i.e., satisfies
tjJ(fg) = tjJ(f)tjJ(g).
We write n(f) = degf From unique factorization we have the formula
where the product is taken over all monic irreducible polynomials over F.
Suppose f is of degree 1, say f( X) = X + c. Then we see that
L tjJ(f)xn(f) = SU, A)X.

nUl= 1
On the other hand, if n ~ 2 we have
L
n(f)=n
tjJ(f)xn = O.
Indeed,
and the sum over Cl in F on the right is 0, as desired.
20
§5. Gauss Sums Over Extension Fields
Therefore we find
(1)
Mutatis mutandis, using the variable xm instead of X, we get
(2)
where the product is taken over all monic irreducible polynomials Q over E,
and
We shall write the product over Q as
TI=TITI·
Q P QIP
Each irreducible polynomial P splits in E into a product
Let n = n(P) = deg P. Then

deg Q = nlr.
If a is any root of P, then [F(a):F] = n and the field F(a) is independent of

the chosen root. We have the following lattice of fields.
E F(a)
~~
F'
En F(a)
=
,1
F
All the polynomials Qi are conjugate over F, and their coefficients generate
the field F' = En F(a), of degree rover F. We have
r = (m, n).
These facts are all obvious from elementary field theory. Since
and
21
1. Character Sums
we get
l/IE(Q) = ex(CO(P»Jl(Cl(P»))lE:F'l
= l/I(p)m/r.
With a view towards (2), we conclude that
(3) TI (1 -l/IE(Q)xmn(Q») = (1 - l/I(p)m/rxmn/ry

Q\P
= TI
{mf'~ 1
(1 - l/I(P)(xny
= TI (l - l/I(p)(eX)n).
,m=l
For this last step, we observe that the map
gives a surjection of {lm -7- {lm/n and the inverse image of any element of
{lm/r is a coset of {lr since r = (m, n). This makes the last step obvious.
Substituting (3) in (2), we now find
1 + SE(XE, JlE)X m = JI IJ (1 - l/I(p\(ex)n(p»
= TI
~m~l
(1 + Sex, Jl)eX)
This proves the theorem..
§6. Application to the Fermat Curve

Although we do not return in this book to the applications of Gauss sums to
algebraic geometry, we cannot resist giving the application of Davenport-
Hasse [D-H], Hua-Vandiver [Hu-V], and Weil [We 1], [We 2], [We 3] to
the computation of the zeta function of a Fermat curve.
We keep things to their simplest case, the method applies much more
generally. We consider the Fermet curve V = V(d) defined by
with d ~ 2, defined over a finite field F with q elements. Again for simplicity,
we suppose that d divides q - 1, and therefore dth roots of unity are con-
tained in F.
We let w: F* -7- {lq-l be the Teichmuller character, and
x= character such that X(u) = W(U)<q-l)/d.
22
If a is an integer mod d, we let Xa(u) have the usual value if u # 0, and for
u = 0 we let:
xa(o) = 1 if a = 0,
Xa(o) = 0 if a # O.
For u in F, we let:
NaCu) = number of solutions x E F such that Xci = u.

Then
ifu = 0
if u # 0, u is not dth power in F
if u # 0, u is dth power in F.
Therefore
NaCu) = L
amodci
Xa(u).
Theorem 6.1. Let N be the number of points of V(d) (in affine space) in the
field F. Then
The sum is taken over integers a, b satisfying 0 < a < d and 0 < b < d,
and a + b "¥- 0 (mod d).
Proof We have
N = L
a,b,c
L
L(u,v,w)= 0
Xa(u)Xb(v)XC(w)
where the sum over u, v, w is taken over triples of elements of F lying on the
line
u +v+w= O.
The sum over a, b, c is taken over elements in Z mod d.

The term for which a = b = c = 0 yields a contribution of q2, that is the
number of points on the line in F.
Next, suppose that in the remaining sum, one of a, b, c is 0 but not all are
o in ZjdZ. Say a = 0 but b # O. Then we may write the sum
L
u+v+w=o
L
certain 'U, W
Xa(u)Xc(w) L
all veF
Xb(V),
and the sum on the far right is O. This shows that all the terms in the sum
23
1. Character Sums
with one, but not all, of a, b, c equal to 0 give a contribution O. Hence we get
N = q2 + 2:
O<a,b,c<d Il+V+W=O
2: Xa(u)Xb(v)XC(w)
where the sum over a, b, c is taken over positive integers satisfying the in-
dicated inequality.
If w = 0 then XC(w) = O. We may therefore assume that in the inner sum,
we have w t= O. We then put
u = u'w and v = v'w.
The inner sum then has the form
2: Xa+b+C(w) 2: Xa(u')Xb(v').
w"o Il'+v'=-l
If a + b + c ~ 0 mod d, then the sum on the left is O. Otherwise it is q - 1,

which we assume from now on. Since 0 < a, b, c < d, there is no such triple
(a, b, c) with a + b == 0 mod d, because any accompanying c would have to
equal d. Hence the sum over a, b, c is for a + b ~ 0 mod d, and then c is
uniquely determined. Changing back the variables u', v' to u" = -u', v" =
- v' and taking into account the value of the Jacobi sum yields the expression
as stated in the theorem.
Let N be the number of points of V(d) in projective space in the field F.

Then
N = 1 + (q - 1)N.
Therefore we obtain:
Corollary. N = 1 +q - 2: cta,b
where cta,b = Xa+b( -1)J(xa, Xb), and (a, b) are as in Theorem 6.1.
Let Nv be the number of points of V(d) in projective space over the field
Fv of degree v over F. The theorem applied to Fv instead of F yields an
analogous expression, the character X being replaced by Xv such that for
uEFv ,
This last expression is nothing but X composed with the norm map, in other
words, it is precisely the character lifted to the extension as in the preceding
section. The additive character is also lifted in a similar fashion. Therefore
by Theorem 5.1 we find
24
Note that the power of x( -1) also behaves in the same way as J when lifted
to Fv' Indeed, if q is odd then
1 + q + ... + q v -1 == v mod 2,
and if q is even, then 1 = - 1 in F.

The zeta function Z(V, T) is defined by the conditions
Z'IZ(T) =- LNT v v -1 and Z(O) = 1.
It is then immediate that

_ T1 (l - lY.a.b T )
Z(V(d), T) - (1 _ T)(1 - qT)
This is best seen by taking the logarithmic derivative of the last expression
on the right-hand side. The operator
Jf-+ I'IJ
is a homomorphism, so we take the operator for each linear term. Inverting
a geometric series we see that the logarithmic derivative of the last expression
on the right-hand side has precisely the power series
Since it has the value 1 at T = 0, it is the unique function having the desired
properties.
If finally one starts with the Fermat curve defined over the field of dth
roots of unity, and one reduces mod primes ,p not dividing d, one can take
the product of the zeta functions for the reduced curve over the correspond-
ing finite field. Then as Weil remarked, since the Jacobi sums are Hecke
characters, it follows that the Hasse zeta function
,eYed), s) = I1 Z(V(d), N,p-S)

Il.ra
is equal to a Hecke L-series (up to the obvious factors of the zeta function of
Q(J1a) at sand s - 1).
The computation of solutions in finite fields works in essentially the same
way for diagonal equations
as in Hua-Vandiver [Hu-V] and Weil [We 1,2,3]. The additional connection

with the Hasse zeta function for the curve over number fields was made by
Weil.
25
2
Stickelberger Ideals and
Bernoulli Distributions
The study of ideal classes or units in cyclotomic fields, or number fields

(Iwasawa, Leopoldt), of divisor classes on modular curves (e.g., as in [KL]),
of higher K-groups (Coates-Sinnott [Co 1], [Co 2], [C-S]) has led to purely
algebraic theorems concerned with group rings and certain ideals, formed
with Bernoulli numbers (somewhat generalized, as by Leopoldt). Such ideals
happen to annihilate these groups, but in many cases it is still conjectural
that the groups in question are isomorphic to the factor group of the group
ring by such ideals.
However, it is possible to study these ideals, the structure of their factor
group, and the orders of the factor groups in the group ring, without any
allusion to the applications to ideal classes, divisors, or units. This chapter
gives the foundations for such study, applicable to many contexts.
The first section gives Iwasawa's computation of the index of the Stickel-
berger ideal for k = 1, directly applicable to the ideal class group in cyclo-
tomic fields. Next we deal with the basic theory of Bernoulli numbers and
polynomials, and especially integrality theorems of Mazur and Coates-
Sinnott. The sections concerning Stickelberger ideals for k ~ 2 are taken
from Kubert-Lang [KL 8]. The last sections on distribution relations are
from [KL 5] and Kubert [Ku].
For a discussion of conjectures in the case of totally real number fields, cf.
Coates [Co 3], [Co 4], and the very general conjectures in Coates-Lichten-
baum [C-L].
The present chapter is organized so that a reader interested especially in
the structure of the ideal class group in the cyclotomic tower (the basic sub-
stantial example of the theory) can read the first section, and then can go
immediately to Chapter 3, followed by Chapter 5 without impairing the
logical understanding of the material. I followed this pattern when I taught
the course in 1977.
26
§1. The Index of the First Stickelberger Ideal
On the other hand, a reader especially eager to get into p-adic L-functions
can concentrate on this chapter and then read Chapter 4 as a continuation
omitting Chapter 3. Only the section on the p-adic regulator in Chapter 4 is
related to Chapter 3. Chapter 2 may then be interpreted as giving the basic
congruence properties of Bernoulli distributions, and Chapter 4 gives
essentially more (p-adically) global measure theoretic properties.
A third alternative is to see Chapters 3 and 4 as forming a pair, describing
side by side the complex and p-adic class number and regulator formulas
originally conceived by Leopoldt.

Let G ~ Z(m)* be the Galois group of Q(Pm), and assume that m is the
conductor of that field, so that m > 1, m is odd, or m is divisible by 4. We
let
M = !- order of G =,!-cp(m).
We let
R = Z[G],
For any G-module, we let A - be the (-I)-eigenspace for a -1. Then multipli-
cation by e- is the projection operator on this eigenspace (provided 2 is
invertible), and e- is the associated idempotent in the group algebra.
Lemma 1. We have R- = 2e- R = (l - a -l)R and
Proof The inclusion (1 - a -l)R c R- is clear. Conversely, let P be a set

of representatives in Z(m)* for Z(m)*j ± 1. Let
rx = L z(c)a c- 1 E R-
with coefficients z(c) E Z. Thus a -lrx = -rx. Then z( -c) = -z(c). If we let
[J = L z(c)a';-1,
ceP
then rx = (l - a -1)/3, thereby proving the lemma, because e- R is a free

abelian group of rank M.
We recall the primitive Stickelberger element
(J,' = L.,
'"
ceZ(m)'
<c) -
m
ac-1 •
27
2. Stickelberger Ideals and Bernoulli Distributions
We have written ()' instead of () because we are now setting more permanent
notation, and there is a more canonical element which has priority, namely
It is immediately verified that
(*) e-()' = (), andso () = ()-.

We are interested in R() n R. The next lemma does away with a possible
alternative definition of this ideal.
Lemma 2. R() n R = (R()' n R)-.
Proof Let T = R()' n R. Clearly
T- c e- R() = R() and T- c R,
so the inclusion ~ is obvious. Conversely, let IY. E R() n R. It will suffice to

prove that IY. E R()' (because IY. E Rand IY. = IY.-). Write
From the hypothesis that IY. has integral coefficients, we conclude that
L z(b) (bem -
b
!)2 == 0 (mod Z)
for all e prime to m, so that
- L z(b)b == -21 L z(b) (mod Z).

1
m b b
We contend that
L z(b)b == 0 (mod m) and L z(b) == 0 (mod 2).

This is obvious if m is odd. Suppose m even, so m is divisible by 4. Write
m = 4mo. Each b is odd, and
L z(b)b == 0 (mod 2mo)

so 2: z(b) is even. Then
L z(b)b == I L z(b)(mod mZ),

thus proving also the first congruence. Only the second will be used.
28
Now let s(G) = 2: U be the sum of the elements of G in the group ring,
and note that
e+8' = 1-s(G) and (1 + U-1)8' = s(G).
Then
IX = 2: Z(b)Ube-(J' = 2: Z(b)Ub(1 - e+)8'

= 2: Z(b)Ub8' - 2: Z(b)Ube+8'
= 2: Z(b)Ub8' - 2: z(bHs(G).
Substituting s(G) = (1 + U -1)8' on the right and using 2: z(b) even shows
that IX lies in R8', and concludes the proof.
It is of interest to determine the index arising from Lemma 2. This is done

in the next lemma. We let as usual:
w = number of roots of unity in Q(Jlm).
Lemma 3. (R8 : R8 n R) = w.
Proof We define a homomorphism
1
T: R8-+- ZjZ
w
by mapping an element of the group algebra on its first coefficient mod Z.

In other words, if
we let TIX = a(I). Note that
1 1
T(8) == In - 2 (mod Z),
and therefore that T is surjective. It now suffices to prove that its kernel is
R8 n R. But we have
whence for odd b prime to m, and IX E R, we get
T(UbIX8) == bT(1X8) (mod Z).

29
If Cf.() is in the kernel of T, it follows that Cf.() also lies in R, thereby proving the
lemma.
We now assume that m = pn is a prime power. Then
!/ = R() () R
is called the Stickelberger ideal We want to determine the index
Define
for any character Xon Z(m)*. Let X' be the primitive character associated with
X, and let m' be its conductor. Then it is easy to verify that if we replace m by
m' and X by X' in the right-hand side, we obtain the same value, so B 1 • x is
independent of whether we view X as primitive character, or simply a charac-
ter on Z(m)*. (The above fact is a special case of the distribution relation,
discussed in the next section.)
Next, we shall use the fact that
for odd characters X. For primitive X the non-vanishing of B 1 •X comes from

its relation with the L-series, and will be briefly recalled in Chapter 3. Cf.
also [L 3], Chapter 14, Corollary of Theorem 2.2.
Lemma 4. (R(): Rm()) = mM •

Proof This is obvious if one can show that R() is a free abelian group of
rank M. When m is a prime power, this results from the fact that for odd X
we have
We shall analyze (R- :!/) by the sequence of groups and subgroups

shown in the following diagram.
e- R 2M
R- __1- !/
::::> ::::>
mMTI -BI.' lu
'odd ul w
c
Rm() mM
R()
30
We have shown the inclusion relations, and we have also indicated the in-
dices. All of them have been proved, except the one on the left-hand side.
This will be the item in the final lemma, and we then find:
Theorem 1.1 (Iwasawa). Assume that m is a prime power. Then
Remark. Even though some inclusions go opposite to each other in the

diagram, to compute indices one still has multiplicativity, with opposite
inclusions occurring with opposite exponents. Cf. §4 if you don't find this
obvious.
Lemma 5. (e- R: e- Rme) = ±mM f1

xodd
B1,X"
Proof First observe that the sign is whatever is needed to make the right-
hand side positive. Multiplication bye-me is an endomorphism of QR-,
which is a semisimple algebra, decomposing into a product of I-dimensional
algebras corresponding to the odd characters. Consequently we find
det(e-me) = f1
xodd
X(me) = m M f1
;todd
B1,x'
On the other hand, e-me maps e- R into itself, and by standard elementary
linear algebra, the index is given by the absolute value of the determinant.
This proves the lemma, and the theorem.
Remark. In Chapter 3 we shall prove that the index computed in Theorem

1.1 is the order of the ( - 1)-eigenspace of the ideal class group in the cyclo-
tomic field, denoted by h -. The analytic class number formula will show that
the product of - B1,x yields the positive sign.
The theorem and its proof are due to Iwasawa [Iw 7]. It was generalized
to composite levels m by Sinnott [Silo In the composite case, one cannot deal
any more with a single element e, but one has to deal with the module
generated by Stickelberger elements of all intermediate levels
for all divisors d of m. A similar situation had already arisen in the analogous
situation in dimension one higher, concerning the Stickelberger elements
formed with B2 rather than, Bl> in the Kubert-Lang series [KL 2], [KL 3],
[KL 5].
31
§2. Bernoulli Numbers

We recall first some general notions concerning distributions, defined by
Mazur following the work of Iwasawa.
Let {X,,} be a sequence of finite sets, and suppose given a sequence of
surjective maps
so that we can consider the projective limit
For convenience, we took our family of sets indexed by the positive

integers. In applications, it often occurs that the sets are ordered by the
positive integers ordered by divisibility. For instance, the family of sets
Z/N", arises in the sequel. We shall also consider the projective family
with a fixed prime number p, and n = 0, 1, 2, .... In each case, the connecting
homomorphism
for MIN is reduction mod M, denoted by rM'

This type of projective family will also arise in isomorphic form as
follows. We have an isomorphism
1
NZ/Z~Z/NZ
given by multiplication with N. We then have a commutative diagram
1
N Z/Z~Z/NZ
NIMI I,M
I
MZ/Z~Z/MZ
where the left vertical arrow is multiplication with N/ M, and the right arrow
is reduction mod M. Thus the system
is also a projective system, ordered by divisibility.
32
Let us now return to the general projective system {Xn }. For each n suppose
given a function CPn of Xn into an abelian group V. We say that the family
{CPn} is compatible if for each n and x E Xn we have
CPn(X) = 2:
:7t n -1Y=X
CPn+l(Y)'
The sum is taken over all the elements of Xn+1lying above x. In what follows,
we often omit the subscripts, and write ny = x, for instance.
Let K be a ring of operators on V. Let f be a function on Xm for some
integer m, with values in K. If n ;::: m, then we view f as defined on Xn through
the natural projection on X m • We conclude at once from the compatibility
relation that
2: f(x)CPn(x) = 2:
xeX Il xeXm
f(x)CPm(x).
Let X be the projective limit
with the limit topology, so that X is a compact space. For each n we have a
surjective map
For each x E Xn the inverse image r;l(x) is an open set in X, and the totality
of such open sets for all n, x is a basis for the topology of X.
A functionf on X is called locally constant if and only if there exists n such
that f factors through X n • Such functions are also called step functions, and
their group is denoted by St(X, K). For each such function, we can define its
I
integral
f dcp = 2:
xeXn
f(x)CPn(X),
independent of the choice of n such that f factors through X n . We then call

the family {CPn}, or the functional dcp, a distribution on X. It is an additive map
dcp: St(X, K) -+ V.
Examples of such maps will be given later with Bernoulli numbers.

Let K be a complete field with respect to a non-Archimedean valuation,
and suppose that V is a non-Archimedean Banach space over K, i.e., V is a
complete vector space, with a norm
satisfying
Iv + wi ::; max{lvl, Iwl} v, WE V
Icvlv = IcIKlviv c E K, vE V.
33
If <p is bounded, i.e., I<Pn(x) I is bounded for all n, x E Xn> then we say that <P
is bounded, or quasi-integral for the valuation. For any IE St(X, K) we have
where IIIII is the sup norm off, and 11<p11 is the sup norm of the values I(Pn(x)l.
Indeed, if I factors through Xn. then
by the non-Archimedean property, so our assertion is clear.

In particular, if IE ceX) is a continuous function on X, then we can
approximate I uniformly by a sequence {fn} of step functions, and since
III --.., Inll ~ 0, we get
Illn - Imll ~ 0
for m, n ~ 00. Hence the integrals
f In d<p
converge, and define the integral
for such a continuous function, provided that <p is bounded. This will be the
case in important examples, and bounded distributions are also called
measures.
All this is preliminary to defining the distributions which are of importance
to us, namely the Bernoulli distributions. If x E ZeN) then x/N can be viewed
t
as an element of Q/Z. For any E R/Z we let <t>
be the smallest real number
~ 0 in the residue class of t mod Z. What we want is for each positive integer
k a polynomial PIc with rational coefficients, leading coefficient 1, such that
the functions
form a distribution on the projective system {Z/NZ}. Such polynomials will

be given by the Bernoulli polynomials. Let the Bernoulli numbers B" be
defined by the power series
tOOt"
B 1. F(t) = -t- =
e - 1 "=0
B""
k.
2:
34
Then for instance
Bo = 1,
Observe that
F( -t) - F(t) = t,
so that F is almost even, and in particular, we have
Bk = 0 if k is odd, k # 1.
We define the Bernoulli polynomials Bk(X) be the expansion
B2.
Then it is clear that the Bernoulli numbers are the constant terms of the
Bernoulli polynomials, that is
We find:
Bo(X) = 1,
The desired distribution relation is implied by the next formula.
B 3.
Proof On one hand, we have
N..;:;;/ te(x+a)t = l N~l Nte[(X+a)/N]NT
a~o e Nt - 1 N a~o e Nt - 1
On the other hand, summing the geometric series L eat directly from a = 0
to a = N - 1 and using the definition of the Bernoulli polynomials shows
that the coefficient of tkfk! is precisely Bk(X), thereby proving the desired
identity.
35
Relation B 3 can also be written in the form
for y E R/Z. This can be interpreted as follows.
On the projective system
{~Z/Z}
the association
1
x H>- Mk-1B k«x») for x E M Z/Z
defines a distribution.
Proof If y E (l/MN)Z/Z is one element such that Ny = x, then all
elements in the inverse image of x by the mapping (N· id) -1 consist of
y + ~, with t mod N.
Multiplying B 4 by Mk-1 yields precisely the distribution relation.

Since the system {(l/M)Z/Z} is isomorphic to the system {Z/MZ}, we can
also express the distribution relation on the latter. It is convenient to norm-
alize this distribution further and to give it a special symbol. For x E Z/NZ
we define
Then the family {Ef!")} forms a distribution on {Z/ NZ}.
Remark. Historically, this distribution arose in the context of the partial

zeta functions. Indeed, if x E (Z/ NZ)*, define
CN(X, s) = L n-
nEX
S•
n>O
The Dirichlet series converges only for Re(s) > 1, but it is classical and
elementary that it can be analytically continued to the whole complex plane,
and Hurwitz has shown that
36
Furthermore the partial zeta functions themselves satisfy the distribution

relation. For a further discussion, cf. Example 4 at the end of the chapter. For
distributions associated with zeta functions in connection with Cartan groups,
see [KL 10].
For the applications, we shall use one more formula concerning the
Bernoulli polynomials, namely
B 5. BiX) = Xk - tkXk-l + lower terms.

This is obvious by the direct multiplication of the series
For what we have in mind, we don't care about the lower terms, which have
rational coefficients.
Let N be a positive integer, and letfbe a function on Z/NZ. We form the

polynomial
N-l te(a + X)t
Flt, X) = a~o f(a) eNt _ 1 .
We define the generalized Bernoulli polynomials (relative to the function f)

by
B 6.
In particular, the constant term of Bk.tCX) is the generalized Bernoulli number
For instance, fmay be a Dirichlet character X on Z(N)*, extended to Z/NZ

by the value 0 on integers not prime to N. Then Bk,x is the generalized Ber-
noulli number of Leopoldt. Directly from the definition, we then find the
expression
B 7.
In terms of the distribution relation, this can be written
37
2. Stickel berger Ideals and Bernoulli Distributions
The distribution {E?)} is rational valued. We shall be interested in its

p-adic integrality properties for a prime p. For this purpose, we describe a
process which integralizes this distribution. For historical comments, see
below, after Theorem 2.1.
Let c be a rational number. For N prime to e (i.e., prime to the numerator
and denominator of e) we define
for x E ZeN). Multiplication by e or e- 1 is well defined on ZeN) so our

expression makes sense. If N is a power of a prime p, then we could also take
e to be a p-adic unit. We can write symbolically
This distribution satisfies the following properties.
E 1. Ei~~(x) = <~) _e<e~x) + e; 1.

-eBl(
Proof We have
Er.~(x) = Bl( <~») <e~x»)

= __
<~) ~ e«e~x) _~)
whence the assertion is clear.
E 2. E£~~(x) == Xk-1Ei~~(x) mod k%ck) Z[e, lie],
where D(k) is a least common multiple of the denominators of the coeffi-

cients of the polynomial Bk(X),
Proof We work with a representative integer x such that
O~x~N-1.
We write
with an integer b satisfying 0 ~ b ~ N - 1 and Y E Zpje]. Then
e-1x b
--=-+y= < b ) +z
-+y
N N N
38
with some integer z. Since Bk(X) = Xk - tkXk-l + lower terms, we find

the following congruences mod NI(D(k))Z[c, lie]:
== kxk-l(X
N - c <c-1X) c- 1)
-sr- + -2-
and Property E 2 follows by using E 1.
The values of Ek~~ are in

I
kD(k) Z[c, lie].
They will be called N-integral if they are p-integral for every prime dividing
N.
Theorem 2.1. (i) The values of Ek~~ are N-integral.

(ii) We have the congruence for every prime p dividing N:
(iii) If c is an integer prime to 2kN and to the denominators of the Bernoulli

polynomial Bk(X), then the values of Ek~~ lie in Z.
Proof For large integer v the values NVlkD(k) are N-integral. Let M = NV.
The distribution relation yields
Ek~XX) = L Ekz.;;(y)
y
where the sum is taken over, those y mod M which reduce to x mod N. The
expression for Eiz.;; is obviously N-integral except possibly for the term
39
(c - 1)/2. But if N is even then c is odd, so (c - 1)/2 is N-integral, and if N

is odd, then (c - 1)/2 is N-integral. If we apply E 2 to each term Ek~d(y)
then we see that the first two assertions are proved.
For case (iii), we take M = (NkD(k))" for large v. The argument then
proceeds as before, because the only denominators occurring in
or
contain only primes dividing NkD(k).
For k = 1 the integraIizing process already appears in the Stickelberger

theorem, and was used extensively by Iwasawa. For k > 1, Coates-Sinnott
obtained integral elements in group rings by this process [C-S 2], Theorem
1.3 and [C-S 3], Theorem 1. Mazur formulated this integralizing process
in terms of measure theory and the distribution relation, which allows the
jacking up argument used to prove Theorem 2.1.
For the rest of this section, we let N = pn with some fixed prime number p,
so the distributions are defined on the projective limit of Z(pn), which is
none other than the p-adic integers Zp. We view the values of the distributions
to be in C p , the completion of the algebraic closure of the p-adic numbers.
We may express Theorem 2. 1(ii) in the limit as follows.
Theorem 2.2. Let c be a p-adic unit. Then
We shall now express Bernoulli numbers in terms of the integralized

distributions.
Theorem 2.3. Let c E Z; and let k be an integer ;::: 1 such that c k =1= 1.
Then
Proof By definition,
On the last integral to the right, we make the change of variable
X I--?> ex,
40
which gives
The formula we want drops out by using Theorem 2.2.
Corollary 1 (Kummer Congruence). Let IX be a residue class mod p - 1 and

IX "# O. Then for even positive integers k == IX mod p - 1, the values (ljk)Bk
are all congruent mod p, and are p-integral.
Proof Select c to be a primitive root mod p so that
c k :;E 1 mod p.
Then 1 - ck is a unit at p. The values 1 - ck and Xk -1 mod p are independent

of the choice of k in the residue class mod p - 1, and the corollary then
follows from the expression of (1jk)Bk as the integral of the theorem.
Corollary 2 (Von Staudt Congruence). Let k == 0 mod p - 1, and k even.

Then
Proof Suppose p odd for simplicity. Let c = 1 + p. An easy induction

shows that
Hence
1 1
-1- k = --k (1
-c p
+ O(p» ,
and so
because the integral over pZp is == 0 mod p. An approximating sum mod p

for the integral over Z; is
p-1 X k - 1
X~l «x)
P - <c-
1
p X) +
c
1) c-
-2- .
Since c = 1 + P we have
41
The desired congruence follows from the fact that
2:
p-1
Xk == -1 modp.
x;l
We leave p = 2 as an exercise. We merely wanted to show how classical

congruences can be handled systematically from integration theory.
Let/be any function on Z/NZ. We defined
In terms of the distribution notation, this can be written
We shall apply this when / is a character of finite order on Z;, so that / is

an ordinary Dirichlet character on Z(pn)* for some positive integer n. As
usual, for such a character, we define its value to be 0 on elements of Z(pn)
which are not prime to p. Then by definition, for any character l/J of finite
order on Z; we have the formula for the Bernoulli-Leopoldt numbers
!n B n.", = Jrzpl/J dEn.
Note: When l/J = 1 we co not have (l/n)B n.", = (l/n)Bn because l/J is 0 on
pZp by definition.
Theorem 2.4. Let l/J be a character 0/finite order on Z;. Then
Proof We write dEn = dEn • c + cn dEnoc-l, or in other words
~ B n.", = f l/J dEn.c + f l/J(x)cn dEn(c-1x).
Integrals are taken over Z;. We let x t-+ cx in the second integral. Then
l/J(c) comes out as a factor. Using Theorem 2.2 concludes the proof.
Theorem 2.5. Let 2 ::; k ::; p - 2. Let w: Z(p)* --* Z; be the Teichmuller
character such that
w(a) == a (modp).
42
§3. Integral Stickelberger Ideals
For any integer n ;::: 1 we have
1 1
n-B",CD1<-" -= -k Bk (modp).
Proof Let l/I = ol'-". Choose c to be a primitive root modp, so that
ck ¢ 1 mod p. By Theorem 2.3 we get
By Theorem 2.4 we have the congruence mod p:
because I - l/I(c)c" and 1 - c k are p-units. Since the expression in brackets

under the integral sign is == 0 (modp), the theorem follows.
The next sections, §3 through §7, takenfrom Kubert-Lang [KL 8], deal further
with the integrality properties of Stickelberger ideals.

Let k be an integer ;::: 2. Let N = p" be a prime power with p ;::: 3 until §S.
We let:
G= Z(N)* if k is odd
G= Z(N)*/ ± 1 if k is even.
R = RG = Z[G] and Rp = Zp[G].
deg: R -+ Z is the augmentation homomorphism, such that
deg(L
aeG
mtt<!) = Lma •
This augmentation homomorphism extends to the complex group algebra
by linearity.
Rm = ideal of R consisting of those elements whose degree is == 0 mod m.
If I is an ideal of R, we let 1m = In Rm.
card G= IGI.
s(G) = L
aeG
(1.
For any, E R we have
,s(G) = (deg ')s(G).

If J is an ideal of R, we write d = deg J to mean that d is the smallest
integer ;::: 0 which generates'the Z-ideal of elements deg , with, in J.
43
Let B/c(X) be the kth Bernoulli polynomial. We let
()/c(N) = N/C-l Ia~ B/c( <~) )a;l
()~(N) = N"-l Ia~ (B/c( <~») B/c(o»)a;l -

where B" = Bk(O) is the kth Bernoulli number. We have:
deg () =1= 0 and deg ()' =1= 0, for k even.
In fact, these degrees can be computed easily. We need only that they are
=1=0 for k even, but the computation is as follows. Suppose k is odd. We use
the distribution relation. Summing over all primitive elements, i.e., elements
of pn yields the value of the distribution summed over all elements of level
pn-l. Continuing in this fashion reduces the computation to level 1. But
The degree of () arises from the same sum but with the term a = 0 omitted.
Hence
1 _ pk-l
deg () = k B"
and
deg ()' =
N"-l
deg () - -k- B"I GI
or
These formulas would also be valid for k even, except for our convention to
take G = Z(N)*/ ± 1. This requires dividing the formulas by 2 to get deg ()
and similarly for ()'. The non-vanishing for k even comes from the functional
equation of the zeta function.
Next we give the ideals used in integralizing the distribution.
44
J(k)(N) = ideal of elements ~ m(b)ub such that
L m(b)bk == 0 (mod N)
I(k)(N) = ideal of elements Ue - ck with integers c prime to N.
Since k and N remain fixed, we often write lJ and lJ' instead of lJk(N) and
lJ~(N). Similarly, we write J(k) and I(k), or J and 1. It is obvious that
We shall determine the extent to which J =F I in §2.

We have:
degI(k)(N) = pt, where t is the maximum integer such that k == 0 mod cp(pt).
This is obvious, because deg I(k)(N) is generated by the integers 1 - ck with

c prime to p.
Theorem 3.1. (i) We-have
RlJ~ n R = I(k)lJ~.
In jact, if an element, E R is such that ,lJ' E R, then, E I(k).

(ii) On the other hand, letting I~k) = ZP/(k), we have
If an element, E Rp is such that ,lJ E Rp then, E I~k).

Proof First we prove that for any prime ~ 2, we have
IlJ' c R, and IplJ c Rp.
A similar property is due to Mazur and Coates-Sinnott, as mentioned before.

Indeed, we have
U;l(UC - Ck)lJk = L Ek~~(a)u;;l

a.eG
where
The p-integrality then follows from Theorem 2.1(i). For other primes we
need a lemma.
45
Lemma 1. The polynomial (l/k)(Bk(X) - BiO)) maps Z into Z and maps

Zz into Zl for every prime I.
Proof A standard property of Bernoulli polynomials states that
Hence for any integer m we see recursively that the first assertion of the
lemma is true. The second, concerning l-adic integers, follows by continuity.
The lemma is also valid for p = 2.
We may define E~,c by using Bk(X) - Bk(O) instead of Bk(X) in the

definition of Ek,c' The lemma shows that /8' c R.
For convenience we let
Lemma 2. (i) Let, E R and suppose that eO' E Zp[G] = Rp. Then, EJ.
(ii) Let, E Rp and suppose that ,8 E Rp. Then, E Jp = ZpJ.

Proof Write, = L z(b)a b with integral coefficients z(b). Then
and therefore
Nk-l ~ z(b)B~ « b
-k- N» ) is p-integral.
But an elementary formula for Bernoulli polynomials, obtained directly

from the definition, gives for an integer b,
Nk-l
- - Bk (b)
- ~-
-_ L.. N k-
- 1 (k) B;-
(b)k-I
k N ;=0 kiN .
Comparing the leading term modulo all the lower order terms, and taking
into account that Bl = -! is p-integral (here we use p # 2), and the Kummer
theorem that B/ is p-integral for i < p - 1, we find
Lz(b)bk _ 1
kN = Omod'j(Zp,
Multiplying both sides by kN proves the lemma.
46
§3 Integral Stickelberger Ideals
Lemma 3. Let pS be the smallest power of p such that pS8~ is p-integral.

Then
s = n + ordp k.
We have J<k) () Z = (pS).
Proof The argument uses the same expression for the Bernoulli polynomial
as in the previous lemma. We see that
Nk-l (k) (1 )k-l

pS 2. -k- i Bl N is p-integral.
The leading term is pS/kN. The Bernoulli numbers Bi are p-integral for
i < p - 1 by Kummer, and for i ;:::: p - 1 the power N k - 1 in front in-
tegralizes (l/N)k-i. It follows that
pS.. I
kN IS p-mtegra ,
whence s has the stated value. Since we have already seen that 18' c R, it
follows that the p-contribution of I () Z is exactly pS. It is clear that I () Z
is equal to (pS), because we can always select
c == 1 mod Nand c == 0 mod 1
for any prime 1 -# p to see that I () Z contains elements prime to I. This

proves the lemma.
Lemma 4. We have J = I + ZN, and (J: I) = ps-n = pordk.

Proof It is clear that N E J. Conversely, write an element of J in the form
The first term is in I, and the second term is an integral multiple of N. This
proves the lemma.
We may now conclude the proof of the theorem. We prove (i). Suppose
~ E R and ~8' E R. By Lemma 2, ~ E J. By Lemma 4, we know that
~ == zNmod I for some z E Z.
We know that 18' c R. Hence zN8' E R. By Lemma 3, it follows that pS

divides zN, so ~ E I, and the theorem (i) is proved. The part (ii) is proved the
same way.
47
§4. General Comments on Indices

Let V be a finite dimensional vector space over the rationals, and let A, B
be lattices in V, that is free Z-modules of the same rank as the dimension
of V. Let C be a lattice containing both of them. We define the index
(A . B) = (C: B).
. (C:A)
It is an easy exercise to prove that this index is independent of the choice of

C, and satisfies the usual multiplicativity property
(A: D)(D: B) = (A: B).
Furthermore, if E is a lattice contained in both A and B then
(A . B) = (A : E) .
. (B: E)
We leave the proofs to the reader.

Suppose that A is not only a lattice, but is an algebra over Z. Let 0 be an
element of QA = V and let m be a positive integer such that mO E A. Assume
that 0 is invertible in QA. Then
(A : AO) = ± de~A 0,
where the determinant is taken for the linear transformation of QA equal to

multiplication by O. This is easily seen, because
(A : AO) = (A : AmO)(AmO : AO)
and
(AmO: AO) = (AO: AmO)-l.
Since mO lies in A, the index (A : AmO) is given by the absolute value of the
determinant of mO, which is mY det 0, where r is the rank of A. This power
mT then cancels the other index.
Note that the determinant can be computed in the extension of scalars by
the complex numbers. In particular, if A is a semisimple algebra, and is
commutative, then
det 0 = nX(O)
48
§5. The Index for k Even
where X ranges over all the characters of the algebra, counted with their
mUltiplicities. In the applications, the algebra is essentially a group ring, so
the multiplicities are 1, and the characters come from characters of the group.
This will be applied to the case when () = ()(k). We recall the definition of
generalized Bernoulli numbers according to Leopoldt:
Thus
1
X«() = k Bk,x'
Note that the Bernoulli number is defined with respect to G, so that for keven,
we are summing over Z(N)* / ± 1. This convention is the most useful for
present 'applications in §5 and §6. (We revert to the other convention in §7.)
For even k, it gives half the other values.
The classical theorem about the non-vanishing of Bk,x when k and X have
the same parity gives the desired invertibility of the Stickelberger element ()k
in the corresponding part of the group algebra over Q.
§5. The Index for k Even

We let s = n + ordp k, and t is defined as in §3, to be the maximum integer
such that k == 0 mod l/J(pt). We regard Ro n R() (for k even) as the Stickel-
berger ideal. We shall prove:
Theorem 5.1.
(R o·• R 0 n R() = Mpordk-t TI + k! B

- k,X'
X,ol
First observe that since deg () and deg ()' :f= 0 we have
Ro n R() = Ro n R()'.
By Theorem 2.1, we conclude that
R()' n R = J()', and hence R()' n Ro = Jo()'.
But Ro + JO' = Ra where
d =' deg J()' = (deg 1)(deg ()').
49
In §3 we had noted deg t = pt. The factor deg 0' will cancel ultimately. In
any case, we have:
(Ro : Ro () RO) = eRo : Ro () RO')

= (Ro : 10 0')
= (Rd: 10')
(R : Ie')
(R: R d )
= ~ (R : RO')(RO' : 10')
= ~ f1 X(O')(R : I).
The product is taken over all characters X of G. We separate this product
into a factor with the trivial character, giving deg 0', canceling that same
factor in d, and the product over the non-trivial characters. For X non-trivial,
we have X(O) = X(O').
In the final step we also wrote (RO' : 10') = (R : I). This is because e' is
invertible in the group algebra over Q. Hence the map ~ 1-+ ~O' induces an
isomorphism on R.
We are therefore reduced to proving a final lemma.
Lemma. (R: I) = pS where s = n + ordp k.

Proof We have (R : I) = (R : J)(J: I). Any element ~ in R can be written
in the form
From this it is clear that (R : J) = N, and the index (J: I) is obvious, thus
concluding the proof.
Remark. Of course we have not determined the sign occurring in the

product of the Bernoulli numbers. It is the sign which makes the product
come out positive, and which one determines easily from the functional
equation of the zeta function and the factorization in L-series. This is irrel-
evant for our purposes here.
§6. The Index for k Odd

Assume k is odd. Note that 0 = 0'. Let
a- = -t(I - 0"-1)
be the idempotent which projects on the (-I)-eigenspace. It is immediate
from the definition that 0 is odd, that is,
a-O = O.
50
§7. Twistings and Stickel berger Ideals
The Stickelberger ideal in this case is R() n R = I(), and is odd. We shall
prove:
Theorem 6.1.
(R- : R() n R) = Npordk TI ± 2~ Bk.,l.'
X odd
The rest of the section is devoted to the proof.
Lemma 1. We have R- = 2e- Rand (e- R : R-) = 2t/><N)/2.

Proof This is the same as Lemma 1 of §1.
We then proceed as in the even case. First we write
and then
(e- R : e- I() = (e- R : e- R()(e- R() : e-I()
= TI x«()(e- R : e- I)
xodd
because () is invertible in e-Q[G]. Furthermore,
(e- R : e- I) = (e- R : R-)(R- : 2e- I)(2e- I: e- I)

= (R-: 2e-I)
because (2e- I: e- I) = 2-t/><N)/2 since e- lis free of rank ¢(N)j2. Finally,
Lemma 2. (R- : 2e- I) = p' where s = n + ord p k.
Proof The group 2e- I is generated by elements of the form
e
An element E R- lies in Zeal - a -1) mod I. Hence the same argument as
in the past case gives the desired index.
§7. Twistings and Stickelberger Ideals

The Stickelberger elements ()k should really be indexed by the groups to
which they correspond. We now want to compare factor groups of the group
ring by various Stickelberger ideals, twisted in various ways. Consequently,
it is not useful any more to have G different in the even or odd case. For this
section, we let N = pn still, and we allow p = 2. We let
51
2.. Stickelberger Ideals and Bernoulli Distributions
We define
Ok.e(pn) = a;;l(a c - Ck)Ok(pn) E Z(pn)[GnJ.
This makes sense since we know from §l that Ok,e(pn) is p-integral.

Let V be a Z(Pn)[GnJ-module. We define its twist to be the tensor product
with the roots of unity,
V(l) = V0 /IN'
Then a in G operates diagonally,
a(v 0 y) = av 0 ay, and aa(v 0 "I) = a(aav 0 "I)'

We let "I be a basis for /IN over ZeN). Note that the element a on the right
makes sense as an element of ZeN) since V 0 JlN is a module over ZeN).
From the definitions we then get the formula
TW1.
resulting from Theorem 2.1 (iii),
The distribution relation allows us in E 2 to replace N by high powers of N

at a higher level, and then return to level N to get this congruence.
In particular, if Ok-I,e annihilates V, then Ok,e annihilates V(l). The argu-
ment simply extracts in a general context the argument given by Coates-
Sinnott [C-S 2J in connection with the ideal Class groups in cyclotomic fields,
see their Theorem 2.1.
Take V to be Z(pn)[GnJ itself, so that V(I) is generated by a single
element al 0 "I. The map
gives an isomorphism
Let ~(pn) = ideal of Z(pn)[GnJ generated by the elements Ok,cCpn). Then

the isomorphism induces a bijection
Hence we get an isomorphism
TW2.
where An = Z(pn)[Gnl is the group ring.
52
§8. Stickelberger Elements as Distributions
We may then pass to the projective limit. The limit of An is the Iwasawa
algebra. We let,?", be the ideal generated by the elements ()k,c (projective limit
of ()k,cCpn)). We obtain an isomorphism with the twist,
A/,?", -+ A(I)/'?"'_l(I).
This isomorphism permutes the eigenspaces for the action of flp-b and this
can be interpreted in terms of congruence relations between Bernoulli-
Leopoldt numbers (with characters) in the obvious manner.
We now make remarks concerning twistings, ideal classes, and modular
curves. We assume that the reader is acquainted with the latter. Suppose
N = p is prime '# 2, 3. The Iwasawa-Leopoldt conjecture predicts an iso-
morphism
where C:- is the p-primary part of the (-I)-eigenspace of the ideal class group
in Q(flp). On the other hand, Kubert-Lang [KL 7] establish an isomorphism
where 'i&'O(Xl(P)) is the cuspidal divisor class group on the modular curve
X1(p), generated by the cusps lying above the relational cusp on Xo(p),
Consequently, we expect a commutative diagram:
It remains a problem to give a direct isomorphism at the bottom, from some

sort of geometric construction. This may in fact lead to a proof of the
Iwasawa-Leopoldt conjecture.

In this section we follow Kubert-Lang [KL 5] to describe a "Stickelberger
distribution" associated with a distribution on Q/Z, and to give its basic
properties.
Let h be a function on Q/Z (with values in some abelian group, but for
the rest of this section, we shall take values in some algebraically closed field
F of characteristic 0). We say that h is an ordinary distribution if it satisfies the
relation
her) = 2: h(t)
Dt=r
for every element r E Q/Z, and positive integer D. The sum is taken over
those elements t such that Dt = r. In the application we have in mind, h
53
will be obtained from the first Bernoulli polynomial, and generalizations on

(Q/ZYk ) lead to the higher Bernoulli polynomials. See [KL 5] for k > 1.
We let G(N) c:: Z(N)*, writing the isomorphism as a f--+ (Ta. We let h be an
ordinary distribution as above. We define
hN(x) = h( <~») for x E ZeN).
For any functionf on G(N) we define (as usual)
SN(J, hN) = 2: f(a)hN(a),

a
with the sum taken over a E Z(N)*. If we define f on ZeN) to be 0 outside

G(N) then we see that
SN(J, hN) = Jf dh.
By abuse of notation, we often write a E G(N) instead of a E Z(N)*.

Let ZN = (l/N)Z/Z and let r E ZN. We define
If the values of h are in the field F, then the values of gN are in the group
algebra F[G(N)]. It is clear that if M is a denominator for r, i.e., r E ZM and
M divides N, then the image of gN(r) under the canonical homomorphism
G(N) --?>- G(M) is equal to gM(r). Thus we may define
g(r) = limgN(r)
in the injective limit of the group algebras (as vector spaces over F), ordered
by divisibility, with the injections from one level to a higher one given by
sending one group element to the sum of all the group elements lying above
it under the canonical homomorphism.
Theorem S.l. The function g: Q/Z --?>-lim F[G(N)] is an ordinary distri-

bution.
Proof Immediate from the definitions.
We define g to be the Stickelberger distribution associated with h.
Let AN be the vector space generated by the values g(r) with r E ZN

(essentially the same as the vector space generated by the values gN(r)). We
observe that g(O) is a constant multiple of the augmentation element, that is
h(O) "
g(O) = IG(N) I aeG(N)
L (T.
54
Let X be a character of G(N) and let m = m(x) be its conductor. We define
where Xm is the character on G(m) determined by X. We let
G,,(N) = set of characters X such that SCi, h) -=f O.
Theorem 8.2. The dimension of AN is equal to the cardinality of G,,(N).

Proof The space generated by the elements gN(r) with r E ZN is clearly a
G(N)-module since
We let the idempotent associated with X be the usual
If M is the conductor of X, then
as one sees at once from the fact that ra depends only on the residue class of
a mod M, for a E G(N). Hence AN has a non-trivial x-component if SCi, h)
-=f O. This shows that the dimension of AN it at least that which we asserted.
On the other hand, let r E ZN and suppose r has exact period M. Let X be
any character of G(N). Then
gN(r)ex = IG(~)I 2:
aeG(N)
h(ra)i(a)ex
= IG(~)I 2:
beG(M)
herb) 2:
redMa = b
i(a)ex'
If the conductor of X does not divide M, then X is non-trivial on the kernel

of the reduction map
red M : G(N) --+ G(M),
and the sum on the right is O. If the conductor of X divides M, then i(a) =
i(b) on the right, so
1 "" IG(N)I _
gN(r)ex = IG(N)I be~) herb) IG(M)I x(b)ex
= IG(~1)1 2:
beG(M)
h(rb)i(b)ex'
55
Since we can write r = a/ M with some a prime to M, a change of variables

in the sums shows that up to a non-zero constant factor, gN(r )ex is equal to
We now have to analyze this sum. The next lemma will show that this sum
is equal to some factor times SmUm, hm).
Lemma. Let X be a character of G(N) with conductor m.

(i) If every prime dividing N also divides m then
(ii) Let p be a prime dividing N but not dividing m. Write N = pnM with
pi M. Then
SN(XN, hN) = (1 - Xm(p))SMClM, hM)'
Proof The first statement is immediate from the distribution relation. Let
us prove (ii). We have
L
aeZ(N)'
x(a)hN(a) = L
beZ(M)'
x(b)
=
L
a b(M)
h(!!"').
N
aeZ(N)'
By the distribution relation, we know that
(b)
h-
M
= ~
L
xeZ(N)
(x)
h-
N
= ~
L
aeZ(N)'
(a) +
h-
N
~
L
a~Z(N)'
(a)
h-·
N
x=b(M) a=b(M) a=b(M)
The elements a in ZeN) which are not primitive but are == b mod M are in
bijection with the elements c E Z(N/p) satisfying the conditions
a = pc and c == p -la mod M,

under the map
C f-? pc
which sends Z(Njp) into pZ/NZ c ZjNZ. Therefore the sum over primitive
elements lying above a given b can be expressed as a difference
~
Lh(a)- - h(b-)- L~ h (c-
aeZ(N)' N - M c Njp
)
a=b(M)
(where the sum is taken over c E Z(N/p), c == p-1a mod M)
56
§9. Universal Distributions
(by the distribution relation). Plugging this into the first relation, and making
a change of variables b f-+ pb, we find
SNU, hN) = SMUM, hM) - L

lIeZCM)'
X(b)h (P-1b)
AT
= (1 - XM(P))SMUM, hM)'
This concludes the proof of the lemma.
In applying the lemma to the theorem, we note that the x-component is
at most one-dimensional, and has exactly dimension 1 under the stated
condition SNCiN, hN) =F O. This concludes the proof of the theorem.
A distribution can be decomposed as a direct sum of an odd and an even
distribution, provided that its image is contained in some module on which
multiplication by 2 is invertible.
In the next section, we shall prove that the rank of the values on ZN is
at most 'IZ~I, where Z: is the set of primitive elements in ZN'
If we take for h the distribution arising from the Bernoulli polynomial
h(r) = B1«r») if r =F 0, h(O) = 0
then the non-vanishing of B 1 •X for odd characters X shows that h has the
maximal attainable rank for an odd distribution. Consequently, we find:
Thoerem 8.3. The Stickelberger distribution g associated with h(r) as
above is the universal odd ordinary distribution into modules on which
multiplication by 2 is invertible.
So far, Theorem 8.3 has been proved only for distributions with values in
a field of characteristic zero. However, the next section will give a result of
Kubert showing that the universal distribution is generated on ZN by free
generators whose cardinality is IZ:I. This will take care of the additional
integrality possibilities allowed in the statement of Theorem 8.3.
Later in the book, we shall see that the cyclotomic units in the cyclotomic
field form an even distribution, which has maximal rank by the class number-
regulator formula, cf. Chapter 3, §3, and Chapter 6, §3.
The direct sum then yields a distribution of maximal attainable rank.
This is one method to show th~t the universal distribution in Theorem 9.1(ii)
has rank IZ:I.

In this section we give a theorem of Kubert [Ku I], [Ku 2], constructing a free
basis for the universal distribution on the projective system {(1jN)ZjZ}. In
[Ku 2] Kubert gives a complete treatment of the ordinary universal distribu-
tion on QkjZk for arbitrary k, as a GLk(Az)-module, where Az is the ring of
integral finite adeles. Here we limit ourselves to k = 1, and give only the
abelian group structure. .
57
For simplicity of notation we let

1 1
ZN = N ZjZ, and eN = Nmod Z.
We let
g: QjZ --+ some abelian group
be an ordinary distribution, in other words we suppose that for r E QjZ,

and a positive integer D we have
2: g(t) =
Dt=r
g(r).
It is clear that such distributions form a category, and we wish to construct

the universal distribution.
We let ztbe the set of primitive elements in ZN, i.e., elements having
period exactly N in ZN.
The prime power case

Let N = pn be a prime power and write N = MD, a factorization with
M > 1. Let r E Z~. If Dt = r then it is immediate that t E (N = prime zt
power is used here). The distribution relation shows that g(r) is an integral
linear combination of the images of the primitive elements g(t). Hence 0 and
these primitive elements generate the universal distribution, at level N.
We have
2: g(t) =
teZN
g(O) and 2:
tezN/v
g(t) = g(O).
Hence we get one relation among primitive elements,
2: g(t) =
teZ/;
o.
Let
Let
Theorem 9.1. (i) The elements g(TN) generate the abelian group generated
by g(ZN).
(ii) If g is the universal distribution then the elements g(t) with t E TN are
free generators.
(iii) The cardinality of TN is equal to that ofZj{.
Proof The first statement is obvious from the preceding remarks. The
cardinality of TN is clearly equal to that of zt.
For (ii), we may consider the
58
free abelian group generated by the elements of Z~ and {O}, modulo the
single linear relation
2:
teZN
(t) = O.
We can then define g on Z~ to be the canonical homomorphism in the factor

group, and for r E Z~ with M -# N and MIN we can define
g(r) = 2:
Dt=r
get), with D = N/M.
It is then clear that g defines a mapping on ZN satisfying the distribution

relation.
The proof of (ii) is in some sense natural, but in many ways it is better
to exhibit mappings which are distributions and which have the appropriate
rank to get the lower bound for the rank of the universal distribution. Cf.
the end of §8, where we exhibit natural distributions in the theory of cyclo-
tomic fields which have such rank.
The composite case

To state the theorem concerning the universal distribution in the com-
posite case, we shall write elements of ZN according to their partial fraction
decomposition. Let
N = [Jpfl.
j;,: 1
Then
1 1
- Z/Z =
N '-P -Z/Z
Q) pfl
and
-a =
N
2: pfla· mod Z
---2.
where ai is well defined mod pfl, while a is well defined mod N. We let:
TN = set of elements a/N as above, such that either ai is prime to Pi and
aj -# 1, or ai = O.
It is then clear that TN has cardinality ¢(N).
Theorem 9.2. The preceding theorem holds with this definition of TN ,for com-
posite N.
59
Proof The proof will be a simplification of Kubert's proof by Katz. Let

AN be the abelian group generated by g(ZN). A distribution having the lower
bound ,peN) for its rank has been exhibited in §8. Since TN has this cardinality,
it will suffice to prove that g(TN) generates AN. We first show that the elements
g(a/N) with a such that ai is prime to p, or al = 0, generate AN. We do this by
induction.
Let
be an arbitrary element of ZN. Write hl = p~al where al is 0 or prime to p.

If al = 0 then we are through by induction, so we can assume that al is prime
to p, and 1::;; r < nl. Then:
by the distribution relation. Since r < nl it follows that
where a~ is prime to p.
Inductively, we may now repeat the same argument with respect to P2,
Pa, . . .. It merely suffices to observe the following. In the first step of the
argument, when we factored out pi, thus changing hi to Ch if hj is prime to p
then CI is prime to p. Thus performing the same argument inductively on the
other primes does not destroy the desired property for those primes which
have already been taken care of. This concludes the first part of the proof.
Secondly, we show that we can recover those elements a/N for which aj
may be equal to 1 from the prescribed set TN. Let
1
yE N' Z/Z.
From the distribution relation, we find:
60
§1O. The Davenport-Hasse Distribution
Subtracting yields
2: g( '~l + Y) == 0 mod AN',

jmod P11 Pl /
(f,p) = 1
where AN' is the group generated by g(ZN')' This yields
-g(~
p~l
+ Y) == 2: g(a;l
a,,' 1Pl
+ Y) mod AN"
(al.Pl)=l
Observe that the same quantity y occurs on both sides of this relation. We
may now repeat the procedure inductively on the partial fraction decomposi-
tion of y. If we write
and say a2 = 1, we get a similar congruence
-g(~
p~l
+ Yl) ==
where N" = Nlp~2. In this way we reduce the proof to the case when N con-
tains fewer prime factors, and then can apply induction with respect to the
number of prime factors to conclude the proof.
§10. The Davenport-Hasse Distribution

In this section we give a relation of Davenport-Hasse [D-H]. Let Fq be the
field with q = pn elements, and let q == 1 mod m. We follow the notation of
Chapter 1, §l. We let p be a prime in Q(f..lq-l) lying above p, and let <.p be a
prime in Q(f..lq-l> f..lp) lying above p. We write as usual
a == f3 mod* <.p
to mean that rxfJ-l == 1 mod msp, where msp is the maximal ideal in the local
ring at <;p. We use similar notation mod* p or mod* p. We let X, l/! be characters
on F:, and put
rex) = - sex, 2).

Theorem 10.1. (Davenport-Hasse) We have
where e(l/!, m) = l/!(m-m) T1 rex).

xm=l
61
Proof Let um("') be the quotient of the left-hand side by the right-hand side,
that is
We have to show umC"') = 1. First note that um("') lies in Q(J.-!Q-1)' This is
immediate by looking at the action of 0'1,v, cf. GS5 of Chapter 1, §l. From
the fact that IS("')I = YC; if'" # 1 and IS("')I = 1 if'" = 1, we conclude that
Ium("') I = 1. Similarly, all conjugates of um("') have absolute value 1. Since
S(", )S(Iii) = ± q, we know that only primes dividing p occur in the factoriza-
tion of S(",). We shall prove that
(1) Um( "') == 1 mod '-P.

This will imply that um("') is a unit, and therefore a root of unity. If p # 2,
this congruence (1) implies that um( "') = 1. If p = 2, we shall give the argu-
ment at the end of the proof.
To prove the congruence, we simplify the expression in Stickelberger's
theorem. For any integer k we had defined s(k) = sik) and y(k) = yik) in
Chapter 1, §2. We let r(k) = rik) be the unique integer such that
o :::; r(k) < q - 1 and k == r(k) mod q - 1.

Lemma 1. Let 0 :::; k < q - 1. Then
k-s(k)
k! == (-p)p::::T y(k) mod* p.
Proof By induction. Suppose first that p t k. Then ko :::: 1, and
s(k) = s(k - 1) + 1, y(k) = y(k - l)ko.
The assertion is then obvious from the inductive step for k - 1. Next sup-
pose p Ik, so k = pk'. Since
ordp k! = [~J
p + ... + [~J
pn-1
and similarly for k', we see that
ordp k! - ordp k'! = k'.
In k! = (k'p)!, the factors not divisible by p give a contribution of
(p - I)! == -1 modp,
taken k' times. The product of the factors divisible by p yields k'! pt, where
t = ord p k'!. The lemma is then immediate.
62
§1O. The Davenport-Hasse Distribution
As in Chapter 1, we let 8 = e21tiip. We let n = 8 - 1. Then from

(8 - 1 + I)" - 1 = 0
we see at· once that
nP - 1 == -p mod* n.
From Stickelberger's theorem and Lemma 1, we conclude that
(2)
This reduces the proof of the congruence relation (1) to the proof of such a
congruence for the expressions on the right-hand side of (2), corresponding
to the way um(ljJ) is made up from expression r(w- k ) for appropriate values
of k. We shall prove two relations for the residue function, namely:
(3) L rex + y) = r(my) + L rex)
n n
mx~O mx~O
(4) rex + y)! == r(my)! m-r(m y ) rex)!

mx~O mx~O
In these relations, sums and products are taken over elements x mod q - 1
such that mx == 0 mod q - 1. The theorem is immediate from these relations,
taking into account
applied to y such that 1jJ.= w- Y •

We prove the two relation& (3) and (4). To begin with, we note that the
left-hand side and right-hand side of each relation is unchanged when we
change y in a residue class mod (q - I)Jm. Consequently we may assume that
q - 1
0::::; y < - - .
m
We choose the obvious representatives
q- 1
x = v - - with v = 0, 1, ... , m - 1.
m
Then
rex) = x, rex + y) = x + y, r(my) = my.
This makes (3) obvious, and (4) takes the form:
mmy n(Y+V q - 1)!

.m = n (v~)! 1
mod* p.
(5)
(my)! m
63
°: : ;
The products are taken for v = 0, 1, ... ,m - 1 and y is taken as above, with
y < (q - l)jm. Let F(y) be the left-hand side of (5). Then the right-
hand side of (5) is equal to F(O), and consequently, it suffices to prove that
F(y) _ *
F(y _ 1) = 1 mod p,
with 1 ::::; Y < (q - l)jm, or equivalently
mm
n (y + v m 1) q -
I1 (my - v) ==lmod*p,
or also
mn-l my + v(q - 1) - 1
= mo
d*
p.
v=o my - v
For this it will suffice to prove that each factor in the product is == 1 mod p.
But the power of p entering in my - v is at most pn-l. Dividing numerator
and denominator of each factor by my - v shows that
m
_-"-y_-_v_+_v-,,q =- 1 mo d* p.
my - v
This proves the theorem except whenp = 2, when we know only that um(ljJ) =
± 1. In this case we argue further as in [D-H].
Let I be a prime dividing q - 1. Let
be the decomposition of IjJ into a product of a character of I-power order, and

a character of order prime to I. Then
is the corresponding decomposition for IjJm. Let IW be the highest power of I

dividing q - 1, and let 'zw be a primitive IWth root of unity. Let
Since IjJz == 1 mod A., it follows that IjJ == IjJz' mod A.. Therefore
,(IjJ) == (ljJz') mod A. and ,(ljJz) == 1 mod A..
In particular,
64
Appendix. Distributions
Since Um(t/J) = ± 1, it follows that um(t/J) = 1, thereby proving the theorem.
Remark. In [Ya], Yamamoto shows that the Gauss sums form the univer-
sal odd distribution modulo 2-torsion.
Appendix
In this chapter we have looked at the distributions which are especially
relevant to the cyclotomic theory discussed in the rest of the book. It is
worthwhile to give here a number of examples of distributions occurring
throughout mathematics, involving various classical objects. We make a list
of a general nature, including those we have already met.
(1) The Bernoulli distribution, which is essentially given by a polynomial.
(2) The Fourier-Bernoulli distribution, giving rise to the Bernoulli distri-

bution, as follows. For real (J we have the Fourier expansion
Thus we may even define Bk on RjZ, and through this Fourier series, the
function given at level N by
satisfies the distribution relation.
(3) The holomorphic Bernoulli distribution. Let
and restrict z to the unit circle, Z = e2ni8 • Then {Nk-Yk} defines a distribution.
The real part for k even and imaginary part for k odd are mere homomorphic
images of this one, and give ri~e to the Bernoulli distribution of (2).
(4) The partial zeta functions. Let
,(s, u) = 60
00
(n
1
+ u)S
be the Hurwitz zeta function, for 0 < u :::; 1. For each real number t, let {t}
be the unique number congruent to t mod Z, and such that
o < {t} :::; 1.
65
Then for a E Z(M), the function
satisfies the distribution relation, namely
The sum on the left is taken for b in ZeN) reducing to a mod M.
(5) The gamma distribution. Define

1
G(z) = . /_ r(z).
v2n
We view G as defined on Q/Z with the origin deleted, but then with values in
the factor group
G: Q/Z - {O} --'? C*/Q:
of the multiplicative group of complex numbers, modulo the multiplicative

group of all algebraic numbers. The classical identity
fl.v /-
N-l
;=0
1
2n
r
(
z +.:Ji.) 1
=. /- r(Nz)N!-NZ
v 2n
shows that G defines a distribution.

RohrIich has conjectured that G is then the universal odd distribution,
with values in groups where mUltiplication by 2 is invertible. This is a con-
jecture in the theory of transcendental numbers. It also leads to the question
(in algebraic independence) whether the distribution relations, the oddness
relations and the functional equations generate an ideal of definition over the
algebraic numbers for all algebraic relations among the values of the gamma
function (l/V2n)r, with rational arguments.
(6) The cyclotomic units, which we have discussed.
(7) The modular units, which may be defined by their q-expansions,

namely
fl (1 -
00
g(a) = _q%B2<alle21lia2<al-ll/2(1 - qz) qnqz)(l - qn/qz)

n =1
66
Appendix. Distributions
and where the value g(a) is to be taken in the multiplicative group of the
modular function field modulo roots of unity, cf. [KL 3], following the work
or Ramachandra and Robert. The association
a H>- g(a)
is the universal even ordinary distribution on Q2jZ2 - {O}. The ordinary

Bernoulli distribution (with k = 2) then appears as a homomorphic image of
this one.
In the last three examples, the distribution is not defined at O. In such cases,
it is useful terminology to refer to the distribution as punctured.
Roughly speaking, I expect that in any classical situation where a distribu-
tion arises naturally, it is universal (odd, even, punctured, as the case may be),
always subject to taking values in groups where 2 is invertible.
(8) The Lobatchevski distribution. I am indebted to Milnor for the

following brief comments which might inspire the reader. Define the
Lobatchevski function
o
A(O) = - iologl2 sin t I dt.
This is essentially the same as the integral
o
- fo logle 2ntt - 11 dt.
Since the function t H>- Ie 2"it - 11 satisfies the distribution relation, one sees
at once that A(O) satisfies the distribution relation in the sense that on
{(1jN)ZjZ} the family {NA(O)} is a distribution, which is odd.
Let H be hyperbolic 3-space. This is the set of points
(Xl' X 2 , Y) E R x R x R+
so (Xl> x 2) is an ordinary point in the plane, and y > o. We endow H with

the metric
dx~ + dx~ + dy2

y2
Select four distinct points in the plane, and let T be the tetrahedron in H
whose vertices are at these points. Then it can be shown that opposite dihedral
67
2; Stickelberger Ideals and Bernoulli Distributions
angles are equal. (The dihedral angles are the angles between the faces of the
tetrahedron.) Let a, p, y be the dihedral angles. Then
a + p+ y = 11:,
and the volume ofthe tetrahedron is precisely given in terms of the Lobatchev-
ski function by
fff dXl ~:2 dy = Vol T = 2(a) + 2{f3) + 2(y).
The search for relations among such volumes had led Milnor to consider the
Lobatchevski function and its relations, now known as distribution relations,
and to show that it had the maximum rank (its values being viewed as con-
tained in a vector space over the rationals). Of course, Kubert's construction
in fact gives free generators over Z.
Finally, let 0 = Z[C], where C is a primitive cube root of unity. Then
where Aut H is the group of automorphisms for the Riemannian structure,

orientation preserving. The tetrahedron is essentially a fundamental domain
for PSL2 (o). This point of view leads into the problem of determining all
relations for volumes offundamental domains in the higher dimensional case.
68
Complex Analytic Class Number F onnulas 3
The complex analytic class number formulas date back to the 19th century.
They relate class numbers of cyclotomic fields and units. They arise by
factoring the zeta function of a cyclotomic field in L-series, and looking at the
factorization of the residue.
§1. Gauss Sums on Z/mZ

We have to redo the properties developed in Chapter 1, for the ring with
divisors of zero Z(m) = Z/mZ. The only additional feature arises from the
presence of non-zero elements which are not units. We let m = f1 p"(P) be
the prime power product. We then have product decompositions
Z(m) = I1 Z(p,,(p» and Z(m)* = I1 Z(p"(P»*.

From the product, for any character X on Z(m)* and any character A. on
Z(m) we have a decomposition
If x E Z(m) and x is not prime to m, we define X(x) = o. We let, be a

primitive mth root of unity (chosen to be e21tI / m over the complex numbers),
and
Observe that Z(m) is self dual under the pairing
69
3. Complex Analytic Class Number Formulas
Let dim. We have a natural surjective homomorphism
Z(m)-+Z(d)
and also a surjective homomorphism
Z(m)* -+ Z(d)*.
If there does not exist dim and d =1= m such that X factors through Z(d)*, then
we call X primitive. Again to determine the smallest d such that a given
character factors through Z(d)*, we may look at prime powers.
Suppose m = p" is a prime power, and X is a character on Z(P"). Let p'
be the smallest power of p such that X is trivial on
For convenience, let us abbreviate
so 1 + p'vA is a group for any positive integer v. The following criterion is

immediate.
X is primitive if and only if r = n.
The power p' = pT(P) is called the conductor of X.

In the composite case, we let the conductor be defined by the product
c(x) = cond(x) = ITp'(P).

plm
It is then clear that c(x) is the smallest d such that X factors through Z(d)*.
We define
Sex) = Sex, A) = 2: X(X)A(X),
:;c
and the sum could be taken only over those x E Z(m)*. It is then obvious that
we have a decomposition
Sex, A) = IT Sp(xp, Ap)

Jl
where the sum Sp is taken over Z(P"(P»*.

If d is an integer prime to m, then, as with Gauss sums over finite fields,
we have
Sex, Aod) = X(d)S(x, A),
by making the change of variables x f4 d-1x.
70
§1. Gauss Sums on ZjmZ
On the other hand, if d is not prime to m, we have one new significant

feature.
Theorem 1.1. If X is primitive and d is not prime to m, then
Sex, ;"od) = 0.
Proof Using the prime power decomposition, we may assume without loss
of generality that m = pn is a prime power. Abbreviate
Also without loss of generality, we may assume d = pT for some integer

r ~ I, and r < n. Form a coset decomposition
Then
Sex, )..opT) = L L
i xepn-TA
X(Ut)x(1 + x»)..(prUi)
= L X(ui»)..(prUt) L x(1 + x).

i x
Since X is assumed primitive, it is non-trivial on 1 + pn-TA, and the sum on

the right is 0, thus proving the theorem.
From here on we have the same formalism as for Gauss sums over finite
fields. For any functionf on Z(m) we define its Fourier transform
Tf(y) = L
XEZ(m)
f(x»)..(-xy).
Theorem 1.2. (i) We have T2f = mj-.

(ii) If X is primitive, then
TX = x(-I)S(x)[l.
(iii) Again if X is primitive, then
S(x)S(x) = m.
prime to m, then TX(Y) = °

Proof Part (i) is proved as for the finite field case. For (ii), if y is not
by Theorem 1.1. If y is prime to m then we can
make the usual change of variables to get the right answer. Part (iii) is then
proved as in the finite field case.
71
§2. Primitive L-series

Let X be a character mod m. We consider the DirichletL-series for Re(s) > 1:
L(s, X) = ~ -x(n)
L., 8 = "
L., " s'
x(a) L., 1
n= 1 n aeZ(m)' nEa n
Let ( again be a primitive mth root of unity. Then we have
~ L (a - n)x = {O if n :; a mod m
m xeZ(m) 1 if n ~ a mod m.
Indeed, if a ~ n (mod m), then the character x H- (a-n)x is non-trivial on

Z(m). Consequently we can write the L-series in the form
L(s, X) = L
aeZ(m)
x(a) -
m
1
LL 00
n=l x
(a-n)x
1
s
n
whence also
1
L L
00 (-nx
L(s, X) = - Sex, ..lox) -8•
m xeZ(m) n=l n
Theorem 2.1. Assume that X is a primitive character mod m. Then

1
L(s, X) = - Sex)
m beZ(m)'
L
i(b)
00 (-no
-8•
n=l n
L
Proof If x is not prime to m then the Gauss sum is 0 by Theorem 1.1.
If b is prime to m, we can make the change of variables which yields the
desired expression.
So far we have worked with Re(s) > 1. We now want to have the value
of the L-series at s = 1. It is not difficult to prove that the L-series has an
analytic continuation for Re(s) > O. Of course, it is also known (and a little
more involved) how to prove the analytic continuation to the whole complex
plane. For our purposes, to get the value at 1, we can work ad hoc, let s be
real> 1, and take the limit as s approaches 1. Then we don't need anything
else here.
We recall a lemma about series.
Lemma. Let {an} be a decreasing sequence of positive numbers, whose limit

is 0 as n -+ 00. Let {b n} be a sequence of complex numbers, and assume that
there is a number C > 0 such that for all n,
72
§2. Primitive L-series
i.e., the partial sums of the series {bn} are bounded. Then the series L: anbn
converges, and in fact
The proof is immediate using summation by parts.
We apply the lemma to the series with b n = ,-nb and an = Iin s with s
real > O. The partial sums of the bn are clearly bounded (they are periodic).
Let
For Izl < 1 we have
L: :...
co n
-log(I - z) =
n=1 n
As z ~ zo, -log(I - z) approaches -log(1 - zo). On the other hand, let
z = rzo with 0 < r :::; 1.
Then the series L: znln converges to L: z~/n as z tends to Zo along the ray
(that is, r tends to 1). This is again obvious by estimating the tail end of the
series using the lemma. Consequently, we find:
Theorem 2.2. If X is a primitive character, then
L(I, X) = - Sex)
m
L:
beZ(m)'
X(b) 10g(I - ,-b).
The picture of the roots of unity looks like in the figure.
If
73
then the picture shows that
The branch of the logarithm is determined so that
-~ <
2
e <~.
2
Observe that we do not change the sum
if we replace b with -b. We shall distinguish two cases.

We say that X is even if X( -1) = 1, and that X is odd if X( -1) = -1.
We assume m > 2, and m = m(x) is the conductor of X.
Case 1. X is even.
In this case, adding the sum with band -b yields
22x(b)10g(I - (-b) = 2X(b)[10g(1- (b) + 10g(I _ (-b)].
With X even, we obtain the formula
L(l,X) = _Sex)
m
2
beZ(m)'
x(b)logil _ (bl.
Case 2. X is odd.
In this case, we let
(= e21ti/m and b = l, ... , m - 1.
Then
log(l - (-b) = 10gl1 - (-bl + i(~ - ~)

log(l - (b) = logil - (bl - i(~ - ~).
Thus with X odd, we obtain the formula
L(I
,X
) = n:iS(x)
,m
m~l
b~ X
-(b) (~ _
m
!)2 = niS(x) B
m
_.
l,X
74
§3. Decomposition of L-series
Remark. Let m be an integer > 1 and let X be a non-trivial character on

Z(m)*. Then either the conductor of X is odd, or it is even, in which case it is
divisible by 4. Hence for a primitive character, we cannot have m = 2.
This is in line with a field theoretic property. Consider the field
Let m be the smallest positive integer for which we can write K in this fashion.
Then either m is odd or m is divisible by 4. If m is odd, then the group of
roots of unity 11K in K consists of ±11m. If m is even, then 11K = 11m.

For the applications we have in mind, we have to deal with two types of
fields: The cyclotomic field Q(l1m) for some integer m > 2, and its maximal
real subfield, over which it is of degree 2. We shall use a language which
applies to the more general situation of an arbitrary abelian extension of the
rationals (known to be contained in a cyclotomic field), but the reader may
limit his attention to the two cases mentioned above. Certain proofs can be
given ad hoc in these cases, while it is easiest to use general class field theory
to deal with the general situation. I hope that the extent to which I recall
certain proofs here will make the material readable to any reader not
acquainted with class field theory.
Let K therefore be an abelian extension of Q, and let K+ be its real
subfield. We let m be the smallest positive integer such that K C Q(l1m) (we
call m the conductor of K). We assume K i= Q, and as said above, you may
assume K = Q(f1m) or K = Q(l1m) + • We have a surjective homomorphism
Z(m)* --+ Gal(K/Q) = GK1Q
Any character X of GK1Q gives rise to a character on Z(m)*, also denoted by X.

We let m(x) be its conductor. We may view X as factored through Z(m(x))*,
in which case we speak of X as the corresponding primitive character. If we
need to make a distinction between X as character on Z(m)* or the corre-
sponding primitive character on Z(m(x))*, then we denote this primitive
character by Xo. The context should always make clear which is meant.
Let
be the zeta function associated with K. It is a fact that there is a decom-

position
(K(S) = f1 L(s, X),

')(
75
where the product is taken over all the primitive characters induced by
the characters of G K1Q • We reproduce the proof in the case K = QVtm). In the
last section we dealt with the L-series in its additive form. Here we use the
multiplicative form
L(s, X) = n (1 _X~»)-1
where the product is taken over all primes p not dividing mex). All these series
and products converge absolutely for Re(s) > 1, and what is to be proved
amounts to formal identities, localized at each prime p. Specifically, the
decomposition is equivalent to proving for each prime number p:
f1 (1 - .l.) = n (1 - X(~»).
Pip N'fJ x P
It is therefore convenient to let t = p-s. As usual, let
be the decomposition of p in prime ideals in K. Then
efr = [K: Q].
The identity to be shown is then equivalent to
(1 - tryT = n (1 -
x
X(P)t).
Suppose first that pt m. Then e = 1. The prime p generates a cyclic subgroup

of order fin Z(m)*,
Z(m)* ~ {p} ~ {I}.
The value of a character X viewed as character on Z(m)* or as primitive

character are the same on p. There are f distinct characters on the cyclic
group {p}, corresponding to the fth roots of unity, each such character
assigning one of these roots of unity to p. Each one of these characters then
extends in r possible ways to Z(m)*. Since trivially we have the factorization
1 - tf = n (1 -
s'=1
,t),
we have proved our identity in the case ptm. The argument is, by the way,
entirely similar if K"# Q(/lm).
Suppose secondly that plm. Write m = pkm ' with (p, m') = 1. If plm(x)
76
then by definition X(p) = o. If ptm(X) then X factors through Z(m')*. We are

therefore reduced to proving the identity
(1 - tty = n (l - X(P)t)
where the product is taken over those X whose conductor m(x) is not divisible
by p, and hence such that X factors through Z(m')*. The arguments are then
identical with the preceding arguments, replacing m by m'. This concludes
the proof. (Cf. [L 1], Chapter XII, §1.)
As usual, we let rio r 2 be the number of real and complex conjugate

embeddings of K.
If K is real then rl = = o.
[K: Q], r2
If K is not real, then r1 = 0 and r2 = ·HK: Q].
We let·
N = [K: Q] = N K •
We assume known the analytic continuation of the zeta function and
L-series at 1 (cf. [L 1], Chapter VII, [L 2], Chapter XIV). By comparing
residues, we have the class number formula:
CNF.
As usual:
W = WK = number of roots of unity in K.

h = hK = class number of K.
R = RK = regulator of K.
d = dK = absolute value of the discriminant.
If K is real, so r z = 0, then W = 2 and the formula reads:
~1f- =
vd
n
X;<l
1;L(I, X)·
Leopoldt's p-adic analogue will be given in the next chapter. If K is not real,
then we let h +, R + denote the class number and regulator of its real subfield
and N+ is the degree of the real subfield,
N+ = NI2 = r 2 •
We shall also need another fact whose proof is somewhat more delicate.
77
Theorem 3.1. We have product expressions;
(i) Ilmex) = d
x .. 1
(ii)
D Sex) =
{
d lf2
iT2d 1l2
if K is real
if K is not real.
Proof It is possible to give essentially algebraic proofs for these facts

(although the sign of the Gauss sums is always a little delicate, involving
something about the complex numbers). The best way to see the theorem,
however, is probably as in Hasse [Ha 1], using the functional equations of the
zeta function and L-series. Indeed, under the change s f-+ I - s, the functions
if K is real
dSI2(n-SI2r(s/2»NI2(n-SI2rC ; s) YI 2(K(S) if Kimaginary
are invariant. On the other hand, under the transformation
S f-+ 1 - s and X f-+ i,
the following functions (for non-trivial X)
mex)SI2(n-SI2r(s/2»L(s, X) if X is even
take on the factor
v' X( -1)mex)
Sex)
Dividing the functional equation of the zeta function by the functional
equation of the L-series, one sees that under s f-+ 1 - s,
( mdex»)SI2 takes on the factor v'X( -I )m(x)

SeX)
The theorem then follows at once.
If we combine the residue formula, Theorem 3.1, and the expressions for
the values L(I, X) for primitive characters X found in the last section, we then
get the following factorizations for the product hR in the two cases.
K real.
2N - 1 hR = TI L
x"l bmodm(x)
-X(b) logll - (~(x)l·
78
Warning: In this case, N = N+, h = h +, R = R + and characters are even.
K imaginary.
w TI L
X even b mod m(x)
-X(b)logll - '~<x>I· TI
X odd
-BI,x·
X*1
In the real case, we observe that all characters are even. Also the number of
roots of unity in K when K is real is equal to 2. Otherwise, the formulas are
just obtained by plugging in.
It will be convenient to reformulate them slightly, to make the connection
between imaginary K and the maximal real subfield clearer. We let:
E = EK = group of units in K
E+ = EK + = group of units in K+
11K = group of roots of unity in K
CK = group of ideal classes in K.
Lemma. We have the index
Proof This is obvious by computing the regulator of the units in K+ with

respect to K, where local factors of 2 occur in each row of the determinant
expressing the regulator, whereas a local factor of 1 occurs in the corre-
sponding determinant giving the regulator of the units in K+ .
Following Hasse, we give a symbol for the index in the lemma, calling it
the unit index:
Reading the class number formula in the real case applied to K+, we find:
Theorem 3.2. For imaginary K,
h = h+ QW2- NI2 TI
xodd
-BI,x·
In the next section, we shall analyze more closely the decomposition
79
where h- is defined as h/h+, and we shall see that h- is an integer. In any

case, we have the class number formula:
CNF-.
In the next section, we shall prove that Q = I if K = Q(fJ.m) and m is a

prime power. In addition, h- will be interpreted as the order of the (-1)-
eigenspace of the ideal class group. From Theorem 1. I of Chapter 2, we find:
Theorem 3.3. If m is a prime power, K = Q(fJ.m), G = Gal(K/Q), and.? is

the Stickelberger ideal, then
Let p be a prime number. If A is an abelian group, we denote by A(p) its

p-primary part. As Iwasawa observed [Iw 7], knowing the index immediately
shows that:
The group Ci(P) is generated by one element over Z[G] if and only if there
is a Z[G]-isomorphism
Indeed, we know that the Stickelberger ideal annihilates the ideal classes,
so the isomorphism is obvious if there exists one generating element by
Theorem 1.1 of Chapter 2.
Let m = p itself. Iwasawa [Iw 7] and Leopoldt [Le 5], [Le 10] have shown
that ifthe Vandiver conjecture h+ prime to p is true, then the cyclicity follows
for the p-primary part of C-. (See Chapter 6, §4.) Proving the Vandiver
conjecture, or the Iwasawa-Leopoldt conjecture that Ci(P) is cyclic over the
group ring is therefore one of the major problems of algebraic number theory
today.
In the Iwasawa-Leopoldt conjecture it is necessary in general to restrict
the conjecture to the p-primary component. For example, let Fbe an imaginary
quadratic field Q( v' - p ), and suppose p is such that F is contained in the
cyclotomic field Q(fJ.p) = K. Then Kover F is totally ramified above p, and
the Hilbert class field of F lifts to an unramified extension of K of the same
degree, so the ideal class group CF is a factor group of the ideal class group
CK , and CF = CF -. It is known that there exist such fields, e.g., Q(v' - 3299),
for which CF contains a group of type (3,3), see Scholz-Taus sky [S-T].
Furthermore, 3 does not divide p - 1, with p = 3299. Consequently all the
non-trivial eigenspaces for characters of Z(p)* of the local ring group Z3[G]
are cyclic over Z3. This shows that there cannot be an isomorphism
80
§4. The (± l)-eigenspaces
and thus in general, the Iwasawa-Leopoldt conjecture has to be restricted to

the p-primary component.
For h+ we also get a formula, and it is convenient to introduce the group
G = Z(m)*/ ± 1,
and for each even character X the group
Gx = Z(mCl))*/ ± 1.
Then there are exactly N/2 even characters and (N/2) - 1 non-trivial even
characters. Therefore we obtain the other class number formula:
CNF+.
The product over X i= 1 is taken over the non-trivial characters of Gx , or

equivalently the non-trivial even characters of Z(m)*. This product will be
interpreted as a determinant of certain units in §5, and it will follow that h+
is equal to the index of a certain subgroup of the units in the group of all
units.
§4. The (± 1)-eigenspaces

In this section we analyze in greater detail the factors h + and h - of the
class number, and the corresponding unit index. We assume that m is odd
or m == o mod 4.
Theorem 4.1. Let K = Q(l1m). Then QK = 1 if m is a prime power, and 2

if m is not a prime power.
Proof Let E = EK be the unit group in K. For each unit u in E, the
quotient u/u is a unit, of absolute value 1, and for any automorphism a of
Kover Q, we have
a(u/u) = au/au
because a commutes with complex conjugation (abelian Galois group).

Hence all conjugates of u/u have absolute value 1. Hence u/u is a root of
unity. Let
({J: E -7 11 = 11K
be the homomorphism ({J(u) = u/u. Then
81
because if u is a root of unity, then q>(u) = u- 2 so the image of q> contains

the squares. Hence the index of q>(E) in J1 is I or 2 because J1 is cyclic. Further-
more, we see at once:
QK = 2 if and only if q> is surjective, i.e., q>(E) = J1.

Assume that m is composite. Let ( be a generator of J1 if m is even, and a
generator of the odd part of J1 if m is odd. Then I - ( is a unit (elementary
fact, and easy exercise), and q>(1 - 0 = -(-l, so q>(E) = J1, in other words
q> is surjective. On the other hand, q>(E+ J1) = J12, so the index is 2 in this case.
Suppose next that m is a prime power, m = pn. We contend that q>(E) i= J1.
It will follow that q>(E) = J12, and since the kernel of q> is E+ the theorem also
follows in this case. Suppose q>(E) = J1. Let ( be a primitive mth root of
unity, and let u be a unit such that
Let
1- (
a = --.
u
Then aj-a. = 1 so a = -a. and a is real. But 1 - ( is a prime element above p

in K and so a is also a prime element, which cannot lie in the real subfield.
This proves the theorem.
Theorem 4.2. Let K = Q(J1m). The natural map
of ideal classes in K+ into the ideal class group of K is injective.

Proof Let a be an ideal of K+ and suppose a = (a) with a in K. Then
-a.ja is a unit, and in fact a root of unity as one sees by an argument similar
to that in Theorem 4.1. Suppose that m is composite. By Theorem 4.1, we
know that QK = 2 and q> is surjective, so there exists a unit u such that
uju = -a.ja.
Then au is real, and generates the same ideal as a, thus proving the theorem
in this case.
Suppose that m = pn is a prime power. Let ( be a primitive mth root of
unity, and let A = 1 - " so A is a prime element above pin K. We can write
-a. = az with some root of unity, and Ail = - ( is a generator of J1. Hence
for some positive integer k. Then aA k is real. Since the ideal generated by IX
comes from K+, and since p is totally ramified in K, it follows that k is even.
82
§4. The (± l)-eigenspaces
Hence z is a square in /1, and therefore in the image of cp, say z = u/u for
some unit u. Then au is real, and generates the same ideal as a, thus proving
that a is principal, and also proving the theorem.
Theorems 4.1 and 4.2 are classical, see for instance Hasse [Ha I], Chapter 3,
for more general results. The elegant proofs given here are due to Iwasawa.
Theorem 4.3. Let K be an imaginary abelian extension ofQ. Then the norm
map
on the ideal class group is surjective.

Proof We have to use class field theory, which gives the more general
statement:
Lemma. Let K be an abelian extension of a number field F. Let H be the

Hilbert class field of F (maximal abelian unramified extension of F). If
K n H = F then the norm map N K1P : CK C p is surjective.
---';>-
Proof For any ideal class c in K, the properties of the Artin symbol show
that
(c, KH/K) restricted to H = (NK1PC, H/F).
We have natural isomorphisms of Galois groups:
Gal(KH/K) ~ Gal(H/F)
KnH=F
Hence the group (NK1PCK, H/F) is the whole Galois group Gal(H/F), whence
N K1PCK = Cp since the Artin symbol gives an isomorphism of the ideal class
group with the Galois group. This proves the lemma.
The theorem follows at once, because Kover K+ is ramified at the

archimedean primes, and hence cannot intersect the Hilbert class field of F
except in F.
Let T denote complex conjugation. Let
Cli = (-I)-eigenspace of CK
= {c E CK such that C1H = I}.
83
Theorem 4.4. Let K = Q(Pm). Then the sequence
is exact.
Proof We consider the norm map followed by the injection,
CK -norm
--+
CK+ ~
inj CK·
The kernel of this composite map is Ci by definition, so the theorem is

obvious by what had already been proved.
Corollary. The quotient h/h+ is an integer, which is the order of the group
Ci.
Remark. The integer h - is called the first factor, and h + is called the second
factor of the class number, in older literature. This is poor terminology since
the ordering seems arbitrary, and for several years this has been replaced by
the plus and minus terminology.
§5. Cyclotomic Units

Let m again be the conductor of the cyclotomic field Q(l1m), so either m is
odd> 1 or m is divisible by 4. Let ( be a primitive mth root of unity. For b
prime to m we let
Then gb is a unit called a cyclotomic unit. It is easy to see that gb is equal to a

real unit times a root of unity. Indeed, without loss of generality we may
assume that b is odd, since (b depends only on the residue class of b mod m.
Then
b- 1
(-Vgb for v =- 2-
is real (i.e., fixed under () -1), as one sees immediately from the definitions.
We let g: be this real unit, uniquely determined up to sign, and call it the
real cyclotomic unit.
We let cff be the group of units in Q(l1m) generated by the roots of unity
and the cyclotomic units. We let cff+ be the group of units in Q(l1m) + generated
by ± 1 and the real cyclotomic units. Then
Observe that gb and g -b differ by a root of unity.
84
As before, let N = [QCJ.lm): Q] and let
N
r=2"-1.
Then r is the rank of E, and also the rank of E+. If Ih, ••• , lOr is a basis for
E+ (mod roots of unity), then the regulator R+ is the absolute value of the
determinant
R(E) = R+ = ± det logluaejl

a.f
where j = 1, ... , r and a E Z(m)*/ ± 1 and a ~ ± 1 (mod m). It is convenient

to let
G = Z(m)*/± 1
so we may view a E G, a i= 1 in G.
On the other hand, we may form the cyclotomic regulator
again with a, bEG, and of course it does not matter if we write gb or g:

since the absolute value of a root of unity is 1.
For composite levels m the cyclotomic units are not necessarily inde-
pendent, and so we now turn to prime power level,
We shall prove in this case that the cyclotomic units are independent.
Interpreting the regulator as the volume of a fundamental domain for the
lattice generated by the log vectors of units in Rr, we see that
Remark. For composite m, as with the index of the Stickelberger ideal,

it is necessary to consider the group generated by cyclotomic units of all
intermediate levels to get a group of units of the right rank.
Theorem 5.1. Let K = Q(flm) and h = hK . Assume m = pn is a prime power.

Then
85
~3. Complex Analytic Class Number Formulas
Proof Let G be any finite abelian group. Then we have the Frobenius
determinant formula for any function f on G:
nL
X,H aeG
x(a)f(a- 1 ) = det [f(ab- 1 )
a.b¢ 1
- f(a)].
The proof will be recalled later for the convenience of the reader. It is already
clear that up to minor changes, this formula yields the theorem, taking into
account the expression for h+ obtained at the end of §3. We now make these
changes explicit.
Lemma 1. We have for G = Z(m)*j ± 1:
± det logluagbl
a.b;"l
= n L X(b) 10gl1 -
X;"l beG
Cbl
= n L X(b) loglgbl·
X¢l beG
Proof The first expression comes from the Frobenius determinant formula
(Theorem 6.2), and the second comes from the fact that for non-trivial X,
L XCb) 10gl1 - CI = o.
Lemma 2. Let Gx = Z(m(x»*j ± 1. For prime power m = pn, we have
L X(b)logll - C~(x)1 = LX(b)logll - C~I·

N~ NG
Proof Let mCl) = pS. We write residue classes in Z(pn)* in the form
y = b + pSc, with 0 ::s; c < pn-s,
and b ranges over a fixed set of representatives for residue classes of Z(pS)*.
Instead of the sums over Gx and G respectively, it is easier now to work with
sums over Z(pS)* and Z(pn)* respectively, and then divide by 2. The desired
relation is then immediate from the identity
n
Am~l
(X - ,iY) = xm - ym,
because we get
L
ymodpn
x(y) 10gl1 - C~nl = L
bmodp-
X(b)logll - Ct-I·
This proves the lemma.
Theorem 5.1 is then immediate from the lemmas, and the class number
formula for h+ obtained from the L-series.
86
It is generally believed that the coincidence of group orders in Theorem 5.1

does not correspond to an isomorphism of the groups involved. Iwasawa has
a counterexample at least that C+ is not isomorphic to Ejtf as Galois module.
Mazur has pointed out that the analogous statement for the case of elliptic
curves with complex multiplication is definitely false.
We conclude this section by mentioning the most classical case of the
quadratic subfield. For our purposes we are interested in the case of the real
quadratic subfield. Thus for the end of this section, we let
m = p withp prime =? 2, 3
and such that K = Q(flm) contains a real subfield F = Q(VD) with D > 0,
so D = p, and D is the discriminant. Let e > 1 be a fundamental unit of F,
and hF the class number. From
where X has order 2, we get
2hF log e = L(1

VD ,x)
S( )
= _--..X.
D
L i(a) logll -
D-l
a=l
,al.
It is a simple matter of the theory of quadratic fields that the conductor m(x)
is exactly D (assumed> 0). The explicit value S(x) can be determined in any
number of ways (via functional equation, via Dirichlet's method as in my
Algebraic Number Theory, Chapter IV, §3, etc.), and we have S(x) = VD.
Thus we find:
Theorem 5.2. For a real quadratic field F = Q(VD) as above,
2hF log e = - L
amodD
i(a) logl! - ,al·
We have the tower of fields:

K
] ' } Z(D)'/± I
I
Q
87
Let C be the group of cyclotomic units in K+ and let CF be ± its norm group
into F, so
Then CF (mod ± 1) is infinite cyclic.

Theorem 5.3.
Proof Let
(£ = I1 (l - ,a) and (£' = I1 (1 - ,a),

~~-l ~~--1
where' is a fixed primitive Dth root of unity. Note that the character X is
even, so a and - a occur simultaneously in each product. Therefore the norm
from K+ to F of any real cyclotomic unit
is a unit in F, and the group generated by these norms (mod ± 1) is infinite

cyclic, generated by a unit 1'/ > 0 such that
From Theorem 5.2 we conclude that
hF log e = log 1'/.
Thus 1'/ = e\ and since 1'/ (mod ± 1) generates the norms of cyclotomic units
in K+ , this proves the index relation of Theorem 5.3.
This index relation is analogous to that of Theorem 5.1 for the full cyclo-
tomic field. Since K+ is totally ramified over F (at the prime p) it follows from
class field theory that
hF divides ht.
(Proof: Let HF be the Hilbert class field of F. Then HF n K+ = F, so
Since HFK+ /K+ is unramified, it follows by class field theory that hF divides
M.)
We shall see later that the Vandiver conjecture asserts that ht is prime to
p. It would then follow that hF is prime to p also. For tables of some hF' see
Borevich-Shafarevich, 'Number Theory, Academic Press, p. 424. It has been
88
§6. The Dedekind Determinant
observed for a long time that hF has very small values, and grows very
slowly. It is unknown if there are infinitely many real quadratic fields of class
number 1.
§6. The Dedekind Determinant

Let G be a finite abelian group and G= {X} its character group. We have the
Dedekind determinant relation:
Theorem 6.1. Let f be any (complex valued) function on G. Then
n 2: x(a)f(a- 1) = detf(a-1b).
XEG aEG a.b
Proof Let Fbe the space offunctions on G. It is a finite dimensional vector
space whose dimension is the order of G. It has two natural bases. First, the
characters {X}, and second the functions {c5 b }, bEG, where
c5 b(x) = 1 if x = b
c5 b(x) = 0 if x i= b.
For each a E G let Tafbe the function such that Taf(x) = f(ax). Then
(TaX)(b) = x(ab) = x(a)x(b),
so that
TaX = x(a)x·
So X is an eigenvector of Ta. Let
T = 2: f(a-1)Ta.
aEG
Then T is a linear map on F, and for each character X, we have
TX = [2:
aEG
x(a)f(a- 1)] x·
Therefore X is an eigenvector of T, and consequently the determinant of Tis

equal to the product over all X occurring on the left-hand side of the equality
in Theorem 6.1.
On the other hand, we look at the effect of T on the other basis. We have
so that Tac5b is the characteristic function of a-lb, and
89
Consequently
TO b = L f(a- 1)oa- 1b
aeG
= L f(a-1b)oa.
aeG
From this we find an expression for the determinant of T which is precisely

the right-hand side in Theorem 4.1. This proves our theorem.
Theorem 6.2. The determinant in Theorem 4.1 splits into
Therefore
L x(a)f(a- 1) =
f1 aeG
l;<l
det [[(ab- 1 )
a.b;<l
- f(a)].
Proof. Let a1 = 1, ... , an be the elements of G. In the determinant
add the last n - 1 rows to the first. Then all elements of the new first row are
equal to 2. f(a- 1 ) = 2. f(a). Factoring this out yields
1 1 1
[~f(a)] f(a2~;1)
f(ana;l) f(a na2 1) .. -I(ana;;l)
Recall that al is chosen to be 1. Subtract the first column from each one of the
other columns. You get the first statement.
On the other hand, the function f can be selected so that the elements
{f(a)}, a E G, are algebraically independent over Q, and therefore the factori-
zation given in this first statement for the determinant is applicable in the
polynomial ring generated over Z by the variables f(a). Combining the first
statement with Theorem 6.1 yields the second relation where the product is
taken only over X =1= 1.
Serre has pointed out to me that the determinant relation is due to Dede-
kind, February 1896, who communicated it to Frobenius in March. Cf.
90
§7. Bounds for Class Numbers
T. Hawkins, "New light on Frobenius ... ", Archive for History of Exact
Sciences 12 (1974), p. 223.

In this book we have not emphasized questions having to do with the size
of the class number. We shall here make some brief remarks concerning
various possibilities to obtain bounds. We let hm = class number of Q(llm),
and p is prime ;;:: 3.
To begin we derive the expression ofthe class number h; as a determinant
following Carlitz-Olson [Ca-O]. We start with the expression
h; = 2p TI - !B
X odd
1 •X
= 2p TI -!
x odd
2:
aeZ(p)' .
x(a)<~),
P
because the characters are non-trivial, and the term with! drops out. We try
to rewrite this as a Dedekind determinant over the group
G = Z(p)*/ ± 1.
We have
Let ro be the Teichmuller character such that ro(a) == a mod p. We write odd
characters as products
x = roljJ
-<
where ljJ is even. Then we find
h; = 2p 1) 4Ia if/(a)ro(a) ( <~) ~a»),
_<
and this makes sense because the function
f(a) = ro(a) ( <~) ~a»)

on Z(p)* is actually well defined mod ± I, so is defined on G. This expression
is now in the form where we can apply the Dedekind determinant, thus getting
91
The size of the determinant is

p-l
-2-·
Let ( be a primitive (p - I)-root of unity. Representatives for elements of G

are given by the powers (i with 1 ::::; i ::::; (p - 1)/2. The determinant may then
be taken over indices
.. 1 P - 1
l,J = ""'-2-
and w(ab) = (1+1. In the expansion of the determinant, every term contains a
factor arising from these (i+1, whose product is obviously 1. Consequently
the determinant is the same as the determinant obtained by omitting these
(1+ i from each term.
Let R(a) be the smallest positive integer in the residue class of a mod p.
Then R(a) is an integer::::; p - 1, and
<~) = R;a).
We use the notation R«(I+') and R( _(i+l) to denote similarly the smallest
positive integers in the residue class of (I+i and _(I+i respectively. Then we
have proved the following theorem.
Theorem 7.1.
where
Observe that each entry in the determinant Dp is an integer of absolute

value::::; p - 1.
The absolute value of the determinant is the volume of the fundamental
domain of its row vectors, say. This volume is bounded by the product of the
Euclidean lengths of these vectors (Hadamard inequality). Carlitz [Cal
observed that this gives the bound
As Carlitz-Olson relate it, the history of the determinant in Theorem 7.1

is amusing. The determinant
p - 1
det R(ab- 1 ), a, b = 1, ... , -2-
was known classically as the Maillet determinant, conjectured to be =1= 0 by

Maillet. Malo computed it for p ::::; 13, and found it equal to the appropriate
92
power of p. He conjectured that it was always so equal, but Carlitz-Olson

computed a bit further, and found extra factors. They derived that the
Maillet determinant is equal to the determinant of Theorem 7.1 (up to the
obvious power of 2), and then also to the class number times that power of
p by using the expression of the class number as a product of generalized
Bernoulli numbers (not called that at the time). Thus Malo had missed out
the class number factor.
In this book we have entirely left out questions having to do with the size
of the class number (as distinguished from certain congruence properties).
We refer the reader to recent papers for these, notably the following which
contain more extensive bibliographies than we give here. We let h m = class
number of Q(J1m), and p is prime.
On the one hand, Carlitz and Carlitz-Olson [Cal, [Ca-O] write h; in
the form ± D/p(P-3)/2 where D is a certain determinant of dimension (p - 1)/2
whose entries are integers between 1 and p - 1. Carlitz pointed out that
Hadamard's inequality immediately gives
Masley-Montgomery [M-M] also prove the inequalities
for primes p > 200. Thus the Carlitz bound is reasonably sharp. For applica-
tions of this see Ribet [Ri].
For primes p, it has been proved by Uchida [Uch] that h p = 1 if and only
if p ::; 19. More generally, Masley and Montgomery [M-M] subsequently
proved that h m = 1 for precisely 29 distinct values of m (always assumed
;;E 2 mod 4), the largest of which is m = 84. Masley [Mas 2] shows that hm = 2
if and only if m = 39, 56. For Euclidean cyclotomic fields, see also [Mas 1].
93
4
The p-adic L-function
In this chapter we return to p-adic integration theory, and give Mazur's

formulation of the p-adic L-function as Mellin transform. It turns out to be
more convenient as a basic definition, than Iwasawa's previous formulation
in terms of power series. The connection is made via Example 2 of §1. We
derive further analytic properties, whlch allow us to make explicit its value
at s = 1, thereby obtaining Leopoldt's formula in the p-adic case, analogous
to that of the complex case. We also give Leopoldt's version of the p-adic
class number formula and regulator.
The basic arguments are due to Leopoldt [Le 11]. However, we shall
follow in §l and §2 a course of Katz, which developed systematically opera-
tions on measures and their corresponding formulation on power series in the
Iwasawa algebra. In this manner, constructions which appear slightly tricky
in Leopoldt's paper here become completely natural, and even forced from
these measure theoretic operations.
The Leopoldt transform then appears as an extension of an integral
transform to a somewhat wider class of power series than those with p-adic
integral coefficients. No use will be made of this, since only integral valued
measures occur in the analysis of the p-adic L-function, but we include
Leopoldt's results for completeness, for convenience of reference if the need
ever arises for them.
The p-adic L-function in the case of elliptic curves is discussed in Robert
[Ro], and especially Coates-Wiles [C-W 2], [C-W 3]. See also Lichtenbaum
[Li 3], and Katz [Ka] for general comments concerning its connection with
formal groups. For the case of totally real fields, Shintani's evaluation of the
zeta function [Sh] presumably allows a development of the L-function similar
to that of the cyclotomic case.
This chapter is used only in Chapter 7, and it can therefore be omitted
94
§1. Measures and Power Series
without loss of the logical connections. On the other hand, if one leaves out
the section on the p-adic regulator, then the chapter appears as a natural
continuation of Chapter 2, and is essentially measure theoretic, independent
of Chapter 3.
Throughout, we need the fact that if 0 is the ring of integers in a p-adic
field, then there is a natural isomorphism
lim o[XJj«(1 + X)P" - 1) ~ o[X]].

+-
The limit is the projective limit, and is called the Iwasawa algebra. This is a
basic fact of algebra. In the next chapter, we need further facts about this
algebra and modules over it. For the convenience of the reader, all these
facts and their proofs will be placed in the next chapter.

Let C p be the completion of the algebraic closure of Qp, and let 0 = 0cp
be the ring of p-integers in Cpo By a measure J.1 we shall mean an o-valued
distribution on Zp. This means that for each integer n :2: 0 we have a function
such that the family {J.1n} is a distribution on the projective system Z(pn).
Let Cont(Zp, 0) or C(Zp, 0) be the space of continuous functions on Zp
into 0, with sup norm. As usual, there is a bijection between measures and
bounded functionals
A: Cont(Zp, 0) --7 O.
[A Zp-linear map A is called bounded if there exists C > 0 such that
IA(cp) I :0:; Cilcpli for all cP E Cont(Zp, 0).
The inf of such C is called the norm of A, and denoted by IIAII. The bounded
functionals form a p-adic space. J Indeed, it is clear that any measureJ.1 gives
rise to a functional
On the other hand, suppose A. is a bounded functional. If x E Z(pn), let CPx

be the characteristic function of the set of elements y E Zp such that
y == xmodpn.
Define
J.1,,(x) = A(CPx)'
95
4. The p-adic L-function
It is then clear that {Iln} defines a measure. Since any continuous function
on Zp can be uniformly approximated by step functions, it follows easily
that the correspondence
is a bijection from o-valued measures on Zp to bounded functionals.

Furthermore, define the norm
111111 = sup Illn(x)l,

n,'"
taken for x E Z(pn) and all n. Then the map Ill--? dll is easily verified to be
norm preserving.
The Iwasawa algebra is obtained as the projective limit
Ao = lim o[X]/«l + X)pft - 1) ~ o[[Xll,
and
o[X]/«1 + X)pn - 1) = o[T]/(Tpft - 1)
where T = 1 + X. Let Yn = T mod (pn - I), so y~n = 1. Let as usual
( r) = r(r - 1) ... (r - k + 1).

k k!
The function Iln on Z(pn) can be viewed as an element of the group algebra
o[Yn], namely
p~: Iln(r)y~ = p~: Iln(r) :~ol C)Xk

= p~: C~l Iln(r) G) )Xk
where the right-hand side is read mod (1 + X)pn - 1. Thus
pn_l pn_l
2:
r=O
Jln(r)y~ = 2:
k=O
Cn,k Xk = Pn(X),
where the coefficients Cn,k are given by
The canonical homomorphism Z(pn+l) ~ Z(pn) maps
96
and we let
P(X) = limPn(X)
be the projective limit of these elements in the Iwasawa algebra. We call

P(X) the power series associated to J1, and also denote it by (PJ1)(X) or
PJ1(X). Thus
P: o-valued measures on Zp _ o[[X]]
is an o-linear map. Conversely, any power series IE o[[X]] defines a com-

patible system of elements in the group algebras o[Yn], so the map P is
bijective. We write
1= PJ1 or J1 = J1f
to mean that I is the power series associated to J1 as above. We call P the

Iwasawa isomorphism.
For any x E Zp let
Ck(x) =
x(x - 1)·· ·(x - k
k!
+ 1) (X ).
= k
Since Ck(r) is an integer for any positive integer r, and since Z + is dense in
Zp, it follows that
is a polynomial map of Zp int;) itself, and in particular is continuous.

For fixed n, define
C <n)( ) = r(r - 1) .. ·(r - k + 1)

k x k!
where 0 ::::: r ::::: pn - 1, and r == x mod pn. Then Cfcn) is a step function,
defined at level n, and
lim Cfcn) = Ck uniformly.

n-+CO
Since the coefficients Cn,k in the polynomial Pn(X) are given by the sum of
products of J1n and the binomial coefficient, we obtain:
Theorem 1.1. Let I(X) = L: CkXk E o[[X]]. Then
97
Theorem 1.2. The power series PJ1 is the unique power series f such that for
z in the maximal ideal of 0, we have
1
Zp
(1 + z)X dJ1(x) = fez).
Proof We have
We can interchange the sum and integral, apply Theorem 1.1, and we see
thatPJ1 has the desired property. Uniqueness is obvious since any power series
is determined by its values.
Example 1. Let J1 be the Dirac measure at a point s E Zp, that is
1Zp
<p dJ1 = <pes).
Then the associated power series f is
f(X) = ~ (S)Xk = (1 + X)s.

k=O k
Example 2. Let v be a measure on Zp whose support lies in the open

closed subset 1 + pZp. Let y be a topological generator of 1 + pZp, for
instance y = 1 + p. There is an isomorphism
such that
By pull back, there exists a unique measure J1 = J1f on Zp such that
fl+PZp
US dv(u) = 1 Zp
yS:': dJ1f(X).
By Theorem 1.2, writing yS = 1 + z, we get
,f l+pZp
US dv(u) = f(yS - 1).
98
The power seriesfis not easily determined in terms of v. Iwasawa expressed

his results on p-adic L-functions in terms of the power series f Mazur gave
the formulation in terms of the integral, see §3 below.
if there exist elements an E 0 such that °

Theorem 1.3. (Mahler) A function cp from Zp into 0 is continuous if and only
lanl -+ and
cp(x) = I
n=o
an (X)
n
.
The sequence {an} is uniquely determined by cpo
Proof Given a sequence {an} as above, it is clear that the function
is continuous. For uniqueness, let
Acp(x) = cp(x + 1) - cp(x).
Then cp(o) = co, and furthermore
and
This proves uniqueness.

We now prove existence. In the applications, the measures will take values
in the ring of p-adic integers in a finite extension of Qp. An argument using
tensor products reduces the general case to this case, and we omit it since we
have no use for it. The case of a finite extension is then reduced to the case
when the measure is Zp-valued by taking a basis for the ring of values over
Zp and projecting on the coordinates. We now handle this case.
Let B be the Banach space of sequences (an) with an E Zp, and lanl -+ 0,
under the sup norm. We have a Zp-linear map
We have to show it is surjective. By completeness of C(Zp, Zp) it suffices to

prove that a given f E C(Zp, Zp) is congruent to the image of an element
99
of B mod pn for each n, and by a simple recursion, it suffices to do this mod p.

In other words, it suffices to prove that the map
given by the same formula as above, is surjective. But
C(Zp, F p) = U Maps(Z(pN), Fp)

N
because F p is discrete and finite.
Lemma. Let 0 ::; k ::; pN. Then the function
of Zp into F p is periodic of period pN.

Proof We have to show
Since
we prove the lemma by comparing the coefficients of Tk.
Now we are reduced to showing that
is bijective. Since both spaces have Fp-dimension pN, the surjectivity follows
from injectivity, which is proved the same way we proved that the function
ep(x) has uniquely determined coefficients an. This proves Mahler's theorem.
Corollary. Iff(X) = .L cnxn and
then
so
Irep dll f I ::; (suplani)llfll ::; Ilfll·

100
§2. Operations on Measures and Power Series
We define the norms:
Ilfll = sup
n
lenl
11f111 = sup lf1n(X) 1as before.
n.:>:
Theorem 1.4. We have Ilfll = 11f1111.

Proof Since
we get trivially Ilfll :::; 11f1/11. Conversely, given a level pn, let Xo E Z(pn) and
let cp be the locally constant function such that
cp(Xo) = 1, and cp(x) = 0 if x of Xo, X E Z(pn).
Then
and on the other hand, from the corollary of Theorem 1.3, we get
If cp df11 I : :; Ilfll,
so 11f1/11 :::; Ilfll as desired.

We shall give a list of integration formulas, or better, a list of operations on
measures and their corresponding operations on power series.
Meas O. i
Zp
df1f = f(O).
Proof Special case of Theorem 1.2 with z = O.
For the next property, we let
t/I.,(x) = (1 + z)'"
if Z E m = maximal ideal of o. Also (with formal groups in mind) we write
x [+] Z = X + Z + zX = (1 + z)(l, + X) - 1.
Meas 1. t/lzf1f = /lg, where g(X) = f(X [+ ] z).
101
Proof For WE m we have
Jfzp l/Iw d(l/IzJl,) = Jfzp l/Iwl/lz dJl,

= L (1 + w)"(l + z)" dJl,(x)
zp
= L (I + w + z + wz)" dJl,(x).
zp
The property is then clear from the definitions.
In particular, let Cbe a pRth root of unity, and let z = C- 1. Then
and we find:
Meas2.
where g(X) = f(C(1 + X) - I) = f(X[+ HC - 1».
As before, putting T = 1 + X, andf(X) = fGm(T) iffis a rational func-

tion, we can write the power series g(X) in Meas 2 in the form
Moreover, let ({J be a step function, constant on cosets mod pR. Write the
Fourier expansion
({J{x) = 2:
,,""=1
rPmc"
We find:
Meas3. ({JJl, = Jlg
where g(X) = 2:
,P"=l
rPmf(C(1 + X) - 1).
Iff(X) = fGm(T) is a rational function, then
Let Up = U be the operator

Uf(X) := f(X) -! 2: f(C(1 + X) -
P ,P=l
1).
102
We call U the unitization operator because of the next property.
Meas 4. If ({J = characteristic function of Z~, then
({JJi, = JiUl·
Proof. We compute trivially the Fourier expansion of ({J:
Then Meas 3 gives
g(X) = 2: 4>(C)f(C(1 + X) - 1) = Uf(X),

'''=1
as was to be shown.
Remark. Let A be the formal multiplicative group (cf. Chapter 8). In the
notation of such groups, we can write the unitization operator in the form
Uf(X) = f(X) - ! 2: f(X [ + ] z).

P 2eA"
Meas 5, Let X be a character on Z:, of finite order with conductor N =
power of p. Let Cbe a primitive Nth root of unity, and let
sex, C) = 2:
aeZ(N)'
x(a}Ca.
Then
XJi, = Jig
where
g(X) = s<t C) 2:
aeZ(N)'
x(a)f(C-a(X + 1) - 1).
Iff is a rational function, then
Proof. It suffices to apply Meas 3 and to compute the Fourier transform

of X. This is trivial, and we have
if x == Omodp
if x == a ¢ 0 modp.
103
4.· The p-adic L-function
Meas 6.
where D = (1 + X)D x . In particular
Proof Note that
x = lim (1 + z)X - 1 = lim ljIzCx) - 1.

2->0 Z 2->0 Z
Hence for any step function cp we get
f xcp(x) dJ1f(X) = lim

2->0
f ljIzCx) - 1 cp(x) dJ1f(X)
Z
= lim
2->0
f cp(x) dJ1g.(X) (by Meas 1)
where
gz(X) = I(X + Z + :X) - ICX) = (1 + X)f'(X) mod z
by Taylor's formula. The desired result follows by taking the limit as z - ? 0

and using the non-trivial part of Theorem 1.4, that is:
Remark. We shall deal throughout with three variables. Let T be the

variable on the "multiplicative group." We put
x= T - 1, T=l+X.
Then Z is the corresponding variable on the additive group. For any power
series I(X) (with coefficients in a field of characteristic 0) there is a corre-
sponding power series denoted by /*(Z) or IGaCZ) such that
This last equality makes sense only when I is a rational function.

The differential operator D then can be expressed in terms of the three
variables,
(1 + X)Dx = Dz = TD T •
The expression in terms of T applies only to rational functions of T (rational
functions of X). The first two expressions in terms of X and Z apply to
104
§3. The Mellin Transform and p-adic L-function
arbitrary power series, and for any positive integer k, the expression D"f
makes sense whether we view f as power series in X or Z. Furthermore,
Iffis a rational function, this is also equal to (TD T )kfGm(1).
Meas 7. Let g = Ug so flg is a measure on Z:. Then
where h is any power series such that Dh = g.

Proof Since a-1flg(a) is a measure on Z:, there exists a power series f
such thatfE o[[X]],
Then
whence by Meas 6,
g = Df= DUf
We let h = Ufto conclude the proof.

Let (j) be the Teichmuller character. If p is odd, then
(j): Z: -+ flp-l
is the character such that (j)(a) == a mod p. If p = 2, then we define (j)(a) = ±1
such that
(j)(a) == a mod 4.
Then we can write uniquely an element a E Z: as
a = (j)(a)<a) ,
where <a) == 1 mod p if p is odd, and <a) == 1 mod 4 if p = 2.

Let fl be a measure. We define its Gamma transform as a function on Zp
i
by the integral
rpfl(S) = <a)S dfl(a),
z·p
and we define its Mellin transform, also as function on Zp, by
Mpfl(S) = iz~
<a)Sa- 1 dfl(a).
105
4.· The p-adic L-function
It is clear that rpJi and MpJi are continuous in s. (For analyticity, see below.)
Since the integral is taken on Z;, MpJi depends only on the restriction of
Ji to ZO;, so if Ji = Jir, then
If Ji = Ji" we write sometimes Mpf instead of MpJir, and similarly for the
Gamma transform.
Note that a-I dJi(a) for a E Z; is also the functional associated with a
measure, so that the Mellin transform is actually a special case of the Gamma
transform (of another measure).
Theorem 3.1. Let g E oUX]] be such that Ug = g, and let h be a power

series such that Dh = g. Then Uh E o[[X]] and
rpUh = MpJig.
Proof This is an immediate application of Meas 7, after integrating the

function (a)s.
We now consider the analyticity properties.
Lemma. Let Ji be a measure on Z;. Then there exists a power series

hE Zp[[s]],
2: bns n
00
h(s) =
n=O
such that bn -+ 0 as n -+ 00, with the property that for all s E Zp,
h(s) = r
Jz;
(a)S dJi.
Proof The integral can be written as a sum of integrals over cosets of

+ pZp (or 1 + 4Z 2 if P = 2). Changing the measure appropriately with
respect to each coset, we are reduced to proving (say for odd p) that for any
measure Ji, the integral
has the desired analyticity property. We note that
= f .
~
L. s(s - 1)·· ·(s - n
l+PZp n=O
+ 1)
(a - l)n
,
n.
dJi(a).
106
But a-I == 0 mod p, and so (a - l)njn! is p-integral for all n. Furthermore,

(a - l)njn! tends to 0 p-adically as n - 00. Hence we can interchange the
sum and integral to yield
where Pn is a polynomial of degree n with integral coefficients, and
Cn = f (
l+PZ"
a - l)n
,dJl(a)
n.
is p-integral, and Cn - O. It is then clear that 2: Pn(s)cn can be written as a

power series h(s) whose coefficients bn tend to 0 as desired.
We had the measure E1 •c in Chapter 2, with C E Z:. Let s be a p-adic

variable in Zp. For any c such that X(c)(c)S is not identically 1 we define the
p-adic L-function Lp by
-1
Lp(1 - s, X) = 1 _ X(C)(C)8 MP(XE1,c)(s)
= 1_ ~~)(C)8 fz. p
(a)Sx(a)a- 1 dE1,c(a).
By the lemma, the integral is analytic as a function of s. The factor in front

is analytic except when
X(C)(C)8 = 1.
If X is non-trivial, we can select C such that X(c) # 1, and then the factor in
front is also analytic at s = O.
Theorem 3.2. The value of Lp(l - s, X) is independent of the choice of c,

and for any positive integer k,
In particular, if k == 0 mod p - 1, and p is odd, then
Proof Since the set of sufficiently large integers k == 0 mod p - 1 is dense
107
in Zp, we see that the first assertion follows from the explicit values given at
integers of the form 1 - k as described. For these, we have:
1
= (1 - xw-k(C)C k) k Bk'l.W- k by Theorem 2.4 of Chapter 2
= (1 - X(C)(C)k) ~ Bk'l.W-k.
Theorem 3.3. Let g = Ug and let h be the power series such that
Dh = g and h(O) = O.
Then
Proof By Meas 7 we have
The formula is then clear from the definition of U.
To compute Lp(l, X) we have to work out the power series associated to

E 1 ,c and then apply the formalism of the preceding section systematically
to get the answer, with s = 0 in Lil - s, X), using Theorem 3.3.
Proposition 3.4. Let c E Z:. The power series associated with the measure
E 1 ,c is
1
1 + X.
C
f1,c = T - 1 - TC - l' with T =
Proof It is immediate to verify that as power series in X the expression

on the right-hand side is holomorphic at X = 0, and that its coefficients are
p-integral because c is a p-unit. Let
I'.(T) = log T _ clog T .

J! T-l T"-1
108
Putting T = eZ we find
/(T) = /*(Z) = eZ ~ 1 - ec:~ 1
On the other hand, letA,c be the power series associated with E 1 ,c, and write
Since
it follows from Meas 6 that
so
Z'l* c(Z) = '(1 - ck)Bk Zk = /*(Z) = 2.. - ~.

'./:. L k! T-I T"-l
It follows that
as desired.
Proposition 3.5. Let X be a non-trivial character on Z; with conductor N.

The power series associated with XE1,c is
where
GxCT) = SeX, n L x(a) 1

N aeZ(N)* ,aT - 1
Proof. Immediate from Meas 5.
We shall now assume that c is an integer> 1 prime to p.

Written in full, the power series for gx,c is
109
If we let
where:
-l ranges over cth roots of unity "1= 1,

a ranges over Z(N)*,
then it is easy to see (and we carry out the computation below) that
Furthermore,
The situation is then set up to apply Theorem 3.3.

We now prove that Dhx.c = gx. c• Observe that since 2: x(a) = 0, we have
G (T) = S(x, 0 "" x-(a) _Ca _ = SCl,O "" x-(a) _T_.

x N -f T - Ca N -f - T - Ca
Proof Taking the logarithmic derivative of
TC - cae = n (T -
.I.
-lca)
we obtain
Multiplying by sex, O/N proves the lemma.

The assertion
follows by using
and differentiating naively using D = TD T •
110
We shaII recaII below how it is possible to extend the definition of the

p-adic logarithm uniquely to a continuous function on aII of C; such that
log p = O. This is the log with which we deal in the next theorem, giving us
Leopoldt's value of the L-function Lp(s, X) at S = 1.
Theorem 3.6. Let X be a primitive Dirichlet character with conductor N equal

to a power of p. Then
Lil, X) = - sex,N 0 2:
aeZ(N)'
x(a) 10g(1 _ ,a).
Proof By Theorem 3.3, Proposition 3.5, and the definition of Lp(1, X), we
find:
But
Using the fact that N is the conductor of X and that the sum of a non-trivial
character over a group is 0, we leave to the reader the verification that
2: x(a) 10g(1 -
a
A,ap ) = o.
It follows that
_ sex, 0 ~ -( ) I
- ---y;.r- -f X a og 1 _
1- ,ca
,a
= set 0 exec) - 1) 2: x(a) 10g(1 _
a
,a),
as was to be shown.
Appendix. The p-adic Logarithm

We recall briefly how to extend the p-adic log to the multiplicative group
C;. The p-adic log is defined first by the usual series
x 2
logi1 + x) = x - 2" + ....
111
We shall omit the p as subscript. The series converges for Ixl < I (the
absolute value is that on C p , the completion of the algebraic closure of Qp).
We extend the log to all units of C:
as follows. The units have a product
decomposition
where j1[p] is the group of roots of unity in F of order prime to p, and U1 is

the group of units == 1 (mod p), and pip in C:. For each unit u we let (u)
be its projection on U1 , and we define
log u = log (u).
Thus the log has been extended to all units, and it is clear that this extension
is continuous, and is a homomorphism.
It· is even possible to extend the log to the whole multiplicative group
C: (following Iwasawa). We let P be a subgroup of C:
containing the
powers of p, and one tth root of p for each positive rational number t. Then
the multiplicative group of C:
has the product decomposition
Again we define the log of an element CI. E C:

to be the log of its projection
on U1 •
We leave it as an exercise to the reader to verify that this extension is
continuous. It is obviously a homomorphism. In particular, log p = O.
As for uniqueness, suppose the log has been extended to a continuous
function on C;, which is a homomorphism into the additive group. Then the
log has to vanish on all roots of unity. If logp = 0 then 10gpT = 0 for all
rational numbers r. Given a E C:
there exists r such that apT is a unit. Hence
the extension is determined by its values on units. Furthermore there exists
a root of unity, such that apT' is == I mod m, where m is the maximal ideal
of the integers of Cpo Hence the log is determined by its values on elements
== 1 mod m, where it is defined by the usual power series. This proves
uniqueness.
§4. The p-adic Regulator

Let K be a totally real number field, and let E = EK be the group of units
of K. Let p be a prime number. Let Ub ... , UT be a family of independent
units in K, and let
0"1: K -+ Cp , i = 1, ... , r + I
be the embeddings of k in the p-adic complex numbers (completion of the
112
§4. The p-adic Regulator
algebraic closure of Qp). We suppose lTT+1 = id. We define the p-adic regulator
up to sign,
Rp(Uh •.. , uT) = ±det log IT,Uj'
If Uh ... , U r are a basis for the units (mod roots of unity), we simply call it
the p-adic regulator, and write
If K is the real subfield of Q(f.1m), and Iff is the group generated by the real
cyclotomic units, then we let
where U1, ••. , U r generate these cyclotomic units, mod ± 1.

We leave it to the reader to verify:
Theorem 4.1. Let K be the real subfield of Q(f.1m). Then
We know from Theorem 5.1 of the preceding chapter that
h+ = (E: Iff).
Let ga (a prime to m) be the cyclotomic units, and g; the corresponding

real cyclotomic units. From our definition of the p-adic log, we know that
for any embedding IT: Q(f.1m) -7- Cp ,
Thus in writing down the regulator, we can use the usual form for the
cyclotomic units, without bothering to write down the extra root of unity.
We may write the p-adic cyclotomic regulator by the Frobenius determinant
formula,
Rp(lff) = det log lT~b

a.b;H
= n 2:
X"l aeG
i(a) log ga,
where G = Z(m)*J ± 1. Since
113
and since LaeG x(a) = 0, we may also write this formula in the form
Rp(cff) = nL
x*l aeG
x(a) log((a - 1).
The product is taken over all non-trivial characters of Z(m)* j ± 1.
Theorem 4.2 (Brumer). We have Rp =I

Q(Pm)+.
°
for the real cyclotomic field
Proof The cyclotomic units are algebraic, and it is a known theorem from
the theory of transcendental numbers that the logs (p-adic or otherwise) of
multiplicatively independent algebraic numbers are linearly independent over
the algebraic numbers. The proof is the p-adic analogue of Baker's proof for
the corresponding result over the complex numbers, see Brumer [Br], or
[L 4], before Chapter VIII, Introduction to the Baker method. The proof given
there applies p-adically. The factorization of the regulator into a product of
linear forms in logarithms then shows that the regulator is not 0.
Theorem 4.3 (Leopoldt p-adic Class Number-regulator Formula). Let

m = pn be a prime power, and K+ = Q(flm) + . Then
Proof From Theorem 4.1 and the complexly derived index
of Theorem 5.1 in the preceding chapter, we find:
±h+ RP(E) = ±Rp(cff) = n n logp((a -

X*l aeG
x(a) 1)
X even
= n Sexm(x)2) 21 LP(1,
X*l
-
,
x)
X even
by Theorem 3.6, the!- appearing because G = Z(m)*j ± 1,
by using the complex Theorem 3.1 of Chapter 3. Selecting the sign of the
regulator Rp appropriately yields the desired formula.
114
§5. The Formal Leopoldt Transform
Remark. The proof of the formula involves the complex case. Presumably
there is a direct proof, which is valid for all totally real number fields K.
Cf. the Appendix of Coates Durham lectures [Co 3], where such a proof is
given for the characteristic polynomial of a certain Iwasawa module. The
extent to which analogues of cyclotomic units will ultimately play a role in
such proofs is not clear at present.
§5. The Formal Leopoldt Transform

Let K be a field of characteristic O. Let T be the variable on the multiplicative
group. We put:
x=T- 1, T=X+1.
Then Z is the corresponding variable on the additive group. Note that
Changing variables gives rise to the notation
This last equality makes sense only whenfis a rational function.

For any power series f E K[[X]] we define the Leopoldt transform rf as a
function on integers ;;::: 0 by
fG.(Z) = 2: rf(k) Zk
k!'
As before we let
Dx = d/dX, Dz = d/dZ, DT = dJdT.
Then
Dz = (1 + X)Dx = TD T,
and for any integer k ;;::: 0,
Define coefficients Yn(k) by

Zk
2: Yn(k) "k.
00
(e Z - l)n = k=l
Then
Yn(k) = D~(eZ - l)nlz=o = (TDTY'(T - l)nIT=l
Yn(k) = 0 if k < n.
115
Lemma. (a) We have
Yn(k) = i
1=0
(_l)n-I(~) i".
I
(b) Each integer Yn(k) is divisible by n!.

Proof As to the first assertion, it is immediate by induction that
Hence
has the expression as stated. On the other hand by induction it follows that
given an integer n, for each integer k there exist integers an> ... , an-Ie such
that at = 0 if i > 0, and
(TDT)"(T - l)n = an(T - l)n + nan-leT - 1)n-l + ...

+ n(n - 1)·· ·(n - k + l)an_,,(T - 1)n-".
Putting T = 1 yields the second assertion.
In the light of the lemma, a power series f(X) has the Z-expression
f(X) = i i
n=o "=0
an Yn(~)
n. k.
Z:.
Consequently,
'" '" Yn(k) Z"

ZDzf = n~"~ an fi! (k - 1)!
= ~ ~ a k Yn(k)Z".
n~o,,~o n n! k!
These formulas can be summarized in the following theorem. We let CCZ, K)

denote the space of functions from Z into K.
Theorem 5.1. There exists a unique linear map
r: K[X] -,)0 C(Z, K)
satisfying anyone of the following equivalent conditions:
n. r«l + x)m)(k) = m" for all integers m ~ O.

116
§6. The p-adic Leopoldt Transform
I'2. r(xn)(k) = Yn(k).

n! n!
I'3.
This map also satisfies:
I'4. r(ZDzf)(k) = krf(k).

Observe that the Leopoldt transform is defined by I'3 for power series.
The other two conditions I'1 and I'2 do not make sense for power series.
However, in the next section, we shall work over a p-adic field K where these
other conditions do make sense for a suitably restricted set of power series,
with certain convergence conditions.

For simplicity, we suppose that p is an odd prime number. Let K be finite over
Qp. Let C p = completion of the algebraic closure of K. We denote the p-adic
absolute value by I I = lip, normalized so that
Ipi = lip·
We define the Leopoldt space:
2 = 2K = space of power series
xn
f(X) = .2 an-,'
n.
such that
We define the Leopoldt norm
Ilfliz = max
n
lanl·
Then2 is a Banach space,and a Banach algebra because Ilfg liz ~ Ilfllzllgllz.
Theorem 6.1. Iff E 2 then f converges on the disc of elements
For such x we have
117
Proof Obvious, because
Ipln,(p-ll :::; In!! and so Ian ~~ I: :; lanl·

We let C(Zp, K) = Banach space of continuous functions on Zp with
values in K, and the sup norm.
If a E Zp and pia, we let <a) = o.

If a E Z: we write
a = C<a) = w(a)<a)
where CE jlp-l and <a) == 1 mod p. The Teichmuller character w by definition

is such that
w(a) = C.
If s E Zp, then
<a)S = lim <a)k
k-+s
is defined in the usual way, where k ranges over positive integers approaching
s p-adically. If a is not prime to p, then we let <a)S = 0 for all s.
If X is a character on Z:, as usual we put x(m) = 0 if m is divisible by p,
so
x(m)<m)S = 0 if plm.
If X = wa. where a is a residue class modp - 1, and k == a modp - 1, then
for any positive integer i such that pti, we have
Theorem 6.2. Let a be a residue class mod p - 1. There exists a unique

continuous linear map
satisfying anyone of the following three equivalent conditions:
r .. 1. ra.((1 + x)m)(s) = wa.(m)<m)S,

for any integer m ;::: o.
r .. 2. ra.(x,n)(S)
n.
= ~
n.t=o
i (-I)n-t(~)wa.(i)<i)S
I
rex 3. r J(s) = lim

m
rf(m) = lim
m
L anr (X;)
n n.
(m)
118
where the limit is taken over positive integers m satisfying:
(*) m --7 00, m --7 S p-adica/ly, m == Q( mod p - 1.
This map also satisfies
II r .. f I ~ Ilfllo2"
and for s E Zp,
r .. 4. r ..«ZDz)f)(s) = srJ(s).
Proof. Any continuous linear map on the space of polynomials (with
Leopoldt norm) extends uniquely by continuity to the Leopoldt Banach
algebra. We shall prove that the linear map
with values
is continuous, and has the other properties. Uniqueness is obvious.

If f(X) = ~ anCXnjn!) lies in !l', then lanl--7 0 as n --7 00. To prove that
r 01 is continuous, it will therefore suffice to prove that the values
are bounded. In fact, we shall see that r,.(Xnjn !)(s) is p-integral. This will also
prove that
Fix the integer n. Let m range over integers as in r .. 3. Such integers are
dense in Zp, so it suffices to prove that
r,.(!;)(m) is p-integral for such m.
If i == 0 mod p, then imjn! is p-integral for large m. If i ;¢; 0 mod p, and
i = oo(i)(i),
then for m close to s p-adicaUy,
119
The Lemma (b) of §2 then concludes the proof that Ilralii ~ Ilfll9', and the
arguments also show that 1'.. 1 and 1'.. 3 are satisfied. It is clear that r fJ. also
satisfies l'a 4, thereby concluding the proof of the theorem.
We can r 0 the p-adic Leopoldt transform •.

The Leopoldt transform p-adically with characters of was first used by
Lichtenbaum [Li 3] to deal with elliptic curves.
In Theorem 6.5 below we shall prove that r 0 is the p-adic Gamma
transform already mentioned in §3 when applied to a power series with
coefficients in o. Hence we may then write
We recall the operator
Uf(X) = f(X) - !
P
L f('(X + 1) -
'P=l
1),
where the sum is taken over all pth roots of unity,. Then for the special
polynomial (1 + x)n we have
U«(1 + x)n) = (1 + x)n - ! L ,n(1 + x)n

p ,
= (1 + X)n (1 - ~ f ,n),
and
,n = {Po if pin
if ptn.
Hence
o if pin
Ul+xn-
« » - {(l + X)n if ptn,
and in particular, U is a projection operator, i.e.,
The next lemma describes the continuity property of the operator U for
the Leopoldt norm.
120
Lemma.
Proof
The coefficient of Xkjk! in the above sum is either 0 or
and Ie - lln-k < I(n - k)q, so the coefficient of an is not a unit atp. But
is a rational integer, and is therefore =0 modp. Hence
I
2:
P- {;"l /(eX + e- I)
1
= p- 2: (<Xl2: anbn. ) -k'
k n=O
k
Xk
•
where b n •k E Z and b n •k =0 mod p. It is then immediate that

as desired.
The next theorems prove for the Leopoldt transform on the Leopoldt
space results which have already been proved for measures.
Theorem 6.3. Let m be an integer :2: 0, and m = IX mod p - 1. Then
ruf(m) = ru/(m).
Proof The two maps
of K[X1 ~ C(Zp, K) are equal on the polynomials (1 + xy. For a fixed m

the maps
/1-+ raf(m) and /1-+ rV/(m)
are continuous, so the theorem follows by continuity.
121
4: The p-adic L-function
Theorem 6.4. For f E .IRK we have
rof(O) = Uf(O) = f(O) - ! 2:

P ,.. =1
f(C - 1).
Proof The power series for Ufin terms of X or Z have the same constant
term. Hence
ruf(O) = Uf(O).
Taking ex = 0, the theorem is obvious from Theorem 6.3, and the fact that
C- 1 lies in the domain of convergence off by Theorem 6.1.
The next theorem resulted from a conversation with Ribet.
Theorem 6.5. For S EZp andfE o[[X]] we have
i"z·
<a)8 dpAa) = rof(s).
Proof By continuity in s, it suffices to prove the theorem when s = k is

an integer ~ 1 and k == 0 mod p - 1. Let qJ be the characteristic function of
Z:. Then
i Z;'
<a)k dJlf(X) = i z"
XkqJ(X) dJlf(X)
= i Z"
Xk dJlutCx)
= DkUf(O)
= rUf(k)
= rof(k).
We now see that the Leopoldt transform is an extension of the Gamma

transform to the Leopoldt space.
122
5
Iwasawa Theory and Ideal Class Groups
We shall now study Iwasawa's theory concerning proJective limits in Zp-

extensions.
The first three sections establish purely algebraic facts about projective
limits, and finitely generated modules over the power series ring Zp[[X]]
which appears as the limit of p-adic group rings of cyclic groups. The situation
is quite similar to modules over principal rings when considering finitely
generated modules over integrally closed Noetherian domains. Cf. Bourbaki,
Commutative Algebra, Chapter VII, §4, where a general structure theorem is
given. For 2-dimensionallocal rings, this was complemented by Serre [Se 1]
who showed that reflexive modules in that case are free, thus getting a com-
plete result for Zp[[X]]. Here we shall follow Paul Cohen's proof analogous
to finding elementary divisors by row and column operations.
We shall also follow Serre's exposition [Se 1], giving the asymptotic estimate
for the orders of the factor modules. This is applied afterwards to the orders
of ideal class groups. Iwasawa's original proofs were rather complicated, and
his point of view was that of projective limits of finite abelian p-groups on
which r ~ Zp operates continuously, and which are of topologically finite
type for this action. See [Iw 1], [Iw 6]. Serre [Se 1] saw that there was an
isomorphism of categories between these objects and Zp[[X]]-modules of
finite type. He introduced this point of view which simplified the proofs and
also proved successful in subsequent applications.
The next three sections deal with arithmetic situations arising as special
cases (but which historically motivated the general results). We consider
several modules over the Iwasawa algebra. First, we deal with the projective
limit of ideal class groups. Class field theory identifies this projective limit
with a Galois group. The reader unacquainted with class field theory can
simply take for granted the' isomorphism, which is described as we need it.
123
5.· Iwasawa Theory and Ideal Class Groups
We follow mostly Serre's exposition [Se 1]. The results are valid for arbitrary
Zp-extensions, not necessarily cyclotomic ones.
The final two sections go further into certain Galois groups as modules
over the Iwasawa algebra, and also describe all possible Zp-extensions of a
given number field in class field theoretic terms. The Leopoldt conjecture
'2
would imply that there are precisely + 1 independent ones. This depends
on the Zp-rank of the closure of the global units in the local units. See §5,
Theorem 5.2.
§1. The Iwasawa Algebra

Let r be a topological group isomorphic to Zp. We write r multiplicatively,
and let y be a fixed generator, so that the isomorphism may be written
Let
Then rn is cyclic of order pn, generated by the image of y. Conversely, a

compatible system {Yn} of generators in a projective system {r n} of cyclic
groups of order pn would give rise to a generator y in their projective limit.
We have a commutative diagram
where T is a variable. Let X = T - 1, T = X + 1. Then Zp[T] = Zp[X],

and
Let
Then
and all coefficients other than the leading coefficient are divisible by p. Such
a polynomial is called distinguished.
We wish to establish an isomorphism
124
Let
We first note that if h is any distinguished polynomial, then
Zp[X]j(h) ~ A/hA.
This is immediate from the Euclidean algorithm (see Theorem 3.1), which
shows that AjhA is free of rank deg hover Zp, and similarly Zp[X]j(h) is free
of the same rank over Zp. Furthermore this same algorithm shows that the
natural map
is surjective, so is an isomorphism.
We thus obtain a natural map for each n,
whence a homomorphism
Theorem 1.1. The homomorphism 13 is an isomorphism.

Proof A trivial induction shows that
hn = (1 + X)pn - 1 E (p, X)n+l
where (p, X) denotes the maximal ideal of Zp[X], generated by p and X.

It follows that the intersection of the ideals hnA must be 0, whence the kernel
of 13 is 0. Since 13 is clearly surjective, this proves the theorem.
Note that the isomorphism 13 depends on the original choice of generator 1.

The projective limit
is called the Iwasawa algebra. Given a choice of generator 1, it is identified

with Zp[[X]] by Theorem 1.1, and then we also call Zp[[X]] the Iwasawa
algebra.
We now consider modules over the Iwasawa algebra.
For each n let Vn be a module over Zp[rn], and suppose we have homo-
morphisms
125
5. Iwasawa Theory and Ideal Class Groups
compatible with the action of the group rings Zp[rn+ 1 ] and Zp[rn] respec-
tively. We may form the projective limit
which is then a A-module.
Examples. In §4 of this chapter, Vn = Cn is the p-primary part of the ideal

class group, and so the projective limit C is a module over A = Zp[[X]]. In
Chapter 7, we shall consider projective systems oflocal units as modules over
the Iwasawa algebra.
If each Vn is a finite abelian group, or is compact, then the projective limit

V is compact, and Zp[[X]] operates continuously on V, which is then what we
call a topological module over Zp[[X]]. (Here and in the sequel, compact
means compact Hausdorff.) Note that Zp[[X]] itself is compact.
Nakayama's lemma. Let 0 be a local ring with maximal ideal m, and m-adic
topology. Let V be a compact topological o-module.
(i) IfmV = V then V = O.
(ii) If 0 is compact, and Vim V is finitely generated, then V is finitely
generated by any set of representatives of Vim v.
Proof Let U be a neighborhood of 0 in V. Since V is a topological 0-
module, for each x E V there exists an open neighborhood Ux of x and a
positive integer n(x) such that
A finite number of neighborhoods Ux cover V. Hence there exists an integer

n such that mn V c U. But m V = V implies mn V = V, and hence V c U for
all U. Since V is Hausdorff, it follows that V = 0, which proves (i).
For (ii), let Xl>' •. , Xs be representatives of Vim V, and let W be the
o-submodule generated by them. Then W is a continuous image of 0(8), and
is therefore compact, and closed. Then V/W is compact, and we have
m(V/W) = V/W.
Hence V/W = 0, and V = W, thereby proving (ii).
Next we pass to certain results concerning finitely generated modules over

the Iwasawa algebra. These will be applied to computing orders of certain
factor groups (which in §4 will be ideal class groups). The reader may omit
the rest of this section if he wishes to disregard such computations for the
moment and merely wishes to concentrate on general structural results.
Two modules V, V' are said to be quasi-isomorphic if there is a homo-
morphism
V __ V'
126
with finite kernel and cokernel. It will be shown in §3 that any finitely
generated V has a quasi-isomorphism with a finite product
(*)
where the jj are distinguished. The first factor A(T) is the free part. The other
factors are A-torsion modules.
Suppose now that V is a torsion module such that V/hnV is finite for all
n. We wish to get an asymptotic formula for the order of V/hnV, Such a
formula does not change under a quasi-isomorphism, so we are reduced to
consider the two cases when
V=A/pm and V=A/f
for some positive integer m, andfis distinguished.

In the firs't case we have
the power series ring over Z(pm). In the second case, A/f is a free module
over Zp, whose rank is degf It may happen in this second case that
is not finite. We shall first make the assumption of finiteness to get the
formula for the order, which is a power of p, so we put
Theorem 1.2. (i) If V = A/pm then en = mpn.

(ii) Let V = A/f where f is distinguished of degree d, and assume Vn finite
for all n. Then for all n sufficiently large,
en = dn.
(iii) If V is finitely generated over A such that Vn is finite for all n, then
there exists a constant c such that
for all n sufficiently large. In the representation of V as in (*) with

r = 0, we have
m = .L mi and d = .L degjj.
127
Proof In case (i) we have
and this is just Z(pm)[TJj(pn - 1), which is a free module of rank p1l over
Z(pm). Thus the computation of the order is obvious.
Let us now look at case (ii). For any h E A we let hv be the endomorphism
of V induced by h. We let
pn
Y1I = Yv, Yv = Xv + 1.
We have
ypn _ 1 = (X + l)pn - 1 == Xp" (modp),
f== X d (modp).
Hence there exists no such that for n > no we have
Xpn-l == 0 mod (f,p) and xt- 1 == 0 modp,

and therefore
Y1I-1 = Yt:"-l == 1 (modp).

It follows that
Now
Y1I+1 - 1 = (1 + Y1I + ... + y~-1)(Y1I - 1)
= (1 + 1 + ... + 1 + O(p2))(Y1I - 1)
= pU(Y1I - 1)
where u is invertible. We have therefore shown that
Furthermore, (Yno - 1) V is of finite index in V, and is therefore a free

module over Zp of the same rank d as V. This proves (ii). Case (iii) is then
obvious, thus proving the theorem.
Next we consider the case when Vjh 1l V is not necessarily finite, but make
additional hypotheses which still allow us to compute the orders of certain
factor groups asymptotically, and which are satisfied in the application to
ideal class groups.
We let
128
§2. Weierstrass Preparation Theorem
We say that V is of Iwasawa type if there exist elements Vb ••• , VB E V such

that, if we put
Uo = Zp-submodule of V generated by (')I - l)Vand Vi,"" v.,

U" = g"Uo,
then VjU" is finite for all n. In particular, VjUo is finite. For a module of
Iwasawa type, we let
V" = VjU".
Theorem 1.3. Assume that V is of Iwasawa type. Then the conclusions of
Theorem 1.2(i), {ii), (iii) remain valid, except that in 1.2(ii) we have to write
the exponent
e" = dn + Co
with some constant co.
Proof Note that Case (i) is unchanged, only Case (ii) is now slightly
different, but the proof runs along entirely similar lines as follows. In this
case, V is Zp-free of rank d. An argument similar to that of Theorem 1.2(ii)
shows that
for all n ~ no. Let W = g"o V. Then
e(Vjg"V) = e(VjW) + e(Wjp"-"oW) = Cl + den - no)
for some constant Cl> since W has the same Zp-rank as V. This proves the
theorem.
§2. Weierstrass Preparation Theorem

The proof ofthe Weierstrass theorem in this section is due to Manin [Man 1].
We start with the Euclidean algorithm.
Theorem 2.1. Let 0 be a complete local ring with maximal ideal m. Let
2: alX!
<Xl
f(X) =
!=o
be a power series in 0 [[ X]], such that not all ai lie in m. Sayao, ... , an -1 E m,
and an E 0* is a unit. Given g E o[[X]] we can solve the equation uniquely
g = qf+ r,
with q E o[[X]], r E o[X], and deg r ::; n - 1.
129
Proof Let a and 't' be the projections on the beginning and tail end of the
series, given by
L bjXjl-+ L bjXj = bo + b1 + ... + b"_1 X "-1

,,-1
a:
1=0
L bjX'1-+ L bjX'-" = b" + b,,+1X + b,,+2 X2 + ....

00
't':
1="
Note that 't'(X"h) = h for any h E o[[X]], and h is a polynomial of degree <n
if and only if't'(h) = O.
The existence of q, r is equivalent with the condition that there exists q
such that
't'(g) = 't'(qf).
But
f = af + X"'t'(f).
Hence our problem is equivalent with solving
't'(g) = 't'(qa(f» + 't'(qX"'t'(f) = 't'(qa(f) + q't'(f).

Note that 't'(f) is invertible. Put Z = q't'(f). Then the above equation is
equivalent with
Note that
a(f).
't' 0 't'(f). o[[X]] ~ mo[[X]],
because a(f)/'t'(f) E mo[[X]]. We can therefore invert to find Z, namely
which proves both existence and uniqueness and concludes the proof.
Theorem 2.2 (Weierstrass Preparation). The power series f in the previous

theorem can be written in the form
where hj E m, and u is a unit in o[[X]].
130
§3. Modules over Zp[[X]]
Proof Write
X" = qf + r,
by the Euclidean algorithm. Then q is invertible because
q = + c1X + .. .
Co
f=··· + a"X" + .. .
so that
1 == coa" (mod m),
and Co is a unit in o. We obtain qf = X" - r, and
f = q-l(X" - r),
with r == 0 (mod m). This proves the theorem.
The integer n in Theorems 2.1 and 2.2 is called the Weierstrass degree off,
and is denoted by
We see that a power series not all of whose coefficients lie in m can be
expressed as a product of a polynomial having the given Weierstrass degree,
times a unit in the power series ring. Furthermore, all the coefficients of the
polynomial except the leading one lie in the maximal ideal. Such a polynomial
is called distinguished.

The structure of finitely generated modules over Zp[[X]] was first determined
by Serre [Se 1] who introduced this point of view in Iwasawa theory. As
already mentioned, cf. Bourbaki for general structure theorems over integrally
closed Noetherian domains. Paul Cohen showed how one could give a proof
along the standard lines of row and column operations, cf. [L 3]. Robert
Coleman pointed out to me that the inductive step as given in [L 3] had to be
modified, and I am indebted to him for the exposition given in the lemma and
Theorem 3.2 below.
We let A = o[[X]], where 0 is a complete discrete valuation ring. We
denote by p a prime element of o. By a finite module over 0 we mean a
finitely generated module annihilated by some power pk and some distin-
guished element A. If 0 = Zp, then "finite" has the usual meaning.
By a quasi-isomorphism we mean a homomorphism with finite kernel and
cokernel. We denote a quasi-isomorphism by the sign
M,...,M'.
131
Theorem 3.1. Let M be a finitely generated A-module. There exists a quasi-

isomorphism
where each jj is a distinguished polynomial, irreducible in o[X], i, j range

over finite sets o/indices, andA(T) is the product 0/A taken r times,/or some
integer r.
The rest of this section is devoted to the proof.

Suppose that M has generators Ub .•. , Un. Relative to such generators we
can form the matrix of relations, whose rows are vectors
such that
Since A is Noetherian, a finite number of the rows generate aU of them.

Performing the usual row and column operations on the matrix amounts
to changing the generators of the module. We shall describe other operations,
corresponding to embedding the module in a bigger one with finite cokernel.
An element A E A is called p-free if A does not lie in pA, in other words, if
we can apply the Weierstrass preparation theorem to it.
Suppose that there is a relation of the form
where ..11 is p-free. We can form the new module M' obtained by adjoining a
new generator v with the relations
This can be formalized by considering a direct sum
M EB (v)
modulo the desired relations, i.e., modulo the submodule generated by the
elements
It is then immediately verified that the canonical map of M into the factor
module is injective. The factor module M'jM is annihilated by p and ..11,
whence is finite. Furthermore, the elements v, U2, •.. , Un generate M', and
have the relation
132
§3. Modules over Zp[[X)]
In terms of the relation matrix, this means that we shall allow the following
operations, replacing the matrix R by a matrix R'.
o 1. If R contains a row (..1.1> PA2' ... , pAn) with ..1.1 not divisible by p, then
we let R' be the matrix whose rows consist of
and the rows of R with first element multiplied by p.
Observe that in this first operation, we may have ..1.2 = ... = An = O.

Next suppose that some power p" (k :2: 1) divides all elements of R, but
that there exists one relation
p"(Al, ... , An)
such that ..1.1 is distinguished (or equivalently, ..1.1 is not divisible by p). We may
then form the module M' obtained by adjoining a new element v with the
relations
Again, it is easily verified that M is embedded in M' and that M'jM is finite.
Note that p"(v - Ul) = O. The relations of the submodule
are generated by R and the additional relation
We have a direct sum decomposition
and the relations of v - Ul are generated by p". To prove the theorem, it

suffices to consider the first component of M'. Thus our second operation is
described as follows.
o 2. If all elements of the first column in R are divisible by p", and if there
exists one relation (p" ..1.1, ... , p" An) such that ..1.1 is not divisible by p, then
we let R' consist of R and the new row
Finally we allow one more operation:
o 3. If R has a relation of the form
133
and there exists an element A not divisible by p such that
is also a relation, then we may replace R by the matrix R' having the same
rows as R, except that the row pk(Al' ... , An) is replaced by
This operation corresponds to the surjection with finite kernel
Row or column operations, or 0 1, 0 2, 0 3 will be called admissible

operations.
Given a matrix R over A, we define
where (a;j) ranges over all admissible transformations of R which leave

unaltered the components of the first k - 1 rows.
Remark. If R' is obtained from R by admissible operations leaving the

values in the first k - 1 rows unaltered, then
Let r ;::: 1 be an integer. Suppose that R has the form
All 0... 0 0... 0)

(
b ... iT - 1. T- 1 ~ . . . ~
* ... * * ... *
and
( ATT ..• ATn) ""' O.

* ... *
Assume also that Ali for i = 1, ... , r - 1 is a distinguished polynomial with
the property that
deg(k)(R) = deg Akk for k = 1, ... , r - 1.
Then we shall say that R is in (r - 1)-normal form. If r = 1 then this condition

is vacuously satisfied, and is the starting point for the induction of the
following lemma.
134
Lemma. Suppose that R is in (r - 1)-normal form, with its first r - 1

diagonal elements All, ... , ..1.r - 1 ,r-1' Then by admissible transformations, we
can transform R into a matrix which is in r-normal form, and has the same
first r - 1 diagonal elements.
Proof Using 0 1 with respect to each of the first r - 1 rows, we may
assume without loss of generality that any given power pk (k ~ 0) divides
all components Ali with i ~ rand j = 1, ... , r - 1, that is all components
lying below the portion of the matrix which has already been diagonalized.
Using 0 2, we may then arrange that p does not divide some ..1.11 with i ~ r,
and j ~ r. After a succession of admissible transformations on the lower right
matrix
( ..1.rr •.• Ar)

*********:
induced by admissible transformations of R which leave the first r - 1 rows
fixed elementwise, we may then find some element ..1.i1 with i ~ rand j ~ r
such that
degw Ati = deg(r)(R).
The Weierstrass preparation theorem allows us to assume that this element

..1.;; is a distinguished polynomial, and
deg Ali = deg(r)(R).
Finally, row and column interchanges which do not involve the elements Ai!
(i = 1, ... , r - 1) allow us to assume that..1.;; = ..1.rr-
There remains to show that we can make all other elements on the rth
row equal to 0 after appropriate transformations. By the Euclidean Algorithm,
we may assume that
deg ..1.Tj < deg ..1.TT for r #- j

deg ..1.Tj < deg ..1.jj for j < r.
We first deal with the elements to the right of ..1.TT on the rth row. We may
assume that ..1.Tj with j > r is divisible by p, otherwise we contradict the
minimality of the degree of ArT' Using 0 1 repeatedly as before with respect
to the first r - 1 rows, we may then assume that all elements ..1.T1 with j < r
are divisible by a high power pk. We then use 0 1 with respect to the rth row,
to divide all elements ..1.Tj (j #- r) by successive powers of p, thus leading to
some element ..1.r1' withj' > r not divisible by p, a contradiction of deg ..1. rr =
deg(r)(R). Thus ..1.0 = 0 for j > r.
For elements ..1.ri to the left of Am that is withj < r, if some such element
is not 0, then we may use 0 1 with respect to the rth row to divide by p,
135
until we are in the situation where there exists j < r such that ATi is not
divisible by p, contradicting the facts that
deg AT; < deg Aii and deg Ajj = deg(J)(R).
Thus we have put the matrix in r-normal form, and proved the lemma.
Theorem 3.2. If R is a matrix of relations, we can transform R with a finite

number of admissible operations into a matrix R' of the form
( o~1l ~ ••. ~ ~
O"'Arr 0···0
... ~)
00···00···0
where Ali are distinguished polynomials.

Proof By the lemma, we can replace R by a matrix R' of the form
where All are distinguished polynomials, and
deg AI! = deg(l)(R) for i = 1, ... , r.
By the Euclidean Algorithm, we may assume that Ail = 0 or
deg Aji < deg Ali for j -# i.
In fact, we contend that Ail = 0 for all j -# i. Suppose otherwise, so that

Ail -# 0 for some j > r > i, so we have a relation
(Ail> ... , Ajr> 0, ... , 0)
not identically O. Let
Then A is not divisible by p, and AUt = 0 for i = 1, ... , r, so
(AA!l' ... , AAi" 0, ... , 0)
is also a relation. By 0 3 we may assume without loss of generality that some

Ail> ... , Ajr is not divisible by p, and then contradict the minimality condition
on the Ali' This proves the theorem.
136
§4. Zp-extensions and Ideal Class Groups
We return to the module interpretation, to see that Theorem 3.2 implies

the theorem. Indeed, any module with matrix of relations R' as in Theorem
3.2 is isomorphic to
r
A,,-r EEl EB AI(Ali)'
1=1
Finally, iff, g are distinguished and relatively prime, the map
A(fg) -7 AI! EEl AIg
is an embedding with finite cokernel. This allows us to decompose the factors

AIAli into a direct sum of factors
where jj .is distinguished and irreducible, thereby concluding the proof of

Theorem 3.1.

Let Ko be a number field. An extension Koo of Ko is called a Zp-extension if
it is abelian, and its Galois group is isomorphic to Zp. To give such an
extension is the same as to give a tower of fields
""
K"" = U K" :::> ••• :::> K" :::> ••• :::> Ko
,,=0
such that K" is cyclic over Ko of degree p".
Examples. Let p be a prime number. Let
Kn = Q(JlpR+l) if P is odd
Kn = Q(JlpR+2) if P is even.
This gives the cyclotomic Zp-extension over the field Ko.

More generally, let K be any number field, let
be the extension obtained by adjoining all p-power roots of unity. Then K(p)
is abelian over K, and it is easy to see that the fixed field of the torsion sub-
group of Gal(K(P)IK) is a Zp-extension of Ko = K, called the cyclotomic
Zp-extension. We study it later in the book. Note that a non-totally real field
K always has non-cyclotomic Zp-extensions, cf. §5. Natural examples can be
constructed with elliptic curves having complex multiplication, cf. [C-W].
We say that a prime ideal .);lo of Ko is almost totally ramified in a Galois
137
extension K' if the inertia group of a prime .p in K' over .po is of finite index
in Gal(K'jKo). We say .po is almost uruamilied if its inertia group is finite.
We consider the following condition of Iwasawa.
IW. K" is totally ramified over Ko over a finite number of prime ideals
.pb ... , .ps lying above p, and is unramified over all other prime ideals.
Lemma. Let Koo/Ko be a Zp-extension. Then:

(i) Only a finite number of prime ideals of Ko ramify in Koo, they lie above
p, and they are almost totally ramified.
(ii) For some positive integer d, the extension Koo/Kct is a Zp-extension
satisfying IW.
Proof Some prime ideal .p of Ko must ramify in Kf:> because class field
theory says the maximal unramified abelian extension of Ko is finite. Let I
be the inertia group. It is a closed subgroup of r, and #0, hence equal to
pmzp for some m, so that .p is almost totally ramified. Over the completion
Ko.p, the maximal tamely ramified abelian extension is finite. Hence the wild
ramification group is of finite index in r, thus showing that .p lies above p.
This proves (i). If we let .pb ... , .ps be the finite number of primes which are
almost totally ramified; and let
be the inertia groups, and d = max dt> then Koo/Kd satisfies condition IW
as desired.
Assume that condition IW is satisfied.

The same lemma as in Chapter 3, §4 shows that the norm map between
any two successive steps in the tower is surjective on the ideal class groups.
We let Cn = C:!') be the p-primary part of the ideal class group in Kn. Then
we have a surjective sequence
and we let
C = lim proj Cn
be the projective limit. We may view C as consisting of all sequences
with Cn E Cn and Cn +l mapping on Cn under the norm map.

Let Mn be the maximal p-primary abelian unramified extension of Kn, in
138
other words the p-primary part of the Hilbert class field of Kn. There is an
isomorphism given by class field theory
such that the following diagram is commutative.
I
Cn + 1 -? Gal(Mn +dKn + 1)
Normi Restriction
Cn ----* Gal(M,,/Kn)
Since K" is totally ramified over K n , it follows that M" is linearly disjoint
from Koo over Kn. The lattice of fields looks as follows. We let Moo = U M".
We let
Remark. If we replace Ko by Kl then Koo over Kl satisfies the same con-

dition IW, so a number of results proved for Koo over Ko apply a fortiori to
Koo over K 1 • Observe that if}' is a topological generator for r, then
Theorem 4.1. Assume first that IW is satisfied with s = 1. Let [ be the

inertia group of any prime above .p in G. Then:
(i) G = [Gc is a semidirect product, and the restriction of [ to Koo gives
an isomorphism of [ and r.
(ii) The commutator group G' = Gb- 1 •
(iii) We have isomorphisms
139
Proof We have an exact sequence
The image of I in r by restriction to K", is surjective because K", is totally

ramified over Ko. It is injective because M", is unramified over K", and so
In Go = {I}.
This proves (i). If U E Go then u Y- 1 is a commutator because r operates on

Go by conjugation. Hence Gb- 1 c G'. On the other hand, G/Gb- 1 is abelian,
so the reverse inclusion also holds and (ii) is proved. Finally, Mo is the
maximal p-primary abelian unramified extension of Ko and so Gal(M",/ Mo)
is the smallest subgroup of G containing G' and the inertia group 1. Since G
is the semidirect product of I and Go, we see that (iii) follows from (ii), and
conclude the proof of the theorem.
Corollary. We have an isomorphism
Cn ~ Gal(Mn/Kn) ~ C/CyPfI_ l ~ Go/GlIn - l •
Proof Apply the theorem to the situation where Ko is replaced by Kn.

Next consider the general situation with a finite number of primes.
Theorem 4.2. Assume that IW is satisfied, with primes %h, ... , ~s. Let Ii be
the inertia group of ~j in G. Then:
(i) There is a semidirect product decomposition
and G' = Gb- 1.

(ii) Let Uj be a generator for I" and write
Then
Co ~ GO/(7:1' ... , 7:., G"/:-1).

Proof Identical with that of Theorem 4.1, except that in the present more
general situation, we have to look at the smallest subgroup of G containing
the commutator group G' and all the inertia groups I j instead of a single
inertia group 1.
Corollary. Let Uo be the Zp-submodule of Go generated by the elements

7:1; ... ,7:8 and Gb- 1. Let
140
Then
Cn ~ Gc/Un •
Proof We apply the theorem to K"" as Zp-extension of Kn. This has the
effect of replacing y by yp" and O"j by O"r.
Then 'rj is replaced by ('rjY", because
for every positive integer k, we have
whence for k = pn we obtain
Then
Un -- GyP"-l(
C 'rlo ••• , 'rs)gn
where ('rlo"" 'r.) is the group generated over Zp by 'rlo"" 'r•• Since
gn(Y - 1) = y',n - 1,
we find Un = Ugn, which proves the corollary.
It will be easily shown below in Theorem 4.4 that C is finitely generated

over the Iwasawa algebra. Then Theorem 4.2 and its corollary show that C
is of Iwasawa type as defined in §I, so that one can apply the counting
procedure given there, to get an asymptotic formula for the orders of the
groups Cn.
I wasawa has conjectured that m = 0 in the case of the cyclotomic tower
Q(Jl.pn), so that in this case, the order ofthe ideal class group (p-primary part)
would have the form
Card Cn = pan+c
for n sufficiently large, in analogy with the orders of points of p-power order
on abelian varieties. This conjecture has recently been proved by Ferrero and
Washington. On the other hand, he has given examples of non-cyclotomic
Zp-extensions K"" over Ko for which m > O.
Theorem 4.3. Assume that IW is satisfied with one prime. If Co = {I}, then
Cn = {I} for all n.
Proof If Co = I, then Theorem 4.1 shows that C = CY-l. Viewing Cas
module over Zp[[X]], this means that C = XC. But X is contained in the
maximal ideal of Zp[[X]]. By Nakayama's lemma, it follows that C = {I},
as desired.
141
Remark. We could let Kn = Q()lp n+l)+ be the real subfield of the cyclo-
tomic field. It is a conjecture of Vandiver in that case that Co is trivial.
[Remember: By definition, Co = CIf) in this chapter.] Thus the Vandiver
conjecture may also be formulated by saying that
ht is prime to p,
and Theorem 4.3 shows that if this is the case, then hi; is also prime to p
for all n.
It is probably so that Vandiver actually never came out in print with the
statement: "I conjecture etc." In [Va], he proves that if the Fermat conjecture
is false for x P + yP = zP with relatively prime integers x, y, z such that xyz
is prime to p, then h+ is divisible by p. Later in the paper, he states: "The
theorem last mentioned as well as Theorem I indicates that much of the
writer's work concerning Fermat's last theorem is tending toward the possible
conclusion that if the second factor of the class number of k«() is prime to I,
then Fermat's Last Theorem is true." The terminology "Vandiver's con-
jecture" seemed appropriate to me. In any case, I believe it.
Theorem 4.4. For any Zp-extension the module Cover Zp[[X]] is a finitely
generated torsion module.
Proof Suppose first for simplicity that condition IW is satisfied with only
one prime. Then C/mC is a factor group of C/cr-l, which is none other than
Co by Theorem 4.1, and is therefore finite. That C is finitely generated is a
special case of Nakayama's lemma.
In general, when IW is satisfied but with several primes, then we have to
use another argument. By Theorem 4.2 we know that
is finitely generated over Zp of rank uniformly bounded by s. Nakayama's

lemma again shows that C is finitely generated over the Iwasawa algebra.
Furthermore, applying Theorem 4.2 to K",/Kn, that is replacing y by ypn,
shows that
C/(ypn - l)C
is also finitely generated over Zp with a similar bound s for the rank. By the
structure theorem of §3, if V is finitely generated over A, then there is a quasi-
isomorphism
(*)
where jj are distinguished. We use this with V = C, writing Vadditively.

The uniform bound on the rank immediately shows that there cannot be any
free part, i.e., r = O. This proves Theorem 4.4.
142
§5. The Maximal p-abelian p-ramified Extension
§5. The Maximal p-abeIian p-ramified Extension

The next two sections describe in class-field theoretic terms some properties
of the Galois group of the maximal p-abelian p- ramified extension of a number
field, and describe its Zp-extensions.
Let K be a number field. We let:
MP(K) = the maximal p-abelian p-ramified extension of K.
MiJr(K) = the maximal p-abelian unramified extension of K.
We fix the prime number p and the field K, so we sometimes omit reference
to them in the notation.
J = JK = ideles of K, and U is the group of unit ideles,
U = n Up
lJ
and JOO = n K;.
veSco
In the first product, ,p ranges over the prime ideals of K. We write
E = EK = units in K. We have an embedding on the diagonal:
G~b(K) = Gal(Mp(K)jK).
By class field theory, an abelian extension of K is unramified at primes

dividing 1 if and only if its associated group in the ideles contains Uz• Conse-
quently we have an isomorphism
where the bar denotes closure in the idele topology. We have the inclusions
The first factor group
is isomorphic to the Galois group of the Hilbert class field, and is finite.
The second factor group is equal to
143
Lemma.
Proof Let Uif) be the group of units in Up which are == 1 mod pt>. Then
the groups
form a fundamental system of neighborhoods for U[P]Joo K*, and their inter-
section is this closure. Intersecting with Up (whose elements have component
1 at all primes not dividing p) shows that
Taking the intersection for all n proves the lemma.
Theorem 5.1. Let H be the p-Hilbert class field of K. Then we have an

isomorphism
Gal(Mp(K)jH) ~ p-part of Upj(JpE

= U~l)j(U~l) n (JpE).
Again, as p is fixed, we write simply Up(E. By a quasi-isomorphism, we

shall mean a homomorphism with finite kernel and cokernel. We denote a
quasi-isomorphism by a single "'. The theorem yields a quasi-isomorphism
Furthermore, since Up contains an open subgroup of finite index isomorphic

to Z~K :Ql, by means of the exponential map, say, we have a quasi-isomorphism
G&b(K)
p '" Z[K:Ql-r"
p , where rp = rankz"It = rp
(E).
The Leopoldt conjecture states that rp = r = r1 + r2 - 1.

Let ZP(K) = composite of all Zp-extensions of K. From the quasi-
isomorphism we find:
Theorem 5.2. Assume the Leopoldt conjecture for K. Then we have a quasi-
isomorphism
G~b(K) '" Z~2+1 ~ Gal(Zp(K)jK).

Proof The first statement comes from the definitions and
For the second statement; we note that the composite of all Zp-extensions of
144
§6. The Galois Group as Module over the Iwasawa Algebra
K has a Galois group embedded in the product of Zp with itself, and as such
is a torsion free finitely generated module over Zp, whose rank is exactly
'2 + 1 by the first statement.
Example. Let K.o = QCJ.l<P»). Let Un be the local units in the completion
of K n , congruent to 1 mod .):In' Let U~ be the subgroup of units whose norm
to Qp is 1. Assume the Vandiver conjecture. Let the notation be as in §4 of
the preceding chapter. Then we obtain an isomorphism
Without assuming the Vandiver conjecture, we shall study the projective limit
of the local groups in Chapter 7.

Let Ko be a number field, K",/Ko any Zp-extension, with Galois group
r = {y},
with topological generator y. Let Q be a p-abelian extension of K", which is

also Galois over Ko. For each n we let Q n be the maximal subfield of Q which
JI
is abelian over Kn.
lG·G:
) {In
l rn I
Ko
The Galois groups are denoted by the letters shown on the diagram. Since Q
is assumed Galois over Ko and is abelian over K"" it follows that the com-
mutator subgroup is
in other words, it consists of all elements
It is frequently useful to view G as an additive module over the Iwasawa
145
algebra. Indeed, rn operates by conjugation on Gal(Q/Kn), and hence on the

commutator group
Hence
lim Gn = G is a compact module over A = Zp[[X]] = lim Zp[rn].
Taking Kn as ground field instead of K o, we obtain mutatis mutandis
Thus in terms of the Iwasawa algebra, we find
We denote by the sign'" a quasi-isomorphism of A-modules.
Theorem 6.1. Let Q be the maximal p-abelian p-ramified extension of K~,.

Then:
(i) G = Gal(Q/Koo) is finitely generated over the Iwasawa algebra, and in
fact
G/GY-l '" Z~ where p = [Ko : Q] - rp - 1.
(ii) If Ko satisfies the Leopoldt conjecture, then p = r2, and
Proof By definition,
and the rank over Zp of a subgroup of finite index in its Galois group was
determined to be [Ko: Q] - rp in Theorem 5.2. Taking into account r itself
shows that G/ XG '" Z~ where p is as stated. Nakayama's lemma then proves
the first assertion, and (i). Part (ii) is then a matter of definitions.
Theorem 6.2. Assume that Ko is totally imaginary, and that each Kn satisfies
the Leopoldt conjecture (namely
Then there is a quasi-isomorphism
where Gtor is the A-torsion submodule of G.
146
Proof From the structure theorem, we know that
On the other hand,
By Theorem 5.2 we know that
From the structure theorem, one sees easily that this is possible only if t = r2,
as desired.
The above theorems give a sample of Iwasawa's results [Iw 12]. It is

possible to vary some of the hypotheses to obtain variants. For instance,
one need not assume the full Leopoldt Gonjecture in Theorem 6.2, merely
assume that the defect in that conjecture is bounded as function of n. For
the cyclotomic Zp-extension, this can be proved easily, see for instance Green-
berg [Gr 4].
147
6
Kummer Theory over Cyclotomic
Zp-extensions
In the last chapter we studied the ideal class groups in a Zp-extension of a

number field. Here we shall consider especially the cyclotomic Zp-extension,
and then Kummer extensions above it, as in Iwasawa [Iw 12], obtained by
adjoining pnth roots of units, p-units, and ideal classes of p-power order.
We also give the Leopoldt-Iwasawa theorem that the Vandiver conjecture
implies that c- is cyclic, in a precise version for the cyclotomic extension of Q,
following Kubert-Lang [KL 9]. We prove that the Galois group of the
Kummer extension obtained by adjoining p-power roots of p-units is 1-
dimensional free over the Iwasawa algebra. As a consequence, we see that
c- is a quotient of this free module. See Leopoldt [Le 5] and the last Satz
in [Le 10], as well as Iwasawa [Iw 7], Theorem 2. In the limit, there is an
analogous (but less precise) statement of Greenberg [Gr 4], see also Coates
[Co 3], Theorem 5.7.
For a discussion of the case of totally real number fields, cf. Coates [Co 3],
[Co 4]. In this connection it is likely that the units conjectured by Stark [St]
(see also Lichtenbaum's conjectures [Li 2]) will playa significant role similar
to the one played by the cyclotomic units, to clarify the situation.
§1. The Cyclotomic Zp-extension

Let flIP) be the group of p- power roots of unity. Then Q(fl(P)) is the composite
of an extension of degree p - 1 if p is odd, Q(i) if p = 2, and a Zp-extension
which is uniquely determined as the fixed field of the (finite) torsion group of
the Galois group, and will be called the cyclotomic Zp-extension. We denote
it by ZiQ). It is real. If K is a number field, we let
148
be the composite of K and the cyclotomic Zp-extension. Then CyciK) is a

Zp-extension of K. If K is totally real, then this cyclotomic Zp-extension is
also totally real.
Suppose on the other hand that K contains the pth roots of unity if p is
odd, and contains i if p = 2. Let qo be the power of p such that the qoth roots
of unity lie in K. Let
Then [Kn+l : Knl = p, and
is Zp-extension of K. Let r= Gal(Koo/Ko) and let
,,: r ~ 1 + qoZp
be the canonical representation such that for any pnth root of unity 'we have
A Galois extension is called p-abelian if its Galois group is a projective

limit of finite p-abelian groups. We now discuss properties of such extensions
of Koo which are Galois over Ko.
For the rest of this section, we assume that Ko contains the pth roots of
unity if p is odd and i if p = 2. Let An be a subgroup of K;, and let
We assume that An is stable under r no
Ordinary Kummer theory gives a pairing
expressed by the symbol
for 0' in the Galois group and C( E A~/pn. If y Ern then
where 0'1 = yay -\ and y is any extension of y to Kn(A~/pn). Indeed,
149
6. Kummer Theory over Cyclotomic Zp-extensions
We may also write
The group A~/P" mod A~/P" n K; has exponent pn, so exponentiating with a
p-adic integer is well defined. In particular, we may rewrite the functorial
formula in the form
We wish to pass to the limit. We could have taken the Kummer pairing on
writing the symbol <u, rx) without an index n, defined by the same formula.
The Galois group on the left can be identified with a subgroup of
Gal(Kn(A~/P")/ K n), arising from the change of base of the Kummer extension
from Kn to K",. Let Gn be the Galois group on the left, so
The field diagram is as follows.
The group Gn is a rn-module, hence a Z(pn)[rnl = An-module. Hence via the

natural homomorphism, it is a A-module, where
A = limAn
is the Iwasawa algebra, isomorphic to Zp[[X]], and X = Yo - 1, where Yo is a
fixed generator of r.
Let
be an element of A. We define the Iwasawa involution
150
Then X* is also in the maximal ideal (p, X) of A, and
is an automorphism of A. The functorial formula for the action of Y E r on

the Kummer symbol can then be expressed in terms of the involution by
for (J E Gn and IX E A~/pn/(A~/pn () K!).

In the applications, we also pass to the limit on n for the Kummer pairing.
We suppose that
Let
We have a compatible system of pairings for m ~ n:
The Galois groups on the left form a projective system, and the Kummer
groups of field elements on the right of the pairing form an injective system.
At each finite level, we have a compact-discrete duality. In the limit, we have a
similar compact-discrete duality
GA x lim A~/pn/(Ak,pn () K!) - ? Jl(P)

-+
with values in the p-primary roots of unity.

The action of An on Gn is compatible in the projective limit, so the limit
group GA is a topological compactA-module. We shall investigate its structure
for various systems {An} obtained from units and ideal classes in the next
sections. It will also happen that we consider two groups, say
A ;:::)B,
in which case Q A ;:::) QB' It is clear in each case that Gal(QA/Q B) is aA-module,
151
and that the Kummer pairings and involution described above also apply to
this intermediate situation.
§2. The Maximal p-abelian p-ramffied Extension of the

Cyclotomic Zp-extension
A Galois extension is called p-abelian if its Galois group is a projective limit
of finite p-abelian groups. It is called p-ramified if it is unramified at all primes
(including infinity) not dividing p. We let:
Mp(K) = the maximal p-abelian p-ramified extension of K.

.Mi}r(K) = the maximal p-abelian unramified extension of K.
We fix the prime number p and the field K, so we sometimes omit reference
to tl;lem in the notation.
Even if K is infinite over Q we may define Mp(K) and M~r(K) as above.
It is then immediate that
where the union is taken over a family of subfields F of K finite over Q, whose
union is K, and which is cofinal with the family of all subfields of K finite
over Q. For instance, if K is finite over Q, and Koo is a Zp-extension, then
Mp(Koo) = Un Mp(Kn).
A similar remark applies for M~r(Kco).
Throughout this section, we let:
Koo = cyclotomic Zp-extension of K o, and we assume that Ko contains

the pth roots of unity if p is odd, contains i if p = 2.
D = maximal p-abelian p-ramified extension of Koo.
En = units in Kn and E = U En.
DE = Koo(E1/P"')
An = group of elements IX in K: such that (IX) = apn where a is (frac-
tional) ideal prime to p, and A = U An.
DA = U K.,(A~/pn)
Bn = p-units in Kn = group of elements whose ideal factorization con-
tains only ideals dividing p.
DB = K.,(BI/P"') .
152
§2. The Maximal p-abelian p-ramified Extension of the Cyclotomic Zp-extension
We have the following diagram of fields.
It is clear that DA and DB both contain DE' In fact, both A and B contain E.
Lemma 1. D = DAD B.
Proof By Kummer theory, D is a composite of cyclic extensions. Let
K",(a l / pm ) c D for some a E K",. Then a E Kn for some n. We take n ~ m
and also such that
Kn(a l / pm) is p-ramified over Kn.
Then a necessarily has an ideal factorization
where b is p-primary and a is prime to p. Let h be the class number of Kn ,

and write h = prd with d prime to p. Then
where (al) = air. and (ft) = blr.. Furthermore,
and also
K n+r(alr./pn+') c 17
An+r
(l/pn+r pl/pn+r El/pn+')
al, ,n+r'
This proves the lemma.
Theorem 2.1. The Galois groups Gal(DA/DE) and Gal(DB/DE) are A-torsion
modules. So Gal(D/DE) is a A-torsion module.
Proof We shall analyze each Galois group separately, and get a closer
view of its structure.
153
The extension DA/DE.

Let GAlE = Gal(DA/D E). For now abbreviate GAlE = G, and let
The field diagram is as follows.
It is clear that G = projective limit of the groups Gn , and that Gn is a

Z(pn)[rnl-module, so in the limit, G is a A-module.
As in Chapter 5, let:
Cn = Cl(P)(Kn) = p-primary subgroup of ideal class group of Kn.
Then we have a homomorphism
given by
if (ot:) = apn • If u is a unit in D such that u pn E An, then u pn E En. The kernel
of our homomorphism is therefore precisely E~/pn, so we have an injective
homomorphism
which is also a A-homomorphism. Let JOt" be its image. Then the Kummer
pairing is isomorphic to a pairing with An, namely:
I I
G" x .91" -------;.. J1,pn
In addition, this isomorphism is compatible with the limiting process:

G n + 1 x .9/,,+1 -?-
J1,pn+l
1 1
Gn x d n --+
1
Il pn
154
§2. The Maximal p-abelian p-ramified Extension of the Cyclotomic Zp-extension
Hence we get a compact-discrete duality
G x d -+ JlO»
where d = direct limit of d,.. By Chapter 5, Theorem 4.4, there exists

A E A such that CA = 1, so C~ = 1 for all n, and A* = 1 for all n. By Kummer
duality, for (l E G we get
Hence (lAO = 1, so A* annihilates G, which is therefore a torsion module over

A as desired.
In addition, we note that the direct limits
lim A~/pn/E~/pn = lim d,. and lim Cn = Coo

-7 -7 -7
are equal since any element in Coo has a representative ideal prime to p.
Consequently we get the additional information:
Theorem 2.2. The Kummer pairing gives rise to a compact-discrete duality
Remark. Iwasawa has also shown that C = lim Cn is quasi-isomorphic to

+-
Hom(lim Cn, Qp/Zp) (see Theorem 11, p. 266 of [Iw 12]).
-7
The extension DB/DE'

Let GBIE = Gal(QB/DE)' For now abbreviate GBIE = G. By the Lemma of
Chapter 5, §1 we know that there is only a finite number of primes ~b ••• , ~s
dividingp in some finite extension K d , such that ~b . . . , ~s are totally ramified
in Koo. Let h be the class number of K d • Let
Then
n _ n ( Ill'''' ,···,1I: Ill'''') •
~~B-~(;E1I:l s
It is immediate that
In particular, the structure theorem for finitely generated A-modules implies

that G cannot have any free part, so is a A-torsion module. This proves
Theorem 2.1.
155
For additional information concerning the fixed field of
(for referring to A-torsion), cf. for instance Coates [Co 1], Theorem 5. Iwasawa
has an example showing that there are cases when the fixed field is not
necessarily DE'
Theorem 2.3. Assume that there is only one prime in Koo lying above p.
Then
where Ep is the group olp-units in K oo , and
Proof We consider the diagram offields:
The ideal above p in Qn is principal, say generated by the element An =

1 - (n' The degree [Kn : Qp] is bounded independently of n, and we have
where en is the ramification index, bounded by this degree. Taking e to be the

least common multiple of the integers en shows that .p~ is principal for all n.
We apply this to the previous discussion of the extension DB = DE(nj:'P"').
As ideals we have
for n sufficiently large, and jn is divisible by arbitrary large powers of p as

n -+ 00. Furthermore
It is then clear that DB = DE'
156
§3. Cyclotomic Units as a Universal Distribution

Letp be a prime number.
Let 81' be the group generated by ±p.<1') (p-power roots of unity) and by
the elements
, - 1, with, E p.<1') and, :I: 1.
We call 81' the cyclotomic p-units. They satisfy the following relations:
CU 1. U-1(' - 1) = _,-1(, - 1)
CU2. I1 (''1 -
.,1'=1
1) = ,1' - 1 if ,1' :I: 1.
CU3. n (' - 1) =p.

{I''' = 1
{primitive
For this last one, note that the pth roots of unity satisfy
X1'-l + ... + 1 = O.
Replacing X by X + 1 yields the equation for' - 1, where, is a pth root

of unity. The constant term is then p. Replacing X by X1'n - 1 yields the
equation for the general case, proving CU 3. The other properties are obvious.
We may rewrite these relations to fit the formalism of distributions as
follows. Let a E (Q/ZY1') and a :I: O. Define
Let V = 81'/ ±p.<1') be the factor group of cyclotomic p-units by roots of unity.
Theorem 3.1. The association of (Q/Z)<1') -+ V given by
a 1-+ ga (mod roots of unity)
satisfies the distribution relations except at O.
The theorem means that for a :I: 0 we have
and is obvious in the light of CU 2.
Let Vn = 81'.10/ ± p.1''' be the factor group by roots of unity of p-units at

level ~n, i.e., generated by the roots of unity, and the elements, - 1 where
, is a pnth root of unity :I: 1. Whether we take , to be primitive or not to
generate Vn is immaterial since the distribution relation shows that we get
all of them from the primitive ones.
157
Let <§;t = Gal(Q{Jlp")/Q) mod U -1' The next theorem is due to Bass [Ba].
Theorem 3.2. The group <§;t operates simply transitively on the primitive
elements of Vn, and the induced homomorphism
is an isomorphism.
Proof The homomorphism is obviously surjective. It is injective because
Z[<§;t] is torsion free, andthe ranks of the two groups are equal. This proves
the theorem.
Theorem 3.3. The factor group Vm/Vnfor m ;::: n has no torsion.

Proof The embedding of Vn into Vm corresponds to the embedding of
group rings
Z[<§;t] ~ Z[<§~]
which sends an element Uc on the element L Ub, where the sum is taken over
Ub E<§~ (c), the set of elements in <§~ which project on U c under the canonical
map
If an element
2: 2: k(b)Ub' k(b) E Z,
c be~!(c)
is a torsion element with respect to Z[<§;t], then all the coefficients k(b) for
b E <§~(c) must be equal to each other, and hence the element already lies
in Z[<§;t], as was to be shown.
Analogues of Theorem 3.2 and 3.3 in the modular case are proved in the
Kubert-Lang series [KL 2, 3, 4, 5]. In that case, it is also shown that there are
no units except the modular ones. Here in the cyclotomic case, say for the
p-primary component, it is the Vandiver conjecture whether the factor group
E/~ is without p-torsion.
For the rest of this section, it is convenient to use ~n to denote the proper
group of cyclotomic units, i.e., the group of units of the form
rca.,
where rc = , - 1, , is a primitive pnth root of unity, and ct = LbeZ(p")' k(b)Ub

is an element of Z[<§n]o of degree 0, i.e.,
2:k(b) = 0.
158
Theorem 3.4. Let p be odd, and let c be a primitive root mod p2. Then En
is generated over Z[~n]o by the element
(e _ 1
v = <1enln = - - ,
(- 1
Proof We write an element IX of degree 0 in the form
IX = 2: k(b)(<1b - 1),
and observe that <1b - 1 is divisible in the integral group ring by <1e - 1
because <1e is a generator of the cyclic group ~n. This proves the theorem.
For p = 2 one has an analogous result using for c an element == 1 mod 4

such that c generates 1 + 4Z 2 • The group ~n = Gal(Q(j.tpn)jQ) is not cyclic
but a pro~uct of a cyclic group of order 2 and ~;t, which is cyclic.
It is convenient t6 reformulate the above theorem by passing to ~;t.
Theorem 3.5. Let c be a generator of I + 4Z2 if p = 2, and a primitive

root mod p2 if p > 2. Let
. if (e - 1 . V.
Vn = lmage 0 y=-r In n,
and let vg be the subgroup of Vn represented by units (not just p-units).

Then we have an isomorphism
so V~ is free of rank 1 over Z[~;t ]0.
Proof Clear.
The composite case.

Let V be the group of cyclotomic units of all levels, modulo the group of roots
of unity. Then V is torsion free.
Let c = ( ... , Cp , ••• ) be a vector with a component Cp E Z: for each p
such that Cp is a primitive root mod p2 if p is odd, and C2 == 1 mod 4 and
generates 1 + 4Z 2 if P = 2. We have an associated automorphism <1e on the
full cyclotomic extension of Q.
Let a E Q/Z and a ¢ Z. We define
. . <1e(e21t1a - 1)
h(a) = Image III Vof 21tla
e - 1·
Then h is an ordinary distribution in the sense of Chapter 2, §8 if we define

h(O) = o. .
159
6; Kummer Theory over Cyclotomic Zp-extensions
Theorem 3.6. This distribution is the universal even ordinary distribution with
value 0 at 0, and values into abelian groups on which multiplication by 2 is
invertible.
Proof On (l/N)Z/Z the group generated by the image of h has rank
!IZ(N)*I - 1,
which according to Kubert's Theorem 9.1(iii) of Chapter 2 is the maximal
possible rank (the value 0 at 0 gives rise to the -1). The Kubert generators
in TN/ ± 1 must therefore be free generators, and the canonical map from,the
universal distribution to h must be an isomorphism.
The above is more or less Bass' theorem in a different formulation. (Also,

as Bass states it, there is some difficulty with 2-torsion.) The idea of interpret-
ing it in terms of the cyclotomic units forming a universal distribution is due
to Kubert-Lang [KL 3], where a similar result is proved for the modular
units. The essential step in the proof here is of course Kubert's theorem cited
above, combined with the independence of the units. In the cyclotomic case,
this comes back to the non-vanishing of the regulator, i.e., L(1, X) ¥- O. In the
modular case, see [KL 2] and [KL 5].
§4. The Iwasawa-Leopoldt Theorem and the Vandiver Conjecture

For simplicity throughout this section we assume that p is an odd prime. Also
throughout this section, we let:
K" = Q(P(Pl) and Ko = Q(/lp) ,

'§ = Gal(K",/Q),
'§n = Gal(Kn/Q),
'§;t = '§n mod 0' -1 as in the preceding section.
Rn = Z(pn)['§n]' and R = projective limit of Rn.
Remark. It is easy to see that
R ~ A[Go]
where A is the usual Iwasawa algebra.
h;t = class number of K;t.

We assume the Vandiver conjecture that h;t = (En: tS'n) is prime to p.
Qn = K",(V~/pn) = K",(g~~~n. By the Vandiver conjecture,
Qn = K",(Ei~~n).
Q = U Q n = K",(Ei/ P"') = K",(tff~/P"').

Gn = Gal(Qn/K;,) and G = Gal(Q/K",).
160
We shall now develop the theory of Gas R-module following the exposition
of [KL 9]. We use an upper minus sign to denote, as usual, the (-I)-eigen-
space. This applies for instance to R -, G -, etc.
Theorem 4.1. Assuming the Vandiver conjecture, we have G = G-, and G

is a I-dimensional free module over R - .
Proof. By the Vandiver conjecture and Theorem 3.3 we have
Let us abbreviate for simplicity
Then the Kummer theory pairing discussed in §l can be described more

explicitly as follows. From Theorem 3.2 we write an isomorphism
using formal linear combinations with coefficients in (l/p'n)Zp/Zp. We have a

model for Kummer duality, through the pairing
such that
This pairing induces a perfect duality
as follows immediately from the formula
where p is complex conjugation, or for that matter any element of C§'n'

Therefore we have an isomorphism of the Kummer pairing in terms of the
group rings,
161
and this isomorphism is compatible with the limiting process, which can be
represented by a diagram in terms of the group rings for m ~ n:
R;;' X
1 R+m
-;;r --7- J-lp"
P
1 r r
R;; x l.
p71 R+
n
--7- J-lpn
It is then clear that

G = lim Gn :::: lim R;; = R-,
as desired.
One would expect the units whose existence is conjectured by Stark [St]
to playa similar role over totally real fields.
Theorem 4.2. Let Cn = CI(Pl(Kn ) be the p-primary part of the ideal class
group of Kn and let C = projective limit of the Cn under the norm map.
Under the Vandiver conjecture, we have C = C-, and C- is cyclic as a
A-module. In fact, the maximal un ramified p-abelian extension of K" is
contained in Q, so we have a natural surjective map
G --7- Go :::: C,
the first map by restriction and the second by class field theory.
Proof The field diagram (once the theorem is proved) is as follows.
GiL} \
K",
Go
What we have to do is to show that the maximal p-abelian unramified exten-

sion of K", is in fact contained in Q. The rest of the theorem is then obvious
from Theorem 4.1. It will suffice to prove that a finite cyclic unramified
p-abelian extension of K", is contained in Q.
Let K",((X) be unramified, with some element (X such that (Xp! lies in K",.
We first show that we may select (X to be real. By Vandiver's conjecture, we
have G6 = 1 so Go = G c. Let a be a generator for Gal(K", ((X)j K",). Then
a(X = '(X for some pth root of unity,.

Let p be complex conjugation. Then pap-l = pap = a- 1 since G = G-.
Hence pap(X = ,-1(X, and therefore
aa = ,a,
162
so that u(aja.) = a/a.. Thus a/a. = b lies in K". But the norm of b from K"
to K:' is 1 (obvious), so by Hilbert's Theorem 90, there exists p E Koo such
that PIP = b. Then a.p is real, and Koo(a.) = K,,(a.P). This shows that we may
assume a. real.
For n sufficiently large, a.pt lies in Kn+, and Kn+(a.) is unramified over K;
because p is odd. Hence we have an ideal factorization
for some fractional ideal a in Kn+. The class of this ideal is principal by
Vandiver's conjecture. It is then immediate that
for some unit u, thereby concluding the proof.
In the next theorem we let
Theorem 4.3. (i) For m ~ n we have an injection
(ii) Under the Vandiver conjecture we have Hl(r m,m Em) = 0, so
is injective.
Proof Let a be an ideal representing an element of en, becoming principal
in Km, say a = (a.) with a. E Km. For any element u E r m,n we have ua = a.
Hence ua. is equal to a. times some unit. The association
a 1-+ cocycle class of (ua./a.)
is a homomorphism of Ker( en -i>- em) into HI of the units, which is immedi-

ately verified to be injective.
Assume now the Vandiver conjecture. Let em be the group of cyclotomic
units. Then Em/em has order prime to p, so
For simplicity let Wm = J1 p m+1. Then we have exact sequences
and
163
whence exact cohomology sequences
and
Since Hl(Z) is trivial and Hl(Vm) is trivial (by elementary facts of cohomology
of finite groups, and Theorem 3.2), it will now suffice to prove that Hl(Wm)
is trivial. By the theory of the Herbrand quotient (cf. for instance Chapter IX,
§I of my Algebraic Number Theory), the orders of Hl(Wm) and HO(Wm) are
equal. However,
where N m•n is the norI1l. Thus finally it suffices to prove that every p-power
roo~ of unity in Kn is the norm of an element in W m• Let , be a generator of
Wm • The elements of the Galois group are represented by p-adic integers of
the form
1 + xpn+l with x E Zjpm-nz.
Taking the norm yields
N m•n' = n
X
,1+xpn = ,pm-n
which is a primitive element in Wn and thus shows that HO(Wm) is trivial.

This concludes the proof of the theorem.
Let K be a number field and let K' be an abelian extension with Galois
group G. We assume that K and K' are stable under complex conjugation.
We say that the extension K' of K is odd (resp. even) if its Galois group is in
the (-I)-eigenspace (resp. the I-eigenspace) for complex conjugation.
Lemma. If IXpn + 1 E Kn+, then Kn(lX)jKn is an odd extension.

Proof Clear.
From the lemma, it follows that QE/Koo is an odd extension, because the
units are generated by real cyclotomic units and roots of unity.
Theorem 4.4. Under the Vandiver conjecture, the Kummer duality gives rise
to a compact discrete duality
and also
164
Proof By Theorems 2.2, 2.3 and Lemma 1 of §2 we know that Q = QA

and that
Gal(QjQ E ) is dual to C",.
Taking eigenspaces for complex conjugation yields the first assertion. As to

the second, we know from Theorem 4.2 that
and Q E is an odd extension of Qnr. Again considering the eigenspaces yields

the second assertion.
165
7
Iwasawa Theory of Local Units
Iwasawa [Iw 8], [Iw 10] developed a theory of local units analogous to the
global theory, taking projective limits, especially in the cyclotomic tower, and
getting the structure of this projective limit modulo the closure of the cyclo-
tomic units. He considers eigenspaces for the characters of Gal(KoJQ p ) where
Ko = Qi() with a primitive pth root of unity (. Since the cyclotomic units are
essentially real, we consider only even non-trivial characters. Then the eigen-
space is isomorphic to AJ(g), where g is a power series which is essentially the
p-adic L-function.
The first section deals with the classical Kummer-Takagi exponents at the
first level QiO, where ( is a primitive pth root of unity, p odd. This is used in
combination with Nakayama's lemma afterwards to get corresponding results
in the cyclotomic tower. Throughout this chapter we assume that p is odd.
Coates-Wiles [C-W 4] have extended this theory to the case of elliptic
curves with complex multiplication. In the process they have found sub-
stantial simplifications for Iwasawa's proofs, and the exposition of this
chapter is essentially due to them. Note especially their generalization of the
Kummer homomorphism to all levels-a key to the whole theory. Such a
homomorphism extends to other formal groups besides the multiplicative
group, and a quite general statement has also been given by Coleman [Col].
On the whole, this chapter may be viewed as giving a good introduction
to the theories of Coates-Wiles. I am much indebted to them for keeping me
up on their work.
§1. The Kummer-Takagi Exponents

Let ( be a primitive pth root of unity, where p is an odd prime. Let Ko =
QiO. We let 0, ~ be the integers and prime ideal of Ko respectively, and
166
let
n='-l.
Let Uo be the group of units == 1 mod ,p in Ko. We let Go = Gal(Ko/Qp), and
be the homomorphism such that
For simplicity of typography in this section we shall write % instead of %0'

LetfE Zp[[X]]. We recall the variables
T= 1 + X = eZ ,
and the differential operator
D = (1 + X)Dx = Dz = TD T •
This last equality holds only for rational functions of X (or T).
Let U E Uo so U == 1 mod,p. Let f be a power series == 1 mod (p, X) such
that
U = f(n).
We then say thatfis a power series associated with u. Such a power series is
well defined up to a multiple of the irreducible polynomial h(X) of n over Zp.
Let J, fl be associated with u, so f == fl mod h. Since fl is a unit power series,
there exists a power series g such that
Then
and
Dflf = Dfllfl + multiples of h and h'.

Since h is an Eisenstein polynomial, it follows that
Dk-l(Dflf)(O) [= Dk 10gf(O)] is well defined mod p for 1 :0; k :0; P - 2.
167
7. Iwasawa Theory of Local Units
We define the Kummer homomorphism for these values of k by
((lk(U) = Dk-I(Dflf)(O) modp.
It is indeed clear that
is a homomorphism. By the change of variables X = eZ - 1, the formula
Dk -I( Dflf)(O)
is also valid for f as function of Z, i.e., if we set
then
We now develop systematically certain properties of the Kummer homo-

morphism. These will be extended in the Coates-Wiles manner later to all
levels.
K 1. If h, f2 are associated with units Ul, U2, then hf2 is associated with
U1 U2. Iff is associated with U and a E Zp, then f(x)a is associated
with ua •
Proof The homomorphic property is clear, and has already been mentioned.
The statement for a E Zp follows from positive integers by continuity.
K 2. Iff is associated with u, then a power series associated with u(1 is
f«(1 + X)"«1) - 1).
Proof Ifu = fen) andf= 1 + ... , then
u(1 = f«l + n)>«<1) - 1).
So the property is obvious. Furthermore,
f«1 + X)'«tT) - 1) = f(e"«1)Z - 1).
The next property then follows from the chain rule in terms of the
variable Z.
K 3.
168
Let X be a character of Go. Let
eCl) =
P
~ 1 L i(a)a
aeGo
be the corresponding idempotent in the group algebra Zp[GoJ. Write X = x!1.

for some residue class oc mod p - 1. Let u == 1 mod p. Put
ue(X) = uCl)·
K 4(i). If k == oc mod p - 1 then <Pk(uCl)) = <piu).
K 4(ii). If k ';iE oc mod p - 1 then <Pk(uCl)) = o.

Proof By K 2 and K 3 we find
<Pk(Ue(X») = xk(eCl))<pk(U).
The property follows by orthogonality of characters.
The units
1 - nk for k = 1, 2, ...
generate Uo topologically. We shall be especially interested in the values of k

satisfying 1 :$ k :$ P - 2, and we shall orthogonalize these units with respect
to the characters of Go. We let
where we abbreviate
Lemma. We have 11k == 1 - n k mod n H !, for 1 :$ k :$ P - 2.

Proof We have
11k == n (1 - (C l)k)-X- mod n

a - lc(a) P- 1
== n (1 + x-k(a)(C l)k). a -
as was to be shown.
169
Theorem 1.1. Let 1 ~ k, j ~ p - 2.
(i) <Pk(1]k) = - k mod p.

(ii) <Pk(1]j) = 0 if k =? j.
Proof The second assertion is a special case of K 4(ii). As to the first,
An associated power series of 1 - n k isJ(X) = 1 - Xk, and
-k
f'IJ(X) = 1 _ Xk
Then
2: XVk
00
DflJ(X) = -k(l + X)
v=o
2: (e
00
= -ke z Z - l)vk
v=o
Hence
Dk-l(DJIJ)(O) = -k
as desired.
By the lemma, and a trivial recursion procedure, any unit == 1 mod .p has
a product expression
and the exponents tk are called the Kummer-Takagi exponents. They are well
defined modp.
Theorem 1.2. Let u be as above. Then
Proof Immediate, from K 4 and the fact that <Pk is a Zp-morphism.
The Kummer Generators

The rest of this section will not be needed, but is included for completeness
of reference, and as an introduction to [C-W 2], [C-W 4]. It is convenient to
170
phrase the results in terms of the Lubin-Tate formal groups, so for the rest
of this section, we assume that the reader is acquainted with the basic facts
of these groups as explained in §1 and §2 of the next chapter, as well as the
existence of the logarithm on such groups, as explained in §6 of the next
chapter.
Let A be a Lubin-Tate formal group over the -IJ-adic field K, and associated
prime 1t. We let B be the basic Lubin-Tate group associated with the
Frobenius polynomial
Let Wbe the local parameter on B, and Z the parameter on the additive group,
so we have
Z = lB(W) == Wmod Wq-l
by Lemma 2 of §6 in the next chapter.

Let
ga(W) = bo + b1W + ...
be a power series with coefficients in K, and let
be the power series obtained by putting gB(W) = gGa(la(W)). Then it is clear

that
bk = dk for k = 0, ... , q - 1.
Taking the logarithmic derivative, i.e., the operator
g~g'/g
for any power series g, we then obtain:
Lemma. If
2: Ck,BWk-l
00
g~/ga(W) =
k=l
2: Ck,GaZk- 1
00
g;;a/gGa(Z) =
k=l
then
Ck,B = Ck,G a for k = 1, ... , q - 2.
171
Proof This is trivial from the chain rule, writing
Let Wo be an element of E" such that
wg- 1 + 1t = O.
The field
is tamely ramified, with different .pg-2.

Let U be a unit in Ko, and let g E o[[W]] be a power series such that
u = g(wo).
If glis another such power series, then
g(W) = gl(W)h(W)
where heW) is the irreducible polynomial of Wo over K. From this it is

immediate that
g'/g(wo) is well defined modulo .pg-2, and lies in o.
In particular, if we write
2: c/cw~-\
00
g'/g(wo) = with C/c E 0,

/C=l
then C/c is well defined mod .p for 1 .:=:; k .:=:; q - 2. We define
«J/c(U) = C/c for 1 .:=:; k .:=:; q - 2.
Then it is clear that
is a homomorphism, which we shall call the Kummer homomorphism of degree

k. We shall determine the value of this homomorphism in special interesting
cases.
The units
1- w~, with k = 1,2, ...

form a topological system of generators for the units == 1 mod .po in 00' Let
Go = Gal(Ko/K), so Go ;:::; Ilq-l'
172
There is a character X of Go into f1q -1 such that
awo = x(a)wo.
To avoid technical complications, we now assume that K = Qp so that

o = Zp. We let
be the idempotent in the group ring Zp[Gol for the character Xk. If u is a
unit ~ 1 mod we' we can define
in the obvious manner. We pick a sequence of integers mE Z approaching t

p-adically, and the ordinary powers urn approach a limit, which is by
definition ut•
We orthogonalize a basis for the units. We let
17k = (1 - W~)"k, for k = 1, .. . ,p - 1.
Then:
(i) 17k == 1 - w~ mod ~~+1.
(ii) a17k = 17~k«1), i.e., 17k E Uo(k), where Uo(k) is the xk-eigenspace of Uo, and
Uo are the units == 1 mod ~o in K o•
The second statement is obvious by the standard properties of the idem-
potent ek. For the first, we simply expand the product
17k == n (1 - w~)-x-k(a)a mod w~+1
== n (1 - x(a)kw~)-x-k(a)
== (1 + W~)P-1
== 1 - w~
as was to be shown.
Theorem 1.3. Let j, k = 1, ... , p - 2. Then:
(i) ({Jk(17J) = 0 if k # j.
(ii) ({Jk(17k) = -k modp.
Ifu is a unit == 1 mod ~o and
173
then
1
Ik == -" ({Jk(U) modp.
Proof Taking the logarithmic derivative formally, we have:
d10g
- - 11k -_ -
dwO
d10-
dwO
g
n (1 -
a
Xk( cr) Wok)-xk(a)
_ kXk(cr)Wk -1
_ "-k(cr)
-7 -x 0
1 - Xk(cr)W~
L L Xki(cr)W~(f+1)-1.
<Xl
= k
i=O a
Forj = 0 we get a term k(p - 1)w~-1 == _kW~-1 mod wg- 2 • This shows that
((Jil'fk) = -k. On the other hand, if k(j + 1) - 1 s p - 3, we have
k(j + 1) S P - 2 so kj s p - 3.
Then
xki is not trivial,
so the orthogonality relations show that the coefficient of the corresponding
power of Wo is O. This proves (i) and (ii). The last assertion then follows from
the homomorphic property of the map ({Jk, thus proving the theorem.
Let d = {at} be a finite family of integers prime to p, and let.AI = {ni}

be a finite family of integers satisfying
Let
Then u is a cyclotomic unit, and u == 1 mod -Po.
Theorem 1.4. Let
Then
1k = i2 Bk L nta~ mod p.
where Bk is the Bernoulli number.
174
§2. Projective Limit of the Unit Groups
Proof Let A be the formal multiplicative group, B the special Lubin-

Tate group associated with it. The power series
corresponds to the power series gB(W) such that
where' is a pth root of unity. Directly from the definition of the Bernoulli
numbers,
Z oo ZIc
ez _ 1 = 2: Bk -k,
Ic=O •
it follows trivially that
g'aa/gGa(Z) = a + Ic~O ~! BIc ak ZIc-l.

Since the operation g r+ g' /g sends multiplication to addition, the theorem
follows from Lemma 1 of §6 in the next chapter, and Theorem 1.3.

Let:
Kn = Qp(Wn), Wn = {tpn+l.
On, i:'n = integers and maximal ideal in Kn respectively.
Un = units == 1 mod i:'n in Kn.
U~ = units whose norms to Qp are equal to 1
= units which are infinitely divisible in the projective system of
units under the norm maps N m,n with m ~ n.
We have given two conditions describing U~, and it is easy to prove that they
are equivalent. Indeed, we have the formula for the norm residue symbol:
where N n is the norm from Kn to Qp. If Nnu = 1 then u is a norm from Km

for every m. Conversely, if the left-hand side is 1 for all m, then
Nnu == 1 modpm
for all positive integers m, so Nnu = 1. This proves the equivalence.
Let:
Gn = Gal(Kn/Qp) and Goo = Gal(Koo/Qp).
rn = Gal(Kn/Ko) and r = lim rn = Gal(KoojKo).
175
Note that Go operates on K n, Un, U~ and
Gn ~ rn X Go, while G'" ~ r x Go.

Let'l' be a topological generator of r. Then P" = Gal(Koo/Kn ) is generated
by')"'".
We have an exact sequence of Galois modules
1- U~ _ Un Nt» subgroup of Z:-1.
From this sequence we conclude that for each character X of Go and X =F 1,
In the next lemma, by rankzp we mean (as usual) the rank of a module modulo
torsion over Zp.
Lemma 1. (i) rankzp Un(x) = pn.

(ii) If X =F 1, Xo then
Un(x) ~ Z':").
Proof The integers On contain a free submodule over the group ring
and for large r, 1 + pTon is Galois-isomorphic to the above submodule under

the exponential map, and is contained in Un with finite index, so the first
part of the lemma is clear.
For the second part, the only torsion in Un consists of the roots of unity
Wn , which is a xo-eigenspace. Hence for X =F 1, Xo we have an isomorphism
as desired.
We consider the groups Un as forming a projective system under the norm

maps, and we let
U = lim Un
+-
be the projective limit. Then from the definition of U~ we see that also
U = lim U~.
+-
176
Note that U is a topological, compact Zp-module, and also aA-module, where
If U is an element of U, then we view U as a vector
U = ( ... , Un,"')
with components Un E Un such that Nm.nu m = Un for m ;::: n, and we also write
Lemma 2. U has no Zp-torsion.

Proof Otherwise there exists a fixed power pT and an element U = lim Un
such tha,t uf = I for all n. Then Un is a root of unity, and if Un =1= 1 for some n,
then the order of Um becomes arbitrarily large as m becomes large, which is
impossible.
Theorem 2.1. For each character X =1= 1, Xo of Go there is a A-isomorphism
U(x) ~ A.
In other words, U(x) is free of dimension 1 over A.
The proof will occupy the rest of this section, and will result from a
sequence of lemmas. A "natural" basis element for U(x) over A will be given
in the next section.
We shall apply Galois and class field theory in a manner similar to the
global case. For simplicity of notation, if X is a r-module, we let:
X(n) = Xj(ypn - 1)X and x(n) = fixed elements under i,n.
For simplicity of notation, throughout this section, denote by Kit b the

maximal p-abelian extension of' K n , and similarly let K~ be the maximal
p-abelian extension of Koo. We have a tower of fields
Recall that r operates by conjugation on Gal(K':,bJKoo).

Lemma 3. If X =1= 1 then we have isomorphisms
Gal(K:bjKoo)(x) ~ Gal(K':,bJKoo\n)(x)
~ Gal(Kit b jKn)(x).
177
Proof This is clear from the fact that K:b is the maximal abelian extension
of Kn contained in K:,b, together with the exact sequence
together with the fact that the last term is ~ Zp.
The Galois group Gal(K:b/Kn) is isomorphic by class field theory with the
completion of K:under the topology of subgroups of finite index. There is a
topological isomorphism as abelian groups
Given a choice of prime element 1t in KfI> the isomorphism has the form
The completion of K: in the topology of subgroups offinite index is therefore
as abelian groups (not Galois modules), where
Z = rI ZI (product taken over all primes I).

I
On the other hand we have an exact sequence of Galois modules
Since Go operates trivially on Z, for each X =f: 1 of Go we have an isomorphism
The isomorphism of local class field theory
K:/PP-l ~ Gal(K:b/Kn)
preserves the r and Go structures of both groups. Passing to the projective
limit over n, it follows from the previous isomorphism that
whence we obtain the next theorem from Lemma 3.
Theorem 2.2. For X =f: 1 we have an isomorphism
178
§3. A Basis for UuJ over A
Lemma 4. Let M be a finitely generated A-module such that
M/(yJ,n - I)M
is free over Zp of rank pn for all n. Then M is quasi-isomorphic to A.

Proof Obvious from the structure theorem in Chapter 5.
The lemma is applied to the unit groups, using Lemma 1 and Theorem 2.2.
We therefore conclude that there is an exact sequence of A-modules
0-+ A -+ U(x) -+A -+ B-+ 0
where A, B are finite. Since U has no Zp-torsion by Lemma 2, it follows that

A = O. The next lemma will conclude the proof.
Lemma 5. In the exact sequence, we have B = O,/or X # 1, "0'
Proof From the exact sequence
o-+ U(x) -+ A -+ B -+ 0
we get the exact (cohomology) sequence
[This is no big deal in the present instance. The last map is obtained by taking
an element b E B(n), lifting back to any c E A, and sending c r+ (ypn - 1)c. This
is well defined in UCx)/(ypn - I)UCx), and the sequence is trivially verified to
be exact.]
Trivially A(n) = O. Hence we obtain an injection
o-+ B(n) -+ U(x)(n) ~ UnCx) by Theorem 2.2.
But UnCx) has no torsion by Lemma l(ii). Hence
B(n) = 0 for all n.
Since B is finite, this implies that B = 0, and proves Theorem 2.1.
§3. A Basis for UCx) over A

For each n ;::: 0 we let Wn = f1.pn+l, and we fix a family of primitive pn+lth
roots of unity Cn E Wn such that
We let
Xn = Cn - 1.
179
The notation remains that of the preceding section. If (j E G ctJ then there is
an isomorphism
such that for all n,
As before, we let
be the corresponding isomorphism at the first level. If X is a character of Go,

with values in !lp-l, then
X= x~
for some k determined mod p - 1.

Given X # 1, Xo we shall construct an element ~ E U such that the element
is a basis of U(x) over A. It is natural to construct ~n of the form
where b E Z;, and w is the Teichmuller character. We have divided by web)

so that ~n == 1 mod ,).In' For each n :;:>: 1 we want that
But Xn is a root of (1 + X)P = Cn-l so the equation for b - Xn over K n - 1 is
(1 +b- y)P - Cn-l = 0,
and from the constant term we see that
Thus
{l + b)P = 1 + b, so (1 + b)P-l = 1.
Therefore we select any
b = A-I, with any A # 1 and AE !lp-l
180
§3. A Basis for Uu) over A
to get the desired b. A choice of A determines such en and we write
e~) instead of en
if we wish to emphasize the dependence on A. Since
w(b)1' = w(b),
it follows that for n ~ 1,
and so the elements en form a projective system in U.

Lemma 1. Given X = x~ 1= 1, Xo there exists A E !lp-l such that if we let
b = A-I, and
then:
(ii) eoU) = eU)· eo generates UoU) over Zp.

Proof We shall check below that for a suitable choice of A (depending
on k) the Kummer-Takagi exponent given by Theorem 1.2 is ¢. 0 mod p.
Then eoU) generates Uo(X)/UoU)P, and hence generates UoU) over Zp by
Nakayama's lemma.
Now for the computation of the Kummer-Takagi exponents, we need only
compute IPk(eO) by K 4. We have
b
eo = web) (1 - xo/b).
The associated power series is
b
J(X) = web) (1 - X/b).
Then
J'/J(X) = -~ 1 _IX/b = -~ v~ (X/bY,

We want to prove that
Dkd(DflJ) ¢. 0 mod p.
181
We have
, X+ 1 1
Dflf= (1 + X)f If(X) = - - = 1
X-b
+- -
T-l .
But D = TD T • Hence it will suffice to prove that for 2 :$; k :$; p - 2 we have
By induction, it is immediately shown that
( TD )m(_I_) = + n m - 1 + Pm(T, 1)
T T - 1 - (T - l)m+l
where P m(T, 1) is a polynomial in 1 of degree :$; m - 2, with coefficients in

Z[T]. Hence
The numerator l k - 2 + Pk-1(l, 1) is a polynomial in 1 of degree:$; p - 4.

It is clearly not identically zero mod p, and so it has at most p - 4 roots mod p.
We can therefore choose 1 # 1 in !1p-l such that A. is not a root of the
polynomial mod p. This completes the proof.
Theorem 3.1. Let X # 1, Xo. We can choose A. E !1p-l such that the element
generates U(x) over A, i.e.,
U(x) = A· e(x),
and such an element is a free basis for U(x) over A.

Proof We know from Theorem 2.2 that
Uo(x) = U(X)/(y - l)U(x),

e
and so by Lemma 1, e(x)· generates U(x) mod rnA' U(X). By Nakayama's
lemma, it follows that e(X) generates U(x) over A. Since U(X) ~ A by Theorem
2.1, such a generator is also a basis, thereby proving the theorem.
§4. The Coates-Wiles Homomorphism

In this section we give the extension of the Kummer homomorphism to all
levels, based on a refinement of the associated power series.
182
§4. The Coates-Wiles Homomorphism
Theorem 4.1. To every element u E U there is a unique power series f E Zp[[ X]]
such that
This power series satisfies fiX) == 1 mod (p, X), and the map
is a homomorphism of U into the multiplicative group of power series

== 1 mod (p, X).
We first note that uniqueness is obvious since a power series has only a
finite number of zeros (Weierstrass preparation theorem).
The proof of existence will proceed via several steps, which also develop
systematically other properties of these series. First:
1
cwo. fiX) = -
web)
(b - X)
is the power series associated with our element ~. Indeed,
Next we note two formal properties of the power series fu which is called the
associated power series to u.
CW 1. If a E Zp then the power series associated with ua is fu(x)a.

Proof This is first obvious when a is a positive integer, and is then true
for all a E Zp by continuity.
CW 2. If h is associated with u, and (J E Goo, then there is a power series

associated with u", namely
fua(X) = h«(1 + X)>«a) - 1).
so the property is obvious from the definitions.
We are now ready to prove the theorem, i.e., we must show that every u
has an associated power series. The two properties CW 1 and CW 2 show
that the set of elements in U having an associated power series is a A[GoJ-
submodule of U, and contaips ~. So it contains A[GoK In particular, taking
183
'7. Iwasawa Theory of Local Units
Theorem 3.1 into account, we have already shown that it contains U(yJ for
all X of 1, Xo. This is enough for the applications we have in mind.
For the remaining eigenspaces of X = 1 or Xo, the element does not e
suffice to generate these spaces, and one must show how to find an associated
power series for additional elements generating these spaces. This is not too
difficult and will be left to the reader, especially since a generalization of the
associated power series to all Lubin-Tate formal groups has been given by
Coleman [Col].
Let as usual
D = (1 + X)Dx = Dz .
Since fu ffu has coefficients in Zp, it is clear that

l
For each integer k ;::: 1 we define the Coates-Wiles homomorphism ({Jk on

Uby
We see that ({Jk: U -l>- Zp maps U into Zp by the preceding remark.
CW 3. For a E Goo, ((Jk(U") = x(a)k({Jk(u),

Proof If Uf--+ u" then CW 2 gives the power series associated with u", and
the assertion is then obvious by the chain rule applied to
CW 4. Let X = x~ where ('f. is a residue class mod p - 1.

(i) If k == ('f. mod p - 1, then ({Jk(U(x» = ({Jk(U),
(ii) If k ;t= ('f. mod p - 1, then ((Jk(U(x» = O.

Proof Let e(x) be the idempotent associated with X. Then by CW 1 and
CW 3 we find
The property follows by orthogonality of characters.
CW 5. For UE U and g E A = Zp[[X]], we have
((Jk(g'U) = g(X(y)k - 1)({Jk(u),

Proof The assertion is true when g(X) = 1, and when g(X) = 1 + X =Y
by CW 3. Thus it is true when g(X) = X, i.e.,
184
§4.. The Coates-Wiles Homomorphism
The property follows for arbitrary polynomials by induction, and arbitrary g

by continuity.
Theorem 4.2. Given a congruence class r:J. mod p - 1, there exists a power
series ha such that for any k == r:J. modp - 1, we have
If r:J. is even :f= 0 mod p - 1, then we can choose A such that
is a unit, and ha is a unit in Z1'[[X]].

Proof Let
f1(X) = D log~(X) = (1 + X)f~/~(X).
Then by Meas 6 of Chapter 4,
CfJk(e) = Dk-IJ1(O) = f Z"

X k - 1 d/1h(x).
Then a computation shows that
for some measure /1. By decomposing the integral over co sets of /11'-1 in Z:
we can write
fZj,
w(a)a<a)S d/1(a) = 2:
rel'''_l
f
l+1'Z"
as d/1r(a)
where /1r is a measure with support in 1 + pZ1'. By Example 2 of Chapter 4,

§1, we conclude that for each r there is a power series J,. such that
fl+1'Z"
as d/1r(a) = J,.(X(y)8 - 1),
and we let
h= 2:
rel',,_l
J,..
185
Then for k == CI. modp - 1, we get
h(x(y)k - 1) = L. p
w(ay<a)k dJ1(a)
This concludes the proof of the existence of h. There remains to show that
({)k(eo.,» is a unit-it is then trivial that h is a unit power series. But this last
property is clear from Lemma 1 of §3, as was to be shown.
§5. The Closure of the Cyclotomic Units

Let rffn be the group of cyclotomic units, i.e., the group generated by
Wn = ± J1 p n+1 and elements
where ( is primitive pn + lth root of unity. Let:
Vn = closure of rffn n Un in Un = ;jn,l

V = lim Vn•
+-
If we wish to preserve the rff-notation, then we may write
V = lim ;jn,l = ;j00,1'

+-
The group Vn is a Zp[Gol-module, and V is a A-module. Since rffnl Wn comes

from the real subfield, x-eigenspaces occur only for even characters X of Go.
Before analyzing the projective limit of Vn , we recall in the p-adic context
some facts about finite levels. In the global fields, we have an isomorphism
where the index on the right indicates the augmentation ideal. Cf. Theorem 3.2
of Chapter 6. Since the cyclotomic units are independent over Zp by the
non-vanishing of the p-adic regulator, we obtain a Gn-isomorphism
Hence for each even X i= 1,
186
Let c E Z; be a primitive root mod p2, and let

Cn - 1
Vn = w(c) C:;' _ l'
Then Vn == 1 mod lJn. Furthermore for any even character X # 1 the element
lies in tfn' so in Vn(X). (The root of unity w(c) disappears when we project on
the eigenspace for X.) Note that the elements Vn form a compatible system in
the cyclotomic tower, that is
We let
Then
Vex) = lim vn(x) = lim v~(X).
Theorem 5.1. For each even character X # 1 we have
and hence
Vex) = A· vex).
Proof Immediate from Theorem 3.2 of Chapter 6.
We note that the power series associated with the element v is
x
heX) = w(c) (1 + xy - 1
Let X be an even character of Go, and X # 1. Let = e(l) be the con- e

structed element of U such that e(x) is a basis for U(x) over A. We have
U(x) = A· e(x) and Vex) = A . vex)·

Let us write
187
Theorem 5.2. Let X = x~ be an even character =I=- 1. We have an isomorphism
U(y)/V(x) ~ A/gxA.
The power series gx (determined up to a unit in A) can be selected such that

it is equal to the power series g satisfying
for any even positive integer k such that k == IX mod p - 1.

Proof We have
q>iv) = q>1e(v(x)) byCW4

= gix(y)1e - l)q>Ie(~(x)) byCW5
= gix(y)1e - 1)q>Ie(~) byCW4
= g(x(y)1e - 1)(1 _ ple-l)-l,
where g = gxha., and ha. is the power series of Theorem 4.2. Since ha. is a unit
power series, g and gx generate the same Ae(x)-ideal. There remains to prove
But
, 1 + X c(1 + xy
(1 + X)f,,/f,,(X) = -X - (1 + xy - 1
eZ cecz
- eZ - 1 - eCz - 1
= Df"/f,,.
So for k ~ 2,
q>1e(v) = Die - l( Df"ff,,)(O)
1
= (1 - cle)"kBIe
by the definition of Bernoulli numbers. This concludes the proof.
The values of the power series g show that it is essentially the p-adic
L-function.
Remark. In this chapter we have proved a local statement which would be

immediate if one had the global Vandiver conjecture, as explained in Chapter 6,
§4. The corresponding global conjecture can also be formulated as follows.
188
Let Kn = Q(p,(P»), and let K':,b be the maximal p-abelian p-ramified

extension of Kn. Let X be any even character 1= 1. Then there is a quasi-
isomorphism
where gx is as in Theorem 5.2.
189
8
Lubin-Tate Theory
This chapter reproduces with little change the approach to local class field
theory given by Lubin-Tate [L-T]. Using special power series associated with
prime elements in a p-adic field, they construct maximal abelian totally
ramified extensions by means of torsion points on formal groups, thus
obtaining a merging of class field theory and Kummer theory by means of
these groups.
The theory applies in particular to the cyclotomic case. The p"th torsion
point on a suitable group will be seen to be the classical cyclotomic numbers
, - 1
where' is a p"th root of unity.
§1. Lubin-Tate Groups

Let 0 be a ring. By a formal group over 0 we mean a power series
F(X, Y) E o[[X, Y]]
in two variables satisfying the three conditions:
FG 1. F(X, Y) = X + Y(mod degree 2).
FG 2. F(X, F(Y,Z)) = F(F(X, Y), Z).
FG 3. F(X, Y) = F(Y, X).
Strictly speaking, our formal groups should be called commutative one-

parameter formal groups, but we won't deal with any others. The expression
190
mod degree 2 means modulo the power series of degree ;::: 2. Using the
associativity with Y = Z = 0 it follows at once that
F(X, Y) == X + Ymod XY,

i.e., F(X, 0) = X and F(O, Y) = Y.
It is an easy matter to show recursively that given a formal group as above,
there exists a unique power series "l(X) such that
"l(X) == - X (mod degree 2)

and
F(X, "l(X)) = F("l(X), X) = o.

If this could not be proved, we would assume it as an axiom. We leave the
proof as an exercise. For the more extensive foundations of formal groups in
any number of variables, cf. Frohlich [Fr].
Example. The formal multiplicative group Gm is defined by
F(X, Y) = X + Y + XY = (1 + X)(1 + Y) - 1.
If a is a positive integer, and [a] denotes "addition" on G m a times, then
[aleX) = (1 + x)a - 1.
If M is an algebra over 0 (always assumed commutative) and M is nilpotent

(in the sense that every element of M is nilpotent-some positive power of
the element is 0) then the formal group F defines an additive group law on the
set of elements of M, by the association
(x, y) I--? F(x, y)
for x, y in M. Instead of F(x, y) we would also write
F(x, y) = X +F y, or x [+ ] y, or x [+ ]F y.
The set of elements of M with this group law could be denoted by M F • On

the other hand, it is useful to use a slightly different notation. We view F as
defining a functor
We may also denote this functor by a letter like A (or AF if we wish to make
the reference to F explicit), and then denote
MF = A(M)
191
8. Lubin-Tate Theory
to be the set of points of A in M. As a set, it consists of the elements of M,

and it is also an additive group with the group law determined as above.
Suppose that 0 is a complete valuation ring, with quotient field K and
maximal ideal mK' We also write 0 = OK' Then m K = m is topologically
nilpotent, in the sense that arbitrarily large powers of an element tend to O.
For any positive integer k, m/mk is a nilpotent o-algebra, and A(m/mk) is a
group, as we saw. By continuity, it follows that A(m) is also a group. Addition
between elements x, y in m is again given by
(x, y) 1-+ F(x, y).
Let L be any algebraic extension with valuation ring 0L and maximal ideal
mL' Then we also have the completion OL if L is infinite over K, with maximal
ideal m L , and it is clear that A(mL ) can again be defined as group with the
group law given by the same formula as above.
By an endomorphism of the formal group F (or A F ), we mean a power
series I( X) such that
I(F(X, Y» = F(f(X),/(Y».
We say that I is defined over 0 if the coefficients of I lie in o. It is then clear

that such an endomorphism defines an endomorphism of A(m) by the
association
X 1-+ I(x) , for x in m.
Similarly, a homomorphism I of a formal group F into a formal group F' is a
power series such that
I(F(X, Y» = F'(f(X),f(Y».
This relation could also be written
I(X +F Y) = I(X) +r I( Y).

Such homomorphism induces a group homomorphism
A(m)....,.. A'(m),
where A'(m) is the group whose underlying set is m, and whose group law is
that determined by F'.
We shall be interested in a special kind of formal group. From now on,
we assume that OK is a discrete valuation ring, and we let n be a prime
element in mK' We assume that oK/mK is finite with q elements. We let:
~ = set of power series lEO [[ X]] such that
I(X) == nXmod degree 2
I(X) == xq mod n.
192
Example. The power series (polynomial) f(X) = xq + nX is an element

of ff,., actually its simplest element, which will be called the special or basic
Frobenius polynomial associated with n.
Example. Let
f(X) = (1 + X)P - 1 = XP + ... + px.

Thenf(X) is an element of~.
The elements of ff,. will be called the Frobenius power series determined
by n.
Theorem 1.1. To each Frobenius power series f in ff,. there exists a unique
formal group Ft (defined over 0) such that f is an endomorphism of Ft.
The formal group associated withf(X) = xq + nX in Theorem 1.1 will be

called the special or basic Lubin-Tate group associated with the. prime n.
The proof of this theorem will follow from a general lemma, as will the
fact that the formal group Ft then admits 0 in a natural way as a ring of
endomorphisms commuting withf
Lemma. Let f and g be Frobenius power series in ff,.. Let
be a linear form with coefficients aj i~ o. There exists a unique series

F(XI , ... , Xn) E O[[XI' ... , Xn]] such that
and
f(F(XI , ... , Xn)) = F(g(XI ), ... , g(Xn))·

Proof We abbreviate X = (Xl' ... , Xn) and g(X) = (g(Xl ), ... , g(Xn)).
We show by induction on r that the congruences
FrCX) == L(X) mod deg 2 and f(Fr(X)) == Fr(g(X)) mod deg r + 1
have a solution Fr(X) in o[X] which is unique mod deg r + 1. This is true
for r = 1 with Fl(X) = L(X). Suppose it true for r ~ 1. We let
Fr+l = Fr + Hr+l
193
where Hr+l is a homogeneous polynomial of degree r + 1 with coefficients

in o. We have:
f(Fr+l(X» == f(F,(X» + nHr+1(X) mod deg r + 2

Fr+l(g(X» == Fr(g(X» + n r+1H r+1(X) mod deg r + 2.
To satisfy the desired relation up to degree r + 1, we must take

H (X) = f(F,(X» - Fr(g(X» mod deg r + 2.
r+1 nr+1 _ n
The coefficients are in 0 because
f(Fr(X» - F,(g(X» == (Fr(X»q - Fr(Xq) == 0 (mod n).

It is then clear that
F(X) = lim Fr(X) E o[[X]]
is the desired unique solution satisfying the conditions of the lemma.
Addendum to the lemma. The completeness of 0 was not assumed or used

in the proof Furthermore, the proof shows that F, is the only power series
with coefficients in an extension field of K satisfying the conditions of the
lemma.
Theorem 1.1 is immediate from the lemma. Indeed, F, is the unique power
series F(X, Y) such that
F(X, Y) == X + Ymod deg 2,

and
f(F(X, Y» = F(f(X), f( Y».
The other two formal group properties are seen to be satisfied by showing
that the left-hand side and right-hand side of FG 2 (resp. FG 3) are each the
unique solution of a system of conditions as in the lemma.
We call F, the Lubin-Tate formal group associated withf If we want to
use the other notation, we also write it A(f), or simply A if the reference to
f is clear from the context.
We shall now see that F t admits 0 as a ring of endomorphisms in a natural
way. We prove slightly more. For each pair of elements f, g E ~ and a E 0,
we let a"g or [a]t,g denote the unique solution of
a"g(X) == aXmod deg 2

fo a"g = a"g g.
0
We write a, or [a], instead of at" for simplicity.
194
Theorem 1.2. The association a 1-+ af is an injective ring homomorphism of

o into End(Ff), such that
More generally, the association a 1-+ af,g is an injective additive homo-

morphism ofo into Hom(Fg, F,), satisfying the composition rule
and
[a + bJ"g(X) = F,(a"g(X), b"g(X»

= a"g(X) +F/ b"g(X).
Proof In each case, one checks immediately that both the left-hand side
and right-hand side of the desired identity are solutions of the type given in
the Lemina, whose solution is unique.
It is clear that ifJ, g E:F.. then the element 1,,g is an isomorphism between
Fg and F,. Thus the isomorphism class of F, is uniquely determined by n.
Furthermore, from Theorem 1.2, we may also view F, as an o-module via
the operation a, for a E 0, and the above isomorphism is obviously an
o-isomorphism.
As a matter of notation, we shall use the three notations
to denote the same power series. After a while, the polynomial f in :F..
becomes mostly irrelevant, and we think in terms of the group law A. Thus
aA or [aJ when A is fixed become more satisfying to work with.
Let L be the completion of an algebraic extension of K. Then we may view
A(mL ) as an o-module in the obvious way. The operation of 0 on A(mL ) is
given by
Of course, if L is finite over K, then L is equal to its own completion. By

functoriality, we also see that the formal isomorphisms 1"g induce isomor-
phisms .
In view of this isomorphism, it is often convenient to omit f or g from the

notation and write [aJ(x) for the operation of 0 on A(md for x in mL'
Let L be a Galois extension, with Galois group Gover K. The operations
of elements of G on L extend to the completion by continuity, so we may
195
replace L by its completion. Since the power series aI,g, au, Ft have coeffi-
cients in 0, it is then clear that the operations which they define on mL
commute with the action of G on m L •
§2. Formal p-adic Multiplication

Again we let OK be a discrete valuation ring with quotient field K, which we
assume complete. We let m K be the maximal ideal, and let q = card oK/m K
be finite, a power of the prime p. We let/be a Frobenius power series over 0,
associated with the prime element n in m K , and we let Ft or A = A(f) be
the corresponding Lubin-Tate group.
For each a E 0, we let Aa(f) be the set of elements x in the maximal ideal
mKa. of the algebraic closure of K such that
In other words, Aa is the kernel of [aJ. If a is a unit, then Ker at = 0, so we

are really concerned with A"n for positive integers n. Of course, Aa depends
on / so we should write Aa( /). However, if g is another Frobenius power
°
series in associated with the same n, then the isomorphism I"g induces an
isomorphism
which commutes with Galois isomorphisms. Further, if 0' is an automorphism

of Ka over K, then
so Aa(f) is a Galois module, and the extension
K(Aa(f» over K
is independent of the choice of/in~. We shall see in a moment that it is a

separable extension, whence it is a Galois extension, and is finite for a 1=
because a non-zero power series has only a finite number ofzeros (Weierstrass
°
preparation, or more naively, use the power series X q + nX in ~).
Consider the case n = I, so consider K(A,,). Then
K(A,,) = K(x)
where x is a root of xq + nX = 0, x 1= 0, or in other words, x is a root of
xq-l + n = O.
196
Thus K(An) is a Kummer extension (since the (q - l)th roots of unity are
in K), with abelian Galois group, cyclic of order q - I, and totally ramified
over n.
Let x E An and x =1= O. The map
gives a homomorphism of 0 into An, whose kernel is obviously no. Since

An has q elements, it follows that
An ~ o/no
as o-module. In particular, Endn An ~ o/no, and
Autn An ~ (o/no)*.
We have a representation
x: Go = Gal(K(An)/K) -7 Autn A" ~ (o/no)*.
Since Go and (o/no)* have the same cardinality, namely q - I~ it follows that
this representation is an isomorphism.
We have similar results in the nn-tower.
Theorem 2.1. (i) The group Ann is a free I-dimensional module over o/nno.
(ii) K(A"n) is abelian over K, totally ramified, and we have a natural
isomorphism
x: Gal(K(A"n)/K) ~ (o/nno)*.
Proof Let be a sequence with Xk E A,h such that
(Xl> X2, ... , xn) Xl =1= 0
and ntCxk) = Xk-l' Without loss of generality we may assume
f(X) = xq + nX.
For k > I we see that Xk is a root of
Relatively to the field K(A"n-l) this is an Eisenstein equation, and so we have

shown inductively that K(A"n) is totally ramified. Since Ann is stable under
the Galois action, and since the equation
is separable, it follows that K(A"n)/K is Galois. As before, we get a repre-

sentation of the Galois group in Autn Ann. The map
197
8; Lubin-Tate Theory
induces an injection of 0/1C"0 into A"n, whence an o-isomorphism by counting

cardinalities, and it follows as for n = I that we have an isomorphism as in
(ii), thus proving the theorem.
Passing to the limit, we may form the projective limit T,,(A), consisting of
all infinite vectors
such that 1Clxn ) = Xn -1 and 1Cf(XO) = o. It is then immediate that T,,(A) is

a free I-dimensional module over o.
Let
Kn = K(A "n +1), Koo = U Kn·
Then Koo is an abelian, totally ramified extension of K, and
x: Gal(Koo/K) ~ 0*
in the natural way. If u is a unit in 0*, then we have a corresponding element

of the Galois group, denoted by uu, which is such that
Uu = [U]f-1
in the representation on T,,(A(f». If we wish to omit the reference to f, we

simply write [u]. Thus on a vector as above, we have
It is also convenient to have a notation for the representation of the Galois

group in 0*. We let
be this representation, where GK = Gal(Ka./K), such that
I ax = [x(u)lr(x) for x E A(1t).
Example. We shall now give the standard example with the formal
multiplicative group
F(X, Y) = X + Y + XY.
Over the p-adic integers 0 = Zp we have the Frobenius series given by
f(X) := (1 + X)P - I = XP + ... + pX

198
associated with the prime p. Let A be the corresponding Lubin-Tate formal

group. Then in fact A is defined by the power series
F,(X, Y) = F(X, Y),
i.e., A is the formal multiplicative group. Then Apn consists of those elements
in the maximal ideal of the algebraic closure satisfying the equation
(1 + Xr - 1 = 0
and these elements are none other than
, - 1,
where, is a-p"th root of unity.
Theorem 2.2. The prime n is a norm/rom every extension K(Ann).

Proof Consider first the bottom level of the tower K(An) over K, obtained
from the equation
xq-l + n = O.
Let Ct be a root. Then
(_I)q-l N(Ct) = n.
If q is odd then n is the norm of Ct. If q is even then q - 1 is odd, the degree
[K(An) : K] is odd, and - 1 itself is a norm. Hence n is the norm of - Ct.
This proves the theorem in case n = 1. For the proof in general, let n = XQ,
and let
be such that
Thus Xl is an element of An' Xl :f: 0, and X,,_l is a norm of ±x" from the
field K(x,,) over K(X"_l)' The argument is similar and equally trivial, as
desired.
Theorem 2.3. Let B be the special Lubin-Tate group associated with the
prime n and the Frobenius polynomial xq + nX. Let' E !iq-l' Then:
(i)
199
(ii) If F(X, Y) is the group law for B, and
F(X, Y) = X+ Y I ajjXi yf,
then Ojj = 0 unless i + j == 1 (mod p - 1).

Proof Let Xn be a generator of B"n+l such that
Since Xo is a root of Xq-l + re = 0 it follows by a trivial recursion that the

irreducible polynomial for Xn over K is a polynomial in Xq-l. Therefore we
can find an automorphism
such that O"nXn = 'xn. Since elements of the Galois group commute with [re],
there exists an element 0" E Gal(K,,,/K) such that
By Theorem 2.1 there exists a E oi such that O"Xn = [a](xn) for all n. Since
O"q-l = 1, it follows that a is a (q - l)th root of unity. But also
so , - a == 0 mod X n • This is impossible since both " a are in J!q-l> unless

, = a, thereby proving (i).
Secondly, for every' E J!q-l we have
Then (ii) follows immediately.
§3. Changing the Prime

We shall now analyze what happens when going from one primere to another
prime re' = reu where u is a unit of OK' Since we have to refer to the primes,
we let
We let A' be the Lubin-Tate group associated with re', so
We let L be the completion of the maximal unramified extension of K, with

ring of integers 0L' We let Knr be the maximal unramified extension of K,
200
§3. Changing the Prime
and 0nr its valuation ring, with maximal ideal mnr • We let cp be the Frobenius
automorphism of K nr , extended by continuity to L.
Theorem 3.1. Let J, f' be Frobenius power series over 0, associated with the
primes n, n' respectively. Let e be a unit of 0L such that e"'/e = u. (Such
units exist.) Then there exists a unique isomorphism
defined over 0L which commutes with the operation ofo, that isfor all a in 0.
o 0 af = af' 0,
0
and such that

O(X) == eXmod deg 2.
This power series 0 satisfies
0'" = 00 UfO
Proof The existence of the unit e such that e"'/e = u is easily obtained by a
recursive procedure, and is left to the reader. We then construct a power series
O(X) = eX + ... to satisfy 0'" = 00 Uf as follows. Let Ol(X) = eX. Suppose
we have found Or(X) of degree r such that
Of (X) == Or(u,(X)) mod deg r + 1.
We wish to find some element b E Jh such that the series
satisfies the same congruence to one higher degree. We have:
Or+ leX) ,= Of (X) + b'" xr+l

Or+l 0 Uf(X) = Or 0 u,(X) + bUf(XY+1.
The condition on b is therefore that
b'" - bur+l = c,
where c is the (r + l)th coefficient of Or 0 u,(X) - Of (X). Write b = aeT+l.

The condition on b is equivalent with the condition
201
A recursive procedure, letting a,,+l = a" + xnn+1 shows that we can solve
for x at each step to make the equation valid mod n", whence solve the
equation for a in 0L. This concludes the construction of e.
We shall now see that e almost satisfies the other conditions of the theorem,
and that it is easy to adjust it to get these other conditions exactly. Let
g = eo n/ 0 e- 1.
It is obvious that eo Ff e- 1 commutes with eo n/ 0 e- 1 = g. We contend

0
that g is a Frobenius power series associated with n', and has coefficients in
° = OK. Once we have proved this contention, we then conclude that the
power series
has the properties characterizing FiX, Y) (it is obvious that this power series
is congruent to X + Y mod deg 2). The Lemma and addendum to Theorem 1.1
show that the power series is equal to FiX, Y). Similarly, we verify that
eo af e- 1 has the properties which characterize ag, so is equal to ago In this
0
manner, we have proved the theorem except for the fact that
is an isomorphism from F f to Fg. Replacing e by Il',g 0 e then concludes the

proof.
All that remains to be done is settle the contention. We have:
Also,
en/e- 1(X) = eUfnfe-1(X) = e<P(f(e- 1(X)))

== e<P(e- 1(X)Q) mod n
== e<P(e-<p(xq)) mod n
modn.
There remains only to prove that the coefficients of en;e -1 lie in 0, and it
suffices to show that they are fixed under the Frobenius automorphism cp.
We have:
which concludes the proof of the theorem.
202
§4. The Reciprocity Law
§4. The Reciprocity Law

Let Knr as before be the maximal unramified abelian extension of K. Local
class field theory would immediately show that the composite field
is the maximal abelian extension of K. On K("l we have a good model for the
Galois group given by the association
af-+ [a], aEo*,
and on Knr we have the Frobenius automorphism, which generates the

Galois group. We wish now to give an independent proof that the field K("l Knr
is independent of the choice of n, and that the structure of the Galois group
is in fact determined independently of that choice also.
Theorem 4.1. The field K("lKnr is independent ofn. Let a E K*. Write
for some unit u, and some integer m. Let r ,,(a) be the automorphism of
K("lKnr such that:
ria) = 0',. on K("l

ria) = <pm (<p = Frobenius) on Knr .
Then the association a f-+ ria) is independent of the choice of n.

Proof Let L be the completion of Knr as in the preceding section. Let A
be the Lubin-Tate formal groups associated with the prime n, and let A'
be associated with the prime n'. Since A and A' are isomorphic over L by
Theorem 3.1, it is clear that
However, Knr is algebraically closed in L. The two totally ramified extensions
become equal when lifted to L. By elementary field theory, they must be

equal as extensions of K nr , thus proving the first assertion in the theorem.
The set of prime elements n' generates the multiplicative group K*. To
prove the independence of ria) from the choice of n, it will therefore suffice
to prove that for all n',
r ,,(n') = r".(n').
203
These automorphisms coincide on Knr since they both give rise to the
Frobenius element. It will therefore suffice to prove that they coincide on
K(1<'l. We keep the notation of Theorem 3.1.' Write n' = un with some unit u.
Since r 1C,(n') = identity on K(1C'l, we are reduced to showing this same property
for rin').
Let f be a Frobenius power series associated with n. The field K(1t'l is
generated by the elements O(x) with x E A(1C l(f). Hence we are reduced to
showing that such elements are fixed by rin'). Indeed:
r 1C(n')e(x) = r iu)r in)e(x)

= r ,,(u)e'P(x)
= O'P(u, -l(X))
= O(x).
This concludes the proof of the theorem.
One may now use local class field theory to guarantee that K(1C lKnr is the
maximal abelian extension of K. Let
be the norm residue symbol mapping K* into Gal(Kab/ K) from local class
field theory. Then we find
r(a) = (a, Kab/K).
Indeed, both automorphisms induce the Frobenius automorphism on Knr>

and for any prime element n, both automorphisms induce the identity on
K(1C>, since n is a norm from every finite subextension of K(1C l by Theorem 2.2.
Since r(a) and (a, Kab/K) coincide on all prime elements, they coincide on
K*.
§5. The Kummer Pairing

It should be clear that the formalism offormal groups is completely analogous
to the classical formalism on the multiplicative group or the group of Witt
vectors. In a similar way, one can develop "Kummer theory" completely
analogously to the standard way (cf. for instance Algebra, Chapter VIII, §8),
or the way Witt did it in characteristic p for his vectors (Crelle 1935-1936).
The possibility of doing this in the context of Lubin-Tate groups was first
noted by Frohlich [Fr]. Of course, some new phenomena arise. Applications
to explicit reciprocity laws as in Coates-Wiles [C-W 2] and Wiles [Wi] will
be postponed to a later chapter.
Let A be a Lubin-Tate group associated with the prime n. We let
204
so Ko = K(A,J We let 0" be the ring of integral elements in K" and let V"
be its maximal ideal. We write A(V,,) as usual for the set of elements V" with
the group law defined by A. We define a pairing
A(V,,) x K: _ A"n+l.
Let x E A(V,,), let t be an element of A(V",) such that [n"+l]t = x, and let
a E K,,*. Note that actually t E A(V 2"+1). Define the symbol
where G'rz = (a, K:b/K,,) is the automorphism of K:b over K" arising from
local class field theory. Then it is clear that (x, a) lies in A"n+l and is inde-
pendent of the choice of t such that [n"+ l]t = x. We call it the local (Kummer)
symbol (relative to the formal group A and the multiplicative group). If we
want to specify A in the notation we write
<x, a)~.
Example. The formal multiplicative group. For p :; 1 mod V" and a E K~

we define the classical norm residue symbol
(p, a)" = <p - 1, a)~ + 1,

where A is the formal multiplicative group.
The local symbol trivially satisfies the following properties.
LS 1. It is oK-linear in x and multiplicative in a.
In particular, the symbol induces a pairing
and it is clear that in the pairing
A(V,,) x K: _ A"n+l
the kernel on the left is exactly [n"+l]A(V,,), because if x is not a [n,,+l]_
multiple in A(V,,), then its [n,,+l] root t generates a proper extension of K n,
so the Galois group operates non-trivially.
LS 2. If e: A - A is an isomorphism over 0 between two Lubin-Tate groups

I
associated with the same prime n, then
(x, a)~' = e«e-1(x), a)~).
205
LS 3. If (J is an automorphism of Kab over K then
LS 4. <x, oc)" = 0 if and only ifoc is a norm from K,,(t), where [n"+l]t = x.
This last property uses a fact from local class field theory which could be
proved from the Lubin-Tate formalism, but which we shall take for granted.
Otherwise, the other properties are obvious.
As an application of LS 4, let N m,,, denote the norm from Km to K" for

m ~ n. Then the orthogonal complement of A"n+l under the pairing is given
by
LS 5. Let m ~ n and let N m,,, denote the norm from Km to K". Then for
x E A(.):),,) and oc E K: we have
In other words, N m,,, is the transpose of [n m-,,].
Again this is clear from the functorial properties of the norm residue symbol
which we assume. We can then define the symbol in a limit situation as
follows. We let
T(K!) = group of sequences (oc o, OCl, • , .) with oc" E K: such that for
all n ~ 0,
Thus T(K!) is just the projective limit of the groups K;' under the norm
mappings. We may then define a pairing
A(.):)co) x T(K!) -+ T,.{A)

by letting
On the right-hand side, the components <x, ocm)m are defined for all m
sufficiently large, i.e., m ~ n such that x E A(p,,). The components <x, OCk)k
for k < n may then be defined by applying the appropriate power of [n] to
the nth component. Property LS 5 shows that this is well defined. The pairing
is oK-linear in x and multiplicative in oc E T(K!).
206
Of course it is a considerable restriction on an element a: n in K; to be

liftable to an infinite vector of consecutive norms. In fact, let N n be the norm
from Kn to K, and let
N;; l(nZ) = group of elements a: E K: whose norm Nna: is a power of n.

Lemma 1. We have N;;l(nZ) = n Nm.nK:;.
mO!:n
Proof Suppose Nna: E n Z. Then
because n is a norm from each extension Km by Theorem 2.2. Hence a: is a

norm froll). K m , thus proving one inclusion.
Conversely, suppose that a: is a norm from each Km for m ~ n. Then
Let Nn!Y. = nTu where u is a unit in K. Since n is a norm from Km by

Theorem 2.2, we conclude that
(u, Km/K) = 1 for all m.
Hence u == 1 mod Vm for all m, so u = 1. This proves the reverse inclusion,

and proves the lemma.
LS 6. If p is odd then
(i) <a:, a:)n = O.
Whether p is odd or even, we have
(ii) <a:, -a:)n = 0,
(iii) <x, -l)n = 0 iford" X ~ e+ 1
and so
(iv) <a:, a:)n = 0 if ord" a: ~ e + 1.
Proof Letfn(X) be a Weierstrass polynomial such that
[nn+l](X) = fn(X)g(X)
207
8. . Lubin-Tate Theory
where g(X) is a unit power series in o[[X]]. Let t be a root of fn(X) = IX.
The extension Kn(t) is independent of the choice of t. Thus, if we factor
fn(X) - IX into irreducible polynomials over K n, say
f1 fn.iX),
S
fn(X) - IX =
i=1
then for each j we obtain Kit) by adjoining a single root of fnJX) to Kn.
Therefore, if Cn,i denotes the constant term of fn,lX), and d is the common
degree of the irreducible polynomials fn,i then we conclude that (- 1)dcn'i
is a norm. But
f1 Cn,i'
S
-IX =
i=1
Hence -IX = (_1)dS times a norm, and (_I)dS = (_I)qn+l is a norm. This
proves the first two assertions. For the last two (relevant only for p = 2),
it is clear from bilinearity that
[2]<x, -I) = o.
If ordn IX ~ e + 1, then IX = [2]y for some y, and so <IX, -I) = 0, thus
proving the last two assertions.
The following lemma gives information on the factor group
by showing that near the origin, the operator [7t] operates very regularly. We
let 7tn denote a prime element in Kn.
Lemma 2. Assume k > qn. Then A(7t7t~on) = [7t]A(7t~on)'
Proof The inclusion ~ is obvious. We prove the reverse inclusion. Let

z with t E On. We must solve
= 7t7t~t
~ + 7tX = z with x = 7t~y and y EOn'
This is equivalent to
But k > qn implies that qk > qn(q - 1) + k, so we are reduced to solving

fey) = ayq + y - t = 0
with a divisible by 7t n • Sincef(t) == 0 mod 7t n , andf'(t) == 1, Hensel's lemma
does it.
208
As consequences of the lemma, we find for instance:
A(nn+ 20n) c [nn+l]A(iJn)

A(n~n+l+lon) c [n]A(iJn).
This second inclusion comeS from using k = q" + 1.
Theorem 5.1. Let Wj be elements of iJ" for i = 1, ... , q"+l, such that
ordlln Wi = i. Then these elements generate A(iJn) mod [n]A(iJ,,), and therefore
generate A(iJn) mod[nn+l]A(iJn) over o.
Proof Since X +A Y == X + Y mod deg 2, given x E iJn we can find
a1 E 0 such that
x -A [adw1 == 0 mod iJ~,

because, [adw1 == a1w1 mod iJ~. We may then proceed recursively to find
a2' ... , aq n + 1 such that
By the lemma, this congruence also holds mod[n]A(iJn). Hence the Wj generate
A(iJn) mod[n]A(iJn), whence by Nakayama's lemma, they also generate
A(iJn) mod[n"+1]A(iJn)· This proves the theorem.
The special case when K = Qp is of importance in the cyclotomic theory

(and elsewhere), and some refined statements can be given as in the next two
theorems due to Coates-Wiles [C-W 1], [C-W 2].
Theorem 5.2. Assume K = Qp. Then A(iJn)/[n,,+1]A(iJn) isfree over 0/n"+10.

Suppose that A is the basic Lubin-Tate group. Let I be the set o/integers i
satisfying
1 ::::; i < pn+1 with (i, p) = 1, or i = pn+1.
Let Xo be a non-zero element of A" and let
(Xci, Xl>"" Xn, ... )
be an element ofT,,(A), that is [n]xk+1 = Xk' Then the elements
{x!,} with i E I
form a basis.
Proof For i E I we let Wi = x~. On the other hand, if pT divides i exactly,
we take
Wi = [nT](xUP').
209
This shows that the elements x~ with i E I generate A(.pn) mod[nn+lJA(.pn),

over o.
There remains to prove that they are linearly independent mod nn+lo.
We first show that they are linearly independent in A(.pn)/[n]A(.pn) over o/no.
Suppose we have a relation
(*)
where 2:A indicates the sum with respect to the group law of A, and some
coefficient a i is a unit, say ak. We may assume ak = 1.
Case 1. k < pHI. For any x E.pn we know that
[n]x = x P + ... + px.

Either the term x P dominates this expression, in which case ord pn x is divisible
by p, or some other term dominates, which means that
p. ord x > pn(p - 1) + ord x,

so ord x > pn, and ord [n]x > pn+l. This implies that we cannot have a
relation of congruence (*) because as with ordinary addition, if y, y' E A(.pn)
and ord y i= ord y' then
ord(y +A y') = min(ord y, ord y').
Hence there cannot be any cancellation in the sum of the left-hand side of
(*), thus concluding the proof in Case 1.
Case 2. k = pn +1. Then by Case 1 we may suppose that ai is divisible by

n for all i i= k, and therefore x~n+l lies in [n]A(.pn). We use the hypothesis
that A is the basic Lubin-Tate group, and then there exists y E .pn such that
But x~n ~ Xo and xg -1 = - n. The above equation is clearly impossible if y

is not divisible by Xo because the orders on the left-hand side cannot match
the order on the right-hand side. Then we divide by xoP to find
Reading this equation mod .pn yields a solution of
yP - y == unit modp,
in the residue class field Z/pZ, which is impossible. This proves the theorem.
210
§6. The Logarithm
Theorem 5.3. Assume that K = Qp, and that K(A,,) does not contain the
pth roots of unity. Then the local pairing
is exact on both sides, i.e., the kernels are 0 on both sides.

Proof In Theorem 5.2 we have determined the order of
It is a standard exercise of local algebraic number theory [L 1], Chapter II,

§3 to determine that
where pT is the order of the group of p-power roots of unity in Kn. If K(A,,)
does not contain /lp then neither does Kn. Hence
has the same order as A(Pn)/[nn + l]A(Pn), and we know that the kernel on the
left is trivial. Since A"n+l is cyclic in the present case, it follows by the duality
of finite abelian groups that the kernel on the right of the pairing must also
be trivial, as desired.
Remark. When the pth roots of unity are in K(A,,), in particular when
A = Gm, the above argument definitely shows that the kernel on the right is
non-zero.
§6. The Logarithm

Let A be a formal group, defined by a power series F(X, Y) over some ring 0
with quotient field K of characteristic O. It can be shown that there exists an
isomorphism
with the additive group, i.e., a power series with coefficients in some extension
of K such that
A(X +F Y) = A(X) + A( Y),

A(X) == X mod deg 2.
The + sign on the right-hand side is the ordinary addition. That power series
is then uniquely determined, and its coefficients lie in K. It is called the
logarithm on A, and we write A = AA if we need to refer to A explicitly.
211
Example. Suppose A = G m is the formal multiplicative group. Then the

log is given by
A(X) = 10g(1 + X),
where the log here is the usual series from calculus.
It is easy to show that any endomorphism of G a is given by multiplication

with a scalar, i.e., if a power series h satisfies
h(X + Y) = h(X) + h(Y),

then h(X) = aX for some constant a. Hence the uniqueness of the log AA
follows at once. In this section we shall prove its existence for Lubin-Tate
groups, and additional properties, following Wiles [W] in preparation for the
explicit reciprocity laws.
Lemma 1. The limit
A(X) = lim ~ n~(X)

n
exists, and gives a formal isomorphism of the Lubin-Tate formal group A

with the additive group Ga.
Remark. The limit is to be understood in the following sense. Each term
is a power series. By the existence of the limit, we mean that for each k,
exists as n ---7 00, and then A(X) is defined to be .I CkX k. The convergence will
not be uniform in k.
Proof of the lemma. We look at the difference
LT n~+T(X) - ~ n~(X) = LT (nA 0 n~(X) - nTn~(X)).

n n n
Let
where
j 2. 2,
212
§6. The Logarithm
and the coefficients al7) are in OK. Then the right-hand side of the required
difference is equal to
We are interested in the coefficients of monomials of degree::;; k for a fixed k.

Reading all expressions mod Xk + 1 we see that we may assume
i ;::; n + 1 - (k + I).
Hence nnx + gn(X) is divisible by nn-k. Similarly, the power series expression
for gr is divisible at least by n r- k • Since gr has degree ;::; 2 it follows that
glnn X + gn(X» is divisible at least by
Dividing by nn+r shows that the required difference tends to 0 as n --l>- 00.
This proves that the desired limit exists.
It is clear that A(X) == X mod deg 2.

There remains to prove that A satisfies the homomorphism property. We
have:
~
n
n1(X +A Y) = ~
n
(n1X +A n1 Y)
where C lj are the coefficients of the formal group
For each fixed k, m the coefficient of Xk ym in the sum on the right-hand side
tends to 0 as n tends to infinity, so the additivity follows.
Lemma 2. The log AA commutes with the action of 0, that is,
For the basic Lubin-Tate group B, if
L ai Xi
<Xl
AB(X) = X +
1=2
then ai = 0 unless i == I·mod q - 1.
213
Proof The function X 1---+ AA(aA(X)) is an additive formal power series.

such that
AiaiX)) == aX mod deg 2.
The uniqueness of the logarithm shows that this function is aAA'

For the basic Lubin-Tate group, we take a = , where , is a primitive
(q - l)th root of unity, and apply Theorem 2.3 to conclude the proof.
Lemma 3. (i) Let A' denote the formal derivative dAfdX. Then A~(X) has
coefficients in o.
(ii) The series AiX) can be written in the form
where gj(X) E o[[X]]. In particular, it converges on the maximal ideal.
(iii) Suppose q ~ 3 and let x E Ka have ord" x ~ 1. Then
Aix) == x mod x 2.
Proof For (i) we differentiate with respect to Y the relation
and get
A~(F(X, Y))D2F(X, Y) = A~( Y).
We then put Y = 0, and find
But from F(X, Y) == X + Y mod deg 2, it follows that D2F(X, 0) is a power

series whose constant term is 1, and with coefficients in o. This proves the
first assertion.
As for (ii), it suffices to prove the result for the basic Lubin-Tate group
whose Frobenius power series is given by
because if",: A --'>- B is an isomorphism such that ",(X) == X mod deg 2, then
AA = AB 0 "'.
214
§6. The Logarithm
It also clearly suffices to prove the following statement:
The power series [nn](X) lies in the module
withj ~ 0, k ~ 1 andj + [logq k] ~ n.
We prove this by induction. It is obvious for n = 1. Assume it for n. Let
Then
It is immediate that nfn{X) satisfies the induction hypothesis with respect to

n + 1. For"the termfn(X)q, it will consist of cross terms which binomial-type
coefficients divisible by p, hence by n, thus satisfying the desired conditions
on the exponents, or terms
The logq of the exponent of X is increased by one, and so the desired in-
equality is also satisfied. This proves (ii).
Part (iii) is obvious from (ii).
Observe that in the simplest case of the ordinary log,

X2
log(l + X) = X - "2 + ....
If P = 2, the first term after X gives trouble. If p = 3, the next term which
might give trouble is
X3
3'
but in this case, the assumption ordp x ~ 1 shows that (iii) holds. After that,
things only get better.
Lemma 4. Let eACZ) be the power series (with coefficients in K) which is

the inverse of AA(X). Let D be the disc in mK a consisting of those elements z
such that
1
ord z > - - .
" q- 1
215
s. Lubin-Tate Theory
Then eA(Z) converges on this disc, and induces the inverse isomorphism to
AA, on the groups
For z in this disc we have
ord z = ord Aiz) = ord eiz).

Proof Let y ED, Y # O. Define
1
2y(X) = - AiyX).
y
Then for i > 0,

1 yql I . qi - 1 .
ord - - l = (q - 1) ord y - 1 > - - - 1.
Y n q - 1
By Lemma 3, it follows that 2y(X) has integral coefficients. Let Ey be the

power series such that Ey o2y(X) = X. Replacing X with y 1 X we see that
Ey(Z) = y1eiyZ). Since Ey has integral coefficients (because Ay(X) =
X + higher terms), we conclude that
1
- eiyZ)
y
has integral coefficients. Let eiZ) = 2: anZn. Then anyn-l is integral for all
n and all yin D. It follows that in fact, anyn tends to 0 (p-adically) as n tends
to infinity for each y in D. Furthermore, we then conclude that
for all y in D, and in particular,
ord eiy) ;::; ord y.
On the other hand, again using Lemma 4, it is immediate that for x in D,

we have
Since eA and AA give inverse mappings, we get
thus proving Lemma 4. '
216
§7. Application of the Logarithm to the Local Symbol
We may then recover immediately a lemma proved in connection with

Theorem 5.1.
(ii) AAA(.\ln) ~ nOn'

Proof Clear.
Lemma 5. The kernel of AA in the maximal ideal of the algebraic closure of

K is precisely A tor , the group of torsion points on A, or in other words, the
group A(n>.
Proof A point x is a torsion point if and only if [nn]x is a torsion point
for some positive integer n, or for every large positive integer n. But [nn]x
approaches 0, and for large n, lies in the neighborhood of 0 where the
exponential and log on A give inverse mappings. Since on the additive group,
there are no elements of finite order, it follows that the kernel of AA is precisely
A(">.
§7. Application of the Logarithm to tile Local Symbol

We recall that the finite extension Kn is self dual as a vector space over K
by means of the trace. This means we have a non-degenerate K-linear pairing
given by
(x, y) J--+ Tn(xy).
Let a be an ideal in Kn. We denote by a.l the set of elements y E Kn such that
Tixy) E 0 = OK'
Of course, we have the notion of perpendicularity with respect to any given
pairing, and the context will always make clear which is meant. We have
a.l = Homo(a, 0).
Indeed, let ljJ: a ~ 0 be a o-homomorphism. Then ljJ can trivially be extended

to a K-linear functional Kn ~ K denoted by the same letter. But then for some
a E Kn we have
ljJ(x) = Tn(xa) for all x E Km
217
and CI. E a.L by assumption. Our identification of a.L with Homo(a,o) then
follows at once. If'Il" = 'Il KnIK is the different, then
We return to the pairing given by the local symbol
We had already noted as a consequence of LS 4 that
Observe that we are dealing with two orthogonality signs: One referring to
the local symbol, and one referring to the duality
(where the trace is viewed as having values in the factor group Kjo), applied
in particular to an ideal
Then we have the pairing
Since Ker AA = A tor , we have Ker AA n K" = A"n+1. Applying the log map
of A, we get a pairing
Let a = AA(~n)' so the factor group on the left is ajn"+la. In the light of the
exact duality
there exists a unique group homomorphism
such that for x E A(~,,) and CI. E Ain+1 we have
LS 7.
218
§7. Application of the Logarithm to the Local Symbol
This formula has been written without abbreviations, but of course in the
future we frequently omit indices A, n, etc. If U E Gal(KRjK) and U 1--+ x(u)
is its representation in 0* on T,,(A), then
LS 8.
Proof We have
and also
Butuxn = [X(U)]Xnby the definition of x, withx(u) EO*. Using [ab] = [a][b]

and Tn(uy) = Tn(Y) concludes the proof of IS 8.
219
9
Explicit Reciprocity Laws
Iwasawa [Iw 8} proved general explicit reciprocity laws extending the classical
results of Artin-Hasse, for applications to the study of units in cyclotomic
fields. These were extended by Coates-Wiles [CW 1] and Wiles [Wi] to
arbitrary Lubin-Tate groups. Although Wiles follows Iwasawa to a large
extent, it turns out his proofs are simpler because of the formalism of the
Lubin-Tate formal groups. We essentially reprod'uce his paper in the present
chapter.
We assume that K is a finite extension of Qp (i.e. has characteristic 0)
because we want to use the logarithm.
We allow p = 2, and I am indebted to R. Coleman for showing me how
Wiles' paper extends with essentially no change to that case, by using (ii),
(iii), (iv) of LS 6, and the minus sign in DL 6.
We let:
A = Lubin-Tate group associated with the prime element n.
We let:
(Xo, Xl, ... ) E T,,(A) with Xo =F O.
N n = norm from Kn to K, and N m.n = norm from Km to Kn for m :::: n.

Tn = trace from Kn to K.
We abbreviate <,
)n to <, ) unless we wish to specify the level at which
the symbol is taken.
220
§1. Statement of the Reciprocity Laws
N;.l(rcZ ) = subgroup of K: consisting of those elements whose norm to K

lies in rcZ (i.e., is a power of rc).
= n Nm.nK! by Lemma 1 of Chapter 8, §5.
m.,n
T(K!) = group of vectors (cto, ctl> •.. ) with ctn E K: such that
§1. Statement of the Reciprocity Laws

Theorem 1.1. Suppose ct E On and ct == 1 mod .\:.In. Then
and
<Xn, ct)n = [rc Ll (Nnct- 1 - l)]Xn.
Proof By the formalism of the norm residue symbol, we know that
Hence [Nnct]xn = Xn by the Lubin-Tate theory, so the first assertion is clear.

We choose t = X2n+l so that [rcn+1]t = Xn. Then
Since (ct, Kn(t)/Kn) = (Nnct, Kn(t)/K) we obtain from Lubin-Tate theory

<Xn, ct) = [Nnct-1]t -A t
= [Nnct-1 - l]t.
Using the first congruence and the fact that [rcn+1]t = Xn yields the theorem.
Coronary 1. Let ct == 1 mod .\:.In. Assume that K is unramified over Qp. Then
where the log is the ordinary log on the multiplicative group.

Proof Since n is unramified, we can write
221
9. Explicit Reciprocity Laws
where z == 0 modpn+l. Since p -# 2 it follows that
Hence
Since K is unramified over Qp, we have 1t '" p, and the corollary follows.
Corollary 2. Let A = G m be the formal multiplicative group. Let , be a

primitive pn + lth root of unity, and let ct == 1 mod Pn' Then
Proof Special case of Corollary 1.
The law of Corollary 2 is one of Artin-Hasse's laws, obtained here by

Wiles as a special case of the Lubin-Tate formalism. We have written the
symbol with the usual parentheses, transfering its meaning to the multiplica-
tive group.
We shall now state the main result of this chapter. Let ct E K~. Let
r = ord vn ct. Let g(X) = crr + ... be a power series in o[[X]] with a unit
Cr such that
Of course, there exist infinitely many such power series. Let
D = d/dX
be the ordinary derivative of formal power series, so that
Dg(X) = g'(X).
Define
The operator DnL depends on the choice of element
(X Q , Xl, ... ) E TiA),
and it depends on the choice of g. We shall see later to what extent it does
not depend on g.
222
§l. Statement of the Reciprocity Laws
We define
Again this dep~nds on the choice of g.

Let x E A(Vn) and let O(m E K:f:. In Lemma 3.2 we shall give conditions
under which the symbol
is well defined mod nn + 1 independently of m. These conditions involve either

x being sufficiently divisible in A(Vn), or m being sufficiently large. The value
of the symbol lies a priori in K/nn+lo, but it will turn out that under suitable
conditions, its value lies in o/nn+lo, so that it can be viewed as an operator
on A"n+1. This was the reason for selecting the bracket in the notation.
Precisely, the conditions are as follows.
Condition (i) m ~ 2n + 1 and there exists an integer
k ~ [nI2] + 2(n + 1)
k ~ max{[n/2] + 2(n + 1), e + I} ifp = 2,
Condition (li) m ~ nand ord" x ~ [n/2] + 2.

Theorem 1.2. Let x E A(Vn) and 0( E K:. Suppose 0( = Nm.nO(mfor some
O(m E K;:. Under either one of the conditions (i), (ii), the symbol [x, O(m]m
has value in o/nn+lo and we have equality
An important case is that when
0( = (0(0, 0(10 ••• ) E T(K!).
Thus O(n satisfies an infinitely regressive norm-divisibility condition. In that

case we may define the symbol
[x, O(] = [x, O(m]m
223
for arbitrarily large m, and its value will be the same for any m ;:::: 2n + 1.
It gives the formula
Example. The formal multiplicative group. For P == 1 mod fl" and IX E K:

we defined the classical norm residue symbol
([3, lX)n = <P - 1, IX)~ + 1,
where A = G m• Consider the special case when K = Qp and 11: = p. Let ,

be a primitive pn+lth root of unity. Then
xn = , - 1.
We have AA(X) = 10g(1 + X), so
1'( ) _ 1 _ Y-l
AA Xn - 1 + Xn - . " •
Let x E flri = Gm(fln). Then we find for m ;:::: 2n + 1:

1
(x, lX)n = ,. where v = pm+l T m (~log (1 + x) DXm log IX).
This is Iwasawa's formula [Iw 10].
Finally there is another Artin-Hasse reciprocity law generalized by Coates-

Wiles to the Lubin-Tate case for level O.
Theorem 1.3. Let x E A(fl5) and IX E Kt. Then the symbol [x, lX]o has values
in oireo, and we have
The rest of the sections will be devoted to the proofs.
§2. The Logarithmic Derivative

In this section we investigate systematically the logarithmic derivative, when
it is well defined (modulo certain powers of the prime), and also to what
extent the mapping On(lX) is well defined. We let:
'il n = different of K" over K;

'il m,,, = different of Km over K" for m ;:::: n.
Then
224
These are immediate by considering the basic Lubin-Tate generators Wn ,

satisfying the equations:
wg- 1 + n = 0 and W~+l + nWn+l - Wn = O.
The relative difference is obtained by taking the derivative, and evaluating at

Wo and Wn+l respectively. The given values fall out.
Let rx E Kn and write rx as a power series
rx = g(xn) with g(X) = crX" + higher terms

and Cr equal to a unit. Let g(X) = X"h(X). Then
g'/g(X) = i + h'/h(X).
Hence g'/g(xn) is integral if r = 0, and in any case lies in V;l. If
are two power series whose values at Xn are equal to rx, and f(X) is the
irreducible polynomial of Xn over K, then
for some power series cp(X) E o[[X]]. Hence
This shows that g' /g(xn) is well defined modulo the different.
DL 1. The map DnL is a homomorphism
and the image of the units lies in On mod :!In.
This is obvious from the previous discussion. Since A'(X) is a power series
starting with I and with integral coefficients, it follows that A'(xn) is a unit.
Hence from the definition of bn. we find:
DL 2. The map bn is a homomorphism
Its image lies in n-nV;l mod :!loon, and the image of the units lies in
n-n mod :!loon.
225
As the elements of Kt are generated by powers of Xn and units, the com-

putation of ~n is reduced to ~n(xn) and ~n(units). Note:
DL 3. For a E Gal(Ks,/K), and IX E Kt,
~,,(IX(J) = x(a)~,,(IX)"'.
Proof Write IX = g(x,,) as usual. Then

alX = g(ax,,) = g([x(a)]x,,).
Thus we let g.,.(X) = g([x(a)](X», so that
On one hand, we have
(1) n" 0 (IX)'" = _1_ g'/g(ax ) = _1_ g'([x(a)1x,,)

" A.'(ax,,) "A'(ax,,) alX
On the other hand, since A [x(a)](X) 0 = x(a)A(X), we find
(2) X(axn)[x(a)]'(x,,) = x(a)A.'(x,,).

Furthermore
(3)
and
g.,.'(X) = g'([x(a)J(X)[x(a)]'(X).
Putting (1), (2), (3), (4) together yields the desired property.
DL 4. Let m :2: n and let IX be a unit in Kt. Then
Om(lX) == ~,,(IX) mod :!loom.

Proof The proof is similar to DL 3. We know that
so by the chain rule,
226
We have the representations
where
Since IY. is assumed to be a unit, the power series g starts with a unit, and so
does the power series gm, so both these power series can be used to compute
the logarithmic derivative. The rest of the proof then follows immediately
from the chain rule and the definitions.
DL 5. Let m :?: n, and IY.mE K;:. Then
bn(Nm.nlY.m) == Tm.nbm(lY.m) mod On'

Proof Without loss of generality we may assume that m = n + 1. We
first deal with the case when IY.m is a unit. W,e find:
(}n(Nm.nlY.m) = bm(1JIY.~) == ~bm(IY.~) byDL4
= L x(u)bm(lY.m)", by DL 3.
"
The sums are taken over U E Gal(Km/Kn). For such U we must have
X(U) == 1 mod xn+l
because [x(u)] is the identity on A"n+!. Since bm(lY.m)" lies in x-mom = x-<n+l)om,
the desired congruence follows.
It will then suffice to prove DL 5 next for IY.m = X m , because of the multi-
plicativity of the function (). For simplicity, let us first suppose that the
Frobenius power series associated with the Lubin-Tate group is in fact a
polynomial,
[x](X) = f(X) = xX + ... + xq,
and that the coefficient of xq is exactly 1. For instance, the basic Lubin-Tate
group and the formal multiplicative group are of this type. Under this
additional assumption, we have in fact the stronger property with equality
instead of the congruence:
DL6.
Remark. The minus signs are there to take care of the case p = 2. If p i= 2,
they can be omitted.
227
Proof We may again suppose m = n + 1. We have (as in the proof of

Theorem 2.2 of Chapter 8)
The formula to be proved amounts to
We have Xn = feXn+l), and since A(f(X» = 1l:A(X), we have
(A 0 f)'(X) = A'(f(X»f'(X) = 1l:A'(X).
We put X = Xn+b and see that the formula amounts to
We replace Xn by f(x n+1) = 1l:Xn+l + ... + X~+1' Let a = Xn+l' Standard

orthogonality relations of elementary algebra (see for instance Algebra,
Chapter VII, §6) yield
Tn+l.n(;:~a;) = 1 if i =q
= 0 if i 'I:( q.
This proves what we wanted.
The proof of DL 5 in general when am = Xm follows exactly the same

pattern, but we end up only with the asserted congruence. We give the details.
By the Weierstrass theorem, we may factor in On,
f(X) - Xn = g(X)h(X)
where
g(X) = b o + ... + bq_1xq-l + xq,
heX) = Co + clX + ... is a unit power series, Co E o~.
Then f'(Xn+l) = g'(xn+1)h(xn+l)' Proceeding as before, we are reduced to

proving the congruence
228
§3. A Local Pairing with the Logarithmic Derivative
Again we replace x" by f(x n +1)' From the factorization we have
Hence
From the orthogonality relation, we obtain a contribution of 1 from the trace

of one term. From the definition of the different (which is precisely g'(xn + 1))
it is then clear that the traces of all the other terms are == 0 mod .jJn, as desired.
Property DL 5 can be expressed in the projective limit as usual. Let
T(Ko%",) = projective limit of the additive groups Kn/on under the

trace maps,
= group of vectors (zo, Zl> •.. ) with Zn E Kn/on such that
Then the map
given by
is well defined, and is a homomorphism.

Having derived the necessary formalism for the values of Dn(ct), we may now
combine this with the logarithm on A to define the symbol
Lemma 3.1. The symbol [x, ctl m is well defined mod nn+l in each ofthefollow-
ing cases:
(i) x E A(.jJ,,) and m ;::: 2n + 1;
(ii) x E A(.jJ~qn) and m ;::: n.
Proof By DL 2 we know that Dm(ct) is well defined mod :Doom. Hence the
symbol is defined mod n" + 1 if
229
By the definition of the different 'Il,;;-l = n- m'Il o1om, this is equivalent with:
nm-n-2A(X)'Il~ is integral.
It will even be shown that in case (i), n m-n-2A(x)'Il o is integral.

For future reference, we prove congruences which imply the above, and
list them systematically.
C 1. If x E A(Vn) and m ~ 2n + 1 then
TmG A(X)Om) c nn+1o.
C 2. If x E A(v~qn) and m ~ n then
Tm(~ A(X)'IloOm ) c nn+1o.
C 3. If x E A(v~qn'Ilo) and m ~ n then
Observe that the 'Ilo does not occur inside the trace in eland C 3. Only C 1
and C 2 are needed for Lemma 3.1 but C 3 will be needed for Lemma 3.2.
We now give the proofs.
Suppose first that m = 2n + 1 (the worst case of (i)). We have to verify
that
nn -1 A(X)'Il o is integral.
Recall that ord" 'Ilo = (q - 2)/(q - 1).

By Chapter 8, §6, Lemma 3 we know that A(X) is a power series in x whose
terms are either integral, or at worst with a factor
~I •
j' and 1 ~ 1.
n
Suppose x E A(VlI). Then
lI-l1()<"1'"\ q- 2
ord n A x ""'0 ~ n - 1 + q j qn( q 1- 1) - 1. + --1·
q-
We need the right-hand side ~ O. For i ::; n - 1 this is obvious, because

n - 1 - i ~ 0 and the other terms are positive. For i ~ n the estimate is
equally easy.
Suppose next that x E A(v~qn) and only that m ~ n, say m = n which is
the worst case. Then x ·lies in the disc of" good" convergence for the log and
230
exponent, and thus

2
ord A(X) = ord x ~ --I'
q-
For C 2, it suffices to verify that n - 2 A(X)'l) 5 is integral, or equivalently
2
-2 +- - + 2q-2
q-2
--
q-l
~ 0,
which is obviously the case. The proof for C 3 is the same.
We remark that in the range where the symbol is well defined, it is OK-linear
in x and multiplicative in IX. In any case, within the ranges of Lemma 1, we
view the symbol as having values in
The next lemma will show that the value [x, IXm]m is independent of m when
IXm is the component of an infinite vector
a = (ao, al> ... ) E T(K!),
and m is sufficiently large. We define [x, IX] to be this value.
Lemma 3.2. Let k ~ m ~ n. Let IXm E K;: and ak E Kit be such that
Then
in either case."
(i) x E A(Vn) and m ~ 2n + 1;
(ii) x E A(V~qn'l)o) and m ~ n.
Proof We have:
== n1 Tm(A(X)t>mCNk.mak)) mod nn+l
== nTm(A(X)Tk.mt>k(a k))
1
by DL 5 and C 1 or C 3
= !n Tk(A(X)t>kCak))
= [x, adk,
as was to be shown.
231
§4. The Main Lemma for Higbly Divisible x and IX = Xn
Only the statement of the main lemma will be used later, and we recommend
to the reader to read the next sections before reading the proof of the main
lemma.
Lemma 4.1. Let x E A(Vn). Suppose that
ord" x ~ [nI2] + 1 + 2e,

where e is the ramification index of Kover QI'. Then
<x, xn>n = [x, xn]~(xn).

Proof First we remark that for the applications, the exact nature of the
order condition on x is irrelevant, and the reader will find it easier just to
assume that the lemma is being proved under the assumption
ord" x ~ [nI2] + 50,OOOe,

or any other high multiple of e, which would do just as well. Also, instead
of [nI2], any expression like [n l - e] would do just as well. In the next section
it will be shown how to use a duality to lower back such expressions to the
precise ones which we ultimately want.
We let ord" x ~ 't(n) , and derive sufficient conditions (also more or less
necessary) for the method of proof to yield the lemma.
During the course of the proof we shall constantly be interchanging
logarithms with the first term in the expansion. If ord" x ~ 1 then
x2
A.(x) == x mod -
1t
If p is odd, we even have
A.(x) == x mod x 2 •
Furthermore, if ord" y ~ e then

2
log(l + y) == y mod ~ .
The formula to be proved is
232
§4. The Main Lemma for Highly Divisible x and IY. = Xn
We first want to replace A(X) by x on the right-hand side. It suffices for this
that
Since A'(xn) is a unit, this is equivalent with
q- 2 1
(1) 2..(n) - n - 3 + --1
q- - q n( q- 1) ~ o.
Certainly ..(n) ~ (nI2) + 2 suffices. Again since A'(Xn) is a unit, proving the
formula for all x is equivalent to proving
This will be done by the sequence of following steps.
Step 1. <A'(Xn)X, x n> = <x + Xm x n>

Step 2. <xn + x, xn> = [-I]<xn' Xn + x>
Step 3. <Xn, Xn + x> = < Xn, 1 + :n)· We then apply the basic re-
ciprocity law of Theorem 1.1 to show that this is equal to
to conclude the proof.
Step 1. We shall use the expression involving the mapping l/!n in Chapter 8,
§7, formula LS 7, namely
>
<x, x n = [Tn(A(X)l/!n(xn)]ixn),
=:= [Tn(xl/!n(xn)1A(xn).
Indeed, A(X) == x mod nnt(n) and Tn(n 2t(n)l/!n(x n» == 0 mod nn+\ because the
image of l/!n is contained in AA(Pn).L, and
Therefore we need
233
For this it suffices that

(2) 2't(n) ~ n + 2.
Again 'ten) ~ (nI2) + 2 suffices.
Since A'(xn) is a unit, we may replace x by A'(xn)x to obtain
°
By Taylor's formula, using the fact that A'(X) has integral coefficients, and
A(Xn) = we get
provided that also 'ten) ~ e. So let us make
n
'ten) ~ 2 + 1 + 2e.
As we have already seen that
we may replace A'(xn)x by A(X + xn) to conclude the proof of Step 1.
Step 2. Formally, this is just the alternating property LS 6 of the symbol

<0(,0(> = 0, but the proof has to be adjusted because the groups involved on
the right and left do not playa symmetric role. We have:
°= <xn + x, Xn + x> = <Xn + x, 1 + :n) [+ ] <xn + x, xn>

«Xn + x)[-]xn' 1 + :n)
<
=
[+ ] Xn, 1 + :n) [ + ] <xn + x, xn>
= < (xn + x)[ -] Xm 1 + :n)
[ + ] <xm Xn + x> [ + ] <xn + x, xn>.

There remains to prove that the first term on the right is 0, and for this it
suffices to show that
234
§4. The Main Lemma for Highly Divisible x and t:I. = Xn
with positive integers r, d such that
(3) r + ed;;::: n + 1.
Let F(X, Y) be the group law on A. Since F(O, Y) = Yand F(X, 0) = X

we see that
F(X, Y) == X + Ymod XY.
It follows at once that
(x + xn) [-] Xn == 0 mod x == 0 mod n«n).

But then we may take
(4) r = 'ten) - 1.
To solve 1 + y = uPd with some unit u, for y = x/xn , we use the fact that
the ordinary log and exp preserve the order ~m the disc of elements z such that
ordp Z > l/(p - 1). It follows that we can take any integer d satisfying
1
o~ d < ordp y - --1'
p-
or in terms of n,
'ten) 1 1
(5) O~d<-- n( 1)---1'
e eqq- p-
For instance it suffices that
1
o~ d < - 'ten) - 1.
e
Picking't(n) = [n/2] + 1 + 2e suffices. This concludes Step 2.
Step 3. Let y = x/xno and t:I. = 1 + y. We have:
<Xn, Xn + x) = <xno 1 + y)
= [n:+1 (N (1 + y)-l -
n l)LeXn)'
We contend that
whence it follows that
<Xn, Xn + x) = [ - n L1 Tn(y)L(Xn),
thereby concluding the proof of the main lemma.
235
To prove the contention, write N n(1 + y)-l = 1 + Z, with
(Cf. Theorem 1.1.) Then
log Nn(l + y)-l = Tn(log(1 + y)-l)

== -Tn(Y) mod TnHy2on).
This amounts to the same type of congruence as before, and is obviously

satisfied. This concludes the proof of the main lemma.
§5. The Main Theorem for the Symbol <x, xn)n

Theorem 5.1. We have the equality
<x, xn)n = [x, xm]~(xn)
under either of the following conditions:

(i) x E A(.):ln) and m ~ 2n + 1
(ii) x E A(.):l~qn~o) and m ~ n.
Proof By LS 5 of §5 in the preceding chapter, we have for m ~ n
This shifts the burden of the proof to level m, and [1tm-n]x is divisible
approximately of order m (asymptotically for m -+ (0). Specifically, to apply
the main lemma of the preceding section, we want to take m so large that
ord,,[1t m- n ]X ~ [mI2] + 1 + 2e.

Since
it suffices to take m ~ 4(n + 1) + 4e.

Let x' = [1t m - n]x. We apply the main lemma of the preceding section to
x' instead of x and m instead of n, to conclude that
<x, IXn)n [[1t m- n]x, IXm]A(Xm)
t
=
= [1tm -n ~ Tm(A(X)Om(xm) (xm).
236
§5. The Main Theorem for the Symbol <x, xn>n
We write this as
This means in particular that nm-n a is integral. On the other hand, there
exists b E 0 such that
Hence we have the congruence
whence it follows that a is integral and also that
which concludes the proof of the equality between the symbols under either
condition (i) or (ii) according to Lemma 3.2.
The next theorem shows how one can refine the conditions of Theorem 4.1,
with a more precise definition ofthe symbol bn(Xn) for the special case rx = Xn.
Theorem 5.2. Assume that the Frobenius power series associated with the
Lubin-Tate group A has the form
f(X) = xq + ... + nX,

i.e., is a polynomial of degree q with leading coefficient 1. Define more
precisely
Then for x E A(.j:)n} we have
Proof First observe that the elements - Xm form a vector
(-Xo, -Xl> ... ) E T(K!),
i.e., each is the norm of the successive one. Instead of using Lemma 3.2,
however, which relied on DL5, we may now use directly the more precise
237
relation DL 6 which gives an equality implying the stronger statement:
1
= it Tn(A(x)Tm.nc5m( -xm»
1
= n- Tn(A(x)c5 n( - xn»,
The value c5 m( -xm) is here taken to be specifically (l/n m)(1/A'(xm)xm ), rather

than up to a congruence.
Example 1. Take A = Gm to be the formal multiplicative group. Then it

satisfies the hypothesis of Theorem 5.2, and we obtain another reciprocity
law of Artin-Hasse:
where
[x, -xn] =
p
Ll Tn (.IXn log(l + x»)
Xn = , - 1,
and, is a primitive pn+lth root of unity.
Example 2. Take again A = Gm , let p #- 2 and let B be the special Lubin-

Tate group, corresponding to the Frobenius polynomial
x p- 1 + p = o.
We contend that:
<xh, xo> = 0 if i is an integer prime to p, or i > p.

<xg, xo> = Xo.
Proof For the first statement, we have by multiplicativity:
o = <xh, xh> = [i]<xh, xo>.

If i is prime to p, this proves our assertion because Ap is a p-group. If i > p,
then one sees from the formula with the trace that the symbol gives O.
The more interesting case is <xg, xo). We could work directly on the
multiplicative group as was done classically, but the functorial formula LS 2
238
§6. The Main Theorem for Divisible x and rx = unit
of §5 in Chapter 8 shows that it suffices to prove the result on the special

group, where
xg- 1 = -po
By Lemma 2 of §6 in Chapter 7 (the logarithm for the special group) we know

that
Aix) = x + terms of degree ;::: p.
Hence
= To( -1) (mod p)

= 1 (modp).
This proves the assertion.
§6. The Main Theorem for Divisible x and rx = nnit

Theorem 6.1. Let x E A(.):l,,) and suppose
ord" x ;::: [nI2] + 2, ifp is odd

ord" x ;::: max{[nI2] + 2, e + I} if p = 2.
Then for any unit rx we have
<x, rx>" = [x, rx]~(x,,).

Proof The units are generated by f.lq -1 and the units == 1 mod .):l". If
rx then both sides of the equation are trivially equal to O. So we deal
E f.lq -1
with the units == 1 mod .):l". A (topological) set of generators for these units
consists of the elements
1 - ex", with j = 1, 2, ... ,
where e is a (q - 1)th root of unity. The elements x~ form additive generators

for A(,pn) over 0, and in our case, we need consider just those powers with i
satisfying
ord" x~ ;::: [nI2] + 2.

239
It suffices to prove the lemma for the symbol (x~, 1 - eX~)n with such values
of i. We shall reduce the proof to the case (x, xn).
We start with the symbol [x~, 1 - ex~]. Since
-jex~-l -j ~
"n{l - ex~) = ..1.'(Xn)(1 _ Xni) = X() L.. eTx"'./,
Xn Xn r=l
we find
[X~, 1 - ex~] = i
-::11 r=l Tn(;_(Xni) ..1.'(Xn1)Xn eTx"'./).
n
_ ~
-
n T=l JI. Xn Xn
(i
~ Tn Xn , '( 1 ) er XnTi)
n + 1 L..
L [eTx~+TJ, Xn].
00
=-j
r=l
The above formal steps are obviously justified. First the sums taken mod nn + 1
are actually finite, and second we have replaced A(Y) by y and vice versa
twice in the range where this applies. The equality takes place in Klnn+1o
where the symbol [x, 0:] takes its values.
By Theorem 5.1 we know that
[eTxin
+TJ, n
XA] (x
n
) = (~TXi+TJ
li'n'n"
X)
Therefore
LA
00
[x~, 1 - ex~]ixn) = [- j] (erx~+TI, Xn),

r=l
and this latter sum is taken on A. Since (x, -I) = 0 by LS 6 if ord" x is big
enough, it will therefore suffice to prove the next and final result.
Theorem 6.2. Suppose

ord" x~ ~ [nI2] +2 ifp is odd
ord" x~ ~ max{[n/2] + 2, e + I} if p = 2.
Let j ~ 1. Then
1> - [ .] ""
00
(X ino eXin - n - L..A (r

e Xn+ rl ,Xn)
- ] i n'
r=l
Proof Let F be the group law on A. Since
.F(X, y) == X + Y mod XY,

240
§6. The Main Theorem for Divisible x and a = unit
we obtain for x, y E .):l,,,
x [ +] y == x + y mod xy and x [ - ] y == x - y mod xy.
This will be applied when ord" x and ord" y ~ [nI2] + 2, so that
ord" xy ~ n + 3.
In that case, A(n"+3 0 ,,) c [n"+l]A(.):l,,), so addition on A and addition on Ga
are interchangeable on the left of the symbol
(x,a)"
under the condition ord" x ~ [nI2] + 2.

This b~ing said, we find:
o = <x~(ex' - 1), x~(ex' - 1»

= <ex~+J - x~, x~) [ + ] <ex~+i - x~, ex' - I)
= <ex~+i, x~) [+ ] <ex~+J, ex' - I) [- ] <x~, ex~ - I)
whence
<x~, ex' - I) = <ex~+J, ex~ - I) [+ ] <ex~+I, x~).
Note that <x, e) = 0 for all e E jlq-l.

Recursively we obtain
LA <erx~+rJ, X~+<r-l)J)
co
<x~, ex' - I) =
r=l
co
= "" <erxl + rJ x-J)
L..A
r=l "'"
LA <erx~=rJ, x,,).
co
= [-j]
r=l
Observe that the sums are in fact finite since for large r the left-hand side of
each symbol in the sum is highly divisible, so the symbol is o. Hence the above
formal steps are valid, and the theorem is proved.
The special case when n = 0 is often interesting for its own sake. The
next lemma may be omitted in the proof of Theorem 7.1 or Theorem 1.2,
but is useful for the proof of. Theorem 7.2.
241
Lemma 6.3. Assume that B is the special Lubin-Tate group with Frobenius
power series xq + nX. Let Wo E B". Then for i ~ 2 and j ~ 1 we have the
same formula with n = 0 as in the previous lemma, namely
2:B <wi/ri, Wo>o·

co
<wi, W6 - 1>0 = [ - j]
r=1
Proof By Theorem 2.3(ii) of Chapter 8, we know that the group law on

B satisfies
F(X, Y) = X + Y + terms of degree ~ q.
If x, y E .):l02 it follows that
But .):l~q c [n]B(.):lo). Hence again addition on A and addition on G a are

interchangeable on the left of the symbol <x, IX>o under the stated conditions,
and the rest of the proof is then identical with that of Theorem 6.2.
§7. End of the Proof of the Main Theorems

Theorem 7.1. Let x E A(.):l,,) and let
Assume that
k ~ [nj2]+ 2(n + 1) ifp is odd

k ~ max{[nj2] + 2(n + 1), e + I} ifp = 2.
Then [x, IXm] lies in ojn'H1 o for 2n + 1 ~ m ~ k, and we have
for such m.
Proof The theorem has already been proved when IX is a power of Xn or
when IX E jiq-1' We may therefore assume that IX is a unit == 1 mod .):In' We
reduce the theorem to the result of the preceding section in exactly the same
manner that Theorem 5.1 was reduced to the main Lemma 4.1. This time
we need
ord,,[nk-n]x ~ [nj2] +2 if p is odd

ord,,[nk-nJx ~ max{[nj2] + 2, e + I} if p = 2.
242
§7. End of the Proof of the Main Theorems
Clearly then, the lower bound on k given above is sufficient. The proof is
then identical with that already given for Theorem 5.1, as desired.
For the case n = 0, special features arise, and we next give a generalization
to Lubin-Tate groups of another explicit formula of Artin-Hasse.
Theorem 7.2. Assume p l' 2. Let x E A(p~) and a E Kt. Then

1
[x, aJo = - To(A{x)bo(a»
n
is well defined mod n, lies in o/no, and we have
<x, a)o = [x, aJg(xo).

Proof Since A(Xo) = 0 it follows that A(Po) = A(p~) = p~. This shows
that [x, aJi> is well defined mod n because <5o(a) is well defined in POl mod<Il o.
By formula LS 2 of Chapter 8, §5 it suffices to prove the theorem when
A = B is the special Lubin-Tate group. In that case we already know the
result when a = Xo or when a E jlq-l' We may therefore assume that a is a
unit == I mod Po. In that case, we follow the same arguments as in the proof
of Theorem 6.1 using Lemma 6.3 instead of Theorem 6.2. The reader can
check that each step is valid, to conclude the proof.
243
Bibliography
[A-F] Y. AMICE and J. FRESNEL, Fonctions zetap-adiques des corps de

nombres abeliens reels, Acta Arith. XX (1972) pp. 353-384
[A-H] E. ARTIN and H. HASSE, Die beiden Erganzungssatze zum
Reziprozitatsgesetz der In-ten Potenzreste im Korper der In-ten
Einheitswurzeln, Abh. Math. Sem. Hamburg 6 (1928) pp. 146-162
[Ba] H. BASS, Generators and relations for cyclotomic units, Nagoya
Math. J. 27 (1966) pp. 401-407
[Br] A. BRUMMER, On the units of algebraic number fields, Mathe-
matika, 14 (1967) pp. 121-124
[Cal L. CARLITZ, A generalization of Maillet's determinant and a bound
for the first factor of the class number, Proc. AMS 12 (1961) pp.
256-261
rCa-OJ L. CARLITZ and F. R. OLSON, MaiIett's determinant, Proc. AMS
6 (1955) pp. 265-269
[Co 1] J. COATES, On K2 and some classical conjectures in algebraic
number theory, Ann. of Math. 95 (1972) pp. 99-116
[Co 2] J. COATES, K-theory and Iwasawa's analogue of the Jacobian,
Algebraic K-theory II, Springer Lecture Notes 342 (1973) pp. 502-520
[Co 3] J. COATES, p-adic L-functions and Iwasawa's theory, Durham con-
ference on algebraic number theory and class field theory, 1976
[Co 4] J. COATES, Fonctions zeta partieIIes d'un corps de nombres totale-
ment reel, Seminaire Delange-Pisot-Poitou, 1974-75
[C-L] J. COATES and S. LICHTENBAUM, On I-adic zeta functions, Ann.
of Math. 98 (1973) pp 498-550
[C-S 1] J. COATES and W. SINNOTT, On p-adic L-functions over real
quadratic fields, Invent. Math. 25 (1974) pp. 253-279
[C-S 2] J. COATES and W. SINNOTT, An analogue of Stickelberger's
theorem for higher K-groups, Invent. Math. 24 (1974) pp. 149-161
244
Bibliography
[C-S 3] J. COATES and W. SINNOTT, Integrality properties of the values

of partial zeta functions, Proc. London Math. Soc. 1977 pp. 365-384
[C-W 1] J. COATES and A. WILES, Explicit Reciprocity laws, Proceedings,
Conference at Caen, Soc. Math. France Astrisque 41-42 (1977) pp.
7-17
[C-W 2] J. COATES and A. WILES, On the conjecture of Birch and Swinner-
ton-Dyer, Invent. Math. 39 (1977) pp. 223-251
[C-W 3] J. COATES and A. WILES, Kummer's criterion for Hurwitz num-
bers, Kyoto Conference on Algebra Number Theory, 1977
[C-W 4] J. COATES and A. WILES, On the conjecture of Birch-Swinnerton-
Dyer II, to appear
[Col] R. COLEMAN, Some modules attached to Lubin-Tate groups, to
appear
[D-H] H. DAVENPORT and H. HASSE, Die Nullstellen der Kongruenz-
zetafunktionenin gewissen zyklischen Fiillen, J. reine angew. Math.
172 (1935) pp. 151-182
[Fe] B. FERRERO, The cyclotomic Z2-extension of imaginary quadratic
fields, to appear
[Fre] J. FRESNEL, Nombres de Bernoulli et fonctions L p-adiques, Ann.
Inst. Fourier 17 (1967) pp. 281-333
[Fro] A. FROHLICH, Formal groups, Springer Lectures Notes in Math-
ematics 74, 1968
[Gi] R. GILLARD, Unites cyclotomiques, unites semi-locales et ZI-
extensions, Ann. Inst. Fourier, to appear
[Gr 1] R. GREENBERG, A generalization of Kummer's criterion, Invent.
Math. 21 (1973) pp. 247-254
[Gr 2] R. GREENBERG, On a certain /-adic representation, Invent. Math.
21 (1973) pp. 117-124
[Gr 3] R. GREENBERG, The Iwasawa invariants of r-extensions of a
fixed number field, Amer. J. Math. 95 (1973) pp. 204-214
[Gr 4] R. GREENBERG, On the Iwasawa invariants of totally real number
fields, Amer. J. Math. 93 (1976) pp. 263-284
[Gr 5] R. GREENBERG, On p-adic L-functions and cyclotomic fields,
Nagoya Math. J. 56 (1974) pp. 61-77
[Gr 6] R. GREENBERG, On p-adic L-functions and cyclotomic fields II,
to appear
[Gr 7] R. GREENBERG, A note on K2 and the theory of Zp-extensions,
to appear
[Ha 1] H. HASSE, Uber die Klassenzahl abelschen Zahlkorpet, Akademie
Verlag, Berlin, 1952
[Ha 2] H. HASSE, Bericht •.. , Teil II, Reziprozitatsgesetz, Reprinted,
Physica Verlag, Wiirzburg, Wien, 1965
[Ha 3] H. HASSE, Theorie der relativ zyklischen algebraischen Funktion-
enkorper insbesonderen bei endlichen Konstantenkorper, J. reine
angew. Math. 172 (1935) pp. 37-54
[Ho] J. HORN, Cyclotomic units and p-adic L-functions, PhD thesis,
Stanford University, 1976
245
Bibliography
[Hu-V] L. K. HUA and H. S. VANDIVER, Proc. Nat. A cad. Sci. USA 34

(1948) pp. 258-263
[Iw 1] K. IWASAWA, On T-extensions of algebraic number fields, Bull.
Amer. Math. Soc. 65 (1959) pp. 183-226
[Iw 2] K. IWASA W A, A note on the group of units of an algebraic number
field, J. Math. pures et app. 35 (1956) pp. 189-192
[Iw 3] K. IWASAWA, Sheaves for algebraic number fields, Ann. of Math.
69 (1959) pp. 408-413
[Iw 4] K. IW ASAWA, On some properties of T-finite modules, Ann. of
Math. 70 (1959) pp. 291-312
[Iw 5] K. IWASA W A, On the theory of cyclotomic fields, Ann. of Math. 70
(1959) pp. 530-561
[Iw 6] K. IWASA W A, On some invariants of cyclotomic fields, Amer. J.
Math. 80 (1958) pp. 773-783
[Iw 7] K. IW ASA W A, A class number formula for cyclotomic fields, Ann.
of Math. 76 (1962) pp. 171-179
[Iw 8] K. IWASAWA, On some modules in the theory of cyclotomic fields,
J. Math. Soc. Japan Vol. 16, No.1 (1964) pp. 42-82
[Iw 9] K. IWASAWA, Some results in the theory of cyclotomic fields, Sym-
posia in Pure Math. Vol. VIII, AMS 1965, pp. 66-69
[Iw 10] K. IW ASAWA, On explicit formulas for the norm residue symbol,
J. Math. Soc. Japan Vol. 20, Nos. 1-2 (1968) pp. 151-165
[Iw 11] K. IWASA W A, On p-adic L-functions, Ann. of Math. 89 (1969) pp.
198-205
[Iw 12] K. IWASAWA, On Zl-extensions of algebraic number fields, Ann. of
Math. 98 (1973) pp. 246-326
[Iw 13] K. IW ASAWA, A note on cyclotomic fields, Invent. Math. 36 (1976)
pp. 115-123
[Iw 14] K. IW ASAWA, Lectures on p-adic L-functions, Ann. of Math. to
appear
[Ka] N. KATZ, Formal groups and p-adic interpolation, to appear
[Ku 1] D. KUBERT, A system of free generators for the universal even
ordinary Z(2) distribution on Q2kjZ2k, Math. Ann. 224 (1976) pp. 21-31
[Ku2] D. KUBERT, The universal ordinary distribution, to appear.
[KL 1] D. KUBERT and S. LANG, Units in the modular function field, I,
Diophantine applications, Math. Ann. 218 (1975) pp. 67-96
[KL2] D. KUBERT and S. LANG, Idem II, A full set of units, pp. 175-189
[KL 3] D. KUBERT and S. LANG, Idem III, Distribution relations, pp'
273-285
[KL4] D. KUBERT and S. LANG, Idem IV, The Siegel functions are
generators, Math. Ann. 227 (1977) pp. 223-242
[KL5] D. KUBER T and S. LANG, Distributions on toroidal groups, Math.
Zeit. (1976) pp. 33-51
[KL 6] D. KUBERT and S. LANG, The p-primary component of the
cuspidal divisor class group on the modular curve X(P), Math. Ann.
to appear
246
Bibliography
[KL 7] D. KUBERT and S. LANG, The index of Stickeiberger ideals of

order 2 and cuspidal class numbers, to appear
[KL 8] D. KUBERT and S. LANG, Stickelberger ideals, to appear
[KL9] D. KUBERT and S. LANG, Iwasawa theory in the modular tower,
to appear
[Ku-L] T. KUBOTA and H. LEOPOLDT, Eine p-adische Theorie der
Zetawerte, J. reine angew. Math. 214/215 (1964) pp. 328-339
[L 1] S. LANG, Algebraic Number Theory, Addison-Wesley, 1970
[L 2] S. LANG, Elliptic functions, Addison-Wesley, 1973
[L 3] S. LANG, Introduction to modular forms, Springer-Verlag, 1976
[L 4] S. LANG, Elliptic curves: Diophantine analysis, Springer-Verlag,
1978
[Le 1] H. W. LEOPOLDT, Zur Geschlechterhteorie in abelschen Zahl-
korpern, Math. Nach. 9, 6 (1953) pp. 351-363
[Le 2] H. W. LEOPOLDT, Uber Einheitengruppe und Klassenzahl reeIIer
Zahlkorper, Abh. Deutschen Akad. Wiss. Berlin, Akademie Verlag,
Verlag, Berlin, 1954
[Le 3] H. W. LEOPOLDT, Uber ein Fundamentalproblem der Theorie der
Einheiten algebraischer Zahlkorper, Sitzungsbericht Bayerischen
Akademie Wiss. (1956), pp. 41-48
[Le 4] H. W. LEOPOLDT, Eine VeraIIgemeinerung der BernouIlischen
Zahlen, Abh. Math. Sem. Hamburg (1958) pp. 131-140
[Le 5] H. W. LEOPOLDT, Zur Struktur der l-Klassengruppe galoisscher
Zahlkorper, J. reine angew. Math. (1958) pp. 165-174
[Le 6] H. W. LEOPOLDT, Uber Klassenzahlprimteiler reeller abel scher
Zahlkorper, Abh. Math. Sem. Hamburg (1959) pp. 36-47
[Le 7] H. W. LEOPOLDT, Uber die Hauptordnung der ganzen Elemente
eines abelschen Zahlkorpers, J. reine angew. Math. 201 (1959) pp.
113-118
[Le 8] H. W. LEOPOLDT, Uber Fermatquotienten von Kreiseinheiten und
Klassenzahlformeln modulo p, Rend. Circ. Mat. Palermo (1960) pp.
1-12
[Le 9] H. W. LEOPOLDT, Zur approximation des p-adischen Logarithmus,
Abh. Math. Sem. Hamburg 25 (1961) pp. 77-81
[Le 10] H. W. LEOPOLDT, Zur Arithmetik in abelschen Zahlkorpern, J.
reine angew. Math. 209 (1962) pp. 54-71
[Le 11] H. W. LEOPOLDT, Eine p-adische Theorie der Zetawerte II, J. reine
angew. Math. 274-275 (1975) pp. 224-239
[Li 1] S. LICHTENBAUM, Values of zeta functions, Male cohomology,
and algebraic K-theory, Algebraic K-Theory II, Springer-Verlag
Lecture Notes in Mathematics 342 (1973) pp. 489-501
[Li 2] S. LICHTENBAUM, Values of zeta and L-functions at zero, Soc.
Math. France Asterisque 24-25 (1975)
[Li 3] S. LICHTENBAUM, On p-adic L-functions associated to elliptic
curves, to appear
[Lu] J. LUBIN, One, parameter formal Lie groups over p-adic integer
rings, Ann. of Math. 80 (1964) pp. 464-484
247
Bibliography
[L-T] J. LUBIN and J. TATE, Formal complex multiplication in local

fields, Ann. of Math. 8 (1965) pp. 380-387
[Man] J. MANIN, Cyclotomic fields and modular curves, Russian Math.
Surveys Vol. 26, No.6, Nov-Dec 1971, pp. 7-78
[Mas 1] J. MASLEY, On Euclidean rings of integers in cyclotomic fields, J.
reine angew. Math. 272 (1975) pp. 45-48
[Mas 2] J. MASLEY, Solution of the class number two problem for cyclo-
tomic fields, Invent. Math. 28 (1975) pp. 243-244
[M-M] J. MASLEY and H. MONTGOMERY, Cyclotomic fields with
unique factorization, J. reine angew. Math. 286 (1976) pp. 248-
256
[Mi] J. MILNOR, Introduction to algebraic K-theory, Ann. of Math.
Studies 72 (1971)
[Maz] B. MAZUR, Analyse p-adique, Bourbaki report, 1972
[M-SwD] B. MAZUR and H. SWINNERTON-DYER, Arithmetic of Weil
curves, Invent. Math. 18 (1972) pp. 183-266
[No.] . A. P. NOVIKOV, Sur Ie nombre de classes des extensions abeliennes
d'un corps quadratique imaginaire, Izv. Akad. Nauk SSSR 31 (1967)
pp.717-726
[Po] F. POLLACZEK, Uber die irregularen Kreiskorper der [-ten ond
[2-ten Einheitswurzeln, Math. Zeit. 21 (1924) pp. 1-38
[Qu 1] D. QUILLEN, Finite generation of the groups Kz of rings of algebraic
integers, Algebraic K-Theory 1, Springer Lecture Notes 341 (1973)
pp. 179-198
[Qu2] D. QUILLEN, Higher algebraic K-theory I, in Algebraic K-theory
I, Springer Lecture Notes in Mathematics 341 (1973) pp. 85-147
[Ri] K. RIBET, A modular construction of unramified p-extensions of
Q{j.tp), Invent. Math. 34 (1976) pp. 151-162
[Ro] G. ROBERT, Nombres de Hurwitz et unites elliptiques, to appear
[8-T] A. SCHOLZ and o. TAUSSKY, Die Hauptideale der kubischen
Klassenkorper imaginar-quadratischer Zahlkorper, J. reine angew.
Math. 171 (1934) pp. 19-41
[Se 1] J.-P. SERRE, Classes des corps cyclotomiques, d'apres Iwasawa,
Seminaire Bourbaki, 1958
[Se 2] J.-P. SERRE, Formes modulaires et fonctions zeta p-adiques,
Modular functions in one variable III, Springer Lecture Notes 350
(1973)
[Sh] T. SHINTANI, On evaluation of zeta functions of totally real
algebraic number fields at non-positive integers, J. Fac. Sci. Univ.
Tokyo IA Vol. 23, No.2 (1976) pp. 393-417
[Si] C. L. SIEGEL, Uber die Fourierschen Koeffizienten von Modul-
formen, Gottingen Nachrichten 3 (1970) pp. 15-56
[Sin] W. SINNOTT, On the Stickelberger ideal and the circular units, to
appear in Annals of Mathematics
[St] H. STARK, L-functions at s = 1, III: Totally real fields and Hilbert's
twelfth problem, Adv. Math. Vol. 22, No.1 (1976) pp. 64-84
[Ta 1] J. TATE, Letter to Iwasawa on a relation between K2 and Galois
cohomology, in Algebraic K-Theory II, Springer Lecture Notes 342
(1973) pp. 524-527
248
Bibliography
[Ta 2] J. TATE, Relations between K2 and Galois cohomology, Invent.

Math. 36 (1976) pp. 257-274
[Ta 3] J. TATE, Symbols in arithmetic, Actes Congres Intern. Math. 1970,
Tome 1, pp. 201-211
[Va] H. S. VANDIVER, Fermat's last theorem and the second factor in
the cyclotomic class number, Bull. AMS 40 (1934) pp. 118-126
[Wa 1] L. WASHINGTON, Class numbers of elliptic function fields and the
distribution of prime numbers, Acta Arith. XXVII (1975) pp. 111-114
[Wa 2] L. WASHINGTON, Class numbers and Zp-extensions, Math. Ann.
214 (1975) pp. 177-193
[Wa 3] L. WASHINGTON, A note onp-adicL-functions, J. Number Theory 8
Vol. 2 (1976) pp. 245-250
[Wa4] L. WASHINGTON, On Fermat's last theorem, J. reine angew. Math.
to appear
[Wa 5] L. WASHINGTON, Units of irregular cyclotomic fields, to appear
[Wa6] L. WASHINGTON, The class number of the field of 5n th roots of
unity, Proc. AMS, to appear
[Wa 7] L. WASHINGTON, The calculation of Lp(1, x), J. Number Theory,
to appear
[We 1] A. WElL, Number of solutions of equations in finite fields, Bull.
AMS 55 (1949) pp. 497-508
[We 2] A. WElL, Jacobi sums as Grossencharaktere, Trans. AMS 73 (1952)
pp. 487-495
[We 3] A. WElL, Sommes de Jacobi et caracteres de Heeke, Gott. Nach.
(1974) pp. 1-14
[We 4] A. WElL, On some exponential sums, Proc. Nat. A cad. Sci. USA 34,
No.5 (1948) pp. 204-207
[Wi] A. WILES, Higher explicit reciprocity laws, to appear
[Ya 1] K. YAMAMOTO, The gap group of multiplicative relationships of
Gaussian sums, Symposia Mathematica No. 15, (1975) pp. 427-440
[Va 2] K. YAMAMOTO, On a conjecture of Hass concerning multiplicative
relations of Gaussian sums, J. Combin. Theory 1 (1966) pp. 476-489
249
Index
A Conjugate 2
Cyclotomic units 84, 157
Admissible operation 134 Cyclotomic Zp-extension 137, 148
Almost unramified 138
Associated power series 167, 183
D
B Davenport-Hasse distribution 62
Davenport-Hasse theorem 20
Basic Lubin-Tate group 193 Dedekind determinant 89
Bass theorem 160 Distinguished 131
Bernoulli distribution 65 Distribution 33
Bernoulli numbers 15, 34
Bernoulli polynomials 15, 35
E
c Ek,c 38
Euclidean algorithm 129
Carlitz bound 92
Class number formula,p-adic 114
Class number formulas 77 F
Coates-Wiles homomorphism
182, 184 Fermat curve 23
Compatible system 33 First factor 84
Conductor 75 Formal group 190
251
Index
Formal multiplicative group 191 L

Fourier transform 2, 71
Leopoldt space 117
Leopoldt transform 115, 120
G L-functions 74
L-function,p-adic 107
Gamma distribution 66 Lobatchevsky distribution 67
Gauss sum 3, 70 Local symbol 205, 217
Local units 175
Logarithm 211
H Logarithmic derivative 224
Lubin-Tate group 194
Hecke character 16
Herbrand theorem 16
M
I Mahler's theorem 100

Maximal p-abelian p-unramified
Index of Stickelberger ideal 31 extension 143
Integral Stickelberger ideal 43 Measures 34, 95
Involution 150 Mellin transform 105
Iwasawa algebra 95, 125
Iwasawa index 3
Iwasawa involution 150 N
Iwasawa isomorphism 97
Iwasawa-Leopoldt conjecture 80 Nakayama's lemma 126
Iwasawa theory of units 166
o
J
Ordinary distribution 53
Jacobi- sum 4
p
K
p-adic class number formula 114
Kubert's theorem 60 p-adic L-function 107
Kummer congruence 41 p-adic regulator 113
Kummer homomorphism 168, 172 Partial zeta function 65
Kummer pairing 155, 205 Power series associated with
Kummer-Takagi exponent 170 measure 97
252
Index
power series associated with unit T

167, 183
p-ramified 152 Teichmuller character 6, 105
Primitive Stickelberger element 27 Twist 52
Q u
QiCindex 79 Unitization operator 103
Quasi-isomorphism 126, 144 Universal distribution 57
v
R
Vandiver conjecture 142
Reciprocity laws 221 Von Staudt congruence 41
Regulator 85
Regulator, p-adic 113
w
Weierstrass preparation theorem
s 130
Weil's theorem 19
Second factor 84
Special Lubin-Tate group 193
Stickelberger distribution 54 z
Stickelberger element 9; 27
Stickelberger ideal 12, 43 Zp-extension 137
Stickelberger's theorem 6 Zeta function of Fermat curve 25
253
Graduate Texts in Mathematics
Soft and hard cover editions are available for each volume up to vol. 14, hard cover
only from Vol. 15
1 TAKEUTI/ZARING. Introduction to Axiomatic Set Theory.

vii, 250 pages. 1971.
2 OXTOBY. Measure andeategory. viii, 95 pages. 1971. (Hard cover edition only.)
3 SCHAEFFER. Topological Vector Spaces. xi, 294 pages. 1971.
4 HILTON/STAMM BACH. A Course in Homological Algebra.
ix, 338 pages. 1971. (Hard cover edition only)
5 MACLANE. Categories for the Working Mathematician.
ix, 262 pages. 1972.
6 HUGHES/PIPER. Projective Planes. xii, 291 pages. 1973.
7 SERRE. A course in Arithmetic. x, 115 pages. 1973. (Hard cover edition only.)
8 TAKEUTI/ZARING. Axiomatic Set Theory. viii, 238 pages. 1973.
9 HUMPHREYS. Introduction to Lie Algebras and Representation Theory. 2nd
printing, revised. xiv, 171 pages. 1978. (Hard cover edition only.)
10 COHEN. A Course in Simple Homotopy Theory. xii, 114 pages. 1973.
11 CONWAY. Functions of One Complex Variable. 2nd ed. approx. 330 pages. 1978.
(Hard cover edition only.)
12 BEALS. Advanced Mathematical Analysis. xi, 230 pages. 1973.
13 ANDERSON/FuLLER. Rings and Categories of Modules.
ix, 339 pages. 1974.
14 GOLUBITSKY/GuILLEMIN. Stable Mappings and Their Singularities.
x, 211 pages. 1974.
15 BERBERIAN. Lectures in Functional Analysis and Operator Theory.
x, 356 pages. 1974.
16 WINTER. The Structure of Fields. xiii, 205 pages. 1974.
17 ROSENBLATT. Random Processes. 2nd ed. x, 228 pages. 1974.
18 HALMOS. Measure Theory. xi, 304 pages. 1974.
19 HALMOS. A Hilbert Space Problem Book. xvii, 365 pages. 1974.
20 HUSEMOLLER. Fibre Bundles. 2nd ed. xvi, 344 pages. 1975.
21 HUMPHREYS. Linear Algebraic Groups. xiv. 272 pages. 1975.
22 BARNES/MACK. An Algebraic Introduction to Mathematical Logic.
x, 137 pages. 1975.
23 GREUB. Linear Algebra. 4th ed. xvii, 451 pages. 1975.
24 HOLMES. Geometric Functional Analysis and Its Applications.
x, 246 pages. 1975.
25 HEWITI/STROMBERG. Real and Abstract Analysis. 4th printing. viii,476
pages. 1978.
26 MANES. Algebraic Theories. x, 356 pages. 1976.
27 KELLEY. General Topology. xiv, 298 pages. 1975.
28 ZARISKI/SAMUEL. Commutative_Alg~bra 1. xi, 329 pages. 1975.
29 ZARISKI/SAMUEL. Commutative Algebra II. x, 414 pages. 1976.
30 JACOBSON. Lectures in Abstract Alegbra I: Basic Concepts.
xii, 205 pages! 1976.
31 JACOBSON. Lectures in Abstract Algebra II: Linear Algebra.
xii, 280 pages. "1975.
32 JACOBSON. Lectures in Abstract Algebra III: Theory of Fields and
Galois Theory. ix, 324 pages. 1976.
33 HIRSCH. Differential Topology. x, 222 pages. 1976.
34 SPITZER. Principles of Random Walk. 2nd ed. xiii, 408 pages. 1976.
35 WERMER. Banach Algebras and Several Complex Va~iables. 2nd ed.
xiv, 162 pages. 1976.
36 KELLEY/NAMIOKA. Linear Topological Spaces. xv, 256 pages. 1976.
37 MONK. Mathematical Logic. x, 531 pages. 1976.
38 GRAUERT/FRITZSCHE. Several Complex Variables. viii, 207 pages. 1976
39 ARVESON. An Invitation ,to C*-Algebras. x, 106 pages. 1976.
40 KEMENy/SNELL/KNAPP. Denumerable Markov Chains. 2nd ed.
xii, 484 pages. 1976.
41 ApOSTOL. Modular Functions and Dirichlet Series in Number Theory.
x, 198 pages. 1976.
42 SERRE. Linear Representations of Finite Groups. 176 pages. 1977.
43 GILLMAN/JERISON. Rings of Continuous Functions. xiii, 300 pages. 1976.
44 KENDIG. Elementary Algebraic Geometry. viii, 309 pages. 1977.
45 LOEVE. Probability Theory. 4th ed. Vol. 1. xvii, 425 pages. 1977.
46 LOEVE. Probability Theory. 4th ed. \obI. 2. xvi, 413 pages. 1978.
47 MOISE. Geometric Topology in Dimensions 2 and 3. x, 262 pages. 1977.
48 SACHS/WU. General Relativity for Mathematicians. xii, 291 pages. 1977.
49 GRUENBERG/WEIR. Linear Geometry. 2nd ed. x, 198 pages. 1977.
50 EDWARDS. Fermat's Last Theorem. xv, 410 pages. 1977.
51 KLINGENBERG. A Course in Differential Geometry. xii, 192 pages. 1978.
52 HARTSHORNE. Algebraic Geometry. xvi, 496 pages. 1977.
53 MANIN. A Course in Mathematical Logic. xiii, 286 pages. 1977.
54 GRAVER/WATKINS. 'Combinatorics with Emphasis on the Theory of
Graphs. xv, 368 pages. 1977.
55 BROWN/PEARCY. Introduction to Operator Theory. Vol. 1: Elements
of Functional Analysis. xiv, 474 pages. 1977.
56 MASSEY. Algebraic Topology: An Introduction. xxi, 261 pages. 1977.
57 CROWELL/Fox. In~roduction to Knot Theory. x, 182 pages. 1977.
58 KOBLITZ. p-adic Numbers, p-adic Analysis, and Zeta-Functions.
x, 122 pages. 1977.
59 LANG. Cyclotomic Fields. xi, 272 pages. 1978.
60 ARNOLD. Mathematical Methods in Classical Mechanics.
approx. 350 pages. 1978.
61 WHITEHEAD. Elements of Homotopy Theory. approx. 500 pages. 1978.

Cycloides Field-Serge Lang

Uploaded by

Copyright:

Available Formats

Cycloides Field-Serge Lang

Uploaded by

Document Information

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

Cycloides Field-Serge Lang

Uploaded by

Copyright:

Available Formats

Graduate Texts in Mathematics

P. R. Halmos F. W. Gehring C. C. Moore

AMS Subject Classification: 12C20, 12B30, 14G20

Library of Congress Cataloging in Publication Data

All rights reserved.

No part of this book may be translated or reproduced in any form

© 1978 by Springer-Verlag, New York Inc.

Birch-Swinnerton-Dyer conjecture, new insight into the explicit reciprocity

New Haven, Connecticut SERGE LANG

3. Changing the Prime 200

ZeN) = integers mod N = Zj NZ.

§1. Character Sums Over Finite Fields

X: F* -i>- Jlq -1 denotes a character of the multiplicative group.

Let/, g be functions on Fwith values in a fixed algebraically closed field of

We define the Fourier transform Tfby

Tf(y) = L f(x)Je( -xy) = Lf(x)e-Tr<XY).

Then Tf is again a function on F, identified with its character group by Je,

T2f(z) = L Lf(x)Je( - yx)Je( -zy)

We define the convolutionf * g between functions by the formula

(f * g)(y) = L f(x)g(y - x).

A change of variables shows that

Theorem 1.2. For functions J, g on F we have

Proof For the first formula we have

T(f*g)(z) = L(f*g)(Y)A(-ZY) = LLf(x)g(y - X)A(-ZY).

We change the order of summation, let t = Y - x, Y = x + t, and find

= LfCx)A( -zx) Lg(t)A( -zt)

The second formula follows from the first because T is an isomorphism

We shall be concerned with the Gauss sums (Lagrange resolvant)

Sex, A) = sex) = L XCU)A(U)

GS O. Let Xl be the trivial character 1 on F*. Then

SCI) = SCI, A),

GS 1. For any character X '=f. 1, we have TX = X( -I)S(x)X- l •

TX(Y) = L.. x(x),.l( - yx) .

GS 2. We have S(i) = X( -1)S(x) andfor X "# 1, S(x)S(i) = X( -1)q, so

Over the complex numbers, we obtain the absolute value

We define the Jacobi sum

J(xI> X2) = - L.. Xl(X)X2(1 - x) .

Observe the minus sign, a most useful convention. We have

J(1, 1) = -(q - 2).

GS 3. If XIX2 "# 1 then

JUI> X2) = Xl( -1).

Thus S(xP) is obtained from S(X) by permuting the elements of F under

Let m be a positive integer dividing q - 1, and suppose that X has order m,

Then the values of X are in Q(l1m) and

For any integer c prime to m we have an automorphism (le,l of Q(l1m, I1p)

(l e,l: ( 1-7 (e and (l e,l is identity on I1p.

For any integer v prime to p, we have an automorphism (ll,v such that

The diagram of fields is as follows.

Theorem 1.3. Assume that X has order m.

§2. Stickelberger's Theorem

The Teichmuller character generates the character group of F*, so any

because elementary algebraic number theory shows that p is totally ramified

For an arbitrary integer k, we define s(k) to be periodic mod q - 1, and

with 1 instead of -Ion the right-hand side.

Proof of Theorem 2.1. If k = 0 then the relation of Theorem 2.1 is clear

(interpreting Tr u as an integer in the given residue class mod p). But

W(U)-I Tr(u) == u-I(u + uP + ... + upn - 1 ) mod ~

L W(U)-I Tr(u) == q - 1 == -1 (mod ~)

thus proving the theorem for k = 1.

Assume now the result proved for k - 1, and write

T(fg)(z) = L(fg)(Y)A(-ZY) = LLf(x)g(y - X)A(-ZY).